2017-01-26 00:13:56 +00:00
// Copyright 2016 Mozilla
//
// Licensed under the Apache License, Version 2.0 (the "License"); you may not use
// this file except in compliance with the License. You may obtain a copy of the
// License at http://www.apache.org/licenses/LICENSE-2.0
// Unless required by applicable law or agreed to in writing, software distributed
// under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR
// CONDITIONS OF ANY KIND, either express or implied. See the License for the
// specific language governing permissions and limitations under the License.
#![ allow(dead_code) ]
2019-07-17 15:46:47 +00:00
use failure ::ResultExt ;
2018-06-06 01:23:59 +00:00
2019-07-17 15:46:47 +00:00
use std ::collections ::hash_map ::Entry ;
2017-02-15 00:50:40 +00:00
use std ::collections ::HashMap ;
use std ::iter ::{ once , repeat } ;
use std ::ops ::Range ;
2017-02-09 20:23:04 +00:00
use std ::path ::Path ;
2017-02-08 22:04:32 +00:00
use itertools ;
use itertools ::Itertools ;
2017-01-26 00:13:56 +00:00
use rusqlite ;
2017-03-13 16:39:19 +00:00
use rusqlite ::limits ::Limit ;
2017-12-05 15:58:24 +00:00
use rusqlite ::types ::{ ToSql , ToSqlOutput } ;
2019-07-17 15:46:47 +00:00
use rusqlite ::TransactionBehavior ;
2017-01-26 00:13:56 +00:00
2020-08-06 03:03:58 +00:00
use crate ::bootstrap ;
use crate ::{ repeat_values , to_namespaced_keyword } ;
2017-04-29 03:11:55 +00:00
2019-07-17 15:46:47 +00:00
use edn ::{ DateTime , Utc , Uuid , Value } ;
2017-04-29 03:11:55 +00:00
2020-08-06 03:03:58 +00:00
use crate ::entids ;
2018-08-08 17:35:06 +00:00
2019-07-17 15:46:47 +00:00
use core_traits ::{ attribute , Attribute , AttributeBitFlags , Entid , TypedValue , ValueType } ;
use mentat_core ::{ AttributeMap , FromMicros , IdentMap , Schema , ToMicros , ValueRc } ;
use db_traits ::errors ::{ DbErrorKind , Result } ;
2018-08-08 17:37:59 +00:00
2020-08-06 03:03:58 +00:00
use crate ::metadata ;
use crate ::schema ::SchemaBuilding ;
use crate ::tx ::transact ;
use crate ::types ::{ AVMap , AVPair , Partition , PartitionMap , DB } ;
2017-01-26 00:13:56 +00:00
2020-08-06 03:03:58 +00:00
use crate ::watcher ::NullWatcher ;
2020-01-14 15:46:21 +00:00
use std ::convert ::TryInto ;
2018-03-06 17:01:20 +00:00
2018-06-13 15:49:40 +00:00
// In PRAGMA foo='bar', `'bar'` must be a constant string (it cannot be a
// bound parameter), so we need to escape manually. According to
// https://www.sqlite.org/faq.html, the only character that must be escaped is
// the single quote, which is escaped by placing two single quotes in a row.
fn escape_string_for_pragma ( s : & str ) -> String {
s . replace ( " ' " , " '' " )
}
2019-07-17 15:46:47 +00:00
fn make_connection (
uri : & Path ,
maybe_encryption_key : Option < & str > ,
) -> rusqlite ::Result < rusqlite ::Connection > {
2018-06-13 15:49:40 +00:00
let conn = match uri . to_string_lossy ( ) . len ( ) {
2017-02-09 20:23:04 +00:00
0 = > rusqlite ::Connection ::open_in_memory ( ) ? ,
_ = > rusqlite ::Connection ::open ( uri ) ? ,
} ;
2018-06-13 15:49:40 +00:00
let page_size = 32768 ;
let initial_pragmas = if let Some ( encryption_key ) = maybe_encryption_key {
2020-01-23 18:16:19 +00:00
if ! cfg! ( feature = " sqlcipher " ) {
panic! ( " This function shouldn't be called with a key unless we have sqlcipher support " ) ;
}
2018-06-13 15:49:40 +00:00
// Important: The `cipher_page_size` cannot be changed without breaking
// the ability to open databases that were written when using a
// different `cipher_page_size`. Additionally, it (AFAICT) must be a
// positive multiple of `page_size`. We use the same value for both here.
2019-07-17 15:46:47 +00:00
format! (
"
2018-06-13 15:49:40 +00:00
PRAGMA key = ' { } ' ;
PRAGMA cipher_page_size = { } ;
2019-07-17 15:46:47 +00:00
" ,
escape_string_for_pragma ( encryption_key ) ,
page_size
)
2018-06-13 15:49:40 +00:00
} else {
String ::new ( )
} ;
2018-01-17 22:36:15 +00:00
// See https://github.com/mozilla/mentat/issues/505 for details on temp_store
// pragma and how it might interact together with consumers such as Firefox.
// temp_store=2 is currently present to force SQLite to store temp files in memory.
// Some of the platforms we support do not have a tmp partition (e.g. Android)
// necessary to store temp files on disk. Ideally, consumers should be able to
// override this behaviour (see issue 505).
2019-07-17 15:46:47 +00:00
conn . execute_batch ( & format! (
"
2018-06-13 15:49:40 +00:00
{ }
2017-02-09 20:23:04 +00:00
PRAGMA journal_mode = wal ;
PRAGMA wal_autocheckpoint = 32 ;
PRAGMA journal_size_limit = 3145728 ;
PRAGMA foreign_keys = ON ;
2017-12-20 06:52:09 +00:00
PRAGMA temp_store = 2 ;
2019-07-17 15:46:47 +00:00
" ,
initial_pragmas
) ) ? ;
2017-02-09 20:23:04 +00:00
Ok ( conn )
2017-01-26 00:13:56 +00:00
}
2019-07-17 15:46:47 +00:00
pub fn new_connection < T > ( uri : T ) -> rusqlite ::Result < rusqlite ::Connection >
where
T : AsRef < Path > ,
{
2018-06-13 15:49:40 +00:00
make_connection ( uri . as_ref ( ) , None )
}
#[ cfg(feature = " sqlcipher " ) ]
2019-07-17 15:46:47 +00:00
pub fn new_connection_with_key < P , S > (
uri : P ,
encryption_key : S ,
) -> rusqlite ::Result < rusqlite ::Connection >
where
P : AsRef < Path > ,
S : AsRef < str > ,
{
2018-06-13 15:49:40 +00:00
make_connection ( uri . as_ref ( ) , Some ( encryption_key . as_ref ( ) ) )
}
#[ cfg(feature = " sqlcipher " ) ]
2019-07-17 15:46:47 +00:00
pub fn change_encryption_key < S > (
conn : & rusqlite ::Connection ,
encryption_key : S ,
) -> rusqlite ::Result < ( ) >
where
S : AsRef < str > ,
{
2018-06-13 15:49:40 +00:00
let escaped = escape_string_for_pragma ( encryption_key . as_ref ( ) ) ;
// `conn.execute` complains that this returns a result, and using a query
// for it requires more boilerplate.
conn . execute_batch ( & format! ( " PRAGMA rekey = ' {} '; " , escaped ) )
}
2017-01-26 00:13:56 +00:00
/// Version history:
///
2017-04-29 03:11:55 +00:00
/// 1: initial Rust Mentat schema.
pub const CURRENT_VERSION : i32 = 1 ;
2017-01-26 00:13:56 +00:00
2017-03-21 16:50:31 +00:00
/// MIN_SQLITE_VERSION should be changed when there's a new minimum version of sqlite required
/// for the project to work.
2020-01-23 18:16:19 +00:00
const MIN_SQLITE_VERSION : i32 = 3_008_000 ;
2017-03-21 16:50:31 +00:00
2020-01-23 18:16:19 +00:00
const TRUE : & bool = & true ;
const FALSE : & bool = & false ;
2017-01-26 00:13:56 +00:00
/// Turn an owned bool into a static reference to a bool.
///
/// `rusqlite` is designed around references to values; this lets us use computed bools easily.
#[ inline(always) ]
fn to_bool_ref ( x : bool ) -> & 'static bool {
2019-07-17 15:46:47 +00:00
if x {
TRUE
} else {
FALSE
}
2017-01-26 00:13:56 +00:00
}
lazy_static! {
2017-04-29 03:11:55 +00:00
/// SQL statements to be executed, in order, to create the Mentat SQL schema (version 1).
2017-01-26 00:13:56 +00:00
#[ cfg_attr(rustfmt, rustfmt_skip) ]
2017-04-29 03:11:55 +00:00
static ref V1_STATEMENTS : Vec < & 'static str > = { vec! [
2017-01-26 00:13:56 +00:00
r #" CREATE TABLE datoms (e INTEGER NOT NULL, a SMALLINT NOT NULL, v BLOB NOT NULL, tx INTEGER NOT NULL,
value_type_tag SMALLINT NOT NULL ,
index_avet TINYINT NOT NULL DEFAULT 0 , index_vaet TINYINT NOT NULL DEFAULT 0 ,
index_fulltext TINYINT NOT NULL DEFAULT 0 ,
unique_value TINYINT NOT NULL DEFAULT 0 ) " #,
r # "CREATE UNIQUE INDEX idx_datoms_eavt ON datoms (e, a, value_type_tag, v)"# ,
r # "CREATE UNIQUE INDEX idx_datoms_aevt ON datoms (a, e, value_type_tag, v)"# ,
// Opt-in index: only if a has :db/index true.
r # "CREATE UNIQUE INDEX idx_datoms_avet ON datoms (a, value_type_tag, v, e) WHERE index_avet IS NOT 0"# ,
// Opt-in index: only if a has :db/valueType :db.type/ref. No need for tag here since all
// indexed elements are refs.
r # "CREATE UNIQUE INDEX idx_datoms_vaet ON datoms (v, a, e) WHERE index_vaet IS NOT 0"# ,
// Opt-in index: only if a has :db/fulltext true; thus, it has :db/valueType :db.type/string,
// which is not :db/valueType :db.type/ref. That is, index_vaet and index_fulltext are mutually
// exclusive.
r # "CREATE INDEX idx_datoms_fulltext ON datoms (value_type_tag, v, a, e) WHERE index_fulltext IS NOT 0"# ,
// TODO: possibly remove this index. :db.unique/{value,identity} should be asserted by the
// transactor in all cases, but the index may speed up some of SQLite's query planning. For now,
// it serves to validate the transactor implementation. Note that tag is needed here to
// differentiate, e.g., keywords and strings.
r # "CREATE UNIQUE INDEX idx_datoms_unique_value ON datoms (a, value_type_tag, v) WHERE unique_value IS NOT 0"# ,
2018-07-11 01:20:38 +00:00
r # "CREATE TABLE timelined_transactions (e INTEGER NOT NULL, a SMALLINT NOT NULL, v BLOB NOT NULL, tx INTEGER NOT NULL, added TINYINT NOT NULL DEFAULT 1, value_type_tag SMALLINT NOT NULL, timeline TINYINT NOT NULL DEFAULT 0)"# ,
r # "CREATE INDEX idx_timelined_transactions_timeline ON timelined_transactions (timeline)"# ,
r # "CREATE VIEW transactions AS SELECT e, a, v, value_type_tag, tx, added FROM timelined_transactions WHERE timeline IS 0"# ,
2017-01-26 00:13:56 +00:00
// Fulltext indexing.
// A fulltext indexed value v is an integer rowid referencing fulltext_values.
// Optional settings:
// tokenize="porter"#,
// prefix='2,3'
// By default we use Unicode-aware tokenizing (particularly for case folding), but preserve
// diacritics.
r #" CREATE VIRTUAL TABLE fulltext_values
USING FTS4 ( text NOT NULL , searchid INT , tokenize = unicode61 " remove_diacritics=0 " ) " #,
// This combination of view and triggers allows you to transparently
// update-or-insert into FTS. Just INSERT INTO fulltext_values_view (text, searchid).
r # "CREATE VIEW fulltext_values_view AS SELECT * FROM fulltext_values"# ,
r #" CREATE TRIGGER replace_fulltext_searchid
INSTEAD OF INSERT ON fulltext_values_view
WHEN EXISTS ( SELECT 1 FROM fulltext_values WHERE text = new . text )
BEGIN
UPDATE fulltext_values SET searchid = new . searchid WHERE text = new . text ;
END " #,
r #" CREATE TRIGGER insert_fulltext_searchid
INSTEAD OF INSERT ON fulltext_values_view
WHEN NOT EXISTS ( SELECT 1 FROM fulltext_values WHERE text = new . text )
BEGIN
INSERT INTO fulltext_values ( text , searchid ) VALUES ( new . text , new . searchid ) ;
END " #,
// A view transparently interpolating fulltext indexed values into the datom structure.
r #" CREATE VIEW fulltext_datoms AS
SELECT e , a , fulltext_values . text AS v , tx , value_type_tag , index_avet , index_vaet , index_fulltext , unique_value
FROM datoms , fulltext_values
WHERE datoms . index_fulltext IS NOT 0 AND datoms . v = fulltext_values . rowid " #,
// A view transparently interpolating all entities (fulltext and non-fulltext) into the datom structure.
r #" CREATE VIEW all_datoms AS
SELECT e , a , v , tx , value_type_tag , index_avet , index_vaet , index_fulltext , unique_value
FROM datoms
WHERE index_fulltext IS 0
UNION ALL
SELECT e , a , v , tx , value_type_tag , index_avet , index_vaet , index_fulltext , unique_value
FROM fulltext_datoms " #,
Schema alteration. Fixes #294 and #295. (#370) r=rnewman
* Pre: Don't retract :db/ident in test.
Datomic (and eventually Mentat) don't allow to retract :db/ident in
this way, so this runs afoul of future work to support mutating
metadata.
* Pre: s/VALUETYPE/VALUE_TYPE/.
This is consistent with the capitalization (which is "valueType") and
the other identifier.
* Pre: Remove some single quotes from error output.
* Part 1: Make materialized views be uniform [e a v value_type_tag].
This looks ahead to a time when we could support arbitrary
user-defined materialized views. For now, the "idents" materialized
view is those datoms of the form [e :db/ident :namespaced/keyword] and
the "schema" materialized view is those datoms of the form [e a v]
where a is in a particular set of attributes that will become clear in
the following commits.
This change is not backwards compatible, so I'm removing the open
current (really, v2) test. It'll be re-instated when we get to
https://github.com/mozilla/mentat/issues/194.
* Pre: Map TypedValue::Ref to TypedValue::Keyword in debug output.
* Part 3: Separate `schema_to_mutate` from the `schema` used to interpret.
This is just to keep track of the expected changes during
bootstrapping. I want bootstrap metadata mutations to flow through
the same code path as metadata mutations during regular transactions;
by differentiating the schema used for interpretation from the schema
that will be updated I expect to be able to apply bootstrap metadata
mutations to an empty schema and have things like materialized views
created (using the regular code paths).
This commit has been re-ordered for conceptual clarity, but it won't
compile because it references the metadata module. It's possible to
make it compile -- the functionality is there in the schema module --
but it's not worth the rebasing effort until after review (and
possibly not even then, since we'll squash down to a single commit to
land).
* Part 2: Maintain entids separately from idents.
In order to support historical idents, we need to distinguish the
"current" map from entid -> ident from the "complete historical" map
ident -> entid. This is what Datomic does; in Datomic, an ident is
never retracted (although it can be replaced). This approach is an
important part of allowing multiple consumers to share a schema
fragment as it migrates forward.
This fixes a limitation of the Clojure implementation, which did not
handle historical idents across knowledge base close and re-open.
The "entids" materialized view is naturally a slice of the "datoms"
table. The "idents" materialized view is a slice of the
"transactions" table. I hope that representing in this way, and
casting the problem in this light, might generalize to future
materialized views.
* Pre: Add DiffSet.
* Part 4: Collect mutations to a `Schema`.
I haven't taken your review comment about consuming AttributeBuilder
during each fluent function. If you read my response and still want
this, I'm happy to do it in review.
* Part 5: Handle :db/ident and :db.{install,alter}/attribute.
This "loops" the committed datoms out of the SQL store and back
through the metadata (schema, but in future also partition map)
processor. The metadata processor updates the schema and produces a
report of what changed; that report is then used to update the SQL
store. That update includes:
- the materialized views ("entids", "idents", and "schema");
- if needed, a subset of the datoms themselves (as flags change).
I've left a TODO for handling attribute retraction in the cases that
it makes sense. I expect that to be straight-forward.
* Review comment: Rename DiffSet to AddRetractAlterSet.
Also adds a little more commentary and a simple test.
* Review comment: Use ToIdent trait.
* Review comment: partially revert "Part 2: Maintain entids separately from idents."
This reverts commit 23a91df9c35e14398f2ddbd1ba25315821e67401.
Following our discussion, this removes the "entids" materialized
view. The next commit will remove historical idents from the "idents"
materialized view.
* Post: Use custom Either rather than std::result::Result.
This is not necessary, but it was suggested that we might be paying an
overhead creating Err instances while using error_chain. That seems
not to be the case, but this change shows that we don't actually use
any of the Result helper methods, so there's no reason to overload
Result. This change might avoid some future confusion, so I'm going
to land it anyway.
Signed-off-by: Nick Alexander <nalexander@mozilla.com>
* Review comment: Don't preserve historical idents.
* Review comment: More prepared statements when updating materialized views.
* Post: Test altering :db/cardinality and :db/unique.
These tests fail due to a Datomic limitation, namely that the marker
flag :db.alter/attribute can only be asserted once for an attribute!
That is, [:db.part/db :db.alter/attribute :attribute] will only be
transacted at most once. Since older versions of Datomic required the
:db.alter/attribute flag, I can only imagine they either never wrote
:db.alter/attribute to the store, or they handled it specially. I'll
need to remove the marker flag system from Mentat in order to address
this fundamental limitation.
* Post: Remove some more single quotes from error output.
* Post: Add assert_transact! macro to unwrap safely.
I was finding it very difficult to track unwrapping errors while
making changes, due to an underlying Mac OS X symbolication issue that
makes running tests with RUST_BACKTRACE=1 so slow that they all time
out.
* Post: Don't expect or recognize :db.{install,alter}/attribute.
I had this all working... except we will never see a repeated
`[:db.part/db :db.alter/attribute :attribute]` assertion in the store!
That means my approach would let you alter an attribute at most one
time. It's not worth hacking around this; it's better to just stop
expecting (and recognizing) the marker flags. (We have all the data
to distinguish the various cases that we need without the marker
flags.)
This brings Mentat in line with the thrust of newer Datomic versions,
but isn't compatible with Datomic, because (if I understand correctly)
Datomic automatically adds :db.{install,alter}/attribute assertions to
transactions.
I haven't purged the corresponding :db/ident and schema fragments just
yet:
- we might want them back
- we might want them in order to upgrade v1 and v2 databases to the
new on-disk layout we're fleshing out (v3?).
* Post: Don't make :db/unique :db.unique/* imply :db/index true.
This patch avoids a potential bug with the "schema" materialized view.
If :db/unique :db.unique/value implies :db/index true, then what
happens when you _retract_ :db.unique/value? I think Datomic defines
this in some way, but I really want the "schema" materialized view to
be a slice of "datoms" and not have these sort of ambiguities and
persistent effects. Therefore, to ensure that we don't retract a
schema characteristic and accidentally change more than we intended
to, this patch stops having any schema characteristic imply any other
schema characteristic(s). To achieve that, I added an
Option<Unique::{Value,Identity}> type to Attribute; this helps with
this patch, and also looks ahead to when we allow to retract
:db/unique attributes.
* Post: Allow to retract :db/ident.
* Post: Include more details about invalid schema changes.
The tests use strings, so they hide the chained errors which do in
fact provide more detail.
* Review comment: Fix outdated comment.
* Review comment: s/_SET/_SQL_LIST/.
* Review comment: Use a sub-select for checking cardinality.
This might be faster in practice.
* Review comment: Put `attribute::Unique` into its own namespace.
2017-03-20 20:18:59 +00:00
// Materialized views of the metadata.
r # "CREATE TABLE idents (e INTEGER NOT NULL, a SMALLINT NOT NULL, v BLOB NOT NULL, value_type_tag SMALLINT NOT NULL)"# ,
r # "CREATE INDEX idx_idents_unique ON idents (e, a, v, value_type_tag)"# ,
r # "CREATE TABLE schema (e INTEGER NOT NULL, a SMALLINT NOT NULL, v BLOB NOT NULL, value_type_tag SMALLINT NOT NULL)"# ,
r # "CREATE INDEX idx_schema_unique ON schema (e, a, v, value_type_tag)"# ,
2018-07-18 00:54:13 +00:00
2017-02-15 00:50:40 +00:00
// TODO: store entid instead of ident for partition name.
2018-07-18 00:54:13 +00:00
r # "CREATE TABLE known_parts (part TEXT NOT NULL PRIMARY KEY, start INTEGER NOT NULL, end INTEGER NOT NULL, allow_excision SMALLINT NOT NULL)"# ,
2017-01-26 00:13:56 +00:00
]
} ;
}
/// Set the SQLite user version.
///
/// Mentat manages its own SQL schema version using the user version. See the [SQLite
/// documentation](https://www.sqlite.org/pragma.html#pragma_user_version).
fn set_user_version ( conn : & rusqlite ::Connection , version : i32 ) -> Result < ( ) > {
2020-01-14 15:46:21 +00:00
conn . execute (
& format! ( " PRAGMA user_version = {} " , version ) ,
rusqlite ::params! [ ] ,
)
. context ( DbErrorKind ::CouldNotSetVersionPragma ) ? ;
2018-06-06 01:23:59 +00:00
Ok ( ( ) )
2017-01-26 00:13:56 +00:00
}
/// Get the SQLite user version.
///
/// Mentat manages its own SQL schema version using the user version. See the [SQLite
/// documentation](https://www.sqlite.org/pragma.html#pragma_user_version).
fn get_user_version ( conn : & rusqlite ::Connection ) -> Result < i32 > {
2019-07-17 15:46:47 +00:00
let v = conn
2019-07-22 11:40:59 +00:00
. query_row ( " PRAGMA user_version " , rusqlite ::params! [ ] , | row | row . get ( 0 ) )
2019-07-17 15:46:47 +00:00
. context ( DbErrorKind ::CouldNotGetVersionPragma ) ? ;
2018-06-06 01:23:59 +00:00
Ok ( v )
2017-01-26 00:13:56 +00:00
}
2018-02-16 10:01:00 +00:00
/// Do just enough work that either `create_current_version` or sync can populate the DB.
2019-07-17 15:46:47 +00:00
pub fn create_empty_current_version (
conn : & mut rusqlite ::Connection ,
) -> Result < ( rusqlite ::Transaction , DB ) > {
2017-12-05 15:58:24 +00:00
let tx = conn . transaction_with_behavior ( TransactionBehavior ::Exclusive ) ? ;
2017-01-26 00:13:56 +00:00
2017-04-29 03:11:55 +00:00
for statement in ( & V1_STATEMENTS ) . iter ( ) {
2019-07-22 11:40:59 +00:00
tx . execute ( statement , rusqlite ::params! [ ] ) ? ;
2017-01-26 00:13:56 +00:00
}
2018-02-16 10:01:00 +00:00
set_user_version ( & tx , CURRENT_VERSION ) ? ;
let bootstrap_schema = bootstrap ::bootstrap_schema ( ) ;
2017-01-26 00:13:56 +00:00
let bootstrap_partition_map = bootstrap ::bootstrap_partition_map ( ) ;
2018-02-16 10:01:00 +00:00
Ok ( ( tx , DB ::new ( bootstrap_partition_map , bootstrap_schema ) ) )
}
2018-07-18 00:54:13 +00:00
/// Creates a partition map view for the main timeline based on partitions
/// defined in 'known_parts'.
fn create_current_partition_view ( conn : & rusqlite ::Connection ) -> Result < ( ) > {
let mut stmt = conn . prepare ( " SELECT part, end FROM known_parts ORDER BY end ASC " ) ? ;
2019-07-17 15:46:47 +00:00
let known_parts : Result < Vec < ( String , i64 ) > > = stmt
2019-07-22 11:40:59 +00:00
. query_and_then ( rusqlite ::params! [ ] , | row | Ok ( ( row . get ( 0 ) ? , row . get ( 1 ) ? ) ) ) ?
2019-07-17 15:46:47 +00:00
. collect ( ) ;
2018-07-18 00:54:13 +00:00
let mut case = vec! [ ] ;
for & ( ref part , ref end ) in known_parts ? . iter ( ) {
case . push ( format! ( r # "WHEN e <= {} THEN "{}""# , end , part ) ) ;
}
2019-07-17 15:46:47 +00:00
let view_stmt = format! (
" CREATE VIEW parts AS
2018-07-18 00:54:13 +00:00
SELECT
CASE { } END AS part ,
min ( e ) AS start ,
max ( e ) + 1 AS idx
FROM timelined_transactions WHERE timeline = { } GROUP BY part " ,
2019-07-17 15:46:47 +00:00
case . join ( " " ) ,
2020-08-06 03:03:58 +00:00
crate ::TIMELINE_MAIN
2018-07-18 00:54:13 +00:00
) ;
2019-07-22 11:40:59 +00:00
conn . execute ( & view_stmt , rusqlite ::params! [ ] ) ? ;
2018-07-18 00:54:13 +00:00
Ok ( ( ) )
}
2018-02-16 10:01:00 +00:00
// TODO: rename "SQL" functions to align with "datoms" functions.
pub fn create_current_version ( conn : & mut rusqlite ::Connection ) -> Result < DB > {
let ( tx , mut db ) = create_empty_current_version ( conn ) ? ;
2020-01-16 15:55:05 +00:00
// TODO: think more carefully about allocating new parts and bit-masking part ranges.
2017-01-26 00:13:56 +00:00
// TODO: install these using bootstrap assertions. It's tricky because the part ranges are implicit.
// TODO: one insert, chunk into 999/3 sections, for safety.
2017-02-15 00:50:40 +00:00
// This is necessary: `transact` will only UPDATE parts, not INSERT them if they're missing.
2018-02-16 10:01:00 +00:00
for ( part , partition ) in db . partition_map . iter ( ) {
2017-01-26 00:13:56 +00:00
// TODO: Convert "keyword" part to SQL using Value conversion.
2019-07-17 15:46:47 +00:00
tx . execute (
" INSERT INTO known_parts (part, start, end, allow_excision) VALUES (?, ?, ?, ?) " ,
& [
part ,
& partition . start . to_string ( ) ,
& partition . end . to_string ( ) ,
2020-01-16 15:55:05 +00:00
& ( partition . allow_excision as i8 ) . to_string ( ) ,
2019-07-17 15:46:47 +00:00
] ,
) ? ;
2017-01-26 00:13:56 +00:00
}
2018-07-18 00:54:13 +00:00
create_current_partition_view ( & tx ) ? ;
2017-02-15 00:50:40 +00:00
// TODO: return to transact_internal to self-manage the encompassing SQLite transaction.
Schema alteration. Fixes #294 and #295. (#370) r=rnewman
* Pre: Don't retract :db/ident in test.
Datomic (and eventually Mentat) don't allow to retract :db/ident in
this way, so this runs afoul of future work to support mutating
metadata.
* Pre: s/VALUETYPE/VALUE_TYPE/.
This is consistent with the capitalization (which is "valueType") and
the other identifier.
* Pre: Remove some single quotes from error output.
* Part 1: Make materialized views be uniform [e a v value_type_tag].
This looks ahead to a time when we could support arbitrary
user-defined materialized views. For now, the "idents" materialized
view is those datoms of the form [e :db/ident :namespaced/keyword] and
the "schema" materialized view is those datoms of the form [e a v]
where a is in a particular set of attributes that will become clear in
the following commits.
This change is not backwards compatible, so I'm removing the open
current (really, v2) test. It'll be re-instated when we get to
https://github.com/mozilla/mentat/issues/194.
* Pre: Map TypedValue::Ref to TypedValue::Keyword in debug output.
* Part 3: Separate `schema_to_mutate` from the `schema` used to interpret.
This is just to keep track of the expected changes during
bootstrapping. I want bootstrap metadata mutations to flow through
the same code path as metadata mutations during regular transactions;
by differentiating the schema used for interpretation from the schema
that will be updated I expect to be able to apply bootstrap metadata
mutations to an empty schema and have things like materialized views
created (using the regular code paths).
This commit has been re-ordered for conceptual clarity, but it won't
compile because it references the metadata module. It's possible to
make it compile -- the functionality is there in the schema module --
but it's not worth the rebasing effort until after review (and
possibly not even then, since we'll squash down to a single commit to
land).
* Part 2: Maintain entids separately from idents.
In order to support historical idents, we need to distinguish the
"current" map from entid -> ident from the "complete historical" map
ident -> entid. This is what Datomic does; in Datomic, an ident is
never retracted (although it can be replaced). This approach is an
important part of allowing multiple consumers to share a schema
fragment as it migrates forward.
This fixes a limitation of the Clojure implementation, which did not
handle historical idents across knowledge base close and re-open.
The "entids" materialized view is naturally a slice of the "datoms"
table. The "idents" materialized view is a slice of the
"transactions" table. I hope that representing in this way, and
casting the problem in this light, might generalize to future
materialized views.
* Pre: Add DiffSet.
* Part 4: Collect mutations to a `Schema`.
I haven't taken your review comment about consuming AttributeBuilder
during each fluent function. If you read my response and still want
this, I'm happy to do it in review.
* Part 5: Handle :db/ident and :db.{install,alter}/attribute.
This "loops" the committed datoms out of the SQL store and back
through the metadata (schema, but in future also partition map)
processor. The metadata processor updates the schema and produces a
report of what changed; that report is then used to update the SQL
store. That update includes:
- the materialized views ("entids", "idents", and "schema");
- if needed, a subset of the datoms themselves (as flags change).
I've left a TODO for handling attribute retraction in the cases that
it makes sense. I expect that to be straight-forward.
* Review comment: Rename DiffSet to AddRetractAlterSet.
Also adds a little more commentary and a simple test.
* Review comment: Use ToIdent trait.
* Review comment: partially revert "Part 2: Maintain entids separately from idents."
This reverts commit 23a91df9c35e14398f2ddbd1ba25315821e67401.
Following our discussion, this removes the "entids" materialized
view. The next commit will remove historical idents from the "idents"
materialized view.
* Post: Use custom Either rather than std::result::Result.
This is not necessary, but it was suggested that we might be paying an
overhead creating Err instances while using error_chain. That seems
not to be the case, but this change shows that we don't actually use
any of the Result helper methods, so there's no reason to overload
Result. This change might avoid some future confusion, so I'm going
to land it anyway.
Signed-off-by: Nick Alexander <nalexander@mozilla.com>
* Review comment: Don't preserve historical idents.
* Review comment: More prepared statements when updating materialized views.
* Post: Test altering :db/cardinality and :db/unique.
These tests fail due to a Datomic limitation, namely that the marker
flag :db.alter/attribute can only be asserted once for an attribute!
That is, [:db.part/db :db.alter/attribute :attribute] will only be
transacted at most once. Since older versions of Datomic required the
:db.alter/attribute flag, I can only imagine they either never wrote
:db.alter/attribute to the store, or they handled it specially. I'll
need to remove the marker flag system from Mentat in order to address
this fundamental limitation.
* Post: Remove some more single quotes from error output.
* Post: Add assert_transact! macro to unwrap safely.
I was finding it very difficult to track unwrapping errors while
making changes, due to an underlying Mac OS X symbolication issue that
makes running tests with RUST_BACKTRACE=1 so slow that they all time
out.
* Post: Don't expect or recognize :db.{install,alter}/attribute.
I had this all working... except we will never see a repeated
`[:db.part/db :db.alter/attribute :attribute]` assertion in the store!
That means my approach would let you alter an attribute at most one
time. It's not worth hacking around this; it's better to just stop
expecting (and recognizing) the marker flags. (We have all the data
to distinguish the various cases that we need without the marker
flags.)
This brings Mentat in line with the thrust of newer Datomic versions,
but isn't compatible with Datomic, because (if I understand correctly)
Datomic automatically adds :db.{install,alter}/attribute assertions to
transactions.
I haven't purged the corresponding :db/ident and schema fragments just
yet:
- we might want them back
- we might want them in order to upgrade v1 and v2 databases to the
new on-disk layout we're fleshing out (v3?).
* Post: Don't make :db/unique :db.unique/* imply :db/index true.
This patch avoids a potential bug with the "schema" materialized view.
If :db/unique :db.unique/value implies :db/index true, then what
happens when you _retract_ :db.unique/value? I think Datomic defines
this in some way, but I really want the "schema" materialized view to
be a slice of "datoms" and not have these sort of ambiguities and
persistent effects. Therefore, to ensure that we don't retract a
schema characteristic and accidentally change more than we intended
to, this patch stops having any schema characteristic imply any other
schema characteristic(s). To achieve that, I added an
Option<Unique::{Value,Identity}> type to Attribute; this helps with
this patch, and also looks ahead to when we allow to retract
:db/unique attributes.
* Post: Allow to retract :db/ident.
* Post: Include more details about invalid schema changes.
The tests use strings, so they hide the chained errors which do in
fact provide more detail.
* Review comment: Fix outdated comment.
* Review comment: s/_SET/_SQL_LIST/.
* Review comment: Use a sub-select for checking cardinality.
This might be faster in practice.
* Review comment: Put `attribute::Unique` into its own namespace.
2017-03-20 20:18:59 +00:00
let bootstrap_schema_for_mutation = Schema ::default ( ) ; // The bootstrap transaction will populate this schema.
2018-02-16 10:01:00 +00:00
2019-07-17 15:46:47 +00:00
let ( _report , next_partition_map , next_schema , _watcher ) = transact (
& tx ,
db . partition_map ,
& bootstrap_schema_for_mutation ,
& db . schema ,
NullWatcher ( ) ,
bootstrap ::bootstrap_entities ( ) ,
) ? ;
2018-03-06 17:01:20 +00:00
Schema alteration. Fixes #294 and #295. (#370) r=rnewman
* Pre: Don't retract :db/ident in test.
Datomic (and eventually Mentat) don't allow to retract :db/ident in
this way, so this runs afoul of future work to support mutating
metadata.
* Pre: s/VALUETYPE/VALUE_TYPE/.
This is consistent with the capitalization (which is "valueType") and
the other identifier.
* Pre: Remove some single quotes from error output.
* Part 1: Make materialized views be uniform [e a v value_type_tag].
This looks ahead to a time when we could support arbitrary
user-defined materialized views. For now, the "idents" materialized
view is those datoms of the form [e :db/ident :namespaced/keyword] and
the "schema" materialized view is those datoms of the form [e a v]
where a is in a particular set of attributes that will become clear in
the following commits.
This change is not backwards compatible, so I'm removing the open
current (really, v2) test. It'll be re-instated when we get to
https://github.com/mozilla/mentat/issues/194.
* Pre: Map TypedValue::Ref to TypedValue::Keyword in debug output.
* Part 3: Separate `schema_to_mutate` from the `schema` used to interpret.
This is just to keep track of the expected changes during
bootstrapping. I want bootstrap metadata mutations to flow through
the same code path as metadata mutations during regular transactions;
by differentiating the schema used for interpretation from the schema
that will be updated I expect to be able to apply bootstrap metadata
mutations to an empty schema and have things like materialized views
created (using the regular code paths).
This commit has been re-ordered for conceptual clarity, but it won't
compile because it references the metadata module. It's possible to
make it compile -- the functionality is there in the schema module --
but it's not worth the rebasing effort until after review (and
possibly not even then, since we'll squash down to a single commit to
land).
* Part 2: Maintain entids separately from idents.
In order to support historical idents, we need to distinguish the
"current" map from entid -> ident from the "complete historical" map
ident -> entid. This is what Datomic does; in Datomic, an ident is
never retracted (although it can be replaced). This approach is an
important part of allowing multiple consumers to share a schema
fragment as it migrates forward.
This fixes a limitation of the Clojure implementation, which did not
handle historical idents across knowledge base close and re-open.
The "entids" materialized view is naturally a slice of the "datoms"
table. The "idents" materialized view is a slice of the
"transactions" table. I hope that representing in this way, and
casting the problem in this light, might generalize to future
materialized views.
* Pre: Add DiffSet.
* Part 4: Collect mutations to a `Schema`.
I haven't taken your review comment about consuming AttributeBuilder
during each fluent function. If you read my response and still want
this, I'm happy to do it in review.
* Part 5: Handle :db/ident and :db.{install,alter}/attribute.
This "loops" the committed datoms out of the SQL store and back
through the metadata (schema, but in future also partition map)
processor. The metadata processor updates the schema and produces a
report of what changed; that report is then used to update the SQL
store. That update includes:
- the materialized views ("entids", "idents", and "schema");
- if needed, a subset of the datoms themselves (as flags change).
I've left a TODO for handling attribute retraction in the cases that
it makes sense. I expect that to be straight-forward.
* Review comment: Rename DiffSet to AddRetractAlterSet.
Also adds a little more commentary and a simple test.
* Review comment: Use ToIdent trait.
* Review comment: partially revert "Part 2: Maintain entids separately from idents."
This reverts commit 23a91df9c35e14398f2ddbd1ba25315821e67401.
Following our discussion, this removes the "entids" materialized
view. The next commit will remove historical idents from the "idents"
materialized view.
* Post: Use custom Either rather than std::result::Result.
This is not necessary, but it was suggested that we might be paying an
overhead creating Err instances while using error_chain. That seems
not to be the case, but this change shows that we don't actually use
any of the Result helper methods, so there's no reason to overload
Result. This change might avoid some future confusion, so I'm going
to land it anyway.
Signed-off-by: Nick Alexander <nalexander@mozilla.com>
* Review comment: Don't preserve historical idents.
* Review comment: More prepared statements when updating materialized views.
* Post: Test altering :db/cardinality and :db/unique.
These tests fail due to a Datomic limitation, namely that the marker
flag :db.alter/attribute can only be asserted once for an attribute!
That is, [:db.part/db :db.alter/attribute :attribute] will only be
transacted at most once. Since older versions of Datomic required the
:db.alter/attribute flag, I can only imagine they either never wrote
:db.alter/attribute to the store, or they handled it specially. I'll
need to remove the marker flag system from Mentat in order to address
this fundamental limitation.
* Post: Remove some more single quotes from error output.
* Post: Add assert_transact! macro to unwrap safely.
I was finding it very difficult to track unwrapping errors while
making changes, due to an underlying Mac OS X symbolication issue that
makes running tests with RUST_BACKTRACE=1 so slow that they all time
out.
* Post: Don't expect or recognize :db.{install,alter}/attribute.
I had this all working... except we will never see a repeated
`[:db.part/db :db.alter/attribute :attribute]` assertion in the store!
That means my approach would let you alter an attribute at most one
time. It's not worth hacking around this; it's better to just stop
expecting (and recognizing) the marker flags. (We have all the data
to distinguish the various cases that we need without the marker
flags.)
This brings Mentat in line with the thrust of newer Datomic versions,
but isn't compatible with Datomic, because (if I understand correctly)
Datomic automatically adds :db.{install,alter}/attribute assertions to
transactions.
I haven't purged the corresponding :db/ident and schema fragments just
yet:
- we might want them back
- we might want them in order to upgrade v1 and v2 databases to the
new on-disk layout we're fleshing out (v3?).
* Post: Don't make :db/unique :db.unique/* imply :db/index true.
This patch avoids a potential bug with the "schema" materialized view.
If :db/unique :db.unique/value implies :db/index true, then what
happens when you _retract_ :db.unique/value? I think Datomic defines
this in some way, but I really want the "schema" materialized view to
be a slice of "datoms" and not have these sort of ambiguities and
persistent effects. Therefore, to ensure that we don't retract a
schema characteristic and accidentally change more than we intended
to, this patch stops having any schema characteristic imply any other
schema characteristic(s). To achieve that, I added an
Option<Unique::{Value,Identity}> type to Attribute; this helps with
this patch, and also looks ahead to when we allow to retract
:db/unique attributes.
* Post: Allow to retract :db/ident.
* Post: Include more details about invalid schema changes.
The tests use strings, so they hide the chained errors which do in
fact provide more detail.
* Review comment: Fix outdated comment.
* Review comment: s/_SET/_SQL_LIST/.
* Review comment: Use a sub-select for checking cardinality.
This might be faster in practice.
* Review comment: Put `attribute::Unique` into its own namespace.
2017-03-20 20:18:59 +00:00
// TODO: validate metadata mutations that aren't schema related, like additional partitions.
if let Some ( next_schema ) = next_schema {
2018-02-16 10:01:00 +00:00
if next_schema ! = db . schema {
2020-01-23 18:16:19 +00:00
bail! ( DbErrorKind ::NotYetImplemented (
2019-07-17 15:46:47 +00:00
" Initial bootstrap transaction did not produce expected bootstrap schema "
2020-01-23 18:16:19 +00:00
. to_string ( )
) ) ;
Schema alteration. Fixes #294 and #295. (#370) r=rnewman
* Pre: Don't retract :db/ident in test.
Datomic (and eventually Mentat) don't allow to retract :db/ident in
this way, so this runs afoul of future work to support mutating
metadata.
* Pre: s/VALUETYPE/VALUE_TYPE/.
This is consistent with the capitalization (which is "valueType") and
the other identifier.
* Pre: Remove some single quotes from error output.
* Part 1: Make materialized views be uniform [e a v value_type_tag].
This looks ahead to a time when we could support arbitrary
user-defined materialized views. For now, the "idents" materialized
view is those datoms of the form [e :db/ident :namespaced/keyword] and
the "schema" materialized view is those datoms of the form [e a v]
where a is in a particular set of attributes that will become clear in
the following commits.
This change is not backwards compatible, so I'm removing the open
current (really, v2) test. It'll be re-instated when we get to
https://github.com/mozilla/mentat/issues/194.
* Pre: Map TypedValue::Ref to TypedValue::Keyword in debug output.
* Part 3: Separate `schema_to_mutate` from the `schema` used to interpret.
This is just to keep track of the expected changes during
bootstrapping. I want bootstrap metadata mutations to flow through
the same code path as metadata mutations during regular transactions;
by differentiating the schema used for interpretation from the schema
that will be updated I expect to be able to apply bootstrap metadata
mutations to an empty schema and have things like materialized views
created (using the regular code paths).
This commit has been re-ordered for conceptual clarity, but it won't
compile because it references the metadata module. It's possible to
make it compile -- the functionality is there in the schema module --
but it's not worth the rebasing effort until after review (and
possibly not even then, since we'll squash down to a single commit to
land).
* Part 2: Maintain entids separately from idents.
In order to support historical idents, we need to distinguish the
"current" map from entid -> ident from the "complete historical" map
ident -> entid. This is what Datomic does; in Datomic, an ident is
never retracted (although it can be replaced). This approach is an
important part of allowing multiple consumers to share a schema
fragment as it migrates forward.
This fixes a limitation of the Clojure implementation, which did not
handle historical idents across knowledge base close and re-open.
The "entids" materialized view is naturally a slice of the "datoms"
table. The "idents" materialized view is a slice of the
"transactions" table. I hope that representing in this way, and
casting the problem in this light, might generalize to future
materialized views.
* Pre: Add DiffSet.
* Part 4: Collect mutations to a `Schema`.
I haven't taken your review comment about consuming AttributeBuilder
during each fluent function. If you read my response and still want
this, I'm happy to do it in review.
* Part 5: Handle :db/ident and :db.{install,alter}/attribute.
This "loops" the committed datoms out of the SQL store and back
through the metadata (schema, but in future also partition map)
processor. The metadata processor updates the schema and produces a
report of what changed; that report is then used to update the SQL
store. That update includes:
- the materialized views ("entids", "idents", and "schema");
- if needed, a subset of the datoms themselves (as flags change).
I've left a TODO for handling attribute retraction in the cases that
it makes sense. I expect that to be straight-forward.
* Review comment: Rename DiffSet to AddRetractAlterSet.
Also adds a little more commentary and a simple test.
* Review comment: Use ToIdent trait.
* Review comment: partially revert "Part 2: Maintain entids separately from idents."
This reverts commit 23a91df9c35e14398f2ddbd1ba25315821e67401.
Following our discussion, this removes the "entids" materialized
view. The next commit will remove historical idents from the "idents"
materialized view.
* Post: Use custom Either rather than std::result::Result.
This is not necessary, but it was suggested that we might be paying an
overhead creating Err instances while using error_chain. That seems
not to be the case, but this change shows that we don't actually use
any of the Result helper methods, so there's no reason to overload
Result. This change might avoid some future confusion, so I'm going
to land it anyway.
Signed-off-by: Nick Alexander <nalexander@mozilla.com>
* Review comment: Don't preserve historical idents.
* Review comment: More prepared statements when updating materialized views.
* Post: Test altering :db/cardinality and :db/unique.
These tests fail due to a Datomic limitation, namely that the marker
flag :db.alter/attribute can only be asserted once for an attribute!
That is, [:db.part/db :db.alter/attribute :attribute] will only be
transacted at most once. Since older versions of Datomic required the
:db.alter/attribute flag, I can only imagine they either never wrote
:db.alter/attribute to the store, or they handled it specially. I'll
need to remove the marker flag system from Mentat in order to address
this fundamental limitation.
* Post: Remove some more single quotes from error output.
* Post: Add assert_transact! macro to unwrap safely.
I was finding it very difficult to track unwrapping errors while
making changes, due to an underlying Mac OS X symbolication issue that
makes running tests with RUST_BACKTRACE=1 so slow that they all time
out.
* Post: Don't expect or recognize :db.{install,alter}/attribute.
I had this all working... except we will never see a repeated
`[:db.part/db :db.alter/attribute :attribute]` assertion in the store!
That means my approach would let you alter an attribute at most one
time. It's not worth hacking around this; it's better to just stop
expecting (and recognizing) the marker flags. (We have all the data
to distinguish the various cases that we need without the marker
flags.)
This brings Mentat in line with the thrust of newer Datomic versions,
but isn't compatible with Datomic, because (if I understand correctly)
Datomic automatically adds :db.{install,alter}/attribute assertions to
transactions.
I haven't purged the corresponding :db/ident and schema fragments just
yet:
- we might want them back
- we might want them in order to upgrade v1 and v2 databases to the
new on-disk layout we're fleshing out (v3?).
* Post: Don't make :db/unique :db.unique/* imply :db/index true.
This patch avoids a potential bug with the "schema" materialized view.
If :db/unique :db.unique/value implies :db/index true, then what
happens when you _retract_ :db.unique/value? I think Datomic defines
this in some way, but I really want the "schema" materialized view to
be a slice of "datoms" and not have these sort of ambiguities and
persistent effects. Therefore, to ensure that we don't retract a
schema characteristic and accidentally change more than we intended
to, this patch stops having any schema characteristic imply any other
schema characteristic(s). To achieve that, I added an
Option<Unique::{Value,Identity}> type to Attribute; this helps with
this patch, and also looks ahead to when we allow to retract
:db/unique attributes.
* Post: Allow to retract :db/ident.
* Post: Include more details about invalid schema changes.
The tests use strings, so they hide the chained errors which do in
fact provide more detail.
* Review comment: Fix outdated comment.
* Review comment: s/_SET/_SQL_LIST/.
* Review comment: Use a sub-select for checking cardinality.
This might be faster in practice.
* Review comment: Put `attribute::Unique` into its own namespace.
2017-03-20 20:18:59 +00:00
}
Extract partial storage abstraction; use error-chain throughout. Fixes #328. r=rnewman (#341)
* Pre: Drop unneeded tx0 from search results.
* Pre: Don't require a schema in some of the DB code.
The idea is to separate the transaction applying code, which is
schema-aware, from the concrete storage code, which is just concerned
with getting bits onto disk.
* Pre: Only reference Schema, not DB, in debug module.
This is part of a larger separation of the volatile PartitionMap,
which is modified every transaction, from the stable Schema, which is
infrequently modified.
* Pre: Fix indentation.
* Extract part of DB to new SchemaTypeChecking trait.
* Extract part of DB to new PartitionMapping trait.
* Pre: Don't expect :db.part/tx partition to advance when tx fails.
This fails right now, because we allocate tx IDs even when we shouldn't.
* Sketch a db interface without DB.
* Add ValueParseError; use error-chain in tx-parser.
This can be simplified when
https://github.com/Marwes/combine/issues/86 makes it to a published
release, but this unblocks us for now. This converts the `combine`
error type `ParseError<&'a [edn::Value]>` to a type with owned
`Vec<edn::Value>` collections, re-using `edn::Value::Vector` for
making them `Display`.
* Pre: Accept Borrow<Schema> instead of just &Schema in debug module.
This makes it easy to use Rc<Schema> or Arc<Schema> without inserting
&* sigils throughout the code.
* Use error-chain in query-parser.
There are a few things to point out here:
- the fine grained error types have been flattened into one crate-wide
error type; it's pretty easy to regain the granularity as needed.
- edn::ParseError is automatically lifted to
mentat_query_parser::errors::Error;
- we use mentat_parser_utils::ValueParser to maintain parsing error
information from `combine`.
* Patch up top-level.
* Review comment: Only `borrow()` once.
2017-02-24 23:32:41 +00:00
}
2017-01-26 00:13:56 +00:00
// TODO: use the drop semantics to do this automagically?
tx . commit ( ) ? ;
Extract partial storage abstraction; use error-chain throughout. Fixes #328. r=rnewman (#341)
* Pre: Drop unneeded tx0 from search results.
* Pre: Don't require a schema in some of the DB code.
The idea is to separate the transaction applying code, which is
schema-aware, from the concrete storage code, which is just concerned
with getting bits onto disk.
* Pre: Only reference Schema, not DB, in debug module.
This is part of a larger separation of the volatile PartitionMap,
which is modified every transaction, from the stable Schema, which is
infrequently modified.
* Pre: Fix indentation.
* Extract part of DB to new SchemaTypeChecking trait.
* Extract part of DB to new PartitionMapping trait.
* Pre: Don't expect :db.part/tx partition to advance when tx fails.
This fails right now, because we allocate tx IDs even when we shouldn't.
* Sketch a db interface without DB.
* Add ValueParseError; use error-chain in tx-parser.
This can be simplified when
https://github.com/Marwes/combine/issues/86 makes it to a published
release, but this unblocks us for now. This converts the `combine`
error type `ParseError<&'a [edn::Value]>` to a type with owned
`Vec<edn::Value>` collections, re-using `edn::Value::Vector` for
making them `Display`.
* Pre: Accept Borrow<Schema> instead of just &Schema in debug module.
This makes it easy to use Rc<Schema> or Arc<Schema> without inserting
&* sigils throughout the code.
* Use error-chain in query-parser.
There are a few things to point out here:
- the fine grained error types have been flattened into one crate-wide
error type; it's pretty easy to regain the granularity as needed.
- edn::ParseError is automatically lifted to
mentat_query_parser::errors::Error;
- we use mentat_parser_utils::ValueParser to maintain parsing error
information from `combine`.
* Patch up top-level.
* Review comment: Only `borrow()` once.
2017-02-24 23:32:41 +00:00
2018-02-16 10:01:00 +00:00
db . partition_map = next_partition_map ;
Ok ( db )
2017-01-26 00:13:56 +00:00
}
2017-02-15 00:50:40 +00:00
pub fn ensure_current_version ( conn : & mut rusqlite ::Connection ) -> Result < DB > {
2017-03-21 16:50:31 +00:00
if rusqlite ::version_number ( ) < MIN_SQLITE_VERSION {
panic! ( " Mentat requires at least sqlite {} " , MIN_SQLITE_VERSION ) ;
}
2017-01-26 00:13:56 +00:00
let user_version = get_user_version ( & conn ) ? ;
match user_version {
2019-07-17 15:46:47 +00:00
0 = > create_current_version ( conn ) ,
2017-05-09 16:42:35 +00:00
CURRENT_VERSION = > read_db ( conn ) ,
// TODO: support updating an existing store.
2019-07-17 15:46:47 +00:00
v = > bail! ( DbErrorKind ::NotYetImplemented ( format! (
" Opening databases with Mentat version: {} " ,
v
) ) ) ,
2017-01-26 00:13:56 +00:00
}
}
2017-02-08 21:59:56 +00:00
pub trait TypedSQLValue {
2019-07-17 15:46:47 +00:00
fn from_sql_value_pair (
value : rusqlite ::types ::Value ,
value_type_tag : i32 ,
) -> Result < TypedValue > ;
2020-01-23 18:16:19 +00:00
fn to_sql_value_pair ( & self ) -> ( ToSqlOutput , i32 ) ;
2017-02-08 21:59:56 +00:00
fn from_edn_value ( value : & Value ) -> Option < TypedValue > ;
fn to_edn_value_pair ( & self ) -> ( Value , ValueType ) ;
}
impl TypedSQLValue for TypedValue {
2017-01-26 00:13:56 +00:00
/// Given a SQLite `value` and a `value_type_tag`, return the corresponding `TypedValue`.
2019-07-17 15:46:47 +00:00
fn from_sql_value_pair (
value : rusqlite ::types ::Value ,
value_type_tag : i32 ,
) -> Result < TypedValue > {
2017-02-08 22:04:32 +00:00
match ( value_type_tag , value ) {
2017-01-26 00:13:56 +00:00
( 0 , rusqlite ::types ::Value ::Integer ( x ) ) = > Ok ( TypedValue ::Ref ( x ) ) ,
( 1 , rusqlite ::types ::Value ::Integer ( x ) ) = > Ok ( TypedValue ::Boolean ( 0 ! = x ) ) ,
2017-04-29 03:11:55 +00:00
// Negative integers are simply times before 1970.
2019-07-17 15:46:47 +00:00
( 4 , rusqlite ::types ::Value ::Integer ( x ) ) = > {
Ok ( TypedValue ::Instant ( DateTime ::< Utc > ::from_micros ( x ) ) )
}
2017-04-29 03:11:55 +00:00
2017-01-26 00:13:56 +00:00
// SQLite distinguishes integral from decimal types, allowing long and double to
// share a tag.
( 5 , rusqlite ::types ::Value ::Integer ( x ) ) = > Ok ( TypedValue ::Long ( x ) ) ,
( 5 , rusqlite ::types ::Value ::Real ( x ) ) = > Ok ( TypedValue ::Double ( x . into ( ) ) ) ,
2018-04-25 21:23:27 +00:00
( 10 , rusqlite ::types ::Value ::Text ( x ) ) = > Ok ( x . into ( ) ) ,
2017-04-29 03:11:55 +00:00
( 11 , rusqlite ::types ::Value ::Blob ( x ) ) = > {
2020-01-14 15:46:21 +00:00
let u = Uuid ::from_bytes ( x . as_slice ( ) . try_into ( ) . unwrap ( ) ) ;
if u . is_nil ( ) {
2017-04-29 03:11:55 +00:00
// Rather than exposing Uuid's ParseError…
2019-07-17 15:46:47 +00:00
bail! ( DbErrorKind ::BadSQLValuePair (
rusqlite ::types ::Value ::Blob ( x ) ,
value_type_tag
) ) ;
2017-04-29 03:11:55 +00:00
}
2020-01-14 15:46:21 +00:00
Ok ( TypedValue ::Uuid ( u ) )
2019-07-17 15:46:47 +00:00
}
( 13 , rusqlite ::types ::Value ::Text ( x ) ) = > to_namespaced_keyword ( & x ) . map ( | k | k . into ( ) ) ,
2018-06-27 00:17:01 +00:00
( _ , value ) = > bail! ( DbErrorKind ::BadSQLValuePair ( value , value_type_tag ) ) ,
2017-01-26 00:13:56 +00:00
}
}
/// Given an EDN `value`, return a corresponding Mentat `TypedValue`.
///
/// An EDN `Value` does not encode a unique Mentat `ValueType`, so the composition
/// `from_edn_value(first(to_edn_value_pair(...)))` loses information. Additionally, there are
/// EDN values which are not Mentat typed values.
///
/// This function is deterministic.
2017-02-08 21:59:56 +00:00
fn from_edn_value ( value : & Value ) -> Option < TypedValue > {
2017-01-26 00:13:56 +00:00
match value {
2020-01-23 18:16:19 +00:00
Value ::Boolean ( x ) = > Some ( TypedValue ::Boolean ( * x ) ) ,
Value ::Instant ( x ) = > Some ( TypedValue ::Instant ( * x ) ) ,
Value ::Integer ( x ) = > Some ( TypedValue ::Long ( * x ) ) ,
Value ::Uuid ( x ) = > Some ( TypedValue ::Uuid ( * x ) ) ,
Value ::Float ( ref x ) = > Some ( TypedValue ::Double ( * x ) ) ,
Value ::Text ( ref x ) = > Some ( x . clone ( ) . into ( ) ) ,
Value ::Keyword ( ref x ) = > Some ( x . clone ( ) . into ( ) ) ,
2019-07-17 15:46:47 +00:00
_ = > None ,
2017-01-26 00:13:56 +00:00
}
}
/// Return the corresponding SQLite `value` and `value_type_tag` pair.
2020-01-23 18:16:19 +00:00
fn to_sql_value_pair ( & self ) -> ( ToSqlOutput , i32 ) {
2017-01-26 00:13:56 +00:00
match self {
2020-01-23 18:16:19 +00:00
TypedValue ::Ref ( x ) = > ( ( * x ) . into ( ) , 0 ) ,
TypedValue ::Boolean ( x ) = > ( ( if * x { 1 } else { 0 } ) . into ( ) , 1 ) ,
TypedValue ::Instant ( x ) = > ( x . to_micros ( ) . into ( ) , 4 ) ,
2017-01-26 00:13:56 +00:00
// SQLite distinguishes integral from decimal types, allowing long and double to share a tag.
2020-01-23 18:16:19 +00:00
TypedValue ::Long ( x ) = > ( ( * x ) . into ( ) , 5 ) ,
TypedValue ::Double ( x ) = > ( x . into_inner ( ) . into ( ) , 5 ) ,
TypedValue ::String ( ref x ) = > ( x . as_str ( ) . into ( ) , 10 ) ,
TypedValue ::Uuid ( ref u ) = > ( u . as_bytes ( ) . to_vec ( ) . into ( ) , 11 ) ,
TypedValue ::Keyword ( ref x ) = > ( x . to_string ( ) . into ( ) , 13 ) ,
2017-01-26 00:13:56 +00:00
}
}
/// Return the corresponding EDN `value` and `value_type` pair.
2017-02-08 21:59:56 +00:00
fn to_edn_value_pair ( & self ) -> ( Value , ValueType ) {
2017-01-26 00:13:56 +00:00
match self {
2020-01-23 18:16:19 +00:00
TypedValue ::Ref ( x ) = > ( Value ::Integer ( * x ) , ValueType ::Ref ) ,
TypedValue ::Boolean ( x ) = > ( Value ::Boolean ( * x ) , ValueType ::Boolean ) ,
TypedValue ::Instant ( x ) = > ( Value ::Instant ( * x ) , ValueType ::Instant ) ,
TypedValue ::Long ( x ) = > ( Value ::Integer ( * x ) , ValueType ::Long ) ,
TypedValue ::Double ( x ) = > ( Value ::Float ( * x ) , ValueType ::Double ) ,
TypedValue ::String ( ref x ) = > ( Value ::Text ( x . as_ref ( ) . clone ( ) ) , ValueType ::String ) ,
TypedValue ::Uuid ( ref u ) = > ( Value ::Uuid ( * u ) , ValueType ::Uuid ) ,
TypedValue ::Keyword ( ref x ) = > ( Value ::Keyword ( x . as_ref ( ) . clone ( ) ) , ValueType ::Keyword ) ,
2017-01-26 00:13:56 +00:00
}
}
}
Schema alteration. Fixes #294 and #295. (#370) r=rnewman
* Pre: Don't retract :db/ident in test.
Datomic (and eventually Mentat) don't allow to retract :db/ident in
this way, so this runs afoul of future work to support mutating
metadata.
* Pre: s/VALUETYPE/VALUE_TYPE/.
This is consistent with the capitalization (which is "valueType") and
the other identifier.
* Pre: Remove some single quotes from error output.
* Part 1: Make materialized views be uniform [e a v value_type_tag].
This looks ahead to a time when we could support arbitrary
user-defined materialized views. For now, the "idents" materialized
view is those datoms of the form [e :db/ident :namespaced/keyword] and
the "schema" materialized view is those datoms of the form [e a v]
where a is in a particular set of attributes that will become clear in
the following commits.
This change is not backwards compatible, so I'm removing the open
current (really, v2) test. It'll be re-instated when we get to
https://github.com/mozilla/mentat/issues/194.
* Pre: Map TypedValue::Ref to TypedValue::Keyword in debug output.
* Part 3: Separate `schema_to_mutate` from the `schema` used to interpret.
This is just to keep track of the expected changes during
bootstrapping. I want bootstrap metadata mutations to flow through
the same code path as metadata mutations during regular transactions;
by differentiating the schema used for interpretation from the schema
that will be updated I expect to be able to apply bootstrap metadata
mutations to an empty schema and have things like materialized views
created (using the regular code paths).
This commit has been re-ordered for conceptual clarity, but it won't
compile because it references the metadata module. It's possible to
make it compile -- the functionality is there in the schema module --
but it's not worth the rebasing effort until after review (and
possibly not even then, since we'll squash down to a single commit to
land).
* Part 2: Maintain entids separately from idents.
In order to support historical idents, we need to distinguish the
"current" map from entid -> ident from the "complete historical" map
ident -> entid. This is what Datomic does; in Datomic, an ident is
never retracted (although it can be replaced). This approach is an
important part of allowing multiple consumers to share a schema
fragment as it migrates forward.
This fixes a limitation of the Clojure implementation, which did not
handle historical idents across knowledge base close and re-open.
The "entids" materialized view is naturally a slice of the "datoms"
table. The "idents" materialized view is a slice of the
"transactions" table. I hope that representing in this way, and
casting the problem in this light, might generalize to future
materialized views.
* Pre: Add DiffSet.
* Part 4: Collect mutations to a `Schema`.
I haven't taken your review comment about consuming AttributeBuilder
during each fluent function. If you read my response and still want
this, I'm happy to do it in review.
* Part 5: Handle :db/ident and :db.{install,alter}/attribute.
This "loops" the committed datoms out of the SQL store and back
through the metadata (schema, but in future also partition map)
processor. The metadata processor updates the schema and produces a
report of what changed; that report is then used to update the SQL
store. That update includes:
- the materialized views ("entids", "idents", and "schema");
- if needed, a subset of the datoms themselves (as flags change).
I've left a TODO for handling attribute retraction in the cases that
it makes sense. I expect that to be straight-forward.
* Review comment: Rename DiffSet to AddRetractAlterSet.
Also adds a little more commentary and a simple test.
* Review comment: Use ToIdent trait.
* Review comment: partially revert "Part 2: Maintain entids separately from idents."
This reverts commit 23a91df9c35e14398f2ddbd1ba25315821e67401.
Following our discussion, this removes the "entids" materialized
view. The next commit will remove historical idents from the "idents"
materialized view.
* Post: Use custom Either rather than std::result::Result.
This is not necessary, but it was suggested that we might be paying an
overhead creating Err instances while using error_chain. That seems
not to be the case, but this change shows that we don't actually use
any of the Result helper methods, so there's no reason to overload
Result. This change might avoid some future confusion, so I'm going
to land it anyway.
Signed-off-by: Nick Alexander <nalexander@mozilla.com>
* Review comment: Don't preserve historical idents.
* Review comment: More prepared statements when updating materialized views.
* Post: Test altering :db/cardinality and :db/unique.
These tests fail due to a Datomic limitation, namely that the marker
flag :db.alter/attribute can only be asserted once for an attribute!
That is, [:db.part/db :db.alter/attribute :attribute] will only be
transacted at most once. Since older versions of Datomic required the
:db.alter/attribute flag, I can only imagine they either never wrote
:db.alter/attribute to the store, or they handled it specially. I'll
need to remove the marker flag system from Mentat in order to address
this fundamental limitation.
* Post: Remove some more single quotes from error output.
* Post: Add assert_transact! macro to unwrap safely.
I was finding it very difficult to track unwrapping errors while
making changes, due to an underlying Mac OS X symbolication issue that
makes running tests with RUST_BACKTRACE=1 so slow that they all time
out.
* Post: Don't expect or recognize :db.{install,alter}/attribute.
I had this all working... except we will never see a repeated
`[:db.part/db :db.alter/attribute :attribute]` assertion in the store!
That means my approach would let you alter an attribute at most one
time. It's not worth hacking around this; it's better to just stop
expecting (and recognizing) the marker flags. (We have all the data
to distinguish the various cases that we need without the marker
flags.)
This brings Mentat in line with the thrust of newer Datomic versions,
but isn't compatible with Datomic, because (if I understand correctly)
Datomic automatically adds :db.{install,alter}/attribute assertions to
transactions.
I haven't purged the corresponding :db/ident and schema fragments just
yet:
- we might want them back
- we might want them in order to upgrade v1 and v2 databases to the
new on-disk layout we're fleshing out (v3?).
* Post: Don't make :db/unique :db.unique/* imply :db/index true.
This patch avoids a potential bug with the "schema" materialized view.
If :db/unique :db.unique/value implies :db/index true, then what
happens when you _retract_ :db.unique/value? I think Datomic defines
this in some way, but I really want the "schema" materialized view to
be a slice of "datoms" and not have these sort of ambiguities and
persistent effects. Therefore, to ensure that we don't retract a
schema characteristic and accidentally change more than we intended
to, this patch stops having any schema characteristic imply any other
schema characteristic(s). To achieve that, I added an
Option<Unique::{Value,Identity}> type to Attribute; this helps with
this patch, and also looks ahead to when we allow to retract
:db/unique attributes.
* Post: Allow to retract :db/ident.
* Post: Include more details about invalid schema changes.
The tests use strings, so they hide the chained errors which do in
fact provide more detail.
* Review comment: Fix outdated comment.
* Review comment: s/_SET/_SQL_LIST/.
* Review comment: Use a sub-select for checking cardinality.
This might be faster in practice.
* Review comment: Put `attribute::Unique` into its own namespace.
2017-03-20 20:18:59 +00:00
/// Read an arbitrary [e a v value_type_tag] materialized view from the given table in the SQL
/// store.
2019-07-17 15:46:47 +00:00
pub ( crate ) fn read_materialized_view (
conn : & rusqlite ::Connection ,
table : & str ,
) -> Result < Vec < ( Entid , Entid , TypedValue ) > > {
let mut stmt : rusqlite ::Statement =
conn . prepare ( format! ( " SELECT e, a, v, value_type_tag FROM {} " , table ) . as_str ( ) ) ? ;
2020-01-14 15:46:21 +00:00
let m : Result < Vec < _ > > = stmt
. query_and_then ( rusqlite ::params! [ ] , row_to_datom_assertion ) ?
. collect ( ) ;
2017-01-26 00:13:56 +00:00
m
}
/// Read the partition map materialized view from the given SQL store.
2018-07-25 18:30:21 +00:00
pub fn read_partition_map ( conn : & rusqlite ::Connection ) -> Result < PartitionMap > {
// An obviously expensive query, but we use it infrequently:
// - on first start,
// - while moving timelines,
// - during sync.
2018-07-18 00:54:13 +00:00
// First part of the union sprinkles 'allow_excision' into the 'parts' view.
// Second part of the union takes care of partitions which are known
// but don't have any transactions.
2020-01-23 18:16:19 +00:00
conn . prepare (
2019-07-17 15:46:47 +00:00
"
2018-07-18 00:54:13 +00:00
SELECT
known_parts . part ,
known_parts . start ,
known_parts . end ,
parts . idx ,
known_parts . allow_excision
FROM
parts
INNER JOIN
known_parts
ON parts . part = known_parts . part
UNION
SELECT
part ,
start ,
end ,
start ,
allow_excision
FROM
known_parts
WHERE
2019-07-17 15:46:47 +00:00
part NOT IN ( SELECT part FROM parts ) " ,
2020-01-23 18:16:19 +00:00
) ?
. query_and_then ( rusqlite ::params! [ ] , | row | -> Result < ( String , Partition ) > {
Ok ( (
row . get ( 0 ) ? ,
Partition ::new ( row . get ( 1 ) ? , row . get ( 2 ) ? , row . get ( 3 ) ? , row . get ( 4 ) ? ) ,
) )
} ) ?
. collect ( )
2017-01-26 00:13:56 +00:00
}
Schema alteration. Fixes #294 and #295. (#370) r=rnewman
* Pre: Don't retract :db/ident in test.
Datomic (and eventually Mentat) don't allow to retract :db/ident in
this way, so this runs afoul of future work to support mutating
metadata.
* Pre: s/VALUETYPE/VALUE_TYPE/.
This is consistent with the capitalization (which is "valueType") and
the other identifier.
* Pre: Remove some single quotes from error output.
* Part 1: Make materialized views be uniform [e a v value_type_tag].
This looks ahead to a time when we could support arbitrary
user-defined materialized views. For now, the "idents" materialized
view is those datoms of the form [e :db/ident :namespaced/keyword] and
the "schema" materialized view is those datoms of the form [e a v]
where a is in a particular set of attributes that will become clear in
the following commits.
This change is not backwards compatible, so I'm removing the open
current (really, v2) test. It'll be re-instated when we get to
https://github.com/mozilla/mentat/issues/194.
* Pre: Map TypedValue::Ref to TypedValue::Keyword in debug output.
* Part 3: Separate `schema_to_mutate` from the `schema` used to interpret.
This is just to keep track of the expected changes during
bootstrapping. I want bootstrap metadata mutations to flow through
the same code path as metadata mutations during regular transactions;
by differentiating the schema used for interpretation from the schema
that will be updated I expect to be able to apply bootstrap metadata
mutations to an empty schema and have things like materialized views
created (using the regular code paths).
This commit has been re-ordered for conceptual clarity, but it won't
compile because it references the metadata module. It's possible to
make it compile -- the functionality is there in the schema module --
but it's not worth the rebasing effort until after review (and
possibly not even then, since we'll squash down to a single commit to
land).
* Part 2: Maintain entids separately from idents.
In order to support historical idents, we need to distinguish the
"current" map from entid -> ident from the "complete historical" map
ident -> entid. This is what Datomic does; in Datomic, an ident is
never retracted (although it can be replaced). This approach is an
important part of allowing multiple consumers to share a schema
fragment as it migrates forward.
This fixes a limitation of the Clojure implementation, which did not
handle historical idents across knowledge base close and re-open.
The "entids" materialized view is naturally a slice of the "datoms"
table. The "idents" materialized view is a slice of the
"transactions" table. I hope that representing in this way, and
casting the problem in this light, might generalize to future
materialized views.
* Pre: Add DiffSet.
* Part 4: Collect mutations to a `Schema`.
I haven't taken your review comment about consuming AttributeBuilder
during each fluent function. If you read my response and still want
this, I'm happy to do it in review.
* Part 5: Handle :db/ident and :db.{install,alter}/attribute.
This "loops" the committed datoms out of the SQL store and back
through the metadata (schema, but in future also partition map)
processor. The metadata processor updates the schema and produces a
report of what changed; that report is then used to update the SQL
store. That update includes:
- the materialized views ("entids", "idents", and "schema");
- if needed, a subset of the datoms themselves (as flags change).
I've left a TODO for handling attribute retraction in the cases that
it makes sense. I expect that to be straight-forward.
* Review comment: Rename DiffSet to AddRetractAlterSet.
Also adds a little more commentary and a simple test.
* Review comment: Use ToIdent trait.
* Review comment: partially revert "Part 2: Maintain entids separately from idents."
This reverts commit 23a91df9c35e14398f2ddbd1ba25315821e67401.
Following our discussion, this removes the "entids" materialized
view. The next commit will remove historical idents from the "idents"
materialized view.
* Post: Use custom Either rather than std::result::Result.
This is not necessary, but it was suggested that we might be paying an
overhead creating Err instances while using error_chain. That seems
not to be the case, but this change shows that we don't actually use
any of the Result helper methods, so there's no reason to overload
Result. This change might avoid some future confusion, so I'm going
to land it anyway.
Signed-off-by: Nick Alexander <nalexander@mozilla.com>
* Review comment: Don't preserve historical idents.
* Review comment: More prepared statements when updating materialized views.
* Post: Test altering :db/cardinality and :db/unique.
These tests fail due to a Datomic limitation, namely that the marker
flag :db.alter/attribute can only be asserted once for an attribute!
That is, [:db.part/db :db.alter/attribute :attribute] will only be
transacted at most once. Since older versions of Datomic required the
:db.alter/attribute flag, I can only imagine they either never wrote
:db.alter/attribute to the store, or they handled it specially. I'll
need to remove the marker flag system from Mentat in order to address
this fundamental limitation.
* Post: Remove some more single quotes from error output.
* Post: Add assert_transact! macro to unwrap safely.
I was finding it very difficult to track unwrapping errors while
making changes, due to an underlying Mac OS X symbolication issue that
makes running tests with RUST_BACKTRACE=1 so slow that they all time
out.
* Post: Don't expect or recognize :db.{install,alter}/attribute.
I had this all working... except we will never see a repeated
`[:db.part/db :db.alter/attribute :attribute]` assertion in the store!
That means my approach would let you alter an attribute at most one
time. It's not worth hacking around this; it's better to just stop
expecting (and recognizing) the marker flags. (We have all the data
to distinguish the various cases that we need without the marker
flags.)
This brings Mentat in line with the thrust of newer Datomic versions,
but isn't compatible with Datomic, because (if I understand correctly)
Datomic automatically adds :db.{install,alter}/attribute assertions to
transactions.
I haven't purged the corresponding :db/ident and schema fragments just
yet:
- we might want them back
- we might want them in order to upgrade v1 and v2 databases to the
new on-disk layout we're fleshing out (v3?).
* Post: Don't make :db/unique :db.unique/* imply :db/index true.
This patch avoids a potential bug with the "schema" materialized view.
If :db/unique :db.unique/value implies :db/index true, then what
happens when you _retract_ :db.unique/value? I think Datomic defines
this in some way, but I really want the "schema" materialized view to
be a slice of "datoms" and not have these sort of ambiguities and
persistent effects. Therefore, to ensure that we don't retract a
schema characteristic and accidentally change more than we intended
to, this patch stops having any schema characteristic imply any other
schema characteristic(s). To achieve that, I added an
Option<Unique::{Value,Identity}> type to Attribute; this helps with
this patch, and also looks ahead to when we allow to retract
:db/unique attributes.
* Post: Allow to retract :db/ident.
* Post: Include more details about invalid schema changes.
The tests use strings, so they hide the chained errors which do in
fact provide more detail.
* Review comment: Fix outdated comment.
* Review comment: s/_SET/_SQL_LIST/.
* Review comment: Use a sub-select for checking cardinality.
This might be faster in practice.
* Review comment: Put `attribute::Unique` into its own namespace.
2017-03-20 20:18:59 +00:00
/// Read the ident map materialized view from the given SQL store.
2018-07-16 20:58:34 +00:00
pub ( crate ) fn read_ident_map ( conn : & rusqlite ::Connection ) -> Result < IdentMap > {
Schema alteration. Fixes #294 and #295. (#370) r=rnewman
* Pre: Don't retract :db/ident in test.
Datomic (and eventually Mentat) don't allow to retract :db/ident in
this way, so this runs afoul of future work to support mutating
metadata.
* Pre: s/VALUETYPE/VALUE_TYPE/.
This is consistent with the capitalization (which is "valueType") and
the other identifier.
* Pre: Remove some single quotes from error output.
* Part 1: Make materialized views be uniform [e a v value_type_tag].
This looks ahead to a time when we could support arbitrary
user-defined materialized views. For now, the "idents" materialized
view is those datoms of the form [e :db/ident :namespaced/keyword] and
the "schema" materialized view is those datoms of the form [e a v]
where a is in a particular set of attributes that will become clear in
the following commits.
This change is not backwards compatible, so I'm removing the open
current (really, v2) test. It'll be re-instated when we get to
https://github.com/mozilla/mentat/issues/194.
* Pre: Map TypedValue::Ref to TypedValue::Keyword in debug output.
* Part 3: Separate `schema_to_mutate` from the `schema` used to interpret.
This is just to keep track of the expected changes during
bootstrapping. I want bootstrap metadata mutations to flow through
the same code path as metadata mutations during regular transactions;
by differentiating the schema used for interpretation from the schema
that will be updated I expect to be able to apply bootstrap metadata
mutations to an empty schema and have things like materialized views
created (using the regular code paths).
This commit has been re-ordered for conceptual clarity, but it won't
compile because it references the metadata module. It's possible to
make it compile -- the functionality is there in the schema module --
but it's not worth the rebasing effort until after review (and
possibly not even then, since we'll squash down to a single commit to
land).
* Part 2: Maintain entids separately from idents.
In order to support historical idents, we need to distinguish the
"current" map from entid -> ident from the "complete historical" map
ident -> entid. This is what Datomic does; in Datomic, an ident is
never retracted (although it can be replaced). This approach is an
important part of allowing multiple consumers to share a schema
fragment as it migrates forward.
This fixes a limitation of the Clojure implementation, which did not
handle historical idents across knowledge base close and re-open.
The "entids" materialized view is naturally a slice of the "datoms"
table. The "idents" materialized view is a slice of the
"transactions" table. I hope that representing in this way, and
casting the problem in this light, might generalize to future
materialized views.
* Pre: Add DiffSet.
* Part 4: Collect mutations to a `Schema`.
I haven't taken your review comment about consuming AttributeBuilder
during each fluent function. If you read my response and still want
this, I'm happy to do it in review.
* Part 5: Handle :db/ident and :db.{install,alter}/attribute.
This "loops" the committed datoms out of the SQL store and back
through the metadata (schema, but in future also partition map)
processor. The metadata processor updates the schema and produces a
report of what changed; that report is then used to update the SQL
store. That update includes:
- the materialized views ("entids", "idents", and "schema");
- if needed, a subset of the datoms themselves (as flags change).
I've left a TODO for handling attribute retraction in the cases that
it makes sense. I expect that to be straight-forward.
* Review comment: Rename DiffSet to AddRetractAlterSet.
Also adds a little more commentary and a simple test.
* Review comment: Use ToIdent trait.
* Review comment: partially revert "Part 2: Maintain entids separately from idents."
This reverts commit 23a91df9c35e14398f2ddbd1ba25315821e67401.
Following our discussion, this removes the "entids" materialized
view. The next commit will remove historical idents from the "idents"
materialized view.
* Post: Use custom Either rather than std::result::Result.
This is not necessary, but it was suggested that we might be paying an
overhead creating Err instances while using error_chain. That seems
not to be the case, but this change shows that we don't actually use
any of the Result helper methods, so there's no reason to overload
Result. This change might avoid some future confusion, so I'm going
to land it anyway.
Signed-off-by: Nick Alexander <nalexander@mozilla.com>
* Review comment: Don't preserve historical idents.
* Review comment: More prepared statements when updating materialized views.
* Post: Test altering :db/cardinality and :db/unique.
These tests fail due to a Datomic limitation, namely that the marker
flag :db.alter/attribute can only be asserted once for an attribute!
That is, [:db.part/db :db.alter/attribute :attribute] will only be
transacted at most once. Since older versions of Datomic required the
:db.alter/attribute flag, I can only imagine they either never wrote
:db.alter/attribute to the store, or they handled it specially. I'll
need to remove the marker flag system from Mentat in order to address
this fundamental limitation.
* Post: Remove some more single quotes from error output.
* Post: Add assert_transact! macro to unwrap safely.
I was finding it very difficult to track unwrapping errors while
making changes, due to an underlying Mac OS X symbolication issue that
makes running tests with RUST_BACKTRACE=1 so slow that they all time
out.
* Post: Don't expect or recognize :db.{install,alter}/attribute.
I had this all working... except we will never see a repeated
`[:db.part/db :db.alter/attribute :attribute]` assertion in the store!
That means my approach would let you alter an attribute at most one
time. It's not worth hacking around this; it's better to just stop
expecting (and recognizing) the marker flags. (We have all the data
to distinguish the various cases that we need without the marker
flags.)
This brings Mentat in line with the thrust of newer Datomic versions,
but isn't compatible with Datomic, because (if I understand correctly)
Datomic automatically adds :db.{install,alter}/attribute assertions to
transactions.
I haven't purged the corresponding :db/ident and schema fragments just
yet:
- we might want them back
- we might want them in order to upgrade v1 and v2 databases to the
new on-disk layout we're fleshing out (v3?).
* Post: Don't make :db/unique :db.unique/* imply :db/index true.
This patch avoids a potential bug with the "schema" materialized view.
If :db/unique :db.unique/value implies :db/index true, then what
happens when you _retract_ :db.unique/value? I think Datomic defines
this in some way, but I really want the "schema" materialized view to
be a slice of "datoms" and not have these sort of ambiguities and
persistent effects. Therefore, to ensure that we don't retract a
schema characteristic and accidentally change more than we intended
to, this patch stops having any schema characteristic imply any other
schema characteristic(s). To achieve that, I added an
Option<Unique::{Value,Identity}> type to Attribute; this helps with
this patch, and also looks ahead to when we allow to retract
:db/unique attributes.
* Post: Allow to retract :db/ident.
* Post: Include more details about invalid schema changes.
The tests use strings, so they hide the chained errors which do in
fact provide more detail.
* Review comment: Fix outdated comment.
* Review comment: s/_SET/_SQL_LIST/.
* Review comment: Use a sub-select for checking cardinality.
This might be faster in practice.
* Review comment: Put `attribute::Unique` into its own namespace.
2017-03-20 20:18:59 +00:00
let v = read_materialized_view ( conn , " idents " ) ? ;
v . into_iter ( ) . map ( | ( e , a , typed_value ) | {
if a ! = entids ::DB_IDENT {
2018-06-27 00:17:01 +00:00
bail! ( DbErrorKind ::NotYetImplemented ( format! ( " bad idents materialized view: expected :db/ident but got {} " , a ) ) ) ;
2017-02-17 21:55:36 +00:00
}
Schema alteration. Fixes #294 and #295. (#370) r=rnewman
* Pre: Don't retract :db/ident in test.
Datomic (and eventually Mentat) don't allow to retract :db/ident in
this way, so this runs afoul of future work to support mutating
metadata.
* Pre: s/VALUETYPE/VALUE_TYPE/.
This is consistent with the capitalization (which is "valueType") and
the other identifier.
* Pre: Remove some single quotes from error output.
* Part 1: Make materialized views be uniform [e a v value_type_tag].
This looks ahead to a time when we could support arbitrary
user-defined materialized views. For now, the "idents" materialized
view is those datoms of the form [e :db/ident :namespaced/keyword] and
the "schema" materialized view is those datoms of the form [e a v]
where a is in a particular set of attributes that will become clear in
the following commits.
This change is not backwards compatible, so I'm removing the open
current (really, v2) test. It'll be re-instated when we get to
https://github.com/mozilla/mentat/issues/194.
* Pre: Map TypedValue::Ref to TypedValue::Keyword in debug output.
* Part 3: Separate `schema_to_mutate` from the `schema` used to interpret.
This is just to keep track of the expected changes during
bootstrapping. I want bootstrap metadata mutations to flow through
the same code path as metadata mutations during regular transactions;
by differentiating the schema used for interpretation from the schema
that will be updated I expect to be able to apply bootstrap metadata
mutations to an empty schema and have things like materialized views
created (using the regular code paths).
This commit has been re-ordered for conceptual clarity, but it won't
compile because it references the metadata module. It's possible to
make it compile -- the functionality is there in the schema module --
but it's not worth the rebasing effort until after review (and
possibly not even then, since we'll squash down to a single commit to
land).
* Part 2: Maintain entids separately from idents.
In order to support historical idents, we need to distinguish the
"current" map from entid -> ident from the "complete historical" map
ident -> entid. This is what Datomic does; in Datomic, an ident is
never retracted (although it can be replaced). This approach is an
important part of allowing multiple consumers to share a schema
fragment as it migrates forward.
This fixes a limitation of the Clojure implementation, which did not
handle historical idents across knowledge base close and re-open.
The "entids" materialized view is naturally a slice of the "datoms"
table. The "idents" materialized view is a slice of the
"transactions" table. I hope that representing in this way, and
casting the problem in this light, might generalize to future
materialized views.
* Pre: Add DiffSet.
* Part 4: Collect mutations to a `Schema`.
I haven't taken your review comment about consuming AttributeBuilder
during each fluent function. If you read my response and still want
this, I'm happy to do it in review.
* Part 5: Handle :db/ident and :db.{install,alter}/attribute.
This "loops" the committed datoms out of the SQL store and back
through the metadata (schema, but in future also partition map)
processor. The metadata processor updates the schema and produces a
report of what changed; that report is then used to update the SQL
store. That update includes:
- the materialized views ("entids", "idents", and "schema");
- if needed, a subset of the datoms themselves (as flags change).
I've left a TODO for handling attribute retraction in the cases that
it makes sense. I expect that to be straight-forward.
* Review comment: Rename DiffSet to AddRetractAlterSet.
Also adds a little more commentary and a simple test.
* Review comment: Use ToIdent trait.
* Review comment: partially revert "Part 2: Maintain entids separately from idents."
This reverts commit 23a91df9c35e14398f2ddbd1ba25315821e67401.
Following our discussion, this removes the "entids" materialized
view. The next commit will remove historical idents from the "idents"
materialized view.
* Post: Use custom Either rather than std::result::Result.
This is not necessary, but it was suggested that we might be paying an
overhead creating Err instances while using error_chain. That seems
not to be the case, but this change shows that we don't actually use
any of the Result helper methods, so there's no reason to overload
Result. This change might avoid some future confusion, so I'm going
to land it anyway.
Signed-off-by: Nick Alexander <nalexander@mozilla.com>
* Review comment: Don't preserve historical idents.
* Review comment: More prepared statements when updating materialized views.
* Post: Test altering :db/cardinality and :db/unique.
These tests fail due to a Datomic limitation, namely that the marker
flag :db.alter/attribute can only be asserted once for an attribute!
That is, [:db.part/db :db.alter/attribute :attribute] will only be
transacted at most once. Since older versions of Datomic required the
:db.alter/attribute flag, I can only imagine they either never wrote
:db.alter/attribute to the store, or they handled it specially. I'll
need to remove the marker flag system from Mentat in order to address
this fundamental limitation.
* Post: Remove some more single quotes from error output.
* Post: Add assert_transact! macro to unwrap safely.
I was finding it very difficult to track unwrapping errors while
making changes, due to an underlying Mac OS X symbolication issue that
makes running tests with RUST_BACKTRACE=1 so slow that they all time
out.
* Post: Don't expect or recognize :db.{install,alter}/attribute.
I had this all working... except we will never see a repeated
`[:db.part/db :db.alter/attribute :attribute]` assertion in the store!
That means my approach would let you alter an attribute at most one
time. It's not worth hacking around this; it's better to just stop
expecting (and recognizing) the marker flags. (We have all the data
to distinguish the various cases that we need without the marker
flags.)
This brings Mentat in line with the thrust of newer Datomic versions,
but isn't compatible with Datomic, because (if I understand correctly)
Datomic automatically adds :db.{install,alter}/attribute assertions to
transactions.
I haven't purged the corresponding :db/ident and schema fragments just
yet:
- we might want them back
- we might want them in order to upgrade v1 and v2 databases to the
new on-disk layout we're fleshing out (v3?).
* Post: Don't make :db/unique :db.unique/* imply :db/index true.
This patch avoids a potential bug with the "schema" materialized view.
If :db/unique :db.unique/value implies :db/index true, then what
happens when you _retract_ :db.unique/value? I think Datomic defines
this in some way, but I really want the "schema" materialized view to
be a slice of "datoms" and not have these sort of ambiguities and
persistent effects. Therefore, to ensure that we don't retract a
schema characteristic and accidentally change more than we intended
to, this patch stops having any schema characteristic imply any other
schema characteristic(s). To achieve that, I added an
Option<Unique::{Value,Identity}> type to Attribute; this helps with
this patch, and also looks ahead to when we allow to retract
:db/unique attributes.
* Post: Allow to retract :db/ident.
* Post: Include more details about invalid schema changes.
The tests use strings, so they hide the chained errors which do in
fact provide more detail.
* Review comment: Fix outdated comment.
* Review comment: s/_SET/_SQL_LIST/.
* Review comment: Use a sub-select for checking cardinality.
This might be faster in practice.
* Review comment: Put `attribute::Unique` into its own namespace.
2017-03-20 20:18:59 +00:00
if let TypedValue ::Keyword ( keyword ) = typed_value {
Use Rc for TypedValue, Variable, and query Ident keywords. (#395) r=nalexander
Part 1, core: use Rc for String and Keyword.
Part 2, query: use Rc for Variable.
Part 3, sql: use Rc for args in SQLiteQueryBuilder.
Part 4, query-algebrizer: use Rc.
Part 5, db: use Rc.
Part 6, query-parser: use Rc.
Part 7, query-projector: use Rc.
Part 8, query-translator: use Rc.
Part 9, top level: use Rc.
Part 10: intern Ident and IdentOrKeyword.
2017-03-29 20:18:17 +00:00
Ok ( ( keyword . as_ref ( ) . clone ( ) , e ) )
Schema alteration. Fixes #294 and #295. (#370) r=rnewman
* Pre: Don't retract :db/ident in test.
Datomic (and eventually Mentat) don't allow to retract :db/ident in
this way, so this runs afoul of future work to support mutating
metadata.
* Pre: s/VALUETYPE/VALUE_TYPE/.
This is consistent with the capitalization (which is "valueType") and
the other identifier.
* Pre: Remove some single quotes from error output.
* Part 1: Make materialized views be uniform [e a v value_type_tag].
This looks ahead to a time when we could support arbitrary
user-defined materialized views. For now, the "idents" materialized
view is those datoms of the form [e :db/ident :namespaced/keyword] and
the "schema" materialized view is those datoms of the form [e a v]
where a is in a particular set of attributes that will become clear in
the following commits.
This change is not backwards compatible, so I'm removing the open
current (really, v2) test. It'll be re-instated when we get to
https://github.com/mozilla/mentat/issues/194.
* Pre: Map TypedValue::Ref to TypedValue::Keyword in debug output.
* Part 3: Separate `schema_to_mutate` from the `schema` used to interpret.
This is just to keep track of the expected changes during
bootstrapping. I want bootstrap metadata mutations to flow through
the same code path as metadata mutations during regular transactions;
by differentiating the schema used for interpretation from the schema
that will be updated I expect to be able to apply bootstrap metadata
mutations to an empty schema and have things like materialized views
created (using the regular code paths).
This commit has been re-ordered for conceptual clarity, but it won't
compile because it references the metadata module. It's possible to
make it compile -- the functionality is there in the schema module --
but it's not worth the rebasing effort until after review (and
possibly not even then, since we'll squash down to a single commit to
land).
* Part 2: Maintain entids separately from idents.
In order to support historical idents, we need to distinguish the
"current" map from entid -> ident from the "complete historical" map
ident -> entid. This is what Datomic does; in Datomic, an ident is
never retracted (although it can be replaced). This approach is an
important part of allowing multiple consumers to share a schema
fragment as it migrates forward.
This fixes a limitation of the Clojure implementation, which did not
handle historical idents across knowledge base close and re-open.
The "entids" materialized view is naturally a slice of the "datoms"
table. The "idents" materialized view is a slice of the
"transactions" table. I hope that representing in this way, and
casting the problem in this light, might generalize to future
materialized views.
* Pre: Add DiffSet.
* Part 4: Collect mutations to a `Schema`.
I haven't taken your review comment about consuming AttributeBuilder
during each fluent function. If you read my response and still want
this, I'm happy to do it in review.
* Part 5: Handle :db/ident and :db.{install,alter}/attribute.
This "loops" the committed datoms out of the SQL store and back
through the metadata (schema, but in future also partition map)
processor. The metadata processor updates the schema and produces a
report of what changed; that report is then used to update the SQL
store. That update includes:
- the materialized views ("entids", "idents", and "schema");
- if needed, a subset of the datoms themselves (as flags change).
I've left a TODO for handling attribute retraction in the cases that
it makes sense. I expect that to be straight-forward.
* Review comment: Rename DiffSet to AddRetractAlterSet.
Also adds a little more commentary and a simple test.
* Review comment: Use ToIdent trait.
* Review comment: partially revert "Part 2: Maintain entids separately from idents."
This reverts commit 23a91df9c35e14398f2ddbd1ba25315821e67401.
Following our discussion, this removes the "entids" materialized
view. The next commit will remove historical idents from the "idents"
materialized view.
* Post: Use custom Either rather than std::result::Result.
This is not necessary, but it was suggested that we might be paying an
overhead creating Err instances while using error_chain. That seems
not to be the case, but this change shows that we don't actually use
any of the Result helper methods, so there's no reason to overload
Result. This change might avoid some future confusion, so I'm going
to land it anyway.
Signed-off-by: Nick Alexander <nalexander@mozilla.com>
* Review comment: Don't preserve historical idents.
* Review comment: More prepared statements when updating materialized views.
* Post: Test altering :db/cardinality and :db/unique.
These tests fail due to a Datomic limitation, namely that the marker
flag :db.alter/attribute can only be asserted once for an attribute!
That is, [:db.part/db :db.alter/attribute :attribute] will only be
transacted at most once. Since older versions of Datomic required the
:db.alter/attribute flag, I can only imagine they either never wrote
:db.alter/attribute to the store, or they handled it specially. I'll
need to remove the marker flag system from Mentat in order to address
this fundamental limitation.
* Post: Remove some more single quotes from error output.
* Post: Add assert_transact! macro to unwrap safely.
I was finding it very difficult to track unwrapping errors while
making changes, due to an underlying Mac OS X symbolication issue that
makes running tests with RUST_BACKTRACE=1 so slow that they all time
out.
* Post: Don't expect or recognize :db.{install,alter}/attribute.
I had this all working... except we will never see a repeated
`[:db.part/db :db.alter/attribute :attribute]` assertion in the store!
That means my approach would let you alter an attribute at most one
time. It's not worth hacking around this; it's better to just stop
expecting (and recognizing) the marker flags. (We have all the data
to distinguish the various cases that we need without the marker
flags.)
This brings Mentat in line with the thrust of newer Datomic versions,
but isn't compatible with Datomic, because (if I understand correctly)
Datomic automatically adds :db.{install,alter}/attribute assertions to
transactions.
I haven't purged the corresponding :db/ident and schema fragments just
yet:
- we might want them back
- we might want them in order to upgrade v1 and v2 databases to the
new on-disk layout we're fleshing out (v3?).
* Post: Don't make :db/unique :db.unique/* imply :db/index true.
This patch avoids a potential bug with the "schema" materialized view.
If :db/unique :db.unique/value implies :db/index true, then what
happens when you _retract_ :db.unique/value? I think Datomic defines
this in some way, but I really want the "schema" materialized view to
be a slice of "datoms" and not have these sort of ambiguities and
persistent effects. Therefore, to ensure that we don't retract a
schema characteristic and accidentally change more than we intended
to, this patch stops having any schema characteristic imply any other
schema characteristic(s). To achieve that, I added an
Option<Unique::{Value,Identity}> type to Attribute; this helps with
this patch, and also looks ahead to when we allow to retract
:db/unique attributes.
* Post: Allow to retract :db/ident.
* Post: Include more details about invalid schema changes.
The tests use strings, so they hide the chained errors which do in
fact provide more detail.
* Review comment: Fix outdated comment.
* Review comment: s/_SET/_SQL_LIST/.
* Review comment: Use a sub-select for checking cardinality.
This might be faster in practice.
* Review comment: Put `attribute::Unique` into its own namespace.
2017-03-20 20:18:59 +00:00
} else {
2018-06-27 00:17:01 +00:00
bail! ( DbErrorKind ::NotYetImplemented ( format! ( " bad idents materialized view: expected [entid :db/ident keyword] but got [entid :db/ident {:?} ] " , typed_value ) ) ) ;
Schema alteration. Fixes #294 and #295. (#370) r=rnewman
* Pre: Don't retract :db/ident in test.
Datomic (and eventually Mentat) don't allow to retract :db/ident in
this way, so this runs afoul of future work to support mutating
metadata.
* Pre: s/VALUETYPE/VALUE_TYPE/.
This is consistent with the capitalization (which is "valueType") and
the other identifier.
* Pre: Remove some single quotes from error output.
* Part 1: Make materialized views be uniform [e a v value_type_tag].
This looks ahead to a time when we could support arbitrary
user-defined materialized views. For now, the "idents" materialized
view is those datoms of the form [e :db/ident :namespaced/keyword] and
the "schema" materialized view is those datoms of the form [e a v]
where a is in a particular set of attributes that will become clear in
the following commits.
This change is not backwards compatible, so I'm removing the open
current (really, v2) test. It'll be re-instated when we get to
https://github.com/mozilla/mentat/issues/194.
* Pre: Map TypedValue::Ref to TypedValue::Keyword in debug output.
* Part 3: Separate `schema_to_mutate` from the `schema` used to interpret.
This is just to keep track of the expected changes during
bootstrapping. I want bootstrap metadata mutations to flow through
the same code path as metadata mutations during regular transactions;
by differentiating the schema used for interpretation from the schema
that will be updated I expect to be able to apply bootstrap metadata
mutations to an empty schema and have things like materialized views
created (using the regular code paths).
This commit has been re-ordered for conceptual clarity, but it won't
compile because it references the metadata module. It's possible to
make it compile -- the functionality is there in the schema module --
but it's not worth the rebasing effort until after review (and
possibly not even then, since we'll squash down to a single commit to
land).
* Part 2: Maintain entids separately from idents.
In order to support historical idents, we need to distinguish the
"current" map from entid -> ident from the "complete historical" map
ident -> entid. This is what Datomic does; in Datomic, an ident is
never retracted (although it can be replaced). This approach is an
important part of allowing multiple consumers to share a schema
fragment as it migrates forward.
This fixes a limitation of the Clojure implementation, which did not
handle historical idents across knowledge base close and re-open.
The "entids" materialized view is naturally a slice of the "datoms"
table. The "idents" materialized view is a slice of the
"transactions" table. I hope that representing in this way, and
casting the problem in this light, might generalize to future
materialized views.
* Pre: Add DiffSet.
* Part 4: Collect mutations to a `Schema`.
I haven't taken your review comment about consuming AttributeBuilder
during each fluent function. If you read my response and still want
this, I'm happy to do it in review.
* Part 5: Handle :db/ident and :db.{install,alter}/attribute.
This "loops" the committed datoms out of the SQL store and back
through the metadata (schema, but in future also partition map)
processor. The metadata processor updates the schema and produces a
report of what changed; that report is then used to update the SQL
store. That update includes:
- the materialized views ("entids", "idents", and "schema");
- if needed, a subset of the datoms themselves (as flags change).
I've left a TODO for handling attribute retraction in the cases that
it makes sense. I expect that to be straight-forward.
* Review comment: Rename DiffSet to AddRetractAlterSet.
Also adds a little more commentary and a simple test.
* Review comment: Use ToIdent trait.
* Review comment: partially revert "Part 2: Maintain entids separately from idents."
This reverts commit 23a91df9c35e14398f2ddbd1ba25315821e67401.
Following our discussion, this removes the "entids" materialized
view. The next commit will remove historical idents from the "idents"
materialized view.
* Post: Use custom Either rather than std::result::Result.
This is not necessary, but it was suggested that we might be paying an
overhead creating Err instances while using error_chain. That seems
not to be the case, but this change shows that we don't actually use
any of the Result helper methods, so there's no reason to overload
Result. This change might avoid some future confusion, so I'm going
to land it anyway.
Signed-off-by: Nick Alexander <nalexander@mozilla.com>
* Review comment: Don't preserve historical idents.
* Review comment: More prepared statements when updating materialized views.
* Post: Test altering :db/cardinality and :db/unique.
These tests fail due to a Datomic limitation, namely that the marker
flag :db.alter/attribute can only be asserted once for an attribute!
That is, [:db.part/db :db.alter/attribute :attribute] will only be
transacted at most once. Since older versions of Datomic required the
:db.alter/attribute flag, I can only imagine they either never wrote
:db.alter/attribute to the store, or they handled it specially. I'll
need to remove the marker flag system from Mentat in order to address
this fundamental limitation.
* Post: Remove some more single quotes from error output.
* Post: Add assert_transact! macro to unwrap safely.
I was finding it very difficult to track unwrapping errors while
making changes, due to an underlying Mac OS X symbolication issue that
makes running tests with RUST_BACKTRACE=1 so slow that they all time
out.
* Post: Don't expect or recognize :db.{install,alter}/attribute.
I had this all working... except we will never see a repeated
`[:db.part/db :db.alter/attribute :attribute]` assertion in the store!
That means my approach would let you alter an attribute at most one
time. It's not worth hacking around this; it's better to just stop
expecting (and recognizing) the marker flags. (We have all the data
to distinguish the various cases that we need without the marker
flags.)
This brings Mentat in line with the thrust of newer Datomic versions,
but isn't compatible with Datomic, because (if I understand correctly)
Datomic automatically adds :db.{install,alter}/attribute assertions to
transactions.
I haven't purged the corresponding :db/ident and schema fragments just
yet:
- we might want them back
- we might want them in order to upgrade v1 and v2 databases to the
new on-disk layout we're fleshing out (v3?).
* Post: Don't make :db/unique :db.unique/* imply :db/index true.
This patch avoids a potential bug with the "schema" materialized view.
If :db/unique :db.unique/value implies :db/index true, then what
happens when you _retract_ :db.unique/value? I think Datomic defines
this in some way, but I really want the "schema" materialized view to
be a slice of "datoms" and not have these sort of ambiguities and
persistent effects. Therefore, to ensure that we don't retract a
schema characteristic and accidentally change more than we intended
to, this patch stops having any schema characteristic imply any other
schema characteristic(s). To achieve that, I added an
Option<Unique::{Value,Identity}> type to Attribute; this helps with
this patch, and also looks ahead to when we allow to retract
:db/unique attributes.
* Post: Allow to retract :db/ident.
* Post: Include more details about invalid schema changes.
The tests use strings, so they hide the chained errors which do in
fact provide more detail.
* Review comment: Fix outdated comment.
* Review comment: s/_SET/_SQL_LIST/.
* Review comment: Use a sub-select for checking cardinality.
This might be faster in practice.
* Review comment: Put `attribute::Unique` into its own namespace.
2017-03-20 20:18:59 +00:00
}
} ) . collect ( )
}
2017-01-26 00:13:56 +00:00
Schema alteration. Fixes #294 and #295. (#370) r=rnewman
* Pre: Don't retract :db/ident in test.
Datomic (and eventually Mentat) don't allow to retract :db/ident in
this way, so this runs afoul of future work to support mutating
metadata.
* Pre: s/VALUETYPE/VALUE_TYPE/.
This is consistent with the capitalization (which is "valueType") and
the other identifier.
* Pre: Remove some single quotes from error output.
* Part 1: Make materialized views be uniform [e a v value_type_tag].
This looks ahead to a time when we could support arbitrary
user-defined materialized views. For now, the "idents" materialized
view is those datoms of the form [e :db/ident :namespaced/keyword] and
the "schema" materialized view is those datoms of the form [e a v]
where a is in a particular set of attributes that will become clear in
the following commits.
This change is not backwards compatible, so I'm removing the open
current (really, v2) test. It'll be re-instated when we get to
https://github.com/mozilla/mentat/issues/194.
* Pre: Map TypedValue::Ref to TypedValue::Keyword in debug output.
* Part 3: Separate `schema_to_mutate` from the `schema` used to interpret.
This is just to keep track of the expected changes during
bootstrapping. I want bootstrap metadata mutations to flow through
the same code path as metadata mutations during regular transactions;
by differentiating the schema used for interpretation from the schema
that will be updated I expect to be able to apply bootstrap metadata
mutations to an empty schema and have things like materialized views
created (using the regular code paths).
This commit has been re-ordered for conceptual clarity, but it won't
compile because it references the metadata module. It's possible to
make it compile -- the functionality is there in the schema module --
but it's not worth the rebasing effort until after review (and
possibly not even then, since we'll squash down to a single commit to
land).
* Part 2: Maintain entids separately from idents.
In order to support historical idents, we need to distinguish the
"current" map from entid -> ident from the "complete historical" map
ident -> entid. This is what Datomic does; in Datomic, an ident is
never retracted (although it can be replaced). This approach is an
important part of allowing multiple consumers to share a schema
fragment as it migrates forward.
This fixes a limitation of the Clojure implementation, which did not
handle historical idents across knowledge base close and re-open.
The "entids" materialized view is naturally a slice of the "datoms"
table. The "idents" materialized view is a slice of the
"transactions" table. I hope that representing in this way, and
casting the problem in this light, might generalize to future
materialized views.
* Pre: Add DiffSet.
* Part 4: Collect mutations to a `Schema`.
I haven't taken your review comment about consuming AttributeBuilder
during each fluent function. If you read my response and still want
this, I'm happy to do it in review.
* Part 5: Handle :db/ident and :db.{install,alter}/attribute.
This "loops" the committed datoms out of the SQL store and back
through the metadata (schema, but in future also partition map)
processor. The metadata processor updates the schema and produces a
report of what changed; that report is then used to update the SQL
store. That update includes:
- the materialized views ("entids", "idents", and "schema");
- if needed, a subset of the datoms themselves (as flags change).
I've left a TODO for handling attribute retraction in the cases that
it makes sense. I expect that to be straight-forward.
* Review comment: Rename DiffSet to AddRetractAlterSet.
Also adds a little more commentary and a simple test.
* Review comment: Use ToIdent trait.
* Review comment: partially revert "Part 2: Maintain entids separately from idents."
This reverts commit 23a91df9c35e14398f2ddbd1ba25315821e67401.
Following our discussion, this removes the "entids" materialized
view. The next commit will remove historical idents from the "idents"
materialized view.
* Post: Use custom Either rather than std::result::Result.
This is not necessary, but it was suggested that we might be paying an
overhead creating Err instances while using error_chain. That seems
not to be the case, but this change shows that we don't actually use
any of the Result helper methods, so there's no reason to overload
Result. This change might avoid some future confusion, so I'm going
to land it anyway.
Signed-off-by: Nick Alexander <nalexander@mozilla.com>
* Review comment: Don't preserve historical idents.
* Review comment: More prepared statements when updating materialized views.
* Post: Test altering :db/cardinality and :db/unique.
These tests fail due to a Datomic limitation, namely that the marker
flag :db.alter/attribute can only be asserted once for an attribute!
That is, [:db.part/db :db.alter/attribute :attribute] will only be
transacted at most once. Since older versions of Datomic required the
:db.alter/attribute flag, I can only imagine they either never wrote
:db.alter/attribute to the store, or they handled it specially. I'll
need to remove the marker flag system from Mentat in order to address
this fundamental limitation.
* Post: Remove some more single quotes from error output.
* Post: Add assert_transact! macro to unwrap safely.
I was finding it very difficult to track unwrapping errors while
making changes, due to an underlying Mac OS X symbolication issue that
makes running tests with RUST_BACKTRACE=1 so slow that they all time
out.
* Post: Don't expect or recognize :db.{install,alter}/attribute.
I had this all working... except we will never see a repeated
`[:db.part/db :db.alter/attribute :attribute]` assertion in the store!
That means my approach would let you alter an attribute at most one
time. It's not worth hacking around this; it's better to just stop
expecting (and recognizing) the marker flags. (We have all the data
to distinguish the various cases that we need without the marker
flags.)
This brings Mentat in line with the thrust of newer Datomic versions,
but isn't compatible with Datomic, because (if I understand correctly)
Datomic automatically adds :db.{install,alter}/attribute assertions to
transactions.
I haven't purged the corresponding :db/ident and schema fragments just
yet:
- we might want them back
- we might want them in order to upgrade v1 and v2 databases to the
new on-disk layout we're fleshing out (v3?).
* Post: Don't make :db/unique :db.unique/* imply :db/index true.
This patch avoids a potential bug with the "schema" materialized view.
If :db/unique :db.unique/value implies :db/index true, then what
happens when you _retract_ :db.unique/value? I think Datomic defines
this in some way, but I really want the "schema" materialized view to
be a slice of "datoms" and not have these sort of ambiguities and
persistent effects. Therefore, to ensure that we don't retract a
schema characteristic and accidentally change more than we intended
to, this patch stops having any schema characteristic imply any other
schema characteristic(s). To achieve that, I added an
Option<Unique::{Value,Identity}> type to Attribute; this helps with
this patch, and also looks ahead to when we allow to retract
:db/unique attributes.
* Post: Allow to retract :db/ident.
* Post: Include more details about invalid schema changes.
The tests use strings, so they hide the chained errors which do in
fact provide more detail.
* Review comment: Fix outdated comment.
* Review comment: s/_SET/_SQL_LIST/.
* Review comment: Use a sub-select for checking cardinality.
This might be faster in practice.
* Review comment: Put `attribute::Unique` into its own namespace.
2017-03-20 20:18:59 +00:00
/// Read the schema materialized view from the given SQL store.
2018-07-16 20:58:34 +00:00
pub ( crate ) fn read_attribute_map ( conn : & rusqlite ::Connection ) -> Result < AttributeMap > {
Schema alteration. Fixes #294 and #295. (#370) r=rnewman
* Pre: Don't retract :db/ident in test.
Datomic (and eventually Mentat) don't allow to retract :db/ident in
this way, so this runs afoul of future work to support mutating
metadata.
* Pre: s/VALUETYPE/VALUE_TYPE/.
This is consistent with the capitalization (which is "valueType") and
the other identifier.
* Pre: Remove some single quotes from error output.
* Part 1: Make materialized views be uniform [e a v value_type_tag].
This looks ahead to a time when we could support arbitrary
user-defined materialized views. For now, the "idents" materialized
view is those datoms of the form [e :db/ident :namespaced/keyword] and
the "schema" materialized view is those datoms of the form [e a v]
where a is in a particular set of attributes that will become clear in
the following commits.
This change is not backwards compatible, so I'm removing the open
current (really, v2) test. It'll be re-instated when we get to
https://github.com/mozilla/mentat/issues/194.
* Pre: Map TypedValue::Ref to TypedValue::Keyword in debug output.
* Part 3: Separate `schema_to_mutate` from the `schema` used to interpret.
This is just to keep track of the expected changes during
bootstrapping. I want bootstrap metadata mutations to flow through
the same code path as metadata mutations during regular transactions;
by differentiating the schema used for interpretation from the schema
that will be updated I expect to be able to apply bootstrap metadata
mutations to an empty schema and have things like materialized views
created (using the regular code paths).
This commit has been re-ordered for conceptual clarity, but it won't
compile because it references the metadata module. It's possible to
make it compile -- the functionality is there in the schema module --
but it's not worth the rebasing effort until after review (and
possibly not even then, since we'll squash down to a single commit to
land).
* Part 2: Maintain entids separately from idents.
In order to support historical idents, we need to distinguish the
"current" map from entid -> ident from the "complete historical" map
ident -> entid. This is what Datomic does; in Datomic, an ident is
never retracted (although it can be replaced). This approach is an
important part of allowing multiple consumers to share a schema
fragment as it migrates forward.
This fixes a limitation of the Clojure implementation, which did not
handle historical idents across knowledge base close and re-open.
The "entids" materialized view is naturally a slice of the "datoms"
table. The "idents" materialized view is a slice of the
"transactions" table. I hope that representing in this way, and
casting the problem in this light, might generalize to future
materialized views.
* Pre: Add DiffSet.
* Part 4: Collect mutations to a `Schema`.
I haven't taken your review comment about consuming AttributeBuilder
during each fluent function. If you read my response and still want
this, I'm happy to do it in review.
* Part 5: Handle :db/ident and :db.{install,alter}/attribute.
This "loops" the committed datoms out of the SQL store and back
through the metadata (schema, but in future also partition map)
processor. The metadata processor updates the schema and produces a
report of what changed; that report is then used to update the SQL
store. That update includes:
- the materialized views ("entids", "idents", and "schema");
- if needed, a subset of the datoms themselves (as flags change).
I've left a TODO for handling attribute retraction in the cases that
it makes sense. I expect that to be straight-forward.
* Review comment: Rename DiffSet to AddRetractAlterSet.
Also adds a little more commentary and a simple test.
* Review comment: Use ToIdent trait.
* Review comment: partially revert "Part 2: Maintain entids separately from idents."
This reverts commit 23a91df9c35e14398f2ddbd1ba25315821e67401.
Following our discussion, this removes the "entids" materialized
view. The next commit will remove historical idents from the "idents"
materialized view.
* Post: Use custom Either rather than std::result::Result.
This is not necessary, but it was suggested that we might be paying an
overhead creating Err instances while using error_chain. That seems
not to be the case, but this change shows that we don't actually use
any of the Result helper methods, so there's no reason to overload
Result. This change might avoid some future confusion, so I'm going
to land it anyway.
Signed-off-by: Nick Alexander <nalexander@mozilla.com>
* Review comment: Don't preserve historical idents.
* Review comment: More prepared statements when updating materialized views.
* Post: Test altering :db/cardinality and :db/unique.
These tests fail due to a Datomic limitation, namely that the marker
flag :db.alter/attribute can only be asserted once for an attribute!
That is, [:db.part/db :db.alter/attribute :attribute] will only be
transacted at most once. Since older versions of Datomic required the
:db.alter/attribute flag, I can only imagine they either never wrote
:db.alter/attribute to the store, or they handled it specially. I'll
need to remove the marker flag system from Mentat in order to address
this fundamental limitation.
* Post: Remove some more single quotes from error output.
* Post: Add assert_transact! macro to unwrap safely.
I was finding it very difficult to track unwrapping errors while
making changes, due to an underlying Mac OS X symbolication issue that
makes running tests with RUST_BACKTRACE=1 so slow that they all time
out.
* Post: Don't expect or recognize :db.{install,alter}/attribute.
I had this all working... except we will never see a repeated
`[:db.part/db :db.alter/attribute :attribute]` assertion in the store!
That means my approach would let you alter an attribute at most one
time. It's not worth hacking around this; it's better to just stop
expecting (and recognizing) the marker flags. (We have all the data
to distinguish the various cases that we need without the marker
flags.)
This brings Mentat in line with the thrust of newer Datomic versions,
but isn't compatible with Datomic, because (if I understand correctly)
Datomic automatically adds :db.{install,alter}/attribute assertions to
transactions.
I haven't purged the corresponding :db/ident and schema fragments just
yet:
- we might want them back
- we might want them in order to upgrade v1 and v2 databases to the
new on-disk layout we're fleshing out (v3?).
* Post: Don't make :db/unique :db.unique/* imply :db/index true.
This patch avoids a potential bug with the "schema" materialized view.
If :db/unique :db.unique/value implies :db/index true, then what
happens when you _retract_ :db.unique/value? I think Datomic defines
this in some way, but I really want the "schema" materialized view to
be a slice of "datoms" and not have these sort of ambiguities and
persistent effects. Therefore, to ensure that we don't retract a
schema characteristic and accidentally change more than we intended
to, this patch stops having any schema characteristic imply any other
schema characteristic(s). To achieve that, I added an
Option<Unique::{Value,Identity}> type to Attribute; this helps with
this patch, and also looks ahead to when we allow to retract
:db/unique attributes.
* Post: Allow to retract :db/ident.
* Post: Include more details about invalid schema changes.
The tests use strings, so they hide the chained errors which do in
fact provide more detail.
* Review comment: Fix outdated comment.
* Review comment: s/_SET/_SQL_LIST/.
* Review comment: Use a sub-select for checking cardinality.
This might be faster in practice.
* Review comment: Put `attribute::Unique` into its own namespace.
2017-03-20 20:18:59 +00:00
let entid_triples = read_materialized_view ( conn , " schema " ) ? ;
2018-01-23 16:23:37 +00:00
let mut attribute_map = AttributeMap ::default ( ) ;
2018-07-18 23:55:31 +00:00
metadata ::update_attribute_map_from_entid_triples ( & mut attribute_map , entid_triples , vec! [ ] ) ? ;
2018-01-23 16:23:37 +00:00
Ok ( attribute_map )
2017-01-26 00:13:56 +00:00
}
/// Read the materialized views from the given SQL store and return a Mentat `DB` for querying and
/// applying transactions.
2018-07-16 20:58:34 +00:00
pub ( crate ) fn read_db ( conn : & rusqlite ::Connection ) -> Result < DB > {
2017-01-26 00:13:56 +00:00
let partition_map = read_partition_map ( conn ) ? ;
let ident_map = read_ident_map ( conn ) ? ;
2018-01-23 16:23:37 +00:00
let attribute_map = read_attribute_map ( conn ) ? ;
let schema = Schema ::from_ident_map_and_attribute_map ( ident_map , attribute_map ) ? ;
2017-01-26 00:13:56 +00:00
Ok ( DB ::new ( partition_map , schema ) )
}
2017-02-08 22:04:32 +00:00
/// Internal representation of an [e a v added] datom, ready to be transacted against the store.
Extract partial storage abstraction; use error-chain throughout. Fixes #328. r=rnewman (#341)
* Pre: Drop unneeded tx0 from search results.
* Pre: Don't require a schema in some of the DB code.
The idea is to separate the transaction applying code, which is
schema-aware, from the concrete storage code, which is just concerned
with getting bits onto disk.
* Pre: Only reference Schema, not DB, in debug module.
This is part of a larger separation of the volatile PartitionMap,
which is modified every transaction, from the stable Schema, which is
infrequently modified.
* Pre: Fix indentation.
* Extract part of DB to new SchemaTypeChecking trait.
* Extract part of DB to new PartitionMapping trait.
* Pre: Don't expect :db.part/tx partition to advance when tx fails.
This fails right now, because we allocate tx IDs even when we shouldn't.
* Sketch a db interface without DB.
* Add ValueParseError; use error-chain in tx-parser.
This can be simplified when
https://github.com/Marwes/combine/issues/86 makes it to a published
release, but this unblocks us for now. This converts the `combine`
error type `ParseError<&'a [edn::Value]>` to a type with owned
`Vec<edn::Value>` collections, re-using `edn::Value::Vector` for
making them `Display`.
* Pre: Accept Borrow<Schema> instead of just &Schema in debug module.
This makes it easy to use Rc<Schema> or Arc<Schema> without inserting
&* sigils throughout the code.
* Use error-chain in query-parser.
There are a few things to point out here:
- the fine grained error types have been flattened into one crate-wide
error type; it's pretty easy to regain the granularity as needed.
- edn::ParseError is automatically lifted to
mentat_query_parser::errors::Error;
- we use mentat_parser_utils::ValueParser to maintain parsing error
information from `combine`.
* Patch up top-level.
* Review comment: Only `borrow()` once.
2017-02-24 23:32:41 +00:00
pub type ReducedEntity < ' a > = ( Entid , Entid , & ' a Attribute , TypedValue , bool ) ;
2017-02-08 22:04:32 +00:00
2019-07-17 15:46:47 +00:00
#[ derive(Clone, Debug, Eq, Hash, Ord, PartialOrd, PartialEq) ]
2017-02-08 22:04:32 +00:00
pub enum SearchType {
Exact ,
Inexact ,
}
Extract partial storage abstraction; use error-chain throughout. Fixes #328. r=rnewman (#341)
* Pre: Drop unneeded tx0 from search results.
* Pre: Don't require a schema in some of the DB code.
The idea is to separate the transaction applying code, which is
schema-aware, from the concrete storage code, which is just concerned
with getting bits onto disk.
* Pre: Only reference Schema, not DB, in debug module.
This is part of a larger separation of the volatile PartitionMap,
which is modified every transaction, from the stable Schema, which is
infrequently modified.
* Pre: Fix indentation.
* Extract part of DB to new SchemaTypeChecking trait.
* Extract part of DB to new PartitionMapping trait.
* Pre: Don't expect :db.part/tx partition to advance when tx fails.
This fails right now, because we allocate tx IDs even when we shouldn't.
* Sketch a db interface without DB.
* Add ValueParseError; use error-chain in tx-parser.
This can be simplified when
https://github.com/Marwes/combine/issues/86 makes it to a published
release, but this unblocks us for now. This converts the `combine`
error type `ParseError<&'a [edn::Value]>` to a type with owned
`Vec<edn::Value>` collections, re-using `edn::Value::Vector` for
making them `Display`.
* Pre: Accept Borrow<Schema> instead of just &Schema in debug module.
This makes it easy to use Rc<Schema> or Arc<Schema> without inserting
&* sigils throughout the code.
* Use error-chain in query-parser.
There are a few things to point out here:
- the fine grained error types have been flattened into one crate-wide
error type; it's pretty easy to regain the granularity as needed.
- edn::ParseError is automatically lifted to
mentat_query_parser::errors::Error;
- we use mentat_parser_utils::ValueParser to maintain parsing error
information from `combine`.
* Patch up top-level.
* Review comment: Only `borrow()` once.
2017-02-24 23:32:41 +00:00
/// `MentatStoring` will be the trait that encapsulates the storage layer. It is consumed by the
/// transaction processing layer.
///
/// Right now, the only implementation of `MentatStoring` is the SQLite-specific SQL schema. In the
/// future, we might consider other SQL engines (perhaps with different fulltext indexing), or
/// entirely different data stores, say ones shaped like key-value stores.
pub trait MentatStoring {
2017-02-15 00:50:40 +00:00
/// Given a slice of [a v] lookup-refs, look up the corresponding [e a v] triples.
///
/// It is assumed that the attribute `a` in each lookup-ref is `:db/unique`, so that at most one
/// matching [e a v] triple exists. (If this is not true, some matching entid `e` will be
/// chosen non-deterministically, if one exists.)
///
/// Returns a map &(a, v) -> e, to avoid cloning potentially large values. The keys of the map
Extract partial storage abstraction; use error-chain throughout. Fixes #328. r=rnewman (#341)
* Pre: Drop unneeded tx0 from search results.
* Pre: Don't require a schema in some of the DB code.
The idea is to separate the transaction applying code, which is
schema-aware, from the concrete storage code, which is just concerned
with getting bits onto disk.
* Pre: Only reference Schema, not DB, in debug module.
This is part of a larger separation of the volatile PartitionMap,
which is modified every transaction, from the stable Schema, which is
infrequently modified.
* Pre: Fix indentation.
* Extract part of DB to new SchemaTypeChecking trait.
* Extract part of DB to new PartitionMapping trait.
* Pre: Don't expect :db.part/tx partition to advance when tx fails.
This fails right now, because we allocate tx IDs even when we shouldn't.
* Sketch a db interface without DB.
* Add ValueParseError; use error-chain in tx-parser.
This can be simplified when
https://github.com/Marwes/combine/issues/86 makes it to a published
release, but this unblocks us for now. This converts the `combine`
error type `ParseError<&'a [edn::Value]>` to a type with owned
`Vec<edn::Value>` collections, re-using `edn::Value::Vector` for
making them `Display`.
* Pre: Accept Borrow<Schema> instead of just &Schema in debug module.
This makes it easy to use Rc<Schema> or Arc<Schema> without inserting
&* sigils throughout the code.
* Use error-chain in query-parser.
There are a few things to point out here:
- the fine grained error types have been flattened into one crate-wide
error type; it's pretty easy to regain the granularity as needed.
- edn::ParseError is automatically lifted to
mentat_query_parser::errors::Error;
- we use mentat_parser_utils::ValueParser to maintain parsing error
information from `combine`.
* Patch up top-level.
* Review comment: Only `borrow()` once.
2017-02-24 23:32:41 +00:00
/// are exactly those (a, v) pairs that have an assertion [e a v] in the store.
fn resolve_avs < ' a > ( & self , avs : & ' a [ & ' a AVPair ] ) -> Result < AVMap < ' a > > ;
/// Begin (or prepare) the underlying storage layer for a new Mentat transaction.
///
/// Use this to create temporary tables, prepare indices, set pragmas, etc, before the initial
/// `insert_non_fts_searches` invocation.
2017-12-05 15:58:24 +00:00
fn begin_tx_application ( & self ) -> Result < ( ) > ;
Extract partial storage abstraction; use error-chain throughout. Fixes #328. r=rnewman (#341)
* Pre: Drop unneeded tx0 from search results.
* Pre: Don't require a schema in some of the DB code.
The idea is to separate the transaction applying code, which is
schema-aware, from the concrete storage code, which is just concerned
with getting bits onto disk.
* Pre: Only reference Schema, not DB, in debug module.
This is part of a larger separation of the volatile PartitionMap,
which is modified every transaction, from the stable Schema, which is
infrequently modified.
* Pre: Fix indentation.
* Extract part of DB to new SchemaTypeChecking trait.
* Extract part of DB to new PartitionMapping trait.
* Pre: Don't expect :db.part/tx partition to advance when tx fails.
This fails right now, because we allocate tx IDs even when we shouldn't.
* Sketch a db interface without DB.
* Add ValueParseError; use error-chain in tx-parser.
This can be simplified when
https://github.com/Marwes/combine/issues/86 makes it to a published
release, but this unblocks us for now. This converts the `combine`
error type `ParseError<&'a [edn::Value]>` to a type with owned
`Vec<edn::Value>` collections, re-using `edn::Value::Vector` for
making them `Display`.
* Pre: Accept Borrow<Schema> instead of just &Schema in debug module.
This makes it easy to use Rc<Schema> or Arc<Schema> without inserting
&* sigils throughout the code.
* Use error-chain in query-parser.
There are a few things to point out here:
- the fine grained error types have been flattened into one crate-wide
error type; it's pretty easy to regain the granularity as needed.
- edn::ParseError is automatically lifted to
mentat_query_parser::errors::Error;
- we use mentat_parser_utils::ValueParser to maintain parsing error
information from `combine`.
* Patch up top-level.
* Review comment: Only `borrow()` once.
2017-02-24 23:32:41 +00:00
// TODO: this is not a reasonable abstraction, but I don't want to really consider non-SQL storage just yet.
2019-07-17 15:46:47 +00:00
fn insert_non_fts_searches < ' a > (
& self ,
entities : & ' a [ ReducedEntity ] ,
search_type : SearchType ,
) -> Result < ( ) > ;
fn insert_fts_searches < ' a > (
& self ,
entities : & ' a [ ReducedEntity ] ,
search_type : SearchType ,
) -> Result < ( ) > ;
Extract partial storage abstraction; use error-chain throughout. Fixes #328. r=rnewman (#341)
* Pre: Drop unneeded tx0 from search results.
* Pre: Don't require a schema in some of the DB code.
The idea is to separate the transaction applying code, which is
schema-aware, from the concrete storage code, which is just concerned
with getting bits onto disk.
* Pre: Only reference Schema, not DB, in debug module.
This is part of a larger separation of the volatile PartitionMap,
which is modified every transaction, from the stable Schema, which is
infrequently modified.
* Pre: Fix indentation.
* Extract part of DB to new SchemaTypeChecking trait.
* Extract part of DB to new PartitionMapping trait.
* Pre: Don't expect :db.part/tx partition to advance when tx fails.
This fails right now, because we allocate tx IDs even when we shouldn't.
* Sketch a db interface without DB.
* Add ValueParseError; use error-chain in tx-parser.
This can be simplified when
https://github.com/Marwes/combine/issues/86 makes it to a published
release, but this unblocks us for now. This converts the `combine`
error type `ParseError<&'a [edn::Value]>` to a type with owned
`Vec<edn::Value>` collections, re-using `edn::Value::Vector` for
making them `Display`.
* Pre: Accept Borrow<Schema> instead of just &Schema in debug module.
This makes it easy to use Rc<Schema> or Arc<Schema> without inserting
&* sigils throughout the code.
* Use error-chain in query-parser.
There are a few things to point out here:
- the fine grained error types have been flattened into one crate-wide
error type; it's pretty easy to regain the granularity as needed.
- edn::ParseError is automatically lifted to
mentat_query_parser::errors::Error;
- we use mentat_parser_utils::ValueParser to maintain parsing error
information from `combine`.
* Patch up top-level.
* Review comment: Only `borrow()` once.
2017-02-24 23:32:41 +00:00
2018-07-17 19:18:53 +00:00
/// Prepare the underlying storage layer for finalization after a Mentat transaction.
Extract partial storage abstraction; use error-chain throughout. Fixes #328. r=rnewman (#341)
* Pre: Drop unneeded tx0 from search results.
* Pre: Don't require a schema in some of the DB code.
The idea is to separate the transaction applying code, which is
schema-aware, from the concrete storage code, which is just concerned
with getting bits onto disk.
* Pre: Only reference Schema, not DB, in debug module.
This is part of a larger separation of the volatile PartitionMap,
which is modified every transaction, from the stable Schema, which is
infrequently modified.
* Pre: Fix indentation.
* Extract part of DB to new SchemaTypeChecking trait.
* Extract part of DB to new PartitionMapping trait.
* Pre: Don't expect :db.part/tx partition to advance when tx fails.
This fails right now, because we allocate tx IDs even when we shouldn't.
* Sketch a db interface without DB.
* Add ValueParseError; use error-chain in tx-parser.
This can be simplified when
https://github.com/Marwes/combine/issues/86 makes it to a published
release, but this unblocks us for now. This converts the `combine`
error type `ParseError<&'a [edn::Value]>` to a type with owned
`Vec<edn::Value>` collections, re-using `edn::Value::Vector` for
making them `Display`.
* Pre: Accept Borrow<Schema> instead of just &Schema in debug module.
This makes it easy to use Rc<Schema> or Arc<Schema> without inserting
&* sigils throughout the code.
* Use error-chain in query-parser.
There are a few things to point out here:
- the fine grained error types have been flattened into one crate-wide
error type; it's pretty easy to regain the granularity as needed.
- edn::ParseError is automatically lifted to
mentat_query_parser::errors::Error;
- we use mentat_parser_utils::ValueParser to maintain parsing error
information from `combine`.
* Patch up top-level.
* Review comment: Only `borrow()` once.
2017-02-24 23:32:41 +00:00
///
/// Use this to finalize temporary tables, complete indices, revert pragmas, etc, after the
/// final `insert_non_fts_searches` invocation.
2018-07-17 19:18:53 +00:00
fn materialize_mentat_transaction ( & self , tx_id : Entid ) -> Result < ( ) > ;
/// Finalize the underlying storage layer after a Mentat transaction.
///
/// This is a final step in performing a transaction.
fn commit_mentat_transaction ( & self , tx_id : Entid ) -> Result < ( ) > ;
Schema alteration. Fixes #294 and #295. (#370) r=rnewman
* Pre: Don't retract :db/ident in test.
Datomic (and eventually Mentat) don't allow to retract :db/ident in
this way, so this runs afoul of future work to support mutating
metadata.
* Pre: s/VALUETYPE/VALUE_TYPE/.
This is consistent with the capitalization (which is "valueType") and
the other identifier.
* Pre: Remove some single quotes from error output.
* Part 1: Make materialized views be uniform [e a v value_type_tag].
This looks ahead to a time when we could support arbitrary
user-defined materialized views. For now, the "idents" materialized
view is those datoms of the form [e :db/ident :namespaced/keyword] and
the "schema" materialized view is those datoms of the form [e a v]
where a is in a particular set of attributes that will become clear in
the following commits.
This change is not backwards compatible, so I'm removing the open
current (really, v2) test. It'll be re-instated when we get to
https://github.com/mozilla/mentat/issues/194.
* Pre: Map TypedValue::Ref to TypedValue::Keyword in debug output.
* Part 3: Separate `schema_to_mutate` from the `schema` used to interpret.
This is just to keep track of the expected changes during
bootstrapping. I want bootstrap metadata mutations to flow through
the same code path as metadata mutations during regular transactions;
by differentiating the schema used for interpretation from the schema
that will be updated I expect to be able to apply bootstrap metadata
mutations to an empty schema and have things like materialized views
created (using the regular code paths).
This commit has been re-ordered for conceptual clarity, but it won't
compile because it references the metadata module. It's possible to
make it compile -- the functionality is there in the schema module --
but it's not worth the rebasing effort until after review (and
possibly not even then, since we'll squash down to a single commit to
land).
* Part 2: Maintain entids separately from idents.
In order to support historical idents, we need to distinguish the
"current" map from entid -> ident from the "complete historical" map
ident -> entid. This is what Datomic does; in Datomic, an ident is
never retracted (although it can be replaced). This approach is an
important part of allowing multiple consumers to share a schema
fragment as it migrates forward.
This fixes a limitation of the Clojure implementation, which did not
handle historical idents across knowledge base close and re-open.
The "entids" materialized view is naturally a slice of the "datoms"
table. The "idents" materialized view is a slice of the
"transactions" table. I hope that representing in this way, and
casting the problem in this light, might generalize to future
materialized views.
* Pre: Add DiffSet.
* Part 4: Collect mutations to a `Schema`.
I haven't taken your review comment about consuming AttributeBuilder
during each fluent function. If you read my response and still want
this, I'm happy to do it in review.
* Part 5: Handle :db/ident and :db.{install,alter}/attribute.
This "loops" the committed datoms out of the SQL store and back
through the metadata (schema, but in future also partition map)
processor. The metadata processor updates the schema and produces a
report of what changed; that report is then used to update the SQL
store. That update includes:
- the materialized views ("entids", "idents", and "schema");
- if needed, a subset of the datoms themselves (as flags change).
I've left a TODO for handling attribute retraction in the cases that
it makes sense. I expect that to be straight-forward.
* Review comment: Rename DiffSet to AddRetractAlterSet.
Also adds a little more commentary and a simple test.
* Review comment: Use ToIdent trait.
* Review comment: partially revert "Part 2: Maintain entids separately from idents."
This reverts commit 23a91df9c35e14398f2ddbd1ba25315821e67401.
Following our discussion, this removes the "entids" materialized
view. The next commit will remove historical idents from the "idents"
materialized view.
* Post: Use custom Either rather than std::result::Result.
This is not necessary, but it was suggested that we might be paying an
overhead creating Err instances while using error_chain. That seems
not to be the case, but this change shows that we don't actually use
any of the Result helper methods, so there's no reason to overload
Result. This change might avoid some future confusion, so I'm going
to land it anyway.
Signed-off-by: Nick Alexander <nalexander@mozilla.com>
* Review comment: Don't preserve historical idents.
* Review comment: More prepared statements when updating materialized views.
* Post: Test altering :db/cardinality and :db/unique.
These tests fail due to a Datomic limitation, namely that the marker
flag :db.alter/attribute can only be asserted once for an attribute!
That is, [:db.part/db :db.alter/attribute :attribute] will only be
transacted at most once. Since older versions of Datomic required the
:db.alter/attribute flag, I can only imagine they either never wrote
:db.alter/attribute to the store, or they handled it specially. I'll
need to remove the marker flag system from Mentat in order to address
this fundamental limitation.
* Post: Remove some more single quotes from error output.
* Post: Add assert_transact! macro to unwrap safely.
I was finding it very difficult to track unwrapping errors while
making changes, due to an underlying Mac OS X symbolication issue that
makes running tests with RUST_BACKTRACE=1 so slow that they all time
out.
* Post: Don't expect or recognize :db.{install,alter}/attribute.
I had this all working... except we will never see a repeated
`[:db.part/db :db.alter/attribute :attribute]` assertion in the store!
That means my approach would let you alter an attribute at most one
time. It's not worth hacking around this; it's better to just stop
expecting (and recognizing) the marker flags. (We have all the data
to distinguish the various cases that we need without the marker
flags.)
This brings Mentat in line with the thrust of newer Datomic versions,
but isn't compatible with Datomic, because (if I understand correctly)
Datomic automatically adds :db.{install,alter}/attribute assertions to
transactions.
I haven't purged the corresponding :db/ident and schema fragments just
yet:
- we might want them back
- we might want them in order to upgrade v1 and v2 databases to the
new on-disk layout we're fleshing out (v3?).
* Post: Don't make :db/unique :db.unique/* imply :db/index true.
This patch avoids a potential bug with the "schema" materialized view.
If :db/unique :db.unique/value implies :db/index true, then what
happens when you _retract_ :db.unique/value? I think Datomic defines
this in some way, but I really want the "schema" materialized view to
be a slice of "datoms" and not have these sort of ambiguities and
persistent effects. Therefore, to ensure that we don't retract a
schema characteristic and accidentally change more than we intended
to, this patch stops having any schema characteristic imply any other
schema characteristic(s). To achieve that, I added an
Option<Unique::{Value,Identity}> type to Attribute; this helps with
this patch, and also looks ahead to when we allow to retract
:db/unique attributes.
* Post: Allow to retract :db/ident.
* Post: Include more details about invalid schema changes.
The tests use strings, so they hide the chained errors which do in
fact provide more detail.
* Review comment: Fix outdated comment.
* Review comment: s/_SET/_SQL_LIST/.
* Review comment: Use a sub-select for checking cardinality.
This might be faster in practice.
* Review comment: Put `attribute::Unique` into its own namespace.
2017-03-20 20:18:59 +00:00
2018-07-17 19:18:53 +00:00
/// Extract metadata-related [e a typed_value added] datoms resolved in the last
/// materialized transaction.
fn resolved_metadata_assertions ( & self ) -> Result < Vec < ( Entid , Entid , TypedValue , bool ) > > ;
Extract partial storage abstraction; use error-chain throughout. Fixes #328. r=rnewman (#341)
* Pre: Drop unneeded tx0 from search results.
* Pre: Don't require a schema in some of the DB code.
The idea is to separate the transaction applying code, which is
schema-aware, from the concrete storage code, which is just concerned
with getting bits onto disk.
* Pre: Only reference Schema, not DB, in debug module.
This is part of a larger separation of the volatile PartitionMap,
which is modified every transaction, from the stable Schema, which is
infrequently modified.
* Pre: Fix indentation.
* Extract part of DB to new SchemaTypeChecking trait.
* Extract part of DB to new PartitionMapping trait.
* Pre: Don't expect :db.part/tx partition to advance when tx fails.
This fails right now, because we allocate tx IDs even when we shouldn't.
* Sketch a db interface without DB.
* Add ValueParseError; use error-chain in tx-parser.
This can be simplified when
https://github.com/Marwes/combine/issues/86 makes it to a published
release, but this unblocks us for now. This converts the `combine`
error type `ParseError<&'a [edn::Value]>` to a type with owned
`Vec<edn::Value>` collections, re-using `edn::Value::Vector` for
making them `Display`.
* Pre: Accept Borrow<Schema> instead of just &Schema in debug module.
This makes it easy to use Rc<Schema> or Arc<Schema> without inserting
&* sigils throughout the code.
* Use error-chain in query-parser.
There are a few things to point out here:
- the fine grained error types have been flattened into one crate-wide
error type; it's pretty easy to regain the granularity as needed.
- edn::ParseError is automatically lifted to
mentat_query_parser::errors::Error;
- we use mentat_parser_utils::ValueParser to maintain parsing error
information from `combine`.
* Patch up top-level.
* Review comment: Only `borrow()` once.
2017-02-24 23:32:41 +00:00
}
/// Take search rows and complete `temp.search_results`.
///
/// See https://github.com/mozilla/mentat/wiki/Transacting:-entity-to-SQL-translation.
fn search ( conn : & rusqlite ::Connection ) -> Result < ( ) > {
// First is fast, only one table walk: lookup by exact eav.
// Second is slower, but still only one table walk: lookup old value by ea.
let s = r #"
INSERT INTO temp . search_results
SELECT t . e0 , t . a0 , t . v0 , t . value_type_tag0 , t . added0 , t . flags0 , ' :db . cardinality / many ' , d . rowid , d . v
FROM temp . exact_searches AS t
LEFT JOIN datoms AS d
ON t . e0 = d . e AND
t . a0 = d . a AND
t . value_type_tag0 = d . value_type_tag AND
t . v0 = d . v
UNION ALL
SELECT t . e0 , t . a0 , t . v0 , t . value_type_tag0 , t . added0 , t . flags0 , ' :db . cardinality / one ' , d . rowid , d . v
FROM temp . inexact_searches AS t
LEFT JOIN datoms AS d
ON t . e0 = d . e AND
t . a0 = d . a " #;
let mut stmt = conn . prepare_cached ( s ) ? ;
2020-01-14 15:46:21 +00:00
stmt . execute ( rusqlite ::params! [ ] )
. context ( DbErrorKind ::CouldNotSearch ) ? ;
2018-06-06 01:23:59 +00:00
Ok ( ( ) )
Extract partial storage abstraction; use error-chain throughout. Fixes #328. r=rnewman (#341)
* Pre: Drop unneeded tx0 from search results.
* Pre: Don't require a schema in some of the DB code.
The idea is to separate the transaction applying code, which is
schema-aware, from the concrete storage code, which is just concerned
with getting bits onto disk.
* Pre: Only reference Schema, not DB, in debug module.
This is part of a larger separation of the volatile PartitionMap,
which is modified every transaction, from the stable Schema, which is
infrequently modified.
* Pre: Fix indentation.
* Extract part of DB to new SchemaTypeChecking trait.
* Extract part of DB to new PartitionMapping trait.
* Pre: Don't expect :db.part/tx partition to advance when tx fails.
This fails right now, because we allocate tx IDs even when we shouldn't.
* Sketch a db interface without DB.
* Add ValueParseError; use error-chain in tx-parser.
This can be simplified when
https://github.com/Marwes/combine/issues/86 makes it to a published
release, but this unblocks us for now. This converts the `combine`
error type `ParseError<&'a [edn::Value]>` to a type with owned
`Vec<edn::Value>` collections, re-using `edn::Value::Vector` for
making them `Display`.
* Pre: Accept Borrow<Schema> instead of just &Schema in debug module.
This makes it easy to use Rc<Schema> or Arc<Schema> without inserting
&* sigils throughout the code.
* Use error-chain in query-parser.
There are a few things to point out here:
- the fine grained error types have been flattened into one crate-wide
error type; it's pretty easy to regain the granularity as needed.
- edn::ParseError is automatically lifted to
mentat_query_parser::errors::Error;
- we use mentat_parser_utils::ValueParser to maintain parsing error
information from `combine`.
* Patch up top-level.
* Review comment: Only `borrow()` once.
2017-02-24 23:32:41 +00:00
}
/// Insert the new transaction into the `transactions` table.
///
/// This turns the contents of `search_results` into a new transaction.
///
/// See https://github.com/mozilla/mentat/wiki/Transacting:-entity-to-SQL-translation.
fn insert_transaction ( conn : & rusqlite ::Connection , tx : Entid ) -> Result < ( ) > {
Add type checking and constraint checking to the transactor. (#663, #532, #679)
This should address #663, by re-inserting type checking in the
transactor stack after the entry point used by the term builder.
Before this commit, we were using an SQLite UNIQUE index to assert
that no `[e a]` pair, with `a` a cardinality one attribute, was
asserted more than once. However, that's not in line with Datomic,
which treats transaction inputs as a set and allows a single datom
like `[e a v]` to appear multiple times. It's both awkward and not
particularly efficient to look for _distinct_ repetitions in SQL, so
we accept some runtime cost in order to check for repetitions in the
transactor. This will allow us to address #532, which is really about
whether we treat inputs as sets. A side benefit is that we can
provide more helpful error messages when the transactor does detect
that the input truly violates the cardinality constraints of the
schema.
This commit builds a trie while error checking and collecting final
terms, which should be fairly efficient. It also allows a simpler
expression of input-provided :db/txInstant datoms, which in turn
uncovered a small issue with the transaction watcher, where-by the
watcher would not see non-input-provided :db/txInstant datoms.
This transition to Datomic-like input-as-set semantics allows us to
address #532. Previously, two tempids that upserted to the same entid
would produce duplicate datoms, and that would have been rejected by
the transactor -- correctly, since we did not allow duplicate datoms
under the input-as-list semantics. With input-as-set semantics,
duplicate datoms are allowed; and that means that we must allow
tempids to be equivalent, i.e., to resolve to the same tempid.
To achieve this, we:
- index the set of tempids
- identify tempid indices that share an upsert
- map tempids to a dense set of contiguous integer labels
We use the well-known union-find algorithm, as implemented by
petgraph, to efficiently manage the set of equivalent tempids.
Along the way, I've fixed and added tests for two small errors in the
transactor. First, don't drop datoms resolved by upsert (#679).
Second, ensure that complex upserts are allocated.
I don't know quite what happened here. The Clojure implementation
correctly kept complex upserts that hadn't resolved as complex
upserts (see
https://github.com/mozilla/mentat/blob/9a9dfb502acf5e4cdb1059d4aac831d7603063c8/src/common/datomish/transact.cljc#L436)
and then allocated complex upserts if they didn't resolve (see
https://github.com/mozilla/mentat/blob/9a9dfb502acf5e4cdb1059d4aac831d7603063c8/src/common/datomish/transact.cljc#L509).
Based on the code comments, I think the Rust implementation must have
incorrectly tried to optimize by handling all complex upserts in at
most a single generation of evolution, and that's just not correct.
We're effectively implementing a topological sort, using very specific
domain knowledge, and its not true that a node in a topological sort
can be considered only once!
2018-04-30 22:16:05 +00:00
// Mentat follows Datomic and treats its input as a set. That means it is okay to transact the
// same [e a v] twice in one transaction. However, we don't want to represent the transacted
// datom twice. Therefore, the transactor unifies repeated datoms, and in addition we add
// indices to the search inputs and search results to ensure that we don't see repeated datoms
// at this point.
Extract partial storage abstraction; use error-chain throughout. Fixes #328. r=rnewman (#341)
* Pre: Drop unneeded tx0 from search results.
* Pre: Don't require a schema in some of the DB code.
The idea is to separate the transaction applying code, which is
schema-aware, from the concrete storage code, which is just concerned
with getting bits onto disk.
* Pre: Only reference Schema, not DB, in debug module.
This is part of a larger separation of the volatile PartitionMap,
which is modified every transaction, from the stable Schema, which is
infrequently modified.
* Pre: Fix indentation.
* Extract part of DB to new SchemaTypeChecking trait.
* Extract part of DB to new PartitionMapping trait.
* Pre: Don't expect :db.part/tx partition to advance when tx fails.
This fails right now, because we allocate tx IDs even when we shouldn't.
* Sketch a db interface without DB.
* Add ValueParseError; use error-chain in tx-parser.
This can be simplified when
https://github.com/Marwes/combine/issues/86 makes it to a published
release, but this unblocks us for now. This converts the `combine`
error type `ParseError<&'a [edn::Value]>` to a type with owned
`Vec<edn::Value>` collections, re-using `edn::Value::Vector` for
making them `Display`.
* Pre: Accept Borrow<Schema> instead of just &Schema in debug module.
This makes it easy to use Rc<Schema> or Arc<Schema> without inserting
&* sigils throughout the code.
* Use error-chain in query-parser.
There are a few things to point out here:
- the fine grained error types have been flattened into one crate-wide
error type; it's pretty easy to regain the granularity as needed.
- edn::ParseError is automatically lifted to
mentat_query_parser::errors::Error;
- we use mentat_parser_utils::ValueParser to maintain parsing error
information from `combine`.
* Patch up top-level.
* Review comment: Only `borrow()` once.
2017-02-24 23:32:41 +00:00
let s = r #"
2018-07-11 01:20:38 +00:00
INSERT INTO timelined_transactions ( e , a , v , tx , added , value_type_tag )
Extract partial storage abstraction; use error-chain throughout. Fixes #328. r=rnewman (#341)
* Pre: Drop unneeded tx0 from search results.
* Pre: Don't require a schema in some of the DB code.
The idea is to separate the transaction applying code, which is
schema-aware, from the concrete storage code, which is just concerned
with getting bits onto disk.
* Pre: Only reference Schema, not DB, in debug module.
This is part of a larger separation of the volatile PartitionMap,
which is modified every transaction, from the stable Schema, which is
infrequently modified.
* Pre: Fix indentation.
* Extract part of DB to new SchemaTypeChecking trait.
* Extract part of DB to new PartitionMapping trait.
* Pre: Don't expect :db.part/tx partition to advance when tx fails.
This fails right now, because we allocate tx IDs even when we shouldn't.
* Sketch a db interface without DB.
* Add ValueParseError; use error-chain in tx-parser.
This can be simplified when
https://github.com/Marwes/combine/issues/86 makes it to a published
release, but this unblocks us for now. This converts the `combine`
error type `ParseError<&'a [edn::Value]>` to a type with owned
`Vec<edn::Value>` collections, re-using `edn::Value::Vector` for
making them `Display`.
* Pre: Accept Borrow<Schema> instead of just &Schema in debug module.
This makes it easy to use Rc<Schema> or Arc<Schema> without inserting
&* sigils throughout the code.
* Use error-chain in query-parser.
There are a few things to point out here:
- the fine grained error types have been flattened into one crate-wide
error type; it's pretty easy to regain the granularity as needed.
- edn::ParseError is automatically lifted to
mentat_query_parser::errors::Error;
- we use mentat_parser_utils::ValueParser to maintain parsing error
information from `combine`.
* Patch up top-level.
* Review comment: Only `borrow()` once.
2017-02-24 23:32:41 +00:00
SELECT e0 , a0 , v0 , ? , 1 , value_type_tag0
FROM temp . search_results
WHERE added0 IS 1 AND ( ( rid IS NULL ) OR ( ( rid IS NOT NULL ) AND ( v0 IS NOT v ) ) ) " #;
let mut stmt = conn . prepare_cached ( s ) ? ;
2019-07-17 15:46:47 +00:00
stmt . execute ( & [ & tx ] )
. context ( DbErrorKind ::TxInsertFailedToAddMissingDatoms ) ? ;
Extract partial storage abstraction; use error-chain throughout. Fixes #328. r=rnewman (#341)
* Pre: Drop unneeded tx0 from search results.
* Pre: Don't require a schema in some of the DB code.
The idea is to separate the transaction applying code, which is
schema-aware, from the concrete storage code, which is just concerned
with getting bits onto disk.
* Pre: Only reference Schema, not DB, in debug module.
This is part of a larger separation of the volatile PartitionMap,
which is modified every transaction, from the stable Schema, which is
infrequently modified.
* Pre: Fix indentation.
* Extract part of DB to new SchemaTypeChecking trait.
* Extract part of DB to new PartitionMapping trait.
* Pre: Don't expect :db.part/tx partition to advance when tx fails.
This fails right now, because we allocate tx IDs even when we shouldn't.
* Sketch a db interface without DB.
* Add ValueParseError; use error-chain in tx-parser.
This can be simplified when
https://github.com/Marwes/combine/issues/86 makes it to a published
release, but this unblocks us for now. This converts the `combine`
error type `ParseError<&'a [edn::Value]>` to a type with owned
`Vec<edn::Value>` collections, re-using `edn::Value::Vector` for
making them `Display`.
* Pre: Accept Borrow<Schema> instead of just &Schema in debug module.
This makes it easy to use Rc<Schema> or Arc<Schema> without inserting
&* sigils throughout the code.
* Use error-chain in query-parser.
There are a few things to point out here:
- the fine grained error types have been flattened into one crate-wide
error type; it's pretty easy to regain the granularity as needed.
- edn::ParseError is automatically lifted to
mentat_query_parser::errors::Error;
- we use mentat_parser_utils::ValueParser to maintain parsing error
information from `combine`.
* Patch up top-level.
* Review comment: Only `borrow()` once.
2017-02-24 23:32:41 +00:00
let s = r #"
2018-07-11 01:20:38 +00:00
INSERT INTO timelined_transactions ( e , a , v , tx , added , value_type_tag )
2018-09-07 20:12:28 +00:00
SELECT DISTINCT e0 , a0 , v , ? , 0 , value_type_tag0
Extract partial storage abstraction; use error-chain throughout. Fixes #328. r=rnewman (#341)
* Pre: Drop unneeded tx0 from search results.
* Pre: Don't require a schema in some of the DB code.
The idea is to separate the transaction applying code, which is
schema-aware, from the concrete storage code, which is just concerned
with getting bits onto disk.
* Pre: Only reference Schema, not DB, in debug module.
This is part of a larger separation of the volatile PartitionMap,
which is modified every transaction, from the stable Schema, which is
infrequently modified.
* Pre: Fix indentation.
* Extract part of DB to new SchemaTypeChecking trait.
* Extract part of DB to new PartitionMapping trait.
* Pre: Don't expect :db.part/tx partition to advance when tx fails.
This fails right now, because we allocate tx IDs even when we shouldn't.
* Sketch a db interface without DB.
* Add ValueParseError; use error-chain in tx-parser.
This can be simplified when
https://github.com/Marwes/combine/issues/86 makes it to a published
release, but this unblocks us for now. This converts the `combine`
error type `ParseError<&'a [edn::Value]>` to a type with owned
`Vec<edn::Value>` collections, re-using `edn::Value::Vector` for
making them `Display`.
* Pre: Accept Borrow<Schema> instead of just &Schema in debug module.
This makes it easy to use Rc<Schema> or Arc<Schema> without inserting
&* sigils throughout the code.
* Use error-chain in query-parser.
There are a few things to point out here:
- the fine grained error types have been flattened into one crate-wide
error type; it's pretty easy to regain the granularity as needed.
- edn::ParseError is automatically lifted to
mentat_query_parser::errors::Error;
- we use mentat_parser_utils::ValueParser to maintain parsing error
information from `combine`.
* Patch up top-level.
* Review comment: Only `borrow()` once.
2017-02-24 23:32:41 +00:00
FROM temp . search_results
WHERE rid IS NOT NULL AND
( ( added0 IS 0 ) OR
( added0 IS 1 AND search_type IS ' :db . cardinality / one ' AND v0 IS NOT v ) ) " #;
let mut stmt = conn . prepare_cached ( s ) ? ;
2019-07-17 15:46:47 +00:00
stmt . execute ( & [ & tx ] )
. context ( DbErrorKind ::TxInsertFailedToRetractDatoms ) ? ;
Extract partial storage abstraction; use error-chain throughout. Fixes #328. r=rnewman (#341)
* Pre: Drop unneeded tx0 from search results.
* Pre: Don't require a schema in some of the DB code.
The idea is to separate the transaction applying code, which is
schema-aware, from the concrete storage code, which is just concerned
with getting bits onto disk.
* Pre: Only reference Schema, not DB, in debug module.
This is part of a larger separation of the volatile PartitionMap,
which is modified every transaction, from the stable Schema, which is
infrequently modified.
* Pre: Fix indentation.
* Extract part of DB to new SchemaTypeChecking trait.
* Extract part of DB to new PartitionMapping trait.
* Pre: Don't expect :db.part/tx partition to advance when tx fails.
This fails right now, because we allocate tx IDs even when we shouldn't.
* Sketch a db interface without DB.
* Add ValueParseError; use error-chain in tx-parser.
This can be simplified when
https://github.com/Marwes/combine/issues/86 makes it to a published
release, but this unblocks us for now. This converts the `combine`
error type `ParseError<&'a [edn::Value]>` to a type with owned
`Vec<edn::Value>` collections, re-using `edn::Value::Vector` for
making them `Display`.
* Pre: Accept Borrow<Schema> instead of just &Schema in debug module.
This makes it easy to use Rc<Schema> or Arc<Schema> without inserting
&* sigils throughout the code.
* Use error-chain in query-parser.
There are a few things to point out here:
- the fine grained error types have been flattened into one crate-wide
error type; it's pretty easy to regain the granularity as needed.
- edn::ParseError is automatically lifted to
mentat_query_parser::errors::Error;
- we use mentat_parser_utils::ValueParser to maintain parsing error
information from `combine`.
* Patch up top-level.
* Review comment: Only `borrow()` once.
2017-02-24 23:32:41 +00:00
Ok ( ( ) )
}
/// Update the contents of the `datoms` materialized view with the new transaction.
///
/// This applies the contents of `search_results` to the `datoms` table (in place).
///
/// See https://github.com/mozilla/mentat/wiki/Transacting:-entity-to-SQL-translation.
fn update_datoms ( conn : & rusqlite ::Connection , tx : Entid ) -> Result < ( ) > {
// Delete datoms that were retracted, or those that were :db.cardinality/one and will be
// replaced.
let s = r #"
WITH ids AS ( SELECT rid
FROM temp . search_results
WHERE rid IS NOT NULL AND
( ( added0 IS 0 ) OR
( added0 IS 1 AND search_type IS ' :db . cardinality / one ' AND v0 IS NOT v ) ) )
DELETE FROM datoms WHERE rowid IN ids " #;
let mut stmt = conn . prepare_cached ( s ) ? ;
2019-07-22 11:40:59 +00:00
stmt . execute ( rusqlite ::params! [ ] )
2019-07-17 15:46:47 +00:00
. context ( DbErrorKind ::DatomsUpdateFailedToRetract ) ? ;
Extract partial storage abstraction; use error-chain throughout. Fixes #328. r=rnewman (#341)
* Pre: Drop unneeded tx0 from search results.
* Pre: Don't require a schema in some of the DB code.
The idea is to separate the transaction applying code, which is
schema-aware, from the concrete storage code, which is just concerned
with getting bits onto disk.
* Pre: Only reference Schema, not DB, in debug module.
This is part of a larger separation of the volatile PartitionMap,
which is modified every transaction, from the stable Schema, which is
infrequently modified.
* Pre: Fix indentation.
* Extract part of DB to new SchemaTypeChecking trait.
* Extract part of DB to new PartitionMapping trait.
* Pre: Don't expect :db.part/tx partition to advance when tx fails.
This fails right now, because we allocate tx IDs even when we shouldn't.
* Sketch a db interface without DB.
* Add ValueParseError; use error-chain in tx-parser.
This can be simplified when
https://github.com/Marwes/combine/issues/86 makes it to a published
release, but this unblocks us for now. This converts the `combine`
error type `ParseError<&'a [edn::Value]>` to a type with owned
`Vec<edn::Value>` collections, re-using `edn::Value::Vector` for
making them `Display`.
* Pre: Accept Borrow<Schema> instead of just &Schema in debug module.
This makes it easy to use Rc<Schema> or Arc<Schema> without inserting
&* sigils throughout the code.
* Use error-chain in query-parser.
There are a few things to point out here:
- the fine grained error types have been flattened into one crate-wide
error type; it's pretty easy to regain the granularity as needed.
- edn::ParseError is automatically lifted to
mentat_query_parser::errors::Error;
- we use mentat_parser_utils::ValueParser to maintain parsing error
information from `combine`.
* Patch up top-level.
* Review comment: Only `borrow()` once.
2017-02-24 23:32:41 +00:00
Add type checking and constraint checking to the transactor. (#663, #532, #679)
This should address #663, by re-inserting type checking in the
transactor stack after the entry point used by the term builder.
Before this commit, we were using an SQLite UNIQUE index to assert
that no `[e a]` pair, with `a` a cardinality one attribute, was
asserted more than once. However, that's not in line with Datomic,
which treats transaction inputs as a set and allows a single datom
like `[e a v]` to appear multiple times. It's both awkward and not
particularly efficient to look for _distinct_ repetitions in SQL, so
we accept some runtime cost in order to check for repetitions in the
transactor. This will allow us to address #532, which is really about
whether we treat inputs as sets. A side benefit is that we can
provide more helpful error messages when the transactor does detect
that the input truly violates the cardinality constraints of the
schema.
This commit builds a trie while error checking and collecting final
terms, which should be fairly efficient. It also allows a simpler
expression of input-provided :db/txInstant datoms, which in turn
uncovered a small issue with the transaction watcher, where-by the
watcher would not see non-input-provided :db/txInstant datoms.
This transition to Datomic-like input-as-set semantics allows us to
address #532. Previously, two tempids that upserted to the same entid
would produce duplicate datoms, and that would have been rejected by
the transactor -- correctly, since we did not allow duplicate datoms
under the input-as-list semantics. With input-as-set semantics,
duplicate datoms are allowed; and that means that we must allow
tempids to be equivalent, i.e., to resolve to the same tempid.
To achieve this, we:
- index the set of tempids
- identify tempid indices that share an upsert
- map tempids to a dense set of contiguous integer labels
We use the well-known union-find algorithm, as implemented by
petgraph, to efficiently manage the set of equivalent tempids.
Along the way, I've fixed and added tests for two small errors in the
transactor. First, don't drop datoms resolved by upsert (#679).
Second, ensure that complex upserts are allocated.
I don't know quite what happened here. The Clojure implementation
correctly kept complex upserts that hadn't resolved as complex
upserts (see
https://github.com/mozilla/mentat/blob/9a9dfb502acf5e4cdb1059d4aac831d7603063c8/src/common/datomish/transact.cljc#L436)
and then allocated complex upserts if they didn't resolve (see
https://github.com/mozilla/mentat/blob/9a9dfb502acf5e4cdb1059d4aac831d7603063c8/src/common/datomish/transact.cljc#L509).
Based on the code comments, I think the Rust implementation must have
incorrectly tried to optimize by handling all complex upserts in at
most a single generation of evolution, and that's just not correct.
We're effectively implementing a topological sort, using very specific
domain knowledge, and its not true that a node in a topological sort
can be considered only once!
2018-04-30 22:16:05 +00:00
// Insert datoms that were added and not already present. We also must expand our bitfield into
// flags. Since Mentat follows Datomic and treats its input as a set, it is okay to transact
// the same [e a v] twice in one transaction, but we don't want to represent the transacted
// datom twice in datoms. The transactor unifies repeated datoms, and in addition we add
// indices to the search inputs and search results to ensure that we don't see repeated datoms
// at this point.
2020-01-14 15:46:21 +00:00
let s = format! (
r #"
Extract partial storage abstraction; use error-chain throughout. Fixes #328. r=rnewman (#341)
* Pre: Drop unneeded tx0 from search results.
* Pre: Don't require a schema in some of the DB code.
The idea is to separate the transaction applying code, which is
schema-aware, from the concrete storage code, which is just concerned
with getting bits onto disk.
* Pre: Only reference Schema, not DB, in debug module.
This is part of a larger separation of the volatile PartitionMap,
which is modified every transaction, from the stable Schema, which is
infrequently modified.
* Pre: Fix indentation.
* Extract part of DB to new SchemaTypeChecking trait.
* Extract part of DB to new PartitionMapping trait.
* Pre: Don't expect :db.part/tx partition to advance when tx fails.
This fails right now, because we allocate tx IDs even when we shouldn't.
* Sketch a db interface without DB.
* Add ValueParseError; use error-chain in tx-parser.
This can be simplified when
https://github.com/Marwes/combine/issues/86 makes it to a published
release, but this unblocks us for now. This converts the `combine`
error type `ParseError<&'a [edn::Value]>` to a type with owned
`Vec<edn::Value>` collections, re-using `edn::Value::Vector` for
making them `Display`.
* Pre: Accept Borrow<Schema> instead of just &Schema in debug module.
This makes it easy to use Rc<Schema> or Arc<Schema> without inserting
&* sigils throughout the code.
* Use error-chain in query-parser.
There are a few things to point out here:
- the fine grained error types have been flattened into one crate-wide
error type; it's pretty easy to regain the granularity as needed.
- edn::ParseError is automatically lifted to
mentat_query_parser::errors::Error;
- we use mentat_parser_utils::ValueParser to maintain parsing error
information from `combine`.
* Patch up top-level.
* Review comment: Only `borrow()` once.
2017-02-24 23:32:41 +00:00
INSERT INTO datoms ( e , a , v , tx , value_type_tag , index_avet , index_vaet , index_fulltext , unique_value )
SELECT e0 , a0 , v0 , ? , value_type_tag0 ,
flags0 & { } IS NOT 0 ,
flags0 & { } IS NOT 0 ,
flags0 & { } IS NOT 0 ,
flags0 & { } IS NOT 0
FROM temp . search_results
WHERE added0 IS 1 AND ( ( rid IS NULL ) OR ( ( rid IS NOT NULL ) AND ( v0 IS NOT v ) ) ) " #,
2020-01-14 15:46:21 +00:00
AttributeBitFlags ::IndexAVET as u8 ,
AttributeBitFlags ::IndexVAET as u8 ,
AttributeBitFlags ::IndexFulltext as u8 ,
AttributeBitFlags ::UniqueValue as u8
) ;
Extract partial storage abstraction; use error-chain throughout. Fixes #328. r=rnewman (#341)
* Pre: Drop unneeded tx0 from search results.
* Pre: Don't require a schema in some of the DB code.
The idea is to separate the transaction applying code, which is
schema-aware, from the concrete storage code, which is just concerned
with getting bits onto disk.
* Pre: Only reference Schema, not DB, in debug module.
This is part of a larger separation of the volatile PartitionMap,
which is modified every transaction, from the stable Schema, which is
infrequently modified.
* Pre: Fix indentation.
* Extract part of DB to new SchemaTypeChecking trait.
* Extract part of DB to new PartitionMapping trait.
* Pre: Don't expect :db.part/tx partition to advance when tx fails.
This fails right now, because we allocate tx IDs even when we shouldn't.
* Sketch a db interface without DB.
* Add ValueParseError; use error-chain in tx-parser.
This can be simplified when
https://github.com/Marwes/combine/issues/86 makes it to a published
release, but this unblocks us for now. This converts the `combine`
error type `ParseError<&'a [edn::Value]>` to a type with owned
`Vec<edn::Value>` collections, re-using `edn::Value::Vector` for
making them `Display`.
* Pre: Accept Borrow<Schema> instead of just &Schema in debug module.
This makes it easy to use Rc<Schema> or Arc<Schema> without inserting
&* sigils throughout the code.
* Use error-chain in query-parser.
There are a few things to point out here:
- the fine grained error types have been flattened into one crate-wide
error type; it's pretty easy to regain the granularity as needed.
- edn::ParseError is automatically lifted to
mentat_query_parser::errors::Error;
- we use mentat_parser_utils::ValueParser to maintain parsing error
information from `combine`.
* Patch up top-level.
* Review comment: Only `borrow()` once.
2017-02-24 23:32:41 +00:00
let mut stmt = conn . prepare_cached ( & s ) ? ;
2019-07-17 15:46:47 +00:00
stmt . execute ( & [ & tx ] )
. context ( DbErrorKind ::DatomsUpdateFailedToAdd ) ? ;
Extract partial storage abstraction; use error-chain throughout. Fixes #328. r=rnewman (#341)
* Pre: Drop unneeded tx0 from search results.
* Pre: Don't require a schema in some of the DB code.
The idea is to separate the transaction applying code, which is
schema-aware, from the concrete storage code, which is just concerned
with getting bits onto disk.
* Pre: Only reference Schema, not DB, in debug module.
This is part of a larger separation of the volatile PartitionMap,
which is modified every transaction, from the stable Schema, which is
infrequently modified.
* Pre: Fix indentation.
* Extract part of DB to new SchemaTypeChecking trait.
* Extract part of DB to new PartitionMapping trait.
* Pre: Don't expect :db.part/tx partition to advance when tx fails.
This fails right now, because we allocate tx IDs even when we shouldn't.
* Sketch a db interface without DB.
* Add ValueParseError; use error-chain in tx-parser.
This can be simplified when
https://github.com/Marwes/combine/issues/86 makes it to a published
release, but this unblocks us for now. This converts the `combine`
error type `ParseError<&'a [edn::Value]>` to a type with owned
`Vec<edn::Value>` collections, re-using `edn::Value::Vector` for
making them `Display`.
* Pre: Accept Borrow<Schema> instead of just &Schema in debug module.
This makes it easy to use Rc<Schema> or Arc<Schema> without inserting
&* sigils throughout the code.
* Use error-chain in query-parser.
There are a few things to point out here:
- the fine grained error types have been flattened into one crate-wide
error type; it's pretty easy to regain the granularity as needed.
- edn::ParseError is automatically lifted to
mentat_query_parser::errors::Error;
- we use mentat_parser_utils::ValueParser to maintain parsing error
information from `combine`.
* Patch up top-level.
* Review comment: Only `borrow()` once.
2017-02-24 23:32:41 +00:00
Ok ( ( ) )
}
impl MentatStoring for rusqlite ::Connection {
fn resolve_avs < ' a > ( & self , avs : & ' a [ & ' a AVPair ] ) -> Result < AVMap < ' a > > {
2017-02-15 00:50:40 +00:00
// Start search_id's at some identifiable number.
let initial_search_id = 2000 ;
let bindings_per_statement = 4 ;
// We map [a v] -> numeric search_id -> e, and then we use the search_id lookups to finally
// produce the map [a v] -> e.
//
// TODO: `collect` into a HashSet so that any (a, v) is resolved at most once.
2017-03-13 16:39:19 +00:00
let max_vars = self . limit ( Limit ::SQLITE_LIMIT_VARIABLE_NUMBER ) as usize ;
2020-01-23 18:16:19 +00:00
let chunks : itertools ::IntoChunks < _ > = avs . iter ( ) . enumerate ( ) . chunks ( max_vars / 4 ) ;
2017-02-15 00:50:40 +00:00
// We'd like to `flat_map` here, but it's not obvious how to `flat_map` across `Result`.
// Alternatively, this is a `fold`, and it might be wise to express it as such.
let results : Result < Vec < Vec < _ > > > = chunks . into_iter ( ) . map ( | chunk | -> Result < Vec < _ > > {
let mut count = 0 ;
// We must keep these computed values somewhere to reference them later, so we can't
// combine this `map` and the subsequent `flat_map`.
let block : Vec < ( i64 , i64 , ToSqlOutput < ' a > , i32 ) > = chunk . map ( | ( index , & & ( a , ref v ) ) | {
count + = 1 ;
let search_id : i64 = initial_search_id + index as i64 ;
let ( value , value_type_tag ) = v . to_sql_value_pair ( ) ;
( search_id , a , value , value_type_tag )
} ) . collect ( ) ;
// `params` reference computed values in `block`.
2020-01-14 15:46:21 +00:00
let params : Vec < & dyn ToSql > = block . iter ( ) . flat_map ( | & ( ref searchid , ref a , ref value , ref value_type_tag ) | {
2017-02-15 00:50:40 +00:00
// Avoid inner heap allocation.
2020-01-14 15:46:21 +00:00
once ( searchid as & dyn ToSql )
. chain ( once ( a as & dyn ToSql )
. chain ( once ( value as & dyn ToSql )
. chain ( once ( value_type_tag as & dyn ToSql ) ) ) )
2017-02-15 00:50:40 +00:00
} ) . collect ( ) ;
// TODO: cache these statements for selected values of `count`.
// TODO: query against `datoms` and UNION ALL with `fulltext_datoms` rather than
// querying against `all_datoms`. We know all the attributes, and in the common case,
// where most unique attributes will not be fulltext-indexed, we'll be querying just
2017-03-13 16:39:19 +00:00
// `datoms`, which will be much faster.ˇ
assert! ( bindings_per_statement * count < max_vars , " Too many values: {} * {} >= {} " , bindings_per_statement , count , max_vars ) ;
2017-02-15 00:50:40 +00:00
let values : String = repeat_values ( bindings_per_statement , count ) ;
let s : String = format! ( " WITH t(search_id, a, v, value_type_tag) AS (VALUES {} ) SELECT t.search_id, d.e \
FROM t , all_datoms AS d \
WHERE d . index_avet IS NOT 0 AND d . a = t . a AND d . value_type_tag = t . value_type_tag AND d . v = t . v " ,
values ) ;
Extract partial storage abstraction; use error-chain throughout. Fixes #328. r=rnewman (#341)
* Pre: Drop unneeded tx0 from search results.
* Pre: Don't require a schema in some of the DB code.
The idea is to separate the transaction applying code, which is
schema-aware, from the concrete storage code, which is just concerned
with getting bits onto disk.
* Pre: Only reference Schema, not DB, in debug module.
This is part of a larger separation of the volatile PartitionMap,
which is modified every transaction, from the stable Schema, which is
infrequently modified.
* Pre: Fix indentation.
* Extract part of DB to new SchemaTypeChecking trait.
* Extract part of DB to new PartitionMapping trait.
* Pre: Don't expect :db.part/tx partition to advance when tx fails.
This fails right now, because we allocate tx IDs even when we shouldn't.
* Sketch a db interface without DB.
* Add ValueParseError; use error-chain in tx-parser.
This can be simplified when
https://github.com/Marwes/combine/issues/86 makes it to a published
release, but this unblocks us for now. This converts the `combine`
error type `ParseError<&'a [edn::Value]>` to a type with owned
`Vec<edn::Value>` collections, re-using `edn::Value::Vector` for
making them `Display`.
* Pre: Accept Borrow<Schema> instead of just &Schema in debug module.
This makes it easy to use Rc<Schema> or Arc<Schema> without inserting
&* sigils throughout the code.
* Use error-chain in query-parser.
There are a few things to point out here:
- the fine grained error types have been flattened into one crate-wide
error type; it's pretty easy to regain the granularity as needed.
- edn::ParseError is automatically lifted to
mentat_query_parser::errors::Error;
- we use mentat_parser_utils::ValueParser to maintain parsing error
information from `combine`.
* Patch up top-level.
* Review comment: Only `borrow()` once.
2017-02-24 23:32:41 +00:00
let mut stmt : rusqlite ::Statement = self . prepare ( s . as_str ( ) ) ? ;
2017-02-15 00:50:40 +00:00
let m : Result < Vec < ( i64 , Entid ) > > = stmt . query_and_then ( & params , | row | -> Result < ( i64 , Entid ) > {
2019-07-17 15:46:47 +00:00
Ok ( ( row . get ( 0 ) ? , row . get ( 1 ) ? ) )
2017-02-15 00:50:40 +00:00
} ) ? . collect ( ) ;
m
} ) . collect ::< Result < Vec < Vec < ( i64 , Entid ) > > > > ( ) ;
// Flatten.
let results : Vec < ( i64 , Entid ) > = results ? . as_slice ( ) . concat ( ) ;
// Create map [a v] -> e.
2019-07-17 15:46:47 +00:00
let m : HashMap < & ' a AVPair , Entid > = results
. into_iter ( )
. map ( | ( search_id , entid ) | {
let index : usize = ( search_id - initial_search_id ) as usize ;
( avs [ index ] , entid )
} )
. collect ( ) ;
2017-02-15 00:50:40 +00:00
Ok ( m )
}
2017-02-08 22:04:32 +00:00
/// Create empty temporary tables for search parameters and search results.
2017-12-05 15:58:24 +00:00
fn begin_tx_application ( & self ) -> Result < ( ) > {
2017-02-08 22:04:32 +00:00
// We can't do this in one shot, since we can't prepare a batch statement.
let statements = [
r # "DROP TABLE IF EXISTS temp.exact_searches"# ,
2017-02-10 20:03:49 +00:00
// Note that `flags0` is a bitfield of several flags compressed via
// `AttributeBitFlags.flags()` in the temporary search tables, later
// expanded in the `datoms` insertion.
2017-02-08 22:04:32 +00:00
r #" CREATE TABLE temp.exact_searches (
e0 INTEGER NOT NULL ,
a0 SMALLINT NOT NULL ,
v0 BLOB NOT NULL ,
value_type_tag0 SMALLINT NOT NULL ,
added0 TINYINT NOT NULL ,
2017-02-10 20:03:49 +00:00
flags0 TINYINT NOT NULL ) " #,
2017-02-08 22:04:32 +00:00
// There's no real need to split exact and inexact searches, so long as we keep things
// in the correct place and performant. Splitting has the advantage of being explicit
// and slightly easier to read, so we'll do that to start.
r # "DROP TABLE IF EXISTS temp.inexact_searches"# ,
r #" CREATE TABLE temp.inexact_searches (
e0 INTEGER NOT NULL ,
a0 SMALLINT NOT NULL ,
v0 BLOB NOT NULL ,
value_type_tag0 SMALLINT NOT NULL ,
added0 TINYINT NOT NULL ,
2017-02-10 20:03:49 +00:00
flags0 TINYINT NOT NULL ) " #,
Add type checking and constraint checking to the transactor. (#663, #532, #679)
This should address #663, by re-inserting type checking in the
transactor stack after the entry point used by the term builder.
Before this commit, we were using an SQLite UNIQUE index to assert
that no `[e a]` pair, with `a` a cardinality one attribute, was
asserted more than once. However, that's not in line with Datomic,
which treats transaction inputs as a set and allows a single datom
like `[e a v]` to appear multiple times. It's both awkward and not
particularly efficient to look for _distinct_ repetitions in SQL, so
we accept some runtime cost in order to check for repetitions in the
transactor. This will allow us to address #532, which is really about
whether we treat inputs as sets. A side benefit is that we can
provide more helpful error messages when the transactor does detect
that the input truly violates the cardinality constraints of the
schema.
This commit builds a trie while error checking and collecting final
terms, which should be fairly efficient. It also allows a simpler
expression of input-provided :db/txInstant datoms, which in turn
uncovered a small issue with the transaction watcher, where-by the
watcher would not see non-input-provided :db/txInstant datoms.
This transition to Datomic-like input-as-set semantics allows us to
address #532. Previously, two tempids that upserted to the same entid
would produce duplicate datoms, and that would have been rejected by
the transactor -- correctly, since we did not allow duplicate datoms
under the input-as-list semantics. With input-as-set semantics,
duplicate datoms are allowed; and that means that we must allow
tempids to be equivalent, i.e., to resolve to the same tempid.
To achieve this, we:
- index the set of tempids
- identify tempid indices that share an upsert
- map tempids to a dense set of contiguous integer labels
We use the well-known union-find algorithm, as implemented by
petgraph, to efficiently manage the set of equivalent tempids.
Along the way, I've fixed and added tests for two small errors in the
transactor. First, don't drop datoms resolved by upsert (#679).
Second, ensure that complex upserts are allocated.
I don't know quite what happened here. The Clojure implementation
correctly kept complex upserts that hadn't resolved as complex
upserts (see
https://github.com/mozilla/mentat/blob/9a9dfb502acf5e4cdb1059d4aac831d7603063c8/src/common/datomish/transact.cljc#L436)
and then allocated complex upserts if they didn't resolve (see
https://github.com/mozilla/mentat/blob/9a9dfb502acf5e4cdb1059d4aac831d7603063c8/src/common/datomish/transact.cljc#L509).
Based on the code comments, I think the Rust implementation must have
incorrectly tried to optimize by handling all complex upserts in at
most a single generation of evolution, and that's just not correct.
We're effectively implementing a topological sort, using very specific
domain knowledge, and its not true that a node in a topological sort
can be considered only once!
2018-04-30 22:16:05 +00:00
// It is fine to transact the same [e a v] twice in one transaction, but the transaction
// processor should unify such repeated datoms. This index will cause insertion to fail
// if the transaction processor incorrectly tries to assert the same (cardinality one)
// datom twice. (Sadly, the failure is opaque.)
2018-01-23 23:11:38 +00:00
r # "CREATE UNIQUE INDEX IF NOT EXISTS temp.inexact_searches_unique ON inexact_searches (e0, a0) WHERE added0 = 1"# ,
2017-02-08 22:04:32 +00:00
r # "DROP TABLE IF EXISTS temp.search_results"# ,
// TODO: don't encode search_type as a STRING. This is explicit and much easier to read
// than another flag, so we'll do it to start, and optimize later.
r #" CREATE TABLE temp.search_results (
e0 INTEGER NOT NULL ,
a0 SMALLINT NOT NULL ,
v0 BLOB NOT NULL ,
value_type_tag0 SMALLINT NOT NULL ,
added0 TINYINT NOT NULL ,
2017-02-10 20:03:49 +00:00
flags0 TINYINT NOT NULL ,
2017-02-08 22:04:32 +00:00
search_type STRING NOT NULL ,
rid INTEGER ,
v BLOB ) " #,
Add type checking and constraint checking to the transactor. (#663, #532, #679)
This should address #663, by re-inserting type checking in the
transactor stack after the entry point used by the term builder.
Before this commit, we were using an SQLite UNIQUE index to assert
that no `[e a]` pair, with `a` a cardinality one attribute, was
asserted more than once. However, that's not in line with Datomic,
which treats transaction inputs as a set and allows a single datom
like `[e a v]` to appear multiple times. It's both awkward and not
particularly efficient to look for _distinct_ repetitions in SQL, so
we accept some runtime cost in order to check for repetitions in the
transactor. This will allow us to address #532, which is really about
whether we treat inputs as sets. A side benefit is that we can
provide more helpful error messages when the transactor does detect
that the input truly violates the cardinality constraints of the
schema.
This commit builds a trie while error checking and collecting final
terms, which should be fairly efficient. It also allows a simpler
expression of input-provided :db/txInstant datoms, which in turn
uncovered a small issue with the transaction watcher, where-by the
watcher would not see non-input-provided :db/txInstant datoms.
This transition to Datomic-like input-as-set semantics allows us to
address #532. Previously, two tempids that upserted to the same entid
would produce duplicate datoms, and that would have been rejected by
the transactor -- correctly, since we did not allow duplicate datoms
under the input-as-list semantics. With input-as-set semantics,
duplicate datoms are allowed; and that means that we must allow
tempids to be equivalent, i.e., to resolve to the same tempid.
To achieve this, we:
- index the set of tempids
- identify tempid indices that share an upsert
- map tempids to a dense set of contiguous integer labels
We use the well-known union-find algorithm, as implemented by
petgraph, to efficiently manage the set of equivalent tempids.
Along the way, I've fixed and added tests for two small errors in the
transactor. First, don't drop datoms resolved by upsert (#679).
Second, ensure that complex upserts are allocated.
I don't know quite what happened here. The Clojure implementation
correctly kept complex upserts that hadn't resolved as complex
upserts (see
https://github.com/mozilla/mentat/blob/9a9dfb502acf5e4cdb1059d4aac831d7603063c8/src/common/datomish/transact.cljc#L436)
and then allocated complex upserts if they didn't resolve (see
https://github.com/mozilla/mentat/blob/9a9dfb502acf5e4cdb1059d4aac831d7603063c8/src/common/datomish/transact.cljc#L509).
Based on the code comments, I think the Rust implementation must have
incorrectly tried to optimize by handling all complex upserts in at
most a single generation of evolution, and that's just not correct.
We're effectively implementing a topological sort, using very specific
domain knowledge, and its not true that a node in a topological sort
can be considered only once!
2018-04-30 22:16:05 +00:00
// It is fine to transact the same [e a v] twice in one transaction, but the transaction
// processor should identify those datoms. This index will cause insertion to fail if
// the internals of the database searching code incorrectly find the same datom twice.
// (Sadly, the failure is opaque.)
2017-02-08 22:04:32 +00:00
//
// N.b.: temp goes on index name, not table name. See http://stackoverflow.com/a/22308016.
r # "CREATE UNIQUE INDEX IF NOT EXISTS temp.search_results_unique ON search_results (e0, a0, v0, value_type_tag0)"# ,
] ;
for statement in & statements {
Extract partial storage abstraction; use error-chain throughout. Fixes #328. r=rnewman (#341)
* Pre: Drop unneeded tx0 from search results.
* Pre: Don't require a schema in some of the DB code.
The idea is to separate the transaction applying code, which is
schema-aware, from the concrete storage code, which is just concerned
with getting bits onto disk.
* Pre: Only reference Schema, not DB, in debug module.
This is part of a larger separation of the volatile PartitionMap,
which is modified every transaction, from the stable Schema, which is
infrequently modified.
* Pre: Fix indentation.
* Extract part of DB to new SchemaTypeChecking trait.
* Extract part of DB to new PartitionMapping trait.
* Pre: Don't expect :db.part/tx partition to advance when tx fails.
This fails right now, because we allocate tx IDs even when we shouldn't.
* Sketch a db interface without DB.
* Add ValueParseError; use error-chain in tx-parser.
This can be simplified when
https://github.com/Marwes/combine/issues/86 makes it to a published
release, but this unblocks us for now. This converts the `combine`
error type `ParseError<&'a [edn::Value]>` to a type with owned
`Vec<edn::Value>` collections, re-using `edn::Value::Vector` for
making them `Display`.
* Pre: Accept Borrow<Schema> instead of just &Schema in debug module.
This makes it easy to use Rc<Schema> or Arc<Schema> without inserting
&* sigils throughout the code.
* Use error-chain in query-parser.
There are a few things to point out here:
- the fine grained error types have been flattened into one crate-wide
error type; it's pretty easy to regain the granularity as needed.
- edn::ParseError is automatically lifted to
mentat_query_parser::errors::Error;
- we use mentat_parser_utils::ValueParser to maintain parsing error
information from `combine`.
* Patch up top-level.
* Review comment: Only `borrow()` once.
2017-02-24 23:32:41 +00:00
let mut stmt = self . prepare_cached ( statement ) ? ;
2019-07-22 11:40:59 +00:00
stmt . execute ( rusqlite ::params! [ ] )
2019-07-17 15:46:47 +00:00
. context ( DbErrorKind ::FailedToCreateTempTables ) ? ;
2017-02-08 22:04:32 +00:00
}
Ok ( ( ) )
}
/// Insert search rows into temporary search tables.
///
/// Eventually, the details of this approach will be captured in
/// https://github.com/mozilla/mentat/wiki/Transacting:-entity-to-SQL-translation.
2019-07-17 15:46:47 +00:00
fn insert_non_fts_searches < ' a > (
& self ,
entities : & ' a [ ReducedEntity < ' a > ] ,
search_type : SearchType ,
) -> Result < ( ) > {
Extract partial storage abstraction; use error-chain throughout. Fixes #328. r=rnewman (#341)
* Pre: Drop unneeded tx0 from search results.
* Pre: Don't require a schema in some of the DB code.
The idea is to separate the transaction applying code, which is
schema-aware, from the concrete storage code, which is just concerned
with getting bits onto disk.
* Pre: Only reference Schema, not DB, in debug module.
This is part of a larger separation of the volatile PartitionMap,
which is modified every transaction, from the stable Schema, which is
infrequently modified.
* Pre: Fix indentation.
* Extract part of DB to new SchemaTypeChecking trait.
* Extract part of DB to new PartitionMapping trait.
* Pre: Don't expect :db.part/tx partition to advance when tx fails.
This fails right now, because we allocate tx IDs even when we shouldn't.
* Sketch a db interface without DB.
* Add ValueParseError; use error-chain in tx-parser.
This can be simplified when
https://github.com/Marwes/combine/issues/86 makes it to a published
release, but this unblocks us for now. This converts the `combine`
error type `ParseError<&'a [edn::Value]>` to a type with owned
`Vec<edn::Value>` collections, re-using `edn::Value::Vector` for
making them `Display`.
* Pre: Accept Borrow<Schema> instead of just &Schema in debug module.
This makes it easy to use Rc<Schema> or Arc<Schema> without inserting
&* sigils throughout the code.
* Use error-chain in query-parser.
There are a few things to point out here:
- the fine grained error types have been flattened into one crate-wide
error type; it's pretty easy to regain the granularity as needed.
- edn::ParseError is automatically lifted to
mentat_query_parser::errors::Error;
- we use mentat_parser_utils::ValueParser to maintain parsing error
information from `combine`.
* Patch up top-level.
* Review comment: Only `borrow()` once.
2017-02-24 23:32:41 +00:00
let bindings_per_statement = 6 ;
2017-02-08 22:04:32 +00:00
2017-03-13 16:39:19 +00:00
let max_vars = self . limit ( Limit ::SQLITE_LIMIT_VARIABLE_NUMBER ) as usize ;
2020-01-23 18:16:19 +00:00
let chunks : itertools ::IntoChunks < _ > =
entities . iter ( ) . chunks ( max_vars / bindings_per_statement ) ;
2017-02-08 22:04:32 +00:00
// We'd like to flat_map here, but it's not obvious how to flat_map across Result.
let results : Result < Vec < ( ) > > = chunks . into_iter ( ) . map ( | chunk | -> Result < ( ) > {
let mut count = 0 ;
// We must keep these computed values somewhere to reference them later, so we can't
// combine this map and the subsequent flat_map.
2017-02-10 20:03:49 +00:00
// (e0, a0, v0, value_type_tag0, added0, flags0)
2020-08-06 03:03:58 +00:00
#[ allow(clippy::type_complexity) ]
Extract partial storage abstraction; use error-chain throughout. Fixes #328. r=rnewman (#341)
* Pre: Drop unneeded tx0 from search results.
* Pre: Don't require a schema in some of the DB code.
The idea is to separate the transaction applying code, which is
schema-aware, from the concrete storage code, which is just concerned
with getting bits onto disk.
* Pre: Only reference Schema, not DB, in debug module.
This is part of a larger separation of the volatile PartitionMap,
which is modified every transaction, from the stable Schema, which is
infrequently modified.
* Pre: Fix indentation.
* Extract part of DB to new SchemaTypeChecking trait.
* Extract part of DB to new PartitionMapping trait.
* Pre: Don't expect :db.part/tx partition to advance when tx fails.
This fails right now, because we allocate tx IDs even when we shouldn't.
* Sketch a db interface without DB.
* Add ValueParseError; use error-chain in tx-parser.
This can be simplified when
https://github.com/Marwes/combine/issues/86 makes it to a published
release, but this unblocks us for now. This converts the `combine`
error type `ParseError<&'a [edn::Value]>` to a type with owned
`Vec<edn::Value>` collections, re-using `edn::Value::Vector` for
making them `Display`.
* Pre: Accept Borrow<Schema> instead of just &Schema in debug module.
This makes it easy to use Rc<Schema> or Arc<Schema> without inserting
&* sigils throughout the code.
* Use error-chain in query-parser.
There are a few things to point out here:
- the fine grained error types have been flattened into one crate-wide
error type; it's pretty easy to regain the granularity as needed.
- edn::ParseError is automatically lifted to
mentat_query_parser::errors::Error;
- we use mentat_parser_utils::ValueParser to maintain parsing error
information from `combine`.
* Patch up top-level.
* Review comment: Only `borrow()` once.
2017-02-24 23:32:41 +00:00
let block : Result < Vec < ( i64 /* e */ ,
i64 /* a */ ,
ToSqlOutput < ' a > /* value */ ,
i32 /* value_type_tag */ ,
bool , /* added0 */
u8 /* flags0 */ ) > > = chunk . map ( | & ( e , a , ref attribute , ref typed_value , added ) | {
2017-02-08 22:04:32 +00:00
count + = 1 ;
// Now we can represent the typed value as an SQL value.
let ( value , value_type_tag ) : ( ToSqlOutput , i32 ) = typed_value . to_sql_value_pair ( ) ;
Extract partial storage abstraction; use error-chain throughout. Fixes #328. r=rnewman (#341)
* Pre: Drop unneeded tx0 from search results.
* Pre: Don't require a schema in some of the DB code.
The idea is to separate the transaction applying code, which is
schema-aware, from the concrete storage code, which is just concerned
with getting bits onto disk.
* Pre: Only reference Schema, not DB, in debug module.
This is part of a larger separation of the volatile PartitionMap,
which is modified every transaction, from the stable Schema, which is
infrequently modified.
* Pre: Fix indentation.
* Extract part of DB to new SchemaTypeChecking trait.
* Extract part of DB to new PartitionMapping trait.
* Pre: Don't expect :db.part/tx partition to advance when tx fails.
This fails right now, because we allocate tx IDs even when we shouldn't.
* Sketch a db interface without DB.
* Add ValueParseError; use error-chain in tx-parser.
This can be simplified when
https://github.com/Marwes/combine/issues/86 makes it to a published
release, but this unblocks us for now. This converts the `combine`
error type `ParseError<&'a [edn::Value]>` to a type with owned
`Vec<edn::Value>` collections, re-using `edn::Value::Vector` for
making them `Display`.
* Pre: Accept Borrow<Schema> instead of just &Schema in debug module.
This makes it easy to use Rc<Schema> or Arc<Schema> without inserting
&* sigils throughout the code.
* Use error-chain in query-parser.
There are a few things to point out here:
- the fine grained error types have been flattened into one crate-wide
error type; it's pretty easy to regain the granularity as needed.
- edn::ParseError is automatically lifted to
mentat_query_parser::errors::Error;
- we use mentat_parser_utils::ValueParser to maintain parsing error
information from `combine`.
* Patch up top-level.
* Review comment: Only `borrow()` once.
2017-02-24 23:32:41 +00:00
Ok ( ( e , a , value , value_type_tag , added , attribute . flags ( ) ) )
2017-02-08 22:04:32 +00:00
} ) . collect ( ) ;
let block = block ? ;
// `params` reference computed values in `block`.
2020-01-14 15:46:21 +00:00
let params : Vec < & dyn ToSql > = block . iter ( ) . flat_map ( | & ( ref e , ref a , ref value , ref value_type_tag , added , ref flags ) | {
2017-02-08 22:04:32 +00:00
// Avoid inner heap allocation.
// TODO: extract some finite length iterator to make this less indented!
2020-01-14 15:46:21 +00:00
once ( e as & dyn ToSql )
. chain ( once ( a as & dyn ToSql )
. chain ( once ( value as & dyn ToSql )
. chain ( once ( value_type_tag as & dyn ToSql )
. chain ( once ( to_bool_ref ( added ) as & dyn ToSql )
. chain ( once ( flags as & dyn ToSql ) ) ) ) ) )
2017-02-08 22:04:32 +00:00
} ) . collect ( ) ;
Lookup refs, nested vector values, map notation. Fixes #180, fixes #183, fixes #284. (#382) r=rnewman
* Pre: Fix error in parser macros.
* Pre: Make test unwrapping more verbose.
* Pre: Make lookup refs be (lookup-ref a v) in the entity position.
This has the advantage of being explicit in all situations and
unambiguous at parse-time. This choice agrees with the Clojure
implementation but not with Datomic. Datomic treats [a v] as a lookup
ref, is ambiguous at parse-time, and is disambiguated in ways I do not
understand at transaction time. We mooted making lookup refs [[a v]]
and outlawing nested value vectors in transactions, but after
implementing that approach I decided it was better to handle lookup
refs at parse time and therefore outlawing nested value vectors is not
necessary.
* Handle lookup refs in the entity and value columns. Fixes #183.
* Pre 0a: Use a stack instead of into_iter.
* Pre 0b: Dedent.
* Pre 0c: Handle `e` after `v`.
This allows to use the original `e` while handling `v`.
* Explode value lists for :db.cardinality/many attributes. Fixes #284.
* Parse and accept map notation. Fixes #180.
* Pre: Modernize add() and retract() into one add_or_retract().
* Pre: Add is_collection and is_atom to edn::Value.
* Pre: Differentiate atoms from lookup-refs in value position.
Initially, I expected to accept arbitrary edn::Value instances in the
value position, and to differentiate in the transactor. However, the
implementation quickly became a two-stage parser, since we always
wanted to parse the resulting value position into some other known
thing using the tx-parser. To save calls into the parser and to allow
the parser to move forward with a smaller API surface, I push as much
of this parsing as possible into the initial parse.
* Pre: Modernize entities().
* Pre: Quote edn::Value::Text in Display.
* Review comment: Add and use edn::Value::into_atom.
* Review comment: Use skip(eof()) throughout.
* Review comment: VecDeque instead of Vec.
* Review comment: Part 0: Rename TempId to TempIdHandle.
* Review comment: Part 1: Differentiate internal and external tempids.
This breaks an abstraction boundary by pushing the Internal/External
split up to the Entity level in tx/ and tx-parser/. This just makes
it easier to explode Entity map notation instances into Entity
instances, taking an existing External tempid :db/id or generating a
new Internal tempid as appropriate. To do this without breaking the
abstraction boundary would require adding flexibility to the
transaction processor: we'd need to be able to turn Entity instances
into some internal enum and handle the two cases independently. It
wouldn't be too hard, but this reduces the combinatorial type
explosion.
2017-03-27 23:30:04 +00:00
// TODO: cache this for selected values of count.
2017-03-13 16:39:19 +00:00
assert! ( bindings_per_statement * count < max_vars , " Too many values: {} * {} >= {} " , bindings_per_statement , count , max_vars ) ;
2017-02-08 22:04:32 +00:00
let values : String = repeat_values ( bindings_per_statement , count ) ;
let s : String = if search_type = = SearchType ::Exact {
Extract partial storage abstraction; use error-chain throughout. Fixes #328. r=rnewman (#341)
* Pre: Drop unneeded tx0 from search results.
* Pre: Don't require a schema in some of the DB code.
The idea is to separate the transaction applying code, which is
schema-aware, from the concrete storage code, which is just concerned
with getting bits onto disk.
* Pre: Only reference Schema, not DB, in debug module.
This is part of a larger separation of the volatile PartitionMap,
which is modified every transaction, from the stable Schema, which is
infrequently modified.
* Pre: Fix indentation.
* Extract part of DB to new SchemaTypeChecking trait.
* Extract part of DB to new PartitionMapping trait.
* Pre: Don't expect :db.part/tx partition to advance when tx fails.
This fails right now, because we allocate tx IDs even when we shouldn't.
* Sketch a db interface without DB.
* Add ValueParseError; use error-chain in tx-parser.
This can be simplified when
https://github.com/Marwes/combine/issues/86 makes it to a published
release, but this unblocks us for now. This converts the `combine`
error type `ParseError<&'a [edn::Value]>` to a type with owned
`Vec<edn::Value>` collections, re-using `edn::Value::Vector` for
making them `Display`.
* Pre: Accept Borrow<Schema> instead of just &Schema in debug module.
This makes it easy to use Rc<Schema> or Arc<Schema> without inserting
&* sigils throughout the code.
* Use error-chain in query-parser.
There are a few things to point out here:
- the fine grained error types have been flattened into one crate-wide
error type; it's pretty easy to regain the granularity as needed.
- edn::ParseError is automatically lifted to
mentat_query_parser::errors::Error;
- we use mentat_parser_utils::ValueParser to maintain parsing error
information from `combine`.
* Patch up top-level.
* Review comment: Only `borrow()` once.
2017-02-24 23:32:41 +00:00
format! ( " INSERT INTO temp.exact_searches (e0, a0, v0, value_type_tag0, added0, flags0) VALUES {} " , values )
2017-02-08 22:04:32 +00:00
} else {
2018-01-23 23:11:38 +00:00
// This will err for duplicates within the tx.
Extract partial storage abstraction; use error-chain throughout. Fixes #328. r=rnewman (#341)
* Pre: Drop unneeded tx0 from search results.
* Pre: Don't require a schema in some of the DB code.
The idea is to separate the transaction applying code, which is
schema-aware, from the concrete storage code, which is just concerned
with getting bits onto disk.
* Pre: Only reference Schema, not DB, in debug module.
This is part of a larger separation of the volatile PartitionMap,
which is modified every transaction, from the stable Schema, which is
infrequently modified.
* Pre: Fix indentation.
* Extract part of DB to new SchemaTypeChecking trait.
* Extract part of DB to new PartitionMapping trait.
* Pre: Don't expect :db.part/tx partition to advance when tx fails.
This fails right now, because we allocate tx IDs even when we shouldn't.
* Sketch a db interface without DB.
* Add ValueParseError; use error-chain in tx-parser.
This can be simplified when
https://github.com/Marwes/combine/issues/86 makes it to a published
release, but this unblocks us for now. This converts the `combine`
error type `ParseError<&'a [edn::Value]>` to a type with owned
`Vec<edn::Value>` collections, re-using `edn::Value::Vector` for
making them `Display`.
* Pre: Accept Borrow<Schema> instead of just &Schema in debug module.
This makes it easy to use Rc<Schema> or Arc<Schema> without inserting
&* sigils throughout the code.
* Use error-chain in query-parser.
There are a few things to point out here:
- the fine grained error types have been flattened into one crate-wide
error type; it's pretty easy to regain the granularity as needed.
- edn::ParseError is automatically lifted to
mentat_query_parser::errors::Error;
- we use mentat_parser_utils::ValueParser to maintain parsing error
information from `combine`.
* Patch up top-level.
* Review comment: Only `borrow()` once.
2017-02-24 23:32:41 +00:00
format! ( " INSERT INTO temp.inexact_searches (e0, a0, v0, value_type_tag0, added0, flags0) VALUES {} " , values )
2017-02-08 22:04:32 +00:00
} ;
// TODO: consider ensuring we inserted the expected number of rows.
Extract partial storage abstraction; use error-chain throughout. Fixes #328. r=rnewman (#341)
* Pre: Drop unneeded tx0 from search results.
* Pre: Don't require a schema in some of the DB code.
The idea is to separate the transaction applying code, which is
schema-aware, from the concrete storage code, which is just concerned
with getting bits onto disk.
* Pre: Only reference Schema, not DB, in debug module.
This is part of a larger separation of the volatile PartitionMap,
which is modified every transaction, from the stable Schema, which is
infrequently modified.
* Pre: Fix indentation.
* Extract part of DB to new SchemaTypeChecking trait.
* Extract part of DB to new PartitionMapping trait.
* Pre: Don't expect :db.part/tx partition to advance when tx fails.
This fails right now, because we allocate tx IDs even when we shouldn't.
* Sketch a db interface without DB.
* Add ValueParseError; use error-chain in tx-parser.
This can be simplified when
https://github.com/Marwes/combine/issues/86 makes it to a published
release, but this unblocks us for now. This converts the `combine`
error type `ParseError<&'a [edn::Value]>` to a type with owned
`Vec<edn::Value>` collections, re-using `edn::Value::Vector` for
making them `Display`.
* Pre: Accept Borrow<Schema> instead of just &Schema in debug module.
This makes it easy to use Rc<Schema> or Arc<Schema> without inserting
&* sigils throughout the code.
* Use error-chain in query-parser.
There are a few things to point out here:
- the fine grained error types have been flattened into one crate-wide
error type; it's pretty easy to regain the granularity as needed.
- edn::ParseError is automatically lifted to
mentat_query_parser::errors::Error;
- we use mentat_parser_utils::ValueParser to maintain parsing error
information from `combine`.
* Patch up top-level.
* Review comment: Only `borrow()` once.
2017-02-24 23:32:41 +00:00
let mut stmt = self . prepare_cached ( s . as_str ( ) ) ? ;
2017-02-08 22:04:32 +00:00
stmt . execute ( & params )
2018-06-27 00:17:01 +00:00
. context ( DbErrorKind ::NonFtsInsertionIntoTempSearchTableFailed )
2018-06-06 01:23:59 +00:00
. map_err ( | e | e . into ( ) )
2017-02-08 22:04:32 +00:00
. map ( | _c | ( ) )
} ) . collect ::< Result < Vec < ( ) > > > ( ) ;
results . map ( | _ | ( ) )
}
2017-03-21 20:12:10 +00:00
/// Insert search rows into temporary search tables.
///
/// Eventually, the details of this approach will be captured in
/// https://github.com/mozilla/mentat/wiki/Transacting:-entity-to-SQL-translation.
2019-07-17 15:46:47 +00:00
fn insert_fts_searches < ' a > (
& self ,
entities : & ' a [ ReducedEntity < ' a > ] ,
search_type : SearchType ,
) -> Result < ( ) > {
2017-03-21 20:12:10 +00:00
let max_vars = self . limit ( Limit ::SQLITE_LIMIT_VARIABLE_NUMBER ) as usize ;
let bindings_per_statement = 6 ;
let mut outer_searchid = 2000 ;
2020-01-23 18:16:19 +00:00
let chunks : itertools ::IntoChunks < _ > =
entities . iter ( ) . chunks ( max_vars / bindings_per_statement ) ;
2017-03-21 20:12:10 +00:00
2018-02-21 03:52:49 +00:00
// From string to (searchid, value_type_tag).
2018-04-25 21:23:27 +00:00
let mut seen : HashMap < ValueRc < String > , ( i64 , i32 ) > = HashMap ::with_capacity ( entities . len ( ) ) ;
2018-02-21 03:52:49 +00:00
2017-03-21 20:12:10 +00:00
// We'd like to flat_map here, but it's not obvious how to flat_map across Result.
let results : Result < Vec < ( ) > > = chunks . into_iter ( ) . map ( | chunk | -> Result < ( ) > {
2018-02-21 03:52:49 +00:00
let mut datom_count = 0 ;
let mut string_count = 0 ;
2017-03-21 20:12:10 +00:00
// We must keep these computed values somewhere to reference them later, so we can't
// combine this map and the subsequent flat_map.
// (e0, a0, v0, value_type_tag0, added0, flags0)
2020-08-06 03:03:58 +00:00
#[ allow(clippy::type_complexity) ]
2017-03-21 20:12:10 +00:00
let block : Result < Vec < ( i64 /* e */ ,
i64 /* a */ ,
2018-02-21 03:52:49 +00:00
Option < ToSqlOutput < ' a > > /* value */ ,
2017-03-21 20:12:10 +00:00
i32 /* value_type_tag */ ,
bool /* added0 */ ,
u8 /* flags0 */ ,
i64 /* searchid */ ) > > = chunk . map ( | & ( e , a , ref attribute , ref typed_value , added ) | {
2018-02-21 03:52:49 +00:00
match typed_value {
2020-01-23 18:16:19 +00:00
TypedValue ::String ( ref rc ) = > {
2018-02-21 03:52:49 +00:00
datom_count + = 1 ;
let entry = seen . entry ( rc . clone ( ) ) ;
match entry {
Entry ::Occupied ( entry ) = > {
let & ( searchid , value_type_tag ) = entry . get ( ) ;
Ok ( ( e , a , None , value_type_tag , added , attribute . flags ( ) , searchid ) )
} ,
Entry ::Vacant ( entry ) = > {
outer_searchid + = 1 ;
string_count + = 1 ;
// Now we can represent the typed value as an SQL value.
let ( value , value_type_tag ) : ( ToSqlOutput , i32 ) = typed_value . to_sql_value_pair ( ) ;
entry . insert ( ( outer_searchid , value_type_tag ) ) ;
Ok ( ( e , a , Some ( value ) , value_type_tag , added , attribute . flags ( ) , outer_searchid ) )
}
}
} ,
_ = > {
2018-06-27 00:17:01 +00:00
bail! ( DbErrorKind ::WrongTypeValueForFtsAssertion ) ;
2018-02-21 03:52:49 +00:00
} ,
2017-03-21 20:12:10 +00:00
}
} ) . collect ( ) ;
let block = block ? ;
// First, insert all fulltext string values.
// `fts_params` reference computed values in `block`.
2020-01-14 15:46:21 +00:00
let fts_params : Vec < & dyn ToSql > =
2018-02-14 00:51:21 +00:00
block . iter ( )
. filter ( | & & ( ref _e , ref _a , ref value , ref _value_type_tag , _added , ref _flags , ref _searchid ) | {
value . is_some ( )
} )
. flat_map ( | & ( ref _e , ref _a , ref value , ref _value_type_tag , _added , ref _flags , ref searchid ) | {
// Avoid inner heap allocation.
2020-01-14 15:46:21 +00:00
once ( value as & dyn ToSql )
. chain ( once ( searchid as & dyn ToSql ) )
2018-02-14 00:51:21 +00:00
} ) . collect ( ) ;
2017-03-21 20:12:10 +00:00
2018-02-21 03:52:49 +00:00
// TODO: make this maximally efficient. It's not terribly inefficient right now.
let fts_values : String = repeat_values ( 2 , string_count ) ;
2017-03-21 20:12:10 +00:00
let fts_s : String = format! ( " INSERT INTO fulltext_values_view (text, searchid) VALUES {} " , fts_values ) ;
// TODO: consider ensuring we inserted the expected number of rows.
let mut stmt = self . prepare_cached ( fts_s . as_str ( ) ) ? ;
2018-06-27 00:17:01 +00:00
stmt . execute ( & fts_params ) . context ( DbErrorKind ::FtsInsertionFailed ) ? ;
2017-03-21 20:12:10 +00:00
// Second, insert searches.
// `params` reference computed values in `block`.
2020-01-14 15:46:21 +00:00
let params : Vec < & dyn ToSql > = block . iter ( ) . flat_map ( | & ( ref e , ref a , ref _value , ref value_type_tag , added , ref flags , ref searchid ) | {
2017-03-21 20:12:10 +00:00
// Avoid inner heap allocation.
// TODO: extract some finite length iterator to make this less indented!
2020-01-14 15:46:21 +00:00
once ( e as & dyn ToSql )
. chain ( once ( a as & dyn ToSql )
. chain ( once ( searchid as & dyn ToSql )
. chain ( once ( value_type_tag as & dyn ToSql )
. chain ( once ( to_bool_ref ( added ) as & dyn ToSql )
. chain ( once ( flags as & dyn ToSql ) ) ) ) ) )
2017-03-21 20:12:10 +00:00
} ) . collect ( ) ;
// TODO: cache this for selected values of count.
2018-02-21 03:52:49 +00:00
assert! ( bindings_per_statement * datom_count < max_vars , " Too many values: {} * {} >= {} " , bindings_per_statement , datom_count , max_vars ) ;
2017-03-21 20:12:10 +00:00
let inner = " (?, ?, (SELECT rowid FROM fulltext_values WHERE searchid = ?), ?, ?, ?) " . to_string ( ) ;
// Like "(?, ?, (SELECT rowid FROM fulltext_values WHERE searchid = ?), ?, ?, ?), (?, ?, (SELECT rowid FROM fulltext_values WHERE searchid = ?), ?, ?, ?)".
2018-02-21 03:52:49 +00:00
let fts_values : String = repeat ( inner ) . take ( datom_count ) . join ( " , " ) ;
2017-03-21 20:12:10 +00:00
let s : String = if search_type = = SearchType ::Exact {
format! ( " INSERT INTO temp.exact_searches (e0, a0, v0, value_type_tag0, added0, flags0) VALUES {} " , fts_values )
} else {
format! ( " INSERT INTO temp.inexact_searches (e0, a0, v0, value_type_tag0, added0, flags0) VALUES {} " , fts_values )
} ;
// TODO: consider ensuring we inserted the expected number of rows.
let mut stmt = self . prepare_cached ( s . as_str ( ) ) ? ;
2018-06-27 00:17:01 +00:00
stmt . execute ( & params ) . context ( DbErrorKind ::FtsInsertionIntoTempSearchTableFailed )
2018-06-06 01:23:59 +00:00
. map_err ( | e | e . into ( ) )
2017-03-21 20:12:10 +00:00
. map ( | _c | ( ) )
} ) . collect ::< Result < Vec < ( ) > > > ( ) ;
// Finally, clean up temporary searchids.
2019-07-17 15:46:47 +00:00
let mut stmt = self . prepare_cached (
" UPDATE fulltext_values SET searchid = NULL WHERE searchid IS NOT NULL " ,
) ? ;
2019-07-22 11:40:59 +00:00
stmt . execute ( rusqlite ::params! [ ] )
2019-07-17 15:46:47 +00:00
. context ( DbErrorKind ::FtsFailedToDropSearchIds ) ? ;
2017-03-21 20:12:10 +00:00
results . map ( | _ | ( ) )
}
2018-07-17 19:18:53 +00:00
fn commit_mentat_transaction ( & self , tx_id : Entid ) -> Result < ( ) > {
Extract partial storage abstraction; use error-chain throughout. Fixes #328. r=rnewman (#341)
* Pre: Drop unneeded tx0 from search results.
* Pre: Don't require a schema in some of the DB code.
The idea is to separate the transaction applying code, which is
schema-aware, from the concrete storage code, which is just concerned
with getting bits onto disk.
* Pre: Only reference Schema, not DB, in debug module.
This is part of a larger separation of the volatile PartitionMap,
which is modified every transaction, from the stable Schema, which is
infrequently modified.
* Pre: Fix indentation.
* Extract part of DB to new SchemaTypeChecking trait.
* Extract part of DB to new PartitionMapping trait.
* Pre: Don't expect :db.part/tx partition to advance when tx fails.
This fails right now, because we allocate tx IDs even when we shouldn't.
* Sketch a db interface without DB.
* Add ValueParseError; use error-chain in tx-parser.
This can be simplified when
https://github.com/Marwes/combine/issues/86 makes it to a published
release, but this unblocks us for now. This converts the `combine`
error type `ParseError<&'a [edn::Value]>` to a type with owned
`Vec<edn::Value>` collections, re-using `edn::Value::Vector` for
making them `Display`.
* Pre: Accept Borrow<Schema> instead of just &Schema in debug module.
This makes it easy to use Rc<Schema> or Arc<Schema> without inserting
&* sigils throughout the code.
* Use error-chain in query-parser.
There are a few things to point out here:
- the fine grained error types have been flattened into one crate-wide
error type; it's pretty easy to regain the granularity as needed.
- edn::ParseError is automatically lifted to
mentat_query_parser::errors::Error;
- we use mentat_parser_utils::ValueParser to maintain parsing error
information from `combine`.
* Patch up top-level.
* Review comment: Only `borrow()` once.
2017-02-24 23:32:41 +00:00
insert_transaction ( & self , tx_id ) ? ;
2018-07-17 19:18:53 +00:00
Ok ( ( ) )
}
fn materialize_mentat_transaction ( & self , tx_id : Entid ) -> Result < ( ) > {
search ( & self ) ? ;
Extract partial storage abstraction; use error-chain throughout. Fixes #328. r=rnewman (#341)
* Pre: Drop unneeded tx0 from search results.
* Pre: Don't require a schema in some of the DB code.
The idea is to separate the transaction applying code, which is
schema-aware, from the concrete storage code, which is just concerned
with getting bits onto disk.
* Pre: Only reference Schema, not DB, in debug module.
This is part of a larger separation of the volatile PartitionMap,
which is modified every transaction, from the stable Schema, which is
infrequently modified.
* Pre: Fix indentation.
* Extract part of DB to new SchemaTypeChecking trait.
* Extract part of DB to new PartitionMapping trait.
* Pre: Don't expect :db.part/tx partition to advance when tx fails.
This fails right now, because we allocate tx IDs even when we shouldn't.
* Sketch a db interface without DB.
* Add ValueParseError; use error-chain in tx-parser.
This can be simplified when
https://github.com/Marwes/combine/issues/86 makes it to a published
release, but this unblocks us for now. This converts the `combine`
error type `ParseError<&'a [edn::Value]>` to a type with owned
`Vec<edn::Value>` collections, re-using `edn::Value::Vector` for
making them `Display`.
* Pre: Accept Borrow<Schema> instead of just &Schema in debug module.
This makes it easy to use Rc<Schema> or Arc<Schema> without inserting
&* sigils throughout the code.
* Use error-chain in query-parser.
There are a few things to point out here:
- the fine grained error types have been flattened into one crate-wide
error type; it's pretty easy to regain the granularity as needed.
- edn::ParseError is automatically lifted to
mentat_query_parser::errors::Error;
- we use mentat_parser_utils::ValueParser to maintain parsing error
information from `combine`.
* Patch up top-level.
* Review comment: Only `borrow()` once.
2017-02-24 23:32:41 +00:00
update_datoms ( & self , tx_id ) ? ;
2017-02-08 22:04:32 +00:00
Ok ( ( ) )
}
Schema alteration. Fixes #294 and #295. (#370) r=rnewman
* Pre: Don't retract :db/ident in test.
Datomic (and eventually Mentat) don't allow to retract :db/ident in
this way, so this runs afoul of future work to support mutating
metadata.
* Pre: s/VALUETYPE/VALUE_TYPE/.
This is consistent with the capitalization (which is "valueType") and
the other identifier.
* Pre: Remove some single quotes from error output.
* Part 1: Make materialized views be uniform [e a v value_type_tag].
This looks ahead to a time when we could support arbitrary
user-defined materialized views. For now, the "idents" materialized
view is those datoms of the form [e :db/ident :namespaced/keyword] and
the "schema" materialized view is those datoms of the form [e a v]
where a is in a particular set of attributes that will become clear in
the following commits.
This change is not backwards compatible, so I'm removing the open
current (really, v2) test. It'll be re-instated when we get to
https://github.com/mozilla/mentat/issues/194.
* Pre: Map TypedValue::Ref to TypedValue::Keyword in debug output.
* Part 3: Separate `schema_to_mutate` from the `schema` used to interpret.
This is just to keep track of the expected changes during
bootstrapping. I want bootstrap metadata mutations to flow through
the same code path as metadata mutations during regular transactions;
by differentiating the schema used for interpretation from the schema
that will be updated I expect to be able to apply bootstrap metadata
mutations to an empty schema and have things like materialized views
created (using the regular code paths).
This commit has been re-ordered for conceptual clarity, but it won't
compile because it references the metadata module. It's possible to
make it compile -- the functionality is there in the schema module --
but it's not worth the rebasing effort until after review (and
possibly not even then, since we'll squash down to a single commit to
land).
* Part 2: Maintain entids separately from idents.
In order to support historical idents, we need to distinguish the
"current" map from entid -> ident from the "complete historical" map
ident -> entid. This is what Datomic does; in Datomic, an ident is
never retracted (although it can be replaced). This approach is an
important part of allowing multiple consumers to share a schema
fragment as it migrates forward.
This fixes a limitation of the Clojure implementation, which did not
handle historical idents across knowledge base close and re-open.
The "entids" materialized view is naturally a slice of the "datoms"
table. The "idents" materialized view is a slice of the
"transactions" table. I hope that representing in this way, and
casting the problem in this light, might generalize to future
materialized views.
* Pre: Add DiffSet.
* Part 4: Collect mutations to a `Schema`.
I haven't taken your review comment about consuming AttributeBuilder
during each fluent function. If you read my response and still want
this, I'm happy to do it in review.
* Part 5: Handle :db/ident and :db.{install,alter}/attribute.
This "loops" the committed datoms out of the SQL store and back
through the metadata (schema, but in future also partition map)
processor. The metadata processor updates the schema and produces a
report of what changed; that report is then used to update the SQL
store. That update includes:
- the materialized views ("entids", "idents", and "schema");
- if needed, a subset of the datoms themselves (as flags change).
I've left a TODO for handling attribute retraction in the cases that
it makes sense. I expect that to be straight-forward.
* Review comment: Rename DiffSet to AddRetractAlterSet.
Also adds a little more commentary and a simple test.
* Review comment: Use ToIdent trait.
* Review comment: partially revert "Part 2: Maintain entids separately from idents."
This reverts commit 23a91df9c35e14398f2ddbd1ba25315821e67401.
Following our discussion, this removes the "entids" materialized
view. The next commit will remove historical idents from the "idents"
materialized view.
* Post: Use custom Either rather than std::result::Result.
This is not necessary, but it was suggested that we might be paying an
overhead creating Err instances while using error_chain. That seems
not to be the case, but this change shows that we don't actually use
any of the Result helper methods, so there's no reason to overload
Result. This change might avoid some future confusion, so I'm going
to land it anyway.
Signed-off-by: Nick Alexander <nalexander@mozilla.com>
* Review comment: Don't preserve historical idents.
* Review comment: More prepared statements when updating materialized views.
* Post: Test altering :db/cardinality and :db/unique.
These tests fail due to a Datomic limitation, namely that the marker
flag :db.alter/attribute can only be asserted once for an attribute!
That is, [:db.part/db :db.alter/attribute :attribute] will only be
transacted at most once. Since older versions of Datomic required the
:db.alter/attribute flag, I can only imagine they either never wrote
:db.alter/attribute to the store, or they handled it specially. I'll
need to remove the marker flag system from Mentat in order to address
this fundamental limitation.
* Post: Remove some more single quotes from error output.
* Post: Add assert_transact! macro to unwrap safely.
I was finding it very difficult to track unwrapping errors while
making changes, due to an underlying Mac OS X symbolication issue that
makes running tests with RUST_BACKTRACE=1 so slow that they all time
out.
* Post: Don't expect or recognize :db.{install,alter}/attribute.
I had this all working... except we will never see a repeated
`[:db.part/db :db.alter/attribute :attribute]` assertion in the store!
That means my approach would let you alter an attribute at most one
time. It's not worth hacking around this; it's better to just stop
expecting (and recognizing) the marker flags. (We have all the data
to distinguish the various cases that we need without the marker
flags.)
This brings Mentat in line with the thrust of newer Datomic versions,
but isn't compatible with Datomic, because (if I understand correctly)
Datomic automatically adds :db.{install,alter}/attribute assertions to
transactions.
I haven't purged the corresponding :db/ident and schema fragments just
yet:
- we might want them back
- we might want them in order to upgrade v1 and v2 databases to the
new on-disk layout we're fleshing out (v3?).
* Post: Don't make :db/unique :db.unique/* imply :db/index true.
This patch avoids a potential bug with the "schema" materialized view.
If :db/unique :db.unique/value implies :db/index true, then what
happens when you _retract_ :db.unique/value? I think Datomic defines
this in some way, but I really want the "schema" materialized view to
be a slice of "datoms" and not have these sort of ambiguities and
persistent effects. Therefore, to ensure that we don't retract a
schema characteristic and accidentally change more than we intended
to, this patch stops having any schema characteristic imply any other
schema characteristic(s). To achieve that, I added an
Option<Unique::{Value,Identity}> type to Attribute; this helps with
this patch, and also looks ahead to when we allow to retract
:db/unique attributes.
* Post: Allow to retract :db/ident.
* Post: Include more details about invalid schema changes.
The tests use strings, so they hide the chained errors which do in
fact provide more detail.
* Review comment: Fix outdated comment.
* Review comment: s/_SET/_SQL_LIST/.
* Review comment: Use a sub-select for checking cardinality.
This might be faster in practice.
* Review comment: Put `attribute::Unique` into its own namespace.
2017-03-20 20:18:59 +00:00
2019-07-17 15:46:47 +00:00
fn resolved_metadata_assertions ( & self ) -> Result < Vec < ( Entid , Entid , TypedValue , bool ) > > {
let sql_stmt = format! (
r #"
2018-07-17 19:18:53 +00:00
SELECT e , a , v , value_type_tag , added FROM
(
SELECT e0 as e , a0 as a , v0 as v , value_type_tag0 as value_type_tag , 1 as added
FROM temp . search_results
WHERE a0 IN { } AND added0 IS 1 AND ( ( rid IS NULL ) OR
( ( rid IS NOT NULL ) AND ( v0 IS NOT v ) ) )
UNION
SELECT e0 as e , a0 as a , v , value_type_tag0 as value_type_tag , 0 as added
FROM temp . search_results
WHERE a0 in { } AND rid IS NOT NULL AND
( ( added0 IS 0 ) OR
( added0 IS 1 AND search_type IS ' :db . cardinality / one ' AND v0 IS NOT v ) )
) ORDER BY e , a , v , value_type_tag , added " #,
2019-07-17 15:46:47 +00:00
entids ::METADATA_SQL_LIST . as_str ( ) ,
entids ::METADATA_SQL_LIST . as_str ( )
2018-07-17 19:18:53 +00:00
) ;
let mut stmt = self . prepare_cached ( & sql_stmt ) ? ;
2019-07-17 15:46:47 +00:00
let m : Result < Vec < _ > > = stmt
2019-07-22 11:40:59 +00:00
. query_and_then ( rusqlite ::params! [ ] , row_to_transaction_assertion ) ?
2019-07-17 15:46:47 +00:00
. collect ( ) ;
Schema alteration. Fixes #294 and #295. (#370) r=rnewman
* Pre: Don't retract :db/ident in test.
Datomic (and eventually Mentat) don't allow to retract :db/ident in
this way, so this runs afoul of future work to support mutating
metadata.
* Pre: s/VALUETYPE/VALUE_TYPE/.
This is consistent with the capitalization (which is "valueType") and
the other identifier.
* Pre: Remove some single quotes from error output.
* Part 1: Make materialized views be uniform [e a v value_type_tag].
This looks ahead to a time when we could support arbitrary
user-defined materialized views. For now, the "idents" materialized
view is those datoms of the form [e :db/ident :namespaced/keyword] and
the "schema" materialized view is those datoms of the form [e a v]
where a is in a particular set of attributes that will become clear in
the following commits.
This change is not backwards compatible, so I'm removing the open
current (really, v2) test. It'll be re-instated when we get to
https://github.com/mozilla/mentat/issues/194.
* Pre: Map TypedValue::Ref to TypedValue::Keyword in debug output.
* Part 3: Separate `schema_to_mutate` from the `schema` used to interpret.
This is just to keep track of the expected changes during
bootstrapping. I want bootstrap metadata mutations to flow through
the same code path as metadata mutations during regular transactions;
by differentiating the schema used for interpretation from the schema
that will be updated I expect to be able to apply bootstrap metadata
mutations to an empty schema and have things like materialized views
created (using the regular code paths).
This commit has been re-ordered for conceptual clarity, but it won't
compile because it references the metadata module. It's possible to
make it compile -- the functionality is there in the schema module --
but it's not worth the rebasing effort until after review (and
possibly not even then, since we'll squash down to a single commit to
land).
* Part 2: Maintain entids separately from idents.
In order to support historical idents, we need to distinguish the
"current" map from entid -> ident from the "complete historical" map
ident -> entid. This is what Datomic does; in Datomic, an ident is
never retracted (although it can be replaced). This approach is an
important part of allowing multiple consumers to share a schema
fragment as it migrates forward.
This fixes a limitation of the Clojure implementation, which did not
handle historical idents across knowledge base close and re-open.
The "entids" materialized view is naturally a slice of the "datoms"
table. The "idents" materialized view is a slice of the
"transactions" table. I hope that representing in this way, and
casting the problem in this light, might generalize to future
materialized views.
* Pre: Add DiffSet.
* Part 4: Collect mutations to a `Schema`.
I haven't taken your review comment about consuming AttributeBuilder
during each fluent function. If you read my response and still want
this, I'm happy to do it in review.
* Part 5: Handle :db/ident and :db.{install,alter}/attribute.
This "loops" the committed datoms out of the SQL store and back
through the metadata (schema, but in future also partition map)
processor. The metadata processor updates the schema and produces a
report of what changed; that report is then used to update the SQL
store. That update includes:
- the materialized views ("entids", "idents", and "schema");
- if needed, a subset of the datoms themselves (as flags change).
I've left a TODO for handling attribute retraction in the cases that
it makes sense. I expect that to be straight-forward.
* Review comment: Rename DiffSet to AddRetractAlterSet.
Also adds a little more commentary and a simple test.
* Review comment: Use ToIdent trait.
* Review comment: partially revert "Part 2: Maintain entids separately from idents."
This reverts commit 23a91df9c35e14398f2ddbd1ba25315821e67401.
Following our discussion, this removes the "entids" materialized
view. The next commit will remove historical idents from the "idents"
materialized view.
* Post: Use custom Either rather than std::result::Result.
This is not necessary, but it was suggested that we might be paying an
overhead creating Err instances while using error_chain. That seems
not to be the case, but this change shows that we don't actually use
any of the Result helper methods, so there's no reason to overload
Result. This change might avoid some future confusion, so I'm going
to land it anyway.
Signed-off-by: Nick Alexander <nalexander@mozilla.com>
* Review comment: Don't preserve historical idents.
* Review comment: More prepared statements when updating materialized views.
* Post: Test altering :db/cardinality and :db/unique.
These tests fail due to a Datomic limitation, namely that the marker
flag :db.alter/attribute can only be asserted once for an attribute!
That is, [:db.part/db :db.alter/attribute :attribute] will only be
transacted at most once. Since older versions of Datomic required the
:db.alter/attribute flag, I can only imagine they either never wrote
:db.alter/attribute to the store, or they handled it specially. I'll
need to remove the marker flag system from Mentat in order to address
this fundamental limitation.
* Post: Remove some more single quotes from error output.
* Post: Add assert_transact! macro to unwrap safely.
I was finding it very difficult to track unwrapping errors while
making changes, due to an underlying Mac OS X symbolication issue that
makes running tests with RUST_BACKTRACE=1 so slow that they all time
out.
* Post: Don't expect or recognize :db.{install,alter}/attribute.
I had this all working... except we will never see a repeated
`[:db.part/db :db.alter/attribute :attribute]` assertion in the store!
That means my approach would let you alter an attribute at most one
time. It's not worth hacking around this; it's better to just stop
expecting (and recognizing) the marker flags. (We have all the data
to distinguish the various cases that we need without the marker
flags.)
This brings Mentat in line with the thrust of newer Datomic versions,
but isn't compatible with Datomic, because (if I understand correctly)
Datomic automatically adds :db.{install,alter}/attribute assertions to
transactions.
I haven't purged the corresponding :db/ident and schema fragments just
yet:
- we might want them back
- we might want them in order to upgrade v1 and v2 databases to the
new on-disk layout we're fleshing out (v3?).
* Post: Don't make :db/unique :db.unique/* imply :db/index true.
This patch avoids a potential bug with the "schema" materialized view.
If :db/unique :db.unique/value implies :db/index true, then what
happens when you _retract_ :db.unique/value? I think Datomic defines
this in some way, but I really want the "schema" materialized view to
be a slice of "datoms" and not have these sort of ambiguities and
persistent effects. Therefore, to ensure that we don't retract a
schema characteristic and accidentally change more than we intended
to, this patch stops having any schema characteristic imply any other
schema characteristic(s). To achieve that, I added an
Option<Unique::{Value,Identity}> type to Attribute; this helps with
this patch, and also looks ahead to when we allow to retract
:db/unique attributes.
* Post: Allow to retract :db/ident.
* Post: Include more details about invalid schema changes.
The tests use strings, so they hide the chained errors which do in
fact provide more detail.
* Review comment: Fix outdated comment.
* Review comment: s/_SET/_SQL_LIST/.
* Review comment: Use a sub-select for checking cardinality.
This might be faster in practice.
* Review comment: Put `attribute::Unique` into its own namespace.
2017-03-20 20:18:59 +00:00
m
}
Extract partial storage abstraction; use error-chain throughout. Fixes #328. r=rnewman (#341)
* Pre: Drop unneeded tx0 from search results.
* Pre: Don't require a schema in some of the DB code.
The idea is to separate the transaction applying code, which is
schema-aware, from the concrete storage code, which is just concerned
with getting bits onto disk.
* Pre: Only reference Schema, not DB, in debug module.
This is part of a larger separation of the volatile PartitionMap,
which is modified every transaction, from the stable Schema, which is
infrequently modified.
* Pre: Fix indentation.
* Extract part of DB to new SchemaTypeChecking trait.
* Extract part of DB to new PartitionMapping trait.
* Pre: Don't expect :db.part/tx partition to advance when tx fails.
This fails right now, because we allocate tx IDs even when we shouldn't.
* Sketch a db interface without DB.
* Add ValueParseError; use error-chain in tx-parser.
This can be simplified when
https://github.com/Marwes/combine/issues/86 makes it to a published
release, but this unblocks us for now. This converts the `combine`
error type `ParseError<&'a [edn::Value]>` to a type with owned
`Vec<edn::Value>` collections, re-using `edn::Value::Vector` for
making them `Display`.
* Pre: Accept Borrow<Schema> instead of just &Schema in debug module.
This makes it easy to use Rc<Schema> or Arc<Schema> without inserting
&* sigils throughout the code.
* Use error-chain in query-parser.
There are a few things to point out here:
- the fine grained error types have been flattened into one crate-wide
error type; it's pretty easy to regain the granularity as needed.
- edn::ParseError is automatically lifted to
mentat_query_parser::errors::Error;
- we use mentat_parser_utils::ValueParser to maintain parsing error
information from `combine`.
* Patch up top-level.
* Review comment: Only `borrow()` once.
2017-02-24 23:32:41 +00:00
}
2017-02-08 22:04:32 +00:00
2018-07-17 19:18:53 +00:00
/// Extract metadata-related [e a typed_value added] datoms committed in the given transaction.
2019-07-17 15:46:47 +00:00
pub fn committed_metadata_assertions (
conn : & rusqlite ::Connection ,
tx_id : Entid ,
) -> Result < Vec < ( Entid , Entid , TypedValue , bool ) > > {
let sql_stmt = format! (
r #"
2018-07-17 19:18:53 +00:00
SELECT e , a , v , value_type_tag , added
FROM transactions
WHERE tx = ? AND a IN { }
ORDER BY e , a , v , value_type_tag , added " #,
entids ::METADATA_SQL_LIST . as_str ( )
) ;
let mut stmt = conn . prepare_cached ( & sql_stmt ) ? ;
2019-07-17 15:46:47 +00:00
let m : Result < Vec < _ > > = stmt
2020-01-14 15:46:21 +00:00
. query_and_then ( & [ & tx_id as & dyn ToSql ] , row_to_transaction_assertion ) ?
2019-07-17 15:46:47 +00:00
. collect ( ) ;
2018-07-17 19:18:53 +00:00
m
}
/// Takes a row, produces a transaction quadruple.
fn row_to_transaction_assertion ( row : & rusqlite ::Row ) -> Result < ( Entid , Entid , TypedValue , bool ) > {
Ok ( (
2019-07-17 15:46:47 +00:00
row . get ( 0 ) ? ,
row . get ( 1 ) ? ,
TypedValue ::from_sql_value_pair ( row . get ( 2 ) ? , row . get ( 3 ) ? ) ? ,
row . get ( 4 ) ? ,
2018-07-17 19:18:53 +00:00
) )
}
/// Takes a row, produces a datom quadruple.
fn row_to_datom_assertion ( row : & rusqlite ::Row ) -> Result < ( Entid , Entid , TypedValue ) > {
Ok ( (
2019-07-17 15:46:47 +00:00
row . get ( 0 ) ? ,
row . get ( 1 ) ? ,
TypedValue ::from_sql_value_pair ( row . get ( 2 ) ? , row . get ( 3 ) ? ) ? ,
2018-07-17 19:18:53 +00:00
) )
}
Schema alteration. Fixes #294 and #295. (#370) r=rnewman
* Pre: Don't retract :db/ident in test.
Datomic (and eventually Mentat) don't allow to retract :db/ident in
this way, so this runs afoul of future work to support mutating
metadata.
* Pre: s/VALUETYPE/VALUE_TYPE/.
This is consistent with the capitalization (which is "valueType") and
the other identifier.
* Pre: Remove some single quotes from error output.
* Part 1: Make materialized views be uniform [e a v value_type_tag].
This looks ahead to a time when we could support arbitrary
user-defined materialized views. For now, the "idents" materialized
view is those datoms of the form [e :db/ident :namespaced/keyword] and
the "schema" materialized view is those datoms of the form [e a v]
where a is in a particular set of attributes that will become clear in
the following commits.
This change is not backwards compatible, so I'm removing the open
current (really, v2) test. It'll be re-instated when we get to
https://github.com/mozilla/mentat/issues/194.
* Pre: Map TypedValue::Ref to TypedValue::Keyword in debug output.
* Part 3: Separate `schema_to_mutate` from the `schema` used to interpret.
This is just to keep track of the expected changes during
bootstrapping. I want bootstrap metadata mutations to flow through
the same code path as metadata mutations during regular transactions;
by differentiating the schema used for interpretation from the schema
that will be updated I expect to be able to apply bootstrap metadata
mutations to an empty schema and have things like materialized views
created (using the regular code paths).
This commit has been re-ordered for conceptual clarity, but it won't
compile because it references the metadata module. It's possible to
make it compile -- the functionality is there in the schema module --
but it's not worth the rebasing effort until after review (and
possibly not even then, since we'll squash down to a single commit to
land).
* Part 2: Maintain entids separately from idents.
In order to support historical idents, we need to distinguish the
"current" map from entid -> ident from the "complete historical" map
ident -> entid. This is what Datomic does; in Datomic, an ident is
never retracted (although it can be replaced). This approach is an
important part of allowing multiple consumers to share a schema
fragment as it migrates forward.
This fixes a limitation of the Clojure implementation, which did not
handle historical idents across knowledge base close and re-open.
The "entids" materialized view is naturally a slice of the "datoms"
table. The "idents" materialized view is a slice of the
"transactions" table. I hope that representing in this way, and
casting the problem in this light, might generalize to future
materialized views.
* Pre: Add DiffSet.
* Part 4: Collect mutations to a `Schema`.
I haven't taken your review comment about consuming AttributeBuilder
during each fluent function. If you read my response and still want
this, I'm happy to do it in review.
* Part 5: Handle :db/ident and :db.{install,alter}/attribute.
This "loops" the committed datoms out of the SQL store and back
through the metadata (schema, but in future also partition map)
processor. The metadata processor updates the schema and produces a
report of what changed; that report is then used to update the SQL
store. That update includes:
- the materialized views ("entids", "idents", and "schema");
- if needed, a subset of the datoms themselves (as flags change).
I've left a TODO for handling attribute retraction in the cases that
it makes sense. I expect that to be straight-forward.
* Review comment: Rename DiffSet to AddRetractAlterSet.
Also adds a little more commentary and a simple test.
* Review comment: Use ToIdent trait.
* Review comment: partially revert "Part 2: Maintain entids separately from idents."
This reverts commit 23a91df9c35e14398f2ddbd1ba25315821e67401.
Following our discussion, this removes the "entids" materialized
view. The next commit will remove historical idents from the "idents"
materialized view.
* Post: Use custom Either rather than std::result::Result.
This is not necessary, but it was suggested that we might be paying an
overhead creating Err instances while using error_chain. That seems
not to be the case, but this change shows that we don't actually use
any of the Result helper methods, so there's no reason to overload
Result. This change might avoid some future confusion, so I'm going
to land it anyway.
Signed-off-by: Nick Alexander <nalexander@mozilla.com>
* Review comment: Don't preserve historical idents.
* Review comment: More prepared statements when updating materialized views.
* Post: Test altering :db/cardinality and :db/unique.
These tests fail due to a Datomic limitation, namely that the marker
flag :db.alter/attribute can only be asserted once for an attribute!
That is, [:db.part/db :db.alter/attribute :attribute] will only be
transacted at most once. Since older versions of Datomic required the
:db.alter/attribute flag, I can only imagine they either never wrote
:db.alter/attribute to the store, or they handled it specially. I'll
need to remove the marker flag system from Mentat in order to address
this fundamental limitation.
* Post: Remove some more single quotes from error output.
* Post: Add assert_transact! macro to unwrap safely.
I was finding it very difficult to track unwrapping errors while
making changes, due to an underlying Mac OS X symbolication issue that
makes running tests with RUST_BACKTRACE=1 so slow that they all time
out.
* Post: Don't expect or recognize :db.{install,alter}/attribute.
I had this all working... except we will never see a repeated
`[:db.part/db :db.alter/attribute :attribute]` assertion in the store!
That means my approach would let you alter an attribute at most one
time. It's not worth hacking around this; it's better to just stop
expecting (and recognizing) the marker flags. (We have all the data
to distinguish the various cases that we need without the marker
flags.)
This brings Mentat in line with the thrust of newer Datomic versions,
but isn't compatible with Datomic, because (if I understand correctly)
Datomic automatically adds :db.{install,alter}/attribute assertions to
transactions.
I haven't purged the corresponding :db/ident and schema fragments just
yet:
- we might want them back
- we might want them in order to upgrade v1 and v2 databases to the
new on-disk layout we're fleshing out (v3?).
* Post: Don't make :db/unique :db.unique/* imply :db/index true.
This patch avoids a potential bug with the "schema" materialized view.
If :db/unique :db.unique/value implies :db/index true, then what
happens when you _retract_ :db.unique/value? I think Datomic defines
this in some way, but I really want the "schema" materialized view to
be a slice of "datoms" and not have these sort of ambiguities and
persistent effects. Therefore, to ensure that we don't retract a
schema characteristic and accidentally change more than we intended
to, this patch stops having any schema characteristic imply any other
schema characteristic(s). To achieve that, I added an
Option<Unique::{Value,Identity}> type to Attribute; this helps with
this patch, and also looks ahead to when we allow to retract
:db/unique attributes.
* Post: Allow to retract :db/ident.
* Post: Include more details about invalid schema changes.
The tests use strings, so they hide the chained errors which do in
fact provide more detail.
* Review comment: Fix outdated comment.
* Review comment: s/_SET/_SQL_LIST/.
* Review comment: Use a sub-select for checking cardinality.
This might be faster in practice.
* Review comment: Put `attribute::Unique` into its own namespace.
2017-03-20 20:18:59 +00:00
/// Update the metadata materialized views based on the given metadata report.
///
/// This updates the "entids", "idents", and "schema" materialized views, copying directly from the
/// "datoms" and "transactions" table as appropriate.
2019-07-17 15:46:47 +00:00
pub fn update_metadata (
conn : & rusqlite ::Connection ,
_old_schema : & Schema ,
new_schema : & Schema ,
metadata_report : & metadata ::MetadataReport ,
) -> Result < ( ) > {
2020-08-06 03:03:58 +00:00
use crate ::metadata ::AttributeAlteration ::* ;
Schema alteration. Fixes #294 and #295. (#370) r=rnewman
* Pre: Don't retract :db/ident in test.
Datomic (and eventually Mentat) don't allow to retract :db/ident in
this way, so this runs afoul of future work to support mutating
metadata.
* Pre: s/VALUETYPE/VALUE_TYPE/.
This is consistent with the capitalization (which is "valueType") and
the other identifier.
* Pre: Remove some single quotes from error output.
* Part 1: Make materialized views be uniform [e a v value_type_tag].
This looks ahead to a time when we could support arbitrary
user-defined materialized views. For now, the "idents" materialized
view is those datoms of the form [e :db/ident :namespaced/keyword] and
the "schema" materialized view is those datoms of the form [e a v]
where a is in a particular set of attributes that will become clear in
the following commits.
This change is not backwards compatible, so I'm removing the open
current (really, v2) test. It'll be re-instated when we get to
https://github.com/mozilla/mentat/issues/194.
* Pre: Map TypedValue::Ref to TypedValue::Keyword in debug output.
* Part 3: Separate `schema_to_mutate` from the `schema` used to interpret.
This is just to keep track of the expected changes during
bootstrapping. I want bootstrap metadata mutations to flow through
the same code path as metadata mutations during regular transactions;
by differentiating the schema used for interpretation from the schema
that will be updated I expect to be able to apply bootstrap metadata
mutations to an empty schema and have things like materialized views
created (using the regular code paths).
This commit has been re-ordered for conceptual clarity, but it won't
compile because it references the metadata module. It's possible to
make it compile -- the functionality is there in the schema module --
but it's not worth the rebasing effort until after review (and
possibly not even then, since we'll squash down to a single commit to
land).
* Part 2: Maintain entids separately from idents.
In order to support historical idents, we need to distinguish the
"current" map from entid -> ident from the "complete historical" map
ident -> entid. This is what Datomic does; in Datomic, an ident is
never retracted (although it can be replaced). This approach is an
important part of allowing multiple consumers to share a schema
fragment as it migrates forward.
This fixes a limitation of the Clojure implementation, which did not
handle historical idents across knowledge base close and re-open.
The "entids" materialized view is naturally a slice of the "datoms"
table. The "idents" materialized view is a slice of the
"transactions" table. I hope that representing in this way, and
casting the problem in this light, might generalize to future
materialized views.
* Pre: Add DiffSet.
* Part 4: Collect mutations to a `Schema`.
I haven't taken your review comment about consuming AttributeBuilder
during each fluent function. If you read my response and still want
this, I'm happy to do it in review.
* Part 5: Handle :db/ident and :db.{install,alter}/attribute.
This "loops" the committed datoms out of the SQL store and back
through the metadata (schema, but in future also partition map)
processor. The metadata processor updates the schema and produces a
report of what changed; that report is then used to update the SQL
store. That update includes:
- the materialized views ("entids", "idents", and "schema");
- if needed, a subset of the datoms themselves (as flags change).
I've left a TODO for handling attribute retraction in the cases that
it makes sense. I expect that to be straight-forward.
* Review comment: Rename DiffSet to AddRetractAlterSet.
Also adds a little more commentary and a simple test.
* Review comment: Use ToIdent trait.
* Review comment: partially revert "Part 2: Maintain entids separately from idents."
This reverts commit 23a91df9c35e14398f2ddbd1ba25315821e67401.
Following our discussion, this removes the "entids" materialized
view. The next commit will remove historical idents from the "idents"
materialized view.
* Post: Use custom Either rather than std::result::Result.
This is not necessary, but it was suggested that we might be paying an
overhead creating Err instances while using error_chain. That seems
not to be the case, but this change shows that we don't actually use
any of the Result helper methods, so there's no reason to overload
Result. This change might avoid some future confusion, so I'm going
to land it anyway.
Signed-off-by: Nick Alexander <nalexander@mozilla.com>
* Review comment: Don't preserve historical idents.
* Review comment: More prepared statements when updating materialized views.
* Post: Test altering :db/cardinality and :db/unique.
These tests fail due to a Datomic limitation, namely that the marker
flag :db.alter/attribute can only be asserted once for an attribute!
That is, [:db.part/db :db.alter/attribute :attribute] will only be
transacted at most once. Since older versions of Datomic required the
:db.alter/attribute flag, I can only imagine they either never wrote
:db.alter/attribute to the store, or they handled it specially. I'll
need to remove the marker flag system from Mentat in order to address
this fundamental limitation.
* Post: Remove some more single quotes from error output.
* Post: Add assert_transact! macro to unwrap safely.
I was finding it very difficult to track unwrapping errors while
making changes, due to an underlying Mac OS X symbolication issue that
makes running tests with RUST_BACKTRACE=1 so slow that they all time
out.
* Post: Don't expect or recognize :db.{install,alter}/attribute.
I had this all working... except we will never see a repeated
`[:db.part/db :db.alter/attribute :attribute]` assertion in the store!
That means my approach would let you alter an attribute at most one
time. It's not worth hacking around this; it's better to just stop
expecting (and recognizing) the marker flags. (We have all the data
to distinguish the various cases that we need without the marker
flags.)
This brings Mentat in line with the thrust of newer Datomic versions,
but isn't compatible with Datomic, because (if I understand correctly)
Datomic automatically adds :db.{install,alter}/attribute assertions to
transactions.
I haven't purged the corresponding :db/ident and schema fragments just
yet:
- we might want them back
- we might want them in order to upgrade v1 and v2 databases to the
new on-disk layout we're fleshing out (v3?).
* Post: Don't make :db/unique :db.unique/* imply :db/index true.
This patch avoids a potential bug with the "schema" materialized view.
If :db/unique :db.unique/value implies :db/index true, then what
happens when you _retract_ :db.unique/value? I think Datomic defines
this in some way, but I really want the "schema" materialized view to
be a slice of "datoms" and not have these sort of ambiguities and
persistent effects. Therefore, to ensure that we don't retract a
schema characteristic and accidentally change more than we intended
to, this patch stops having any schema characteristic imply any other
schema characteristic(s). To achieve that, I added an
Option<Unique::{Value,Identity}> type to Attribute; this helps with
this patch, and also looks ahead to when we allow to retract
:db/unique attributes.
* Post: Allow to retract :db/ident.
* Post: Include more details about invalid schema changes.
The tests use strings, so they hide the chained errors which do in
fact provide more detail.
* Review comment: Fix outdated comment.
* Review comment: s/_SET/_SQL_LIST/.
* Review comment: Use a sub-select for checking cardinality.
This might be faster in practice.
* Review comment: Put `attribute::Unique` into its own namespace.
2017-03-20 20:18:59 +00:00
// Populate the materialized view directly from datoms (and, potentially in the future,
// transactions). This might generalize nicely as we expand the set of materialized views.
// TODO: consider doing this in fewer SQLite execute() invocations.
// TODO: use concat! to avoid creating String instances.
if ! metadata_report . idents_altered . is_empty ( ) {
// Idents is the materialized view of the [entid :db/ident ident] slice of datoms.
2020-01-23 18:16:19 +00:00
conn . execute (
" DELETE FROM idents " . to_string ( ) . as_str ( ) ,
rusqlite ::params! [ ] ,
) ? ;
2019-07-17 15:46:47 +00:00
conn . execute (
format! (
" INSERT INTO idents SELECT e, a, v, value_type_tag FROM datoms WHERE a IN {} " ,
entids ::IDENTS_SQL_LIST . as_str ( )
)
. as_str ( ) ,
2019-07-22 11:40:59 +00:00
rusqlite ::params! [ ] ,
2019-07-17 15:46:47 +00:00
) ? ;
Schema alteration. Fixes #294 and #295. (#370) r=rnewman
* Pre: Don't retract :db/ident in test.
Datomic (and eventually Mentat) don't allow to retract :db/ident in
this way, so this runs afoul of future work to support mutating
metadata.
* Pre: s/VALUETYPE/VALUE_TYPE/.
This is consistent with the capitalization (which is "valueType") and
the other identifier.
* Pre: Remove some single quotes from error output.
* Part 1: Make materialized views be uniform [e a v value_type_tag].
This looks ahead to a time when we could support arbitrary
user-defined materialized views. For now, the "idents" materialized
view is those datoms of the form [e :db/ident :namespaced/keyword] and
the "schema" materialized view is those datoms of the form [e a v]
where a is in a particular set of attributes that will become clear in
the following commits.
This change is not backwards compatible, so I'm removing the open
current (really, v2) test. It'll be re-instated when we get to
https://github.com/mozilla/mentat/issues/194.
* Pre: Map TypedValue::Ref to TypedValue::Keyword in debug output.
* Part 3: Separate `schema_to_mutate` from the `schema` used to interpret.
This is just to keep track of the expected changes during
bootstrapping. I want bootstrap metadata mutations to flow through
the same code path as metadata mutations during regular transactions;
by differentiating the schema used for interpretation from the schema
that will be updated I expect to be able to apply bootstrap metadata
mutations to an empty schema and have things like materialized views
created (using the regular code paths).
This commit has been re-ordered for conceptual clarity, but it won't
compile because it references the metadata module. It's possible to
make it compile -- the functionality is there in the schema module --
but it's not worth the rebasing effort until after review (and
possibly not even then, since we'll squash down to a single commit to
land).
* Part 2: Maintain entids separately from idents.
In order to support historical idents, we need to distinguish the
"current" map from entid -> ident from the "complete historical" map
ident -> entid. This is what Datomic does; in Datomic, an ident is
never retracted (although it can be replaced). This approach is an
important part of allowing multiple consumers to share a schema
fragment as it migrates forward.
This fixes a limitation of the Clojure implementation, which did not
handle historical idents across knowledge base close and re-open.
The "entids" materialized view is naturally a slice of the "datoms"
table. The "idents" materialized view is a slice of the
"transactions" table. I hope that representing in this way, and
casting the problem in this light, might generalize to future
materialized views.
* Pre: Add DiffSet.
* Part 4: Collect mutations to a `Schema`.
I haven't taken your review comment about consuming AttributeBuilder
during each fluent function. If you read my response and still want
this, I'm happy to do it in review.
* Part 5: Handle :db/ident and :db.{install,alter}/attribute.
This "loops" the committed datoms out of the SQL store and back
through the metadata (schema, but in future also partition map)
processor. The metadata processor updates the schema and produces a
report of what changed; that report is then used to update the SQL
store. That update includes:
- the materialized views ("entids", "idents", and "schema");
- if needed, a subset of the datoms themselves (as flags change).
I've left a TODO for handling attribute retraction in the cases that
it makes sense. I expect that to be straight-forward.
* Review comment: Rename DiffSet to AddRetractAlterSet.
Also adds a little more commentary and a simple test.
* Review comment: Use ToIdent trait.
* Review comment: partially revert "Part 2: Maintain entids separately from idents."
This reverts commit 23a91df9c35e14398f2ddbd1ba25315821e67401.
Following our discussion, this removes the "entids" materialized
view. The next commit will remove historical idents from the "idents"
materialized view.
* Post: Use custom Either rather than std::result::Result.
This is not necessary, but it was suggested that we might be paying an
overhead creating Err instances while using error_chain. That seems
not to be the case, but this change shows that we don't actually use
any of the Result helper methods, so there's no reason to overload
Result. This change might avoid some future confusion, so I'm going
to land it anyway.
Signed-off-by: Nick Alexander <nalexander@mozilla.com>
* Review comment: Don't preserve historical idents.
* Review comment: More prepared statements when updating materialized views.
* Post: Test altering :db/cardinality and :db/unique.
These tests fail due to a Datomic limitation, namely that the marker
flag :db.alter/attribute can only be asserted once for an attribute!
That is, [:db.part/db :db.alter/attribute :attribute] will only be
transacted at most once. Since older versions of Datomic required the
:db.alter/attribute flag, I can only imagine they either never wrote
:db.alter/attribute to the store, or they handled it specially. I'll
need to remove the marker flag system from Mentat in order to address
this fundamental limitation.
* Post: Remove some more single quotes from error output.
* Post: Add assert_transact! macro to unwrap safely.
I was finding it very difficult to track unwrapping errors while
making changes, due to an underlying Mac OS X symbolication issue that
makes running tests with RUST_BACKTRACE=1 so slow that they all time
out.
* Post: Don't expect or recognize :db.{install,alter}/attribute.
I had this all working... except we will never see a repeated
`[:db.part/db :db.alter/attribute :attribute]` assertion in the store!
That means my approach would let you alter an attribute at most one
time. It's not worth hacking around this; it's better to just stop
expecting (and recognizing) the marker flags. (We have all the data
to distinguish the various cases that we need without the marker
flags.)
This brings Mentat in line with the thrust of newer Datomic versions,
but isn't compatible with Datomic, because (if I understand correctly)
Datomic automatically adds :db.{install,alter}/attribute assertions to
transactions.
I haven't purged the corresponding :db/ident and schema fragments just
yet:
- we might want them back
- we might want them in order to upgrade v1 and v2 databases to the
new on-disk layout we're fleshing out (v3?).
* Post: Don't make :db/unique :db.unique/* imply :db/index true.
This patch avoids a potential bug with the "schema" materialized view.
If :db/unique :db.unique/value implies :db/index true, then what
happens when you _retract_ :db.unique/value? I think Datomic defines
this in some way, but I really want the "schema" materialized view to
be a slice of "datoms" and not have these sort of ambiguities and
persistent effects. Therefore, to ensure that we don't retract a
schema characteristic and accidentally change more than we intended
to, this patch stops having any schema characteristic imply any other
schema characteristic(s). To achieve that, I added an
Option<Unique::{Value,Identity}> type to Attribute; this helps with
this patch, and also looks ahead to when we allow to retract
:db/unique attributes.
* Post: Allow to retract :db/ident.
* Post: Include more details about invalid schema changes.
The tests use strings, so they hide the chained errors which do in
fact provide more detail.
* Review comment: Fix outdated comment.
* Review comment: s/_SET/_SQL_LIST/.
* Review comment: Use a sub-select for checking cardinality.
This might be faster in practice.
* Review comment: Put `attribute::Unique` into its own namespace.
2017-03-20 20:18:59 +00:00
}
2018-07-18 23:55:31 +00:00
// Populate the materialized view directly from datoms.
// It's possible that an "ident" was removed, along with its attributes.
// That's not counted as an "alteration" of attributes, so we explicitly check
// for non-emptiness of 'idents_altered'.
Schema alteration. Fixes #294 and #295. (#370) r=rnewman
* Pre: Don't retract :db/ident in test.
Datomic (and eventually Mentat) don't allow to retract :db/ident in
this way, so this runs afoul of future work to support mutating
metadata.
* Pre: s/VALUETYPE/VALUE_TYPE/.
This is consistent with the capitalization (which is "valueType") and
the other identifier.
* Pre: Remove some single quotes from error output.
* Part 1: Make materialized views be uniform [e a v value_type_tag].
This looks ahead to a time when we could support arbitrary
user-defined materialized views. For now, the "idents" materialized
view is those datoms of the form [e :db/ident :namespaced/keyword] and
the "schema" materialized view is those datoms of the form [e a v]
where a is in a particular set of attributes that will become clear in
the following commits.
This change is not backwards compatible, so I'm removing the open
current (really, v2) test. It'll be re-instated when we get to
https://github.com/mozilla/mentat/issues/194.
* Pre: Map TypedValue::Ref to TypedValue::Keyword in debug output.
* Part 3: Separate `schema_to_mutate` from the `schema` used to interpret.
This is just to keep track of the expected changes during
bootstrapping. I want bootstrap metadata mutations to flow through
the same code path as metadata mutations during regular transactions;
by differentiating the schema used for interpretation from the schema
that will be updated I expect to be able to apply bootstrap metadata
mutations to an empty schema and have things like materialized views
created (using the regular code paths).
This commit has been re-ordered for conceptual clarity, but it won't
compile because it references the metadata module. It's possible to
make it compile -- the functionality is there in the schema module --
but it's not worth the rebasing effort until after review (and
possibly not even then, since we'll squash down to a single commit to
land).
* Part 2: Maintain entids separately from idents.
In order to support historical idents, we need to distinguish the
"current" map from entid -> ident from the "complete historical" map
ident -> entid. This is what Datomic does; in Datomic, an ident is
never retracted (although it can be replaced). This approach is an
important part of allowing multiple consumers to share a schema
fragment as it migrates forward.
This fixes a limitation of the Clojure implementation, which did not
handle historical idents across knowledge base close and re-open.
The "entids" materialized view is naturally a slice of the "datoms"
table. The "idents" materialized view is a slice of the
"transactions" table. I hope that representing in this way, and
casting the problem in this light, might generalize to future
materialized views.
* Pre: Add DiffSet.
* Part 4: Collect mutations to a `Schema`.
I haven't taken your review comment about consuming AttributeBuilder
during each fluent function. If you read my response and still want
this, I'm happy to do it in review.
* Part 5: Handle :db/ident and :db.{install,alter}/attribute.
This "loops" the committed datoms out of the SQL store and back
through the metadata (schema, but in future also partition map)
processor. The metadata processor updates the schema and produces a
report of what changed; that report is then used to update the SQL
store. That update includes:
- the materialized views ("entids", "idents", and "schema");
- if needed, a subset of the datoms themselves (as flags change).
I've left a TODO for handling attribute retraction in the cases that
it makes sense. I expect that to be straight-forward.
* Review comment: Rename DiffSet to AddRetractAlterSet.
Also adds a little more commentary and a simple test.
* Review comment: Use ToIdent trait.
* Review comment: partially revert "Part 2: Maintain entids separately from idents."
This reverts commit 23a91df9c35e14398f2ddbd1ba25315821e67401.
Following our discussion, this removes the "entids" materialized
view. The next commit will remove historical idents from the "idents"
materialized view.
* Post: Use custom Either rather than std::result::Result.
This is not necessary, but it was suggested that we might be paying an
overhead creating Err instances while using error_chain. That seems
not to be the case, but this change shows that we don't actually use
any of the Result helper methods, so there's no reason to overload
Result. This change might avoid some future confusion, so I'm going
to land it anyway.
Signed-off-by: Nick Alexander <nalexander@mozilla.com>
* Review comment: Don't preserve historical idents.
* Review comment: More prepared statements when updating materialized views.
* Post: Test altering :db/cardinality and :db/unique.
These tests fail due to a Datomic limitation, namely that the marker
flag :db.alter/attribute can only be asserted once for an attribute!
That is, [:db.part/db :db.alter/attribute :attribute] will only be
transacted at most once. Since older versions of Datomic required the
:db.alter/attribute flag, I can only imagine they either never wrote
:db.alter/attribute to the store, or they handled it specially. I'll
need to remove the marker flag system from Mentat in order to address
this fundamental limitation.
* Post: Remove some more single quotes from error output.
* Post: Add assert_transact! macro to unwrap safely.
I was finding it very difficult to track unwrapping errors while
making changes, due to an underlying Mac OS X symbolication issue that
makes running tests with RUST_BACKTRACE=1 so slow that they all time
out.
* Post: Don't expect or recognize :db.{install,alter}/attribute.
I had this all working... except we will never see a repeated
`[:db.part/db :db.alter/attribute :attribute]` assertion in the store!
That means my approach would let you alter an attribute at most one
time. It's not worth hacking around this; it's better to just stop
expecting (and recognizing) the marker flags. (We have all the data
to distinguish the various cases that we need without the marker
flags.)
This brings Mentat in line with the thrust of newer Datomic versions,
but isn't compatible with Datomic, because (if I understand correctly)
Datomic automatically adds :db.{install,alter}/attribute assertions to
transactions.
I haven't purged the corresponding :db/ident and schema fragments just
yet:
- we might want them back
- we might want them in order to upgrade v1 and v2 databases to the
new on-disk layout we're fleshing out (v3?).
* Post: Don't make :db/unique :db.unique/* imply :db/index true.
This patch avoids a potential bug with the "schema" materialized view.
If :db/unique :db.unique/value implies :db/index true, then what
happens when you _retract_ :db.unique/value? I think Datomic defines
this in some way, but I really want the "schema" materialized view to
be a slice of "datoms" and not have these sort of ambiguities and
persistent effects. Therefore, to ensure that we don't retract a
schema characteristic and accidentally change more than we intended
to, this patch stops having any schema characteristic imply any other
schema characteristic(s). To achieve that, I added an
Option<Unique::{Value,Identity}> type to Attribute; this helps with
this patch, and also looks ahead to when we allow to retract
:db/unique attributes.
* Post: Allow to retract :db/ident.
* Post: Include more details about invalid schema changes.
The tests use strings, so they hide the chained errors which do in
fact provide more detail.
* Review comment: Fix outdated comment.
* Review comment: s/_SET/_SQL_LIST/.
* Review comment: Use a sub-select for checking cardinality.
This might be faster in practice.
* Review comment: Put `attribute::Unique` into its own namespace.
2017-03-20 20:18:59 +00:00
2018-07-18 23:55:31 +00:00
// TODO expand metadata report to allow for better signaling for the above.
if ! metadata_report . attributes_installed . is_empty ( )
| | ! metadata_report . attributes_altered . is_empty ( )
2019-07-17 15:46:47 +00:00
| | ! metadata_report . idents_altered . is_empty ( )
{
2020-01-23 18:16:19 +00:00
conn . execute (
" DELETE FROM schema " . to_string ( ) . as_str ( ) ,
rusqlite ::params! [ ] ,
) ? ;
2018-07-18 23:55:31 +00:00
// NB: we're using :db/valueType as a placeholder for the entire schema-defining set.
2019-07-17 15:46:47 +00:00
let s = format! (
r #"
2018-07-18 23:55:31 +00:00
WITH s ( e ) AS ( SELECT e FROM datoms WHERE a = { } )
INSERT INTO schema
SELECT s . e , a , v , value_type_tag
FROM datoms , s
WHERE s . e = datoms . e AND a IN { }
2019-07-17 15:46:47 +00:00
" #,
entids ::DB_VALUE_TYPE ,
entids ::SCHEMA_SQL_LIST . as_str ( )
) ;
2019-07-22 11:40:59 +00:00
conn . execute ( & s , rusqlite ::params! [ ] ) ? ;
Schema alteration. Fixes #294 and #295. (#370) r=rnewman
* Pre: Don't retract :db/ident in test.
Datomic (and eventually Mentat) don't allow to retract :db/ident in
this way, so this runs afoul of future work to support mutating
metadata.
* Pre: s/VALUETYPE/VALUE_TYPE/.
This is consistent with the capitalization (which is "valueType") and
the other identifier.
* Pre: Remove some single quotes from error output.
* Part 1: Make materialized views be uniform [e a v value_type_tag].
This looks ahead to a time when we could support arbitrary
user-defined materialized views. For now, the "idents" materialized
view is those datoms of the form [e :db/ident :namespaced/keyword] and
the "schema" materialized view is those datoms of the form [e a v]
where a is in a particular set of attributes that will become clear in
the following commits.
This change is not backwards compatible, so I'm removing the open
current (really, v2) test. It'll be re-instated when we get to
https://github.com/mozilla/mentat/issues/194.
* Pre: Map TypedValue::Ref to TypedValue::Keyword in debug output.
* Part 3: Separate `schema_to_mutate` from the `schema` used to interpret.
This is just to keep track of the expected changes during
bootstrapping. I want bootstrap metadata mutations to flow through
the same code path as metadata mutations during regular transactions;
by differentiating the schema used for interpretation from the schema
that will be updated I expect to be able to apply bootstrap metadata
mutations to an empty schema and have things like materialized views
created (using the regular code paths).
This commit has been re-ordered for conceptual clarity, but it won't
compile because it references the metadata module. It's possible to
make it compile -- the functionality is there in the schema module --
but it's not worth the rebasing effort until after review (and
possibly not even then, since we'll squash down to a single commit to
land).
* Part 2: Maintain entids separately from idents.
In order to support historical idents, we need to distinguish the
"current" map from entid -> ident from the "complete historical" map
ident -> entid. This is what Datomic does; in Datomic, an ident is
never retracted (although it can be replaced). This approach is an
important part of allowing multiple consumers to share a schema
fragment as it migrates forward.
This fixes a limitation of the Clojure implementation, which did not
handle historical idents across knowledge base close and re-open.
The "entids" materialized view is naturally a slice of the "datoms"
table. The "idents" materialized view is a slice of the
"transactions" table. I hope that representing in this way, and
casting the problem in this light, might generalize to future
materialized views.
* Pre: Add DiffSet.
* Part 4: Collect mutations to a `Schema`.
I haven't taken your review comment about consuming AttributeBuilder
during each fluent function. If you read my response and still want
this, I'm happy to do it in review.
* Part 5: Handle :db/ident and :db.{install,alter}/attribute.
This "loops" the committed datoms out of the SQL store and back
through the metadata (schema, but in future also partition map)
processor. The metadata processor updates the schema and produces a
report of what changed; that report is then used to update the SQL
store. That update includes:
- the materialized views ("entids", "idents", and "schema");
- if needed, a subset of the datoms themselves (as flags change).
I've left a TODO for handling attribute retraction in the cases that
it makes sense. I expect that to be straight-forward.
* Review comment: Rename DiffSet to AddRetractAlterSet.
Also adds a little more commentary and a simple test.
* Review comment: Use ToIdent trait.
* Review comment: partially revert "Part 2: Maintain entids separately from idents."
This reverts commit 23a91df9c35e14398f2ddbd1ba25315821e67401.
Following our discussion, this removes the "entids" materialized
view. The next commit will remove historical idents from the "idents"
materialized view.
* Post: Use custom Either rather than std::result::Result.
This is not necessary, but it was suggested that we might be paying an
overhead creating Err instances while using error_chain. That seems
not to be the case, but this change shows that we don't actually use
any of the Result helper methods, so there's no reason to overload
Result. This change might avoid some future confusion, so I'm going
to land it anyway.
Signed-off-by: Nick Alexander <nalexander@mozilla.com>
* Review comment: Don't preserve historical idents.
* Review comment: More prepared statements when updating materialized views.
* Post: Test altering :db/cardinality and :db/unique.
These tests fail due to a Datomic limitation, namely that the marker
flag :db.alter/attribute can only be asserted once for an attribute!
That is, [:db.part/db :db.alter/attribute :attribute] will only be
transacted at most once. Since older versions of Datomic required the
:db.alter/attribute flag, I can only imagine they either never wrote
:db.alter/attribute to the store, or they handled it specially. I'll
need to remove the marker flag system from Mentat in order to address
this fundamental limitation.
* Post: Remove some more single quotes from error output.
* Post: Add assert_transact! macro to unwrap safely.
I was finding it very difficult to track unwrapping errors while
making changes, due to an underlying Mac OS X symbolication issue that
makes running tests with RUST_BACKTRACE=1 so slow that they all time
out.
* Post: Don't expect or recognize :db.{install,alter}/attribute.
I had this all working... except we will never see a repeated
`[:db.part/db :db.alter/attribute :attribute]` assertion in the store!
That means my approach would let you alter an attribute at most one
time. It's not worth hacking around this; it's better to just stop
expecting (and recognizing) the marker flags. (We have all the data
to distinguish the various cases that we need without the marker
flags.)
This brings Mentat in line with the thrust of newer Datomic versions,
but isn't compatible with Datomic, because (if I understand correctly)
Datomic automatically adds :db.{install,alter}/attribute assertions to
transactions.
I haven't purged the corresponding :db/ident and schema fragments just
yet:
- we might want them back
- we might want them in order to upgrade v1 and v2 databases to the
new on-disk layout we're fleshing out (v3?).
* Post: Don't make :db/unique :db.unique/* imply :db/index true.
This patch avoids a potential bug with the "schema" materialized view.
If :db/unique :db.unique/value implies :db/index true, then what
happens when you _retract_ :db.unique/value? I think Datomic defines
this in some way, but I really want the "schema" materialized view to
be a slice of "datoms" and not have these sort of ambiguities and
persistent effects. Therefore, to ensure that we don't retract a
schema characteristic and accidentally change more than we intended
to, this patch stops having any schema characteristic imply any other
schema characteristic(s). To achieve that, I added an
Option<Unique::{Value,Identity}> type to Attribute; this helps with
this patch, and also looks ahead to when we allow to retract
:db/unique attributes.
* Post: Allow to retract :db/ident.
* Post: Include more details about invalid schema changes.
The tests use strings, so they hide the chained errors which do in
fact provide more detail.
* Review comment: Fix outdated comment.
* Review comment: s/_SET/_SQL_LIST/.
* Review comment: Use a sub-select for checking cardinality.
This might be faster in practice.
* Review comment: Put `attribute::Unique` into its own namespace.
2017-03-20 20:18:59 +00:00
}
let mut index_stmt = conn . prepare ( " UPDATE datoms SET index_avet = ? WHERE a = ? " ) ? ;
let mut unique_value_stmt = conn . prepare ( " UPDATE datoms SET unique_value = ? WHERE a = ? " ) ? ;
2019-07-17 15:46:47 +00:00
let mut cardinality_stmt = conn . prepare (
r #"
Schema alteration. Fixes #294 and #295. (#370) r=rnewman
* Pre: Don't retract :db/ident in test.
Datomic (and eventually Mentat) don't allow to retract :db/ident in
this way, so this runs afoul of future work to support mutating
metadata.
* Pre: s/VALUETYPE/VALUE_TYPE/.
This is consistent with the capitalization (which is "valueType") and
the other identifier.
* Pre: Remove some single quotes from error output.
* Part 1: Make materialized views be uniform [e a v value_type_tag].
This looks ahead to a time when we could support arbitrary
user-defined materialized views. For now, the "idents" materialized
view is those datoms of the form [e :db/ident :namespaced/keyword] and
the "schema" materialized view is those datoms of the form [e a v]
where a is in a particular set of attributes that will become clear in
the following commits.
This change is not backwards compatible, so I'm removing the open
current (really, v2) test. It'll be re-instated when we get to
https://github.com/mozilla/mentat/issues/194.
* Pre: Map TypedValue::Ref to TypedValue::Keyword in debug output.
* Part 3: Separate `schema_to_mutate` from the `schema` used to interpret.
This is just to keep track of the expected changes during
bootstrapping. I want bootstrap metadata mutations to flow through
the same code path as metadata mutations during regular transactions;
by differentiating the schema used for interpretation from the schema
that will be updated I expect to be able to apply bootstrap metadata
mutations to an empty schema and have things like materialized views
created (using the regular code paths).
This commit has been re-ordered for conceptual clarity, but it won't
compile because it references the metadata module. It's possible to
make it compile -- the functionality is there in the schema module --
but it's not worth the rebasing effort until after review (and
possibly not even then, since we'll squash down to a single commit to
land).
* Part 2: Maintain entids separately from idents.
In order to support historical idents, we need to distinguish the
"current" map from entid -> ident from the "complete historical" map
ident -> entid. This is what Datomic does; in Datomic, an ident is
never retracted (although it can be replaced). This approach is an
important part of allowing multiple consumers to share a schema
fragment as it migrates forward.
This fixes a limitation of the Clojure implementation, which did not
handle historical idents across knowledge base close and re-open.
The "entids" materialized view is naturally a slice of the "datoms"
table. The "idents" materialized view is a slice of the
"transactions" table. I hope that representing in this way, and
casting the problem in this light, might generalize to future
materialized views.
* Pre: Add DiffSet.
* Part 4: Collect mutations to a `Schema`.
I haven't taken your review comment about consuming AttributeBuilder
during each fluent function. If you read my response and still want
this, I'm happy to do it in review.
* Part 5: Handle :db/ident and :db.{install,alter}/attribute.
This "loops" the committed datoms out of the SQL store and back
through the metadata (schema, but in future also partition map)
processor. The metadata processor updates the schema and produces a
report of what changed; that report is then used to update the SQL
store. That update includes:
- the materialized views ("entids", "idents", and "schema");
- if needed, a subset of the datoms themselves (as flags change).
I've left a TODO for handling attribute retraction in the cases that
it makes sense. I expect that to be straight-forward.
* Review comment: Rename DiffSet to AddRetractAlterSet.
Also adds a little more commentary and a simple test.
* Review comment: Use ToIdent trait.
* Review comment: partially revert "Part 2: Maintain entids separately from idents."
This reverts commit 23a91df9c35e14398f2ddbd1ba25315821e67401.
Following our discussion, this removes the "entids" materialized
view. The next commit will remove historical idents from the "idents"
materialized view.
* Post: Use custom Either rather than std::result::Result.
This is not necessary, but it was suggested that we might be paying an
overhead creating Err instances while using error_chain. That seems
not to be the case, but this change shows that we don't actually use
any of the Result helper methods, so there's no reason to overload
Result. This change might avoid some future confusion, so I'm going
to land it anyway.
Signed-off-by: Nick Alexander <nalexander@mozilla.com>
* Review comment: Don't preserve historical idents.
* Review comment: More prepared statements when updating materialized views.
* Post: Test altering :db/cardinality and :db/unique.
These tests fail due to a Datomic limitation, namely that the marker
flag :db.alter/attribute can only be asserted once for an attribute!
That is, [:db.part/db :db.alter/attribute :attribute] will only be
transacted at most once. Since older versions of Datomic required the
:db.alter/attribute flag, I can only imagine they either never wrote
:db.alter/attribute to the store, or they handled it specially. I'll
need to remove the marker flag system from Mentat in order to address
this fundamental limitation.
* Post: Remove some more single quotes from error output.
* Post: Add assert_transact! macro to unwrap safely.
I was finding it very difficult to track unwrapping errors while
making changes, due to an underlying Mac OS X symbolication issue that
makes running tests with RUST_BACKTRACE=1 so slow that they all time
out.
* Post: Don't expect or recognize :db.{install,alter}/attribute.
I had this all working... except we will never see a repeated
`[:db.part/db :db.alter/attribute :attribute]` assertion in the store!
That means my approach would let you alter an attribute at most one
time. It's not worth hacking around this; it's better to just stop
expecting (and recognizing) the marker flags. (We have all the data
to distinguish the various cases that we need without the marker
flags.)
This brings Mentat in line with the thrust of newer Datomic versions,
but isn't compatible with Datomic, because (if I understand correctly)
Datomic automatically adds :db.{install,alter}/attribute assertions to
transactions.
I haven't purged the corresponding :db/ident and schema fragments just
yet:
- we might want them back
- we might want them in order to upgrade v1 and v2 databases to the
new on-disk layout we're fleshing out (v3?).
* Post: Don't make :db/unique :db.unique/* imply :db/index true.
This patch avoids a potential bug with the "schema" materialized view.
If :db/unique :db.unique/value implies :db/index true, then what
happens when you _retract_ :db.unique/value? I think Datomic defines
this in some way, but I really want the "schema" materialized view to
be a slice of "datoms" and not have these sort of ambiguities and
persistent effects. Therefore, to ensure that we don't retract a
schema characteristic and accidentally change more than we intended
to, this patch stops having any schema characteristic imply any other
schema characteristic(s). To achieve that, I added an
Option<Unique::{Value,Identity}> type to Attribute; this helps with
this patch, and also looks ahead to when we allow to retract
:db/unique attributes.
* Post: Allow to retract :db/ident.
* Post: Include more details about invalid schema changes.
The tests use strings, so they hide the chained errors which do in
fact provide more detail.
* Review comment: Fix outdated comment.
* Review comment: s/_SET/_SQL_LIST/.
* Review comment: Use a sub-select for checking cardinality.
This might be faster in practice.
* Review comment: Put `attribute::Unique` into its own namespace.
2017-03-20 20:18:59 +00:00
SELECT EXISTS
( SELECT 1
FROM datoms AS left , datoms AS right
WHERE left . a = ? AND
left . a = right . a AND
left . e = right . e AND
2019-07-17 15:46:47 +00:00
left . v < > right . v ) " #,
) ? ;
Schema alteration. Fixes #294 and #295. (#370) r=rnewman
* Pre: Don't retract :db/ident in test.
Datomic (and eventually Mentat) don't allow to retract :db/ident in
this way, so this runs afoul of future work to support mutating
metadata.
* Pre: s/VALUETYPE/VALUE_TYPE/.
This is consistent with the capitalization (which is "valueType") and
the other identifier.
* Pre: Remove some single quotes from error output.
* Part 1: Make materialized views be uniform [e a v value_type_tag].
This looks ahead to a time when we could support arbitrary
user-defined materialized views. For now, the "idents" materialized
view is those datoms of the form [e :db/ident :namespaced/keyword] and
the "schema" materialized view is those datoms of the form [e a v]
where a is in a particular set of attributes that will become clear in
the following commits.
This change is not backwards compatible, so I'm removing the open
current (really, v2) test. It'll be re-instated when we get to
https://github.com/mozilla/mentat/issues/194.
* Pre: Map TypedValue::Ref to TypedValue::Keyword in debug output.
* Part 3: Separate `schema_to_mutate` from the `schema` used to interpret.
This is just to keep track of the expected changes during
bootstrapping. I want bootstrap metadata mutations to flow through
the same code path as metadata mutations during regular transactions;
by differentiating the schema used for interpretation from the schema
that will be updated I expect to be able to apply bootstrap metadata
mutations to an empty schema and have things like materialized views
created (using the regular code paths).
This commit has been re-ordered for conceptual clarity, but it won't
compile because it references the metadata module. It's possible to
make it compile -- the functionality is there in the schema module --
but it's not worth the rebasing effort until after review (and
possibly not even then, since we'll squash down to a single commit to
land).
* Part 2: Maintain entids separately from idents.
In order to support historical idents, we need to distinguish the
"current" map from entid -> ident from the "complete historical" map
ident -> entid. This is what Datomic does; in Datomic, an ident is
never retracted (although it can be replaced). This approach is an
important part of allowing multiple consumers to share a schema
fragment as it migrates forward.
This fixes a limitation of the Clojure implementation, which did not
handle historical idents across knowledge base close and re-open.
The "entids" materialized view is naturally a slice of the "datoms"
table. The "idents" materialized view is a slice of the
"transactions" table. I hope that representing in this way, and
casting the problem in this light, might generalize to future
materialized views.
* Pre: Add DiffSet.
* Part 4: Collect mutations to a `Schema`.
I haven't taken your review comment about consuming AttributeBuilder
during each fluent function. If you read my response and still want
this, I'm happy to do it in review.
* Part 5: Handle :db/ident and :db.{install,alter}/attribute.
This "loops" the committed datoms out of the SQL store and back
through the metadata (schema, but in future also partition map)
processor. The metadata processor updates the schema and produces a
report of what changed; that report is then used to update the SQL
store. That update includes:
- the materialized views ("entids", "idents", and "schema");
- if needed, a subset of the datoms themselves (as flags change).
I've left a TODO for handling attribute retraction in the cases that
it makes sense. I expect that to be straight-forward.
* Review comment: Rename DiffSet to AddRetractAlterSet.
Also adds a little more commentary and a simple test.
* Review comment: Use ToIdent trait.
* Review comment: partially revert "Part 2: Maintain entids separately from idents."
This reverts commit 23a91df9c35e14398f2ddbd1ba25315821e67401.
Following our discussion, this removes the "entids" materialized
view. The next commit will remove historical idents from the "idents"
materialized view.
* Post: Use custom Either rather than std::result::Result.
This is not necessary, but it was suggested that we might be paying an
overhead creating Err instances while using error_chain. That seems
not to be the case, but this change shows that we don't actually use
any of the Result helper methods, so there's no reason to overload
Result. This change might avoid some future confusion, so I'm going
to land it anyway.
Signed-off-by: Nick Alexander <nalexander@mozilla.com>
* Review comment: Don't preserve historical idents.
* Review comment: More prepared statements when updating materialized views.
* Post: Test altering :db/cardinality and :db/unique.
These tests fail due to a Datomic limitation, namely that the marker
flag :db.alter/attribute can only be asserted once for an attribute!
That is, [:db.part/db :db.alter/attribute :attribute] will only be
transacted at most once. Since older versions of Datomic required the
:db.alter/attribute flag, I can only imagine they either never wrote
:db.alter/attribute to the store, or they handled it specially. I'll
need to remove the marker flag system from Mentat in order to address
this fundamental limitation.
* Post: Remove some more single quotes from error output.
* Post: Add assert_transact! macro to unwrap safely.
I was finding it very difficult to track unwrapping errors while
making changes, due to an underlying Mac OS X symbolication issue that
makes running tests with RUST_BACKTRACE=1 so slow that they all time
out.
* Post: Don't expect or recognize :db.{install,alter}/attribute.
I had this all working... except we will never see a repeated
`[:db.part/db :db.alter/attribute :attribute]` assertion in the store!
That means my approach would let you alter an attribute at most one
time. It's not worth hacking around this; it's better to just stop
expecting (and recognizing) the marker flags. (We have all the data
to distinguish the various cases that we need without the marker
flags.)
This brings Mentat in line with the thrust of newer Datomic versions,
but isn't compatible with Datomic, because (if I understand correctly)
Datomic automatically adds :db.{install,alter}/attribute assertions to
transactions.
I haven't purged the corresponding :db/ident and schema fragments just
yet:
- we might want them back
- we might want them in order to upgrade v1 and v2 databases to the
new on-disk layout we're fleshing out (v3?).
* Post: Don't make :db/unique :db.unique/* imply :db/index true.
This patch avoids a potential bug with the "schema" materialized view.
If :db/unique :db.unique/value implies :db/index true, then what
happens when you _retract_ :db.unique/value? I think Datomic defines
this in some way, but I really want the "schema" materialized view to
be a slice of "datoms" and not have these sort of ambiguities and
persistent effects. Therefore, to ensure that we don't retract a
schema characteristic and accidentally change more than we intended
to, this patch stops having any schema characteristic imply any other
schema characteristic(s). To achieve that, I added an
Option<Unique::{Value,Identity}> type to Attribute; this helps with
this patch, and also looks ahead to when we allow to retract
:db/unique attributes.
* Post: Allow to retract :db/ident.
* Post: Include more details about invalid schema changes.
The tests use strings, so they hide the chained errors which do in
fact provide more detail.
* Review comment: Fix outdated comment.
* Review comment: s/_SET/_SQL_LIST/.
* Review comment: Use a sub-select for checking cardinality.
This might be faster in practice.
* Review comment: Put `attribute::Unique` into its own namespace.
2017-03-20 20:18:59 +00:00
for ( & entid , alterations ) in & metadata_report . attributes_altered {
let attribute = new_schema . require_attribute_for_entid ( entid ) ? ;
for alteration in alterations {
match alteration {
& Index = > {
// This should always succeed.
2020-01-14 15:46:21 +00:00
index_stmt . execute ( & [ & attribute . index , & entid as & dyn ToSql ] ) ? ;
2019-07-17 15:46:47 +00:00
}
Schema alteration. Fixes #294 and #295. (#370) r=rnewman
* Pre: Don't retract :db/ident in test.
Datomic (and eventually Mentat) don't allow to retract :db/ident in
this way, so this runs afoul of future work to support mutating
metadata.
* Pre: s/VALUETYPE/VALUE_TYPE/.
This is consistent with the capitalization (which is "valueType") and
the other identifier.
* Pre: Remove some single quotes from error output.
* Part 1: Make materialized views be uniform [e a v value_type_tag].
This looks ahead to a time when we could support arbitrary
user-defined materialized views. For now, the "idents" materialized
view is those datoms of the form [e :db/ident :namespaced/keyword] and
the "schema" materialized view is those datoms of the form [e a v]
where a is in a particular set of attributes that will become clear in
the following commits.
This change is not backwards compatible, so I'm removing the open
current (really, v2) test. It'll be re-instated when we get to
https://github.com/mozilla/mentat/issues/194.
* Pre: Map TypedValue::Ref to TypedValue::Keyword in debug output.
* Part 3: Separate `schema_to_mutate` from the `schema` used to interpret.
This is just to keep track of the expected changes during
bootstrapping. I want bootstrap metadata mutations to flow through
the same code path as metadata mutations during regular transactions;
by differentiating the schema used for interpretation from the schema
that will be updated I expect to be able to apply bootstrap metadata
mutations to an empty schema and have things like materialized views
created (using the regular code paths).
This commit has been re-ordered for conceptual clarity, but it won't
compile because it references the metadata module. It's possible to
make it compile -- the functionality is there in the schema module --
but it's not worth the rebasing effort until after review (and
possibly not even then, since we'll squash down to a single commit to
land).
* Part 2: Maintain entids separately from idents.
In order to support historical idents, we need to distinguish the
"current" map from entid -> ident from the "complete historical" map
ident -> entid. This is what Datomic does; in Datomic, an ident is
never retracted (although it can be replaced). This approach is an
important part of allowing multiple consumers to share a schema
fragment as it migrates forward.
This fixes a limitation of the Clojure implementation, which did not
handle historical idents across knowledge base close and re-open.
The "entids" materialized view is naturally a slice of the "datoms"
table. The "idents" materialized view is a slice of the
"transactions" table. I hope that representing in this way, and
casting the problem in this light, might generalize to future
materialized views.
* Pre: Add DiffSet.
* Part 4: Collect mutations to a `Schema`.
I haven't taken your review comment about consuming AttributeBuilder
during each fluent function. If you read my response and still want
this, I'm happy to do it in review.
* Part 5: Handle :db/ident and :db.{install,alter}/attribute.
This "loops" the committed datoms out of the SQL store and back
through the metadata (schema, but in future also partition map)
processor. The metadata processor updates the schema and produces a
report of what changed; that report is then used to update the SQL
store. That update includes:
- the materialized views ("entids", "idents", and "schema");
- if needed, a subset of the datoms themselves (as flags change).
I've left a TODO for handling attribute retraction in the cases that
it makes sense. I expect that to be straight-forward.
* Review comment: Rename DiffSet to AddRetractAlterSet.
Also adds a little more commentary and a simple test.
* Review comment: Use ToIdent trait.
* Review comment: partially revert "Part 2: Maintain entids separately from idents."
This reverts commit 23a91df9c35e14398f2ddbd1ba25315821e67401.
Following our discussion, this removes the "entids" materialized
view. The next commit will remove historical idents from the "idents"
materialized view.
* Post: Use custom Either rather than std::result::Result.
This is not necessary, but it was suggested that we might be paying an
overhead creating Err instances while using error_chain. That seems
not to be the case, but this change shows that we don't actually use
any of the Result helper methods, so there's no reason to overload
Result. This change might avoid some future confusion, so I'm going
to land it anyway.
Signed-off-by: Nick Alexander <nalexander@mozilla.com>
* Review comment: Don't preserve historical idents.
* Review comment: More prepared statements when updating materialized views.
* Post: Test altering :db/cardinality and :db/unique.
These tests fail due to a Datomic limitation, namely that the marker
flag :db.alter/attribute can only be asserted once for an attribute!
That is, [:db.part/db :db.alter/attribute :attribute] will only be
transacted at most once. Since older versions of Datomic required the
:db.alter/attribute flag, I can only imagine they either never wrote
:db.alter/attribute to the store, or they handled it specially. I'll
need to remove the marker flag system from Mentat in order to address
this fundamental limitation.
* Post: Remove some more single quotes from error output.
* Post: Add assert_transact! macro to unwrap safely.
I was finding it very difficult to track unwrapping errors while
making changes, due to an underlying Mac OS X symbolication issue that
makes running tests with RUST_BACKTRACE=1 so slow that they all time
out.
* Post: Don't expect or recognize :db.{install,alter}/attribute.
I had this all working... except we will never see a repeated
`[:db.part/db :db.alter/attribute :attribute]` assertion in the store!
That means my approach would let you alter an attribute at most one
time. It's not worth hacking around this; it's better to just stop
expecting (and recognizing) the marker flags. (We have all the data
to distinguish the various cases that we need without the marker
flags.)
This brings Mentat in line with the thrust of newer Datomic versions,
but isn't compatible with Datomic, because (if I understand correctly)
Datomic automatically adds :db.{install,alter}/attribute assertions to
transactions.
I haven't purged the corresponding :db/ident and schema fragments just
yet:
- we might want them back
- we might want them in order to upgrade v1 and v2 databases to the
new on-disk layout we're fleshing out (v3?).
* Post: Don't make :db/unique :db.unique/* imply :db/index true.
This patch avoids a potential bug with the "schema" materialized view.
If :db/unique :db.unique/value implies :db/index true, then what
happens when you _retract_ :db.unique/value? I think Datomic defines
this in some way, but I really want the "schema" materialized view to
be a slice of "datoms" and not have these sort of ambiguities and
persistent effects. Therefore, to ensure that we don't retract a
schema characteristic and accidentally change more than we intended
to, this patch stops having any schema characteristic imply any other
schema characteristic(s). To achieve that, I added an
Option<Unique::{Value,Identity}> type to Attribute; this helps with
this patch, and also looks ahead to when we allow to retract
:db/unique attributes.
* Post: Allow to retract :db/ident.
* Post: Include more details about invalid schema changes.
The tests use strings, so they hide the chained errors which do in
fact provide more detail.
* Review comment: Fix outdated comment.
* Review comment: s/_SET/_SQL_LIST/.
* Review comment: Use a sub-select for checking cardinality.
This might be faster in practice.
* Review comment: Put `attribute::Unique` into its own namespace.
2017-03-20 20:18:59 +00:00
& Unique = > {
// TODO: This can fail if there are conflicting values; give a more helpful
// error message in this case.
2019-07-17 15:46:47 +00:00
if unique_value_stmt
2020-01-14 15:46:21 +00:00
. execute ( & [
to_bool_ref ( attribute . unique . is_some ( ) ) ,
& entid as & dyn ToSql ,
] )
2019-07-17 15:46:47 +00:00
. is_err ( )
{
Schema alteration. Fixes #294 and #295. (#370) r=rnewman
* Pre: Don't retract :db/ident in test.
Datomic (and eventually Mentat) don't allow to retract :db/ident in
this way, so this runs afoul of future work to support mutating
metadata.
* Pre: s/VALUETYPE/VALUE_TYPE/.
This is consistent with the capitalization (which is "valueType") and
the other identifier.
* Pre: Remove some single quotes from error output.
* Part 1: Make materialized views be uniform [e a v value_type_tag].
This looks ahead to a time when we could support arbitrary
user-defined materialized views. For now, the "idents" materialized
view is those datoms of the form [e :db/ident :namespaced/keyword] and
the "schema" materialized view is those datoms of the form [e a v]
where a is in a particular set of attributes that will become clear in
the following commits.
This change is not backwards compatible, so I'm removing the open
current (really, v2) test. It'll be re-instated when we get to
https://github.com/mozilla/mentat/issues/194.
* Pre: Map TypedValue::Ref to TypedValue::Keyword in debug output.
* Part 3: Separate `schema_to_mutate` from the `schema` used to interpret.
This is just to keep track of the expected changes during
bootstrapping. I want bootstrap metadata mutations to flow through
the same code path as metadata mutations during regular transactions;
by differentiating the schema used for interpretation from the schema
that will be updated I expect to be able to apply bootstrap metadata
mutations to an empty schema and have things like materialized views
created (using the regular code paths).
This commit has been re-ordered for conceptual clarity, but it won't
compile because it references the metadata module. It's possible to
make it compile -- the functionality is there in the schema module --
but it's not worth the rebasing effort until after review (and
possibly not even then, since we'll squash down to a single commit to
land).
* Part 2: Maintain entids separately from idents.
In order to support historical idents, we need to distinguish the
"current" map from entid -> ident from the "complete historical" map
ident -> entid. This is what Datomic does; in Datomic, an ident is
never retracted (although it can be replaced). This approach is an
important part of allowing multiple consumers to share a schema
fragment as it migrates forward.
This fixes a limitation of the Clojure implementation, which did not
handle historical idents across knowledge base close and re-open.
The "entids" materialized view is naturally a slice of the "datoms"
table. The "idents" materialized view is a slice of the
"transactions" table. I hope that representing in this way, and
casting the problem in this light, might generalize to future
materialized views.
* Pre: Add DiffSet.
* Part 4: Collect mutations to a `Schema`.
I haven't taken your review comment about consuming AttributeBuilder
during each fluent function. If you read my response and still want
this, I'm happy to do it in review.
* Part 5: Handle :db/ident and :db.{install,alter}/attribute.
This "loops" the committed datoms out of the SQL store and back
through the metadata (schema, but in future also partition map)
processor. The metadata processor updates the schema and produces a
report of what changed; that report is then used to update the SQL
store. That update includes:
- the materialized views ("entids", "idents", and "schema");
- if needed, a subset of the datoms themselves (as flags change).
I've left a TODO for handling attribute retraction in the cases that
it makes sense. I expect that to be straight-forward.
* Review comment: Rename DiffSet to AddRetractAlterSet.
Also adds a little more commentary and a simple test.
* Review comment: Use ToIdent trait.
* Review comment: partially revert "Part 2: Maintain entids separately from idents."
This reverts commit 23a91df9c35e14398f2ddbd1ba25315821e67401.
Following our discussion, this removes the "entids" materialized
view. The next commit will remove historical idents from the "idents"
materialized view.
* Post: Use custom Either rather than std::result::Result.
This is not necessary, but it was suggested that we might be paying an
overhead creating Err instances while using error_chain. That seems
not to be the case, but this change shows that we don't actually use
any of the Result helper methods, so there's no reason to overload
Result. This change might avoid some future confusion, so I'm going
to land it anyway.
Signed-off-by: Nick Alexander <nalexander@mozilla.com>
* Review comment: Don't preserve historical idents.
* Review comment: More prepared statements when updating materialized views.
* Post: Test altering :db/cardinality and :db/unique.
These tests fail due to a Datomic limitation, namely that the marker
flag :db.alter/attribute can only be asserted once for an attribute!
That is, [:db.part/db :db.alter/attribute :attribute] will only be
transacted at most once. Since older versions of Datomic required the
:db.alter/attribute flag, I can only imagine they either never wrote
:db.alter/attribute to the store, or they handled it specially. I'll
need to remove the marker flag system from Mentat in order to address
this fundamental limitation.
* Post: Remove some more single quotes from error output.
* Post: Add assert_transact! macro to unwrap safely.
I was finding it very difficult to track unwrapping errors while
making changes, due to an underlying Mac OS X symbolication issue that
makes running tests with RUST_BACKTRACE=1 so slow that they all time
out.
* Post: Don't expect or recognize :db.{install,alter}/attribute.
I had this all working... except we will never see a repeated
`[:db.part/db :db.alter/attribute :attribute]` assertion in the store!
That means my approach would let you alter an attribute at most one
time. It's not worth hacking around this; it's better to just stop
expecting (and recognizing) the marker flags. (We have all the data
to distinguish the various cases that we need without the marker
flags.)
This brings Mentat in line with the thrust of newer Datomic versions,
but isn't compatible with Datomic, because (if I understand correctly)
Datomic automatically adds :db.{install,alter}/attribute assertions to
transactions.
I haven't purged the corresponding :db/ident and schema fragments just
yet:
- we might want them back
- we might want them in order to upgrade v1 and v2 databases to the
new on-disk layout we're fleshing out (v3?).
* Post: Don't make :db/unique :db.unique/* imply :db/index true.
This patch avoids a potential bug with the "schema" materialized view.
If :db/unique :db.unique/value implies :db/index true, then what
happens when you _retract_ :db.unique/value? I think Datomic defines
this in some way, but I really want the "schema" materialized view to
be a slice of "datoms" and not have these sort of ambiguities and
persistent effects. Therefore, to ensure that we don't retract a
schema characteristic and accidentally change more than we intended
to, this patch stops having any schema characteristic imply any other
schema characteristic(s). To achieve that, I added an
Option<Unique::{Value,Identity}> type to Attribute; this helps with
this patch, and also looks ahead to when we allow to retract
:db/unique attributes.
* Post: Allow to retract :db/ident.
* Post: Include more details about invalid schema changes.
The tests use strings, so they hide the chained errors which do in
fact provide more detail.
* Review comment: Fix outdated comment.
* Review comment: s/_SET/_SQL_LIST/.
* Review comment: Use a sub-select for checking cardinality.
This might be faster in practice.
* Review comment: Put `attribute::Unique` into its own namespace.
2017-03-20 20:18:59 +00:00
match attribute . unique {
2019-07-17 15:46:47 +00:00
Some ( attribute ::Unique ::Value ) = > {
bail! ( DbErrorKind ::SchemaAlterationFailed ( format! (
" Cannot alter schema attribute {} to be :db.unique/value " ,
entid
) ) )
}
Some ( attribute ::Unique ::Identity ) = > {
bail! ( DbErrorKind ::SchemaAlterationFailed ( format! (
" Cannot alter schema attribute {} to be :db.unique/identity " ,
entid
) ) )
}
Schema alteration. Fixes #294 and #295. (#370) r=rnewman
* Pre: Don't retract :db/ident in test.
Datomic (and eventually Mentat) don't allow to retract :db/ident in
this way, so this runs afoul of future work to support mutating
metadata.
* Pre: s/VALUETYPE/VALUE_TYPE/.
This is consistent with the capitalization (which is "valueType") and
the other identifier.
* Pre: Remove some single quotes from error output.
* Part 1: Make materialized views be uniform [e a v value_type_tag].
This looks ahead to a time when we could support arbitrary
user-defined materialized views. For now, the "idents" materialized
view is those datoms of the form [e :db/ident :namespaced/keyword] and
the "schema" materialized view is those datoms of the form [e a v]
where a is in a particular set of attributes that will become clear in
the following commits.
This change is not backwards compatible, so I'm removing the open
current (really, v2) test. It'll be re-instated when we get to
https://github.com/mozilla/mentat/issues/194.
* Pre: Map TypedValue::Ref to TypedValue::Keyword in debug output.
* Part 3: Separate `schema_to_mutate` from the `schema` used to interpret.
This is just to keep track of the expected changes during
bootstrapping. I want bootstrap metadata mutations to flow through
the same code path as metadata mutations during regular transactions;
by differentiating the schema used for interpretation from the schema
that will be updated I expect to be able to apply bootstrap metadata
mutations to an empty schema and have things like materialized views
created (using the regular code paths).
This commit has been re-ordered for conceptual clarity, but it won't
compile because it references the metadata module. It's possible to
make it compile -- the functionality is there in the schema module --
but it's not worth the rebasing effort until after review (and
possibly not even then, since we'll squash down to a single commit to
land).
* Part 2: Maintain entids separately from idents.
In order to support historical idents, we need to distinguish the
"current" map from entid -> ident from the "complete historical" map
ident -> entid. This is what Datomic does; in Datomic, an ident is
never retracted (although it can be replaced). This approach is an
important part of allowing multiple consumers to share a schema
fragment as it migrates forward.
This fixes a limitation of the Clojure implementation, which did not
handle historical idents across knowledge base close and re-open.
The "entids" materialized view is naturally a slice of the "datoms"
table. The "idents" materialized view is a slice of the
"transactions" table. I hope that representing in this way, and
casting the problem in this light, might generalize to future
materialized views.
* Pre: Add DiffSet.
* Part 4: Collect mutations to a `Schema`.
I haven't taken your review comment about consuming AttributeBuilder
during each fluent function. If you read my response and still want
this, I'm happy to do it in review.
* Part 5: Handle :db/ident and :db.{install,alter}/attribute.
This "loops" the committed datoms out of the SQL store and back
through the metadata (schema, but in future also partition map)
processor. The metadata processor updates the schema and produces a
report of what changed; that report is then used to update the SQL
store. That update includes:
- the materialized views ("entids", "idents", and "schema");
- if needed, a subset of the datoms themselves (as flags change).
I've left a TODO for handling attribute retraction in the cases that
it makes sense. I expect that to be straight-forward.
* Review comment: Rename DiffSet to AddRetractAlterSet.
Also adds a little more commentary and a simple test.
* Review comment: Use ToIdent trait.
* Review comment: partially revert "Part 2: Maintain entids separately from idents."
This reverts commit 23a91df9c35e14398f2ddbd1ba25315821e67401.
Following our discussion, this removes the "entids" materialized
view. The next commit will remove historical idents from the "idents"
materialized view.
* Post: Use custom Either rather than std::result::Result.
This is not necessary, but it was suggested that we might be paying an
overhead creating Err instances while using error_chain. That seems
not to be the case, but this change shows that we don't actually use
any of the Result helper methods, so there's no reason to overload
Result. This change might avoid some future confusion, so I'm going
to land it anyway.
Signed-off-by: Nick Alexander <nalexander@mozilla.com>
* Review comment: Don't preserve historical idents.
* Review comment: More prepared statements when updating materialized views.
* Post: Test altering :db/cardinality and :db/unique.
These tests fail due to a Datomic limitation, namely that the marker
flag :db.alter/attribute can only be asserted once for an attribute!
That is, [:db.part/db :db.alter/attribute :attribute] will only be
transacted at most once. Since older versions of Datomic required the
:db.alter/attribute flag, I can only imagine they either never wrote
:db.alter/attribute to the store, or they handled it specially. I'll
need to remove the marker flag system from Mentat in order to address
this fundamental limitation.
* Post: Remove some more single quotes from error output.
* Post: Add assert_transact! macro to unwrap safely.
I was finding it very difficult to track unwrapping errors while
making changes, due to an underlying Mac OS X symbolication issue that
makes running tests with RUST_BACKTRACE=1 so slow that they all time
out.
* Post: Don't expect or recognize :db.{install,alter}/attribute.
I had this all working... except we will never see a repeated
`[:db.part/db :db.alter/attribute :attribute]` assertion in the store!
That means my approach would let you alter an attribute at most one
time. It's not worth hacking around this; it's better to just stop
expecting (and recognizing) the marker flags. (We have all the data
to distinguish the various cases that we need without the marker
flags.)
This brings Mentat in line with the thrust of newer Datomic versions,
but isn't compatible with Datomic, because (if I understand correctly)
Datomic automatically adds :db.{install,alter}/attribute assertions to
transactions.
I haven't purged the corresponding :db/ident and schema fragments just
yet:
- we might want them back
- we might want them in order to upgrade v1 and v2 databases to the
new on-disk layout we're fleshing out (v3?).
* Post: Don't make :db/unique :db.unique/* imply :db/index true.
This patch avoids a potential bug with the "schema" materialized view.
If :db/unique :db.unique/value implies :db/index true, then what
happens when you _retract_ :db.unique/value? I think Datomic defines
this in some way, but I really want the "schema" materialized view to
be a slice of "datoms" and not have these sort of ambiguities and
persistent effects. Therefore, to ensure that we don't retract a
schema characteristic and accidentally change more than we intended
to, this patch stops having any schema characteristic imply any other
schema characteristic(s). To achieve that, I added an
Option<Unique::{Value,Identity}> type to Attribute; this helps with
this patch, and also looks ahead to when we allow to retract
:db/unique attributes.
* Post: Allow to retract :db/ident.
* Post: Include more details about invalid schema changes.
The tests use strings, so they hide the chained errors which do in
fact provide more detail.
* Review comment: Fix outdated comment.
* Review comment: s/_SET/_SQL_LIST/.
* Review comment: Use a sub-select for checking cardinality.
This might be faster in practice.
* Review comment: Put `attribute::Unique` into its own namespace.
2017-03-20 20:18:59 +00:00
None = > unreachable! ( ) , // This shouldn't happen, even after we support removing :db/unique.
}
}
2019-07-17 15:46:47 +00:00
}
Schema alteration. Fixes #294 and #295. (#370) r=rnewman
* Pre: Don't retract :db/ident in test.
Datomic (and eventually Mentat) don't allow to retract :db/ident in
this way, so this runs afoul of future work to support mutating
metadata.
* Pre: s/VALUETYPE/VALUE_TYPE/.
This is consistent with the capitalization (which is "valueType") and
the other identifier.
* Pre: Remove some single quotes from error output.
* Part 1: Make materialized views be uniform [e a v value_type_tag].
This looks ahead to a time when we could support arbitrary
user-defined materialized views. For now, the "idents" materialized
view is those datoms of the form [e :db/ident :namespaced/keyword] and
the "schema" materialized view is those datoms of the form [e a v]
where a is in a particular set of attributes that will become clear in
the following commits.
This change is not backwards compatible, so I'm removing the open
current (really, v2) test. It'll be re-instated when we get to
https://github.com/mozilla/mentat/issues/194.
* Pre: Map TypedValue::Ref to TypedValue::Keyword in debug output.
* Part 3: Separate `schema_to_mutate` from the `schema` used to interpret.
This is just to keep track of the expected changes during
bootstrapping. I want bootstrap metadata mutations to flow through
the same code path as metadata mutations during regular transactions;
by differentiating the schema used for interpretation from the schema
that will be updated I expect to be able to apply bootstrap metadata
mutations to an empty schema and have things like materialized views
created (using the regular code paths).
This commit has been re-ordered for conceptual clarity, but it won't
compile because it references the metadata module. It's possible to
make it compile -- the functionality is there in the schema module --
but it's not worth the rebasing effort until after review (and
possibly not even then, since we'll squash down to a single commit to
land).
* Part 2: Maintain entids separately from idents.
In order to support historical idents, we need to distinguish the
"current" map from entid -> ident from the "complete historical" map
ident -> entid. This is what Datomic does; in Datomic, an ident is
never retracted (although it can be replaced). This approach is an
important part of allowing multiple consumers to share a schema
fragment as it migrates forward.
This fixes a limitation of the Clojure implementation, which did not
handle historical idents across knowledge base close and re-open.
The "entids" materialized view is naturally a slice of the "datoms"
table. The "idents" materialized view is a slice of the
"transactions" table. I hope that representing in this way, and
casting the problem in this light, might generalize to future
materialized views.
* Pre: Add DiffSet.
* Part 4: Collect mutations to a `Schema`.
I haven't taken your review comment about consuming AttributeBuilder
during each fluent function. If you read my response and still want
this, I'm happy to do it in review.
* Part 5: Handle :db/ident and :db.{install,alter}/attribute.
This "loops" the committed datoms out of the SQL store and back
through the metadata (schema, but in future also partition map)
processor. The metadata processor updates the schema and produces a
report of what changed; that report is then used to update the SQL
store. That update includes:
- the materialized views ("entids", "idents", and "schema");
- if needed, a subset of the datoms themselves (as flags change).
I've left a TODO for handling attribute retraction in the cases that
it makes sense. I expect that to be straight-forward.
* Review comment: Rename DiffSet to AddRetractAlterSet.
Also adds a little more commentary and a simple test.
* Review comment: Use ToIdent trait.
* Review comment: partially revert "Part 2: Maintain entids separately from idents."
This reverts commit 23a91df9c35e14398f2ddbd1ba25315821e67401.
Following our discussion, this removes the "entids" materialized
view. The next commit will remove historical idents from the "idents"
materialized view.
* Post: Use custom Either rather than std::result::Result.
This is not necessary, but it was suggested that we might be paying an
overhead creating Err instances while using error_chain. That seems
not to be the case, but this change shows that we don't actually use
any of the Result helper methods, so there's no reason to overload
Result. This change might avoid some future confusion, so I'm going
to land it anyway.
Signed-off-by: Nick Alexander <nalexander@mozilla.com>
* Review comment: Don't preserve historical idents.
* Review comment: More prepared statements when updating materialized views.
* Post: Test altering :db/cardinality and :db/unique.
These tests fail due to a Datomic limitation, namely that the marker
flag :db.alter/attribute can only be asserted once for an attribute!
That is, [:db.part/db :db.alter/attribute :attribute] will only be
transacted at most once. Since older versions of Datomic required the
:db.alter/attribute flag, I can only imagine they either never wrote
:db.alter/attribute to the store, or they handled it specially. I'll
need to remove the marker flag system from Mentat in order to address
this fundamental limitation.
* Post: Remove some more single quotes from error output.
* Post: Add assert_transact! macro to unwrap safely.
I was finding it very difficult to track unwrapping errors while
making changes, due to an underlying Mac OS X symbolication issue that
makes running tests with RUST_BACKTRACE=1 so slow that they all time
out.
* Post: Don't expect or recognize :db.{install,alter}/attribute.
I had this all working... except we will never see a repeated
`[:db.part/db :db.alter/attribute :attribute]` assertion in the store!
That means my approach would let you alter an attribute at most one
time. It's not worth hacking around this; it's better to just stop
expecting (and recognizing) the marker flags. (We have all the data
to distinguish the various cases that we need without the marker
flags.)
This brings Mentat in line with the thrust of newer Datomic versions,
but isn't compatible with Datomic, because (if I understand correctly)
Datomic automatically adds :db.{install,alter}/attribute assertions to
transactions.
I haven't purged the corresponding :db/ident and schema fragments just
yet:
- we might want them back
- we might want them in order to upgrade v1 and v2 databases to the
new on-disk layout we're fleshing out (v3?).
* Post: Don't make :db/unique :db.unique/* imply :db/index true.
This patch avoids a potential bug with the "schema" materialized view.
If :db/unique :db.unique/value implies :db/index true, then what
happens when you _retract_ :db.unique/value? I think Datomic defines
this in some way, but I really want the "schema" materialized view to
be a slice of "datoms" and not have these sort of ambiguities and
persistent effects. Therefore, to ensure that we don't retract a
schema characteristic and accidentally change more than we intended
to, this patch stops having any schema characteristic imply any other
schema characteristic(s). To achieve that, I added an
Option<Unique::{Value,Identity}> type to Attribute; this helps with
this patch, and also looks ahead to when we allow to retract
:db/unique attributes.
* Post: Allow to retract :db/ident.
* Post: Include more details about invalid schema changes.
The tests use strings, so they hide the chained errors which do in
fact provide more detail.
* Review comment: Fix outdated comment.
* Review comment: s/_SET/_SQL_LIST/.
* Review comment: Use a sub-select for checking cardinality.
This might be faster in practice.
* Review comment: Put `attribute::Unique` into its own namespace.
2017-03-20 20:18:59 +00:00
& Cardinality = > {
// We can always go from :db.cardinality/one to :db.cardinality many. It's
// :db.cardinality/many to :db.cardinality/one that can fail.
//
// TODO: improve the failure message. Perhaps try to mimic what Datomic says in
// this case?
if ! attribute . multival {
2020-01-14 15:46:21 +00:00
let mut rows = cardinality_stmt . query ( & [ & entid as & dyn ToSql ] ) ? ;
2019-07-17 15:46:47 +00:00
if rows . next ( ) ? . is_some ( ) {
bail! ( DbErrorKind ::SchemaAlterationFailed ( format! (
" Cannot alter schema attribute {} to be :db.cardinality/one " ,
entid
) ) ) ;
Schema alteration. Fixes #294 and #295. (#370) r=rnewman
* Pre: Don't retract :db/ident in test.
Datomic (and eventually Mentat) don't allow to retract :db/ident in
this way, so this runs afoul of future work to support mutating
metadata.
* Pre: s/VALUETYPE/VALUE_TYPE/.
This is consistent with the capitalization (which is "valueType") and
the other identifier.
* Pre: Remove some single quotes from error output.
* Part 1: Make materialized views be uniform [e a v value_type_tag].
This looks ahead to a time when we could support arbitrary
user-defined materialized views. For now, the "idents" materialized
view is those datoms of the form [e :db/ident :namespaced/keyword] and
the "schema" materialized view is those datoms of the form [e a v]
where a is in a particular set of attributes that will become clear in
the following commits.
This change is not backwards compatible, so I'm removing the open
current (really, v2) test. It'll be re-instated when we get to
https://github.com/mozilla/mentat/issues/194.
* Pre: Map TypedValue::Ref to TypedValue::Keyword in debug output.
* Part 3: Separate `schema_to_mutate` from the `schema` used to interpret.
This is just to keep track of the expected changes during
bootstrapping. I want bootstrap metadata mutations to flow through
the same code path as metadata mutations during regular transactions;
by differentiating the schema used for interpretation from the schema
that will be updated I expect to be able to apply bootstrap metadata
mutations to an empty schema and have things like materialized views
created (using the regular code paths).
This commit has been re-ordered for conceptual clarity, but it won't
compile because it references the metadata module. It's possible to
make it compile -- the functionality is there in the schema module --
but it's not worth the rebasing effort until after review (and
possibly not even then, since we'll squash down to a single commit to
land).
* Part 2: Maintain entids separately from idents.
In order to support historical idents, we need to distinguish the
"current" map from entid -> ident from the "complete historical" map
ident -> entid. This is what Datomic does; in Datomic, an ident is
never retracted (although it can be replaced). This approach is an
important part of allowing multiple consumers to share a schema
fragment as it migrates forward.
This fixes a limitation of the Clojure implementation, which did not
handle historical idents across knowledge base close and re-open.
The "entids" materialized view is naturally a slice of the "datoms"
table. The "idents" materialized view is a slice of the
"transactions" table. I hope that representing in this way, and
casting the problem in this light, might generalize to future
materialized views.
* Pre: Add DiffSet.
* Part 4: Collect mutations to a `Schema`.
I haven't taken your review comment about consuming AttributeBuilder
during each fluent function. If you read my response and still want
this, I'm happy to do it in review.
* Part 5: Handle :db/ident and :db.{install,alter}/attribute.
This "loops" the committed datoms out of the SQL store and back
through the metadata (schema, but in future also partition map)
processor. The metadata processor updates the schema and produces a
report of what changed; that report is then used to update the SQL
store. That update includes:
- the materialized views ("entids", "idents", and "schema");
- if needed, a subset of the datoms themselves (as flags change).
I've left a TODO for handling attribute retraction in the cases that
it makes sense. I expect that to be straight-forward.
* Review comment: Rename DiffSet to AddRetractAlterSet.
Also adds a little more commentary and a simple test.
* Review comment: Use ToIdent trait.
* Review comment: partially revert "Part 2: Maintain entids separately from idents."
This reverts commit 23a91df9c35e14398f2ddbd1ba25315821e67401.
Following our discussion, this removes the "entids" materialized
view. The next commit will remove historical idents from the "idents"
materialized view.
* Post: Use custom Either rather than std::result::Result.
This is not necessary, but it was suggested that we might be paying an
overhead creating Err instances while using error_chain. That seems
not to be the case, but this change shows that we don't actually use
any of the Result helper methods, so there's no reason to overload
Result. This change might avoid some future confusion, so I'm going
to land it anyway.
Signed-off-by: Nick Alexander <nalexander@mozilla.com>
* Review comment: Don't preserve historical idents.
* Review comment: More prepared statements when updating materialized views.
* Post: Test altering :db/cardinality and :db/unique.
These tests fail due to a Datomic limitation, namely that the marker
flag :db.alter/attribute can only be asserted once for an attribute!
That is, [:db.part/db :db.alter/attribute :attribute] will only be
transacted at most once. Since older versions of Datomic required the
:db.alter/attribute flag, I can only imagine they either never wrote
:db.alter/attribute to the store, or they handled it specially. I'll
need to remove the marker flag system from Mentat in order to address
this fundamental limitation.
* Post: Remove some more single quotes from error output.
* Post: Add assert_transact! macro to unwrap safely.
I was finding it very difficult to track unwrapping errors while
making changes, due to an underlying Mac OS X symbolication issue that
makes running tests with RUST_BACKTRACE=1 so slow that they all time
out.
* Post: Don't expect or recognize :db.{install,alter}/attribute.
I had this all working... except we will never see a repeated
`[:db.part/db :db.alter/attribute :attribute]` assertion in the store!
That means my approach would let you alter an attribute at most one
time. It's not worth hacking around this; it's better to just stop
expecting (and recognizing) the marker flags. (We have all the data
to distinguish the various cases that we need without the marker
flags.)
This brings Mentat in line with the thrust of newer Datomic versions,
but isn't compatible with Datomic, because (if I understand correctly)
Datomic automatically adds :db.{install,alter}/attribute assertions to
transactions.
I haven't purged the corresponding :db/ident and schema fragments just
yet:
- we might want them back
- we might want them in order to upgrade v1 and v2 databases to the
new on-disk layout we're fleshing out (v3?).
* Post: Don't make :db/unique :db.unique/* imply :db/index true.
This patch avoids a potential bug with the "schema" materialized view.
If :db/unique :db.unique/value implies :db/index true, then what
happens when you _retract_ :db.unique/value? I think Datomic defines
this in some way, but I really want the "schema" materialized view to
be a slice of "datoms" and not have these sort of ambiguities and
persistent effects. Therefore, to ensure that we don't retract a
schema characteristic and accidentally change more than we intended
to, this patch stops having any schema characteristic imply any other
schema characteristic(s). To achieve that, I added an
Option<Unique::{Value,Identity}> type to Attribute; this helps with
this patch, and also looks ahead to when we allow to retract
:db/unique attributes.
* Post: Allow to retract :db/ident.
* Post: Include more details about invalid schema changes.
The tests use strings, so they hide the chained errors which do in
fact provide more detail.
* Review comment: Fix outdated comment.
* Review comment: s/_SET/_SQL_LIST/.
* Review comment: Use a sub-select for checking cardinality.
This might be faster in practice.
* Review comment: Put `attribute::Unique` into its own namespace.
2017-03-20 20:18:59 +00:00
}
}
2019-07-17 15:46:47 +00:00
}
Schema alteration. Fixes #294 and #295. (#370) r=rnewman
* Pre: Don't retract :db/ident in test.
Datomic (and eventually Mentat) don't allow to retract :db/ident in
this way, so this runs afoul of future work to support mutating
metadata.
* Pre: s/VALUETYPE/VALUE_TYPE/.
This is consistent with the capitalization (which is "valueType") and
the other identifier.
* Pre: Remove some single quotes from error output.
* Part 1: Make materialized views be uniform [e a v value_type_tag].
This looks ahead to a time when we could support arbitrary
user-defined materialized views. For now, the "idents" materialized
view is those datoms of the form [e :db/ident :namespaced/keyword] and
the "schema" materialized view is those datoms of the form [e a v]
where a is in a particular set of attributes that will become clear in
the following commits.
This change is not backwards compatible, so I'm removing the open
current (really, v2) test. It'll be re-instated when we get to
https://github.com/mozilla/mentat/issues/194.
* Pre: Map TypedValue::Ref to TypedValue::Keyword in debug output.
* Part 3: Separate `schema_to_mutate` from the `schema` used to interpret.
This is just to keep track of the expected changes during
bootstrapping. I want bootstrap metadata mutations to flow through
the same code path as metadata mutations during regular transactions;
by differentiating the schema used for interpretation from the schema
that will be updated I expect to be able to apply bootstrap metadata
mutations to an empty schema and have things like materialized views
created (using the regular code paths).
This commit has been re-ordered for conceptual clarity, but it won't
compile because it references the metadata module. It's possible to
make it compile -- the functionality is there in the schema module --
but it's not worth the rebasing effort until after review (and
possibly not even then, since we'll squash down to a single commit to
land).
* Part 2: Maintain entids separately from idents.
In order to support historical idents, we need to distinguish the
"current" map from entid -> ident from the "complete historical" map
ident -> entid. This is what Datomic does; in Datomic, an ident is
never retracted (although it can be replaced). This approach is an
important part of allowing multiple consumers to share a schema
fragment as it migrates forward.
This fixes a limitation of the Clojure implementation, which did not
handle historical idents across knowledge base close and re-open.
The "entids" materialized view is naturally a slice of the "datoms"
table. The "idents" materialized view is a slice of the
"transactions" table. I hope that representing in this way, and
casting the problem in this light, might generalize to future
materialized views.
* Pre: Add DiffSet.
* Part 4: Collect mutations to a `Schema`.
I haven't taken your review comment about consuming AttributeBuilder
during each fluent function. If you read my response and still want
this, I'm happy to do it in review.
* Part 5: Handle :db/ident and :db.{install,alter}/attribute.
This "loops" the committed datoms out of the SQL store and back
through the metadata (schema, but in future also partition map)
processor. The metadata processor updates the schema and produces a
report of what changed; that report is then used to update the SQL
store. That update includes:
- the materialized views ("entids", "idents", and "schema");
- if needed, a subset of the datoms themselves (as flags change).
I've left a TODO for handling attribute retraction in the cases that
it makes sense. I expect that to be straight-forward.
* Review comment: Rename DiffSet to AddRetractAlterSet.
Also adds a little more commentary and a simple test.
* Review comment: Use ToIdent trait.
* Review comment: partially revert "Part 2: Maintain entids separately from idents."
This reverts commit 23a91df9c35e14398f2ddbd1ba25315821e67401.
Following our discussion, this removes the "entids" materialized
view. The next commit will remove historical idents from the "idents"
materialized view.
* Post: Use custom Either rather than std::result::Result.
This is not necessary, but it was suggested that we might be paying an
overhead creating Err instances while using error_chain. That seems
not to be the case, but this change shows that we don't actually use
any of the Result helper methods, so there's no reason to overload
Result. This change might avoid some future confusion, so I'm going
to land it anyway.
Signed-off-by: Nick Alexander <nalexander@mozilla.com>
* Review comment: Don't preserve historical idents.
* Review comment: More prepared statements when updating materialized views.
* Post: Test altering :db/cardinality and :db/unique.
These tests fail due to a Datomic limitation, namely that the marker
flag :db.alter/attribute can only be asserted once for an attribute!
That is, [:db.part/db :db.alter/attribute :attribute] will only be
transacted at most once. Since older versions of Datomic required the
:db.alter/attribute flag, I can only imagine they either never wrote
:db.alter/attribute to the store, or they handled it specially. I'll
need to remove the marker flag system from Mentat in order to address
this fundamental limitation.
* Post: Remove some more single quotes from error output.
* Post: Add assert_transact! macro to unwrap safely.
I was finding it very difficult to track unwrapping errors while
making changes, due to an underlying Mac OS X symbolication issue that
makes running tests with RUST_BACKTRACE=1 so slow that they all time
out.
* Post: Don't expect or recognize :db.{install,alter}/attribute.
I had this all working... except we will never see a repeated
`[:db.part/db :db.alter/attribute :attribute]` assertion in the store!
That means my approach would let you alter an attribute at most one
time. It's not worth hacking around this; it's better to just stop
expecting (and recognizing) the marker flags. (We have all the data
to distinguish the various cases that we need without the marker
flags.)
This brings Mentat in line with the thrust of newer Datomic versions,
but isn't compatible with Datomic, because (if I understand correctly)
Datomic automatically adds :db.{install,alter}/attribute assertions to
transactions.
I haven't purged the corresponding :db/ident and schema fragments just
yet:
- we might want them back
- we might want them in order to upgrade v1 and v2 databases to the
new on-disk layout we're fleshing out (v3?).
* Post: Don't make :db/unique :db.unique/* imply :db/index true.
This patch avoids a potential bug with the "schema" materialized view.
If :db/unique :db.unique/value implies :db/index true, then what
happens when you _retract_ :db.unique/value? I think Datomic defines
this in some way, but I really want the "schema" materialized view to
be a slice of "datoms" and not have these sort of ambiguities and
persistent effects. Therefore, to ensure that we don't retract a
schema characteristic and accidentally change more than we intended
to, this patch stops having any schema characteristic imply any other
schema characteristic(s). To achieve that, I added an
Option<Unique::{Value,Identity}> type to Attribute; this helps with
this patch, and also looks ahead to when we allow to retract
:db/unique attributes.
* Post: Allow to retract :db/ident.
* Post: Include more details about invalid schema changes.
The tests use strings, so they hide the chained errors which do in
fact provide more detail.
* Review comment: Fix outdated comment.
* Review comment: s/_SET/_SQL_LIST/.
* Review comment: Use a sub-select for checking cardinality.
This might be faster in practice.
* Review comment: Put `attribute::Unique` into its own namespace.
2017-03-20 20:18:59 +00:00
& NoHistory | & IsComponent = > {
// There's no on disk change required for either of these.
2019-07-17 15:46:47 +00:00
}
Schema alteration. Fixes #294 and #295. (#370) r=rnewman
* Pre: Don't retract :db/ident in test.
Datomic (and eventually Mentat) don't allow to retract :db/ident in
this way, so this runs afoul of future work to support mutating
metadata.
* Pre: s/VALUETYPE/VALUE_TYPE/.
This is consistent with the capitalization (which is "valueType") and
the other identifier.
* Pre: Remove some single quotes from error output.
* Part 1: Make materialized views be uniform [e a v value_type_tag].
This looks ahead to a time when we could support arbitrary
user-defined materialized views. For now, the "idents" materialized
view is those datoms of the form [e :db/ident :namespaced/keyword] and
the "schema" materialized view is those datoms of the form [e a v]
where a is in a particular set of attributes that will become clear in
the following commits.
This change is not backwards compatible, so I'm removing the open
current (really, v2) test. It'll be re-instated when we get to
https://github.com/mozilla/mentat/issues/194.
* Pre: Map TypedValue::Ref to TypedValue::Keyword in debug output.
* Part 3: Separate `schema_to_mutate` from the `schema` used to interpret.
This is just to keep track of the expected changes during
bootstrapping. I want bootstrap metadata mutations to flow through
the same code path as metadata mutations during regular transactions;
by differentiating the schema used for interpretation from the schema
that will be updated I expect to be able to apply bootstrap metadata
mutations to an empty schema and have things like materialized views
created (using the regular code paths).
This commit has been re-ordered for conceptual clarity, but it won't
compile because it references the metadata module. It's possible to
make it compile -- the functionality is there in the schema module --
but it's not worth the rebasing effort until after review (and
possibly not even then, since we'll squash down to a single commit to
land).
* Part 2: Maintain entids separately from idents.
In order to support historical idents, we need to distinguish the
"current" map from entid -> ident from the "complete historical" map
ident -> entid. This is what Datomic does; in Datomic, an ident is
never retracted (although it can be replaced). This approach is an
important part of allowing multiple consumers to share a schema
fragment as it migrates forward.
This fixes a limitation of the Clojure implementation, which did not
handle historical idents across knowledge base close and re-open.
The "entids" materialized view is naturally a slice of the "datoms"
table. The "idents" materialized view is a slice of the
"transactions" table. I hope that representing in this way, and
casting the problem in this light, might generalize to future
materialized views.
* Pre: Add DiffSet.
* Part 4: Collect mutations to a `Schema`.
I haven't taken your review comment about consuming AttributeBuilder
during each fluent function. If you read my response and still want
this, I'm happy to do it in review.
* Part 5: Handle :db/ident and :db.{install,alter}/attribute.
This "loops" the committed datoms out of the SQL store and back
through the metadata (schema, but in future also partition map)
processor. The metadata processor updates the schema and produces a
report of what changed; that report is then used to update the SQL
store. That update includes:
- the materialized views ("entids", "idents", and "schema");
- if needed, a subset of the datoms themselves (as flags change).
I've left a TODO for handling attribute retraction in the cases that
it makes sense. I expect that to be straight-forward.
* Review comment: Rename DiffSet to AddRetractAlterSet.
Also adds a little more commentary and a simple test.
* Review comment: Use ToIdent trait.
* Review comment: partially revert "Part 2: Maintain entids separately from idents."
This reverts commit 23a91df9c35e14398f2ddbd1ba25315821e67401.
Following our discussion, this removes the "entids" materialized
view. The next commit will remove historical idents from the "idents"
materialized view.
* Post: Use custom Either rather than std::result::Result.
This is not necessary, but it was suggested that we might be paying an
overhead creating Err instances while using error_chain. That seems
not to be the case, but this change shows that we don't actually use
any of the Result helper methods, so there's no reason to overload
Result. This change might avoid some future confusion, so I'm going
to land it anyway.
Signed-off-by: Nick Alexander <nalexander@mozilla.com>
* Review comment: Don't preserve historical idents.
* Review comment: More prepared statements when updating materialized views.
* Post: Test altering :db/cardinality and :db/unique.
These tests fail due to a Datomic limitation, namely that the marker
flag :db.alter/attribute can only be asserted once for an attribute!
That is, [:db.part/db :db.alter/attribute :attribute] will only be
transacted at most once. Since older versions of Datomic required the
:db.alter/attribute flag, I can only imagine they either never wrote
:db.alter/attribute to the store, or they handled it specially. I'll
need to remove the marker flag system from Mentat in order to address
this fundamental limitation.
* Post: Remove some more single quotes from error output.
* Post: Add assert_transact! macro to unwrap safely.
I was finding it very difficult to track unwrapping errors while
making changes, due to an underlying Mac OS X symbolication issue that
makes running tests with RUST_BACKTRACE=1 so slow that they all time
out.
* Post: Don't expect or recognize :db.{install,alter}/attribute.
I had this all working... except we will never see a repeated
`[:db.part/db :db.alter/attribute :attribute]` assertion in the store!
That means my approach would let you alter an attribute at most one
time. It's not worth hacking around this; it's better to just stop
expecting (and recognizing) the marker flags. (We have all the data
to distinguish the various cases that we need without the marker
flags.)
This brings Mentat in line with the thrust of newer Datomic versions,
but isn't compatible with Datomic, because (if I understand correctly)
Datomic automatically adds :db.{install,alter}/attribute assertions to
transactions.
I haven't purged the corresponding :db/ident and schema fragments just
yet:
- we might want them back
- we might want them in order to upgrade v1 and v2 databases to the
new on-disk layout we're fleshing out (v3?).
* Post: Don't make :db/unique :db.unique/* imply :db/index true.
This patch avoids a potential bug with the "schema" materialized view.
If :db/unique :db.unique/value implies :db/index true, then what
happens when you _retract_ :db.unique/value? I think Datomic defines
this in some way, but I really want the "schema" materialized view to
be a slice of "datoms" and not have these sort of ambiguities and
persistent effects. Therefore, to ensure that we don't retract a
schema characteristic and accidentally change more than we intended
to, this patch stops having any schema characteristic imply any other
schema characteristic(s). To achieve that, I added an
Option<Unique::{Value,Identity}> type to Attribute; this helps with
this patch, and also looks ahead to when we allow to retract
:db/unique attributes.
* Post: Allow to retract :db/ident.
* Post: Include more details about invalid schema changes.
The tests use strings, so they hide the chained errors which do in
fact provide more detail.
* Review comment: Fix outdated comment.
* Review comment: s/_SET/_SQL_LIST/.
* Review comment: Use a sub-select for checking cardinality.
This might be faster in practice.
* Review comment: Put `attribute::Unique` into its own namespace.
2017-03-20 20:18:59 +00:00
}
}
}
Ok ( ( ) )
}
2018-07-12 23:50:08 +00:00
impl PartitionMap {
2017-02-15 00:50:40 +00:00
/// Allocate a single fresh entid in the given `partition`.
2018-07-12 23:50:08 +00:00
pub ( crate ) fn allocate_entid ( & mut self , partition : & str ) -> i64 {
2017-02-15 00:50:40 +00:00
self . allocate_entids ( partition , 1 ) . start
}
2017-02-08 22:04:32 +00:00
2017-02-15 00:50:40 +00:00
/// Allocate `n` fresh entids in the given `partition`.
2018-07-12 23:50:08 +00:00
pub ( crate ) fn allocate_entids ( & mut self , partition : & str , n : usize ) -> Range < i64 > {
Extract partial storage abstraction; use error-chain throughout. Fixes #328. r=rnewman (#341)
* Pre: Drop unneeded tx0 from search results.
* Pre: Don't require a schema in some of the DB code.
The idea is to separate the transaction applying code, which is
schema-aware, from the concrete storage code, which is just concerned
with getting bits onto disk.
* Pre: Only reference Schema, not DB, in debug module.
This is part of a larger separation of the volatile PartitionMap,
which is modified every transaction, from the stable Schema, which is
infrequently modified.
* Pre: Fix indentation.
* Extract part of DB to new SchemaTypeChecking trait.
* Extract part of DB to new PartitionMapping trait.
* Pre: Don't expect :db.part/tx partition to advance when tx fails.
This fails right now, because we allocate tx IDs even when we shouldn't.
* Sketch a db interface without DB.
* Add ValueParseError; use error-chain in tx-parser.
This can be simplified when
https://github.com/Marwes/combine/issues/86 makes it to a published
release, but this unblocks us for now. This converts the `combine`
error type `ParseError<&'a [edn::Value]>` to a type with owned
`Vec<edn::Value>` collections, re-using `edn::Value::Vector` for
making them `Display`.
* Pre: Accept Borrow<Schema> instead of just &Schema in debug module.
This makes it easy to use Rc<Schema> or Arc<Schema> without inserting
&* sigils throughout the code.
* Use error-chain in query-parser.
There are a few things to point out here:
- the fine grained error types have been flattened into one crate-wide
error type; it's pretty easy to regain the granularity as needed.
- edn::ParseError is automatically lifted to
mentat_query_parser::errors::Error;
- we use mentat_parser_utils::ValueParser to maintain parsing error
information from `combine`.
* Patch up top-level.
* Review comment: Only `borrow()` once.
2017-02-24 23:32:41 +00:00
match self . get_mut ( partition ) {
2018-07-17 01:33:40 +00:00
Some ( partition ) = > partition . allocate_entids ( n ) ,
2019-07-17 15:46:47 +00:00
None = > panic! (
" Cannot allocate entid from unknown partition: {} " ,
partition
) ,
2017-02-08 22:04:32 +00:00
}
2017-02-15 00:50:40 +00:00
}
2017-05-08 03:00:04 +00:00
2018-07-12 23:50:08 +00:00
pub ( crate ) fn contains_entid ( & self , entid : Entid ) -> bool {
2019-07-17 15:46:47 +00:00
self . values ( )
. any ( | partition | partition . contains_entid ( entid ) )
2017-05-08 03:00:04 +00:00
}
2017-01-26 00:13:56 +00:00
}
#[ cfg(test) ]
mod tests {
Add type checking and constraint checking to the transactor. (#663, #532, #679)
This should address #663, by re-inserting type checking in the
transactor stack after the entry point used by the term builder.
Before this commit, we were using an SQLite UNIQUE index to assert
that no `[e a]` pair, with `a` a cardinality one attribute, was
asserted more than once. However, that's not in line with Datomic,
which treats transaction inputs as a set and allows a single datom
like `[e a v]` to appear multiple times. It's both awkward and not
particularly efficient to look for _distinct_ repetitions in SQL, so
we accept some runtime cost in order to check for repetitions in the
transactor. This will allow us to address #532, which is really about
whether we treat inputs as sets. A side benefit is that we can
provide more helpful error messages when the transactor does detect
that the input truly violates the cardinality constraints of the
schema.
This commit builds a trie while error checking and collecting final
terms, which should be fairly efficient. It also allows a simpler
expression of input-provided :db/txInstant datoms, which in turn
uncovered a small issue with the transaction watcher, where-by the
watcher would not see non-input-provided :db/txInstant datoms.
This transition to Datomic-like input-as-set semantics allows us to
address #532. Previously, two tempids that upserted to the same entid
would produce duplicate datoms, and that would have been rejected by
the transactor -- correctly, since we did not allow duplicate datoms
under the input-as-list semantics. With input-as-set semantics,
duplicate datoms are allowed; and that means that we must allow
tempids to be equivalent, i.e., to resolve to the same tempid.
To achieve this, we:
- index the set of tempids
- identify tempid indices that share an upsert
- map tempids to a dense set of contiguous integer labels
We use the well-known union-find algorithm, as implemented by
petgraph, to efficiently manage the set of equivalent tempids.
Along the way, I've fixed and added tests for two small errors in the
transactor. First, don't drop datoms resolved by upsert (#679).
Second, ensure that complex upserts are allocated.
I don't know quite what happened here. The Clojure implementation
correctly kept complex upserts that hadn't resolved as complex
upserts (see
https://github.com/mozilla/mentat/blob/9a9dfb502acf5e4cdb1059d4aac831d7603063c8/src/common/datomish/transact.cljc#L436)
and then allocated complex upserts if they didn't resolve (see
https://github.com/mozilla/mentat/blob/9a9dfb502acf5e4cdb1059d4aac831d7603063c8/src/common/datomish/transact.cljc#L509).
Based on the code comments, I think the Rust implementation must have
incorrectly tried to optimize by handling all complex upserts in at
most a single generation of evolution, and that's just not correct.
We're effectively implementing a topological sort, using very specific
domain knowledge, and its not true that a node in a topological sort
can be considered only once!
2018-04-30 22:16:05 +00:00
extern crate env_logger ;
2019-07-17 15:46:47 +00:00
use std ::borrow ::Borrow ;
2018-07-12 23:50:08 +00:00
2017-01-26 00:13:56 +00:00
use super ::* ;
2020-08-06 03:03:58 +00:00
use crate ::debug ::{ tempids , TestConn } ;
use crate ::internal_types ::Term ;
2019-07-17 15:46:47 +00:00
use core_traits ::{ attribute , KnownEntid } ;
use db_traits ::errors ;
use edn ::entities ::OpType ;
use edn ::{ self , InternSet } ;
Add type checking and constraint checking to the transactor. (#663, #532, #679)
This should address #663, by re-inserting type checking in the
transactor stack after the entry point used by the term builder.
Before this commit, we were using an SQLite UNIQUE index to assert
that no `[e a]` pair, with `a` a cardinality one attribute, was
asserted more than once. However, that's not in line with Datomic,
which treats transaction inputs as a set and allows a single datom
like `[e a v]` to appear multiple times. It's both awkward and not
particularly efficient to look for _distinct_ repetitions in SQL, so
we accept some runtime cost in order to check for repetitions in the
transactor. This will allow us to address #532, which is really about
whether we treat inputs as sets. A side benefit is that we can
provide more helpful error messages when the transactor does detect
that the input truly violates the cardinality constraints of the
schema.
This commit builds a trie while error checking and collecting final
terms, which should be fairly efficient. It also allows a simpler
expression of input-provided :db/txInstant datoms, which in turn
uncovered a small issue with the transaction watcher, where-by the
watcher would not see non-input-provided :db/txInstant datoms.
This transition to Datomic-like input-as-set semantics allows us to
address #532. Previously, two tempids that upserted to the same entid
would produce duplicate datoms, and that would have been rejected by
the transactor -- correctly, since we did not allow duplicate datoms
under the input-as-list semantics. With input-as-set semantics,
duplicate datoms are allowed; and that means that we must allow
tempids to be equivalent, i.e., to resolve to the same tempid.
To achieve this, we:
- index the set of tempids
- identify tempid indices that share an upsert
- map tempids to a dense set of contiguous integer labels
We use the well-known union-find algorithm, as implemented by
petgraph, to efficiently manage the set of equivalent tempids.
Along the way, I've fixed and added tests for two small errors in the
transactor. First, don't drop datoms resolved by upsert (#679).
Second, ensure that complex upserts are allocated.
I don't know quite what happened here. The Clojure implementation
correctly kept complex upserts that hadn't resolved as complex
upserts (see
https://github.com/mozilla/mentat/blob/9a9dfb502acf5e4cdb1059d4aac831d7603063c8/src/common/datomish/transact.cljc#L436)
and then allocated complex upserts if they didn't resolve (see
https://github.com/mozilla/mentat/blob/9a9dfb502acf5e4cdb1059d4aac831d7603063c8/src/common/datomish/transact.cljc#L509).
Based on the code comments, I think the Rust implementation must have
incorrectly tried to optimize by handling all complex upserts in at
most a single generation of evolution, and that's just not correct.
We're effectively implementing a topological sort, using very specific
domain knowledge, and its not true that a node in a topological sort
can be considered only once!
2018-04-30 22:16:05 +00:00
use mentat_core ::util ::Either ::* ;
2019-07-17 15:46:47 +00:00
use mentat_core ::{ HasSchema , Keyword } ;
use std ::collections ::BTreeMap ;
2017-03-20 18:29:17 +00:00
2018-06-13 15:49:40 +00:00
fn run_test_add ( mut conn : TestConn ) {
2017-03-20 18:29:17 +00:00
// Test inserting :db.cardinality/one elements.
2019-07-17 15:46:47 +00:00
assert_transact! (
conn ,
" [[:db/add 100 :db.schema/version 1]
[ :db / add 101 :db . schema / version 2 ] ] "
) ;
assert_matches! (
conn . last_transaction ( ) ,
" [[100 :db.schema/version 1 ?tx true]
Schema alteration. Fixes #294 and #295. (#370) r=rnewman
* Pre: Don't retract :db/ident in test.
Datomic (and eventually Mentat) don't allow to retract :db/ident in
this way, so this runs afoul of future work to support mutating
metadata.
* Pre: s/VALUETYPE/VALUE_TYPE/.
This is consistent with the capitalization (which is "valueType") and
the other identifier.
* Pre: Remove some single quotes from error output.
* Part 1: Make materialized views be uniform [e a v value_type_tag].
This looks ahead to a time when we could support arbitrary
user-defined materialized views. For now, the "idents" materialized
view is those datoms of the form [e :db/ident :namespaced/keyword] and
the "schema" materialized view is those datoms of the form [e a v]
where a is in a particular set of attributes that will become clear in
the following commits.
This change is not backwards compatible, so I'm removing the open
current (really, v2) test. It'll be re-instated when we get to
https://github.com/mozilla/mentat/issues/194.
* Pre: Map TypedValue::Ref to TypedValue::Keyword in debug output.
* Part 3: Separate `schema_to_mutate` from the `schema` used to interpret.
This is just to keep track of the expected changes during
bootstrapping. I want bootstrap metadata mutations to flow through
the same code path as metadata mutations during regular transactions;
by differentiating the schema used for interpretation from the schema
that will be updated I expect to be able to apply bootstrap metadata
mutations to an empty schema and have things like materialized views
created (using the regular code paths).
This commit has been re-ordered for conceptual clarity, but it won't
compile because it references the metadata module. It's possible to
make it compile -- the functionality is there in the schema module --
but it's not worth the rebasing effort until after review (and
possibly not even then, since we'll squash down to a single commit to
land).
* Part 2: Maintain entids separately from idents.
In order to support historical idents, we need to distinguish the
"current" map from entid -> ident from the "complete historical" map
ident -> entid. This is what Datomic does; in Datomic, an ident is
never retracted (although it can be replaced). This approach is an
important part of allowing multiple consumers to share a schema
fragment as it migrates forward.
This fixes a limitation of the Clojure implementation, which did not
handle historical idents across knowledge base close and re-open.
The "entids" materialized view is naturally a slice of the "datoms"
table. The "idents" materialized view is a slice of the
"transactions" table. I hope that representing in this way, and
casting the problem in this light, might generalize to future
materialized views.
* Pre: Add DiffSet.
* Part 4: Collect mutations to a `Schema`.
I haven't taken your review comment about consuming AttributeBuilder
during each fluent function. If you read my response and still want
this, I'm happy to do it in review.
* Part 5: Handle :db/ident and :db.{install,alter}/attribute.
This "loops" the committed datoms out of the SQL store and back
through the metadata (schema, but in future also partition map)
processor. The metadata processor updates the schema and produces a
report of what changed; that report is then used to update the SQL
store. That update includes:
- the materialized views ("entids", "idents", and "schema");
- if needed, a subset of the datoms themselves (as flags change).
I've left a TODO for handling attribute retraction in the cases that
it makes sense. I expect that to be straight-forward.
* Review comment: Rename DiffSet to AddRetractAlterSet.
Also adds a little more commentary and a simple test.
* Review comment: Use ToIdent trait.
* Review comment: partially revert "Part 2: Maintain entids separately from idents."
This reverts commit 23a91df9c35e14398f2ddbd1ba25315821e67401.
Following our discussion, this removes the "entids" materialized
view. The next commit will remove historical idents from the "idents"
materialized view.
* Post: Use custom Either rather than std::result::Result.
This is not necessary, but it was suggested that we might be paying an
overhead creating Err instances while using error_chain. That seems
not to be the case, but this change shows that we don't actually use
any of the Result helper methods, so there's no reason to overload
Result. This change might avoid some future confusion, so I'm going
to land it anyway.
Signed-off-by: Nick Alexander <nalexander@mozilla.com>
* Review comment: Don't preserve historical idents.
* Review comment: More prepared statements when updating materialized views.
* Post: Test altering :db/cardinality and :db/unique.
These tests fail due to a Datomic limitation, namely that the marker
flag :db.alter/attribute can only be asserted once for an attribute!
That is, [:db.part/db :db.alter/attribute :attribute] will only be
transacted at most once. Since older versions of Datomic required the
:db.alter/attribute flag, I can only imagine they either never wrote
:db.alter/attribute to the store, or they handled it specially. I'll
need to remove the marker flag system from Mentat in order to address
this fundamental limitation.
* Post: Remove some more single quotes from error output.
* Post: Add assert_transact! macro to unwrap safely.
I was finding it very difficult to track unwrapping errors while
making changes, due to an underlying Mac OS X symbolication issue that
makes running tests with RUST_BACKTRACE=1 so slow that they all time
out.
* Post: Don't expect or recognize :db.{install,alter}/attribute.
I had this all working... except we will never see a repeated
`[:db.part/db :db.alter/attribute :attribute]` assertion in the store!
That means my approach would let you alter an attribute at most one
time. It's not worth hacking around this; it's better to just stop
expecting (and recognizing) the marker flags. (We have all the data
to distinguish the various cases that we need without the marker
flags.)
This brings Mentat in line with the thrust of newer Datomic versions,
but isn't compatible with Datomic, because (if I understand correctly)
Datomic automatically adds :db.{install,alter}/attribute assertions to
transactions.
I haven't purged the corresponding :db/ident and schema fragments just
yet:
- we might want them back
- we might want them in order to upgrade v1 and v2 databases to the
new on-disk layout we're fleshing out (v3?).
* Post: Don't make :db/unique :db.unique/* imply :db/index true.
This patch avoids a potential bug with the "schema" materialized view.
If :db/unique :db.unique/value implies :db/index true, then what
happens when you _retract_ :db.unique/value? I think Datomic defines
this in some way, but I really want the "schema" materialized view to
be a slice of "datoms" and not have these sort of ambiguities and
persistent effects. Therefore, to ensure that we don't retract a
schema characteristic and accidentally change more than we intended
to, this patch stops having any schema characteristic imply any other
schema characteristic(s). To achieve that, I added an
Option<Unique::{Value,Identity}> type to Attribute; this helps with
this patch, and also looks ahead to when we allow to retract
:db/unique attributes.
* Post: Allow to retract :db/ident.
* Post: Include more details about invalid schema changes.
The tests use strings, so they hide the chained errors which do in
fact provide more detail.
* Review comment: Fix outdated comment.
* Review comment: s/_SET/_SQL_LIST/.
* Review comment: Use a sub-select for checking cardinality.
This might be faster in practice.
* Review comment: Put `attribute::Unique` into its own namespace.
2017-03-20 20:18:59 +00:00
[ 101 :db . schema / version 2 ? tx true ]
2019-07-17 15:46:47 +00:00
[ ? tx :db / txInstant ? ms ? tx true ] ] "
) ;
assert_matches! (
conn . datoms ( ) ,
" [[100 :db.schema/version 1]
[ 101 :db . schema / version 2 ] ] "
) ;
2017-03-20 18:29:17 +00:00
// Test inserting :db.cardinality/many elements.
2019-07-17 15:46:47 +00:00
assert_transact! (
conn ,
" [[:db/add 200 :db.schema/attribute 100]
[ :db / add 200 :db . schema / attribute 101 ] ] "
) ;
assert_matches! (
conn . last_transaction ( ) ,
" [[200 :db.schema/attribute 100 ?tx true]
2017-03-20 18:29:17 +00:00
[ 200 :db . schema / attribute 101 ? tx true ]
2019-07-17 15:46:47 +00:00
[ ? tx :db / txInstant ? ms ? tx true ] ] "
) ;
assert_matches! (
conn . datoms ( ) ,
" [[100 :db.schema/version 1]
Schema alteration. Fixes #294 and #295. (#370) r=rnewman
* Pre: Don't retract :db/ident in test.
Datomic (and eventually Mentat) don't allow to retract :db/ident in
this way, so this runs afoul of future work to support mutating
metadata.
* Pre: s/VALUETYPE/VALUE_TYPE/.
This is consistent with the capitalization (which is "valueType") and
the other identifier.
* Pre: Remove some single quotes from error output.
* Part 1: Make materialized views be uniform [e a v value_type_tag].
This looks ahead to a time when we could support arbitrary
user-defined materialized views. For now, the "idents" materialized
view is those datoms of the form [e :db/ident :namespaced/keyword] and
the "schema" materialized view is those datoms of the form [e a v]
where a is in a particular set of attributes that will become clear in
the following commits.
This change is not backwards compatible, so I'm removing the open
current (really, v2) test. It'll be re-instated when we get to
https://github.com/mozilla/mentat/issues/194.
* Pre: Map TypedValue::Ref to TypedValue::Keyword in debug output.
* Part 3: Separate `schema_to_mutate` from the `schema` used to interpret.
This is just to keep track of the expected changes during
bootstrapping. I want bootstrap metadata mutations to flow through
the same code path as metadata mutations during regular transactions;
by differentiating the schema used for interpretation from the schema
that will be updated I expect to be able to apply bootstrap metadata
mutations to an empty schema and have things like materialized views
created (using the regular code paths).
This commit has been re-ordered for conceptual clarity, but it won't
compile because it references the metadata module. It's possible to
make it compile -- the functionality is there in the schema module --
but it's not worth the rebasing effort until after review (and
possibly not even then, since we'll squash down to a single commit to
land).
* Part 2: Maintain entids separately from idents.
In order to support historical idents, we need to distinguish the
"current" map from entid -> ident from the "complete historical" map
ident -> entid. This is what Datomic does; in Datomic, an ident is
never retracted (although it can be replaced). This approach is an
important part of allowing multiple consumers to share a schema
fragment as it migrates forward.
This fixes a limitation of the Clojure implementation, which did not
handle historical idents across knowledge base close and re-open.
The "entids" materialized view is naturally a slice of the "datoms"
table. The "idents" materialized view is a slice of the
"transactions" table. I hope that representing in this way, and
casting the problem in this light, might generalize to future
materialized views.
* Pre: Add DiffSet.
* Part 4: Collect mutations to a `Schema`.
I haven't taken your review comment about consuming AttributeBuilder
during each fluent function. If you read my response and still want
this, I'm happy to do it in review.
* Part 5: Handle :db/ident and :db.{install,alter}/attribute.
This "loops" the committed datoms out of the SQL store and back
through the metadata (schema, but in future also partition map)
processor. The metadata processor updates the schema and produces a
report of what changed; that report is then used to update the SQL
store. That update includes:
- the materialized views ("entids", "idents", and "schema");
- if needed, a subset of the datoms themselves (as flags change).
I've left a TODO for handling attribute retraction in the cases that
it makes sense. I expect that to be straight-forward.
* Review comment: Rename DiffSet to AddRetractAlterSet.
Also adds a little more commentary and a simple test.
* Review comment: Use ToIdent trait.
* Review comment: partially revert "Part 2: Maintain entids separately from idents."
This reverts commit 23a91df9c35e14398f2ddbd1ba25315821e67401.
Following our discussion, this removes the "entids" materialized
view. The next commit will remove historical idents from the "idents"
materialized view.
* Post: Use custom Either rather than std::result::Result.
This is not necessary, but it was suggested that we might be paying an
overhead creating Err instances while using error_chain. That seems
not to be the case, but this change shows that we don't actually use
any of the Result helper methods, so there's no reason to overload
Result. This change might avoid some future confusion, so I'm going
to land it anyway.
Signed-off-by: Nick Alexander <nalexander@mozilla.com>
* Review comment: Don't preserve historical idents.
* Review comment: More prepared statements when updating materialized views.
* Post: Test altering :db/cardinality and :db/unique.
These tests fail due to a Datomic limitation, namely that the marker
flag :db.alter/attribute can only be asserted once for an attribute!
That is, [:db.part/db :db.alter/attribute :attribute] will only be
transacted at most once. Since older versions of Datomic required the
:db.alter/attribute flag, I can only imagine they either never wrote
:db.alter/attribute to the store, or they handled it specially. I'll
need to remove the marker flag system from Mentat in order to address
this fundamental limitation.
* Post: Remove some more single quotes from error output.
* Post: Add assert_transact! macro to unwrap safely.
I was finding it very difficult to track unwrapping errors while
making changes, due to an underlying Mac OS X symbolication issue that
makes running tests with RUST_BACKTRACE=1 so slow that they all time
out.
* Post: Don't expect or recognize :db.{install,alter}/attribute.
I had this all working... except we will never see a repeated
`[:db.part/db :db.alter/attribute :attribute]` assertion in the store!
That means my approach would let you alter an attribute at most one
time. It's not worth hacking around this; it's better to just stop
expecting (and recognizing) the marker flags. (We have all the data
to distinguish the various cases that we need without the marker
flags.)
This brings Mentat in line with the thrust of newer Datomic versions,
but isn't compatible with Datomic, because (if I understand correctly)
Datomic automatically adds :db.{install,alter}/attribute assertions to
transactions.
I haven't purged the corresponding :db/ident and schema fragments just
yet:
- we might want them back
- we might want them in order to upgrade v1 and v2 databases to the
new on-disk layout we're fleshing out (v3?).
* Post: Don't make :db/unique :db.unique/* imply :db/index true.
This patch avoids a potential bug with the "schema" materialized view.
If :db/unique :db.unique/value implies :db/index true, then what
happens when you _retract_ :db.unique/value? I think Datomic defines
this in some way, but I really want the "schema" materialized view to
be a slice of "datoms" and not have these sort of ambiguities and
persistent effects. Therefore, to ensure that we don't retract a
schema characteristic and accidentally change more than we intended
to, this patch stops having any schema characteristic imply any other
schema characteristic(s). To achieve that, I added an
Option<Unique::{Value,Identity}> type to Attribute; this helps with
this patch, and also looks ahead to when we allow to retract
:db/unique attributes.
* Post: Allow to retract :db/ident.
* Post: Include more details about invalid schema changes.
The tests use strings, so they hide the chained errors which do in
fact provide more detail.
* Review comment: Fix outdated comment.
* Review comment: s/_SET/_SQL_LIST/.
* Review comment: Use a sub-select for checking cardinality.
This might be faster in practice.
* Review comment: Put `attribute::Unique` into its own namespace.
2017-03-20 20:18:59 +00:00
[ 101 :db . schema / version 2 ]
2017-03-20 18:29:17 +00:00
[ 200 :db . schema / attribute 100 ]
2019-07-17 15:46:47 +00:00
[ 200 :db . schema / attribute 101 ] ] "
) ;
2017-03-20 18:29:17 +00:00
// Test replacing existing :db.cardinality/one elements.
2019-07-17 15:46:47 +00:00
assert_transact! (
conn ,
" [[:db/add 100 :db.schema/version 11]
[ :db / add 101 :db . schema / version 22 ] ] "
) ;
assert_matches! (
conn . last_transaction ( ) ,
" [[100 :db.schema/version 1 ?tx false]
Schema alteration. Fixes #294 and #295. (#370) r=rnewman
* Pre: Don't retract :db/ident in test.
Datomic (and eventually Mentat) don't allow to retract :db/ident in
this way, so this runs afoul of future work to support mutating
metadata.
* Pre: s/VALUETYPE/VALUE_TYPE/.
This is consistent with the capitalization (which is "valueType") and
the other identifier.
* Pre: Remove some single quotes from error output.
* Part 1: Make materialized views be uniform [e a v value_type_tag].
This looks ahead to a time when we could support arbitrary
user-defined materialized views. For now, the "idents" materialized
view is those datoms of the form [e :db/ident :namespaced/keyword] and
the "schema" materialized view is those datoms of the form [e a v]
where a is in a particular set of attributes that will become clear in
the following commits.
This change is not backwards compatible, so I'm removing the open
current (really, v2) test. It'll be re-instated when we get to
https://github.com/mozilla/mentat/issues/194.
* Pre: Map TypedValue::Ref to TypedValue::Keyword in debug output.
* Part 3: Separate `schema_to_mutate` from the `schema` used to interpret.
This is just to keep track of the expected changes during
bootstrapping. I want bootstrap metadata mutations to flow through
the same code path as metadata mutations during regular transactions;
by differentiating the schema used for interpretation from the schema
that will be updated I expect to be able to apply bootstrap metadata
mutations to an empty schema and have things like materialized views
created (using the regular code paths).
This commit has been re-ordered for conceptual clarity, but it won't
compile because it references the metadata module. It's possible to
make it compile -- the functionality is there in the schema module --
but it's not worth the rebasing effort until after review (and
possibly not even then, since we'll squash down to a single commit to
land).
* Part 2: Maintain entids separately from idents.
In order to support historical idents, we need to distinguish the
"current" map from entid -> ident from the "complete historical" map
ident -> entid. This is what Datomic does; in Datomic, an ident is
never retracted (although it can be replaced). This approach is an
important part of allowing multiple consumers to share a schema
fragment as it migrates forward.
This fixes a limitation of the Clojure implementation, which did not
handle historical idents across knowledge base close and re-open.
The "entids" materialized view is naturally a slice of the "datoms"
table. The "idents" materialized view is a slice of the
"transactions" table. I hope that representing in this way, and
casting the problem in this light, might generalize to future
materialized views.
* Pre: Add DiffSet.
* Part 4: Collect mutations to a `Schema`.
I haven't taken your review comment about consuming AttributeBuilder
during each fluent function. If you read my response and still want
this, I'm happy to do it in review.
* Part 5: Handle :db/ident and :db.{install,alter}/attribute.
This "loops" the committed datoms out of the SQL store and back
through the metadata (schema, but in future also partition map)
processor. The metadata processor updates the schema and produces a
report of what changed; that report is then used to update the SQL
store. That update includes:
- the materialized views ("entids", "idents", and "schema");
- if needed, a subset of the datoms themselves (as flags change).
I've left a TODO for handling attribute retraction in the cases that
it makes sense. I expect that to be straight-forward.
* Review comment: Rename DiffSet to AddRetractAlterSet.
Also adds a little more commentary and a simple test.
* Review comment: Use ToIdent trait.
* Review comment: partially revert "Part 2: Maintain entids separately from idents."
This reverts commit 23a91df9c35e14398f2ddbd1ba25315821e67401.
Following our discussion, this removes the "entids" materialized
view. The next commit will remove historical idents from the "idents"
materialized view.
* Post: Use custom Either rather than std::result::Result.
This is not necessary, but it was suggested that we might be paying an
overhead creating Err instances while using error_chain. That seems
not to be the case, but this change shows that we don't actually use
any of the Result helper methods, so there's no reason to overload
Result. This change might avoid some future confusion, so I'm going
to land it anyway.
Signed-off-by: Nick Alexander <nalexander@mozilla.com>
* Review comment: Don't preserve historical idents.
* Review comment: More prepared statements when updating materialized views.
* Post: Test altering :db/cardinality and :db/unique.
These tests fail due to a Datomic limitation, namely that the marker
flag :db.alter/attribute can only be asserted once for an attribute!
That is, [:db.part/db :db.alter/attribute :attribute] will only be
transacted at most once. Since older versions of Datomic required the
:db.alter/attribute flag, I can only imagine they either never wrote
:db.alter/attribute to the store, or they handled it specially. I'll
need to remove the marker flag system from Mentat in order to address
this fundamental limitation.
* Post: Remove some more single quotes from error output.
* Post: Add assert_transact! macro to unwrap safely.
I was finding it very difficult to track unwrapping errors while
making changes, due to an underlying Mac OS X symbolication issue that
makes running tests with RUST_BACKTRACE=1 so slow that they all time
out.
* Post: Don't expect or recognize :db.{install,alter}/attribute.
I had this all working... except we will never see a repeated
`[:db.part/db :db.alter/attribute :attribute]` assertion in the store!
That means my approach would let you alter an attribute at most one
time. It's not worth hacking around this; it's better to just stop
expecting (and recognizing) the marker flags. (We have all the data
to distinguish the various cases that we need without the marker
flags.)
This brings Mentat in line with the thrust of newer Datomic versions,
but isn't compatible with Datomic, because (if I understand correctly)
Datomic automatically adds :db.{install,alter}/attribute assertions to
transactions.
I haven't purged the corresponding :db/ident and schema fragments just
yet:
- we might want them back
- we might want them in order to upgrade v1 and v2 databases to the
new on-disk layout we're fleshing out (v3?).
* Post: Don't make :db/unique :db.unique/* imply :db/index true.
This patch avoids a potential bug with the "schema" materialized view.
If :db/unique :db.unique/value implies :db/index true, then what
happens when you _retract_ :db.unique/value? I think Datomic defines
this in some way, but I really want the "schema" materialized view to
be a slice of "datoms" and not have these sort of ambiguities and
persistent effects. Therefore, to ensure that we don't retract a
schema characteristic and accidentally change more than we intended
to, this patch stops having any schema characteristic imply any other
schema characteristic(s). To achieve that, I added an
Option<Unique::{Value,Identity}> type to Attribute; this helps with
this patch, and also looks ahead to when we allow to retract
:db/unique attributes.
* Post: Allow to retract :db/ident.
* Post: Include more details about invalid schema changes.
The tests use strings, so they hide the chained errors which do in
fact provide more detail.
* Review comment: Fix outdated comment.
* Review comment: s/_SET/_SQL_LIST/.
* Review comment: Use a sub-select for checking cardinality.
This might be faster in practice.
* Review comment: Put `attribute::Unique` into its own namespace.
2017-03-20 20:18:59 +00:00
[ 100 :db . schema / version 11 ? tx true ]
[ 101 :db . schema / version 2 ? tx false ]
[ 101 :db . schema / version 22 ? tx true ]
2019-07-17 15:46:47 +00:00
[ ? tx :db / txInstant ? ms ? tx true ] ] "
) ;
assert_matches! (
conn . datoms ( ) ,
" [[100 :db.schema/version 11]
Schema alteration. Fixes #294 and #295. (#370) r=rnewman
* Pre: Don't retract :db/ident in test.
Datomic (and eventually Mentat) don't allow to retract :db/ident in
this way, so this runs afoul of future work to support mutating
metadata.
* Pre: s/VALUETYPE/VALUE_TYPE/.
This is consistent with the capitalization (which is "valueType") and
the other identifier.
* Pre: Remove some single quotes from error output.
* Part 1: Make materialized views be uniform [e a v value_type_tag].
This looks ahead to a time when we could support arbitrary
user-defined materialized views. For now, the "idents" materialized
view is those datoms of the form [e :db/ident :namespaced/keyword] and
the "schema" materialized view is those datoms of the form [e a v]
where a is in a particular set of attributes that will become clear in
the following commits.
This change is not backwards compatible, so I'm removing the open
current (really, v2) test. It'll be re-instated when we get to
https://github.com/mozilla/mentat/issues/194.
* Pre: Map TypedValue::Ref to TypedValue::Keyword in debug output.
* Part 3: Separate `schema_to_mutate` from the `schema` used to interpret.
This is just to keep track of the expected changes during
bootstrapping. I want bootstrap metadata mutations to flow through
the same code path as metadata mutations during regular transactions;
by differentiating the schema used for interpretation from the schema
that will be updated I expect to be able to apply bootstrap metadata
mutations to an empty schema and have things like materialized views
created (using the regular code paths).
This commit has been re-ordered for conceptual clarity, but it won't
compile because it references the metadata module. It's possible to
make it compile -- the functionality is there in the schema module --
but it's not worth the rebasing effort until after review (and
possibly not even then, since we'll squash down to a single commit to
land).
* Part 2: Maintain entids separately from idents.
In order to support historical idents, we need to distinguish the
"current" map from entid -> ident from the "complete historical" map
ident -> entid. This is what Datomic does; in Datomic, an ident is
never retracted (although it can be replaced). This approach is an
important part of allowing multiple consumers to share a schema
fragment as it migrates forward.
This fixes a limitation of the Clojure implementation, which did not
handle historical idents across knowledge base close and re-open.
The "entids" materialized view is naturally a slice of the "datoms"
table. The "idents" materialized view is a slice of the
"transactions" table. I hope that representing in this way, and
casting the problem in this light, might generalize to future
materialized views.
* Pre: Add DiffSet.
* Part 4: Collect mutations to a `Schema`.
I haven't taken your review comment about consuming AttributeBuilder
during each fluent function. If you read my response and still want
this, I'm happy to do it in review.
* Part 5: Handle :db/ident and :db.{install,alter}/attribute.
This "loops" the committed datoms out of the SQL store and back
through the metadata (schema, but in future also partition map)
processor. The metadata processor updates the schema and produces a
report of what changed; that report is then used to update the SQL
store. That update includes:
- the materialized views ("entids", "idents", and "schema");
- if needed, a subset of the datoms themselves (as flags change).
I've left a TODO for handling attribute retraction in the cases that
it makes sense. I expect that to be straight-forward.
* Review comment: Rename DiffSet to AddRetractAlterSet.
Also adds a little more commentary and a simple test.
* Review comment: Use ToIdent trait.
* Review comment: partially revert "Part 2: Maintain entids separately from idents."
This reverts commit 23a91df9c35e14398f2ddbd1ba25315821e67401.
Following our discussion, this removes the "entids" materialized
view. The next commit will remove historical idents from the "idents"
materialized view.
* Post: Use custom Either rather than std::result::Result.
This is not necessary, but it was suggested that we might be paying an
overhead creating Err instances while using error_chain. That seems
not to be the case, but this change shows that we don't actually use
any of the Result helper methods, so there's no reason to overload
Result. This change might avoid some future confusion, so I'm going
to land it anyway.
Signed-off-by: Nick Alexander <nalexander@mozilla.com>
* Review comment: Don't preserve historical idents.
* Review comment: More prepared statements when updating materialized views.
* Post: Test altering :db/cardinality and :db/unique.
These tests fail due to a Datomic limitation, namely that the marker
flag :db.alter/attribute can only be asserted once for an attribute!
That is, [:db.part/db :db.alter/attribute :attribute] will only be
transacted at most once. Since older versions of Datomic required the
:db.alter/attribute flag, I can only imagine they either never wrote
:db.alter/attribute to the store, or they handled it specially. I'll
need to remove the marker flag system from Mentat in order to address
this fundamental limitation.
* Post: Remove some more single quotes from error output.
* Post: Add assert_transact! macro to unwrap safely.
I was finding it very difficult to track unwrapping errors while
making changes, due to an underlying Mac OS X symbolication issue that
makes running tests with RUST_BACKTRACE=1 so slow that they all time
out.
* Post: Don't expect or recognize :db.{install,alter}/attribute.
I had this all working... except we will never see a repeated
`[:db.part/db :db.alter/attribute :attribute]` assertion in the store!
That means my approach would let you alter an attribute at most one
time. It's not worth hacking around this; it's better to just stop
expecting (and recognizing) the marker flags. (We have all the data
to distinguish the various cases that we need without the marker
flags.)
This brings Mentat in line with the thrust of newer Datomic versions,
but isn't compatible with Datomic, because (if I understand correctly)
Datomic automatically adds :db.{install,alter}/attribute assertions to
transactions.
I haven't purged the corresponding :db/ident and schema fragments just
yet:
- we might want them back
- we might want them in order to upgrade v1 and v2 databases to the
new on-disk layout we're fleshing out (v3?).
* Post: Don't make :db/unique :db.unique/* imply :db/index true.
This patch avoids a potential bug with the "schema" materialized view.
If :db/unique :db.unique/value implies :db/index true, then what
happens when you _retract_ :db.unique/value? I think Datomic defines
this in some way, but I really want the "schema" materialized view to
be a slice of "datoms" and not have these sort of ambiguities and
persistent effects. Therefore, to ensure that we don't retract a
schema characteristic and accidentally change more than we intended
to, this patch stops having any schema characteristic imply any other
schema characteristic(s). To achieve that, I added an
Option<Unique::{Value,Identity}> type to Attribute; this helps with
this patch, and also looks ahead to when we allow to retract
:db/unique attributes.
* Post: Allow to retract :db/ident.
* Post: Include more details about invalid schema changes.
The tests use strings, so they hide the chained errors which do in
fact provide more detail.
* Review comment: Fix outdated comment.
* Review comment: s/_SET/_SQL_LIST/.
* Review comment: Use a sub-select for checking cardinality.
This might be faster in practice.
* Review comment: Put `attribute::Unique` into its own namespace.
2017-03-20 20:18:59 +00:00
[ 101 :db . schema / version 22 ]
2017-03-20 18:29:17 +00:00
[ 200 :db . schema / attribute 100 ]
2019-07-17 15:46:47 +00:00
[ 200 :db . schema / attribute 101 ] ] "
) ;
2017-03-20 18:29:17 +00:00
// Test that asserting existing :db.cardinality/one elements doesn't change the store.
2019-07-17 15:46:47 +00:00
assert_transact! (
conn ,
" [[:db/add 100 :db.schema/version 11]
[ :db / add 101 :db . schema / version 22 ] ] "
) ;
assert_matches! (
conn . last_transaction ( ) ,
" [[?tx :db/txInstant ?ms ?tx true]] "
) ;
assert_matches! (
conn . datoms ( ) ,
" [[100 :db.schema/version 11]
Schema alteration. Fixes #294 and #295. (#370) r=rnewman
* Pre: Don't retract :db/ident in test.
Datomic (and eventually Mentat) don't allow to retract :db/ident in
this way, so this runs afoul of future work to support mutating
metadata.
* Pre: s/VALUETYPE/VALUE_TYPE/.
This is consistent with the capitalization (which is "valueType") and
the other identifier.
* Pre: Remove some single quotes from error output.
* Part 1: Make materialized views be uniform [e a v value_type_tag].
This looks ahead to a time when we could support arbitrary
user-defined materialized views. For now, the "idents" materialized
view is those datoms of the form [e :db/ident :namespaced/keyword] and
the "schema" materialized view is those datoms of the form [e a v]
where a is in a particular set of attributes that will become clear in
the following commits.
This change is not backwards compatible, so I'm removing the open
current (really, v2) test. It'll be re-instated when we get to
https://github.com/mozilla/mentat/issues/194.
* Pre: Map TypedValue::Ref to TypedValue::Keyword in debug output.
* Part 3: Separate `schema_to_mutate` from the `schema` used to interpret.
This is just to keep track of the expected changes during
bootstrapping. I want bootstrap metadata mutations to flow through
the same code path as metadata mutations during regular transactions;
by differentiating the schema used for interpretation from the schema
that will be updated I expect to be able to apply bootstrap metadata
mutations to an empty schema and have things like materialized views
created (using the regular code paths).
This commit has been re-ordered for conceptual clarity, but it won't
compile because it references the metadata module. It's possible to
make it compile -- the functionality is there in the schema module --
but it's not worth the rebasing effort until after review (and
possibly not even then, since we'll squash down to a single commit to
land).
* Part 2: Maintain entids separately from idents.
In order to support historical idents, we need to distinguish the
"current" map from entid -> ident from the "complete historical" map
ident -> entid. This is what Datomic does; in Datomic, an ident is
never retracted (although it can be replaced). This approach is an
important part of allowing multiple consumers to share a schema
fragment as it migrates forward.
This fixes a limitation of the Clojure implementation, which did not
handle historical idents across knowledge base close and re-open.
The "entids" materialized view is naturally a slice of the "datoms"
table. The "idents" materialized view is a slice of the
"transactions" table. I hope that representing in this way, and
casting the problem in this light, might generalize to future
materialized views.
* Pre: Add DiffSet.
* Part 4: Collect mutations to a `Schema`.
I haven't taken your review comment about consuming AttributeBuilder
during each fluent function. If you read my response and still want
this, I'm happy to do it in review.
* Part 5: Handle :db/ident and :db.{install,alter}/attribute.
This "loops" the committed datoms out of the SQL store and back
through the metadata (schema, but in future also partition map)
processor. The metadata processor updates the schema and produces a
report of what changed; that report is then used to update the SQL
store. That update includes:
- the materialized views ("entids", "idents", and "schema");
- if needed, a subset of the datoms themselves (as flags change).
I've left a TODO for handling attribute retraction in the cases that
it makes sense. I expect that to be straight-forward.
* Review comment: Rename DiffSet to AddRetractAlterSet.
Also adds a little more commentary and a simple test.
* Review comment: Use ToIdent trait.
* Review comment: partially revert "Part 2: Maintain entids separately from idents."
This reverts commit 23a91df9c35e14398f2ddbd1ba25315821e67401.
Following our discussion, this removes the "entids" materialized
view. The next commit will remove historical idents from the "idents"
materialized view.
* Post: Use custom Either rather than std::result::Result.
This is not necessary, but it was suggested that we might be paying an
overhead creating Err instances while using error_chain. That seems
not to be the case, but this change shows that we don't actually use
any of the Result helper methods, so there's no reason to overload
Result. This change might avoid some future confusion, so I'm going
to land it anyway.
Signed-off-by: Nick Alexander <nalexander@mozilla.com>
* Review comment: Don't preserve historical idents.
* Review comment: More prepared statements when updating materialized views.
* Post: Test altering :db/cardinality and :db/unique.
These tests fail due to a Datomic limitation, namely that the marker
flag :db.alter/attribute can only be asserted once for an attribute!
That is, [:db.part/db :db.alter/attribute :attribute] will only be
transacted at most once. Since older versions of Datomic required the
:db.alter/attribute flag, I can only imagine they either never wrote
:db.alter/attribute to the store, or they handled it specially. I'll
need to remove the marker flag system from Mentat in order to address
this fundamental limitation.
* Post: Remove some more single quotes from error output.
* Post: Add assert_transact! macro to unwrap safely.
I was finding it very difficult to track unwrapping errors while
making changes, due to an underlying Mac OS X symbolication issue that
makes running tests with RUST_BACKTRACE=1 so slow that they all time
out.
* Post: Don't expect or recognize :db.{install,alter}/attribute.
I had this all working... except we will never see a repeated
`[:db.part/db :db.alter/attribute :attribute]` assertion in the store!
That means my approach would let you alter an attribute at most one
time. It's not worth hacking around this; it's better to just stop
expecting (and recognizing) the marker flags. (We have all the data
to distinguish the various cases that we need without the marker
flags.)
This brings Mentat in line with the thrust of newer Datomic versions,
but isn't compatible with Datomic, because (if I understand correctly)
Datomic automatically adds :db.{install,alter}/attribute assertions to
transactions.
I haven't purged the corresponding :db/ident and schema fragments just
yet:
- we might want them back
- we might want them in order to upgrade v1 and v2 databases to the
new on-disk layout we're fleshing out (v3?).
* Post: Don't make :db/unique :db.unique/* imply :db/index true.
This patch avoids a potential bug with the "schema" materialized view.
If :db/unique :db.unique/value implies :db/index true, then what
happens when you _retract_ :db.unique/value? I think Datomic defines
this in some way, but I really want the "schema" materialized view to
be a slice of "datoms" and not have these sort of ambiguities and
persistent effects. Therefore, to ensure that we don't retract a
schema characteristic and accidentally change more than we intended
to, this patch stops having any schema characteristic imply any other
schema characteristic(s). To achieve that, I added an
Option<Unique::{Value,Identity}> type to Attribute; this helps with
this patch, and also looks ahead to when we allow to retract
:db/unique attributes.
* Post: Allow to retract :db/ident.
* Post: Include more details about invalid schema changes.
The tests use strings, so they hide the chained errors which do in
fact provide more detail.
* Review comment: Fix outdated comment.
* Review comment: s/_SET/_SQL_LIST/.
* Review comment: Use a sub-select for checking cardinality.
This might be faster in practice.
* Review comment: Put `attribute::Unique` into its own namespace.
2017-03-20 20:18:59 +00:00
[ 101 :db . schema / version 22 ]
2017-03-20 18:29:17 +00:00
[ 200 :db . schema / attribute 100 ]
2019-07-17 15:46:47 +00:00
[ 200 :db . schema / attribute 101 ] ] "
) ;
2017-03-20 18:29:17 +00:00
// Test that asserting existing :db.cardinality/many elements doesn't change the store.
2019-07-17 15:46:47 +00:00
assert_transact! (
conn ,
" [[:db/add 200 :db.schema/attribute 100]
[ :db / add 200 :db . schema / attribute 101 ] ] "
) ;
assert_matches! (
conn . last_transaction ( ) ,
" [[?tx :db/txInstant ?ms ?tx true]] "
) ;
assert_matches! (
conn . datoms ( ) ,
" [[100 :db.schema/version 11]
Schema alteration. Fixes #294 and #295. (#370) r=rnewman
* Pre: Don't retract :db/ident in test.
Datomic (and eventually Mentat) don't allow to retract :db/ident in
this way, so this runs afoul of future work to support mutating
metadata.
* Pre: s/VALUETYPE/VALUE_TYPE/.
This is consistent with the capitalization (which is "valueType") and
the other identifier.
* Pre: Remove some single quotes from error output.
* Part 1: Make materialized views be uniform [e a v value_type_tag].
This looks ahead to a time when we could support arbitrary
user-defined materialized views. For now, the "idents" materialized
view is those datoms of the form [e :db/ident :namespaced/keyword] and
the "schema" materialized view is those datoms of the form [e a v]
where a is in a particular set of attributes that will become clear in
the following commits.
This change is not backwards compatible, so I'm removing the open
current (really, v2) test. It'll be re-instated when we get to
https://github.com/mozilla/mentat/issues/194.
* Pre: Map TypedValue::Ref to TypedValue::Keyword in debug output.
* Part 3: Separate `schema_to_mutate` from the `schema` used to interpret.
This is just to keep track of the expected changes during
bootstrapping. I want bootstrap metadata mutations to flow through
the same code path as metadata mutations during regular transactions;
by differentiating the schema used for interpretation from the schema
that will be updated I expect to be able to apply bootstrap metadata
mutations to an empty schema and have things like materialized views
created (using the regular code paths).
This commit has been re-ordered for conceptual clarity, but it won't
compile because it references the metadata module. It's possible to
make it compile -- the functionality is there in the schema module --
but it's not worth the rebasing effort until after review (and
possibly not even then, since we'll squash down to a single commit to
land).
* Part 2: Maintain entids separately from idents.
In order to support historical idents, we need to distinguish the
"current" map from entid -> ident from the "complete historical" map
ident -> entid. This is what Datomic does; in Datomic, an ident is
never retracted (although it can be replaced). This approach is an
important part of allowing multiple consumers to share a schema
fragment as it migrates forward.
This fixes a limitation of the Clojure implementation, which did not
handle historical idents across knowledge base close and re-open.
The "entids" materialized view is naturally a slice of the "datoms"
table. The "idents" materialized view is a slice of the
"transactions" table. I hope that representing in this way, and
casting the problem in this light, might generalize to future
materialized views.
* Pre: Add DiffSet.
* Part 4: Collect mutations to a `Schema`.
I haven't taken your review comment about consuming AttributeBuilder
during each fluent function. If you read my response and still want
this, I'm happy to do it in review.
* Part 5: Handle :db/ident and :db.{install,alter}/attribute.
This "loops" the committed datoms out of the SQL store and back
through the metadata (schema, but in future also partition map)
processor. The metadata processor updates the schema and produces a
report of what changed; that report is then used to update the SQL
store. That update includes:
- the materialized views ("entids", "idents", and "schema");
- if needed, a subset of the datoms themselves (as flags change).
I've left a TODO for handling attribute retraction in the cases that
it makes sense. I expect that to be straight-forward.
* Review comment: Rename DiffSet to AddRetractAlterSet.
Also adds a little more commentary and a simple test.
* Review comment: Use ToIdent trait.
* Review comment: partially revert "Part 2: Maintain entids separately from idents."
This reverts commit 23a91df9c35e14398f2ddbd1ba25315821e67401.
Following our discussion, this removes the "entids" materialized
view. The next commit will remove historical idents from the "idents"
materialized view.
* Post: Use custom Either rather than std::result::Result.
This is not necessary, but it was suggested that we might be paying an
overhead creating Err instances while using error_chain. That seems
not to be the case, but this change shows that we don't actually use
any of the Result helper methods, so there's no reason to overload
Result. This change might avoid some future confusion, so I'm going
to land it anyway.
Signed-off-by: Nick Alexander <nalexander@mozilla.com>
* Review comment: Don't preserve historical idents.
* Review comment: More prepared statements when updating materialized views.
* Post: Test altering :db/cardinality and :db/unique.
These tests fail due to a Datomic limitation, namely that the marker
flag :db.alter/attribute can only be asserted once for an attribute!
That is, [:db.part/db :db.alter/attribute :attribute] will only be
transacted at most once. Since older versions of Datomic required the
:db.alter/attribute flag, I can only imagine they either never wrote
:db.alter/attribute to the store, or they handled it specially. I'll
need to remove the marker flag system from Mentat in order to address
this fundamental limitation.
* Post: Remove some more single quotes from error output.
* Post: Add assert_transact! macro to unwrap safely.
I was finding it very difficult to track unwrapping errors while
making changes, due to an underlying Mac OS X symbolication issue that
makes running tests with RUST_BACKTRACE=1 so slow that they all time
out.
* Post: Don't expect or recognize :db.{install,alter}/attribute.
I had this all working... except we will never see a repeated
`[:db.part/db :db.alter/attribute :attribute]` assertion in the store!
That means my approach would let you alter an attribute at most one
time. It's not worth hacking around this; it's better to just stop
expecting (and recognizing) the marker flags. (We have all the data
to distinguish the various cases that we need without the marker
flags.)
This brings Mentat in line with the thrust of newer Datomic versions,
but isn't compatible with Datomic, because (if I understand correctly)
Datomic automatically adds :db.{install,alter}/attribute assertions to
transactions.
I haven't purged the corresponding :db/ident and schema fragments just
yet:
- we might want them back
- we might want them in order to upgrade v1 and v2 databases to the
new on-disk layout we're fleshing out (v3?).
* Post: Don't make :db/unique :db.unique/* imply :db/index true.
This patch avoids a potential bug with the "schema" materialized view.
If :db/unique :db.unique/value implies :db/index true, then what
happens when you _retract_ :db.unique/value? I think Datomic defines
this in some way, but I really want the "schema" materialized view to
be a slice of "datoms" and not have these sort of ambiguities and
persistent effects. Therefore, to ensure that we don't retract a
schema characteristic and accidentally change more than we intended
to, this patch stops having any schema characteristic imply any other
schema characteristic(s). To achieve that, I added an
Option<Unique::{Value,Identity}> type to Attribute; this helps with
this patch, and also looks ahead to when we allow to retract
:db/unique attributes.
* Post: Allow to retract :db/ident.
* Post: Include more details about invalid schema changes.
The tests use strings, so they hide the chained errors which do in
fact provide more detail.
* Review comment: Fix outdated comment.
* Review comment: s/_SET/_SQL_LIST/.
* Review comment: Use a sub-select for checking cardinality.
This might be faster in practice.
* Review comment: Put `attribute::Unique` into its own namespace.
2017-03-20 20:18:59 +00:00
[ 101 :db . schema / version 22 ]
2017-03-20 18:29:17 +00:00
[ 200 :db . schema / attribute 100 ]
2019-07-17 15:46:47 +00:00
[ 200 :db . schema / attribute 101 ] ] "
) ;
2017-02-08 22:04:32 +00:00
}
2018-06-13 15:49:40 +00:00
#[ test ]
fn test_add ( ) {
run_test_add ( TestConn ::default ( ) ) ;
}
2018-01-22 18:17:12 +00:00
#[ test ]
fn test_tx_assertions ( ) {
let mut conn = TestConn ::default ( ) ;
// Test that txInstant can be asserted.
2019-07-17 15:46:47 +00:00
assert_transact! (
conn ,
" [[:db/add (transaction-tx) :db/txInstant #inst \" 2017-06-16T00:56:41.257Z \" ]
2018-01-22 18:17:12 +00:00
[ :db / add 100 :db / ident :name / Ivan ]
2019-07-17 15:46:47 +00:00
[ :db / add 101 :db / ident :name / Petr ] ] "
) ;
assert_matches! (
conn . last_transaction ( ) ,
" [[100 :db/ident :name/Ivan ?tx true]
2018-01-22 18:17:12 +00:00
[ 101 :db / ident :name / Petr ? tx true ]
2019-07-17 15:46:47 +00:00
[ ? tx :db / txInstant #inst \ " 2017-06-16T00:56:41.257Z \" ?tx true]] "
) ;
2018-01-22 18:17:12 +00:00
// Test multiple txInstant with different values should fail.
2018-04-26 22:48:27 +00:00
assert_transact! ( conn , " [[:db/add (transaction-tx) :db/txInstant #inst \" 2017-06-16T00:59:11.257Z \" ]
[ :db / add ( transaction - tx ) :db / txInstant #inst \ " 2017-06-16T00:59:11.752Z \" ]
2018-01-22 18:17:12 +00:00
[ :db / add 102 :db / ident :name / Vlad ] ] " ,
Add type checking and constraint checking to the transactor. (#663, #532, #679)
This should address #663, by re-inserting type checking in the
transactor stack after the entry point used by the term builder.
Before this commit, we were using an SQLite UNIQUE index to assert
that no `[e a]` pair, with `a` a cardinality one attribute, was
asserted more than once. However, that's not in line with Datomic,
which treats transaction inputs as a set and allows a single datom
like `[e a v]` to appear multiple times. It's both awkward and not
particularly efficient to look for _distinct_ repetitions in SQL, so
we accept some runtime cost in order to check for repetitions in the
transactor. This will allow us to address #532, which is really about
whether we treat inputs as sets. A side benefit is that we can
provide more helpful error messages when the transactor does detect
that the input truly violates the cardinality constraints of the
schema.
This commit builds a trie while error checking and collecting final
terms, which should be fairly efficient. It also allows a simpler
expression of input-provided :db/txInstant datoms, which in turn
uncovered a small issue with the transaction watcher, where-by the
watcher would not see non-input-provided :db/txInstant datoms.
This transition to Datomic-like input-as-set semantics allows us to
address #532. Previously, two tempids that upserted to the same entid
would produce duplicate datoms, and that would have been rejected by
the transactor -- correctly, since we did not allow duplicate datoms
under the input-as-list semantics. With input-as-set semantics,
duplicate datoms are allowed; and that means that we must allow
tempids to be equivalent, i.e., to resolve to the same tempid.
To achieve this, we:
- index the set of tempids
- identify tempid indices that share an upsert
- map tempids to a dense set of contiguous integer labels
We use the well-known union-find algorithm, as implemented by
petgraph, to efficiently manage the set of equivalent tempids.
Along the way, I've fixed and added tests for two small errors in the
transactor. First, don't drop datoms resolved by upsert (#679).
Second, ensure that complex upserts are allocated.
I don't know quite what happened here. The Clojure implementation
correctly kept complex upserts that hadn't resolved as complex
upserts (see
https://github.com/mozilla/mentat/blob/9a9dfb502acf5e4cdb1059d4aac831d7603063c8/src/common/datomish/transact.cljc#L436)
and then allocated complex upserts if they didn't resolve (see
https://github.com/mozilla/mentat/blob/9a9dfb502acf5e4cdb1059d4aac831d7603063c8/src/common/datomish/transact.cljc#L509).
Based on the code comments, I think the Rust implementation must have
incorrectly tried to optimize by handling all complex upserts in at
most a single generation of evolution, and that's just not correct.
We're effectively implementing a topological sort, using very specific
domain knowledge, and its not true that a node in a topological sort
can be considered only once!
2018-04-30 22:16:05 +00:00
Err ( " schema constraint violation: cardinality conflicts: \n CardinalityOneAddConflict { e: 268435458, a: 3, vs: {Instant(2017-06-16T00:59:11.257Z), Instant(2017-06-16T00:59:11.752Z)} } \n " ) ) ;
2018-01-22 18:17:12 +00:00
// Test multiple txInstants with the same value.
2018-04-26 22:48:27 +00:00
assert_transact! ( conn , " [[:db/add (transaction-tx) :db/txInstant #inst \" 2017-06-16T00:59:11.257Z \" ]
[ :db / add ( transaction - tx ) :db / txInstant #inst \ " 2017-06-16T00:59:11.257Z \" ]
2018-01-22 18:17:12 +00:00
[ :db / add 103 :db / ident :name / Dimitri ]
[ :db / add 104 :db / ident :name / Anton ] ] " );
2019-07-17 15:46:47 +00:00
assert_matches! (
conn . last_transaction ( ) ,
" [[103 :db/ident :name/Dimitri ?tx true]
2018-01-22 18:17:12 +00:00
[ 104 :db / ident :name / Anton ? tx true ]
2019-07-17 15:46:47 +00:00
[ ? tx :db / txInstant #inst \ " 2017-06-16T00:59:11.257Z \" ?tx true]] "
) ;
2018-01-22 18:17:12 +00:00
2018-04-26 22:48:27 +00:00
// We need a few attributes to work with.
2019-07-17 15:46:47 +00:00
assert_transact! (
conn ,
" [[:db/add 111 :db/ident :test/str]
2018-04-26 22:48:27 +00:00
[ :db / add 111 :db / valueType :db . type / string ]
[ :db / add 222 :db / ident :test / ref ]
2019-07-17 15:46:47 +00:00
[ :db / add 222 :db / valueType :db . type / ref ] ] "
) ;
2018-04-26 22:48:27 +00:00
// Test that we can assert metadata about the current transaction.
2019-07-17 15:46:47 +00:00
assert_transact! (
conn ,
" [[:db/add (transaction-tx) :test/str \" We want metadata! \" ]] "
) ;
assert_matches! (
conn . last_transaction ( ) ,
" [[?tx :db/txInstant ?ms ?tx true]
[ ? tx :test / str \ " We want metadata! \" ?tx true]] "
) ;
2018-04-26 22:48:27 +00:00
// Test that we can use (transaction-tx) as a value.
assert_transact! ( conn , " [[:db/add 333 :test/ref (transaction-tx)]] " ) ;
2019-07-17 15:46:47 +00:00
assert_matches! (
conn . last_transaction ( ) ,
" [[333 :test/ref ?tx ?tx true]
[ ? tx :db / txInstant ? ms ? tx true ] ] "
) ;
2018-04-26 22:48:27 +00:00
// Test that we type-check properly. In the value position, (transaction-tx) yields a ref;
// :db/ident expects a keyword.
assert_transact! ( conn , " [[:db/add 444 :db/ident (transaction-tx)]] " ,
Err ( " not yet implemented: Transaction function transaction-tx produced value of type :db.type/ref but expected type :db.type/keyword " ) ) ;
// Test that we can assert metadata about the current transaction.
2019-07-17 15:46:47 +00:00
assert_transact! (
conn ,
" [[:db/add (transaction-tx) :test/ref (transaction-tx)]] "
) ;
assert_matches! (
conn . last_transaction ( ) ,
" [[?tx :db/txInstant ?ms ?tx true]
[ ? tx :test / ref ? tx ? tx true ] ] "
) ;
2018-01-22 18:17:12 +00:00
}
2017-02-08 22:04:32 +00:00
#[ test ]
fn test_retract ( ) {
2017-03-20 18:29:17 +00:00
let mut conn = TestConn ::default ( ) ;
// Insert a few :db.cardinality/one elements.
2019-07-17 15:46:47 +00:00
assert_transact! (
conn ,
" [[:db/add 100 :db.schema/version 1]
[ :db / add 101 :db . schema / version 2 ] ] "
) ;
assert_matches! (
conn . last_transaction ( ) ,
" [[100 :db.schema/version 1 ?tx true]
Schema alteration. Fixes #294 and #295. (#370) r=rnewman
* Pre: Don't retract :db/ident in test.
Datomic (and eventually Mentat) don't allow to retract :db/ident in
this way, so this runs afoul of future work to support mutating
metadata.
* Pre: s/VALUETYPE/VALUE_TYPE/.
This is consistent with the capitalization (which is "valueType") and
the other identifier.
* Pre: Remove some single quotes from error output.
* Part 1: Make materialized views be uniform [e a v value_type_tag].
This looks ahead to a time when we could support arbitrary
user-defined materialized views. For now, the "idents" materialized
view is those datoms of the form [e :db/ident :namespaced/keyword] and
the "schema" materialized view is those datoms of the form [e a v]
where a is in a particular set of attributes that will become clear in
the following commits.
This change is not backwards compatible, so I'm removing the open
current (really, v2) test. It'll be re-instated when we get to
https://github.com/mozilla/mentat/issues/194.
* Pre: Map TypedValue::Ref to TypedValue::Keyword in debug output.
* Part 3: Separate `schema_to_mutate` from the `schema` used to interpret.
This is just to keep track of the expected changes during
bootstrapping. I want bootstrap metadata mutations to flow through
the same code path as metadata mutations during regular transactions;
by differentiating the schema used for interpretation from the schema
that will be updated I expect to be able to apply bootstrap metadata
mutations to an empty schema and have things like materialized views
created (using the regular code paths).
This commit has been re-ordered for conceptual clarity, but it won't
compile because it references the metadata module. It's possible to
make it compile -- the functionality is there in the schema module --
but it's not worth the rebasing effort until after review (and
possibly not even then, since we'll squash down to a single commit to
land).
* Part 2: Maintain entids separately from idents.
In order to support historical idents, we need to distinguish the
"current" map from entid -> ident from the "complete historical" map
ident -> entid. This is what Datomic does; in Datomic, an ident is
never retracted (although it can be replaced). This approach is an
important part of allowing multiple consumers to share a schema
fragment as it migrates forward.
This fixes a limitation of the Clojure implementation, which did not
handle historical idents across knowledge base close and re-open.
The "entids" materialized view is naturally a slice of the "datoms"
table. The "idents" materialized view is a slice of the
"transactions" table. I hope that representing in this way, and
casting the problem in this light, might generalize to future
materialized views.
* Pre: Add DiffSet.
* Part 4: Collect mutations to a `Schema`.
I haven't taken your review comment about consuming AttributeBuilder
during each fluent function. If you read my response and still want
this, I'm happy to do it in review.
* Part 5: Handle :db/ident and :db.{install,alter}/attribute.
This "loops" the committed datoms out of the SQL store and back
through the metadata (schema, but in future also partition map)
processor. The metadata processor updates the schema and produces a
report of what changed; that report is then used to update the SQL
store. That update includes:
- the materialized views ("entids", "idents", and "schema");
- if needed, a subset of the datoms themselves (as flags change).
I've left a TODO for handling attribute retraction in the cases that
it makes sense. I expect that to be straight-forward.
* Review comment: Rename DiffSet to AddRetractAlterSet.
Also adds a little more commentary and a simple test.
* Review comment: Use ToIdent trait.
* Review comment: partially revert "Part 2: Maintain entids separately from idents."
This reverts commit 23a91df9c35e14398f2ddbd1ba25315821e67401.
Following our discussion, this removes the "entids" materialized
view. The next commit will remove historical idents from the "idents"
materialized view.
* Post: Use custom Either rather than std::result::Result.
This is not necessary, but it was suggested that we might be paying an
overhead creating Err instances while using error_chain. That seems
not to be the case, but this change shows that we don't actually use
any of the Result helper methods, so there's no reason to overload
Result. This change might avoid some future confusion, so I'm going
to land it anyway.
Signed-off-by: Nick Alexander <nalexander@mozilla.com>
* Review comment: Don't preserve historical idents.
* Review comment: More prepared statements when updating materialized views.
* Post: Test altering :db/cardinality and :db/unique.
These tests fail due to a Datomic limitation, namely that the marker
flag :db.alter/attribute can only be asserted once for an attribute!
That is, [:db.part/db :db.alter/attribute :attribute] will only be
transacted at most once. Since older versions of Datomic required the
:db.alter/attribute flag, I can only imagine they either never wrote
:db.alter/attribute to the store, or they handled it specially. I'll
need to remove the marker flag system from Mentat in order to address
this fundamental limitation.
* Post: Remove some more single quotes from error output.
* Post: Add assert_transact! macro to unwrap safely.
I was finding it very difficult to track unwrapping errors while
making changes, due to an underlying Mac OS X symbolication issue that
makes running tests with RUST_BACKTRACE=1 so slow that they all time
out.
* Post: Don't expect or recognize :db.{install,alter}/attribute.
I had this all working... except we will never see a repeated
`[:db.part/db :db.alter/attribute :attribute]` assertion in the store!
That means my approach would let you alter an attribute at most one
time. It's not worth hacking around this; it's better to just stop
expecting (and recognizing) the marker flags. (We have all the data
to distinguish the various cases that we need without the marker
flags.)
This brings Mentat in line with the thrust of newer Datomic versions,
but isn't compatible with Datomic, because (if I understand correctly)
Datomic automatically adds :db.{install,alter}/attribute assertions to
transactions.
I haven't purged the corresponding :db/ident and schema fragments just
yet:
- we might want them back
- we might want them in order to upgrade v1 and v2 databases to the
new on-disk layout we're fleshing out (v3?).
* Post: Don't make :db/unique :db.unique/* imply :db/index true.
This patch avoids a potential bug with the "schema" materialized view.
If :db/unique :db.unique/value implies :db/index true, then what
happens when you _retract_ :db.unique/value? I think Datomic defines
this in some way, but I really want the "schema" materialized view to
be a slice of "datoms" and not have these sort of ambiguities and
persistent effects. Therefore, to ensure that we don't retract a
schema characteristic and accidentally change more than we intended
to, this patch stops having any schema characteristic imply any other
schema characteristic(s). To achieve that, I added an
Option<Unique::{Value,Identity}> type to Attribute; this helps with
this patch, and also looks ahead to when we allow to retract
:db/unique attributes.
* Post: Allow to retract :db/ident.
* Post: Include more details about invalid schema changes.
The tests use strings, so they hide the chained errors which do in
fact provide more detail.
* Review comment: Fix outdated comment.
* Review comment: s/_SET/_SQL_LIST/.
* Review comment: Use a sub-select for checking cardinality.
This might be faster in practice.
* Review comment: Put `attribute::Unique` into its own namespace.
2017-03-20 20:18:59 +00:00
[ 101 :db . schema / version 2 ? tx true ]
2019-07-17 15:46:47 +00:00
[ ? tx :db / txInstant ? ms ? tx true ] ] "
) ;
assert_matches! (
conn . datoms ( ) ,
" [[100 :db.schema/version 1]
[ 101 :db . schema / version 2 ] ] "
) ;
2017-03-20 18:29:17 +00:00
// And a few :db.cardinality/many elements.
2019-07-17 15:46:47 +00:00
assert_transact! (
conn ,
" [[:db/add 200 :db.schema/attribute 100]
[ :db / add 200 :db . schema / attribute 101 ] ] "
) ;
assert_matches! (
conn . last_transaction ( ) ,
" [[200 :db.schema/attribute 100 ?tx true]
2017-03-20 18:29:17 +00:00
[ 200 :db . schema / attribute 101 ? tx true ]
2019-07-17 15:46:47 +00:00
[ ? tx :db / txInstant ? ms ? tx true ] ] "
) ;
assert_matches! (
conn . datoms ( ) ,
" [[100 :db.schema/version 1]
Schema alteration. Fixes #294 and #295. (#370) r=rnewman
* Pre: Don't retract :db/ident in test.
Datomic (and eventually Mentat) don't allow to retract :db/ident in
this way, so this runs afoul of future work to support mutating
metadata.
* Pre: s/VALUETYPE/VALUE_TYPE/.
This is consistent with the capitalization (which is "valueType") and
the other identifier.
* Pre: Remove some single quotes from error output.
* Part 1: Make materialized views be uniform [e a v value_type_tag].
This looks ahead to a time when we could support arbitrary
user-defined materialized views. For now, the "idents" materialized
view is those datoms of the form [e :db/ident :namespaced/keyword] and
the "schema" materialized view is those datoms of the form [e a v]
where a is in a particular set of attributes that will become clear in
the following commits.
This change is not backwards compatible, so I'm removing the open
current (really, v2) test. It'll be re-instated when we get to
https://github.com/mozilla/mentat/issues/194.
* Pre: Map TypedValue::Ref to TypedValue::Keyword in debug output.
* Part 3: Separate `schema_to_mutate` from the `schema` used to interpret.
This is just to keep track of the expected changes during
bootstrapping. I want bootstrap metadata mutations to flow through
the same code path as metadata mutations during regular transactions;
by differentiating the schema used for interpretation from the schema
that will be updated I expect to be able to apply bootstrap metadata
mutations to an empty schema and have things like materialized views
created (using the regular code paths).
This commit has been re-ordered for conceptual clarity, but it won't
compile because it references the metadata module. It's possible to
make it compile -- the functionality is there in the schema module --
but it's not worth the rebasing effort until after review (and
possibly not even then, since we'll squash down to a single commit to
land).
* Part 2: Maintain entids separately from idents.
In order to support historical idents, we need to distinguish the
"current" map from entid -> ident from the "complete historical" map
ident -> entid. This is what Datomic does; in Datomic, an ident is
never retracted (although it can be replaced). This approach is an
important part of allowing multiple consumers to share a schema
fragment as it migrates forward.
This fixes a limitation of the Clojure implementation, which did not
handle historical idents across knowledge base close and re-open.
The "entids" materialized view is naturally a slice of the "datoms"
table. The "idents" materialized view is a slice of the
"transactions" table. I hope that representing in this way, and
casting the problem in this light, might generalize to future
materialized views.
* Pre: Add DiffSet.
* Part 4: Collect mutations to a `Schema`.
I haven't taken your review comment about consuming AttributeBuilder
during each fluent function. If you read my response and still want
this, I'm happy to do it in review.
* Part 5: Handle :db/ident and :db.{install,alter}/attribute.
This "loops" the committed datoms out of the SQL store and back
through the metadata (schema, but in future also partition map)
processor. The metadata processor updates the schema and produces a
report of what changed; that report is then used to update the SQL
store. That update includes:
- the materialized views ("entids", "idents", and "schema");
- if needed, a subset of the datoms themselves (as flags change).
I've left a TODO for handling attribute retraction in the cases that
it makes sense. I expect that to be straight-forward.
* Review comment: Rename DiffSet to AddRetractAlterSet.
Also adds a little more commentary and a simple test.
* Review comment: Use ToIdent trait.
* Review comment: partially revert "Part 2: Maintain entids separately from idents."
This reverts commit 23a91df9c35e14398f2ddbd1ba25315821e67401.
Following our discussion, this removes the "entids" materialized
view. The next commit will remove historical idents from the "idents"
materialized view.
* Post: Use custom Either rather than std::result::Result.
This is not necessary, but it was suggested that we might be paying an
overhead creating Err instances while using error_chain. That seems
not to be the case, but this change shows that we don't actually use
any of the Result helper methods, so there's no reason to overload
Result. This change might avoid some future confusion, so I'm going
to land it anyway.
Signed-off-by: Nick Alexander <nalexander@mozilla.com>
* Review comment: Don't preserve historical idents.
* Review comment: More prepared statements when updating materialized views.
* Post: Test altering :db/cardinality and :db/unique.
These tests fail due to a Datomic limitation, namely that the marker
flag :db.alter/attribute can only be asserted once for an attribute!
That is, [:db.part/db :db.alter/attribute :attribute] will only be
transacted at most once. Since older versions of Datomic required the
:db.alter/attribute flag, I can only imagine they either never wrote
:db.alter/attribute to the store, or they handled it specially. I'll
need to remove the marker flag system from Mentat in order to address
this fundamental limitation.
* Post: Remove some more single quotes from error output.
* Post: Add assert_transact! macro to unwrap safely.
I was finding it very difficult to track unwrapping errors while
making changes, due to an underlying Mac OS X symbolication issue that
makes running tests with RUST_BACKTRACE=1 so slow that they all time
out.
* Post: Don't expect or recognize :db.{install,alter}/attribute.
I had this all working... except we will never see a repeated
`[:db.part/db :db.alter/attribute :attribute]` assertion in the store!
That means my approach would let you alter an attribute at most one
time. It's not worth hacking around this; it's better to just stop
expecting (and recognizing) the marker flags. (We have all the data
to distinguish the various cases that we need without the marker
flags.)
This brings Mentat in line with the thrust of newer Datomic versions,
but isn't compatible with Datomic, because (if I understand correctly)
Datomic automatically adds :db.{install,alter}/attribute assertions to
transactions.
I haven't purged the corresponding :db/ident and schema fragments just
yet:
- we might want them back
- we might want them in order to upgrade v1 and v2 databases to the
new on-disk layout we're fleshing out (v3?).
* Post: Don't make :db/unique :db.unique/* imply :db/index true.
This patch avoids a potential bug with the "schema" materialized view.
If :db/unique :db.unique/value implies :db/index true, then what
happens when you _retract_ :db.unique/value? I think Datomic defines
this in some way, but I really want the "schema" materialized view to
be a slice of "datoms" and not have these sort of ambiguities and
persistent effects. Therefore, to ensure that we don't retract a
schema characteristic and accidentally change more than we intended
to, this patch stops having any schema characteristic imply any other
schema characteristic(s). To achieve that, I added an
Option<Unique::{Value,Identity}> type to Attribute; this helps with
this patch, and also looks ahead to when we allow to retract
:db/unique attributes.
* Post: Allow to retract :db/ident.
* Post: Include more details about invalid schema changes.
The tests use strings, so they hide the chained errors which do in
fact provide more detail.
* Review comment: Fix outdated comment.
* Review comment: s/_SET/_SQL_LIST/.
* Review comment: Use a sub-select for checking cardinality.
This might be faster in practice.
* Review comment: Put `attribute::Unique` into its own namespace.
2017-03-20 20:18:59 +00:00
[ 101 :db . schema / version 2 ]
2017-03-20 18:29:17 +00:00
[ 200 :db . schema / attribute 100 ]
2019-07-17 15:46:47 +00:00
[ 200 :db . schema / attribute 101 ] ] "
) ;
2017-03-20 18:29:17 +00:00
// Test that we can retract :db.cardinality/one elements.
Schema alteration. Fixes #294 and #295. (#370) r=rnewman
* Pre: Don't retract :db/ident in test.
Datomic (and eventually Mentat) don't allow to retract :db/ident in
this way, so this runs afoul of future work to support mutating
metadata.
* Pre: s/VALUETYPE/VALUE_TYPE/.
This is consistent with the capitalization (which is "valueType") and
the other identifier.
* Pre: Remove some single quotes from error output.
* Part 1: Make materialized views be uniform [e a v value_type_tag].
This looks ahead to a time when we could support arbitrary
user-defined materialized views. For now, the "idents" materialized
view is those datoms of the form [e :db/ident :namespaced/keyword] and
the "schema" materialized view is those datoms of the form [e a v]
where a is in a particular set of attributes that will become clear in
the following commits.
This change is not backwards compatible, so I'm removing the open
current (really, v2) test. It'll be re-instated when we get to
https://github.com/mozilla/mentat/issues/194.
* Pre: Map TypedValue::Ref to TypedValue::Keyword in debug output.
* Part 3: Separate `schema_to_mutate` from the `schema` used to interpret.
This is just to keep track of the expected changes during
bootstrapping. I want bootstrap metadata mutations to flow through
the same code path as metadata mutations during regular transactions;
by differentiating the schema used for interpretation from the schema
that will be updated I expect to be able to apply bootstrap metadata
mutations to an empty schema and have things like materialized views
created (using the regular code paths).
This commit has been re-ordered for conceptual clarity, but it won't
compile because it references the metadata module. It's possible to
make it compile -- the functionality is there in the schema module --
but it's not worth the rebasing effort until after review (and
possibly not even then, since we'll squash down to a single commit to
land).
* Part 2: Maintain entids separately from idents.
In order to support historical idents, we need to distinguish the
"current" map from entid -> ident from the "complete historical" map
ident -> entid. This is what Datomic does; in Datomic, an ident is
never retracted (although it can be replaced). This approach is an
important part of allowing multiple consumers to share a schema
fragment as it migrates forward.
This fixes a limitation of the Clojure implementation, which did not
handle historical idents across knowledge base close and re-open.
The "entids" materialized view is naturally a slice of the "datoms"
table. The "idents" materialized view is a slice of the
"transactions" table. I hope that representing in this way, and
casting the problem in this light, might generalize to future
materialized views.
* Pre: Add DiffSet.
* Part 4: Collect mutations to a `Schema`.
I haven't taken your review comment about consuming AttributeBuilder
during each fluent function. If you read my response and still want
this, I'm happy to do it in review.
* Part 5: Handle :db/ident and :db.{install,alter}/attribute.
This "loops" the committed datoms out of the SQL store and back
through the metadata (schema, but in future also partition map)
processor. The metadata processor updates the schema and produces a
report of what changed; that report is then used to update the SQL
store. That update includes:
- the materialized views ("entids", "idents", and "schema");
- if needed, a subset of the datoms themselves (as flags change).
I've left a TODO for handling attribute retraction in the cases that
it makes sense. I expect that to be straight-forward.
* Review comment: Rename DiffSet to AddRetractAlterSet.
Also adds a little more commentary and a simple test.
* Review comment: Use ToIdent trait.
* Review comment: partially revert "Part 2: Maintain entids separately from idents."
This reverts commit 23a91df9c35e14398f2ddbd1ba25315821e67401.
Following our discussion, this removes the "entids" materialized
view. The next commit will remove historical idents from the "idents"
materialized view.
* Post: Use custom Either rather than std::result::Result.
This is not necessary, but it was suggested that we might be paying an
overhead creating Err instances while using error_chain. That seems
not to be the case, but this change shows that we don't actually use
any of the Result helper methods, so there's no reason to overload
Result. This change might avoid some future confusion, so I'm going
to land it anyway.
Signed-off-by: Nick Alexander <nalexander@mozilla.com>
* Review comment: Don't preserve historical idents.
* Review comment: More prepared statements when updating materialized views.
* Post: Test altering :db/cardinality and :db/unique.
These tests fail due to a Datomic limitation, namely that the marker
flag :db.alter/attribute can only be asserted once for an attribute!
That is, [:db.part/db :db.alter/attribute :attribute] will only be
transacted at most once. Since older versions of Datomic required the
:db.alter/attribute flag, I can only imagine they either never wrote
:db.alter/attribute to the store, or they handled it specially. I'll
need to remove the marker flag system from Mentat in order to address
this fundamental limitation.
* Post: Remove some more single quotes from error output.
* Post: Add assert_transact! macro to unwrap safely.
I was finding it very difficult to track unwrapping errors while
making changes, due to an underlying Mac OS X symbolication issue that
makes running tests with RUST_BACKTRACE=1 so slow that they all time
out.
* Post: Don't expect or recognize :db.{install,alter}/attribute.
I had this all working... except we will never see a repeated
`[:db.part/db :db.alter/attribute :attribute]` assertion in the store!
That means my approach would let you alter an attribute at most one
time. It's not worth hacking around this; it's better to just stop
expecting (and recognizing) the marker flags. (We have all the data
to distinguish the various cases that we need without the marker
flags.)
This brings Mentat in line with the thrust of newer Datomic versions,
but isn't compatible with Datomic, because (if I understand correctly)
Datomic automatically adds :db.{install,alter}/attribute assertions to
transactions.
I haven't purged the corresponding :db/ident and schema fragments just
yet:
- we might want them back
- we might want them in order to upgrade v1 and v2 databases to the
new on-disk layout we're fleshing out (v3?).
* Post: Don't make :db/unique :db.unique/* imply :db/index true.
This patch avoids a potential bug with the "schema" materialized view.
If :db/unique :db.unique/value implies :db/index true, then what
happens when you _retract_ :db.unique/value? I think Datomic defines
this in some way, but I really want the "schema" materialized view to
be a slice of "datoms" and not have these sort of ambiguities and
persistent effects. Therefore, to ensure that we don't retract a
schema characteristic and accidentally change more than we intended
to, this patch stops having any schema characteristic imply any other
schema characteristic(s). To achieve that, I added an
Option<Unique::{Value,Identity}> type to Attribute; this helps with
this patch, and also looks ahead to when we allow to retract
:db/unique attributes.
* Post: Allow to retract :db/ident.
* Post: Include more details about invalid schema changes.
The tests use strings, so they hide the chained errors which do in
fact provide more detail.
* Review comment: Fix outdated comment.
* Review comment: s/_SET/_SQL_LIST/.
* Review comment: Use a sub-select for checking cardinality.
This might be faster in practice.
* Review comment: Put `attribute::Unique` into its own namespace.
2017-03-20 20:18:59 +00:00
assert_transact! ( conn , " [[:db/retract 100 :db.schema/version 1]] " ) ;
2019-07-17 15:46:47 +00:00
assert_matches! (
conn . last_transaction ( ) ,
" [[100 :db.schema/version 1 ?tx false]
[ ? tx :db / txInstant ? ms ? tx true ] ] "
) ;
assert_matches! (
conn . datoms ( ) ,
" [[101 :db.schema/version 2]
2017-03-20 18:29:17 +00:00
[ 200 :db . schema / attribute 100 ]
2019-07-17 15:46:47 +00:00
[ 200 :db . schema / attribute 101 ] ] "
) ;
2017-03-20 18:29:17 +00:00
// Test that we can retract :db.cardinality/many elements.
Schema alteration. Fixes #294 and #295. (#370) r=rnewman
* Pre: Don't retract :db/ident in test.
Datomic (and eventually Mentat) don't allow to retract :db/ident in
this way, so this runs afoul of future work to support mutating
metadata.
* Pre: s/VALUETYPE/VALUE_TYPE/.
This is consistent with the capitalization (which is "valueType") and
the other identifier.
* Pre: Remove some single quotes from error output.
* Part 1: Make materialized views be uniform [e a v value_type_tag].
This looks ahead to a time when we could support arbitrary
user-defined materialized views. For now, the "idents" materialized
view is those datoms of the form [e :db/ident :namespaced/keyword] and
the "schema" materialized view is those datoms of the form [e a v]
where a is in a particular set of attributes that will become clear in
the following commits.
This change is not backwards compatible, so I'm removing the open
current (really, v2) test. It'll be re-instated when we get to
https://github.com/mozilla/mentat/issues/194.
* Pre: Map TypedValue::Ref to TypedValue::Keyword in debug output.
* Part 3: Separate `schema_to_mutate` from the `schema` used to interpret.
This is just to keep track of the expected changes during
bootstrapping. I want bootstrap metadata mutations to flow through
the same code path as metadata mutations during regular transactions;
by differentiating the schema used for interpretation from the schema
that will be updated I expect to be able to apply bootstrap metadata
mutations to an empty schema and have things like materialized views
created (using the regular code paths).
This commit has been re-ordered for conceptual clarity, but it won't
compile because it references the metadata module. It's possible to
make it compile -- the functionality is there in the schema module --
but it's not worth the rebasing effort until after review (and
possibly not even then, since we'll squash down to a single commit to
land).
* Part 2: Maintain entids separately from idents.
In order to support historical idents, we need to distinguish the
"current" map from entid -> ident from the "complete historical" map
ident -> entid. This is what Datomic does; in Datomic, an ident is
never retracted (although it can be replaced). This approach is an
important part of allowing multiple consumers to share a schema
fragment as it migrates forward.
This fixes a limitation of the Clojure implementation, which did not
handle historical idents across knowledge base close and re-open.
The "entids" materialized view is naturally a slice of the "datoms"
table. The "idents" materialized view is a slice of the
"transactions" table. I hope that representing in this way, and
casting the problem in this light, might generalize to future
materialized views.
* Pre: Add DiffSet.
* Part 4: Collect mutations to a `Schema`.
I haven't taken your review comment about consuming AttributeBuilder
during each fluent function. If you read my response and still want
this, I'm happy to do it in review.
* Part 5: Handle :db/ident and :db.{install,alter}/attribute.
This "loops" the committed datoms out of the SQL store and back
through the metadata (schema, but in future also partition map)
processor. The metadata processor updates the schema and produces a
report of what changed; that report is then used to update the SQL
store. That update includes:
- the materialized views ("entids", "idents", and "schema");
- if needed, a subset of the datoms themselves (as flags change).
I've left a TODO for handling attribute retraction in the cases that
it makes sense. I expect that to be straight-forward.
* Review comment: Rename DiffSet to AddRetractAlterSet.
Also adds a little more commentary and a simple test.
* Review comment: Use ToIdent trait.
* Review comment: partially revert "Part 2: Maintain entids separately from idents."
This reverts commit 23a91df9c35e14398f2ddbd1ba25315821e67401.
Following our discussion, this removes the "entids" materialized
view. The next commit will remove historical idents from the "idents"
materialized view.
* Post: Use custom Either rather than std::result::Result.
This is not necessary, but it was suggested that we might be paying an
overhead creating Err instances while using error_chain. That seems
not to be the case, but this change shows that we don't actually use
any of the Result helper methods, so there's no reason to overload
Result. This change might avoid some future confusion, so I'm going
to land it anyway.
Signed-off-by: Nick Alexander <nalexander@mozilla.com>
* Review comment: Don't preserve historical idents.
* Review comment: More prepared statements when updating materialized views.
* Post: Test altering :db/cardinality and :db/unique.
These tests fail due to a Datomic limitation, namely that the marker
flag :db.alter/attribute can only be asserted once for an attribute!
That is, [:db.part/db :db.alter/attribute :attribute] will only be
transacted at most once. Since older versions of Datomic required the
:db.alter/attribute flag, I can only imagine they either never wrote
:db.alter/attribute to the store, or they handled it specially. I'll
need to remove the marker flag system from Mentat in order to address
this fundamental limitation.
* Post: Remove some more single quotes from error output.
* Post: Add assert_transact! macro to unwrap safely.
I was finding it very difficult to track unwrapping errors while
making changes, due to an underlying Mac OS X symbolication issue that
makes running tests with RUST_BACKTRACE=1 so slow that they all time
out.
* Post: Don't expect or recognize :db.{install,alter}/attribute.
I had this all working... except we will never see a repeated
`[:db.part/db :db.alter/attribute :attribute]` assertion in the store!
That means my approach would let you alter an attribute at most one
time. It's not worth hacking around this; it's better to just stop
expecting (and recognizing) the marker flags. (We have all the data
to distinguish the various cases that we need without the marker
flags.)
This brings Mentat in line with the thrust of newer Datomic versions,
but isn't compatible with Datomic, because (if I understand correctly)
Datomic automatically adds :db.{install,alter}/attribute assertions to
transactions.
I haven't purged the corresponding :db/ident and schema fragments just
yet:
- we might want them back
- we might want them in order to upgrade v1 and v2 databases to the
new on-disk layout we're fleshing out (v3?).
* Post: Don't make :db/unique :db.unique/* imply :db/index true.
This patch avoids a potential bug with the "schema" materialized view.
If :db/unique :db.unique/value implies :db/index true, then what
happens when you _retract_ :db.unique/value? I think Datomic defines
this in some way, but I really want the "schema" materialized view to
be a slice of "datoms" and not have these sort of ambiguities and
persistent effects. Therefore, to ensure that we don't retract a
schema characteristic and accidentally change more than we intended
to, this patch stops having any schema characteristic imply any other
schema characteristic(s). To achieve that, I added an
Option<Unique::{Value,Identity}> type to Attribute; this helps with
this patch, and also looks ahead to when we allow to retract
:db/unique attributes.
* Post: Allow to retract :db/ident.
* Post: Include more details about invalid schema changes.
The tests use strings, so they hide the chained errors which do in
fact provide more detail.
* Review comment: Fix outdated comment.
* Review comment: s/_SET/_SQL_LIST/.
* Review comment: Use a sub-select for checking cardinality.
This might be faster in practice.
* Review comment: Put `attribute::Unique` into its own namespace.
2017-03-20 20:18:59 +00:00
assert_transact! ( conn , " [[:db/retract 200 :db.schema/attribute 100]] " ) ;
2019-07-17 15:46:47 +00:00
assert_matches! (
conn . last_transaction ( ) ,
" [[200 :db.schema/attribute 100 ?tx false]
[ ? tx :db / txInstant ? ms ? tx true ] ] "
) ;
assert_matches! (
conn . datoms ( ) ,
" [[101 :db.schema/version 2]
[ 200 :db . schema / attribute 101 ] ] "
) ;
2017-03-20 18:29:17 +00:00
// Verify that retracting :db.cardinality/{one,many} elements that are not present doesn't
// change the store.
2019-07-17 15:46:47 +00:00
assert_transact! (
conn ,
" [[:db/retract 100 :db.schema/version 1]
[ :db / retract 200 :db . schema / attribute 100 ] ] "
) ;
assert_matches! (
conn . last_transaction ( ) ,
" [[?tx :db/txInstant ?ms ?tx true]] "
) ;
assert_matches! (
conn . datoms ( ) ,
" [[101 :db.schema/version 2]
[ 200 :db . schema / attribute 101 ] ] "
) ;
2017-02-15 00:50:40 +00:00
}
2018-07-13 21:26:10 +00:00
#[ test ]
fn test_db_doc_is_not_schema ( ) {
let mut conn = TestConn ::default ( ) ;
// Neither transaction below is defining a new attribute. That is, it's fine to use :db/doc
// to describe any entity in the system, not just attributes. And in particular, including
// :db/doc shouldn't make the transactor consider the entity a schema attribute.
2019-07-17 15:46:47 +00:00
assert_transact! (
conn ,
r #"
2018-07-13 21:26:10 +00:00
[ { :db / doc " test " } ]
2019-07-17 15:46:47 +00:00
" #
) ;
2018-07-13 21:26:10 +00:00
2019-07-17 15:46:47 +00:00
assert_transact! (
conn ,
r #"
2018-07-13 21:26:10 +00:00
[ { :db / ident :test / id :db / doc " test " } ]
2019-07-17 15:46:47 +00:00
" #
) ;
2018-07-13 21:26:10 +00:00
}
2018-02-01 17:06:01 +00:00
// Unique is required!
#[ test ]
fn test_upsert_issue_538 ( ) {
let mut conn = TestConn ::default ( ) ;
assert_transact! ( conn , "
[ { :db / ident :person / name
:db / valueType :db . type / string
:db / cardinality :db . cardinality / many }
{ :db / ident :person / age
:db / valueType :db . type / long
:db / cardinality :db . cardinality / one }
{ :db / ident :person / email
:db / valueType :db . type / string
:db / unique :db . unique / identity
:db / cardinality :db . cardinality / many } ] " ,
Err ( " bad schema assertion: :db/unique :db/unique_identity without :db/index true for entid: 65538 " ) ) ;
}
Schema alteration. Fixes #294 and #295. (#370) r=rnewman
* Pre: Don't retract :db/ident in test.
Datomic (and eventually Mentat) don't allow to retract :db/ident in
this way, so this runs afoul of future work to support mutating
metadata.
* Pre: s/VALUETYPE/VALUE_TYPE/.
This is consistent with the capitalization (which is "valueType") and
the other identifier.
* Pre: Remove some single quotes from error output.
* Part 1: Make materialized views be uniform [e a v value_type_tag].
This looks ahead to a time when we could support arbitrary
user-defined materialized views. For now, the "idents" materialized
view is those datoms of the form [e :db/ident :namespaced/keyword] and
the "schema" materialized view is those datoms of the form [e a v]
where a is in a particular set of attributes that will become clear in
the following commits.
This change is not backwards compatible, so I'm removing the open
current (really, v2) test. It'll be re-instated when we get to
https://github.com/mozilla/mentat/issues/194.
* Pre: Map TypedValue::Ref to TypedValue::Keyword in debug output.
* Part 3: Separate `schema_to_mutate` from the `schema` used to interpret.
This is just to keep track of the expected changes during
bootstrapping. I want bootstrap metadata mutations to flow through
the same code path as metadata mutations during regular transactions;
by differentiating the schema used for interpretation from the schema
that will be updated I expect to be able to apply bootstrap metadata
mutations to an empty schema and have things like materialized views
created (using the regular code paths).
This commit has been re-ordered for conceptual clarity, but it won't
compile because it references the metadata module. It's possible to
make it compile -- the functionality is there in the schema module --
but it's not worth the rebasing effort until after review (and
possibly not even then, since we'll squash down to a single commit to
land).
* Part 2: Maintain entids separately from idents.
In order to support historical idents, we need to distinguish the
"current" map from entid -> ident from the "complete historical" map
ident -> entid. This is what Datomic does; in Datomic, an ident is
never retracted (although it can be replaced). This approach is an
important part of allowing multiple consumers to share a schema
fragment as it migrates forward.
This fixes a limitation of the Clojure implementation, which did not
handle historical idents across knowledge base close and re-open.
The "entids" materialized view is naturally a slice of the "datoms"
table. The "idents" materialized view is a slice of the
"transactions" table. I hope that representing in this way, and
casting the problem in this light, might generalize to future
materialized views.
* Pre: Add DiffSet.
* Part 4: Collect mutations to a `Schema`.
I haven't taken your review comment about consuming AttributeBuilder
during each fluent function. If you read my response and still want
this, I'm happy to do it in review.
* Part 5: Handle :db/ident and :db.{install,alter}/attribute.
This "loops" the committed datoms out of the SQL store and back
through the metadata (schema, but in future also partition map)
processor. The metadata processor updates the schema and produces a
report of what changed; that report is then used to update the SQL
store. That update includes:
- the materialized views ("entids", "idents", and "schema");
- if needed, a subset of the datoms themselves (as flags change).
I've left a TODO for handling attribute retraction in the cases that
it makes sense. I expect that to be straight-forward.
* Review comment: Rename DiffSet to AddRetractAlterSet.
Also adds a little more commentary and a simple test.
* Review comment: Use ToIdent trait.
* Review comment: partially revert "Part 2: Maintain entids separately from idents."
This reverts commit 23a91df9c35e14398f2ddbd1ba25315821e67401.
Following our discussion, this removes the "entids" materialized
view. The next commit will remove historical idents from the "idents"
materialized view.
* Post: Use custom Either rather than std::result::Result.
This is not necessary, but it was suggested that we might be paying an
overhead creating Err instances while using error_chain. That seems
not to be the case, but this change shows that we don't actually use
any of the Result helper methods, so there's no reason to overload
Result. This change might avoid some future confusion, so I'm going
to land it anyway.
Signed-off-by: Nick Alexander <nalexander@mozilla.com>
* Review comment: Don't preserve historical idents.
* Review comment: More prepared statements when updating materialized views.
* Post: Test altering :db/cardinality and :db/unique.
These tests fail due to a Datomic limitation, namely that the marker
flag :db.alter/attribute can only be asserted once for an attribute!
That is, [:db.part/db :db.alter/attribute :attribute] will only be
transacted at most once. Since older versions of Datomic required the
:db.alter/attribute flag, I can only imagine they either never wrote
:db.alter/attribute to the store, or they handled it specially. I'll
need to remove the marker flag system from Mentat in order to address
this fundamental limitation.
* Post: Remove some more single quotes from error output.
* Post: Add assert_transact! macro to unwrap safely.
I was finding it very difficult to track unwrapping errors while
making changes, due to an underlying Mac OS X symbolication issue that
makes running tests with RUST_BACKTRACE=1 so slow that they all time
out.
* Post: Don't expect or recognize :db.{install,alter}/attribute.
I had this all working... except we will never see a repeated
`[:db.part/db :db.alter/attribute :attribute]` assertion in the store!
That means my approach would let you alter an attribute at most one
time. It's not worth hacking around this; it's better to just stop
expecting (and recognizing) the marker flags. (We have all the data
to distinguish the various cases that we need without the marker
flags.)
This brings Mentat in line with the thrust of newer Datomic versions,
but isn't compatible with Datomic, because (if I understand correctly)
Datomic automatically adds :db.{install,alter}/attribute assertions to
transactions.
I haven't purged the corresponding :db/ident and schema fragments just
yet:
- we might want them back
- we might want them in order to upgrade v1 and v2 databases to the
new on-disk layout we're fleshing out (v3?).
* Post: Don't make :db/unique :db.unique/* imply :db/index true.
This patch avoids a potential bug with the "schema" materialized view.
If :db/unique :db.unique/value implies :db/index true, then what
happens when you _retract_ :db.unique/value? I think Datomic defines
this in some way, but I really want the "schema" materialized view to
be a slice of "datoms" and not have these sort of ambiguities and
persistent effects. Therefore, to ensure that we don't retract a
schema characteristic and accidentally change more than we intended
to, this patch stops having any schema characteristic imply any other
schema characteristic(s). To achieve that, I added an
Option<Unique::{Value,Identity}> type to Attribute; this helps with
this patch, and also looks ahead to when we allow to retract
:db/unique attributes.
* Post: Allow to retract :db/ident.
* Post: Include more details about invalid schema changes.
The tests use strings, so they hide the chained errors which do in
fact provide more detail.
* Review comment: Fix outdated comment.
* Review comment: s/_SET/_SQL_LIST/.
* Review comment: Use a sub-select for checking cardinality.
This might be faster in practice.
* Review comment: Put `attribute::Unique` into its own namespace.
2017-03-20 20:18:59 +00:00
// TODO: don't use :db/ident to test upserts!
2017-02-15 00:50:40 +00:00
#[ test ]
fn test_upsert_vector ( ) {
2017-03-20 18:29:17 +00:00
let mut conn = TestConn ::default ( ) ;
// Insert some :db.unique/identity elements.
2019-07-17 15:46:47 +00:00
assert_transact! (
conn ,
" [[:db/add 100 :db/ident :name/Ivan]
[ :db / add 101 :db / ident :name / Petr ] ] "
) ;
assert_matches! (
conn . last_transaction ( ) ,
" [[100 :db/ident :name/Ivan ?tx true]
2017-03-20 18:29:17 +00:00
[ 101 :db / ident :name / Petr ? tx true ]
2019-07-17 15:46:47 +00:00
[ ? tx :db / txInstant ? ms ? tx true ] ] "
) ;
assert_matches! (
conn . datoms ( ) ,
" [[100 :db/ident :name/Ivan]
[ 101 :db / ident :name / Petr ] ] "
) ;
2017-03-20 18:29:17 +00:00
// Upserting two tempids to the same entid works.
2019-07-17 15:46:47 +00:00
let report = assert_transact! (
conn ,
" [[:db/add \" t1 \" :db/ident :name/Ivan]
Schema alteration. Fixes #294 and #295. (#370) r=rnewman
* Pre: Don't retract :db/ident in test.
Datomic (and eventually Mentat) don't allow to retract :db/ident in
this way, so this runs afoul of future work to support mutating
metadata.
* Pre: s/VALUETYPE/VALUE_TYPE/.
This is consistent with the capitalization (which is "valueType") and
the other identifier.
* Pre: Remove some single quotes from error output.
* Part 1: Make materialized views be uniform [e a v value_type_tag].
This looks ahead to a time when we could support arbitrary
user-defined materialized views. For now, the "idents" materialized
view is those datoms of the form [e :db/ident :namespaced/keyword] and
the "schema" materialized view is those datoms of the form [e a v]
where a is in a particular set of attributes that will become clear in
the following commits.
This change is not backwards compatible, so I'm removing the open
current (really, v2) test. It'll be re-instated when we get to
https://github.com/mozilla/mentat/issues/194.
* Pre: Map TypedValue::Ref to TypedValue::Keyword in debug output.
* Part 3: Separate `schema_to_mutate` from the `schema` used to interpret.
This is just to keep track of the expected changes during
bootstrapping. I want bootstrap metadata mutations to flow through
the same code path as metadata mutations during regular transactions;
by differentiating the schema used for interpretation from the schema
that will be updated I expect to be able to apply bootstrap metadata
mutations to an empty schema and have things like materialized views
created (using the regular code paths).
This commit has been re-ordered for conceptual clarity, but it won't
compile because it references the metadata module. It's possible to
make it compile -- the functionality is there in the schema module --
but it's not worth the rebasing effort until after review (and
possibly not even then, since we'll squash down to a single commit to
land).
* Part 2: Maintain entids separately from idents.
In order to support historical idents, we need to distinguish the
"current" map from entid -> ident from the "complete historical" map
ident -> entid. This is what Datomic does; in Datomic, an ident is
never retracted (although it can be replaced). This approach is an
important part of allowing multiple consumers to share a schema
fragment as it migrates forward.
This fixes a limitation of the Clojure implementation, which did not
handle historical idents across knowledge base close and re-open.
The "entids" materialized view is naturally a slice of the "datoms"
table. The "idents" materialized view is a slice of the
"transactions" table. I hope that representing in this way, and
casting the problem in this light, might generalize to future
materialized views.
* Pre: Add DiffSet.
* Part 4: Collect mutations to a `Schema`.
I haven't taken your review comment about consuming AttributeBuilder
during each fluent function. If you read my response and still want
this, I'm happy to do it in review.
* Part 5: Handle :db/ident and :db.{install,alter}/attribute.
This "loops" the committed datoms out of the SQL store and back
through the metadata (schema, but in future also partition map)
processor. The metadata processor updates the schema and produces a
report of what changed; that report is then used to update the SQL
store. That update includes:
- the materialized views ("entids", "idents", and "schema");
- if needed, a subset of the datoms themselves (as flags change).
I've left a TODO for handling attribute retraction in the cases that
it makes sense. I expect that to be straight-forward.
* Review comment: Rename DiffSet to AddRetractAlterSet.
Also adds a little more commentary and a simple test.
* Review comment: Use ToIdent trait.
* Review comment: partially revert "Part 2: Maintain entids separately from idents."
This reverts commit 23a91df9c35e14398f2ddbd1ba25315821e67401.
Following our discussion, this removes the "entids" materialized
view. The next commit will remove historical idents from the "idents"
materialized view.
* Post: Use custom Either rather than std::result::Result.
This is not necessary, but it was suggested that we might be paying an
overhead creating Err instances while using error_chain. That seems
not to be the case, but this change shows that we don't actually use
any of the Result helper methods, so there's no reason to overload
Result. This change might avoid some future confusion, so I'm going
to land it anyway.
Signed-off-by: Nick Alexander <nalexander@mozilla.com>
* Review comment: Don't preserve historical idents.
* Review comment: More prepared statements when updating materialized views.
* Post: Test altering :db/cardinality and :db/unique.
These tests fail due to a Datomic limitation, namely that the marker
flag :db.alter/attribute can only be asserted once for an attribute!
That is, [:db.part/db :db.alter/attribute :attribute] will only be
transacted at most once. Since older versions of Datomic required the
:db.alter/attribute flag, I can only imagine they either never wrote
:db.alter/attribute to the store, or they handled it specially. I'll
need to remove the marker flag system from Mentat in order to address
this fundamental limitation.
* Post: Remove some more single quotes from error output.
* Post: Add assert_transact! macro to unwrap safely.
I was finding it very difficult to track unwrapping errors while
making changes, due to an underlying Mac OS X symbolication issue that
makes running tests with RUST_BACKTRACE=1 so slow that they all time
out.
* Post: Don't expect or recognize :db.{install,alter}/attribute.
I had this all working... except we will never see a repeated
`[:db.part/db :db.alter/attribute :attribute]` assertion in the store!
That means my approach would let you alter an attribute at most one
time. It's not worth hacking around this; it's better to just stop
expecting (and recognizing) the marker flags. (We have all the data
to distinguish the various cases that we need without the marker
flags.)
This brings Mentat in line with the thrust of newer Datomic versions,
but isn't compatible with Datomic, because (if I understand correctly)
Datomic automatically adds :db.{install,alter}/attribute assertions to
transactions.
I haven't purged the corresponding :db/ident and schema fragments just
yet:
- we might want them back
- we might want them in order to upgrade v1 and v2 databases to the
new on-disk layout we're fleshing out (v3?).
* Post: Don't make :db/unique :db.unique/* imply :db/index true.
This patch avoids a potential bug with the "schema" materialized view.
If :db/unique :db.unique/value implies :db/index true, then what
happens when you _retract_ :db.unique/value? I think Datomic defines
this in some way, but I really want the "schema" materialized view to
be a slice of "datoms" and not have these sort of ambiguities and
persistent effects. Therefore, to ensure that we don't retract a
schema characteristic and accidentally change more than we intended
to, this patch stops having any schema characteristic imply any other
schema characteristic(s). To achieve that, I added an
Option<Unique::{Value,Identity}> type to Attribute; this helps with
this patch, and also looks ahead to when we allow to retract
:db/unique attributes.
* Post: Allow to retract :db/ident.
* Post: Include more details about invalid schema changes.
The tests use strings, so they hide the chained errors which do in
fact provide more detail.
* Review comment: Fix outdated comment.
* Review comment: s/_SET/_SQL_LIST/.
* Review comment: Use a sub-select for checking cardinality.
This might be faster in practice.
* Review comment: Put `attribute::Unique` into its own namespace.
2017-03-20 20:18:59 +00:00
[ :db / add \ " t1 \" :db.schema/attribute 100]
[ :db / add \ " t2 \" :db/ident :name/Petr]
2019-07-17 15:46:47 +00:00
[ :db / add \ " t2 \" :db.schema/attribute 101]] "
) ;
assert_matches! (
conn . last_transaction ( ) ,
" [[100 :db.schema/attribute :name/Ivan ?tx true]
Schema alteration. Fixes #294 and #295. (#370) r=rnewman
* Pre: Don't retract :db/ident in test.
Datomic (and eventually Mentat) don't allow to retract :db/ident in
this way, so this runs afoul of future work to support mutating
metadata.
* Pre: s/VALUETYPE/VALUE_TYPE/.
This is consistent with the capitalization (which is "valueType") and
the other identifier.
* Pre: Remove some single quotes from error output.
* Part 1: Make materialized views be uniform [e a v value_type_tag].
This looks ahead to a time when we could support arbitrary
user-defined materialized views. For now, the "idents" materialized
view is those datoms of the form [e :db/ident :namespaced/keyword] and
the "schema" materialized view is those datoms of the form [e a v]
where a is in a particular set of attributes that will become clear in
the following commits.
This change is not backwards compatible, so I'm removing the open
current (really, v2) test. It'll be re-instated when we get to
https://github.com/mozilla/mentat/issues/194.
* Pre: Map TypedValue::Ref to TypedValue::Keyword in debug output.
* Part 3: Separate `schema_to_mutate` from the `schema` used to interpret.
This is just to keep track of the expected changes during
bootstrapping. I want bootstrap metadata mutations to flow through
the same code path as metadata mutations during regular transactions;
by differentiating the schema used for interpretation from the schema
that will be updated I expect to be able to apply bootstrap metadata
mutations to an empty schema and have things like materialized views
created (using the regular code paths).
This commit has been re-ordered for conceptual clarity, but it won't
compile because it references the metadata module. It's possible to
make it compile -- the functionality is there in the schema module --
but it's not worth the rebasing effort until after review (and
possibly not even then, since we'll squash down to a single commit to
land).
* Part 2: Maintain entids separately from idents.
In order to support historical idents, we need to distinguish the
"current" map from entid -> ident from the "complete historical" map
ident -> entid. This is what Datomic does; in Datomic, an ident is
never retracted (although it can be replaced). This approach is an
important part of allowing multiple consumers to share a schema
fragment as it migrates forward.
This fixes a limitation of the Clojure implementation, which did not
handle historical idents across knowledge base close and re-open.
The "entids" materialized view is naturally a slice of the "datoms"
table. The "idents" materialized view is a slice of the
"transactions" table. I hope that representing in this way, and
casting the problem in this light, might generalize to future
materialized views.
* Pre: Add DiffSet.
* Part 4: Collect mutations to a `Schema`.
I haven't taken your review comment about consuming AttributeBuilder
during each fluent function. If you read my response and still want
this, I'm happy to do it in review.
* Part 5: Handle :db/ident and :db.{install,alter}/attribute.
This "loops" the committed datoms out of the SQL store and back
through the metadata (schema, but in future also partition map)
processor. The metadata processor updates the schema and produces a
report of what changed; that report is then used to update the SQL
store. That update includes:
- the materialized views ("entids", "idents", and "schema");
- if needed, a subset of the datoms themselves (as flags change).
I've left a TODO for handling attribute retraction in the cases that
it makes sense. I expect that to be straight-forward.
* Review comment: Rename DiffSet to AddRetractAlterSet.
Also adds a little more commentary and a simple test.
* Review comment: Use ToIdent trait.
* Review comment: partially revert "Part 2: Maintain entids separately from idents."
This reverts commit 23a91df9c35e14398f2ddbd1ba25315821e67401.
Following our discussion, this removes the "entids" materialized
view. The next commit will remove historical idents from the "idents"
materialized view.
* Post: Use custom Either rather than std::result::Result.
This is not necessary, but it was suggested that we might be paying an
overhead creating Err instances while using error_chain. That seems
not to be the case, but this change shows that we don't actually use
any of the Result helper methods, so there's no reason to overload
Result. This change might avoid some future confusion, so I'm going
to land it anyway.
Signed-off-by: Nick Alexander <nalexander@mozilla.com>
* Review comment: Don't preserve historical idents.
* Review comment: More prepared statements when updating materialized views.
* Post: Test altering :db/cardinality and :db/unique.
These tests fail due to a Datomic limitation, namely that the marker
flag :db.alter/attribute can only be asserted once for an attribute!
That is, [:db.part/db :db.alter/attribute :attribute] will only be
transacted at most once. Since older versions of Datomic required the
:db.alter/attribute flag, I can only imagine they either never wrote
:db.alter/attribute to the store, or they handled it specially. I'll
need to remove the marker flag system from Mentat in order to address
this fundamental limitation.
* Post: Remove some more single quotes from error output.
* Post: Add assert_transact! macro to unwrap safely.
I was finding it very difficult to track unwrapping errors while
making changes, due to an underlying Mac OS X symbolication issue that
makes running tests with RUST_BACKTRACE=1 so slow that they all time
out.
* Post: Don't expect or recognize :db.{install,alter}/attribute.
I had this all working... except we will never see a repeated
`[:db.part/db :db.alter/attribute :attribute]` assertion in the store!
That means my approach would let you alter an attribute at most one
time. It's not worth hacking around this; it's better to just stop
expecting (and recognizing) the marker flags. (We have all the data
to distinguish the various cases that we need without the marker
flags.)
This brings Mentat in line with the thrust of newer Datomic versions,
but isn't compatible with Datomic, because (if I understand correctly)
Datomic automatically adds :db.{install,alter}/attribute assertions to
transactions.
I haven't purged the corresponding :db/ident and schema fragments just
yet:
- we might want them back
- we might want them in order to upgrade v1 and v2 databases to the
new on-disk layout we're fleshing out (v3?).
* Post: Don't make :db/unique :db.unique/* imply :db/index true.
This patch avoids a potential bug with the "schema" materialized view.
If :db/unique :db.unique/value implies :db/index true, then what
happens when you _retract_ :db.unique/value? I think Datomic defines
this in some way, but I really want the "schema" materialized view to
be a slice of "datoms" and not have these sort of ambiguities and
persistent effects. Therefore, to ensure that we don't retract a
schema characteristic and accidentally change more than we intended
to, this patch stops having any schema characteristic imply any other
schema characteristic(s). To achieve that, I added an
Option<Unique::{Value,Identity}> type to Attribute; this helps with
this patch, and also looks ahead to when we allow to retract
:db/unique attributes.
* Post: Allow to retract :db/ident.
* Post: Include more details about invalid schema changes.
The tests use strings, so they hide the chained errors which do in
fact provide more detail.
* Review comment: Fix outdated comment.
* Review comment: s/_SET/_SQL_LIST/.
* Review comment: Use a sub-select for checking cardinality.
This might be faster in practice.
* Review comment: Put `attribute::Unique` into its own namespace.
2017-03-20 20:18:59 +00:00
[ 101 :db . schema / attribute :name / Petr ? tx true ]
2019-07-17 15:46:47 +00:00
[ ? tx :db / txInstant ? ms ? tx true ] ] "
) ;
assert_matches! (
conn . datoms ( ) ,
" [[100 :db/ident :name/Ivan]
Schema alteration. Fixes #294 and #295. (#370) r=rnewman
* Pre: Don't retract :db/ident in test.
Datomic (and eventually Mentat) don't allow to retract :db/ident in
this way, so this runs afoul of future work to support mutating
metadata.
* Pre: s/VALUETYPE/VALUE_TYPE/.
This is consistent with the capitalization (which is "valueType") and
the other identifier.
* Pre: Remove some single quotes from error output.
* Part 1: Make materialized views be uniform [e a v value_type_tag].
This looks ahead to a time when we could support arbitrary
user-defined materialized views. For now, the "idents" materialized
view is those datoms of the form [e :db/ident :namespaced/keyword] and
the "schema" materialized view is those datoms of the form [e a v]
where a is in a particular set of attributes that will become clear in
the following commits.
This change is not backwards compatible, so I'm removing the open
current (really, v2) test. It'll be re-instated when we get to
https://github.com/mozilla/mentat/issues/194.
* Pre: Map TypedValue::Ref to TypedValue::Keyword in debug output.
* Part 3: Separate `schema_to_mutate` from the `schema` used to interpret.
This is just to keep track of the expected changes during
bootstrapping. I want bootstrap metadata mutations to flow through
the same code path as metadata mutations during regular transactions;
by differentiating the schema used for interpretation from the schema
that will be updated I expect to be able to apply bootstrap metadata
mutations to an empty schema and have things like materialized views
created (using the regular code paths).
This commit has been re-ordered for conceptual clarity, but it won't
compile because it references the metadata module. It's possible to
make it compile -- the functionality is there in the schema module --
but it's not worth the rebasing effort until after review (and
possibly not even then, since we'll squash down to a single commit to
land).
* Part 2: Maintain entids separately from idents.
In order to support historical idents, we need to distinguish the
"current" map from entid -> ident from the "complete historical" map
ident -> entid. This is what Datomic does; in Datomic, an ident is
never retracted (although it can be replaced). This approach is an
important part of allowing multiple consumers to share a schema
fragment as it migrates forward.
This fixes a limitation of the Clojure implementation, which did not
handle historical idents across knowledge base close and re-open.
The "entids" materialized view is naturally a slice of the "datoms"
table. The "idents" materialized view is a slice of the
"transactions" table. I hope that representing in this way, and
casting the problem in this light, might generalize to future
materialized views.
* Pre: Add DiffSet.
* Part 4: Collect mutations to a `Schema`.
I haven't taken your review comment about consuming AttributeBuilder
during each fluent function. If you read my response and still want
this, I'm happy to do it in review.
* Part 5: Handle :db/ident and :db.{install,alter}/attribute.
This "loops" the committed datoms out of the SQL store and back
through the metadata (schema, but in future also partition map)
processor. The metadata processor updates the schema and produces a
report of what changed; that report is then used to update the SQL
store. That update includes:
- the materialized views ("entids", "idents", and "schema");
- if needed, a subset of the datoms themselves (as flags change).
I've left a TODO for handling attribute retraction in the cases that
it makes sense. I expect that to be straight-forward.
* Review comment: Rename DiffSet to AddRetractAlterSet.
Also adds a little more commentary and a simple test.
* Review comment: Use ToIdent trait.
* Review comment: partially revert "Part 2: Maintain entids separately from idents."
This reverts commit 23a91df9c35e14398f2ddbd1ba25315821e67401.
Following our discussion, this removes the "entids" materialized
view. The next commit will remove historical idents from the "idents"
materialized view.
* Post: Use custom Either rather than std::result::Result.
This is not necessary, but it was suggested that we might be paying an
overhead creating Err instances while using error_chain. That seems
not to be the case, but this change shows that we don't actually use
any of the Result helper methods, so there's no reason to overload
Result. This change might avoid some future confusion, so I'm going
to land it anyway.
Signed-off-by: Nick Alexander <nalexander@mozilla.com>
* Review comment: Don't preserve historical idents.
* Review comment: More prepared statements when updating materialized views.
* Post: Test altering :db/cardinality and :db/unique.
These tests fail due to a Datomic limitation, namely that the marker
flag :db.alter/attribute can only be asserted once for an attribute!
That is, [:db.part/db :db.alter/attribute :attribute] will only be
transacted at most once. Since older versions of Datomic required the
:db.alter/attribute flag, I can only imagine they either never wrote
:db.alter/attribute to the store, or they handled it specially. I'll
need to remove the marker flag system from Mentat in order to address
this fundamental limitation.
* Post: Remove some more single quotes from error output.
* Post: Add assert_transact! macro to unwrap safely.
I was finding it very difficult to track unwrapping errors while
making changes, due to an underlying Mac OS X symbolication issue that
makes running tests with RUST_BACKTRACE=1 so slow that they all time
out.
* Post: Don't expect or recognize :db.{install,alter}/attribute.
I had this all working... except we will never see a repeated
`[:db.part/db :db.alter/attribute :attribute]` assertion in the store!
That means my approach would let you alter an attribute at most one
time. It's not worth hacking around this; it's better to just stop
expecting (and recognizing) the marker flags. (We have all the data
to distinguish the various cases that we need without the marker
flags.)
This brings Mentat in line with the thrust of newer Datomic versions,
but isn't compatible with Datomic, because (if I understand correctly)
Datomic automatically adds :db.{install,alter}/attribute assertions to
transactions.
I haven't purged the corresponding :db/ident and schema fragments just
yet:
- we might want them back
- we might want them in order to upgrade v1 and v2 databases to the
new on-disk layout we're fleshing out (v3?).
* Post: Don't make :db/unique :db.unique/* imply :db/index true.
This patch avoids a potential bug with the "schema" materialized view.
If :db/unique :db.unique/value implies :db/index true, then what
happens when you _retract_ :db.unique/value? I think Datomic defines
this in some way, but I really want the "schema" materialized view to
be a slice of "datoms" and not have these sort of ambiguities and
persistent effects. Therefore, to ensure that we don't retract a
schema characteristic and accidentally change more than we intended
to, this patch stops having any schema characteristic imply any other
schema characteristic(s). To achieve that, I added an
Option<Unique::{Value,Identity}> type to Attribute; this helps with
this patch, and also looks ahead to when we allow to retract
:db/unique attributes.
* Post: Allow to retract :db/ident.
* Post: Include more details about invalid schema changes.
The tests use strings, so they hide the chained errors which do in
fact provide more detail.
* Review comment: Fix outdated comment.
* Review comment: s/_SET/_SQL_LIST/.
* Review comment: Use a sub-select for checking cardinality.
This might be faster in practice.
* Review comment: Put `attribute::Unique` into its own namespace.
2017-03-20 20:18:59 +00:00
[ 100 :db . schema / attribute :name / Ivan ]
2017-03-20 18:29:17 +00:00
[ 101 :db / ident :name / Petr ]
2019-07-17 15:46:47 +00:00
[ 101 :db . schema / attribute :name / Petr ] ] "
) ;
assert_matches! (
tempids ( & report ) ,
" { \" t1 \" 100
\ " t2 \" 101} "
) ;
2017-03-20 18:29:17 +00:00
// Upserting a tempid works. The ref doesn't have to exist (at this time), but we can't
// reuse an existing ref due to :db/unique :db.unique/value.
2019-07-17 15:46:47 +00:00
let report = assert_transact! (
conn ,
" [[:db/add \" t1 \" :db/ident :name/Ivan]
[ :db / add \ " t1 \" :db.schema/attribute 102]] "
) ;
assert_matches! (
conn . last_transaction ( ) ,
" [[100 :db.schema/attribute 102 ?tx true]
[ ? true :db / txInstant ? ms ? tx true ] ] "
) ;
assert_matches! (
conn . datoms ( ) ,
" [[100 :db/ident :name/Ivan]
Schema alteration. Fixes #294 and #295. (#370) r=rnewman
* Pre: Don't retract :db/ident in test.
Datomic (and eventually Mentat) don't allow to retract :db/ident in
this way, so this runs afoul of future work to support mutating
metadata.
* Pre: s/VALUETYPE/VALUE_TYPE/.
This is consistent with the capitalization (which is "valueType") and
the other identifier.
* Pre: Remove some single quotes from error output.
* Part 1: Make materialized views be uniform [e a v value_type_tag].
This looks ahead to a time when we could support arbitrary
user-defined materialized views. For now, the "idents" materialized
view is those datoms of the form [e :db/ident :namespaced/keyword] and
the "schema" materialized view is those datoms of the form [e a v]
where a is in a particular set of attributes that will become clear in
the following commits.
This change is not backwards compatible, so I'm removing the open
current (really, v2) test. It'll be re-instated when we get to
https://github.com/mozilla/mentat/issues/194.
* Pre: Map TypedValue::Ref to TypedValue::Keyword in debug output.
* Part 3: Separate `schema_to_mutate` from the `schema` used to interpret.
This is just to keep track of the expected changes during
bootstrapping. I want bootstrap metadata mutations to flow through
the same code path as metadata mutations during regular transactions;
by differentiating the schema used for interpretation from the schema
that will be updated I expect to be able to apply bootstrap metadata
mutations to an empty schema and have things like materialized views
created (using the regular code paths).
This commit has been re-ordered for conceptual clarity, but it won't
compile because it references the metadata module. It's possible to
make it compile -- the functionality is there in the schema module --
but it's not worth the rebasing effort until after review (and
possibly not even then, since we'll squash down to a single commit to
land).
* Part 2: Maintain entids separately from idents.
In order to support historical idents, we need to distinguish the
"current" map from entid -> ident from the "complete historical" map
ident -> entid. This is what Datomic does; in Datomic, an ident is
never retracted (although it can be replaced). This approach is an
important part of allowing multiple consumers to share a schema
fragment as it migrates forward.
This fixes a limitation of the Clojure implementation, which did not
handle historical idents across knowledge base close and re-open.
The "entids" materialized view is naturally a slice of the "datoms"
table. The "idents" materialized view is a slice of the
"transactions" table. I hope that representing in this way, and
casting the problem in this light, might generalize to future
materialized views.
* Pre: Add DiffSet.
* Part 4: Collect mutations to a `Schema`.
I haven't taken your review comment about consuming AttributeBuilder
during each fluent function. If you read my response and still want
this, I'm happy to do it in review.
* Part 5: Handle :db/ident and :db.{install,alter}/attribute.
This "loops" the committed datoms out of the SQL store and back
through the metadata (schema, but in future also partition map)
processor. The metadata processor updates the schema and produces a
report of what changed; that report is then used to update the SQL
store. That update includes:
- the materialized views ("entids", "idents", and "schema");
- if needed, a subset of the datoms themselves (as flags change).
I've left a TODO for handling attribute retraction in the cases that
it makes sense. I expect that to be straight-forward.
* Review comment: Rename DiffSet to AddRetractAlterSet.
Also adds a little more commentary and a simple test.
* Review comment: Use ToIdent trait.
* Review comment: partially revert "Part 2: Maintain entids separately from idents."
This reverts commit 23a91df9c35e14398f2ddbd1ba25315821e67401.
Following our discussion, this removes the "entids" materialized
view. The next commit will remove historical idents from the "idents"
materialized view.
* Post: Use custom Either rather than std::result::Result.
This is not necessary, but it was suggested that we might be paying an
overhead creating Err instances while using error_chain. That seems
not to be the case, but this change shows that we don't actually use
any of the Result helper methods, so there's no reason to overload
Result. This change might avoid some future confusion, so I'm going
to land it anyway.
Signed-off-by: Nick Alexander <nalexander@mozilla.com>
* Review comment: Don't preserve historical idents.
* Review comment: More prepared statements when updating materialized views.
* Post: Test altering :db/cardinality and :db/unique.
These tests fail due to a Datomic limitation, namely that the marker
flag :db.alter/attribute can only be asserted once for an attribute!
That is, [:db.part/db :db.alter/attribute :attribute] will only be
transacted at most once. Since older versions of Datomic required the
:db.alter/attribute flag, I can only imagine they either never wrote
:db.alter/attribute to the store, or they handled it specially. I'll
need to remove the marker flag system from Mentat in order to address
this fundamental limitation.
* Post: Remove some more single quotes from error output.
* Post: Add assert_transact! macro to unwrap safely.
I was finding it very difficult to track unwrapping errors while
making changes, due to an underlying Mac OS X symbolication issue that
makes running tests with RUST_BACKTRACE=1 so slow that they all time
out.
* Post: Don't expect or recognize :db.{install,alter}/attribute.
I had this all working... except we will never see a repeated
`[:db.part/db :db.alter/attribute :attribute]` assertion in the store!
That means my approach would let you alter an attribute at most one
time. It's not worth hacking around this; it's better to just stop
expecting (and recognizing) the marker flags. (We have all the data
to distinguish the various cases that we need without the marker
flags.)
This brings Mentat in line with the thrust of newer Datomic versions,
but isn't compatible with Datomic, because (if I understand correctly)
Datomic automatically adds :db.{install,alter}/attribute assertions to
transactions.
I haven't purged the corresponding :db/ident and schema fragments just
yet:
- we might want them back
- we might want them in order to upgrade v1 and v2 databases to the
new on-disk layout we're fleshing out (v3?).
* Post: Don't make :db/unique :db.unique/* imply :db/index true.
This patch avoids a potential bug with the "schema" materialized view.
If :db/unique :db.unique/value implies :db/index true, then what
happens when you _retract_ :db.unique/value? I think Datomic defines
this in some way, but I really want the "schema" materialized view to
be a slice of "datoms" and not have these sort of ambiguities and
persistent effects. Therefore, to ensure that we don't retract a
schema characteristic and accidentally change more than we intended
to, this patch stops having any schema characteristic imply any other
schema characteristic(s). To achieve that, I added an
Option<Unique::{Value,Identity}> type to Attribute; this helps with
this patch, and also looks ahead to when we allow to retract
:db/unique attributes.
* Post: Allow to retract :db/ident.
* Post: Include more details about invalid schema changes.
The tests use strings, so they hide the chained errors which do in
fact provide more detail.
* Review comment: Fix outdated comment.
* Review comment: s/_SET/_SQL_LIST/.
* Review comment: Use a sub-select for checking cardinality.
This might be faster in practice.
* Review comment: Put `attribute::Unique` into its own namespace.
2017-03-20 20:18:59 +00:00
[ 100 :db . schema / attribute :name / Ivan ]
2017-03-20 18:29:17 +00:00
[ 100 :db . schema / attribute 102 ]
[ 101 :db / ident :name / Petr ]
2019-07-17 15:46:47 +00:00
[ 101 :db . schema / attribute :name / Petr ] ] "
) ;
assert_matches! ( tempids ( & report ) , " { \" t1 \" 100} " ) ;
2017-03-20 18:34:38 +00:00
2017-03-20 18:29:17 +00:00
// A single complex upsert allocates a new entid.
Schema alteration. Fixes #294 and #295. (#370) r=rnewman
* Pre: Don't retract :db/ident in test.
Datomic (and eventually Mentat) don't allow to retract :db/ident in
this way, so this runs afoul of future work to support mutating
metadata.
* Pre: s/VALUETYPE/VALUE_TYPE/.
This is consistent with the capitalization (which is "valueType") and
the other identifier.
* Pre: Remove some single quotes from error output.
* Part 1: Make materialized views be uniform [e a v value_type_tag].
This looks ahead to a time when we could support arbitrary
user-defined materialized views. For now, the "idents" materialized
view is those datoms of the form [e :db/ident :namespaced/keyword] and
the "schema" materialized view is those datoms of the form [e a v]
where a is in a particular set of attributes that will become clear in
the following commits.
This change is not backwards compatible, so I'm removing the open
current (really, v2) test. It'll be re-instated when we get to
https://github.com/mozilla/mentat/issues/194.
* Pre: Map TypedValue::Ref to TypedValue::Keyword in debug output.
* Part 3: Separate `schema_to_mutate` from the `schema` used to interpret.
This is just to keep track of the expected changes during
bootstrapping. I want bootstrap metadata mutations to flow through
the same code path as metadata mutations during regular transactions;
by differentiating the schema used for interpretation from the schema
that will be updated I expect to be able to apply bootstrap metadata
mutations to an empty schema and have things like materialized views
created (using the regular code paths).
This commit has been re-ordered for conceptual clarity, but it won't
compile because it references the metadata module. It's possible to
make it compile -- the functionality is there in the schema module --
but it's not worth the rebasing effort until after review (and
possibly not even then, since we'll squash down to a single commit to
land).
* Part 2: Maintain entids separately from idents.
In order to support historical idents, we need to distinguish the
"current" map from entid -> ident from the "complete historical" map
ident -> entid. This is what Datomic does; in Datomic, an ident is
never retracted (although it can be replaced). This approach is an
important part of allowing multiple consumers to share a schema
fragment as it migrates forward.
This fixes a limitation of the Clojure implementation, which did not
handle historical idents across knowledge base close and re-open.
The "entids" materialized view is naturally a slice of the "datoms"
table. The "idents" materialized view is a slice of the
"transactions" table. I hope that representing in this way, and
casting the problem in this light, might generalize to future
materialized views.
* Pre: Add DiffSet.
* Part 4: Collect mutations to a `Schema`.
I haven't taken your review comment about consuming AttributeBuilder
during each fluent function. If you read my response and still want
this, I'm happy to do it in review.
* Part 5: Handle :db/ident and :db.{install,alter}/attribute.
This "loops" the committed datoms out of the SQL store and back
through the metadata (schema, but in future also partition map)
processor. The metadata processor updates the schema and produces a
report of what changed; that report is then used to update the SQL
store. That update includes:
- the materialized views ("entids", "idents", and "schema");
- if needed, a subset of the datoms themselves (as flags change).
I've left a TODO for handling attribute retraction in the cases that
it makes sense. I expect that to be straight-forward.
* Review comment: Rename DiffSet to AddRetractAlterSet.
Also adds a little more commentary and a simple test.
* Review comment: Use ToIdent trait.
* Review comment: partially revert "Part 2: Maintain entids separately from idents."
This reverts commit 23a91df9c35e14398f2ddbd1ba25315821e67401.
Following our discussion, this removes the "entids" materialized
view. The next commit will remove historical idents from the "idents"
materialized view.
* Post: Use custom Either rather than std::result::Result.
This is not necessary, but it was suggested that we might be paying an
overhead creating Err instances while using error_chain. That seems
not to be the case, but this change shows that we don't actually use
any of the Result helper methods, so there's no reason to overload
Result. This change might avoid some future confusion, so I'm going
to land it anyway.
Signed-off-by: Nick Alexander <nalexander@mozilla.com>
* Review comment: Don't preserve historical idents.
* Review comment: More prepared statements when updating materialized views.
* Post: Test altering :db/cardinality and :db/unique.
These tests fail due to a Datomic limitation, namely that the marker
flag :db.alter/attribute can only be asserted once for an attribute!
That is, [:db.part/db :db.alter/attribute :attribute] will only be
transacted at most once. Since older versions of Datomic required the
:db.alter/attribute flag, I can only imagine they either never wrote
:db.alter/attribute to the store, or they handled it specially. I'll
need to remove the marker flag system from Mentat in order to address
this fundamental limitation.
* Post: Remove some more single quotes from error output.
* Post: Add assert_transact! macro to unwrap safely.
I was finding it very difficult to track unwrapping errors while
making changes, due to an underlying Mac OS X symbolication issue that
makes running tests with RUST_BACKTRACE=1 so slow that they all time
out.
* Post: Don't expect or recognize :db.{install,alter}/attribute.
I had this all working... except we will never see a repeated
`[:db.part/db :db.alter/attribute :attribute]` assertion in the store!
That means my approach would let you alter an attribute at most one
time. It's not worth hacking around this; it's better to just stop
expecting (and recognizing) the marker flags. (We have all the data
to distinguish the various cases that we need without the marker
flags.)
This brings Mentat in line with the thrust of newer Datomic versions,
but isn't compatible with Datomic, because (if I understand correctly)
Datomic automatically adds :db.{install,alter}/attribute assertions to
transactions.
I haven't purged the corresponding :db/ident and schema fragments just
yet:
- we might want them back
- we might want them in order to upgrade v1 and v2 databases to the
new on-disk layout we're fleshing out (v3?).
* Post: Don't make :db/unique :db.unique/* imply :db/index true.
This patch avoids a potential bug with the "schema" materialized view.
If :db/unique :db.unique/value implies :db/index true, then what
happens when you _retract_ :db.unique/value? I think Datomic defines
this in some way, but I really want the "schema" materialized view to
be a slice of "datoms" and not have these sort of ambiguities and
persistent effects. Therefore, to ensure that we don't retract a
schema characteristic and accidentally change more than we intended
to, this patch stops having any schema characteristic imply any other
schema characteristic(s). To achieve that, I added an
Option<Unique::{Value,Identity}> type to Attribute; this helps with
this patch, and also looks ahead to when we allow to retract
:db/unique attributes.
* Post: Allow to retract :db/ident.
* Post: Include more details about invalid schema changes.
The tests use strings, so they hide the chained errors which do in
fact provide more detail.
* Review comment: Fix outdated comment.
* Review comment: s/_SET/_SQL_LIST/.
* Review comment: Use a sub-select for checking cardinality.
This might be faster in practice.
* Review comment: Put `attribute::Unique` into its own namespace.
2017-03-20 20:18:59 +00:00
let report = assert_transact! ( conn , " [[:db/add \" t1 \" :db.schema/attribute \" t2 \" ]] " ) ;
2019-07-17 15:46:47 +00:00
assert_matches! (
conn . last_transaction ( ) ,
" [[65536 :db.schema/attribute 65537 ?tx true]
[ ? tx :db / txInstant ? ms ? tx true ] ] "
) ;
assert_matches! (
tempids ( & report ) ,
" { \" t1 \" 65536
\ " t2 \" 65537} "
) ;
2017-03-20 18:29:17 +00:00
// Conflicting upserts fail.
Schema alteration. Fixes #294 and #295. (#370) r=rnewman
* Pre: Don't retract :db/ident in test.
Datomic (and eventually Mentat) don't allow to retract :db/ident in
this way, so this runs afoul of future work to support mutating
metadata.
* Pre: s/VALUETYPE/VALUE_TYPE/.
This is consistent with the capitalization (which is "valueType") and
the other identifier.
* Pre: Remove some single quotes from error output.
* Part 1: Make materialized views be uniform [e a v value_type_tag].
This looks ahead to a time when we could support arbitrary
user-defined materialized views. For now, the "idents" materialized
view is those datoms of the form [e :db/ident :namespaced/keyword] and
the "schema" materialized view is those datoms of the form [e a v]
where a is in a particular set of attributes that will become clear in
the following commits.
This change is not backwards compatible, so I'm removing the open
current (really, v2) test. It'll be re-instated when we get to
https://github.com/mozilla/mentat/issues/194.
* Pre: Map TypedValue::Ref to TypedValue::Keyword in debug output.
* Part 3: Separate `schema_to_mutate` from the `schema` used to interpret.
This is just to keep track of the expected changes during
bootstrapping. I want bootstrap metadata mutations to flow through
the same code path as metadata mutations during regular transactions;
by differentiating the schema used for interpretation from the schema
that will be updated I expect to be able to apply bootstrap metadata
mutations to an empty schema and have things like materialized views
created (using the regular code paths).
This commit has been re-ordered for conceptual clarity, but it won't
compile because it references the metadata module. It's possible to
make it compile -- the functionality is there in the schema module --
but it's not worth the rebasing effort until after review (and
possibly not even then, since we'll squash down to a single commit to
land).
* Part 2: Maintain entids separately from idents.
In order to support historical idents, we need to distinguish the
"current" map from entid -> ident from the "complete historical" map
ident -> entid. This is what Datomic does; in Datomic, an ident is
never retracted (although it can be replaced). This approach is an
important part of allowing multiple consumers to share a schema
fragment as it migrates forward.
This fixes a limitation of the Clojure implementation, which did not
handle historical idents across knowledge base close and re-open.
The "entids" materialized view is naturally a slice of the "datoms"
table. The "idents" materialized view is a slice of the
"transactions" table. I hope that representing in this way, and
casting the problem in this light, might generalize to future
materialized views.
* Pre: Add DiffSet.
* Part 4: Collect mutations to a `Schema`.
I haven't taken your review comment about consuming AttributeBuilder
during each fluent function. If you read my response and still want
this, I'm happy to do it in review.
* Part 5: Handle :db/ident and :db.{install,alter}/attribute.
This "loops" the committed datoms out of the SQL store and back
through the metadata (schema, but in future also partition map)
processor. The metadata processor updates the schema and produces a
report of what changed; that report is then used to update the SQL
store. That update includes:
- the materialized views ("entids", "idents", and "schema");
- if needed, a subset of the datoms themselves (as flags change).
I've left a TODO for handling attribute retraction in the cases that
it makes sense. I expect that to be straight-forward.
* Review comment: Rename DiffSet to AddRetractAlterSet.
Also adds a little more commentary and a simple test.
* Review comment: Use ToIdent trait.
* Review comment: partially revert "Part 2: Maintain entids separately from idents."
This reverts commit 23a91df9c35e14398f2ddbd1ba25315821e67401.
Following our discussion, this removes the "entids" materialized
view. The next commit will remove historical idents from the "idents"
materialized view.
* Post: Use custom Either rather than std::result::Result.
This is not necessary, but it was suggested that we might be paying an
overhead creating Err instances while using error_chain. That seems
not to be the case, but this change shows that we don't actually use
any of the Result helper methods, so there's no reason to overload
Result. This change might avoid some future confusion, so I'm going
to land it anyway.
Signed-off-by: Nick Alexander <nalexander@mozilla.com>
* Review comment: Don't preserve historical idents.
* Review comment: More prepared statements when updating materialized views.
* Post: Test altering :db/cardinality and :db/unique.
These tests fail due to a Datomic limitation, namely that the marker
flag :db.alter/attribute can only be asserted once for an attribute!
That is, [:db.part/db :db.alter/attribute :attribute] will only be
transacted at most once. Since older versions of Datomic required the
:db.alter/attribute flag, I can only imagine they either never wrote
:db.alter/attribute to the store, or they handled it specially. I'll
need to remove the marker flag system from Mentat in order to address
this fundamental limitation.
* Post: Remove some more single quotes from error output.
* Post: Add assert_transact! macro to unwrap safely.
I was finding it very difficult to track unwrapping errors while
making changes, due to an underlying Mac OS X symbolication issue that
makes running tests with RUST_BACKTRACE=1 so slow that they all time
out.
* Post: Don't expect or recognize :db.{install,alter}/attribute.
I had this all working... except we will never see a repeated
`[:db.part/db :db.alter/attribute :attribute]` assertion in the store!
That means my approach would let you alter an attribute at most one
time. It's not worth hacking around this; it's better to just stop
expecting (and recognizing) the marker flags. (We have all the data
to distinguish the various cases that we need without the marker
flags.)
This brings Mentat in line with the thrust of newer Datomic versions,
but isn't compatible with Datomic, because (if I understand correctly)
Datomic automatically adds :db.{install,alter}/attribute assertions to
transactions.
I haven't purged the corresponding :db/ident and schema fragments just
yet:
- we might want them back
- we might want them in order to upgrade v1 and v2 databases to the
new on-disk layout we're fleshing out (v3?).
* Post: Don't make :db/unique :db.unique/* imply :db/index true.
This patch avoids a potential bug with the "schema" materialized view.
If :db/unique :db.unique/value implies :db/index true, then what
happens when you _retract_ :db.unique/value? I think Datomic defines
this in some way, but I really want the "schema" materialized view to
be a slice of "datoms" and not have these sort of ambiguities and
persistent effects. Therefore, to ensure that we don't retract a
schema characteristic and accidentally change more than we intended
to, this patch stops having any schema characteristic imply any other
schema characteristic(s). To achieve that, I added an
Option<Unique::{Value,Identity}> type to Attribute; this helps with
this patch, and also looks ahead to when we allow to retract
:db/unique attributes.
* Post: Allow to retract :db/ident.
* Post: Include more details about invalid schema changes.
The tests use strings, so they hide the chained errors which do in
fact provide more detail.
* Review comment: Fix outdated comment.
* Review comment: s/_SET/_SQL_LIST/.
* Review comment: Use a sub-select for checking cardinality.
This might be faster in practice.
* Review comment: Put `attribute::Unique` into its own namespace.
2017-03-20 20:18:59 +00:00
assert_transact! ( conn , " [[:db/add \" t1 \" :db/ident :name/Ivan]
2018-05-01 20:47:33 +00:00
[ :db / add \ " t1 \" :db/ident :name/Petr]] " ,
Err ( " schema constraint violation: conflicting upserts: \n tempid External( \" t1 \" ) upserts to {KnownEntid(100), KnownEntid(101)} \n " ) ) ;
// The error messages of conflicting upserts gives information about all failing upserts (in a particular generation).
assert_transact! ( conn , " [[:db/add \" t2 \" :db/ident :name/Grigory]
[ :db / add \ " t2 \" :db/ident :name/Petr]
[ :db / add \ " t2 \" :db/ident :name/Ivan]
[ :db / add \ " t1 \" :db/ident :name/Ivan]
[ :db / add \ " t1 \" :db/ident :name/Petr]] " ,
Err ( " schema constraint violation: conflicting upserts: \n tempid External( \" t1 \" ) upserts to {KnownEntid(100), KnownEntid(101)} \n tempid External( \" t2 \" ) upserts to {KnownEntid(100), KnownEntid(101)} \n " ) ) ;
2017-03-20 18:29:17 +00:00
// tempids in :db/retract that don't upsert fail.
Schema alteration. Fixes #294 and #295. (#370) r=rnewman
* Pre: Don't retract :db/ident in test.
Datomic (and eventually Mentat) don't allow to retract :db/ident in
this way, so this runs afoul of future work to support mutating
metadata.
* Pre: s/VALUETYPE/VALUE_TYPE/.
This is consistent with the capitalization (which is "valueType") and
the other identifier.
* Pre: Remove some single quotes from error output.
* Part 1: Make materialized views be uniform [e a v value_type_tag].
This looks ahead to a time when we could support arbitrary
user-defined materialized views. For now, the "idents" materialized
view is those datoms of the form [e :db/ident :namespaced/keyword] and
the "schema" materialized view is those datoms of the form [e a v]
where a is in a particular set of attributes that will become clear in
the following commits.
This change is not backwards compatible, so I'm removing the open
current (really, v2) test. It'll be re-instated when we get to
https://github.com/mozilla/mentat/issues/194.
* Pre: Map TypedValue::Ref to TypedValue::Keyword in debug output.
* Part 3: Separate `schema_to_mutate` from the `schema` used to interpret.
This is just to keep track of the expected changes during
bootstrapping. I want bootstrap metadata mutations to flow through
the same code path as metadata mutations during regular transactions;
by differentiating the schema used for interpretation from the schema
that will be updated I expect to be able to apply bootstrap metadata
mutations to an empty schema and have things like materialized views
created (using the regular code paths).
This commit has been re-ordered for conceptual clarity, but it won't
compile because it references the metadata module. It's possible to
make it compile -- the functionality is there in the schema module --
but it's not worth the rebasing effort until after review (and
possibly not even then, since we'll squash down to a single commit to
land).
* Part 2: Maintain entids separately from idents.
In order to support historical idents, we need to distinguish the
"current" map from entid -> ident from the "complete historical" map
ident -> entid. This is what Datomic does; in Datomic, an ident is
never retracted (although it can be replaced). This approach is an
important part of allowing multiple consumers to share a schema
fragment as it migrates forward.
This fixes a limitation of the Clojure implementation, which did not
handle historical idents across knowledge base close and re-open.
The "entids" materialized view is naturally a slice of the "datoms"
table. The "idents" materialized view is a slice of the
"transactions" table. I hope that representing in this way, and
casting the problem in this light, might generalize to future
materialized views.
* Pre: Add DiffSet.
* Part 4: Collect mutations to a `Schema`.
I haven't taken your review comment about consuming AttributeBuilder
during each fluent function. If you read my response and still want
this, I'm happy to do it in review.
* Part 5: Handle :db/ident and :db.{install,alter}/attribute.
This "loops" the committed datoms out of the SQL store and back
through the metadata (schema, but in future also partition map)
processor. The metadata processor updates the schema and produces a
report of what changed; that report is then used to update the SQL
store. That update includes:
- the materialized views ("entids", "idents", and "schema");
- if needed, a subset of the datoms themselves (as flags change).
I've left a TODO for handling attribute retraction in the cases that
it makes sense. I expect that to be straight-forward.
* Review comment: Rename DiffSet to AddRetractAlterSet.
Also adds a little more commentary and a simple test.
* Review comment: Use ToIdent trait.
* Review comment: partially revert "Part 2: Maintain entids separately from idents."
This reverts commit 23a91df9c35e14398f2ddbd1ba25315821e67401.
Following our discussion, this removes the "entids" materialized
view. The next commit will remove historical idents from the "idents"
materialized view.
* Post: Use custom Either rather than std::result::Result.
This is not necessary, but it was suggested that we might be paying an
overhead creating Err instances while using error_chain. That seems
not to be the case, but this change shows that we don't actually use
any of the Result helper methods, so there's no reason to overload
Result. This change might avoid some future confusion, so I'm going
to land it anyway.
Signed-off-by: Nick Alexander <nalexander@mozilla.com>
* Review comment: Don't preserve historical idents.
* Review comment: More prepared statements when updating materialized views.
* Post: Test altering :db/cardinality and :db/unique.
These tests fail due to a Datomic limitation, namely that the marker
flag :db.alter/attribute can only be asserted once for an attribute!
That is, [:db.part/db :db.alter/attribute :attribute] will only be
transacted at most once. Since older versions of Datomic required the
:db.alter/attribute flag, I can only imagine they either never wrote
:db.alter/attribute to the store, or they handled it specially. I'll
need to remove the marker flag system from Mentat in order to address
this fundamental limitation.
* Post: Remove some more single quotes from error output.
* Post: Add assert_transact! macro to unwrap safely.
I was finding it very difficult to track unwrapping errors while
making changes, due to an underlying Mac OS X symbolication issue that
makes running tests with RUST_BACKTRACE=1 so slow that they all time
out.
* Post: Don't expect or recognize :db.{install,alter}/attribute.
I had this all working... except we will never see a repeated
`[:db.part/db :db.alter/attribute :attribute]` assertion in the store!
That means my approach would let you alter an attribute at most one
time. It's not worth hacking around this; it's better to just stop
expecting (and recognizing) the marker flags. (We have all the data
to distinguish the various cases that we need without the marker
flags.)
This brings Mentat in line with the thrust of newer Datomic versions,
but isn't compatible with Datomic, because (if I understand correctly)
Datomic automatically adds :db.{install,alter}/attribute assertions to
transactions.
I haven't purged the corresponding :db/ident and schema fragments just
yet:
- we might want them back
- we might want them in order to upgrade v1 and v2 databases to the
new on-disk layout we're fleshing out (v3?).
* Post: Don't make :db/unique :db.unique/* imply :db/index true.
This patch avoids a potential bug with the "schema" materialized view.
If :db/unique :db.unique/value implies :db/index true, then what
happens when you _retract_ :db.unique/value? I think Datomic defines
this in some way, but I really want the "schema" materialized view to
be a slice of "datoms" and not have these sort of ambiguities and
persistent effects. Therefore, to ensure that we don't retract a
schema characteristic and accidentally change more than we intended
to, this patch stops having any schema characteristic imply any other
schema characteristic(s). To achieve that, I added an
Option<Unique::{Value,Identity}> type to Attribute; this helps with
this patch, and also looks ahead to when we allow to retract
:db/unique attributes.
* Post: Allow to retract :db/ident.
* Post: Include more details about invalid schema changes.
The tests use strings, so they hide the chained errors which do in
fact provide more detail.
* Review comment: Fix outdated comment.
* Review comment: s/_SET/_SQL_LIST/.
* Review comment: Use a sub-select for checking cardinality.
This might be faster in practice.
* Review comment: Put `attribute::Unique` into its own namespace.
2017-03-20 20:18:59 +00:00
assert_transact! ( conn , " [[:db/retract \" t1 \" :db/ident :name/Anonymous]] " ,
Err ( " not yet implemented: [:db/retract ...] entity referenced tempid that did not upsert: t1 " ) ) ;
2017-03-20 18:29:17 +00:00
// tempids in :db/retract that do upsert are retracted. The ref given doesn't exist, so the
// assertion will be ignored.
2019-07-17 15:46:47 +00:00
let report = assert_transact! (
conn ,
" [[:db/add \" t1 \" :db/ident :name/Ivan]
[ :db / retract \ " t1 \" :db.schema/attribute 103]] "
) ;
assert_matches! (
conn . last_transaction ( ) ,
" [[?tx :db/txInstant ?ms ?tx true]] "
) ;
assert_matches! ( tempids ( & report ) , " { \" t1 \" 100} " ) ;
2017-03-20 18:29:17 +00:00
// A multistep upsert. The upsert algorithm will first try to resolve "t1", fail, and then
// allocate both "t1" and "t2".
2019-07-17 15:46:47 +00:00
let report = assert_transact! (
conn ,
" [[:db/add \" t1 \" :db/ident :name/Josef]
[ :db / add \ " t2 \" :db.schema/attribute \" t1 \" ]] "
) ;
assert_matches! (
conn . last_transaction ( ) ,
" [[65538 :db/ident :name/Josef ?tx true]
Schema alteration. Fixes #294 and #295. (#370) r=rnewman
* Pre: Don't retract :db/ident in test.
Datomic (and eventually Mentat) don't allow to retract :db/ident in
this way, so this runs afoul of future work to support mutating
metadata.
* Pre: s/VALUETYPE/VALUE_TYPE/.
This is consistent with the capitalization (which is "valueType") and
the other identifier.
* Pre: Remove some single quotes from error output.
* Part 1: Make materialized views be uniform [e a v value_type_tag].
This looks ahead to a time when we could support arbitrary
user-defined materialized views. For now, the "idents" materialized
view is those datoms of the form [e :db/ident :namespaced/keyword] and
the "schema" materialized view is those datoms of the form [e a v]
where a is in a particular set of attributes that will become clear in
the following commits.
This change is not backwards compatible, so I'm removing the open
current (really, v2) test. It'll be re-instated when we get to
https://github.com/mozilla/mentat/issues/194.
* Pre: Map TypedValue::Ref to TypedValue::Keyword in debug output.
* Part 3: Separate `schema_to_mutate` from the `schema` used to interpret.
This is just to keep track of the expected changes during
bootstrapping. I want bootstrap metadata mutations to flow through
the same code path as metadata mutations during regular transactions;
by differentiating the schema used for interpretation from the schema
that will be updated I expect to be able to apply bootstrap metadata
mutations to an empty schema and have things like materialized views
created (using the regular code paths).
This commit has been re-ordered for conceptual clarity, but it won't
compile because it references the metadata module. It's possible to
make it compile -- the functionality is there in the schema module --
but it's not worth the rebasing effort until after review (and
possibly not even then, since we'll squash down to a single commit to
land).
* Part 2: Maintain entids separately from idents.
In order to support historical idents, we need to distinguish the
"current" map from entid -> ident from the "complete historical" map
ident -> entid. This is what Datomic does; in Datomic, an ident is
never retracted (although it can be replaced). This approach is an
important part of allowing multiple consumers to share a schema
fragment as it migrates forward.
This fixes a limitation of the Clojure implementation, which did not
handle historical idents across knowledge base close and re-open.
The "entids" materialized view is naturally a slice of the "datoms"
table. The "idents" materialized view is a slice of the
"transactions" table. I hope that representing in this way, and
casting the problem in this light, might generalize to future
materialized views.
* Pre: Add DiffSet.
* Part 4: Collect mutations to a `Schema`.
I haven't taken your review comment about consuming AttributeBuilder
during each fluent function. If you read my response and still want
this, I'm happy to do it in review.
* Part 5: Handle :db/ident and :db.{install,alter}/attribute.
This "loops" the committed datoms out of the SQL store and back
through the metadata (schema, but in future also partition map)
processor. The metadata processor updates the schema and produces a
report of what changed; that report is then used to update the SQL
store. That update includes:
- the materialized views ("entids", "idents", and "schema");
- if needed, a subset of the datoms themselves (as flags change).
I've left a TODO for handling attribute retraction in the cases that
it makes sense. I expect that to be straight-forward.
* Review comment: Rename DiffSet to AddRetractAlterSet.
Also adds a little more commentary and a simple test.
* Review comment: Use ToIdent trait.
* Review comment: partially revert "Part 2: Maintain entids separately from idents."
This reverts commit 23a91df9c35e14398f2ddbd1ba25315821e67401.
Following our discussion, this removes the "entids" materialized
view. The next commit will remove historical idents from the "idents"
materialized view.
* Post: Use custom Either rather than std::result::Result.
This is not necessary, but it was suggested that we might be paying an
overhead creating Err instances while using error_chain. That seems
not to be the case, but this change shows that we don't actually use
any of the Result helper methods, so there's no reason to overload
Result. This change might avoid some future confusion, so I'm going
to land it anyway.
Signed-off-by: Nick Alexander <nalexander@mozilla.com>
* Review comment: Don't preserve historical idents.
* Review comment: More prepared statements when updating materialized views.
* Post: Test altering :db/cardinality and :db/unique.
These tests fail due to a Datomic limitation, namely that the marker
flag :db.alter/attribute can only be asserted once for an attribute!
That is, [:db.part/db :db.alter/attribute :attribute] will only be
transacted at most once. Since older versions of Datomic required the
:db.alter/attribute flag, I can only imagine they either never wrote
:db.alter/attribute to the store, or they handled it specially. I'll
need to remove the marker flag system from Mentat in order to address
this fundamental limitation.
* Post: Remove some more single quotes from error output.
* Post: Add assert_transact! macro to unwrap safely.
I was finding it very difficult to track unwrapping errors while
making changes, due to an underlying Mac OS X symbolication issue that
makes running tests with RUST_BACKTRACE=1 so slow that they all time
out.
* Post: Don't expect or recognize :db.{install,alter}/attribute.
I had this all working... except we will never see a repeated
`[:db.part/db :db.alter/attribute :attribute]` assertion in the store!
That means my approach would let you alter an attribute at most one
time. It's not worth hacking around this; it's better to just stop
expecting (and recognizing) the marker flags. (We have all the data
to distinguish the various cases that we need without the marker
flags.)
This brings Mentat in line with the thrust of newer Datomic versions,
but isn't compatible with Datomic, because (if I understand correctly)
Datomic automatically adds :db.{install,alter}/attribute assertions to
transactions.
I haven't purged the corresponding :db/ident and schema fragments just
yet:
- we might want them back
- we might want them in order to upgrade v1 and v2 databases to the
new on-disk layout we're fleshing out (v3?).
* Post: Don't make :db/unique :db.unique/* imply :db/index true.
This patch avoids a potential bug with the "schema" materialized view.
If :db/unique :db.unique/value implies :db/index true, then what
happens when you _retract_ :db.unique/value? I think Datomic defines
this in some way, but I really want the "schema" materialized view to
be a slice of "datoms" and not have these sort of ambiguities and
persistent effects. Therefore, to ensure that we don't retract a
schema characteristic and accidentally change more than we intended
to, this patch stops having any schema characteristic imply any other
schema characteristic(s). To achieve that, I added an
Option<Unique::{Value,Identity}> type to Attribute; this helps with
this patch, and also looks ahead to when we allow to retract
:db/unique attributes.
* Post: Allow to retract :db/ident.
* Post: Include more details about invalid schema changes.
The tests use strings, so they hide the chained errors which do in
fact provide more detail.
* Review comment: Fix outdated comment.
* Review comment: s/_SET/_SQL_LIST/.
* Review comment: Use a sub-select for checking cardinality.
This might be faster in practice.
* Review comment: Put `attribute::Unique` into its own namespace.
2017-03-20 20:18:59 +00:00
[ 65539 :db . schema / attribute :name / Josef ? tx true ]
2019-07-17 15:46:47 +00:00
[ ? tx :db / txInstant ? ms ? tx true ] ] "
) ;
assert_matches! (
tempids ( & report ) ,
" { \" t1 \" 65538
\ " t2 \" 65539} "
) ;
2017-03-20 18:29:17 +00:00
// A multistep insert. This time, we can resolve both, but we have to try "t1", succeed,
// and then resolve "t2".
// TODO: We can't quite test this without more schema elements.
// conn.transact("[[:db/add \"t1\" :db/ident :name/Josef]
// [:db/add \"t2\" :db/ident \"t1\"]]");
// assert_matches!(conn.last_transaction(),
// "[[65538 :db/ident :name/Josef]
// [65538 :db/ident :name/Karl]
// [?tx :db/txInstant ?ms ?tx true]]");
2017-01-26 00:13:56 +00:00
}
2017-03-13 16:39:19 +00:00
Add type checking and constraint checking to the transactor. (#663, #532, #679)
This should address #663, by re-inserting type checking in the
transactor stack after the entry point used by the term builder.
Before this commit, we were using an SQLite UNIQUE index to assert
that no `[e a]` pair, with `a` a cardinality one attribute, was
asserted more than once. However, that's not in line with Datomic,
which treats transaction inputs as a set and allows a single datom
like `[e a v]` to appear multiple times. It's both awkward and not
particularly efficient to look for _distinct_ repetitions in SQL, so
we accept some runtime cost in order to check for repetitions in the
transactor. This will allow us to address #532, which is really about
whether we treat inputs as sets. A side benefit is that we can
provide more helpful error messages when the transactor does detect
that the input truly violates the cardinality constraints of the
schema.
This commit builds a trie while error checking and collecting final
terms, which should be fairly efficient. It also allows a simpler
expression of input-provided :db/txInstant datoms, which in turn
uncovered a small issue with the transaction watcher, where-by the
watcher would not see non-input-provided :db/txInstant datoms.
This transition to Datomic-like input-as-set semantics allows us to
address #532. Previously, two tempids that upserted to the same entid
would produce duplicate datoms, and that would have been rejected by
the transactor -- correctly, since we did not allow duplicate datoms
under the input-as-list semantics. With input-as-set semantics,
duplicate datoms are allowed; and that means that we must allow
tempids to be equivalent, i.e., to resolve to the same tempid.
To achieve this, we:
- index the set of tempids
- identify tempid indices that share an upsert
- map tempids to a dense set of contiguous integer labels
We use the well-known union-find algorithm, as implemented by
petgraph, to efficiently manage the set of equivalent tempids.
Along the way, I've fixed and added tests for two small errors in the
transactor. First, don't drop datoms resolved by upsert (#679).
Second, ensure that complex upserts are allocated.
I don't know quite what happened here. The Clojure implementation
correctly kept complex upserts that hadn't resolved as complex
upserts (see
https://github.com/mozilla/mentat/blob/9a9dfb502acf5e4cdb1059d4aac831d7603063c8/src/common/datomish/transact.cljc#L436)
and then allocated complex upserts if they didn't resolve (see
https://github.com/mozilla/mentat/blob/9a9dfb502acf5e4cdb1059d4aac831d7603063c8/src/common/datomish/transact.cljc#L509).
Based on the code comments, I think the Rust implementation must have
incorrectly tried to optimize by handling all complex upserts in at
most a single generation of evolution, and that's just not correct.
We're effectively implementing a topological sort, using very specific
domain knowledge, and its not true that a node in a topological sort
can be considered only once!
2018-04-30 22:16:05 +00:00
#[ test ]
fn test_resolved_upserts ( ) {
let mut conn = TestConn ::default ( ) ;
2019-07-17 15:46:47 +00:00
assert_transact! (
conn ,
" [
Add type checking and constraint checking to the transactor. (#663, #532, #679)
This should address #663, by re-inserting type checking in the
transactor stack after the entry point used by the term builder.
Before this commit, we were using an SQLite UNIQUE index to assert
that no `[e a]` pair, with `a` a cardinality one attribute, was
asserted more than once. However, that's not in line with Datomic,
which treats transaction inputs as a set and allows a single datom
like `[e a v]` to appear multiple times. It's both awkward and not
particularly efficient to look for _distinct_ repetitions in SQL, so
we accept some runtime cost in order to check for repetitions in the
transactor. This will allow us to address #532, which is really about
whether we treat inputs as sets. A side benefit is that we can
provide more helpful error messages when the transactor does detect
that the input truly violates the cardinality constraints of the
schema.
This commit builds a trie while error checking and collecting final
terms, which should be fairly efficient. It also allows a simpler
expression of input-provided :db/txInstant datoms, which in turn
uncovered a small issue with the transaction watcher, where-by the
watcher would not see non-input-provided :db/txInstant datoms.
This transition to Datomic-like input-as-set semantics allows us to
address #532. Previously, two tempids that upserted to the same entid
would produce duplicate datoms, and that would have been rejected by
the transactor -- correctly, since we did not allow duplicate datoms
under the input-as-list semantics. With input-as-set semantics,
duplicate datoms are allowed; and that means that we must allow
tempids to be equivalent, i.e., to resolve to the same tempid.
To achieve this, we:
- index the set of tempids
- identify tempid indices that share an upsert
- map tempids to a dense set of contiguous integer labels
We use the well-known union-find algorithm, as implemented by
petgraph, to efficiently manage the set of equivalent tempids.
Along the way, I've fixed and added tests for two small errors in the
transactor. First, don't drop datoms resolved by upsert (#679).
Second, ensure that complex upserts are allocated.
I don't know quite what happened here. The Clojure implementation
correctly kept complex upserts that hadn't resolved as complex
upserts (see
https://github.com/mozilla/mentat/blob/9a9dfb502acf5e4cdb1059d4aac831d7603063c8/src/common/datomish/transact.cljc#L436)
and then allocated complex upserts if they didn't resolve (see
https://github.com/mozilla/mentat/blob/9a9dfb502acf5e4cdb1059d4aac831d7603063c8/src/common/datomish/transact.cljc#L509).
Based on the code comments, I think the Rust implementation must have
incorrectly tried to optimize by handling all complex upserts in at
most a single generation of evolution, and that's just not correct.
We're effectively implementing a topological sort, using very specific
domain knowledge, and its not true that a node in a topological sort
can be considered only once!
2018-04-30 22:16:05 +00:00
{ :db / ident :test / id
:db / valueType :db . type / string
:db / unique :db . unique / identity
:db / index true
:db / cardinality :db . cardinality / one }
{ :db / ident :test / ref
:db / valueType :db . type / ref
:db / unique :db . unique / identity
:db / index true
:db / cardinality :db . cardinality / one }
2019-07-17 15:46:47 +00:00
] "
) ;
Add type checking and constraint checking to the transactor. (#663, #532, #679)
This should address #663, by re-inserting type checking in the
transactor stack after the entry point used by the term builder.
Before this commit, we were using an SQLite UNIQUE index to assert
that no `[e a]` pair, with `a` a cardinality one attribute, was
asserted more than once. However, that's not in line with Datomic,
which treats transaction inputs as a set and allows a single datom
like `[e a v]` to appear multiple times. It's both awkward and not
particularly efficient to look for _distinct_ repetitions in SQL, so
we accept some runtime cost in order to check for repetitions in the
transactor. This will allow us to address #532, which is really about
whether we treat inputs as sets. A side benefit is that we can
provide more helpful error messages when the transactor does detect
that the input truly violates the cardinality constraints of the
schema.
This commit builds a trie while error checking and collecting final
terms, which should be fairly efficient. It also allows a simpler
expression of input-provided :db/txInstant datoms, which in turn
uncovered a small issue with the transaction watcher, where-by the
watcher would not see non-input-provided :db/txInstant datoms.
This transition to Datomic-like input-as-set semantics allows us to
address #532. Previously, two tempids that upserted to the same entid
would produce duplicate datoms, and that would have been rejected by
the transactor -- correctly, since we did not allow duplicate datoms
under the input-as-list semantics. With input-as-set semantics,
duplicate datoms are allowed; and that means that we must allow
tempids to be equivalent, i.e., to resolve to the same tempid.
To achieve this, we:
- index the set of tempids
- identify tempid indices that share an upsert
- map tempids to a dense set of contiguous integer labels
We use the well-known union-find algorithm, as implemented by
petgraph, to efficiently manage the set of equivalent tempids.
Along the way, I've fixed and added tests for two small errors in the
transactor. First, don't drop datoms resolved by upsert (#679).
Second, ensure that complex upserts are allocated.
I don't know quite what happened here. The Clojure implementation
correctly kept complex upserts that hadn't resolved as complex
upserts (see
https://github.com/mozilla/mentat/blob/9a9dfb502acf5e4cdb1059d4aac831d7603063c8/src/common/datomish/transact.cljc#L436)
and then allocated complex upserts if they didn't resolve (see
https://github.com/mozilla/mentat/blob/9a9dfb502acf5e4cdb1059d4aac831d7603063c8/src/common/datomish/transact.cljc#L509).
Based on the code comments, I think the Rust implementation must have
incorrectly tried to optimize by handling all complex upserts in at
most a single generation of evolution, and that's just not correct.
We're effectively implementing a topological sort, using very specific
domain knowledge, and its not true that a node in a topological sort
can be considered only once!
2018-04-30 22:16:05 +00:00
// Partial data for :test/id, links via :test/ref.
2019-07-17 15:46:47 +00:00
assert_transact! (
conn ,
r #" [
Add type checking and constraint checking to the transactor. (#663, #532, #679)
This should address #663, by re-inserting type checking in the
transactor stack after the entry point used by the term builder.
Before this commit, we were using an SQLite UNIQUE index to assert
that no `[e a]` pair, with `a` a cardinality one attribute, was
asserted more than once. However, that's not in line with Datomic,
which treats transaction inputs as a set and allows a single datom
like `[e a v]` to appear multiple times. It's both awkward and not
particularly efficient to look for _distinct_ repetitions in SQL, so
we accept some runtime cost in order to check for repetitions in the
transactor. This will allow us to address #532, which is really about
whether we treat inputs as sets. A side benefit is that we can
provide more helpful error messages when the transactor does detect
that the input truly violates the cardinality constraints of the
schema.
This commit builds a trie while error checking and collecting final
terms, which should be fairly efficient. It also allows a simpler
expression of input-provided :db/txInstant datoms, which in turn
uncovered a small issue with the transaction watcher, where-by the
watcher would not see non-input-provided :db/txInstant datoms.
This transition to Datomic-like input-as-set semantics allows us to
address #532. Previously, two tempids that upserted to the same entid
would produce duplicate datoms, and that would have been rejected by
the transactor -- correctly, since we did not allow duplicate datoms
under the input-as-list semantics. With input-as-set semantics,
duplicate datoms are allowed; and that means that we must allow
tempids to be equivalent, i.e., to resolve to the same tempid.
To achieve this, we:
- index the set of tempids
- identify tempid indices that share an upsert
- map tempids to a dense set of contiguous integer labels
We use the well-known union-find algorithm, as implemented by
petgraph, to efficiently manage the set of equivalent tempids.
Along the way, I've fixed and added tests for two small errors in the
transactor. First, don't drop datoms resolved by upsert (#679).
Second, ensure that complex upserts are allocated.
I don't know quite what happened here. The Clojure implementation
correctly kept complex upserts that hadn't resolved as complex
upserts (see
https://github.com/mozilla/mentat/blob/9a9dfb502acf5e4cdb1059d4aac831d7603063c8/src/common/datomish/transact.cljc#L436)
and then allocated complex upserts if they didn't resolve (see
https://github.com/mozilla/mentat/blob/9a9dfb502acf5e4cdb1059d4aac831d7603063c8/src/common/datomish/transact.cljc#L509).
Based on the code comments, I think the Rust implementation must have
incorrectly tried to optimize by handling all complex upserts in at
most a single generation of evolution, and that's just not correct.
We're effectively implementing a topological sort, using very specific
domain knowledge, and its not true that a node in a topological sort
can be considered only once!
2018-04-30 22:16:05 +00:00
[ :db / add 100 :test / id " 0 " ]
[ :db / add 101 :test / ref 100 ]
[ :db / add 102 :test / ref 101 ]
[ :db / add 103 :test / ref 102 ]
2019-07-17 15:46:47 +00:00
] " #
) ;
Add type checking and constraint checking to the transactor. (#663, #532, #679)
This should address #663, by re-inserting type checking in the
transactor stack after the entry point used by the term builder.
Before this commit, we were using an SQLite UNIQUE index to assert
that no `[e a]` pair, with `a` a cardinality one attribute, was
asserted more than once. However, that's not in line with Datomic,
which treats transaction inputs as a set and allows a single datom
like `[e a v]` to appear multiple times. It's both awkward and not
particularly efficient to look for _distinct_ repetitions in SQL, so
we accept some runtime cost in order to check for repetitions in the
transactor. This will allow us to address #532, which is really about
whether we treat inputs as sets. A side benefit is that we can
provide more helpful error messages when the transactor does detect
that the input truly violates the cardinality constraints of the
schema.
This commit builds a trie while error checking and collecting final
terms, which should be fairly efficient. It also allows a simpler
expression of input-provided :db/txInstant datoms, which in turn
uncovered a small issue with the transaction watcher, where-by the
watcher would not see non-input-provided :db/txInstant datoms.
This transition to Datomic-like input-as-set semantics allows us to
address #532. Previously, two tempids that upserted to the same entid
would produce duplicate datoms, and that would have been rejected by
the transactor -- correctly, since we did not allow duplicate datoms
under the input-as-list semantics. With input-as-set semantics,
duplicate datoms are allowed; and that means that we must allow
tempids to be equivalent, i.e., to resolve to the same tempid.
To achieve this, we:
- index the set of tempids
- identify tempid indices that share an upsert
- map tempids to a dense set of contiguous integer labels
We use the well-known union-find algorithm, as implemented by
petgraph, to efficiently manage the set of equivalent tempids.
Along the way, I've fixed and added tests for two small errors in the
transactor. First, don't drop datoms resolved by upsert (#679).
Second, ensure that complex upserts are allocated.
I don't know quite what happened here. The Clojure implementation
correctly kept complex upserts that hadn't resolved as complex
upserts (see
https://github.com/mozilla/mentat/blob/9a9dfb502acf5e4cdb1059d4aac831d7603063c8/src/common/datomish/transact.cljc#L436)
and then allocated complex upserts if they didn't resolve (see
https://github.com/mozilla/mentat/blob/9a9dfb502acf5e4cdb1059d4aac831d7603063c8/src/common/datomish/transact.cljc#L509).
Based on the code comments, I think the Rust implementation must have
incorrectly tried to optimize by handling all complex upserts in at
most a single generation of evolution, and that's just not correct.
We're effectively implementing a topological sort, using very specific
domain knowledge, and its not true that a node in a topological sort
can be considered only once!
2018-04-30 22:16:05 +00:00
// Fill in the rest of the data for :test/id, using the links of :test/ref.
2019-07-17 15:46:47 +00:00
let report = assert_transact! (
conn ,
r #" [
Add type checking and constraint checking to the transactor. (#663, #532, #679)
This should address #663, by re-inserting type checking in the
transactor stack after the entry point used by the term builder.
Before this commit, we were using an SQLite UNIQUE index to assert
that no `[e a]` pair, with `a` a cardinality one attribute, was
asserted more than once. However, that's not in line with Datomic,
which treats transaction inputs as a set and allows a single datom
like `[e a v]` to appear multiple times. It's both awkward and not
particularly efficient to look for _distinct_ repetitions in SQL, so
we accept some runtime cost in order to check for repetitions in the
transactor. This will allow us to address #532, which is really about
whether we treat inputs as sets. A side benefit is that we can
provide more helpful error messages when the transactor does detect
that the input truly violates the cardinality constraints of the
schema.
This commit builds a trie while error checking and collecting final
terms, which should be fairly efficient. It also allows a simpler
expression of input-provided :db/txInstant datoms, which in turn
uncovered a small issue with the transaction watcher, where-by the
watcher would not see non-input-provided :db/txInstant datoms.
This transition to Datomic-like input-as-set semantics allows us to
address #532. Previously, two tempids that upserted to the same entid
would produce duplicate datoms, and that would have been rejected by
the transactor -- correctly, since we did not allow duplicate datoms
under the input-as-list semantics. With input-as-set semantics,
duplicate datoms are allowed; and that means that we must allow
tempids to be equivalent, i.e., to resolve to the same tempid.
To achieve this, we:
- index the set of tempids
- identify tempid indices that share an upsert
- map tempids to a dense set of contiguous integer labels
We use the well-known union-find algorithm, as implemented by
petgraph, to efficiently manage the set of equivalent tempids.
Along the way, I've fixed and added tests for two small errors in the
transactor. First, don't drop datoms resolved by upsert (#679).
Second, ensure that complex upserts are allocated.
I don't know quite what happened here. The Clojure implementation
correctly kept complex upserts that hadn't resolved as complex
upserts (see
https://github.com/mozilla/mentat/blob/9a9dfb502acf5e4cdb1059d4aac831d7603063c8/src/common/datomish/transact.cljc#L436)
and then allocated complex upserts if they didn't resolve (see
https://github.com/mozilla/mentat/blob/9a9dfb502acf5e4cdb1059d4aac831d7603063c8/src/common/datomish/transact.cljc#L509).
Based on the code comments, I think the Rust implementation must have
incorrectly tried to optimize by handling all complex upserts in at
most a single generation of evolution, and that's just not correct.
We're effectively implementing a topological sort, using very specific
domain knowledge, and its not true that a node in a topological sort
can be considered only once!
2018-04-30 22:16:05 +00:00
{ :db / id " a " :test / id " 0 " }
{ :db / id " b " :test / id " 1 " :test / ref " a " }
{ :db / id " c " :test / id " 2 " :test / ref " b " }
{ :db / id " d " :test / id " 3 " :test / ref " c " }
2019-07-17 15:46:47 +00:00
] " #
) ;
Add type checking and constraint checking to the transactor. (#663, #532, #679)
This should address #663, by re-inserting type checking in the
transactor stack after the entry point used by the term builder.
Before this commit, we were using an SQLite UNIQUE index to assert
that no `[e a]` pair, with `a` a cardinality one attribute, was
asserted more than once. However, that's not in line with Datomic,
which treats transaction inputs as a set and allows a single datom
like `[e a v]` to appear multiple times. It's both awkward and not
particularly efficient to look for _distinct_ repetitions in SQL, so
we accept some runtime cost in order to check for repetitions in the
transactor. This will allow us to address #532, which is really about
whether we treat inputs as sets. A side benefit is that we can
provide more helpful error messages when the transactor does detect
that the input truly violates the cardinality constraints of the
schema.
This commit builds a trie while error checking and collecting final
terms, which should be fairly efficient. It also allows a simpler
expression of input-provided :db/txInstant datoms, which in turn
uncovered a small issue with the transaction watcher, where-by the
watcher would not see non-input-provided :db/txInstant datoms.
This transition to Datomic-like input-as-set semantics allows us to
address #532. Previously, two tempids that upserted to the same entid
would produce duplicate datoms, and that would have been rejected by
the transactor -- correctly, since we did not allow duplicate datoms
under the input-as-list semantics. With input-as-set semantics,
duplicate datoms are allowed; and that means that we must allow
tempids to be equivalent, i.e., to resolve to the same tempid.
To achieve this, we:
- index the set of tempids
- identify tempid indices that share an upsert
- map tempids to a dense set of contiguous integer labels
We use the well-known union-find algorithm, as implemented by
petgraph, to efficiently manage the set of equivalent tempids.
Along the way, I've fixed and added tests for two small errors in the
transactor. First, don't drop datoms resolved by upsert (#679).
Second, ensure that complex upserts are allocated.
I don't know quite what happened here. The Clojure implementation
correctly kept complex upserts that hadn't resolved as complex
upserts (see
https://github.com/mozilla/mentat/blob/9a9dfb502acf5e4cdb1059d4aac831d7603063c8/src/common/datomish/transact.cljc#L436)
and then allocated complex upserts if they didn't resolve (see
https://github.com/mozilla/mentat/blob/9a9dfb502acf5e4cdb1059d4aac831d7603063c8/src/common/datomish/transact.cljc#L509).
Based on the code comments, I think the Rust implementation must have
incorrectly tried to optimize by handling all complex upserts in at
most a single generation of evolution, and that's just not correct.
We're effectively implementing a topological sort, using very specific
domain knowledge, and its not true that a node in a topological sort
can be considered only once!
2018-04-30 22:16:05 +00:00
2019-07-17 15:46:47 +00:00
assert_matches! (
tempids ( & report ) ,
r #" {
Add type checking and constraint checking to the transactor. (#663, #532, #679)
This should address #663, by re-inserting type checking in the
transactor stack after the entry point used by the term builder.
Before this commit, we were using an SQLite UNIQUE index to assert
that no `[e a]` pair, with `a` a cardinality one attribute, was
asserted more than once. However, that's not in line with Datomic,
which treats transaction inputs as a set and allows a single datom
like `[e a v]` to appear multiple times. It's both awkward and not
particularly efficient to look for _distinct_ repetitions in SQL, so
we accept some runtime cost in order to check for repetitions in the
transactor. This will allow us to address #532, which is really about
whether we treat inputs as sets. A side benefit is that we can
provide more helpful error messages when the transactor does detect
that the input truly violates the cardinality constraints of the
schema.
This commit builds a trie while error checking and collecting final
terms, which should be fairly efficient. It also allows a simpler
expression of input-provided :db/txInstant datoms, which in turn
uncovered a small issue with the transaction watcher, where-by the
watcher would not see non-input-provided :db/txInstant datoms.
This transition to Datomic-like input-as-set semantics allows us to
address #532. Previously, two tempids that upserted to the same entid
would produce duplicate datoms, and that would have been rejected by
the transactor -- correctly, since we did not allow duplicate datoms
under the input-as-list semantics. With input-as-set semantics,
duplicate datoms are allowed; and that means that we must allow
tempids to be equivalent, i.e., to resolve to the same tempid.
To achieve this, we:
- index the set of tempids
- identify tempid indices that share an upsert
- map tempids to a dense set of contiguous integer labels
We use the well-known union-find algorithm, as implemented by
petgraph, to efficiently manage the set of equivalent tempids.
Along the way, I've fixed and added tests for two small errors in the
transactor. First, don't drop datoms resolved by upsert (#679).
Second, ensure that complex upserts are allocated.
I don't know quite what happened here. The Clojure implementation
correctly kept complex upserts that hadn't resolved as complex
upserts (see
https://github.com/mozilla/mentat/blob/9a9dfb502acf5e4cdb1059d4aac831d7603063c8/src/common/datomish/transact.cljc#L436)
and then allocated complex upserts if they didn't resolve (see
https://github.com/mozilla/mentat/blob/9a9dfb502acf5e4cdb1059d4aac831d7603063c8/src/common/datomish/transact.cljc#L509).
Based on the code comments, I think the Rust implementation must have
incorrectly tried to optimize by handling all complex upserts in at
most a single generation of evolution, and that's just not correct.
We're effectively implementing a topological sort, using very specific
domain knowledge, and its not true that a node in a topological sort
can be considered only once!
2018-04-30 22:16:05 +00:00
" a " 100
" b " 101
" c " 102
" d " 103
2019-07-17 15:46:47 +00:00
} " #
) ;
Add type checking and constraint checking to the transactor. (#663, #532, #679)
This should address #663, by re-inserting type checking in the
transactor stack after the entry point used by the term builder.
Before this commit, we were using an SQLite UNIQUE index to assert
that no `[e a]` pair, with `a` a cardinality one attribute, was
asserted more than once. However, that's not in line with Datomic,
which treats transaction inputs as a set and allows a single datom
like `[e a v]` to appear multiple times. It's both awkward and not
particularly efficient to look for _distinct_ repetitions in SQL, so
we accept some runtime cost in order to check for repetitions in the
transactor. This will allow us to address #532, which is really about
whether we treat inputs as sets. A side benefit is that we can
provide more helpful error messages when the transactor does detect
that the input truly violates the cardinality constraints of the
schema.
This commit builds a trie while error checking and collecting final
terms, which should be fairly efficient. It also allows a simpler
expression of input-provided :db/txInstant datoms, which in turn
uncovered a small issue with the transaction watcher, where-by the
watcher would not see non-input-provided :db/txInstant datoms.
This transition to Datomic-like input-as-set semantics allows us to
address #532. Previously, two tempids that upserted to the same entid
would produce duplicate datoms, and that would have been rejected by
the transactor -- correctly, since we did not allow duplicate datoms
under the input-as-list semantics. With input-as-set semantics,
duplicate datoms are allowed; and that means that we must allow
tempids to be equivalent, i.e., to resolve to the same tempid.
To achieve this, we:
- index the set of tempids
- identify tempid indices that share an upsert
- map tempids to a dense set of contiguous integer labels
We use the well-known union-find algorithm, as implemented by
petgraph, to efficiently manage the set of equivalent tempids.
Along the way, I've fixed and added tests for two small errors in the
transactor. First, don't drop datoms resolved by upsert (#679).
Second, ensure that complex upserts are allocated.
I don't know quite what happened here. The Clojure implementation
correctly kept complex upserts that hadn't resolved as complex
upserts (see
https://github.com/mozilla/mentat/blob/9a9dfb502acf5e4cdb1059d4aac831d7603063c8/src/common/datomish/transact.cljc#L436)
and then allocated complex upserts if they didn't resolve (see
https://github.com/mozilla/mentat/blob/9a9dfb502acf5e4cdb1059d4aac831d7603063c8/src/common/datomish/transact.cljc#L509).
Based on the code comments, I think the Rust implementation must have
incorrectly tried to optimize by handling all complex upserts in at
most a single generation of evolution, and that's just not correct.
We're effectively implementing a topological sort, using very specific
domain knowledge, and its not true that a node in a topological sort
can be considered only once!
2018-04-30 22:16:05 +00:00
2019-07-17 15:46:47 +00:00
assert_matches! (
conn . last_transaction ( ) ,
r #" [
Add type checking and constraint checking to the transactor. (#663, #532, #679)
This should address #663, by re-inserting type checking in the
transactor stack after the entry point used by the term builder.
Before this commit, we were using an SQLite UNIQUE index to assert
that no `[e a]` pair, with `a` a cardinality one attribute, was
asserted more than once. However, that's not in line with Datomic,
which treats transaction inputs as a set and allows a single datom
like `[e a v]` to appear multiple times. It's both awkward and not
particularly efficient to look for _distinct_ repetitions in SQL, so
we accept some runtime cost in order to check for repetitions in the
transactor. This will allow us to address #532, which is really about
whether we treat inputs as sets. A side benefit is that we can
provide more helpful error messages when the transactor does detect
that the input truly violates the cardinality constraints of the
schema.
This commit builds a trie while error checking and collecting final
terms, which should be fairly efficient. It also allows a simpler
expression of input-provided :db/txInstant datoms, which in turn
uncovered a small issue with the transaction watcher, where-by the
watcher would not see non-input-provided :db/txInstant datoms.
This transition to Datomic-like input-as-set semantics allows us to
address #532. Previously, two tempids that upserted to the same entid
would produce duplicate datoms, and that would have been rejected by
the transactor -- correctly, since we did not allow duplicate datoms
under the input-as-list semantics. With input-as-set semantics,
duplicate datoms are allowed; and that means that we must allow
tempids to be equivalent, i.e., to resolve to the same tempid.
To achieve this, we:
- index the set of tempids
- identify tempid indices that share an upsert
- map tempids to a dense set of contiguous integer labels
We use the well-known union-find algorithm, as implemented by
petgraph, to efficiently manage the set of equivalent tempids.
Along the way, I've fixed and added tests for two small errors in the
transactor. First, don't drop datoms resolved by upsert (#679).
Second, ensure that complex upserts are allocated.
I don't know quite what happened here. The Clojure implementation
correctly kept complex upserts that hadn't resolved as complex
upserts (see
https://github.com/mozilla/mentat/blob/9a9dfb502acf5e4cdb1059d4aac831d7603063c8/src/common/datomish/transact.cljc#L436)
and then allocated complex upserts if they didn't resolve (see
https://github.com/mozilla/mentat/blob/9a9dfb502acf5e4cdb1059d4aac831d7603063c8/src/common/datomish/transact.cljc#L509).
Based on the code comments, I think the Rust implementation must have
incorrectly tried to optimize by handling all complex upserts in at
most a single generation of evolution, and that's just not correct.
We're effectively implementing a topological sort, using very specific
domain knowledge, and its not true that a node in a topological sort
can be considered only once!
2018-04-30 22:16:05 +00:00
[ 101 :test / id " 1 " ? tx true ]
[ 102 :test / id " 2 " ? tx true ]
[ 103 :test / id " 3 " ? tx true ]
[ ? tx :db / txInstant ? ms ? tx true ]
2019-07-17 15:46:47 +00:00
] " #
) ;
Add type checking and constraint checking to the transactor. (#663, #532, #679)
This should address #663, by re-inserting type checking in the
transactor stack after the entry point used by the term builder.
Before this commit, we were using an SQLite UNIQUE index to assert
that no `[e a]` pair, with `a` a cardinality one attribute, was
asserted more than once. However, that's not in line with Datomic,
which treats transaction inputs as a set and allows a single datom
like `[e a v]` to appear multiple times. It's both awkward and not
particularly efficient to look for _distinct_ repetitions in SQL, so
we accept some runtime cost in order to check for repetitions in the
transactor. This will allow us to address #532, which is really about
whether we treat inputs as sets. A side benefit is that we can
provide more helpful error messages when the transactor does detect
that the input truly violates the cardinality constraints of the
schema.
This commit builds a trie while error checking and collecting final
terms, which should be fairly efficient. It also allows a simpler
expression of input-provided :db/txInstant datoms, which in turn
uncovered a small issue with the transaction watcher, where-by the
watcher would not see non-input-provided :db/txInstant datoms.
This transition to Datomic-like input-as-set semantics allows us to
address #532. Previously, two tempids that upserted to the same entid
would produce duplicate datoms, and that would have been rejected by
the transactor -- correctly, since we did not allow duplicate datoms
under the input-as-list semantics. With input-as-set semantics,
duplicate datoms are allowed; and that means that we must allow
tempids to be equivalent, i.e., to resolve to the same tempid.
To achieve this, we:
- index the set of tempids
- identify tempid indices that share an upsert
- map tempids to a dense set of contiguous integer labels
We use the well-known union-find algorithm, as implemented by
petgraph, to efficiently manage the set of equivalent tempids.
Along the way, I've fixed and added tests for two small errors in the
transactor. First, don't drop datoms resolved by upsert (#679).
Second, ensure that complex upserts are allocated.
I don't know quite what happened here. The Clojure implementation
correctly kept complex upserts that hadn't resolved as complex
upserts (see
https://github.com/mozilla/mentat/blob/9a9dfb502acf5e4cdb1059d4aac831d7603063c8/src/common/datomish/transact.cljc#L436)
and then allocated complex upserts if they didn't resolve (see
https://github.com/mozilla/mentat/blob/9a9dfb502acf5e4cdb1059d4aac831d7603063c8/src/common/datomish/transact.cljc#L509).
Based on the code comments, I think the Rust implementation must have
incorrectly tried to optimize by handling all complex upserts in at
most a single generation of evolution, and that's just not correct.
We're effectively implementing a topological sort, using very specific
domain knowledge, and its not true that a node in a topological sort
can be considered only once!
2018-04-30 22:16:05 +00:00
}
2017-03-13 16:39:19 +00:00
#[ test ]
fn test_sqlite_limit ( ) {
let conn = new_connection ( " " ) . expect ( " Couldn't open in-memory db " ) ;
let initial = conn . limit ( Limit ::SQLITE_LIMIT_VARIABLE_NUMBER ) ;
// Sanity check.
assert! ( initial > 500 ) ;
// Make sure setting works.
conn . set_limit ( Limit ::SQLITE_LIMIT_VARIABLE_NUMBER , 222 ) ;
assert_eq! ( 222 , conn . limit ( Limit ::SQLITE_LIMIT_VARIABLE_NUMBER ) ) ;
}
Schema alteration. Fixes #294 and #295. (#370) r=rnewman
* Pre: Don't retract :db/ident in test.
Datomic (and eventually Mentat) don't allow to retract :db/ident in
this way, so this runs afoul of future work to support mutating
metadata.
* Pre: s/VALUETYPE/VALUE_TYPE/.
This is consistent with the capitalization (which is "valueType") and
the other identifier.
* Pre: Remove some single quotes from error output.
* Part 1: Make materialized views be uniform [e a v value_type_tag].
This looks ahead to a time when we could support arbitrary
user-defined materialized views. For now, the "idents" materialized
view is those datoms of the form [e :db/ident :namespaced/keyword] and
the "schema" materialized view is those datoms of the form [e a v]
where a is in a particular set of attributes that will become clear in
the following commits.
This change is not backwards compatible, so I'm removing the open
current (really, v2) test. It'll be re-instated when we get to
https://github.com/mozilla/mentat/issues/194.
* Pre: Map TypedValue::Ref to TypedValue::Keyword in debug output.
* Part 3: Separate `schema_to_mutate` from the `schema` used to interpret.
This is just to keep track of the expected changes during
bootstrapping. I want bootstrap metadata mutations to flow through
the same code path as metadata mutations during regular transactions;
by differentiating the schema used for interpretation from the schema
that will be updated I expect to be able to apply bootstrap metadata
mutations to an empty schema and have things like materialized views
created (using the regular code paths).
This commit has been re-ordered for conceptual clarity, but it won't
compile because it references the metadata module. It's possible to
make it compile -- the functionality is there in the schema module --
but it's not worth the rebasing effort until after review (and
possibly not even then, since we'll squash down to a single commit to
land).
* Part 2: Maintain entids separately from idents.
In order to support historical idents, we need to distinguish the
"current" map from entid -> ident from the "complete historical" map
ident -> entid. This is what Datomic does; in Datomic, an ident is
never retracted (although it can be replaced). This approach is an
important part of allowing multiple consumers to share a schema
fragment as it migrates forward.
This fixes a limitation of the Clojure implementation, which did not
handle historical idents across knowledge base close and re-open.
The "entids" materialized view is naturally a slice of the "datoms"
table. The "idents" materialized view is a slice of the
"transactions" table. I hope that representing in this way, and
casting the problem in this light, might generalize to future
materialized views.
* Pre: Add DiffSet.
* Part 4: Collect mutations to a `Schema`.
I haven't taken your review comment about consuming AttributeBuilder
during each fluent function. If you read my response and still want
this, I'm happy to do it in review.
* Part 5: Handle :db/ident and :db.{install,alter}/attribute.
This "loops" the committed datoms out of the SQL store and back
through the metadata (schema, but in future also partition map)
processor. The metadata processor updates the schema and produces a
report of what changed; that report is then used to update the SQL
store. That update includes:
- the materialized views ("entids", "idents", and "schema");
- if needed, a subset of the datoms themselves (as flags change).
I've left a TODO for handling attribute retraction in the cases that
it makes sense. I expect that to be straight-forward.
* Review comment: Rename DiffSet to AddRetractAlterSet.
Also adds a little more commentary and a simple test.
* Review comment: Use ToIdent trait.
* Review comment: partially revert "Part 2: Maintain entids separately from idents."
This reverts commit 23a91df9c35e14398f2ddbd1ba25315821e67401.
Following our discussion, this removes the "entids" materialized
view. The next commit will remove historical idents from the "idents"
materialized view.
* Post: Use custom Either rather than std::result::Result.
This is not necessary, but it was suggested that we might be paying an
overhead creating Err instances while using error_chain. That seems
not to be the case, but this change shows that we don't actually use
any of the Result helper methods, so there's no reason to overload
Result. This change might avoid some future confusion, so I'm going
to land it anyway.
Signed-off-by: Nick Alexander <nalexander@mozilla.com>
* Review comment: Don't preserve historical idents.
* Review comment: More prepared statements when updating materialized views.
* Post: Test altering :db/cardinality and :db/unique.
These tests fail due to a Datomic limitation, namely that the marker
flag :db.alter/attribute can only be asserted once for an attribute!
That is, [:db.part/db :db.alter/attribute :attribute] will only be
transacted at most once. Since older versions of Datomic required the
:db.alter/attribute flag, I can only imagine they either never wrote
:db.alter/attribute to the store, or they handled it specially. I'll
need to remove the marker flag system from Mentat in order to address
this fundamental limitation.
* Post: Remove some more single quotes from error output.
* Post: Add assert_transact! macro to unwrap safely.
I was finding it very difficult to track unwrapping errors while
making changes, due to an underlying Mac OS X symbolication issue that
makes running tests with RUST_BACKTRACE=1 so slow that they all time
out.
* Post: Don't expect or recognize :db.{install,alter}/attribute.
I had this all working... except we will never see a repeated
`[:db.part/db :db.alter/attribute :attribute]` assertion in the store!
That means my approach would let you alter an attribute at most one
time. It's not worth hacking around this; it's better to just stop
expecting (and recognizing) the marker flags. (We have all the data
to distinguish the various cases that we need without the marker
flags.)
This brings Mentat in line with the thrust of newer Datomic versions,
but isn't compatible with Datomic, because (if I understand correctly)
Datomic automatically adds :db.{install,alter}/attribute assertions to
transactions.
I haven't purged the corresponding :db/ident and schema fragments just
yet:
- we might want them back
- we might want them in order to upgrade v1 and v2 databases to the
new on-disk layout we're fleshing out (v3?).
* Post: Don't make :db/unique :db.unique/* imply :db/index true.
This patch avoids a potential bug with the "schema" materialized view.
If :db/unique :db.unique/value implies :db/index true, then what
happens when you _retract_ :db.unique/value? I think Datomic defines
this in some way, but I really want the "schema" materialized view to
be a slice of "datoms" and not have these sort of ambiguities and
persistent effects. Therefore, to ensure that we don't retract a
schema characteristic and accidentally change more than we intended
to, this patch stops having any schema characteristic imply any other
schema characteristic(s). To achieve that, I added an
Option<Unique::{Value,Identity}> type to Attribute; this helps with
this patch, and also looks ahead to when we allow to retract
:db/unique attributes.
* Post: Allow to retract :db/ident.
* Post: Include more details about invalid schema changes.
The tests use strings, so they hide the chained errors which do in
fact provide more detail.
* Review comment: Fix outdated comment.
* Review comment: s/_SET/_SQL_LIST/.
* Review comment: Use a sub-select for checking cardinality.
This might be faster in practice.
* Review comment: Put `attribute::Unique` into its own namespace.
2017-03-20 20:18:59 +00:00
#[ test ]
fn test_db_install ( ) {
let mut conn = TestConn ::default ( ) ;
2018-07-18 23:55:31 +00:00
// We're missing some tests here, since our implementation is incomplete.
// See https://github.com/mozilla/mentat/issues/797
// We can assert a new schema attribute.
2019-07-17 15:46:47 +00:00
assert_transact! (
conn ,
" [[:db/add 100 :db/ident :test/ident]
Schema alteration. Fixes #294 and #295. (#370) r=rnewman
* Pre: Don't retract :db/ident in test.
Datomic (and eventually Mentat) don't allow to retract :db/ident in
this way, so this runs afoul of future work to support mutating
metadata.
* Pre: s/VALUETYPE/VALUE_TYPE/.
This is consistent with the capitalization (which is "valueType") and
the other identifier.
* Pre: Remove some single quotes from error output.
* Part 1: Make materialized views be uniform [e a v value_type_tag].
This looks ahead to a time when we could support arbitrary
user-defined materialized views. For now, the "idents" materialized
view is those datoms of the form [e :db/ident :namespaced/keyword] and
the "schema" materialized view is those datoms of the form [e a v]
where a is in a particular set of attributes that will become clear in
the following commits.
This change is not backwards compatible, so I'm removing the open
current (really, v2) test. It'll be re-instated when we get to
https://github.com/mozilla/mentat/issues/194.
* Pre: Map TypedValue::Ref to TypedValue::Keyword in debug output.
* Part 3: Separate `schema_to_mutate` from the `schema` used to interpret.
This is just to keep track of the expected changes during
bootstrapping. I want bootstrap metadata mutations to flow through
the same code path as metadata mutations during regular transactions;
by differentiating the schema used for interpretation from the schema
that will be updated I expect to be able to apply bootstrap metadata
mutations to an empty schema and have things like materialized views
created (using the regular code paths).
This commit has been re-ordered for conceptual clarity, but it won't
compile because it references the metadata module. It's possible to
make it compile -- the functionality is there in the schema module --
but it's not worth the rebasing effort until after review (and
possibly not even then, since we'll squash down to a single commit to
land).
* Part 2: Maintain entids separately from idents.
In order to support historical idents, we need to distinguish the
"current" map from entid -> ident from the "complete historical" map
ident -> entid. This is what Datomic does; in Datomic, an ident is
never retracted (although it can be replaced). This approach is an
important part of allowing multiple consumers to share a schema
fragment as it migrates forward.
This fixes a limitation of the Clojure implementation, which did not
handle historical idents across knowledge base close and re-open.
The "entids" materialized view is naturally a slice of the "datoms"
table. The "idents" materialized view is a slice of the
"transactions" table. I hope that representing in this way, and
casting the problem in this light, might generalize to future
materialized views.
* Pre: Add DiffSet.
* Part 4: Collect mutations to a `Schema`.
I haven't taken your review comment about consuming AttributeBuilder
during each fluent function. If you read my response and still want
this, I'm happy to do it in review.
* Part 5: Handle :db/ident and :db.{install,alter}/attribute.
This "loops" the committed datoms out of the SQL store and back
through the metadata (schema, but in future also partition map)
processor. The metadata processor updates the schema and produces a
report of what changed; that report is then used to update the SQL
store. That update includes:
- the materialized views ("entids", "idents", and "schema");
- if needed, a subset of the datoms themselves (as flags change).
I've left a TODO for handling attribute retraction in the cases that
it makes sense. I expect that to be straight-forward.
* Review comment: Rename DiffSet to AddRetractAlterSet.
Also adds a little more commentary and a simple test.
* Review comment: Use ToIdent trait.
* Review comment: partially revert "Part 2: Maintain entids separately from idents."
This reverts commit 23a91df9c35e14398f2ddbd1ba25315821e67401.
Following our discussion, this removes the "entids" materialized
view. The next commit will remove historical idents from the "idents"
materialized view.
* Post: Use custom Either rather than std::result::Result.
This is not necessary, but it was suggested that we might be paying an
overhead creating Err instances while using error_chain. That seems
not to be the case, but this change shows that we don't actually use
any of the Result helper methods, so there's no reason to overload
Result. This change might avoid some future confusion, so I'm going
to land it anyway.
Signed-off-by: Nick Alexander <nalexander@mozilla.com>
* Review comment: Don't preserve historical idents.
* Review comment: More prepared statements when updating materialized views.
* Post: Test altering :db/cardinality and :db/unique.
These tests fail due to a Datomic limitation, namely that the marker
flag :db.alter/attribute can only be asserted once for an attribute!
That is, [:db.part/db :db.alter/attribute :attribute] will only be
transacted at most once. Since older versions of Datomic required the
:db.alter/attribute flag, I can only imagine they either never wrote
:db.alter/attribute to the store, or they handled it specially. I'll
need to remove the marker flag system from Mentat in order to address
this fundamental limitation.
* Post: Remove some more single quotes from error output.
* Post: Add assert_transact! macro to unwrap safely.
I was finding it very difficult to track unwrapping errors while
making changes, due to an underlying Mac OS X symbolication issue that
makes running tests with RUST_BACKTRACE=1 so slow that they all time
out.
* Post: Don't expect or recognize :db.{install,alter}/attribute.
I had this all working... except we will never see a repeated
`[:db.part/db :db.alter/attribute :attribute]` assertion in the store!
That means my approach would let you alter an attribute at most one
time. It's not worth hacking around this; it's better to just stop
expecting (and recognizing) the marker flags. (We have all the data
to distinguish the various cases that we need without the marker
flags.)
This brings Mentat in line with the thrust of newer Datomic versions,
but isn't compatible with Datomic, because (if I understand correctly)
Datomic automatically adds :db.{install,alter}/attribute assertions to
transactions.
I haven't purged the corresponding :db/ident and schema fragments just
yet:
- we might want them back
- we might want them in order to upgrade v1 and v2 databases to the
new on-disk layout we're fleshing out (v3?).
* Post: Don't make :db/unique :db.unique/* imply :db/index true.
This patch avoids a potential bug with the "schema" materialized view.
If :db/unique :db.unique/value implies :db/index true, then what
happens when you _retract_ :db.unique/value? I think Datomic defines
this in some way, but I really want the "schema" materialized view to
be a slice of "datoms" and not have these sort of ambiguities and
persistent effects. Therefore, to ensure that we don't retract a
schema characteristic and accidentally change more than we intended
to, this patch stops having any schema characteristic imply any other
schema characteristic(s). To achieve that, I added an
Option<Unique::{Value,Identity}> type to Attribute; this helps with
this patch, and also looks ahead to when we allow to retract
:db/unique attributes.
* Post: Allow to retract :db/ident.
* Post: Include more details about invalid schema changes.
The tests use strings, so they hide the chained errors which do in
fact provide more detail.
* Review comment: Fix outdated comment.
* Review comment: s/_SET/_SQL_LIST/.
* Review comment: Use a sub-select for checking cardinality.
This might be faster in practice.
* Review comment: Put `attribute::Unique` into its own namespace.
2017-03-20 20:18:59 +00:00
[ :db / add 100 :db / valueType :db . type / long ]
2019-07-17 15:46:47 +00:00
[ :db / add 100 :db / cardinality :db . cardinality / many ] ] "
) ;
Schema alteration. Fixes #294 and #295. (#370) r=rnewman
* Pre: Don't retract :db/ident in test.
Datomic (and eventually Mentat) don't allow to retract :db/ident in
this way, so this runs afoul of future work to support mutating
metadata.
* Pre: s/VALUETYPE/VALUE_TYPE/.
This is consistent with the capitalization (which is "valueType") and
the other identifier.
* Pre: Remove some single quotes from error output.
* Part 1: Make materialized views be uniform [e a v value_type_tag].
This looks ahead to a time when we could support arbitrary
user-defined materialized views. For now, the "idents" materialized
view is those datoms of the form [e :db/ident :namespaced/keyword] and
the "schema" materialized view is those datoms of the form [e a v]
where a is in a particular set of attributes that will become clear in
the following commits.
This change is not backwards compatible, so I'm removing the open
current (really, v2) test. It'll be re-instated when we get to
https://github.com/mozilla/mentat/issues/194.
* Pre: Map TypedValue::Ref to TypedValue::Keyword in debug output.
* Part 3: Separate `schema_to_mutate` from the `schema` used to interpret.
This is just to keep track of the expected changes during
bootstrapping. I want bootstrap metadata mutations to flow through
the same code path as metadata mutations during regular transactions;
by differentiating the schema used for interpretation from the schema
that will be updated I expect to be able to apply bootstrap metadata
mutations to an empty schema and have things like materialized views
created (using the regular code paths).
This commit has been re-ordered for conceptual clarity, but it won't
compile because it references the metadata module. It's possible to
make it compile -- the functionality is there in the schema module --
but it's not worth the rebasing effort until after review (and
possibly not even then, since we'll squash down to a single commit to
land).
* Part 2: Maintain entids separately from idents.
In order to support historical idents, we need to distinguish the
"current" map from entid -> ident from the "complete historical" map
ident -> entid. This is what Datomic does; in Datomic, an ident is
never retracted (although it can be replaced). This approach is an
important part of allowing multiple consumers to share a schema
fragment as it migrates forward.
This fixes a limitation of the Clojure implementation, which did not
handle historical idents across knowledge base close and re-open.
The "entids" materialized view is naturally a slice of the "datoms"
table. The "idents" materialized view is a slice of the
"transactions" table. I hope that representing in this way, and
casting the problem in this light, might generalize to future
materialized views.
* Pre: Add DiffSet.
* Part 4: Collect mutations to a `Schema`.
I haven't taken your review comment about consuming AttributeBuilder
during each fluent function. If you read my response and still want
this, I'm happy to do it in review.
* Part 5: Handle :db/ident and :db.{install,alter}/attribute.
This "loops" the committed datoms out of the SQL store and back
through the metadata (schema, but in future also partition map)
processor. The metadata processor updates the schema and produces a
report of what changed; that report is then used to update the SQL
store. That update includes:
- the materialized views ("entids", "idents", and "schema");
- if needed, a subset of the datoms themselves (as flags change).
I've left a TODO for handling attribute retraction in the cases that
it makes sense. I expect that to be straight-forward.
* Review comment: Rename DiffSet to AddRetractAlterSet.
Also adds a little more commentary and a simple test.
* Review comment: Use ToIdent trait.
* Review comment: partially revert "Part 2: Maintain entids separately from idents."
This reverts commit 23a91df9c35e14398f2ddbd1ba25315821e67401.
Following our discussion, this removes the "entids" materialized
view. The next commit will remove historical idents from the "idents"
materialized view.
* Post: Use custom Either rather than std::result::Result.
This is not necessary, but it was suggested that we might be paying an
overhead creating Err instances while using error_chain. That seems
not to be the case, but this change shows that we don't actually use
any of the Result helper methods, so there's no reason to overload
Result. This change might avoid some future confusion, so I'm going
to land it anyway.
Signed-off-by: Nick Alexander <nalexander@mozilla.com>
* Review comment: Don't preserve historical idents.
* Review comment: More prepared statements when updating materialized views.
* Post: Test altering :db/cardinality and :db/unique.
These tests fail due to a Datomic limitation, namely that the marker
flag :db.alter/attribute can only be asserted once for an attribute!
That is, [:db.part/db :db.alter/attribute :attribute] will only be
transacted at most once. Since older versions of Datomic required the
:db.alter/attribute flag, I can only imagine they either never wrote
:db.alter/attribute to the store, or they handled it specially. I'll
need to remove the marker flag system from Mentat in order to address
this fundamental limitation.
* Post: Remove some more single quotes from error output.
* Post: Add assert_transact! macro to unwrap safely.
I was finding it very difficult to track unwrapping errors while
making changes, due to an underlying Mac OS X symbolication issue that
makes running tests with RUST_BACKTRACE=1 so slow that they all time
out.
* Post: Don't expect or recognize :db.{install,alter}/attribute.
I had this all working... except we will never see a repeated
`[:db.part/db :db.alter/attribute :attribute]` assertion in the store!
That means my approach would let you alter an attribute at most one
time. It's not worth hacking around this; it's better to just stop
expecting (and recognizing) the marker flags. (We have all the data
to distinguish the various cases that we need without the marker
flags.)
This brings Mentat in line with the thrust of newer Datomic versions,
but isn't compatible with Datomic, because (if I understand correctly)
Datomic automatically adds :db.{install,alter}/attribute assertions to
transactions.
I haven't purged the corresponding :db/ident and schema fragments just
yet:
- we might want them back
- we might want them in order to upgrade v1 and v2 databases to the
new on-disk layout we're fleshing out (v3?).
* Post: Don't make :db/unique :db.unique/* imply :db/index true.
This patch avoids a potential bug with the "schema" materialized view.
If :db/unique :db.unique/value implies :db/index true, then what
happens when you _retract_ :db.unique/value? I think Datomic defines
this in some way, but I really want the "schema" materialized view to
be a slice of "datoms" and not have these sort of ambiguities and
persistent effects. Therefore, to ensure that we don't retract a
schema characteristic and accidentally change more than we intended
to, this patch stops having any schema characteristic imply any other
schema characteristic(s). To achieve that, I added an
Option<Unique::{Value,Identity}> type to Attribute; this helps with
this patch, and also looks ahead to when we allow to retract
:db/unique attributes.
* Post: Allow to retract :db/ident.
* Post: Include more details about invalid schema changes.
The tests use strings, so they hide the chained errors which do in
fact provide more detail.
* Review comment: Fix outdated comment.
* Review comment: s/_SET/_SQL_LIST/.
* Review comment: Use a sub-select for checking cardinality.
This might be faster in practice.
* Review comment: Put `attribute::Unique` into its own namespace.
2017-03-20 20:18:59 +00:00
2019-07-17 15:46:47 +00:00
assert_eq! (
conn . schema . entid_map . get ( & 100 ) . cloned ( ) . unwrap ( ) ,
to_namespaced_keyword ( " :test/ident " ) . unwrap ( )
) ;
assert_eq! (
conn . schema
. ident_map
. get ( & to_namespaced_keyword ( " :test/ident " ) . unwrap ( ) )
. cloned ( )
. unwrap ( ) ,
100
) ;
Schema alteration. Fixes #294 and #295. (#370) r=rnewman
* Pre: Don't retract :db/ident in test.
Datomic (and eventually Mentat) don't allow to retract :db/ident in
this way, so this runs afoul of future work to support mutating
metadata.
* Pre: s/VALUETYPE/VALUE_TYPE/.
This is consistent with the capitalization (which is "valueType") and
the other identifier.
* Pre: Remove some single quotes from error output.
* Part 1: Make materialized views be uniform [e a v value_type_tag].
This looks ahead to a time when we could support arbitrary
user-defined materialized views. For now, the "idents" materialized
view is those datoms of the form [e :db/ident :namespaced/keyword] and
the "schema" materialized view is those datoms of the form [e a v]
where a is in a particular set of attributes that will become clear in
the following commits.
This change is not backwards compatible, so I'm removing the open
current (really, v2) test. It'll be re-instated when we get to
https://github.com/mozilla/mentat/issues/194.
* Pre: Map TypedValue::Ref to TypedValue::Keyword in debug output.
* Part 3: Separate `schema_to_mutate` from the `schema` used to interpret.
This is just to keep track of the expected changes during
bootstrapping. I want bootstrap metadata mutations to flow through
the same code path as metadata mutations during regular transactions;
by differentiating the schema used for interpretation from the schema
that will be updated I expect to be able to apply bootstrap metadata
mutations to an empty schema and have things like materialized views
created (using the regular code paths).
This commit has been re-ordered for conceptual clarity, but it won't
compile because it references the metadata module. It's possible to
make it compile -- the functionality is there in the schema module --
but it's not worth the rebasing effort until after review (and
possibly not even then, since we'll squash down to a single commit to
land).
* Part 2: Maintain entids separately from idents.
In order to support historical idents, we need to distinguish the
"current" map from entid -> ident from the "complete historical" map
ident -> entid. This is what Datomic does; in Datomic, an ident is
never retracted (although it can be replaced). This approach is an
important part of allowing multiple consumers to share a schema
fragment as it migrates forward.
This fixes a limitation of the Clojure implementation, which did not
handle historical idents across knowledge base close and re-open.
The "entids" materialized view is naturally a slice of the "datoms"
table. The "idents" materialized view is a slice of the
"transactions" table. I hope that representing in this way, and
casting the problem in this light, might generalize to future
materialized views.
* Pre: Add DiffSet.
* Part 4: Collect mutations to a `Schema`.
I haven't taken your review comment about consuming AttributeBuilder
during each fluent function. If you read my response and still want
this, I'm happy to do it in review.
* Part 5: Handle :db/ident and :db.{install,alter}/attribute.
This "loops" the committed datoms out of the SQL store and back
through the metadata (schema, but in future also partition map)
processor. The metadata processor updates the schema and produces a
report of what changed; that report is then used to update the SQL
store. That update includes:
- the materialized views ("entids", "idents", and "schema");
- if needed, a subset of the datoms themselves (as flags change).
I've left a TODO for handling attribute retraction in the cases that
it makes sense. I expect that to be straight-forward.
* Review comment: Rename DiffSet to AddRetractAlterSet.
Also adds a little more commentary and a simple test.
* Review comment: Use ToIdent trait.
* Review comment: partially revert "Part 2: Maintain entids separately from idents."
This reverts commit 23a91df9c35e14398f2ddbd1ba25315821e67401.
Following our discussion, this removes the "entids" materialized
view. The next commit will remove historical idents from the "idents"
materialized view.
* Post: Use custom Either rather than std::result::Result.
This is not necessary, but it was suggested that we might be paying an
overhead creating Err instances while using error_chain. That seems
not to be the case, but this change shows that we don't actually use
any of the Result helper methods, so there's no reason to overload
Result. This change might avoid some future confusion, so I'm going
to land it anyway.
Signed-off-by: Nick Alexander <nalexander@mozilla.com>
* Review comment: Don't preserve historical idents.
* Review comment: More prepared statements when updating materialized views.
* Post: Test altering :db/cardinality and :db/unique.
These tests fail due to a Datomic limitation, namely that the marker
flag :db.alter/attribute can only be asserted once for an attribute!
That is, [:db.part/db :db.alter/attribute :attribute] will only be
transacted at most once. Since older versions of Datomic required the
:db.alter/attribute flag, I can only imagine they either never wrote
:db.alter/attribute to the store, or they handled it specially. I'll
need to remove the marker flag system from Mentat in order to address
this fundamental limitation.
* Post: Remove some more single quotes from error output.
* Post: Add assert_transact! macro to unwrap safely.
I was finding it very difficult to track unwrapping errors while
making changes, due to an underlying Mac OS X symbolication issue that
makes running tests with RUST_BACKTRACE=1 so slow that they all time
out.
* Post: Don't expect or recognize :db.{install,alter}/attribute.
I had this all working... except we will never see a repeated
`[:db.part/db :db.alter/attribute :attribute]` assertion in the store!
That means my approach would let you alter an attribute at most one
time. It's not worth hacking around this; it's better to just stop
expecting (and recognizing) the marker flags. (We have all the data
to distinguish the various cases that we need without the marker
flags.)
This brings Mentat in line with the thrust of newer Datomic versions,
but isn't compatible with Datomic, because (if I understand correctly)
Datomic automatically adds :db.{install,alter}/attribute assertions to
transactions.
I haven't purged the corresponding :db/ident and schema fragments just
yet:
- we might want them back
- we might want them in order to upgrade v1 and v2 databases to the
new on-disk layout we're fleshing out (v3?).
* Post: Don't make :db/unique :db.unique/* imply :db/index true.
This patch avoids a potential bug with the "schema" materialized view.
If :db/unique :db.unique/value implies :db/index true, then what
happens when you _retract_ :db.unique/value? I think Datomic defines
this in some way, but I really want the "schema" materialized view to
be a slice of "datoms" and not have these sort of ambiguities and
persistent effects. Therefore, to ensure that we don't retract a
schema characteristic and accidentally change more than we intended
to, this patch stops having any schema characteristic imply any other
schema characteristic(s). To achieve that, I added an
Option<Unique::{Value,Identity}> type to Attribute; this helps with
this patch, and also looks ahead to when we allow to retract
:db/unique attributes.
* Post: Allow to retract :db/ident.
* Post: Include more details about invalid schema changes.
The tests use strings, so they hide the chained errors which do in
fact provide more detail.
* Review comment: Fix outdated comment.
* Review comment: s/_SET/_SQL_LIST/.
* Review comment: Use a sub-select for checking cardinality.
This might be faster in practice.
* Review comment: Put `attribute::Unique` into its own namespace.
2017-03-20 20:18:59 +00:00
let attribute = conn . schema . attribute_for_entid ( 100 ) . unwrap ( ) . clone ( ) ;
assert_eq! ( attribute . value_type , ValueType ::Long ) ;
assert_eq! ( attribute . multival , true ) ;
assert_eq! ( attribute . fulltext , false ) ;
2019-07-17 15:46:47 +00:00
assert_matches! (
conn . last_transaction ( ) ,
" [[100 :db/ident :test/ident ?tx true]
Schema alteration. Fixes #294 and #295. (#370) r=rnewman
* Pre: Don't retract :db/ident in test.
Datomic (and eventually Mentat) don't allow to retract :db/ident in
this way, so this runs afoul of future work to support mutating
metadata.
* Pre: s/VALUETYPE/VALUE_TYPE/.
This is consistent with the capitalization (which is "valueType") and
the other identifier.
* Pre: Remove some single quotes from error output.
* Part 1: Make materialized views be uniform [e a v value_type_tag].
This looks ahead to a time when we could support arbitrary
user-defined materialized views. For now, the "idents" materialized
view is those datoms of the form [e :db/ident :namespaced/keyword] and
the "schema" materialized view is those datoms of the form [e a v]
where a is in a particular set of attributes that will become clear in
the following commits.
This change is not backwards compatible, so I'm removing the open
current (really, v2) test. It'll be re-instated when we get to
https://github.com/mozilla/mentat/issues/194.
* Pre: Map TypedValue::Ref to TypedValue::Keyword in debug output.
* Part 3: Separate `schema_to_mutate` from the `schema` used to interpret.
This is just to keep track of the expected changes during
bootstrapping. I want bootstrap metadata mutations to flow through
the same code path as metadata mutations during regular transactions;
by differentiating the schema used for interpretation from the schema
that will be updated I expect to be able to apply bootstrap metadata
mutations to an empty schema and have things like materialized views
created (using the regular code paths).
This commit has been re-ordered for conceptual clarity, but it won't
compile because it references the metadata module. It's possible to
make it compile -- the functionality is there in the schema module --
but it's not worth the rebasing effort until after review (and
possibly not even then, since we'll squash down to a single commit to
land).
* Part 2: Maintain entids separately from idents.
In order to support historical idents, we need to distinguish the
"current" map from entid -> ident from the "complete historical" map
ident -> entid. This is what Datomic does; in Datomic, an ident is
never retracted (although it can be replaced). This approach is an
important part of allowing multiple consumers to share a schema
fragment as it migrates forward.
This fixes a limitation of the Clojure implementation, which did not
handle historical idents across knowledge base close and re-open.
The "entids" materialized view is naturally a slice of the "datoms"
table. The "idents" materialized view is a slice of the
"transactions" table. I hope that representing in this way, and
casting the problem in this light, might generalize to future
materialized views.
* Pre: Add DiffSet.
* Part 4: Collect mutations to a `Schema`.
I haven't taken your review comment about consuming AttributeBuilder
during each fluent function. If you read my response and still want
this, I'm happy to do it in review.
* Part 5: Handle :db/ident and :db.{install,alter}/attribute.
This "loops" the committed datoms out of the SQL store and back
through the metadata (schema, but in future also partition map)
processor. The metadata processor updates the schema and produces a
report of what changed; that report is then used to update the SQL
store. That update includes:
- the materialized views ("entids", "idents", and "schema");
- if needed, a subset of the datoms themselves (as flags change).
I've left a TODO for handling attribute retraction in the cases that
it makes sense. I expect that to be straight-forward.
* Review comment: Rename DiffSet to AddRetractAlterSet.
Also adds a little more commentary and a simple test.
* Review comment: Use ToIdent trait.
* Review comment: partially revert "Part 2: Maintain entids separately from idents."
This reverts commit 23a91df9c35e14398f2ddbd1ba25315821e67401.
Following our discussion, this removes the "entids" materialized
view. The next commit will remove historical idents from the "idents"
materialized view.
* Post: Use custom Either rather than std::result::Result.
This is not necessary, but it was suggested that we might be paying an
overhead creating Err instances while using error_chain. That seems
not to be the case, but this change shows that we don't actually use
any of the Result helper methods, so there's no reason to overload
Result. This change might avoid some future confusion, so I'm going
to land it anyway.
Signed-off-by: Nick Alexander <nalexander@mozilla.com>
* Review comment: Don't preserve historical idents.
* Review comment: More prepared statements when updating materialized views.
* Post: Test altering :db/cardinality and :db/unique.
These tests fail due to a Datomic limitation, namely that the marker
flag :db.alter/attribute can only be asserted once for an attribute!
That is, [:db.part/db :db.alter/attribute :attribute] will only be
transacted at most once. Since older versions of Datomic required the
:db.alter/attribute flag, I can only imagine they either never wrote
:db.alter/attribute to the store, or they handled it specially. I'll
need to remove the marker flag system from Mentat in order to address
this fundamental limitation.
* Post: Remove some more single quotes from error output.
* Post: Add assert_transact! macro to unwrap safely.
I was finding it very difficult to track unwrapping errors while
making changes, due to an underlying Mac OS X symbolication issue that
makes running tests with RUST_BACKTRACE=1 so slow that they all time
out.
* Post: Don't expect or recognize :db.{install,alter}/attribute.
I had this all working... except we will never see a repeated
`[:db.part/db :db.alter/attribute :attribute]` assertion in the store!
That means my approach would let you alter an attribute at most one
time. It's not worth hacking around this; it's better to just stop
expecting (and recognizing) the marker flags. (We have all the data
to distinguish the various cases that we need without the marker
flags.)
This brings Mentat in line with the thrust of newer Datomic versions,
but isn't compatible with Datomic, because (if I understand correctly)
Datomic automatically adds :db.{install,alter}/attribute assertions to
transactions.
I haven't purged the corresponding :db/ident and schema fragments just
yet:
- we might want them back
- we might want them in order to upgrade v1 and v2 databases to the
new on-disk layout we're fleshing out (v3?).
* Post: Don't make :db/unique :db.unique/* imply :db/index true.
This patch avoids a potential bug with the "schema" materialized view.
If :db/unique :db.unique/value implies :db/index true, then what
happens when you _retract_ :db.unique/value? I think Datomic defines
this in some way, but I really want the "schema" materialized view to
be a slice of "datoms" and not have these sort of ambiguities and
persistent effects. Therefore, to ensure that we don't retract a
schema characteristic and accidentally change more than we intended
to, this patch stops having any schema characteristic imply any other
schema characteristic(s). To achieve that, I added an
Option<Unique::{Value,Identity}> type to Attribute; this helps with
this patch, and also looks ahead to when we allow to retract
:db/unique attributes.
* Post: Allow to retract :db/ident.
* Post: Include more details about invalid schema changes.
The tests use strings, so they hide the chained errors which do in
fact provide more detail.
* Review comment: Fix outdated comment.
* Review comment: s/_SET/_SQL_LIST/.
* Review comment: Use a sub-select for checking cardinality.
This might be faster in practice.
* Review comment: Put `attribute::Unique` into its own namespace.
2017-03-20 20:18:59 +00:00
[ 100 :db / valueType :db . type / long ? tx true ]
[ 100 :db / cardinality :db . cardinality / many ? tx true ]
2019-07-17 15:46:47 +00:00
[ ? tx :db / txInstant ? ms ? tx true ] ] "
) ;
assert_matches! (
conn . datoms ( ) ,
" [[100 :db/ident :test/ident]
Schema alteration. Fixes #294 and #295. (#370) r=rnewman
* Pre: Don't retract :db/ident in test.
Datomic (and eventually Mentat) don't allow to retract :db/ident in
this way, so this runs afoul of future work to support mutating
metadata.
* Pre: s/VALUETYPE/VALUE_TYPE/.
This is consistent with the capitalization (which is "valueType") and
the other identifier.
* Pre: Remove some single quotes from error output.
* Part 1: Make materialized views be uniform [e a v value_type_tag].
This looks ahead to a time when we could support arbitrary
user-defined materialized views. For now, the "idents" materialized
view is those datoms of the form [e :db/ident :namespaced/keyword] and
the "schema" materialized view is those datoms of the form [e a v]
where a is in a particular set of attributes that will become clear in
the following commits.
This change is not backwards compatible, so I'm removing the open
current (really, v2) test. It'll be re-instated when we get to
https://github.com/mozilla/mentat/issues/194.
* Pre: Map TypedValue::Ref to TypedValue::Keyword in debug output.
* Part 3: Separate `schema_to_mutate` from the `schema` used to interpret.
This is just to keep track of the expected changes during
bootstrapping. I want bootstrap metadata mutations to flow through
the same code path as metadata mutations during regular transactions;
by differentiating the schema used for interpretation from the schema
that will be updated I expect to be able to apply bootstrap metadata
mutations to an empty schema and have things like materialized views
created (using the regular code paths).
This commit has been re-ordered for conceptual clarity, but it won't
compile because it references the metadata module. It's possible to
make it compile -- the functionality is there in the schema module --
but it's not worth the rebasing effort until after review (and
possibly not even then, since we'll squash down to a single commit to
land).
* Part 2: Maintain entids separately from idents.
In order to support historical idents, we need to distinguish the
"current" map from entid -> ident from the "complete historical" map
ident -> entid. This is what Datomic does; in Datomic, an ident is
never retracted (although it can be replaced). This approach is an
important part of allowing multiple consumers to share a schema
fragment as it migrates forward.
This fixes a limitation of the Clojure implementation, which did not
handle historical idents across knowledge base close and re-open.
The "entids" materialized view is naturally a slice of the "datoms"
table. The "idents" materialized view is a slice of the
"transactions" table. I hope that representing in this way, and
casting the problem in this light, might generalize to future
materialized views.
* Pre: Add DiffSet.
* Part 4: Collect mutations to a `Schema`.
I haven't taken your review comment about consuming AttributeBuilder
during each fluent function. If you read my response and still want
this, I'm happy to do it in review.
* Part 5: Handle :db/ident and :db.{install,alter}/attribute.
This "loops" the committed datoms out of the SQL store and back
through the metadata (schema, but in future also partition map)
processor. The metadata processor updates the schema and produces a
report of what changed; that report is then used to update the SQL
store. That update includes:
- the materialized views ("entids", "idents", and "schema");
- if needed, a subset of the datoms themselves (as flags change).
I've left a TODO for handling attribute retraction in the cases that
it makes sense. I expect that to be straight-forward.
* Review comment: Rename DiffSet to AddRetractAlterSet.
Also adds a little more commentary and a simple test.
* Review comment: Use ToIdent trait.
* Review comment: partially revert "Part 2: Maintain entids separately from idents."
This reverts commit 23a91df9c35e14398f2ddbd1ba25315821e67401.
Following our discussion, this removes the "entids" materialized
view. The next commit will remove historical idents from the "idents"
materialized view.
* Post: Use custom Either rather than std::result::Result.
This is not necessary, but it was suggested that we might be paying an
overhead creating Err instances while using error_chain. That seems
not to be the case, but this change shows that we don't actually use
any of the Result helper methods, so there's no reason to overload
Result. This change might avoid some future confusion, so I'm going
to land it anyway.
Signed-off-by: Nick Alexander <nalexander@mozilla.com>
* Review comment: Don't preserve historical idents.
* Review comment: More prepared statements when updating materialized views.
* Post: Test altering :db/cardinality and :db/unique.
These tests fail due to a Datomic limitation, namely that the marker
flag :db.alter/attribute can only be asserted once for an attribute!
That is, [:db.part/db :db.alter/attribute :attribute] will only be
transacted at most once. Since older versions of Datomic required the
:db.alter/attribute flag, I can only imagine they either never wrote
:db.alter/attribute to the store, or they handled it specially. I'll
need to remove the marker flag system from Mentat in order to address
this fundamental limitation.
* Post: Remove some more single quotes from error output.
* Post: Add assert_transact! macro to unwrap safely.
I was finding it very difficult to track unwrapping errors while
making changes, due to an underlying Mac OS X symbolication issue that
makes running tests with RUST_BACKTRACE=1 so slow that they all time
out.
* Post: Don't expect or recognize :db.{install,alter}/attribute.
I had this all working... except we will never see a repeated
`[:db.part/db :db.alter/attribute :attribute]` assertion in the store!
That means my approach would let you alter an attribute at most one
time. It's not worth hacking around this; it's better to just stop
expecting (and recognizing) the marker flags. (We have all the data
to distinguish the various cases that we need without the marker
flags.)
This brings Mentat in line with the thrust of newer Datomic versions,
but isn't compatible with Datomic, because (if I understand correctly)
Datomic automatically adds :db.{install,alter}/attribute assertions to
transactions.
I haven't purged the corresponding :db/ident and schema fragments just
yet:
- we might want them back
- we might want them in order to upgrade v1 and v2 databases to the
new on-disk layout we're fleshing out (v3?).
* Post: Don't make :db/unique :db.unique/* imply :db/index true.
This patch avoids a potential bug with the "schema" materialized view.
If :db/unique :db.unique/value implies :db/index true, then what
happens when you _retract_ :db.unique/value? I think Datomic defines
this in some way, but I really want the "schema" materialized view to
be a slice of "datoms" and not have these sort of ambiguities and
persistent effects. Therefore, to ensure that we don't retract a
schema characteristic and accidentally change more than we intended
to, this patch stops having any schema characteristic imply any other
schema characteristic(s). To achieve that, I added an
Option<Unique::{Value,Identity}> type to Attribute; this helps with
this patch, and also looks ahead to when we allow to retract
:db/unique attributes.
* Post: Allow to retract :db/ident.
* Post: Include more details about invalid schema changes.
The tests use strings, so they hide the chained errors which do in
fact provide more detail.
* Review comment: Fix outdated comment.
* Review comment: s/_SET/_SQL_LIST/.
* Review comment: Use a sub-select for checking cardinality.
This might be faster in practice.
* Review comment: Put `attribute::Unique` into its own namespace.
2017-03-20 20:18:59 +00:00
[ 100 :db / valueType :db . type / long ]
2019-07-17 15:46:47 +00:00
[ 100 :db / cardinality :db . cardinality / many ] ] "
) ;
Schema alteration. Fixes #294 and #295. (#370) r=rnewman
* Pre: Don't retract :db/ident in test.
Datomic (and eventually Mentat) don't allow to retract :db/ident in
this way, so this runs afoul of future work to support mutating
metadata.
* Pre: s/VALUETYPE/VALUE_TYPE/.
This is consistent with the capitalization (which is "valueType") and
the other identifier.
* Pre: Remove some single quotes from error output.
* Part 1: Make materialized views be uniform [e a v value_type_tag].
This looks ahead to a time when we could support arbitrary
user-defined materialized views. For now, the "idents" materialized
view is those datoms of the form [e :db/ident :namespaced/keyword] and
the "schema" materialized view is those datoms of the form [e a v]
where a is in a particular set of attributes that will become clear in
the following commits.
This change is not backwards compatible, so I'm removing the open
current (really, v2) test. It'll be re-instated when we get to
https://github.com/mozilla/mentat/issues/194.
* Pre: Map TypedValue::Ref to TypedValue::Keyword in debug output.
* Part 3: Separate `schema_to_mutate` from the `schema` used to interpret.
This is just to keep track of the expected changes during
bootstrapping. I want bootstrap metadata mutations to flow through
the same code path as metadata mutations during regular transactions;
by differentiating the schema used for interpretation from the schema
that will be updated I expect to be able to apply bootstrap metadata
mutations to an empty schema and have things like materialized views
created (using the regular code paths).
This commit has been re-ordered for conceptual clarity, but it won't
compile because it references the metadata module. It's possible to
make it compile -- the functionality is there in the schema module --
but it's not worth the rebasing effort until after review (and
possibly not even then, since we'll squash down to a single commit to
land).
* Part 2: Maintain entids separately from idents.
In order to support historical idents, we need to distinguish the
"current" map from entid -> ident from the "complete historical" map
ident -> entid. This is what Datomic does; in Datomic, an ident is
never retracted (although it can be replaced). This approach is an
important part of allowing multiple consumers to share a schema
fragment as it migrates forward.
This fixes a limitation of the Clojure implementation, which did not
handle historical idents across knowledge base close and re-open.
The "entids" materialized view is naturally a slice of the "datoms"
table. The "idents" materialized view is a slice of the
"transactions" table. I hope that representing in this way, and
casting the problem in this light, might generalize to future
materialized views.
* Pre: Add DiffSet.
* Part 4: Collect mutations to a `Schema`.
I haven't taken your review comment about consuming AttributeBuilder
during each fluent function. If you read my response and still want
this, I'm happy to do it in review.
* Part 5: Handle :db/ident and :db.{install,alter}/attribute.
This "loops" the committed datoms out of the SQL store and back
through the metadata (schema, but in future also partition map)
processor. The metadata processor updates the schema and produces a
report of what changed; that report is then used to update the SQL
store. That update includes:
- the materialized views ("entids", "idents", and "schema");
- if needed, a subset of the datoms themselves (as flags change).
I've left a TODO for handling attribute retraction in the cases that
it makes sense. I expect that to be straight-forward.
* Review comment: Rename DiffSet to AddRetractAlterSet.
Also adds a little more commentary and a simple test.
* Review comment: Use ToIdent trait.
* Review comment: partially revert "Part 2: Maintain entids separately from idents."
This reverts commit 23a91df9c35e14398f2ddbd1ba25315821e67401.
Following our discussion, this removes the "entids" materialized
view. The next commit will remove historical idents from the "idents"
materialized view.
* Post: Use custom Either rather than std::result::Result.
This is not necessary, but it was suggested that we might be paying an
overhead creating Err instances while using error_chain. That seems
not to be the case, but this change shows that we don't actually use
any of the Result helper methods, so there's no reason to overload
Result. This change might avoid some future confusion, so I'm going
to land it anyway.
Signed-off-by: Nick Alexander <nalexander@mozilla.com>
* Review comment: Don't preserve historical idents.
* Review comment: More prepared statements when updating materialized views.
* Post: Test altering :db/cardinality and :db/unique.
These tests fail due to a Datomic limitation, namely that the marker
flag :db.alter/attribute can only be asserted once for an attribute!
That is, [:db.part/db :db.alter/attribute :attribute] will only be
transacted at most once. Since older versions of Datomic required the
:db.alter/attribute flag, I can only imagine they either never wrote
:db.alter/attribute to the store, or they handled it specially. I'll
need to remove the marker flag system from Mentat in order to address
this fundamental limitation.
* Post: Remove some more single quotes from error output.
* Post: Add assert_transact! macro to unwrap safely.
I was finding it very difficult to track unwrapping errors while
making changes, due to an underlying Mac OS X symbolication issue that
makes running tests with RUST_BACKTRACE=1 so slow that they all time
out.
* Post: Don't expect or recognize :db.{install,alter}/attribute.
I had this all working... except we will never see a repeated
`[:db.part/db :db.alter/attribute :attribute]` assertion in the store!
That means my approach would let you alter an attribute at most one
time. It's not worth hacking around this; it's better to just stop
expecting (and recognizing) the marker flags. (We have all the data
to distinguish the various cases that we need without the marker
flags.)
This brings Mentat in line with the thrust of newer Datomic versions,
but isn't compatible with Datomic, because (if I understand correctly)
Datomic automatically adds :db.{install,alter}/attribute assertions to
transactions.
I haven't purged the corresponding :db/ident and schema fragments just
yet:
- we might want them back
- we might want them in order to upgrade v1 and v2 databases to the
new on-disk layout we're fleshing out (v3?).
* Post: Don't make :db/unique :db.unique/* imply :db/index true.
This patch avoids a potential bug with the "schema" materialized view.
If :db/unique :db.unique/value implies :db/index true, then what
happens when you _retract_ :db.unique/value? I think Datomic defines
this in some way, but I really want the "schema" materialized view to
be a slice of "datoms" and not have these sort of ambiguities and
persistent effects. Therefore, to ensure that we don't retract a
schema characteristic and accidentally change more than we intended
to, this patch stops having any schema characteristic imply any other
schema characteristic(s). To achieve that, I added an
Option<Unique::{Value,Identity}> type to Attribute; this helps with
this patch, and also looks ahead to when we allow to retract
:db/unique attributes.
* Post: Allow to retract :db/ident.
* Post: Include more details about invalid schema changes.
The tests use strings, so they hide the chained errors which do in
fact provide more detail.
* Review comment: Fix outdated comment.
* Review comment: s/_SET/_SQL_LIST/.
* Review comment: Use a sub-select for checking cardinality.
This might be faster in practice.
* Review comment: Put `attribute::Unique` into its own namespace.
2017-03-20 20:18:59 +00:00
// Let's check we actually have the schema characteristics we expect.
let attribute = conn . schema . attribute_for_entid ( 100 ) . unwrap ( ) . clone ( ) ;
assert_eq! ( attribute . value_type , ValueType ::Long ) ;
assert_eq! ( attribute . multival , true ) ;
assert_eq! ( attribute . fulltext , false ) ;
// Let's check that we can use the freshly installed attribute.
2019-07-17 15:46:47 +00:00
assert_transact! (
conn ,
" [[:db/add 101 100 -10]
[ :db / add 101 :test / ident - 9 ] ] "
) ;
Schema alteration. Fixes #294 and #295. (#370) r=rnewman
* Pre: Don't retract :db/ident in test.
Datomic (and eventually Mentat) don't allow to retract :db/ident in
this way, so this runs afoul of future work to support mutating
metadata.
* Pre: s/VALUETYPE/VALUE_TYPE/.
This is consistent with the capitalization (which is "valueType") and
the other identifier.
* Pre: Remove some single quotes from error output.
* Part 1: Make materialized views be uniform [e a v value_type_tag].
This looks ahead to a time when we could support arbitrary
user-defined materialized views. For now, the "idents" materialized
view is those datoms of the form [e :db/ident :namespaced/keyword] and
the "schema" materialized view is those datoms of the form [e a v]
where a is in a particular set of attributes that will become clear in
the following commits.
This change is not backwards compatible, so I'm removing the open
current (really, v2) test. It'll be re-instated when we get to
https://github.com/mozilla/mentat/issues/194.
* Pre: Map TypedValue::Ref to TypedValue::Keyword in debug output.
* Part 3: Separate `schema_to_mutate` from the `schema` used to interpret.
This is just to keep track of the expected changes during
bootstrapping. I want bootstrap metadata mutations to flow through
the same code path as metadata mutations during regular transactions;
by differentiating the schema used for interpretation from the schema
that will be updated I expect to be able to apply bootstrap metadata
mutations to an empty schema and have things like materialized views
created (using the regular code paths).
This commit has been re-ordered for conceptual clarity, but it won't
compile because it references the metadata module. It's possible to
make it compile -- the functionality is there in the schema module --
but it's not worth the rebasing effort until after review (and
possibly not even then, since we'll squash down to a single commit to
land).
* Part 2: Maintain entids separately from idents.
In order to support historical idents, we need to distinguish the
"current" map from entid -> ident from the "complete historical" map
ident -> entid. This is what Datomic does; in Datomic, an ident is
never retracted (although it can be replaced). This approach is an
important part of allowing multiple consumers to share a schema
fragment as it migrates forward.
This fixes a limitation of the Clojure implementation, which did not
handle historical idents across knowledge base close and re-open.
The "entids" materialized view is naturally a slice of the "datoms"
table. The "idents" materialized view is a slice of the
"transactions" table. I hope that representing in this way, and
casting the problem in this light, might generalize to future
materialized views.
* Pre: Add DiffSet.
* Part 4: Collect mutations to a `Schema`.
I haven't taken your review comment about consuming AttributeBuilder
during each fluent function. If you read my response and still want
this, I'm happy to do it in review.
* Part 5: Handle :db/ident and :db.{install,alter}/attribute.
This "loops" the committed datoms out of the SQL store and back
through the metadata (schema, but in future also partition map)
processor. The metadata processor updates the schema and produces a
report of what changed; that report is then used to update the SQL
store. That update includes:
- the materialized views ("entids", "idents", and "schema");
- if needed, a subset of the datoms themselves (as flags change).
I've left a TODO for handling attribute retraction in the cases that
it makes sense. I expect that to be straight-forward.
* Review comment: Rename DiffSet to AddRetractAlterSet.
Also adds a little more commentary and a simple test.
* Review comment: Use ToIdent trait.
* Review comment: partially revert "Part 2: Maintain entids separately from idents."
This reverts commit 23a91df9c35e14398f2ddbd1ba25315821e67401.
Following our discussion, this removes the "entids" materialized
view. The next commit will remove historical idents from the "idents"
materialized view.
* Post: Use custom Either rather than std::result::Result.
This is not necessary, but it was suggested that we might be paying an
overhead creating Err instances while using error_chain. That seems
not to be the case, but this change shows that we don't actually use
any of the Result helper methods, so there's no reason to overload
Result. This change might avoid some future confusion, so I'm going
to land it anyway.
Signed-off-by: Nick Alexander <nalexander@mozilla.com>
* Review comment: Don't preserve historical idents.
* Review comment: More prepared statements when updating materialized views.
* Post: Test altering :db/cardinality and :db/unique.
These tests fail due to a Datomic limitation, namely that the marker
flag :db.alter/attribute can only be asserted once for an attribute!
That is, [:db.part/db :db.alter/attribute :attribute] will only be
transacted at most once. Since older versions of Datomic required the
:db.alter/attribute flag, I can only imagine they either never wrote
:db.alter/attribute to the store, or they handled it specially. I'll
need to remove the marker flag system from Mentat in order to address
this fundamental limitation.
* Post: Remove some more single quotes from error output.
* Post: Add assert_transact! macro to unwrap safely.
I was finding it very difficult to track unwrapping errors while
making changes, due to an underlying Mac OS X symbolication issue that
makes running tests with RUST_BACKTRACE=1 so slow that they all time
out.
* Post: Don't expect or recognize :db.{install,alter}/attribute.
I had this all working... except we will never see a repeated
`[:db.part/db :db.alter/attribute :attribute]` assertion in the store!
That means my approach would let you alter an attribute at most one
time. It's not worth hacking around this; it's better to just stop
expecting (and recognizing) the marker flags. (We have all the data
to distinguish the various cases that we need without the marker
flags.)
This brings Mentat in line with the thrust of newer Datomic versions,
but isn't compatible with Datomic, because (if I understand correctly)
Datomic automatically adds :db.{install,alter}/attribute assertions to
transactions.
I haven't purged the corresponding :db/ident and schema fragments just
yet:
- we might want them back
- we might want them in order to upgrade v1 and v2 databases to the
new on-disk layout we're fleshing out (v3?).
* Post: Don't make :db/unique :db.unique/* imply :db/index true.
This patch avoids a potential bug with the "schema" materialized view.
If :db/unique :db.unique/value implies :db/index true, then what
happens when you _retract_ :db.unique/value? I think Datomic defines
this in some way, but I really want the "schema" materialized view to
be a slice of "datoms" and not have these sort of ambiguities and
persistent effects. Therefore, to ensure that we don't retract a
schema characteristic and accidentally change more than we intended
to, this patch stops having any schema characteristic imply any other
schema characteristic(s). To achieve that, I added an
Option<Unique::{Value,Identity}> type to Attribute; this helps with
this patch, and also looks ahead to when we allow to retract
:db/unique attributes.
* Post: Allow to retract :db/ident.
* Post: Include more details about invalid schema changes.
The tests use strings, so they hide the chained errors which do in
fact provide more detail.
* Review comment: Fix outdated comment.
* Review comment: s/_SET/_SQL_LIST/.
* Review comment: Use a sub-select for checking cardinality.
This might be faster in practice.
* Review comment: Put `attribute::Unique` into its own namespace.
2017-03-20 20:18:59 +00:00
2019-07-17 15:46:47 +00:00
assert_matches! (
conn . last_transaction ( ) ,
" [[101 :test/ident -10 ?tx true]
Schema alteration. Fixes #294 and #295. (#370) r=rnewman
* Pre: Don't retract :db/ident in test.
Datomic (and eventually Mentat) don't allow to retract :db/ident in
this way, so this runs afoul of future work to support mutating
metadata.
* Pre: s/VALUETYPE/VALUE_TYPE/.
This is consistent with the capitalization (which is "valueType") and
the other identifier.
* Pre: Remove some single quotes from error output.
* Part 1: Make materialized views be uniform [e a v value_type_tag].
This looks ahead to a time when we could support arbitrary
user-defined materialized views. For now, the "idents" materialized
view is those datoms of the form [e :db/ident :namespaced/keyword] and
the "schema" materialized view is those datoms of the form [e a v]
where a is in a particular set of attributes that will become clear in
the following commits.
This change is not backwards compatible, so I'm removing the open
current (really, v2) test. It'll be re-instated when we get to
https://github.com/mozilla/mentat/issues/194.
* Pre: Map TypedValue::Ref to TypedValue::Keyword in debug output.
* Part 3: Separate `schema_to_mutate` from the `schema` used to interpret.
This is just to keep track of the expected changes during
bootstrapping. I want bootstrap metadata mutations to flow through
the same code path as metadata mutations during regular transactions;
by differentiating the schema used for interpretation from the schema
that will be updated I expect to be able to apply bootstrap metadata
mutations to an empty schema and have things like materialized views
created (using the regular code paths).
This commit has been re-ordered for conceptual clarity, but it won't
compile because it references the metadata module. It's possible to
make it compile -- the functionality is there in the schema module --
but it's not worth the rebasing effort until after review (and
possibly not even then, since we'll squash down to a single commit to
land).
* Part 2: Maintain entids separately from idents.
In order to support historical idents, we need to distinguish the
"current" map from entid -> ident from the "complete historical" map
ident -> entid. This is what Datomic does; in Datomic, an ident is
never retracted (although it can be replaced). This approach is an
important part of allowing multiple consumers to share a schema
fragment as it migrates forward.
This fixes a limitation of the Clojure implementation, which did not
handle historical idents across knowledge base close and re-open.
The "entids" materialized view is naturally a slice of the "datoms"
table. The "idents" materialized view is a slice of the
"transactions" table. I hope that representing in this way, and
casting the problem in this light, might generalize to future
materialized views.
* Pre: Add DiffSet.
* Part 4: Collect mutations to a `Schema`.
I haven't taken your review comment about consuming AttributeBuilder
during each fluent function. If you read my response and still want
this, I'm happy to do it in review.
* Part 5: Handle :db/ident and :db.{install,alter}/attribute.
This "loops" the committed datoms out of the SQL store and back
through the metadata (schema, but in future also partition map)
processor. The metadata processor updates the schema and produces a
report of what changed; that report is then used to update the SQL
store. That update includes:
- the materialized views ("entids", "idents", and "schema");
- if needed, a subset of the datoms themselves (as flags change).
I've left a TODO for handling attribute retraction in the cases that
it makes sense. I expect that to be straight-forward.
* Review comment: Rename DiffSet to AddRetractAlterSet.
Also adds a little more commentary and a simple test.
* Review comment: Use ToIdent trait.
* Review comment: partially revert "Part 2: Maintain entids separately from idents."
This reverts commit 23a91df9c35e14398f2ddbd1ba25315821e67401.
Following our discussion, this removes the "entids" materialized
view. The next commit will remove historical idents from the "idents"
materialized view.
* Post: Use custom Either rather than std::result::Result.
This is not necessary, but it was suggested that we might be paying an
overhead creating Err instances while using error_chain. That seems
not to be the case, but this change shows that we don't actually use
any of the Result helper methods, so there's no reason to overload
Result. This change might avoid some future confusion, so I'm going
to land it anyway.
Signed-off-by: Nick Alexander <nalexander@mozilla.com>
* Review comment: Don't preserve historical idents.
* Review comment: More prepared statements when updating materialized views.
* Post: Test altering :db/cardinality and :db/unique.
These tests fail due to a Datomic limitation, namely that the marker
flag :db.alter/attribute can only be asserted once for an attribute!
That is, [:db.part/db :db.alter/attribute :attribute] will only be
transacted at most once. Since older versions of Datomic required the
:db.alter/attribute flag, I can only imagine they either never wrote
:db.alter/attribute to the store, or they handled it specially. I'll
need to remove the marker flag system from Mentat in order to address
this fundamental limitation.
* Post: Remove some more single quotes from error output.
* Post: Add assert_transact! macro to unwrap safely.
I was finding it very difficult to track unwrapping errors while
making changes, due to an underlying Mac OS X symbolication issue that
makes running tests with RUST_BACKTRACE=1 so slow that they all time
out.
* Post: Don't expect or recognize :db.{install,alter}/attribute.
I had this all working... except we will never see a repeated
`[:db.part/db :db.alter/attribute :attribute]` assertion in the store!
That means my approach would let you alter an attribute at most one
time. It's not worth hacking around this; it's better to just stop
expecting (and recognizing) the marker flags. (We have all the data
to distinguish the various cases that we need without the marker
flags.)
This brings Mentat in line with the thrust of newer Datomic versions,
but isn't compatible with Datomic, because (if I understand correctly)
Datomic automatically adds :db.{install,alter}/attribute assertions to
transactions.
I haven't purged the corresponding :db/ident and schema fragments just
yet:
- we might want them back
- we might want them in order to upgrade v1 and v2 databases to the
new on-disk layout we're fleshing out (v3?).
* Post: Don't make :db/unique :db.unique/* imply :db/index true.
This patch avoids a potential bug with the "schema" materialized view.
If :db/unique :db.unique/value implies :db/index true, then what
happens when you _retract_ :db.unique/value? I think Datomic defines
this in some way, but I really want the "schema" materialized view to
be a slice of "datoms" and not have these sort of ambiguities and
persistent effects. Therefore, to ensure that we don't retract a
schema characteristic and accidentally change more than we intended
to, this patch stops having any schema characteristic imply any other
schema characteristic(s). To achieve that, I added an
Option<Unique::{Value,Identity}> type to Attribute; this helps with
this patch, and also looks ahead to when we allow to retract
:db/unique attributes.
* Post: Allow to retract :db/ident.
* Post: Include more details about invalid schema changes.
The tests use strings, so they hide the chained errors which do in
fact provide more detail.
* Review comment: Fix outdated comment.
* Review comment: s/_SET/_SQL_LIST/.
* Review comment: Use a sub-select for checking cardinality.
This might be faster in practice.
* Review comment: Put `attribute::Unique` into its own namespace.
2017-03-20 20:18:59 +00:00
[ 101 :test / ident - 9 ? tx true ]
2019-07-17 15:46:47 +00:00
[ ? tx :db / txInstant ? ms ? tx true ] ] "
) ;
Schema alteration. Fixes #294 and #295. (#370) r=rnewman
* Pre: Don't retract :db/ident in test.
Datomic (and eventually Mentat) don't allow to retract :db/ident in
this way, so this runs afoul of future work to support mutating
metadata.
* Pre: s/VALUETYPE/VALUE_TYPE/.
This is consistent with the capitalization (which is "valueType") and
the other identifier.
* Pre: Remove some single quotes from error output.
* Part 1: Make materialized views be uniform [e a v value_type_tag].
This looks ahead to a time when we could support arbitrary
user-defined materialized views. For now, the "idents" materialized
view is those datoms of the form [e :db/ident :namespaced/keyword] and
the "schema" materialized view is those datoms of the form [e a v]
where a is in a particular set of attributes that will become clear in
the following commits.
This change is not backwards compatible, so I'm removing the open
current (really, v2) test. It'll be re-instated when we get to
https://github.com/mozilla/mentat/issues/194.
* Pre: Map TypedValue::Ref to TypedValue::Keyword in debug output.
* Part 3: Separate `schema_to_mutate` from the `schema` used to interpret.
This is just to keep track of the expected changes during
bootstrapping. I want bootstrap metadata mutations to flow through
the same code path as metadata mutations during regular transactions;
by differentiating the schema used for interpretation from the schema
that will be updated I expect to be able to apply bootstrap metadata
mutations to an empty schema and have things like materialized views
created (using the regular code paths).
This commit has been re-ordered for conceptual clarity, but it won't
compile because it references the metadata module. It's possible to
make it compile -- the functionality is there in the schema module --
but it's not worth the rebasing effort until after review (and
possibly not even then, since we'll squash down to a single commit to
land).
* Part 2: Maintain entids separately from idents.
In order to support historical idents, we need to distinguish the
"current" map from entid -> ident from the "complete historical" map
ident -> entid. This is what Datomic does; in Datomic, an ident is
never retracted (although it can be replaced). This approach is an
important part of allowing multiple consumers to share a schema
fragment as it migrates forward.
This fixes a limitation of the Clojure implementation, which did not
handle historical idents across knowledge base close and re-open.
The "entids" materialized view is naturally a slice of the "datoms"
table. The "idents" materialized view is a slice of the
"transactions" table. I hope that representing in this way, and
casting the problem in this light, might generalize to future
materialized views.
* Pre: Add DiffSet.
* Part 4: Collect mutations to a `Schema`.
I haven't taken your review comment about consuming AttributeBuilder
during each fluent function. If you read my response and still want
this, I'm happy to do it in review.
* Part 5: Handle :db/ident and :db.{install,alter}/attribute.
This "loops" the committed datoms out of the SQL store and back
through the metadata (schema, but in future also partition map)
processor. The metadata processor updates the schema and produces a
report of what changed; that report is then used to update the SQL
store. That update includes:
- the materialized views ("entids", "idents", and "schema");
- if needed, a subset of the datoms themselves (as flags change).
I've left a TODO for handling attribute retraction in the cases that
it makes sense. I expect that to be straight-forward.
* Review comment: Rename DiffSet to AddRetractAlterSet.
Also adds a little more commentary and a simple test.
* Review comment: Use ToIdent trait.
* Review comment: partially revert "Part 2: Maintain entids separately from idents."
This reverts commit 23a91df9c35e14398f2ddbd1ba25315821e67401.
Following our discussion, this removes the "entids" materialized
view. The next commit will remove historical idents from the "idents"
materialized view.
* Post: Use custom Either rather than std::result::Result.
This is not necessary, but it was suggested that we might be paying an
overhead creating Err instances while using error_chain. That seems
not to be the case, but this change shows that we don't actually use
any of the Result helper methods, so there's no reason to overload
Result. This change might avoid some future confusion, so I'm going
to land it anyway.
Signed-off-by: Nick Alexander <nalexander@mozilla.com>
* Review comment: Don't preserve historical idents.
* Review comment: More prepared statements when updating materialized views.
* Post: Test altering :db/cardinality and :db/unique.
These tests fail due to a Datomic limitation, namely that the marker
flag :db.alter/attribute can only be asserted once for an attribute!
That is, [:db.part/db :db.alter/attribute :attribute] will only be
transacted at most once. Since older versions of Datomic required the
:db.alter/attribute flag, I can only imagine they either never wrote
:db.alter/attribute to the store, or they handled it specially. I'll
need to remove the marker flag system from Mentat in order to address
this fundamental limitation.
* Post: Remove some more single quotes from error output.
* Post: Add assert_transact! macro to unwrap safely.
I was finding it very difficult to track unwrapping errors while
making changes, due to an underlying Mac OS X symbolication issue that
makes running tests with RUST_BACKTRACE=1 so slow that they all time
out.
* Post: Don't expect or recognize :db.{install,alter}/attribute.
I had this all working... except we will never see a repeated
`[:db.part/db :db.alter/attribute :attribute]` assertion in the store!
That means my approach would let you alter an attribute at most one
time. It's not worth hacking around this; it's better to just stop
expecting (and recognizing) the marker flags. (We have all the data
to distinguish the various cases that we need without the marker
flags.)
This brings Mentat in line with the thrust of newer Datomic versions,
but isn't compatible with Datomic, because (if I understand correctly)
Datomic automatically adds :db.{install,alter}/attribute assertions to
transactions.
I haven't purged the corresponding :db/ident and schema fragments just
yet:
- we might want them back
- we might want them in order to upgrade v1 and v2 databases to the
new on-disk layout we're fleshing out (v3?).
* Post: Don't make :db/unique :db.unique/* imply :db/index true.
This patch avoids a potential bug with the "schema" materialized view.
If :db/unique :db.unique/value implies :db/index true, then what
happens when you _retract_ :db.unique/value? I think Datomic defines
this in some way, but I really want the "schema" materialized view to
be a slice of "datoms" and not have these sort of ambiguities and
persistent effects. Therefore, to ensure that we don't retract a
schema characteristic and accidentally change more than we intended
to, this patch stops having any schema characteristic imply any other
schema characteristic(s). To achieve that, I added an
Option<Unique::{Value,Identity}> type to Attribute; this helps with
this patch, and also looks ahead to when we allow to retract
:db/unique attributes.
* Post: Allow to retract :db/ident.
* Post: Include more details about invalid schema changes.
The tests use strings, so they hide the chained errors which do in
fact provide more detail.
* Review comment: Fix outdated comment.
* Review comment: s/_SET/_SQL_LIST/.
* Review comment: Use a sub-select for checking cardinality.
This might be faster in practice.
* Review comment: Put `attribute::Unique` into its own namespace.
2017-03-20 20:18:59 +00:00
2018-07-18 23:55:31 +00:00
// Cannot retract a single characteristic of an installed attribute.
2019-07-17 15:46:47 +00:00
assert_transact! (
conn ,
" [[:db/retract 100 :db/cardinality :db.cardinality/many]] " ,
Err ( " bad schema assertion: Retracting attribute 8 for entity 100 not permitted. " )
) ;
Schema alteration. Fixes #294 and #295. (#370) r=rnewman
* Pre: Don't retract :db/ident in test.
Datomic (and eventually Mentat) don't allow to retract :db/ident in
this way, so this runs afoul of future work to support mutating
metadata.
* Pre: s/VALUETYPE/VALUE_TYPE/.
This is consistent with the capitalization (which is "valueType") and
the other identifier.
* Pre: Remove some single quotes from error output.
* Part 1: Make materialized views be uniform [e a v value_type_tag].
This looks ahead to a time when we could support arbitrary
user-defined materialized views. For now, the "idents" materialized
view is those datoms of the form [e :db/ident :namespaced/keyword] and
the "schema" materialized view is those datoms of the form [e a v]
where a is in a particular set of attributes that will become clear in
the following commits.
This change is not backwards compatible, so I'm removing the open
current (really, v2) test. It'll be re-instated when we get to
https://github.com/mozilla/mentat/issues/194.
* Pre: Map TypedValue::Ref to TypedValue::Keyword in debug output.
* Part 3: Separate `schema_to_mutate` from the `schema` used to interpret.
This is just to keep track of the expected changes during
bootstrapping. I want bootstrap metadata mutations to flow through
the same code path as metadata mutations during regular transactions;
by differentiating the schema used for interpretation from the schema
that will be updated I expect to be able to apply bootstrap metadata
mutations to an empty schema and have things like materialized views
created (using the regular code paths).
This commit has been re-ordered for conceptual clarity, but it won't
compile because it references the metadata module. It's possible to
make it compile -- the functionality is there in the schema module --
but it's not worth the rebasing effort until after review (and
possibly not even then, since we'll squash down to a single commit to
land).
* Part 2: Maintain entids separately from idents.
In order to support historical idents, we need to distinguish the
"current" map from entid -> ident from the "complete historical" map
ident -> entid. This is what Datomic does; in Datomic, an ident is
never retracted (although it can be replaced). This approach is an
important part of allowing multiple consumers to share a schema
fragment as it migrates forward.
This fixes a limitation of the Clojure implementation, which did not
handle historical idents across knowledge base close and re-open.
The "entids" materialized view is naturally a slice of the "datoms"
table. The "idents" materialized view is a slice of the
"transactions" table. I hope that representing in this way, and
casting the problem in this light, might generalize to future
materialized views.
* Pre: Add DiffSet.
* Part 4: Collect mutations to a `Schema`.
I haven't taken your review comment about consuming AttributeBuilder
during each fluent function. If you read my response and still want
this, I'm happy to do it in review.
* Part 5: Handle :db/ident and :db.{install,alter}/attribute.
This "loops" the committed datoms out of the SQL store and back
through the metadata (schema, but in future also partition map)
processor. The metadata processor updates the schema and produces a
report of what changed; that report is then used to update the SQL
store. That update includes:
- the materialized views ("entids", "idents", and "schema");
- if needed, a subset of the datoms themselves (as flags change).
I've left a TODO for handling attribute retraction in the cases that
it makes sense. I expect that to be straight-forward.
* Review comment: Rename DiffSet to AddRetractAlterSet.
Also adds a little more commentary and a simple test.
* Review comment: Use ToIdent trait.
* Review comment: partially revert "Part 2: Maintain entids separately from idents."
This reverts commit 23a91df9c35e14398f2ddbd1ba25315821e67401.
Following our discussion, this removes the "entids" materialized
view. The next commit will remove historical idents from the "idents"
materialized view.
* Post: Use custom Either rather than std::result::Result.
This is not necessary, but it was suggested that we might be paying an
overhead creating Err instances while using error_chain. That seems
not to be the case, but this change shows that we don't actually use
any of the Result helper methods, so there's no reason to overload
Result. This change might avoid some future confusion, so I'm going
to land it anyway.
Signed-off-by: Nick Alexander <nalexander@mozilla.com>
* Review comment: Don't preserve historical idents.
* Review comment: More prepared statements when updating materialized views.
* Post: Test altering :db/cardinality and :db/unique.
These tests fail due to a Datomic limitation, namely that the marker
flag :db.alter/attribute can only be asserted once for an attribute!
That is, [:db.part/db :db.alter/attribute :attribute] will only be
transacted at most once. Since older versions of Datomic required the
:db.alter/attribute flag, I can only imagine they either never wrote
:db.alter/attribute to the store, or they handled it specially. I'll
need to remove the marker flag system from Mentat in order to address
this fundamental limitation.
* Post: Remove some more single quotes from error output.
* Post: Add assert_transact! macro to unwrap safely.
I was finding it very difficult to track unwrapping errors while
making changes, due to an underlying Mac OS X symbolication issue that
makes running tests with RUST_BACKTRACE=1 so slow that they all time
out.
* Post: Don't expect or recognize :db.{install,alter}/attribute.
I had this all working... except we will never see a repeated
`[:db.part/db :db.alter/attribute :attribute]` assertion in the store!
That means my approach would let you alter an attribute at most one
time. It's not worth hacking around this; it's better to just stop
expecting (and recognizing) the marker flags. (We have all the data
to distinguish the various cases that we need without the marker
flags.)
This brings Mentat in line with the thrust of newer Datomic versions,
but isn't compatible with Datomic, because (if I understand correctly)
Datomic automatically adds :db.{install,alter}/attribute assertions to
transactions.
I haven't purged the corresponding :db/ident and schema fragments just
yet:
- we might want them back
- we might want them in order to upgrade v1 and v2 databases to the
new on-disk layout we're fleshing out (v3?).
* Post: Don't make :db/unique :db.unique/* imply :db/index true.
This patch avoids a potential bug with the "schema" materialized view.
If :db/unique :db.unique/value implies :db/index true, then what
happens when you _retract_ :db.unique/value? I think Datomic defines
this in some way, but I really want the "schema" materialized view to
be a slice of "datoms" and not have these sort of ambiguities and
persistent effects. Therefore, to ensure that we don't retract a
schema characteristic and accidentally change more than we intended
to, this patch stops having any schema characteristic imply any other
schema characteristic(s). To achieve that, I added an
Option<Unique::{Value,Identity}> type to Attribute; this helps with
this patch, and also looks ahead to when we allow to retract
:db/unique attributes.
* Post: Allow to retract :db/ident.
* Post: Include more details about invalid schema changes.
The tests use strings, so they hide the chained errors which do in
fact provide more detail.
* Review comment: Fix outdated comment.
* Review comment: s/_SET/_SQL_LIST/.
* Review comment: Use a sub-select for checking cardinality.
This might be faster in practice.
* Review comment: Put `attribute::Unique` into its own namespace.
2017-03-20 20:18:59 +00:00
2018-07-18 23:55:31 +00:00
// Cannot retract a single characteristic of an installed attribute.
2019-07-17 15:46:47 +00:00
assert_transact! (
conn ,
" [[:db/retract 100 :db/valueType :db.type/long]] " ,
Err ( " bad schema assertion: Retracting attribute 7 for entity 100 not permitted. " )
) ;
2018-07-18 23:55:31 +00:00
// Cannot retract a non-defining set of characteristics of an installed attribute.
assert_transact! ( conn ,
" [[:db/retract 100 :db/valueType :db.type/long]
[ :db / retract 100 :db / cardinality :db . cardinality / many ] ] " ,
Err ( " bad schema assertion: Retracting defining attributes of a schema without retracting its :db/ident is not permitted. " ) ) ;
// See https://github.com/mozilla/mentat/issues/796.
// assert_transact!(conn,
// "[[:db/retract 100 :db/ident :test/ident]]",
// Err("bad schema assertion: Retracting :db/ident of a schema without retracting its defining attributes is not permitted."));
// Can retract all of characterists of an installed attribute in one go.
2019-07-17 15:46:47 +00:00
assert_transact! (
conn ,
" [[:db/retract 100 :db/cardinality :db.cardinality/many]
2018-07-18 23:55:31 +00:00
[ :db / retract 100 :db / valueType :db . type / long ]
2019-07-17 15:46:47 +00:00
[ :db / retract 100 :db / ident :test / ident ] ] "
) ;
2018-07-18 23:55:31 +00:00
Schema alteration. Fixes #294 and #295. (#370) r=rnewman
* Pre: Don't retract :db/ident in test.
Datomic (and eventually Mentat) don't allow to retract :db/ident in
this way, so this runs afoul of future work to support mutating
metadata.
* Pre: s/VALUETYPE/VALUE_TYPE/.
This is consistent with the capitalization (which is "valueType") and
the other identifier.
* Pre: Remove some single quotes from error output.
* Part 1: Make materialized views be uniform [e a v value_type_tag].
This looks ahead to a time when we could support arbitrary
user-defined materialized views. For now, the "idents" materialized
view is those datoms of the form [e :db/ident :namespaced/keyword] and
the "schema" materialized view is those datoms of the form [e a v]
where a is in a particular set of attributes that will become clear in
the following commits.
This change is not backwards compatible, so I'm removing the open
current (really, v2) test. It'll be re-instated when we get to
https://github.com/mozilla/mentat/issues/194.
* Pre: Map TypedValue::Ref to TypedValue::Keyword in debug output.
* Part 3: Separate `schema_to_mutate` from the `schema` used to interpret.
This is just to keep track of the expected changes during
bootstrapping. I want bootstrap metadata mutations to flow through
the same code path as metadata mutations during regular transactions;
by differentiating the schema used for interpretation from the schema
that will be updated I expect to be able to apply bootstrap metadata
mutations to an empty schema and have things like materialized views
created (using the regular code paths).
This commit has been re-ordered for conceptual clarity, but it won't
compile because it references the metadata module. It's possible to
make it compile -- the functionality is there in the schema module --
but it's not worth the rebasing effort until after review (and
possibly not even then, since we'll squash down to a single commit to
land).
* Part 2: Maintain entids separately from idents.
In order to support historical idents, we need to distinguish the
"current" map from entid -> ident from the "complete historical" map
ident -> entid. This is what Datomic does; in Datomic, an ident is
never retracted (although it can be replaced). This approach is an
important part of allowing multiple consumers to share a schema
fragment as it migrates forward.
This fixes a limitation of the Clojure implementation, which did not
handle historical idents across knowledge base close and re-open.
The "entids" materialized view is naturally a slice of the "datoms"
table. The "idents" materialized view is a slice of the
"transactions" table. I hope that representing in this way, and
casting the problem in this light, might generalize to future
materialized views.
* Pre: Add DiffSet.
* Part 4: Collect mutations to a `Schema`.
I haven't taken your review comment about consuming AttributeBuilder
during each fluent function. If you read my response and still want
this, I'm happy to do it in review.
* Part 5: Handle :db/ident and :db.{install,alter}/attribute.
This "loops" the committed datoms out of the SQL store and back
through the metadata (schema, but in future also partition map)
processor. The metadata processor updates the schema and produces a
report of what changed; that report is then used to update the SQL
store. That update includes:
- the materialized views ("entids", "idents", and "schema");
- if needed, a subset of the datoms themselves (as flags change).
I've left a TODO for handling attribute retraction in the cases that
it makes sense. I expect that to be straight-forward.
* Review comment: Rename DiffSet to AddRetractAlterSet.
Also adds a little more commentary and a simple test.
* Review comment: Use ToIdent trait.
* Review comment: partially revert "Part 2: Maintain entids separately from idents."
This reverts commit 23a91df9c35e14398f2ddbd1ba25315821e67401.
Following our discussion, this removes the "entids" materialized
view. The next commit will remove historical idents from the "idents"
materialized view.
* Post: Use custom Either rather than std::result::Result.
This is not necessary, but it was suggested that we might be paying an
overhead creating Err instances while using error_chain. That seems
not to be the case, but this change shows that we don't actually use
any of the Result helper methods, so there's no reason to overload
Result. This change might avoid some future confusion, so I'm going
to land it anyway.
Signed-off-by: Nick Alexander <nalexander@mozilla.com>
* Review comment: Don't preserve historical idents.
* Review comment: More prepared statements when updating materialized views.
* Post: Test altering :db/cardinality and :db/unique.
These tests fail due to a Datomic limitation, namely that the marker
flag :db.alter/attribute can only be asserted once for an attribute!
That is, [:db.part/db :db.alter/attribute :attribute] will only be
transacted at most once. Since older versions of Datomic required the
:db.alter/attribute flag, I can only imagine they either never wrote
:db.alter/attribute to the store, or they handled it specially. I'll
need to remove the marker flag system from Mentat in order to address
this fundamental limitation.
* Post: Remove some more single quotes from error output.
* Post: Add assert_transact! macro to unwrap safely.
I was finding it very difficult to track unwrapping errors while
making changes, due to an underlying Mac OS X symbolication issue that
makes running tests with RUST_BACKTRACE=1 so slow that they all time
out.
* Post: Don't expect or recognize :db.{install,alter}/attribute.
I had this all working... except we will never see a repeated
`[:db.part/db :db.alter/attribute :attribute]` assertion in the store!
That means my approach would let you alter an attribute at most one
time. It's not worth hacking around this; it's better to just stop
expecting (and recognizing) the marker flags. (We have all the data
to distinguish the various cases that we need without the marker
flags.)
This brings Mentat in line with the thrust of newer Datomic versions,
but isn't compatible with Datomic, because (if I understand correctly)
Datomic automatically adds :db.{install,alter}/attribute assertions to
transactions.
I haven't purged the corresponding :db/ident and schema fragments just
yet:
- we might want them back
- we might want them in order to upgrade v1 and v2 databases to the
new on-disk layout we're fleshing out (v3?).
* Post: Don't make :db/unique :db.unique/* imply :db/index true.
This patch avoids a potential bug with the "schema" materialized view.
If :db/unique :db.unique/value implies :db/index true, then what
happens when you _retract_ :db.unique/value? I think Datomic defines
this in some way, but I really want the "schema" materialized view to
be a slice of "datoms" and not have these sort of ambiguities and
persistent effects. Therefore, to ensure that we don't retract a
schema characteristic and accidentally change more than we intended
to, this patch stops having any schema characteristic imply any other
schema characteristic(s). To achieve that, I added an
Option<Unique::{Value,Identity}> type to Attribute; this helps with
this patch, and also looks ahead to when we allow to retract
:db/unique attributes.
* Post: Allow to retract :db/ident.
* Post: Include more details about invalid schema changes.
The tests use strings, so they hide the chained errors which do in
fact provide more detail.
* Review comment: Fix outdated comment.
* Review comment: s/_SET/_SQL_LIST/.
* Review comment: Use a sub-select for checking cardinality.
This might be faster in practice.
* Review comment: Put `attribute::Unique` into its own namespace.
2017-03-20 20:18:59 +00:00
// Trying to install an attribute without a :db/ident is allowed.
2019-07-17 15:46:47 +00:00
assert_transact! (
conn ,
" [[:db/add 101 :db/valueType :db.type/long]
[ :db / add 101 :db / cardinality :db . cardinality / many ] ] "
) ;
Schema alteration. Fixes #294 and #295. (#370) r=rnewman
* Pre: Don't retract :db/ident in test.
Datomic (and eventually Mentat) don't allow to retract :db/ident in
this way, so this runs afoul of future work to support mutating
metadata.
* Pre: s/VALUETYPE/VALUE_TYPE/.
This is consistent with the capitalization (which is "valueType") and
the other identifier.
* Pre: Remove some single quotes from error output.
* Part 1: Make materialized views be uniform [e a v value_type_tag].
This looks ahead to a time when we could support arbitrary
user-defined materialized views. For now, the "idents" materialized
view is those datoms of the form [e :db/ident :namespaced/keyword] and
the "schema" materialized view is those datoms of the form [e a v]
where a is in a particular set of attributes that will become clear in
the following commits.
This change is not backwards compatible, so I'm removing the open
current (really, v2) test. It'll be re-instated when we get to
https://github.com/mozilla/mentat/issues/194.
* Pre: Map TypedValue::Ref to TypedValue::Keyword in debug output.
* Part 3: Separate `schema_to_mutate` from the `schema` used to interpret.
This is just to keep track of the expected changes during
bootstrapping. I want bootstrap metadata mutations to flow through
the same code path as metadata mutations during regular transactions;
by differentiating the schema used for interpretation from the schema
that will be updated I expect to be able to apply bootstrap metadata
mutations to an empty schema and have things like materialized views
created (using the regular code paths).
This commit has been re-ordered for conceptual clarity, but it won't
compile because it references the metadata module. It's possible to
make it compile -- the functionality is there in the schema module --
but it's not worth the rebasing effort until after review (and
possibly not even then, since we'll squash down to a single commit to
land).
* Part 2: Maintain entids separately from idents.
In order to support historical idents, we need to distinguish the
"current" map from entid -> ident from the "complete historical" map
ident -> entid. This is what Datomic does; in Datomic, an ident is
never retracted (although it can be replaced). This approach is an
important part of allowing multiple consumers to share a schema
fragment as it migrates forward.
This fixes a limitation of the Clojure implementation, which did not
handle historical idents across knowledge base close and re-open.
The "entids" materialized view is naturally a slice of the "datoms"
table. The "idents" materialized view is a slice of the
"transactions" table. I hope that representing in this way, and
casting the problem in this light, might generalize to future
materialized views.
* Pre: Add DiffSet.
* Part 4: Collect mutations to a `Schema`.
I haven't taken your review comment about consuming AttributeBuilder
during each fluent function. If you read my response and still want
this, I'm happy to do it in review.
* Part 5: Handle :db/ident and :db.{install,alter}/attribute.
This "loops" the committed datoms out of the SQL store and back
through the metadata (schema, but in future also partition map)
processor. The metadata processor updates the schema and produces a
report of what changed; that report is then used to update the SQL
store. That update includes:
- the materialized views ("entids", "idents", and "schema");
- if needed, a subset of the datoms themselves (as flags change).
I've left a TODO for handling attribute retraction in the cases that
it makes sense. I expect that to be straight-forward.
* Review comment: Rename DiffSet to AddRetractAlterSet.
Also adds a little more commentary and a simple test.
* Review comment: Use ToIdent trait.
* Review comment: partially revert "Part 2: Maintain entids separately from idents."
This reverts commit 23a91df9c35e14398f2ddbd1ba25315821e67401.
Following our discussion, this removes the "entids" materialized
view. The next commit will remove historical idents from the "idents"
materialized view.
* Post: Use custom Either rather than std::result::Result.
This is not necessary, but it was suggested that we might be paying an
overhead creating Err instances while using error_chain. That seems
not to be the case, but this change shows that we don't actually use
any of the Result helper methods, so there's no reason to overload
Result. This change might avoid some future confusion, so I'm going
to land it anyway.
Signed-off-by: Nick Alexander <nalexander@mozilla.com>
* Review comment: Don't preserve historical idents.
* Review comment: More prepared statements when updating materialized views.
* Post: Test altering :db/cardinality and :db/unique.
These tests fail due to a Datomic limitation, namely that the marker
flag :db.alter/attribute can only be asserted once for an attribute!
That is, [:db.part/db :db.alter/attribute :attribute] will only be
transacted at most once. Since older versions of Datomic required the
:db.alter/attribute flag, I can only imagine they either never wrote
:db.alter/attribute to the store, or they handled it specially. I'll
need to remove the marker flag system from Mentat in order to address
this fundamental limitation.
* Post: Remove some more single quotes from error output.
* Post: Add assert_transact! macro to unwrap safely.
I was finding it very difficult to track unwrapping errors while
making changes, due to an underlying Mac OS X symbolication issue that
makes running tests with RUST_BACKTRACE=1 so slow that they all time
out.
* Post: Don't expect or recognize :db.{install,alter}/attribute.
I had this all working... except we will never see a repeated
`[:db.part/db :db.alter/attribute :attribute]` assertion in the store!
That means my approach would let you alter an attribute at most one
time. It's not worth hacking around this; it's better to just stop
expecting (and recognizing) the marker flags. (We have all the data
to distinguish the various cases that we need without the marker
flags.)
This brings Mentat in line with the thrust of newer Datomic versions,
but isn't compatible with Datomic, because (if I understand correctly)
Datomic automatically adds :db.{install,alter}/attribute assertions to
transactions.
I haven't purged the corresponding :db/ident and schema fragments just
yet:
- we might want them back
- we might want them in order to upgrade v1 and v2 databases to the
new on-disk layout we're fleshing out (v3?).
* Post: Don't make :db/unique :db.unique/* imply :db/index true.
This patch avoids a potential bug with the "schema" materialized view.
If :db/unique :db.unique/value implies :db/index true, then what
happens when you _retract_ :db.unique/value? I think Datomic defines
this in some way, but I really want the "schema" materialized view to
be a slice of "datoms" and not have these sort of ambiguities and
persistent effects. Therefore, to ensure that we don't retract a
schema characteristic and accidentally change more than we intended
to, this patch stops having any schema characteristic imply any other
schema characteristic(s). To achieve that, I added an
Option<Unique::{Value,Identity}> type to Attribute; this helps with
this patch, and also looks ahead to when we allow to retract
:db/unique attributes.
* Post: Allow to retract :db/ident.
* Post: Include more details about invalid schema changes.
The tests use strings, so they hide the chained errors which do in
fact provide more detail.
* Review comment: Fix outdated comment.
* Review comment: s/_SET/_SQL_LIST/.
* Review comment: Use a sub-select for checking cardinality.
This might be faster in practice.
* Review comment: Put `attribute::Unique` into its own namespace.
2017-03-20 20:18:59 +00:00
}
#[ test ]
fn test_db_alter ( ) {
let mut conn = TestConn ::default ( ) ;
// Start by installing a :db.cardinality/one attribute.
2019-07-17 15:46:47 +00:00
assert_transact! (
conn ,
" [[:db/add 100 :db/ident :test/ident]
Schema alteration. Fixes #294 and #295. (#370) r=rnewman
* Pre: Don't retract :db/ident in test.
Datomic (and eventually Mentat) don't allow to retract :db/ident in
this way, so this runs afoul of future work to support mutating
metadata.
* Pre: s/VALUETYPE/VALUE_TYPE/.
This is consistent with the capitalization (which is "valueType") and
the other identifier.
* Pre: Remove some single quotes from error output.
* Part 1: Make materialized views be uniform [e a v value_type_tag].
This looks ahead to a time when we could support arbitrary
user-defined materialized views. For now, the "idents" materialized
view is those datoms of the form [e :db/ident :namespaced/keyword] and
the "schema" materialized view is those datoms of the form [e a v]
where a is in a particular set of attributes that will become clear in
the following commits.
This change is not backwards compatible, so I'm removing the open
current (really, v2) test. It'll be re-instated when we get to
https://github.com/mozilla/mentat/issues/194.
* Pre: Map TypedValue::Ref to TypedValue::Keyword in debug output.
* Part 3: Separate `schema_to_mutate` from the `schema` used to interpret.
This is just to keep track of the expected changes during
bootstrapping. I want bootstrap metadata mutations to flow through
the same code path as metadata mutations during regular transactions;
by differentiating the schema used for interpretation from the schema
that will be updated I expect to be able to apply bootstrap metadata
mutations to an empty schema and have things like materialized views
created (using the regular code paths).
This commit has been re-ordered for conceptual clarity, but it won't
compile because it references the metadata module. It's possible to
make it compile -- the functionality is there in the schema module --
but it's not worth the rebasing effort until after review (and
possibly not even then, since we'll squash down to a single commit to
land).
* Part 2: Maintain entids separately from idents.
In order to support historical idents, we need to distinguish the
"current" map from entid -> ident from the "complete historical" map
ident -> entid. This is what Datomic does; in Datomic, an ident is
never retracted (although it can be replaced). This approach is an
important part of allowing multiple consumers to share a schema
fragment as it migrates forward.
This fixes a limitation of the Clojure implementation, which did not
handle historical idents across knowledge base close and re-open.
The "entids" materialized view is naturally a slice of the "datoms"
table. The "idents" materialized view is a slice of the
"transactions" table. I hope that representing in this way, and
casting the problem in this light, might generalize to future
materialized views.
* Pre: Add DiffSet.
* Part 4: Collect mutations to a `Schema`.
I haven't taken your review comment about consuming AttributeBuilder
during each fluent function. If you read my response and still want
this, I'm happy to do it in review.
* Part 5: Handle :db/ident and :db.{install,alter}/attribute.
This "loops" the committed datoms out of the SQL store and back
through the metadata (schema, but in future also partition map)
processor. The metadata processor updates the schema and produces a
report of what changed; that report is then used to update the SQL
store. That update includes:
- the materialized views ("entids", "idents", and "schema");
- if needed, a subset of the datoms themselves (as flags change).
I've left a TODO for handling attribute retraction in the cases that
it makes sense. I expect that to be straight-forward.
* Review comment: Rename DiffSet to AddRetractAlterSet.
Also adds a little more commentary and a simple test.
* Review comment: Use ToIdent trait.
* Review comment: partially revert "Part 2: Maintain entids separately from idents."
This reverts commit 23a91df9c35e14398f2ddbd1ba25315821e67401.
Following our discussion, this removes the "entids" materialized
view. The next commit will remove historical idents from the "idents"
materialized view.
* Post: Use custom Either rather than std::result::Result.
This is not necessary, but it was suggested that we might be paying an
overhead creating Err instances while using error_chain. That seems
not to be the case, but this change shows that we don't actually use
any of the Result helper methods, so there's no reason to overload
Result. This change might avoid some future confusion, so I'm going
to land it anyway.
Signed-off-by: Nick Alexander <nalexander@mozilla.com>
* Review comment: Don't preserve historical idents.
* Review comment: More prepared statements when updating materialized views.
* Post: Test altering :db/cardinality and :db/unique.
These tests fail due to a Datomic limitation, namely that the marker
flag :db.alter/attribute can only be asserted once for an attribute!
That is, [:db.part/db :db.alter/attribute :attribute] will only be
transacted at most once. Since older versions of Datomic required the
:db.alter/attribute flag, I can only imagine they either never wrote
:db.alter/attribute to the store, or they handled it specially. I'll
need to remove the marker flag system from Mentat in order to address
this fundamental limitation.
* Post: Remove some more single quotes from error output.
* Post: Add assert_transact! macro to unwrap safely.
I was finding it very difficult to track unwrapping errors while
making changes, due to an underlying Mac OS X symbolication issue that
makes running tests with RUST_BACKTRACE=1 so slow that they all time
out.
* Post: Don't expect or recognize :db.{install,alter}/attribute.
I had this all working... except we will never see a repeated
`[:db.part/db :db.alter/attribute :attribute]` assertion in the store!
That means my approach would let you alter an attribute at most one
time. It's not worth hacking around this; it's better to just stop
expecting (and recognizing) the marker flags. (We have all the data
to distinguish the various cases that we need without the marker
flags.)
This brings Mentat in line with the thrust of newer Datomic versions,
but isn't compatible with Datomic, because (if I understand correctly)
Datomic automatically adds :db.{install,alter}/attribute assertions to
transactions.
I haven't purged the corresponding :db/ident and schema fragments just
yet:
- we might want them back
- we might want them in order to upgrade v1 and v2 databases to the
new on-disk layout we're fleshing out (v3?).
* Post: Don't make :db/unique :db.unique/* imply :db/index true.
This patch avoids a potential bug with the "schema" materialized view.
If :db/unique :db.unique/value implies :db/index true, then what
happens when you _retract_ :db.unique/value? I think Datomic defines
this in some way, but I really want the "schema" materialized view to
be a slice of "datoms" and not have these sort of ambiguities and
persistent effects. Therefore, to ensure that we don't retract a
schema characteristic and accidentally change more than we intended
to, this patch stops having any schema characteristic imply any other
schema characteristic(s). To achieve that, I added an
Option<Unique::{Value,Identity}> type to Attribute; this helps with
this patch, and also looks ahead to when we allow to retract
:db/unique attributes.
* Post: Allow to retract :db/ident.
* Post: Include more details about invalid schema changes.
The tests use strings, so they hide the chained errors which do in
fact provide more detail.
* Review comment: Fix outdated comment.
* Review comment: s/_SET/_SQL_LIST/.
* Review comment: Use a sub-select for checking cardinality.
This might be faster in practice.
* Review comment: Put `attribute::Unique` into its own namespace.
2017-03-20 20:18:59 +00:00
[ :db / add 100 :db / valueType :db . type / keyword ]
2019-07-17 15:46:47 +00:00
[ :db / add 100 :db / cardinality :db . cardinality / one ] ] "
) ;
Schema alteration. Fixes #294 and #295. (#370) r=rnewman
* Pre: Don't retract :db/ident in test.
Datomic (and eventually Mentat) don't allow to retract :db/ident in
this way, so this runs afoul of future work to support mutating
metadata.
* Pre: s/VALUETYPE/VALUE_TYPE/.
This is consistent with the capitalization (which is "valueType") and
the other identifier.
* Pre: Remove some single quotes from error output.
* Part 1: Make materialized views be uniform [e a v value_type_tag].
This looks ahead to a time when we could support arbitrary
user-defined materialized views. For now, the "idents" materialized
view is those datoms of the form [e :db/ident :namespaced/keyword] and
the "schema" materialized view is those datoms of the form [e a v]
where a is in a particular set of attributes that will become clear in
the following commits.
This change is not backwards compatible, so I'm removing the open
current (really, v2) test. It'll be re-instated when we get to
https://github.com/mozilla/mentat/issues/194.
* Pre: Map TypedValue::Ref to TypedValue::Keyword in debug output.
* Part 3: Separate `schema_to_mutate` from the `schema` used to interpret.
This is just to keep track of the expected changes during
bootstrapping. I want bootstrap metadata mutations to flow through
the same code path as metadata mutations during regular transactions;
by differentiating the schema used for interpretation from the schema
that will be updated I expect to be able to apply bootstrap metadata
mutations to an empty schema and have things like materialized views
created (using the regular code paths).
This commit has been re-ordered for conceptual clarity, but it won't
compile because it references the metadata module. It's possible to
make it compile -- the functionality is there in the schema module --
but it's not worth the rebasing effort until after review (and
possibly not even then, since we'll squash down to a single commit to
land).
* Part 2: Maintain entids separately from idents.
In order to support historical idents, we need to distinguish the
"current" map from entid -> ident from the "complete historical" map
ident -> entid. This is what Datomic does; in Datomic, an ident is
never retracted (although it can be replaced). This approach is an
important part of allowing multiple consumers to share a schema
fragment as it migrates forward.
This fixes a limitation of the Clojure implementation, which did not
handle historical idents across knowledge base close and re-open.
The "entids" materialized view is naturally a slice of the "datoms"
table. The "idents" materialized view is a slice of the
"transactions" table. I hope that representing in this way, and
casting the problem in this light, might generalize to future
materialized views.
* Pre: Add DiffSet.
* Part 4: Collect mutations to a `Schema`.
I haven't taken your review comment about consuming AttributeBuilder
during each fluent function. If you read my response and still want
this, I'm happy to do it in review.
* Part 5: Handle :db/ident and :db.{install,alter}/attribute.
This "loops" the committed datoms out of the SQL store and back
through the metadata (schema, but in future also partition map)
processor. The metadata processor updates the schema and produces a
report of what changed; that report is then used to update the SQL
store. That update includes:
- the materialized views ("entids", "idents", and "schema");
- if needed, a subset of the datoms themselves (as flags change).
I've left a TODO for handling attribute retraction in the cases that
it makes sense. I expect that to be straight-forward.
* Review comment: Rename DiffSet to AddRetractAlterSet.
Also adds a little more commentary and a simple test.
* Review comment: Use ToIdent trait.
* Review comment: partially revert "Part 2: Maintain entids separately from idents."
This reverts commit 23a91df9c35e14398f2ddbd1ba25315821e67401.
Following our discussion, this removes the "entids" materialized
view. The next commit will remove historical idents from the "idents"
materialized view.
* Post: Use custom Either rather than std::result::Result.
This is not necessary, but it was suggested that we might be paying an
overhead creating Err instances while using error_chain. That seems
not to be the case, but this change shows that we don't actually use
any of the Result helper methods, so there's no reason to overload
Result. This change might avoid some future confusion, so I'm going
to land it anyway.
Signed-off-by: Nick Alexander <nalexander@mozilla.com>
* Review comment: Don't preserve historical idents.
* Review comment: More prepared statements when updating materialized views.
* Post: Test altering :db/cardinality and :db/unique.
These tests fail due to a Datomic limitation, namely that the marker
flag :db.alter/attribute can only be asserted once for an attribute!
That is, [:db.part/db :db.alter/attribute :attribute] will only be
transacted at most once. Since older versions of Datomic required the
:db.alter/attribute flag, I can only imagine they either never wrote
:db.alter/attribute to the store, or they handled it specially. I'll
need to remove the marker flag system from Mentat in order to address
this fundamental limitation.
* Post: Remove some more single quotes from error output.
* Post: Add assert_transact! macro to unwrap safely.
I was finding it very difficult to track unwrapping errors while
making changes, due to an underlying Mac OS X symbolication issue that
makes running tests with RUST_BACKTRACE=1 so slow that they all time
out.
* Post: Don't expect or recognize :db.{install,alter}/attribute.
I had this all working... except we will never see a repeated
`[:db.part/db :db.alter/attribute :attribute]` assertion in the store!
That means my approach would let you alter an attribute at most one
time. It's not worth hacking around this; it's better to just stop
expecting (and recognizing) the marker flags. (We have all the data
to distinguish the various cases that we need without the marker
flags.)
This brings Mentat in line with the thrust of newer Datomic versions,
but isn't compatible with Datomic, because (if I understand correctly)
Datomic automatically adds :db.{install,alter}/attribute assertions to
transactions.
I haven't purged the corresponding :db/ident and schema fragments just
yet:
- we might want them back
- we might want them in order to upgrade v1 and v2 databases to the
new on-disk layout we're fleshing out (v3?).
* Post: Don't make :db/unique :db.unique/* imply :db/index true.
This patch avoids a potential bug with the "schema" materialized view.
If :db/unique :db.unique/value implies :db/index true, then what
happens when you _retract_ :db.unique/value? I think Datomic defines
this in some way, but I really want the "schema" materialized view to
be a slice of "datoms" and not have these sort of ambiguities and
persistent effects. Therefore, to ensure that we don't retract a
schema characteristic and accidentally change more than we intended
to, this patch stops having any schema characteristic imply any other
schema characteristic(s). To achieve that, I added an
Option<Unique::{Value,Identity}> type to Attribute; this helps with
this patch, and also looks ahead to when we allow to retract
:db/unique attributes.
* Post: Allow to retract :db/ident.
* Post: Include more details about invalid schema changes.
The tests use strings, so they hide the chained errors which do in
fact provide more detail.
* Review comment: Fix outdated comment.
* Review comment: s/_SET/_SQL_LIST/.
* Review comment: Use a sub-select for checking cardinality.
This might be faster in practice.
* Review comment: Put `attribute::Unique` into its own namespace.
2017-03-20 20:18:59 +00:00
// Trying to alter the :db/valueType will fail.
assert_transact! ( conn , " [[:db/add 100 :db/valueType :db.type/long]] " ,
Err ( " bad schema assertion: Schema alteration for existing attribute with entid 100 is not valid " ) ) ;
// But we can alter the cardinality.
assert_transact! ( conn , " [[:db/add 100 :db/cardinality :db.cardinality/many]] " ) ;
2019-07-17 15:46:47 +00:00
assert_matches! (
conn . last_transaction ( ) ,
" [[100 :db/cardinality :db.cardinality/one ?tx false]
Schema alteration. Fixes #294 and #295. (#370) r=rnewman
* Pre: Don't retract :db/ident in test.
Datomic (and eventually Mentat) don't allow to retract :db/ident in
this way, so this runs afoul of future work to support mutating
metadata.
* Pre: s/VALUETYPE/VALUE_TYPE/.
This is consistent with the capitalization (which is "valueType") and
the other identifier.
* Pre: Remove some single quotes from error output.
* Part 1: Make materialized views be uniform [e a v value_type_tag].
This looks ahead to a time when we could support arbitrary
user-defined materialized views. For now, the "idents" materialized
view is those datoms of the form [e :db/ident :namespaced/keyword] and
the "schema" materialized view is those datoms of the form [e a v]
where a is in a particular set of attributes that will become clear in
the following commits.
This change is not backwards compatible, so I'm removing the open
current (really, v2) test. It'll be re-instated when we get to
https://github.com/mozilla/mentat/issues/194.
* Pre: Map TypedValue::Ref to TypedValue::Keyword in debug output.
* Part 3: Separate `schema_to_mutate` from the `schema` used to interpret.
This is just to keep track of the expected changes during
bootstrapping. I want bootstrap metadata mutations to flow through
the same code path as metadata mutations during regular transactions;
by differentiating the schema used for interpretation from the schema
that will be updated I expect to be able to apply bootstrap metadata
mutations to an empty schema and have things like materialized views
created (using the regular code paths).
This commit has been re-ordered for conceptual clarity, but it won't
compile because it references the metadata module. It's possible to
make it compile -- the functionality is there in the schema module --
but it's not worth the rebasing effort until after review (and
possibly not even then, since we'll squash down to a single commit to
land).
* Part 2: Maintain entids separately from idents.
In order to support historical idents, we need to distinguish the
"current" map from entid -> ident from the "complete historical" map
ident -> entid. This is what Datomic does; in Datomic, an ident is
never retracted (although it can be replaced). This approach is an
important part of allowing multiple consumers to share a schema
fragment as it migrates forward.
This fixes a limitation of the Clojure implementation, which did not
handle historical idents across knowledge base close and re-open.
The "entids" materialized view is naturally a slice of the "datoms"
table. The "idents" materialized view is a slice of the
"transactions" table. I hope that representing in this way, and
casting the problem in this light, might generalize to future
materialized views.
* Pre: Add DiffSet.
* Part 4: Collect mutations to a `Schema`.
I haven't taken your review comment about consuming AttributeBuilder
during each fluent function. If you read my response and still want
this, I'm happy to do it in review.
* Part 5: Handle :db/ident and :db.{install,alter}/attribute.
This "loops" the committed datoms out of the SQL store and back
through the metadata (schema, but in future also partition map)
processor. The metadata processor updates the schema and produces a
report of what changed; that report is then used to update the SQL
store. That update includes:
- the materialized views ("entids", "idents", and "schema");
- if needed, a subset of the datoms themselves (as flags change).
I've left a TODO for handling attribute retraction in the cases that
it makes sense. I expect that to be straight-forward.
* Review comment: Rename DiffSet to AddRetractAlterSet.
Also adds a little more commentary and a simple test.
* Review comment: Use ToIdent trait.
* Review comment: partially revert "Part 2: Maintain entids separately from idents."
This reverts commit 23a91df9c35e14398f2ddbd1ba25315821e67401.
Following our discussion, this removes the "entids" materialized
view. The next commit will remove historical idents from the "idents"
materialized view.
* Post: Use custom Either rather than std::result::Result.
This is not necessary, but it was suggested that we might be paying an
overhead creating Err instances while using error_chain. That seems
not to be the case, but this change shows that we don't actually use
any of the Result helper methods, so there's no reason to overload
Result. This change might avoid some future confusion, so I'm going
to land it anyway.
Signed-off-by: Nick Alexander <nalexander@mozilla.com>
* Review comment: Don't preserve historical idents.
* Review comment: More prepared statements when updating materialized views.
* Post: Test altering :db/cardinality and :db/unique.
These tests fail due to a Datomic limitation, namely that the marker
flag :db.alter/attribute can only be asserted once for an attribute!
That is, [:db.part/db :db.alter/attribute :attribute] will only be
transacted at most once. Since older versions of Datomic required the
:db.alter/attribute flag, I can only imagine they either never wrote
:db.alter/attribute to the store, or they handled it specially. I'll
need to remove the marker flag system from Mentat in order to address
this fundamental limitation.
* Post: Remove some more single quotes from error output.
* Post: Add assert_transact! macro to unwrap safely.
I was finding it very difficult to track unwrapping errors while
making changes, due to an underlying Mac OS X symbolication issue that
makes running tests with RUST_BACKTRACE=1 so slow that they all time
out.
* Post: Don't expect or recognize :db.{install,alter}/attribute.
I had this all working... except we will never see a repeated
`[:db.part/db :db.alter/attribute :attribute]` assertion in the store!
That means my approach would let you alter an attribute at most one
time. It's not worth hacking around this; it's better to just stop
expecting (and recognizing) the marker flags. (We have all the data
to distinguish the various cases that we need without the marker
flags.)
This brings Mentat in line with the thrust of newer Datomic versions,
but isn't compatible with Datomic, because (if I understand correctly)
Datomic automatically adds :db.{install,alter}/attribute assertions to
transactions.
I haven't purged the corresponding :db/ident and schema fragments just
yet:
- we might want them back
- we might want them in order to upgrade v1 and v2 databases to the
new on-disk layout we're fleshing out (v3?).
* Post: Don't make :db/unique :db.unique/* imply :db/index true.
This patch avoids a potential bug with the "schema" materialized view.
If :db/unique :db.unique/value implies :db/index true, then what
happens when you _retract_ :db.unique/value? I think Datomic defines
this in some way, but I really want the "schema" materialized view to
be a slice of "datoms" and not have these sort of ambiguities and
persistent effects. Therefore, to ensure that we don't retract a
schema characteristic and accidentally change more than we intended
to, this patch stops having any schema characteristic imply any other
schema characteristic(s). To achieve that, I added an
Option<Unique::{Value,Identity}> type to Attribute; this helps with
this patch, and also looks ahead to when we allow to retract
:db/unique attributes.
* Post: Allow to retract :db/ident.
* Post: Include more details about invalid schema changes.
The tests use strings, so they hide the chained errors which do in
fact provide more detail.
* Review comment: Fix outdated comment.
* Review comment: s/_SET/_SQL_LIST/.
* Review comment: Use a sub-select for checking cardinality.
This might be faster in practice.
* Review comment: Put `attribute::Unique` into its own namespace.
2017-03-20 20:18:59 +00:00
[ 100 :db / cardinality :db . cardinality / many ? tx true ]
2019-07-17 15:46:47 +00:00
[ ? tx :db / txInstant ? ms ? tx true ] ] "
) ;
assert_matches! (
conn . datoms ( ) ,
" [[100 :db/ident :test/ident]
Schema alteration. Fixes #294 and #295. (#370) r=rnewman
* Pre: Don't retract :db/ident in test.
Datomic (and eventually Mentat) don't allow to retract :db/ident in
this way, so this runs afoul of future work to support mutating
metadata.
* Pre: s/VALUETYPE/VALUE_TYPE/.
This is consistent with the capitalization (which is "valueType") and
the other identifier.
* Pre: Remove some single quotes from error output.
* Part 1: Make materialized views be uniform [e a v value_type_tag].
This looks ahead to a time when we could support arbitrary
user-defined materialized views. For now, the "idents" materialized
view is those datoms of the form [e :db/ident :namespaced/keyword] and
the "schema" materialized view is those datoms of the form [e a v]
where a is in a particular set of attributes that will become clear in
the following commits.
This change is not backwards compatible, so I'm removing the open
current (really, v2) test. It'll be re-instated when we get to
https://github.com/mozilla/mentat/issues/194.
* Pre: Map TypedValue::Ref to TypedValue::Keyword in debug output.
* Part 3: Separate `schema_to_mutate` from the `schema` used to interpret.
This is just to keep track of the expected changes during
bootstrapping. I want bootstrap metadata mutations to flow through
the same code path as metadata mutations during regular transactions;
by differentiating the schema used for interpretation from the schema
that will be updated I expect to be able to apply bootstrap metadata
mutations to an empty schema and have things like materialized views
created (using the regular code paths).
This commit has been re-ordered for conceptual clarity, but it won't
compile because it references the metadata module. It's possible to
make it compile -- the functionality is there in the schema module --
but it's not worth the rebasing effort until after review (and
possibly not even then, since we'll squash down to a single commit to
land).
* Part 2: Maintain entids separately from idents.
In order to support historical idents, we need to distinguish the
"current" map from entid -> ident from the "complete historical" map
ident -> entid. This is what Datomic does; in Datomic, an ident is
never retracted (although it can be replaced). This approach is an
important part of allowing multiple consumers to share a schema
fragment as it migrates forward.
This fixes a limitation of the Clojure implementation, which did not
handle historical idents across knowledge base close and re-open.
The "entids" materialized view is naturally a slice of the "datoms"
table. The "idents" materialized view is a slice of the
"transactions" table. I hope that representing in this way, and
casting the problem in this light, might generalize to future
materialized views.
* Pre: Add DiffSet.
* Part 4: Collect mutations to a `Schema`.
I haven't taken your review comment about consuming AttributeBuilder
during each fluent function. If you read my response and still want
this, I'm happy to do it in review.
* Part 5: Handle :db/ident and :db.{install,alter}/attribute.
This "loops" the committed datoms out of the SQL store and back
through the metadata (schema, but in future also partition map)
processor. The metadata processor updates the schema and produces a
report of what changed; that report is then used to update the SQL
store. That update includes:
- the materialized views ("entids", "idents", and "schema");
- if needed, a subset of the datoms themselves (as flags change).
I've left a TODO for handling attribute retraction in the cases that
it makes sense. I expect that to be straight-forward.
* Review comment: Rename DiffSet to AddRetractAlterSet.
Also adds a little more commentary and a simple test.
* Review comment: Use ToIdent trait.
* Review comment: partially revert "Part 2: Maintain entids separately from idents."
This reverts commit 23a91df9c35e14398f2ddbd1ba25315821e67401.
Following our discussion, this removes the "entids" materialized
view. The next commit will remove historical idents from the "idents"
materialized view.
* Post: Use custom Either rather than std::result::Result.
This is not necessary, but it was suggested that we might be paying an
overhead creating Err instances while using error_chain. That seems
not to be the case, but this change shows that we don't actually use
any of the Result helper methods, so there's no reason to overload
Result. This change might avoid some future confusion, so I'm going
to land it anyway.
Signed-off-by: Nick Alexander <nalexander@mozilla.com>
* Review comment: Don't preserve historical idents.
* Review comment: More prepared statements when updating materialized views.
* Post: Test altering :db/cardinality and :db/unique.
These tests fail due to a Datomic limitation, namely that the marker
flag :db.alter/attribute can only be asserted once for an attribute!
That is, [:db.part/db :db.alter/attribute :attribute] will only be
transacted at most once. Since older versions of Datomic required the
:db.alter/attribute flag, I can only imagine they either never wrote
:db.alter/attribute to the store, or they handled it specially. I'll
need to remove the marker flag system from Mentat in order to address
this fundamental limitation.
* Post: Remove some more single quotes from error output.
* Post: Add assert_transact! macro to unwrap safely.
I was finding it very difficult to track unwrapping errors while
making changes, due to an underlying Mac OS X symbolication issue that
makes running tests with RUST_BACKTRACE=1 so slow that they all time
out.
* Post: Don't expect or recognize :db.{install,alter}/attribute.
I had this all working... except we will never see a repeated
`[:db.part/db :db.alter/attribute :attribute]` assertion in the store!
That means my approach would let you alter an attribute at most one
time. It's not worth hacking around this; it's better to just stop
expecting (and recognizing) the marker flags. (We have all the data
to distinguish the various cases that we need without the marker
flags.)
This brings Mentat in line with the thrust of newer Datomic versions,
but isn't compatible with Datomic, because (if I understand correctly)
Datomic automatically adds :db.{install,alter}/attribute assertions to
transactions.
I haven't purged the corresponding :db/ident and schema fragments just
yet:
- we might want them back
- we might want them in order to upgrade v1 and v2 databases to the
new on-disk layout we're fleshing out (v3?).
* Post: Don't make :db/unique :db.unique/* imply :db/index true.
This patch avoids a potential bug with the "schema" materialized view.
If :db/unique :db.unique/value implies :db/index true, then what
happens when you _retract_ :db.unique/value? I think Datomic defines
this in some way, but I really want the "schema" materialized view to
be a slice of "datoms" and not have these sort of ambiguities and
persistent effects. Therefore, to ensure that we don't retract a
schema characteristic and accidentally change more than we intended
to, this patch stops having any schema characteristic imply any other
schema characteristic(s). To achieve that, I added an
Option<Unique::{Value,Identity}> type to Attribute; this helps with
this patch, and also looks ahead to when we allow to retract
:db/unique attributes.
* Post: Allow to retract :db/ident.
* Post: Include more details about invalid schema changes.
The tests use strings, so they hide the chained errors which do in
fact provide more detail.
* Review comment: Fix outdated comment.
* Review comment: s/_SET/_SQL_LIST/.
* Review comment: Use a sub-select for checking cardinality.
This might be faster in practice.
* Review comment: Put `attribute::Unique` into its own namespace.
2017-03-20 20:18:59 +00:00
[ 100 :db / valueType :db . type / keyword ]
2019-07-17 15:46:47 +00:00
[ 100 :db / cardinality :db . cardinality / many ] ] "
) ;
Schema alteration. Fixes #294 and #295. (#370) r=rnewman
* Pre: Don't retract :db/ident in test.
Datomic (and eventually Mentat) don't allow to retract :db/ident in
this way, so this runs afoul of future work to support mutating
metadata.
* Pre: s/VALUETYPE/VALUE_TYPE/.
This is consistent with the capitalization (which is "valueType") and
the other identifier.
* Pre: Remove some single quotes from error output.
* Part 1: Make materialized views be uniform [e a v value_type_tag].
This looks ahead to a time when we could support arbitrary
user-defined materialized views. For now, the "idents" materialized
view is those datoms of the form [e :db/ident :namespaced/keyword] and
the "schema" materialized view is those datoms of the form [e a v]
where a is in a particular set of attributes that will become clear in
the following commits.
This change is not backwards compatible, so I'm removing the open
current (really, v2) test. It'll be re-instated when we get to
https://github.com/mozilla/mentat/issues/194.
* Pre: Map TypedValue::Ref to TypedValue::Keyword in debug output.
* Part 3: Separate `schema_to_mutate` from the `schema` used to interpret.
This is just to keep track of the expected changes during
bootstrapping. I want bootstrap metadata mutations to flow through
the same code path as metadata mutations during regular transactions;
by differentiating the schema used for interpretation from the schema
that will be updated I expect to be able to apply bootstrap metadata
mutations to an empty schema and have things like materialized views
created (using the regular code paths).
This commit has been re-ordered for conceptual clarity, but it won't
compile because it references the metadata module. It's possible to
make it compile -- the functionality is there in the schema module --
but it's not worth the rebasing effort until after review (and
possibly not even then, since we'll squash down to a single commit to
land).
* Part 2: Maintain entids separately from idents.
In order to support historical idents, we need to distinguish the
"current" map from entid -> ident from the "complete historical" map
ident -> entid. This is what Datomic does; in Datomic, an ident is
never retracted (although it can be replaced). This approach is an
important part of allowing multiple consumers to share a schema
fragment as it migrates forward.
This fixes a limitation of the Clojure implementation, which did not
handle historical idents across knowledge base close and re-open.
The "entids" materialized view is naturally a slice of the "datoms"
table. The "idents" materialized view is a slice of the
"transactions" table. I hope that representing in this way, and
casting the problem in this light, might generalize to future
materialized views.
* Pre: Add DiffSet.
* Part 4: Collect mutations to a `Schema`.
I haven't taken your review comment about consuming AttributeBuilder
during each fluent function. If you read my response and still want
this, I'm happy to do it in review.
* Part 5: Handle :db/ident and :db.{install,alter}/attribute.
This "loops" the committed datoms out of the SQL store and back
through the metadata (schema, but in future also partition map)
processor. The metadata processor updates the schema and produces a
report of what changed; that report is then used to update the SQL
store. That update includes:
- the materialized views ("entids", "idents", and "schema");
- if needed, a subset of the datoms themselves (as flags change).
I've left a TODO for handling attribute retraction in the cases that
it makes sense. I expect that to be straight-forward.
* Review comment: Rename DiffSet to AddRetractAlterSet.
Also adds a little more commentary and a simple test.
* Review comment: Use ToIdent trait.
* Review comment: partially revert "Part 2: Maintain entids separately from idents."
This reverts commit 23a91df9c35e14398f2ddbd1ba25315821e67401.
Following our discussion, this removes the "entids" materialized
view. The next commit will remove historical idents from the "idents"
materialized view.
* Post: Use custom Either rather than std::result::Result.
This is not necessary, but it was suggested that we might be paying an
overhead creating Err instances while using error_chain. That seems
not to be the case, but this change shows that we don't actually use
any of the Result helper methods, so there's no reason to overload
Result. This change might avoid some future confusion, so I'm going
to land it anyway.
Signed-off-by: Nick Alexander <nalexander@mozilla.com>
* Review comment: Don't preserve historical idents.
* Review comment: More prepared statements when updating materialized views.
* Post: Test altering :db/cardinality and :db/unique.
These tests fail due to a Datomic limitation, namely that the marker
flag :db.alter/attribute can only be asserted once for an attribute!
That is, [:db.part/db :db.alter/attribute :attribute] will only be
transacted at most once. Since older versions of Datomic required the
:db.alter/attribute flag, I can only imagine they either never wrote
:db.alter/attribute to the store, or they handled it specially. I'll
need to remove the marker flag system from Mentat in order to address
this fundamental limitation.
* Post: Remove some more single quotes from error output.
* Post: Add assert_transact! macro to unwrap safely.
I was finding it very difficult to track unwrapping errors while
making changes, due to an underlying Mac OS X symbolication issue that
makes running tests with RUST_BACKTRACE=1 so slow that they all time
out.
* Post: Don't expect or recognize :db.{install,alter}/attribute.
I had this all working... except we will never see a repeated
`[:db.part/db :db.alter/attribute :attribute]` assertion in the store!
That means my approach would let you alter an attribute at most one
time. It's not worth hacking around this; it's better to just stop
expecting (and recognizing) the marker flags. (We have all the data
to distinguish the various cases that we need without the marker
flags.)
This brings Mentat in line with the thrust of newer Datomic versions,
but isn't compatible with Datomic, because (if I understand correctly)
Datomic automatically adds :db.{install,alter}/attribute assertions to
transactions.
I haven't purged the corresponding :db/ident and schema fragments just
yet:
- we might want them back
- we might want them in order to upgrade v1 and v2 databases to the
new on-disk layout we're fleshing out (v3?).
* Post: Don't make :db/unique :db.unique/* imply :db/index true.
This patch avoids a potential bug with the "schema" materialized view.
If :db/unique :db.unique/value implies :db/index true, then what
happens when you _retract_ :db.unique/value? I think Datomic defines
this in some way, but I really want the "schema" materialized view to
be a slice of "datoms" and not have these sort of ambiguities and
persistent effects. Therefore, to ensure that we don't retract a
schema characteristic and accidentally change more than we intended
to, this patch stops having any schema characteristic imply any other
schema characteristic(s). To achieve that, I added an
Option<Unique::{Value,Identity}> type to Attribute; this helps with
this patch, and also looks ahead to when we allow to retract
:db/unique attributes.
* Post: Allow to retract :db/ident.
* Post: Include more details about invalid schema changes.
The tests use strings, so they hide the chained errors which do in
fact provide more detail.
* Review comment: Fix outdated comment.
* Review comment: s/_SET/_SQL_LIST/.
* Review comment: Use a sub-select for checking cardinality.
This might be faster in practice.
* Review comment: Put `attribute::Unique` into its own namespace.
2017-03-20 20:18:59 +00:00
// Let's check we actually have the schema characteristics we expect.
let attribute = conn . schema . attribute_for_entid ( 100 ) . unwrap ( ) . clone ( ) ;
assert_eq! ( attribute . value_type , ValueType ::Keyword ) ;
assert_eq! ( attribute . multival , true ) ;
assert_eq! ( attribute . fulltext , false ) ;
// Let's check that we can use the freshly altered attribute's new characteristic.
2019-07-17 15:46:47 +00:00
assert_transact! (
conn ,
" [[:db/add 101 100 :test/value1]
[ :db / add 101 :test / ident :test / value2 ] ] "
) ;
Schema alteration. Fixes #294 and #295. (#370) r=rnewman
* Pre: Don't retract :db/ident in test.
Datomic (and eventually Mentat) don't allow to retract :db/ident in
this way, so this runs afoul of future work to support mutating
metadata.
* Pre: s/VALUETYPE/VALUE_TYPE/.
This is consistent with the capitalization (which is "valueType") and
the other identifier.
* Pre: Remove some single quotes from error output.
* Part 1: Make materialized views be uniform [e a v value_type_tag].
This looks ahead to a time when we could support arbitrary
user-defined materialized views. For now, the "idents" materialized
view is those datoms of the form [e :db/ident :namespaced/keyword] and
the "schema" materialized view is those datoms of the form [e a v]
where a is in a particular set of attributes that will become clear in
the following commits.
This change is not backwards compatible, so I'm removing the open
current (really, v2) test. It'll be re-instated when we get to
https://github.com/mozilla/mentat/issues/194.
* Pre: Map TypedValue::Ref to TypedValue::Keyword in debug output.
* Part 3: Separate `schema_to_mutate` from the `schema` used to interpret.
This is just to keep track of the expected changes during
bootstrapping. I want bootstrap metadata mutations to flow through
the same code path as metadata mutations during regular transactions;
by differentiating the schema used for interpretation from the schema
that will be updated I expect to be able to apply bootstrap metadata
mutations to an empty schema and have things like materialized views
created (using the regular code paths).
This commit has been re-ordered for conceptual clarity, but it won't
compile because it references the metadata module. It's possible to
make it compile -- the functionality is there in the schema module --
but it's not worth the rebasing effort until after review (and
possibly not even then, since we'll squash down to a single commit to
land).
* Part 2: Maintain entids separately from idents.
In order to support historical idents, we need to distinguish the
"current" map from entid -> ident from the "complete historical" map
ident -> entid. This is what Datomic does; in Datomic, an ident is
never retracted (although it can be replaced). This approach is an
important part of allowing multiple consumers to share a schema
fragment as it migrates forward.
This fixes a limitation of the Clojure implementation, which did not
handle historical idents across knowledge base close and re-open.
The "entids" materialized view is naturally a slice of the "datoms"
table. The "idents" materialized view is a slice of the
"transactions" table. I hope that representing in this way, and
casting the problem in this light, might generalize to future
materialized views.
* Pre: Add DiffSet.
* Part 4: Collect mutations to a `Schema`.
I haven't taken your review comment about consuming AttributeBuilder
during each fluent function. If you read my response and still want
this, I'm happy to do it in review.
* Part 5: Handle :db/ident and :db.{install,alter}/attribute.
This "loops" the committed datoms out of the SQL store and back
through the metadata (schema, but in future also partition map)
processor. The metadata processor updates the schema and produces a
report of what changed; that report is then used to update the SQL
store. That update includes:
- the materialized views ("entids", "idents", and "schema");
- if needed, a subset of the datoms themselves (as flags change).
I've left a TODO for handling attribute retraction in the cases that
it makes sense. I expect that to be straight-forward.
* Review comment: Rename DiffSet to AddRetractAlterSet.
Also adds a little more commentary and a simple test.
* Review comment: Use ToIdent trait.
* Review comment: partially revert "Part 2: Maintain entids separately from idents."
This reverts commit 23a91df9c35e14398f2ddbd1ba25315821e67401.
Following our discussion, this removes the "entids" materialized
view. The next commit will remove historical idents from the "idents"
materialized view.
* Post: Use custom Either rather than std::result::Result.
This is not necessary, but it was suggested that we might be paying an
overhead creating Err instances while using error_chain. That seems
not to be the case, but this change shows that we don't actually use
any of the Result helper methods, so there's no reason to overload
Result. This change might avoid some future confusion, so I'm going
to land it anyway.
Signed-off-by: Nick Alexander <nalexander@mozilla.com>
* Review comment: Don't preserve historical idents.
* Review comment: More prepared statements when updating materialized views.
* Post: Test altering :db/cardinality and :db/unique.
These tests fail due to a Datomic limitation, namely that the marker
flag :db.alter/attribute can only be asserted once for an attribute!
That is, [:db.part/db :db.alter/attribute :attribute] will only be
transacted at most once. Since older versions of Datomic required the
:db.alter/attribute flag, I can only imagine they either never wrote
:db.alter/attribute to the store, or they handled it specially. I'll
need to remove the marker flag system from Mentat in order to address
this fundamental limitation.
* Post: Remove some more single quotes from error output.
* Post: Add assert_transact! macro to unwrap safely.
I was finding it very difficult to track unwrapping errors while
making changes, due to an underlying Mac OS X symbolication issue that
makes running tests with RUST_BACKTRACE=1 so slow that they all time
out.
* Post: Don't expect or recognize :db.{install,alter}/attribute.
I had this all working... except we will never see a repeated
`[:db.part/db :db.alter/attribute :attribute]` assertion in the store!
That means my approach would let you alter an attribute at most one
time. It's not worth hacking around this; it's better to just stop
expecting (and recognizing) the marker flags. (We have all the data
to distinguish the various cases that we need without the marker
flags.)
This brings Mentat in line with the thrust of newer Datomic versions,
but isn't compatible with Datomic, because (if I understand correctly)
Datomic automatically adds :db.{install,alter}/attribute assertions to
transactions.
I haven't purged the corresponding :db/ident and schema fragments just
yet:
- we might want them back
- we might want them in order to upgrade v1 and v2 databases to the
new on-disk layout we're fleshing out (v3?).
* Post: Don't make :db/unique :db.unique/* imply :db/index true.
This patch avoids a potential bug with the "schema" materialized view.
If :db/unique :db.unique/value implies :db/index true, then what
happens when you _retract_ :db.unique/value? I think Datomic defines
this in some way, but I really want the "schema" materialized view to
be a slice of "datoms" and not have these sort of ambiguities and
persistent effects. Therefore, to ensure that we don't retract a
schema characteristic and accidentally change more than we intended
to, this patch stops having any schema characteristic imply any other
schema characteristic(s). To achieve that, I added an
Option<Unique::{Value,Identity}> type to Attribute; this helps with
this patch, and also looks ahead to when we allow to retract
:db/unique attributes.
* Post: Allow to retract :db/ident.
* Post: Include more details about invalid schema changes.
The tests use strings, so they hide the chained errors which do in
fact provide more detail.
* Review comment: Fix outdated comment.
* Review comment: s/_SET/_SQL_LIST/.
* Review comment: Use a sub-select for checking cardinality.
This might be faster in practice.
* Review comment: Put `attribute::Unique` into its own namespace.
2017-03-20 20:18:59 +00:00
2019-07-17 15:46:47 +00:00
assert_matches! (
conn . last_transaction ( ) ,
" [[101 :test/ident :test/value1 ?tx true]
Schema alteration. Fixes #294 and #295. (#370) r=rnewman
* Pre: Don't retract :db/ident in test.
Datomic (and eventually Mentat) don't allow to retract :db/ident in
this way, so this runs afoul of future work to support mutating
metadata.
* Pre: s/VALUETYPE/VALUE_TYPE/.
This is consistent with the capitalization (which is "valueType") and
the other identifier.
* Pre: Remove some single quotes from error output.
* Part 1: Make materialized views be uniform [e a v value_type_tag].
This looks ahead to a time when we could support arbitrary
user-defined materialized views. For now, the "idents" materialized
view is those datoms of the form [e :db/ident :namespaced/keyword] and
the "schema" materialized view is those datoms of the form [e a v]
where a is in a particular set of attributes that will become clear in
the following commits.
This change is not backwards compatible, so I'm removing the open
current (really, v2) test. It'll be re-instated when we get to
https://github.com/mozilla/mentat/issues/194.
* Pre: Map TypedValue::Ref to TypedValue::Keyword in debug output.
* Part 3: Separate `schema_to_mutate` from the `schema` used to interpret.
This is just to keep track of the expected changes during
bootstrapping. I want bootstrap metadata mutations to flow through
the same code path as metadata mutations during regular transactions;
by differentiating the schema used for interpretation from the schema
that will be updated I expect to be able to apply bootstrap metadata
mutations to an empty schema and have things like materialized views
created (using the regular code paths).
This commit has been re-ordered for conceptual clarity, but it won't
compile because it references the metadata module. It's possible to
make it compile -- the functionality is there in the schema module --
but it's not worth the rebasing effort until after review (and
possibly not even then, since we'll squash down to a single commit to
land).
* Part 2: Maintain entids separately from idents.
In order to support historical idents, we need to distinguish the
"current" map from entid -> ident from the "complete historical" map
ident -> entid. This is what Datomic does; in Datomic, an ident is
never retracted (although it can be replaced). This approach is an
important part of allowing multiple consumers to share a schema
fragment as it migrates forward.
This fixes a limitation of the Clojure implementation, which did not
handle historical idents across knowledge base close and re-open.
The "entids" materialized view is naturally a slice of the "datoms"
table. The "idents" materialized view is a slice of the
"transactions" table. I hope that representing in this way, and
casting the problem in this light, might generalize to future
materialized views.
* Pre: Add DiffSet.
* Part 4: Collect mutations to a `Schema`.
I haven't taken your review comment about consuming AttributeBuilder
during each fluent function. If you read my response and still want
this, I'm happy to do it in review.
* Part 5: Handle :db/ident and :db.{install,alter}/attribute.
This "loops" the committed datoms out of the SQL store and back
through the metadata (schema, but in future also partition map)
processor. The metadata processor updates the schema and produces a
report of what changed; that report is then used to update the SQL
store. That update includes:
- the materialized views ("entids", "idents", and "schema");
- if needed, a subset of the datoms themselves (as flags change).
I've left a TODO for handling attribute retraction in the cases that
it makes sense. I expect that to be straight-forward.
* Review comment: Rename DiffSet to AddRetractAlterSet.
Also adds a little more commentary and a simple test.
* Review comment: Use ToIdent trait.
* Review comment: partially revert "Part 2: Maintain entids separately from idents."
This reverts commit 23a91df9c35e14398f2ddbd1ba25315821e67401.
Following our discussion, this removes the "entids" materialized
view. The next commit will remove historical idents from the "idents"
materialized view.
* Post: Use custom Either rather than std::result::Result.
This is not necessary, but it was suggested that we might be paying an
overhead creating Err instances while using error_chain. That seems
not to be the case, but this change shows that we don't actually use
any of the Result helper methods, so there's no reason to overload
Result. This change might avoid some future confusion, so I'm going
to land it anyway.
Signed-off-by: Nick Alexander <nalexander@mozilla.com>
* Review comment: Don't preserve historical idents.
* Review comment: More prepared statements when updating materialized views.
* Post: Test altering :db/cardinality and :db/unique.
These tests fail due to a Datomic limitation, namely that the marker
flag :db.alter/attribute can only be asserted once for an attribute!
That is, [:db.part/db :db.alter/attribute :attribute] will only be
transacted at most once. Since older versions of Datomic required the
:db.alter/attribute flag, I can only imagine they either never wrote
:db.alter/attribute to the store, or they handled it specially. I'll
need to remove the marker flag system from Mentat in order to address
this fundamental limitation.
* Post: Remove some more single quotes from error output.
* Post: Add assert_transact! macro to unwrap safely.
I was finding it very difficult to track unwrapping errors while
making changes, due to an underlying Mac OS X symbolication issue that
makes running tests with RUST_BACKTRACE=1 so slow that they all time
out.
* Post: Don't expect or recognize :db.{install,alter}/attribute.
I had this all working... except we will never see a repeated
`[:db.part/db :db.alter/attribute :attribute]` assertion in the store!
That means my approach would let you alter an attribute at most one
time. It's not worth hacking around this; it's better to just stop
expecting (and recognizing) the marker flags. (We have all the data
to distinguish the various cases that we need without the marker
flags.)
This brings Mentat in line with the thrust of newer Datomic versions,
but isn't compatible with Datomic, because (if I understand correctly)
Datomic automatically adds :db.{install,alter}/attribute assertions to
transactions.
I haven't purged the corresponding :db/ident and schema fragments just
yet:
- we might want them back
- we might want them in order to upgrade v1 and v2 databases to the
new on-disk layout we're fleshing out (v3?).
* Post: Don't make :db/unique :db.unique/* imply :db/index true.
This patch avoids a potential bug with the "schema" materialized view.
If :db/unique :db.unique/value implies :db/index true, then what
happens when you _retract_ :db.unique/value? I think Datomic defines
this in some way, but I really want the "schema" materialized view to
be a slice of "datoms" and not have these sort of ambiguities and
persistent effects. Therefore, to ensure that we don't retract a
schema characteristic and accidentally change more than we intended
to, this patch stops having any schema characteristic imply any other
schema characteristic(s). To achieve that, I added an
Option<Unique::{Value,Identity}> type to Attribute; this helps with
this patch, and also looks ahead to when we allow to retract
:db/unique attributes.
* Post: Allow to retract :db/ident.
* Post: Include more details about invalid schema changes.
The tests use strings, so they hide the chained errors which do in
fact provide more detail.
* Review comment: Fix outdated comment.
* Review comment: s/_SET/_SQL_LIST/.
* Review comment: Use a sub-select for checking cardinality.
This might be faster in practice.
* Review comment: Put `attribute::Unique` into its own namespace.
2017-03-20 20:18:59 +00:00
[ 101 :test / ident :test / value2 ? tx true ]
2019-07-17 15:46:47 +00:00
[ ? tx :db / txInstant ? ms ? tx true ] ] "
) ;
Schema alteration. Fixes #294 and #295. (#370) r=rnewman
* Pre: Don't retract :db/ident in test.
Datomic (and eventually Mentat) don't allow to retract :db/ident in
this way, so this runs afoul of future work to support mutating
metadata.
* Pre: s/VALUETYPE/VALUE_TYPE/.
This is consistent with the capitalization (which is "valueType") and
the other identifier.
* Pre: Remove some single quotes from error output.
* Part 1: Make materialized views be uniform [e a v value_type_tag].
This looks ahead to a time when we could support arbitrary
user-defined materialized views. For now, the "idents" materialized
view is those datoms of the form [e :db/ident :namespaced/keyword] and
the "schema" materialized view is those datoms of the form [e a v]
where a is in a particular set of attributes that will become clear in
the following commits.
This change is not backwards compatible, so I'm removing the open
current (really, v2) test. It'll be re-instated when we get to
https://github.com/mozilla/mentat/issues/194.
* Pre: Map TypedValue::Ref to TypedValue::Keyword in debug output.
* Part 3: Separate `schema_to_mutate` from the `schema` used to interpret.
This is just to keep track of the expected changes during
bootstrapping. I want bootstrap metadata mutations to flow through
the same code path as metadata mutations during regular transactions;
by differentiating the schema used for interpretation from the schema
that will be updated I expect to be able to apply bootstrap metadata
mutations to an empty schema and have things like materialized views
created (using the regular code paths).
This commit has been re-ordered for conceptual clarity, but it won't
compile because it references the metadata module. It's possible to
make it compile -- the functionality is there in the schema module --
but it's not worth the rebasing effort until after review (and
possibly not even then, since we'll squash down to a single commit to
land).
* Part 2: Maintain entids separately from idents.
In order to support historical idents, we need to distinguish the
"current" map from entid -> ident from the "complete historical" map
ident -> entid. This is what Datomic does; in Datomic, an ident is
never retracted (although it can be replaced). This approach is an
important part of allowing multiple consumers to share a schema
fragment as it migrates forward.
This fixes a limitation of the Clojure implementation, which did not
handle historical idents across knowledge base close and re-open.
The "entids" materialized view is naturally a slice of the "datoms"
table. The "idents" materialized view is a slice of the
"transactions" table. I hope that representing in this way, and
casting the problem in this light, might generalize to future
materialized views.
* Pre: Add DiffSet.
* Part 4: Collect mutations to a `Schema`.
I haven't taken your review comment about consuming AttributeBuilder
during each fluent function. If you read my response and still want
this, I'm happy to do it in review.
* Part 5: Handle :db/ident and :db.{install,alter}/attribute.
This "loops" the committed datoms out of the SQL store and back
through the metadata (schema, but in future also partition map)
processor. The metadata processor updates the schema and produces a
report of what changed; that report is then used to update the SQL
store. That update includes:
- the materialized views ("entids", "idents", and "schema");
- if needed, a subset of the datoms themselves (as flags change).
I've left a TODO for handling attribute retraction in the cases that
it makes sense. I expect that to be straight-forward.
* Review comment: Rename DiffSet to AddRetractAlterSet.
Also adds a little more commentary and a simple test.
* Review comment: Use ToIdent trait.
* Review comment: partially revert "Part 2: Maintain entids separately from idents."
This reverts commit 23a91df9c35e14398f2ddbd1ba25315821e67401.
Following our discussion, this removes the "entids" materialized
view. The next commit will remove historical idents from the "idents"
materialized view.
* Post: Use custom Either rather than std::result::Result.
This is not necessary, but it was suggested that we might be paying an
overhead creating Err instances while using error_chain. That seems
not to be the case, but this change shows that we don't actually use
any of the Result helper methods, so there's no reason to overload
Result. This change might avoid some future confusion, so I'm going
to land it anyway.
Signed-off-by: Nick Alexander <nalexander@mozilla.com>
* Review comment: Don't preserve historical idents.
* Review comment: More prepared statements when updating materialized views.
* Post: Test altering :db/cardinality and :db/unique.
These tests fail due to a Datomic limitation, namely that the marker
flag :db.alter/attribute can only be asserted once for an attribute!
That is, [:db.part/db :db.alter/attribute :attribute] will only be
transacted at most once. Since older versions of Datomic required the
:db.alter/attribute flag, I can only imagine they either never wrote
:db.alter/attribute to the store, or they handled it specially. I'll
need to remove the marker flag system from Mentat in order to address
this fundamental limitation.
* Post: Remove some more single quotes from error output.
* Post: Add assert_transact! macro to unwrap safely.
I was finding it very difficult to track unwrapping errors while
making changes, due to an underlying Mac OS X symbolication issue that
makes running tests with RUST_BACKTRACE=1 so slow that they all time
out.
* Post: Don't expect or recognize :db.{install,alter}/attribute.
I had this all working... except we will never see a repeated
`[:db.part/db :db.alter/attribute :attribute]` assertion in the store!
That means my approach would let you alter an attribute at most one
time. It's not worth hacking around this; it's better to just stop
expecting (and recognizing) the marker flags. (We have all the data
to distinguish the various cases that we need without the marker
flags.)
This brings Mentat in line with the thrust of newer Datomic versions,
but isn't compatible with Datomic, because (if I understand correctly)
Datomic automatically adds :db.{install,alter}/attribute assertions to
transactions.
I haven't purged the corresponding :db/ident and schema fragments just
yet:
- we might want them back
- we might want them in order to upgrade v1 and v2 databases to the
new on-disk layout we're fleshing out (v3?).
* Post: Don't make :db/unique :db.unique/* imply :db/index true.
This patch avoids a potential bug with the "schema" materialized view.
If :db/unique :db.unique/value implies :db/index true, then what
happens when you _retract_ :db.unique/value? I think Datomic defines
this in some way, but I really want the "schema" materialized view to
be a slice of "datoms" and not have these sort of ambiguities and
persistent effects. Therefore, to ensure that we don't retract a
schema characteristic and accidentally change more than we intended
to, this patch stops having any schema characteristic imply any other
schema characteristic(s). To achieve that, I added an
Option<Unique::{Value,Identity}> type to Attribute; this helps with
this patch, and also looks ahead to when we allow to retract
:db/unique attributes.
* Post: Allow to retract :db/ident.
* Post: Include more details about invalid schema changes.
The tests use strings, so they hide the chained errors which do in
fact provide more detail.
* Review comment: Fix outdated comment.
* Review comment: s/_SET/_SQL_LIST/.
* Review comment: Use a sub-select for checking cardinality.
This might be faster in practice.
* Review comment: Put `attribute::Unique` into its own namespace.
2017-03-20 20:18:59 +00:00
}
#[ test ]
fn test_db_ident ( ) {
let mut conn = TestConn ::default ( ) ;
// We can assert a new :db/ident.
assert_transact! ( conn , " [[:db/add 100 :db/ident :name/Ivan]] " ) ;
2019-07-17 15:46:47 +00:00
assert_matches! (
conn . last_transaction ( ) ,
" [[100 :db/ident :name/Ivan ?tx true]
[ ? tx :db / txInstant ? ms ? tx true ] ] "
) ;
assert_matches! ( conn . datoms ( ) , " [[100 :db/ident :name/Ivan]] " ) ;
assert_eq! (
conn . schema . entid_map . get ( & 100 ) . cloned ( ) . unwrap ( ) ,
to_namespaced_keyword ( " :name/Ivan " ) . unwrap ( )
) ;
assert_eq! (
conn . schema
. ident_map
. get ( & to_namespaced_keyword ( " :name/Ivan " ) . unwrap ( ) )
. cloned ( )
. unwrap ( ) ,
100
) ;
Schema alteration. Fixes #294 and #295. (#370) r=rnewman
* Pre: Don't retract :db/ident in test.
Datomic (and eventually Mentat) don't allow to retract :db/ident in
this way, so this runs afoul of future work to support mutating
metadata.
* Pre: s/VALUETYPE/VALUE_TYPE/.
This is consistent with the capitalization (which is "valueType") and
the other identifier.
* Pre: Remove some single quotes from error output.
* Part 1: Make materialized views be uniform [e a v value_type_tag].
This looks ahead to a time when we could support arbitrary
user-defined materialized views. For now, the "idents" materialized
view is those datoms of the form [e :db/ident :namespaced/keyword] and
the "schema" materialized view is those datoms of the form [e a v]
where a is in a particular set of attributes that will become clear in
the following commits.
This change is not backwards compatible, so I'm removing the open
current (really, v2) test. It'll be re-instated when we get to
https://github.com/mozilla/mentat/issues/194.
* Pre: Map TypedValue::Ref to TypedValue::Keyword in debug output.
* Part 3: Separate `schema_to_mutate` from the `schema` used to interpret.
This is just to keep track of the expected changes during
bootstrapping. I want bootstrap metadata mutations to flow through
the same code path as metadata mutations during regular transactions;
by differentiating the schema used for interpretation from the schema
that will be updated I expect to be able to apply bootstrap metadata
mutations to an empty schema and have things like materialized views
created (using the regular code paths).
This commit has been re-ordered for conceptual clarity, but it won't
compile because it references the metadata module. It's possible to
make it compile -- the functionality is there in the schema module --
but it's not worth the rebasing effort until after review (and
possibly not even then, since we'll squash down to a single commit to
land).
* Part 2: Maintain entids separately from idents.
In order to support historical idents, we need to distinguish the
"current" map from entid -> ident from the "complete historical" map
ident -> entid. This is what Datomic does; in Datomic, an ident is
never retracted (although it can be replaced). This approach is an
important part of allowing multiple consumers to share a schema
fragment as it migrates forward.
This fixes a limitation of the Clojure implementation, which did not
handle historical idents across knowledge base close and re-open.
The "entids" materialized view is naturally a slice of the "datoms"
table. The "idents" materialized view is a slice of the
"transactions" table. I hope that representing in this way, and
casting the problem in this light, might generalize to future
materialized views.
* Pre: Add DiffSet.
* Part 4: Collect mutations to a `Schema`.
I haven't taken your review comment about consuming AttributeBuilder
during each fluent function. If you read my response and still want
this, I'm happy to do it in review.
* Part 5: Handle :db/ident and :db.{install,alter}/attribute.
This "loops" the committed datoms out of the SQL store and back
through the metadata (schema, but in future also partition map)
processor. The metadata processor updates the schema and produces a
report of what changed; that report is then used to update the SQL
store. That update includes:
- the materialized views ("entids", "idents", and "schema");
- if needed, a subset of the datoms themselves (as flags change).
I've left a TODO for handling attribute retraction in the cases that
it makes sense. I expect that to be straight-forward.
* Review comment: Rename DiffSet to AddRetractAlterSet.
Also adds a little more commentary and a simple test.
* Review comment: Use ToIdent trait.
* Review comment: partially revert "Part 2: Maintain entids separately from idents."
This reverts commit 23a91df9c35e14398f2ddbd1ba25315821e67401.
Following our discussion, this removes the "entids" materialized
view. The next commit will remove historical idents from the "idents"
materialized view.
* Post: Use custom Either rather than std::result::Result.
This is not necessary, but it was suggested that we might be paying an
overhead creating Err instances while using error_chain. That seems
not to be the case, but this change shows that we don't actually use
any of the Result helper methods, so there's no reason to overload
Result. This change might avoid some future confusion, so I'm going
to land it anyway.
Signed-off-by: Nick Alexander <nalexander@mozilla.com>
* Review comment: Don't preserve historical idents.
* Review comment: More prepared statements when updating materialized views.
* Post: Test altering :db/cardinality and :db/unique.
These tests fail due to a Datomic limitation, namely that the marker
flag :db.alter/attribute can only be asserted once for an attribute!
That is, [:db.part/db :db.alter/attribute :attribute] will only be
transacted at most once. Since older versions of Datomic required the
:db.alter/attribute flag, I can only imagine they either never wrote
:db.alter/attribute to the store, or they handled it specially. I'll
need to remove the marker flag system from Mentat in order to address
this fundamental limitation.
* Post: Remove some more single quotes from error output.
* Post: Add assert_transact! macro to unwrap safely.
I was finding it very difficult to track unwrapping errors while
making changes, due to an underlying Mac OS X symbolication issue that
makes running tests with RUST_BACKTRACE=1 so slow that they all time
out.
* Post: Don't expect or recognize :db.{install,alter}/attribute.
I had this all working... except we will never see a repeated
`[:db.part/db :db.alter/attribute :attribute]` assertion in the store!
That means my approach would let you alter an attribute at most one
time. It's not worth hacking around this; it's better to just stop
expecting (and recognizing) the marker flags. (We have all the data
to distinguish the various cases that we need without the marker
flags.)
This brings Mentat in line with the thrust of newer Datomic versions,
but isn't compatible with Datomic, because (if I understand correctly)
Datomic automatically adds :db.{install,alter}/attribute assertions to
transactions.
I haven't purged the corresponding :db/ident and schema fragments just
yet:
- we might want them back
- we might want them in order to upgrade v1 and v2 databases to the
new on-disk layout we're fleshing out (v3?).
* Post: Don't make :db/unique :db.unique/* imply :db/index true.
This patch avoids a potential bug with the "schema" materialized view.
If :db/unique :db.unique/value implies :db/index true, then what
happens when you _retract_ :db.unique/value? I think Datomic defines
this in some way, but I really want the "schema" materialized view to
be a slice of "datoms" and not have these sort of ambiguities and
persistent effects. Therefore, to ensure that we don't retract a
schema characteristic and accidentally change more than we intended
to, this patch stops having any schema characteristic imply any other
schema characteristic(s). To achieve that, I added an
Option<Unique::{Value,Identity}> type to Attribute; this helps with
this patch, and also looks ahead to when we allow to retract
:db/unique attributes.
* Post: Allow to retract :db/ident.
* Post: Include more details about invalid schema changes.
The tests use strings, so they hide the chained errors which do in
fact provide more detail.
* Review comment: Fix outdated comment.
* Review comment: s/_SET/_SQL_LIST/.
* Review comment: Use a sub-select for checking cardinality.
This might be faster in practice.
* Review comment: Put `attribute::Unique` into its own namespace.
2017-03-20 20:18:59 +00:00
// We can re-assert an existing :db/ident.
assert_transact! ( conn , " [[:db/add 100 :db/ident :name/Ivan]] " ) ;
2019-07-17 15:46:47 +00:00
assert_matches! (
conn . last_transaction ( ) ,
" [[?tx :db/txInstant ?ms ?tx true]] "
) ;
assert_matches! ( conn . datoms ( ) , " [[100 :db/ident :name/Ivan]] " ) ;
assert_eq! (
conn . schema . entid_map . get ( & 100 ) . cloned ( ) . unwrap ( ) ,
to_namespaced_keyword ( " :name/Ivan " ) . unwrap ( )
) ;
assert_eq! (
conn . schema
. ident_map
. get ( & to_namespaced_keyword ( " :name/Ivan " ) . unwrap ( ) )
. cloned ( )
. unwrap ( ) ,
100
) ;
Schema alteration. Fixes #294 and #295. (#370) r=rnewman
* Pre: Don't retract :db/ident in test.
Datomic (and eventually Mentat) don't allow to retract :db/ident in
this way, so this runs afoul of future work to support mutating
metadata.
* Pre: s/VALUETYPE/VALUE_TYPE/.
This is consistent with the capitalization (which is "valueType") and
the other identifier.
* Pre: Remove some single quotes from error output.
* Part 1: Make materialized views be uniform [e a v value_type_tag].
This looks ahead to a time when we could support arbitrary
user-defined materialized views. For now, the "idents" materialized
view is those datoms of the form [e :db/ident :namespaced/keyword] and
the "schema" materialized view is those datoms of the form [e a v]
where a is in a particular set of attributes that will become clear in
the following commits.
This change is not backwards compatible, so I'm removing the open
current (really, v2) test. It'll be re-instated when we get to
https://github.com/mozilla/mentat/issues/194.
* Pre: Map TypedValue::Ref to TypedValue::Keyword in debug output.
* Part 3: Separate `schema_to_mutate` from the `schema` used to interpret.
This is just to keep track of the expected changes during
bootstrapping. I want bootstrap metadata mutations to flow through
the same code path as metadata mutations during regular transactions;
by differentiating the schema used for interpretation from the schema
that will be updated I expect to be able to apply bootstrap metadata
mutations to an empty schema and have things like materialized views
created (using the regular code paths).
This commit has been re-ordered for conceptual clarity, but it won't
compile because it references the metadata module. It's possible to
make it compile -- the functionality is there in the schema module --
but it's not worth the rebasing effort until after review (and
possibly not even then, since we'll squash down to a single commit to
land).
* Part 2: Maintain entids separately from idents.
In order to support historical idents, we need to distinguish the
"current" map from entid -> ident from the "complete historical" map
ident -> entid. This is what Datomic does; in Datomic, an ident is
never retracted (although it can be replaced). This approach is an
important part of allowing multiple consumers to share a schema
fragment as it migrates forward.
This fixes a limitation of the Clojure implementation, which did not
handle historical idents across knowledge base close and re-open.
The "entids" materialized view is naturally a slice of the "datoms"
table. The "idents" materialized view is a slice of the
"transactions" table. I hope that representing in this way, and
casting the problem in this light, might generalize to future
materialized views.
* Pre: Add DiffSet.
* Part 4: Collect mutations to a `Schema`.
I haven't taken your review comment about consuming AttributeBuilder
during each fluent function. If you read my response and still want
this, I'm happy to do it in review.
* Part 5: Handle :db/ident and :db.{install,alter}/attribute.
This "loops" the committed datoms out of the SQL store and back
through the metadata (schema, but in future also partition map)
processor. The metadata processor updates the schema and produces a
report of what changed; that report is then used to update the SQL
store. That update includes:
- the materialized views ("entids", "idents", and "schema");
- if needed, a subset of the datoms themselves (as flags change).
I've left a TODO for handling attribute retraction in the cases that
it makes sense. I expect that to be straight-forward.
* Review comment: Rename DiffSet to AddRetractAlterSet.
Also adds a little more commentary and a simple test.
* Review comment: Use ToIdent trait.
* Review comment: partially revert "Part 2: Maintain entids separately from idents."
This reverts commit 23a91df9c35e14398f2ddbd1ba25315821e67401.
Following our discussion, this removes the "entids" materialized
view. The next commit will remove historical idents from the "idents"
materialized view.
* Post: Use custom Either rather than std::result::Result.
This is not necessary, but it was suggested that we might be paying an
overhead creating Err instances while using error_chain. That seems
not to be the case, but this change shows that we don't actually use
any of the Result helper methods, so there's no reason to overload
Result. This change might avoid some future confusion, so I'm going
to land it anyway.
Signed-off-by: Nick Alexander <nalexander@mozilla.com>
* Review comment: Don't preserve historical idents.
* Review comment: More prepared statements when updating materialized views.
* Post: Test altering :db/cardinality and :db/unique.
These tests fail due to a Datomic limitation, namely that the marker
flag :db.alter/attribute can only be asserted once for an attribute!
That is, [:db.part/db :db.alter/attribute :attribute] will only be
transacted at most once. Since older versions of Datomic required the
:db.alter/attribute flag, I can only imagine they either never wrote
:db.alter/attribute to the store, or they handled it specially. I'll
need to remove the marker flag system from Mentat in order to address
this fundamental limitation.
* Post: Remove some more single quotes from error output.
* Post: Add assert_transact! macro to unwrap safely.
I was finding it very difficult to track unwrapping errors while
making changes, due to an underlying Mac OS X symbolication issue that
makes running tests with RUST_BACKTRACE=1 so slow that they all time
out.
* Post: Don't expect or recognize :db.{install,alter}/attribute.
I had this all working... except we will never see a repeated
`[:db.part/db :db.alter/attribute :attribute]` assertion in the store!
That means my approach would let you alter an attribute at most one
time. It's not worth hacking around this; it's better to just stop
expecting (and recognizing) the marker flags. (We have all the data
to distinguish the various cases that we need without the marker
flags.)
This brings Mentat in line with the thrust of newer Datomic versions,
but isn't compatible with Datomic, because (if I understand correctly)
Datomic automatically adds :db.{install,alter}/attribute assertions to
transactions.
I haven't purged the corresponding :db/ident and schema fragments just
yet:
- we might want them back
- we might want them in order to upgrade v1 and v2 databases to the
new on-disk layout we're fleshing out (v3?).
* Post: Don't make :db/unique :db.unique/* imply :db/index true.
This patch avoids a potential bug with the "schema" materialized view.
If :db/unique :db.unique/value implies :db/index true, then what
happens when you _retract_ :db.unique/value? I think Datomic defines
this in some way, but I really want the "schema" materialized view to
be a slice of "datoms" and not have these sort of ambiguities and
persistent effects. Therefore, to ensure that we don't retract a
schema characteristic and accidentally change more than we intended
to, this patch stops having any schema characteristic imply any other
schema characteristic(s). To achieve that, I added an
Option<Unique::{Value,Identity}> type to Attribute; this helps with
this patch, and also looks ahead to when we allow to retract
:db/unique attributes.
* Post: Allow to retract :db/ident.
* Post: Include more details about invalid schema changes.
The tests use strings, so they hide the chained errors which do in
fact provide more detail.
* Review comment: Fix outdated comment.
* Review comment: s/_SET/_SQL_LIST/.
* Review comment: Use a sub-select for checking cardinality.
This might be faster in practice.
* Review comment: Put `attribute::Unique` into its own namespace.
2017-03-20 20:18:59 +00:00
// We can alter an existing :db/ident to have a new keyword.
assert_transact! ( conn , " [[:db/add :name/Ivan :db/ident :name/Petr]] " ) ;
2019-07-17 15:46:47 +00:00
assert_matches! (
conn . last_transaction ( ) ,
" [[100 :db/ident :name/Ivan ?tx false]
Schema alteration. Fixes #294 and #295. (#370) r=rnewman
* Pre: Don't retract :db/ident in test.
Datomic (and eventually Mentat) don't allow to retract :db/ident in
this way, so this runs afoul of future work to support mutating
metadata.
* Pre: s/VALUETYPE/VALUE_TYPE/.
This is consistent with the capitalization (which is "valueType") and
the other identifier.
* Pre: Remove some single quotes from error output.
* Part 1: Make materialized views be uniform [e a v value_type_tag].
This looks ahead to a time when we could support arbitrary
user-defined materialized views. For now, the "idents" materialized
view is those datoms of the form [e :db/ident :namespaced/keyword] and
the "schema" materialized view is those datoms of the form [e a v]
where a is in a particular set of attributes that will become clear in
the following commits.
This change is not backwards compatible, so I'm removing the open
current (really, v2) test. It'll be re-instated when we get to
https://github.com/mozilla/mentat/issues/194.
* Pre: Map TypedValue::Ref to TypedValue::Keyword in debug output.
* Part 3: Separate `schema_to_mutate` from the `schema` used to interpret.
This is just to keep track of the expected changes during
bootstrapping. I want bootstrap metadata mutations to flow through
the same code path as metadata mutations during regular transactions;
by differentiating the schema used for interpretation from the schema
that will be updated I expect to be able to apply bootstrap metadata
mutations to an empty schema and have things like materialized views
created (using the regular code paths).
This commit has been re-ordered for conceptual clarity, but it won't
compile because it references the metadata module. It's possible to
make it compile -- the functionality is there in the schema module --
but it's not worth the rebasing effort until after review (and
possibly not even then, since we'll squash down to a single commit to
land).
* Part 2: Maintain entids separately from idents.
In order to support historical idents, we need to distinguish the
"current" map from entid -> ident from the "complete historical" map
ident -> entid. This is what Datomic does; in Datomic, an ident is
never retracted (although it can be replaced). This approach is an
important part of allowing multiple consumers to share a schema
fragment as it migrates forward.
This fixes a limitation of the Clojure implementation, which did not
handle historical idents across knowledge base close and re-open.
The "entids" materialized view is naturally a slice of the "datoms"
table. The "idents" materialized view is a slice of the
"transactions" table. I hope that representing in this way, and
casting the problem in this light, might generalize to future
materialized views.
* Pre: Add DiffSet.
* Part 4: Collect mutations to a `Schema`.
I haven't taken your review comment about consuming AttributeBuilder
during each fluent function. If you read my response and still want
this, I'm happy to do it in review.
* Part 5: Handle :db/ident and :db.{install,alter}/attribute.
This "loops" the committed datoms out of the SQL store and back
through the metadata (schema, but in future also partition map)
processor. The metadata processor updates the schema and produces a
report of what changed; that report is then used to update the SQL
store. That update includes:
- the materialized views ("entids", "idents", and "schema");
- if needed, a subset of the datoms themselves (as flags change).
I've left a TODO for handling attribute retraction in the cases that
it makes sense. I expect that to be straight-forward.
* Review comment: Rename DiffSet to AddRetractAlterSet.
Also adds a little more commentary and a simple test.
* Review comment: Use ToIdent trait.
* Review comment: partially revert "Part 2: Maintain entids separately from idents."
This reverts commit 23a91df9c35e14398f2ddbd1ba25315821e67401.
Following our discussion, this removes the "entids" materialized
view. The next commit will remove historical idents from the "idents"
materialized view.
* Post: Use custom Either rather than std::result::Result.
This is not necessary, but it was suggested that we might be paying an
overhead creating Err instances while using error_chain. That seems
not to be the case, but this change shows that we don't actually use
any of the Result helper methods, so there's no reason to overload
Result. This change might avoid some future confusion, so I'm going
to land it anyway.
Signed-off-by: Nick Alexander <nalexander@mozilla.com>
* Review comment: Don't preserve historical idents.
* Review comment: More prepared statements when updating materialized views.
* Post: Test altering :db/cardinality and :db/unique.
These tests fail due to a Datomic limitation, namely that the marker
flag :db.alter/attribute can only be asserted once for an attribute!
That is, [:db.part/db :db.alter/attribute :attribute] will only be
transacted at most once. Since older versions of Datomic required the
:db.alter/attribute flag, I can only imagine they either never wrote
:db.alter/attribute to the store, or they handled it specially. I'll
need to remove the marker flag system from Mentat in order to address
this fundamental limitation.
* Post: Remove some more single quotes from error output.
* Post: Add assert_transact! macro to unwrap safely.
I was finding it very difficult to track unwrapping errors while
making changes, due to an underlying Mac OS X symbolication issue that
makes running tests with RUST_BACKTRACE=1 so slow that they all time
out.
* Post: Don't expect or recognize :db.{install,alter}/attribute.
I had this all working... except we will never see a repeated
`[:db.part/db :db.alter/attribute :attribute]` assertion in the store!
That means my approach would let you alter an attribute at most one
time. It's not worth hacking around this; it's better to just stop
expecting (and recognizing) the marker flags. (We have all the data
to distinguish the various cases that we need without the marker
flags.)
This brings Mentat in line with the thrust of newer Datomic versions,
but isn't compatible with Datomic, because (if I understand correctly)
Datomic automatically adds :db.{install,alter}/attribute assertions to
transactions.
I haven't purged the corresponding :db/ident and schema fragments just
yet:
- we might want them back
- we might want them in order to upgrade v1 and v2 databases to the
new on-disk layout we're fleshing out (v3?).
* Post: Don't make :db/unique :db.unique/* imply :db/index true.
This patch avoids a potential bug with the "schema" materialized view.
If :db/unique :db.unique/value implies :db/index true, then what
happens when you _retract_ :db.unique/value? I think Datomic defines
this in some way, but I really want the "schema" materialized view to
be a slice of "datoms" and not have these sort of ambiguities and
persistent effects. Therefore, to ensure that we don't retract a
schema characteristic and accidentally change more than we intended
to, this patch stops having any schema characteristic imply any other
schema characteristic(s). To achieve that, I added an
Option<Unique::{Value,Identity}> type to Attribute; this helps with
this patch, and also looks ahead to when we allow to retract
:db/unique attributes.
* Post: Allow to retract :db/ident.
* Post: Include more details about invalid schema changes.
The tests use strings, so they hide the chained errors which do in
fact provide more detail.
* Review comment: Fix outdated comment.
* Review comment: s/_SET/_SQL_LIST/.
* Review comment: Use a sub-select for checking cardinality.
This might be faster in practice.
* Review comment: Put `attribute::Unique` into its own namespace.
2017-03-20 20:18:59 +00:00
[ 100 :db / ident :name / Petr ? tx true ]
2019-07-17 15:46:47 +00:00
[ ? tx :db / txInstant ? ms ? tx true ] ] "
) ;
assert_matches! ( conn . datoms ( ) , " [[100 :db/ident :name/Petr]] " ) ;
Schema alteration. Fixes #294 and #295. (#370) r=rnewman
* Pre: Don't retract :db/ident in test.
Datomic (and eventually Mentat) don't allow to retract :db/ident in
this way, so this runs afoul of future work to support mutating
metadata.
* Pre: s/VALUETYPE/VALUE_TYPE/.
This is consistent with the capitalization (which is "valueType") and
the other identifier.
* Pre: Remove some single quotes from error output.
* Part 1: Make materialized views be uniform [e a v value_type_tag].
This looks ahead to a time when we could support arbitrary
user-defined materialized views. For now, the "idents" materialized
view is those datoms of the form [e :db/ident :namespaced/keyword] and
the "schema" materialized view is those datoms of the form [e a v]
where a is in a particular set of attributes that will become clear in
the following commits.
This change is not backwards compatible, so I'm removing the open
current (really, v2) test. It'll be re-instated when we get to
https://github.com/mozilla/mentat/issues/194.
* Pre: Map TypedValue::Ref to TypedValue::Keyword in debug output.
* Part 3: Separate `schema_to_mutate` from the `schema` used to interpret.
This is just to keep track of the expected changes during
bootstrapping. I want bootstrap metadata mutations to flow through
the same code path as metadata mutations during regular transactions;
by differentiating the schema used for interpretation from the schema
that will be updated I expect to be able to apply bootstrap metadata
mutations to an empty schema and have things like materialized views
created (using the regular code paths).
This commit has been re-ordered for conceptual clarity, but it won't
compile because it references the metadata module. It's possible to
make it compile -- the functionality is there in the schema module --
but it's not worth the rebasing effort until after review (and
possibly not even then, since we'll squash down to a single commit to
land).
* Part 2: Maintain entids separately from idents.
In order to support historical idents, we need to distinguish the
"current" map from entid -> ident from the "complete historical" map
ident -> entid. This is what Datomic does; in Datomic, an ident is
never retracted (although it can be replaced). This approach is an
important part of allowing multiple consumers to share a schema
fragment as it migrates forward.
This fixes a limitation of the Clojure implementation, which did not
handle historical idents across knowledge base close and re-open.
The "entids" materialized view is naturally a slice of the "datoms"
table. The "idents" materialized view is a slice of the
"transactions" table. I hope that representing in this way, and
casting the problem in this light, might generalize to future
materialized views.
* Pre: Add DiffSet.
* Part 4: Collect mutations to a `Schema`.
I haven't taken your review comment about consuming AttributeBuilder
during each fluent function. If you read my response and still want
this, I'm happy to do it in review.
* Part 5: Handle :db/ident and :db.{install,alter}/attribute.
This "loops" the committed datoms out of the SQL store and back
through the metadata (schema, but in future also partition map)
processor. The metadata processor updates the schema and produces a
report of what changed; that report is then used to update the SQL
store. That update includes:
- the materialized views ("entids", "idents", and "schema");
- if needed, a subset of the datoms themselves (as flags change).
I've left a TODO for handling attribute retraction in the cases that
it makes sense. I expect that to be straight-forward.
* Review comment: Rename DiffSet to AddRetractAlterSet.
Also adds a little more commentary and a simple test.
* Review comment: Use ToIdent trait.
* Review comment: partially revert "Part 2: Maintain entids separately from idents."
This reverts commit 23a91df9c35e14398f2ddbd1ba25315821e67401.
Following our discussion, this removes the "entids" materialized
view. The next commit will remove historical idents from the "idents"
materialized view.
* Post: Use custom Either rather than std::result::Result.
This is not necessary, but it was suggested that we might be paying an
overhead creating Err instances while using error_chain. That seems
not to be the case, but this change shows that we don't actually use
any of the Result helper methods, so there's no reason to overload
Result. This change might avoid some future confusion, so I'm going
to land it anyway.
Signed-off-by: Nick Alexander <nalexander@mozilla.com>
* Review comment: Don't preserve historical idents.
* Review comment: More prepared statements when updating materialized views.
* Post: Test altering :db/cardinality and :db/unique.
These tests fail due to a Datomic limitation, namely that the marker
flag :db.alter/attribute can only be asserted once for an attribute!
That is, [:db.part/db :db.alter/attribute :attribute] will only be
transacted at most once. Since older versions of Datomic required the
:db.alter/attribute flag, I can only imagine they either never wrote
:db.alter/attribute to the store, or they handled it specially. I'll
need to remove the marker flag system from Mentat in order to address
this fundamental limitation.
* Post: Remove some more single quotes from error output.
* Post: Add assert_transact! macro to unwrap safely.
I was finding it very difficult to track unwrapping errors while
making changes, due to an underlying Mac OS X symbolication issue that
makes running tests with RUST_BACKTRACE=1 so slow that they all time
out.
* Post: Don't expect or recognize :db.{install,alter}/attribute.
I had this all working... except we will never see a repeated
`[:db.part/db :db.alter/attribute :attribute]` assertion in the store!
That means my approach would let you alter an attribute at most one
time. It's not worth hacking around this; it's better to just stop
expecting (and recognizing) the marker flags. (We have all the data
to distinguish the various cases that we need without the marker
flags.)
This brings Mentat in line with the thrust of newer Datomic versions,
but isn't compatible with Datomic, because (if I understand correctly)
Datomic automatically adds :db.{install,alter}/attribute assertions to
transactions.
I haven't purged the corresponding :db/ident and schema fragments just
yet:
- we might want them back
- we might want them in order to upgrade v1 and v2 databases to the
new on-disk layout we're fleshing out (v3?).
* Post: Don't make :db/unique :db.unique/* imply :db/index true.
This patch avoids a potential bug with the "schema" materialized view.
If :db/unique :db.unique/value implies :db/index true, then what
happens when you _retract_ :db.unique/value? I think Datomic defines
this in some way, but I really want the "schema" materialized view to
be a slice of "datoms" and not have these sort of ambiguities and
persistent effects. Therefore, to ensure that we don't retract a
schema characteristic and accidentally change more than we intended
to, this patch stops having any schema characteristic imply any other
schema characteristic(s). To achieve that, I added an
Option<Unique::{Value,Identity}> type to Attribute; this helps with
this patch, and also looks ahead to when we allow to retract
:db/unique attributes.
* Post: Allow to retract :db/ident.
* Post: Include more details about invalid schema changes.
The tests use strings, so they hide the chained errors which do in
fact provide more detail.
* Review comment: Fix outdated comment.
* Review comment: s/_SET/_SQL_LIST/.
* Review comment: Use a sub-select for checking cardinality.
This might be faster in practice.
* Review comment: Put `attribute::Unique` into its own namespace.
2017-03-20 20:18:59 +00:00
// Entid map is updated.
2019-07-17 15:46:47 +00:00
assert_eq! (
conn . schema . entid_map . get ( & 100 ) . cloned ( ) . unwrap ( ) ,
to_namespaced_keyword ( " :name/Petr " ) . unwrap ( )
) ;
Schema alteration. Fixes #294 and #295. (#370) r=rnewman
* Pre: Don't retract :db/ident in test.
Datomic (and eventually Mentat) don't allow to retract :db/ident in
this way, so this runs afoul of future work to support mutating
metadata.
* Pre: s/VALUETYPE/VALUE_TYPE/.
This is consistent with the capitalization (which is "valueType") and
the other identifier.
* Pre: Remove some single quotes from error output.
* Part 1: Make materialized views be uniform [e a v value_type_tag].
This looks ahead to a time when we could support arbitrary
user-defined materialized views. For now, the "idents" materialized
view is those datoms of the form [e :db/ident :namespaced/keyword] and
the "schema" materialized view is those datoms of the form [e a v]
where a is in a particular set of attributes that will become clear in
the following commits.
This change is not backwards compatible, so I'm removing the open
current (really, v2) test. It'll be re-instated when we get to
https://github.com/mozilla/mentat/issues/194.
* Pre: Map TypedValue::Ref to TypedValue::Keyword in debug output.
* Part 3: Separate `schema_to_mutate` from the `schema` used to interpret.
This is just to keep track of the expected changes during
bootstrapping. I want bootstrap metadata mutations to flow through
the same code path as metadata mutations during regular transactions;
by differentiating the schema used for interpretation from the schema
that will be updated I expect to be able to apply bootstrap metadata
mutations to an empty schema and have things like materialized views
created (using the regular code paths).
This commit has been re-ordered for conceptual clarity, but it won't
compile because it references the metadata module. It's possible to
make it compile -- the functionality is there in the schema module --
but it's not worth the rebasing effort until after review (and
possibly not even then, since we'll squash down to a single commit to
land).
* Part 2: Maintain entids separately from idents.
In order to support historical idents, we need to distinguish the
"current" map from entid -> ident from the "complete historical" map
ident -> entid. This is what Datomic does; in Datomic, an ident is
never retracted (although it can be replaced). This approach is an
important part of allowing multiple consumers to share a schema
fragment as it migrates forward.
This fixes a limitation of the Clojure implementation, which did not
handle historical idents across knowledge base close and re-open.
The "entids" materialized view is naturally a slice of the "datoms"
table. The "idents" materialized view is a slice of the
"transactions" table. I hope that representing in this way, and
casting the problem in this light, might generalize to future
materialized views.
* Pre: Add DiffSet.
* Part 4: Collect mutations to a `Schema`.
I haven't taken your review comment about consuming AttributeBuilder
during each fluent function. If you read my response and still want
this, I'm happy to do it in review.
* Part 5: Handle :db/ident and :db.{install,alter}/attribute.
This "loops" the committed datoms out of the SQL store and back
through the metadata (schema, but in future also partition map)
processor. The metadata processor updates the schema and produces a
report of what changed; that report is then used to update the SQL
store. That update includes:
- the materialized views ("entids", "idents", and "schema");
- if needed, a subset of the datoms themselves (as flags change).
I've left a TODO for handling attribute retraction in the cases that
it makes sense. I expect that to be straight-forward.
* Review comment: Rename DiffSet to AddRetractAlterSet.
Also adds a little more commentary and a simple test.
* Review comment: Use ToIdent trait.
* Review comment: partially revert "Part 2: Maintain entids separately from idents."
This reverts commit 23a91df9c35e14398f2ddbd1ba25315821e67401.
Following our discussion, this removes the "entids" materialized
view. The next commit will remove historical idents from the "idents"
materialized view.
* Post: Use custom Either rather than std::result::Result.
This is not necessary, but it was suggested that we might be paying an
overhead creating Err instances while using error_chain. That seems
not to be the case, but this change shows that we don't actually use
any of the Result helper methods, so there's no reason to overload
Result. This change might avoid some future confusion, so I'm going
to land it anyway.
Signed-off-by: Nick Alexander <nalexander@mozilla.com>
* Review comment: Don't preserve historical idents.
* Review comment: More prepared statements when updating materialized views.
* Post: Test altering :db/cardinality and :db/unique.
These tests fail due to a Datomic limitation, namely that the marker
flag :db.alter/attribute can only be asserted once for an attribute!
That is, [:db.part/db :db.alter/attribute :attribute] will only be
transacted at most once. Since older versions of Datomic required the
:db.alter/attribute flag, I can only imagine they either never wrote
:db.alter/attribute to the store, or they handled it specially. I'll
need to remove the marker flag system from Mentat in order to address
this fundamental limitation.
* Post: Remove some more single quotes from error output.
* Post: Add assert_transact! macro to unwrap safely.
I was finding it very difficult to track unwrapping errors while
making changes, due to an underlying Mac OS X symbolication issue that
makes running tests with RUST_BACKTRACE=1 so slow that they all time
out.
* Post: Don't expect or recognize :db.{install,alter}/attribute.
I had this all working... except we will never see a repeated
`[:db.part/db :db.alter/attribute :attribute]` assertion in the store!
That means my approach would let you alter an attribute at most one
time. It's not worth hacking around this; it's better to just stop
expecting (and recognizing) the marker flags. (We have all the data
to distinguish the various cases that we need without the marker
flags.)
This brings Mentat in line with the thrust of newer Datomic versions,
but isn't compatible with Datomic, because (if I understand correctly)
Datomic automatically adds :db.{install,alter}/attribute assertions to
transactions.
I haven't purged the corresponding :db/ident and schema fragments just
yet:
- we might want them back
- we might want them in order to upgrade v1 and v2 databases to the
new on-disk layout we're fleshing out (v3?).
* Post: Don't make :db/unique :db.unique/* imply :db/index true.
This patch avoids a potential bug with the "schema" materialized view.
If :db/unique :db.unique/value implies :db/index true, then what
happens when you _retract_ :db.unique/value? I think Datomic defines
this in some way, but I really want the "schema" materialized view to
be a slice of "datoms" and not have these sort of ambiguities and
persistent effects. Therefore, to ensure that we don't retract a
schema characteristic and accidentally change more than we intended
to, this patch stops having any schema characteristic imply any other
schema characteristic(s). To achieve that, I added an
Option<Unique::{Value,Identity}> type to Attribute; this helps with
this patch, and also looks ahead to when we allow to retract
:db/unique attributes.
* Post: Allow to retract :db/ident.
* Post: Include more details about invalid schema changes.
The tests use strings, so they hide the chained errors which do in
fact provide more detail.
* Review comment: Fix outdated comment.
* Review comment: s/_SET/_SQL_LIST/.
* Review comment: Use a sub-select for checking cardinality.
This might be faster in practice.
* Review comment: Put `attribute::Unique` into its own namespace.
2017-03-20 20:18:59 +00:00
// Ident map contains the new ident.
2019-07-17 15:46:47 +00:00
assert_eq! (
conn . schema
. ident_map
. get ( & to_namespaced_keyword ( " :name/Petr " ) . unwrap ( ) )
. cloned ( )
. unwrap ( ) ,
100
) ;
Schema alteration. Fixes #294 and #295. (#370) r=rnewman
* Pre: Don't retract :db/ident in test.
Datomic (and eventually Mentat) don't allow to retract :db/ident in
this way, so this runs afoul of future work to support mutating
metadata.
* Pre: s/VALUETYPE/VALUE_TYPE/.
This is consistent with the capitalization (which is "valueType") and
the other identifier.
* Pre: Remove some single quotes from error output.
* Part 1: Make materialized views be uniform [e a v value_type_tag].
This looks ahead to a time when we could support arbitrary
user-defined materialized views. For now, the "idents" materialized
view is those datoms of the form [e :db/ident :namespaced/keyword] and
the "schema" materialized view is those datoms of the form [e a v]
where a is in a particular set of attributes that will become clear in
the following commits.
This change is not backwards compatible, so I'm removing the open
current (really, v2) test. It'll be re-instated when we get to
https://github.com/mozilla/mentat/issues/194.
* Pre: Map TypedValue::Ref to TypedValue::Keyword in debug output.
* Part 3: Separate `schema_to_mutate` from the `schema` used to interpret.
This is just to keep track of the expected changes during
bootstrapping. I want bootstrap metadata mutations to flow through
the same code path as metadata mutations during regular transactions;
by differentiating the schema used for interpretation from the schema
that will be updated I expect to be able to apply bootstrap metadata
mutations to an empty schema and have things like materialized views
created (using the regular code paths).
This commit has been re-ordered for conceptual clarity, but it won't
compile because it references the metadata module. It's possible to
make it compile -- the functionality is there in the schema module --
but it's not worth the rebasing effort until after review (and
possibly not even then, since we'll squash down to a single commit to
land).
* Part 2: Maintain entids separately from idents.
In order to support historical idents, we need to distinguish the
"current" map from entid -> ident from the "complete historical" map
ident -> entid. This is what Datomic does; in Datomic, an ident is
never retracted (although it can be replaced). This approach is an
important part of allowing multiple consumers to share a schema
fragment as it migrates forward.
This fixes a limitation of the Clojure implementation, which did not
handle historical idents across knowledge base close and re-open.
The "entids" materialized view is naturally a slice of the "datoms"
table. The "idents" materialized view is a slice of the
"transactions" table. I hope that representing in this way, and
casting the problem in this light, might generalize to future
materialized views.
* Pre: Add DiffSet.
* Part 4: Collect mutations to a `Schema`.
I haven't taken your review comment about consuming AttributeBuilder
during each fluent function. If you read my response and still want
this, I'm happy to do it in review.
* Part 5: Handle :db/ident and :db.{install,alter}/attribute.
This "loops" the committed datoms out of the SQL store and back
through the metadata (schema, but in future also partition map)
processor. The metadata processor updates the schema and produces a
report of what changed; that report is then used to update the SQL
store. That update includes:
- the materialized views ("entids", "idents", and "schema");
- if needed, a subset of the datoms themselves (as flags change).
I've left a TODO for handling attribute retraction in the cases that
it makes sense. I expect that to be straight-forward.
* Review comment: Rename DiffSet to AddRetractAlterSet.
Also adds a little more commentary and a simple test.
* Review comment: Use ToIdent trait.
* Review comment: partially revert "Part 2: Maintain entids separately from idents."
This reverts commit 23a91df9c35e14398f2ddbd1ba25315821e67401.
Following our discussion, this removes the "entids" materialized
view. The next commit will remove historical idents from the "idents"
materialized view.
* Post: Use custom Either rather than std::result::Result.
This is not necessary, but it was suggested that we might be paying an
overhead creating Err instances while using error_chain. That seems
not to be the case, but this change shows that we don't actually use
any of the Result helper methods, so there's no reason to overload
Result. This change might avoid some future confusion, so I'm going
to land it anyway.
Signed-off-by: Nick Alexander <nalexander@mozilla.com>
* Review comment: Don't preserve historical idents.
* Review comment: More prepared statements when updating materialized views.
* Post: Test altering :db/cardinality and :db/unique.
These tests fail due to a Datomic limitation, namely that the marker
flag :db.alter/attribute can only be asserted once for an attribute!
That is, [:db.part/db :db.alter/attribute :attribute] will only be
transacted at most once. Since older versions of Datomic required the
:db.alter/attribute flag, I can only imagine they either never wrote
:db.alter/attribute to the store, or they handled it specially. I'll
need to remove the marker flag system from Mentat in order to address
this fundamental limitation.
* Post: Remove some more single quotes from error output.
* Post: Add assert_transact! macro to unwrap safely.
I was finding it very difficult to track unwrapping errors while
making changes, due to an underlying Mac OS X symbolication issue that
makes running tests with RUST_BACKTRACE=1 so slow that they all time
out.
* Post: Don't expect or recognize :db.{install,alter}/attribute.
I had this all working... except we will never see a repeated
`[:db.part/db :db.alter/attribute :attribute]` assertion in the store!
That means my approach would let you alter an attribute at most one
time. It's not worth hacking around this; it's better to just stop
expecting (and recognizing) the marker flags. (We have all the data
to distinguish the various cases that we need without the marker
flags.)
This brings Mentat in line with the thrust of newer Datomic versions,
but isn't compatible with Datomic, because (if I understand correctly)
Datomic automatically adds :db.{install,alter}/attribute assertions to
transactions.
I haven't purged the corresponding :db/ident and schema fragments just
yet:
- we might want them back
- we might want them in order to upgrade v1 and v2 databases to the
new on-disk layout we're fleshing out (v3?).
* Post: Don't make :db/unique :db.unique/* imply :db/index true.
This patch avoids a potential bug with the "schema" materialized view.
If :db/unique :db.unique/value implies :db/index true, then what
happens when you _retract_ :db.unique/value? I think Datomic defines
this in some way, but I really want the "schema" materialized view to
be a slice of "datoms" and not have these sort of ambiguities and
persistent effects. Therefore, to ensure that we don't retract a
schema characteristic and accidentally change more than we intended
to, this patch stops having any schema characteristic imply any other
schema characteristic(s). To achieve that, I added an
Option<Unique::{Value,Identity}> type to Attribute; this helps with
this patch, and also looks ahead to when we allow to retract
:db/unique attributes.
* Post: Allow to retract :db/ident.
* Post: Include more details about invalid schema changes.
The tests use strings, so they hide the chained errors which do in
fact provide more detail.
* Review comment: Fix outdated comment.
* Review comment: s/_SET/_SQL_LIST/.
* Review comment: Use a sub-select for checking cardinality.
This might be faster in practice.
* Review comment: Put `attribute::Unique` into its own namespace.
2017-03-20 20:18:59 +00:00
// Ident map no longer contains the old ident.
2019-07-17 15:46:47 +00:00
assert! ( conn
. schema
. ident_map
. get ( & to_namespaced_keyword ( " :name/Ivan " ) . unwrap ( ) )
. is_none ( ) ) ;
Schema alteration. Fixes #294 and #295. (#370) r=rnewman
* Pre: Don't retract :db/ident in test.
Datomic (and eventually Mentat) don't allow to retract :db/ident in
this way, so this runs afoul of future work to support mutating
metadata.
* Pre: s/VALUETYPE/VALUE_TYPE/.
This is consistent with the capitalization (which is "valueType") and
the other identifier.
* Pre: Remove some single quotes from error output.
* Part 1: Make materialized views be uniform [e a v value_type_tag].
This looks ahead to a time when we could support arbitrary
user-defined materialized views. For now, the "idents" materialized
view is those datoms of the form [e :db/ident :namespaced/keyword] and
the "schema" materialized view is those datoms of the form [e a v]
where a is in a particular set of attributes that will become clear in
the following commits.
This change is not backwards compatible, so I'm removing the open
current (really, v2) test. It'll be re-instated when we get to
https://github.com/mozilla/mentat/issues/194.
* Pre: Map TypedValue::Ref to TypedValue::Keyword in debug output.
* Part 3: Separate `schema_to_mutate` from the `schema` used to interpret.
This is just to keep track of the expected changes during
bootstrapping. I want bootstrap metadata mutations to flow through
the same code path as metadata mutations during regular transactions;
by differentiating the schema used for interpretation from the schema
that will be updated I expect to be able to apply bootstrap metadata
mutations to an empty schema and have things like materialized views
created (using the regular code paths).
This commit has been re-ordered for conceptual clarity, but it won't
compile because it references the metadata module. It's possible to
make it compile -- the functionality is there in the schema module --
but it's not worth the rebasing effort until after review (and
possibly not even then, since we'll squash down to a single commit to
land).
* Part 2: Maintain entids separately from idents.
In order to support historical idents, we need to distinguish the
"current" map from entid -> ident from the "complete historical" map
ident -> entid. This is what Datomic does; in Datomic, an ident is
never retracted (although it can be replaced). This approach is an
important part of allowing multiple consumers to share a schema
fragment as it migrates forward.
This fixes a limitation of the Clojure implementation, which did not
handle historical idents across knowledge base close and re-open.
The "entids" materialized view is naturally a slice of the "datoms"
table. The "idents" materialized view is a slice of the
"transactions" table. I hope that representing in this way, and
casting the problem in this light, might generalize to future
materialized views.
* Pre: Add DiffSet.
* Part 4: Collect mutations to a `Schema`.
I haven't taken your review comment about consuming AttributeBuilder
during each fluent function. If you read my response and still want
this, I'm happy to do it in review.
* Part 5: Handle :db/ident and :db.{install,alter}/attribute.
This "loops" the committed datoms out of the SQL store and back
through the metadata (schema, but in future also partition map)
processor. The metadata processor updates the schema and produces a
report of what changed; that report is then used to update the SQL
store. That update includes:
- the materialized views ("entids", "idents", and "schema");
- if needed, a subset of the datoms themselves (as flags change).
I've left a TODO for handling attribute retraction in the cases that
it makes sense. I expect that to be straight-forward.
* Review comment: Rename DiffSet to AddRetractAlterSet.
Also adds a little more commentary and a simple test.
* Review comment: Use ToIdent trait.
* Review comment: partially revert "Part 2: Maintain entids separately from idents."
This reverts commit 23a91df9c35e14398f2ddbd1ba25315821e67401.
Following our discussion, this removes the "entids" materialized
view. The next commit will remove historical idents from the "idents"
materialized view.
* Post: Use custom Either rather than std::result::Result.
This is not necessary, but it was suggested that we might be paying an
overhead creating Err instances while using error_chain. That seems
not to be the case, but this change shows that we don't actually use
any of the Result helper methods, so there's no reason to overload
Result. This change might avoid some future confusion, so I'm going
to land it anyway.
Signed-off-by: Nick Alexander <nalexander@mozilla.com>
* Review comment: Don't preserve historical idents.
* Review comment: More prepared statements when updating materialized views.
* Post: Test altering :db/cardinality and :db/unique.
These tests fail due to a Datomic limitation, namely that the marker
flag :db.alter/attribute can only be asserted once for an attribute!
That is, [:db.part/db :db.alter/attribute :attribute] will only be
transacted at most once. Since older versions of Datomic required the
:db.alter/attribute flag, I can only imagine they either never wrote
:db.alter/attribute to the store, or they handled it specially. I'll
need to remove the marker flag system from Mentat in order to address
this fundamental limitation.
* Post: Remove some more single quotes from error output.
* Post: Add assert_transact! macro to unwrap safely.
I was finding it very difficult to track unwrapping errors while
making changes, due to an underlying Mac OS X symbolication issue that
makes running tests with RUST_BACKTRACE=1 so slow that they all time
out.
* Post: Don't expect or recognize :db.{install,alter}/attribute.
I had this all working... except we will never see a repeated
`[:db.part/db :db.alter/attribute :attribute]` assertion in the store!
That means my approach would let you alter an attribute at most one
time. It's not worth hacking around this; it's better to just stop
expecting (and recognizing) the marker flags. (We have all the data
to distinguish the various cases that we need without the marker
flags.)
This brings Mentat in line with the thrust of newer Datomic versions,
but isn't compatible with Datomic, because (if I understand correctly)
Datomic automatically adds :db.{install,alter}/attribute assertions to
transactions.
I haven't purged the corresponding :db/ident and schema fragments just
yet:
- we might want them back
- we might want them in order to upgrade v1 and v2 databases to the
new on-disk layout we're fleshing out (v3?).
* Post: Don't make :db/unique :db.unique/* imply :db/index true.
This patch avoids a potential bug with the "schema" materialized view.
If :db/unique :db.unique/value implies :db/index true, then what
happens when you _retract_ :db.unique/value? I think Datomic defines
this in some way, but I really want the "schema" materialized view to
be a slice of "datoms" and not have these sort of ambiguities and
persistent effects. Therefore, to ensure that we don't retract a
schema characteristic and accidentally change more than we intended
to, this patch stops having any schema characteristic imply any other
schema characteristic(s). To achieve that, I added an
Option<Unique::{Value,Identity}> type to Attribute; this helps with
this patch, and also looks ahead to when we allow to retract
:db/unique attributes.
* Post: Allow to retract :db/ident.
* Post: Include more details about invalid schema changes.
The tests use strings, so they hide the chained errors which do in
fact provide more detail.
* Review comment: Fix outdated comment.
* Review comment: s/_SET/_SQL_LIST/.
* Review comment: Use a sub-select for checking cardinality.
This might be faster in practice.
* Review comment: Put `attribute::Unique` into its own namespace.
2017-03-20 20:18:59 +00:00
// We can re-purpose an old ident.
assert_transact! ( conn , " [[:db/add 101 :db/ident :name/Ivan]] " ) ;
2019-07-17 15:46:47 +00:00
assert_matches! (
conn . last_transaction ( ) ,
" [[101 :db/ident :name/Ivan ?tx true]
[ ? tx :db / txInstant ? ms ? tx true ] ] "
) ;
assert_matches! (
conn . datoms ( ) ,
" [[100 :db/ident :name/Petr]
[ 101 :db / ident :name / Ivan ] ] "
) ;
Schema alteration. Fixes #294 and #295. (#370) r=rnewman
* Pre: Don't retract :db/ident in test.
Datomic (and eventually Mentat) don't allow to retract :db/ident in
this way, so this runs afoul of future work to support mutating
metadata.
* Pre: s/VALUETYPE/VALUE_TYPE/.
This is consistent with the capitalization (which is "valueType") and
the other identifier.
* Pre: Remove some single quotes from error output.
* Part 1: Make materialized views be uniform [e a v value_type_tag].
This looks ahead to a time when we could support arbitrary
user-defined materialized views. For now, the "idents" materialized
view is those datoms of the form [e :db/ident :namespaced/keyword] and
the "schema" materialized view is those datoms of the form [e a v]
where a is in a particular set of attributes that will become clear in
the following commits.
This change is not backwards compatible, so I'm removing the open
current (really, v2) test. It'll be re-instated when we get to
https://github.com/mozilla/mentat/issues/194.
* Pre: Map TypedValue::Ref to TypedValue::Keyword in debug output.
* Part 3: Separate `schema_to_mutate` from the `schema` used to interpret.
This is just to keep track of the expected changes during
bootstrapping. I want bootstrap metadata mutations to flow through
the same code path as metadata mutations during regular transactions;
by differentiating the schema used for interpretation from the schema
that will be updated I expect to be able to apply bootstrap metadata
mutations to an empty schema and have things like materialized views
created (using the regular code paths).
This commit has been re-ordered for conceptual clarity, but it won't
compile because it references the metadata module. It's possible to
make it compile -- the functionality is there in the schema module --
but it's not worth the rebasing effort until after review (and
possibly not even then, since we'll squash down to a single commit to
land).
* Part 2: Maintain entids separately from idents.
In order to support historical idents, we need to distinguish the
"current" map from entid -> ident from the "complete historical" map
ident -> entid. This is what Datomic does; in Datomic, an ident is
never retracted (although it can be replaced). This approach is an
important part of allowing multiple consumers to share a schema
fragment as it migrates forward.
This fixes a limitation of the Clojure implementation, which did not
handle historical idents across knowledge base close and re-open.
The "entids" materialized view is naturally a slice of the "datoms"
table. The "idents" materialized view is a slice of the
"transactions" table. I hope that representing in this way, and
casting the problem in this light, might generalize to future
materialized views.
* Pre: Add DiffSet.
* Part 4: Collect mutations to a `Schema`.
I haven't taken your review comment about consuming AttributeBuilder
during each fluent function. If you read my response and still want
this, I'm happy to do it in review.
* Part 5: Handle :db/ident and :db.{install,alter}/attribute.
This "loops" the committed datoms out of the SQL store and back
through the metadata (schema, but in future also partition map)
processor. The metadata processor updates the schema and produces a
report of what changed; that report is then used to update the SQL
store. That update includes:
- the materialized views ("entids", "idents", and "schema");
- if needed, a subset of the datoms themselves (as flags change).
I've left a TODO for handling attribute retraction in the cases that
it makes sense. I expect that to be straight-forward.
* Review comment: Rename DiffSet to AddRetractAlterSet.
Also adds a little more commentary and a simple test.
* Review comment: Use ToIdent trait.
* Review comment: partially revert "Part 2: Maintain entids separately from idents."
This reverts commit 23a91df9c35e14398f2ddbd1ba25315821e67401.
Following our discussion, this removes the "entids" materialized
view. The next commit will remove historical idents from the "idents"
materialized view.
* Post: Use custom Either rather than std::result::Result.
This is not necessary, but it was suggested that we might be paying an
overhead creating Err instances while using error_chain. That seems
not to be the case, but this change shows that we don't actually use
any of the Result helper methods, so there's no reason to overload
Result. This change might avoid some future confusion, so I'm going
to land it anyway.
Signed-off-by: Nick Alexander <nalexander@mozilla.com>
* Review comment: Don't preserve historical idents.
* Review comment: More prepared statements when updating materialized views.
* Post: Test altering :db/cardinality and :db/unique.
These tests fail due to a Datomic limitation, namely that the marker
flag :db.alter/attribute can only be asserted once for an attribute!
That is, [:db.part/db :db.alter/attribute :attribute] will only be
transacted at most once. Since older versions of Datomic required the
:db.alter/attribute flag, I can only imagine they either never wrote
:db.alter/attribute to the store, or they handled it specially. I'll
need to remove the marker flag system from Mentat in order to address
this fundamental limitation.
* Post: Remove some more single quotes from error output.
* Post: Add assert_transact! macro to unwrap safely.
I was finding it very difficult to track unwrapping errors while
making changes, due to an underlying Mac OS X symbolication issue that
makes running tests with RUST_BACKTRACE=1 so slow that they all time
out.
* Post: Don't expect or recognize :db.{install,alter}/attribute.
I had this all working... except we will never see a repeated
`[:db.part/db :db.alter/attribute :attribute]` assertion in the store!
That means my approach would let you alter an attribute at most one
time. It's not worth hacking around this; it's better to just stop
expecting (and recognizing) the marker flags. (We have all the data
to distinguish the various cases that we need without the marker
flags.)
This brings Mentat in line with the thrust of newer Datomic versions,
but isn't compatible with Datomic, because (if I understand correctly)
Datomic automatically adds :db.{install,alter}/attribute assertions to
transactions.
I haven't purged the corresponding :db/ident and schema fragments just
yet:
- we might want them back
- we might want them in order to upgrade v1 and v2 databases to the
new on-disk layout we're fleshing out (v3?).
* Post: Don't make :db/unique :db.unique/* imply :db/index true.
This patch avoids a potential bug with the "schema" materialized view.
If :db/unique :db.unique/value implies :db/index true, then what
happens when you _retract_ :db.unique/value? I think Datomic defines
this in some way, but I really want the "schema" materialized view to
be a slice of "datoms" and not have these sort of ambiguities and
persistent effects. Therefore, to ensure that we don't retract a
schema characteristic and accidentally change more than we intended
to, this patch stops having any schema characteristic imply any other
schema characteristic(s). To achieve that, I added an
Option<Unique::{Value,Identity}> type to Attribute; this helps with
this patch, and also looks ahead to when we allow to retract
:db/unique attributes.
* Post: Allow to retract :db/ident.
* Post: Include more details about invalid schema changes.
The tests use strings, so they hide the chained errors which do in
fact provide more detail.
* Review comment: Fix outdated comment.
* Review comment: s/_SET/_SQL_LIST/.
* Review comment: Use a sub-select for checking cardinality.
This might be faster in practice.
* Review comment: Put `attribute::Unique` into its own namespace.
2017-03-20 20:18:59 +00:00
// Entid map contains both entids.
2019-07-17 15:46:47 +00:00
assert_eq! (
conn . schema . entid_map . get ( & 100 ) . cloned ( ) . unwrap ( ) ,
to_namespaced_keyword ( " :name/Petr " ) . unwrap ( )
) ;
assert_eq! (
conn . schema . entid_map . get ( & 101 ) . cloned ( ) . unwrap ( ) ,
to_namespaced_keyword ( " :name/Ivan " ) . unwrap ( )
) ;
Schema alteration. Fixes #294 and #295. (#370) r=rnewman
* Pre: Don't retract :db/ident in test.
Datomic (and eventually Mentat) don't allow to retract :db/ident in
this way, so this runs afoul of future work to support mutating
metadata.
* Pre: s/VALUETYPE/VALUE_TYPE/.
This is consistent with the capitalization (which is "valueType") and
the other identifier.
* Pre: Remove some single quotes from error output.
* Part 1: Make materialized views be uniform [e a v value_type_tag].
This looks ahead to a time when we could support arbitrary
user-defined materialized views. For now, the "idents" materialized
view is those datoms of the form [e :db/ident :namespaced/keyword] and
the "schema" materialized view is those datoms of the form [e a v]
where a is in a particular set of attributes that will become clear in
the following commits.
This change is not backwards compatible, so I'm removing the open
current (really, v2) test. It'll be re-instated when we get to
https://github.com/mozilla/mentat/issues/194.
* Pre: Map TypedValue::Ref to TypedValue::Keyword in debug output.
* Part 3: Separate `schema_to_mutate` from the `schema` used to interpret.
This is just to keep track of the expected changes during
bootstrapping. I want bootstrap metadata mutations to flow through
the same code path as metadata mutations during regular transactions;
by differentiating the schema used for interpretation from the schema
that will be updated I expect to be able to apply bootstrap metadata
mutations to an empty schema and have things like materialized views
created (using the regular code paths).
This commit has been re-ordered for conceptual clarity, but it won't
compile because it references the metadata module. It's possible to
make it compile -- the functionality is there in the schema module --
but it's not worth the rebasing effort until after review (and
possibly not even then, since we'll squash down to a single commit to
land).
* Part 2: Maintain entids separately from idents.
In order to support historical idents, we need to distinguish the
"current" map from entid -> ident from the "complete historical" map
ident -> entid. This is what Datomic does; in Datomic, an ident is
never retracted (although it can be replaced). This approach is an
important part of allowing multiple consumers to share a schema
fragment as it migrates forward.
This fixes a limitation of the Clojure implementation, which did not
handle historical idents across knowledge base close and re-open.
The "entids" materialized view is naturally a slice of the "datoms"
table. The "idents" materialized view is a slice of the
"transactions" table. I hope that representing in this way, and
casting the problem in this light, might generalize to future
materialized views.
* Pre: Add DiffSet.
* Part 4: Collect mutations to a `Schema`.
I haven't taken your review comment about consuming AttributeBuilder
during each fluent function. If you read my response and still want
this, I'm happy to do it in review.
* Part 5: Handle :db/ident and :db.{install,alter}/attribute.
This "loops" the committed datoms out of the SQL store and back
through the metadata (schema, but in future also partition map)
processor. The metadata processor updates the schema and produces a
report of what changed; that report is then used to update the SQL
store. That update includes:
- the materialized views ("entids", "idents", and "schema");
- if needed, a subset of the datoms themselves (as flags change).
I've left a TODO for handling attribute retraction in the cases that
it makes sense. I expect that to be straight-forward.
* Review comment: Rename DiffSet to AddRetractAlterSet.
Also adds a little more commentary and a simple test.
* Review comment: Use ToIdent trait.
* Review comment: partially revert "Part 2: Maintain entids separately from idents."
This reverts commit 23a91df9c35e14398f2ddbd1ba25315821e67401.
Following our discussion, this removes the "entids" materialized
view. The next commit will remove historical idents from the "idents"
materialized view.
* Post: Use custom Either rather than std::result::Result.
This is not necessary, but it was suggested that we might be paying an
overhead creating Err instances while using error_chain. That seems
not to be the case, but this change shows that we don't actually use
any of the Result helper methods, so there's no reason to overload
Result. This change might avoid some future confusion, so I'm going
to land it anyway.
Signed-off-by: Nick Alexander <nalexander@mozilla.com>
* Review comment: Don't preserve historical idents.
* Review comment: More prepared statements when updating materialized views.
* Post: Test altering :db/cardinality and :db/unique.
These tests fail due to a Datomic limitation, namely that the marker
flag :db.alter/attribute can only be asserted once for an attribute!
That is, [:db.part/db :db.alter/attribute :attribute] will only be
transacted at most once. Since older versions of Datomic required the
:db.alter/attribute flag, I can only imagine they either never wrote
:db.alter/attribute to the store, or they handled it specially. I'll
need to remove the marker flag system from Mentat in order to address
this fundamental limitation.
* Post: Remove some more single quotes from error output.
* Post: Add assert_transact! macro to unwrap safely.
I was finding it very difficult to track unwrapping errors while
making changes, due to an underlying Mac OS X symbolication issue that
makes running tests with RUST_BACKTRACE=1 so slow that they all time
out.
* Post: Don't expect or recognize :db.{install,alter}/attribute.
I had this all working... except we will never see a repeated
`[:db.part/db :db.alter/attribute :attribute]` assertion in the store!
That means my approach would let you alter an attribute at most one
time. It's not worth hacking around this; it's better to just stop
expecting (and recognizing) the marker flags. (We have all the data
to distinguish the various cases that we need without the marker
flags.)
This brings Mentat in line with the thrust of newer Datomic versions,
but isn't compatible with Datomic, because (if I understand correctly)
Datomic automatically adds :db.{install,alter}/attribute assertions to
transactions.
I haven't purged the corresponding :db/ident and schema fragments just
yet:
- we might want them back
- we might want them in order to upgrade v1 and v2 databases to the
new on-disk layout we're fleshing out (v3?).
* Post: Don't make :db/unique :db.unique/* imply :db/index true.
This patch avoids a potential bug with the "schema" materialized view.
If :db/unique :db.unique/value implies :db/index true, then what
happens when you _retract_ :db.unique/value? I think Datomic defines
this in some way, but I really want the "schema" materialized view to
be a slice of "datoms" and not have these sort of ambiguities and
persistent effects. Therefore, to ensure that we don't retract a
schema characteristic and accidentally change more than we intended
to, this patch stops having any schema characteristic imply any other
schema characteristic(s). To achieve that, I added an
Option<Unique::{Value,Identity}> type to Attribute; this helps with
this patch, and also looks ahead to when we allow to retract
:db/unique attributes.
* Post: Allow to retract :db/ident.
* Post: Include more details about invalid schema changes.
The tests use strings, so they hide the chained errors which do in
fact provide more detail.
* Review comment: Fix outdated comment.
* Review comment: s/_SET/_SQL_LIST/.
* Review comment: Use a sub-select for checking cardinality.
This might be faster in practice.
* Review comment: Put `attribute::Unique` into its own namespace.
2017-03-20 20:18:59 +00:00
// Ident map contains the new ident.
2019-07-17 15:46:47 +00:00
assert_eq! (
conn . schema
. ident_map
. get ( & to_namespaced_keyword ( " :name/Petr " ) . unwrap ( ) )
. cloned ( )
. unwrap ( ) ,
100
) ;
Schema alteration. Fixes #294 and #295. (#370) r=rnewman
* Pre: Don't retract :db/ident in test.
Datomic (and eventually Mentat) don't allow to retract :db/ident in
this way, so this runs afoul of future work to support mutating
metadata.
* Pre: s/VALUETYPE/VALUE_TYPE/.
This is consistent with the capitalization (which is "valueType") and
the other identifier.
* Pre: Remove some single quotes from error output.
* Part 1: Make materialized views be uniform [e a v value_type_tag].
This looks ahead to a time when we could support arbitrary
user-defined materialized views. For now, the "idents" materialized
view is those datoms of the form [e :db/ident :namespaced/keyword] and
the "schema" materialized view is those datoms of the form [e a v]
where a is in a particular set of attributes that will become clear in
the following commits.
This change is not backwards compatible, so I'm removing the open
current (really, v2) test. It'll be re-instated when we get to
https://github.com/mozilla/mentat/issues/194.
* Pre: Map TypedValue::Ref to TypedValue::Keyword in debug output.
* Part 3: Separate `schema_to_mutate` from the `schema` used to interpret.
This is just to keep track of the expected changes during
bootstrapping. I want bootstrap metadata mutations to flow through
the same code path as metadata mutations during regular transactions;
by differentiating the schema used for interpretation from the schema
that will be updated I expect to be able to apply bootstrap metadata
mutations to an empty schema and have things like materialized views
created (using the regular code paths).
This commit has been re-ordered for conceptual clarity, but it won't
compile because it references the metadata module. It's possible to
make it compile -- the functionality is there in the schema module --
but it's not worth the rebasing effort until after review (and
possibly not even then, since we'll squash down to a single commit to
land).
* Part 2: Maintain entids separately from idents.
In order to support historical idents, we need to distinguish the
"current" map from entid -> ident from the "complete historical" map
ident -> entid. This is what Datomic does; in Datomic, an ident is
never retracted (although it can be replaced). This approach is an
important part of allowing multiple consumers to share a schema
fragment as it migrates forward.
This fixes a limitation of the Clojure implementation, which did not
handle historical idents across knowledge base close and re-open.
The "entids" materialized view is naturally a slice of the "datoms"
table. The "idents" materialized view is a slice of the
"transactions" table. I hope that representing in this way, and
casting the problem in this light, might generalize to future
materialized views.
* Pre: Add DiffSet.
* Part 4: Collect mutations to a `Schema`.
I haven't taken your review comment about consuming AttributeBuilder
during each fluent function. If you read my response and still want
this, I'm happy to do it in review.
* Part 5: Handle :db/ident and :db.{install,alter}/attribute.
This "loops" the committed datoms out of the SQL store and back
through the metadata (schema, but in future also partition map)
processor. The metadata processor updates the schema and produces a
report of what changed; that report is then used to update the SQL
store. That update includes:
- the materialized views ("entids", "idents", and "schema");
- if needed, a subset of the datoms themselves (as flags change).
I've left a TODO for handling attribute retraction in the cases that
it makes sense. I expect that to be straight-forward.
* Review comment: Rename DiffSet to AddRetractAlterSet.
Also adds a little more commentary and a simple test.
* Review comment: Use ToIdent trait.
* Review comment: partially revert "Part 2: Maintain entids separately from idents."
This reverts commit 23a91df9c35e14398f2ddbd1ba25315821e67401.
Following our discussion, this removes the "entids" materialized
view. The next commit will remove historical idents from the "idents"
materialized view.
* Post: Use custom Either rather than std::result::Result.
This is not necessary, but it was suggested that we might be paying an
overhead creating Err instances while using error_chain. That seems
not to be the case, but this change shows that we don't actually use
any of the Result helper methods, so there's no reason to overload
Result. This change might avoid some future confusion, so I'm going
to land it anyway.
Signed-off-by: Nick Alexander <nalexander@mozilla.com>
* Review comment: Don't preserve historical idents.
* Review comment: More prepared statements when updating materialized views.
* Post: Test altering :db/cardinality and :db/unique.
These tests fail due to a Datomic limitation, namely that the marker
flag :db.alter/attribute can only be asserted once for an attribute!
That is, [:db.part/db :db.alter/attribute :attribute] will only be
transacted at most once. Since older versions of Datomic required the
:db.alter/attribute flag, I can only imagine they either never wrote
:db.alter/attribute to the store, or they handled it specially. I'll
need to remove the marker flag system from Mentat in order to address
this fundamental limitation.
* Post: Remove some more single quotes from error output.
* Post: Add assert_transact! macro to unwrap safely.
I was finding it very difficult to track unwrapping errors while
making changes, due to an underlying Mac OS X symbolication issue that
makes running tests with RUST_BACKTRACE=1 so slow that they all time
out.
* Post: Don't expect or recognize :db.{install,alter}/attribute.
I had this all working... except we will never see a repeated
`[:db.part/db :db.alter/attribute :attribute]` assertion in the store!
That means my approach would let you alter an attribute at most one
time. It's not worth hacking around this; it's better to just stop
expecting (and recognizing) the marker flags. (We have all the data
to distinguish the various cases that we need without the marker
flags.)
This brings Mentat in line with the thrust of newer Datomic versions,
but isn't compatible with Datomic, because (if I understand correctly)
Datomic automatically adds :db.{install,alter}/attribute assertions to
transactions.
I haven't purged the corresponding :db/ident and schema fragments just
yet:
- we might want them back
- we might want them in order to upgrade v1 and v2 databases to the
new on-disk layout we're fleshing out (v3?).
* Post: Don't make :db/unique :db.unique/* imply :db/index true.
This patch avoids a potential bug with the "schema" materialized view.
If :db/unique :db.unique/value implies :db/index true, then what
happens when you _retract_ :db.unique/value? I think Datomic defines
this in some way, but I really want the "schema" materialized view to
be a slice of "datoms" and not have these sort of ambiguities and
persistent effects. Therefore, to ensure that we don't retract a
schema characteristic and accidentally change more than we intended
to, this patch stops having any schema characteristic imply any other
schema characteristic(s). To achieve that, I added an
Option<Unique::{Value,Identity}> type to Attribute; this helps with
this patch, and also looks ahead to when we allow to retract
:db/unique attributes.
* Post: Allow to retract :db/ident.
* Post: Include more details about invalid schema changes.
The tests use strings, so they hide the chained errors which do in
fact provide more detail.
* Review comment: Fix outdated comment.
* Review comment: s/_SET/_SQL_LIST/.
* Review comment: Use a sub-select for checking cardinality.
This might be faster in practice.
* Review comment: Put `attribute::Unique` into its own namespace.
2017-03-20 20:18:59 +00:00
// Ident map contains the old ident, but re-purposed to the new entid.
2019-07-17 15:46:47 +00:00
assert_eq! (
conn . schema
. ident_map
. get ( & to_namespaced_keyword ( " :name/Ivan " ) . unwrap ( ) )
. cloned ( )
. unwrap ( ) ,
101
) ;
Schema alteration. Fixes #294 and #295. (#370) r=rnewman
* Pre: Don't retract :db/ident in test.
Datomic (and eventually Mentat) don't allow to retract :db/ident in
this way, so this runs afoul of future work to support mutating
metadata.
* Pre: s/VALUETYPE/VALUE_TYPE/.
This is consistent with the capitalization (which is "valueType") and
the other identifier.
* Pre: Remove some single quotes from error output.
* Part 1: Make materialized views be uniform [e a v value_type_tag].
This looks ahead to a time when we could support arbitrary
user-defined materialized views. For now, the "idents" materialized
view is those datoms of the form [e :db/ident :namespaced/keyword] and
the "schema" materialized view is those datoms of the form [e a v]
where a is in a particular set of attributes that will become clear in
the following commits.
This change is not backwards compatible, so I'm removing the open
current (really, v2) test. It'll be re-instated when we get to
https://github.com/mozilla/mentat/issues/194.
* Pre: Map TypedValue::Ref to TypedValue::Keyword in debug output.
* Part 3: Separate `schema_to_mutate` from the `schema` used to interpret.
This is just to keep track of the expected changes during
bootstrapping. I want bootstrap metadata mutations to flow through
the same code path as metadata mutations during regular transactions;
by differentiating the schema used for interpretation from the schema
that will be updated I expect to be able to apply bootstrap metadata
mutations to an empty schema and have things like materialized views
created (using the regular code paths).
This commit has been re-ordered for conceptual clarity, but it won't
compile because it references the metadata module. It's possible to
make it compile -- the functionality is there in the schema module --
but it's not worth the rebasing effort until after review (and
possibly not even then, since we'll squash down to a single commit to
land).
* Part 2: Maintain entids separately from idents.
In order to support historical idents, we need to distinguish the
"current" map from entid -> ident from the "complete historical" map
ident -> entid. This is what Datomic does; in Datomic, an ident is
never retracted (although it can be replaced). This approach is an
important part of allowing multiple consumers to share a schema
fragment as it migrates forward.
This fixes a limitation of the Clojure implementation, which did not
handle historical idents across knowledge base close and re-open.
The "entids" materialized view is naturally a slice of the "datoms"
table. The "idents" materialized view is a slice of the
"transactions" table. I hope that representing in this way, and
casting the problem in this light, might generalize to future
materialized views.
* Pre: Add DiffSet.
* Part 4: Collect mutations to a `Schema`.
I haven't taken your review comment about consuming AttributeBuilder
during each fluent function. If you read my response and still want
this, I'm happy to do it in review.
* Part 5: Handle :db/ident and :db.{install,alter}/attribute.
This "loops" the committed datoms out of the SQL store and back
through the metadata (schema, but in future also partition map)
processor. The metadata processor updates the schema and produces a
report of what changed; that report is then used to update the SQL
store. That update includes:
- the materialized views ("entids", "idents", and "schema");
- if needed, a subset of the datoms themselves (as flags change).
I've left a TODO for handling attribute retraction in the cases that
it makes sense. I expect that to be straight-forward.
* Review comment: Rename DiffSet to AddRetractAlterSet.
Also adds a little more commentary and a simple test.
* Review comment: Use ToIdent trait.
* Review comment: partially revert "Part 2: Maintain entids separately from idents."
This reverts commit 23a91df9c35e14398f2ddbd1ba25315821e67401.
Following our discussion, this removes the "entids" materialized
view. The next commit will remove historical idents from the "idents"
materialized view.
* Post: Use custom Either rather than std::result::Result.
This is not necessary, but it was suggested that we might be paying an
overhead creating Err instances while using error_chain. That seems
not to be the case, but this change shows that we don't actually use
any of the Result helper methods, so there's no reason to overload
Result. This change might avoid some future confusion, so I'm going
to land it anyway.
Signed-off-by: Nick Alexander <nalexander@mozilla.com>
* Review comment: Don't preserve historical idents.
* Review comment: More prepared statements when updating materialized views.
* Post: Test altering :db/cardinality and :db/unique.
These tests fail due to a Datomic limitation, namely that the marker
flag :db.alter/attribute can only be asserted once for an attribute!
That is, [:db.part/db :db.alter/attribute :attribute] will only be
transacted at most once. Since older versions of Datomic required the
:db.alter/attribute flag, I can only imagine they either never wrote
:db.alter/attribute to the store, or they handled it specially. I'll
need to remove the marker flag system from Mentat in order to address
this fundamental limitation.
* Post: Remove some more single quotes from error output.
* Post: Add assert_transact! macro to unwrap safely.
I was finding it very difficult to track unwrapping errors while
making changes, due to an underlying Mac OS X symbolication issue that
makes running tests with RUST_BACKTRACE=1 so slow that they all time
out.
* Post: Don't expect or recognize :db.{install,alter}/attribute.
I had this all working... except we will never see a repeated
`[:db.part/db :db.alter/attribute :attribute]` assertion in the store!
That means my approach would let you alter an attribute at most one
time. It's not worth hacking around this; it's better to just stop
expecting (and recognizing) the marker flags. (We have all the data
to distinguish the various cases that we need without the marker
flags.)
This brings Mentat in line with the thrust of newer Datomic versions,
but isn't compatible with Datomic, because (if I understand correctly)
Datomic automatically adds :db.{install,alter}/attribute assertions to
transactions.
I haven't purged the corresponding :db/ident and schema fragments just
yet:
- we might want them back
- we might want them in order to upgrade v1 and v2 databases to the
new on-disk layout we're fleshing out (v3?).
* Post: Don't make :db/unique :db.unique/* imply :db/index true.
This patch avoids a potential bug with the "schema" materialized view.
If :db/unique :db.unique/value implies :db/index true, then what
happens when you _retract_ :db.unique/value? I think Datomic defines
this in some way, but I really want the "schema" materialized view to
be a slice of "datoms" and not have these sort of ambiguities and
persistent effects. Therefore, to ensure that we don't retract a
schema characteristic and accidentally change more than we intended
to, this patch stops having any schema characteristic imply any other
schema characteristic(s). To achieve that, I added an
Option<Unique::{Value,Identity}> type to Attribute; this helps with
this patch, and also looks ahead to when we allow to retract
:db/unique attributes.
* Post: Allow to retract :db/ident.
* Post: Include more details about invalid schema changes.
The tests use strings, so they hide the chained errors which do in
fact provide more detail.
* Review comment: Fix outdated comment.
* Review comment: s/_SET/_SQL_LIST/.
* Review comment: Use a sub-select for checking cardinality.
This might be faster in practice.
* Review comment: Put `attribute::Unique` into its own namespace.
2017-03-20 20:18:59 +00:00
// We can retract an existing :db/ident.
assert_transact! ( conn , " [[:db/retract :name/Petr :db/ident :name/Petr]] " ) ;
// It's really gone.
assert! ( conn . schema . entid_map . get ( & 100 ) . is_none ( ) ) ;
2019-07-17 15:46:47 +00:00
assert! ( conn
. schema
. ident_map
. get ( & to_namespaced_keyword ( " :name/Petr " ) . unwrap ( ) )
. is_none ( ) ) ;
Schema alteration. Fixes #294 and #295. (#370) r=rnewman
* Pre: Don't retract :db/ident in test.
Datomic (and eventually Mentat) don't allow to retract :db/ident in
this way, so this runs afoul of future work to support mutating
metadata.
* Pre: s/VALUETYPE/VALUE_TYPE/.
This is consistent with the capitalization (which is "valueType") and
the other identifier.
* Pre: Remove some single quotes from error output.
* Part 1: Make materialized views be uniform [e a v value_type_tag].
This looks ahead to a time when we could support arbitrary
user-defined materialized views. For now, the "idents" materialized
view is those datoms of the form [e :db/ident :namespaced/keyword] and
the "schema" materialized view is those datoms of the form [e a v]
where a is in a particular set of attributes that will become clear in
the following commits.
This change is not backwards compatible, so I'm removing the open
current (really, v2) test. It'll be re-instated when we get to
https://github.com/mozilla/mentat/issues/194.
* Pre: Map TypedValue::Ref to TypedValue::Keyword in debug output.
* Part 3: Separate `schema_to_mutate` from the `schema` used to interpret.
This is just to keep track of the expected changes during
bootstrapping. I want bootstrap metadata mutations to flow through
the same code path as metadata mutations during regular transactions;
by differentiating the schema used for interpretation from the schema
that will be updated I expect to be able to apply bootstrap metadata
mutations to an empty schema and have things like materialized views
created (using the regular code paths).
This commit has been re-ordered for conceptual clarity, but it won't
compile because it references the metadata module. It's possible to
make it compile -- the functionality is there in the schema module --
but it's not worth the rebasing effort until after review (and
possibly not even then, since we'll squash down to a single commit to
land).
* Part 2: Maintain entids separately from idents.
In order to support historical idents, we need to distinguish the
"current" map from entid -> ident from the "complete historical" map
ident -> entid. This is what Datomic does; in Datomic, an ident is
never retracted (although it can be replaced). This approach is an
important part of allowing multiple consumers to share a schema
fragment as it migrates forward.
This fixes a limitation of the Clojure implementation, which did not
handle historical idents across knowledge base close and re-open.
The "entids" materialized view is naturally a slice of the "datoms"
table. The "idents" materialized view is a slice of the
"transactions" table. I hope that representing in this way, and
casting the problem in this light, might generalize to future
materialized views.
* Pre: Add DiffSet.
* Part 4: Collect mutations to a `Schema`.
I haven't taken your review comment about consuming AttributeBuilder
during each fluent function. If you read my response and still want
this, I'm happy to do it in review.
* Part 5: Handle :db/ident and :db.{install,alter}/attribute.
This "loops" the committed datoms out of the SQL store and back
through the metadata (schema, but in future also partition map)
processor. The metadata processor updates the schema and produces a
report of what changed; that report is then used to update the SQL
store. That update includes:
- the materialized views ("entids", "idents", and "schema");
- if needed, a subset of the datoms themselves (as flags change).
I've left a TODO for handling attribute retraction in the cases that
it makes sense. I expect that to be straight-forward.
* Review comment: Rename DiffSet to AddRetractAlterSet.
Also adds a little more commentary and a simple test.
* Review comment: Use ToIdent trait.
* Review comment: partially revert "Part 2: Maintain entids separately from idents."
This reverts commit 23a91df9c35e14398f2ddbd1ba25315821e67401.
Following our discussion, this removes the "entids" materialized
view. The next commit will remove historical idents from the "idents"
materialized view.
* Post: Use custom Either rather than std::result::Result.
This is not necessary, but it was suggested that we might be paying an
overhead creating Err instances while using error_chain. That seems
not to be the case, but this change shows that we don't actually use
any of the Result helper methods, so there's no reason to overload
Result. This change might avoid some future confusion, so I'm going
to land it anyway.
Signed-off-by: Nick Alexander <nalexander@mozilla.com>
* Review comment: Don't preserve historical idents.
* Review comment: More prepared statements when updating materialized views.
* Post: Test altering :db/cardinality and :db/unique.
These tests fail due to a Datomic limitation, namely that the marker
flag :db.alter/attribute can only be asserted once for an attribute!
That is, [:db.part/db :db.alter/attribute :attribute] will only be
transacted at most once. Since older versions of Datomic required the
:db.alter/attribute flag, I can only imagine they either never wrote
:db.alter/attribute to the store, or they handled it specially. I'll
need to remove the marker flag system from Mentat in order to address
this fundamental limitation.
* Post: Remove some more single quotes from error output.
* Post: Add assert_transact! macro to unwrap safely.
I was finding it very difficult to track unwrapping errors while
making changes, due to an underlying Mac OS X symbolication issue that
makes running tests with RUST_BACKTRACE=1 so slow that they all time
out.
* Post: Don't expect or recognize :db.{install,alter}/attribute.
I had this all working... except we will never see a repeated
`[:db.part/db :db.alter/attribute :attribute]` assertion in the store!
That means my approach would let you alter an attribute at most one
time. It's not worth hacking around this; it's better to just stop
expecting (and recognizing) the marker flags. (We have all the data
to distinguish the various cases that we need without the marker
flags.)
This brings Mentat in line with the thrust of newer Datomic versions,
but isn't compatible with Datomic, because (if I understand correctly)
Datomic automatically adds :db.{install,alter}/attribute assertions to
transactions.
I haven't purged the corresponding :db/ident and schema fragments just
yet:
- we might want them back
- we might want them in order to upgrade v1 and v2 databases to the
new on-disk layout we're fleshing out (v3?).
* Post: Don't make :db/unique :db.unique/* imply :db/index true.
This patch avoids a potential bug with the "schema" materialized view.
If :db/unique :db.unique/value implies :db/index true, then what
happens when you _retract_ :db.unique/value? I think Datomic defines
this in some way, but I really want the "schema" materialized view to
be a slice of "datoms" and not have these sort of ambiguities and
persistent effects. Therefore, to ensure that we don't retract a
schema characteristic and accidentally change more than we intended
to, this patch stops having any schema characteristic imply any other
schema characteristic(s). To achieve that, I added an
Option<Unique::{Value,Identity}> type to Attribute; this helps with
this patch, and also looks ahead to when we allow to retract
:db/unique attributes.
* Post: Allow to retract :db/ident.
* Post: Include more details about invalid schema changes.
The tests use strings, so they hide the chained errors which do in
fact provide more detail.
* Review comment: Fix outdated comment.
* Review comment: s/_SET/_SQL_LIST/.
* Review comment: Use a sub-select for checking cardinality.
This might be faster in practice.
* Review comment: Put `attribute::Unique` into its own namespace.
2017-03-20 20:18:59 +00:00
}
#[ test ]
fn test_db_alter_cardinality ( ) {
let mut conn = TestConn ::default ( ) ;
// Start by installing a :db.cardinality/one attribute.
2019-07-17 15:46:47 +00:00
assert_transact! (
conn ,
" [[:db/add 100 :db/ident :test/ident]
Schema alteration. Fixes #294 and #295. (#370) r=rnewman
* Pre: Don't retract :db/ident in test.
Datomic (and eventually Mentat) don't allow to retract :db/ident in
this way, so this runs afoul of future work to support mutating
metadata.
* Pre: s/VALUETYPE/VALUE_TYPE/.
This is consistent with the capitalization (which is "valueType") and
the other identifier.
* Pre: Remove some single quotes from error output.
* Part 1: Make materialized views be uniform [e a v value_type_tag].
This looks ahead to a time when we could support arbitrary
user-defined materialized views. For now, the "idents" materialized
view is those datoms of the form [e :db/ident :namespaced/keyword] and
the "schema" materialized view is those datoms of the form [e a v]
where a is in a particular set of attributes that will become clear in
the following commits.
This change is not backwards compatible, so I'm removing the open
current (really, v2) test. It'll be re-instated when we get to
https://github.com/mozilla/mentat/issues/194.
* Pre: Map TypedValue::Ref to TypedValue::Keyword in debug output.
* Part 3: Separate `schema_to_mutate` from the `schema` used to interpret.
This is just to keep track of the expected changes during
bootstrapping. I want bootstrap metadata mutations to flow through
the same code path as metadata mutations during regular transactions;
by differentiating the schema used for interpretation from the schema
that will be updated I expect to be able to apply bootstrap metadata
mutations to an empty schema and have things like materialized views
created (using the regular code paths).
This commit has been re-ordered for conceptual clarity, but it won't
compile because it references the metadata module. It's possible to
make it compile -- the functionality is there in the schema module --
but it's not worth the rebasing effort until after review (and
possibly not even then, since we'll squash down to a single commit to
land).
* Part 2: Maintain entids separately from idents.
In order to support historical idents, we need to distinguish the
"current" map from entid -> ident from the "complete historical" map
ident -> entid. This is what Datomic does; in Datomic, an ident is
never retracted (although it can be replaced). This approach is an
important part of allowing multiple consumers to share a schema
fragment as it migrates forward.
This fixes a limitation of the Clojure implementation, which did not
handle historical idents across knowledge base close and re-open.
The "entids" materialized view is naturally a slice of the "datoms"
table. The "idents" materialized view is a slice of the
"transactions" table. I hope that representing in this way, and
casting the problem in this light, might generalize to future
materialized views.
* Pre: Add DiffSet.
* Part 4: Collect mutations to a `Schema`.
I haven't taken your review comment about consuming AttributeBuilder
during each fluent function. If you read my response and still want
this, I'm happy to do it in review.
* Part 5: Handle :db/ident and :db.{install,alter}/attribute.
This "loops" the committed datoms out of the SQL store and back
through the metadata (schema, but in future also partition map)
processor. The metadata processor updates the schema and produces a
report of what changed; that report is then used to update the SQL
store. That update includes:
- the materialized views ("entids", "idents", and "schema");
- if needed, a subset of the datoms themselves (as flags change).
I've left a TODO for handling attribute retraction in the cases that
it makes sense. I expect that to be straight-forward.
* Review comment: Rename DiffSet to AddRetractAlterSet.
Also adds a little more commentary and a simple test.
* Review comment: Use ToIdent trait.
* Review comment: partially revert "Part 2: Maintain entids separately from idents."
This reverts commit 23a91df9c35e14398f2ddbd1ba25315821e67401.
Following our discussion, this removes the "entids" materialized
view. The next commit will remove historical idents from the "idents"
materialized view.
* Post: Use custom Either rather than std::result::Result.
This is not necessary, but it was suggested that we might be paying an
overhead creating Err instances while using error_chain. That seems
not to be the case, but this change shows that we don't actually use
any of the Result helper methods, so there's no reason to overload
Result. This change might avoid some future confusion, so I'm going
to land it anyway.
Signed-off-by: Nick Alexander <nalexander@mozilla.com>
* Review comment: Don't preserve historical idents.
* Review comment: More prepared statements when updating materialized views.
* Post: Test altering :db/cardinality and :db/unique.
These tests fail due to a Datomic limitation, namely that the marker
flag :db.alter/attribute can only be asserted once for an attribute!
That is, [:db.part/db :db.alter/attribute :attribute] will only be
transacted at most once. Since older versions of Datomic required the
:db.alter/attribute flag, I can only imagine they either never wrote
:db.alter/attribute to the store, or they handled it specially. I'll
need to remove the marker flag system from Mentat in order to address
this fundamental limitation.
* Post: Remove some more single quotes from error output.
* Post: Add assert_transact! macro to unwrap safely.
I was finding it very difficult to track unwrapping errors while
making changes, due to an underlying Mac OS X symbolication issue that
makes running tests with RUST_BACKTRACE=1 so slow that they all time
out.
* Post: Don't expect or recognize :db.{install,alter}/attribute.
I had this all working... except we will never see a repeated
`[:db.part/db :db.alter/attribute :attribute]` assertion in the store!
That means my approach would let you alter an attribute at most one
time. It's not worth hacking around this; it's better to just stop
expecting (and recognizing) the marker flags. (We have all the data
to distinguish the various cases that we need without the marker
flags.)
This brings Mentat in line with the thrust of newer Datomic versions,
but isn't compatible with Datomic, because (if I understand correctly)
Datomic automatically adds :db.{install,alter}/attribute assertions to
transactions.
I haven't purged the corresponding :db/ident and schema fragments just
yet:
- we might want them back
- we might want them in order to upgrade v1 and v2 databases to the
new on-disk layout we're fleshing out (v3?).
* Post: Don't make :db/unique :db.unique/* imply :db/index true.
This patch avoids a potential bug with the "schema" materialized view.
If :db/unique :db.unique/value implies :db/index true, then what
happens when you _retract_ :db.unique/value? I think Datomic defines
this in some way, but I really want the "schema" materialized view to
be a slice of "datoms" and not have these sort of ambiguities and
persistent effects. Therefore, to ensure that we don't retract a
schema characteristic and accidentally change more than we intended
to, this patch stops having any schema characteristic imply any other
schema characteristic(s). To achieve that, I added an
Option<Unique::{Value,Identity}> type to Attribute; this helps with
this patch, and also looks ahead to when we allow to retract
:db/unique attributes.
* Post: Allow to retract :db/ident.
* Post: Include more details about invalid schema changes.
The tests use strings, so they hide the chained errors which do in
fact provide more detail.
* Review comment: Fix outdated comment.
* Review comment: s/_SET/_SQL_LIST/.
* Review comment: Use a sub-select for checking cardinality.
This might be faster in practice.
* Review comment: Put `attribute::Unique` into its own namespace.
2017-03-20 20:18:59 +00:00
[ :db / add 100 :db / valueType :db . type / long ]
2019-07-17 15:46:47 +00:00
[ :db / add 100 :db / cardinality :db . cardinality / one ] ] "
) ;
Schema alteration. Fixes #294 and #295. (#370) r=rnewman
* Pre: Don't retract :db/ident in test.
Datomic (and eventually Mentat) don't allow to retract :db/ident in
this way, so this runs afoul of future work to support mutating
metadata.
* Pre: s/VALUETYPE/VALUE_TYPE/.
This is consistent with the capitalization (which is "valueType") and
the other identifier.
* Pre: Remove some single quotes from error output.
* Part 1: Make materialized views be uniform [e a v value_type_tag].
This looks ahead to a time when we could support arbitrary
user-defined materialized views. For now, the "idents" materialized
view is those datoms of the form [e :db/ident :namespaced/keyword] and
the "schema" materialized view is those datoms of the form [e a v]
where a is in a particular set of attributes that will become clear in
the following commits.
This change is not backwards compatible, so I'm removing the open
current (really, v2) test. It'll be re-instated when we get to
https://github.com/mozilla/mentat/issues/194.
* Pre: Map TypedValue::Ref to TypedValue::Keyword in debug output.
* Part 3: Separate `schema_to_mutate` from the `schema` used to interpret.
This is just to keep track of the expected changes during
bootstrapping. I want bootstrap metadata mutations to flow through
the same code path as metadata mutations during regular transactions;
by differentiating the schema used for interpretation from the schema
that will be updated I expect to be able to apply bootstrap metadata
mutations to an empty schema and have things like materialized views
created (using the regular code paths).
This commit has been re-ordered for conceptual clarity, but it won't
compile because it references the metadata module. It's possible to
make it compile -- the functionality is there in the schema module --
but it's not worth the rebasing effort until after review (and
possibly not even then, since we'll squash down to a single commit to
land).
* Part 2: Maintain entids separately from idents.
In order to support historical idents, we need to distinguish the
"current" map from entid -> ident from the "complete historical" map
ident -> entid. This is what Datomic does; in Datomic, an ident is
never retracted (although it can be replaced). This approach is an
important part of allowing multiple consumers to share a schema
fragment as it migrates forward.
This fixes a limitation of the Clojure implementation, which did not
handle historical idents across knowledge base close and re-open.
The "entids" materialized view is naturally a slice of the "datoms"
table. The "idents" materialized view is a slice of the
"transactions" table. I hope that representing in this way, and
casting the problem in this light, might generalize to future
materialized views.
* Pre: Add DiffSet.
* Part 4: Collect mutations to a `Schema`.
I haven't taken your review comment about consuming AttributeBuilder
during each fluent function. If you read my response and still want
this, I'm happy to do it in review.
* Part 5: Handle :db/ident and :db.{install,alter}/attribute.
This "loops" the committed datoms out of the SQL store and back
through the metadata (schema, but in future also partition map)
processor. The metadata processor updates the schema and produces a
report of what changed; that report is then used to update the SQL
store. That update includes:
- the materialized views ("entids", "idents", and "schema");
- if needed, a subset of the datoms themselves (as flags change).
I've left a TODO for handling attribute retraction in the cases that
it makes sense. I expect that to be straight-forward.
* Review comment: Rename DiffSet to AddRetractAlterSet.
Also adds a little more commentary and a simple test.
* Review comment: Use ToIdent trait.
* Review comment: partially revert "Part 2: Maintain entids separately from idents."
This reverts commit 23a91df9c35e14398f2ddbd1ba25315821e67401.
Following our discussion, this removes the "entids" materialized
view. The next commit will remove historical idents from the "idents"
materialized view.
* Post: Use custom Either rather than std::result::Result.
This is not necessary, but it was suggested that we might be paying an
overhead creating Err instances while using error_chain. That seems
not to be the case, but this change shows that we don't actually use
any of the Result helper methods, so there's no reason to overload
Result. This change might avoid some future confusion, so I'm going
to land it anyway.
Signed-off-by: Nick Alexander <nalexander@mozilla.com>
* Review comment: Don't preserve historical idents.
* Review comment: More prepared statements when updating materialized views.
* Post: Test altering :db/cardinality and :db/unique.
These tests fail due to a Datomic limitation, namely that the marker
flag :db.alter/attribute can only be asserted once for an attribute!
That is, [:db.part/db :db.alter/attribute :attribute] will only be
transacted at most once. Since older versions of Datomic required the
:db.alter/attribute flag, I can only imagine they either never wrote
:db.alter/attribute to the store, or they handled it specially. I'll
need to remove the marker flag system from Mentat in order to address
this fundamental limitation.
* Post: Remove some more single quotes from error output.
* Post: Add assert_transact! macro to unwrap safely.
I was finding it very difficult to track unwrapping errors while
making changes, due to an underlying Mac OS X symbolication issue that
makes running tests with RUST_BACKTRACE=1 so slow that they all time
out.
* Post: Don't expect or recognize :db.{install,alter}/attribute.
I had this all working... except we will never see a repeated
`[:db.part/db :db.alter/attribute :attribute]` assertion in the store!
That means my approach would let you alter an attribute at most one
time. It's not worth hacking around this; it's better to just stop
expecting (and recognizing) the marker flags. (We have all the data
to distinguish the various cases that we need without the marker
flags.)
This brings Mentat in line with the thrust of newer Datomic versions,
but isn't compatible with Datomic, because (if I understand correctly)
Datomic automatically adds :db.{install,alter}/attribute assertions to
transactions.
I haven't purged the corresponding :db/ident and schema fragments just
yet:
- we might want them back
- we might want them in order to upgrade v1 and v2 databases to the
new on-disk layout we're fleshing out (v3?).
* Post: Don't make :db/unique :db.unique/* imply :db/index true.
This patch avoids a potential bug with the "schema" materialized view.
If :db/unique :db.unique/value implies :db/index true, then what
happens when you _retract_ :db.unique/value? I think Datomic defines
this in some way, but I really want the "schema" materialized view to
be a slice of "datoms" and not have these sort of ambiguities and
persistent effects. Therefore, to ensure that we don't retract a
schema characteristic and accidentally change more than we intended
to, this patch stops having any schema characteristic imply any other
schema characteristic(s). To achieve that, I added an
Option<Unique::{Value,Identity}> type to Attribute; this helps with
this patch, and also looks ahead to when we allow to retract
:db/unique attributes.
* Post: Allow to retract :db/ident.
* Post: Include more details about invalid schema changes.
The tests use strings, so they hide the chained errors which do in
fact provide more detail.
* Review comment: Fix outdated comment.
* Review comment: s/_SET/_SQL_LIST/.
* Review comment: Use a sub-select for checking cardinality.
This might be faster in practice.
* Review comment: Put `attribute::Unique` into its own namespace.
2017-03-20 20:18:59 +00:00
assert_transact! ( conn , " [[:db/add 200 :test/ident 1]] " ) ;
// We can always go from :db.cardinality/one to :db.cardinality/many.
assert_transact! ( conn , " [[:db/add 100 :db/cardinality :db.cardinality/many]] " ) ;
assert_transact! ( conn , " [[:db/add 200 :test/ident 2]] " ) ;
2019-07-17 15:46:47 +00:00
assert_matches! (
conn . datoms ( ) ,
" [[100 :db/ident :test/ident]
Schema alteration. Fixes #294 and #295. (#370) r=rnewman
* Pre: Don't retract :db/ident in test.
Datomic (and eventually Mentat) don't allow to retract :db/ident in
this way, so this runs afoul of future work to support mutating
metadata.
* Pre: s/VALUETYPE/VALUE_TYPE/.
This is consistent with the capitalization (which is "valueType") and
the other identifier.
* Pre: Remove some single quotes from error output.
* Part 1: Make materialized views be uniform [e a v value_type_tag].
This looks ahead to a time when we could support arbitrary
user-defined materialized views. For now, the "idents" materialized
view is those datoms of the form [e :db/ident :namespaced/keyword] and
the "schema" materialized view is those datoms of the form [e a v]
where a is in a particular set of attributes that will become clear in
the following commits.
This change is not backwards compatible, so I'm removing the open
current (really, v2) test. It'll be re-instated when we get to
https://github.com/mozilla/mentat/issues/194.
* Pre: Map TypedValue::Ref to TypedValue::Keyword in debug output.
* Part 3: Separate `schema_to_mutate` from the `schema` used to interpret.
This is just to keep track of the expected changes during
bootstrapping. I want bootstrap metadata mutations to flow through
the same code path as metadata mutations during regular transactions;
by differentiating the schema used for interpretation from the schema
that will be updated I expect to be able to apply bootstrap metadata
mutations to an empty schema and have things like materialized views
created (using the regular code paths).
This commit has been re-ordered for conceptual clarity, but it won't
compile because it references the metadata module. It's possible to
make it compile -- the functionality is there in the schema module --
but it's not worth the rebasing effort until after review (and
possibly not even then, since we'll squash down to a single commit to
land).
* Part 2: Maintain entids separately from idents.
In order to support historical idents, we need to distinguish the
"current" map from entid -> ident from the "complete historical" map
ident -> entid. This is what Datomic does; in Datomic, an ident is
never retracted (although it can be replaced). This approach is an
important part of allowing multiple consumers to share a schema
fragment as it migrates forward.
This fixes a limitation of the Clojure implementation, which did not
handle historical idents across knowledge base close and re-open.
The "entids" materialized view is naturally a slice of the "datoms"
table. The "idents" materialized view is a slice of the
"transactions" table. I hope that representing in this way, and
casting the problem in this light, might generalize to future
materialized views.
* Pre: Add DiffSet.
* Part 4: Collect mutations to a `Schema`.
I haven't taken your review comment about consuming AttributeBuilder
during each fluent function. If you read my response and still want
this, I'm happy to do it in review.
* Part 5: Handle :db/ident and :db.{install,alter}/attribute.
This "loops" the committed datoms out of the SQL store and back
through the metadata (schema, but in future also partition map)
processor. The metadata processor updates the schema and produces a
report of what changed; that report is then used to update the SQL
store. That update includes:
- the materialized views ("entids", "idents", and "schema");
- if needed, a subset of the datoms themselves (as flags change).
I've left a TODO for handling attribute retraction in the cases that
it makes sense. I expect that to be straight-forward.
* Review comment: Rename DiffSet to AddRetractAlterSet.
Also adds a little more commentary and a simple test.
* Review comment: Use ToIdent trait.
* Review comment: partially revert "Part 2: Maintain entids separately from idents."
This reverts commit 23a91df9c35e14398f2ddbd1ba25315821e67401.
Following our discussion, this removes the "entids" materialized
view. The next commit will remove historical idents from the "idents"
materialized view.
* Post: Use custom Either rather than std::result::Result.
This is not necessary, but it was suggested that we might be paying an
overhead creating Err instances while using error_chain. That seems
not to be the case, but this change shows that we don't actually use
any of the Result helper methods, so there's no reason to overload
Result. This change might avoid some future confusion, so I'm going
to land it anyway.
Signed-off-by: Nick Alexander <nalexander@mozilla.com>
* Review comment: Don't preserve historical idents.
* Review comment: More prepared statements when updating materialized views.
* Post: Test altering :db/cardinality and :db/unique.
These tests fail due to a Datomic limitation, namely that the marker
flag :db.alter/attribute can only be asserted once for an attribute!
That is, [:db.part/db :db.alter/attribute :attribute] will only be
transacted at most once. Since older versions of Datomic required the
:db.alter/attribute flag, I can only imagine they either never wrote
:db.alter/attribute to the store, or they handled it specially. I'll
need to remove the marker flag system from Mentat in order to address
this fundamental limitation.
* Post: Remove some more single quotes from error output.
* Post: Add assert_transact! macro to unwrap safely.
I was finding it very difficult to track unwrapping errors while
making changes, due to an underlying Mac OS X symbolication issue that
makes running tests with RUST_BACKTRACE=1 so slow that they all time
out.
* Post: Don't expect or recognize :db.{install,alter}/attribute.
I had this all working... except we will never see a repeated
`[:db.part/db :db.alter/attribute :attribute]` assertion in the store!
That means my approach would let you alter an attribute at most one
time. It's not worth hacking around this; it's better to just stop
expecting (and recognizing) the marker flags. (We have all the data
to distinguish the various cases that we need without the marker
flags.)
This brings Mentat in line with the thrust of newer Datomic versions,
but isn't compatible with Datomic, because (if I understand correctly)
Datomic automatically adds :db.{install,alter}/attribute assertions to
transactions.
I haven't purged the corresponding :db/ident and schema fragments just
yet:
- we might want them back
- we might want them in order to upgrade v1 and v2 databases to the
new on-disk layout we're fleshing out (v3?).
* Post: Don't make :db/unique :db.unique/* imply :db/index true.
This patch avoids a potential bug with the "schema" materialized view.
If :db/unique :db.unique/value implies :db/index true, then what
happens when you _retract_ :db.unique/value? I think Datomic defines
this in some way, but I really want the "schema" materialized view to
be a slice of "datoms" and not have these sort of ambiguities and
persistent effects. Therefore, to ensure that we don't retract a
schema characteristic and accidentally change more than we intended
to, this patch stops having any schema characteristic imply any other
schema characteristic(s). To achieve that, I added an
Option<Unique::{Value,Identity}> type to Attribute; this helps with
this patch, and also looks ahead to when we allow to retract
:db/unique attributes.
* Post: Allow to retract :db/ident.
* Post: Include more details about invalid schema changes.
The tests use strings, so they hide the chained errors which do in
fact provide more detail.
* Review comment: Fix outdated comment.
* Review comment: s/_SET/_SQL_LIST/.
* Review comment: Use a sub-select for checking cardinality.
This might be faster in practice.
* Review comment: Put `attribute::Unique` into its own namespace.
2017-03-20 20:18:59 +00:00
[ 100 :db / valueType :db . type / long ]
[ 100 :db / cardinality :db . cardinality / many ]
[ 200 :test / ident 1 ]
2019-07-17 15:46:47 +00:00
[ 200 :test / ident 2 ] ] "
) ;
Schema alteration. Fixes #294 and #295. (#370) r=rnewman
* Pre: Don't retract :db/ident in test.
Datomic (and eventually Mentat) don't allow to retract :db/ident in
this way, so this runs afoul of future work to support mutating
metadata.
* Pre: s/VALUETYPE/VALUE_TYPE/.
This is consistent with the capitalization (which is "valueType") and
the other identifier.
* Pre: Remove some single quotes from error output.
* Part 1: Make materialized views be uniform [e a v value_type_tag].
This looks ahead to a time when we could support arbitrary
user-defined materialized views. For now, the "idents" materialized
view is those datoms of the form [e :db/ident :namespaced/keyword] and
the "schema" materialized view is those datoms of the form [e a v]
where a is in a particular set of attributes that will become clear in
the following commits.
This change is not backwards compatible, so I'm removing the open
current (really, v2) test. It'll be re-instated when we get to
https://github.com/mozilla/mentat/issues/194.
* Pre: Map TypedValue::Ref to TypedValue::Keyword in debug output.
* Part 3: Separate `schema_to_mutate` from the `schema` used to interpret.
This is just to keep track of the expected changes during
bootstrapping. I want bootstrap metadata mutations to flow through
the same code path as metadata mutations during regular transactions;
by differentiating the schema used for interpretation from the schema
that will be updated I expect to be able to apply bootstrap metadata
mutations to an empty schema and have things like materialized views
created (using the regular code paths).
This commit has been re-ordered for conceptual clarity, but it won't
compile because it references the metadata module. It's possible to
make it compile -- the functionality is there in the schema module --
but it's not worth the rebasing effort until after review (and
possibly not even then, since we'll squash down to a single commit to
land).
* Part 2: Maintain entids separately from idents.
In order to support historical idents, we need to distinguish the
"current" map from entid -> ident from the "complete historical" map
ident -> entid. This is what Datomic does; in Datomic, an ident is
never retracted (although it can be replaced). This approach is an
important part of allowing multiple consumers to share a schema
fragment as it migrates forward.
This fixes a limitation of the Clojure implementation, which did not
handle historical idents across knowledge base close and re-open.
The "entids" materialized view is naturally a slice of the "datoms"
table. The "idents" materialized view is a slice of the
"transactions" table. I hope that representing in this way, and
casting the problem in this light, might generalize to future
materialized views.
* Pre: Add DiffSet.
* Part 4: Collect mutations to a `Schema`.
I haven't taken your review comment about consuming AttributeBuilder
during each fluent function. If you read my response and still want
this, I'm happy to do it in review.
* Part 5: Handle :db/ident and :db.{install,alter}/attribute.
This "loops" the committed datoms out of the SQL store and back
through the metadata (schema, but in future also partition map)
processor. The metadata processor updates the schema and produces a
report of what changed; that report is then used to update the SQL
store. That update includes:
- the materialized views ("entids", "idents", and "schema");
- if needed, a subset of the datoms themselves (as flags change).
I've left a TODO for handling attribute retraction in the cases that
it makes sense. I expect that to be straight-forward.
* Review comment: Rename DiffSet to AddRetractAlterSet.
Also adds a little more commentary and a simple test.
* Review comment: Use ToIdent trait.
* Review comment: partially revert "Part 2: Maintain entids separately from idents."
This reverts commit 23a91df9c35e14398f2ddbd1ba25315821e67401.
Following our discussion, this removes the "entids" materialized
view. The next commit will remove historical idents from the "idents"
materialized view.
* Post: Use custom Either rather than std::result::Result.
This is not necessary, but it was suggested that we might be paying an
overhead creating Err instances while using error_chain. That seems
not to be the case, but this change shows that we don't actually use
any of the Result helper methods, so there's no reason to overload
Result. This change might avoid some future confusion, so I'm going
to land it anyway.
Signed-off-by: Nick Alexander <nalexander@mozilla.com>
* Review comment: Don't preserve historical idents.
* Review comment: More prepared statements when updating materialized views.
* Post: Test altering :db/cardinality and :db/unique.
These tests fail due to a Datomic limitation, namely that the marker
flag :db.alter/attribute can only be asserted once for an attribute!
That is, [:db.part/db :db.alter/attribute :attribute] will only be
transacted at most once. Since older versions of Datomic required the
:db.alter/attribute flag, I can only imagine they either never wrote
:db.alter/attribute to the store, or they handled it specially. I'll
need to remove the marker flag system from Mentat in order to address
this fundamental limitation.
* Post: Remove some more single quotes from error output.
* Post: Add assert_transact! macro to unwrap safely.
I was finding it very difficult to track unwrapping errors while
making changes, due to an underlying Mac OS X symbolication issue that
makes running tests with RUST_BACKTRACE=1 so slow that they all time
out.
* Post: Don't expect or recognize :db.{install,alter}/attribute.
I had this all working... except we will never see a repeated
`[:db.part/db :db.alter/attribute :attribute]` assertion in the store!
That means my approach would let you alter an attribute at most one
time. It's not worth hacking around this; it's better to just stop
expecting (and recognizing) the marker flags. (We have all the data
to distinguish the various cases that we need without the marker
flags.)
This brings Mentat in line with the thrust of newer Datomic versions,
but isn't compatible with Datomic, because (if I understand correctly)
Datomic automatically adds :db.{install,alter}/attribute assertions to
transactions.
I haven't purged the corresponding :db/ident and schema fragments just
yet:
- we might want them back
- we might want them in order to upgrade v1 and v2 databases to the
new on-disk layout we're fleshing out (v3?).
* Post: Don't make :db/unique :db.unique/* imply :db/index true.
This patch avoids a potential bug with the "schema" materialized view.
If :db/unique :db.unique/value implies :db/index true, then what
happens when you _retract_ :db.unique/value? I think Datomic defines
this in some way, but I really want the "schema" materialized view to
be a slice of "datoms" and not have these sort of ambiguities and
persistent effects. Therefore, to ensure that we don't retract a
schema characteristic and accidentally change more than we intended
to, this patch stops having any schema characteristic imply any other
schema characteristic(s). To achieve that, I added an
Option<Unique::{Value,Identity}> type to Attribute; this helps with
this patch, and also looks ahead to when we allow to retract
:db/unique attributes.
* Post: Allow to retract :db/ident.
* Post: Include more details about invalid schema changes.
The tests use strings, so they hide the chained errors which do in
fact provide more detail.
* Review comment: Fix outdated comment.
* Review comment: s/_SET/_SQL_LIST/.
* Review comment: Use a sub-select for checking cardinality.
This might be faster in practice.
* Review comment: Put `attribute::Unique` into its own namespace.
2017-03-20 20:18:59 +00:00
// We can't always go from :db.cardinality/many to :db.cardinality/one.
assert_transact! ( conn , " [[:db/add 100 :db/cardinality :db.cardinality/one]] " ,
// TODO: give more helpful error details.
2018-04-03 20:18:56 +00:00
Err ( " schema alteration failed: Cannot alter schema attribute 100 to be :db.cardinality/one " ) ) ;
Schema alteration. Fixes #294 and #295. (#370) r=rnewman
* Pre: Don't retract :db/ident in test.
Datomic (and eventually Mentat) don't allow to retract :db/ident in
this way, so this runs afoul of future work to support mutating
metadata.
* Pre: s/VALUETYPE/VALUE_TYPE/.
This is consistent with the capitalization (which is "valueType") and
the other identifier.
* Pre: Remove some single quotes from error output.
* Part 1: Make materialized views be uniform [e a v value_type_tag].
This looks ahead to a time when we could support arbitrary
user-defined materialized views. For now, the "idents" materialized
view is those datoms of the form [e :db/ident :namespaced/keyword] and
the "schema" materialized view is those datoms of the form [e a v]
where a is in a particular set of attributes that will become clear in
the following commits.
This change is not backwards compatible, so I'm removing the open
current (really, v2) test. It'll be re-instated when we get to
https://github.com/mozilla/mentat/issues/194.
* Pre: Map TypedValue::Ref to TypedValue::Keyword in debug output.
* Part 3: Separate `schema_to_mutate` from the `schema` used to interpret.
This is just to keep track of the expected changes during
bootstrapping. I want bootstrap metadata mutations to flow through
the same code path as metadata mutations during regular transactions;
by differentiating the schema used for interpretation from the schema
that will be updated I expect to be able to apply bootstrap metadata
mutations to an empty schema and have things like materialized views
created (using the regular code paths).
This commit has been re-ordered for conceptual clarity, but it won't
compile because it references the metadata module. It's possible to
make it compile -- the functionality is there in the schema module --
but it's not worth the rebasing effort until after review (and
possibly not even then, since we'll squash down to a single commit to
land).
* Part 2: Maintain entids separately from idents.
In order to support historical idents, we need to distinguish the
"current" map from entid -> ident from the "complete historical" map
ident -> entid. This is what Datomic does; in Datomic, an ident is
never retracted (although it can be replaced). This approach is an
important part of allowing multiple consumers to share a schema
fragment as it migrates forward.
This fixes a limitation of the Clojure implementation, which did not
handle historical idents across knowledge base close and re-open.
The "entids" materialized view is naturally a slice of the "datoms"
table. The "idents" materialized view is a slice of the
"transactions" table. I hope that representing in this way, and
casting the problem in this light, might generalize to future
materialized views.
* Pre: Add DiffSet.
* Part 4: Collect mutations to a `Schema`.
I haven't taken your review comment about consuming AttributeBuilder
during each fluent function. If you read my response and still want
this, I'm happy to do it in review.
* Part 5: Handle :db/ident and :db.{install,alter}/attribute.
This "loops" the committed datoms out of the SQL store and back
through the metadata (schema, but in future also partition map)
processor. The metadata processor updates the schema and produces a
report of what changed; that report is then used to update the SQL
store. That update includes:
- the materialized views ("entids", "idents", and "schema");
- if needed, a subset of the datoms themselves (as flags change).
I've left a TODO for handling attribute retraction in the cases that
it makes sense. I expect that to be straight-forward.
* Review comment: Rename DiffSet to AddRetractAlterSet.
Also adds a little more commentary and a simple test.
* Review comment: Use ToIdent trait.
* Review comment: partially revert "Part 2: Maintain entids separately from idents."
This reverts commit 23a91df9c35e14398f2ddbd1ba25315821e67401.
Following our discussion, this removes the "entids" materialized
view. The next commit will remove historical idents from the "idents"
materialized view.
* Post: Use custom Either rather than std::result::Result.
This is not necessary, but it was suggested that we might be paying an
overhead creating Err instances while using error_chain. That seems
not to be the case, but this change shows that we don't actually use
any of the Result helper methods, so there's no reason to overload
Result. This change might avoid some future confusion, so I'm going
to land it anyway.
Signed-off-by: Nick Alexander <nalexander@mozilla.com>
* Review comment: Don't preserve historical idents.
* Review comment: More prepared statements when updating materialized views.
* Post: Test altering :db/cardinality and :db/unique.
These tests fail due to a Datomic limitation, namely that the marker
flag :db.alter/attribute can only be asserted once for an attribute!
That is, [:db.part/db :db.alter/attribute :attribute] will only be
transacted at most once. Since older versions of Datomic required the
:db.alter/attribute flag, I can only imagine they either never wrote
:db.alter/attribute to the store, or they handled it specially. I'll
need to remove the marker flag system from Mentat in order to address
this fundamental limitation.
* Post: Remove some more single quotes from error output.
* Post: Add assert_transact! macro to unwrap safely.
I was finding it very difficult to track unwrapping errors while
making changes, due to an underlying Mac OS X symbolication issue that
makes running tests with RUST_BACKTRACE=1 so slow that they all time
out.
* Post: Don't expect or recognize :db.{install,alter}/attribute.
I had this all working... except we will never see a repeated
`[:db.part/db :db.alter/attribute :attribute]` assertion in the store!
That means my approach would let you alter an attribute at most one
time. It's not worth hacking around this; it's better to just stop
expecting (and recognizing) the marker flags. (We have all the data
to distinguish the various cases that we need without the marker
flags.)
This brings Mentat in line with the thrust of newer Datomic versions,
but isn't compatible with Datomic, because (if I understand correctly)
Datomic automatically adds :db.{install,alter}/attribute assertions to
transactions.
I haven't purged the corresponding :db/ident and schema fragments just
yet:
- we might want them back
- we might want them in order to upgrade v1 and v2 databases to the
new on-disk layout we're fleshing out (v3?).
* Post: Don't make :db/unique :db.unique/* imply :db/index true.
This patch avoids a potential bug with the "schema" materialized view.
If :db/unique :db.unique/value implies :db/index true, then what
happens when you _retract_ :db.unique/value? I think Datomic defines
this in some way, but I really want the "schema" materialized view to
be a slice of "datoms" and not have these sort of ambiguities and
persistent effects. Therefore, to ensure that we don't retract a
schema characteristic and accidentally change more than we intended
to, this patch stops having any schema characteristic imply any other
schema characteristic(s). To achieve that, I added an
Option<Unique::{Value,Identity}> type to Attribute; this helps with
this patch, and also looks ahead to when we allow to retract
:db/unique attributes.
* Post: Allow to retract :db/ident.
* Post: Include more details about invalid schema changes.
The tests use strings, so they hide the chained errors which do in
fact provide more detail.
* Review comment: Fix outdated comment.
* Review comment: s/_SET/_SQL_LIST/.
* Review comment: Use a sub-select for checking cardinality.
This might be faster in practice.
* Review comment: Put `attribute::Unique` into its own namespace.
2017-03-20 20:18:59 +00:00
}
#[ test ]
fn test_db_alter_unique_value ( ) {
let mut conn = TestConn ::default ( ) ;
// Start by installing a :db.cardinality/one attribute.
2019-07-17 15:46:47 +00:00
assert_transact! (
conn ,
" [[:db/add 100 :db/ident :test/ident]
Schema alteration. Fixes #294 and #295. (#370) r=rnewman
* Pre: Don't retract :db/ident in test.
Datomic (and eventually Mentat) don't allow to retract :db/ident in
this way, so this runs afoul of future work to support mutating
metadata.
* Pre: s/VALUETYPE/VALUE_TYPE/.
This is consistent with the capitalization (which is "valueType") and
the other identifier.
* Pre: Remove some single quotes from error output.
* Part 1: Make materialized views be uniform [e a v value_type_tag].
This looks ahead to a time when we could support arbitrary
user-defined materialized views. For now, the "idents" materialized
view is those datoms of the form [e :db/ident :namespaced/keyword] and
the "schema" materialized view is those datoms of the form [e a v]
where a is in a particular set of attributes that will become clear in
the following commits.
This change is not backwards compatible, so I'm removing the open
current (really, v2) test. It'll be re-instated when we get to
https://github.com/mozilla/mentat/issues/194.
* Pre: Map TypedValue::Ref to TypedValue::Keyword in debug output.
* Part 3: Separate `schema_to_mutate` from the `schema` used to interpret.
This is just to keep track of the expected changes during
bootstrapping. I want bootstrap metadata mutations to flow through
the same code path as metadata mutations during regular transactions;
by differentiating the schema used for interpretation from the schema
that will be updated I expect to be able to apply bootstrap metadata
mutations to an empty schema and have things like materialized views
created (using the regular code paths).
This commit has been re-ordered for conceptual clarity, but it won't
compile because it references the metadata module. It's possible to
make it compile -- the functionality is there in the schema module --
but it's not worth the rebasing effort until after review (and
possibly not even then, since we'll squash down to a single commit to
land).
* Part 2: Maintain entids separately from idents.
In order to support historical idents, we need to distinguish the
"current" map from entid -> ident from the "complete historical" map
ident -> entid. This is what Datomic does; in Datomic, an ident is
never retracted (although it can be replaced). This approach is an
important part of allowing multiple consumers to share a schema
fragment as it migrates forward.
This fixes a limitation of the Clojure implementation, which did not
handle historical idents across knowledge base close and re-open.
The "entids" materialized view is naturally a slice of the "datoms"
table. The "idents" materialized view is a slice of the
"transactions" table. I hope that representing in this way, and
casting the problem in this light, might generalize to future
materialized views.
* Pre: Add DiffSet.
* Part 4: Collect mutations to a `Schema`.
I haven't taken your review comment about consuming AttributeBuilder
during each fluent function. If you read my response and still want
this, I'm happy to do it in review.
* Part 5: Handle :db/ident and :db.{install,alter}/attribute.
This "loops" the committed datoms out of the SQL store and back
through the metadata (schema, but in future also partition map)
processor. The metadata processor updates the schema and produces a
report of what changed; that report is then used to update the SQL
store. That update includes:
- the materialized views ("entids", "idents", and "schema");
- if needed, a subset of the datoms themselves (as flags change).
I've left a TODO for handling attribute retraction in the cases that
it makes sense. I expect that to be straight-forward.
* Review comment: Rename DiffSet to AddRetractAlterSet.
Also adds a little more commentary and a simple test.
* Review comment: Use ToIdent trait.
* Review comment: partially revert "Part 2: Maintain entids separately from idents."
This reverts commit 23a91df9c35e14398f2ddbd1ba25315821e67401.
Following our discussion, this removes the "entids" materialized
view. The next commit will remove historical idents from the "idents"
materialized view.
* Post: Use custom Either rather than std::result::Result.
This is not necessary, but it was suggested that we might be paying an
overhead creating Err instances while using error_chain. That seems
not to be the case, but this change shows that we don't actually use
any of the Result helper methods, so there's no reason to overload
Result. This change might avoid some future confusion, so I'm going
to land it anyway.
Signed-off-by: Nick Alexander <nalexander@mozilla.com>
* Review comment: Don't preserve historical idents.
* Review comment: More prepared statements when updating materialized views.
* Post: Test altering :db/cardinality and :db/unique.
These tests fail due to a Datomic limitation, namely that the marker
flag :db.alter/attribute can only be asserted once for an attribute!
That is, [:db.part/db :db.alter/attribute :attribute] will only be
transacted at most once. Since older versions of Datomic required the
:db.alter/attribute flag, I can only imagine they either never wrote
:db.alter/attribute to the store, or they handled it specially. I'll
need to remove the marker flag system from Mentat in order to address
this fundamental limitation.
* Post: Remove some more single quotes from error output.
* Post: Add assert_transact! macro to unwrap safely.
I was finding it very difficult to track unwrapping errors while
making changes, due to an underlying Mac OS X symbolication issue that
makes running tests with RUST_BACKTRACE=1 so slow that they all time
out.
* Post: Don't expect or recognize :db.{install,alter}/attribute.
I had this all working... except we will never see a repeated
`[:db.part/db :db.alter/attribute :attribute]` assertion in the store!
That means my approach would let you alter an attribute at most one
time. It's not worth hacking around this; it's better to just stop
expecting (and recognizing) the marker flags. (We have all the data
to distinguish the various cases that we need without the marker
flags.)
This brings Mentat in line with the thrust of newer Datomic versions,
but isn't compatible with Datomic, because (if I understand correctly)
Datomic automatically adds :db.{install,alter}/attribute assertions to
transactions.
I haven't purged the corresponding :db/ident and schema fragments just
yet:
- we might want them back
- we might want them in order to upgrade v1 and v2 databases to the
new on-disk layout we're fleshing out (v3?).
* Post: Don't make :db/unique :db.unique/* imply :db/index true.
This patch avoids a potential bug with the "schema" materialized view.
If :db/unique :db.unique/value implies :db/index true, then what
happens when you _retract_ :db.unique/value? I think Datomic defines
this in some way, but I really want the "schema" materialized view to
be a slice of "datoms" and not have these sort of ambiguities and
persistent effects. Therefore, to ensure that we don't retract a
schema characteristic and accidentally change more than we intended
to, this patch stops having any schema characteristic imply any other
schema characteristic(s). To achieve that, I added an
Option<Unique::{Value,Identity}> type to Attribute; this helps with
this patch, and also looks ahead to when we allow to retract
:db/unique attributes.
* Post: Allow to retract :db/ident.
* Post: Include more details about invalid schema changes.
The tests use strings, so they hide the chained errors which do in
fact provide more detail.
* Review comment: Fix outdated comment.
* Review comment: s/_SET/_SQL_LIST/.
* Review comment: Use a sub-select for checking cardinality.
This might be faster in practice.
* Review comment: Put `attribute::Unique` into its own namespace.
2017-03-20 20:18:59 +00:00
[ :db / add 100 :db / valueType :db . type / long ]
2019-07-17 15:46:47 +00:00
[ :db / add 100 :db / cardinality :db . cardinality / one ] ] "
) ;
Schema alteration. Fixes #294 and #295. (#370) r=rnewman
* Pre: Don't retract :db/ident in test.
Datomic (and eventually Mentat) don't allow to retract :db/ident in
this way, so this runs afoul of future work to support mutating
metadata.
* Pre: s/VALUETYPE/VALUE_TYPE/.
This is consistent with the capitalization (which is "valueType") and
the other identifier.
* Pre: Remove some single quotes from error output.
* Part 1: Make materialized views be uniform [e a v value_type_tag].
This looks ahead to a time when we could support arbitrary
user-defined materialized views. For now, the "idents" materialized
view is those datoms of the form [e :db/ident :namespaced/keyword] and
the "schema" materialized view is those datoms of the form [e a v]
where a is in a particular set of attributes that will become clear in
the following commits.
This change is not backwards compatible, so I'm removing the open
current (really, v2) test. It'll be re-instated when we get to
https://github.com/mozilla/mentat/issues/194.
* Pre: Map TypedValue::Ref to TypedValue::Keyword in debug output.
* Part 3: Separate `schema_to_mutate` from the `schema` used to interpret.
This is just to keep track of the expected changes during
bootstrapping. I want bootstrap metadata mutations to flow through
the same code path as metadata mutations during regular transactions;
by differentiating the schema used for interpretation from the schema
that will be updated I expect to be able to apply bootstrap metadata
mutations to an empty schema and have things like materialized views
created (using the regular code paths).
This commit has been re-ordered for conceptual clarity, but it won't
compile because it references the metadata module. It's possible to
make it compile -- the functionality is there in the schema module --
but it's not worth the rebasing effort until after review (and
possibly not even then, since we'll squash down to a single commit to
land).
* Part 2: Maintain entids separately from idents.
In order to support historical idents, we need to distinguish the
"current" map from entid -> ident from the "complete historical" map
ident -> entid. This is what Datomic does; in Datomic, an ident is
never retracted (although it can be replaced). This approach is an
important part of allowing multiple consumers to share a schema
fragment as it migrates forward.
This fixes a limitation of the Clojure implementation, which did not
handle historical idents across knowledge base close and re-open.
The "entids" materialized view is naturally a slice of the "datoms"
table. The "idents" materialized view is a slice of the
"transactions" table. I hope that representing in this way, and
casting the problem in this light, might generalize to future
materialized views.
* Pre: Add DiffSet.
* Part 4: Collect mutations to a `Schema`.
I haven't taken your review comment about consuming AttributeBuilder
during each fluent function. If you read my response and still want
this, I'm happy to do it in review.
* Part 5: Handle :db/ident and :db.{install,alter}/attribute.
This "loops" the committed datoms out of the SQL store and back
through the metadata (schema, but in future also partition map)
processor. The metadata processor updates the schema and produces a
report of what changed; that report is then used to update the SQL
store. That update includes:
- the materialized views ("entids", "idents", and "schema");
- if needed, a subset of the datoms themselves (as flags change).
I've left a TODO for handling attribute retraction in the cases that
it makes sense. I expect that to be straight-forward.
* Review comment: Rename DiffSet to AddRetractAlterSet.
Also adds a little more commentary and a simple test.
* Review comment: Use ToIdent trait.
* Review comment: partially revert "Part 2: Maintain entids separately from idents."
This reverts commit 23a91df9c35e14398f2ddbd1ba25315821e67401.
Following our discussion, this removes the "entids" materialized
view. The next commit will remove historical idents from the "idents"
materialized view.
* Post: Use custom Either rather than std::result::Result.
This is not necessary, but it was suggested that we might be paying an
overhead creating Err instances while using error_chain. That seems
not to be the case, but this change shows that we don't actually use
any of the Result helper methods, so there's no reason to overload
Result. This change might avoid some future confusion, so I'm going
to land it anyway.
Signed-off-by: Nick Alexander <nalexander@mozilla.com>
* Review comment: Don't preserve historical idents.
* Review comment: More prepared statements when updating materialized views.
* Post: Test altering :db/cardinality and :db/unique.
These tests fail due to a Datomic limitation, namely that the marker
flag :db.alter/attribute can only be asserted once for an attribute!
That is, [:db.part/db :db.alter/attribute :attribute] will only be
transacted at most once. Since older versions of Datomic required the
:db.alter/attribute flag, I can only imagine they either never wrote
:db.alter/attribute to the store, or they handled it specially. I'll
need to remove the marker flag system from Mentat in order to address
this fundamental limitation.
* Post: Remove some more single quotes from error output.
* Post: Add assert_transact! macro to unwrap safely.
I was finding it very difficult to track unwrapping errors while
making changes, due to an underlying Mac OS X symbolication issue that
makes running tests with RUST_BACKTRACE=1 so slow that they all time
out.
* Post: Don't expect or recognize :db.{install,alter}/attribute.
I had this all working... except we will never see a repeated
`[:db.part/db :db.alter/attribute :attribute]` assertion in the store!
That means my approach would let you alter an attribute at most one
time. It's not worth hacking around this; it's better to just stop
expecting (and recognizing) the marker flags. (We have all the data
to distinguish the various cases that we need without the marker
flags.)
This brings Mentat in line with the thrust of newer Datomic versions,
but isn't compatible with Datomic, because (if I understand correctly)
Datomic automatically adds :db.{install,alter}/attribute assertions to
transactions.
I haven't purged the corresponding :db/ident and schema fragments just
yet:
- we might want them back
- we might want them in order to upgrade v1 and v2 databases to the
new on-disk layout we're fleshing out (v3?).
* Post: Don't make :db/unique :db.unique/* imply :db/index true.
This patch avoids a potential bug with the "schema" materialized view.
If :db/unique :db.unique/value implies :db/index true, then what
happens when you _retract_ :db.unique/value? I think Datomic defines
this in some way, but I really want the "schema" materialized view to
be a slice of "datoms" and not have these sort of ambiguities and
persistent effects. Therefore, to ensure that we don't retract a
schema characteristic and accidentally change more than we intended
to, this patch stops having any schema characteristic imply any other
schema characteristic(s). To achieve that, I added an
Option<Unique::{Value,Identity}> type to Attribute; this helps with
this patch, and also looks ahead to when we allow to retract
:db/unique attributes.
* Post: Allow to retract :db/ident.
* Post: Include more details about invalid schema changes.
The tests use strings, so they hide the chained errors which do in
fact provide more detail.
* Review comment: Fix outdated comment.
* Review comment: s/_SET/_SQL_LIST/.
* Review comment: Use a sub-select for checking cardinality.
This might be faster in practice.
* Review comment: Put `attribute::Unique` into its own namespace.
2017-03-20 20:18:59 +00:00
2019-07-17 15:46:47 +00:00
assert_transact! (
conn ,
" [[:db/add 200 :test/ident 1]
[ :db / add 201 :test / ident 1 ] ] "
) ;
Schema alteration. Fixes #294 and #295. (#370) r=rnewman
* Pre: Don't retract :db/ident in test.
Datomic (and eventually Mentat) don't allow to retract :db/ident in
this way, so this runs afoul of future work to support mutating
metadata.
* Pre: s/VALUETYPE/VALUE_TYPE/.
This is consistent with the capitalization (which is "valueType") and
the other identifier.
* Pre: Remove some single quotes from error output.
* Part 1: Make materialized views be uniform [e a v value_type_tag].
This looks ahead to a time when we could support arbitrary
user-defined materialized views. For now, the "idents" materialized
view is those datoms of the form [e :db/ident :namespaced/keyword] and
the "schema" materialized view is those datoms of the form [e a v]
where a is in a particular set of attributes that will become clear in
the following commits.
This change is not backwards compatible, so I'm removing the open
current (really, v2) test. It'll be re-instated when we get to
https://github.com/mozilla/mentat/issues/194.
* Pre: Map TypedValue::Ref to TypedValue::Keyword in debug output.
* Part 3: Separate `schema_to_mutate` from the `schema` used to interpret.
This is just to keep track of the expected changes during
bootstrapping. I want bootstrap metadata mutations to flow through
the same code path as metadata mutations during regular transactions;
by differentiating the schema used for interpretation from the schema
that will be updated I expect to be able to apply bootstrap metadata
mutations to an empty schema and have things like materialized views
created (using the regular code paths).
This commit has been re-ordered for conceptual clarity, but it won't
compile because it references the metadata module. It's possible to
make it compile -- the functionality is there in the schema module --
but it's not worth the rebasing effort until after review (and
possibly not even then, since we'll squash down to a single commit to
land).
* Part 2: Maintain entids separately from idents.
In order to support historical idents, we need to distinguish the
"current" map from entid -> ident from the "complete historical" map
ident -> entid. This is what Datomic does; in Datomic, an ident is
never retracted (although it can be replaced). This approach is an
important part of allowing multiple consumers to share a schema
fragment as it migrates forward.
This fixes a limitation of the Clojure implementation, which did not
handle historical idents across knowledge base close and re-open.
The "entids" materialized view is naturally a slice of the "datoms"
table. The "idents" materialized view is a slice of the
"transactions" table. I hope that representing in this way, and
casting the problem in this light, might generalize to future
materialized views.
* Pre: Add DiffSet.
* Part 4: Collect mutations to a `Schema`.
I haven't taken your review comment about consuming AttributeBuilder
during each fluent function. If you read my response and still want
this, I'm happy to do it in review.
* Part 5: Handle :db/ident and :db.{install,alter}/attribute.
This "loops" the committed datoms out of the SQL store and back
through the metadata (schema, but in future also partition map)
processor. The metadata processor updates the schema and produces a
report of what changed; that report is then used to update the SQL
store. That update includes:
- the materialized views ("entids", "idents", and "schema");
- if needed, a subset of the datoms themselves (as flags change).
I've left a TODO for handling attribute retraction in the cases that
it makes sense. I expect that to be straight-forward.
* Review comment: Rename DiffSet to AddRetractAlterSet.
Also adds a little more commentary and a simple test.
* Review comment: Use ToIdent trait.
* Review comment: partially revert "Part 2: Maintain entids separately from idents."
This reverts commit 23a91df9c35e14398f2ddbd1ba25315821e67401.
Following our discussion, this removes the "entids" materialized
view. The next commit will remove historical idents from the "idents"
materialized view.
* Post: Use custom Either rather than std::result::Result.
This is not necessary, but it was suggested that we might be paying an
overhead creating Err instances while using error_chain. That seems
not to be the case, but this change shows that we don't actually use
any of the Result helper methods, so there's no reason to overload
Result. This change might avoid some future confusion, so I'm going
to land it anyway.
Signed-off-by: Nick Alexander <nalexander@mozilla.com>
* Review comment: Don't preserve historical idents.
* Review comment: More prepared statements when updating materialized views.
* Post: Test altering :db/cardinality and :db/unique.
These tests fail due to a Datomic limitation, namely that the marker
flag :db.alter/attribute can only be asserted once for an attribute!
That is, [:db.part/db :db.alter/attribute :attribute] will only be
transacted at most once. Since older versions of Datomic required the
:db.alter/attribute flag, I can only imagine they either never wrote
:db.alter/attribute to the store, or they handled it specially. I'll
need to remove the marker flag system from Mentat in order to address
this fundamental limitation.
* Post: Remove some more single quotes from error output.
* Post: Add assert_transact! macro to unwrap safely.
I was finding it very difficult to track unwrapping errors while
making changes, due to an underlying Mac OS X symbolication issue that
makes running tests with RUST_BACKTRACE=1 so slow that they all time
out.
* Post: Don't expect or recognize :db.{install,alter}/attribute.
I had this all working... except we will never see a repeated
`[:db.part/db :db.alter/attribute :attribute]` assertion in the store!
That means my approach would let you alter an attribute at most one
time. It's not worth hacking around this; it's better to just stop
expecting (and recognizing) the marker flags. (We have all the data
to distinguish the various cases that we need without the marker
flags.)
This brings Mentat in line with the thrust of newer Datomic versions,
but isn't compatible with Datomic, because (if I understand correctly)
Datomic automatically adds :db.{install,alter}/attribute assertions to
transactions.
I haven't purged the corresponding :db/ident and schema fragments just
yet:
- we might want them back
- we might want them in order to upgrade v1 and v2 databases to the
new on-disk layout we're fleshing out (v3?).
* Post: Don't make :db/unique :db.unique/* imply :db/index true.
This patch avoids a potential bug with the "schema" materialized view.
If :db/unique :db.unique/value implies :db/index true, then what
happens when you _retract_ :db.unique/value? I think Datomic defines
this in some way, but I really want the "schema" materialized view to
be a slice of "datoms" and not have these sort of ambiguities and
persistent effects. Therefore, to ensure that we don't retract a
schema characteristic and accidentally change more than we intended
to, this patch stops having any schema characteristic imply any other
schema characteristic(s). To achieve that, I added an
Option<Unique::{Value,Identity}> type to Attribute; this helps with
this patch, and also looks ahead to when we allow to retract
:db/unique attributes.
* Post: Allow to retract :db/ident.
* Post: Include more details about invalid schema changes.
The tests use strings, so they hide the chained errors which do in
fact provide more detail.
* Review comment: Fix outdated comment.
* Review comment: s/_SET/_SQL_LIST/.
* Review comment: Use a sub-select for checking cardinality.
This might be faster in practice.
* Review comment: Put `attribute::Unique` into its own namespace.
2017-03-20 20:18:59 +00:00
// We can't always migrate to be :db.unique/value.
assert_transact! ( conn , " [[:db/add :test/ident :db/unique :db.unique/value]] " ,
// TODO: give more helpful error details.
2018-04-03 20:18:56 +00:00
Err ( " schema alteration failed: Cannot alter schema attribute 100 to be :db.unique/value " ) ) ;
Schema alteration. Fixes #294 and #295. (#370) r=rnewman
* Pre: Don't retract :db/ident in test.
Datomic (and eventually Mentat) don't allow to retract :db/ident in
this way, so this runs afoul of future work to support mutating
metadata.
* Pre: s/VALUETYPE/VALUE_TYPE/.
This is consistent with the capitalization (which is "valueType") and
the other identifier.
* Pre: Remove some single quotes from error output.
* Part 1: Make materialized views be uniform [e a v value_type_tag].
This looks ahead to a time when we could support arbitrary
user-defined materialized views. For now, the "idents" materialized
view is those datoms of the form [e :db/ident :namespaced/keyword] and
the "schema" materialized view is those datoms of the form [e a v]
where a is in a particular set of attributes that will become clear in
the following commits.
This change is not backwards compatible, so I'm removing the open
current (really, v2) test. It'll be re-instated when we get to
https://github.com/mozilla/mentat/issues/194.
* Pre: Map TypedValue::Ref to TypedValue::Keyword in debug output.
* Part 3: Separate `schema_to_mutate` from the `schema` used to interpret.
This is just to keep track of the expected changes during
bootstrapping. I want bootstrap metadata mutations to flow through
the same code path as metadata mutations during regular transactions;
by differentiating the schema used for interpretation from the schema
that will be updated I expect to be able to apply bootstrap metadata
mutations to an empty schema and have things like materialized views
created (using the regular code paths).
This commit has been re-ordered for conceptual clarity, but it won't
compile because it references the metadata module. It's possible to
make it compile -- the functionality is there in the schema module --
but it's not worth the rebasing effort until after review (and
possibly not even then, since we'll squash down to a single commit to
land).
* Part 2: Maintain entids separately from idents.
In order to support historical idents, we need to distinguish the
"current" map from entid -> ident from the "complete historical" map
ident -> entid. This is what Datomic does; in Datomic, an ident is
never retracted (although it can be replaced). This approach is an
important part of allowing multiple consumers to share a schema
fragment as it migrates forward.
This fixes a limitation of the Clojure implementation, which did not
handle historical idents across knowledge base close and re-open.
The "entids" materialized view is naturally a slice of the "datoms"
table. The "idents" materialized view is a slice of the
"transactions" table. I hope that representing in this way, and
casting the problem in this light, might generalize to future
materialized views.
* Pre: Add DiffSet.
* Part 4: Collect mutations to a `Schema`.
I haven't taken your review comment about consuming AttributeBuilder
during each fluent function. If you read my response and still want
this, I'm happy to do it in review.
* Part 5: Handle :db/ident and :db.{install,alter}/attribute.
This "loops" the committed datoms out of the SQL store and back
through the metadata (schema, but in future also partition map)
processor. The metadata processor updates the schema and produces a
report of what changed; that report is then used to update the SQL
store. That update includes:
- the materialized views ("entids", "idents", and "schema");
- if needed, a subset of the datoms themselves (as flags change).
I've left a TODO for handling attribute retraction in the cases that
it makes sense. I expect that to be straight-forward.
* Review comment: Rename DiffSet to AddRetractAlterSet.
Also adds a little more commentary and a simple test.
* Review comment: Use ToIdent trait.
* Review comment: partially revert "Part 2: Maintain entids separately from idents."
This reverts commit 23a91df9c35e14398f2ddbd1ba25315821e67401.
Following our discussion, this removes the "entids" materialized
view. The next commit will remove historical idents from the "idents"
materialized view.
* Post: Use custom Either rather than std::result::Result.
This is not necessary, but it was suggested that we might be paying an
overhead creating Err instances while using error_chain. That seems
not to be the case, but this change shows that we don't actually use
any of the Result helper methods, so there's no reason to overload
Result. This change might avoid some future confusion, so I'm going
to land it anyway.
Signed-off-by: Nick Alexander <nalexander@mozilla.com>
* Review comment: Don't preserve historical idents.
* Review comment: More prepared statements when updating materialized views.
* Post: Test altering :db/cardinality and :db/unique.
These tests fail due to a Datomic limitation, namely that the marker
flag :db.alter/attribute can only be asserted once for an attribute!
That is, [:db.part/db :db.alter/attribute :attribute] will only be
transacted at most once. Since older versions of Datomic required the
:db.alter/attribute flag, I can only imagine they either never wrote
:db.alter/attribute to the store, or they handled it specially. I'll
need to remove the marker flag system from Mentat in order to address
this fundamental limitation.
* Post: Remove some more single quotes from error output.
* Post: Add assert_transact! macro to unwrap safely.
I was finding it very difficult to track unwrapping errors while
making changes, due to an underlying Mac OS X symbolication issue that
makes running tests with RUST_BACKTRACE=1 so slow that they all time
out.
* Post: Don't expect or recognize :db.{install,alter}/attribute.
I had this all working... except we will never see a repeated
`[:db.part/db :db.alter/attribute :attribute]` assertion in the store!
That means my approach would let you alter an attribute at most one
time. It's not worth hacking around this; it's better to just stop
expecting (and recognizing) the marker flags. (We have all the data
to distinguish the various cases that we need without the marker
flags.)
This brings Mentat in line with the thrust of newer Datomic versions,
but isn't compatible with Datomic, because (if I understand correctly)
Datomic automatically adds :db.{install,alter}/attribute assertions to
transactions.
I haven't purged the corresponding :db/ident and schema fragments just
yet:
- we might want them back
- we might want them in order to upgrade v1 and v2 databases to the
new on-disk layout we're fleshing out (v3?).
* Post: Don't make :db/unique :db.unique/* imply :db/index true.
This patch avoids a potential bug with the "schema" materialized view.
If :db/unique :db.unique/value implies :db/index true, then what
happens when you _retract_ :db.unique/value? I think Datomic defines
this in some way, but I really want the "schema" materialized view to
be a slice of "datoms" and not have these sort of ambiguities and
persistent effects. Therefore, to ensure that we don't retract a
schema characteristic and accidentally change more than we intended
to, this patch stops having any schema characteristic imply any other
schema characteristic(s). To achieve that, I added an
Option<Unique::{Value,Identity}> type to Attribute; this helps with
this patch, and also looks ahead to when we allow to retract
:db/unique attributes.
* Post: Allow to retract :db/ident.
* Post: Include more details about invalid schema changes.
The tests use strings, so they hide the chained errors which do in
fact provide more detail.
* Review comment: Fix outdated comment.
* Review comment: s/_SET/_SQL_LIST/.
* Review comment: Use a sub-select for checking cardinality.
This might be faster in practice.
* Review comment: Put `attribute::Unique` into its own namespace.
2017-03-20 20:18:59 +00:00
// Not even indirectly!
assert_transact! ( conn , " [[:db/add :test/ident :db/unique :db.unique/identity]] " ,
// TODO: give more helpful error details.
2018-04-03 20:18:56 +00:00
Err ( " schema alteration failed: Cannot alter schema attribute 100 to be :db.unique/identity " ) ) ;
Schema alteration. Fixes #294 and #295. (#370) r=rnewman
* Pre: Don't retract :db/ident in test.
Datomic (and eventually Mentat) don't allow to retract :db/ident in
this way, so this runs afoul of future work to support mutating
metadata.
* Pre: s/VALUETYPE/VALUE_TYPE/.
This is consistent with the capitalization (which is "valueType") and
the other identifier.
* Pre: Remove some single quotes from error output.
* Part 1: Make materialized views be uniform [e a v value_type_tag].
This looks ahead to a time when we could support arbitrary
user-defined materialized views. For now, the "idents" materialized
view is those datoms of the form [e :db/ident :namespaced/keyword] and
the "schema" materialized view is those datoms of the form [e a v]
where a is in a particular set of attributes that will become clear in
the following commits.
This change is not backwards compatible, so I'm removing the open
current (really, v2) test. It'll be re-instated when we get to
https://github.com/mozilla/mentat/issues/194.
* Pre: Map TypedValue::Ref to TypedValue::Keyword in debug output.
* Part 3: Separate `schema_to_mutate` from the `schema` used to interpret.
This is just to keep track of the expected changes during
bootstrapping. I want bootstrap metadata mutations to flow through
the same code path as metadata mutations during regular transactions;
by differentiating the schema used for interpretation from the schema
that will be updated I expect to be able to apply bootstrap metadata
mutations to an empty schema and have things like materialized views
created (using the regular code paths).
This commit has been re-ordered for conceptual clarity, but it won't
compile because it references the metadata module. It's possible to
make it compile -- the functionality is there in the schema module --
but it's not worth the rebasing effort until after review (and
possibly not even then, since we'll squash down to a single commit to
land).
* Part 2: Maintain entids separately from idents.
In order to support historical idents, we need to distinguish the
"current" map from entid -> ident from the "complete historical" map
ident -> entid. This is what Datomic does; in Datomic, an ident is
never retracted (although it can be replaced). This approach is an
important part of allowing multiple consumers to share a schema
fragment as it migrates forward.
This fixes a limitation of the Clojure implementation, which did not
handle historical idents across knowledge base close and re-open.
The "entids" materialized view is naturally a slice of the "datoms"
table. The "idents" materialized view is a slice of the
"transactions" table. I hope that representing in this way, and
casting the problem in this light, might generalize to future
materialized views.
* Pre: Add DiffSet.
* Part 4: Collect mutations to a `Schema`.
I haven't taken your review comment about consuming AttributeBuilder
during each fluent function. If you read my response and still want
this, I'm happy to do it in review.
* Part 5: Handle :db/ident and :db.{install,alter}/attribute.
This "loops" the committed datoms out of the SQL store and back
through the metadata (schema, but in future also partition map)
processor. The metadata processor updates the schema and produces a
report of what changed; that report is then used to update the SQL
store. That update includes:
- the materialized views ("entids", "idents", and "schema");
- if needed, a subset of the datoms themselves (as flags change).
I've left a TODO for handling attribute retraction in the cases that
it makes sense. I expect that to be straight-forward.
* Review comment: Rename DiffSet to AddRetractAlterSet.
Also adds a little more commentary and a simple test.
* Review comment: Use ToIdent trait.
* Review comment: partially revert "Part 2: Maintain entids separately from idents."
This reverts commit 23a91df9c35e14398f2ddbd1ba25315821e67401.
Following our discussion, this removes the "entids" materialized
view. The next commit will remove historical idents from the "idents"
materialized view.
* Post: Use custom Either rather than std::result::Result.
This is not necessary, but it was suggested that we might be paying an
overhead creating Err instances while using error_chain. That seems
not to be the case, but this change shows that we don't actually use
any of the Result helper methods, so there's no reason to overload
Result. This change might avoid some future confusion, so I'm going
to land it anyway.
Signed-off-by: Nick Alexander <nalexander@mozilla.com>
* Review comment: Don't preserve historical idents.
* Review comment: More prepared statements when updating materialized views.
* Post: Test altering :db/cardinality and :db/unique.
These tests fail due to a Datomic limitation, namely that the marker
flag :db.alter/attribute can only be asserted once for an attribute!
That is, [:db.part/db :db.alter/attribute :attribute] will only be
transacted at most once. Since older versions of Datomic required the
:db.alter/attribute flag, I can only imagine they either never wrote
:db.alter/attribute to the store, or they handled it specially. I'll
need to remove the marker flag system from Mentat in order to address
this fundamental limitation.
* Post: Remove some more single quotes from error output.
* Post: Add assert_transact! macro to unwrap safely.
I was finding it very difficult to track unwrapping errors while
making changes, due to an underlying Mac OS X symbolication issue that
makes running tests with RUST_BACKTRACE=1 so slow that they all time
out.
* Post: Don't expect or recognize :db.{install,alter}/attribute.
I had this all working... except we will never see a repeated
`[:db.part/db :db.alter/attribute :attribute]` assertion in the store!
That means my approach would let you alter an attribute at most one
time. It's not worth hacking around this; it's better to just stop
expecting (and recognizing) the marker flags. (We have all the data
to distinguish the various cases that we need without the marker
flags.)
This brings Mentat in line with the thrust of newer Datomic versions,
but isn't compatible with Datomic, because (if I understand correctly)
Datomic automatically adds :db.{install,alter}/attribute assertions to
transactions.
I haven't purged the corresponding :db/ident and schema fragments just
yet:
- we might want them back
- we might want them in order to upgrade v1 and v2 databases to the
new on-disk layout we're fleshing out (v3?).
* Post: Don't make :db/unique :db.unique/* imply :db/index true.
This patch avoids a potential bug with the "schema" materialized view.
If :db/unique :db.unique/value implies :db/index true, then what
happens when you _retract_ :db.unique/value? I think Datomic defines
this in some way, but I really want the "schema" materialized view to
be a slice of "datoms" and not have these sort of ambiguities and
persistent effects. Therefore, to ensure that we don't retract a
schema characteristic and accidentally change more than we intended
to, this patch stops having any schema characteristic imply any other
schema characteristic(s). To achieve that, I added an
Option<Unique::{Value,Identity}> type to Attribute; this helps with
this patch, and also looks ahead to when we allow to retract
:db/unique attributes.
* Post: Allow to retract :db/ident.
* Post: Include more details about invalid schema changes.
The tests use strings, so they hide the chained errors which do in
fact provide more detail.
* Review comment: Fix outdated comment.
* Review comment: s/_SET/_SQL_LIST/.
* Review comment: Use a sub-select for checking cardinality.
This might be faster in practice.
* Review comment: Put `attribute::Unique` into its own namespace.
2017-03-20 20:18:59 +00:00
// But we can if we make sure there's no repeated [a v] pair.
assert_transact! ( conn , " [[:db/add 201 :test/ident 2]] " ) ;
2019-07-17 15:46:47 +00:00
assert_transact! (
conn ,
" [[:db/add :test/ident :db/index true]
Schema alteration. Fixes #294 and #295. (#370) r=rnewman
* Pre: Don't retract :db/ident in test.
Datomic (and eventually Mentat) don't allow to retract :db/ident in
this way, so this runs afoul of future work to support mutating
metadata.
* Pre: s/VALUETYPE/VALUE_TYPE/.
This is consistent with the capitalization (which is "valueType") and
the other identifier.
* Pre: Remove some single quotes from error output.
* Part 1: Make materialized views be uniform [e a v value_type_tag].
This looks ahead to a time when we could support arbitrary
user-defined materialized views. For now, the "idents" materialized
view is those datoms of the form [e :db/ident :namespaced/keyword] and
the "schema" materialized view is those datoms of the form [e a v]
where a is in a particular set of attributes that will become clear in
the following commits.
This change is not backwards compatible, so I'm removing the open
current (really, v2) test. It'll be re-instated when we get to
https://github.com/mozilla/mentat/issues/194.
* Pre: Map TypedValue::Ref to TypedValue::Keyword in debug output.
* Part 3: Separate `schema_to_mutate` from the `schema` used to interpret.
This is just to keep track of the expected changes during
bootstrapping. I want bootstrap metadata mutations to flow through
the same code path as metadata mutations during regular transactions;
by differentiating the schema used for interpretation from the schema
that will be updated I expect to be able to apply bootstrap metadata
mutations to an empty schema and have things like materialized views
created (using the regular code paths).
This commit has been re-ordered for conceptual clarity, but it won't
compile because it references the metadata module. It's possible to
make it compile -- the functionality is there in the schema module --
but it's not worth the rebasing effort until after review (and
possibly not even then, since we'll squash down to a single commit to
land).
* Part 2: Maintain entids separately from idents.
In order to support historical idents, we need to distinguish the
"current" map from entid -> ident from the "complete historical" map
ident -> entid. This is what Datomic does; in Datomic, an ident is
never retracted (although it can be replaced). This approach is an
important part of allowing multiple consumers to share a schema
fragment as it migrates forward.
This fixes a limitation of the Clojure implementation, which did not
handle historical idents across knowledge base close and re-open.
The "entids" materialized view is naturally a slice of the "datoms"
table. The "idents" materialized view is a slice of the
"transactions" table. I hope that representing in this way, and
casting the problem in this light, might generalize to future
materialized views.
* Pre: Add DiffSet.
* Part 4: Collect mutations to a `Schema`.
I haven't taken your review comment about consuming AttributeBuilder
during each fluent function. If you read my response and still want
this, I'm happy to do it in review.
* Part 5: Handle :db/ident and :db.{install,alter}/attribute.
This "loops" the committed datoms out of the SQL store and back
through the metadata (schema, but in future also partition map)
processor. The metadata processor updates the schema and produces a
report of what changed; that report is then used to update the SQL
store. That update includes:
- the materialized views ("entids", "idents", and "schema");
- if needed, a subset of the datoms themselves (as flags change).
I've left a TODO for handling attribute retraction in the cases that
it makes sense. I expect that to be straight-forward.
* Review comment: Rename DiffSet to AddRetractAlterSet.
Also adds a little more commentary and a simple test.
* Review comment: Use ToIdent trait.
* Review comment: partially revert "Part 2: Maintain entids separately from idents."
This reverts commit 23a91df9c35e14398f2ddbd1ba25315821e67401.
Following our discussion, this removes the "entids" materialized
view. The next commit will remove historical idents from the "idents"
materialized view.
* Post: Use custom Either rather than std::result::Result.
This is not necessary, but it was suggested that we might be paying an
overhead creating Err instances while using error_chain. That seems
not to be the case, but this change shows that we don't actually use
any of the Result helper methods, so there's no reason to overload
Result. This change might avoid some future confusion, so I'm going
to land it anyway.
Signed-off-by: Nick Alexander <nalexander@mozilla.com>
* Review comment: Don't preserve historical idents.
* Review comment: More prepared statements when updating materialized views.
* Post: Test altering :db/cardinality and :db/unique.
These tests fail due to a Datomic limitation, namely that the marker
flag :db.alter/attribute can only be asserted once for an attribute!
That is, [:db.part/db :db.alter/attribute :attribute] will only be
transacted at most once. Since older versions of Datomic required the
:db.alter/attribute flag, I can only imagine they either never wrote
:db.alter/attribute to the store, or they handled it specially. I'll
need to remove the marker flag system from Mentat in order to address
this fundamental limitation.
* Post: Remove some more single quotes from error output.
* Post: Add assert_transact! macro to unwrap safely.
I was finding it very difficult to track unwrapping errors while
making changes, due to an underlying Mac OS X symbolication issue that
makes running tests with RUST_BACKTRACE=1 so slow that they all time
out.
* Post: Don't expect or recognize :db.{install,alter}/attribute.
I had this all working... except we will never see a repeated
`[:db.part/db :db.alter/attribute :attribute]` assertion in the store!
That means my approach would let you alter an attribute at most one
time. It's not worth hacking around this; it's better to just stop
expecting (and recognizing) the marker flags. (We have all the data
to distinguish the various cases that we need without the marker
flags.)
This brings Mentat in line with the thrust of newer Datomic versions,
but isn't compatible with Datomic, because (if I understand correctly)
Datomic automatically adds :db.{install,alter}/attribute assertions to
transactions.
I haven't purged the corresponding :db/ident and schema fragments just
yet:
- we might want them back
- we might want them in order to upgrade v1 and v2 databases to the
new on-disk layout we're fleshing out (v3?).
* Post: Don't make :db/unique :db.unique/* imply :db/index true.
This patch avoids a potential bug with the "schema" materialized view.
If :db/unique :db.unique/value implies :db/index true, then what
happens when you _retract_ :db.unique/value? I think Datomic defines
this in some way, but I really want the "schema" materialized view to
be a slice of "datoms" and not have these sort of ambiguities and
persistent effects. Therefore, to ensure that we don't retract a
schema characteristic and accidentally change more than we intended
to, this patch stops having any schema characteristic imply any other
schema characteristic(s). To achieve that, I added an
Option<Unique::{Value,Identity}> type to Attribute; this helps with
this patch, and also looks ahead to when we allow to retract
:db/unique attributes.
* Post: Allow to retract :db/ident.
* Post: Include more details about invalid schema changes.
The tests use strings, so they hide the chained errors which do in
fact provide more detail.
* Review comment: Fix outdated comment.
* Review comment: s/_SET/_SQL_LIST/.
* Review comment: Use a sub-select for checking cardinality.
This might be faster in practice.
* Review comment: Put `attribute::Unique` into its own namespace.
2017-03-20 20:18:59 +00:00
[ :db / add :test / ident :db / unique :db . unique / value ]
2019-07-17 15:46:47 +00:00
[ :db / add :db . part / db :db . alter / attribute 100 ] ] "
) ;
2018-04-03 20:18:56 +00:00
// We can also retract the uniqueness constraint altogether.
2019-07-17 15:46:47 +00:00
assert_transact! (
conn ,
" [[:db/retract :test/ident :db/unique :db.unique/value]] "
) ;
2018-04-03 20:18:56 +00:00
// Once we've done so, the schema shows it's not unique…
{
2019-07-17 15:46:47 +00:00
let attr = conn
. schema
. attribute_for_ident ( & Keyword ::namespaced ( " test " , " ident " ) )
. unwrap ( )
. 0 ;
2018-04-03 20:18:56 +00:00
assert_eq! ( None , attr . unique ) ;
}
// … and we can add more assertions with duplicate values.
2019-07-17 15:46:47 +00:00
assert_transact! (
conn ,
" [[:db/add 121 :test/ident 1]
[ :db / add 221 :test / ident 2 ] ] "
) ;
Schema alteration. Fixes #294 and #295. (#370) r=rnewman
* Pre: Don't retract :db/ident in test.
Datomic (and eventually Mentat) don't allow to retract :db/ident in
this way, so this runs afoul of future work to support mutating
metadata.
* Pre: s/VALUETYPE/VALUE_TYPE/.
This is consistent with the capitalization (which is "valueType") and
the other identifier.
* Pre: Remove some single quotes from error output.
* Part 1: Make materialized views be uniform [e a v value_type_tag].
This looks ahead to a time when we could support arbitrary
user-defined materialized views. For now, the "idents" materialized
view is those datoms of the form [e :db/ident :namespaced/keyword] and
the "schema" materialized view is those datoms of the form [e a v]
where a is in a particular set of attributes that will become clear in
the following commits.
This change is not backwards compatible, so I'm removing the open
current (really, v2) test. It'll be re-instated when we get to
https://github.com/mozilla/mentat/issues/194.
* Pre: Map TypedValue::Ref to TypedValue::Keyword in debug output.
* Part 3: Separate `schema_to_mutate` from the `schema` used to interpret.
This is just to keep track of the expected changes during
bootstrapping. I want bootstrap metadata mutations to flow through
the same code path as metadata mutations during regular transactions;
by differentiating the schema used for interpretation from the schema
that will be updated I expect to be able to apply bootstrap metadata
mutations to an empty schema and have things like materialized views
created (using the regular code paths).
This commit has been re-ordered for conceptual clarity, but it won't
compile because it references the metadata module. It's possible to
make it compile -- the functionality is there in the schema module --
but it's not worth the rebasing effort until after review (and
possibly not even then, since we'll squash down to a single commit to
land).
* Part 2: Maintain entids separately from idents.
In order to support historical idents, we need to distinguish the
"current" map from entid -> ident from the "complete historical" map
ident -> entid. This is what Datomic does; in Datomic, an ident is
never retracted (although it can be replaced). This approach is an
important part of allowing multiple consumers to share a schema
fragment as it migrates forward.
This fixes a limitation of the Clojure implementation, which did not
handle historical idents across knowledge base close and re-open.
The "entids" materialized view is naturally a slice of the "datoms"
table. The "idents" materialized view is a slice of the
"transactions" table. I hope that representing in this way, and
casting the problem in this light, might generalize to future
materialized views.
* Pre: Add DiffSet.
* Part 4: Collect mutations to a `Schema`.
I haven't taken your review comment about consuming AttributeBuilder
during each fluent function. If you read my response and still want
this, I'm happy to do it in review.
* Part 5: Handle :db/ident and :db.{install,alter}/attribute.
This "loops" the committed datoms out of the SQL store and back
through the metadata (schema, but in future also partition map)
processor. The metadata processor updates the schema and produces a
report of what changed; that report is then used to update the SQL
store. That update includes:
- the materialized views ("entids", "idents", and "schema");
- if needed, a subset of the datoms themselves (as flags change).
I've left a TODO for handling attribute retraction in the cases that
it makes sense. I expect that to be straight-forward.
* Review comment: Rename DiffSet to AddRetractAlterSet.
Also adds a little more commentary and a simple test.
* Review comment: Use ToIdent trait.
* Review comment: partially revert "Part 2: Maintain entids separately from idents."
This reverts commit 23a91df9c35e14398f2ddbd1ba25315821e67401.
Following our discussion, this removes the "entids" materialized
view. The next commit will remove historical idents from the "idents"
materialized view.
* Post: Use custom Either rather than std::result::Result.
This is not necessary, but it was suggested that we might be paying an
overhead creating Err instances while using error_chain. That seems
not to be the case, but this change shows that we don't actually use
any of the Result helper methods, so there's no reason to overload
Result. This change might avoid some future confusion, so I'm going
to land it anyway.
Signed-off-by: Nick Alexander <nalexander@mozilla.com>
* Review comment: Don't preserve historical idents.
* Review comment: More prepared statements when updating materialized views.
* Post: Test altering :db/cardinality and :db/unique.
These tests fail due to a Datomic limitation, namely that the marker
flag :db.alter/attribute can only be asserted once for an attribute!
That is, [:db.part/db :db.alter/attribute :attribute] will only be
transacted at most once. Since older versions of Datomic required the
:db.alter/attribute flag, I can only imagine they either never wrote
:db.alter/attribute to the store, or they handled it specially. I'll
need to remove the marker flag system from Mentat in order to address
this fundamental limitation.
* Post: Remove some more single quotes from error output.
* Post: Add assert_transact! macro to unwrap safely.
I was finding it very difficult to track unwrapping errors while
making changes, due to an underlying Mac OS X symbolication issue that
makes running tests with RUST_BACKTRACE=1 so slow that they all time
out.
* Post: Don't expect or recognize :db.{install,alter}/attribute.
I had this all working... except we will never see a repeated
`[:db.part/db :db.alter/attribute :attribute]` assertion in the store!
That means my approach would let you alter an attribute at most one
time. It's not worth hacking around this; it's better to just stop
expecting (and recognizing) the marker flags. (We have all the data
to distinguish the various cases that we need without the marker
flags.)
This brings Mentat in line with the thrust of newer Datomic versions,
but isn't compatible with Datomic, because (if I understand correctly)
Datomic automatically adds :db.{install,alter}/attribute assertions to
transactions.
I haven't purged the corresponding :db/ident and schema fragments just
yet:
- we might want them back
- we might want them in order to upgrade v1 and v2 databases to the
new on-disk layout we're fleshing out (v3?).
* Post: Don't make :db/unique :db.unique/* imply :db/index true.
This patch avoids a potential bug with the "schema" materialized view.
If :db/unique :db.unique/value implies :db/index true, then what
happens when you _retract_ :db.unique/value? I think Datomic defines
this in some way, but I really want the "schema" materialized view to
be a slice of "datoms" and not have these sort of ambiguities and
persistent effects. Therefore, to ensure that we don't retract a
schema characteristic and accidentally change more than we intended
to, this patch stops having any schema characteristic imply any other
schema characteristic(s). To achieve that, I added an
Option<Unique::{Value,Identity}> type to Attribute; this helps with
this patch, and also looks ahead to when we allow to retract
:db/unique attributes.
* Post: Allow to retract :db/ident.
* Post: Include more details about invalid schema changes.
The tests use strings, so they hide the chained errors which do in
fact provide more detail.
* Review comment: Fix outdated comment.
* Review comment: s/_SET/_SQL_LIST/.
* Review comment: Use a sub-select for checking cardinality.
This might be faster in practice.
* Review comment: Put `attribute::Unique` into its own namespace.
2017-03-20 20:18:59 +00:00
}
2017-03-21 20:12:10 +00:00
2018-09-07 20:12:28 +00:00
#[ test ]
fn test_db_double_retraction_issue_818 ( ) {
let mut conn = TestConn ::default ( ) ;
// Start by installing a :db.cardinality/one attribute.
2019-07-17 15:46:47 +00:00
assert_transact! (
conn ,
" [[:db/add 100 :db/ident :test/ident]
2018-09-07 20:12:28 +00:00
[ :db / add 100 :db / valueType :db . type / string ]
[ :db / add 100 :db / cardinality :db . cardinality / one ]
[ :db / add 100 :db / unique :db . unique / identity ]
2019-07-17 15:46:47 +00:00
[ :db / add 100 :db / index true ] ] "
) ;
2018-09-07 20:12:28 +00:00
assert_transact! ( conn , " [[:db/add 200 :test/ident \" Oi \" ]] " ) ;
2019-07-17 15:46:47 +00:00
assert_transact! (
conn ,
" [[:db/add 200 :test/ident \" Ai! \" ]
[ :db / retract 200 :test / ident \ " Oi \" ]] "
) ;
2018-09-07 20:12:28 +00:00
2019-07-17 15:46:47 +00:00
assert_matches! (
conn . last_transaction ( ) ,
" [[200 :test/ident \" Ai! \" ?tx true]
2018-09-07 20:12:28 +00:00
[ 200 :test / ident \ " Oi \" ?tx false]
2019-07-17 15:46:47 +00:00
[ ? tx :db / txInstant ? ms ? tx true ] ] "
) ;
2018-09-07 20:12:28 +00:00
2019-07-17 15:46:47 +00:00
assert_matches! (
conn . datoms ( ) ,
" [[100 :db/ident :test/ident]
2018-09-07 20:12:28 +00:00
[ 100 :db / valueType :db . type / string ]
[ 100 :db / cardinality :db . cardinality / one ]
[ 100 :db / unique :db . unique / identity ]
[ 100 :db / index true ]
2019-07-17 15:46:47 +00:00
[ 200 :test / ident \ " Ai! \" ]] "
) ;
2018-09-07 20:12:28 +00:00
}
2017-03-21 20:12:10 +00:00
/// Verify that we can't alter :db/fulltext schema characteristics at all.
#[ test ]
fn test_db_alter_fulltext ( ) {
let mut conn = TestConn ::default ( ) ;
// Start by installing a :db/fulltext true and a :db/fulltext unset attribute.
2019-07-17 15:46:47 +00:00
assert_transact! (
conn ,
" [[:db/add 111 :db/ident :test/fulltext]
2017-03-21 20:12:10 +00:00
[ :db / add 111 :db / valueType :db . type / string ]
[ :db / add 111 :db / unique :db . unique / identity ]
[ :db / add 111 :db / index true ]
[ :db / add 111 :db / fulltext true ]
[ :db / add 222 :db / ident :test / string ]
[ :db / add 222 :db / cardinality :db . cardinality / one ]
[ :db / add 222 :db / valueType :db . type / string ]
2019-07-17 15:46:47 +00:00
[ :db / add 222 :db / index true ] ] "
) ;
2017-03-21 20:12:10 +00:00
2019-07-17 15:46:47 +00:00
assert_transact! (
conn ,
" [[:db/retract 111 :db/fulltext true]] " ,
Err ( " bad schema assertion: Retracting attribute 12 for entity 111 not permitted. " )
) ;
2017-03-21 20:12:10 +00:00
assert_transact! ( conn ,
" [[:db/add 222 :db/fulltext true]] " ,
Err ( " bad schema assertion: Schema alteration for existing attribute with entid 222 is not valid " ) ) ;
}
#[ test ]
fn test_db_fulltext ( ) {
let mut conn = TestConn ::default ( ) ;
// Start by installing a few :db/fulltext true attributes.
2019-07-17 15:46:47 +00:00
assert_transact! (
conn ,
" [[:db/add 111 :db/ident :test/fulltext]
2017-03-21 20:12:10 +00:00
[ :db / add 111 :db / valueType :db . type / string ]
[ :db / add 111 :db / unique :db . unique / identity ]
[ :db / add 111 :db / index true ]
[ :db / add 111 :db / fulltext true ]
[ :db / add 222 :db / ident :test / other ]
[ :db / add 222 :db / cardinality :db . cardinality / one ]
[ :db / add 222 :db / valueType :db . type / string ]
[ :db / add 222 :db / index true ]
2019-07-17 15:46:47 +00:00
[ :db / add 222 :db / fulltext true ] ] "
) ;
2017-03-21 20:12:10 +00:00
// Let's check we actually have the schema characteristics we expect.
2019-07-17 15:46:47 +00:00
let fulltext = conn
. schema
. attribute_for_entid ( 111 )
. cloned ( )
. expect ( " :test/fulltext " ) ;
2017-03-21 20:12:10 +00:00
assert_eq! ( fulltext . value_type , ValueType ::String ) ;
assert_eq! ( fulltext . fulltext , true ) ;
assert_eq! ( fulltext . multival , false ) ;
assert_eq! ( fulltext . unique , Some ( attribute ::Unique ::Identity ) ) ;
2019-07-17 15:46:47 +00:00
let other = conn
. schema
. attribute_for_entid ( 222 )
. cloned ( )
. expect ( " :test/other " ) ;
2017-03-21 20:12:10 +00:00
assert_eq! ( other . value_type , ValueType ::String ) ;
assert_eq! ( other . fulltext , true ) ;
assert_eq! ( other . multival , false ) ;
assert_eq! ( other . unique , None ) ;
// We can add fulltext indexed datoms.
assert_transact! ( conn , " [[:db/add 301 :test/fulltext \" test this \" ]] " ) ;
// value column is rowid into fulltext table.
2019-07-17 15:46:47 +00:00
assert_matches! ( conn . fulltext_values ( ) , " [[1 \" test this \" ]] " ) ;
assert_matches! (
conn . last_transaction ( ) ,
" [[301 :test/fulltext 1 ?tx true]
[ ? tx :db / txInstant ? ms ? tx true ] ] "
) ;
assert_matches! (
conn . datoms ( ) ,
" [[111 :db/ident :test/fulltext]
2017-03-21 20:12:10 +00:00
[ 111 :db / valueType :db . type / string ]
[ 111 :db / unique :db . unique / identity ]
[ 111 :db / index true ]
[ 111 :db / fulltext true ]
[ 222 :db / ident :test / other ]
[ 222 :db / valueType :db . type / string ]
[ 222 :db / cardinality :db . cardinality / one ]
[ 222 :db / index true ]
[ 222 :db / fulltext true ]
2019-07-17 15:46:47 +00:00
[ 301 :test / fulltext 1 ] ] "
) ;
2017-03-21 20:12:10 +00:00
// We can replace existing fulltext indexed datoms.
assert_transact! ( conn , " [[:db/add 301 :test/fulltext \" alternate thing \" ]] " ) ;
// value column is rowid into fulltext table.
2019-07-17 15:46:47 +00:00
assert_matches! (
conn . fulltext_values ( ) ,
" [[1 \" test this \" ]
[ 2 \ " alternate thing \" ]] "
) ;
assert_matches! (
conn . last_transaction ( ) ,
" [[301 :test/fulltext 1 ?tx false]
2017-03-21 20:12:10 +00:00
[ 301 :test / fulltext 2 ? tx true ]
2019-07-17 15:46:47 +00:00
[ ? tx :db / txInstant ? ms ? tx true ] ] "
) ;
assert_matches! (
conn . datoms ( ) ,
" [[111 :db/ident :test/fulltext]
2017-03-21 20:12:10 +00:00
[ 111 :db / valueType :db . type / string ]
[ 111 :db / unique :db . unique / identity ]
[ 111 :db / index true ]
[ 111 :db / fulltext true ]
[ 222 :db / ident :test / other ]
[ 222 :db / valueType :db . type / string ]
[ 222 :db / cardinality :db . cardinality / one ]
[ 222 :db / index true ]
[ 222 :db / fulltext true ]
2019-07-17 15:46:47 +00:00
[ 301 :test / fulltext 2 ] ] "
) ;
2017-03-21 20:12:10 +00:00
// We can upsert keyed by fulltext indexed datoms.
2019-07-17 15:46:47 +00:00
assert_transact! (
conn ,
" [[:db/add \" t \" :test/fulltext \" alternate thing \" ]
[ :db / add \ " t \" :test/other \" other \" ]] "
) ;
2017-03-21 20:12:10 +00:00
// value column is rowid into fulltext table.
2019-07-17 15:46:47 +00:00
assert_matches! (
conn . fulltext_values ( ) ,
" [[1 \" test this \" ]
2017-03-21 20:12:10 +00:00
[ 2 \ " alternate thing \" ]
2019-07-17 15:46:47 +00:00
[ 3 \ " other \" ]] "
) ;
assert_matches! (
conn . last_transaction ( ) ,
" [[301 :test/other 3 ?tx true]
[ ? tx :db / txInstant ? ms ? tx true ] ] "
) ;
assert_matches! (
conn . datoms ( ) ,
" [[111 :db/ident :test/fulltext]
2017-03-21 20:12:10 +00:00
[ 111 :db / valueType :db . type / string ]
[ 111 :db / unique :db . unique / identity ]
[ 111 :db / index true ]
[ 111 :db / fulltext true ]
[ 222 :db / ident :test / other ]
[ 222 :db / valueType :db . type / string ]
[ 222 :db / cardinality :db . cardinality / one ]
[ 222 :db / index true ]
[ 222 :db / fulltext true ]
[ 301 :test / fulltext 2 ]
2019-07-17 15:46:47 +00:00
[ 301 :test / other 3 ] ] "
) ;
2017-03-21 20:12:10 +00:00
// We can re-use fulltext values; they won't be added to the fulltext values table twice.
assert_transact! ( conn , " [[:db/add 302 :test/other \" alternate thing \" ]] " ) ;
// value column is rowid into fulltext table.
2019-07-17 15:46:47 +00:00
assert_matches! (
conn . fulltext_values ( ) ,
" [[1 \" test this \" ]
2017-03-21 20:12:10 +00:00
[ 2 \ " alternate thing \" ]
2019-07-17 15:46:47 +00:00
[ 3 \ " other \" ]] "
) ;
assert_matches! (
conn . last_transaction ( ) ,
" [[302 :test/other 2 ?tx true]
[ ? tx :db / txInstant ? ms ? tx true ] ] "
) ;
assert_matches! (
conn . datoms ( ) ,
" [[111 :db/ident :test/fulltext]
2017-03-21 20:12:10 +00:00
[ 111 :db / valueType :db . type / string ]
[ 111 :db / unique :db . unique / identity ]
[ 111 :db / index true ]
[ 111 :db / fulltext true ]
[ 222 :db / ident :test / other ]
[ 222 :db / valueType :db . type / string ]
[ 222 :db / cardinality :db . cardinality / one ]
[ 222 :db / index true ]
[ 222 :db / fulltext true ]
[ 301 :test / fulltext 2 ]
[ 301 :test / other 3 ]
2019-07-17 15:46:47 +00:00
[ 302 :test / other 2 ] ] "
) ;
2017-03-21 20:12:10 +00:00
// We can retract fulltext indexed datoms. The underlying fulltext value remains -- indeed,
// it might still be in use.
assert_transact! ( conn , " [[:db/retract 302 :test/other \" alternate thing \" ]] " ) ;
// value column is rowid into fulltext table.
2019-07-17 15:46:47 +00:00
assert_matches! (
conn . fulltext_values ( ) ,
" [[1 \" test this \" ]
2017-03-21 20:12:10 +00:00
[ 2 \ " alternate thing \" ]
2019-07-17 15:46:47 +00:00
[ 3 \ " other \" ]] "
) ;
assert_matches! (
conn . last_transaction ( ) ,
" [[302 :test/other 2 ?tx false]
[ ? tx :db / txInstant ? ms ? tx true ] ] "
) ;
assert_matches! (
conn . datoms ( ) ,
" [[111 :db/ident :test/fulltext]
2017-03-21 20:12:10 +00:00
[ 111 :db / valueType :db . type / string ]
[ 111 :db / unique :db . unique / identity ]
[ 111 :db / index true ]
[ 111 :db / fulltext true ]
[ 222 :db / ident :test / other ]
[ 222 :db / valueType :db . type / string ]
[ 222 :db / cardinality :db . cardinality / one ]
[ 222 :db / index true ]
[ 222 :db / fulltext true ]
[ 301 :test / fulltext 2 ]
2019-07-17 15:46:47 +00:00
[ 301 :test / other 3 ] ] "
) ;
2017-03-21 20:12:10 +00:00
}
Lookup refs, nested vector values, map notation. Fixes #180, fixes #183, fixes #284. (#382) r=rnewman
* Pre: Fix error in parser macros.
* Pre: Make test unwrapping more verbose.
* Pre: Make lookup refs be (lookup-ref a v) in the entity position.
This has the advantage of being explicit in all situations and
unambiguous at parse-time. This choice agrees with the Clojure
implementation but not with Datomic. Datomic treats [a v] as a lookup
ref, is ambiguous at parse-time, and is disambiguated in ways I do not
understand at transaction time. We mooted making lookup refs [[a v]]
and outlawing nested value vectors in transactions, but after
implementing that approach I decided it was better to handle lookup
refs at parse time and therefore outlawing nested value vectors is not
necessary.
* Handle lookup refs in the entity and value columns. Fixes #183.
* Pre 0a: Use a stack instead of into_iter.
* Pre 0b: Dedent.
* Pre 0c: Handle `e` after `v`.
This allows to use the original `e` while handling `v`.
* Explode value lists for :db.cardinality/many attributes. Fixes #284.
* Parse and accept map notation. Fixes #180.
* Pre: Modernize add() and retract() into one add_or_retract().
* Pre: Add is_collection and is_atom to edn::Value.
* Pre: Differentiate atoms from lookup-refs in value position.
Initially, I expected to accept arbitrary edn::Value instances in the
value position, and to differentiate in the transactor. However, the
implementation quickly became a two-stage parser, since we always
wanted to parse the resulting value position into some other known
thing using the tx-parser. To save calls into the parser and to allow
the parser to move forward with a smaller API surface, I push as much
of this parsing as possible into the initial parse.
* Pre: Modernize entities().
* Pre: Quote edn::Value::Text in Display.
* Review comment: Add and use edn::Value::into_atom.
* Review comment: Use skip(eof()) throughout.
* Review comment: VecDeque instead of Vec.
* Review comment: Part 0: Rename TempId to TempIdHandle.
* Review comment: Part 1: Differentiate internal and external tempids.
This breaks an abstraction boundary by pushing the Internal/External
split up to the Entity level in tx/ and tx-parser/. This just makes
it easier to explode Entity map notation instances into Entity
instances, taking an existing External tempid :db/id or generating a
new Internal tempid as appropriate. To do this without breaking the
abstraction boundary would require adding flexibility to the
transaction processor: we'd need to be able to turn Entity instances
into some internal enum and handle the two cases independently. It
wouldn't be too hard, but this reduces the combinatorial type
explosion.
2017-03-27 23:30:04 +00:00
#[ test ]
fn test_lookup_refs_entity_column ( ) {
let mut conn = TestConn ::default ( ) ;
// Start by installing a few attributes.
2019-07-17 15:46:47 +00:00
assert_transact! (
conn ,
" [[:db/add 111 :db/ident :test/unique_value]
Lookup refs, nested vector values, map notation. Fixes #180, fixes #183, fixes #284. (#382) r=rnewman
* Pre: Fix error in parser macros.
* Pre: Make test unwrapping more verbose.
* Pre: Make lookup refs be (lookup-ref a v) in the entity position.
This has the advantage of being explicit in all situations and
unambiguous at parse-time. This choice agrees with the Clojure
implementation but not with Datomic. Datomic treats [a v] as a lookup
ref, is ambiguous at parse-time, and is disambiguated in ways I do not
understand at transaction time. We mooted making lookup refs [[a v]]
and outlawing nested value vectors in transactions, but after
implementing that approach I decided it was better to handle lookup
refs at parse time and therefore outlawing nested value vectors is not
necessary.
* Handle lookup refs in the entity and value columns. Fixes #183.
* Pre 0a: Use a stack instead of into_iter.
* Pre 0b: Dedent.
* Pre 0c: Handle `e` after `v`.
This allows to use the original `e` while handling `v`.
* Explode value lists for :db.cardinality/many attributes. Fixes #284.
* Parse and accept map notation. Fixes #180.
* Pre: Modernize add() and retract() into one add_or_retract().
* Pre: Add is_collection and is_atom to edn::Value.
* Pre: Differentiate atoms from lookup-refs in value position.
Initially, I expected to accept arbitrary edn::Value instances in the
value position, and to differentiate in the transactor. However, the
implementation quickly became a two-stage parser, since we always
wanted to parse the resulting value position into some other known
thing using the tx-parser. To save calls into the parser and to allow
the parser to move forward with a smaller API surface, I push as much
of this parsing as possible into the initial parse.
* Pre: Modernize entities().
* Pre: Quote edn::Value::Text in Display.
* Review comment: Add and use edn::Value::into_atom.
* Review comment: Use skip(eof()) throughout.
* Review comment: VecDeque instead of Vec.
* Review comment: Part 0: Rename TempId to TempIdHandle.
* Review comment: Part 1: Differentiate internal and external tempids.
This breaks an abstraction boundary by pushing the Internal/External
split up to the Entity level in tx/ and tx-parser/. This just makes
it easier to explode Entity map notation instances into Entity
instances, taking an existing External tempid :db/id or generating a
new Internal tempid as appropriate. To do this without breaking the
abstraction boundary would require adding flexibility to the
transaction processor: we'd need to be able to turn Entity instances
into some internal enum and handle the two cases independently. It
wouldn't be too hard, but this reduces the combinatorial type
explosion.
2017-03-27 23:30:04 +00:00
[ :db / add 111 :db / valueType :db . type / string ]
[ :db / add 111 :db / unique :db . unique / value ]
[ :db / add 111 :db / index true ]
[ :db / add 222 :db / ident :test / unique_identity ]
[ :db / add 222 :db / valueType :db . type / long ]
[ :db / add 222 :db / unique :db . unique / identity ]
[ :db / add 222 :db / index true ]
[ :db / add 333 :db / ident :test / not_unique ]
[ :db / add 333 :db / cardinality :db . cardinality / one ]
[ :db / add 333 :db / valueType :db . type / keyword ]
2019-07-17 15:46:47 +00:00
[ :db / add 333 :db / index true ] ] "
) ;
Lookup refs, nested vector values, map notation. Fixes #180, fixes #183, fixes #284. (#382) r=rnewman
* Pre: Fix error in parser macros.
* Pre: Make test unwrapping more verbose.
* Pre: Make lookup refs be (lookup-ref a v) in the entity position.
This has the advantage of being explicit in all situations and
unambiguous at parse-time. This choice agrees with the Clojure
implementation but not with Datomic. Datomic treats [a v] as a lookup
ref, is ambiguous at parse-time, and is disambiguated in ways I do not
understand at transaction time. We mooted making lookup refs [[a v]]
and outlawing nested value vectors in transactions, but after
implementing that approach I decided it was better to handle lookup
refs at parse time and therefore outlawing nested value vectors is not
necessary.
* Handle lookup refs in the entity and value columns. Fixes #183.
* Pre 0a: Use a stack instead of into_iter.
* Pre 0b: Dedent.
* Pre 0c: Handle `e` after `v`.
This allows to use the original `e` while handling `v`.
* Explode value lists for :db.cardinality/many attributes. Fixes #284.
* Parse and accept map notation. Fixes #180.
* Pre: Modernize add() and retract() into one add_or_retract().
* Pre: Add is_collection and is_atom to edn::Value.
* Pre: Differentiate atoms from lookup-refs in value position.
Initially, I expected to accept arbitrary edn::Value instances in the
value position, and to differentiate in the transactor. However, the
implementation quickly became a two-stage parser, since we always
wanted to parse the resulting value position into some other known
thing using the tx-parser. To save calls into the parser and to allow
the parser to move forward with a smaller API surface, I push as much
of this parsing as possible into the initial parse.
* Pre: Modernize entities().
* Pre: Quote edn::Value::Text in Display.
* Review comment: Add and use edn::Value::into_atom.
* Review comment: Use skip(eof()) throughout.
* Review comment: VecDeque instead of Vec.
* Review comment: Part 0: Rename TempId to TempIdHandle.
* Review comment: Part 1: Differentiate internal and external tempids.
This breaks an abstraction boundary by pushing the Internal/External
split up to the Entity level in tx/ and tx-parser/. This just makes
it easier to explode Entity map notation instances into Entity
instances, taking an existing External tempid :db/id or generating a
new Internal tempid as appropriate. To do this without breaking the
abstraction boundary would require adding flexibility to the
transaction processor: we'd need to be able to turn Entity instances
into some internal enum and handle the two cases independently. It
wouldn't be too hard, but this reduces the combinatorial type
explosion.
2017-03-27 23:30:04 +00:00
// And a few datoms to match against.
2019-07-17 15:46:47 +00:00
assert_transact! (
conn ,
" [[:db/add 501 :test/unique_value \" test this \" ]
Lookup refs, nested vector values, map notation. Fixes #180, fixes #183, fixes #284. (#382) r=rnewman
* Pre: Fix error in parser macros.
* Pre: Make test unwrapping more verbose.
* Pre: Make lookup refs be (lookup-ref a v) in the entity position.
This has the advantage of being explicit in all situations and
unambiguous at parse-time. This choice agrees with the Clojure
implementation but not with Datomic. Datomic treats [a v] as a lookup
ref, is ambiguous at parse-time, and is disambiguated in ways I do not
understand at transaction time. We mooted making lookup refs [[a v]]
and outlawing nested value vectors in transactions, but after
implementing that approach I decided it was better to handle lookup
refs at parse time and therefore outlawing nested value vectors is not
necessary.
* Handle lookup refs in the entity and value columns. Fixes #183.
* Pre 0a: Use a stack instead of into_iter.
* Pre 0b: Dedent.
* Pre 0c: Handle `e` after `v`.
This allows to use the original `e` while handling `v`.
* Explode value lists for :db.cardinality/many attributes. Fixes #284.
* Parse and accept map notation. Fixes #180.
* Pre: Modernize add() and retract() into one add_or_retract().
* Pre: Add is_collection and is_atom to edn::Value.
* Pre: Differentiate atoms from lookup-refs in value position.
Initially, I expected to accept arbitrary edn::Value instances in the
value position, and to differentiate in the transactor. However, the
implementation quickly became a two-stage parser, since we always
wanted to parse the resulting value position into some other known
thing using the tx-parser. To save calls into the parser and to allow
the parser to move forward with a smaller API surface, I push as much
of this parsing as possible into the initial parse.
* Pre: Modernize entities().
* Pre: Quote edn::Value::Text in Display.
* Review comment: Add and use edn::Value::into_atom.
* Review comment: Use skip(eof()) throughout.
* Review comment: VecDeque instead of Vec.
* Review comment: Part 0: Rename TempId to TempIdHandle.
* Review comment: Part 1: Differentiate internal and external tempids.
This breaks an abstraction boundary by pushing the Internal/External
split up to the Entity level in tx/ and tx-parser/. This just makes
it easier to explode Entity map notation instances into Entity
instances, taking an existing External tempid :db/id or generating a
new Internal tempid as appropriate. To do this without breaking the
abstraction boundary would require adding flexibility to the
transaction processor: we'd need to be able to turn Entity instances
into some internal enum and handle the two cases independently. It
wouldn't be too hard, but this reduces the combinatorial type
explosion.
2017-03-27 23:30:04 +00:00
[ :db / add 502 :test / unique_value \ " other \" ]
[ :db / add 503 :test / unique_identity - 10 ]
[ :db / add 504 :test / unique_identity - 20 ]
[ :db / add 505 :test / not_unique :test / keyword ]
2019-07-17 15:46:47 +00:00
[ :db / add 506 :test / not_unique :test / keyword ] ] "
) ;
Lookup refs, nested vector values, map notation. Fixes #180, fixes #183, fixes #284. (#382) r=rnewman
* Pre: Fix error in parser macros.
* Pre: Make test unwrapping more verbose.
* Pre: Make lookup refs be (lookup-ref a v) in the entity position.
This has the advantage of being explicit in all situations and
unambiguous at parse-time. This choice agrees with the Clojure
implementation but not with Datomic. Datomic treats [a v] as a lookup
ref, is ambiguous at parse-time, and is disambiguated in ways I do not
understand at transaction time. We mooted making lookup refs [[a v]]
and outlawing nested value vectors in transactions, but after
implementing that approach I decided it was better to handle lookup
refs at parse time and therefore outlawing nested value vectors is not
necessary.
* Handle lookup refs in the entity and value columns. Fixes #183.
* Pre 0a: Use a stack instead of into_iter.
* Pre 0b: Dedent.
* Pre 0c: Handle `e` after `v`.
This allows to use the original `e` while handling `v`.
* Explode value lists for :db.cardinality/many attributes. Fixes #284.
* Parse and accept map notation. Fixes #180.
* Pre: Modernize add() and retract() into one add_or_retract().
* Pre: Add is_collection and is_atom to edn::Value.
* Pre: Differentiate atoms from lookup-refs in value position.
Initially, I expected to accept arbitrary edn::Value instances in the
value position, and to differentiate in the transactor. However, the
implementation quickly became a two-stage parser, since we always
wanted to parse the resulting value position into some other known
thing using the tx-parser. To save calls into the parser and to allow
the parser to move forward with a smaller API surface, I push as much
of this parsing as possible into the initial parse.
* Pre: Modernize entities().
* Pre: Quote edn::Value::Text in Display.
* Review comment: Add and use edn::Value::into_atom.
* Review comment: Use skip(eof()) throughout.
* Review comment: VecDeque instead of Vec.
* Review comment: Part 0: Rename TempId to TempIdHandle.
* Review comment: Part 1: Differentiate internal and external tempids.
This breaks an abstraction boundary by pushing the Internal/External
split up to the Entity level in tx/ and tx-parser/. This just makes
it easier to explode Entity map notation instances into Entity
instances, taking an existing External tempid :db/id or generating a
new Internal tempid as appropriate. To do this without breaking the
abstraction boundary would require adding flexibility to the
transaction processor: we'd need to be able to turn Entity instances
into some internal enum and handle the two cases independently. It
wouldn't be too hard, but this reduces the combinatorial type
explosion.
2017-03-27 23:30:04 +00:00
// We can resolve lookup refs in the entity column, referring to the attribute as an entid or an ident.
assert_transact! ( conn , " [[:db/add (lookup-ref :test/unique_value \" test this \" ) :test/not_unique :test/keyword]
[ :db / add ( lookup - ref 111 \ " other \" ) :test/not_unique :test/keyword]
[ :db / add ( lookup - ref :test / unique_identity - 10 ) :test / not_unique :test / keyword ]
[ :db / add ( lookup - ref 222 - 20 ) :test / not_unique :test / keyword ] ] " );
2019-07-17 15:46:47 +00:00
assert_matches! (
conn . last_transaction ( ) ,
" [[501 :test/not_unique :test/keyword ?tx true]
Lookup refs, nested vector values, map notation. Fixes #180, fixes #183, fixes #284. (#382) r=rnewman
* Pre: Fix error in parser macros.
* Pre: Make test unwrapping more verbose.
* Pre: Make lookup refs be (lookup-ref a v) in the entity position.
This has the advantage of being explicit in all situations and
unambiguous at parse-time. This choice agrees with the Clojure
implementation but not with Datomic. Datomic treats [a v] as a lookup
ref, is ambiguous at parse-time, and is disambiguated in ways I do not
understand at transaction time. We mooted making lookup refs [[a v]]
and outlawing nested value vectors in transactions, but after
implementing that approach I decided it was better to handle lookup
refs at parse time and therefore outlawing nested value vectors is not
necessary.
* Handle lookup refs in the entity and value columns. Fixes #183.
* Pre 0a: Use a stack instead of into_iter.
* Pre 0b: Dedent.
* Pre 0c: Handle `e` after `v`.
This allows to use the original `e` while handling `v`.
* Explode value lists for :db.cardinality/many attributes. Fixes #284.
* Parse and accept map notation. Fixes #180.
* Pre: Modernize add() and retract() into one add_or_retract().
* Pre: Add is_collection and is_atom to edn::Value.
* Pre: Differentiate atoms from lookup-refs in value position.
Initially, I expected to accept arbitrary edn::Value instances in the
value position, and to differentiate in the transactor. However, the
implementation quickly became a two-stage parser, since we always
wanted to parse the resulting value position into some other known
thing using the tx-parser. To save calls into the parser and to allow
the parser to move forward with a smaller API surface, I push as much
of this parsing as possible into the initial parse.
* Pre: Modernize entities().
* Pre: Quote edn::Value::Text in Display.
* Review comment: Add and use edn::Value::into_atom.
* Review comment: Use skip(eof()) throughout.
* Review comment: VecDeque instead of Vec.
* Review comment: Part 0: Rename TempId to TempIdHandle.
* Review comment: Part 1: Differentiate internal and external tempids.
This breaks an abstraction boundary by pushing the Internal/External
split up to the Entity level in tx/ and tx-parser/. This just makes
it easier to explode Entity map notation instances into Entity
instances, taking an existing External tempid :db/id or generating a
new Internal tempid as appropriate. To do this without breaking the
abstraction boundary would require adding flexibility to the
transaction processor: we'd need to be able to turn Entity instances
into some internal enum and handle the two cases independently. It
wouldn't be too hard, but this reduces the combinatorial type
explosion.
2017-03-27 23:30:04 +00:00
[ 502 :test / not_unique :test / keyword ? tx true ]
[ 503 :test / not_unique :test / keyword ? tx true ]
[ 504 :test / not_unique :test / keyword ? tx true ]
2019-07-17 15:46:47 +00:00
[ ? tx :db / txInstant ? ms ? tx true ] ] "
) ;
Lookup refs, nested vector values, map notation. Fixes #180, fixes #183, fixes #284. (#382) r=rnewman
* Pre: Fix error in parser macros.
* Pre: Make test unwrapping more verbose.
* Pre: Make lookup refs be (lookup-ref a v) in the entity position.
This has the advantage of being explicit in all situations and
unambiguous at parse-time. This choice agrees with the Clojure
implementation but not with Datomic. Datomic treats [a v] as a lookup
ref, is ambiguous at parse-time, and is disambiguated in ways I do not
understand at transaction time. We mooted making lookup refs [[a v]]
and outlawing nested value vectors in transactions, but after
implementing that approach I decided it was better to handle lookup
refs at parse time and therefore outlawing nested value vectors is not
necessary.
* Handle lookup refs in the entity and value columns. Fixes #183.
* Pre 0a: Use a stack instead of into_iter.
* Pre 0b: Dedent.
* Pre 0c: Handle `e` after `v`.
This allows to use the original `e` while handling `v`.
* Explode value lists for :db.cardinality/many attributes. Fixes #284.
* Parse and accept map notation. Fixes #180.
* Pre: Modernize add() and retract() into one add_or_retract().
* Pre: Add is_collection and is_atom to edn::Value.
* Pre: Differentiate atoms from lookup-refs in value position.
Initially, I expected to accept arbitrary edn::Value instances in the
value position, and to differentiate in the transactor. However, the
implementation quickly became a two-stage parser, since we always
wanted to parse the resulting value position into some other known
thing using the tx-parser. To save calls into the parser and to allow
the parser to move forward with a smaller API surface, I push as much
of this parsing as possible into the initial parse.
* Pre: Modernize entities().
* Pre: Quote edn::Value::Text in Display.
* Review comment: Add and use edn::Value::into_atom.
* Review comment: Use skip(eof()) throughout.
* Review comment: VecDeque instead of Vec.
* Review comment: Part 0: Rename TempId to TempIdHandle.
* Review comment: Part 1: Differentiate internal and external tempids.
This breaks an abstraction boundary by pushing the Internal/External
split up to the Entity level in tx/ and tx-parser/. This just makes
it easier to explode Entity map notation instances into Entity
instances, taking an existing External tempid :db/id or generating a
new Internal tempid as appropriate. To do this without breaking the
abstraction boundary would require adding flexibility to the
transaction processor: we'd need to be able to turn Entity instances
into some internal enum and handle the two cases independently. It
wouldn't be too hard, but this reduces the combinatorial type
explosion.
2017-03-27 23:30:04 +00:00
// We cannot resolve lookup refs that aren't :db/unique.
assert_transact! ( conn ,
" [[:db/add (lookup-ref :test/not_unique :test/keyword) :test/not_unique :test/keyword]] " ,
2018-05-15 07:43:07 +00:00
Err ( " not yet implemented: Cannot resolve (lookup-ref 333 Keyword(Keyword(NamespaceableName { namespace: Some( \" test \" ), name: \" keyword \" }))) with attribute that is not :db/unique " ) ) ;
Lookup refs, nested vector values, map notation. Fixes #180, fixes #183, fixes #284. (#382) r=rnewman
* Pre: Fix error in parser macros.
* Pre: Make test unwrapping more verbose.
* Pre: Make lookup refs be (lookup-ref a v) in the entity position.
This has the advantage of being explicit in all situations and
unambiguous at parse-time. This choice agrees with the Clojure
implementation but not with Datomic. Datomic treats [a v] as a lookup
ref, is ambiguous at parse-time, and is disambiguated in ways I do not
understand at transaction time. We mooted making lookup refs [[a v]]
and outlawing nested value vectors in transactions, but after
implementing that approach I decided it was better to handle lookup
refs at parse time and therefore outlawing nested value vectors is not
necessary.
* Handle lookup refs in the entity and value columns. Fixes #183.
* Pre 0a: Use a stack instead of into_iter.
* Pre 0b: Dedent.
* Pre 0c: Handle `e` after `v`.
This allows to use the original `e` while handling `v`.
* Explode value lists for :db.cardinality/many attributes. Fixes #284.
* Parse and accept map notation. Fixes #180.
* Pre: Modernize add() and retract() into one add_or_retract().
* Pre: Add is_collection and is_atom to edn::Value.
* Pre: Differentiate atoms from lookup-refs in value position.
Initially, I expected to accept arbitrary edn::Value instances in the
value position, and to differentiate in the transactor. However, the
implementation quickly became a two-stage parser, since we always
wanted to parse the resulting value position into some other known
thing using the tx-parser. To save calls into the parser and to allow
the parser to move forward with a smaller API surface, I push as much
of this parsing as possible into the initial parse.
* Pre: Modernize entities().
* Pre: Quote edn::Value::Text in Display.
* Review comment: Add and use edn::Value::into_atom.
* Review comment: Use skip(eof()) throughout.
* Review comment: VecDeque instead of Vec.
* Review comment: Part 0: Rename TempId to TempIdHandle.
* Review comment: Part 1: Differentiate internal and external tempids.
This breaks an abstraction boundary by pushing the Internal/External
split up to the Entity level in tx/ and tx-parser/. This just makes
it easier to explode Entity map notation instances into Entity
instances, taking an existing External tempid :db/id or generating a
new Internal tempid as appropriate. To do this without breaking the
abstraction boundary would require adding flexibility to the
transaction processor: we'd need to be able to turn Entity instances
into some internal enum and handle the two cases independently. It
wouldn't be too hard, but this reduces the combinatorial type
explosion.
2017-03-27 23:30:04 +00:00
// We type check the lookup ref's value against the lookup ref's attribute.
assert_transact! ( conn ,
" [[:db/add (lookup-ref :test/unique_value :test/not_a_string) :test/not_unique :test/keyword]] " ,
2018-05-15 07:43:07 +00:00
Err ( " value \' :test/not_a_string \' is not the expected Mentat value type String " ) ) ;
Lookup refs, nested vector values, map notation. Fixes #180, fixes #183, fixes #284. (#382) r=rnewman
* Pre: Fix error in parser macros.
* Pre: Make test unwrapping more verbose.
* Pre: Make lookup refs be (lookup-ref a v) in the entity position.
This has the advantage of being explicit in all situations and
unambiguous at parse-time. This choice agrees with the Clojure
implementation but not with Datomic. Datomic treats [a v] as a lookup
ref, is ambiguous at parse-time, and is disambiguated in ways I do not
understand at transaction time. We mooted making lookup refs [[a v]]
and outlawing nested value vectors in transactions, but after
implementing that approach I decided it was better to handle lookup
refs at parse time and therefore outlawing nested value vectors is not
necessary.
* Handle lookup refs in the entity and value columns. Fixes #183.
* Pre 0a: Use a stack instead of into_iter.
* Pre 0b: Dedent.
* Pre 0c: Handle `e` after `v`.
This allows to use the original `e` while handling `v`.
* Explode value lists for :db.cardinality/many attributes. Fixes #284.
* Parse and accept map notation. Fixes #180.
* Pre: Modernize add() and retract() into one add_or_retract().
* Pre: Add is_collection and is_atom to edn::Value.
* Pre: Differentiate atoms from lookup-refs in value position.
Initially, I expected to accept arbitrary edn::Value instances in the
value position, and to differentiate in the transactor. However, the
implementation quickly became a two-stage parser, since we always
wanted to parse the resulting value position into some other known
thing using the tx-parser. To save calls into the parser and to allow
the parser to move forward with a smaller API surface, I push as much
of this parsing as possible into the initial parse.
* Pre: Modernize entities().
* Pre: Quote edn::Value::Text in Display.
* Review comment: Add and use edn::Value::into_atom.
* Review comment: Use skip(eof()) throughout.
* Review comment: VecDeque instead of Vec.
* Review comment: Part 0: Rename TempId to TempIdHandle.
* Review comment: Part 1: Differentiate internal and external tempids.
This breaks an abstraction boundary by pushing the Internal/External
split up to the Entity level in tx/ and tx-parser/. This just makes
it easier to explode Entity map notation instances into Entity
instances, taking an existing External tempid :db/id or generating a
new Internal tempid as appropriate. To do this without breaking the
abstraction boundary would require adding flexibility to the
transaction processor: we'd need to be able to turn Entity instances
into some internal enum and handle the two cases independently. It
wouldn't be too hard, but this reduces the combinatorial type
explosion.
2017-03-27 23:30:04 +00:00
// Each lookup ref in the entity column must resolve
assert_transact! ( conn ,
" [[:db/add (lookup-ref :test/unique_value \" unmatched string value \" ) :test/not_unique :test/keyword]] " ,
Err ( " no entid found for ident: couldn \' t lookup [a v]: (111, String( \" unmatched string value \" )) " ) ) ;
}
#[ test ]
fn test_lookup_refs_value_column ( ) {
let mut conn = TestConn ::default ( ) ;
// Start by installing a few attributes.
2019-07-17 15:46:47 +00:00
assert_transact! (
conn ,
" [[:db/add 111 :db/ident :test/unique_value]
Lookup refs, nested vector values, map notation. Fixes #180, fixes #183, fixes #284. (#382) r=rnewman
* Pre: Fix error in parser macros.
* Pre: Make test unwrapping more verbose.
* Pre: Make lookup refs be (lookup-ref a v) in the entity position.
This has the advantage of being explicit in all situations and
unambiguous at parse-time. This choice agrees with the Clojure
implementation but not with Datomic. Datomic treats [a v] as a lookup
ref, is ambiguous at parse-time, and is disambiguated in ways I do not
understand at transaction time. We mooted making lookup refs [[a v]]
and outlawing nested value vectors in transactions, but after
implementing that approach I decided it was better to handle lookup
refs at parse time and therefore outlawing nested value vectors is not
necessary.
* Handle lookup refs in the entity and value columns. Fixes #183.
* Pre 0a: Use a stack instead of into_iter.
* Pre 0b: Dedent.
* Pre 0c: Handle `e` after `v`.
This allows to use the original `e` while handling `v`.
* Explode value lists for :db.cardinality/many attributes. Fixes #284.
* Parse and accept map notation. Fixes #180.
* Pre: Modernize add() and retract() into one add_or_retract().
* Pre: Add is_collection and is_atom to edn::Value.
* Pre: Differentiate atoms from lookup-refs in value position.
Initially, I expected to accept arbitrary edn::Value instances in the
value position, and to differentiate in the transactor. However, the
implementation quickly became a two-stage parser, since we always
wanted to parse the resulting value position into some other known
thing using the tx-parser. To save calls into the parser and to allow
the parser to move forward with a smaller API surface, I push as much
of this parsing as possible into the initial parse.
* Pre: Modernize entities().
* Pre: Quote edn::Value::Text in Display.
* Review comment: Add and use edn::Value::into_atom.
* Review comment: Use skip(eof()) throughout.
* Review comment: VecDeque instead of Vec.
* Review comment: Part 0: Rename TempId to TempIdHandle.
* Review comment: Part 1: Differentiate internal and external tempids.
This breaks an abstraction boundary by pushing the Internal/External
split up to the Entity level in tx/ and tx-parser/. This just makes
it easier to explode Entity map notation instances into Entity
instances, taking an existing External tempid :db/id or generating a
new Internal tempid as appropriate. To do this without breaking the
abstraction boundary would require adding flexibility to the
transaction processor: we'd need to be able to turn Entity instances
into some internal enum and handle the two cases independently. It
wouldn't be too hard, but this reduces the combinatorial type
explosion.
2017-03-27 23:30:04 +00:00
[ :db / add 111 :db / valueType :db . type / string ]
[ :db / add 111 :db / unique :db . unique / value ]
[ :db / add 111 :db / index true ]
[ :db / add 222 :db / ident :test / unique_identity ]
[ :db / add 222 :db / valueType :db . type / long ]
[ :db / add 222 :db / unique :db . unique / identity ]
[ :db / add 222 :db / index true ]
[ :db / add 333 :db / ident :test / not_unique ]
[ :db / add 333 :db / cardinality :db . cardinality / one ]
[ :db / add 333 :db / valueType :db . type / keyword ]
[ :db / add 333 :db / index true ]
[ :db / add 444 :db / ident :test / ref ]
[ :db / add 444 :db / valueType :db . type / ref ]
[ :db / add 444 :db / unique :db . unique / identity ]
2019-07-17 15:46:47 +00:00
[ :db / add 444 :db / index true ] ] "
) ;
Lookup refs, nested vector values, map notation. Fixes #180, fixes #183, fixes #284. (#382) r=rnewman
* Pre: Fix error in parser macros.
* Pre: Make test unwrapping more verbose.
* Pre: Make lookup refs be (lookup-ref a v) in the entity position.
This has the advantage of being explicit in all situations and
unambiguous at parse-time. This choice agrees with the Clojure
implementation but not with Datomic. Datomic treats [a v] as a lookup
ref, is ambiguous at parse-time, and is disambiguated in ways I do not
understand at transaction time. We mooted making lookup refs [[a v]]
and outlawing nested value vectors in transactions, but after
implementing that approach I decided it was better to handle lookup
refs at parse time and therefore outlawing nested value vectors is not
necessary.
* Handle lookup refs in the entity and value columns. Fixes #183.
* Pre 0a: Use a stack instead of into_iter.
* Pre 0b: Dedent.
* Pre 0c: Handle `e` after `v`.
This allows to use the original `e` while handling `v`.
* Explode value lists for :db.cardinality/many attributes. Fixes #284.
* Parse and accept map notation. Fixes #180.
* Pre: Modernize add() and retract() into one add_or_retract().
* Pre: Add is_collection and is_atom to edn::Value.
* Pre: Differentiate atoms from lookup-refs in value position.
Initially, I expected to accept arbitrary edn::Value instances in the
value position, and to differentiate in the transactor. However, the
implementation quickly became a two-stage parser, since we always
wanted to parse the resulting value position into some other known
thing using the tx-parser. To save calls into the parser and to allow
the parser to move forward with a smaller API surface, I push as much
of this parsing as possible into the initial parse.
* Pre: Modernize entities().
* Pre: Quote edn::Value::Text in Display.
* Review comment: Add and use edn::Value::into_atom.
* Review comment: Use skip(eof()) throughout.
* Review comment: VecDeque instead of Vec.
* Review comment: Part 0: Rename TempId to TempIdHandle.
* Review comment: Part 1: Differentiate internal and external tempids.
This breaks an abstraction boundary by pushing the Internal/External
split up to the Entity level in tx/ and tx-parser/. This just makes
it easier to explode Entity map notation instances into Entity
instances, taking an existing External tempid :db/id or generating a
new Internal tempid as appropriate. To do this without breaking the
abstraction boundary would require adding flexibility to the
transaction processor: we'd need to be able to turn Entity instances
into some internal enum and handle the two cases independently. It
wouldn't be too hard, but this reduces the combinatorial type
explosion.
2017-03-27 23:30:04 +00:00
// And a few datoms to match against.
2019-07-17 15:46:47 +00:00
assert_transact! (
conn ,
" [[:db/add 501 :test/unique_value \" test this \" ]
Lookup refs, nested vector values, map notation. Fixes #180, fixes #183, fixes #284. (#382) r=rnewman
* Pre: Fix error in parser macros.
* Pre: Make test unwrapping more verbose.
* Pre: Make lookup refs be (lookup-ref a v) in the entity position.
This has the advantage of being explicit in all situations and
unambiguous at parse-time. This choice agrees with the Clojure
implementation but not with Datomic. Datomic treats [a v] as a lookup
ref, is ambiguous at parse-time, and is disambiguated in ways I do not
understand at transaction time. We mooted making lookup refs [[a v]]
and outlawing nested value vectors in transactions, but after
implementing that approach I decided it was better to handle lookup
refs at parse time and therefore outlawing nested value vectors is not
necessary.
* Handle lookup refs in the entity and value columns. Fixes #183.
* Pre 0a: Use a stack instead of into_iter.
* Pre 0b: Dedent.
* Pre 0c: Handle `e` after `v`.
This allows to use the original `e` while handling `v`.
* Explode value lists for :db.cardinality/many attributes. Fixes #284.
* Parse and accept map notation. Fixes #180.
* Pre: Modernize add() and retract() into one add_or_retract().
* Pre: Add is_collection and is_atom to edn::Value.
* Pre: Differentiate atoms from lookup-refs in value position.
Initially, I expected to accept arbitrary edn::Value instances in the
value position, and to differentiate in the transactor. However, the
implementation quickly became a two-stage parser, since we always
wanted to parse the resulting value position into some other known
thing using the tx-parser. To save calls into the parser and to allow
the parser to move forward with a smaller API surface, I push as much
of this parsing as possible into the initial parse.
* Pre: Modernize entities().
* Pre: Quote edn::Value::Text in Display.
* Review comment: Add and use edn::Value::into_atom.
* Review comment: Use skip(eof()) throughout.
* Review comment: VecDeque instead of Vec.
* Review comment: Part 0: Rename TempId to TempIdHandle.
* Review comment: Part 1: Differentiate internal and external tempids.
This breaks an abstraction boundary by pushing the Internal/External
split up to the Entity level in tx/ and tx-parser/. This just makes
it easier to explode Entity map notation instances into Entity
instances, taking an existing External tempid :db/id or generating a
new Internal tempid as appropriate. To do this without breaking the
abstraction boundary would require adding flexibility to the
transaction processor: we'd need to be able to turn Entity instances
into some internal enum and handle the two cases independently. It
wouldn't be too hard, but this reduces the combinatorial type
explosion.
2017-03-27 23:30:04 +00:00
[ :db / add 502 :test / unique_value \ " other \" ]
[ :db / add 503 :test / unique_identity - 10 ]
[ :db / add 504 :test / unique_identity - 20 ]
[ :db / add 505 :test / not_unique :test / keyword ]
2019-07-17 15:46:47 +00:00
[ :db / add 506 :test / not_unique :test / keyword ] ] "
) ;
Lookup refs, nested vector values, map notation. Fixes #180, fixes #183, fixes #284. (#382) r=rnewman
* Pre: Fix error in parser macros.
* Pre: Make test unwrapping more verbose.
* Pre: Make lookup refs be (lookup-ref a v) in the entity position.
This has the advantage of being explicit in all situations and
unambiguous at parse-time. This choice agrees with the Clojure
implementation but not with Datomic. Datomic treats [a v] as a lookup
ref, is ambiguous at parse-time, and is disambiguated in ways I do not
understand at transaction time. We mooted making lookup refs [[a v]]
and outlawing nested value vectors in transactions, but after
implementing that approach I decided it was better to handle lookup
refs at parse time and therefore outlawing nested value vectors is not
necessary.
* Handle lookup refs in the entity and value columns. Fixes #183.
* Pre 0a: Use a stack instead of into_iter.
* Pre 0b: Dedent.
* Pre 0c: Handle `e` after `v`.
This allows to use the original `e` while handling `v`.
* Explode value lists for :db.cardinality/many attributes. Fixes #284.
* Parse and accept map notation. Fixes #180.
* Pre: Modernize add() and retract() into one add_or_retract().
* Pre: Add is_collection and is_atom to edn::Value.
* Pre: Differentiate atoms from lookup-refs in value position.
Initially, I expected to accept arbitrary edn::Value instances in the
value position, and to differentiate in the transactor. However, the
implementation quickly became a two-stage parser, since we always
wanted to parse the resulting value position into some other known
thing using the tx-parser. To save calls into the parser and to allow
the parser to move forward with a smaller API surface, I push as much
of this parsing as possible into the initial parse.
* Pre: Modernize entities().
* Pre: Quote edn::Value::Text in Display.
* Review comment: Add and use edn::Value::into_atom.
* Review comment: Use skip(eof()) throughout.
* Review comment: VecDeque instead of Vec.
* Review comment: Part 0: Rename TempId to TempIdHandle.
* Review comment: Part 1: Differentiate internal and external tempids.
This breaks an abstraction boundary by pushing the Internal/External
split up to the Entity level in tx/ and tx-parser/. This just makes
it easier to explode Entity map notation instances into Entity
instances, taking an existing External tempid :db/id or generating a
new Internal tempid as appropriate. To do this without breaking the
abstraction boundary would require adding flexibility to the
transaction processor: we'd need to be able to turn Entity instances
into some internal enum and handle the two cases independently. It
wouldn't be too hard, but this reduces the combinatorial type
explosion.
2017-03-27 23:30:04 +00:00
// We can resolve lookup refs in the entity column, referring to the attribute as an entid or an ident.
2019-07-17 15:46:47 +00:00
assert_transact! (
conn ,
" [[:db/add 601 :test/ref (lookup-ref :test/unique_value \" test this \" )]
Lookup refs, nested vector values, map notation. Fixes #180, fixes #183, fixes #284. (#382) r=rnewman
* Pre: Fix error in parser macros.
* Pre: Make test unwrapping more verbose.
* Pre: Make lookup refs be (lookup-ref a v) in the entity position.
This has the advantage of being explicit in all situations and
unambiguous at parse-time. This choice agrees with the Clojure
implementation but not with Datomic. Datomic treats [a v] as a lookup
ref, is ambiguous at parse-time, and is disambiguated in ways I do not
understand at transaction time. We mooted making lookup refs [[a v]]
and outlawing nested value vectors in transactions, but after
implementing that approach I decided it was better to handle lookup
refs at parse time and therefore outlawing nested value vectors is not
necessary.
* Handle lookup refs in the entity and value columns. Fixes #183.
* Pre 0a: Use a stack instead of into_iter.
* Pre 0b: Dedent.
* Pre 0c: Handle `e` after `v`.
This allows to use the original `e` while handling `v`.
* Explode value lists for :db.cardinality/many attributes. Fixes #284.
* Parse and accept map notation. Fixes #180.
* Pre: Modernize add() and retract() into one add_or_retract().
* Pre: Add is_collection and is_atom to edn::Value.
* Pre: Differentiate atoms from lookup-refs in value position.
Initially, I expected to accept arbitrary edn::Value instances in the
value position, and to differentiate in the transactor. However, the
implementation quickly became a two-stage parser, since we always
wanted to parse the resulting value position into some other known
thing using the tx-parser. To save calls into the parser and to allow
the parser to move forward with a smaller API surface, I push as much
of this parsing as possible into the initial parse.
* Pre: Modernize entities().
* Pre: Quote edn::Value::Text in Display.
* Review comment: Add and use edn::Value::into_atom.
* Review comment: Use skip(eof()) throughout.
* Review comment: VecDeque instead of Vec.
* Review comment: Part 0: Rename TempId to TempIdHandle.
* Review comment: Part 1: Differentiate internal and external tempids.
This breaks an abstraction boundary by pushing the Internal/External
split up to the Entity level in tx/ and tx-parser/. This just makes
it easier to explode Entity map notation instances into Entity
instances, taking an existing External tempid :db/id or generating a
new Internal tempid as appropriate. To do this without breaking the
abstraction boundary would require adding flexibility to the
transaction processor: we'd need to be able to turn Entity instances
into some internal enum and handle the two cases independently. It
wouldn't be too hard, but this reduces the combinatorial type
explosion.
2017-03-27 23:30:04 +00:00
[ :db / add 602 :test / ref ( lookup - ref 111 \ " other \" )]
[ :db / add 603 :test / ref ( lookup - ref :test / unique_identity - 10 ) ]
2019-07-17 15:46:47 +00:00
[ :db / add 604 :test / ref ( lookup - ref 222 - 20 ) ] ] "
) ;
assert_matches! (
conn . last_transaction ( ) ,
" [[601 :test/ref 501 ?tx true]
Lookup refs, nested vector values, map notation. Fixes #180, fixes #183, fixes #284. (#382) r=rnewman
* Pre: Fix error in parser macros.
* Pre: Make test unwrapping more verbose.
* Pre: Make lookup refs be (lookup-ref a v) in the entity position.
This has the advantage of being explicit in all situations and
unambiguous at parse-time. This choice agrees with the Clojure
implementation but not with Datomic. Datomic treats [a v] as a lookup
ref, is ambiguous at parse-time, and is disambiguated in ways I do not
understand at transaction time. We mooted making lookup refs [[a v]]
and outlawing nested value vectors in transactions, but after
implementing that approach I decided it was better to handle lookup
refs at parse time and therefore outlawing nested value vectors is not
necessary.
* Handle lookup refs in the entity and value columns. Fixes #183.
* Pre 0a: Use a stack instead of into_iter.
* Pre 0b: Dedent.
* Pre 0c: Handle `e` after `v`.
This allows to use the original `e` while handling `v`.
* Explode value lists for :db.cardinality/many attributes. Fixes #284.
* Parse and accept map notation. Fixes #180.
* Pre: Modernize add() and retract() into one add_or_retract().
* Pre: Add is_collection and is_atom to edn::Value.
* Pre: Differentiate atoms from lookup-refs in value position.
Initially, I expected to accept arbitrary edn::Value instances in the
value position, and to differentiate in the transactor. However, the
implementation quickly became a two-stage parser, since we always
wanted to parse the resulting value position into some other known
thing using the tx-parser. To save calls into the parser and to allow
the parser to move forward with a smaller API surface, I push as much
of this parsing as possible into the initial parse.
* Pre: Modernize entities().
* Pre: Quote edn::Value::Text in Display.
* Review comment: Add and use edn::Value::into_atom.
* Review comment: Use skip(eof()) throughout.
* Review comment: VecDeque instead of Vec.
* Review comment: Part 0: Rename TempId to TempIdHandle.
* Review comment: Part 1: Differentiate internal and external tempids.
This breaks an abstraction boundary by pushing the Internal/External
split up to the Entity level in tx/ and tx-parser/. This just makes
it easier to explode Entity map notation instances into Entity
instances, taking an existing External tempid :db/id or generating a
new Internal tempid as appropriate. To do this without breaking the
abstraction boundary would require adding flexibility to the
transaction processor: we'd need to be able to turn Entity instances
into some internal enum and handle the two cases independently. It
wouldn't be too hard, but this reduces the combinatorial type
explosion.
2017-03-27 23:30:04 +00:00
[ 602 :test / ref 502 ? tx true ]
[ 603 :test / ref 503 ? tx true ]
[ 604 :test / ref 504 ? tx true ]
2019-07-17 15:46:47 +00:00
[ ? tx :db / txInstant ? ms ? tx true ] ] "
) ;
Lookup refs, nested vector values, map notation. Fixes #180, fixes #183, fixes #284. (#382) r=rnewman
* Pre: Fix error in parser macros.
* Pre: Make test unwrapping more verbose.
* Pre: Make lookup refs be (lookup-ref a v) in the entity position.
This has the advantage of being explicit in all situations and
unambiguous at parse-time. This choice agrees with the Clojure
implementation but not with Datomic. Datomic treats [a v] as a lookup
ref, is ambiguous at parse-time, and is disambiguated in ways I do not
understand at transaction time. We mooted making lookup refs [[a v]]
and outlawing nested value vectors in transactions, but after
implementing that approach I decided it was better to handle lookup
refs at parse time and therefore outlawing nested value vectors is not
necessary.
* Handle lookup refs in the entity and value columns. Fixes #183.
* Pre 0a: Use a stack instead of into_iter.
* Pre 0b: Dedent.
* Pre 0c: Handle `e` after `v`.
This allows to use the original `e` while handling `v`.
* Explode value lists for :db.cardinality/many attributes. Fixes #284.
* Parse and accept map notation. Fixes #180.
* Pre: Modernize add() and retract() into one add_or_retract().
* Pre: Add is_collection and is_atom to edn::Value.
* Pre: Differentiate atoms from lookup-refs in value position.
Initially, I expected to accept arbitrary edn::Value instances in the
value position, and to differentiate in the transactor. However, the
implementation quickly became a two-stage parser, since we always
wanted to parse the resulting value position into some other known
thing using the tx-parser. To save calls into the parser and to allow
the parser to move forward with a smaller API surface, I push as much
of this parsing as possible into the initial parse.
* Pre: Modernize entities().
* Pre: Quote edn::Value::Text in Display.
* Review comment: Add and use edn::Value::into_atom.
* Review comment: Use skip(eof()) throughout.
* Review comment: VecDeque instead of Vec.
* Review comment: Part 0: Rename TempId to TempIdHandle.
* Review comment: Part 1: Differentiate internal and external tempids.
This breaks an abstraction boundary by pushing the Internal/External
split up to the Entity level in tx/ and tx-parser/. This just makes
it easier to explode Entity map notation instances into Entity
instances, taking an existing External tempid :db/id or generating a
new Internal tempid as appropriate. To do this without breaking the
abstraction boundary would require adding flexibility to the
transaction processor: we'd need to be able to turn Entity instances
into some internal enum and handle the two cases independently. It
wouldn't be too hard, but this reduces the combinatorial type
explosion.
2017-03-27 23:30:04 +00:00
// We cannot resolve lookup refs for attributes that aren't :db/ref.
assert_transact! ( conn ,
" [[:db/add \" t \" :test/not_unique (lookup-ref :test/unique_value \" test this \" )]] " ,
Err ( " not yet implemented: Cannot resolve value lookup ref for attribute 333 that is not :db/valueType :db.type/ref " ) ) ;
// If a value column lookup ref resolves, we can upsert against it. Here, the lookup ref
// resolves to 501, which upserts "t" to 601.
2019-07-17 15:46:47 +00:00
assert_transact! (
conn ,
" [[:db/add \" t \" :test/ref (lookup-ref :test/unique_value \" test this \" )]
[ :db / add \ " t \" :test/not_unique :test/keyword]] "
) ;
assert_matches! (
conn . last_transaction ( ) ,
" [[601 :test/not_unique :test/keyword ?tx true]
[ ? tx :db / txInstant ? ms ? tx true ] ] "
) ;
Lookup refs, nested vector values, map notation. Fixes #180, fixes #183, fixes #284. (#382) r=rnewman
* Pre: Fix error in parser macros.
* Pre: Make test unwrapping more verbose.
* Pre: Make lookup refs be (lookup-ref a v) in the entity position.
This has the advantage of being explicit in all situations and
unambiguous at parse-time. This choice agrees with the Clojure
implementation but not with Datomic. Datomic treats [a v] as a lookup
ref, is ambiguous at parse-time, and is disambiguated in ways I do not
understand at transaction time. We mooted making lookup refs [[a v]]
and outlawing nested value vectors in transactions, but after
implementing that approach I decided it was better to handle lookup
refs at parse time and therefore outlawing nested value vectors is not
necessary.
* Handle lookup refs in the entity and value columns. Fixes #183.
* Pre 0a: Use a stack instead of into_iter.
* Pre 0b: Dedent.
* Pre 0c: Handle `e` after `v`.
This allows to use the original `e` while handling `v`.
* Explode value lists for :db.cardinality/many attributes. Fixes #284.
* Parse and accept map notation. Fixes #180.
* Pre: Modernize add() and retract() into one add_or_retract().
* Pre: Add is_collection and is_atom to edn::Value.
* Pre: Differentiate atoms from lookup-refs in value position.
Initially, I expected to accept arbitrary edn::Value instances in the
value position, and to differentiate in the transactor. However, the
implementation quickly became a two-stage parser, since we always
wanted to parse the resulting value position into some other known
thing using the tx-parser. To save calls into the parser and to allow
the parser to move forward with a smaller API surface, I push as much
of this parsing as possible into the initial parse.
* Pre: Modernize entities().
* Pre: Quote edn::Value::Text in Display.
* Review comment: Add and use edn::Value::into_atom.
* Review comment: Use skip(eof()) throughout.
* Review comment: VecDeque instead of Vec.
* Review comment: Part 0: Rename TempId to TempIdHandle.
* Review comment: Part 1: Differentiate internal and external tempids.
This breaks an abstraction boundary by pushing the Internal/External
split up to the Entity level in tx/ and tx-parser/. This just makes
it easier to explode Entity map notation instances into Entity
instances, taking an existing External tempid :db/id or generating a
new Internal tempid as appropriate. To do this without breaking the
abstraction boundary would require adding flexibility to the
transaction processor: we'd need to be able to turn Entity instances
into some internal enum and handle the two cases independently. It
wouldn't be too hard, but this reduces the combinatorial type
explosion.
2017-03-27 23:30:04 +00:00
// Each lookup ref in the value column must resolve
assert_transact! ( conn ,
" [[:db/add \" t \" :test/ref (lookup-ref :test/unique_value \" unmatched string value \" )]] " ,
Err ( " no entid found for ident: couldn \' t lookup [a v]: (111, String( \" unmatched string value \" )) " ) ) ;
}
#[ test ]
fn test_explode_value_lists ( ) {
let mut conn = TestConn ::default ( ) ;
// Start by installing a few attributes.
2019-07-17 15:46:47 +00:00
assert_transact! (
conn ,
" [[:db/add 111 :db/ident :test/many]
Lookup refs, nested vector values, map notation. Fixes #180, fixes #183, fixes #284. (#382) r=rnewman
* Pre: Fix error in parser macros.
* Pre: Make test unwrapping more verbose.
* Pre: Make lookup refs be (lookup-ref a v) in the entity position.
This has the advantage of being explicit in all situations and
unambiguous at parse-time. This choice agrees with the Clojure
implementation but not with Datomic. Datomic treats [a v] as a lookup
ref, is ambiguous at parse-time, and is disambiguated in ways I do not
understand at transaction time. We mooted making lookup refs [[a v]]
and outlawing nested value vectors in transactions, but after
implementing that approach I decided it was better to handle lookup
refs at parse time and therefore outlawing nested value vectors is not
necessary.
* Handle lookup refs in the entity and value columns. Fixes #183.
* Pre 0a: Use a stack instead of into_iter.
* Pre 0b: Dedent.
* Pre 0c: Handle `e` after `v`.
This allows to use the original `e` while handling `v`.
* Explode value lists for :db.cardinality/many attributes. Fixes #284.
* Parse and accept map notation. Fixes #180.
* Pre: Modernize add() and retract() into one add_or_retract().
* Pre: Add is_collection and is_atom to edn::Value.
* Pre: Differentiate atoms from lookup-refs in value position.
Initially, I expected to accept arbitrary edn::Value instances in the
value position, and to differentiate in the transactor. However, the
implementation quickly became a two-stage parser, since we always
wanted to parse the resulting value position into some other known
thing using the tx-parser. To save calls into the parser and to allow
the parser to move forward with a smaller API surface, I push as much
of this parsing as possible into the initial parse.
* Pre: Modernize entities().
* Pre: Quote edn::Value::Text in Display.
* Review comment: Add and use edn::Value::into_atom.
* Review comment: Use skip(eof()) throughout.
* Review comment: VecDeque instead of Vec.
* Review comment: Part 0: Rename TempId to TempIdHandle.
* Review comment: Part 1: Differentiate internal and external tempids.
This breaks an abstraction boundary by pushing the Internal/External
split up to the Entity level in tx/ and tx-parser/. This just makes
it easier to explode Entity map notation instances into Entity
instances, taking an existing External tempid :db/id or generating a
new Internal tempid as appropriate. To do this without breaking the
abstraction boundary would require adding flexibility to the
transaction processor: we'd need to be able to turn Entity instances
into some internal enum and handle the two cases independently. It
wouldn't be too hard, but this reduces the combinatorial type
explosion.
2017-03-27 23:30:04 +00:00
[ :db / add 111 :db / valueType :db . type / long ]
[ :db / add 111 :db / cardinality :db . cardinality / many ]
[ :db / add 222 :db / ident :test / one ]
[ :db / add 222 :db / valueType :db . type / long ]
2019-07-17 15:46:47 +00:00
[ :db / add 222 :db / cardinality :db . cardinality / one ] ] "
) ;
Lookup refs, nested vector values, map notation. Fixes #180, fixes #183, fixes #284. (#382) r=rnewman
* Pre: Fix error in parser macros.
* Pre: Make test unwrapping more verbose.
* Pre: Make lookup refs be (lookup-ref a v) in the entity position.
This has the advantage of being explicit in all situations and
unambiguous at parse-time. This choice agrees with the Clojure
implementation but not with Datomic. Datomic treats [a v] as a lookup
ref, is ambiguous at parse-time, and is disambiguated in ways I do not
understand at transaction time. We mooted making lookup refs [[a v]]
and outlawing nested value vectors in transactions, but after
implementing that approach I decided it was better to handle lookup
refs at parse time and therefore outlawing nested value vectors is not
necessary.
* Handle lookup refs in the entity and value columns. Fixes #183.
* Pre 0a: Use a stack instead of into_iter.
* Pre 0b: Dedent.
* Pre 0c: Handle `e` after `v`.
This allows to use the original `e` while handling `v`.
* Explode value lists for :db.cardinality/many attributes. Fixes #284.
* Parse and accept map notation. Fixes #180.
* Pre: Modernize add() and retract() into one add_or_retract().
* Pre: Add is_collection and is_atom to edn::Value.
* Pre: Differentiate atoms from lookup-refs in value position.
Initially, I expected to accept arbitrary edn::Value instances in the
value position, and to differentiate in the transactor. However, the
implementation quickly became a two-stage parser, since we always
wanted to parse the resulting value position into some other known
thing using the tx-parser. To save calls into the parser and to allow
the parser to move forward with a smaller API surface, I push as much
of this parsing as possible into the initial parse.
* Pre: Modernize entities().
* Pre: Quote edn::Value::Text in Display.
* Review comment: Add and use edn::Value::into_atom.
* Review comment: Use skip(eof()) throughout.
* Review comment: VecDeque instead of Vec.
* Review comment: Part 0: Rename TempId to TempIdHandle.
* Review comment: Part 1: Differentiate internal and external tempids.
This breaks an abstraction boundary by pushing the Internal/External
split up to the Entity level in tx/ and tx-parser/. This just makes
it easier to explode Entity map notation instances into Entity
instances, taking an existing External tempid :db/id or generating a
new Internal tempid as appropriate. To do this without breaking the
abstraction boundary would require adding flexibility to the
transaction processor: we'd need to be able to turn Entity instances
into some internal enum and handle the two cases independently. It
wouldn't be too hard, but this reduces the combinatorial type
explosion.
2017-03-27 23:30:04 +00:00
// Check that we can explode vectors for :db.cardinality/many attributes.
2019-07-17 15:46:47 +00:00
assert_transact! (
conn ,
" [[:db/add 501 :test/many [1]]
Lookup refs, nested vector values, map notation. Fixes #180, fixes #183, fixes #284. (#382) r=rnewman
* Pre: Fix error in parser macros.
* Pre: Make test unwrapping more verbose.
* Pre: Make lookup refs be (lookup-ref a v) in the entity position.
This has the advantage of being explicit in all situations and
unambiguous at parse-time. This choice agrees with the Clojure
implementation but not with Datomic. Datomic treats [a v] as a lookup
ref, is ambiguous at parse-time, and is disambiguated in ways I do not
understand at transaction time. We mooted making lookup refs [[a v]]
and outlawing nested value vectors in transactions, but after
implementing that approach I decided it was better to handle lookup
refs at parse time and therefore outlawing nested value vectors is not
necessary.
* Handle lookup refs in the entity and value columns. Fixes #183.
* Pre 0a: Use a stack instead of into_iter.
* Pre 0b: Dedent.
* Pre 0c: Handle `e` after `v`.
This allows to use the original `e` while handling `v`.
* Explode value lists for :db.cardinality/many attributes. Fixes #284.
* Parse and accept map notation. Fixes #180.
* Pre: Modernize add() and retract() into one add_or_retract().
* Pre: Add is_collection and is_atom to edn::Value.
* Pre: Differentiate atoms from lookup-refs in value position.
Initially, I expected to accept arbitrary edn::Value instances in the
value position, and to differentiate in the transactor. However, the
implementation quickly became a two-stage parser, since we always
wanted to parse the resulting value position into some other known
thing using the tx-parser. To save calls into the parser and to allow
the parser to move forward with a smaller API surface, I push as much
of this parsing as possible into the initial parse.
* Pre: Modernize entities().
* Pre: Quote edn::Value::Text in Display.
* Review comment: Add and use edn::Value::into_atom.
* Review comment: Use skip(eof()) throughout.
* Review comment: VecDeque instead of Vec.
* Review comment: Part 0: Rename TempId to TempIdHandle.
* Review comment: Part 1: Differentiate internal and external tempids.
This breaks an abstraction boundary by pushing the Internal/External
split up to the Entity level in tx/ and tx-parser/. This just makes
it easier to explode Entity map notation instances into Entity
instances, taking an existing External tempid :db/id or generating a
new Internal tempid as appropriate. To do this without breaking the
abstraction boundary would require adding flexibility to the
transaction processor: we'd need to be able to turn Entity instances
into some internal enum and handle the two cases independently. It
wouldn't be too hard, but this reduces the combinatorial type
explosion.
2017-03-27 23:30:04 +00:00
[ :db / add 502 :test / many [ 2 3 ] ]
2019-07-17 15:46:47 +00:00
[ :db / add 503 :test / many [ 4 5 6 ] ] ] "
) ;
assert_matches! (
conn . last_transaction ( ) ,
" [[501 :test/many 1 ?tx true]
Lookup refs, nested vector values, map notation. Fixes #180, fixes #183, fixes #284. (#382) r=rnewman
* Pre: Fix error in parser macros.
* Pre: Make test unwrapping more verbose.
* Pre: Make lookup refs be (lookup-ref a v) in the entity position.
This has the advantage of being explicit in all situations and
unambiguous at parse-time. This choice agrees with the Clojure
implementation but not with Datomic. Datomic treats [a v] as a lookup
ref, is ambiguous at parse-time, and is disambiguated in ways I do not
understand at transaction time. We mooted making lookup refs [[a v]]
and outlawing nested value vectors in transactions, but after
implementing that approach I decided it was better to handle lookup
refs at parse time and therefore outlawing nested value vectors is not
necessary.
* Handle lookup refs in the entity and value columns. Fixes #183.
* Pre 0a: Use a stack instead of into_iter.
* Pre 0b: Dedent.
* Pre 0c: Handle `e` after `v`.
This allows to use the original `e` while handling `v`.
* Explode value lists for :db.cardinality/many attributes. Fixes #284.
* Parse and accept map notation. Fixes #180.
* Pre: Modernize add() and retract() into one add_or_retract().
* Pre: Add is_collection and is_atom to edn::Value.
* Pre: Differentiate atoms from lookup-refs in value position.
Initially, I expected to accept arbitrary edn::Value instances in the
value position, and to differentiate in the transactor. However, the
implementation quickly became a two-stage parser, since we always
wanted to parse the resulting value position into some other known
thing using the tx-parser. To save calls into the parser and to allow
the parser to move forward with a smaller API surface, I push as much
of this parsing as possible into the initial parse.
* Pre: Modernize entities().
* Pre: Quote edn::Value::Text in Display.
* Review comment: Add and use edn::Value::into_atom.
* Review comment: Use skip(eof()) throughout.
* Review comment: VecDeque instead of Vec.
* Review comment: Part 0: Rename TempId to TempIdHandle.
* Review comment: Part 1: Differentiate internal and external tempids.
This breaks an abstraction boundary by pushing the Internal/External
split up to the Entity level in tx/ and tx-parser/. This just makes
it easier to explode Entity map notation instances into Entity
instances, taking an existing External tempid :db/id or generating a
new Internal tempid as appropriate. To do this without breaking the
abstraction boundary would require adding flexibility to the
transaction processor: we'd need to be able to turn Entity instances
into some internal enum and handle the two cases independently. It
wouldn't be too hard, but this reduces the combinatorial type
explosion.
2017-03-27 23:30:04 +00:00
[ 502 :test / many 2 ? tx true ]
[ 502 :test / many 3 ? tx true ]
[ 503 :test / many 4 ? tx true ]
[ 503 :test / many 5 ? tx true ]
[ 503 :test / many 6 ? tx true ]
2019-07-17 15:46:47 +00:00
[ ? tx :db / txInstant ? ms ? tx true ] ] "
) ;
Lookup refs, nested vector values, map notation. Fixes #180, fixes #183, fixes #284. (#382) r=rnewman
* Pre: Fix error in parser macros.
* Pre: Make test unwrapping more verbose.
* Pre: Make lookup refs be (lookup-ref a v) in the entity position.
This has the advantage of being explicit in all situations and
unambiguous at parse-time. This choice agrees with the Clojure
implementation but not with Datomic. Datomic treats [a v] as a lookup
ref, is ambiguous at parse-time, and is disambiguated in ways I do not
understand at transaction time. We mooted making lookup refs [[a v]]
and outlawing nested value vectors in transactions, but after
implementing that approach I decided it was better to handle lookup
refs at parse time and therefore outlawing nested value vectors is not
necessary.
* Handle lookup refs in the entity and value columns. Fixes #183.
* Pre 0a: Use a stack instead of into_iter.
* Pre 0b: Dedent.
* Pre 0c: Handle `e` after `v`.
This allows to use the original `e` while handling `v`.
* Explode value lists for :db.cardinality/many attributes. Fixes #284.
* Parse and accept map notation. Fixes #180.
* Pre: Modernize add() and retract() into one add_or_retract().
* Pre: Add is_collection and is_atom to edn::Value.
* Pre: Differentiate atoms from lookup-refs in value position.
Initially, I expected to accept arbitrary edn::Value instances in the
value position, and to differentiate in the transactor. However, the
implementation quickly became a two-stage parser, since we always
wanted to parse the resulting value position into some other known
thing using the tx-parser. To save calls into the parser and to allow
the parser to move forward with a smaller API surface, I push as much
of this parsing as possible into the initial parse.
* Pre: Modernize entities().
* Pre: Quote edn::Value::Text in Display.
* Review comment: Add and use edn::Value::into_atom.
* Review comment: Use skip(eof()) throughout.
* Review comment: VecDeque instead of Vec.
* Review comment: Part 0: Rename TempId to TempIdHandle.
* Review comment: Part 1: Differentiate internal and external tempids.
This breaks an abstraction boundary by pushing the Internal/External
split up to the Entity level in tx/ and tx-parser/. This just makes
it easier to explode Entity map notation instances into Entity
instances, taking an existing External tempid :db/id or generating a
new Internal tempid as appropriate. To do this without breaking the
abstraction boundary would require adding flexibility to the
transaction processor: we'd need to be able to turn Entity instances
into some internal enum and handle the two cases independently. It
wouldn't be too hard, but this reduces the combinatorial type
explosion.
2017-03-27 23:30:04 +00:00
// Check that we can explode nested vectors for :db.cardinality/many attributes.
assert_transact! ( conn , " [[:db/add 600 :test/many [1 [2] [[3] [4]] []]]] " ) ;
2019-07-17 15:46:47 +00:00
assert_matches! (
conn . last_transaction ( ) ,
" [[600 :test/many 1 ?tx true]
Lookup refs, nested vector values, map notation. Fixes #180, fixes #183, fixes #284. (#382) r=rnewman
* Pre: Fix error in parser macros.
* Pre: Make test unwrapping more verbose.
* Pre: Make lookup refs be (lookup-ref a v) in the entity position.
This has the advantage of being explicit in all situations and
unambiguous at parse-time. This choice agrees with the Clojure
implementation but not with Datomic. Datomic treats [a v] as a lookup
ref, is ambiguous at parse-time, and is disambiguated in ways I do not
understand at transaction time. We mooted making lookup refs [[a v]]
and outlawing nested value vectors in transactions, but after
implementing that approach I decided it was better to handle lookup
refs at parse time and therefore outlawing nested value vectors is not
necessary.
* Handle lookup refs in the entity and value columns. Fixes #183.
* Pre 0a: Use a stack instead of into_iter.
* Pre 0b: Dedent.
* Pre 0c: Handle `e` after `v`.
This allows to use the original `e` while handling `v`.
* Explode value lists for :db.cardinality/many attributes. Fixes #284.
* Parse and accept map notation. Fixes #180.
* Pre: Modernize add() and retract() into one add_or_retract().
* Pre: Add is_collection and is_atom to edn::Value.
* Pre: Differentiate atoms from lookup-refs in value position.
Initially, I expected to accept arbitrary edn::Value instances in the
value position, and to differentiate in the transactor. However, the
implementation quickly became a two-stage parser, since we always
wanted to parse the resulting value position into some other known
thing using the tx-parser. To save calls into the parser and to allow
the parser to move forward with a smaller API surface, I push as much
of this parsing as possible into the initial parse.
* Pre: Modernize entities().
* Pre: Quote edn::Value::Text in Display.
* Review comment: Add and use edn::Value::into_atom.
* Review comment: Use skip(eof()) throughout.
* Review comment: VecDeque instead of Vec.
* Review comment: Part 0: Rename TempId to TempIdHandle.
* Review comment: Part 1: Differentiate internal and external tempids.
This breaks an abstraction boundary by pushing the Internal/External
split up to the Entity level in tx/ and tx-parser/. This just makes
it easier to explode Entity map notation instances into Entity
instances, taking an existing External tempid :db/id or generating a
new Internal tempid as appropriate. To do this without breaking the
abstraction boundary would require adding flexibility to the
transaction processor: we'd need to be able to turn Entity instances
into some internal enum and handle the two cases independently. It
wouldn't be too hard, but this reduces the combinatorial type
explosion.
2017-03-27 23:30:04 +00:00
[ 600 :test / many 2 ? tx true ]
[ 600 :test / many 3 ? tx true ]
[ 600 :test / many 4 ? tx true ]
2019-07-17 15:46:47 +00:00
[ ? tx :db / txInstant ? ms ? tx true ] ] "
) ;
Lookup refs, nested vector values, map notation. Fixes #180, fixes #183, fixes #284. (#382) r=rnewman
* Pre: Fix error in parser macros.
* Pre: Make test unwrapping more verbose.
* Pre: Make lookup refs be (lookup-ref a v) in the entity position.
This has the advantage of being explicit in all situations and
unambiguous at parse-time. This choice agrees with the Clojure
implementation but not with Datomic. Datomic treats [a v] as a lookup
ref, is ambiguous at parse-time, and is disambiguated in ways I do not
understand at transaction time. We mooted making lookup refs [[a v]]
and outlawing nested value vectors in transactions, but after
implementing that approach I decided it was better to handle lookup
refs at parse time and therefore outlawing nested value vectors is not
necessary.
* Handle lookup refs in the entity and value columns. Fixes #183.
* Pre 0a: Use a stack instead of into_iter.
* Pre 0b: Dedent.
* Pre 0c: Handle `e` after `v`.
This allows to use the original `e` while handling `v`.
* Explode value lists for :db.cardinality/many attributes. Fixes #284.
* Parse and accept map notation. Fixes #180.
* Pre: Modernize add() and retract() into one add_or_retract().
* Pre: Add is_collection and is_atom to edn::Value.
* Pre: Differentiate atoms from lookup-refs in value position.
Initially, I expected to accept arbitrary edn::Value instances in the
value position, and to differentiate in the transactor. However, the
implementation quickly became a two-stage parser, since we always
wanted to parse the resulting value position into some other known
thing using the tx-parser. To save calls into the parser and to allow
the parser to move forward with a smaller API surface, I push as much
of this parsing as possible into the initial parse.
* Pre: Modernize entities().
* Pre: Quote edn::Value::Text in Display.
* Review comment: Add and use edn::Value::into_atom.
* Review comment: Use skip(eof()) throughout.
* Review comment: VecDeque instead of Vec.
* Review comment: Part 0: Rename TempId to TempIdHandle.
* Review comment: Part 1: Differentiate internal and external tempids.
This breaks an abstraction boundary by pushing the Internal/External
split up to the Entity level in tx/ and tx-parser/. This just makes
it easier to explode Entity map notation instances into Entity
instances, taking an existing External tempid :db/id or generating a
new Internal tempid as appropriate. To do this without breaking the
abstraction boundary would require adding flexibility to the
transaction processor: we'd need to be able to turn Entity instances
into some internal enum and handle the two cases independently. It
wouldn't be too hard, but this reduces the combinatorial type
explosion.
2017-03-27 23:30:04 +00:00
// Check that we cannot explode vectors for :db.cardinality/one attributes.
assert_transact! ( conn ,
" [[:db/add 501 :test/one [1]]] " ,
Err ( " not yet implemented: Cannot explode vector value for attribute 222 that is not :db.cardinality :db.cardinality/many " ) ) ;
assert_transact! ( conn ,
" [[:db/add 501 :test/one [2 3]]] " ,
Err ( " not yet implemented: Cannot explode vector value for attribute 222 that is not :db.cardinality :db.cardinality/many " ) ) ;
}
#[ test ]
fn test_explode_map_notation ( ) {
let mut conn = TestConn ::default ( ) ;
// Start by installing a few attributes.
2019-07-17 15:46:47 +00:00
assert_transact! (
conn ,
" [[:db/add 111 :db/ident :test/many]
Lookup refs, nested vector values, map notation. Fixes #180, fixes #183, fixes #284. (#382) r=rnewman
* Pre: Fix error in parser macros.
* Pre: Make test unwrapping more verbose.
* Pre: Make lookup refs be (lookup-ref a v) in the entity position.
This has the advantage of being explicit in all situations and
unambiguous at parse-time. This choice agrees with the Clojure
implementation but not with Datomic. Datomic treats [a v] as a lookup
ref, is ambiguous at parse-time, and is disambiguated in ways I do not
understand at transaction time. We mooted making lookup refs [[a v]]
and outlawing nested value vectors in transactions, but after
implementing that approach I decided it was better to handle lookup
refs at parse time and therefore outlawing nested value vectors is not
necessary.
* Handle lookup refs in the entity and value columns. Fixes #183.
* Pre 0a: Use a stack instead of into_iter.
* Pre 0b: Dedent.
* Pre 0c: Handle `e` after `v`.
This allows to use the original `e` while handling `v`.
* Explode value lists for :db.cardinality/many attributes. Fixes #284.
* Parse and accept map notation. Fixes #180.
* Pre: Modernize add() and retract() into one add_or_retract().
* Pre: Add is_collection and is_atom to edn::Value.
* Pre: Differentiate atoms from lookup-refs in value position.
Initially, I expected to accept arbitrary edn::Value instances in the
value position, and to differentiate in the transactor. However, the
implementation quickly became a two-stage parser, since we always
wanted to parse the resulting value position into some other known
thing using the tx-parser. To save calls into the parser and to allow
the parser to move forward with a smaller API surface, I push as much
of this parsing as possible into the initial parse.
* Pre: Modernize entities().
* Pre: Quote edn::Value::Text in Display.
* Review comment: Add and use edn::Value::into_atom.
* Review comment: Use skip(eof()) throughout.
* Review comment: VecDeque instead of Vec.
* Review comment: Part 0: Rename TempId to TempIdHandle.
* Review comment: Part 1: Differentiate internal and external tempids.
This breaks an abstraction boundary by pushing the Internal/External
split up to the Entity level in tx/ and tx-parser/. This just makes
it easier to explode Entity map notation instances into Entity
instances, taking an existing External tempid :db/id or generating a
new Internal tempid as appropriate. To do this without breaking the
abstraction boundary would require adding flexibility to the
transaction processor: we'd need to be able to turn Entity instances
into some internal enum and handle the two cases independently. It
wouldn't be too hard, but this reduces the combinatorial type
explosion.
2017-03-27 23:30:04 +00:00
[ :db / add 111 :db / valueType :db . type / long ]
[ :db / add 111 :db / cardinality :db . cardinality / many ]
[ :db / add 222 :db / ident :test / component ]
[ :db / add 222 :db / isComponent true ]
[ :db / add 222 :db / valueType :db . type / ref ]
[ :db / add 333 :db / ident :test / unique ]
[ :db / add 333 :db / unique :db . unique / identity ]
[ :db / add 333 :db / index true ]
[ :db / add 333 :db / valueType :db . type / long ]
[ :db / add 444 :db / ident :test / dangling ]
2019-07-17 15:46:47 +00:00
[ :db / add 444 :db / valueType :db . type / ref ] ] "
) ;
Lookup refs, nested vector values, map notation. Fixes #180, fixes #183, fixes #284. (#382) r=rnewman
* Pre: Fix error in parser macros.
* Pre: Make test unwrapping more verbose.
* Pre: Make lookup refs be (lookup-ref a v) in the entity position.
This has the advantage of being explicit in all situations and
unambiguous at parse-time. This choice agrees with the Clojure
implementation but not with Datomic. Datomic treats [a v] as a lookup
ref, is ambiguous at parse-time, and is disambiguated in ways I do not
understand at transaction time. We mooted making lookup refs [[a v]]
and outlawing nested value vectors in transactions, but after
implementing that approach I decided it was better to handle lookup
refs at parse time and therefore outlawing nested value vectors is not
necessary.
* Handle lookup refs in the entity and value columns. Fixes #183.
* Pre 0a: Use a stack instead of into_iter.
* Pre 0b: Dedent.
* Pre 0c: Handle `e` after `v`.
This allows to use the original `e` while handling `v`.
* Explode value lists for :db.cardinality/many attributes. Fixes #284.
* Parse and accept map notation. Fixes #180.
* Pre: Modernize add() and retract() into one add_or_retract().
* Pre: Add is_collection and is_atom to edn::Value.
* Pre: Differentiate atoms from lookup-refs in value position.
Initially, I expected to accept arbitrary edn::Value instances in the
value position, and to differentiate in the transactor. However, the
implementation quickly became a two-stage parser, since we always
wanted to parse the resulting value position into some other known
thing using the tx-parser. To save calls into the parser and to allow
the parser to move forward with a smaller API surface, I push as much
of this parsing as possible into the initial parse.
* Pre: Modernize entities().
* Pre: Quote edn::Value::Text in Display.
* Review comment: Add and use edn::Value::into_atom.
* Review comment: Use skip(eof()) throughout.
* Review comment: VecDeque instead of Vec.
* Review comment: Part 0: Rename TempId to TempIdHandle.
* Review comment: Part 1: Differentiate internal and external tempids.
This breaks an abstraction boundary by pushing the Internal/External
split up to the Entity level in tx/ and tx-parser/. This just makes
it easier to explode Entity map notation instances into Entity
instances, taking an existing External tempid :db/id or generating a
new Internal tempid as appropriate. To do this without breaking the
abstraction boundary would require adding flexibility to the
transaction processor: we'd need to be able to turn Entity instances
into some internal enum and handle the two cases independently. It
wouldn't be too hard, but this reduces the combinatorial type
explosion.
2017-03-27 23:30:04 +00:00
// Check that we can explode map notation without :db/id.
let report = assert_transact! ( conn , " [{:test/many 1}] " ) ;
2019-07-17 15:46:47 +00:00
assert_matches! (
conn . last_transaction ( ) ,
" [[?e :test/many 1 ?tx true]
[ ? tx :db / txInstant ? ms ? tx true ] ] "
) ;
assert_matches! ( tempids ( & report ) , " {} " ) ;
Lookup refs, nested vector values, map notation. Fixes #180, fixes #183, fixes #284. (#382) r=rnewman
* Pre: Fix error in parser macros.
* Pre: Make test unwrapping more verbose.
* Pre: Make lookup refs be (lookup-ref a v) in the entity position.
This has the advantage of being explicit in all situations and
unambiguous at parse-time. This choice agrees with the Clojure
implementation but not with Datomic. Datomic treats [a v] as a lookup
ref, is ambiguous at parse-time, and is disambiguated in ways I do not
understand at transaction time. We mooted making lookup refs [[a v]]
and outlawing nested value vectors in transactions, but after
implementing that approach I decided it was better to handle lookup
refs at parse time and therefore outlawing nested value vectors is not
necessary.
* Handle lookup refs in the entity and value columns. Fixes #183.
* Pre 0a: Use a stack instead of into_iter.
* Pre 0b: Dedent.
* Pre 0c: Handle `e` after `v`.
This allows to use the original `e` while handling `v`.
* Explode value lists for :db.cardinality/many attributes. Fixes #284.
* Parse and accept map notation. Fixes #180.
* Pre: Modernize add() and retract() into one add_or_retract().
* Pre: Add is_collection and is_atom to edn::Value.
* Pre: Differentiate atoms from lookup-refs in value position.
Initially, I expected to accept arbitrary edn::Value instances in the
value position, and to differentiate in the transactor. However, the
implementation quickly became a two-stage parser, since we always
wanted to parse the resulting value position into some other known
thing using the tx-parser. To save calls into the parser and to allow
the parser to move forward with a smaller API surface, I push as much
of this parsing as possible into the initial parse.
* Pre: Modernize entities().
* Pre: Quote edn::Value::Text in Display.
* Review comment: Add and use edn::Value::into_atom.
* Review comment: Use skip(eof()) throughout.
* Review comment: VecDeque instead of Vec.
* Review comment: Part 0: Rename TempId to TempIdHandle.
* Review comment: Part 1: Differentiate internal and external tempids.
This breaks an abstraction boundary by pushing the Internal/External
split up to the Entity level in tx/ and tx-parser/. This just makes
it easier to explode Entity map notation instances into Entity
instances, taking an existing External tempid :db/id or generating a
new Internal tempid as appropriate. To do this without breaking the
abstraction boundary would require adding flexibility to the
transaction processor: we'd need to be able to turn Entity instances
into some internal enum and handle the two cases independently. It
wouldn't be too hard, but this reduces the combinatorial type
explosion.
2017-03-27 23:30:04 +00:00
// Check that we can explode map notation with :db/id, as an entid, ident, and tempid.
2019-07-17 15:46:47 +00:00
let report = assert_transact! (
conn ,
" [{:db/id :db/ident :test/many 1}
Lookup refs, nested vector values, map notation. Fixes #180, fixes #183, fixes #284. (#382) r=rnewman
* Pre: Fix error in parser macros.
* Pre: Make test unwrapping more verbose.
* Pre: Make lookup refs be (lookup-ref a v) in the entity position.
This has the advantage of being explicit in all situations and
unambiguous at parse-time. This choice agrees with the Clojure
implementation but not with Datomic. Datomic treats [a v] as a lookup
ref, is ambiguous at parse-time, and is disambiguated in ways I do not
understand at transaction time. We mooted making lookup refs [[a v]]
and outlawing nested value vectors in transactions, but after
implementing that approach I decided it was better to handle lookup
refs at parse time and therefore outlawing nested value vectors is not
necessary.
* Handle lookup refs in the entity and value columns. Fixes #183.
* Pre 0a: Use a stack instead of into_iter.
* Pre 0b: Dedent.
* Pre 0c: Handle `e` after `v`.
This allows to use the original `e` while handling `v`.
* Explode value lists for :db.cardinality/many attributes. Fixes #284.
* Parse and accept map notation. Fixes #180.
* Pre: Modernize add() and retract() into one add_or_retract().
* Pre: Add is_collection and is_atom to edn::Value.
* Pre: Differentiate atoms from lookup-refs in value position.
Initially, I expected to accept arbitrary edn::Value instances in the
value position, and to differentiate in the transactor. However, the
implementation quickly became a two-stage parser, since we always
wanted to parse the resulting value position into some other known
thing using the tx-parser. To save calls into the parser and to allow
the parser to move forward with a smaller API surface, I push as much
of this parsing as possible into the initial parse.
* Pre: Modernize entities().
* Pre: Quote edn::Value::Text in Display.
* Review comment: Add and use edn::Value::into_atom.
* Review comment: Use skip(eof()) throughout.
* Review comment: VecDeque instead of Vec.
* Review comment: Part 0: Rename TempId to TempIdHandle.
* Review comment: Part 1: Differentiate internal and external tempids.
This breaks an abstraction boundary by pushing the Internal/External
split up to the Entity level in tx/ and tx-parser/. This just makes
it easier to explode Entity map notation instances into Entity
instances, taking an existing External tempid :db/id or generating a
new Internal tempid as appropriate. To do this without breaking the
abstraction boundary would require adding flexibility to the
transaction processor: we'd need to be able to turn Entity instances
into some internal enum and handle the two cases independently. It
wouldn't be too hard, but this reduces the combinatorial type
explosion.
2017-03-27 23:30:04 +00:00
{ :db / id 500 :test / many 2 }
2019-07-17 15:46:47 +00:00
{ :db / id \ " t \" :test/many 3}] "
) ;
assert_matches! (
conn . last_transaction ( ) ,
" [[1 :test/many 1 ?tx true]
Lookup refs, nested vector values, map notation. Fixes #180, fixes #183, fixes #284. (#382) r=rnewman
* Pre: Fix error in parser macros.
* Pre: Make test unwrapping more verbose.
* Pre: Make lookup refs be (lookup-ref a v) in the entity position.
This has the advantage of being explicit in all situations and
unambiguous at parse-time. This choice agrees with the Clojure
implementation but not with Datomic. Datomic treats [a v] as a lookup
ref, is ambiguous at parse-time, and is disambiguated in ways I do not
understand at transaction time. We mooted making lookup refs [[a v]]
and outlawing nested value vectors in transactions, but after
implementing that approach I decided it was better to handle lookup
refs at parse time and therefore outlawing nested value vectors is not
necessary.
* Handle lookup refs in the entity and value columns. Fixes #183.
* Pre 0a: Use a stack instead of into_iter.
* Pre 0b: Dedent.
* Pre 0c: Handle `e` after `v`.
This allows to use the original `e` while handling `v`.
* Explode value lists for :db.cardinality/many attributes. Fixes #284.
* Parse and accept map notation. Fixes #180.
* Pre: Modernize add() and retract() into one add_or_retract().
* Pre: Add is_collection and is_atom to edn::Value.
* Pre: Differentiate atoms from lookup-refs in value position.
Initially, I expected to accept arbitrary edn::Value instances in the
value position, and to differentiate in the transactor. However, the
implementation quickly became a two-stage parser, since we always
wanted to parse the resulting value position into some other known
thing using the tx-parser. To save calls into the parser and to allow
the parser to move forward with a smaller API surface, I push as much
of this parsing as possible into the initial parse.
* Pre: Modernize entities().
* Pre: Quote edn::Value::Text in Display.
* Review comment: Add and use edn::Value::into_atom.
* Review comment: Use skip(eof()) throughout.
* Review comment: VecDeque instead of Vec.
* Review comment: Part 0: Rename TempId to TempIdHandle.
* Review comment: Part 1: Differentiate internal and external tempids.
This breaks an abstraction boundary by pushing the Internal/External
split up to the Entity level in tx/ and tx-parser/. This just makes
it easier to explode Entity map notation instances into Entity
instances, taking an existing External tempid :db/id or generating a
new Internal tempid as appropriate. To do this without breaking the
abstraction boundary would require adding flexibility to the
transaction processor: we'd need to be able to turn Entity instances
into some internal enum and handle the two cases independently. It
wouldn't be too hard, but this reduces the combinatorial type
explosion.
2017-03-27 23:30:04 +00:00
[ 500 :test / many 2 ? tx true ]
[ ? e :test / many 3 ? tx true ]
2019-07-17 15:46:47 +00:00
[ ? tx :db / txInstant ? ms ? tx true ] ] "
) ;
assert_matches! ( tempids ( & report ) , " { \" t \" 65537} " ) ;
Lookup refs, nested vector values, map notation. Fixes #180, fixes #183, fixes #284. (#382) r=rnewman
* Pre: Fix error in parser macros.
* Pre: Make test unwrapping more verbose.
* Pre: Make lookup refs be (lookup-ref a v) in the entity position.
This has the advantage of being explicit in all situations and
unambiguous at parse-time. This choice agrees with the Clojure
implementation but not with Datomic. Datomic treats [a v] as a lookup
ref, is ambiguous at parse-time, and is disambiguated in ways I do not
understand at transaction time. We mooted making lookup refs [[a v]]
and outlawing nested value vectors in transactions, but after
implementing that approach I decided it was better to handle lookup
refs at parse time and therefore outlawing nested value vectors is not
necessary.
* Handle lookup refs in the entity and value columns. Fixes #183.
* Pre 0a: Use a stack instead of into_iter.
* Pre 0b: Dedent.
* Pre 0c: Handle `e` after `v`.
This allows to use the original `e` while handling `v`.
* Explode value lists for :db.cardinality/many attributes. Fixes #284.
* Parse and accept map notation. Fixes #180.
* Pre: Modernize add() and retract() into one add_or_retract().
* Pre: Add is_collection and is_atom to edn::Value.
* Pre: Differentiate atoms from lookup-refs in value position.
Initially, I expected to accept arbitrary edn::Value instances in the
value position, and to differentiate in the transactor. However, the
implementation quickly became a two-stage parser, since we always
wanted to parse the resulting value position into some other known
thing using the tx-parser. To save calls into the parser and to allow
the parser to move forward with a smaller API surface, I push as much
of this parsing as possible into the initial parse.
* Pre: Modernize entities().
* Pre: Quote edn::Value::Text in Display.
* Review comment: Add and use edn::Value::into_atom.
* Review comment: Use skip(eof()) throughout.
* Review comment: VecDeque instead of Vec.
* Review comment: Part 0: Rename TempId to TempIdHandle.
* Review comment: Part 1: Differentiate internal and external tempids.
This breaks an abstraction boundary by pushing the Internal/External
split up to the Entity level in tx/ and tx-parser/. This just makes
it easier to explode Entity map notation instances into Entity
instances, taking an existing External tempid :db/id or generating a
new Internal tempid as appropriate. To do this without breaking the
abstraction boundary would require adding flexibility to the
transaction processor: we'd need to be able to turn Entity instances
into some internal enum and handle the two cases independently. It
wouldn't be too hard, but this reduces the combinatorial type
explosion.
2017-03-27 23:30:04 +00:00
2018-05-15 07:43:07 +00:00
// Check that we can explode map notation with :db/id as a lookup-ref or tx-function.
2019-07-17 15:46:47 +00:00
let report = assert_transact! (
conn ,
" [{:db/id (lookup-ref :db/ident :db/ident) :test/many 4}
{ :db / id ( transaction - tx ) :test / many 5 } ] "
) ;
assert_matches! (
conn . last_transaction ( ) ,
" [[1 :test/many 4 ?tx true]
2018-05-15 07:43:07 +00:00
[ ? tx :db / txInstant ? ms ? tx true ]
2019-07-17 15:46:47 +00:00
[ ? tx :test / many 5 ? tx true ] ] "
) ;
assert_matches! ( tempids ( & report ) , " {} " ) ;
2018-05-15 07:43:07 +00:00
Lookup refs, nested vector values, map notation. Fixes #180, fixes #183, fixes #284. (#382) r=rnewman
* Pre: Fix error in parser macros.
* Pre: Make test unwrapping more verbose.
* Pre: Make lookup refs be (lookup-ref a v) in the entity position.
This has the advantage of being explicit in all situations and
unambiguous at parse-time. This choice agrees with the Clojure
implementation but not with Datomic. Datomic treats [a v] as a lookup
ref, is ambiguous at parse-time, and is disambiguated in ways I do not
understand at transaction time. We mooted making lookup refs [[a v]]
and outlawing nested value vectors in transactions, but after
implementing that approach I decided it was better to handle lookup
refs at parse time and therefore outlawing nested value vectors is not
necessary.
* Handle lookup refs in the entity and value columns. Fixes #183.
* Pre 0a: Use a stack instead of into_iter.
* Pre 0b: Dedent.
* Pre 0c: Handle `e` after `v`.
This allows to use the original `e` while handling `v`.
* Explode value lists for :db.cardinality/many attributes. Fixes #284.
* Parse and accept map notation. Fixes #180.
* Pre: Modernize add() and retract() into one add_or_retract().
* Pre: Add is_collection and is_atom to edn::Value.
* Pre: Differentiate atoms from lookup-refs in value position.
Initially, I expected to accept arbitrary edn::Value instances in the
value position, and to differentiate in the transactor. However, the
implementation quickly became a two-stage parser, since we always
wanted to parse the resulting value position into some other known
thing using the tx-parser. To save calls into the parser and to allow
the parser to move forward with a smaller API surface, I push as much
of this parsing as possible into the initial parse.
* Pre: Modernize entities().
* Pre: Quote edn::Value::Text in Display.
* Review comment: Add and use edn::Value::into_atom.
* Review comment: Use skip(eof()) throughout.
* Review comment: VecDeque instead of Vec.
* Review comment: Part 0: Rename TempId to TempIdHandle.
* Review comment: Part 1: Differentiate internal and external tempids.
This breaks an abstraction boundary by pushing the Internal/External
split up to the Entity level in tx/ and tx-parser/. This just makes
it easier to explode Entity map notation instances into Entity
instances, taking an existing External tempid :db/id or generating a
new Internal tempid as appropriate. To do this without breaking the
abstraction boundary would require adding flexibility to the
transaction processor: we'd need to be able to turn Entity instances
into some internal enum and handle the two cases independently. It
wouldn't be too hard, but this reduces the combinatorial type
explosion.
2017-03-27 23:30:04 +00:00
// Check that we can explode map notation with nested vector values.
let report = assert_transact! ( conn , " [{:test/many [1 2]}] " ) ;
2019-07-17 15:46:47 +00:00
assert_matches! (
conn . last_transaction ( ) ,
" [[?e :test/many 1 ?tx true]
Lookup refs, nested vector values, map notation. Fixes #180, fixes #183, fixes #284. (#382) r=rnewman
* Pre: Fix error in parser macros.
* Pre: Make test unwrapping more verbose.
* Pre: Make lookup refs be (lookup-ref a v) in the entity position.
This has the advantage of being explicit in all situations and
unambiguous at parse-time. This choice agrees with the Clojure
implementation but not with Datomic. Datomic treats [a v] as a lookup
ref, is ambiguous at parse-time, and is disambiguated in ways I do not
understand at transaction time. We mooted making lookup refs [[a v]]
and outlawing nested value vectors in transactions, but after
implementing that approach I decided it was better to handle lookup
refs at parse time and therefore outlawing nested value vectors is not
necessary.
* Handle lookup refs in the entity and value columns. Fixes #183.
* Pre 0a: Use a stack instead of into_iter.
* Pre 0b: Dedent.
* Pre 0c: Handle `e` after `v`.
This allows to use the original `e` while handling `v`.
* Explode value lists for :db.cardinality/many attributes. Fixes #284.
* Parse and accept map notation. Fixes #180.
* Pre: Modernize add() and retract() into one add_or_retract().
* Pre: Add is_collection and is_atom to edn::Value.
* Pre: Differentiate atoms from lookup-refs in value position.
Initially, I expected to accept arbitrary edn::Value instances in the
value position, and to differentiate in the transactor. However, the
implementation quickly became a two-stage parser, since we always
wanted to parse the resulting value position into some other known
thing using the tx-parser. To save calls into the parser and to allow
the parser to move forward with a smaller API surface, I push as much
of this parsing as possible into the initial parse.
* Pre: Modernize entities().
* Pre: Quote edn::Value::Text in Display.
* Review comment: Add and use edn::Value::into_atom.
* Review comment: Use skip(eof()) throughout.
* Review comment: VecDeque instead of Vec.
* Review comment: Part 0: Rename TempId to TempIdHandle.
* Review comment: Part 1: Differentiate internal and external tempids.
This breaks an abstraction boundary by pushing the Internal/External
split up to the Entity level in tx/ and tx-parser/. This just makes
it easier to explode Entity map notation instances into Entity
instances, taking an existing External tempid :db/id or generating a
new Internal tempid as appropriate. To do this without breaking the
abstraction boundary would require adding flexibility to the
transaction processor: we'd need to be able to turn Entity instances
into some internal enum and handle the two cases independently. It
wouldn't be too hard, but this reduces the combinatorial type
explosion.
2017-03-27 23:30:04 +00:00
[ ? e :test / many 2 ? tx true ]
2019-07-17 15:46:47 +00:00
[ ? tx :db / txInstant ? ms ? tx true ] ] "
) ;
assert_matches! ( tempids ( & report ) , " {} " ) ;
Lookup refs, nested vector values, map notation. Fixes #180, fixes #183, fixes #284. (#382) r=rnewman
* Pre: Fix error in parser macros.
* Pre: Make test unwrapping more verbose.
* Pre: Make lookup refs be (lookup-ref a v) in the entity position.
This has the advantage of being explicit in all situations and
unambiguous at parse-time. This choice agrees with the Clojure
implementation but not with Datomic. Datomic treats [a v] as a lookup
ref, is ambiguous at parse-time, and is disambiguated in ways I do not
understand at transaction time. We mooted making lookup refs [[a v]]
and outlawing nested value vectors in transactions, but after
implementing that approach I decided it was better to handle lookup
refs at parse time and therefore outlawing nested value vectors is not
necessary.
* Handle lookup refs in the entity and value columns. Fixes #183.
* Pre 0a: Use a stack instead of into_iter.
* Pre 0b: Dedent.
* Pre 0c: Handle `e` after `v`.
This allows to use the original `e` while handling `v`.
* Explode value lists for :db.cardinality/many attributes. Fixes #284.
* Parse and accept map notation. Fixes #180.
* Pre: Modernize add() and retract() into one add_or_retract().
* Pre: Add is_collection and is_atom to edn::Value.
* Pre: Differentiate atoms from lookup-refs in value position.
Initially, I expected to accept arbitrary edn::Value instances in the
value position, and to differentiate in the transactor. However, the
implementation quickly became a two-stage parser, since we always
wanted to parse the resulting value position into some other known
thing using the tx-parser. To save calls into the parser and to allow
the parser to move forward with a smaller API surface, I push as much
of this parsing as possible into the initial parse.
* Pre: Modernize entities().
* Pre: Quote edn::Value::Text in Display.
* Review comment: Add and use edn::Value::into_atom.
* Review comment: Use skip(eof()) throughout.
* Review comment: VecDeque instead of Vec.
* Review comment: Part 0: Rename TempId to TempIdHandle.
* Review comment: Part 1: Differentiate internal and external tempids.
This breaks an abstraction boundary by pushing the Internal/External
split up to the Entity level in tx/ and tx-parser/. This just makes
it easier to explode Entity map notation instances into Entity
instances, taking an existing External tempid :db/id or generating a
new Internal tempid as appropriate. To do this without breaking the
abstraction boundary would require adding flexibility to the
transaction processor: we'd need to be able to turn Entity instances
into some internal enum and handle the two cases independently. It
wouldn't be too hard, but this reduces the combinatorial type
explosion.
2017-03-27 23:30:04 +00:00
// Check that we can explode map notation with nested maps if the attribute is
// :db/isComponent true.
let report = assert_transact! ( conn , " [{:test/component {:test/many 1}}] " ) ;
2019-07-17 15:46:47 +00:00
assert_matches! (
conn . last_transaction ( ) ,
" [[?e :test/component ?f ?tx true]
Lookup refs, nested vector values, map notation. Fixes #180, fixes #183, fixes #284. (#382) r=rnewman
* Pre: Fix error in parser macros.
* Pre: Make test unwrapping more verbose.
* Pre: Make lookup refs be (lookup-ref a v) in the entity position.
This has the advantage of being explicit in all situations and
unambiguous at parse-time. This choice agrees with the Clojure
implementation but not with Datomic. Datomic treats [a v] as a lookup
ref, is ambiguous at parse-time, and is disambiguated in ways I do not
understand at transaction time. We mooted making lookup refs [[a v]]
and outlawing nested value vectors in transactions, but after
implementing that approach I decided it was better to handle lookup
refs at parse time and therefore outlawing nested value vectors is not
necessary.
* Handle lookup refs in the entity and value columns. Fixes #183.
* Pre 0a: Use a stack instead of into_iter.
* Pre 0b: Dedent.
* Pre 0c: Handle `e` after `v`.
This allows to use the original `e` while handling `v`.
* Explode value lists for :db.cardinality/many attributes. Fixes #284.
* Parse and accept map notation. Fixes #180.
* Pre: Modernize add() and retract() into one add_or_retract().
* Pre: Add is_collection and is_atom to edn::Value.
* Pre: Differentiate atoms from lookup-refs in value position.
Initially, I expected to accept arbitrary edn::Value instances in the
value position, and to differentiate in the transactor. However, the
implementation quickly became a two-stage parser, since we always
wanted to parse the resulting value position into some other known
thing using the tx-parser. To save calls into the parser and to allow
the parser to move forward with a smaller API surface, I push as much
of this parsing as possible into the initial parse.
* Pre: Modernize entities().
* Pre: Quote edn::Value::Text in Display.
* Review comment: Add and use edn::Value::into_atom.
* Review comment: Use skip(eof()) throughout.
* Review comment: VecDeque instead of Vec.
* Review comment: Part 0: Rename TempId to TempIdHandle.
* Review comment: Part 1: Differentiate internal and external tempids.
This breaks an abstraction boundary by pushing the Internal/External
split up to the Entity level in tx/ and tx-parser/. This just makes
it easier to explode Entity map notation instances into Entity
instances, taking an existing External tempid :db/id or generating a
new Internal tempid as appropriate. To do this without breaking the
abstraction boundary would require adding flexibility to the
transaction processor: we'd need to be able to turn Entity instances
into some internal enum and handle the two cases independently. It
wouldn't be too hard, but this reduces the combinatorial type
explosion.
2017-03-27 23:30:04 +00:00
[ ? f :test / many 1 ? tx true ]
2019-07-17 15:46:47 +00:00
[ ? tx :db / txInstant ? ms ? tx true ] ] "
) ;
assert_matches! ( tempids ( & report ) , " {} " ) ;
Lookup refs, nested vector values, map notation. Fixes #180, fixes #183, fixes #284. (#382) r=rnewman
* Pre: Fix error in parser macros.
* Pre: Make test unwrapping more verbose.
* Pre: Make lookup refs be (lookup-ref a v) in the entity position.
This has the advantage of being explicit in all situations and
unambiguous at parse-time. This choice agrees with the Clojure
implementation but not with Datomic. Datomic treats [a v] as a lookup
ref, is ambiguous at parse-time, and is disambiguated in ways I do not
understand at transaction time. We mooted making lookup refs [[a v]]
and outlawing nested value vectors in transactions, but after
implementing that approach I decided it was better to handle lookup
refs at parse time and therefore outlawing nested value vectors is not
necessary.
* Handle lookup refs in the entity and value columns. Fixes #183.
* Pre 0a: Use a stack instead of into_iter.
* Pre 0b: Dedent.
* Pre 0c: Handle `e` after `v`.
This allows to use the original `e` while handling `v`.
* Explode value lists for :db.cardinality/many attributes. Fixes #284.
* Parse and accept map notation. Fixes #180.
* Pre: Modernize add() and retract() into one add_or_retract().
* Pre: Add is_collection and is_atom to edn::Value.
* Pre: Differentiate atoms from lookup-refs in value position.
Initially, I expected to accept arbitrary edn::Value instances in the
value position, and to differentiate in the transactor. However, the
implementation quickly became a two-stage parser, since we always
wanted to parse the resulting value position into some other known
thing using the tx-parser. To save calls into the parser and to allow
the parser to move forward with a smaller API surface, I push as much
of this parsing as possible into the initial parse.
* Pre: Modernize entities().
* Pre: Quote edn::Value::Text in Display.
* Review comment: Add and use edn::Value::into_atom.
* Review comment: Use skip(eof()) throughout.
* Review comment: VecDeque instead of Vec.
* Review comment: Part 0: Rename TempId to TempIdHandle.
* Review comment: Part 1: Differentiate internal and external tempids.
This breaks an abstraction boundary by pushing the Internal/External
split up to the Entity level in tx/ and tx-parser/. This just makes
it easier to explode Entity map notation instances into Entity
instances, taking an existing External tempid :db/id or generating a
new Internal tempid as appropriate. To do this without breaking the
abstraction boundary would require adding flexibility to the
transaction processor: we'd need to be able to turn Entity instances
into some internal enum and handle the two cases independently. It
wouldn't be too hard, but this reduces the combinatorial type
explosion.
2017-03-27 23:30:04 +00:00
// Check that we can explode map notation with nested maps if the inner map contains a
// :db/unique :db.unique/identity attribute.
let report = assert_transact! ( conn , " [{:test/dangling {:test/unique 10}}] " ) ;
2019-07-17 15:46:47 +00:00
assert_matches! (
conn . last_transaction ( ) ,
" [[?e :test/dangling ?f ?tx true]
Lookup refs, nested vector values, map notation. Fixes #180, fixes #183, fixes #284. (#382) r=rnewman
* Pre: Fix error in parser macros.
* Pre: Make test unwrapping more verbose.
* Pre: Make lookup refs be (lookup-ref a v) in the entity position.
This has the advantage of being explicit in all situations and
unambiguous at parse-time. This choice agrees with the Clojure
implementation but not with Datomic. Datomic treats [a v] as a lookup
ref, is ambiguous at parse-time, and is disambiguated in ways I do not
understand at transaction time. We mooted making lookup refs [[a v]]
and outlawing nested value vectors in transactions, but after
implementing that approach I decided it was better to handle lookup
refs at parse time and therefore outlawing nested value vectors is not
necessary.
* Handle lookup refs in the entity and value columns. Fixes #183.
* Pre 0a: Use a stack instead of into_iter.
* Pre 0b: Dedent.
* Pre 0c: Handle `e` after `v`.
This allows to use the original `e` while handling `v`.
* Explode value lists for :db.cardinality/many attributes. Fixes #284.
* Parse and accept map notation. Fixes #180.
* Pre: Modernize add() and retract() into one add_or_retract().
* Pre: Add is_collection and is_atom to edn::Value.
* Pre: Differentiate atoms from lookup-refs in value position.
Initially, I expected to accept arbitrary edn::Value instances in the
value position, and to differentiate in the transactor. However, the
implementation quickly became a two-stage parser, since we always
wanted to parse the resulting value position into some other known
thing using the tx-parser. To save calls into the parser and to allow
the parser to move forward with a smaller API surface, I push as much
of this parsing as possible into the initial parse.
* Pre: Modernize entities().
* Pre: Quote edn::Value::Text in Display.
* Review comment: Add and use edn::Value::into_atom.
* Review comment: Use skip(eof()) throughout.
* Review comment: VecDeque instead of Vec.
* Review comment: Part 0: Rename TempId to TempIdHandle.
* Review comment: Part 1: Differentiate internal and external tempids.
This breaks an abstraction boundary by pushing the Internal/External
split up to the Entity level in tx/ and tx-parser/. This just makes
it easier to explode Entity map notation instances into Entity
instances, taking an existing External tempid :db/id or generating a
new Internal tempid as appropriate. To do this without breaking the
abstraction boundary would require adding flexibility to the
transaction processor: we'd need to be able to turn Entity instances
into some internal enum and handle the two cases independently. It
wouldn't be too hard, but this reduces the combinatorial type
explosion.
2017-03-27 23:30:04 +00:00
[ ? f :test / unique 10 ? tx true ]
2019-07-17 15:46:47 +00:00
[ ? tx :db / txInstant ? ms ? tx true ] ] "
) ;
assert_matches! ( tempids ( & report ) , " {} " ) ;
Lookup refs, nested vector values, map notation. Fixes #180, fixes #183, fixes #284. (#382) r=rnewman
* Pre: Fix error in parser macros.
* Pre: Make test unwrapping more verbose.
* Pre: Make lookup refs be (lookup-ref a v) in the entity position.
This has the advantage of being explicit in all situations and
unambiguous at parse-time. This choice agrees with the Clojure
implementation but not with Datomic. Datomic treats [a v] as a lookup
ref, is ambiguous at parse-time, and is disambiguated in ways I do not
understand at transaction time. We mooted making lookup refs [[a v]]
and outlawing nested value vectors in transactions, but after
implementing that approach I decided it was better to handle lookup
refs at parse time and therefore outlawing nested value vectors is not
necessary.
* Handle lookup refs in the entity and value columns. Fixes #183.
* Pre 0a: Use a stack instead of into_iter.
* Pre 0b: Dedent.
* Pre 0c: Handle `e` after `v`.
This allows to use the original `e` while handling `v`.
* Explode value lists for :db.cardinality/many attributes. Fixes #284.
* Parse and accept map notation. Fixes #180.
* Pre: Modernize add() and retract() into one add_or_retract().
* Pre: Add is_collection and is_atom to edn::Value.
* Pre: Differentiate atoms from lookup-refs in value position.
Initially, I expected to accept arbitrary edn::Value instances in the
value position, and to differentiate in the transactor. However, the
implementation quickly became a two-stage parser, since we always
wanted to parse the resulting value position into some other known
thing using the tx-parser. To save calls into the parser and to allow
the parser to move forward with a smaller API surface, I push as much
of this parsing as possible into the initial parse.
* Pre: Modernize entities().
* Pre: Quote edn::Value::Text in Display.
* Review comment: Add and use edn::Value::into_atom.
* Review comment: Use skip(eof()) throughout.
* Review comment: VecDeque instead of Vec.
* Review comment: Part 0: Rename TempId to TempIdHandle.
* Review comment: Part 1: Differentiate internal and external tempids.
This breaks an abstraction boundary by pushing the Internal/External
split up to the Entity level in tx/ and tx-parser/. This just makes
it easier to explode Entity map notation instances into Entity
instances, taking an existing External tempid :db/id or generating a
new Internal tempid as appropriate. To do this without breaking the
abstraction boundary would require adding flexibility to the
transaction processor: we'd need to be able to turn Entity instances
into some internal enum and handle the two cases independently. It
wouldn't be too hard, but this reduces the combinatorial type
explosion.
2017-03-27 23:30:04 +00:00
// Verify that we can't explode map notation with nested maps if the inner map would be
// dangling.
assert_transact! ( conn ,
" [{:test/dangling {:test/many 11}}] " ,
Err ( " not yet implemented: Cannot explode nested map value that would lead to dangling entity for attribute 444 " ) ) ;
// Verify that we can explode map notation with nested maps, even if the inner map would be
// dangling, if we give a :db/id explicitly.
2017-06-07 21:17:56 +00:00
assert_transact! ( conn , " [{:test/dangling {:db/id \" t \" :test/many 12}}] " ) ;
Lookup refs, nested vector values, map notation. Fixes #180, fixes #183, fixes #284. (#382) r=rnewman
* Pre: Fix error in parser macros.
* Pre: Make test unwrapping more verbose.
* Pre: Make lookup refs be (lookup-ref a v) in the entity position.
This has the advantage of being explicit in all situations and
unambiguous at parse-time. This choice agrees with the Clojure
implementation but not with Datomic. Datomic treats [a v] as a lookup
ref, is ambiguous at parse-time, and is disambiguated in ways I do not
understand at transaction time. We mooted making lookup refs [[a v]]
and outlawing nested value vectors in transactions, but after
implementing that approach I decided it was better to handle lookup
refs at parse time and therefore outlawing nested value vectors is not
necessary.
* Handle lookup refs in the entity and value columns. Fixes #183.
* Pre 0a: Use a stack instead of into_iter.
* Pre 0b: Dedent.
* Pre 0c: Handle `e` after `v`.
This allows to use the original `e` while handling `v`.
* Explode value lists for :db.cardinality/many attributes. Fixes #284.
* Parse and accept map notation. Fixes #180.
* Pre: Modernize add() and retract() into one add_or_retract().
* Pre: Add is_collection and is_atom to edn::Value.
* Pre: Differentiate atoms from lookup-refs in value position.
Initially, I expected to accept arbitrary edn::Value instances in the
value position, and to differentiate in the transactor. However, the
implementation quickly became a two-stage parser, since we always
wanted to parse the resulting value position into some other known
thing using the tx-parser. To save calls into the parser and to allow
the parser to move forward with a smaller API surface, I push as much
of this parsing as possible into the initial parse.
* Pre: Modernize entities().
* Pre: Quote edn::Value::Text in Display.
* Review comment: Add and use edn::Value::into_atom.
* Review comment: Use skip(eof()) throughout.
* Review comment: VecDeque instead of Vec.
* Review comment: Part 0: Rename TempId to TempIdHandle.
* Review comment: Part 1: Differentiate internal and external tempids.
This breaks an abstraction boundary by pushing the Internal/External
split up to the Entity level in tx/ and tx-parser/. This just makes
it easier to explode Entity map notation instances into Entity
instances, taking an existing External tempid :db/id or generating a
new Internal tempid as appropriate. To do this without breaking the
abstraction boundary would require adding flexibility to the
transaction processor: we'd need to be able to turn Entity instances
into some internal enum and handle the two cases independently. It
wouldn't be too hard, but this reduces the combinatorial type
explosion.
2017-03-27 23:30:04 +00:00
}
2017-06-06 22:21:51 +00:00
#[ test ]
fn test_explode_reversed_notation ( ) {
let mut conn = TestConn ::default ( ) ;
// Start by installing a few attributes.
2019-07-17 15:46:47 +00:00
assert_transact! (
conn ,
" [[:db/add 111 :db/ident :test/many]
2017-06-06 22:21:51 +00:00
[ :db / add 111 :db / valueType :db . type / long ]
[ :db / add 111 :db / cardinality :db . cardinality / many ]
[ :db / add 222 :db / ident :test / component ]
[ :db / add 222 :db / isComponent true ]
[ :db / add 222 :db / valueType :db . type / ref ]
[ :db / add 333 :db / ident :test / unique ]
[ :db / add 333 :db / unique :db . unique / identity ]
[ :db / add 333 :db / index true ]
[ :db / add 333 :db / valueType :db . type / long ]
[ :db / add 444 :db / ident :test / dangling ]
2019-07-17 15:46:47 +00:00
[ :db / add 444 :db / valueType :db . type / ref ] ] "
) ;
2017-06-06 22:21:51 +00:00
// Check that we can explode direct reversed notation, entids.
let report = assert_transact! ( conn , " [[:db/add 100 :test/_dangling 200]] " ) ;
2019-07-17 15:46:47 +00:00
assert_matches! (
conn . last_transaction ( ) ,
" [[200 :test/dangling 100 ?tx true]
[ ? tx :db / txInstant ? ms ? tx true ] ] "
) ;
assert_matches! ( tempids ( & report ) , " {} " ) ;
2017-06-06 22:21:51 +00:00
// Check that we can explode direct reversed notation, idents.
let report = assert_transact! ( conn , " [[:db/add :test/many :test/_dangling :test/unique]] " ) ;
2019-07-17 15:46:47 +00:00
assert_matches! (
conn . last_transaction ( ) ,
" [[333 :test/dangling :test/many ?tx true]
[ ? tx :db / txInstant ? ms ? tx true ] ] "
) ;
assert_matches! ( tempids ( & report ) , " {} " ) ;
2017-06-06 22:21:51 +00:00
// Check that we can explode direct reversed notation, tempids.
let report = assert_transact! ( conn , " [[:db/add \" s \" :test/_dangling \" t \" ]] " ) ;
2019-07-17 15:46:47 +00:00
assert_matches! (
conn . last_transaction ( ) ,
" [[65537 :test/dangling 65536 ?tx true]
[ ? tx :db / txInstant ? ms ? tx true ] ] "
) ;
2017-06-06 22:21:51 +00:00
// This is implementation specific, but it should be deterministic.
2019-07-17 15:46:47 +00:00
assert_matches! (
tempids ( & report ) ,
" { \" s \" 65536
\ " t \" 65537} "
) ;
2017-06-06 22:21:51 +00:00
// Check that we can explode reversed notation in map notation without :db/id.
2019-07-17 15:46:47 +00:00
let report = assert_transact! (
conn ,
" [{:test/_dangling 501}
2017-06-06 22:21:51 +00:00
{ :test / _dangling :test / many }
2019-07-17 15:46:47 +00:00
{ :test / _dangling \ " t \" }] "
) ;
assert_matches! (
conn . last_transaction ( ) ,
" [[111 :test/dangling ?e1 ?tx true]
2017-06-06 22:21:51 +00:00
[ 501 :test / dangling ? e2 ? tx true ]
[ 65538 :test / dangling ? e3 ? tx true ]
2019-07-17 15:46:47 +00:00
[ ? tx :db / txInstant ? ms ? tx true ] ] "
) ;
assert_matches! ( tempids ( & report ) , " { \" t \" 65538} " ) ;
2017-06-06 22:21:51 +00:00
// Check that we can explode reversed notation in map notation with :db/id, entid.
let report = assert_transact! ( conn , " [{:db/id 600 :test/_dangling 601}] " ) ;
2019-07-17 15:46:47 +00:00
assert_matches! (
conn . last_transaction ( ) ,
" [[601 :test/dangling 600 ?tx true]
[ ? tx :db / txInstant ? ms ? tx true ] ] "
) ;
assert_matches! ( tempids ( & report ) , " {} " ) ;
2017-06-06 22:21:51 +00:00
// Check that we can explode reversed notation in map notation with :db/id, ident.
2019-07-17 15:46:47 +00:00
let report = assert_transact! (
conn ,
" [{:db/id :test/component :test/_dangling :test/component}] "
) ;
assert_matches! (
conn . last_transaction ( ) ,
" [[222 :test/dangling :test/component ?tx true]
[ ? tx :db / txInstant ? ms ? tx true ] ] "
) ;
assert_matches! ( tempids ( & report ) , " {} " ) ;
2017-06-06 22:21:51 +00:00
// Check that we can explode reversed notation in map notation with :db/id, tempid.
let report = assert_transact! ( conn , " [{:db/id \" s \" :test/_dangling \" t \" }] " ) ;
2019-07-17 15:46:47 +00:00
assert_matches! (
conn . last_transaction ( ) ,
" [[65543 :test/dangling 65542 ?tx true]
[ ? tx :db / txInstant ? ms ? tx true ] ] "
) ;
2017-06-06 22:21:51 +00:00
// This is implementation specific, but it should be deterministic.
2019-07-17 15:46:47 +00:00
assert_matches! (
tempids ( & report ) ,
" { \" s \" 65542
\ " t \" 65543} "
) ;
2017-06-06 22:21:51 +00:00
2017-06-07 20:37:40 +00:00
// Check that we can use the same attribute in both forward and backward form in the same
// transaction.
2019-07-17 15:46:47 +00:00
let report = assert_transact! (
conn ,
" [[:db/add 888 :test/dangling 889]
[ :db / add 888 :test / _dangling 889 ] ] "
) ;
assert_matches! (
conn . last_transaction ( ) ,
" [[888 :test/dangling 889 ?tx true]
2017-06-07 20:37:40 +00:00
[ 889 :test / dangling 888 ? tx true ]
2019-07-17 15:46:47 +00:00
[ ? tx :db / txInstant ? ms ? tx true ] ] "
) ;
assert_matches! ( tempids ( & report ) , " {} " ) ;
2017-06-07 20:37:40 +00:00
// Check that we can use the same attribute in both forward and backward form in the same
// transaction in map notation.
2019-07-17 15:46:47 +00:00
let report = assert_transact! (
conn ,
" [{:db/id 998 :test/dangling 999 :test/_dangling 999}] "
) ;
assert_matches! (
conn . last_transaction ( ) ,
" [[998 :test/dangling 999 ?tx true]
2017-06-07 20:37:40 +00:00
[ 999 :test / dangling 998 ? tx true ]
2019-07-17 15:46:47 +00:00
[ ? tx :db / txInstant ? ms ? tx true ] ] "
) ;
assert_matches! ( tempids ( & report ) , " {} " ) ;
2017-06-07 21:17:56 +00:00
}
#[ test ]
fn test_explode_reversed_notation_errors ( ) {
let mut conn = TestConn ::default ( ) ;
// Start by installing a few attributes.
2019-07-17 15:46:47 +00:00
assert_transact! (
conn ,
" [[:db/add 111 :db/ident :test/many]
2017-06-07 21:17:56 +00:00
[ :db / add 111 :db / valueType :db . type / long ]
[ :db / add 111 :db / cardinality :db . cardinality / many ]
[ :db / add 222 :db / ident :test / component ]
[ :db / add 222 :db / isComponent true ]
[ :db / add 222 :db / valueType :db . type / ref ]
[ :db / add 333 :db / ident :test / unique ]
[ :db / add 333 :db / unique :db . unique / identity ]
[ :db / add 333 :db / index true ]
[ :db / add 333 :db / valueType :db . type / long ]
[ :db / add 444 :db / ident :test / dangling ]
2019-07-17 15:46:47 +00:00
[ :db / add 444 :db / valueType :db . type / ref ] ] "
) ;
2017-06-07 21:17:56 +00:00
// `tx-parser` should fail to parse direct reverse notation with nested value maps and
// nested value vectors, so we only test things that "get through" to the map notation
// dynamic processor here.
2017-06-06 22:21:51 +00:00
// Verify that we can't explode reverse notation in map notation with nested value maps.
assert_transact! ( conn ,
2017-06-07 21:17:56 +00:00
" [{:test/_dangling {:test/many 14}}] " ,
2017-06-06 22:21:51 +00:00
Err ( " not yet implemented: Cannot explode map notation value in :attr/_reversed notation for attribute 444 " ) ) ;
// Verify that we can't explode reverse notation in map notation with nested value vectors.
assert_transact! ( conn ,
" [{:test/_dangling [:test/many]}] " ,
Err ( " not yet implemented: Cannot explode vector value in :attr/_reversed notation for attribute 444 " ) ) ;
// Verify that we can't use reverse notation with non-:db.type/ref attributes.
assert_transact! ( conn ,
" [{:test/_unique 500}] " ,
Err ( " not yet implemented: Cannot use :attr/_reversed notation for attribute 333 that is not :db/valueType :db.type/ref " ) ) ;
// Verify that we can't use reverse notation with unrecognized attributes.
2019-07-17 15:46:47 +00:00
assert_transact! (
conn ,
" [{:test/_unknown 500}] " ,
Err ( " no entid found for ident: :test/unknown " )
) ; // TODO: make this error reference the original :test/_unknown.
2017-06-06 22:21:51 +00:00
// Verify that we can't use reverse notation with bad value types: here, an unknown keyword
// that can't be coerced to a ref.
2019-07-17 15:46:47 +00:00
assert_transact! (
conn ,
" [{:test/_dangling :test/unknown}] " ,
Err ( " no entid found for ident: :test/unknown " )
) ;
2017-06-06 22:21:51 +00:00
// And here, a float.
2019-07-17 15:46:47 +00:00
assert_transact! (
conn ,
" [{:test/_dangling 1.23}] " ,
Err ( " value \' 1.23 \' is not the expected Mentat value type Ref " )
) ;
2017-06-06 22:21:51 +00:00
}
2018-01-23 23:11:38 +00:00
#[ test ]
fn test_cardinality_one_violation_existing_entity ( ) {
let mut conn = TestConn ::default ( ) ;
// Start by installing a few attributes.
2019-07-17 15:46:47 +00:00
assert_transact! (
conn ,
r #" [
2018-01-23 23:11:38 +00:00
[ :db / add 111 :db / ident :test / one ]
[ :db / add 111 :db / valueType :db . type / long ]
[ :db / add 111 :db / cardinality :db . cardinality / one ]
[ :db / add 112 :db / ident :test / unique ]
[ :db / add 112 :db / index true ]
[ :db / add 112 :db / valueType :db . type / string ]
[ :db / add 112 :db / cardinality :db . cardinality / one ]
[ :db / add 112 :db / unique :db . unique / identity ]
2019-07-17 15:46:47 +00:00
] " #
) ;
2018-01-23 23:11:38 +00:00
2019-07-17 15:46:47 +00:00
assert_transact! (
conn ,
r #" [
2018-01-23 23:11:38 +00:00
[ :db / add " foo " :test / unique " x " ]
2019-07-17 15:46:47 +00:00
] " #
) ;
2018-01-23 23:11:38 +00:00
// You can try to assert two values for the same entity and attribute,
// but you'll get an error.
assert_transact! ( conn , r #" [
[ :db / add " foo " :test / unique " x " ]
[ :db / add " foo " :test / one 123 ]
[ :db / add " bar " :test / unique " x " ]
[ :db / add " bar " :test / one 124 ]
] " #,
Add type checking and constraint checking to the transactor. (#663, #532, #679)
This should address #663, by re-inserting type checking in the
transactor stack after the entry point used by the term builder.
Before this commit, we were using an SQLite UNIQUE index to assert
that no `[e a]` pair, with `a` a cardinality one attribute, was
asserted more than once. However, that's not in line with Datomic,
which treats transaction inputs as a set and allows a single datom
like `[e a v]` to appear multiple times. It's both awkward and not
particularly efficient to look for _distinct_ repetitions in SQL, so
we accept some runtime cost in order to check for repetitions in the
transactor. This will allow us to address #532, which is really about
whether we treat inputs as sets. A side benefit is that we can
provide more helpful error messages when the transactor does detect
that the input truly violates the cardinality constraints of the
schema.
This commit builds a trie while error checking and collecting final
terms, which should be fairly efficient. It also allows a simpler
expression of input-provided :db/txInstant datoms, which in turn
uncovered a small issue with the transaction watcher, where-by the
watcher would not see non-input-provided :db/txInstant datoms.
This transition to Datomic-like input-as-set semantics allows us to
address #532. Previously, two tempids that upserted to the same entid
would produce duplicate datoms, and that would have been rejected by
the transactor -- correctly, since we did not allow duplicate datoms
under the input-as-list semantics. With input-as-set semantics,
duplicate datoms are allowed; and that means that we must allow
tempids to be equivalent, i.e., to resolve to the same tempid.
To achieve this, we:
- index the set of tempids
- identify tempid indices that share an upsert
- map tempids to a dense set of contiguous integer labels
We use the well-known union-find algorithm, as implemented by
petgraph, to efficiently manage the set of equivalent tempids.
Along the way, I've fixed and added tests for two small errors in the
transactor. First, don't drop datoms resolved by upsert (#679).
Second, ensure that complex upserts are allocated.
I don't know quite what happened here. The Clojure implementation
correctly kept complex upserts that hadn't resolved as complex
upserts (see
https://github.com/mozilla/mentat/blob/9a9dfb502acf5e4cdb1059d4aac831d7603063c8/src/common/datomish/transact.cljc#L436)
and then allocated complex upserts if they didn't resolve (see
https://github.com/mozilla/mentat/blob/9a9dfb502acf5e4cdb1059d4aac831d7603063c8/src/common/datomish/transact.cljc#L509).
Based on the code comments, I think the Rust implementation must have
incorrectly tried to optimize by handling all complex upserts in at
most a single generation of evolution, and that's just not correct.
We're effectively implementing a topological sort, using very specific
domain knowledge, and its not true that a node in a topological sort
can be considered only once!
2018-04-30 22:16:05 +00:00
// This is implementation specific (due to the allocated entid), but it should be deterministic.
Err ( " schema constraint violation: cardinality conflicts: \n CardinalityOneAddConflict { e: 65536, a: 111, vs: {Long(123), Long(124)} } \n " ) ) ;
2018-01-23 23:11:38 +00:00
// It also fails for map notation.
assert_transact! ( conn , r #" [
{ :test / unique " x " , :test / one 123 }
{ :test / unique " x " , :test / one 124 }
] " #,
Add type checking and constraint checking to the transactor. (#663, #532, #679)
This should address #663, by re-inserting type checking in the
transactor stack after the entry point used by the term builder.
Before this commit, we were using an SQLite UNIQUE index to assert
that no `[e a]` pair, with `a` a cardinality one attribute, was
asserted more than once. However, that's not in line with Datomic,
which treats transaction inputs as a set and allows a single datom
like `[e a v]` to appear multiple times. It's both awkward and not
particularly efficient to look for _distinct_ repetitions in SQL, so
we accept some runtime cost in order to check for repetitions in the
transactor. This will allow us to address #532, which is really about
whether we treat inputs as sets. A side benefit is that we can
provide more helpful error messages when the transactor does detect
that the input truly violates the cardinality constraints of the
schema.
This commit builds a trie while error checking and collecting final
terms, which should be fairly efficient. It also allows a simpler
expression of input-provided :db/txInstant datoms, which in turn
uncovered a small issue with the transaction watcher, where-by the
watcher would not see non-input-provided :db/txInstant datoms.
This transition to Datomic-like input-as-set semantics allows us to
address #532. Previously, two tempids that upserted to the same entid
would produce duplicate datoms, and that would have been rejected by
the transactor -- correctly, since we did not allow duplicate datoms
under the input-as-list semantics. With input-as-set semantics,
duplicate datoms are allowed; and that means that we must allow
tempids to be equivalent, i.e., to resolve to the same tempid.
To achieve this, we:
- index the set of tempids
- identify tempid indices that share an upsert
- map tempids to a dense set of contiguous integer labels
We use the well-known union-find algorithm, as implemented by
petgraph, to efficiently manage the set of equivalent tempids.
Along the way, I've fixed and added tests for two small errors in the
transactor. First, don't drop datoms resolved by upsert (#679).
Second, ensure that complex upserts are allocated.
I don't know quite what happened here. The Clojure implementation
correctly kept complex upserts that hadn't resolved as complex
upserts (see
https://github.com/mozilla/mentat/blob/9a9dfb502acf5e4cdb1059d4aac831d7603063c8/src/common/datomish/transact.cljc#L436)
and then allocated complex upserts if they didn't resolve (see
https://github.com/mozilla/mentat/blob/9a9dfb502acf5e4cdb1059d4aac831d7603063c8/src/common/datomish/transact.cljc#L509).
Based on the code comments, I think the Rust implementation must have
incorrectly tried to optimize by handling all complex upserts in at
most a single generation of evolution, and that's just not correct.
We're effectively implementing a topological sort, using very specific
domain knowledge, and its not true that a node in a topological sort
can be considered only once!
2018-04-30 22:16:05 +00:00
// This is implementation specific (due to the allocated entid), but it should be deterministic.
Err ( " schema constraint violation: cardinality conflicts: \n CardinalityOneAddConflict { e: 65536, a: 111, vs: {Long(123), Long(124)} } \n " ) ) ;
2018-01-23 23:11:38 +00:00
}
2018-05-01 20:47:33 +00:00
#[ test ]
fn test_conflicting_upserts ( ) {
let mut conn = TestConn ::default ( ) ;
2020-01-14 15:46:21 +00:00
assert_transact! (
conn ,
r #" [
2018-05-01 20:47:33 +00:00
{ :db / ident :page / id :db / valueType :db . type / string :db / index true :db / unique :db . unique / identity }
{ :db / ident :page / ref :db / valueType :db . type / ref :db / index true :db / unique :db . unique / identity }
{ :db / ident :page / title :db / valueType :db . type / string :db / cardinality :db . cardinality / many }
2020-01-14 15:46:21 +00:00
] " #
) ;
2018-05-01 20:47:33 +00:00
// Let's test some conflicting upserts. First, valid data to work with -- note self references.
2019-07-17 15:46:47 +00:00
assert_transact! (
conn ,
r #" [
2018-05-01 20:47:33 +00:00
[ :db / add 111 :page / id " 1 " ]
[ :db / add 111 :page / ref 111 ]
[ :db / add 222 :page / id " 2 " ]
[ :db / add 222 :page / ref 222 ]
2019-07-17 15:46:47 +00:00
] " #
) ;
2018-05-01 20:47:33 +00:00
// Now valid upserts. Note the references are valid.
2019-07-17 15:46:47 +00:00
let report = assert_transact! (
conn ,
r #" [
2018-05-01 20:47:33 +00:00
[ :db / add " a " :page / id " 1 " ]
[ :db / add " a " :page / ref " a " ]
[ :db / add " b " :page / id " 2 " ]
[ :db / add " b " :page / ref " b " ]
2019-07-17 15:46:47 +00:00
] " #
) ;
assert_matches! (
tempids ( & report ) ,
" { \" a \" 111
\ " b \" 222} "
) ;
2018-05-01 20:47:33 +00:00
// Now conflicting upserts. Note the references are reversed. This example is interesting
// because the first round `UpsertE` instances upsert, and this resolves all of the tempids
// in the `UpsertEV` instances. However, those `UpsertEV` instances lead to conflicting
// upserts! This tests that we don't resolve too far, giving a chance for those upserts to
// fail. This error message is crossing generations, although it's not reflected in the
// error data structure.
assert_transact! ( conn , r #" [
[ :db / add " a " :page / id " 1 " ]
[ :db / add " a " :page / ref " b " ]
[ :db / add " b " :page / id " 2 " ]
[ :db / add " b " :page / ref " a " ]
] " #,
Err ( " schema constraint violation: conflicting upserts: \n tempid External( \" a \" ) upserts to {KnownEntid(111), KnownEntid(222)} \n tempid External( \" b \" ) upserts to {KnownEntid(111), KnownEntid(222)} \n " ) ) ;
// Here's a case where the upsert is not resolved, just allocated, but leads to conflicting
// cardinality one datoms.
assert_transact! ( conn , r #" [
[ :db / add " x " :page / ref 333 ]
[ :db / add " x " :page / ref 444 ]
] " #,
Add type checking and constraint checking to the transactor. (#663, #532, #679)
This should address #663, by re-inserting type checking in the
transactor stack after the entry point used by the term builder.
Before this commit, we were using an SQLite UNIQUE index to assert
that no `[e a]` pair, with `a` a cardinality one attribute, was
asserted more than once. However, that's not in line with Datomic,
which treats transaction inputs as a set and allows a single datom
like `[e a v]` to appear multiple times. It's both awkward and not
particularly efficient to look for _distinct_ repetitions in SQL, so
we accept some runtime cost in order to check for repetitions in the
transactor. This will allow us to address #532, which is really about
whether we treat inputs as sets. A side benefit is that we can
provide more helpful error messages when the transactor does detect
that the input truly violates the cardinality constraints of the
schema.
This commit builds a trie while error checking and collecting final
terms, which should be fairly efficient. It also allows a simpler
expression of input-provided :db/txInstant datoms, which in turn
uncovered a small issue with the transaction watcher, where-by the
watcher would not see non-input-provided :db/txInstant datoms.
This transition to Datomic-like input-as-set semantics allows us to
address #532. Previously, two tempids that upserted to the same entid
would produce duplicate datoms, and that would have been rejected by
the transactor -- correctly, since we did not allow duplicate datoms
under the input-as-list semantics. With input-as-set semantics,
duplicate datoms are allowed; and that means that we must allow
tempids to be equivalent, i.e., to resolve to the same tempid.
To achieve this, we:
- index the set of tempids
- identify tempid indices that share an upsert
- map tempids to a dense set of contiguous integer labels
We use the well-known union-find algorithm, as implemented by
petgraph, to efficiently manage the set of equivalent tempids.
Along the way, I've fixed and added tests for two small errors in the
transactor. First, don't drop datoms resolved by upsert (#679).
Second, ensure that complex upserts are allocated.
I don't know quite what happened here. The Clojure implementation
correctly kept complex upserts that hadn't resolved as complex
upserts (see
https://github.com/mozilla/mentat/blob/9a9dfb502acf5e4cdb1059d4aac831d7603063c8/src/common/datomish/transact.cljc#L436)
and then allocated complex upserts if they didn't resolve (see
https://github.com/mozilla/mentat/blob/9a9dfb502acf5e4cdb1059d4aac831d7603063c8/src/common/datomish/transact.cljc#L509).
Based on the code comments, I think the Rust implementation must have
incorrectly tried to optimize by handling all complex upserts in at
most a single generation of evolution, and that's just not correct.
We're effectively implementing a topological sort, using very specific
domain knowledge, and its not true that a node in a topological sort
can be considered only once!
2018-04-30 22:16:05 +00:00
Err ( " schema constraint violation: cardinality conflicts: \n CardinalityOneAddConflict { e: 65539, a: 65537, vs: {Ref(333), Ref(444)} } \n " ) ) ;
}
#[ test ]
fn test_upsert_issue_532 ( ) {
let mut conn = TestConn ::default ( ) ;
2020-01-14 15:46:21 +00:00
assert_transact! (
conn ,
r #" [
Add type checking and constraint checking to the transactor. (#663, #532, #679)
This should address #663, by re-inserting type checking in the
transactor stack after the entry point used by the term builder.
Before this commit, we were using an SQLite UNIQUE index to assert
that no `[e a]` pair, with `a` a cardinality one attribute, was
asserted more than once. However, that's not in line with Datomic,
which treats transaction inputs as a set and allows a single datom
like `[e a v]` to appear multiple times. It's both awkward and not
particularly efficient to look for _distinct_ repetitions in SQL, so
we accept some runtime cost in order to check for repetitions in the
transactor. This will allow us to address #532, which is really about
whether we treat inputs as sets. A side benefit is that we can
provide more helpful error messages when the transactor does detect
that the input truly violates the cardinality constraints of the
schema.
This commit builds a trie while error checking and collecting final
terms, which should be fairly efficient. It also allows a simpler
expression of input-provided :db/txInstant datoms, which in turn
uncovered a small issue with the transaction watcher, where-by the
watcher would not see non-input-provided :db/txInstant datoms.
This transition to Datomic-like input-as-set semantics allows us to
address #532. Previously, two tempids that upserted to the same entid
would produce duplicate datoms, and that would have been rejected by
the transactor -- correctly, since we did not allow duplicate datoms
under the input-as-list semantics. With input-as-set semantics,
duplicate datoms are allowed; and that means that we must allow
tempids to be equivalent, i.e., to resolve to the same tempid.
To achieve this, we:
- index the set of tempids
- identify tempid indices that share an upsert
- map tempids to a dense set of contiguous integer labels
We use the well-known union-find algorithm, as implemented by
petgraph, to efficiently manage the set of equivalent tempids.
Along the way, I've fixed and added tests for two small errors in the
transactor. First, don't drop datoms resolved by upsert (#679).
Second, ensure that complex upserts are allocated.
I don't know quite what happened here. The Clojure implementation
correctly kept complex upserts that hadn't resolved as complex
upserts (see
https://github.com/mozilla/mentat/blob/9a9dfb502acf5e4cdb1059d4aac831d7603063c8/src/common/datomish/transact.cljc#L436)
and then allocated complex upserts if they didn't resolve (see
https://github.com/mozilla/mentat/blob/9a9dfb502acf5e4cdb1059d4aac831d7603063c8/src/common/datomish/transact.cljc#L509).
Based on the code comments, I think the Rust implementation must have
incorrectly tried to optimize by handling all complex upserts in at
most a single generation of evolution, and that's just not correct.
We're effectively implementing a topological sort, using very specific
domain knowledge, and its not true that a node in a topological sort
can be considered only once!
2018-04-30 22:16:05 +00:00
{ :db / ident :page / id :db / valueType :db . type / string :db / index true :db / unique :db . unique / identity }
{ :db / ident :page / ref :db / valueType :db . type / ref :db / index true :db / unique :db . unique / identity }
{ :db / ident :page / title :db / valueType :db . type / string :db / cardinality :db . cardinality / many }
2020-01-14 15:46:21 +00:00
] " #
) ;
Add type checking and constraint checking to the transactor. (#663, #532, #679)
This should address #663, by re-inserting type checking in the
transactor stack after the entry point used by the term builder.
Before this commit, we were using an SQLite UNIQUE index to assert
that no `[e a]` pair, with `a` a cardinality one attribute, was
asserted more than once. However, that's not in line with Datomic,
which treats transaction inputs as a set and allows a single datom
like `[e a v]` to appear multiple times. It's both awkward and not
particularly efficient to look for _distinct_ repetitions in SQL, so
we accept some runtime cost in order to check for repetitions in the
transactor. This will allow us to address #532, which is really about
whether we treat inputs as sets. A side benefit is that we can
provide more helpful error messages when the transactor does detect
that the input truly violates the cardinality constraints of the
schema.
This commit builds a trie while error checking and collecting final
terms, which should be fairly efficient. It also allows a simpler
expression of input-provided :db/txInstant datoms, which in turn
uncovered a small issue with the transaction watcher, where-by the
watcher would not see non-input-provided :db/txInstant datoms.
This transition to Datomic-like input-as-set semantics allows us to
address #532. Previously, two tempids that upserted to the same entid
would produce duplicate datoms, and that would have been rejected by
the transactor -- correctly, since we did not allow duplicate datoms
under the input-as-list semantics. With input-as-set semantics,
duplicate datoms are allowed; and that means that we must allow
tempids to be equivalent, i.e., to resolve to the same tempid.
To achieve this, we:
- index the set of tempids
- identify tempid indices that share an upsert
- map tempids to a dense set of contiguous integer labels
We use the well-known union-find algorithm, as implemented by
petgraph, to efficiently manage the set of equivalent tempids.
Along the way, I've fixed and added tests for two small errors in the
transactor. First, don't drop datoms resolved by upsert (#679).
Second, ensure that complex upserts are allocated.
I don't know quite what happened here. The Clojure implementation
correctly kept complex upserts that hadn't resolved as complex
upserts (see
https://github.com/mozilla/mentat/blob/9a9dfb502acf5e4cdb1059d4aac831d7603063c8/src/common/datomish/transact.cljc#L436)
and then allocated complex upserts if they didn't resolve (see
https://github.com/mozilla/mentat/blob/9a9dfb502acf5e4cdb1059d4aac831d7603063c8/src/common/datomish/transact.cljc#L509).
Based on the code comments, I think the Rust implementation must have
incorrectly tried to optimize by handling all complex upserts in at
most a single generation of evolution, and that's just not correct.
We're effectively implementing a topological sort, using very specific
domain knowledge, and its not true that a node in a topological sort
can be considered only once!
2018-04-30 22:16:05 +00:00
// Observe that "foo" and "zot" upsert to the same entid, and that doesn't cause a
// cardinality conflict, because we treat the input with set semantics and accept
// duplicate datoms.
2019-07-17 15:46:47 +00:00
let report = assert_transact! (
conn ,
r #" [
Add type checking and constraint checking to the transactor. (#663, #532, #679)
This should address #663, by re-inserting type checking in the
transactor stack after the entry point used by the term builder.
Before this commit, we were using an SQLite UNIQUE index to assert
that no `[e a]` pair, with `a` a cardinality one attribute, was
asserted more than once. However, that's not in line with Datomic,
which treats transaction inputs as a set and allows a single datom
like `[e a v]` to appear multiple times. It's both awkward and not
particularly efficient to look for _distinct_ repetitions in SQL, so
we accept some runtime cost in order to check for repetitions in the
transactor. This will allow us to address #532, which is really about
whether we treat inputs as sets. A side benefit is that we can
provide more helpful error messages when the transactor does detect
that the input truly violates the cardinality constraints of the
schema.
This commit builds a trie while error checking and collecting final
terms, which should be fairly efficient. It also allows a simpler
expression of input-provided :db/txInstant datoms, which in turn
uncovered a small issue with the transaction watcher, where-by the
watcher would not see non-input-provided :db/txInstant datoms.
This transition to Datomic-like input-as-set semantics allows us to
address #532. Previously, two tempids that upserted to the same entid
would produce duplicate datoms, and that would have been rejected by
the transactor -- correctly, since we did not allow duplicate datoms
under the input-as-list semantics. With input-as-set semantics,
duplicate datoms are allowed; and that means that we must allow
tempids to be equivalent, i.e., to resolve to the same tempid.
To achieve this, we:
- index the set of tempids
- identify tempid indices that share an upsert
- map tempids to a dense set of contiguous integer labels
We use the well-known union-find algorithm, as implemented by
petgraph, to efficiently manage the set of equivalent tempids.
Along the way, I've fixed and added tests for two small errors in the
transactor. First, don't drop datoms resolved by upsert (#679).
Second, ensure that complex upserts are allocated.
I don't know quite what happened here. The Clojure implementation
correctly kept complex upserts that hadn't resolved as complex
upserts (see
https://github.com/mozilla/mentat/blob/9a9dfb502acf5e4cdb1059d4aac831d7603063c8/src/common/datomish/transact.cljc#L436)
and then allocated complex upserts if they didn't resolve (see
https://github.com/mozilla/mentat/blob/9a9dfb502acf5e4cdb1059d4aac831d7603063c8/src/common/datomish/transact.cljc#L509).
Based on the code comments, I think the Rust implementation must have
incorrectly tried to optimize by handling all complex upserts in at
most a single generation of evolution, and that's just not correct.
We're effectively implementing a topological sort, using very specific
domain knowledge, and its not true that a node in a topological sort
can be considered only once!
2018-04-30 22:16:05 +00:00
[ :db / add " bar " :page / id " z " ]
[ :db / add " foo " :page / ref " bar " ]
[ :db / add " foo " :page / title " x " ]
[ :db / add " zot " :page / ref " bar " ]
[ :db / add " zot " :db / ident :other / ident ]
2019-07-17 15:46:47 +00:00
] " #
) ;
assert_matches! (
tempids ( & report ) ,
" { \" bar \" ?b
Add type checking and constraint checking to the transactor. (#663, #532, #679)
This should address #663, by re-inserting type checking in the
transactor stack after the entry point used by the term builder.
Before this commit, we were using an SQLite UNIQUE index to assert
that no `[e a]` pair, with `a` a cardinality one attribute, was
asserted more than once. However, that's not in line with Datomic,
which treats transaction inputs as a set and allows a single datom
like `[e a v]` to appear multiple times. It's both awkward and not
particularly efficient to look for _distinct_ repetitions in SQL, so
we accept some runtime cost in order to check for repetitions in the
transactor. This will allow us to address #532, which is really about
whether we treat inputs as sets. A side benefit is that we can
provide more helpful error messages when the transactor does detect
that the input truly violates the cardinality constraints of the
schema.
This commit builds a trie while error checking and collecting final
terms, which should be fairly efficient. It also allows a simpler
expression of input-provided :db/txInstant datoms, which in turn
uncovered a small issue with the transaction watcher, where-by the
watcher would not see non-input-provided :db/txInstant datoms.
This transition to Datomic-like input-as-set semantics allows us to
address #532. Previously, two tempids that upserted to the same entid
would produce duplicate datoms, and that would have been rejected by
the transactor -- correctly, since we did not allow duplicate datoms
under the input-as-list semantics. With input-as-set semantics,
duplicate datoms are allowed; and that means that we must allow
tempids to be equivalent, i.e., to resolve to the same tempid.
To achieve this, we:
- index the set of tempids
- identify tempid indices that share an upsert
- map tempids to a dense set of contiguous integer labels
We use the well-known union-find algorithm, as implemented by
petgraph, to efficiently manage the set of equivalent tempids.
Along the way, I've fixed and added tests for two small errors in the
transactor. First, don't drop datoms resolved by upsert (#679).
Second, ensure that complex upserts are allocated.
I don't know quite what happened here. The Clojure implementation
correctly kept complex upserts that hadn't resolved as complex
upserts (see
https://github.com/mozilla/mentat/blob/9a9dfb502acf5e4cdb1059d4aac831d7603063c8/src/common/datomish/transact.cljc#L436)
and then allocated complex upserts if they didn't resolve (see
https://github.com/mozilla/mentat/blob/9a9dfb502acf5e4cdb1059d4aac831d7603063c8/src/common/datomish/transact.cljc#L509).
Based on the code comments, I think the Rust implementation must have
incorrectly tried to optimize by handling all complex upserts in at
most a single generation of evolution, and that's just not correct.
We're effectively implementing a topological sort, using very specific
domain knowledge, and its not true that a node in a topological sort
can be considered only once!
2018-04-30 22:16:05 +00:00
\ " foo \" ?f
2019-07-17 15:46:47 +00:00
\ " zot \" ?f} "
) ;
assert_matches! (
conn . last_transaction ( ) ,
" [[?b :page/id \" z \" ?tx true]
Add type checking and constraint checking to the transactor. (#663, #532, #679)
This should address #663, by re-inserting type checking in the
transactor stack after the entry point used by the term builder.
Before this commit, we were using an SQLite UNIQUE index to assert
that no `[e a]` pair, with `a` a cardinality one attribute, was
asserted more than once. However, that's not in line with Datomic,
which treats transaction inputs as a set and allows a single datom
like `[e a v]` to appear multiple times. It's both awkward and not
particularly efficient to look for _distinct_ repetitions in SQL, so
we accept some runtime cost in order to check for repetitions in the
transactor. This will allow us to address #532, which is really about
whether we treat inputs as sets. A side benefit is that we can
provide more helpful error messages when the transactor does detect
that the input truly violates the cardinality constraints of the
schema.
This commit builds a trie while error checking and collecting final
terms, which should be fairly efficient. It also allows a simpler
expression of input-provided :db/txInstant datoms, which in turn
uncovered a small issue with the transaction watcher, where-by the
watcher would not see non-input-provided :db/txInstant datoms.
This transition to Datomic-like input-as-set semantics allows us to
address #532. Previously, two tempids that upserted to the same entid
would produce duplicate datoms, and that would have been rejected by
the transactor -- correctly, since we did not allow duplicate datoms
under the input-as-list semantics. With input-as-set semantics,
duplicate datoms are allowed; and that means that we must allow
tempids to be equivalent, i.e., to resolve to the same tempid.
To achieve this, we:
- index the set of tempids
- identify tempid indices that share an upsert
- map tempids to a dense set of contiguous integer labels
We use the well-known union-find algorithm, as implemented by
petgraph, to efficiently manage the set of equivalent tempids.
Along the way, I've fixed and added tests for two small errors in the
transactor. First, don't drop datoms resolved by upsert (#679).
Second, ensure that complex upserts are allocated.
I don't know quite what happened here. The Clojure implementation
correctly kept complex upserts that hadn't resolved as complex
upserts (see
https://github.com/mozilla/mentat/blob/9a9dfb502acf5e4cdb1059d4aac831d7603063c8/src/common/datomish/transact.cljc#L436)
and then allocated complex upserts if they didn't resolve (see
https://github.com/mozilla/mentat/blob/9a9dfb502acf5e4cdb1059d4aac831d7603063c8/src/common/datomish/transact.cljc#L509).
Based on the code comments, I think the Rust implementation must have
incorrectly tried to optimize by handling all complex upserts in at
most a single generation of evolution, and that's just not correct.
We're effectively implementing a topological sort, using very specific
domain knowledge, and its not true that a node in a topological sort
can be considered only once!
2018-04-30 22:16:05 +00:00
[ ? f :db / ident :other / ident ? tx true ]
[ ? f :page / ref ? b ? tx true ]
[ ? f :page / title \ " x \" ?tx true]
2019-07-17 15:46:47 +00:00
[ ? tx :db / txInstant ? ms ? tx true ] ] "
) ;
Add type checking and constraint checking to the transactor. (#663, #532, #679)
This should address #663, by re-inserting type checking in the
transactor stack after the entry point used by the term builder.
Before this commit, we were using an SQLite UNIQUE index to assert
that no `[e a]` pair, with `a` a cardinality one attribute, was
asserted more than once. However, that's not in line with Datomic,
which treats transaction inputs as a set and allows a single datom
like `[e a v]` to appear multiple times. It's both awkward and not
particularly efficient to look for _distinct_ repetitions in SQL, so
we accept some runtime cost in order to check for repetitions in the
transactor. This will allow us to address #532, which is really about
whether we treat inputs as sets. A side benefit is that we can
provide more helpful error messages when the transactor does detect
that the input truly violates the cardinality constraints of the
schema.
This commit builds a trie while error checking and collecting final
terms, which should be fairly efficient. It also allows a simpler
expression of input-provided :db/txInstant datoms, which in turn
uncovered a small issue with the transaction watcher, where-by the
watcher would not see non-input-provided :db/txInstant datoms.
This transition to Datomic-like input-as-set semantics allows us to
address #532. Previously, two tempids that upserted to the same entid
would produce duplicate datoms, and that would have been rejected by
the transactor -- correctly, since we did not allow duplicate datoms
under the input-as-list semantics. With input-as-set semantics,
duplicate datoms are allowed; and that means that we must allow
tempids to be equivalent, i.e., to resolve to the same tempid.
To achieve this, we:
- index the set of tempids
- identify tempid indices that share an upsert
- map tempids to a dense set of contiguous integer labels
We use the well-known union-find algorithm, as implemented by
petgraph, to efficiently manage the set of equivalent tempids.
Along the way, I've fixed and added tests for two small errors in the
transactor. First, don't drop datoms resolved by upsert (#679).
Second, ensure that complex upserts are allocated.
I don't know quite what happened here. The Clojure implementation
correctly kept complex upserts that hadn't resolved as complex
upserts (see
https://github.com/mozilla/mentat/blob/9a9dfb502acf5e4cdb1059d4aac831d7603063c8/src/common/datomish/transact.cljc#L436)
and then allocated complex upserts if they didn't resolve (see
https://github.com/mozilla/mentat/blob/9a9dfb502acf5e4cdb1059d4aac831d7603063c8/src/common/datomish/transact.cljc#L509).
Based on the code comments, I think the Rust implementation must have
incorrectly tried to optimize by handling all complex upserts in at
most a single generation of evolution, and that's just not correct.
We're effectively implementing a topological sort, using very specific
domain knowledge, and its not true that a node in a topological sort
can be considered only once!
2018-04-30 22:16:05 +00:00
2019-07-17 15:46:47 +00:00
let report = assert_transact! (
conn ,
r #" [
Add type checking and constraint checking to the transactor. (#663, #532, #679)
This should address #663, by re-inserting type checking in the
transactor stack after the entry point used by the term builder.
Before this commit, we were using an SQLite UNIQUE index to assert
that no `[e a]` pair, with `a` a cardinality one attribute, was
asserted more than once. However, that's not in line with Datomic,
which treats transaction inputs as a set and allows a single datom
like `[e a v]` to appear multiple times. It's both awkward and not
particularly efficient to look for _distinct_ repetitions in SQL, so
we accept some runtime cost in order to check for repetitions in the
transactor. This will allow us to address #532, which is really about
whether we treat inputs as sets. A side benefit is that we can
provide more helpful error messages when the transactor does detect
that the input truly violates the cardinality constraints of the
schema.
This commit builds a trie while error checking and collecting final
terms, which should be fairly efficient. It also allows a simpler
expression of input-provided :db/txInstant datoms, which in turn
uncovered a small issue with the transaction watcher, where-by the
watcher would not see non-input-provided :db/txInstant datoms.
This transition to Datomic-like input-as-set semantics allows us to
address #532. Previously, two tempids that upserted to the same entid
would produce duplicate datoms, and that would have been rejected by
the transactor -- correctly, since we did not allow duplicate datoms
under the input-as-list semantics. With input-as-set semantics,
duplicate datoms are allowed; and that means that we must allow
tempids to be equivalent, i.e., to resolve to the same tempid.
To achieve this, we:
- index the set of tempids
- identify tempid indices that share an upsert
- map tempids to a dense set of contiguous integer labels
We use the well-known union-find algorithm, as implemented by
petgraph, to efficiently manage the set of equivalent tempids.
Along the way, I've fixed and added tests for two small errors in the
transactor. First, don't drop datoms resolved by upsert (#679).
Second, ensure that complex upserts are allocated.
I don't know quite what happened here. The Clojure implementation
correctly kept complex upserts that hadn't resolved as complex
upserts (see
https://github.com/mozilla/mentat/blob/9a9dfb502acf5e4cdb1059d4aac831d7603063c8/src/common/datomish/transact.cljc#L436)
and then allocated complex upserts if they didn't resolve (see
https://github.com/mozilla/mentat/blob/9a9dfb502acf5e4cdb1059d4aac831d7603063c8/src/common/datomish/transact.cljc#L509).
Based on the code comments, I think the Rust implementation must have
incorrectly tried to optimize by handling all complex upserts in at
most a single generation of evolution, and that's just not correct.
We're effectively implementing a topological sort, using very specific
domain knowledge, and its not true that a node in a topological sort
can be considered only once!
2018-04-30 22:16:05 +00:00
[ :db / add " foo " :page / id " x " ]
[ :db / add " foo " :page / title " x " ]
[ :db / add " bar " :page / id " x " ]
[ :db / add " bar " :page / title " y " ]
2019-07-17 15:46:47 +00:00
] " #
) ;
assert_matches! (
tempids ( & report ) ,
" { \" foo \" ?e
\ " bar \" ?e} "
) ;
Add type checking and constraint checking to the transactor. (#663, #532, #679)
This should address #663, by re-inserting type checking in the
transactor stack after the entry point used by the term builder.
Before this commit, we were using an SQLite UNIQUE index to assert
that no `[e a]` pair, with `a` a cardinality one attribute, was
asserted more than once. However, that's not in line with Datomic,
which treats transaction inputs as a set and allows a single datom
like `[e a v]` to appear multiple times. It's both awkward and not
particularly efficient to look for _distinct_ repetitions in SQL, so
we accept some runtime cost in order to check for repetitions in the
transactor. This will allow us to address #532, which is really about
whether we treat inputs as sets. A side benefit is that we can
provide more helpful error messages when the transactor does detect
that the input truly violates the cardinality constraints of the
schema.
This commit builds a trie while error checking and collecting final
terms, which should be fairly efficient. It also allows a simpler
expression of input-provided :db/txInstant datoms, which in turn
uncovered a small issue with the transaction watcher, where-by the
watcher would not see non-input-provided :db/txInstant datoms.
This transition to Datomic-like input-as-set semantics allows us to
address #532. Previously, two tempids that upserted to the same entid
would produce duplicate datoms, and that would have been rejected by
the transactor -- correctly, since we did not allow duplicate datoms
under the input-as-list semantics. With input-as-set semantics,
duplicate datoms are allowed; and that means that we must allow
tempids to be equivalent, i.e., to resolve to the same tempid.
To achieve this, we:
- index the set of tempids
- identify tempid indices that share an upsert
- map tempids to a dense set of contiguous integer labels
We use the well-known union-find algorithm, as implemented by
petgraph, to efficiently manage the set of equivalent tempids.
Along the way, I've fixed and added tests for two small errors in the
transactor. First, don't drop datoms resolved by upsert (#679).
Second, ensure that complex upserts are allocated.
I don't know quite what happened here. The Clojure implementation
correctly kept complex upserts that hadn't resolved as complex
upserts (see
https://github.com/mozilla/mentat/blob/9a9dfb502acf5e4cdb1059d4aac831d7603063c8/src/common/datomish/transact.cljc#L436)
and then allocated complex upserts if they didn't resolve (see
https://github.com/mozilla/mentat/blob/9a9dfb502acf5e4cdb1059d4aac831d7603063c8/src/common/datomish/transact.cljc#L509).
Based on the code comments, I think the Rust implementation must have
incorrectly tried to optimize by handling all complex upserts in at
most a single generation of evolution, and that's just not correct.
We're effectively implementing a topological sort, using very specific
domain knowledge, and its not true that a node in a topological sort
can be considered only once!
2018-04-30 22:16:05 +00:00
// One entity, two page titles.
2019-07-17 15:46:47 +00:00
assert_matches! (
conn . last_transaction ( ) ,
" [[?e :page/id \" x \" ?tx true]
Add type checking and constraint checking to the transactor. (#663, #532, #679)
This should address #663, by re-inserting type checking in the
transactor stack after the entry point used by the term builder.
Before this commit, we were using an SQLite UNIQUE index to assert
that no `[e a]` pair, with `a` a cardinality one attribute, was
asserted more than once. However, that's not in line with Datomic,
which treats transaction inputs as a set and allows a single datom
like `[e a v]` to appear multiple times. It's both awkward and not
particularly efficient to look for _distinct_ repetitions in SQL, so
we accept some runtime cost in order to check for repetitions in the
transactor. This will allow us to address #532, which is really about
whether we treat inputs as sets. A side benefit is that we can
provide more helpful error messages when the transactor does detect
that the input truly violates the cardinality constraints of the
schema.
This commit builds a trie while error checking and collecting final
terms, which should be fairly efficient. It also allows a simpler
expression of input-provided :db/txInstant datoms, which in turn
uncovered a small issue with the transaction watcher, where-by the
watcher would not see non-input-provided :db/txInstant datoms.
This transition to Datomic-like input-as-set semantics allows us to
address #532. Previously, two tempids that upserted to the same entid
would produce duplicate datoms, and that would have been rejected by
the transactor -- correctly, since we did not allow duplicate datoms
under the input-as-list semantics. With input-as-set semantics,
duplicate datoms are allowed; and that means that we must allow
tempids to be equivalent, i.e., to resolve to the same tempid.
To achieve this, we:
- index the set of tempids
- identify tempid indices that share an upsert
- map tempids to a dense set of contiguous integer labels
We use the well-known union-find algorithm, as implemented by
petgraph, to efficiently manage the set of equivalent tempids.
Along the way, I've fixed and added tests for two small errors in the
transactor. First, don't drop datoms resolved by upsert (#679).
Second, ensure that complex upserts are allocated.
I don't know quite what happened here. The Clojure implementation
correctly kept complex upserts that hadn't resolved as complex
upserts (see
https://github.com/mozilla/mentat/blob/9a9dfb502acf5e4cdb1059d4aac831d7603063c8/src/common/datomish/transact.cljc#L436)
and then allocated complex upserts if they didn't resolve (see
https://github.com/mozilla/mentat/blob/9a9dfb502acf5e4cdb1059d4aac831d7603063c8/src/common/datomish/transact.cljc#L509).
Based on the code comments, I think the Rust implementation must have
incorrectly tried to optimize by handling all complex upserts in at
most a single generation of evolution, and that's just not correct.
We're effectively implementing a topological sort, using very specific
domain knowledge, and its not true that a node in a topological sort
can be considered only once!
2018-04-30 22:16:05 +00:00
[ ? e :page / title \ " x \" ?tx true]
[ ? e :page / title \ " y \" ?tx true]
2019-07-17 15:46:47 +00:00
[ ? tx :db / txInstant ? ms ? tx true ] ] "
) ;
Add type checking and constraint checking to the transactor. (#663, #532, #679)
This should address #663, by re-inserting type checking in the
transactor stack after the entry point used by the term builder.
Before this commit, we were using an SQLite UNIQUE index to assert
that no `[e a]` pair, with `a` a cardinality one attribute, was
asserted more than once. However, that's not in line with Datomic,
which treats transaction inputs as a set and allows a single datom
like `[e a v]` to appear multiple times. It's both awkward and not
particularly efficient to look for _distinct_ repetitions in SQL, so
we accept some runtime cost in order to check for repetitions in the
transactor. This will allow us to address #532, which is really about
whether we treat inputs as sets. A side benefit is that we can
provide more helpful error messages when the transactor does detect
that the input truly violates the cardinality constraints of the
schema.
This commit builds a trie while error checking and collecting final
terms, which should be fairly efficient. It also allows a simpler
expression of input-provided :db/txInstant datoms, which in turn
uncovered a small issue with the transaction watcher, where-by the
watcher would not see non-input-provided :db/txInstant datoms.
This transition to Datomic-like input-as-set semantics allows us to
address #532. Previously, two tempids that upserted to the same entid
would produce duplicate datoms, and that would have been rejected by
the transactor -- correctly, since we did not allow duplicate datoms
under the input-as-list semantics. With input-as-set semantics,
duplicate datoms are allowed; and that means that we must allow
tempids to be equivalent, i.e., to resolve to the same tempid.
To achieve this, we:
- index the set of tempids
- identify tempid indices that share an upsert
- map tempids to a dense set of contiguous integer labels
We use the well-known union-find algorithm, as implemented by
petgraph, to efficiently manage the set of equivalent tempids.
Along the way, I've fixed and added tests for two small errors in the
transactor. First, don't drop datoms resolved by upsert (#679).
Second, ensure that complex upserts are allocated.
I don't know quite what happened here. The Clojure implementation
correctly kept complex upserts that hadn't resolved as complex
upserts (see
https://github.com/mozilla/mentat/blob/9a9dfb502acf5e4cdb1059d4aac831d7603063c8/src/common/datomish/transact.cljc#L436)
and then allocated complex upserts if they didn't resolve (see
https://github.com/mozilla/mentat/blob/9a9dfb502acf5e4cdb1059d4aac831d7603063c8/src/common/datomish/transact.cljc#L509).
Based on the code comments, I think the Rust implementation must have
incorrectly tried to optimize by handling all complex upserts in at
most a single generation of evolution, and that's just not correct.
We're effectively implementing a topological sort, using very specific
domain knowledge, and its not true that a node in a topological sort
can be considered only once!
2018-04-30 22:16:05 +00:00
// Here, "foo", "bar", and "baz", all refer to the same reference, but none of them actually
// upsert to existing entities.
2019-07-17 15:46:47 +00:00
let report = assert_transact! (
conn ,
r #" [
Add type checking and constraint checking to the transactor. (#663, #532, #679)
This should address #663, by re-inserting type checking in the
transactor stack after the entry point used by the term builder.
Before this commit, we were using an SQLite UNIQUE index to assert
that no `[e a]` pair, with `a` a cardinality one attribute, was
asserted more than once. However, that's not in line with Datomic,
which treats transaction inputs as a set and allows a single datom
like `[e a v]` to appear multiple times. It's both awkward and not
particularly efficient to look for _distinct_ repetitions in SQL, so
we accept some runtime cost in order to check for repetitions in the
transactor. This will allow us to address #532, which is really about
whether we treat inputs as sets. A side benefit is that we can
provide more helpful error messages when the transactor does detect
that the input truly violates the cardinality constraints of the
schema.
This commit builds a trie while error checking and collecting final
terms, which should be fairly efficient. It also allows a simpler
expression of input-provided :db/txInstant datoms, which in turn
uncovered a small issue with the transaction watcher, where-by the
watcher would not see non-input-provided :db/txInstant datoms.
This transition to Datomic-like input-as-set semantics allows us to
address #532. Previously, two tempids that upserted to the same entid
would produce duplicate datoms, and that would have been rejected by
the transactor -- correctly, since we did not allow duplicate datoms
under the input-as-list semantics. With input-as-set semantics,
duplicate datoms are allowed; and that means that we must allow
tempids to be equivalent, i.e., to resolve to the same tempid.
To achieve this, we:
- index the set of tempids
- identify tempid indices that share an upsert
- map tempids to a dense set of contiguous integer labels
We use the well-known union-find algorithm, as implemented by
petgraph, to efficiently manage the set of equivalent tempids.
Along the way, I've fixed and added tests for two small errors in the
transactor. First, don't drop datoms resolved by upsert (#679).
Second, ensure that complex upserts are allocated.
I don't know quite what happened here. The Clojure implementation
correctly kept complex upserts that hadn't resolved as complex
upserts (see
https://github.com/mozilla/mentat/blob/9a9dfb502acf5e4cdb1059d4aac831d7603063c8/src/common/datomish/transact.cljc#L436)
and then allocated complex upserts if they didn't resolve (see
https://github.com/mozilla/mentat/blob/9a9dfb502acf5e4cdb1059d4aac831d7603063c8/src/common/datomish/transact.cljc#L509).
Based on the code comments, I think the Rust implementation must have
incorrectly tried to optimize by handling all complex upserts in at
most a single generation of evolution, and that's just not correct.
We're effectively implementing a topological sort, using very specific
domain knowledge, and its not true that a node in a topological sort
can be considered only once!
2018-04-30 22:16:05 +00:00
[ :db / add " foo " :page / id " id " ]
[ :db / add " bar " :db / ident :bar / bar ]
{ :db / id " baz " :page / id " id " :db / ident :bar / bar }
2019-07-17 15:46:47 +00:00
] " #
) ;
assert_matches! (
tempids ( & report ) ,
" { \" foo \" ?e
Add type checking and constraint checking to the transactor. (#663, #532, #679)
This should address #663, by re-inserting type checking in the
transactor stack after the entry point used by the term builder.
Before this commit, we were using an SQLite UNIQUE index to assert
that no `[e a]` pair, with `a` a cardinality one attribute, was
asserted more than once. However, that's not in line with Datomic,
which treats transaction inputs as a set and allows a single datom
like `[e a v]` to appear multiple times. It's both awkward and not
particularly efficient to look for _distinct_ repetitions in SQL, so
we accept some runtime cost in order to check for repetitions in the
transactor. This will allow us to address #532, which is really about
whether we treat inputs as sets. A side benefit is that we can
provide more helpful error messages when the transactor does detect
that the input truly violates the cardinality constraints of the
schema.
This commit builds a trie while error checking and collecting final
terms, which should be fairly efficient. It also allows a simpler
expression of input-provided :db/txInstant datoms, which in turn
uncovered a small issue with the transaction watcher, where-by the
watcher would not see non-input-provided :db/txInstant datoms.
This transition to Datomic-like input-as-set semantics allows us to
address #532. Previously, two tempids that upserted to the same entid
would produce duplicate datoms, and that would have been rejected by
the transactor -- correctly, since we did not allow duplicate datoms
under the input-as-list semantics. With input-as-set semantics,
duplicate datoms are allowed; and that means that we must allow
tempids to be equivalent, i.e., to resolve to the same tempid.
To achieve this, we:
- index the set of tempids
- identify tempid indices that share an upsert
- map tempids to a dense set of contiguous integer labels
We use the well-known union-find algorithm, as implemented by
petgraph, to efficiently manage the set of equivalent tempids.
Along the way, I've fixed and added tests for two small errors in the
transactor. First, don't drop datoms resolved by upsert (#679).
Second, ensure that complex upserts are allocated.
I don't know quite what happened here. The Clojure implementation
correctly kept complex upserts that hadn't resolved as complex
upserts (see
https://github.com/mozilla/mentat/blob/9a9dfb502acf5e4cdb1059d4aac831d7603063c8/src/common/datomish/transact.cljc#L436)
and then allocated complex upserts if they didn't resolve (see
https://github.com/mozilla/mentat/blob/9a9dfb502acf5e4cdb1059d4aac831d7603063c8/src/common/datomish/transact.cljc#L509).
Based on the code comments, I think the Rust implementation must have
incorrectly tried to optimize by handling all complex upserts in at
most a single generation of evolution, and that's just not correct.
We're effectively implementing a topological sort, using very specific
domain knowledge, and its not true that a node in a topological sort
can be considered only once!
2018-04-30 22:16:05 +00:00
\ " bar \" ?e
2019-07-17 15:46:47 +00:00
\ " baz \" ?e} "
) ;
Add type checking and constraint checking to the transactor. (#663, #532, #679)
This should address #663, by re-inserting type checking in the
transactor stack after the entry point used by the term builder.
Before this commit, we were using an SQLite UNIQUE index to assert
that no `[e a]` pair, with `a` a cardinality one attribute, was
asserted more than once. However, that's not in line with Datomic,
which treats transaction inputs as a set and allows a single datom
like `[e a v]` to appear multiple times. It's both awkward and not
particularly efficient to look for _distinct_ repetitions in SQL, so
we accept some runtime cost in order to check for repetitions in the
transactor. This will allow us to address #532, which is really about
whether we treat inputs as sets. A side benefit is that we can
provide more helpful error messages when the transactor does detect
that the input truly violates the cardinality constraints of the
schema.
This commit builds a trie while error checking and collecting final
terms, which should be fairly efficient. It also allows a simpler
expression of input-provided :db/txInstant datoms, which in turn
uncovered a small issue with the transaction watcher, where-by the
watcher would not see non-input-provided :db/txInstant datoms.
This transition to Datomic-like input-as-set semantics allows us to
address #532. Previously, two tempids that upserted to the same entid
would produce duplicate datoms, and that would have been rejected by
the transactor -- correctly, since we did not allow duplicate datoms
under the input-as-list semantics. With input-as-set semantics,
duplicate datoms are allowed; and that means that we must allow
tempids to be equivalent, i.e., to resolve to the same tempid.
To achieve this, we:
- index the set of tempids
- identify tempid indices that share an upsert
- map tempids to a dense set of contiguous integer labels
We use the well-known union-find algorithm, as implemented by
petgraph, to efficiently manage the set of equivalent tempids.
Along the way, I've fixed and added tests for two small errors in the
transactor. First, don't drop datoms resolved by upsert (#679).
Second, ensure that complex upserts are allocated.
I don't know quite what happened here. The Clojure implementation
correctly kept complex upserts that hadn't resolved as complex
upserts (see
https://github.com/mozilla/mentat/blob/9a9dfb502acf5e4cdb1059d4aac831d7603063c8/src/common/datomish/transact.cljc#L436)
and then allocated complex upserts if they didn't resolve (see
https://github.com/mozilla/mentat/blob/9a9dfb502acf5e4cdb1059d4aac831d7603063c8/src/common/datomish/transact.cljc#L509).
Based on the code comments, I think the Rust implementation must have
incorrectly tried to optimize by handling all complex upserts in at
most a single generation of evolution, and that's just not correct.
We're effectively implementing a topological sort, using very specific
domain knowledge, and its not true that a node in a topological sort
can be considered only once!
2018-04-30 22:16:05 +00:00
2019-07-17 15:46:47 +00:00
assert_matches! (
conn . last_transaction ( ) ,
" [[?e :db/ident :bar/bar ?tx true]
Add type checking and constraint checking to the transactor. (#663, #532, #679)
This should address #663, by re-inserting type checking in the
transactor stack after the entry point used by the term builder.
Before this commit, we were using an SQLite UNIQUE index to assert
that no `[e a]` pair, with `a` a cardinality one attribute, was
asserted more than once. However, that's not in line with Datomic,
which treats transaction inputs as a set and allows a single datom
like `[e a v]` to appear multiple times. It's both awkward and not
particularly efficient to look for _distinct_ repetitions in SQL, so
we accept some runtime cost in order to check for repetitions in the
transactor. This will allow us to address #532, which is really about
whether we treat inputs as sets. A side benefit is that we can
provide more helpful error messages when the transactor does detect
that the input truly violates the cardinality constraints of the
schema.
This commit builds a trie while error checking and collecting final
terms, which should be fairly efficient. It also allows a simpler
expression of input-provided :db/txInstant datoms, which in turn
uncovered a small issue with the transaction watcher, where-by the
watcher would not see non-input-provided :db/txInstant datoms.
This transition to Datomic-like input-as-set semantics allows us to
address #532. Previously, two tempids that upserted to the same entid
would produce duplicate datoms, and that would have been rejected by
the transactor -- correctly, since we did not allow duplicate datoms
under the input-as-list semantics. With input-as-set semantics,
duplicate datoms are allowed; and that means that we must allow
tempids to be equivalent, i.e., to resolve to the same tempid.
To achieve this, we:
- index the set of tempids
- identify tempid indices that share an upsert
- map tempids to a dense set of contiguous integer labels
We use the well-known union-find algorithm, as implemented by
petgraph, to efficiently manage the set of equivalent tempids.
Along the way, I've fixed and added tests for two small errors in the
transactor. First, don't drop datoms resolved by upsert (#679).
Second, ensure that complex upserts are allocated.
I don't know quite what happened here. The Clojure implementation
correctly kept complex upserts that hadn't resolved as complex
upserts (see
https://github.com/mozilla/mentat/blob/9a9dfb502acf5e4cdb1059d4aac831d7603063c8/src/common/datomish/transact.cljc#L436)
and then allocated complex upserts if they didn't resolve (see
https://github.com/mozilla/mentat/blob/9a9dfb502acf5e4cdb1059d4aac831d7603063c8/src/common/datomish/transact.cljc#L509).
Based on the code comments, I think the Rust implementation must have
incorrectly tried to optimize by handling all complex upserts in at
most a single generation of evolution, and that's just not correct.
We're effectively implementing a topological sort, using very specific
domain knowledge, and its not true that a node in a topological sort
can be considered only once!
2018-04-30 22:16:05 +00:00
[ ? e :page / id \ " id \" ?tx true]
2019-07-17 15:46:47 +00:00
[ ? tx :db / txInstant ? ms ? tx true ] ] "
) ;
Add type checking and constraint checking to the transactor. (#663, #532, #679)
This should address #663, by re-inserting type checking in the
transactor stack after the entry point used by the term builder.
Before this commit, we were using an SQLite UNIQUE index to assert
that no `[e a]` pair, with `a` a cardinality one attribute, was
asserted more than once. However, that's not in line with Datomic,
which treats transaction inputs as a set and allows a single datom
like `[e a v]` to appear multiple times. It's both awkward and not
particularly efficient to look for _distinct_ repetitions in SQL, so
we accept some runtime cost in order to check for repetitions in the
transactor. This will allow us to address #532, which is really about
whether we treat inputs as sets. A side benefit is that we can
provide more helpful error messages when the transactor does detect
that the input truly violates the cardinality constraints of the
schema.
This commit builds a trie while error checking and collecting final
terms, which should be fairly efficient. It also allows a simpler
expression of input-provided :db/txInstant datoms, which in turn
uncovered a small issue with the transaction watcher, where-by the
watcher would not see non-input-provided :db/txInstant datoms.
This transition to Datomic-like input-as-set semantics allows us to
address #532. Previously, two tempids that upserted to the same entid
would produce duplicate datoms, and that would have been rejected by
the transactor -- correctly, since we did not allow duplicate datoms
under the input-as-list semantics. With input-as-set semantics,
duplicate datoms are allowed; and that means that we must allow
tempids to be equivalent, i.e., to resolve to the same tempid.
To achieve this, we:
- index the set of tempids
- identify tempid indices that share an upsert
- map tempids to a dense set of contiguous integer labels
We use the well-known union-find algorithm, as implemented by
petgraph, to efficiently manage the set of equivalent tempids.
Along the way, I've fixed and added tests for two small errors in the
transactor. First, don't drop datoms resolved by upsert (#679).
Second, ensure that complex upserts are allocated.
I don't know quite what happened here. The Clojure implementation
correctly kept complex upserts that hadn't resolved as complex
upserts (see
https://github.com/mozilla/mentat/blob/9a9dfb502acf5e4cdb1059d4aac831d7603063c8/src/common/datomish/transact.cljc#L436)
and then allocated complex upserts if they didn't resolve (see
https://github.com/mozilla/mentat/blob/9a9dfb502acf5e4cdb1059d4aac831d7603063c8/src/common/datomish/transact.cljc#L509).
Based on the code comments, I think the Rust implementation must have
incorrectly tried to optimize by handling all complex upserts in at
most a single generation of evolution, and that's just not correct.
We're effectively implementing a topological sort, using very specific
domain knowledge, and its not true that a node in a topological sort
can be considered only once!
2018-04-30 22:16:05 +00:00
// If we do it again, everything resolves to the same IDs.
2019-07-17 15:46:47 +00:00
let report = assert_transact! (
conn ,
r #" [
Add type checking and constraint checking to the transactor. (#663, #532, #679)
This should address #663, by re-inserting type checking in the
transactor stack after the entry point used by the term builder.
Before this commit, we were using an SQLite UNIQUE index to assert
that no `[e a]` pair, with `a` a cardinality one attribute, was
asserted more than once. However, that's not in line with Datomic,
which treats transaction inputs as a set and allows a single datom
like `[e a v]` to appear multiple times. It's both awkward and not
particularly efficient to look for _distinct_ repetitions in SQL, so
we accept some runtime cost in order to check for repetitions in the
transactor. This will allow us to address #532, which is really about
whether we treat inputs as sets. A side benefit is that we can
provide more helpful error messages when the transactor does detect
that the input truly violates the cardinality constraints of the
schema.
This commit builds a trie while error checking and collecting final
terms, which should be fairly efficient. It also allows a simpler
expression of input-provided :db/txInstant datoms, which in turn
uncovered a small issue with the transaction watcher, where-by the
watcher would not see non-input-provided :db/txInstant datoms.
This transition to Datomic-like input-as-set semantics allows us to
address #532. Previously, two tempids that upserted to the same entid
would produce duplicate datoms, and that would have been rejected by
the transactor -- correctly, since we did not allow duplicate datoms
under the input-as-list semantics. With input-as-set semantics,
duplicate datoms are allowed; and that means that we must allow
tempids to be equivalent, i.e., to resolve to the same tempid.
To achieve this, we:
- index the set of tempids
- identify tempid indices that share an upsert
- map tempids to a dense set of contiguous integer labels
We use the well-known union-find algorithm, as implemented by
petgraph, to efficiently manage the set of equivalent tempids.
Along the way, I've fixed and added tests for two small errors in the
transactor. First, don't drop datoms resolved by upsert (#679).
Second, ensure that complex upserts are allocated.
I don't know quite what happened here. The Clojure implementation
correctly kept complex upserts that hadn't resolved as complex
upserts (see
https://github.com/mozilla/mentat/blob/9a9dfb502acf5e4cdb1059d4aac831d7603063c8/src/common/datomish/transact.cljc#L436)
and then allocated complex upserts if they didn't resolve (see
https://github.com/mozilla/mentat/blob/9a9dfb502acf5e4cdb1059d4aac831d7603063c8/src/common/datomish/transact.cljc#L509).
Based on the code comments, I think the Rust implementation must have
incorrectly tried to optimize by handling all complex upserts in at
most a single generation of evolution, and that's just not correct.
We're effectively implementing a topological sort, using very specific
domain knowledge, and its not true that a node in a topological sort
can be considered only once!
2018-04-30 22:16:05 +00:00
[ :db / add " foo " :page / id " id " ]
[ :db / add " bar " :db / ident :bar / bar ]
{ :db / id " baz " :page / id " id " :db / ident :bar / bar }
2019-07-17 15:46:47 +00:00
] " #
) ;
assert_matches! (
tempids ( & report ) ,
" { \" foo \" ?e
Add type checking and constraint checking to the transactor. (#663, #532, #679)
This should address #663, by re-inserting type checking in the
transactor stack after the entry point used by the term builder.
Before this commit, we were using an SQLite UNIQUE index to assert
that no `[e a]` pair, with `a` a cardinality one attribute, was
asserted more than once. However, that's not in line with Datomic,
which treats transaction inputs as a set and allows a single datom
like `[e a v]` to appear multiple times. It's both awkward and not
particularly efficient to look for _distinct_ repetitions in SQL, so
we accept some runtime cost in order to check for repetitions in the
transactor. This will allow us to address #532, which is really about
whether we treat inputs as sets. A side benefit is that we can
provide more helpful error messages when the transactor does detect
that the input truly violates the cardinality constraints of the
schema.
This commit builds a trie while error checking and collecting final
terms, which should be fairly efficient. It also allows a simpler
expression of input-provided :db/txInstant datoms, which in turn
uncovered a small issue with the transaction watcher, where-by the
watcher would not see non-input-provided :db/txInstant datoms.
This transition to Datomic-like input-as-set semantics allows us to
address #532. Previously, two tempids that upserted to the same entid
would produce duplicate datoms, and that would have been rejected by
the transactor -- correctly, since we did not allow duplicate datoms
under the input-as-list semantics. With input-as-set semantics,
duplicate datoms are allowed; and that means that we must allow
tempids to be equivalent, i.e., to resolve to the same tempid.
To achieve this, we:
- index the set of tempids
- identify tempid indices that share an upsert
- map tempids to a dense set of contiguous integer labels
We use the well-known union-find algorithm, as implemented by
petgraph, to efficiently manage the set of equivalent tempids.
Along the way, I've fixed and added tests for two small errors in the
transactor. First, don't drop datoms resolved by upsert (#679).
Second, ensure that complex upserts are allocated.
I don't know quite what happened here. The Clojure implementation
correctly kept complex upserts that hadn't resolved as complex
upserts (see
https://github.com/mozilla/mentat/blob/9a9dfb502acf5e4cdb1059d4aac831d7603063c8/src/common/datomish/transact.cljc#L436)
and then allocated complex upserts if they didn't resolve (see
https://github.com/mozilla/mentat/blob/9a9dfb502acf5e4cdb1059d4aac831d7603063c8/src/common/datomish/transact.cljc#L509).
Based on the code comments, I think the Rust implementation must have
incorrectly tried to optimize by handling all complex upserts in at
most a single generation of evolution, and that's just not correct.
We're effectively implementing a topological sort, using very specific
domain knowledge, and its not true that a node in a topological sort
can be considered only once!
2018-04-30 22:16:05 +00:00
\ " bar \" ?e
2019-07-17 15:46:47 +00:00
\ " baz \" ?e} "
) ;
Add type checking and constraint checking to the transactor. (#663, #532, #679)
This should address #663, by re-inserting type checking in the
transactor stack after the entry point used by the term builder.
Before this commit, we were using an SQLite UNIQUE index to assert
that no `[e a]` pair, with `a` a cardinality one attribute, was
asserted more than once. However, that's not in line with Datomic,
which treats transaction inputs as a set and allows a single datom
like `[e a v]` to appear multiple times. It's both awkward and not
particularly efficient to look for _distinct_ repetitions in SQL, so
we accept some runtime cost in order to check for repetitions in the
transactor. This will allow us to address #532, which is really about
whether we treat inputs as sets. A side benefit is that we can
provide more helpful error messages when the transactor does detect
that the input truly violates the cardinality constraints of the
schema.
This commit builds a trie while error checking and collecting final
terms, which should be fairly efficient. It also allows a simpler
expression of input-provided :db/txInstant datoms, which in turn
uncovered a small issue with the transaction watcher, where-by the
watcher would not see non-input-provided :db/txInstant datoms.
This transition to Datomic-like input-as-set semantics allows us to
address #532. Previously, two tempids that upserted to the same entid
would produce duplicate datoms, and that would have been rejected by
the transactor -- correctly, since we did not allow duplicate datoms
under the input-as-list semantics. With input-as-set semantics,
duplicate datoms are allowed; and that means that we must allow
tempids to be equivalent, i.e., to resolve to the same tempid.
To achieve this, we:
- index the set of tempids
- identify tempid indices that share an upsert
- map tempids to a dense set of contiguous integer labels
We use the well-known union-find algorithm, as implemented by
petgraph, to efficiently manage the set of equivalent tempids.
Along the way, I've fixed and added tests for two small errors in the
transactor. First, don't drop datoms resolved by upsert (#679).
Second, ensure that complex upserts are allocated.
I don't know quite what happened here. The Clojure implementation
correctly kept complex upserts that hadn't resolved as complex
upserts (see
https://github.com/mozilla/mentat/blob/9a9dfb502acf5e4cdb1059d4aac831d7603063c8/src/common/datomish/transact.cljc#L436)
and then allocated complex upserts if they didn't resolve (see
https://github.com/mozilla/mentat/blob/9a9dfb502acf5e4cdb1059d4aac831d7603063c8/src/common/datomish/transact.cljc#L509).
Based on the code comments, I think the Rust implementation must have
incorrectly tried to optimize by handling all complex upserts in at
most a single generation of evolution, and that's just not correct.
We're effectively implementing a topological sort, using very specific
domain knowledge, and its not true that a node in a topological sort
can be considered only once!
2018-04-30 22:16:05 +00:00
2019-07-17 15:46:47 +00:00
assert_matches! (
conn . last_transaction ( ) ,
" [[?tx :db/txInstant ?ms ?tx true]] "
) ;
Add type checking and constraint checking to the transactor. (#663, #532, #679)
This should address #663, by re-inserting type checking in the
transactor stack after the entry point used by the term builder.
Before this commit, we were using an SQLite UNIQUE index to assert
that no `[e a]` pair, with `a` a cardinality one attribute, was
asserted more than once. However, that's not in line with Datomic,
which treats transaction inputs as a set and allows a single datom
like `[e a v]` to appear multiple times. It's both awkward and not
particularly efficient to look for _distinct_ repetitions in SQL, so
we accept some runtime cost in order to check for repetitions in the
transactor. This will allow us to address #532, which is really about
whether we treat inputs as sets. A side benefit is that we can
provide more helpful error messages when the transactor does detect
that the input truly violates the cardinality constraints of the
schema.
This commit builds a trie while error checking and collecting final
terms, which should be fairly efficient. It also allows a simpler
expression of input-provided :db/txInstant datoms, which in turn
uncovered a small issue with the transaction watcher, where-by the
watcher would not see non-input-provided :db/txInstant datoms.
This transition to Datomic-like input-as-set semantics allows us to
address #532. Previously, two tempids that upserted to the same entid
would produce duplicate datoms, and that would have been rejected by
the transactor -- correctly, since we did not allow duplicate datoms
under the input-as-list semantics. With input-as-set semantics,
duplicate datoms are allowed; and that means that we must allow
tempids to be equivalent, i.e., to resolve to the same tempid.
To achieve this, we:
- index the set of tempids
- identify tempid indices that share an upsert
- map tempids to a dense set of contiguous integer labels
We use the well-known union-find algorithm, as implemented by
petgraph, to efficiently manage the set of equivalent tempids.
Along the way, I've fixed and added tests for two small errors in the
transactor. First, don't drop datoms resolved by upsert (#679).
Second, ensure that complex upserts are allocated.
I don't know quite what happened here. The Clojure implementation
correctly kept complex upserts that hadn't resolved as complex
upserts (see
https://github.com/mozilla/mentat/blob/9a9dfb502acf5e4cdb1059d4aac831d7603063c8/src/common/datomish/transact.cljc#L436)
and then allocated complex upserts if they didn't resolve (see
https://github.com/mozilla/mentat/blob/9a9dfb502acf5e4cdb1059d4aac831d7603063c8/src/common/datomish/transact.cljc#L509).
Based on the code comments, I think the Rust implementation must have
incorrectly tried to optimize by handling all complex upserts in at
most a single generation of evolution, and that's just not correct.
We're effectively implementing a topological sort, using very specific
domain knowledge, and its not true that a node in a topological sort
can be considered only once!
2018-04-30 22:16:05 +00:00
}
#[ test ]
fn test_term_typechecking_issue_663 ( ) {
// The builder interfaces provide untrusted `Term` instances to the transactor, bypassing
// the typechecking layers invoked in the schema-aware coercion from `edn::Value` into
// `TypedValue`. Typechecking now happens lower in the stack (as well as higher in the
// stack) so we shouldn't be able to insert bad data into the store.
let mut conn = TestConn ::default ( ) ;
let mut terms = vec! [ ] ;
2019-07-17 15:46:47 +00:00
terms . push ( Term ::AddOrRetract (
OpType ::Add ,
Left ( KnownEntid ( 200 ) ) ,
entids ::DB_IDENT ,
Left ( TypedValue ::typed_string ( " test " ) ) ,
) ) ;
terms . push ( Term ::AddOrRetract (
OpType ::Retract ,
Left ( KnownEntid ( 100 ) ) ,
entids ::DB_TX_INSTANT ,
Left ( TypedValue ::Long ( - 1 ) ) ,
) ) ;
Add type checking and constraint checking to the transactor. (#663, #532, #679)
This should address #663, by re-inserting type checking in the
transactor stack after the entry point used by the term builder.
Before this commit, we were using an SQLite UNIQUE index to assert
that no `[e a]` pair, with `a` a cardinality one attribute, was
asserted more than once. However, that's not in line with Datomic,
which treats transaction inputs as a set and allows a single datom
like `[e a v]` to appear multiple times. It's both awkward and not
particularly efficient to look for _distinct_ repetitions in SQL, so
we accept some runtime cost in order to check for repetitions in the
transactor. This will allow us to address #532, which is really about
whether we treat inputs as sets. A side benefit is that we can
provide more helpful error messages when the transactor does detect
that the input truly violates the cardinality constraints of the
schema.
This commit builds a trie while error checking and collecting final
terms, which should be fairly efficient. It also allows a simpler
expression of input-provided :db/txInstant datoms, which in turn
uncovered a small issue with the transaction watcher, where-by the
watcher would not see non-input-provided :db/txInstant datoms.
This transition to Datomic-like input-as-set semantics allows us to
address #532. Previously, two tempids that upserted to the same entid
would produce duplicate datoms, and that would have been rejected by
the transactor -- correctly, since we did not allow duplicate datoms
under the input-as-list semantics. With input-as-set semantics,
duplicate datoms are allowed; and that means that we must allow
tempids to be equivalent, i.e., to resolve to the same tempid.
To achieve this, we:
- index the set of tempids
- identify tempid indices that share an upsert
- map tempids to a dense set of contiguous integer labels
We use the well-known union-find algorithm, as implemented by
petgraph, to efficiently manage the set of equivalent tempids.
Along the way, I've fixed and added tests for two small errors in the
transactor. First, don't drop datoms resolved by upsert (#679).
Second, ensure that complex upserts are allocated.
I don't know quite what happened here. The Clojure implementation
correctly kept complex upserts that hadn't resolved as complex
upserts (see
https://github.com/mozilla/mentat/blob/9a9dfb502acf5e4cdb1059d4aac831d7603063c8/src/common/datomish/transact.cljc#L436)
and then allocated complex upserts if they didn't resolve (see
https://github.com/mozilla/mentat/blob/9a9dfb502acf5e4cdb1059d4aac831d7603063c8/src/common/datomish/transact.cljc#L509).
Based on the code comments, I think the Rust implementation must have
incorrectly tried to optimize by handling all complex upserts in at
most a single generation of evolution, and that's just not correct.
We're effectively implementing a topological sort, using very specific
domain knowledge, and its not true that a node in a topological sort
can be considered only once!
2018-04-30 22:16:05 +00:00
let report = conn . transact_simple_terms ( terms , InternSet ::new ( ) ) ;
2018-06-27 00:17:01 +00:00
match report . err ( ) . map ( | e | e . kind ( ) ) {
2019-07-17 15:46:47 +00:00
Some ( DbErrorKind ::SchemaConstraintViolation (
errors ::SchemaConstraintViolation ::TypeDisagreements {
ref conflicting_datoms ,
} ,
) ) = > {
Add type checking and constraint checking to the transactor. (#663, #532, #679)
This should address #663, by re-inserting type checking in the
transactor stack after the entry point used by the term builder.
Before this commit, we were using an SQLite UNIQUE index to assert
that no `[e a]` pair, with `a` a cardinality one attribute, was
asserted more than once. However, that's not in line with Datomic,
which treats transaction inputs as a set and allows a single datom
like `[e a v]` to appear multiple times. It's both awkward and not
particularly efficient to look for _distinct_ repetitions in SQL, so
we accept some runtime cost in order to check for repetitions in the
transactor. This will allow us to address #532, which is really about
whether we treat inputs as sets. A side benefit is that we can
provide more helpful error messages when the transactor does detect
that the input truly violates the cardinality constraints of the
schema.
This commit builds a trie while error checking and collecting final
terms, which should be fairly efficient. It also allows a simpler
expression of input-provided :db/txInstant datoms, which in turn
uncovered a small issue with the transaction watcher, where-by the
watcher would not see non-input-provided :db/txInstant datoms.
This transition to Datomic-like input-as-set semantics allows us to
address #532. Previously, two tempids that upserted to the same entid
would produce duplicate datoms, and that would have been rejected by
the transactor -- correctly, since we did not allow duplicate datoms
under the input-as-list semantics. With input-as-set semantics,
duplicate datoms are allowed; and that means that we must allow
tempids to be equivalent, i.e., to resolve to the same tempid.
To achieve this, we:
- index the set of tempids
- identify tempid indices that share an upsert
- map tempids to a dense set of contiguous integer labels
We use the well-known union-find algorithm, as implemented by
petgraph, to efficiently manage the set of equivalent tempids.
Along the way, I've fixed and added tests for two small errors in the
transactor. First, don't drop datoms resolved by upsert (#679).
Second, ensure that complex upserts are allocated.
I don't know quite what happened here. The Clojure implementation
correctly kept complex upserts that hadn't resolved as complex
upserts (see
https://github.com/mozilla/mentat/blob/9a9dfb502acf5e4cdb1059d4aac831d7603063c8/src/common/datomish/transact.cljc#L436)
and then allocated complex upserts if they didn't resolve (see
https://github.com/mozilla/mentat/blob/9a9dfb502acf5e4cdb1059d4aac831d7603063c8/src/common/datomish/transact.cljc#L509).
Based on the code comments, I think the Rust implementation must have
incorrectly tried to optimize by handling all complex upserts in at
most a single generation of evolution, and that's just not correct.
We're effectively implementing a topological sort, using very specific
domain knowledge, and its not true that a node in a topological sort
can be considered only once!
2018-04-30 22:16:05 +00:00
let mut map = BTreeMap ::default ( ) ;
2019-07-17 15:46:47 +00:00
map . insert (
( 100 , entids ::DB_TX_INSTANT , TypedValue ::Long ( - 1 ) ) ,
ValueType ::Instant ,
) ;
map . insert (
( 200 , entids ::DB_IDENT , TypedValue ::typed_string ( " test " ) ) ,
ValueType ::Keyword ,
) ;
Add type checking and constraint checking to the transactor. (#663, #532, #679)
This should address #663, by re-inserting type checking in the
transactor stack after the entry point used by the term builder.
Before this commit, we were using an SQLite UNIQUE index to assert
that no `[e a]` pair, with `a` a cardinality one attribute, was
asserted more than once. However, that's not in line with Datomic,
which treats transaction inputs as a set and allows a single datom
like `[e a v]` to appear multiple times. It's both awkward and not
particularly efficient to look for _distinct_ repetitions in SQL, so
we accept some runtime cost in order to check for repetitions in the
transactor. This will allow us to address #532, which is really about
whether we treat inputs as sets. A side benefit is that we can
provide more helpful error messages when the transactor does detect
that the input truly violates the cardinality constraints of the
schema.
This commit builds a trie while error checking and collecting final
terms, which should be fairly efficient. It also allows a simpler
expression of input-provided :db/txInstant datoms, which in turn
uncovered a small issue with the transaction watcher, where-by the
watcher would not see non-input-provided :db/txInstant datoms.
This transition to Datomic-like input-as-set semantics allows us to
address #532. Previously, two tempids that upserted to the same entid
would produce duplicate datoms, and that would have been rejected by
the transactor -- correctly, since we did not allow duplicate datoms
under the input-as-list semantics. With input-as-set semantics,
duplicate datoms are allowed; and that means that we must allow
tempids to be equivalent, i.e., to resolve to the same tempid.
To achieve this, we:
- index the set of tempids
- identify tempid indices that share an upsert
- map tempids to a dense set of contiguous integer labels
We use the well-known union-find algorithm, as implemented by
petgraph, to efficiently manage the set of equivalent tempids.
Along the way, I've fixed and added tests for two small errors in the
transactor. First, don't drop datoms resolved by upsert (#679).
Second, ensure that complex upserts are allocated.
I don't know quite what happened here. The Clojure implementation
correctly kept complex upserts that hadn't resolved as complex
upserts (see
https://github.com/mozilla/mentat/blob/9a9dfb502acf5e4cdb1059d4aac831d7603063c8/src/common/datomish/transact.cljc#L436)
and then allocated complex upserts if they didn't resolve (see
https://github.com/mozilla/mentat/blob/9a9dfb502acf5e4cdb1059d4aac831d7603063c8/src/common/datomish/transact.cljc#L509).
Based on the code comments, I think the Rust implementation must have
incorrectly tried to optimize by handling all complex upserts in at
most a single generation of evolution, and that's just not correct.
We're effectively implementing a topological sort, using very specific
domain knowledge, and its not true that a node in a topological sort
can be considered only once!
2018-04-30 22:16:05 +00:00
2018-06-27 00:17:01 +00:00
assert_eq! ( conflicting_datoms , & map ) ;
2019-07-17 15:46:47 +00:00
}
Add type checking and constraint checking to the transactor. (#663, #532, #679)
This should address #663, by re-inserting type checking in the
transactor stack after the entry point used by the term builder.
Before this commit, we were using an SQLite UNIQUE index to assert
that no `[e a]` pair, with `a` a cardinality one attribute, was
asserted more than once. However, that's not in line with Datomic,
which treats transaction inputs as a set and allows a single datom
like `[e a v]` to appear multiple times. It's both awkward and not
particularly efficient to look for _distinct_ repetitions in SQL, so
we accept some runtime cost in order to check for repetitions in the
transactor. This will allow us to address #532, which is really about
whether we treat inputs as sets. A side benefit is that we can
provide more helpful error messages when the transactor does detect
that the input truly violates the cardinality constraints of the
schema.
This commit builds a trie while error checking and collecting final
terms, which should be fairly efficient. It also allows a simpler
expression of input-provided :db/txInstant datoms, which in turn
uncovered a small issue with the transaction watcher, where-by the
watcher would not see non-input-provided :db/txInstant datoms.
This transition to Datomic-like input-as-set semantics allows us to
address #532. Previously, two tempids that upserted to the same entid
would produce duplicate datoms, and that would have been rejected by
the transactor -- correctly, since we did not allow duplicate datoms
under the input-as-list semantics. With input-as-set semantics,
duplicate datoms are allowed; and that means that we must allow
tempids to be equivalent, i.e., to resolve to the same tempid.
To achieve this, we:
- index the set of tempids
- identify tempid indices that share an upsert
- map tempids to a dense set of contiguous integer labels
We use the well-known union-find algorithm, as implemented by
petgraph, to efficiently manage the set of equivalent tempids.
Along the way, I've fixed and added tests for two small errors in the
transactor. First, don't drop datoms resolved by upsert (#679).
Second, ensure that complex upserts are allocated.
I don't know quite what happened here. The Clojure implementation
correctly kept complex upserts that hadn't resolved as complex
upserts (see
https://github.com/mozilla/mentat/blob/9a9dfb502acf5e4cdb1059d4aac831d7603063c8/src/common/datomish/transact.cljc#L436)
and then allocated complex upserts if they didn't resolve (see
https://github.com/mozilla/mentat/blob/9a9dfb502acf5e4cdb1059d4aac831d7603063c8/src/common/datomish/transact.cljc#L509).
Based on the code comments, I think the Rust implementation must have
incorrectly tried to optimize by handling all complex upserts in at
most a single generation of evolution, and that's just not correct.
We're effectively implementing a topological sort, using very specific
domain knowledge, and its not true that a node in a topological sort
can be considered only once!
2018-04-30 22:16:05 +00:00
x = > panic! ( " expected schema constraint violation, got {:?} " , x ) ,
}
}
#[ test ]
fn test_cardinality_constraints ( ) {
let mut conn = TestConn ::default ( ) ;
2020-01-14 15:46:21 +00:00
assert_transact! (
conn ,
r #" [
Add type checking and constraint checking to the transactor. (#663, #532, #679)
This should address #663, by re-inserting type checking in the
transactor stack after the entry point used by the term builder.
Before this commit, we were using an SQLite UNIQUE index to assert
that no `[e a]` pair, with `a` a cardinality one attribute, was
asserted more than once. However, that's not in line with Datomic,
which treats transaction inputs as a set and allows a single datom
like `[e a v]` to appear multiple times. It's both awkward and not
particularly efficient to look for _distinct_ repetitions in SQL, so
we accept some runtime cost in order to check for repetitions in the
transactor. This will allow us to address #532, which is really about
whether we treat inputs as sets. A side benefit is that we can
provide more helpful error messages when the transactor does detect
that the input truly violates the cardinality constraints of the
schema.
This commit builds a trie while error checking and collecting final
terms, which should be fairly efficient. It also allows a simpler
expression of input-provided :db/txInstant datoms, which in turn
uncovered a small issue with the transaction watcher, where-by the
watcher would not see non-input-provided :db/txInstant datoms.
This transition to Datomic-like input-as-set semantics allows us to
address #532. Previously, two tempids that upserted to the same entid
would produce duplicate datoms, and that would have been rejected by
the transactor -- correctly, since we did not allow duplicate datoms
under the input-as-list semantics. With input-as-set semantics,
duplicate datoms are allowed; and that means that we must allow
tempids to be equivalent, i.e., to resolve to the same tempid.
To achieve this, we:
- index the set of tempids
- identify tempid indices that share an upsert
- map tempids to a dense set of contiguous integer labels
We use the well-known union-find algorithm, as implemented by
petgraph, to efficiently manage the set of equivalent tempids.
Along the way, I've fixed and added tests for two small errors in the
transactor. First, don't drop datoms resolved by upsert (#679).
Second, ensure that complex upserts are allocated.
I don't know quite what happened here. The Clojure implementation
correctly kept complex upserts that hadn't resolved as complex
upserts (see
https://github.com/mozilla/mentat/blob/9a9dfb502acf5e4cdb1059d4aac831d7603063c8/src/common/datomish/transact.cljc#L436)
and then allocated complex upserts if they didn't resolve (see
https://github.com/mozilla/mentat/blob/9a9dfb502acf5e4cdb1059d4aac831d7603063c8/src/common/datomish/transact.cljc#L509).
Based on the code comments, I think the Rust implementation must have
incorrectly tried to optimize by handling all complex upserts in at
most a single generation of evolution, and that's just not correct.
We're effectively implementing a topological sort, using very specific
domain knowledge, and its not true that a node in a topological sort
can be considered only once!
2018-04-30 22:16:05 +00:00
{ :db / id 200 :db / ident :test / one :db / valueType :db . type / long :db / cardinality :db . cardinality / one }
{ :db / id 201 :db / ident :test / many :db / valueType :db . type / long :db / cardinality :db . cardinality / many }
2020-01-14 15:46:21 +00:00
] " #
) ;
Add type checking and constraint checking to the transactor. (#663, #532, #679)
This should address #663, by re-inserting type checking in the
transactor stack after the entry point used by the term builder.
Before this commit, we were using an SQLite UNIQUE index to assert
that no `[e a]` pair, with `a` a cardinality one attribute, was
asserted more than once. However, that's not in line with Datomic,
which treats transaction inputs as a set and allows a single datom
like `[e a v]` to appear multiple times. It's both awkward and not
particularly efficient to look for _distinct_ repetitions in SQL, so
we accept some runtime cost in order to check for repetitions in the
transactor. This will allow us to address #532, which is really about
whether we treat inputs as sets. A side benefit is that we can
provide more helpful error messages when the transactor does detect
that the input truly violates the cardinality constraints of the
schema.
This commit builds a trie while error checking and collecting final
terms, which should be fairly efficient. It also allows a simpler
expression of input-provided :db/txInstant datoms, which in turn
uncovered a small issue with the transaction watcher, where-by the
watcher would not see non-input-provided :db/txInstant datoms.
This transition to Datomic-like input-as-set semantics allows us to
address #532. Previously, two tempids that upserted to the same entid
would produce duplicate datoms, and that would have been rejected by
the transactor -- correctly, since we did not allow duplicate datoms
under the input-as-list semantics. With input-as-set semantics,
duplicate datoms are allowed; and that means that we must allow
tempids to be equivalent, i.e., to resolve to the same tempid.
To achieve this, we:
- index the set of tempids
- identify tempid indices that share an upsert
- map tempids to a dense set of contiguous integer labels
We use the well-known union-find algorithm, as implemented by
petgraph, to efficiently manage the set of equivalent tempids.
Along the way, I've fixed and added tests for two small errors in the
transactor. First, don't drop datoms resolved by upsert (#679).
Second, ensure that complex upserts are allocated.
I don't know quite what happened here. The Clojure implementation
correctly kept complex upserts that hadn't resolved as complex
upserts (see
https://github.com/mozilla/mentat/blob/9a9dfb502acf5e4cdb1059d4aac831d7603063c8/src/common/datomish/transact.cljc#L436)
and then allocated complex upserts if they didn't resolve (see
https://github.com/mozilla/mentat/blob/9a9dfb502acf5e4cdb1059d4aac831d7603063c8/src/common/datomish/transact.cljc#L509).
Based on the code comments, I think the Rust implementation must have
incorrectly tried to optimize by handling all complex upserts in at
most a single generation of evolution, and that's just not correct.
We're effectively implementing a topological sort, using very specific
domain knowledge, and its not true that a node in a topological sort
can be considered only once!
2018-04-30 22:16:05 +00:00
// Can add the same datom multiple times for an attribute, regardless of cardinality.
2019-07-17 15:46:47 +00:00
assert_transact! (
conn ,
r #" [
Add type checking and constraint checking to the transactor. (#663, #532, #679)
This should address #663, by re-inserting type checking in the
transactor stack after the entry point used by the term builder.
Before this commit, we were using an SQLite UNIQUE index to assert
that no `[e a]` pair, with `a` a cardinality one attribute, was
asserted more than once. However, that's not in line with Datomic,
which treats transaction inputs as a set and allows a single datom
like `[e a v]` to appear multiple times. It's both awkward and not
particularly efficient to look for _distinct_ repetitions in SQL, so
we accept some runtime cost in order to check for repetitions in the
transactor. This will allow us to address #532, which is really about
whether we treat inputs as sets. A side benefit is that we can
provide more helpful error messages when the transactor does detect
that the input truly violates the cardinality constraints of the
schema.
This commit builds a trie while error checking and collecting final
terms, which should be fairly efficient. It also allows a simpler
expression of input-provided :db/txInstant datoms, which in turn
uncovered a small issue with the transaction watcher, where-by the
watcher would not see non-input-provided :db/txInstant datoms.
This transition to Datomic-like input-as-set semantics allows us to
address #532. Previously, two tempids that upserted to the same entid
would produce duplicate datoms, and that would have been rejected by
the transactor -- correctly, since we did not allow duplicate datoms
under the input-as-list semantics. With input-as-set semantics,
duplicate datoms are allowed; and that means that we must allow
tempids to be equivalent, i.e., to resolve to the same tempid.
To achieve this, we:
- index the set of tempids
- identify tempid indices that share an upsert
- map tempids to a dense set of contiguous integer labels
We use the well-known union-find algorithm, as implemented by
petgraph, to efficiently manage the set of equivalent tempids.
Along the way, I've fixed and added tests for two small errors in the
transactor. First, don't drop datoms resolved by upsert (#679).
Second, ensure that complex upserts are allocated.
I don't know quite what happened here. The Clojure implementation
correctly kept complex upserts that hadn't resolved as complex
upserts (see
https://github.com/mozilla/mentat/blob/9a9dfb502acf5e4cdb1059d4aac831d7603063c8/src/common/datomish/transact.cljc#L436)
and then allocated complex upserts if they didn't resolve (see
https://github.com/mozilla/mentat/blob/9a9dfb502acf5e4cdb1059d4aac831d7603063c8/src/common/datomish/transact.cljc#L509).
Based on the code comments, I think the Rust implementation must have
incorrectly tried to optimize by handling all complex upserts in at
most a single generation of evolution, and that's just not correct.
We're effectively implementing a topological sort, using very specific
domain knowledge, and its not true that a node in a topological sort
can be considered only once!
2018-04-30 22:16:05 +00:00
[ :db / add 100 :test / one 1 ]
[ :db / add 100 :test / one 1 ]
[ :db / add 100 :test / many 2 ]
[ :db / add 100 :test / many 2 ]
2019-07-17 15:46:47 +00:00
] " #
) ;
Add type checking and constraint checking to the transactor. (#663, #532, #679)
This should address #663, by re-inserting type checking in the
transactor stack after the entry point used by the term builder.
Before this commit, we were using an SQLite UNIQUE index to assert
that no `[e a]` pair, with `a` a cardinality one attribute, was
asserted more than once. However, that's not in line with Datomic,
which treats transaction inputs as a set and allows a single datom
like `[e a v]` to appear multiple times. It's both awkward and not
particularly efficient to look for _distinct_ repetitions in SQL, so
we accept some runtime cost in order to check for repetitions in the
transactor. This will allow us to address #532, which is really about
whether we treat inputs as sets. A side benefit is that we can
provide more helpful error messages when the transactor does detect
that the input truly violates the cardinality constraints of the
schema.
This commit builds a trie while error checking and collecting final
terms, which should be fairly efficient. It also allows a simpler
expression of input-provided :db/txInstant datoms, which in turn
uncovered a small issue with the transaction watcher, where-by the
watcher would not see non-input-provided :db/txInstant datoms.
This transition to Datomic-like input-as-set semantics allows us to
address #532. Previously, two tempids that upserted to the same entid
would produce duplicate datoms, and that would have been rejected by
the transactor -- correctly, since we did not allow duplicate datoms
under the input-as-list semantics. With input-as-set semantics,
duplicate datoms are allowed; and that means that we must allow
tempids to be equivalent, i.e., to resolve to the same tempid.
To achieve this, we:
- index the set of tempids
- identify tempid indices that share an upsert
- map tempids to a dense set of contiguous integer labels
We use the well-known union-find algorithm, as implemented by
petgraph, to efficiently manage the set of equivalent tempids.
Along the way, I've fixed and added tests for two small errors in the
transactor. First, don't drop datoms resolved by upsert (#679).
Second, ensure that complex upserts are allocated.
I don't know quite what happened here. The Clojure implementation
correctly kept complex upserts that hadn't resolved as complex
upserts (see
https://github.com/mozilla/mentat/blob/9a9dfb502acf5e4cdb1059d4aac831d7603063c8/src/common/datomish/transact.cljc#L436)
and then allocated complex upserts if they didn't resolve (see
https://github.com/mozilla/mentat/blob/9a9dfb502acf5e4cdb1059d4aac831d7603063c8/src/common/datomish/transact.cljc#L509).
Based on the code comments, I think the Rust implementation must have
incorrectly tried to optimize by handling all complex upserts in at
most a single generation of evolution, and that's just not correct.
We're effectively implementing a topological sort, using very specific
domain knowledge, and its not true that a node in a topological sort
can be considered only once!
2018-04-30 22:16:05 +00:00
// Can retract the same datom multiple times for an attribute, regardless of cardinality.
2019-07-17 15:46:47 +00:00
assert_transact! (
conn ,
r #" [
Add type checking and constraint checking to the transactor. (#663, #532, #679)
This should address #663, by re-inserting type checking in the
transactor stack after the entry point used by the term builder.
Before this commit, we were using an SQLite UNIQUE index to assert
that no `[e a]` pair, with `a` a cardinality one attribute, was
asserted more than once. However, that's not in line with Datomic,
which treats transaction inputs as a set and allows a single datom
like `[e a v]` to appear multiple times. It's both awkward and not
particularly efficient to look for _distinct_ repetitions in SQL, so
we accept some runtime cost in order to check for repetitions in the
transactor. This will allow us to address #532, which is really about
whether we treat inputs as sets. A side benefit is that we can
provide more helpful error messages when the transactor does detect
that the input truly violates the cardinality constraints of the
schema.
This commit builds a trie while error checking and collecting final
terms, which should be fairly efficient. It also allows a simpler
expression of input-provided :db/txInstant datoms, which in turn
uncovered a small issue with the transaction watcher, where-by the
watcher would not see non-input-provided :db/txInstant datoms.
This transition to Datomic-like input-as-set semantics allows us to
address #532. Previously, two tempids that upserted to the same entid
would produce duplicate datoms, and that would have been rejected by
the transactor -- correctly, since we did not allow duplicate datoms
under the input-as-list semantics. With input-as-set semantics,
duplicate datoms are allowed; and that means that we must allow
tempids to be equivalent, i.e., to resolve to the same tempid.
To achieve this, we:
- index the set of tempids
- identify tempid indices that share an upsert
- map tempids to a dense set of contiguous integer labels
We use the well-known union-find algorithm, as implemented by
petgraph, to efficiently manage the set of equivalent tempids.
Along the way, I've fixed and added tests for two small errors in the
transactor. First, don't drop datoms resolved by upsert (#679).
Second, ensure that complex upserts are allocated.
I don't know quite what happened here. The Clojure implementation
correctly kept complex upserts that hadn't resolved as complex
upserts (see
https://github.com/mozilla/mentat/blob/9a9dfb502acf5e4cdb1059d4aac831d7603063c8/src/common/datomish/transact.cljc#L436)
and then allocated complex upserts if they didn't resolve (see
https://github.com/mozilla/mentat/blob/9a9dfb502acf5e4cdb1059d4aac831d7603063c8/src/common/datomish/transact.cljc#L509).
Based on the code comments, I think the Rust implementation must have
incorrectly tried to optimize by handling all complex upserts in at
most a single generation of evolution, and that's just not correct.
We're effectively implementing a topological sort, using very specific
domain knowledge, and its not true that a node in a topological sort
can be considered only once!
2018-04-30 22:16:05 +00:00
[ :db / retract 100 :test / one 1 ]
[ :db / retract 100 :test / one 1 ]
[ :db / retract 100 :test / many 2 ]
[ :db / retract 100 :test / many 2 ]
2019-07-17 15:46:47 +00:00
] " #
) ;
Add type checking and constraint checking to the transactor. (#663, #532, #679)
This should address #663, by re-inserting type checking in the
transactor stack after the entry point used by the term builder.
Before this commit, we were using an SQLite UNIQUE index to assert
that no `[e a]` pair, with `a` a cardinality one attribute, was
asserted more than once. However, that's not in line with Datomic,
which treats transaction inputs as a set and allows a single datom
like `[e a v]` to appear multiple times. It's both awkward and not
particularly efficient to look for _distinct_ repetitions in SQL, so
we accept some runtime cost in order to check for repetitions in the
transactor. This will allow us to address #532, which is really about
whether we treat inputs as sets. A side benefit is that we can
provide more helpful error messages when the transactor does detect
that the input truly violates the cardinality constraints of the
schema.
This commit builds a trie while error checking and collecting final
terms, which should be fairly efficient. It also allows a simpler
expression of input-provided :db/txInstant datoms, which in turn
uncovered a small issue with the transaction watcher, where-by the
watcher would not see non-input-provided :db/txInstant datoms.
This transition to Datomic-like input-as-set semantics allows us to
address #532. Previously, two tempids that upserted to the same entid
would produce duplicate datoms, and that would have been rejected by
the transactor -- correctly, since we did not allow duplicate datoms
under the input-as-list semantics. With input-as-set semantics,
duplicate datoms are allowed; and that means that we must allow
tempids to be equivalent, i.e., to resolve to the same tempid.
To achieve this, we:
- index the set of tempids
- identify tempid indices that share an upsert
- map tempids to a dense set of contiguous integer labels
We use the well-known union-find algorithm, as implemented by
petgraph, to efficiently manage the set of equivalent tempids.
Along the way, I've fixed and added tests for two small errors in the
transactor. First, don't drop datoms resolved by upsert (#679).
Second, ensure that complex upserts are allocated.
I don't know quite what happened here. The Clojure implementation
correctly kept complex upserts that hadn't resolved as complex
upserts (see
https://github.com/mozilla/mentat/blob/9a9dfb502acf5e4cdb1059d4aac831d7603063c8/src/common/datomish/transact.cljc#L436)
and then allocated complex upserts if they didn't resolve (see
https://github.com/mozilla/mentat/blob/9a9dfb502acf5e4cdb1059d4aac831d7603063c8/src/common/datomish/transact.cljc#L509).
Based on the code comments, I think the Rust implementation must have
incorrectly tried to optimize by handling all complex upserts in at
most a single generation of evolution, and that's just not correct.
We're effectively implementing a topological sort, using very specific
domain knowledge, and its not true that a node in a topological sort
can be considered only once!
2018-04-30 22:16:05 +00:00
// Can't transact multiple datoms for a cardinality one attribute.
assert_transact! ( conn , r #" [
[ :db / add 100 :test / one 3 ]
[ :db / add 100 :test / one 4 ]
] " #,
Err ( " schema constraint violation: cardinality conflicts: \n CardinalityOneAddConflict { e: 100, a: 200, vs: {Long(3), Long(4)} } \n " ) ) ;
// Can transact multiple datoms for a cardinality many attribute.
2019-07-17 15:46:47 +00:00
assert_transact! (
conn ,
r #" [
Add type checking and constraint checking to the transactor. (#663, #532, #679)
This should address #663, by re-inserting type checking in the
transactor stack after the entry point used by the term builder.
Before this commit, we were using an SQLite UNIQUE index to assert
that no `[e a]` pair, with `a` a cardinality one attribute, was
asserted more than once. However, that's not in line with Datomic,
which treats transaction inputs as a set and allows a single datom
like `[e a v]` to appear multiple times. It's both awkward and not
particularly efficient to look for _distinct_ repetitions in SQL, so
we accept some runtime cost in order to check for repetitions in the
transactor. This will allow us to address #532, which is really about
whether we treat inputs as sets. A side benefit is that we can
provide more helpful error messages when the transactor does detect
that the input truly violates the cardinality constraints of the
schema.
This commit builds a trie while error checking and collecting final
terms, which should be fairly efficient. It also allows a simpler
expression of input-provided :db/txInstant datoms, which in turn
uncovered a small issue with the transaction watcher, where-by the
watcher would not see non-input-provided :db/txInstant datoms.
This transition to Datomic-like input-as-set semantics allows us to
address #532. Previously, two tempids that upserted to the same entid
would produce duplicate datoms, and that would have been rejected by
the transactor -- correctly, since we did not allow duplicate datoms
under the input-as-list semantics. With input-as-set semantics,
duplicate datoms are allowed; and that means that we must allow
tempids to be equivalent, i.e., to resolve to the same tempid.
To achieve this, we:
- index the set of tempids
- identify tempid indices that share an upsert
- map tempids to a dense set of contiguous integer labels
We use the well-known union-find algorithm, as implemented by
petgraph, to efficiently manage the set of equivalent tempids.
Along the way, I've fixed and added tests for two small errors in the
transactor. First, don't drop datoms resolved by upsert (#679).
Second, ensure that complex upserts are allocated.
I don't know quite what happened here. The Clojure implementation
correctly kept complex upserts that hadn't resolved as complex
upserts (see
https://github.com/mozilla/mentat/blob/9a9dfb502acf5e4cdb1059d4aac831d7603063c8/src/common/datomish/transact.cljc#L436)
and then allocated complex upserts if they didn't resolve (see
https://github.com/mozilla/mentat/blob/9a9dfb502acf5e4cdb1059d4aac831d7603063c8/src/common/datomish/transact.cljc#L509).
Based on the code comments, I think the Rust implementation must have
incorrectly tried to optimize by handling all complex upserts in at
most a single generation of evolution, and that's just not correct.
We're effectively implementing a topological sort, using very specific
domain knowledge, and its not true that a node in a topological sort
can be considered only once!
2018-04-30 22:16:05 +00:00
[ :db / add 100 :test / many 5 ]
[ :db / add 100 :test / many 6 ]
2019-07-17 15:46:47 +00:00
] " #
) ;
Add type checking and constraint checking to the transactor. (#663, #532, #679)
This should address #663, by re-inserting type checking in the
transactor stack after the entry point used by the term builder.
Before this commit, we were using an SQLite UNIQUE index to assert
that no `[e a]` pair, with `a` a cardinality one attribute, was
asserted more than once. However, that's not in line with Datomic,
which treats transaction inputs as a set and allows a single datom
like `[e a v]` to appear multiple times. It's both awkward and not
particularly efficient to look for _distinct_ repetitions in SQL, so
we accept some runtime cost in order to check for repetitions in the
transactor. This will allow us to address #532, which is really about
whether we treat inputs as sets. A side benefit is that we can
provide more helpful error messages when the transactor does detect
that the input truly violates the cardinality constraints of the
schema.
This commit builds a trie while error checking and collecting final
terms, which should be fairly efficient. It also allows a simpler
expression of input-provided :db/txInstant datoms, which in turn
uncovered a small issue with the transaction watcher, where-by the
watcher would not see non-input-provided :db/txInstant datoms.
This transition to Datomic-like input-as-set semantics allows us to
address #532. Previously, two tempids that upserted to the same entid
would produce duplicate datoms, and that would have been rejected by
the transactor -- correctly, since we did not allow duplicate datoms
under the input-as-list semantics. With input-as-set semantics,
duplicate datoms are allowed; and that means that we must allow
tempids to be equivalent, i.e., to resolve to the same tempid.
To achieve this, we:
- index the set of tempids
- identify tempid indices that share an upsert
- map tempids to a dense set of contiguous integer labels
We use the well-known union-find algorithm, as implemented by
petgraph, to efficiently manage the set of equivalent tempids.
Along the way, I've fixed and added tests for two small errors in the
transactor. First, don't drop datoms resolved by upsert (#679).
Second, ensure that complex upserts are allocated.
I don't know quite what happened here. The Clojure implementation
correctly kept complex upserts that hadn't resolved as complex
upserts (see
https://github.com/mozilla/mentat/blob/9a9dfb502acf5e4cdb1059d4aac831d7603063c8/src/common/datomish/transact.cljc#L436)
and then allocated complex upserts if they didn't resolve (see
https://github.com/mozilla/mentat/blob/9a9dfb502acf5e4cdb1059d4aac831d7603063c8/src/common/datomish/transact.cljc#L509).
Based on the code comments, I think the Rust implementation must have
incorrectly tried to optimize by handling all complex upserts in at
most a single generation of evolution, and that's just not correct.
We're effectively implementing a topological sort, using very specific
domain knowledge, and its not true that a node in a topological sort
can be considered only once!
2018-04-30 22:16:05 +00:00
// Can't add and retract the same datom for an attribute, regardless of cardinality.
assert_transact! ( conn , r #" [
[ :db / add 100 :test / one 7 ]
[ :db / retract 100 :test / one 7 ]
[ :db / add 100 :test / many 8 ]
[ :db / retract 100 :test / many 8 ]
] " #,
Err ( " schema constraint violation: cardinality conflicts: \n AddRetractConflict { e: 100, a: 200, vs: {Long(7)} } \n AddRetractConflict { e: 100, a: 201, vs: {Long(8)} } \n " ) ) ;
2018-05-01 20:47:33 +00:00
}
2018-06-13 15:49:40 +00:00
#[ test ]
#[ cfg(feature = " sqlcipher " ) ]
fn test_sqlcipher_openable ( ) {
let secret_key = " key " ;
2019-07-17 15:46:47 +00:00
let sqlite = new_connection_with_key ( " ../fixtures/v1encrypted.db " , secret_key )
. expect ( " Failed to find test DB " ) ;
sqlite
2020-01-14 15:46:21 +00:00
. query_row (
" SELECT COUNT(*) FROM sqlite_master " ,
rusqlite ::params! [ ] ,
| row | row . get ::< _ , i64 > ( 0 ) ,
)
2018-06-13 15:49:40 +00:00
. expect ( " Failed to execute sql query on encrypted DB " ) ;
}
#[ cfg(feature = " sqlcipher " ) ]
2019-07-17 15:46:47 +00:00
fn test_open_fail < F > ( opener : F )
where
F : FnOnce ( ) -> rusqlite ::Result < rusqlite ::Connection > ,
{
2018-06-13 15:49:40 +00:00
let err = opener ( ) . expect_err ( " Should fail to open encrypted DB " ) ;
match err {
rusqlite ::Error ::SqliteFailure ( err , .. ) = > {
2019-07-17 15:46:47 +00:00
assert_eq! (
err . extended_code , 26 ,
" Should get error code 26 (not a database). "
) ;
}
2018-06-13 15:49:40 +00:00
err = > {
panic! ( " Wrong error type! {} " , err ) ;
}
}
}
#[ test ]
#[ cfg(feature = " sqlcipher " ) ]
fn test_sqlcipher_requires_key ( ) {
// Don't use a key.
test_open_fail ( | | new_connection ( " ../fixtures/v1encrypted.db " ) ) ;
}
#[ test ]
#[ cfg(feature = " sqlcipher " ) ]
fn test_sqlcipher_requires_correct_key ( ) {
// Use a key, but the wrong one.
test_open_fail ( | | new_connection_with_key ( " ../fixtures/v1encrypted.db " , " wrong key " ) ) ;
}
#[ test ]
#[ cfg(feature = " sqlcipher " ) ]
fn test_sqlcipher_some_transactions ( ) {
2019-07-17 15:46:47 +00:00
let sqlite =
new_connection_with_key ( " " , " hunter2 " ) . expect ( " Failed to create encrypted connection " ) ;
2018-06-13 15:49:40 +00:00
// Run a basic test as a sanity check.
run_test_add ( TestConn ::with_sqlite ( sqlite ) ) ;
}
2017-01-26 00:13:56 +00:00
}