Simple aggregates. (#584) r=emily

* Pre: use debugcli in VSCode.
* Pre: wrap subqueries in parentheses in output SQL.
* Pre: add ExistingColumn.

This lets us make reference to columns by name, rather than only
pointing to qualified aliases.

* Pre: add Into for &str to TypedValue.
* Pre: add Store.transact.
* Pre: cleanup.
* Parse and algebrize simple aggregates. (#312)
* Follow-up: print aggregate columns more neatly in the CLI.
* Useful ValueTypeSet helpers.
* Allow for entity inequalities.
* Add 'differ', which is a ref-specialized not-equals.
* Add 'unpermute', a function for getting unique, distinct pairs from bindings.
* Review comments.
* Add 'the' pseudo-aggregation operator.

This allows for a corresponding value to be returned when a query
includes one 'min' or 'max' aggregate.
This commit is contained in:
Richard Newman 2018-03-12 15:18:50 -07:00 committed by GitHub
parent 46835885e4
commit 833ff92436
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
20 changed files with 1676 additions and 141 deletions

2
.vscode/tasks.json vendored
View file

@ -21,7 +21,7 @@
"label": "Run CLI",
"command": "cargo",
"args": [
"cli",
"debugcli",
],
"problemMatcher": [
"$rustc"

View file

@ -293,6 +293,12 @@ impl From<Uuid> for TypedValue {
}
}
impl<'a> From<&'a str> for TypedValue {
fn from(value: &'a str) -> TypedValue {
TypedValue::String(Rc::new(value.to_string()))
}
}
impl From<String> for TypedValue {
fn from(value: String) -> TypedValue {
TypedValue::String(Rc::new(value))
@ -449,6 +455,15 @@ impl ValueTypeSet {
ValueTypeSet(EnumSet::of_both(ValueType::Double, ValueType::Long))
}
/// Return a set containing `Double`, `Long`, and `Instant`.
pub fn of_numeric_and_instant_types() -> ValueTypeSet {
let mut s = EnumSet::new();
s.insert(ValueType::Double);
s.insert(ValueType::Long);
s.insert(ValueType::Instant);
ValueTypeSet(s)
}
/// Return a set containing `Ref` and `Keyword`.
pub fn of_keywords() -> ValueTypeSet {
ValueTypeSet(EnumSet::of_both(ValueType::Ref, ValueType::Keyword))
@ -516,6 +531,18 @@ impl ValueTypeSet {
}
}
impl From<ValueType> for ValueTypeSet {
fn from(t: ValueType) -> Self {
ValueTypeSet::of_one(t)
}
}
impl ValueTypeSet {
pub fn is_only_numeric(&self) -> bool {
self.is_subset(&ValueTypeSet::of_numeric_types())
}
}
impl IntoIterator for ValueTypeSet {
type Item = ValueType;
type IntoIter = ::enum_set::Iter<ValueType>;
@ -541,10 +568,16 @@ impl ::std::iter::Extend<ValueType> for ValueTypeSet {
}
}
/// We have an enum of types, `ValueType`. It can be collected into a set, `ValueTypeSet`. Each type
/// is associated with a type tag, which is how a type is represented in, e.g., SQL storage. Types
/// can share type tags, because backing SQL storage is able to differentiate between some types
/// (e.g., longs and doubles), and so distinct tags aren't necessary. That association is defined by
/// `SQLValueType`. That trait similarly extends to `ValueTypeSet`, which maps a collection of types
/// into a collection of tags.
pub trait SQLValueTypeSet {
fn value_type_tags(&self) -> BTreeSet<ValueTypeTag>;
fn has_unique_type_code(&self) -> bool;
fn unique_type_code(&self) -> Option<ValueTypeTag>;
fn has_unique_type_tag(&self) -> bool;
fn unique_type_tag(&self) -> Option<ValueTypeTag>;
}
impl SQLValueTypeSet for ValueTypeSet {
@ -557,15 +590,15 @@ impl SQLValueTypeSet for ValueTypeSet {
out
}
fn unique_type_code(&self) -> Option<ValueTypeTag> {
if self.is_unit() || self.has_unique_type_code() {
fn unique_type_tag(&self) -> Option<ValueTypeTag> {
if self.is_unit() || self.has_unique_type_tag() {
self.exemplar().map(|t| t.value_type_tag())
} else {
None
}
}
fn has_unique_type_code(&self) -> bool {
fn has_unique_type_tag(&self) -> bool {
if self.is_unit() {
return true;
}

View file

@ -94,7 +94,7 @@ impl ConjoiningClauses {
// TODO: process source variables.
match args.next().unwrap() {
FnArg::SrcVar(SrcVar::DefaultSrc) => {},
_ => bail!(ErrorKind::InvalidArgument(where_fn.operator.clone(), "source variable".into(), 0)),
_ => bail!(ErrorKind::InvalidArgument(where_fn.operator.clone(), "source variable", 0)),
}
let schema = known.schema;
@ -127,8 +127,11 @@ impl ConjoiningClauses {
// An unknown ident, or an entity that isn't present in the store, or isn't a fulltext
// attribute, is likely enough to be a coding error that we choose to bail instead of
// marking the pattern as known-empty.
let a = a.ok_or(ErrorKind::InvalidArgument(where_fn.operator.clone(), "attribute".into(), 1))?;
let attribute = schema.attribute_for_entid(a).cloned().ok_or(ErrorKind::InvalidArgument(where_fn.operator.clone(), "attribute".into(), 1))?;
let a = a.ok_or(ErrorKind::InvalidArgument(where_fn.operator.clone(), "attribute", 1))?;
let attribute = schema.attribute_for_entid(a)
.cloned()
.ok_or(ErrorKind::InvalidArgument(where_fn.operator.clone(),
"attribute", 1))?;
if !attribute.fulltext {
// We can never get results from a non-fulltext attribute!
@ -166,12 +169,12 @@ impl ConjoiningClauses {
FnArg::Variable(in_var) => {
match self.bound_value(&in_var) {
Some(t @ TypedValue::String(_)) => Either::Left(t),
Some(_) => bail!(ErrorKind::InvalidArgument(where_fn.operator.clone(), "string".into(), 2)),
Some(_) => bail!(ErrorKind::InvalidArgument(where_fn.operator.clone(), "string", 2)),
None => {
// Regardless of whether we'll be providing a string later, or the value
// comes from a column, it must be a string.
if self.known_type(&in_var) != Some(ValueType::String) {
bail!(ErrorKind::InvalidArgument(where_fn.operator.clone(), "string".into(), 2));
bail!(ErrorKind::InvalidArgument(where_fn.operator.clone(), "string", 2));
}
if self.input_variables.contains(&in_var) {
@ -192,7 +195,7 @@ impl ConjoiningClauses {
},
}
},
_ => bail!(ErrorKind::InvalidArgument(where_fn.operator.clone(), "string".into(), 2)),
_ => bail!(ErrorKind::InvalidArgument(where_fn.operator.clone(), "string", 2)),
};
let qv = match search {

View file

@ -286,6 +286,18 @@ impl Default for ConjoiningClauses {
}
}
pub struct VariableIterator<'a>(
::std::collections::btree_map::Keys<'a, Variable, TypedValue>,
);
impl<'a> Iterator for VariableIterator<'a> {
type Item = &'a Variable;
fn next(&mut self) -> Option<&'a Variable> {
self.0.next()
}
}
impl ConjoiningClauses {
/// Construct a new `ConjoiningClauses` with the provided alias counter. This allows a caller
/// to share a counter with an enclosing scope, and to start counting at a particular offset
@ -390,7 +402,7 @@ impl ConjoiningClauses {
self.value_bindings.get(var).cloned()
}
pub(crate) fn is_value_bound(&self, var: &Variable) -> bool {
pub fn is_value_bound(&self, var: &Variable) -> bool {
self.value_bindings.contains_key(var)
}
@ -398,9 +410,14 @@ impl ConjoiningClauses {
self.value_bindings.with_intersected_keys(variables)
}
/// Return an iterator over the variables externally bound to values.
pub fn value_bound_variables(&self) -> VariableIterator {
VariableIterator(self.value_bindings.keys())
}
/// Return a set of the variables externally bound to values.
pub(crate) fn value_bound_variable_set(&self) -> BTreeSet<Variable> {
self.value_bindings.keys().cloned().collect()
pub fn value_bound_variable_set(&self) -> BTreeSet<Variable> {
self.value_bound_variables().cloned().collect()
}
/// Return a single `ValueType` if the given variable is known to have a precise type.
@ -414,7 +431,7 @@ impl ConjoiningClauses {
}
}
pub(crate) fn known_type_set(&self, var: &Variable) -> ValueTypeSet {
pub fn known_type_set(&self, var: &Variable) -> ValueTypeSet {
self.known_types.get(var).cloned().unwrap_or(ValueTypeSet::any())
}

View file

@ -92,13 +92,13 @@ impl ConjoiningClauses {
let mut left_types = self.potential_types(known.schema, &left)?
.intersection(&supported_types);
if left_types.is_empty() {
bail!(ErrorKind::InvalidArgument(predicate.operator.clone(), "numeric or instant", 0));
bail!(ErrorKind::InvalidArgumentType(predicate.operator.clone(), supported_types, 0));
}
let mut right_types = self.potential_types(known.schema, &right)?
.intersection(&supported_types);
if right_types.is_empty() {
bail!(ErrorKind::InvalidArgument(predicate.operator.clone(), "numeric or instant", 1));
bail!(ErrorKind::InvalidArgumentType(predicate.operator.clone(), supported_types, 1));
}
// We would like to allow longs to compare to doubles.
@ -134,14 +134,18 @@ impl ConjoiningClauses {
// We expect the intersection to be Long, Long+Double, Double, or Instant.
let left_v;
let right_v;
if shared_types == ValueTypeSet::of_one(ValueType::Instant) {
left_v = self.resolve_instant_argument(&predicate.operator, 0, left)?;
right_v = self.resolve_instant_argument(&predicate.operator, 1, right)?;
} else if !shared_types.is_empty() && shared_types.is_subset(&ValueTypeSet::of_numeric_types()) {
} else if shared_types.is_only_numeric() {
left_v = self.resolve_numeric_argument(&predicate.operator, 0, left)?;
right_v = self.resolve_numeric_argument(&predicate.operator, 1, right)?;
} else if shared_types == ValueTypeSet::of_one(ValueType::Ref) {
left_v = self.resolve_ref_argument(known.schema, &predicate.operator, 0, left)?;
right_v = self.resolve_ref_argument(known.schema, &predicate.operator, 1, right)?;
} else {
bail!(ErrorKind::InvalidArgument(predicate.operator.clone(), "numeric or instant", 0));
bail!(ErrorKind::InvalidArgumentType(predicate.operator.clone(), supported_types, 0));
}
// These arguments must be variables or instant/numeric constants.

View file

@ -9,6 +9,8 @@
// specific language governing permissions and limitations under the License.
use mentat_core::{
HasSchema,
Schema,
TypedValue,
ValueType,
};
@ -92,11 +94,49 @@ impl ConjoiningClauses {
Constant(NonIntegerConstant::BigInteger(_)) |
Vector(_) => {
self.mark_known_empty(EmptyBecause::NonInstantArgument);
bail!(ErrorKind::InvalidArgument(function.clone(), "instant", position));
bail!(ErrorKind::InvalidArgumentType(function.clone(), ValueType::Instant.into(), position));
},
}
}
/// Take a function argument and turn it into a `QueryValue` suitable for use in a concrete
/// constraint.
pub(crate) fn resolve_ref_argument(&mut self, schema: &Schema, function: &PlainSymbol, position: usize, arg: FnArg) -> Result<QueryValue> {
use self::FnArg::*;
match arg {
FnArg::Variable(var) => {
self.constrain_var_to_type(var.clone(), ValueType::Ref);
if let Some(TypedValue::Ref(e)) = self.bound_value(&var) {
// Incorrect types will be handled by the constraint, above.
Ok(QueryValue::Entid(e))
} else {
self.column_bindings
.get(&var)
.and_then(|cols| cols.first().map(|col| QueryValue::Column(col.clone())))
.ok_or_else(|| Error::from_kind(ErrorKind::UnboundVariable(var.name())))
}
},
EntidOrInteger(i) => Ok(QueryValue::TypedValue(TypedValue::Ref(i))),
IdentOrKeyword(i) => {
schema.get_entid(&i)
.map(|known_entid| QueryValue::Entid(known_entid.into()))
.ok_or_else(|| Error::from_kind(ErrorKind::UnrecognizedIdent(i.to_string())))
},
Constant(NonIntegerConstant::Boolean(_)) |
Constant(NonIntegerConstant::Float(_)) |
Constant(NonIntegerConstant::Text(_)) |
Constant(NonIntegerConstant::Uuid(_)) |
Constant(NonIntegerConstant::Instant(_)) |
Constant(NonIntegerConstant::BigInteger(_)) |
SrcVar(_) |
Vector(_) => {
self.mark_known_empty(EmptyBecause::NonEntityArgument);
bail!(ErrorKind::InvalidArgumentType(function.clone(), ValueType::Ref.into(), position));
},
}
}
/// Take a function argument and turn it into a `QueryValue` suitable for use in a concrete
/// constraint.
#[allow(dead_code)]

View file

@ -12,6 +12,7 @@ extern crate mentat_query;
use mentat_core::{
ValueType,
ValueTypeSet,
};
use self::mentat_query::{
@ -49,6 +50,11 @@ error_chain! {
display("value of type {} provided for var {}, expected {}", provided, var, declared)
}
UnrecognizedIdent(ident: String) {
description("no entid found for ident")
display("no entid found for ident: {}", ident)
}
UnknownFunction(name: PlainSymbol) {
description("no such function")
display("no function named {}", name)
@ -80,9 +86,14 @@ error_chain! {
display("invalid expression in ground constant")
}
InvalidArgument(function: PlainSymbol, expected_type: &'static str, position: usize) {
InvalidArgument(function: PlainSymbol, expected: &'static str, position: usize) {
description("invalid argument")
display("invalid argument to {}: expected {} in position {}.", function, expected_type, position)
display("invalid argument to {}: expected {} in position {}.", function, expected, position)
}
InvalidArgumentType(function: PlainSymbol, expected_types: ValueTypeSet, position: usize) {
description("invalid argument")
display("invalid argument to {}: expected one of {:?} in position {}.", function, expected_types, position)
}
InvalidLimit(val: String, kind: ValueType) {

View file

@ -8,6 +8,8 @@
// CONDITIONS OF ANY KIND, either express or implied. See the License for the
// specific language governing permissions and limitations under the License.
#![recursion_limit="128"]
#[macro_use]
extern crate error_chain;
@ -130,7 +132,19 @@ pub struct AlgebraicQuery {
default_source: SrcVar,
pub find_spec: Rc<FindSpec>,
has_aggregates: bool,
/// The set of variables that the caller wishes to be used for grouping when aggregating.
/// These are specified in the query input, as `:with`, and are then chewed up during projection.
/// If no variables are supplied, then no additional grouping is necessary beyond the
/// non-aggregated projection list.
pub with: BTreeSet<Variable>,
/// Some query features, such as ordering, are implemented by implicit reference to SQL columns.
/// In order for these references to be 'live', those columns must be projected.
/// This is the set of variables that must be so projected.
/// This is not necessarily every variable that will be so required -- some variables
/// will already be in the projection list.
pub named_projection: BTreeSet<Variable>,
pub order: Option<Vec<OrderBy>>,
pub limit: Limit,
pub cc: clauses::ConjoiningClauses,
@ -147,7 +161,12 @@ impl AlgebraicQuery {
self.find_spec
.columns()
.all(|e| match e {
&Element::Variable(ref var) => self.cc.is_value_bound(var),
&Element::Variable(ref var) |
&Element::Corresponding(ref var) => self.cc.is_value_bound(var),
// For now, we pretend that aggregate functions are never fully bound:
// we don't statically compute them, even if we know the value of the var.
&Element::Aggregate(ref _fn) => false,
})
}
@ -270,7 +289,6 @@ pub fn algebrize_with_inputs(known: Known,
cc.process_required_types()?;
let (order, extra_vars) = validate_and_simplify_order(&cc, parsed.order)?;
let with: BTreeSet<Variable> = parsed.with.into_iter().chain(extra_vars.into_iter()).collect();
// This might leave us with an unused `:in` variable.
let limit = if parsed.find_spec.is_unit_limited() { Limit::Fixed(1) } else { parsed.limit };
@ -278,7 +296,8 @@ pub fn algebrize_with_inputs(known: Known,
default_source: parsed.default_source,
find_spec: Rc::new(parsed.find_spec),
has_aggregates: false, // TODO: we don't parse them yet.
with: with,
with: parsed.with,
named_projection: extra_vars,
order: order,
limit: limit,
cc: cc,

View file

@ -283,6 +283,10 @@ pub enum Inequality {
GreaterThan,
GreaterThanOrEquals,
NotEquals,
// Ref operators.
Unpermute,
Differ,
}
impl Inequality {
@ -294,6 +298,9 @@ impl Inequality {
GreaterThan => ">",
GreaterThanOrEquals => ">=",
NotEquals => "<>",
Unpermute => "<",
Differ => "<>",
}
}
@ -304,15 +311,31 @@ impl Inequality {
">" => Some(Inequality::GreaterThan),
">=" => Some(Inequality::GreaterThanOrEquals),
"!=" => Some(Inequality::NotEquals),
_ => None,
"unpermute" => Some(Inequality::Unpermute),
"differ" => Some(Inequality::Differ),
_ => None,
}
}
// The built-in inequality operators apply to Long, Double, and Instant.
pub fn supported_types(&self) -> ValueTypeSet {
let mut ts = ValueTypeSet::of_numeric_types();
ts.insert(ValueType::Instant);
ts
use self::Inequality::*;
match self {
&LessThan |
&LessThanOrEquals |
&GreaterThan |
&GreaterThanOrEquals |
&NotEquals => {
let mut ts = ValueTypeSet::of_numeric_types();
ts.insert(ValueType::Instant);
ts
},
&Unpermute |
&Differ => {
ValueTypeSet::of_one(ValueType::Ref)
},
}
}
}
@ -325,6 +348,9 @@ impl Debug for Inequality {
&GreaterThan => ">",
&GreaterThanOrEquals => ">=",
&NotEquals => "!=", // Datalog uses !=. SQL uses <>.
&Unpermute => "<",
&Differ => "<>",
})
}
}
@ -505,6 +531,7 @@ pub enum EmptyBecause {
NonAttributeArgument,
NonInstantArgument,
NonNumericArgument,
NonEntityArgument,
NonStringFulltextValue,
NonFulltextAttribute(Entid),
UnresolvedIdent(NamespacedKeyword),
@ -546,6 +573,9 @@ impl Debug for EmptyBecause {
&NonInstantArgument => {
write!(f, "Non-instant argument in instant place")
},
&NonEntityArgument => {
write!(f, "Non-entity argument in entity place")
},
&NonNumericArgument => {
write!(f, "Non-numeric argument in numeric place")
},

View file

@ -69,9 +69,9 @@ fn test_instant_predicates_require_instants() {
[?e :foo/date ?t]
[(> ?t "2017-06-16T00:56:41.257Z")]]"#;
match bails(known, query).0 {
ErrorKind::InvalidArgument(op, why, idx) => {
ErrorKind::InvalidArgumentType(op, why, idx) => {
assert_eq!(op, PlainSymbol::new(">"));
assert_eq!(why, "numeric or instant");
assert_eq!(why, ValueTypeSet::of_numeric_and_instant_types());
assert_eq!(idx, 1);
},
_ => panic!("Expected InvalidArgument."),
@ -82,9 +82,9 @@ fn test_instant_predicates_require_instants() {
[?e :foo/date ?t]
[(> "2017-06-16T00:56:41.257Z", ?t)]]"#;
match bails(known, query).0 {
ErrorKind::InvalidArgument(op, why, idx) => {
ErrorKind::InvalidArgumentType(op, why, idx) => {
assert_eq!(op, PlainSymbol::new(">"));
assert_eq!(why, "numeric or instant");
assert_eq!(why, ValueTypeSet::of_numeric_and_instant_types());
assert_eq!(idx, 0); // We get this right.
},
_ => panic!("Expected InvalidArgument."),

View file

@ -41,6 +41,7 @@ use self::mentat_parser_utils::value_and_span::{
};
use self::mentat_query::{
Aggregate,
Binding,
Direction,
Element,
@ -170,6 +171,8 @@ def_parser!(Query, order, Order, {
.or(Query::variable().map(|v| Order(Direction::Ascending, v)))
});
def_matches_plain_symbol!(Query, the, "the");
pub struct Where<'a>(std::marker::PhantomData<&'a ()>);
def_parser!(Where, pattern_value_place, PatternValuePlace, {
@ -274,6 +277,13 @@ def_parser!(Query, func, (QueryFunction, Vec<FnArg>), {
(Query::query_function(), Query::arguments())
});
def_parser!(Query, aggregate, Aggregate, {
seq().of_exactly(Query::func())
.map(|(func, args)| Aggregate {
func, args,
})
});
/// A vector containing just a parenthesized filter expression.
def_parser!(Where, pred, WhereClause, {
// Accept either a nested list or a nested vector here:
@ -417,10 +427,25 @@ def_matches_plain_symbol!(Find, ellipsis, "...");
def_matches_plain_symbol!(Find, placeholder, "_");
def_parser!(Find, elem, Element, {
def_parser!(Find, variable_element, Element, {
Query::variable().map(Element::Variable)
});
def_parser!(Find, corresponding_element, Element, {
seq().of_exactly(Query::the().with(Query::variable()))
.map(Element::Corresponding)
});
def_parser!(Find, aggregate_element, Element, {
Query::aggregate().map(Element::Aggregate)
});
def_parser!(Find, elem, Element, {
choice([try(Find::variable_element()),
try(Find::corresponding_element()),
try(Find::aggregate_element())])
});
def_parser!(Find, find_scalar, FindSpec, {
Find::elem().skip(Find::period())
.map(FindSpec::FindScalar)
@ -955,6 +980,45 @@ mod test {
]));
}
#[test]
fn test_the() {
assert_edn_parses_to!(Find::corresponding_element,
"(the ?y)",
Element::Corresponding(Variable::from_valid_name("?y")));
assert_edn_parses_to!(Find::find_tuple,
"[(the ?x) ?y]",
FindSpec::FindTuple(vec![Element::Corresponding(Variable::from_valid_name("?x")),
Element::Variable(Variable::from_valid_name("?y"))]));
assert_edn_parses_to!(Find::spec,
"[(the ?x) ?y]",
FindSpec::FindTuple(vec![Element::Corresponding(Variable::from_valid_name("?x")),
Element::Variable(Variable::from_valid_name("?y"))]));
let expected_query =
FindQuery {
find_spec: FindSpec::FindTuple(vec![Element::Corresponding(Variable::from_valid_name("?x")),
Element::Variable(Variable::from_valid_name("?y"))]),
where_clauses: vec![
WhereClause::Pattern(Pattern {
source: None,
entity: PatternNonValuePlace::Variable(Variable::from_valid_name("?x")),
attribute: PatternNonValuePlace::Placeholder,
value: PatternValuePlace::Variable(Variable::from_valid_name("?y")),
tx: PatternNonValuePlace::Placeholder,
})],
default_source: SrcVar::DefaultSrc,
with: Default::default(),
in_vars: Default::default(),
in_sources: Default::default(),
limit: Limit::None,
order: None,
};
assert_edn_parses_to!(Find::query,
"[:find [(the ?x) ?y]
:where [?x _ ?y]]",
expected_query);
}
#[test]
fn test_where_fn() {
assert_edn_parses_to!(Where::where_fn,

View file

@ -5,6 +5,7 @@ workspace = ".."
[dependencies]
error-chain = { git = "https://github.com/rnewman/error-chain", branch = "rnewman/sync" }
indexmap = "0.4"
[dependencies.rusqlite]
version = "0.13"

View file

@ -10,6 +10,7 @@
#[macro_use]
extern crate error_chain;
extern crate indexmap;
extern crate rusqlite;
extern crate mentat_core;
@ -24,8 +25,13 @@ use std::collections::{
};
use std::iter;
use std::rc::Rc;
use indexmap::{
IndexSet,
};
use rusqlite::{
Row,
Rows,
@ -33,8 +39,10 @@ use rusqlite::{
use mentat_core::{
SQLValueType,
SQLValueTypeSet,
TypedValue,
ValueType,
ValueTypeSet,
ValueTypeTag,
};
@ -47,9 +55,12 @@ use mentat_db::{
};
use mentat_query::{
Aggregate,
Element,
FindSpec,
Limit,
PlainSymbol,
QueryFunction,
Variable,
};
@ -57,12 +68,15 @@ use mentat_query_algebrizer::{
AlgebraicQuery,
ColumnName,
ConjoiningClauses,
QualifiedAlias,
VariableBindings,
VariableColumn,
};
use mentat_query_sql::{
ColumnOrExpression,
Expression,
GroupBy,
Name,
Projection,
ProjectedColumn,
@ -73,6 +87,39 @@ error_chain! {
Error, ErrorKind, ResultExt, Result;
}
errors {
/// We're just not done yet. Message that the feature is recognized but not yet
/// implemented.
NotYetImplemented(t: String) {
description("not yet implemented")
display("not yet implemented: {}", t)
}
CannotProjectImpossibleBinding(op: SimpleAggregationOp) {
description("no possible types for variable in projection list")
display("no possible types for value provided to {:?}", op)
}
CannotApplyAggregateOperationToTypes(op: SimpleAggregationOp, types: ValueTypeSet) {
description("cannot apply projection operation to types")
display("cannot apply projection operation {:?} to types {:?}", op, types)
}
UnboundVariable(var: PlainSymbol) {
description("cannot project unbound variable")
display("cannot project unbound variable {:?}", var)
}
NoTypeAvailableForVariable(var: PlainSymbol) {
description("cannot find type for variable")
display("cannot find type for variable {:?}", var)
}
UnexpectedResultsType(actual: &'static str, expected: &'static str) {
description("unexpected query results type")
display("expected {}, got {}", expected, actual)
}
AmbiguousAggregates(min_max_count: usize, corresponding_count: usize) {
description("ambiguous aggregates")
display("min/max expressions: {} (max 1), corresponding: {}", min_max_count, corresponding_count)
}
}
foreign_links {
Rusqlite(rusqlite::Error);
}
@ -80,13 +127,6 @@ error_chain! {
links {
DbError(mentat_db::Error, mentat_db::ErrorKind);
}
errors {
UnexpectedResultsType(actual: &'static str, expected: &'static str) {
description("unexpected query results type")
display("expected {}, got {}", expected, actual)
}
}
}
#[derive(Debug, PartialEq, Eq)]
@ -146,23 +186,54 @@ impl QueryOutput {
pub fn from_constants(spec: &Rc<FindSpec>, bindings: VariableBindings) -> QueryResults {
use self::FindSpec::*;
match &**spec {
&FindScalar(Element::Variable(ref var)) => {
&FindScalar(Element::Variable(ref var)) |
&FindScalar(Element::Corresponding(ref var)) => {
let val = bindings.get(var).cloned();
QueryResults::Scalar(val)
},
&FindScalar(Element::Aggregate(ref _agg)) => {
// TODO
unimplemented!();
},
&FindTuple(ref elements) => {
let values = elements.iter().map(|e| match e {
&Element::Variable(ref var) => bindings.get(var).cloned().expect("every var to have a binding"),
}).collect();
let values = elements.iter()
.map(|e| match e {
&Element::Variable(ref var) |
&Element::Corresponding(ref var) => {
bindings.get(var).cloned().expect("every var to have a binding")
},
&Element::Aggregate(ref _agg) => {
// TODO: static computation of aggregates, then
// implement the condition in `is_fully_bound`.
unreachable!();
},
})
.collect();
QueryResults::Tuple(Some(values))
},
&FindColl(Element::Variable(ref var)) => {
&FindColl(Element::Variable(ref var)) |
&FindColl(Element::Corresponding(ref var)) => {
let val = bindings.get(var).cloned().expect("every var to have a binding");
QueryResults::Coll(vec![val])
},
&FindColl(Element::Aggregate(ref _agg)) => {
// Does it even make sense to write
// [:find [(max ?x) ...] :where [_ :foo/bar ?x]]
// ?
// TODO
unimplemented!();
},
&FindRel(ref elements) => {
let values = elements.iter().map(|e| match e {
&Element::Variable(ref var) => bindings.get(var).cloned().expect("every var to have a binding"),
&Element::Variable(ref var) |
&Element::Corresponding(ref var) => {
bindings.get(var).cloned().expect("every var to have a binding")
},
&Element::Aggregate(ref _agg) => {
// TODO: static computation of aggregates, then
// implement the condition in `is_fully_bound`.
unreachable!();
},
}).collect();
QueryResults::Rel(vec![values])
},
@ -254,16 +325,16 @@ impl TypedIndex {
/// Look up this index and type(index) pair in the provided row.
/// This function will panic if:
///
/// - This is an `Unknown` and the retrieved type code isn't an i32.
/// - This is an `Unknown` and the retrieved type tag isn't an i32.
/// - If the retrieved value can't be coerced to a rusqlite `Value`.
/// - Either index is out of bounds.
///
/// Because we construct our SQL projection list, the code that stored the data, and this
/// Because we construct our SQL projection list, the tag that stored the data, and this
/// consumer, a panic here implies that we have a bad bug — we put data of a very wrong type in
/// a row, and thus can't coerce to Value, we're retrieving from the wrong place, or our
/// generated SQL is junk.
///
/// This function will return a runtime error if the type code is unknown, or the value is
/// This function will return a runtime error if the type tag is unknown, or the value is
/// otherwise not convertible by the DB layer.
fn lookup<'a, 'stmt>(&self, row: &Row<'a, 'stmt>) -> Result<TypedValue> {
use TypedIndex::*;
@ -282,17 +353,22 @@ impl TypedIndex {
}
}
fn candidate_column(cc: &ConjoiningClauses, var: &Variable) -> (ColumnOrExpression, Name) {
fn cc_column(cc: &ConjoiningClauses, var: &Variable) -> Result<QualifiedAlias> {
cc.column_bindings
.get(var)
.and_then(|cols| cols.get(0).cloned())
.ok_or_else(|| ErrorKind::UnboundVariable(var.name()).into())
}
fn candidate_column(cc: &ConjoiningClauses, var: &Variable) -> Result<(ColumnOrExpression, Name)> {
// Every variable should be bound by the top-level CC to at least
// one column in the query. If that constraint is violated it's a
// bug in our code, so it's appropriate to panic here.
let columns = cc.column_bindings
.get(var)
.expect(format!("Every variable should have a binding, but {:?} does not", var).as_str());
let qa = columns[0].clone();
let name = VariableColumn::Variable(var.clone()).column_name();
(ColumnOrExpression::Column(qa), name)
cc_column(cc, var)
.map(|qa| {
let name = VariableColumn::Variable(var.clone()).column_name();
(ColumnOrExpression::Column(qa), name)
})
}
fn candidate_type_column(cc: &ConjoiningClauses, var: &Variable) -> (ColumnOrExpression, Name) {
@ -304,24 +380,216 @@ fn candidate_type_column(cc: &ConjoiningClauses, var: &Variable) -> (ColumnOrExp
}
/// Return the projected column -- that is, a value or SQL column and an associated name -- for a
/// given variable. Also return the type, if known.
/// given variable. Also return the type.
/// Callers are expected to determine whether to project a type tag as an additional SQL column.
pub fn projected_column_for_var(var: &Variable, cc: &ConjoiningClauses) -> (ProjectedColumn, Option<ValueType>) {
pub fn projected_column_for_var(var: &Variable, cc: &ConjoiningClauses) -> Result<(ProjectedColumn, ValueTypeSet)> {
if let Some(value) = cc.bound_value(&var) {
// If we already know the value, then our lives are easy.
let tag = value.value_type();
let name = VariableColumn::Variable(var.clone()).column_name();
(ProjectedColumn(ColumnOrExpression::Value(value.clone()), name), Some(tag))
Ok((ProjectedColumn(ColumnOrExpression::Value(value.clone()), name), ValueTypeSet::of_one(tag)))
} else {
// If we don't, then the CC *must* have bound the variable.
let (column, name) = candidate_column(cc, var);
(ProjectedColumn(column, name), cc.known_type(var))
let (column, name) = candidate_column(cc, var)?;
Ok((ProjectedColumn(column, name), cc.known_type_set(var)))
}
}
/// Returns two values:
/// - The `ColumnOrExpression` to use in the query. This will always refer to other
/// variables by name; never to a datoms column.
/// - The known type of that value.
fn projected_column_for_simple_aggregate(simple: &SimpleAggregate, cc: &ConjoiningClauses) -> Result<(ProjectedColumn, ValueType)> {
let known_types = cc.known_type_set(&simple.var);
let return_type = simple.op.is_applicable_to_types(known_types)?;
let projected_column_or_expression =
if let Some(value) = cc.bound_value(&simple.var) {
// Oh, we already know the value!
if simple.use_static_value() {
// We can statically compute the aggregate result for some operators -- not count or
// sum, but avg/max/min are OK.
ColumnOrExpression::Value(value)
} else {
let expression = Expression::Unary {
sql_op: simple.op.to_sql(),
arg: ColumnOrExpression::Value(value),
};
ColumnOrExpression::Expression(Box::new(expression), return_type)
}
} else {
// The common case: the values are bound during execution.
let name = VariableColumn::Variable(simple.var.clone()).column_name();
let expression = Expression::Unary {
sql_op: simple.op.to_sql(),
arg: ColumnOrExpression::ExistingColumn(name),
};
ColumnOrExpression::Expression(Box::new(expression), return_type)
};
Ok((ProjectedColumn(projected_column_or_expression, simple.column_name()), return_type))
}
#[derive(Clone, Copy, Debug, Eq, PartialEq)]
pub enum SimpleAggregationOp {
Avg,
Count,
Max,
Min,
Sum,
}
impl SimpleAggregationOp {
fn to_sql(&self) -> &'static str {
use SimpleAggregationOp::*;
match self {
&Avg => "avg",
&Count => "count",
&Max => "max",
&Min => "min",
&Sum => "sum",
}
}
fn for_function(function: &QueryFunction) -> Option<SimpleAggregationOp> {
match function.0.plain_name() {
"avg" => Some(SimpleAggregationOp::Avg),
"count" => Some(SimpleAggregationOp::Count),
"max" => Some(SimpleAggregationOp::Max),
"min" => Some(SimpleAggregationOp::Min),
"sum" => Some(SimpleAggregationOp::Sum),
_ => None,
}
}
/// With knowledge of the types to which a variable might be bound,
/// return a `Result` to determine whether this aggregation is suitable.
/// For example, it's valid to take the `Avg` of `{Double, Long}`, invalid
/// to take `Sum` of `{Instant}`, valid to take (lexicographic) `Max` of `{String}`,
/// but invalid to take `Max` of `{Uuid, String}`.
///
/// The returned type is the type of the result of the aggregation.
fn is_applicable_to_types(&self, possibilities: ValueTypeSet) -> Result<ValueType> {
use SimpleAggregationOp::*;
if possibilities.is_empty() {
bail!(ErrorKind::CannotProjectImpossibleBinding(*self))
}
match self {
// One can always count results.
&Count => Ok(ValueType::Long),
// Only numeric types can be averaged or summed.
&Avg => {
if possibilities.is_only_numeric() {
// The mean of a set of numeric values will always, for our purposes, be a double.
Ok(ValueType::Double)
} else {
bail!(ErrorKind::CannotApplyAggregateOperationToTypes(*self, possibilities))
}
},
&Sum => {
if possibilities.is_only_numeric() {
if possibilities.contains(ValueType::Double) {
Ok(ValueType::Double)
} else {
// TODO: BigInt.
Ok(ValueType::Long)
}
} else {
bail!(ErrorKind::CannotApplyAggregateOperationToTypes(*self, possibilities))
}
},
&Max | &Min => {
if possibilities.is_unit() {
use ValueType::*;
let the_type = possibilities.exemplar().expect("a type");
match the_type {
// These types are numerically ordered.
Double | Long | Instant => Ok(the_type),
// Boolean: false < true.
Boolean => Ok(the_type),
// String: lexicographic order.
String => Ok(the_type),
// These types are unordered.
Keyword | Ref | Uuid => {
bail!(ErrorKind::CannotApplyAggregateOperationToTypes(*self, possibilities))
},
}
} else {
// It cannot be empty -- we checked.
// The only types that are valid to compare cross-type are numbers.
if possibilities.is_only_numeric() {
// Note that if the max/min is a Long, it will be returned as a Double!
if possibilities.contains(ValueType::Double) {
Ok(ValueType::Double)
} else {
// TODO: BigInt.
Ok(ValueType::Long)
}
} else {
bail!(ErrorKind::CannotApplyAggregateOperationToTypes(*self, possibilities))
}
}
},
}
}
}
struct SimpleAggregate {
op: SimpleAggregationOp,
var: Variable,
}
impl SimpleAggregate {
fn column_name(&self) -> Name {
format!("({} {})", self.op.to_sql(), self.var.name())
}
fn use_static_value(&self) -> bool {
use SimpleAggregationOp::*;
match self.op {
Avg | Max | Min => true,
Count | Sum => false,
}
}
}
trait SimpleAggregation {
fn to_simple(&self) -> Option<SimpleAggregate>;
}
impl SimpleAggregation for Aggregate {
fn to_simple(&self) -> Option<SimpleAggregate> {
if self.args.len() != 1 {
return None;
}
self.args[0]
.as_variable()
.and_then(|v| SimpleAggregationOp::for_function(&self.func)
.map(|op| SimpleAggregate { op, var: v.clone(), }))
}
}
/// An internal temporary struct to pass between the projection 'walk' and the
/// resultant projector.
/// Projection accumulates four things:
/// - Two SQL projection lists. We need two because aggregate queries are nested
/// in order to apply DISTINCT to values prior to aggregation.
/// - A collection of templates for the projector to use to extract values.
/// - A list of columns to use for grouping. Grouping is a property of the projection!
struct ProjectedElements {
sql_projection: Projection,
pre_aggregate_projection: Option<Projection>,
templates: Vec<TypedIndex>,
group_by: Vec<GroupBy>,
}
/// Walk an iterator of `Element`s, collecting projector templates and columns.
///
/// Returns a pair: the SQL projection (which should always be a `Projection::Columns`)
/// Returns a `ProjectedElements`, which combines SQL projections
/// and a `Vec` of `TypedIndex` 'keys' to use when looking up values.
///
/// Callers must ensure that every `Element` is distinct -- a query like
@ -334,26 +602,56 @@ pub fn projected_column_for_var(var: &Variable, cc: &ConjoiningClauses) -> (Proj
fn project_elements<'a, I: IntoIterator<Item = &'a Element>>(
count: usize,
elements: I,
query: &AlgebraicQuery) -> Result<(Projection, Vec<TypedIndex>)> {
query: &AlgebraicQuery) -> Result<ProjectedElements> {
// Give a little padding for type tags.
let mut inner_projection = Vec::with_capacity(count + 2);
// Everything in the outer query will _either_ be an aggregate operation
// _or_ a reference to a name projected from the inner.
// We'll expand them later.
let mut outer_projection: Vec<Either<Name, ProjectedColumn>> = Vec::with_capacity(count + 2);
let mut cols = Vec::with_capacity(count);
let mut i: i32 = 0;
let mut min_max_count: usize = 0;
let mut corresponding_count: usize = 0;
let mut templates = vec![];
let mut with = query.with.clone();
let mut aggregates = false;
// Any variable that appears intact in the :find clause, not inside an aggregate expression.
// "Query variables not in aggregate expressions will group the results and appear intact
// in the result."
// We use an ordered set here so that we group in the correct order.
let mut outer_variables = IndexSet::new();
// Any variable that we are projecting from the inner query.
let mut inner_variables = BTreeSet::new();
for e in elements {
if let &Element::Corresponding(_) = e {
corresponding_count += 1;
}
match e {
// Each time we come across a variable, we push a SQL column
// into the SQL projection, aliased to the name of the variable,
// and we push an annotated index into the projector.
&Element::Variable(ref var) => {
// If we're projecting this, we don't need it in :with.
with.remove(var);
&Element::Variable(ref var) |
&Element::Corresponding(ref var) => {
if outer_variables.contains(var) {
eprintln!("Warning: duplicate variable {} in query.", var);
}
let (projected_column, maybe_type) = projected_column_for_var(&var, &query.cc);
cols.push(projected_column);
if let Some(ty) = maybe_type {
let tag = ty.value_type_tag();
// TODO: it's an error to have `[:find ?x (the ?x) …]`.
outer_variables.insert(var.clone());
inner_variables.insert(var.clone());
let (projected_column, type_set) = projected_column_for_var(&var, &query.cc)?;
outer_projection.push(Either::Left(projected_column.1.clone()));
inner_projection.push(projected_column);
if let Some(tag) = type_set.unique_type_tag() {
templates.push(TypedIndex::Known(i, tag));
i += 1; // We used one SQL column.
} else {
@ -362,25 +660,213 @@ fn project_elements<'a, I: IntoIterator<Item = &'a Element>>(
// Also project the type from the SQL query.
let (type_column, type_name) = candidate_type_column(&query.cc, &var);
cols.push(ProjectedColumn(type_column, type_name));
inner_projection.push(ProjectedColumn(type_column, type_name.clone()));
outer_projection.push(Either::Left(type_name));
}
},
&Element::Aggregate(ref a) => {
if let Some(simple) = a.to_simple() {
aggregates = true;
use SimpleAggregationOp::*;
match simple.op {
Max | Min => {
min_max_count += 1;
},
Avg | Count | Sum => (),
}
// When we encounter a simple aggregate -- one in which the aggregation can be
// implemented in SQL, on a single variable -- we just push the SQL aggregation op.
// We must ensure the following:
// - There's a column for the var.
// - The type of the var is known to be restricted to a sensible input set
// (not necessarily a single type, but e.g., all vals must be Double or Long).
// - The type set must be appropriate for the operation. E.g., `Sum` is not a
// meaningful operation on instants.
let (projected_column, return_type) = projected_column_for_simple_aggregate(&simple, &query.cc)?;
outer_projection.push(Either::Right(projected_column));
if !inner_variables.contains(&simple.var) {
inner_variables.insert(simple.var.clone());
let (projected_column, _type_set) = projected_column_for_var(&simple.var, &query.cc)?;
inner_projection.push(projected_column);
if query.cc.known_type_set(&simple.var).unique_type_tag().is_none() {
// Also project the type from the SQL query.
let (type_column, type_name) = candidate_type_column(&query.cc, &simple.var);
inner_projection.push(ProjectedColumn(type_column, type_name.clone()));
}
}
// We might regret using the type tag here instead of the `ValueType`.
templates.push(TypedIndex::Known(i, return_type.value_type_tag()));
i += 1;
} else {
// TODO: complex aggregates.
bail!(ErrorKind::NotYetImplemented("complex aggregates".into()));
}
},
}
}
match (min_max_count, corresponding_count) {
(0, 0) | (_, 0) => {},
(0, _) => {
eprintln!("Warning: used `(the ?var)` without `min` or `max`.");
},
(1, _) => {
// This is the success case!
},
(n, c) => {
bail!(ErrorKind::AmbiguousAggregates(n, c));
},
}
// Anything used in ORDER BY (which we're given in `named_projection`)
// needs to be in the SQL column list so we can refer to it by name.
//
// They don't affect projection.
//
// If a variable is of a non-fixed type, also project the type tag column, so we don't
// accidentally unify across types when considering uniqueness!
for var in query.named_projection.iter() {
if outer_variables.contains(var) {
continue;
}
// If it's a fixed value, we need do nothing further.
if query.cc.is_value_bound(&var) {
continue;
}
let already_inner = inner_variables.contains(&var);
let (column, name) = candidate_column(&query.cc, &var)?;
if !already_inner {
inner_projection.push(ProjectedColumn(column, name.clone()));
inner_variables.insert(var.clone());
}
outer_projection.push(Either::Left(name));
outer_variables.insert(var.clone());
// We don't care if a column has a single _type_, we care if it has a single type _tag_,
// because that's what we'll use if we're projecting. E.g., Long and Double.
// Single type implies single type tag, and is cheaper, so we check that first.
let types = query.cc.known_type_set(&var);
if !types.has_unique_type_tag() {
let (type_column, type_name) = candidate_type_column(&query.cc, &var);
if !already_inner {
inner_projection.push(ProjectedColumn(type_column, type_name.clone()));
}
outer_projection.push(Either::Left(type_name));
}
}
if !aggregates {
// We're done -- we never need to group unless we're aggregating.
return Ok(ProjectedElements {
sql_projection: Projection::Columns(inner_projection),
pre_aggregate_projection: None,
templates,
group_by: vec![],
});
}
// OK, on to aggregates.
// We need to produce two SQL projection lists: one for an inner query and one for the outer.
//
// The inner serves these purposes:
// - Projecting variables to avoid duplicates being elided. (:with)
// - Making bindings available to the outermost query for projection, ordering, and grouping.
//
// The outer is consumed by the projector.
//
// We will also be producing:
// - A GROUP BY list to group the output of the inner query by non-aggregate variables
// so that it can be correctly aggregated.
// Turn this collection of vars into a collection of columns from the query.
// We don't allow grouping on anything but a variable bound in the query.
// We group by tag if necessary.
let mut group_by = Vec::with_capacity(outer_variables.len() + 2);
for var in outer_variables.into_iter() {
if query.cc.is_value_bound(&var) {
continue;
}
// The GROUP BY goes outside, but it needs every variable and type tag to be
// projected from inside. Collect in both directions here.
let name = VariableColumn::Variable(var.clone()).column_name();
group_by.push(GroupBy::ProjectedColumn(name));
let needs_type_projection = !query.cc.known_type_set(&var).has_unique_type_tag();
let already_inner = inner_variables.contains(&var);
if !already_inner {
let (column, name) = candidate_column(&query.cc, &var)?;
inner_projection.push(ProjectedColumn(column, name.clone()));
}
if needs_type_projection {
let type_name = VariableColumn::VariableTypeTag(var.clone()).column_name();
if !already_inner {
let type_col = query.cc
.extracted_types
.get(&var)
.cloned()
.ok_or_else(|| ErrorKind::NoTypeAvailableForVariable(var.name().clone()))?;
inner_projection.push(ProjectedColumn(ColumnOrExpression::Column(type_col), type_name.clone()));
}
group_by.push(GroupBy::ProjectedColumn(type_name));
};
}
for var in query.with.iter() {
// We never need to project a constant.
if query.cc.is_value_bound(&var) {
continue;
}
// We don't need to add inner projections for :with if they are already there.
if !inner_variables.contains(&var) {
let (projected_column, type_set) = projected_column_for_var(&var, &query.cc)?;
inner_projection.push(projected_column);
if type_set.unique_type_tag().is_none() {
// Also project the type from the SQL query.
let (type_column, type_name) = candidate_type_column(&query.cc, &var);
inner_projection.push(ProjectedColumn(type_column, type_name.clone()));
}
}
}
for var in with {
// We need to collect these into the SQL column list, but they don't affect projection.
// If a variable is of a non-fixed type, also project the type tag column, so we don't
// accidentally unify across types when considering uniqueness!
let (column, name) = candidate_column(&query.cc, &var);
cols.push(ProjectedColumn(column, name));
if query.cc.known_type(&var).is_none() {
let (type_column, type_name) = candidate_type_column(&query.cc, &var);
cols.push(ProjectedColumn(type_column, type_name));
}
}
// At this point we know we have a double-layer projection. Collect the outer.
//
// If we have an inner and outer layer, the inner layer will name its
// variables, and the outer will re-project them.
// If we only have one layer, then the outer will do the naming.
// (We could try to not use names in the inner query, but then what would we do for
// `ground` and known values?)
// Walk the projection, switching the outer columns to use the inner names.
Ok((Projection::Columns(cols), templates))
let outer_projection = outer_projection.into_iter().map(|c| {
match c {
Either::Left(name) => {
ProjectedColumn(ColumnOrExpression::ExistingColumn(name.clone()),
name)
},
Either::Right(pc) => pc,
}
}).collect();
Ok(ProjectedElements {
sql_projection: Projection::Columns(outer_projection),
pre_aggregate_projection: Some(Projection::Columns(inner_projection)),
templates,
group_by,
})
}
pub trait Projector {
@ -436,12 +922,14 @@ impl ScalarProjector {
}
}
fn combine(spec: Rc<FindSpec>, sql: Projection, mut templates: Vec<TypedIndex>) -> Result<CombinedProjection> {
let template = templates.pop().expect("Expected a single template");
fn combine(spec: Rc<FindSpec>, mut elements: ProjectedElements) -> Result<CombinedProjection> {
let template = elements.templates.pop().expect("Expected a single template");
Ok(CombinedProjection {
sql_projection: sql,
sql_projection: elements.sql_projection,
pre_aggregate_projection: elements.pre_aggregate_projection,
datalog_projector: Box::new(ScalarProjector::with_template(spec, template)),
distinct: false,
group_by_cols: elements.group_by,
})
}
}
@ -486,6 +974,8 @@ impl TupleProjector {
// This is exactly the same as for rel.
fn collect_bindings<'a, 'stmt>(&self, row: Row<'a, 'stmt>) -> Result<Vec<TypedValue>> {
// There will be at least as many SQL columns as Datalog columns.
// gte 'cos we might be querying extra columns for ordering.
// The templates will take care of ignoring columns.
assert!(row.column_count() >= self.len as i32);
self.templates
.iter()
@ -493,12 +983,14 @@ impl TupleProjector {
.collect::<Result<Vec<TypedValue>>>()
}
fn combine(spec: Rc<FindSpec>, column_count: usize, sql: Projection, templates: Vec<TypedIndex>) -> Result<CombinedProjection> {
let p = TupleProjector::with_templates(spec, column_count, templates);
fn combine(spec: Rc<FindSpec>, column_count: usize, elements: ProjectedElements) -> Result<CombinedProjection> {
let p = TupleProjector::with_templates(spec, column_count, elements.templates);
Ok(CombinedProjection {
sql_projection: sql,
sql_projection: elements.sql_projection,
pre_aggregate_projection: elements.pre_aggregate_projection,
datalog_projector: Box::new(p),
distinct: false,
group_by_cols: elements.group_by,
})
}
}
@ -546,6 +1038,8 @@ impl RelProjector {
fn collect_bindings<'a, 'stmt>(&self, row: Row<'a, 'stmt>) -> Result<Vec<TypedValue>> {
// There will be at least as many SQL columns as Datalog columns.
// gte 'cos we might be querying extra columns for ordering.
// The templates will take care of ignoring columns.
assert!(row.column_count() >= self.len as i32);
self.templates
.iter()
@ -553,12 +1047,21 @@ impl RelProjector {
.collect::<Result<Vec<TypedValue>>>()
}
fn combine(spec: Rc<FindSpec>, column_count: usize, sql: Projection, templates: Vec<TypedIndex>) -> Result<CombinedProjection> {
let p = RelProjector::with_templates(spec, column_count, templates);
fn combine(spec: Rc<FindSpec>, column_count: usize, elements: ProjectedElements) -> Result<CombinedProjection> {
let p = RelProjector::with_templates(spec, column_count, elements.templates);
// If every column yields only one value, or if this is an aggregate query
// (because by definition every column in an aggregate query is either
// aggregated or is a variable _upon which we group_), then don't bother
// with DISTINCT.
let already_distinct = elements.pre_aggregate_projection.is_some() ||
p.columns().all(|e| e.is_unit());
Ok(CombinedProjection {
sql_projection: sql,
sql_projection: elements.sql_projection,
pre_aggregate_projection: elements.pre_aggregate_projection,
datalog_projector: Box::new(p),
distinct: true,
distinct: !already_distinct,
group_by_cols: elements.group_by,
})
}
}
@ -597,12 +1100,22 @@ impl CollProjector {
}
}
fn combine(spec: Rc<FindSpec>, sql: Projection, mut templates: Vec<TypedIndex>) -> Result<CombinedProjection> {
let template = templates.pop().expect("Expected a single template");
fn combine(spec: Rc<FindSpec>, mut elements: ProjectedElements) -> Result<CombinedProjection> {
let template = elements.templates.pop().expect("Expected a single template");
let p = CollProjector::with_template(spec, template);
// If every column yields only one value, or if this is an aggregate query
// (because by definition every column in an aggregate query is either
// aggregated or is a variable _upon which we group_), then don't bother
// with DISTINCT.
let already_distinct = elements.pre_aggregate_projection.is_some() ||
p.columns().all(|e| e.is_unit());
Ok(CombinedProjection {
sql_projection: sql,
datalog_projector: Box::new(CollProjector::with_template(spec, template)),
distinct: true,
sql_projection: elements.sql_projection,
pre_aggregate_projection: elements.pre_aggregate_projection,
datalog_projector: Box::new(p),
distinct: !already_distinct,
group_by_cols: elements.group_by,
})
}
}
@ -626,19 +1139,33 @@ impl Projector for CollProjector {
}
}
/// Combines the two things you need to turn a query into SQL and turn its results into
/// `QueryResults`.
/// Combines the things you need to turn a query into SQL and turn its results into
/// `QueryResults`: SQL-related projection information (`DISTINCT`, columns, etc.) and
/// a Datalog projector that turns SQL into structures.
pub struct CombinedProjection {
/// A SQL projection, mapping columns mentioned in the body of the query to columns in the
/// output.
pub sql_projection: Projection,
/// If a query contains aggregates, we need to generate a nested subquery: an inner query
/// that returns our distinct variable bindings (and any `:with` vars), and an outer query
/// that applies aggregation. That's so we can put `DISTINCT` in the inner query and apply
/// aggregation afterwards -- `SELECT DISTINCT count(foo)` counts _then_ uniques, and we need
/// the opposite to implement Datalog distinct semantics.
/// If this is the case, `sql_projection` will be the outer query's projection list, and
/// `pre_aggregate_projection` will be the inner.
/// If the query doesn't use aggregation, this field will be `None`.
pub pre_aggregate_projection: Option<Projection>,
/// A Datalog projection. This consumes rows of the appropriate shape (as defined by
/// the SQL projection) to yield one of the four kinds of Datalog query result.
pub datalog_projector: Box<Projector>,
/// True if this query requires the SQL query to include DISTINCT.
pub distinct: bool,
// A list of column names to use as a GROUP BY clause.
pub group_by_cols: Vec<GroupBy>,
}
impl CombinedProjection {
@ -665,7 +1192,17 @@ pub fn query_projection(query: &AlgebraicQuery) -> Result<Either<ConstantProject
if query.is_fully_unit_bound() {
// Do a few gyrations to produce empty results of the right kind for the query.
let variables: BTreeSet<Variable> = spec.columns().map(|e| match e { &Element::Variable(ref var) => var.clone() }).collect();
let variables: BTreeSet<Variable> = spec.columns()
.map(|e| match e {
&Element::Variable(ref var) |
&Element::Corresponding(ref var) => var.clone(),
&Element::Aggregate(ref _agg) => {
// TODO: static computation of aggregates, then
// implement the condition in `is_fully_bound`.
unreachable!();
},
})
.collect();
// TODO: error handling
let results = QueryOutput::from_constants(&spec, query.cc.value_bindings(&variables));
@ -679,25 +1216,25 @@ pub fn query_projection(query: &AlgebraicQuery) -> Result<Either<ConstantProject
} else {
match *query.find_spec {
FindColl(ref element) => {
let (cols, templates) = project_elements(1, iter::once(element), query)?;
CollProjector::combine(spec, cols, templates).map(|p| p.flip_distinct_for_limit(&query.limit))
let elements = project_elements(1, iter::once(element), query)?;
CollProjector::combine(spec, elements).map(|p| p.flip_distinct_for_limit(&query.limit))
},
FindScalar(ref element) => {
let (cols, templates) = project_elements(1, iter::once(element), query)?;
ScalarProjector::combine(spec, cols, templates)
let elements = project_elements(1, iter::once(element), query)?;
ScalarProjector::combine(spec, elements)
},
FindRel(ref elements) => {
let column_count = query.find_spec.expected_column_count();
let (cols, templates) = project_elements(column_count, elements, query)?;
RelProjector::combine(spec, column_count, cols, templates).map(|p| p.flip_distinct_for_limit(&query.limit))
let elements = project_elements(column_count, elements, query)?;
RelProjector::combine(spec, column_count, elements).map(|p| p.flip_distinct_for_limit(&query.limit))
},
FindTuple(ref elements) => {
let column_count = query.find_spec.expected_column_count();
let (cols, templates) = project_elements(column_count, elements, query)?;
TupleProjector::combine(spec, column_count, cols, templates)
let elements = project_elements(column_count, elements, query)?;
TupleProjector::combine(spec, column_count, elements)
},
}.map(Either::Right)
}

View file

@ -0,0 +1,88 @@
// 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.
extern crate mentat_core;
extern crate mentat_query;
extern crate mentat_query_algebrizer;
extern crate mentat_query_parser;
extern crate mentat_query_projector;
use mentat_core::{
Attribute,
Entid,
Schema,
ValueType,
};
use mentat_query_parser::{
parse_find_string,
};
use mentat_query::{
NamespacedKeyword,
};
use mentat_query_algebrizer::{
Known,
algebrize,
};
use mentat_query_projector::{
query_projection,
};
// These are helpers that tests use to build Schema instances.
fn associate_ident(schema: &mut Schema, i: NamespacedKeyword, e: Entid) {
schema.entid_map.insert(e, i.clone());
schema.ident_map.insert(i.clone(), e);
}
fn add_attribute(schema: &mut Schema, e: Entid, a: Attribute) {
schema.attribute_map.insert(e, a);
}
fn prepopulated_schema() -> Schema {
let mut schema = Schema::default();
associate_ident(&mut schema, NamespacedKeyword::new("foo", "name"), 65);
associate_ident(&mut schema, NamespacedKeyword::new("foo", "age"), 68);
associate_ident(&mut schema, NamespacedKeyword::new("foo", "height"), 69);
add_attribute(&mut schema, 65, Attribute {
value_type: ValueType::String,
multival: false,
..Default::default()
});
add_attribute(&mut schema, 68, Attribute {
value_type: ValueType::Long,
multival: false,
..Default::default()
});
add_attribute(&mut schema, 69, Attribute {
value_type: ValueType::Long,
multival: false,
..Default::default()
});
schema
}
#[test]
fn test_aggregate_unsuitable_type() {
let schema = prepopulated_schema();
let query = r#"[:find (avg ?e)
:where
[?e :foo/age ?a]]"#;
// While the query itself algebrizes and parses…
let parsed = parse_find_string(query).expect("query input to have parsed");
let algebrized = algebrize(Known::for_schema(&schema), parsed).expect("query algebrizes");
// … when we look at the projection list, we cannot reconcile the types.
assert!(query_projection(&algebrized).is_err());
}

View file

@ -15,11 +15,11 @@ extern crate mentat_query_algebrizer;
extern crate mentat_sql;
use std::boxed::Box;
use mentat_core::{
Entid,
TypedValue,
SQLTypeAffinity,
TypedValue,
ValueType,
};
use mentat_query::{
@ -57,10 +57,16 @@ use mentat_sql::{
/// implementation for each storage backend. Passing `TypedValue`s here allows for that.
pub enum ColumnOrExpression {
Column(QualifiedAlias),
ExistingColumn(Name),
Entid(Entid), // Because it's so common.
Integer(i32), // We use these for type codes etc.
Long(i64),
Value(TypedValue),
Expression(Box<Expression>, ValueType), // Track the return type.
}
pub enum Expression {
Unary { sql_op: &'static str, arg: ColumnOrExpression },
}
/// `QueryValue` and `ColumnOrExpression` are almost identical… merge somehow?
@ -85,6 +91,26 @@ pub enum Projection {
One,
}
#[derive(Debug, PartialEq, Eq)]
pub enum GroupBy {
ProjectedColumn(Name),
QueryColumn(QualifiedAlias),
// TODO: non-projected expressions, etc.
}
impl QueryFragment for GroupBy {
fn push_sql(&self, out: &mut QueryBuilder) -> BuildQueryResult {
match self {
&GroupBy::ProjectedColumn(ref name) => {
out.push_identifier(name.as_str())
},
&GroupBy::QueryColumn(ref qa) => {
qualified_alias_push_sql(out, qa)
},
}
}
}
#[derive(Copy, Clone)]
pub struct Op(pub &'static str); // TODO: we can do better than this!
@ -190,6 +216,7 @@ pub struct SelectQuery {
pub projection: Projection,
pub from: FromClause,
pub constraints: Vec<Constraint>,
pub group_by: Vec<GroupBy>,
pub order: Vec<OrderBy>,
pub limit: Limit,
}
@ -262,10 +289,11 @@ impl QueryFragment for ColumnOrExpression {
fn push_sql(&self, out: &mut QueryBuilder) -> BuildQueryResult {
use self::ColumnOrExpression::*;
match self {
&Column(QualifiedAlias(ref table, ref column)) => {
out.push_identifier(table.as_str())?;
out.push_sql(".");
push_column(out, column)
&Column(ref qa) => {
qualified_alias_push_sql(out, qa)
},
&ExistingColumn(ref alias) => {
out.push_identifier(alias.as_str())
},
&Entid(entid) => {
out.push_sql(entid.to_string().as_str());
@ -282,6 +310,23 @@ impl QueryFragment for ColumnOrExpression {
&Value(ref v) => {
out.push_typed_value(v)
},
&Expression(ref e, _) => {
e.push_sql(out)
},
}
}
}
impl QueryFragment for Expression {
fn push_sql(&self, out: &mut QueryBuilder) -> BuildQueryResult {
match self {
&Expression::Unary { ref sql_op, ref arg } => {
out.push_sql(sql_op); // No need to escape built-ins.
out.push_sql("(");
arg.push_sql(out)?;
out.push_sql(")");
Ok(())
},
}
}
}
@ -368,10 +413,8 @@ impl QueryFragment for Constraint {
Ok(())
},
&NotExists { ref subquery } => {
out.push_sql("NOT EXISTS (");
subquery.push_sql(out)?;
out.push_sql(")");
Ok(())
out.push_sql("NOT EXISTS ");
subquery.push_sql(out)
},
&TypeCheck { ref value, ref affinity } => {
out.push_sql("typeof(");
@ -397,6 +440,13 @@ impl QueryFragment for JoinOp {
}
}
// We don't own QualifiedAlias or QueryFragment, so we can't implement the trait.
fn qualified_alias_push_sql(out: &mut QueryBuilder, qa: &QualifiedAlias) -> BuildQueryResult {
out.push_identifier(qa.0.as_str())?;
out.push_sql(".");
push_column(out, &qa.1)
}
// We don't own SourceAlias or QueryFragment, so we can't implement the trait.
fn source_alias_push_sql(out: &mut QueryBuilder, sa: &SourceAlias) -> BuildQueryResult {
let &SourceAlias(ref table, ref alias) = sa;
@ -440,7 +490,10 @@ impl QueryFragment for TableOrSubquery {
out.push_identifier(table_alias.as_str())
},
&Subquery(ref subquery) => {
subquery.push_sql(out)
out.push_sql("(");
subquery.push_sql(out)?;
out.push_sql(")");
Ok(())
},
&Values(ref values, ref table_alias) => {
// XXX: does this work for Values::Unnamed?
@ -545,6 +598,16 @@ impl QueryFragment for SelectQuery {
{ out.push_sql(" AND ") });
}
match &self.group_by {
group_by if !group_by.is_empty() => {
out.push_sql(" GROUP BY ");
interpose!(group, group_by,
{ group.push_sql(out)? },
{ out.push_sql(", ") });
},
_ => {},
}
if !self.order.is_empty() {
out.push_sql(" ORDER BY ");
interpose!(&OrderBy(ref dir, ref var), self.order,
@ -745,6 +808,7 @@ mod tests {
right: ColumnOrExpression::Entid(65536),
},
],
group_by: vec![],
order: vec![],
limit: Limit::None,
};

View file

@ -11,6 +11,7 @@
use mentat_core::{
SQLTypeAffinity,
SQLValueType,
SQLValueTypeSet,
TypedValue,
ValueType,
ValueTypeTag,
@ -56,6 +57,7 @@ use mentat_query_sql::{
ColumnOrExpression,
Constraint,
FromClause,
GroupBy,
Op,
ProjectedColumn,
Projection,
@ -287,7 +289,8 @@ fn table_for_computed(computed: ComputedTable, alias: TableAlias) -> TableOrSubq
// project it as the variable name.
// E.g., SELECT datoms03.v AS `?x`.
for var in projection.iter() {
let (projected_column, maybe_type) = projected_column_for_var(var, &cc);
// TODO: chain results out.
let (projected_column, type_set) = projected_column_for_var(var, &cc).expect("every var to be bound");
columns.push(projected_column);
// Similarly, project type tags if they're not known conclusively in the
@ -295,10 +298,10 @@ fn table_for_computed(computed: ComputedTable, alias: TableAlias) -> TableOrSubq
// Assumption: we'll never need to project a tag without projecting the value of a variable.
if type_extraction.contains(var) {
let expression =
if let Some(ty) = maybe_type {
if let Some(tag) = type_set.unique_type_tag() {
// If we know the type for sure, just project the constant.
// SELECT datoms03.v AS `?x`, 10 AS `?x_value_type_tag`
ColumnOrExpression::Integer(ty.value_type_tag())
ColumnOrExpression::Integer(tag)
} else {
// Otherwise, we'll have an established type binding! This'll be
// either a datoms table or, recursively, a subquery. Project
@ -319,7 +322,7 @@ fn table_for_computed(computed: ComputedTable, alias: TableAlias) -> TableOrSubq
// Each arm simply turns into a subquery.
// The SQL translation will stuff "UNION" between each arm.
let projection = Projection::Columns(columns);
cc_to_select_query(projection, cc, false, None, Limit::None)
cc_to_select_query(projection, cc, false, vec![], None, Limit::None)
}).collect(),
alias)
},
@ -340,6 +343,7 @@ fn empty_query() -> SelectQuery {
distinct: false,
projection: Projection::One,
from: FromClause::Nothing,
group_by: vec![],
constraints: vec![],
order: vec![],
limit: Limit::None,
@ -352,6 +356,7 @@ fn empty_query() -> SelectQuery {
fn cc_to_select_query(projection: Projection,
cc: ConjoiningClauses,
distinct: bool,
group_by: Vec<GroupBy>,
order: Option<Vec<OrderBy>>,
limit: Limit) -> SelectQuery {
let from = if cc.from.is_empty() {
@ -387,6 +392,7 @@ fn cc_to_select_query(projection: Projection,
distinct: distinct,
projection: projection,
from: from,
group_by: group_by,
constraints: cc.wheres
.into_iter()
.map(|c| c.to_constraint())
@ -403,7 +409,31 @@ pub fn cc_to_exists(cc: ConjoiningClauses) -> SelectQuery {
// In this case we can produce a very simple query that returns no results.
empty_query()
} else {
cc_to_select_query(Projection::One, cc, false, None, Limit::None)
cc_to_select_query(Projection::One, cc, false, vec![], None, Limit::None)
}
}
/// Take a query and wrap it as a subquery of a new query with the provided projection list.
/// All limits, ordering, and grouping move to the outer query. The inner query is marked as
/// distinct.
fn re_project(mut inner: SelectQuery, projection: Projection) -> SelectQuery {
let outer_distinct = inner.distinct;
inner.distinct = true;
let group_by = inner.group_by;
inner.group_by = vec![];
let order_by = inner.order;
inner.order = vec![];
let limit = inner.limit;
inner.limit = Limit::None;
SelectQuery {
distinct: outer_distinct,
projection: projection,
from: FromClause::TableList(TableList(vec![TableOrSubquery::Subquery(Box::new(inner))])),
constraints: vec![],
group_by: group_by,
order: order_by,
limit: limit,
}
}
@ -414,10 +444,30 @@ pub fn query_to_select(query: AlgebraicQuery) -> Result<ProjectedSelect> {
// SQL-based aggregation -- `SELECT SUM(datoms00.e)` -- is fine.
query_projection(&query).map(|e| match e {
Either::Left(constant) => ProjectedSelect::Constant(constant),
Either::Right(CombinedProjection { sql_projection, datalog_projector, distinct, }) => {
let q = cc_to_select_query(sql_projection, query.cc, distinct, query.order, query.limit);
Either::Right(CombinedProjection {
sql_projection,
pre_aggregate_projection,
datalog_projector,
distinct,
group_by_cols,
}) => {
ProjectedSelect::Query {
query: q,
query: match pre_aggregate_projection {
// If we know we need a nested query for aggregation, build that first.
Some(pre_aggregate) => {
let inner = cc_to_select_query(pre_aggregate,
query.cc,
distinct,
group_by_cols,
query.order,
query.limit);
let outer = re_project(inner, sql_projection);
outer
},
None => {
cc_to_select_query(sql_projection, query.cc, distinct, group_by_cols, query.order, query.limit)
},
},
projector: datalog_projector,
}
},

View file

@ -662,13 +662,13 @@ fn test_with_without_aggregate() {
// Known type.
let query = r#"[:find ?x :with ?y :where [?x :foo/bar ?y]]"#;
let SQLQuery { sql, args } = translate(&schema, query);
assert_eq!(sql, "SELECT DISTINCT `datoms00`.e AS `?x`, `datoms00`.v AS `?y` FROM `datoms` AS `datoms00` WHERE `datoms00`.a = 99");
assert_eq!(sql, "SELECT DISTINCT `datoms00`.e AS `?x` FROM `datoms` AS `datoms00` WHERE `datoms00`.a = 99");
assert_eq!(args, vec![]);
// Unknown type.
let query = r#"[:find ?x :with ?y :where [?x _ ?y]]"#;
let SQLQuery { sql, args } = translate(&schema, query);
assert_eq!(sql, "SELECT DISTINCT `all_datoms00`.e AS `?x`, `all_datoms00`.v AS `?y`, `all_datoms00`.value_type_tag AS `?y_value_type_tag` FROM `all_datoms` AS `all_datoms00`");
assert_eq!(sql, "SELECT DISTINCT `all_datoms00`.e AS `?x` FROM `all_datoms` AS `all_datoms00`");
assert_eq!(args, vec![]);
}
@ -1081,3 +1081,37 @@ fn test_instant_range() {
AND `datoms00`.v > 1497574601257000");
assert_eq!(args, vec![]);
}
#[test]
fn test_project_aggregates() {
let schema = prepopulated_typed_schema(ValueType::Long);
let query = r#"[:find ?e (max ?t)
:where
[?e :foo/bar ?t]]"#;
let SQLQuery { sql, args } = translate(&schema, query);
// No outer DISTINCT: we aggregate or group by every variable.
assert_eq!(sql, "SELECT `?e` AS `?e`, max(`?t`) AS `(max ?t)` \
FROM \
(SELECT DISTINCT \
`datoms00`.e AS `?e`, \
`datoms00`.v AS `?t` \
FROM `datoms` AS `datoms00` \
WHERE `datoms00`.a = 99) \
GROUP BY `?e`");
assert_eq!(args, vec![]);
let query = r#"[:find (max ?t)
:with ?e
:where
[?e :foo/bar ?t]]"#;
let SQLQuery { sql, args } = translate(&schema, query);
assert_eq!(sql, "SELECT max(`?t`) AS `(max ?t)` \
FROM \
(SELECT DISTINCT \
`datoms00`.v AS `?t`, \
`datoms00`.e AS `?e` \
FROM `datoms` AS `datoms00` \
WHERE `datoms00`.a = 99)");
assert_eq!(args, vec![]);
}

View file

@ -153,6 +153,12 @@ impl QueryFunction {
}
}
impl std::fmt::Display for QueryFunction {
fn fmt(&self, f: &mut std::fmt::Formatter) -> std::fmt::Result {
write!(f, "{}", self.0)
}
}
#[derive(Clone, Debug, Eq, PartialEq)]
pub enum Direction {
Ascending,
@ -266,6 +272,26 @@ impl FromValue<FnArg> for FnArg {
}
}
// For display in column headings in the repl.
impl std::fmt::Display for FnArg {
fn fmt(&self, f: &mut std::fmt::Formatter) -> std::fmt::Result {
match self {
&FnArg::Variable(ref var) => write!(f, "{}", var),
&FnArg::SrcVar(ref var) => {
if var == &SrcVar::DefaultSrc {
write!(f, "$")
} else {
write!(f, "{:?}", var)
}
},
&FnArg::EntidOrInteger(entid) => write!(f, "{}", entid),
&FnArg::IdentOrKeyword(ref kw) => write!(f, "{}", kw),
&FnArg::Constant(ref constant) => write!(f, "{:?}", constant),
&FnArg::Vector(ref vec) => write!(f, "{:?}", vec),
}
}
}
impl FnArg {
pub fn as_variable(&self) -> Option<&Variable> {
match self {
@ -435,20 +461,37 @@ pub struct Pull {
}
*/
/*
#[derive(Debug, Eq, PartialEq)]
pub struct Aggregate {
pub fn_name: String,
pub func: QueryFunction,
pub args: Vec<FnArg>,
}
*/
#[derive(Debug, Eq, PartialEq)]
pub enum Element {
Variable(Variable),
// Aggregate(Aggregate), // TODO
Aggregate(Aggregate),
/// In a query with a `max` or `min` aggregate, a corresponding variable
/// (indicated in the query with `(the ?var)`, is guaranteed to come from
/// the row that provided the max or min value. Queries with more than one
/// `max` or `min` cannot yield predictable behavior, and will err during
/// algebrizing.
Corresponding(Variable),
// Pull(Pull), // TODO
}
impl Element {
/// Returns true if the element must yield only one value.
pub fn is_unit(&self) -> bool {
match self {
&Element::Variable(_) => false,
&Element::Aggregate(_) => true,
&Element::Corresponding(_) => true,
}
}
}
impl From<Variable> for Element {
fn from(x: Variable) -> Element {
Element::Variable(x)
@ -461,6 +504,16 @@ impl std::fmt::Display for Element {
&Element::Variable(ref var) => {
write!(f, "{}", var)
},
&Element::Aggregate(ref agg) => {
match agg.args.len() {
0 => write!(f, "({})", agg.func),
1 => write!(f, "({} {})", agg.func, agg.args[0]),
_ => write!(f, "({} {:?})", agg.func, agg.args),
}
},
&Element::Corresponding(ref var) => {
write!(f, "(the {})", var)
},
}
}
}

View file

@ -173,6 +173,13 @@ impl Store {
sqlite: connection,
})
}
pub fn transact(&mut self, transaction: &str) -> Result<TxReport> {
let mut ip = self.begin_transaction()?;
let report = ip.transact(transaction)?;
ip.commit()?;
Ok(report)
}
}
pub trait Queryable {

View file

@ -15,7 +15,11 @@ extern crate time;
extern crate mentat;
extern crate mentat_core;
extern crate mentat_db;
// TODO: when we switch to `failure`, make this more humane.
extern crate mentat_query_algebrizer; // For errors.
extern crate mentat_query_projector; // For errors.
extern crate mentat_query_translator; // For errors.
use std::str::FromStr;
@ -32,10 +36,13 @@ use mentat_core::{
};
use mentat::{
IntoResult,
NamespacedKeyword,
PlainSymbol,
QueryInputs,
Queryable,
QueryResults,
Store,
Variable,
new_connection,
};
@ -381,7 +388,7 @@ fn test_fulltext() {
_ => panic!("Unexpected results."),
}
},
_ => panic!("Expected query to work."),
r => panic!("Unexpected results {:?}.", r),
}
let a = conn.transact(&mut c, r#"[[:db/add "a" :foo/term "talk"]]"#)
@ -522,7 +529,6 @@ fn test_lookup() {
let fetched_many = conn.lookup_value_for_attribute(&c, *entid, &foo_many).unwrap().unwrap();
assert!(two_longs.contains(&fetched_many));
}
#[test]
fn test_type_reqs() {
let mut c = new_connection("").expect("Couldn't open conn.");
@ -621,3 +627,477 @@ fn test_type_reqs() {
}
};
}
#[test]
fn test_monster_head_aggregates() {
let mut store = Store::open("").expect("opened");
let mut in_progress = store.begin_transaction().expect("began");
in_progress.transact(r#"[
{:db/ident :monster/heads
:db/valueType :db.type/long
:db/cardinality :db.cardinality/one}
{:db/ident :monster/name
:db/valueType :db.type/string
:db/cardinality :db.cardinality/one
:db/index true
:db/unique :db.unique/identity}
{:db/ident :monster/weapon
:db/valueType :db.type/string
:db/cardinality :db.cardinality/many}
]"#).expect("transacted");
in_progress.transact(r#"[
{:monster/heads 1
:monster/name "Medusa"
:monster/weapon "Stony gaze"}
{:monster/heads 1
:monster/name "Cyclops"
:monster/weapon ["Large club" "Mighty arms" "Stompy feet"]}
{:monster/heads 1
:monster/name "Chimera"
:monster/weapon "Goat-like agility"}
{:monster/heads 3
:monster/name "Cerberus"
:monster/weapon ["8-foot Kong®" "Deadly drool"]}
]"#).expect("transacted");
// Without :with, uniqueness applies prior to aggregation, so we get 1 + 3 = 4.
let res = in_progress.q_once("[:find (sum ?heads) . :where [?monster :monster/heads ?heads]]", None)
.expect("results")
.into();
match res {
QueryResults::Scalar(Some(TypedValue::Long(count))) => {
assert_eq!(count, 4);
},
r => panic!("Unexpected result {:?}", r),
};
// With :with, uniqueness includes the monster, so we get 1 + 1 + 1 + 3 = 6.
let res = in_progress.q_once("[:find (sum ?heads) . :with ?monster :where [?monster :monster/heads ?heads]]", None)
.expect("results")
.into();
match res {
QueryResults::Scalar(Some(TypedValue::Long(count))) => {
assert_eq!(count, 6);
},
r => panic!("Unexpected result {:?}", r),
};
// Aggregates group.
let res = in_progress.q_once(r#"[:find ?name (count ?weapon)
:with ?monster
:order (asc ?name)
:where [?monster :monster/name ?name]
[?monster :monster/weapon ?weapon]]"#,
None)
.expect("results")
.into();
match res {
QueryResults::Rel(vals) => {
let expected = vec![
vec!["Cerberus".into(), TypedValue::Long(2)],
vec!["Chimera".into(), TypedValue::Long(1)],
vec!["Cyclops".into(), TypedValue::Long(3)],
vec!["Medusa".into(), TypedValue::Long(1)],
];
assert_eq!(vals, expected);
},
r => panic!("Unexpected result {:?}", r),
};
in_progress.rollback().expect("rolled back");
}
#[test]
fn test_basic_aggregates() {
let mut store = Store::open("").expect("opened");
store.transact(r#"[
{:db/ident :foo/is-vegetarian :db/valueType :db.type/boolean :db/cardinality :db.cardinality/one}
{:db/ident :foo/age :db/valueType :db.type/long :db/cardinality :db.cardinality/one}
{:db/ident :foo/name :db/valueType :db.type/string :db/cardinality :db.cardinality/one}
]"#).unwrap();
let _ids = store.transact(r#"[
[:db/add "a" :foo/name "Alice"]
[:db/add "b" :foo/name "Beli"]
[:db/add "c" :foo/name "Carlos"]
[:db/add "d" :foo/name "Diana"]
[:db/add "a" :foo/is-vegetarian true]
[:db/add "b" :foo/is-vegetarian true]
[:db/add "c" :foo/is-vegetarian false]
[:db/add "d" :foo/is-vegetarian false]
[:db/add "a" :foo/age 14]
[:db/add "b" :foo/age 22]
[:db/add "c" :foo/age 42]
[:db/add "d" :foo/age 28]
]"#).unwrap().tempids;
// Count the number of distinct bindings of `?veg` that are `true` -- namely, one.
// This is not the same as `count-distinct`: note the distinction between
// including `:with` and not.
// In this case, the `DISTINCT` must occur inside the aggregation, not outside it.
/*
Rather than:
SELECT DISTINCT count(1) AS `(count ?veg)`
FROM `datoms` AS `datoms00`
WHERE `datoms00`.a = 65536
AND `datoms00`.v = 1;
our query should be
SELECT DISTINCT count(`?veg`) AS `(count ?veg)`
FROM (
SELECT DISTINCT 1 AS `?veg`
FROM `datoms` AS `datoms00`
WHERE `datoms00`.a = 65536
AND `datoms00`.v = 1
);
*/
let r = store.q_once(r#"[:find (count ?veg)
:where
[_ :foo/is-vegetarian ?veg]
[(ground true) ?veg]]"#, None)
.expect("results")
.into();
match r {
QueryResults::Rel(vals) => {
assert_eq!(vals, vec![vec![TypedValue::Long(1)]]);
},
_ => panic!("Expected rel."),
}
// And this should be
/*
SELECT DISTINCT count(`?veg`) AS `(count ?veg)`
FROM (
SELECT DISTINCT 1 AS `?veg`, `datoms00`.e AS `?person`
FROM `datoms` AS `datoms00`
WHERE `datoms00`.a = 65536
AND `datoms00`.v = 1
);
*/
let r = store.q_once(r#"[:find (count ?veg) .
:with ?person
:where
[?person :foo/is-vegetarian ?veg]
[(ground true) ?veg]]"#, None)
.expect("results")
.into();
match r {
QueryResults::Scalar(Some(val)) => {
assert_eq!(val, TypedValue::Long(2));
},
_ => panic!("Expected scalar."),
}
// What are the oldest and youngest ages?
let r = store.q_once(r#"[:find [(min ?age) (max ?age)]
:where
[_ :foo/age ?age]]"#, None)
.expect("results")
.into();
match r {
QueryResults::Tuple(Some(vals)) => {
assert_eq!(vals,
vec![TypedValue::Long(14),
TypedValue::Long(42)]);
},
_ => panic!("Expected tuple."),
}
// Who's youngest, via order?
let r = store.q_once(r#"[:find [?name ?age]
:order (asc ?age)
:where
[?x :foo/age ?age]
[?x :foo/name ?name]]"#, None)
.expect("results")
.into();
match r {
QueryResults::Tuple(Some(vals)) => {
assert_eq!(vals,
vec![TypedValue::String("Alice".to_string().into()),
TypedValue::Long(14)]);
},
r => panic!("Unexpected results {:?}", r),
}
// Who's oldest, via order?
let r = store.q_once(r#"[:find [?name ?age]
:order (desc ?age)
:where
[?x :foo/age ?age]
[?x :foo/name ?name]]"#, None)
.expect("results")
.into();
match r {
QueryResults::Tuple(Some(vals)) => {
assert_eq!(vals,
vec![TypedValue::String("Carlos".to_string().into()),
TypedValue::Long(42)]);
},
_ => panic!("Expected tuple."),
}
// How many of each age do we have?
// Add an extra person to make this interesting.
store.transact(r#"[{:foo/name "Medusa", :foo/age 28}]"#).expect("transacted");
// If we omit the 'with', we'll get the wrong answer:
let r = store.q_once(r#"[:find ?age (count ?age)
:order (asc ?age)
:where [_ :foo/age ?age]]"#, None)
.expect("results")
.into();
match r {
QueryResults::Rel(vals) => {
assert_eq!(vals, vec![
vec![TypedValue::Long(14), TypedValue::Long(1)],
vec![TypedValue::Long(22), TypedValue::Long(1)],
vec![TypedValue::Long(28), TypedValue::Long(1)],
vec![TypedValue::Long(42), TypedValue::Long(1)],
]);
},
_ => panic!("Expected rel."),
}
// If we include it, we'll get the right one:
let r = store.q_once(r#"[:find ?age (count ?age)
:with ?person
:order (asc ?age)
:where [?person :foo/age ?age]]"#, None)
.expect("results")
.into();
match r {
QueryResults::Rel(vals) => {
assert_eq!(vals, vec![
vec![TypedValue::Long(14), TypedValue::Long(1)],
vec![TypedValue::Long(22), TypedValue::Long(1)],
vec![TypedValue::Long(28), TypedValue::Long(2)],
vec![TypedValue::Long(42), TypedValue::Long(1)],
]);
},
_ => panic!("Expected rel."),
}
}
#[test]
fn test_combinatorial() {
let mut store = Store::open("").expect("opened");
store.transact(r#"[
[:db/add "a" :db/ident :foo/name]
[:db/add "a" :db/valueType :db.type/string]
[:db/add "a" :db/cardinality :db.cardinality/one]
[:db/add "b" :db/ident :foo/dance]
[:db/add "b" :db/valueType :db.type/ref]
[:db/add "b" :db/cardinality :db.cardinality/many]
[:db/add "b" :db/index true]
]"#).unwrap();
store.transact(r#"[
[:db/add "a" :foo/name "Alice"]
[:db/add "b" :foo/name "Beli"]
[:db/add "c" :foo/name "Carlos"]
[:db/add "d" :foo/name "Diana"]
;; Alice danced with Beli twice.
[:db/add "a" :foo/dance "ab"]
[:db/add "b" :foo/dance "ab"]
[:db/add "a" :foo/dance "ba"]
[:db/add "b" :foo/dance "ba"]
;; Carlos danced with Diana.
[:db/add "c" :foo/dance "cd"]
[:db/add "d" :foo/dance "cd"]
;; Alice danced with Diana.
[:db/add "a" :foo/dance "ad"]
[:db/add "d" :foo/dance "ad"]
]"#).unwrap();
// How many different pairings of dancers were there?
// If we just use `!=` (or `differ`), the number is doubled because of symmetry!
assert_eq!(TypedValue::Long(6),
store.q_once(r#"[:find (count ?right) .
:with ?left
:where
[?left :foo/dance ?dance]
[?right :foo/dance ?dance]
[(differ ?left ?right)]]"#, None)
.into_scalar_result()
.expect("scalar results").unwrap());
// SQL addresses this by using `<` instead of `!=` -- by imposing
// an order on values, we can ensure that each pair only appears once, not
// once per permutation.
// It's far from ideal to expose an ordering on entids, because developers
// will come to rely on it. Instead we expose a specific operator: `unpermute`.
// When used in a query that generates permuted pairs of references, this
// ensures that only one permutation is returned for a given pair.
assert_eq!(TypedValue::Long(3),
store.q_once(r#"[:find (count ?right) .
:with ?left
:where
[?left :foo/dance ?dance]
[?right :foo/dance ?dance]
[(unpermute ?left ?right)]]"#, None)
.into_scalar_result()
.expect("scalar results").unwrap());
}
#[test]
fn test_aggregation_implicit_grouping() {
let mut store = Store::open("").expect("opened");
store.transact(r#"[
[:db/add "a" :db/ident :foo/score]
[:db/add "a" :db/valueType :db.type/long]
[:db/add "a" :db/cardinality :db.cardinality/one]
[:db/add "b" :db/ident :foo/name]
[:db/add "b" :db/valueType :db.type/string]
[:db/add "b" :db/cardinality :db.cardinality/one]
[:db/add "c" :db/ident :foo/is-vegetarian]
[:db/add "c" :db/valueType :db.type/boolean]
[:db/add "c" :db/cardinality :db.cardinality/one]
[:db/add "d" :db/ident :foo/play]
[:db/add "d" :db/valueType :db.type/ref]
[:db/add "d" :db/cardinality :db.cardinality/many]
[:db/add "d" :db/index true]
[:db/add "d" :db/unique :db.unique/value]
]"#).unwrap();
let ids = store.transact(r#"[
[:db/add "a" :foo/name "Alice"]
[:db/add "b" :foo/name "Beli"]
[:db/add "c" :foo/name "Carlos"]
[:db/add "d" :foo/name "Diana"]
[:db/add "a" :foo/is-vegetarian true]
[:db/add "b" :foo/is-vegetarian true]
[:db/add "c" :foo/is-vegetarian false]
[:db/add "d" :foo/is-vegetarian false]
[:db/add "aa" :foo/score 14]
[:db/add "ab" :foo/score 99]
[:db/add "ac" :foo/score 14]
[:db/add "ba" :foo/score 22]
[:db/add "bb" :foo/score 11]
[:db/add "ca" :foo/score 42]
[:db/add "da" :foo/score 5]
[:db/add "db" :foo/score 28]
[:db/add "d" :foo/play "da"]
[:db/add "d" :foo/play "db"]
[:db/add "a" :foo/play "aa"]
[:db/add "a" :foo/play "ab"]
[:db/add "a" :foo/play "ac"]
[:db/add "b" :foo/play "ba"]
[:db/add "b" :foo/play "bb"]
[:db/add "c" :foo/play "ca"]
]"#).unwrap().tempids;
// How many different scores were there?
assert_eq!(TypedValue::Long(7),
store.q_once(r#"[:find (count ?score) .
:where
[?game :foo/score ?score]]"#, None)
.into_scalar_result()
.expect("scalar results").unwrap());
// How many different games resulted in scores?
// '14' appears twice.
assert_eq!(TypedValue::Long(8),
store.q_once(r#"[:find (count ?score) .
:with ?game
:where
[?game :foo/score ?score]]"#, None)
.into_scalar_result()
.expect("scalar results").unwrap());
// Who's the highest-scoring vegetarian?
assert_eq!(vec!["Alice".into(), TypedValue::Long(99)],
store.q_once(r#"[:find [(the ?name) (max ?score)]
:where
[?game :foo/score ?score]
[?person :foo/play ?game]
[?person :foo/is-vegetarian true]
[?person :foo/name ?name]]"#, None)
.into_tuple_result()
.expect("tuple results").unwrap());
// We can't run an ambiguous correspondence.
let res = store.q_once(r#"[:find [(the ?name) (min ?score) (max ?score)]
:where
[?game :foo/score ?score]
[?person :foo/play ?game]
[?person :foo/is-vegetarian true]
[?person :foo/name ?name]]"#, None);
match res {
Result::Err(
Error(
ErrorKind::TranslatorError(
::mentat_query_translator::ErrorKind::ProjectorError(
::mentat_query_projector::ErrorKind::AmbiguousAggregates(mmc, cc)
)
), _)) => {
assert_eq!(mmc, 2);
assert_eq!(cc, 1);
},
r => {
panic!("Unexpected result {:?}.", r);
},
}
// Max scores for vegetarians.
assert_eq!(vec![vec!["Alice".into(), TypedValue::Long(99)],
vec!["Beli".into(), TypedValue::Long(22)]],
store.q_once(r#"[:find ?name (max ?score)
:where
[?game :foo/score ?score]
[?person :foo/play ?game]
[?person :foo/is-vegetarian true]
[?person :foo/name ?name]]"#, None)
.into_rel_result()
.expect("rel results"));
// We can combine these aggregates.
let r = store.q_once(r#"[:find ?x ?name (max ?score) (count ?score) (avg ?score)
:with ?game ; So we don't discard duplicate scores!
:where
[?x :foo/name ?name]
[?x :foo/play ?game]
[?game :foo/score ?score]]"#, None)
.expect("results")
.into();
match r {
QueryResults::Rel(vals) => {
assert_eq!(vals,
vec![
vec![TypedValue::Ref(ids.get("a").cloned().unwrap()),
TypedValue::String("Alice".to_string().into()),
TypedValue::Long(99),
TypedValue::Long(3),
TypedValue::Double((127f64 / 3f64).into())],
vec![TypedValue::Ref(ids.get("b").cloned().unwrap()),
TypedValue::String("Beli".to_string().into()),
TypedValue::Long(22),
TypedValue::Long(2),
TypedValue::Double((33f64 / 2f64).into())],
vec![TypedValue::Ref(ids.get("c").cloned().unwrap()),
TypedValue::String("Carlos".to_string().into()),
TypedValue::Long(42),
TypedValue::Long(1),
TypedValue::Double(42f64.into())],
vec![TypedValue::Ref(ids.get("d").cloned().unwrap()),
TypedValue::String("Diana".to_string().into()),
TypedValue::Long(28),
TypedValue::Long(2),
TypedValue::Double((33f64 / 2f64).into())]]);
},
x => panic!("Got unexpected results {:?}", x),
}
}