sparsemap/tests/soak.c

#include <assert.h>
#include <errno.h>
#include <stdbool.h>
#include <stddef.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <unistd.h>

#include "../include/common.h"
#include "../include/roaring.h"
#include "../include/sparsemap.h"
#include "../include/tdigest.h"

#include "midl.c"

typedef size_t pgno_t;

char *
bytes_as(double bytes, char *s, size_t size)
{
  const char *units[] = { "b", "KiB", "MiB", "GiB", "TiB", "PiB", "EiB", "ZiB", "YiB" };
  size_t i = 0;

  while (bytes >= 1024 && i < sizeof(units) / sizeof(units[0]) - 1) {
    bytes /= 1024;
    i++;
  }

  snprintf(s, size, "%.2f %s", bytes, units[i]);
  return s;
}

/**
 * A "coin toss" function that is critical to the proper operation of the
 * Skiplist.  For example, when `max = 6` this function returns 0 with
 * probability 0.5, 1 with 0.25, 2 with 0.125, etc. until 6 with 0.5^7.
 */
static int
toss(size_t max)
{
  size_t level = 0;
  double probability = 0.5;

  double random_value = (double)xorshift32() / RAND_MAX;
  while (random_value < probability && level < max) {
    level++;
    probability *= 0.5;
  }
  return level;
}

bool recording = true;

static void
record_set_mutation(FILE *out, pgno_t pg)
{
  if (recording) {
    fprintf(out, "set %lu\n", pg);
  }
}

static void
record_clear_mutation(FILE *out, pgno_t pg)
{
  if (recording) {
    fprintf(out, "clear %lu\n", pg);
  }
}

static void
record_take_span_mutation(FILE *out, pgno_t pg, unsigned len)
{
  if (recording) {
    fprintf(out, "take %lu %u\n", pg, len);
  }
}

static void
record_release_span_mutation(FILE *out, pgno_t pg, unsigned len)
{
  if (recording) {
    fprintf(out, "release %lu %u\n", pg, len);
  }
}

static void
__scan_record_offsets(sm_idx_t v[], size_t n, void *aux)
{
  FILE *out = (FILE *)aux;
  for (size_t i = 0; i < n; i++) {
    fprintf(out, "%u ", v[i]);
  }
}

static void
record_merge_mutation(FILE *out, void *handle)
{
  if (recording) {
    sparsemap_t *map = (sparsemap_t *)handle;
    fprintf(out, "merge %zu ", sparsemap_get_ending_offset(map));
    sparsemap_scan(map, __scan_record_offsets, 0, (void *)out);
    fprintf(out, "\n");
  }
}

/* sparsemap ------------------------------------------------------------- */

static sparsemap_idx_t
_sparsemap_set(sparsemap_t **map, sparsemap_idx_t idx, bool value)
{
  do {
    sparsemap_idx_t l = sparsemap_set(*map, idx, value);
    if (l != idx) {
      if (errno == ENOSPC) {
        *map = sparsemap_set_data_size(*map, NULL, sparsemap_get_capacity(*map) + 64);
        assert(*map != NULL);
        errno = 0;
      } else {
        assert(false);
      }
    } else {
      return l;
    }
  } while (true);
}

static void *
__sm_alloc(size_t capacity)
{
  sparsemap_t *map = sparsemap(capacity);
  assert(map != NULL);
  return map;
}

static void
__sm_free(void *handle)
{
  sparsemap_t *map = (sparsemap_t *)handle;
  free(map);
}

static pgno_t
__sm_set(void **handle, pgno_t pg)
{
  sparsemap_t **map = (sparsemap_t **)handle;
  return (pgno_t)_sparsemap_set(map, pg, true);
}

static bool
__sm_is_set(void *handle, pgno_t pg)
{
  sparsemap_t *map = (sparsemap_t *)handle;
  return sparsemap_is_set(map, pg);
}

static pgno_t
__sm_clear(void **handle, pgno_t pg)
{
  sparsemap_t **map = (sparsemap_t **)handle;
  return (pgno_t)_sparsemap_set(map, pg, false);
}

static pgno_t
__sm_find_span(void *handle, unsigned len)
{
  sparsemap_t *map = (sparsemap_t *)handle;
  pgno_t pgno = (pgno_t)sparsemap_span(map, 0, len, true);
  assert(SPARSEMAP_NOT_FOUND(pgno) == false);
  return pgno;
}

static bool
__sm_take_span(void **handle, pgno_t pg, unsigned len)
{
  sparsemap_t **map = (sparsemap_t **)handle;
  for (pgno_t i = pg; i < pg + len; i++) {
    assert(_sparsemap_set(map, i, false) == i);
  }
  return true;
}

static bool
__sm_release_span(void **handle, pgno_t pg, unsigned len)
{
  sparsemap_t **map = (sparsemap_t **)handle;
  for (pgno_t i = pg; i < pg + len; i++) {
    assert(_sparsemap_set(map, i, true) == i);
  }
  return true;
}

static bool
__sm_is_span(void *handle, pgno_t pg, unsigned len)
{
  sparsemap_t *map = (sparsemap_t *)handle;
  for (pgno_t i = pg; i < pg + len; i++) {
    if (sparsemap_is_set(map, i) != true) {
      return false;
    }
  }
  return true;
}

static bool
__sm_is_empty(void *handle, pgno_t pg, unsigned len)
{
  sparsemap_t *map = (sparsemap_t *)handle;
  for (pgno_t i = 0; i < len; i++) {
    if (sparsemap_is_set(map, pg + i) != false) {
      return false;
    }
  }
  return true;
}

static bool
__sm_is_first(void *handle, pgno_t pg, unsigned len)
{
  sparsemap_t *map = (sparsemap_t *)handle;
  for (sparsemap_idx_t i = 0; i < pg + len; i++) {
    sparsemap_idx_t j = 0;
    while (sparsemap_is_set(map, i + j) == true && j < len) {
      j++;
    }
    if (j == len) {
      return i == pg;
    }
  }
  return false;
}

static bool
__sm_merge(void **handle, void *other_handle)
{
  sparsemap_t **map = (sparsemap_t **)handle;
  sparsemap_t *other = (sparsemap_t *)other_handle;
  do {
    int retval = sparsemap_merge(*map, other);
    if (retval != 0) {
      if (errno == ENOSPC) {
        size_t new_size = retval + (64 - (retval % 64)) + 64;
        *map = sparsemap_set_data_size(*map, NULL, sparsemap_get_capacity(*map) + new_size);
        assert(*map != NULL);
        errno = 0;
      } else {
        assert(false);
      }
    } else {
      break;
    }
  } while (true);
  return true;
}

static size_t
__sm_size(void *handle)
{
  sparsemap_t *map = (sparsemap_t *)handle;
  return sparsemap_get_size(map);
}

static size_t
__sm_count(void *handle)
{
  sparsemap_t *map = (sparsemap_t *)handle;
  return sparsemap_rank(map, 0, SPARSEMAP_IDX_MAX, true);
}

/* midl ------------------------------------------------------------------ */

static void *
__midl_alloc(size_t capacity)
{
  MDB_IDL list = mdb_midl_alloc(capacity);
  assert(list != NULL);
  return (void *)list;
}

static void
__midl_free(void *handle)
{
  MDB_IDL list = (MDB_IDL)handle;
  mdb_midl_free(list);
}

static pgno_t
__midl_set(void **handle, pgno_t pg)
{
  MDB_IDL *_list = (MDB_IDL *)handle, list = *_list;
  if (list[0] + 1 == list[-1]) {
    mdb_midl_need(_list, list[-1] + 1);
    list = *_list;
  }
  mdb_midl_insert(list, pg);
  return pg;
}

static bool
__midl_is_set(void *handle, pgno_t pg)
{
  MDB_IDL list = (MDB_IDL)handle;
  pgno_t i = mdb_midl_search(list, pg);
  return i <= list[0] && list[i] == pg;
}

static pgno_t
__midl_clear(void **handle, pgno_t pg)
{
  MDB_IDL list = *(MDB_IDL *)handle;
  unsigned len = list[0];
  list[0] = len -= 1;
  for (unsigned j = pg - 1; j < len;)
    list[++j] = list[++pg];
  for (unsigned j = len + 1; j <= list[-1]; j++)
    list[j] = 0;
  return pg;
}

static pgno_t
__midl_find_span(void *handle, unsigned len)
{
  MDB_IDL list = (MDB_IDL)handle;

  /* Seek a big enough contiguous page range. Prefer
   * pages at the tail, just truncating the list.
   */
  int retry = 1;
  unsigned i = 0;
  pgno_t pgno = 0, *mop = list;
  unsigned n2 = len, mop_len = mop[0];
  do {
    if (mop_len > n2) {
      i = mop_len;
      do {
        pgno = mop[i];
        if (mop[i - n2] == pgno + n2)
          goto search_done;
      } while (--i > n2);
      if (--retry < 0)
        break;
    }
  } while (1);
search_done:;
  return pgno;
}

static bool
__midl_take_span(void **handle, pgno_t pg, unsigned len)
{
  MDB_IDL list = *(MDB_IDL *)handle;
  int i = list[list[0]] == pg ? list[0] : mdb_midl_search(list, pg);
  unsigned j, num = len;
  pgno_t *mop = list;
  unsigned mop_len = mop[0];

  mop[0] = mop_len -= num;
  /* Move any stragglers down */
  for (j = i - num; j < mop_len;)
    mop[++j] = mop[++i];
  /* Set all unused values in the array to 0
  for (j = mop_len + 1; j <= mop[-1]; j++)
    mop[j] = 0; */
  return true;
}

static bool
__midl_release_span(void **handle, pgno_t pg, unsigned len)
{
  MDB_IDL *_list = (MDB_IDL *)handle, list = *_list;
  if (list[0] + len >= list[-1]) {
    mdb_midl_need(_list, list[-1] + len);
    list = *_list;
  }
  for (size_t i = pg; i < pg + len; i++) {
    mdb_midl_insert(list, i);
  }
  mdb_midl_sort(list);
  return true;
}

static bool
__midl_is_span(void *handle, pgno_t pg, unsigned len)
{
  MDB_IDL list = (MDB_IDL)handle;
  pgno_t idx = mdb_midl_search(list, pg);
  bool found = idx <= list[0] && list[idx] == pg;
  if (!found)
    return false;
  if (len == 1)
    return true;
  if (list[len] + 1 != list[len - 1])
    return false;
  return true;
}

static bool
__midl_is_empty(void *handle, pgno_t pg, unsigned len)
{
  MDB_IDL list = (MDB_IDL)handle;
  for (pgno_t i = pg; i < pg + len; i++) {
    pgno_t idx = mdb_midl_search(list, pg);
    bool found = idx <= list[0] && list[idx] == pg;
    if (found)
      return false;
  }
  return true;
}

static bool
__midl_merge(void **handle, void *other_handle)
{
  MDB_IDL *_list = (MDB_IDL *)handle, list = *_list;
  MDB_IDL other = (MDB_IDL)other_handle;
  if (list[0] + other[0] >= list[-1]) {
    mdb_midl_need(_list, list[-1] + other[0]);
    list = *_list;
  }
  mdb_midl_append_list(_list, other);
  list = *_list;
  mdb_midl_sort(list);
  return true;
}

static size_t
__midl_size(void *handle)
{
  MDB_IDL list = (MDB_IDL)handle;
  return list[0] * sizeof(pgno_t);
}

static size_t
__midl_count(void *handle)
{
  MDB_IDL list = (MDB_IDL)handle;
  return list[0];
}

static bool
__midl_validate(void *handle)
{
  MDB_IDL list = (MDB_IDL)handle;
  pgno_t id = 1;
  while (id < list[0]) {
    if (list[id] >= list[id + 1])
      return false;
    id++;
  }
  return true;
}

/* roaring --------------------------------------------------------------- */

static void *
__roar_alloc(size_t capacity)
{
  roaring_bitmap_t *map = roaring_bitmap_create();
  assert(map != NULL);
  return map;
}

static void
__roar_free(void *handle)
{
  roaring_bitmap_t *rbm = (roaring_bitmap_t *)handle;
  roaring_free(rbm);
}

static pgno_t
__roar_set(void **handle, pgno_t pg)
{
  roaring_bitmap_t **_rbm = (roaring_bitmap_t **)handle, *rbm = *_rbm;
  assert(roaring_bitmap_add_checked(rbm, pg) == true);
  return pg;
}

static bool
__roar_is_set(void *handle, pgno_t pg)
{
  roaring_bitmap_t *rbm = (roaring_bitmap_t *)handle;
  return roaring_bitmap_contains(rbm, pg);
}

static pgno_t
__roar_clear(void **handle, pgno_t pg)
{
  roaring_bitmap_t **_rbm = (roaring_bitmap_t **)handle, *rbm = *_rbm;
  roaring_bitmap_remove(rbm, pg);
  return pg;
}

static pgno_t
__roar_find_span(void *handle, unsigned len)
{
  roaring_bitmap_t *rbm = (roaring_bitmap_t *)handle;
  uint64_t max = roaring_bitmap_maximum(rbm);
  uint64_t offset = roaring_bitmap_minimum(rbm);
  do {
    if (len == 1 || roaring_bitmap_range_cardinality(rbm, offset, offset + len) == len) {
      break;
    }
    offset++;
  } while (offset <= max);
  return offset;
}

static bool
__roar_take_span(void **handle, pgno_t pg, unsigned len)
{
  roaring_bitmap_t **_rbm = (roaring_bitmap_t **)handle, *rbm = *_rbm;
  roaring_bitmap_remove_range(rbm, pg, pg + len);
  return true;
}

static bool
__roar_release_span(void **handle, pgno_t pg, unsigned len)
{
  roaring_bitmap_t **_rbm = (roaring_bitmap_t **)handle, *rbm = *_rbm;
  for (size_t i = pg; i < pg + len; i++) {
    assert(roaring_bitmap_add_checked(rbm, i) == true);
  }
  return true;
}

static bool
__roar_is_span(void *handle, pgno_t pg, unsigned len)
{
  roaring_bitmap_t *rbm = (roaring_bitmap_t *)handle;
  for (pgno_t i = pg; i < pg + len; i++) {
    if (roaring_bitmap_contains(rbm, i) != true) {
      return false;
    }
  }
  return true;
}

static bool
__roar_is_empty(void *handle, pgno_t pg, unsigned len)
{
  roaring_bitmap_t *rbm = (roaring_bitmap_t *)handle;
  for (pgno_t i = 0; i < len; i++) {
    if (roaring_bitmap_contains(rbm, pg + i) != false) {
      return false;
    }
  }
  return true;
}

static bool
__roar_merge(void **handle, void *other_handle)
{
  roaring_bitmap_t **_rbm = (roaring_bitmap_t **)handle, *rbm = *_rbm;
  roaring_bitmap_t *other = (roaring_bitmap_t *)other_handle;
  roaring_bitmap_or_inplace(rbm, other);
  return true;
}

static size_t
__roar_size(void *handle)
{
  // TODO
  return 0;
}

static size_t
__roar_count(void *handle)
{
  // TODO
  return 0;
}

static bool
__roar_validate(void *handle)
{
  roaring_bitmap_t *rbm = (roaring_bitmap_t *)handle;
  roaring_bitmap_run_optimize(rbm);
  return true;
}

/* ----------------------------------------------------------------------- */

typedef enum { SM, ML, RB } container_impl_t;

typedef struct container {
  const char *name;
  void *(*alloc)(size_t capacity);
  void (*free)(void *handle);
  pgno_t (*set)(void **handle, pgno_t pg);
  bool (*is_set)(void *handle, pgno_t pg);
  pgno_t (*clear)(void **handle, pgno_t pg);
  pgno_t (*find_span)(void *handle, unsigned len);
  bool (*take_span)(void **handle, pgno_t pg, unsigned len);
  bool (*release_span)(void **handle, pgno_t pg, unsigned len);
  bool (*is_span)(void *handle, pgno_t pg, unsigned len);
  bool (*is_empty)(void *handle, pgno_t pg, unsigned len);
  bool (*is_first)(void *handle, pgno_t pg, unsigned len);
  bool (*merge)(void **handle, void *other_handle);
  size_t (*size)(void *handle);
  size_t (*count)(void *handle);
  bool (*validate)(void *handle);
} container_t;

// clang-format off
container_t containers[] = {
  { "sparsemap",
    .alloc = __sm_alloc,
    .free = __sm_free,
    .set = __sm_set,
    .is_set = __sm_is_set,
    .clear = __sm_clear,
    .find_span = __sm_find_span,
    .take_span = __sm_take_span,
    .release_span = __sm_release_span,
    .is_span = __sm_is_span,
    .is_empty = __sm_is_empty,
    .is_first = __sm_is_first,
    .merge = __sm_merge,
    .size = __sm_size,
    .count = __sm_count,
    .validate = NULL
  },
  { "midl",
    .alloc = __midl_alloc,
    .free = __midl_free,
    .set = __midl_set,
    .is_set = __midl_is_set,
    .clear = __midl_clear,
    .find_span = __midl_find_span,
    .take_span = __midl_take_span,
    .release_span = __midl_release_span,
    .is_span = __midl_is_span,
    .is_empty = __midl_is_empty,
    .is_first = NULL,
    .merge = __midl_merge,
    .size = __midl_size,
    .count = __midl_count,
    .validate = __midl_validate
  },
  { "roaring",
    .alloc = __roar_alloc,
    .free = __roar_free,
    .set = __roar_set,
    .is_set = __roar_is_set,
    .clear = __roar_clear,
    .find_span = __roar_find_span,
    .take_span = __roar_take_span,
    .release_span = __roar_release_span,
    .is_span = __roar_is_span,
    .is_empty = __roar_is_empty,
    .is_first = NULL,
    .merge = __roar_merge,
    .size = __roar_size,
    .count = __roar_count,
    .validate = __roar_validate,
  },
};
// clang-format on

void *handles[(sizeof((containers)) / sizeof((containers)[0]))];
void *new_handles[(sizeof((containers)) / sizeof((containers)[0]))];
FILE *fp;

#define alloc(type, size) containers[type].alloc(size);
#define cast(type, fn, ...) \
  if (containers[type].fn)  \
  containers[type].fn(handles[type], ##__VA_ARGS__)
#define invoke(type, fn, ...) containers[type].fn(handles[type], __VA_ARGS__)
#define mutate(type, fn, ...) (type == 0) ? record_##fn##_mutation(fp, __VA_ARGS__) : (void)0, containers[type].fn(&handles[type], __VA_ARGS__)
#define foreach(set) for (unsigned type = 0; type < (sizeof((set)) / sizeof((set)[0])); type++)
#define compare(set)                                                           \
  for (unsigned type = 1; type < (sizeof((set)) / sizeof((set)[0])); type++) { \
    verify_eq(0, handles[0], type, handles[type]);                             \
  }

bool
verify_sm_eq_rb(sparsemap_t *map, roaring_bitmap_t *rbm)
{
  uint64_t max = roaring_bitmap_maximum(rbm);
  roaring_uint32_iterator_t iter;
  roaring_iterator_init(rbm, &iter);
  for (uint64_t i = 0; i <= max; i++) {
    if (i == iter.current_value) {
      assert(sparsemap_is_set(map, i) == true);
      roaring_uint32_iterator_advance(&iter);
    } else {
      assert(sparsemap_is_set(map, i) == false);
    }
  }
  return true;
}

bool
verify_sm_eq_ml(sparsemap_t *map, MDB_IDL list)
{
  for (MDB_ID i = 1; i <= list[0]; i++) {
    pgno_t pg = list[i];
    unsigned skipped = i == 1 ? 0 : list[i - 1] - list[i] - 1;
    if (skipped) {
      for (MDB_ID j = list[i - 1]; j > list[i]; j--) {
        if (sparsemap_is_set(map, pg - j) != false) {
          __diag("%zu\n", pg - j);
          return false;
        }
      }
    }
    if (sparsemap_is_set(map, pg) != true) {
      __diag("%zu\n", pg);
      return false;
    }
  }
  return true;
}

bool
verify_eq(unsigned a, void *ad, unsigned b, void *bd)
{
  bool ret = true;

  // 'a' should always be a Sparsemap
  switch (b) {
  case ML:
    assert((ret = verify_sm_eq_ml((sparsemap_t *)ad, (MDB_IDL)bd)) == true);
    break;
  case RB:
    assert((ret = verify_sm_eq_rb((sparsemap_t *)ad, (roaring_bitmap_t *)bd)) == true);
    break;
  default:
    break;
  }
  return ret;
}

td_histogram_t *l_span_loc;
td_histogram_t *b_span_loc;
td_histogram_t *l_span_take;
td_histogram_t *b_span_take;
td_histogram_t *l_span_merge;
td_histogram_t *b_span_merge;

void
stats_header(void)
{
  printf(
    "timestamp,iterations,idl_cap,idl_used,idl_bytes,sm_cap,sm_used,idl_loc_p50,idl_loc_p75,idl_loc_p90,idl_loc_p99,idl_loc_p999,sm_loc_p50,sm_loc_p75,sm_loc_p90,sm_loc_p99,sm_loc_p999,idl_take_p50,idl_take_p75,idl_take_p90,idl_take_p99,idl_take_p999,sm_take_p50,sm_take_p75,sm_take_p90,sm_take_p99,sm_take_p999,idl_merge_p50,idl_merge_p75,idl_merge_p90,idl_merge_p99,idl_merge_p999,sm_merge_p50,sm_merge_p75,sm_merge_p90,sm_merge_p99,sm_merge_p999\n");
}

void
stats(size_t iterations, sparsemap_t *map, MDB_IDL list)
{
  if (iterations < 10)
    return;

  td_compress(l_span_loc);
  td_compress(b_span_loc);
  td_compress(l_span_take);
  td_compress(b_span_take);
  td_compress(l_span_merge);
  td_compress(b_span_merge);

  printf(
    "%f,%zu,%zu,%zu,%zu,%zu,%zu,%.10f,%.10f,%.10f,%.10f,%.10f,%.10f,%.10f,%.10f,%.10f,%.10f,%.10f,%.10f,%.10f,%.10f,%.10f,%.10f,%.10f,%.10f,%.10f,%.10f,%.10f,%.10f,%.10f,%.10f,%.10f,%.10f,%.10f,%.10f,%.10f,%.10f\n",
    nsts(), iterations, list[-1], list[0], MDB_IDL_SIZEOF(list), sparsemap_get_capacity(map), sparsemap_get_size(map), td_quantile(l_span_loc, .5),
    td_quantile(l_span_loc, .75), td_quantile(l_span_loc, .90), td_quantile(l_span_loc, .99), td_quantile(l_span_loc, .999), td_quantile(b_span_loc, .5),
    td_quantile(b_span_loc, .75), td_quantile(b_span_loc, .90), td_quantile(b_span_loc, .99), td_quantile(b_span_loc, .999), td_quantile(l_span_take, .5),
    td_quantile(l_span_take, .75), td_quantile(l_span_take, .90), td_quantile(l_span_take, .99), td_quantile(l_span_take, .999), td_quantile(b_span_take, .5),
    td_quantile(b_span_take, .75), td_quantile(b_span_take, .90), td_quantile(b_span_take, .99), td_quantile(b_span_take, .999), td_quantile(l_span_merge, .5),
    td_quantile(l_span_merge, .75), td_quantile(l_span_merge, .90), td_quantile(l_span_merge, .99), td_quantile(l_span_merge, .999),
    td_quantile(b_span_merge, .5), td_quantile(b_span_merge, .75), td_quantile(b_span_merge, .90), td_quantile(b_span_merge, .99),
    td_quantile(b_span_merge, .999));
}

#define INITIAL_AMOUNT 1024 * 2

#define SHORT_OPT "r:fa:bh"
#define LONG_OPT "record:,force,amount:,buffer,help"

void
print_usage(const char *program_name)
{
  printf("Usage: %s [OPTIONS]\n", program_name);
  printf("  -r, --record <file>    Path to the file for recording (optional)\n");
  printf("  -f, --force            Force overwrite of existing file (optional)\n");
  printf("  -b, --buffer           Disable buffering writes to stdout/err (optional)\n");
  printf("  -a, --amount <number>  Specify the number of entries to record (must be positive, optional)\n");
  printf("  -h, --help             Print this help message\n");
}

int
main(int argc, char *argv[])
{
  int opt;
  const char *record_file = NULL;
  int force_flag = 0;
  size_t left, amt = INITIAL_AMOUNT;
  bool buffer = true;

  fp = stdout;

  while ((opt = getopt(argc, argv, SHORT_OPT LONG_OPT)) != -1) {
    switch (opt) {
    case 'r':
      record_file = optarg;
      break;
    case 'f':
      force_flag = 1;
      break;
    case 'b':
      buffer = false;
      break;
    case 'a':
      amt = atoi(optarg);
      if (amt <= 0) {
        fprintf(stderr, "Error: Invalid amount. Amount must be a positive number.\n");
        return 1;
      }
      break;
    case 'h':
      print_usage(argv[0]);
      return 0;
    case '?':
      fprintf(stderr, "Unknown option: %c\n", optopt);
      return 1;
    default:
      break;
    }
  }

  // Check if record file is specified
  if (record_file == NULL) {
    recording = true;//TODO
  } else {
    // Check for existing file without force flag
    if (access(record_file, F_OK) == 0 && !force_flag) {
      fprintf(stderr, "Warning: File '%s' already exists. Use -f or --force to overwrite.\n", record_file);
      return 1;
    }

    // Open the file for writing (truncate if force flag is set)
    fp = fopen(record_file, force_flag ? "w" : "a");
    if (fp == NULL) {
      perror("Error opening file");
      return 1;
    }
  }

  // disable buffering
  if (!buffer) {
    setvbuf(stdout, NULL, _IONBF, 0);
    setvbuf(fp, NULL, _IONBF, 0);
  }
  unsigned types[] = { SM, ML, RB };
  unsigned num_types = (sizeof((types)) / sizeof((types)[0]));

  /* Setup: add an amt of bits to each container. */
  foreach(types)
  {
    handles[type] = alloc(type, amt);
    for (size_t i = 0; i < amt; i++) {
      assert(invoke(type, is_set, i) == false);
      assert(mutate(type, set, i) == i);
      assert(invoke(type, is_set, i) == true);
    }
    cast(type, validate);
  }
  compare(types);
  left = amt;

  while (true) {
    // the an amount [1, 16] of pages to find preferring smaller sizes
    unsigned len = toss(15) + 1;
    pgno_t loc[num_types];
    foreach(types)
    {
      loc[type] = invoke(type, find_span, len);
    }
    for (unsigned n = 0; n < num_types; n++) {
      foreach(types)
      {
        assert(invoke(type, is_span, loc[n], len));
      }
    }
    foreach(types)
    {
      cast(type, validate);
    }

    unsigned which_loc = (unsigned)xorshift32() % num_types;
    foreach(types)
    {
      assert(mutate(type, take_span, loc[which_loc], len));
      cast(type, validate);
    }
    compare(types);
    left -= len;

    // Once we've used 1/10th of the free list, let's replenish it a bit.
    if (amt - left > amt / 10) {
      do {
        pgno_t pgno;
        size_t len, retries = amt;
        // Find a hole in the map to replenish.
        do {
          len = toss(15) + 1;
          pgno = sparsemap_span(handles[SM], 0, len, false);
        } while (SPARSEMAP_NOT_FOUND(pgno) && --retries);
        if (retries == 0) {
          goto larger_please;
        }
        if (SPARSEMAP_FOUND(pgno)) {
          foreach(types)
          {
            assert(invoke(type, is_empty, pgno, len));
          }
          compare(types);
          foreach(types)
          {
            assert(invoke(type, is_span, pgno, len) == false);
            assert(mutate(type, release_span, pgno, len));
            assert(invoke(type, is_span, pgno, len) == true);
            cast(type, validate);
          }
          compare(types);
          left += len;
        }
      } while (amt - left > amt / 100);
    }

    if (toss(1000) > 800) {
      size_t new_offset, new_amt;
      pgno_t max;
    larger_please:
      new_amt = 1024 + (xorshift32() % 2048) + toss(1024);
      new_offset = xorshift32() % 4096 + 1024;
      max = sparsemap_get_ending_offset(handles[SM]);

      // Build a new container to merge with the existing one.
      foreach(types)
      {
        new_handles[type] = alloc(type, new_amt);
        for (size_t i = 0; i < new_amt; i++) {
          // We don't want to record and we're using new_handles not
          // handles, so call fn directly.
          assert(containers[type].is_set(new_handles[type], i + new_offset) == false);
          containers[type].set(&new_handles[type], i + new_offset);
          assert(containers[type].is_set(new_handles[type], i + new_offset) == true);
        }
      }
      foreach(types)
      {
        assert(mutate(type, merge, new_handles[type]));
        cast(type, validate);
      }
      compare(types);
      left += new_amt;
      amt += new_amt;
      foreach(types)
      {
        containers[type].free(new_handles[type]);
      }
    }
  }

  return 0;
}

#if 0
/*
 * A "soak test" that tries to replicate behavior in LMDB for page allocation.
 */
int
main(void)
{
  size_t replenish = 0, iterations = 0;

  // disable buffering
#ifdef DEBUG
  setvbuf(stdout, NULL, _IONBF, 0);
  setvbuf(stderr, NULL, _IONBF, 0);
#endif

  l_span_loc = td_new(100);
  b_span_loc = td_new(100);
  l_span_take = td_new(100);
  b_span_take = td_new(100);
  l_span_merge = td_new(100);
  b_span_merge = td_new(100);

  stats_header();

  sparsemap_idx_t amt = INITIAL_AMOUNT;
  MDB_IDL list = mdb_midl_alloc(amt);
  sparsemap_t *map = sparsemap(INITIAL_AMOUNT);
  roaring_bitmap_t *rbm = roaring_bitmap_create();

  // start with 2GiB of 4KiB free pages to track:
  //  - MDB_IDL requires one int for each free page
  //  - Sparsemap will compress the set bits using less memory
  mdb_midl_need(&list, amt);
  for (sparsemap_idx_t pg = 0; pg < amt; pg++) {
    // We list every free (unallocated) page in the IDL, while...
    mdb_midl_xappend(list, pg);
    // ... true (unset in the bitmap) indicates free in the bitmap, ...
    assert(_sparsemap_set(&map, pg, true) == pg);
    assert(roaring_bitmap_add_checked(rbm, pg));
  }
  mdb_midl_sort(list);
  roaring_bitmap_run_optimize(rbm);
  assert(verify_sm_eq_ml(map, list));
  assert(verify_sm_eq_rm(map, rbm));

  double b, e;
  while (1) {
    unsigned mi;
    pgno_t ml, sl, rl;

    // get an amount [1, 16] of pages to find preferring smaller sizes
    unsigned n = toss(15) + 1;

    // find a set of pages using the MDB_IDL
    {
      b = nsts();
      /* Seek a big enough contiguous page range. Prefer
       * pages at the tail, just truncating the list.
       */
      int retry = 1;
      unsigned i = 0;
      pgno_t pgno = 0, *mop = list;
      unsigned n2 = len, mop_len = mop[0];
      if (mop_len > n2) {
        i = mop_len;
        do {
          pgno = mop[i];
          if (mop[i - n2] == pgno + n2)
            goto search_done;
        } while (--i > n2);
        if (--retry < 0)
          break;
      }
    search_done:;
      ml = pgno;
      mi = i;
      e = nsts();
      td_add(l_span_loc, e - b, 1);
    }
    assert(verify_span_midl(list, ml, n));
    assert(verify_span_sparsemap(map, ml, n));
    assert(verify_span_roaring(rbm, ml, n));

    // find a set of pages using the Sparsemap
    {
      b = nsts();
      pgno_t pgno = sparsemap_span(map, 0, n, true);
      assert(SPARSEMAP_NOT_FOUND(pgno) == false);
      sl = pgno;
      e = nsts();
      td_add(b_span_loc, e - b, 1);
      assert(verify_sm_is_first_available_span(map, pgno, n, true));
    }
    assert(verify_span_midl(list, sl, n));
    assert(verify_span_sparsemap(map, sl, n));
    assert(verify_span_roaring(rbm, sl, n));

    // find a set of pages using the Roaring Bitmap
    {
      b = nsts();
      uint64_t max = roaring_bitmap_maximum(rbm);
      uint64_t offset = roaring_bitmap_minimum(rbm);
      do {
        if (n == 1 || roaring_bitmap_range_cardinality(rbm, offset, offset + n) == n) {
          break;
        }
        offset++;
      } while (offset <= max);
      rl = offset;
      e = nsts();
    }
    /*
    if (rl != sl) {
      assert(verify_span_midl(list, rl, n));
      assert(verify_span_sparsemap(map, rl, n));
      assert(verify_span_roaring(rbm, rl, n));
    }
    */
    assert(rl == sl);

    bool prefer_mdb_idl_loc = (bool)xorshift32() % 2;

    // acquire the set of pages within the list
    if (prefer_mdb_idl_loc) {
      b = nsts();
      unsigned j, num = n;
      int i = mi;
      pgno_t *mop = list;
      unsigned mop_len = mop[0];

      mop[0] = mop_len -= num;
      /* Move any stragglers down */
      for (j = i - num; j < mop_len;)
        mop[++j] = mop[++i];
      e = nsts();
      for (j = mop_len + 1; j <= mop[-1]; j++)
        mop[j] = 0;
      td_add(l_span_take, e - b, 1);
    } else {
      b = nsts();
      unsigned j, num = n;
      int i = mdb_midl_search(list, sl) + num;
      pgno_t *mop = list;
      unsigned mop_len = mop[0];

      mop[0] = mop_len -= num;
      /* Move any stragglers down */
      for (j = i - num; j < mop_len;)
        mop[++j] = mop[++i];
      e = nsts();
      for (j = mop_len + 1; j <= mop[-1]; j++)
        mop[j] = 0;
      td_add(l_span_take, e - b, 1);
    }

    // acquire the set of pages within the sparsemap
    if (prefer_mdb_idl_loc) {
      b = nsts();
      for (pgno_t i = ml; i < ml + n; i++) {
        assert(_sparsemap_set(&map, i, false) == i);
      }
      e = nsts();
      td_add(b_span_take, e - b, 1);
    } else {
      b = nsts();
      for (pgno_t i = sl; i <= sl + n; i++) {
        assert(_sparsemap_set(&map, i, false) == i);
      }
      e = nsts();
      td_add(b_span_take, e - b, 1);
    }

    // acquire the set of pages within the roaring bitmap
    if (prefer_mdb_idl_loc) {
      b = nsts();
      roaring_bitmap_remove_range(rbm, ml, ml + n);
      e = nsts();
    } else {
      b = nsts();
      roaring_bitmap_remove_range(rbm, sl, sl + n);
      e = nsts();
    }
    roaring_bitmap_run_optimize(rbm);

    assert(verify_sm_eq_ml(map, list));
    assert(verify_sm_eq_rm(map, rbm));

    // Once we've used a tenth of the free list, let's replenish it a bit.
    if (list[0] < amt / 10) {
      do {
        pgno_t pgno;
        size_t len, retries = amt;
        do {
          len = toss(15) + 1;
          pgno = sparsemap_span(map, 0, len, false);
          assert(verify_sm_is_first_available_span(map, pgno, n, false));
          //__diag("%zu\t%zu,%zu\n", iterations, replenish, retries);
        } while (SPARSEMAP_NOT_FOUND(pgno) && --retries);
        if (retries == 0) {
          goto larger_please;
        }
        if (SPARSEMAP_FOUND(pgno)) {
          assert(verify_empty_midl(list, pgno, len));
          assert(verify_empty_sparsemap(map, pgno, len));
          assert(verify_empty_roaring(rbm, pgno, len));
          assert(verify_sm_eq_ml(map, list));
          assert(verify_sm_eq_rm(map, rbm));
          if (list[-1] - list[0] < len) {
            mdb_midl_need(&list, list[-1] + len);
          }
          for (size_t i = pgno; i < pgno + len; i++) {
            assert(verify_midl_contains(list, i) == false);
            assert(sparsemap_is_set(map, i) == false);
            mdb_midl_insert(list, i);
            assert(verify_midl_contains(list, i) == true);
            assert(_sparsemap_set(&map, i, true) == i);
            assert(sparsemap_is_set(map, i) == true);
            assert(roaring_bitmap_add_checked(rbm, i) == true);
          }
          mdb_midl_sort(list);
          assert(verify_midl_nodups(list));
          assert(verify_span_midl(list, pgno, len));
          assert(verify_span_sparsemap(map, pgno, len));
          assert(verify_span_roaring(rbm, pgno, len));
        }
        assert(verify_sm_eq_ml(map, list));
        assert(verify_sm_eq_rm(map, rbm));
        replenish++;
      } while (list[0] < amt - 32);
    }
    replenish = 0;

    // every so often, either ...
    if (iterations % 1000 == 0) {
    larger_please:;
      size_t COUNT = xorshift32() % 3586 + 513;
      // ... add some amount of 4KiB pages, or
      size_t len = COUNT;
      // The largest page is at list[1] because this is a reverse sorted list.
      pgno_t pg = list[0] ? list[1] + 1 : 0;
      if (true) { // disable shrinking for now... (toss(6) + 1 < 7)
        MDB_IDL new_list = mdb_midl_alloc(len);
        sparsemap_t *new_map = sparsemap(INITIAL_AMOUNT);
        roaring_bitmap_t *new_rbm = roaring_bitmap_create();
        for (size_t i = 0; i < len; i++) {
          pgno_t gp = (pg + len) - i;
          new_list[i + 1] = gp;
          new_list[0]++;
          assert(verify_midl_contains(new_list, gp) == true);
          assert(_sparsemap_set(&new_map, gp, true) == gp);
          assert(sparsemap_is_set(new_map, gp));
          assert(roaring_bitmap_add_checked(new_rbm, gp));
          assert(roaring_bitmap_contains(new_rbm, gp));
        }
        assert(verify_sm_eq_ml(new_map, new_list));
        assert(verify_sm_eq_rm(new_map, new_rbm));
        {
          b = nsts();
          mdb_midl_append_list(&list, new_list);
          mdb_midl_sort(list);
          e = nsts();
          td_add(l_span_merge, e - b, 1);
        }
        for (size_t i = 0; i < len; i++) {
          pgno_t gp = (pg + len) - i;
          assert(verify_midl_contains(list, gp) == true);
        }
        {
          b = nsts();
          _sparsemap_merge(&map, new_map);
          e = nsts();
          td_add(b_span_merge, e - b, 1);
        }
        for (size_t i = 0; i < len; i++) {
          pgno_t gp = (pg + len) - i;
          assert(sparsemap_is_set(map, gp));
        }
        free(new_map);
        {
          b = nsts();
          roaring_bitmap_or_inplace(rbm, new_rbm);
          e = nsts();
        }
        for (size_t i = 0; i < len; i++) {
          pgno_t gp = (pg + len) - i;
          assert(roaring_bitmap_contains(rbm, gp));
        }
        roaring_free(new_rbm);
      } else {
        if (list[-1] > INITIAL_AMOUNT) {
          // ... a fraction of the time, remove COUNT / 2 of 4KiB pages.
          {
            pgno_t pg;
            for (size_t i = 0; i < COUNT; i++) {
              pg = list[list[0] - i];
              assert(sparsemap_is_set(map, pg) == true);
              assert(_sparsemap_set(&map, pg, false) == pg);
            }
          }
          {
            roaring_bitmap_remove_range_closed(rbm, list[list[0] - COUNT], list[list[0]]);
          }
          {
            mdb_midl_shrink_to(&list, list[0] - COUNT);
          }
          assert(list[list[0]] != pg);
          assert(verify_midl_nodups(list));
          verify_sm_eq_ml(map, list);
          verify_sm_eq_rm(map, rbm);
        }
      }
    }
    stats(iterations, map, list);
    // printf("\033[K%zu\r", iterations);
    iterations++;
  }

  return 0;
}
#endif