| 1 |
/*------------------------------------------------------------------------- |
| 2 |
* |
| 3 |
* mvdistinct.c |
| 4 |
* POSTGRES multivariate ndistinct coefficients |
| 5 |
* |
| 6 |
* Estimating number of groups in a combination of columns (e.g. for GROUP BY) |
| 7 |
* is tricky, and the estimation error is often significant. |
| 8 |
|
| 9 |
* The multivariate ndistinct coefficients address this by storing ndistinct |
| 10 |
* estimates for combinations of the user-specified columns. So for example |
| 11 |
* given a statistics object on three columns (a,b,c), this module estimates |
| 12 |
* and stores n-distinct for (a,b), (a,c), (b,c) and (a,b,c). The per-column |
| 13 |
* estimates are already available in pg_statistic. |
| 14 |
* |
| 15 |
* |
| 16 |
* Portions Copyright (c) 1996-2025, PostgreSQL Global Development Group |
| 17 |
* Portions Copyright (c) 1994, Regents of the University of California |
| 18 |
* |
| 19 |
* IDENTIFICATION |
| 20 |
* src/backend/statistics/mvdistinct.c |
| 21 |
* |
| 22 |
*------------------------------------------------------------------------- |
| 23 |
*/ |
| 24 |
#include "postgres.h" |
| 25 |
|
| 26 |
#include <math.h> |
| 27 |
|
| 28 |
#include "catalog/pg_statistic_ext.h" |
| 29 |
#include "catalog/pg_statistic_ext_data.h" |
| 30 |
#include "lib/stringinfo.h" |
| 31 |
#include "statistics/extended_stats_internal.h" |
| 32 |
#include "statistics/statistics.h" |
| 33 |
#include "utils/fmgrprotos.h" |
| 34 |
#include "utils/syscache.h" |
| 35 |
#include "utils/typcache.h" |
| 36 |
#include "varatt.h" |
| 37 |
|
| 38 |
static double ndistinct_for_combination(double totalrows, StatsBuildData *data, |
| 39 |
int k, int *combination); |
| 40 |
static double estimate_ndistinct(double totalrows, int numrows, int d, int f1); |
| 41 |
static int n_choose_k(int n, int k); |
| 42 |
static int num_combinations(int n); |
| 43 |
|
| 44 |
/* size of the struct header fields (magic, type, nitems) */ |
| 45 |
#define SizeOfHeader (3 * sizeof(uint32)) |
| 46 |
|
| 47 |
/* size of a serialized ndistinct item (coefficient, natts, atts) */ |
| 48 |
#define SizeOfItem(natts) \ |
| 49 |
(sizeof(double) + sizeof(int) + (natts) * sizeof(AttrNumber)) |
| 50 |
|
| 51 |
/* minimal size of a ndistinct item (with two attributes) */ |
| 52 |
#define MinSizeOfItem SizeOfItem(2) |
| 53 |
|
| 54 |
/* minimal size of mvndistinct, when all items are minimal */ |
| 55 |
#define MinSizeOfItems(nitems) \ |
| 56 |
(SizeOfHeader + (nitems) * MinSizeOfItem) |
| 57 |
|
| 58 |
/* Combination generator API */ |
| 59 |
|
| 60 |
/* internal state for generator of k-combinations of n elements */ |
| 61 |
typedef struct CombinationGenerator |
| 62 |
{ |
| 63 |
int k; /* size of the combination */ |
| 64 |
int n; /* total number of elements */ |
| 65 |
int current; /* index of the next combination to return */ |
| 66 |
int ncombinations; /* number of combinations (size of array) */ |
| 67 |
int *combinations; /* array of pre-built combinations */ |
| 68 |
} CombinationGenerator; |
| 69 |
|
| 70 |
static CombinationGenerator *generator_init(int n, int k); |
| 71 |
static void generator_free(CombinationGenerator *state); |
| 72 |
static int *generator_next(CombinationGenerator *state); |
| 73 |
static void generate_combinations(CombinationGenerator *state); |
| 74 |
|
| 75 |
|
| 76 |
/* |
| 77 |
* statext_ndistinct_build |
| 78 |
* Compute ndistinct coefficient for the combination of attributes. |
| 79 |
* |
| 80 |
* This computes the ndistinct estimate using the same estimator used |
| 81 |
* in analyze.c and then computes the coefficient. |
| 82 |
* |
| 83 |
* To handle expressions easily, we treat them as system attributes with |
| 84 |
* negative attnums, and offset everything by number of expressions to |
| 85 |
* allow using Bitmapsets. |
| 86 |
*/ |
| 87 |
MVNDistinct * |
| 88 |
statext_ndistinct_build(double totalrows, StatsBuildData *data) |
| 89 |
{ |
| 90 |
MVNDistinct *result; |
| 91 |
int k; |
| 92 |
int itemcnt; |
| 93 |
int numattrs = data->nattnums; |
| 94 |
int numcombs = num_combinations(numattrs); |
| 95 |
|
| 96 |
result = palloc(offsetof(MVNDistinct, items) + |
| 97 |
numcombs * sizeof(MVNDistinctItem)); |
| 98 |
result->magic = STATS_NDISTINCT_MAGIC; |
| 99 |
result->type = STATS_NDISTINCT_TYPE_BASIC; |
| 100 |
result->nitems = numcombs; |
| 101 |
|
| 102 |
itemcnt = 0; |
| 103 |
for (k = 2; k <= numattrs; k++) |
| 104 |
{ |
| 105 |
int *combination; |
| 106 |
CombinationGenerator *generator; |
| 107 |
|
| 108 |
/* generate combinations of K out of N elements */ |
| 109 |
generator = generator_init(numattrs, k); |
| 110 |
|
| 111 |
while ((combination = generator_next(generator))) |
| 112 |
{ |
| 113 |
MVNDistinctItem *item = &result->items[itemcnt]; |
| 114 |
int j; |
| 115 |
|
| 116 |
item->attributes = palloc(sizeof(AttrNumber) * k); |
| 117 |
item->nattributes = k; |
| 118 |
|
| 119 |
/* translate the indexes to attnums */ |
| 120 |
for (j = 0; j < k; j++) |
| 121 |
{ |
| 122 |
item->attributes[j] = data->attnums[combination[j]]; |
| 123 |
|
| 124 |
Assert(AttributeNumberIsValid(item->attributes[j])); |
| 125 |
} |
| 126 |
|
| 127 |
item->ndistinct = |
| 128 |
ndistinct_for_combination(totalrows, data, k, combination); |
| 129 |
|
| 130 |
itemcnt++; |
| 131 |
Assert(itemcnt <= result->nitems); |
| 132 |
} |
| 133 |
|
| 134 |
generator_free(generator); |
| 135 |
} |
| 136 |
|
| 137 |
/* must consume exactly the whole output array */ |
| 138 |
Assert(itemcnt == result->nitems); |
| 139 |
|
| 140 |
return result; |
| 141 |
} |
| 142 |
|
| 143 |
/* |
| 144 |
* statext_ndistinct_load |
| 145 |
* Load the ndistinct value for the indicated pg_statistic_ext tuple |
| 146 |
*/ |
| 147 |
MVNDistinct * |
| 148 |
statext_ndistinct_load(Oid mvoid, bool inh) |
| 149 |
{ |
| 150 |
MVNDistinct *result; |
| 151 |
bool isnull; |
| 152 |
Datum ndist; |
| 153 |
HeapTuple htup; |
| 154 |
|
| 155 |
htup = SearchSysCache2(STATEXTDATASTXOID, |
| 156 |
ObjectIdGetDatum(mvoid), BoolGetDatum(inh)); |
| 157 |
if (!HeapTupleIsValid(htup)) |
| 158 |
elog(ERROR, "cache lookup failed for statistics object %u", mvoid); |
| 159 |
|
| 160 |
ndist = SysCacheGetAttr(STATEXTDATASTXOID, htup, |
| 161 |
Anum_pg_statistic_ext_data_stxdndistinct, &isnull); |
| 162 |
if (isnull) |
| 163 |
elog(ERROR, |
| 164 |
"requested statistics kind \"%c\" is not yet built for statistics object %u", |
| 165 |
STATS_EXT_NDISTINCT, mvoid); |
| 166 |
|
| 167 |
result = statext_ndistinct_deserialize(DatumGetByteaPP(ndist)); |
| 168 |
|
| 169 |
ReleaseSysCache(htup); |
| 170 |
|
| 171 |
return result; |
| 172 |
} |
| 173 |
|
| 174 |
/* |
| 175 |
* statext_ndistinct_serialize |
| 176 |
* serialize ndistinct to the on-disk bytea format |
| 177 |
*/ |
| 178 |
bytea * |
| 179 |
statext_ndistinct_serialize(MVNDistinct *ndistinct) |
| 180 |
{ |
| 181 |
int i; |
| 182 |
bytea *output; |
| 183 |
char *tmp; |
| 184 |
Size len; |
| 185 |
|
| 186 |
Assert(ndistinct->magic == STATS_NDISTINCT_MAGIC); |
| 187 |
Assert(ndistinct->type == STATS_NDISTINCT_TYPE_BASIC); |
| 188 |
|
| 189 |
/* |
| 190 |
* Base size is size of scalar fields in the struct, plus one base struct |
| 191 |
* for each item, including number of items for each. |
| 192 |
*/ |
| 193 |
len = VARHDRSZ + SizeOfHeader; |
| 194 |
|
| 195 |
/* and also include space for the actual attribute numbers */ |
| 196 |
for (i = 0; i < ndistinct->nitems; i++) |
| 197 |
{ |
| 198 |
int nmembers; |
| 199 |
|
| 200 |
nmembers = ndistinct->items[i].nattributes; |
| 201 |
Assert(nmembers >= 2); |
| 202 |
|
| 203 |
len += SizeOfItem(nmembers); |
| 204 |
} |
| 205 |
|
| 206 |
output = (bytea *) palloc(len); |
| 207 |
SET_VARSIZE(output, len); |
| 208 |
|
| 209 |
tmp = VARDATA(output); |
| 210 |
|
| 211 |
/* Store the base struct values (magic, type, nitems) */ |
| 212 |
memcpy(tmp, &ndistinct->magic, sizeof(uint32)); |
| 213 |
tmp += sizeof(uint32); |
| 214 |
memcpy(tmp, &ndistinct->type, sizeof(uint32)); |
| 215 |
tmp += sizeof(uint32); |
| 216 |
memcpy(tmp, &ndistinct->nitems, sizeof(uint32)); |
| 217 |
tmp += sizeof(uint32); |
| 218 |
|
| 219 |
/* |
| 220 |
* store number of attributes and attribute numbers for each entry |
| 221 |
*/ |
| 222 |
for (i = 0; i < ndistinct->nitems; i++) |
| 223 |
{ |
| 224 |
MVNDistinctItem item = ndistinct->items[i]; |
| 225 |
int nmembers = item.nattributes; |
| 226 |
|
| 227 |
memcpy(tmp, &item.ndistinct, sizeof(double)); |
| 228 |
tmp += sizeof(double); |
| 229 |
memcpy(tmp, &nmembers, sizeof(int)); |
| 230 |
tmp += sizeof(int); |
| 231 |
|
| 232 |
memcpy(tmp, item.attributes, sizeof(AttrNumber) * nmembers); |
| 233 |
tmp += nmembers * sizeof(AttrNumber); |
| 234 |
|
| 235 |
/* protect against overflows */ |
| 236 |
Assert(tmp <= ((char *) output + len)); |
| 237 |
} |
| 238 |
|
| 239 |
/* check we used exactly the expected space */ |
| 240 |
Assert(tmp == ((char *) output + len)); |
| 241 |
|
| 242 |
return output; |
| 243 |
} |
| 244 |
|
| 245 |
/* |
| 246 |
* statext_ndistinct_deserialize |
| 247 |
* Read an on-disk bytea format MVNDistinct to in-memory format |
| 248 |
*/ |
| 249 |
MVNDistinct * |
| 250 |
statext_ndistinct_deserialize(bytea *data) |
| 251 |
{ |
| 252 |
int i; |
| 253 |
Size minimum_size; |
| 254 |
MVNDistinct ndist; |
| 255 |
MVNDistinct *ndistinct; |
| 256 |
char *tmp; |
| 257 |
|
| 258 |
if (data == NULL) |
| 259 |
return NULL; |
| 260 |
|
| 261 |
/* we expect at least the basic fields of MVNDistinct struct */ |
| 262 |
if (VARSIZE_ANY_EXHDR(data) < SizeOfHeader) |
| 263 |
elog(ERROR, "invalid MVNDistinct size %zu (expected at least %zu)", |
| 264 |
VARSIZE_ANY_EXHDR(data), SizeOfHeader); |
| 265 |
|
| 266 |
/* initialize pointer to the data part (skip the varlena header) */ |
| 267 |
tmp = VARDATA_ANY(data); |
| 268 |
|
| 269 |
/* read the header fields and perform basic sanity checks */ |
| 270 |
memcpy(&ndist.magic, tmp, sizeof(uint32)); |
| 271 |
tmp += sizeof(uint32); |
| 272 |
memcpy(&ndist.type, tmp, sizeof(uint32)); |
| 273 |
tmp += sizeof(uint32); |
| 274 |
memcpy(&ndist.nitems, tmp, sizeof(uint32)); |
| 275 |
tmp += sizeof(uint32); |
| 276 |
|
| 277 |
if (ndist.magic != STATS_NDISTINCT_MAGIC) |
| 278 |
elog(ERROR, "invalid ndistinct magic %08x (expected %08x)", |
| 279 |
ndist.magic, STATS_NDISTINCT_MAGIC); |
| 280 |
if (ndist.type != STATS_NDISTINCT_TYPE_BASIC) |
| 281 |
elog(ERROR, "invalid ndistinct type %d (expected %d)", |
| 282 |
ndist.type, STATS_NDISTINCT_TYPE_BASIC); |
| 283 |
if (ndist.nitems == 0) |
| 284 |
elog(ERROR, "invalid zero-length item array in MVNDistinct"); |
| 285 |
|
| 286 |
/* what minimum bytea size do we expect for those parameters */ |
| 287 |
minimum_size = MinSizeOfItems(ndist.nitems); |
| 288 |
if (VARSIZE_ANY_EXHDR(data) < minimum_size) |
| 289 |
elog(ERROR, "invalid MVNDistinct size %zu (expected at least %zu)", |
| 290 |
VARSIZE_ANY_EXHDR(data), minimum_size); |
| 291 |
|
| 292 |
/* |
| 293 |
* Allocate space for the ndistinct items (no space for each item's |
| 294 |
* attnos: those live in bitmapsets allocated separately) |
| 295 |
*/ |
| 296 |
ndistinct = palloc0(MAXALIGN(offsetof(MVNDistinct, items)) + |
| 297 |
(ndist.nitems * sizeof(MVNDistinctItem))); |
| 298 |
ndistinct->magic = ndist.magic; |
| 299 |
ndistinct->type = ndist.type; |
| 300 |
ndistinct->nitems = ndist.nitems; |
| 301 |
|
| 302 |
for (i = 0; i < ndistinct->nitems; i++) |
| 303 |
{ |
| 304 |
MVNDistinctItem *item = &ndistinct->items[i]; |
| 305 |
|
| 306 |
/* ndistinct value */ |
| 307 |
memcpy(&item->ndistinct, tmp, sizeof(double)); |
| 308 |
tmp += sizeof(double); |
| 309 |
|
| 310 |
/* number of attributes */ |
| 311 |
memcpy(&item->nattributes, tmp, sizeof(int)); |
| 312 |
tmp += sizeof(int); |
| 313 |
Assert((item->nattributes >= 2) && (item->nattributes <= STATS_MAX_DIMENSIONS)); |
| 314 |
|
| 315 |
item->attributes |
| 316 |
= (AttrNumber *) palloc(item->nattributes * sizeof(AttrNumber)); |
| 317 |
|
| 318 |
memcpy(item->attributes, tmp, sizeof(AttrNumber) * item->nattributes); |
| 319 |
tmp += sizeof(AttrNumber) * item->nattributes; |
| 320 |
|
| 321 |
/* still within the bytea */ |
| 322 |
Assert(tmp <= ((char *) data + VARSIZE_ANY(data))); |
| 323 |
} |
| 324 |
|
| 325 |
/* we should have consumed the whole bytea exactly */ |
| 326 |
Assert(tmp == ((char *) data + VARSIZE_ANY(data))); |
| 327 |
|
| 328 |
return ndistinct; |
| 329 |
} |
| 330 |
|
| 331 |
/* |
| 332 |
* pg_ndistinct_in |
| 333 |
* input routine for type pg_ndistinct |
| 334 |
* |
| 335 |
* pg_ndistinct is real enough to be a table column, but it has no |
| 336 |
* operations of its own, and disallows input (just like pg_node_tree). |
| 337 |
*/ |
| 338 |
Datum |
| 339 |
pg_ndistinct_in(PG_FUNCTION_ARGS) |
| 340 |
{ |
| 341 |
ereport(ERROR, |
| 342 |
(errcode(ERRCODE_FEATURE_NOT_SUPPORTED), |
| 343 |
errmsg("cannot accept a value of type %s", "pg_ndistinct"))); |
| 344 |
|
| 345 |
PG_RETURN_VOID(); /* keep compiler quiet */ |
| 346 |
} |
| 347 |
|
| 348 |
/* |
| 349 |
* pg_ndistinct |
| 350 |
* output routine for type pg_ndistinct |
| 351 |
* |
| 352 |
* Produces a human-readable representation of the value. |
| 353 |
*/ |
| 354 |
Datum |
| 355 |
pg_ndistinct_out(PG_FUNCTION_ARGS) |
| 356 |
{ |
| 357 |
bytea *data = PG_GETARG_BYTEA_PP(0); |
| 358 |
MVNDistinct *ndist = statext_ndistinct_deserialize(data); |
| 359 |
int i; |
| 360 |
StringInfoData str; |
| 361 |
|
| 362 |
initStringInfo(&str); |
| 363 |
appendStringInfoChar(&str, '{'); |
| 364 |
|
| 365 |
for (i = 0; i < ndist->nitems; i++) |
| 366 |
{ |
| 367 |
int j; |
| 368 |
MVNDistinctItem item = ndist->items[i]; |
| 369 |
|
| 370 |
if (i > 0) |
| 371 |
appendStringInfoString(&str, ", "); |
| 372 |
|
| 373 |
for (j = 0; j < item.nattributes; j++) |
| 374 |
{ |
| 375 |
AttrNumber attnum = item.attributes[j]; |
| 376 |
|
| 377 |
appendStringInfo(&str, "%s%d", (j == 0) ? "\"" : ", ", attnum); |
| 378 |
} |
| 379 |
appendStringInfo(&str, "\": %d", (int) item.ndistinct); |
| 380 |
} |
| 381 |
|
| 382 |
appendStringInfoChar(&str, '}'); |
| 383 |
|
| 384 |
PG_RETURN_CSTRING(str.data); |
| 385 |
} |
| 386 |
|
| 387 |
/* |
| 388 |
* pg_ndistinct_recv |
| 389 |
* binary input routine for type pg_ndistinct |
| 390 |
*/ |
| 391 |
Datum |
| 392 |
pg_ndistinct_recv(PG_FUNCTION_ARGS) |
| 393 |
{ |
| 394 |
ereport(ERROR, |
| 395 |
(errcode(ERRCODE_FEATURE_NOT_SUPPORTED), |
| 396 |
errmsg("cannot accept a value of type %s", "pg_ndistinct"))); |
| 397 |
|
| 398 |
PG_RETURN_VOID(); /* keep compiler quiet */ |
| 399 |
} |
| 400 |
|
| 401 |
/* |
| 402 |
* pg_ndistinct_send |
| 403 |
* binary output routine for type pg_ndistinct |
| 404 |
* |
| 405 |
* n-distinct is serialized into a bytea value, so let's send that. |
| 406 |
*/ |
| 407 |
Datum |
| 408 |
pg_ndistinct_send(PG_FUNCTION_ARGS) |
| 409 |
{ |
| 410 |
return byteasend(fcinfo); |
| 411 |
} |
| 412 |
|
| 413 |
/* |
| 414 |
* ndistinct_for_combination |
| 415 |
* Estimates number of distinct values in a combination of columns. |
| 416 |
* |
| 417 |
* This uses the same ndistinct estimator as compute_scalar_stats() in |
| 418 |
* ANALYZE, i.e., |
| 419 |
* n*d / (n - f1 + f1*n/N) |
| 420 |
* |
| 421 |
* except that instead of values in a single column we are dealing with |
| 422 |
* combination of multiple columns. |
| 423 |
*/ |
| 424 |
static double |
| 425 |
ndistinct_for_combination(double totalrows, StatsBuildData *data, |
| 426 |
int k, int *combination) |
| 427 |
{ |
| 428 |
int i, |
| 429 |
j; |
| 430 |
int f1, |
| 431 |
cnt, |
| 432 |
d; |
| 433 |
bool *isnull; |
| 434 |
Datum *values; |
| 435 |
SortItem *items; |
| 436 |
MultiSortSupport mss; |
| 437 |
int numrows = data->numrows; |
| 438 |
|
| 439 |
mss = multi_sort_init(k); |
| 440 |
|
| 441 |
/* |
| 442 |
* In order to determine the number of distinct elements, create separate |
| 443 |
* values[]/isnull[] arrays with all the data we have, then sort them |
| 444 |
* using the specified column combination as dimensions. We could try to |
| 445 |
* sort in place, but it'd probably be more complex and bug-prone. |
| 446 |
*/ |
| 447 |
items = (SortItem *) palloc(numrows * sizeof(SortItem)); |
| 448 |
values = (Datum *) palloc0(sizeof(Datum) * numrows * k); |
| 449 |
isnull = (bool *) palloc0(sizeof(bool) * numrows * k); |
| 450 |
|
| 451 |
for (i = 0; i < numrows; i++) |
| 452 |
{ |
| 453 |
items[i].values = &values[i * k]; |
| 454 |
items[i].isnull = &isnull[i * k]; |
| 455 |
} |
| 456 |
|
| 457 |
/* |
| 458 |
* For each dimension, set up sort-support and fill in the values from the |
| 459 |
* sample data. |
| 460 |
* |
| 461 |
* We use the column data types' default sort operators and collations; |
| 462 |
* perhaps at some point it'd be worth using column-specific collations? |
| 463 |
*/ |
| 464 |
for (i = 0; i < k; i++) |
| 465 |
{ |
| 466 |
Oid typid; |
| 467 |
TypeCacheEntry *type; |
| 468 |
Oid collid = InvalidOid; |
| 469 |
VacAttrStats *colstat = data->stats[combination[i]]; |
| 470 |
|
| 471 |
typid = colstat->attrtypid; |
| 472 |
collid = colstat->attrcollid; |
| 473 |
|
| 474 |
type = lookup_type_cache(typid, TYPECACHE_LT_OPR); |
| 475 |
if (type->lt_opr == InvalidOid) /* shouldn't happen */ |
| 476 |
elog(ERROR, "cache lookup failed for ordering operator for type %u", |
| 477 |
typid); |
| 478 |
|
| 479 |
/* prepare the sort function for this dimension */ |
| 480 |
multi_sort_add_dimension(mss, i, type->lt_opr, collid); |
| 481 |
|
| 482 |
/* accumulate all the data for this dimension into the arrays */ |
| 483 |
for (j = 0; j < numrows; j++) |
| 484 |
{ |
| 485 |
items[j].values[i] = data->values[combination[i]][j]; |
| 486 |
items[j].isnull[i] = data->nulls[combination[i]][j]; |
| 487 |
} |
| 488 |
} |
| 489 |
|
| 490 |
/* We can sort the array now ... */ |
| 491 |
qsort_interruptible(items, numrows, sizeof(SortItem), |
| 492 |
multi_sort_compare, mss); |
| 493 |
|
| 494 |
/* ... and count the number of distinct combinations */ |
| 495 |
|
| 496 |
f1 = 0; |
| 497 |
cnt = 1; |
| 498 |
d = 1; |
| 499 |
for (i = 1; i < numrows; i++) |
| 500 |
{ |
| 501 |
if (multi_sort_compare(&items[i], &items[i - 1], mss) != 0) |
| 502 |
{ |
| 503 |
if (cnt == 1) |
| 504 |
f1 += 1; |
| 505 |
|
| 506 |
d++; |
| 507 |
cnt = 0; |
| 508 |
} |
| 509 |
|
| 510 |
cnt += 1; |
| 511 |
} |
| 512 |
|
| 513 |
if (cnt == 1) |
| 514 |
f1 += 1; |
| 515 |
|
| 516 |
return estimate_ndistinct(totalrows, numrows, d, f1); |
| 517 |
} |
| 518 |
|
| 519 |
/* The Duj1 estimator (already used in analyze.c). */ |
| 520 |
static double |
| 521 |
estimate_ndistinct(double totalrows, int numrows, int d, int f1) |
| 522 |
{ |
| 523 |
double numer, |
| 524 |
denom, |
| 525 |
ndistinct; |
| 526 |
|
| 527 |
numer = (double) numrows * (double) d; |
| 528 |
|
| 529 |
denom = (double) (numrows - f1) + |
| 530 |
(double) f1 * (double) numrows / totalrows; |
| 531 |
|
| 532 |
ndistinct = numer / denom; |
| 533 |
|
| 534 |
/* Clamp to sane range in case of roundoff error */ |
| 535 |
if (ndistinct < (double) d) |
| 536 |
ndistinct = (double) d; |
| 537 |
|
| 538 |
if (ndistinct > totalrows) |
| 539 |
ndistinct = totalrows; |
| 540 |
|
| 541 |
return floor(ndistinct + 0.5); |
| 542 |
} |
| 543 |
|
| 544 |
/* |
| 545 |
* n_choose_k |
| 546 |
* computes binomial coefficients using an algorithm that is both |
| 547 |
* efficient and prevents overflows |
| 548 |
*/ |
| 549 |
static int |
| 550 |
n_choose_k(int n, int k) |
| 551 |
{ |
| 552 |
int d, |
| 553 |
r; |
| 554 |
|
| 555 |
Assert((k > 0) && (n >= k)); |
| 556 |
|
| 557 |
/* use symmetry of the binomial coefficients */ |
| 558 |
k = Min(k, n - k); |
| 559 |
|
| 560 |
r = 1; |
| 561 |
for (d = 1; d <= k; ++d) |
| 562 |
{ |
| 563 |
r *= n--; |
| 564 |
r /= d; |
| 565 |
} |
| 566 |
|
| 567 |
return r; |
| 568 |
} |
| 569 |
|
| 570 |
/* |
| 571 |
* num_combinations |
| 572 |
* number of combinations, excluding single-value combinations |
| 573 |
*/ |
| 574 |
static int |
| 575 |
num_combinations(int n) |
| 576 |
{ |
| 577 |
return (1 << n) - (n + 1); |
| 578 |
} |
| 579 |
|
| 580 |
/* |
| 581 |
* generator_init |
| 582 |
* initialize the generator of combinations |
| 583 |
* |
| 584 |
* The generator produces combinations of K elements in the interval (0..N). |
| 585 |
* We prebuild all the combinations in this method, which is simpler than |
| 586 |
* generating them on the fly. |
| 587 |
*/ |
| 588 |
static CombinationGenerator * |
| 589 |
generator_init(int n, int k) |
| 590 |
{ |
| 591 |
CombinationGenerator *state; |
| 592 |
|
| 593 |
Assert((n >= k) && (k > 0)); |
| 594 |
|
| 595 |
/* allocate the generator state as a single chunk of memory */ |
| 596 |
state = (CombinationGenerator *) palloc(sizeof(CombinationGenerator)); |
| 597 |
|
| 598 |
state->ncombinations = n_choose_k(n, k); |
| 599 |
|
| 600 |
/* pre-allocate space for all combinations */ |
| 601 |
state->combinations = (int *) palloc(sizeof(int) * k * state->ncombinations); |
| 602 |
|
| 603 |
state->current = 0; |
| 604 |
state->k = k; |
| 605 |
state->n = n; |
| 606 |
|
| 607 |
/* now actually pre-generate all the combinations of K elements */ |
| 608 |
generate_combinations(state); |
| 609 |
|
| 610 |
/* make sure we got the expected number of combinations */ |
| 611 |
Assert(state->current == state->ncombinations); |
| 612 |
|
| 613 |
/* reset the number, so we start with the first one */ |
| 614 |
state->current = 0; |
| 615 |
|
| 616 |
return state; |
| 617 |
} |
| 618 |
|
| 619 |
/* |
| 620 |
* generator_next |
| 621 |
* returns the next combination from the prebuilt list |
| 622 |
* |
| 623 |
* Returns a combination of K array indexes (0 .. N), as specified to |
| 624 |
* generator_init), or NULL when there are no more combination. |
| 625 |
*/ |
| 626 |
static int * |
| 627 |
generator_next(CombinationGenerator *state) |
| 628 |
{ |
| 629 |
if (state->current == state->ncombinations) |
| 630 |
return NULL; |
| 631 |
|
| 632 |
return &state->combinations[state->k * state->current++]; |
| 633 |
} |
| 634 |
|
| 635 |
/* |
| 636 |
* generator_free |
| 637 |
* free the internal state of the generator |
| 638 |
* |
| 639 |
* Releases the generator internal state (pre-built combinations). |
| 640 |
*/ |
| 641 |
static void |
| 642 |
generator_free(CombinationGenerator *state) |
| 643 |
{ |
| 644 |
pfree(state->combinations); |
| 645 |
pfree(state); |
| 646 |
} |
| 647 |
|
| 648 |
/* |
| 649 |
* generate_combinations_recurse |
| 650 |
* given a prefix, generate all possible combinations |
| 651 |
* |
| 652 |
* Given a prefix (first few elements of the combination), generate following |
| 653 |
* elements recursively. We generate the combinations in lexicographic order, |
| 654 |
* which eliminates permutations of the same combination. |
| 655 |
*/ |
| 656 |
static void |
| 657 |
generate_combinations_recurse(CombinationGenerator *state, |
| 658 |
int index, int start, int *current) |
| 659 |
{ |
| 660 |
/* If we haven't filled all the elements, simply recurse. */ |
| 661 |
if (index < state->k) |
| 662 |
{ |
| 663 |
int i; |
| 664 |
|
| 665 |
/* |
| 666 |
* The values have to be in ascending order, so make sure we start |
| 667 |
* with the value passed by parameter. |
| 668 |
*/ |
| 669 |
|
| 670 |
for (i = start; i < state->n; i++) |
| 671 |
{ |
| 672 |
current[index] = i; |
| 673 |
generate_combinations_recurse(state, (index + 1), (i + 1), current); |
| 674 |
} |
| 675 |
|
| 676 |
return; |
| 677 |
} |
| 678 |
else |
| 679 |
{ |
| 680 |
/* we got a valid combination, add it to the array */ |
| 681 |
memcpy(&state->combinations[(state->k * state->current)], |
| 682 |
current, state->k * sizeof(int)); |
| 683 |
state->current++; |
| 684 |
} |
| 685 |
} |
| 686 |
|
| 687 |
/* |
| 688 |
* generate_combinations |
| 689 |
* generate all k-combinations of N elements |
| 690 |
*/ |
| 691 |
static void |
| 692 |
generate_combinations(CombinationGenerator *state) |
| 693 |
{ |
| 694 |
int *current = (int *) palloc0(sizeof(int) * state->k); |
| 695 |
|
| 696 |
generate_combinations_recurse(state, 0, 0, current); |
| 697 |
|
| 698 |
pfree(current); |
| 699 |
} |