PostgreSQL Source Code  git master
ts_typanalyze.c
Go to the documentation of this file.
1 /*-------------------------------------------------------------------------
2  *
3  * ts_typanalyze.c
4  * functions for gathering statistics from tsvector columns
5  *
6  * Portions Copyright (c) 1996-2017, PostgreSQL Global Development Group
7  *
8  *
9  * IDENTIFICATION
10  * src/backend/tsearch/ts_typanalyze.c
11  *
12  *-------------------------------------------------------------------------
13  */
14 #include "postgres.h"
15 
16 #include "access/hash.h"
17 #include "catalog/pg_operator.h"
18 #include "commands/vacuum.h"
19 #include "tsearch/ts_type.h"
20 #include "utils/builtins.h"
21 
22 
23 /* A hash key for lexemes */
24 typedef struct
25 {
26  char *lexeme; /* lexeme (not NULL terminated!) */
27  int length; /* its length in bytes */
29 
30 /* A hash table entry for the Lossy Counting algorithm */
31 typedef struct
32 {
33  LexemeHashKey key; /* This is 'e' from the LC algorithm. */
34  int frequency; /* This is 'f'. */
35  int delta; /* And this is 'delta'. */
36 } TrackItem;
37 
38 static void compute_tsvector_stats(VacAttrStats *stats,
39  AnalyzeAttrFetchFunc fetchfunc,
40  int samplerows,
41  double totalrows);
42 static void prune_lexemes_hashtable(HTAB *lexemes_tab, int b_current);
43 static uint32 lexeme_hash(const void *key, Size keysize);
44 static int lexeme_match(const void *key1, const void *key2, Size keysize);
45 static int lexeme_compare(const void *key1, const void *key2);
46 static int trackitem_compare_frequencies_desc(const void *e1, const void *e2);
47 static int trackitem_compare_lexemes(const void *e1, const void *e2);
48 
49 
50 /*
51  * ts_typanalyze -- a custom typanalyze function for tsvector columns
52  */
53 Datum
55 {
57  Form_pg_attribute attr = stats->attr;
58 
59  /* If the attstattarget column is negative, use the default value */
60  /* NB: it is okay to scribble on stats->attr since it's a copy */
61  if (attr->attstattarget < 0)
62  attr->attstattarget = default_statistics_target;
63 
65  /* see comment about the choice of minrows in commands/analyze.c */
66  stats->minrows = 300 * attr->attstattarget;
67 
68  PG_RETURN_BOOL(true);
69 }
70 
71 /*
72  * compute_tsvector_stats() -- compute statistics for a tsvector column
73  *
74  * This functions computes statistics that are useful for determining @@
75  * operations' selectivity, along with the fraction of non-null rows and
76  * average width.
77  *
78  * Instead of finding the most common values, as we do for most datatypes,
79  * we're looking for the most common lexemes. This is more useful, because
80  * there most probably won't be any two rows with the same tsvector and thus
81  * the notion of a MCV is a bit bogus with this datatype. With a list of the
82  * most common lexemes we can do a better job at figuring out @@ selectivity.
83  *
84  * For the same reasons we assume that tsvector columns are unique when
85  * determining the number of distinct values.
86  *
87  * The algorithm used is Lossy Counting, as proposed in the paper "Approximate
88  * frequency counts over data streams" by G. S. Manku and R. Motwani, in
89  * Proceedings of the 28th International Conference on Very Large Data Bases,
90  * Hong Kong, China, August 2002, section 4.2. The paper is available at
91  * http://www.vldb.org/conf/2002/S10P03.pdf
92  *
93  * The Lossy Counting (aka LC) algorithm goes like this:
94  * Let s be the threshold frequency for an item (the minimum frequency we
95  * are interested in) and epsilon the error margin for the frequency. Let D
96  * be a set of triples (e, f, delta), where e is an element value, f is that
97  * element's frequency (actually, its current occurrence count) and delta is
98  * the maximum error in f. We start with D empty and process the elements in
99  * batches of size w. (The batch size is also known as "bucket size" and is
100  * equal to 1/epsilon.) Let the current batch number be b_current, starting
101  * with 1. For each element e we either increment its f count, if it's
102  * already in D, or insert a new triple into D with values (e, 1, b_current
103  * - 1). After processing each batch we prune D, by removing from it all
104  * elements with f + delta <= b_current. After the algorithm finishes we
105  * suppress all elements from D that do not satisfy f >= (s - epsilon) * N,
106  * where N is the total number of elements in the input. We emit the
107  * remaining elements with estimated frequency f/N. The LC paper proves
108  * that this algorithm finds all elements with true frequency at least s,
109  * and that no frequency is overestimated or is underestimated by more than
110  * epsilon. Furthermore, given reasonable assumptions about the input
111  * distribution, the required table size is no more than about 7 times w.
112  *
113  * We set s to be the estimated frequency of the K'th word in a natural
114  * language's frequency table, where K is the target number of entries in
115  * the MCELEM array plus an arbitrary constant, meant to reflect the fact
116  * that the most common words in any language would usually be stopwords
117  * so we will not actually see them in the input. We assume that the
118  * distribution of word frequencies (including the stopwords) follows Zipf's
119  * law with an exponent of 1.
120  *
121  * Assuming Zipfian distribution, the frequency of the K'th word is equal
122  * to 1/(K * H(W)) where H(n) is 1/2 + 1/3 + ... + 1/n and W is the number of
123  * words in the language. Putting W as one million, we get roughly 0.07/K.
124  * Assuming top 10 words are stopwords gives s = 0.07/(K + 10). We set
125  * epsilon = s/10, which gives bucket width w = (K + 10)/0.007 and
126  * maximum expected hashtable size of about 1000 * (K + 10).
127  *
128  * Note: in the above discussion, s, epsilon, and f/N are in terms of a
129  * lexeme's frequency as a fraction of all lexemes seen in the input.
130  * However, what we actually want to store in the finished pg_statistic
131  * entry is each lexeme's frequency as a fraction of all rows that it occurs
132  * in. Assuming that the input tsvectors are correctly constructed, no
133  * lexeme occurs more than once per tsvector, so the final count f is a
134  * correct estimate of the number of input tsvectors it occurs in, and we
135  * need only change the divisor from N to nonnull_cnt to get the number we
136  * want.
137  */
138 static void
140  AnalyzeAttrFetchFunc fetchfunc,
141  int samplerows,
142  double totalrows)
143 {
144  int num_mcelem;
145  int null_cnt = 0;
146  double total_width = 0;
147 
148  /* This is D from the LC algorithm. */
149  HTAB *lexemes_tab;
150  HASHCTL hash_ctl;
151  HASH_SEQ_STATUS scan_status;
152 
153  /* This is the current bucket number from the LC algorithm */
154  int b_current;
155 
156  /* This is 'w' from the LC algorithm */
157  int bucket_width;
158  int vector_no,
159  lexeme_no;
161  TrackItem *item;
162 
163  /*
164  * We want statistics_target * 10 lexemes in the MCELEM array. This
165  * multiplier is pretty arbitrary, but is meant to reflect the fact that
166  * the number of individual lexeme values tracked in pg_statistic ought to
167  * be more than the number of values for a simple scalar column.
168  */
169  num_mcelem = stats->attr->attstattarget * 10;
170 
171  /*
172  * We set bucket width equal to (num_mcelem + 10) / 0.007 as per the
173  * comment above.
174  */
175  bucket_width = (num_mcelem + 10) * 1000 / 7;
176 
177  /*
178  * Create the hashtable. It will be in local memory, so we don't need to
179  * worry about overflowing the initial size. Also we don't need to pay any
180  * attention to locking and memory management.
181  */
182  MemSet(&hash_ctl, 0, sizeof(hash_ctl));
183  hash_ctl.keysize = sizeof(LexemeHashKey);
184  hash_ctl.entrysize = sizeof(TrackItem);
185  hash_ctl.hash = lexeme_hash;
186  hash_ctl.match = lexeme_match;
187  hash_ctl.hcxt = CurrentMemoryContext;
188  lexemes_tab = hash_create("Analyzed lexemes table",
189  num_mcelem,
190  &hash_ctl,
192 
193  /* Initialize counters. */
194  b_current = 1;
195  lexeme_no = 0;
196 
197  /* Loop over the tsvectors. */
198  for (vector_no = 0; vector_no < samplerows; vector_no++)
199  {
200  Datum value;
201  bool isnull;
202  TSVector vector;
203  WordEntry *curentryptr;
204  char *lexemesptr;
205  int j;
206 
208 
209  value = fetchfunc(stats, vector_no, &isnull);
210 
211  /*
212  * Check for null/nonnull.
213  */
214  if (isnull)
215  {
216  null_cnt++;
217  continue;
218  }
219 
220  /*
221  * Add up widths for average-width calculation. Since it's a
222  * tsvector, we know it's varlena. As in the regular
223  * compute_minimal_stats function, we use the toasted width for this
224  * calculation.
225  */
226  total_width += VARSIZE_ANY(DatumGetPointer(value));
227 
228  /*
229  * Now detoast the tsvector if needed.
230  */
231  vector = DatumGetTSVector(value);
232 
233  /*
234  * We loop through the lexemes in the tsvector and add them to our
235  * tracking hashtable.
236  */
237  lexemesptr = STRPTR(vector);
238  curentryptr = ARRPTR(vector);
239  for (j = 0; j < vector->size; j++)
240  {
241  bool found;
242 
243  /*
244  * Construct a hash key. The key points into the (detoasted)
245  * tsvector value at this point, but if a new entry is created, we
246  * make a copy of it. This way we can free the tsvector value
247  * once we've processed all its lexemes.
248  */
249  hash_key.lexeme = lexemesptr + curentryptr->pos;
250  hash_key.length = curentryptr->len;
251 
252  /* Lookup current lexeme in hashtable, adding it if new */
253  item = (TrackItem *) hash_search(lexemes_tab,
254  (const void *) &hash_key,
255  HASH_ENTER, &found);
256 
257  if (found)
258  {
259  /* The lexeme is already on the tracking list */
260  item->frequency++;
261  }
262  else
263  {
264  /* Initialize new tracking list element */
265  item->frequency = 1;
266  item->delta = b_current - 1;
267 
268  item->key.lexeme = palloc(hash_key.length);
269  memcpy(item->key.lexeme, hash_key.lexeme, hash_key.length);
270  }
271 
272  /* lexeme_no is the number of elements processed (ie N) */
273  lexeme_no++;
274 
275  /* We prune the D structure after processing each bucket */
276  if (lexeme_no % bucket_width == 0)
277  {
278  prune_lexemes_hashtable(lexemes_tab, b_current);
279  b_current++;
280  }
281 
282  /* Advance to the next WordEntry in the tsvector */
283  curentryptr++;
284  }
285 
286  /* If the vector was toasted, free the detoasted copy. */
287  if (TSVectorGetDatum(vector) != value)
288  pfree(vector);
289  }
290 
291  /* We can only compute real stats if we found some non-null values. */
292  if (null_cnt < samplerows)
293  {
294  int nonnull_cnt = samplerows - null_cnt;
295  int i;
296  TrackItem **sort_table;
297  int track_len;
298  int cutoff_freq;
299  int minfreq,
300  maxfreq;
301 
302  stats->stats_valid = true;
303  /* Do the simple null-frac and average width stats */
304  stats->stanullfrac = (double) null_cnt / (double) samplerows;
305  stats->stawidth = total_width / (double) nonnull_cnt;
306 
307  /* Assume it's a unique column (see notes above) */
308  stats->stadistinct = -1.0 * (1.0 - stats->stanullfrac);
309 
310  /*
311  * Construct an array of the interesting hashtable items, that is,
312  * those meeting the cutoff frequency (s - epsilon)*N. Also identify
313  * the minimum and maximum frequencies among these items.
314  *
315  * Since epsilon = s/10 and bucket_width = 1/epsilon, the cutoff
316  * frequency is 9*N / bucket_width.
317  */
318  cutoff_freq = 9 * lexeme_no / bucket_width;
319 
320  i = hash_get_num_entries(lexemes_tab); /* surely enough space */
321  sort_table = (TrackItem **) palloc(sizeof(TrackItem *) * i);
322 
323  hash_seq_init(&scan_status, lexemes_tab);
324  track_len = 0;
325  minfreq = lexeme_no;
326  maxfreq = 0;
327  while ((item = (TrackItem *) hash_seq_search(&scan_status)) != NULL)
328  {
329  if (item->frequency > cutoff_freq)
330  {
331  sort_table[track_len++] = item;
332  minfreq = Min(minfreq, item->frequency);
333  maxfreq = Max(maxfreq, item->frequency);
334  }
335  }
336  Assert(track_len <= i);
337 
338  /* emit some statistics for debug purposes */
339  elog(DEBUG3, "tsvector_stats: target # mces = %d, bucket width = %d, "
340  "# lexemes = %d, hashtable size = %d, usable entries = %d",
341  num_mcelem, bucket_width, lexeme_no, i, track_len);
342 
343  /*
344  * If we obtained more lexemes than we really want, get rid of those
345  * with least frequencies. The easiest way is to qsort the array into
346  * descending frequency order and truncate the array.
347  */
348  if (num_mcelem < track_len)
349  {
350  qsort(sort_table, track_len, sizeof(TrackItem *),
352  /* reset minfreq to the smallest frequency we're keeping */
353  minfreq = sort_table[num_mcelem - 1]->frequency;
354  }
355  else
356  num_mcelem = track_len;
357 
358  /* Generate MCELEM slot entry */
359  if (num_mcelem > 0)
360  {
361  MemoryContext old_context;
362  Datum *mcelem_values;
363  float4 *mcelem_freqs;
364 
365  /*
366  * We want to store statistics sorted on the lexeme value using
367  * first length, then byte-for-byte comparison. The reason for
368  * doing length comparison first is that we don't care about the
369  * ordering so long as it's consistent, and comparing lengths
370  * first gives us a chance to avoid a strncmp() call.
371  *
372  * This is different from what we do with scalar statistics --
373  * they get sorted on frequencies. The rationale is that we
374  * usually search through most common elements looking for a
375  * specific value, so we can grab its frequency. When values are
376  * presorted we can employ binary search for that. See
377  * ts_selfuncs.c for a real usage scenario.
378  */
379  qsort(sort_table, num_mcelem, sizeof(TrackItem *),
381 
382  /* Must copy the target values into anl_context */
383  old_context = MemoryContextSwitchTo(stats->anl_context);
384 
385  /*
386  * We sorted statistics on the lexeme value, but we want to be
387  * able to find out the minimal and maximal frequency without
388  * going through all the values. We keep those two extra
389  * frequencies in two extra cells in mcelem_freqs.
390  *
391  * (Note: the MCELEM statistics slot definition allows for a third
392  * extra number containing the frequency of nulls, but we don't
393  * create that for a tsvector column, since null elements aren't
394  * possible.)
395  */
396  mcelem_values = (Datum *) palloc(num_mcelem * sizeof(Datum));
397  mcelem_freqs = (float4 *) palloc((num_mcelem + 2) * sizeof(float4));
398 
399  /*
400  * See comments above about use of nonnull_cnt as the divisor for
401  * the final frequency estimates.
402  */
403  for (i = 0; i < num_mcelem; i++)
404  {
405  TrackItem *item = sort_table[i];
406 
407  mcelem_values[i] =
409  item->key.length));
410  mcelem_freqs[i] = (double) item->frequency / (double) nonnull_cnt;
411  }
412  mcelem_freqs[i++] = (double) minfreq / (double) nonnull_cnt;
413  mcelem_freqs[i] = (double) maxfreq / (double) nonnull_cnt;
414  MemoryContextSwitchTo(old_context);
415 
416  stats->stakind[0] = STATISTIC_KIND_MCELEM;
417  stats->staop[0] = TextEqualOperator;
418  stats->stanumbers[0] = mcelem_freqs;
419  /* See above comment about two extra frequency fields */
420  stats->numnumbers[0] = num_mcelem + 2;
421  stats->stavalues[0] = mcelem_values;
422  stats->numvalues[0] = num_mcelem;
423  /* We are storing text values */
424  stats->statypid[0] = TEXTOID;
425  stats->statyplen[0] = -1; /* typlen, -1 for varlena */
426  stats->statypbyval[0] = false;
427  stats->statypalign[0] = 'i';
428  }
429  }
430  else
431  {
432  /* We found only nulls; assume the column is entirely null */
433  stats->stats_valid = true;
434  stats->stanullfrac = 1.0;
435  stats->stawidth = 0; /* "unknown" */
436  stats->stadistinct = 0.0; /* "unknown" */
437  }
438 
439  /*
440  * We don't need to bother cleaning up any of our temporary palloc's. The
441  * hashtable should also go away, as it used a child memory context.
442  */
443 }
444 
445 /*
446  * A function to prune the D structure from the Lossy Counting algorithm.
447  * Consult compute_tsvector_stats() for wider explanation.
448  */
449 static void
450 prune_lexemes_hashtable(HTAB *lexemes_tab, int b_current)
451 {
452  HASH_SEQ_STATUS scan_status;
453  TrackItem *item;
454 
455  hash_seq_init(&scan_status, lexemes_tab);
456  while ((item = (TrackItem *) hash_seq_search(&scan_status)) != NULL)
457  {
458  if (item->frequency + item->delta <= b_current)
459  {
460  char *lexeme = item->key.lexeme;
461 
462  if (hash_search(lexemes_tab, (const void *) &item->key,
463  HASH_REMOVE, NULL) == NULL)
464  elog(ERROR, "hash table corrupted");
465  pfree(lexeme);
466  }
467  }
468 }
469 
470 /*
471  * Hash functions for lexemes. They are strings, but not NULL terminated,
472  * so we need a special hash function.
473  */
474 static uint32
475 lexeme_hash(const void *key, Size keysize)
476 {
477  const LexemeHashKey *l = (const LexemeHashKey *) key;
478 
479  return DatumGetUInt32(hash_any((const unsigned char *) l->lexeme,
480  l->length));
481 }
482 
483 /*
484  * Matching function for lexemes, to be used in hashtable lookups.
485  */
486 static int
487 lexeme_match(const void *key1, const void *key2, Size keysize)
488 {
489  /* The keysize parameter is superfluous, the keys store their lengths */
490  return lexeme_compare(key1, key2);
491 }
492 
493 /*
494  * Comparison function for lexemes.
495  */
496 static int
497 lexeme_compare(const void *key1, const void *key2)
498 {
499  const LexemeHashKey *d1 = (const LexemeHashKey *) key1;
500  const LexemeHashKey *d2 = (const LexemeHashKey *) key2;
501 
502  /* First, compare by length */
503  if (d1->length > d2->length)
504  return 1;
505  else if (d1->length < d2->length)
506  return -1;
507  /* Lengths are equal, do a byte-by-byte comparison */
508  return strncmp(d1->lexeme, d2->lexeme, d1->length);
509 }
510 
511 /*
512  * qsort() comparator for sorting TrackItems on frequencies (descending sort)
513  */
514 static int
515 trackitem_compare_frequencies_desc(const void *e1, const void *e2)
516 {
517  const TrackItem *const *t1 = (const TrackItem *const *) e1;
518  const TrackItem *const *t2 = (const TrackItem *const *) e2;
519 
520  return (*t2)->frequency - (*t1)->frequency;
521 }
522 
523 /*
524  * qsort() comparator for sorting TrackItems on lexemes
525  */
526 static int
527 trackitem_compare_lexemes(const void *e1, const void *e2)
528 {
529  const TrackItem *const *t1 = (const TrackItem *const *) e1;
530  const TrackItem *const *t2 = (const TrackItem *const *) e2;
531 
532  return lexeme_compare(&(*t1)->key, &(*t2)->key);
533 }
#define DatumGetUInt32(X)
Definition: postgres.h:492
int minrows
Definition: vacuum.h:92
#define HASH_CONTEXT
Definition: hsearch.h:93
#define HASH_ELEM
Definition: hsearch.h:87
MemoryContext hcxt
Definition: hsearch.h:78
#define DEBUG3
Definition: elog.h:23
#define TEXTOID
Definition: pg_type.h:324
#define PointerGetDatum(X)
Definition: postgres.h:562
#define Min(x, y)
Definition: c.h:802
static int lexeme_compare(const void *key1, const void *key2)
Datum * stavalues[STATISTIC_NUM_SLOTS]
Definition: vacuum.h:108
static MemoryContext MemoryContextSwitchTo(MemoryContext context)
Definition: palloc.h:109
#define TextEqualOperator
Definition: pg_operator.h:136
Size entrysize
Definition: hsearch.h:73
uint32 len
Definition: ts_type.h:44
#define MemSet(start, val, len)
Definition: c.h:853
long hash_get_num_entries(HTAB *hashp)
Definition: dynahash.c:1331
#define PG_GETARG_POINTER(n)
Definition: fmgr.h:241
bool statypbyval[STATISTIC_NUM_SLOTS]
Definition: vacuum.h:118
void * hash_search(HTAB *hashp, const void *keyPtr, HASHACTION action, bool *foundPtr)
Definition: dynahash.c:902
char statypalign[STATISTIC_NUM_SLOTS]
Definition: vacuum.h:119
Form_pg_attribute attr
Definition: vacuum.h:81
Definition: dynahash.c:208
void pfree(void *pointer)
Definition: mcxt.c:949
#define ERROR
Definition: elog.h:43
int32 stawidth
Definition: vacuum.h:101
static struct @121 value
text * cstring_to_text_with_len(const char *s, int len)
Definition: varlena.c:161
int numnumbers[STATISTIC_NUM_SLOTS]
Definition: vacuum.h:105
LexemeHashKey key
Definition: ts_typanalyze.c:33
int32 size
Definition: ts_type.h:93
static int lexeme_match(const void *key1, const void *key2, Size keysize)
static void compute_tsvector_stats(VacAttrStats *stats, AnalyzeAttrFetchFunc fetchfunc, int samplerows, double totalrows)
static void prune_lexemes_hashtable(HTAB *lexemes_tab, int b_current)
FormData_pg_attribute * Form_pg_attribute
Definition: pg_attribute.h:187
unsigned int uint32
Definition: c.h:296
#define TSVectorGetDatum(X)
Definition: ts_type.h:119
MemoryContext CurrentMemoryContext
Definition: mcxt.c:37
#define DatumGetTSVector(X)
Definition: ts_type.h:117
float4 stanullfrac
Definition: vacuum.h:100
Datum ts_typanalyze(PG_FUNCTION_ARGS)
Definition: ts_typanalyze.c:54
Oid staop[STATISTIC_NUM_SLOTS]
Definition: vacuum.h:104
bool stats_valid
Definition: vacuum.h:99
float float4
Definition: c.h:428
#define PG_RETURN_BOOL(x)
Definition: fmgr.h:319
HTAB * hash_create(const char *tabname, long nelem, HASHCTL *info, int flags)
Definition: dynahash.c:316
uintptr_t Datum
Definition: postgres.h:372
int16 stakind[STATISTIC_NUM_SLOTS]
Definition: vacuum.h:103
Size keysize
Definition: hsearch.h:72
HashCompareFunc match
Definition: hsearch.h:75
#define VARSIZE_ANY(PTR)
Definition: postgres.h:334
Oid statypid[STATISTIC_NUM_SLOTS]
Definition: vacuum.h:116
static uint32 lexeme_hash(const void *key, Size keysize)
#define Max(x, y)
Definition: c.h:796
#define Assert(condition)
Definition: c.h:670
static int trackitem_compare_frequencies_desc(const void *e1, const void *e2)
float4 * stanumbers[STATISTIC_NUM_SLOTS]
Definition: vacuum.h:106
Datum hash_any(register const unsigned char *k, register int keylen)
Definition: hashfunc.c:428
#define HASH_COMPARE
Definition: hsearch.h:90
Datum(* AnalyzeAttrFetchFunc)(VacAttrStatsP stats, int rownum, bool *isNull)
Definition: vacuum.h:61
size_t Size
Definition: c.h:404
void * hash_seq_search(HASH_SEQ_STATUS *status)
Definition: dynahash.c:1385
uint32 pos
Definition: ts_type.h:44
void hash_seq_init(HASH_SEQ_STATUS *status, HTAB *hashp)
Definition: dynahash.c:1375
MemoryContext anl_context
Definition: vacuum.h:85
static int trackitem_compare_lexemes(const void *e1, const void *e2)
#define DatumGetPointer(X)
Definition: postgres.h:555
static dshash_hash hash_key(dshash_table *hash_table, const void *key)
Definition: dshash.c:885
int numvalues[STATISTIC_NUM_SLOTS]
Definition: vacuum.h:107
void * palloc(Size size)
Definition: mcxt.c:848
#define STRPTR(x)
Definition: hstore.h:76
int16 statyplen[STATISTIC_NUM_SLOTS]
Definition: vacuum.h:117
#define STATISTIC_KIND_MCELEM
Definition: pg_statistic.h:257
int i
AnalyzeAttrComputeStatsFunc compute_stats
Definition: vacuum.h:91
#define PG_FUNCTION_ARGS
Definition: fmgr.h:158
#define ARRPTR(x)
Definition: cube.c:26
#define elog
Definition: elog.h:219
#define qsort(a, b, c, d)
Definition: port.h:408
void vacuum_delay_point(void)
Definition: vacuum.c:1658
int default_statistics_target
Definition: analyze.c:77
HashValueFunc hash
Definition: hsearch.h:74
#define HASH_FUNCTION
Definition: hsearch.h:89
float4 stadistinct
Definition: vacuum.h:102