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) 1996-2023, 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 "catalog/pg_collation.h"
17 #include "catalog/pg_operator.h"
18 #include "commands/vacuum.h"
19 #include "common/hashfn.h"
20 #include "tsearch/ts_type.h"
21 #include "utils/builtins.h"
22 #include "varatt.h"
23 
24 
25 /* A hash key for lexemes */
26 typedef struct
27 {
28  char *lexeme; /* lexeme (not NULL terminated!) */
29  int length; /* its length in bytes */
31 
32 /* A hash table entry for the Lossy Counting algorithm */
33 typedef struct
34 {
35  LexemeHashKey key; /* This is 'e' from the LC algorithm. */
36  int frequency; /* This is 'f'. */
37  int delta; /* And this is 'delta'. */
38 } TrackItem;
39 
40 static void compute_tsvector_stats(VacAttrStats *stats,
41  AnalyzeAttrFetchFunc fetchfunc,
42  int samplerows,
43  double totalrows);
44 static void prune_lexemes_hashtable(HTAB *lexemes_tab, int b_current);
45 static uint32 lexeme_hash(const void *key, Size keysize);
46 static int lexeme_match(const void *key1, const void *key2, Size keysize);
47 static int lexeme_compare(const void *key1, const void *key2);
48 static int trackitem_compare_frequencies_desc(const void *e1, const void *e2,
49  void *arg);
50 static int trackitem_compare_lexemes(const void *e1, const void *e2,
51  void *arg);
52 
53 
54 /*
55  * ts_typanalyze -- a custom typanalyze function for tsvector columns
56  */
57 Datum
59 {
61  Form_pg_attribute attr = stats->attr;
62 
63  /* If the attstattarget column is negative, use the default value */
64  /* NB: it is okay to scribble on stats->attr since it's a copy */
65  if (attr->attstattarget < 0)
66  attr->attstattarget = default_statistics_target;
67 
69  /* see comment about the choice of minrows in commands/analyze.c */
70  stats->minrows = 300 * attr->attstattarget;
71 
72  PG_RETURN_BOOL(true);
73 }
74 
75 /*
76  * compute_tsvector_stats() -- compute statistics for a tsvector column
77  *
78  * This functions computes statistics that are useful for determining @@
79  * operations' selectivity, along with the fraction of non-null rows and
80  * average width.
81  *
82  * Instead of finding the most common values, as we do for most datatypes,
83  * we're looking for the most common lexemes. This is more useful, because
84  * there most probably won't be any two rows with the same tsvector and thus
85  * the notion of a MCV is a bit bogus with this datatype. With a list of the
86  * most common lexemes we can do a better job at figuring out @@ selectivity.
87  *
88  * For the same reasons we assume that tsvector columns are unique when
89  * determining the number of distinct values.
90  *
91  * The algorithm used is Lossy Counting, as proposed in the paper "Approximate
92  * frequency counts over data streams" by G. S. Manku and R. Motwani, in
93  * Proceedings of the 28th International Conference on Very Large Data Bases,
94  * Hong Kong, China, August 2002, section 4.2. The paper is available at
95  * http://www.vldb.org/conf/2002/S10P03.pdf
96  *
97  * The Lossy Counting (aka LC) algorithm goes like this:
98  * Let s be the threshold frequency for an item (the minimum frequency we
99  * are interested in) and epsilon the error margin for the frequency. Let D
100  * be a set of triples (e, f, delta), where e is an element value, f is that
101  * element's frequency (actually, its current occurrence count) and delta is
102  * the maximum error in f. We start with D empty and process the elements in
103  * batches of size w. (The batch size is also known as "bucket size" and is
104  * equal to 1/epsilon.) Let the current batch number be b_current, starting
105  * with 1. For each element e we either increment its f count, if it's
106  * already in D, or insert a new triple into D with values (e, 1, b_current
107  * - 1). After processing each batch we prune D, by removing from it all
108  * elements with f + delta <= b_current. After the algorithm finishes we
109  * suppress all elements from D that do not satisfy f >= (s - epsilon) * N,
110  * where N is the total number of elements in the input. We emit the
111  * remaining elements with estimated frequency f/N. The LC paper proves
112  * that this algorithm finds all elements with true frequency at least s,
113  * and that no frequency is overestimated or is underestimated by more than
114  * epsilon. Furthermore, given reasonable assumptions about the input
115  * distribution, the required table size is no more than about 7 times w.
116  *
117  * We set s to be the estimated frequency of the K'th word in a natural
118  * language's frequency table, where K is the target number of entries in
119  * the MCELEM array plus an arbitrary constant, meant to reflect the fact
120  * that the most common words in any language would usually be stopwords
121  * so we will not actually see them in the input. We assume that the
122  * distribution of word frequencies (including the stopwords) follows Zipf's
123  * law with an exponent of 1.
124  *
125  * Assuming Zipfian distribution, the frequency of the K'th word is equal
126  * to 1/(K * H(W)) where H(n) is 1/2 + 1/3 + ... + 1/n and W is the number of
127  * words in the language. Putting W as one million, we get roughly 0.07/K.
128  * Assuming top 10 words are stopwords gives s = 0.07/(K + 10). We set
129  * epsilon = s/10, which gives bucket width w = (K + 10)/0.007 and
130  * maximum expected hashtable size of about 1000 * (K + 10).
131  *
132  * Note: in the above discussion, s, epsilon, and f/N are in terms of a
133  * lexeme's frequency as a fraction of all lexemes seen in the input.
134  * However, what we actually want to store in the finished pg_statistic
135  * entry is each lexeme's frequency as a fraction of all rows that it occurs
136  * in. Assuming that the input tsvectors are correctly constructed, no
137  * lexeme occurs more than once per tsvector, so the final count f is a
138  * correct estimate of the number of input tsvectors it occurs in, and we
139  * need only change the divisor from N to nonnull_cnt to get the number we
140  * want.
141  */
142 static void
144  AnalyzeAttrFetchFunc fetchfunc,
145  int samplerows,
146  double totalrows)
147 {
148  int num_mcelem;
149  int null_cnt = 0;
150  double total_width = 0;
151 
152  /* This is D from the LC algorithm. */
153  HTAB *lexemes_tab;
154  HASHCTL hash_ctl;
155  HASH_SEQ_STATUS scan_status;
156 
157  /* This is the current bucket number from the LC algorithm */
158  int b_current;
159 
160  /* This is 'w' from the LC algorithm */
161  int bucket_width;
162  int vector_no,
163  lexeme_no;
165 
166  /*
167  * We want statistics_target * 10 lexemes in the MCELEM array. This
168  * multiplier is pretty arbitrary, but is meant to reflect the fact that
169  * the number of individual lexeme values tracked in pg_statistic ought to
170  * be more than the number of values for a simple scalar column.
171  */
172  num_mcelem = stats->attr->attstattarget * 10;
173 
174  /*
175  * We set bucket width equal to (num_mcelem + 10) / 0.007 as per the
176  * comment above.
177  */
178  bucket_width = (num_mcelem + 10) * 1000 / 7;
179 
180  /*
181  * Create the hashtable. It will be in local memory, so we don't need to
182  * worry about overflowing the initial size. Also we don't need to pay any
183  * attention to locking and memory management.
184  */
185  hash_ctl.keysize = sizeof(LexemeHashKey);
186  hash_ctl.entrysize = sizeof(TrackItem);
187  hash_ctl.hash = lexeme_hash;
188  hash_ctl.match = lexeme_match;
189  hash_ctl.hcxt = CurrentMemoryContext;
190  lexemes_tab = hash_create("Analyzed lexemes table",
191  num_mcelem,
192  &hash_ctl,
194 
195  /* Initialize counters. */
196  b_current = 1;
197  lexeme_no = 0;
198 
199  /* Loop over the tsvectors. */
200  for (vector_no = 0; vector_no < samplerows; vector_no++)
201  {
202  Datum value;
203  bool isnull;
204  TSVector vector;
205  WordEntry *curentryptr;
206  char *lexemesptr;
207  int j;
208 
210 
211  value = fetchfunc(stats, vector_no, &isnull);
212 
213  /*
214  * Check for null/nonnull.
215  */
216  if (isnull)
217  {
218  null_cnt++;
219  continue;
220  }
221 
222  /*
223  * Add up widths for average-width calculation. Since it's a
224  * tsvector, we know it's varlena. As in the regular
225  * compute_minimal_stats function, we use the toasted width for this
226  * calculation.
227  */
228  total_width += VARSIZE_ANY(DatumGetPointer(value));
229 
230  /*
231  * Now detoast the tsvector if needed.
232  */
233  vector = DatumGetTSVector(value);
234 
235  /*
236  * We loop through the lexemes in the tsvector and add them to our
237  * tracking hashtable.
238  */
239  lexemesptr = STRPTR(vector);
240  curentryptr = ARRPTR(vector);
241  for (j = 0; j < vector->size; j++)
242  {
243  TrackItem *item;
244  bool found;
245 
246  /*
247  * Construct a hash key. The key points into the (detoasted)
248  * tsvector value at this point, but if a new entry is created, we
249  * make a copy of it. This way we can free the tsvector value
250  * once we've processed all its lexemes.
251  */
252  hash_key.lexeme = lexemesptr + curentryptr->pos;
253  hash_key.length = curentryptr->len;
254 
255  /* Lookup current lexeme in hashtable, adding it if new */
256  item = (TrackItem *) hash_search(lexemes_tab,
257  &hash_key,
258  HASH_ENTER, &found);
259 
260  if (found)
261  {
262  /* The lexeme is already on the tracking list */
263  item->frequency++;
264  }
265  else
266  {
267  /* Initialize new tracking list element */
268  item->frequency = 1;
269  item->delta = b_current - 1;
270 
271  item->key.lexeme = palloc(hash_key.length);
272  memcpy(item->key.lexeme, hash_key.lexeme, hash_key.length);
273  }
274 
275  /* lexeme_no is the number of elements processed (ie N) */
276  lexeme_no++;
277 
278  /* We prune the D structure after processing each bucket */
279  if (lexeme_no % bucket_width == 0)
280  {
281  prune_lexemes_hashtable(lexemes_tab, b_current);
282  b_current++;
283  }
284 
285  /* Advance to the next WordEntry in the tsvector */
286  curentryptr++;
287  }
288 
289  /* If the vector was toasted, free the detoasted copy. */
290  if (TSVectorGetDatum(vector) != value)
291  pfree(vector);
292  }
293 
294  /* We can only compute real stats if we found some non-null values. */
295  if (null_cnt < samplerows)
296  {
297  int nonnull_cnt = samplerows - null_cnt;
298  int i;
299  TrackItem **sort_table;
300  TrackItem *item;
301  int track_len;
302  int cutoff_freq;
303  int minfreq,
304  maxfreq;
305 
306  stats->stats_valid = true;
307  /* Do the simple null-frac and average width stats */
308  stats->stanullfrac = (double) null_cnt / (double) samplerows;
309  stats->stawidth = total_width / (double) nonnull_cnt;
310 
311  /* Assume it's a unique column (see notes above) */
312  stats->stadistinct = -1.0 * (1.0 - stats->stanullfrac);
313 
314  /*
315  * Construct an array of the interesting hashtable items, that is,
316  * those meeting the cutoff frequency (s - epsilon)*N. Also identify
317  * the minimum and maximum frequencies among these items.
318  *
319  * Since epsilon = s/10 and bucket_width = 1/epsilon, the cutoff
320  * frequency is 9*N / bucket_width.
321  */
322  cutoff_freq = 9 * lexeme_no / bucket_width;
323 
324  i = hash_get_num_entries(lexemes_tab); /* surely enough space */
325  sort_table = (TrackItem **) palloc(sizeof(TrackItem *) * i);
326 
327  hash_seq_init(&scan_status, lexemes_tab);
328  track_len = 0;
329  minfreq = lexeme_no;
330  maxfreq = 0;
331  while ((item = (TrackItem *) hash_seq_search(&scan_status)) != NULL)
332  {
333  if (item->frequency > cutoff_freq)
334  {
335  sort_table[track_len++] = item;
336  minfreq = Min(minfreq, item->frequency);
337  maxfreq = Max(maxfreq, item->frequency);
338  }
339  }
340  Assert(track_len <= i);
341 
342  /* emit some statistics for debug purposes */
343  elog(DEBUG3, "tsvector_stats: target # mces = %d, bucket width = %d, "
344  "# lexemes = %d, hashtable size = %d, usable entries = %d",
345  num_mcelem, bucket_width, lexeme_no, i, track_len);
346 
347  /*
348  * If we obtained more lexemes than we really want, get rid of those
349  * with least frequencies. The easiest way is to qsort the array into
350  * descending frequency order and truncate the array.
351  */
352  if (num_mcelem < track_len)
353  {
354  qsort_interruptible(sort_table, track_len, sizeof(TrackItem *),
356  /* reset minfreq to the smallest frequency we're keeping */
357  minfreq = sort_table[num_mcelem - 1]->frequency;
358  }
359  else
360  num_mcelem = track_len;
361 
362  /* Generate MCELEM slot entry */
363  if (num_mcelem > 0)
364  {
365  MemoryContext old_context;
366  Datum *mcelem_values;
367  float4 *mcelem_freqs;
368 
369  /*
370  * We want to store statistics sorted on the lexeme value using
371  * first length, then byte-for-byte comparison. The reason for
372  * doing length comparison first is that we don't care about the
373  * ordering so long as it's consistent, and comparing lengths
374  * first gives us a chance to avoid a strncmp() call.
375  *
376  * This is different from what we do with scalar statistics --
377  * they get sorted on frequencies. The rationale is that we
378  * usually search through most common elements looking for a
379  * specific value, so we can grab its frequency. When values are
380  * presorted we can employ binary search for that. See
381  * ts_selfuncs.c for a real usage scenario.
382  */
383  qsort_interruptible(sort_table, num_mcelem, sizeof(TrackItem *),
385 
386  /* Must copy the target values into anl_context */
387  old_context = MemoryContextSwitchTo(stats->anl_context);
388 
389  /*
390  * We sorted statistics on the lexeme value, but we want to be
391  * able to find out the minimal and maximal frequency without
392  * going through all the values. We keep those two extra
393  * frequencies in two extra cells in mcelem_freqs.
394  *
395  * (Note: the MCELEM statistics slot definition allows for a third
396  * extra number containing the frequency of nulls, but we don't
397  * create that for a tsvector column, since null elements aren't
398  * possible.)
399  */
400  mcelem_values = (Datum *) palloc(num_mcelem * sizeof(Datum));
401  mcelem_freqs = (float4 *) palloc((num_mcelem + 2) * sizeof(float4));
402 
403  /*
404  * See comments above about use of nonnull_cnt as the divisor for
405  * the final frequency estimates.
406  */
407  for (i = 0; i < num_mcelem; i++)
408  {
409  TrackItem *titem = sort_table[i];
410 
411  mcelem_values[i] =
413  titem->key.length));
414  mcelem_freqs[i] = (double) titem->frequency / (double) nonnull_cnt;
415  }
416  mcelem_freqs[i++] = (double) minfreq / (double) nonnull_cnt;
417  mcelem_freqs[i] = (double) maxfreq / (double) nonnull_cnt;
418  MemoryContextSwitchTo(old_context);
419 
420  stats->stakind[0] = STATISTIC_KIND_MCELEM;
421  stats->staop[0] = TextEqualOperator;
422  stats->stacoll[0] = DEFAULT_COLLATION_OID;
423  stats->stanumbers[0] = mcelem_freqs;
424  /* See above comment about two extra frequency fields */
425  stats->numnumbers[0] = num_mcelem + 2;
426  stats->stavalues[0] = mcelem_values;
427  stats->numvalues[0] = num_mcelem;
428  /* We are storing text values */
429  stats->statypid[0] = TEXTOID;
430  stats->statyplen[0] = -1; /* typlen, -1 for varlena */
431  stats->statypbyval[0] = false;
432  stats->statypalign[0] = 'i';
433  }
434  }
435  else
436  {
437  /* We found only nulls; assume the column is entirely null */
438  stats->stats_valid = true;
439  stats->stanullfrac = 1.0;
440  stats->stawidth = 0; /* "unknown" */
441  stats->stadistinct = 0.0; /* "unknown" */
442  }
443 
444  /*
445  * We don't need to bother cleaning up any of our temporary palloc's. The
446  * hashtable should also go away, as it used a child memory context.
447  */
448 }
449 
450 /*
451  * A function to prune the D structure from the Lossy Counting algorithm.
452  * Consult compute_tsvector_stats() for wider explanation.
453  */
454 static void
455 prune_lexemes_hashtable(HTAB *lexemes_tab, int b_current)
456 {
457  HASH_SEQ_STATUS scan_status;
458  TrackItem *item;
459 
460  hash_seq_init(&scan_status, lexemes_tab);
461  while ((item = (TrackItem *) hash_seq_search(&scan_status)) != NULL)
462  {
463  if (item->frequency + item->delta <= b_current)
464  {
465  char *lexeme = item->key.lexeme;
466 
467  if (hash_search(lexemes_tab, &item->key,
468  HASH_REMOVE, NULL) == NULL)
469  elog(ERROR, "hash table corrupted");
470  pfree(lexeme);
471  }
472  }
473 }
474 
475 /*
476  * Hash functions for lexemes. They are strings, but not NULL terminated,
477  * so we need a special hash function.
478  */
479 static uint32
480 lexeme_hash(const void *key, Size keysize)
481 {
482  const LexemeHashKey *l = (const LexemeHashKey *) key;
483 
484  return DatumGetUInt32(hash_any((const unsigned char *) l->lexeme,
485  l->length));
486 }
487 
488 /*
489  * Matching function for lexemes, to be used in hashtable lookups.
490  */
491 static int
492 lexeme_match(const void *key1, const void *key2, Size keysize)
493 {
494  /* The keysize parameter is superfluous, the keys store their lengths */
495  return lexeme_compare(key1, key2);
496 }
497 
498 /*
499  * Comparison function for lexemes.
500  */
501 static int
502 lexeme_compare(const void *key1, const void *key2)
503 {
504  const LexemeHashKey *d1 = (const LexemeHashKey *) key1;
505  const LexemeHashKey *d2 = (const LexemeHashKey *) key2;
506 
507  /* First, compare by length */
508  if (d1->length > d2->length)
509  return 1;
510  else if (d1->length < d2->length)
511  return -1;
512  /* Lengths are equal, do a byte-by-byte comparison */
513  return strncmp(d1->lexeme, d2->lexeme, d1->length);
514 }
515 
516 /*
517  * Comparator for sorting TrackItems on frequencies (descending sort)
518  */
519 static int
520 trackitem_compare_frequencies_desc(const void *e1, const void *e2, void *arg)
521 {
522  const TrackItem *const *t1 = (const TrackItem *const *) e1;
523  const TrackItem *const *t2 = (const TrackItem *const *) e2;
524 
525  return (*t2)->frequency - (*t1)->frequency;
526 }
527 
528 /*
529  * Comparator for sorting TrackItems on lexemes
530  */
531 static int
532 trackitem_compare_lexemes(const void *e1, const void *e2, void *arg)
533 {
534  const TrackItem *const *t1 = (const TrackItem *const *) e1;
535  const TrackItem *const *t2 = (const TrackItem *const *) e2;
536 
537  return lexeme_compare(&(*t1)->key, &(*t2)->key);
538 }
unsigned int uint32
Definition: c.h:490
#define Min(x, y)
Definition: c.h:988
#define Max(x, y)
Definition: c.h:982
float float4
Definition: c.h:613
size_t Size
Definition: c.h:589
int default_statistics_target
Definition: analyze.c:83
#define ARRPTR(x)
Definition: cube.c:25
static dshash_hash hash_key(dshash_table *hash_table, const void *key)
Definition: dshash.c:1018
void * hash_search(HTAB *hashp, const void *keyPtr, HASHACTION action, bool *foundPtr)
Definition: dynahash.c:953
HTAB * hash_create(const char *tabname, long nelem, const HASHCTL *info, int flags)
Definition: dynahash.c:350
long hash_get_num_entries(HTAB *hashp)
Definition: dynahash.c:1377
void * hash_seq_search(HASH_SEQ_STATUS *status)
Definition: dynahash.c:1431
void hash_seq_init(HASH_SEQ_STATUS *status, HTAB *hashp)
Definition: dynahash.c:1421
#define DEBUG3
Definition: elog.h:28
#define ERROR
Definition: elog.h:39
#define PG_GETARG_POINTER(n)
Definition: fmgr.h:276
#define PG_FUNCTION_ARGS
Definition: fmgr.h:193
#define PG_RETURN_BOOL(x)
Definition: fmgr.h:359
static Datum hash_any(const unsigned char *k, int keylen)
Definition: hashfn.h:31
@ HASH_REMOVE
Definition: hsearch.h:115
@ HASH_ENTER
Definition: hsearch.h:114
#define HASH_CONTEXT
Definition: hsearch.h:102
#define HASH_ELEM
Definition: hsearch.h:95
#define HASH_COMPARE
Definition: hsearch.h:99
#define HASH_FUNCTION
Definition: hsearch.h:98
#define STRPTR(x)
Definition: hstore.h:76
static struct @147 value
int j
Definition: isn.c:74
int i
Definition: isn.c:73
Assert(fmt[strlen(fmt) - 1] !='\n')
void pfree(void *pointer)
Definition: mcxt.c:1456
MemoryContext CurrentMemoryContext
Definition: mcxt.c:135
void * palloc(Size size)
Definition: mcxt.c:1226
static MemoryContext MemoryContextSwitchTo(MemoryContext context)
Definition: palloc.h:138
FormData_pg_attribute * Form_pg_attribute
Definition: pg_attribute.h:209
void * arg
void qsort_interruptible(void *base, size_t nel, size_t elsize, qsort_arg_comparator cmp, void *arg)
static uint32 DatumGetUInt32(Datum X)
Definition: postgres.h:222
static Datum PointerGetDatum(const void *X)
Definition: postgres.h:322
uintptr_t Datum
Definition: postgres.h:64
static Pointer DatumGetPointer(Datum X)
Definition: postgres.h:312
Size keysize
Definition: hsearch.h:75
HashValueFunc hash
Definition: hsearch.h:78
Size entrysize
Definition: hsearch.h:76
HashCompareFunc match
Definition: hsearch.h:80
MemoryContext hcxt
Definition: hsearch.h:86
Definition: dynahash.c:220
int32 size
Definition: ts_type.h:93
LexemeHashKey key
Definition: ts_typanalyze.c:35
bool stats_valid
Definition: vacuum.h:147
float4 stanullfrac
Definition: vacuum.h:148
int16 stakind[STATISTIC_NUM_SLOTS]
Definition: vacuum.h:151
MemoryContext anl_context
Definition: vacuum.h:133
Oid statypid[STATISTIC_NUM_SLOTS]
Definition: vacuum.h:165
Oid staop[STATISTIC_NUM_SLOTS]
Definition: vacuum.h:152
Oid stacoll[STATISTIC_NUM_SLOTS]
Definition: vacuum.h:153
char statypalign[STATISTIC_NUM_SLOTS]
Definition: vacuum.h:168
float4 * stanumbers[STATISTIC_NUM_SLOTS]
Definition: vacuum.h:155
int minrows
Definition: vacuum.h:140
Form_pg_attribute attr
Definition: vacuum.h:128
int32 stawidth
Definition: vacuum.h:149
bool statypbyval[STATISTIC_NUM_SLOTS]
Definition: vacuum.h:167
int16 statyplen[STATISTIC_NUM_SLOTS]
Definition: vacuum.h:166
int numvalues[STATISTIC_NUM_SLOTS]
Definition: vacuum.h:156
Datum * stavalues[STATISTIC_NUM_SLOTS]
Definition: vacuum.h:157
float4 stadistinct
Definition: vacuum.h:150
int numnumbers[STATISTIC_NUM_SLOTS]
Definition: vacuum.h:154
AnalyzeAttrComputeStatsFunc compute_stats
Definition: vacuum.h:139
uint32 pos
Definition: ts_type.h:46
uint32 len
Definition: ts_type.h:45
static int trackitem_compare_frequencies_desc(const void *e1, const void *e2, void *arg)
static void prune_lexemes_hashtable(HTAB *lexemes_tab, int b_current)
static int trackitem_compare_lexemes(const void *e1, const void *e2, void *arg)
static int lexeme_compare(const void *key1, const void *key2)
static int lexeme_match(const void *key1, const void *key2, Size keysize)
Datum ts_typanalyze(PG_FUNCTION_ARGS)
Definition: ts_typanalyze.c:58
static void compute_tsvector_stats(VacAttrStats *stats, AnalyzeAttrFetchFunc fetchfunc, int samplerows, double totalrows)
static uint32 lexeme_hash(const void *key, Size keysize)
static TSVector DatumGetTSVector(Datum X)
Definition: ts_type.h:118
static Datum TSVectorGetDatum(const TSVectorData *X)
Definition: ts_type.h:130
void vacuum_delay_point(void)
Definition: vacuum.c:2326
Datum(* AnalyzeAttrFetchFunc)(VacAttrStatsP stats, int rownum, bool *isNull)
Definition: vacuum.h:107
#define VARSIZE_ANY(PTR)
Definition: varatt.h:311
text * cstring_to_text_with_len(const char *s, int len)
Definition: varlena.c:194