PostgreSQL Source Code  git master
 All Data Structures Namespaces Files Functions Variables Typedefs Enumerations Enumerator Macros
ts_selfuncs.c File Reference
#include "postgres.h"
#include "access/htup_details.h"
#include "catalog/pg_statistic.h"
#include "catalog/pg_type.h"
#include "miscadmin.h"
#include "nodes/nodes.h"
#include "tsearch/ts_type.h"
#include "utils/builtins.h"
#include "utils/lsyscache.h"
#include "utils/selfuncs.h"
#include "utils/syscache.h"
Include dependency graph for ts_selfuncs.c:

Go to the source code of this file.

Data Structures

struct  TextFreq
 
struct  LexemeKey
 

Macros

#define DEFAULT_TS_MATCH_SEL   0.005
 
#define tsquery_opr_selec_no_stats(query)   tsquery_opr_selec(GETQUERY(query), GETOPERAND(query), NULL, 0, 0)
 

Functions

static Selectivity tsquerysel (VariableStatData *vardata, Datum constval)
 
static Selectivity mcelem_tsquery_selec (TSQuery query, Datum *mcelem, int nmcelem, float4 *numbers, int nnumbers)
 
static Selectivity tsquery_opr_selec (QueryItem *item, char *operand, TextFreq *lookup, int length, float4 minfreq)
 
static int compare_lexeme_textfreq (const void *e1, const void *e2)
 
Datum tsmatchsel (PG_FUNCTION_ARGS)
 
Datum tsmatchjoinsel (PG_FUNCTION_ARGS)
 

Macro Definition Documentation

#define DEFAULT_TS_MATCH_SEL   0.005

Definition at line 33 of file ts_selfuncs.c.

Referenced by tsmatchjoinsel(), tsmatchsel(), and tsquery_opr_selec().

#define tsquery_opr_selec_no_stats (   query)    tsquery_opr_selec(GETQUERY(query), GETOPERAND(query), NULL, 0, 0)

Definition at line 57 of file ts_selfuncs.c.

Referenced by mcelem_tsquery_selec(), and tsquerysel().

Function Documentation

static int compare_lexeme_textfreq ( const void *  e1,
const void *  e2 
)
static

Definition at line 435 of file ts_selfuncs.c.

References TextFreq::element, LexemeKey::length, LexemeKey::lexeme, VARDATA_ANY, and VARSIZE_ANY_EXHDR.

Referenced by tsquery_opr_selec().

436 {
437  const LexemeKey *key = (const LexemeKey *) e1;
438  const TextFreq *t = (const TextFreq *) e2;
439  int len1,
440  len2;
441 
442  len1 = key->length;
443  len2 = VARSIZE_ANY_EXHDR(t->element);
444 
445  /* Compare lengths first, possibly avoiding a strncmp call */
446  if (len1 > len2)
447  return 1;
448  else if (len1 < len2)
449  return -1;
450 
451  /* Fall back on byte-for-byte comparison */
452  return strncmp(key->lexeme, VARDATA_ANY(t->element), len1);
453 }
#define VARDATA_ANY(PTR)
Definition: postgres.h:347
char * lexeme
Definition: ts_selfuncs.c:45
int length
Definition: ts_selfuncs.c:46
text * element
Definition: ts_selfuncs.c:38
#define VARSIZE_ANY_EXHDR(PTR)
Definition: postgres.h:340
static Selectivity mcelem_tsquery_selec ( TSQuery  query,
Datum mcelem,
int  nmcelem,
float4 numbers,
int  nnumbers 
)
static

Definition at line 208 of file ts_selfuncs.c.

References Assert, DatumGetPointer, TextFreq::element, TextFreq::frequency, GETOPERAND, GETQUERY, i, palloc(), pfree(), tsquery_opr_selec(), tsquery_opr_selec_no_stats, VARATT_IS_COMPRESSED, and VARATT_IS_EXTERNAL.

Referenced by tsquerysel().

210 {
211  float4 minfreq;
212  TextFreq *lookup;
213  Selectivity selec;
214  int i;
215 
216  /*
217  * There should be two more Numbers than Values, because the last two
218  * cells are taken for minimal and maximal frequency. Punt if not.
219  *
220  * (Note: the MCELEM statistics slot definition allows for a third extra
221  * number containing the frequency of nulls, but we're not expecting that
222  * to appear for a tsvector column.)
223  */
224  if (nnumbers != nmcelem + 2)
225  return tsquery_opr_selec_no_stats(query);
226 
227  /*
228  * Transpose the data into a single array so we can use bsearch().
229  */
230  lookup = (TextFreq *) palloc(sizeof(TextFreq) * nmcelem);
231  for (i = 0; i < nmcelem; i++)
232  {
233  /*
234  * The text Datums came from an array, so it cannot be compressed or
235  * stored out-of-line -- it's safe to use VARSIZE_ANY*.
236  */
237  Assert(!VARATT_IS_COMPRESSED(mcelem[i]) && !VARATT_IS_EXTERNAL(mcelem[i]));
238  lookup[i].element = (text *) DatumGetPointer(mcelem[i]);
239  lookup[i].frequency = numbers[i];
240  }
241 
242  /*
243  * Grab the lowest frequency. compute_tsvector_stats() stored it for us in
244  * the one before the last cell of the Numbers array. See ts_typanalyze.c
245  */
246  minfreq = numbers[nnumbers - 2];
247 
248  selec = tsquery_opr_selec(GETQUERY(query), GETOPERAND(query), lookup,
249  nmcelem, minfreq);
250 
251  pfree(lookup);
252 
253  return selec;
254 }
static Selectivity tsquery_opr_selec(QueryItem *item, char *operand, TextFreq *lookup, int length, float4 minfreq)
Definition: ts_selfuncs.c:279
#define VARATT_IS_COMPRESSED(PTR)
Definition: postgres.h:313
double Selectivity
Definition: nodes.h:640
#define GETQUERY(x)
Definition: _int.h:142
#define GETOPERAND(x)
Definition: ltree.h:118
#define VARATT_IS_EXTERNAL(PTR)
Definition: postgres.h:314
void pfree(void *pointer)
Definition: mcxt.c:949
float4 frequency
Definition: ts_selfuncs.c:39
Definition: zic.c:289
float float4
Definition: c.h:374
text * element
Definition: ts_selfuncs.c:38
#define Assert(condition)
Definition: c.h:681
#define DatumGetPointer(X)
Definition: postgres.h:555
void * palloc(Size size)
Definition: mcxt.c:848
int i
Definition: c.h:433
#define tsquery_opr_selec_no_stats(query)
Definition: ts_selfuncs.c:57
Datum tsmatchjoinsel ( PG_FUNCTION_ARGS  )

Definition at line 140 of file ts_selfuncs.c.

References DEFAULT_TS_MATCH_SEL, and PG_RETURN_FLOAT8.

141 {
142  /* for the moment we just punt */
144 }
#define PG_RETURN_FLOAT8(x)
Definition: fmgr.h:326
#define DEFAULT_TS_MATCH_SEL
Definition: ts_selfuncs.c:33
Datum tsmatchsel ( PG_FUNCTION_ARGS  )

Definition at line 68 of file ts_selfuncs.c.

References generate_unaccent_rules::args, Assert, CLAMP_PROBABILITY, DEFAULT_TS_MATCH_SEL, get_restriction_variable(), IsA, PG_GETARG_INT32, PG_GETARG_OID, PG_GETARG_POINTER, PG_RETURN_FLOAT8, ReleaseVariableStats, TSQUERYOID, tsquerysel(), TSVECTOROID, and VariableStatData::vartype.

69 {
71 
72 #ifdef NOT_USED
73  Oid operator = PG_GETARG_OID(1);
74 #endif
75  List *args = (List *) PG_GETARG_POINTER(2);
76  int varRelid = PG_GETARG_INT32(3);
77  VariableStatData vardata;
78  Node *other;
79  bool varonleft;
80  Selectivity selec;
81 
82  /*
83  * If expression is not variable = something or something = variable, then
84  * punt and return a default estimate.
85  */
86  if (!get_restriction_variable(root, args, varRelid,
87  &vardata, &other, &varonleft))
89 
90  /*
91  * Can't do anything useful if the something is not a constant, either.
92  */
93  if (!IsA(other, Const))
94  {
95  ReleaseVariableStats(vardata);
97  }
98 
99  /*
100  * The "@@" operator is strict, so we can cope with NULL right away
101  */
102  if (((Const *) other)->constisnull)
103  {
104  ReleaseVariableStats(vardata);
105  PG_RETURN_FLOAT8(0.0);
106  }
107 
108  /*
109  * OK, there's a Var and a Const we're dealing with here. We need the
110  * Const to be a TSQuery, else we can't do anything useful. We have to
111  * check this because the Var might be the TSQuery not the TSVector.
112  */
113  if (((Const *) other)->consttype == TSQUERYOID)
114  {
115  /* tsvector @@ tsquery or the other way around */
116  Assert(vardata.vartype == TSVECTOROID);
117 
118  selec = tsquerysel(&vardata, ((Const *) other)->constvalue);
119  }
120  else
121  {
122  /* If we can't see the query structure, must punt */
123  selec = DEFAULT_TS_MATCH_SEL;
124  }
125 
126  ReleaseVariableStats(vardata);
127 
128  CLAMP_PROBABILITY(selec);
129 
130  PG_RETURN_FLOAT8((float8) selec);
131 }
#define PG_GETARG_INT32(n)
Definition: fmgr.h:234
#define TSQUERYOID
Definition: pg_type.h:621
#define IsA(nodeptr, _type_)
Definition: nodes.h:561
bool get_restriction_variable(PlannerInfo *root, List *args, int varRelid, VariableStatData *vardata, Node **other, bool *varonleft)
Definition: selfuncs.c:4541
#define PG_RETURN_FLOAT8(x)
Definition: fmgr.h:326
static Selectivity tsquerysel(VariableStatData *vardata, Datum constval)
Definition: ts_selfuncs.c:151
Definition: nodes.h:510
#define DEFAULT_TS_MATCH_SEL
Definition: ts_selfuncs.c:33
#define PG_GETARG_POINTER(n)
Definition: fmgr.h:241
double Selectivity
Definition: nodes.h:640
unsigned int Oid
Definition: postgres_ext.h:31
#define CLAMP_PROBABILITY(p)
Definition: selfuncs.h:57
double float8
Definition: c.h:375
#define PG_GETARG_OID(n)
Definition: fmgr.h:240
#define Assert(condition)
Definition: c.h:681
#define ReleaseVariableStats(vardata)
Definition: selfuncs.h:81
#define TSVECTOROID
Definition: pg_type.h:615
Definition: pg_list.h:45
static Selectivity tsquery_opr_selec ( QueryItem item,
char *  operand,
TextFreq lookup,
int  length,
float4  minfreq 
)
static

Definition at line 279 of file ts_selfuncs.c.

References check_stack_depth(), CLAMP_PROBABILITY, compare_lexeme_textfreq(), DEFAULT_TS_MATCH_SEL, QueryOperand::distance, TextFreq::element, elog, ERROR, TextFreq::frequency, i, QueryOperator::left, LexemeKey::length, QueryOperand::length, length(), LexemeKey::lexeme, Max, Min, OP_AND, OP_NOT, OP_OR, OP_PHRASE, QueryOperator::oper, oper(), QueryOperand::prefix, QI_VAL, QueryItem::qoperator, s1, s2, QueryItem::type, VARDATA_ANY, and VARSIZE_ANY_EXHDR.

Referenced by mcelem_tsquery_selec().

281 {
282  Selectivity selec;
283 
284  /* since this function recurses, it could be driven to stack overflow */
286 
287  if (item->type == QI_VAL)
288  {
289  QueryOperand *oper = (QueryOperand *) item;
290  LexemeKey key;
291 
292  /*
293  * Prepare the key for bsearch().
294  */
295  key.lexeme = operand + oper->distance;
296  key.length = oper->length;
297 
298  if (oper->prefix)
299  {
300  /* Prefix match, ie the query item is lexeme:* */
301  Selectivity matched,
302  allmces;
303  int i,
304  n_matched;
305 
306  /*
307  * Our strategy is to scan through the MCELEM list and combine the
308  * frequencies of the ones that match the prefix. We then
309  * extrapolate the fraction of matching MCELEMs to the remaining
310  * rows, assuming that the MCELEMs are representative of the whole
311  * lexeme population in this respect. (Compare
312  * histogram_selectivity().) Note that these are most common
313  * elements not most common values, so they're not mutually
314  * exclusive. We treat occurrences as independent events.
315  *
316  * This is only a good plan if we have a pretty fair number of
317  * MCELEMs available; we set the threshold at 100. If no stats or
318  * insufficient stats, arbitrarily use DEFAULT_TS_MATCH_SEL*4.
319  */
320  if (lookup == NULL || length < 100)
321  return (Selectivity) (DEFAULT_TS_MATCH_SEL * 4);
322 
323  matched = allmces = 0;
324  n_matched = 0;
325  for (i = 0; i < length; i++)
326  {
327  TextFreq *t = lookup + i;
328  int tlen = VARSIZE_ANY_EXHDR(t->element);
329 
330  if (tlen >= key.length &&
331  strncmp(key.lexeme, VARDATA_ANY(t->element),
332  key.length) == 0)
333  {
334  matched += t->frequency - matched * t->frequency;
335  n_matched++;
336  }
337  allmces += t->frequency - allmces * t->frequency;
338  }
339 
340  /* Clamp to ensure sanity in the face of roundoff error */
341  CLAMP_PROBABILITY(matched);
342  CLAMP_PROBABILITY(allmces);
343 
344  selec = matched + (1.0 - allmces) * ((double) n_matched / length);
345 
346  /*
347  * In any case, never believe that a prefix match has selectivity
348  * less than we would assign for a non-MCELEM lexeme. This
349  * preserves the property that "word:*" should be estimated to
350  * match at least as many rows as "word" would be.
351  */
352  selec = Max(Min(DEFAULT_TS_MATCH_SEL, minfreq / 2), selec);
353  }
354  else
355  {
356  /* Regular exact lexeme match */
357  TextFreq *searchres;
358 
359  /* If no stats for the variable, use DEFAULT_TS_MATCH_SEL */
360  if (lookup == NULL)
362 
363  searchres = (TextFreq *) bsearch(&key, lookup, length,
364  sizeof(TextFreq),
366 
367  if (searchres)
368  {
369  /*
370  * The element is in MCELEM. Return precise selectivity (or
371  * at least as precise as ANALYZE could find out).
372  */
373  selec = searchres->frequency;
374  }
375  else
376  {
377  /*
378  * The element is not in MCELEM. Punt, but assume that the
379  * selectivity cannot be more than minfreq / 2.
380  */
381  selec = Min(DEFAULT_TS_MATCH_SEL, minfreq / 2);
382  }
383  }
384  }
385  else
386  {
387  /* Current TSQuery node is an operator */
388  Selectivity s1,
389  s2;
390 
391  switch (item->qoperator.oper)
392  {
393  case OP_NOT:
394  selec = 1.0 - tsquery_opr_selec(item + 1, operand,
395  lookup, length, minfreq);
396  break;
397 
398  case OP_PHRASE:
399  case OP_AND:
400  s1 = tsquery_opr_selec(item + 1, operand,
401  lookup, length, minfreq);
402  s2 = tsquery_opr_selec(item + item->qoperator.left, operand,
403  lookup, length, minfreq);
404  selec = s1 * s2;
405  break;
406 
407  case OP_OR:
408  s1 = tsquery_opr_selec(item + 1, operand,
409  lookup, length, minfreq);
410  s2 = tsquery_opr_selec(item + item->qoperator.left, operand,
411  lookup, length, minfreq);
412  selec = s1 + s2 - s1 * s2;
413  break;
414 
415  default:
416  elog(ERROR, "unrecognized operator: %d", item->qoperator.oper);
417  selec = 0; /* keep compiler quiet */
418  break;
419  }
420  }
421 
422  /* Clamp intermediate results to stay sane despite roundoff error */
423  CLAMP_PROBABILITY(selec);
424 
425  return selec;
426 }
QueryOperator qoperator
Definition: ts_type.h:196
static Selectivity tsquery_opr_selec(QueryItem *item, char *operand, TextFreq *lookup, int length, float4 minfreq)
Definition: ts_selfuncs.c:279
int length(const List *list)
Definition: list.c:1271
#define VARDATA_ANY(PTR)
Definition: postgres.h:347
char * lexeme
Definition: ts_selfuncs.c:45
#define Min(x, y)
Definition: c.h:812
#define DEFAULT_TS_MATCH_SEL
Definition: ts_selfuncs.c:33
#define QI_VAL
Definition: ts_type.h:134
double Selectivity
Definition: nodes.h:640
uint32 distance
Definition: ts_type.h:158
#define OP_OR
Definition: ts_type.h:168
#define CLAMP_PROBABILITY(p)
Definition: selfuncs.h:57
static int compare_lexeme_textfreq(const void *e1, const void *e2)
Definition: ts_selfuncs.c:435
#define OP_AND
Definition: ts_type.h:167
#define ERROR
Definition: elog.h:43
char * s1
float4 frequency
Definition: ts_selfuncs.c:39
void check_stack_depth(void)
Definition: postgres.c:3150
int length
Definition: ts_selfuncs.c:46
QueryItemType type
Definition: ts_type.h:195
char * s2
text * element
Definition: ts_selfuncs.c:38
#define Max(x, y)
Definition: c.h:806
#define OP_PHRASE
Definition: ts_type.h:169
uint32 length
Definition: ts_type.h:158
uint32 left
Definition: ts_type.h:184
#define VARSIZE_ANY_EXHDR(PTR)
Definition: postgres.h:340
int i
bool prefix
Definition: ts_type.h:150
#define elog
Definition: elog.h:219
Operator oper(ParseState *pstate, List *opname, Oid ltypeId, Oid rtypeId, bool noError, int location)
Definition: parse_oper.c:377
#define OP_NOT
Definition: ts_type.h:166
static Selectivity tsquerysel ( VariableStatData vardata,
Datum  constval 
)
static

Definition at line 151 of file ts_selfuncs.c.

References ATTSTATSSLOT_NUMBERS, ATTSTATSSLOT_VALUES, DatumGetTSQuery, free_attstatsslot(), get_attstatsslot(), GETSTRUCT, HeapTupleIsValid, InvalidOid, mcelem_tsquery_selec(), AttStatsSlot::nnumbers, AttStatsSlot::numbers, AttStatsSlot::nvalues, TSQueryData::size, STATISTIC_KIND_MCELEM, VariableStatData::statsTuple, tsquery_opr_selec_no_stats, and AttStatsSlot::values.

Referenced by tsmatchsel().

152 {
153  Selectivity selec;
154  TSQuery query;
155 
156  /* The caller made sure the const is a TSQuery, so get it now */
157  query = DatumGetTSQuery(constval);
158 
159  /* Empty query matches nothing */
160  if (query->size == 0)
161  return (Selectivity) 0.0;
162 
163  if (HeapTupleIsValid(vardata->statsTuple))
164  {
165  Form_pg_statistic stats;
166  AttStatsSlot sslot;
167 
168  stats = (Form_pg_statistic) GETSTRUCT(vardata->statsTuple);
169 
170  /* MCELEM will be an array of TEXT elements for a tsvector column */
171  if (get_attstatsslot(&sslot, vardata->statsTuple,
174  {
175  /*
176  * There is a most-common-elements slot for the tsvector Var, so
177  * use that.
178  */
179  selec = mcelem_tsquery_selec(query, sslot.values, sslot.nvalues,
180  sslot.numbers, sslot.nnumbers);
181  free_attstatsslot(&sslot);
182  }
183  else
184  {
185  /* No most-common-elements info, so do without */
186  selec = tsquery_opr_selec_no_stats(query);
187  }
188 
189  /*
190  * MCE stats count only non-null rows, so adjust for null rows.
191  */
192  selec *= (1.0 - stats->stanullfrac);
193  }
194  else
195  {
196  /* No stats at all, so do without */
197  selec = tsquery_opr_selec_no_stats(query);
198  /* we assume no nulls here, so no stanullfrac correction */
199  }
200 
201  return selec;
202 }
#define DatumGetTSQuery(X)
Definition: ts_type.h:235
#define GETSTRUCT(TUP)
Definition: htup_details.h:656
#define ATTSTATSSLOT_VALUES
Definition: lsyscache.h:39
HeapTuple statsTuple
Definition: selfuncs.h:71
int nnumbers
Definition: lsyscache.h:53
double Selectivity
Definition: nodes.h:640
FormData_pg_statistic * Form_pg_statistic
Definition: pg_statistic.h:129
#define ATTSTATSSLOT_NUMBERS
Definition: lsyscache.h:40
float4 * numbers
Definition: lsyscache.h:52
static Selectivity mcelem_tsquery_selec(TSQuery query, Datum *mcelem, int nmcelem, float4 *numbers, int nnumbers)
Definition: ts_selfuncs.c:208
#define InvalidOid
Definition: postgres_ext.h:36
#define HeapTupleIsValid(tuple)
Definition: htup.h:77
bool get_attstatsslot(AttStatsSlot *sslot, HeapTuple statstuple, int reqkind, Oid reqop, int flags)
Definition: lsyscache.c:2895
Datum * values
Definition: lsyscache.h:49
int32 size
Definition: ts_type.h:208
#define STATISTIC_KIND_MCELEM
Definition: pg_statistic.h:257
#define tsquery_opr_selec_no_stats(query)
Definition: ts_selfuncs.c:57
void free_attstatsslot(AttStatsSlot *sslot)
Definition: lsyscache.c:3011