Line data Source code
1 : /*-------------------------------------------------------------------------
2 : *
3 : * tsvector_op.c
4 : * operations over tsvector
5 : *
6 : * Portions Copyright (c) 1996-2025, PostgreSQL Global Development Group
7 : *
8 : *
9 : * IDENTIFICATION
10 : * src/backend/utils/adt/tsvector_op.c
11 : *
12 : *-------------------------------------------------------------------------
13 : */
14 : #include "postgres.h"
15 :
16 : #include <limits.h>
17 :
18 : #include "access/htup_details.h"
19 : #include "catalog/namespace.h"
20 : #include "catalog/pg_type.h"
21 : #include "commands/trigger.h"
22 : #include "common/int.h"
23 : #include "executor/spi.h"
24 : #include "funcapi.h"
25 : #include "lib/qunique.h"
26 : #include "mb/pg_wchar.h"
27 : #include "miscadmin.h"
28 : #include "parser/parse_coerce.h"
29 : #include "tsearch/ts_utils.h"
30 : #include "utils/array.h"
31 : #include "utils/builtins.h"
32 : #include "utils/regproc.h"
33 : #include "utils/rel.h"
34 :
35 :
36 : typedef struct
37 : {
38 : WordEntry *arrb;
39 : WordEntry *arre;
40 : char *values;
41 : char *operand;
42 : } CHKVAL;
43 :
44 :
45 : typedef struct StatEntry
46 : {
47 : uint32 ndoc; /* zero indicates that we were already here
48 : * while walking through the tree */
49 : uint32 nentry;
50 : struct StatEntry *left;
51 : struct StatEntry *right;
52 : uint32 lenlexeme;
53 : char lexeme[FLEXIBLE_ARRAY_MEMBER];
54 : } StatEntry;
55 :
56 : #define STATENTRYHDRSZ (offsetof(StatEntry, lexeme))
57 :
58 : typedef struct
59 : {
60 : int32 weight;
61 :
62 : uint32 maxdepth;
63 :
64 : StatEntry **stack;
65 : uint32 stackpos;
66 :
67 : StatEntry *root;
68 : } TSVectorStat;
69 :
70 :
71 : static TSTernaryValue TS_execute_recurse(QueryItem *curitem, void *arg,
72 : uint32 flags,
73 : TSExecuteCallback chkcond);
74 : static bool TS_execute_locations_recurse(QueryItem *curitem,
75 : void *arg,
76 : TSExecuteCallback chkcond,
77 : List **locations);
78 : static int tsvector_bsearch(const TSVector tsv, char *lexeme, int lexeme_len);
79 : static Datum tsvector_update_trigger(PG_FUNCTION_ARGS, bool config_column);
80 :
81 :
82 : /*
83 : * Order: haspos, len, word, for all positions (pos, weight)
84 : */
85 : static int
86 2 : silly_cmp_tsvector(const TSVector a, const TSVector b)
87 : {
88 2 : if (VARSIZE(a) < VARSIZE(b))
89 0 : return -1;
90 2 : else if (VARSIZE(a) > VARSIZE(b))
91 0 : return 1;
92 2 : else if (a->size < b->size)
93 0 : return -1;
94 2 : else if (a->size > b->size)
95 0 : return 1;
96 : else
97 : {
98 2 : WordEntry *aptr = ARRPTR(a);
99 2 : WordEntry *bptr = ARRPTR(b);
100 2 : int i = 0;
101 : int res;
102 :
103 :
104 8 : for (i = 0; i < a->size; i++)
105 : {
106 6 : if (aptr->haspos != bptr->haspos)
107 : {
108 0 : return (aptr->haspos > bptr->haspos) ? -1 : 1;
109 : }
110 6 : else if ((res = tsCompareString(STRPTR(a) + aptr->pos, aptr->len, STRPTR(b) + bptr->pos, bptr->len, false)) != 0)
111 : {
112 0 : return res;
113 : }
114 6 : else if (aptr->haspos)
115 : {
116 0 : WordEntryPos *ap = POSDATAPTR(a, aptr);
117 0 : WordEntryPos *bp = POSDATAPTR(b, bptr);
118 : int j;
119 :
120 0 : if (POSDATALEN(a, aptr) != POSDATALEN(b, bptr))
121 0 : return (POSDATALEN(a, aptr) > POSDATALEN(b, bptr)) ? -1 : 1;
122 :
123 0 : for (j = 0; j < POSDATALEN(a, aptr); j++)
124 : {
125 0 : if (WEP_GETPOS(*ap) != WEP_GETPOS(*bp))
126 : {
127 0 : return (WEP_GETPOS(*ap) > WEP_GETPOS(*bp)) ? -1 : 1;
128 : }
129 0 : else if (WEP_GETWEIGHT(*ap) != WEP_GETWEIGHT(*bp))
130 : {
131 0 : return (WEP_GETWEIGHT(*ap) > WEP_GETWEIGHT(*bp)) ? -1 : 1;
132 : }
133 0 : ap++, bp++;
134 : }
135 : }
136 :
137 6 : aptr++;
138 6 : bptr++;
139 : }
140 : }
141 :
142 2 : return 0;
143 : }
144 :
145 : #define TSVECTORCMPFUNC( type, action, ret ) \
146 : Datum \
147 : tsvector_##type(PG_FUNCTION_ARGS) \
148 : { \
149 : TSVector a = PG_GETARG_TSVECTOR(0); \
150 : TSVector b = PG_GETARG_TSVECTOR(1); \
151 : int res = silly_cmp_tsvector(a, b); \
152 : PG_FREE_IF_COPY(a,0); \
153 : PG_FREE_IF_COPY(b,1); \
154 : PG_RETURN_##ret( res action 0 ); \
155 : } \
156 : /* keep compiler quiet - no extra ; */ \
157 : extern int no_such_variable
158 :
159 0 : TSVECTORCMPFUNC(lt, <, BOOL);
160 0 : TSVECTORCMPFUNC(le, <=, BOOL);
161 2 : TSVECTORCMPFUNC(eq, ==, BOOL);
162 0 : TSVECTORCMPFUNC(ge, >=, BOOL);
163 0 : TSVECTORCMPFUNC(gt, >, BOOL);
164 0 : TSVECTORCMPFUNC(ne, !=, BOOL);
165 0 : TSVECTORCMPFUNC(cmp, +, INT32);
166 :
167 : Datum
168 90 : tsvector_strip(PG_FUNCTION_ARGS)
169 : {
170 90 : TSVector in = PG_GETARG_TSVECTOR(0);
171 : TSVector out;
172 : int i,
173 90 : len = 0;
174 90 : WordEntry *arrin = ARRPTR(in),
175 : *arrout;
176 : char *cur;
177 :
178 318 : for (i = 0; i < in->size; i++)
179 228 : len += arrin[i].len;
180 :
181 90 : len = CALCDATASIZE(in->size, len);
182 90 : out = (TSVector) palloc0(len);
183 90 : SET_VARSIZE(out, len);
184 90 : out->size = in->size;
185 90 : arrout = ARRPTR(out);
186 90 : cur = STRPTR(out);
187 318 : for (i = 0; i < in->size; i++)
188 : {
189 228 : memcpy(cur, STRPTR(in) + arrin[i].pos, arrin[i].len);
190 228 : arrout[i].haspos = 0;
191 228 : arrout[i].len = arrin[i].len;
192 228 : arrout[i].pos = cur - STRPTR(out);
193 228 : cur += arrout[i].len;
194 : }
195 :
196 90 : PG_FREE_IF_COPY(in, 0);
197 90 : PG_RETURN_POINTER(out);
198 : }
199 :
200 : Datum
201 10 : tsvector_length(PG_FUNCTION_ARGS)
202 : {
203 10 : TSVector in = PG_GETARG_TSVECTOR(0);
204 10 : int32 ret = in->size;
205 :
206 10 : PG_FREE_IF_COPY(in, 0);
207 10 : PG_RETURN_INT32(ret);
208 : }
209 :
210 : Datum
211 12 : tsvector_setweight(PG_FUNCTION_ARGS)
212 : {
213 12 : TSVector in = PG_GETARG_TSVECTOR(0);
214 12 : char cw = PG_GETARG_CHAR(1);
215 : TSVector out;
216 : int i,
217 : j;
218 : WordEntry *entry;
219 : WordEntryPos *p;
220 12 : int w = 0;
221 :
222 12 : switch (cw)
223 : {
224 0 : case 'A':
225 : case 'a':
226 0 : w = 3;
227 0 : break;
228 0 : case 'B':
229 : case 'b':
230 0 : w = 2;
231 0 : break;
232 12 : case 'C':
233 : case 'c':
234 12 : w = 1;
235 12 : break;
236 0 : case 'D':
237 : case 'd':
238 0 : w = 0;
239 0 : break;
240 0 : default:
241 : /* internal error */
242 0 : elog(ERROR, "unrecognized weight: %d", cw);
243 : }
244 :
245 12 : out = (TSVector) palloc(VARSIZE(in));
246 12 : memcpy(out, in, VARSIZE(in));
247 12 : entry = ARRPTR(out);
248 12 : i = out->size;
249 60 : while (i--)
250 : {
251 48 : if ((j = POSDATALEN(out, entry)) != 0)
252 : {
253 48 : p = POSDATAPTR(out, entry);
254 168 : while (j--)
255 : {
256 120 : WEP_SETWEIGHT(*p, w);
257 120 : p++;
258 : }
259 : }
260 48 : entry++;
261 : }
262 :
263 12 : PG_FREE_IF_COPY(in, 0);
264 12 : PG_RETURN_POINTER(out);
265 : }
266 :
267 : /*
268 : * setweight(tsin tsvector, char_weight "char", lexemes "text"[])
269 : *
270 : * Assign weight w to elements of tsin that are listed in lexemes.
271 : */
272 : Datum
273 24 : tsvector_setweight_by_filter(PG_FUNCTION_ARGS)
274 : {
275 24 : TSVector tsin = PG_GETARG_TSVECTOR(0);
276 24 : char char_weight = PG_GETARG_CHAR(1);
277 24 : ArrayType *lexemes = PG_GETARG_ARRAYTYPE_P(2);
278 :
279 : TSVector tsout;
280 : int i,
281 : j,
282 : nlexemes,
283 : weight;
284 : WordEntry *entry;
285 : Datum *dlexemes;
286 : bool *nulls;
287 :
288 24 : switch (char_weight)
289 : {
290 0 : case 'A':
291 : case 'a':
292 0 : weight = 3;
293 0 : break;
294 0 : case 'B':
295 : case 'b':
296 0 : weight = 2;
297 0 : break;
298 24 : case 'C':
299 : case 'c':
300 24 : weight = 1;
301 24 : break;
302 0 : case 'D':
303 : case 'd':
304 0 : weight = 0;
305 0 : break;
306 0 : default:
307 : /* internal error */
308 0 : elog(ERROR, "unrecognized weight: %c", char_weight);
309 : }
310 :
311 24 : tsout = (TSVector) palloc(VARSIZE(tsin));
312 24 : memcpy(tsout, tsin, VARSIZE(tsin));
313 24 : entry = ARRPTR(tsout);
314 :
315 24 : deconstruct_array_builtin(lexemes, TEXTOID, &dlexemes, &nulls, &nlexemes);
316 :
317 : /*
318 : * Assuming that lexemes array is significantly shorter than tsvector we
319 : * can iterate through lexemes performing binary search of each lexeme
320 : * from lexemes in tsvector.
321 : */
322 72 : for (i = 0; i < nlexemes; i++)
323 : {
324 : char *lex;
325 : int lex_len,
326 : lex_pos;
327 :
328 : /* Ignore null array elements, they surely don't match */
329 48 : if (nulls[i])
330 6 : continue;
331 :
332 42 : lex = VARDATA(dlexemes[i]);
333 42 : lex_len = VARSIZE(dlexemes[i]) - VARHDRSZ;
334 42 : lex_pos = tsvector_bsearch(tsout, lex, lex_len);
335 :
336 42 : if (lex_pos >= 0 && (j = POSDATALEN(tsout, entry + lex_pos)) != 0)
337 : {
338 24 : WordEntryPos *p = POSDATAPTR(tsout, entry + lex_pos);
339 :
340 78 : while (j--)
341 : {
342 54 : WEP_SETWEIGHT(*p, weight);
343 54 : p++;
344 : }
345 : }
346 : }
347 :
348 24 : PG_FREE_IF_COPY(tsin, 0);
349 24 : PG_FREE_IF_COPY(lexemes, 2);
350 :
351 24 : PG_RETURN_POINTER(tsout);
352 : }
353 :
354 : #define compareEntry(pa, a, pb, b) \
355 : tsCompareString((pa) + (a)->pos, (a)->len, \
356 : (pb) + (b)->pos, (b)->len, \
357 : false)
358 :
359 : /*
360 : * Add positions from src to dest after offsetting them by maxpos.
361 : * Return the number added (might be less than expected due to overflow)
362 : */
363 : static int32
364 12 : add_pos(TSVector src, WordEntry *srcptr,
365 : TSVector dest, WordEntry *destptr,
366 : int32 maxpos)
367 : {
368 12 : uint16 *clen = &_POSVECPTR(dest, destptr)->npos;
369 : int i;
370 12 : uint16 slen = POSDATALEN(src, srcptr),
371 : startlen;
372 12 : WordEntryPos *spos = POSDATAPTR(src, srcptr),
373 12 : *dpos = POSDATAPTR(dest, destptr);
374 :
375 12 : if (!destptr->haspos)
376 0 : *clen = 0;
377 :
378 12 : startlen = *clen;
379 12 : for (i = 0;
380 24 : i < slen && *clen < MAXNUMPOS &&
381 12 : (*clen == 0 || WEP_GETPOS(dpos[*clen - 1]) != MAXENTRYPOS - 1);
382 12 : i++)
383 : {
384 12 : WEP_SETWEIGHT(dpos[*clen], WEP_GETWEIGHT(spos[i]));
385 12 : WEP_SETPOS(dpos[*clen], LIMITPOS(WEP_GETPOS(spos[i]) + maxpos));
386 12 : (*clen)++;
387 : }
388 :
389 12 : if (*clen != startlen)
390 12 : destptr->haspos = 1;
391 12 : return *clen - startlen;
392 : }
393 :
394 : /*
395 : * Perform binary search of given lexeme in TSVector.
396 : * Returns lexeme position in TSVector's entry array or -1 if lexeme wasn't
397 : * found.
398 : */
399 : static int
400 198 : tsvector_bsearch(const TSVector tsv, char *lexeme, int lexeme_len)
401 : {
402 198 : WordEntry *arrin = ARRPTR(tsv);
403 198 : int StopLow = 0,
404 198 : StopHigh = tsv->size,
405 : StopMiddle,
406 : cmp;
407 :
408 522 : while (StopLow < StopHigh)
409 : {
410 462 : StopMiddle = (StopLow + StopHigh) / 2;
411 :
412 462 : cmp = tsCompareString(lexeme, lexeme_len,
413 462 : STRPTR(tsv) + arrin[StopMiddle].pos,
414 462 : arrin[StopMiddle].len,
415 : false);
416 :
417 462 : if (cmp < 0)
418 216 : StopHigh = StopMiddle;
419 246 : else if (cmp > 0)
420 108 : StopLow = StopMiddle + 1;
421 : else /* found it */
422 138 : return StopMiddle;
423 : }
424 :
425 60 : return -1;
426 : }
427 :
428 : /*
429 : * qsort comparator functions
430 : */
431 :
432 : static int
433 78 : compare_int(const void *va, const void *vb)
434 : {
435 78 : int a = *((const int *) va);
436 78 : int b = *((const int *) vb);
437 :
438 78 : return pg_cmp_s32(a, b);
439 : }
440 :
441 : static int
442 102 : compare_text_lexemes(const void *va, const void *vb)
443 : {
444 102 : Datum a = *((const Datum *) va);
445 102 : Datum b = *((const Datum *) vb);
446 102 : char *alex = VARDATA_ANY(a);
447 102 : int alex_len = VARSIZE_ANY_EXHDR(a);
448 102 : char *blex = VARDATA_ANY(b);
449 102 : int blex_len = VARSIZE_ANY_EXHDR(b);
450 :
451 102 : return tsCompareString(alex, alex_len, blex, blex_len, false);
452 : }
453 :
454 : /*
455 : * Internal routine to delete lexemes from TSVector by array of offsets.
456 : *
457 : * int *indices_to_delete -- array of lexeme offsets to delete (modified here!)
458 : * int indices_count -- size of that array
459 : *
460 : * Returns new TSVector without given lexemes along with their positions
461 : * and weights.
462 : */
463 : static TSVector
464 66 : tsvector_delete_by_indices(TSVector tsv, int *indices_to_delete,
465 : int indices_count)
466 : {
467 : TSVector tsout;
468 66 : WordEntry *arrin = ARRPTR(tsv),
469 : *arrout;
470 66 : char *data = STRPTR(tsv),
471 : *dataout;
472 : int i, /* index in arrin */
473 : j, /* index in arrout */
474 : k, /* index in indices_to_delete */
475 : curoff; /* index in dataout area */
476 :
477 : /*
478 : * Sort the filter array to simplify membership checks below. Also, get
479 : * rid of any duplicate entries, so that we can assume that indices_count
480 : * is exactly equal to the number of lexemes that will be removed.
481 : */
482 66 : if (indices_count > 1)
483 : {
484 30 : qsort(indices_to_delete, indices_count, sizeof(int), compare_int);
485 30 : indices_count = qunique(indices_to_delete, indices_count, sizeof(int),
486 : compare_int);
487 : }
488 :
489 : /*
490 : * Here we overestimate tsout size, since we don't know how much space is
491 : * used by the deleted lexeme(s). We will set exact size below.
492 : */
493 66 : tsout = (TSVector) palloc0(VARSIZE(tsv));
494 :
495 : /* This count must be correct because STRPTR(tsout) relies on it. */
496 66 : tsout->size = tsv->size - indices_count;
497 :
498 : /*
499 : * Copy tsv to tsout, skipping lexemes listed in indices_to_delete.
500 : */
501 66 : arrout = ARRPTR(tsout);
502 66 : dataout = STRPTR(tsout);
503 66 : curoff = 0;
504 396 : for (i = j = k = 0; i < tsv->size; i++)
505 : {
506 : /*
507 : * If current i is present in indices_to_delete, skip this lexeme.
508 : * Since indices_to_delete is already sorted, we only need to check
509 : * the current (k'th) entry.
510 : */
511 330 : if (k < indices_count && i == indices_to_delete[k])
512 : {
513 96 : k++;
514 96 : continue;
515 : }
516 :
517 : /* Copy lexeme and its positions and weights */
518 234 : memcpy(dataout + curoff, data + arrin[i].pos, arrin[i].len);
519 234 : arrout[j].haspos = arrin[i].haspos;
520 234 : arrout[j].len = arrin[i].len;
521 234 : arrout[j].pos = curoff;
522 234 : curoff += arrin[i].len;
523 234 : if (arrin[i].haspos)
524 : {
525 156 : int len = POSDATALEN(tsv, arrin + i) * sizeof(WordEntryPos)
526 156 : + sizeof(uint16);
527 :
528 156 : curoff = SHORTALIGN(curoff);
529 156 : memcpy(dataout + curoff,
530 156 : STRPTR(tsv) + SHORTALIGN(arrin[i].pos + arrin[i].len),
531 : len);
532 156 : curoff += len;
533 : }
534 :
535 234 : j++;
536 : }
537 :
538 : /*
539 : * k should now be exactly equal to indices_count. If it isn't then the
540 : * caller provided us with indices outside of [0, tsv->size) range and
541 : * estimation of tsout's size is wrong.
542 : */
543 : Assert(k == indices_count);
544 :
545 66 : SET_VARSIZE(tsout, CALCDATASIZE(tsout->size, curoff));
546 66 : return tsout;
547 : }
548 :
549 : /*
550 : * Delete given lexeme from tsvector.
551 : * Implementation of user-level ts_delete(tsvector, text).
552 : */
553 : Datum
554 36 : tsvector_delete_str(PG_FUNCTION_ARGS)
555 : {
556 36 : TSVector tsin = PG_GETARG_TSVECTOR(0),
557 : tsout;
558 36 : text *tlexeme = PG_GETARG_TEXT_PP(1);
559 36 : char *lexeme = VARDATA_ANY(tlexeme);
560 36 : int lexeme_len = VARSIZE_ANY_EXHDR(tlexeme),
561 : skip_index;
562 :
563 36 : if ((skip_index = tsvector_bsearch(tsin, lexeme, lexeme_len)) == -1)
564 12 : PG_RETURN_POINTER(tsin);
565 :
566 24 : tsout = tsvector_delete_by_indices(tsin, &skip_index, 1);
567 :
568 24 : PG_FREE_IF_COPY(tsin, 0);
569 24 : PG_FREE_IF_COPY(tlexeme, 1);
570 24 : PG_RETURN_POINTER(tsout);
571 : }
572 :
573 : /*
574 : * Delete given array of lexemes from tsvector.
575 : * Implementation of user-level ts_delete(tsvector, text[]).
576 : */
577 : Datum
578 42 : tsvector_delete_arr(PG_FUNCTION_ARGS)
579 : {
580 42 : TSVector tsin = PG_GETARG_TSVECTOR(0),
581 : tsout;
582 42 : ArrayType *lexemes = PG_GETARG_ARRAYTYPE_P(1);
583 : int i,
584 : nlex,
585 : skip_count,
586 : *skip_indices;
587 : Datum *dlexemes;
588 : bool *nulls;
589 :
590 42 : deconstruct_array_builtin(lexemes, TEXTOID, &dlexemes, &nulls, &nlex);
591 :
592 : /*
593 : * In typical use case array of lexemes to delete is relatively small. So
594 : * here we optimize things for that scenario: iterate through lexarr
595 : * performing binary search of each lexeme from lexarr in tsvector.
596 : */
597 42 : skip_indices = palloc0(nlex * sizeof(int));
598 168 : for (i = skip_count = 0; i < nlex; i++)
599 : {
600 : char *lex;
601 : int lex_len,
602 : lex_pos;
603 :
604 : /* Ignore null array elements, they surely don't match */
605 126 : if (nulls[i])
606 6 : continue;
607 :
608 120 : lex = VARDATA(dlexemes[i]);
609 120 : lex_len = VARSIZE(dlexemes[i]) - VARHDRSZ;
610 120 : lex_pos = tsvector_bsearch(tsin, lex, lex_len);
611 :
612 120 : if (lex_pos >= 0)
613 78 : skip_indices[skip_count++] = lex_pos;
614 : }
615 :
616 42 : tsout = tsvector_delete_by_indices(tsin, skip_indices, skip_count);
617 :
618 42 : pfree(skip_indices);
619 42 : PG_FREE_IF_COPY(tsin, 0);
620 42 : PG_FREE_IF_COPY(lexemes, 1);
621 :
622 42 : PG_RETURN_POINTER(tsout);
623 : }
624 :
625 : /*
626 : * Expand tsvector as table with following columns:
627 : * lexeme: lexeme text
628 : * positions: integer array of lexeme positions
629 : * weights: char array of weights corresponding to positions
630 : */
631 : Datum
632 180 : tsvector_unnest(PG_FUNCTION_ARGS)
633 : {
634 : FuncCallContext *funcctx;
635 : TSVector tsin;
636 :
637 180 : if (SRF_IS_FIRSTCALL())
638 : {
639 : MemoryContext oldcontext;
640 : TupleDesc tupdesc;
641 :
642 30 : funcctx = SRF_FIRSTCALL_INIT();
643 30 : oldcontext = MemoryContextSwitchTo(funcctx->multi_call_memory_ctx);
644 :
645 30 : tupdesc = CreateTemplateTupleDesc(3);
646 30 : TupleDescInitEntry(tupdesc, (AttrNumber) 1, "lexeme",
647 : TEXTOID, -1, 0);
648 30 : TupleDescInitEntry(tupdesc, (AttrNumber) 2, "positions",
649 : INT2ARRAYOID, -1, 0);
650 30 : TupleDescInitEntry(tupdesc, (AttrNumber) 3, "weights",
651 : TEXTARRAYOID, -1, 0);
652 30 : if (get_call_result_type(fcinfo, NULL, &tupdesc) != TYPEFUNC_COMPOSITE)
653 0 : elog(ERROR, "return type must be a row type");
654 30 : funcctx->tuple_desc = tupdesc;
655 :
656 30 : funcctx->user_fctx = PG_GETARG_TSVECTOR_COPY(0);
657 :
658 30 : MemoryContextSwitchTo(oldcontext);
659 : }
660 :
661 180 : funcctx = SRF_PERCALL_SETUP();
662 180 : tsin = (TSVector) funcctx->user_fctx;
663 :
664 180 : if (funcctx->call_cntr < tsin->size)
665 : {
666 150 : WordEntry *arrin = ARRPTR(tsin);
667 150 : char *data = STRPTR(tsin);
668 : HeapTuple tuple;
669 : int j,
670 150 : i = funcctx->call_cntr;
671 150 : bool nulls[] = {false, false, false};
672 : Datum values[3];
673 :
674 150 : values[0] = PointerGetDatum(cstring_to_text_with_len(data + arrin[i].pos, arrin[i].len));
675 :
676 150 : if (arrin[i].haspos)
677 : {
678 : WordEntryPosVector *posv;
679 : Datum *positions;
680 : Datum *weights;
681 : char weight;
682 :
683 : /*
684 : * Internally tsvector stores position and weight in the same
685 : * uint16 (2 bits for weight, 14 for position). Here we extract
686 : * that in two separate arrays.
687 : */
688 90 : posv = _POSVECPTR(tsin, arrin + i);
689 90 : positions = palloc(posv->npos * sizeof(Datum));
690 90 : weights = palloc(posv->npos * sizeof(Datum));
691 252 : for (j = 0; j < posv->npos; j++)
692 : {
693 162 : positions[j] = Int16GetDatum(WEP_GETPOS(posv->pos[j]));
694 162 : weight = 'D' - WEP_GETWEIGHT(posv->pos[j]);
695 162 : weights[j] = PointerGetDatum(cstring_to_text_with_len(&weight,
696 : 1));
697 : }
698 :
699 90 : values[1] = PointerGetDatum(construct_array_builtin(positions, posv->npos, INT2OID));
700 90 : values[2] = PointerGetDatum(construct_array_builtin(weights, posv->npos, TEXTOID));
701 : }
702 : else
703 : {
704 60 : nulls[1] = nulls[2] = true;
705 : }
706 :
707 150 : tuple = heap_form_tuple(funcctx->tuple_desc, values, nulls);
708 150 : SRF_RETURN_NEXT(funcctx, HeapTupleGetDatum(tuple));
709 : }
710 : else
711 : {
712 30 : SRF_RETURN_DONE(funcctx);
713 : }
714 : }
715 :
716 : /*
717 : * Convert tsvector to array of lexemes.
718 : */
719 : Datum
720 12 : tsvector_to_array(PG_FUNCTION_ARGS)
721 : {
722 12 : TSVector tsin = PG_GETARG_TSVECTOR(0);
723 12 : WordEntry *arrin = ARRPTR(tsin);
724 : Datum *elements;
725 : int i;
726 : ArrayType *array;
727 :
728 12 : elements = palloc(tsin->size * sizeof(Datum));
729 :
730 72 : for (i = 0; i < tsin->size; i++)
731 : {
732 60 : elements[i] = PointerGetDatum(cstring_to_text_with_len(STRPTR(tsin) + arrin[i].pos,
733 60 : arrin[i].len));
734 : }
735 :
736 12 : array = construct_array_builtin(elements, tsin->size, TEXTOID);
737 :
738 12 : pfree(elements);
739 12 : PG_FREE_IF_COPY(tsin, 0);
740 12 : PG_RETURN_POINTER(array);
741 : }
742 :
743 : /*
744 : * Build tsvector from array of lexemes.
745 : */
746 : Datum
747 24 : array_to_tsvector(PG_FUNCTION_ARGS)
748 : {
749 24 : ArrayType *v = PG_GETARG_ARRAYTYPE_P(0);
750 : TSVector tsout;
751 : Datum *dlexemes;
752 : WordEntry *arrout;
753 : bool *nulls;
754 : int nitems,
755 : i,
756 : tslen,
757 24 : datalen = 0;
758 : char *cur;
759 :
760 24 : deconstruct_array_builtin(v, TEXTOID, &dlexemes, &nulls, &nitems);
761 :
762 : /*
763 : * Reject nulls and zero length strings (maybe we should just ignore them,
764 : * instead?)
765 : */
766 126 : for (i = 0; i < nitems; i++)
767 : {
768 114 : if (nulls[i])
769 6 : ereport(ERROR,
770 : (errcode(ERRCODE_NULL_VALUE_NOT_ALLOWED),
771 : errmsg("lexeme array may not contain nulls")));
772 :
773 108 : if (VARSIZE(dlexemes[i]) - VARHDRSZ == 0)
774 6 : ereport(ERROR,
775 : (errcode(ERRCODE_ZERO_LENGTH_CHARACTER_STRING),
776 : errmsg("lexeme array may not contain empty strings")));
777 : }
778 :
779 : /* Sort and de-dup, because this is required for a valid tsvector. */
780 12 : if (nitems > 1)
781 : {
782 12 : qsort(dlexemes, nitems, sizeof(Datum), compare_text_lexemes);
783 12 : nitems = qunique(dlexemes, nitems, sizeof(Datum),
784 : compare_text_lexemes);
785 : }
786 :
787 : /* Calculate space needed for surviving lexemes. */
788 60 : for (i = 0; i < nitems; i++)
789 48 : datalen += VARSIZE(dlexemes[i]) - VARHDRSZ;
790 12 : tslen = CALCDATASIZE(nitems, datalen);
791 :
792 : /* Allocate and fill tsvector. */
793 12 : tsout = (TSVector) palloc0(tslen);
794 12 : SET_VARSIZE(tsout, tslen);
795 12 : tsout->size = nitems;
796 :
797 12 : arrout = ARRPTR(tsout);
798 12 : cur = STRPTR(tsout);
799 60 : for (i = 0; i < nitems; i++)
800 : {
801 48 : char *lex = VARDATA(dlexemes[i]);
802 48 : int lex_len = VARSIZE(dlexemes[i]) - VARHDRSZ;
803 :
804 48 : memcpy(cur, lex, lex_len);
805 48 : arrout[i].haspos = 0;
806 48 : arrout[i].len = lex_len;
807 48 : arrout[i].pos = cur - STRPTR(tsout);
808 48 : cur += lex_len;
809 : }
810 :
811 12 : PG_FREE_IF_COPY(v, 0);
812 12 : PG_RETURN_POINTER(tsout);
813 : }
814 :
815 : /*
816 : * ts_filter(): keep only lexemes with given weights in tsvector.
817 : */
818 : Datum
819 18 : tsvector_filter(PG_FUNCTION_ARGS)
820 : {
821 18 : TSVector tsin = PG_GETARG_TSVECTOR(0),
822 : tsout;
823 18 : ArrayType *weights = PG_GETARG_ARRAYTYPE_P(1);
824 18 : WordEntry *arrin = ARRPTR(tsin),
825 : *arrout;
826 18 : char *datain = STRPTR(tsin),
827 : *dataout;
828 : Datum *dweights;
829 : bool *nulls;
830 : int nweights;
831 : int i,
832 : j;
833 18 : int cur_pos = 0;
834 18 : char mask = 0;
835 :
836 18 : deconstruct_array_builtin(weights, CHAROID, &dweights, &nulls, &nweights);
837 :
838 42 : for (i = 0; i < nweights; i++)
839 : {
840 : char char_weight;
841 :
842 30 : if (nulls[i])
843 6 : ereport(ERROR,
844 : (errcode(ERRCODE_NULL_VALUE_NOT_ALLOWED),
845 : errmsg("weight array may not contain nulls")));
846 :
847 24 : char_weight = DatumGetChar(dweights[i]);
848 24 : switch (char_weight)
849 : {
850 18 : case 'A':
851 : case 'a':
852 18 : mask = mask | 8;
853 18 : break;
854 6 : case 'B':
855 : case 'b':
856 6 : mask = mask | 4;
857 6 : break;
858 0 : case 'C':
859 : case 'c':
860 0 : mask = mask | 2;
861 0 : break;
862 0 : case 'D':
863 : case 'd':
864 0 : mask = mask | 1;
865 0 : break;
866 0 : default:
867 0 : ereport(ERROR,
868 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
869 : errmsg("unrecognized weight: \"%c\"", char_weight)));
870 : }
871 : }
872 :
873 12 : tsout = (TSVector) palloc0(VARSIZE(tsin));
874 12 : tsout->size = tsin->size;
875 12 : arrout = ARRPTR(tsout);
876 12 : dataout = STRPTR(tsout);
877 :
878 108 : for (i = j = 0; i < tsin->size; i++)
879 : {
880 : WordEntryPosVector *posvin,
881 : *posvout;
882 96 : int npos = 0;
883 : int k;
884 :
885 96 : if (!arrin[i].haspos)
886 30 : continue;
887 :
888 66 : posvin = _POSVECPTR(tsin, arrin + i);
889 66 : posvout = (WordEntryPosVector *)
890 66 : (dataout + SHORTALIGN(cur_pos + arrin[i].len));
891 :
892 132 : for (k = 0; k < posvin->npos; k++)
893 : {
894 66 : if (mask & (1 << WEP_GETWEIGHT(posvin->pos[k])))
895 30 : posvout->pos[npos++] = posvin->pos[k];
896 : }
897 :
898 : /* if no satisfactory positions found, skip lexeme */
899 66 : if (!npos)
900 36 : continue;
901 :
902 30 : arrout[j].haspos = true;
903 30 : arrout[j].len = arrin[i].len;
904 30 : arrout[j].pos = cur_pos;
905 :
906 30 : memcpy(dataout + cur_pos, datain + arrin[i].pos, arrin[i].len);
907 30 : posvout->npos = npos;
908 30 : cur_pos += SHORTALIGN(arrin[i].len);
909 30 : cur_pos += POSDATALEN(tsout, arrout + j) * sizeof(WordEntryPos) +
910 : sizeof(uint16);
911 30 : j++;
912 : }
913 :
914 12 : tsout->size = j;
915 12 : if (dataout != STRPTR(tsout))
916 12 : memmove(STRPTR(tsout), dataout, cur_pos);
917 :
918 12 : SET_VARSIZE(tsout, CALCDATASIZE(tsout->size, cur_pos));
919 :
920 12 : PG_FREE_IF_COPY(tsin, 0);
921 12 : PG_RETURN_POINTER(tsout);
922 : }
923 :
924 : Datum
925 12 : tsvector_concat(PG_FUNCTION_ARGS)
926 : {
927 12 : TSVector in1 = PG_GETARG_TSVECTOR(0);
928 12 : TSVector in2 = PG_GETARG_TSVECTOR(1);
929 : TSVector out;
930 : WordEntry *ptr;
931 : WordEntry *ptr1,
932 : *ptr2;
933 : WordEntryPos *p;
934 12 : int maxpos = 0,
935 : i,
936 : j,
937 : i1,
938 : i2,
939 : dataoff,
940 : output_bytes,
941 : output_size;
942 : char *data,
943 : *data1,
944 : *data2;
945 :
946 : /* Get max position in in1; we'll need this to offset in2's positions */
947 12 : ptr = ARRPTR(in1);
948 12 : i = in1->size;
949 30 : while (i--)
950 : {
951 18 : if ((j = POSDATALEN(in1, ptr)) != 0)
952 : {
953 18 : p = POSDATAPTR(in1, ptr);
954 36 : while (j--)
955 : {
956 18 : if (WEP_GETPOS(*p) > maxpos)
957 12 : maxpos = WEP_GETPOS(*p);
958 18 : p++;
959 : }
960 : }
961 18 : ptr++;
962 : }
963 :
964 12 : ptr1 = ARRPTR(in1);
965 12 : ptr2 = ARRPTR(in2);
966 12 : data1 = STRPTR(in1);
967 12 : data2 = STRPTR(in2);
968 12 : i1 = in1->size;
969 12 : i2 = in2->size;
970 :
971 : /*
972 : * Conservative estimate of space needed. We might need all the data in
973 : * both inputs, and conceivably add a pad byte before position data for
974 : * each item where there was none before.
975 : */
976 12 : output_bytes = VARSIZE(in1) + VARSIZE(in2) + i1 + i2;
977 :
978 12 : out = (TSVector) palloc0(output_bytes);
979 12 : SET_VARSIZE(out, output_bytes);
980 :
981 : /*
982 : * We must make out->size valid so that STRPTR(out) is sensible. We'll
983 : * collapse out any unused space at the end.
984 : */
985 12 : out->size = in1->size + in2->size;
986 :
987 12 : ptr = ARRPTR(out);
988 12 : data = STRPTR(out);
989 12 : dataoff = 0;
990 30 : while (i1 && i2)
991 : {
992 18 : int cmp = compareEntry(data1, ptr1, data2, ptr2);
993 :
994 18 : if (cmp < 0)
995 : { /* in1 first */
996 6 : ptr->haspos = ptr1->haspos;
997 6 : ptr->len = ptr1->len;
998 6 : memcpy(data + dataoff, data1 + ptr1->pos, ptr1->len);
999 6 : ptr->pos = dataoff;
1000 6 : dataoff += ptr1->len;
1001 6 : if (ptr->haspos)
1002 : {
1003 6 : dataoff = SHORTALIGN(dataoff);
1004 6 : memcpy(data + dataoff, _POSVECPTR(in1, ptr1), POSDATALEN(in1, ptr1) * sizeof(WordEntryPos) + sizeof(uint16));
1005 6 : dataoff += POSDATALEN(in1, ptr1) * sizeof(WordEntryPos) + sizeof(uint16);
1006 : }
1007 :
1008 6 : ptr++;
1009 6 : ptr1++;
1010 6 : i1--;
1011 : }
1012 12 : else if (cmp > 0)
1013 : { /* in2 first */
1014 6 : ptr->haspos = ptr2->haspos;
1015 6 : ptr->len = ptr2->len;
1016 6 : memcpy(data + dataoff, data2 + ptr2->pos, ptr2->len);
1017 6 : ptr->pos = dataoff;
1018 6 : dataoff += ptr2->len;
1019 6 : if (ptr->haspos)
1020 : {
1021 0 : int addlen = add_pos(in2, ptr2, out, ptr, maxpos);
1022 :
1023 0 : if (addlen == 0)
1024 0 : ptr->haspos = 0;
1025 : else
1026 : {
1027 0 : dataoff = SHORTALIGN(dataoff);
1028 0 : dataoff += addlen * sizeof(WordEntryPos) + sizeof(uint16);
1029 : }
1030 : }
1031 :
1032 6 : ptr++;
1033 6 : ptr2++;
1034 6 : i2--;
1035 : }
1036 : else
1037 : {
1038 6 : ptr->haspos = ptr1->haspos | ptr2->haspos;
1039 6 : ptr->len = ptr1->len;
1040 6 : memcpy(data + dataoff, data1 + ptr1->pos, ptr1->len);
1041 6 : ptr->pos = dataoff;
1042 6 : dataoff += ptr1->len;
1043 6 : if (ptr->haspos)
1044 : {
1045 6 : if (ptr1->haspos)
1046 : {
1047 6 : dataoff = SHORTALIGN(dataoff);
1048 6 : memcpy(data + dataoff, _POSVECPTR(in1, ptr1), POSDATALEN(in1, ptr1) * sizeof(WordEntryPos) + sizeof(uint16));
1049 6 : dataoff += POSDATALEN(in1, ptr1) * sizeof(WordEntryPos) + sizeof(uint16);
1050 6 : if (ptr2->haspos)
1051 6 : dataoff += add_pos(in2, ptr2, out, ptr, maxpos) * sizeof(WordEntryPos);
1052 : }
1053 : else /* must have ptr2->haspos */
1054 : {
1055 0 : int addlen = add_pos(in2, ptr2, out, ptr, maxpos);
1056 :
1057 0 : if (addlen == 0)
1058 0 : ptr->haspos = 0;
1059 : else
1060 : {
1061 0 : dataoff = SHORTALIGN(dataoff);
1062 0 : dataoff += addlen * sizeof(WordEntryPos) + sizeof(uint16);
1063 : }
1064 : }
1065 : }
1066 :
1067 6 : ptr++;
1068 6 : ptr1++;
1069 6 : ptr2++;
1070 6 : i1--;
1071 6 : i2--;
1072 : }
1073 : }
1074 :
1075 18 : while (i1)
1076 : {
1077 6 : ptr->haspos = ptr1->haspos;
1078 6 : ptr->len = ptr1->len;
1079 6 : memcpy(data + dataoff, data1 + ptr1->pos, ptr1->len);
1080 6 : ptr->pos = dataoff;
1081 6 : dataoff += ptr1->len;
1082 6 : if (ptr->haspos)
1083 : {
1084 6 : dataoff = SHORTALIGN(dataoff);
1085 6 : memcpy(data + dataoff, _POSVECPTR(in1, ptr1), POSDATALEN(in1, ptr1) * sizeof(WordEntryPos) + sizeof(uint16));
1086 6 : dataoff += POSDATALEN(in1, ptr1) * sizeof(WordEntryPos) + sizeof(uint16);
1087 : }
1088 :
1089 6 : ptr++;
1090 6 : ptr1++;
1091 6 : i1--;
1092 : }
1093 :
1094 18 : while (i2)
1095 : {
1096 6 : ptr->haspos = ptr2->haspos;
1097 6 : ptr->len = ptr2->len;
1098 6 : memcpy(data + dataoff, data2 + ptr2->pos, ptr2->len);
1099 6 : ptr->pos = dataoff;
1100 6 : dataoff += ptr2->len;
1101 6 : if (ptr->haspos)
1102 : {
1103 6 : int addlen = add_pos(in2, ptr2, out, ptr, maxpos);
1104 :
1105 6 : if (addlen == 0)
1106 0 : ptr->haspos = 0;
1107 : else
1108 : {
1109 6 : dataoff = SHORTALIGN(dataoff);
1110 6 : dataoff += addlen * sizeof(WordEntryPos) + sizeof(uint16);
1111 : }
1112 : }
1113 :
1114 6 : ptr++;
1115 6 : ptr2++;
1116 6 : i2--;
1117 : }
1118 :
1119 : /*
1120 : * Instead of checking each offset individually, we check for overflow of
1121 : * pos fields once at the end.
1122 : */
1123 12 : if (dataoff > MAXSTRPOS)
1124 0 : ereport(ERROR,
1125 : (errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
1126 : errmsg("string is too long for tsvector (%d bytes, max %d bytes)", dataoff, MAXSTRPOS)));
1127 :
1128 : /*
1129 : * Adjust sizes (asserting that we didn't overrun the original estimates)
1130 : * and collapse out any unused array entries.
1131 : */
1132 12 : output_size = ptr - ARRPTR(out);
1133 : Assert(output_size <= out->size);
1134 12 : out->size = output_size;
1135 12 : if (data != STRPTR(out))
1136 6 : memmove(STRPTR(out), data, dataoff);
1137 12 : output_bytes = CALCDATASIZE(out->size, dataoff);
1138 : Assert(output_bytes <= VARSIZE(out));
1139 12 : SET_VARSIZE(out, output_bytes);
1140 :
1141 12 : PG_FREE_IF_COPY(in1, 0);
1142 12 : PG_FREE_IF_COPY(in2, 1);
1143 12 : PG_RETURN_POINTER(out);
1144 : }
1145 :
1146 : /*
1147 : * Compare two strings by tsvector rules.
1148 : *
1149 : * if prefix = true then it returns zero value iff b has prefix a
1150 : */
1151 : int32
1152 6874706 : tsCompareString(char *a, int lena, char *b, int lenb, bool prefix)
1153 : {
1154 : int cmp;
1155 :
1156 6874706 : if (lena == 0)
1157 : {
1158 36 : if (prefix)
1159 0 : cmp = 0; /* empty string is prefix of anything */
1160 : else
1161 36 : cmp = (lenb > 0) ? -1 : 0;
1162 : }
1163 6874670 : else if (lenb == 0)
1164 : {
1165 0 : cmp = (lena > 0) ? 1 : 0;
1166 : }
1167 : else
1168 : {
1169 6874670 : cmp = memcmp(a, b, Min((unsigned int) lena, (unsigned int) lenb));
1170 :
1171 6874670 : if (prefix)
1172 : {
1173 16458 : if (cmp == 0 && lena > lenb)
1174 0 : cmp = 1; /* a is longer, so not a prefix of b */
1175 : }
1176 6858212 : else if (cmp == 0 && lena != lenb)
1177 : {
1178 32446 : cmp = (lena < lenb) ? -1 : 1;
1179 : }
1180 : }
1181 :
1182 6874706 : return cmp;
1183 : }
1184 :
1185 : /*
1186 : * Check weight info or/and fill 'data' with the required positions
1187 : */
1188 : static TSTernaryValue
1189 68082 : checkclass_str(CHKVAL *chkval, WordEntry *entry, QueryOperand *val,
1190 : ExecPhraseData *data)
1191 : {
1192 68082 : TSTernaryValue result = TS_NO;
1193 :
1194 : Assert(data == NULL || data->npos == 0);
1195 :
1196 68082 : if (entry->haspos)
1197 : {
1198 : WordEntryPosVector *posvec;
1199 :
1200 : /*
1201 : * We can't use the _POSVECPTR macro here because the pointer to the
1202 : * tsvector's lexeme storage is already contained in chkval->values.
1203 : */
1204 4488 : posvec = (WordEntryPosVector *)
1205 4488 : (chkval->values + SHORTALIGN(entry->pos + entry->len));
1206 :
1207 4488 : if (val->weight && data)
1208 48 : {
1209 48 : WordEntryPos *posvec_iter = posvec->pos;
1210 : WordEntryPos *dptr;
1211 :
1212 : /*
1213 : * Filter position information by weights
1214 : */
1215 48 : dptr = data->pos = palloc(sizeof(WordEntryPos) * posvec->npos);
1216 48 : data->allocated = true;
1217 :
1218 : /* Is there a position with a matching weight? */
1219 96 : while (posvec_iter < posvec->pos + posvec->npos)
1220 : {
1221 : /* If true, append this position to the data->pos */
1222 48 : if (val->weight & (1 << WEP_GETWEIGHT(*posvec_iter)))
1223 : {
1224 24 : *dptr = WEP_GETPOS(*posvec_iter);
1225 24 : dptr++;
1226 : }
1227 :
1228 48 : posvec_iter++;
1229 : }
1230 :
1231 48 : data->npos = dptr - data->pos;
1232 :
1233 48 : if (data->npos > 0)
1234 24 : result = TS_YES;
1235 : else
1236 : {
1237 24 : pfree(data->pos);
1238 24 : data->pos = NULL;
1239 24 : data->allocated = false;
1240 : }
1241 : }
1242 4440 : else if (val->weight)
1243 : {
1244 456 : WordEntryPos *posvec_iter = posvec->pos;
1245 :
1246 : /* Is there a position with a matching weight? */
1247 690 : while (posvec_iter < posvec->pos + posvec->npos)
1248 : {
1249 504 : if (val->weight & (1 << WEP_GETWEIGHT(*posvec_iter)))
1250 : {
1251 270 : result = TS_YES;
1252 270 : break; /* no need to go further */
1253 : }
1254 :
1255 234 : posvec_iter++;
1256 : }
1257 : }
1258 3984 : else if (data)
1259 : {
1260 2274 : data->npos = posvec->npos;
1261 2274 : data->pos = posvec->pos;
1262 2274 : data->allocated = false;
1263 2274 : result = TS_YES;
1264 : }
1265 : else
1266 : {
1267 : /* simplest case: no weight check, positions not needed */
1268 1710 : result = TS_YES;
1269 : }
1270 : }
1271 : else
1272 : {
1273 : /*
1274 : * Position info is lacking, so if the caller requires it, we can only
1275 : * say that maybe there is a match.
1276 : *
1277 : * Notice, however, that we *don't* check val->weight here.
1278 : * Historically, stripped tsvectors are considered to match queries
1279 : * whether or not the query has a weight restriction; that's a little
1280 : * dubious but we'll preserve the behavior.
1281 : */
1282 63594 : if (data)
1283 23058 : result = TS_MAYBE;
1284 : else
1285 40536 : result = TS_YES;
1286 : }
1287 :
1288 68082 : return result;
1289 : }
1290 :
1291 : /*
1292 : * TS_execute callback for matching a tsquery operand to plain tsvector data
1293 : */
1294 : static TSTernaryValue
1295 284022 : checkcondition_str(void *checkval, QueryOperand *val, ExecPhraseData *data)
1296 : {
1297 284022 : CHKVAL *chkval = (CHKVAL *) checkval;
1298 284022 : WordEntry *StopLow = chkval->arrb;
1299 284022 : WordEntry *StopHigh = chkval->arre;
1300 284022 : WordEntry *StopMiddle = StopHigh;
1301 284022 : TSTernaryValue res = TS_NO;
1302 :
1303 : /* Loop invariant: StopLow <= val < StopHigh */
1304 1786806 : while (StopLow < StopHigh)
1305 : {
1306 : int difference;
1307 :
1308 1555830 : StopMiddle = StopLow + (StopHigh - StopLow) / 2;
1309 1555830 : difference = tsCompareString(chkval->operand + val->distance,
1310 1555830 : val->length,
1311 1555830 : chkval->values + StopMiddle->pos,
1312 1555830 : StopMiddle->len,
1313 : false);
1314 :
1315 1555830 : if (difference == 0)
1316 : {
1317 : /* Check weight info & fill 'data' with positions */
1318 53046 : res = checkclass_str(chkval, StopMiddle, val, data);
1319 53046 : break;
1320 : }
1321 1502784 : else if (difference > 0)
1322 847512 : StopLow = StopMiddle + 1;
1323 : else
1324 655272 : StopHigh = StopMiddle;
1325 : }
1326 :
1327 : /*
1328 : * If it's a prefix search, we should also consider lexemes that the
1329 : * search term is a prefix of (which will necessarily immediately follow
1330 : * the place we found in the above loop). But we can skip them if there
1331 : * was a definite match on the exact term AND the caller doesn't need
1332 : * position info.
1333 : */
1334 284022 : if (val->prefix && (res != TS_YES || data))
1335 : {
1336 16524 : WordEntryPos *allpos = NULL;
1337 16524 : int npos = 0,
1338 16524 : totalpos = 0;
1339 :
1340 : /* adjust start position for corner case */
1341 16524 : if (StopLow >= StopHigh)
1342 16512 : StopMiddle = StopHigh;
1343 :
1344 : /* we don't try to re-use any data from the initial match */
1345 16524 : if (data)
1346 : {
1347 36 : if (data->allocated)
1348 0 : pfree(data->pos);
1349 36 : data->pos = NULL;
1350 36 : data->allocated = false;
1351 36 : data->npos = 0;
1352 : }
1353 16524 : res = TS_NO;
1354 :
1355 31458 : while ((res != TS_YES || data) &&
1356 47490 : StopMiddle < chkval->arre &&
1357 15930 : tsCompareString(chkval->operand + val->distance,
1358 15930 : val->length,
1359 15930 : chkval->values + StopMiddle->pos,
1360 15930 : StopMiddle->len,
1361 : true) == 0)
1362 : {
1363 : TSTernaryValue subres;
1364 :
1365 15036 : subres = checkclass_str(chkval, StopMiddle, val, data);
1366 :
1367 15036 : if (subres != TS_NO)
1368 : {
1369 14976 : if (data)
1370 : {
1371 : /*
1372 : * We need to join position information
1373 : */
1374 42 : if (subres == TS_MAYBE)
1375 : {
1376 : /*
1377 : * No position info for this match, so we must report
1378 : * MAYBE overall.
1379 : */
1380 0 : res = TS_MAYBE;
1381 : /* forget any previous positions */
1382 0 : npos = 0;
1383 : /* don't leak storage */
1384 0 : if (allpos)
1385 0 : pfree(allpos);
1386 0 : break;
1387 : }
1388 :
1389 78 : while (npos + data->npos > totalpos)
1390 : {
1391 36 : if (totalpos == 0)
1392 : {
1393 36 : totalpos = 256;
1394 36 : allpos = palloc(sizeof(WordEntryPos) * totalpos);
1395 : }
1396 : else
1397 : {
1398 0 : totalpos *= 2;
1399 0 : allpos = repalloc(allpos, sizeof(WordEntryPos) * totalpos);
1400 : }
1401 : }
1402 :
1403 42 : memcpy(allpos + npos, data->pos, sizeof(WordEntryPos) * data->npos);
1404 42 : npos += data->npos;
1405 :
1406 : /* don't leak storage from individual matches */
1407 42 : if (data->allocated)
1408 24 : pfree(data->pos);
1409 42 : data->pos = NULL;
1410 42 : data->allocated = false;
1411 : /* it's important to reset data->npos before next loop */
1412 42 : data->npos = 0;
1413 : }
1414 : else
1415 : {
1416 : /* Don't need positions, just handle YES/MAYBE */
1417 14934 : if (subres == TS_YES || res == TS_NO)
1418 14934 : res = subres;
1419 : }
1420 : }
1421 :
1422 15036 : StopMiddle++;
1423 : }
1424 :
1425 16524 : if (data && npos > 0)
1426 : {
1427 : /* Sort and make unique array of found positions */
1428 36 : data->pos = allpos;
1429 36 : qsort(data->pos, npos, sizeof(WordEntryPos), compareWordEntryPos);
1430 36 : data->npos = qunique(data->pos, npos, sizeof(WordEntryPos),
1431 : compareWordEntryPos);
1432 36 : data->allocated = true;
1433 36 : res = TS_YES;
1434 : }
1435 : }
1436 :
1437 284022 : return res;
1438 : }
1439 :
1440 : /*
1441 : * Compute output position list for a tsquery operator in phrase mode.
1442 : *
1443 : * Merge the position lists in Ldata and Rdata as specified by "emit",
1444 : * returning the result list into *data. The input position lists must be
1445 : * sorted and unique, and the output will be as well.
1446 : *
1447 : * data: pointer to initially-all-zeroes output struct, or NULL
1448 : * Ldata, Rdata: input position lists
1449 : * emit: bitmask of TSPO_XXX flags
1450 : * Loffset: offset to be added to Ldata positions before comparing/outputting
1451 : * Roffset: offset to be added to Rdata positions before comparing/outputting
1452 : * max_npos: maximum possible required size of output position array
1453 : *
1454 : * Loffset and Roffset should not be negative, else we risk trying to output
1455 : * negative positions, which won't fit into WordEntryPos.
1456 : *
1457 : * The result is boolean (TS_YES or TS_NO), but for the caller's convenience
1458 : * we return it as TSTernaryValue.
1459 : *
1460 : * Returns TS_YES if any positions were emitted to *data; or if data is NULL,
1461 : * returns TS_YES if any positions would have been emitted.
1462 : */
1463 : #define TSPO_L_ONLY 0x01 /* emit positions appearing only in L */
1464 : #define TSPO_R_ONLY 0x02 /* emit positions appearing only in R */
1465 : #define TSPO_BOTH 0x04 /* emit positions appearing in both L&R */
1466 :
1467 : static TSTernaryValue
1468 29966 : TS_phrase_output(ExecPhraseData *data,
1469 : ExecPhraseData *Ldata,
1470 : ExecPhraseData *Rdata,
1471 : int emit,
1472 : int Loffset,
1473 : int Roffset,
1474 : int max_npos)
1475 : {
1476 : int Lindex,
1477 : Rindex;
1478 :
1479 : /* Loop until both inputs are exhausted */
1480 29966 : Lindex = Rindex = 0;
1481 30998 : while (Lindex < Ldata->npos || Rindex < Rdata->npos)
1482 : {
1483 : int Lpos,
1484 : Rpos;
1485 2334 : int output_pos = 0;
1486 :
1487 : /*
1488 : * Fetch current values to compare. WEP_GETPOS() is needed because
1489 : * ExecPhraseData->data can point to a tsvector's WordEntryPosVector.
1490 : */
1491 2334 : if (Lindex < Ldata->npos)
1492 1686 : Lpos = WEP_GETPOS(Ldata->pos[Lindex]) + Loffset;
1493 : else
1494 : {
1495 : /* L array exhausted, so we're done if R_ONLY isn't set */
1496 648 : if (!(emit & TSPO_R_ONLY))
1497 150 : break;
1498 498 : Lpos = INT_MAX;
1499 : }
1500 2184 : if (Rindex < Rdata->npos)
1501 1938 : Rpos = WEP_GETPOS(Rdata->pos[Rindex]) + Roffset;
1502 : else
1503 : {
1504 : /* R array exhausted, so we're done if L_ONLY isn't set */
1505 246 : if (!(emit & TSPO_L_ONLY))
1506 162 : break;
1507 84 : Rpos = INT_MAX;
1508 : }
1509 :
1510 : /* Merge-join the two input lists */
1511 2022 : if (Lpos < Rpos)
1512 : {
1513 : /* Lpos is not matched in Rdata, should we output it? */
1514 486 : if (emit & TSPO_L_ONLY)
1515 144 : output_pos = Lpos;
1516 486 : Lindex++;
1517 : }
1518 1536 : else if (Lpos == Rpos)
1519 : {
1520 : /* Lpos and Rpos match ... should we output it? */
1521 798 : if (emit & TSPO_BOTH)
1522 702 : output_pos = Rpos;
1523 798 : Lindex++;
1524 798 : Rindex++;
1525 : }
1526 : else /* Lpos > Rpos */
1527 : {
1528 : /* Rpos is not matched in Ldata, should we output it? */
1529 738 : if (emit & TSPO_R_ONLY)
1530 540 : output_pos = Rpos;
1531 738 : Rindex++;
1532 : }
1533 :
1534 2022 : if (output_pos > 0)
1535 : {
1536 1386 : if (data)
1537 : {
1538 : /* Store position, first allocating output array if needed */
1539 396 : if (data->pos == NULL)
1540 : {
1541 318 : data->pos = (WordEntryPos *)
1542 318 : palloc(max_npos * sizeof(WordEntryPos));
1543 318 : data->allocated = true;
1544 : }
1545 396 : data->pos[data->npos++] = output_pos;
1546 : }
1547 : else
1548 : {
1549 : /*
1550 : * Exact positions not needed, so return TS_YES as soon as we
1551 : * know there is at least one.
1552 : */
1553 990 : return TS_YES;
1554 : }
1555 : }
1556 : }
1557 :
1558 28976 : if (data && data->npos > 0)
1559 : {
1560 : /* Let's assert we didn't overrun the array */
1561 : Assert(data->npos <= max_npos);
1562 318 : return TS_YES;
1563 : }
1564 28658 : return TS_NO;
1565 : }
1566 :
1567 : /*
1568 : * Execute tsquery at or below an OP_PHRASE operator.
1569 : *
1570 : * This handles tsquery execution at recursion levels where we need to care
1571 : * about match locations.
1572 : *
1573 : * In addition to the same arguments used for TS_execute, the caller may pass
1574 : * a preinitialized-to-zeroes ExecPhraseData struct, to be filled with lexeme
1575 : * match position info on success. data == NULL if no position data need be
1576 : * returned.
1577 : * Note: the function assumes data != NULL for operators other than OP_PHRASE.
1578 : * This is OK because an outside call always starts from an OP_PHRASE node,
1579 : * and all internal recursion cases pass data != NULL.
1580 : *
1581 : * The detailed semantics of the match data, given that the function returned
1582 : * TS_YES (successful match), are:
1583 : *
1584 : * npos > 0, negate = false:
1585 : * query is matched at specified position(s) (and only those positions)
1586 : * npos > 0, negate = true:
1587 : * query is matched at all positions *except* specified position(s)
1588 : * npos = 0, negate = true:
1589 : * query is matched at all positions
1590 : * npos = 0, negate = false:
1591 : * disallowed (this should result in TS_NO or TS_MAYBE, as appropriate)
1592 : *
1593 : * Successful matches also return a "width" value which is the match width in
1594 : * lexemes, less one. Hence, "width" is zero for simple one-lexeme matches,
1595 : * and is the sum of the phrase operator distances for phrase matches. Note
1596 : * that when width > 0, the listed positions represent the ends of matches not
1597 : * the starts. (This unintuitive rule is needed to avoid possibly generating
1598 : * negative positions, which wouldn't fit into the WordEntryPos arrays.)
1599 : *
1600 : * If the TSExecuteCallback function reports that an operand is present
1601 : * but fails to provide position(s) for it, we will return TS_MAYBE when
1602 : * it is possible but not certain that the query is matched.
1603 : *
1604 : * When the function returns TS_NO or TS_MAYBE, it must return npos = 0,
1605 : * negate = false (which is the state initialized by the caller); but the
1606 : * "width" output in such cases is undefined.
1607 : */
1608 : static TSTernaryValue
1609 704600 : TS_phrase_execute(QueryItem *curitem, void *arg, uint32 flags,
1610 : TSExecuteCallback chkcond,
1611 : ExecPhraseData *data)
1612 : {
1613 : ExecPhraseData Ldata,
1614 : Rdata;
1615 : TSTernaryValue lmatch,
1616 : rmatch;
1617 : int Loffset,
1618 : Roffset,
1619 : maxwidth;
1620 :
1621 : /* since this function recurses, it could be driven to stack overflow */
1622 704600 : check_stack_depth();
1623 :
1624 : /* ... and let's check for query cancel while we're at it */
1625 704600 : CHECK_FOR_INTERRUPTS();
1626 :
1627 704600 : if (curitem->type == QI_VAL)
1628 346652 : return chkcond(arg, (QueryOperand *) curitem, data);
1629 :
1630 357948 : switch (curitem->qoperator.oper)
1631 : {
1632 121174 : case OP_NOT:
1633 :
1634 : /*
1635 : * We need not touch data->width, since a NOT operation does not
1636 : * change the match width.
1637 : */
1638 121174 : if (flags & TS_EXEC_SKIP_NOT)
1639 : {
1640 : /* with SKIP_NOT, report NOT as "match everywhere" */
1641 : Assert(data->npos == 0 && !data->negate);
1642 0 : data->negate = true;
1643 0 : return TS_YES;
1644 : }
1645 121174 : switch (TS_phrase_execute(curitem + 1, arg, flags, chkcond, data))
1646 : {
1647 105896 : case TS_NO:
1648 : /* change "match nowhere" to "match everywhere" */
1649 : Assert(data->npos == 0 && !data->negate);
1650 105896 : data->negate = true;
1651 105896 : return TS_YES;
1652 390 : case TS_YES:
1653 390 : if (data->npos > 0)
1654 : {
1655 : /* we have some positions, invert negate flag */
1656 384 : data->negate = !data->negate;
1657 384 : return TS_YES;
1658 : }
1659 6 : else if (data->negate)
1660 : {
1661 : /* change "match everywhere" to "match nowhere" */
1662 6 : data->negate = false;
1663 6 : return TS_NO;
1664 : }
1665 : /* Should not get here if result was TS_YES */
1666 : Assert(false);
1667 0 : break;
1668 14888 : case TS_MAYBE:
1669 : /* match positions are, and remain, uncertain */
1670 14888 : return TS_MAYBE;
1671 : }
1672 0 : break;
1673 :
1674 236618 : case OP_PHRASE:
1675 : case OP_AND:
1676 236618 : memset(&Ldata, 0, sizeof(Ldata));
1677 236618 : memset(&Rdata, 0, sizeof(Rdata));
1678 :
1679 236618 : lmatch = TS_phrase_execute(curitem + curitem->qoperator.left,
1680 : arg, flags, chkcond, &Ldata);
1681 236618 : if (lmatch == TS_NO)
1682 126488 : return TS_NO;
1683 :
1684 110130 : rmatch = TS_phrase_execute(curitem + 1,
1685 : arg, flags, chkcond, &Rdata);
1686 110130 : if (rmatch == TS_NO)
1687 54446 : return TS_NO;
1688 :
1689 : /*
1690 : * If either operand has no position information, then we can't
1691 : * return reliable position data, only a MAYBE result.
1692 : */
1693 55684 : if (lmatch == TS_MAYBE || rmatch == TS_MAYBE)
1694 25874 : return TS_MAYBE;
1695 :
1696 29810 : if (curitem->qoperator.oper == OP_PHRASE)
1697 : {
1698 : /*
1699 : * Compute Loffset and Roffset suitable for phrase match, and
1700 : * compute overall width of whole phrase match.
1701 : */
1702 29804 : Loffset = curitem->qoperator.distance + Rdata.width;
1703 29804 : Roffset = 0;
1704 29804 : if (data)
1705 186 : data->width = curitem->qoperator.distance +
1706 186 : Ldata.width + Rdata.width;
1707 : }
1708 : else
1709 : {
1710 : /*
1711 : * For OP_AND, set output width and alignment like OP_OR (see
1712 : * comment below)
1713 : */
1714 6 : maxwidth = Max(Ldata.width, Rdata.width);
1715 6 : Loffset = maxwidth - Ldata.width;
1716 6 : Roffset = maxwidth - Rdata.width;
1717 6 : if (data)
1718 6 : data->width = maxwidth;
1719 : }
1720 :
1721 29810 : if (Ldata.negate && Rdata.negate)
1722 : {
1723 : /* !L & !R: treat as !(L | R) */
1724 28436 : (void) TS_phrase_output(data, &Ldata, &Rdata,
1725 : TSPO_BOTH | TSPO_L_ONLY | TSPO_R_ONLY,
1726 : Loffset, Roffset,
1727 28436 : Ldata.npos + Rdata.npos);
1728 28436 : if (data)
1729 0 : data->negate = true;
1730 28436 : return TS_YES;
1731 : }
1732 1374 : else if (Ldata.negate)
1733 : {
1734 : /* !L & R */
1735 450 : return TS_phrase_output(data, &Ldata, &Rdata,
1736 : TSPO_R_ONLY,
1737 : Loffset, Roffset,
1738 : Rdata.npos);
1739 : }
1740 924 : else if (Rdata.negate)
1741 : {
1742 : /* L & !R */
1743 6 : return TS_phrase_output(data, &Ldata, &Rdata,
1744 : TSPO_L_ONLY,
1745 : Loffset, Roffset,
1746 : Ldata.npos);
1747 : }
1748 : else
1749 : {
1750 : /* straight AND */
1751 918 : return TS_phrase_output(data, &Ldata, &Rdata,
1752 : TSPO_BOTH,
1753 : Loffset, Roffset,
1754 918 : Min(Ldata.npos, Rdata.npos));
1755 : }
1756 :
1757 156 : case OP_OR:
1758 156 : memset(&Ldata, 0, sizeof(Ldata));
1759 156 : memset(&Rdata, 0, sizeof(Rdata));
1760 :
1761 156 : lmatch = TS_phrase_execute(curitem + curitem->qoperator.left,
1762 : arg, flags, chkcond, &Ldata);
1763 156 : rmatch = TS_phrase_execute(curitem + 1,
1764 : arg, flags, chkcond, &Rdata);
1765 :
1766 156 : if (lmatch == TS_NO && rmatch == TS_NO)
1767 12 : return TS_NO;
1768 :
1769 : /*
1770 : * If either operand has no position information, then we can't
1771 : * return reliable position data, only a MAYBE result.
1772 : */
1773 144 : if (lmatch == TS_MAYBE || rmatch == TS_MAYBE)
1774 0 : return TS_MAYBE;
1775 :
1776 : /*
1777 : * Cope with undefined output width from failed submatch. (This
1778 : * takes less code than trying to ensure that all failure returns
1779 : * set data->width to zero.)
1780 : */
1781 144 : if (lmatch == TS_NO)
1782 18 : Ldata.width = 0;
1783 144 : if (rmatch == TS_NO)
1784 84 : Rdata.width = 0;
1785 :
1786 : /*
1787 : * For OP_AND and OP_OR, report the width of the wider of the two
1788 : * inputs, and align the narrower input's positions to the right
1789 : * end of that width. This rule deals at least somewhat
1790 : * reasonably with cases like "x <-> (y | z <-> q)".
1791 : */
1792 144 : maxwidth = Max(Ldata.width, Rdata.width);
1793 144 : Loffset = maxwidth - Ldata.width;
1794 144 : Roffset = maxwidth - Rdata.width;
1795 144 : data->width = maxwidth;
1796 :
1797 144 : if (Ldata.negate && Rdata.negate)
1798 : {
1799 : /* !L | !R: treat as !(L & R) */
1800 6 : (void) TS_phrase_output(data, &Ldata, &Rdata,
1801 : TSPO_BOTH,
1802 : Loffset, Roffset,
1803 6 : Min(Ldata.npos, Rdata.npos));
1804 6 : data->negate = true;
1805 6 : return TS_YES;
1806 : }
1807 138 : else if (Ldata.negate)
1808 : {
1809 : /* !L | R: treat as !(L & !R) */
1810 30 : (void) TS_phrase_output(data, &Ldata, &Rdata,
1811 : TSPO_L_ONLY,
1812 : Loffset, Roffset,
1813 : Ldata.npos);
1814 30 : data->negate = true;
1815 30 : return TS_YES;
1816 : }
1817 108 : else if (Rdata.negate)
1818 : {
1819 : /* L | !R: treat as !(!L & R) */
1820 6 : (void) TS_phrase_output(data, &Ldata, &Rdata,
1821 : TSPO_R_ONLY,
1822 : Loffset, Roffset,
1823 : Rdata.npos);
1824 6 : data->negate = true;
1825 6 : return TS_YES;
1826 : }
1827 : else
1828 : {
1829 : /* straight OR */
1830 102 : return TS_phrase_output(data, &Ldata, &Rdata,
1831 : TSPO_BOTH | TSPO_L_ONLY | TSPO_R_ONLY,
1832 : Loffset, Roffset,
1833 102 : Ldata.npos + Rdata.npos);
1834 : }
1835 :
1836 0 : default:
1837 0 : elog(ERROR, "unrecognized operator: %d", curitem->qoperator.oper);
1838 : }
1839 :
1840 : /* not reachable, but keep compiler quiet */
1841 0 : return TS_NO;
1842 : }
1843 :
1844 :
1845 : /*
1846 : * Evaluate tsquery boolean expression.
1847 : *
1848 : * curitem: current tsquery item (initially, the first one)
1849 : * arg: opaque value to pass through to callback function
1850 : * flags: bitmask of flag bits shown in ts_utils.h
1851 : * chkcond: callback function to check whether a primitive value is present
1852 : */
1853 : bool
1854 520692 : TS_execute(QueryItem *curitem, void *arg, uint32 flags,
1855 : TSExecuteCallback chkcond)
1856 : {
1857 : /*
1858 : * If we get TS_MAYBE from the recursion, return true. We could only see
1859 : * that result if the caller passed TS_EXEC_PHRASE_NO_POS, so there's no
1860 : * need to check again.
1861 : */
1862 520692 : return TS_execute_recurse(curitem, arg, flags, chkcond) != TS_NO;
1863 : }
1864 :
1865 : /*
1866 : * Evaluate tsquery boolean expression.
1867 : *
1868 : * This is the same as TS_execute except that TS_MAYBE is returned as-is.
1869 : */
1870 : TSTernaryValue
1871 36942 : TS_execute_ternary(QueryItem *curitem, void *arg, uint32 flags,
1872 : TSExecuteCallback chkcond)
1873 : {
1874 36942 : return TS_execute_recurse(curitem, arg, flags, chkcond);
1875 : }
1876 :
1877 : /*
1878 : * TS_execute recursion for operators above any phrase operator. Here we do
1879 : * not need to worry about lexeme positions. As soon as we hit an OP_PHRASE
1880 : * operator, we pass it off to TS_phrase_execute which does worry.
1881 : */
1882 : static TSTernaryValue
1883 1055428 : TS_execute_recurse(QueryItem *curitem, void *arg, uint32 flags,
1884 : TSExecuteCallback chkcond)
1885 : {
1886 : TSTernaryValue lmatch;
1887 :
1888 : /* since this function recurses, it could be driven to stack overflow */
1889 1055428 : check_stack_depth();
1890 :
1891 : /* ... and let's check for query cancel while we're at it */
1892 1055428 : CHECK_FOR_INTERRUPTS();
1893 :
1894 1055428 : if (curitem->type == QI_VAL)
1895 423312 : return chkcond(arg, (QueryOperand *) curitem,
1896 : NULL /* don't need position info */ );
1897 :
1898 632116 : switch (curitem->qoperator.oper)
1899 : {
1900 203256 : case OP_NOT:
1901 203256 : if (flags & TS_EXEC_SKIP_NOT)
1902 0 : return TS_YES;
1903 203256 : switch (TS_execute_recurse(curitem + 1, arg, flags, chkcond))
1904 : {
1905 191736 : case TS_NO:
1906 191736 : return TS_YES;
1907 4890 : case TS_YES:
1908 4890 : return TS_NO;
1909 6630 : case TS_MAYBE:
1910 6630 : return TS_MAYBE;
1911 : }
1912 0 : break;
1913 :
1914 83588 : case OP_AND:
1915 83588 : lmatch = TS_execute_recurse(curitem + curitem->qoperator.left, arg,
1916 : flags, chkcond);
1917 83588 : if (lmatch == TS_NO)
1918 66390 : return TS_NO;
1919 17198 : switch (TS_execute_recurse(curitem + 1, arg, flags, chkcond))
1920 : {
1921 10110 : case TS_NO:
1922 10110 : return TS_NO;
1923 3300 : case TS_YES:
1924 3300 : return lmatch;
1925 3788 : case TS_MAYBE:
1926 3788 : return TS_MAYBE;
1927 : }
1928 0 : break;
1929 :
1930 108966 : case OP_OR:
1931 108966 : lmatch = TS_execute_recurse(curitem + curitem->qoperator.left, arg,
1932 : flags, chkcond);
1933 108966 : if (lmatch == TS_YES)
1934 24180 : return TS_YES;
1935 84786 : switch (TS_execute_recurse(curitem + 1, arg, flags, chkcond))
1936 : {
1937 57506 : case TS_NO:
1938 57506 : return lmatch;
1939 7416 : case TS_YES:
1940 7416 : return TS_YES;
1941 19864 : case TS_MAYBE:
1942 19864 : return TS_MAYBE;
1943 : }
1944 0 : break;
1945 :
1946 236306 : case OP_PHRASE:
1947 :
1948 : /*
1949 : * If we get a MAYBE result, and the caller doesn't want that,
1950 : * convert it to NO. It would be more consistent, perhaps, to
1951 : * return the result of TS_phrase_execute() verbatim and then
1952 : * convert MAYBE results at the top of the recursion. But
1953 : * converting at the topmost phrase operator gives results that
1954 : * are bug-compatible with the old implementation, so do it like
1955 : * this for now.
1956 : */
1957 236306 : switch (TS_phrase_execute(curitem, arg, flags, chkcond, NULL))
1958 : {
1959 181126 : case TS_NO:
1960 181126 : return TS_NO;
1961 29312 : case TS_YES:
1962 29312 : return TS_YES;
1963 25868 : case TS_MAYBE:
1964 25868 : return (flags & TS_EXEC_PHRASE_NO_POS) ? TS_MAYBE : TS_NO;
1965 : }
1966 0 : break;
1967 :
1968 0 : default:
1969 0 : elog(ERROR, "unrecognized operator: %d", curitem->qoperator.oper);
1970 : }
1971 :
1972 : /* not reachable, but keep compiler quiet */
1973 0 : return TS_NO;
1974 : }
1975 :
1976 : /*
1977 : * Evaluate tsquery and report locations of matching terms.
1978 : *
1979 : * This is like TS_execute except that it returns match locations not just
1980 : * success/failure status. The callback function is required to provide
1981 : * position data (we report failure if it doesn't).
1982 : *
1983 : * On successful match, the result is a List of ExecPhraseData structs, one
1984 : * for each AND'ed term or phrase operator in the query. Each struct includes
1985 : * a sorted array of lexeme positions matching that term. (Recall that for
1986 : * phrase operators, the match includes width+1 lexemes, and the recorded
1987 : * position is that of the rightmost lexeme.)
1988 : *
1989 : * OR subexpressions are handled by union'ing their match locations into a
1990 : * single List element, which is valid since any of those locations contains
1991 : * a match. However, when some of the OR'ed terms are phrase operators, we
1992 : * report the maximum width of any of the OR'ed terms, making such cases
1993 : * slightly imprecise in the conservative direction. (For example, if the
1994 : * tsquery is "(A <-> B) | C", an occurrence of C in the data would be
1995 : * reported as though it includes the lexeme to the left of C.)
1996 : *
1997 : * Locations of NOT subexpressions are not reported. (Obviously, there can
1998 : * be no successful NOT matches at top level, or the match would have failed.
1999 : * So this amounts to ignoring NOTs underneath ORs.)
2000 : *
2001 : * The result is NIL if no match, or if position data was not returned.
2002 : *
2003 : * Arguments are the same as for TS_execute, although flags is currently
2004 : * vestigial since none of the defined bits are sensible here.
2005 : */
2006 : List *
2007 362 : TS_execute_locations(QueryItem *curitem, void *arg,
2008 : uint32 flags,
2009 : TSExecuteCallback chkcond)
2010 : {
2011 : List *result;
2012 :
2013 : /* No flags supported, as yet */
2014 : Assert(flags == TS_EXEC_EMPTY);
2015 362 : if (TS_execute_locations_recurse(curitem, arg, chkcond, &result))
2016 128 : return result;
2017 234 : return NIL;
2018 : }
2019 :
2020 : /*
2021 : * TS_execute_locations recursion for operators above any phrase operator.
2022 : * OP_PHRASE subexpressions can be passed off to TS_phrase_execute.
2023 : */
2024 : static bool
2025 1070 : TS_execute_locations_recurse(QueryItem *curitem, void *arg,
2026 : TSExecuteCallback chkcond,
2027 : List **locations)
2028 : {
2029 : bool lmatch,
2030 : rmatch;
2031 : List *llocations,
2032 : *rlocations;
2033 : ExecPhraseData *data;
2034 :
2035 : /* since this function recurses, it could be driven to stack overflow */
2036 1070 : check_stack_depth();
2037 :
2038 : /* ... and let's check for query cancel while we're at it */
2039 1070 : CHECK_FOR_INTERRUPTS();
2040 :
2041 : /* Default locations result is empty */
2042 1070 : *locations = NIL;
2043 :
2044 1070 : if (curitem->type == QI_VAL)
2045 : {
2046 446 : data = palloc0_object(ExecPhraseData);
2047 446 : if (chkcond(arg, (QueryOperand *) curitem, data) == TS_YES)
2048 : {
2049 212 : *locations = list_make1(data);
2050 212 : return true;
2051 : }
2052 234 : pfree(data);
2053 234 : return false;
2054 : }
2055 :
2056 624 : switch (curitem->qoperator.oper)
2057 : {
2058 12 : case OP_NOT:
2059 12 : if (!TS_execute_locations_recurse(curitem + 1, arg, chkcond,
2060 : &llocations))
2061 0 : return true; /* we don't pass back any locations */
2062 12 : return false;
2063 :
2064 528 : case OP_AND:
2065 528 : if (!TS_execute_locations_recurse(curitem + curitem->qoperator.left,
2066 : arg, chkcond,
2067 : &llocations))
2068 408 : return false;
2069 120 : if (!TS_execute_locations_recurse(curitem + 1,
2070 : arg, chkcond,
2071 : &rlocations))
2072 54 : return false;
2073 66 : *locations = list_concat(llocations, rlocations);
2074 66 : return true;
2075 :
2076 24 : case OP_OR:
2077 24 : lmatch = TS_execute_locations_recurse(curitem + curitem->qoperator.left,
2078 : arg, chkcond,
2079 : &llocations);
2080 24 : rmatch = TS_execute_locations_recurse(curitem + 1,
2081 : arg, chkcond,
2082 : &rlocations);
2083 24 : if (lmatch || rmatch)
2084 : {
2085 : /*
2086 : * We generate an AND'able location struct from each
2087 : * combination of sub-matches, following the disjunctive law
2088 : * (A & B) | (C & D) = (A | C) & (A | D) & (B | C) & (B | D).
2089 : *
2090 : * However, if either input didn't produce locations (i.e., it
2091 : * failed or was a NOT), we must just return the other list.
2092 : */
2093 24 : if (llocations == NIL)
2094 0 : *locations = rlocations;
2095 24 : else if (rlocations == NIL)
2096 12 : *locations = llocations;
2097 : else
2098 : {
2099 : ListCell *ll;
2100 :
2101 24 : foreach(ll, llocations)
2102 : {
2103 12 : ExecPhraseData *ldata = (ExecPhraseData *) lfirst(ll);
2104 : ListCell *lr;
2105 :
2106 24 : foreach(lr, rlocations)
2107 : {
2108 12 : ExecPhraseData *rdata = (ExecPhraseData *) lfirst(lr);
2109 :
2110 12 : data = palloc0_object(ExecPhraseData);
2111 12 : (void) TS_phrase_output(data, ldata, rdata,
2112 : TSPO_BOTH | TSPO_L_ONLY | TSPO_R_ONLY,
2113 : 0, 0,
2114 12 : ldata->npos + rdata->npos);
2115 : /* Report the larger width, as explained above. */
2116 12 : data->width = Max(ldata->width, rdata->width);
2117 12 : *locations = lappend(*locations, data);
2118 : }
2119 : }
2120 : }
2121 :
2122 24 : return true;
2123 : }
2124 0 : return false;
2125 :
2126 60 : case OP_PHRASE:
2127 : /* We can hand this off to TS_phrase_execute */
2128 60 : data = palloc0_object(ExecPhraseData);
2129 60 : if (TS_phrase_execute(curitem, arg, TS_EXEC_EMPTY, chkcond,
2130 : data) == TS_YES)
2131 : {
2132 60 : if (!data->negate)
2133 60 : *locations = list_make1(data);
2134 60 : return true;
2135 : }
2136 0 : pfree(data);
2137 0 : return false;
2138 :
2139 0 : default:
2140 0 : elog(ERROR, "unrecognized operator: %d", curitem->qoperator.oper);
2141 : }
2142 :
2143 : /* not reachable, but keep compiler quiet */
2144 : return false;
2145 : }
2146 :
2147 : /*
2148 : * Detect whether a tsquery boolean expression requires any positive matches
2149 : * to values shown in the tsquery.
2150 : *
2151 : * This is needed to know whether a GIN index search requires full index scan.
2152 : * For example, 'x & !y' requires a match of x, so it's sufficient to scan
2153 : * entries for x; but 'x | !y' could match rows containing neither x nor y.
2154 : */
2155 : bool
2156 834 : tsquery_requires_match(QueryItem *curitem)
2157 : {
2158 : /* since this function recurses, it could be driven to stack overflow */
2159 834 : check_stack_depth();
2160 :
2161 834 : if (curitem->type == QI_VAL)
2162 396 : return true;
2163 :
2164 438 : switch (curitem->qoperator.oper)
2165 : {
2166 168 : case OP_NOT:
2167 :
2168 : /*
2169 : * Assume there are no required matches underneath a NOT. For
2170 : * some cases with nested NOTs, we could prove there's a required
2171 : * match, but it seems unlikely to be worth the trouble.
2172 : */
2173 168 : return false;
2174 :
2175 204 : case OP_PHRASE:
2176 :
2177 : /*
2178 : * Treat OP_PHRASE as OP_AND here
2179 : */
2180 : case OP_AND:
2181 : /* If either side requires a match, we're good */
2182 204 : if (tsquery_requires_match(curitem + curitem->qoperator.left))
2183 156 : return true;
2184 : else
2185 48 : return tsquery_requires_match(curitem + 1);
2186 :
2187 66 : case OP_OR:
2188 : /* Both sides must require a match */
2189 66 : if (tsquery_requires_match(curitem + curitem->qoperator.left))
2190 66 : return tsquery_requires_match(curitem + 1);
2191 : else
2192 0 : return false;
2193 :
2194 0 : default:
2195 0 : elog(ERROR, "unrecognized operator: %d", curitem->qoperator.oper);
2196 : }
2197 :
2198 : /* not reachable, but keep compiler quiet */
2199 : return false;
2200 : }
2201 :
2202 : /*
2203 : * boolean operations
2204 : */
2205 : Datum
2206 60 : ts_match_qv(PG_FUNCTION_ARGS)
2207 : {
2208 60 : PG_RETURN_DATUM(DirectFunctionCall2(ts_match_vq,
2209 : PG_GETARG_DATUM(1),
2210 : PG_GETARG_DATUM(0)));
2211 : }
2212 :
2213 : Datum
2214 220080 : ts_match_vq(PG_FUNCTION_ARGS)
2215 : {
2216 220080 : TSVector val = PG_GETARG_TSVECTOR(0);
2217 220080 : TSQuery query = PG_GETARG_TSQUERY(1);
2218 : CHKVAL chkval;
2219 : bool result;
2220 :
2221 : /* empty query matches nothing */
2222 220080 : if (!query->size)
2223 : {
2224 0 : PG_FREE_IF_COPY(val, 0);
2225 0 : PG_FREE_IF_COPY(query, 1);
2226 0 : PG_RETURN_BOOL(false);
2227 : }
2228 :
2229 220080 : chkval.arrb = ARRPTR(val);
2230 220080 : chkval.arre = chkval.arrb + val->size;
2231 220080 : chkval.values = STRPTR(val);
2232 220080 : chkval.operand = GETOPERAND(query);
2233 220080 : result = TS_execute(GETQUERY(query),
2234 : &chkval,
2235 : TS_EXEC_EMPTY,
2236 : checkcondition_str);
2237 :
2238 220080 : PG_FREE_IF_COPY(val, 0);
2239 220080 : PG_FREE_IF_COPY(query, 1);
2240 220080 : PG_RETURN_BOOL(result);
2241 : }
2242 :
2243 : Datum
2244 0 : ts_match_tt(PG_FUNCTION_ARGS)
2245 : {
2246 : TSVector vector;
2247 : TSQuery query;
2248 : bool res;
2249 :
2250 0 : vector = DatumGetTSVector(DirectFunctionCall1(to_tsvector,
2251 : PG_GETARG_DATUM(0)));
2252 0 : query = DatumGetTSQuery(DirectFunctionCall1(plainto_tsquery,
2253 : PG_GETARG_DATUM(1)));
2254 :
2255 0 : res = DatumGetBool(DirectFunctionCall2(ts_match_vq,
2256 : TSVectorGetDatum(vector),
2257 : TSQueryGetDatum(query)));
2258 :
2259 0 : pfree(vector);
2260 0 : pfree(query);
2261 :
2262 0 : PG_RETURN_BOOL(res);
2263 : }
2264 :
2265 : Datum
2266 0 : ts_match_tq(PG_FUNCTION_ARGS)
2267 : {
2268 : TSVector vector;
2269 0 : TSQuery query = PG_GETARG_TSQUERY(1);
2270 : bool res;
2271 :
2272 0 : vector = DatumGetTSVector(DirectFunctionCall1(to_tsvector,
2273 : PG_GETARG_DATUM(0)));
2274 :
2275 0 : res = DatumGetBool(DirectFunctionCall2(ts_match_vq,
2276 : TSVectorGetDatum(vector),
2277 : TSQueryGetDatum(query)));
2278 :
2279 0 : pfree(vector);
2280 0 : PG_FREE_IF_COPY(query, 1);
2281 :
2282 0 : PG_RETURN_BOOL(res);
2283 : }
2284 :
2285 : /*
2286 : * ts_stat statistic function support
2287 : */
2288 :
2289 :
2290 : /*
2291 : * Returns the number of positions in value 'wptr' within tsvector 'txt',
2292 : * that have a weight equal to one of the weights in 'weight' bitmask.
2293 : */
2294 : static int
2295 8178 : check_weight(TSVector txt, WordEntry *wptr, int8 weight)
2296 : {
2297 8178 : int len = POSDATALEN(txt, wptr);
2298 8178 : int num = 0;
2299 8178 : WordEntryPos *ptr = POSDATAPTR(txt, wptr);
2300 :
2301 16650 : while (len--)
2302 : {
2303 8472 : if (weight & (1 << WEP_GETWEIGHT(*ptr)))
2304 12 : num++;
2305 8472 : ptr++;
2306 : }
2307 8178 : return num;
2308 : }
2309 :
2310 : #define compareStatWord(a,e,t) \
2311 : tsCompareString((a)->lexeme, (a)->lenlexeme, \
2312 : STRPTR(t) + (e)->pos, (e)->len, \
2313 : false)
2314 :
2315 : static void
2316 345624 : insertStatEntry(MemoryContext persistentContext, TSVectorStat *stat, TSVector txt, uint32 off)
2317 : {
2318 345624 : WordEntry *we = ARRPTR(txt) + off;
2319 345624 : StatEntry *node = stat->root,
2320 345624 : *pnode = NULL;
2321 : int n,
2322 345624 : res = 0;
2323 345624 : uint32 depth = 1;
2324 :
2325 345624 : if (stat->weight == 0)
2326 172812 : n = (we->haspos) ? POSDATALEN(txt, we) : 1;
2327 : else
2328 172812 : n = (we->haspos) ? check_weight(txt, we, stat->weight) : 0;
2329 :
2330 345624 : if (n == 0)
2331 172806 : return; /* nothing to insert */
2332 :
2333 1745394 : while (node)
2334 : {
2335 1738530 : res = compareStatWord(node, we, txt);
2336 :
2337 1738530 : if (res == 0)
2338 : {
2339 165954 : break;
2340 : }
2341 : else
2342 : {
2343 1572576 : pnode = node;
2344 1572576 : node = (res < 0) ? node->left : node->right;
2345 : }
2346 1572576 : depth++;
2347 : }
2348 :
2349 172818 : if (depth > stat->maxdepth)
2350 126 : stat->maxdepth = depth;
2351 :
2352 172818 : if (node == NULL)
2353 : {
2354 6864 : node = MemoryContextAlloc(persistentContext, STATENTRYHDRSZ + we->len);
2355 6864 : node->left = node->right = NULL;
2356 6864 : node->ndoc = 1;
2357 6864 : node->nentry = n;
2358 6864 : node->lenlexeme = we->len;
2359 6864 : memcpy(node->lexeme, STRPTR(txt) + we->pos, node->lenlexeme);
2360 :
2361 6864 : if (pnode == NULL)
2362 : {
2363 12 : stat->root = node;
2364 : }
2365 : else
2366 : {
2367 6852 : if (res < 0)
2368 3380 : pnode->left = node;
2369 : else
2370 3472 : pnode->right = node;
2371 : }
2372 : }
2373 : else
2374 : {
2375 165954 : node->ndoc++;
2376 165954 : node->nentry += n;
2377 : }
2378 : }
2379 :
2380 : static void
2381 495384 : chooseNextStatEntry(MemoryContext persistentContext, TSVectorStat *stat, TSVector txt,
2382 : uint32 low, uint32 high, uint32 offset)
2383 : {
2384 : uint32 pos;
2385 495384 : uint32 middle = (low + high) >> 1;
2386 :
2387 495384 : pos = (low + middle) >> 1;
2388 495384 : if (low != middle && pos >= offset && pos - offset < txt->size)
2389 170328 : insertStatEntry(persistentContext, stat, txt, pos - offset);
2390 495384 : pos = (high + middle + 1) >> 1;
2391 495384 : if (middle + 1 != high && pos >= offset && pos - offset < txt->size)
2392 169284 : insertStatEntry(persistentContext, stat, txt, pos - offset);
2393 :
2394 495384 : if (low != middle)
2395 247692 : chooseNextStatEntry(persistentContext, stat, txt, low, middle, offset);
2396 495384 : if (high != middle + 1)
2397 241680 : chooseNextStatEntry(persistentContext, stat, txt, middle + 1, high, offset);
2398 495384 : }
2399 :
2400 : /*
2401 : * This is written like a custom aggregate function, because the
2402 : * original plan was to do just that. Unfortunately, an aggregate function
2403 : * can't return a set, so that plan was abandoned. If that limitation is
2404 : * lifted in the future, ts_stat could be a real aggregate function so that
2405 : * you could use it like this:
2406 : *
2407 : * SELECT ts_stat(vector_column) FROM vector_table;
2408 : *
2409 : * where vector_column is a tsvector-type column in vector_table.
2410 : */
2411 :
2412 : static TSVectorStat *
2413 6108 : ts_accum(MemoryContext persistentContext, TSVectorStat *stat, Datum data)
2414 : {
2415 6108 : TSVector txt = DatumGetTSVector(data);
2416 : uint32 i,
2417 6108 : nbit = 0,
2418 : offset;
2419 :
2420 6108 : if (stat == NULL)
2421 : { /* Init in first */
2422 0 : stat = MemoryContextAllocZero(persistentContext, sizeof(TSVectorStat));
2423 0 : stat->maxdepth = 1;
2424 : }
2425 :
2426 : /* simple check of correctness */
2427 6108 : if (txt == NULL || txt->size == 0)
2428 : {
2429 96 : if (txt && txt != (TSVector) DatumGetPointer(data))
2430 96 : pfree(txt);
2431 96 : return stat;
2432 : }
2433 :
2434 6012 : i = txt->size - 1;
2435 42720 : for (; i > 0; i >>= 1)
2436 36708 : nbit++;
2437 :
2438 6012 : nbit = 1 << nbit;
2439 6012 : offset = (nbit - txt->size) / 2;
2440 :
2441 6012 : insertStatEntry(persistentContext, stat, txt, (nbit >> 1) - offset);
2442 6012 : chooseNextStatEntry(persistentContext, stat, txt, 0, nbit, offset);
2443 :
2444 6012 : return stat;
2445 : }
2446 :
2447 : static void
2448 12 : ts_setup_firstcall(FunctionCallInfo fcinfo, FuncCallContext *funcctx,
2449 : TSVectorStat *stat)
2450 : {
2451 : TupleDesc tupdesc;
2452 : MemoryContext oldcontext;
2453 : StatEntry *node;
2454 :
2455 12 : funcctx->user_fctx = stat;
2456 :
2457 12 : oldcontext = MemoryContextSwitchTo(funcctx->multi_call_memory_ctx);
2458 :
2459 12 : stat->stack = palloc0(sizeof(StatEntry *) * (stat->maxdepth + 1));
2460 12 : stat->stackpos = 0;
2461 :
2462 12 : node = stat->root;
2463 : /* find leftmost value */
2464 12 : if (node == NULL)
2465 0 : stat->stack[stat->stackpos] = NULL;
2466 : else
2467 : for (;;)
2468 : {
2469 48 : stat->stack[stat->stackpos] = node;
2470 48 : if (node->left)
2471 : {
2472 36 : stat->stackpos++;
2473 36 : node = node->left;
2474 : }
2475 : else
2476 12 : break;
2477 : }
2478 : Assert(stat->stackpos <= stat->maxdepth);
2479 :
2480 12 : if (get_call_result_type(fcinfo, NULL, &tupdesc) != TYPEFUNC_COMPOSITE)
2481 0 : elog(ERROR, "return type must be a row type");
2482 12 : funcctx->tuple_desc = tupdesc;
2483 12 : funcctx->attinmeta = TupleDescGetAttInMetadata(tupdesc);
2484 :
2485 12 : MemoryContextSwitchTo(oldcontext);
2486 12 : }
2487 :
2488 : static StatEntry *
2489 13728 : walkStatEntryTree(TSVectorStat *stat)
2490 : {
2491 13728 : StatEntry *node = stat->stack[stat->stackpos];
2492 :
2493 13728 : if (node == NULL)
2494 0 : return NULL;
2495 :
2496 13728 : if (node->ndoc != 0)
2497 : {
2498 : /* return entry itself: we already was at left sublink */
2499 3392 : return node;
2500 : }
2501 10336 : else if (node->right && node->right != stat->stack[stat->stackpos + 1])
2502 : {
2503 : /* go on right sublink */
2504 3472 : stat->stackpos++;
2505 3472 : node = node->right;
2506 :
2507 : /* find most-left value */
2508 : for (;;)
2509 : {
2510 6816 : stat->stack[stat->stackpos] = node;
2511 6816 : if (node->left)
2512 : {
2513 3344 : stat->stackpos++;
2514 3344 : node = node->left;
2515 : }
2516 : else
2517 3472 : break;
2518 : }
2519 3472 : Assert(stat->stackpos <= stat->maxdepth);
2520 : }
2521 : else
2522 : {
2523 : /* we already return all left subtree, itself and right subtree */
2524 6864 : if (stat->stackpos == 0)
2525 12 : return NULL;
2526 :
2527 6852 : stat->stackpos--;
2528 6852 : return walkStatEntryTree(stat);
2529 : }
2530 :
2531 3472 : return node;
2532 : }
2533 :
2534 : static Datum
2535 6876 : ts_process_call(FuncCallContext *funcctx)
2536 : {
2537 : TSVectorStat *st;
2538 : StatEntry *entry;
2539 :
2540 6876 : st = (TSVectorStat *) funcctx->user_fctx;
2541 :
2542 6876 : entry = walkStatEntryTree(st);
2543 :
2544 6876 : if (entry != NULL)
2545 : {
2546 : Datum result;
2547 : char *values[3];
2548 : char ndoc[16];
2549 : char nentry[16];
2550 : HeapTuple tuple;
2551 :
2552 6864 : values[0] = palloc(entry->lenlexeme + 1);
2553 6864 : memcpy(values[0], entry->lexeme, entry->lenlexeme);
2554 6864 : (values[0])[entry->lenlexeme] = '\0';
2555 6864 : sprintf(ndoc, "%d", entry->ndoc);
2556 6864 : values[1] = ndoc;
2557 6864 : sprintf(nentry, "%d", entry->nentry);
2558 6864 : values[2] = nentry;
2559 :
2560 6864 : tuple = BuildTupleFromCStrings(funcctx->attinmeta, values);
2561 6864 : result = HeapTupleGetDatum(tuple);
2562 :
2563 6864 : pfree(values[0]);
2564 :
2565 : /* mark entry as already visited */
2566 6864 : entry->ndoc = 0;
2567 :
2568 6864 : return result;
2569 : }
2570 :
2571 12 : return (Datum) 0;
2572 : }
2573 :
2574 : static TSVectorStat *
2575 12 : ts_stat_sql(MemoryContext persistentContext, text *txt, text *ws)
2576 : {
2577 12 : char *query = text_to_cstring(txt);
2578 : TSVectorStat *stat;
2579 : bool isnull;
2580 : Portal portal;
2581 : SPIPlanPtr plan;
2582 :
2583 12 : if ((plan = SPI_prepare(query, 0, NULL)) == NULL)
2584 : /* internal error */
2585 0 : elog(ERROR, "SPI_prepare(\"%s\") failed", query);
2586 :
2587 12 : if ((portal = SPI_cursor_open(NULL, plan, NULL, NULL, true)) == NULL)
2588 : /* internal error */
2589 0 : elog(ERROR, "SPI_cursor_open(\"%s\") failed", query);
2590 :
2591 12 : SPI_cursor_fetch(portal, true, 100);
2592 :
2593 12 : if (SPI_tuptable == NULL ||
2594 12 : SPI_tuptable->tupdesc->natts != 1 ||
2595 12 : !IsBinaryCoercible(SPI_gettypeid(SPI_tuptable->tupdesc, 1),
2596 : TSVECTOROID))
2597 0 : ereport(ERROR,
2598 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
2599 : errmsg("ts_stat query must return one tsvector column")));
2600 :
2601 12 : stat = MemoryContextAllocZero(persistentContext, sizeof(TSVectorStat));
2602 12 : stat->maxdepth = 1;
2603 :
2604 12 : if (ws)
2605 : {
2606 : char *buf;
2607 :
2608 6 : buf = VARDATA_ANY(ws);
2609 18 : while (buf - VARDATA_ANY(ws) < VARSIZE_ANY_EXHDR(ws))
2610 : {
2611 12 : if (pg_mblen(buf) == 1)
2612 : {
2613 12 : switch (*buf)
2614 : {
2615 6 : case 'A':
2616 : case 'a':
2617 6 : stat->weight |= 1 << 3;
2618 6 : break;
2619 6 : case 'B':
2620 : case 'b':
2621 6 : stat->weight |= 1 << 2;
2622 6 : break;
2623 0 : case 'C':
2624 : case 'c':
2625 0 : stat->weight |= 1 << 1;
2626 0 : break;
2627 0 : case 'D':
2628 : case 'd':
2629 0 : stat->weight |= 1;
2630 0 : break;
2631 0 : default:
2632 0 : stat->weight |= 0;
2633 : }
2634 : }
2635 12 : buf += pg_mblen(buf);
2636 : }
2637 : }
2638 :
2639 84 : while (SPI_processed > 0)
2640 : {
2641 : uint64 i;
2642 :
2643 6180 : for (i = 0; i < SPI_processed; i++)
2644 : {
2645 6108 : Datum data = SPI_getbinval(SPI_tuptable->vals[i], SPI_tuptable->tupdesc, 1, &isnull);
2646 :
2647 6108 : if (!isnull)
2648 6108 : stat = ts_accum(persistentContext, stat, data);
2649 : }
2650 :
2651 72 : SPI_freetuptable(SPI_tuptable);
2652 72 : SPI_cursor_fetch(portal, true, 100);
2653 : }
2654 :
2655 12 : SPI_freetuptable(SPI_tuptable);
2656 12 : SPI_cursor_close(portal);
2657 12 : SPI_freeplan(plan);
2658 12 : pfree(query);
2659 :
2660 12 : return stat;
2661 : }
2662 :
2663 : Datum
2664 6864 : ts_stat1(PG_FUNCTION_ARGS)
2665 : {
2666 : FuncCallContext *funcctx;
2667 : Datum result;
2668 :
2669 6864 : if (SRF_IS_FIRSTCALL())
2670 : {
2671 : TSVectorStat *stat;
2672 6 : text *txt = PG_GETARG_TEXT_PP(0);
2673 :
2674 6 : funcctx = SRF_FIRSTCALL_INIT();
2675 6 : SPI_connect();
2676 6 : stat = ts_stat_sql(funcctx->multi_call_memory_ctx, txt, NULL);
2677 6 : PG_FREE_IF_COPY(txt, 0);
2678 6 : ts_setup_firstcall(fcinfo, funcctx, stat);
2679 6 : SPI_finish();
2680 : }
2681 :
2682 6864 : funcctx = SRF_PERCALL_SETUP();
2683 6864 : if ((result = ts_process_call(funcctx)) != (Datum) 0)
2684 6858 : SRF_RETURN_NEXT(funcctx, result);
2685 6 : SRF_RETURN_DONE(funcctx);
2686 : }
2687 :
2688 : Datum
2689 12 : ts_stat2(PG_FUNCTION_ARGS)
2690 : {
2691 : FuncCallContext *funcctx;
2692 : Datum result;
2693 :
2694 12 : if (SRF_IS_FIRSTCALL())
2695 : {
2696 : TSVectorStat *stat;
2697 6 : text *txt = PG_GETARG_TEXT_PP(0);
2698 6 : text *ws = PG_GETARG_TEXT_PP(1);
2699 :
2700 6 : funcctx = SRF_FIRSTCALL_INIT();
2701 6 : SPI_connect();
2702 6 : stat = ts_stat_sql(funcctx->multi_call_memory_ctx, txt, ws);
2703 6 : PG_FREE_IF_COPY(txt, 0);
2704 6 : PG_FREE_IF_COPY(ws, 1);
2705 6 : ts_setup_firstcall(fcinfo, funcctx, stat);
2706 6 : SPI_finish();
2707 : }
2708 :
2709 12 : funcctx = SRF_PERCALL_SETUP();
2710 12 : if ((result = ts_process_call(funcctx)) != (Datum) 0)
2711 6 : SRF_RETURN_NEXT(funcctx, result);
2712 6 : SRF_RETURN_DONE(funcctx);
2713 : }
2714 :
2715 :
2716 : /*
2717 : * Triggers for automatic update of a tsvector column from text column(s)
2718 : *
2719 : * Trigger arguments are either
2720 : * name of tsvector col, name of tsconfig to use, name(s) of text col(s)
2721 : * name of tsvector col, name of regconfig col, name(s) of text col(s)
2722 : * ie, tsconfig can either be specified by name, or indirectly as the
2723 : * contents of a regconfig field in the row. If the name is used, it must
2724 : * be explicitly schema-qualified.
2725 : */
2726 : Datum
2727 18 : tsvector_update_trigger_byid(PG_FUNCTION_ARGS)
2728 : {
2729 18 : return tsvector_update_trigger(fcinfo, false);
2730 : }
2731 :
2732 : Datum
2733 0 : tsvector_update_trigger_bycolumn(PG_FUNCTION_ARGS)
2734 : {
2735 0 : return tsvector_update_trigger(fcinfo, true);
2736 : }
2737 :
2738 : static Datum
2739 18 : tsvector_update_trigger(PG_FUNCTION_ARGS, bool config_column)
2740 : {
2741 : TriggerData *trigdata;
2742 : Trigger *trigger;
2743 : Relation rel;
2744 18 : HeapTuple rettuple = NULL;
2745 : int tsvector_attr_num,
2746 : i;
2747 : ParsedText prs;
2748 : Datum datum;
2749 : bool isnull;
2750 : text *txt;
2751 : Oid cfgId;
2752 : bool update_needed;
2753 :
2754 : /* Check call context */
2755 18 : if (!CALLED_AS_TRIGGER(fcinfo)) /* internal error */
2756 0 : elog(ERROR, "tsvector_update_trigger: not fired by trigger manager");
2757 :
2758 18 : trigdata = (TriggerData *) fcinfo->context;
2759 18 : if (!TRIGGER_FIRED_FOR_ROW(trigdata->tg_event))
2760 0 : elog(ERROR, "tsvector_update_trigger: must be fired for row");
2761 18 : if (!TRIGGER_FIRED_BEFORE(trigdata->tg_event))
2762 0 : elog(ERROR, "tsvector_update_trigger: must be fired BEFORE event");
2763 :
2764 18 : if (TRIGGER_FIRED_BY_INSERT(trigdata->tg_event))
2765 : {
2766 12 : rettuple = trigdata->tg_trigtuple;
2767 12 : update_needed = true;
2768 : }
2769 6 : else if (TRIGGER_FIRED_BY_UPDATE(trigdata->tg_event))
2770 : {
2771 6 : rettuple = trigdata->tg_newtuple;
2772 6 : update_needed = false; /* computed below */
2773 : }
2774 : else
2775 0 : elog(ERROR, "tsvector_update_trigger: must be fired for INSERT or UPDATE");
2776 :
2777 18 : trigger = trigdata->tg_trigger;
2778 18 : rel = trigdata->tg_relation;
2779 :
2780 18 : if (trigger->tgnargs < 3)
2781 0 : elog(ERROR, "tsvector_update_trigger: arguments must be tsvector_field, ts_config, text_field1, ...)");
2782 :
2783 : /* Find the target tsvector column */
2784 18 : tsvector_attr_num = SPI_fnumber(rel->rd_att, trigger->tgargs[0]);
2785 18 : if (tsvector_attr_num == SPI_ERROR_NOATTRIBUTE)
2786 0 : ereport(ERROR,
2787 : (errcode(ERRCODE_UNDEFINED_COLUMN),
2788 : errmsg("tsvector column \"%s\" does not exist",
2789 : trigger->tgargs[0])));
2790 : /* This will effectively reject system columns, so no separate test: */
2791 18 : if (!IsBinaryCoercible(SPI_gettypeid(rel->rd_att, tsvector_attr_num),
2792 : TSVECTOROID))
2793 0 : ereport(ERROR,
2794 : (errcode(ERRCODE_DATATYPE_MISMATCH),
2795 : errmsg("column \"%s\" is not of tsvector type",
2796 : trigger->tgargs[0])));
2797 :
2798 : /* Find the configuration to use */
2799 18 : if (config_column)
2800 : {
2801 : int config_attr_num;
2802 :
2803 0 : config_attr_num = SPI_fnumber(rel->rd_att, trigger->tgargs[1]);
2804 0 : if (config_attr_num == SPI_ERROR_NOATTRIBUTE)
2805 0 : ereport(ERROR,
2806 : (errcode(ERRCODE_UNDEFINED_COLUMN),
2807 : errmsg("configuration column \"%s\" does not exist",
2808 : trigger->tgargs[1])));
2809 0 : if (!IsBinaryCoercible(SPI_gettypeid(rel->rd_att, config_attr_num),
2810 : REGCONFIGOID))
2811 0 : ereport(ERROR,
2812 : (errcode(ERRCODE_DATATYPE_MISMATCH),
2813 : errmsg("column \"%s\" is not of regconfig type",
2814 : trigger->tgargs[1])));
2815 :
2816 0 : datum = SPI_getbinval(rettuple, rel->rd_att, config_attr_num, &isnull);
2817 0 : if (isnull)
2818 0 : ereport(ERROR,
2819 : (errcode(ERRCODE_NULL_VALUE_NOT_ALLOWED),
2820 : errmsg("configuration column \"%s\" must not be null",
2821 : trigger->tgargs[1])));
2822 0 : cfgId = DatumGetObjectId(datum);
2823 : }
2824 : else
2825 : {
2826 : List *names;
2827 :
2828 18 : names = stringToQualifiedNameList(trigger->tgargs[1], NULL);
2829 : /* require a schema so that results are not search path dependent */
2830 18 : if (list_length(names) < 2)
2831 0 : ereport(ERROR,
2832 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
2833 : errmsg("text search configuration name \"%s\" must be schema-qualified",
2834 : trigger->tgargs[1])));
2835 18 : cfgId = get_ts_config_oid(names, false);
2836 : }
2837 :
2838 : /* initialize parse state */
2839 18 : prs.lenwords = 32;
2840 18 : prs.curwords = 0;
2841 18 : prs.pos = 0;
2842 18 : prs.words = (ParsedWord *) palloc(sizeof(ParsedWord) * prs.lenwords);
2843 :
2844 : /* find all words in indexable column(s) */
2845 36 : for (i = 2; i < trigger->tgnargs; i++)
2846 : {
2847 : int numattr;
2848 :
2849 18 : numattr = SPI_fnumber(rel->rd_att, trigger->tgargs[i]);
2850 18 : if (numattr == SPI_ERROR_NOATTRIBUTE)
2851 0 : ereport(ERROR,
2852 : (errcode(ERRCODE_UNDEFINED_COLUMN),
2853 : errmsg("column \"%s\" does not exist",
2854 : trigger->tgargs[i])));
2855 18 : if (!IsBinaryCoercible(SPI_gettypeid(rel->rd_att, numattr), TEXTOID))
2856 0 : ereport(ERROR,
2857 : (errcode(ERRCODE_DATATYPE_MISMATCH),
2858 : errmsg("column \"%s\" is not of a character type",
2859 : trigger->tgargs[i])));
2860 :
2861 18 : if (bms_is_member(numattr - FirstLowInvalidHeapAttributeNumber, trigdata->tg_updatedcols))
2862 6 : update_needed = true;
2863 :
2864 18 : datum = SPI_getbinval(rettuple, rel->rd_att, numattr, &isnull);
2865 18 : if (isnull)
2866 6 : continue;
2867 :
2868 12 : txt = DatumGetTextPP(datum);
2869 :
2870 12 : parsetext(cfgId, &prs, VARDATA_ANY(txt), VARSIZE_ANY_EXHDR(txt));
2871 :
2872 12 : if (txt != (text *) DatumGetPointer(datum))
2873 0 : pfree(txt);
2874 : }
2875 :
2876 18 : if (update_needed)
2877 : {
2878 : /* make tsvector value */
2879 18 : datum = TSVectorGetDatum(make_tsvector(&prs));
2880 18 : isnull = false;
2881 :
2882 : /* and insert it into tuple */
2883 18 : rettuple = heap_modify_tuple_by_cols(rettuple, rel->rd_att,
2884 : 1, &tsvector_attr_num,
2885 : &datum, &isnull);
2886 :
2887 18 : pfree(DatumGetPointer(datum));
2888 : }
2889 :
2890 18 : return PointerGetDatum(rettuple);
2891 : }
|
