Line data Source code
1 : /*-------------------------------------------------------------------------
2 : *
3 : * tsvector_op.c
4 : * operations over tsvector
5 : *
6 : * Portions Copyright (c) 1996-2023, 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 "executor/spi.h"
23 : #include "funcapi.h"
24 : #include "lib/qunique.h"
25 : #include "mb/pg_wchar.h"
26 : #include "miscadmin.h"
27 : #include "parser/parse_coerce.h"
28 : #include "tsearch/ts_utils.h"
29 : #include "utils/array.h"
30 : #include "utils/builtins.h"
31 : #include "utils/lsyscache.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 : if (a == b)
439 12 : return 0;
440 66 : return (a > b) ? 1 : -1;
441 : }
442 :
443 : static int
444 102 : compare_text_lexemes(const void *va, const void *vb)
445 : {
446 102 : Datum a = *((const Datum *) va);
447 102 : Datum b = *((const Datum *) vb);
448 102 : char *alex = VARDATA_ANY(a);
449 102 : int alex_len = VARSIZE_ANY_EXHDR(a);
450 102 : char *blex = VARDATA_ANY(b);
451 102 : int blex_len = VARSIZE_ANY_EXHDR(b);
452 :
453 102 : return tsCompareString(alex, alex_len, blex, blex_len, false);
454 : }
455 :
456 : /*
457 : * Internal routine to delete lexemes from TSVector by array of offsets.
458 : *
459 : * int *indices_to_delete -- array of lexeme offsets to delete (modified here!)
460 : * int indices_count -- size of that array
461 : *
462 : * Returns new TSVector without given lexemes along with their positions
463 : * and weights.
464 : */
465 : static TSVector
466 66 : tsvector_delete_by_indices(TSVector tsv, int *indices_to_delete,
467 : int indices_count)
468 : {
469 : TSVector tsout;
470 66 : WordEntry *arrin = ARRPTR(tsv),
471 : *arrout;
472 66 : char *data = STRPTR(tsv),
473 : *dataout;
474 : int i, /* index in arrin */
475 : j, /* index in arrout */
476 : k, /* index in indices_to_delete */
477 : curoff; /* index in dataout area */
478 :
479 : /*
480 : * Sort the filter array to simplify membership checks below. Also, get
481 : * rid of any duplicate entries, so that we can assume that indices_count
482 : * is exactly equal to the number of lexemes that will be removed.
483 : */
484 66 : if (indices_count > 1)
485 : {
486 30 : qsort(indices_to_delete, indices_count, sizeof(int), compare_int);
487 30 : indices_count = qunique(indices_to_delete, indices_count, sizeof(int),
488 : compare_int);
489 : }
490 :
491 : /*
492 : * Here we overestimate tsout size, since we don't know how much space is
493 : * used by the deleted lexeme(s). We will set exact size below.
494 : */
495 66 : tsout = (TSVector) palloc0(VARSIZE(tsv));
496 :
497 : /* This count must be correct because STRPTR(tsout) relies on it. */
498 66 : tsout->size = tsv->size - indices_count;
499 :
500 : /*
501 : * Copy tsv to tsout, skipping lexemes listed in indices_to_delete.
502 : */
503 66 : arrout = ARRPTR(tsout);
504 66 : dataout = STRPTR(tsout);
505 66 : curoff = 0;
506 396 : for (i = j = k = 0; i < tsv->size; i++)
507 : {
508 : /*
509 : * If current i is present in indices_to_delete, skip this lexeme.
510 : * Since indices_to_delete is already sorted, we only need to check
511 : * the current (k'th) entry.
512 : */
513 330 : if (k < indices_count && i == indices_to_delete[k])
514 : {
515 96 : k++;
516 96 : continue;
517 : }
518 :
519 : /* Copy lexeme and its positions and weights */
520 234 : memcpy(dataout + curoff, data + arrin[i].pos, arrin[i].len);
521 234 : arrout[j].haspos = arrin[i].haspos;
522 234 : arrout[j].len = arrin[i].len;
523 234 : arrout[j].pos = curoff;
524 234 : curoff += arrin[i].len;
525 234 : if (arrin[i].haspos)
526 : {
527 156 : int len = POSDATALEN(tsv, arrin + i) * sizeof(WordEntryPos)
528 156 : + sizeof(uint16);
529 :
530 156 : curoff = SHORTALIGN(curoff);
531 156 : memcpy(dataout + curoff,
532 156 : STRPTR(tsv) + SHORTALIGN(arrin[i].pos + arrin[i].len),
533 : len);
534 156 : curoff += len;
535 : }
536 :
537 234 : j++;
538 : }
539 :
540 : /*
541 : * k should now be exactly equal to indices_count. If it isn't then the
542 : * caller provided us with indices outside of [0, tsv->size) range and
543 : * estimation of tsout's size is wrong.
544 : */
545 : Assert(k == indices_count);
546 :
547 66 : SET_VARSIZE(tsout, CALCDATASIZE(tsout->size, curoff));
548 66 : return tsout;
549 : }
550 :
551 : /*
552 : * Delete given lexeme from tsvector.
553 : * Implementation of user-level ts_delete(tsvector, text).
554 : */
555 : Datum
556 36 : tsvector_delete_str(PG_FUNCTION_ARGS)
557 : {
558 36 : TSVector tsin = PG_GETARG_TSVECTOR(0),
559 : tsout;
560 36 : text *tlexeme = PG_GETARG_TEXT_PP(1);
561 36 : char *lexeme = VARDATA_ANY(tlexeme);
562 36 : int lexeme_len = VARSIZE_ANY_EXHDR(tlexeme),
563 : skip_index;
564 :
565 36 : if ((skip_index = tsvector_bsearch(tsin, lexeme, lexeme_len)) == -1)
566 12 : PG_RETURN_POINTER(tsin);
567 :
568 24 : tsout = tsvector_delete_by_indices(tsin, &skip_index, 1);
569 :
570 24 : PG_FREE_IF_COPY(tsin, 0);
571 24 : PG_FREE_IF_COPY(tlexeme, 1);
572 24 : PG_RETURN_POINTER(tsout);
573 : }
574 :
575 : /*
576 : * Delete given array of lexemes from tsvector.
577 : * Implementation of user-level ts_delete(tsvector, text[]).
578 : */
579 : Datum
580 42 : tsvector_delete_arr(PG_FUNCTION_ARGS)
581 : {
582 42 : TSVector tsin = PG_GETARG_TSVECTOR(0),
583 : tsout;
584 42 : ArrayType *lexemes = PG_GETARG_ARRAYTYPE_P(1);
585 : int i,
586 : nlex,
587 : skip_count,
588 : *skip_indices;
589 : Datum *dlexemes;
590 : bool *nulls;
591 :
592 42 : deconstruct_array_builtin(lexemes, TEXTOID, &dlexemes, &nulls, &nlex);
593 :
594 : /*
595 : * In typical use case array of lexemes to delete is relatively small. So
596 : * here we optimize things for that scenario: iterate through lexarr
597 : * performing binary search of each lexeme from lexarr in tsvector.
598 : */
599 42 : skip_indices = palloc0(nlex * sizeof(int));
600 168 : for (i = skip_count = 0; i < nlex; i++)
601 : {
602 : char *lex;
603 : int lex_len,
604 : lex_pos;
605 :
606 : /* Ignore null array elements, they surely don't match */
607 126 : if (nulls[i])
608 6 : continue;
609 :
610 120 : lex = VARDATA(dlexemes[i]);
611 120 : lex_len = VARSIZE(dlexemes[i]) - VARHDRSZ;
612 120 : lex_pos = tsvector_bsearch(tsin, lex, lex_len);
613 :
614 120 : if (lex_pos >= 0)
615 78 : skip_indices[skip_count++] = lex_pos;
616 : }
617 :
618 42 : tsout = tsvector_delete_by_indices(tsin, skip_indices, skip_count);
619 :
620 42 : pfree(skip_indices);
621 42 : PG_FREE_IF_COPY(tsin, 0);
622 42 : PG_FREE_IF_COPY(lexemes, 1);
623 :
624 42 : PG_RETURN_POINTER(tsout);
625 : }
626 :
627 : /*
628 : * Expand tsvector as table with following columns:
629 : * lexeme: lexeme text
630 : * positions: integer array of lexeme positions
631 : * weights: char array of weights corresponding to positions
632 : */
633 : Datum
634 180 : tsvector_unnest(PG_FUNCTION_ARGS)
635 : {
636 : FuncCallContext *funcctx;
637 : TSVector tsin;
638 :
639 180 : if (SRF_IS_FIRSTCALL())
640 : {
641 : MemoryContext oldcontext;
642 : TupleDesc tupdesc;
643 :
644 30 : funcctx = SRF_FIRSTCALL_INIT();
645 30 : oldcontext = MemoryContextSwitchTo(funcctx->multi_call_memory_ctx);
646 :
647 30 : tupdesc = CreateTemplateTupleDesc(3);
648 30 : TupleDescInitEntry(tupdesc, (AttrNumber) 1, "lexeme",
649 : TEXTOID, -1, 0);
650 30 : TupleDescInitEntry(tupdesc, (AttrNumber) 2, "positions",
651 : INT2ARRAYOID, -1, 0);
652 30 : TupleDescInitEntry(tupdesc, (AttrNumber) 3, "weights",
653 : TEXTARRAYOID, -1, 0);
654 30 : if (get_call_result_type(fcinfo, NULL, &tupdesc) != TYPEFUNC_COMPOSITE)
655 0 : elog(ERROR, "return type must be a row type");
656 30 : funcctx->tuple_desc = tupdesc;
657 :
658 30 : funcctx->user_fctx = PG_GETARG_TSVECTOR_COPY(0);
659 :
660 30 : MemoryContextSwitchTo(oldcontext);
661 : }
662 :
663 180 : funcctx = SRF_PERCALL_SETUP();
664 180 : tsin = (TSVector) funcctx->user_fctx;
665 :
666 180 : if (funcctx->call_cntr < tsin->size)
667 : {
668 150 : WordEntry *arrin = ARRPTR(tsin);
669 150 : char *data = STRPTR(tsin);
670 : HeapTuple tuple;
671 : int j,
672 150 : i = funcctx->call_cntr;
673 150 : bool nulls[] = {false, false, false};
674 : Datum values[3];
675 :
676 150 : values[0] = PointerGetDatum(cstring_to_text_with_len(data + arrin[i].pos, arrin[i].len));
677 :
678 150 : if (arrin[i].haspos)
679 : {
680 : WordEntryPosVector *posv;
681 : Datum *positions;
682 : Datum *weights;
683 : char weight;
684 :
685 : /*
686 : * Internally tsvector stores position and weight in the same
687 : * uint16 (2 bits for weight, 14 for position). Here we extract
688 : * that in two separate arrays.
689 : */
690 90 : posv = _POSVECPTR(tsin, arrin + i);
691 90 : positions = palloc(posv->npos * sizeof(Datum));
692 90 : weights = palloc(posv->npos * sizeof(Datum));
693 252 : for (j = 0; j < posv->npos; j++)
694 : {
695 162 : positions[j] = Int16GetDatum(WEP_GETPOS(posv->pos[j]));
696 162 : weight = 'D' - WEP_GETWEIGHT(posv->pos[j]);
697 162 : weights[j] = PointerGetDatum(cstring_to_text_with_len(&weight,
698 : 1));
699 : }
700 :
701 90 : values[1] = PointerGetDatum(construct_array_builtin(positions, posv->npos, INT2OID));
702 90 : values[2] = PointerGetDatum(construct_array_builtin(weights, posv->npos, TEXTOID));
703 : }
704 : else
705 : {
706 60 : nulls[1] = nulls[2] = true;
707 : }
708 :
709 150 : tuple = heap_form_tuple(funcctx->tuple_desc, values, nulls);
710 150 : SRF_RETURN_NEXT(funcctx, HeapTupleGetDatum(tuple));
711 : }
712 : else
713 : {
714 30 : SRF_RETURN_DONE(funcctx);
715 : }
716 : }
717 :
718 : /*
719 : * Convert tsvector to array of lexemes.
720 : */
721 : Datum
722 12 : tsvector_to_array(PG_FUNCTION_ARGS)
723 : {
724 12 : TSVector tsin = PG_GETARG_TSVECTOR(0);
725 12 : WordEntry *arrin = ARRPTR(tsin);
726 : Datum *elements;
727 : int i;
728 : ArrayType *array;
729 :
730 12 : elements = palloc(tsin->size * sizeof(Datum));
731 :
732 72 : for (i = 0; i < tsin->size; i++)
733 : {
734 60 : elements[i] = PointerGetDatum(cstring_to_text_with_len(STRPTR(tsin) + arrin[i].pos,
735 60 : arrin[i].len));
736 : }
737 :
738 12 : array = construct_array_builtin(elements, tsin->size, TEXTOID);
739 :
740 12 : pfree(elements);
741 12 : PG_FREE_IF_COPY(tsin, 0);
742 12 : PG_RETURN_POINTER(array);
743 : }
744 :
745 : /*
746 : * Build tsvector from array of lexemes.
747 : */
748 : Datum
749 24 : array_to_tsvector(PG_FUNCTION_ARGS)
750 : {
751 24 : ArrayType *v = PG_GETARG_ARRAYTYPE_P(0);
752 : TSVector tsout;
753 : Datum *dlexemes;
754 : WordEntry *arrout;
755 : bool *nulls;
756 : int nitems,
757 : i,
758 : tslen,
759 24 : datalen = 0;
760 : char *cur;
761 :
762 24 : deconstruct_array_builtin(v, TEXTOID, &dlexemes, &nulls, &nitems);
763 :
764 : /*
765 : * Reject nulls and zero length strings (maybe we should just ignore them,
766 : * instead?)
767 : */
768 126 : for (i = 0; i < nitems; i++)
769 : {
770 114 : if (nulls[i])
771 6 : ereport(ERROR,
772 : (errcode(ERRCODE_NULL_VALUE_NOT_ALLOWED),
773 : errmsg("lexeme array may not contain nulls")));
774 :
775 108 : if (VARSIZE(dlexemes[i]) - VARHDRSZ == 0)
776 6 : ereport(ERROR,
777 : (errcode(ERRCODE_ZERO_LENGTH_CHARACTER_STRING),
778 : errmsg("lexeme array may not contain empty strings")));
779 : }
780 :
781 : /* Sort and de-dup, because this is required for a valid tsvector. */
782 12 : if (nitems > 1)
783 : {
784 12 : qsort(dlexemes, nitems, sizeof(Datum), compare_text_lexemes);
785 12 : nitems = qunique(dlexemes, nitems, sizeof(Datum),
786 : compare_text_lexemes);
787 : }
788 :
789 : /* Calculate space needed for surviving lexemes. */
790 60 : for (i = 0; i < nitems; i++)
791 48 : datalen += VARSIZE(dlexemes[i]) - VARHDRSZ;
792 12 : tslen = CALCDATASIZE(nitems, datalen);
793 :
794 : /* Allocate and fill tsvector. */
795 12 : tsout = (TSVector) palloc0(tslen);
796 12 : SET_VARSIZE(tsout, tslen);
797 12 : tsout->size = nitems;
798 :
799 12 : arrout = ARRPTR(tsout);
800 12 : cur = STRPTR(tsout);
801 60 : for (i = 0; i < nitems; i++)
802 : {
803 48 : char *lex = VARDATA(dlexemes[i]);
804 48 : int lex_len = VARSIZE(dlexemes[i]) - VARHDRSZ;
805 :
806 48 : memcpy(cur, lex, lex_len);
807 48 : arrout[i].haspos = 0;
808 48 : arrout[i].len = lex_len;
809 48 : arrout[i].pos = cur - STRPTR(tsout);
810 48 : cur += lex_len;
811 : }
812 :
813 12 : PG_FREE_IF_COPY(v, 0);
814 12 : PG_RETURN_POINTER(tsout);
815 : }
816 :
817 : /*
818 : * ts_filter(): keep only lexemes with given weights in tsvector.
819 : */
820 : Datum
821 18 : tsvector_filter(PG_FUNCTION_ARGS)
822 : {
823 18 : TSVector tsin = PG_GETARG_TSVECTOR(0),
824 : tsout;
825 18 : ArrayType *weights = PG_GETARG_ARRAYTYPE_P(1);
826 18 : WordEntry *arrin = ARRPTR(tsin),
827 : *arrout;
828 18 : char *datain = STRPTR(tsin),
829 : *dataout;
830 : Datum *dweights;
831 : bool *nulls;
832 : int nweights;
833 : int i,
834 : j;
835 18 : int cur_pos = 0;
836 18 : char mask = 0;
837 :
838 18 : deconstruct_array_builtin(weights, CHAROID, &dweights, &nulls, &nweights);
839 :
840 42 : for (i = 0; i < nweights; i++)
841 : {
842 : char char_weight;
843 :
844 30 : if (nulls[i])
845 6 : ereport(ERROR,
846 : (errcode(ERRCODE_NULL_VALUE_NOT_ALLOWED),
847 : errmsg("weight array may not contain nulls")));
848 :
849 24 : char_weight = DatumGetChar(dweights[i]);
850 24 : switch (char_weight)
851 : {
852 18 : case 'A':
853 : case 'a':
854 18 : mask = mask | 8;
855 18 : break;
856 6 : case 'B':
857 : case 'b':
858 6 : mask = mask | 4;
859 6 : break;
860 0 : case 'C':
861 : case 'c':
862 0 : mask = mask | 2;
863 0 : break;
864 0 : case 'D':
865 : case 'd':
866 0 : mask = mask | 1;
867 0 : break;
868 0 : default:
869 0 : ereport(ERROR,
870 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
871 : errmsg("unrecognized weight: \"%c\"", char_weight)));
872 : }
873 : }
874 :
875 12 : tsout = (TSVector) palloc0(VARSIZE(tsin));
876 12 : tsout->size = tsin->size;
877 12 : arrout = ARRPTR(tsout);
878 12 : dataout = STRPTR(tsout);
879 :
880 108 : for (i = j = 0; i < tsin->size; i++)
881 : {
882 : WordEntryPosVector *posvin,
883 : *posvout;
884 96 : int npos = 0;
885 : int k;
886 :
887 96 : if (!arrin[i].haspos)
888 30 : continue;
889 :
890 66 : posvin = _POSVECPTR(tsin, arrin + i);
891 66 : posvout = (WordEntryPosVector *)
892 66 : (dataout + SHORTALIGN(cur_pos + arrin[i].len));
893 :
894 132 : for (k = 0; k < posvin->npos; k++)
895 : {
896 66 : if (mask & (1 << WEP_GETWEIGHT(posvin->pos[k])))
897 30 : posvout->pos[npos++] = posvin->pos[k];
898 : }
899 :
900 : /* if no satisfactory positions found, skip lexeme */
901 66 : if (!npos)
902 36 : continue;
903 :
904 30 : arrout[j].haspos = true;
905 30 : arrout[j].len = arrin[i].len;
906 30 : arrout[j].pos = cur_pos;
907 :
908 30 : memcpy(dataout + cur_pos, datain + arrin[i].pos, arrin[i].len);
909 30 : posvout->npos = npos;
910 30 : cur_pos += SHORTALIGN(arrin[i].len);
911 30 : cur_pos += POSDATALEN(tsout, arrout + j) * sizeof(WordEntryPos) +
912 : sizeof(uint16);
913 30 : j++;
914 : }
915 :
916 12 : tsout->size = j;
917 12 : if (dataout != STRPTR(tsout))
918 12 : memmove(STRPTR(tsout), dataout, cur_pos);
919 :
920 12 : SET_VARSIZE(tsout, CALCDATASIZE(tsout->size, cur_pos));
921 :
922 12 : PG_FREE_IF_COPY(tsin, 0);
923 12 : PG_RETURN_POINTER(tsout);
924 : }
925 :
926 : Datum
927 12 : tsvector_concat(PG_FUNCTION_ARGS)
928 : {
929 12 : TSVector in1 = PG_GETARG_TSVECTOR(0);
930 12 : TSVector in2 = PG_GETARG_TSVECTOR(1);
931 : TSVector out;
932 : WordEntry *ptr;
933 : WordEntry *ptr1,
934 : *ptr2;
935 : WordEntryPos *p;
936 12 : int maxpos = 0,
937 : i,
938 : j,
939 : i1,
940 : i2,
941 : dataoff,
942 : output_bytes,
943 : output_size;
944 : char *data,
945 : *data1,
946 : *data2;
947 :
948 : /* Get max position in in1; we'll need this to offset in2's positions */
949 12 : ptr = ARRPTR(in1);
950 12 : i = in1->size;
951 30 : while (i--)
952 : {
953 18 : if ((j = POSDATALEN(in1, ptr)) != 0)
954 : {
955 18 : p = POSDATAPTR(in1, ptr);
956 36 : while (j--)
957 : {
958 18 : if (WEP_GETPOS(*p) > maxpos)
959 12 : maxpos = WEP_GETPOS(*p);
960 18 : p++;
961 : }
962 : }
963 18 : ptr++;
964 : }
965 :
966 12 : ptr1 = ARRPTR(in1);
967 12 : ptr2 = ARRPTR(in2);
968 12 : data1 = STRPTR(in1);
969 12 : data2 = STRPTR(in2);
970 12 : i1 = in1->size;
971 12 : i2 = in2->size;
972 :
973 : /*
974 : * Conservative estimate of space needed. We might need all the data in
975 : * both inputs, and conceivably add a pad byte before position data for
976 : * each item where there was none before.
977 : */
978 12 : output_bytes = VARSIZE(in1) + VARSIZE(in2) + i1 + i2;
979 :
980 12 : out = (TSVector) palloc0(output_bytes);
981 12 : SET_VARSIZE(out, output_bytes);
982 :
983 : /*
984 : * We must make out->size valid so that STRPTR(out) is sensible. We'll
985 : * collapse out any unused space at the end.
986 : */
987 12 : out->size = in1->size + in2->size;
988 :
989 12 : ptr = ARRPTR(out);
990 12 : data = STRPTR(out);
991 12 : dataoff = 0;
992 30 : while (i1 && i2)
993 : {
994 18 : int cmp = compareEntry(data1, ptr1, data2, ptr2);
995 :
996 18 : if (cmp < 0)
997 : { /* in1 first */
998 6 : ptr->haspos = ptr1->haspos;
999 6 : ptr->len = ptr1->len;
1000 6 : memcpy(data + dataoff, data1 + ptr1->pos, ptr1->len);
1001 6 : ptr->pos = dataoff;
1002 6 : dataoff += ptr1->len;
1003 6 : if (ptr->haspos)
1004 : {
1005 6 : dataoff = SHORTALIGN(dataoff);
1006 6 : memcpy(data + dataoff, _POSVECPTR(in1, ptr1), POSDATALEN(in1, ptr1) * sizeof(WordEntryPos) + sizeof(uint16));
1007 6 : dataoff += POSDATALEN(in1, ptr1) * sizeof(WordEntryPos) + sizeof(uint16);
1008 : }
1009 :
1010 6 : ptr++;
1011 6 : ptr1++;
1012 6 : i1--;
1013 : }
1014 12 : else if (cmp > 0)
1015 : { /* in2 first */
1016 6 : ptr->haspos = ptr2->haspos;
1017 6 : ptr->len = ptr2->len;
1018 6 : memcpy(data + dataoff, data2 + ptr2->pos, ptr2->len);
1019 6 : ptr->pos = dataoff;
1020 6 : dataoff += ptr2->len;
1021 6 : if (ptr->haspos)
1022 : {
1023 0 : int addlen = add_pos(in2, ptr2, out, ptr, maxpos);
1024 :
1025 0 : if (addlen == 0)
1026 0 : ptr->haspos = 0;
1027 : else
1028 : {
1029 0 : dataoff = SHORTALIGN(dataoff);
1030 0 : dataoff += addlen * sizeof(WordEntryPos) + sizeof(uint16);
1031 : }
1032 : }
1033 :
1034 6 : ptr++;
1035 6 : ptr2++;
1036 6 : i2--;
1037 : }
1038 : else
1039 : {
1040 6 : ptr->haspos = ptr1->haspos | ptr2->haspos;
1041 6 : ptr->len = ptr1->len;
1042 6 : memcpy(data + dataoff, data1 + ptr1->pos, ptr1->len);
1043 6 : ptr->pos = dataoff;
1044 6 : dataoff += ptr1->len;
1045 6 : if (ptr->haspos)
1046 : {
1047 6 : if (ptr1->haspos)
1048 : {
1049 6 : dataoff = SHORTALIGN(dataoff);
1050 6 : memcpy(data + dataoff, _POSVECPTR(in1, ptr1), POSDATALEN(in1, ptr1) * sizeof(WordEntryPos) + sizeof(uint16));
1051 6 : dataoff += POSDATALEN(in1, ptr1) * sizeof(WordEntryPos) + sizeof(uint16);
1052 6 : if (ptr2->haspos)
1053 6 : dataoff += add_pos(in2, ptr2, out, ptr, maxpos) * sizeof(WordEntryPos);
1054 : }
1055 : else /* must have ptr2->haspos */
1056 : {
1057 0 : int addlen = add_pos(in2, ptr2, out, ptr, maxpos);
1058 :
1059 0 : if (addlen == 0)
1060 0 : ptr->haspos = 0;
1061 : else
1062 : {
1063 0 : dataoff = SHORTALIGN(dataoff);
1064 0 : dataoff += addlen * sizeof(WordEntryPos) + sizeof(uint16);
1065 : }
1066 : }
1067 : }
1068 :
1069 6 : ptr++;
1070 6 : ptr1++;
1071 6 : ptr2++;
1072 6 : i1--;
1073 6 : i2--;
1074 : }
1075 : }
1076 :
1077 18 : while (i1)
1078 : {
1079 6 : ptr->haspos = ptr1->haspos;
1080 6 : ptr->len = ptr1->len;
1081 6 : memcpy(data + dataoff, data1 + ptr1->pos, ptr1->len);
1082 6 : ptr->pos = dataoff;
1083 6 : dataoff += ptr1->len;
1084 6 : if (ptr->haspos)
1085 : {
1086 6 : dataoff = SHORTALIGN(dataoff);
1087 6 : memcpy(data + dataoff, _POSVECPTR(in1, ptr1), POSDATALEN(in1, ptr1) * sizeof(WordEntryPos) + sizeof(uint16));
1088 6 : dataoff += POSDATALEN(in1, ptr1) * sizeof(WordEntryPos) + sizeof(uint16);
1089 : }
1090 :
1091 6 : ptr++;
1092 6 : ptr1++;
1093 6 : i1--;
1094 : }
1095 :
1096 18 : while (i2)
1097 : {
1098 6 : ptr->haspos = ptr2->haspos;
1099 6 : ptr->len = ptr2->len;
1100 6 : memcpy(data + dataoff, data2 + ptr2->pos, ptr2->len);
1101 6 : ptr->pos = dataoff;
1102 6 : dataoff += ptr2->len;
1103 6 : if (ptr->haspos)
1104 : {
1105 6 : int addlen = add_pos(in2, ptr2, out, ptr, maxpos);
1106 :
1107 6 : if (addlen == 0)
1108 0 : ptr->haspos = 0;
1109 : else
1110 : {
1111 6 : dataoff = SHORTALIGN(dataoff);
1112 6 : dataoff += addlen * sizeof(WordEntryPos) + sizeof(uint16);
1113 : }
1114 : }
1115 :
1116 6 : ptr++;
1117 6 : ptr2++;
1118 6 : i2--;
1119 : }
1120 :
1121 : /*
1122 : * Instead of checking each offset individually, we check for overflow of
1123 : * pos fields once at the end.
1124 : */
1125 12 : if (dataoff > MAXSTRPOS)
1126 0 : ereport(ERROR,
1127 : (errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
1128 : errmsg("string is too long for tsvector (%d bytes, max %d bytes)", dataoff, MAXSTRPOS)));
1129 :
1130 : /*
1131 : * Adjust sizes (asserting that we didn't overrun the original estimates)
1132 : * and collapse out any unused array entries.
1133 : */
1134 12 : output_size = ptr - ARRPTR(out);
1135 : Assert(output_size <= out->size);
1136 12 : out->size = output_size;
1137 12 : if (data != STRPTR(out))
1138 6 : memmove(STRPTR(out), data, dataoff);
1139 12 : output_bytes = CALCDATASIZE(out->size, dataoff);
1140 : Assert(output_bytes <= VARSIZE(out));
1141 12 : SET_VARSIZE(out, output_bytes);
1142 :
1143 12 : PG_FREE_IF_COPY(in1, 0);
1144 12 : PG_FREE_IF_COPY(in2, 1);
1145 12 : PG_RETURN_POINTER(out);
1146 : }
1147 :
1148 : /*
1149 : * Compare two strings by tsvector rules.
1150 : *
1151 : * if prefix = true then it returns zero value iff b has prefix a
1152 : */
1153 : int32
1154 6215358 : tsCompareString(char *a, int lena, char *b, int lenb, bool prefix)
1155 : {
1156 : int cmp;
1157 :
1158 6215358 : if (lena == 0)
1159 : {
1160 36 : if (prefix)
1161 0 : cmp = 0; /* empty string is prefix of anything */
1162 : else
1163 36 : cmp = (lenb > 0) ? -1 : 0;
1164 : }
1165 6215322 : else if (lenb == 0)
1166 : {
1167 0 : cmp = (lena > 0) ? 1 : 0;
1168 : }
1169 : else
1170 : {
1171 6215322 : cmp = memcmp(a, b, Min((unsigned int) lena, (unsigned int) lenb));
1172 :
1173 6215322 : if (prefix)
1174 : {
1175 16458 : if (cmp == 0 && lena > lenb)
1176 0 : cmp = 1; /* a is longer, so not a prefix of b */
1177 : }
1178 6198864 : else if (cmp == 0 && lena != lenb)
1179 : {
1180 32268 : cmp = (lena < lenb) ? -1 : 1;
1181 : }
1182 : }
1183 :
1184 6215358 : return cmp;
1185 : }
1186 :
1187 : /*
1188 : * Check weight info or/and fill 'data' with the required positions
1189 : */
1190 : static TSTernaryValue
1191 68082 : checkclass_str(CHKVAL *chkval, WordEntry *entry, QueryOperand *val,
1192 : ExecPhraseData *data)
1193 : {
1194 68082 : TSTernaryValue result = TS_NO;
1195 :
1196 : Assert(data == NULL || data->npos == 0);
1197 :
1198 68082 : if (entry->haspos)
1199 : {
1200 : WordEntryPosVector *posvec;
1201 :
1202 : /*
1203 : * We can't use the _POSVECPTR macro here because the pointer to the
1204 : * tsvector's lexeme storage is already contained in chkval->values.
1205 : */
1206 4488 : posvec = (WordEntryPosVector *)
1207 4488 : (chkval->values + SHORTALIGN(entry->pos + entry->len));
1208 :
1209 4488 : if (val->weight && data)
1210 48 : {
1211 48 : WordEntryPos *posvec_iter = posvec->pos;
1212 : WordEntryPos *dptr;
1213 :
1214 : /*
1215 : * Filter position information by weights
1216 : */
1217 48 : dptr = data->pos = palloc(sizeof(WordEntryPos) * posvec->npos);
1218 48 : data->allocated = true;
1219 :
1220 : /* Is there a position with a matching weight? */
1221 96 : while (posvec_iter < posvec->pos + posvec->npos)
1222 : {
1223 : /* If true, append this position to the data->pos */
1224 48 : if (val->weight & (1 << WEP_GETWEIGHT(*posvec_iter)))
1225 : {
1226 24 : *dptr = WEP_GETPOS(*posvec_iter);
1227 24 : dptr++;
1228 : }
1229 :
1230 48 : posvec_iter++;
1231 : }
1232 :
1233 48 : data->npos = dptr - data->pos;
1234 :
1235 48 : if (data->npos > 0)
1236 24 : result = TS_YES;
1237 : else
1238 : {
1239 24 : pfree(data->pos);
1240 24 : data->pos = NULL;
1241 24 : data->allocated = false;
1242 : }
1243 : }
1244 4440 : else if (val->weight)
1245 : {
1246 456 : WordEntryPos *posvec_iter = posvec->pos;
1247 :
1248 : /* Is there a position with a matching weight? */
1249 690 : while (posvec_iter < posvec->pos + posvec->npos)
1250 : {
1251 504 : if (val->weight & (1 << WEP_GETWEIGHT(*posvec_iter)))
1252 : {
1253 270 : result = TS_YES;
1254 270 : break; /* no need to go further */
1255 : }
1256 :
1257 234 : posvec_iter++;
1258 : }
1259 : }
1260 3984 : else if (data)
1261 : {
1262 2274 : data->npos = posvec->npos;
1263 2274 : data->pos = posvec->pos;
1264 2274 : data->allocated = false;
1265 2274 : result = TS_YES;
1266 : }
1267 : else
1268 : {
1269 : /* simplest case: no weight check, positions not needed */
1270 1710 : result = TS_YES;
1271 : }
1272 : }
1273 : else
1274 : {
1275 : /*
1276 : * Position info is lacking, so if the caller requires it, we can only
1277 : * say that maybe there is a match.
1278 : *
1279 : * Notice, however, that we *don't* check val->weight here.
1280 : * Historically, stripped tsvectors are considered to match queries
1281 : * whether or not the query has a weight restriction; that's a little
1282 : * dubious but we'll preserve the behavior.
1283 : */
1284 63594 : if (data)
1285 23058 : result = TS_MAYBE;
1286 : else
1287 40536 : result = TS_YES;
1288 : }
1289 :
1290 68082 : return result;
1291 : }
1292 :
1293 : /*
1294 : * TS_execute callback for matching a tsquery operand to plain tsvector data
1295 : */
1296 : static TSTernaryValue
1297 284022 : checkcondition_str(void *checkval, QueryOperand *val, ExecPhraseData *data)
1298 : {
1299 284022 : CHKVAL *chkval = (CHKVAL *) checkval;
1300 284022 : WordEntry *StopLow = chkval->arrb;
1301 284022 : WordEntry *StopHigh = chkval->arre;
1302 284022 : WordEntry *StopMiddle = StopHigh;
1303 284022 : TSTernaryValue res = TS_NO;
1304 :
1305 : /* Loop invariant: StopLow <= val < StopHigh */
1306 1786806 : while (StopLow < StopHigh)
1307 : {
1308 : int difference;
1309 :
1310 1555830 : StopMiddle = StopLow + (StopHigh - StopLow) / 2;
1311 1555830 : difference = tsCompareString(chkval->operand + val->distance,
1312 1555830 : val->length,
1313 1555830 : chkval->values + StopMiddle->pos,
1314 1555830 : StopMiddle->len,
1315 : false);
1316 :
1317 1555830 : if (difference == 0)
1318 : {
1319 : /* Check weight info & fill 'data' with positions */
1320 53046 : res = checkclass_str(chkval, StopMiddle, val, data);
1321 53046 : break;
1322 : }
1323 1502784 : else if (difference > 0)
1324 847512 : StopLow = StopMiddle + 1;
1325 : else
1326 655272 : StopHigh = StopMiddle;
1327 : }
1328 :
1329 : /*
1330 : * If it's a prefix search, we should also consider lexemes that the
1331 : * search term is a prefix of (which will necessarily immediately follow
1332 : * the place we found in the above loop). But we can skip them if there
1333 : * was a definite match on the exact term AND the caller doesn't need
1334 : * position info.
1335 : */
1336 284022 : if (val->prefix && (res != TS_YES || data))
1337 : {
1338 16524 : WordEntryPos *allpos = NULL;
1339 16524 : int npos = 0,
1340 16524 : totalpos = 0;
1341 :
1342 : /* adjust start position for corner case */
1343 16524 : if (StopLow >= StopHigh)
1344 16512 : StopMiddle = StopHigh;
1345 :
1346 : /* we don't try to re-use any data from the initial match */
1347 16524 : if (data)
1348 : {
1349 36 : if (data->allocated)
1350 0 : pfree(data->pos);
1351 36 : data->pos = NULL;
1352 36 : data->allocated = false;
1353 36 : data->npos = 0;
1354 : }
1355 16524 : res = TS_NO;
1356 :
1357 31560 : while ((res != TS_YES || data) &&
1358 32556 : StopMiddle < chkval->arre &&
1359 15930 : tsCompareString(chkval->operand + val->distance,
1360 15930 : val->length,
1361 15930 : chkval->values + StopMiddle->pos,
1362 15930 : StopMiddle->len,
1363 : true) == 0)
1364 : {
1365 : TSTernaryValue subres;
1366 :
1367 15036 : subres = checkclass_str(chkval, StopMiddle, val, data);
1368 :
1369 15036 : if (subres != TS_NO)
1370 : {
1371 14976 : if (data)
1372 : {
1373 : /*
1374 : * We need to join position information
1375 : */
1376 42 : if (subres == TS_MAYBE)
1377 : {
1378 : /*
1379 : * No position info for this match, so we must report
1380 : * MAYBE overall.
1381 : */
1382 0 : res = TS_MAYBE;
1383 : /* forget any previous positions */
1384 0 : npos = 0;
1385 : /* don't leak storage */
1386 0 : if (allpos)
1387 0 : pfree(allpos);
1388 0 : break;
1389 : }
1390 :
1391 78 : while (npos + data->npos > totalpos)
1392 : {
1393 36 : if (totalpos == 0)
1394 : {
1395 36 : totalpos = 256;
1396 36 : allpos = palloc(sizeof(WordEntryPos) * totalpos);
1397 : }
1398 : else
1399 : {
1400 0 : totalpos *= 2;
1401 0 : allpos = repalloc(allpos, sizeof(WordEntryPos) * totalpos);
1402 : }
1403 : }
1404 :
1405 42 : memcpy(allpos + npos, data->pos, sizeof(WordEntryPos) * data->npos);
1406 42 : npos += data->npos;
1407 :
1408 : /* don't leak storage from individual matches */
1409 42 : if (data->allocated)
1410 24 : pfree(data->pos);
1411 42 : data->pos = NULL;
1412 42 : data->allocated = false;
1413 : /* it's important to reset data->npos before next loop */
1414 42 : data->npos = 0;
1415 : }
1416 : else
1417 : {
1418 : /* Don't need positions, just handle YES/MAYBE */
1419 14934 : if (subres == TS_YES || res == TS_NO)
1420 14934 : res = subres;
1421 : }
1422 : }
1423 :
1424 15036 : StopMiddle++;
1425 : }
1426 :
1427 16524 : if (data && npos > 0)
1428 : {
1429 : /* Sort and make unique array of found positions */
1430 36 : data->pos = allpos;
1431 36 : qsort(data->pos, npos, sizeof(WordEntryPos), compareWordEntryPos);
1432 36 : data->npos = qunique(data->pos, npos, sizeof(WordEntryPos),
1433 : compareWordEntryPos);
1434 36 : data->allocated = true;
1435 36 : res = TS_YES;
1436 : }
1437 : }
1438 :
1439 284022 : return res;
1440 : }
1441 :
1442 : /*
1443 : * Compute output position list for a tsquery operator in phrase mode.
1444 : *
1445 : * Merge the position lists in Ldata and Rdata as specified by "emit",
1446 : * returning the result list into *data. The input position lists must be
1447 : * sorted and unique, and the output will be as well.
1448 : *
1449 : * data: pointer to initially-all-zeroes output struct, or NULL
1450 : * Ldata, Rdata: input position lists
1451 : * emit: bitmask of TSPO_XXX flags
1452 : * Loffset: offset to be added to Ldata positions before comparing/outputting
1453 : * Roffset: offset to be added to Rdata positions before comparing/outputting
1454 : * max_npos: maximum possible required size of output position array
1455 : *
1456 : * Loffset and Roffset should not be negative, else we risk trying to output
1457 : * negative positions, which won't fit into WordEntryPos.
1458 : *
1459 : * The result is boolean (TS_YES or TS_NO), but for the caller's convenience
1460 : * we return it as TSTernaryValue.
1461 : *
1462 : * Returns TS_YES if any positions were emitted to *data; or if data is NULL,
1463 : * returns TS_YES if any positions would have been emitted.
1464 : */
1465 : #define TSPO_L_ONLY 0x01 /* emit positions appearing only in L */
1466 : #define TSPO_R_ONLY 0x02 /* emit positions appearing only in R */
1467 : #define TSPO_BOTH 0x04 /* emit positions appearing in both L&R */
1468 :
1469 : static TSTernaryValue
1470 29970 : TS_phrase_output(ExecPhraseData *data,
1471 : ExecPhraseData *Ldata,
1472 : ExecPhraseData *Rdata,
1473 : int emit,
1474 : int Loffset,
1475 : int Roffset,
1476 : int max_npos)
1477 : {
1478 : int Lindex,
1479 : Rindex;
1480 :
1481 : /* Loop until both inputs are exhausted */
1482 29970 : Lindex = Rindex = 0;
1483 31002 : while (Lindex < Ldata->npos || Rindex < Rdata->npos)
1484 : {
1485 : int Lpos,
1486 : Rpos;
1487 2334 : int output_pos = 0;
1488 :
1489 : /*
1490 : * Fetch current values to compare. WEP_GETPOS() is needed because
1491 : * ExecPhraseData->data can point to a tsvector's WordEntryPosVector.
1492 : */
1493 2334 : if (Lindex < Ldata->npos)
1494 1686 : Lpos = WEP_GETPOS(Ldata->pos[Lindex]) + Loffset;
1495 : else
1496 : {
1497 : /* L array exhausted, so we're done if R_ONLY isn't set */
1498 648 : if (!(emit & TSPO_R_ONLY))
1499 150 : break;
1500 498 : Lpos = INT_MAX;
1501 : }
1502 2184 : if (Rindex < Rdata->npos)
1503 1938 : Rpos = WEP_GETPOS(Rdata->pos[Rindex]) + Roffset;
1504 : else
1505 : {
1506 : /* R array exhausted, so we're done if L_ONLY isn't set */
1507 246 : if (!(emit & TSPO_L_ONLY))
1508 162 : break;
1509 84 : Rpos = INT_MAX;
1510 : }
1511 :
1512 : /* Merge-join the two input lists */
1513 2022 : if (Lpos < Rpos)
1514 : {
1515 : /* Lpos is not matched in Rdata, should we output it? */
1516 486 : if (emit & TSPO_L_ONLY)
1517 144 : output_pos = Lpos;
1518 486 : Lindex++;
1519 : }
1520 1536 : else if (Lpos == Rpos)
1521 : {
1522 : /* Lpos and Rpos match ... should we output it? */
1523 798 : if (emit & TSPO_BOTH)
1524 702 : output_pos = Rpos;
1525 798 : Lindex++;
1526 798 : Rindex++;
1527 : }
1528 : else /* Lpos > Rpos */
1529 : {
1530 : /* Rpos is not matched in Ldata, should we output it? */
1531 738 : if (emit & TSPO_R_ONLY)
1532 540 : output_pos = Rpos;
1533 738 : Rindex++;
1534 : }
1535 :
1536 2022 : if (output_pos > 0)
1537 : {
1538 1386 : if (data)
1539 : {
1540 : /* Store position, first allocating output array if needed */
1541 396 : if (data->pos == NULL)
1542 : {
1543 318 : data->pos = (WordEntryPos *)
1544 318 : palloc(max_npos * sizeof(WordEntryPos));
1545 318 : data->allocated = true;
1546 : }
1547 396 : data->pos[data->npos++] = output_pos;
1548 : }
1549 : else
1550 : {
1551 : /*
1552 : * Exact positions not needed, so return TS_YES as soon as we
1553 : * know there is at least one.
1554 : */
1555 990 : return TS_YES;
1556 : }
1557 : }
1558 : }
1559 :
1560 28980 : if (data && data->npos > 0)
1561 : {
1562 : /* Let's assert we didn't overrun the array */
1563 : Assert(data->npos <= max_npos);
1564 318 : return TS_YES;
1565 : }
1566 28662 : return TS_NO;
1567 : }
1568 :
1569 : /*
1570 : * Execute tsquery at or below an OP_PHRASE operator.
1571 : *
1572 : * This handles tsquery execution at recursion levels where we need to care
1573 : * about match locations.
1574 : *
1575 : * In addition to the same arguments used for TS_execute, the caller may pass
1576 : * a preinitialized-to-zeroes ExecPhraseData struct, to be filled with lexeme
1577 : * match position info on success. data == NULL if no position data need be
1578 : * returned.
1579 : * Note: the function assumes data != NULL for operators other than OP_PHRASE.
1580 : * This is OK because an outside call always starts from an OP_PHRASE node,
1581 : * and all internal recursion cases pass data != NULL.
1582 : *
1583 : * The detailed semantics of the match data, given that the function returned
1584 : * TS_YES (successful match), are:
1585 : *
1586 : * npos > 0, negate = false:
1587 : * query is matched at specified position(s) (and only those positions)
1588 : * npos > 0, negate = true:
1589 : * query is matched at all positions *except* specified position(s)
1590 : * npos = 0, negate = true:
1591 : * query is matched at all positions
1592 : * npos = 0, negate = false:
1593 : * disallowed (this should result in TS_NO or TS_MAYBE, as appropriate)
1594 : *
1595 : * Successful matches also return a "width" value which is the match width in
1596 : * lexemes, less one. Hence, "width" is zero for simple one-lexeme matches,
1597 : * and is the sum of the phrase operator distances for phrase matches. Note
1598 : * that when width > 0, the listed positions represent the ends of matches not
1599 : * the starts. (This unintuitive rule is needed to avoid possibly generating
1600 : * negative positions, which wouldn't fit into the WordEntryPos arrays.)
1601 : *
1602 : * If the TSExecuteCallback function reports that an operand is present
1603 : * but fails to provide position(s) for it, we will return TS_MAYBE when
1604 : * it is possible but not certain that the query is matched.
1605 : *
1606 : * When the function returns TS_NO or TS_MAYBE, it must return npos = 0,
1607 : * negate = false (which is the state initialized by the caller); but the
1608 : * "width" output in such cases is undefined.
1609 : */
1610 : static TSTernaryValue
1611 702606 : TS_phrase_execute(QueryItem *curitem, void *arg, uint32 flags,
1612 : TSExecuteCallback chkcond,
1613 : ExecPhraseData *data)
1614 : {
1615 : ExecPhraseData Ldata,
1616 : Rdata;
1617 : TSTernaryValue lmatch,
1618 : rmatch;
1619 : int Loffset,
1620 : Roffset,
1621 : maxwidth;
1622 :
1623 : /* since this function recurses, it could be driven to stack overflow */
1624 702606 : check_stack_depth();
1625 :
1626 : /* ... and let's check for query cancel while we're at it */
1627 702606 : CHECK_FOR_INTERRUPTS();
1628 :
1629 702606 : if (curitem->type == QI_VAL)
1630 345622 : return chkcond(arg, (QueryOperand *) curitem, data);
1631 :
1632 356984 : switch (curitem->qoperator.oper)
1633 : {
1634 120868 : case OP_NOT:
1635 :
1636 : /*
1637 : * We need not touch data->width, since a NOT operation does not
1638 : * change the match width.
1639 : */
1640 120868 : if (flags & TS_EXEC_SKIP_NOT)
1641 : {
1642 : /* with SKIP_NOT, report NOT as "match everywhere" */
1643 : Assert(data->npos == 0 && !data->negate);
1644 0 : data->negate = true;
1645 0 : return TS_YES;
1646 : }
1647 120868 : switch (TS_phrase_execute(curitem + 1, arg, flags, chkcond, data))
1648 : {
1649 105654 : case TS_NO:
1650 : /* change "match nowhere" to "match everywhere" */
1651 : Assert(data->npos == 0 && !data->negate);
1652 105654 : data->negate = true;
1653 105654 : return TS_YES;
1654 390 : case TS_YES:
1655 390 : if (data->npos > 0)
1656 : {
1657 : /* we have some positions, invert negate flag */
1658 384 : data->negate = !data->negate;
1659 384 : return TS_YES;
1660 : }
1661 6 : else if (data->negate)
1662 : {
1663 : /* change "match everywhere" to "match nowhere" */
1664 6 : data->negate = false;
1665 6 : return TS_NO;
1666 : }
1667 : /* Should not get here if result was TS_YES */
1668 : Assert(false);
1669 0 : break;
1670 14824 : case TS_MAYBE:
1671 : /* match positions are, and remain, uncertain */
1672 14824 : return TS_MAYBE;
1673 : }
1674 0 : break;
1675 :
1676 235960 : case OP_PHRASE:
1677 : case OP_AND:
1678 235960 : memset(&Ldata, 0, sizeof(Ldata));
1679 235960 : memset(&Rdata, 0, sizeof(Rdata));
1680 :
1681 235960 : lmatch = TS_phrase_execute(curitem + curitem->qoperator.left,
1682 : arg, flags, chkcond, &Ldata);
1683 235960 : if (lmatch == TS_NO)
1684 126202 : return TS_NO;
1685 :
1686 109758 : rmatch = TS_phrase_execute(curitem + 1,
1687 : arg, flags, chkcond, &Rdata);
1688 109758 : if (rmatch == TS_NO)
1689 54172 : return TS_NO;
1690 :
1691 : /*
1692 : * If either operand has no position information, then we can't
1693 : * return reliable position data, only a MAYBE result.
1694 : */
1695 55586 : if (lmatch == TS_MAYBE || rmatch == TS_MAYBE)
1696 25772 : return TS_MAYBE;
1697 :
1698 29814 : if (curitem->qoperator.oper == OP_PHRASE)
1699 : {
1700 : /*
1701 : * Compute Loffset and Roffset suitable for phrase match, and
1702 : * compute overall width of whole phrase match.
1703 : */
1704 29808 : Loffset = curitem->qoperator.distance + Rdata.width;
1705 29808 : Roffset = 0;
1706 29808 : if (data)
1707 186 : data->width = curitem->qoperator.distance +
1708 186 : Ldata.width + Rdata.width;
1709 : }
1710 : else
1711 : {
1712 : /*
1713 : * For OP_AND, set output width and alignment like OP_OR (see
1714 : * comment below)
1715 : */
1716 6 : maxwidth = Max(Ldata.width, Rdata.width);
1717 6 : Loffset = maxwidth - Ldata.width;
1718 6 : Roffset = maxwidth - Rdata.width;
1719 6 : if (data)
1720 6 : data->width = maxwidth;
1721 : }
1722 :
1723 29814 : if (Ldata.negate && Rdata.negate)
1724 : {
1725 : /* !L & !R: treat as !(L | R) */
1726 28440 : (void) TS_phrase_output(data, &Ldata, &Rdata,
1727 : TSPO_BOTH | TSPO_L_ONLY | TSPO_R_ONLY,
1728 : Loffset, Roffset,
1729 28440 : Ldata.npos + Rdata.npos);
1730 28440 : if (data)
1731 0 : data->negate = true;
1732 28440 : return TS_YES;
1733 : }
1734 1374 : else if (Ldata.negate)
1735 : {
1736 : /* !L & R */
1737 450 : return TS_phrase_output(data, &Ldata, &Rdata,
1738 : TSPO_R_ONLY,
1739 : Loffset, Roffset,
1740 : Rdata.npos);
1741 : }
1742 924 : else if (Rdata.negate)
1743 : {
1744 : /* L & !R */
1745 6 : return TS_phrase_output(data, &Ldata, &Rdata,
1746 : TSPO_L_ONLY,
1747 : Loffset, Roffset,
1748 : Ldata.npos);
1749 : }
1750 : else
1751 : {
1752 : /* straight AND */
1753 918 : return TS_phrase_output(data, &Ldata, &Rdata,
1754 : TSPO_BOTH,
1755 : Loffset, Roffset,
1756 918 : Min(Ldata.npos, Rdata.npos));
1757 : }
1758 :
1759 156 : case OP_OR:
1760 156 : memset(&Ldata, 0, sizeof(Ldata));
1761 156 : memset(&Rdata, 0, sizeof(Rdata));
1762 :
1763 156 : lmatch = TS_phrase_execute(curitem + curitem->qoperator.left,
1764 : arg, flags, chkcond, &Ldata);
1765 156 : rmatch = TS_phrase_execute(curitem + 1,
1766 : arg, flags, chkcond, &Rdata);
1767 :
1768 156 : if (lmatch == TS_NO && rmatch == TS_NO)
1769 12 : return TS_NO;
1770 :
1771 : /*
1772 : * If either operand has no position information, then we can't
1773 : * return reliable position data, only a MAYBE result.
1774 : */
1775 144 : if (lmatch == TS_MAYBE || rmatch == TS_MAYBE)
1776 0 : return TS_MAYBE;
1777 :
1778 : /*
1779 : * Cope with undefined output width from failed submatch. (This
1780 : * takes less code than trying to ensure that all failure returns
1781 : * set data->width to zero.)
1782 : */
1783 144 : if (lmatch == TS_NO)
1784 18 : Ldata.width = 0;
1785 144 : if (rmatch == TS_NO)
1786 84 : Rdata.width = 0;
1787 :
1788 : /*
1789 : * For OP_AND and OP_OR, report the width of the wider of the two
1790 : * inputs, and align the narrower input's positions to the right
1791 : * end of that width. This rule deals at least somewhat
1792 : * reasonably with cases like "x <-> (y | z <-> q)".
1793 : */
1794 144 : maxwidth = Max(Ldata.width, Rdata.width);
1795 144 : Loffset = maxwidth - Ldata.width;
1796 144 : Roffset = maxwidth - Rdata.width;
1797 144 : data->width = maxwidth;
1798 :
1799 144 : if (Ldata.negate && Rdata.negate)
1800 : {
1801 : /* !L | !R: treat as !(L & R) */
1802 6 : (void) TS_phrase_output(data, &Ldata, &Rdata,
1803 : TSPO_BOTH,
1804 : Loffset, Roffset,
1805 6 : Min(Ldata.npos, Rdata.npos));
1806 6 : data->negate = true;
1807 6 : return TS_YES;
1808 : }
1809 138 : else if (Ldata.negate)
1810 : {
1811 : /* !L | R: treat as !(L & !R) */
1812 30 : (void) TS_phrase_output(data, &Ldata, &Rdata,
1813 : TSPO_L_ONLY,
1814 : Loffset, Roffset,
1815 : Ldata.npos);
1816 30 : data->negate = true;
1817 30 : return TS_YES;
1818 : }
1819 108 : else if (Rdata.negate)
1820 : {
1821 : /* L | !R: treat as !(!L & R) */
1822 6 : (void) TS_phrase_output(data, &Ldata, &Rdata,
1823 : TSPO_R_ONLY,
1824 : Loffset, Roffset,
1825 : Rdata.npos);
1826 6 : data->negate = true;
1827 6 : return TS_YES;
1828 : }
1829 : else
1830 : {
1831 : /* straight OR */
1832 102 : return TS_phrase_output(data, &Ldata, &Rdata,
1833 : TSPO_BOTH | TSPO_L_ONLY | TSPO_R_ONLY,
1834 : Loffset, Roffset,
1835 102 : Ldata.npos + Rdata.npos);
1836 : }
1837 :
1838 0 : default:
1839 0 : elog(ERROR, "unrecognized operator: %d", curitem->qoperator.oper);
1840 : }
1841 :
1842 : /* not reachable, but keep compiler quiet */
1843 0 : return TS_NO;
1844 : }
1845 :
1846 :
1847 : /*
1848 : * Evaluate tsquery boolean expression.
1849 : *
1850 : * curitem: current tsquery item (initially, the first one)
1851 : * arg: opaque value to pass through to callback function
1852 : * flags: bitmask of flag bits shown in ts_utils.h
1853 : * chkcond: callback function to check whether a primitive value is present
1854 : */
1855 : bool
1856 520930 : TS_execute(QueryItem *curitem, void *arg, uint32 flags,
1857 : TSExecuteCallback chkcond)
1858 : {
1859 : /*
1860 : * If we get TS_MAYBE from the recursion, return true. We could only see
1861 : * that result if the caller passed TS_EXEC_PHRASE_NO_POS, so there's no
1862 : * need to check again.
1863 : */
1864 520930 : return TS_execute_recurse(curitem, arg, flags, chkcond) != TS_NO;
1865 : }
1866 :
1867 : /*
1868 : * Evaluate tsquery boolean expression.
1869 : *
1870 : * This is the same as TS_execute except that TS_MAYBE is returned as-is.
1871 : */
1872 : TSTernaryValue
1873 36942 : TS_execute_ternary(QueryItem *curitem, void *arg, uint32 flags,
1874 : TSExecuteCallback chkcond)
1875 : {
1876 36942 : return TS_execute_recurse(curitem, arg, flags, chkcond);
1877 : }
1878 :
1879 : /*
1880 : * TS_execute recursion for operators above any phrase operator. Here we do
1881 : * not need to worry about lexeme positions. As soon as we hit an OP_PHRASE
1882 : * operator, we pass it off to TS_phrase_execute which does worry.
1883 : */
1884 : static TSTernaryValue
1885 1055344 : TS_execute_recurse(QueryItem *curitem, void *arg, uint32 flags,
1886 : TSExecuteCallback chkcond)
1887 : {
1888 : TSTernaryValue lmatch;
1889 :
1890 : /* since this function recurses, it could be driven to stack overflow */
1891 1055344 : check_stack_depth();
1892 :
1893 : /* ... and let's check for query cancel while we're at it */
1894 1055344 : CHECK_FOR_INTERRUPTS();
1895 :
1896 1055344 : if (curitem->type == QI_VAL)
1897 424116 : return chkcond(arg, (QueryOperand *) curitem,
1898 : NULL /* don't need position info */ );
1899 :
1900 631228 : switch (curitem->qoperator.oper)
1901 : {
1902 203208 : case OP_NOT:
1903 203208 : if (flags & TS_EXEC_SKIP_NOT)
1904 0 : return TS_YES;
1905 203208 : switch (TS_execute_recurse(curitem + 1, arg, flags, chkcond))
1906 : {
1907 191696 : case TS_NO:
1908 191696 : return TS_YES;
1909 4890 : case TS_YES:
1910 4890 : return TS_NO;
1911 6622 : case TS_MAYBE:
1912 6622 : return TS_MAYBE;
1913 : }
1914 0 : break;
1915 :
1916 83454 : case OP_AND:
1917 83454 : lmatch = TS_execute_recurse(curitem + curitem->qoperator.left, arg,
1918 : flags, chkcond);
1919 83454 : if (lmatch == TS_NO)
1920 66300 : return TS_NO;
1921 17154 : switch (TS_execute_recurse(curitem + 1, arg, flags, chkcond))
1922 : {
1923 10114 : case TS_NO:
1924 10114 : return TS_NO;
1925 3300 : case TS_YES:
1926 3300 : return lmatch;
1927 3740 : case TS_MAYBE:
1928 3740 : return TS_MAYBE;
1929 : }
1930 0 : break;
1931 :
1932 108918 : case OP_OR:
1933 108918 : lmatch = TS_execute_recurse(curitem + curitem->qoperator.left, arg,
1934 : flags, chkcond);
1935 108918 : if (lmatch == TS_YES)
1936 24180 : return TS_YES;
1937 84738 : switch (TS_execute_recurse(curitem + 1, arg, flags, chkcond))
1938 : {
1939 57512 : case TS_NO:
1940 57512 : return lmatch;
1941 7416 : case TS_YES:
1942 7416 : return TS_YES;
1943 19810 : case TS_MAYBE:
1944 19810 : return TS_MAYBE;
1945 : }
1946 0 : break;
1947 :
1948 235648 : case OP_PHRASE:
1949 :
1950 : /*
1951 : * If we get a MAYBE result, and the caller doesn't want that,
1952 : * convert it to NO. It would be more consistent, perhaps, to
1953 : * return the result of TS_phrase_execute() verbatim and then
1954 : * convert MAYBE results at the top of the recursion. But
1955 : * converting at the topmost phrase operator gives results that
1956 : * are bug-compatible with the old implementation, so do it like
1957 : * this for now.
1958 : */
1959 235648 : switch (TS_phrase_execute(curitem, arg, flags, chkcond, NULL))
1960 : {
1961 180566 : case TS_NO:
1962 180566 : return TS_NO;
1963 29316 : case TS_YES:
1964 29316 : return TS_YES;
1965 25766 : case TS_MAYBE:
1966 25766 : return (flags & TS_EXEC_PHRASE_NO_POS) ? TS_MAYBE : TS_NO;
1967 : }
1968 0 : break;
1969 :
1970 0 : default:
1971 0 : elog(ERROR, "unrecognized operator: %d", curitem->qoperator.oper);
1972 : }
1973 :
1974 : /* not reachable, but keep compiler quiet */
1975 0 : return TS_NO;
1976 : }
1977 :
1978 : /*
1979 : * Evaluate tsquery and report locations of matching terms.
1980 : *
1981 : * This is like TS_execute except that it returns match locations not just
1982 : * success/failure status. The callback function is required to provide
1983 : * position data (we report failure if it doesn't).
1984 : *
1985 : * On successful match, the result is a List of ExecPhraseData structs, one
1986 : * for each AND'ed term or phrase operator in the query. Each struct includes
1987 : * a sorted array of lexeme positions matching that term. (Recall that for
1988 : * phrase operators, the match includes width+1 lexemes, and the recorded
1989 : * position is that of the rightmost lexeme.)
1990 : *
1991 : * OR subexpressions are handled by union'ing their match locations into a
1992 : * single List element, which is valid since any of those locations contains
1993 : * a match. However, when some of the OR'ed terms are phrase operators, we
1994 : * report the maximum width of any of the OR'ed terms, making such cases
1995 : * slightly imprecise in the conservative direction. (For example, if the
1996 : * tsquery is "(A <-> B) | C", an occurrence of C in the data would be
1997 : * reported as though it includes the lexeme to the left of C.)
1998 : *
1999 : * Locations of NOT subexpressions are not reported. (Obviously, there can
2000 : * be no successful NOT matches at top level, or the match would have failed.
2001 : * So this amounts to ignoring NOTs underneath ORs.)
2002 : *
2003 : * The result is NIL if no match, or if position data was not returned.
2004 : *
2005 : * Arguments are the same as for TS_execute, although flags is currently
2006 : * vestigial since none of the defined bits are sensible here.
2007 : */
2008 : List *
2009 362 : TS_execute_locations(QueryItem *curitem, void *arg,
2010 : uint32 flags,
2011 : TSExecuteCallback chkcond)
2012 : {
2013 : List *result;
2014 :
2015 : /* No flags supported, as yet */
2016 : Assert(flags == TS_EXEC_EMPTY);
2017 362 : if (TS_execute_locations_recurse(curitem, arg, chkcond, &result))
2018 128 : return result;
2019 234 : return NIL;
2020 : }
2021 :
2022 : /*
2023 : * TS_execute_locations recursion for operators above any phrase operator.
2024 : * OP_PHRASE subexpressions can be passed off to TS_phrase_execute.
2025 : */
2026 : static bool
2027 1070 : TS_execute_locations_recurse(QueryItem *curitem, void *arg,
2028 : TSExecuteCallback chkcond,
2029 : List **locations)
2030 : {
2031 : bool lmatch,
2032 : rmatch;
2033 : List *llocations,
2034 : *rlocations;
2035 : ExecPhraseData *data;
2036 :
2037 : /* since this function recurses, it could be driven to stack overflow */
2038 1070 : check_stack_depth();
2039 :
2040 : /* ... and let's check for query cancel while we're at it */
2041 1070 : CHECK_FOR_INTERRUPTS();
2042 :
2043 : /* Default locations result is empty */
2044 1070 : *locations = NIL;
2045 :
2046 1070 : if (curitem->type == QI_VAL)
2047 : {
2048 446 : data = palloc0_object(ExecPhraseData);
2049 446 : if (chkcond(arg, (QueryOperand *) curitem, data) == TS_YES)
2050 : {
2051 212 : *locations = list_make1(data);
2052 212 : return true;
2053 : }
2054 234 : pfree(data);
2055 234 : return false;
2056 : }
2057 :
2058 624 : switch (curitem->qoperator.oper)
2059 : {
2060 12 : case OP_NOT:
2061 12 : if (!TS_execute_locations_recurse(curitem + 1, arg, chkcond,
2062 : &llocations))
2063 0 : return true; /* we don't pass back any locations */
2064 12 : return false;
2065 :
2066 528 : case OP_AND:
2067 528 : if (!TS_execute_locations_recurse(curitem + curitem->qoperator.left,
2068 : arg, chkcond,
2069 : &llocations))
2070 408 : return false;
2071 120 : if (!TS_execute_locations_recurse(curitem + 1,
2072 : arg, chkcond,
2073 : &rlocations))
2074 54 : return false;
2075 66 : *locations = list_concat(llocations, rlocations);
2076 66 : return true;
2077 :
2078 24 : case OP_OR:
2079 24 : lmatch = TS_execute_locations_recurse(curitem + curitem->qoperator.left,
2080 : arg, chkcond,
2081 : &llocations);
2082 24 : rmatch = TS_execute_locations_recurse(curitem + 1,
2083 : arg, chkcond,
2084 : &rlocations);
2085 24 : if (lmatch || rmatch)
2086 : {
2087 : /*
2088 : * We generate an AND'able location struct from each
2089 : * combination of sub-matches, following the disjunctive law
2090 : * (A & B) | (C & D) = (A | C) & (A | D) & (B | C) & (B | D).
2091 : *
2092 : * However, if either input didn't produce locations (i.e., it
2093 : * failed or was a NOT), we must just return the other list.
2094 : */
2095 24 : if (llocations == NIL)
2096 0 : *locations = rlocations;
2097 24 : else if (rlocations == NIL)
2098 12 : *locations = llocations;
2099 : else
2100 : {
2101 : ListCell *ll;
2102 :
2103 24 : foreach(ll, llocations)
2104 : {
2105 12 : ExecPhraseData *ldata = (ExecPhraseData *) lfirst(ll);
2106 : ListCell *lr;
2107 :
2108 24 : foreach(lr, rlocations)
2109 : {
2110 12 : ExecPhraseData *rdata = (ExecPhraseData *) lfirst(lr);
2111 :
2112 12 : data = palloc0_object(ExecPhraseData);
2113 12 : (void) TS_phrase_output(data, ldata, rdata,
2114 : TSPO_BOTH | TSPO_L_ONLY | TSPO_R_ONLY,
2115 : 0, 0,
2116 12 : ldata->npos + rdata->npos);
2117 : /* Report the larger width, as explained above. */
2118 12 : data->width = Max(ldata->width, rdata->width);
2119 12 : *locations = lappend(*locations, data);
2120 : }
2121 : }
2122 : }
2123 :
2124 24 : return true;
2125 : }
2126 0 : return false;
2127 :
2128 60 : case OP_PHRASE:
2129 : /* We can hand this off to TS_phrase_execute */
2130 60 : data = palloc0_object(ExecPhraseData);
2131 60 : if (TS_phrase_execute(curitem, arg, TS_EXEC_EMPTY, chkcond,
2132 : data) == TS_YES)
2133 : {
2134 60 : if (!data->negate)
2135 60 : *locations = list_make1(data);
2136 60 : return true;
2137 : }
2138 0 : pfree(data);
2139 0 : return false;
2140 :
2141 0 : default:
2142 0 : elog(ERROR, "unrecognized operator: %d", curitem->qoperator.oper);
2143 : }
2144 :
2145 : /* not reachable, but keep compiler quiet */
2146 : return false;
2147 : }
2148 :
2149 : /*
2150 : * Detect whether a tsquery boolean expression requires any positive matches
2151 : * to values shown in the tsquery.
2152 : *
2153 : * This is needed to know whether a GIN index search requires full index scan.
2154 : * For example, 'x & !y' requires a match of x, so it's sufficient to scan
2155 : * entries for x; but 'x | !y' could match rows containing neither x nor y.
2156 : */
2157 : bool
2158 834 : tsquery_requires_match(QueryItem *curitem)
2159 : {
2160 : /* since this function recurses, it could be driven to stack overflow */
2161 834 : check_stack_depth();
2162 :
2163 834 : if (curitem->type == QI_VAL)
2164 396 : return true;
2165 :
2166 438 : switch (curitem->qoperator.oper)
2167 : {
2168 168 : case OP_NOT:
2169 :
2170 : /*
2171 : * Assume there are no required matches underneath a NOT. For
2172 : * some cases with nested NOTs, we could prove there's a required
2173 : * match, but it seems unlikely to be worth the trouble.
2174 : */
2175 168 : return false;
2176 :
2177 204 : case OP_PHRASE:
2178 :
2179 : /*
2180 : * Treat OP_PHRASE as OP_AND here
2181 : */
2182 : case OP_AND:
2183 : /* If either side requires a match, we're good */
2184 204 : if (tsquery_requires_match(curitem + curitem->qoperator.left))
2185 156 : return true;
2186 : else
2187 48 : return tsquery_requires_match(curitem + 1);
2188 :
2189 66 : case OP_OR:
2190 : /* Both sides must require a match */
2191 66 : if (tsquery_requires_match(curitem + curitem->qoperator.left))
2192 66 : return tsquery_requires_match(curitem + 1);
2193 : else
2194 0 : return false;
2195 :
2196 0 : default:
2197 0 : elog(ERROR, "unrecognized operator: %d", curitem->qoperator.oper);
2198 : }
2199 :
2200 : /* not reachable, but keep compiler quiet */
2201 : return false;
2202 : }
2203 :
2204 : /*
2205 : * boolean operations
2206 : */
2207 : Datum
2208 60 : ts_match_qv(PG_FUNCTION_ARGS)
2209 : {
2210 60 : PG_RETURN_DATUM(DirectFunctionCall2(ts_match_vq,
2211 : PG_GETARG_DATUM(1),
2212 : PG_GETARG_DATUM(0)));
2213 : }
2214 :
2215 : Datum
2216 220080 : ts_match_vq(PG_FUNCTION_ARGS)
2217 : {
2218 220080 : TSVector val = PG_GETARG_TSVECTOR(0);
2219 220080 : TSQuery query = PG_GETARG_TSQUERY(1);
2220 : CHKVAL chkval;
2221 : bool result;
2222 :
2223 : /* empty query matches nothing */
2224 220080 : if (!query->size)
2225 : {
2226 0 : PG_FREE_IF_COPY(val, 0);
2227 0 : PG_FREE_IF_COPY(query, 1);
2228 0 : PG_RETURN_BOOL(false);
2229 : }
2230 :
2231 220080 : chkval.arrb = ARRPTR(val);
2232 220080 : chkval.arre = chkval.arrb + val->size;
2233 220080 : chkval.values = STRPTR(val);
2234 220080 : chkval.operand = GETOPERAND(query);
2235 220080 : result = TS_execute(GETQUERY(query),
2236 : &chkval,
2237 : TS_EXEC_EMPTY,
2238 : checkcondition_str);
2239 :
2240 220080 : PG_FREE_IF_COPY(val, 0);
2241 220080 : PG_FREE_IF_COPY(query, 1);
2242 220080 : PG_RETURN_BOOL(result);
2243 : }
2244 :
2245 : Datum
2246 0 : ts_match_tt(PG_FUNCTION_ARGS)
2247 : {
2248 : TSVector vector;
2249 : TSQuery query;
2250 : bool res;
2251 :
2252 0 : vector = DatumGetTSVector(DirectFunctionCall1(to_tsvector,
2253 : PG_GETARG_DATUM(0)));
2254 0 : query = DatumGetTSQuery(DirectFunctionCall1(plainto_tsquery,
2255 : PG_GETARG_DATUM(1)));
2256 :
2257 0 : res = DatumGetBool(DirectFunctionCall2(ts_match_vq,
2258 : TSVectorGetDatum(vector),
2259 : TSQueryGetDatum(query)));
2260 :
2261 0 : pfree(vector);
2262 0 : pfree(query);
2263 :
2264 0 : PG_RETURN_BOOL(res);
2265 : }
2266 :
2267 : Datum
2268 0 : ts_match_tq(PG_FUNCTION_ARGS)
2269 : {
2270 : TSVector vector;
2271 0 : TSQuery query = PG_GETARG_TSQUERY(1);
2272 : bool res;
2273 :
2274 0 : vector = DatumGetTSVector(DirectFunctionCall1(to_tsvector,
2275 : PG_GETARG_DATUM(0)));
2276 :
2277 0 : res = DatumGetBool(DirectFunctionCall2(ts_match_vq,
2278 : TSVectorGetDatum(vector),
2279 : TSQueryGetDatum(query)));
2280 :
2281 0 : pfree(vector);
2282 0 : PG_FREE_IF_COPY(query, 1);
2283 :
2284 0 : PG_RETURN_BOOL(res);
2285 : }
2286 :
2287 : /*
2288 : * ts_stat statistic function support
2289 : */
2290 :
2291 :
2292 : /*
2293 : * Returns the number of positions in value 'wptr' within tsvector 'txt',
2294 : * that have a weight equal to one of the weights in 'weight' bitmask.
2295 : */
2296 : static int
2297 8178 : check_weight(TSVector txt, WordEntry *wptr, int8 weight)
2298 : {
2299 8178 : int len = POSDATALEN(txt, wptr);
2300 8178 : int num = 0;
2301 8178 : WordEntryPos *ptr = POSDATAPTR(txt, wptr);
2302 :
2303 16650 : while (len--)
2304 : {
2305 8472 : if (weight & (1 << WEP_GETWEIGHT(*ptr)))
2306 12 : num++;
2307 8472 : ptr++;
2308 : }
2309 8178 : return num;
2310 : }
2311 :
2312 : #define compareStatWord(a,e,t) \
2313 : tsCompareString((a)->lexeme, (a)->lenlexeme, \
2314 : STRPTR(t) + (e)->pos, (e)->len, \
2315 : false)
2316 :
2317 : static void
2318 345624 : insertStatEntry(MemoryContext persistentContext, TSVectorStat *stat, TSVector txt, uint32 off)
2319 : {
2320 345624 : WordEntry *we = ARRPTR(txt) + off;
2321 345624 : StatEntry *node = stat->root,
2322 345624 : *pnode = NULL;
2323 : int n,
2324 345624 : res = 0;
2325 345624 : uint32 depth = 1;
2326 :
2327 345624 : if (stat->weight == 0)
2328 172812 : n = (we->haspos) ? POSDATALEN(txt, we) : 1;
2329 : else
2330 172812 : n = (we->haspos) ? check_weight(txt, we, stat->weight) : 0;
2331 :
2332 345624 : if (n == 0)
2333 172806 : return; /* nothing to insert */
2334 :
2335 1745382 : while (node)
2336 : {
2337 1738518 : res = compareStatWord(node, we, txt);
2338 :
2339 1738518 : if (res == 0)
2340 : {
2341 165954 : break;
2342 : }
2343 : else
2344 : {
2345 1572564 : pnode = node;
2346 1572564 : node = (res < 0) ? node->left : node->right;
2347 : }
2348 1572564 : depth++;
2349 : }
2350 :
2351 172818 : if (depth > stat->maxdepth)
2352 126 : stat->maxdepth = depth;
2353 :
2354 172818 : if (node == NULL)
2355 : {
2356 6864 : node = MemoryContextAlloc(persistentContext, STATENTRYHDRSZ + we->len);
2357 6864 : node->left = node->right = NULL;
2358 6864 : node->ndoc = 1;
2359 6864 : node->nentry = n;
2360 6864 : node->lenlexeme = we->len;
2361 6864 : memcpy(node->lexeme, STRPTR(txt) + we->pos, node->lenlexeme);
2362 :
2363 6864 : if (pnode == NULL)
2364 : {
2365 12 : stat->root = node;
2366 : }
2367 : else
2368 : {
2369 6852 : if (res < 0)
2370 3380 : pnode->left = node;
2371 : else
2372 3472 : pnode->right = node;
2373 : }
2374 : }
2375 : else
2376 : {
2377 165954 : node->ndoc++;
2378 165954 : node->nentry += n;
2379 : }
2380 : }
2381 :
2382 : static void
2383 495384 : chooseNextStatEntry(MemoryContext persistentContext, TSVectorStat *stat, TSVector txt,
2384 : uint32 low, uint32 high, uint32 offset)
2385 : {
2386 : uint32 pos;
2387 495384 : uint32 middle = (low + high) >> 1;
2388 :
2389 495384 : pos = (low + middle) >> 1;
2390 495384 : if (low != middle && pos >= offset && pos - offset < txt->size)
2391 170328 : insertStatEntry(persistentContext, stat, txt, pos - offset);
2392 495384 : pos = (high + middle + 1) >> 1;
2393 495384 : if (middle + 1 != high && pos >= offset && pos - offset < txt->size)
2394 169284 : insertStatEntry(persistentContext, stat, txt, pos - offset);
2395 :
2396 495384 : if (low != middle)
2397 247692 : chooseNextStatEntry(persistentContext, stat, txt, low, middle, offset);
2398 495384 : if (high != middle + 1)
2399 241680 : chooseNextStatEntry(persistentContext, stat, txt, middle + 1, high, offset);
2400 495384 : }
2401 :
2402 : /*
2403 : * This is written like a custom aggregate function, because the
2404 : * original plan was to do just that. Unfortunately, an aggregate function
2405 : * can't return a set, so that plan was abandoned. If that limitation is
2406 : * lifted in the future, ts_stat could be a real aggregate function so that
2407 : * you could use it like this:
2408 : *
2409 : * SELECT ts_stat(vector_column) FROM vector_table;
2410 : *
2411 : * where vector_column is a tsvector-type column in vector_table.
2412 : */
2413 :
2414 : static TSVectorStat *
2415 6108 : ts_accum(MemoryContext persistentContext, TSVectorStat *stat, Datum data)
2416 : {
2417 6108 : TSVector txt = DatumGetTSVector(data);
2418 : uint32 i,
2419 6108 : nbit = 0,
2420 : offset;
2421 :
2422 6108 : if (stat == NULL)
2423 : { /* Init in first */
2424 0 : stat = MemoryContextAllocZero(persistentContext, sizeof(TSVectorStat));
2425 0 : stat->maxdepth = 1;
2426 : }
2427 :
2428 : /* simple check of correctness */
2429 6108 : if (txt == NULL || txt->size == 0)
2430 : {
2431 96 : if (txt && txt != (TSVector) DatumGetPointer(data))
2432 96 : pfree(txt);
2433 96 : return stat;
2434 : }
2435 :
2436 6012 : i = txt->size - 1;
2437 42720 : for (; i > 0; i >>= 1)
2438 36708 : nbit++;
2439 :
2440 6012 : nbit = 1 << nbit;
2441 6012 : offset = (nbit - txt->size) / 2;
2442 :
2443 6012 : insertStatEntry(persistentContext, stat, txt, (nbit >> 1) - offset);
2444 6012 : chooseNextStatEntry(persistentContext, stat, txt, 0, nbit, offset);
2445 :
2446 6012 : return stat;
2447 : }
2448 :
2449 : static void
2450 12 : ts_setup_firstcall(FunctionCallInfo fcinfo, FuncCallContext *funcctx,
2451 : TSVectorStat *stat)
2452 : {
2453 : TupleDesc tupdesc;
2454 : MemoryContext oldcontext;
2455 : StatEntry *node;
2456 :
2457 12 : funcctx->user_fctx = (void *) stat;
2458 :
2459 12 : oldcontext = MemoryContextSwitchTo(funcctx->multi_call_memory_ctx);
2460 :
2461 12 : stat->stack = palloc0(sizeof(StatEntry *) * (stat->maxdepth + 1));
2462 12 : stat->stackpos = 0;
2463 :
2464 12 : node = stat->root;
2465 : /* find leftmost value */
2466 12 : if (node == NULL)
2467 0 : stat->stack[stat->stackpos] = NULL;
2468 : else
2469 : for (;;)
2470 : {
2471 48 : stat->stack[stat->stackpos] = node;
2472 48 : if (node->left)
2473 : {
2474 36 : stat->stackpos++;
2475 36 : node = node->left;
2476 : }
2477 : else
2478 12 : break;
2479 : }
2480 : Assert(stat->stackpos <= stat->maxdepth);
2481 :
2482 12 : if (get_call_result_type(fcinfo, NULL, &tupdesc) != TYPEFUNC_COMPOSITE)
2483 0 : elog(ERROR, "return type must be a row type");
2484 12 : funcctx->tuple_desc = tupdesc;
2485 12 : funcctx->attinmeta = TupleDescGetAttInMetadata(tupdesc);
2486 :
2487 12 : MemoryContextSwitchTo(oldcontext);
2488 12 : }
2489 :
2490 : static StatEntry *
2491 13728 : walkStatEntryTree(TSVectorStat *stat)
2492 : {
2493 13728 : StatEntry *node = stat->stack[stat->stackpos];
2494 :
2495 13728 : if (node == NULL)
2496 0 : return NULL;
2497 :
2498 13728 : if (node->ndoc != 0)
2499 : {
2500 : /* return entry itself: we already was at left sublink */
2501 3392 : return node;
2502 : }
2503 10336 : else if (node->right && node->right != stat->stack[stat->stackpos + 1])
2504 : {
2505 : /* go on right sublink */
2506 3472 : stat->stackpos++;
2507 3472 : node = node->right;
2508 :
2509 : /* find most-left value */
2510 : for (;;)
2511 : {
2512 6816 : stat->stack[stat->stackpos] = node;
2513 6816 : if (node->left)
2514 : {
2515 3344 : stat->stackpos++;
2516 3344 : node = node->left;
2517 : }
2518 : else
2519 3472 : break;
2520 : }
2521 3472 : Assert(stat->stackpos <= stat->maxdepth);
2522 : }
2523 : else
2524 : {
2525 : /* we already return all left subtree, itself and right subtree */
2526 6864 : if (stat->stackpos == 0)
2527 12 : return NULL;
2528 :
2529 6852 : stat->stackpos--;
2530 6852 : return walkStatEntryTree(stat);
2531 : }
2532 :
2533 3472 : return node;
2534 : }
2535 :
2536 : static Datum
2537 6876 : ts_process_call(FuncCallContext *funcctx)
2538 : {
2539 : TSVectorStat *st;
2540 : StatEntry *entry;
2541 :
2542 6876 : st = (TSVectorStat *) funcctx->user_fctx;
2543 :
2544 6876 : entry = walkStatEntryTree(st);
2545 :
2546 6876 : if (entry != NULL)
2547 : {
2548 : Datum result;
2549 : char *values[3];
2550 : char ndoc[16];
2551 : char nentry[16];
2552 : HeapTuple tuple;
2553 :
2554 6864 : values[0] = palloc(entry->lenlexeme + 1);
2555 6864 : memcpy(values[0], entry->lexeme, entry->lenlexeme);
2556 6864 : (values[0])[entry->lenlexeme] = '\0';
2557 6864 : sprintf(ndoc, "%d", entry->ndoc);
2558 6864 : values[1] = ndoc;
2559 6864 : sprintf(nentry, "%d", entry->nentry);
2560 6864 : values[2] = nentry;
2561 :
2562 6864 : tuple = BuildTupleFromCStrings(funcctx->attinmeta, values);
2563 6864 : result = HeapTupleGetDatum(tuple);
2564 :
2565 6864 : pfree(values[0]);
2566 :
2567 : /* mark entry as already visited */
2568 6864 : entry->ndoc = 0;
2569 :
2570 6864 : return result;
2571 : }
2572 :
2573 12 : return (Datum) 0;
2574 : }
2575 :
2576 : static TSVectorStat *
2577 12 : ts_stat_sql(MemoryContext persistentContext, text *txt, text *ws)
2578 : {
2579 12 : char *query = text_to_cstring(txt);
2580 : TSVectorStat *stat;
2581 : bool isnull;
2582 : Portal portal;
2583 : SPIPlanPtr plan;
2584 :
2585 12 : if ((plan = SPI_prepare(query, 0, NULL)) == NULL)
2586 : /* internal error */
2587 0 : elog(ERROR, "SPI_prepare(\"%s\") failed", query);
2588 :
2589 12 : if ((portal = SPI_cursor_open(NULL, plan, NULL, NULL, true)) == NULL)
2590 : /* internal error */
2591 0 : elog(ERROR, "SPI_cursor_open(\"%s\") failed", query);
2592 :
2593 12 : SPI_cursor_fetch(portal, true, 100);
2594 :
2595 12 : if (SPI_tuptable == NULL ||
2596 12 : SPI_tuptable->tupdesc->natts != 1 ||
2597 12 : !IsBinaryCoercible(SPI_gettypeid(SPI_tuptable->tupdesc, 1),
2598 : TSVECTOROID))
2599 0 : ereport(ERROR,
2600 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
2601 : errmsg("ts_stat query must return one tsvector column")));
2602 :
2603 12 : stat = MemoryContextAllocZero(persistentContext, sizeof(TSVectorStat));
2604 12 : stat->maxdepth = 1;
2605 :
2606 12 : if (ws)
2607 : {
2608 : char *buf;
2609 :
2610 6 : buf = VARDATA_ANY(ws);
2611 18 : while (buf - VARDATA_ANY(ws) < VARSIZE_ANY_EXHDR(ws))
2612 : {
2613 12 : if (pg_mblen(buf) == 1)
2614 : {
2615 12 : switch (*buf)
2616 : {
2617 6 : case 'A':
2618 : case 'a':
2619 6 : stat->weight |= 1 << 3;
2620 6 : break;
2621 6 : case 'B':
2622 : case 'b':
2623 6 : stat->weight |= 1 << 2;
2624 6 : break;
2625 0 : case 'C':
2626 : case 'c':
2627 0 : stat->weight |= 1 << 1;
2628 0 : break;
2629 0 : case 'D':
2630 : case 'd':
2631 0 : stat->weight |= 1;
2632 0 : break;
2633 0 : default:
2634 0 : stat->weight |= 0;
2635 : }
2636 0 : }
2637 12 : buf += pg_mblen(buf);
2638 : }
2639 : }
2640 :
2641 84 : while (SPI_processed > 0)
2642 : {
2643 : uint64 i;
2644 :
2645 6180 : for (i = 0; i < SPI_processed; i++)
2646 : {
2647 6108 : Datum data = SPI_getbinval(SPI_tuptable->vals[i], SPI_tuptable->tupdesc, 1, &isnull);
2648 :
2649 6108 : if (!isnull)
2650 6108 : stat = ts_accum(persistentContext, stat, data);
2651 : }
2652 :
2653 72 : SPI_freetuptable(SPI_tuptable);
2654 72 : SPI_cursor_fetch(portal, true, 100);
2655 : }
2656 :
2657 12 : SPI_freetuptable(SPI_tuptable);
2658 12 : SPI_cursor_close(portal);
2659 12 : SPI_freeplan(plan);
2660 12 : pfree(query);
2661 :
2662 12 : return stat;
2663 : }
2664 :
2665 : Datum
2666 6864 : ts_stat1(PG_FUNCTION_ARGS)
2667 : {
2668 : FuncCallContext *funcctx;
2669 : Datum result;
2670 :
2671 6864 : if (SRF_IS_FIRSTCALL())
2672 : {
2673 : TSVectorStat *stat;
2674 6 : text *txt = PG_GETARG_TEXT_PP(0);
2675 :
2676 6 : funcctx = SRF_FIRSTCALL_INIT();
2677 6 : SPI_connect();
2678 6 : stat = ts_stat_sql(funcctx->multi_call_memory_ctx, txt, NULL);
2679 6 : PG_FREE_IF_COPY(txt, 0);
2680 6 : ts_setup_firstcall(fcinfo, funcctx, stat);
2681 6 : SPI_finish();
2682 : }
2683 :
2684 6864 : funcctx = SRF_PERCALL_SETUP();
2685 6864 : if ((result = ts_process_call(funcctx)) != (Datum) 0)
2686 6858 : SRF_RETURN_NEXT(funcctx, result);
2687 6 : SRF_RETURN_DONE(funcctx);
2688 : }
2689 :
2690 : Datum
2691 12 : ts_stat2(PG_FUNCTION_ARGS)
2692 : {
2693 : FuncCallContext *funcctx;
2694 : Datum result;
2695 :
2696 12 : if (SRF_IS_FIRSTCALL())
2697 : {
2698 : TSVectorStat *stat;
2699 6 : text *txt = PG_GETARG_TEXT_PP(0);
2700 6 : text *ws = PG_GETARG_TEXT_PP(1);
2701 :
2702 6 : funcctx = SRF_FIRSTCALL_INIT();
2703 6 : SPI_connect();
2704 6 : stat = ts_stat_sql(funcctx->multi_call_memory_ctx, txt, ws);
2705 6 : PG_FREE_IF_COPY(txt, 0);
2706 6 : PG_FREE_IF_COPY(ws, 1);
2707 6 : ts_setup_firstcall(fcinfo, funcctx, stat);
2708 6 : SPI_finish();
2709 : }
2710 :
2711 12 : funcctx = SRF_PERCALL_SETUP();
2712 12 : if ((result = ts_process_call(funcctx)) != (Datum) 0)
2713 6 : SRF_RETURN_NEXT(funcctx, result);
2714 6 : SRF_RETURN_DONE(funcctx);
2715 : }
2716 :
2717 :
2718 : /*
2719 : * Triggers for automatic update of a tsvector column from text column(s)
2720 : *
2721 : * Trigger arguments are either
2722 : * name of tsvector col, name of tsconfig to use, name(s) of text col(s)
2723 : * name of tsvector col, name of regconfig col, name(s) of text col(s)
2724 : * ie, tsconfig can either be specified by name, or indirectly as the
2725 : * contents of a regconfig field in the row. If the name is used, it must
2726 : * be explicitly schema-qualified.
2727 : */
2728 : Datum
2729 18 : tsvector_update_trigger_byid(PG_FUNCTION_ARGS)
2730 : {
2731 18 : return tsvector_update_trigger(fcinfo, false);
2732 : }
2733 :
2734 : Datum
2735 0 : tsvector_update_trigger_bycolumn(PG_FUNCTION_ARGS)
2736 : {
2737 0 : return tsvector_update_trigger(fcinfo, true);
2738 : }
2739 :
2740 : static Datum
2741 18 : tsvector_update_trigger(PG_FUNCTION_ARGS, bool config_column)
2742 : {
2743 : TriggerData *trigdata;
2744 : Trigger *trigger;
2745 : Relation rel;
2746 18 : HeapTuple rettuple = NULL;
2747 : int tsvector_attr_num,
2748 : i;
2749 : ParsedText prs;
2750 : Datum datum;
2751 : bool isnull;
2752 : text *txt;
2753 : Oid cfgId;
2754 : bool update_needed;
2755 :
2756 : /* Check call context */
2757 18 : if (!CALLED_AS_TRIGGER(fcinfo)) /* internal error */
2758 0 : elog(ERROR, "tsvector_update_trigger: not fired by trigger manager");
2759 :
2760 18 : trigdata = (TriggerData *) fcinfo->context;
2761 18 : if (!TRIGGER_FIRED_FOR_ROW(trigdata->tg_event))
2762 0 : elog(ERROR, "tsvector_update_trigger: must be fired for row");
2763 18 : if (!TRIGGER_FIRED_BEFORE(trigdata->tg_event))
2764 0 : elog(ERROR, "tsvector_update_trigger: must be fired BEFORE event");
2765 :
2766 18 : if (TRIGGER_FIRED_BY_INSERT(trigdata->tg_event))
2767 : {
2768 12 : rettuple = trigdata->tg_trigtuple;
2769 12 : update_needed = true;
2770 : }
2771 6 : else if (TRIGGER_FIRED_BY_UPDATE(trigdata->tg_event))
2772 : {
2773 6 : rettuple = trigdata->tg_newtuple;
2774 6 : update_needed = false; /* computed below */
2775 : }
2776 : else
2777 0 : elog(ERROR, "tsvector_update_trigger: must be fired for INSERT or UPDATE");
2778 :
2779 18 : trigger = trigdata->tg_trigger;
2780 18 : rel = trigdata->tg_relation;
2781 :
2782 18 : if (trigger->tgnargs < 3)
2783 0 : elog(ERROR, "tsvector_update_trigger: arguments must be tsvector_field, ts_config, text_field1, ...)");
2784 :
2785 : /* Find the target tsvector column */
2786 18 : tsvector_attr_num = SPI_fnumber(rel->rd_att, trigger->tgargs[0]);
2787 18 : if (tsvector_attr_num == SPI_ERROR_NOATTRIBUTE)
2788 0 : ereport(ERROR,
2789 : (errcode(ERRCODE_UNDEFINED_COLUMN),
2790 : errmsg("tsvector column \"%s\" does not exist",
2791 : trigger->tgargs[0])));
2792 : /* This will effectively reject system columns, so no separate test: */
2793 18 : if (!IsBinaryCoercible(SPI_gettypeid(rel->rd_att, tsvector_attr_num),
2794 : TSVECTOROID))
2795 0 : ereport(ERROR,
2796 : (errcode(ERRCODE_DATATYPE_MISMATCH),
2797 : errmsg("column \"%s\" is not of tsvector type",
2798 : trigger->tgargs[0])));
2799 :
2800 : /* Find the configuration to use */
2801 18 : if (config_column)
2802 : {
2803 : int config_attr_num;
2804 :
2805 0 : config_attr_num = SPI_fnumber(rel->rd_att, trigger->tgargs[1]);
2806 0 : if (config_attr_num == SPI_ERROR_NOATTRIBUTE)
2807 0 : ereport(ERROR,
2808 : (errcode(ERRCODE_UNDEFINED_COLUMN),
2809 : errmsg("configuration column \"%s\" does not exist",
2810 : trigger->tgargs[1])));
2811 0 : if (!IsBinaryCoercible(SPI_gettypeid(rel->rd_att, config_attr_num),
2812 : REGCONFIGOID))
2813 0 : ereport(ERROR,
2814 : (errcode(ERRCODE_DATATYPE_MISMATCH),
2815 : errmsg("column \"%s\" is not of regconfig type",
2816 : trigger->tgargs[1])));
2817 :
2818 0 : datum = SPI_getbinval(rettuple, rel->rd_att, config_attr_num, &isnull);
2819 0 : if (isnull)
2820 0 : ereport(ERROR,
2821 : (errcode(ERRCODE_NULL_VALUE_NOT_ALLOWED),
2822 : errmsg("configuration column \"%s\" must not be null",
2823 : trigger->tgargs[1])));
2824 0 : cfgId = DatumGetObjectId(datum);
2825 : }
2826 : else
2827 : {
2828 : List *names;
2829 :
2830 18 : names = stringToQualifiedNameList(trigger->tgargs[1], NULL);
2831 : /* require a schema so that results are not search path dependent */
2832 18 : if (list_length(names) < 2)
2833 0 : ereport(ERROR,
2834 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
2835 : errmsg("text search configuration name \"%s\" must be schema-qualified",
2836 : trigger->tgargs[1])));
2837 18 : cfgId = get_ts_config_oid(names, false);
2838 : }
2839 :
2840 : /* initialize parse state */
2841 18 : prs.lenwords = 32;
2842 18 : prs.curwords = 0;
2843 18 : prs.pos = 0;
2844 18 : prs.words = (ParsedWord *) palloc(sizeof(ParsedWord) * prs.lenwords);
2845 :
2846 : /* find all words in indexable column(s) */
2847 36 : for (i = 2; i < trigger->tgnargs; i++)
2848 : {
2849 : int numattr;
2850 :
2851 18 : numattr = SPI_fnumber(rel->rd_att, trigger->tgargs[i]);
2852 18 : if (numattr == SPI_ERROR_NOATTRIBUTE)
2853 0 : ereport(ERROR,
2854 : (errcode(ERRCODE_UNDEFINED_COLUMN),
2855 : errmsg("column \"%s\" does not exist",
2856 : trigger->tgargs[i])));
2857 18 : if (!IsBinaryCoercible(SPI_gettypeid(rel->rd_att, numattr), TEXTOID))
2858 0 : ereport(ERROR,
2859 : (errcode(ERRCODE_DATATYPE_MISMATCH),
2860 : errmsg("column \"%s\" is not of a character type",
2861 : trigger->tgargs[i])));
2862 :
2863 18 : if (bms_is_member(numattr - FirstLowInvalidHeapAttributeNumber, trigdata->tg_updatedcols))
2864 6 : update_needed = true;
2865 :
2866 18 : datum = SPI_getbinval(rettuple, rel->rd_att, numattr, &isnull);
2867 18 : if (isnull)
2868 6 : continue;
2869 :
2870 12 : txt = DatumGetTextPP(datum);
2871 :
2872 12 : parsetext(cfgId, &prs, VARDATA_ANY(txt), VARSIZE_ANY_EXHDR(txt));
2873 :
2874 12 : if (txt != (text *) DatumGetPointer(datum))
2875 0 : pfree(txt);
2876 : }
2877 :
2878 18 : if (update_needed)
2879 : {
2880 : /* make tsvector value */
2881 18 : datum = TSVectorGetDatum(make_tsvector(&prs));
2882 18 : isnull = false;
2883 :
2884 : /* and insert it into tuple */
2885 18 : rettuple = heap_modify_tuple_by_cols(rettuple, rel->rd_att,
2886 : 1, &tsvector_attr_num,
2887 : &datum, &isnull);
2888 :
2889 18 : pfree(DatumGetPointer(datum));
2890 : }
2891 :
2892 18 : return PointerGetDatum(rettuple);
2893 : }
|
