Line data Source code
1 : /*-------------------------------------------------------------------------
2 : *
3 : * tsvector_op.c
4 : * operations over tsvector
5 : *
6 : * Portions Copyright (c) 1996-2021, 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 int tsvector_bsearch(const TSVector tsv, char *lexeme, int lexeme_len);
75 : static Datum tsvector_update_trigger(PG_FUNCTION_ARGS, bool config_column);
76 :
77 :
78 : /*
79 : * Order: haspos, len, word, for all positions (pos, weight)
80 : */
81 : static int
82 2 : silly_cmp_tsvector(const TSVector a, const TSVector b)
83 : {
84 2 : if (VARSIZE(a) < VARSIZE(b))
85 0 : return -1;
86 2 : else if (VARSIZE(a) > VARSIZE(b))
87 0 : return 1;
88 2 : else if (a->size < b->size)
89 0 : return -1;
90 2 : else if (a->size > b->size)
91 0 : return 1;
92 : else
93 : {
94 2 : WordEntry *aptr = ARRPTR(a);
95 2 : WordEntry *bptr = ARRPTR(b);
96 2 : int i = 0;
97 : int res;
98 :
99 :
100 8 : for (i = 0; i < a->size; i++)
101 : {
102 6 : if (aptr->haspos != bptr->haspos)
103 : {
104 0 : return (aptr->haspos > bptr->haspos) ? -1 : 1;
105 : }
106 6 : else if ((res = tsCompareString(STRPTR(a) + aptr->pos, aptr->len, STRPTR(b) + bptr->pos, bptr->len, false)) != 0)
107 : {
108 0 : return res;
109 : }
110 6 : else if (aptr->haspos)
111 : {
112 0 : WordEntryPos *ap = POSDATAPTR(a, aptr);
113 0 : WordEntryPos *bp = POSDATAPTR(b, bptr);
114 : int j;
115 :
116 0 : if (POSDATALEN(a, aptr) != POSDATALEN(b, bptr))
117 0 : return (POSDATALEN(a, aptr) > POSDATALEN(b, bptr)) ? -1 : 1;
118 :
119 0 : for (j = 0; j < POSDATALEN(a, aptr); j++)
120 : {
121 0 : if (WEP_GETPOS(*ap) != WEP_GETPOS(*bp))
122 : {
123 0 : return (WEP_GETPOS(*ap) > WEP_GETPOS(*bp)) ? -1 : 1;
124 : }
125 0 : else if (WEP_GETWEIGHT(*ap) != WEP_GETWEIGHT(*bp))
126 : {
127 0 : return (WEP_GETWEIGHT(*ap) > WEP_GETWEIGHT(*bp)) ? -1 : 1;
128 : }
129 0 : ap++, bp++;
130 : }
131 : }
132 :
133 6 : aptr++;
134 6 : bptr++;
135 : }
136 : }
137 :
138 2 : return 0;
139 : }
140 :
141 : #define TSVECTORCMPFUNC( type, action, ret ) \
142 : Datum \
143 : tsvector_##type(PG_FUNCTION_ARGS) \
144 : { \
145 : TSVector a = PG_GETARG_TSVECTOR(0); \
146 : TSVector b = PG_GETARG_TSVECTOR(1); \
147 : int res = silly_cmp_tsvector(a, b); \
148 : PG_FREE_IF_COPY(a,0); \
149 : PG_FREE_IF_COPY(b,1); \
150 : PG_RETURN_##ret( res action 0 ); \
151 : } \
152 : /* keep compiler quiet - no extra ; */ \
153 : extern int no_such_variable
154 :
155 0 : TSVECTORCMPFUNC(lt, <, BOOL);
156 0 : TSVECTORCMPFUNC(le, <=, BOOL);
157 2 : TSVECTORCMPFUNC(eq, ==, BOOL);
158 0 : TSVECTORCMPFUNC(ge, >=, BOOL);
159 0 : TSVECTORCMPFUNC(gt, >, BOOL);
160 0 : TSVECTORCMPFUNC(ne, !=, BOOL);
161 0 : TSVECTORCMPFUNC(cmp, +, INT32);
162 :
163 : Datum
164 60 : tsvector_strip(PG_FUNCTION_ARGS)
165 : {
166 60 : TSVector in = PG_GETARG_TSVECTOR(0);
167 : TSVector out;
168 : int i,
169 60 : len = 0;
170 60 : WordEntry *arrin = ARRPTR(in),
171 : *arrout;
172 : char *cur;
173 :
174 212 : for (i = 0; i < in->size; i++)
175 152 : len += arrin[i].len;
176 :
177 60 : len = CALCDATASIZE(in->size, len);
178 60 : out = (TSVector) palloc0(len);
179 60 : SET_VARSIZE(out, len);
180 60 : out->size = in->size;
181 60 : arrout = ARRPTR(out);
182 60 : cur = STRPTR(out);
183 212 : for (i = 0; i < in->size; i++)
184 : {
185 152 : memcpy(cur, STRPTR(in) + arrin[i].pos, arrin[i].len);
186 152 : arrout[i].haspos = 0;
187 152 : arrout[i].len = arrin[i].len;
188 152 : arrout[i].pos = cur - STRPTR(out);
189 152 : cur += arrout[i].len;
190 : }
191 :
192 60 : PG_FREE_IF_COPY(in, 0);
193 60 : PG_RETURN_POINTER(out);
194 : }
195 :
196 : Datum
197 4 : tsvector_length(PG_FUNCTION_ARGS)
198 : {
199 4 : TSVector in = PG_GETARG_TSVECTOR(0);
200 4 : int32 ret = in->size;
201 :
202 4 : PG_FREE_IF_COPY(in, 0);
203 4 : PG_RETURN_INT32(ret);
204 : }
205 :
206 : Datum
207 8 : tsvector_setweight(PG_FUNCTION_ARGS)
208 : {
209 8 : TSVector in = PG_GETARG_TSVECTOR(0);
210 8 : char cw = PG_GETARG_CHAR(1);
211 : TSVector out;
212 : int i,
213 : j;
214 : WordEntry *entry;
215 : WordEntryPos *p;
216 8 : int w = 0;
217 :
218 8 : switch (cw)
219 : {
220 0 : case 'A':
221 : case 'a':
222 0 : w = 3;
223 0 : break;
224 0 : case 'B':
225 : case 'b':
226 0 : w = 2;
227 0 : break;
228 8 : case 'C':
229 : case 'c':
230 8 : w = 1;
231 8 : break;
232 0 : case 'D':
233 : case 'd':
234 0 : w = 0;
235 0 : break;
236 0 : default:
237 : /* internal error */
238 0 : elog(ERROR, "unrecognized weight: %d", cw);
239 : }
240 :
241 8 : out = (TSVector) palloc(VARSIZE(in));
242 8 : memcpy(out, in, VARSIZE(in));
243 8 : entry = ARRPTR(out);
244 8 : i = out->size;
245 40 : while (i--)
246 : {
247 32 : if ((j = POSDATALEN(out, entry)) != 0)
248 : {
249 32 : p = POSDATAPTR(out, entry);
250 112 : while (j--)
251 : {
252 80 : WEP_SETWEIGHT(*p, w);
253 80 : p++;
254 : }
255 : }
256 32 : entry++;
257 : }
258 :
259 8 : PG_FREE_IF_COPY(in, 0);
260 8 : PG_RETURN_POINTER(out);
261 : }
262 :
263 : /*
264 : * setweight(tsin tsvector, char_weight "char", lexemes "text"[])
265 : *
266 : * Assign weight w to elements of tsin that are listed in lexemes.
267 : */
268 : Datum
269 20 : tsvector_setweight_by_filter(PG_FUNCTION_ARGS)
270 : {
271 20 : TSVector tsin = PG_GETARG_TSVECTOR(0);
272 20 : char char_weight = PG_GETARG_CHAR(1);
273 20 : ArrayType *lexemes = PG_GETARG_ARRAYTYPE_P(2);
274 :
275 : TSVector tsout;
276 : int i,
277 : j,
278 : nlexemes,
279 : weight;
280 : WordEntry *entry;
281 : Datum *dlexemes;
282 : bool *nulls;
283 :
284 20 : switch (char_weight)
285 : {
286 0 : case 'A':
287 : case 'a':
288 0 : weight = 3;
289 0 : break;
290 0 : case 'B':
291 : case 'b':
292 0 : weight = 2;
293 0 : break;
294 20 : case 'C':
295 : case 'c':
296 20 : weight = 1;
297 20 : break;
298 0 : case 'D':
299 : case 'd':
300 0 : weight = 0;
301 0 : break;
302 0 : default:
303 : /* internal error */
304 0 : elog(ERROR, "unrecognized weight: %c", char_weight);
305 : }
306 :
307 20 : tsout = (TSVector) palloc(VARSIZE(tsin));
308 20 : memcpy(tsout, tsin, VARSIZE(tsin));
309 20 : entry = ARRPTR(tsout);
310 :
311 20 : deconstruct_array(lexemes, TEXTOID, -1, false, TYPALIGN_INT,
312 : &dlexemes, &nulls, &nlexemes);
313 :
314 : /*
315 : * Assuming that lexemes array is significantly shorter than tsvector we
316 : * can iterate through lexemes performing binary search of each lexeme
317 : * from lexemes in tsvector.
318 : */
319 52 : for (i = 0; i < nlexemes; i++)
320 : {
321 : char *lex;
322 : int lex_len,
323 : lex_pos;
324 :
325 36 : if (nulls[i])
326 4 : ereport(ERROR,
327 : (errcode(ERRCODE_NULL_VALUE_NOT_ALLOWED),
328 : errmsg("lexeme array may not contain nulls")));
329 :
330 32 : lex = VARDATA(dlexemes[i]);
331 32 : lex_len = VARSIZE(dlexemes[i]) - VARHDRSZ;
332 32 : lex_pos = tsvector_bsearch(tsout, lex, lex_len);
333 :
334 32 : if (lex_pos >= 0 && (j = POSDATALEN(tsout, entry + lex_pos)) != 0)
335 : {
336 16 : WordEntryPos *p = POSDATAPTR(tsout, entry + lex_pos);
337 :
338 52 : while (j--)
339 : {
340 36 : WEP_SETWEIGHT(*p, weight);
341 36 : p++;
342 : }
343 : }
344 : }
345 :
346 16 : PG_FREE_IF_COPY(tsin, 0);
347 16 : PG_FREE_IF_COPY(lexemes, 2);
348 :
349 16 : PG_RETURN_POINTER(tsout);
350 : }
351 :
352 : #define compareEntry(pa, a, pb, b) \
353 : tsCompareString((pa) + (a)->pos, (a)->len, \
354 : (pb) + (b)->pos, (b)->len, \
355 : false)
356 :
357 : /*
358 : * Add positions from src to dest after offsetting them by maxpos.
359 : * Return the number added (might be less than expected due to overflow)
360 : */
361 : static int32
362 8 : add_pos(TSVector src, WordEntry *srcptr,
363 : TSVector dest, WordEntry *destptr,
364 : int32 maxpos)
365 : {
366 8 : uint16 *clen = &_POSVECPTR(dest, destptr)->npos;
367 : int i;
368 8 : uint16 slen = POSDATALEN(src, srcptr),
369 : startlen;
370 8 : WordEntryPos *spos = POSDATAPTR(src, srcptr),
371 8 : *dpos = POSDATAPTR(dest, destptr);
372 :
373 8 : if (!destptr->haspos)
374 0 : *clen = 0;
375 :
376 8 : startlen = *clen;
377 8 : for (i = 0;
378 16 : i < slen && *clen < MAXNUMPOS &&
379 8 : (*clen == 0 || WEP_GETPOS(dpos[*clen - 1]) != MAXENTRYPOS - 1);
380 8 : i++)
381 : {
382 8 : WEP_SETWEIGHT(dpos[*clen], WEP_GETWEIGHT(spos[i]));
383 8 : WEP_SETPOS(dpos[*clen], LIMITPOS(WEP_GETPOS(spos[i]) + maxpos));
384 8 : (*clen)++;
385 : }
386 :
387 8 : if (*clen != startlen)
388 8 : destptr->haspos = 1;
389 8 : return *clen - startlen;
390 : }
391 :
392 : /*
393 : * Perform binary search of given lexeme in TSVector.
394 : * Returns lexeme position in TSVector's entry array or -1 if lexeme wasn't
395 : * found.
396 : */
397 : static int
398 132 : tsvector_bsearch(const TSVector tsv, char *lexeme, int lexeme_len)
399 : {
400 132 : WordEntry *arrin = ARRPTR(tsv);
401 132 : int StopLow = 0,
402 132 : StopHigh = tsv->size,
403 : StopMiddle,
404 : cmp;
405 :
406 336 : while (StopLow < StopHigh)
407 : {
408 304 : StopMiddle = (StopLow + StopHigh) / 2;
409 :
410 608 : cmp = tsCompareString(lexeme, lexeme_len,
411 304 : STRPTR(tsv) + arrin[StopMiddle].pos,
412 304 : arrin[StopMiddle].len,
413 : false);
414 :
415 304 : if (cmp < 0)
416 128 : StopHigh = StopMiddle;
417 176 : else if (cmp > 0)
418 76 : StopLow = StopMiddle + 1;
419 : else /* found it */
420 100 : return StopMiddle;
421 : }
422 :
423 32 : return -1;
424 : }
425 :
426 : /*
427 : * qsort comparator functions
428 : */
429 :
430 : static int
431 44 : compare_int(const void *va, const void *vb)
432 : {
433 44 : int a = *((const int *) va);
434 44 : int b = *((const int *) vb);
435 :
436 44 : if (a == b)
437 8 : return 0;
438 36 : return (a > b) ? 1 : -1;
439 : }
440 :
441 : static int
442 68 : compare_text_lexemes(const void *va, const void *vb)
443 : {
444 68 : Datum a = *((const Datum *) va);
445 68 : Datum b = *((const Datum *) vb);
446 68 : char *alex = VARDATA_ANY(a);
447 68 : int alex_len = VARSIZE_ANY_EXHDR(a);
448 68 : char *blex = VARDATA_ANY(b);
449 68 : int blex_len = VARSIZE_ANY_EXHDR(b);
450 :
451 68 : 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 40 : tsvector_delete_by_indices(TSVector tsv, int *indices_to_delete,
465 : int indices_count)
466 : {
467 : TSVector tsout;
468 40 : WordEntry *arrin = ARRPTR(tsv),
469 : *arrout;
470 40 : 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 40 : if (indices_count > 1)
483 : {
484 16 : qsort(indices_to_delete, indices_count, sizeof(int), compare_int);
485 16 : 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 40 : tsout = (TSVector) palloc0(VARSIZE(tsv));
494 :
495 : /* This count must be correct because STRPTR(tsout) relies on it. */
496 40 : tsout->size = tsv->size - indices_count;
497 :
498 : /*
499 : * Copy tsv to tsout, skipping lexemes listed in indices_to_delete.
500 : */
501 40 : arrout = ARRPTR(tsout);
502 40 : dataout = STRPTR(tsout);
503 40 : curoff = 0;
504 240 : 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 : */
511 200 : if (k < indices_count && i == indices_to_delete[k])
512 : {
513 56 : k++;
514 56 : continue;
515 : }
516 :
517 : /* Copy lexeme and its positions and weights */
518 144 : memcpy(dataout + curoff, data + arrin[i].pos, arrin[i].len);
519 144 : arrout[j].haspos = arrin[i].haspos;
520 144 : arrout[j].len = arrin[i].len;
521 144 : arrout[j].pos = curoff;
522 144 : curoff += arrin[i].len;
523 144 : if (arrin[i].haspos)
524 : {
525 104 : int len = POSDATALEN(tsv, arrin + i) * sizeof(WordEntryPos)
526 : + sizeof(uint16);
527 :
528 104 : curoff = SHORTALIGN(curoff);
529 208 : memcpy(dataout + curoff,
530 104 : STRPTR(tsv) + SHORTALIGN(arrin[i].pos + arrin[i].len),
531 : len);
532 104 : curoff += len;
533 : }
534 :
535 144 : 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 40 : SET_VARSIZE(tsout, CALCDATASIZE(tsout->size, curoff));
546 40 : return tsout;
547 : }
548 :
549 : /*
550 : * Delete given lexeme from tsvector.
551 : * Implementation of user-level ts_delete(tsvector, text).
552 : */
553 : Datum
554 24 : tsvector_delete_str(PG_FUNCTION_ARGS)
555 : {
556 24 : TSVector tsin = PG_GETARG_TSVECTOR(0),
557 : tsout;
558 24 : text *tlexeme = PG_GETARG_TEXT_PP(1);
559 24 : char *lexeme = VARDATA_ANY(tlexeme);
560 24 : int lexeme_len = VARSIZE_ANY_EXHDR(tlexeme),
561 : skip_index;
562 :
563 24 : if ((skip_index = tsvector_bsearch(tsin, lexeme, lexeme_len)) == -1)
564 8 : PG_RETURN_POINTER(tsin);
565 :
566 16 : tsout = tsvector_delete_by_indices(tsin, &skip_index, 1);
567 :
568 16 : PG_FREE_IF_COPY(tsin, 0);
569 16 : PG_FREE_IF_COPY(tlexeme, 1);
570 16 : 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 28 : tsvector_delete_arr(PG_FUNCTION_ARGS)
579 : {
580 28 : TSVector tsin = PG_GETARG_TSVECTOR(0),
581 : tsout;
582 28 : 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 28 : deconstruct_array(lexemes, TEXTOID, -1, false, TYPALIGN_INT,
591 : &dlexemes, &nulls, &nlex);
592 :
593 : /*
594 : * In typical use case array of lexemes to delete is relatively small. So
595 : * here we optimize things for that scenario: iterate through lexarr
596 : * performing binary search of each lexeme from lexarr in tsvector.
597 : */
598 28 : skip_indices = palloc0(nlex * sizeof(int));
599 104 : for (i = skip_count = 0; i < nlex; i++)
600 : {
601 : char *lex;
602 : int lex_len,
603 : lex_pos;
604 :
605 80 : if (nulls[i])
606 4 : ereport(ERROR,
607 : (errcode(ERRCODE_NULL_VALUE_NOT_ALLOWED),
608 : errmsg("lexeme array may not contain nulls")));
609 :
610 76 : lex = VARDATA(dlexemes[i]);
611 76 : lex_len = VARSIZE(dlexemes[i]) - VARHDRSZ;
612 76 : lex_pos = tsvector_bsearch(tsin, lex, lex_len);
613 :
614 76 : if (lex_pos >= 0)
615 52 : skip_indices[skip_count++] = lex_pos;
616 : }
617 :
618 24 : tsout = tsvector_delete_by_indices(tsin, skip_indices, skip_count);
619 :
620 24 : pfree(skip_indices);
621 24 : PG_FREE_IF_COPY(tsin, 0);
622 24 : PG_FREE_IF_COPY(lexemes, 1);
623 :
624 24 : 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
634 120 : tsvector_unnest(PG_FUNCTION_ARGS)
635 : {
636 : FuncCallContext *funcctx;
637 : TSVector tsin;
638 :
639 120 : if (SRF_IS_FIRSTCALL())
640 : {
641 : MemoryContext oldcontext;
642 : TupleDesc tupdesc;
643 :
644 20 : funcctx = SRF_FIRSTCALL_INIT();
645 20 : oldcontext = MemoryContextSwitchTo(funcctx->multi_call_memory_ctx);
646 :
647 20 : tupdesc = CreateTemplateTupleDesc(3);
648 20 : TupleDescInitEntry(tupdesc, (AttrNumber) 1, "lexeme",
649 : TEXTOID, -1, 0);
650 20 : TupleDescInitEntry(tupdesc, (AttrNumber) 2, "positions",
651 : INT2ARRAYOID, -1, 0);
652 20 : TupleDescInitEntry(tupdesc, (AttrNumber) 3, "weights",
653 : TEXTARRAYOID, -1, 0);
654 20 : funcctx->tuple_desc = BlessTupleDesc(tupdesc);
655 :
656 20 : funcctx->user_fctx = PG_GETARG_TSVECTOR_COPY(0);
657 :
658 20 : MemoryContextSwitchTo(oldcontext);
659 : }
660 :
661 120 : funcctx = SRF_PERCALL_SETUP();
662 120 : tsin = (TSVector) funcctx->user_fctx;
663 :
664 120 : if (funcctx->call_cntr < tsin->size)
665 : {
666 100 : WordEntry *arrin = ARRPTR(tsin);
667 100 : char *data = STRPTR(tsin);
668 : HeapTuple tuple;
669 : int j,
670 100 : i = funcctx->call_cntr;
671 100 : bool nulls[] = {false, false, false};
672 : Datum values[3];
673 :
674 100 : values[0] = PointerGetDatum(cstring_to_text_with_len(data + arrin[i].pos, arrin[i].len));
675 :
676 100 : 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 60 : posv = _POSVECPTR(tsin, arrin + i);
689 60 : positions = palloc(posv->npos * sizeof(Datum));
690 60 : weights = palloc(posv->npos * sizeof(Datum));
691 168 : for (j = 0; j < posv->npos; j++)
692 : {
693 108 : positions[j] = Int16GetDatum(WEP_GETPOS(posv->pos[j]));
694 108 : weight = 'D' - WEP_GETWEIGHT(posv->pos[j]);
695 108 : weights[j] = PointerGetDatum(cstring_to_text_with_len(&weight,
696 : 1));
697 : }
698 :
699 60 : values[1] = PointerGetDatum(construct_array(positions, posv->npos,
700 : INT2OID, 2, true, TYPALIGN_SHORT));
701 60 : values[2] = PointerGetDatum(construct_array(weights, posv->npos,
702 : TEXTOID, -1, false, TYPALIGN_INT));
703 : }
704 : else
705 : {
706 40 : nulls[1] = nulls[2] = true;
707 : }
708 :
709 100 : tuple = heap_form_tuple(funcctx->tuple_desc, values, nulls);
710 100 : SRF_RETURN_NEXT(funcctx, HeapTupleGetDatum(tuple));
711 : }
712 : else
713 : {
714 20 : SRF_RETURN_DONE(funcctx);
715 : }
716 : }
717 :
718 : /*
719 : * Convert tsvector to array of lexemes.
720 : */
721 : Datum
722 8 : tsvector_to_array(PG_FUNCTION_ARGS)
723 : {
724 8 : TSVector tsin = PG_GETARG_TSVECTOR(0);
725 8 : WordEntry *arrin = ARRPTR(tsin);
726 : Datum *elements;
727 : int i;
728 : ArrayType *array;
729 :
730 8 : elements = palloc(tsin->size * sizeof(Datum));
731 :
732 48 : for (i = 0; i < tsin->size; i++)
733 : {
734 40 : elements[i] = PointerGetDatum(cstring_to_text_with_len(STRPTR(tsin) + arrin[i].pos,
735 : arrin[i].len));
736 : }
737 :
738 8 : array = construct_array(elements, tsin->size, TEXTOID, -1, false, TYPALIGN_INT);
739 :
740 8 : pfree(elements);
741 8 : PG_FREE_IF_COPY(tsin, 0);
742 8 : PG_RETURN_POINTER(array);
743 : }
744 :
745 : /*
746 : * Build tsvector from array of lexemes.
747 : */
748 : Datum
749 12 : array_to_tsvector(PG_FUNCTION_ARGS)
750 : {
751 12 : 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 12 : datalen = 0;
760 : char *cur;
761 :
762 12 : deconstruct_array(v, TEXTOID, -1, false, TYPALIGN_INT, &dlexemes, &nulls, &nitems);
763 :
764 : /* Reject nulls (maybe we should just ignore them, instead?) */
765 64 : for (i = 0; i < nitems; i++)
766 : {
767 56 : if (nulls[i])
768 4 : ereport(ERROR,
769 : (errcode(ERRCODE_NULL_VALUE_NOT_ALLOWED),
770 : errmsg("lexeme array may not contain nulls")));
771 : }
772 :
773 : /* Sort and de-dup, because this is required for a valid tsvector. */
774 8 : if (nitems > 1)
775 : {
776 8 : qsort(dlexemes, nitems, sizeof(Datum), compare_text_lexemes);
777 8 : nitems = qunique(dlexemes, nitems, sizeof(Datum),
778 : compare_text_lexemes);
779 : }
780 :
781 : /* Calculate space needed for surviving lexemes. */
782 40 : for (i = 0; i < nitems; i++)
783 32 : datalen += VARSIZE(dlexemes[i]) - VARHDRSZ;
784 8 : tslen = CALCDATASIZE(nitems, datalen);
785 :
786 : /* Allocate and fill tsvector. */
787 8 : tsout = (TSVector) palloc0(tslen);
788 8 : SET_VARSIZE(tsout, tslen);
789 8 : tsout->size = nitems;
790 :
791 8 : arrout = ARRPTR(tsout);
792 8 : cur = STRPTR(tsout);
793 40 : for (i = 0; i < nitems; i++)
794 : {
795 32 : char *lex = VARDATA(dlexemes[i]);
796 32 : int lex_len = VARSIZE(dlexemes[i]) - VARHDRSZ;
797 :
798 32 : memcpy(cur, lex, lex_len);
799 32 : arrout[i].haspos = 0;
800 32 : arrout[i].len = lex_len;
801 32 : arrout[i].pos = cur - STRPTR(tsout);
802 32 : cur += lex_len;
803 : }
804 :
805 8 : PG_FREE_IF_COPY(v, 0);
806 8 : PG_RETURN_POINTER(tsout);
807 : }
808 :
809 : /*
810 : * ts_filter(): keep only lexemes with given weights in tsvector.
811 : */
812 : Datum
813 12 : tsvector_filter(PG_FUNCTION_ARGS)
814 : {
815 12 : TSVector tsin = PG_GETARG_TSVECTOR(0),
816 : tsout;
817 12 : ArrayType *weights = PG_GETARG_ARRAYTYPE_P(1);
818 12 : WordEntry *arrin = ARRPTR(tsin),
819 : *arrout;
820 12 : char *datain = STRPTR(tsin),
821 : *dataout;
822 : Datum *dweights;
823 : bool *nulls;
824 : int nweights;
825 : int i,
826 : j;
827 12 : int cur_pos = 0;
828 12 : char mask = 0;
829 :
830 12 : deconstruct_array(weights, CHAROID, 1, true, TYPALIGN_CHAR,
831 : &dweights, &nulls, &nweights);
832 :
833 28 : for (i = 0; i < nweights; i++)
834 : {
835 : char char_weight;
836 :
837 20 : if (nulls[i])
838 4 : ereport(ERROR,
839 : (errcode(ERRCODE_NULL_VALUE_NOT_ALLOWED),
840 : errmsg("weight array may not contain nulls")));
841 :
842 16 : char_weight = DatumGetChar(dweights[i]);
843 16 : switch (char_weight)
844 : {
845 12 : case 'A':
846 : case 'a':
847 12 : mask = mask | 8;
848 12 : break;
849 4 : case 'B':
850 : case 'b':
851 4 : mask = mask | 4;
852 4 : break;
853 0 : case 'C':
854 : case 'c':
855 0 : mask = mask | 2;
856 0 : break;
857 0 : case 'D':
858 : case 'd':
859 0 : mask = mask | 1;
860 0 : break;
861 0 : default:
862 0 : ereport(ERROR,
863 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
864 : errmsg("unrecognized weight: \"%c\"", char_weight)));
865 : }
866 : }
867 :
868 8 : tsout = (TSVector) palloc0(VARSIZE(tsin));
869 8 : tsout->size = tsin->size;
870 8 : arrout = ARRPTR(tsout);
871 8 : dataout = STRPTR(tsout);
872 :
873 72 : for (i = j = 0; i < tsin->size; i++)
874 : {
875 : WordEntryPosVector *posvin,
876 : *posvout;
877 64 : int npos = 0;
878 : int k;
879 :
880 64 : if (!arrin[i].haspos)
881 20 : continue;
882 :
883 44 : posvin = _POSVECPTR(tsin, arrin + i);
884 44 : posvout = (WordEntryPosVector *)
885 44 : (dataout + SHORTALIGN(cur_pos + arrin[i].len));
886 :
887 88 : for (k = 0; k < posvin->npos; k++)
888 : {
889 44 : if (mask & (1 << WEP_GETWEIGHT(posvin->pos[k])))
890 20 : posvout->pos[npos++] = posvin->pos[k];
891 : }
892 :
893 : /* if no satisfactory positions found, skip lexeme */
894 44 : if (!npos)
895 24 : continue;
896 :
897 20 : arrout[j].haspos = true;
898 20 : arrout[j].len = arrin[i].len;
899 20 : arrout[j].pos = cur_pos;
900 :
901 20 : memcpy(dataout + cur_pos, datain + arrin[i].pos, arrin[i].len);
902 20 : posvout->npos = npos;
903 20 : cur_pos += SHORTALIGN(arrin[i].len);
904 20 : cur_pos += POSDATALEN(tsout, arrout + j) * sizeof(WordEntryPos) +
905 : sizeof(uint16);
906 20 : j++;
907 : }
908 :
909 8 : tsout->size = j;
910 8 : if (dataout != STRPTR(tsout))
911 8 : memmove(STRPTR(tsout), dataout, cur_pos);
912 :
913 8 : SET_VARSIZE(tsout, CALCDATASIZE(tsout->size, cur_pos));
914 :
915 8 : PG_FREE_IF_COPY(tsin, 0);
916 8 : PG_RETURN_POINTER(tsout);
917 : }
918 :
919 : Datum
920 8 : tsvector_concat(PG_FUNCTION_ARGS)
921 : {
922 8 : TSVector in1 = PG_GETARG_TSVECTOR(0);
923 8 : TSVector in2 = PG_GETARG_TSVECTOR(1);
924 : TSVector out;
925 : WordEntry *ptr;
926 : WordEntry *ptr1,
927 : *ptr2;
928 : WordEntryPos *p;
929 8 : int maxpos = 0,
930 : i,
931 : j,
932 : i1,
933 : i2,
934 : dataoff,
935 : output_bytes,
936 : output_size;
937 : char *data,
938 : *data1,
939 : *data2;
940 :
941 : /* Get max position in in1; we'll need this to offset in2's positions */
942 8 : ptr = ARRPTR(in1);
943 8 : i = in1->size;
944 20 : while (i--)
945 : {
946 12 : if ((j = POSDATALEN(in1, ptr)) != 0)
947 : {
948 12 : p = POSDATAPTR(in1, ptr);
949 24 : while (j--)
950 : {
951 12 : if (WEP_GETPOS(*p) > maxpos)
952 8 : maxpos = WEP_GETPOS(*p);
953 12 : p++;
954 : }
955 : }
956 12 : ptr++;
957 : }
958 :
959 8 : ptr1 = ARRPTR(in1);
960 8 : ptr2 = ARRPTR(in2);
961 8 : data1 = STRPTR(in1);
962 8 : data2 = STRPTR(in2);
963 8 : i1 = in1->size;
964 8 : i2 = in2->size;
965 :
966 : /*
967 : * Conservative estimate of space needed. We might need all the data in
968 : * both inputs, and conceivably add a pad byte before position data for
969 : * each item where there was none before.
970 : */
971 8 : output_bytes = VARSIZE(in1) + VARSIZE(in2) + i1 + i2;
972 :
973 8 : out = (TSVector) palloc0(output_bytes);
974 8 : SET_VARSIZE(out, output_bytes);
975 :
976 : /*
977 : * We must make out->size valid so that STRPTR(out) is sensible. We'll
978 : * collapse out any unused space at the end.
979 : */
980 8 : out->size = in1->size + in2->size;
981 :
982 8 : ptr = ARRPTR(out);
983 8 : data = STRPTR(out);
984 8 : dataoff = 0;
985 20 : while (i1 && i2)
986 : {
987 12 : int cmp = compareEntry(data1, ptr1, data2, ptr2);
988 :
989 12 : if (cmp < 0)
990 : { /* in1 first */
991 4 : ptr->haspos = ptr1->haspos;
992 4 : ptr->len = ptr1->len;
993 4 : memcpy(data + dataoff, data1 + ptr1->pos, ptr1->len);
994 4 : ptr->pos = dataoff;
995 4 : dataoff += ptr1->len;
996 4 : if (ptr->haspos)
997 : {
998 4 : dataoff = SHORTALIGN(dataoff);
999 4 : memcpy(data + dataoff, _POSVECPTR(in1, ptr1), POSDATALEN(in1, ptr1) * sizeof(WordEntryPos) + sizeof(uint16));
1000 4 : dataoff += POSDATALEN(in1, ptr1) * sizeof(WordEntryPos) + sizeof(uint16);
1001 : }
1002 :
1003 4 : ptr++;
1004 4 : ptr1++;
1005 4 : i1--;
1006 : }
1007 8 : else if (cmp > 0)
1008 : { /* in2 first */
1009 4 : ptr->haspos = ptr2->haspos;
1010 4 : ptr->len = ptr2->len;
1011 4 : memcpy(data + dataoff, data2 + ptr2->pos, ptr2->len);
1012 4 : ptr->pos = dataoff;
1013 4 : dataoff += ptr2->len;
1014 4 : if (ptr->haspos)
1015 : {
1016 0 : int addlen = add_pos(in2, ptr2, out, ptr, maxpos);
1017 :
1018 0 : if (addlen == 0)
1019 0 : ptr->haspos = 0;
1020 : else
1021 : {
1022 0 : dataoff = SHORTALIGN(dataoff);
1023 0 : dataoff += addlen * sizeof(WordEntryPos) + sizeof(uint16);
1024 : }
1025 : }
1026 :
1027 4 : ptr++;
1028 4 : ptr2++;
1029 4 : i2--;
1030 : }
1031 : else
1032 : {
1033 4 : ptr->haspos = ptr1->haspos | ptr2->haspos;
1034 4 : ptr->len = ptr1->len;
1035 4 : memcpy(data + dataoff, data1 + ptr1->pos, ptr1->len);
1036 4 : ptr->pos = dataoff;
1037 4 : dataoff += ptr1->len;
1038 4 : if (ptr->haspos)
1039 : {
1040 4 : if (ptr1->haspos)
1041 : {
1042 4 : dataoff = SHORTALIGN(dataoff);
1043 4 : memcpy(data + dataoff, _POSVECPTR(in1, ptr1), POSDATALEN(in1, ptr1) * sizeof(WordEntryPos) + sizeof(uint16));
1044 4 : dataoff += POSDATALEN(in1, ptr1) * sizeof(WordEntryPos) + sizeof(uint16);
1045 4 : if (ptr2->haspos)
1046 4 : dataoff += add_pos(in2, ptr2, out, ptr, maxpos) * sizeof(WordEntryPos);
1047 : }
1048 : else /* must have ptr2->haspos */
1049 : {
1050 0 : int addlen = add_pos(in2, ptr2, out, ptr, maxpos);
1051 :
1052 0 : if (addlen == 0)
1053 0 : ptr->haspos = 0;
1054 : else
1055 : {
1056 0 : dataoff = SHORTALIGN(dataoff);
1057 0 : dataoff += addlen * sizeof(WordEntryPos) + sizeof(uint16);
1058 : }
1059 : }
1060 : }
1061 :
1062 4 : ptr++;
1063 4 : ptr1++;
1064 4 : ptr2++;
1065 4 : i1--;
1066 4 : i2--;
1067 : }
1068 : }
1069 :
1070 12 : while (i1)
1071 : {
1072 4 : ptr->haspos = ptr1->haspos;
1073 4 : ptr->len = ptr1->len;
1074 4 : memcpy(data + dataoff, data1 + ptr1->pos, ptr1->len);
1075 4 : ptr->pos = dataoff;
1076 4 : dataoff += ptr1->len;
1077 4 : if (ptr->haspos)
1078 : {
1079 4 : dataoff = SHORTALIGN(dataoff);
1080 4 : memcpy(data + dataoff, _POSVECPTR(in1, ptr1), POSDATALEN(in1, ptr1) * sizeof(WordEntryPos) + sizeof(uint16));
1081 4 : dataoff += POSDATALEN(in1, ptr1) * sizeof(WordEntryPos) + sizeof(uint16);
1082 : }
1083 :
1084 4 : ptr++;
1085 4 : ptr1++;
1086 4 : i1--;
1087 : }
1088 :
1089 12 : while (i2)
1090 : {
1091 4 : ptr->haspos = ptr2->haspos;
1092 4 : ptr->len = ptr2->len;
1093 4 : memcpy(data + dataoff, data2 + ptr2->pos, ptr2->len);
1094 4 : ptr->pos = dataoff;
1095 4 : dataoff += ptr2->len;
1096 4 : if (ptr->haspos)
1097 : {
1098 4 : int addlen = add_pos(in2, ptr2, out, ptr, maxpos);
1099 :
1100 4 : if (addlen == 0)
1101 0 : ptr->haspos = 0;
1102 : else
1103 : {
1104 4 : dataoff = SHORTALIGN(dataoff);
1105 4 : dataoff += addlen * sizeof(WordEntryPos) + sizeof(uint16);
1106 : }
1107 : }
1108 :
1109 4 : ptr++;
1110 4 : ptr2++;
1111 4 : i2--;
1112 : }
1113 :
1114 : /*
1115 : * Instead of checking each offset individually, we check for overflow of
1116 : * pos fields once at the end.
1117 : */
1118 8 : if (dataoff > MAXSTRPOS)
1119 0 : ereport(ERROR,
1120 : (errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
1121 : errmsg("string is too long for tsvector (%d bytes, max %d bytes)", dataoff, MAXSTRPOS)));
1122 :
1123 : /*
1124 : * Adjust sizes (asserting that we didn't overrun the original estimates)
1125 : * and collapse out any unused array entries.
1126 : */
1127 8 : output_size = ptr - ARRPTR(out);
1128 : Assert(output_size <= out->size);
1129 8 : out->size = output_size;
1130 8 : if (data != STRPTR(out))
1131 4 : memmove(STRPTR(out), data, dataoff);
1132 8 : output_bytes = CALCDATASIZE(out->size, dataoff);
1133 : Assert(output_bytes <= VARSIZE(out));
1134 8 : SET_VARSIZE(out, output_bytes);
1135 :
1136 8 : PG_FREE_IF_COPY(in1, 0);
1137 8 : PG_FREE_IF_COPY(in2, 1);
1138 8 : PG_RETURN_POINTER(out);
1139 : }
1140 :
1141 : /*
1142 : * Compare two strings by tsvector rules.
1143 : *
1144 : * if prefix = true then it returns zero value iff b has prefix a
1145 : */
1146 : int32
1147 4143838 : tsCompareString(char *a, int lena, char *b, int lenb, bool prefix)
1148 : {
1149 : int cmp;
1150 :
1151 4143838 : if (lena == 0)
1152 : {
1153 0 : if (prefix)
1154 0 : cmp = 0; /* empty string is prefix of anything */
1155 : else
1156 0 : cmp = (lenb > 0) ? -1 : 0;
1157 : }
1158 4143838 : else if (lenb == 0)
1159 : {
1160 0 : cmp = (lena > 0) ? 1 : 0;
1161 : }
1162 : else
1163 : {
1164 4143838 : cmp = memcmp(a, b, Min(lena, lenb));
1165 :
1166 4143838 : if (prefix)
1167 : {
1168 10972 : if (cmp == 0 && lena > lenb)
1169 0 : cmp = 1; /* a is longer, so not a prefix of b */
1170 : }
1171 4132866 : else if (cmp == 0 && lena != lenb)
1172 : {
1173 21500 : cmp = (lena < lenb) ? -1 : 1;
1174 : }
1175 : }
1176 :
1177 4143838 : return cmp;
1178 : }
1179 :
1180 : /*
1181 : * Check weight info or/and fill 'data' with the required positions
1182 : */
1183 : static TSTernaryValue
1184 45412 : checkclass_str(CHKVAL *chkval, WordEntry *entry, QueryOperand *val,
1185 : ExecPhraseData *data)
1186 : {
1187 45412 : TSTernaryValue result = TS_NO;
1188 :
1189 : Assert(data == NULL || data->npos == 0);
1190 :
1191 45412 : if (entry->haspos)
1192 : {
1193 : WordEntryPosVector *posvec;
1194 :
1195 : /*
1196 : * We can't use the _POSVECPTR macro here because the pointer to the
1197 : * tsvector's lexeme storage is already contained in chkval->values.
1198 : */
1199 2992 : posvec = (WordEntryPosVector *)
1200 2992 : (chkval->values + SHORTALIGN(entry->pos + entry->len));
1201 :
1202 2992 : if (val->weight && data)
1203 32 : {
1204 32 : WordEntryPos *posvec_iter = posvec->pos;
1205 : WordEntryPos *dptr;
1206 :
1207 : /*
1208 : * Filter position information by weights
1209 : */
1210 32 : dptr = data->pos = palloc(sizeof(WordEntryPos) * posvec->npos);
1211 32 : data->allocated = true;
1212 :
1213 : /* Is there a position with a matching weight? */
1214 64 : while (posvec_iter < posvec->pos + posvec->npos)
1215 : {
1216 : /* If true, append this position to the data->pos */
1217 32 : if (val->weight & (1 << WEP_GETWEIGHT(*posvec_iter)))
1218 : {
1219 16 : *dptr = WEP_GETPOS(*posvec_iter);
1220 16 : dptr++;
1221 : }
1222 :
1223 32 : posvec_iter++;
1224 : }
1225 :
1226 32 : data->npos = dptr - data->pos;
1227 :
1228 32 : if (data->npos > 0)
1229 16 : result = TS_YES;
1230 : else
1231 : {
1232 16 : pfree(data->pos);
1233 16 : data->pos = NULL;
1234 16 : data->allocated = false;
1235 : }
1236 : }
1237 2960 : else if (val->weight)
1238 : {
1239 304 : WordEntryPos *posvec_iter = posvec->pos;
1240 :
1241 : /* Is there a position with a matching weight? */
1242 460 : while (posvec_iter < posvec->pos + posvec->npos)
1243 : {
1244 336 : if (val->weight & (1 << WEP_GETWEIGHT(*posvec_iter)))
1245 : {
1246 180 : result = TS_YES;
1247 180 : break; /* no need to go further */
1248 : }
1249 :
1250 156 : posvec_iter++;
1251 : }
1252 : }
1253 2656 : else if (data)
1254 : {
1255 1516 : data->npos = posvec->npos;
1256 1516 : data->pos = posvec->pos;
1257 1516 : data->allocated = false;
1258 1516 : result = TS_YES;
1259 : }
1260 : else
1261 : {
1262 : /* simplest case: no weight check, positions not needed */
1263 1140 : result = TS_YES;
1264 : }
1265 : }
1266 : else
1267 : {
1268 : /*
1269 : * Position info is lacking, so if the caller requires it, we can only
1270 : * say that maybe there is a match.
1271 : *
1272 : * Notice, however, that we *don't* check val->weight here.
1273 : * Historically, stripped tsvectors are considered to match queries
1274 : * whether or not the query has a weight restriction; that's a little
1275 : * dubious but we'll preserve the behavior.
1276 : */
1277 42420 : if (data)
1278 15396 : result = TS_MAYBE;
1279 : else
1280 27024 : result = TS_YES;
1281 : }
1282 :
1283 45412 : return result;
1284 : }
1285 :
1286 : /*
1287 : * TS_execute callback for matching a tsquery operand to plain tsvector data
1288 : */
1289 : static TSTernaryValue
1290 189788 : checkcondition_str(void *checkval, QueryOperand *val, ExecPhraseData *data)
1291 : {
1292 189788 : CHKVAL *chkval = (CHKVAL *) checkval;
1293 189788 : WordEntry *StopLow = chkval->arrb;
1294 189788 : WordEntry *StopHigh = chkval->arre;
1295 189788 : WordEntry *StopMiddle = StopHigh;
1296 189788 : TSTernaryValue res = TS_NO;
1297 :
1298 : /* Loop invariant: StopLow <= val < StopHigh */
1299 1193668 : while (StopLow < StopHigh)
1300 : {
1301 : int difference;
1302 :
1303 1039268 : StopMiddle = StopLow + (StopHigh - StopLow) / 2;
1304 3117804 : difference = tsCompareString(chkval->operand + val->distance,
1305 1039268 : val->length,
1306 1039268 : chkval->values + StopMiddle->pos,
1307 1039268 : StopMiddle->len,
1308 : false);
1309 :
1310 1039268 : if (difference == 0)
1311 : {
1312 : /* Check weight info & fill 'data' with positions */
1313 35388 : res = checkclass_str(chkval, StopMiddle, val, data);
1314 35388 : break;
1315 : }
1316 1003880 : else if (difference > 0)
1317 566180 : StopLow = StopMiddle + 1;
1318 : else
1319 437700 : StopHigh = StopMiddle;
1320 : }
1321 :
1322 : /*
1323 : * If it's a prefix search, we should also consider lexemes that the
1324 : * search term is a prefix of (which will necessarily immediately follow
1325 : * the place we found in the above loop). But we can skip them if there
1326 : * was a definite match on the exact term AND the caller doesn't need
1327 : * position info.
1328 : */
1329 189788 : if (val->prefix && (res != TS_YES || data))
1330 : {
1331 11016 : WordEntryPos *allpos = NULL;
1332 11016 : int npos = 0,
1333 11016 : totalpos = 0;
1334 :
1335 : /* adjust start position for corner case */
1336 11016 : if (StopLow >= StopHigh)
1337 11008 : StopMiddle = StopHigh;
1338 :
1339 : /* we don't try to re-use any data from the initial match */
1340 11016 : if (data)
1341 : {
1342 24 : if (data->allocated)
1343 0 : pfree(data->pos);
1344 24 : data->pos = NULL;
1345 24 : data->allocated = false;
1346 24 : data->npos = 0;
1347 : }
1348 11016 : res = TS_NO;
1349 :
1350 21040 : while ((res != TS_YES || data) &&
1351 21704 : StopMiddle < chkval->arre &&
1352 31860 : tsCompareString(chkval->operand + val->distance,
1353 10620 : val->length,
1354 10620 : chkval->values + StopMiddle->pos,
1355 10620 : StopMiddle->len,
1356 : true) == 0)
1357 : {
1358 : TSTernaryValue subres;
1359 :
1360 10024 : subres = checkclass_str(chkval, StopMiddle, val, data);
1361 :
1362 10024 : if (subres != TS_NO)
1363 : {
1364 9984 : if (data)
1365 : {
1366 : /*
1367 : * We need to join position information
1368 : */
1369 28 : if (subres == TS_MAYBE)
1370 : {
1371 : /*
1372 : * No position info for this match, so we must report
1373 : * MAYBE overall.
1374 : */
1375 0 : res = TS_MAYBE;
1376 : /* forget any previous positions */
1377 0 : npos = 0;
1378 : /* don't leak storage */
1379 0 : if (allpos)
1380 0 : pfree(allpos);
1381 0 : break;
1382 : }
1383 :
1384 52 : while (npos + data->npos > totalpos)
1385 : {
1386 24 : if (totalpos == 0)
1387 : {
1388 24 : totalpos = 256;
1389 24 : allpos = palloc(sizeof(WordEntryPos) * totalpos);
1390 : }
1391 : else
1392 : {
1393 0 : totalpos *= 2;
1394 0 : allpos = repalloc(allpos, sizeof(WordEntryPos) * totalpos);
1395 : }
1396 : }
1397 :
1398 28 : memcpy(allpos + npos, data->pos, sizeof(WordEntryPos) * data->npos);
1399 28 : npos += data->npos;
1400 :
1401 : /* don't leak storage from individual matches */
1402 28 : if (data->allocated)
1403 16 : pfree(data->pos);
1404 28 : data->pos = NULL;
1405 28 : data->allocated = false;
1406 : /* it's important to reset data->npos before next loop */
1407 28 : data->npos = 0;
1408 : }
1409 : else
1410 : {
1411 : /* Don't need positions, just handle YES/MAYBE */
1412 9956 : if (subres == TS_YES || res == TS_NO)
1413 9956 : res = subres;
1414 : }
1415 : }
1416 :
1417 10024 : StopMiddle++;
1418 : }
1419 :
1420 11016 : if (data && npos > 0)
1421 : {
1422 : /* Sort and make unique array of found positions */
1423 24 : data->pos = allpos;
1424 24 : qsort(data->pos, npos, sizeof(WordEntryPos), compareWordEntryPos);
1425 24 : data->npos = qunique(data->pos, npos, sizeof(WordEntryPos),
1426 : compareWordEntryPos);
1427 24 : data->allocated = true;
1428 24 : res = TS_YES;
1429 : }
1430 : }
1431 :
1432 189788 : return res;
1433 : }
1434 :
1435 : /*
1436 : * Compute output position list for a tsquery operator in phrase mode.
1437 : *
1438 : * Merge the position lists in Ldata and Rdata as specified by "emit",
1439 : * returning the result list into *data. The input position lists must be
1440 : * sorted and unique, and the output will be as well.
1441 : *
1442 : * data: pointer to initially-all-zeroes output struct, or NULL
1443 : * Ldata, Rdata: input position lists
1444 : * emit: bitmask of TSPO_XXX flags
1445 : * Loffset: offset to be added to Ldata positions before comparing/outputting
1446 : * Roffset: offset to be added to Rdata positions before comparing/outputting
1447 : * max_npos: maximum possible required size of output position array
1448 : *
1449 : * Loffset and Roffset should not be negative, else we risk trying to output
1450 : * negative positions, which won't fit into WordEntryPos.
1451 : *
1452 : * The result is boolean (TS_YES or TS_NO), but for the caller's convenience
1453 : * we return it as TSTernaryValue.
1454 : *
1455 : * Returns TS_YES if any positions were emitted to *data; or if data is NULL,
1456 : * returns TS_YES if any positions would have been emitted.
1457 : */
1458 : #define TSPO_L_ONLY 0x01 /* emit positions appearing only in L */
1459 : #define TSPO_R_ONLY 0x02 /* emit positions appearing only in R */
1460 : #define TSPO_BOTH 0x04 /* emit positions appearing in both L&R */
1461 :
1462 : static TSTernaryValue
1463 19964 : TS_phrase_output(ExecPhraseData *data,
1464 : ExecPhraseData *Ldata,
1465 : ExecPhraseData *Rdata,
1466 : int emit,
1467 : int Loffset,
1468 : int Roffset,
1469 : int max_npos)
1470 : {
1471 : int Lindex,
1472 : Rindex;
1473 :
1474 : /* Loop until both inputs are exhausted */
1475 19964 : Lindex = Rindex = 0;
1476 20616 : while (Lindex < Ldata->npos || Rindex < Rdata->npos)
1477 : {
1478 : int Lpos,
1479 : Rpos;
1480 1536 : int output_pos = 0;
1481 :
1482 : /*
1483 : * Fetch current values to compare. WEP_GETPOS() is needed because
1484 : * ExecPhraseData->data can point to a tsvector's WordEntryPosVector.
1485 : */
1486 1536 : if (Lindex < Ldata->npos)
1487 1108 : Lpos = WEP_GETPOS(Ldata->pos[Lindex]) + Loffset;
1488 : else
1489 : {
1490 : /* L array exhausted, so we're done if R_ONLY isn't set */
1491 428 : if (!(emit & TSPO_R_ONLY))
1492 104 : break;
1493 324 : Lpos = INT_MAX;
1494 : }
1495 1432 : if (Rindex < Rdata->npos)
1496 1260 : Rpos = WEP_GETPOS(Rdata->pos[Rindex]) + Roffset;
1497 : else
1498 : {
1499 : /* R array exhausted, so we're done if L_ONLY isn't set */
1500 172 : if (!(emit & TSPO_L_ONLY))
1501 116 : break;
1502 56 : Rpos = INT_MAX;
1503 : }
1504 :
1505 : /* Merge-join the two input lists */
1506 1316 : if (Lpos < Rpos)
1507 : {
1508 : /* Lpos is not matched in Rdata, should we output it? */
1509 320 : if (emit & TSPO_L_ONLY)
1510 76 : output_pos = Lpos;
1511 320 : Lindex++;
1512 : }
1513 996 : else if (Lpos == Rpos)
1514 : {
1515 : /* Lpos and Rpos match ... should we output it? */
1516 496 : if (emit & TSPO_BOTH)
1517 432 : output_pos = Rpos;
1518 496 : Lindex++;
1519 496 : Rindex++;
1520 : }
1521 : else /* Lpos > Rpos */
1522 : {
1523 : /* Rpos is not matched in Ldata, should we output it? */
1524 500 : if (emit & TSPO_R_ONLY)
1525 352 : output_pos = Rpos;
1526 500 : Rindex++;
1527 : }
1528 :
1529 1316 : if (output_pos > 0)
1530 : {
1531 860 : if (data)
1532 : {
1533 : /* Store position, first allocating output array if needed */
1534 196 : if (data->pos == NULL)
1535 : {
1536 164 : data->pos = (WordEntryPos *)
1537 164 : palloc(max_npos * sizeof(WordEntryPos));
1538 164 : data->allocated = true;
1539 : }
1540 196 : data->pos[data->npos++] = output_pos;
1541 : }
1542 : else
1543 : {
1544 : /*
1545 : * Exact positions not needed, so return TS_YES as soon as we
1546 : * know there is at least one.
1547 : */
1548 664 : return TS_YES;
1549 : }
1550 : }
1551 : }
1552 :
1553 19300 : if (data && data->npos > 0)
1554 : {
1555 : /* Let's assert we didn't overrun the array */
1556 : Assert(data->npos <= max_npos);
1557 164 : return TS_YES;
1558 : }
1559 19136 : return TS_NO;
1560 : }
1561 :
1562 : /*
1563 : * Execute tsquery at or below an OP_PHRASE operator.
1564 : *
1565 : * This handles tsquery execution at recursion levels where we need to care
1566 : * about match locations.
1567 : *
1568 : * In addition to the same arguments used for TS_execute, the caller may pass
1569 : * a preinitialized-to-zeroes ExecPhraseData struct, to be filled with lexeme
1570 : * match position info on success. data == NULL if no position data need be
1571 : * returned. (In practice, outside callers pass NULL, and only the internal
1572 : * recursion cases pass a data pointer.)
1573 : * Note: the function assumes data != NULL for operators other than OP_PHRASE.
1574 : * This is OK because an outside call always starts from an OP_PHRASE node.
1575 : *
1576 : * The detailed semantics of the match data, given that the function returned
1577 : * TS_YES (successful match), are:
1578 : *
1579 : * npos > 0, negate = false:
1580 : * query is matched at specified position(s) (and only those positions)
1581 : * npos > 0, negate = true:
1582 : * query is matched at all positions *except* specified position(s)
1583 : * npos = 0, negate = true:
1584 : * query is matched at all positions
1585 : * npos = 0, negate = false:
1586 : * disallowed (this should result in TS_NO or TS_MAYBE, as appropriate)
1587 : *
1588 : * Successful matches also return a "width" value which is the match width in
1589 : * lexemes, less one. Hence, "width" is zero for simple one-lexeme matches,
1590 : * and is the sum of the phrase operator distances for phrase matches. Note
1591 : * that when width > 0, the listed positions represent the ends of matches not
1592 : * the starts. (This unintuitive rule is needed to avoid possibly generating
1593 : * negative positions, which wouldn't fit into the WordEntryPos arrays.)
1594 : *
1595 : * If the TSExecuteCallback function reports that an operand is present
1596 : * but fails to provide position(s) for it, we will return TS_MAYBE when
1597 : * it is possible but not certain that the query is matched.
1598 : *
1599 : * When the function returns TS_NO or TS_MAYBE, it must return npos = 0,
1600 : * negate = false (which is the state initialized by the caller); but the
1601 : * "width" output in such cases is undefined.
1602 : */
1603 : static TSTernaryValue
1604 469190 : TS_phrase_execute(QueryItem *curitem, void *arg, uint32 flags,
1605 : TSExecuteCallback chkcond,
1606 : ExecPhraseData *data)
1607 : {
1608 : ExecPhraseData Ldata,
1609 : Rdata;
1610 : TSTernaryValue lmatch,
1611 : rmatch;
1612 : int Loffset,
1613 : Roffset,
1614 : maxwidth;
1615 :
1616 : /* since this function recurses, it could be driven to stack overflow */
1617 469190 : check_stack_depth();
1618 :
1619 469190 : if (curitem->type == QI_VAL)
1620 230826 : return chkcond(arg, (QueryOperand *) curitem, data);
1621 :
1622 238364 : switch (curitem->qoperator.oper)
1623 : {
1624 80554 : case OP_NOT:
1625 :
1626 : /*
1627 : * We need not touch data->width, since a NOT operation does not
1628 : * change the match width.
1629 : */
1630 80554 : if (flags & TS_EXEC_SKIP_NOT)
1631 : {
1632 : /* with SKIP_NOT, report NOT as "match everywhere" */
1633 : Assert(data->npos == 0 && !data->negate);
1634 0 : data->negate = true;
1635 0 : return TS_YES;
1636 : }
1637 80554 : switch (TS_phrase_execute(curitem + 1, arg, flags, chkcond, data))
1638 : {
1639 70414 : case TS_NO:
1640 : /* change "match nowhere" to "match everywhere" */
1641 : Assert(data->npos == 0 && !data->negate);
1642 70414 : data->negate = true;
1643 70414 : return TS_YES;
1644 260 : case TS_YES:
1645 260 : if (data->npos > 0)
1646 : {
1647 : /* we have some positions, invert negate flag */
1648 256 : data->negate = !data->negate;
1649 256 : return TS_YES;
1650 : }
1651 4 : else if (data->negate)
1652 : {
1653 : /* change "match everywhere" to "match nowhere" */
1654 4 : data->negate = false;
1655 4 : return TS_NO;
1656 : }
1657 : /* Should not get here if result was TS_YES */
1658 : Assert(false);
1659 0 : break;
1660 9880 : case TS_MAYBE:
1661 : /* match positions are, and remain, uncertain */
1662 9880 : return TS_MAYBE;
1663 : }
1664 0 : break;
1665 :
1666 157706 : case OP_PHRASE:
1667 : case OP_AND:
1668 157706 : memset(&Ldata, 0, sizeof(Ldata));
1669 157706 : memset(&Rdata, 0, sizeof(Rdata));
1670 :
1671 157706 : lmatch = TS_phrase_execute(curitem + curitem->qoperator.left,
1672 : arg, flags, chkcond, &Ldata);
1673 157706 : if (lmatch == TS_NO)
1674 84510 : return TS_NO;
1675 :
1676 73196 : rmatch = TS_phrase_execute(curitem + 1,
1677 : arg, flags, chkcond, &Rdata);
1678 73196 : if (rmatch == TS_NO)
1679 36168 : return TS_NO;
1680 :
1681 : /*
1682 : * If either operand has no position information, then we can't
1683 : * return reliable position data, only a MAYBE result.
1684 : */
1685 37028 : if (lmatch == TS_MAYBE || rmatch == TS_MAYBE)
1686 17160 : return TS_MAYBE;
1687 :
1688 19868 : if (curitem->qoperator.oper == OP_PHRASE)
1689 : {
1690 : /*
1691 : * Compute Loffset and Roffset suitable for phrase match, and
1692 : * compute overall width of whole phrase match.
1693 : */
1694 19864 : Loffset = curitem->qoperator.distance + Rdata.width;
1695 19864 : Roffset = 0;
1696 19864 : if (data)
1697 252 : data->width = curitem->qoperator.distance +
1698 84 : Ldata.width + Rdata.width;
1699 : }
1700 : else
1701 : {
1702 : /*
1703 : * For OP_AND, set output width and alignment like OP_OR (see
1704 : * comment below)
1705 : */
1706 4 : maxwidth = Max(Ldata.width, Rdata.width);
1707 4 : Loffset = maxwidth - Ldata.width;
1708 4 : Roffset = maxwidth - Rdata.width;
1709 4 : if (data)
1710 4 : data->width = maxwidth;
1711 : }
1712 :
1713 19868 : if (Ldata.negate && Rdata.negate)
1714 : {
1715 : /* !L & !R: treat as !(L | R) */
1716 18960 : (void) TS_phrase_output(data, &Ldata, &Rdata,
1717 : TSPO_BOTH | TSPO_L_ONLY | TSPO_R_ONLY,
1718 : Loffset, Roffset,
1719 18960 : Ldata.npos + Rdata.npos);
1720 18960 : if (data)
1721 0 : data->negate = true;
1722 18960 : return TS_YES;
1723 : }
1724 908 : else if (Ldata.negate)
1725 : {
1726 : /* !L & R */
1727 300 : return TS_phrase_output(data, &Ldata, &Rdata,
1728 : TSPO_R_ONLY,
1729 : Loffset, Roffset,
1730 : Rdata.npos);
1731 : }
1732 608 : else if (Rdata.negate)
1733 : {
1734 : /* L & !R */
1735 4 : return TS_phrase_output(data, &Ldata, &Rdata,
1736 : TSPO_L_ONLY,
1737 : Loffset, Roffset,
1738 : Ldata.npos);
1739 : }
1740 : else
1741 : {
1742 : /* straight AND */
1743 604 : return TS_phrase_output(data, &Ldata, &Rdata,
1744 : TSPO_BOTH,
1745 : Loffset, Roffset,
1746 604 : Min(Ldata.npos, Rdata.npos));
1747 : }
1748 :
1749 104 : case OP_OR:
1750 104 : memset(&Ldata, 0, sizeof(Ldata));
1751 104 : memset(&Rdata, 0, sizeof(Rdata));
1752 :
1753 104 : lmatch = TS_phrase_execute(curitem + curitem->qoperator.left,
1754 : arg, flags, chkcond, &Ldata);
1755 104 : rmatch = TS_phrase_execute(curitem + 1,
1756 : arg, flags, chkcond, &Rdata);
1757 :
1758 104 : if (lmatch == TS_NO && rmatch == TS_NO)
1759 8 : return TS_NO;
1760 :
1761 : /*
1762 : * If either operand has no position information, then we can't
1763 : * return reliable position data, only a MAYBE result.
1764 : */
1765 96 : if (lmatch == TS_MAYBE || rmatch == TS_MAYBE)
1766 0 : return TS_MAYBE;
1767 :
1768 : /*
1769 : * Cope with undefined output width from failed submatch. (This
1770 : * takes less code than trying to ensure that all failure returns
1771 : * set data->width to zero.)
1772 : */
1773 96 : if (lmatch == TS_NO)
1774 12 : Ldata.width = 0;
1775 96 : if (rmatch == TS_NO)
1776 56 : Rdata.width = 0;
1777 :
1778 : /*
1779 : * For OP_AND and OP_OR, report the width of the wider of the two
1780 : * inputs, and align the narrower input's positions to the right
1781 : * end of that width. This rule deals at least somewhat
1782 : * reasonably with cases like "x <-> (y | z <-> q)".
1783 : */
1784 96 : maxwidth = Max(Ldata.width, Rdata.width);
1785 96 : Loffset = maxwidth - Ldata.width;
1786 96 : Roffset = maxwidth - Rdata.width;
1787 96 : data->width = maxwidth;
1788 :
1789 96 : if (Ldata.negate && Rdata.negate)
1790 : {
1791 : /* !L | !R: treat as !(L & R) */
1792 4 : (void) TS_phrase_output(data, &Ldata, &Rdata,
1793 : TSPO_BOTH,
1794 : Loffset, Roffset,
1795 4 : Min(Ldata.npos, Rdata.npos));
1796 4 : data->negate = true;
1797 4 : return TS_YES;
1798 : }
1799 92 : else if (Ldata.negate)
1800 : {
1801 : /* !L | R: treat as !(L & !R) */
1802 20 : (void) TS_phrase_output(data, &Ldata, &Rdata,
1803 : TSPO_L_ONLY,
1804 : Loffset, Roffset,
1805 : Ldata.npos);
1806 20 : data->negate = true;
1807 20 : return TS_YES;
1808 : }
1809 72 : else if (Rdata.negate)
1810 : {
1811 : /* L | !R: treat as !(!L & R) */
1812 4 : (void) TS_phrase_output(data, &Ldata, &Rdata,
1813 : TSPO_R_ONLY,
1814 : Loffset, Roffset,
1815 : Rdata.npos);
1816 4 : data->negate = true;
1817 4 : return TS_YES;
1818 : }
1819 : else
1820 : {
1821 : /* straight OR */
1822 68 : return TS_phrase_output(data, &Ldata, &Rdata,
1823 : TSPO_BOTH | TSPO_L_ONLY | TSPO_R_ONLY,
1824 : Loffset, Roffset,
1825 68 : Ldata.npos + Rdata.npos);
1826 : }
1827 :
1828 0 : default:
1829 0 : elog(ERROR, "unrecognized operator: %d", curitem->qoperator.oper);
1830 : }
1831 :
1832 : /* not reachable, but keep compiler quiet */
1833 0 : return TS_NO;
1834 : }
1835 :
1836 :
1837 : /*
1838 : * Evaluate tsquery boolean expression.
1839 : *
1840 : * curitem: current tsquery item (initially, the first one)
1841 : * arg: opaque value to pass through to callback function
1842 : * flags: bitmask of flag bits shown in ts_utils.h
1843 : * chkcond: callback function to check whether a primitive value is present
1844 : */
1845 : bool
1846 372310 : TS_execute(QueryItem *curitem, void *arg, uint32 flags,
1847 : TSExecuteCallback chkcond)
1848 : {
1849 : /*
1850 : * If we get TS_MAYBE from the recursion, return true. We could only see
1851 : * that result if the caller passed TS_EXEC_PHRASE_NO_POS, so there's no
1852 : * need to check again.
1853 : */
1854 372310 : return TS_execute_recurse(curitem, arg, flags, chkcond) != TS_NO;
1855 : }
1856 :
1857 : /*
1858 : * TS_execute recursion for operators above any phrase operator. Here we do
1859 : * not need to worry about lexeme positions. As soon as we hit an OP_PHRASE
1860 : * operator, we pass it off to TS_phrase_execute which does worry.
1861 : */
1862 : static TSTernaryValue
1863 704840 : TS_execute_recurse(QueryItem *curitem, void *arg, uint32 flags,
1864 : TSExecuteCallback chkcond)
1865 : {
1866 : TSTernaryValue lmatch;
1867 :
1868 : /* since this function recurses, it could be driven to stack overflow */
1869 704840 : check_stack_depth();
1870 :
1871 : /* ... and let's check for query cancel while we're at it */
1872 704840 : CHECK_FOR_INTERRUPTS();
1873 :
1874 704840 : if (curitem->type == QI_VAL)
1875 282760 : return chkcond(arg, (QueryOperand *) curitem,
1876 : NULL /* don't need position info */ );
1877 :
1878 422080 : switch (curitem->qoperator.oper)
1879 : {
1880 135868 : case OP_NOT:
1881 135868 : if (flags & TS_EXEC_SKIP_NOT)
1882 0 : return TS_YES;
1883 135868 : switch (TS_execute_recurse(curitem + 1, arg, flags, chkcond))
1884 : {
1885 128216 : case TS_NO:
1886 128216 : return TS_YES;
1887 3260 : case TS_YES:
1888 3260 : return TS_NO;
1889 4392 : case TS_MAYBE:
1890 4392 : return TS_MAYBE;
1891 : }
1892 0 : break;
1893 :
1894 56150 : case OP_AND:
1895 56150 : lmatch = TS_execute_recurse(curitem + curitem->qoperator.left, arg,
1896 : flags, chkcond);
1897 56150 : if (lmatch == TS_NO)
1898 44630 : return TS_NO;
1899 11520 : switch (TS_execute_recurse(curitem + 1, arg, flags, chkcond))
1900 : {
1901 6840 : case TS_NO:
1902 6840 : return TS_NO;
1903 2200 : case TS_YES:
1904 2200 : return lmatch;
1905 2480 : case TS_MAYBE:
1906 2480 : return TS_MAYBE;
1907 : }
1908 0 : break;
1909 :
1910 72536 : case OP_OR:
1911 72536 : lmatch = TS_execute_recurse(curitem + curitem->qoperator.left, arg,
1912 : flags, chkcond);
1913 72536 : if (lmatch == TS_YES)
1914 16080 : return TS_YES;
1915 56456 : switch (TS_execute_recurse(curitem + 1, arg, flags, chkcond))
1916 : {
1917 38338 : case TS_NO:
1918 38338 : return lmatch;
1919 4928 : case TS_YES:
1920 4928 : return TS_YES;
1921 13190 : case TS_MAYBE:
1922 13190 : return TS_MAYBE;
1923 : }
1924 0 : break;
1925 :
1926 157526 : case OP_PHRASE:
1927 :
1928 : /*
1929 : * If we get a MAYBE result, and the caller doesn't want that,
1930 : * convert it to NO. It would be more consistent, perhaps, to
1931 : * return the result of TS_phrase_execute() verbatim and then
1932 : * convert MAYBE results at the top of the recursion. But
1933 : * converting at the topmost phrase operator gives results that
1934 : * are bug-compatible with the old implementation, so do it like
1935 : * this for now.
1936 : */
1937 157526 : switch (TS_phrase_execute(curitem, arg, flags, chkcond, NULL))
1938 : {
1939 120822 : case TS_NO:
1940 120822 : return TS_NO;
1941 19548 : case TS_YES:
1942 19548 : return TS_YES;
1943 17156 : case TS_MAYBE:
1944 17156 : return (flags & TS_EXEC_PHRASE_NO_POS) ? TS_MAYBE : TS_NO;
1945 : }
1946 0 : break;
1947 :
1948 0 : default:
1949 0 : elog(ERROR, "unrecognized operator: %d", curitem->qoperator.oper);
1950 : }
1951 :
1952 : /* not reachable, but keep compiler quiet */
1953 0 : return TS_NO;
1954 : }
1955 :
1956 : /*
1957 : * Detect whether a tsquery boolean expression requires any positive matches
1958 : * to values shown in the tsquery.
1959 : *
1960 : * This is needed to know whether a GIN index search requires full index scan.
1961 : * For example, 'x & !y' requires a match of x, so it's sufficient to scan
1962 : * entries for x; but 'x | !y' could match rows containing neither x nor y.
1963 : */
1964 : bool
1965 556 : tsquery_requires_match(QueryItem *curitem)
1966 : {
1967 : /* since this function recurses, it could be driven to stack overflow */
1968 556 : check_stack_depth();
1969 :
1970 556 : if (curitem->type == QI_VAL)
1971 264 : return true;
1972 :
1973 292 : switch (curitem->qoperator.oper)
1974 : {
1975 112 : case OP_NOT:
1976 :
1977 : /*
1978 : * Assume there are no required matches underneath a NOT. For
1979 : * some cases with nested NOTs, we could prove there's a required
1980 : * match, but it seems unlikely to be worth the trouble.
1981 : */
1982 112 : return false;
1983 :
1984 136 : case OP_PHRASE:
1985 :
1986 : /*
1987 : * Treat OP_PHRASE as OP_AND here
1988 : */
1989 : case OP_AND:
1990 : /* If either side requires a match, we're good */
1991 136 : if (tsquery_requires_match(curitem + curitem->qoperator.left))
1992 104 : return true;
1993 : else
1994 32 : return tsquery_requires_match(curitem + 1);
1995 :
1996 44 : case OP_OR:
1997 : /* Both sides must require a match */
1998 44 : if (tsquery_requires_match(curitem + curitem->qoperator.left))
1999 44 : return tsquery_requires_match(curitem + 1);
2000 : else
2001 0 : return false;
2002 :
2003 0 : default:
2004 0 : elog(ERROR, "unrecognized operator: %d", curitem->qoperator.oper);
2005 : }
2006 :
2007 : /* not reachable, but keep compiler quiet */
2008 : return false;
2009 : }
2010 :
2011 : /*
2012 : * boolean operations
2013 : */
2014 : Datum
2015 40 : ts_match_qv(PG_FUNCTION_ARGS)
2016 : {
2017 40 : PG_RETURN_DATUM(DirectFunctionCall2(ts_match_vq,
2018 : PG_GETARG_DATUM(1),
2019 : PG_GETARG_DATUM(0)));
2020 : }
2021 :
2022 : Datum
2023 147136 : ts_match_vq(PG_FUNCTION_ARGS)
2024 : {
2025 147136 : TSVector val = PG_GETARG_TSVECTOR(0);
2026 147136 : TSQuery query = PG_GETARG_TSQUERY(1);
2027 : CHKVAL chkval;
2028 : bool result;
2029 :
2030 : /* empty query matches nothing */
2031 147136 : if (!query->size)
2032 : {
2033 0 : PG_FREE_IF_COPY(val, 0);
2034 0 : PG_FREE_IF_COPY(query, 1);
2035 0 : PG_RETURN_BOOL(false);
2036 : }
2037 :
2038 147136 : chkval.arrb = ARRPTR(val);
2039 147136 : chkval.arre = chkval.arrb + val->size;
2040 147136 : chkval.values = STRPTR(val);
2041 147136 : chkval.operand = GETOPERAND(query);
2042 147136 : result = TS_execute(GETQUERY(query),
2043 : &chkval,
2044 : TS_EXEC_EMPTY,
2045 : checkcondition_str);
2046 :
2047 147136 : PG_FREE_IF_COPY(val, 0);
2048 147136 : PG_FREE_IF_COPY(query, 1);
2049 147136 : PG_RETURN_BOOL(result);
2050 : }
2051 :
2052 : Datum
2053 0 : ts_match_tt(PG_FUNCTION_ARGS)
2054 : {
2055 : TSVector vector;
2056 : TSQuery query;
2057 : bool res;
2058 :
2059 0 : vector = DatumGetTSVector(DirectFunctionCall1(to_tsvector,
2060 : PG_GETARG_DATUM(0)));
2061 0 : query = DatumGetTSQuery(DirectFunctionCall1(plainto_tsquery,
2062 : PG_GETARG_DATUM(1)));
2063 :
2064 0 : res = DatumGetBool(DirectFunctionCall2(ts_match_vq,
2065 : TSVectorGetDatum(vector),
2066 : TSQueryGetDatum(query)));
2067 :
2068 0 : pfree(vector);
2069 0 : pfree(query);
2070 :
2071 0 : PG_RETURN_BOOL(res);
2072 : }
2073 :
2074 : Datum
2075 0 : ts_match_tq(PG_FUNCTION_ARGS)
2076 : {
2077 : TSVector vector;
2078 0 : TSQuery query = PG_GETARG_TSQUERY(1);
2079 : bool res;
2080 :
2081 0 : vector = DatumGetTSVector(DirectFunctionCall1(to_tsvector,
2082 : PG_GETARG_DATUM(0)));
2083 :
2084 0 : res = DatumGetBool(DirectFunctionCall2(ts_match_vq,
2085 : TSVectorGetDatum(vector),
2086 : TSQueryGetDatum(query)));
2087 :
2088 0 : pfree(vector);
2089 0 : PG_FREE_IF_COPY(query, 1);
2090 :
2091 0 : PG_RETURN_BOOL(res);
2092 : }
2093 :
2094 : /*
2095 : * ts_stat statistic function support
2096 : */
2097 :
2098 :
2099 : /*
2100 : * Returns the number of positions in value 'wptr' within tsvector 'txt',
2101 : * that have a weight equal to one of the weights in 'weight' bitmask.
2102 : */
2103 : static int
2104 5452 : check_weight(TSVector txt, WordEntry *wptr, int8 weight)
2105 : {
2106 5452 : int len = POSDATALEN(txt, wptr);
2107 5452 : int num = 0;
2108 5452 : WordEntryPos *ptr = POSDATAPTR(txt, wptr);
2109 :
2110 11100 : while (len--)
2111 : {
2112 5648 : if (weight & (1 << WEP_GETWEIGHT(*ptr)))
2113 8 : num++;
2114 5648 : ptr++;
2115 : }
2116 5452 : return num;
2117 : }
2118 :
2119 : #define compareStatWord(a,e,t) \
2120 : tsCompareString((a)->lexeme, (a)->lenlexeme, \
2121 : STRPTR(t) + (e)->pos, (e)->len, \
2122 : false)
2123 :
2124 : static void
2125 230416 : insertStatEntry(MemoryContext persistentContext, TSVectorStat *stat, TSVector txt, uint32 off)
2126 : {
2127 230416 : WordEntry *we = ARRPTR(txt) + off;
2128 230416 : StatEntry *node = stat->root,
2129 230416 : *pnode = NULL;
2130 : int n,
2131 230416 : res = 0;
2132 230416 : uint32 depth = 1;
2133 :
2134 230416 : if (stat->weight == 0)
2135 115208 : n = (we->haspos) ? POSDATALEN(txt, we) : 1;
2136 : else
2137 115208 : n = (we->haspos) ? check_weight(txt, we, stat->weight) : 0;
2138 :
2139 230416 : if (n == 0)
2140 115204 : return; /* nothing to insert */
2141 :
2142 1163640 : while (node)
2143 : {
2144 1159064 : res = compareStatWord(node, we, txt);
2145 :
2146 1159064 : if (res == 0)
2147 : {
2148 110636 : break;
2149 : }
2150 : else
2151 : {
2152 1048428 : pnode = node;
2153 1048428 : node = (res < 0) ? node->left : node->right;
2154 : }
2155 1048428 : depth++;
2156 : }
2157 :
2158 115212 : if (depth > stat->maxdepth)
2159 84 : stat->maxdepth = depth;
2160 :
2161 115212 : if (node == NULL)
2162 : {
2163 4576 : node = MemoryContextAlloc(persistentContext, STATENTRYHDRSZ + we->len);
2164 4576 : node->left = node->right = NULL;
2165 4576 : node->ndoc = 1;
2166 4576 : node->nentry = n;
2167 4576 : node->lenlexeme = we->len;
2168 4576 : memcpy(node->lexeme, STRPTR(txt) + we->pos, node->lenlexeme);
2169 :
2170 4576 : if (pnode == NULL)
2171 : {
2172 8 : stat->root = node;
2173 : }
2174 : else
2175 : {
2176 4568 : if (res < 0)
2177 2248 : pnode->left = node;
2178 : else
2179 2320 : pnode->right = node;
2180 : }
2181 :
2182 : }
2183 : else
2184 : {
2185 110636 : node->ndoc++;
2186 110636 : node->nentry += n;
2187 : }
2188 : }
2189 :
2190 : static void
2191 330256 : chooseNextStatEntry(MemoryContext persistentContext, TSVectorStat *stat, TSVector txt,
2192 : uint32 low, uint32 high, uint32 offset)
2193 : {
2194 : uint32 pos;
2195 330256 : uint32 middle = (low + high) >> 1;
2196 :
2197 330256 : pos = (low + middle) >> 1;
2198 330256 : if (low != middle && pos >= offset && pos - offset < txt->size)
2199 113552 : insertStatEntry(persistentContext, stat, txt, pos - offset);
2200 330256 : pos = (high + middle + 1) >> 1;
2201 330256 : if (middle + 1 != high && pos >= offset && pos - offset < txt->size)
2202 112856 : insertStatEntry(persistentContext, stat, txt, pos - offset);
2203 :
2204 330256 : if (low != middle)
2205 165128 : chooseNextStatEntry(persistentContext, stat, txt, low, middle, offset);
2206 330256 : if (high != middle + 1)
2207 161120 : chooseNextStatEntry(persistentContext, stat, txt, middle + 1, high, offset);
2208 330256 : }
2209 :
2210 : /*
2211 : * This is written like a custom aggregate function, because the
2212 : * original plan was to do just that. Unfortunately, an aggregate function
2213 : * can't return a set, so that plan was abandoned. If that limitation is
2214 : * lifted in the future, ts_stat could be a real aggregate function so that
2215 : * you could use it like this:
2216 : *
2217 : * SELECT ts_stat(vector_column) FROM vector_table;
2218 : *
2219 : * where vector_column is a tsvector-type column in vector_table.
2220 : */
2221 :
2222 : static TSVectorStat *
2223 4072 : ts_accum(MemoryContext persistentContext, TSVectorStat *stat, Datum data)
2224 : {
2225 4072 : TSVector txt = DatumGetTSVector(data);
2226 : uint32 i,
2227 4072 : nbit = 0,
2228 : offset;
2229 :
2230 4072 : if (stat == NULL)
2231 : { /* Init in first */
2232 0 : stat = MemoryContextAllocZero(persistentContext, sizeof(TSVectorStat));
2233 0 : stat->maxdepth = 1;
2234 : }
2235 :
2236 : /* simple check of correctness */
2237 4072 : if (txt == NULL || txt->size == 0)
2238 : {
2239 64 : if (txt && txt != (TSVector) DatumGetPointer(data))
2240 64 : pfree(txt);
2241 64 : return stat;
2242 : }
2243 :
2244 4008 : i = txt->size - 1;
2245 28480 : for (; i > 0; i >>= 1)
2246 24472 : nbit++;
2247 :
2248 4008 : nbit = 1 << nbit;
2249 4008 : offset = (nbit - txt->size) / 2;
2250 :
2251 4008 : insertStatEntry(persistentContext, stat, txt, (nbit >> 1) - offset);
2252 4008 : chooseNextStatEntry(persistentContext, stat, txt, 0, nbit, offset);
2253 :
2254 4008 : return stat;
2255 : }
2256 :
2257 : static void
2258 8 : ts_setup_firstcall(FunctionCallInfo fcinfo, FuncCallContext *funcctx,
2259 : TSVectorStat *stat)
2260 : {
2261 : TupleDesc tupdesc;
2262 : MemoryContext oldcontext;
2263 : StatEntry *node;
2264 :
2265 8 : funcctx->user_fctx = (void *) stat;
2266 :
2267 8 : oldcontext = MemoryContextSwitchTo(funcctx->multi_call_memory_ctx);
2268 :
2269 8 : stat->stack = palloc0(sizeof(StatEntry *) * (stat->maxdepth + 1));
2270 8 : stat->stackpos = 0;
2271 :
2272 8 : node = stat->root;
2273 : /* find leftmost value */
2274 8 : if (node == NULL)
2275 0 : stat->stack[stat->stackpos] = NULL;
2276 : else
2277 : for (;;)
2278 : {
2279 24 : stat->stack[stat->stackpos] = node;
2280 32 : if (node->left)
2281 : {
2282 24 : stat->stackpos++;
2283 24 : node = node->left;
2284 : }
2285 : else
2286 8 : break;
2287 : }
2288 : Assert(stat->stackpos <= stat->maxdepth);
2289 :
2290 8 : tupdesc = CreateTemplateTupleDesc(3);
2291 8 : TupleDescInitEntry(tupdesc, (AttrNumber) 1, "word",
2292 : TEXTOID, -1, 0);
2293 8 : TupleDescInitEntry(tupdesc, (AttrNumber) 2, "ndoc",
2294 : INT4OID, -1, 0);
2295 8 : TupleDescInitEntry(tupdesc, (AttrNumber) 3, "nentry",
2296 : INT4OID, -1, 0);
2297 8 : funcctx->tuple_desc = BlessTupleDesc(tupdesc);
2298 8 : funcctx->attinmeta = TupleDescGetAttInMetadata(tupdesc);
2299 :
2300 8 : MemoryContextSwitchTo(oldcontext);
2301 8 : }
2302 :
2303 : static StatEntry *
2304 9152 : walkStatEntryTree(TSVectorStat *stat)
2305 : {
2306 9152 : StatEntry *node = stat->stack[stat->stackpos];
2307 :
2308 9152 : if (node == NULL)
2309 0 : return NULL;
2310 :
2311 9152 : if (node->ndoc != 0)
2312 : {
2313 : /* return entry itself: we already was at left sublink */
2314 2256 : return node;
2315 : }
2316 6896 : else if (node->right && node->right != stat->stack[stat->stackpos + 1])
2317 : {
2318 : /* go on right sublink */
2319 2320 : stat->stackpos++;
2320 2320 : node = node->right;
2321 :
2322 : /* find most-left value */
2323 : for (;;)
2324 : {
2325 2224 : stat->stack[stat->stackpos] = node;
2326 4544 : if (node->left)
2327 : {
2328 2224 : stat->stackpos++;
2329 2224 : node = node->left;
2330 : }
2331 : else
2332 2320 : break;
2333 : }
2334 2320 : Assert(stat->stackpos <= stat->maxdepth);
2335 : }
2336 : else
2337 : {
2338 : /* we already return all left subtree, itself and right subtree */
2339 4576 : if (stat->stackpos == 0)
2340 8 : return NULL;
2341 :
2342 4568 : stat->stackpos--;
2343 4568 : return walkStatEntryTree(stat);
2344 : }
2345 :
2346 2320 : return node;
2347 : }
2348 :
2349 : static Datum
2350 4584 : ts_process_call(FuncCallContext *funcctx)
2351 : {
2352 : TSVectorStat *st;
2353 : StatEntry *entry;
2354 :
2355 4584 : st = (TSVectorStat *) funcctx->user_fctx;
2356 :
2357 4584 : entry = walkStatEntryTree(st);
2358 :
2359 4584 : if (entry != NULL)
2360 : {
2361 : Datum result;
2362 : char *values[3];
2363 : char ndoc[16];
2364 : char nentry[16];
2365 : HeapTuple tuple;
2366 :
2367 4576 : values[0] = palloc(entry->lenlexeme + 1);
2368 4576 : memcpy(values[0], entry->lexeme, entry->lenlexeme);
2369 4576 : (values[0])[entry->lenlexeme] = '\0';
2370 4576 : sprintf(ndoc, "%d", entry->ndoc);
2371 4576 : values[1] = ndoc;
2372 4576 : sprintf(nentry, "%d", entry->nentry);
2373 4576 : values[2] = nentry;
2374 :
2375 4576 : tuple = BuildTupleFromCStrings(funcctx->attinmeta, values);
2376 4576 : result = HeapTupleGetDatum(tuple);
2377 :
2378 4576 : pfree(values[0]);
2379 :
2380 : /* mark entry as already visited */
2381 4576 : entry->ndoc = 0;
2382 :
2383 4576 : return result;
2384 : }
2385 :
2386 8 : return (Datum) 0;
2387 : }
2388 :
2389 : static TSVectorStat *
2390 8 : ts_stat_sql(MemoryContext persistentContext, text *txt, text *ws)
2391 : {
2392 8 : char *query = text_to_cstring(txt);
2393 : TSVectorStat *stat;
2394 : bool isnull;
2395 : Portal portal;
2396 : SPIPlanPtr plan;
2397 :
2398 8 : if ((plan = SPI_prepare(query, 0, NULL)) == NULL)
2399 : /* internal error */
2400 0 : elog(ERROR, "SPI_prepare(\"%s\") failed", query);
2401 :
2402 8 : if ((portal = SPI_cursor_open(NULL, plan, NULL, NULL, true)) == NULL)
2403 : /* internal error */
2404 0 : elog(ERROR, "SPI_cursor_open(\"%s\") failed", query);
2405 :
2406 8 : SPI_cursor_fetch(portal, true, 100);
2407 :
2408 8 : if (SPI_tuptable == NULL ||
2409 8 : SPI_tuptable->tupdesc->natts != 1 ||
2410 8 : !IsBinaryCoercible(SPI_gettypeid(SPI_tuptable->tupdesc, 1),
2411 : TSVECTOROID))
2412 0 : ereport(ERROR,
2413 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
2414 : errmsg("ts_stat query must return one tsvector column")));
2415 :
2416 8 : stat = MemoryContextAllocZero(persistentContext, sizeof(TSVectorStat));
2417 8 : stat->maxdepth = 1;
2418 :
2419 8 : if (ws)
2420 : {
2421 : char *buf;
2422 :
2423 4 : buf = VARDATA_ANY(ws);
2424 12 : while (buf - VARDATA_ANY(ws) < VARSIZE_ANY_EXHDR(ws))
2425 : {
2426 8 : if (pg_mblen(buf) == 1)
2427 : {
2428 8 : switch (*buf)
2429 : {
2430 4 : case 'A':
2431 : case 'a':
2432 4 : stat->weight |= 1 << 3;
2433 4 : break;
2434 4 : case 'B':
2435 : case 'b':
2436 4 : stat->weight |= 1 << 2;
2437 4 : break;
2438 0 : case 'C':
2439 : case 'c':
2440 0 : stat->weight |= 1 << 1;
2441 0 : break;
2442 0 : case 'D':
2443 : case 'd':
2444 0 : stat->weight |= 1;
2445 0 : break;
2446 0 : default:
2447 0 : stat->weight |= 0;
2448 : }
2449 0 : }
2450 8 : buf += pg_mblen(buf);
2451 : }
2452 : }
2453 :
2454 56 : while (SPI_processed > 0)
2455 : {
2456 : uint64 i;
2457 :
2458 4120 : for (i = 0; i < SPI_processed; i++)
2459 : {
2460 4072 : Datum data = SPI_getbinval(SPI_tuptable->vals[i], SPI_tuptable->tupdesc, 1, &isnull);
2461 :
2462 4072 : if (!isnull)
2463 4072 : stat = ts_accum(persistentContext, stat, data);
2464 : }
2465 :
2466 48 : SPI_freetuptable(SPI_tuptable);
2467 48 : SPI_cursor_fetch(portal, true, 100);
2468 : }
2469 :
2470 8 : SPI_freetuptable(SPI_tuptable);
2471 8 : SPI_cursor_close(portal);
2472 8 : SPI_freeplan(plan);
2473 8 : pfree(query);
2474 :
2475 8 : return stat;
2476 : }
2477 :
2478 : Datum
2479 4576 : ts_stat1(PG_FUNCTION_ARGS)
2480 : {
2481 : FuncCallContext *funcctx;
2482 : Datum result;
2483 :
2484 4576 : if (SRF_IS_FIRSTCALL())
2485 : {
2486 : TSVectorStat *stat;
2487 4 : text *txt = PG_GETARG_TEXT_PP(0);
2488 :
2489 4 : funcctx = SRF_FIRSTCALL_INIT();
2490 4 : SPI_connect();
2491 4 : stat = ts_stat_sql(funcctx->multi_call_memory_ctx, txt, NULL);
2492 4 : PG_FREE_IF_COPY(txt, 0);
2493 4 : ts_setup_firstcall(fcinfo, funcctx, stat);
2494 4 : SPI_finish();
2495 : }
2496 :
2497 4576 : funcctx = SRF_PERCALL_SETUP();
2498 4576 : if ((result = ts_process_call(funcctx)) != (Datum) 0)
2499 4572 : SRF_RETURN_NEXT(funcctx, result);
2500 4 : SRF_RETURN_DONE(funcctx);
2501 : }
2502 :
2503 : Datum
2504 8 : ts_stat2(PG_FUNCTION_ARGS)
2505 : {
2506 : FuncCallContext *funcctx;
2507 : Datum result;
2508 :
2509 8 : if (SRF_IS_FIRSTCALL())
2510 : {
2511 : TSVectorStat *stat;
2512 4 : text *txt = PG_GETARG_TEXT_PP(0);
2513 4 : text *ws = PG_GETARG_TEXT_PP(1);
2514 :
2515 4 : funcctx = SRF_FIRSTCALL_INIT();
2516 4 : SPI_connect();
2517 4 : stat = ts_stat_sql(funcctx->multi_call_memory_ctx, txt, ws);
2518 4 : PG_FREE_IF_COPY(txt, 0);
2519 4 : PG_FREE_IF_COPY(ws, 1);
2520 4 : ts_setup_firstcall(fcinfo, funcctx, stat);
2521 4 : SPI_finish();
2522 : }
2523 :
2524 8 : funcctx = SRF_PERCALL_SETUP();
2525 8 : if ((result = ts_process_call(funcctx)) != (Datum) 0)
2526 4 : SRF_RETURN_NEXT(funcctx, result);
2527 4 : SRF_RETURN_DONE(funcctx);
2528 : }
2529 :
2530 :
2531 : /*
2532 : * Triggers for automatic update of a tsvector column from text column(s)
2533 : *
2534 : * Trigger arguments are either
2535 : * name of tsvector col, name of tsconfig to use, name(s) of text col(s)
2536 : * name of tsvector col, name of regconfig col, name(s) of text col(s)
2537 : * ie, tsconfig can either be specified by name, or indirectly as the
2538 : * contents of a regconfig field in the row. If the name is used, it must
2539 : * be explicitly schema-qualified.
2540 : */
2541 : Datum
2542 12 : tsvector_update_trigger_byid(PG_FUNCTION_ARGS)
2543 : {
2544 12 : return tsvector_update_trigger(fcinfo, false);
2545 : }
2546 :
2547 : Datum
2548 0 : tsvector_update_trigger_bycolumn(PG_FUNCTION_ARGS)
2549 : {
2550 0 : return tsvector_update_trigger(fcinfo, true);
2551 : }
2552 :
2553 : static Datum
2554 12 : tsvector_update_trigger(PG_FUNCTION_ARGS, bool config_column)
2555 : {
2556 : TriggerData *trigdata;
2557 : Trigger *trigger;
2558 : Relation rel;
2559 12 : HeapTuple rettuple = NULL;
2560 : int tsvector_attr_num,
2561 : i;
2562 : ParsedText prs;
2563 : Datum datum;
2564 : bool isnull;
2565 : text *txt;
2566 : Oid cfgId;
2567 : bool update_needed;
2568 :
2569 : /* Check call context */
2570 12 : if (!CALLED_AS_TRIGGER(fcinfo)) /* internal error */
2571 0 : elog(ERROR, "tsvector_update_trigger: not fired by trigger manager");
2572 :
2573 12 : trigdata = (TriggerData *) fcinfo->context;
2574 12 : if (!TRIGGER_FIRED_FOR_ROW(trigdata->tg_event))
2575 0 : elog(ERROR, "tsvector_update_trigger: must be fired for row");
2576 12 : if (!TRIGGER_FIRED_BEFORE(trigdata->tg_event))
2577 0 : elog(ERROR, "tsvector_update_trigger: must be fired BEFORE event");
2578 :
2579 12 : if (TRIGGER_FIRED_BY_INSERT(trigdata->tg_event))
2580 : {
2581 8 : rettuple = trigdata->tg_trigtuple;
2582 8 : update_needed = true;
2583 : }
2584 4 : else if (TRIGGER_FIRED_BY_UPDATE(trigdata->tg_event))
2585 : {
2586 4 : rettuple = trigdata->tg_newtuple;
2587 4 : update_needed = false; /* computed below */
2588 : }
2589 : else
2590 0 : elog(ERROR, "tsvector_update_trigger: must be fired for INSERT or UPDATE");
2591 :
2592 12 : trigger = trigdata->tg_trigger;
2593 12 : rel = trigdata->tg_relation;
2594 :
2595 12 : if (trigger->tgnargs < 3)
2596 0 : elog(ERROR, "tsvector_update_trigger: arguments must be tsvector_field, ts_config, text_field1, ...)");
2597 :
2598 : /* Find the target tsvector column */
2599 12 : tsvector_attr_num = SPI_fnumber(rel->rd_att, trigger->tgargs[0]);
2600 12 : if (tsvector_attr_num == SPI_ERROR_NOATTRIBUTE)
2601 0 : ereport(ERROR,
2602 : (errcode(ERRCODE_UNDEFINED_COLUMN),
2603 : errmsg("tsvector column \"%s\" does not exist",
2604 : trigger->tgargs[0])));
2605 : /* This will effectively reject system columns, so no separate test: */
2606 12 : if (!IsBinaryCoercible(SPI_gettypeid(rel->rd_att, tsvector_attr_num),
2607 : TSVECTOROID))
2608 0 : ereport(ERROR,
2609 : (errcode(ERRCODE_DATATYPE_MISMATCH),
2610 : errmsg("column \"%s\" is not of tsvector type",
2611 : trigger->tgargs[0])));
2612 :
2613 : /* Find the configuration to use */
2614 12 : if (config_column)
2615 : {
2616 : int config_attr_num;
2617 :
2618 0 : config_attr_num = SPI_fnumber(rel->rd_att, trigger->tgargs[1]);
2619 0 : if (config_attr_num == SPI_ERROR_NOATTRIBUTE)
2620 0 : ereport(ERROR,
2621 : (errcode(ERRCODE_UNDEFINED_COLUMN),
2622 : errmsg("configuration column \"%s\" does not exist",
2623 : trigger->tgargs[1])));
2624 0 : if (!IsBinaryCoercible(SPI_gettypeid(rel->rd_att, config_attr_num),
2625 : REGCONFIGOID))
2626 0 : ereport(ERROR,
2627 : (errcode(ERRCODE_DATATYPE_MISMATCH),
2628 : errmsg("column \"%s\" is not of regconfig type",
2629 : trigger->tgargs[1])));
2630 :
2631 0 : datum = SPI_getbinval(rettuple, rel->rd_att, config_attr_num, &isnull);
2632 0 : if (isnull)
2633 0 : ereport(ERROR,
2634 : (errcode(ERRCODE_NULL_VALUE_NOT_ALLOWED),
2635 : errmsg("configuration column \"%s\" must not be null",
2636 : trigger->tgargs[1])));
2637 0 : cfgId = DatumGetObjectId(datum);
2638 : }
2639 : else
2640 : {
2641 : List *names;
2642 :
2643 12 : names = stringToQualifiedNameList(trigger->tgargs[1]);
2644 : /* require a schema so that results are not search path dependent */
2645 12 : if (list_length(names) < 2)
2646 0 : ereport(ERROR,
2647 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
2648 : errmsg("text search configuration name \"%s\" must be schema-qualified",
2649 : trigger->tgargs[1])));
2650 12 : cfgId = get_ts_config_oid(names, false);
2651 : }
2652 :
2653 : /* initialize parse state */
2654 12 : prs.lenwords = 32;
2655 12 : prs.curwords = 0;
2656 12 : prs.pos = 0;
2657 12 : prs.words = (ParsedWord *) palloc(sizeof(ParsedWord) * prs.lenwords);
2658 :
2659 : /* find all words in indexable column(s) */
2660 24 : for (i = 2; i < trigger->tgnargs; i++)
2661 : {
2662 : int numattr;
2663 :
2664 12 : numattr = SPI_fnumber(rel->rd_att, trigger->tgargs[i]);
2665 12 : if (numattr == SPI_ERROR_NOATTRIBUTE)
2666 0 : ereport(ERROR,
2667 : (errcode(ERRCODE_UNDEFINED_COLUMN),
2668 : errmsg("column \"%s\" does not exist",
2669 : trigger->tgargs[i])));
2670 12 : if (!IsBinaryCoercible(SPI_gettypeid(rel->rd_att, numattr), TEXTOID))
2671 0 : ereport(ERROR,
2672 : (errcode(ERRCODE_DATATYPE_MISMATCH),
2673 : errmsg("column \"%s\" is not of a character type",
2674 : trigger->tgargs[i])));
2675 :
2676 12 : if (bms_is_member(numattr - FirstLowInvalidHeapAttributeNumber, trigdata->tg_updatedcols))
2677 4 : update_needed = true;
2678 :
2679 12 : datum = SPI_getbinval(rettuple, rel->rd_att, numattr, &isnull);
2680 12 : if (isnull)
2681 4 : continue;
2682 :
2683 8 : txt = DatumGetTextPP(datum);
2684 :
2685 8 : parsetext(cfgId, &prs, VARDATA_ANY(txt), VARSIZE_ANY_EXHDR(txt));
2686 :
2687 8 : if (txt != (text *) DatumGetPointer(datum))
2688 0 : pfree(txt);
2689 : }
2690 :
2691 12 : if (update_needed)
2692 : {
2693 : /* make tsvector value */
2694 12 : datum = TSVectorGetDatum(make_tsvector(&prs));
2695 12 : isnull = false;
2696 :
2697 : /* and insert it into tuple */
2698 12 : rettuple = heap_modify_tuple_by_cols(rettuple, rel->rd_att,
2699 : 1, &tsvector_attr_num,
2700 : &datum, &isnull);
2701 :
2702 12 : pfree(DatumGetPointer(datum));
2703 : }
2704 :
2705 12 : return PointerGetDatum(rettuple);
2706 : }
|
