Line data Source code
1 : /*-------------------------------------------------------------------------
2 : *
3 : * regexp.c
4 : * Postgres' interface to the regular expression package.
5 : *
6 : * Portions Copyright (c) 1996-2025, PostgreSQL Global Development Group
7 : * Portions Copyright (c) 1994, Regents of the University of California
8 : *
9 : *
10 : * IDENTIFICATION
11 : * src/backend/utils/adt/regexp.c
12 : *
13 : * Alistair Crooks added the code for the regex caching
14 : * agc - cached the regular expressions used - there's a good chance
15 : * that we'll get a hit, so this saves a compile step for every
16 : * attempted match. I haven't actually measured the speed improvement,
17 : * but it `looks' a lot quicker visually when watching regression
18 : * test output.
19 : *
20 : * agc - incorporated Keith Bostic's Berkeley regex code into
21 : * the tree for all ports. To distinguish this regex code from any that
22 : * is existent on a platform, I've prepended the string "pg_" to
23 : * the functions regcomp, regerror, regexec and regfree.
24 : * Fixed a bug that was originally a typo by me, where `i' was used
25 : * instead of `oldest' when compiling regular expressions - benign
26 : * results mostly, although occasionally it bit you...
27 : *
28 : *-------------------------------------------------------------------------
29 : */
30 : #include "postgres.h"
31 :
32 : #include "catalog/pg_type.h"
33 : #include "funcapi.h"
34 : #include "regex/regex.h"
35 : #include "utils/array.h"
36 : #include "utils/builtins.h"
37 : #include "utils/memutils.h"
38 : #include "utils/varlena.h"
39 :
40 : #define PG_GETARG_TEXT_PP_IF_EXISTS(_n) \
41 : (PG_NARGS() > (_n) ? PG_GETARG_TEXT_PP(_n) : NULL)
42 :
43 :
44 : /* all the options of interest for regex functions */
45 : typedef struct pg_re_flags
46 : {
47 : int cflags; /* compile flags for Spencer's regex code */
48 : bool glob; /* do it globally (for each occurrence) */
49 : } pg_re_flags;
50 :
51 : /* cross-call state for regexp_match and regexp_split functions */
52 : typedef struct regexp_matches_ctx
53 : {
54 : text *orig_str; /* data string in original TEXT form */
55 : int nmatches; /* number of places where pattern matched */
56 : int npatterns; /* number of capturing subpatterns */
57 : /* We store start char index and end+1 char index for each match */
58 : /* so the number of entries in match_locs is nmatches * npatterns * 2 */
59 : int *match_locs; /* 0-based character indexes */
60 : int next_match; /* 0-based index of next match to process */
61 : /* workspace for build_regexp_match_result() */
62 : Datum *elems; /* has npatterns elements */
63 : bool *nulls; /* has npatterns elements */
64 : pg_wchar *wide_str; /* wide-char version of original string */
65 : char *conv_buf; /* conversion buffer, if needed */
66 : int conv_bufsiz; /* size thereof */
67 : } regexp_matches_ctx;
68 :
69 : /*
70 : * We cache precompiled regular expressions using a "self organizing list"
71 : * structure, in which recently-used items tend to be near the front.
72 : * Whenever we use an entry, it's moved up to the front of the list.
73 : * Over time, an item's average position corresponds to its frequency of use.
74 : *
75 : * When we first create an entry, it's inserted at the front of
76 : * the array, dropping the entry at the end of the array if necessary to
77 : * make room. (This might seem to be weighting the new entry too heavily,
78 : * but if we insert new entries further back, we'll be unable to adjust to
79 : * a sudden shift in the query mix where we are presented with MAX_CACHED_RES
80 : * never-before-seen items used circularly. We ought to be able to handle
81 : * that case, so we have to insert at the front.)
82 : *
83 : * Knuth mentions a variant strategy in which a used item is moved up just
84 : * one place in the list. Although he says this uses fewer comparisons on
85 : * average, it seems not to adapt very well to the situation where you have
86 : * both some reusable patterns and a steady stream of non-reusable patterns.
87 : * A reusable pattern that isn't used at least as often as non-reusable
88 : * patterns are seen will "fail to keep up" and will drop off the end of the
89 : * cache. With move-to-front, a reusable pattern is guaranteed to stay in
90 : * the cache as long as it's used at least once in every MAX_CACHED_RES uses.
91 : */
92 :
93 : /* this is the maximum number of cached regular expressions */
94 : #ifndef MAX_CACHED_RES
95 : #define MAX_CACHED_RES 32
96 : #endif
97 :
98 : /* A parent memory context for regular expressions. */
99 : static MemoryContext RegexpCacheMemoryContext;
100 :
101 : /* this structure describes one cached regular expression */
102 : typedef struct cached_re_str
103 : {
104 : MemoryContext cre_context; /* memory context for this regexp */
105 : char *cre_pat; /* original RE (not null terminated!) */
106 : int cre_pat_len; /* length of original RE, in bytes */
107 : int cre_flags; /* compile flags: extended,icase etc */
108 : Oid cre_collation; /* collation to use */
109 : regex_t cre_re; /* the compiled regular expression */
110 : } cached_re_str;
111 :
112 : static int num_res = 0; /* # of cached re's */
113 : static cached_re_str re_array[MAX_CACHED_RES]; /* cached re's */
114 :
115 :
116 : /* Local functions */
117 : static regexp_matches_ctx *setup_regexp_matches(text *orig_str, text *pattern,
118 : pg_re_flags *re_flags,
119 : int start_search,
120 : Oid collation,
121 : bool use_subpatterns,
122 : bool ignore_degenerate,
123 : bool fetching_unmatched);
124 : static ArrayType *build_regexp_match_result(regexp_matches_ctx *matchctx);
125 : static Datum build_regexp_split_result(regexp_matches_ctx *splitctx);
126 :
127 :
128 : /*
129 : * RE_compile_and_cache - compile a RE, caching if possible
130 : *
131 : * Returns regex_t *
132 : *
133 : * text_re --- the pattern, expressed as a TEXT object
134 : * cflags --- compile options for the pattern
135 : * collation --- collation to use for LC_CTYPE-dependent behavior
136 : *
137 : * Pattern is given in the database encoding. We internally convert to
138 : * an array of pg_wchar, which is what Spencer's regex package wants.
139 : */
140 : regex_t *
141 7132018 : RE_compile_and_cache(text *text_re, int cflags, Oid collation)
142 : {
143 7132018 : int text_re_len = VARSIZE_ANY_EXHDR(text_re);
144 7132018 : char *text_re_val = VARDATA_ANY(text_re);
145 : pg_wchar *pattern;
146 : int pattern_len;
147 : int i;
148 : int regcomp_result;
149 : cached_re_str re_temp;
150 : char errMsg[100];
151 : MemoryContext oldcontext;
152 :
153 : /*
154 : * Look for a match among previously compiled REs. Since the data
155 : * structure is self-organizing with most-used entries at the front, our
156 : * search strategy can just be to scan from the front.
157 : */
158 7726336 : for (i = 0; i < num_res; i++)
159 : {
160 7720012 : if (re_array[i].cre_pat_len == text_re_len &&
161 7140650 : re_array[i].cre_flags == cflags &&
162 7139452 : re_array[i].cre_collation == collation &&
163 7139076 : memcmp(re_array[i].cre_pat, text_re_val, text_re_len) == 0)
164 : {
165 : /*
166 : * Found a match; move it to front if not there already.
167 : */
168 7125694 : if (i > 0)
169 : {
170 455870 : re_temp = re_array[i];
171 455870 : memmove(&re_array[1], &re_array[0], i * sizeof(cached_re_str));
172 455870 : re_array[0] = re_temp;
173 : }
174 :
175 7125694 : return &re_array[0].cre_re;
176 : }
177 : }
178 :
179 : /* Set up the cache memory on first go through. */
180 6324 : if (unlikely(RegexpCacheMemoryContext == NULL))
181 1612 : RegexpCacheMemoryContext =
182 1612 : AllocSetContextCreate(TopMemoryContext,
183 : "RegexpCacheMemoryContext",
184 : ALLOCSET_SMALL_SIZES);
185 :
186 : /*
187 : * Couldn't find it, so try to compile the new RE. To avoid leaking
188 : * resources on failure, we build into the re_temp local.
189 : */
190 :
191 : /* Convert pattern string to wide characters */
192 6324 : pattern = (pg_wchar *) palloc((text_re_len + 1) * sizeof(pg_wchar));
193 6324 : pattern_len = pg_mb2wchar_with_len(text_re_val,
194 : pattern,
195 : text_re_len);
196 :
197 : /*
198 : * Make a memory context for this compiled regexp. This is initially a
199 : * child of the current memory context, so it will be cleaned up
200 : * automatically if compilation is interrupted and throws an ERROR. We'll
201 : * re-parent it under the longer lived cache context if we make it to the
202 : * bottom of this function.
203 : */
204 6324 : re_temp.cre_context = AllocSetContextCreate(CurrentMemoryContext,
205 : "RegexpMemoryContext",
206 : ALLOCSET_SMALL_SIZES);
207 6324 : oldcontext = MemoryContextSwitchTo(re_temp.cre_context);
208 :
209 6324 : regcomp_result = pg_regcomp(&re_temp.cre_re,
210 : pattern,
211 : pattern_len,
212 : cflags,
213 : collation);
214 :
215 6300 : pfree(pattern);
216 :
217 6300 : if (regcomp_result != REG_OKAY)
218 : {
219 : /* re didn't compile (no need for pg_regfree, if so) */
220 36 : pg_regerror(regcomp_result, &re_temp.cre_re, errMsg, sizeof(errMsg));
221 36 : ereport(ERROR,
222 : (errcode(ERRCODE_INVALID_REGULAR_EXPRESSION),
223 : errmsg("invalid regular expression: %s", errMsg)));
224 : }
225 :
226 : /* Copy the pattern into the per-regexp memory context. */
227 6264 : re_temp.cre_pat = palloc(text_re_len + 1);
228 6264 : memcpy(re_temp.cre_pat, text_re_val, text_re_len);
229 :
230 : /*
231 : * NUL-terminate it only for the benefit of the identifier used for the
232 : * memory context, visible in the pg_backend_memory_contexts view.
233 : */
234 6264 : re_temp.cre_pat[text_re_len] = 0;
235 6264 : MemoryContextSetIdentifier(re_temp.cre_context, re_temp.cre_pat);
236 :
237 6264 : re_temp.cre_pat_len = text_re_len;
238 6264 : re_temp.cre_flags = cflags;
239 6264 : re_temp.cre_collation = collation;
240 :
241 : /*
242 : * Okay, we have a valid new item in re_temp; insert it into the storage
243 : * array. Discard last entry if needed.
244 : */
245 6264 : if (num_res >= MAX_CACHED_RES)
246 : {
247 870 : --num_res;
248 : Assert(num_res < MAX_CACHED_RES);
249 : /* Delete the memory context holding the regexp and pattern. */
250 870 : MemoryContextDelete(re_array[num_res].cre_context);
251 : }
252 :
253 : /* Re-parent the memory context to our long-lived cache context. */
254 6264 : MemoryContextSetParent(re_temp.cre_context, RegexpCacheMemoryContext);
255 :
256 6264 : if (num_res > 0)
257 4652 : memmove(&re_array[1], &re_array[0], num_res * sizeof(cached_re_str));
258 :
259 6264 : re_array[0] = re_temp;
260 6264 : num_res++;
261 :
262 6264 : MemoryContextSwitchTo(oldcontext);
263 :
264 6264 : return &re_array[0].cre_re;
265 : }
266 :
267 : /*
268 : * RE_wchar_execute - execute a RE on pg_wchar data
269 : *
270 : * Returns true on match, false on no match
271 : *
272 : * re --- the compiled pattern as returned by RE_compile_and_cache
273 : * data --- the data to match against (need not be null-terminated)
274 : * data_len --- the length of the data string
275 : * start_search -- the offset in the data to start searching
276 : * nmatch, pmatch --- optional return area for match details
277 : *
278 : * Data is given as array of pg_wchar which is what Spencer's regex package
279 : * wants.
280 : */
281 : static bool
282 7988540 : RE_wchar_execute(regex_t *re, pg_wchar *data, int data_len,
283 : int start_search, int nmatch, regmatch_t *pmatch)
284 : {
285 : int regexec_result;
286 : char errMsg[100];
287 :
288 : /* Perform RE match and return result */
289 7988540 : regexec_result = pg_regexec(re,
290 : data,
291 : data_len,
292 : start_search,
293 : NULL, /* no details */
294 : nmatch,
295 : pmatch,
296 : 0);
297 :
298 7988540 : if (regexec_result != REG_OKAY && regexec_result != REG_NOMATCH)
299 : {
300 : /* re failed??? */
301 0 : pg_regerror(regexec_result, re, errMsg, sizeof(errMsg));
302 0 : ereport(ERROR,
303 : (errcode(ERRCODE_INVALID_REGULAR_EXPRESSION),
304 : errmsg("regular expression failed: %s", errMsg)));
305 : }
306 :
307 7988540 : return (regexec_result == REG_OKAY);
308 : }
309 :
310 : /*
311 : * RE_execute - execute a RE
312 : *
313 : * Returns true on match, false on no match
314 : *
315 : * re --- the compiled pattern as returned by RE_compile_and_cache
316 : * dat --- the data to match against (need not be null-terminated)
317 : * dat_len --- the length of the data string
318 : * nmatch, pmatch --- optional return area for match details
319 : *
320 : * Data is given in the database encoding. We internally
321 : * convert to array of pg_wchar which is what Spencer's regex package wants.
322 : */
323 : static bool
324 6891570 : RE_execute(regex_t *re, char *dat, int dat_len,
325 : int nmatch, regmatch_t *pmatch)
326 : {
327 : pg_wchar *data;
328 : int data_len;
329 : bool match;
330 :
331 : /* Convert data string to wide characters */
332 6891570 : data = (pg_wchar *) palloc((dat_len + 1) * sizeof(pg_wchar));
333 6891570 : data_len = pg_mb2wchar_with_len(dat, data, dat_len);
334 :
335 : /* Perform RE match and return result */
336 6891570 : match = RE_wchar_execute(re, data, data_len, 0, nmatch, pmatch);
337 :
338 6891570 : pfree(data);
339 6891570 : return match;
340 : }
341 :
342 : /*
343 : * RE_compile_and_execute - compile and execute a RE
344 : *
345 : * Returns true on match, false on no match
346 : *
347 : * text_re --- the pattern, expressed as a TEXT object
348 : * dat --- the data to match against (need not be null-terminated)
349 : * dat_len --- the length of the data string
350 : * cflags --- compile options for the pattern
351 : * collation --- collation to use for LC_CTYPE-dependent behavior
352 : * nmatch, pmatch --- optional return area for match details
353 : *
354 : * Both pattern and data are given in the database encoding. We internally
355 : * convert to array of pg_wchar which is what Spencer's regex package wants.
356 : */
357 : bool
358 6889956 : RE_compile_and_execute(text *text_re, char *dat, int dat_len,
359 : int cflags, Oid collation,
360 : int nmatch, regmatch_t *pmatch)
361 : {
362 : regex_t *re;
363 :
364 : /* Use REG_NOSUB if caller does not want sub-match details */
365 6889956 : if (nmatch < 2)
366 6889956 : cflags |= REG_NOSUB;
367 :
368 : /* Compile RE */
369 6889956 : re = RE_compile_and_cache(text_re, cflags, collation);
370 :
371 6889932 : return RE_execute(re, dat, dat_len, nmatch, pmatch);
372 : }
373 :
374 :
375 : /*
376 : * parse_re_flags - parse the options argument of regexp_match and friends
377 : *
378 : * flags --- output argument, filled with desired options
379 : * opts --- TEXT object, or NULL for defaults
380 : *
381 : * This accepts all the options allowed by any of the callers; callers that
382 : * don't want some have to reject them after the fact.
383 : */
384 : static void
385 209606 : parse_re_flags(pg_re_flags *flags, text *opts)
386 : {
387 : /* regex flavor is always folded into the compile flags */
388 209606 : flags->cflags = REG_ADVANCED;
389 209606 : flags->glob = false;
390 :
391 209606 : if (opts)
392 : {
393 4702 : char *opt_p = VARDATA_ANY(opts);
394 4702 : int opt_len = VARSIZE_ANY_EXHDR(opts);
395 : int i;
396 :
397 10586 : for (i = 0; i < opt_len; i++)
398 : {
399 5908 : switch (opt_p[i])
400 : {
401 4380 : case 'g':
402 4380 : flags->glob = true;
403 4380 : break;
404 0 : case 'b': /* BREs (but why???) */
405 0 : flags->cflags &= ~(REG_ADVANCED | REG_EXTENDED | REG_QUOTE);
406 0 : break;
407 10 : case 'c': /* case sensitive */
408 10 : flags->cflags &= ~REG_ICASE;
409 10 : break;
410 0 : case 'e': /* plain EREs */
411 0 : flags->cflags |= REG_EXTENDED;
412 0 : flags->cflags &= ~(REG_ADVANCED | REG_QUOTE);
413 0 : break;
414 292 : case 'i': /* case insensitive */
415 292 : flags->cflags |= REG_ICASE;
416 292 : break;
417 1184 : case 'm': /* Perloid synonym for n */
418 : case 'n': /* \n affects ^ $ . [^ */
419 1184 : flags->cflags |= REG_NEWLINE;
420 1184 : break;
421 0 : case 'p': /* ~Perl, \n affects . [^ */
422 0 : flags->cflags |= REG_NLSTOP;
423 0 : flags->cflags &= ~REG_NLANCH;
424 0 : break;
425 0 : case 'q': /* literal string */
426 0 : flags->cflags |= REG_QUOTE;
427 0 : flags->cflags &= ~(REG_ADVANCED | REG_EXTENDED);
428 0 : break;
429 12 : case 's': /* single line, \n ordinary */
430 12 : flags->cflags &= ~REG_NEWLINE;
431 12 : break;
432 0 : case 't': /* tight syntax */
433 0 : flags->cflags &= ~REG_EXPANDED;
434 0 : break;
435 0 : case 'w': /* weird, \n affects ^ $ only */
436 0 : flags->cflags &= ~REG_NLSTOP;
437 0 : flags->cflags |= REG_NLANCH;
438 0 : break;
439 6 : case 'x': /* expanded syntax */
440 6 : flags->cflags |= REG_EXPANDED;
441 6 : break;
442 24 : default:
443 24 : ereport(ERROR,
444 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
445 : errmsg("invalid regular expression option: \"%.*s\"",
446 : pg_mblen(opt_p + i), opt_p + i)));
447 : break;
448 : }
449 : }
450 : }
451 209582 : }
452 :
453 :
454 : /*
455 : * interface routines called by the function manager
456 : */
457 :
458 : Datum
459 6430206 : nameregexeq(PG_FUNCTION_ARGS)
460 : {
461 6430206 : Name n = PG_GETARG_NAME(0);
462 6430206 : text *p = PG_GETARG_TEXT_PP(1);
463 :
464 6430206 : PG_RETURN_BOOL(RE_compile_and_execute(p,
465 : NameStr(*n),
466 : strlen(NameStr(*n)),
467 : REG_ADVANCED,
468 : PG_GET_COLLATION(),
469 : 0, NULL));
470 : }
471 :
472 : Datum
473 27354 : nameregexne(PG_FUNCTION_ARGS)
474 : {
475 27354 : Name n = PG_GETARG_NAME(0);
476 27354 : text *p = PG_GETARG_TEXT_PP(1);
477 :
478 27354 : PG_RETURN_BOOL(!RE_compile_and_execute(p,
479 : NameStr(*n),
480 : strlen(NameStr(*n)),
481 : REG_ADVANCED,
482 : PG_GET_COLLATION(),
483 : 0, NULL));
484 : }
485 :
486 : Datum
487 390222 : textregexeq(PG_FUNCTION_ARGS)
488 : {
489 390222 : text *s = PG_GETARG_TEXT_PP(0);
490 390222 : text *p = PG_GETARG_TEXT_PP(1);
491 :
492 390222 : PG_RETURN_BOOL(RE_compile_and_execute(p,
493 : VARDATA_ANY(s),
494 : VARSIZE_ANY_EXHDR(s),
495 : REG_ADVANCED,
496 : PG_GET_COLLATION(),
497 : 0, NULL));
498 : }
499 :
500 : Datum
501 34152 : textregexne(PG_FUNCTION_ARGS)
502 : {
503 34152 : text *s = PG_GETARG_TEXT_PP(0);
504 34152 : text *p = PG_GETARG_TEXT_PP(1);
505 :
506 34152 : PG_RETURN_BOOL(!RE_compile_and_execute(p,
507 : VARDATA_ANY(s),
508 : VARSIZE_ANY_EXHDR(s),
509 : REG_ADVANCED,
510 : PG_GET_COLLATION(),
511 : 0, NULL));
512 : }
513 :
514 :
515 : /*
516 : * routines that use the regexp stuff, but ignore the case.
517 : * for this, we use the REG_ICASE flag to pg_regcomp
518 : */
519 :
520 :
521 : Datum
522 7468 : nameicregexeq(PG_FUNCTION_ARGS)
523 : {
524 7468 : Name n = PG_GETARG_NAME(0);
525 7468 : text *p = PG_GETARG_TEXT_PP(1);
526 :
527 7468 : PG_RETURN_BOOL(RE_compile_and_execute(p,
528 : NameStr(*n),
529 : strlen(NameStr(*n)),
530 : REG_ADVANCED | REG_ICASE,
531 : PG_GET_COLLATION(),
532 : 0, NULL));
533 : }
534 :
535 : Datum
536 6 : nameicregexne(PG_FUNCTION_ARGS)
537 : {
538 6 : Name n = PG_GETARG_NAME(0);
539 6 : text *p = PG_GETARG_TEXT_PP(1);
540 :
541 6 : PG_RETURN_BOOL(!RE_compile_and_execute(p,
542 : NameStr(*n),
543 : strlen(NameStr(*n)),
544 : REG_ADVANCED | REG_ICASE,
545 : PG_GET_COLLATION(),
546 : 0, NULL));
547 : }
548 :
549 : Datum
550 220 : texticregexeq(PG_FUNCTION_ARGS)
551 : {
552 220 : text *s = PG_GETARG_TEXT_PP(0);
553 220 : text *p = PG_GETARG_TEXT_PP(1);
554 :
555 220 : PG_RETURN_BOOL(RE_compile_and_execute(p,
556 : VARDATA_ANY(s),
557 : VARSIZE_ANY_EXHDR(s),
558 : REG_ADVANCED | REG_ICASE,
559 : PG_GET_COLLATION(),
560 : 0, NULL));
561 : }
562 :
563 : Datum
564 28 : texticregexne(PG_FUNCTION_ARGS)
565 : {
566 28 : text *s = PG_GETARG_TEXT_PP(0);
567 28 : text *p = PG_GETARG_TEXT_PP(1);
568 :
569 28 : PG_RETURN_BOOL(!RE_compile_and_execute(p,
570 : VARDATA_ANY(s),
571 : VARSIZE_ANY_EXHDR(s),
572 : REG_ADVANCED | REG_ICASE,
573 : PG_GET_COLLATION(),
574 : 0, NULL));
575 : }
576 :
577 :
578 : /*
579 : * textregexsubstr()
580 : * Return a substring matched by a regular expression.
581 : */
582 : Datum
583 1638 : textregexsubstr(PG_FUNCTION_ARGS)
584 : {
585 1638 : text *s = PG_GETARG_TEXT_PP(0);
586 1638 : text *p = PG_GETARG_TEXT_PP(1);
587 : regex_t *re;
588 : regmatch_t pmatch[2];
589 : int so,
590 : eo;
591 :
592 : /* Compile RE */
593 1638 : re = RE_compile_and_cache(p, REG_ADVANCED, PG_GET_COLLATION());
594 :
595 : /*
596 : * We pass two regmatch_t structs to get info about the overall match and
597 : * the match for the first parenthesized subexpression (if any). If there
598 : * is a parenthesized subexpression, we return what it matched; else
599 : * return what the whole regexp matched.
600 : */
601 1638 : if (!RE_execute(re,
602 1638 : VARDATA_ANY(s), VARSIZE_ANY_EXHDR(s),
603 : 2, pmatch))
604 6 : PG_RETURN_NULL(); /* definitely no match */
605 :
606 1632 : if (re->re_nsub > 0)
607 : {
608 : /* has parenthesized subexpressions, use the first one */
609 1522 : so = pmatch[1].rm_so;
610 1522 : eo = pmatch[1].rm_eo;
611 : }
612 : else
613 : {
614 : /* no parenthesized subexpression, use whole match */
615 110 : so = pmatch[0].rm_so;
616 110 : eo = pmatch[0].rm_eo;
617 : }
618 :
619 : /*
620 : * It is possible to have a match to the whole pattern but no match for a
621 : * subexpression; for example 'foo(bar)?' is considered to match 'foo' but
622 : * there is no subexpression match. So this extra test for match failure
623 : * is not redundant.
624 : */
625 1632 : if (so < 0 || eo < 0)
626 6 : PG_RETURN_NULL();
627 :
628 1626 : return DirectFunctionCall3(text_substr,
629 : PointerGetDatum(s),
630 : Int32GetDatum(so + 1),
631 : Int32GetDatum(eo - so));
632 : }
633 :
634 : /*
635 : * textregexreplace_noopt()
636 : * Return a string matched by a regular expression, with replacement.
637 : *
638 : * This version doesn't have an option argument: we default to case
639 : * sensitive match, replace the first instance only.
640 : */
641 : Datum
642 14412 : textregexreplace_noopt(PG_FUNCTION_ARGS)
643 : {
644 14412 : text *s = PG_GETARG_TEXT_PP(0);
645 14412 : text *p = PG_GETARG_TEXT_PP(1);
646 14412 : text *r = PG_GETARG_TEXT_PP(2);
647 :
648 14412 : PG_RETURN_TEXT_P(replace_text_regexp(s, p, r,
649 : REG_ADVANCED, PG_GET_COLLATION(),
650 : 0, 1));
651 : }
652 :
653 : /*
654 : * textregexreplace()
655 : * Return a string matched by a regular expression, with replacement.
656 : */
657 : Datum
658 4308 : textregexreplace(PG_FUNCTION_ARGS)
659 : {
660 4308 : text *s = PG_GETARG_TEXT_PP(0);
661 4308 : text *p = PG_GETARG_TEXT_PP(1);
662 4308 : text *r = PG_GETARG_TEXT_PP(2);
663 4308 : text *opt = PG_GETARG_TEXT_PP(3);
664 : pg_re_flags flags;
665 :
666 : /*
667 : * regexp_replace() with four arguments will be preferentially resolved as
668 : * this form when the fourth argument is of type UNKNOWN. However, the
669 : * user might have intended to call textregexreplace_extended_no_n. If we
670 : * see flags that look like an integer, emit the same error that
671 : * parse_re_flags would, but add a HINT about how to fix it.
672 : */
673 4308 : if (VARSIZE_ANY_EXHDR(opt) > 0)
674 : {
675 4308 : char *opt_p = VARDATA_ANY(opt);
676 :
677 4308 : if (*opt_p >= '0' && *opt_p <= '9')
678 6 : ereport(ERROR,
679 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
680 : errmsg("invalid regular expression option: \"%.*s\"",
681 : pg_mblen(opt_p), opt_p),
682 : errhint("If you meant to use regexp_replace() with a start parameter, cast the fourth argument to integer explicitly.")));
683 : }
684 :
685 4302 : parse_re_flags(&flags, opt);
686 :
687 4296 : PG_RETURN_TEXT_P(replace_text_regexp(s, p, r,
688 : flags.cflags, PG_GET_COLLATION(),
689 : 0, flags.glob ? 0 : 1));
690 : }
691 :
692 : /*
693 : * textregexreplace_extended()
694 : * Return a string matched by a regular expression, with replacement.
695 : * Extends textregexreplace by allowing a start position and the
696 : * choice of the occurrence to replace (0 means all occurrences).
697 : */
698 : Datum
699 66 : textregexreplace_extended(PG_FUNCTION_ARGS)
700 : {
701 66 : text *s = PG_GETARG_TEXT_PP(0);
702 66 : text *p = PG_GETARG_TEXT_PP(1);
703 66 : text *r = PG_GETARG_TEXT_PP(2);
704 66 : int start = 1;
705 66 : int n = 1;
706 66 : text *flags = PG_GETARG_TEXT_PP_IF_EXISTS(5);
707 : pg_re_flags re_flags;
708 :
709 : /* Collect optional parameters */
710 66 : if (PG_NARGS() > 3)
711 : {
712 66 : start = PG_GETARG_INT32(3);
713 66 : if (start <= 0)
714 6 : ereport(ERROR,
715 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
716 : errmsg("invalid value for parameter \"%s\": %d",
717 : "start", start)));
718 : }
719 60 : if (PG_NARGS() > 4)
720 : {
721 54 : n = PG_GETARG_INT32(4);
722 54 : if (n < 0)
723 6 : ereport(ERROR,
724 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
725 : errmsg("invalid value for parameter \"%s\": %d",
726 : "n", n)));
727 : }
728 :
729 : /* Determine options */
730 54 : parse_re_flags(&re_flags, flags);
731 :
732 : /* If N was not specified, deduce it from the 'g' flag */
733 54 : if (PG_NARGS() <= 4)
734 6 : n = re_flags.glob ? 0 : 1;
735 :
736 : /* Do the replacement(s) */
737 54 : PG_RETURN_TEXT_P(replace_text_regexp(s, p, r,
738 : re_flags.cflags, PG_GET_COLLATION(),
739 : start - 1, n));
740 : }
741 :
742 : /* This is separate to keep the opr_sanity regression test from complaining */
743 : Datum
744 6 : textregexreplace_extended_no_n(PG_FUNCTION_ARGS)
745 : {
746 6 : return textregexreplace_extended(fcinfo);
747 : }
748 :
749 : /* This is separate to keep the opr_sanity regression test from complaining */
750 : Datum
751 6 : textregexreplace_extended_no_flags(PG_FUNCTION_ARGS)
752 : {
753 6 : return textregexreplace_extended(fcinfo);
754 : }
755 :
756 : /*
757 : * similar_to_escape(), similar_escape()
758 : *
759 : * Convert a SQL "SIMILAR TO" regexp pattern to POSIX style, so it can be
760 : * used by our regexp engine.
761 : *
762 : * similar_escape_internal() is the common workhorse for three SQL-exposed
763 : * functions. esc_text can be passed as NULL to select the default escape
764 : * (which is '\'), or as an empty string to select no escape character.
765 : */
766 : static text *
767 186 : similar_escape_internal(text *pat_text, text *esc_text)
768 : {
769 : text *result;
770 : char *p,
771 : *e,
772 : *r;
773 : int plen,
774 : elen;
775 186 : bool afterescape = false;
776 186 : int nquotes = 0;
777 186 : int bracket_depth = 0; /* square bracket nesting level */
778 186 : int charclass_pos = 0; /* position inside a character class */
779 :
780 186 : p = VARDATA_ANY(pat_text);
781 186 : plen = VARSIZE_ANY_EXHDR(pat_text);
782 186 : if (esc_text == NULL)
783 : {
784 : /* No ESCAPE clause provided; default to backslash as escape */
785 88 : e = "\\";
786 88 : elen = 1;
787 : }
788 : else
789 : {
790 98 : e = VARDATA_ANY(esc_text);
791 98 : elen = VARSIZE_ANY_EXHDR(esc_text);
792 98 : if (elen == 0)
793 6 : e = NULL; /* no escape character */
794 92 : else if (elen > 1)
795 : {
796 6 : int escape_mblen = pg_mbstrlen_with_len(e, elen);
797 :
798 6 : if (escape_mblen > 1)
799 6 : ereport(ERROR,
800 : (errcode(ERRCODE_INVALID_ESCAPE_SEQUENCE),
801 : errmsg("invalid escape string"),
802 : errhint("Escape string must be empty or one character.")));
803 : }
804 : }
805 :
806 : /*----------
807 : * We surround the transformed input string with
808 : * ^(?: ... )$
809 : * which requires some explanation. We need "^" and "$" to force
810 : * the pattern to match the entire input string as per the SQL spec.
811 : * The "(?:" and ")" are a non-capturing set of parens; we have to have
812 : * parens in case the string contains "|", else the "^" and "$" will
813 : * be bound into the first and last alternatives which is not what we
814 : * want, and the parens must be non capturing because we don't want them
815 : * to count when selecting output for SUBSTRING.
816 : *
817 : * When the pattern is divided into three parts by escape-double-quotes,
818 : * what we emit is
819 : * ^(?:part1){1,1}?(part2){1,1}(?:part3)$
820 : * which requires even more explanation. The "{1,1}?" on part1 makes it
821 : * non-greedy so that it will match the smallest possible amount of text
822 : * not the largest, as required by SQL. The plain parens around part2
823 : * are capturing parens so that that part is what controls the result of
824 : * SUBSTRING. The "{1,1}" forces part2 to be greedy, so that it matches
825 : * the largest possible amount of text; hence part3 must match the
826 : * smallest amount of text, as required by SQL. We don't need an explicit
827 : * greediness marker on part3. Note that this also confines the effects
828 : * of any "|" characters to the respective part, which is what we want.
829 : *
830 : * The SQL spec says that SUBSTRING's pattern must contain exactly two
831 : * escape-double-quotes, but we only complain if there's more than two.
832 : * With none, we act as though part1 and part3 are empty; with one, we
833 : * act as though part3 is empty. Both behaviors fall out of omitting
834 : * the relevant part separators in the above expansion. If the result
835 : * of this function is used in a plain regexp match (SIMILAR TO), the
836 : * escape-double-quotes have no effect on the match behavior.
837 : *
838 : * While we don't fully validate character classes (bracket expressions),
839 : * we do need to parse them well enough to know where they end.
840 : * "charclass_pos" tracks where we are in a character class.
841 : * Its value is uninteresting when bracket_depth is 0.
842 : * But when bracket_depth > 0, it will be
843 : * 1: right after the opening '[' (a following '^' will negate
844 : * the class, while ']' is a literal character)
845 : * 2: right after a '^' after the opening '[' (']' is still a literal
846 : * character)
847 : * 3 or more: further inside the character class (']' ends the class)
848 : *----------
849 : */
850 :
851 : /*
852 : * We need room for the prefix/postfix and part separators, plus as many
853 : * as 3 output bytes per input byte; since the input is at most 1GB this
854 : * can't overflow size_t.
855 : */
856 180 : result = (text *) palloc(VARHDRSZ + 23 + 3 * (size_t) plen);
857 180 : r = VARDATA(result);
858 :
859 180 : *r++ = '^';
860 180 : *r++ = '(';
861 180 : *r++ = '?';
862 180 : *r++ = ':';
863 :
864 1840 : while (plen > 0)
865 : {
866 1666 : char pchar = *p;
867 :
868 : /*
869 : * If both the escape character and the current character from the
870 : * pattern are multi-byte, we need to take the slow path.
871 : *
872 : * But if one of them is single-byte, we can process the pattern one
873 : * byte at a time, ignoring multi-byte characters. (This works
874 : * because all server-encodings have the property that a valid
875 : * multi-byte character representation cannot contain the
876 : * representation of a valid single-byte character.)
877 : */
878 :
879 1666 : if (elen > 1)
880 : {
881 0 : int mblen = pg_mblen(p);
882 :
883 0 : if (mblen > 1)
884 : {
885 : /* slow, multi-byte path */
886 0 : if (afterescape)
887 : {
888 0 : *r++ = '\\';
889 0 : memcpy(r, p, mblen);
890 0 : r += mblen;
891 0 : afterescape = false;
892 : }
893 0 : else if (e && elen == mblen && memcmp(e, p, mblen) == 0)
894 : {
895 : /* SQL escape character; do not send to output */
896 0 : afterescape = true;
897 : }
898 : else
899 : {
900 : /*
901 : * We know it's a multi-byte character, so we don't need
902 : * to do all the comparisons to single-byte characters
903 : * that we do below.
904 : */
905 0 : memcpy(r, p, mblen);
906 0 : r += mblen;
907 : }
908 :
909 0 : p += mblen;
910 0 : plen -= mblen;
911 :
912 0 : continue;
913 : }
914 : }
915 :
916 : /* fast path */
917 1666 : if (afterescape)
918 : {
919 166 : if (pchar == '"' && bracket_depth < 1) /* escape-double-quote? */
920 : {
921 : /* emit appropriate part separator, per notes above */
922 124 : if (nquotes == 0)
923 : {
924 62 : *r++ = ')';
925 62 : *r++ = '{';
926 62 : *r++ = '1';
927 62 : *r++ = ',';
928 62 : *r++ = '1';
929 62 : *r++ = '}';
930 62 : *r++ = '?';
931 62 : *r++ = '(';
932 : }
933 62 : else if (nquotes == 1)
934 : {
935 56 : *r++ = ')';
936 56 : *r++ = '{';
937 56 : *r++ = '1';
938 56 : *r++ = ',';
939 56 : *r++ = '1';
940 56 : *r++ = '}';
941 56 : *r++ = '(';
942 56 : *r++ = '?';
943 56 : *r++ = ':';
944 : }
945 : else
946 6 : ereport(ERROR,
947 : (errcode(ERRCODE_INVALID_USE_OF_ESCAPE_CHARACTER),
948 : errmsg("SQL regular expression may not contain more than two escape-double-quote separators")));
949 118 : nquotes++;
950 : }
951 : else
952 : {
953 : /*
954 : * We allow any character at all to be escaped; notably, this
955 : * allows access to POSIX character-class escapes such as
956 : * "\d". The SQL spec is considerably more restrictive.
957 : */
958 42 : *r++ = '\\';
959 42 : *r++ = pchar;
960 :
961 : /*
962 : * If we encounter an escaped character in a character class,
963 : * we are no longer at the beginning.
964 : */
965 42 : charclass_pos = 3;
966 : }
967 160 : afterescape = false;
968 : }
969 1500 : else if (e && pchar == *e)
970 : {
971 : /* SQL escape character; do not send to output */
972 166 : afterescape = true;
973 : }
974 1334 : else if (bracket_depth > 0)
975 : {
976 : /* inside a character class */
977 612 : if (pchar == '\\')
978 : {
979 : /*
980 : * If we're here, backslash is not the SQL escape character,
981 : * so treat it as a literal class element, which requires
982 : * doubling it. (This matches our behavior for backslashes
983 : * outside character classes.)
984 : */
985 0 : *r++ = '\\';
986 : }
987 612 : *r++ = pchar;
988 :
989 : /* parse the character class well enough to identify ending ']' */
990 612 : if (pchar == ']' && charclass_pos > 2)
991 : {
992 : /* found the real end of a bracket pair */
993 138 : bracket_depth--;
994 : /* don't reset charclass_pos, this may be an inner bracket */
995 : }
996 474 : else if (pchar == '[')
997 : {
998 : /* start of a nested bracket pair */
999 72 : bracket_depth++;
1000 :
1001 : /*
1002 : * We are no longer at the beginning of a character class.
1003 : * (The nested bracket pair is a collating element, not a
1004 : * character class in its own right.)
1005 : */
1006 72 : charclass_pos = 3;
1007 : }
1008 402 : else if (pchar == '^')
1009 : {
1010 : /*
1011 : * A caret right after the opening bracket negates the
1012 : * character class. In that case, the following will
1013 : * increment charclass_pos from 1 to 2, so that a following
1014 : * ']' is still a literal character and does not end the
1015 : * character class. If we are further inside a character
1016 : * class, charclass_pos might get incremented past 3, which is
1017 : * fine.
1018 : */
1019 60 : charclass_pos++;
1020 : }
1021 : else
1022 : {
1023 : /*
1024 : * Anything else (including a backslash or leading ']') is an
1025 : * element of the character class, so we are no longer at the
1026 : * beginning of the class.
1027 : */
1028 342 : charclass_pos = 3;
1029 : }
1030 : }
1031 722 : else if (pchar == '[')
1032 : {
1033 : /* start of a character class */
1034 66 : *r++ = pchar;
1035 66 : bracket_depth = 1;
1036 66 : charclass_pos = 1;
1037 : }
1038 656 : else if (pchar == '%')
1039 : {
1040 132 : *r++ = '.';
1041 132 : *r++ = '*';
1042 : }
1043 524 : else if (pchar == '_')
1044 64 : *r++ = '.';
1045 460 : else if (pchar == '(')
1046 : {
1047 : /* convert to non-capturing parenthesis */
1048 30 : *r++ = '(';
1049 30 : *r++ = '?';
1050 30 : *r++ = ':';
1051 : }
1052 430 : else if (pchar == '\\' || pchar == '.' ||
1053 390 : pchar == '^' || pchar == '$')
1054 : {
1055 52 : *r++ = '\\';
1056 52 : *r++ = pchar;
1057 : }
1058 : else
1059 378 : *r++ = pchar;
1060 1660 : p++, plen--;
1061 : }
1062 :
1063 174 : *r++ = ')';
1064 174 : *r++ = '$';
1065 :
1066 174 : SET_VARSIZE(result, r - ((char *) result));
1067 :
1068 174 : return result;
1069 : }
1070 :
1071 : /*
1072 : * similar_to_escape(pattern, escape)
1073 : */
1074 : Datum
1075 98 : similar_to_escape_2(PG_FUNCTION_ARGS)
1076 : {
1077 98 : text *pat_text = PG_GETARG_TEXT_PP(0);
1078 98 : text *esc_text = PG_GETARG_TEXT_PP(1);
1079 : text *result;
1080 :
1081 98 : result = similar_escape_internal(pat_text, esc_text);
1082 :
1083 86 : PG_RETURN_TEXT_P(result);
1084 : }
1085 :
1086 : /*
1087 : * similar_to_escape(pattern)
1088 : * Inserts a default escape character.
1089 : */
1090 : Datum
1091 88 : similar_to_escape_1(PG_FUNCTION_ARGS)
1092 : {
1093 88 : text *pat_text = PG_GETARG_TEXT_PP(0);
1094 : text *result;
1095 :
1096 88 : result = similar_escape_internal(pat_text, NULL);
1097 :
1098 88 : PG_RETURN_TEXT_P(result);
1099 : }
1100 :
1101 : /*
1102 : * similar_escape(pattern, escape)
1103 : *
1104 : * Legacy function for compatibility with views stored using the
1105 : * pre-v13 expansion of SIMILAR TO. Unlike the above functions, this
1106 : * is non-strict, which leads to not-per-spec handling of "ESCAPE NULL".
1107 : */
1108 : Datum
1109 0 : similar_escape(PG_FUNCTION_ARGS)
1110 : {
1111 : text *pat_text;
1112 : text *esc_text;
1113 : text *result;
1114 :
1115 : /* This function is not strict, so must test explicitly */
1116 0 : if (PG_ARGISNULL(0))
1117 0 : PG_RETURN_NULL();
1118 0 : pat_text = PG_GETARG_TEXT_PP(0);
1119 :
1120 0 : if (PG_ARGISNULL(1))
1121 0 : esc_text = NULL; /* use default escape character */
1122 : else
1123 0 : esc_text = PG_GETARG_TEXT_PP(1);
1124 :
1125 0 : result = similar_escape_internal(pat_text, esc_text);
1126 :
1127 0 : PG_RETURN_TEXT_P(result);
1128 : }
1129 :
1130 : /*
1131 : * regexp_count()
1132 : * Return the number of matches of a pattern within a string.
1133 : */
1134 : Datum
1135 48 : regexp_count(PG_FUNCTION_ARGS)
1136 : {
1137 48 : text *str = PG_GETARG_TEXT_PP(0);
1138 48 : text *pattern = PG_GETARG_TEXT_PP(1);
1139 48 : int start = 1;
1140 48 : text *flags = PG_GETARG_TEXT_PP_IF_EXISTS(3);
1141 : pg_re_flags re_flags;
1142 : regexp_matches_ctx *matchctx;
1143 :
1144 : /* Collect optional parameters */
1145 48 : if (PG_NARGS() > 2)
1146 : {
1147 42 : start = PG_GETARG_INT32(2);
1148 42 : if (start <= 0)
1149 12 : ereport(ERROR,
1150 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
1151 : errmsg("invalid value for parameter \"%s\": %d",
1152 : "start", start)));
1153 : }
1154 :
1155 : /* Determine options */
1156 36 : parse_re_flags(&re_flags, flags);
1157 : /* User mustn't specify 'g' */
1158 36 : if (re_flags.glob)
1159 0 : ereport(ERROR,
1160 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
1161 : /* translator: %s is a SQL function name */
1162 : errmsg("%s does not support the \"global\" option",
1163 : "regexp_count()")));
1164 : /* But we find all the matches anyway */
1165 36 : re_flags.glob = true;
1166 :
1167 : /* Do the matching */
1168 36 : matchctx = setup_regexp_matches(str, pattern, &re_flags, start - 1,
1169 : PG_GET_COLLATION(),
1170 : false, /* can ignore subexprs */
1171 : false, false);
1172 :
1173 36 : PG_RETURN_INT32(matchctx->nmatches);
1174 : }
1175 :
1176 : /* This is separate to keep the opr_sanity regression test from complaining */
1177 : Datum
1178 6 : regexp_count_no_start(PG_FUNCTION_ARGS)
1179 : {
1180 6 : return regexp_count(fcinfo);
1181 : }
1182 :
1183 : /* This is separate to keep the opr_sanity regression test from complaining */
1184 : Datum
1185 30 : regexp_count_no_flags(PG_FUNCTION_ARGS)
1186 : {
1187 30 : return regexp_count(fcinfo);
1188 : }
1189 :
1190 : /*
1191 : * regexp_instr()
1192 : * Return the match's position within the string
1193 : */
1194 : Datum
1195 156 : regexp_instr(PG_FUNCTION_ARGS)
1196 : {
1197 156 : text *str = PG_GETARG_TEXT_PP(0);
1198 156 : text *pattern = PG_GETARG_TEXT_PP(1);
1199 156 : int start = 1;
1200 156 : int n = 1;
1201 156 : int endoption = 0;
1202 156 : text *flags = PG_GETARG_TEXT_PP_IF_EXISTS(5);
1203 156 : int subexpr = 0;
1204 : int pos;
1205 : pg_re_flags re_flags;
1206 : regexp_matches_ctx *matchctx;
1207 :
1208 : /* Collect optional parameters */
1209 156 : if (PG_NARGS() > 2)
1210 : {
1211 138 : start = PG_GETARG_INT32(2);
1212 138 : if (start <= 0)
1213 6 : ereport(ERROR,
1214 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
1215 : errmsg("invalid value for parameter \"%s\": %d",
1216 : "start", start)));
1217 : }
1218 150 : if (PG_NARGS() > 3)
1219 : {
1220 126 : n = PG_GETARG_INT32(3);
1221 126 : if (n <= 0)
1222 6 : ereport(ERROR,
1223 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
1224 : errmsg("invalid value for parameter \"%s\": %d",
1225 : "n", n)));
1226 : }
1227 144 : if (PG_NARGS() > 4)
1228 : {
1229 108 : endoption = PG_GETARG_INT32(4);
1230 108 : if (endoption != 0 && endoption != 1)
1231 12 : ereport(ERROR,
1232 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
1233 : errmsg("invalid value for parameter \"%s\": %d",
1234 : "endoption", endoption)));
1235 : }
1236 132 : if (PG_NARGS() > 6)
1237 : {
1238 84 : subexpr = PG_GETARG_INT32(6);
1239 84 : if (subexpr < 0)
1240 6 : ereport(ERROR,
1241 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
1242 : errmsg("invalid value for parameter \"%s\": %d",
1243 : "subexpr", subexpr)));
1244 : }
1245 :
1246 : /* Determine options */
1247 126 : parse_re_flags(&re_flags, flags);
1248 : /* User mustn't specify 'g' */
1249 126 : if (re_flags.glob)
1250 6 : ereport(ERROR,
1251 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
1252 : /* translator: %s is a SQL function name */
1253 : errmsg("%s does not support the \"global\" option",
1254 : "regexp_instr()")));
1255 : /* But we find all the matches anyway */
1256 120 : re_flags.glob = true;
1257 :
1258 : /* Do the matching */
1259 120 : matchctx = setup_regexp_matches(str, pattern, &re_flags, start - 1,
1260 : PG_GET_COLLATION(),
1261 : (subexpr > 0), /* need submatches? */
1262 : false, false);
1263 :
1264 : /* When n exceeds matches return 0 (includes case of no matches) */
1265 120 : if (n > matchctx->nmatches)
1266 12 : PG_RETURN_INT32(0);
1267 :
1268 : /* When subexpr exceeds number of subexpressions return 0 */
1269 108 : if (subexpr > matchctx->npatterns)
1270 12 : PG_RETURN_INT32(0);
1271 :
1272 : /* Select the appropriate match position to return */
1273 96 : pos = (n - 1) * matchctx->npatterns;
1274 96 : if (subexpr > 0)
1275 54 : pos += subexpr - 1;
1276 96 : pos *= 2;
1277 96 : if (endoption == 1)
1278 30 : pos += 1;
1279 :
1280 96 : if (matchctx->match_locs[pos] >= 0)
1281 90 : PG_RETURN_INT32(matchctx->match_locs[pos] + 1);
1282 : else
1283 6 : PG_RETURN_INT32(0); /* position not identifiable */
1284 : }
1285 :
1286 : /* This is separate to keep the opr_sanity regression test from complaining */
1287 : Datum
1288 18 : regexp_instr_no_start(PG_FUNCTION_ARGS)
1289 : {
1290 18 : return regexp_instr(fcinfo);
1291 : }
1292 :
1293 : /* This is separate to keep the opr_sanity regression test from complaining */
1294 : Datum
1295 6 : regexp_instr_no_n(PG_FUNCTION_ARGS)
1296 : {
1297 6 : return regexp_instr(fcinfo);
1298 : }
1299 :
1300 : /* This is separate to keep the opr_sanity regression test from complaining */
1301 : Datum
1302 24 : regexp_instr_no_endoption(PG_FUNCTION_ARGS)
1303 : {
1304 24 : return regexp_instr(fcinfo);
1305 : }
1306 :
1307 : /* This is separate to keep the opr_sanity regression test from complaining */
1308 : Datum
1309 12 : regexp_instr_no_flags(PG_FUNCTION_ARGS)
1310 : {
1311 12 : return regexp_instr(fcinfo);
1312 : }
1313 :
1314 : /* This is separate to keep the opr_sanity regression test from complaining */
1315 : Datum
1316 12 : regexp_instr_no_subexpr(PG_FUNCTION_ARGS)
1317 : {
1318 12 : return regexp_instr(fcinfo);
1319 : }
1320 :
1321 : /*
1322 : * regexp_like()
1323 : * Test for a pattern match within a string.
1324 : */
1325 : Datum
1326 30 : regexp_like(PG_FUNCTION_ARGS)
1327 : {
1328 30 : text *str = PG_GETARG_TEXT_PP(0);
1329 30 : text *pattern = PG_GETARG_TEXT_PP(1);
1330 30 : text *flags = PG_GETARG_TEXT_PP_IF_EXISTS(2);
1331 : pg_re_flags re_flags;
1332 :
1333 : /* Determine options */
1334 30 : parse_re_flags(&re_flags, flags);
1335 : /* User mustn't specify 'g' */
1336 30 : if (re_flags.glob)
1337 6 : ereport(ERROR,
1338 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
1339 : /* translator: %s is a SQL function name */
1340 : errmsg("%s does not support the \"global\" option",
1341 : "regexp_like()")));
1342 :
1343 : /* Otherwise it's like textregexeq/texticregexeq */
1344 24 : PG_RETURN_BOOL(RE_compile_and_execute(pattern,
1345 : VARDATA_ANY(str),
1346 : VARSIZE_ANY_EXHDR(str),
1347 : re_flags.cflags,
1348 : PG_GET_COLLATION(),
1349 : 0, NULL));
1350 : }
1351 :
1352 : /* This is separate to keep the opr_sanity regression test from complaining */
1353 : Datum
1354 6 : regexp_like_no_flags(PG_FUNCTION_ARGS)
1355 : {
1356 6 : return regexp_like(fcinfo);
1357 : }
1358 :
1359 : /*
1360 : * regexp_match()
1361 : * Return the first substring(s) matching a pattern within a string.
1362 : */
1363 : Datum
1364 2540 : regexp_match(PG_FUNCTION_ARGS)
1365 : {
1366 2540 : text *orig_str = PG_GETARG_TEXT_PP(0);
1367 2540 : text *pattern = PG_GETARG_TEXT_PP(1);
1368 2540 : text *flags = PG_GETARG_TEXT_PP_IF_EXISTS(2);
1369 : pg_re_flags re_flags;
1370 : regexp_matches_ctx *matchctx;
1371 :
1372 : /* Determine options */
1373 2540 : parse_re_flags(&re_flags, flags);
1374 : /* User mustn't specify 'g' */
1375 2540 : if (re_flags.glob)
1376 8 : ereport(ERROR,
1377 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
1378 : /* translator: %s is a SQL function name */
1379 : errmsg("%s does not support the \"global\" option",
1380 : "regexp_match()"),
1381 : errhint("Use the regexp_matches function instead.")));
1382 :
1383 2532 : matchctx = setup_regexp_matches(orig_str, pattern, &re_flags, 0,
1384 : PG_GET_COLLATION(), true, false, false);
1385 :
1386 2532 : if (matchctx->nmatches == 0)
1387 130 : PG_RETURN_NULL();
1388 :
1389 : Assert(matchctx->nmatches == 1);
1390 :
1391 : /* Create workspace that build_regexp_match_result needs */
1392 2402 : matchctx->elems = (Datum *) palloc(sizeof(Datum) * matchctx->npatterns);
1393 2402 : matchctx->nulls = (bool *) palloc(sizeof(bool) * matchctx->npatterns);
1394 :
1395 2402 : PG_RETURN_DATUM(PointerGetDatum(build_regexp_match_result(matchctx)));
1396 : }
1397 :
1398 : /* This is separate to keep the opr_sanity regression test from complaining */
1399 : Datum
1400 2510 : regexp_match_no_flags(PG_FUNCTION_ARGS)
1401 : {
1402 2510 : return regexp_match(fcinfo);
1403 : }
1404 :
1405 : /*
1406 : * regexp_matches()
1407 : * Return a table of all matches of a pattern within a string.
1408 : */
1409 : Datum
1410 2610 : regexp_matches(PG_FUNCTION_ARGS)
1411 : {
1412 : FuncCallContext *funcctx;
1413 : regexp_matches_ctx *matchctx;
1414 :
1415 2610 : if (SRF_IS_FIRSTCALL())
1416 : {
1417 1950 : text *pattern = PG_GETARG_TEXT_PP(1);
1418 1950 : text *flags = PG_GETARG_TEXT_PP_IF_EXISTS(2);
1419 : pg_re_flags re_flags;
1420 : MemoryContext oldcontext;
1421 :
1422 1950 : funcctx = SRF_FIRSTCALL_INIT();
1423 1950 : oldcontext = MemoryContextSwitchTo(funcctx->multi_call_memory_ctx);
1424 :
1425 : /* Determine options */
1426 1950 : parse_re_flags(&re_flags, flags);
1427 :
1428 : /* be sure to copy the input string into the multi-call ctx */
1429 1944 : matchctx = setup_regexp_matches(PG_GETARG_TEXT_P_COPY(0), pattern,
1430 : &re_flags, 0,
1431 : PG_GET_COLLATION(),
1432 : true, false, false);
1433 :
1434 : /* Pre-create workspace that build_regexp_match_result needs */
1435 1932 : matchctx->elems = (Datum *) palloc(sizeof(Datum) * matchctx->npatterns);
1436 1932 : matchctx->nulls = (bool *) palloc(sizeof(bool) * matchctx->npatterns);
1437 :
1438 1932 : MemoryContextSwitchTo(oldcontext);
1439 1932 : funcctx->user_fctx = matchctx;
1440 : }
1441 :
1442 2592 : funcctx = SRF_PERCALL_SETUP();
1443 2592 : matchctx = (regexp_matches_ctx *) funcctx->user_fctx;
1444 :
1445 2592 : if (matchctx->next_match < matchctx->nmatches)
1446 : {
1447 : ArrayType *result_ary;
1448 :
1449 660 : result_ary = build_regexp_match_result(matchctx);
1450 660 : matchctx->next_match++;
1451 660 : SRF_RETURN_NEXT(funcctx, PointerGetDatum(result_ary));
1452 : }
1453 :
1454 1932 : SRF_RETURN_DONE(funcctx);
1455 : }
1456 :
1457 : /* This is separate to keep the opr_sanity regression test from complaining */
1458 : Datum
1459 2286 : regexp_matches_no_flags(PG_FUNCTION_ARGS)
1460 : {
1461 2286 : return regexp_matches(fcinfo);
1462 : }
1463 :
1464 : /*
1465 : * setup_regexp_matches --- do the initial matching for regexp_match,
1466 : * regexp_split, and related functions
1467 : *
1468 : * To avoid having to re-find the compiled pattern on each call, we do
1469 : * all the matching in one swoop. The returned regexp_matches_ctx contains
1470 : * the locations of all the substrings matching the pattern.
1471 : *
1472 : * start_search: the character (not byte) offset in orig_str at which to
1473 : * begin the search. Returned positions are relative to orig_str anyway.
1474 : * use_subpatterns: collect data about matches to parenthesized subexpressions.
1475 : * ignore_degenerate: ignore zero-length matches.
1476 : * fetching_unmatched: caller wants to fetch unmatched substrings.
1477 : *
1478 : * We don't currently assume that fetching_unmatched is exclusive of fetching
1479 : * the matched text too; if it's set, the conversion buffer is large enough to
1480 : * fetch any single matched or unmatched string, but not any larger
1481 : * substring. (In practice, when splitting the matches are usually small
1482 : * anyway, and it didn't seem worth complicating the code further.)
1483 : */
1484 : static regexp_matches_ctx *
1485 205170 : setup_regexp_matches(text *orig_str, text *pattern, pg_re_flags *re_flags,
1486 : int start_search,
1487 : Oid collation,
1488 : bool use_subpatterns,
1489 : bool ignore_degenerate,
1490 : bool fetching_unmatched)
1491 : {
1492 205170 : regexp_matches_ctx *matchctx = palloc0(sizeof(regexp_matches_ctx));
1493 205170 : int eml = pg_database_encoding_max_length();
1494 : int orig_len;
1495 : pg_wchar *wide_str;
1496 : int wide_len;
1497 : int cflags;
1498 : regex_t *cpattern;
1499 : regmatch_t *pmatch;
1500 : int pmatch_len;
1501 : int array_len;
1502 : int array_idx;
1503 : int prev_match_end;
1504 : int prev_valid_match_end;
1505 205170 : int maxlen = 0; /* largest fetch length in characters */
1506 :
1507 : /* save original string --- we'll extract result substrings from it */
1508 205170 : matchctx->orig_str = orig_str;
1509 :
1510 : /* convert string to pg_wchar form for matching */
1511 205170 : orig_len = VARSIZE_ANY_EXHDR(orig_str);
1512 205170 : wide_str = (pg_wchar *) palloc(sizeof(pg_wchar) * (orig_len + 1));
1513 205170 : wide_len = pg_mb2wchar_with_len(VARDATA_ANY(orig_str), wide_str, orig_len);
1514 :
1515 : /* set up the compiled pattern */
1516 205170 : cflags = re_flags->cflags;
1517 205170 : if (!use_subpatterns)
1518 200592 : cflags |= REG_NOSUB;
1519 205170 : cpattern = RE_compile_and_cache(pattern, cflags, collation);
1520 :
1521 : /* do we want to remember subpatterns? */
1522 205158 : if (use_subpatterns && cpattern->re_nsub > 0)
1523 : {
1524 2694 : matchctx->npatterns = cpattern->re_nsub;
1525 2694 : pmatch_len = cpattern->re_nsub + 1;
1526 : }
1527 : else
1528 : {
1529 202464 : use_subpatterns = false;
1530 202464 : matchctx->npatterns = 1;
1531 202464 : pmatch_len = 1;
1532 : }
1533 :
1534 : /* temporary output space for RE package */
1535 205158 : pmatch = palloc(sizeof(regmatch_t) * pmatch_len);
1536 :
1537 : /*
1538 : * the real output space (grown dynamically if needed)
1539 : *
1540 : * use values 2^n-1, not 2^n, so that we hit the limit at 2^28-1 rather
1541 : * than at 2^27
1542 : */
1543 205158 : array_len = re_flags->glob ? 255 : 31;
1544 205158 : matchctx->match_locs = (int *) palloc(sizeof(int) * array_len);
1545 205158 : array_idx = 0;
1546 :
1547 : /* search for the pattern, perhaps repeatedly */
1548 205158 : prev_match_end = 0;
1549 205158 : prev_valid_match_end = 0;
1550 1096970 : while (RE_wchar_execute(cpattern, wide_str, wide_len, start_search,
1551 : pmatch_len, pmatch))
1552 : {
1553 : /*
1554 : * If requested, ignore degenerate matches, which are zero-length
1555 : * matches occurring at the start or end of a string or just after a
1556 : * previous match.
1557 : */
1558 894720 : if (!ignore_degenerate ||
1559 891262 : (pmatch[0].rm_so < wide_len &&
1560 891220 : pmatch[0].rm_eo > prev_match_end))
1561 : {
1562 : /* enlarge output space if needed */
1563 894900 : while (array_idx + matchctx->npatterns * 2 + 1 > array_len)
1564 : {
1565 360 : array_len += array_len + 1; /* 2^n-1 => 2^(n+1)-1 */
1566 360 : if (array_len > MaxAllocSize / sizeof(int))
1567 0 : ereport(ERROR,
1568 : (errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
1569 : errmsg("too many regular expression matches")));
1570 360 : matchctx->match_locs = (int *) repalloc(matchctx->match_locs,
1571 : sizeof(int) * array_len);
1572 : }
1573 :
1574 : /* save this match's locations */
1575 894540 : if (use_subpatterns)
1576 : {
1577 : int i;
1578 :
1579 7944 : for (i = 1; i <= matchctx->npatterns; i++)
1580 : {
1581 5362 : int so = pmatch[i].rm_so;
1582 5362 : int eo = pmatch[i].rm_eo;
1583 :
1584 5362 : matchctx->match_locs[array_idx++] = so;
1585 5362 : matchctx->match_locs[array_idx++] = eo;
1586 5362 : if (so >= 0 && eo >= 0 && (eo - so) > maxlen)
1587 3444 : maxlen = (eo - so);
1588 : }
1589 : }
1590 : else
1591 : {
1592 891958 : int so = pmatch[0].rm_so;
1593 891958 : int eo = pmatch[0].rm_eo;
1594 :
1595 891958 : matchctx->match_locs[array_idx++] = so;
1596 891958 : matchctx->match_locs[array_idx++] = eo;
1597 891958 : if (so >= 0 && eo >= 0 && (eo - so) > maxlen)
1598 200906 : maxlen = (eo - so);
1599 : }
1600 894540 : matchctx->nmatches++;
1601 :
1602 : /*
1603 : * check length of unmatched portion between end of previous valid
1604 : * (nondegenerate, or degenerate but not ignored) match and start
1605 : * of current one
1606 : */
1607 894540 : if (fetching_unmatched &&
1608 891082 : pmatch[0].rm_so >= 0 &&
1609 891082 : (pmatch[0].rm_so - prev_valid_match_end) > maxlen)
1610 380706 : maxlen = (pmatch[0].rm_so - prev_valid_match_end);
1611 894540 : prev_valid_match_end = pmatch[0].rm_eo;
1612 : }
1613 894720 : prev_match_end = pmatch[0].rm_eo;
1614 :
1615 : /* if not glob, stop after one match */
1616 894720 : if (!re_flags->glob)
1617 2842 : break;
1618 :
1619 : /*
1620 : * Advance search position. Normally we start the next search at the
1621 : * end of the previous match; but if the match was of zero length, we
1622 : * have to advance by one character, or we'd just find the same match
1623 : * again.
1624 : */
1625 891878 : start_search = prev_match_end;
1626 891878 : if (pmatch[0].rm_so == pmatch[0].rm_eo)
1627 1176 : start_search++;
1628 891878 : if (start_search > wide_len)
1629 66 : break;
1630 : }
1631 :
1632 : /*
1633 : * check length of unmatched portion between end of last match and end of
1634 : * input string
1635 : */
1636 205158 : if (fetching_unmatched &&
1637 200454 : (wide_len - prev_valid_match_end) > maxlen)
1638 72 : maxlen = (wide_len - prev_valid_match_end);
1639 :
1640 : /*
1641 : * Keep a note of the end position of the string for the benefit of
1642 : * splitting code.
1643 : */
1644 205158 : matchctx->match_locs[array_idx] = wide_len;
1645 :
1646 205158 : if (eml > 1)
1647 : {
1648 205158 : int64 maxsiz = eml * (int64) maxlen;
1649 : int conv_bufsiz;
1650 :
1651 : /*
1652 : * Make the conversion buffer large enough for any substring of
1653 : * interest.
1654 : *
1655 : * Worst case: assume we need the maximum size (maxlen*eml), but take
1656 : * advantage of the fact that the original string length in bytes is
1657 : * an upper bound on the byte length of any fetched substring (and we
1658 : * know that len+1 is safe to allocate because the varlena header is
1659 : * longer than 1 byte).
1660 : */
1661 205158 : if (maxsiz > orig_len)
1662 200820 : conv_bufsiz = orig_len + 1;
1663 : else
1664 4338 : conv_bufsiz = maxsiz + 1; /* safe since maxsiz < 2^30 */
1665 :
1666 205158 : matchctx->conv_buf = palloc(conv_bufsiz);
1667 205158 : matchctx->conv_bufsiz = conv_bufsiz;
1668 205158 : matchctx->wide_str = wide_str;
1669 : }
1670 : else
1671 : {
1672 : /* No need to keep the wide string if we're in a single-byte charset. */
1673 0 : pfree(wide_str);
1674 0 : matchctx->wide_str = NULL;
1675 0 : matchctx->conv_buf = NULL;
1676 0 : matchctx->conv_bufsiz = 0;
1677 : }
1678 :
1679 : /* Clean up temp storage */
1680 205158 : pfree(pmatch);
1681 :
1682 205158 : return matchctx;
1683 : }
1684 :
1685 : /*
1686 : * build_regexp_match_result - build output array for current match
1687 : */
1688 : static ArrayType *
1689 3062 : build_regexp_match_result(regexp_matches_ctx *matchctx)
1690 : {
1691 3062 : char *buf = matchctx->conv_buf;
1692 3062 : Datum *elems = matchctx->elems;
1693 3062 : bool *nulls = matchctx->nulls;
1694 : int dims[1];
1695 : int lbs[1];
1696 : int loc;
1697 : int i;
1698 :
1699 : /* Extract matching substrings from the original string */
1700 3062 : loc = matchctx->next_match * matchctx->npatterns * 2;
1701 8634 : for (i = 0; i < matchctx->npatterns; i++)
1702 : {
1703 5572 : int so = matchctx->match_locs[loc++];
1704 5572 : int eo = matchctx->match_locs[loc++];
1705 :
1706 5572 : if (so < 0 || eo < 0)
1707 : {
1708 6 : elems[i] = (Datum) 0;
1709 6 : nulls[i] = true;
1710 : }
1711 5566 : else if (buf)
1712 : {
1713 5566 : int len = pg_wchar2mb_with_len(matchctx->wide_str + so,
1714 : buf,
1715 : eo - so);
1716 :
1717 : Assert(len < matchctx->conv_bufsiz);
1718 5566 : elems[i] = PointerGetDatum(cstring_to_text_with_len(buf, len));
1719 5566 : nulls[i] = false;
1720 : }
1721 : else
1722 : {
1723 0 : elems[i] = DirectFunctionCall3(text_substr,
1724 : PointerGetDatum(matchctx->orig_str),
1725 : Int32GetDatum(so + 1),
1726 : Int32GetDatum(eo - so));
1727 0 : nulls[i] = false;
1728 : }
1729 : }
1730 :
1731 : /* And form an array */
1732 3062 : dims[0] = matchctx->npatterns;
1733 3062 : lbs[0] = 1;
1734 : /* XXX: this hardcodes assumptions about the text type */
1735 3062 : return construct_md_array(elems, nulls, 1, dims, lbs,
1736 : TEXTOID, -1, false, TYPALIGN_INT);
1737 : }
1738 :
1739 : /*
1740 : * regexp_split_to_table()
1741 : * Split the string at matches of the pattern, returning the
1742 : * split-out substrings as a table.
1743 : */
1744 : Datum
1745 622 : regexp_split_to_table(PG_FUNCTION_ARGS)
1746 : {
1747 : FuncCallContext *funcctx;
1748 : regexp_matches_ctx *splitctx;
1749 :
1750 622 : if (SRF_IS_FIRSTCALL())
1751 : {
1752 52 : text *pattern = PG_GETARG_TEXT_PP(1);
1753 52 : text *flags = PG_GETARG_TEXT_PP_IF_EXISTS(2);
1754 : pg_re_flags re_flags;
1755 : MemoryContext oldcontext;
1756 :
1757 52 : funcctx = SRF_FIRSTCALL_INIT();
1758 52 : oldcontext = MemoryContextSwitchTo(funcctx->multi_call_memory_ctx);
1759 :
1760 : /* Determine options */
1761 52 : parse_re_flags(&re_flags, flags);
1762 : /* User mustn't specify 'g' */
1763 46 : if (re_flags.glob)
1764 6 : ereport(ERROR,
1765 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
1766 : /* translator: %s is a SQL function name */
1767 : errmsg("%s does not support the \"global\" option",
1768 : "regexp_split_to_table()")));
1769 : /* But we find all the matches anyway */
1770 40 : re_flags.glob = true;
1771 :
1772 : /* be sure to copy the input string into the multi-call ctx */
1773 40 : splitctx = setup_regexp_matches(PG_GETARG_TEXT_P_COPY(0), pattern,
1774 : &re_flags, 0,
1775 : PG_GET_COLLATION(),
1776 : false, true, true);
1777 :
1778 40 : MemoryContextSwitchTo(oldcontext);
1779 40 : funcctx->user_fctx = splitctx;
1780 : }
1781 :
1782 610 : funcctx = SRF_PERCALL_SETUP();
1783 610 : splitctx = (regexp_matches_ctx *) funcctx->user_fctx;
1784 :
1785 610 : if (splitctx->next_match <= splitctx->nmatches)
1786 : {
1787 570 : Datum result = build_regexp_split_result(splitctx);
1788 :
1789 570 : splitctx->next_match++;
1790 570 : SRF_RETURN_NEXT(funcctx, result);
1791 : }
1792 :
1793 40 : SRF_RETURN_DONE(funcctx);
1794 : }
1795 :
1796 : /* This is separate to keep the opr_sanity regression test from complaining */
1797 : Datum
1798 552 : regexp_split_to_table_no_flags(PG_FUNCTION_ARGS)
1799 : {
1800 552 : return regexp_split_to_table(fcinfo);
1801 : }
1802 :
1803 : /*
1804 : * regexp_split_to_array()
1805 : * Split the string at matches of the pattern, returning the
1806 : * split-out substrings as an array.
1807 : */
1808 : Datum
1809 200426 : regexp_split_to_array(PG_FUNCTION_ARGS)
1810 : {
1811 200426 : ArrayBuildState *astate = NULL;
1812 : pg_re_flags re_flags;
1813 : regexp_matches_ctx *splitctx;
1814 :
1815 : /* Determine options */
1816 200426 : parse_re_flags(&re_flags, PG_GETARG_TEXT_PP_IF_EXISTS(2));
1817 : /* User mustn't specify 'g' */
1818 200420 : if (re_flags.glob)
1819 6 : ereport(ERROR,
1820 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
1821 : /* translator: %s is a SQL function name */
1822 : errmsg("%s does not support the \"global\" option",
1823 : "regexp_split_to_array()")));
1824 : /* But we find all the matches anyway */
1825 200414 : re_flags.glob = true;
1826 :
1827 200414 : splitctx = setup_regexp_matches(PG_GETARG_TEXT_PP(0),
1828 200414 : PG_GETARG_TEXT_PP(1),
1829 : &re_flags, 0,
1830 : PG_GET_COLLATION(),
1831 : false, true, true);
1832 :
1833 1291380 : while (splitctx->next_match <= splitctx->nmatches)
1834 : {
1835 1090966 : astate = accumArrayResult(astate,
1836 : build_regexp_split_result(splitctx),
1837 : false,
1838 : TEXTOID,
1839 : CurrentMemoryContext);
1840 1090966 : splitctx->next_match++;
1841 : }
1842 :
1843 200414 : PG_RETURN_DATUM(makeArrayResult(astate, CurrentMemoryContext));
1844 : }
1845 :
1846 : /* This is separate to keep the opr_sanity regression test from complaining */
1847 : Datum
1848 200384 : regexp_split_to_array_no_flags(PG_FUNCTION_ARGS)
1849 : {
1850 200384 : return regexp_split_to_array(fcinfo);
1851 : }
1852 :
1853 : /*
1854 : * build_regexp_split_result - build output string for current match
1855 : *
1856 : * We return the string between the current match and the previous one,
1857 : * or the string after the last match when next_match == nmatches.
1858 : */
1859 : static Datum
1860 1091536 : build_regexp_split_result(regexp_matches_ctx *splitctx)
1861 : {
1862 1091536 : char *buf = splitctx->conv_buf;
1863 : int startpos;
1864 : int endpos;
1865 :
1866 1091536 : if (splitctx->next_match > 0)
1867 891082 : startpos = splitctx->match_locs[splitctx->next_match * 2 - 1];
1868 : else
1869 200454 : startpos = 0;
1870 1091536 : if (startpos < 0)
1871 0 : elog(ERROR, "invalid match ending position");
1872 :
1873 1091536 : endpos = splitctx->match_locs[splitctx->next_match * 2];
1874 1091536 : if (endpos < startpos)
1875 0 : elog(ERROR, "invalid match starting position");
1876 :
1877 1091536 : if (buf)
1878 : {
1879 : int len;
1880 :
1881 1091536 : len = pg_wchar2mb_with_len(splitctx->wide_str + startpos,
1882 : buf,
1883 : endpos - startpos);
1884 : Assert(len < splitctx->conv_bufsiz);
1885 1091536 : return PointerGetDatum(cstring_to_text_with_len(buf, len));
1886 : }
1887 : else
1888 : {
1889 0 : return DirectFunctionCall3(text_substr,
1890 : PointerGetDatum(splitctx->orig_str),
1891 : Int32GetDatum(startpos + 1),
1892 : Int32GetDatum(endpos - startpos));
1893 : }
1894 : }
1895 :
1896 : /*
1897 : * regexp_substr()
1898 : * Return the substring that matches a regular expression pattern
1899 : */
1900 : Datum
1901 108 : regexp_substr(PG_FUNCTION_ARGS)
1902 : {
1903 108 : text *str = PG_GETARG_TEXT_PP(0);
1904 108 : text *pattern = PG_GETARG_TEXT_PP(1);
1905 108 : int start = 1;
1906 108 : int n = 1;
1907 108 : text *flags = PG_GETARG_TEXT_PP_IF_EXISTS(4);
1908 108 : int subexpr = 0;
1909 : int so,
1910 : eo,
1911 : pos;
1912 : pg_re_flags re_flags;
1913 : regexp_matches_ctx *matchctx;
1914 :
1915 : /* Collect optional parameters */
1916 108 : if (PG_NARGS() > 2)
1917 : {
1918 90 : start = PG_GETARG_INT32(2);
1919 90 : if (start <= 0)
1920 6 : ereport(ERROR,
1921 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
1922 : errmsg("invalid value for parameter \"%s\": %d",
1923 : "start", start)));
1924 : }
1925 102 : if (PG_NARGS() > 3)
1926 : {
1927 78 : n = PG_GETARG_INT32(3);
1928 78 : if (n <= 0)
1929 6 : ereport(ERROR,
1930 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
1931 : errmsg("invalid value for parameter \"%s\": %d",
1932 : "n", n)));
1933 : }
1934 96 : if (PG_NARGS() > 5)
1935 : {
1936 48 : subexpr = PG_GETARG_INT32(5);
1937 48 : if (subexpr < 0)
1938 6 : ereport(ERROR,
1939 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
1940 : errmsg("invalid value for parameter \"%s\": %d",
1941 : "subexpr", subexpr)));
1942 : }
1943 :
1944 : /* Determine options */
1945 90 : parse_re_flags(&re_flags, flags);
1946 : /* User mustn't specify 'g' */
1947 90 : if (re_flags.glob)
1948 6 : ereport(ERROR,
1949 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
1950 : /* translator: %s is a SQL function name */
1951 : errmsg("%s does not support the \"global\" option",
1952 : "regexp_substr()")));
1953 : /* But we find all the matches anyway */
1954 84 : re_flags.glob = true;
1955 :
1956 : /* Do the matching */
1957 84 : matchctx = setup_regexp_matches(str, pattern, &re_flags, start - 1,
1958 : PG_GET_COLLATION(),
1959 : (subexpr > 0), /* need submatches? */
1960 : false, false);
1961 :
1962 : /* When n exceeds matches return NULL (includes case of no matches) */
1963 84 : if (n > matchctx->nmatches)
1964 12 : PG_RETURN_NULL();
1965 :
1966 : /* When subexpr exceeds number of subexpressions return NULL */
1967 72 : if (subexpr > matchctx->npatterns)
1968 6 : PG_RETURN_NULL();
1969 :
1970 : /* Select the appropriate match position to return */
1971 66 : pos = (n - 1) * matchctx->npatterns;
1972 66 : if (subexpr > 0)
1973 30 : pos += subexpr - 1;
1974 66 : pos *= 2;
1975 66 : so = matchctx->match_locs[pos];
1976 66 : eo = matchctx->match_locs[pos + 1];
1977 :
1978 66 : if (so < 0 || eo < 0)
1979 6 : PG_RETURN_NULL(); /* unidentifiable location */
1980 :
1981 60 : PG_RETURN_DATUM(DirectFunctionCall3(text_substr,
1982 : PointerGetDatum(matchctx->orig_str),
1983 : Int32GetDatum(so + 1),
1984 : Int32GetDatum(eo - so)));
1985 : }
1986 :
1987 : /* This is separate to keep the opr_sanity regression test from complaining */
1988 : Datum
1989 18 : regexp_substr_no_start(PG_FUNCTION_ARGS)
1990 : {
1991 18 : return regexp_substr(fcinfo);
1992 : }
1993 :
1994 : /* This is separate to keep the opr_sanity regression test from complaining */
1995 : Datum
1996 6 : regexp_substr_no_n(PG_FUNCTION_ARGS)
1997 : {
1998 6 : return regexp_substr(fcinfo);
1999 : }
2000 :
2001 : /* This is separate to keep the opr_sanity regression test from complaining */
2002 : Datum
2003 24 : regexp_substr_no_flags(PG_FUNCTION_ARGS)
2004 : {
2005 24 : return regexp_substr(fcinfo);
2006 : }
2007 :
2008 : /* This is separate to keep the opr_sanity regression test from complaining */
2009 : Datum
2010 12 : regexp_substr_no_subexpr(PG_FUNCTION_ARGS)
2011 : {
2012 12 : return regexp_substr(fcinfo);
2013 : }
2014 :
2015 : /*
2016 : * regexp_fixed_prefix - extract fixed prefix, if any, for a regexp
2017 : *
2018 : * The result is NULL if there is no fixed prefix, else a palloc'd string.
2019 : * If it is an exact match, not just a prefix, *exact is returned as true.
2020 : */
2021 : char *
2022 16492 : regexp_fixed_prefix(text *text_re, bool case_insensitive, Oid collation,
2023 : bool *exact)
2024 : {
2025 : char *result;
2026 : regex_t *re;
2027 : int cflags;
2028 : int re_result;
2029 : pg_wchar *str;
2030 : size_t slen;
2031 : size_t maxlen;
2032 : char errMsg[100];
2033 :
2034 16492 : *exact = false; /* default result */
2035 :
2036 : /* Compile RE */
2037 16492 : cflags = REG_ADVANCED;
2038 16492 : if (case_insensitive)
2039 62 : cflags |= REG_ICASE;
2040 :
2041 16492 : re = RE_compile_and_cache(text_re, cflags | REG_NOSUB, collation);
2042 :
2043 : /* Examine it to see if there's a fixed prefix */
2044 16468 : re_result = pg_regprefix(re, &str, &slen);
2045 :
2046 16468 : switch (re_result)
2047 : {
2048 762 : case REG_NOMATCH:
2049 762 : return NULL;
2050 :
2051 2948 : case REG_PREFIX:
2052 : /* continue with wchar conversion */
2053 2948 : break;
2054 :
2055 12758 : case REG_EXACT:
2056 12758 : *exact = true;
2057 : /* continue with wchar conversion */
2058 12758 : break;
2059 :
2060 0 : default:
2061 : /* re failed??? */
2062 0 : pg_regerror(re_result, re, errMsg, sizeof(errMsg));
2063 0 : ereport(ERROR,
2064 : (errcode(ERRCODE_INVALID_REGULAR_EXPRESSION),
2065 : errmsg("regular expression failed: %s", errMsg)));
2066 : break;
2067 : }
2068 :
2069 : /* Convert pg_wchar result back to database encoding */
2070 15706 : maxlen = pg_database_encoding_max_length() * slen + 1;
2071 15706 : result = (char *) palloc(maxlen);
2072 15706 : slen = pg_wchar2mb_with_len(str, result, slen);
2073 : Assert(slen < maxlen);
2074 :
2075 15706 : pfree(str);
2076 :
2077 15706 : return result;
2078 : }
|