Line data Source code
1 : /*-------------------------------------------------------------------------
2 : *
3 : * regexp.c
4 : * Postgres' interface to the regular expression package.
5 : *
6 : * Portions Copyright (c) 1996-2024, 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 1470286 : RE_compile_and_cache(text *text_re, int cflags, Oid collation)
142 : {
143 1470286 : int text_re_len = VARSIZE_ANY_EXHDR(text_re);
144 1470286 : 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 1995098 : for (i = 0; i < num_res; i++)
159 : {
160 1989538 : if (re_array[i].cre_pat_len == text_re_len &&
161 1478590 : re_array[i].cre_flags == cflags &&
162 1477386 : re_array[i].cre_collation == collation &&
163 1477190 : 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 1464726 : if (i > 0)
169 : {
170 430142 : re_temp = re_array[i];
171 430142 : memmove(&re_array[1], &re_array[0], i * sizeof(cached_re_str));
172 430142 : re_array[0] = re_temp;
173 : }
174 :
175 1464726 : return &re_array[0].cre_re;
176 : }
177 : }
178 :
179 : /* Set up the cache memory on first go through. */
180 5560 : if (unlikely(RegexpCacheMemoryContext == NULL))
181 1320 : RegexpCacheMemoryContext =
182 1320 : 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 5560 : pattern = (pg_wchar *) palloc((text_re_len + 1) * sizeof(pg_wchar));
193 5560 : 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 5560 : re_temp.cre_context = AllocSetContextCreate(CurrentMemoryContext,
205 : "RegexpMemoryContext",
206 : ALLOCSET_SMALL_SIZES);
207 5560 : oldcontext = MemoryContextSwitchTo(re_temp.cre_context);
208 :
209 5560 : regcomp_result = pg_regcomp(&re_temp.cre_re,
210 : pattern,
211 : pattern_len,
212 : cflags,
213 : collation);
214 :
215 5536 : pfree(pattern);
216 :
217 5536 : 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 5500 : re_temp.cre_pat = palloc(text_re_len + 1);
228 5500 : 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 5500 : re_temp.cre_pat[text_re_len] = 0;
235 5500 : MemoryContextSetIdentifier(re_temp.cre_context, re_temp.cre_pat);
236 :
237 5500 : re_temp.cre_pat_len = text_re_len;
238 5500 : re_temp.cre_flags = cflags;
239 5500 : 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 5500 : if (num_res >= MAX_CACHED_RES)
246 : {
247 792 : --num_res;
248 : Assert(num_res < MAX_CACHED_RES);
249 : /* Delete the memory context holding the regexp and pattern. */
250 792 : MemoryContextDelete(re_array[num_res].cre_context);
251 : }
252 :
253 : /* Re-parent the memory context to our long-lived cache context. */
254 5500 : MemoryContextSetParent(re_temp.cre_context, RegexpCacheMemoryContext);
255 :
256 5500 : if (num_res > 0)
257 4180 : memmove(&re_array[1], &re_array[0], num_res * sizeof(cached_re_str));
258 :
259 5500 : re_array[0] = re_temp;
260 5500 : num_res++;
261 :
262 5500 : MemoryContextSwitchTo(oldcontext);
263 :
264 5500 : 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 2334392 : 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 2334392 : 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 2334392 : 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 2334392 : 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 1239074 : 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 1239074 : data = (pg_wchar *) palloc((dat_len + 1) * sizeof(pg_wchar));
333 1239074 : data_len = pg_mb2wchar_with_len(dat, data, dat_len);
334 :
335 : /* Perform RE match and return result */
336 1239074 : match = RE_wchar_execute(re, data, data_len, 0, nmatch, pmatch);
337 :
338 1239074 : pfree(data);
339 1239074 : 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 1237470 : 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 1237470 : if (nmatch < 2)
366 1237470 : cflags |= REG_NOSUB;
367 :
368 : /* Compile RE */
369 1237470 : re = RE_compile_and_cache(text_re, cflags, collation);
370 :
371 1237446 : 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 206486 : parse_re_flags(pg_re_flags *flags, text *opts)
386 : {
387 : /* regex flavor is always folded into the compile flags */
388 206486 : flags->cflags = REG_ADVANCED;
389 206486 : flags->glob = false;
390 :
391 206486 : if (opts)
392 : {
393 3230 : char *opt_p = VARDATA_ANY(opts);
394 3230 : int opt_len = VARSIZE_ANY_EXHDR(opts);
395 : int i;
396 :
397 7424 : for (i = 0; i < opt_len; i++)
398 : {
399 4218 : switch (opt_p[i])
400 : {
401 2908 : case 'g':
402 2908 : flags->glob = true;
403 2908 : 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;
417 966 : case 'm': /* Perloid synonym for n */
418 : case 'n': /* \n affects ^ $ . [^ */
419 966 : flags->cflags |= REG_NEWLINE;
420 966 : 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 206462 : }
452 :
453 :
454 : /*
455 : * interface routines called by the function manager
456 : */
457 :
458 : Datum
459 775956 : nameregexeq(PG_FUNCTION_ARGS)
460 : {
461 775956 : Name n = PG_GETARG_NAME(0);
462 775956 : text *p = PG_GETARG_TEXT_PP(1);
463 :
464 775956 : 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 22372 : nameregexne(PG_FUNCTION_ARGS)
474 : {
475 22372 : Name n = PG_GETARG_NAME(0);
476 22372 : text *p = PG_GETARG_TEXT_PP(1);
477 :
478 22372 : 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 397328 : textregexeq(PG_FUNCTION_ARGS)
488 : {
489 397328 : text *s = PG_GETARG_TEXT_PP(0);
490 397328 : text *p = PG_GETARG_TEXT_PP(1);
491 :
492 397328 : 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 34134 : textregexne(PG_FUNCTION_ARGS)
502 : {
503 34134 : text *s = PG_GETARG_TEXT_PP(0);
504 34134 : text *p = PG_GETARG_TEXT_PP(1);
505 :
506 34134 : 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 7156 : nameicregexeq(PG_FUNCTION_ARGS)
523 : {
524 7156 : Name n = PG_GETARG_NAME(0);
525 7156 : text *p = PG_GETARG_TEXT_PP(1);
526 :
527 7156 : 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
550 196 : texticregexeq(PG_FUNCTION_ARGS)
551 : {
552 196 : text *s = PG_GETARG_TEXT_PP(0);
553 196 : text *p = PG_GETARG_TEXT_PP(1);
554 :
555 196 : 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 22 : texticregexne(PG_FUNCTION_ARGS)
565 : {
566 22 : text *s = PG_GETARG_TEXT_PP(0);
567 22 : text *p = PG_GETARG_TEXT_PP(1);
568 :
569 22 : 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 1628 : textregexsubstr(PG_FUNCTION_ARGS)
584 : {
585 1628 : text *s = PG_GETARG_TEXT_PP(0);
586 1628 : 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 1628 : 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 3256 : if (!RE_execute(re,
602 3256 : VARDATA_ANY(s), VARSIZE_ANY_EXHDR(s),
603 : 2, pmatch))
604 6 : PG_RETURN_NULL(); /* definitely no match */
605 :
606 1622 : 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 */
615 100 : so = pmatch[0].rm_so;
616 100 : 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 1622 : if (so < 0 || eo < 0)
626 6 : PG_RETURN_NULL();
627 :
628 1616 : 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 9942 : textregexreplace_noopt(PG_FUNCTION_ARGS)
643 : {
644 9942 : text *s = PG_GETARG_TEXT_PP(0);
645 9942 : text *p = PG_GETARG_TEXT_PP(1);
646 9942 : text *r = PG_GETARG_TEXT_PP(2);
647 :
648 9942 : 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 2836 : textregexreplace(PG_FUNCTION_ARGS)
659 : {
660 2836 : text *s = PG_GETARG_TEXT_PP(0);
661 2836 : text *p = PG_GETARG_TEXT_PP(1);
662 2836 : text *r = PG_GETARG_TEXT_PP(2);
663 2836 : 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 2836 : if (VARSIZE_ANY_EXHDR(opt) > 0)
674 : {
675 2836 : char *opt_p = VARDATA_ANY(opt);
676 :
677 2836 : 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 2830 : parse_re_flags(&flags, opt);
686 :
687 2824 : 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 *
767 138 : 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;
775 138 : bool afterescape = false;
776 138 : bool incharclass = false;
777 138 : int nquotes = 0;
778 :
779 138 : p = VARDATA_ANY(pat_text);
780 138 : plen = VARSIZE_ANY_EXHDR(pat_text);
781 138 : if (esc_text == NULL)
782 : {
783 : /* No ESCAPE clause provided; default to backslash as escape */
784 40 : e = "\\";
785 40 : elen = 1;
786 : }
787 : else
788 : {
789 98 : e = VARDATA_ANY(esc_text);
790 98 : elen = VARSIZE_ANY_EXHDR(esc_text);
791 98 : if (elen == 0)
792 6 : e = NULL; /* no escape character */
793 92 : else if (elen > 1)
794 : {
795 6 : int escape_mblen = pg_mbstrlen_with_len(e, elen);
796 :
797 6 : if (escape_mblen > 1)
798 6 : ereport(ERROR,
799 : (errcode(ERRCODE_INVALID_ESCAPE_SEQUENCE),
800 : errmsg("invalid escape string"),
801 : errhint("Escape string must be empty or one character.")));
802 : }
803 : }
804 :
805 : /*----------
806 : * We surround the transformed input string with
807 : * ^(?: ... )$
808 : * which requires some explanation. We need "^" and "$" to force
809 : * the pattern to match the entire input string as per the SQL spec.
810 : * The "(?:" and ")" are a non-capturing set of parens; we have to have
811 : * parens in case the string contains "|", else the "^" and "$" will
812 : * be bound into the first and last alternatives which is not what we
813 : * want, and the parens must be non capturing because we don't want them
814 : * to count when selecting output for SUBSTRING.
815 : *
816 : * When the pattern is divided into three parts by escape-double-quotes,
817 : * what we emit is
818 : * ^(?:part1){1,1}?(part2){1,1}(?:part3)$
819 : * which requires even more explanation. The "{1,1}?" on part1 makes it
820 : * non-greedy so that it will match the smallest possible amount of text
821 : * not the largest, as required by SQL. The plain parens around part2
822 : * are capturing parens so that that part is what controls the result of
823 : * SUBSTRING. The "{1,1}" forces part2 to be greedy, so that it matches
824 : * the largest possible amount of text; hence part3 must match the
825 : * smallest amount of text, as required by SQL. We don't need an explicit
826 : * greediness marker on part3. Note that this also confines the effects
827 : * of any "|" characters to the respective part, which is what we want.
828 : *
829 : * The SQL spec says that SUBSTRING's pattern must contain exactly two
830 : * escape-double-quotes, but we only complain if there's more than two.
831 : * With none, we act as though part1 and part3 are empty; with one, we
832 : * act as though part3 is empty. Both behaviors fall out of omitting
833 : * the relevant part separators in the above expansion. If the result
834 : * of this function is used in a plain regexp match (SIMILAR TO), the
835 : * escape-double-quotes have no effect on the match behavior.
836 : *----------
837 : */
838 :
839 : /*
840 : * We need room for the prefix/postfix and part separators, plus as many
841 : * as 3 output bytes per input byte; since the input is at most 1GB this
842 : * can't overflow size_t.
843 : */
844 132 : result = (text *) palloc(VARHDRSZ + 23 + 3 * (size_t) plen);
845 132 : r = VARDATA(result);
846 :
847 132 : *r++ = '^';
848 132 : *r++ = '(';
849 132 : *r++ = '?';
850 132 : *r++ = ':';
851 :
852 1096 : while (plen > 0)
853 : {
854 970 : char pchar = *p;
855 :
856 : /*
857 : * If both the escape character and the current character from the
858 : * pattern are multi-byte, we need to take the slow path.
859 : *
860 : * But if one of them is single-byte, we can process the pattern one
861 : * byte at a time, ignoring multi-byte characters. (This works
862 : * because all server-encodings have the property that a valid
863 : * multi-byte character representation cannot contain the
864 : * representation of a valid single-byte character.)
865 : */
866 :
867 970 : if (elen > 1)
868 : {
869 0 : int mblen = pg_mblen(p);
870 :
871 0 : if (mblen > 1)
872 : {
873 : /* slow, multi-byte path */
874 0 : if (afterescape)
875 : {
876 0 : *r++ = '\\';
877 0 : memcpy(r, p, mblen);
878 0 : r += mblen;
879 0 : afterescape = false;
880 : }
881 0 : else if (e && elen == mblen && memcmp(e, p, mblen) == 0)
882 : {
883 : /* SQL escape character; do not send to output */
884 0 : afterescape = true;
885 : }
886 : else
887 : {
888 : /*
889 : * We know it's a multi-byte character, so we don't need
890 : * to do all the comparisons to single-byte characters
891 : * that we do below.
892 : */
893 0 : memcpy(r, p, mblen);
894 0 : r += mblen;
895 : }
896 :
897 0 : p += mblen;
898 0 : plen -= mblen;
899 :
900 0 : continue;
901 : }
902 : }
903 :
904 : /* fast path */
905 970 : if (afterescape)
906 : {
907 154 : if (pchar == '"' && !incharclass) /* escape-double-quote? */
908 : {
909 : /* emit appropriate part separator, per notes above */
910 136 : if (nquotes == 0)
911 : {
912 68 : *r++ = ')';
913 68 : *r++ = '{';
914 68 : *r++ = '1';
915 68 : *r++ = ',';
916 68 : *r++ = '1';
917 68 : *r++ = '}';
918 68 : *r++ = '?';
919 68 : *r++ = '(';
920 : }
921 68 : else if (nquotes == 1)
922 : {
923 62 : *r++ = ')';
924 62 : *r++ = '{';
925 62 : *r++ = '1';
926 62 : *r++ = ',';
927 62 : *r++ = '1';
928 62 : *r++ = '}';
929 62 : *r++ = '(';
930 62 : *r++ = '?';
931 62 : *r++ = ':';
932 : }
933 : else
934 6 : ereport(ERROR,
935 : (errcode(ERRCODE_INVALID_USE_OF_ESCAPE_CHARACTER),
936 : errmsg("SQL regular expression may not contain more than two escape-double-quote separators")));
937 130 : nquotes++;
938 : }
939 : else
940 : {
941 : /*
942 : * We allow any character at all to be escaped; notably, this
943 : * allows access to POSIX character-class escapes such as
944 : * "\d". The SQL spec is considerably more restrictive.
945 : */
946 18 : *r++ = '\\';
947 18 : *r++ = pchar;
948 : }
949 148 : afterescape = false;
950 : }
951 816 : else if (e && pchar == *e)
952 : {
953 : /* SQL escape character; do not send to output */
954 154 : afterescape = true;
955 : }
956 662 : else if (incharclass)
957 : {
958 36 : if (pchar == '\\')
959 0 : *r++ = '\\';
960 36 : *r++ = pchar;
961 36 : if (pchar == ']')
962 6 : incharclass = false;
963 : }
964 626 : else if (pchar == '[')
965 : {
966 6 : *r++ = pchar;
967 6 : incharclass = true;
968 : }
969 620 : else if (pchar == '%')
970 : {
971 114 : *r++ = '.';
972 114 : *r++ = '*';
973 : }
974 506 : else if (pchar == '_')
975 52 : *r++ = '.';
976 454 : else if (pchar == '(')
977 : {
978 : /* convert to non-capturing parenthesis */
979 18 : *r++ = '(';
980 18 : *r++ = '?';
981 18 : *r++ = ':';
982 : }
983 436 : else if (pchar == '\\' || pchar == '.' ||
984 426 : pchar == '^' || pchar == '$')
985 : {
986 10 : *r++ = '\\';
987 10 : *r++ = pchar;
988 : }
989 : else
990 426 : *r++ = pchar;
991 964 : p++, plen--;
992 : }
993 :
994 126 : *r++ = ')';
995 126 : *r++ = '$';
996 :
997 126 : SET_VARSIZE(result, r - ((char *) result));
998 :
999 126 : return result;
1000 : }
1001 :
1002 : /*
1003 : * similar_to_escape(pattern, escape)
1004 : */
1005 : Datum
1006 98 : similar_to_escape_2(PG_FUNCTION_ARGS)
1007 : {
1008 98 : text *pat_text = PG_GETARG_TEXT_PP(0);
1009 98 : text *esc_text = PG_GETARG_TEXT_PP(1);
1010 : text *result;
1011 :
1012 98 : result = similar_escape_internal(pat_text, esc_text);
1013 :
1014 86 : PG_RETURN_TEXT_P(result);
1015 : }
1016 :
1017 : /*
1018 : * similar_to_escape(pattern)
1019 : * Inserts a default escape character.
1020 : */
1021 : Datum
1022 40 : similar_to_escape_1(PG_FUNCTION_ARGS)
1023 : {
1024 40 : text *pat_text = PG_GETARG_TEXT_PP(0);
1025 : text *result;
1026 :
1027 40 : result = similar_escape_internal(pat_text, NULL);
1028 :
1029 40 : PG_RETURN_TEXT_P(result);
1030 : }
1031 :
1032 : /*
1033 : * similar_escape(pattern, escape)
1034 : *
1035 : * Legacy function for compatibility with views stored using the
1036 : * pre-v13 expansion of SIMILAR TO. Unlike the above functions, this
1037 : * is non-strict, which leads to not-per-spec handling of "ESCAPE NULL".
1038 : */
1039 : Datum
1040 0 : similar_escape(PG_FUNCTION_ARGS)
1041 : {
1042 : text *pat_text;
1043 : text *esc_text;
1044 : text *result;
1045 :
1046 : /* This function is not strict, so must test explicitly */
1047 0 : if (PG_ARGISNULL(0))
1048 0 : PG_RETURN_NULL();
1049 0 : pat_text = PG_GETARG_TEXT_PP(0);
1050 :
1051 0 : if (PG_ARGISNULL(1))
1052 0 : esc_text = NULL; /* use default escape character */
1053 : else
1054 0 : esc_text = PG_GETARG_TEXT_PP(1);
1055 :
1056 0 : result = similar_escape_internal(pat_text, esc_text);
1057 :
1058 0 : PG_RETURN_TEXT_P(result);
1059 : }
1060 :
1061 : /*
1062 : * regexp_count()
1063 : * Return the number of matches of a pattern within a string.
1064 : */
1065 : Datum
1066 48 : regexp_count(PG_FUNCTION_ARGS)
1067 : {
1068 48 : text *str = PG_GETARG_TEXT_PP(0);
1069 48 : text *pattern = PG_GETARG_TEXT_PP(1);
1070 48 : int start = 1;
1071 48 : text *flags = PG_GETARG_TEXT_PP_IF_EXISTS(3);
1072 : pg_re_flags re_flags;
1073 : regexp_matches_ctx *matchctx;
1074 :
1075 : /* Collect optional parameters */
1076 48 : if (PG_NARGS() > 2)
1077 : {
1078 42 : start = PG_GETARG_INT32(2);
1079 42 : if (start <= 0)
1080 12 : ereport(ERROR,
1081 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
1082 : errmsg("invalid value for parameter \"%s\": %d",
1083 : "start", start)));
1084 : }
1085 :
1086 : /* Determine options */
1087 36 : parse_re_flags(&re_flags, flags);
1088 : /* User mustn't specify 'g' */
1089 36 : if (re_flags.glob)
1090 0 : ereport(ERROR,
1091 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
1092 : /* translator: %s is a SQL function name */
1093 : errmsg("%s does not support the \"global\" option",
1094 : "regexp_count()")));
1095 : /* But we find all the matches anyway */
1096 36 : re_flags.glob = true;
1097 :
1098 : /* Do the matching */
1099 36 : matchctx = setup_regexp_matches(str, pattern, &re_flags, start - 1,
1100 : PG_GET_COLLATION(),
1101 : false, /* can ignore subexprs */
1102 : false, false);
1103 :
1104 36 : PG_RETURN_INT32(matchctx->nmatches);
1105 : }
1106 :
1107 : /* This is separate to keep the opr_sanity regression test from complaining */
1108 : Datum
1109 6 : regexp_count_no_start(PG_FUNCTION_ARGS)
1110 : {
1111 6 : return regexp_count(fcinfo);
1112 : }
1113 :
1114 : /* This is separate to keep the opr_sanity regression test from complaining */
1115 : Datum
1116 30 : regexp_count_no_flags(PG_FUNCTION_ARGS)
1117 : {
1118 30 : return regexp_count(fcinfo);
1119 : }
1120 :
1121 : /*
1122 : * regexp_instr()
1123 : * Return the match's position within the string
1124 : */
1125 : Datum
1126 156 : regexp_instr(PG_FUNCTION_ARGS)
1127 : {
1128 156 : text *str = PG_GETARG_TEXT_PP(0);
1129 156 : text *pattern = PG_GETARG_TEXT_PP(1);
1130 156 : int start = 1;
1131 156 : int n = 1;
1132 156 : int endoption = 0;
1133 156 : text *flags = PG_GETARG_TEXT_PP_IF_EXISTS(5);
1134 156 : int subexpr = 0;
1135 : int pos;
1136 : pg_re_flags re_flags;
1137 : regexp_matches_ctx *matchctx;
1138 :
1139 : /* Collect optional parameters */
1140 156 : if (PG_NARGS() > 2)
1141 : {
1142 138 : start = PG_GETARG_INT32(2);
1143 138 : if (start <= 0)
1144 6 : ereport(ERROR,
1145 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
1146 : errmsg("invalid value for parameter \"%s\": %d",
1147 : "start", start)));
1148 : }
1149 150 : if (PG_NARGS() > 3)
1150 : {
1151 126 : n = PG_GETARG_INT32(3);
1152 126 : if (n <= 0)
1153 6 : ereport(ERROR,
1154 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
1155 : errmsg("invalid value for parameter \"%s\": %d",
1156 : "n", n)));
1157 : }
1158 144 : if (PG_NARGS() > 4)
1159 : {
1160 108 : endoption = PG_GETARG_INT32(4);
1161 108 : if (endoption != 0 && endoption != 1)
1162 12 : ereport(ERROR,
1163 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
1164 : errmsg("invalid value for parameter \"%s\": %d",
1165 : "endoption", endoption)));
1166 : }
1167 132 : if (PG_NARGS() > 6)
1168 : {
1169 84 : subexpr = PG_GETARG_INT32(6);
1170 84 : if (subexpr < 0)
1171 6 : ereport(ERROR,
1172 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
1173 : errmsg("invalid value for parameter \"%s\": %d",
1174 : "subexpr", subexpr)));
1175 : }
1176 :
1177 : /* Determine options */
1178 126 : parse_re_flags(&re_flags, flags);
1179 : /* User mustn't specify 'g' */
1180 126 : if (re_flags.glob)
1181 6 : ereport(ERROR,
1182 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
1183 : /* translator: %s is a SQL function name */
1184 : errmsg("%s does not support the \"global\" option",
1185 : "regexp_instr()")));
1186 : /* But we find all the matches anyway */
1187 120 : re_flags.glob = true;
1188 :
1189 : /* Do the matching */
1190 120 : matchctx = setup_regexp_matches(str, pattern, &re_flags, start - 1,
1191 : PG_GET_COLLATION(),
1192 : (subexpr > 0), /* need submatches? */
1193 : false, false);
1194 :
1195 : /* When n exceeds matches return 0 (includes case of no matches) */
1196 120 : if (n > matchctx->nmatches)
1197 12 : PG_RETURN_INT32(0);
1198 :
1199 : /* When subexpr exceeds number of subexpressions return 0 */
1200 108 : if (subexpr > matchctx->npatterns)
1201 12 : PG_RETURN_INT32(0);
1202 :
1203 : /* Select the appropriate match position to return */
1204 96 : pos = (n - 1) * matchctx->npatterns;
1205 96 : if (subexpr > 0)
1206 54 : pos += subexpr - 1;
1207 96 : pos *= 2;
1208 96 : if (endoption == 1)
1209 30 : pos += 1;
1210 :
1211 96 : if (matchctx->match_locs[pos] >= 0)
1212 90 : PG_RETURN_INT32(matchctx->match_locs[pos] + 1);
1213 : else
1214 6 : PG_RETURN_INT32(0); /* position not identifiable */
1215 : }
1216 :
1217 : /* This is separate to keep the opr_sanity regression test from complaining */
1218 : Datum
1219 18 : regexp_instr_no_start(PG_FUNCTION_ARGS)
1220 : {
1221 18 : return regexp_instr(fcinfo);
1222 : }
1223 :
1224 : /* This is separate to keep the opr_sanity regression test from complaining */
1225 : Datum
1226 6 : regexp_instr_no_n(PG_FUNCTION_ARGS)
1227 : {
1228 6 : return regexp_instr(fcinfo);
1229 : }
1230 :
1231 : /* This is separate to keep the opr_sanity regression test from complaining */
1232 : Datum
1233 24 : regexp_instr_no_endoption(PG_FUNCTION_ARGS)
1234 : {
1235 24 : return regexp_instr(fcinfo);
1236 : }
1237 :
1238 : /* This is separate to keep the opr_sanity regression test from complaining */
1239 : Datum
1240 12 : regexp_instr_no_flags(PG_FUNCTION_ARGS)
1241 : {
1242 12 : return regexp_instr(fcinfo);
1243 : }
1244 :
1245 : /* This is separate to keep the opr_sanity regression test from complaining */
1246 : Datum
1247 12 : regexp_instr_no_subexpr(PG_FUNCTION_ARGS)
1248 : {
1249 12 : return regexp_instr(fcinfo);
1250 : }
1251 :
1252 : /*
1253 : * regexp_like()
1254 : * Test for a pattern match within a string.
1255 : */
1256 : Datum
1257 30 : regexp_like(PG_FUNCTION_ARGS)
1258 : {
1259 30 : text *str = PG_GETARG_TEXT_PP(0);
1260 30 : text *pattern = PG_GETARG_TEXT_PP(1);
1261 30 : text *flags = PG_GETARG_TEXT_PP_IF_EXISTS(2);
1262 : pg_re_flags re_flags;
1263 :
1264 : /* Determine options */
1265 30 : parse_re_flags(&re_flags, flags);
1266 : /* User mustn't specify 'g' */
1267 30 : if (re_flags.glob)
1268 6 : ereport(ERROR,
1269 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
1270 : /* translator: %s is a SQL function name */
1271 : errmsg("%s does not support the \"global\" option",
1272 : "regexp_like()")));
1273 :
1274 : /* Otherwise it's like textregexeq/texticregexeq */
1275 24 : PG_RETURN_BOOL(RE_compile_and_execute(pattern,
1276 : VARDATA_ANY(str),
1277 : VARSIZE_ANY_EXHDR(str),
1278 : re_flags.cflags,
1279 : PG_GET_COLLATION(),
1280 : 0, NULL));
1281 : }
1282 :
1283 : /* This is separate to keep the opr_sanity regression test from complaining */
1284 : Datum
1285 6 : regexp_like_no_flags(PG_FUNCTION_ARGS)
1286 : {
1287 6 : return regexp_like(fcinfo);
1288 : }
1289 :
1290 : /*
1291 : * regexp_match()
1292 : * Return the first substring(s) matching a pattern within a string.
1293 : */
1294 : Datum
1295 2556 : regexp_match(PG_FUNCTION_ARGS)
1296 : {
1297 2556 : text *orig_str = PG_GETARG_TEXT_PP(0);
1298 2556 : text *pattern = PG_GETARG_TEXT_PP(1);
1299 2556 : text *flags = PG_GETARG_TEXT_PP_IF_EXISTS(2);
1300 : pg_re_flags re_flags;
1301 : regexp_matches_ctx *matchctx;
1302 :
1303 : /* Determine options */
1304 2556 : parse_re_flags(&re_flags, flags);
1305 : /* User mustn't specify 'g' */
1306 2556 : if (re_flags.glob)
1307 8 : ereport(ERROR,
1308 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
1309 : /* translator: %s is a SQL function name */
1310 : errmsg("%s does not support the \"global\" option",
1311 : "regexp_match()"),
1312 : errhint("Use the regexp_matches function instead.")));
1313 :
1314 2548 : matchctx = setup_regexp_matches(orig_str, pattern, &re_flags, 0,
1315 : PG_GET_COLLATION(), true, false, false);
1316 :
1317 2548 : if (matchctx->nmatches == 0)
1318 170 : PG_RETURN_NULL();
1319 :
1320 : Assert(matchctx->nmatches == 1);
1321 :
1322 : /* Create workspace that build_regexp_match_result needs */
1323 2378 : matchctx->elems = (Datum *) palloc(sizeof(Datum) * matchctx->npatterns);
1324 2378 : matchctx->nulls = (bool *) palloc(sizeof(bool) * matchctx->npatterns);
1325 :
1326 2378 : PG_RETURN_DATUM(PointerGetDatum(build_regexp_match_result(matchctx)));
1327 : }
1328 :
1329 : /* This is separate to keep the opr_sanity regression test from complaining */
1330 : Datum
1331 2526 : regexp_match_no_flags(PG_FUNCTION_ARGS)
1332 : {
1333 2526 : return regexp_match(fcinfo);
1334 : }
1335 :
1336 : /*
1337 : * regexp_matches()
1338 : * Return a table of all matches of a pattern within a string.
1339 : */
1340 : Datum
1341 678 : regexp_matches(PG_FUNCTION_ARGS)
1342 : {
1343 : FuncCallContext *funcctx;
1344 : regexp_matches_ctx *matchctx;
1345 :
1346 678 : if (SRF_IS_FIRSTCALL())
1347 : {
1348 288 : text *pattern = PG_GETARG_TEXT_PP(1);
1349 288 : text *flags = PG_GETARG_TEXT_PP_IF_EXISTS(2);
1350 : pg_re_flags re_flags;
1351 : MemoryContext oldcontext;
1352 :
1353 288 : funcctx = SRF_FIRSTCALL_INIT();
1354 288 : oldcontext = MemoryContextSwitchTo(funcctx->multi_call_memory_ctx);
1355 :
1356 : /* Determine options */
1357 288 : parse_re_flags(&re_flags, flags);
1358 :
1359 : /* be sure to copy the input string into the multi-call ctx */
1360 282 : matchctx = setup_regexp_matches(PG_GETARG_TEXT_P_COPY(0), pattern,
1361 : &re_flags, 0,
1362 : PG_GET_COLLATION(),
1363 : true, false, false);
1364 :
1365 : /* Pre-create workspace that build_regexp_match_result needs */
1366 270 : matchctx->elems = (Datum *) palloc(sizeof(Datum) * matchctx->npatterns);
1367 270 : matchctx->nulls = (bool *) palloc(sizeof(bool) * matchctx->npatterns);
1368 :
1369 270 : MemoryContextSwitchTo(oldcontext);
1370 270 : funcctx->user_fctx = (void *) matchctx;
1371 : }
1372 :
1373 660 : funcctx = SRF_PERCALL_SETUP();
1374 660 : matchctx = (regexp_matches_ctx *) funcctx->user_fctx;
1375 :
1376 660 : if (matchctx->next_match < matchctx->nmatches)
1377 : {
1378 : ArrayType *result_ary;
1379 :
1380 390 : result_ary = build_regexp_match_result(matchctx);
1381 390 : matchctx->next_match++;
1382 390 : SRF_RETURN_NEXT(funcctx, PointerGetDatum(result_ary));
1383 : }
1384 :
1385 270 : SRF_RETURN_DONE(funcctx);
1386 : }
1387 :
1388 : /* This is separate to keep the opr_sanity regression test from complaining */
1389 : Datum
1390 354 : regexp_matches_no_flags(PG_FUNCTION_ARGS)
1391 : {
1392 354 : return regexp_matches(fcinfo);
1393 : }
1394 :
1395 : /*
1396 : * setup_regexp_matches --- do the initial matching for regexp_match,
1397 : * regexp_split, and related functions
1398 : *
1399 : * To avoid having to re-find the compiled pattern on each call, we do
1400 : * all the matching in one swoop. The returned regexp_matches_ctx contains
1401 : * the locations of all the substrings matching the pattern.
1402 : *
1403 : * start_search: the character (not byte) offset in orig_str at which to
1404 : * begin the search. Returned positions are relative to orig_str anyway.
1405 : * use_subpatterns: collect data about matches to parenthesized subexpressions.
1406 : * ignore_degenerate: ignore zero-length matches.
1407 : * fetching_unmatched: caller wants to fetch unmatched substrings.
1408 : *
1409 : * We don't currently assume that fetching_unmatched is exclusive of fetching
1410 : * the matched text too; if it's set, the conversion buffer is large enough to
1411 : * fetch any single matched or unmatched string, but not any larger
1412 : * substring. (In practice, when splitting the matches are usually small
1413 : * anyway, and it didn't seem worth complicating the code further.)
1414 : */
1415 : static regexp_matches_ctx *
1416 203522 : setup_regexp_matches(text *orig_str, text *pattern, pg_re_flags *re_flags,
1417 : int start_search,
1418 : Oid collation,
1419 : bool use_subpatterns,
1420 : bool ignore_degenerate,
1421 : bool fetching_unmatched)
1422 : {
1423 203522 : regexp_matches_ctx *matchctx = palloc0(sizeof(regexp_matches_ctx));
1424 203522 : int eml = pg_database_encoding_max_length();
1425 : int orig_len;
1426 : pg_wchar *wide_str;
1427 : int wide_len;
1428 : int cflags;
1429 : regex_t *cpattern;
1430 : regmatch_t *pmatch;
1431 : int pmatch_len;
1432 : int array_len;
1433 : int array_idx;
1434 : int prev_match_end;
1435 : int prev_valid_match_end;
1436 203522 : int maxlen = 0; /* largest fetch length in characters */
1437 :
1438 : /* save original string --- we'll extract result substrings from it */
1439 203522 : matchctx->orig_str = orig_str;
1440 :
1441 : /* convert string to pg_wchar form for matching */
1442 203522 : orig_len = VARSIZE_ANY_EXHDR(orig_str);
1443 203522 : wide_str = (pg_wchar *) palloc(sizeof(pg_wchar) * (orig_len + 1));
1444 203522 : wide_len = pg_mb2wchar_with_len(VARDATA_ANY(orig_str), wide_str, orig_len);
1445 :
1446 : /* set up the compiled pattern */
1447 203522 : cflags = re_flags->cflags;
1448 203522 : if (!use_subpatterns)
1449 200590 : cflags |= REG_NOSUB;
1450 203522 : cpattern = RE_compile_and_cache(pattern, cflags, collation);
1451 :
1452 : /* do we want to remember subpatterns? */
1453 203510 : if (use_subpatterns && cpattern->re_nsub > 0)
1454 : {
1455 2710 : matchctx->npatterns = cpattern->re_nsub;
1456 2710 : pmatch_len = cpattern->re_nsub + 1;
1457 : }
1458 : else
1459 : {
1460 200800 : use_subpatterns = false;
1461 200800 : matchctx->npatterns = 1;
1462 200800 : pmatch_len = 1;
1463 : }
1464 :
1465 : /* temporary output space for RE package */
1466 203510 : pmatch = palloc(sizeof(regmatch_t) * pmatch_len);
1467 :
1468 : /*
1469 : * the real output space (grown dynamically if needed)
1470 : *
1471 : * use values 2^n-1, not 2^n, so that we hit the limit at 2^28-1 rather
1472 : * than at 2^27
1473 : */
1474 203510 : array_len = re_flags->glob ? 255 : 31;
1475 203510 : matchctx->match_locs = (int *) palloc(sizeof(int) * array_len);
1476 203510 : array_idx = 0;
1477 :
1478 : /* search for the pattern, perhaps repeatedly */
1479 203510 : prev_match_end = 0;
1480 203510 : prev_valid_match_end = 0;
1481 1095318 : while (RE_wchar_execute(cpattern, wide_str, wide_len, start_search,
1482 : pmatch_len, pmatch))
1483 : {
1484 : /*
1485 : * If requested, ignore degenerate matches, which are zero-length
1486 : * matches occurring at the start or end of a string or just after a
1487 : * previous match.
1488 : */
1489 894422 : if (!ignore_degenerate ||
1490 891258 : (pmatch[0].rm_so < wide_len &&
1491 891216 : pmatch[0].rm_eo > prev_match_end))
1492 : {
1493 : /* enlarge output space if needed */
1494 894602 : while (array_idx + matchctx->npatterns * 2 + 1 > array_len)
1495 : {
1496 360 : array_len += array_len + 1; /* 2^n-1 => 2^(n+1)-1 */
1497 360 : if (array_len > MaxAllocSize / sizeof(int))
1498 0 : ereport(ERROR,
1499 : (errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
1500 : errmsg("too many regular expression matches")));
1501 360 : matchctx->match_locs = (int *) repalloc(matchctx->match_locs,
1502 : sizeof(int) * array_len);
1503 : }
1504 :
1505 : /* save this match's locations */
1506 894242 : if (use_subpatterns)
1507 : {
1508 : int i;
1509 :
1510 7872 : for (i = 1; i <= matchctx->npatterns; i++)
1511 : {
1512 5314 : int so = pmatch[i].rm_so;
1513 5314 : int eo = pmatch[i].rm_eo;
1514 :
1515 5314 : matchctx->match_locs[array_idx++] = so;
1516 5314 : matchctx->match_locs[array_idx++] = eo;
1517 5314 : if (so >= 0 && eo >= 0 && (eo - so) > maxlen)
1518 3420 : maxlen = (eo - so);
1519 : }
1520 : }
1521 : else
1522 : {
1523 891684 : int so = pmatch[0].rm_so;
1524 891684 : int eo = pmatch[0].rm_eo;
1525 :
1526 891684 : matchctx->match_locs[array_idx++] = so;
1527 891684 : matchctx->match_locs[array_idx++] = eo;
1528 891684 : if (so >= 0 && eo >= 0 && (eo - so) > maxlen)
1529 200634 : maxlen = (eo - so);
1530 : }
1531 894242 : matchctx->nmatches++;
1532 :
1533 : /*
1534 : * check length of unmatched portion between end of previous valid
1535 : * (nondegenerate, or degenerate but not ignored) match and start
1536 : * of current one
1537 : */
1538 894242 : if (fetching_unmatched &&
1539 891078 : pmatch[0].rm_so >= 0 &&
1540 891078 : (pmatch[0].rm_so - prev_valid_match_end) > maxlen)
1541 380878 : maxlen = (pmatch[0].rm_so - prev_valid_match_end);
1542 894242 : prev_valid_match_end = pmatch[0].rm_eo;
1543 : }
1544 894422 : prev_match_end = pmatch[0].rm_eo;
1545 :
1546 : /* if not glob, stop after one match */
1547 894422 : if (!re_flags->glob)
1548 2548 : break;
1549 :
1550 : /*
1551 : * Advance search position. Normally we start the next search at the
1552 : * end of the previous match; but if the match was of zero length, we
1553 : * have to advance by one character, or we'd just find the same match
1554 : * again.
1555 : */
1556 891874 : start_search = prev_match_end;
1557 891874 : if (pmatch[0].rm_so == pmatch[0].rm_eo)
1558 1176 : start_search++;
1559 891874 : if (start_search > wide_len)
1560 66 : break;
1561 : }
1562 :
1563 : /*
1564 : * check length of unmatched portion between end of last match and end of
1565 : * input string
1566 : */
1567 203510 : if (fetching_unmatched &&
1568 200452 : (wide_len - prev_valid_match_end) > maxlen)
1569 76 : maxlen = (wide_len - prev_valid_match_end);
1570 :
1571 : /*
1572 : * Keep a note of the end position of the string for the benefit of
1573 : * splitting code.
1574 : */
1575 203510 : matchctx->match_locs[array_idx] = wide_len;
1576 :
1577 203510 : if (eml > 1)
1578 : {
1579 203510 : int64 maxsiz = eml * (int64) maxlen;
1580 : int conv_bufsiz;
1581 :
1582 : /*
1583 : * Make the conversion buffer large enough for any substring of
1584 : * interest.
1585 : *
1586 : * Worst case: assume we need the maximum size (maxlen*eml), but take
1587 : * advantage of the fact that the original string length in bytes is
1588 : * an upper bound on the byte length of any fetched substring (and we
1589 : * know that len+1 is safe to allocate because the varlena header is
1590 : * longer than 1 byte).
1591 : */
1592 203510 : if (maxsiz > orig_len)
1593 200548 : conv_bufsiz = orig_len + 1;
1594 : else
1595 2962 : conv_bufsiz = maxsiz + 1; /* safe since maxsiz < 2^30 */
1596 :
1597 203510 : matchctx->conv_buf = palloc(conv_bufsiz);
1598 203510 : matchctx->conv_bufsiz = conv_bufsiz;
1599 203510 : matchctx->wide_str = wide_str;
1600 : }
1601 : else
1602 : {
1603 : /* No need to keep the wide string if we're in a single-byte charset. */
1604 0 : pfree(wide_str);
1605 0 : matchctx->wide_str = NULL;
1606 0 : matchctx->conv_buf = NULL;
1607 0 : matchctx->conv_bufsiz = 0;
1608 : }
1609 :
1610 : /* Clean up temp storage */
1611 203510 : pfree(pmatch);
1612 :
1613 203510 : return matchctx;
1614 : }
1615 :
1616 : /*
1617 : * build_regexp_match_result - build output array for current match
1618 : */
1619 : static ArrayType *
1620 2768 : build_regexp_match_result(regexp_matches_ctx *matchctx)
1621 : {
1622 2768 : char *buf = matchctx->conv_buf;
1623 2768 : Datum *elems = matchctx->elems;
1624 2768 : bool *nulls = matchctx->nulls;
1625 : int dims[1];
1626 : int lbs[1];
1627 : int loc;
1628 : int i;
1629 :
1630 : /* Extract matching substrings from the original string */
1631 2768 : loc = matchctx->next_match * matchctx->npatterns * 2;
1632 8022 : for (i = 0; i < matchctx->npatterns; i++)
1633 : {
1634 5254 : int so = matchctx->match_locs[loc++];
1635 5254 : int eo = matchctx->match_locs[loc++];
1636 :
1637 5254 : if (so < 0 || eo < 0)
1638 : {
1639 6 : elems[i] = (Datum) 0;
1640 6 : nulls[i] = true;
1641 : }
1642 5248 : else if (buf)
1643 : {
1644 5248 : int len = pg_wchar2mb_with_len(matchctx->wide_str + so,
1645 : buf,
1646 : eo - so);
1647 :
1648 : Assert(len < matchctx->conv_bufsiz);
1649 5248 : elems[i] = PointerGetDatum(cstring_to_text_with_len(buf, len));
1650 5248 : nulls[i] = false;
1651 : }
1652 : else
1653 : {
1654 0 : elems[i] = DirectFunctionCall3(text_substr,
1655 : PointerGetDatum(matchctx->orig_str),
1656 : Int32GetDatum(so + 1),
1657 : Int32GetDatum(eo - so));
1658 0 : nulls[i] = false;
1659 : }
1660 : }
1661 :
1662 : /* And form an array */
1663 2768 : dims[0] = matchctx->npatterns;
1664 2768 : lbs[0] = 1;
1665 : /* XXX: this hardcodes assumptions about the text type */
1666 2768 : return construct_md_array(elems, nulls, 1, dims, lbs,
1667 : TEXTOID, -1, false, TYPALIGN_INT);
1668 : }
1669 :
1670 : /*
1671 : * regexp_split_to_table()
1672 : * Split the string at matches of the pattern, returning the
1673 : * split-out substrings as a table.
1674 : */
1675 : Datum
1676 622 : regexp_split_to_table(PG_FUNCTION_ARGS)
1677 : {
1678 : FuncCallContext *funcctx;
1679 : regexp_matches_ctx *splitctx;
1680 :
1681 622 : if (SRF_IS_FIRSTCALL())
1682 : {
1683 52 : text *pattern = PG_GETARG_TEXT_PP(1);
1684 52 : text *flags = PG_GETARG_TEXT_PP_IF_EXISTS(2);
1685 : pg_re_flags re_flags;
1686 : MemoryContext oldcontext;
1687 :
1688 52 : funcctx = SRF_FIRSTCALL_INIT();
1689 52 : oldcontext = MemoryContextSwitchTo(funcctx->multi_call_memory_ctx);
1690 :
1691 : /* Determine options */
1692 52 : parse_re_flags(&re_flags, flags);
1693 : /* User mustn't specify 'g' */
1694 46 : if (re_flags.glob)
1695 6 : ereport(ERROR,
1696 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
1697 : /* translator: %s is a SQL function name */
1698 : errmsg("%s does not support the \"global\" option",
1699 : "regexp_split_to_table()")));
1700 : /* But we find all the matches anyway */
1701 40 : re_flags.glob = true;
1702 :
1703 : /* be sure to copy the input string into the multi-call ctx */
1704 40 : splitctx = setup_regexp_matches(PG_GETARG_TEXT_P_COPY(0), pattern,
1705 : &re_flags, 0,
1706 : PG_GET_COLLATION(),
1707 : false, true, true);
1708 :
1709 40 : MemoryContextSwitchTo(oldcontext);
1710 40 : funcctx->user_fctx = (void *) splitctx;
1711 : }
1712 :
1713 610 : funcctx = SRF_PERCALL_SETUP();
1714 610 : splitctx = (regexp_matches_ctx *) funcctx->user_fctx;
1715 :
1716 610 : if (splitctx->next_match <= splitctx->nmatches)
1717 : {
1718 570 : Datum result = build_regexp_split_result(splitctx);
1719 :
1720 570 : splitctx->next_match++;
1721 570 : SRF_RETURN_NEXT(funcctx, result);
1722 : }
1723 :
1724 40 : SRF_RETURN_DONE(funcctx);
1725 : }
1726 :
1727 : /* This is separate to keep the opr_sanity regression test from complaining */
1728 : Datum
1729 552 : regexp_split_to_table_no_flags(PG_FUNCTION_ARGS)
1730 : {
1731 552 : return regexp_split_to_table(fcinfo);
1732 : }
1733 :
1734 : /*
1735 : * regexp_split_to_array()
1736 : * Split the string at matches of the pattern, returning the
1737 : * split-out substrings as an array.
1738 : */
1739 : Datum
1740 200424 : regexp_split_to_array(PG_FUNCTION_ARGS)
1741 : {
1742 200424 : ArrayBuildState *astate = NULL;
1743 : pg_re_flags re_flags;
1744 : regexp_matches_ctx *splitctx;
1745 :
1746 : /* Determine options */
1747 200424 : parse_re_flags(&re_flags, PG_GETARG_TEXT_PP_IF_EXISTS(2));
1748 : /* User mustn't specify 'g' */
1749 200418 : if (re_flags.glob)
1750 6 : ereport(ERROR,
1751 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
1752 : /* translator: %s is a SQL function name */
1753 : errmsg("%s does not support the \"global\" option",
1754 : "regexp_split_to_array()")));
1755 : /* But we find all the matches anyway */
1756 200412 : re_flags.glob = true;
1757 :
1758 200412 : splitctx = setup_regexp_matches(PG_GETARG_TEXT_PP(0),
1759 200412 : PG_GETARG_TEXT_PP(1),
1760 : &re_flags, 0,
1761 : PG_GET_COLLATION(),
1762 : false, true, true);
1763 :
1764 1291372 : while (splitctx->next_match <= splitctx->nmatches)
1765 : {
1766 1090960 : astate = accumArrayResult(astate,
1767 : build_regexp_split_result(splitctx),
1768 : false,
1769 : TEXTOID,
1770 : CurrentMemoryContext);
1771 1090960 : splitctx->next_match++;
1772 : }
1773 :
1774 200412 : PG_RETURN_DATUM(makeArrayResult(astate, CurrentMemoryContext));
1775 : }
1776 :
1777 : /* This is separate to keep the opr_sanity regression test from complaining */
1778 : Datum
1779 200382 : regexp_split_to_array_no_flags(PG_FUNCTION_ARGS)
1780 : {
1781 200382 : return regexp_split_to_array(fcinfo);
1782 : }
1783 :
1784 : /*
1785 : * build_regexp_split_result - build output string for current match
1786 : *
1787 : * We return the string between the current match and the previous one,
1788 : * or the string after the last match when next_match == nmatches.
1789 : */
1790 : static Datum
1791 1091530 : build_regexp_split_result(regexp_matches_ctx *splitctx)
1792 : {
1793 1091530 : char *buf = splitctx->conv_buf;
1794 : int startpos;
1795 : int endpos;
1796 :
1797 1091530 : if (splitctx->next_match > 0)
1798 891078 : startpos = splitctx->match_locs[splitctx->next_match * 2 - 1];
1799 : else
1800 200452 : startpos = 0;
1801 1091530 : if (startpos < 0)
1802 0 : elog(ERROR, "invalid match ending position");
1803 :
1804 1091530 : endpos = splitctx->match_locs[splitctx->next_match * 2];
1805 1091530 : if (endpos < startpos)
1806 0 : elog(ERROR, "invalid match starting position");
1807 :
1808 1091530 : if (buf)
1809 : {
1810 : int len;
1811 :
1812 1091530 : len = pg_wchar2mb_with_len(splitctx->wide_str + startpos,
1813 : buf,
1814 : endpos - startpos);
1815 : Assert(len < splitctx->conv_bufsiz);
1816 1091530 : return PointerGetDatum(cstring_to_text_with_len(buf, len));
1817 : }
1818 : else
1819 : {
1820 0 : return DirectFunctionCall3(text_substr,
1821 : PointerGetDatum(splitctx->orig_str),
1822 : Int32GetDatum(startpos + 1),
1823 : Int32GetDatum(endpos - startpos));
1824 : }
1825 : }
1826 :
1827 : /*
1828 : * regexp_substr()
1829 : * Return the substring that matches a regular expression pattern
1830 : */
1831 : Datum
1832 108 : regexp_substr(PG_FUNCTION_ARGS)
1833 : {
1834 108 : text *str = PG_GETARG_TEXT_PP(0);
1835 108 : text *pattern = PG_GETARG_TEXT_PP(1);
1836 108 : int start = 1;
1837 108 : int n = 1;
1838 108 : text *flags = PG_GETARG_TEXT_PP_IF_EXISTS(4);
1839 108 : int subexpr = 0;
1840 : int so,
1841 : eo,
1842 : pos;
1843 : pg_re_flags re_flags;
1844 : regexp_matches_ctx *matchctx;
1845 :
1846 : /* Collect optional parameters */
1847 108 : if (PG_NARGS() > 2)
1848 : {
1849 90 : start = PG_GETARG_INT32(2);
1850 90 : if (start <= 0)
1851 6 : ereport(ERROR,
1852 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
1853 : errmsg("invalid value for parameter \"%s\": %d",
1854 : "start", start)));
1855 : }
1856 102 : if (PG_NARGS() > 3)
1857 : {
1858 78 : n = PG_GETARG_INT32(3);
1859 78 : if (n <= 0)
1860 6 : ereport(ERROR,
1861 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
1862 : errmsg("invalid value for parameter \"%s\": %d",
1863 : "n", n)));
1864 : }
1865 96 : if (PG_NARGS() > 5)
1866 : {
1867 48 : subexpr = PG_GETARG_INT32(5);
1868 48 : if (subexpr < 0)
1869 6 : ereport(ERROR,
1870 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
1871 : errmsg("invalid value for parameter \"%s\": %d",
1872 : "subexpr", subexpr)));
1873 : }
1874 :
1875 : /* Determine options */
1876 90 : parse_re_flags(&re_flags, flags);
1877 : /* User mustn't specify 'g' */
1878 90 : if (re_flags.glob)
1879 6 : ereport(ERROR,
1880 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
1881 : /* translator: %s is a SQL function name */
1882 : errmsg("%s does not support the \"global\" option",
1883 : "regexp_substr()")));
1884 : /* But we find all the matches anyway */
1885 84 : re_flags.glob = true;
1886 :
1887 : /* Do the matching */
1888 84 : matchctx = setup_regexp_matches(str, pattern, &re_flags, start - 1,
1889 : PG_GET_COLLATION(),
1890 : (subexpr > 0), /* need submatches? */
1891 : false, false);
1892 :
1893 : /* When n exceeds matches return NULL (includes case of no matches) */
1894 84 : if (n > matchctx->nmatches)
1895 12 : PG_RETURN_NULL();
1896 :
1897 : /* When subexpr exceeds number of subexpressions return NULL */
1898 72 : if (subexpr > matchctx->npatterns)
1899 6 : PG_RETURN_NULL();
1900 :
1901 : /* Select the appropriate match position to return */
1902 66 : pos = (n - 1) * matchctx->npatterns;
1903 66 : if (subexpr > 0)
1904 30 : pos += subexpr - 1;
1905 66 : pos *= 2;
1906 66 : so = matchctx->match_locs[pos];
1907 66 : eo = matchctx->match_locs[pos + 1];
1908 :
1909 66 : if (so < 0 || eo < 0)
1910 6 : PG_RETURN_NULL(); /* unidentifiable location */
1911 :
1912 60 : PG_RETURN_DATUM(DirectFunctionCall3(text_substr,
1913 : PointerGetDatum(matchctx->orig_str),
1914 : Int32GetDatum(so + 1),
1915 : Int32GetDatum(eo - so)));
1916 : }
1917 :
1918 : /* This is separate to keep the opr_sanity regression test from complaining */
1919 : Datum
1920 18 : regexp_substr_no_start(PG_FUNCTION_ARGS)
1921 : {
1922 18 : return regexp_substr(fcinfo);
1923 : }
1924 :
1925 : /* This is separate to keep the opr_sanity regression test from complaining */
1926 : Datum
1927 6 : regexp_substr_no_n(PG_FUNCTION_ARGS)
1928 : {
1929 6 : return regexp_substr(fcinfo);
1930 : }
1931 :
1932 : /* This is separate to keep the opr_sanity regression test from complaining */
1933 : Datum
1934 24 : regexp_substr_no_flags(PG_FUNCTION_ARGS)
1935 : {
1936 24 : return regexp_substr(fcinfo);
1937 : }
1938 :
1939 : /* This is separate to keep the opr_sanity regression test from complaining */
1940 : Datum
1941 12 : regexp_substr_no_subexpr(PG_FUNCTION_ARGS)
1942 : {
1943 12 : return regexp_substr(fcinfo);
1944 : }
1945 :
1946 : /*
1947 : * regexp_fixed_prefix - extract fixed prefix, if any, for a regexp
1948 : *
1949 : * The result is NULL if there is no fixed prefix, else a palloc'd string.
1950 : * If it is an exact match, not just a prefix, *exact is returned as true.
1951 : */
1952 : char *
1953 14846 : regexp_fixed_prefix(text *text_re, bool case_insensitive, Oid collation,
1954 : bool *exact)
1955 : {
1956 : char *result;
1957 : regex_t *re;
1958 : int cflags;
1959 : int re_result;
1960 : pg_wchar *str;
1961 : size_t slen;
1962 : size_t maxlen;
1963 : char errMsg[100];
1964 :
1965 14846 : *exact = false; /* default result */
1966 :
1967 : /* Compile RE */
1968 14846 : cflags = REG_ADVANCED;
1969 14846 : if (case_insensitive)
1970 62 : cflags |= REG_ICASE;
1971 :
1972 14846 : re = RE_compile_and_cache(text_re, cflags | REG_NOSUB, collation);
1973 :
1974 : /* Examine it to see if there's a fixed prefix */
1975 14822 : re_result = pg_regprefix(re, &str, &slen);
1976 :
1977 14822 : switch (re_result)
1978 : {
1979 726 : case REG_NOMATCH:
1980 726 : return NULL;
1981 :
1982 1910 : case REG_PREFIX:
1983 : /* continue with wchar conversion */
1984 1910 : break;
1985 :
1986 12186 : case REG_EXACT:
1987 12186 : *exact = true;
1988 : /* continue with wchar conversion */
1989 12186 : break;
1990 :
1991 0 : default:
1992 : /* re failed??? */
1993 0 : pg_regerror(re_result, re, errMsg, sizeof(errMsg));
1994 0 : ereport(ERROR,
1995 : (errcode(ERRCODE_INVALID_REGULAR_EXPRESSION),
1996 : errmsg("regular expression failed: %s", errMsg)));
1997 : break;
1998 : }
1999 :
2000 : /* Convert pg_wchar result back to database encoding */
2001 14096 : maxlen = pg_database_encoding_max_length() * slen + 1;
2002 14096 : result = (char *) palloc(maxlen);
2003 14096 : slen = pg_wchar2mb_with_len(str, result, slen);
2004 : Assert(slen < maxlen);
2005 :
2006 14096 : pfree(str);
2007 :
2008 14096 : return result;
2009 : }
|
