Line data Source code
1 : /*-------------------------------------------------------------------------
2 : *
3 : * regexp.c
4 : * Postgres' interface to the regular expression package.
5 : *
6 : * Portions Copyright (c) 1996-2026, 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 3709154 : RE_compile_and_cache(text *text_re, int cflags, Oid collation)
142 : {
143 3709154 : int text_re_len = VARSIZE_ANY_EXHDR(text_re);
144 3709154 : 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 4013894 : for (i = 0; i < num_res; i++)
159 : {
160 4010629 : if (re_array[i].cre_pat_len == text_re_len &&
161 3713415 : re_array[i].cre_flags == cflags &&
162 3712816 : re_array[i].cre_collation == collation &&
163 3712628 : 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 3705889 : if (i > 0)
169 : {
170 235105 : re_temp = re_array[i];
171 235105 : memmove(&re_array[1], &re_array[0], i * sizeof(cached_re_str));
172 235105 : re_array[0] = re_temp;
173 : }
174 :
175 3705889 : return &re_array[0].cre_re;
176 : }
177 : }
178 :
179 : /* Set up the cache memory on first go through. */
180 3265 : if (unlikely(RegexpCacheMemoryContext == NULL))
181 854 : RegexpCacheMemoryContext =
182 854 : 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 3265 : pattern = palloc_array(pg_wchar, text_re_len + 1);
193 3265 : 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 3265 : re_temp.cre_context = AllocSetContextCreate(CurrentMemoryContext,
205 : "RegexpMemoryContext",
206 : ALLOCSET_SMALL_SIZES);
207 3265 : oldcontext = MemoryContextSwitchTo(re_temp.cre_context);
208 :
209 3265 : regcomp_result = pg_regcomp(&re_temp.cre_re,
210 : pattern,
211 : pattern_len,
212 : cflags,
213 : collation);
214 :
215 3253 : pfree(pattern);
216 :
217 3253 : if (regcomp_result != REG_OKAY)
218 : {
219 : /* re didn't compile (no need for pg_regfree, if so) */
220 18 : pg_regerror(regcomp_result, &re_temp.cre_re, errMsg, sizeof(errMsg));
221 18 : 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 3235 : re_temp.cre_pat = palloc(text_re_len + 1);
228 3235 : 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 3235 : re_temp.cre_pat[text_re_len] = 0;
235 3235 : MemoryContextSetIdentifier(re_temp.cre_context, re_temp.cre_pat);
236 :
237 3235 : re_temp.cre_pat_len = text_re_len;
238 3235 : re_temp.cre_flags = cflags;
239 3235 : 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 3235 : if (num_res >= MAX_CACHED_RES)
246 : {
247 444 : --num_res;
248 : Assert(num_res < MAX_CACHED_RES);
249 : /* Delete the memory context holding the regexp and pattern. */
250 444 : MemoryContextDelete(re_array[num_res].cre_context);
251 : }
252 :
253 : /* Re-parent the memory context to our long-lived cache context. */
254 3235 : MemoryContextSetParent(re_temp.cre_context, RegexpCacheMemoryContext);
255 :
256 3235 : if (num_res > 0)
257 2381 : memmove(&re_array[1], &re_array[0], num_res * sizeof(cached_re_str));
258 :
259 3235 : re_array[0] = re_temp;
260 3235 : num_res++;
261 :
262 3235 : MemoryContextSwitchTo(oldcontext);
263 :
264 3235 : 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 4137161 : 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 4137161 : 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 4137161 : 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 4137161 : 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 3588676 : 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 3588676 : data = palloc_array(pg_wchar, dat_len + 1);
333 3588676 : data_len = pg_mb2wchar_with_len(dat, data, dat_len);
334 :
335 : /* Perform RE match and return result */
336 3588676 : match = RE_wchar_execute(re, data, data_len, 0, nmatch, pmatch);
337 :
338 3588676 : pfree(data);
339 3588676 : 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 3587865 : 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 3587865 : if (nmatch < 2)
366 3587865 : cflags |= REG_NOSUB;
367 :
368 : /* Compile RE */
369 3587865 : re = RE_compile_and_cache(text_re, cflags, collation);
370 :
371 3587853 : 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 104802 : parse_re_flags(pg_re_flags *flags, text *opts)
386 : {
387 : /* regex flavor is always folded into the compile flags */
388 104802 : flags->cflags = REG_ADVANCED;
389 104802 : flags->glob = false;
390 :
391 104802 : if (opts)
392 : {
393 2351 : char *opt_p = VARDATA_ANY(opts);
394 2351 : int opt_len = VARSIZE_ANY_EXHDR(opts);
395 : int i;
396 :
397 5290 : for (i = 0; i < opt_len; i++)
398 : {
399 2951 : switch (opt_p[i])
400 : {
401 2190 : case 'g':
402 2190 : flags->glob = true;
403 2190 : break;
404 0 : case 'b': /* BREs (but why???) */
405 0 : flags->cflags &= ~(REG_ADVANCED | REG_EXTENDED | REG_QUOTE);
406 0 : break;
407 5 : case 'c': /* case sensitive */
408 5 : flags->cflags &= ~REG_ICASE;
409 5 : break;
410 0 : case 'e': /* plain EREs */
411 0 : flags->cflags |= REG_EXTENDED;
412 0 : flags->cflags &= ~(REG_ADVANCED | REG_QUOTE);
413 0 : break;
414 146 : case 'i': /* case insensitive */
415 146 : flags->cflags |= REG_ICASE;
416 146 : break;
417 589 : case 'm': /* Perloid synonym for n */
418 : case 'n': /* \n affects ^ $ . [^ */
419 589 : flags->cflags |= REG_NEWLINE;
420 589 : 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 6 : case 's': /* single line, \n ordinary */
430 6 : flags->cflags &= ~REG_NEWLINE;
431 6 : 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 3 : case 'x': /* expanded syntax */
440 3 : flags->cflags |= REG_EXPANDED;
441 3 : break;
442 12 : default:
443 12 : ereport(ERROR,
444 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
445 : errmsg("invalid regular expression option: \"%.*s\"",
446 : pg_mblen_range(opt_p + i, opt_p + opt_len), opt_p + i)));
447 : break;
448 : }
449 : }
450 : }
451 104790 : }
452 :
453 :
454 : /*
455 : * interface routines called by the function manager
456 : */
457 :
458 : Datum
459 3356164 : nameregexeq(PG_FUNCTION_ARGS)
460 : {
461 3356164 : Name n = PG_GETARG_NAME(0);
462 3356164 : text *p = PG_GETARG_TEXT_PP(1);
463 :
464 3356164 : 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 13715 : nameregexne(PG_FUNCTION_ARGS)
474 : {
475 13715 : Name n = PG_GETARG_NAME(0);
476 13715 : text *p = PG_GETARG_TEXT_PP(1);
477 :
478 13715 : 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 196868 : textregexeq(PG_FUNCTION_ARGS)
488 : {
489 196868 : text *s = PG_GETARG_TEXT_PP(0);
490 196868 : text *p = PG_GETARG_TEXT_PP(1);
491 :
492 196868 : 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 17076 : textregexne(PG_FUNCTION_ARGS)
502 : {
503 17076 : text *s = PG_GETARG_TEXT_PP(0);
504 17076 : text *p = PG_GETARG_TEXT_PP(1);
505 :
506 17076 : 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 3765 : nameicregexeq(PG_FUNCTION_ARGS)
523 : {
524 3765 : Name n = PG_GETARG_NAME(0);
525 3765 : text *p = PG_GETARG_TEXT_PP(1);
526 :
527 3765 : 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 3 : nameicregexne(PG_FUNCTION_ARGS)
537 : {
538 3 : Name n = PG_GETARG_NAME(0);
539 3 : text *p = PG_GETARG_TEXT_PP(1);
540 :
541 3 : 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
550 110 : texticregexeq(PG_FUNCTION_ARGS)
551 : {
552 110 : text *s = PG_GETARG_TEXT_PP(0);
553 110 : text *p = PG_GETARG_TEXT_PP(1);
554 :
555 110 : 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 14 : texticregexne(PG_FUNCTION_ARGS)
565 : {
566 14 : text *s = PG_GETARG_TEXT_PP(0);
567 14 : text *p = PG_GETARG_TEXT_PP(1);
568 :
569 14 : 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
583 823 : textregexsubstr(PG_FUNCTION_ARGS)
584 : {
585 823 : text *s = PG_GETARG_TEXT_PP(0);
586 823 : 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 */
593 823 : 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 : */
601 823 : if (!RE_execute(re,
602 823 : VARDATA_ANY(s), VARSIZE_ANY_EXHDR(s),
603 : 2, pmatch))
604 6 : PG_RETURN_NULL(); /* definitely no match */
605 :
606 817 : if (re->re_nsub > 0)
607 : {
608 : /* has parenthesized subexpressions, use the first one */
609 761 : so = pmatch[1].rm_so;
610 761 : eo = pmatch[1].rm_eo;
611 : }
612 : else
613 : {
614 : /* no parenthesized subexpression, use whole match */
615 56 : so = pmatch[0].rm_so;
616 56 : 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 : */
625 817 : if (so < 0 || eo < 0)
626 3 : PG_RETURN_NULL();
627 :
628 814 : 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 7218 : textregexreplace_noopt(PG_FUNCTION_ARGS)
643 : {
644 7218 : text *s = PG_GETARG_TEXT_PP(0);
645 7218 : text *p = PG_GETARG_TEXT_PP(1);
646 7218 : text *r = PG_GETARG_TEXT_PP(2);
647 :
648 7218 : 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 2154 : textregexreplace(PG_FUNCTION_ARGS)
659 : {
660 2154 : text *s = PG_GETARG_TEXT_PP(0);
661 2154 : text *p = PG_GETARG_TEXT_PP(1);
662 2154 : text *r = PG_GETARG_TEXT_PP(2);
663 2154 : 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 2154 : if (VARSIZE_ANY_EXHDR(opt) > 0)
674 : {
675 2154 : char *opt_p = VARDATA_ANY(opt);
676 2154 : const char *end_p = opt_p + VARSIZE_ANY_EXHDR(opt);
677 :
678 2154 : if (*opt_p >= '0' && *opt_p <= '9')
679 3 : ereport(ERROR,
680 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
681 : errmsg("invalid regular expression option: \"%.*s\"",
682 : pg_mblen_range(opt_p, end_p), opt_p),
683 : errhint("If you meant to use regexp_replace() with a start parameter, cast the fourth argument to integer explicitly.")));
684 : }
685 :
686 2151 : parse_re_flags(&flags, opt);
687 :
688 2148 : PG_RETURN_TEXT_P(replace_text_regexp(s, p, r,
689 : flags.cflags, PG_GET_COLLATION(),
690 : 0, flags.glob ? 0 : 1));
691 : }
692 :
693 : /*
694 : * textregexreplace_extended()
695 : * Return a string matched by a regular expression, with replacement.
696 : * Extends textregexreplace by allowing a start position and the
697 : * choice of the occurrence to replace (0 means all occurrences).
698 : */
699 : Datum
700 33 : textregexreplace_extended(PG_FUNCTION_ARGS)
701 : {
702 33 : text *s = PG_GETARG_TEXT_PP(0);
703 33 : text *p = PG_GETARG_TEXT_PP(1);
704 33 : text *r = PG_GETARG_TEXT_PP(2);
705 33 : int start = 1;
706 33 : int n = 1;
707 33 : text *flags = PG_GETARG_TEXT_PP_IF_EXISTS(5);
708 : pg_re_flags re_flags;
709 :
710 : /* Collect optional parameters */
711 33 : if (PG_NARGS() > 3)
712 : {
713 33 : start = PG_GETARG_INT32(3);
714 33 : if (start <= 0)
715 3 : ereport(ERROR,
716 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
717 : errmsg("invalid value for parameter \"%s\": %d",
718 : "start", start)));
719 : }
720 30 : if (PG_NARGS() > 4)
721 : {
722 27 : n = PG_GETARG_INT32(4);
723 27 : if (n < 0)
724 3 : ereport(ERROR,
725 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
726 : errmsg("invalid value for parameter \"%s\": %d",
727 : "n", n)));
728 : }
729 :
730 : /* Determine options */
731 27 : parse_re_flags(&re_flags, flags);
732 :
733 : /* If N was not specified, deduce it from the 'g' flag */
734 27 : if (PG_NARGS() <= 4)
735 3 : n = re_flags.glob ? 0 : 1;
736 :
737 : /* Do the replacement(s) */
738 27 : PG_RETURN_TEXT_P(replace_text_regexp(s, p, r,
739 : re_flags.cflags, PG_GET_COLLATION(),
740 : start - 1, n));
741 : }
742 :
743 : /* This is separate to keep the opr_sanity regression test from complaining */
744 : Datum
745 3 : textregexreplace_extended_no_n(PG_FUNCTION_ARGS)
746 : {
747 3 : return textregexreplace_extended(fcinfo);
748 : }
749 :
750 : /* This is separate to keep the opr_sanity regression test from complaining */
751 : Datum
752 3 : textregexreplace_extended_no_flags(PG_FUNCTION_ARGS)
753 : {
754 3 : return textregexreplace_extended(fcinfo);
755 : }
756 :
757 : /*
758 : * similar_to_escape(), similar_escape()
759 : *
760 : * Convert a SQL "SIMILAR TO" regexp pattern to POSIX style, so it can be
761 : * used by our regexp engine.
762 : *
763 : * similar_escape_internal() is the common workhorse for three SQL-exposed
764 : * functions. esc_text can be passed as NULL to select the default escape
765 : * (which is '\'), or as an empty string to select no escape character.
766 : */
767 : static text *
768 96 : similar_escape_internal(text *pat_text, text *esc_text)
769 : {
770 : text *result;
771 : char *p,
772 : *e,
773 : *r;
774 : int plen,
775 : elen;
776 : const char *pend;
777 96 : bool afterescape = false;
778 96 : int nquotes = 0;
779 96 : int bracket_depth = 0; /* square bracket nesting level */
780 96 : int charclass_pos = 0; /* position inside a character class */
781 :
782 96 : p = VARDATA_ANY(pat_text);
783 96 : plen = VARSIZE_ANY_EXHDR(pat_text);
784 96 : pend = p + plen;
785 96 : if (esc_text == NULL)
786 : {
787 : /* No ESCAPE clause provided; default to backslash as escape */
788 44 : e = "\\";
789 44 : elen = 1;
790 : }
791 : else
792 : {
793 52 : e = VARDATA_ANY(esc_text);
794 52 : elen = VARSIZE_ANY_EXHDR(esc_text);
795 52 : if (elen == 0)
796 3 : e = NULL; /* no escape character */
797 49 : else if (elen > 1)
798 : {
799 6 : int escape_mblen = pg_mbstrlen_with_len(e, elen);
800 :
801 6 : if (escape_mblen > 1)
802 3 : ereport(ERROR,
803 : (errcode(ERRCODE_INVALID_ESCAPE_SEQUENCE),
804 : errmsg("invalid escape string"),
805 : errhint("Escape string must be empty or one character.")));
806 : }
807 : }
808 :
809 : /*----------
810 : * We surround the transformed input string with
811 : * ^(?: ... )$
812 : * which requires some explanation. We need "^" and "$" to force
813 : * the pattern to match the entire input string as per the SQL spec.
814 : * The "(?:" and ")" are a non-capturing set of parens; we have to have
815 : * parens in case the string contains "|", else the "^" and "$" will
816 : * be bound into the first and last alternatives which is not what we
817 : * want, and the parens must be non capturing because we don't want them
818 : * to count when selecting output for SUBSTRING.
819 : *
820 : * When the pattern is divided into three parts by escape-double-quotes,
821 : * what we emit is
822 : * ^(?:part1){1,1}?(part2){1,1}(?:part3)$
823 : * which requires even more explanation. The "{1,1}?" on part1 makes it
824 : * non-greedy so that it will match the smallest possible amount of text
825 : * not the largest, as required by SQL. The plain parens around part2
826 : * are capturing parens so that that part is what controls the result of
827 : * SUBSTRING. The "{1,1}" forces part2 to be greedy, so that it matches
828 : * the largest possible amount of text; hence part3 must match the
829 : * smallest amount of text, as required by SQL. We don't need an explicit
830 : * greediness marker on part3. Note that this also confines the effects
831 : * of any "|" characters to the respective part, which is what we want.
832 : *
833 : * The SQL spec says that SUBSTRING's pattern must contain exactly two
834 : * escape-double-quotes, but we only complain if there's more than two.
835 : * With none, we act as though part1 and part3 are empty; with one, we
836 : * act as though part3 is empty. Both behaviors fall out of omitting
837 : * the relevant part separators in the above expansion. If the result
838 : * of this function is used in a plain regexp match (SIMILAR TO), the
839 : * escape-double-quotes have no effect on the match behavior.
840 : *
841 : * While we don't fully validate character classes (bracket expressions),
842 : * we do need to parse them well enough to know where they end.
843 : * "charclass_pos" tracks where we are in a character class.
844 : * Its value is uninteresting when bracket_depth is 0.
845 : * But when bracket_depth > 0, it will be
846 : * 1: right after the opening '[' (a following '^' will negate
847 : * the class, while ']' is a literal character)
848 : * 2: right after a '^' after the opening '[' (']' is still a literal
849 : * character)
850 : * 3 or more: further inside the character class (']' ends the class)
851 : *----------
852 : */
853 :
854 : /*
855 : * We need room for the prefix/postfix and part separators, plus as many
856 : * as 3 output bytes per input byte; since the input is at most 1GB this
857 : * can't overflow size_t.
858 : */
859 93 : result = (text *) palloc(VARHDRSZ + 23 + 3 * (size_t) plen);
860 93 : r = VARDATA(result);
861 :
862 93 : *r++ = '^';
863 93 : *r++ = '(';
864 93 : *r++ = '?';
865 93 : *r++ = ':';
866 :
867 926 : while (plen > 0)
868 : {
869 836 : char pchar = *p;
870 :
871 : /*
872 : * If both the escape character and the current character from the
873 : * pattern are multi-byte, we need to take the slow path.
874 : *
875 : * But if one of them is single-byte, we can process the pattern one
876 : * byte at a time, ignoring multi-byte characters. (This works
877 : * because all server-encodings have the property that a valid
878 : * multi-byte character representation cannot contain the
879 : * representation of a valid single-byte character.)
880 : */
881 :
882 836 : if (elen > 1)
883 : {
884 3 : int mblen = pg_mblen_range(p, pend);
885 :
886 3 : if (mblen > 1)
887 : {
888 : /* slow, multi-byte path */
889 3 : if (afterescape)
890 : {
891 0 : *r++ = '\\';
892 0 : memcpy(r, p, mblen);
893 0 : r += mblen;
894 0 : afterescape = false;
895 : }
896 3 : else if (e && elen == mblen && memcmp(e, p, mblen) == 0)
897 : {
898 : /* SQL escape character; do not send to output */
899 0 : afterescape = true;
900 : }
901 : else
902 : {
903 : /*
904 : * We know it's a multi-byte character, so we don't need
905 : * to do all the comparisons to single-byte characters
906 : * that we do below.
907 : */
908 3 : memcpy(r, p, mblen);
909 3 : r += mblen;
910 : }
911 :
912 3 : p += mblen;
913 3 : plen -= mblen;
914 :
915 3 : continue;
916 : }
917 : }
918 :
919 : /* fast path */
920 833 : if (afterescape)
921 : {
922 83 : if (pchar == '"' && bracket_depth < 1) /* escape-double-quote? */
923 : {
924 : /* emit appropriate part separator, per notes above */
925 62 : if (nquotes == 0)
926 : {
927 31 : *r++ = ')';
928 31 : *r++ = '{';
929 31 : *r++ = '1';
930 31 : *r++ = ',';
931 31 : *r++ = '1';
932 31 : *r++ = '}';
933 31 : *r++ = '?';
934 31 : *r++ = '(';
935 : }
936 31 : else if (nquotes == 1)
937 : {
938 28 : *r++ = ')';
939 28 : *r++ = '{';
940 28 : *r++ = '1';
941 28 : *r++ = ',';
942 28 : *r++ = '1';
943 28 : *r++ = '}';
944 28 : *r++ = '(';
945 28 : *r++ = '?';
946 28 : *r++ = ':';
947 : }
948 : else
949 3 : ereport(ERROR,
950 : (errcode(ERRCODE_INVALID_USE_OF_ESCAPE_CHARACTER),
951 : errmsg("SQL regular expression may not contain more than two escape-double-quote separators")));
952 59 : nquotes++;
953 : }
954 : else
955 : {
956 : /*
957 : * We allow any character at all to be escaped; notably, this
958 : * allows access to POSIX character-class escapes such as
959 : * "\d". The SQL spec is considerably more restrictive.
960 : */
961 21 : *r++ = '\\';
962 21 : *r++ = pchar;
963 :
964 : /*
965 : * If we encounter an escaped character in a character class,
966 : * we are no longer at the beginning.
967 : */
968 21 : charclass_pos = 3;
969 : }
970 80 : afterescape = false;
971 : }
972 750 : else if (e && pchar == *e)
973 : {
974 : /* SQL escape character; do not send to output */
975 83 : afterescape = true;
976 : }
977 667 : else if (bracket_depth > 0)
978 : {
979 : /* inside a character class */
980 306 : if (pchar == '\\')
981 : {
982 : /*
983 : * If we're here, backslash is not the SQL escape character,
984 : * so treat it as a literal class element, which requires
985 : * doubling it. (This matches our behavior for backslashes
986 : * outside character classes.)
987 : */
988 0 : *r++ = '\\';
989 : }
990 306 : *r++ = pchar;
991 :
992 : /* parse the character class well enough to identify ending ']' */
993 306 : if (pchar == ']' && charclass_pos > 2)
994 : {
995 : /* found the real end of a bracket pair */
996 69 : bracket_depth--;
997 : /* don't reset charclass_pos, this may be an inner bracket */
998 : }
999 237 : else if (pchar == '[')
1000 : {
1001 : /* start of a nested bracket pair */
1002 36 : bracket_depth++;
1003 :
1004 : /*
1005 : * We are no longer at the beginning of a character class.
1006 : * (The nested bracket pair is a collating element, not a
1007 : * character class in its own right.)
1008 : */
1009 36 : charclass_pos = 3;
1010 : }
1011 201 : else if (pchar == '^')
1012 : {
1013 : /*
1014 : * A caret right after the opening bracket negates the
1015 : * character class. In that case, the following will
1016 : * increment charclass_pos from 1 to 2, so that a following
1017 : * ']' is still a literal character and does not end the
1018 : * character class. If we are further inside a character
1019 : * class, charclass_pos might get incremented past 3, which is
1020 : * fine.
1021 : */
1022 30 : charclass_pos++;
1023 : }
1024 : else
1025 : {
1026 : /*
1027 : * Anything else (including a backslash or leading ']') is an
1028 : * element of the character class, so we are no longer at the
1029 : * beginning of the class.
1030 : */
1031 171 : charclass_pos = 3;
1032 : }
1033 : }
1034 361 : else if (pchar == '[')
1035 : {
1036 : /* start of a character class */
1037 33 : *r++ = pchar;
1038 33 : bracket_depth = 1;
1039 33 : charclass_pos = 1;
1040 : }
1041 328 : else if (pchar == '%')
1042 : {
1043 66 : *r++ = '.';
1044 66 : *r++ = '*';
1045 : }
1046 262 : else if (pchar == '_')
1047 32 : *r++ = '.';
1048 230 : else if (pchar == '(')
1049 : {
1050 : /* convert to non-capturing parenthesis */
1051 15 : *r++ = '(';
1052 15 : *r++ = '?';
1053 15 : *r++ = ':';
1054 : }
1055 215 : else if (pchar == '\\' || pchar == '.' ||
1056 195 : pchar == '^' || pchar == '$')
1057 : {
1058 26 : *r++ = '\\';
1059 26 : *r++ = pchar;
1060 : }
1061 : else
1062 189 : *r++ = pchar;
1063 830 : p++, plen--;
1064 : }
1065 :
1066 90 : *r++ = ')';
1067 90 : *r++ = '$';
1068 :
1069 90 : SET_VARSIZE(result, r - ((char *) result));
1070 :
1071 90 : return result;
1072 : }
1073 :
1074 : /*
1075 : * similar_to_escape(pattern, escape)
1076 : */
1077 : Datum
1078 52 : similar_to_escape_2(PG_FUNCTION_ARGS)
1079 : {
1080 52 : text *pat_text = PG_GETARG_TEXT_PP(0);
1081 52 : text *esc_text = PG_GETARG_TEXT_PP(1);
1082 : text *result;
1083 :
1084 52 : result = similar_escape_internal(pat_text, esc_text);
1085 :
1086 46 : PG_RETURN_TEXT_P(result);
1087 : }
1088 :
1089 : /*
1090 : * similar_to_escape(pattern)
1091 : * Inserts a default escape character.
1092 : */
1093 : Datum
1094 44 : similar_to_escape_1(PG_FUNCTION_ARGS)
1095 : {
1096 44 : text *pat_text = PG_GETARG_TEXT_PP(0);
1097 : text *result;
1098 :
1099 44 : result = similar_escape_internal(pat_text, NULL);
1100 :
1101 44 : PG_RETURN_TEXT_P(result);
1102 : }
1103 :
1104 : /*
1105 : * similar_escape(pattern, escape)
1106 : *
1107 : * Legacy function for compatibility with views stored using the
1108 : * pre-v13 expansion of SIMILAR TO. Unlike the above functions, this
1109 : * is non-strict, which leads to not-per-spec handling of "ESCAPE NULL".
1110 : */
1111 : Datum
1112 0 : similar_escape(PG_FUNCTION_ARGS)
1113 : {
1114 : text *pat_text;
1115 : text *esc_text;
1116 : text *result;
1117 :
1118 : /* This function is not strict, so must test explicitly */
1119 0 : if (PG_ARGISNULL(0))
1120 0 : PG_RETURN_NULL();
1121 0 : pat_text = PG_GETARG_TEXT_PP(0);
1122 :
1123 0 : if (PG_ARGISNULL(1))
1124 0 : esc_text = NULL; /* use default escape character */
1125 : else
1126 0 : esc_text = PG_GETARG_TEXT_PP(1);
1127 :
1128 0 : result = similar_escape_internal(pat_text, esc_text);
1129 :
1130 0 : PG_RETURN_TEXT_P(result);
1131 : }
1132 :
1133 : /*
1134 : * regexp_count()
1135 : * Return the number of matches of a pattern within a string.
1136 : */
1137 : Datum
1138 24 : regexp_count(PG_FUNCTION_ARGS)
1139 : {
1140 24 : text *str = PG_GETARG_TEXT_PP(0);
1141 24 : text *pattern = PG_GETARG_TEXT_PP(1);
1142 24 : int start = 1;
1143 24 : text *flags = PG_GETARG_TEXT_PP_IF_EXISTS(3);
1144 : pg_re_flags re_flags;
1145 : regexp_matches_ctx *matchctx;
1146 :
1147 : /* Collect optional parameters */
1148 24 : if (PG_NARGS() > 2)
1149 : {
1150 21 : start = PG_GETARG_INT32(2);
1151 21 : if (start <= 0)
1152 6 : ereport(ERROR,
1153 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
1154 : errmsg("invalid value for parameter \"%s\": %d",
1155 : "start", start)));
1156 : }
1157 :
1158 : /* Determine options */
1159 18 : parse_re_flags(&re_flags, flags);
1160 : /* User mustn't specify 'g' */
1161 18 : if (re_flags.glob)
1162 0 : ereport(ERROR,
1163 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
1164 : /* translator: %s is a SQL function name */
1165 : errmsg("%s does not support the \"global\" option",
1166 : "regexp_count()")));
1167 : /* But we find all the matches anyway */
1168 18 : re_flags.glob = true;
1169 :
1170 : /* Do the matching */
1171 18 : matchctx = setup_regexp_matches(str, pattern, &re_flags, start - 1,
1172 : PG_GET_COLLATION(),
1173 : false, /* can ignore subexprs */
1174 : false, false);
1175 :
1176 18 : PG_RETURN_INT32(matchctx->nmatches);
1177 : }
1178 :
1179 : /* This is separate to keep the opr_sanity regression test from complaining */
1180 : Datum
1181 3 : regexp_count_no_start(PG_FUNCTION_ARGS)
1182 : {
1183 3 : return regexp_count(fcinfo);
1184 : }
1185 :
1186 : /* This is separate to keep the opr_sanity regression test from complaining */
1187 : Datum
1188 15 : regexp_count_no_flags(PG_FUNCTION_ARGS)
1189 : {
1190 15 : return regexp_count(fcinfo);
1191 : }
1192 :
1193 : /*
1194 : * regexp_instr()
1195 : * Return the match's position within the string
1196 : */
1197 : Datum
1198 78 : regexp_instr(PG_FUNCTION_ARGS)
1199 : {
1200 78 : text *str = PG_GETARG_TEXT_PP(0);
1201 78 : text *pattern = PG_GETARG_TEXT_PP(1);
1202 78 : int start = 1;
1203 78 : int n = 1;
1204 78 : int endoption = 0;
1205 78 : text *flags = PG_GETARG_TEXT_PP_IF_EXISTS(5);
1206 78 : int subexpr = 0;
1207 : int pos;
1208 : pg_re_flags re_flags;
1209 : regexp_matches_ctx *matchctx;
1210 :
1211 : /* Collect optional parameters */
1212 78 : if (PG_NARGS() > 2)
1213 : {
1214 69 : start = PG_GETARG_INT32(2);
1215 69 : if (start <= 0)
1216 3 : ereport(ERROR,
1217 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
1218 : errmsg("invalid value for parameter \"%s\": %d",
1219 : "start", start)));
1220 : }
1221 75 : if (PG_NARGS() > 3)
1222 : {
1223 63 : n = PG_GETARG_INT32(3);
1224 63 : if (n <= 0)
1225 3 : ereport(ERROR,
1226 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
1227 : errmsg("invalid value for parameter \"%s\": %d",
1228 : "n", n)));
1229 : }
1230 72 : if (PG_NARGS() > 4)
1231 : {
1232 54 : endoption = PG_GETARG_INT32(4);
1233 54 : if (endoption != 0 && endoption != 1)
1234 6 : ereport(ERROR,
1235 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
1236 : errmsg("invalid value for parameter \"%s\": %d",
1237 : "endoption", endoption)));
1238 : }
1239 66 : if (PG_NARGS() > 6)
1240 : {
1241 42 : subexpr = PG_GETARG_INT32(6);
1242 42 : if (subexpr < 0)
1243 3 : ereport(ERROR,
1244 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
1245 : errmsg("invalid value for parameter \"%s\": %d",
1246 : "subexpr", subexpr)));
1247 : }
1248 :
1249 : /* Determine options */
1250 63 : parse_re_flags(&re_flags, flags);
1251 : /* User mustn't specify 'g' */
1252 63 : if (re_flags.glob)
1253 3 : ereport(ERROR,
1254 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
1255 : /* translator: %s is a SQL function name */
1256 : errmsg("%s does not support the \"global\" option",
1257 : "regexp_instr()")));
1258 : /* But we find all the matches anyway */
1259 60 : re_flags.glob = true;
1260 :
1261 : /* Do the matching */
1262 60 : matchctx = setup_regexp_matches(str, pattern, &re_flags, start - 1,
1263 : PG_GET_COLLATION(),
1264 : (subexpr > 0), /* need submatches? */
1265 : false, false);
1266 :
1267 : /* When n exceeds matches return 0 (includes case of no matches) */
1268 60 : if (n > matchctx->nmatches)
1269 6 : PG_RETURN_INT32(0);
1270 :
1271 : /* When subexpr exceeds number of subexpressions return 0 */
1272 54 : if (subexpr > matchctx->npatterns)
1273 6 : PG_RETURN_INT32(0);
1274 :
1275 : /* Select the appropriate match position to return */
1276 48 : pos = (n - 1) * matchctx->npatterns;
1277 48 : if (subexpr > 0)
1278 27 : pos += subexpr - 1;
1279 48 : pos *= 2;
1280 48 : if (endoption == 1)
1281 15 : pos += 1;
1282 :
1283 48 : if (matchctx->match_locs[pos] >= 0)
1284 45 : PG_RETURN_INT32(matchctx->match_locs[pos] + 1);
1285 : else
1286 3 : PG_RETURN_INT32(0); /* position not identifiable */
1287 : }
1288 :
1289 : /* This is separate to keep the opr_sanity regression test from complaining */
1290 : Datum
1291 9 : regexp_instr_no_start(PG_FUNCTION_ARGS)
1292 : {
1293 9 : return regexp_instr(fcinfo);
1294 : }
1295 :
1296 : /* This is separate to keep the opr_sanity regression test from complaining */
1297 : Datum
1298 3 : regexp_instr_no_n(PG_FUNCTION_ARGS)
1299 : {
1300 3 : return regexp_instr(fcinfo);
1301 : }
1302 :
1303 : /* This is separate to keep the opr_sanity regression test from complaining */
1304 : Datum
1305 12 : regexp_instr_no_endoption(PG_FUNCTION_ARGS)
1306 : {
1307 12 : return regexp_instr(fcinfo);
1308 : }
1309 :
1310 : /* This is separate to keep the opr_sanity regression test from complaining */
1311 : Datum
1312 6 : regexp_instr_no_flags(PG_FUNCTION_ARGS)
1313 : {
1314 6 : return regexp_instr(fcinfo);
1315 : }
1316 :
1317 : /* This is separate to keep the opr_sanity regression test from complaining */
1318 : Datum
1319 6 : regexp_instr_no_subexpr(PG_FUNCTION_ARGS)
1320 : {
1321 6 : return regexp_instr(fcinfo);
1322 : }
1323 :
1324 : /*
1325 : * regexp_like()
1326 : * Test for a pattern match within a string.
1327 : */
1328 : Datum
1329 15 : regexp_like(PG_FUNCTION_ARGS)
1330 : {
1331 15 : text *str = PG_GETARG_TEXT_PP(0);
1332 15 : text *pattern = PG_GETARG_TEXT_PP(1);
1333 15 : text *flags = PG_GETARG_TEXT_PP_IF_EXISTS(2);
1334 : pg_re_flags re_flags;
1335 :
1336 : /* Determine options */
1337 15 : parse_re_flags(&re_flags, flags);
1338 : /* User mustn't specify 'g' */
1339 15 : if (re_flags.glob)
1340 3 : ereport(ERROR,
1341 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
1342 : /* translator: %s is a SQL function name */
1343 : errmsg("%s does not support the \"global\" option",
1344 : "regexp_like()")));
1345 :
1346 : /* Otherwise it's like textregexeq/texticregexeq */
1347 12 : PG_RETURN_BOOL(RE_compile_and_execute(pattern,
1348 : VARDATA_ANY(str),
1349 : VARSIZE_ANY_EXHDR(str),
1350 : re_flags.cflags,
1351 : PG_GET_COLLATION(),
1352 : 0, NULL));
1353 : }
1354 :
1355 : /* This is separate to keep the opr_sanity regression test from complaining */
1356 : Datum
1357 3 : regexp_like_no_flags(PG_FUNCTION_ARGS)
1358 : {
1359 3 : return regexp_like(fcinfo);
1360 : }
1361 :
1362 : /*
1363 : * regexp_match()
1364 : * Return the first substring(s) matching a pattern within a string.
1365 : */
1366 : Datum
1367 1270 : regexp_match(PG_FUNCTION_ARGS)
1368 : {
1369 1270 : text *orig_str = PG_GETARG_TEXT_PP(0);
1370 1270 : text *pattern = PG_GETARG_TEXT_PP(1);
1371 1270 : text *flags = PG_GETARG_TEXT_PP_IF_EXISTS(2);
1372 : pg_re_flags re_flags;
1373 : regexp_matches_ctx *matchctx;
1374 :
1375 : /* Determine options */
1376 1270 : parse_re_flags(&re_flags, flags);
1377 : /* User mustn't specify 'g' */
1378 1270 : if (re_flags.glob)
1379 4 : ereport(ERROR,
1380 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
1381 : /* translator: %s is a SQL function name */
1382 : errmsg("%s does not support the \"global\" option",
1383 : "regexp_match()"),
1384 : errhint("Use the regexp_matches function instead.")));
1385 :
1386 1266 : matchctx = setup_regexp_matches(orig_str, pattern, &re_flags, 0,
1387 : PG_GET_COLLATION(), true, false, false);
1388 :
1389 1266 : if (matchctx->nmatches == 0)
1390 65 : PG_RETURN_NULL();
1391 :
1392 : Assert(matchctx->nmatches == 1);
1393 :
1394 : /* Create workspace that build_regexp_match_result needs */
1395 1201 : matchctx->elems = palloc_array(Datum, matchctx->npatterns);
1396 1201 : matchctx->nulls = palloc_array(bool, matchctx->npatterns);
1397 :
1398 1201 : PG_RETURN_DATUM(PointerGetDatum(build_regexp_match_result(matchctx)));
1399 : }
1400 :
1401 : /* This is separate to keep the opr_sanity regression test from complaining */
1402 : Datum
1403 1255 : regexp_match_no_flags(PG_FUNCTION_ARGS)
1404 : {
1405 1255 : return regexp_match(fcinfo);
1406 : }
1407 :
1408 : /*
1409 : * regexp_matches()
1410 : * Return a table of all matches of a pattern within a string.
1411 : */
1412 : Datum
1413 1305 : regexp_matches(PG_FUNCTION_ARGS)
1414 : {
1415 : FuncCallContext *funcctx;
1416 : regexp_matches_ctx *matchctx;
1417 :
1418 1305 : if (SRF_IS_FIRSTCALL())
1419 : {
1420 975 : text *pattern = PG_GETARG_TEXT_PP(1);
1421 975 : text *flags = PG_GETARG_TEXT_PP_IF_EXISTS(2);
1422 : pg_re_flags re_flags;
1423 : MemoryContext oldcontext;
1424 :
1425 975 : funcctx = SRF_FIRSTCALL_INIT();
1426 975 : oldcontext = MemoryContextSwitchTo(funcctx->multi_call_memory_ctx);
1427 :
1428 : /* Determine options */
1429 975 : parse_re_flags(&re_flags, flags);
1430 :
1431 : /* be sure to copy the input string into the multi-call ctx */
1432 972 : matchctx = setup_regexp_matches(PG_GETARG_TEXT_P_COPY(0), pattern,
1433 : &re_flags, 0,
1434 : PG_GET_COLLATION(),
1435 : true, false, false);
1436 :
1437 : /* Pre-create workspace that build_regexp_match_result needs */
1438 966 : matchctx->elems = palloc_array(Datum, matchctx->npatterns);
1439 966 : matchctx->nulls = palloc_array(bool, matchctx->npatterns);
1440 :
1441 966 : MemoryContextSwitchTo(oldcontext);
1442 966 : funcctx->user_fctx = matchctx;
1443 : }
1444 :
1445 1296 : funcctx = SRF_PERCALL_SETUP();
1446 1296 : matchctx = (regexp_matches_ctx *) funcctx->user_fctx;
1447 :
1448 1296 : if (matchctx->next_match < matchctx->nmatches)
1449 : {
1450 : ArrayType *result_ary;
1451 :
1452 330 : result_ary = build_regexp_match_result(matchctx);
1453 330 : matchctx->next_match++;
1454 330 : SRF_RETURN_NEXT(funcctx, PointerGetDatum(result_ary));
1455 : }
1456 :
1457 966 : SRF_RETURN_DONE(funcctx);
1458 : }
1459 :
1460 : /* This is separate to keep the opr_sanity regression test from complaining */
1461 : Datum
1462 1143 : regexp_matches_no_flags(PG_FUNCTION_ARGS)
1463 : {
1464 1143 : return regexp_matches(fcinfo);
1465 : }
1466 :
1467 : /*
1468 : * setup_regexp_matches --- do the initial matching for regexp_match,
1469 : * regexp_split, and related functions
1470 : *
1471 : * To avoid having to re-find the compiled pattern on each call, we do
1472 : * all the matching in one swoop. The returned regexp_matches_ctx contains
1473 : * the locations of all the substrings matching the pattern.
1474 : *
1475 : * start_search: the character (not byte) offset in orig_str at which to
1476 : * begin the search. Returned positions are relative to orig_str anyway.
1477 : * use_subpatterns: collect data about matches to parenthesized subexpressions.
1478 : * ignore_degenerate: ignore zero-length matches.
1479 : * fetching_unmatched: caller wants to fetch unmatched substrings.
1480 : *
1481 : * We don't currently assume that fetching_unmatched is exclusive of fetching
1482 : * the matched text too; if it's set, the conversion buffer is large enough to
1483 : * fetch any single matched or unmatched string, but not any larger
1484 : * substring. (In practice, when splitting the matches are usually small
1485 : * anyway, and it didn't seem worth complicating the code further.)
1486 : */
1487 : static regexp_matches_ctx *
1488 102584 : setup_regexp_matches(text *orig_str, text *pattern, pg_re_flags *re_flags,
1489 : int start_search,
1490 : Oid collation,
1491 : bool use_subpatterns,
1492 : bool ignore_degenerate,
1493 : bool fetching_unmatched)
1494 : {
1495 102584 : regexp_matches_ctx *matchctx = palloc0_object(regexp_matches_ctx);
1496 102584 : int eml = pg_database_encoding_max_length();
1497 : int orig_len;
1498 : pg_wchar *wide_str;
1499 : int wide_len;
1500 : int cflags;
1501 : regex_t *cpattern;
1502 : regmatch_t *pmatch;
1503 : int pmatch_len;
1504 : int array_len;
1505 : int array_idx;
1506 : int prev_match_end;
1507 : int prev_valid_match_end;
1508 102584 : int maxlen = 0; /* largest fetch length in characters */
1509 :
1510 : /* save original string --- we'll extract result substrings from it */
1511 102584 : matchctx->orig_str = orig_str;
1512 :
1513 : /* convert string to pg_wchar form for matching */
1514 102584 : orig_len = VARSIZE_ANY_EXHDR(orig_str);
1515 102584 : wide_str = palloc_array(pg_wchar, orig_len + 1);
1516 102584 : wide_len = pg_mb2wchar_with_len(VARDATA_ANY(orig_str), wide_str, orig_len);
1517 :
1518 : /* set up the compiled pattern */
1519 102584 : cflags = re_flags->cflags;
1520 102584 : if (!use_subpatterns)
1521 100295 : cflags |= REG_NOSUB;
1522 102584 : cpattern = RE_compile_and_cache(pattern, cflags, collation);
1523 :
1524 : /* do we want to remember subpatterns? */
1525 102578 : if (use_subpatterns && cpattern->re_nsub > 0)
1526 : {
1527 1347 : matchctx->npatterns = cpattern->re_nsub;
1528 1347 : pmatch_len = cpattern->re_nsub + 1;
1529 : }
1530 : else
1531 : {
1532 101231 : use_subpatterns = false;
1533 101231 : matchctx->npatterns = 1;
1534 101231 : pmatch_len = 1;
1535 : }
1536 :
1537 : /* temporary output space for RE package */
1538 102578 : pmatch = palloc_array(regmatch_t, pmatch_len);
1539 :
1540 : /*
1541 : * the real output space (grown dynamically if needed)
1542 : *
1543 : * use values 2^n-1, not 2^n, so that we hit the limit at 2^28-1 rather
1544 : * than at 2^27
1545 : */
1546 102578 : array_len = re_flags->glob ? 255 : 31;
1547 102578 : matchctx->match_locs = palloc_array(int, array_len);
1548 102578 : array_idx = 0;
1549 :
1550 : /* search for the pattern, perhaps repeatedly */
1551 102578 : prev_match_end = 0;
1552 102578 : prev_valid_match_end = 0;
1553 548485 : while (RE_wchar_execute(cpattern, wide_str, wide_len, start_search,
1554 : pmatch_len, pmatch))
1555 : {
1556 : /*
1557 : * If requested, ignore degenerate matches, which are zero-length
1558 : * matches occurring at the start or end of a string or just after a
1559 : * previous match.
1560 : */
1561 447361 : if (!ignore_degenerate ||
1562 445632 : (pmatch[0].rm_so < wide_len &&
1563 445611 : pmatch[0].rm_eo > prev_match_end))
1564 : {
1565 : /* enlarge output space if needed */
1566 447451 : while (array_idx + matchctx->npatterns * 2 + 1 > array_len)
1567 : {
1568 180 : array_len += array_len + 1; /* 2^n-1 => 2^(n+1)-1 */
1569 180 : if (array_len > MaxAllocSize / sizeof(int))
1570 0 : ereport(ERROR,
1571 : (errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
1572 : errmsg("too many regular expression matches")));
1573 180 : matchctx->match_locs = (int *) repalloc(matchctx->match_locs,
1574 : sizeof(int) * array_len);
1575 : }
1576 :
1577 : /* save this match's locations */
1578 447271 : if (use_subpatterns)
1579 : {
1580 : int i;
1581 :
1582 3972 : for (i = 1; i <= matchctx->npatterns; i++)
1583 : {
1584 2681 : int so = pmatch[i].rm_so;
1585 2681 : int eo = pmatch[i].rm_eo;
1586 :
1587 2681 : matchctx->match_locs[array_idx++] = so;
1588 2681 : matchctx->match_locs[array_idx++] = eo;
1589 2681 : if (so >= 0 && eo >= 0 && (eo - so) > maxlen)
1590 1722 : maxlen = (eo - so);
1591 : }
1592 : }
1593 : else
1594 : {
1595 445980 : int so = pmatch[0].rm_so;
1596 445980 : int eo = pmatch[0].rm_eo;
1597 :
1598 445980 : matchctx->match_locs[array_idx++] = so;
1599 445980 : matchctx->match_locs[array_idx++] = eo;
1600 445980 : if (so >= 0 && eo >= 0 && (eo - so) > maxlen)
1601 100452 : maxlen = (eo - so);
1602 : }
1603 447271 : matchctx->nmatches++;
1604 :
1605 : /*
1606 : * check length of unmatched portion between end of previous valid
1607 : * (nondegenerate, or degenerate but not ignored) match and start
1608 : * of current one
1609 : */
1610 447271 : if (fetching_unmatched &&
1611 445542 : pmatch[0].rm_so >= 0 &&
1612 445542 : (pmatch[0].rm_so - prev_valid_match_end) > maxlen)
1613 190450 : maxlen = (pmatch[0].rm_so - prev_valid_match_end);
1614 447271 : prev_valid_match_end = pmatch[0].rm_eo;
1615 : }
1616 447361 : prev_match_end = pmatch[0].rm_eo;
1617 :
1618 : /* if not glob, stop after one match */
1619 447361 : if (!re_flags->glob)
1620 1421 : break;
1621 :
1622 : /*
1623 : * Advance search position. Normally we start the next search at the
1624 : * end of the previous match; but if the match was of zero length, we
1625 : * have to advance by one character, or we'd just find the same match
1626 : * again.
1627 : */
1628 445940 : start_search = prev_match_end;
1629 445940 : if (pmatch[0].rm_so == pmatch[0].rm_eo)
1630 588 : start_search++;
1631 445940 : if (start_search > wide_len)
1632 33 : break;
1633 : }
1634 :
1635 : /*
1636 : * check length of unmatched portion between end of last match and end of
1637 : * input string
1638 : */
1639 102578 : if (fetching_unmatched &&
1640 100226 : (wide_len - prev_valid_match_end) > maxlen)
1641 35 : maxlen = (wide_len - prev_valid_match_end);
1642 :
1643 : /*
1644 : * Keep a note of the end position of the string for the benefit of
1645 : * splitting code.
1646 : */
1647 102578 : matchctx->match_locs[array_idx] = wide_len;
1648 :
1649 102578 : if (eml > 1)
1650 : {
1651 102578 : int64 maxsiz = eml * (int64) maxlen;
1652 : int conv_bufsiz;
1653 :
1654 : /*
1655 : * Make the conversion buffer large enough for any substring of
1656 : * interest.
1657 : *
1658 : * Worst case: assume we need the maximum size (maxlen*eml), but take
1659 : * advantage of the fact that the original string length in bytes is
1660 : * an upper bound on the byte length of any fetched substring (and we
1661 : * know that len+1 is safe to allocate because the varlena header is
1662 : * longer than 1 byte).
1663 : */
1664 102578 : if (maxsiz > orig_len)
1665 100409 : conv_bufsiz = orig_len + 1;
1666 : else
1667 2169 : conv_bufsiz = maxsiz + 1; /* safe since maxsiz < 2^30 */
1668 :
1669 102578 : matchctx->conv_buf = palloc(conv_bufsiz);
1670 102578 : matchctx->conv_bufsiz = conv_bufsiz;
1671 102578 : matchctx->wide_str = wide_str;
1672 : }
1673 : else
1674 : {
1675 : /* No need to keep the wide string if we're in a single-byte charset. */
1676 0 : pfree(wide_str);
1677 0 : matchctx->wide_str = NULL;
1678 0 : matchctx->conv_buf = NULL;
1679 0 : matchctx->conv_bufsiz = 0;
1680 : }
1681 :
1682 : /* Clean up temp storage */
1683 102578 : pfree(pmatch);
1684 :
1685 102578 : return matchctx;
1686 : }
1687 :
1688 : /*
1689 : * build_regexp_match_result - build output array for current match
1690 : */
1691 : static ArrayType *
1692 1531 : build_regexp_match_result(regexp_matches_ctx *matchctx)
1693 : {
1694 1531 : char *buf = matchctx->conv_buf;
1695 1531 : Datum *elems = matchctx->elems;
1696 1531 : bool *nulls = matchctx->nulls;
1697 : int dims[1];
1698 : int lbs[1];
1699 : int loc;
1700 : int i;
1701 :
1702 : /* Extract matching substrings from the original string */
1703 1531 : loc = matchctx->next_match * matchctx->npatterns * 2;
1704 4317 : for (i = 0; i < matchctx->npatterns; i++)
1705 : {
1706 2786 : int so = matchctx->match_locs[loc++];
1707 2786 : int eo = matchctx->match_locs[loc++];
1708 :
1709 2786 : if (so < 0 || eo < 0)
1710 : {
1711 3 : elems[i] = (Datum) 0;
1712 3 : nulls[i] = true;
1713 : }
1714 2783 : else if (buf)
1715 : {
1716 2783 : int len = pg_wchar2mb_with_len(matchctx->wide_str + so,
1717 : buf,
1718 : eo - so);
1719 :
1720 : Assert(len < matchctx->conv_bufsiz);
1721 2783 : elems[i] = PointerGetDatum(cstring_to_text_with_len(buf, len));
1722 2783 : nulls[i] = false;
1723 : }
1724 : else
1725 : {
1726 0 : elems[i] = DirectFunctionCall3(text_substr,
1727 : PointerGetDatum(matchctx->orig_str),
1728 : Int32GetDatum(so + 1),
1729 : Int32GetDatum(eo - so));
1730 0 : nulls[i] = false;
1731 : }
1732 : }
1733 :
1734 : /* And form an array */
1735 1531 : dims[0] = matchctx->npatterns;
1736 1531 : lbs[0] = 1;
1737 : /* XXX: this hardcodes assumptions about the text type */
1738 1531 : return construct_md_array(elems, nulls, 1, dims, lbs,
1739 : TEXTOID, -1, false, TYPALIGN_INT);
1740 : }
1741 :
1742 : /*
1743 : * regexp_split_to_table()
1744 : * Split the string at matches of the pattern, returning the
1745 : * split-out substrings as a table.
1746 : */
1747 : Datum
1748 311 : regexp_split_to_table(PG_FUNCTION_ARGS)
1749 : {
1750 : FuncCallContext *funcctx;
1751 : regexp_matches_ctx *splitctx;
1752 :
1753 311 : if (SRF_IS_FIRSTCALL())
1754 : {
1755 26 : text *pattern = PG_GETARG_TEXT_PP(1);
1756 26 : text *flags = PG_GETARG_TEXT_PP_IF_EXISTS(2);
1757 : pg_re_flags re_flags;
1758 : MemoryContext oldcontext;
1759 :
1760 26 : funcctx = SRF_FIRSTCALL_INIT();
1761 26 : oldcontext = MemoryContextSwitchTo(funcctx->multi_call_memory_ctx);
1762 :
1763 : /* Determine options */
1764 26 : parse_re_flags(&re_flags, flags);
1765 : /* User mustn't specify 'g' */
1766 23 : if (re_flags.glob)
1767 3 : ereport(ERROR,
1768 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
1769 : /* translator: %s is a SQL function name */
1770 : errmsg("%s does not support the \"global\" option",
1771 : "regexp_split_to_table()")));
1772 : /* But we find all the matches anyway */
1773 20 : re_flags.glob = true;
1774 :
1775 : /* be sure to copy the input string into the multi-call ctx */
1776 20 : splitctx = setup_regexp_matches(PG_GETARG_TEXT_P_COPY(0), pattern,
1777 : &re_flags, 0,
1778 : PG_GET_COLLATION(),
1779 : false, true, true);
1780 :
1781 20 : MemoryContextSwitchTo(oldcontext);
1782 20 : funcctx->user_fctx = splitctx;
1783 : }
1784 :
1785 305 : funcctx = SRF_PERCALL_SETUP();
1786 305 : splitctx = (regexp_matches_ctx *) funcctx->user_fctx;
1787 :
1788 305 : if (splitctx->next_match <= splitctx->nmatches)
1789 : {
1790 285 : Datum result = build_regexp_split_result(splitctx);
1791 :
1792 285 : splitctx->next_match++;
1793 285 : SRF_RETURN_NEXT(funcctx, result);
1794 : }
1795 :
1796 20 : SRF_RETURN_DONE(funcctx);
1797 : }
1798 :
1799 : /* This is separate to keep the opr_sanity regression test from complaining */
1800 : Datum
1801 276 : regexp_split_to_table_no_flags(PG_FUNCTION_ARGS)
1802 : {
1803 276 : return regexp_split_to_table(fcinfo);
1804 : }
1805 :
1806 : /*
1807 : * regexp_split_to_array()
1808 : * Split the string at matches of the pattern, returning the
1809 : * split-out substrings as an array.
1810 : */
1811 : Datum
1812 100212 : regexp_split_to_array(PG_FUNCTION_ARGS)
1813 : {
1814 100212 : ArrayBuildState *astate = NULL;
1815 : pg_re_flags re_flags;
1816 : regexp_matches_ctx *splitctx;
1817 :
1818 : /* Determine options */
1819 100212 : parse_re_flags(&re_flags, PG_GETARG_TEXT_PP_IF_EXISTS(2));
1820 : /* User mustn't specify 'g' */
1821 100209 : if (re_flags.glob)
1822 3 : ereport(ERROR,
1823 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
1824 : /* translator: %s is a SQL function name */
1825 : errmsg("%s does not support the \"global\" option",
1826 : "regexp_split_to_array()")));
1827 : /* But we find all the matches anyway */
1828 100206 : re_flags.glob = true;
1829 :
1830 100206 : splitctx = setup_regexp_matches(PG_GETARG_TEXT_PP(0),
1831 100206 : PG_GETARG_TEXT_PP(1),
1832 : &re_flags, 0,
1833 : PG_GET_COLLATION(),
1834 : false, true, true);
1835 :
1836 645689 : while (splitctx->next_match <= splitctx->nmatches)
1837 : {
1838 545483 : astate = accumArrayResult(astate,
1839 : build_regexp_split_result(splitctx),
1840 : false,
1841 : TEXTOID,
1842 : CurrentMemoryContext);
1843 545483 : splitctx->next_match++;
1844 : }
1845 :
1846 100206 : PG_RETURN_DATUM(makeArrayResult(astate, CurrentMemoryContext));
1847 : }
1848 :
1849 : /* This is separate to keep the opr_sanity regression test from complaining */
1850 : Datum
1851 100191 : regexp_split_to_array_no_flags(PG_FUNCTION_ARGS)
1852 : {
1853 100191 : return regexp_split_to_array(fcinfo);
1854 : }
1855 :
1856 : /*
1857 : * build_regexp_split_result - build output string for current match
1858 : *
1859 : * We return the string between the current match and the previous one,
1860 : * or the string after the last match when next_match == nmatches.
1861 : */
1862 : static Datum
1863 545768 : build_regexp_split_result(regexp_matches_ctx *splitctx)
1864 : {
1865 545768 : char *buf = splitctx->conv_buf;
1866 : int startpos;
1867 : int endpos;
1868 :
1869 545768 : if (splitctx->next_match > 0)
1870 445542 : startpos = splitctx->match_locs[splitctx->next_match * 2 - 1];
1871 : else
1872 100226 : startpos = 0;
1873 545768 : if (startpos < 0)
1874 0 : elog(ERROR, "invalid match ending position");
1875 :
1876 545768 : endpos = splitctx->match_locs[splitctx->next_match * 2];
1877 545768 : if (endpos < startpos)
1878 0 : elog(ERROR, "invalid match starting position");
1879 :
1880 545768 : if (buf)
1881 : {
1882 : int len;
1883 :
1884 545768 : len = pg_wchar2mb_with_len(splitctx->wide_str + startpos,
1885 : buf,
1886 : endpos - startpos);
1887 : Assert(len < splitctx->conv_bufsiz);
1888 545768 : return PointerGetDatum(cstring_to_text_with_len(buf, len));
1889 : }
1890 : else
1891 : {
1892 0 : return DirectFunctionCall3(text_substr,
1893 : PointerGetDatum(splitctx->orig_str),
1894 : Int32GetDatum(startpos + 1),
1895 : Int32GetDatum(endpos - startpos));
1896 : }
1897 : }
1898 :
1899 : /*
1900 : * regexp_substr()
1901 : * Return the substring that matches a regular expression pattern
1902 : */
1903 : Datum
1904 54 : regexp_substr(PG_FUNCTION_ARGS)
1905 : {
1906 54 : text *str = PG_GETARG_TEXT_PP(0);
1907 54 : text *pattern = PG_GETARG_TEXT_PP(1);
1908 54 : int start = 1;
1909 54 : int n = 1;
1910 54 : text *flags = PG_GETARG_TEXT_PP_IF_EXISTS(4);
1911 54 : int subexpr = 0;
1912 : int so,
1913 : eo,
1914 : pos;
1915 : pg_re_flags re_flags;
1916 : regexp_matches_ctx *matchctx;
1917 :
1918 : /* Collect optional parameters */
1919 54 : if (PG_NARGS() > 2)
1920 : {
1921 45 : start = PG_GETARG_INT32(2);
1922 45 : if (start <= 0)
1923 3 : ereport(ERROR,
1924 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
1925 : errmsg("invalid value for parameter \"%s\": %d",
1926 : "start", start)));
1927 : }
1928 51 : if (PG_NARGS() > 3)
1929 : {
1930 39 : n = PG_GETARG_INT32(3);
1931 39 : if (n <= 0)
1932 3 : ereport(ERROR,
1933 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
1934 : errmsg("invalid value for parameter \"%s\": %d",
1935 : "n", n)));
1936 : }
1937 48 : if (PG_NARGS() > 5)
1938 : {
1939 24 : subexpr = PG_GETARG_INT32(5);
1940 24 : if (subexpr < 0)
1941 3 : ereport(ERROR,
1942 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
1943 : errmsg("invalid value for parameter \"%s\": %d",
1944 : "subexpr", subexpr)));
1945 : }
1946 :
1947 : /* Determine options */
1948 45 : parse_re_flags(&re_flags, flags);
1949 : /* User mustn't specify 'g' */
1950 45 : if (re_flags.glob)
1951 3 : ereport(ERROR,
1952 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
1953 : /* translator: %s is a SQL function name */
1954 : errmsg("%s does not support the \"global\" option",
1955 : "regexp_substr()")));
1956 : /* But we find all the matches anyway */
1957 42 : re_flags.glob = true;
1958 :
1959 : /* Do the matching */
1960 42 : matchctx = setup_regexp_matches(str, pattern, &re_flags, start - 1,
1961 : PG_GET_COLLATION(),
1962 : (subexpr > 0), /* need submatches? */
1963 : false, false);
1964 :
1965 : /* When n exceeds matches return NULL (includes case of no matches) */
1966 42 : if (n > matchctx->nmatches)
1967 6 : PG_RETURN_NULL();
1968 :
1969 : /* When subexpr exceeds number of subexpressions return NULL */
1970 36 : if (subexpr > matchctx->npatterns)
1971 3 : PG_RETURN_NULL();
1972 :
1973 : /* Select the appropriate match position to return */
1974 33 : pos = (n - 1) * matchctx->npatterns;
1975 33 : if (subexpr > 0)
1976 15 : pos += subexpr - 1;
1977 33 : pos *= 2;
1978 33 : so = matchctx->match_locs[pos];
1979 33 : eo = matchctx->match_locs[pos + 1];
1980 :
1981 33 : if (so < 0 || eo < 0)
1982 3 : PG_RETURN_NULL(); /* unidentifiable location */
1983 :
1984 30 : PG_RETURN_DATUM(DirectFunctionCall3(text_substr,
1985 : PointerGetDatum(matchctx->orig_str),
1986 : Int32GetDatum(so + 1),
1987 : Int32GetDatum(eo - so)));
1988 : }
1989 :
1990 : /* This is separate to keep the opr_sanity regression test from complaining */
1991 : Datum
1992 9 : regexp_substr_no_start(PG_FUNCTION_ARGS)
1993 : {
1994 9 : return regexp_substr(fcinfo);
1995 : }
1996 :
1997 : /* This is separate to keep the opr_sanity regression test from complaining */
1998 : Datum
1999 3 : regexp_substr_no_n(PG_FUNCTION_ARGS)
2000 : {
2001 3 : return regexp_substr(fcinfo);
2002 : }
2003 :
2004 : /* This is separate to keep the opr_sanity regression test from complaining */
2005 : Datum
2006 12 : regexp_substr_no_flags(PG_FUNCTION_ARGS)
2007 : {
2008 12 : return regexp_substr(fcinfo);
2009 : }
2010 :
2011 : /* This is separate to keep the opr_sanity regression test from complaining */
2012 : Datum
2013 6 : regexp_substr_no_subexpr(PG_FUNCTION_ARGS)
2014 : {
2015 6 : return regexp_substr(fcinfo);
2016 : }
2017 :
2018 : /*
2019 : * regexp_fixed_prefix - extract fixed prefix, if any, for a regexp
2020 : *
2021 : * The result is NULL if there is no fixed prefix, else a palloc'd string.
2022 : * If it is an exact match, not just a prefix, *exact is returned as true.
2023 : */
2024 : char *
2025 8489 : regexp_fixed_prefix(text *text_re, bool case_insensitive, Oid collation,
2026 : bool *exact)
2027 : {
2028 : char *result;
2029 : regex_t *re;
2030 : int cflags;
2031 : int re_result;
2032 : pg_wchar *str;
2033 : size_t slen;
2034 : size_t maxlen;
2035 : char errMsg[100];
2036 :
2037 8489 : *exact = false; /* default result */
2038 :
2039 : /* Compile RE */
2040 8489 : cflags = REG_ADVANCED;
2041 8489 : if (case_insensitive)
2042 31 : cflags |= REG_ICASE;
2043 :
2044 8489 : re = RE_compile_and_cache(text_re, cflags | REG_NOSUB, collation);
2045 :
2046 : /* Examine it to see if there's a fixed prefix */
2047 8477 : re_result = pg_regprefix(re, &str, &slen);
2048 :
2049 8477 : switch (re_result)
2050 : {
2051 385 : case REG_NOMATCH:
2052 385 : return NULL;
2053 :
2054 1579 : case REG_PREFIX:
2055 : /* continue with wchar conversion */
2056 1579 : break;
2057 :
2058 6513 : case REG_EXACT:
2059 6513 : *exact = true;
2060 : /* continue with wchar conversion */
2061 6513 : break;
2062 :
2063 0 : default:
2064 : /* re failed??? */
2065 0 : pg_regerror(re_result, re, errMsg, sizeof(errMsg));
2066 0 : ereport(ERROR,
2067 : (errcode(ERRCODE_INVALID_REGULAR_EXPRESSION),
2068 : errmsg("regular expression failed: %s", errMsg)));
2069 : break;
2070 : }
2071 :
2072 : /* Convert pg_wchar result back to database encoding */
2073 8092 : maxlen = pg_database_encoding_max_length() * slen + 1;
2074 8092 : result = (char *) palloc(maxlen);
2075 8092 : slen = pg_wchar2mb_with_len(str, result, slen);
2076 : Assert(slen < maxlen);
2077 :
2078 8092 : pfree(str);
2079 :
2080 8092 : return result;
2081 : }
|
