1/*-------------------------------------------------------------------------
2 *
3 * regexp.c
4 * Postgres' interface to the regular expression package.
5 *
6 * Portions Copyright (c) 1996-2019, 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 "miscadmin.h"
35#include "regex/regex.h"
36#include "utils/array.h"
37#include "utils/builtins.h"
38#include "utils/memutils.h"
39#include "utils/varlena.h"
40
41#define PG_GETARG_TEXT_PP_IF_EXISTS(_n) \
42 (PG_NARGS() > (_n) ? PG_GETARG_TEXT_PP(_n) : NULL)
43
44
45/* all the options of interest for regex functions */
46typedef struct pg_re_flags
47{
48 int cflags; /* compile flags for Spencer's regex code */
49 bool glob; /* do it globally (for each occurrence) */
50} pg_re_flags;
51
52/* cross-call state for regexp_match and regexp_split functions */
53typedef struct regexp_matches_ctx
54{
55 text *orig_str; /* data string in original TEXT form */
56 int nmatches; /* number of places where pattern matched */
57 int npatterns; /* number of capturing subpatterns */
58 /* We store start char index and end+1 char index for each match */
59 /* so the number of entries in match_locs is nmatches * npatterns * 2 */
60 int *match_locs; /* 0-based character indexes */
61 int next_match; /* 0-based index of next match to process */
62 /* workspace for build_regexp_match_result() */
63 Datum *elems; /* has npatterns elements */
64 bool *nulls; /* has npatterns elements */
65 pg_wchar *wide_str; /* wide-char version of original string */
66 char *conv_buf; /* conversion buffer */
67 int conv_bufsiz; /* size thereof */
68} regexp_matches_ctx;
69
70/*
71 * We cache precompiled regular expressions using a "self organizing list"
72 * structure, in which recently-used items tend to be near the front.
73 * Whenever we use an entry, it's moved up to the front of the list.
74 * Over time, an item's average position corresponds to its frequency of use.
75 *
76 * When we first create an entry, it's inserted at the front of
77 * the array, dropping the entry at the end of the array if necessary to
78 * make room. (This might seem to be weighting the new entry too heavily,
79 * but if we insert new entries further back, we'll be unable to adjust to
80 * a sudden shift in the query mix where we are presented with MAX_CACHED_RES
81 * never-before-seen items used circularly. We ought to be able to handle
82 * that case, so we have to insert at the front.)
83 *
84 * Knuth mentions a variant strategy in which a used item is moved up just
85 * one place in the list. Although he says this uses fewer comparisons on
86 * average, it seems not to adapt very well to the situation where you have
87 * both some reusable patterns and a steady stream of non-reusable patterns.
88 * A reusable pattern that isn't used at least as often as non-reusable
89 * patterns are seen will "fail to keep up" and will drop off the end of the
90 * cache. With move-to-front, a reusable pattern is guaranteed to stay in
91 * the cache as long as it's used at least once in every MAX_CACHED_RES uses.
92 */
93
94/* this is the maximum number of cached regular expressions */
95#ifndef MAX_CACHED_RES
96#define MAX_CACHED_RES 32
97#endif
98
99/* this structure describes one cached regular expression */
100typedef struct cached_re_str
101{
102 char *cre_pat; /* original RE (not null terminated!) */
103 int cre_pat_len; /* length of original RE, in bytes */
104 int cre_flags; /* compile flags: extended,icase etc */
105 Oid cre_collation; /* collation to use */
106 regex_t cre_re; /* the compiled regular expression */
107} cached_re_str;
108
109static int num_res = 0; /* # of cached re's */
110static cached_re_str re_array[MAX_CACHED_RES]; /* cached re's */
111
112
113/* Local functions */
114static regexp_matches_ctx *setup_regexp_matches(text *orig_str, text *pattern,
115 pg_re_flags *flags,
116 Oid collation,
117 bool use_subpatterns,
118 bool ignore_degenerate,
119 bool fetching_unmatched);
120static ArrayType *build_regexp_match_result(regexp_matches_ctx *matchctx);
121static Datum build_regexp_split_result(regexp_matches_ctx *splitctx);
122
123
124/*
125 * RE_compile_and_cache - compile a RE, caching if possible
126 *
127 * Returns regex_t *
128 *
129 * text_re --- the pattern, expressed as a TEXT object
130 * cflags --- compile options for the pattern
131 * collation --- collation to use for LC_CTYPE-dependent behavior
132 *
133 * Pattern is given in the database encoding. We internally convert to
134 * an array of pg_wchar, which is what Spencer's regex package wants.
135 */
136regex_t *
137RE_compile_and_cache(text *text_re, int cflags, Oid collation)
138{
139 int text_re_len = VARSIZE_ANY_EXHDR(text_re);
140 char *text_re_val = VARDATA_ANY(text_re);
141 pg_wchar *pattern;
142 int pattern_len;
143 int i;
144 int regcomp_result;
145 cached_re_str re_temp;
146 char errMsg[100];
147
148 /*
149 * Look for a match among previously compiled REs. Since the data
150 * structure is self-organizing with most-used entries at the front, our
151 * search strategy can just be to scan from the front.
152 */
153 for (i = 0; i < num_res; i++)
154 {
155 if (re_array[i].cre_pat_len == text_re_len &&
156 re_array[i].cre_flags == cflags &&
157 re_array[i].cre_collation == collation &&
158 memcmp(re_array[i].cre_pat, text_re_val, text_re_len) == 0)
159 {
160 /*
161 * Found a match; move it to front if not there already.
162 */
163 if (i > 0)
164 {
165 re_temp = re_array[i];
166 memmove(&re_array[1], &re_array[0], i * sizeof(cached_re_str));
167 re_array[0] = re_temp;
168 }
169
170 return &re_array[0].cre_re;
171 }
172 }
173
174 /*
175 * Couldn't find it, so try to compile the new RE. To avoid leaking
176 * resources on failure, we build into the re_temp local.
177 */
178
179 /* Convert pattern string to wide characters */
180 pattern = (pg_wchar *) palloc((text_re_len + 1) * sizeof(pg_wchar));
181 pattern_len = pg_mb2wchar_with_len(text_re_val,
182 pattern,
183 text_re_len);
184
185 regcomp_result = pg_regcomp(&re_temp.cre_re,
186 pattern,
187 pattern_len,
188 cflags,
189 collation);
190
191 pfree(pattern);
192
193 if (regcomp_result != REG_OKAY)
194 {
195 /* re didn't compile (no need for pg_regfree, if so) */
196
197 /*
198 * Here and in other places in this file, do CHECK_FOR_INTERRUPTS
199 * before reporting a regex error. This is so that if the regex
200 * library aborts and returns REG_CANCEL, we don't print an error
201 * message that implies the regex was invalid.
202 */
203 CHECK_FOR_INTERRUPTS();
204
205 pg_regerror(regcomp_result, &re_temp.cre_re, errMsg, sizeof(errMsg));
206 ereport(ERROR,
207 (errcode(ERRCODE_INVALID_REGULAR_EXPRESSION),
208 errmsg("invalid regular expression: %s", errMsg)));
209 }
210
211 /*
212 * We use malloc/free for the cre_pat field because the storage has to
213 * persist across transactions, and because we want to get control back on
214 * out-of-memory. The Max() is because some malloc implementations return
215 * NULL for malloc(0).
216 */
217 re_temp.cre_pat = malloc(Max(text_re_len, 1));
218 if (re_temp.cre_pat == NULL)
219 {
220 pg_regfree(&re_temp.cre_re);
221 ereport(ERROR,
222 (errcode(ERRCODE_OUT_OF_MEMORY),
223 errmsg("out of memory")));
224 }
225 memcpy(re_temp.cre_pat, text_re_val, text_re_len);
226 re_temp.cre_pat_len = text_re_len;
227 re_temp.cre_flags = cflags;
228 re_temp.cre_collation = collation;
229
230 /*
231 * Okay, we have a valid new item in re_temp; insert it into the storage
232 * array. Discard last entry if needed.
233 */
234 if (num_res >= MAX_CACHED_RES)
235 {
236 --num_res;
237 Assert(num_res < MAX_CACHED_RES);
238 pg_regfree(&re_array[num_res].cre_re);
239 free(re_array[num_res].cre_pat);
240 }
241
242 if (num_res > 0)
243 memmove(&re_array[1], &re_array[0], num_res * sizeof(cached_re_str));
244
245 re_array[0] = re_temp;
246 num_res++;
247
248 return &re_array[0].cre_re;
249}
250
251/*
252 * RE_wchar_execute - execute a RE on pg_wchar data
253 *
254 * Returns true on match, false on no match
255 *
256 * re --- the compiled pattern as returned by RE_compile_and_cache
257 * data --- the data to match against (need not be null-terminated)
258 * data_len --- the length of the data string
259 * start_search -- the offset in the data to start searching
260 * nmatch, pmatch --- optional return area for match details
261 *
262 * Data is given as array of pg_wchar which is what Spencer's regex package
263 * wants.
264 */
265static bool
266RE_wchar_execute(regex_t *re, pg_wchar *data, int data_len,
267 int start_search, int nmatch, regmatch_t *pmatch)
268{
269 int regexec_result;
270 char errMsg[100];
271
272 /* Perform RE match and return result */
273 regexec_result = pg_regexec(re,
274 data,
275 data_len,
276 start_search,
277 NULL, /* no details */
278 nmatch,
279 pmatch,
280 0);
281
282 if (regexec_result != REG_OKAY && regexec_result != REG_NOMATCH)
283 {
284 /* re failed??? */
285 CHECK_FOR_INTERRUPTS();
286 pg_regerror(regexec_result, re, errMsg, sizeof(errMsg));
287 ereport(ERROR,
288 (errcode(ERRCODE_INVALID_REGULAR_EXPRESSION),
289 errmsg("regular expression failed: %s", errMsg)));
290 }
291
292 return (regexec_result == REG_OKAY);
293}
294
295/*
296 * RE_execute - execute a RE
297 *
298 * Returns true on match, false on no match
299 *
300 * re --- the compiled pattern as returned by RE_compile_and_cache
301 * dat --- the data to match against (need not be null-terminated)
302 * dat_len --- the length of the data string
303 * nmatch, pmatch --- optional return area for match details
304 *
305 * Data is given in the database encoding. We internally
306 * convert to array of pg_wchar which is what Spencer's regex package wants.
307 */
308static bool
309RE_execute(regex_t *re, char *dat, int dat_len,
310 int nmatch, regmatch_t *pmatch)
311{
312 pg_wchar *data;
313 int data_len;
314 bool match;
315
316 /* Convert data string to wide characters */
317 data = (pg_wchar *) palloc((dat_len + 1) * sizeof(pg_wchar));
318 data_len = pg_mb2wchar_with_len(dat, data, dat_len);
319
320 /* Perform RE match and return result */
321 match = RE_wchar_execute(re, data, data_len, 0, nmatch, pmatch);
322
323 pfree(data);
324 return match;
325}
326
327/*
328 * RE_compile_and_execute - compile and execute a RE
329 *
330 * Returns true on match, false on no match
331 *
332 * text_re --- the pattern, expressed as a TEXT object
333 * dat --- the data to match against (need not be null-terminated)
334 * dat_len --- the length of the data string
335 * cflags --- compile options for the pattern
336 * collation --- collation to use for LC_CTYPE-dependent behavior
337 * nmatch, pmatch --- optional return area for match details
338 *
339 * Both pattern and data are given in the database encoding. We internally
340 * convert to array of pg_wchar which is what Spencer's regex package wants.
341 */
342bool
343RE_compile_and_execute(text *text_re, char *dat, int dat_len,
344 int cflags, Oid collation,
345 int nmatch, regmatch_t *pmatch)
346{
347 regex_t *re;
348
349 /* Compile RE */
350 re = RE_compile_and_cache(text_re, cflags, collation);
351
352 return RE_execute(re, dat, dat_len, nmatch, pmatch);
353}
354
355
356/*
357 * parse_re_flags - parse the options argument of regexp_match and friends
358 *
359 * flags --- output argument, filled with desired options
360 * opts --- TEXT object, or NULL for defaults
361 *
362 * This accepts all the options allowed by any of the callers; callers that
363 * don't want some have to reject them after the fact.
364 */
365static void
366parse_re_flags(pg_re_flags *flags, text *opts)
367{
368 /* regex flavor is always folded into the compile flags */
369 flags->cflags = REG_ADVANCED;
370 flags->glob = false;
371
372 if (opts)
373 {
374 char *opt_p = VARDATA_ANY(opts);
375 int opt_len = VARSIZE_ANY_EXHDR(opts);
376 int i;
377
378 for (i = 0; i < opt_len; i++)
379 {
380 switch (opt_p[i])
381 {
382 case 'g':
383 flags->glob = true;
384 break;
385 case 'b': /* BREs (but why???) */
386 flags->cflags &= ~(REG_ADVANCED | REG_EXTENDED | REG_QUOTE);
387 break;
388 case 'c': /* case sensitive */
389 flags->cflags &= ~REG_ICASE;
390 break;
391 case 'e': /* plain EREs */
392 flags->cflags |= REG_EXTENDED;
393 flags->cflags &= ~(REG_ADVANCED | REG_QUOTE);
394 break;
395 case 'i': /* case insensitive */
396 flags->cflags |= REG_ICASE;
397 break;
398 case 'm': /* Perloid synonym for n */
399 case 'n': /* \n affects ^ $ . [^ */
400 flags->cflags |= REG_NEWLINE;
401 break;
402 case 'p': /* ~Perl, \n affects . [^ */
403 flags->cflags |= REG_NLSTOP;
404 flags->cflags &= ~REG_NLANCH;
405 break;
406 case 'q': /* literal string */
407 flags->cflags |= REG_QUOTE;
408 flags->cflags &= ~(REG_ADVANCED | REG_EXTENDED);
409 break;
410 case 's': /* single line, \n ordinary */
411 flags->cflags &= ~REG_NEWLINE;
412 break;
413 case 't': /* tight syntax */
414 flags->cflags &= ~REG_EXPANDED;
415 break;
416 case 'w': /* weird, \n affects ^ $ only */
417 flags->cflags &= ~REG_NLSTOP;
418 flags->cflags |= REG_NLANCH;
419 break;
420 case 'x': /* expanded syntax */
421 flags->cflags |= REG_EXPANDED;
422 break;
423 default:
424 ereport(ERROR,
425 (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
426 errmsg("invalid regular expression option: \"%c\"",
427 opt_p[i])));
428 break;
429 }
430 }
431 }
432}
433
434
435/*
436 * interface routines called by the function manager
437 */
438
439Datum
440nameregexeq(PG_FUNCTION_ARGS)
441{
442 Name n = PG_GETARG_NAME(0);
443 text *p = PG_GETARG_TEXT_PP(1);
444
445 PG_RETURN_BOOL(RE_compile_and_execute(p,
446 NameStr(*n),
447 strlen(NameStr(*n)),
448 REG_ADVANCED,
449 PG_GET_COLLATION(),
450 0, NULL));
451}
452
453Datum
454nameregexne(PG_FUNCTION_ARGS)
455{
456 Name n = PG_GETARG_NAME(0);
457 text *p = PG_GETARG_TEXT_PP(1);
458
459 PG_RETURN_BOOL(!RE_compile_and_execute(p,
460 NameStr(*n),
461 strlen(NameStr(*n)),
462 REG_ADVANCED,
463 PG_GET_COLLATION(),
464 0, NULL));
465}
466
467Datum
468textregexeq(PG_FUNCTION_ARGS)
469{
470 text *s = PG_GETARG_TEXT_PP(0);
471 text *p = PG_GETARG_TEXT_PP(1);
472
473 PG_RETURN_BOOL(RE_compile_and_execute(p,
474 VARDATA_ANY(s),
475 VARSIZE_ANY_EXHDR(s),
476 REG_ADVANCED,
477 PG_GET_COLLATION(),
478 0, NULL));
479}
480
481Datum
482textregexne(PG_FUNCTION_ARGS)
483{
484 text *s = PG_GETARG_TEXT_PP(0);
485 text *p = PG_GETARG_TEXT_PP(1);
486
487 PG_RETURN_BOOL(!RE_compile_and_execute(p,
488 VARDATA_ANY(s),
489 VARSIZE_ANY_EXHDR(s),
490 REG_ADVANCED,
491 PG_GET_COLLATION(),
492 0, NULL));
493}
494
495
496/*
497 * routines that use the regexp stuff, but ignore the case.
498 * for this, we use the REG_ICASE flag to pg_regcomp
499 */
500
501
502Datum
503nameicregexeq(PG_FUNCTION_ARGS)
504{
505 Name n = PG_GETARG_NAME(0);
506 text *p = PG_GETARG_TEXT_PP(1);
507
508 PG_RETURN_BOOL(RE_compile_and_execute(p,
509 NameStr(*n),
510 strlen(NameStr(*n)),
511 REG_ADVANCED | REG_ICASE,
512 PG_GET_COLLATION(),
513 0, NULL));
514}
515
516Datum
517nameicregexne(PG_FUNCTION_ARGS)
518{
519 Name n = PG_GETARG_NAME(0);
520 text *p = PG_GETARG_TEXT_PP(1);
521
522 PG_RETURN_BOOL(!RE_compile_and_execute(p,
523 NameStr(*n),
524 strlen(NameStr(*n)),
525 REG_ADVANCED | REG_ICASE,
526 PG_GET_COLLATION(),
527 0, NULL));
528}
529
530Datum
531texticregexeq(PG_FUNCTION_ARGS)
532{
533 text *s = PG_GETARG_TEXT_PP(0);
534 text *p = PG_GETARG_TEXT_PP(1);
535
536 PG_RETURN_BOOL(RE_compile_and_execute(p,
537 VARDATA_ANY(s),
538 VARSIZE_ANY_EXHDR(s),
539 REG_ADVANCED | REG_ICASE,
540 PG_GET_COLLATION(),
541 0, NULL));
542}
543
544Datum
545texticregexne(PG_FUNCTION_ARGS)
546{
547 text *s = PG_GETARG_TEXT_PP(0);
548 text *p = PG_GETARG_TEXT_PP(1);
549
550 PG_RETURN_BOOL(!RE_compile_and_execute(p,
551 VARDATA_ANY(s),
552 VARSIZE_ANY_EXHDR(s),
553 REG_ADVANCED | REG_ICASE,
554 PG_GET_COLLATION(),
555 0, NULL));
556}
557
558
559/*
560 * textregexsubstr()
561 * Return a substring matched by a regular expression.
562 */
563Datum
564textregexsubstr(PG_FUNCTION_ARGS)
565{
566 text *s = PG_GETARG_TEXT_PP(0);
567 text *p = PG_GETARG_TEXT_PP(1);
568 regex_t *re;
569 regmatch_t pmatch[2];
570 int so,
571 eo;
572
573 /* Compile RE */
574 re = RE_compile_and_cache(p, REG_ADVANCED, PG_GET_COLLATION());
575
576 /*
577 * We pass two regmatch_t structs to get info about the overall match and
578 * the match for the first parenthesized subexpression (if any). If there
579 * is a parenthesized subexpression, we return what it matched; else
580 * return what the whole regexp matched.
581 */
582 if (!RE_execute(re,
583 VARDATA_ANY(s), VARSIZE_ANY_EXHDR(s),
584 2, pmatch))
585 PG_RETURN_NULL(); /* definitely no match */
586
587 if (re->re_nsub > 0)
588 {
589 /* has parenthesized subexpressions, use the first one */
590 so = pmatch[1].rm_so;
591 eo = pmatch[1].rm_eo;
592 }
593 else
594 {
595 /* no parenthesized subexpression, use whole match */
596 so = pmatch[0].rm_so;
597 eo = pmatch[0].rm_eo;
598 }
599
600 /*
601 * It is possible to have a match to the whole pattern but no match for a
602 * subexpression; for example 'foo(bar)?' is considered to match 'foo' but
603 * there is no subexpression match. So this extra test for match failure
604 * is not redundant.
605 */
606 if (so < 0 || eo < 0)
607 PG_RETURN_NULL();
608
609 return DirectFunctionCall3(text_substr,
610 PointerGetDatum(s),
611 Int32GetDatum(so + 1),
612 Int32GetDatum(eo - so));
613}
614
615/*
616 * textregexreplace_noopt()
617 * Return a string matched by a regular expression, with replacement.
618 *
619 * This version doesn't have an option argument: we default to case
620 * sensitive match, replace the first instance only.
621 */
622Datum
623textregexreplace_noopt(PG_FUNCTION_ARGS)
624{
625 text *s = PG_GETARG_TEXT_PP(0);
626 text *p = PG_GETARG_TEXT_PP(1);
627 text *r = PG_GETARG_TEXT_PP(2);
628 regex_t *re;
629
630 re = RE_compile_and_cache(p, REG_ADVANCED, PG_GET_COLLATION());
631
632 PG_RETURN_TEXT_P(replace_text_regexp(s, (void *) re, r, false));
633}
634
635/*
636 * textregexreplace()
637 * Return a string matched by a regular expression, with replacement.
638 */
639Datum
640textregexreplace(PG_FUNCTION_ARGS)
641{
642 text *s = PG_GETARG_TEXT_PP(0);
643 text *p = PG_GETARG_TEXT_PP(1);
644 text *r = PG_GETARG_TEXT_PP(2);
645 text *opt = PG_GETARG_TEXT_PP(3);
646 regex_t *re;
647 pg_re_flags flags;
648
649 parse_re_flags(&flags, opt);
650
651 re = RE_compile_and_cache(p, flags.cflags, PG_GET_COLLATION());
652
653 PG_RETURN_TEXT_P(replace_text_regexp(s, (void *) re, r, flags.glob));
654}
655
656/*
657 * similar_escape()
658 * Convert a SQL:2008 regexp pattern to POSIX style, so it can be used by
659 * our regexp engine.
660 */
661Datum
662similar_escape(PG_FUNCTION_ARGS)
663{
664 text *pat_text;
665 text *esc_text;
666 text *result;
667 char *p,
668 *e,
669 *r;
670 int plen,
671 elen;
672 bool afterescape = false;
673 bool incharclass = false;
674 int nquotes = 0;
675
676 /* This function is not strict, so must test explicitly */
677 if (PG_ARGISNULL(0))
678 PG_RETURN_NULL();
679 pat_text = PG_GETARG_TEXT_PP(0);
680 p = VARDATA_ANY(pat_text);
681 plen = VARSIZE_ANY_EXHDR(pat_text);
682 if (PG_ARGISNULL(1))
683 {
684 /* No ESCAPE clause provided; default to backslash as escape */
685 e = "\\";
686 elen = 1;
687 }
688 else
689 {
690 esc_text = PG_GETARG_TEXT_PP(1);
691 e = VARDATA_ANY(esc_text);
692 elen = VARSIZE_ANY_EXHDR(esc_text);
693 if (elen == 0)
694 e = NULL; /* no escape character */
695 else
696 {
697 int escape_mblen = pg_mbstrlen_with_len(e, elen);
698
699 if (escape_mblen > 1)
700 ereport(ERROR,
701 (errcode(ERRCODE_INVALID_ESCAPE_SEQUENCE),
702 errmsg("invalid escape string"),
703 errhint("Escape string must be empty or one character.")));
704 }
705 }
706
707 /*----------
708 * We surround the transformed input string with
709 * ^(?: ... )$
710 * which requires some explanation. We need "^" and "$" to force
711 * the pattern to match the entire input string as per the SQL spec.
712 * The "(?:" and ")" are a non-capturing set of parens; we have to have
713 * parens in case the string contains "|", else the "^" and "$" will
714 * be bound into the first and last alternatives which is not what we
715 * want, and the parens must be non capturing because we don't want them
716 * to count when selecting output for SUBSTRING.
717 *
718 * When the pattern is divided into three parts by escape-double-quotes,
719 * what we emit is
720 * ^(?:part1){1,1}?(part2){1,1}(?:part3)$
721 * which requires even more explanation. The "{1,1}?" on part1 makes it
722 * non-greedy so that it will match the smallest possible amount of text
723 * not the largest, as required by SQL. The plain parens around part2
724 * are capturing parens so that that part is what controls the result of
725 * SUBSTRING. The "{1,1}" forces part2 to be greedy, so that it matches
726 * the largest possible amount of text; hence part3 must match the
727 * smallest amount of text, as required by SQL. We don't need an explicit
728 * greediness marker on part3. Note that this also confines the effects
729 * of any "|" characters to the respective part, which is what we want.
730 *
731 * The SQL spec says that SUBSTRING's pattern must contain exactly two
732 * escape-double-quotes, but we only complain if there's more than two.
733 * With none, we act as though part1 and part3 are empty; with one, we
734 * act as though part3 is empty. Both behaviors fall out of omitting
735 * the relevant part separators in the above expansion. If the result
736 * of this function is used in a plain regexp match (SIMILAR TO), the
737 * escape-double-quotes have no effect on the match behavior.
738 *----------
739 */
740
741 /*
742 * We need room for the prefix/postfix and part separators, plus as many
743 * as 3 output bytes per input byte; since the input is at most 1GB this
744 * can't overflow size_t.
745 */
746 result = (text *) palloc(VARHDRSZ + 23 + 3 * (size_t) plen);
747 r = VARDATA(result);
748
749 *r++ = '^';
750 *r++ = '(';
751 *r++ = '?';
752 *r++ = ':';
753
754 while (plen > 0)
755 {
756 char pchar = *p;
757
758 /*
759 * If both the escape character and the current character from the
760 * pattern are multi-byte, we need to take the slow path.
761 *
762 * But if one of them is single-byte, we can process the pattern one
763 * byte at a time, ignoring multi-byte characters. (This works
764 * because all server-encodings have the property that a valid
765 * multi-byte character representation cannot contain the
766 * representation of a valid single-byte character.)
767 */
768
769 if (elen > 1)
770 {
771 int mblen = pg_mblen(p);
772
773 if (mblen > 1)
774 {
775 /* slow, multi-byte path */
776 if (afterescape)
777 {
778 *r++ = '\\';
779 memcpy(r, p, mblen);
780 r += mblen;
781 afterescape = false;
782 }
783 else if (e && elen == mblen && memcmp(e, p, mblen) == 0)
784 {
785 /* SQL escape character; do not send to output */
786 afterescape = true;
787 }
788 else
789 {
790 /*
791 * We know it's a multi-byte character, so we don't need
792 * to do all the comparisons to single-byte characters
793 * that we do below.
794 */
795 memcpy(r, p, mblen);
796 r += mblen;
797 }
798
799 p += mblen;
800 plen -= mblen;
801
802 continue;
803 }
804 }
805
806 /* fast path */
807 if (afterescape)
808 {
809 if (pchar == '"' && !incharclass) /* escape-double-quote? */
810 {
811 /* emit appropriate part separator, per notes above */
812 if (nquotes == 0)
813 {
814 *r++ = ')';
815 *r++ = '{';
816 *r++ = '1';
817 *r++ = ',';
818 *r++ = '1';
819 *r++ = '}';
820 *r++ = '?';
821 *r++ = '(';
822 }
823 else if (nquotes == 1)
824 {
825 *r++ = ')';
826 *r++ = '{';
827 *r++ = '1';
828 *r++ = ',';
829 *r++ = '1';
830 *r++ = '}';
831 *r++ = '(';
832 *r++ = '?';
833 *r++ = ':';
834 }
835 else
836 ereport(ERROR,
837 (errcode(ERRCODE_INVALID_USE_OF_ESCAPE_CHARACTER),
838 errmsg("SQL regular expression may not contain more than two escape-double-quote separators")));
839 nquotes++;
840 }
841 else
842 {
843 /*
844 * We allow any character at all to be escaped; notably, this
845 * allows access to POSIX character-class escapes such as
846 * "\d". The SQL spec is considerably more restrictive.
847 */
848 *r++ = '\\';
849 *r++ = pchar;
850 }
851 afterescape = false;
852 }
853 else if (e && pchar == *e)
854 {
855 /* SQL escape character; do not send to output */
856 afterescape = true;
857 }
858 else if (incharclass)
859 {
860 if (pchar == '\\')
861 *r++ = '\\';
862 *r++ = pchar;
863 if (pchar == ']')
864 incharclass = false;
865 }
866 else if (pchar == '[')
867 {
868 *r++ = pchar;
869 incharclass = true;
870 }
871 else if (pchar == '%')
872 {
873 *r++ = '.';
874 *r++ = '*';
875 }
876 else if (pchar == '_')
877 *r++ = '.';
878 else if (pchar == '(')
879 {
880 /* convert to non-capturing parenthesis */
881 *r++ = '(';
882 *r++ = '?';
883 *r++ = ':';
884 }
885 else if (pchar == '\\' || pchar == '.' ||
886 pchar == '^' || pchar == '$')
887 {
888 *r++ = '\\';
889 *r++ = pchar;
890 }
891 else
892 *r++ = pchar;
893 p++, plen--;
894 }
895
896 *r++ = ')';
897 *r++ = '$';
898
899 SET_VARSIZE(result, r - ((char *) result));
900
901 PG_RETURN_TEXT_P(result);
902}
903
904/*
905 * regexp_match()
906 * Return the first substring(s) matching a pattern within a string.
907 */
908Datum
909regexp_match(PG_FUNCTION_ARGS)
910{
911 text *orig_str = PG_GETARG_TEXT_PP(0);
912 text *pattern = PG_GETARG_TEXT_PP(1);
913 text *flags = PG_GETARG_TEXT_PP_IF_EXISTS(2);
914 pg_re_flags re_flags;
915 regexp_matches_ctx *matchctx;
916
917 /* Determine options */
918 parse_re_flags(&re_flags, flags);
919 /* User mustn't specify 'g' */
920 if (re_flags.glob)
921 ereport(ERROR,
922 (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
923 /* translator: %s is a SQL function name */
924 errmsg("%s does not support the \"global\" option",
925 "regexp_match()"),
926 errhint("Use the regexp_matches function instead.")));
927
928 matchctx = setup_regexp_matches(orig_str, pattern, &re_flags,
929 PG_GET_COLLATION(), true, false, false);
930
931 if (matchctx->nmatches == 0)
932 PG_RETURN_NULL();
933
934 Assert(matchctx->nmatches == 1);
935
936 /* Create workspace that build_regexp_match_result needs */
937 matchctx->elems = (Datum *) palloc(sizeof(Datum) * matchctx->npatterns);
938 matchctx->nulls = (bool *) palloc(sizeof(bool) * matchctx->npatterns);
939
940 PG_RETURN_DATUM(PointerGetDatum(build_regexp_match_result(matchctx)));
941}
942
943/* This is separate to keep the opr_sanity regression test from complaining */
944Datum
945regexp_match_no_flags(PG_FUNCTION_ARGS)
946{
947 return regexp_match(fcinfo);
948}
949
950/*
951 * regexp_matches()
952 * Return a table of all matches of a pattern within a string.
953 */
954Datum
955regexp_matches(PG_FUNCTION_ARGS)
956{
957 FuncCallContext *funcctx;
958 regexp_matches_ctx *matchctx;
959
960 if (SRF_IS_FIRSTCALL())
961 {
962 text *pattern = PG_GETARG_TEXT_PP(1);
963 text *flags = PG_GETARG_TEXT_PP_IF_EXISTS(2);
964 pg_re_flags re_flags;
965 MemoryContext oldcontext;
966
967 funcctx = SRF_FIRSTCALL_INIT();
968 oldcontext = MemoryContextSwitchTo(funcctx->multi_call_memory_ctx);
969
970 /* Determine options */
971 parse_re_flags(&re_flags, flags);
972
973 /* be sure to copy the input string into the multi-call ctx */
974 matchctx = setup_regexp_matches(PG_GETARG_TEXT_P_COPY(0), pattern,
975 &re_flags,
976 PG_GET_COLLATION(),
977 true, false, false);
978
979 /* Pre-create workspace that build_regexp_match_result needs */
980 matchctx->elems = (Datum *) palloc(sizeof(Datum) * matchctx->npatterns);
981 matchctx->nulls = (bool *) palloc(sizeof(bool) * matchctx->npatterns);
982
983 MemoryContextSwitchTo(oldcontext);
984 funcctx->user_fctx = (void *) matchctx;
985 }
986
987 funcctx = SRF_PERCALL_SETUP();
988 matchctx = (regexp_matches_ctx *) funcctx->user_fctx;
989
990 if (matchctx->next_match < matchctx->nmatches)
991 {
992 ArrayType *result_ary;
993
994 result_ary = build_regexp_match_result(matchctx);
995 matchctx->next_match++;
996 SRF_RETURN_NEXT(funcctx, PointerGetDatum(result_ary));
997 }
998
999 SRF_RETURN_DONE(funcctx);
1000}
1001
1002/* This is separate to keep the opr_sanity regression test from complaining */
1003Datum
1004regexp_matches_no_flags(PG_FUNCTION_ARGS)
1005{
1006 return regexp_matches(fcinfo);
1007}
1008
1009/*
1010 * setup_regexp_matches --- do the initial matching for regexp_match
1011 * and regexp_split functions
1012 *
1013 * To avoid having to re-find the compiled pattern on each call, we do
1014 * all the matching in one swoop. The returned regexp_matches_ctx contains
1015 * the locations of all the substrings matching the pattern.
1016 *
1017 * The three bool parameters have only two patterns (one for matching, one for
1018 * splitting) but it seems clearer to distinguish the functionality this way
1019 * than to key it all off one "is_split" flag. We don't currently assume that
1020 * fetching_unmatched is exclusive of fetching the matched text too; if it's
1021 * set, the conversion buffer is large enough to fetch any single matched or
1022 * unmatched string, but not any larger substring. (In practice, when splitting
1023 * the matches are usually small anyway, and it didn't seem worth complicating
1024 * the code further.)
1025 */
1026static regexp_matches_ctx *
1027setup_regexp_matches(text *orig_str, text *pattern, pg_re_flags *re_flags,
1028 Oid collation,
1029 bool use_subpatterns,
1030 bool ignore_degenerate,
1031 bool fetching_unmatched)
1032{
1033 regexp_matches_ctx *matchctx = palloc0(sizeof(regexp_matches_ctx));
1034 int eml = pg_database_encoding_max_length();
1035 int orig_len;
1036 pg_wchar *wide_str;
1037 int wide_len;
1038 regex_t *cpattern;
1039 regmatch_t *pmatch;
1040 int pmatch_len;
1041 int array_len;
1042 int array_idx;
1043 int prev_match_end;
1044 int prev_valid_match_end;
1045 int start_search;
1046 int maxlen = 0; /* largest fetch length in characters */
1047
1048 /* save original string --- we'll extract result substrings from it */
1049 matchctx->orig_str = orig_str;
1050
1051 /* convert string to pg_wchar form for matching */
1052 orig_len = VARSIZE_ANY_EXHDR(orig_str);
1053 wide_str = (pg_wchar *) palloc(sizeof(pg_wchar) * (orig_len + 1));
1054 wide_len = pg_mb2wchar_with_len(VARDATA_ANY(orig_str), wide_str, orig_len);
1055
1056 /* set up the compiled pattern */
1057 cpattern = RE_compile_and_cache(pattern, re_flags->cflags, collation);
1058
1059 /* do we want to remember subpatterns? */
1060 if (use_subpatterns && cpattern->re_nsub > 0)
1061 {
1062 matchctx->npatterns = cpattern->re_nsub;
1063 pmatch_len = cpattern->re_nsub + 1;
1064 }
1065 else
1066 {
1067 use_subpatterns = false;
1068 matchctx->npatterns = 1;
1069 pmatch_len = 1;
1070 }
1071
1072 /* temporary output space for RE package */
1073 pmatch = palloc(sizeof(regmatch_t) * pmatch_len);
1074
1075 /*
1076 * the real output space (grown dynamically if needed)
1077 *
1078 * use values 2^n-1, not 2^n, so that we hit the limit at 2^28-1 rather
1079 * than at 2^27
1080 */
1081 array_len = re_flags->glob ? 255 : 31;
1082 matchctx->match_locs = (int *) palloc(sizeof(int) * array_len);
1083 array_idx = 0;
1084
1085 /* search for the pattern, perhaps repeatedly */
1086 prev_match_end = 0;
1087 prev_valid_match_end = 0;
1088 start_search = 0;
1089 while (RE_wchar_execute(cpattern, wide_str, wide_len, start_search,
1090 pmatch_len, pmatch))
1091 {
1092 /*
1093 * If requested, ignore degenerate matches, which are zero-length
1094 * matches occurring at the start or end of a string or just after a
1095 * previous match.
1096 */
1097 if (!ignore_degenerate ||
1098 (pmatch[0].rm_so < wide_len &&
1099 pmatch[0].rm_eo > prev_match_end))
1100 {
1101 /* enlarge output space if needed */
1102 while (array_idx + matchctx->npatterns * 2 + 1 > array_len)
1103 {
1104 array_len += array_len + 1; /* 2^n-1 => 2^(n+1)-1 */
1105 if (array_len > MaxAllocSize / sizeof(int))
1106 ereport(ERROR,
1107 (errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
1108 errmsg("too many regular expression matches")));
1109 matchctx->match_locs = (int *) repalloc(matchctx->match_locs,
1110 sizeof(int) * array_len);
1111 }
1112
1113 /* save this match's locations */
1114 if (use_subpatterns)
1115 {
1116 int i;
1117
1118 for (i = 1; i <= matchctx->npatterns; i++)
1119 {
1120 int so = pmatch[i].rm_so;
1121 int eo = pmatch[i].rm_eo;
1122
1123 matchctx->match_locs[array_idx++] = so;
1124 matchctx->match_locs[array_idx++] = eo;
1125 if (so >= 0 && eo >= 0 && (eo - so) > maxlen)
1126 maxlen = (eo - so);
1127 }
1128 }
1129 else
1130 {
1131 int so = pmatch[0].rm_so;
1132 int eo = pmatch[0].rm_eo;
1133
1134 matchctx->match_locs[array_idx++] = so;
1135 matchctx->match_locs[array_idx++] = eo;
1136 if (so >= 0 && eo >= 0 && (eo - so) > maxlen)
1137 maxlen = (eo - so);
1138 }
1139 matchctx->nmatches++;
1140
1141 /*
1142 * check length of unmatched portion between end of previous valid
1143 * (nondegenerate, or degenerate but not ignored) match and start
1144 * of current one
1145 */
1146 if (fetching_unmatched &&
1147 pmatch[0].rm_so >= 0 &&
1148 (pmatch[0].rm_so - prev_valid_match_end) > maxlen)
1149 maxlen = (pmatch[0].rm_so - prev_valid_match_end);
1150 prev_valid_match_end = pmatch[0].rm_eo;
1151 }
1152 prev_match_end = pmatch[0].rm_eo;
1153
1154 /* if not glob, stop after one match */
1155 if (!re_flags->glob)
1156 break;
1157
1158 /*
1159 * Advance search position. Normally we start the next search at the
1160 * end of the previous match; but if the match was of zero length, we
1161 * have to advance by one character, or we'd just find the same match
1162 * again.
1163 */
1164 start_search = prev_match_end;
1165 if (pmatch[0].rm_so == pmatch[0].rm_eo)
1166 start_search++;
1167 if (start_search > wide_len)
1168 break;
1169 }
1170
1171 /*
1172 * check length of unmatched portion between end of last match and end of
1173 * input string
1174 */
1175 if (fetching_unmatched &&
1176 (wide_len - prev_valid_match_end) > maxlen)
1177 maxlen = (wide_len - prev_valid_match_end);
1178
1179 /*
1180 * Keep a note of the end position of the string for the benefit of
1181 * splitting code.
1182 */
1183 matchctx->match_locs[array_idx] = wide_len;
1184
1185 if (eml > 1)
1186 {
1187 int64 maxsiz = eml * (int64) maxlen;
1188 int conv_bufsiz;
1189
1190 /*
1191 * Make the conversion buffer large enough for any substring of
1192 * interest.
1193 *
1194 * Worst case: assume we need the maximum size (maxlen*eml), but take
1195 * advantage of the fact that the original string length in bytes is
1196 * an upper bound on the byte length of any fetched substring (and we
1197 * know that len+1 is safe to allocate because the varlena header is
1198 * longer than 1 byte).
1199 */
1200 if (maxsiz > orig_len)
1201 conv_bufsiz = orig_len + 1;
1202 else
1203 conv_bufsiz = maxsiz + 1; /* safe since maxsiz < 2^30 */
1204
1205 matchctx->conv_buf = palloc(conv_bufsiz);
1206 matchctx->conv_bufsiz = conv_bufsiz;
1207 matchctx->wide_str = wide_str;
1208 }
1209 else
1210 {
1211 /* No need to keep the wide string if we're in a single-byte charset. */
1212 pfree(wide_str);
1213 matchctx->wide_str = NULL;
1214 matchctx->conv_buf = NULL;
1215 matchctx->conv_bufsiz = 0;
1216 }
1217
1218 /* Clean up temp storage */
1219 pfree(pmatch);
1220
1221 return matchctx;
1222}
1223
1224/*
1225 * build_regexp_match_result - build output array for current match
1226 */
1227static ArrayType *
1228build_regexp_match_result(regexp_matches_ctx *matchctx)
1229{
1230 char *buf = matchctx->conv_buf;
1231 int bufsiz PG_USED_FOR_ASSERTS_ONLY = matchctx->conv_bufsiz;
1232 Datum *elems = matchctx->elems;
1233 bool *nulls = matchctx->nulls;
1234 int dims[1];
1235 int lbs[1];
1236 int loc;
1237 int i;
1238
1239 /* Extract matching substrings from the original string */
1240 loc = matchctx->next_match * matchctx->npatterns * 2;
1241 for (i = 0; i < matchctx->npatterns; i++)
1242 {
1243 int so = matchctx->match_locs[loc++];
1244 int eo = matchctx->match_locs[loc++];
1245
1246 if (so < 0 || eo < 0)
1247 {
1248 elems[i] = (Datum) 0;
1249 nulls[i] = true;
1250 }
1251 else if (buf)
1252 {
1253 int len = pg_wchar2mb_with_len(matchctx->wide_str + so,
1254 buf,
1255 eo - so);
1256
1257 Assert(len < bufsiz);
1258 elems[i] = PointerGetDatum(cstring_to_text_with_len(buf, len));
1259 nulls[i] = false;
1260 }
1261 else
1262 {
1263 elems[i] = DirectFunctionCall3(text_substr,
1264 PointerGetDatum(matchctx->orig_str),
1265 Int32GetDatum(so + 1),
1266 Int32GetDatum(eo - so));
1267 nulls[i] = false;
1268 }
1269 }
1270
1271 /* And form an array */
1272 dims[0] = matchctx->npatterns;
1273 lbs[0] = 1;
1274 /* XXX: this hardcodes assumptions about the text type */
1275 return construct_md_array(elems, nulls, 1, dims, lbs,
1276 TEXTOID, -1, false, 'i');
1277}
1278
1279/*
1280 * regexp_split_to_table()
1281 * Split the string at matches of the pattern, returning the
1282 * split-out substrings as a table.
1283 */
1284Datum
1285regexp_split_to_table(PG_FUNCTION_ARGS)
1286{
1287 FuncCallContext *funcctx;
1288 regexp_matches_ctx *splitctx;
1289
1290 if (SRF_IS_FIRSTCALL())
1291 {
1292 text *pattern = PG_GETARG_TEXT_PP(1);
1293 text *flags = PG_GETARG_TEXT_PP_IF_EXISTS(2);
1294 pg_re_flags re_flags;
1295 MemoryContext oldcontext;
1296
1297 funcctx = SRF_FIRSTCALL_INIT();
1298 oldcontext = MemoryContextSwitchTo(funcctx->multi_call_memory_ctx);
1299
1300 /* Determine options */
1301 parse_re_flags(&re_flags, flags);
1302 /* User mustn't specify 'g' */
1303 if (re_flags.glob)
1304 ereport(ERROR,
1305 (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
1306 /* translator: %s is a SQL function name */
1307 errmsg("%s does not support the \"global\" option",
1308 "regexp_split_to_table()")));
1309 /* But we find all the matches anyway */
1310 re_flags.glob = true;
1311
1312 /* be sure to copy the input string into the multi-call ctx */
1313 splitctx = setup_regexp_matches(PG_GETARG_TEXT_P_COPY(0), pattern,
1314 &re_flags,
1315 PG_GET_COLLATION(),
1316 false, true, true);
1317
1318 MemoryContextSwitchTo(oldcontext);
1319 funcctx->user_fctx = (void *) splitctx;
1320 }
1321
1322 funcctx = SRF_PERCALL_SETUP();
1323 splitctx = (regexp_matches_ctx *) funcctx->user_fctx;
1324
1325 if (splitctx->next_match <= splitctx->nmatches)
1326 {
1327 Datum result = build_regexp_split_result(splitctx);
1328
1329 splitctx->next_match++;
1330 SRF_RETURN_NEXT(funcctx, result);
1331 }
1332
1333 SRF_RETURN_DONE(funcctx);
1334}
1335
1336/* This is separate to keep the opr_sanity regression test from complaining */
1337Datum
1338regexp_split_to_table_no_flags(PG_FUNCTION_ARGS)
1339{
1340 return regexp_split_to_table(fcinfo);
1341}
1342
1343/*
1344 * regexp_split_to_array()
1345 * Split the string at matches of the pattern, returning the
1346 * split-out substrings as an array.
1347 */
1348Datum
1349regexp_split_to_array(PG_FUNCTION_ARGS)
1350{
1351 ArrayBuildState *astate = NULL;
1352 pg_re_flags re_flags;
1353 regexp_matches_ctx *splitctx;
1354
1355 /* Determine options */
1356 parse_re_flags(&re_flags, PG_GETARG_TEXT_PP_IF_EXISTS(2));
1357 /* User mustn't specify 'g' */
1358 if (re_flags.glob)
1359 ereport(ERROR,
1360 (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
1361 /* translator: %s is a SQL function name */
1362 errmsg("%s does not support the \"global\" option",
1363 "regexp_split_to_array()")));
1364 /* But we find all the matches anyway */
1365 re_flags.glob = true;
1366
1367 splitctx = setup_regexp_matches(PG_GETARG_TEXT_PP(0),
1368 PG_GETARG_TEXT_PP(1),
1369 &re_flags,
1370 PG_GET_COLLATION(),
1371 false, true, true);
1372
1373 while (splitctx->next_match <= splitctx->nmatches)
1374 {
1375 astate = accumArrayResult(astate,
1376 build_regexp_split_result(splitctx),
1377 false,
1378 TEXTOID,
1379 CurrentMemoryContext);
1380 splitctx->next_match++;
1381 }
1382
1383 PG_RETURN_ARRAYTYPE_P(makeArrayResult(astate, CurrentMemoryContext));
1384}
1385
1386/* This is separate to keep the opr_sanity regression test from complaining */
1387Datum
1388regexp_split_to_array_no_flags(PG_FUNCTION_ARGS)
1389{
1390 return regexp_split_to_array(fcinfo);
1391}
1392
1393/*
1394 * build_regexp_split_result - build output string for current match
1395 *
1396 * We return the string between the current match and the previous one,
1397 * or the string after the last match when next_match == nmatches.
1398 */
1399static Datum
1400build_regexp_split_result(regexp_matches_ctx *splitctx)
1401{
1402 char *buf = splitctx->conv_buf;
1403 int startpos;
1404 int endpos;
1405
1406 if (splitctx->next_match > 0)
1407 startpos = splitctx->match_locs[splitctx->next_match * 2 - 1];
1408 else
1409 startpos = 0;
1410 if (startpos < 0)
1411 elog(ERROR, "invalid match ending position");
1412
1413 if (buf)
1414 {
1415 int bufsiz PG_USED_FOR_ASSERTS_ONLY = splitctx->conv_bufsiz;
1416 int len;
1417
1418 endpos = splitctx->match_locs[splitctx->next_match * 2];
1419 if (endpos < startpos)
1420 elog(ERROR, "invalid match starting position");
1421 len = pg_wchar2mb_with_len(splitctx->wide_str + startpos,
1422 buf,
1423 endpos - startpos);
1424 Assert(len < bufsiz);
1425 return PointerGetDatum(cstring_to_text_with_len(buf, len));
1426 }
1427 else
1428 {
1429 endpos = splitctx->match_locs[splitctx->next_match * 2];
1430 if (endpos < startpos)
1431 elog(ERROR, "invalid match starting position");
1432 return DirectFunctionCall3(text_substr,
1433 PointerGetDatum(splitctx->orig_str),
1434 Int32GetDatum(startpos + 1),
1435 Int32GetDatum(endpos - startpos));
1436 }
1437}
1438
1439/*
1440 * regexp_fixed_prefix - extract fixed prefix, if any, for a regexp
1441 *
1442 * The result is NULL if there is no fixed prefix, else a palloc'd string.
1443 * If it is an exact match, not just a prefix, *exact is returned as true.
1444 */
1445char *
1446regexp_fixed_prefix(text *text_re, bool case_insensitive, Oid collation,
1447 bool *exact)
1448{
1449 char *result;
1450 regex_t *re;
1451 int cflags;
1452 int re_result;
1453 pg_wchar *str;
1454 size_t slen;
1455 size_t maxlen;
1456 char errMsg[100];
1457
1458 *exact = false; /* default result */
1459
1460 /* Compile RE */
1461 cflags = REG_ADVANCED;
1462 if (case_insensitive)
1463 cflags |= REG_ICASE;
1464
1465 re = RE_compile_and_cache(text_re, cflags, collation);
1466
1467 /* Examine it to see if there's a fixed prefix */
1468 re_result = pg_regprefix(re, &str, &slen);
1469
1470 switch (re_result)
1471 {
1472 case REG_NOMATCH:
1473 return NULL;
1474
1475 case REG_PREFIX:
1476 /* continue with wchar conversion */
1477 break;
1478
1479 case REG_EXACT:
1480 *exact = true;
1481 /* continue with wchar conversion */
1482 break;
1483
1484 default:
1485 /* re failed??? */
1486 CHECK_FOR_INTERRUPTS();
1487 pg_regerror(re_result, re, errMsg, sizeof(errMsg));
1488 ereport(ERROR,
1489 (errcode(ERRCODE_INVALID_REGULAR_EXPRESSION),
1490 errmsg("regular expression failed: %s", errMsg)));
1491 break;
1492 }
1493
1494 /* Convert pg_wchar result back to database encoding */
1495 maxlen = pg_database_encoding_max_length() * slen + 1;
1496 result = (char *) palloc(maxlen);
1497 slen = pg_wchar2mb_with_len(str, result, slen);
1498 Assert(slen < maxlen);
1499
1500 free(str);
1501
1502 return result;
1503}
1504