Branch data Line data Source code
1 : : /*-------------------------------------------------------------------------
2 : : *
3 : : * like_match.c
4 : : * LIKE pattern matching internal code.
5 : : *
6 : : * This file is included by like.c four times, to provide matching code for
7 : : * (1) single-byte encodings, (2) UTF8, (3) other multi-byte encodings,
8 : : * and (4) case insensitive matches in single-byte encodings.
9 : : * (UTF8 is a special case because we can use a much more efficient version
10 : : * of NextChar than can be used for general multi-byte encodings.)
11 : : *
12 : : * Before the inclusion, we need to define the following macros:
13 : : *
14 : : * NextChar
15 : : * MatchText - to name of function wanted
16 : : * do_like_escape - name of function if wanted - needs CHAREQ and CopyAdvChar
17 : : * MATCH_LOWER - define for case (4) to specify case folding for 1-byte chars
18 : : *
19 : : * Copyright (c) 1996-2026, PostgreSQL Global Development Group
20 : : *
21 : : * IDENTIFICATION
22 : : * src/backend/utils/adt/like_match.c
23 : : *
24 : : *-------------------------------------------------------------------------
25 : : */
26 : :
27 : : /*
28 : : * Originally written by Rich $alz, mirror!rs, Wed Nov 26 19:03:17 EST 1986.
29 : : * Rich $alz is now <rsalz@bbn.com>.
30 : : * Special thanks to Lars Mathiesen <thorinn@diku.dk> for the
31 : : * LIKE_ABORT code.
32 : : *
33 : : * This code was shamelessly stolen from the "pql" code by myself and
34 : : * slightly modified :)
35 : : *
36 : : * All references to the word "star" were replaced by "percent"
37 : : * All references to the word "wild" were replaced by "like"
38 : : *
39 : : * All the nice shell RE matching stuff was replaced by just "_" and "%"
40 : : *
41 : : * As I don't have a copy of the SQL standard handy I wasn't sure whether
42 : : * to leave in the '\' escape character handling.
43 : : *
44 : : * Keith Parks. <keith@mtcc.demon.co.uk>
45 : : *
46 : : * SQL lets you specify the escape character by saying
47 : : * LIKE <pattern> ESCAPE <escape character>. We are a small operation
48 : : * so we force you to use '\'. - ay 7/95
49 : : *
50 : : * Now we have the like_escape() function that converts patterns with
51 : : * any specified escape character (or none at all) to the internal
52 : : * default escape character, which is still '\'. - tgl 9/2000
53 : : *
54 : : * The code is rewritten to avoid requiring null-terminated strings,
55 : : * which in turn allows us to leave out some memcpy() operations.
56 : : * This code should be faster and take less memory, but no promises...
57 : : * - thomas 2000-08-06
58 : : */
59 : :
60 : :
61 : : /*--------------------
62 : : * Match text and pattern, return LIKE_TRUE, LIKE_FALSE, or LIKE_ABORT.
63 : : *
64 : : * LIKE_TRUE: they match
65 : : * LIKE_FALSE: they don't match
66 : : * LIKE_ABORT: not only don't they match, but the text is too short.
67 : : *
68 : : * If LIKE_ABORT is returned, then no suffix of the text can match the
69 : : * pattern either, so an upper-level % scan can stop scanning now.
70 : : *--------------------
71 : : */
72 : :
73 : : /*
74 : : * MATCH_LOWER is defined for ILIKE in the C locale as an optimization. Other
75 : : * locales must casefold the inputs before matching.
76 : : */
77 : : #ifdef MATCH_LOWER
78 : : #define GETCHAR(t) pg_ascii_tolower(t)
79 : : #else
80 : : #define GETCHAR(t) (t)
81 : : #endif
82 : :
83 : : static int
84 : 816025 : MatchText(const char *t, int tlen, const char *p, int plen, pg_locale_t locale)
85 : : {
86 [ + + + + ]: 816025 : bool nondeterministic = (locale && !locale->deterministic);
87 : :
88 : : /* Fast path for match-everything pattern */
89 [ + + + + ]: 816025 : if (plen == 1 && *p == '%')
90 : 146 : return LIKE_TRUE;
91 : :
92 : : /* Since this function recurses, it could be driven to stack overflow */
93 : 815879 : check_stack_depth();
94 : :
95 : : /*
96 : : * In this loop, we advance by char when matching wildcards (and thus on
97 : : * recursive entry to this function we are properly char-synced). On other
98 : : * occasions it is safe to advance by byte, as the text and pattern will
99 : : * be in lockstep. This allows us to perform all comparisons between the
100 : : * text and pattern on a byte by byte basis, even for multi-byte
101 : : * encodings. (But that doesn't work in a nondeterministic locale, so the
102 : : * nondeterministic case below has to advance the text by chars.)
103 : : */
104 [ + + + + ]: 1204028 : while (tlen > 0 && plen > 0)
105 : : {
106 : : /*
107 : : * At the top of this loop, we are not positioned immediately after an
108 : : * escape, so we may take wildcards at face value.
109 : : */
110 [ + + ]: 1196762 : if (*p == '%')
111 : : {
112 : : char firstpat;
113 : :
114 : : /*
115 : : * % processing is essentially a search for a text position at
116 : : * which the remainder of the text matches the remainder of the
117 : : * pattern, using a recursive call to check each potential match.
118 : : *
119 : : * If there are wildcards immediately following the %, we can skip
120 : : * over them first, using the idea that any sequence of N _'s and
121 : : * one or more %'s is equivalent to N _'s and one % (ie, it will
122 : : * match any sequence of at least N text characters). In this way
123 : : * we will always run the recursive search loop using a pattern
124 : : * fragment that begins with a literal character-to-match, thereby
125 : : * not recursing more than we have to.
126 : : */
127 : 94395 : NextByte(p, plen);
128 : :
129 [ + + ]: 94744 : while (plen > 0)
130 : : {
131 [ + + ]: 73555 : if (*p == '%')
132 : 15 : NextByte(p, plen);
133 [ + + ]: 73540 : else if (*p == '_')
134 : : {
135 : : /* If not enough text left to match the pattern, ABORT */
136 [ + + ]: 339 : if (tlen <= 0)
137 : 5 : return LIKE_ABORT;
138 [ + + - + ]: 334 : NextChar(t, tlen);
139 : 334 : NextByte(p, plen);
140 : : }
141 : : else
142 : 73201 : break; /* Reached a non-wildcard pattern char */
143 : : }
144 : :
145 : : /*
146 : : * If we're at end of pattern, match: we have a trailing % which
147 : : * matches any remaining text string.
148 : : */
149 [ + + ]: 94390 : if (plen <= 0)
150 : 21189 : return LIKE_TRUE;
151 : :
152 : : /*
153 : : * Otherwise, scan for a text position at which we can match the
154 : : * rest of the pattern. The first remaining pattern char is known
155 : : * to be a regular or escaped literal character, so we can compare
156 : : * the first pattern byte to each text byte to avoid recursing
157 : : * more than we have to. This fact also guarantees that we don't
158 : : * have to consider a match to the zero-length substring at the
159 : : * end of the text. But with a nondeterministic locale, we can't
160 : : * rely on the first byte of a match being equal, so we have to
161 : : * recurse in any case.
162 : : */
163 [ + + ]: 73201 : if (*p == '\\')
164 : : {
165 [ - + ]: 2 : if (plen < 2)
166 [ # # ]: 0 : ereport(ERROR,
167 : : (errcode(ERRCODE_INVALID_ESCAPE_SEQUENCE),
168 : : errmsg("LIKE pattern must not end with escape character")));
169 : 2 : firstpat = GETCHAR(p[1]);
170 : : }
171 : : else
172 : 73199 : firstpat = GETCHAR(*p);
173 : :
174 [ + + ]: 2152931 : while (tlen > 0)
175 : : {
176 [ + + + + ]: 2095627 : if (GETCHAR(*t) == firstpat || nondeterministic)
177 : : {
178 : 58984 : int matched = MatchText(t, tlen, p, plen, locale);
179 : :
180 [ + + ]: 58984 : if (matched != LIKE_FALSE)
181 : 15897 : return matched; /* TRUE or ABORT */
182 : : }
183 : :
184 [ + + + + ]: 2079768 : NextChar(t, tlen);
185 : : }
186 : :
187 : : /*
188 : : * End of text with no match, so no point in trying later places
189 : : * to start matching this pattern.
190 : : */
191 : 57304 : return LIKE_ABORT;
192 : : }
193 [ + + ]: 1102367 : else if (*p == '_')
194 : : {
195 : : /* _ matches any single character, and we know there is one */
196 [ + + + + ]: 10154 : NextChar(t, tlen);
197 : 10134 : NextByte(p, plen);
198 : 10134 : continue;
199 : : }
200 [ + + ]: 1092233 : else if (nondeterministic)
201 : : {
202 : : /*
203 : : * For nondeterministic locales, we find the next substring of the
204 : : * pattern that does not contain wildcards and try to find a
205 : : * matching substring in the text. Crucially, we cannot do this
206 : : * character by character, as in the normal case, but must do it
207 : : * substring by substring, partitioned by the wildcard characters.
208 : : * (This is per SQL standard.)
209 : : */
210 : : const char *p1;
211 : : size_t p1len;
212 : : const char *t1;
213 : : size_t t1len;
214 : : bool found_escape;
215 : : const char *subpat;
216 : : size_t subpatlen;
217 : 277 : char *buf = NULL;
218 : :
219 : : /*
220 : : * Determine length of substring of pattern without wildcards. p
221 : : * is the start of the subpattern, p1 will advance to one past its
222 : : * last byte. Also track if we found an escape character.
223 : : */
224 : 277 : p1 = p;
225 : 277 : p1len = plen;
226 : 277 : found_escape = false;
227 [ + + ]: 923 : while (p1len > 0)
228 : : {
229 [ + + ]: 776 : if (*p1 == '\\')
230 : : {
231 : 64 : found_escape = true;
232 : 64 : NextByte(p1, p1len);
233 [ + + ]: 64 : if (p1len == 0)
234 [ + - ]: 4 : ereport(ERROR,
235 : : (errcode(ERRCODE_INVALID_ESCAPE_SEQUENCE),
236 : : errmsg("LIKE pattern must not end with escape character")));
237 : : }
238 [ + + + + ]: 712 : else if (*p1 == '_' || *p1 == '%')
239 : : break;
240 : : /* Advance over regular or escaped character */
241 : 646 : NextByte(p1, p1len);
242 : : }
243 : :
244 : : /*
245 : : * If we found an escape character, then make a de-escaped copy of
246 : : * the subpattern that we can use to match literally. Otherwise
247 : : * we can use the subpattern in-place. (buf holds the de-escaped
248 : : * copy; be sure to pfree it before returning.)
249 : : */
250 [ + + ]: 273 : if (found_escape)
251 : : {
252 : : char *b;
253 : :
254 : 50 : b = buf = palloc(p1 - p);
255 [ + + ]: 265 : for (const char *c = p; c < p1; c++)
256 : : {
257 [ + + ]: 215 : if (*c == '\\')
258 : 60 : c++; /* we already checked this isn't the end */
259 : 215 : *(b++) = *c;
260 : : }
261 : :
262 : 50 : subpat = buf;
263 : 50 : subpatlen = b - buf;
264 : : }
265 : : else
266 : : {
267 : 223 : subpat = p;
268 : 223 : subpatlen = p1 - p;
269 : : }
270 : :
271 : : /*
272 : : * Shortcut: If this is the end of the pattern, then the rest of
273 : : * the text has to match the rest of the pattern.
274 : : */
275 [ + + ]: 273 : if (p1len == 0)
276 : : {
277 : : int cmp;
278 : :
279 : 147 : cmp = pg_strncoll(subpat, subpatlen, t, tlen, locale);
280 : :
281 [ + + ]: 147 : if (buf)
282 : 25 : pfree(buf);
283 [ + + ]: 147 : if (cmp == 0)
284 : 97 : return LIKE_TRUE;
285 : : else
286 : 50 : return LIKE_FALSE;
287 : : }
288 : :
289 : : /*
290 : : * Consider each successively-longer substring of the remaining
291 : : * text and try to match it against the subpattern. t is the
292 : : * start of the substring, t1 is one past its last byte. We start
293 : : * with a zero-length substring.
294 : : */
295 : 126 : t1 = t;
296 : 126 : t1len = tlen;
297 : : for (;;)
298 : 336 : {
299 : : int cmp;
300 : :
301 : : /* This could be slow, so allow interrupts */
302 [ - + ]: 462 : CHECK_FOR_INTERRUPTS();
303 : :
304 : 462 : cmp = pg_strncoll(subpat, subpatlen, t, (t1 - t), locale);
305 : :
306 : : /*
307 : : * If we found a match, we have to test if the rest of pattern
308 : : * can match against the rest of the text. If not, we have to
309 : : * continue and try the next longer substring. (This is
310 : : * similar to the recursion for the '%' wildcard above.)
311 : : *
312 : : * Note that we can't just wind forward p and t and continue
313 : : * with the main loop. This would fail for example with
314 : : *
315 : : * U&'\0061\0308bc' LIKE U&'\00E4_c' COLLATE ignore_accents
316 : : *
317 : : * You'd find that t=\0061 matches p=\00E4, but then the rest
318 : : * won't match; but t=\0061\0308 also matches p=\00E4, and
319 : : * then the rest will match.
320 : : */
321 [ + + ]: 462 : if (cmp == 0)
322 : : {
323 : 100 : int matched = MatchText(t1, t1len, p1, p1len, locale);
324 : :
325 [ + + ]: 100 : if (matched == LIKE_TRUE)
326 : : {
327 [ + + ]: 90 : if (buf)
328 : 20 : pfree(buf);
329 : 90 : return matched;
330 : : }
331 : : }
332 : :
333 : : /*
334 : : * Didn't match. If we used up the whole text, then the match
335 : : * fails. Otherwise, try again with a longer substring.
336 : : */
337 [ + + ]: 372 : if (t1len == 0)
338 : : {
339 [ + + ]: 36 : if (buf)
340 : 5 : pfree(buf);
341 : 36 : return LIKE_FALSE;
342 : : }
343 : : else
344 [ + + + + ]: 366 : NextChar(t1, t1len);
345 : : } /* end loop over substrings starting at t */
346 : : }
347 : : /* the rest of this loop considers only deterministic cases */
348 [ + + ]: 1091956 : else if (*p == '\\')
349 : : {
350 : : /* Next pattern byte must match literally, whatever it is */
351 : 9006 : NextByte(p, plen);
352 : : /* ... and there had better be one, per SQL standard */
353 [ - + ]: 9006 : if (plen <= 0)
354 [ # # ]: 0 : ereport(ERROR,
355 : : (errcode(ERRCODE_INVALID_ESCAPE_SEQUENCE),
356 : : errmsg("LIKE pattern must not end with escape character")));
357 [ + + ]: 9006 : if (GETCHAR(*p) != GETCHAR(*t))
358 : 2446 : return LIKE_FALSE;
359 : : }
360 [ + + ]: 1082950 : else if (GETCHAR(*p) != GETCHAR(*t))
361 : : {
362 : : /* non-wildcard pattern char fails to match text char */
363 : 711495 : return LIKE_FALSE;
364 : : }
365 : :
366 : : /*
367 : : * Pattern and text match, so advance.
368 : : *
369 : : * It is safe to use NextByte instead of NextChar here, even for
370 : : * multi-byte character sets, because we are not following immediately
371 : : * after a wildcard character. If we are in the middle of a multibyte
372 : : * character, we must already have matched at least one byte of the
373 : : * character from both text and pattern; so we cannot get out-of-sync
374 : : * on character boundaries. And we know that no backend-legal
375 : : * encoding allows ASCII characters such as '%' to appear as non-first
376 : : * bytes of characters, so we won't mistakenly detect a new wildcard.
377 : : */
378 : 378015 : NextByte(t, tlen);
379 : 378015 : NextByte(p, plen);
380 : : }
381 : :
382 [ + + ]: 7266 : if (tlen > 0)
383 : 259 : return LIKE_FALSE; /* end of pattern, but not of text */
384 : :
385 : : /*
386 : : * End of text, but perhaps not of pattern. Match iff the remaining
387 : : * pattern can match a zero-length string, ie, it's zero or more %'s.
388 : : */
389 [ + + + + ]: 7409 : while (plen > 0 && *p == '%')
390 : 402 : NextByte(p, plen);
391 [ + + ]: 7007 : if (plen <= 0)
392 : 3365 : return LIKE_TRUE;
393 : :
394 : : /*
395 : : * End of text with no match, so no point in trying later places to start
396 : : * matching this pattern.
397 : : */
398 : 3642 : return LIKE_ABORT;
399 : : } /* MatchText() */
400 : :
401 : : /*
402 : : * like_escape() --- given a pattern and an ESCAPE string,
403 : : * convert the pattern to use Postgres' standard backslash escape convention.
404 : : */
405 : : #ifdef do_like_escape
406 : :
407 : : static text *
408 : 196 : do_like_escape(text *pat, text *esc)
409 : : {
410 : : text *result;
411 : : char *p,
412 : : *e,
413 : : *r;
414 : : int plen,
415 : : elen;
416 : : bool afterescape;
417 : :
418 : 196 : p = VARDATA_ANY(pat);
419 : 196 : plen = VARSIZE_ANY_EXHDR(pat);
420 : 196 : e = VARDATA_ANY(esc);
421 : 196 : elen = VARSIZE_ANY_EXHDR(esc);
422 : :
423 : : /*
424 : : * Worst-case pattern growth is 2x --- unlikely, but it's hardly worth
425 : : * trying to calculate the size more accurately than that.
426 : : */
427 : 196 : result = (text *) palloc(plen * 2 + VARHDRSZ);
428 : 196 : r = VARDATA(result);
429 : :
430 [ + + ]: 196 : if (elen == 0)
431 : : {
432 : : /*
433 : : * No escape character is wanted. Double any backslashes in the
434 : : * pattern to make them act like ordinary characters.
435 : : */
436 [ + + ]: 64 : while (plen > 0)
437 : : {
438 [ - + ]: 48 : if (*p == '\\')
439 : 0 : *r++ = '\\';
440 [ + + ]: 96 : CopyAdvChar(r, p, plen);
441 : : }
442 : : }
443 : : else
444 : : {
445 : : /*
446 : : * The specified escape must be only a single character.
447 : : */
448 : 180 : NextChar(e, elen);
449 [ - + ]: 180 : if (elen != 0)
450 [ # # ]: 0 : ereport(ERROR,
451 : : (errcode(ERRCODE_INVALID_ESCAPE_SEQUENCE),
452 : : errmsg("invalid escape string"),
453 : : errhint("Escape string must be empty or one character.")));
454 : :
455 : 180 : e = VARDATA_ANY(esc);
456 : 180 : elen = VARSIZE_ANY_EXHDR(esc);
457 : :
458 : : /*
459 : : * If specified escape is '\', just copy the pattern as-is.
460 : : */
461 [ - + ]: 180 : if (*e == '\\')
462 : : {
463 : 0 : memcpy(result, pat, VARSIZE_ANY(pat));
464 : 0 : return result;
465 : : }
466 : :
467 : : /*
468 : : * Otherwise, convert occurrences of the specified escape character to
469 : : * '\', and double occurrences of '\' --- unless they immediately
470 : : * follow an escape character!
471 : : */
472 : 180 : afterescape = false;
473 [ + + ]: 1130 : while (plen > 0)
474 : : {
475 [ + + + + ]: 950 : if (CHAREQ(p, plen, e, elen) && !afterescape)
476 : : {
477 : 175 : *r++ = '\\';
478 : 175 : NextChar(p, plen);
479 : 175 : afterescape = true;
480 : : }
481 [ + + ]: 775 : else if (*p == '\\')
482 : : {
483 : 15 : *r++ = '\\';
484 [ + + ]: 15 : if (!afterescape)
485 : 10 : *r++ = '\\';
486 : 15 : NextChar(p, plen);
487 : 15 : afterescape = false;
488 : : }
489 : : else
490 : : {
491 [ + + ]: 1500 : CopyAdvChar(r, p, plen);
492 : 760 : afterescape = false;
493 : : }
494 : : }
495 : : }
496 : :
497 : 196 : SET_VARSIZE(result, r - ((char *) result));
498 : :
499 : 196 : return result;
500 : : }
501 : : #endif /* do_like_escape */
502 : :
503 : : #ifdef CHAREQ
504 : : #undef CHAREQ
505 : : #endif
506 : :
507 : : #undef NextChar
508 : : #undef CopyAdvChar
509 : : #undef MatchText
510 : :
511 : : #ifdef do_like_escape
512 : : #undef do_like_escape
513 : : #endif
514 : :
515 : : #undef GETCHAR
516 : :
517 : : #ifdef MATCH_LOWER
518 : : #undef MATCH_LOWER
519 : :
520 : : #endif
|
