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) 1996-2025, 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 1281820 : MatchText(const char *t, int tlen, const char *p, int plen, pg_locale_t locale)
85 : {
86 : /* Fast path for match-everything pattern */
87 1281820 : if (plen == 1 && *p == '%')
88 244 : return LIKE_TRUE;
89 :
90 : /* Since this function recurses, it could be driven to stack overflow */
91 1281576 : check_stack_depth();
92 :
93 : /*
94 : * In this loop, we advance by char when matching wildcards (and thus on
95 : * recursive entry to this function we are properly char-synced). On other
96 : * occasions it is safe to advance by byte, as the text and pattern will
97 : * be in lockstep. This allows us to perform all comparisons between the
98 : * text and pattern on a byte by byte basis, even for multi-byte
99 : * encodings.
100 : */
101 1907744 : while (tlen > 0 && plen > 0)
102 : {
103 1897202 : if (*p == '\\')
104 : {
105 : /* Next pattern byte must match literally, whatever it is */
106 12966 : NextByte(p, plen);
107 : /* ... and there had better be one, per SQL standard */
108 12966 : if (plen <= 0)
109 0 : ereport(ERROR,
110 : (errcode(ERRCODE_INVALID_ESCAPE_SEQUENCE),
111 : errmsg("LIKE pattern must not end with escape character")));
112 12966 : if (GETCHAR(*p) != GETCHAR(*t))
113 3442 : return LIKE_FALSE;
114 : }
115 1884236 : else if (*p == '%')
116 : {
117 : char firstpat;
118 :
119 : /*
120 : * % processing is essentially a search for a text position at
121 : * which the remainder of the text matches the remainder of the
122 : * pattern, using a recursive call to check each potential match.
123 : *
124 : * If there are wildcards immediately following the %, we can skip
125 : * over them first, using the idea that any sequence of N _'s and
126 : * one or more %'s is equivalent to N _'s and one % (ie, it will
127 : * match any sequence of at least N text characters). In this way
128 : * we will always run the recursive search loop using a pattern
129 : * fragment that begins with a literal character-to-match, thereby
130 : * not recursing more than we have to.
131 : */
132 167456 : NextByte(p, plen);
133 :
134 167954 : while (plen > 0)
135 : {
136 132080 : if (*p == '%')
137 18 : NextByte(p, plen);
138 132062 : else if (*p == '_')
139 : {
140 : /* If not enough text left to match the pattern, ABORT */
141 486 : if (tlen <= 0)
142 6 : return LIKE_ABORT;
143 480 : NextChar(t, tlen);
144 480 : NextByte(p, plen);
145 : }
146 : else
147 131576 : break; /* Reached a non-wildcard pattern char */
148 : }
149 :
150 : /*
151 : * If we're at end of pattern, match: we have a trailing % which
152 : * matches any remaining text string.
153 : */
154 167450 : if (plen <= 0)
155 35874 : return LIKE_TRUE;
156 :
157 : /*
158 : * Otherwise, scan for a text position at which we can match the
159 : * rest of the pattern. The first remaining pattern char is known
160 : * to be a regular or escaped literal character, so we can compare
161 : * the first pattern byte to each text byte to avoid recursing
162 : * more than we have to. This fact also guarantees that we don't
163 : * have to consider a match to the zero-length substring at the
164 : * end of the text. With a nondeterministic collation, we can't
165 : * rely on the first bytes being equal, so we have to recurse in
166 : * any case.
167 : */
168 131576 : if (*p == '\\')
169 : {
170 4 : if (plen < 2)
171 0 : ereport(ERROR,
172 : (errcode(ERRCODE_INVALID_ESCAPE_SEQUENCE),
173 : errmsg("LIKE pattern must not end with escape character")));
174 4 : firstpat = GETCHAR(p[1]);
175 : }
176 : else
177 131572 : firstpat = GETCHAR(*p);
178 :
179 3840798 : while (tlen > 0)
180 : {
181 3738234 : if (GETCHAR(*t) == firstpat || (locale && !locale->deterministic))
182 : {
183 108140 : int matched = MatchText(t, tlen, p, plen, locale);
184 :
185 108140 : if (matched != LIKE_FALSE)
186 29012 : return matched; /* TRUE or ABORT */
187 : }
188 :
189 3709270 : NextChar(t, tlen);
190 : }
191 :
192 : /*
193 : * End of text with no match, so no point in trying later places
194 : * to start matching this pattern.
195 : */
196 102564 : return LIKE_ABORT;
197 : }
198 1716780 : else if (*p == '_')
199 : {
200 : /* _ matches any single character, and we know there is one */
201 13212 : NextChar(t, tlen);
202 13194 : NextByte(p, plen);
203 13194 : continue;
204 : }
205 1703586 : else if (locale && !locale->deterministic)
206 : {
207 : /*
208 : * For nondeterministic locales, we find the next substring of the
209 : * pattern that does not contain wildcards and try to find a
210 : * matching substring in the text. Crucially, we cannot do this
211 : * character by character, as in the normal case, but must do it
212 : * substring by substring, partitioned by the wildcard characters.
213 : * (This is per SQL standard.)
214 : */
215 : const char *p1;
216 : size_t p1len;
217 : const char *t1;
218 : size_t t1len;
219 : bool found_escape;
220 : const char *subpat;
221 : size_t subpatlen;
222 282 : char *buf = NULL;
223 :
224 : /*
225 : * Determine next substring of pattern without wildcards. p is
226 : * the start of the subpattern, p1 is one past the last byte. Also
227 : * track if we found an escape character.
228 : */
229 282 : p1 = p;
230 282 : p1len = plen;
231 282 : found_escape = false;
232 834 : while (p1len > 0)
233 : {
234 690 : if (*p1 == '\\')
235 : {
236 12 : found_escape = true;
237 12 : NextByte(p1, p1len);
238 12 : if (p1len == 0)
239 6 : ereport(ERROR,
240 : (errcode(ERRCODE_INVALID_ESCAPE_SEQUENCE),
241 : errmsg("LIKE pattern must not end with escape character")));
242 : }
243 678 : else if (*p1 == '_' || *p1 == '%')
244 : break;
245 552 : NextByte(p1, p1len);
246 : }
247 :
248 : /*
249 : * If we found an escape character, then make an unescaped copy of
250 : * the subpattern.
251 : */
252 276 : if (found_escape)
253 : {
254 : char *b;
255 :
256 6 : b = buf = palloc(p1 - p);
257 30 : for (const char *c = p; c < p1; c++)
258 : {
259 24 : if (*c == '\\')
260 : ;
261 : else
262 18 : *(b++) = *c;
263 : }
264 :
265 6 : subpat = buf;
266 6 : subpatlen = b - buf;
267 : }
268 : else
269 : {
270 270 : subpat = p;
271 270 : subpatlen = p1 - p;
272 : }
273 :
274 : /*
275 : * Shortcut: If this is the end of the pattern, then the rest of
276 : * the text has to match the rest of the pattern.
277 : */
278 276 : if (p1len == 0)
279 : {
280 : int cmp;
281 :
282 144 : cmp = pg_strncoll(subpat, subpatlen, t, tlen, locale);
283 :
284 144 : if (buf)
285 6 : pfree(buf);
286 144 : if (cmp == 0)
287 90 : return LIKE_TRUE;
288 : else
289 54 : return LIKE_FALSE;
290 : }
291 :
292 : /*
293 : * Now build a substring of the text and try to match it against
294 : * the subpattern. t is the start of the text, t1 is one past the
295 : * last byte. We start with a zero-length string.
296 : */
297 132 : t1 = t;
298 132 : t1len = tlen;
299 : for (;;)
300 258 : {
301 : int cmp;
302 :
303 390 : CHECK_FOR_INTERRUPTS();
304 :
305 390 : cmp = pg_strncoll(subpat, subpatlen, t, (t1 - t), locale);
306 :
307 : /*
308 : * If we found a match, we have to test if the rest of pattern
309 : * can match against the rest of the string. Otherwise we
310 : * have to continue here try matching with a longer substring.
311 : * (This is similar to the recursion for the '%' wildcard
312 : * above.)
313 : *
314 : * Note that we can't just wind forward p and t and continue
315 : * with the main loop. This would fail for example with
316 : *
317 : * U&'\0061\0308bc' LIKE U&'\00E4_c' COLLATE ignore_accents
318 : *
319 : * You'd find that t=\0061 matches p=\00E4, but then the rest
320 : * won't match; but t=\0061\0308 also matches p=\00E4, and
321 : * then the rest will match.
322 : */
323 390 : if (cmp == 0)
324 : {
325 102 : int matched = MatchText(t1, t1len, p1, p1len, locale);
326 :
327 102 : if (matched == LIKE_TRUE)
328 : {
329 90 : if (buf)
330 0 : pfree(buf);
331 90 : return matched;
332 : }
333 : }
334 :
335 : /*
336 : * Didn't match. If we used up the whole text, then the match
337 : * fails. Otherwise, try again with a longer substring.
338 : */
339 300 : if (t1len == 0)
340 : {
341 42 : if (buf)
342 0 : pfree(buf);
343 42 : return LIKE_FALSE;
344 : }
345 : else
346 288 : NextChar(t1, t1len);
347 : }
348 : }
349 1703304 : else if (GETCHAR(*p) != GETCHAR(*t))
350 : {
351 : /* non-wildcard pattern char fails to match text char */
352 1099854 : return LIKE_FALSE;
353 : }
354 :
355 : /*
356 : * Pattern and text match, so advance.
357 : *
358 : * It is safe to use NextByte instead of NextChar here, even for
359 : * multi-byte character sets, because we are not following immediately
360 : * after a wildcard character. If we are in the middle of a multibyte
361 : * character, we must already have matched at least one byte of the
362 : * character from both text and pattern; so we cannot get out-of-sync
363 : * on character boundaries. And we know that no backend-legal
364 : * encoding allows ASCII characters such as '%' to appear as non-first
365 : * bytes of characters, so we won't mistakenly detect a new wildcard.
366 : */
367 612974 : NextByte(t, tlen);
368 612974 : NextByte(p, plen);
369 : }
370 :
371 10542 : if (tlen > 0)
372 306 : return LIKE_FALSE; /* end of pattern, but not of text */
373 :
374 : /*
375 : * End of text, but perhaps not of pattern. Match iff the remaining
376 : * pattern can match a zero-length string, ie, it's zero or more %'s.
377 : */
378 10828 : while (plen > 0 && *p == '%')
379 592 : NextByte(p, plen);
380 10236 : if (plen <= 0)
381 4672 : return LIKE_TRUE;
382 :
383 : /*
384 : * End of text with no match, so no point in trying later places to start
385 : * matching this pattern.
386 : */
387 5564 : return LIKE_ABORT;
388 : } /* MatchText() */
389 :
390 : /*
391 : * like_escape() --- given a pattern and an ESCAPE string,
392 : * convert the pattern to use Postgres' standard backslash escape convention.
393 : */
394 : #ifdef do_like_escape
395 :
396 : static text *
397 224 : do_like_escape(text *pat, text *esc)
398 : {
399 : text *result;
400 : char *p,
401 : *e,
402 : *r;
403 : int plen,
404 : elen;
405 : bool afterescape;
406 :
407 224 : p = VARDATA_ANY(pat);
408 224 : plen = VARSIZE_ANY_EXHDR(pat);
409 224 : e = VARDATA_ANY(esc);
410 224 : elen = VARSIZE_ANY_EXHDR(esc);
411 :
412 : /*
413 : * Worst-case pattern growth is 2x --- unlikely, but it's hardly worth
414 : * trying to calculate the size more accurately than that.
415 : */
416 224 : result = (text *) palloc(plen * 2 + VARHDRSZ);
417 224 : r = VARDATA(result);
418 :
419 224 : if (elen == 0)
420 : {
421 : /*
422 : * No escape character is wanted. Double any backslashes in the
423 : * pattern to make them act like ordinary characters.
424 : */
425 128 : while (plen > 0)
426 : {
427 96 : if (*p == '\\')
428 0 : *r++ = '\\';
429 192 : CopyAdvChar(r, p, plen);
430 : }
431 : }
432 : else
433 : {
434 : /*
435 : * The specified escape must be only a single character.
436 : */
437 192 : NextChar(e, elen);
438 192 : if (elen != 0)
439 0 : ereport(ERROR,
440 : (errcode(ERRCODE_INVALID_ESCAPE_SEQUENCE),
441 : errmsg("invalid escape string"),
442 : errhint("Escape string must be empty or one character.")));
443 :
444 192 : e = VARDATA_ANY(esc);
445 :
446 : /*
447 : * If specified escape is '\', just copy the pattern as-is.
448 : */
449 192 : if (*e == '\\')
450 : {
451 0 : memcpy(result, pat, VARSIZE_ANY(pat));
452 0 : return result;
453 : }
454 :
455 : /*
456 : * Otherwise, convert occurrences of the specified escape character to
457 : * '\', and double occurrences of '\' --- unless they immediately
458 : * follow an escape character!
459 : */
460 192 : afterescape = false;
461 1164 : while (plen > 0)
462 : {
463 972 : if (CHAREQ(p, e) && !afterescape)
464 : {
465 192 : *r++ = '\\';
466 192 : NextChar(p, plen);
467 192 : afterescape = true;
468 : }
469 780 : else if (*p == '\\')
470 : {
471 0 : *r++ = '\\';
472 0 : if (!afterescape)
473 0 : *r++ = '\\';
474 0 : NextChar(p, plen);
475 0 : afterescape = false;
476 : }
477 : else
478 : {
479 1524 : CopyAdvChar(r, p, plen);
480 780 : afterescape = false;
481 : }
482 : }
483 : }
484 :
485 224 : SET_VARSIZE(result, r - ((char *) result));
486 :
487 224 : return result;
488 : }
489 : #endif /* do_like_escape */
490 :
491 : #ifdef CHAREQ
492 : #undef CHAREQ
493 : #endif
494 :
495 : #undef NextChar
496 : #undef CopyAdvChar
497 : #undef MatchText
498 :
499 : #ifdef do_like_escape
500 : #undef do_like_escape
501 : #endif
502 :
503 : #undef GETCHAR
504 :
505 : #ifdef MATCH_LOWER
506 : #undef MATCH_LOWER
507 :
508 : #endif
|
