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) 1996-2024, 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 : #ifdef MATCH_LOWER
74 : #define GETCHAR(t, locale) MATCH_LOWER(t, locale)
75 : #else
76 : #define GETCHAR(t, locale) (t)
77 : #endif
78 :
79 : static int
80 1199178 : MatchText(const char *t, int tlen, const char *p, int plen, pg_locale_t locale)
81 : {
82 : /* Fast path for match-everything pattern */
83 1199178 : if (plen == 1 && *p == '%')
84 202 : return LIKE_TRUE;
85 :
86 : /* Since this function recurses, it could be driven to stack overflow */
87 1198976 : check_stack_depth();
88 :
89 : /*
90 : * In this loop, we advance by char when matching wildcards (and thus on
91 : * recursive entry to this function we are properly char-synced). On other
92 : * occasions it is safe to advance by byte, as the text and pattern will
93 : * be in lockstep. This allows us to perform all comparisons between the
94 : * text and pattern on a byte by byte basis, even for multi-byte
95 : * encodings.
96 : */
97 1802386 : while (tlen > 0 && plen > 0)
98 : {
99 1791972 : if (*p == '\\')
100 : {
101 : /* Next pattern byte must match literally, whatever it is */
102 12416 : NextByte(p, plen);
103 : /* ... and there had better be one, per SQL standard */
104 12416 : if (plen <= 0)
105 0 : ereport(ERROR,
106 : (errcode(ERRCODE_INVALID_ESCAPE_SEQUENCE),
107 : errmsg("LIKE pattern must not end with escape character")));
108 12416 : if (GETCHAR(*p, locale) != GETCHAR(*t, locale))
109 3452 : return LIKE_FALSE;
110 : }
111 1779556 : else if (*p == '%')
112 : {
113 : char firstpat;
114 :
115 : /*
116 : * % processing is essentially a search for a text position at
117 : * which the remainder of the text matches the remainder of the
118 : * pattern, using a recursive call to check each potential match.
119 : *
120 : * If there are wildcards immediately following the %, we can skip
121 : * over them first, using the idea that any sequence of N _'s and
122 : * one or more %'s is equivalent to N _'s and one % (ie, it will
123 : * match any sequence of at least N text characters). In this way
124 : * we will always run the recursive search loop using a pattern
125 : * fragment that begins with a literal character-to-match, thereby
126 : * not recursing more than we have to.
127 : */
128 165646 : NextByte(p, plen);
129 :
130 166138 : while (plen > 0)
131 : {
132 131718 : if (*p == '%')
133 6 : NextByte(p, plen);
134 131712 : else if (*p == '_')
135 : {
136 : /* If not enough text left to match the pattern, ABORT */
137 492 : if (tlen <= 0)
138 6 : return LIKE_ABORT;
139 486 : NextChar(t, tlen);
140 486 : NextByte(p, plen);
141 : }
142 : else
143 131220 : break; /* Reached a non-wildcard pattern char */
144 : }
145 :
146 : /*
147 : * If we're at end of pattern, match: we have a trailing % which
148 : * matches any remaining text string.
149 : */
150 165640 : if (plen <= 0)
151 34420 : return LIKE_TRUE;
152 :
153 : /*
154 : * Otherwise, scan for a text position at which we can match the
155 : * rest of the pattern. The first remaining pattern char is known
156 : * to be a regular or escaped literal character, so we can compare
157 : * the first pattern byte to each text byte to avoid recursing
158 : * more than we have to. This fact also guarantees that we don't
159 : * have to consider a match to the zero-length substring at the
160 : * end of the text.
161 : */
162 131220 : if (*p == '\\')
163 : {
164 4 : if (plen < 2)
165 0 : ereport(ERROR,
166 : (errcode(ERRCODE_INVALID_ESCAPE_SEQUENCE),
167 : errmsg("LIKE pattern must not end with escape character")));
168 4 : firstpat = GETCHAR(p[1], locale);
169 : }
170 : else
171 131216 : firstpat = GETCHAR(*p, locale);
172 :
173 3842620 : while (tlen > 0)
174 : {
175 3740286 : if (GETCHAR(*t, locale) == firstpat)
176 : {
177 107308 : int matched = MatchText(t, tlen, p, plen, locale);
178 :
179 107308 : if (matched != LIKE_FALSE)
180 28886 : return matched; /* TRUE or ABORT */
181 : }
182 :
183 3711424 : NextChar(t, tlen);
184 : }
185 :
186 : /*
187 : * End of text with no match, so no point in trying later places
188 : * to start matching this pattern.
189 : */
190 102334 : return LIKE_ABORT;
191 : }
192 1613910 : else if (*p == '_')
193 : {
194 : /* _ matches any single character, and we know there is one */
195 10724 : NextChar(t, tlen);
196 10724 : NextByte(p, plen);
197 10724 : continue;
198 : }
199 1603186 : else if (GETCHAR(*p, locale) != GETCHAR(*t, locale))
200 : {
201 : /* non-wildcard pattern char fails to match text char */
202 1019464 : return LIKE_FALSE;
203 : }
204 :
205 : /*
206 : * Pattern and text match, so advance.
207 : *
208 : * It is safe to use NextByte instead of NextChar here, even for
209 : * multi-byte character sets, because we are not following immediately
210 : * after a wildcard character. If we are in the middle of a multibyte
211 : * character, we must already have matched at least one byte of the
212 : * character from both text and pattern; so we cannot get out-of-sync
213 : * on character boundaries. And we know that no backend-legal
214 : * encoding allows ASCII characters such as '%' to appear as non-first
215 : * bytes of characters, so we won't mistakenly detect a new wildcard.
216 : */
217 592686 : NextByte(t, tlen);
218 592686 : NextByte(p, plen);
219 : }
220 :
221 10414 : if (tlen > 0)
222 312 : return LIKE_FALSE; /* end of pattern, but not of text */
223 :
224 : /*
225 : * End of text, but perhaps not of pattern. Match iff the remaining
226 : * pattern can match a zero-length string, ie, it's zero or more %'s.
227 : */
228 10676 : while (plen > 0 && *p == '%')
229 574 : NextByte(p, plen);
230 10102 : if (plen <= 0)
231 4622 : return LIKE_TRUE;
232 :
233 : /*
234 : * End of text with no match, so no point in trying later places to start
235 : * matching this pattern.
236 : */
237 5480 : return LIKE_ABORT;
238 : } /* MatchText() */
239 :
240 : /*
241 : * like_escape() --- given a pattern and an ESCAPE string,
242 : * convert the pattern to use Postgres' standard backslash escape convention.
243 : */
244 : #ifdef do_like_escape
245 :
246 : static text *
247 224 : do_like_escape(text *pat, text *esc)
248 : {
249 : text *result;
250 : char *p,
251 : *e,
252 : *r;
253 : int plen,
254 : elen;
255 : bool afterescape;
256 :
257 224 : p = VARDATA_ANY(pat);
258 224 : plen = VARSIZE_ANY_EXHDR(pat);
259 224 : e = VARDATA_ANY(esc);
260 224 : elen = VARSIZE_ANY_EXHDR(esc);
261 :
262 : /*
263 : * Worst-case pattern growth is 2x --- unlikely, but it's hardly worth
264 : * trying to calculate the size more accurately than that.
265 : */
266 224 : result = (text *) palloc(plen * 2 + VARHDRSZ);
267 224 : r = VARDATA(result);
268 :
269 224 : if (elen == 0)
270 : {
271 : /*
272 : * No escape character is wanted. Double any backslashes in the
273 : * pattern to make them act like ordinary characters.
274 : */
275 128 : while (plen > 0)
276 : {
277 96 : if (*p == '\\')
278 0 : *r++ = '\\';
279 192 : CopyAdvChar(r, p, plen);
280 : }
281 : }
282 : else
283 : {
284 : /*
285 : * The specified escape must be only a single character.
286 : */
287 192 : NextChar(e, elen);
288 192 : if (elen != 0)
289 0 : ereport(ERROR,
290 : (errcode(ERRCODE_INVALID_ESCAPE_SEQUENCE),
291 : errmsg("invalid escape string"),
292 : errhint("Escape string must be empty or one character.")));
293 :
294 192 : e = VARDATA_ANY(esc);
295 :
296 : /*
297 : * If specified escape is '\', just copy the pattern as-is.
298 : */
299 192 : if (*e == '\\')
300 : {
301 0 : memcpy(result, pat, VARSIZE_ANY(pat));
302 0 : return result;
303 : }
304 :
305 : /*
306 : * Otherwise, convert occurrences of the specified escape character to
307 : * '\', and double occurrences of '\' --- unless they immediately
308 : * follow an escape character!
309 : */
310 192 : afterescape = false;
311 1164 : while (plen > 0)
312 : {
313 972 : if (CHAREQ(p, e) && !afterescape)
314 : {
315 192 : *r++ = '\\';
316 192 : NextChar(p, plen);
317 192 : afterescape = true;
318 : }
319 780 : else if (*p == '\\')
320 : {
321 0 : *r++ = '\\';
322 0 : if (!afterescape)
323 0 : *r++ = '\\';
324 0 : NextChar(p, plen);
325 0 : afterescape = false;
326 : }
327 : else
328 : {
329 1524 : CopyAdvChar(r, p, plen);
330 780 : afterescape = false;
331 : }
332 : }
333 : }
334 :
335 224 : SET_VARSIZE(result, r - ((char *) result));
336 :
337 224 : return result;
338 : }
339 : #endif /* do_like_escape */
340 :
341 : #ifdef CHAREQ
342 : #undef CHAREQ
343 : #endif
344 :
345 : #undef NextChar
346 : #undef CopyAdvChar
347 : #undef MatchText
348 :
349 : #ifdef do_like_escape
350 : #undef do_like_escape
351 : #endif
352 :
353 : #undef GETCHAR
354 :
355 : #ifdef MATCH_LOWER
356 : #undef MATCH_LOWER
357 :
358 : #endif
|
