Age Owner Branch data TLA 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) MATCH_LOWER(t)
75 : : #else
76 : : #define GETCHAR(t) (t)
77 : : #endif
78 : :
79 : : static int
1892 peter@eisentraut.org 80 :CBC 576361 : MatchText(const char *t, int tlen, const char *p, int plen,
81 : : pg_locale_t locale, bool locale_is_c)
82 : : {
83 : : /* Fast path for match-everything pattern */
5439 tgl@sss.pgh.pa.us 84 [ + + + + ]: 576361 : if (plen == 1 && *p == '%')
8228 ishii@postgresql.org 85 : 401 : return LIKE_TRUE;
86 : :
87 : : /* Since this function recurses, it could be driven to stack overflow */
3117 tgl@sss.pgh.pa.us 88 : 575960 : check_stack_depth();
89 : :
90 : : /*
91 : : * In this loop, we advance by char when matching wildcards (and thus on
92 : : * recursive entry to this function we are properly char-synced). On other
93 : : * occasions it is safe to advance by byte, as the text and pattern will
94 : : * be in lockstep. This allows us to perform all comparisons between the
95 : : * text and pattern on a byte by byte basis, even for multi-byte
96 : : * encodings.
97 : : */
5439 98 [ + + + + ]: 876288 : while (tlen > 0 && plen > 0)
99 : : {
8228 ishii@postgresql.org 100 [ + + ]: 871078 : if (*p == '\\')
101 : : {
102 : : /* Next pattern byte must match literally, whatever it is */
6161 andrew@dunslane.net 103 : 6268 : NextByte(p, plen);
104 : : /* ... and there had better be one, per SQL standard */
5679 tgl@sss.pgh.pa.us 105 [ - + ]: 6268 : if (plen <= 0)
5679 tgl@sss.pgh.pa.us 106 [ # # ]:UBC 0 : ereport(ERROR,
107 : : (errcode(ERRCODE_INVALID_ESCAPE_SEQUENCE),
108 : : errmsg("LIKE pattern must not end with escape character")));
5070 tgl@sss.pgh.pa.us 109 [ + + ]:CBC 6268 : if (GETCHAR(*p) != GETCHAR(*t))
8228 ishii@postgresql.org 110 : 1663 : return LIKE_FALSE;
111 : : }
112 [ + + ]: 864810 : else if (*p == '%')
113 : : {
114 : : char firstpat;
115 : :
116 : : /*
117 : : * % processing is essentially a search for a text position at
118 : : * which the remainder of the text matches the remainder of the
119 : : * pattern, using a recursive call to check each potential match.
120 : : *
121 : : * If there are wildcards immediately following the %, we can skip
122 : : * over them first, using the idea that any sequence of N _'s and
123 : : * one or more %'s is equivalent to N _'s and one % (ie, it will
124 : : * match any sequence of at least N text characters). In this way
125 : : * we will always run the recursive search loop using a pattern
126 : : * fragment that begins with a literal character-to-match, thereby
127 : : * not recursing more than we have to.
128 : : */
5070 tgl@sss.pgh.pa.us 129 : 82975 : NextByte(p, plen);
130 : :
131 [ + + ]: 83222 : while (plen > 0)
132 : : {
133 [ + + ]: 65963 : if (*p == '%')
134 : 3 : NextByte(p, plen);
135 [ + + ]: 65960 : else if (*p == '_')
136 : : {
137 : : /* If not enough text left to match the pattern, ABORT */
138 [ + + ]: 247 : if (tlen <= 0)
139 : 3 : return LIKE_ABORT;
140 [ + + - + ]: 244 : NextChar(t, tlen);
141 : 244 : NextByte(p, plen);
142 : : }
143 : : else
144 : 65713 : break; /* Reached a non-wildcard pattern char */
145 : : }
146 : :
147 : : /*
148 : : * If we're at end of pattern, match: we have a trailing % which
149 : : * matches any remaining text string.
150 : : */
8228 ishii@postgresql.org 151 [ + + ]: 82972 : if (plen <= 0)
152 : 17259 : return LIKE_TRUE;
153 : :
154 : : /*
155 : : * Otherwise, scan for a text position at which we can match the
156 : : * rest of the pattern. The first remaining pattern char is known
157 : : * to be a regular or escaped literal character, so we can compare
158 : : * the first pattern byte to each text byte to avoid recursing
159 : : * more than we have to. This fact also guarantees that we don't
160 : : * have to consider a match to the zero-length substring at the
161 : : * end of the text.
162 : : */
5070 tgl@sss.pgh.pa.us 163 [ + + ]: 65713 : if (*p == '\\')
164 : : {
165 [ - + ]: 2 : if (plen < 2)
5070 tgl@sss.pgh.pa.us 166 [ # # ]:UBC 0 : ereport(ERROR,
167 : : (errcode(ERRCODE_INVALID_ESCAPE_SEQUENCE),
168 : : errmsg("LIKE pattern must not end with escape character")));
5070 tgl@sss.pgh.pa.us 169 :CBC 2 : firstpat = GETCHAR(p[1]);
170 : : }
171 : : else
172 : 65711 : firstpat = GETCHAR(*p);
173 : :
174 [ + + ]: 1922413 : while (tlen > 0)
175 : : {
176 [ + + ]: 1871139 : if (GETCHAR(*t) == firstpat)
177 : : {
4754 178 : 53673 : int matched = MatchText(t, tlen, p, plen,
179 : : locale, locale_is_c);
180 : :
6161 andrew@dunslane.net 181 [ + + ]: 53673 : if (matched != LIKE_FALSE)
5031 bruce@momjian.us 182 : 14439 : return matched; /* TRUE or ABORT */
183 : : }
184 : :
5070 tgl@sss.pgh.pa.us 185 [ + + + + ]: 1856712 : NextChar(t, tlen);
186 : : }
187 : :
188 : : /*
189 : : * End of text with no match, so no point in trying later places
190 : : * to start matching this pattern.
191 : : */
8228 ishii@postgresql.org 192 : 51274 : return LIKE_ABORT;
193 : : }
6161 andrew@dunslane.net 194 [ + + ]: 781835 : else if (*p == '_')
195 : : {
196 : : /* _ matches any single character, and we know there is one */
197 [ + + - + ]: 5336 : NextChar(t, tlen);
198 : 5336 : NextByte(p, plen);
199 : 5336 : continue;
200 : : }
5070 tgl@sss.pgh.pa.us 201 [ + + ]: 776499 : else if (GETCHAR(*p) != GETCHAR(*t))
202 : : {
203 : : /* non-wildcard pattern char fails to match text char */
8228 ishii@postgresql.org 204 : 486112 : return LIKE_FALSE;
205 : : }
206 : :
207 : : /*
208 : : * Pattern and text match, so advance.
209 : : *
210 : : * It is safe to use NextByte instead of NextChar here, even for
211 : : * multi-byte character sets, because we are not following immediately
212 : : * after a wildcard character. If we are in the middle of a multibyte
213 : : * character, we must already have matched at least one byte of the
214 : : * character from both text and pattern; so we cannot get out-of-sync
215 : : * on character boundaries. And we know that no backend-legal
216 : : * encoding allows ASCII characters such as '%' to appear as non-first
217 : : * bytes of characters, so we won't mistakenly detect a new wildcard.
218 : : */
6161 andrew@dunslane.net 219 : 294992 : NextByte(t, tlen);
220 : 294992 : NextByte(p, plen);
221 : : }
222 : :
8228 ishii@postgresql.org 223 [ + + ]: 5210 : if (tlen > 0)
224 : 162 : return LIKE_FALSE; /* end of pattern, but not of text */
225 : :
226 : : /*
227 : : * End of text, but perhaps not of pattern. Match iff the remaining
228 : : * pattern can match a zero-length string, ie, it's zero or more %'s.
229 : : */
5070 tgl@sss.pgh.pa.us 230 [ + + + + ]: 5322 : while (plen > 0 && *p == '%')
6161 andrew@dunslane.net 231 : 274 : NextByte(p, plen);
8228 ishii@postgresql.org 232 [ + + ]: 5048 : if (plen <= 0)
233 : 2301 : return LIKE_TRUE;
234 : :
235 : : /*
236 : : * End of text with no match, so no point in trying later places to start
237 : : * matching this pattern.
238 : : */
239 : 2747 : return LIKE_ABORT;
240 : : } /* MatchText() */
241 : :
242 : : /*
243 : : * like_escape() --- given a pattern and an ESCAPE string,
244 : : * convert the pattern to use Postgres' standard backslash escape convention.
245 : : */
246 : : #ifdef do_like_escape
247 : :
248 : : static text *
249 : 112 : do_like_escape(text *pat, text *esc)
250 : : {
251 : : text *result;
252 : : char *p,
253 : : *e,
254 : : *r;
255 : : int plen,
256 : : elen;
257 : : bool afterescape;
258 : :
6050 tgl@sss.pgh.pa.us 259 [ + + ]: 112 : p = VARDATA_ANY(pat);
260 [ - + - - : 112 : plen = VARSIZE_ANY_EXHDR(pat);
- - - - +
+ ]
261 [ - + ]: 112 : e = VARDATA_ANY(esc);
262 [ - + - - : 112 : elen = VARSIZE_ANY_EXHDR(esc);
- - - - -
+ ]
263 : :
264 : : /*
265 : : * Worst-case pattern growth is 2x --- unlikely, but it's hardly worth
266 : : * trying to calculate the size more accurately than that.
267 : : */
8228 ishii@postgresql.org 268 : 112 : result = (text *) palloc(plen * 2 + VARHDRSZ);
269 : 112 : r = VARDATA(result);
270 : :
271 [ + + ]: 112 : if (elen == 0)
272 : : {
273 : : /*
274 : : * No escape character is wanted. Double any backslashes in the
275 : : * pattern to make them act like ordinary characters.
276 : : */
277 [ + + ]: 64 : while (plen > 0)
278 : : {
279 [ - + ]: 48 : if (*p == '\\')
8228 ishii@postgresql.org 280 :UBC 0 : *r++ = '\\';
8228 ishii@postgresql.org 281 [ + + ]:CBC 96 : CopyAdvChar(r, p, plen);
282 : : }
283 : : }
284 : : else
285 : : {
286 : : /*
287 : : * The specified escape must be only a single character.
288 : : */
289 : 96 : NextChar(e, elen);
290 [ - + ]: 96 : if (elen != 0)
7567 tgl@sss.pgh.pa.us 291 [ # # ]:UBC 0 : ereport(ERROR,
292 : : (errcode(ERRCODE_INVALID_ESCAPE_SEQUENCE),
293 : : errmsg("invalid escape string"),
294 : : errhint("Escape string must be empty or one character.")));
295 : :
6050 tgl@sss.pgh.pa.us 296 [ - + ]:CBC 96 : e = VARDATA_ANY(esc);
297 : :
298 : : /*
299 : : * If specified escape is '\', just copy the pattern as-is.
300 : : */
8228 ishii@postgresql.org 301 [ - + ]: 96 : if (*e == '\\')
302 : : {
6050 tgl@sss.pgh.pa.us 303 [ # # # # :UBC 0 : memcpy(result, pat, VARSIZE_ANY(pat));
# # # # #
# ]
8228 ishii@postgresql.org 304 : 0 : return result;
305 : : }
306 : :
307 : : /*
308 : : * Otherwise, convert occurrences of the specified escape character to
309 : : * '\', and double occurrences of '\' --- unless they immediately
310 : : * follow an escape character!
311 : : */
8228 ishii@postgresql.org 312 :CBC 96 : afterescape = false;
313 [ + + ]: 582 : while (plen > 0)
314 : : {
315 [ + + + + ]: 486 : if (CHAREQ(p, e) && !afterescape)
316 : : {
317 : 96 : *r++ = '\\';
318 : 96 : NextChar(p, plen);
319 : 96 : afterescape = true;
320 : : }
321 [ - + ]: 390 : else if (*p == '\\')
322 : : {
8228 ishii@postgresql.org 323 :UBC 0 : *r++ = '\\';
324 [ # # ]: 0 : if (!afterescape)
325 : 0 : *r++ = '\\';
326 : 0 : NextChar(p, plen);
327 : 0 : afterescape = false;
328 : : }
329 : : else
330 : : {
8228 ishii@postgresql.org 331 [ + + ]:CBC 762 : CopyAdvChar(r, p, plen);
332 : 390 : afterescape = false;
333 : : }
334 : : }
335 : : }
336 : :
6256 tgl@sss.pgh.pa.us 337 : 112 : SET_VARSIZE(result, r - ((char *) result));
338 : :
8228 ishii@postgresql.org 339 : 112 : return result;
340 : : }
341 : : #endif /* do_like_escape */
342 : :
343 : : #ifdef CHAREQ
344 : : #undef CHAREQ
345 : : #endif
346 : :
347 : : #undef NextChar
348 : : #undef CopyAdvChar
349 : : #undef MatchText
350 : :
351 : : #ifdef do_like_escape
352 : : #undef do_like_escape
353 : : #endif
354 : :
355 : : #undef GETCHAR
356 : :
357 : : #ifdef MATCH_LOWER
358 : : #undef MATCH_LOWER
359 : :
360 : : #endif
|
