Commit | Line | Data |
---|---|---|
e318085a | 1 | /* Extended regular expression matching and search library, version |
0b32bf0e | 2 | 0.12. (Implements POSIX draft P1003.2/D11.2, except for some of the |
bc78d348 KB |
3 | internationalization features.) |
4 | ||
0b5538bd | 5 | Copyright (C) 1993, 1994, 1995, 1996, 1997, 1998, 1999, 2000, 2001, |
114f9c96 | 6 | 2002, 2003, 2004, 2005, 2006, 2007, 2008, 2009, 2010 |
e468b87f | 7 | Free Software Foundation, Inc. |
bc78d348 | 8 | |
fa9a63c5 RM |
9 | This program is free software; you can redistribute it and/or modify |
10 | it under the terms of the GNU General Public License as published by | |
e468b87f | 11 | the Free Software Foundation; either version 3, or (at your option) |
fa9a63c5 RM |
12 | any later version. |
13 | ||
14 | This program is distributed in the hope that it will be useful, | |
15 | but WITHOUT ANY WARRANTY; without even the implied warranty of | |
7814e705 | 16 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
fa9a63c5 RM |
17 | GNU General Public License for more details. |
18 | ||
19 | You should have received a copy of the GNU General Public License | |
20 | along with this program; if not, write to the Free Software | |
4fc5845f | 21 | Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, |
7814e705 | 22 | USA. */ |
fa9a63c5 | 23 | |
6df42991 | 24 | /* TODO: |
505bde11 | 25 | - structure the opcode space into opcode+flag. |
dc1e502d | 26 | - merge with glibc's regex.[ch]. |
01618498 | 27 | - replace (succeed_n + jump_n + set_number_at) with something that doesn't |
6dcf2d0e SM |
28 | need to modify the compiled regexp so that re_match can be reentrant. |
29 | - get rid of on_failure_jump_smart by doing the optimization in re_comp | |
30 | rather than at run-time, so that re_match can be reentrant. | |
01618498 | 31 | */ |
505bde11 | 32 | |
fa9a63c5 | 33 | /* AIX requires this to be the first thing in the file. */ |
0b32bf0e | 34 | #if defined _AIX && !defined REGEX_MALLOC |
fa9a63c5 RM |
35 | #pragma alloca |
36 | #endif | |
37 | ||
fa9a63c5 | 38 | #ifdef HAVE_CONFIG_H |
0b32bf0e | 39 | # include <config.h> |
fa9a63c5 RM |
40 | #endif |
41 | ||
4bb91c68 SM |
42 | #if defined STDC_HEADERS && !defined emacs |
43 | # include <stddef.h> | |
44 | #else | |
45 | /* We need this for `regex.h', and perhaps for the Emacs include files. */ | |
46 | # include <sys/types.h> | |
47 | #endif | |
fa9a63c5 | 48 | |
14473664 SM |
49 | /* Whether to use ISO C Amendment 1 wide char functions. |
50 | Those should not be used for Emacs since it uses its own. */ | |
5e5388f6 GM |
51 | #if defined _LIBC |
52 | #define WIDE_CHAR_SUPPORT 1 | |
53 | #else | |
14473664 | 54 | #define WIDE_CHAR_SUPPORT \ |
5e5388f6 GM |
55 | (HAVE_WCTYPE_H && HAVE_WCHAR_H && HAVE_BTOWC && !emacs) |
56 | #endif | |
14473664 SM |
57 | |
58 | /* For platform which support the ISO C amendement 1 functionality we | |
59 | support user defined character classes. */ | |
a0ad02f7 | 60 | #if WIDE_CHAR_SUPPORT |
14473664 SM |
61 | /* Solaris 2.5 has a bug: <wchar.h> must be included before <wctype.h>. */ |
62 | # include <wchar.h> | |
63 | # include <wctype.h> | |
64 | #endif | |
65 | ||
c0f9ea08 SM |
66 | #ifdef _LIBC |
67 | /* We have to keep the namespace clean. */ | |
68 | # define regfree(preg) __regfree (preg) | |
69 | # define regexec(pr, st, nm, pm, ef) __regexec (pr, st, nm, pm, ef) | |
70 | # define regcomp(preg, pattern, cflags) __regcomp (preg, pattern, cflags) | |
ec869672 JR |
71 | # define regerror(err_code, preg, errbuf, errbuf_size) \ |
72 | __regerror(err_code, preg, errbuf, errbuf_size) | |
c0f9ea08 SM |
73 | # define re_set_registers(bu, re, nu, st, en) \ |
74 | __re_set_registers (bu, re, nu, st, en) | |
75 | # define re_match_2(bufp, string1, size1, string2, size2, pos, regs, stop) \ | |
76 | __re_match_2 (bufp, string1, size1, string2, size2, pos, regs, stop) | |
77 | # define re_match(bufp, string, size, pos, regs) \ | |
78 | __re_match (bufp, string, size, pos, regs) | |
79 | # define re_search(bufp, string, size, startpos, range, regs) \ | |
80 | __re_search (bufp, string, size, startpos, range, regs) | |
81 | # define re_compile_pattern(pattern, length, bufp) \ | |
82 | __re_compile_pattern (pattern, length, bufp) | |
83 | # define re_set_syntax(syntax) __re_set_syntax (syntax) | |
84 | # define re_search_2(bufp, st1, s1, st2, s2, startpos, range, regs, stop) \ | |
85 | __re_search_2 (bufp, st1, s1, st2, s2, startpos, range, regs, stop) | |
86 | # define re_compile_fastmap(bufp) __re_compile_fastmap (bufp) | |
87 | ||
14473664 SM |
88 | /* Make sure we call libc's function even if the user overrides them. */ |
89 | # define btowc __btowc | |
90 | # define iswctype __iswctype | |
91 | # define wctype __wctype | |
92 | ||
c0f9ea08 SM |
93 | # define WEAK_ALIAS(a,b) weak_alias (a, b) |
94 | ||
95 | /* We are also using some library internals. */ | |
96 | # include <locale/localeinfo.h> | |
97 | # include <locale/elem-hash.h> | |
98 | # include <langinfo.h> | |
99 | #else | |
100 | # define WEAK_ALIAS(a,b) | |
101 | #endif | |
102 | ||
4bb91c68 | 103 | /* This is for other GNU distributions with internationalized messages. */ |
0b32bf0e | 104 | #if HAVE_LIBINTL_H || defined _LIBC |
fa9a63c5 RM |
105 | # include <libintl.h> |
106 | #else | |
107 | # define gettext(msgid) (msgid) | |
108 | #endif | |
109 | ||
5e69f11e RM |
110 | #ifndef gettext_noop |
111 | /* This define is so xgettext can find the internationalizable | |
112 | strings. */ | |
0b32bf0e | 113 | # define gettext_noop(String) String |
5e69f11e RM |
114 | #endif |
115 | ||
fa9a63c5 RM |
116 | /* The `emacs' switch turns on certain matching commands |
117 | that make sense only in Emacs. */ | |
118 | #ifdef emacs | |
119 | ||
d7306fe6 | 120 | # include <setjmp.h> |
0b32bf0e SM |
121 | # include "lisp.h" |
122 | # include "buffer.h" | |
b18215fc RS |
123 | |
124 | /* Make syntax table lookup grant data in gl_state. */ | |
0b32bf0e | 125 | # define SYNTAX_ENTRY_VIA_PROPERTY |
b18215fc | 126 | |
0b32bf0e | 127 | # include "syntax.h" |
9117d724 | 128 | # include "character.h" |
0b32bf0e | 129 | # include "category.h" |
fa9a63c5 | 130 | |
7689ef0b EZ |
131 | # ifdef malloc |
132 | # undef malloc | |
133 | # endif | |
0b32bf0e | 134 | # define malloc xmalloc |
7689ef0b EZ |
135 | # ifdef realloc |
136 | # undef realloc | |
137 | # endif | |
0b32bf0e | 138 | # define realloc xrealloc |
7689ef0b EZ |
139 | # ifdef free |
140 | # undef free | |
141 | # endif | |
0b32bf0e | 142 | # define free xfree |
9abbd165 | 143 | |
7814e705 | 144 | /* Converts the pointer to the char to BEG-based offset from the start. */ |
0b32bf0e SM |
145 | # define PTR_TO_OFFSET(d) POS_AS_IN_BUFFER (POINTER_TO_OFFSET (d)) |
146 | # define POS_AS_IN_BUFFER(p) ((p) + (NILP (re_match_object) || BUFFERP (re_match_object))) | |
147 | ||
148 | # define RE_MULTIBYTE_P(bufp) ((bufp)->multibyte) | |
bf216479 | 149 | # define RE_TARGET_MULTIBYTE_P(bufp) ((bufp)->target_multibyte) |
62a6e103 AS |
150 | # define RE_STRING_CHAR(p, multibyte) \ |
151 | (multibyte ? (STRING_CHAR (p)) : (*(p))) | |
152 | # define RE_STRING_CHAR_AND_LENGTH(p, len, multibyte) \ | |
153 | (multibyte ? (STRING_CHAR_AND_LENGTH (p, len)) : ((len) = 1, *(p))) | |
2d1675e4 | 154 | |
4c0354d7 | 155 | # define RE_CHAR_TO_MULTIBYTE(c) UNIBYTE_TO_CHAR (c) |
cf9c99bc | 156 | |
2afc21f5 | 157 | # define RE_CHAR_TO_UNIBYTE(c) CHAR_TO_BYTE_SAFE (c) |
cf9c99bc | 158 | |
6fdd04b0 KH |
159 | /* Set C a (possibly converted to multibyte) character before P. P |
160 | points into a string which is the virtual concatenation of STR1 | |
161 | (which ends at END1) or STR2 (which ends at END2). */ | |
bf216479 KH |
162 | # define GET_CHAR_BEFORE_2(c, p, str1, end1, str2, end2) \ |
163 | do { \ | |
02cb78b5 | 164 | if (target_multibyte) \ |
bf216479 KH |
165 | { \ |
166 | re_char *dtemp = (p) == (str2) ? (end1) : (p); \ | |
167 | re_char *dlimit = ((p) > (str2) && (p) <= (end2)) ? (str2) : (str1); \ | |
168 | while (dtemp-- > dlimit && !CHAR_HEAD_P (*dtemp)); \ | |
62a6e103 | 169 | c = STRING_CHAR (dtemp); \ |
bf216479 KH |
170 | } \ |
171 | else \ | |
172 | { \ | |
173 | (c = ((p) == (str2) ? (end1) : (p))[-1]); \ | |
cf9c99bc | 174 | (c) = RE_CHAR_TO_MULTIBYTE (c); \ |
bf216479 | 175 | } \ |
2d1675e4 SM |
176 | } while (0) |
177 | ||
6fdd04b0 KH |
178 | /* Set C a (possibly converted to multibyte) character at P, and set |
179 | LEN to the byte length of that character. */ | |
180 | # define GET_CHAR_AFTER(c, p, len) \ | |
181 | do { \ | |
02cb78b5 | 182 | if (target_multibyte) \ |
62a6e103 | 183 | (c) = STRING_CHAR_AND_LENGTH (p, len); \ |
6fdd04b0 KH |
184 | else \ |
185 | { \ | |
cf9c99bc | 186 | (c) = *p; \ |
6fdd04b0 | 187 | len = 1; \ |
cf9c99bc | 188 | (c) = RE_CHAR_TO_MULTIBYTE (c); \ |
6fdd04b0 | 189 | } \ |
8f924df7 | 190 | } while (0) |
4e8a9132 | 191 | |
fa9a63c5 RM |
192 | #else /* not emacs */ |
193 | ||
194 | /* If we are not linking with Emacs proper, | |
195 | we can't use the relocating allocator | |
196 | even if config.h says that we can. */ | |
0b32bf0e | 197 | # undef REL_ALLOC |
fa9a63c5 | 198 | |
d2762c86 DN |
199 | # ifdef HAVE_UNISTD_H |
200 | # include <unistd.h> | |
0b32bf0e | 201 | # endif |
fa9a63c5 | 202 | |
a77f947b CY |
203 | /* When used in Emacs's lib-src, we need xmalloc and xrealloc. */ |
204 | ||
205 | void * | |
d2762c86 | 206 | xmalloc (size_t size) |
a77f947b CY |
207 | { |
208 | register void *val; | |
209 | val = (void *) malloc (size); | |
210 | if (!val && size) | |
211 | { | |
212 | write (2, "virtual memory exhausted\n", 25); | |
213 | exit (1); | |
214 | } | |
215 | return val; | |
216 | } | |
217 | ||
218 | void * | |
d2762c86 | 219 | xrealloc (void *block, size_t size) |
a77f947b CY |
220 | { |
221 | register void *val; | |
222 | /* We must call malloc explicitly when BLOCK is 0, since some | |
223 | reallocs don't do this. */ | |
224 | if (! block) | |
225 | val = (void *) malloc (size); | |
226 | else | |
227 | val = (void *) realloc (block, size); | |
228 | if (!val && size) | |
229 | { | |
230 | write (2, "virtual memory exhausted\n", 25); | |
231 | exit (1); | |
232 | } | |
233 | return val; | |
234 | } | |
235 | ||
a073faa6 CY |
236 | # ifdef malloc |
237 | # undef malloc | |
238 | # endif | |
239 | # define malloc xmalloc | |
240 | # ifdef realloc | |
241 | # undef realloc | |
242 | # endif | |
243 | # define realloc xrealloc | |
244 | ||
72af86bd AS |
245 | /* This is the normal way of making sure we have memcpy, memcmp and memset. */ |
246 | # if defined HAVE_STRING_H || defined STDC_HEADERS || defined _LIBC | |
247 | # include <string.h> | |
248 | # else | |
249 | # include <strings.h> | |
250 | # ifndef memcmp | |
251 | # define memcmp(s1, s2, n) bcmp (s1, s2, n) | |
0b32bf0e | 252 | # endif |
72af86bd AS |
253 | # ifndef memcpy |
254 | # define memcpy(d, s, n) (bcopy (s, d, n), (d)) | |
0b32bf0e SM |
255 | # endif |
256 | # endif | |
fa9a63c5 RM |
257 | |
258 | /* Define the syntax stuff for \<, \>, etc. */ | |
259 | ||
990b2375 | 260 | /* Sword must be nonzero for the wordchar pattern commands in re_match_2. */ |
669fa600 | 261 | enum syntaxcode { Swhitespace = 0, Sword = 1, Ssymbol = 2 }; |
fa9a63c5 | 262 | |
0b32bf0e | 263 | # define SWITCH_ENUM_CAST(x) (x) |
fa9a63c5 | 264 | |
e934739e | 265 | /* Dummy macros for non-Emacs environments. */ |
0b32bf0e SM |
266 | # define CHAR_CHARSET(c) 0 |
267 | # define CHARSET_LEADING_CODE_BASE(c) 0 | |
268 | # define MAX_MULTIBYTE_LENGTH 1 | |
269 | # define RE_MULTIBYTE_P(x) 0 | |
bf216479 | 270 | # define RE_TARGET_MULTIBYTE_P(x) 0 |
0b32bf0e SM |
271 | # define WORD_BOUNDARY_P(c1, c2) (0) |
272 | # define CHAR_HEAD_P(p) (1) | |
273 | # define SINGLE_BYTE_CHAR_P(c) (1) | |
274 | # define SAME_CHARSET_P(c1, c2) (1) | |
aa3830c4 | 275 | # define BYTES_BY_CHAR_HEAD(p) (1) |
70806df6 | 276 | # define PREV_CHAR_BOUNDARY(p, limit) ((p)--) |
62a6e103 AS |
277 | # define STRING_CHAR(p) (*(p)) |
278 | # define RE_STRING_CHAR(p, multibyte) STRING_CHAR (p) | |
0b32bf0e | 279 | # define CHAR_STRING(c, s) (*(s) = (c), 1) |
62a6e103 AS |
280 | # define STRING_CHAR_AND_LENGTH(p, actual_len) ((actual_len) = 1, *(p)) |
281 | # define RE_STRING_CHAR_AND_LENGTH(p, len, multibyte) STRING_CHAR_AND_LENGTH (p, len) | |
cf9c99bc KH |
282 | # define RE_CHAR_TO_MULTIBYTE(c) (c) |
283 | # define RE_CHAR_TO_UNIBYTE(c) (c) | |
0b32bf0e | 284 | # define GET_CHAR_BEFORE_2(c, p, str1, end1, str2, end2) \ |
b18215fc | 285 | (c = ((p) == (str2) ? *((end1) - 1) : *((p) - 1))) |
6fdd04b0 KH |
286 | # define GET_CHAR_AFTER(c, p, len) \ |
287 | (c = *p, len = 1) | |
0b32bf0e | 288 | # define MAKE_CHAR(charset, c1, c2) (c1) |
9117d724 KH |
289 | # define BYTE8_TO_CHAR(c) (c) |
290 | # define CHAR_BYTE8_P(c) (0) | |
bf216479 | 291 | # define CHAR_LEADING_CODE(c) (c) |
8f924df7 | 292 | |
fa9a63c5 | 293 | #endif /* not emacs */ |
4e8a9132 SM |
294 | |
295 | #ifndef RE_TRANSLATE | |
0b32bf0e SM |
296 | # define RE_TRANSLATE(TBL, C) ((unsigned char)(TBL)[C]) |
297 | # define RE_TRANSLATE_P(TBL) (TBL) | |
4e8a9132 | 298 | #endif |
fa9a63c5 RM |
299 | \f |
300 | /* Get the interface, including the syntax bits. */ | |
301 | #include "regex.h" | |
302 | ||
f71b19b6 DL |
303 | /* isalpha etc. are used for the character classes. */ |
304 | #include <ctype.h> | |
fa9a63c5 | 305 | |
f71b19b6 | 306 | #ifdef emacs |
fa9a63c5 | 307 | |
f71b19b6 | 308 | /* 1 if C is an ASCII character. */ |
0b32bf0e | 309 | # define IS_REAL_ASCII(c) ((c) < 0200) |
fa9a63c5 | 310 | |
f71b19b6 | 311 | /* 1 if C is a unibyte character. */ |
0b32bf0e | 312 | # define ISUNIBYTE(c) (SINGLE_BYTE_CHAR_P ((c))) |
96cc36cc | 313 | |
f71b19b6 | 314 | /* The Emacs definitions should not be directly affected by locales. */ |
96cc36cc | 315 | |
f71b19b6 | 316 | /* In Emacs, these are only used for single-byte characters. */ |
0b32bf0e SM |
317 | # define ISDIGIT(c) ((c) >= '0' && (c) <= '9') |
318 | # define ISCNTRL(c) ((c) < ' ') | |
319 | # define ISXDIGIT(c) (((c) >= '0' && (c) <= '9') \ | |
f71b19b6 DL |
320 | || ((c) >= 'a' && (c) <= 'f') \ |
321 | || ((c) >= 'A' && (c) <= 'F')) | |
96cc36cc RS |
322 | |
323 | /* This is only used for single-byte characters. */ | |
0b32bf0e | 324 | # define ISBLANK(c) ((c) == ' ' || (c) == '\t') |
96cc36cc RS |
325 | |
326 | /* The rest must handle multibyte characters. */ | |
327 | ||
0b32bf0e | 328 | # define ISGRAPH(c) (SINGLE_BYTE_CHAR_P (c) \ |
f71b19b6 | 329 | ? (c) > ' ' && !((c) >= 0177 && (c) <= 0237) \ |
96cc36cc RS |
330 | : 1) |
331 | ||
14473664 | 332 | # define ISPRINT(c) (SINGLE_BYTE_CHAR_P (c) \ |
f71b19b6 | 333 | ? (c) >= ' ' && !((c) >= 0177 && (c) <= 0237) \ |
96cc36cc RS |
334 | : 1) |
335 | ||
0b32bf0e | 336 | # define ISALNUM(c) (IS_REAL_ASCII (c) \ |
f71b19b6 DL |
337 | ? (((c) >= 'a' && (c) <= 'z') \ |
338 | || ((c) >= 'A' && (c) <= 'Z') \ | |
339 | || ((c) >= '0' && (c) <= '9')) \ | |
96cc36cc RS |
340 | : SYNTAX (c) == Sword) |
341 | ||
0b32bf0e | 342 | # define ISALPHA(c) (IS_REAL_ASCII (c) \ |
f71b19b6 DL |
343 | ? (((c) >= 'a' && (c) <= 'z') \ |
344 | || ((c) >= 'A' && (c) <= 'Z')) \ | |
96cc36cc RS |
345 | : SYNTAX (c) == Sword) |
346 | ||
0b32bf0e | 347 | # define ISLOWER(c) (LOWERCASEP (c)) |
96cc36cc | 348 | |
0b32bf0e | 349 | # define ISPUNCT(c) (IS_REAL_ASCII (c) \ |
f71b19b6 DL |
350 | ? ((c) > ' ' && (c) < 0177 \ |
351 | && !(((c) >= 'a' && (c) <= 'z') \ | |
4bb91c68 SM |
352 | || ((c) >= 'A' && (c) <= 'Z') \ |
353 | || ((c) >= '0' && (c) <= '9'))) \ | |
96cc36cc RS |
354 | : SYNTAX (c) != Sword) |
355 | ||
0b32bf0e | 356 | # define ISSPACE(c) (SYNTAX (c) == Swhitespace) |
96cc36cc | 357 | |
0b32bf0e | 358 | # define ISUPPER(c) (UPPERCASEP (c)) |
96cc36cc | 359 | |
0b32bf0e | 360 | # define ISWORD(c) (SYNTAX (c) == Sword) |
96cc36cc RS |
361 | |
362 | #else /* not emacs */ | |
363 | ||
f71b19b6 DL |
364 | /* Jim Meyering writes: |
365 | ||
366 | "... Some ctype macros are valid only for character codes that | |
367 | isascii says are ASCII (SGI's IRIX-4.0.5 is one such system --when | |
368 | using /bin/cc or gcc but without giving an ansi option). So, all | |
4bb91c68 | 369 | ctype uses should be through macros like ISPRINT... If |
f71b19b6 DL |
370 | STDC_HEADERS is defined, then autoconf has verified that the ctype |
371 | macros don't need to be guarded with references to isascii. ... | |
372 | Defining isascii to 1 should let any compiler worth its salt | |
4bb91c68 SM |
373 | eliminate the && through constant folding." |
374 | Solaris defines some of these symbols so we must undefine them first. */ | |
f71b19b6 | 375 | |
4bb91c68 | 376 | # undef ISASCII |
0b32bf0e SM |
377 | # if defined STDC_HEADERS || (!defined isascii && !defined HAVE_ISASCII) |
378 | # define ISASCII(c) 1 | |
379 | # else | |
380 | # define ISASCII(c) isascii(c) | |
381 | # endif | |
f71b19b6 DL |
382 | |
383 | /* 1 if C is an ASCII character. */ | |
0b32bf0e | 384 | # define IS_REAL_ASCII(c) ((c) < 0200) |
f71b19b6 DL |
385 | |
386 | /* This distinction is not meaningful, except in Emacs. */ | |
0b32bf0e SM |
387 | # define ISUNIBYTE(c) 1 |
388 | ||
389 | # ifdef isblank | |
390 | # define ISBLANK(c) (ISASCII (c) && isblank (c)) | |
391 | # else | |
392 | # define ISBLANK(c) ((c) == ' ' || (c) == '\t') | |
393 | # endif | |
394 | # ifdef isgraph | |
395 | # define ISGRAPH(c) (ISASCII (c) && isgraph (c)) | |
396 | # else | |
397 | # define ISGRAPH(c) (ISASCII (c) && isprint (c) && !isspace (c)) | |
398 | # endif | |
399 | ||
4bb91c68 | 400 | # undef ISPRINT |
0b32bf0e SM |
401 | # define ISPRINT(c) (ISASCII (c) && isprint (c)) |
402 | # define ISDIGIT(c) (ISASCII (c) && isdigit (c)) | |
403 | # define ISALNUM(c) (ISASCII (c) && isalnum (c)) | |
404 | # define ISALPHA(c) (ISASCII (c) && isalpha (c)) | |
405 | # define ISCNTRL(c) (ISASCII (c) && iscntrl (c)) | |
406 | # define ISLOWER(c) (ISASCII (c) && islower (c)) | |
407 | # define ISPUNCT(c) (ISASCII (c) && ispunct (c)) | |
408 | # define ISSPACE(c) (ISASCII (c) && isspace (c)) | |
409 | # define ISUPPER(c) (ISASCII (c) && isupper (c)) | |
410 | # define ISXDIGIT(c) (ISASCII (c) && isxdigit (c)) | |
411 | ||
412 | # define ISWORD(c) ISALPHA(c) | |
413 | ||
4bb91c68 SM |
414 | # ifdef _tolower |
415 | # define TOLOWER(c) _tolower(c) | |
416 | # else | |
417 | # define TOLOWER(c) tolower(c) | |
418 | # endif | |
419 | ||
420 | /* How many characters in the character set. */ | |
421 | # define CHAR_SET_SIZE 256 | |
422 | ||
0b32bf0e | 423 | # ifdef SYNTAX_TABLE |
f71b19b6 | 424 | |
0b32bf0e | 425 | extern char *re_syntax_table; |
f71b19b6 | 426 | |
0b32bf0e SM |
427 | # else /* not SYNTAX_TABLE */ |
428 | ||
0b32bf0e SM |
429 | static char re_syntax_table[CHAR_SET_SIZE]; |
430 | ||
431 | static void | |
d2762c86 | 432 | init_syntax_once (void) |
0b32bf0e SM |
433 | { |
434 | register int c; | |
435 | static int done = 0; | |
436 | ||
437 | if (done) | |
438 | return; | |
439 | ||
72af86bd | 440 | memset (re_syntax_table, 0, sizeof re_syntax_table); |
0b32bf0e | 441 | |
4bb91c68 SM |
442 | for (c = 0; c < CHAR_SET_SIZE; ++c) |
443 | if (ISALNUM (c)) | |
444 | re_syntax_table[c] = Sword; | |
fa9a63c5 | 445 | |
669fa600 | 446 | re_syntax_table['_'] = Ssymbol; |
fa9a63c5 | 447 | |
0b32bf0e SM |
448 | done = 1; |
449 | } | |
450 | ||
451 | # endif /* not SYNTAX_TABLE */ | |
96cc36cc | 452 | |
4bb91c68 SM |
453 | # define SYNTAX(c) re_syntax_table[(c)] |
454 | ||
96cc36cc RS |
455 | #endif /* not emacs */ |
456 | \f | |
fa9a63c5 | 457 | #ifndef NULL |
0b32bf0e | 458 | # define NULL (void *)0 |
fa9a63c5 RM |
459 | #endif |
460 | ||
461 | /* We remove any previous definition of `SIGN_EXTEND_CHAR', | |
462 | since ours (we hope) works properly with all combinations of | |
463 | machines, compilers, `char' and `unsigned char' argument types. | |
4bb91c68 | 464 | (Per Bothner suggested the basic approach.) */ |
fa9a63c5 RM |
465 | #undef SIGN_EXTEND_CHAR |
466 | #if __STDC__ | |
0b32bf0e | 467 | # define SIGN_EXTEND_CHAR(c) ((signed char) (c)) |
fa9a63c5 RM |
468 | #else /* not __STDC__ */ |
469 | /* As in Harbison and Steele. */ | |
0b32bf0e | 470 | # define SIGN_EXTEND_CHAR(c) ((((unsigned char) (c)) ^ 128) - 128) |
fa9a63c5 RM |
471 | #endif |
472 | \f | |
473 | /* Should we use malloc or alloca? If REGEX_MALLOC is not defined, we | |
474 | use `alloca' instead of `malloc'. This is because using malloc in | |
475 | re_search* or re_match* could cause memory leaks when C-g is used in | |
476 | Emacs; also, malloc is slower and causes storage fragmentation. On | |
5e69f11e RM |
477 | the other hand, malloc is more portable, and easier to debug. |
478 | ||
fa9a63c5 RM |
479 | Because we sometimes use alloca, some routines have to be macros, |
480 | not functions -- `alloca'-allocated space disappears at the end of the | |
481 | function it is called in. */ | |
482 | ||
483 | #ifdef REGEX_MALLOC | |
484 | ||
0b32bf0e SM |
485 | # define REGEX_ALLOCATE malloc |
486 | # define REGEX_REALLOCATE(source, osize, nsize) realloc (source, nsize) | |
487 | # define REGEX_FREE free | |
fa9a63c5 RM |
488 | |
489 | #else /* not REGEX_MALLOC */ | |
490 | ||
491 | /* Emacs already defines alloca, sometimes. */ | |
0b32bf0e | 492 | # ifndef alloca |
fa9a63c5 RM |
493 | |
494 | /* Make alloca work the best possible way. */ | |
0b32bf0e SM |
495 | # ifdef __GNUC__ |
496 | # define alloca __builtin_alloca | |
497 | # else /* not __GNUC__ */ | |
7f585e7a | 498 | # ifdef HAVE_ALLOCA_H |
0b32bf0e SM |
499 | # include <alloca.h> |
500 | # endif /* HAVE_ALLOCA_H */ | |
501 | # endif /* not __GNUC__ */ | |
fa9a63c5 | 502 | |
0b32bf0e | 503 | # endif /* not alloca */ |
fa9a63c5 | 504 | |
0b32bf0e | 505 | # define REGEX_ALLOCATE alloca |
fa9a63c5 RM |
506 | |
507 | /* Assumes a `char *destination' variable. */ | |
0b32bf0e | 508 | # define REGEX_REALLOCATE(source, osize, nsize) \ |
fa9a63c5 | 509 | (destination = (char *) alloca (nsize), \ |
4bb91c68 | 510 | memcpy (destination, source, osize)) |
fa9a63c5 RM |
511 | |
512 | /* No need to do anything to free, after alloca. */ | |
0b32bf0e | 513 | # define REGEX_FREE(arg) ((void)0) /* Do nothing! But inhibit gcc warning. */ |
fa9a63c5 RM |
514 | |
515 | #endif /* not REGEX_MALLOC */ | |
516 | ||
517 | /* Define how to allocate the failure stack. */ | |
518 | ||
0b32bf0e | 519 | #if defined REL_ALLOC && defined REGEX_MALLOC |
4297555e | 520 | |
0b32bf0e | 521 | # define REGEX_ALLOCATE_STACK(size) \ |
fa9a63c5 | 522 | r_alloc (&failure_stack_ptr, (size)) |
0b32bf0e | 523 | # define REGEX_REALLOCATE_STACK(source, osize, nsize) \ |
fa9a63c5 | 524 | r_re_alloc (&failure_stack_ptr, (nsize)) |
0b32bf0e | 525 | # define REGEX_FREE_STACK(ptr) \ |
fa9a63c5 RM |
526 | r_alloc_free (&failure_stack_ptr) |
527 | ||
4297555e | 528 | #else /* not using relocating allocator */ |
fa9a63c5 | 529 | |
0b32bf0e | 530 | # ifdef REGEX_MALLOC |
fa9a63c5 | 531 | |
0b32bf0e SM |
532 | # define REGEX_ALLOCATE_STACK malloc |
533 | # define REGEX_REALLOCATE_STACK(source, osize, nsize) realloc (source, nsize) | |
534 | # define REGEX_FREE_STACK free | |
fa9a63c5 | 535 | |
0b32bf0e | 536 | # else /* not REGEX_MALLOC */ |
fa9a63c5 | 537 | |
0b32bf0e | 538 | # define REGEX_ALLOCATE_STACK alloca |
fa9a63c5 | 539 | |
0b32bf0e | 540 | # define REGEX_REALLOCATE_STACK(source, osize, nsize) \ |
fa9a63c5 | 541 | REGEX_REALLOCATE (source, osize, nsize) |
7814e705 | 542 | /* No need to explicitly free anything. */ |
0b32bf0e | 543 | # define REGEX_FREE_STACK(arg) ((void)0) |
fa9a63c5 | 544 | |
0b32bf0e | 545 | # endif /* not REGEX_MALLOC */ |
4297555e | 546 | #endif /* not using relocating allocator */ |
fa9a63c5 RM |
547 | |
548 | ||
549 | /* True if `size1' is non-NULL and PTR is pointing anywhere inside | |
550 | `string1' or just past its end. This works if PTR is NULL, which is | |
551 | a good thing. */ | |
25fe55af | 552 | #define FIRST_STRING_P(ptr) \ |
fa9a63c5 RM |
553 | (size1 && string1 <= (ptr) && (ptr) <= string1 + size1) |
554 | ||
555 | /* (Re)Allocate N items of type T using malloc, or fail. */ | |
556 | #define TALLOC(n, t) ((t *) malloc ((n) * sizeof (t))) | |
557 | #define RETALLOC(addr, n, t) ((addr) = (t *) realloc (addr, (n) * sizeof (t))) | |
558 | #define RETALLOC_IF(addr, n, t) \ | |
559 | if (addr) RETALLOC((addr), (n), t); else (addr) = TALLOC ((n), t) | |
560 | #define REGEX_TALLOC(n, t) ((t *) REGEX_ALLOCATE ((n) * sizeof (t))) | |
561 | ||
4bb91c68 | 562 | #define BYTEWIDTH 8 /* In bits. */ |
fa9a63c5 RM |
563 | |
564 | #define STREQ(s1, s2) ((strcmp (s1, s2) == 0)) | |
565 | ||
566 | #undef MAX | |
567 | #undef MIN | |
568 | #define MAX(a, b) ((a) > (b) ? (a) : (b)) | |
569 | #define MIN(a, b) ((a) < (b) ? (a) : (b)) | |
570 | ||
66f0296e SM |
571 | /* Type of source-pattern and string chars. */ |
572 | typedef const unsigned char re_char; | |
573 | ||
fa9a63c5 RM |
574 | typedef char boolean; |
575 | #define false 0 | |
576 | #define true 1 | |
577 | ||
4bb91c68 SM |
578 | static int re_match_2_internal _RE_ARGS ((struct re_pattern_buffer *bufp, |
579 | re_char *string1, int size1, | |
580 | re_char *string2, int size2, | |
581 | int pos, | |
582 | struct re_registers *regs, | |
583 | int stop)); | |
fa9a63c5 RM |
584 | \f |
585 | /* These are the command codes that appear in compiled regular | |
4bb91c68 | 586 | expressions. Some opcodes are followed by argument bytes. A |
fa9a63c5 RM |
587 | command code can specify any interpretation whatsoever for its |
588 | arguments. Zero bytes may appear in the compiled regular expression. */ | |
589 | ||
590 | typedef enum | |
591 | { | |
592 | no_op = 0, | |
593 | ||
4bb91c68 | 594 | /* Succeed right away--no more backtracking. */ |
fa9a63c5 RM |
595 | succeed, |
596 | ||
25fe55af | 597 | /* Followed by one byte giving n, then by n literal bytes. */ |
fa9a63c5 RM |
598 | exactn, |
599 | ||
25fe55af | 600 | /* Matches any (more or less) character. */ |
fa9a63c5 RM |
601 | anychar, |
602 | ||
25fe55af RS |
603 | /* Matches any one char belonging to specified set. First |
604 | following byte is number of bitmap bytes. Then come bytes | |
605 | for a bitmap saying which chars are in. Bits in each byte | |
606 | are ordered low-bit-first. A character is in the set if its | |
607 | bit is 1. A character too large to have a bit in the map is | |
96cc36cc RS |
608 | automatically not in the set. |
609 | ||
610 | If the length byte has the 0x80 bit set, then that stuff | |
611 | is followed by a range table: | |
612 | 2 bytes of flags for character sets (low 8 bits, high 8 bits) | |
0b32bf0e | 613 | See RANGE_TABLE_WORK_BITS below. |
01618498 | 614 | 2 bytes, the number of pairs that follow (upto 32767) |
96cc36cc | 615 | pairs, each 2 multibyte characters, |
0b32bf0e | 616 | each multibyte character represented as 3 bytes. */ |
fa9a63c5 RM |
617 | charset, |
618 | ||
25fe55af | 619 | /* Same parameters as charset, but match any character that is |
4bb91c68 | 620 | not one of those specified. */ |
fa9a63c5 RM |
621 | charset_not, |
622 | ||
25fe55af RS |
623 | /* Start remembering the text that is matched, for storing in a |
624 | register. Followed by one byte with the register number, in | |
625 | the range 0 to one less than the pattern buffer's re_nsub | |
505bde11 | 626 | field. */ |
fa9a63c5 RM |
627 | start_memory, |
628 | ||
25fe55af RS |
629 | /* Stop remembering the text that is matched and store it in a |
630 | memory register. Followed by one byte with the register | |
631 | number, in the range 0 to one less than `re_nsub' in the | |
505bde11 | 632 | pattern buffer. */ |
fa9a63c5 RM |
633 | stop_memory, |
634 | ||
25fe55af | 635 | /* Match a duplicate of something remembered. Followed by one |
4bb91c68 | 636 | byte containing the register number. */ |
fa9a63c5 RM |
637 | duplicate, |
638 | ||
25fe55af | 639 | /* Fail unless at beginning of line. */ |
fa9a63c5 RM |
640 | begline, |
641 | ||
4bb91c68 | 642 | /* Fail unless at end of line. */ |
fa9a63c5 RM |
643 | endline, |
644 | ||
25fe55af RS |
645 | /* Succeeds if at beginning of buffer (if emacs) or at beginning |
646 | of string to be matched (if not). */ | |
fa9a63c5 RM |
647 | begbuf, |
648 | ||
25fe55af | 649 | /* Analogously, for end of buffer/string. */ |
fa9a63c5 | 650 | endbuf, |
5e69f11e | 651 | |
25fe55af | 652 | /* Followed by two byte relative address to which to jump. */ |
5e69f11e | 653 | jump, |
fa9a63c5 | 654 | |
25fe55af | 655 | /* Followed by two-byte relative address of place to resume at |
7814e705 | 656 | in case of failure. */ |
fa9a63c5 | 657 | on_failure_jump, |
5e69f11e | 658 | |
25fe55af RS |
659 | /* Like on_failure_jump, but pushes a placeholder instead of the |
660 | current string position when executed. */ | |
fa9a63c5 | 661 | on_failure_keep_string_jump, |
5e69f11e | 662 | |
505bde11 SM |
663 | /* Just like `on_failure_jump', except that it checks that we |
664 | don't get stuck in an infinite loop (matching an empty string | |
665 | indefinitely). */ | |
666 | on_failure_jump_loop, | |
667 | ||
0683b6fa SM |
668 | /* Just like `on_failure_jump_loop', except that it checks for |
669 | a different kind of loop (the kind that shows up with non-greedy | |
670 | operators). This operation has to be immediately preceded | |
671 | by a `no_op'. */ | |
672 | on_failure_jump_nastyloop, | |
673 | ||
0b32bf0e | 674 | /* A smart `on_failure_jump' used for greedy * and + operators. |
505bde11 SM |
675 | It analyses the loop before which it is put and if the |
676 | loop does not require backtracking, it changes itself to | |
4e8a9132 SM |
677 | `on_failure_keep_string_jump' and short-circuits the loop, |
678 | else it just defaults to changing itself into `on_failure_jump'. | |
679 | It assumes that it is pointing to just past a `jump'. */ | |
505bde11 | 680 | on_failure_jump_smart, |
fa9a63c5 | 681 | |
25fe55af | 682 | /* Followed by two-byte relative address and two-byte number n. |
ed0767d8 SM |
683 | After matching N times, jump to the address upon failure. |
684 | Does not work if N starts at 0: use on_failure_jump_loop | |
685 | instead. */ | |
fa9a63c5 RM |
686 | succeed_n, |
687 | ||
25fe55af RS |
688 | /* Followed by two-byte relative address, and two-byte number n. |
689 | Jump to the address N times, then fail. */ | |
fa9a63c5 RM |
690 | jump_n, |
691 | ||
25fe55af | 692 | /* Set the following two-byte relative address to the |
7814e705 | 693 | subsequent two-byte number. The address *includes* the two |
25fe55af | 694 | bytes of number. */ |
fa9a63c5 RM |
695 | set_number_at, |
696 | ||
fa9a63c5 RM |
697 | wordbeg, /* Succeeds if at word beginning. */ |
698 | wordend, /* Succeeds if at word end. */ | |
699 | ||
700 | wordbound, /* Succeeds if at a word boundary. */ | |
7814e705 | 701 | notwordbound, /* Succeeds if not at a word boundary. */ |
fa9a63c5 | 702 | |
669fa600 SM |
703 | symbeg, /* Succeeds if at symbol beginning. */ |
704 | symend, /* Succeeds if at symbol end. */ | |
705 | ||
fa9a63c5 | 706 | /* Matches any character whose syntax is specified. Followed by |
25fe55af | 707 | a byte which contains a syntax code, e.g., Sword. */ |
fa9a63c5 RM |
708 | syntaxspec, |
709 | ||
710 | /* Matches any character whose syntax is not that specified. */ | |
1fb352e0 SM |
711 | notsyntaxspec |
712 | ||
713 | #ifdef emacs | |
714 | ,before_dot, /* Succeeds if before point. */ | |
715 | at_dot, /* Succeeds if at point. */ | |
716 | after_dot, /* Succeeds if after point. */ | |
b18215fc RS |
717 | |
718 | /* Matches any character whose category-set contains the specified | |
7814e705 JB |
719 | category. The operator is followed by a byte which contains a |
720 | category code (mnemonic ASCII character). */ | |
b18215fc RS |
721 | categoryspec, |
722 | ||
723 | /* Matches any character whose category-set does not contain the | |
724 | specified category. The operator is followed by a byte which | |
725 | contains the category code (mnemonic ASCII character). */ | |
726 | notcategoryspec | |
fa9a63c5 RM |
727 | #endif /* emacs */ |
728 | } re_opcode_t; | |
729 | \f | |
730 | /* Common operations on the compiled pattern. */ | |
731 | ||
732 | /* Store NUMBER in two contiguous bytes starting at DESTINATION. */ | |
733 | ||
734 | #define STORE_NUMBER(destination, number) \ | |
735 | do { \ | |
736 | (destination)[0] = (number) & 0377; \ | |
737 | (destination)[1] = (number) >> 8; \ | |
738 | } while (0) | |
739 | ||
740 | /* Same as STORE_NUMBER, except increment DESTINATION to | |
741 | the byte after where the number is stored. Therefore, DESTINATION | |
742 | must be an lvalue. */ | |
743 | ||
744 | #define STORE_NUMBER_AND_INCR(destination, number) \ | |
745 | do { \ | |
746 | STORE_NUMBER (destination, number); \ | |
747 | (destination) += 2; \ | |
748 | } while (0) | |
749 | ||
750 | /* Put into DESTINATION a number stored in two contiguous bytes starting | |
751 | at SOURCE. */ | |
752 | ||
753 | #define EXTRACT_NUMBER(destination, source) \ | |
754 | do { \ | |
755 | (destination) = *(source) & 0377; \ | |
756 | (destination) += SIGN_EXTEND_CHAR (*((source) + 1)) << 8; \ | |
757 | } while (0) | |
758 | ||
759 | #ifdef DEBUG | |
4bb91c68 | 760 | static void extract_number _RE_ARGS ((int *dest, re_char *source)); |
fa9a63c5 RM |
761 | static void |
762 | extract_number (dest, source) | |
763 | int *dest; | |
01618498 | 764 | re_char *source; |
fa9a63c5 | 765 | { |
5e69f11e | 766 | int temp = SIGN_EXTEND_CHAR (*(source + 1)); |
fa9a63c5 RM |
767 | *dest = *source & 0377; |
768 | *dest += temp << 8; | |
769 | } | |
770 | ||
4bb91c68 | 771 | # ifndef EXTRACT_MACROS /* To debug the macros. */ |
0b32bf0e SM |
772 | # undef EXTRACT_NUMBER |
773 | # define EXTRACT_NUMBER(dest, src) extract_number (&dest, src) | |
774 | # endif /* not EXTRACT_MACROS */ | |
fa9a63c5 RM |
775 | |
776 | #endif /* DEBUG */ | |
777 | ||
778 | /* Same as EXTRACT_NUMBER, except increment SOURCE to after the number. | |
779 | SOURCE must be an lvalue. */ | |
780 | ||
781 | #define EXTRACT_NUMBER_AND_INCR(destination, source) \ | |
782 | do { \ | |
783 | EXTRACT_NUMBER (destination, source); \ | |
25fe55af | 784 | (source) += 2; \ |
fa9a63c5 RM |
785 | } while (0) |
786 | ||
787 | #ifdef DEBUG | |
4bb91c68 SM |
788 | static void extract_number_and_incr _RE_ARGS ((int *destination, |
789 | re_char **source)); | |
fa9a63c5 RM |
790 | static void |
791 | extract_number_and_incr (destination, source) | |
792 | int *destination; | |
01618498 | 793 | re_char **source; |
5e69f11e | 794 | { |
fa9a63c5 RM |
795 | extract_number (destination, *source); |
796 | *source += 2; | |
797 | } | |
798 | ||
0b32bf0e SM |
799 | # ifndef EXTRACT_MACROS |
800 | # undef EXTRACT_NUMBER_AND_INCR | |
801 | # define EXTRACT_NUMBER_AND_INCR(dest, src) \ | |
fa9a63c5 | 802 | extract_number_and_incr (&dest, &src) |
0b32bf0e | 803 | # endif /* not EXTRACT_MACROS */ |
fa9a63c5 RM |
804 | |
805 | #endif /* DEBUG */ | |
806 | \f | |
b18215fc RS |
807 | /* Store a multibyte character in three contiguous bytes starting |
808 | DESTINATION, and increment DESTINATION to the byte after where the | |
7814e705 | 809 | character is stored. Therefore, DESTINATION must be an lvalue. */ |
b18215fc RS |
810 | |
811 | #define STORE_CHARACTER_AND_INCR(destination, character) \ | |
812 | do { \ | |
813 | (destination)[0] = (character) & 0377; \ | |
814 | (destination)[1] = ((character) >> 8) & 0377; \ | |
815 | (destination)[2] = (character) >> 16; \ | |
816 | (destination) += 3; \ | |
817 | } while (0) | |
818 | ||
819 | /* Put into DESTINATION a character stored in three contiguous bytes | |
7814e705 | 820 | starting at SOURCE. */ |
b18215fc RS |
821 | |
822 | #define EXTRACT_CHARACTER(destination, source) \ | |
823 | do { \ | |
824 | (destination) = ((source)[0] \ | |
825 | | ((source)[1] << 8) \ | |
826 | | ((source)[2] << 16)); \ | |
827 | } while (0) | |
828 | ||
829 | ||
830 | /* Macros for charset. */ | |
831 | ||
832 | /* Size of bitmap of charset P in bytes. P is a start of charset, | |
833 | i.e. *P is (re_opcode_t) charset or (re_opcode_t) charset_not. */ | |
834 | #define CHARSET_BITMAP_SIZE(p) ((p)[1] & 0x7F) | |
835 | ||
836 | /* Nonzero if charset P has range table. */ | |
25fe55af | 837 | #define CHARSET_RANGE_TABLE_EXISTS_P(p) ((p)[1] & 0x80) |
b18215fc RS |
838 | |
839 | /* Return the address of range table of charset P. But not the start | |
840 | of table itself, but the before where the number of ranges is | |
96cc36cc RS |
841 | stored. `2 +' means to skip re_opcode_t and size of bitmap, |
842 | and the 2 bytes of flags at the start of the range table. */ | |
843 | #define CHARSET_RANGE_TABLE(p) (&(p)[4 + CHARSET_BITMAP_SIZE (p)]) | |
844 | ||
845 | /* Extract the bit flags that start a range table. */ | |
846 | #define CHARSET_RANGE_TABLE_BITS(p) \ | |
847 | ((p)[2 + CHARSET_BITMAP_SIZE (p)] \ | |
848 | + (p)[3 + CHARSET_BITMAP_SIZE (p)] * 0x100) | |
b18215fc RS |
849 | |
850 | /* Test if C is listed in the bitmap of charset P. */ | |
851 | #define CHARSET_LOOKUP_BITMAP(p, c) \ | |
852 | ((c) < CHARSET_BITMAP_SIZE (p) * BYTEWIDTH \ | |
853 | && (p)[2 + (c) / BYTEWIDTH] & (1 << ((c) % BYTEWIDTH))) | |
854 | ||
855 | /* Return the address of end of RANGE_TABLE. COUNT is number of | |
7814e705 JB |
856 | ranges (which is a pair of (start, end)) in the RANGE_TABLE. `* 2' |
857 | is start of range and end of range. `* 3' is size of each start | |
b18215fc RS |
858 | and end. */ |
859 | #define CHARSET_RANGE_TABLE_END(range_table, count) \ | |
860 | ((range_table) + (count) * 2 * 3) | |
861 | ||
7814e705 | 862 | /* Test if C is in RANGE_TABLE. A flag NOT is negated if C is in. |
b18215fc RS |
863 | COUNT is number of ranges in RANGE_TABLE. */ |
864 | #define CHARSET_LOOKUP_RANGE_TABLE_RAW(not, c, range_table, count) \ | |
865 | do \ | |
866 | { \ | |
01618498 SM |
867 | re_wchar_t range_start, range_end; \ |
868 | re_char *p; \ | |
869 | re_char *range_table_end \ | |
b18215fc RS |
870 | = CHARSET_RANGE_TABLE_END ((range_table), (count)); \ |
871 | \ | |
872 | for (p = (range_table); p < range_table_end; p += 2 * 3) \ | |
873 | { \ | |
874 | EXTRACT_CHARACTER (range_start, p); \ | |
875 | EXTRACT_CHARACTER (range_end, p + 3); \ | |
876 | \ | |
877 | if (range_start <= (c) && (c) <= range_end) \ | |
878 | { \ | |
879 | (not) = !(not); \ | |
880 | break; \ | |
881 | } \ | |
882 | } \ | |
883 | } \ | |
884 | while (0) | |
885 | ||
886 | /* Test if C is in range table of CHARSET. The flag NOT is negated if | |
887 | C is listed in it. */ | |
888 | #define CHARSET_LOOKUP_RANGE_TABLE(not, c, charset) \ | |
889 | do \ | |
890 | { \ | |
891 | /* Number of ranges in range table. */ \ | |
892 | int count; \ | |
01618498 SM |
893 | re_char *range_table = CHARSET_RANGE_TABLE (charset); \ |
894 | \ | |
b18215fc RS |
895 | EXTRACT_NUMBER_AND_INCR (count, range_table); \ |
896 | CHARSET_LOOKUP_RANGE_TABLE_RAW ((not), (c), range_table, count); \ | |
897 | } \ | |
898 | while (0) | |
899 | \f | |
fa9a63c5 RM |
900 | /* If DEBUG is defined, Regex prints many voluminous messages about what |
901 | it is doing (if the variable `debug' is nonzero). If linked with the | |
902 | main program in `iregex.c', you can enter patterns and strings | |
903 | interactively. And if linked with the main program in `main.c' and | |
4bb91c68 | 904 | the other test files, you can run the already-written tests. */ |
fa9a63c5 RM |
905 | |
906 | #ifdef DEBUG | |
907 | ||
908 | /* We use standard I/O for debugging. */ | |
0b32bf0e | 909 | # include <stdio.h> |
fa9a63c5 RM |
910 | |
911 | /* It is useful to test things that ``must'' be true when debugging. */ | |
0b32bf0e | 912 | # include <assert.h> |
fa9a63c5 | 913 | |
99633e97 | 914 | static int debug = -100000; |
fa9a63c5 | 915 | |
0b32bf0e SM |
916 | # define DEBUG_STATEMENT(e) e |
917 | # define DEBUG_PRINT1(x) if (debug > 0) printf (x) | |
918 | # define DEBUG_PRINT2(x1, x2) if (debug > 0) printf (x1, x2) | |
919 | # define DEBUG_PRINT3(x1, x2, x3) if (debug > 0) printf (x1, x2, x3) | |
920 | # define DEBUG_PRINT4(x1, x2, x3, x4) if (debug > 0) printf (x1, x2, x3, x4) | |
921 | # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) \ | |
99633e97 | 922 | if (debug > 0) print_partial_compiled_pattern (s, e) |
0b32bf0e | 923 | # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) \ |
99633e97 | 924 | if (debug > 0) print_double_string (w, s1, sz1, s2, sz2) |
fa9a63c5 RM |
925 | |
926 | ||
927 | /* Print the fastmap in human-readable form. */ | |
928 | ||
929 | void | |
930 | print_fastmap (fastmap) | |
931 | char *fastmap; | |
932 | { | |
933 | unsigned was_a_range = 0; | |
5e69f11e RM |
934 | unsigned i = 0; |
935 | ||
fa9a63c5 RM |
936 | while (i < (1 << BYTEWIDTH)) |
937 | { | |
938 | if (fastmap[i++]) | |
939 | { | |
940 | was_a_range = 0; | |
25fe55af RS |
941 | putchar (i - 1); |
942 | while (i < (1 << BYTEWIDTH) && fastmap[i]) | |
943 | { | |
944 | was_a_range = 1; | |
945 | i++; | |
946 | } | |
fa9a63c5 | 947 | if (was_a_range) |
25fe55af RS |
948 | { |
949 | printf ("-"); | |
950 | putchar (i - 1); | |
951 | } | |
952 | } | |
fa9a63c5 | 953 | } |
5e69f11e | 954 | putchar ('\n'); |
fa9a63c5 RM |
955 | } |
956 | ||
957 | ||
958 | /* Print a compiled pattern string in human-readable form, starting at | |
959 | the START pointer into it and ending just before the pointer END. */ | |
960 | ||
961 | void | |
962 | print_partial_compiled_pattern (start, end) | |
01618498 SM |
963 | re_char *start; |
964 | re_char *end; | |
fa9a63c5 RM |
965 | { |
966 | int mcnt, mcnt2; | |
01618498 SM |
967 | re_char *p = start; |
968 | re_char *pend = end; | |
fa9a63c5 RM |
969 | |
970 | if (start == NULL) | |
971 | { | |
a1a052df | 972 | fprintf (stderr, "(null)\n"); |
fa9a63c5 RM |
973 | return; |
974 | } | |
5e69f11e | 975 | |
fa9a63c5 RM |
976 | /* Loop over pattern commands. */ |
977 | while (p < pend) | |
978 | { | |
a1a052df | 979 | fprintf (stderr, "%d:\t", p - start); |
fa9a63c5 RM |
980 | |
981 | switch ((re_opcode_t) *p++) | |
982 | { | |
25fe55af | 983 | case no_op: |
a1a052df | 984 | fprintf (stderr, "/no_op"); |
25fe55af | 985 | break; |
fa9a63c5 | 986 | |
99633e97 | 987 | case succeed: |
a1a052df | 988 | fprintf (stderr, "/succeed"); |
99633e97 SM |
989 | break; |
990 | ||
fa9a63c5 RM |
991 | case exactn: |
992 | mcnt = *p++; | |
a1a052df | 993 | fprintf (stderr, "/exactn/%d", mcnt); |
25fe55af | 994 | do |
fa9a63c5 | 995 | { |
a1a052df | 996 | fprintf (stderr, "/%c", *p++); |
25fe55af RS |
997 | } |
998 | while (--mcnt); | |
999 | break; | |
fa9a63c5 RM |
1000 | |
1001 | case start_memory: | |
a1a052df | 1002 | fprintf (stderr, "/start_memory/%d", *p++); |
25fe55af | 1003 | break; |
fa9a63c5 RM |
1004 | |
1005 | case stop_memory: | |
a1a052df | 1006 | fprintf (stderr, "/stop_memory/%d", *p++); |
25fe55af | 1007 | break; |
fa9a63c5 RM |
1008 | |
1009 | case duplicate: | |
a1a052df | 1010 | fprintf (stderr, "/duplicate/%d", *p++); |
fa9a63c5 RM |
1011 | break; |
1012 | ||
1013 | case anychar: | |
a1a052df | 1014 | fprintf (stderr, "/anychar"); |
fa9a63c5 RM |
1015 | break; |
1016 | ||
1017 | case charset: | |
25fe55af RS |
1018 | case charset_not: |
1019 | { | |
1020 | register int c, last = -100; | |
fa9a63c5 | 1021 | register int in_range = 0; |
99633e97 SM |
1022 | int length = CHARSET_BITMAP_SIZE (p - 1); |
1023 | int has_range_table = CHARSET_RANGE_TABLE_EXISTS_P (p - 1); | |
fa9a63c5 | 1024 | |
a1a052df | 1025 | fprintf (stderr, "/charset [%s", |
839966f3 | 1026 | (re_opcode_t) *(p - 1) == charset_not ? "^" : ""); |
5e69f11e | 1027 | |
839966f3 KH |
1028 | if (p + *p >= pend) |
1029 | fprintf (stderr, " !extends past end of pattern! "); | |
fa9a63c5 | 1030 | |
25fe55af | 1031 | for (c = 0; c < 256; c++) |
96cc36cc | 1032 | if (c / 8 < length |
fa9a63c5 RM |
1033 | && (p[1 + (c/8)] & (1 << (c % 8)))) |
1034 | { | |
1035 | /* Are we starting a range? */ | |
1036 | if (last + 1 == c && ! in_range) | |
1037 | { | |
a1a052df | 1038 | fprintf (stderr, "-"); |
fa9a63c5 RM |
1039 | in_range = 1; |
1040 | } | |
1041 | /* Have we broken a range? */ | |
1042 | else if (last + 1 != c && in_range) | |
96cc36cc | 1043 | { |
a1a052df | 1044 | fprintf (stderr, "%c", last); |
fa9a63c5 RM |
1045 | in_range = 0; |
1046 | } | |
5e69f11e | 1047 | |
fa9a63c5 | 1048 | if (! in_range) |
a1a052df | 1049 | fprintf (stderr, "%c", c); |
fa9a63c5 RM |
1050 | |
1051 | last = c; | |
25fe55af | 1052 | } |
fa9a63c5 RM |
1053 | |
1054 | if (in_range) | |
a1a052df | 1055 | fprintf (stderr, "%c", last); |
fa9a63c5 | 1056 | |
a1a052df | 1057 | fprintf (stderr, "]"); |
fa9a63c5 | 1058 | |
99633e97 | 1059 | p += 1 + length; |
96cc36cc | 1060 | |
96cc36cc | 1061 | if (has_range_table) |
99633e97 SM |
1062 | { |
1063 | int count; | |
a1a052df | 1064 | fprintf (stderr, "has-range-table"); |
99633e97 SM |
1065 | |
1066 | /* ??? Should print the range table; for now, just skip it. */ | |
1067 | p += 2; /* skip range table bits */ | |
1068 | EXTRACT_NUMBER_AND_INCR (count, p); | |
1069 | p = CHARSET_RANGE_TABLE_END (p, count); | |
1070 | } | |
fa9a63c5 RM |
1071 | } |
1072 | break; | |
1073 | ||
1074 | case begline: | |
a1a052df | 1075 | fprintf (stderr, "/begline"); |
25fe55af | 1076 | break; |
fa9a63c5 RM |
1077 | |
1078 | case endline: | |
a1a052df | 1079 | fprintf (stderr, "/endline"); |
25fe55af | 1080 | break; |
fa9a63c5 RM |
1081 | |
1082 | case on_failure_jump: | |
25fe55af | 1083 | extract_number_and_incr (&mcnt, &p); |
a1a052df | 1084 | fprintf (stderr, "/on_failure_jump to %d", p + mcnt - start); |
25fe55af | 1085 | break; |
fa9a63c5 RM |
1086 | |
1087 | case on_failure_keep_string_jump: | |
25fe55af | 1088 | extract_number_and_incr (&mcnt, &p); |
a1a052df | 1089 | fprintf (stderr, "/on_failure_keep_string_jump to %d", p + mcnt - start); |
25fe55af | 1090 | break; |
fa9a63c5 | 1091 | |
0683b6fa SM |
1092 | case on_failure_jump_nastyloop: |
1093 | extract_number_and_incr (&mcnt, &p); | |
a1a052df | 1094 | fprintf (stderr, "/on_failure_jump_nastyloop to %d", p + mcnt - start); |
0683b6fa SM |
1095 | break; |
1096 | ||
505bde11 | 1097 | case on_failure_jump_loop: |
fa9a63c5 | 1098 | extract_number_and_incr (&mcnt, &p); |
a1a052df | 1099 | fprintf (stderr, "/on_failure_jump_loop to %d", p + mcnt - start); |
5e69f11e RM |
1100 | break; |
1101 | ||
505bde11 | 1102 | case on_failure_jump_smart: |
fa9a63c5 | 1103 | extract_number_and_incr (&mcnt, &p); |
a1a052df | 1104 | fprintf (stderr, "/on_failure_jump_smart to %d", p + mcnt - start); |
5e69f11e RM |
1105 | break; |
1106 | ||
25fe55af | 1107 | case jump: |
fa9a63c5 | 1108 | extract_number_and_incr (&mcnt, &p); |
a1a052df | 1109 | fprintf (stderr, "/jump to %d", p + mcnt - start); |
fa9a63c5 RM |
1110 | break; |
1111 | ||
25fe55af RS |
1112 | case succeed_n: |
1113 | extract_number_and_incr (&mcnt, &p); | |
1114 | extract_number_and_incr (&mcnt2, &p); | |
a1a052df | 1115 | fprintf (stderr, "/succeed_n to %d, %d times", p - 2 + mcnt - start, mcnt2); |
25fe55af | 1116 | break; |
5e69f11e | 1117 | |
25fe55af RS |
1118 | case jump_n: |
1119 | extract_number_and_incr (&mcnt, &p); | |
1120 | extract_number_and_incr (&mcnt2, &p); | |
a1a052df | 1121 | fprintf (stderr, "/jump_n to %d, %d times", p - 2 + mcnt - start, mcnt2); |
25fe55af | 1122 | break; |
5e69f11e | 1123 | |
25fe55af RS |
1124 | case set_number_at: |
1125 | extract_number_and_incr (&mcnt, &p); | |
1126 | extract_number_and_incr (&mcnt2, &p); | |
a1a052df | 1127 | fprintf (stderr, "/set_number_at location %d to %d", p - 2 + mcnt - start, mcnt2); |
25fe55af | 1128 | break; |
5e69f11e | 1129 | |
25fe55af | 1130 | case wordbound: |
a1a052df | 1131 | fprintf (stderr, "/wordbound"); |
fa9a63c5 RM |
1132 | break; |
1133 | ||
1134 | case notwordbound: | |
a1a052df | 1135 | fprintf (stderr, "/notwordbound"); |
25fe55af | 1136 | break; |
fa9a63c5 RM |
1137 | |
1138 | case wordbeg: | |
a1a052df | 1139 | fprintf (stderr, "/wordbeg"); |
fa9a63c5 | 1140 | break; |
5e69f11e | 1141 | |
fa9a63c5 | 1142 | case wordend: |
a1a052df | 1143 | fprintf (stderr, "/wordend"); |
e2543b02 | 1144 | break; |
5e69f11e | 1145 | |
669fa600 | 1146 | case symbeg: |
e2543b02 | 1147 | fprintf (stderr, "/symbeg"); |
669fa600 SM |
1148 | break; |
1149 | ||
1150 | case symend: | |
e2543b02 | 1151 | fprintf (stderr, "/symend"); |
669fa600 | 1152 | break; |
5e69f11e | 1153 | |
1fb352e0 | 1154 | case syntaxspec: |
a1a052df | 1155 | fprintf (stderr, "/syntaxspec"); |
1fb352e0 | 1156 | mcnt = *p++; |
a1a052df | 1157 | fprintf (stderr, "/%d", mcnt); |
1fb352e0 SM |
1158 | break; |
1159 | ||
1160 | case notsyntaxspec: | |
a1a052df | 1161 | fprintf (stderr, "/notsyntaxspec"); |
1fb352e0 | 1162 | mcnt = *p++; |
a1a052df | 1163 | fprintf (stderr, "/%d", mcnt); |
1fb352e0 SM |
1164 | break; |
1165 | ||
0b32bf0e | 1166 | # ifdef emacs |
fa9a63c5 | 1167 | case before_dot: |
a1a052df | 1168 | fprintf (stderr, "/before_dot"); |
25fe55af | 1169 | break; |
fa9a63c5 RM |
1170 | |
1171 | case at_dot: | |
a1a052df | 1172 | fprintf (stderr, "/at_dot"); |
25fe55af | 1173 | break; |
fa9a63c5 RM |
1174 | |
1175 | case after_dot: | |
a1a052df | 1176 | fprintf (stderr, "/after_dot"); |
25fe55af | 1177 | break; |
fa9a63c5 | 1178 | |
1fb352e0 | 1179 | case categoryspec: |
a1a052df | 1180 | fprintf (stderr, "/categoryspec"); |
fa9a63c5 | 1181 | mcnt = *p++; |
a1a052df | 1182 | fprintf (stderr, "/%d", mcnt); |
25fe55af | 1183 | break; |
5e69f11e | 1184 | |
1fb352e0 | 1185 | case notcategoryspec: |
a1a052df | 1186 | fprintf (stderr, "/notcategoryspec"); |
fa9a63c5 | 1187 | mcnt = *p++; |
a1a052df | 1188 | fprintf (stderr, "/%d", mcnt); |
fa9a63c5 | 1189 | break; |
0b32bf0e | 1190 | # endif /* emacs */ |
fa9a63c5 | 1191 | |
fa9a63c5 | 1192 | case begbuf: |
a1a052df | 1193 | fprintf (stderr, "/begbuf"); |
25fe55af | 1194 | break; |
fa9a63c5 RM |
1195 | |
1196 | case endbuf: | |
a1a052df | 1197 | fprintf (stderr, "/endbuf"); |
25fe55af | 1198 | break; |
fa9a63c5 | 1199 | |
25fe55af | 1200 | default: |
a1a052df | 1201 | fprintf (stderr, "?%d", *(p-1)); |
fa9a63c5 RM |
1202 | } |
1203 | ||
a1a052df | 1204 | fprintf (stderr, "\n"); |
fa9a63c5 RM |
1205 | } |
1206 | ||
a1a052df | 1207 | fprintf (stderr, "%d:\tend of pattern.\n", p - start); |
fa9a63c5 RM |
1208 | } |
1209 | ||
1210 | ||
1211 | void | |
1212 | print_compiled_pattern (bufp) | |
1213 | struct re_pattern_buffer *bufp; | |
1214 | { | |
01618498 | 1215 | re_char *buffer = bufp->buffer; |
fa9a63c5 RM |
1216 | |
1217 | print_partial_compiled_pattern (buffer, buffer + bufp->used); | |
4bb91c68 SM |
1218 | printf ("%ld bytes used/%ld bytes allocated.\n", |
1219 | bufp->used, bufp->allocated); | |
fa9a63c5 RM |
1220 | |
1221 | if (bufp->fastmap_accurate && bufp->fastmap) | |
1222 | { | |
1223 | printf ("fastmap: "); | |
1224 | print_fastmap (bufp->fastmap); | |
1225 | } | |
1226 | ||
1227 | printf ("re_nsub: %d\t", bufp->re_nsub); | |
1228 | printf ("regs_alloc: %d\t", bufp->regs_allocated); | |
1229 | printf ("can_be_null: %d\t", bufp->can_be_null); | |
fa9a63c5 RM |
1230 | printf ("no_sub: %d\t", bufp->no_sub); |
1231 | printf ("not_bol: %d\t", bufp->not_bol); | |
1232 | printf ("not_eol: %d\t", bufp->not_eol); | |
4bb91c68 | 1233 | printf ("syntax: %lx\n", bufp->syntax); |
505bde11 | 1234 | fflush (stdout); |
fa9a63c5 RM |
1235 | /* Perhaps we should print the translate table? */ |
1236 | } | |
1237 | ||
1238 | ||
1239 | void | |
1240 | print_double_string (where, string1, size1, string2, size2) | |
66f0296e SM |
1241 | re_char *where; |
1242 | re_char *string1; | |
1243 | re_char *string2; | |
fa9a63c5 RM |
1244 | int size1; |
1245 | int size2; | |
1246 | { | |
4bb91c68 | 1247 | int this_char; |
5e69f11e | 1248 | |
fa9a63c5 RM |
1249 | if (where == NULL) |
1250 | printf ("(null)"); | |
1251 | else | |
1252 | { | |
1253 | if (FIRST_STRING_P (where)) | |
25fe55af RS |
1254 | { |
1255 | for (this_char = where - string1; this_char < size1; this_char++) | |
1256 | putchar (string1[this_char]); | |
fa9a63c5 | 1257 | |
25fe55af RS |
1258 | where = string2; |
1259 | } | |
fa9a63c5 RM |
1260 | |
1261 | for (this_char = where - string2; this_char < size2; this_char++) | |
25fe55af | 1262 | putchar (string2[this_char]); |
fa9a63c5 RM |
1263 | } |
1264 | } | |
1265 | ||
1266 | #else /* not DEBUG */ | |
1267 | ||
0b32bf0e SM |
1268 | # undef assert |
1269 | # define assert(e) | |
fa9a63c5 | 1270 | |
0b32bf0e SM |
1271 | # define DEBUG_STATEMENT(e) |
1272 | # define DEBUG_PRINT1(x) | |
1273 | # define DEBUG_PRINT2(x1, x2) | |
1274 | # define DEBUG_PRINT3(x1, x2, x3) | |
1275 | # define DEBUG_PRINT4(x1, x2, x3, x4) | |
1276 | # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) | |
1277 | # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) | |
fa9a63c5 RM |
1278 | |
1279 | #endif /* not DEBUG */ | |
1280 | \f | |
1281 | /* Set by `re_set_syntax' to the current regexp syntax to recognize. Can | |
1282 | also be assigned to arbitrarily: each pattern buffer stores its own | |
1283 | syntax, so it can be changed between regex compilations. */ | |
1284 | /* This has no initializer because initialized variables in Emacs | |
1285 | become read-only after dumping. */ | |
1286 | reg_syntax_t re_syntax_options; | |
1287 | ||
1288 | ||
1289 | /* Specify the precise syntax of regexps for compilation. This provides | |
1290 | for compatibility for various utilities which historically have | |
1291 | different, incompatible syntaxes. | |
1292 | ||
1293 | The argument SYNTAX is a bit mask comprised of the various bits | |
4bb91c68 | 1294 | defined in regex.h. We return the old syntax. */ |
fa9a63c5 RM |
1295 | |
1296 | reg_syntax_t | |
971de7fb | 1297 | re_set_syntax (reg_syntax_t syntax) |
fa9a63c5 RM |
1298 | { |
1299 | reg_syntax_t ret = re_syntax_options; | |
5e69f11e | 1300 | |
fa9a63c5 RM |
1301 | re_syntax_options = syntax; |
1302 | return ret; | |
1303 | } | |
c0f9ea08 | 1304 | WEAK_ALIAS (__re_set_syntax, re_set_syntax) |
f9b0fd99 RS |
1305 | |
1306 | /* Regexp to use to replace spaces, or NULL meaning don't. */ | |
1307 | static re_char *whitespace_regexp; | |
1308 | ||
1309 | void | |
971de7fb | 1310 | re_set_whitespace_regexp (const char *regexp) |
f9b0fd99 | 1311 | { |
6470ea05 | 1312 | whitespace_regexp = (re_char *) regexp; |
f9b0fd99 RS |
1313 | } |
1314 | WEAK_ALIAS (__re_set_syntax, re_set_syntax) | |
fa9a63c5 RM |
1315 | \f |
1316 | /* This table gives an error message for each of the error codes listed | |
4bb91c68 | 1317 | in regex.h. Obviously the order here has to be same as there. |
fa9a63c5 | 1318 | POSIX doesn't require that we do anything for REG_NOERROR, |
4bb91c68 | 1319 | but why not be nice? */ |
fa9a63c5 RM |
1320 | |
1321 | static const char *re_error_msgid[] = | |
5e69f11e RM |
1322 | { |
1323 | gettext_noop ("Success"), /* REG_NOERROR */ | |
1324 | gettext_noop ("No match"), /* REG_NOMATCH */ | |
1325 | gettext_noop ("Invalid regular expression"), /* REG_BADPAT */ | |
1326 | gettext_noop ("Invalid collation character"), /* REG_ECOLLATE */ | |
1327 | gettext_noop ("Invalid character class name"), /* REG_ECTYPE */ | |
1328 | gettext_noop ("Trailing backslash"), /* REG_EESCAPE */ | |
1329 | gettext_noop ("Invalid back reference"), /* REG_ESUBREG */ | |
1330 | gettext_noop ("Unmatched [ or [^"), /* REG_EBRACK */ | |
1331 | gettext_noop ("Unmatched ( or \\("), /* REG_EPAREN */ | |
1332 | gettext_noop ("Unmatched \\{"), /* REG_EBRACE */ | |
1333 | gettext_noop ("Invalid content of \\{\\}"), /* REG_BADBR */ | |
1334 | gettext_noop ("Invalid range end"), /* REG_ERANGE */ | |
1335 | gettext_noop ("Memory exhausted"), /* REG_ESPACE */ | |
1336 | gettext_noop ("Invalid preceding regular expression"), /* REG_BADRPT */ | |
1337 | gettext_noop ("Premature end of regular expression"), /* REG_EEND */ | |
1338 | gettext_noop ("Regular expression too big"), /* REG_ESIZE */ | |
1339 | gettext_noop ("Unmatched ) or \\)"), /* REG_ERPAREN */ | |
b3e4c897 | 1340 | gettext_noop ("Range striding over charsets") /* REG_ERANGEX */ |
fa9a63c5 RM |
1341 | }; |
1342 | \f | |
4bb91c68 | 1343 | /* Avoiding alloca during matching, to placate r_alloc. */ |
fa9a63c5 RM |
1344 | |
1345 | /* Define MATCH_MAY_ALLOCATE unless we need to make sure that the | |
1346 | searching and matching functions should not call alloca. On some | |
1347 | systems, alloca is implemented in terms of malloc, and if we're | |
1348 | using the relocating allocator routines, then malloc could cause a | |
1349 | relocation, which might (if the strings being searched are in the | |
1350 | ralloc heap) shift the data out from underneath the regexp | |
1351 | routines. | |
1352 | ||
5e69f11e | 1353 | Here's another reason to avoid allocation: Emacs |
fa9a63c5 RM |
1354 | processes input from X in a signal handler; processing X input may |
1355 | call malloc; if input arrives while a matching routine is calling | |
1356 | malloc, then we're scrod. But Emacs can't just block input while | |
1357 | calling matching routines; then we don't notice interrupts when | |
1358 | they come in. So, Emacs blocks input around all regexp calls | |
1359 | except the matching calls, which it leaves unprotected, in the | |
1360 | faith that they will not malloc. */ | |
1361 | ||
1362 | /* Normally, this is fine. */ | |
1363 | #define MATCH_MAY_ALLOCATE | |
1364 | ||
fa9a63c5 RM |
1365 | /* The match routines may not allocate if (1) they would do it with malloc |
1366 | and (2) it's not safe for them to use malloc. | |
1367 | Note that if REL_ALLOC is defined, matching would not use malloc for the | |
1368 | failure stack, but we would still use it for the register vectors; | |
4bb91c68 | 1369 | so REL_ALLOC should not affect this. */ |
b588157e | 1370 | #if defined REGEX_MALLOC && defined emacs |
0b32bf0e | 1371 | # undef MATCH_MAY_ALLOCATE |
fa9a63c5 RM |
1372 | #endif |
1373 | ||
1374 | \f | |
1375 | /* Failure stack declarations and macros; both re_compile_fastmap and | |
1376 | re_match_2 use a failure stack. These have to be macros because of | |
1377 | REGEX_ALLOCATE_STACK. */ | |
5e69f11e | 1378 | |
fa9a63c5 | 1379 | |
320a2a73 | 1380 | /* Approximate number of failure points for which to initially allocate space |
fa9a63c5 RM |
1381 | when matching. If this number is exceeded, we allocate more |
1382 | space, so it is not a hard limit. */ | |
1383 | #ifndef INIT_FAILURE_ALLOC | |
0b32bf0e | 1384 | # define INIT_FAILURE_ALLOC 20 |
fa9a63c5 RM |
1385 | #endif |
1386 | ||
1387 | /* Roughly the maximum number of failure points on the stack. Would be | |
320a2a73 | 1388 | exactly that if always used TYPICAL_FAILURE_SIZE items each time we failed. |
fa9a63c5 | 1389 | This is a variable only so users of regex can assign to it; we never |
ada30c0e SM |
1390 | change it ourselves. We always multiply it by TYPICAL_FAILURE_SIZE |
1391 | before using it, so it should probably be a byte-count instead. */ | |
c0f9ea08 SM |
1392 | # if defined MATCH_MAY_ALLOCATE |
1393 | /* Note that 4400 was enough to cause a crash on Alpha OSF/1, | |
320a2a73 KH |
1394 | whose default stack limit is 2mb. In order for a larger |
1395 | value to work reliably, you have to try to make it accord | |
1396 | with the process stack limit. */ | |
c0f9ea08 SM |
1397 | size_t re_max_failures = 40000; |
1398 | # else | |
1399 | size_t re_max_failures = 4000; | |
1400 | # endif | |
fa9a63c5 RM |
1401 | |
1402 | union fail_stack_elt | |
1403 | { | |
01618498 | 1404 | re_char *pointer; |
c0f9ea08 SM |
1405 | /* This should be the biggest `int' that's no bigger than a pointer. */ |
1406 | long integer; | |
fa9a63c5 RM |
1407 | }; |
1408 | ||
1409 | typedef union fail_stack_elt fail_stack_elt_t; | |
1410 | ||
1411 | typedef struct | |
1412 | { | |
1413 | fail_stack_elt_t *stack; | |
c0f9ea08 SM |
1414 | size_t size; |
1415 | size_t avail; /* Offset of next open position. */ | |
1416 | size_t frame; /* Offset of the cur constructed frame. */ | |
fa9a63c5 RM |
1417 | } fail_stack_type; |
1418 | ||
505bde11 | 1419 | #define FAIL_STACK_EMPTY() (fail_stack.frame == 0) |
fa9a63c5 RM |
1420 | #define FAIL_STACK_FULL() (fail_stack.avail == fail_stack.size) |
1421 | ||
1422 | ||
1423 | /* Define macros to initialize and free the failure stack. | |
1424 | Do `return -2' if the alloc fails. */ | |
1425 | ||
1426 | #ifdef MATCH_MAY_ALLOCATE | |
0b32bf0e | 1427 | # define INIT_FAIL_STACK() \ |
fa9a63c5 RM |
1428 | do { \ |
1429 | fail_stack.stack = (fail_stack_elt_t *) \ | |
320a2a73 KH |
1430 | REGEX_ALLOCATE_STACK (INIT_FAILURE_ALLOC * TYPICAL_FAILURE_SIZE \ |
1431 | * sizeof (fail_stack_elt_t)); \ | |
fa9a63c5 RM |
1432 | \ |
1433 | if (fail_stack.stack == NULL) \ | |
1434 | return -2; \ | |
1435 | \ | |
1436 | fail_stack.size = INIT_FAILURE_ALLOC; \ | |
1437 | fail_stack.avail = 0; \ | |
505bde11 | 1438 | fail_stack.frame = 0; \ |
fa9a63c5 RM |
1439 | } while (0) |
1440 | ||
0b32bf0e | 1441 | # define RESET_FAIL_STACK() REGEX_FREE_STACK (fail_stack.stack) |
fa9a63c5 | 1442 | #else |
0b32bf0e | 1443 | # define INIT_FAIL_STACK() \ |
fa9a63c5 RM |
1444 | do { \ |
1445 | fail_stack.avail = 0; \ | |
505bde11 | 1446 | fail_stack.frame = 0; \ |
fa9a63c5 RM |
1447 | } while (0) |
1448 | ||
0b32bf0e | 1449 | # define RESET_FAIL_STACK() ((void)0) |
fa9a63c5 RM |
1450 | #endif |
1451 | ||
1452 | ||
320a2a73 KH |
1453 | /* Double the size of FAIL_STACK, up to a limit |
1454 | which allows approximately `re_max_failures' items. | |
fa9a63c5 RM |
1455 | |
1456 | Return 1 if succeeds, and 0 if either ran out of memory | |
5e69f11e RM |
1457 | allocating space for it or it was already too large. |
1458 | ||
4bb91c68 | 1459 | REGEX_REALLOCATE_STACK requires `destination' be declared. */ |
fa9a63c5 | 1460 | |
320a2a73 KH |
1461 | /* Factor to increase the failure stack size by |
1462 | when we increase it. | |
1463 | This used to be 2, but 2 was too wasteful | |
1464 | because the old discarded stacks added up to as much space | |
1465 | were as ultimate, maximum-size stack. */ | |
1466 | #define FAIL_STACK_GROWTH_FACTOR 4 | |
1467 | ||
1468 | #define GROW_FAIL_STACK(fail_stack) \ | |
eead07d6 KH |
1469 | (((fail_stack).size * sizeof (fail_stack_elt_t) \ |
1470 | >= re_max_failures * TYPICAL_FAILURE_SIZE) \ | |
fa9a63c5 | 1471 | ? 0 \ |
320a2a73 KH |
1472 | : ((fail_stack).stack \ |
1473 | = (fail_stack_elt_t *) \ | |
25fe55af RS |
1474 | REGEX_REALLOCATE_STACK ((fail_stack).stack, \ |
1475 | (fail_stack).size * sizeof (fail_stack_elt_t), \ | |
320a2a73 KH |
1476 | MIN (re_max_failures * TYPICAL_FAILURE_SIZE, \ |
1477 | ((fail_stack).size * sizeof (fail_stack_elt_t) \ | |
1478 | * FAIL_STACK_GROWTH_FACTOR))), \ | |
fa9a63c5 RM |
1479 | \ |
1480 | (fail_stack).stack == NULL \ | |
1481 | ? 0 \ | |
6453db45 KH |
1482 | : ((fail_stack).size \ |
1483 | = (MIN (re_max_failures * TYPICAL_FAILURE_SIZE, \ | |
1484 | ((fail_stack).size * sizeof (fail_stack_elt_t) \ | |
1485 | * FAIL_STACK_GROWTH_FACTOR)) \ | |
1486 | / sizeof (fail_stack_elt_t)), \ | |
25fe55af | 1487 | 1))) |
fa9a63c5 RM |
1488 | |
1489 | ||
fa9a63c5 RM |
1490 | /* Push a pointer value onto the failure stack. |
1491 | Assumes the variable `fail_stack'. Probably should only | |
4bb91c68 | 1492 | be called from within `PUSH_FAILURE_POINT'. */ |
fa9a63c5 | 1493 | #define PUSH_FAILURE_POINTER(item) \ |
01618498 | 1494 | fail_stack.stack[fail_stack.avail++].pointer = (item) |
fa9a63c5 RM |
1495 | |
1496 | /* This pushes an integer-valued item onto the failure stack. | |
1497 | Assumes the variable `fail_stack'. Probably should only | |
4bb91c68 | 1498 | be called from within `PUSH_FAILURE_POINT'. */ |
fa9a63c5 RM |
1499 | #define PUSH_FAILURE_INT(item) \ |
1500 | fail_stack.stack[fail_stack.avail++].integer = (item) | |
1501 | ||
1502 | /* Push a fail_stack_elt_t value onto the failure stack. | |
1503 | Assumes the variable `fail_stack'. Probably should only | |
4bb91c68 | 1504 | be called from within `PUSH_FAILURE_POINT'. */ |
fa9a63c5 RM |
1505 | #define PUSH_FAILURE_ELT(item) \ |
1506 | fail_stack.stack[fail_stack.avail++] = (item) | |
1507 | ||
1508 | /* These three POP... operations complement the three PUSH... operations. | |
1509 | All assume that `fail_stack' is nonempty. */ | |
1510 | #define POP_FAILURE_POINTER() fail_stack.stack[--fail_stack.avail].pointer | |
1511 | #define POP_FAILURE_INT() fail_stack.stack[--fail_stack.avail].integer | |
1512 | #define POP_FAILURE_ELT() fail_stack.stack[--fail_stack.avail] | |
1513 | ||
505bde11 SM |
1514 | /* Individual items aside from the registers. */ |
1515 | #define NUM_NONREG_ITEMS 3 | |
1516 | ||
1517 | /* Used to examine the stack (to detect infinite loops). */ | |
1518 | #define FAILURE_PAT(h) fail_stack.stack[(h) - 1].pointer | |
66f0296e | 1519 | #define FAILURE_STR(h) (fail_stack.stack[(h) - 2].pointer) |
505bde11 SM |
1520 | #define NEXT_FAILURE_HANDLE(h) fail_stack.stack[(h) - 3].integer |
1521 | #define TOP_FAILURE_HANDLE() fail_stack.frame | |
fa9a63c5 RM |
1522 | |
1523 | ||
505bde11 SM |
1524 | #define ENSURE_FAIL_STACK(space) \ |
1525 | while (REMAINING_AVAIL_SLOTS <= space) { \ | |
1526 | if (!GROW_FAIL_STACK (fail_stack)) \ | |
1527 | return -2; \ | |
1528 | DEBUG_PRINT2 ("\n Doubled stack; size now: %d\n", (fail_stack).size);\ | |
1529 | DEBUG_PRINT2 (" slots available: %d\n", REMAINING_AVAIL_SLOTS);\ | |
1530 | } | |
1531 | ||
1532 | /* Push register NUM onto the stack. */ | |
1533 | #define PUSH_FAILURE_REG(num) \ | |
1534 | do { \ | |
1535 | char *destination; \ | |
1536 | ENSURE_FAIL_STACK(3); \ | |
1537 | DEBUG_PRINT4 (" Push reg %d (spanning %p -> %p)\n", \ | |
1538 | num, regstart[num], regend[num]); \ | |
1539 | PUSH_FAILURE_POINTER (regstart[num]); \ | |
1540 | PUSH_FAILURE_POINTER (regend[num]); \ | |
1541 | PUSH_FAILURE_INT (num); \ | |
1542 | } while (0) | |
1543 | ||
01618498 SM |
1544 | /* Change the counter's value to VAL, but make sure that it will |
1545 | be reset when backtracking. */ | |
1546 | #define PUSH_NUMBER(ptr,val) \ | |
dc1e502d SM |
1547 | do { \ |
1548 | char *destination; \ | |
1549 | int c; \ | |
1550 | ENSURE_FAIL_STACK(3); \ | |
1551 | EXTRACT_NUMBER (c, ptr); \ | |
01618498 | 1552 | DEBUG_PRINT4 (" Push number %p = %d -> %d\n", ptr, c, val); \ |
dc1e502d SM |
1553 | PUSH_FAILURE_INT (c); \ |
1554 | PUSH_FAILURE_POINTER (ptr); \ | |
1555 | PUSH_FAILURE_INT (-1); \ | |
01618498 | 1556 | STORE_NUMBER (ptr, val); \ |
dc1e502d SM |
1557 | } while (0) |
1558 | ||
505bde11 | 1559 | /* Pop a saved register off the stack. */ |
dc1e502d | 1560 | #define POP_FAILURE_REG_OR_COUNT() \ |
505bde11 SM |
1561 | do { \ |
1562 | int reg = POP_FAILURE_INT (); \ | |
dc1e502d SM |
1563 | if (reg == -1) \ |
1564 | { \ | |
1565 | /* It's a counter. */ \ | |
6dcf2d0e SM |
1566 | /* Here, we discard `const', making re_match non-reentrant. */ \ |
1567 | unsigned char *ptr = (unsigned char*) POP_FAILURE_POINTER (); \ | |
dc1e502d SM |
1568 | reg = POP_FAILURE_INT (); \ |
1569 | STORE_NUMBER (ptr, reg); \ | |
1570 | DEBUG_PRINT3 (" Pop counter %p = %d\n", ptr, reg); \ | |
1571 | } \ | |
1572 | else \ | |
1573 | { \ | |
1574 | regend[reg] = POP_FAILURE_POINTER (); \ | |
1575 | regstart[reg] = POP_FAILURE_POINTER (); \ | |
1576 | DEBUG_PRINT4 (" Pop reg %d (spanning %p -> %p)\n", \ | |
1577 | reg, regstart[reg], regend[reg]); \ | |
1578 | } \ | |
505bde11 SM |
1579 | } while (0) |
1580 | ||
1581 | /* Check that we are not stuck in an infinite loop. */ | |
1582 | #define CHECK_INFINITE_LOOP(pat_cur, string_place) \ | |
1583 | do { \ | |
f6df485f | 1584 | int failure = TOP_FAILURE_HANDLE (); \ |
505bde11 | 1585 | /* Check for infinite matching loops */ \ |
f6df485f RS |
1586 | while (failure > 0 \ |
1587 | && (FAILURE_STR (failure) == string_place \ | |
1588 | || FAILURE_STR (failure) == NULL)) \ | |
505bde11 SM |
1589 | { \ |
1590 | assert (FAILURE_PAT (failure) >= bufp->buffer \ | |
66f0296e | 1591 | && FAILURE_PAT (failure) <= bufp->buffer + bufp->used); \ |
505bde11 | 1592 | if (FAILURE_PAT (failure) == pat_cur) \ |
f6df485f | 1593 | { \ |
6df42991 SM |
1594 | cycle = 1; \ |
1595 | break; \ | |
f6df485f | 1596 | } \ |
66f0296e | 1597 | DEBUG_PRINT2 (" Other pattern: %p\n", FAILURE_PAT (failure)); \ |
505bde11 SM |
1598 | failure = NEXT_FAILURE_HANDLE(failure); \ |
1599 | } \ | |
1600 | DEBUG_PRINT2 (" Other string: %p\n", FAILURE_STR (failure)); \ | |
1601 | } while (0) | |
6df42991 | 1602 | |
fa9a63c5 | 1603 | /* Push the information about the state we will need |
5e69f11e RM |
1604 | if we ever fail back to it. |
1605 | ||
505bde11 | 1606 | Requires variables fail_stack, regstart, regend and |
320a2a73 | 1607 | num_regs be declared. GROW_FAIL_STACK requires `destination' be |
fa9a63c5 | 1608 | declared. |
5e69f11e | 1609 | |
fa9a63c5 RM |
1610 | Does `return FAILURE_CODE' if runs out of memory. */ |
1611 | ||
505bde11 SM |
1612 | #define PUSH_FAILURE_POINT(pattern, string_place) \ |
1613 | do { \ | |
1614 | char *destination; \ | |
1615 | /* Must be int, so when we don't save any registers, the arithmetic \ | |
1616 | of 0 + -1 isn't done as unsigned. */ \ | |
1617 | \ | |
505bde11 | 1618 | DEBUG_STATEMENT (nfailure_points_pushed++); \ |
4bb91c68 | 1619 | DEBUG_PRINT1 ("\nPUSH_FAILURE_POINT:\n"); \ |
505bde11 SM |
1620 | DEBUG_PRINT2 (" Before push, next avail: %d\n", (fail_stack).avail); \ |
1621 | DEBUG_PRINT2 (" size: %d\n", (fail_stack).size);\ | |
1622 | \ | |
1623 | ENSURE_FAIL_STACK (NUM_NONREG_ITEMS); \ | |
1624 | \ | |
1625 | DEBUG_PRINT1 ("\n"); \ | |
1626 | \ | |
1627 | DEBUG_PRINT2 (" Push frame index: %d\n", fail_stack.frame); \ | |
1628 | PUSH_FAILURE_INT (fail_stack.frame); \ | |
1629 | \ | |
1630 | DEBUG_PRINT2 (" Push string %p: `", string_place); \ | |
1631 | DEBUG_PRINT_DOUBLE_STRING (string_place, string1, size1, string2, size2);\ | |
1632 | DEBUG_PRINT1 ("'\n"); \ | |
1633 | PUSH_FAILURE_POINTER (string_place); \ | |
1634 | \ | |
1635 | DEBUG_PRINT2 (" Push pattern %p: ", pattern); \ | |
1636 | DEBUG_PRINT_COMPILED_PATTERN (bufp, pattern, pend); \ | |
1637 | PUSH_FAILURE_POINTER (pattern); \ | |
1638 | \ | |
1639 | /* Close the frame by moving the frame pointer past it. */ \ | |
1640 | fail_stack.frame = fail_stack.avail; \ | |
1641 | } while (0) | |
fa9a63c5 | 1642 | |
320a2a73 KH |
1643 | /* Estimate the size of data pushed by a typical failure stack entry. |
1644 | An estimate is all we need, because all we use this for | |
1645 | is to choose a limit for how big to make the failure stack. */ | |
ada30c0e | 1646 | /* BEWARE, the value `20' is hard-coded in emacs.c:main(). */ |
320a2a73 | 1647 | #define TYPICAL_FAILURE_SIZE 20 |
fa9a63c5 | 1648 | |
fa9a63c5 RM |
1649 | /* How many items can still be added to the stack without overflowing it. */ |
1650 | #define REMAINING_AVAIL_SLOTS ((fail_stack).size - (fail_stack).avail) | |
1651 | ||
1652 | ||
1653 | /* Pops what PUSH_FAIL_STACK pushes. | |
1654 | ||
1655 | We restore into the parameters, all of which should be lvalues: | |
1656 | STR -- the saved data position. | |
1657 | PAT -- the saved pattern position. | |
fa9a63c5 | 1658 | REGSTART, REGEND -- arrays of string positions. |
5e69f11e | 1659 | |
fa9a63c5 | 1660 | Also assumes the variables `fail_stack' and (if debugging), `bufp', |
7814e705 | 1661 | `pend', `string1', `size1', `string2', and `size2'. */ |
fa9a63c5 | 1662 | |
505bde11 SM |
1663 | #define POP_FAILURE_POINT(str, pat) \ |
1664 | do { \ | |
fa9a63c5 RM |
1665 | assert (!FAIL_STACK_EMPTY ()); \ |
1666 | \ | |
1667 | /* Remove failure points and point to how many regs pushed. */ \ | |
1668 | DEBUG_PRINT1 ("POP_FAILURE_POINT:\n"); \ | |
1669 | DEBUG_PRINT2 (" Before pop, next avail: %d\n", fail_stack.avail); \ | |
25fe55af | 1670 | DEBUG_PRINT2 (" size: %d\n", fail_stack.size); \ |
fa9a63c5 | 1671 | \ |
505bde11 SM |
1672 | /* Pop the saved registers. */ \ |
1673 | while (fail_stack.frame < fail_stack.avail) \ | |
dc1e502d | 1674 | POP_FAILURE_REG_OR_COUNT (); \ |
fa9a63c5 | 1675 | \ |
01618498 | 1676 | pat = POP_FAILURE_POINTER (); \ |
505bde11 SM |
1677 | DEBUG_PRINT2 (" Popping pattern %p: ", pat); \ |
1678 | DEBUG_PRINT_COMPILED_PATTERN (bufp, pat, pend); \ | |
fa9a63c5 RM |
1679 | \ |
1680 | /* If the saved string location is NULL, it came from an \ | |
1681 | on_failure_keep_string_jump opcode, and we want to throw away the \ | |
1682 | saved NULL, thus retaining our current position in the string. */ \ | |
01618498 | 1683 | str = POP_FAILURE_POINTER (); \ |
505bde11 | 1684 | DEBUG_PRINT2 (" Popping string %p: `", str); \ |
fa9a63c5 RM |
1685 | DEBUG_PRINT_DOUBLE_STRING (str, string1, size1, string2, size2); \ |
1686 | DEBUG_PRINT1 ("'\n"); \ | |
1687 | \ | |
505bde11 SM |
1688 | fail_stack.frame = POP_FAILURE_INT (); \ |
1689 | DEBUG_PRINT2 (" Popping frame index: %d\n", fail_stack.frame); \ | |
fa9a63c5 | 1690 | \ |
505bde11 SM |
1691 | assert (fail_stack.avail >= 0); \ |
1692 | assert (fail_stack.frame <= fail_stack.avail); \ | |
fa9a63c5 | 1693 | \ |
fa9a63c5 | 1694 | DEBUG_STATEMENT (nfailure_points_popped++); \ |
505bde11 | 1695 | } while (0) /* POP_FAILURE_POINT */ |
fa9a63c5 RM |
1696 | |
1697 | ||
1698 | \f | |
fa9a63c5 | 1699 | /* Registers are set to a sentinel when they haven't yet matched. */ |
4bb91c68 | 1700 | #define REG_UNSET(e) ((e) == NULL) |
fa9a63c5 RM |
1701 | \f |
1702 | /* Subroutine declarations and macros for regex_compile. */ | |
1703 | ||
4bb91c68 SM |
1704 | static reg_errcode_t regex_compile _RE_ARGS ((re_char *pattern, size_t size, |
1705 | reg_syntax_t syntax, | |
1706 | struct re_pattern_buffer *bufp)); | |
1707 | static void store_op1 _RE_ARGS ((re_opcode_t op, unsigned char *loc, int arg)); | |
1708 | static void store_op2 _RE_ARGS ((re_opcode_t op, unsigned char *loc, | |
1709 | int arg1, int arg2)); | |
1710 | static void insert_op1 _RE_ARGS ((re_opcode_t op, unsigned char *loc, | |
1711 | int arg, unsigned char *end)); | |
1712 | static void insert_op2 _RE_ARGS ((re_opcode_t op, unsigned char *loc, | |
1713 | int arg1, int arg2, unsigned char *end)); | |
01618498 SM |
1714 | static boolean at_begline_loc_p _RE_ARGS ((re_char *pattern, |
1715 | re_char *p, | |
4bb91c68 | 1716 | reg_syntax_t syntax)); |
01618498 SM |
1717 | static boolean at_endline_loc_p _RE_ARGS ((re_char *p, |
1718 | re_char *pend, | |
4bb91c68 | 1719 | reg_syntax_t syntax)); |
01618498 SM |
1720 | static re_char *skip_one_char _RE_ARGS ((re_char *p)); |
1721 | static int analyse_first _RE_ARGS ((re_char *p, re_char *pend, | |
4bb91c68 | 1722 | char *fastmap, const int multibyte)); |
fa9a63c5 | 1723 | |
fa9a63c5 | 1724 | /* Fetch the next character in the uncompiled pattern, with no |
4bb91c68 | 1725 | translation. */ |
36595814 | 1726 | #define PATFETCH(c) \ |
2d1675e4 SM |
1727 | do { \ |
1728 | int len; \ | |
1729 | if (p == pend) return REG_EEND; \ | |
62a6e103 | 1730 | c = RE_STRING_CHAR_AND_LENGTH (p, len, multibyte); \ |
2d1675e4 | 1731 | p += len; \ |
fa9a63c5 RM |
1732 | } while (0) |
1733 | ||
fa9a63c5 RM |
1734 | |
1735 | /* If `translate' is non-null, return translate[D], else just D. We | |
1736 | cast the subscript to translate because some data is declared as | |
1737 | `char *', to avoid warnings when a string constant is passed. But | |
1738 | when we use a character as a subscript we must make it unsigned. */ | |
6676cb1c | 1739 | #ifndef TRANSLATE |
0b32bf0e | 1740 | # define TRANSLATE(d) \ |
66f0296e | 1741 | (RE_TRANSLATE_P (translate) ? RE_TRANSLATE (translate, (d)) : (d)) |
6676cb1c | 1742 | #endif |
fa9a63c5 RM |
1743 | |
1744 | ||
1745 | /* Macros for outputting the compiled pattern into `buffer'. */ | |
1746 | ||
1747 | /* If the buffer isn't allocated when it comes in, use this. */ | |
1748 | #define INIT_BUF_SIZE 32 | |
1749 | ||
4bb91c68 | 1750 | /* Make sure we have at least N more bytes of space in buffer. */ |
fa9a63c5 | 1751 | #define GET_BUFFER_SPACE(n) \ |
01618498 | 1752 | while ((size_t) (b - bufp->buffer + (n)) > bufp->allocated) \ |
fa9a63c5 RM |
1753 | EXTEND_BUFFER () |
1754 | ||
1755 | /* Make sure we have one more byte of buffer space and then add C to it. */ | |
1756 | #define BUF_PUSH(c) \ | |
1757 | do { \ | |
1758 | GET_BUFFER_SPACE (1); \ | |
1759 | *b++ = (unsigned char) (c); \ | |
1760 | } while (0) | |
1761 | ||
1762 | ||
1763 | /* Ensure we have two more bytes of buffer space and then append C1 and C2. */ | |
1764 | #define BUF_PUSH_2(c1, c2) \ | |
1765 | do { \ | |
1766 | GET_BUFFER_SPACE (2); \ | |
1767 | *b++ = (unsigned char) (c1); \ | |
1768 | *b++ = (unsigned char) (c2); \ | |
1769 | } while (0) | |
1770 | ||
1771 | ||
4bb91c68 | 1772 | /* As with BUF_PUSH_2, except for three bytes. */ |
fa9a63c5 RM |
1773 | #define BUF_PUSH_3(c1, c2, c3) \ |
1774 | do { \ | |
1775 | GET_BUFFER_SPACE (3); \ | |
1776 | *b++ = (unsigned char) (c1); \ | |
1777 | *b++ = (unsigned char) (c2); \ | |
1778 | *b++ = (unsigned char) (c3); \ | |
1779 | } while (0) | |
1780 | ||
1781 | ||
1782 | /* Store a jump with opcode OP at LOC to location TO. We store a | |
4bb91c68 | 1783 | relative address offset by the three bytes the jump itself occupies. */ |
fa9a63c5 RM |
1784 | #define STORE_JUMP(op, loc, to) \ |
1785 | store_op1 (op, loc, (to) - (loc) - 3) | |
1786 | ||
1787 | /* Likewise, for a two-argument jump. */ | |
1788 | #define STORE_JUMP2(op, loc, to, arg) \ | |
1789 | store_op2 (op, loc, (to) - (loc) - 3, arg) | |
1790 | ||
4bb91c68 | 1791 | /* Like `STORE_JUMP', but for inserting. Assume `b' is the buffer end. */ |
fa9a63c5 RM |
1792 | #define INSERT_JUMP(op, loc, to) \ |
1793 | insert_op1 (op, loc, (to) - (loc) - 3, b) | |
1794 | ||
1795 | /* Like `STORE_JUMP2', but for inserting. Assume `b' is the buffer end. */ | |
1796 | #define INSERT_JUMP2(op, loc, to, arg) \ | |
1797 | insert_op2 (op, loc, (to) - (loc) - 3, arg, b) | |
1798 | ||
1799 | ||
1800 | /* This is not an arbitrary limit: the arguments which represent offsets | |
839966f3 | 1801 | into the pattern are two bytes long. So if 2^15 bytes turns out to |
fa9a63c5 | 1802 | be too small, many things would have to change. */ |
839966f3 KH |
1803 | # define MAX_BUF_SIZE (1L << 15) |
1804 | ||
1805 | #if 0 /* This is when we thought it could be 2^16 bytes. */ | |
4bb91c68 SM |
1806 | /* Any other compiler which, like MSC, has allocation limit below 2^16 |
1807 | bytes will have to use approach similar to what was done below for | |
1808 | MSC and drop MAX_BUF_SIZE a bit. Otherwise you may end up | |
1809 | reallocating to 0 bytes. Such thing is not going to work too well. | |
1810 | You have been warned!! */ | |
1811 | #if defined _MSC_VER && !defined WIN32 | |
1812 | /* Microsoft C 16-bit versions limit malloc to approx 65512 bytes. */ | |
1813 | # define MAX_BUF_SIZE 65500L | |
1814 | #else | |
1815 | # define MAX_BUF_SIZE (1L << 16) | |
1816 | #endif | |
839966f3 | 1817 | #endif /* 0 */ |
fa9a63c5 RM |
1818 | |
1819 | /* Extend the buffer by twice its current size via realloc and | |
1820 | reset the pointers that pointed into the old block to point to the | |
1821 | correct places in the new one. If extending the buffer results in it | |
4bb91c68 SM |
1822 | being larger than MAX_BUF_SIZE, then flag memory exhausted. */ |
1823 | #if __BOUNDED_POINTERS__ | |
1824 | # define SET_HIGH_BOUND(P) (__ptrhigh (P) = __ptrlow (P) + bufp->allocated) | |
381880b0 CY |
1825 | # define MOVE_BUFFER_POINTER(P) \ |
1826 | (__ptrlow (P) = new_buffer + (__ptrlow (P) - old_buffer), \ | |
1827 | SET_HIGH_BOUND (P), \ | |
1828 | __ptrvalue (P) = new_buffer + (__ptrvalue (P) - old_buffer)) | |
4bb91c68 SM |
1829 | # define ELSE_EXTEND_BUFFER_HIGH_BOUND \ |
1830 | else \ | |
1831 | { \ | |
1832 | SET_HIGH_BOUND (b); \ | |
1833 | SET_HIGH_BOUND (begalt); \ | |
1834 | if (fixup_alt_jump) \ | |
1835 | SET_HIGH_BOUND (fixup_alt_jump); \ | |
1836 | if (laststart) \ | |
1837 | SET_HIGH_BOUND (laststart); \ | |
1838 | if (pending_exact) \ | |
1839 | SET_HIGH_BOUND (pending_exact); \ | |
1840 | } | |
1841 | #else | |
381880b0 | 1842 | # define MOVE_BUFFER_POINTER(P) ((P) = new_buffer + ((P) - old_buffer)) |
4bb91c68 SM |
1843 | # define ELSE_EXTEND_BUFFER_HIGH_BOUND |
1844 | #endif | |
fa9a63c5 | 1845 | #define EXTEND_BUFFER() \ |
25fe55af | 1846 | do { \ |
381880b0 | 1847 | unsigned char *old_buffer = bufp->buffer; \ |
25fe55af | 1848 | if (bufp->allocated == MAX_BUF_SIZE) \ |
fa9a63c5 RM |
1849 | return REG_ESIZE; \ |
1850 | bufp->allocated <<= 1; \ | |
1851 | if (bufp->allocated > MAX_BUF_SIZE) \ | |
25fe55af | 1852 | bufp->allocated = MAX_BUF_SIZE; \ |
01618498 | 1853 | RETALLOC (bufp->buffer, bufp->allocated, unsigned char); \ |
fa9a63c5 RM |
1854 | if (bufp->buffer == NULL) \ |
1855 | return REG_ESPACE; \ | |
1856 | /* If the buffer moved, move all the pointers into it. */ \ | |
1857 | if (old_buffer != bufp->buffer) \ | |
1858 | { \ | |
381880b0 | 1859 | unsigned char *new_buffer = bufp->buffer; \ |
4bb91c68 SM |
1860 | MOVE_BUFFER_POINTER (b); \ |
1861 | MOVE_BUFFER_POINTER (begalt); \ | |
25fe55af | 1862 | if (fixup_alt_jump) \ |
4bb91c68 | 1863 | MOVE_BUFFER_POINTER (fixup_alt_jump); \ |
25fe55af | 1864 | if (laststart) \ |
4bb91c68 | 1865 | MOVE_BUFFER_POINTER (laststart); \ |
25fe55af | 1866 | if (pending_exact) \ |
4bb91c68 | 1867 | MOVE_BUFFER_POINTER (pending_exact); \ |
fa9a63c5 | 1868 | } \ |
4bb91c68 | 1869 | ELSE_EXTEND_BUFFER_HIGH_BOUND \ |
fa9a63c5 RM |
1870 | } while (0) |
1871 | ||
1872 | ||
1873 | /* Since we have one byte reserved for the register number argument to | |
1874 | {start,stop}_memory, the maximum number of groups we can report | |
1875 | things about is what fits in that byte. */ | |
1876 | #define MAX_REGNUM 255 | |
1877 | ||
1878 | /* But patterns can have more than `MAX_REGNUM' registers. We just | |
1879 | ignore the excess. */ | |
098d42af | 1880 | typedef int regnum_t; |
fa9a63c5 RM |
1881 | |
1882 | ||
1883 | /* Macros for the compile stack. */ | |
1884 | ||
1885 | /* Since offsets can go either forwards or backwards, this type needs to | |
4bb91c68 SM |
1886 | be able to hold values from -(MAX_BUF_SIZE - 1) to MAX_BUF_SIZE - 1. */ |
1887 | /* int may be not enough when sizeof(int) == 2. */ | |
1888 | typedef long pattern_offset_t; | |
fa9a63c5 RM |
1889 | |
1890 | typedef struct | |
1891 | { | |
1892 | pattern_offset_t begalt_offset; | |
1893 | pattern_offset_t fixup_alt_jump; | |
5e69f11e | 1894 | pattern_offset_t laststart_offset; |
fa9a63c5 RM |
1895 | regnum_t regnum; |
1896 | } compile_stack_elt_t; | |
1897 | ||
1898 | ||
1899 | typedef struct | |
1900 | { | |
1901 | compile_stack_elt_t *stack; | |
1902 | unsigned size; | |
1903 | unsigned avail; /* Offset of next open position. */ | |
1904 | } compile_stack_type; | |
1905 | ||
1906 | ||
1907 | #define INIT_COMPILE_STACK_SIZE 32 | |
1908 | ||
1909 | #define COMPILE_STACK_EMPTY (compile_stack.avail == 0) | |
1910 | #define COMPILE_STACK_FULL (compile_stack.avail == compile_stack.size) | |
1911 | ||
4bb91c68 | 1912 | /* The next available element. */ |
fa9a63c5 RM |
1913 | #define COMPILE_STACK_TOP (compile_stack.stack[compile_stack.avail]) |
1914 | ||
1cee1e27 SM |
1915 | /* Explicit quit checking is only used on NTemacs and whenever we |
1916 | use polling to process input events. */ | |
1917 | #if defined emacs && (defined WINDOWSNT || defined SYNC_INPUT) && defined QUIT | |
77d11aec RS |
1918 | extern int immediate_quit; |
1919 | # define IMMEDIATE_QUIT_CHECK \ | |
1920 | do { \ | |
1921 | if (immediate_quit) QUIT; \ | |
1922 | } while (0) | |
1923 | #else | |
1924 | # define IMMEDIATE_QUIT_CHECK ((void)0) | |
1925 | #endif | |
1926 | \f | |
b18215fc RS |
1927 | /* Structure to manage work area for range table. */ |
1928 | struct range_table_work_area | |
1929 | { | |
1930 | int *table; /* actual work area. */ | |
1931 | int allocated; /* allocated size for work area in bytes. */ | |
7814e705 | 1932 | int used; /* actually used size in words. */ |
96cc36cc | 1933 | int bits; /* flag to record character classes */ |
b18215fc RS |
1934 | }; |
1935 | ||
77d11aec RS |
1936 | /* Make sure that WORK_AREA can hold more N multibyte characters. |
1937 | This is used only in set_image_of_range and set_image_of_range_1. | |
1938 | It expects WORK_AREA to be a pointer. | |
1939 | If it can't get the space, it returns from the surrounding function. */ | |
1940 | ||
1941 | #define EXTEND_RANGE_TABLE(work_area, n) \ | |
1942 | do { \ | |
8f924df7 | 1943 | if (((work_area).used + (n)) * sizeof (int) > (work_area).allocated) \ |
77d11aec | 1944 | { \ |
8f924df7 KH |
1945 | extend_range_table_work_area (&work_area); \ |
1946 | if ((work_area).table == 0) \ | |
77d11aec RS |
1947 | return (REG_ESPACE); \ |
1948 | } \ | |
b18215fc RS |
1949 | } while (0) |
1950 | ||
96cc36cc RS |
1951 | #define SET_RANGE_TABLE_WORK_AREA_BIT(work_area, bit) \ |
1952 | (work_area).bits |= (bit) | |
1953 | ||
14473664 SM |
1954 | /* Bits used to implement the multibyte-part of the various character classes |
1955 | such as [:alnum:] in a charset's range table. */ | |
1956 | #define BIT_WORD 0x1 | |
1957 | #define BIT_LOWER 0x2 | |
1958 | #define BIT_PUNCT 0x4 | |
1959 | #define BIT_SPACE 0x8 | |
1960 | #define BIT_UPPER 0x10 | |
1961 | #define BIT_MULTIBYTE 0x20 | |
96cc36cc | 1962 | |
b18215fc RS |
1963 | /* Set a range (RANGE_START, RANGE_END) to WORK_AREA. */ |
1964 | #define SET_RANGE_TABLE_WORK_AREA(work_area, range_start, range_end) \ | |
77d11aec | 1965 | do { \ |
8f924df7 | 1966 | EXTEND_RANGE_TABLE ((work_area), 2); \ |
b18215fc RS |
1967 | (work_area).table[(work_area).used++] = (range_start); \ |
1968 | (work_area).table[(work_area).used++] = (range_end); \ | |
1969 | } while (0) | |
1970 | ||
7814e705 | 1971 | /* Free allocated memory for WORK_AREA. */ |
b18215fc RS |
1972 | #define FREE_RANGE_TABLE_WORK_AREA(work_area) \ |
1973 | do { \ | |
1974 | if ((work_area).table) \ | |
1975 | free ((work_area).table); \ | |
1976 | } while (0) | |
1977 | ||
96cc36cc | 1978 | #define CLEAR_RANGE_TABLE_WORK_USED(work_area) ((work_area).used = 0, (work_area).bits = 0) |
b18215fc | 1979 | #define RANGE_TABLE_WORK_USED(work_area) ((work_area).used) |
96cc36cc | 1980 | #define RANGE_TABLE_WORK_BITS(work_area) ((work_area).bits) |
b18215fc | 1981 | #define RANGE_TABLE_WORK_ELT(work_area, i) ((work_area).table[i]) |
77d11aec | 1982 | \f |
b18215fc | 1983 | |
fa9a63c5 | 1984 | /* Set the bit for character C in a list. */ |
01618498 | 1985 | #define SET_LIST_BIT(c) (b[((c)) / BYTEWIDTH] |= 1 << ((c) % BYTEWIDTH)) |
fa9a63c5 RM |
1986 | |
1987 | ||
bf216479 KH |
1988 | #ifdef emacs |
1989 | ||
cf9c99bc KH |
1990 | /* Store characters in the range FROM to TO in the bitmap at B (for |
1991 | ASCII and unibyte characters) and WORK_AREA (for multibyte | |
1992 | characters) while translating them and paying attention to the | |
1993 | continuity of translated characters. | |
8f924df7 | 1994 | |
cf9c99bc KH |
1995 | Implementation note: It is better to implement these fairly big |
1996 | macros by a function, but it's not that easy because macros called | |
8f924df7 | 1997 | in this macro assume various local variables already declared. */ |
bf216479 | 1998 | |
cf9c99bc KH |
1999 | /* Both FROM and TO are ASCII characters. */ |
2000 | ||
2001 | #define SETUP_ASCII_RANGE(work_area, FROM, TO) \ | |
2002 | do { \ | |
2003 | int C0, C1; \ | |
2004 | \ | |
2005 | for (C0 = (FROM); C0 <= (TO); C0++) \ | |
2006 | { \ | |
2007 | C1 = TRANSLATE (C0); \ | |
2008 | if (! ASCII_CHAR_P (C1)) \ | |
2009 | { \ | |
2010 | SET_RANGE_TABLE_WORK_AREA ((work_area), C1, C1); \ | |
2011 | if ((C1 = RE_CHAR_TO_UNIBYTE (C1)) < 0) \ | |
2012 | C1 = C0; \ | |
2013 | } \ | |
2014 | SET_LIST_BIT (C1); \ | |
2015 | } \ | |
2016 | } while (0) | |
2017 | ||
2018 | ||
2019 | /* Both FROM and TO are unibyte characters (0x80..0xFF). */ | |
2020 | ||
2021 | #define SETUP_UNIBYTE_RANGE(work_area, FROM, TO) \ | |
2022 | do { \ | |
2023 | int C0, C1, C2, I; \ | |
2024 | int USED = RANGE_TABLE_WORK_USED (work_area); \ | |
2025 | \ | |
2026 | for (C0 = (FROM); C0 <= (TO); C0++) \ | |
2027 | { \ | |
2028 | C1 = RE_CHAR_TO_MULTIBYTE (C0); \ | |
2029 | if (CHAR_BYTE8_P (C1)) \ | |
2030 | SET_LIST_BIT (C0); \ | |
2031 | else \ | |
2032 | { \ | |
2033 | C2 = TRANSLATE (C1); \ | |
2034 | if (C2 == C1 \ | |
2035 | || (C1 = RE_CHAR_TO_UNIBYTE (C2)) < 0) \ | |
2036 | C1 = C0; \ | |
2037 | SET_LIST_BIT (C1); \ | |
2038 | for (I = RANGE_TABLE_WORK_USED (work_area) - 2; I >= USED; I -= 2) \ | |
2039 | { \ | |
2040 | int from = RANGE_TABLE_WORK_ELT (work_area, I); \ | |
2041 | int to = RANGE_TABLE_WORK_ELT (work_area, I + 1); \ | |
2042 | \ | |
2043 | if (C2 >= from - 1 && C2 <= to + 1) \ | |
2044 | { \ | |
2045 | if (C2 == from - 1) \ | |
2046 | RANGE_TABLE_WORK_ELT (work_area, I)--; \ | |
2047 | else if (C2 == to + 1) \ | |
2048 | RANGE_TABLE_WORK_ELT (work_area, I + 1)++; \ | |
2049 | break; \ | |
2050 | } \ | |
2051 | } \ | |
2052 | if (I < USED) \ | |
2053 | SET_RANGE_TABLE_WORK_AREA ((work_area), C2, C2); \ | |
2054 | } \ | |
2055 | } \ | |
2056 | } while (0) | |
2057 | ||
2058 | ||
78edd3b7 | 2059 | /* Both FROM and TO are multibyte characters. */ |
cf9c99bc KH |
2060 | |
2061 | #define SETUP_MULTIBYTE_RANGE(work_area, FROM, TO) \ | |
2062 | do { \ | |
2063 | int C0, C1, C2, I, USED = RANGE_TABLE_WORK_USED (work_area); \ | |
2064 | \ | |
2065 | SET_RANGE_TABLE_WORK_AREA ((work_area), (FROM), (TO)); \ | |
2066 | for (C0 = (FROM); C0 <= (TO); C0++) \ | |
2067 | { \ | |
2068 | C1 = TRANSLATE (C0); \ | |
2069 | if ((C2 = RE_CHAR_TO_UNIBYTE (C1)) >= 0 \ | |
2070 | || (C1 != C0 && (C2 = RE_CHAR_TO_UNIBYTE (C0)) >= 0)) \ | |
2071 | SET_LIST_BIT (C2); \ | |
2072 | if (C1 >= (FROM) && C1 <= (TO)) \ | |
2073 | continue; \ | |
2074 | for (I = RANGE_TABLE_WORK_USED (work_area) - 2; I >= USED; I -= 2) \ | |
2075 | { \ | |
2076 | int from = RANGE_TABLE_WORK_ELT (work_area, I); \ | |
2077 | int to = RANGE_TABLE_WORK_ELT (work_area, I + 1); \ | |
2078 | \ | |
2079 | if (C1 >= from - 1 && C1 <= to + 1) \ | |
2080 | { \ | |
2081 | if (C1 == from - 1) \ | |
2082 | RANGE_TABLE_WORK_ELT (work_area, I)--; \ | |
2083 | else if (C1 == to + 1) \ | |
2084 | RANGE_TABLE_WORK_ELT (work_area, I + 1)++; \ | |
2085 | break; \ | |
2086 | } \ | |
2087 | } \ | |
2088 | if (I < USED) \ | |
2089 | SET_RANGE_TABLE_WORK_AREA ((work_area), C1, C1); \ | |
2090 | } \ | |
bf216479 KH |
2091 | } while (0) |
2092 | ||
2093 | #endif /* emacs */ | |
2094 | ||
fa9a63c5 | 2095 | /* Get the next unsigned number in the uncompiled pattern. */ |
25fe55af | 2096 | #define GET_UNSIGNED_NUMBER(num) \ |
c72b0edd SM |
2097 | do { \ |
2098 | if (p == pend) \ | |
2099 | FREE_STACK_RETURN (REG_EBRACE); \ | |
2100 | else \ | |
2101 | { \ | |
2102 | PATFETCH (c); \ | |
2103 | while ('0' <= c && c <= '9') \ | |
2104 | { \ | |
2105 | int prev; \ | |
2106 | if (num < 0) \ | |
2107 | num = 0; \ | |
2108 | prev = num; \ | |
2109 | num = num * 10 + c - '0'; \ | |
2110 | if (num / 10 != prev) \ | |
2111 | FREE_STACK_RETURN (REG_BADBR); \ | |
2112 | if (p == pend) \ | |
2113 | FREE_STACK_RETURN (REG_EBRACE); \ | |
2114 | PATFETCH (c); \ | |
2115 | } \ | |
2116 | } \ | |
2117 | } while (0) | |
77d11aec | 2118 | \f |
1fdab503 | 2119 | #if ! WIDE_CHAR_SUPPORT |
01618498 | 2120 | |
14473664 | 2121 | /* Map a string to the char class it names (if any). */ |
1fdab503 | 2122 | re_wctype_t |
971de7fb | 2123 | re_wctype (const re_char *str) |
14473664 | 2124 | { |
ada30c0e | 2125 | const char *string = str; |
14473664 SM |
2126 | if (STREQ (string, "alnum")) return RECC_ALNUM; |
2127 | else if (STREQ (string, "alpha")) return RECC_ALPHA; | |
2128 | else if (STREQ (string, "word")) return RECC_WORD; | |
2129 | else if (STREQ (string, "ascii")) return RECC_ASCII; | |
2130 | else if (STREQ (string, "nonascii")) return RECC_NONASCII; | |
2131 | else if (STREQ (string, "graph")) return RECC_GRAPH; | |
2132 | else if (STREQ (string, "lower")) return RECC_LOWER; | |
2133 | else if (STREQ (string, "print")) return RECC_PRINT; | |
2134 | else if (STREQ (string, "punct")) return RECC_PUNCT; | |
2135 | else if (STREQ (string, "space")) return RECC_SPACE; | |
2136 | else if (STREQ (string, "upper")) return RECC_UPPER; | |
2137 | else if (STREQ (string, "unibyte")) return RECC_UNIBYTE; | |
2138 | else if (STREQ (string, "multibyte")) return RECC_MULTIBYTE; | |
2139 | else if (STREQ (string, "digit")) return RECC_DIGIT; | |
2140 | else if (STREQ (string, "xdigit")) return RECC_XDIGIT; | |
2141 | else if (STREQ (string, "cntrl")) return RECC_CNTRL; | |
2142 | else if (STREQ (string, "blank")) return RECC_BLANK; | |
2143 | else return 0; | |
2144 | } | |
2145 | ||
e0f24100 | 2146 | /* True if CH is in the char class CC. */ |
1fdab503 | 2147 | boolean |
971de7fb | 2148 | re_iswctype (int ch, re_wctype_t cc) |
14473664 SM |
2149 | { |
2150 | switch (cc) | |
2151 | { | |
0cdd06f8 SM |
2152 | case RECC_ALNUM: return ISALNUM (ch); |
2153 | case RECC_ALPHA: return ISALPHA (ch); | |
2154 | case RECC_BLANK: return ISBLANK (ch); | |
2155 | case RECC_CNTRL: return ISCNTRL (ch); | |
2156 | case RECC_DIGIT: return ISDIGIT (ch); | |
2157 | case RECC_GRAPH: return ISGRAPH (ch); | |
2158 | case RECC_LOWER: return ISLOWER (ch); | |
2159 | case RECC_PRINT: return ISPRINT (ch); | |
2160 | case RECC_PUNCT: return ISPUNCT (ch); | |
2161 | case RECC_SPACE: return ISSPACE (ch); | |
2162 | case RECC_UPPER: return ISUPPER (ch); | |
2163 | case RECC_XDIGIT: return ISXDIGIT (ch); | |
2164 | case RECC_ASCII: return IS_REAL_ASCII (ch); | |
2165 | case RECC_NONASCII: return !IS_REAL_ASCII (ch); | |
2166 | case RECC_UNIBYTE: return ISUNIBYTE (ch); | |
2167 | case RECC_MULTIBYTE: return !ISUNIBYTE (ch); | |
2168 | case RECC_WORD: return ISWORD (ch); | |
2169 | case RECC_ERROR: return false; | |
2170 | default: | |
2171 | abort(); | |
14473664 SM |
2172 | } |
2173 | } | |
fa9a63c5 | 2174 | |
14473664 SM |
2175 | /* Return a bit-pattern to use in the range-table bits to match multibyte |
2176 | chars of class CC. */ | |
2177 | static int | |
971de7fb | 2178 | re_wctype_to_bit (re_wctype_t cc) |
14473664 SM |
2179 | { |
2180 | switch (cc) | |
2181 | { | |
2182 | case RECC_NONASCII: case RECC_PRINT: case RECC_GRAPH: | |
0cdd06f8 SM |
2183 | case RECC_MULTIBYTE: return BIT_MULTIBYTE; |
2184 | case RECC_ALPHA: case RECC_ALNUM: case RECC_WORD: return BIT_WORD; | |
2185 | case RECC_LOWER: return BIT_LOWER; | |
2186 | case RECC_UPPER: return BIT_UPPER; | |
2187 | case RECC_PUNCT: return BIT_PUNCT; | |
2188 | case RECC_SPACE: return BIT_SPACE; | |
14473664 | 2189 | case RECC_ASCII: case RECC_DIGIT: case RECC_XDIGIT: case RECC_CNTRL: |
0cdd06f8 SM |
2190 | case RECC_BLANK: case RECC_UNIBYTE: case RECC_ERROR: return 0; |
2191 | default: | |
2192 | abort(); | |
14473664 SM |
2193 | } |
2194 | } | |
2195 | #endif | |
77d11aec RS |
2196 | \f |
2197 | /* Filling in the work area of a range. */ | |
2198 | ||
2199 | /* Actually extend the space in WORK_AREA. */ | |
2200 | ||
2201 | static void | |
971de7fb | 2202 | extend_range_table_work_area (struct range_table_work_area *work_area) |
177c0ea7 | 2203 | { |
77d11aec RS |
2204 | work_area->allocated += 16 * sizeof (int); |
2205 | if (work_area->table) | |
2206 | work_area->table | |
2207 | = (int *) realloc (work_area->table, work_area->allocated); | |
2208 | else | |
2209 | work_area->table | |
2210 | = (int *) malloc (work_area->allocated); | |
2211 | } | |
2212 | ||
8f924df7 | 2213 | #if 0 |
77d11aec RS |
2214 | #ifdef emacs |
2215 | ||
2216 | /* Carefully find the ranges of codes that are equivalent | |
2217 | under case conversion to the range start..end when passed through | |
2218 | TRANSLATE. Handle the case where non-letters can come in between | |
2219 | two upper-case letters (which happens in Latin-1). | |
2220 | Also handle the case of groups of more than 2 case-equivalent chars. | |
2221 | ||
2222 | The basic method is to look at consecutive characters and see | |
2223 | if they can form a run that can be handled as one. | |
2224 | ||
2225 | Returns -1 if successful, REG_ESPACE if ran out of space. */ | |
2226 | ||
2227 | static int | |
2228 | set_image_of_range_1 (work_area, start, end, translate) | |
2229 | RE_TRANSLATE_TYPE translate; | |
2230 | struct range_table_work_area *work_area; | |
2231 | re_wchar_t start, end; | |
2232 | { | |
2233 | /* `one_case' indicates a character, or a run of characters, | |
2234 | each of which is an isolate (no case-equivalents). | |
2235 | This includes all ASCII non-letters. | |
2236 | ||
2237 | `two_case' indicates a character, or a run of characters, | |
2238 | each of which has two case-equivalent forms. | |
2239 | This includes all ASCII letters. | |
2240 | ||
2241 | `strange' indicates a character that has more than one | |
2242 | case-equivalent. */ | |
177c0ea7 | 2243 | |
77d11aec RS |
2244 | enum case_type {one_case, two_case, strange}; |
2245 | ||
2246 | /* Describe the run that is in progress, | |
2247 | which the next character can try to extend. | |
2248 | If run_type is strange, that means there really is no run. | |
2249 | If run_type is one_case, then run_start...run_end is the run. | |
2250 | If run_type is two_case, then the run is run_start...run_end, | |
2251 | and the case-equivalents end at run_eqv_end. */ | |
2252 | ||
2253 | enum case_type run_type = strange; | |
2254 | int run_start, run_end, run_eqv_end; | |
2255 | ||
2256 | Lisp_Object eqv_table; | |
2257 | ||
2258 | if (!RE_TRANSLATE_P (translate)) | |
2259 | { | |
b7c12565 | 2260 | EXTEND_RANGE_TABLE (work_area, 2); |
77d11aec RS |
2261 | work_area->table[work_area->used++] = (start); |
2262 | work_area->table[work_area->used++] = (end); | |
b7c12565 | 2263 | return -1; |
77d11aec RS |
2264 | } |
2265 | ||
2266 | eqv_table = XCHAR_TABLE (translate)->extras[2]; | |
99633e97 | 2267 | |
77d11aec RS |
2268 | for (; start <= end; start++) |
2269 | { | |
2270 | enum case_type this_type; | |
2271 | int eqv = RE_TRANSLATE (eqv_table, start); | |
2272 | int minchar, maxchar; | |
2273 | ||
2274 | /* Classify this character */ | |
2275 | if (eqv == start) | |
2276 | this_type = one_case; | |
2277 | else if (RE_TRANSLATE (eqv_table, eqv) == start) | |
2278 | this_type = two_case; | |
2279 | else | |
2280 | this_type = strange; | |
2281 | ||
2282 | if (start < eqv) | |
2283 | minchar = start, maxchar = eqv; | |
2284 | else | |
2285 | minchar = eqv, maxchar = start; | |
2286 | ||
2287 | /* Can this character extend the run in progress? */ | |
2288 | if (this_type == strange || this_type != run_type | |
2289 | || !(minchar == run_end + 1 | |
2290 | && (run_type == two_case | |
2291 | ? maxchar == run_eqv_end + 1 : 1))) | |
2292 | { | |
2293 | /* No, end the run. | |
2294 | Record each of its equivalent ranges. */ | |
2295 | if (run_type == one_case) | |
2296 | { | |
2297 | EXTEND_RANGE_TABLE (work_area, 2); | |
2298 | work_area->table[work_area->used++] = run_start; | |
2299 | work_area->table[work_area->used++] = run_end; | |
2300 | } | |
2301 | else if (run_type == two_case) | |
2302 | { | |
2303 | EXTEND_RANGE_TABLE (work_area, 4); | |
2304 | work_area->table[work_area->used++] = run_start; | |
2305 | work_area->table[work_area->used++] = run_end; | |
2306 | work_area->table[work_area->used++] | |
2307 | = RE_TRANSLATE (eqv_table, run_start); | |
2308 | work_area->table[work_area->used++] | |
2309 | = RE_TRANSLATE (eqv_table, run_end); | |
2310 | } | |
2311 | run_type = strange; | |
2312 | } | |
177c0ea7 | 2313 | |
77d11aec RS |
2314 | if (this_type == strange) |
2315 | { | |
2316 | /* For a strange character, add each of its equivalents, one | |
2317 | by one. Don't start a range. */ | |
2318 | do | |
2319 | { | |
2320 | EXTEND_RANGE_TABLE (work_area, 2); | |
2321 | work_area->table[work_area->used++] = eqv; | |
2322 | work_area->table[work_area->used++] = eqv; | |
2323 | eqv = RE_TRANSLATE (eqv_table, eqv); | |
2324 | } | |
2325 | while (eqv != start); | |
2326 | } | |
2327 | ||
2328 | /* Add this char to the run, or start a new run. */ | |
2329 | else if (run_type == strange) | |
2330 | { | |
2331 | /* Initialize a new range. */ | |
2332 | run_type = this_type; | |
2333 | run_start = start; | |
2334 | run_end = start; | |
2335 | run_eqv_end = RE_TRANSLATE (eqv_table, run_end); | |
2336 | } | |
2337 | else | |
2338 | { | |
2339 | /* Extend a running range. */ | |
2340 | run_end = minchar; | |
2341 | run_eqv_end = RE_TRANSLATE (eqv_table, run_end); | |
2342 | } | |
2343 | } | |
2344 | ||
2345 | /* If a run is still in progress at the end, finish it now | |
2346 | by recording its equivalent ranges. */ | |
2347 | if (run_type == one_case) | |
2348 | { | |
2349 | EXTEND_RANGE_TABLE (work_area, 2); | |
2350 | work_area->table[work_area->used++] = run_start; | |
2351 | work_area->table[work_area->used++] = run_end; | |
2352 | } | |
2353 | else if (run_type == two_case) | |
2354 | { | |
2355 | EXTEND_RANGE_TABLE (work_area, 4); | |
2356 | work_area->table[work_area->used++] = run_start; | |
2357 | work_area->table[work_area->used++] = run_end; | |
2358 | work_area->table[work_area->used++] | |
2359 | = RE_TRANSLATE (eqv_table, run_start); | |
2360 | work_area->table[work_area->used++] | |
2361 | = RE_TRANSLATE (eqv_table, run_end); | |
2362 | } | |
2363 | ||
2364 | return -1; | |
2365 | } | |
36595814 | 2366 | |
77d11aec | 2367 | #endif /* emacs */ |
36595814 | 2368 | |
2b34df4e | 2369 | /* Record the image of the range start..end when passed through |
36595814 SM |
2370 | TRANSLATE. This is not necessarily TRANSLATE(start)..TRANSLATE(end) |
2371 | and is not even necessarily contiguous. | |
b7c12565 RS |
2372 | Normally we approximate it with the smallest contiguous range that contains |
2373 | all the chars we need. However, for Latin-1 we go to extra effort | |
2374 | to do a better job. | |
2375 | ||
2376 | This function is not called for ASCII ranges. | |
77d11aec RS |
2377 | |
2378 | Returns -1 if successful, REG_ESPACE if ran out of space. */ | |
2379 | ||
2380 | static int | |
36595814 SM |
2381 | set_image_of_range (work_area, start, end, translate) |
2382 | RE_TRANSLATE_TYPE translate; | |
2383 | struct range_table_work_area *work_area; | |
2384 | re_wchar_t start, end; | |
2385 | { | |
77d11aec RS |
2386 | re_wchar_t cmin, cmax; |
2387 | ||
2388 | #ifdef emacs | |
2389 | /* For Latin-1 ranges, use set_image_of_range_1 | |
2390 | to get proper handling of ranges that include letters and nonletters. | |
b7c12565 | 2391 | For a range that includes the whole of Latin-1, this is not necessary. |
77d11aec | 2392 | For other character sets, we don't bother to get this right. */ |
b7c12565 RS |
2393 | if (RE_TRANSLATE_P (translate) && start < 04400 |
2394 | && !(start < 04200 && end >= 04377)) | |
77d11aec | 2395 | { |
b7c12565 | 2396 | int newend; |
77d11aec | 2397 | int tem; |
b7c12565 RS |
2398 | newend = end; |
2399 | if (newend > 04377) | |
2400 | newend = 04377; | |
2401 | tem = set_image_of_range_1 (work_area, start, newend, translate); | |
77d11aec RS |
2402 | if (tem > 0) |
2403 | return tem; | |
2404 | ||
2405 | start = 04400; | |
2406 | if (end < 04400) | |
2407 | return -1; | |
2408 | } | |
2409 | #endif | |
2410 | ||
b7c12565 RS |
2411 | EXTEND_RANGE_TABLE (work_area, 2); |
2412 | work_area->table[work_area->used++] = (start); | |
2413 | work_area->table[work_area->used++] = (end); | |
2414 | ||
2415 | cmin = -1, cmax = -1; | |
77d11aec | 2416 | |
36595814 | 2417 | if (RE_TRANSLATE_P (translate)) |
b7c12565 RS |
2418 | { |
2419 | int ch; | |
77d11aec | 2420 | |
b7c12565 RS |
2421 | for (ch = start; ch <= end; ch++) |
2422 | { | |
2423 | re_wchar_t c = TRANSLATE (ch); | |
2424 | if (! (start <= c && c <= end)) | |
2425 | { | |
2426 | if (cmin == -1) | |
2427 | cmin = c, cmax = c; | |
2428 | else | |
2429 | { | |
2430 | cmin = MIN (cmin, c); | |
2431 | cmax = MAX (cmax, c); | |
2432 | } | |
2433 | } | |
2434 | } | |
2435 | ||
2436 | if (cmin != -1) | |
2437 | { | |
2438 | EXTEND_RANGE_TABLE (work_area, 2); | |
2439 | work_area->table[work_area->used++] = (cmin); | |
2440 | work_area->table[work_area->used++] = (cmax); | |
2441 | } | |
2442 | } | |
36595814 | 2443 | |
77d11aec RS |
2444 | return -1; |
2445 | } | |
8f924df7 | 2446 | #endif /* 0 */ |
fa9a63c5 RM |
2447 | \f |
2448 | #ifndef MATCH_MAY_ALLOCATE | |
2449 | ||
2450 | /* If we cannot allocate large objects within re_match_2_internal, | |
2451 | we make the fail stack and register vectors global. | |
2452 | The fail stack, we grow to the maximum size when a regexp | |
2453 | is compiled. | |
2454 | The register vectors, we adjust in size each time we | |
2455 | compile a regexp, according to the number of registers it needs. */ | |
2456 | ||
2457 | static fail_stack_type fail_stack; | |
2458 | ||
2459 | /* Size with which the following vectors are currently allocated. | |
2460 | That is so we can make them bigger as needed, | |
4bb91c68 | 2461 | but never make them smaller. */ |
fa9a63c5 RM |
2462 | static int regs_allocated_size; |
2463 | ||
66f0296e SM |
2464 | static re_char ** regstart, ** regend; |
2465 | static re_char **best_regstart, **best_regend; | |
fa9a63c5 RM |
2466 | |
2467 | /* Make the register vectors big enough for NUM_REGS registers, | |
4bb91c68 | 2468 | but don't make them smaller. */ |
fa9a63c5 RM |
2469 | |
2470 | static | |
2471 | regex_grow_registers (num_regs) | |
2472 | int num_regs; | |
2473 | { | |
2474 | if (num_regs > regs_allocated_size) | |
2475 | { | |
66f0296e SM |
2476 | RETALLOC_IF (regstart, num_regs, re_char *); |
2477 | RETALLOC_IF (regend, num_regs, re_char *); | |
2478 | RETALLOC_IF (best_regstart, num_regs, re_char *); | |
2479 | RETALLOC_IF (best_regend, num_regs, re_char *); | |
fa9a63c5 RM |
2480 | |
2481 | regs_allocated_size = num_regs; | |
2482 | } | |
2483 | } | |
2484 | ||
2485 | #endif /* not MATCH_MAY_ALLOCATE */ | |
2486 | \f | |
99633e97 SM |
2487 | static boolean group_in_compile_stack _RE_ARGS ((compile_stack_type |
2488 | compile_stack, | |
2489 | regnum_t regnum)); | |
2490 | ||
fa9a63c5 RM |
2491 | /* `regex_compile' compiles PATTERN (of length SIZE) according to SYNTAX. |
2492 | Returns one of error codes defined in `regex.h', or zero for success. | |
2493 | ||
2494 | Assumes the `allocated' (and perhaps `buffer') and `translate' | |
2495 | fields are set in BUFP on entry. | |
2496 | ||
2497 | If it succeeds, results are put in BUFP (if it returns an error, the | |
2498 | contents of BUFP are undefined): | |
2499 | `buffer' is the compiled pattern; | |
2500 | `syntax' is set to SYNTAX; | |
2501 | `used' is set to the length of the compiled pattern; | |
2502 | `fastmap_accurate' is zero; | |
2503 | `re_nsub' is the number of subexpressions in PATTERN; | |
2504 | `not_bol' and `not_eol' are zero; | |
5e69f11e | 2505 | |
c0f9ea08 | 2506 | The `fastmap' field is neither examined nor set. */ |
fa9a63c5 | 2507 | |
505bde11 SM |
2508 | /* Insert the `jump' from the end of last alternative to "here". |
2509 | The space for the jump has already been allocated. */ | |
2510 | #define FIXUP_ALT_JUMP() \ | |
2511 | do { \ | |
2512 | if (fixup_alt_jump) \ | |
2513 | STORE_JUMP (jump, fixup_alt_jump, b); \ | |
2514 | } while (0) | |
2515 | ||
2516 | ||
fa9a63c5 RM |
2517 | /* Return, freeing storage we allocated. */ |
2518 | #define FREE_STACK_RETURN(value) \ | |
b18215fc RS |
2519 | do { \ |
2520 | FREE_RANGE_TABLE_WORK_AREA (range_table_work); \ | |
2521 | free (compile_stack.stack); \ | |
2522 | return value; \ | |
2523 | } while (0) | |
fa9a63c5 RM |
2524 | |
2525 | static reg_errcode_t | |
971de7fb | 2526 | regex_compile (const re_char *pattern, size_t size, reg_syntax_t syntax, struct re_pattern_buffer *bufp) |
fa9a63c5 | 2527 | { |
01618498 SM |
2528 | /* We fetch characters from PATTERN here. */ |
2529 | register re_wchar_t c, c1; | |
5e69f11e | 2530 | |
fa9a63c5 | 2531 | /* A random temporary spot in PATTERN. */ |
66f0296e | 2532 | re_char *p1; |
fa9a63c5 RM |
2533 | |
2534 | /* Points to the end of the buffer, where we should append. */ | |
2535 | register unsigned char *b; | |
5e69f11e | 2536 | |
fa9a63c5 RM |
2537 | /* Keeps track of unclosed groups. */ |
2538 | compile_stack_type compile_stack; | |
2539 | ||
2540 | /* Points to the current (ending) position in the pattern. */ | |
22336245 RS |
2541 | #ifdef AIX |
2542 | /* `const' makes AIX compiler fail. */ | |
66f0296e | 2543 | unsigned char *p = pattern; |
22336245 | 2544 | #else |
66f0296e | 2545 | re_char *p = pattern; |
22336245 | 2546 | #endif |
66f0296e | 2547 | re_char *pend = pattern + size; |
5e69f11e | 2548 | |
fa9a63c5 | 2549 | /* How to translate the characters in the pattern. */ |
6676cb1c | 2550 | RE_TRANSLATE_TYPE translate = bufp->translate; |
fa9a63c5 RM |
2551 | |
2552 | /* Address of the count-byte of the most recently inserted `exactn' | |
2553 | command. This makes it possible to tell if a new exact-match | |
2554 | character can be added to that command or if the character requires | |
2555 | a new `exactn' command. */ | |
2556 | unsigned char *pending_exact = 0; | |
2557 | ||
2558 | /* Address of start of the most recently finished expression. | |
2559 | This tells, e.g., postfix * where to find the start of its | |
2560 | operand. Reset at the beginning of groups and alternatives. */ | |
2561 | unsigned char *laststart = 0; | |
2562 | ||
2563 | /* Address of beginning of regexp, or inside of last group. */ | |
2564 | unsigned char *begalt; | |
2565 | ||
2566 | /* Place in the uncompiled pattern (i.e., the {) to | |
2567 | which to go back if the interval is invalid. */ | |
66f0296e | 2568 | re_char *beg_interval; |
5e69f11e | 2569 | |
fa9a63c5 | 2570 | /* Address of the place where a forward jump should go to the end of |
7814e705 | 2571 | the containing expression. Each alternative of an `or' -- except the |
fa9a63c5 RM |
2572 | last -- ends with a forward jump of this sort. */ |
2573 | unsigned char *fixup_alt_jump = 0; | |
2574 | ||
b18215fc RS |
2575 | /* Work area for range table of charset. */ |
2576 | struct range_table_work_area range_table_work; | |
2577 | ||
2d1675e4 SM |
2578 | /* If the object matched can contain multibyte characters. */ |
2579 | const boolean multibyte = RE_MULTIBYTE_P (bufp); | |
2580 | ||
8f924df7 | 2581 | /* If a target of matching can contain multibyte characters. */ |
6fdd04b0 KH |
2582 | const boolean target_multibyte = RE_TARGET_MULTIBYTE_P (bufp); |
2583 | ||
f9b0fd99 RS |
2584 | /* Nonzero if we have pushed down into a subpattern. */ |
2585 | int in_subpattern = 0; | |
2586 | ||
2587 | /* These hold the values of p, pattern, and pend from the main | |
2588 | pattern when we have pushed into a subpattern. */ | |
2589 | re_char *main_p; | |
2590 | re_char *main_pattern; | |
2591 | re_char *main_pend; | |
2592 | ||
fa9a63c5 | 2593 | #ifdef DEBUG |
99633e97 | 2594 | debug++; |
fa9a63c5 | 2595 | DEBUG_PRINT1 ("\nCompiling pattern: "); |
99633e97 | 2596 | if (debug > 0) |
fa9a63c5 RM |
2597 | { |
2598 | unsigned debug_count; | |
5e69f11e | 2599 | |
fa9a63c5 | 2600 | for (debug_count = 0; debug_count < size; debug_count++) |
25fe55af | 2601 | putchar (pattern[debug_count]); |
fa9a63c5 RM |
2602 | putchar ('\n'); |
2603 | } | |
2604 | #endif /* DEBUG */ | |
2605 | ||
2606 | /* Initialize the compile stack. */ | |
2607 | compile_stack.stack = TALLOC (INIT_COMPILE_STACK_SIZE, compile_stack_elt_t); | |
2608 | if (compile_stack.stack == NULL) | |
2609 | return REG_ESPACE; | |
2610 | ||
2611 | compile_stack.size = INIT_COMPILE_STACK_SIZE; | |
2612 | compile_stack.avail = 0; | |
2613 | ||
b18215fc RS |
2614 | range_table_work.table = 0; |
2615 | range_table_work.allocated = 0; | |
2616 | ||
fa9a63c5 RM |
2617 | /* Initialize the pattern buffer. */ |
2618 | bufp->syntax = syntax; | |
2619 | bufp->fastmap_accurate = 0; | |
2620 | bufp->not_bol = bufp->not_eol = 0; | |
6224b623 | 2621 | bufp->used_syntax = 0; |
fa9a63c5 RM |
2622 | |
2623 | /* Set `used' to zero, so that if we return an error, the pattern | |
2624 | printer (for debugging) will think there's no pattern. We reset it | |
2625 | at the end. */ | |
2626 | bufp->used = 0; | |
5e69f11e | 2627 | |
fa9a63c5 | 2628 | /* Always count groups, whether or not bufp->no_sub is set. */ |
5e69f11e | 2629 | bufp->re_nsub = 0; |
fa9a63c5 | 2630 | |
0b32bf0e | 2631 | #if !defined emacs && !defined SYNTAX_TABLE |
fa9a63c5 RM |
2632 | /* Initialize the syntax table. */ |
2633 | init_syntax_once (); | |
2634 | #endif | |
2635 | ||
2636 | if (bufp->allocated == 0) | |
2637 | { | |
2638 | if (bufp->buffer) | |
2639 | { /* If zero allocated, but buffer is non-null, try to realloc | |
25fe55af | 2640 | enough space. This loses if buffer's address is bogus, but |
7814e705 | 2641 | that is the user's responsibility. */ |
25fe55af RS |
2642 | RETALLOC (bufp->buffer, INIT_BUF_SIZE, unsigned char); |
2643 | } | |
fa9a63c5 | 2644 | else |
7814e705 | 2645 | { /* Caller did not allocate a buffer. Do it for them. */ |
25fe55af RS |
2646 | bufp->buffer = TALLOC (INIT_BUF_SIZE, unsigned char); |
2647 | } | |
fa9a63c5 RM |
2648 | if (!bufp->buffer) FREE_STACK_RETURN (REG_ESPACE); |
2649 | ||
2650 | bufp->allocated = INIT_BUF_SIZE; | |
2651 | } | |
2652 | ||
2653 | begalt = b = bufp->buffer; | |
2654 | ||
2655 | /* Loop through the uncompiled pattern until we're at the end. */ | |
f9b0fd99 | 2656 | while (1) |
fa9a63c5 | 2657 | { |
f9b0fd99 RS |
2658 | if (p == pend) |
2659 | { | |
2660 | /* If this is the end of an included regexp, | |
2661 | pop back to the main regexp and try again. */ | |
2662 | if (in_subpattern) | |
2663 | { | |
2664 | in_subpattern = 0; | |
2665 | pattern = main_pattern; | |
2666 | p = main_p; | |
2667 | pend = main_pend; | |
2668 | continue; | |
2669 | } | |
2670 | /* If this is the end of the main regexp, we are done. */ | |
2671 | break; | |
2672 | } | |
2673 | ||
fa9a63c5 RM |
2674 | PATFETCH (c); |
2675 | ||
2676 | switch (c) | |
25fe55af | 2677 | { |
f9b0fd99 RS |
2678 | case ' ': |
2679 | { | |
2680 | re_char *p1 = p; | |
2681 | ||
2682 | /* If there's no special whitespace regexp, treat | |
4fb680cd RS |
2683 | spaces normally. And don't try to do this recursively. */ |
2684 | if (!whitespace_regexp || in_subpattern) | |
f9b0fd99 RS |
2685 | goto normal_char; |
2686 | ||
2687 | /* Peek past following spaces. */ | |
2688 | while (p1 != pend) | |
2689 | { | |
2690 | if (*p1 != ' ') | |
2691 | break; | |
2692 | p1++; | |
2693 | } | |
2694 | /* If the spaces are followed by a repetition op, | |
2695 | treat them normally. */ | |
c721eee5 RS |
2696 | if (p1 != pend |
2697 | && (*p1 == '*' || *p1 == '+' || *p1 == '?' | |
f9b0fd99 RS |
2698 | || (*p1 == '\\' && p1 + 1 != pend && p1[1] == '{'))) |
2699 | goto normal_char; | |
2700 | ||
2701 | /* Replace the spaces with the whitespace regexp. */ | |
2702 | in_subpattern = 1; | |
2703 | main_p = p1; | |
2704 | main_pend = pend; | |
2705 | main_pattern = pattern; | |
2706 | p = pattern = whitespace_regexp; | |
2707 | pend = p + strlen (p); | |
2708 | break; | |
7814e705 | 2709 | } |
f9b0fd99 | 2710 | |
25fe55af RS |
2711 | case '^': |
2712 | { | |
7814e705 | 2713 | if ( /* If at start of pattern, it's an operator. */ |
25fe55af | 2714 | p == pattern + 1 |
7814e705 | 2715 | /* If context independent, it's an operator. */ |
25fe55af | 2716 | || syntax & RE_CONTEXT_INDEP_ANCHORS |
7814e705 | 2717 | /* Otherwise, depends on what's come before. */ |
25fe55af | 2718 | || at_begline_loc_p (pattern, p, syntax)) |
c0f9ea08 | 2719 | BUF_PUSH ((syntax & RE_NO_NEWLINE_ANCHOR) ? begbuf : begline); |
25fe55af RS |
2720 | else |
2721 | goto normal_char; | |
2722 | } | |
2723 | break; | |
2724 | ||
2725 | ||
2726 | case '$': | |
2727 | { | |
2728 | if ( /* If at end of pattern, it's an operator. */ | |
2729 | p == pend | |
7814e705 | 2730 | /* If context independent, it's an operator. */ |
25fe55af RS |
2731 | || syntax & RE_CONTEXT_INDEP_ANCHORS |
2732 | /* Otherwise, depends on what's next. */ | |
2733 | || at_endline_loc_p (p, pend, syntax)) | |
c0f9ea08 | 2734 | BUF_PUSH ((syntax & RE_NO_NEWLINE_ANCHOR) ? endbuf : endline); |
25fe55af RS |
2735 | else |
2736 | goto normal_char; | |
2737 | } | |
2738 | break; | |
fa9a63c5 RM |
2739 | |
2740 | ||
2741 | case '+': | |
25fe55af RS |
2742 | case '?': |
2743 | if ((syntax & RE_BK_PLUS_QM) | |
2744 | || (syntax & RE_LIMITED_OPS)) | |
2745 | goto normal_char; | |
2746 | handle_plus: | |
2747 | case '*': | |
2748 | /* If there is no previous pattern... */ | |
2749 | if (!laststart) | |
2750 | { | |
2751 | if (syntax & RE_CONTEXT_INVALID_OPS) | |
2752 | FREE_STACK_RETURN (REG_BADRPT); | |
2753 | else if (!(syntax & RE_CONTEXT_INDEP_OPS)) | |
2754 | goto normal_char; | |
2755 | } | |
2756 | ||
2757 | { | |
7814e705 | 2758 | /* 1 means zero (many) matches is allowed. */ |
66f0296e SM |
2759 | boolean zero_times_ok = 0, many_times_ok = 0; |
2760 | boolean greedy = 1; | |
25fe55af RS |
2761 | |
2762 | /* If there is a sequence of repetition chars, collapse it | |
2763 | down to just one (the right one). We can't combine | |
2764 | interval operators with these because of, e.g., `a{2}*', | |
7814e705 | 2765 | which should only match an even number of `a's. */ |
25fe55af RS |
2766 | |
2767 | for (;;) | |
2768 | { | |
0b32bf0e | 2769 | if ((syntax & RE_FRUGAL) |
1c8c6d39 DL |
2770 | && c == '?' && (zero_times_ok || many_times_ok)) |
2771 | greedy = 0; | |
2772 | else | |
2773 | { | |
2774 | zero_times_ok |= c != '+'; | |
2775 | many_times_ok |= c != '?'; | |
2776 | } | |
25fe55af RS |
2777 | |
2778 | if (p == pend) | |
2779 | break; | |
ed0767d8 SM |
2780 | else if (*p == '*' |
2781 | || (!(syntax & RE_BK_PLUS_QM) | |
2782 | && (*p == '+' || *p == '?'))) | |
25fe55af | 2783 | ; |
ed0767d8 | 2784 | else if (syntax & RE_BK_PLUS_QM && *p == '\\') |
25fe55af | 2785 | { |
ed0767d8 SM |
2786 | if (p+1 == pend) |
2787 | FREE_STACK_RETURN (REG_EESCAPE); | |
2788 | if (p[1] == '+' || p[1] == '?') | |
2789 | PATFETCH (c); /* Gobble up the backslash. */ | |
2790 | else | |
2791 | break; | |
25fe55af RS |
2792 | } |
2793 | else | |
ed0767d8 | 2794 | break; |
25fe55af | 2795 | /* If we get here, we found another repeat character. */ |
ed0767d8 SM |
2796 | PATFETCH (c); |
2797 | } | |
25fe55af RS |
2798 | |
2799 | /* Star, etc. applied to an empty pattern is equivalent | |
2800 | to an empty pattern. */ | |
4e8a9132 | 2801 | if (!laststart || laststart == b) |
25fe55af RS |
2802 | break; |
2803 | ||
2804 | /* Now we know whether or not zero matches is allowed | |
7814e705 | 2805 | and also whether or not two or more matches is allowed. */ |
1c8c6d39 DL |
2806 | if (greedy) |
2807 | { | |
99633e97 | 2808 | if (many_times_ok) |
4e8a9132 SM |
2809 | { |
2810 | boolean simple = skip_one_char (laststart) == b; | |
2811 | unsigned int startoffset = 0; | |
f6a3f532 | 2812 | re_opcode_t ofj = |
01618498 | 2813 | /* Check if the loop can match the empty string. */ |
6df42991 SM |
2814 | (simple || !analyse_first (laststart, b, NULL, 0)) |
2815 | ? on_failure_jump : on_failure_jump_loop; | |
4e8a9132 | 2816 | assert (skip_one_char (laststart) <= b); |
177c0ea7 | 2817 | |
4e8a9132 SM |
2818 | if (!zero_times_ok && simple) |
2819 | { /* Since simple * loops can be made faster by using | |
2820 | on_failure_keep_string_jump, we turn simple P+ | |
2821 | into PP* if P is simple. */ | |
2822 | unsigned char *p1, *p2; | |
2823 | startoffset = b - laststart; | |
2824 | GET_BUFFER_SPACE (startoffset); | |
2825 | p1 = b; p2 = laststart; | |
2826 | while (p2 < p1) | |
2827 | *b++ = *p2++; | |
2828 | zero_times_ok = 1; | |
99633e97 | 2829 | } |
4e8a9132 SM |
2830 | |
2831 | GET_BUFFER_SPACE (6); | |
2832 | if (!zero_times_ok) | |
2833 | /* A + loop. */ | |
f6a3f532 | 2834 | STORE_JUMP (ofj, b, b + 6); |
99633e97 | 2835 | else |
4e8a9132 SM |
2836 | /* Simple * loops can use on_failure_keep_string_jump |
2837 | depending on what follows. But since we don't know | |
2838 | that yet, we leave the decision up to | |
2839 | on_failure_jump_smart. */ | |
f6a3f532 | 2840 | INSERT_JUMP (simple ? on_failure_jump_smart : ofj, |
4e8a9132 | 2841 | laststart + startoffset, b + 6); |
99633e97 | 2842 | b += 3; |
4e8a9132 | 2843 | STORE_JUMP (jump, b, laststart + startoffset); |
99633e97 SM |
2844 | b += 3; |
2845 | } | |
2846 | else | |
2847 | { | |
4e8a9132 SM |
2848 | /* A simple ? pattern. */ |
2849 | assert (zero_times_ok); | |
2850 | GET_BUFFER_SPACE (3); | |
2851 | INSERT_JUMP (on_failure_jump, laststart, b + 3); | |
99633e97 SM |
2852 | b += 3; |
2853 | } | |
1c8c6d39 DL |
2854 | } |
2855 | else /* not greedy */ | |
2856 | { /* I wish the greedy and non-greedy cases could be merged. */ | |
2857 | ||
0683b6fa | 2858 | GET_BUFFER_SPACE (7); /* We might use less. */ |
1c8c6d39 DL |
2859 | if (many_times_ok) |
2860 | { | |
f6a3f532 SM |
2861 | boolean emptyp = analyse_first (laststart, b, NULL, 0); |
2862 | ||
6df42991 SM |
2863 | /* The non-greedy multiple match looks like |
2864 | a repeat..until: we only need a conditional jump | |
2865 | at the end of the loop. */ | |
f6a3f532 SM |
2866 | if (emptyp) BUF_PUSH (no_op); |
2867 | STORE_JUMP (emptyp ? on_failure_jump_nastyloop | |
2868 | : on_failure_jump, b, laststart); | |
1c8c6d39 DL |
2869 | b += 3; |
2870 | if (zero_times_ok) | |
2871 | { | |
2872 | /* The repeat...until naturally matches one or more. | |
2873 | To also match zero times, we need to first jump to | |
6df42991 | 2874 | the end of the loop (its conditional jump). */ |
1c8c6d39 DL |
2875 | INSERT_JUMP (jump, laststart, b); |
2876 | b += 3; | |
2877 | } | |
2878 | } | |
2879 | else | |
2880 | { | |
2881 | /* non-greedy a?? */ | |
1c8c6d39 DL |
2882 | INSERT_JUMP (jump, laststart, b + 3); |
2883 | b += 3; | |
2884 | INSERT_JUMP (on_failure_jump, laststart, laststart + 6); | |
2885 | b += 3; | |
2886 | } | |
2887 | } | |
2888 | } | |
4e8a9132 | 2889 | pending_exact = 0; |
fa9a63c5 RM |
2890 | break; |
2891 | ||
2892 | ||
2893 | case '.': | |
25fe55af RS |
2894 | laststart = b; |
2895 | BUF_PUSH (anychar); | |
2896 | break; | |
fa9a63c5 RM |
2897 | |
2898 | ||
25fe55af RS |
2899 | case '[': |
2900 | { | |
b18215fc | 2901 | CLEAR_RANGE_TABLE_WORK_USED (range_table_work); |
fa9a63c5 | 2902 | |
25fe55af | 2903 | if (p == pend) FREE_STACK_RETURN (REG_EBRACK); |
fa9a63c5 | 2904 | |
25fe55af RS |
2905 | /* Ensure that we have enough space to push a charset: the |
2906 | opcode, the length count, and the bitset; 34 bytes in all. */ | |
fa9a63c5 RM |
2907 | GET_BUFFER_SPACE (34); |
2908 | ||
25fe55af | 2909 | laststart = b; |
e318085a | 2910 | |
25fe55af | 2911 | /* We test `*p == '^' twice, instead of using an if |
7814e705 | 2912 | statement, so we only need one BUF_PUSH. */ |
25fe55af RS |
2913 | BUF_PUSH (*p == '^' ? charset_not : charset); |
2914 | if (*p == '^') | |
2915 | p++; | |
e318085a | 2916 | |
25fe55af RS |
2917 | /* Remember the first position in the bracket expression. */ |
2918 | p1 = p; | |
e318085a | 2919 | |
7814e705 | 2920 | /* Push the number of bytes in the bitmap. */ |
25fe55af | 2921 | BUF_PUSH ((1 << BYTEWIDTH) / BYTEWIDTH); |
e318085a | 2922 | |
25fe55af | 2923 | /* Clear the whole map. */ |
72af86bd | 2924 | memset (b, 0, (1 << BYTEWIDTH) / BYTEWIDTH); |
e318085a | 2925 | |
25fe55af RS |
2926 | /* charset_not matches newline according to a syntax bit. */ |
2927 | if ((re_opcode_t) b[-2] == charset_not | |
2928 | && (syntax & RE_HAT_LISTS_NOT_NEWLINE)) | |
2929 | SET_LIST_BIT ('\n'); | |
fa9a63c5 | 2930 | |
7814e705 | 2931 | /* Read in characters and ranges, setting map bits. */ |
25fe55af RS |
2932 | for (;;) |
2933 | { | |
b18215fc | 2934 | boolean escaped_char = false; |
2d1675e4 | 2935 | const unsigned char *p2 = p; |
cf9c99bc | 2936 | re_wchar_t ch, c2; |
e318085a | 2937 | |
25fe55af | 2938 | if (p == pend) FREE_STACK_RETURN (REG_EBRACK); |
e318085a | 2939 | |
36595814 SM |
2940 | /* Don't translate yet. The range TRANSLATE(X..Y) cannot |
2941 | always be determined from TRANSLATE(X) and TRANSLATE(Y) | |
2942 | So the translation is done later in a loop. Example: | |
2943 | (let ((case-fold-search t)) (string-match "[A-_]" "A")) */ | |
25fe55af | 2944 | PATFETCH (c); |
e318085a | 2945 | |
25fe55af RS |
2946 | /* \ might escape characters inside [...] and [^...]. */ |
2947 | if ((syntax & RE_BACKSLASH_ESCAPE_IN_LISTS) && c == '\\') | |
2948 | { | |
2949 | if (p == pend) FREE_STACK_RETURN (REG_EESCAPE); | |
e318085a RS |
2950 | |
2951 | PATFETCH (c); | |
b18215fc | 2952 | escaped_char = true; |
25fe55af | 2953 | } |
b18215fc RS |
2954 | else |
2955 | { | |
7814e705 | 2956 | /* Could be the end of the bracket expression. If it's |
657fcfbd RS |
2957 | not (i.e., when the bracket expression is `[]' so |
2958 | far), the ']' character bit gets set way below. */ | |
2d1675e4 | 2959 | if (c == ']' && p2 != p1) |
657fcfbd | 2960 | break; |
25fe55af | 2961 | } |
b18215fc | 2962 | |
25fe55af RS |
2963 | /* See if we're at the beginning of a possible character |
2964 | class. */ | |
b18215fc | 2965 | |
2d1675e4 SM |
2966 | if (!escaped_char && |
2967 | syntax & RE_CHAR_CLASSES && c == '[' && *p == ':') | |
657fcfbd | 2968 | { |
7814e705 | 2969 | /* Leave room for the null. */ |
14473664 | 2970 | unsigned char str[CHAR_CLASS_MAX_LENGTH + 1]; |
ed0767d8 | 2971 | const unsigned char *class_beg; |
b18215fc | 2972 | |
25fe55af RS |
2973 | PATFETCH (c); |
2974 | c1 = 0; | |
ed0767d8 | 2975 | class_beg = p; |
b18215fc | 2976 | |
25fe55af RS |
2977 | /* If pattern is `[[:'. */ |
2978 | if (p == pend) FREE_STACK_RETURN (REG_EBRACK); | |
b18215fc | 2979 | |
25fe55af RS |
2980 | for (;;) |
2981 | { | |
14473664 SM |
2982 | PATFETCH (c); |
2983 | if ((c == ':' && *p == ']') || p == pend) | |
2984 | break; | |
2985 | if (c1 < CHAR_CLASS_MAX_LENGTH) | |
2986 | str[c1++] = c; | |
2987 | else | |
2988 | /* This is in any case an invalid class name. */ | |
2989 | str[0] = '\0'; | |
25fe55af RS |
2990 | } |
2991 | str[c1] = '\0'; | |
b18215fc RS |
2992 | |
2993 | /* If isn't a word bracketed by `[:' and `:]': | |
2994 | undo the ending character, the letters, and | |
2995 | leave the leading `:' and `[' (but set bits for | |
2996 | them). */ | |
25fe55af RS |
2997 | if (c == ':' && *p == ']') |
2998 | { | |
14473664 | 2999 | re_wctype_t cc; |
8f924df7 | 3000 | int limit; |
14473664 SM |
3001 | |
3002 | cc = re_wctype (str); | |
3003 | ||
3004 | if (cc == 0) | |
fa9a63c5 RM |
3005 | FREE_STACK_RETURN (REG_ECTYPE); |
3006 | ||
14473664 SM |
3007 | /* Throw away the ] at the end of the character |
3008 | class. */ | |
3009 | PATFETCH (c); | |
fa9a63c5 | 3010 | |
14473664 | 3011 | if (p == pend) FREE_STACK_RETURN (REG_EBRACK); |
fa9a63c5 | 3012 | |
cf9c99bc KH |
3013 | #ifndef emacs |
3014 | for (ch = 0; ch < (1 << BYTEWIDTH); ++ch) | |
8f924df7 KH |
3015 | if (re_iswctype (btowc (ch), cc)) |
3016 | { | |
3017 | c = TRANSLATE (ch); | |
ed00c2ac KH |
3018 | if (c < (1 << BYTEWIDTH)) |
3019 | SET_LIST_BIT (c); | |
8f924df7 | 3020 | } |
cf9c99bc KH |
3021 | #else /* emacs */ |
3022 | /* Most character classes in a multibyte match | |
3023 | just set a flag. Exceptions are is_blank, | |
3024 | is_digit, is_cntrl, and is_xdigit, since | |
3025 | they can only match ASCII characters. We | |
3026 | don't need to handle them for multibyte. | |
3027 | They are distinguished by a negative wctype. */ | |
96cc36cc | 3028 | |
254c06a8 SM |
3029 | /* Setup the gl_state object to its buffer-defined |
3030 | value. This hardcodes the buffer-global | |
3031 | syntax-table for ASCII chars, while the other chars | |
3032 | will obey syntax-table properties. It's not ideal, | |
3033 | but it's the way it's been done until now. */ | |
d48cd3f4 | 3034 | SETUP_BUFFER_SYNTAX_TABLE (); |
254c06a8 | 3035 | |
cf9c99bc | 3036 | for (ch = 0; ch < 256; ++ch) |
25fe55af | 3037 | { |
cf9c99bc KH |
3038 | c = RE_CHAR_TO_MULTIBYTE (ch); |
3039 | if (! CHAR_BYTE8_P (c) | |
3040 | && re_iswctype (c, cc)) | |
8f924df7 | 3041 | { |
cf9c99bc KH |
3042 | SET_LIST_BIT (ch); |
3043 | c1 = TRANSLATE (c); | |
3044 | if (c1 == c) | |
3045 | continue; | |
3046 | if (ASCII_CHAR_P (c1)) | |
3047 | SET_LIST_BIT (c1); | |
3048 | else if ((c1 = RE_CHAR_TO_UNIBYTE (c1)) >= 0) | |
3049 | SET_LIST_BIT (c1); | |
8f924df7 | 3050 | } |
25fe55af | 3051 | } |
cf9c99bc KH |
3052 | SET_RANGE_TABLE_WORK_AREA_BIT |
3053 | (range_table_work, re_wctype_to_bit (cc)); | |
3054 | #endif /* emacs */ | |
6224b623 SM |
3055 | /* In most cases the matching rule for char classes |
3056 | only uses the syntax table for multibyte chars, | |
3057 | so that the content of the syntax-table it is not | |
3058 | hardcoded in the range_table. SPACE and WORD are | |
3059 | the two exceptions. */ | |
3060 | if ((1 << cc) & ((1 << RECC_SPACE) | (1 << RECC_WORD))) | |
3061 | bufp->used_syntax = 1; | |
3062 | ||
b18215fc RS |
3063 | /* Repeat the loop. */ |
3064 | continue; | |
25fe55af RS |
3065 | } |
3066 | else | |
3067 | { | |
ed0767d8 SM |
3068 | /* Go back to right after the "[:". */ |
3069 | p = class_beg; | |
25fe55af | 3070 | SET_LIST_BIT ('['); |
b18215fc RS |
3071 | |
3072 | /* Because the `:' may starts the range, we | |
3073 | can't simply set bit and repeat the loop. | |
7814e705 | 3074 | Instead, just set it to C and handle below. */ |
b18215fc | 3075 | c = ':'; |
25fe55af RS |
3076 | } |
3077 | } | |
b18215fc RS |
3078 | |
3079 | if (p < pend && p[0] == '-' && p[1] != ']') | |
3080 | { | |
3081 | ||
3082 | /* Discard the `-'. */ | |
3083 | PATFETCH (c1); | |
3084 | ||
3085 | /* Fetch the character which ends the range. */ | |
3086 | PATFETCH (c1); | |
cf9c99bc KH |
3087 | #ifdef emacs |
3088 | if (CHAR_BYTE8_P (c1) | |
3089 | && ! ASCII_CHAR_P (c) && ! CHAR_BYTE8_P (c)) | |
3090 | /* Treat the range from a multibyte character to | |
3091 | raw-byte character as empty. */ | |
3092 | c = c1 + 1; | |
3093 | #endif /* emacs */ | |
e318085a | 3094 | } |
25fe55af | 3095 | else |
b18215fc RS |
3096 | /* Range from C to C. */ |
3097 | c1 = c; | |
3098 | ||
cf9c99bc | 3099 | if (c > c1) |
25fe55af | 3100 | { |
cf9c99bc KH |
3101 | if (syntax & RE_NO_EMPTY_RANGES) |
3102 | FREE_STACK_RETURN (REG_ERANGEX); | |
3103 | /* Else, repeat the loop. */ | |
bf216479 | 3104 | } |
6fdd04b0 | 3105 | else |
25fe55af | 3106 | { |
cf9c99bc KH |
3107 | #ifndef emacs |
3108 | /* Set the range into bitmap */ | |
8f924df7 | 3109 | for (; c <= c1; c++) |
b18215fc | 3110 | { |
cf9c99bc KH |
3111 | ch = TRANSLATE (c); |
3112 | if (ch < (1 << BYTEWIDTH)) | |
3113 | SET_LIST_BIT (ch); | |
3114 | } | |
3115 | #else /* emacs */ | |
3116 | if (c < 128) | |
3117 | { | |
3118 | ch = MIN (127, c1); | |
3119 | SETUP_ASCII_RANGE (range_table_work, c, ch); | |
3120 | c = ch + 1; | |
3121 | if (CHAR_BYTE8_P (c1)) | |
3122 | c = BYTE8_TO_CHAR (128); | |
3123 | } | |
3124 | if (c <= c1) | |
3125 | { | |
3126 | if (CHAR_BYTE8_P (c)) | |
3127 | { | |
3128 | c = CHAR_TO_BYTE8 (c); | |
3129 | c1 = CHAR_TO_BYTE8 (c1); | |
3130 | for (; c <= c1; c++) | |
3131 | SET_LIST_BIT (c); | |
3132 | } | |
3133 | else if (multibyte) | |
3134 | { | |
3135 | SETUP_MULTIBYTE_RANGE (range_table_work, c, c1); | |
3136 | } | |
3137 | else | |
3138 | { | |
3139 | SETUP_UNIBYTE_RANGE (range_table_work, c, c1); | |
3140 | } | |
e934739e | 3141 | } |
cf9c99bc | 3142 | #endif /* emacs */ |
25fe55af | 3143 | } |
e318085a RS |
3144 | } |
3145 | ||
25fe55af | 3146 | /* Discard any (non)matching list bytes that are all 0 at the |
7814e705 | 3147 | end of the map. Decrease the map-length byte too. */ |
25fe55af RS |
3148 | while ((int) b[-1] > 0 && b[b[-1] - 1] == 0) |
3149 | b[-1]--; | |
3150 | b += b[-1]; | |
fa9a63c5 | 3151 | |
96cc36cc RS |
3152 | /* Build real range table from work area. */ |
3153 | if (RANGE_TABLE_WORK_USED (range_table_work) | |
3154 | || RANGE_TABLE_WORK_BITS (range_table_work)) | |
b18215fc RS |
3155 | { |
3156 | int i; | |
3157 | int used = RANGE_TABLE_WORK_USED (range_table_work); | |
fa9a63c5 | 3158 | |
b18215fc | 3159 | /* Allocate space for COUNT + RANGE_TABLE. Needs two |
96cc36cc RS |
3160 | bytes for flags, two for COUNT, and three bytes for |
3161 | each character. */ | |
3162 | GET_BUFFER_SPACE (4 + used * 3); | |
fa9a63c5 | 3163 | |
b18215fc RS |
3164 | /* Indicate the existence of range table. */ |
3165 | laststart[1] |= 0x80; | |
fa9a63c5 | 3166 | |
96cc36cc RS |
3167 | /* Store the character class flag bits into the range table. |
3168 | If not in emacs, these flag bits are always 0. */ | |
3169 | *b++ = RANGE_TABLE_WORK_BITS (range_table_work) & 0xff; | |
3170 | *b++ = RANGE_TABLE_WORK_BITS (range_table_work) >> 8; | |
3171 | ||
b18215fc RS |
3172 | STORE_NUMBER_AND_INCR (b, used / 2); |
3173 | for (i = 0; i < used; i++) | |
3174 | STORE_CHARACTER_AND_INCR | |
3175 | (b, RANGE_TABLE_WORK_ELT (range_table_work, i)); | |
3176 | } | |
25fe55af RS |
3177 | } |
3178 | break; | |
fa9a63c5 RM |
3179 | |
3180 | ||
b18215fc | 3181 | case '(': |
25fe55af RS |
3182 | if (syntax & RE_NO_BK_PARENS) |
3183 | goto handle_open; | |
3184 | else | |
3185 | goto normal_char; | |
fa9a63c5 RM |
3186 | |
3187 | ||
25fe55af RS |
3188 | case ')': |
3189 | if (syntax & RE_NO_BK_PARENS) | |
3190 | goto handle_close; | |
3191 | else | |
3192 | goto normal_char; | |
e318085a RS |
3193 | |
3194 | ||
25fe55af RS |
3195 | case '\n': |
3196 | if (syntax & RE_NEWLINE_ALT) | |
3197 | goto handle_alt; | |
3198 | else | |
3199 | goto normal_char; | |
e318085a RS |
3200 | |
3201 | ||
b18215fc | 3202 | case '|': |
25fe55af RS |
3203 | if (syntax & RE_NO_BK_VBAR) |
3204 | goto handle_alt; | |
3205 | else | |
3206 | goto normal_char; | |
3207 | ||
3208 | ||
3209 | case '{': | |
3210 | if (syntax & RE_INTERVALS && syntax & RE_NO_BK_BRACES) | |
3211 | goto handle_interval; | |
3212 | else | |
3213 | goto normal_char; | |
3214 | ||
3215 | ||
3216 | case '\\': | |
3217 | if (p == pend) FREE_STACK_RETURN (REG_EESCAPE); | |
3218 | ||
3219 | /* Do not translate the character after the \, so that we can | |
3220 | distinguish, e.g., \B from \b, even if we normally would | |
3221 | translate, e.g., B to b. */ | |
36595814 | 3222 | PATFETCH (c); |
25fe55af RS |
3223 | |
3224 | switch (c) | |
3225 | { | |
3226 | case '(': | |
3227 | if (syntax & RE_NO_BK_PARENS) | |
3228 | goto normal_backslash; | |
3229 | ||
3230 | handle_open: | |
505bde11 SM |
3231 | { |
3232 | int shy = 0; | |
c69b0314 | 3233 | regnum_t regnum = 0; |
505bde11 SM |
3234 | if (p+1 < pend) |
3235 | { | |
3236 | /* Look for a special (?...) construct */ | |
ed0767d8 | 3237 | if ((syntax & RE_SHY_GROUPS) && *p == '?') |
505bde11 | 3238 | { |
ed0767d8 | 3239 | PATFETCH (c); /* Gobble up the '?'. */ |
c69b0314 | 3240 | while (!shy) |
505bde11 | 3241 | { |
c69b0314 SM |
3242 | PATFETCH (c); |
3243 | switch (c) | |
3244 | { | |
3245 | case ':': shy = 1; break; | |
3246 | case '0': | |
3247 | /* An explicitly specified regnum must start | |
3248 | with non-0. */ | |
3249 | if (regnum == 0) | |
3250 | FREE_STACK_RETURN (REG_BADPAT); | |
3251 | case '1': case '2': case '3': case '4': | |
3252 | case '5': case '6': case '7': case '8': case '9': | |
3253 | regnum = 10*regnum + (c - '0'); break; | |
3254 | default: | |
3255 | /* Only (?:...) is supported right now. */ | |
3256 | FREE_STACK_RETURN (REG_BADPAT); | |
3257 | } | |
505bde11 SM |
3258 | } |
3259 | } | |
505bde11 SM |
3260 | } |
3261 | ||
3262 | if (!shy) | |
c69b0314 SM |
3263 | regnum = ++bufp->re_nsub; |
3264 | else if (regnum) | |
3265 | { /* It's actually not shy, but explicitly numbered. */ | |
3266 | shy = 0; | |
3267 | if (regnum > bufp->re_nsub) | |
3268 | bufp->re_nsub = regnum; | |
3269 | else if (regnum > bufp->re_nsub | |
3270 | /* Ideally, we'd want to check that the specified | |
3271 | group can't have matched (i.e. all subgroups | |
3272 | using the same regnum are in other branches of | |
3273 | OR patterns), but we don't currently keep track | |
3274 | of enough info to do that easily. */ | |
3275 | || group_in_compile_stack (compile_stack, regnum)) | |
3276 | FREE_STACK_RETURN (REG_BADPAT); | |
505bde11 | 3277 | } |
c69b0314 SM |
3278 | else |
3279 | /* It's really shy. */ | |
3280 | regnum = - bufp->re_nsub; | |
25fe55af | 3281 | |
99633e97 SM |
3282 | if (COMPILE_STACK_FULL) |
3283 | { | |
3284 | RETALLOC (compile_stack.stack, compile_stack.size << 1, | |
3285 | compile_stack_elt_t); | |
3286 | if (compile_stack.stack == NULL) return REG_ESPACE; | |
25fe55af | 3287 | |
99633e97 SM |
3288 | compile_stack.size <<= 1; |
3289 | } | |
25fe55af | 3290 | |
99633e97 | 3291 | /* These are the values to restore when we hit end of this |
7814e705 | 3292 | group. They are all relative offsets, so that if the |
99633e97 SM |
3293 | whole pattern moves because of realloc, they will still |
3294 | be valid. */ | |
3295 | COMPILE_STACK_TOP.begalt_offset = begalt - bufp->buffer; | |
3296 | COMPILE_STACK_TOP.fixup_alt_jump | |
3297 | = fixup_alt_jump ? fixup_alt_jump - bufp->buffer + 1 : 0; | |
3298 | COMPILE_STACK_TOP.laststart_offset = b - bufp->buffer; | |
c69b0314 | 3299 | COMPILE_STACK_TOP.regnum = regnum; |
99633e97 | 3300 | |
c69b0314 SM |
3301 | /* Do not push a start_memory for groups beyond the last one |
3302 | we can represent in the compiled pattern. */ | |
3303 | if (regnum <= MAX_REGNUM && regnum > 0) | |
99633e97 SM |
3304 | BUF_PUSH_2 (start_memory, regnum); |
3305 | ||
3306 | compile_stack.avail++; | |
3307 | ||
3308 | fixup_alt_jump = 0; | |
3309 | laststart = 0; | |
3310 | begalt = b; | |
3311 | /* If we've reached MAX_REGNUM groups, then this open | |
3312 | won't actually generate any code, so we'll have to | |
3313 | clear pending_exact explicitly. */ | |
3314 | pending_exact = 0; | |
3315 | break; | |
505bde11 | 3316 | } |
25fe55af RS |
3317 | |
3318 | case ')': | |
3319 | if (syntax & RE_NO_BK_PARENS) goto normal_backslash; | |
3320 | ||
3321 | if (COMPILE_STACK_EMPTY) | |
505bde11 SM |
3322 | { |
3323 | if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD) | |
3324 | goto normal_backslash; | |
3325 | else | |
3326 | FREE_STACK_RETURN (REG_ERPAREN); | |
3327 | } | |
25fe55af RS |
3328 | |
3329 | handle_close: | |
505bde11 | 3330 | FIXUP_ALT_JUMP (); |
25fe55af RS |
3331 | |
3332 | /* See similar code for backslashed left paren above. */ | |
3333 | if (COMPILE_STACK_EMPTY) | |
505bde11 SM |
3334 | { |
3335 | if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD) | |
3336 | goto normal_char; | |
3337 | else | |
3338 | FREE_STACK_RETURN (REG_ERPAREN); | |
3339 | } | |
25fe55af RS |
3340 | |
3341 | /* Since we just checked for an empty stack above, this | |
3342 | ``can't happen''. */ | |
3343 | assert (compile_stack.avail != 0); | |
3344 | { | |
3345 | /* We don't just want to restore into `regnum', because | |
3346 | later groups should continue to be numbered higher, | |
7814e705 | 3347 | as in `(ab)c(de)' -- the second group is #2. */ |
c69b0314 | 3348 | regnum_t regnum; |
25fe55af RS |
3349 | |
3350 | compile_stack.avail--; | |
3351 | begalt = bufp->buffer + COMPILE_STACK_TOP.begalt_offset; | |
3352 | fixup_alt_jump | |
3353 | = COMPILE_STACK_TOP.fixup_alt_jump | |
3354 | ? bufp->buffer + COMPILE_STACK_TOP.fixup_alt_jump - 1 | |
3355 | : 0; | |
3356 | laststart = bufp->buffer + COMPILE_STACK_TOP.laststart_offset; | |
c69b0314 | 3357 | regnum = COMPILE_STACK_TOP.regnum; |
b18215fc RS |
3358 | /* If we've reached MAX_REGNUM groups, then this open |
3359 | won't actually generate any code, so we'll have to | |
3360 | clear pending_exact explicitly. */ | |
3361 | pending_exact = 0; | |
e318085a | 3362 | |
25fe55af | 3363 | /* We're at the end of the group, so now we know how many |
7814e705 | 3364 | groups were inside this one. */ |
c69b0314 SM |
3365 | if (regnum <= MAX_REGNUM && regnum > 0) |
3366 | BUF_PUSH_2 (stop_memory, regnum); | |
25fe55af RS |
3367 | } |
3368 | break; | |
3369 | ||
3370 | ||
3371 | case '|': /* `\|'. */ | |
3372 | if (syntax & RE_LIMITED_OPS || syntax & RE_NO_BK_VBAR) | |
3373 | goto normal_backslash; | |
3374 | handle_alt: | |
3375 | if (syntax & RE_LIMITED_OPS) | |
3376 | goto normal_char; | |
3377 | ||
3378 | /* Insert before the previous alternative a jump which | |
7814e705 | 3379 | jumps to this alternative if the former fails. */ |
25fe55af RS |
3380 | GET_BUFFER_SPACE (3); |
3381 | INSERT_JUMP (on_failure_jump, begalt, b + 6); | |
3382 | pending_exact = 0; | |
3383 | b += 3; | |
3384 | ||
3385 | /* The alternative before this one has a jump after it | |
3386 | which gets executed if it gets matched. Adjust that | |
3387 | jump so it will jump to this alternative's analogous | |
3388 | jump (put in below, which in turn will jump to the next | |
3389 | (if any) alternative's such jump, etc.). The last such | |
3390 | jump jumps to the correct final destination. A picture: | |
3391 | _____ _____ | |
3392 | | | | | | |
3393 | | v | v | |
3394 | a | b | c | |
3395 | ||
3396 | If we are at `b', then fixup_alt_jump right now points to a | |
3397 | three-byte space after `a'. We'll put in the jump, set | |
3398 | fixup_alt_jump to right after `b', and leave behind three | |
3399 | bytes which we'll fill in when we get to after `c'. */ | |
3400 | ||
505bde11 | 3401 | FIXUP_ALT_JUMP (); |
25fe55af RS |
3402 | |
3403 | /* Mark and leave space for a jump after this alternative, | |
3404 | to be filled in later either by next alternative or | |
3405 | when know we're at the end of a series of alternatives. */ | |
3406 | fixup_alt_jump = b; | |
3407 | GET_BUFFER_SPACE (3); | |
3408 | b += 3; | |
3409 | ||
3410 | laststart = 0; | |
3411 | begalt = b; | |
3412 | break; | |
3413 | ||
3414 | ||
3415 | case '{': | |
3416 | /* If \{ is a literal. */ | |
3417 | if (!(syntax & RE_INTERVALS) | |
3418 | /* If we're at `\{' and it's not the open-interval | |
3419 | operator. */ | |
4bb91c68 | 3420 | || (syntax & RE_NO_BK_BRACES)) |
25fe55af RS |
3421 | goto normal_backslash; |
3422 | ||
3423 | handle_interval: | |
3424 | { | |
3425 | /* If got here, then the syntax allows intervals. */ | |
3426 | ||
3427 | /* At least (most) this many matches must be made. */ | |
99633e97 | 3428 | int lower_bound = 0, upper_bound = -1; |
25fe55af | 3429 | |
ed0767d8 | 3430 | beg_interval = p; |
25fe55af | 3431 | |
25fe55af RS |
3432 | GET_UNSIGNED_NUMBER (lower_bound); |
3433 | ||
3434 | if (c == ',') | |
ed0767d8 | 3435 | GET_UNSIGNED_NUMBER (upper_bound); |
25fe55af RS |
3436 | else |
3437 | /* Interval such as `{1}' => match exactly once. */ | |
3438 | upper_bound = lower_bound; | |
3439 | ||
3440 | if (lower_bound < 0 || upper_bound > RE_DUP_MAX | |
ed0767d8 | 3441 | || (upper_bound >= 0 && lower_bound > upper_bound)) |
4bb91c68 | 3442 | FREE_STACK_RETURN (REG_BADBR); |
25fe55af RS |
3443 | |
3444 | if (!(syntax & RE_NO_BK_BRACES)) | |
3445 | { | |
4bb91c68 SM |
3446 | if (c != '\\') |
3447 | FREE_STACK_RETURN (REG_BADBR); | |
c72b0edd SM |
3448 | if (p == pend) |
3449 | FREE_STACK_RETURN (REG_EESCAPE); | |
25fe55af RS |
3450 | PATFETCH (c); |
3451 | } | |
3452 | ||
3453 | if (c != '}') | |
4bb91c68 | 3454 | FREE_STACK_RETURN (REG_BADBR); |
25fe55af RS |
3455 | |
3456 | /* We just parsed a valid interval. */ | |
3457 | ||
3458 | /* If it's invalid to have no preceding re. */ | |
3459 | if (!laststart) | |
3460 | { | |
3461 | if (syntax & RE_CONTEXT_INVALID_OPS) | |
3462 | FREE_STACK_RETURN (REG_BADRPT); | |
3463 | else if (syntax & RE_CONTEXT_INDEP_OPS) | |
3464 | laststart = b; | |
3465 | else | |
3466 | goto unfetch_interval; | |
3467 | } | |
3468 | ||
6df42991 SM |
3469 | if (upper_bound == 0) |
3470 | /* If the upper bound is zero, just drop the sub pattern | |
3471 | altogether. */ | |
3472 | b = laststart; | |
3473 | else if (lower_bound == 1 && upper_bound == 1) | |
3474 | /* Just match it once: nothing to do here. */ | |
3475 | ; | |
3476 | ||
3477 | /* Otherwise, we have a nontrivial interval. When | |
3478 | we're all done, the pattern will look like: | |
3479 | set_number_at <jump count> <upper bound> | |
3480 | set_number_at <succeed_n count> <lower bound> | |
3481 | succeed_n <after jump addr> <succeed_n count> | |
3482 | <body of loop> | |
3483 | jump_n <succeed_n addr> <jump count> | |
3484 | (The upper bound and `jump_n' are omitted if | |
3485 | `upper_bound' is 1, though.) */ | |
3486 | else | |
3487 | { /* If the upper bound is > 1, we need to insert | |
3488 | more at the end of the loop. */ | |
3489 | unsigned int nbytes = (upper_bound < 0 ? 3 | |
3490 | : upper_bound > 1 ? 5 : 0); | |
3491 | unsigned int startoffset = 0; | |
3492 | ||
3493 | GET_BUFFER_SPACE (20); /* We might use less. */ | |
3494 | ||
3495 | if (lower_bound == 0) | |
3496 | { | |
3497 | /* A succeed_n that starts with 0 is really a | |
3498 | a simple on_failure_jump_loop. */ | |
3499 | INSERT_JUMP (on_failure_jump_loop, laststart, | |
3500 | b + 3 + nbytes); | |
3501 | b += 3; | |
3502 | } | |
3503 | else | |
3504 | { | |
3505 | /* Initialize lower bound of the `succeed_n', even | |
3506 | though it will be set during matching by its | |
3507 | attendant `set_number_at' (inserted next), | |
3508 | because `re_compile_fastmap' needs to know. | |
3509 | Jump to the `jump_n' we might insert below. */ | |
3510 | INSERT_JUMP2 (succeed_n, laststart, | |
3511 | b + 5 + nbytes, | |
3512 | lower_bound); | |
3513 | b += 5; | |
3514 | ||
3515 | /* Code to initialize the lower bound. Insert | |
7814e705 | 3516 | before the `succeed_n'. The `5' is the last two |
6df42991 SM |
3517 | bytes of this `set_number_at', plus 3 bytes of |
3518 | the following `succeed_n'. */ | |
3519 | insert_op2 (set_number_at, laststart, 5, lower_bound, b); | |
3520 | b += 5; | |
3521 | startoffset += 5; | |
3522 | } | |
3523 | ||
3524 | if (upper_bound < 0) | |
3525 | { | |
3526 | /* A negative upper bound stands for infinity, | |
3527 | in which case it degenerates to a plain jump. */ | |
3528 | STORE_JUMP (jump, b, laststart + startoffset); | |
3529 | b += 3; | |
3530 | } | |
3531 | else if (upper_bound > 1) | |
3532 | { /* More than one repetition is allowed, so | |
3533 | append a backward jump to the `succeed_n' | |
3534 | that starts this interval. | |
3535 | ||
3536 | When we've reached this during matching, | |
3537 | we'll have matched the interval once, so | |
3538 | jump back only `upper_bound - 1' times. */ | |
3539 | STORE_JUMP2 (jump_n, b, laststart + startoffset, | |
3540 | upper_bound - 1); | |
3541 | b += 5; | |
3542 | ||
3543 | /* The location we want to set is the second | |
3544 | parameter of the `jump_n'; that is `b-2' as | |
3545 | an absolute address. `laststart' will be | |
3546 | the `set_number_at' we're about to insert; | |
3547 | `laststart+3' the number to set, the source | |
3548 | for the relative address. But we are | |
3549 | inserting into the middle of the pattern -- | |
3550 | so everything is getting moved up by 5. | |
3551 | Conclusion: (b - 2) - (laststart + 3) + 5, | |
3552 | i.e., b - laststart. | |
3553 | ||
3554 | We insert this at the beginning of the loop | |
3555 | so that if we fail during matching, we'll | |
3556 | reinitialize the bounds. */ | |
3557 | insert_op2 (set_number_at, laststart, b - laststart, | |
3558 | upper_bound - 1, b); | |
3559 | b += 5; | |
3560 | } | |
3561 | } | |
25fe55af RS |
3562 | pending_exact = 0; |
3563 | beg_interval = NULL; | |
3564 | } | |
3565 | break; | |
3566 | ||
3567 | unfetch_interval: | |
3568 | /* If an invalid interval, match the characters as literals. */ | |
3569 | assert (beg_interval); | |
3570 | p = beg_interval; | |
3571 | beg_interval = NULL; | |
3572 | ||
3573 | /* normal_char and normal_backslash need `c'. */ | |
ed0767d8 | 3574 | c = '{'; |
25fe55af RS |
3575 | |
3576 | if (!(syntax & RE_NO_BK_BRACES)) | |
3577 | { | |
ed0767d8 SM |
3578 | assert (p > pattern && p[-1] == '\\'); |
3579 | goto normal_backslash; | |
25fe55af | 3580 | } |
ed0767d8 SM |
3581 | else |
3582 | goto normal_char; | |
e318085a | 3583 | |
b18215fc | 3584 | #ifdef emacs |
25fe55af | 3585 | /* There is no way to specify the before_dot and after_dot |
7814e705 | 3586 | operators. rms says this is ok. --karl */ |
25fe55af RS |
3587 | case '=': |
3588 | BUF_PUSH (at_dot); | |
3589 | break; | |
3590 | ||
3591 | case 's': | |
3592 | laststart = b; | |
3593 | PATFETCH (c); | |
3594 | BUF_PUSH_2 (syntaxspec, syntax_spec_code[c]); | |
3595 | break; | |
3596 | ||
3597 | case 'S': | |
3598 | laststart = b; | |
3599 | PATFETCH (c); | |
3600 | BUF_PUSH_2 (notsyntaxspec, syntax_spec_code[c]); | |
3601 | break; | |
b18215fc RS |
3602 | |
3603 | case 'c': | |
3604 | laststart = b; | |
36595814 | 3605 | PATFETCH (c); |
b18215fc RS |
3606 | BUF_PUSH_2 (categoryspec, c); |
3607 | break; | |
e318085a | 3608 | |
b18215fc RS |
3609 | case 'C': |
3610 | laststart = b; | |
36595814 | 3611 | PATFETCH (c); |
b18215fc RS |
3612 | BUF_PUSH_2 (notcategoryspec, c); |
3613 | break; | |
3614 | #endif /* emacs */ | |
e318085a | 3615 | |
e318085a | 3616 | |
25fe55af | 3617 | case 'w': |
4bb91c68 SM |
3618 | if (syntax & RE_NO_GNU_OPS) |
3619 | goto normal_char; | |
25fe55af | 3620 | laststart = b; |
1fb352e0 | 3621 | BUF_PUSH_2 (syntaxspec, Sword); |
25fe55af | 3622 | break; |
e318085a | 3623 | |
e318085a | 3624 | |
25fe55af | 3625 | case 'W': |
4bb91c68 SM |
3626 | if (syntax & RE_NO_GNU_OPS) |
3627 | goto normal_char; | |
25fe55af | 3628 | laststart = b; |
1fb352e0 | 3629 | BUF_PUSH_2 (notsyntaxspec, Sword); |
25fe55af | 3630 | break; |
e318085a RS |
3631 | |
3632 | ||
25fe55af | 3633 | case '<': |
4bb91c68 SM |
3634 | if (syntax & RE_NO_GNU_OPS) |
3635 | goto normal_char; | |
25fe55af RS |
3636 | BUF_PUSH (wordbeg); |
3637 | break; | |
e318085a | 3638 | |
25fe55af | 3639 | case '>': |
4bb91c68 SM |
3640 | if (syntax & RE_NO_GNU_OPS) |
3641 | goto normal_char; | |
25fe55af RS |
3642 | BUF_PUSH (wordend); |
3643 | break; | |
e318085a | 3644 | |
669fa600 SM |
3645 | case '_': |
3646 | if (syntax & RE_NO_GNU_OPS) | |
3647 | goto normal_char; | |
3648 | laststart = b; | |
3649 | PATFETCH (c); | |
3650 | if (c == '<') | |
3651 | BUF_PUSH (symbeg); | |
3652 | else if (c == '>') | |
3653 | BUF_PUSH (symend); | |
3654 | else | |
3655 | FREE_STACK_RETURN (REG_BADPAT); | |
3656 | break; | |
3657 | ||
25fe55af | 3658 | case 'b': |
4bb91c68 SM |
3659 | if (syntax & RE_NO_GNU_OPS) |
3660 | goto normal_char; | |
25fe55af RS |
3661 | BUF_PUSH (wordbound); |
3662 | break; | |
e318085a | 3663 | |
25fe55af | 3664 | case 'B': |
4bb91c68 SM |
3665 | if (syntax & RE_NO_GNU_OPS) |
3666 | goto normal_char; | |
25fe55af RS |
3667 | BUF_PUSH (notwordbound); |
3668 | break; | |
fa9a63c5 | 3669 | |
25fe55af | 3670 | case '`': |
4bb91c68 SM |
3671 | if (syntax & RE_NO_GNU_OPS) |
3672 | goto normal_char; | |
25fe55af RS |
3673 | BUF_PUSH (begbuf); |
3674 | break; | |
e318085a | 3675 | |
25fe55af | 3676 | case '\'': |
4bb91c68 SM |
3677 | if (syntax & RE_NO_GNU_OPS) |
3678 | goto normal_char; | |
25fe55af RS |
3679 | BUF_PUSH (endbuf); |
3680 | break; | |
e318085a | 3681 | |
25fe55af RS |
3682 | case '1': case '2': case '3': case '4': case '5': |
3683 | case '6': case '7': case '8': case '9': | |
0cdd06f8 SM |
3684 | { |
3685 | regnum_t reg; | |
e318085a | 3686 | |
0cdd06f8 SM |
3687 | if (syntax & RE_NO_BK_REFS) |
3688 | goto normal_backslash; | |
e318085a | 3689 | |
0cdd06f8 | 3690 | reg = c - '0'; |
e318085a | 3691 | |
c69b0314 SM |
3692 | if (reg > bufp->re_nsub || reg < 1 |
3693 | /* Can't back reference to a subexp before its end. */ | |
3694 | || group_in_compile_stack (compile_stack, reg)) | |
0cdd06f8 | 3695 | FREE_STACK_RETURN (REG_ESUBREG); |
e318085a | 3696 | |
0cdd06f8 SM |
3697 | laststart = b; |
3698 | BUF_PUSH_2 (duplicate, reg); | |
3699 | } | |
25fe55af | 3700 | break; |
e318085a | 3701 | |
e318085a | 3702 | |
25fe55af RS |
3703 | case '+': |
3704 | case '?': | |
3705 | if (syntax & RE_BK_PLUS_QM) | |
3706 | goto handle_plus; | |
3707 | else | |
3708 | goto normal_backslash; | |
3709 | ||
3710 | default: | |
3711 | normal_backslash: | |
3712 | /* You might think it would be useful for \ to mean | |
3713 | not to translate; but if we don't translate it | |
4bb91c68 | 3714 | it will never match anything. */ |
25fe55af RS |
3715 | goto normal_char; |
3716 | } | |
3717 | break; | |
fa9a63c5 RM |
3718 | |
3719 | ||
3720 | default: | |
25fe55af | 3721 | /* Expects the character in `c'. */ |
fa9a63c5 | 3722 | normal_char: |
36595814 | 3723 | /* If no exactn currently being built. */ |
25fe55af | 3724 | if (!pending_exact |
fa9a63c5 | 3725 | |
25fe55af RS |
3726 | /* If last exactn not at current position. */ |
3727 | || pending_exact + *pending_exact + 1 != b | |
5e69f11e | 3728 | |
25fe55af | 3729 | /* We have only one byte following the exactn for the count. */ |
2d1675e4 | 3730 | || *pending_exact >= (1 << BYTEWIDTH) - MAX_MULTIBYTE_LENGTH |
fa9a63c5 | 3731 | |
7814e705 | 3732 | /* If followed by a repetition operator. */ |
9d99031f | 3733 | || (p != pend && (*p == '*' || *p == '^')) |
fa9a63c5 | 3734 | || ((syntax & RE_BK_PLUS_QM) |
9d99031f RS |
3735 | ? p + 1 < pend && *p == '\\' && (p[1] == '+' || p[1] == '?') |
3736 | : p != pend && (*p == '+' || *p == '?')) | |
fa9a63c5 | 3737 | || ((syntax & RE_INTERVALS) |
25fe55af | 3738 | && ((syntax & RE_NO_BK_BRACES) |
9d99031f RS |
3739 | ? p != pend && *p == '{' |
3740 | : p + 1 < pend && p[0] == '\\' && p[1] == '{'))) | |
fa9a63c5 RM |
3741 | { |
3742 | /* Start building a new exactn. */ | |
5e69f11e | 3743 | |
25fe55af | 3744 | laststart = b; |
fa9a63c5 RM |
3745 | |
3746 | BUF_PUSH_2 (exactn, 0); | |
3747 | pending_exact = b - 1; | |
25fe55af | 3748 | } |
5e69f11e | 3749 | |
2d1675e4 SM |
3750 | GET_BUFFER_SPACE (MAX_MULTIBYTE_LENGTH); |
3751 | { | |
e0277a47 KH |
3752 | int len; |
3753 | ||
cf9c99bc | 3754 | if (multibyte) |
6fdd04b0 | 3755 | { |
cf9c99bc | 3756 | c = TRANSLATE (c); |
6fdd04b0 KH |
3757 | len = CHAR_STRING (c, b); |
3758 | b += len; | |
3759 | } | |
e0277a47 | 3760 | else |
6fdd04b0 | 3761 | { |
cf9c99bc KH |
3762 | c1 = RE_CHAR_TO_MULTIBYTE (c); |
3763 | if (! CHAR_BYTE8_P (c1)) | |
3764 | { | |
3765 | re_wchar_t c2 = TRANSLATE (c1); | |
3766 | ||
3767 | if (c1 != c2 && (c1 = RE_CHAR_TO_UNIBYTE (c2)) >= 0) | |
3768 | c = c1; | |
409f2919 | 3769 | } |
6fdd04b0 KH |
3770 | *b++ = c; |
3771 | len = 1; | |
3772 | } | |
2d1675e4 SM |
3773 | (*pending_exact) += len; |
3774 | } | |
3775 | ||
fa9a63c5 | 3776 | break; |
25fe55af | 3777 | } /* switch (c) */ |
fa9a63c5 RM |
3778 | } /* while p != pend */ |
3779 | ||
5e69f11e | 3780 | |
fa9a63c5 | 3781 | /* Through the pattern now. */ |
5e69f11e | 3782 | |
505bde11 | 3783 | FIXUP_ALT_JUMP (); |
fa9a63c5 | 3784 | |
5e69f11e | 3785 | if (!COMPILE_STACK_EMPTY) |
fa9a63c5 RM |
3786 | FREE_STACK_RETURN (REG_EPAREN); |
3787 | ||
3788 | /* If we don't want backtracking, force success | |
3789 | the first time we reach the end of the compiled pattern. */ | |
3790 | if (syntax & RE_NO_POSIX_BACKTRACKING) | |
3791 | BUF_PUSH (succeed); | |
3792 | ||
fa9a63c5 RM |
3793 | /* We have succeeded; set the length of the buffer. */ |
3794 | bufp->used = b - bufp->buffer; | |
3795 | ||
3796 | #ifdef DEBUG | |
99633e97 | 3797 | if (debug > 0) |
fa9a63c5 | 3798 | { |
505bde11 | 3799 | re_compile_fastmap (bufp); |
fa9a63c5 RM |
3800 | DEBUG_PRINT1 ("\nCompiled pattern: \n"); |
3801 | print_compiled_pattern (bufp); | |
3802 | } | |
99633e97 | 3803 | debug--; |
fa9a63c5 RM |
3804 | #endif /* DEBUG */ |
3805 | ||
3806 | #ifndef MATCH_MAY_ALLOCATE | |
3807 | /* Initialize the failure stack to the largest possible stack. This | |
3808 | isn't necessary unless we're trying to avoid calling alloca in | |
3809 | the search and match routines. */ | |
3810 | { | |
3811 | int num_regs = bufp->re_nsub + 1; | |
3812 | ||
320a2a73 | 3813 | if (fail_stack.size < re_max_failures * TYPICAL_FAILURE_SIZE) |
fa9a63c5 | 3814 | { |
a26f4ccd | 3815 | fail_stack.size = re_max_failures * TYPICAL_FAILURE_SIZE; |
fa9a63c5 | 3816 | |
fa9a63c5 RM |
3817 | if (! fail_stack.stack) |
3818 | fail_stack.stack | |
5e69f11e | 3819 | = (fail_stack_elt_t *) malloc (fail_stack.size |
fa9a63c5 RM |
3820 | * sizeof (fail_stack_elt_t)); |
3821 | else | |
3822 | fail_stack.stack | |
3823 | = (fail_stack_elt_t *) realloc (fail_stack.stack, | |
3824 | (fail_stack.size | |
3825 | * sizeof (fail_stack_elt_t))); | |
fa9a63c5 RM |
3826 | } |
3827 | ||
3828 | regex_grow_registers (num_regs); | |
3829 | } | |
3830 | #endif /* not MATCH_MAY_ALLOCATE */ | |
3831 | ||
839966f3 | 3832 | FREE_STACK_RETURN (REG_NOERROR); |
fa9a63c5 RM |
3833 | } /* regex_compile */ |
3834 | \f | |
3835 | /* Subroutines for `regex_compile'. */ | |
3836 | ||
7814e705 | 3837 | /* Store OP at LOC followed by two-byte integer parameter ARG. */ |
fa9a63c5 RM |
3838 | |
3839 | static void | |
971de7fb | 3840 | store_op1 (re_opcode_t op, unsigned char *loc, int arg) |
fa9a63c5 RM |
3841 | { |
3842 | *loc = (unsigned char) op; | |
3843 | STORE_NUMBER (loc + 1, arg); | |
3844 | } | |
3845 | ||
3846 | ||
3847 | /* Like `store_op1', but for two two-byte parameters ARG1 and ARG2. */ | |
3848 | ||
3849 | static void | |
971de7fb | 3850 | store_op2 (re_opcode_t op, unsigned char *loc, int arg1, int arg2) |
fa9a63c5 RM |
3851 | { |
3852 | *loc = (unsigned char) op; | |
3853 | STORE_NUMBER (loc + 1, arg1); | |
3854 | STORE_NUMBER (loc + 3, arg2); | |
3855 | } | |
3856 | ||
3857 | ||
3858 | /* Copy the bytes from LOC to END to open up three bytes of space at LOC | |
3859 | for OP followed by two-byte integer parameter ARG. */ | |
3860 | ||
3861 | static void | |
971de7fb | 3862 | insert_op1 (re_opcode_t op, unsigned char *loc, int arg, unsigned char *end) |
fa9a63c5 RM |
3863 | { |
3864 | register unsigned char *pfrom = end; | |
3865 | register unsigned char *pto = end + 3; | |
3866 | ||
3867 | while (pfrom != loc) | |
3868 | *--pto = *--pfrom; | |
5e69f11e | 3869 | |
fa9a63c5 RM |
3870 | store_op1 (op, loc, arg); |
3871 | } | |
3872 | ||
3873 | ||
3874 | /* Like `insert_op1', but for two two-byte parameters ARG1 and ARG2. */ | |
3875 | ||
3876 | static void | |
971de7fb | 3877 | insert_op2 (re_opcode_t op, unsigned char *loc, int arg1, int arg2, unsigned char *end) |
fa9a63c5 RM |
3878 | { |
3879 | register unsigned char *pfrom = end; | |
3880 | register unsigned char *pto = end + 5; | |
3881 | ||
3882 | while (pfrom != loc) | |
3883 | *--pto = *--pfrom; | |
5e69f11e | 3884 | |
fa9a63c5 RM |
3885 | store_op2 (op, loc, arg1, arg2); |
3886 | } | |
3887 | ||
3888 | ||
3889 | /* P points to just after a ^ in PATTERN. Return true if that ^ comes | |
3890 | after an alternative or a begin-subexpression. We assume there is at | |
3891 | least one character before the ^. */ | |
3892 | ||
3893 | static boolean | |
971de7fb | 3894 | at_begline_loc_p (const re_char *pattern, const re_char *p, reg_syntax_t syntax) |
fa9a63c5 | 3895 | { |
01618498 | 3896 | re_char *prev = p - 2; |
fa9a63c5 | 3897 | boolean prev_prev_backslash = prev > pattern && prev[-1] == '\\'; |
5e69f11e | 3898 | |
fa9a63c5 RM |
3899 | return |
3900 | /* After a subexpression? */ | |
3901 | (*prev == '(' && (syntax & RE_NO_BK_PARENS || prev_prev_backslash)) | |
25fe55af | 3902 | /* After an alternative? */ |
d2af47df SM |
3903 | || (*prev == '|' && (syntax & RE_NO_BK_VBAR || prev_prev_backslash)) |
3904 | /* After a shy subexpression? */ | |
3905 | || ((syntax & RE_SHY_GROUPS) && prev - 2 >= pattern | |
3906 | && prev[-1] == '?' && prev[-2] == '(' | |
3907 | && (syntax & RE_NO_BK_PARENS | |
3908 | || (prev - 3 >= pattern && prev[-3] == '\\'))); | |
fa9a63c5 RM |
3909 | } |
3910 | ||
3911 | ||
3912 | /* The dual of at_begline_loc_p. This one is for $. We assume there is | |
3913 | at least one character after the $, i.e., `P < PEND'. */ | |
3914 | ||
3915 | static boolean | |
971de7fb | 3916 | at_endline_loc_p (const re_char *p, const re_char *pend, reg_syntax_t syntax) |
fa9a63c5 | 3917 | { |
01618498 | 3918 | re_char *next = p; |
fa9a63c5 | 3919 | boolean next_backslash = *next == '\\'; |
01618498 | 3920 | re_char *next_next = p + 1 < pend ? p + 1 : 0; |
5e69f11e | 3921 | |
fa9a63c5 RM |
3922 | return |
3923 | /* Before a subexpression? */ | |
3924 | (syntax & RE_NO_BK_PARENS ? *next == ')' | |
25fe55af | 3925 | : next_backslash && next_next && *next_next == ')') |
fa9a63c5 RM |
3926 | /* Before an alternative? */ |
3927 | || (syntax & RE_NO_BK_VBAR ? *next == '|' | |
25fe55af | 3928 | : next_backslash && next_next && *next_next == '|'); |
fa9a63c5 RM |
3929 | } |
3930 | ||
3931 | ||
5e69f11e | 3932 | /* Returns true if REGNUM is in one of COMPILE_STACK's elements and |
fa9a63c5 RM |
3933 | false if it's not. */ |
3934 | ||
3935 | static boolean | |
971de7fb | 3936 | group_in_compile_stack (compile_stack_type compile_stack, regnum_t regnum) |
fa9a63c5 RM |
3937 | { |
3938 | int this_element; | |
3939 | ||
5e69f11e RM |
3940 | for (this_element = compile_stack.avail - 1; |
3941 | this_element >= 0; | |
fa9a63c5 RM |
3942 | this_element--) |
3943 | if (compile_stack.stack[this_element].regnum == regnum) | |
3944 | return true; | |
3945 | ||
3946 | return false; | |
3947 | } | |
fa9a63c5 | 3948 | \f |
f6a3f532 SM |
3949 | /* analyse_first. |
3950 | If fastmap is non-NULL, go through the pattern and fill fastmap | |
3951 | with all the possible leading chars. If fastmap is NULL, don't | |
3952 | bother filling it up (obviously) and only return whether the | |
3953 | pattern could potentially match the empty string. | |
3954 | ||
3955 | Return 1 if p..pend might match the empty string. | |
3956 | Return 0 if p..pend matches at least one char. | |
01618498 | 3957 | Return -1 if fastmap was not updated accurately. */ |
f6a3f532 SM |
3958 | |
3959 | static int | |
438105ed | 3960 | analyse_first (const re_char *p, const re_char *pend, char *fastmap, const int multibyte) |
fa9a63c5 | 3961 | { |
505bde11 | 3962 | int j, k; |
1fb352e0 | 3963 | boolean not; |
fa9a63c5 | 3964 | |
b18215fc | 3965 | /* If all elements for base leading-codes in fastmap is set, this |
7814e705 | 3966 | flag is set true. */ |
b18215fc RS |
3967 | boolean match_any_multibyte_characters = false; |
3968 | ||
f6a3f532 | 3969 | assert (p); |
5e69f11e | 3970 | |
505bde11 SM |
3971 | /* The loop below works as follows: |
3972 | - It has a working-list kept in the PATTERN_STACK and which basically | |
3973 | starts by only containing a pointer to the first operation. | |
3974 | - If the opcode we're looking at is a match against some set of | |
3975 | chars, then we add those chars to the fastmap and go on to the | |
3976 | next work element from the worklist (done via `break'). | |
3977 | - If the opcode is a control operator on the other hand, we either | |
3978 | ignore it (if it's meaningless at this point, such as `start_memory') | |
3979 | or execute it (if it's a jump). If the jump has several destinations | |
3980 | (i.e. `on_failure_jump'), then we push the other destination onto the | |
3981 | worklist. | |
3982 | We guarantee termination by ignoring backward jumps (more or less), | |
3983 | so that `p' is monotonically increasing. More to the point, we | |
3984 | never set `p' (or push) anything `<= p1'. */ | |
3985 | ||
01618498 | 3986 | while (p < pend) |
fa9a63c5 | 3987 | { |
505bde11 SM |
3988 | /* `p1' is used as a marker of how far back a `on_failure_jump' |
3989 | can go without being ignored. It is normally equal to `p' | |
3990 | (which prevents any backward `on_failure_jump') except right | |
3991 | after a plain `jump', to allow patterns such as: | |
3992 | 0: jump 10 | |
3993 | 3..9: <body> | |
3994 | 10: on_failure_jump 3 | |
3995 | as used for the *? operator. */ | |
01618498 | 3996 | re_char *p1 = p; |
5e69f11e | 3997 | |
fa9a63c5 RM |
3998 | switch (SWITCH_ENUM_CAST ((re_opcode_t) *p++)) |
3999 | { | |
f6a3f532 | 4000 | case succeed: |
01618498 | 4001 | return 1; |
f6a3f532 | 4002 | continue; |
fa9a63c5 | 4003 | |
fa9a63c5 | 4004 | case duplicate: |
505bde11 SM |
4005 | /* If the first character has to match a backreference, that means |
4006 | that the group was empty (since it already matched). Since this | |
4007 | is the only case that interests us here, we can assume that the | |
4008 | backreference must match the empty string. */ | |
4009 | p++; | |
4010 | continue; | |
fa9a63c5 RM |
4011 | |
4012 | ||
4013 | /* Following are the cases which match a character. These end | |
7814e705 | 4014 | with `break'. */ |
fa9a63c5 RM |
4015 | |
4016 | case exactn: | |
e0277a47 | 4017 | if (fastmap) |
cf9c99bc KH |
4018 | { |
4019 | /* If multibyte is nonzero, the first byte of each | |
4020 | character is an ASCII or a leading code. Otherwise, | |
4021 | each byte is a character. Thus, this works in both | |
4022 | cases. */ | |
4023 | fastmap[p[1]] = 1; | |
4024 | if (! multibyte) | |
4025 | { | |
4026 | /* For the case of matching this unibyte regex | |
4027 | against multibyte, we must set a leading code of | |
4028 | the corresponding multibyte character. */ | |
4029 | int c = RE_CHAR_TO_MULTIBYTE (p[1]); | |
4030 | ||
86e893e3 | 4031 | fastmap[CHAR_LEADING_CODE (c)] = 1; |
cf9c99bc KH |
4032 | } |
4033 | } | |
fa9a63c5 RM |
4034 | break; |
4035 | ||
4036 | ||
1fb352e0 SM |
4037 | case anychar: |
4038 | /* We could put all the chars except for \n (and maybe \0) | |
4039 | but we don't bother since it is generally not worth it. */ | |
f6a3f532 | 4040 | if (!fastmap) break; |
01618498 | 4041 | return -1; |
fa9a63c5 RM |
4042 | |
4043 | ||
b18215fc | 4044 | case charset_not: |
1fb352e0 | 4045 | if (!fastmap) break; |
bf216479 KH |
4046 | { |
4047 | /* Chars beyond end of bitmap are possible matches. */ | |
bf216479 | 4048 | for (j = CHARSET_BITMAP_SIZE (&p[-1]) * BYTEWIDTH; |
cf9c99bc | 4049 | j < (1 << BYTEWIDTH); j++) |
bf216479 KH |
4050 | fastmap[j] = 1; |
4051 | } | |
4052 | ||
1fb352e0 SM |
4053 | /* Fallthrough */ |
4054 | case charset: | |
4055 | if (!fastmap) break; | |
4056 | not = (re_opcode_t) *(p - 1) == charset_not; | |
b18215fc RS |
4057 | for (j = CHARSET_BITMAP_SIZE (&p[-1]) * BYTEWIDTH - 1, p++; |
4058 | j >= 0; j--) | |
1fb352e0 | 4059 | if (!!(p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH))) ^ not) |
49da453b | 4060 | fastmap[j] = 1; |
b18215fc | 4061 | |
6482db2e KH |
4062 | #ifdef emacs |
4063 | if (/* Any leading code can possibly start a character | |
1fb352e0 | 4064 | which doesn't match the specified set of characters. */ |
6482db2e | 4065 | not |
409f2919 | 4066 | || |
6482db2e KH |
4067 | /* If we can match a character class, we can match any |
4068 | multibyte characters. */ | |
4069 | (CHARSET_RANGE_TABLE_EXISTS_P (&p[-2]) | |
4070 | && CHARSET_RANGE_TABLE_BITS (&p[-2]) != 0)) | |
4071 | ||
b18215fc | 4072 | { |
b18215fc RS |
4073 | if (match_any_multibyte_characters == false) |
4074 | { | |
6482db2e KH |
4075 | for (j = MIN_MULTIBYTE_LEADING_CODE; |
4076 | j <= MAX_MULTIBYTE_LEADING_CODE; j++) | |
6fdd04b0 | 4077 | fastmap[j] = 1; |
b18215fc RS |
4078 | match_any_multibyte_characters = true; |
4079 | } | |
4080 | } | |
b18215fc | 4081 | |
1fb352e0 SM |
4082 | else if (!not && CHARSET_RANGE_TABLE_EXISTS_P (&p[-2]) |
4083 | && match_any_multibyte_characters == false) | |
4084 | { | |
bf216479 | 4085 | /* Set fastmap[I] to 1 where I is a leading code of each |
51e4f4a8 | 4086 | multibyte character in the range table. */ |
1fb352e0 | 4087 | int c, count; |
bf216479 | 4088 | unsigned char lc1, lc2; |
b18215fc | 4089 | |
1fb352e0 | 4090 | /* Make P points the range table. `+ 2' is to skip flag |
0b32bf0e | 4091 | bits for a character class. */ |
1fb352e0 | 4092 | p += CHARSET_BITMAP_SIZE (&p[-2]) + 2; |
b18215fc | 4093 | |
1fb352e0 SM |
4094 | /* Extract the number of ranges in range table into COUNT. */ |
4095 | EXTRACT_NUMBER_AND_INCR (count, p); | |
cf9c99bc | 4096 | for (; count > 0; count--, p += 3) |
1fb352e0 | 4097 | { |
9117d724 KH |
4098 | /* Extract the start and end of each range. */ |
4099 | EXTRACT_CHARACTER (c, p); | |
bf216479 | 4100 | lc1 = CHAR_LEADING_CODE (c); |
9117d724 | 4101 | p += 3; |
1fb352e0 | 4102 | EXTRACT_CHARACTER (c, p); |
bf216479 KH |
4103 | lc2 = CHAR_LEADING_CODE (c); |
4104 | for (j = lc1; j <= lc2; j++) | |
9117d724 | 4105 | fastmap[j] = 1; |
1fb352e0 SM |
4106 | } |
4107 | } | |
6482db2e | 4108 | #endif |
b18215fc RS |
4109 | break; |
4110 | ||
1fb352e0 SM |
4111 | case syntaxspec: |
4112 | case notsyntaxspec: | |
4113 | if (!fastmap) break; | |
4114 | #ifndef emacs | |
4115 | not = (re_opcode_t)p[-1] == notsyntaxspec; | |
4116 | k = *p++; | |
4117 | for (j = 0; j < (1 << BYTEWIDTH); j++) | |
990b2375 | 4118 | if ((SYNTAX (j) == (enum syntaxcode) k) ^ not) |
b18215fc | 4119 | fastmap[j] = 1; |
b18215fc | 4120 | break; |
1fb352e0 | 4121 | #else /* emacs */ |
b18215fc RS |
4122 | /* This match depends on text properties. These end with |
4123 | aborting optimizations. */ | |
01618498 | 4124 | return -1; |
b18215fc RS |
4125 | |
4126 | case categoryspec: | |
b18215fc | 4127 | case notcategoryspec: |
1fb352e0 SM |
4128 | if (!fastmap) break; |
4129 | not = (re_opcode_t)p[-1] == notcategoryspec; | |
b18215fc | 4130 | k = *p++; |
6482db2e | 4131 | for (j = (1 << BYTEWIDTH); j >= 0; j--) |
1fb352e0 | 4132 | if ((CHAR_HAS_CATEGORY (j, k)) ^ not) |
b18215fc RS |
4133 | fastmap[j] = 1; |
4134 | ||
6482db2e KH |
4135 | /* Any leading code can possibly start a character which |
4136 | has or doesn't has the specified category. */ | |
4137 | if (match_any_multibyte_characters == false) | |
6fdd04b0 | 4138 | { |
6482db2e KH |
4139 | for (j = MIN_MULTIBYTE_LEADING_CODE; |
4140 | j <= MAX_MULTIBYTE_LEADING_CODE; j++) | |
4141 | fastmap[j] = 1; | |
4142 | match_any_multibyte_characters = true; | |
6fdd04b0 | 4143 | } |
b18215fc RS |
4144 | break; |
4145 | ||
fa9a63c5 | 4146 | /* All cases after this match the empty string. These end with |
25fe55af | 4147 | `continue'. */ |
fa9a63c5 | 4148 | |
fa9a63c5 RM |
4149 | case before_dot: |
4150 | case at_dot: | |
4151 | case after_dot: | |
1fb352e0 | 4152 | #endif /* !emacs */ |
25fe55af RS |
4153 | case no_op: |
4154 | case begline: | |
4155 | case endline: | |
fa9a63c5 RM |
4156 | case begbuf: |
4157 | case endbuf: | |
4158 | case wordbound: | |
4159 | case notwordbound: | |
4160 | case wordbeg: | |
4161 | case wordend: | |
669fa600 SM |
4162 | case symbeg: |
4163 | case symend: | |
25fe55af | 4164 | continue; |
fa9a63c5 RM |
4165 | |
4166 | ||
fa9a63c5 | 4167 | case jump: |
25fe55af | 4168 | EXTRACT_NUMBER_AND_INCR (j, p); |
505bde11 SM |
4169 | if (j < 0) |
4170 | /* Backward jumps can only go back to code that we've already | |
4171 | visited. `re_compile' should make sure this is true. */ | |
4172 | break; | |
25fe55af | 4173 | p += j; |
505bde11 SM |
4174 | switch (SWITCH_ENUM_CAST ((re_opcode_t) *p)) |
4175 | { | |
4176 | case on_failure_jump: | |
4177 | case on_failure_keep_string_jump: | |
505bde11 | 4178 | case on_failure_jump_loop: |
0683b6fa | 4179 | case on_failure_jump_nastyloop: |
505bde11 SM |
4180 | case on_failure_jump_smart: |
4181 | p++; | |
4182 | break; | |
4183 | default: | |
4184 | continue; | |
4185 | }; | |
4186 | /* Keep `p1' to allow the `on_failure_jump' we are jumping to | |
4187 | to jump back to "just after here". */ | |
4188 | /* Fallthrough */ | |
fa9a63c5 | 4189 | |
25fe55af RS |
4190 | case on_failure_jump: |
4191 | case on_failure_keep_string_jump: | |
0683b6fa | 4192 | case on_failure_jump_nastyloop: |
505bde11 SM |
4193 | case on_failure_jump_loop: |
4194 | case on_failure_jump_smart: | |
25fe55af | 4195 | EXTRACT_NUMBER_AND_INCR (j, p); |
505bde11 | 4196 | if (p + j <= p1) |
ed0767d8 | 4197 | ; /* Backward jump to be ignored. */ |
01618498 SM |
4198 | else |
4199 | { /* We have to look down both arms. | |
4200 | We first go down the "straight" path so as to minimize | |
4201 | stack usage when going through alternatives. */ | |
4202 | int r = analyse_first (p, pend, fastmap, multibyte); | |
4203 | if (r) return r; | |
4204 | p += j; | |
4205 | } | |
25fe55af | 4206 | continue; |
fa9a63c5 RM |
4207 | |
4208 | ||
ed0767d8 SM |
4209 | case jump_n: |
4210 | /* This code simply does not properly handle forward jump_n. */ | |
4211 | DEBUG_STATEMENT (EXTRACT_NUMBER (j, p); assert (j < 0)); | |
4212 | p += 4; | |
4213 | /* jump_n can either jump or fall through. The (backward) jump | |
4214 | case has already been handled, so we only need to look at the | |
4215 | fallthrough case. */ | |
4216 | continue; | |
177c0ea7 | 4217 | |
fa9a63c5 | 4218 | case succeed_n: |
ed0767d8 SM |
4219 | /* If N == 0, it should be an on_failure_jump_loop instead. */ |
4220 | DEBUG_STATEMENT (EXTRACT_NUMBER (j, p + 2); assert (j > 0)); | |
4221 | p += 4; | |
4222 | /* We only care about one iteration of the loop, so we don't | |
4223 | need to consider the case where this behaves like an | |
4224 | on_failure_jump. */ | |
25fe55af | 4225 | continue; |
fa9a63c5 RM |
4226 | |
4227 | ||
4228 | case set_number_at: | |
25fe55af RS |
4229 | p += 4; |
4230 | continue; | |
fa9a63c5 RM |
4231 | |
4232 | ||
4233 | case start_memory: | |
25fe55af | 4234 | case stop_memory: |
505bde11 | 4235 | p += 1; |
fa9a63c5 RM |
4236 | continue; |
4237 | ||
4238 | ||
4239 | default: | |
25fe55af RS |
4240 | abort (); /* We have listed all the cases. */ |
4241 | } /* switch *p++ */ | |
fa9a63c5 RM |
4242 | |
4243 | /* Getting here means we have found the possible starting | |
25fe55af | 4244 | characters for one path of the pattern -- and that the empty |
7814e705 | 4245 | string does not match. We need not follow this path further. */ |
01618498 | 4246 | return 0; |
fa9a63c5 RM |
4247 | } /* while p */ |
4248 | ||
01618498 SM |
4249 | /* We reached the end without matching anything. */ |
4250 | return 1; | |
4251 | ||
f6a3f532 SM |
4252 | } /* analyse_first */ |
4253 | \f | |
4254 | /* re_compile_fastmap computes a ``fastmap'' for the compiled pattern in | |
4255 | BUFP. A fastmap records which of the (1 << BYTEWIDTH) possible | |
4256 | characters can start a string that matches the pattern. This fastmap | |
4257 | is used by re_search to skip quickly over impossible starting points. | |
4258 | ||
4259 | Character codes above (1 << BYTEWIDTH) are not represented in the | |
4260 | fastmap, but the leading codes are represented. Thus, the fastmap | |
4261 | indicates which character sets could start a match. | |
4262 | ||
4263 | The caller must supply the address of a (1 << BYTEWIDTH)-byte data | |
4264 | area as BUFP->fastmap. | |
4265 | ||
4266 | We set the `fastmap', `fastmap_accurate', and `can_be_null' fields in | |
4267 | the pattern buffer. | |
4268 | ||
4269 | Returns 0 if we succeed, -2 if an internal error. */ | |
4270 | ||
4271 | int | |
971de7fb | 4272 | re_compile_fastmap (struct re_pattern_buffer *bufp) |
f6a3f532 SM |
4273 | { |
4274 | char *fastmap = bufp->fastmap; | |
4275 | int analysis; | |
4276 | ||
4277 | assert (fastmap && bufp->buffer); | |
4278 | ||
72af86bd | 4279 | memset (fastmap, 0, 1 << BYTEWIDTH); /* Assume nothing's valid. */ |
f6a3f532 SM |
4280 | bufp->fastmap_accurate = 1; /* It will be when we're done. */ |
4281 | ||
4282 | analysis = analyse_first (bufp->buffer, bufp->buffer + bufp->used, | |
2d1675e4 | 4283 | fastmap, RE_MULTIBYTE_P (bufp)); |
c0f9ea08 | 4284 | bufp->can_be_null = (analysis != 0); |
fa9a63c5 RM |
4285 | return 0; |
4286 | } /* re_compile_fastmap */ | |
4287 | \f | |
4288 | /* Set REGS to hold NUM_REGS registers, storing them in STARTS and | |
4289 | ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use | |
4290 | this memory for recording register information. STARTS and ENDS | |
4291 | must be allocated using the malloc library routine, and must each | |
4292 | be at least NUM_REGS * sizeof (regoff_t) bytes long. | |
4293 | ||
4294 | If NUM_REGS == 0, then subsequent matches should allocate their own | |
4295 | register data. | |
4296 | ||
4297 | Unless this function is called, the first search or match using | |
4298 | PATTERN_BUFFER will allocate its own register data, without | |
4299 | freeing the old data. */ | |
4300 | ||
4301 | void | |
971de7fb | 4302 | re_set_registers (struct re_pattern_buffer *bufp, struct re_registers *regs, unsigned int num_regs, regoff_t *starts, regoff_t *ends) |
fa9a63c5 RM |
4303 | { |
4304 | if (num_regs) | |
4305 | { | |
4306 | bufp->regs_allocated = REGS_REALLOCATE; | |
4307 | regs->num_regs = num_regs; | |
4308 | regs->start = starts; | |
4309 | regs->end = ends; | |
4310 | } | |
4311 | else | |
4312 | { | |
4313 | bufp->regs_allocated = REGS_UNALLOCATED; | |
4314 | regs->num_regs = 0; | |
4315 | regs->start = regs->end = (regoff_t *) 0; | |
4316 | } | |
4317 | } | |
c0f9ea08 | 4318 | WEAK_ALIAS (__re_set_registers, re_set_registers) |
fa9a63c5 | 4319 | \f |
7814e705 | 4320 | /* Searching routines. */ |
fa9a63c5 RM |
4321 | |
4322 | /* Like re_search_2, below, but only one string is specified, and | |
4323 | doesn't let you say where to stop matching. */ | |
4324 | ||
4325 | int | |
971de7fb | 4326 | re_search (struct re_pattern_buffer *bufp, const char *string, int size, int startpos, int range, struct re_registers *regs) |
fa9a63c5 | 4327 | { |
5e69f11e | 4328 | return re_search_2 (bufp, NULL, 0, string, size, startpos, range, |
fa9a63c5 RM |
4329 | regs, size); |
4330 | } | |
c0f9ea08 | 4331 | WEAK_ALIAS (__re_search, re_search) |
fa9a63c5 | 4332 | |
70806df6 KH |
4333 | /* Head address of virtual concatenation of string. */ |
4334 | #define HEAD_ADDR_VSTRING(P) \ | |
4335 | (((P) >= size1 ? string2 : string1)) | |
4336 | ||
b18215fc RS |
4337 | /* End address of virtual concatenation of string. */ |
4338 | #define STOP_ADDR_VSTRING(P) \ | |
4339 | (((P) >= size1 ? string2 + size2 : string1 + size1)) | |
4340 | ||
4341 | /* Address of POS in the concatenation of virtual string. */ | |
4342 | #define POS_ADDR_VSTRING(POS) \ | |
4343 | (((POS) >= size1 ? string2 - size1 : string1) + (POS)) | |
fa9a63c5 RM |
4344 | |
4345 | /* Using the compiled pattern in BUFP->buffer, first tries to match the | |
4346 | virtual concatenation of STRING1 and STRING2, starting first at index | |
4347 | STARTPOS, then at STARTPOS + 1, and so on. | |
5e69f11e | 4348 | |
fa9a63c5 | 4349 | STRING1 and STRING2 have length SIZE1 and SIZE2, respectively. |
5e69f11e | 4350 | |
fa9a63c5 RM |
4351 | RANGE is how far to scan while trying to match. RANGE = 0 means try |
4352 | only at STARTPOS; in general, the last start tried is STARTPOS + | |
4353 | RANGE. | |
5e69f11e | 4354 | |
fa9a63c5 RM |
4355 | In REGS, return the indices of the virtual concatenation of STRING1 |
4356 | and STRING2 that matched the entire BUFP->buffer and its contained | |
4357 | subexpressions. | |
5e69f11e | 4358 | |
fa9a63c5 RM |
4359 | Do not consider matching one past the index STOP in the virtual |
4360 | concatenation of STRING1 and STRING2. | |
4361 | ||
4362 | We return either the position in the strings at which the match was | |
4363 | found, -1 if no match, or -2 if error (such as failure | |
4364 | stack overflow). */ | |
4365 | ||
4366 | int | |
971de7fb | 4367 | re_search_2 (struct re_pattern_buffer *bufp, const char *str1, int size1, const char *str2, int size2, int startpos, int range, struct re_registers *regs, int stop) |
fa9a63c5 RM |
4368 | { |
4369 | int val; | |
66f0296e SM |
4370 | re_char *string1 = (re_char*) str1; |
4371 | re_char *string2 = (re_char*) str2; | |
fa9a63c5 | 4372 | register char *fastmap = bufp->fastmap; |
6676cb1c | 4373 | register RE_TRANSLATE_TYPE translate = bufp->translate; |
fa9a63c5 RM |
4374 | int total_size = size1 + size2; |
4375 | int endpos = startpos + range; | |
c0f9ea08 | 4376 | boolean anchored_start; |
cf9c99bc KH |
4377 | /* Nonzero if we are searching multibyte string. */ |
4378 | const boolean multibyte = RE_TARGET_MULTIBYTE_P (bufp); | |
b18215fc | 4379 | |
fa9a63c5 RM |
4380 | /* Check for out-of-range STARTPOS. */ |
4381 | if (startpos < 0 || startpos > total_size) | |
4382 | return -1; | |
5e69f11e | 4383 | |
fa9a63c5 | 4384 | /* Fix up RANGE if it might eventually take us outside |
34597fa9 | 4385 | the virtual concatenation of STRING1 and STRING2. |
5e69f11e | 4386 | Make sure we won't move STARTPOS below 0 or above TOTAL_SIZE. */ |
34597fa9 RS |
4387 | if (endpos < 0) |
4388 | range = 0 - startpos; | |
fa9a63c5 RM |
4389 | else if (endpos > total_size) |
4390 | range = total_size - startpos; | |
4391 | ||
4392 | /* If the search isn't to be a backwards one, don't waste time in a | |
7b140fd7 | 4393 | search for a pattern anchored at beginning of buffer. */ |
fa9a63c5 RM |
4394 | if (bufp->used > 0 && (re_opcode_t) bufp->buffer[0] == begbuf && range > 0) |
4395 | { | |
4396 | if (startpos > 0) | |
4397 | return -1; | |
4398 | else | |
7b140fd7 | 4399 | range = 0; |
fa9a63c5 RM |
4400 | } |
4401 | ||
ae4788a8 RS |
4402 | #ifdef emacs |
4403 | /* In a forward search for something that starts with \=. | |
4404 | don't keep searching past point. */ | |
4405 | if (bufp->used > 0 && (re_opcode_t) bufp->buffer[0] == at_dot && range > 0) | |
4406 | { | |
7b140fd7 RS |
4407 | range = PT_BYTE - BEGV_BYTE - startpos; |
4408 | if (range < 0) | |
ae4788a8 RS |
4409 | return -1; |
4410 | } | |
4411 | #endif /* emacs */ | |
4412 | ||
fa9a63c5 RM |
4413 | /* Update the fastmap now if not correct already. */ |
4414 | if (fastmap && !bufp->fastmap_accurate) | |
01618498 | 4415 | re_compile_fastmap (bufp); |
5e69f11e | 4416 | |
c8499ba5 | 4417 | /* See whether the pattern is anchored. */ |
c0f9ea08 | 4418 | anchored_start = (bufp->buffer[0] == begline); |
c8499ba5 | 4419 | |
b18215fc | 4420 | #ifdef emacs |
d48cd3f4 | 4421 | gl_state.object = re_match_object; /* Used by SYNTAX_TABLE_BYTE_TO_CHAR. */ |
cc9b4df2 | 4422 | { |
99633e97 | 4423 | int charpos = SYNTAX_TABLE_BYTE_TO_CHAR (POS_AS_IN_BUFFER (startpos)); |
cc9b4df2 KH |
4424 | |
4425 | SETUP_SYNTAX_TABLE_FOR_OBJECT (re_match_object, charpos, 1); | |
4426 | } | |
b18215fc RS |
4427 | #endif |
4428 | ||
fa9a63c5 RM |
4429 | /* Loop through the string, looking for a place to start matching. */ |
4430 | for (;;) | |
5e69f11e | 4431 | { |
c8499ba5 RS |
4432 | /* If the pattern is anchored, |
4433 | skip quickly past places we cannot match. | |
4434 | We don't bother to treat startpos == 0 specially | |
4435 | because that case doesn't repeat. */ | |
4436 | if (anchored_start && startpos > 0) | |
4437 | { | |
c0f9ea08 SM |
4438 | if (! ((startpos <= size1 ? string1[startpos - 1] |
4439 | : string2[startpos - size1 - 1]) | |
4440 | == '\n')) | |
c8499ba5 RS |
4441 | goto advance; |
4442 | } | |
4443 | ||
fa9a63c5 | 4444 | /* If a fastmap is supplied, skip quickly over characters that |
25fe55af RS |
4445 | cannot be the start of a match. If the pattern can match the |
4446 | null string, however, we don't need to skip characters; we want | |
7814e705 | 4447 | the first null string. */ |
fa9a63c5 RM |
4448 | if (fastmap && startpos < total_size && !bufp->can_be_null) |
4449 | { | |
66f0296e | 4450 | register re_char *d; |
01618498 | 4451 | register re_wchar_t buf_ch; |
e934739e RS |
4452 | |
4453 | d = POS_ADDR_VSTRING (startpos); | |
4454 | ||
7814e705 | 4455 | if (range > 0) /* Searching forwards. */ |
fa9a63c5 | 4456 | { |
fa9a63c5 RM |
4457 | register int lim = 0; |
4458 | int irange = range; | |
4459 | ||
25fe55af RS |
4460 | if (startpos < size1 && startpos + range >= size1) |
4461 | lim = range - (size1 - startpos); | |
fa9a63c5 | 4462 | |
25fe55af RS |
4463 | /* Written out as an if-else to avoid testing `translate' |
4464 | inside the loop. */ | |
28ae27ae AS |
4465 | if (RE_TRANSLATE_P (translate)) |
4466 | { | |
e934739e RS |
4467 | if (multibyte) |
4468 | while (range > lim) | |
4469 | { | |
4470 | int buf_charlen; | |
4471 | ||
62a6e103 | 4472 | buf_ch = STRING_CHAR_AND_LENGTH (d, buf_charlen); |
e934739e | 4473 | buf_ch = RE_TRANSLATE (translate, buf_ch); |
bf216479 | 4474 | if (fastmap[CHAR_LEADING_CODE (buf_ch)]) |
e934739e RS |
4475 | break; |
4476 | ||
4477 | range -= buf_charlen; | |
4478 | d += buf_charlen; | |
4479 | } | |
4480 | else | |
bf216479 | 4481 | while (range > lim) |
33c46939 | 4482 | { |
cf9c99bc KH |
4483 | register re_wchar_t ch, translated; |
4484 | ||
bf216479 | 4485 | buf_ch = *d; |
cf9c99bc KH |
4486 | ch = RE_CHAR_TO_MULTIBYTE (buf_ch); |
4487 | translated = RE_TRANSLATE (translate, ch); | |
4488 | if (translated != ch | |
4489 | && (ch = RE_CHAR_TO_UNIBYTE (translated)) >= 0) | |
4490 | buf_ch = ch; | |
6fdd04b0 | 4491 | if (fastmap[buf_ch]) |
bf216479 | 4492 | break; |
33c46939 RS |
4493 | d++; |
4494 | range--; | |
4495 | } | |
e934739e | 4496 | } |
fa9a63c5 | 4497 | else |
6fdd04b0 KH |
4498 | { |
4499 | if (multibyte) | |
4500 | while (range > lim) | |
4501 | { | |
4502 | int buf_charlen; | |
fa9a63c5 | 4503 | |
62a6e103 | 4504 | buf_ch = STRING_CHAR_AND_LENGTH (d, buf_charlen); |
6fdd04b0 KH |
4505 | if (fastmap[CHAR_LEADING_CODE (buf_ch)]) |
4506 | break; | |
4507 | range -= buf_charlen; | |
4508 | d += buf_charlen; | |
4509 | } | |
e934739e | 4510 | else |
6fdd04b0 | 4511 | while (range > lim && !fastmap[*d]) |
33c46939 RS |
4512 | { |
4513 | d++; | |
4514 | range--; | |
4515 | } | |
e934739e | 4516 | } |
fa9a63c5 RM |
4517 | startpos += irange - range; |
4518 | } | |
7814e705 | 4519 | else /* Searching backwards. */ |
fa9a63c5 | 4520 | { |
ba5e343c KH |
4521 | if (multibyte) |
4522 | { | |
62a6e103 | 4523 | buf_ch = STRING_CHAR (d); |
ba5e343c KH |
4524 | buf_ch = TRANSLATE (buf_ch); |
4525 | if (! fastmap[CHAR_LEADING_CODE (buf_ch)]) | |
4526 | goto advance; | |
4527 | } | |
4528 | else | |
4529 | { | |
cf9c99bc KH |
4530 | register re_wchar_t ch, translated; |
4531 | ||
4532 | buf_ch = *d; | |
4533 | ch = RE_CHAR_TO_MULTIBYTE (buf_ch); | |
4534 | translated = TRANSLATE (ch); | |
4535 | if (translated != ch | |
4536 | && (ch = RE_CHAR_TO_UNIBYTE (translated)) >= 0) | |
4537 | buf_ch = ch; | |
4538 | if (! fastmap[TRANSLATE (buf_ch)]) | |
ba5e343c KH |
4539 | goto advance; |
4540 | } | |
fa9a63c5 RM |
4541 | } |
4542 | } | |
4543 | ||
4544 | /* If can't match the null string, and that's all we have left, fail. */ | |
4545 | if (range >= 0 && startpos == total_size && fastmap | |
25fe55af | 4546 | && !bufp->can_be_null) |
fa9a63c5 RM |
4547 | return -1; |
4548 | ||
4549 | val = re_match_2_internal (bufp, string1, size1, string2, size2, | |
4550 | startpos, regs, stop); | |
fa9a63c5 RM |
4551 | |
4552 | if (val >= 0) | |
4553 | return startpos; | |
5e69f11e | 4554 | |
fa9a63c5 RM |
4555 | if (val == -2) |
4556 | return -2; | |
4557 | ||
4558 | advance: | |
5e69f11e | 4559 | if (!range) |
25fe55af | 4560 | break; |
5e69f11e | 4561 | else if (range > 0) |
25fe55af | 4562 | { |
b18215fc RS |
4563 | /* Update STARTPOS to the next character boundary. */ |
4564 | if (multibyte) | |
4565 | { | |
66f0296e SM |
4566 | re_char *p = POS_ADDR_VSTRING (startpos); |
4567 | re_char *pend = STOP_ADDR_VSTRING (startpos); | |
aa3830c4 | 4568 | int len = BYTES_BY_CHAR_HEAD (*p); |
b18215fc RS |
4569 | |
4570 | range -= len; | |
4571 | if (range < 0) | |
4572 | break; | |
4573 | startpos += len; | |
4574 | } | |
4575 | else | |
4576 | { | |
b560c397 RS |
4577 | range--; |
4578 | startpos++; | |
4579 | } | |
e318085a | 4580 | } |
fa9a63c5 | 4581 | else |
25fe55af RS |
4582 | { |
4583 | range++; | |
4584 | startpos--; | |
b18215fc RS |
4585 | |
4586 | /* Update STARTPOS to the previous character boundary. */ | |
4587 | if (multibyte) | |
4588 | { | |
70806df6 KH |
4589 | re_char *p = POS_ADDR_VSTRING (startpos) + 1; |
4590 | re_char *p0 = p; | |
4591 | re_char *phead = HEAD_ADDR_VSTRING (startpos); | |
b18215fc RS |
4592 | |
4593 | /* Find the head of multibyte form. */ | |
70806df6 KH |
4594 | PREV_CHAR_BOUNDARY (p, phead); |
4595 | range += p0 - 1 - p; | |
4596 | if (range > 0) | |
4597 | break; | |
b18215fc | 4598 | |
70806df6 | 4599 | startpos -= p0 - 1 - p; |
b18215fc | 4600 | } |
25fe55af | 4601 | } |
fa9a63c5 RM |
4602 | } |
4603 | return -1; | |
4604 | } /* re_search_2 */ | |
c0f9ea08 | 4605 | WEAK_ALIAS (__re_search_2, re_search_2) |
fa9a63c5 RM |
4606 | \f |
4607 | /* Declarations and macros for re_match_2. */ | |
4608 | ||
2d1675e4 SM |
4609 | static int bcmp_translate _RE_ARGS((re_char *s1, re_char *s2, |
4610 | register int len, | |
4611 | RE_TRANSLATE_TYPE translate, | |
4612 | const int multibyte)); | |
fa9a63c5 RM |
4613 | |
4614 | /* This converts PTR, a pointer into one of the search strings `string1' | |
4615 | and `string2' into an offset from the beginning of that string. */ | |
4616 | #define POINTER_TO_OFFSET(ptr) \ | |
4617 | (FIRST_STRING_P (ptr) \ | |
4618 | ? ((regoff_t) ((ptr) - string1)) \ | |
4619 | : ((regoff_t) ((ptr) - string2 + size1))) | |
4620 | ||
fa9a63c5 | 4621 | /* Call before fetching a character with *d. This switches over to |
419d1c74 SM |
4622 | string2 if necessary. |
4623 | Check re_match_2_internal for a discussion of why end_match_2 might | |
4624 | not be within string2 (but be equal to end_match_1 instead). */ | |
fa9a63c5 | 4625 | #define PREFETCH() \ |
25fe55af | 4626 | while (d == dend) \ |
fa9a63c5 RM |
4627 | { \ |
4628 | /* End of string2 => fail. */ \ | |
25fe55af RS |
4629 | if (dend == end_match_2) \ |
4630 | goto fail; \ | |
4bb91c68 | 4631 | /* End of string1 => advance to string2. */ \ |
25fe55af | 4632 | d = string2; \ |
fa9a63c5 RM |
4633 | dend = end_match_2; \ |
4634 | } | |
4635 | ||
f1ad044f SM |
4636 | /* Call before fetching a char with *d if you already checked other limits. |
4637 | This is meant for use in lookahead operations like wordend, etc.. | |
4638 | where we might need to look at parts of the string that might be | |
4639 | outside of the LIMITs (i.e past `stop'). */ | |
4640 | #define PREFETCH_NOLIMIT() \ | |
4641 | if (d == end1) \ | |
4642 | { \ | |
4643 | d = string2; \ | |
4644 | dend = end_match_2; \ | |
4645 | } \ | |
fa9a63c5 RM |
4646 | |
4647 | /* Test if at very beginning or at very end of the virtual concatenation | |
7814e705 | 4648 | of `string1' and `string2'. If only one string, it's `string2'. */ |
fa9a63c5 | 4649 | #define AT_STRINGS_BEG(d) ((d) == (size1 ? string1 : string2) || !size2) |
5e69f11e | 4650 | #define AT_STRINGS_END(d) ((d) == end2) |
fa9a63c5 RM |
4651 | |
4652 | ||
4653 | /* Test if D points to a character which is word-constituent. We have | |
4654 | two special cases to check for: if past the end of string1, look at | |
4655 | the first character in string2; and if before the beginning of | |
4656 | string2, look at the last character in string1. */ | |
4657 | #define WORDCHAR_P(d) \ | |
4658 | (SYNTAX ((d) == end1 ? *string2 \ | |
25fe55af | 4659 | : (d) == string2 - 1 ? *(end1 - 1) : *(d)) \ |
fa9a63c5 RM |
4660 | == Sword) |
4661 | ||
9121ca40 | 4662 | /* Disabled due to a compiler bug -- see comment at case wordbound */ |
b18215fc RS |
4663 | |
4664 | /* The comment at case wordbound is following one, but we don't use | |
4665 | AT_WORD_BOUNDARY anymore to support multibyte form. | |
4666 | ||
4667 | The DEC Alpha C compiler 3.x generates incorrect code for the | |
25fe55af | 4668 | test WORDCHAR_P (d - 1) != WORDCHAR_P (d) in the expansion of |
7814e705 | 4669 | AT_WORD_BOUNDARY, so this code is disabled. Expanding the |
b18215fc RS |
4670 | macro and introducing temporary variables works around the bug. */ |
4671 | ||
9121ca40 | 4672 | #if 0 |
fa9a63c5 RM |
4673 | /* Test if the character before D and the one at D differ with respect |
4674 | to being word-constituent. */ | |
4675 | #define AT_WORD_BOUNDARY(d) \ | |
4676 | (AT_STRINGS_BEG (d) || AT_STRINGS_END (d) \ | |
4677 | || WORDCHAR_P (d - 1) != WORDCHAR_P (d)) | |
9121ca40 | 4678 | #endif |
fa9a63c5 RM |
4679 | |
4680 | /* Free everything we malloc. */ | |
4681 | #ifdef MATCH_MAY_ALLOCATE | |
0b32bf0e SM |
4682 | # define FREE_VAR(var) if (var) { REGEX_FREE (var); var = NULL; } else |
4683 | # define FREE_VARIABLES() \ | |
fa9a63c5 RM |
4684 | do { \ |
4685 | REGEX_FREE_STACK (fail_stack.stack); \ | |
4686 | FREE_VAR (regstart); \ | |
4687 | FREE_VAR (regend); \ | |
fa9a63c5 RM |
4688 | FREE_VAR (best_regstart); \ |
4689 | FREE_VAR (best_regend); \ | |
fa9a63c5 RM |
4690 | } while (0) |
4691 | #else | |
0b32bf0e | 4692 | # define FREE_VARIABLES() ((void)0) /* Do nothing! But inhibit gcc warning. */ |
fa9a63c5 RM |
4693 | #endif /* not MATCH_MAY_ALLOCATE */ |
4694 | ||
505bde11 SM |
4695 | \f |
4696 | /* Optimization routines. */ | |
4697 | ||
4e8a9132 SM |
4698 | /* If the operation is a match against one or more chars, |
4699 | return a pointer to the next operation, else return NULL. */ | |
01618498 | 4700 | static re_char * |
971de7fb | 4701 | skip_one_char (const re_char *p) |
4e8a9132 SM |
4702 | { |
4703 | switch (SWITCH_ENUM_CAST (*p++)) | |
4704 | { | |
4705 | case anychar: | |
4706 | break; | |
177c0ea7 | 4707 | |
4e8a9132 SM |
4708 | case exactn: |
4709 | p += *p + 1; | |
4710 | break; | |
4711 | ||
4712 | case charset_not: | |
4713 | case charset: | |
4714 | if (CHARSET_RANGE_TABLE_EXISTS_P (p - 1)) | |
4715 | { | |
4716 | int mcnt; | |
4717 | p = CHARSET_RANGE_TABLE (p - 1); | |
4718 | EXTRACT_NUMBER_AND_INCR (mcnt, p); | |
4719 | p = CHARSET_RANGE_TABLE_END (p, mcnt); | |
4720 | } | |
4721 | else | |
4722 | p += 1 + CHARSET_BITMAP_SIZE (p - 1); | |
4723 | break; | |
177c0ea7 | 4724 | |
4e8a9132 SM |
4725 | case syntaxspec: |
4726 | case notsyntaxspec: | |
1fb352e0 | 4727 | #ifdef emacs |
4e8a9132 SM |
4728 | case categoryspec: |
4729 | case notcategoryspec: | |
4730 | #endif /* emacs */ | |
4731 | p++; | |
4732 | break; | |
4733 | ||
4734 | default: | |
4735 | p = NULL; | |
4736 | } | |
4737 | return p; | |
4738 | } | |
4739 | ||
4740 | ||
505bde11 | 4741 | /* Jump over non-matching operations. */ |
839966f3 | 4742 | static re_char * |
971de7fb | 4743 | skip_noops (const re_char *p, const re_char *pend) |
505bde11 SM |
4744 | { |
4745 | int mcnt; | |
4746 | while (p < pend) | |
4747 | { | |
4748 | switch (SWITCH_ENUM_CAST ((re_opcode_t) *p)) | |
4749 | { | |
4750 | case start_memory: | |
505bde11 SM |
4751 | case stop_memory: |
4752 | p += 2; break; | |
4753 | case no_op: | |
4754 | p += 1; break; | |
4755 | case jump: | |
4756 | p += 1; | |
4757 | EXTRACT_NUMBER_AND_INCR (mcnt, p); | |
4758 | p += mcnt; | |
4759 | break; | |
4760 | default: | |
4761 | return p; | |
4762 | } | |
4763 | } | |
4764 | assert (p == pend); | |
4765 | return p; | |
4766 | } | |
4767 | ||
4768 | /* Non-zero if "p1 matches something" implies "p2 fails". */ | |
4769 | static int | |
971de7fb | 4770 | mutually_exclusive_p (struct re_pattern_buffer *bufp, const re_char *p1, const re_char *p2) |
505bde11 | 4771 | { |
4e8a9132 | 4772 | re_opcode_t op2; |
2d1675e4 | 4773 | const boolean multibyte = RE_MULTIBYTE_P (bufp); |
505bde11 SM |
4774 | unsigned char *pend = bufp->buffer + bufp->used; |
4775 | ||
4e8a9132 | 4776 | assert (p1 >= bufp->buffer && p1 < pend |
505bde11 SM |
4777 | && p2 >= bufp->buffer && p2 <= pend); |
4778 | ||
4779 | /* Skip over open/close-group commands. | |
4780 | If what follows this loop is a ...+ construct, | |
4781 | look at what begins its body, since we will have to | |
4782 | match at least one of that. */ | |
4e8a9132 SM |
4783 | p2 = skip_noops (p2, pend); |
4784 | /* The same skip can be done for p1, except that this function | |
4785 | is only used in the case where p1 is a simple match operator. */ | |
4786 | /* p1 = skip_noops (p1, pend); */ | |
4787 | ||
4788 | assert (p1 >= bufp->buffer && p1 < pend | |
4789 | && p2 >= bufp->buffer && p2 <= pend); | |
4790 | ||
4791 | op2 = p2 == pend ? succeed : *p2; | |
4792 | ||
4793 | switch (SWITCH_ENUM_CAST (op2)) | |
505bde11 | 4794 | { |
4e8a9132 SM |
4795 | case succeed: |
4796 | case endbuf: | |
4797 | /* If we're at the end of the pattern, we can change. */ | |
4798 | if (skip_one_char (p1)) | |
505bde11 | 4799 | { |
505bde11 SM |
4800 | DEBUG_PRINT1 (" End of pattern: fast loop.\n"); |
4801 | return 1; | |
505bde11 | 4802 | } |
4e8a9132 | 4803 | break; |
177c0ea7 | 4804 | |
4e8a9132 | 4805 | case endline: |
4e8a9132 SM |
4806 | case exactn: |
4807 | { | |
01618498 | 4808 | register re_wchar_t c |
4e8a9132 | 4809 | = (re_opcode_t) *p2 == endline ? '\n' |
62a6e103 | 4810 | : RE_STRING_CHAR (p2 + 2, multibyte); |
505bde11 | 4811 | |
4e8a9132 SM |
4812 | if ((re_opcode_t) *p1 == exactn) |
4813 | { | |
62a6e103 | 4814 | if (c != RE_STRING_CHAR (p1 + 2, multibyte)) |
4e8a9132 SM |
4815 | { |
4816 | DEBUG_PRINT3 (" '%c' != '%c' => fast loop.\n", c, p1[2]); | |
4817 | return 1; | |
4818 | } | |
4819 | } | |
505bde11 | 4820 | |
4e8a9132 SM |
4821 | else if ((re_opcode_t) *p1 == charset |
4822 | || (re_opcode_t) *p1 == charset_not) | |
4823 | { | |
4824 | int not = (re_opcode_t) *p1 == charset_not; | |
505bde11 | 4825 | |
4e8a9132 SM |
4826 | /* Test if C is listed in charset (or charset_not) |
4827 | at `p1'. */ | |
6fdd04b0 | 4828 | if (! multibyte || IS_REAL_ASCII (c)) |
4e8a9132 SM |
4829 | { |
4830 | if (c < CHARSET_BITMAP_SIZE (p1) * BYTEWIDTH | |
4831 | && p1[2 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH))) | |
4832 | not = !not; | |
4833 | } | |
4834 | else if (CHARSET_RANGE_TABLE_EXISTS_P (p1)) | |
4835 | CHARSET_LOOKUP_RANGE_TABLE (not, c, p1); | |
505bde11 | 4836 | |
4e8a9132 SM |
4837 | /* `not' is equal to 1 if c would match, which means |
4838 | that we can't change to pop_failure_jump. */ | |
4839 | if (!not) | |
4840 | { | |
4841 | DEBUG_PRINT1 (" No match => fast loop.\n"); | |
4842 | return 1; | |
4843 | } | |
4844 | } | |
4845 | else if ((re_opcode_t) *p1 == anychar | |
4846 | && c == '\n') | |
4847 | { | |
4848 | DEBUG_PRINT1 (" . != \\n => fast loop.\n"); | |
4849 | return 1; | |
4850 | } | |
4851 | } | |
4852 | break; | |
505bde11 | 4853 | |
4e8a9132 | 4854 | case charset: |
4e8a9132 SM |
4855 | { |
4856 | if ((re_opcode_t) *p1 == exactn) | |
4857 | /* Reuse the code above. */ | |
4858 | return mutually_exclusive_p (bufp, p2, p1); | |
505bde11 | 4859 | |
505bde11 SM |
4860 | /* It is hard to list up all the character in charset |
4861 | P2 if it includes multibyte character. Give up in | |
4862 | such case. */ | |
4863 | else if (!multibyte || !CHARSET_RANGE_TABLE_EXISTS_P (p2)) | |
4864 | { | |
4865 | /* Now, we are sure that P2 has no range table. | |
4866 | So, for the size of bitmap in P2, `p2[1]' is | |
7814e705 | 4867 | enough. But P1 may have range table, so the |
505bde11 SM |
4868 | size of bitmap table of P1 is extracted by |
4869 | using macro `CHARSET_BITMAP_SIZE'. | |
4870 | ||
6fdd04b0 KH |
4871 | In a multibyte case, we know that all the character |
4872 | listed in P2 is ASCII. In a unibyte case, P1 has only a | |
4873 | bitmap table. So, in both cases, it is enough to test | |
4874 | only the bitmap table of P1. */ | |
505bde11 | 4875 | |
411e4203 | 4876 | if ((re_opcode_t) *p1 == charset) |
505bde11 SM |
4877 | { |
4878 | int idx; | |
4879 | /* We win if the charset inside the loop | |
4880 | has no overlap with the one after the loop. */ | |
4881 | for (idx = 0; | |
4882 | (idx < (int) p2[1] | |
4883 | && idx < CHARSET_BITMAP_SIZE (p1)); | |
4884 | idx++) | |
4885 | if ((p2[2 + idx] & p1[2 + idx]) != 0) | |
4886 | break; | |
4887 | ||
4888 | if (idx == p2[1] | |
4889 | || idx == CHARSET_BITMAP_SIZE (p1)) | |
4890 | { | |
4891 | DEBUG_PRINT1 (" No match => fast loop.\n"); | |
4892 | return 1; | |
4893 | } | |
4894 | } | |
411e4203 | 4895 | else if ((re_opcode_t) *p1 == charset_not) |
505bde11 SM |
4896 | { |
4897 | int idx; | |
4898 | /* We win if the charset_not inside the loop lists | |
7814e705 | 4899 | every character listed in the charset after. */ |
505bde11 SM |
4900 | for (idx = 0; idx < (int) p2[1]; idx++) |
4901 | if (! (p2[2 + idx] == 0 | |
4902 | || (idx < CHARSET_BITMAP_SIZE (p1) | |
4903 | && ((p2[2 + idx] & ~ p1[2 + idx]) == 0)))) | |
4904 | break; | |
4905 | ||
4e8a9132 SM |
4906 | if (idx == p2[1]) |
4907 | { | |
4908 | DEBUG_PRINT1 (" No match => fast loop.\n"); | |
4909 | return 1; | |
4910 | } | |
4911 | } | |
4912 | } | |
4913 | } | |
609b757a | 4914 | break; |
177c0ea7 | 4915 | |
411e4203 SM |
4916 | case charset_not: |
4917 | switch (SWITCH_ENUM_CAST (*p1)) | |
4918 | { | |
4919 | case exactn: | |
4920 | case charset: | |
4921 | /* Reuse the code above. */ | |
4922 | return mutually_exclusive_p (bufp, p2, p1); | |
4923 | case charset_not: | |
4924 | /* When we have two charset_not, it's very unlikely that | |
4925 | they don't overlap. The union of the two sets of excluded | |
4926 | chars should cover all possible chars, which, as a matter of | |
4927 | fact, is virtually impossible in multibyte buffers. */ | |
36595814 | 4928 | break; |
411e4203 SM |
4929 | } |
4930 | break; | |
4931 | ||
4e8a9132 | 4932 | case wordend: |
669fa600 SM |
4933 | return ((re_opcode_t) *p1 == syntaxspec && p1[1] == Sword); |
4934 | case symend: | |
4e8a9132 | 4935 | return ((re_opcode_t) *p1 == syntaxspec |
669fa600 SM |
4936 | && (p1[1] == Ssymbol || p1[1] == Sword)); |
4937 | case notsyntaxspec: | |
4938 | return ((re_opcode_t) *p1 == syntaxspec && p1[1] == p2[1]); | |
4e8a9132 SM |
4939 | |
4940 | case wordbeg: | |
669fa600 SM |
4941 | return ((re_opcode_t) *p1 == notsyntaxspec && p1[1] == Sword); |
4942 | case symbeg: | |
4e8a9132 | 4943 | return ((re_opcode_t) *p1 == notsyntaxspec |
669fa600 SM |
4944 | && (p1[1] == Ssymbol || p1[1] == Sword)); |
4945 | case syntaxspec: | |
4946 | return ((re_opcode_t) *p1 == notsyntaxspec && p1[1] == p2[1]); | |
4e8a9132 SM |
4947 | |
4948 | case wordbound: | |
4949 | return (((re_opcode_t) *p1 == notsyntaxspec | |
4950 | || (re_opcode_t) *p1 == syntaxspec) | |
4951 | && p1[1] == Sword); | |
4952 | ||
1fb352e0 | 4953 | #ifdef emacs |
4e8a9132 SM |
4954 | case categoryspec: |
4955 | return ((re_opcode_t) *p1 == notcategoryspec && p1[1] == p2[1]); | |
4956 | case notcategoryspec: | |
4957 | return ((re_opcode_t) *p1 == categoryspec && p1[1] == p2[1]); | |
4958 | #endif /* emacs */ | |
4959 | ||
4960 | default: | |
4961 | ; | |
505bde11 SM |
4962 | } |
4963 | ||
4964 | /* Safe default. */ | |
4965 | return 0; | |
4966 | } | |
4967 | ||
fa9a63c5 RM |
4968 | \f |
4969 | /* Matching routines. */ | |
4970 | ||
25fe55af | 4971 | #ifndef emacs /* Emacs never uses this. */ |
fa9a63c5 RM |
4972 | /* re_match is like re_match_2 except it takes only a single string. */ |
4973 | ||
4974 | int | |
d2762c86 DN |
4975 | re_match (struct re_pattern_buffer *bufp, const char *string, |
4976 | int size, int pos, struct re_registers *regs) | |
fa9a63c5 | 4977 | { |
4bb91c68 | 4978 | int result = re_match_2_internal (bufp, NULL, 0, (re_char*) string, size, |
fa9a63c5 | 4979 | pos, regs, size); |
fa9a63c5 RM |
4980 | return result; |
4981 | } | |
c0f9ea08 | 4982 | WEAK_ALIAS (__re_match, re_match) |
fa9a63c5 RM |
4983 | #endif /* not emacs */ |
4984 | ||
b18215fc RS |
4985 | #ifdef emacs |
4986 | /* In Emacs, this is the string or buffer in which we | |
7814e705 | 4987 | are matching. It is used for looking up syntax properties. */ |
b18215fc RS |
4988 | Lisp_Object re_match_object; |
4989 | #endif | |
fa9a63c5 RM |
4990 | |
4991 | /* re_match_2 matches the compiled pattern in BUFP against the | |
4992 | the (virtual) concatenation of STRING1 and STRING2 (of length SIZE1 | |
4993 | and SIZE2, respectively). We start matching at POS, and stop | |
4994 | matching at STOP. | |
5e69f11e | 4995 | |
fa9a63c5 | 4996 | If REGS is non-null and the `no_sub' field of BUFP is nonzero, we |
7814e705 | 4997 | store offsets for the substring each group matched in REGS. See the |
fa9a63c5 RM |
4998 | documentation for exactly how many groups we fill. |
4999 | ||
5000 | We return -1 if no match, -2 if an internal error (such as the | |
7814e705 | 5001 | failure stack overflowing). Otherwise, we return the length of the |
fa9a63c5 RM |
5002 | matched substring. */ |
5003 | ||
5004 | int | |
971de7fb | 5005 | re_match_2 (struct re_pattern_buffer *bufp, const char *string1, int size1, const char *string2, int size2, int pos, struct re_registers *regs, int stop) |
fa9a63c5 | 5006 | { |
b18215fc | 5007 | int result; |
25fe55af | 5008 | |
b18215fc | 5009 | #ifdef emacs |
cc9b4df2 | 5010 | int charpos; |
d48cd3f4 | 5011 | gl_state.object = re_match_object; /* Used by SYNTAX_TABLE_BYTE_TO_CHAR. */ |
99633e97 | 5012 | charpos = SYNTAX_TABLE_BYTE_TO_CHAR (POS_AS_IN_BUFFER (pos)); |
cc9b4df2 | 5013 | SETUP_SYNTAX_TABLE_FOR_OBJECT (re_match_object, charpos, 1); |
b18215fc RS |
5014 | #endif |
5015 | ||
4bb91c68 SM |
5016 | result = re_match_2_internal (bufp, (re_char*) string1, size1, |
5017 | (re_char*) string2, size2, | |
cc9b4df2 | 5018 | pos, regs, stop); |
fa9a63c5 RM |
5019 | return result; |
5020 | } | |
c0f9ea08 | 5021 | WEAK_ALIAS (__re_match_2, re_match_2) |
fa9a63c5 | 5022 | |
bf216479 | 5023 | |
fa9a63c5 | 5024 | /* This is a separate function so that we can force an alloca cleanup |
7814e705 | 5025 | afterwards. */ |
fa9a63c5 | 5026 | static int |
971de7fb | 5027 | re_match_2_internal (struct re_pattern_buffer *bufp, const re_char *string1, int size1, const re_char *string2, int size2, int pos, struct re_registers *regs, int stop) |
fa9a63c5 RM |
5028 | { |
5029 | /* General temporaries. */ | |
5030 | int mcnt; | |
01618498 | 5031 | size_t reg; |
66f0296e | 5032 | boolean not; |
fa9a63c5 RM |
5033 | |
5034 | /* Just past the end of the corresponding string. */ | |
66f0296e | 5035 | re_char *end1, *end2; |
fa9a63c5 RM |
5036 | |
5037 | /* Pointers into string1 and string2, just past the last characters in | |
7814e705 | 5038 | each to consider matching. */ |
66f0296e | 5039 | re_char *end_match_1, *end_match_2; |
fa9a63c5 RM |
5040 | |
5041 | /* Where we are in the data, and the end of the current string. */ | |
66f0296e | 5042 | re_char *d, *dend; |
5e69f11e | 5043 | |
99633e97 SM |
5044 | /* Used sometimes to remember where we were before starting matching |
5045 | an operator so that we can go back in case of failure. This "atomic" | |
5046 | behavior of matching opcodes is indispensable to the correctness | |
5047 | of the on_failure_keep_string_jump optimization. */ | |
5048 | re_char *dfail; | |
5049 | ||
fa9a63c5 | 5050 | /* Where we are in the pattern, and the end of the pattern. */ |
01618498 SM |
5051 | re_char *p = bufp->buffer; |
5052 | re_char *pend = p + bufp->used; | |
fa9a63c5 | 5053 | |
25fe55af | 5054 | /* We use this to map every character in the string. */ |
6676cb1c | 5055 | RE_TRANSLATE_TYPE translate = bufp->translate; |
fa9a63c5 | 5056 | |
cf9c99bc | 5057 | /* Nonzero if BUFP is setup from a multibyte regex. */ |
2d1675e4 | 5058 | const boolean multibyte = RE_MULTIBYTE_P (bufp); |
b18215fc | 5059 | |
cf9c99bc KH |
5060 | /* Nonzero if STRING1/STRING2 are multibyte. */ |
5061 | const boolean target_multibyte = RE_TARGET_MULTIBYTE_P (bufp); | |
5062 | ||
fa9a63c5 RM |
5063 | /* Failure point stack. Each place that can handle a failure further |
5064 | down the line pushes a failure point on this stack. It consists of | |
505bde11 | 5065 | regstart, and regend for all registers corresponding to |
fa9a63c5 RM |
5066 | the subexpressions we're currently inside, plus the number of such |
5067 | registers, and, finally, two char *'s. The first char * is where | |
5068 | to resume scanning the pattern; the second one is where to resume | |
7814e705 JB |
5069 | scanning the strings. */ |
5070 | #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */ | |
fa9a63c5 RM |
5071 | fail_stack_type fail_stack; |
5072 | #endif | |
5073 | #ifdef DEBUG | |
fa9a63c5 RM |
5074 | unsigned nfailure_points_pushed = 0, nfailure_points_popped = 0; |
5075 | #endif | |
5076 | ||
0b32bf0e | 5077 | #if defined REL_ALLOC && defined REGEX_MALLOC |
fa9a63c5 RM |
5078 | /* This holds the pointer to the failure stack, when |
5079 | it is allocated relocatably. */ | |
5080 | fail_stack_elt_t *failure_stack_ptr; | |
99633e97 | 5081 | #endif |
fa9a63c5 RM |
5082 | |
5083 | /* We fill all the registers internally, independent of what we | |
7814e705 | 5084 | return, for use in backreferences. The number here includes |
fa9a63c5 | 5085 | an element for register zero. */ |
4bb91c68 | 5086 | size_t num_regs = bufp->re_nsub + 1; |
5e69f11e | 5087 | |
fa9a63c5 RM |
5088 | /* Information on the contents of registers. These are pointers into |
5089 | the input strings; they record just what was matched (on this | |
5090 | attempt) by a subexpression part of the pattern, that is, the | |
5091 | regnum-th regstart pointer points to where in the pattern we began | |
5092 | matching and the regnum-th regend points to right after where we | |
5093 | stopped matching the regnum-th subexpression. (The zeroth register | |
5094 | keeps track of what the whole pattern matches.) */ | |
5095 | #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */ | |
66f0296e | 5096 | re_char **regstart, **regend; |
fa9a63c5 RM |
5097 | #endif |
5098 | ||
fa9a63c5 | 5099 | /* The following record the register info as found in the above |
5e69f11e | 5100 | variables when we find a match better than any we've seen before. |
fa9a63c5 RM |
5101 | This happens as we backtrack through the failure points, which in |
5102 | turn happens only if we have not yet matched the entire string. */ | |
5103 | unsigned best_regs_set = false; | |
5104 | #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */ | |
66f0296e | 5105 | re_char **best_regstart, **best_regend; |
fa9a63c5 | 5106 | #endif |
5e69f11e | 5107 | |
fa9a63c5 RM |
5108 | /* Logically, this is `best_regend[0]'. But we don't want to have to |
5109 | allocate space for that if we're not allocating space for anything | |
7814e705 | 5110 | else (see below). Also, we never need info about register 0 for |
fa9a63c5 RM |
5111 | any of the other register vectors, and it seems rather a kludge to |
5112 | treat `best_regend' differently than the rest. So we keep track of | |
5113 | the end of the best match so far in a separate variable. We | |
5114 | initialize this to NULL so that when we backtrack the first time | |
5115 | and need to test it, it's not garbage. */ | |
66f0296e | 5116 | re_char *match_end = NULL; |
fa9a63c5 | 5117 | |
fa9a63c5 RM |
5118 | #ifdef DEBUG |
5119 | /* Counts the total number of registers pushed. */ | |
5e69f11e | 5120 | unsigned num_regs_pushed = 0; |
fa9a63c5 RM |
5121 | #endif |
5122 | ||
5123 | DEBUG_PRINT1 ("\n\nEntering re_match_2.\n"); | |
5e69f11e | 5124 | |
fa9a63c5 | 5125 | INIT_FAIL_STACK (); |
5e69f11e | 5126 | |
fa9a63c5 RM |
5127 | #ifdef MATCH_MAY_ALLOCATE |
5128 | /* Do not bother to initialize all the register variables if there are | |
5129 | no groups in the pattern, as it takes a fair amount of time. If | |
5130 | there are groups, we include space for register 0 (the whole | |
5131 | pattern), even though we never use it, since it simplifies the | |
5132 | array indexing. We should fix this. */ | |
5133 | if (bufp->re_nsub) | |
5134 | { | |
66f0296e SM |
5135 | regstart = REGEX_TALLOC (num_regs, re_char *); |
5136 | regend = REGEX_TALLOC (num_regs, re_char *); | |
5137 | best_regstart = REGEX_TALLOC (num_regs, re_char *); | |
5138 | best_regend = REGEX_TALLOC (num_regs, re_char *); | |
fa9a63c5 | 5139 | |
505bde11 | 5140 | if (!(regstart && regend && best_regstart && best_regend)) |
25fe55af RS |
5141 | { |
5142 | FREE_VARIABLES (); | |
5143 | return -2; | |
5144 | } | |
fa9a63c5 RM |
5145 | } |
5146 | else | |
5147 | { | |
5148 | /* We must initialize all our variables to NULL, so that | |
25fe55af | 5149 | `FREE_VARIABLES' doesn't try to free them. */ |
505bde11 | 5150 | regstart = regend = best_regstart = best_regend = NULL; |
fa9a63c5 RM |
5151 | } |
5152 | #endif /* MATCH_MAY_ALLOCATE */ | |
5153 | ||
5154 | /* The starting position is bogus. */ | |
5155 | if (pos < 0 || pos > size1 + size2) | |
5156 | { | |
5157 | FREE_VARIABLES (); | |
5158 | return -1; | |
5159 | } | |
5e69f11e | 5160 | |
fa9a63c5 RM |
5161 | /* Initialize subexpression text positions to -1 to mark ones that no |
5162 | start_memory/stop_memory has been seen for. Also initialize the | |
5163 | register information struct. */ | |
01618498 SM |
5164 | for (reg = 1; reg < num_regs; reg++) |
5165 | regstart[reg] = regend[reg] = NULL; | |
99633e97 | 5166 | |
fa9a63c5 | 5167 | /* We move `string1' into `string2' if the latter's empty -- but not if |
7814e705 | 5168 | `string1' is null. */ |
fa9a63c5 RM |
5169 | if (size2 == 0 && string1 != NULL) |
5170 | { | |
5171 | string2 = string1; | |
5172 | size2 = size1; | |
5173 | string1 = 0; | |
5174 | size1 = 0; | |
5175 | } | |
5176 | end1 = string1 + size1; | |
5177 | end2 = string2 + size2; | |
5178 | ||
5e69f11e | 5179 | /* `p' scans through the pattern as `d' scans through the data. |
fa9a63c5 RM |
5180 | `dend' is the end of the input string that `d' points within. `d' |
5181 | is advanced into the following input string whenever necessary, but | |
5182 | this happens before fetching; therefore, at the beginning of the | |
5183 | loop, `d' can be pointing at the end of a string, but it cannot | |
5184 | equal `string2'. */ | |
419d1c74 | 5185 | if (pos >= size1) |
fa9a63c5 | 5186 | { |
419d1c74 SM |
5187 | /* Only match within string2. */ |
5188 | d = string2 + pos - size1; | |
5189 | dend = end_match_2 = string2 + stop - size1; | |
5190 | end_match_1 = end1; /* Just to give it a value. */ | |
fa9a63c5 RM |
5191 | } |
5192 | else | |
5193 | { | |
f1ad044f | 5194 | if (stop < size1) |
419d1c74 SM |
5195 | { |
5196 | /* Only match within string1. */ | |
5197 | end_match_1 = string1 + stop; | |
5198 | /* BEWARE! | |
5199 | When we reach end_match_1, PREFETCH normally switches to string2. | |
5200 | But in the present case, this means that just doing a PREFETCH | |
5201 | makes us jump from `stop' to `gap' within the string. | |
5202 | What we really want here is for the search to stop as | |
5203 | soon as we hit end_match_1. That's why we set end_match_2 | |
5204 | to end_match_1 (since PREFETCH fails as soon as we hit | |
5205 | end_match_2). */ | |
5206 | end_match_2 = end_match_1; | |
5207 | } | |
5208 | else | |
f1ad044f SM |
5209 | { /* It's important to use this code when stop == size so that |
5210 | moving `d' from end1 to string2 will not prevent the d == dend | |
5211 | check from catching the end of string. */ | |
419d1c74 SM |
5212 | end_match_1 = end1; |
5213 | end_match_2 = string2 + stop - size1; | |
5214 | } | |
5215 | d = string1 + pos; | |
5216 | dend = end_match_1; | |
fa9a63c5 RM |
5217 | } |
5218 | ||
5219 | DEBUG_PRINT1 ("The compiled pattern is: "); | |
5220 | DEBUG_PRINT_COMPILED_PATTERN (bufp, p, pend); | |
5221 | DEBUG_PRINT1 ("The string to match is: `"); | |
5222 | DEBUG_PRINT_DOUBLE_STRING (d, string1, size1, string2, size2); | |
5223 | DEBUG_PRINT1 ("'\n"); | |
5e69f11e | 5224 | |
7814e705 | 5225 | /* This loops over pattern commands. It exits by returning from the |
fa9a63c5 RM |
5226 | function if the match is complete, or it drops through if the match |
5227 | fails at this starting point in the input data. */ | |
5228 | for (;;) | |
5229 | { | |
505bde11 | 5230 | DEBUG_PRINT2 ("\n%p: ", p); |
fa9a63c5 RM |
5231 | |
5232 | if (p == pend) | |
5233 | { /* End of pattern means we might have succeeded. */ | |
25fe55af | 5234 | DEBUG_PRINT1 ("end of pattern ... "); |
5e69f11e | 5235 | |
fa9a63c5 | 5236 | /* If we haven't matched the entire string, and we want the |
25fe55af RS |
5237 | longest match, try backtracking. */ |
5238 | if (d != end_match_2) | |
fa9a63c5 RM |
5239 | { |
5240 | /* 1 if this match ends in the same string (string1 or string2) | |
5241 | as the best previous match. */ | |
5e69f11e | 5242 | boolean same_str_p = (FIRST_STRING_P (match_end) |
99633e97 | 5243 | == FIRST_STRING_P (d)); |
fa9a63c5 RM |
5244 | /* 1 if this match is the best seen so far. */ |
5245 | boolean best_match_p; | |
5246 | ||
5247 | /* AIX compiler got confused when this was combined | |
7814e705 | 5248 | with the previous declaration. */ |
fa9a63c5 RM |
5249 | if (same_str_p) |
5250 | best_match_p = d > match_end; | |
5251 | else | |
99633e97 | 5252 | best_match_p = !FIRST_STRING_P (d); |
fa9a63c5 | 5253 | |
25fe55af RS |
5254 | DEBUG_PRINT1 ("backtracking.\n"); |
5255 | ||
5256 | if (!FAIL_STACK_EMPTY ()) | |
5257 | { /* More failure points to try. */ | |
5258 | ||
5259 | /* If exceeds best match so far, save it. */ | |
5260 | if (!best_regs_set || best_match_p) | |
5261 | { | |
5262 | best_regs_set = true; | |
5263 | match_end = d; | |
5264 | ||
5265 | DEBUG_PRINT1 ("\nSAVING match as best so far.\n"); | |
5266 | ||
01618498 | 5267 | for (reg = 1; reg < num_regs; reg++) |
25fe55af | 5268 | { |
01618498 SM |
5269 | best_regstart[reg] = regstart[reg]; |
5270 | best_regend[reg] = regend[reg]; | |
25fe55af RS |
5271 | } |
5272 | } | |
5273 | goto fail; | |
5274 | } | |
5275 | ||
5276 | /* If no failure points, don't restore garbage. And if | |
5277 | last match is real best match, don't restore second | |
5278 | best one. */ | |
5279 | else if (best_regs_set && !best_match_p) | |
5280 | { | |
5281 | restore_best_regs: | |
5282 | /* Restore best match. It may happen that `dend == | |
5283 | end_match_1' while the restored d is in string2. | |
5284 | For example, the pattern `x.*y.*z' against the | |
5285 | strings `x-' and `y-z-', if the two strings are | |
7814e705 | 5286 | not consecutive in memory. */ |
25fe55af RS |
5287 | DEBUG_PRINT1 ("Restoring best registers.\n"); |
5288 | ||
5289 | d = match_end; | |
5290 | dend = ((d >= string1 && d <= end1) | |
5291 | ? end_match_1 : end_match_2); | |
fa9a63c5 | 5292 | |
01618498 | 5293 | for (reg = 1; reg < num_regs; reg++) |
fa9a63c5 | 5294 | { |
01618498 SM |
5295 | regstart[reg] = best_regstart[reg]; |
5296 | regend[reg] = best_regend[reg]; | |
fa9a63c5 | 5297 | } |
25fe55af RS |
5298 | } |
5299 | } /* d != end_match_2 */ | |
fa9a63c5 RM |
5300 | |
5301 | succeed_label: | |
25fe55af | 5302 | DEBUG_PRINT1 ("Accepting match.\n"); |
fa9a63c5 | 5303 | |
25fe55af RS |
5304 | /* If caller wants register contents data back, do it. */ |
5305 | if (regs && !bufp->no_sub) | |
fa9a63c5 | 5306 | { |
25fe55af RS |
5307 | /* Have the register data arrays been allocated? */ |
5308 | if (bufp->regs_allocated == REGS_UNALLOCATED) | |
7814e705 | 5309 | { /* No. So allocate them with malloc. We need one |
25fe55af RS |
5310 | extra element beyond `num_regs' for the `-1' marker |
5311 | GNU code uses. */ | |
5312 | regs->num_regs = MAX (RE_NREGS, num_regs + 1); | |
5313 | regs->start = TALLOC (regs->num_regs, regoff_t); | |
5314 | regs->end = TALLOC (regs->num_regs, regoff_t); | |
5315 | if (regs->start == NULL || regs->end == NULL) | |
fa9a63c5 RM |
5316 | { |
5317 | FREE_VARIABLES (); | |
5318 | return -2; | |
5319 | } | |
25fe55af RS |
5320 | bufp->regs_allocated = REGS_REALLOCATE; |
5321 | } | |
5322 | else if (bufp->regs_allocated == REGS_REALLOCATE) | |
5323 | { /* Yes. If we need more elements than were already | |
5324 | allocated, reallocate them. If we need fewer, just | |
5325 | leave it alone. */ | |
5326 | if (regs->num_regs < num_regs + 1) | |
5327 | { | |
5328 | regs->num_regs = num_regs + 1; | |
5329 | RETALLOC (regs->start, regs->num_regs, regoff_t); | |
5330 | RETALLOC (regs->end, regs->num_regs, regoff_t); | |
5331 | if (regs->start == NULL || regs->end == NULL) | |
fa9a63c5 RM |
5332 | { |
5333 | FREE_VARIABLES (); | |
5334 | return -2; | |
5335 | } | |
25fe55af RS |
5336 | } |
5337 | } | |
5338 | else | |
fa9a63c5 RM |
5339 | { |
5340 | /* These braces fend off a "empty body in an else-statement" | |
7814e705 | 5341 | warning under GCC when assert expands to nothing. */ |
fa9a63c5 RM |
5342 | assert (bufp->regs_allocated == REGS_FIXED); |
5343 | } | |
5344 | ||
25fe55af RS |
5345 | /* Convert the pointer data in `regstart' and `regend' to |
5346 | indices. Register zero has to be set differently, | |
5347 | since we haven't kept track of any info for it. */ | |
5348 | if (regs->num_regs > 0) | |
5349 | { | |
5350 | regs->start[0] = pos; | |
99633e97 | 5351 | regs->end[0] = POINTER_TO_OFFSET (d); |
25fe55af | 5352 | } |
5e69f11e | 5353 | |
25fe55af RS |
5354 | /* Go through the first `min (num_regs, regs->num_regs)' |
5355 | registers, since that is all we initialized. */ | |
01618498 | 5356 | for (reg = 1; reg < MIN (num_regs, regs->num_regs); reg++) |
fa9a63c5 | 5357 | { |
01618498 SM |
5358 | if (REG_UNSET (regstart[reg]) || REG_UNSET (regend[reg])) |
5359 | regs->start[reg] = regs->end[reg] = -1; | |
25fe55af RS |
5360 | else |
5361 | { | |
01618498 SM |
5362 | regs->start[reg] |
5363 | = (regoff_t) POINTER_TO_OFFSET (regstart[reg]); | |
5364 | regs->end[reg] | |
5365 | = (regoff_t) POINTER_TO_OFFSET (regend[reg]); | |
25fe55af | 5366 | } |
fa9a63c5 | 5367 | } |
5e69f11e | 5368 | |
25fe55af RS |
5369 | /* If the regs structure we return has more elements than |
5370 | were in the pattern, set the extra elements to -1. If | |
5371 | we (re)allocated the registers, this is the case, | |
5372 | because we always allocate enough to have at least one | |
7814e705 | 5373 | -1 at the end. */ |
01618498 SM |
5374 | for (reg = num_regs; reg < regs->num_regs; reg++) |
5375 | regs->start[reg] = regs->end[reg] = -1; | |
fa9a63c5 RM |
5376 | } /* regs && !bufp->no_sub */ |
5377 | ||
25fe55af RS |
5378 | DEBUG_PRINT4 ("%u failure points pushed, %u popped (%u remain).\n", |
5379 | nfailure_points_pushed, nfailure_points_popped, | |
5380 | nfailure_points_pushed - nfailure_points_popped); | |
5381 | DEBUG_PRINT2 ("%u registers pushed.\n", num_regs_pushed); | |
fa9a63c5 | 5382 | |
99633e97 | 5383 | mcnt = POINTER_TO_OFFSET (d) - pos; |
fa9a63c5 | 5384 | |
25fe55af | 5385 | DEBUG_PRINT2 ("Returning %d from re_match_2.\n", mcnt); |
fa9a63c5 | 5386 | |
25fe55af RS |
5387 | FREE_VARIABLES (); |
5388 | return mcnt; | |
5389 | } | |
fa9a63c5 | 5390 | |
7814e705 | 5391 | /* Otherwise match next pattern command. */ |
fa9a63c5 RM |
5392 | switch (SWITCH_ENUM_CAST ((re_opcode_t) *p++)) |
5393 | { | |
25fe55af RS |
5394 | /* Ignore these. Used to ignore the n of succeed_n's which |
5395 | currently have n == 0. */ | |
5396 | case no_op: | |
5397 | DEBUG_PRINT1 ("EXECUTING no_op.\n"); | |
5398 | break; | |
fa9a63c5 RM |
5399 | |
5400 | case succeed: | |
25fe55af | 5401 | DEBUG_PRINT1 ("EXECUTING succeed.\n"); |
fa9a63c5 RM |
5402 | goto succeed_label; |
5403 | ||
7814e705 | 5404 | /* Match the next n pattern characters exactly. The following |
25fe55af | 5405 | byte in the pattern defines n, and the n bytes after that |
7814e705 | 5406 | are the characters to match. */ |
fa9a63c5 RM |
5407 | case exactn: |
5408 | mcnt = *p++; | |
25fe55af | 5409 | DEBUG_PRINT2 ("EXECUTING exactn %d.\n", mcnt); |
fa9a63c5 | 5410 | |
99633e97 SM |
5411 | /* Remember the start point to rollback upon failure. */ |
5412 | dfail = d; | |
5413 | ||
6fdd04b0 | 5414 | #ifndef emacs |
25fe55af RS |
5415 | /* This is written out as an if-else so we don't waste time |
5416 | testing `translate' inside the loop. */ | |
28703c16 | 5417 | if (RE_TRANSLATE_P (translate)) |
6fdd04b0 KH |
5418 | do |
5419 | { | |
5420 | PREFETCH (); | |
5421 | if (RE_TRANSLATE (translate, *d) != *p++) | |
e934739e | 5422 | { |
6fdd04b0 KH |
5423 | d = dfail; |
5424 | goto fail; | |
e934739e | 5425 | } |
6fdd04b0 KH |
5426 | d++; |
5427 | } | |
5428 | while (--mcnt); | |
fa9a63c5 | 5429 | else |
6fdd04b0 KH |
5430 | do |
5431 | { | |
5432 | PREFETCH (); | |
5433 | if (*d++ != *p++) | |
bf216479 | 5434 | { |
6fdd04b0 KH |
5435 | d = dfail; |
5436 | goto fail; | |
bf216479 | 5437 | } |
6fdd04b0 KH |
5438 | } |
5439 | while (--mcnt); | |
5440 | #else /* emacs */ | |
5441 | /* The cost of testing `translate' is comparatively small. */ | |
cf9c99bc | 5442 | if (target_multibyte) |
6fdd04b0 KH |
5443 | do |
5444 | { | |
5445 | int pat_charlen, buf_charlen; | |
cf9c99bc | 5446 | int pat_ch, buf_ch; |
e934739e | 5447 | |
6fdd04b0 | 5448 | PREFETCH (); |
cf9c99bc | 5449 | if (multibyte) |
62a6e103 | 5450 | pat_ch = STRING_CHAR_AND_LENGTH (p, pat_charlen); |
cf9c99bc KH |
5451 | else |
5452 | { | |
5453 | pat_ch = RE_CHAR_TO_MULTIBYTE (*p); | |
5454 | pat_charlen = 1; | |
5455 | } | |
62a6e103 | 5456 | buf_ch = STRING_CHAR_AND_LENGTH (d, buf_charlen); |
e934739e | 5457 | |
6fdd04b0 | 5458 | if (TRANSLATE (buf_ch) != pat_ch) |
e934739e | 5459 | { |
6fdd04b0 KH |
5460 | d = dfail; |
5461 | goto fail; | |
e934739e | 5462 | } |
bf216479 | 5463 | |
6fdd04b0 KH |
5464 | p += pat_charlen; |
5465 | d += buf_charlen; | |
5466 | mcnt -= pat_charlen; | |
5467 | } | |
5468 | while (mcnt > 0); | |
fa9a63c5 | 5469 | else |
6fdd04b0 KH |
5470 | do |
5471 | { | |
cf9c99bc KH |
5472 | int pat_charlen, buf_charlen; |
5473 | int pat_ch, buf_ch; | |
bf216479 | 5474 | |
6fdd04b0 | 5475 | PREFETCH (); |
cf9c99bc KH |
5476 | if (multibyte) |
5477 | { | |
62a6e103 | 5478 | pat_ch = STRING_CHAR_AND_LENGTH (p, pat_charlen); |
2afc21f5 | 5479 | pat_ch = RE_CHAR_TO_UNIBYTE (pat_ch); |
cf9c99bc KH |
5480 | } |
5481 | else | |
5482 | { | |
5483 | pat_ch = *p; | |
5484 | pat_charlen = 1; | |
5485 | } | |
5486 | buf_ch = RE_CHAR_TO_MULTIBYTE (*d); | |
5487 | if (! CHAR_BYTE8_P (buf_ch)) | |
5488 | { | |
5489 | buf_ch = TRANSLATE (buf_ch); | |
5490 | buf_ch = RE_CHAR_TO_UNIBYTE (buf_ch); | |
5491 | if (buf_ch < 0) | |
5492 | buf_ch = *d; | |
5493 | } | |
0e2501ed AS |
5494 | else |
5495 | buf_ch = *d; | |
cf9c99bc | 5496 | if (buf_ch != pat_ch) |
6fdd04b0 KH |
5497 | { |
5498 | d = dfail; | |
5499 | goto fail; | |
bf216479 | 5500 | } |
cf9c99bc KH |
5501 | p += pat_charlen; |
5502 | d++; | |
6fdd04b0 KH |
5503 | } |
5504 | while (--mcnt); | |
5505 | #endif | |
25fe55af | 5506 | break; |
fa9a63c5 RM |
5507 | |
5508 | ||
25fe55af | 5509 | /* Match any character except possibly a newline or a null. */ |
fa9a63c5 | 5510 | case anychar: |
e934739e RS |
5511 | { |
5512 | int buf_charlen; | |
01618498 | 5513 | re_wchar_t buf_ch; |
fa9a63c5 | 5514 | |
e934739e | 5515 | DEBUG_PRINT1 ("EXECUTING anychar.\n"); |
fa9a63c5 | 5516 | |
e934739e | 5517 | PREFETCH (); |
62a6e103 | 5518 | buf_ch = RE_STRING_CHAR_AND_LENGTH (d, buf_charlen, |
cf9c99bc | 5519 | target_multibyte); |
e934739e RS |
5520 | buf_ch = TRANSLATE (buf_ch); |
5521 | ||
5522 | if ((!(bufp->syntax & RE_DOT_NEWLINE) | |
5523 | && buf_ch == '\n') | |
5524 | || ((bufp->syntax & RE_DOT_NOT_NULL) | |
5525 | && buf_ch == '\000')) | |
5526 | goto fail; | |
5527 | ||
e934739e RS |
5528 | DEBUG_PRINT2 (" Matched `%d'.\n", *d); |
5529 | d += buf_charlen; | |
5530 | } | |
fa9a63c5 RM |
5531 | break; |
5532 | ||
5533 | ||
5534 | case charset: | |
5535 | case charset_not: | |
5536 | { | |
b18215fc | 5537 | register unsigned int c; |
fa9a63c5 | 5538 | boolean not = (re_opcode_t) *(p - 1) == charset_not; |
b18215fc RS |
5539 | int len; |
5540 | ||
5541 | /* Start of actual range_table, or end of bitmap if there is no | |
5542 | range table. */ | |
01618498 | 5543 | re_char *range_table; |
b18215fc | 5544 | |
96cc36cc | 5545 | /* Nonzero if there is a range table. */ |
b18215fc RS |
5546 | int range_table_exists; |
5547 | ||
96cc36cc RS |
5548 | /* Number of ranges of range table. This is not included |
5549 | in the initial byte-length of the command. */ | |
5550 | int count = 0; | |
fa9a63c5 | 5551 | |
f5020181 AS |
5552 | /* Whether matching against a unibyte character. */ |
5553 | boolean unibyte_char = false; | |
5554 | ||
25fe55af | 5555 | DEBUG_PRINT2 ("EXECUTING charset%s.\n", not ? "_not" : ""); |
fa9a63c5 | 5556 | |
b18215fc | 5557 | range_table_exists = CHARSET_RANGE_TABLE_EXISTS_P (&p[-1]); |
96cc36cc | 5558 | |
b18215fc | 5559 | if (range_table_exists) |
96cc36cc RS |
5560 | { |
5561 | range_table = CHARSET_RANGE_TABLE (&p[-1]); /* Past the bitmap. */ | |
5562 | EXTRACT_NUMBER_AND_INCR (count, range_table); | |
5563 | } | |
b18215fc | 5564 | |
2d1675e4 | 5565 | PREFETCH (); |
62a6e103 | 5566 | c = RE_STRING_CHAR_AND_LENGTH (d, len, target_multibyte); |
cf9c99bc KH |
5567 | if (target_multibyte) |
5568 | { | |
5569 | int c1; | |
b18215fc | 5570 | |
cf9c99bc KH |
5571 | c = TRANSLATE (c); |
5572 | c1 = RE_CHAR_TO_UNIBYTE (c); | |
5573 | if (c1 >= 0) | |
f5020181 AS |
5574 | { |
5575 | unibyte_char = true; | |
5576 | c = c1; | |
5577 | } | |
cf9c99bc KH |
5578 | } |
5579 | else | |
5580 | { | |
5581 | int c1 = RE_CHAR_TO_MULTIBYTE (c); | |
5582 | ||
5583 | if (! CHAR_BYTE8_P (c1)) | |
5584 | { | |
5585 | c1 = TRANSLATE (c1); | |
5586 | c1 = RE_CHAR_TO_UNIBYTE (c1); | |
5587 | if (c1 >= 0) | |
f5020181 AS |
5588 | { |
5589 | unibyte_char = true; | |
5590 | c = c1; | |
5591 | } | |
cf9c99bc | 5592 | } |
0b8be006 AS |
5593 | else |
5594 | unibyte_char = true; | |
cf9c99bc KH |
5595 | } |
5596 | ||
f5020181 | 5597 | if (unibyte_char && c < (1 << BYTEWIDTH)) |
b18215fc | 5598 | { /* Lookup bitmap. */ |
b18215fc RS |
5599 | /* Cast to `unsigned' instead of `unsigned char' in |
5600 | case the bit list is a full 32 bytes long. */ | |
5601 | if (c < (unsigned) (CHARSET_BITMAP_SIZE (&p[-1]) * BYTEWIDTH) | |
96cc36cc RS |
5602 | && p[1 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH))) |
5603 | not = !not; | |
b18215fc | 5604 | } |
96cc36cc | 5605 | #ifdef emacs |
b18215fc | 5606 | else if (range_table_exists) |
96cc36cc RS |
5607 | { |
5608 | int class_bits = CHARSET_RANGE_TABLE_BITS (&p[-1]); | |
5609 | ||
14473664 SM |
5610 | if ( (class_bits & BIT_LOWER && ISLOWER (c)) |
5611 | | (class_bits & BIT_MULTIBYTE) | |
96cc36cc RS |
5612 | | (class_bits & BIT_PUNCT && ISPUNCT (c)) |
5613 | | (class_bits & BIT_SPACE && ISSPACE (c)) | |
5614 | | (class_bits & BIT_UPPER && ISUPPER (c)) | |
5615 | | (class_bits & BIT_WORD && ISWORD (c))) | |
5616 | not = !not; | |
5617 | else | |
5618 | CHARSET_LOOKUP_RANGE_TABLE_RAW (not, c, range_table, count); | |
5619 | } | |
5620 | #endif /* emacs */ | |
fa9a63c5 | 5621 | |
96cc36cc RS |
5622 | if (range_table_exists) |
5623 | p = CHARSET_RANGE_TABLE_END (range_table, count); | |
5624 | else | |
5625 | p += CHARSET_BITMAP_SIZE (&p[-1]) + 1; | |
fa9a63c5 RM |
5626 | |
5627 | if (!not) goto fail; | |
5e69f11e | 5628 | |
b18215fc | 5629 | d += len; |
fa9a63c5 RM |
5630 | break; |
5631 | } | |
5632 | ||
5633 | ||
25fe55af | 5634 | /* The beginning of a group is represented by start_memory. |
505bde11 | 5635 | The argument is the register number. The text |
25fe55af | 5636 | matched within the group is recorded (in the internal |
7814e705 | 5637 | registers data structure) under the register number. */ |
25fe55af | 5638 | case start_memory: |
505bde11 SM |
5639 | DEBUG_PRINT2 ("EXECUTING start_memory %d:\n", *p); |
5640 | ||
5641 | /* In case we need to undo this operation (via backtracking). */ | |
5642 | PUSH_FAILURE_REG ((unsigned int)*p); | |
fa9a63c5 | 5643 | |
25fe55af | 5644 | regstart[*p] = d; |
4bb91c68 | 5645 | regend[*p] = NULL; /* probably unnecessary. -sm */ |
fa9a63c5 RM |
5646 | DEBUG_PRINT2 (" regstart: %d\n", POINTER_TO_OFFSET (regstart[*p])); |
5647 | ||
25fe55af | 5648 | /* Move past the register number and inner group count. */ |
505bde11 | 5649 | p += 1; |
25fe55af | 5650 | break; |
fa9a63c5 RM |
5651 | |
5652 | ||
25fe55af | 5653 | /* The stop_memory opcode represents the end of a group. Its |
505bde11 | 5654 | argument is the same as start_memory's: the register number. */ |
fa9a63c5 | 5655 | case stop_memory: |
505bde11 SM |
5656 | DEBUG_PRINT2 ("EXECUTING stop_memory %d:\n", *p); |
5657 | ||
5658 | assert (!REG_UNSET (regstart[*p])); | |
5659 | /* Strictly speaking, there should be code such as: | |
177c0ea7 | 5660 | |
0b32bf0e | 5661 | assert (REG_UNSET (regend[*p])); |
505bde11 SM |
5662 | PUSH_FAILURE_REGSTOP ((unsigned int)*p); |
5663 | ||
5664 | But the only info to be pushed is regend[*p] and it is known to | |
5665 | be UNSET, so there really isn't anything to push. | |
5666 | Not pushing anything, on the other hand deprives us from the | |
5667 | guarantee that regend[*p] is UNSET since undoing this operation | |
5668 | will not reset its value properly. This is not important since | |
5669 | the value will only be read on the next start_memory or at | |
5670 | the very end and both events can only happen if this stop_memory | |
5671 | is *not* undone. */ | |
fa9a63c5 | 5672 | |
25fe55af | 5673 | regend[*p] = d; |
fa9a63c5 RM |
5674 | DEBUG_PRINT2 (" regend: %d\n", POINTER_TO_OFFSET (regend[*p])); |
5675 | ||
25fe55af | 5676 | /* Move past the register number and the inner group count. */ |
505bde11 | 5677 | p += 1; |
25fe55af | 5678 | break; |
fa9a63c5 RM |
5679 | |
5680 | ||
5681 | /* \<digit> has been turned into a `duplicate' command which is | |
25fe55af RS |
5682 | followed by the numeric value of <digit> as the register number. */ |
5683 | case duplicate: | |
fa9a63c5 | 5684 | { |
66f0296e | 5685 | register re_char *d2, *dend2; |
7814e705 | 5686 | int regno = *p++; /* Get which register to match against. */ |
fa9a63c5 RM |
5687 | DEBUG_PRINT2 ("EXECUTING duplicate %d.\n", regno); |
5688 | ||
7814e705 | 5689 | /* Can't back reference a group which we've never matched. */ |
25fe55af RS |
5690 | if (REG_UNSET (regstart[regno]) || REG_UNSET (regend[regno])) |
5691 | goto fail; | |
5e69f11e | 5692 | |
7814e705 | 5693 | /* Where in input to try to start matching. */ |
25fe55af | 5694 | d2 = regstart[regno]; |
5e69f11e | 5695 | |
99633e97 SM |
5696 | /* Remember the start point to rollback upon failure. */ |
5697 | dfail = d; | |
5698 | ||
25fe55af RS |
5699 | /* Where to stop matching; if both the place to start and |
5700 | the place to stop matching are in the same string, then | |
5701 | set to the place to stop, otherwise, for now have to use | |
5702 | the end of the first string. */ | |
fa9a63c5 | 5703 | |
25fe55af | 5704 | dend2 = ((FIRST_STRING_P (regstart[regno]) |
fa9a63c5 RM |
5705 | == FIRST_STRING_P (regend[regno])) |
5706 | ? regend[regno] : end_match_1); | |
5707 | for (;;) | |
5708 | { | |
5709 | /* If necessary, advance to next segment in register | |
25fe55af | 5710 | contents. */ |
fa9a63c5 RM |
5711 | while (d2 == dend2) |
5712 | { | |
5713 | if (dend2 == end_match_2) break; | |
5714 | if (dend2 == regend[regno]) break; | |
5715 | ||
25fe55af RS |
5716 | /* End of string1 => advance to string2. */ |
5717 | d2 = string2; | |
5718 | dend2 = regend[regno]; | |
fa9a63c5 RM |
5719 | } |
5720 | /* At end of register contents => success */ | |
5721 | if (d2 == dend2) break; | |
5722 | ||
5723 | /* If necessary, advance to next segment in data. */ | |
5724 | PREFETCH (); | |
5725 | ||
5726 | /* How many characters left in this segment to match. */ | |
5727 | mcnt = dend - d; | |
5e69f11e | 5728 | |
fa9a63c5 | 5729 | /* Want how many consecutive characters we can match in |
25fe55af RS |
5730 | one shot, so, if necessary, adjust the count. */ |
5731 | if (mcnt > dend2 - d2) | |
fa9a63c5 | 5732 | mcnt = dend2 - d2; |
5e69f11e | 5733 | |
fa9a63c5 | 5734 | /* Compare that many; failure if mismatch, else move |
25fe55af | 5735 | past them. */ |
28703c16 | 5736 | if (RE_TRANSLATE_P (translate) |
02cb78b5 | 5737 | ? bcmp_translate (d, d2, mcnt, translate, target_multibyte) |
4bb91c68 | 5738 | : memcmp (d, d2, mcnt)) |
99633e97 SM |
5739 | { |
5740 | d = dfail; | |
5741 | goto fail; | |
5742 | } | |
fa9a63c5 | 5743 | d += mcnt, d2 += mcnt; |
fa9a63c5 RM |
5744 | } |
5745 | } | |
5746 | break; | |
5747 | ||
5748 | ||
25fe55af | 5749 | /* begline matches the empty string at the beginning of the string |
c0f9ea08 | 5750 | (unless `not_bol' is set in `bufp'), and after newlines. */ |
fa9a63c5 | 5751 | case begline: |
25fe55af | 5752 | DEBUG_PRINT1 ("EXECUTING begline.\n"); |
5e69f11e | 5753 | |
25fe55af RS |
5754 | if (AT_STRINGS_BEG (d)) |
5755 | { | |
5756 | if (!bufp->not_bol) break; | |
5757 | } | |
419d1c74 | 5758 | else |
25fe55af | 5759 | { |
bf216479 | 5760 | unsigned c; |
419d1c74 | 5761 | GET_CHAR_BEFORE_2 (c, d, string1, end1, string2, end2); |
c0f9ea08 | 5762 | if (c == '\n') |
419d1c74 | 5763 | break; |
25fe55af RS |
5764 | } |
5765 | /* In all other cases, we fail. */ | |
5766 | goto fail; | |
fa9a63c5 RM |
5767 | |
5768 | ||
25fe55af | 5769 | /* endline is the dual of begline. */ |
fa9a63c5 | 5770 | case endline: |
25fe55af | 5771 | DEBUG_PRINT1 ("EXECUTING endline.\n"); |
fa9a63c5 | 5772 | |
25fe55af RS |
5773 | if (AT_STRINGS_END (d)) |
5774 | { | |
5775 | if (!bufp->not_eol) break; | |
5776 | } | |
f1ad044f | 5777 | else |
25fe55af | 5778 | { |
f1ad044f | 5779 | PREFETCH_NOLIMIT (); |
c0f9ea08 | 5780 | if (*d == '\n') |
f1ad044f | 5781 | break; |
25fe55af RS |
5782 | } |
5783 | goto fail; | |
fa9a63c5 RM |
5784 | |
5785 | ||
5786 | /* Match at the very beginning of the data. */ | |
25fe55af RS |
5787 | case begbuf: |
5788 | DEBUG_PRINT1 ("EXECUTING begbuf.\n"); | |
5789 | if (AT_STRINGS_BEG (d)) | |
5790 | break; | |
5791 | goto fail; | |
fa9a63c5 RM |
5792 | |
5793 | ||
5794 | /* Match at the very end of the data. */ | |
25fe55af RS |
5795 | case endbuf: |
5796 | DEBUG_PRINT1 ("EXECUTING endbuf.\n"); | |
fa9a63c5 RM |
5797 | if (AT_STRINGS_END (d)) |
5798 | break; | |
25fe55af | 5799 | goto fail; |
5e69f11e | 5800 | |
5e69f11e | 5801 | |
25fe55af RS |
5802 | /* on_failure_keep_string_jump is used to optimize `.*\n'. It |
5803 | pushes NULL as the value for the string on the stack. Then | |
505bde11 | 5804 | `POP_FAILURE_POINT' will keep the current value for the |
25fe55af | 5805 | string, instead of restoring it. To see why, consider |
7814e705 | 5806 | matching `foo\nbar' against `.*\n'. The .* matches the foo; |
25fe55af RS |
5807 | then the . fails against the \n. But the next thing we want |
5808 | to do is match the \n against the \n; if we restored the | |
5809 | string value, we would be back at the foo. | |
5810 | ||
5811 | Because this is used only in specific cases, we don't need to | |
5812 | check all the things that `on_failure_jump' does, to make | |
5813 | sure the right things get saved on the stack. Hence we don't | |
5814 | share its code. The only reason to push anything on the | |
5815 | stack at all is that otherwise we would have to change | |
5816 | `anychar's code to do something besides goto fail in this | |
5817 | case; that seems worse than this. */ | |
5818 | case on_failure_keep_string_jump: | |
505bde11 SM |
5819 | EXTRACT_NUMBER_AND_INCR (mcnt, p); |
5820 | DEBUG_PRINT3 ("EXECUTING on_failure_keep_string_jump %d (to %p):\n", | |
5821 | mcnt, p + mcnt); | |
fa9a63c5 | 5822 | |
505bde11 SM |
5823 | PUSH_FAILURE_POINT (p - 3, NULL); |
5824 | break; | |
5825 | ||
0683b6fa SM |
5826 | /* A nasty loop is introduced by the non-greedy *? and +?. |
5827 | With such loops, the stack only ever contains one failure point | |
5828 | at a time, so that a plain on_failure_jump_loop kind of | |
5829 | cycle detection cannot work. Worse yet, such a detection | |
5830 | can not only fail to detect a cycle, but it can also wrongly | |
5831 | detect a cycle (between different instantiations of the same | |
6df42991 | 5832 | loop). |
0683b6fa SM |
5833 | So the method used for those nasty loops is a little different: |
5834 | We use a special cycle-detection-stack-frame which is pushed | |
5835 | when the on_failure_jump_nastyloop failure-point is *popped*. | |
5836 | This special frame thus marks the beginning of one iteration | |
5837 | through the loop and we can hence easily check right here | |
5838 | whether something matched between the beginning and the end of | |
5839 | the loop. */ | |
5840 | case on_failure_jump_nastyloop: | |
5841 | EXTRACT_NUMBER_AND_INCR (mcnt, p); | |
5842 | DEBUG_PRINT3 ("EXECUTING on_failure_jump_nastyloop %d (to %p):\n", | |
5843 | mcnt, p + mcnt); | |
5844 | ||
5845 | assert ((re_opcode_t)p[-4] == no_op); | |
6df42991 SM |
5846 | { |
5847 | int cycle = 0; | |
5848 | CHECK_INFINITE_LOOP (p - 4, d); | |
5849 | if (!cycle) | |
5850 | /* If there's a cycle, just continue without pushing | |
5851 | this failure point. The failure point is the "try again" | |
5852 | option, which shouldn't be tried. | |
5853 | We want (x?)*?y\1z to match both xxyz and xxyxz. */ | |
5854 | PUSH_FAILURE_POINT (p - 3, d); | |
5855 | } | |
0683b6fa SM |
5856 | break; |
5857 | ||
4e8a9132 SM |
5858 | /* Simple loop detecting on_failure_jump: just check on the |
5859 | failure stack if the same spot was already hit earlier. */ | |
505bde11 SM |
5860 | case on_failure_jump_loop: |
5861 | on_failure: | |
5862 | EXTRACT_NUMBER_AND_INCR (mcnt, p); | |
5863 | DEBUG_PRINT3 ("EXECUTING on_failure_jump_loop %d (to %p):\n", | |
5864 | mcnt, p + mcnt); | |
6df42991 SM |
5865 | { |
5866 | int cycle = 0; | |
5867 | CHECK_INFINITE_LOOP (p - 3, d); | |
5868 | if (cycle) | |
5869 | /* If there's a cycle, get out of the loop, as if the matching | |
5870 | had failed. We used to just `goto fail' here, but that was | |
5871 | aborting the search a bit too early: we want to keep the | |
5872 | empty-loop-match and keep matching after the loop. | |
5873 | We want (x?)*y\1z to match both xxyz and xxyxz. */ | |
5874 | p += mcnt; | |
5875 | else | |
5876 | PUSH_FAILURE_POINT (p - 3, d); | |
5877 | } | |
25fe55af | 5878 | break; |
fa9a63c5 RM |
5879 | |
5880 | ||
5881 | /* Uses of on_failure_jump: | |
5e69f11e | 5882 | |
25fe55af RS |
5883 | Each alternative starts with an on_failure_jump that points |
5884 | to the beginning of the next alternative. Each alternative | |
5885 | except the last ends with a jump that in effect jumps past | |
5886 | the rest of the alternatives. (They really jump to the | |
5887 | ending jump of the following alternative, because tensioning | |
5888 | these jumps is a hassle.) | |
fa9a63c5 | 5889 | |
25fe55af RS |
5890 | Repeats start with an on_failure_jump that points past both |
5891 | the repetition text and either the following jump or | |
5892 | pop_failure_jump back to this on_failure_jump. */ | |
fa9a63c5 | 5893 | case on_failure_jump: |
25fe55af | 5894 | EXTRACT_NUMBER_AND_INCR (mcnt, p); |
505bde11 SM |
5895 | DEBUG_PRINT3 ("EXECUTING on_failure_jump %d (to %p):\n", |
5896 | mcnt, p + mcnt); | |
25fe55af | 5897 | |
505bde11 | 5898 | PUSH_FAILURE_POINT (p -3, d); |
25fe55af RS |
5899 | break; |
5900 | ||
4e8a9132 | 5901 | /* This operation is used for greedy *. |
505bde11 SM |
5902 | Compare the beginning of the repeat with what in the |
5903 | pattern follows its end. If we can establish that there | |
5904 | is nothing that they would both match, i.e., that we | |
5905 | would have to backtrack because of (as in, e.g., `a*a') | |
5906 | then we can use a non-backtracking loop based on | |
4e8a9132 | 5907 | on_failure_keep_string_jump instead of on_failure_jump. */ |
505bde11 | 5908 | case on_failure_jump_smart: |
25fe55af | 5909 | EXTRACT_NUMBER_AND_INCR (mcnt, p); |
505bde11 SM |
5910 | DEBUG_PRINT3 ("EXECUTING on_failure_jump_smart %d (to %p).\n", |
5911 | mcnt, p + mcnt); | |
25fe55af | 5912 | { |
01618498 | 5913 | re_char *p1 = p; /* Next operation. */ |
6dcf2d0e SM |
5914 | /* Here, we discard `const', making re_match non-reentrant. */ |
5915 | unsigned char *p2 = (unsigned char*) p + mcnt; /* Jump dest. */ | |
5916 | unsigned char *p3 = (unsigned char*) p - 3; /* opcode location. */ | |
fa9a63c5 | 5917 | |
505bde11 SM |
5918 | p -= 3; /* Reset so that we will re-execute the |
5919 | instruction once it's been changed. */ | |
fa9a63c5 | 5920 | |
4e8a9132 SM |
5921 | EXTRACT_NUMBER (mcnt, p2 - 2); |
5922 | ||
5923 | /* Ensure this is a indeed the trivial kind of loop | |
5924 | we are expecting. */ | |
5925 | assert (skip_one_char (p1) == p2 - 3); | |
5926 | assert ((re_opcode_t) p2[-3] == jump && p2 + mcnt == p); | |
99633e97 | 5927 | DEBUG_STATEMENT (debug += 2); |
505bde11 | 5928 | if (mutually_exclusive_p (bufp, p1, p2)) |
fa9a63c5 | 5929 | { |
505bde11 | 5930 | /* Use a fast `on_failure_keep_string_jump' loop. */ |
4e8a9132 | 5931 | DEBUG_PRINT1 (" smart exclusive => fast loop.\n"); |
01618498 | 5932 | *p3 = (unsigned char) on_failure_keep_string_jump; |
4e8a9132 | 5933 | STORE_NUMBER (p2 - 2, mcnt + 3); |
25fe55af | 5934 | } |
505bde11 | 5935 | else |
fa9a63c5 | 5936 | { |
505bde11 SM |
5937 | /* Default to a safe `on_failure_jump' loop. */ |
5938 | DEBUG_PRINT1 (" smart default => slow loop.\n"); | |
01618498 | 5939 | *p3 = (unsigned char) on_failure_jump; |
fa9a63c5 | 5940 | } |
99633e97 | 5941 | DEBUG_STATEMENT (debug -= 2); |
25fe55af | 5942 | } |
505bde11 | 5943 | break; |
25fe55af RS |
5944 | |
5945 | /* Unconditionally jump (without popping any failure points). */ | |
5946 | case jump: | |
fa9a63c5 | 5947 | unconditional_jump: |
5b370c2b | 5948 | IMMEDIATE_QUIT_CHECK; |
fa9a63c5 | 5949 | EXTRACT_NUMBER_AND_INCR (mcnt, p); /* Get the amount to jump. */ |
25fe55af | 5950 | DEBUG_PRINT2 ("EXECUTING jump %d ", mcnt); |
7814e705 | 5951 | p += mcnt; /* Do the jump. */ |
505bde11 | 5952 | DEBUG_PRINT2 ("(to %p).\n", p); |
25fe55af RS |
5953 | break; |
5954 | ||
5955 | ||
25fe55af RS |
5956 | /* Have to succeed matching what follows at least n times. |
5957 | After that, handle like `on_failure_jump'. */ | |
5958 | case succeed_n: | |
01618498 | 5959 | /* Signedness doesn't matter since we only compare MCNT to 0. */ |
25fe55af RS |
5960 | EXTRACT_NUMBER (mcnt, p + 2); |
5961 | DEBUG_PRINT2 ("EXECUTING succeed_n %d.\n", mcnt); | |
5e69f11e | 5962 | |
dc1e502d SM |
5963 | /* Originally, mcnt is how many times we HAVE to succeed. */ |
5964 | if (mcnt != 0) | |
25fe55af | 5965 | { |
6dcf2d0e SM |
5966 | /* Here, we discard `const', making re_match non-reentrant. */ |
5967 | unsigned char *p2 = (unsigned char*) p + 2; /* counter loc. */ | |
dc1e502d | 5968 | mcnt--; |
01618498 SM |
5969 | p += 4; |
5970 | PUSH_NUMBER (p2, mcnt); | |
25fe55af | 5971 | } |
dc1e502d SM |
5972 | else |
5973 | /* The two bytes encoding mcnt == 0 are two no_op opcodes. */ | |
5974 | goto on_failure; | |
25fe55af RS |
5975 | break; |
5976 | ||
5977 | case jump_n: | |
01618498 | 5978 | /* Signedness doesn't matter since we only compare MCNT to 0. */ |
25fe55af RS |
5979 | EXTRACT_NUMBER (mcnt, p + 2); |
5980 | DEBUG_PRINT2 ("EXECUTING jump_n %d.\n", mcnt); | |
5981 | ||
5982 | /* Originally, this is how many times we CAN jump. */ | |
dc1e502d | 5983 | if (mcnt != 0) |
25fe55af | 5984 | { |
6dcf2d0e SM |
5985 | /* Here, we discard `const', making re_match non-reentrant. */ |
5986 | unsigned char *p2 = (unsigned char*) p + 2; /* counter loc. */ | |
dc1e502d | 5987 | mcnt--; |
01618498 | 5988 | PUSH_NUMBER (p2, mcnt); |
dc1e502d | 5989 | goto unconditional_jump; |
25fe55af RS |
5990 | } |
5991 | /* If don't have to jump any more, skip over the rest of command. */ | |
5e69f11e RM |
5992 | else |
5993 | p += 4; | |
25fe55af | 5994 | break; |
5e69f11e | 5995 | |
fa9a63c5 RM |
5996 | case set_number_at: |
5997 | { | |
01618498 | 5998 | unsigned char *p2; /* Location of the counter. */ |
25fe55af | 5999 | DEBUG_PRINT1 ("EXECUTING set_number_at.\n"); |
fa9a63c5 | 6000 | |
25fe55af | 6001 | EXTRACT_NUMBER_AND_INCR (mcnt, p); |
6dcf2d0e SM |
6002 | /* Here, we discard `const', making re_match non-reentrant. */ |
6003 | p2 = (unsigned char*) p + mcnt; | |
01618498 | 6004 | /* Signedness doesn't matter since we only copy MCNT's bits . */ |
25fe55af | 6005 | EXTRACT_NUMBER_AND_INCR (mcnt, p); |
01618498 SM |
6006 | DEBUG_PRINT3 (" Setting %p to %d.\n", p2, mcnt); |
6007 | PUSH_NUMBER (p2, mcnt); | |
25fe55af RS |
6008 | break; |
6009 | } | |
9121ca40 KH |
6010 | |
6011 | case wordbound: | |
66f0296e SM |
6012 | case notwordbound: |
6013 | not = (re_opcode_t) *(p - 1) == notwordbound; | |
6014 | DEBUG_PRINT2 ("EXECUTING %swordbound.\n", not?"not":""); | |
fa9a63c5 | 6015 | |
99633e97 | 6016 | /* We SUCCEED (or FAIL) in one of the following cases: */ |
9121ca40 | 6017 | |
b18215fc | 6018 | /* Case 1: D is at the beginning or the end of string. */ |
9121ca40 | 6019 | if (AT_STRINGS_BEG (d) || AT_STRINGS_END (d)) |
66f0296e | 6020 | not = !not; |
b18215fc RS |
6021 | else |
6022 | { | |
6023 | /* C1 is the character before D, S1 is the syntax of C1, C2 | |
6024 | is the character at D, and S2 is the syntax of C2. */ | |
01618498 SM |
6025 | re_wchar_t c1, c2; |
6026 | int s1, s2; | |
bf216479 | 6027 | int dummy; |
b18215fc | 6028 | #ifdef emacs |
2d1675e4 SM |
6029 | int offset = PTR_TO_OFFSET (d - 1); |
6030 | int charpos = SYNTAX_TABLE_BYTE_TO_CHAR (offset); | |
5d967c7a | 6031 | UPDATE_SYNTAX_TABLE (charpos); |
25fe55af | 6032 | #endif |
66f0296e | 6033 | GET_CHAR_BEFORE_2 (c1, d, string1, end1, string2, end2); |
b18215fc RS |
6034 | s1 = SYNTAX (c1); |
6035 | #ifdef emacs | |
5d967c7a | 6036 | UPDATE_SYNTAX_TABLE_FORWARD (charpos + 1); |
25fe55af | 6037 | #endif |
f1ad044f | 6038 | PREFETCH_NOLIMIT (); |
6fdd04b0 | 6039 | GET_CHAR_AFTER (c2, d, dummy); |
b18215fc RS |
6040 | s2 = SYNTAX (c2); |
6041 | ||
6042 | if (/* Case 2: Only one of S1 and S2 is Sword. */ | |
6043 | ((s1 == Sword) != (s2 == Sword)) | |
6044 | /* Case 3: Both of S1 and S2 are Sword, and macro | |
7814e705 | 6045 | WORD_BOUNDARY_P (C1, C2) returns nonzero. */ |
b18215fc | 6046 | || ((s1 == Sword) && WORD_BOUNDARY_P (c1, c2))) |
66f0296e SM |
6047 | not = !not; |
6048 | } | |
6049 | if (not) | |
9121ca40 | 6050 | break; |
b18215fc | 6051 | else |
9121ca40 | 6052 | goto fail; |
fa9a63c5 RM |
6053 | |
6054 | case wordbeg: | |
25fe55af | 6055 | DEBUG_PRINT1 ("EXECUTING wordbeg.\n"); |
fa9a63c5 | 6056 | |
b18215fc RS |
6057 | /* We FAIL in one of the following cases: */ |
6058 | ||
7814e705 | 6059 | /* Case 1: D is at the end of string. */ |
b18215fc | 6060 | if (AT_STRINGS_END (d)) |
99633e97 | 6061 | goto fail; |
b18215fc RS |
6062 | else |
6063 | { | |
6064 | /* C1 is the character before D, S1 is the syntax of C1, C2 | |
6065 | is the character at D, and S2 is the syntax of C2. */ | |
01618498 SM |
6066 | re_wchar_t c1, c2; |
6067 | int s1, s2; | |
bf216479 | 6068 | int dummy; |
fa9a63c5 | 6069 | #ifdef emacs |
2d1675e4 SM |
6070 | int offset = PTR_TO_OFFSET (d); |
6071 | int charpos = SYNTAX_TABLE_BYTE_TO_CHAR (offset); | |
92432794 | 6072 | UPDATE_SYNTAX_TABLE (charpos); |
25fe55af | 6073 | #endif |
99633e97 | 6074 | PREFETCH (); |
6fdd04b0 | 6075 | GET_CHAR_AFTER (c2, d, dummy); |
b18215fc | 6076 | s2 = SYNTAX (c2); |
177c0ea7 | 6077 | |
b18215fc RS |
6078 | /* Case 2: S2 is not Sword. */ |
6079 | if (s2 != Sword) | |
6080 | goto fail; | |
6081 | ||
6082 | /* Case 3: D is not at the beginning of string ... */ | |
6083 | if (!AT_STRINGS_BEG (d)) | |
6084 | { | |
6085 | GET_CHAR_BEFORE_2 (c1, d, string1, end1, string2, end2); | |
6086 | #ifdef emacs | |
5d967c7a | 6087 | UPDATE_SYNTAX_TABLE_BACKWARD (charpos - 1); |
25fe55af | 6088 | #endif |
b18215fc RS |
6089 | s1 = SYNTAX (c1); |
6090 | ||
6091 | /* ... and S1 is Sword, and WORD_BOUNDARY_P (C1, C2) | |
7814e705 | 6092 | returns 0. */ |
b18215fc RS |
6093 | if ((s1 == Sword) && !WORD_BOUNDARY_P (c1, c2)) |
6094 | goto fail; | |
6095 | } | |
6096 | } | |
e318085a RS |
6097 | break; |
6098 | ||
b18215fc | 6099 | case wordend: |
25fe55af | 6100 | DEBUG_PRINT1 ("EXECUTING wordend.\n"); |
b18215fc RS |
6101 | |
6102 | /* We FAIL in one of the following cases: */ | |
6103 | ||
6104 | /* Case 1: D is at the beginning of string. */ | |
6105 | if (AT_STRINGS_BEG (d)) | |
e318085a | 6106 | goto fail; |
b18215fc RS |
6107 | else |
6108 | { | |
6109 | /* C1 is the character before D, S1 is the syntax of C1, C2 | |
6110 | is the character at D, and S2 is the syntax of C2. */ | |
01618498 SM |
6111 | re_wchar_t c1, c2; |
6112 | int s1, s2; | |
bf216479 | 6113 | int dummy; |
5d967c7a | 6114 | #ifdef emacs |
2d1675e4 SM |
6115 | int offset = PTR_TO_OFFSET (d) - 1; |
6116 | int charpos = SYNTAX_TABLE_BYTE_TO_CHAR (offset); | |
92432794 | 6117 | UPDATE_SYNTAX_TABLE (charpos); |
5d967c7a | 6118 | #endif |
99633e97 | 6119 | GET_CHAR_BEFORE_2 (c1, d, string1, end1, string2, end2); |
b18215fc RS |
6120 | s1 = SYNTAX (c1); |
6121 | ||
6122 | /* Case 2: S1 is not Sword. */ | |
6123 | if (s1 != Sword) | |
6124 | goto fail; | |
6125 | ||
6126 | /* Case 3: D is not at the end of string ... */ | |
6127 | if (!AT_STRINGS_END (d)) | |
6128 | { | |
f1ad044f | 6129 | PREFETCH_NOLIMIT (); |
6fdd04b0 | 6130 | GET_CHAR_AFTER (c2, d, dummy); |
5d967c7a RS |
6131 | #ifdef emacs |
6132 | UPDATE_SYNTAX_TABLE_FORWARD (charpos); | |
6133 | #endif | |
b18215fc RS |
6134 | s2 = SYNTAX (c2); |
6135 | ||
6136 | /* ... and S2 is Sword, and WORD_BOUNDARY_P (C1, C2) | |
7814e705 | 6137 | returns 0. */ |
b18215fc | 6138 | if ((s2 == Sword) && !WORD_BOUNDARY_P (c1, c2)) |
25fe55af | 6139 | goto fail; |
b18215fc RS |
6140 | } |
6141 | } | |
e318085a RS |
6142 | break; |
6143 | ||
669fa600 SM |
6144 | case symbeg: |
6145 | DEBUG_PRINT1 ("EXECUTING symbeg.\n"); | |
6146 | ||
6147 | /* We FAIL in one of the following cases: */ | |
6148 | ||
7814e705 | 6149 | /* Case 1: D is at the end of string. */ |
669fa600 SM |
6150 | if (AT_STRINGS_END (d)) |
6151 | goto fail; | |
6152 | else | |
6153 | { | |
6154 | /* C1 is the character before D, S1 is the syntax of C1, C2 | |
6155 | is the character at D, and S2 is the syntax of C2. */ | |
6156 | re_wchar_t c1, c2; | |
6157 | int s1, s2; | |
6158 | #ifdef emacs | |
6159 | int offset = PTR_TO_OFFSET (d); | |
6160 | int charpos = SYNTAX_TABLE_BYTE_TO_CHAR (offset); | |
6161 | UPDATE_SYNTAX_TABLE (charpos); | |
6162 | #endif | |
6163 | PREFETCH (); | |
62a6e103 | 6164 | c2 = RE_STRING_CHAR (d, target_multibyte); |
669fa600 | 6165 | s2 = SYNTAX (c2); |
7814e705 | 6166 | |
669fa600 SM |
6167 | /* Case 2: S2 is neither Sword nor Ssymbol. */ |
6168 | if (s2 != Sword && s2 != Ssymbol) | |
6169 | goto fail; | |
6170 | ||
6171 | /* Case 3: D is not at the beginning of string ... */ | |
6172 | if (!AT_STRINGS_BEG (d)) | |
6173 | { | |
6174 | GET_CHAR_BEFORE_2 (c1, d, string1, end1, string2, end2); | |
6175 | #ifdef emacs | |
6176 | UPDATE_SYNTAX_TABLE_BACKWARD (charpos - 1); | |
6177 | #endif | |
6178 | s1 = SYNTAX (c1); | |
6179 | ||
6180 | /* ... and S1 is Sword or Ssymbol. */ | |
6181 | if (s1 == Sword || s1 == Ssymbol) | |
6182 | goto fail; | |
6183 | } | |
6184 | } | |
6185 | break; | |
6186 | ||
6187 | case symend: | |
6188 | DEBUG_PRINT1 ("EXECUTING symend.\n"); | |
6189 | ||
6190 | /* We FAIL in one of the following cases: */ | |
6191 | ||
6192 | /* Case 1: D is at the beginning of string. */ | |
6193 | if (AT_STRINGS_BEG (d)) | |
6194 | goto fail; | |
6195 | else | |
6196 | { | |
6197 | /* C1 is the character before D, S1 is the syntax of C1, C2 | |
6198 | is the character at D, and S2 is the syntax of C2. */ | |
6199 | re_wchar_t c1, c2; | |
6200 | int s1, s2; | |
6201 | #ifdef emacs | |
6202 | int offset = PTR_TO_OFFSET (d) - 1; | |
6203 | int charpos = SYNTAX_TABLE_BYTE_TO_CHAR (offset); | |
6204 | UPDATE_SYNTAX_TABLE (charpos); | |
6205 | #endif | |
6206 | GET_CHAR_BEFORE_2 (c1, d, string1, end1, string2, end2); | |
6207 | s1 = SYNTAX (c1); | |
6208 | ||
6209 | /* Case 2: S1 is neither Ssymbol nor Sword. */ | |
6210 | if (s1 != Sword && s1 != Ssymbol) | |
6211 | goto fail; | |
6212 | ||
6213 | /* Case 3: D is not at the end of string ... */ | |
6214 | if (!AT_STRINGS_END (d)) | |
6215 | { | |
6216 | PREFETCH_NOLIMIT (); | |
62a6e103 | 6217 | c2 = RE_STRING_CHAR (d, target_multibyte); |
669fa600 | 6218 | #ifdef emacs |
134579f2 | 6219 | UPDATE_SYNTAX_TABLE_FORWARD (charpos + 1); |
669fa600 SM |
6220 | #endif |
6221 | s2 = SYNTAX (c2); | |
6222 | ||
6223 | /* ... and S2 is Sword or Ssymbol. */ | |
6224 | if (s2 == Sword || s2 == Ssymbol) | |
6225 | goto fail; | |
b18215fc RS |
6226 | } |
6227 | } | |
e318085a RS |
6228 | break; |
6229 | ||
fa9a63c5 | 6230 | case syntaxspec: |
1fb352e0 SM |
6231 | case notsyntaxspec: |
6232 | not = (re_opcode_t) *(p - 1) == notsyntaxspec; | |
fa9a63c5 | 6233 | mcnt = *p++; |
1fb352e0 | 6234 | DEBUG_PRINT3 ("EXECUTING %ssyntaxspec %d.\n", not?"not":"", mcnt); |
fa9a63c5 | 6235 | PREFETCH (); |
b18215fc RS |
6236 | #ifdef emacs |
6237 | { | |
2d1675e4 SM |
6238 | int offset = PTR_TO_OFFSET (d); |
6239 | int pos1 = SYNTAX_TABLE_BYTE_TO_CHAR (offset); | |
b18215fc RS |
6240 | UPDATE_SYNTAX_TABLE (pos1); |
6241 | } | |
25fe55af | 6242 | #endif |
b18215fc | 6243 | { |
01618498 SM |
6244 | int len; |
6245 | re_wchar_t c; | |
b18215fc | 6246 | |
6fdd04b0 | 6247 | GET_CHAR_AFTER (c, d, len); |
990b2375 | 6248 | if ((SYNTAX (c) != (enum syntaxcode) mcnt) ^ not) |
1fb352e0 | 6249 | goto fail; |
b18215fc RS |
6250 | d += len; |
6251 | } | |
fa9a63c5 RM |
6252 | break; |
6253 | ||
b18215fc | 6254 | #ifdef emacs |
1fb352e0 SM |
6255 | case before_dot: |
6256 | DEBUG_PRINT1 ("EXECUTING before_dot.\n"); | |
6257 | if (PTR_BYTE_POS (d) >= PT_BYTE) | |
fa9a63c5 | 6258 | goto fail; |
b18215fc RS |
6259 | break; |
6260 | ||
1fb352e0 SM |
6261 | case at_dot: |
6262 | DEBUG_PRINT1 ("EXECUTING at_dot.\n"); | |
6263 | if (PTR_BYTE_POS (d) != PT_BYTE) | |
6264 | goto fail; | |
6265 | break; | |
b18215fc | 6266 | |
1fb352e0 SM |
6267 | case after_dot: |
6268 | DEBUG_PRINT1 ("EXECUTING after_dot.\n"); | |
6269 | if (PTR_BYTE_POS (d) <= PT_BYTE) | |
6270 | goto fail; | |
e318085a | 6271 | break; |
fa9a63c5 | 6272 | |
1fb352e0 | 6273 | case categoryspec: |
b18215fc | 6274 | case notcategoryspec: |
1fb352e0 | 6275 | not = (re_opcode_t) *(p - 1) == notcategoryspec; |
b18215fc | 6276 | mcnt = *p++; |
1fb352e0 | 6277 | DEBUG_PRINT3 ("EXECUTING %scategoryspec %d.\n", not?"not":"", mcnt); |
b18215fc RS |
6278 | PREFETCH (); |
6279 | { | |
01618498 SM |
6280 | int len; |
6281 | re_wchar_t c; | |
6282 | ||
6fdd04b0 | 6283 | GET_CHAR_AFTER (c, d, len); |
1fb352e0 | 6284 | if ((!CHAR_HAS_CATEGORY (c, mcnt)) ^ not) |
b18215fc RS |
6285 | goto fail; |
6286 | d += len; | |
6287 | } | |
fa9a63c5 | 6288 | break; |
5e69f11e | 6289 | |
1fb352e0 | 6290 | #endif /* emacs */ |
5e69f11e | 6291 | |
0b32bf0e SM |
6292 | default: |
6293 | abort (); | |
fa9a63c5 | 6294 | } |
b18215fc | 6295 | continue; /* Successfully executed one pattern command; keep going. */ |
fa9a63c5 RM |
6296 | |
6297 | ||
6298 | /* We goto here if a matching operation fails. */ | |
6299 | fail: | |
5b370c2b | 6300 | IMMEDIATE_QUIT_CHECK; |
fa9a63c5 | 6301 | if (!FAIL_STACK_EMPTY ()) |
505bde11 | 6302 | { |
01618498 | 6303 | re_char *str, *pat; |
505bde11 | 6304 | /* A restart point is known. Restore to that state. */ |
0b32bf0e SM |
6305 | DEBUG_PRINT1 ("\nFAIL:\n"); |
6306 | POP_FAILURE_POINT (str, pat); | |
505bde11 SM |
6307 | switch (SWITCH_ENUM_CAST ((re_opcode_t) *pat++)) |
6308 | { | |
6309 | case on_failure_keep_string_jump: | |
6310 | assert (str == NULL); | |
6311 | goto continue_failure_jump; | |
6312 | ||
0683b6fa SM |
6313 | case on_failure_jump_nastyloop: |
6314 | assert ((re_opcode_t)pat[-2] == no_op); | |
6315 | PUSH_FAILURE_POINT (pat - 2, str); | |
6316 | /* Fallthrough */ | |
6317 | ||
505bde11 SM |
6318 | case on_failure_jump_loop: |
6319 | case on_failure_jump: | |
6320 | case succeed_n: | |
6321 | d = str; | |
6322 | continue_failure_jump: | |
6323 | EXTRACT_NUMBER_AND_INCR (mcnt, pat); | |
6324 | p = pat + mcnt; | |
6325 | break; | |
b18215fc | 6326 | |
0683b6fa SM |
6327 | case no_op: |
6328 | /* A special frame used for nastyloops. */ | |
6329 | goto fail; | |
6330 | ||
505bde11 SM |
6331 | default: |
6332 | abort(); | |
6333 | } | |
fa9a63c5 | 6334 | |
505bde11 | 6335 | assert (p >= bufp->buffer && p <= pend); |
b18215fc | 6336 | |
0b32bf0e | 6337 | if (d >= string1 && d <= end1) |
fa9a63c5 | 6338 | dend = end_match_1; |
0b32bf0e | 6339 | } |
fa9a63c5 | 6340 | else |
0b32bf0e | 6341 | break; /* Matching at this starting point really fails. */ |
fa9a63c5 RM |
6342 | } /* for (;;) */ |
6343 | ||
6344 | if (best_regs_set) | |
6345 | goto restore_best_regs; | |
6346 | ||
6347 | FREE_VARIABLES (); | |
6348 | ||
b18215fc | 6349 | return -1; /* Failure to match. */ |
fa9a63c5 RM |
6350 | } /* re_match_2 */ |
6351 | \f | |
6352 | /* Subroutine definitions for re_match_2. */ | |
6353 | ||
fa9a63c5 RM |
6354 | /* Return zero if TRANSLATE[S1] and TRANSLATE[S2] are identical for LEN |
6355 | bytes; nonzero otherwise. */ | |
5e69f11e | 6356 | |
fa9a63c5 | 6357 | static int |
438105ed JB |
6358 | bcmp_translate (const re_char *s1, const re_char *s2, register int len, |
6359 | RE_TRANSLATE_TYPE translate, const int target_multibyte) | |
fa9a63c5 | 6360 | { |
2d1675e4 SM |
6361 | register re_char *p1 = s1, *p2 = s2; |
6362 | re_char *p1_end = s1 + len; | |
6363 | re_char *p2_end = s2 + len; | |
e934739e | 6364 | |
4bb91c68 SM |
6365 | /* FIXME: Checking both p1 and p2 presumes that the two strings might have |
6366 | different lengths, but relying on a single `len' would break this. -sm */ | |
6367 | while (p1 < p1_end && p2 < p2_end) | |
fa9a63c5 | 6368 | { |
e934739e | 6369 | int p1_charlen, p2_charlen; |
01618498 | 6370 | re_wchar_t p1_ch, p2_ch; |
e934739e | 6371 | |
6fdd04b0 KH |
6372 | GET_CHAR_AFTER (p1_ch, p1, p1_charlen); |
6373 | GET_CHAR_AFTER (p2_ch, p2, p2_charlen); | |
e934739e RS |
6374 | |
6375 | if (RE_TRANSLATE (translate, p1_ch) | |
6376 | != RE_TRANSLATE (translate, p2_ch)) | |
bc192b5b | 6377 | return 1; |
e934739e RS |
6378 | |
6379 | p1 += p1_charlen, p2 += p2_charlen; | |
fa9a63c5 | 6380 | } |
e934739e RS |
6381 | |
6382 | if (p1 != p1_end || p2 != p2_end) | |
6383 | return 1; | |
6384 | ||
fa9a63c5 RM |
6385 | return 0; |
6386 | } | |
6387 | \f | |
6388 | /* Entry points for GNU code. */ | |
6389 | ||
6390 | /* re_compile_pattern is the GNU regular expression compiler: it | |
6391 | compiles PATTERN (of length SIZE) and puts the result in BUFP. | |
6392 | Returns 0 if the pattern was valid, otherwise an error string. | |
5e69f11e | 6393 | |
fa9a63c5 RM |
6394 | Assumes the `allocated' (and perhaps `buffer') and `translate' fields |
6395 | are set in BUFP on entry. | |
5e69f11e | 6396 | |
b18215fc | 6397 | We call regex_compile to do the actual compilation. */ |
fa9a63c5 RM |
6398 | |
6399 | const char * | |
971de7fb | 6400 | re_compile_pattern (const char *pattern, size_t length, struct re_pattern_buffer *bufp) |
fa9a63c5 RM |
6401 | { |
6402 | reg_errcode_t ret; | |
5e69f11e | 6403 | |
fa9a63c5 RM |
6404 | /* GNU code is written to assume at least RE_NREGS registers will be set |
6405 | (and at least one extra will be -1). */ | |
6406 | bufp->regs_allocated = REGS_UNALLOCATED; | |
5e69f11e | 6407 | |
fa9a63c5 RM |
6408 | /* And GNU code determines whether or not to get register information |
6409 | by passing null for the REGS argument to re_match, etc., not by | |
6410 | setting no_sub. */ | |
6411 | bufp->no_sub = 0; | |
5e69f11e | 6412 | |
4bb91c68 | 6413 | ret = regex_compile ((re_char*) pattern, length, re_syntax_options, bufp); |
fa9a63c5 RM |
6414 | |
6415 | if (!ret) | |
6416 | return NULL; | |
6417 | return gettext (re_error_msgid[(int) ret]); | |
5e69f11e | 6418 | } |
c0f9ea08 | 6419 | WEAK_ALIAS (__re_compile_pattern, re_compile_pattern) |
fa9a63c5 | 6420 | \f |
b18215fc RS |
6421 | /* Entry points compatible with 4.2 BSD regex library. We don't define |
6422 | them unless specifically requested. */ | |
fa9a63c5 | 6423 | |
0b32bf0e | 6424 | #if defined _REGEX_RE_COMP || defined _LIBC |
fa9a63c5 RM |
6425 | |
6426 | /* BSD has one and only one pattern buffer. */ | |
6427 | static struct re_pattern_buffer re_comp_buf; | |
6428 | ||
6429 | char * | |
0b32bf0e | 6430 | # ifdef _LIBC |
48afdd44 RM |
6431 | /* Make these definitions weak in libc, so POSIX programs can redefine |
6432 | these names if they don't use our functions, and still use | |
6433 | regcomp/regexec below without link errors. */ | |
6434 | weak_function | |
0b32bf0e | 6435 | # endif |
fa9a63c5 RM |
6436 | re_comp (s) |
6437 | const char *s; | |
6438 | { | |
6439 | reg_errcode_t ret; | |
5e69f11e | 6440 | |
fa9a63c5 RM |
6441 | if (!s) |
6442 | { | |
6443 | if (!re_comp_buf.buffer) | |
0b32bf0e | 6444 | /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */ |
a60198e5 | 6445 | return (char *) gettext ("No previous regular expression"); |
fa9a63c5 RM |
6446 | return 0; |
6447 | } | |
6448 | ||
6449 | if (!re_comp_buf.buffer) | |
6450 | { | |
6451 | re_comp_buf.buffer = (unsigned char *) malloc (200); | |
6452 | if (re_comp_buf.buffer == NULL) | |
0b32bf0e SM |
6453 | /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */ |
6454 | return (char *) gettext (re_error_msgid[(int) REG_ESPACE]); | |
fa9a63c5 RM |
6455 | re_comp_buf.allocated = 200; |
6456 | ||
6457 | re_comp_buf.fastmap = (char *) malloc (1 << BYTEWIDTH); | |
6458 | if (re_comp_buf.fastmap == NULL) | |
a60198e5 SM |
6459 | /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */ |
6460 | return (char *) gettext (re_error_msgid[(int) REG_ESPACE]); | |
fa9a63c5 RM |
6461 | } |
6462 | ||
6463 | /* Since `re_exec' always passes NULL for the `regs' argument, we | |
6464 | don't need to initialize the pattern buffer fields which affect it. */ | |
6465 | ||
fa9a63c5 | 6466 | ret = regex_compile (s, strlen (s), re_syntax_options, &re_comp_buf); |
5e69f11e | 6467 | |
fa9a63c5 RM |
6468 | if (!ret) |
6469 | return NULL; | |
6470 | ||
6471 | /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */ | |
6472 | return (char *) gettext (re_error_msgid[(int) ret]); | |
6473 | } | |
6474 | ||
6475 | ||
6476 | int | |
0b32bf0e | 6477 | # ifdef _LIBC |
48afdd44 | 6478 | weak_function |
0b32bf0e | 6479 | # endif |
fa9a63c5 RM |
6480 | re_exec (s) |
6481 | const char *s; | |
6482 | { | |
6483 | const int len = strlen (s); | |
6484 | return | |
6485 | 0 <= re_search (&re_comp_buf, s, len, 0, len, (struct re_registers *) 0); | |
6486 | } | |
6487 | #endif /* _REGEX_RE_COMP */ | |
6488 | \f | |
6489 | /* POSIX.2 functions. Don't define these for Emacs. */ | |
6490 | ||
6491 | #ifndef emacs | |
6492 | ||
6493 | /* regcomp takes a regular expression as a string and compiles it. | |
6494 | ||
b18215fc | 6495 | PREG is a regex_t *. We do not expect any fields to be initialized, |
fa9a63c5 RM |
6496 | since POSIX says we shouldn't. Thus, we set |
6497 | ||
6498 | `buffer' to the compiled pattern; | |
6499 | `used' to the length of the compiled pattern; | |
6500 | `syntax' to RE_SYNTAX_POSIX_EXTENDED if the | |
6501 | REG_EXTENDED bit in CFLAGS is set; otherwise, to | |
6502 | RE_SYNTAX_POSIX_BASIC; | |
c0f9ea08 SM |
6503 | `fastmap' to an allocated space for the fastmap; |
6504 | `fastmap_accurate' to zero; | |
fa9a63c5 RM |
6505 | `re_nsub' to the number of subexpressions in PATTERN. |
6506 | ||
6507 | PATTERN is the address of the pattern string. | |
6508 | ||
6509 | CFLAGS is a series of bits which affect compilation. | |
6510 | ||
6511 | If REG_EXTENDED is set, we use POSIX extended syntax; otherwise, we | |
6512 | use POSIX basic syntax. | |
6513 | ||
6514 | If REG_NEWLINE is set, then . and [^...] don't match newline. | |
6515 | Also, regexec will try a match beginning after every newline. | |
6516 | ||
6517 | If REG_ICASE is set, then we considers upper- and lowercase | |
6518 | versions of letters to be equivalent when matching. | |
6519 | ||
6520 | If REG_NOSUB is set, then when PREG is passed to regexec, that | |
6521 | routine will report only success or failure, and nothing about the | |
6522 | registers. | |
6523 | ||
b18215fc | 6524 | It returns 0 if it succeeds, nonzero if it doesn't. (See regex.h for |
fa9a63c5 RM |
6525 | the return codes and their meanings.) */ |
6526 | ||
6527 | int | |
d2762c86 DN |
6528 | regcomp (regex_t *__restrict preg, const char *__restrict pattern, |
6529 | int cflags) | |
fa9a63c5 RM |
6530 | { |
6531 | reg_errcode_t ret; | |
4bb91c68 | 6532 | reg_syntax_t syntax |
fa9a63c5 RM |
6533 | = (cflags & REG_EXTENDED) ? |
6534 | RE_SYNTAX_POSIX_EXTENDED : RE_SYNTAX_POSIX_BASIC; | |
6535 | ||
6536 | /* regex_compile will allocate the space for the compiled pattern. */ | |
6537 | preg->buffer = 0; | |
6538 | preg->allocated = 0; | |
6539 | preg->used = 0; | |
5e69f11e | 6540 | |
c0f9ea08 SM |
6541 | /* Try to allocate space for the fastmap. */ |
6542 | preg->fastmap = (char *) malloc (1 << BYTEWIDTH); | |
5e69f11e | 6543 | |
fa9a63c5 RM |
6544 | if (cflags & REG_ICASE) |
6545 | { | |
6546 | unsigned i; | |
5e69f11e | 6547 | |
6676cb1c RS |
6548 | preg->translate |
6549 | = (RE_TRANSLATE_TYPE) malloc (CHAR_SET_SIZE | |
6550 | * sizeof (*(RE_TRANSLATE_TYPE)0)); | |
fa9a63c5 | 6551 | if (preg->translate == NULL) |
0b32bf0e | 6552 | return (int) REG_ESPACE; |
fa9a63c5 RM |
6553 | |
6554 | /* Map uppercase characters to corresponding lowercase ones. */ | |
6555 | for (i = 0; i < CHAR_SET_SIZE; i++) | |
4bb91c68 | 6556 | preg->translate[i] = ISUPPER (i) ? TOLOWER (i) : i; |
fa9a63c5 RM |
6557 | } |
6558 | else | |
6559 | preg->translate = NULL; | |
6560 | ||
6561 | /* If REG_NEWLINE is set, newlines are treated differently. */ | |
6562 | if (cflags & REG_NEWLINE) | |
6563 | { /* REG_NEWLINE implies neither . nor [^...] match newline. */ | |
6564 | syntax &= ~RE_DOT_NEWLINE; | |
6565 | syntax |= RE_HAT_LISTS_NOT_NEWLINE; | |
fa9a63c5 RM |
6566 | } |
6567 | else | |
c0f9ea08 | 6568 | syntax |= RE_NO_NEWLINE_ANCHOR; |
fa9a63c5 RM |
6569 | |
6570 | preg->no_sub = !!(cflags & REG_NOSUB); | |
6571 | ||
5e69f11e | 6572 | /* POSIX says a null character in the pattern terminates it, so we |
fa9a63c5 | 6573 | can use strlen here in compiling the pattern. */ |
4bb91c68 | 6574 | ret = regex_compile ((re_char*) pattern, strlen (pattern), syntax, preg); |
5e69f11e | 6575 | |
fa9a63c5 RM |
6576 | /* POSIX doesn't distinguish between an unmatched open-group and an |
6577 | unmatched close-group: both are REG_EPAREN. */ | |
c0f9ea08 SM |
6578 | if (ret == REG_ERPAREN) |
6579 | ret = REG_EPAREN; | |
6580 | ||
6581 | if (ret == REG_NOERROR && preg->fastmap) | |
6582 | { /* Compute the fastmap now, since regexec cannot modify the pattern | |
6583 | buffer. */ | |
6584 | re_compile_fastmap (preg); | |
6585 | if (preg->can_be_null) | |
6586 | { /* The fastmap can't be used anyway. */ | |
6587 | free (preg->fastmap); | |
6588 | preg->fastmap = NULL; | |
6589 | } | |
6590 | } | |
fa9a63c5 RM |
6591 | return (int) ret; |
6592 | } | |
c0f9ea08 | 6593 | WEAK_ALIAS (__regcomp, regcomp) |
fa9a63c5 RM |
6594 | |
6595 | ||
6596 | /* regexec searches for a given pattern, specified by PREG, in the | |
6597 | string STRING. | |
5e69f11e | 6598 | |
fa9a63c5 | 6599 | If NMATCH is zero or REG_NOSUB was set in the cflags argument to |
b18215fc | 6600 | `regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at |
fa9a63c5 RM |
6601 | least NMATCH elements, and we set them to the offsets of the |
6602 | corresponding matched substrings. | |
5e69f11e | 6603 | |
fa9a63c5 RM |
6604 | EFLAGS specifies `execution flags' which affect matching: if |
6605 | REG_NOTBOL is set, then ^ does not match at the beginning of the | |
6606 | string; if REG_NOTEOL is set, then $ does not match at the end. | |
5e69f11e | 6607 | |
fa9a63c5 RM |
6608 | We return 0 if we find a match and REG_NOMATCH if not. */ |
6609 | ||
6610 | int | |
d2762c86 DN |
6611 | regexec (const regex_t *__restrict preg, const char *__restrict string, |
6612 | size_t nmatch, regmatch_t pmatch[__restrict_arr], int eflags) | |
fa9a63c5 RM |
6613 | { |
6614 | int ret; | |
6615 | struct re_registers regs; | |
6616 | regex_t private_preg; | |
6617 | int len = strlen (string); | |
c0f9ea08 | 6618 | boolean want_reg_info = !preg->no_sub && nmatch > 0 && pmatch; |
fa9a63c5 RM |
6619 | |
6620 | private_preg = *preg; | |
5e69f11e | 6621 | |
fa9a63c5 RM |
6622 | private_preg.not_bol = !!(eflags & REG_NOTBOL); |
6623 | private_preg.not_eol = !!(eflags & REG_NOTEOL); | |
5e69f11e | 6624 | |
fa9a63c5 RM |
6625 | /* The user has told us exactly how many registers to return |
6626 | information about, via `nmatch'. We have to pass that on to the | |
b18215fc | 6627 | matching routines. */ |
fa9a63c5 | 6628 | private_preg.regs_allocated = REGS_FIXED; |
5e69f11e | 6629 | |
fa9a63c5 RM |
6630 | if (want_reg_info) |
6631 | { | |
6632 | regs.num_regs = nmatch; | |
4bb91c68 SM |
6633 | regs.start = TALLOC (nmatch * 2, regoff_t); |
6634 | if (regs.start == NULL) | |
0b32bf0e | 6635 | return (int) REG_NOMATCH; |
4bb91c68 | 6636 | regs.end = regs.start + nmatch; |
fa9a63c5 RM |
6637 | } |
6638 | ||
c0f9ea08 SM |
6639 | /* Instead of using not_eol to implement REG_NOTEOL, we could simply |
6640 | pass (&private_preg, string, len + 1, 0, len, ...) pretending the string | |
6641 | was a little bit longer but still only matching the real part. | |
6642 | This works because the `endline' will check for a '\n' and will find a | |
6643 | '\0', correctly deciding that this is not the end of a line. | |
6644 | But it doesn't work out so nicely for REG_NOTBOL, since we don't have | |
6645 | a convenient '\0' there. For all we know, the string could be preceded | |
6646 | by '\n' which would throw things off. */ | |
6647 | ||
fa9a63c5 RM |
6648 | /* Perform the searching operation. */ |
6649 | ret = re_search (&private_preg, string, len, | |
0b32bf0e SM |
6650 | /* start: */ 0, /* range: */ len, |
6651 | want_reg_info ? ®s : (struct re_registers *) 0); | |
5e69f11e | 6652 | |
fa9a63c5 RM |
6653 | /* Copy the register information to the POSIX structure. */ |
6654 | if (want_reg_info) | |
6655 | { | |
6656 | if (ret >= 0) | |
0b32bf0e SM |
6657 | { |
6658 | unsigned r; | |
fa9a63c5 | 6659 | |
0b32bf0e SM |
6660 | for (r = 0; r < nmatch; r++) |
6661 | { | |
6662 | pmatch[r].rm_so = regs.start[r]; | |
6663 | pmatch[r].rm_eo = regs.end[r]; | |
6664 | } | |
6665 | } | |
fa9a63c5 | 6666 | |
b18215fc | 6667 | /* If we needed the temporary register info, free the space now. */ |
fa9a63c5 | 6668 | free (regs.start); |
fa9a63c5 RM |
6669 | } |
6670 | ||
6671 | /* We want zero return to mean success, unlike `re_search'. */ | |
6672 | return ret >= 0 ? (int) REG_NOERROR : (int) REG_NOMATCH; | |
6673 | } | |
c0f9ea08 | 6674 | WEAK_ALIAS (__regexec, regexec) |
fa9a63c5 RM |
6675 | |
6676 | ||
ec869672 JR |
6677 | /* Returns a message corresponding to an error code, ERR_CODE, returned |
6678 | from either regcomp or regexec. We don't use PREG here. | |
6679 | ||
6680 | ERR_CODE was previously called ERRCODE, but that name causes an | |
6681 | error with msvc8 compiler. */ | |
fa9a63c5 RM |
6682 | |
6683 | size_t | |
d2762c86 | 6684 | regerror (int err_code, const regex_t *preg, char *errbuf, size_t errbuf_size) |
fa9a63c5 RM |
6685 | { |
6686 | const char *msg; | |
6687 | size_t msg_size; | |
6688 | ||
ec869672 JR |
6689 | if (err_code < 0 |
6690 | || err_code >= (sizeof (re_error_msgid) / sizeof (re_error_msgid[0]))) | |
5e69f11e | 6691 | /* Only error codes returned by the rest of the code should be passed |
b18215fc | 6692 | to this routine. If we are given anything else, or if other regex |
fa9a63c5 RM |
6693 | code generates an invalid error code, then the program has a bug. |
6694 | Dump core so we can fix it. */ | |
6695 | abort (); | |
6696 | ||
ec869672 | 6697 | msg = gettext (re_error_msgid[err_code]); |
fa9a63c5 RM |
6698 | |
6699 | msg_size = strlen (msg) + 1; /* Includes the null. */ | |
5e69f11e | 6700 | |
fa9a63c5 RM |
6701 | if (errbuf_size != 0) |
6702 | { | |
6703 | if (msg_size > errbuf_size) | |
0b32bf0e SM |
6704 | { |
6705 | strncpy (errbuf, msg, errbuf_size - 1); | |
6706 | errbuf[errbuf_size - 1] = 0; | |
6707 | } | |
fa9a63c5 | 6708 | else |
0b32bf0e | 6709 | strcpy (errbuf, msg); |
fa9a63c5 RM |
6710 | } |
6711 | ||
6712 | return msg_size; | |
6713 | } | |
c0f9ea08 | 6714 | WEAK_ALIAS (__regerror, regerror) |
fa9a63c5 RM |
6715 | |
6716 | ||
6717 | /* Free dynamically allocated space used by PREG. */ | |
6718 | ||
6719 | void | |
d2762c86 | 6720 | regfree (regex_t *preg) |
fa9a63c5 | 6721 | { |
c2cd06e6 | 6722 | free (preg->buffer); |
fa9a63c5 | 6723 | preg->buffer = NULL; |
5e69f11e | 6724 | |
fa9a63c5 RM |
6725 | preg->allocated = 0; |
6726 | preg->used = 0; | |
6727 | ||
c2cd06e6 | 6728 | free (preg->fastmap); |
fa9a63c5 RM |
6729 | preg->fastmap = NULL; |
6730 | preg->fastmap_accurate = 0; | |
6731 | ||
c2cd06e6 | 6732 | free (preg->translate); |
fa9a63c5 RM |
6733 | preg->translate = NULL; |
6734 | } | |
c0f9ea08 | 6735 | WEAK_ALIAS (__regfree, regfree) |
fa9a63c5 RM |
6736 | |
6737 | #endif /* not emacs */ | |
839966f3 | 6738 |