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 | ||
4bb91c68 | 5 | Copyright (C) 1993,94,95,96,97,98,99,2000 Free Software Foundation, Inc. |
bc78d348 | 6 | |
fa9a63c5 RM |
7 | This program is free software; you can redistribute it and/or modify |
8 | it under the terms of the GNU General Public License as published by | |
9 | the Free Software Foundation; either version 2, or (at your option) | |
10 | any later version. | |
11 | ||
12 | This program is distributed in the hope that it will be useful, | |
13 | but WITHOUT ANY WARRANTY; without even the implied warranty of | |
25fe55af | 14 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
fa9a63c5 RM |
15 | GNU General Public License for more details. |
16 | ||
17 | You should have received a copy of the GNU General Public License | |
18 | along with this program; if not, write to the Free Software | |
ba4a8e51 | 19 | Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, |
25fe55af | 20 | USA. */ |
fa9a63c5 | 21 | |
36595814 SM |
22 | /* BUGS: |
23 | - (x?)*y\1z should match both xxxxyxz and xxxyz. | |
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 | ||
68d96f02 | 38 | #undef _GNU_SOURCE |
fa9a63c5 RM |
39 | #define _GNU_SOURCE |
40 | ||
41 | #ifdef HAVE_CONFIG_H | |
0b32bf0e | 42 | # include <config.h> |
fa9a63c5 RM |
43 | #endif |
44 | ||
4bb91c68 SM |
45 | #if defined STDC_HEADERS && !defined emacs |
46 | # include <stddef.h> | |
47 | #else | |
48 | /* We need this for `regex.h', and perhaps for the Emacs include files. */ | |
49 | # include <sys/types.h> | |
50 | #endif | |
fa9a63c5 | 51 | |
14473664 SM |
52 | /* Whether to use ISO C Amendment 1 wide char functions. |
53 | Those should not be used for Emacs since it uses its own. */ | |
5e5388f6 GM |
54 | #if defined _LIBC |
55 | #define WIDE_CHAR_SUPPORT 1 | |
56 | #else | |
14473664 | 57 | #define WIDE_CHAR_SUPPORT \ |
5e5388f6 GM |
58 | (HAVE_WCTYPE_H && HAVE_WCHAR_H && HAVE_BTOWC && !emacs) |
59 | #endif | |
14473664 SM |
60 | |
61 | /* For platform which support the ISO C amendement 1 functionality we | |
62 | support user defined character classes. */ | |
a0ad02f7 | 63 | #if WIDE_CHAR_SUPPORT |
14473664 SM |
64 | /* Solaris 2.5 has a bug: <wchar.h> must be included before <wctype.h>. */ |
65 | # include <wchar.h> | |
66 | # include <wctype.h> | |
67 | #endif | |
68 | ||
c0f9ea08 SM |
69 | #ifdef _LIBC |
70 | /* We have to keep the namespace clean. */ | |
71 | # define regfree(preg) __regfree (preg) | |
72 | # define regexec(pr, st, nm, pm, ef) __regexec (pr, st, nm, pm, ef) | |
73 | # define regcomp(preg, pattern, cflags) __regcomp (preg, pattern, cflags) | |
74 | # define regerror(errcode, preg, errbuf, errbuf_size) \ | |
75 | __regerror(errcode, preg, errbuf, errbuf_size) | |
76 | # define re_set_registers(bu, re, nu, st, en) \ | |
77 | __re_set_registers (bu, re, nu, st, en) | |
78 | # define re_match_2(bufp, string1, size1, string2, size2, pos, regs, stop) \ | |
79 | __re_match_2 (bufp, string1, size1, string2, size2, pos, regs, stop) | |
80 | # define re_match(bufp, string, size, pos, regs) \ | |
81 | __re_match (bufp, string, size, pos, regs) | |
82 | # define re_search(bufp, string, size, startpos, range, regs) \ | |
83 | __re_search (bufp, string, size, startpos, range, regs) | |
84 | # define re_compile_pattern(pattern, length, bufp) \ | |
85 | __re_compile_pattern (pattern, length, bufp) | |
86 | # define re_set_syntax(syntax) __re_set_syntax (syntax) | |
87 | # define re_search_2(bufp, st1, s1, st2, s2, startpos, range, regs, stop) \ | |
88 | __re_search_2 (bufp, st1, s1, st2, s2, startpos, range, regs, stop) | |
89 | # define re_compile_fastmap(bufp) __re_compile_fastmap (bufp) | |
90 | ||
14473664 SM |
91 | /* Make sure we call libc's function even if the user overrides them. */ |
92 | # define btowc __btowc | |
93 | # define iswctype __iswctype | |
94 | # define wctype __wctype | |
95 | ||
c0f9ea08 SM |
96 | # define WEAK_ALIAS(a,b) weak_alias (a, b) |
97 | ||
98 | /* We are also using some library internals. */ | |
99 | # include <locale/localeinfo.h> | |
100 | # include <locale/elem-hash.h> | |
101 | # include <langinfo.h> | |
102 | #else | |
103 | # define WEAK_ALIAS(a,b) | |
104 | #endif | |
105 | ||
4bb91c68 | 106 | /* This is for other GNU distributions with internationalized messages. */ |
0b32bf0e | 107 | #if HAVE_LIBINTL_H || defined _LIBC |
fa9a63c5 RM |
108 | # include <libintl.h> |
109 | #else | |
110 | # define gettext(msgid) (msgid) | |
111 | #endif | |
112 | ||
5e69f11e RM |
113 | #ifndef gettext_noop |
114 | /* This define is so xgettext can find the internationalizable | |
115 | strings. */ | |
0b32bf0e | 116 | # define gettext_noop(String) String |
5e69f11e RM |
117 | #endif |
118 | ||
fa9a63c5 RM |
119 | /* The `emacs' switch turns on certain matching commands |
120 | that make sense only in Emacs. */ | |
121 | #ifdef emacs | |
122 | ||
0b32bf0e SM |
123 | # include "lisp.h" |
124 | # include "buffer.h" | |
b18215fc RS |
125 | |
126 | /* Make syntax table lookup grant data in gl_state. */ | |
0b32bf0e | 127 | # define SYNTAX_ENTRY_VIA_PROPERTY |
b18215fc | 128 | |
0b32bf0e SM |
129 | # include "syntax.h" |
130 | # include "charset.h" | |
131 | # include "category.h" | |
fa9a63c5 | 132 | |
7689ef0b EZ |
133 | # ifdef malloc |
134 | # undef malloc | |
135 | # endif | |
0b32bf0e | 136 | # define malloc xmalloc |
7689ef0b EZ |
137 | # ifdef realloc |
138 | # undef realloc | |
139 | # endif | |
0b32bf0e | 140 | # define realloc xrealloc |
7689ef0b EZ |
141 | # ifdef free |
142 | # undef free | |
143 | # endif | |
0b32bf0e | 144 | # define free xfree |
9abbd165 | 145 | |
0b32bf0e SM |
146 | /* Converts the pointer to the char to BEG-based offset from the start. */ |
147 | # define PTR_TO_OFFSET(d) POS_AS_IN_BUFFER (POINTER_TO_OFFSET (d)) | |
148 | # define POS_AS_IN_BUFFER(p) ((p) + (NILP (re_match_object) || BUFFERP (re_match_object))) | |
149 | ||
150 | # define RE_MULTIBYTE_P(bufp) ((bufp)->multibyte) | |
151 | # define RE_STRING_CHAR(p, s) \ | |
4e8a9132 | 152 | (multibyte ? (STRING_CHAR (p, s)) : (*(p))) |
0b32bf0e | 153 | # define RE_STRING_CHAR_AND_LENGTH(p, s, len) \ |
2d1675e4 SM |
154 | (multibyte ? (STRING_CHAR_AND_LENGTH (p, s, len)) : ((len) = 1, *(p))) |
155 | ||
156 | /* Set C a (possibly multibyte) character before P. P points into a | |
157 | string which is the virtual concatenation of STR1 (which ends at | |
158 | END1) or STR2 (which ends at END2). */ | |
0b32bf0e | 159 | # define GET_CHAR_BEFORE_2(c, p, str1, end1, str2, end2) \ |
2d1675e4 SM |
160 | do { \ |
161 | if (multibyte) \ | |
162 | { \ | |
0b32bf0e SM |
163 | re_char *dtemp = (p) == (str2) ? (end1) : (p); \ |
164 | re_char *dlimit = ((p) > (str2) && (p) <= (end2)) ? (str2) : (str1); \ | |
165 | while (dtemp-- > dlimit && !CHAR_HEAD_P (*dtemp)); \ | |
166 | c = STRING_CHAR (dtemp, (p) - dtemp); \ | |
2d1675e4 SM |
167 | } \ |
168 | else \ | |
169 | (c = ((p) == (str2) ? (end1) : (p))[-1]); \ | |
170 | } while (0) | |
171 | ||
4e8a9132 | 172 | |
fa9a63c5 RM |
173 | #else /* not emacs */ |
174 | ||
175 | /* If we are not linking with Emacs proper, | |
176 | we can't use the relocating allocator | |
177 | even if config.h says that we can. */ | |
0b32bf0e | 178 | # undef REL_ALLOC |
fa9a63c5 | 179 | |
0b32bf0e SM |
180 | # if defined STDC_HEADERS || defined _LIBC |
181 | # include <stdlib.h> | |
182 | # else | |
fa9a63c5 RM |
183 | char *malloc (); |
184 | char *realloc (); | |
0b32bf0e | 185 | # endif |
fa9a63c5 | 186 | |
9e4ecb26 | 187 | /* When used in Emacs's lib-src, we need to get bzero and bcopy somehow. |
4bb91c68 | 188 | If nothing else has been done, use the method below. */ |
0b32bf0e SM |
189 | # ifdef INHIBIT_STRING_HEADER |
190 | # if !(defined HAVE_BZERO && defined HAVE_BCOPY) | |
191 | # if !defined bzero && !defined bcopy | |
192 | # undef INHIBIT_STRING_HEADER | |
193 | # endif | |
194 | # endif | |
195 | # endif | |
9e4ecb26 | 196 | |
4bb91c68 | 197 | /* This is the normal way of making sure we have memcpy, memcmp and bzero. |
9e4ecb26 KH |
198 | This is used in most programs--a few other programs avoid this |
199 | by defining INHIBIT_STRING_HEADER. */ | |
0b32bf0e SM |
200 | # ifndef INHIBIT_STRING_HEADER |
201 | # if defined HAVE_STRING_H || defined STDC_HEADERS || defined _LIBC | |
202 | # include <string.h> | |
0b32bf0e | 203 | # ifndef bzero |
4bb91c68 SM |
204 | # ifndef _LIBC |
205 | # define bzero(s, n) (memset (s, '\0', n), (s)) | |
206 | # else | |
207 | # define bzero(s, n) __bzero (s, n) | |
208 | # endif | |
0b32bf0e SM |
209 | # endif |
210 | # else | |
211 | # include <strings.h> | |
4bb91c68 SM |
212 | # ifndef memcmp |
213 | # define memcmp(s1, s2, n) bcmp (s1, s2, n) | |
214 | # endif | |
215 | # ifndef memcpy | |
216 | # define memcpy(d, s, n) (bcopy (s, d, n), (d)) | |
217 | # endif | |
0b32bf0e SM |
218 | # endif |
219 | # endif | |
fa9a63c5 RM |
220 | |
221 | /* Define the syntax stuff for \<, \>, etc. */ | |
222 | ||
990b2375 SM |
223 | /* Sword must be nonzero for the wordchar pattern commands in re_match_2. */ |
224 | enum syntaxcode { Swhitespace = 0, Sword = 1 }; | |
fa9a63c5 | 225 | |
0b32bf0e SM |
226 | # ifdef SWITCH_ENUM_BUG |
227 | # define SWITCH_ENUM_CAST(x) ((int)(x)) | |
228 | # else | |
229 | # define SWITCH_ENUM_CAST(x) (x) | |
230 | # endif | |
fa9a63c5 | 231 | |
e934739e | 232 | /* Dummy macros for non-Emacs environments. */ |
0b32bf0e SM |
233 | # define BASE_LEADING_CODE_P(c) (0) |
234 | # define CHAR_CHARSET(c) 0 | |
235 | # define CHARSET_LEADING_CODE_BASE(c) 0 | |
236 | # define MAX_MULTIBYTE_LENGTH 1 | |
237 | # define RE_MULTIBYTE_P(x) 0 | |
238 | # define WORD_BOUNDARY_P(c1, c2) (0) | |
239 | # define CHAR_HEAD_P(p) (1) | |
240 | # define SINGLE_BYTE_CHAR_P(c) (1) | |
241 | # define SAME_CHARSET_P(c1, c2) (1) | |
242 | # define MULTIBYTE_FORM_LENGTH(p, s) (1) | |
243 | # define STRING_CHAR(p, s) (*(p)) | |
244 | # define RE_STRING_CHAR STRING_CHAR | |
245 | # define CHAR_STRING(c, s) (*(s) = (c), 1) | |
246 | # define STRING_CHAR_AND_LENGTH(p, s, actual_len) ((actual_len) = 1, *(p)) | |
247 | # define RE_STRING_CHAR_AND_LENGTH STRING_CHAR_AND_LENGTH | |
248 | # define GET_CHAR_BEFORE_2(c, p, str1, end1, str2, end2) \ | |
b18215fc | 249 | (c = ((p) == (str2) ? *((end1) - 1) : *((p) - 1))) |
0b32bf0e | 250 | # define MAKE_CHAR(charset, c1, c2) (c1) |
fa9a63c5 | 251 | #endif /* not emacs */ |
4e8a9132 SM |
252 | |
253 | #ifndef RE_TRANSLATE | |
0b32bf0e SM |
254 | # define RE_TRANSLATE(TBL, C) ((unsigned char)(TBL)[C]) |
255 | # define RE_TRANSLATE_P(TBL) (TBL) | |
4e8a9132 | 256 | #endif |
fa9a63c5 RM |
257 | \f |
258 | /* Get the interface, including the syntax bits. */ | |
259 | #include "regex.h" | |
260 | ||
f71b19b6 DL |
261 | /* isalpha etc. are used for the character classes. */ |
262 | #include <ctype.h> | |
fa9a63c5 | 263 | |
f71b19b6 | 264 | #ifdef emacs |
fa9a63c5 | 265 | |
f71b19b6 | 266 | /* 1 if C is an ASCII character. */ |
0b32bf0e | 267 | # define IS_REAL_ASCII(c) ((c) < 0200) |
fa9a63c5 | 268 | |
f71b19b6 | 269 | /* 1 if C is a unibyte character. */ |
0b32bf0e | 270 | # define ISUNIBYTE(c) (SINGLE_BYTE_CHAR_P ((c))) |
96cc36cc | 271 | |
f71b19b6 | 272 | /* The Emacs definitions should not be directly affected by locales. */ |
96cc36cc | 273 | |
f71b19b6 | 274 | /* In Emacs, these are only used for single-byte characters. */ |
0b32bf0e SM |
275 | # define ISDIGIT(c) ((c) >= '0' && (c) <= '9') |
276 | # define ISCNTRL(c) ((c) < ' ') | |
277 | # define ISXDIGIT(c) (((c) >= '0' && (c) <= '9') \ | |
f71b19b6 DL |
278 | || ((c) >= 'a' && (c) <= 'f') \ |
279 | || ((c) >= 'A' && (c) <= 'F')) | |
96cc36cc RS |
280 | |
281 | /* This is only used for single-byte characters. */ | |
0b32bf0e | 282 | # define ISBLANK(c) ((c) == ' ' || (c) == '\t') |
96cc36cc RS |
283 | |
284 | /* The rest must handle multibyte characters. */ | |
285 | ||
0b32bf0e | 286 | # define ISGRAPH(c) (SINGLE_BYTE_CHAR_P (c) \ |
f71b19b6 | 287 | ? (c) > ' ' && !((c) >= 0177 && (c) <= 0237) \ |
96cc36cc RS |
288 | : 1) |
289 | ||
14473664 | 290 | # define ISPRINT(c) (SINGLE_BYTE_CHAR_P (c) \ |
f71b19b6 | 291 | ? (c) >= ' ' && !((c) >= 0177 && (c) <= 0237) \ |
96cc36cc RS |
292 | : 1) |
293 | ||
0b32bf0e | 294 | # define ISALNUM(c) (IS_REAL_ASCII (c) \ |
f71b19b6 DL |
295 | ? (((c) >= 'a' && (c) <= 'z') \ |
296 | || ((c) >= 'A' && (c) <= 'Z') \ | |
297 | || ((c) >= '0' && (c) <= '9')) \ | |
96cc36cc RS |
298 | : SYNTAX (c) == Sword) |
299 | ||
0b32bf0e | 300 | # define ISALPHA(c) (IS_REAL_ASCII (c) \ |
f71b19b6 DL |
301 | ? (((c) >= 'a' && (c) <= 'z') \ |
302 | || ((c) >= 'A' && (c) <= 'Z')) \ | |
96cc36cc RS |
303 | : SYNTAX (c) == Sword) |
304 | ||
0b32bf0e | 305 | # define ISLOWER(c) (LOWERCASEP (c)) |
96cc36cc | 306 | |
0b32bf0e | 307 | # define ISPUNCT(c) (IS_REAL_ASCII (c) \ |
f71b19b6 DL |
308 | ? ((c) > ' ' && (c) < 0177 \ |
309 | && !(((c) >= 'a' && (c) <= 'z') \ | |
4bb91c68 SM |
310 | || ((c) >= 'A' && (c) <= 'Z') \ |
311 | || ((c) >= '0' && (c) <= '9'))) \ | |
96cc36cc RS |
312 | : SYNTAX (c) != Sword) |
313 | ||
0b32bf0e | 314 | # define ISSPACE(c) (SYNTAX (c) == Swhitespace) |
96cc36cc | 315 | |
0b32bf0e | 316 | # define ISUPPER(c) (UPPERCASEP (c)) |
96cc36cc | 317 | |
0b32bf0e | 318 | # define ISWORD(c) (SYNTAX (c) == Sword) |
96cc36cc RS |
319 | |
320 | #else /* not emacs */ | |
321 | ||
f71b19b6 DL |
322 | /* Jim Meyering writes: |
323 | ||
324 | "... Some ctype macros are valid only for character codes that | |
325 | isascii says are ASCII (SGI's IRIX-4.0.5 is one such system --when | |
326 | using /bin/cc or gcc but without giving an ansi option). So, all | |
4bb91c68 | 327 | ctype uses should be through macros like ISPRINT... If |
f71b19b6 DL |
328 | STDC_HEADERS is defined, then autoconf has verified that the ctype |
329 | macros don't need to be guarded with references to isascii. ... | |
330 | Defining isascii to 1 should let any compiler worth its salt | |
4bb91c68 SM |
331 | eliminate the && through constant folding." |
332 | Solaris defines some of these symbols so we must undefine them first. */ | |
f71b19b6 | 333 | |
4bb91c68 | 334 | # undef ISASCII |
0b32bf0e SM |
335 | # if defined STDC_HEADERS || (!defined isascii && !defined HAVE_ISASCII) |
336 | # define ISASCII(c) 1 | |
337 | # else | |
338 | # define ISASCII(c) isascii(c) | |
339 | # endif | |
f71b19b6 DL |
340 | |
341 | /* 1 if C is an ASCII character. */ | |
0b32bf0e | 342 | # define IS_REAL_ASCII(c) ((c) < 0200) |
f71b19b6 DL |
343 | |
344 | /* This distinction is not meaningful, except in Emacs. */ | |
0b32bf0e SM |
345 | # define ISUNIBYTE(c) 1 |
346 | ||
347 | # ifdef isblank | |
348 | # define ISBLANK(c) (ISASCII (c) && isblank (c)) | |
349 | # else | |
350 | # define ISBLANK(c) ((c) == ' ' || (c) == '\t') | |
351 | # endif | |
352 | # ifdef isgraph | |
353 | # define ISGRAPH(c) (ISASCII (c) && isgraph (c)) | |
354 | # else | |
355 | # define ISGRAPH(c) (ISASCII (c) && isprint (c) && !isspace (c)) | |
356 | # endif | |
357 | ||
4bb91c68 | 358 | # undef ISPRINT |
0b32bf0e SM |
359 | # define ISPRINT(c) (ISASCII (c) && isprint (c)) |
360 | # define ISDIGIT(c) (ISASCII (c) && isdigit (c)) | |
361 | # define ISALNUM(c) (ISASCII (c) && isalnum (c)) | |
362 | # define ISALPHA(c) (ISASCII (c) && isalpha (c)) | |
363 | # define ISCNTRL(c) (ISASCII (c) && iscntrl (c)) | |
364 | # define ISLOWER(c) (ISASCII (c) && islower (c)) | |
365 | # define ISPUNCT(c) (ISASCII (c) && ispunct (c)) | |
366 | # define ISSPACE(c) (ISASCII (c) && isspace (c)) | |
367 | # define ISUPPER(c) (ISASCII (c) && isupper (c)) | |
368 | # define ISXDIGIT(c) (ISASCII (c) && isxdigit (c)) | |
369 | ||
370 | # define ISWORD(c) ISALPHA(c) | |
371 | ||
4bb91c68 SM |
372 | # ifdef _tolower |
373 | # define TOLOWER(c) _tolower(c) | |
374 | # else | |
375 | # define TOLOWER(c) tolower(c) | |
376 | # endif | |
377 | ||
378 | /* How many characters in the character set. */ | |
379 | # define CHAR_SET_SIZE 256 | |
380 | ||
0b32bf0e | 381 | # ifdef SYNTAX_TABLE |
f71b19b6 | 382 | |
0b32bf0e | 383 | extern char *re_syntax_table; |
f71b19b6 | 384 | |
0b32bf0e SM |
385 | # else /* not SYNTAX_TABLE */ |
386 | ||
0b32bf0e SM |
387 | static char re_syntax_table[CHAR_SET_SIZE]; |
388 | ||
389 | static void | |
390 | init_syntax_once () | |
391 | { | |
392 | register int c; | |
393 | static int done = 0; | |
394 | ||
395 | if (done) | |
396 | return; | |
397 | ||
398 | bzero (re_syntax_table, sizeof re_syntax_table); | |
399 | ||
4bb91c68 SM |
400 | for (c = 0; c < CHAR_SET_SIZE; ++c) |
401 | if (ISALNUM (c)) | |
402 | re_syntax_table[c] = Sword; | |
fa9a63c5 | 403 | |
0b32bf0e | 404 | re_syntax_table['_'] = Sword; |
fa9a63c5 | 405 | |
0b32bf0e SM |
406 | done = 1; |
407 | } | |
408 | ||
409 | # endif /* not SYNTAX_TABLE */ | |
96cc36cc | 410 | |
4bb91c68 SM |
411 | # define SYNTAX(c) re_syntax_table[(c)] |
412 | ||
96cc36cc RS |
413 | #endif /* not emacs */ |
414 | \f | |
fa9a63c5 | 415 | #ifndef NULL |
0b32bf0e | 416 | # define NULL (void *)0 |
fa9a63c5 RM |
417 | #endif |
418 | ||
419 | /* We remove any previous definition of `SIGN_EXTEND_CHAR', | |
420 | since ours (we hope) works properly with all combinations of | |
421 | machines, compilers, `char' and `unsigned char' argument types. | |
4bb91c68 | 422 | (Per Bothner suggested the basic approach.) */ |
fa9a63c5 RM |
423 | #undef SIGN_EXTEND_CHAR |
424 | #if __STDC__ | |
0b32bf0e | 425 | # define SIGN_EXTEND_CHAR(c) ((signed char) (c)) |
fa9a63c5 RM |
426 | #else /* not __STDC__ */ |
427 | /* As in Harbison and Steele. */ | |
0b32bf0e | 428 | # define SIGN_EXTEND_CHAR(c) ((((unsigned char) (c)) ^ 128) - 128) |
fa9a63c5 RM |
429 | #endif |
430 | \f | |
431 | /* Should we use malloc or alloca? If REGEX_MALLOC is not defined, we | |
432 | use `alloca' instead of `malloc'. This is because using malloc in | |
433 | re_search* or re_match* could cause memory leaks when C-g is used in | |
434 | Emacs; also, malloc is slower and causes storage fragmentation. On | |
5e69f11e RM |
435 | the other hand, malloc is more portable, and easier to debug. |
436 | ||
fa9a63c5 RM |
437 | Because we sometimes use alloca, some routines have to be macros, |
438 | not functions -- `alloca'-allocated space disappears at the end of the | |
439 | function it is called in. */ | |
440 | ||
441 | #ifdef REGEX_MALLOC | |
442 | ||
0b32bf0e SM |
443 | # define REGEX_ALLOCATE malloc |
444 | # define REGEX_REALLOCATE(source, osize, nsize) realloc (source, nsize) | |
445 | # define REGEX_FREE free | |
fa9a63c5 RM |
446 | |
447 | #else /* not REGEX_MALLOC */ | |
448 | ||
449 | /* Emacs already defines alloca, sometimes. */ | |
0b32bf0e | 450 | # ifndef alloca |
fa9a63c5 RM |
451 | |
452 | /* Make alloca work the best possible way. */ | |
0b32bf0e SM |
453 | # ifdef __GNUC__ |
454 | # define alloca __builtin_alloca | |
455 | # else /* not __GNUC__ */ | |
456 | # if HAVE_ALLOCA_H | |
457 | # include <alloca.h> | |
458 | # endif /* HAVE_ALLOCA_H */ | |
459 | # endif /* not __GNUC__ */ | |
fa9a63c5 | 460 | |
0b32bf0e | 461 | # endif /* not alloca */ |
fa9a63c5 | 462 | |
0b32bf0e | 463 | # define REGEX_ALLOCATE alloca |
fa9a63c5 RM |
464 | |
465 | /* Assumes a `char *destination' variable. */ | |
0b32bf0e | 466 | # define REGEX_REALLOCATE(source, osize, nsize) \ |
fa9a63c5 | 467 | (destination = (char *) alloca (nsize), \ |
4bb91c68 | 468 | memcpy (destination, source, osize)) |
fa9a63c5 RM |
469 | |
470 | /* No need to do anything to free, after alloca. */ | |
0b32bf0e | 471 | # define REGEX_FREE(arg) ((void)0) /* Do nothing! But inhibit gcc warning. */ |
fa9a63c5 RM |
472 | |
473 | #endif /* not REGEX_MALLOC */ | |
474 | ||
475 | /* Define how to allocate the failure stack. */ | |
476 | ||
0b32bf0e | 477 | #if defined REL_ALLOC && defined REGEX_MALLOC |
4297555e | 478 | |
0b32bf0e | 479 | # define REGEX_ALLOCATE_STACK(size) \ |
fa9a63c5 | 480 | r_alloc (&failure_stack_ptr, (size)) |
0b32bf0e | 481 | # define REGEX_REALLOCATE_STACK(source, osize, nsize) \ |
fa9a63c5 | 482 | r_re_alloc (&failure_stack_ptr, (nsize)) |
0b32bf0e | 483 | # define REGEX_FREE_STACK(ptr) \ |
fa9a63c5 RM |
484 | r_alloc_free (&failure_stack_ptr) |
485 | ||
4297555e | 486 | #else /* not using relocating allocator */ |
fa9a63c5 | 487 | |
0b32bf0e | 488 | # ifdef REGEX_MALLOC |
fa9a63c5 | 489 | |
0b32bf0e SM |
490 | # define REGEX_ALLOCATE_STACK malloc |
491 | # define REGEX_REALLOCATE_STACK(source, osize, nsize) realloc (source, nsize) | |
492 | # define REGEX_FREE_STACK free | |
fa9a63c5 | 493 | |
0b32bf0e | 494 | # else /* not REGEX_MALLOC */ |
fa9a63c5 | 495 | |
0b32bf0e | 496 | # define REGEX_ALLOCATE_STACK alloca |
fa9a63c5 | 497 | |
0b32bf0e | 498 | # define REGEX_REALLOCATE_STACK(source, osize, nsize) \ |
fa9a63c5 | 499 | REGEX_REALLOCATE (source, osize, nsize) |
25fe55af | 500 | /* No need to explicitly free anything. */ |
0b32bf0e | 501 | # define REGEX_FREE_STACK(arg) ((void)0) |
fa9a63c5 | 502 | |
0b32bf0e | 503 | # endif /* not REGEX_MALLOC */ |
4297555e | 504 | #endif /* not using relocating allocator */ |
fa9a63c5 RM |
505 | |
506 | ||
507 | /* True if `size1' is non-NULL and PTR is pointing anywhere inside | |
508 | `string1' or just past its end. This works if PTR is NULL, which is | |
509 | a good thing. */ | |
25fe55af | 510 | #define FIRST_STRING_P(ptr) \ |
fa9a63c5 RM |
511 | (size1 && string1 <= (ptr) && (ptr) <= string1 + size1) |
512 | ||
513 | /* (Re)Allocate N items of type T using malloc, or fail. */ | |
514 | #define TALLOC(n, t) ((t *) malloc ((n) * sizeof (t))) | |
515 | #define RETALLOC(addr, n, t) ((addr) = (t *) realloc (addr, (n) * sizeof (t))) | |
516 | #define RETALLOC_IF(addr, n, t) \ | |
517 | if (addr) RETALLOC((addr), (n), t); else (addr) = TALLOC ((n), t) | |
518 | #define REGEX_TALLOC(n, t) ((t *) REGEX_ALLOCATE ((n) * sizeof (t))) | |
519 | ||
4bb91c68 | 520 | #define BYTEWIDTH 8 /* In bits. */ |
fa9a63c5 RM |
521 | |
522 | #define STREQ(s1, s2) ((strcmp (s1, s2) == 0)) | |
523 | ||
524 | #undef MAX | |
525 | #undef MIN | |
526 | #define MAX(a, b) ((a) > (b) ? (a) : (b)) | |
527 | #define MIN(a, b) ((a) < (b) ? (a) : (b)) | |
528 | ||
66f0296e SM |
529 | /* Type of source-pattern and string chars. */ |
530 | typedef const unsigned char re_char; | |
531 | ||
fa9a63c5 RM |
532 | typedef char boolean; |
533 | #define false 0 | |
534 | #define true 1 | |
535 | ||
4bb91c68 SM |
536 | static int re_match_2_internal _RE_ARGS ((struct re_pattern_buffer *bufp, |
537 | re_char *string1, int size1, | |
538 | re_char *string2, int size2, | |
539 | int pos, | |
540 | struct re_registers *regs, | |
541 | int stop)); | |
fa9a63c5 RM |
542 | \f |
543 | /* These are the command codes that appear in compiled regular | |
4bb91c68 | 544 | expressions. Some opcodes are followed by argument bytes. A |
fa9a63c5 RM |
545 | command code can specify any interpretation whatsoever for its |
546 | arguments. Zero bytes may appear in the compiled regular expression. */ | |
547 | ||
548 | typedef enum | |
549 | { | |
550 | no_op = 0, | |
551 | ||
4bb91c68 | 552 | /* Succeed right away--no more backtracking. */ |
fa9a63c5 RM |
553 | succeed, |
554 | ||
25fe55af | 555 | /* Followed by one byte giving n, then by n literal bytes. */ |
fa9a63c5 RM |
556 | exactn, |
557 | ||
25fe55af | 558 | /* Matches any (more or less) character. */ |
fa9a63c5 RM |
559 | anychar, |
560 | ||
25fe55af RS |
561 | /* Matches any one char belonging to specified set. First |
562 | following byte is number of bitmap bytes. Then come bytes | |
563 | for a bitmap saying which chars are in. Bits in each byte | |
564 | are ordered low-bit-first. A character is in the set if its | |
565 | bit is 1. A character too large to have a bit in the map is | |
96cc36cc RS |
566 | automatically not in the set. |
567 | ||
568 | If the length byte has the 0x80 bit set, then that stuff | |
569 | is followed by a range table: | |
570 | 2 bytes of flags for character sets (low 8 bits, high 8 bits) | |
0b32bf0e | 571 | See RANGE_TABLE_WORK_BITS below. |
01618498 | 572 | 2 bytes, the number of pairs that follow (upto 32767) |
96cc36cc | 573 | pairs, each 2 multibyte characters, |
0b32bf0e | 574 | each multibyte character represented as 3 bytes. */ |
fa9a63c5 RM |
575 | charset, |
576 | ||
25fe55af | 577 | /* Same parameters as charset, but match any character that is |
4bb91c68 | 578 | not one of those specified. */ |
fa9a63c5 RM |
579 | charset_not, |
580 | ||
25fe55af RS |
581 | /* Start remembering the text that is matched, for storing in a |
582 | register. Followed by one byte with the register number, in | |
583 | the range 0 to one less than the pattern buffer's re_nsub | |
505bde11 | 584 | field. */ |
fa9a63c5 RM |
585 | start_memory, |
586 | ||
25fe55af RS |
587 | /* Stop remembering the text that is matched and store it in a |
588 | memory register. Followed by one byte with the register | |
589 | number, in the range 0 to one less than `re_nsub' in the | |
505bde11 | 590 | pattern buffer. */ |
fa9a63c5 RM |
591 | stop_memory, |
592 | ||
25fe55af | 593 | /* Match a duplicate of something remembered. Followed by one |
4bb91c68 | 594 | byte containing the register number. */ |
fa9a63c5 RM |
595 | duplicate, |
596 | ||
25fe55af | 597 | /* Fail unless at beginning of line. */ |
fa9a63c5 RM |
598 | begline, |
599 | ||
4bb91c68 | 600 | /* Fail unless at end of line. */ |
fa9a63c5 RM |
601 | endline, |
602 | ||
25fe55af RS |
603 | /* Succeeds if at beginning of buffer (if emacs) or at beginning |
604 | of string to be matched (if not). */ | |
fa9a63c5 RM |
605 | begbuf, |
606 | ||
25fe55af | 607 | /* Analogously, for end of buffer/string. */ |
fa9a63c5 | 608 | endbuf, |
5e69f11e | 609 | |
25fe55af | 610 | /* Followed by two byte relative address to which to jump. */ |
5e69f11e | 611 | jump, |
fa9a63c5 | 612 | |
25fe55af RS |
613 | /* Followed by two-byte relative address of place to resume at |
614 | in case of failure. */ | |
fa9a63c5 | 615 | on_failure_jump, |
5e69f11e | 616 | |
25fe55af RS |
617 | /* Like on_failure_jump, but pushes a placeholder instead of the |
618 | current string position when executed. */ | |
fa9a63c5 | 619 | on_failure_keep_string_jump, |
5e69f11e | 620 | |
505bde11 SM |
621 | /* Just like `on_failure_jump', except that it checks that we |
622 | don't get stuck in an infinite loop (matching an empty string | |
623 | indefinitely). */ | |
624 | on_failure_jump_loop, | |
625 | ||
0683b6fa SM |
626 | /* Just like `on_failure_jump_loop', except that it checks for |
627 | a different kind of loop (the kind that shows up with non-greedy | |
628 | operators). This operation has to be immediately preceded | |
629 | by a `no_op'. */ | |
630 | on_failure_jump_nastyloop, | |
631 | ||
0b32bf0e | 632 | /* A smart `on_failure_jump' used for greedy * and + operators. |
505bde11 SM |
633 | It analyses the loop before which it is put and if the |
634 | loop does not require backtracking, it changes itself to | |
4e8a9132 SM |
635 | `on_failure_keep_string_jump' and short-circuits the loop, |
636 | else it just defaults to changing itself into `on_failure_jump'. | |
637 | It assumes that it is pointing to just past a `jump'. */ | |
505bde11 | 638 | on_failure_jump_smart, |
fa9a63c5 | 639 | |
25fe55af | 640 | /* Followed by two-byte relative address and two-byte number n. |
ed0767d8 SM |
641 | After matching N times, jump to the address upon failure. |
642 | Does not work if N starts at 0: use on_failure_jump_loop | |
643 | instead. */ | |
fa9a63c5 RM |
644 | succeed_n, |
645 | ||
25fe55af RS |
646 | /* Followed by two-byte relative address, and two-byte number n. |
647 | Jump to the address N times, then fail. */ | |
fa9a63c5 RM |
648 | jump_n, |
649 | ||
25fe55af RS |
650 | /* Set the following two-byte relative address to the |
651 | subsequent two-byte number. The address *includes* the two | |
652 | bytes of number. */ | |
fa9a63c5 RM |
653 | set_number_at, |
654 | ||
fa9a63c5 RM |
655 | wordbeg, /* Succeeds if at word beginning. */ |
656 | wordend, /* Succeeds if at word end. */ | |
657 | ||
658 | wordbound, /* Succeeds if at a word boundary. */ | |
1fb352e0 | 659 | notwordbound, /* Succeeds if not at a word boundary. */ |
fa9a63c5 RM |
660 | |
661 | /* Matches any character whose syntax is specified. Followed by | |
25fe55af | 662 | a byte which contains a syntax code, e.g., Sword. */ |
fa9a63c5 RM |
663 | syntaxspec, |
664 | ||
665 | /* Matches any character whose syntax is not that specified. */ | |
1fb352e0 SM |
666 | notsyntaxspec |
667 | ||
668 | #ifdef emacs | |
669 | ,before_dot, /* Succeeds if before point. */ | |
670 | at_dot, /* Succeeds if at point. */ | |
671 | after_dot, /* Succeeds if after point. */ | |
b18215fc RS |
672 | |
673 | /* Matches any character whose category-set contains the specified | |
25fe55af RS |
674 | category. The operator is followed by a byte which contains a |
675 | category code (mnemonic ASCII character). */ | |
b18215fc RS |
676 | categoryspec, |
677 | ||
678 | /* Matches any character whose category-set does not contain the | |
679 | specified category. The operator is followed by a byte which | |
680 | contains the category code (mnemonic ASCII character). */ | |
681 | notcategoryspec | |
fa9a63c5 RM |
682 | #endif /* emacs */ |
683 | } re_opcode_t; | |
684 | \f | |
685 | /* Common operations on the compiled pattern. */ | |
686 | ||
687 | /* Store NUMBER in two contiguous bytes starting at DESTINATION. */ | |
688 | ||
689 | #define STORE_NUMBER(destination, number) \ | |
690 | do { \ | |
691 | (destination)[0] = (number) & 0377; \ | |
692 | (destination)[1] = (number) >> 8; \ | |
693 | } while (0) | |
694 | ||
695 | /* Same as STORE_NUMBER, except increment DESTINATION to | |
696 | the byte after where the number is stored. Therefore, DESTINATION | |
697 | must be an lvalue. */ | |
698 | ||
699 | #define STORE_NUMBER_AND_INCR(destination, number) \ | |
700 | do { \ | |
701 | STORE_NUMBER (destination, number); \ | |
702 | (destination) += 2; \ | |
703 | } while (0) | |
704 | ||
705 | /* Put into DESTINATION a number stored in two contiguous bytes starting | |
706 | at SOURCE. */ | |
707 | ||
708 | #define EXTRACT_NUMBER(destination, source) \ | |
709 | do { \ | |
710 | (destination) = *(source) & 0377; \ | |
711 | (destination) += SIGN_EXTEND_CHAR (*((source) + 1)) << 8; \ | |
712 | } while (0) | |
713 | ||
714 | #ifdef DEBUG | |
4bb91c68 | 715 | static void extract_number _RE_ARGS ((int *dest, re_char *source)); |
fa9a63c5 RM |
716 | static void |
717 | extract_number (dest, source) | |
718 | int *dest; | |
01618498 | 719 | re_char *source; |
fa9a63c5 | 720 | { |
5e69f11e | 721 | int temp = SIGN_EXTEND_CHAR (*(source + 1)); |
fa9a63c5 RM |
722 | *dest = *source & 0377; |
723 | *dest += temp << 8; | |
724 | } | |
725 | ||
4bb91c68 | 726 | # ifndef EXTRACT_MACROS /* To debug the macros. */ |
0b32bf0e SM |
727 | # undef EXTRACT_NUMBER |
728 | # define EXTRACT_NUMBER(dest, src) extract_number (&dest, src) | |
729 | # endif /* not EXTRACT_MACROS */ | |
fa9a63c5 RM |
730 | |
731 | #endif /* DEBUG */ | |
732 | ||
733 | /* Same as EXTRACT_NUMBER, except increment SOURCE to after the number. | |
734 | SOURCE must be an lvalue. */ | |
735 | ||
736 | #define EXTRACT_NUMBER_AND_INCR(destination, source) \ | |
737 | do { \ | |
738 | EXTRACT_NUMBER (destination, source); \ | |
25fe55af | 739 | (source) += 2; \ |
fa9a63c5 RM |
740 | } while (0) |
741 | ||
742 | #ifdef DEBUG | |
4bb91c68 SM |
743 | static void extract_number_and_incr _RE_ARGS ((int *destination, |
744 | re_char **source)); | |
fa9a63c5 RM |
745 | static void |
746 | extract_number_and_incr (destination, source) | |
747 | int *destination; | |
01618498 | 748 | re_char **source; |
5e69f11e | 749 | { |
fa9a63c5 RM |
750 | extract_number (destination, *source); |
751 | *source += 2; | |
752 | } | |
753 | ||
0b32bf0e SM |
754 | # ifndef EXTRACT_MACROS |
755 | # undef EXTRACT_NUMBER_AND_INCR | |
756 | # define EXTRACT_NUMBER_AND_INCR(dest, src) \ | |
fa9a63c5 | 757 | extract_number_and_incr (&dest, &src) |
0b32bf0e | 758 | # endif /* not EXTRACT_MACROS */ |
fa9a63c5 RM |
759 | |
760 | #endif /* DEBUG */ | |
761 | \f | |
b18215fc RS |
762 | /* Store a multibyte character in three contiguous bytes starting |
763 | DESTINATION, and increment DESTINATION to the byte after where the | |
25fe55af | 764 | character is stored. Therefore, DESTINATION must be an lvalue. */ |
b18215fc RS |
765 | |
766 | #define STORE_CHARACTER_AND_INCR(destination, character) \ | |
767 | do { \ | |
768 | (destination)[0] = (character) & 0377; \ | |
769 | (destination)[1] = ((character) >> 8) & 0377; \ | |
770 | (destination)[2] = (character) >> 16; \ | |
771 | (destination) += 3; \ | |
772 | } while (0) | |
773 | ||
774 | /* Put into DESTINATION a character stored in three contiguous bytes | |
25fe55af | 775 | starting at SOURCE. */ |
b18215fc RS |
776 | |
777 | #define EXTRACT_CHARACTER(destination, source) \ | |
778 | do { \ | |
779 | (destination) = ((source)[0] \ | |
780 | | ((source)[1] << 8) \ | |
781 | | ((source)[2] << 16)); \ | |
782 | } while (0) | |
783 | ||
784 | ||
785 | /* Macros for charset. */ | |
786 | ||
787 | /* Size of bitmap of charset P in bytes. P is a start of charset, | |
788 | i.e. *P is (re_opcode_t) charset or (re_opcode_t) charset_not. */ | |
789 | #define CHARSET_BITMAP_SIZE(p) ((p)[1] & 0x7F) | |
790 | ||
791 | /* Nonzero if charset P has range table. */ | |
25fe55af | 792 | #define CHARSET_RANGE_TABLE_EXISTS_P(p) ((p)[1] & 0x80) |
b18215fc RS |
793 | |
794 | /* Return the address of range table of charset P. But not the start | |
795 | of table itself, but the before where the number of ranges is | |
96cc36cc RS |
796 | stored. `2 +' means to skip re_opcode_t and size of bitmap, |
797 | and the 2 bytes of flags at the start of the range table. */ | |
798 | #define CHARSET_RANGE_TABLE(p) (&(p)[4 + CHARSET_BITMAP_SIZE (p)]) | |
799 | ||
800 | /* Extract the bit flags that start a range table. */ | |
801 | #define CHARSET_RANGE_TABLE_BITS(p) \ | |
802 | ((p)[2 + CHARSET_BITMAP_SIZE (p)] \ | |
803 | + (p)[3 + CHARSET_BITMAP_SIZE (p)] * 0x100) | |
b18215fc RS |
804 | |
805 | /* Test if C is listed in the bitmap of charset P. */ | |
806 | #define CHARSET_LOOKUP_BITMAP(p, c) \ | |
807 | ((c) < CHARSET_BITMAP_SIZE (p) * BYTEWIDTH \ | |
808 | && (p)[2 + (c) / BYTEWIDTH] & (1 << ((c) % BYTEWIDTH))) | |
809 | ||
810 | /* Return the address of end of RANGE_TABLE. COUNT is number of | |
25fe55af RS |
811 | ranges (which is a pair of (start, end)) in the RANGE_TABLE. `* 2' |
812 | is start of range and end of range. `* 3' is size of each start | |
b18215fc RS |
813 | and end. */ |
814 | #define CHARSET_RANGE_TABLE_END(range_table, count) \ | |
815 | ((range_table) + (count) * 2 * 3) | |
816 | ||
25fe55af | 817 | /* Test if C is in RANGE_TABLE. A flag NOT is negated if C is in. |
b18215fc RS |
818 | COUNT is number of ranges in RANGE_TABLE. */ |
819 | #define CHARSET_LOOKUP_RANGE_TABLE_RAW(not, c, range_table, count) \ | |
820 | do \ | |
821 | { \ | |
01618498 SM |
822 | re_wchar_t range_start, range_end; \ |
823 | re_char *p; \ | |
824 | re_char *range_table_end \ | |
b18215fc RS |
825 | = CHARSET_RANGE_TABLE_END ((range_table), (count)); \ |
826 | \ | |
827 | for (p = (range_table); p < range_table_end; p += 2 * 3) \ | |
828 | { \ | |
829 | EXTRACT_CHARACTER (range_start, p); \ | |
830 | EXTRACT_CHARACTER (range_end, p + 3); \ | |
831 | \ | |
832 | if (range_start <= (c) && (c) <= range_end) \ | |
833 | { \ | |
834 | (not) = !(not); \ | |
835 | break; \ | |
836 | } \ | |
837 | } \ | |
838 | } \ | |
839 | while (0) | |
840 | ||
841 | /* Test if C is in range table of CHARSET. The flag NOT is negated if | |
842 | C is listed in it. */ | |
843 | #define CHARSET_LOOKUP_RANGE_TABLE(not, c, charset) \ | |
844 | do \ | |
845 | { \ | |
846 | /* Number of ranges in range table. */ \ | |
847 | int count; \ | |
01618498 SM |
848 | re_char *range_table = CHARSET_RANGE_TABLE (charset); \ |
849 | \ | |
b18215fc RS |
850 | EXTRACT_NUMBER_AND_INCR (count, range_table); \ |
851 | CHARSET_LOOKUP_RANGE_TABLE_RAW ((not), (c), range_table, count); \ | |
852 | } \ | |
853 | while (0) | |
854 | \f | |
fa9a63c5 RM |
855 | /* If DEBUG is defined, Regex prints many voluminous messages about what |
856 | it is doing (if the variable `debug' is nonzero). If linked with the | |
857 | main program in `iregex.c', you can enter patterns and strings | |
858 | interactively. And if linked with the main program in `main.c' and | |
4bb91c68 | 859 | the other test files, you can run the already-written tests. */ |
fa9a63c5 RM |
860 | |
861 | #ifdef DEBUG | |
862 | ||
863 | /* We use standard I/O for debugging. */ | |
0b32bf0e | 864 | # include <stdio.h> |
fa9a63c5 RM |
865 | |
866 | /* It is useful to test things that ``must'' be true when debugging. */ | |
0b32bf0e | 867 | # include <assert.h> |
fa9a63c5 | 868 | |
99633e97 | 869 | static int debug = -100000; |
fa9a63c5 | 870 | |
0b32bf0e SM |
871 | # define DEBUG_STATEMENT(e) e |
872 | # define DEBUG_PRINT1(x) if (debug > 0) printf (x) | |
873 | # define DEBUG_PRINT2(x1, x2) if (debug > 0) printf (x1, x2) | |
874 | # define DEBUG_PRINT3(x1, x2, x3) if (debug > 0) printf (x1, x2, x3) | |
875 | # define DEBUG_PRINT4(x1, x2, x3, x4) if (debug > 0) printf (x1, x2, x3, x4) | |
876 | # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) \ | |
99633e97 | 877 | if (debug > 0) print_partial_compiled_pattern (s, e) |
0b32bf0e | 878 | # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) \ |
99633e97 | 879 | if (debug > 0) print_double_string (w, s1, sz1, s2, sz2) |
fa9a63c5 RM |
880 | |
881 | ||
882 | /* Print the fastmap in human-readable form. */ | |
883 | ||
884 | void | |
885 | print_fastmap (fastmap) | |
886 | char *fastmap; | |
887 | { | |
888 | unsigned was_a_range = 0; | |
5e69f11e RM |
889 | unsigned i = 0; |
890 | ||
fa9a63c5 RM |
891 | while (i < (1 << BYTEWIDTH)) |
892 | { | |
893 | if (fastmap[i++]) | |
894 | { | |
895 | was_a_range = 0; | |
25fe55af RS |
896 | putchar (i - 1); |
897 | while (i < (1 << BYTEWIDTH) && fastmap[i]) | |
898 | { | |
899 | was_a_range = 1; | |
900 | i++; | |
901 | } | |
fa9a63c5 | 902 | if (was_a_range) |
25fe55af RS |
903 | { |
904 | printf ("-"); | |
905 | putchar (i - 1); | |
906 | } | |
907 | } | |
fa9a63c5 | 908 | } |
5e69f11e | 909 | putchar ('\n'); |
fa9a63c5 RM |
910 | } |
911 | ||
912 | ||
913 | /* Print a compiled pattern string in human-readable form, starting at | |
914 | the START pointer into it and ending just before the pointer END. */ | |
915 | ||
916 | void | |
917 | print_partial_compiled_pattern (start, end) | |
01618498 SM |
918 | re_char *start; |
919 | re_char *end; | |
fa9a63c5 RM |
920 | { |
921 | int mcnt, mcnt2; | |
01618498 SM |
922 | re_char *p = start; |
923 | re_char *pend = end; | |
fa9a63c5 RM |
924 | |
925 | if (start == NULL) | |
926 | { | |
927 | printf ("(null)\n"); | |
928 | return; | |
929 | } | |
5e69f11e | 930 | |
fa9a63c5 RM |
931 | /* Loop over pattern commands. */ |
932 | while (p < pend) | |
933 | { | |
934 | printf ("%d:\t", p - start); | |
935 | ||
936 | switch ((re_opcode_t) *p++) | |
937 | { | |
25fe55af RS |
938 | case no_op: |
939 | printf ("/no_op"); | |
940 | break; | |
fa9a63c5 | 941 | |
99633e97 SM |
942 | case succeed: |
943 | printf ("/succeed"); | |
944 | break; | |
945 | ||
fa9a63c5 RM |
946 | case exactn: |
947 | mcnt = *p++; | |
25fe55af RS |
948 | printf ("/exactn/%d", mcnt); |
949 | do | |
fa9a63c5 | 950 | { |
25fe55af | 951 | putchar ('/'); |
fa9a63c5 | 952 | putchar (*p++); |
25fe55af RS |
953 | } |
954 | while (--mcnt); | |
955 | break; | |
fa9a63c5 RM |
956 | |
957 | case start_memory: | |
505bde11 | 958 | printf ("/start_memory/%d", *p++); |
25fe55af | 959 | break; |
fa9a63c5 RM |
960 | |
961 | case stop_memory: | |
505bde11 | 962 | printf ("/stop_memory/%d", *p++); |
25fe55af | 963 | break; |
fa9a63c5 RM |
964 | |
965 | case duplicate: | |
966 | printf ("/duplicate/%d", *p++); | |
967 | break; | |
968 | ||
969 | case anychar: | |
970 | printf ("/anychar"); | |
971 | break; | |
972 | ||
973 | case charset: | |
25fe55af RS |
974 | case charset_not: |
975 | { | |
976 | register int c, last = -100; | |
fa9a63c5 | 977 | register int in_range = 0; |
99633e97 SM |
978 | int length = CHARSET_BITMAP_SIZE (p - 1); |
979 | int has_range_table = CHARSET_RANGE_TABLE_EXISTS_P (p - 1); | |
fa9a63c5 RM |
980 | |
981 | printf ("/charset [%s", | |
25fe55af | 982 | (re_opcode_t) *(p - 1) == charset_not ? "^" : ""); |
5e69f11e | 983 | |
25fe55af | 984 | assert (p + *p < pend); |
fa9a63c5 | 985 | |
25fe55af | 986 | for (c = 0; c < 256; c++) |
96cc36cc | 987 | if (c / 8 < length |
fa9a63c5 RM |
988 | && (p[1 + (c/8)] & (1 << (c % 8)))) |
989 | { | |
990 | /* Are we starting a range? */ | |
991 | if (last + 1 == c && ! in_range) | |
992 | { | |
993 | putchar ('-'); | |
994 | in_range = 1; | |
995 | } | |
996 | /* Have we broken a range? */ | |
997 | else if (last + 1 != c && in_range) | |
96cc36cc | 998 | { |
fa9a63c5 RM |
999 | putchar (last); |
1000 | in_range = 0; | |
1001 | } | |
5e69f11e | 1002 | |
fa9a63c5 RM |
1003 | if (! in_range) |
1004 | putchar (c); | |
1005 | ||
1006 | last = c; | |
25fe55af | 1007 | } |
fa9a63c5 RM |
1008 | |
1009 | if (in_range) | |
1010 | putchar (last); | |
1011 | ||
1012 | putchar (']'); | |
1013 | ||
99633e97 | 1014 | p += 1 + length; |
96cc36cc | 1015 | |
96cc36cc | 1016 | if (has_range_table) |
99633e97 SM |
1017 | { |
1018 | int count; | |
1019 | printf ("has-range-table"); | |
1020 | ||
1021 | /* ??? Should print the range table; for now, just skip it. */ | |
1022 | p += 2; /* skip range table bits */ | |
1023 | EXTRACT_NUMBER_AND_INCR (count, p); | |
1024 | p = CHARSET_RANGE_TABLE_END (p, count); | |
1025 | } | |
fa9a63c5 RM |
1026 | } |
1027 | break; | |
1028 | ||
1029 | case begline: | |
1030 | printf ("/begline"); | |
25fe55af | 1031 | break; |
fa9a63c5 RM |
1032 | |
1033 | case endline: | |
25fe55af RS |
1034 | printf ("/endline"); |
1035 | break; | |
fa9a63c5 RM |
1036 | |
1037 | case on_failure_jump: | |
25fe55af RS |
1038 | extract_number_and_incr (&mcnt, &p); |
1039 | printf ("/on_failure_jump to %d", p + mcnt - start); | |
1040 | break; | |
fa9a63c5 RM |
1041 | |
1042 | case on_failure_keep_string_jump: | |
25fe55af RS |
1043 | extract_number_and_incr (&mcnt, &p); |
1044 | printf ("/on_failure_keep_string_jump to %d", p + mcnt - start); | |
1045 | break; | |
fa9a63c5 | 1046 | |
0683b6fa SM |
1047 | case on_failure_jump_nastyloop: |
1048 | extract_number_and_incr (&mcnt, &p); | |
1049 | printf ("/on_failure_jump_nastyloop to %d", p + mcnt - start); | |
1050 | break; | |
1051 | ||
505bde11 | 1052 | case on_failure_jump_loop: |
fa9a63c5 | 1053 | extract_number_and_incr (&mcnt, &p); |
505bde11 | 1054 | printf ("/on_failure_jump_loop to %d", p + mcnt - start); |
5e69f11e RM |
1055 | break; |
1056 | ||
505bde11 | 1057 | case on_failure_jump_smart: |
fa9a63c5 | 1058 | extract_number_and_incr (&mcnt, &p); |
505bde11 | 1059 | printf ("/on_failure_jump_smart to %d", p + mcnt - start); |
5e69f11e RM |
1060 | break; |
1061 | ||
25fe55af | 1062 | case jump: |
fa9a63c5 | 1063 | extract_number_and_incr (&mcnt, &p); |
25fe55af | 1064 | printf ("/jump to %d", p + mcnt - start); |
fa9a63c5 RM |
1065 | break; |
1066 | ||
25fe55af RS |
1067 | case succeed_n: |
1068 | extract_number_and_incr (&mcnt, &p); | |
1069 | extract_number_and_incr (&mcnt2, &p); | |
99633e97 | 1070 | printf ("/succeed_n to %d, %d times", p - 2 + mcnt - start, mcnt2); |
25fe55af | 1071 | break; |
5e69f11e | 1072 | |
25fe55af RS |
1073 | case jump_n: |
1074 | extract_number_and_incr (&mcnt, &p); | |
1075 | extract_number_and_incr (&mcnt2, &p); | |
99633e97 | 1076 | printf ("/jump_n to %d, %d times", p - 2 + mcnt - start, mcnt2); |
25fe55af | 1077 | break; |
5e69f11e | 1078 | |
25fe55af RS |
1079 | case set_number_at: |
1080 | extract_number_and_incr (&mcnt, &p); | |
1081 | extract_number_and_incr (&mcnt2, &p); | |
99633e97 | 1082 | printf ("/set_number_at location %d to %d", p - 2 + mcnt - start, mcnt2); |
25fe55af | 1083 | break; |
5e69f11e | 1084 | |
25fe55af | 1085 | case wordbound: |
fa9a63c5 RM |
1086 | printf ("/wordbound"); |
1087 | break; | |
1088 | ||
1089 | case notwordbound: | |
1090 | printf ("/notwordbound"); | |
25fe55af | 1091 | break; |
fa9a63c5 RM |
1092 | |
1093 | case wordbeg: | |
1094 | printf ("/wordbeg"); | |
1095 | break; | |
5e69f11e | 1096 | |
fa9a63c5 RM |
1097 | case wordend: |
1098 | printf ("/wordend"); | |
5e69f11e | 1099 | |
1fb352e0 SM |
1100 | case syntaxspec: |
1101 | printf ("/syntaxspec"); | |
1102 | mcnt = *p++; | |
1103 | printf ("/%d", mcnt); | |
1104 | break; | |
1105 | ||
1106 | case notsyntaxspec: | |
1107 | printf ("/notsyntaxspec"); | |
1108 | mcnt = *p++; | |
1109 | printf ("/%d", mcnt); | |
1110 | break; | |
1111 | ||
0b32bf0e | 1112 | # ifdef emacs |
fa9a63c5 RM |
1113 | case before_dot: |
1114 | printf ("/before_dot"); | |
25fe55af | 1115 | break; |
fa9a63c5 RM |
1116 | |
1117 | case at_dot: | |
1118 | printf ("/at_dot"); | |
25fe55af | 1119 | break; |
fa9a63c5 RM |
1120 | |
1121 | case after_dot: | |
1122 | printf ("/after_dot"); | |
25fe55af | 1123 | break; |
fa9a63c5 | 1124 | |
1fb352e0 SM |
1125 | case categoryspec: |
1126 | printf ("/categoryspec"); | |
fa9a63c5 RM |
1127 | mcnt = *p++; |
1128 | printf ("/%d", mcnt); | |
25fe55af | 1129 | break; |
5e69f11e | 1130 | |
1fb352e0 SM |
1131 | case notcategoryspec: |
1132 | printf ("/notcategoryspec"); | |
fa9a63c5 RM |
1133 | mcnt = *p++; |
1134 | printf ("/%d", mcnt); | |
1135 | break; | |
0b32bf0e | 1136 | # endif /* emacs */ |
fa9a63c5 | 1137 | |
fa9a63c5 RM |
1138 | case begbuf: |
1139 | printf ("/begbuf"); | |
25fe55af | 1140 | break; |
fa9a63c5 RM |
1141 | |
1142 | case endbuf: | |
1143 | printf ("/endbuf"); | |
25fe55af | 1144 | break; |
fa9a63c5 | 1145 | |
25fe55af RS |
1146 | default: |
1147 | printf ("?%d", *(p-1)); | |
fa9a63c5 RM |
1148 | } |
1149 | ||
1150 | putchar ('\n'); | |
1151 | } | |
1152 | ||
1153 | printf ("%d:\tend of pattern.\n", p - start); | |
1154 | } | |
1155 | ||
1156 | ||
1157 | void | |
1158 | print_compiled_pattern (bufp) | |
1159 | struct re_pattern_buffer *bufp; | |
1160 | { | |
01618498 | 1161 | re_char *buffer = bufp->buffer; |
fa9a63c5 RM |
1162 | |
1163 | print_partial_compiled_pattern (buffer, buffer + bufp->used); | |
4bb91c68 SM |
1164 | printf ("%ld bytes used/%ld bytes allocated.\n", |
1165 | bufp->used, bufp->allocated); | |
fa9a63c5 RM |
1166 | |
1167 | if (bufp->fastmap_accurate && bufp->fastmap) | |
1168 | { | |
1169 | printf ("fastmap: "); | |
1170 | print_fastmap (bufp->fastmap); | |
1171 | } | |
1172 | ||
1173 | printf ("re_nsub: %d\t", bufp->re_nsub); | |
1174 | printf ("regs_alloc: %d\t", bufp->regs_allocated); | |
1175 | printf ("can_be_null: %d\t", bufp->can_be_null); | |
fa9a63c5 RM |
1176 | printf ("no_sub: %d\t", bufp->no_sub); |
1177 | printf ("not_bol: %d\t", bufp->not_bol); | |
1178 | printf ("not_eol: %d\t", bufp->not_eol); | |
4bb91c68 | 1179 | printf ("syntax: %lx\n", bufp->syntax); |
505bde11 | 1180 | fflush (stdout); |
fa9a63c5 RM |
1181 | /* Perhaps we should print the translate table? */ |
1182 | } | |
1183 | ||
1184 | ||
1185 | void | |
1186 | print_double_string (where, string1, size1, string2, size2) | |
66f0296e SM |
1187 | re_char *where; |
1188 | re_char *string1; | |
1189 | re_char *string2; | |
fa9a63c5 RM |
1190 | int size1; |
1191 | int size2; | |
1192 | { | |
4bb91c68 | 1193 | int this_char; |
5e69f11e | 1194 | |
fa9a63c5 RM |
1195 | if (where == NULL) |
1196 | printf ("(null)"); | |
1197 | else | |
1198 | { | |
1199 | if (FIRST_STRING_P (where)) | |
25fe55af RS |
1200 | { |
1201 | for (this_char = where - string1; this_char < size1; this_char++) | |
1202 | putchar (string1[this_char]); | |
fa9a63c5 | 1203 | |
25fe55af RS |
1204 | where = string2; |
1205 | } | |
fa9a63c5 RM |
1206 | |
1207 | for (this_char = where - string2; this_char < size2; this_char++) | |
25fe55af | 1208 | putchar (string2[this_char]); |
fa9a63c5 RM |
1209 | } |
1210 | } | |
1211 | ||
1212 | #else /* not DEBUG */ | |
1213 | ||
0b32bf0e SM |
1214 | # undef assert |
1215 | # define assert(e) | |
fa9a63c5 | 1216 | |
0b32bf0e SM |
1217 | # define DEBUG_STATEMENT(e) |
1218 | # define DEBUG_PRINT1(x) | |
1219 | # define DEBUG_PRINT2(x1, x2) | |
1220 | # define DEBUG_PRINT3(x1, x2, x3) | |
1221 | # define DEBUG_PRINT4(x1, x2, x3, x4) | |
1222 | # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) | |
1223 | # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) | |
fa9a63c5 RM |
1224 | |
1225 | #endif /* not DEBUG */ | |
1226 | \f | |
1227 | /* Set by `re_set_syntax' to the current regexp syntax to recognize. Can | |
1228 | also be assigned to arbitrarily: each pattern buffer stores its own | |
1229 | syntax, so it can be changed between regex compilations. */ | |
1230 | /* This has no initializer because initialized variables in Emacs | |
1231 | become read-only after dumping. */ | |
1232 | reg_syntax_t re_syntax_options; | |
1233 | ||
1234 | ||
1235 | /* Specify the precise syntax of regexps for compilation. This provides | |
1236 | for compatibility for various utilities which historically have | |
1237 | different, incompatible syntaxes. | |
1238 | ||
1239 | The argument SYNTAX is a bit mask comprised of the various bits | |
4bb91c68 | 1240 | defined in regex.h. We return the old syntax. */ |
fa9a63c5 RM |
1241 | |
1242 | reg_syntax_t | |
1243 | re_set_syntax (syntax) | |
1244 | reg_syntax_t syntax; | |
1245 | { | |
1246 | reg_syntax_t ret = re_syntax_options; | |
5e69f11e | 1247 | |
fa9a63c5 RM |
1248 | re_syntax_options = syntax; |
1249 | return ret; | |
1250 | } | |
c0f9ea08 | 1251 | WEAK_ALIAS (__re_set_syntax, re_set_syntax) |
fa9a63c5 RM |
1252 | \f |
1253 | /* This table gives an error message for each of the error codes listed | |
4bb91c68 | 1254 | in regex.h. Obviously the order here has to be same as there. |
fa9a63c5 | 1255 | POSIX doesn't require that we do anything for REG_NOERROR, |
4bb91c68 | 1256 | but why not be nice? */ |
fa9a63c5 RM |
1257 | |
1258 | static const char *re_error_msgid[] = | |
5e69f11e RM |
1259 | { |
1260 | gettext_noop ("Success"), /* REG_NOERROR */ | |
1261 | gettext_noop ("No match"), /* REG_NOMATCH */ | |
1262 | gettext_noop ("Invalid regular expression"), /* REG_BADPAT */ | |
1263 | gettext_noop ("Invalid collation character"), /* REG_ECOLLATE */ | |
1264 | gettext_noop ("Invalid character class name"), /* REG_ECTYPE */ | |
1265 | gettext_noop ("Trailing backslash"), /* REG_EESCAPE */ | |
1266 | gettext_noop ("Invalid back reference"), /* REG_ESUBREG */ | |
1267 | gettext_noop ("Unmatched [ or [^"), /* REG_EBRACK */ | |
1268 | gettext_noop ("Unmatched ( or \\("), /* REG_EPAREN */ | |
1269 | gettext_noop ("Unmatched \\{"), /* REG_EBRACE */ | |
1270 | gettext_noop ("Invalid content of \\{\\}"), /* REG_BADBR */ | |
1271 | gettext_noop ("Invalid range end"), /* REG_ERANGE */ | |
1272 | gettext_noop ("Memory exhausted"), /* REG_ESPACE */ | |
1273 | gettext_noop ("Invalid preceding regular expression"), /* REG_BADRPT */ | |
1274 | gettext_noop ("Premature end of regular expression"), /* REG_EEND */ | |
1275 | gettext_noop ("Regular expression too big"), /* REG_ESIZE */ | |
1276 | gettext_noop ("Unmatched ) or \\)"), /* REG_ERPAREN */ | |
fa9a63c5 RM |
1277 | }; |
1278 | \f | |
4bb91c68 | 1279 | /* Avoiding alloca during matching, to placate r_alloc. */ |
fa9a63c5 RM |
1280 | |
1281 | /* Define MATCH_MAY_ALLOCATE unless we need to make sure that the | |
1282 | searching and matching functions should not call alloca. On some | |
1283 | systems, alloca is implemented in terms of malloc, and if we're | |
1284 | using the relocating allocator routines, then malloc could cause a | |
1285 | relocation, which might (if the strings being searched are in the | |
1286 | ralloc heap) shift the data out from underneath the regexp | |
1287 | routines. | |
1288 | ||
5e69f11e | 1289 | Here's another reason to avoid allocation: Emacs |
fa9a63c5 RM |
1290 | processes input from X in a signal handler; processing X input may |
1291 | call malloc; if input arrives while a matching routine is calling | |
1292 | malloc, then we're scrod. But Emacs can't just block input while | |
1293 | calling matching routines; then we don't notice interrupts when | |
1294 | they come in. So, Emacs blocks input around all regexp calls | |
1295 | except the matching calls, which it leaves unprotected, in the | |
1296 | faith that they will not malloc. */ | |
1297 | ||
1298 | /* Normally, this is fine. */ | |
1299 | #define MATCH_MAY_ALLOCATE | |
1300 | ||
1301 | /* When using GNU C, we are not REALLY using the C alloca, no matter | |
1302 | what config.h may say. So don't take precautions for it. */ | |
1303 | #ifdef __GNUC__ | |
0b32bf0e | 1304 | # undef C_ALLOCA |
fa9a63c5 RM |
1305 | #endif |
1306 | ||
1307 | /* The match routines may not allocate if (1) they would do it with malloc | |
1308 | and (2) it's not safe for them to use malloc. | |
1309 | Note that if REL_ALLOC is defined, matching would not use malloc for the | |
1310 | failure stack, but we would still use it for the register vectors; | |
4bb91c68 | 1311 | so REL_ALLOC should not affect this. */ |
0b32bf0e SM |
1312 | #if (defined C_ALLOCA || defined REGEX_MALLOC) && defined emacs |
1313 | # undef MATCH_MAY_ALLOCATE | |
fa9a63c5 RM |
1314 | #endif |
1315 | ||
1316 | \f | |
1317 | /* Failure stack declarations and macros; both re_compile_fastmap and | |
1318 | re_match_2 use a failure stack. These have to be macros because of | |
1319 | REGEX_ALLOCATE_STACK. */ | |
5e69f11e | 1320 | |
fa9a63c5 | 1321 | |
320a2a73 | 1322 | /* Approximate number of failure points for which to initially allocate space |
fa9a63c5 RM |
1323 | when matching. If this number is exceeded, we allocate more |
1324 | space, so it is not a hard limit. */ | |
1325 | #ifndef INIT_FAILURE_ALLOC | |
0b32bf0e | 1326 | # define INIT_FAILURE_ALLOC 20 |
fa9a63c5 RM |
1327 | #endif |
1328 | ||
1329 | /* Roughly the maximum number of failure points on the stack. Would be | |
320a2a73 | 1330 | exactly that if always used TYPICAL_FAILURE_SIZE items each time we failed. |
fa9a63c5 | 1331 | This is a variable only so users of regex can assign to it; we never |
ada30c0e SM |
1332 | change it ourselves. We always multiply it by TYPICAL_FAILURE_SIZE |
1333 | before using it, so it should probably be a byte-count instead. */ | |
c0f9ea08 SM |
1334 | # if defined MATCH_MAY_ALLOCATE |
1335 | /* Note that 4400 was enough to cause a crash on Alpha OSF/1, | |
320a2a73 KH |
1336 | whose default stack limit is 2mb. In order for a larger |
1337 | value to work reliably, you have to try to make it accord | |
1338 | with the process stack limit. */ | |
c0f9ea08 SM |
1339 | size_t re_max_failures = 40000; |
1340 | # else | |
1341 | size_t re_max_failures = 4000; | |
1342 | # endif | |
fa9a63c5 RM |
1343 | |
1344 | union fail_stack_elt | |
1345 | { | |
01618498 | 1346 | re_char *pointer; |
c0f9ea08 SM |
1347 | /* This should be the biggest `int' that's no bigger than a pointer. */ |
1348 | long integer; | |
fa9a63c5 RM |
1349 | }; |
1350 | ||
1351 | typedef union fail_stack_elt fail_stack_elt_t; | |
1352 | ||
1353 | typedef struct | |
1354 | { | |
1355 | fail_stack_elt_t *stack; | |
c0f9ea08 SM |
1356 | size_t size; |
1357 | size_t avail; /* Offset of next open position. */ | |
1358 | size_t frame; /* Offset of the cur constructed frame. */ | |
fa9a63c5 RM |
1359 | } fail_stack_type; |
1360 | ||
505bde11 | 1361 | #define FAIL_STACK_EMPTY() (fail_stack.frame == 0) |
fa9a63c5 RM |
1362 | #define FAIL_STACK_FULL() (fail_stack.avail == fail_stack.size) |
1363 | ||
1364 | ||
1365 | /* Define macros to initialize and free the failure stack. | |
1366 | Do `return -2' if the alloc fails. */ | |
1367 | ||
1368 | #ifdef MATCH_MAY_ALLOCATE | |
0b32bf0e | 1369 | # define INIT_FAIL_STACK() \ |
fa9a63c5 RM |
1370 | do { \ |
1371 | fail_stack.stack = (fail_stack_elt_t *) \ | |
320a2a73 KH |
1372 | REGEX_ALLOCATE_STACK (INIT_FAILURE_ALLOC * TYPICAL_FAILURE_SIZE \ |
1373 | * sizeof (fail_stack_elt_t)); \ | |
fa9a63c5 RM |
1374 | \ |
1375 | if (fail_stack.stack == NULL) \ | |
1376 | return -2; \ | |
1377 | \ | |
1378 | fail_stack.size = INIT_FAILURE_ALLOC; \ | |
1379 | fail_stack.avail = 0; \ | |
505bde11 | 1380 | fail_stack.frame = 0; \ |
fa9a63c5 RM |
1381 | } while (0) |
1382 | ||
0b32bf0e | 1383 | # define RESET_FAIL_STACK() REGEX_FREE_STACK (fail_stack.stack) |
fa9a63c5 | 1384 | #else |
0b32bf0e | 1385 | # define INIT_FAIL_STACK() \ |
fa9a63c5 RM |
1386 | do { \ |
1387 | fail_stack.avail = 0; \ | |
505bde11 | 1388 | fail_stack.frame = 0; \ |
fa9a63c5 RM |
1389 | } while (0) |
1390 | ||
0b32bf0e | 1391 | # define RESET_FAIL_STACK() ((void)0) |
fa9a63c5 RM |
1392 | #endif |
1393 | ||
1394 | ||
320a2a73 KH |
1395 | /* Double the size of FAIL_STACK, up to a limit |
1396 | which allows approximately `re_max_failures' items. | |
fa9a63c5 RM |
1397 | |
1398 | Return 1 if succeeds, and 0 if either ran out of memory | |
5e69f11e RM |
1399 | allocating space for it or it was already too large. |
1400 | ||
4bb91c68 | 1401 | REGEX_REALLOCATE_STACK requires `destination' be declared. */ |
fa9a63c5 | 1402 | |
320a2a73 KH |
1403 | /* Factor to increase the failure stack size by |
1404 | when we increase it. | |
1405 | This used to be 2, but 2 was too wasteful | |
1406 | because the old discarded stacks added up to as much space | |
1407 | were as ultimate, maximum-size stack. */ | |
1408 | #define FAIL_STACK_GROWTH_FACTOR 4 | |
1409 | ||
1410 | #define GROW_FAIL_STACK(fail_stack) \ | |
eead07d6 KH |
1411 | (((fail_stack).size * sizeof (fail_stack_elt_t) \ |
1412 | >= re_max_failures * TYPICAL_FAILURE_SIZE) \ | |
fa9a63c5 | 1413 | ? 0 \ |
320a2a73 KH |
1414 | : ((fail_stack).stack \ |
1415 | = (fail_stack_elt_t *) \ | |
25fe55af RS |
1416 | REGEX_REALLOCATE_STACK ((fail_stack).stack, \ |
1417 | (fail_stack).size * sizeof (fail_stack_elt_t), \ | |
320a2a73 KH |
1418 | MIN (re_max_failures * TYPICAL_FAILURE_SIZE, \ |
1419 | ((fail_stack).size * sizeof (fail_stack_elt_t) \ | |
1420 | * FAIL_STACK_GROWTH_FACTOR))), \ | |
fa9a63c5 RM |
1421 | \ |
1422 | (fail_stack).stack == NULL \ | |
1423 | ? 0 \ | |
6453db45 KH |
1424 | : ((fail_stack).size \ |
1425 | = (MIN (re_max_failures * TYPICAL_FAILURE_SIZE, \ | |
1426 | ((fail_stack).size * sizeof (fail_stack_elt_t) \ | |
1427 | * FAIL_STACK_GROWTH_FACTOR)) \ | |
1428 | / sizeof (fail_stack_elt_t)), \ | |
25fe55af | 1429 | 1))) |
fa9a63c5 RM |
1430 | |
1431 | ||
fa9a63c5 RM |
1432 | /* Push a pointer value onto the failure stack. |
1433 | Assumes the variable `fail_stack'. Probably should only | |
4bb91c68 | 1434 | be called from within `PUSH_FAILURE_POINT'. */ |
fa9a63c5 | 1435 | #define PUSH_FAILURE_POINTER(item) \ |
01618498 | 1436 | fail_stack.stack[fail_stack.avail++].pointer = (item) |
fa9a63c5 RM |
1437 | |
1438 | /* This pushes an integer-valued item onto the failure stack. | |
1439 | Assumes the variable `fail_stack'. Probably should only | |
4bb91c68 | 1440 | be called from within `PUSH_FAILURE_POINT'. */ |
fa9a63c5 RM |
1441 | #define PUSH_FAILURE_INT(item) \ |
1442 | fail_stack.stack[fail_stack.avail++].integer = (item) | |
1443 | ||
1444 | /* Push a fail_stack_elt_t value onto the failure stack. | |
1445 | Assumes the variable `fail_stack'. Probably should only | |
4bb91c68 | 1446 | be called from within `PUSH_FAILURE_POINT'. */ |
fa9a63c5 RM |
1447 | #define PUSH_FAILURE_ELT(item) \ |
1448 | fail_stack.stack[fail_stack.avail++] = (item) | |
1449 | ||
1450 | /* These three POP... operations complement the three PUSH... operations. | |
1451 | All assume that `fail_stack' is nonempty. */ | |
1452 | #define POP_FAILURE_POINTER() fail_stack.stack[--fail_stack.avail].pointer | |
1453 | #define POP_FAILURE_INT() fail_stack.stack[--fail_stack.avail].integer | |
1454 | #define POP_FAILURE_ELT() fail_stack.stack[--fail_stack.avail] | |
1455 | ||
505bde11 SM |
1456 | /* Individual items aside from the registers. */ |
1457 | #define NUM_NONREG_ITEMS 3 | |
1458 | ||
1459 | /* Used to examine the stack (to detect infinite loops). */ | |
1460 | #define FAILURE_PAT(h) fail_stack.stack[(h) - 1].pointer | |
66f0296e | 1461 | #define FAILURE_STR(h) (fail_stack.stack[(h) - 2].pointer) |
505bde11 SM |
1462 | #define NEXT_FAILURE_HANDLE(h) fail_stack.stack[(h) - 3].integer |
1463 | #define TOP_FAILURE_HANDLE() fail_stack.frame | |
fa9a63c5 RM |
1464 | |
1465 | ||
505bde11 SM |
1466 | #define ENSURE_FAIL_STACK(space) \ |
1467 | while (REMAINING_AVAIL_SLOTS <= space) { \ | |
1468 | if (!GROW_FAIL_STACK (fail_stack)) \ | |
1469 | return -2; \ | |
1470 | DEBUG_PRINT2 ("\n Doubled stack; size now: %d\n", (fail_stack).size);\ | |
1471 | DEBUG_PRINT2 (" slots available: %d\n", REMAINING_AVAIL_SLOTS);\ | |
1472 | } | |
1473 | ||
1474 | /* Push register NUM onto the stack. */ | |
1475 | #define PUSH_FAILURE_REG(num) \ | |
1476 | do { \ | |
1477 | char *destination; \ | |
1478 | ENSURE_FAIL_STACK(3); \ | |
1479 | DEBUG_PRINT4 (" Push reg %d (spanning %p -> %p)\n", \ | |
1480 | num, regstart[num], regend[num]); \ | |
1481 | PUSH_FAILURE_POINTER (regstart[num]); \ | |
1482 | PUSH_FAILURE_POINTER (regend[num]); \ | |
1483 | PUSH_FAILURE_INT (num); \ | |
1484 | } while (0) | |
1485 | ||
01618498 SM |
1486 | /* Change the counter's value to VAL, but make sure that it will |
1487 | be reset when backtracking. */ | |
1488 | #define PUSH_NUMBER(ptr,val) \ | |
dc1e502d SM |
1489 | do { \ |
1490 | char *destination; \ | |
1491 | int c; \ | |
1492 | ENSURE_FAIL_STACK(3); \ | |
1493 | EXTRACT_NUMBER (c, ptr); \ | |
01618498 | 1494 | DEBUG_PRINT4 (" Push number %p = %d -> %d\n", ptr, c, val); \ |
dc1e502d SM |
1495 | PUSH_FAILURE_INT (c); \ |
1496 | PUSH_FAILURE_POINTER (ptr); \ | |
1497 | PUSH_FAILURE_INT (-1); \ | |
01618498 | 1498 | STORE_NUMBER (ptr, val); \ |
dc1e502d SM |
1499 | } while (0) |
1500 | ||
505bde11 | 1501 | /* Pop a saved register off the stack. */ |
dc1e502d | 1502 | #define POP_FAILURE_REG_OR_COUNT() \ |
505bde11 SM |
1503 | do { \ |
1504 | int reg = POP_FAILURE_INT (); \ | |
dc1e502d SM |
1505 | if (reg == -1) \ |
1506 | { \ | |
1507 | /* It's a counter. */ \ | |
6dcf2d0e SM |
1508 | /* Here, we discard `const', making re_match non-reentrant. */ \ |
1509 | unsigned char *ptr = (unsigned char*) POP_FAILURE_POINTER (); \ | |
dc1e502d SM |
1510 | reg = POP_FAILURE_INT (); \ |
1511 | STORE_NUMBER (ptr, reg); \ | |
1512 | DEBUG_PRINT3 (" Pop counter %p = %d\n", ptr, reg); \ | |
1513 | } \ | |
1514 | else \ | |
1515 | { \ | |
1516 | regend[reg] = POP_FAILURE_POINTER (); \ | |
1517 | regstart[reg] = POP_FAILURE_POINTER (); \ | |
1518 | DEBUG_PRINT4 (" Pop reg %d (spanning %p -> %p)\n", \ | |
1519 | reg, regstart[reg], regend[reg]); \ | |
1520 | } \ | |
505bde11 SM |
1521 | } while (0) |
1522 | ||
f6df485f RS |
1523 | /* Discard a saved register off the stack. */ |
1524 | #define DISCARD_FAILURE_REG_OR_COUNT() \ | |
1525 | do { \ | |
1526 | int reg = POP_FAILURE_INT (); \ | |
1527 | if (reg == -1) \ | |
1528 | { \ | |
1529 | /* It's a counter. */ \ | |
1530 | POP_FAILURE_POINTER (); \ | |
1531 | reg = POP_FAILURE_INT (); \ | |
1532 | DEBUG_PRINT3 (" Discard counter %p = %d\n", ptr, reg); \ | |
1533 | } \ | |
1534 | else \ | |
1535 | { \ | |
1536 | POP_FAILURE_POINTER (); \ | |
1537 | POP_FAILURE_POINTER (); \ | |
1538 | DEBUG_PRINT4 (" Discard reg %d (spanning %p -> %p)\n", \ | |
1539 | reg, regstart[reg], regend[reg]); \ | |
1540 | } \ | |
1541 | } while (0) | |
1542 | ||
505bde11 SM |
1543 | /* Check that we are not stuck in an infinite loop. */ |
1544 | #define CHECK_INFINITE_LOOP(pat_cur, string_place) \ | |
1545 | do { \ | |
f6df485f | 1546 | int failure = TOP_FAILURE_HANDLE (); \ |
505bde11 | 1547 | /* Check for infinite matching loops */ \ |
f6df485f RS |
1548 | while (failure > 0 \ |
1549 | && (FAILURE_STR (failure) == string_place \ | |
1550 | || FAILURE_STR (failure) == NULL)) \ | |
505bde11 SM |
1551 | { \ |
1552 | assert (FAILURE_PAT (failure) >= bufp->buffer \ | |
66f0296e | 1553 | && FAILURE_PAT (failure) <= bufp->buffer + bufp->used); \ |
505bde11 | 1554 | if (FAILURE_PAT (failure) == pat_cur) \ |
f6df485f RS |
1555 | { \ |
1556 | while (fail_stack.frame < fail_stack.avail) \ | |
1557 | DISCARD_FAILURE_REG_OR_COUNT (); \ | |
1558 | goto fail; \ | |
1559 | } \ | |
66f0296e | 1560 | DEBUG_PRINT2 (" Other pattern: %p\n", FAILURE_PAT (failure)); \ |
505bde11 SM |
1561 | failure = NEXT_FAILURE_HANDLE(failure); \ |
1562 | } \ | |
1563 | DEBUG_PRINT2 (" Other string: %p\n", FAILURE_STR (failure)); \ | |
1564 | } while (0) | |
1565 | ||
fa9a63c5 | 1566 | /* Push the information about the state we will need |
5e69f11e RM |
1567 | if we ever fail back to it. |
1568 | ||
505bde11 | 1569 | Requires variables fail_stack, regstart, regend and |
320a2a73 | 1570 | num_regs be declared. GROW_FAIL_STACK requires `destination' be |
fa9a63c5 | 1571 | declared. |
5e69f11e | 1572 | |
fa9a63c5 RM |
1573 | Does `return FAILURE_CODE' if runs out of memory. */ |
1574 | ||
505bde11 SM |
1575 | #define PUSH_FAILURE_POINT(pattern, string_place) \ |
1576 | do { \ | |
1577 | char *destination; \ | |
1578 | /* Must be int, so when we don't save any registers, the arithmetic \ | |
1579 | of 0 + -1 isn't done as unsigned. */ \ | |
1580 | \ | |
505bde11 | 1581 | DEBUG_STATEMENT (nfailure_points_pushed++); \ |
4bb91c68 | 1582 | DEBUG_PRINT1 ("\nPUSH_FAILURE_POINT:\n"); \ |
505bde11 SM |
1583 | DEBUG_PRINT2 (" Before push, next avail: %d\n", (fail_stack).avail); \ |
1584 | DEBUG_PRINT2 (" size: %d\n", (fail_stack).size);\ | |
1585 | \ | |
1586 | ENSURE_FAIL_STACK (NUM_NONREG_ITEMS); \ | |
1587 | \ | |
1588 | DEBUG_PRINT1 ("\n"); \ | |
1589 | \ | |
1590 | DEBUG_PRINT2 (" Push frame index: %d\n", fail_stack.frame); \ | |
1591 | PUSH_FAILURE_INT (fail_stack.frame); \ | |
1592 | \ | |
1593 | DEBUG_PRINT2 (" Push string %p: `", string_place); \ | |
1594 | DEBUG_PRINT_DOUBLE_STRING (string_place, string1, size1, string2, size2);\ | |
1595 | DEBUG_PRINT1 ("'\n"); \ | |
1596 | PUSH_FAILURE_POINTER (string_place); \ | |
1597 | \ | |
1598 | DEBUG_PRINT2 (" Push pattern %p: ", pattern); \ | |
1599 | DEBUG_PRINT_COMPILED_PATTERN (bufp, pattern, pend); \ | |
1600 | PUSH_FAILURE_POINTER (pattern); \ | |
1601 | \ | |
1602 | /* Close the frame by moving the frame pointer past it. */ \ | |
1603 | fail_stack.frame = fail_stack.avail; \ | |
1604 | } while (0) | |
fa9a63c5 | 1605 | |
320a2a73 KH |
1606 | /* Estimate the size of data pushed by a typical failure stack entry. |
1607 | An estimate is all we need, because all we use this for | |
1608 | is to choose a limit for how big to make the failure stack. */ | |
ada30c0e | 1609 | /* BEWARE, the value `20' is hard-coded in emacs.c:main(). */ |
320a2a73 | 1610 | #define TYPICAL_FAILURE_SIZE 20 |
fa9a63c5 | 1611 | |
fa9a63c5 RM |
1612 | /* How many items can still be added to the stack without overflowing it. */ |
1613 | #define REMAINING_AVAIL_SLOTS ((fail_stack).size - (fail_stack).avail) | |
1614 | ||
1615 | ||
1616 | /* Pops what PUSH_FAIL_STACK pushes. | |
1617 | ||
1618 | We restore into the parameters, all of which should be lvalues: | |
1619 | STR -- the saved data position. | |
1620 | PAT -- the saved pattern position. | |
fa9a63c5 | 1621 | REGSTART, REGEND -- arrays of string positions. |
5e69f11e | 1622 | |
fa9a63c5 | 1623 | Also assumes the variables `fail_stack' and (if debugging), `bufp', |
25fe55af | 1624 | `pend', `string1', `size1', `string2', and `size2'. */ |
fa9a63c5 | 1625 | |
505bde11 SM |
1626 | #define POP_FAILURE_POINT(str, pat) \ |
1627 | do { \ | |
fa9a63c5 RM |
1628 | assert (!FAIL_STACK_EMPTY ()); \ |
1629 | \ | |
1630 | /* Remove failure points and point to how many regs pushed. */ \ | |
1631 | DEBUG_PRINT1 ("POP_FAILURE_POINT:\n"); \ | |
1632 | DEBUG_PRINT2 (" Before pop, next avail: %d\n", fail_stack.avail); \ | |
25fe55af | 1633 | DEBUG_PRINT2 (" size: %d\n", fail_stack.size); \ |
fa9a63c5 | 1634 | \ |
505bde11 SM |
1635 | /* Pop the saved registers. */ \ |
1636 | while (fail_stack.frame < fail_stack.avail) \ | |
dc1e502d | 1637 | POP_FAILURE_REG_OR_COUNT (); \ |
fa9a63c5 | 1638 | \ |
01618498 | 1639 | pat = POP_FAILURE_POINTER (); \ |
505bde11 SM |
1640 | DEBUG_PRINT2 (" Popping pattern %p: ", pat); \ |
1641 | DEBUG_PRINT_COMPILED_PATTERN (bufp, pat, pend); \ | |
fa9a63c5 RM |
1642 | \ |
1643 | /* If the saved string location is NULL, it came from an \ | |
1644 | on_failure_keep_string_jump opcode, and we want to throw away the \ | |
1645 | saved NULL, thus retaining our current position in the string. */ \ | |
01618498 | 1646 | str = POP_FAILURE_POINTER (); \ |
505bde11 | 1647 | DEBUG_PRINT2 (" Popping string %p: `", str); \ |
fa9a63c5 RM |
1648 | DEBUG_PRINT_DOUBLE_STRING (str, string1, size1, string2, size2); \ |
1649 | DEBUG_PRINT1 ("'\n"); \ | |
1650 | \ | |
505bde11 SM |
1651 | fail_stack.frame = POP_FAILURE_INT (); \ |
1652 | DEBUG_PRINT2 (" Popping frame index: %d\n", fail_stack.frame); \ | |
fa9a63c5 | 1653 | \ |
505bde11 SM |
1654 | assert (fail_stack.avail >= 0); \ |
1655 | assert (fail_stack.frame <= fail_stack.avail); \ | |
fa9a63c5 | 1656 | \ |
fa9a63c5 | 1657 | DEBUG_STATEMENT (nfailure_points_popped++); \ |
505bde11 | 1658 | } while (0) /* POP_FAILURE_POINT */ |
fa9a63c5 RM |
1659 | |
1660 | ||
1661 | \f | |
fa9a63c5 | 1662 | /* Registers are set to a sentinel when they haven't yet matched. */ |
4bb91c68 | 1663 | #define REG_UNSET(e) ((e) == NULL) |
fa9a63c5 RM |
1664 | \f |
1665 | /* Subroutine declarations and macros for regex_compile. */ | |
1666 | ||
4bb91c68 SM |
1667 | static reg_errcode_t regex_compile _RE_ARGS ((re_char *pattern, size_t size, |
1668 | reg_syntax_t syntax, | |
1669 | struct re_pattern_buffer *bufp)); | |
1670 | static void store_op1 _RE_ARGS ((re_opcode_t op, unsigned char *loc, int arg)); | |
1671 | static void store_op2 _RE_ARGS ((re_opcode_t op, unsigned char *loc, | |
1672 | int arg1, int arg2)); | |
1673 | static void insert_op1 _RE_ARGS ((re_opcode_t op, unsigned char *loc, | |
1674 | int arg, unsigned char *end)); | |
1675 | static void insert_op2 _RE_ARGS ((re_opcode_t op, unsigned char *loc, | |
1676 | int arg1, int arg2, unsigned char *end)); | |
01618498 SM |
1677 | static boolean at_begline_loc_p _RE_ARGS ((re_char *pattern, |
1678 | re_char *p, | |
4bb91c68 | 1679 | reg_syntax_t syntax)); |
01618498 SM |
1680 | static boolean at_endline_loc_p _RE_ARGS ((re_char *p, |
1681 | re_char *pend, | |
4bb91c68 | 1682 | reg_syntax_t syntax)); |
01618498 SM |
1683 | static re_char *skip_one_char _RE_ARGS ((re_char *p)); |
1684 | static int analyse_first _RE_ARGS ((re_char *p, re_char *pend, | |
4bb91c68 | 1685 | char *fastmap, const int multibyte)); |
fa9a63c5 | 1686 | |
fa9a63c5 | 1687 | /* Fetch the next character in the uncompiled pattern, with no |
4bb91c68 | 1688 | translation. */ |
36595814 | 1689 | #define PATFETCH(c) \ |
2d1675e4 SM |
1690 | do { \ |
1691 | int len; \ | |
1692 | if (p == pend) return REG_EEND; \ | |
1693 | c = RE_STRING_CHAR_AND_LENGTH (p, pend - p, len); \ | |
1694 | p += len; \ | |
fa9a63c5 RM |
1695 | } while (0) |
1696 | ||
fa9a63c5 RM |
1697 | |
1698 | /* If `translate' is non-null, return translate[D], else just D. We | |
1699 | cast the subscript to translate because some data is declared as | |
1700 | `char *', to avoid warnings when a string constant is passed. But | |
1701 | when we use a character as a subscript we must make it unsigned. */ | |
6676cb1c | 1702 | #ifndef TRANSLATE |
0b32bf0e | 1703 | # define TRANSLATE(d) \ |
66f0296e | 1704 | (RE_TRANSLATE_P (translate) ? RE_TRANSLATE (translate, (d)) : (d)) |
6676cb1c | 1705 | #endif |
fa9a63c5 RM |
1706 | |
1707 | ||
1708 | /* Macros for outputting the compiled pattern into `buffer'. */ | |
1709 | ||
1710 | /* If the buffer isn't allocated when it comes in, use this. */ | |
1711 | #define INIT_BUF_SIZE 32 | |
1712 | ||
4bb91c68 | 1713 | /* Make sure we have at least N more bytes of space in buffer. */ |
fa9a63c5 | 1714 | #define GET_BUFFER_SPACE(n) \ |
01618498 | 1715 | while ((size_t) (b - bufp->buffer + (n)) > bufp->allocated) \ |
fa9a63c5 RM |
1716 | EXTEND_BUFFER () |
1717 | ||
1718 | /* Make sure we have one more byte of buffer space and then add C to it. */ | |
1719 | #define BUF_PUSH(c) \ | |
1720 | do { \ | |
1721 | GET_BUFFER_SPACE (1); \ | |
1722 | *b++ = (unsigned char) (c); \ | |
1723 | } while (0) | |
1724 | ||
1725 | ||
1726 | /* Ensure we have two more bytes of buffer space and then append C1 and C2. */ | |
1727 | #define BUF_PUSH_2(c1, c2) \ | |
1728 | do { \ | |
1729 | GET_BUFFER_SPACE (2); \ | |
1730 | *b++ = (unsigned char) (c1); \ | |
1731 | *b++ = (unsigned char) (c2); \ | |
1732 | } while (0) | |
1733 | ||
1734 | ||
4bb91c68 | 1735 | /* As with BUF_PUSH_2, except for three bytes. */ |
fa9a63c5 RM |
1736 | #define BUF_PUSH_3(c1, c2, c3) \ |
1737 | do { \ | |
1738 | GET_BUFFER_SPACE (3); \ | |
1739 | *b++ = (unsigned char) (c1); \ | |
1740 | *b++ = (unsigned char) (c2); \ | |
1741 | *b++ = (unsigned char) (c3); \ | |
1742 | } while (0) | |
1743 | ||
1744 | ||
1745 | /* Store a jump with opcode OP at LOC to location TO. We store a | |
4bb91c68 | 1746 | relative address offset by the three bytes the jump itself occupies. */ |
fa9a63c5 RM |
1747 | #define STORE_JUMP(op, loc, to) \ |
1748 | store_op1 (op, loc, (to) - (loc) - 3) | |
1749 | ||
1750 | /* Likewise, for a two-argument jump. */ | |
1751 | #define STORE_JUMP2(op, loc, to, arg) \ | |
1752 | store_op2 (op, loc, (to) - (loc) - 3, arg) | |
1753 | ||
4bb91c68 | 1754 | /* Like `STORE_JUMP', but for inserting. Assume `b' is the buffer end. */ |
fa9a63c5 RM |
1755 | #define INSERT_JUMP(op, loc, to) \ |
1756 | insert_op1 (op, loc, (to) - (loc) - 3, b) | |
1757 | ||
1758 | /* Like `STORE_JUMP2', but for inserting. Assume `b' is the buffer end. */ | |
1759 | #define INSERT_JUMP2(op, loc, to, arg) \ | |
1760 | insert_op2 (op, loc, (to) - (loc) - 3, arg, b) | |
1761 | ||
1762 | ||
1763 | /* This is not an arbitrary limit: the arguments which represent offsets | |
4bb91c68 | 1764 | into the pattern are two bytes long. So if 2^16 bytes turns out to |
fa9a63c5 | 1765 | be too small, many things would have to change. */ |
4bb91c68 SM |
1766 | /* Any other compiler which, like MSC, has allocation limit below 2^16 |
1767 | bytes will have to use approach similar to what was done below for | |
1768 | MSC and drop MAX_BUF_SIZE a bit. Otherwise you may end up | |
1769 | reallocating to 0 bytes. Such thing is not going to work too well. | |
1770 | You have been warned!! */ | |
1771 | #if defined _MSC_VER && !defined WIN32 | |
1772 | /* Microsoft C 16-bit versions limit malloc to approx 65512 bytes. */ | |
1773 | # define MAX_BUF_SIZE 65500L | |
1774 | #else | |
1775 | # define MAX_BUF_SIZE (1L << 16) | |
1776 | #endif | |
fa9a63c5 RM |
1777 | |
1778 | /* Extend the buffer by twice its current size via realloc and | |
1779 | reset the pointers that pointed into the old block to point to the | |
1780 | correct places in the new one. If extending the buffer results in it | |
4bb91c68 SM |
1781 | being larger than MAX_BUF_SIZE, then flag memory exhausted. */ |
1782 | #if __BOUNDED_POINTERS__ | |
1783 | # define SET_HIGH_BOUND(P) (__ptrhigh (P) = __ptrlow (P) + bufp->allocated) | |
1784 | # define MOVE_BUFFER_POINTER(P) \ | |
1785 | (__ptrlow (P) += incr, SET_HIGH_BOUND (P), __ptrvalue (P) += incr) | |
1786 | # define ELSE_EXTEND_BUFFER_HIGH_BOUND \ | |
1787 | else \ | |
1788 | { \ | |
1789 | SET_HIGH_BOUND (b); \ | |
1790 | SET_HIGH_BOUND (begalt); \ | |
1791 | if (fixup_alt_jump) \ | |
1792 | SET_HIGH_BOUND (fixup_alt_jump); \ | |
1793 | if (laststart) \ | |
1794 | SET_HIGH_BOUND (laststart); \ | |
1795 | if (pending_exact) \ | |
1796 | SET_HIGH_BOUND (pending_exact); \ | |
1797 | } | |
1798 | #else | |
1799 | # define MOVE_BUFFER_POINTER(P) (P) += incr | |
1800 | # define ELSE_EXTEND_BUFFER_HIGH_BOUND | |
1801 | #endif | |
fa9a63c5 | 1802 | #define EXTEND_BUFFER() \ |
25fe55af | 1803 | do { \ |
01618498 | 1804 | re_char *old_buffer = bufp->buffer; \ |
25fe55af | 1805 | if (bufp->allocated == MAX_BUF_SIZE) \ |
fa9a63c5 RM |
1806 | return REG_ESIZE; \ |
1807 | bufp->allocated <<= 1; \ | |
1808 | if (bufp->allocated > MAX_BUF_SIZE) \ | |
25fe55af | 1809 | bufp->allocated = MAX_BUF_SIZE; \ |
01618498 | 1810 | RETALLOC (bufp->buffer, bufp->allocated, unsigned char); \ |
fa9a63c5 RM |
1811 | if (bufp->buffer == NULL) \ |
1812 | return REG_ESPACE; \ | |
1813 | /* If the buffer moved, move all the pointers into it. */ \ | |
1814 | if (old_buffer != bufp->buffer) \ | |
1815 | { \ | |
4bb91c68 SM |
1816 | int incr = bufp->buffer - old_buffer; \ |
1817 | MOVE_BUFFER_POINTER (b); \ | |
1818 | MOVE_BUFFER_POINTER (begalt); \ | |
25fe55af | 1819 | if (fixup_alt_jump) \ |
4bb91c68 | 1820 | MOVE_BUFFER_POINTER (fixup_alt_jump); \ |
25fe55af | 1821 | if (laststart) \ |
4bb91c68 | 1822 | MOVE_BUFFER_POINTER (laststart); \ |
25fe55af | 1823 | if (pending_exact) \ |
4bb91c68 | 1824 | MOVE_BUFFER_POINTER (pending_exact); \ |
fa9a63c5 | 1825 | } \ |
4bb91c68 | 1826 | ELSE_EXTEND_BUFFER_HIGH_BOUND \ |
fa9a63c5 RM |
1827 | } while (0) |
1828 | ||
1829 | ||
1830 | /* Since we have one byte reserved for the register number argument to | |
1831 | {start,stop}_memory, the maximum number of groups we can report | |
1832 | things about is what fits in that byte. */ | |
1833 | #define MAX_REGNUM 255 | |
1834 | ||
1835 | /* But patterns can have more than `MAX_REGNUM' registers. We just | |
1836 | ignore the excess. */ | |
1837 | typedef unsigned regnum_t; | |
1838 | ||
1839 | ||
1840 | /* Macros for the compile stack. */ | |
1841 | ||
1842 | /* Since offsets can go either forwards or backwards, this type needs to | |
4bb91c68 SM |
1843 | be able to hold values from -(MAX_BUF_SIZE - 1) to MAX_BUF_SIZE - 1. */ |
1844 | /* int may be not enough when sizeof(int) == 2. */ | |
1845 | typedef long pattern_offset_t; | |
fa9a63c5 RM |
1846 | |
1847 | typedef struct | |
1848 | { | |
1849 | pattern_offset_t begalt_offset; | |
1850 | pattern_offset_t fixup_alt_jump; | |
5e69f11e | 1851 | pattern_offset_t laststart_offset; |
fa9a63c5 RM |
1852 | regnum_t regnum; |
1853 | } compile_stack_elt_t; | |
1854 | ||
1855 | ||
1856 | typedef struct | |
1857 | { | |
1858 | compile_stack_elt_t *stack; | |
1859 | unsigned size; | |
1860 | unsigned avail; /* Offset of next open position. */ | |
1861 | } compile_stack_type; | |
1862 | ||
1863 | ||
1864 | #define INIT_COMPILE_STACK_SIZE 32 | |
1865 | ||
1866 | #define COMPILE_STACK_EMPTY (compile_stack.avail == 0) | |
1867 | #define COMPILE_STACK_FULL (compile_stack.avail == compile_stack.size) | |
1868 | ||
4bb91c68 | 1869 | /* The next available element. */ |
fa9a63c5 RM |
1870 | #define COMPILE_STACK_TOP (compile_stack.stack[compile_stack.avail]) |
1871 | ||
77d11aec RS |
1872 | /* Explicit quit checking is only used on NTemacs. */ |
1873 | #if defined WINDOWSNT && defined emacs && defined QUIT | |
1874 | extern int immediate_quit; | |
1875 | # define IMMEDIATE_QUIT_CHECK \ | |
1876 | do { \ | |
1877 | if (immediate_quit) QUIT; \ | |
1878 | } while (0) | |
1879 | #else | |
1880 | # define IMMEDIATE_QUIT_CHECK ((void)0) | |
1881 | #endif | |
1882 | \f | |
b18215fc RS |
1883 | /* Structure to manage work area for range table. */ |
1884 | struct range_table_work_area | |
1885 | { | |
1886 | int *table; /* actual work area. */ | |
1887 | int allocated; /* allocated size for work area in bytes. */ | |
25fe55af | 1888 | int used; /* actually used size in words. */ |
96cc36cc | 1889 | int bits; /* flag to record character classes */ |
b18215fc RS |
1890 | }; |
1891 | ||
77d11aec RS |
1892 | /* Make sure that WORK_AREA can hold more N multibyte characters. |
1893 | This is used only in set_image_of_range and set_image_of_range_1. | |
1894 | It expects WORK_AREA to be a pointer. | |
1895 | If it can't get the space, it returns from the surrounding function. */ | |
1896 | ||
1897 | #define EXTEND_RANGE_TABLE(work_area, n) \ | |
1898 | do { \ | |
1899 | if (((work_area)->used + (n)) * sizeof (int) > (work_area)->allocated) \ | |
1900 | { \ | |
1901 | extend_range_table_work_area (work_area); \ | |
1902 | if ((work_area)->table == 0) \ | |
1903 | return (REG_ESPACE); \ | |
1904 | } \ | |
b18215fc RS |
1905 | } while (0) |
1906 | ||
96cc36cc RS |
1907 | #define SET_RANGE_TABLE_WORK_AREA_BIT(work_area, bit) \ |
1908 | (work_area).bits |= (bit) | |
1909 | ||
14473664 SM |
1910 | /* Bits used to implement the multibyte-part of the various character classes |
1911 | such as [:alnum:] in a charset's range table. */ | |
1912 | #define BIT_WORD 0x1 | |
1913 | #define BIT_LOWER 0x2 | |
1914 | #define BIT_PUNCT 0x4 | |
1915 | #define BIT_SPACE 0x8 | |
1916 | #define BIT_UPPER 0x10 | |
1917 | #define BIT_MULTIBYTE 0x20 | |
96cc36cc | 1918 | |
36595814 SM |
1919 | /* Set a range START..END to WORK_AREA. |
1920 | The range is passed through TRANSLATE, so START and END | |
1921 | should be untranslated. */ | |
77d11aec RS |
1922 | #define SET_RANGE_TABLE_WORK_AREA(work_area, start, end) \ |
1923 | do { \ | |
1924 | int tem; \ | |
1925 | tem = set_image_of_range (&work_area, start, end, translate); \ | |
1926 | if (tem > 0) \ | |
1927 | FREE_STACK_RETURN (tem); \ | |
b18215fc RS |
1928 | } while (0) |
1929 | ||
25fe55af | 1930 | /* Free allocated memory for WORK_AREA. */ |
b18215fc RS |
1931 | #define FREE_RANGE_TABLE_WORK_AREA(work_area) \ |
1932 | do { \ | |
1933 | if ((work_area).table) \ | |
1934 | free ((work_area).table); \ | |
1935 | } while (0) | |
1936 | ||
96cc36cc | 1937 | #define CLEAR_RANGE_TABLE_WORK_USED(work_area) ((work_area).used = 0, (work_area).bits = 0) |
b18215fc | 1938 | #define RANGE_TABLE_WORK_USED(work_area) ((work_area).used) |
96cc36cc | 1939 | #define RANGE_TABLE_WORK_BITS(work_area) ((work_area).bits) |
b18215fc | 1940 | #define RANGE_TABLE_WORK_ELT(work_area, i) ((work_area).table[i]) |
77d11aec | 1941 | \f |
b18215fc | 1942 | |
fa9a63c5 | 1943 | /* Set the bit for character C in a list. */ |
01618498 | 1944 | #define SET_LIST_BIT(c) (b[((c)) / BYTEWIDTH] |= 1 << ((c) % BYTEWIDTH)) |
fa9a63c5 RM |
1945 | |
1946 | ||
1947 | /* Get the next unsigned number in the uncompiled pattern. */ | |
25fe55af | 1948 | #define GET_UNSIGNED_NUMBER(num) \ |
99633e97 | 1949 | do { if (p != pend) \ |
fa9a63c5 | 1950 | { \ |
25fe55af | 1951 | PATFETCH (c); \ |
060cc8ff RS |
1952 | if (c == ' ') \ |
1953 | FREE_STACK_RETURN (REG_BADBR); \ | |
4bb91c68 | 1954 | while ('0' <= c && c <= '9') \ |
25fe55af | 1955 | { \ |
f6df485f | 1956 | int prev; \ |
25fe55af | 1957 | if (num < 0) \ |
f6df485f RS |
1958 | num = 0; \ |
1959 | prev = num; \ | |
25fe55af | 1960 | num = num * 10 + c - '0'; \ |
f6df485f RS |
1961 | if (num / 10 != prev) \ |
1962 | FREE_STACK_RETURN (REG_BADBR); \ | |
25fe55af | 1963 | if (p == pend) \ |
f6df485f | 1964 | break; \ |
25fe55af RS |
1965 | PATFETCH (c); \ |
1966 | } \ | |
060cc8ff RS |
1967 | if (c == ' ') \ |
1968 | FREE_STACK_RETURN (REG_BADBR); \ | |
25fe55af | 1969 | } \ |
99633e97 | 1970 | } while (0) |
77d11aec | 1971 | \f |
a0ad02f7 | 1972 | #if WIDE_CHAR_SUPPORT |
14473664 SM |
1973 | /* The GNU C library provides support for user-defined character classes |
1974 | and the functions from ISO C amendement 1. */ | |
1975 | # ifdef CHARCLASS_NAME_MAX | |
1976 | # define CHAR_CLASS_MAX_LENGTH CHARCLASS_NAME_MAX | |
1977 | # else | |
1978 | /* This shouldn't happen but some implementation might still have this | |
1979 | problem. Use a reasonable default value. */ | |
1980 | # define CHAR_CLASS_MAX_LENGTH 256 | |
1981 | # endif | |
1982 | typedef wctype_t re_wctype_t; | |
01618498 | 1983 | typedef wchar_t re_wchar_t; |
14473664 SM |
1984 | # define re_wctype wctype |
1985 | # define re_iswctype iswctype | |
1986 | # define re_wctype_to_bit(cc) 0 | |
1987 | #else | |
1988 | # define CHAR_CLASS_MAX_LENGTH 9 /* Namely, `multibyte'. */ | |
1989 | # define btowc(c) c | |
1990 | ||
01618498 | 1991 | /* Character classes. */ |
14473664 SM |
1992 | typedef enum { RECC_ERROR = 0, |
1993 | RECC_ALNUM, RECC_ALPHA, RECC_WORD, | |
1994 | RECC_GRAPH, RECC_PRINT, | |
1995 | RECC_LOWER, RECC_UPPER, | |
1996 | RECC_PUNCT, RECC_CNTRL, | |
1997 | RECC_DIGIT, RECC_XDIGIT, | |
1998 | RECC_BLANK, RECC_SPACE, | |
1999 | RECC_MULTIBYTE, RECC_NONASCII, | |
2000 | RECC_ASCII, RECC_UNIBYTE | |
2001 | } re_wctype_t; | |
2002 | ||
01618498 SM |
2003 | typedef int re_wchar_t; |
2004 | ||
14473664 SM |
2005 | /* Map a string to the char class it names (if any). */ |
2006 | static re_wctype_t | |
ada30c0e SM |
2007 | re_wctype (str) |
2008 | re_char *str; | |
14473664 | 2009 | { |
ada30c0e | 2010 | const char *string = str; |
14473664 SM |
2011 | if (STREQ (string, "alnum")) return RECC_ALNUM; |
2012 | else if (STREQ (string, "alpha")) return RECC_ALPHA; | |
2013 | else if (STREQ (string, "word")) return RECC_WORD; | |
2014 | else if (STREQ (string, "ascii")) return RECC_ASCII; | |
2015 | else if (STREQ (string, "nonascii")) return RECC_NONASCII; | |
2016 | else if (STREQ (string, "graph")) return RECC_GRAPH; | |
2017 | else if (STREQ (string, "lower")) return RECC_LOWER; | |
2018 | else if (STREQ (string, "print")) return RECC_PRINT; | |
2019 | else if (STREQ (string, "punct")) return RECC_PUNCT; | |
2020 | else if (STREQ (string, "space")) return RECC_SPACE; | |
2021 | else if (STREQ (string, "upper")) return RECC_UPPER; | |
2022 | else if (STREQ (string, "unibyte")) return RECC_UNIBYTE; | |
2023 | else if (STREQ (string, "multibyte")) return RECC_MULTIBYTE; | |
2024 | else if (STREQ (string, "digit")) return RECC_DIGIT; | |
2025 | else if (STREQ (string, "xdigit")) return RECC_XDIGIT; | |
2026 | else if (STREQ (string, "cntrl")) return RECC_CNTRL; | |
2027 | else if (STREQ (string, "blank")) return RECC_BLANK; | |
2028 | else return 0; | |
2029 | } | |
2030 | ||
2031 | /* True iff CH is in the char class CC. */ | |
2032 | static boolean | |
2033 | re_iswctype (ch, cc) | |
2034 | int ch; | |
2035 | re_wctype_t cc; | |
2036 | { | |
2037 | switch (cc) | |
2038 | { | |
0cdd06f8 SM |
2039 | case RECC_ALNUM: return ISALNUM (ch); |
2040 | case RECC_ALPHA: return ISALPHA (ch); | |
2041 | case RECC_BLANK: return ISBLANK (ch); | |
2042 | case RECC_CNTRL: return ISCNTRL (ch); | |
2043 | case RECC_DIGIT: return ISDIGIT (ch); | |
2044 | case RECC_GRAPH: return ISGRAPH (ch); | |
2045 | case RECC_LOWER: return ISLOWER (ch); | |
2046 | case RECC_PRINT: return ISPRINT (ch); | |
2047 | case RECC_PUNCT: return ISPUNCT (ch); | |
2048 | case RECC_SPACE: return ISSPACE (ch); | |
2049 | case RECC_UPPER: return ISUPPER (ch); | |
2050 | case RECC_XDIGIT: return ISXDIGIT (ch); | |
2051 | case RECC_ASCII: return IS_REAL_ASCII (ch); | |
2052 | case RECC_NONASCII: return !IS_REAL_ASCII (ch); | |
2053 | case RECC_UNIBYTE: return ISUNIBYTE (ch); | |
2054 | case RECC_MULTIBYTE: return !ISUNIBYTE (ch); | |
2055 | case RECC_WORD: return ISWORD (ch); | |
2056 | case RECC_ERROR: return false; | |
2057 | default: | |
2058 | abort(); | |
14473664 SM |
2059 | } |
2060 | } | |
fa9a63c5 | 2061 | |
14473664 SM |
2062 | /* Return a bit-pattern to use in the range-table bits to match multibyte |
2063 | chars of class CC. */ | |
2064 | static int | |
2065 | re_wctype_to_bit (cc) | |
2066 | re_wctype_t cc; | |
2067 | { | |
2068 | switch (cc) | |
2069 | { | |
2070 | case RECC_NONASCII: case RECC_PRINT: case RECC_GRAPH: | |
0cdd06f8 SM |
2071 | case RECC_MULTIBYTE: return BIT_MULTIBYTE; |
2072 | case RECC_ALPHA: case RECC_ALNUM: case RECC_WORD: return BIT_WORD; | |
2073 | case RECC_LOWER: return BIT_LOWER; | |
2074 | case RECC_UPPER: return BIT_UPPER; | |
2075 | case RECC_PUNCT: return BIT_PUNCT; | |
2076 | case RECC_SPACE: return BIT_SPACE; | |
14473664 | 2077 | case RECC_ASCII: case RECC_DIGIT: case RECC_XDIGIT: case RECC_CNTRL: |
0cdd06f8 SM |
2078 | case RECC_BLANK: case RECC_UNIBYTE: case RECC_ERROR: return 0; |
2079 | default: | |
2080 | abort(); | |
14473664 SM |
2081 | } |
2082 | } | |
2083 | #endif | |
77d11aec RS |
2084 | \f |
2085 | /* Filling in the work area of a range. */ | |
2086 | ||
2087 | /* Actually extend the space in WORK_AREA. */ | |
2088 | ||
2089 | static void | |
2090 | extend_range_table_work_area (work_area) | |
2091 | struct range_table_work_area *work_area; | |
2092 | { | |
2093 | work_area->allocated += 16 * sizeof (int); | |
2094 | if (work_area->table) | |
2095 | work_area->table | |
2096 | = (int *) realloc (work_area->table, work_area->allocated); | |
2097 | else | |
2098 | work_area->table | |
2099 | = (int *) malloc (work_area->allocated); | |
2100 | } | |
2101 | ||
2102 | #ifdef emacs | |
2103 | ||
2104 | /* Carefully find the ranges of codes that are equivalent | |
2105 | under case conversion to the range start..end when passed through | |
2106 | TRANSLATE. Handle the case where non-letters can come in between | |
2107 | two upper-case letters (which happens in Latin-1). | |
2108 | Also handle the case of groups of more than 2 case-equivalent chars. | |
2109 | ||
2110 | The basic method is to look at consecutive characters and see | |
2111 | if they can form a run that can be handled as one. | |
2112 | ||
2113 | Returns -1 if successful, REG_ESPACE if ran out of space. */ | |
2114 | ||
2115 | static int | |
2116 | set_image_of_range_1 (work_area, start, end, translate) | |
2117 | RE_TRANSLATE_TYPE translate; | |
2118 | struct range_table_work_area *work_area; | |
2119 | re_wchar_t start, end; | |
2120 | { | |
2121 | /* `one_case' indicates a character, or a run of characters, | |
2122 | each of which is an isolate (no case-equivalents). | |
2123 | This includes all ASCII non-letters. | |
2124 | ||
2125 | `two_case' indicates a character, or a run of characters, | |
2126 | each of which has two case-equivalent forms. | |
2127 | This includes all ASCII letters. | |
2128 | ||
2129 | `strange' indicates a character that has more than one | |
2130 | case-equivalent. */ | |
2131 | ||
2132 | enum case_type {one_case, two_case, strange}; | |
2133 | ||
2134 | /* Describe the run that is in progress, | |
2135 | which the next character can try to extend. | |
2136 | If run_type is strange, that means there really is no run. | |
2137 | If run_type is one_case, then run_start...run_end is the run. | |
2138 | If run_type is two_case, then the run is run_start...run_end, | |
2139 | and the case-equivalents end at run_eqv_end. */ | |
2140 | ||
2141 | enum case_type run_type = strange; | |
2142 | int run_start, run_end, run_eqv_end; | |
2143 | ||
2144 | Lisp_Object eqv_table; | |
2145 | ||
2146 | if (!RE_TRANSLATE_P (translate)) | |
2147 | { | |
2148 | work_area->table[work_area->used++] = (start); | |
2149 | work_area->table[work_area->used++] = (end); | |
2150 | return; | |
2151 | } | |
2152 | ||
2153 | eqv_table = XCHAR_TABLE (translate)->extras[2]; | |
99633e97 | 2154 | |
77d11aec RS |
2155 | for (; start <= end; start++) |
2156 | { | |
2157 | enum case_type this_type; | |
2158 | int eqv = RE_TRANSLATE (eqv_table, start); | |
2159 | int minchar, maxchar; | |
2160 | ||
2161 | /* Classify this character */ | |
2162 | if (eqv == start) | |
2163 | this_type = one_case; | |
2164 | else if (RE_TRANSLATE (eqv_table, eqv) == start) | |
2165 | this_type = two_case; | |
2166 | else | |
2167 | this_type = strange; | |
2168 | ||
2169 | if (start < eqv) | |
2170 | minchar = start, maxchar = eqv; | |
2171 | else | |
2172 | minchar = eqv, maxchar = start; | |
2173 | ||
2174 | /* Can this character extend the run in progress? */ | |
2175 | if (this_type == strange || this_type != run_type | |
2176 | || !(minchar == run_end + 1 | |
2177 | && (run_type == two_case | |
2178 | ? maxchar == run_eqv_end + 1 : 1))) | |
2179 | { | |
2180 | /* No, end the run. | |
2181 | Record each of its equivalent ranges. */ | |
2182 | if (run_type == one_case) | |
2183 | { | |
2184 | EXTEND_RANGE_TABLE (work_area, 2); | |
2185 | work_area->table[work_area->used++] = run_start; | |
2186 | work_area->table[work_area->used++] = run_end; | |
2187 | } | |
2188 | else if (run_type == two_case) | |
2189 | { | |
2190 | EXTEND_RANGE_TABLE (work_area, 4); | |
2191 | work_area->table[work_area->used++] = run_start; | |
2192 | work_area->table[work_area->used++] = run_end; | |
2193 | work_area->table[work_area->used++] | |
2194 | = RE_TRANSLATE (eqv_table, run_start); | |
2195 | work_area->table[work_area->used++] | |
2196 | = RE_TRANSLATE (eqv_table, run_end); | |
2197 | } | |
2198 | run_type = strange; | |
2199 | } | |
2200 | ||
2201 | if (this_type == strange) | |
2202 | { | |
2203 | /* For a strange character, add each of its equivalents, one | |
2204 | by one. Don't start a range. */ | |
2205 | do | |
2206 | { | |
2207 | EXTEND_RANGE_TABLE (work_area, 2); | |
2208 | work_area->table[work_area->used++] = eqv; | |
2209 | work_area->table[work_area->used++] = eqv; | |
2210 | eqv = RE_TRANSLATE (eqv_table, eqv); | |
2211 | } | |
2212 | while (eqv != start); | |
2213 | } | |
2214 | ||
2215 | /* Add this char to the run, or start a new run. */ | |
2216 | else if (run_type == strange) | |
2217 | { | |
2218 | /* Initialize a new range. */ | |
2219 | run_type = this_type; | |
2220 | run_start = start; | |
2221 | run_end = start; | |
2222 | run_eqv_end = RE_TRANSLATE (eqv_table, run_end); | |
2223 | } | |
2224 | else | |
2225 | { | |
2226 | /* Extend a running range. */ | |
2227 | run_end = minchar; | |
2228 | run_eqv_end = RE_TRANSLATE (eqv_table, run_end); | |
2229 | } | |
2230 | } | |
2231 | ||
2232 | /* If a run is still in progress at the end, finish it now | |
2233 | by recording its equivalent ranges. */ | |
2234 | if (run_type == one_case) | |
2235 | { | |
2236 | EXTEND_RANGE_TABLE (work_area, 2); | |
2237 | work_area->table[work_area->used++] = run_start; | |
2238 | work_area->table[work_area->used++] = run_end; | |
2239 | } | |
2240 | else if (run_type == two_case) | |
2241 | { | |
2242 | EXTEND_RANGE_TABLE (work_area, 4); | |
2243 | work_area->table[work_area->used++] = run_start; | |
2244 | work_area->table[work_area->used++] = run_end; | |
2245 | work_area->table[work_area->used++] | |
2246 | = RE_TRANSLATE (eqv_table, run_start); | |
2247 | work_area->table[work_area->used++] | |
2248 | = RE_TRANSLATE (eqv_table, run_end); | |
2249 | } | |
2250 | ||
2251 | return -1; | |
2252 | } | |
36595814 | 2253 | |
77d11aec | 2254 | #endif /* emacs */ |
36595814 SM |
2255 | |
2256 | /* We need to find the image of the range start..end when passed through | |
2257 | TRANSLATE. This is not necessarily TRANSLATE(start)..TRANSLATE(end) | |
2258 | and is not even necessarily contiguous. | |
2259 | We approximate it with the smallest contiguous range that contains | |
77d11aec RS |
2260 | all the chars we need. However, that is not good enough for Latin-1, |
2261 | so we do a better job in that case. | |
2262 | ||
2263 | Returns -1 if successful, REG_ESPACE if ran out of space. */ | |
2264 | ||
2265 | static int | |
36595814 SM |
2266 | set_image_of_range (work_area, start, end, translate) |
2267 | RE_TRANSLATE_TYPE translate; | |
2268 | struct range_table_work_area *work_area; | |
2269 | re_wchar_t start, end; | |
2270 | { | |
77d11aec RS |
2271 | re_wchar_t cmin, cmax; |
2272 | ||
2273 | #ifdef emacs | |
2274 | /* For Latin-1 ranges, use set_image_of_range_1 | |
2275 | to get proper handling of ranges that include letters and nonletters. | |
2276 | For ASCII, this is not necessary. | |
2277 | For other character sets, we don't bother to get this right. */ | |
2278 | if (start < 04400 && end > 0200) | |
2279 | { | |
2280 | int tem; | |
2281 | tem = set_image_of_range_1 (work_area, start, end, translate); | |
2282 | if (tem > 0) | |
2283 | return tem; | |
2284 | ||
2285 | start = 04400; | |
2286 | if (end < 04400) | |
2287 | return -1; | |
2288 | } | |
2289 | #endif | |
2290 | ||
2291 | cmin = TRANSLATE (start), cmax = TRANSLATE (end); | |
2292 | ||
36595814 SM |
2293 | if (RE_TRANSLATE_P (translate)) |
2294 | for (; start <= end; start++) | |
2295 | { | |
2296 | re_wchar_t c = TRANSLATE (start); | |
2297 | cmin = MIN (cmin, c); | |
2298 | cmax = MAX (cmax, c); | |
2299 | } | |
77d11aec RS |
2300 | |
2301 | EXTEND_RANGE_TABLE (work_area, 2); | |
36595814 SM |
2302 | work_area->table[work_area->used++] = (cmin); |
2303 | work_area->table[work_area->used++] = (cmax); | |
36595814 | 2304 | |
77d11aec RS |
2305 | return -1; |
2306 | } | |
fa9a63c5 RM |
2307 | \f |
2308 | #ifndef MATCH_MAY_ALLOCATE | |
2309 | ||
2310 | /* If we cannot allocate large objects within re_match_2_internal, | |
2311 | we make the fail stack and register vectors global. | |
2312 | The fail stack, we grow to the maximum size when a regexp | |
2313 | is compiled. | |
2314 | The register vectors, we adjust in size each time we | |
2315 | compile a regexp, according to the number of registers it needs. */ | |
2316 | ||
2317 | static fail_stack_type fail_stack; | |
2318 | ||
2319 | /* Size with which the following vectors are currently allocated. | |
2320 | That is so we can make them bigger as needed, | |
4bb91c68 | 2321 | but never make them smaller. */ |
fa9a63c5 RM |
2322 | static int regs_allocated_size; |
2323 | ||
66f0296e SM |
2324 | static re_char ** regstart, ** regend; |
2325 | static re_char **best_regstart, **best_regend; | |
fa9a63c5 RM |
2326 | |
2327 | /* Make the register vectors big enough for NUM_REGS registers, | |
4bb91c68 | 2328 | but don't make them smaller. */ |
fa9a63c5 RM |
2329 | |
2330 | static | |
2331 | regex_grow_registers (num_regs) | |
2332 | int num_regs; | |
2333 | { | |
2334 | if (num_regs > regs_allocated_size) | |
2335 | { | |
66f0296e SM |
2336 | RETALLOC_IF (regstart, num_regs, re_char *); |
2337 | RETALLOC_IF (regend, num_regs, re_char *); | |
2338 | RETALLOC_IF (best_regstart, num_regs, re_char *); | |
2339 | RETALLOC_IF (best_regend, num_regs, re_char *); | |
fa9a63c5 RM |
2340 | |
2341 | regs_allocated_size = num_regs; | |
2342 | } | |
2343 | } | |
2344 | ||
2345 | #endif /* not MATCH_MAY_ALLOCATE */ | |
2346 | \f | |
99633e97 SM |
2347 | static boolean group_in_compile_stack _RE_ARGS ((compile_stack_type |
2348 | compile_stack, | |
2349 | regnum_t regnum)); | |
2350 | ||
fa9a63c5 RM |
2351 | /* `regex_compile' compiles PATTERN (of length SIZE) according to SYNTAX. |
2352 | Returns one of error codes defined in `regex.h', or zero for success. | |
2353 | ||
2354 | Assumes the `allocated' (and perhaps `buffer') and `translate' | |
2355 | fields are set in BUFP on entry. | |
2356 | ||
2357 | If it succeeds, results are put in BUFP (if it returns an error, the | |
2358 | contents of BUFP are undefined): | |
2359 | `buffer' is the compiled pattern; | |
2360 | `syntax' is set to SYNTAX; | |
2361 | `used' is set to the length of the compiled pattern; | |
2362 | `fastmap_accurate' is zero; | |
2363 | `re_nsub' is the number of subexpressions in PATTERN; | |
2364 | `not_bol' and `not_eol' are zero; | |
5e69f11e | 2365 | |
c0f9ea08 | 2366 | The `fastmap' field is neither examined nor set. */ |
fa9a63c5 | 2367 | |
505bde11 SM |
2368 | /* Insert the `jump' from the end of last alternative to "here". |
2369 | The space for the jump has already been allocated. */ | |
2370 | #define FIXUP_ALT_JUMP() \ | |
2371 | do { \ | |
2372 | if (fixup_alt_jump) \ | |
2373 | STORE_JUMP (jump, fixup_alt_jump, b); \ | |
2374 | } while (0) | |
2375 | ||
2376 | ||
fa9a63c5 RM |
2377 | /* Return, freeing storage we allocated. */ |
2378 | #define FREE_STACK_RETURN(value) \ | |
b18215fc RS |
2379 | do { \ |
2380 | FREE_RANGE_TABLE_WORK_AREA (range_table_work); \ | |
2381 | free (compile_stack.stack); \ | |
2382 | return value; \ | |
2383 | } while (0) | |
fa9a63c5 RM |
2384 | |
2385 | static reg_errcode_t | |
2386 | regex_compile (pattern, size, syntax, bufp) | |
66f0296e | 2387 | re_char *pattern; |
4bb91c68 | 2388 | size_t size; |
fa9a63c5 RM |
2389 | reg_syntax_t syntax; |
2390 | struct re_pattern_buffer *bufp; | |
2391 | { | |
01618498 SM |
2392 | /* We fetch characters from PATTERN here. */ |
2393 | register re_wchar_t c, c1; | |
5e69f11e | 2394 | |
fa9a63c5 | 2395 | /* A random temporary spot in PATTERN. */ |
66f0296e | 2396 | re_char *p1; |
fa9a63c5 RM |
2397 | |
2398 | /* Points to the end of the buffer, where we should append. */ | |
2399 | register unsigned char *b; | |
5e69f11e | 2400 | |
fa9a63c5 RM |
2401 | /* Keeps track of unclosed groups. */ |
2402 | compile_stack_type compile_stack; | |
2403 | ||
2404 | /* Points to the current (ending) position in the pattern. */ | |
22336245 RS |
2405 | #ifdef AIX |
2406 | /* `const' makes AIX compiler fail. */ | |
66f0296e | 2407 | unsigned char *p = pattern; |
22336245 | 2408 | #else |
66f0296e | 2409 | re_char *p = pattern; |
22336245 | 2410 | #endif |
66f0296e | 2411 | re_char *pend = pattern + size; |
5e69f11e | 2412 | |
fa9a63c5 | 2413 | /* How to translate the characters in the pattern. */ |
6676cb1c | 2414 | RE_TRANSLATE_TYPE translate = bufp->translate; |
fa9a63c5 RM |
2415 | |
2416 | /* Address of the count-byte of the most recently inserted `exactn' | |
2417 | command. This makes it possible to tell if a new exact-match | |
2418 | character can be added to that command or if the character requires | |
2419 | a new `exactn' command. */ | |
2420 | unsigned char *pending_exact = 0; | |
2421 | ||
2422 | /* Address of start of the most recently finished expression. | |
2423 | This tells, e.g., postfix * where to find the start of its | |
2424 | operand. Reset at the beginning of groups and alternatives. */ | |
2425 | unsigned char *laststart = 0; | |
2426 | ||
2427 | /* Address of beginning of regexp, or inside of last group. */ | |
2428 | unsigned char *begalt; | |
2429 | ||
2430 | /* Place in the uncompiled pattern (i.e., the {) to | |
2431 | which to go back if the interval is invalid. */ | |
66f0296e | 2432 | re_char *beg_interval; |
5e69f11e | 2433 | |
fa9a63c5 | 2434 | /* Address of the place where a forward jump should go to the end of |
25fe55af | 2435 | the containing expression. Each alternative of an `or' -- except the |
fa9a63c5 RM |
2436 | last -- ends with a forward jump of this sort. */ |
2437 | unsigned char *fixup_alt_jump = 0; | |
2438 | ||
2439 | /* Counts open-groups as they are encountered. Remembered for the | |
2440 | matching close-group on the compile stack, so the same register | |
2441 | number is put in the stop_memory as the start_memory. */ | |
2442 | regnum_t regnum = 0; | |
2443 | ||
b18215fc RS |
2444 | /* Work area for range table of charset. */ |
2445 | struct range_table_work_area range_table_work; | |
2446 | ||
2d1675e4 SM |
2447 | /* If the object matched can contain multibyte characters. */ |
2448 | const boolean multibyte = RE_MULTIBYTE_P (bufp); | |
2449 | ||
fa9a63c5 | 2450 | #ifdef DEBUG |
99633e97 | 2451 | debug++; |
fa9a63c5 | 2452 | DEBUG_PRINT1 ("\nCompiling pattern: "); |
99633e97 | 2453 | if (debug > 0) |
fa9a63c5 RM |
2454 | { |
2455 | unsigned debug_count; | |
5e69f11e | 2456 | |
fa9a63c5 | 2457 | for (debug_count = 0; debug_count < size; debug_count++) |
25fe55af | 2458 | putchar (pattern[debug_count]); |
fa9a63c5 RM |
2459 | putchar ('\n'); |
2460 | } | |
2461 | #endif /* DEBUG */ | |
2462 | ||
2463 | /* Initialize the compile stack. */ | |
2464 | compile_stack.stack = TALLOC (INIT_COMPILE_STACK_SIZE, compile_stack_elt_t); | |
2465 | if (compile_stack.stack == NULL) | |
2466 | return REG_ESPACE; | |
2467 | ||
2468 | compile_stack.size = INIT_COMPILE_STACK_SIZE; | |
2469 | compile_stack.avail = 0; | |
2470 | ||
b18215fc RS |
2471 | range_table_work.table = 0; |
2472 | range_table_work.allocated = 0; | |
2473 | ||
fa9a63c5 RM |
2474 | /* Initialize the pattern buffer. */ |
2475 | bufp->syntax = syntax; | |
2476 | bufp->fastmap_accurate = 0; | |
2477 | bufp->not_bol = bufp->not_eol = 0; | |
2478 | ||
2479 | /* Set `used' to zero, so that if we return an error, the pattern | |
2480 | printer (for debugging) will think there's no pattern. We reset it | |
2481 | at the end. */ | |
2482 | bufp->used = 0; | |
5e69f11e | 2483 | |
fa9a63c5 | 2484 | /* Always count groups, whether or not bufp->no_sub is set. */ |
5e69f11e | 2485 | bufp->re_nsub = 0; |
fa9a63c5 | 2486 | |
0b32bf0e | 2487 | #if !defined emacs && !defined SYNTAX_TABLE |
fa9a63c5 RM |
2488 | /* Initialize the syntax table. */ |
2489 | init_syntax_once (); | |
2490 | #endif | |
2491 | ||
2492 | if (bufp->allocated == 0) | |
2493 | { | |
2494 | if (bufp->buffer) | |
2495 | { /* If zero allocated, but buffer is non-null, try to realloc | |
25fe55af RS |
2496 | enough space. This loses if buffer's address is bogus, but |
2497 | that is the user's responsibility. */ | |
2498 | RETALLOC (bufp->buffer, INIT_BUF_SIZE, unsigned char); | |
2499 | } | |
fa9a63c5 | 2500 | else |
25fe55af RS |
2501 | { /* Caller did not allocate a buffer. Do it for them. */ |
2502 | bufp->buffer = TALLOC (INIT_BUF_SIZE, unsigned char); | |
2503 | } | |
fa9a63c5 RM |
2504 | if (!bufp->buffer) FREE_STACK_RETURN (REG_ESPACE); |
2505 | ||
2506 | bufp->allocated = INIT_BUF_SIZE; | |
2507 | } | |
2508 | ||
2509 | begalt = b = bufp->buffer; | |
2510 | ||
2511 | /* Loop through the uncompiled pattern until we're at the end. */ | |
2512 | while (p != pend) | |
2513 | { | |
2514 | PATFETCH (c); | |
2515 | ||
2516 | switch (c) | |
25fe55af RS |
2517 | { |
2518 | case '^': | |
2519 | { | |
2520 | if ( /* If at start of pattern, it's an operator. */ | |
2521 | p == pattern + 1 | |
2522 | /* If context independent, it's an operator. */ | |
2523 | || syntax & RE_CONTEXT_INDEP_ANCHORS | |
2524 | /* Otherwise, depends on what's come before. */ | |
2525 | || at_begline_loc_p (pattern, p, syntax)) | |
c0f9ea08 | 2526 | BUF_PUSH ((syntax & RE_NO_NEWLINE_ANCHOR) ? begbuf : begline); |
25fe55af RS |
2527 | else |
2528 | goto normal_char; | |
2529 | } | |
2530 | break; | |
2531 | ||
2532 | ||
2533 | case '$': | |
2534 | { | |
2535 | if ( /* If at end of pattern, it's an operator. */ | |
2536 | p == pend | |
2537 | /* If context independent, it's an operator. */ | |
2538 | || syntax & RE_CONTEXT_INDEP_ANCHORS | |
2539 | /* Otherwise, depends on what's next. */ | |
2540 | || at_endline_loc_p (p, pend, syntax)) | |
c0f9ea08 | 2541 | BUF_PUSH ((syntax & RE_NO_NEWLINE_ANCHOR) ? endbuf : endline); |
25fe55af RS |
2542 | else |
2543 | goto normal_char; | |
2544 | } | |
2545 | break; | |
fa9a63c5 RM |
2546 | |
2547 | ||
2548 | case '+': | |
25fe55af RS |
2549 | case '?': |
2550 | if ((syntax & RE_BK_PLUS_QM) | |
2551 | || (syntax & RE_LIMITED_OPS)) | |
2552 | goto normal_char; | |
2553 | handle_plus: | |
2554 | case '*': | |
2555 | /* If there is no previous pattern... */ | |
2556 | if (!laststart) | |
2557 | { | |
2558 | if (syntax & RE_CONTEXT_INVALID_OPS) | |
2559 | FREE_STACK_RETURN (REG_BADRPT); | |
2560 | else if (!(syntax & RE_CONTEXT_INDEP_OPS)) | |
2561 | goto normal_char; | |
2562 | } | |
2563 | ||
2564 | { | |
25fe55af | 2565 | /* 1 means zero (many) matches is allowed. */ |
66f0296e SM |
2566 | boolean zero_times_ok = 0, many_times_ok = 0; |
2567 | boolean greedy = 1; | |
25fe55af RS |
2568 | |
2569 | /* If there is a sequence of repetition chars, collapse it | |
2570 | down to just one (the right one). We can't combine | |
2571 | interval operators with these because of, e.g., `a{2}*', | |
2572 | which should only match an even number of `a's. */ | |
2573 | ||
2574 | for (;;) | |
2575 | { | |
0b32bf0e | 2576 | if ((syntax & RE_FRUGAL) |
1c8c6d39 DL |
2577 | && c == '?' && (zero_times_ok || many_times_ok)) |
2578 | greedy = 0; | |
2579 | else | |
2580 | { | |
2581 | zero_times_ok |= c != '+'; | |
2582 | many_times_ok |= c != '?'; | |
2583 | } | |
25fe55af RS |
2584 | |
2585 | if (p == pend) | |
2586 | break; | |
ed0767d8 SM |
2587 | else if (*p == '*' |
2588 | || (!(syntax & RE_BK_PLUS_QM) | |
2589 | && (*p == '+' || *p == '?'))) | |
25fe55af | 2590 | ; |
ed0767d8 | 2591 | else if (syntax & RE_BK_PLUS_QM && *p == '\\') |
25fe55af | 2592 | { |
ed0767d8 SM |
2593 | if (p+1 == pend) |
2594 | FREE_STACK_RETURN (REG_EESCAPE); | |
2595 | if (p[1] == '+' || p[1] == '?') | |
2596 | PATFETCH (c); /* Gobble up the backslash. */ | |
2597 | else | |
2598 | break; | |
25fe55af RS |
2599 | } |
2600 | else | |
ed0767d8 | 2601 | break; |
25fe55af | 2602 | /* If we get here, we found another repeat character. */ |
ed0767d8 SM |
2603 | PATFETCH (c); |
2604 | } | |
25fe55af RS |
2605 | |
2606 | /* Star, etc. applied to an empty pattern is equivalent | |
2607 | to an empty pattern. */ | |
4e8a9132 | 2608 | if (!laststart || laststart == b) |
25fe55af RS |
2609 | break; |
2610 | ||
2611 | /* Now we know whether or not zero matches is allowed | |
2612 | and also whether or not two or more matches is allowed. */ | |
1c8c6d39 DL |
2613 | if (greedy) |
2614 | { | |
99633e97 | 2615 | if (many_times_ok) |
4e8a9132 SM |
2616 | { |
2617 | boolean simple = skip_one_char (laststart) == b; | |
2618 | unsigned int startoffset = 0; | |
f6a3f532 | 2619 | re_opcode_t ofj = |
01618498 | 2620 | /* Check if the loop can match the empty string. */ |
f6a3f532 SM |
2621 | (simple || !analyse_first (laststart, b, NULL, 0)) ? |
2622 | on_failure_jump : on_failure_jump_loop; | |
4e8a9132 SM |
2623 | assert (skip_one_char (laststart) <= b); |
2624 | ||
2625 | if (!zero_times_ok && simple) | |
2626 | { /* Since simple * loops can be made faster by using | |
2627 | on_failure_keep_string_jump, we turn simple P+ | |
2628 | into PP* if P is simple. */ | |
2629 | unsigned char *p1, *p2; | |
2630 | startoffset = b - laststart; | |
2631 | GET_BUFFER_SPACE (startoffset); | |
2632 | p1 = b; p2 = laststart; | |
2633 | while (p2 < p1) | |
2634 | *b++ = *p2++; | |
2635 | zero_times_ok = 1; | |
99633e97 | 2636 | } |
4e8a9132 SM |
2637 | |
2638 | GET_BUFFER_SPACE (6); | |
2639 | if (!zero_times_ok) | |
2640 | /* A + loop. */ | |
f6a3f532 | 2641 | STORE_JUMP (ofj, b, b + 6); |
99633e97 | 2642 | else |
4e8a9132 SM |
2643 | /* Simple * loops can use on_failure_keep_string_jump |
2644 | depending on what follows. But since we don't know | |
2645 | that yet, we leave the decision up to | |
2646 | on_failure_jump_smart. */ | |
f6a3f532 | 2647 | INSERT_JUMP (simple ? on_failure_jump_smart : ofj, |
4e8a9132 | 2648 | laststart + startoffset, b + 6); |
99633e97 | 2649 | b += 3; |
4e8a9132 | 2650 | STORE_JUMP (jump, b, laststart + startoffset); |
99633e97 SM |
2651 | b += 3; |
2652 | } | |
2653 | else | |
2654 | { | |
4e8a9132 SM |
2655 | /* A simple ? pattern. */ |
2656 | assert (zero_times_ok); | |
2657 | GET_BUFFER_SPACE (3); | |
2658 | INSERT_JUMP (on_failure_jump, laststart, b + 3); | |
99633e97 SM |
2659 | b += 3; |
2660 | } | |
1c8c6d39 DL |
2661 | } |
2662 | else /* not greedy */ | |
2663 | { /* I wish the greedy and non-greedy cases could be merged. */ | |
2664 | ||
0683b6fa | 2665 | GET_BUFFER_SPACE (7); /* We might use less. */ |
1c8c6d39 DL |
2666 | if (many_times_ok) |
2667 | { | |
f6a3f532 SM |
2668 | boolean emptyp = analyse_first (laststart, b, NULL, 0); |
2669 | ||
505bde11 | 2670 | /* The non-greedy multiple match looks like a repeat..until: |
1c8c6d39 | 2671 | we only need a conditional jump at the end of the loop */ |
f6a3f532 SM |
2672 | if (emptyp) BUF_PUSH (no_op); |
2673 | STORE_JUMP (emptyp ? on_failure_jump_nastyloop | |
2674 | : on_failure_jump, b, laststart); | |
1c8c6d39 DL |
2675 | b += 3; |
2676 | if (zero_times_ok) | |
2677 | { | |
2678 | /* The repeat...until naturally matches one or more. | |
2679 | To also match zero times, we need to first jump to | |
2680 | the end of the loop (its conditional jump). */ | |
1c8c6d39 DL |
2681 | INSERT_JUMP (jump, laststart, b); |
2682 | b += 3; | |
2683 | } | |
2684 | } | |
2685 | else | |
2686 | { | |
2687 | /* non-greedy a?? */ | |
1c8c6d39 DL |
2688 | INSERT_JUMP (jump, laststart, b + 3); |
2689 | b += 3; | |
2690 | INSERT_JUMP (on_failure_jump, laststart, laststart + 6); | |
2691 | b += 3; | |
2692 | } | |
2693 | } | |
2694 | } | |
4e8a9132 | 2695 | pending_exact = 0; |
fa9a63c5 RM |
2696 | break; |
2697 | ||
2698 | ||
2699 | case '.': | |
25fe55af RS |
2700 | laststart = b; |
2701 | BUF_PUSH (anychar); | |
2702 | break; | |
fa9a63c5 RM |
2703 | |
2704 | ||
25fe55af RS |
2705 | case '[': |
2706 | { | |
b18215fc | 2707 | CLEAR_RANGE_TABLE_WORK_USED (range_table_work); |
fa9a63c5 | 2708 | |
25fe55af | 2709 | if (p == pend) FREE_STACK_RETURN (REG_EBRACK); |
fa9a63c5 | 2710 | |
25fe55af RS |
2711 | /* Ensure that we have enough space to push a charset: the |
2712 | opcode, the length count, and the bitset; 34 bytes in all. */ | |
fa9a63c5 RM |
2713 | GET_BUFFER_SPACE (34); |
2714 | ||
25fe55af | 2715 | laststart = b; |
e318085a | 2716 | |
25fe55af RS |
2717 | /* We test `*p == '^' twice, instead of using an if |
2718 | statement, so we only need one BUF_PUSH. */ | |
2719 | BUF_PUSH (*p == '^' ? charset_not : charset); | |
2720 | if (*p == '^') | |
2721 | p++; | |
e318085a | 2722 | |
25fe55af RS |
2723 | /* Remember the first position in the bracket expression. */ |
2724 | p1 = p; | |
e318085a | 2725 | |
25fe55af RS |
2726 | /* Push the number of bytes in the bitmap. */ |
2727 | BUF_PUSH ((1 << BYTEWIDTH) / BYTEWIDTH); | |
e318085a | 2728 | |
25fe55af RS |
2729 | /* Clear the whole map. */ |
2730 | bzero (b, (1 << BYTEWIDTH) / BYTEWIDTH); | |
e318085a | 2731 | |
25fe55af RS |
2732 | /* charset_not matches newline according to a syntax bit. */ |
2733 | if ((re_opcode_t) b[-2] == charset_not | |
2734 | && (syntax & RE_HAT_LISTS_NOT_NEWLINE)) | |
2735 | SET_LIST_BIT ('\n'); | |
fa9a63c5 | 2736 | |
25fe55af RS |
2737 | /* Read in characters and ranges, setting map bits. */ |
2738 | for (;;) | |
2739 | { | |
b18215fc | 2740 | boolean escaped_char = false; |
2d1675e4 | 2741 | const unsigned char *p2 = p; |
e318085a | 2742 | |
25fe55af | 2743 | if (p == pend) FREE_STACK_RETURN (REG_EBRACK); |
e318085a | 2744 | |
36595814 SM |
2745 | /* Don't translate yet. The range TRANSLATE(X..Y) cannot |
2746 | always be determined from TRANSLATE(X) and TRANSLATE(Y) | |
2747 | So the translation is done later in a loop. Example: | |
2748 | (let ((case-fold-search t)) (string-match "[A-_]" "A")) */ | |
25fe55af | 2749 | PATFETCH (c); |
e318085a | 2750 | |
25fe55af RS |
2751 | /* \ might escape characters inside [...] and [^...]. */ |
2752 | if ((syntax & RE_BACKSLASH_ESCAPE_IN_LISTS) && c == '\\') | |
2753 | { | |
2754 | if (p == pend) FREE_STACK_RETURN (REG_EESCAPE); | |
e318085a RS |
2755 | |
2756 | PATFETCH (c); | |
b18215fc | 2757 | escaped_char = true; |
25fe55af | 2758 | } |
b18215fc RS |
2759 | else |
2760 | { | |
657fcfbd RS |
2761 | /* Could be the end of the bracket expression. If it's |
2762 | not (i.e., when the bracket expression is `[]' so | |
2763 | far), the ']' character bit gets set way below. */ | |
2d1675e4 | 2764 | if (c == ']' && p2 != p1) |
657fcfbd | 2765 | break; |
25fe55af | 2766 | } |
b18215fc | 2767 | |
b18215fc RS |
2768 | /* What should we do for the character which is |
2769 | greater than 0x7F, but not BASE_LEADING_CODE_P? | |
2770 | XXX */ | |
2771 | ||
25fe55af RS |
2772 | /* See if we're at the beginning of a possible character |
2773 | class. */ | |
b18215fc | 2774 | |
2d1675e4 SM |
2775 | if (!escaped_char && |
2776 | syntax & RE_CHAR_CLASSES && c == '[' && *p == ':') | |
657fcfbd RS |
2777 | { |
2778 | /* Leave room for the null. */ | |
14473664 | 2779 | unsigned char str[CHAR_CLASS_MAX_LENGTH + 1]; |
ed0767d8 | 2780 | const unsigned char *class_beg; |
b18215fc | 2781 | |
25fe55af RS |
2782 | PATFETCH (c); |
2783 | c1 = 0; | |
ed0767d8 | 2784 | class_beg = p; |
b18215fc | 2785 | |
25fe55af RS |
2786 | /* If pattern is `[[:'. */ |
2787 | if (p == pend) FREE_STACK_RETURN (REG_EBRACK); | |
b18215fc | 2788 | |
25fe55af RS |
2789 | for (;;) |
2790 | { | |
14473664 SM |
2791 | PATFETCH (c); |
2792 | if ((c == ':' && *p == ']') || p == pend) | |
2793 | break; | |
2794 | if (c1 < CHAR_CLASS_MAX_LENGTH) | |
2795 | str[c1++] = c; | |
2796 | else | |
2797 | /* This is in any case an invalid class name. */ | |
2798 | str[0] = '\0'; | |
25fe55af RS |
2799 | } |
2800 | str[c1] = '\0'; | |
b18215fc RS |
2801 | |
2802 | /* If isn't a word bracketed by `[:' and `:]': | |
2803 | undo the ending character, the letters, and | |
2804 | leave the leading `:' and `[' (but set bits for | |
2805 | them). */ | |
25fe55af RS |
2806 | if (c == ':' && *p == ']') |
2807 | { | |
36595814 | 2808 | re_wchar_t ch; |
14473664 SM |
2809 | re_wctype_t cc; |
2810 | ||
2811 | cc = re_wctype (str); | |
2812 | ||
2813 | if (cc == 0) | |
fa9a63c5 RM |
2814 | FREE_STACK_RETURN (REG_ECTYPE); |
2815 | ||
14473664 SM |
2816 | /* Throw away the ] at the end of the character |
2817 | class. */ | |
2818 | PATFETCH (c); | |
fa9a63c5 | 2819 | |
14473664 | 2820 | if (p == pend) FREE_STACK_RETURN (REG_EBRACK); |
fa9a63c5 | 2821 | |
96cc36cc RS |
2822 | /* Most character classes in a multibyte match |
2823 | just set a flag. Exceptions are is_blank, | |
2824 | is_digit, is_cntrl, and is_xdigit, since | |
2825 | they can only match ASCII characters. We | |
14473664 SM |
2826 | don't need to handle them for multibyte. |
2827 | They are distinguished by a negative wctype. */ | |
96cc36cc | 2828 | |
2d1675e4 | 2829 | if (multibyte) |
14473664 SM |
2830 | SET_RANGE_TABLE_WORK_AREA_BIT (range_table_work, |
2831 | re_wctype_to_bit (cc)); | |
96cc36cc | 2832 | |
14473664 | 2833 | for (ch = 0; ch < 1 << BYTEWIDTH; ++ch) |
25fe55af | 2834 | { |
7ae68633 | 2835 | int translated = TRANSLATE (ch); |
14473664 | 2836 | if (re_iswctype (btowc (ch), cc)) |
96cc36cc | 2837 | SET_LIST_BIT (translated); |
25fe55af | 2838 | } |
b18215fc RS |
2839 | |
2840 | /* Repeat the loop. */ | |
2841 | continue; | |
25fe55af RS |
2842 | } |
2843 | else | |
2844 | { | |
ed0767d8 SM |
2845 | /* Go back to right after the "[:". */ |
2846 | p = class_beg; | |
25fe55af | 2847 | SET_LIST_BIT ('['); |
b18215fc RS |
2848 | |
2849 | /* Because the `:' may starts the range, we | |
2850 | can't simply set bit and repeat the loop. | |
25fe55af | 2851 | Instead, just set it to C and handle below. */ |
b18215fc | 2852 | c = ':'; |
25fe55af RS |
2853 | } |
2854 | } | |
b18215fc RS |
2855 | |
2856 | if (p < pend && p[0] == '-' && p[1] != ']') | |
2857 | { | |
2858 | ||
2859 | /* Discard the `-'. */ | |
2860 | PATFETCH (c1); | |
2861 | ||
2862 | /* Fetch the character which ends the range. */ | |
2863 | PATFETCH (c1); | |
b18215fc | 2864 | |
b54f61ed | 2865 | if (SINGLE_BYTE_CHAR_P (c)) |
e934739e | 2866 | { |
b54f61ed KH |
2867 | if (! SINGLE_BYTE_CHAR_P (c1)) |
2868 | { | |
3ff2446d KH |
2869 | /* Handle a range starting with a |
2870 | character of less than 256, and ending | |
2871 | with a character of not less than 256. | |
2872 | Split that into two ranges, the low one | |
2873 | ending at 0377, and the high one | |
2874 | starting at the smallest character in | |
2875 | the charset of C1 and ending at C1. */ | |
b54f61ed | 2876 | int charset = CHAR_CHARSET (c1); |
36595814 SM |
2877 | re_wchar_t c2 = MAKE_CHAR (charset, 0, 0); |
2878 | ||
b54f61ed KH |
2879 | SET_RANGE_TABLE_WORK_AREA (range_table_work, |
2880 | c2, c1); | |
333526e0 | 2881 | c1 = 0377; |
b54f61ed | 2882 | } |
e934739e RS |
2883 | } |
2884 | else if (!SAME_CHARSET_P (c, c1)) | |
b18215fc | 2885 | FREE_STACK_RETURN (REG_ERANGE); |
e318085a | 2886 | } |
25fe55af | 2887 | else |
b18215fc RS |
2888 | /* Range from C to C. */ |
2889 | c1 = c; | |
2890 | ||
2891 | /* Set the range ... */ | |
2892 | if (SINGLE_BYTE_CHAR_P (c)) | |
2893 | /* ... into bitmap. */ | |
25fe55af | 2894 | { |
01618498 | 2895 | re_wchar_t this_char; |
36595814 | 2896 | re_wchar_t range_start = c, range_end = c1; |
b18215fc RS |
2897 | |
2898 | /* If the start is after the end, the range is empty. */ | |
2899 | if (range_start > range_end) | |
2900 | { | |
2901 | if (syntax & RE_NO_EMPTY_RANGES) | |
2902 | FREE_STACK_RETURN (REG_ERANGE); | |
2903 | /* Else, repeat the loop. */ | |
2904 | } | |
2905 | else | |
2906 | { | |
2907 | for (this_char = range_start; this_char <= range_end; | |
2908 | this_char++) | |
2909 | SET_LIST_BIT (TRANSLATE (this_char)); | |
e934739e | 2910 | } |
25fe55af | 2911 | } |
e318085a | 2912 | else |
b18215fc RS |
2913 | /* ... into range table. */ |
2914 | SET_RANGE_TABLE_WORK_AREA (range_table_work, c, c1); | |
e318085a RS |
2915 | } |
2916 | ||
25fe55af RS |
2917 | /* Discard any (non)matching list bytes that are all 0 at the |
2918 | end of the map. Decrease the map-length byte too. */ | |
2919 | while ((int) b[-1] > 0 && b[b[-1] - 1] == 0) | |
2920 | b[-1]--; | |
2921 | b += b[-1]; | |
fa9a63c5 | 2922 | |
96cc36cc RS |
2923 | /* Build real range table from work area. */ |
2924 | if (RANGE_TABLE_WORK_USED (range_table_work) | |
2925 | || RANGE_TABLE_WORK_BITS (range_table_work)) | |
b18215fc RS |
2926 | { |
2927 | int i; | |
2928 | int used = RANGE_TABLE_WORK_USED (range_table_work); | |
fa9a63c5 | 2929 | |
b18215fc | 2930 | /* Allocate space for COUNT + RANGE_TABLE. Needs two |
96cc36cc RS |
2931 | bytes for flags, two for COUNT, and three bytes for |
2932 | each character. */ | |
2933 | GET_BUFFER_SPACE (4 + used * 3); | |
fa9a63c5 | 2934 | |
b18215fc RS |
2935 | /* Indicate the existence of range table. */ |
2936 | laststart[1] |= 0x80; | |
fa9a63c5 | 2937 | |
96cc36cc RS |
2938 | /* Store the character class flag bits into the range table. |
2939 | If not in emacs, these flag bits are always 0. */ | |
2940 | *b++ = RANGE_TABLE_WORK_BITS (range_table_work) & 0xff; | |
2941 | *b++ = RANGE_TABLE_WORK_BITS (range_table_work) >> 8; | |
2942 | ||
b18215fc RS |
2943 | STORE_NUMBER_AND_INCR (b, used / 2); |
2944 | for (i = 0; i < used; i++) | |
2945 | STORE_CHARACTER_AND_INCR | |
2946 | (b, RANGE_TABLE_WORK_ELT (range_table_work, i)); | |
2947 | } | |
25fe55af RS |
2948 | } |
2949 | break; | |
fa9a63c5 RM |
2950 | |
2951 | ||
b18215fc | 2952 | case '(': |
25fe55af RS |
2953 | if (syntax & RE_NO_BK_PARENS) |
2954 | goto handle_open; | |
2955 | else | |
2956 | goto normal_char; | |
fa9a63c5 RM |
2957 | |
2958 | ||
25fe55af RS |
2959 | case ')': |
2960 | if (syntax & RE_NO_BK_PARENS) | |
2961 | goto handle_close; | |
2962 | else | |
2963 | goto normal_char; | |
e318085a RS |
2964 | |
2965 | ||
25fe55af RS |
2966 | case '\n': |
2967 | if (syntax & RE_NEWLINE_ALT) | |
2968 | goto handle_alt; | |
2969 | else | |
2970 | goto normal_char; | |
e318085a RS |
2971 | |
2972 | ||
b18215fc | 2973 | case '|': |
25fe55af RS |
2974 | if (syntax & RE_NO_BK_VBAR) |
2975 | goto handle_alt; | |
2976 | else | |
2977 | goto normal_char; | |
2978 | ||
2979 | ||
2980 | case '{': | |
2981 | if (syntax & RE_INTERVALS && syntax & RE_NO_BK_BRACES) | |
2982 | goto handle_interval; | |
2983 | else | |
2984 | goto normal_char; | |
2985 | ||
2986 | ||
2987 | case '\\': | |
2988 | if (p == pend) FREE_STACK_RETURN (REG_EESCAPE); | |
2989 | ||
2990 | /* Do not translate the character after the \, so that we can | |
2991 | distinguish, e.g., \B from \b, even if we normally would | |
2992 | translate, e.g., B to b. */ | |
36595814 | 2993 | PATFETCH (c); |
25fe55af RS |
2994 | |
2995 | switch (c) | |
2996 | { | |
2997 | case '(': | |
2998 | if (syntax & RE_NO_BK_PARENS) | |
2999 | goto normal_backslash; | |
3000 | ||
3001 | handle_open: | |
505bde11 SM |
3002 | { |
3003 | int shy = 0; | |
3004 | if (p+1 < pend) | |
3005 | { | |
3006 | /* Look for a special (?...) construct */ | |
ed0767d8 | 3007 | if ((syntax & RE_SHY_GROUPS) && *p == '?') |
505bde11 | 3008 | { |
ed0767d8 | 3009 | PATFETCH (c); /* Gobble up the '?'. */ |
505bde11 SM |
3010 | PATFETCH (c); |
3011 | switch (c) | |
3012 | { | |
3013 | case ':': shy = 1; break; | |
3014 | default: | |
3015 | /* Only (?:...) is supported right now. */ | |
3016 | FREE_STACK_RETURN (REG_BADPAT); | |
3017 | } | |
3018 | } | |
505bde11 SM |
3019 | } |
3020 | ||
3021 | if (!shy) | |
3022 | { | |
3023 | bufp->re_nsub++; | |
3024 | regnum++; | |
3025 | } | |
25fe55af | 3026 | |
99633e97 SM |
3027 | if (COMPILE_STACK_FULL) |
3028 | { | |
3029 | RETALLOC (compile_stack.stack, compile_stack.size << 1, | |
3030 | compile_stack_elt_t); | |
3031 | if (compile_stack.stack == NULL) return REG_ESPACE; | |
25fe55af | 3032 | |
99633e97 SM |
3033 | compile_stack.size <<= 1; |
3034 | } | |
25fe55af | 3035 | |
99633e97 SM |
3036 | /* These are the values to restore when we hit end of this |
3037 | group. They are all relative offsets, so that if the | |
3038 | whole pattern moves because of realloc, they will still | |
3039 | be valid. */ | |
3040 | COMPILE_STACK_TOP.begalt_offset = begalt - bufp->buffer; | |
3041 | COMPILE_STACK_TOP.fixup_alt_jump | |
3042 | = fixup_alt_jump ? fixup_alt_jump - bufp->buffer + 1 : 0; | |
3043 | COMPILE_STACK_TOP.laststart_offset = b - bufp->buffer; | |
3044 | COMPILE_STACK_TOP.regnum = shy ? -regnum : regnum; | |
3045 | ||
3046 | /* Do not push a | |
3047 | start_memory for groups beyond the last one we can | |
3048 | represent in the compiled pattern. */ | |
3049 | if (regnum <= MAX_REGNUM && !shy) | |
3050 | BUF_PUSH_2 (start_memory, regnum); | |
3051 | ||
3052 | compile_stack.avail++; | |
3053 | ||
3054 | fixup_alt_jump = 0; | |
3055 | laststart = 0; | |
3056 | begalt = b; | |
3057 | /* If we've reached MAX_REGNUM groups, then this open | |
3058 | won't actually generate any code, so we'll have to | |
3059 | clear pending_exact explicitly. */ | |
3060 | pending_exact = 0; | |
3061 | break; | |
505bde11 | 3062 | } |
25fe55af RS |
3063 | |
3064 | case ')': | |
3065 | if (syntax & RE_NO_BK_PARENS) goto normal_backslash; | |
3066 | ||
3067 | if (COMPILE_STACK_EMPTY) | |
505bde11 SM |
3068 | { |
3069 | if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD) | |
3070 | goto normal_backslash; | |
3071 | else | |
3072 | FREE_STACK_RETURN (REG_ERPAREN); | |
3073 | } | |
25fe55af RS |
3074 | |
3075 | handle_close: | |
505bde11 | 3076 | FIXUP_ALT_JUMP (); |
25fe55af RS |
3077 | |
3078 | /* See similar code for backslashed left paren above. */ | |
3079 | if (COMPILE_STACK_EMPTY) | |
505bde11 SM |
3080 | { |
3081 | if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD) | |
3082 | goto normal_char; | |
3083 | else | |
3084 | FREE_STACK_RETURN (REG_ERPAREN); | |
3085 | } | |
25fe55af RS |
3086 | |
3087 | /* Since we just checked for an empty stack above, this | |
3088 | ``can't happen''. */ | |
3089 | assert (compile_stack.avail != 0); | |
3090 | { | |
3091 | /* We don't just want to restore into `regnum', because | |
3092 | later groups should continue to be numbered higher, | |
3093 | as in `(ab)c(de)' -- the second group is #2. */ | |
3094 | regnum_t this_group_regnum; | |
3095 | ||
3096 | compile_stack.avail--; | |
3097 | begalt = bufp->buffer + COMPILE_STACK_TOP.begalt_offset; | |
3098 | fixup_alt_jump | |
3099 | = COMPILE_STACK_TOP.fixup_alt_jump | |
3100 | ? bufp->buffer + COMPILE_STACK_TOP.fixup_alt_jump - 1 | |
3101 | : 0; | |
3102 | laststart = bufp->buffer + COMPILE_STACK_TOP.laststart_offset; | |
3103 | this_group_regnum = COMPILE_STACK_TOP.regnum; | |
b18215fc RS |
3104 | /* If we've reached MAX_REGNUM groups, then this open |
3105 | won't actually generate any code, so we'll have to | |
3106 | clear pending_exact explicitly. */ | |
3107 | pending_exact = 0; | |
e318085a | 3108 | |
25fe55af RS |
3109 | /* We're at the end of the group, so now we know how many |
3110 | groups were inside this one. */ | |
505bde11 SM |
3111 | if (this_group_regnum <= MAX_REGNUM && this_group_regnum > 0) |
3112 | BUF_PUSH_2 (stop_memory, this_group_regnum); | |
25fe55af RS |
3113 | } |
3114 | break; | |
3115 | ||
3116 | ||
3117 | case '|': /* `\|'. */ | |
3118 | if (syntax & RE_LIMITED_OPS || syntax & RE_NO_BK_VBAR) | |
3119 | goto normal_backslash; | |
3120 | handle_alt: | |
3121 | if (syntax & RE_LIMITED_OPS) | |
3122 | goto normal_char; | |
3123 | ||
3124 | /* Insert before the previous alternative a jump which | |
3125 | jumps to this alternative if the former fails. */ | |
3126 | GET_BUFFER_SPACE (3); | |
3127 | INSERT_JUMP (on_failure_jump, begalt, b + 6); | |
3128 | pending_exact = 0; | |
3129 | b += 3; | |
3130 | ||
3131 | /* The alternative before this one has a jump after it | |
3132 | which gets executed if it gets matched. Adjust that | |
3133 | jump so it will jump to this alternative's analogous | |
3134 | jump (put in below, which in turn will jump to the next | |
3135 | (if any) alternative's such jump, etc.). The last such | |
3136 | jump jumps to the correct final destination. A picture: | |
3137 | _____ _____ | |
3138 | | | | | | |
3139 | | v | v | |
3140 | a | b | c | |
3141 | ||
3142 | If we are at `b', then fixup_alt_jump right now points to a | |
3143 | three-byte space after `a'. We'll put in the jump, set | |
3144 | fixup_alt_jump to right after `b', and leave behind three | |
3145 | bytes which we'll fill in when we get to after `c'. */ | |
3146 | ||
505bde11 | 3147 | FIXUP_ALT_JUMP (); |
25fe55af RS |
3148 | |
3149 | /* Mark and leave space for a jump after this alternative, | |
3150 | to be filled in later either by next alternative or | |
3151 | when know we're at the end of a series of alternatives. */ | |
3152 | fixup_alt_jump = b; | |
3153 | GET_BUFFER_SPACE (3); | |
3154 | b += 3; | |
3155 | ||
3156 | laststart = 0; | |
3157 | begalt = b; | |
3158 | break; | |
3159 | ||
3160 | ||
3161 | case '{': | |
3162 | /* If \{ is a literal. */ | |
3163 | if (!(syntax & RE_INTERVALS) | |
3164 | /* If we're at `\{' and it's not the open-interval | |
3165 | operator. */ | |
4bb91c68 | 3166 | || (syntax & RE_NO_BK_BRACES)) |
25fe55af RS |
3167 | goto normal_backslash; |
3168 | ||
3169 | handle_interval: | |
3170 | { | |
3171 | /* If got here, then the syntax allows intervals. */ | |
3172 | ||
3173 | /* At least (most) this many matches must be made. */ | |
99633e97 | 3174 | int lower_bound = 0, upper_bound = -1; |
25fe55af | 3175 | |
ed0767d8 | 3176 | beg_interval = p; |
25fe55af RS |
3177 | |
3178 | if (p == pend) | |
4bb91c68 | 3179 | FREE_STACK_RETURN (REG_EBRACE); |
25fe55af RS |
3180 | |
3181 | GET_UNSIGNED_NUMBER (lower_bound); | |
3182 | ||
3183 | if (c == ',') | |
ed0767d8 | 3184 | GET_UNSIGNED_NUMBER (upper_bound); |
25fe55af RS |
3185 | else |
3186 | /* Interval such as `{1}' => match exactly once. */ | |
3187 | upper_bound = lower_bound; | |
3188 | ||
3189 | if (lower_bound < 0 || upper_bound > RE_DUP_MAX | |
ed0767d8 | 3190 | || (upper_bound >= 0 && lower_bound > upper_bound)) |
4bb91c68 | 3191 | FREE_STACK_RETURN (REG_BADBR); |
25fe55af RS |
3192 | |
3193 | if (!(syntax & RE_NO_BK_BRACES)) | |
3194 | { | |
4bb91c68 SM |
3195 | if (c != '\\') |
3196 | FREE_STACK_RETURN (REG_BADBR); | |
25fe55af RS |
3197 | |
3198 | PATFETCH (c); | |
3199 | } | |
3200 | ||
3201 | if (c != '}') | |
4bb91c68 | 3202 | FREE_STACK_RETURN (REG_BADBR); |
25fe55af RS |
3203 | |
3204 | /* We just parsed a valid interval. */ | |
3205 | ||
3206 | /* If it's invalid to have no preceding re. */ | |
3207 | if (!laststart) | |
3208 | { | |
3209 | if (syntax & RE_CONTEXT_INVALID_OPS) | |
3210 | FREE_STACK_RETURN (REG_BADRPT); | |
3211 | else if (syntax & RE_CONTEXT_INDEP_OPS) | |
3212 | laststart = b; | |
3213 | else | |
3214 | goto unfetch_interval; | |
3215 | } | |
3216 | ||
25fe55af | 3217 | if (upper_bound == 0) |
ed0767d8 SM |
3218 | /* If the upper bound is zero, just drop the sub pattern |
3219 | altogether. */ | |
3220 | b = laststart; | |
3221 | else if (lower_bound == 1 && upper_bound == 1) | |
3222 | /* Just match it once: nothing to do here. */ | |
3223 | ; | |
25fe55af RS |
3224 | |
3225 | /* Otherwise, we have a nontrivial interval. When | |
3226 | we're all done, the pattern will look like: | |
3227 | set_number_at <jump count> <upper bound> | |
3228 | set_number_at <succeed_n count> <lower bound> | |
3229 | succeed_n <after jump addr> <succeed_n count> | |
3230 | <body of loop> | |
3231 | jump_n <succeed_n addr> <jump count> | |
3232 | (The upper bound and `jump_n' are omitted if | |
3233 | `upper_bound' is 1, though.) */ | |
3234 | else | |
3235 | { /* If the upper bound is > 1, we need to insert | |
3236 | more at the end of the loop. */ | |
ed0767d8 SM |
3237 | unsigned int nbytes = (upper_bound < 0 ? 3 |
3238 | : upper_bound > 1 ? 5 : 0); | |
3239 | unsigned int startoffset = 0; | |
3240 | ||
3241 | GET_BUFFER_SPACE (20); /* We might use less. */ | |
3242 | ||
3243 | if (lower_bound == 0) | |
3244 | { | |
3245 | /* A succeed_n that starts with 0 is really a | |
3246 | a simple on_failure_jump_loop. */ | |
3247 | INSERT_JUMP (on_failure_jump_loop, laststart, | |
3248 | b + 3 + nbytes); | |
3249 | b += 3; | |
3250 | } | |
3251 | else | |
3252 | { | |
3253 | /* Initialize lower bound of the `succeed_n', even | |
3254 | though it will be set during matching by its | |
3255 | attendant `set_number_at' (inserted next), | |
3256 | because `re_compile_fastmap' needs to know. | |
3257 | Jump to the `jump_n' we might insert below. */ | |
3258 | INSERT_JUMP2 (succeed_n, laststart, | |
3259 | b + 5 + nbytes, | |
3260 | lower_bound); | |
3261 | b += 5; | |
3262 | ||
3263 | /* Code to initialize the lower bound. Insert | |
3264 | before the `succeed_n'. The `5' is the last two | |
3265 | bytes of this `set_number_at', plus 3 bytes of | |
3266 | the following `succeed_n'. */ | |
3267 | insert_op2 (set_number_at, laststart, 5, lower_bound, b); | |
3268 | b += 5; | |
3269 | startoffset += 5; | |
3270 | } | |
3271 | ||
3272 | if (upper_bound < 0) | |
3273 | { | |
3274 | /* A negative upper bound stands for infinity, | |
3275 | in which case it degenerates to a plain jump. */ | |
3276 | STORE_JUMP (jump, b, laststart + startoffset); | |
3277 | b += 3; | |
3278 | } | |
3279 | else if (upper_bound > 1) | |
25fe55af RS |
3280 | { /* More than one repetition is allowed, so |
3281 | append a backward jump to the `succeed_n' | |
3282 | that starts this interval. | |
3283 | ||
3284 | When we've reached this during matching, | |
3285 | we'll have matched the interval once, so | |
3286 | jump back only `upper_bound - 1' times. */ | |
ed0767d8 | 3287 | STORE_JUMP2 (jump_n, b, laststart + startoffset, |
25fe55af RS |
3288 | upper_bound - 1); |
3289 | b += 5; | |
3290 | ||
3291 | /* The location we want to set is the second | |
3292 | parameter of the `jump_n'; that is `b-2' as | |
3293 | an absolute address. `laststart' will be | |
3294 | the `set_number_at' we're about to insert; | |
3295 | `laststart+3' the number to set, the source | |
3296 | for the relative address. But we are | |
3297 | inserting into the middle of the pattern -- | |
3298 | so everything is getting moved up by 5. | |
3299 | Conclusion: (b - 2) - (laststart + 3) + 5, | |
3300 | i.e., b - laststart. | |
3301 | ||
3302 | We insert this at the beginning of the loop | |
3303 | so that if we fail during matching, we'll | |
3304 | reinitialize the bounds. */ | |
3305 | insert_op2 (set_number_at, laststart, b - laststart, | |
3306 | upper_bound - 1, b); | |
3307 | b += 5; | |
3308 | } | |
3309 | } | |
3310 | pending_exact = 0; | |
3311 | beg_interval = NULL; | |
3312 | } | |
3313 | break; | |
3314 | ||
3315 | unfetch_interval: | |
3316 | /* If an invalid interval, match the characters as literals. */ | |
3317 | assert (beg_interval); | |
3318 | p = beg_interval; | |
3319 | beg_interval = NULL; | |
3320 | ||
3321 | /* normal_char and normal_backslash need `c'. */ | |
ed0767d8 | 3322 | c = '{'; |
25fe55af RS |
3323 | |
3324 | if (!(syntax & RE_NO_BK_BRACES)) | |
3325 | { | |
ed0767d8 SM |
3326 | assert (p > pattern && p[-1] == '\\'); |
3327 | goto normal_backslash; | |
25fe55af | 3328 | } |
ed0767d8 SM |
3329 | else |
3330 | goto normal_char; | |
e318085a | 3331 | |
b18215fc | 3332 | #ifdef emacs |
25fe55af RS |
3333 | /* There is no way to specify the before_dot and after_dot |
3334 | operators. rms says this is ok. --karl */ | |
3335 | case '=': | |
3336 | BUF_PUSH (at_dot); | |
3337 | break; | |
3338 | ||
3339 | case 's': | |
3340 | laststart = b; | |
3341 | PATFETCH (c); | |
3342 | BUF_PUSH_2 (syntaxspec, syntax_spec_code[c]); | |
3343 | break; | |
3344 | ||
3345 | case 'S': | |
3346 | laststart = b; | |
3347 | PATFETCH (c); | |
3348 | BUF_PUSH_2 (notsyntaxspec, syntax_spec_code[c]); | |
3349 | break; | |
b18215fc RS |
3350 | |
3351 | case 'c': | |
3352 | laststart = b; | |
36595814 | 3353 | PATFETCH (c); |
b18215fc RS |
3354 | BUF_PUSH_2 (categoryspec, c); |
3355 | break; | |
e318085a | 3356 | |
b18215fc RS |
3357 | case 'C': |
3358 | laststart = b; | |
36595814 | 3359 | PATFETCH (c); |
b18215fc RS |
3360 | BUF_PUSH_2 (notcategoryspec, c); |
3361 | break; | |
3362 | #endif /* emacs */ | |
e318085a | 3363 | |
e318085a | 3364 | |
25fe55af | 3365 | case 'w': |
4bb91c68 SM |
3366 | if (syntax & RE_NO_GNU_OPS) |
3367 | goto normal_char; | |
25fe55af | 3368 | laststart = b; |
1fb352e0 | 3369 | BUF_PUSH_2 (syntaxspec, Sword); |
25fe55af | 3370 | break; |
e318085a | 3371 | |
e318085a | 3372 | |
25fe55af | 3373 | case 'W': |
4bb91c68 SM |
3374 | if (syntax & RE_NO_GNU_OPS) |
3375 | goto normal_char; | |
25fe55af | 3376 | laststart = b; |
1fb352e0 | 3377 | BUF_PUSH_2 (notsyntaxspec, Sword); |
25fe55af | 3378 | break; |
e318085a RS |
3379 | |
3380 | ||
25fe55af | 3381 | case '<': |
4bb91c68 SM |
3382 | if (syntax & RE_NO_GNU_OPS) |
3383 | goto normal_char; | |
25fe55af RS |
3384 | BUF_PUSH (wordbeg); |
3385 | break; | |
e318085a | 3386 | |
25fe55af | 3387 | case '>': |
4bb91c68 SM |
3388 | if (syntax & RE_NO_GNU_OPS) |
3389 | goto normal_char; | |
25fe55af RS |
3390 | BUF_PUSH (wordend); |
3391 | break; | |
e318085a | 3392 | |
25fe55af | 3393 | case 'b': |
4bb91c68 SM |
3394 | if (syntax & RE_NO_GNU_OPS) |
3395 | goto normal_char; | |
25fe55af RS |
3396 | BUF_PUSH (wordbound); |
3397 | break; | |
e318085a | 3398 | |
25fe55af | 3399 | case 'B': |
4bb91c68 SM |
3400 | if (syntax & RE_NO_GNU_OPS) |
3401 | goto normal_char; | |
25fe55af RS |
3402 | BUF_PUSH (notwordbound); |
3403 | break; | |
fa9a63c5 | 3404 | |
25fe55af | 3405 | case '`': |
4bb91c68 SM |
3406 | if (syntax & RE_NO_GNU_OPS) |
3407 | goto normal_char; | |
25fe55af RS |
3408 | BUF_PUSH (begbuf); |
3409 | break; | |
e318085a | 3410 | |
25fe55af | 3411 | case '\'': |
4bb91c68 SM |
3412 | if (syntax & RE_NO_GNU_OPS) |
3413 | goto normal_char; | |
25fe55af RS |
3414 | BUF_PUSH (endbuf); |
3415 | break; | |
e318085a | 3416 | |
25fe55af RS |
3417 | case '1': case '2': case '3': case '4': case '5': |
3418 | case '6': case '7': case '8': case '9': | |
0cdd06f8 SM |
3419 | { |
3420 | regnum_t reg; | |
e318085a | 3421 | |
0cdd06f8 SM |
3422 | if (syntax & RE_NO_BK_REFS) |
3423 | goto normal_backslash; | |
e318085a | 3424 | |
0cdd06f8 | 3425 | reg = c - '0'; |
e318085a | 3426 | |
0cdd06f8 SM |
3427 | /* Can't back reference to a subexpression before its end. */ |
3428 | if (reg > regnum || group_in_compile_stack (compile_stack, reg)) | |
3429 | FREE_STACK_RETURN (REG_ESUBREG); | |
e318085a | 3430 | |
0cdd06f8 SM |
3431 | laststart = b; |
3432 | BUF_PUSH_2 (duplicate, reg); | |
3433 | } | |
25fe55af | 3434 | break; |
e318085a | 3435 | |
e318085a | 3436 | |
25fe55af RS |
3437 | case '+': |
3438 | case '?': | |
3439 | if (syntax & RE_BK_PLUS_QM) | |
3440 | goto handle_plus; | |
3441 | else | |
3442 | goto normal_backslash; | |
3443 | ||
3444 | default: | |
3445 | normal_backslash: | |
3446 | /* You might think it would be useful for \ to mean | |
3447 | not to translate; but if we don't translate it | |
4bb91c68 | 3448 | it will never match anything. */ |
25fe55af RS |
3449 | goto normal_char; |
3450 | } | |
3451 | break; | |
fa9a63c5 RM |
3452 | |
3453 | ||
3454 | default: | |
25fe55af | 3455 | /* Expects the character in `c'. */ |
fa9a63c5 | 3456 | normal_char: |
36595814 | 3457 | /* If no exactn currently being built. */ |
25fe55af | 3458 | if (!pending_exact |
fa9a63c5 | 3459 | |
25fe55af RS |
3460 | /* If last exactn not at current position. */ |
3461 | || pending_exact + *pending_exact + 1 != b | |
5e69f11e | 3462 | |
25fe55af | 3463 | /* We have only one byte following the exactn for the count. */ |
2d1675e4 | 3464 | || *pending_exact >= (1 << BYTEWIDTH) - MAX_MULTIBYTE_LENGTH |
fa9a63c5 | 3465 | |
25fe55af | 3466 | /* If followed by a repetition operator. */ |
9d99031f | 3467 | || (p != pend && (*p == '*' || *p == '^')) |
fa9a63c5 | 3468 | || ((syntax & RE_BK_PLUS_QM) |
9d99031f RS |
3469 | ? p + 1 < pend && *p == '\\' && (p[1] == '+' || p[1] == '?') |
3470 | : p != pend && (*p == '+' || *p == '?')) | |
fa9a63c5 | 3471 | || ((syntax & RE_INTERVALS) |
25fe55af | 3472 | && ((syntax & RE_NO_BK_BRACES) |
9d99031f RS |
3473 | ? p != pend && *p == '{' |
3474 | : p + 1 < pend && p[0] == '\\' && p[1] == '{'))) | |
fa9a63c5 RM |
3475 | { |
3476 | /* Start building a new exactn. */ | |
5e69f11e | 3477 | |
25fe55af | 3478 | laststart = b; |
fa9a63c5 RM |
3479 | |
3480 | BUF_PUSH_2 (exactn, 0); | |
3481 | pending_exact = b - 1; | |
25fe55af | 3482 | } |
5e69f11e | 3483 | |
2d1675e4 SM |
3484 | GET_BUFFER_SPACE (MAX_MULTIBYTE_LENGTH); |
3485 | { | |
e0277a47 KH |
3486 | int len; |
3487 | ||
36595814 | 3488 | c = TRANSLATE (c); |
e0277a47 KH |
3489 | if (multibyte) |
3490 | len = CHAR_STRING (c, b); | |
3491 | else | |
3492 | *b = c, len = 1; | |
2d1675e4 SM |
3493 | b += len; |
3494 | (*pending_exact) += len; | |
3495 | } | |
3496 | ||
fa9a63c5 | 3497 | break; |
25fe55af | 3498 | } /* switch (c) */ |
fa9a63c5 RM |
3499 | } /* while p != pend */ |
3500 | ||
5e69f11e | 3501 | |
fa9a63c5 | 3502 | /* Through the pattern now. */ |
5e69f11e | 3503 | |
505bde11 | 3504 | FIXUP_ALT_JUMP (); |
fa9a63c5 | 3505 | |
5e69f11e | 3506 | if (!COMPILE_STACK_EMPTY) |
fa9a63c5 RM |
3507 | FREE_STACK_RETURN (REG_EPAREN); |
3508 | ||
3509 | /* If we don't want backtracking, force success | |
3510 | the first time we reach the end of the compiled pattern. */ | |
3511 | if (syntax & RE_NO_POSIX_BACKTRACKING) | |
3512 | BUF_PUSH (succeed); | |
3513 | ||
3514 | free (compile_stack.stack); | |
3515 | ||
3516 | /* We have succeeded; set the length of the buffer. */ | |
3517 | bufp->used = b - bufp->buffer; | |
3518 | ||
3519 | #ifdef DEBUG | |
99633e97 | 3520 | if (debug > 0) |
fa9a63c5 | 3521 | { |
505bde11 | 3522 | re_compile_fastmap (bufp); |
fa9a63c5 RM |
3523 | DEBUG_PRINT1 ("\nCompiled pattern: \n"); |
3524 | print_compiled_pattern (bufp); | |
3525 | } | |
99633e97 | 3526 | debug--; |
fa9a63c5 RM |
3527 | #endif /* DEBUG */ |
3528 | ||
3529 | #ifndef MATCH_MAY_ALLOCATE | |
3530 | /* Initialize the failure stack to the largest possible stack. This | |
3531 | isn't necessary unless we're trying to avoid calling alloca in | |
3532 | the search and match routines. */ | |
3533 | { | |
3534 | int num_regs = bufp->re_nsub + 1; | |
3535 | ||
320a2a73 | 3536 | if (fail_stack.size < re_max_failures * TYPICAL_FAILURE_SIZE) |
fa9a63c5 | 3537 | { |
a26f4ccd | 3538 | fail_stack.size = re_max_failures * TYPICAL_FAILURE_SIZE; |
fa9a63c5 | 3539 | |
fa9a63c5 RM |
3540 | if (! fail_stack.stack) |
3541 | fail_stack.stack | |
5e69f11e | 3542 | = (fail_stack_elt_t *) malloc (fail_stack.size |
fa9a63c5 RM |
3543 | * sizeof (fail_stack_elt_t)); |
3544 | else | |
3545 | fail_stack.stack | |
3546 | = (fail_stack_elt_t *) realloc (fail_stack.stack, | |
3547 | (fail_stack.size | |
3548 | * sizeof (fail_stack_elt_t))); | |
fa9a63c5 RM |
3549 | } |
3550 | ||
3551 | regex_grow_registers (num_regs); | |
3552 | } | |
3553 | #endif /* not MATCH_MAY_ALLOCATE */ | |
3554 | ||
3555 | return REG_NOERROR; | |
3556 | } /* regex_compile */ | |
3557 | \f | |
3558 | /* Subroutines for `regex_compile'. */ | |
3559 | ||
25fe55af | 3560 | /* Store OP at LOC followed by two-byte integer parameter ARG. */ |
fa9a63c5 RM |
3561 | |
3562 | static void | |
3563 | store_op1 (op, loc, arg) | |
3564 | re_opcode_t op; | |
3565 | unsigned char *loc; | |
3566 | int arg; | |
3567 | { | |
3568 | *loc = (unsigned char) op; | |
3569 | STORE_NUMBER (loc + 1, arg); | |
3570 | } | |
3571 | ||
3572 | ||
3573 | /* Like `store_op1', but for two two-byte parameters ARG1 and ARG2. */ | |
3574 | ||
3575 | static void | |
3576 | store_op2 (op, loc, arg1, arg2) | |
3577 | re_opcode_t op; | |
3578 | unsigned char *loc; | |
3579 | int arg1, arg2; | |
3580 | { | |
3581 | *loc = (unsigned char) op; | |
3582 | STORE_NUMBER (loc + 1, arg1); | |
3583 | STORE_NUMBER (loc + 3, arg2); | |
3584 | } | |
3585 | ||
3586 | ||
3587 | /* Copy the bytes from LOC to END to open up three bytes of space at LOC | |
3588 | for OP followed by two-byte integer parameter ARG. */ | |
3589 | ||
3590 | static void | |
3591 | insert_op1 (op, loc, arg, end) | |
3592 | re_opcode_t op; | |
3593 | unsigned char *loc; | |
3594 | int arg; | |
5e69f11e | 3595 | unsigned char *end; |
fa9a63c5 RM |
3596 | { |
3597 | register unsigned char *pfrom = end; | |
3598 | register unsigned char *pto = end + 3; | |
3599 | ||
3600 | while (pfrom != loc) | |
3601 | *--pto = *--pfrom; | |
5e69f11e | 3602 | |
fa9a63c5 RM |
3603 | store_op1 (op, loc, arg); |
3604 | } | |
3605 | ||
3606 | ||
3607 | /* Like `insert_op1', but for two two-byte parameters ARG1 and ARG2. */ | |
3608 | ||
3609 | static void | |
3610 | insert_op2 (op, loc, arg1, arg2, end) | |
3611 | re_opcode_t op; | |
3612 | unsigned char *loc; | |
3613 | int arg1, arg2; | |
5e69f11e | 3614 | unsigned char *end; |
fa9a63c5 RM |
3615 | { |
3616 | register unsigned char *pfrom = end; | |
3617 | register unsigned char *pto = end + 5; | |
3618 | ||
3619 | while (pfrom != loc) | |
3620 | *--pto = *--pfrom; | |
5e69f11e | 3621 | |
fa9a63c5 RM |
3622 | store_op2 (op, loc, arg1, arg2); |
3623 | } | |
3624 | ||
3625 | ||
3626 | /* P points to just after a ^ in PATTERN. Return true if that ^ comes | |
3627 | after an alternative or a begin-subexpression. We assume there is at | |
3628 | least one character before the ^. */ | |
3629 | ||
3630 | static boolean | |
3631 | at_begline_loc_p (pattern, p, syntax) | |
01618498 | 3632 | re_char *pattern, *p; |
fa9a63c5 RM |
3633 | reg_syntax_t syntax; |
3634 | { | |
01618498 | 3635 | re_char *prev = p - 2; |
fa9a63c5 | 3636 | boolean prev_prev_backslash = prev > pattern && prev[-1] == '\\'; |
5e69f11e | 3637 | |
fa9a63c5 RM |
3638 | return |
3639 | /* After a subexpression? */ | |
3640 | (*prev == '(' && (syntax & RE_NO_BK_PARENS || prev_prev_backslash)) | |
25fe55af | 3641 | /* After an alternative? */ |
d2af47df SM |
3642 | || (*prev == '|' && (syntax & RE_NO_BK_VBAR || prev_prev_backslash)) |
3643 | /* After a shy subexpression? */ | |
3644 | || ((syntax & RE_SHY_GROUPS) && prev - 2 >= pattern | |
3645 | && prev[-1] == '?' && prev[-2] == '(' | |
3646 | && (syntax & RE_NO_BK_PARENS | |
3647 | || (prev - 3 >= pattern && prev[-3] == '\\'))); | |
fa9a63c5 RM |
3648 | } |
3649 | ||
3650 | ||
3651 | /* The dual of at_begline_loc_p. This one is for $. We assume there is | |
3652 | at least one character after the $, i.e., `P < PEND'. */ | |
3653 | ||
3654 | static boolean | |
3655 | at_endline_loc_p (p, pend, syntax) | |
01618498 | 3656 | re_char *p, *pend; |
99633e97 | 3657 | reg_syntax_t syntax; |
fa9a63c5 | 3658 | { |
01618498 | 3659 | re_char *next = p; |
fa9a63c5 | 3660 | boolean next_backslash = *next == '\\'; |
01618498 | 3661 | re_char *next_next = p + 1 < pend ? p + 1 : 0; |
5e69f11e | 3662 | |
fa9a63c5 RM |
3663 | return |
3664 | /* Before a subexpression? */ | |
3665 | (syntax & RE_NO_BK_PARENS ? *next == ')' | |
25fe55af | 3666 | : next_backslash && next_next && *next_next == ')') |
fa9a63c5 RM |
3667 | /* Before an alternative? */ |
3668 | || (syntax & RE_NO_BK_VBAR ? *next == '|' | |
25fe55af | 3669 | : next_backslash && next_next && *next_next == '|'); |
fa9a63c5 RM |
3670 | } |
3671 | ||
3672 | ||
5e69f11e | 3673 | /* Returns true if REGNUM is in one of COMPILE_STACK's elements and |
fa9a63c5 RM |
3674 | false if it's not. */ |
3675 | ||
3676 | static boolean | |
3677 | group_in_compile_stack (compile_stack, regnum) | |
3678 | compile_stack_type compile_stack; | |
3679 | regnum_t regnum; | |
3680 | { | |
3681 | int this_element; | |
3682 | ||
5e69f11e RM |
3683 | for (this_element = compile_stack.avail - 1; |
3684 | this_element >= 0; | |
fa9a63c5 RM |
3685 | this_element--) |
3686 | if (compile_stack.stack[this_element].regnum == regnum) | |
3687 | return true; | |
3688 | ||
3689 | return false; | |
3690 | } | |
fa9a63c5 | 3691 | \f |
f6a3f532 SM |
3692 | /* analyse_first. |
3693 | If fastmap is non-NULL, go through the pattern and fill fastmap | |
3694 | with all the possible leading chars. If fastmap is NULL, don't | |
3695 | bother filling it up (obviously) and only return whether the | |
3696 | pattern could potentially match the empty string. | |
3697 | ||
3698 | Return 1 if p..pend might match the empty string. | |
3699 | Return 0 if p..pend matches at least one char. | |
01618498 | 3700 | Return -1 if fastmap was not updated accurately. */ |
f6a3f532 SM |
3701 | |
3702 | static int | |
3703 | analyse_first (p, pend, fastmap, multibyte) | |
01618498 | 3704 | re_char *p, *pend; |
f6a3f532 SM |
3705 | char *fastmap; |
3706 | const int multibyte; | |
fa9a63c5 | 3707 | { |
505bde11 | 3708 | int j, k; |
1fb352e0 | 3709 | boolean not; |
fa9a63c5 | 3710 | |
b18215fc | 3711 | /* If all elements for base leading-codes in fastmap is set, this |
25fe55af | 3712 | flag is set true. */ |
b18215fc RS |
3713 | boolean match_any_multibyte_characters = false; |
3714 | ||
f6a3f532 | 3715 | assert (p); |
5e69f11e | 3716 | |
505bde11 SM |
3717 | /* The loop below works as follows: |
3718 | - It has a working-list kept in the PATTERN_STACK and which basically | |
3719 | starts by only containing a pointer to the first operation. | |
3720 | - If the opcode we're looking at is a match against some set of | |
3721 | chars, then we add those chars to the fastmap and go on to the | |
3722 | next work element from the worklist (done via `break'). | |
3723 | - If the opcode is a control operator on the other hand, we either | |
3724 | ignore it (if it's meaningless at this point, such as `start_memory') | |
3725 | or execute it (if it's a jump). If the jump has several destinations | |
3726 | (i.e. `on_failure_jump'), then we push the other destination onto the | |
3727 | worklist. | |
3728 | We guarantee termination by ignoring backward jumps (more or less), | |
3729 | so that `p' is monotonically increasing. More to the point, we | |
3730 | never set `p' (or push) anything `<= p1'. */ | |
3731 | ||
01618498 | 3732 | while (p < pend) |
fa9a63c5 | 3733 | { |
505bde11 SM |
3734 | /* `p1' is used as a marker of how far back a `on_failure_jump' |
3735 | can go without being ignored. It is normally equal to `p' | |
3736 | (which prevents any backward `on_failure_jump') except right | |
3737 | after a plain `jump', to allow patterns such as: | |
3738 | 0: jump 10 | |
3739 | 3..9: <body> | |
3740 | 10: on_failure_jump 3 | |
3741 | as used for the *? operator. */ | |
01618498 | 3742 | re_char *p1 = p; |
5e69f11e | 3743 | |
fa9a63c5 RM |
3744 | switch (SWITCH_ENUM_CAST ((re_opcode_t) *p++)) |
3745 | { | |
f6a3f532 | 3746 | case succeed: |
01618498 | 3747 | return 1; |
f6a3f532 | 3748 | continue; |
fa9a63c5 | 3749 | |
fa9a63c5 | 3750 | case duplicate: |
505bde11 SM |
3751 | /* If the first character has to match a backreference, that means |
3752 | that the group was empty (since it already matched). Since this | |
3753 | is the only case that interests us here, we can assume that the | |
3754 | backreference must match the empty string. */ | |
3755 | p++; | |
3756 | continue; | |
fa9a63c5 RM |
3757 | |
3758 | ||
3759 | /* Following are the cases which match a character. These end | |
25fe55af | 3760 | with `break'. */ |
fa9a63c5 RM |
3761 | |
3762 | case exactn: | |
e0277a47 KH |
3763 | if (fastmap) |
3764 | { | |
3765 | int c = RE_STRING_CHAR (p + 1, pend - p); | |
3766 | ||
3767 | if (SINGLE_BYTE_CHAR_P (c)) | |
3768 | fastmap[c] = 1; | |
3769 | else | |
3770 | fastmap[p[1]] = 1; | |
3771 | } | |
fa9a63c5 RM |
3772 | break; |
3773 | ||
3774 | ||
1fb352e0 SM |
3775 | case anychar: |
3776 | /* We could put all the chars except for \n (and maybe \0) | |
3777 | but we don't bother since it is generally not worth it. */ | |
f6a3f532 | 3778 | if (!fastmap) break; |
01618498 | 3779 | return -1; |
fa9a63c5 RM |
3780 | |
3781 | ||
b18215fc | 3782 | case charset_not: |
ba5c004d RS |
3783 | /* Chars beyond end of bitmap are possible matches. |
3784 | All the single-byte codes can occur in multibyte buffers. | |
3785 | So any that are not listed in the charset | |
3786 | are possible matches, even in multibyte buffers. */ | |
1fb352e0 | 3787 | if (!fastmap) break; |
b18215fc | 3788 | for (j = CHARSET_BITMAP_SIZE (&p[-1]) * BYTEWIDTH; |
1fb352e0 | 3789 | j < (1 << BYTEWIDTH); j++) |
b18215fc | 3790 | fastmap[j] = 1; |
1fb352e0 SM |
3791 | /* Fallthrough */ |
3792 | case charset: | |
3793 | if (!fastmap) break; | |
3794 | not = (re_opcode_t) *(p - 1) == charset_not; | |
b18215fc RS |
3795 | for (j = CHARSET_BITMAP_SIZE (&p[-1]) * BYTEWIDTH - 1, p++; |
3796 | j >= 0; j--) | |
1fb352e0 | 3797 | if (!!(p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH))) ^ not) |
b18215fc RS |
3798 | fastmap[j] = 1; |
3799 | ||
1fb352e0 SM |
3800 | if ((not && multibyte) |
3801 | /* Any character set can possibly contain a character | |
3802 | which doesn't match the specified set of characters. */ | |
3803 | || (CHARSET_RANGE_TABLE_EXISTS_P (&p[-2]) | |
3804 | && CHARSET_RANGE_TABLE_BITS (&p[-2]) != 0)) | |
3805 | /* If we can match a character class, we can match | |
3806 | any character set. */ | |
b18215fc RS |
3807 | { |
3808 | set_fastmap_for_multibyte_characters: | |
3809 | if (match_any_multibyte_characters == false) | |
3810 | { | |
3811 | for (j = 0x80; j < 0xA0; j++) /* XXX */ | |
3812 | if (BASE_LEADING_CODE_P (j)) | |
3813 | fastmap[j] = 1; | |
3814 | match_any_multibyte_characters = true; | |
3815 | } | |
3816 | } | |
b18215fc | 3817 | |
1fb352e0 SM |
3818 | else if (!not && CHARSET_RANGE_TABLE_EXISTS_P (&p[-2]) |
3819 | && match_any_multibyte_characters == false) | |
3820 | { | |
3821 | /* Set fastmap[I] 1 where I is a base leading code of each | |
3822 | multibyte character in the range table. */ | |
3823 | int c, count; | |
b18215fc | 3824 | |
1fb352e0 | 3825 | /* Make P points the range table. `+ 2' is to skip flag |
0b32bf0e | 3826 | bits for a character class. */ |
1fb352e0 | 3827 | p += CHARSET_BITMAP_SIZE (&p[-2]) + 2; |
b18215fc | 3828 | |
1fb352e0 SM |
3829 | /* Extract the number of ranges in range table into COUNT. */ |
3830 | EXTRACT_NUMBER_AND_INCR (count, p); | |
3831 | for (; count > 0; count--, p += 2 * 3) /* XXX */ | |
3832 | { | |
3833 | /* Extract the start of each range. */ | |
3834 | EXTRACT_CHARACTER (c, p); | |
3835 | j = CHAR_CHARSET (c); | |
3836 | fastmap[CHARSET_LEADING_CODE_BASE (j)] = 1; | |
3837 | } | |
3838 | } | |
b18215fc RS |
3839 | break; |
3840 | ||
1fb352e0 SM |
3841 | case syntaxspec: |
3842 | case notsyntaxspec: | |
3843 | if (!fastmap) break; | |
3844 | #ifndef emacs | |
3845 | not = (re_opcode_t)p[-1] == notsyntaxspec; | |
3846 | k = *p++; | |
3847 | for (j = 0; j < (1 << BYTEWIDTH); j++) | |
990b2375 | 3848 | if ((SYNTAX (j) == (enum syntaxcode) k) ^ not) |
b18215fc | 3849 | fastmap[j] = 1; |
b18215fc | 3850 | break; |
1fb352e0 | 3851 | #else /* emacs */ |
b18215fc RS |
3852 | /* This match depends on text properties. These end with |
3853 | aborting optimizations. */ | |
01618498 | 3854 | return -1; |
b18215fc RS |
3855 | |
3856 | case categoryspec: | |
b18215fc | 3857 | case notcategoryspec: |
1fb352e0 SM |
3858 | if (!fastmap) break; |
3859 | not = (re_opcode_t)p[-1] == notcategoryspec; | |
b18215fc | 3860 | k = *p++; |
1fb352e0 SM |
3861 | for (j = 0; j < (1 << BYTEWIDTH); j++) |
3862 | if ((CHAR_HAS_CATEGORY (j, k)) ^ not) | |
b18215fc RS |
3863 | fastmap[j] = 1; |
3864 | ||
1fb352e0 | 3865 | if (multibyte) |
b18215fc | 3866 | /* Any character set can possibly contain a character |
1fb352e0 | 3867 | whose category is K (or not). */ |
b18215fc RS |
3868 | goto set_fastmap_for_multibyte_characters; |
3869 | break; | |
3870 | ||
fa9a63c5 | 3871 | /* All cases after this match the empty string. These end with |
25fe55af | 3872 | `continue'. */ |
fa9a63c5 | 3873 | |
fa9a63c5 RM |
3874 | case before_dot: |
3875 | case at_dot: | |
3876 | case after_dot: | |
1fb352e0 | 3877 | #endif /* !emacs */ |
25fe55af RS |
3878 | case no_op: |
3879 | case begline: | |
3880 | case endline: | |
fa9a63c5 RM |
3881 | case begbuf: |
3882 | case endbuf: | |
3883 | case wordbound: | |
3884 | case notwordbound: | |
3885 | case wordbeg: | |
3886 | case wordend: | |
25fe55af | 3887 | continue; |
fa9a63c5 RM |
3888 | |
3889 | ||
fa9a63c5 | 3890 | case jump: |
25fe55af | 3891 | EXTRACT_NUMBER_AND_INCR (j, p); |
505bde11 SM |
3892 | if (j < 0) |
3893 | /* Backward jumps can only go back to code that we've already | |
3894 | visited. `re_compile' should make sure this is true. */ | |
3895 | break; | |
25fe55af | 3896 | p += j; |
505bde11 SM |
3897 | switch (SWITCH_ENUM_CAST ((re_opcode_t) *p)) |
3898 | { | |
3899 | case on_failure_jump: | |
3900 | case on_failure_keep_string_jump: | |
505bde11 | 3901 | case on_failure_jump_loop: |
0683b6fa | 3902 | case on_failure_jump_nastyloop: |
505bde11 SM |
3903 | case on_failure_jump_smart: |
3904 | p++; | |
3905 | break; | |
3906 | default: | |
3907 | continue; | |
3908 | }; | |
3909 | /* Keep `p1' to allow the `on_failure_jump' we are jumping to | |
3910 | to jump back to "just after here". */ | |
3911 | /* Fallthrough */ | |
fa9a63c5 | 3912 | |
25fe55af RS |
3913 | case on_failure_jump: |
3914 | case on_failure_keep_string_jump: | |
0683b6fa | 3915 | case on_failure_jump_nastyloop: |
505bde11 SM |
3916 | case on_failure_jump_loop: |
3917 | case on_failure_jump_smart: | |
25fe55af | 3918 | EXTRACT_NUMBER_AND_INCR (j, p); |
505bde11 | 3919 | if (p + j <= p1) |
ed0767d8 | 3920 | ; /* Backward jump to be ignored. */ |
01618498 SM |
3921 | else |
3922 | { /* We have to look down both arms. | |
3923 | We first go down the "straight" path so as to minimize | |
3924 | stack usage when going through alternatives. */ | |
3925 | int r = analyse_first (p, pend, fastmap, multibyte); | |
3926 | if (r) return r; | |
3927 | p += j; | |
3928 | } | |
25fe55af | 3929 | continue; |
fa9a63c5 RM |
3930 | |
3931 | ||
ed0767d8 SM |
3932 | case jump_n: |
3933 | /* This code simply does not properly handle forward jump_n. */ | |
3934 | DEBUG_STATEMENT (EXTRACT_NUMBER (j, p); assert (j < 0)); | |
3935 | p += 4; | |
3936 | /* jump_n can either jump or fall through. The (backward) jump | |
3937 | case has already been handled, so we only need to look at the | |
3938 | fallthrough case. */ | |
3939 | continue; | |
3940 | ||
fa9a63c5 | 3941 | case succeed_n: |
ed0767d8 SM |
3942 | /* If N == 0, it should be an on_failure_jump_loop instead. */ |
3943 | DEBUG_STATEMENT (EXTRACT_NUMBER (j, p + 2); assert (j > 0)); | |
3944 | p += 4; | |
3945 | /* We only care about one iteration of the loop, so we don't | |
3946 | need to consider the case where this behaves like an | |
3947 | on_failure_jump. */ | |
25fe55af | 3948 | continue; |
fa9a63c5 RM |
3949 | |
3950 | ||
3951 | case set_number_at: | |
25fe55af RS |
3952 | p += 4; |
3953 | continue; | |
fa9a63c5 RM |
3954 | |
3955 | ||
3956 | case start_memory: | |
25fe55af | 3957 | case stop_memory: |
505bde11 | 3958 | p += 1; |
fa9a63c5 RM |
3959 | continue; |
3960 | ||
3961 | ||
3962 | default: | |
25fe55af RS |
3963 | abort (); /* We have listed all the cases. */ |
3964 | } /* switch *p++ */ | |
fa9a63c5 RM |
3965 | |
3966 | /* Getting here means we have found the possible starting | |
25fe55af | 3967 | characters for one path of the pattern -- and that the empty |
01618498 SM |
3968 | string does not match. We need not follow this path further. */ |
3969 | return 0; | |
fa9a63c5 RM |
3970 | } /* while p */ |
3971 | ||
01618498 SM |
3972 | /* We reached the end without matching anything. */ |
3973 | return 1; | |
3974 | ||
f6a3f532 SM |
3975 | } /* analyse_first */ |
3976 | \f | |
3977 | /* re_compile_fastmap computes a ``fastmap'' for the compiled pattern in | |
3978 | BUFP. A fastmap records which of the (1 << BYTEWIDTH) possible | |
3979 | characters can start a string that matches the pattern. This fastmap | |
3980 | is used by re_search to skip quickly over impossible starting points. | |
3981 | ||
3982 | Character codes above (1 << BYTEWIDTH) are not represented in the | |
3983 | fastmap, but the leading codes are represented. Thus, the fastmap | |
3984 | indicates which character sets could start a match. | |
3985 | ||
3986 | The caller must supply the address of a (1 << BYTEWIDTH)-byte data | |
3987 | area as BUFP->fastmap. | |
3988 | ||
3989 | We set the `fastmap', `fastmap_accurate', and `can_be_null' fields in | |
3990 | the pattern buffer. | |
3991 | ||
3992 | Returns 0 if we succeed, -2 if an internal error. */ | |
3993 | ||
3994 | int | |
3995 | re_compile_fastmap (bufp) | |
3996 | struct re_pattern_buffer *bufp; | |
3997 | { | |
3998 | char *fastmap = bufp->fastmap; | |
3999 | int analysis; | |
4000 | ||
4001 | assert (fastmap && bufp->buffer); | |
4002 | ||
4003 | bzero (fastmap, 1 << BYTEWIDTH); /* Assume nothing's valid. */ | |
4004 | bufp->fastmap_accurate = 1; /* It will be when we're done. */ | |
4005 | ||
4006 | analysis = analyse_first (bufp->buffer, bufp->buffer + bufp->used, | |
2d1675e4 | 4007 | fastmap, RE_MULTIBYTE_P (bufp)); |
c0f9ea08 | 4008 | bufp->can_be_null = (analysis != 0); |
fa9a63c5 RM |
4009 | return 0; |
4010 | } /* re_compile_fastmap */ | |
4011 | \f | |
4012 | /* Set REGS to hold NUM_REGS registers, storing them in STARTS and | |
4013 | ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use | |
4014 | this memory for recording register information. STARTS and ENDS | |
4015 | must be allocated using the malloc library routine, and must each | |
4016 | be at least NUM_REGS * sizeof (regoff_t) bytes long. | |
4017 | ||
4018 | If NUM_REGS == 0, then subsequent matches should allocate their own | |
4019 | register data. | |
4020 | ||
4021 | Unless this function is called, the first search or match using | |
4022 | PATTERN_BUFFER will allocate its own register data, without | |
4023 | freeing the old data. */ | |
4024 | ||
4025 | void | |
4026 | re_set_registers (bufp, regs, num_regs, starts, ends) | |
4027 | struct re_pattern_buffer *bufp; | |
4028 | struct re_registers *regs; | |
4029 | unsigned num_regs; | |
4030 | regoff_t *starts, *ends; | |
4031 | { | |
4032 | if (num_regs) | |
4033 | { | |
4034 | bufp->regs_allocated = REGS_REALLOCATE; | |
4035 | regs->num_regs = num_regs; | |
4036 | regs->start = starts; | |
4037 | regs->end = ends; | |
4038 | } | |
4039 | else | |
4040 | { | |
4041 | bufp->regs_allocated = REGS_UNALLOCATED; | |
4042 | regs->num_regs = 0; | |
4043 | regs->start = regs->end = (regoff_t *) 0; | |
4044 | } | |
4045 | } | |
c0f9ea08 | 4046 | WEAK_ALIAS (__re_set_registers, re_set_registers) |
fa9a63c5 | 4047 | \f |
25fe55af | 4048 | /* Searching routines. */ |
fa9a63c5 RM |
4049 | |
4050 | /* Like re_search_2, below, but only one string is specified, and | |
4051 | doesn't let you say where to stop matching. */ | |
4052 | ||
4053 | int | |
4054 | re_search (bufp, string, size, startpos, range, regs) | |
4055 | struct re_pattern_buffer *bufp; | |
4056 | const char *string; | |
4057 | int size, startpos, range; | |
4058 | struct re_registers *regs; | |
4059 | { | |
5e69f11e | 4060 | return re_search_2 (bufp, NULL, 0, string, size, startpos, range, |
fa9a63c5 RM |
4061 | regs, size); |
4062 | } | |
c0f9ea08 | 4063 | WEAK_ALIAS (__re_search, re_search) |
fa9a63c5 | 4064 | |
b18215fc RS |
4065 | /* End address of virtual concatenation of string. */ |
4066 | #define STOP_ADDR_VSTRING(P) \ | |
4067 | (((P) >= size1 ? string2 + size2 : string1 + size1)) | |
4068 | ||
4069 | /* Address of POS in the concatenation of virtual string. */ | |
4070 | #define POS_ADDR_VSTRING(POS) \ | |
4071 | (((POS) >= size1 ? string2 - size1 : string1) + (POS)) | |
fa9a63c5 RM |
4072 | |
4073 | /* Using the compiled pattern in BUFP->buffer, first tries to match the | |
4074 | virtual concatenation of STRING1 and STRING2, starting first at index | |
4075 | STARTPOS, then at STARTPOS + 1, and so on. | |
5e69f11e | 4076 | |
fa9a63c5 | 4077 | STRING1 and STRING2 have length SIZE1 and SIZE2, respectively. |
5e69f11e | 4078 | |
fa9a63c5 RM |
4079 | RANGE is how far to scan while trying to match. RANGE = 0 means try |
4080 | only at STARTPOS; in general, the last start tried is STARTPOS + | |
4081 | RANGE. | |
5e69f11e | 4082 | |
fa9a63c5 RM |
4083 | In REGS, return the indices of the virtual concatenation of STRING1 |
4084 | and STRING2 that matched the entire BUFP->buffer and its contained | |
4085 | subexpressions. | |
5e69f11e | 4086 | |
fa9a63c5 RM |
4087 | Do not consider matching one past the index STOP in the virtual |
4088 | concatenation of STRING1 and STRING2. | |
4089 | ||
4090 | We return either the position in the strings at which the match was | |
4091 | found, -1 if no match, or -2 if error (such as failure | |
4092 | stack overflow). */ | |
4093 | ||
4094 | int | |
66f0296e | 4095 | re_search_2 (bufp, str1, size1, str2, size2, startpos, range, regs, stop) |
fa9a63c5 | 4096 | struct re_pattern_buffer *bufp; |
66f0296e | 4097 | const char *str1, *str2; |
fa9a63c5 RM |
4098 | int size1, size2; |
4099 | int startpos; | |
4100 | int range; | |
4101 | struct re_registers *regs; | |
4102 | int stop; | |
4103 | { | |
4104 | int val; | |
66f0296e SM |
4105 | re_char *string1 = (re_char*) str1; |
4106 | re_char *string2 = (re_char*) str2; | |
fa9a63c5 | 4107 | register char *fastmap = bufp->fastmap; |
6676cb1c | 4108 | register RE_TRANSLATE_TYPE translate = bufp->translate; |
fa9a63c5 RM |
4109 | int total_size = size1 + size2; |
4110 | int endpos = startpos + range; | |
c0f9ea08 | 4111 | boolean anchored_start; |
fa9a63c5 | 4112 | |
25fe55af | 4113 | /* Nonzero if we have to concern multibyte character. */ |
2d1675e4 | 4114 | const boolean multibyte = RE_MULTIBYTE_P (bufp); |
b18215fc | 4115 | |
fa9a63c5 RM |
4116 | /* Check for out-of-range STARTPOS. */ |
4117 | if (startpos < 0 || startpos > total_size) | |
4118 | return -1; | |
5e69f11e | 4119 | |
fa9a63c5 | 4120 | /* Fix up RANGE if it might eventually take us outside |
34597fa9 | 4121 | the virtual concatenation of STRING1 and STRING2. |
5e69f11e | 4122 | Make sure we won't move STARTPOS below 0 or above TOTAL_SIZE. */ |
34597fa9 RS |
4123 | if (endpos < 0) |
4124 | range = 0 - startpos; | |
fa9a63c5 RM |
4125 | else if (endpos > total_size) |
4126 | range = total_size - startpos; | |
4127 | ||
4128 | /* If the search isn't to be a backwards one, don't waste time in a | |
7b140fd7 | 4129 | search for a pattern anchored at beginning of buffer. */ |
fa9a63c5 RM |
4130 | if (bufp->used > 0 && (re_opcode_t) bufp->buffer[0] == begbuf && range > 0) |
4131 | { | |
4132 | if (startpos > 0) | |
4133 | return -1; | |
4134 | else | |
7b140fd7 | 4135 | range = 0; |
fa9a63c5 RM |
4136 | } |
4137 | ||
ae4788a8 RS |
4138 | #ifdef emacs |
4139 | /* In a forward search for something that starts with \=. | |
4140 | don't keep searching past point. */ | |
4141 | if (bufp->used > 0 && (re_opcode_t) bufp->buffer[0] == at_dot && range > 0) | |
4142 | { | |
7b140fd7 RS |
4143 | range = PT_BYTE - BEGV_BYTE - startpos; |
4144 | if (range < 0) | |
ae4788a8 RS |
4145 | return -1; |
4146 | } | |
4147 | #endif /* emacs */ | |
4148 | ||
fa9a63c5 RM |
4149 | /* Update the fastmap now if not correct already. */ |
4150 | if (fastmap && !bufp->fastmap_accurate) | |
01618498 | 4151 | re_compile_fastmap (bufp); |
5e69f11e | 4152 | |
c8499ba5 | 4153 | /* See whether the pattern is anchored. */ |
c0f9ea08 | 4154 | anchored_start = (bufp->buffer[0] == begline); |
c8499ba5 | 4155 | |
b18215fc | 4156 | #ifdef emacs |
cc9b4df2 KH |
4157 | gl_state.object = re_match_object; |
4158 | { | |
99633e97 | 4159 | int charpos = SYNTAX_TABLE_BYTE_TO_CHAR (POS_AS_IN_BUFFER (startpos)); |
cc9b4df2 KH |
4160 | |
4161 | SETUP_SYNTAX_TABLE_FOR_OBJECT (re_match_object, charpos, 1); | |
4162 | } | |
b18215fc RS |
4163 | #endif |
4164 | ||
fa9a63c5 RM |
4165 | /* Loop through the string, looking for a place to start matching. */ |
4166 | for (;;) | |
5e69f11e | 4167 | { |
c8499ba5 RS |
4168 | /* If the pattern is anchored, |
4169 | skip quickly past places we cannot match. | |
4170 | We don't bother to treat startpos == 0 specially | |
4171 | because that case doesn't repeat. */ | |
4172 | if (anchored_start && startpos > 0) | |
4173 | { | |
c0f9ea08 SM |
4174 | if (! ((startpos <= size1 ? string1[startpos - 1] |
4175 | : string2[startpos - size1 - 1]) | |
4176 | == '\n')) | |
c8499ba5 RS |
4177 | goto advance; |
4178 | } | |
4179 | ||
fa9a63c5 | 4180 | /* If a fastmap is supplied, skip quickly over characters that |
25fe55af RS |
4181 | cannot be the start of a match. If the pattern can match the |
4182 | null string, however, we don't need to skip characters; we want | |
4183 | the first null string. */ | |
fa9a63c5 RM |
4184 | if (fastmap && startpos < total_size && !bufp->can_be_null) |
4185 | { | |
66f0296e | 4186 | register re_char *d; |
01618498 | 4187 | register re_wchar_t buf_ch; |
e934739e RS |
4188 | |
4189 | d = POS_ADDR_VSTRING (startpos); | |
4190 | ||
25fe55af | 4191 | if (range > 0) /* Searching forwards. */ |
fa9a63c5 | 4192 | { |
fa9a63c5 RM |
4193 | register int lim = 0; |
4194 | int irange = range; | |
4195 | ||
25fe55af RS |
4196 | if (startpos < size1 && startpos + range >= size1) |
4197 | lim = range - (size1 - startpos); | |
fa9a63c5 | 4198 | |
25fe55af RS |
4199 | /* Written out as an if-else to avoid testing `translate' |
4200 | inside the loop. */ | |
28ae27ae AS |
4201 | if (RE_TRANSLATE_P (translate)) |
4202 | { | |
e934739e RS |
4203 | if (multibyte) |
4204 | while (range > lim) | |
4205 | { | |
4206 | int buf_charlen; | |
4207 | ||
4208 | buf_ch = STRING_CHAR_AND_LENGTH (d, range - lim, | |
4209 | buf_charlen); | |
4210 | ||
4211 | buf_ch = RE_TRANSLATE (translate, buf_ch); | |
4212 | if (buf_ch >= 0400 | |
4213 | || fastmap[buf_ch]) | |
4214 | break; | |
4215 | ||
4216 | range -= buf_charlen; | |
4217 | d += buf_charlen; | |
4218 | } | |
4219 | else | |
4220 | while (range > lim | |
66f0296e | 4221 | && !fastmap[RE_TRANSLATE (translate, *d)]) |
33c46939 RS |
4222 | { |
4223 | d++; | |
4224 | range--; | |
4225 | } | |
e934739e | 4226 | } |
fa9a63c5 | 4227 | else |
66f0296e | 4228 | while (range > lim && !fastmap[*d]) |
33c46939 RS |
4229 | { |
4230 | d++; | |
4231 | range--; | |
4232 | } | |
fa9a63c5 RM |
4233 | |
4234 | startpos += irange - range; | |
4235 | } | |
25fe55af | 4236 | else /* Searching backwards. */ |
fa9a63c5 | 4237 | { |
2d1675e4 SM |
4238 | int room = (startpos >= size1 |
4239 | ? size2 + size1 - startpos | |
4240 | : size1 - startpos); | |
4241 | buf_ch = RE_STRING_CHAR (d, room); | |
4242 | buf_ch = TRANSLATE (buf_ch); | |
fa9a63c5 | 4243 | |
e934739e RS |
4244 | if (! (buf_ch >= 0400 |
4245 | || fastmap[buf_ch])) | |
fa9a63c5 RM |
4246 | goto advance; |
4247 | } | |
4248 | } | |
4249 | ||
4250 | /* If can't match the null string, and that's all we have left, fail. */ | |
4251 | if (range >= 0 && startpos == total_size && fastmap | |
25fe55af | 4252 | && !bufp->can_be_null) |
fa9a63c5 RM |
4253 | return -1; |
4254 | ||
4255 | val = re_match_2_internal (bufp, string1, size1, string2, size2, | |
4256 | startpos, regs, stop); | |
4257 | #ifndef REGEX_MALLOC | |
0b32bf0e | 4258 | # ifdef C_ALLOCA |
fa9a63c5 | 4259 | alloca (0); |
0b32bf0e | 4260 | # endif |
fa9a63c5 RM |
4261 | #endif |
4262 | ||
4263 | if (val >= 0) | |
4264 | return startpos; | |
5e69f11e | 4265 | |
fa9a63c5 RM |
4266 | if (val == -2) |
4267 | return -2; | |
4268 | ||
4269 | advance: | |
5e69f11e | 4270 | if (!range) |
25fe55af | 4271 | break; |
5e69f11e | 4272 | else if (range > 0) |
25fe55af | 4273 | { |
b18215fc RS |
4274 | /* Update STARTPOS to the next character boundary. */ |
4275 | if (multibyte) | |
4276 | { | |
66f0296e SM |
4277 | re_char *p = POS_ADDR_VSTRING (startpos); |
4278 | re_char *pend = STOP_ADDR_VSTRING (startpos); | |
b18215fc RS |
4279 | int len = MULTIBYTE_FORM_LENGTH (p, pend - p); |
4280 | ||
4281 | range -= len; | |
4282 | if (range < 0) | |
4283 | break; | |
4284 | startpos += len; | |
4285 | } | |
4286 | else | |
4287 | { | |
b560c397 RS |
4288 | range--; |
4289 | startpos++; | |
4290 | } | |
e318085a | 4291 | } |
fa9a63c5 | 4292 | else |
25fe55af RS |
4293 | { |
4294 | range++; | |
4295 | startpos--; | |
b18215fc RS |
4296 | |
4297 | /* Update STARTPOS to the previous character boundary. */ | |
4298 | if (multibyte) | |
4299 | { | |
66f0296e | 4300 | re_char *p = POS_ADDR_VSTRING (startpos); |
b18215fc RS |
4301 | int len = 0; |
4302 | ||
4303 | /* Find the head of multibyte form. */ | |
5d967c7a | 4304 | while (!CHAR_HEAD_P (*p)) |
b18215fc RS |
4305 | p--, len++; |
4306 | ||
4307 | /* Adjust it. */ | |
4308 | #if 0 /* XXX */ | |
4309 | if (MULTIBYTE_FORM_LENGTH (p, len + 1) != (len + 1)) | |
4310 | ; | |
4311 | else | |
4312 | #endif | |
4313 | { | |
4314 | range += len; | |
4315 | if (range > 0) | |
4316 | break; | |
4317 | ||
4318 | startpos -= len; | |
4319 | } | |
4320 | } | |
25fe55af | 4321 | } |
fa9a63c5 RM |
4322 | } |
4323 | return -1; | |
4324 | } /* re_search_2 */ | |
c0f9ea08 | 4325 | WEAK_ALIAS (__re_search_2, re_search_2) |
fa9a63c5 RM |
4326 | \f |
4327 | /* Declarations and macros for re_match_2. */ | |
4328 | ||
2d1675e4 SM |
4329 | static int bcmp_translate _RE_ARGS((re_char *s1, re_char *s2, |
4330 | register int len, | |
4331 | RE_TRANSLATE_TYPE translate, | |
4332 | const int multibyte)); | |
fa9a63c5 RM |
4333 | |
4334 | /* This converts PTR, a pointer into one of the search strings `string1' | |
4335 | and `string2' into an offset from the beginning of that string. */ | |
4336 | #define POINTER_TO_OFFSET(ptr) \ | |
4337 | (FIRST_STRING_P (ptr) \ | |
4338 | ? ((regoff_t) ((ptr) - string1)) \ | |
4339 | : ((regoff_t) ((ptr) - string2 + size1))) | |
4340 | ||
fa9a63c5 | 4341 | /* Call before fetching a character with *d. This switches over to |
419d1c74 SM |
4342 | string2 if necessary. |
4343 | Check re_match_2_internal for a discussion of why end_match_2 might | |
4344 | not be within string2 (but be equal to end_match_1 instead). */ | |
fa9a63c5 | 4345 | #define PREFETCH() \ |
25fe55af | 4346 | while (d == dend) \ |
fa9a63c5 RM |
4347 | { \ |
4348 | /* End of string2 => fail. */ \ | |
25fe55af RS |
4349 | if (dend == end_match_2) \ |
4350 | goto fail; \ | |
4bb91c68 | 4351 | /* End of string1 => advance to string2. */ \ |
25fe55af | 4352 | d = string2; \ |
fa9a63c5 RM |
4353 | dend = end_match_2; \ |
4354 | } | |
4355 | ||
f1ad044f SM |
4356 | /* Call before fetching a char with *d if you already checked other limits. |
4357 | This is meant for use in lookahead operations like wordend, etc.. | |
4358 | where we might need to look at parts of the string that might be | |
4359 | outside of the LIMITs (i.e past `stop'). */ | |
4360 | #define PREFETCH_NOLIMIT() \ | |
4361 | if (d == end1) \ | |
4362 | { \ | |
4363 | d = string2; \ | |
4364 | dend = end_match_2; \ | |
4365 | } \ | |
fa9a63c5 RM |
4366 | |
4367 | /* Test if at very beginning or at very end of the virtual concatenation | |
25fe55af | 4368 | of `string1' and `string2'. If only one string, it's `string2'. */ |
fa9a63c5 | 4369 | #define AT_STRINGS_BEG(d) ((d) == (size1 ? string1 : string2) || !size2) |
5e69f11e | 4370 | #define AT_STRINGS_END(d) ((d) == end2) |
fa9a63c5 RM |
4371 | |
4372 | ||
4373 | /* Test if D points to a character which is word-constituent. We have | |
4374 | two special cases to check for: if past the end of string1, look at | |
4375 | the first character in string2; and if before the beginning of | |
4376 | string2, look at the last character in string1. */ | |
4377 | #define WORDCHAR_P(d) \ | |
4378 | (SYNTAX ((d) == end1 ? *string2 \ | |
25fe55af | 4379 | : (d) == string2 - 1 ? *(end1 - 1) : *(d)) \ |
fa9a63c5 RM |
4380 | == Sword) |
4381 | ||
9121ca40 | 4382 | /* Disabled due to a compiler bug -- see comment at case wordbound */ |
b18215fc RS |
4383 | |
4384 | /* The comment at case wordbound is following one, but we don't use | |
4385 | AT_WORD_BOUNDARY anymore to support multibyte form. | |
4386 | ||
4387 | The DEC Alpha C compiler 3.x generates incorrect code for the | |
25fe55af RS |
4388 | test WORDCHAR_P (d - 1) != WORDCHAR_P (d) in the expansion of |
4389 | AT_WORD_BOUNDARY, so this code is disabled. Expanding the | |
b18215fc RS |
4390 | macro and introducing temporary variables works around the bug. */ |
4391 | ||
9121ca40 | 4392 | #if 0 |
fa9a63c5 RM |
4393 | /* Test if the character before D and the one at D differ with respect |
4394 | to being word-constituent. */ | |
4395 | #define AT_WORD_BOUNDARY(d) \ | |
4396 | (AT_STRINGS_BEG (d) || AT_STRINGS_END (d) \ | |
4397 | || WORDCHAR_P (d - 1) != WORDCHAR_P (d)) | |
9121ca40 | 4398 | #endif |
fa9a63c5 RM |
4399 | |
4400 | /* Free everything we malloc. */ | |
4401 | #ifdef MATCH_MAY_ALLOCATE | |
0b32bf0e SM |
4402 | # define FREE_VAR(var) if (var) { REGEX_FREE (var); var = NULL; } else |
4403 | # define FREE_VARIABLES() \ | |
fa9a63c5 RM |
4404 | do { \ |
4405 | REGEX_FREE_STACK (fail_stack.stack); \ | |
4406 | FREE_VAR (regstart); \ | |
4407 | FREE_VAR (regend); \ | |
fa9a63c5 RM |
4408 | FREE_VAR (best_regstart); \ |
4409 | FREE_VAR (best_regend); \ | |
fa9a63c5 RM |
4410 | } while (0) |
4411 | #else | |
0b32bf0e | 4412 | # define FREE_VARIABLES() ((void)0) /* Do nothing! But inhibit gcc warning. */ |
fa9a63c5 RM |
4413 | #endif /* not MATCH_MAY_ALLOCATE */ |
4414 | ||
505bde11 SM |
4415 | \f |
4416 | /* Optimization routines. */ | |
4417 | ||
4e8a9132 SM |
4418 | /* If the operation is a match against one or more chars, |
4419 | return a pointer to the next operation, else return NULL. */ | |
01618498 | 4420 | static re_char * |
4e8a9132 | 4421 | skip_one_char (p) |
01618498 | 4422 | re_char *p; |
4e8a9132 SM |
4423 | { |
4424 | switch (SWITCH_ENUM_CAST (*p++)) | |
4425 | { | |
4426 | case anychar: | |
4427 | break; | |
4428 | ||
4429 | case exactn: | |
4430 | p += *p + 1; | |
4431 | break; | |
4432 | ||
4433 | case charset_not: | |
4434 | case charset: | |
4435 | if (CHARSET_RANGE_TABLE_EXISTS_P (p - 1)) | |
4436 | { | |
4437 | int mcnt; | |
4438 | p = CHARSET_RANGE_TABLE (p - 1); | |
4439 | EXTRACT_NUMBER_AND_INCR (mcnt, p); | |
4440 | p = CHARSET_RANGE_TABLE_END (p, mcnt); | |
4441 | } | |
4442 | else | |
4443 | p += 1 + CHARSET_BITMAP_SIZE (p - 1); | |
4444 | break; | |
4445 | ||
4e8a9132 SM |
4446 | case syntaxspec: |
4447 | case notsyntaxspec: | |
1fb352e0 | 4448 | #ifdef emacs |
4e8a9132 SM |
4449 | case categoryspec: |
4450 | case notcategoryspec: | |
4451 | #endif /* emacs */ | |
4452 | p++; | |
4453 | break; | |
4454 | ||
4455 | default: | |
4456 | p = NULL; | |
4457 | } | |
4458 | return p; | |
4459 | } | |
4460 | ||
4461 | ||
505bde11 SM |
4462 | /* Jump over non-matching operations. */ |
4463 | static unsigned char * | |
4e8a9132 | 4464 | skip_noops (p, pend) |
505bde11 | 4465 | unsigned char *p, *pend; |
505bde11 SM |
4466 | { |
4467 | int mcnt; | |
4468 | while (p < pend) | |
4469 | { | |
4470 | switch (SWITCH_ENUM_CAST ((re_opcode_t) *p)) | |
4471 | { | |
4472 | case start_memory: | |
505bde11 SM |
4473 | case stop_memory: |
4474 | p += 2; break; | |
4475 | case no_op: | |
4476 | p += 1; break; | |
4477 | case jump: | |
4478 | p += 1; | |
4479 | EXTRACT_NUMBER_AND_INCR (mcnt, p); | |
4480 | p += mcnt; | |
4481 | break; | |
4482 | default: | |
4483 | return p; | |
4484 | } | |
4485 | } | |
4486 | assert (p == pend); | |
4487 | return p; | |
4488 | } | |
4489 | ||
4490 | /* Non-zero if "p1 matches something" implies "p2 fails". */ | |
4491 | static int | |
4492 | mutually_exclusive_p (bufp, p1, p2) | |
4493 | struct re_pattern_buffer *bufp; | |
4494 | unsigned char *p1, *p2; | |
4495 | { | |
4e8a9132 | 4496 | re_opcode_t op2; |
2d1675e4 | 4497 | const boolean multibyte = RE_MULTIBYTE_P (bufp); |
505bde11 SM |
4498 | unsigned char *pend = bufp->buffer + bufp->used; |
4499 | ||
4e8a9132 | 4500 | assert (p1 >= bufp->buffer && p1 < pend |
505bde11 SM |
4501 | && p2 >= bufp->buffer && p2 <= pend); |
4502 | ||
4503 | /* Skip over open/close-group commands. | |
4504 | If what follows this loop is a ...+ construct, | |
4505 | look at what begins its body, since we will have to | |
4506 | match at least one of that. */ | |
4e8a9132 SM |
4507 | p2 = skip_noops (p2, pend); |
4508 | /* The same skip can be done for p1, except that this function | |
4509 | is only used in the case where p1 is a simple match operator. */ | |
4510 | /* p1 = skip_noops (p1, pend); */ | |
4511 | ||
4512 | assert (p1 >= bufp->buffer && p1 < pend | |
4513 | && p2 >= bufp->buffer && p2 <= pend); | |
4514 | ||
4515 | op2 = p2 == pend ? succeed : *p2; | |
4516 | ||
4517 | switch (SWITCH_ENUM_CAST (op2)) | |
505bde11 | 4518 | { |
4e8a9132 SM |
4519 | case succeed: |
4520 | case endbuf: | |
4521 | /* If we're at the end of the pattern, we can change. */ | |
4522 | if (skip_one_char (p1)) | |
505bde11 | 4523 | { |
505bde11 SM |
4524 | DEBUG_PRINT1 (" End of pattern: fast loop.\n"); |
4525 | return 1; | |
505bde11 | 4526 | } |
4e8a9132 SM |
4527 | break; |
4528 | ||
4529 | case endline: | |
4e8a9132 SM |
4530 | case exactn: |
4531 | { | |
01618498 | 4532 | register re_wchar_t c |
4e8a9132 | 4533 | = (re_opcode_t) *p2 == endline ? '\n' |
411e4203 | 4534 | : RE_STRING_CHAR (p2 + 2, pend - p2 - 2); |
505bde11 | 4535 | |
4e8a9132 SM |
4536 | if ((re_opcode_t) *p1 == exactn) |
4537 | { | |
4538 | if (c != RE_STRING_CHAR (p1 + 2, pend - p1 - 2)) | |
4539 | { | |
4540 | DEBUG_PRINT3 (" '%c' != '%c' => fast loop.\n", c, p1[2]); | |
4541 | return 1; | |
4542 | } | |
4543 | } | |
505bde11 | 4544 | |
4e8a9132 SM |
4545 | else if ((re_opcode_t) *p1 == charset |
4546 | || (re_opcode_t) *p1 == charset_not) | |
4547 | { | |
4548 | int not = (re_opcode_t) *p1 == charset_not; | |
505bde11 | 4549 | |
4e8a9132 SM |
4550 | /* Test if C is listed in charset (or charset_not) |
4551 | at `p1'. */ | |
4552 | if (SINGLE_BYTE_CHAR_P (c)) | |
4553 | { | |
4554 | if (c < CHARSET_BITMAP_SIZE (p1) * BYTEWIDTH | |
4555 | && p1[2 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH))) | |
4556 | not = !not; | |
4557 | } | |
4558 | else if (CHARSET_RANGE_TABLE_EXISTS_P (p1)) | |
4559 | CHARSET_LOOKUP_RANGE_TABLE (not, c, p1); | |
505bde11 | 4560 | |
4e8a9132 SM |
4561 | /* `not' is equal to 1 if c would match, which means |
4562 | that we can't change to pop_failure_jump. */ | |
4563 | if (!not) | |
4564 | { | |
4565 | DEBUG_PRINT1 (" No match => fast loop.\n"); | |
4566 | return 1; | |
4567 | } | |
4568 | } | |
4569 | else if ((re_opcode_t) *p1 == anychar | |
4570 | && c == '\n') | |
4571 | { | |
4572 | DEBUG_PRINT1 (" . != \\n => fast loop.\n"); | |
4573 | return 1; | |
4574 | } | |
4575 | } | |
4576 | break; | |
505bde11 | 4577 | |
4e8a9132 | 4578 | case charset: |
4e8a9132 SM |
4579 | { |
4580 | if ((re_opcode_t) *p1 == exactn) | |
4581 | /* Reuse the code above. */ | |
4582 | return mutually_exclusive_p (bufp, p2, p1); | |
505bde11 | 4583 | |
505bde11 SM |
4584 | /* It is hard to list up all the character in charset |
4585 | P2 if it includes multibyte character. Give up in | |
4586 | such case. */ | |
4587 | else if (!multibyte || !CHARSET_RANGE_TABLE_EXISTS_P (p2)) | |
4588 | { | |
4589 | /* Now, we are sure that P2 has no range table. | |
4590 | So, for the size of bitmap in P2, `p2[1]' is | |
4591 | enough. But P1 may have range table, so the | |
4592 | size of bitmap table of P1 is extracted by | |
4593 | using macro `CHARSET_BITMAP_SIZE'. | |
4594 | ||
4595 | Since we know that all the character listed in | |
4596 | P2 is ASCII, it is enough to test only bitmap | |
4597 | table of P1. */ | |
4598 | ||
411e4203 | 4599 | if ((re_opcode_t) *p1 == charset) |
505bde11 SM |
4600 | { |
4601 | int idx; | |
4602 | /* We win if the charset inside the loop | |
4603 | has no overlap with the one after the loop. */ | |
4604 | for (idx = 0; | |
4605 | (idx < (int) p2[1] | |
4606 | && idx < CHARSET_BITMAP_SIZE (p1)); | |
4607 | idx++) | |
4608 | if ((p2[2 + idx] & p1[2 + idx]) != 0) | |
4609 | break; | |
4610 | ||
4611 | if (idx == p2[1] | |
4612 | || idx == CHARSET_BITMAP_SIZE (p1)) | |
4613 | { | |
4614 | DEBUG_PRINT1 (" No match => fast loop.\n"); | |
4615 | return 1; | |
4616 | } | |
4617 | } | |
411e4203 | 4618 | else if ((re_opcode_t) *p1 == charset_not) |
505bde11 SM |
4619 | { |
4620 | int idx; | |
4621 | /* We win if the charset_not inside the loop lists | |
4622 | every character listed in the charset after. */ | |
4623 | for (idx = 0; idx < (int) p2[1]; idx++) | |
4624 | if (! (p2[2 + idx] == 0 | |
4625 | || (idx < CHARSET_BITMAP_SIZE (p1) | |
4626 | && ((p2[2 + idx] & ~ p1[2 + idx]) == 0)))) | |
4627 | break; | |
4628 | ||
4e8a9132 SM |
4629 | if (idx == p2[1]) |
4630 | { | |
4631 | DEBUG_PRINT1 (" No match => fast loop.\n"); | |
4632 | return 1; | |
4633 | } | |
4634 | } | |
4635 | } | |
4636 | } | |
609b757a | 4637 | break; |
4e8a9132 | 4638 | |
411e4203 SM |
4639 | case charset_not: |
4640 | switch (SWITCH_ENUM_CAST (*p1)) | |
4641 | { | |
4642 | case exactn: | |
4643 | case charset: | |
4644 | /* Reuse the code above. */ | |
4645 | return mutually_exclusive_p (bufp, p2, p1); | |
4646 | case charset_not: | |
4647 | /* When we have two charset_not, it's very unlikely that | |
4648 | they don't overlap. The union of the two sets of excluded | |
4649 | chars should cover all possible chars, which, as a matter of | |
4650 | fact, is virtually impossible in multibyte buffers. */ | |
36595814 | 4651 | break; |
411e4203 SM |
4652 | } |
4653 | break; | |
4654 | ||
4e8a9132 SM |
4655 | case wordend: |
4656 | case notsyntaxspec: | |
4657 | return ((re_opcode_t) *p1 == syntaxspec | |
4658 | && p1[1] == (op2 == wordend ? Sword : p2[1])); | |
4659 | ||
4660 | case wordbeg: | |
4661 | case syntaxspec: | |
4662 | return ((re_opcode_t) *p1 == notsyntaxspec | |
4663 | && p1[1] == (op2 == wordend ? Sword : p2[1])); | |
4664 | ||
4665 | case wordbound: | |
4666 | return (((re_opcode_t) *p1 == notsyntaxspec | |
4667 | || (re_opcode_t) *p1 == syntaxspec) | |
4668 | && p1[1] == Sword); | |
4669 | ||
1fb352e0 | 4670 | #ifdef emacs |
4e8a9132 SM |
4671 | case categoryspec: |
4672 | return ((re_opcode_t) *p1 == notcategoryspec && p1[1] == p2[1]); | |
4673 | case notcategoryspec: | |
4674 | return ((re_opcode_t) *p1 == categoryspec && p1[1] == p2[1]); | |
4675 | #endif /* emacs */ | |
4676 | ||
4677 | default: | |
4678 | ; | |
505bde11 SM |
4679 | } |
4680 | ||
4681 | /* Safe default. */ | |
4682 | return 0; | |
4683 | } | |
4684 | ||
fa9a63c5 RM |
4685 | \f |
4686 | /* Matching routines. */ | |
4687 | ||
25fe55af | 4688 | #ifndef emacs /* Emacs never uses this. */ |
fa9a63c5 RM |
4689 | /* re_match is like re_match_2 except it takes only a single string. */ |
4690 | ||
4691 | int | |
4692 | re_match (bufp, string, size, pos, regs) | |
4693 | struct re_pattern_buffer *bufp; | |
4694 | const char *string; | |
4695 | int size, pos; | |
4696 | struct re_registers *regs; | |
4697 | { | |
4bb91c68 | 4698 | int result = re_match_2_internal (bufp, NULL, 0, (re_char*) string, size, |
fa9a63c5 | 4699 | pos, regs, size); |
0b32bf0e | 4700 | # if defined C_ALLOCA && !defined REGEX_MALLOC |
fa9a63c5 | 4701 | alloca (0); |
0b32bf0e | 4702 | # endif |
fa9a63c5 RM |
4703 | return result; |
4704 | } | |
c0f9ea08 | 4705 | WEAK_ALIAS (__re_match, re_match) |
fa9a63c5 RM |
4706 | #endif /* not emacs */ |
4707 | ||
b18215fc RS |
4708 | #ifdef emacs |
4709 | /* In Emacs, this is the string or buffer in which we | |
25fe55af | 4710 | are matching. It is used for looking up syntax properties. */ |
b18215fc RS |
4711 | Lisp_Object re_match_object; |
4712 | #endif | |
fa9a63c5 RM |
4713 | |
4714 | /* re_match_2 matches the compiled pattern in BUFP against the | |
4715 | the (virtual) concatenation of STRING1 and STRING2 (of length SIZE1 | |
4716 | and SIZE2, respectively). We start matching at POS, and stop | |
4717 | matching at STOP. | |
5e69f11e | 4718 | |
fa9a63c5 | 4719 | If REGS is non-null and the `no_sub' field of BUFP is nonzero, we |
25fe55af | 4720 | store offsets for the substring each group matched in REGS. See the |
fa9a63c5 RM |
4721 | documentation for exactly how many groups we fill. |
4722 | ||
4723 | We return -1 if no match, -2 if an internal error (such as the | |
25fe55af | 4724 | failure stack overflowing). Otherwise, we return the length of the |
fa9a63c5 RM |
4725 | matched substring. */ |
4726 | ||
4727 | int | |
4728 | re_match_2 (bufp, string1, size1, string2, size2, pos, regs, stop) | |
4729 | struct re_pattern_buffer *bufp; | |
4730 | const char *string1, *string2; | |
4731 | int size1, size2; | |
4732 | int pos; | |
4733 | struct re_registers *regs; | |
4734 | int stop; | |
4735 | { | |
b18215fc | 4736 | int result; |
25fe55af | 4737 | |
b18215fc | 4738 | #ifdef emacs |
cc9b4df2 KH |
4739 | int charpos; |
4740 | gl_state.object = re_match_object; | |
99633e97 | 4741 | charpos = SYNTAX_TABLE_BYTE_TO_CHAR (POS_AS_IN_BUFFER (pos)); |
cc9b4df2 | 4742 | SETUP_SYNTAX_TABLE_FOR_OBJECT (re_match_object, charpos, 1); |
b18215fc RS |
4743 | #endif |
4744 | ||
4bb91c68 SM |
4745 | result = re_match_2_internal (bufp, (re_char*) string1, size1, |
4746 | (re_char*) string2, size2, | |
cc9b4df2 | 4747 | pos, regs, stop); |
0b32bf0e | 4748 | #if defined C_ALLOCA && !defined REGEX_MALLOC |
fa9a63c5 | 4749 | alloca (0); |
a60198e5 | 4750 | #endif |
fa9a63c5 RM |
4751 | return result; |
4752 | } | |
c0f9ea08 | 4753 | WEAK_ALIAS (__re_match_2, re_match_2) |
fa9a63c5 RM |
4754 | |
4755 | /* This is a separate function so that we can force an alloca cleanup | |
25fe55af | 4756 | afterwards. */ |
fa9a63c5 RM |
4757 | static int |
4758 | re_match_2_internal (bufp, string1, size1, string2, size2, pos, regs, stop) | |
4759 | struct re_pattern_buffer *bufp; | |
66f0296e | 4760 | re_char *string1, *string2; |
fa9a63c5 RM |
4761 | int size1, size2; |
4762 | int pos; | |
4763 | struct re_registers *regs; | |
4764 | int stop; | |
4765 | { | |
4766 | /* General temporaries. */ | |
4767 | int mcnt; | |
01618498 | 4768 | size_t reg; |
66f0296e | 4769 | boolean not; |
fa9a63c5 RM |
4770 | |
4771 | /* Just past the end of the corresponding string. */ | |
66f0296e | 4772 | re_char *end1, *end2; |
fa9a63c5 RM |
4773 | |
4774 | /* Pointers into string1 and string2, just past the last characters in | |
25fe55af | 4775 | each to consider matching. */ |
66f0296e | 4776 | re_char *end_match_1, *end_match_2; |
fa9a63c5 RM |
4777 | |
4778 | /* Where we are in the data, and the end of the current string. */ | |
66f0296e | 4779 | re_char *d, *dend; |
5e69f11e | 4780 | |
99633e97 SM |
4781 | /* Used sometimes to remember where we were before starting matching |
4782 | an operator so that we can go back in case of failure. This "atomic" | |
4783 | behavior of matching opcodes is indispensable to the correctness | |
4784 | of the on_failure_keep_string_jump optimization. */ | |
4785 | re_char *dfail; | |
4786 | ||
fa9a63c5 | 4787 | /* Where we are in the pattern, and the end of the pattern. */ |
01618498 SM |
4788 | re_char *p = bufp->buffer; |
4789 | re_char *pend = p + bufp->used; | |
fa9a63c5 | 4790 | |
25fe55af | 4791 | /* We use this to map every character in the string. */ |
6676cb1c | 4792 | RE_TRANSLATE_TYPE translate = bufp->translate; |
fa9a63c5 | 4793 | |
25fe55af | 4794 | /* Nonzero if we have to concern multibyte character. */ |
2d1675e4 | 4795 | const boolean multibyte = RE_MULTIBYTE_P (bufp); |
b18215fc | 4796 | |
fa9a63c5 RM |
4797 | /* Failure point stack. Each place that can handle a failure further |
4798 | down the line pushes a failure point on this stack. It consists of | |
505bde11 | 4799 | regstart, and regend for all registers corresponding to |
fa9a63c5 RM |
4800 | the subexpressions we're currently inside, plus the number of such |
4801 | registers, and, finally, two char *'s. The first char * is where | |
4802 | to resume scanning the pattern; the second one is where to resume | |
505bde11 | 4803 | scanning the strings. */ |
25fe55af | 4804 | #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */ |
fa9a63c5 RM |
4805 | fail_stack_type fail_stack; |
4806 | #endif | |
4807 | #ifdef DEBUG | |
fa9a63c5 RM |
4808 | unsigned nfailure_points_pushed = 0, nfailure_points_popped = 0; |
4809 | #endif | |
4810 | ||
0b32bf0e | 4811 | #if defined REL_ALLOC && defined REGEX_MALLOC |
fa9a63c5 RM |
4812 | /* This holds the pointer to the failure stack, when |
4813 | it is allocated relocatably. */ | |
4814 | fail_stack_elt_t *failure_stack_ptr; | |
99633e97 | 4815 | #endif |
fa9a63c5 RM |
4816 | |
4817 | /* We fill all the registers internally, independent of what we | |
25fe55af | 4818 | return, for use in backreferences. The number here includes |
fa9a63c5 | 4819 | an element for register zero. */ |
4bb91c68 | 4820 | size_t num_regs = bufp->re_nsub + 1; |
5e69f11e | 4821 | |
fa9a63c5 RM |
4822 | /* Information on the contents of registers. These are pointers into |
4823 | the input strings; they record just what was matched (on this | |
4824 | attempt) by a subexpression part of the pattern, that is, the | |
4825 | regnum-th regstart pointer points to where in the pattern we began | |
4826 | matching and the regnum-th regend points to right after where we | |
4827 | stopped matching the regnum-th subexpression. (The zeroth register | |
4828 | keeps track of what the whole pattern matches.) */ | |
4829 | #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */ | |
66f0296e | 4830 | re_char **regstart, **regend; |
fa9a63c5 RM |
4831 | #endif |
4832 | ||
fa9a63c5 | 4833 | /* The following record the register info as found in the above |
5e69f11e | 4834 | variables when we find a match better than any we've seen before. |
fa9a63c5 RM |
4835 | This happens as we backtrack through the failure points, which in |
4836 | turn happens only if we have not yet matched the entire string. */ | |
4837 | unsigned best_regs_set = false; | |
4838 | #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */ | |
66f0296e | 4839 | re_char **best_regstart, **best_regend; |
fa9a63c5 | 4840 | #endif |
5e69f11e | 4841 | |
fa9a63c5 RM |
4842 | /* Logically, this is `best_regend[0]'. But we don't want to have to |
4843 | allocate space for that if we're not allocating space for anything | |
25fe55af | 4844 | else (see below). Also, we never need info about register 0 for |
fa9a63c5 RM |
4845 | any of the other register vectors, and it seems rather a kludge to |
4846 | treat `best_regend' differently than the rest. So we keep track of | |
4847 | the end of the best match so far in a separate variable. We | |
4848 | initialize this to NULL so that when we backtrack the first time | |
4849 | and need to test it, it's not garbage. */ | |
66f0296e | 4850 | re_char *match_end = NULL; |
fa9a63c5 | 4851 | |
fa9a63c5 RM |
4852 | #ifdef DEBUG |
4853 | /* Counts the total number of registers pushed. */ | |
5e69f11e | 4854 | unsigned num_regs_pushed = 0; |
fa9a63c5 RM |
4855 | #endif |
4856 | ||
4857 | DEBUG_PRINT1 ("\n\nEntering re_match_2.\n"); | |
5e69f11e | 4858 | |
fa9a63c5 | 4859 | INIT_FAIL_STACK (); |
5e69f11e | 4860 | |
fa9a63c5 RM |
4861 | #ifdef MATCH_MAY_ALLOCATE |
4862 | /* Do not bother to initialize all the register variables if there are | |
4863 | no groups in the pattern, as it takes a fair amount of time. If | |
4864 | there are groups, we include space for register 0 (the whole | |
4865 | pattern), even though we never use it, since it simplifies the | |
4866 | array indexing. We should fix this. */ | |
4867 | if (bufp->re_nsub) | |
4868 | { | |
66f0296e SM |
4869 | regstart = REGEX_TALLOC (num_regs, re_char *); |
4870 | regend = REGEX_TALLOC (num_regs, re_char *); | |
4871 | best_regstart = REGEX_TALLOC (num_regs, re_char *); | |
4872 | best_regend = REGEX_TALLOC (num_regs, re_char *); | |
fa9a63c5 | 4873 | |
505bde11 | 4874 | if (!(regstart && regend && best_regstart && best_regend)) |
25fe55af RS |
4875 | { |
4876 | FREE_VARIABLES (); | |
4877 | return -2; | |
4878 | } | |
fa9a63c5 RM |
4879 | } |
4880 | else | |
4881 | { | |
4882 | /* We must initialize all our variables to NULL, so that | |
25fe55af | 4883 | `FREE_VARIABLES' doesn't try to free them. */ |
505bde11 | 4884 | regstart = regend = best_regstart = best_regend = NULL; |
fa9a63c5 RM |
4885 | } |
4886 | #endif /* MATCH_MAY_ALLOCATE */ | |
4887 | ||
4888 | /* The starting position is bogus. */ | |
4889 | if (pos < 0 || pos > size1 + size2) | |
4890 | { | |
4891 | FREE_VARIABLES (); | |
4892 | return -1; | |
4893 | } | |
5e69f11e | 4894 | |
fa9a63c5 RM |
4895 | /* Initialize subexpression text positions to -1 to mark ones that no |
4896 | start_memory/stop_memory has been seen for. Also initialize the | |
4897 | register information struct. */ | |
01618498 SM |
4898 | for (reg = 1; reg < num_regs; reg++) |
4899 | regstart[reg] = regend[reg] = NULL; | |
99633e97 | 4900 | |
fa9a63c5 | 4901 | /* We move `string1' into `string2' if the latter's empty -- but not if |
25fe55af | 4902 | `string1' is null. */ |
fa9a63c5 RM |
4903 | if (size2 == 0 && string1 != NULL) |
4904 | { | |
4905 | string2 = string1; | |
4906 | size2 = size1; | |
4907 | string1 = 0; | |
4908 | size1 = 0; | |
4909 | } | |
4910 | end1 = string1 + size1; | |
4911 | end2 = string2 + size2; | |
4912 | ||
5e69f11e | 4913 | /* `p' scans through the pattern as `d' scans through the data. |
fa9a63c5 RM |
4914 | `dend' is the end of the input string that `d' points within. `d' |
4915 | is advanced into the following input string whenever necessary, but | |
4916 | this happens before fetching; therefore, at the beginning of the | |
4917 | loop, `d' can be pointing at the end of a string, but it cannot | |
4918 | equal `string2'. */ | |
419d1c74 | 4919 | if (pos >= size1) |
fa9a63c5 | 4920 | { |
419d1c74 SM |
4921 | /* Only match within string2. */ |
4922 | d = string2 + pos - size1; | |
4923 | dend = end_match_2 = string2 + stop - size1; | |
4924 | end_match_1 = end1; /* Just to give it a value. */ | |
fa9a63c5 RM |
4925 | } |
4926 | else | |
4927 | { | |
f1ad044f | 4928 | if (stop < size1) |
419d1c74 SM |
4929 | { |
4930 | /* Only match within string1. */ | |
4931 | end_match_1 = string1 + stop; | |
4932 | /* BEWARE! | |
4933 | When we reach end_match_1, PREFETCH normally switches to string2. | |
4934 | But in the present case, this means that just doing a PREFETCH | |
4935 | makes us jump from `stop' to `gap' within the string. | |
4936 | What we really want here is for the search to stop as | |
4937 | soon as we hit end_match_1. That's why we set end_match_2 | |
4938 | to end_match_1 (since PREFETCH fails as soon as we hit | |
4939 | end_match_2). */ | |
4940 | end_match_2 = end_match_1; | |
4941 | } | |
4942 | else | |
f1ad044f SM |
4943 | { /* It's important to use this code when stop == size so that |
4944 | moving `d' from end1 to string2 will not prevent the d == dend | |
4945 | check from catching the end of string. */ | |
419d1c74 SM |
4946 | end_match_1 = end1; |
4947 | end_match_2 = string2 + stop - size1; | |
4948 | } | |
4949 | d = string1 + pos; | |
4950 | dend = end_match_1; | |
fa9a63c5 RM |
4951 | } |
4952 | ||
4953 | DEBUG_PRINT1 ("The compiled pattern is: "); | |
4954 | DEBUG_PRINT_COMPILED_PATTERN (bufp, p, pend); | |
4955 | DEBUG_PRINT1 ("The string to match is: `"); | |
4956 | DEBUG_PRINT_DOUBLE_STRING (d, string1, size1, string2, size2); | |
4957 | DEBUG_PRINT1 ("'\n"); | |
5e69f11e | 4958 | |
25fe55af | 4959 | /* This loops over pattern commands. It exits by returning from the |
fa9a63c5 RM |
4960 | function if the match is complete, or it drops through if the match |
4961 | fails at this starting point in the input data. */ | |
4962 | for (;;) | |
4963 | { | |
505bde11 | 4964 | DEBUG_PRINT2 ("\n%p: ", p); |
fa9a63c5 RM |
4965 | |
4966 | if (p == pend) | |
4967 | { /* End of pattern means we might have succeeded. */ | |
25fe55af | 4968 | DEBUG_PRINT1 ("end of pattern ... "); |
5e69f11e | 4969 | |
fa9a63c5 | 4970 | /* If we haven't matched the entire string, and we want the |
25fe55af RS |
4971 | longest match, try backtracking. */ |
4972 | if (d != end_match_2) | |
fa9a63c5 RM |
4973 | { |
4974 | /* 1 if this match ends in the same string (string1 or string2) | |
4975 | as the best previous match. */ | |
5e69f11e | 4976 | boolean same_str_p = (FIRST_STRING_P (match_end) |
99633e97 | 4977 | == FIRST_STRING_P (d)); |
fa9a63c5 RM |
4978 | /* 1 if this match is the best seen so far. */ |
4979 | boolean best_match_p; | |
4980 | ||
4981 | /* AIX compiler got confused when this was combined | |
25fe55af | 4982 | with the previous declaration. */ |
fa9a63c5 RM |
4983 | if (same_str_p) |
4984 | best_match_p = d > match_end; | |
4985 | else | |
99633e97 | 4986 | best_match_p = !FIRST_STRING_P (d); |
fa9a63c5 | 4987 | |
25fe55af RS |
4988 | DEBUG_PRINT1 ("backtracking.\n"); |
4989 | ||
4990 | if (!FAIL_STACK_EMPTY ()) | |
4991 | { /* More failure points to try. */ | |
4992 | ||
4993 | /* If exceeds best match so far, save it. */ | |
4994 | if (!best_regs_set || best_match_p) | |
4995 | { | |
4996 | best_regs_set = true; | |
4997 | match_end = d; | |
4998 | ||
4999 | DEBUG_PRINT1 ("\nSAVING match as best so far.\n"); | |
5000 | ||
01618498 | 5001 | for (reg = 1; reg < num_regs; reg++) |
25fe55af | 5002 | { |
01618498 SM |
5003 | best_regstart[reg] = regstart[reg]; |
5004 | best_regend[reg] = regend[reg]; | |
25fe55af RS |
5005 | } |
5006 | } | |
5007 | goto fail; | |
5008 | } | |
5009 | ||
5010 | /* If no failure points, don't restore garbage. And if | |
5011 | last match is real best match, don't restore second | |
5012 | best one. */ | |
5013 | else if (best_regs_set && !best_match_p) | |
5014 | { | |
5015 | restore_best_regs: | |
5016 | /* Restore best match. It may happen that `dend == | |
5017 | end_match_1' while the restored d is in string2. | |
5018 | For example, the pattern `x.*y.*z' against the | |
5019 | strings `x-' and `y-z-', if the two strings are | |
5020 | not consecutive in memory. */ | |
5021 | DEBUG_PRINT1 ("Restoring best registers.\n"); | |
5022 | ||
5023 | d = match_end; | |
5024 | dend = ((d >= string1 && d <= end1) | |
5025 | ? end_match_1 : end_match_2); | |
fa9a63c5 | 5026 | |
01618498 | 5027 | for (reg = 1; reg < num_regs; reg++) |
fa9a63c5 | 5028 | { |
01618498 SM |
5029 | regstart[reg] = best_regstart[reg]; |
5030 | regend[reg] = best_regend[reg]; | |
fa9a63c5 | 5031 | } |
25fe55af RS |
5032 | } |
5033 | } /* d != end_match_2 */ | |
fa9a63c5 RM |
5034 | |
5035 | succeed_label: | |
25fe55af | 5036 | DEBUG_PRINT1 ("Accepting match.\n"); |
fa9a63c5 | 5037 | |
25fe55af RS |
5038 | /* If caller wants register contents data back, do it. */ |
5039 | if (regs && !bufp->no_sub) | |
fa9a63c5 | 5040 | { |
25fe55af RS |
5041 | /* Have the register data arrays been allocated? */ |
5042 | if (bufp->regs_allocated == REGS_UNALLOCATED) | |
5043 | { /* No. So allocate them with malloc. We need one | |
5044 | extra element beyond `num_regs' for the `-1' marker | |
5045 | GNU code uses. */ | |
5046 | regs->num_regs = MAX (RE_NREGS, num_regs + 1); | |
5047 | regs->start = TALLOC (regs->num_regs, regoff_t); | |
5048 | regs->end = TALLOC (regs->num_regs, regoff_t); | |
5049 | if (regs->start == NULL || regs->end == NULL) | |
fa9a63c5 RM |
5050 | { |
5051 | FREE_VARIABLES (); | |
5052 | return -2; | |
5053 | } | |
25fe55af RS |
5054 | bufp->regs_allocated = REGS_REALLOCATE; |
5055 | } | |
5056 | else if (bufp->regs_allocated == REGS_REALLOCATE) | |
5057 | { /* Yes. If we need more elements than were already | |
5058 | allocated, reallocate them. If we need fewer, just | |
5059 | leave it alone. */ | |
5060 | if (regs->num_regs < num_regs + 1) | |
5061 | { | |
5062 | regs->num_regs = num_regs + 1; | |
5063 | RETALLOC (regs->start, regs->num_regs, regoff_t); | |
5064 | RETALLOC (regs->end, regs->num_regs, regoff_t); | |
5065 | if (regs->start == NULL || regs->end == NULL) | |
fa9a63c5 RM |
5066 | { |
5067 | FREE_VARIABLES (); | |
5068 | return -2; | |
5069 | } | |
25fe55af RS |
5070 | } |
5071 | } | |
5072 | else | |
fa9a63c5 RM |
5073 | { |
5074 | /* These braces fend off a "empty body in an else-statement" | |
25fe55af | 5075 | warning under GCC when assert expands to nothing. */ |
fa9a63c5 RM |
5076 | assert (bufp->regs_allocated == REGS_FIXED); |
5077 | } | |
5078 | ||
25fe55af RS |
5079 | /* Convert the pointer data in `regstart' and `regend' to |
5080 | indices. Register zero has to be set differently, | |
5081 | since we haven't kept track of any info for it. */ | |
5082 | if (regs->num_regs > 0) | |
5083 | { | |
5084 | regs->start[0] = pos; | |
99633e97 | 5085 | regs->end[0] = POINTER_TO_OFFSET (d); |
25fe55af | 5086 | } |
5e69f11e | 5087 | |
25fe55af RS |
5088 | /* Go through the first `min (num_regs, regs->num_regs)' |
5089 | registers, since that is all we initialized. */ | |
01618498 | 5090 | for (reg = 1; reg < MIN (num_regs, regs->num_regs); reg++) |
fa9a63c5 | 5091 | { |
01618498 SM |
5092 | if (REG_UNSET (regstart[reg]) || REG_UNSET (regend[reg])) |
5093 | regs->start[reg] = regs->end[reg] = -1; | |
25fe55af RS |
5094 | else |
5095 | { | |
01618498 SM |
5096 | regs->start[reg] |
5097 | = (regoff_t) POINTER_TO_OFFSET (regstart[reg]); | |
5098 | regs->end[reg] | |
5099 | = (regoff_t) POINTER_TO_OFFSET (regend[reg]); | |
25fe55af | 5100 | } |
fa9a63c5 | 5101 | } |
5e69f11e | 5102 | |
25fe55af RS |
5103 | /* If the regs structure we return has more elements than |
5104 | were in the pattern, set the extra elements to -1. If | |
5105 | we (re)allocated the registers, this is the case, | |
5106 | because we always allocate enough to have at least one | |
5107 | -1 at the end. */ | |
01618498 SM |
5108 | for (reg = num_regs; reg < regs->num_regs; reg++) |
5109 | regs->start[reg] = regs->end[reg] = -1; | |
fa9a63c5 RM |
5110 | } /* regs && !bufp->no_sub */ |
5111 | ||
25fe55af RS |
5112 | DEBUG_PRINT4 ("%u failure points pushed, %u popped (%u remain).\n", |
5113 | nfailure_points_pushed, nfailure_points_popped, | |
5114 | nfailure_points_pushed - nfailure_points_popped); | |
5115 | DEBUG_PRINT2 ("%u registers pushed.\n", num_regs_pushed); | |
fa9a63c5 | 5116 | |
99633e97 | 5117 | mcnt = POINTER_TO_OFFSET (d) - pos; |
fa9a63c5 | 5118 | |
25fe55af | 5119 | DEBUG_PRINT2 ("Returning %d from re_match_2.\n", mcnt); |
fa9a63c5 | 5120 | |
25fe55af RS |
5121 | FREE_VARIABLES (); |
5122 | return mcnt; | |
5123 | } | |
fa9a63c5 | 5124 | |
25fe55af | 5125 | /* Otherwise match next pattern command. */ |
fa9a63c5 RM |
5126 | switch (SWITCH_ENUM_CAST ((re_opcode_t) *p++)) |
5127 | { | |
25fe55af RS |
5128 | /* Ignore these. Used to ignore the n of succeed_n's which |
5129 | currently have n == 0. */ | |
5130 | case no_op: | |
5131 | DEBUG_PRINT1 ("EXECUTING no_op.\n"); | |
5132 | break; | |
fa9a63c5 RM |
5133 | |
5134 | case succeed: | |
25fe55af | 5135 | DEBUG_PRINT1 ("EXECUTING succeed.\n"); |
fa9a63c5 RM |
5136 | goto succeed_label; |
5137 | ||
25fe55af RS |
5138 | /* Match the next n pattern characters exactly. The following |
5139 | byte in the pattern defines n, and the n bytes after that | |
5140 | are the characters to match. */ | |
fa9a63c5 RM |
5141 | case exactn: |
5142 | mcnt = *p++; | |
25fe55af | 5143 | DEBUG_PRINT2 ("EXECUTING exactn %d.\n", mcnt); |
fa9a63c5 | 5144 | |
99633e97 SM |
5145 | /* Remember the start point to rollback upon failure. */ |
5146 | dfail = d; | |
5147 | ||
25fe55af RS |
5148 | /* This is written out as an if-else so we don't waste time |
5149 | testing `translate' inside the loop. */ | |
28703c16 | 5150 | if (RE_TRANSLATE_P (translate)) |
fa9a63c5 | 5151 | { |
e934739e RS |
5152 | if (multibyte) |
5153 | do | |
5154 | { | |
5155 | int pat_charlen, buf_charlen; | |
e71b1971 | 5156 | unsigned int pat_ch, buf_ch; |
e934739e RS |
5157 | |
5158 | PREFETCH (); | |
5159 | pat_ch = STRING_CHAR_AND_LENGTH (p, pend - p, pat_charlen); | |
5160 | buf_ch = STRING_CHAR_AND_LENGTH (d, dend - d, buf_charlen); | |
5161 | ||
5162 | if (RE_TRANSLATE (translate, buf_ch) | |
5163 | != pat_ch) | |
99633e97 SM |
5164 | { |
5165 | d = dfail; | |
5166 | goto fail; | |
5167 | } | |
e934739e RS |
5168 | |
5169 | p += pat_charlen; | |
5170 | d += buf_charlen; | |
5171 | mcnt -= pat_charlen; | |
5172 | } | |
5173 | while (mcnt > 0); | |
5174 | else | |
e934739e RS |
5175 | do |
5176 | { | |
5177 | PREFETCH (); | |
66f0296e | 5178 | if (RE_TRANSLATE (translate, *d) != *p++) |
99633e97 SM |
5179 | { |
5180 | d = dfail; | |
5181 | goto fail; | |
5182 | } | |
33c46939 | 5183 | d++; |
e934739e RS |
5184 | } |
5185 | while (--mcnt); | |
fa9a63c5 RM |
5186 | } |
5187 | else | |
5188 | { | |
5189 | do | |
5190 | { | |
5191 | PREFETCH (); | |
99633e97 SM |
5192 | if (*d++ != *p++) |
5193 | { | |
5194 | d = dfail; | |
5195 | goto fail; | |
5196 | } | |
fa9a63c5 RM |
5197 | } |
5198 | while (--mcnt); | |
5199 | } | |
25fe55af | 5200 | break; |
fa9a63c5 RM |
5201 | |
5202 | ||
25fe55af | 5203 | /* Match any character except possibly a newline or a null. */ |
fa9a63c5 | 5204 | case anychar: |
e934739e RS |
5205 | { |
5206 | int buf_charlen; | |
01618498 | 5207 | re_wchar_t buf_ch; |
fa9a63c5 | 5208 | |
e934739e | 5209 | DEBUG_PRINT1 ("EXECUTING anychar.\n"); |
fa9a63c5 | 5210 | |
e934739e | 5211 | PREFETCH (); |
2d1675e4 | 5212 | buf_ch = RE_STRING_CHAR_AND_LENGTH (d, dend - d, buf_charlen); |
e934739e RS |
5213 | buf_ch = TRANSLATE (buf_ch); |
5214 | ||
5215 | if ((!(bufp->syntax & RE_DOT_NEWLINE) | |
5216 | && buf_ch == '\n') | |
5217 | || ((bufp->syntax & RE_DOT_NOT_NULL) | |
5218 | && buf_ch == '\000')) | |
5219 | goto fail; | |
5220 | ||
e934739e RS |
5221 | DEBUG_PRINT2 (" Matched `%d'.\n", *d); |
5222 | d += buf_charlen; | |
5223 | } | |
fa9a63c5 RM |
5224 | break; |
5225 | ||
5226 | ||
5227 | case charset: | |
5228 | case charset_not: | |
5229 | { | |
b18215fc | 5230 | register unsigned int c; |
fa9a63c5 | 5231 | boolean not = (re_opcode_t) *(p - 1) == charset_not; |
b18215fc RS |
5232 | int len; |
5233 | ||
5234 | /* Start of actual range_table, or end of bitmap if there is no | |
5235 | range table. */ | |
01618498 | 5236 | re_char *range_table; |
b18215fc | 5237 | |
96cc36cc | 5238 | /* Nonzero if there is a range table. */ |
b18215fc RS |
5239 | int range_table_exists; |
5240 | ||
96cc36cc RS |
5241 | /* Number of ranges of range table. This is not included |
5242 | in the initial byte-length of the command. */ | |
5243 | int count = 0; | |
fa9a63c5 | 5244 | |
25fe55af | 5245 | DEBUG_PRINT2 ("EXECUTING charset%s.\n", not ? "_not" : ""); |
fa9a63c5 | 5246 | |
b18215fc | 5247 | range_table_exists = CHARSET_RANGE_TABLE_EXISTS_P (&p[-1]); |
96cc36cc | 5248 | |
b18215fc | 5249 | if (range_table_exists) |
96cc36cc RS |
5250 | { |
5251 | range_table = CHARSET_RANGE_TABLE (&p[-1]); /* Past the bitmap. */ | |
5252 | EXTRACT_NUMBER_AND_INCR (count, range_table); | |
5253 | } | |
b18215fc | 5254 | |
2d1675e4 SM |
5255 | PREFETCH (); |
5256 | c = RE_STRING_CHAR_AND_LENGTH (d, dend - d, len); | |
5257 | c = TRANSLATE (c); /* The character to match. */ | |
b18215fc RS |
5258 | |
5259 | if (SINGLE_BYTE_CHAR_P (c)) | |
5260 | { /* Lookup bitmap. */ | |
b18215fc RS |
5261 | /* Cast to `unsigned' instead of `unsigned char' in |
5262 | case the bit list is a full 32 bytes long. */ | |
5263 | if (c < (unsigned) (CHARSET_BITMAP_SIZE (&p[-1]) * BYTEWIDTH) | |
96cc36cc RS |
5264 | && p[1 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH))) |
5265 | not = !not; | |
b18215fc | 5266 | } |
96cc36cc | 5267 | #ifdef emacs |
b18215fc | 5268 | else if (range_table_exists) |
96cc36cc RS |
5269 | { |
5270 | int class_bits = CHARSET_RANGE_TABLE_BITS (&p[-1]); | |
5271 | ||
14473664 SM |
5272 | if ( (class_bits & BIT_LOWER && ISLOWER (c)) |
5273 | | (class_bits & BIT_MULTIBYTE) | |
96cc36cc RS |
5274 | | (class_bits & BIT_PUNCT && ISPUNCT (c)) |
5275 | | (class_bits & BIT_SPACE && ISSPACE (c)) | |
5276 | | (class_bits & BIT_UPPER && ISUPPER (c)) | |
5277 | | (class_bits & BIT_WORD && ISWORD (c))) | |
5278 | not = !not; | |
5279 | else | |
5280 | CHARSET_LOOKUP_RANGE_TABLE_RAW (not, c, range_table, count); | |
5281 | } | |
5282 | #endif /* emacs */ | |
fa9a63c5 | 5283 | |
96cc36cc RS |
5284 | if (range_table_exists) |
5285 | p = CHARSET_RANGE_TABLE_END (range_table, count); | |
5286 | else | |
5287 | p += CHARSET_BITMAP_SIZE (&p[-1]) + 1; | |
fa9a63c5 RM |
5288 | |
5289 | if (!not) goto fail; | |
5e69f11e | 5290 | |
b18215fc | 5291 | d += len; |
fa9a63c5 RM |
5292 | break; |
5293 | } | |
5294 | ||
5295 | ||
25fe55af | 5296 | /* The beginning of a group is represented by start_memory. |
505bde11 | 5297 | The argument is the register number. The text |
25fe55af RS |
5298 | matched within the group is recorded (in the internal |
5299 | registers data structure) under the register number. */ | |
5300 | case start_memory: | |
505bde11 SM |
5301 | DEBUG_PRINT2 ("EXECUTING start_memory %d:\n", *p); |
5302 | ||
5303 | /* In case we need to undo this operation (via backtracking). */ | |
5304 | PUSH_FAILURE_REG ((unsigned int)*p); | |
fa9a63c5 | 5305 | |
25fe55af | 5306 | regstart[*p] = d; |
4bb91c68 | 5307 | regend[*p] = NULL; /* probably unnecessary. -sm */ |
fa9a63c5 RM |
5308 | DEBUG_PRINT2 (" regstart: %d\n", POINTER_TO_OFFSET (regstart[*p])); |
5309 | ||
25fe55af | 5310 | /* Move past the register number and inner group count. */ |
505bde11 | 5311 | p += 1; |
25fe55af | 5312 | break; |
fa9a63c5 RM |
5313 | |
5314 | ||
25fe55af | 5315 | /* The stop_memory opcode represents the end of a group. Its |
505bde11 | 5316 | argument is the same as start_memory's: the register number. */ |
fa9a63c5 | 5317 | case stop_memory: |
505bde11 SM |
5318 | DEBUG_PRINT2 ("EXECUTING stop_memory %d:\n", *p); |
5319 | ||
5320 | assert (!REG_UNSET (regstart[*p])); | |
5321 | /* Strictly speaking, there should be code such as: | |
5322 | ||
0b32bf0e | 5323 | assert (REG_UNSET (regend[*p])); |
505bde11 SM |
5324 | PUSH_FAILURE_REGSTOP ((unsigned int)*p); |
5325 | ||
5326 | But the only info to be pushed is regend[*p] and it is known to | |
5327 | be UNSET, so there really isn't anything to push. | |
5328 | Not pushing anything, on the other hand deprives us from the | |
5329 | guarantee that regend[*p] is UNSET since undoing this operation | |
5330 | will not reset its value properly. This is not important since | |
5331 | the value will only be read on the next start_memory or at | |
5332 | the very end and both events can only happen if this stop_memory | |
5333 | is *not* undone. */ | |
fa9a63c5 | 5334 | |
25fe55af | 5335 | regend[*p] = d; |
fa9a63c5 RM |
5336 | DEBUG_PRINT2 (" regend: %d\n", POINTER_TO_OFFSET (regend[*p])); |
5337 | ||
25fe55af | 5338 | /* Move past the register number and the inner group count. */ |
505bde11 | 5339 | p += 1; |
25fe55af | 5340 | break; |
fa9a63c5 RM |
5341 | |
5342 | ||
5343 | /* \<digit> has been turned into a `duplicate' command which is | |
25fe55af RS |
5344 | followed by the numeric value of <digit> as the register number. */ |
5345 | case duplicate: | |
fa9a63c5 | 5346 | { |
66f0296e | 5347 | register re_char *d2, *dend2; |
25fe55af | 5348 | int regno = *p++; /* Get which register to match against. */ |
fa9a63c5 RM |
5349 | DEBUG_PRINT2 ("EXECUTING duplicate %d.\n", regno); |
5350 | ||
25fe55af RS |
5351 | /* Can't back reference a group which we've never matched. */ |
5352 | if (REG_UNSET (regstart[regno]) || REG_UNSET (regend[regno])) | |
5353 | goto fail; | |
5e69f11e | 5354 | |
25fe55af RS |
5355 | /* Where in input to try to start matching. */ |
5356 | d2 = regstart[regno]; | |
5e69f11e | 5357 | |
99633e97 SM |
5358 | /* Remember the start point to rollback upon failure. */ |
5359 | dfail = d; | |
5360 | ||
25fe55af RS |
5361 | /* Where to stop matching; if both the place to start and |
5362 | the place to stop matching are in the same string, then | |
5363 | set to the place to stop, otherwise, for now have to use | |
5364 | the end of the first string. */ | |
fa9a63c5 | 5365 | |
25fe55af | 5366 | dend2 = ((FIRST_STRING_P (regstart[regno]) |
fa9a63c5 RM |
5367 | == FIRST_STRING_P (regend[regno])) |
5368 | ? regend[regno] : end_match_1); | |
5369 | for (;;) | |
5370 | { | |
5371 | /* If necessary, advance to next segment in register | |
25fe55af | 5372 | contents. */ |
fa9a63c5 RM |
5373 | while (d2 == dend2) |
5374 | { | |
5375 | if (dend2 == end_match_2) break; | |
5376 | if (dend2 == regend[regno]) break; | |
5377 | ||
25fe55af RS |
5378 | /* End of string1 => advance to string2. */ |
5379 | d2 = string2; | |
5380 | dend2 = regend[regno]; | |
fa9a63c5 RM |
5381 | } |
5382 | /* At end of register contents => success */ | |
5383 | if (d2 == dend2) break; | |
5384 | ||
5385 | /* If necessary, advance to next segment in data. */ | |
5386 | PREFETCH (); | |
5387 | ||
5388 | /* How many characters left in this segment to match. */ | |
5389 | mcnt = dend - d; | |
5e69f11e | 5390 | |
fa9a63c5 | 5391 | /* Want how many consecutive characters we can match in |
25fe55af RS |
5392 | one shot, so, if necessary, adjust the count. */ |
5393 | if (mcnt > dend2 - d2) | |
fa9a63c5 | 5394 | mcnt = dend2 - d2; |
5e69f11e | 5395 | |
fa9a63c5 | 5396 | /* Compare that many; failure if mismatch, else move |
25fe55af | 5397 | past them. */ |
28703c16 | 5398 | if (RE_TRANSLATE_P (translate) |
2d1675e4 | 5399 | ? bcmp_translate (d, d2, mcnt, translate, multibyte) |
4bb91c68 | 5400 | : memcmp (d, d2, mcnt)) |
99633e97 SM |
5401 | { |
5402 | d = dfail; | |
5403 | goto fail; | |
5404 | } | |
fa9a63c5 | 5405 | d += mcnt, d2 += mcnt; |
fa9a63c5 RM |
5406 | } |
5407 | } | |
5408 | break; | |
5409 | ||
5410 | ||
25fe55af | 5411 | /* begline matches the empty string at the beginning of the string |
c0f9ea08 | 5412 | (unless `not_bol' is set in `bufp'), and after newlines. */ |
fa9a63c5 | 5413 | case begline: |
25fe55af | 5414 | DEBUG_PRINT1 ("EXECUTING begline.\n"); |
5e69f11e | 5415 | |
25fe55af RS |
5416 | if (AT_STRINGS_BEG (d)) |
5417 | { | |
5418 | if (!bufp->not_bol) break; | |
5419 | } | |
419d1c74 | 5420 | else |
25fe55af | 5421 | { |
419d1c74 SM |
5422 | unsigned char c; |
5423 | GET_CHAR_BEFORE_2 (c, d, string1, end1, string2, end2); | |
c0f9ea08 | 5424 | if (c == '\n') |
419d1c74 | 5425 | break; |
25fe55af RS |
5426 | } |
5427 | /* In all other cases, we fail. */ | |
5428 | goto fail; | |
fa9a63c5 RM |
5429 | |
5430 | ||
25fe55af | 5431 | /* endline is the dual of begline. */ |
fa9a63c5 | 5432 | case endline: |
25fe55af | 5433 | DEBUG_PRINT1 ("EXECUTING endline.\n"); |
fa9a63c5 | 5434 | |
25fe55af RS |
5435 | if (AT_STRINGS_END (d)) |
5436 | { | |
5437 | if (!bufp->not_eol) break; | |
5438 | } | |
f1ad044f | 5439 | else |
25fe55af | 5440 | { |
f1ad044f | 5441 | PREFETCH_NOLIMIT (); |
c0f9ea08 | 5442 | if (*d == '\n') |
f1ad044f | 5443 | break; |
25fe55af RS |
5444 | } |
5445 | goto fail; | |
fa9a63c5 RM |
5446 | |
5447 | ||
5448 | /* Match at the very beginning of the data. */ | |
25fe55af RS |
5449 | case begbuf: |
5450 | DEBUG_PRINT1 ("EXECUTING begbuf.\n"); | |
5451 | if (AT_STRINGS_BEG (d)) | |
5452 | break; | |
5453 | goto fail; | |
fa9a63c5 RM |
5454 | |
5455 | ||
5456 | /* Match at the very end of the data. */ | |
25fe55af RS |
5457 | case endbuf: |
5458 | DEBUG_PRINT1 ("EXECUTING endbuf.\n"); | |
fa9a63c5 RM |
5459 | if (AT_STRINGS_END (d)) |
5460 | break; | |
25fe55af | 5461 | goto fail; |
5e69f11e | 5462 | |
5e69f11e | 5463 | |
25fe55af RS |
5464 | /* on_failure_keep_string_jump is used to optimize `.*\n'. It |
5465 | pushes NULL as the value for the string on the stack. Then | |
505bde11 | 5466 | `POP_FAILURE_POINT' will keep the current value for the |
25fe55af RS |
5467 | string, instead of restoring it. To see why, consider |
5468 | matching `foo\nbar' against `.*\n'. The .* matches the foo; | |
5469 | then the . fails against the \n. But the next thing we want | |
5470 | to do is match the \n against the \n; if we restored the | |
5471 | string value, we would be back at the foo. | |
5472 | ||
5473 | Because this is used only in specific cases, we don't need to | |
5474 | check all the things that `on_failure_jump' does, to make | |
5475 | sure the right things get saved on the stack. Hence we don't | |
5476 | share its code. The only reason to push anything on the | |
5477 | stack at all is that otherwise we would have to change | |
5478 | `anychar's code to do something besides goto fail in this | |
5479 | case; that seems worse than this. */ | |
5480 | case on_failure_keep_string_jump: | |
505bde11 SM |
5481 | EXTRACT_NUMBER_AND_INCR (mcnt, p); |
5482 | DEBUG_PRINT3 ("EXECUTING on_failure_keep_string_jump %d (to %p):\n", | |
5483 | mcnt, p + mcnt); | |
fa9a63c5 | 5484 | |
505bde11 SM |
5485 | PUSH_FAILURE_POINT (p - 3, NULL); |
5486 | break; | |
5487 | ||
0683b6fa SM |
5488 | /* A nasty loop is introduced by the non-greedy *? and +?. |
5489 | With such loops, the stack only ever contains one failure point | |
5490 | at a time, so that a plain on_failure_jump_loop kind of | |
5491 | cycle detection cannot work. Worse yet, such a detection | |
5492 | can not only fail to detect a cycle, but it can also wrongly | |
5493 | detect a cycle (between different instantiations of the same | |
5494 | loop. | |
5495 | So the method used for those nasty loops is a little different: | |
5496 | We use a special cycle-detection-stack-frame which is pushed | |
5497 | when the on_failure_jump_nastyloop failure-point is *popped*. | |
5498 | This special frame thus marks the beginning of one iteration | |
5499 | through the loop and we can hence easily check right here | |
5500 | whether something matched between the beginning and the end of | |
5501 | the loop. */ | |
5502 | case on_failure_jump_nastyloop: | |
5503 | EXTRACT_NUMBER_AND_INCR (mcnt, p); | |
5504 | DEBUG_PRINT3 ("EXECUTING on_failure_jump_nastyloop %d (to %p):\n", | |
5505 | mcnt, p + mcnt); | |
5506 | ||
5507 | assert ((re_opcode_t)p[-4] == no_op); | |
5508 | CHECK_INFINITE_LOOP (p - 4, d); | |
5509 | PUSH_FAILURE_POINT (p - 3, d); | |
5510 | break; | |
5511 | ||
505bde11 | 5512 | |
4e8a9132 SM |
5513 | /* Simple loop detecting on_failure_jump: just check on the |
5514 | failure stack if the same spot was already hit earlier. */ | |
505bde11 SM |
5515 | case on_failure_jump_loop: |
5516 | on_failure: | |
5517 | EXTRACT_NUMBER_AND_INCR (mcnt, p); | |
5518 | DEBUG_PRINT3 ("EXECUTING on_failure_jump_loop %d (to %p):\n", | |
5519 | mcnt, p + mcnt); | |
5520 | ||
5521 | CHECK_INFINITE_LOOP (p - 3, d); | |
5522 | PUSH_FAILURE_POINT (p - 3, d); | |
25fe55af | 5523 | break; |
fa9a63c5 RM |
5524 | |
5525 | ||
5526 | /* Uses of on_failure_jump: | |
5e69f11e | 5527 | |
25fe55af RS |
5528 | Each alternative starts with an on_failure_jump that points |
5529 | to the beginning of the next alternative. Each alternative | |
5530 | except the last ends with a jump that in effect jumps past | |
5531 | the rest of the alternatives. (They really jump to the | |
5532 | ending jump of the following alternative, because tensioning | |
5533 | these jumps is a hassle.) | |
fa9a63c5 | 5534 | |
25fe55af RS |
5535 | Repeats start with an on_failure_jump that points past both |
5536 | the repetition text and either the following jump or | |
5537 | pop_failure_jump back to this on_failure_jump. */ | |
fa9a63c5 | 5538 | case on_failure_jump: |
5b370c2b | 5539 | IMMEDIATE_QUIT_CHECK; |
25fe55af | 5540 | EXTRACT_NUMBER_AND_INCR (mcnt, p); |
505bde11 SM |
5541 | DEBUG_PRINT3 ("EXECUTING on_failure_jump %d (to %p):\n", |
5542 | mcnt, p + mcnt); | |
25fe55af | 5543 | |
505bde11 | 5544 | PUSH_FAILURE_POINT (p -3, d); |
25fe55af RS |
5545 | break; |
5546 | ||
4e8a9132 | 5547 | /* This operation is used for greedy *. |
505bde11 SM |
5548 | Compare the beginning of the repeat with what in the |
5549 | pattern follows its end. If we can establish that there | |
5550 | is nothing that they would both match, i.e., that we | |
5551 | would have to backtrack because of (as in, e.g., `a*a') | |
5552 | then we can use a non-backtracking loop based on | |
4e8a9132 | 5553 | on_failure_keep_string_jump instead of on_failure_jump. */ |
505bde11 | 5554 | case on_failure_jump_smart: |
5b370c2b | 5555 | IMMEDIATE_QUIT_CHECK; |
25fe55af | 5556 | EXTRACT_NUMBER_AND_INCR (mcnt, p); |
505bde11 SM |
5557 | DEBUG_PRINT3 ("EXECUTING on_failure_jump_smart %d (to %p).\n", |
5558 | mcnt, p + mcnt); | |
25fe55af | 5559 | { |
01618498 | 5560 | re_char *p1 = p; /* Next operation. */ |
6dcf2d0e SM |
5561 | /* Here, we discard `const', making re_match non-reentrant. */ |
5562 | unsigned char *p2 = (unsigned char*) p + mcnt; /* Jump dest. */ | |
5563 | unsigned char *p3 = (unsigned char*) p - 3; /* opcode location. */ | |
fa9a63c5 | 5564 | |
505bde11 SM |
5565 | p -= 3; /* Reset so that we will re-execute the |
5566 | instruction once it's been changed. */ | |
fa9a63c5 | 5567 | |
4e8a9132 SM |
5568 | EXTRACT_NUMBER (mcnt, p2 - 2); |
5569 | ||
5570 | /* Ensure this is a indeed the trivial kind of loop | |
5571 | we are expecting. */ | |
5572 | assert (skip_one_char (p1) == p2 - 3); | |
5573 | assert ((re_opcode_t) p2[-3] == jump && p2 + mcnt == p); | |
99633e97 | 5574 | DEBUG_STATEMENT (debug += 2); |
505bde11 | 5575 | if (mutually_exclusive_p (bufp, p1, p2)) |
fa9a63c5 | 5576 | { |
505bde11 | 5577 | /* Use a fast `on_failure_keep_string_jump' loop. */ |
4e8a9132 | 5578 | DEBUG_PRINT1 (" smart exclusive => fast loop.\n"); |
01618498 | 5579 | *p3 = (unsigned char) on_failure_keep_string_jump; |
4e8a9132 | 5580 | STORE_NUMBER (p2 - 2, mcnt + 3); |
25fe55af | 5581 | } |
505bde11 | 5582 | else |
fa9a63c5 | 5583 | { |
505bde11 SM |
5584 | /* Default to a safe `on_failure_jump' loop. */ |
5585 | DEBUG_PRINT1 (" smart default => slow loop.\n"); | |
01618498 | 5586 | *p3 = (unsigned char) on_failure_jump; |
fa9a63c5 | 5587 | } |
99633e97 | 5588 | DEBUG_STATEMENT (debug -= 2); |
25fe55af | 5589 | } |
505bde11 | 5590 | break; |
25fe55af RS |
5591 | |
5592 | /* Unconditionally jump (without popping any failure points). */ | |
5593 | case jump: | |
fa9a63c5 | 5594 | unconditional_jump: |
5b370c2b | 5595 | IMMEDIATE_QUIT_CHECK; |
fa9a63c5 | 5596 | EXTRACT_NUMBER_AND_INCR (mcnt, p); /* Get the amount to jump. */ |
25fe55af RS |
5597 | DEBUG_PRINT2 ("EXECUTING jump %d ", mcnt); |
5598 | p += mcnt; /* Do the jump. */ | |
505bde11 | 5599 | DEBUG_PRINT2 ("(to %p).\n", p); |
25fe55af RS |
5600 | break; |
5601 | ||
5602 | ||
25fe55af RS |
5603 | /* Have to succeed matching what follows at least n times. |
5604 | After that, handle like `on_failure_jump'. */ | |
5605 | case succeed_n: | |
01618498 | 5606 | /* Signedness doesn't matter since we only compare MCNT to 0. */ |
25fe55af RS |
5607 | EXTRACT_NUMBER (mcnt, p + 2); |
5608 | DEBUG_PRINT2 ("EXECUTING succeed_n %d.\n", mcnt); | |
5e69f11e | 5609 | |
dc1e502d SM |
5610 | /* Originally, mcnt is how many times we HAVE to succeed. */ |
5611 | if (mcnt != 0) | |
25fe55af | 5612 | { |
6dcf2d0e SM |
5613 | /* Here, we discard `const', making re_match non-reentrant. */ |
5614 | unsigned char *p2 = (unsigned char*) p + 2; /* counter loc. */ | |
dc1e502d | 5615 | mcnt--; |
01618498 SM |
5616 | p += 4; |
5617 | PUSH_NUMBER (p2, mcnt); | |
25fe55af | 5618 | } |
dc1e502d SM |
5619 | else |
5620 | /* The two bytes encoding mcnt == 0 are two no_op opcodes. */ | |
5621 | goto on_failure; | |
25fe55af RS |
5622 | break; |
5623 | ||
5624 | case jump_n: | |
01618498 | 5625 | /* Signedness doesn't matter since we only compare MCNT to 0. */ |
25fe55af RS |
5626 | EXTRACT_NUMBER (mcnt, p + 2); |
5627 | DEBUG_PRINT2 ("EXECUTING jump_n %d.\n", mcnt); | |
5628 | ||
5629 | /* Originally, this is how many times we CAN jump. */ | |
dc1e502d | 5630 | if (mcnt != 0) |
25fe55af | 5631 | { |
6dcf2d0e SM |
5632 | /* Here, we discard `const', making re_match non-reentrant. */ |
5633 | unsigned char *p2 = (unsigned char*) p + 2; /* counter loc. */ | |
dc1e502d | 5634 | mcnt--; |
01618498 | 5635 | PUSH_NUMBER (p2, mcnt); |
dc1e502d | 5636 | goto unconditional_jump; |
25fe55af RS |
5637 | } |
5638 | /* If don't have to jump any more, skip over the rest of command. */ | |
5e69f11e RM |
5639 | else |
5640 | p += 4; | |
25fe55af | 5641 | break; |
5e69f11e | 5642 | |
fa9a63c5 RM |
5643 | case set_number_at: |
5644 | { | |
01618498 | 5645 | unsigned char *p2; /* Location of the counter. */ |
25fe55af | 5646 | DEBUG_PRINT1 ("EXECUTING set_number_at.\n"); |
fa9a63c5 | 5647 | |
25fe55af | 5648 | EXTRACT_NUMBER_AND_INCR (mcnt, p); |
6dcf2d0e SM |
5649 | /* Here, we discard `const', making re_match non-reentrant. */ |
5650 | p2 = (unsigned char*) p + mcnt; | |
01618498 | 5651 | /* Signedness doesn't matter since we only copy MCNT's bits . */ |
25fe55af | 5652 | EXTRACT_NUMBER_AND_INCR (mcnt, p); |
01618498 SM |
5653 | DEBUG_PRINT3 (" Setting %p to %d.\n", p2, mcnt); |
5654 | PUSH_NUMBER (p2, mcnt); | |
25fe55af RS |
5655 | break; |
5656 | } | |
9121ca40 KH |
5657 | |
5658 | case wordbound: | |
66f0296e SM |
5659 | case notwordbound: |
5660 | not = (re_opcode_t) *(p - 1) == notwordbound; | |
5661 | DEBUG_PRINT2 ("EXECUTING %swordbound.\n", not?"not":""); | |
fa9a63c5 | 5662 | |
99633e97 | 5663 | /* We SUCCEED (or FAIL) in one of the following cases: */ |
9121ca40 | 5664 | |
b18215fc | 5665 | /* Case 1: D is at the beginning or the end of string. */ |
9121ca40 | 5666 | if (AT_STRINGS_BEG (d) || AT_STRINGS_END (d)) |
66f0296e | 5667 | not = !not; |
b18215fc RS |
5668 | else |
5669 | { | |
5670 | /* C1 is the character before D, S1 is the syntax of C1, C2 | |
5671 | is the character at D, and S2 is the syntax of C2. */ | |
01618498 SM |
5672 | re_wchar_t c1, c2; |
5673 | int s1, s2; | |
b18215fc | 5674 | #ifdef emacs |
2d1675e4 SM |
5675 | int offset = PTR_TO_OFFSET (d - 1); |
5676 | int charpos = SYNTAX_TABLE_BYTE_TO_CHAR (offset); | |
5d967c7a | 5677 | UPDATE_SYNTAX_TABLE (charpos); |
25fe55af | 5678 | #endif |
66f0296e | 5679 | GET_CHAR_BEFORE_2 (c1, d, string1, end1, string2, end2); |
b18215fc RS |
5680 | s1 = SYNTAX (c1); |
5681 | #ifdef emacs | |
5d967c7a | 5682 | UPDATE_SYNTAX_TABLE_FORWARD (charpos + 1); |
25fe55af | 5683 | #endif |
f1ad044f | 5684 | PREFETCH_NOLIMIT (); |
2d1675e4 | 5685 | c2 = RE_STRING_CHAR (d, dend - d); |
b18215fc RS |
5686 | s2 = SYNTAX (c2); |
5687 | ||
5688 | if (/* Case 2: Only one of S1 and S2 is Sword. */ | |
5689 | ((s1 == Sword) != (s2 == Sword)) | |
5690 | /* Case 3: Both of S1 and S2 are Sword, and macro | |
25fe55af | 5691 | WORD_BOUNDARY_P (C1, C2) returns nonzero. */ |
b18215fc | 5692 | || ((s1 == Sword) && WORD_BOUNDARY_P (c1, c2))) |
66f0296e SM |
5693 | not = !not; |
5694 | } | |
5695 | if (not) | |
9121ca40 | 5696 | break; |
b18215fc | 5697 | else |
9121ca40 | 5698 | goto fail; |
fa9a63c5 RM |
5699 | |
5700 | case wordbeg: | |
25fe55af | 5701 | DEBUG_PRINT1 ("EXECUTING wordbeg.\n"); |
fa9a63c5 | 5702 | |
b18215fc RS |
5703 | /* We FAIL in one of the following cases: */ |
5704 | ||
25fe55af | 5705 | /* Case 1: D is at the end of string. */ |
b18215fc | 5706 | if (AT_STRINGS_END (d)) |
99633e97 | 5707 | goto fail; |
b18215fc RS |
5708 | else |
5709 | { | |
5710 | /* C1 is the character before D, S1 is the syntax of C1, C2 | |
5711 | is the character at D, and S2 is the syntax of C2. */ | |
01618498 SM |
5712 | re_wchar_t c1, c2; |
5713 | int s1, s2; | |
fa9a63c5 | 5714 | #ifdef emacs |
2d1675e4 SM |
5715 | int offset = PTR_TO_OFFSET (d); |
5716 | int charpos = SYNTAX_TABLE_BYTE_TO_CHAR (offset); | |
92432794 | 5717 | UPDATE_SYNTAX_TABLE (charpos); |
25fe55af | 5718 | #endif |
99633e97 | 5719 | PREFETCH (); |
2d1675e4 | 5720 | c2 = RE_STRING_CHAR (d, dend - d); |
b18215fc | 5721 | s2 = SYNTAX (c2); |
25fe55af | 5722 | |
b18215fc RS |
5723 | /* Case 2: S2 is not Sword. */ |
5724 | if (s2 != Sword) | |
5725 | goto fail; | |
5726 | ||
5727 | /* Case 3: D is not at the beginning of string ... */ | |
5728 | if (!AT_STRINGS_BEG (d)) | |
5729 | { | |
5730 | GET_CHAR_BEFORE_2 (c1, d, string1, end1, string2, end2); | |
5731 | #ifdef emacs | |
5d967c7a | 5732 | UPDATE_SYNTAX_TABLE_BACKWARD (charpos - 1); |
25fe55af | 5733 | #endif |
b18215fc RS |
5734 | s1 = SYNTAX (c1); |
5735 | ||
5736 | /* ... and S1 is Sword, and WORD_BOUNDARY_P (C1, C2) | |
25fe55af | 5737 | returns 0. */ |
b18215fc RS |
5738 | if ((s1 == Sword) && !WORD_BOUNDARY_P (c1, c2)) |
5739 | goto fail; | |
5740 | } | |
5741 | } | |
e318085a RS |
5742 | break; |
5743 | ||
b18215fc | 5744 | case wordend: |
25fe55af | 5745 | DEBUG_PRINT1 ("EXECUTING wordend.\n"); |
b18215fc RS |
5746 | |
5747 | /* We FAIL in one of the following cases: */ | |
5748 | ||
5749 | /* Case 1: D is at the beginning of string. */ | |
5750 | if (AT_STRINGS_BEG (d)) | |
e318085a | 5751 | goto fail; |
b18215fc RS |
5752 | else |
5753 | { | |
5754 | /* C1 is the character before D, S1 is the syntax of C1, C2 | |
5755 | is the character at D, and S2 is the syntax of C2. */ | |
01618498 SM |
5756 | re_wchar_t c1, c2; |
5757 | int s1, s2; | |
5d967c7a | 5758 | #ifdef emacs |
2d1675e4 SM |
5759 | int offset = PTR_TO_OFFSET (d) - 1; |
5760 | int charpos = SYNTAX_TABLE_BYTE_TO_CHAR (offset); | |
92432794 | 5761 | UPDATE_SYNTAX_TABLE (charpos); |
5d967c7a | 5762 | #endif |
99633e97 | 5763 | GET_CHAR_BEFORE_2 (c1, d, string1, end1, string2, end2); |
b18215fc RS |
5764 | s1 = SYNTAX (c1); |
5765 | ||
5766 | /* Case 2: S1 is not Sword. */ | |
5767 | if (s1 != Sword) | |
5768 | goto fail; | |
5769 | ||
5770 | /* Case 3: D is not at the end of string ... */ | |
5771 | if (!AT_STRINGS_END (d)) | |
5772 | { | |
f1ad044f | 5773 | PREFETCH_NOLIMIT (); |
2d1675e4 | 5774 | c2 = RE_STRING_CHAR (d, dend - d); |
5d967c7a RS |
5775 | #ifdef emacs |
5776 | UPDATE_SYNTAX_TABLE_FORWARD (charpos); | |
5777 | #endif | |
b18215fc RS |
5778 | s2 = SYNTAX (c2); |
5779 | ||
5780 | /* ... and S2 is Sword, and WORD_BOUNDARY_P (C1, C2) | |
25fe55af | 5781 | returns 0. */ |
b18215fc | 5782 | if ((s2 == Sword) && !WORD_BOUNDARY_P (c1, c2)) |
25fe55af | 5783 | goto fail; |
b18215fc RS |
5784 | } |
5785 | } | |
e318085a RS |
5786 | break; |
5787 | ||
fa9a63c5 | 5788 | case syntaxspec: |
1fb352e0 SM |
5789 | case notsyntaxspec: |
5790 | not = (re_opcode_t) *(p - 1) == notsyntaxspec; | |
fa9a63c5 | 5791 | mcnt = *p++; |
1fb352e0 | 5792 | DEBUG_PRINT3 ("EXECUTING %ssyntaxspec %d.\n", not?"not":"", mcnt); |
fa9a63c5 | 5793 | PREFETCH (); |
b18215fc RS |
5794 | #ifdef emacs |
5795 | { | |
2d1675e4 SM |
5796 | int offset = PTR_TO_OFFSET (d); |
5797 | int pos1 = SYNTAX_TABLE_BYTE_TO_CHAR (offset); | |
b18215fc RS |
5798 | UPDATE_SYNTAX_TABLE (pos1); |
5799 | } | |
25fe55af | 5800 | #endif |
b18215fc | 5801 | { |
01618498 SM |
5802 | int len; |
5803 | re_wchar_t c; | |
b18215fc | 5804 | |
2d1675e4 | 5805 | c = RE_STRING_CHAR_AND_LENGTH (d, dend - d, len); |
b18215fc | 5806 | |
990b2375 | 5807 | if ((SYNTAX (c) != (enum syntaxcode) mcnt) ^ not) |
1fb352e0 | 5808 | goto fail; |
b18215fc RS |
5809 | d += len; |
5810 | } | |
fa9a63c5 RM |
5811 | break; |
5812 | ||
b18215fc | 5813 | #ifdef emacs |
1fb352e0 SM |
5814 | case before_dot: |
5815 | DEBUG_PRINT1 ("EXECUTING before_dot.\n"); | |
5816 | if (PTR_BYTE_POS (d) >= PT_BYTE) | |
fa9a63c5 | 5817 | goto fail; |
b18215fc RS |
5818 | break; |
5819 | ||
1fb352e0 SM |
5820 | case at_dot: |
5821 | DEBUG_PRINT1 ("EXECUTING at_dot.\n"); | |
5822 | if (PTR_BYTE_POS (d) != PT_BYTE) | |
5823 | goto fail; | |
5824 | break; | |
b18215fc | 5825 | |
1fb352e0 SM |
5826 | case after_dot: |
5827 | DEBUG_PRINT1 ("EXECUTING after_dot.\n"); | |
5828 | if (PTR_BYTE_POS (d) <= PT_BYTE) | |
5829 | goto fail; | |
e318085a | 5830 | break; |
fa9a63c5 | 5831 | |
1fb352e0 | 5832 | case categoryspec: |
b18215fc | 5833 | case notcategoryspec: |
1fb352e0 | 5834 | not = (re_opcode_t) *(p - 1) == notcategoryspec; |
b18215fc | 5835 | mcnt = *p++; |
1fb352e0 | 5836 | DEBUG_PRINT3 ("EXECUTING %scategoryspec %d.\n", not?"not":"", mcnt); |
b18215fc RS |
5837 | PREFETCH (); |
5838 | { | |
01618498 SM |
5839 | int len; |
5840 | re_wchar_t c; | |
5841 | ||
2d1675e4 | 5842 | c = RE_STRING_CHAR_AND_LENGTH (d, dend - d, len); |
b18215fc | 5843 | |
1fb352e0 | 5844 | if ((!CHAR_HAS_CATEGORY (c, mcnt)) ^ not) |
b18215fc RS |
5845 | goto fail; |
5846 | d += len; | |
5847 | } | |
fa9a63c5 | 5848 | break; |
5e69f11e | 5849 | |
1fb352e0 | 5850 | #endif /* emacs */ |
5e69f11e | 5851 | |
0b32bf0e SM |
5852 | default: |
5853 | abort (); | |
fa9a63c5 | 5854 | } |
b18215fc | 5855 | continue; /* Successfully executed one pattern command; keep going. */ |
fa9a63c5 RM |
5856 | |
5857 | ||
5858 | /* We goto here if a matching operation fails. */ | |
5859 | fail: | |
5b370c2b | 5860 | IMMEDIATE_QUIT_CHECK; |
fa9a63c5 | 5861 | if (!FAIL_STACK_EMPTY ()) |
505bde11 | 5862 | { |
01618498 | 5863 | re_char *str, *pat; |
505bde11 | 5864 | /* A restart point is known. Restore to that state. */ |
0b32bf0e SM |
5865 | DEBUG_PRINT1 ("\nFAIL:\n"); |
5866 | POP_FAILURE_POINT (str, pat); | |
505bde11 SM |
5867 | switch (SWITCH_ENUM_CAST ((re_opcode_t) *pat++)) |
5868 | { | |
5869 | case on_failure_keep_string_jump: | |
5870 | assert (str == NULL); | |
5871 | goto continue_failure_jump; | |
5872 | ||
0683b6fa SM |
5873 | case on_failure_jump_nastyloop: |
5874 | assert ((re_opcode_t)pat[-2] == no_op); | |
5875 | PUSH_FAILURE_POINT (pat - 2, str); | |
5876 | /* Fallthrough */ | |
5877 | ||
505bde11 SM |
5878 | case on_failure_jump_loop: |
5879 | case on_failure_jump: | |
5880 | case succeed_n: | |
5881 | d = str; | |
5882 | continue_failure_jump: | |
5883 | EXTRACT_NUMBER_AND_INCR (mcnt, pat); | |
5884 | p = pat + mcnt; | |
5885 | break; | |
b18215fc | 5886 | |
0683b6fa SM |
5887 | case no_op: |
5888 | /* A special frame used for nastyloops. */ | |
5889 | goto fail; | |
5890 | ||
505bde11 SM |
5891 | default: |
5892 | abort(); | |
5893 | } | |
fa9a63c5 | 5894 | |
505bde11 | 5895 | assert (p >= bufp->buffer && p <= pend); |
b18215fc | 5896 | |
0b32bf0e | 5897 | if (d >= string1 && d <= end1) |
fa9a63c5 | 5898 | dend = end_match_1; |
0b32bf0e | 5899 | } |
fa9a63c5 | 5900 | else |
0b32bf0e | 5901 | break; /* Matching at this starting point really fails. */ |
fa9a63c5 RM |
5902 | } /* for (;;) */ |
5903 | ||
5904 | if (best_regs_set) | |
5905 | goto restore_best_regs; | |
5906 | ||
5907 | FREE_VARIABLES (); | |
5908 | ||
b18215fc | 5909 | return -1; /* Failure to match. */ |
fa9a63c5 RM |
5910 | } /* re_match_2 */ |
5911 | \f | |
5912 | /* Subroutine definitions for re_match_2. */ | |
5913 | ||
fa9a63c5 RM |
5914 | /* Return zero if TRANSLATE[S1] and TRANSLATE[S2] are identical for LEN |
5915 | bytes; nonzero otherwise. */ | |
5e69f11e | 5916 | |
fa9a63c5 | 5917 | static int |
2d1675e4 SM |
5918 | bcmp_translate (s1, s2, len, translate, multibyte) |
5919 | re_char *s1, *s2; | |
fa9a63c5 | 5920 | register int len; |
6676cb1c | 5921 | RE_TRANSLATE_TYPE translate; |
2d1675e4 | 5922 | const int multibyte; |
fa9a63c5 | 5923 | { |
2d1675e4 SM |
5924 | register re_char *p1 = s1, *p2 = s2; |
5925 | re_char *p1_end = s1 + len; | |
5926 | re_char *p2_end = s2 + len; | |
e934739e | 5927 | |
4bb91c68 SM |
5928 | /* FIXME: Checking both p1 and p2 presumes that the two strings might have |
5929 | different lengths, but relying on a single `len' would break this. -sm */ | |
5930 | while (p1 < p1_end && p2 < p2_end) | |
fa9a63c5 | 5931 | { |
e934739e | 5932 | int p1_charlen, p2_charlen; |
01618498 | 5933 | re_wchar_t p1_ch, p2_ch; |
e934739e | 5934 | |
2d1675e4 SM |
5935 | p1_ch = RE_STRING_CHAR_AND_LENGTH (p1, p1_end - p1, p1_charlen); |
5936 | p2_ch = RE_STRING_CHAR_AND_LENGTH (p2, p2_end - p2, p2_charlen); | |
e934739e RS |
5937 | |
5938 | if (RE_TRANSLATE (translate, p1_ch) | |
5939 | != RE_TRANSLATE (translate, p2_ch)) | |
bc192b5b | 5940 | return 1; |
e934739e RS |
5941 | |
5942 | p1 += p1_charlen, p2 += p2_charlen; | |
fa9a63c5 | 5943 | } |
e934739e RS |
5944 | |
5945 | if (p1 != p1_end || p2 != p2_end) | |
5946 | return 1; | |
5947 | ||
fa9a63c5 RM |
5948 | return 0; |
5949 | } | |
5950 | \f | |
5951 | /* Entry points for GNU code. */ | |
5952 | ||
5953 | /* re_compile_pattern is the GNU regular expression compiler: it | |
5954 | compiles PATTERN (of length SIZE) and puts the result in BUFP. | |
5955 | Returns 0 if the pattern was valid, otherwise an error string. | |
5e69f11e | 5956 | |
fa9a63c5 RM |
5957 | Assumes the `allocated' (and perhaps `buffer') and `translate' fields |
5958 | are set in BUFP on entry. | |
5e69f11e | 5959 | |
b18215fc | 5960 | We call regex_compile to do the actual compilation. */ |
fa9a63c5 RM |
5961 | |
5962 | const char * | |
5963 | re_compile_pattern (pattern, length, bufp) | |
5964 | const char *pattern; | |
0b32bf0e | 5965 | size_t length; |
fa9a63c5 RM |
5966 | struct re_pattern_buffer *bufp; |
5967 | { | |
5968 | reg_errcode_t ret; | |
5e69f11e | 5969 | |
fa9a63c5 RM |
5970 | /* GNU code is written to assume at least RE_NREGS registers will be set |
5971 | (and at least one extra will be -1). */ | |
5972 | bufp->regs_allocated = REGS_UNALLOCATED; | |
5e69f11e | 5973 | |
fa9a63c5 RM |
5974 | /* And GNU code determines whether or not to get register information |
5975 | by passing null for the REGS argument to re_match, etc., not by | |
5976 | setting no_sub. */ | |
5977 | bufp->no_sub = 0; | |
5e69f11e | 5978 | |
4bb91c68 | 5979 | ret = regex_compile ((re_char*) pattern, length, re_syntax_options, bufp); |
fa9a63c5 RM |
5980 | |
5981 | if (!ret) | |
5982 | return NULL; | |
5983 | return gettext (re_error_msgid[(int) ret]); | |
5e69f11e | 5984 | } |
c0f9ea08 | 5985 | WEAK_ALIAS (__re_compile_pattern, re_compile_pattern) |
fa9a63c5 | 5986 | \f |
b18215fc RS |
5987 | /* Entry points compatible with 4.2 BSD regex library. We don't define |
5988 | them unless specifically requested. */ | |
fa9a63c5 | 5989 | |
0b32bf0e | 5990 | #if defined _REGEX_RE_COMP || defined _LIBC |
fa9a63c5 RM |
5991 | |
5992 | /* BSD has one and only one pattern buffer. */ | |
5993 | static struct re_pattern_buffer re_comp_buf; | |
5994 | ||
5995 | char * | |
0b32bf0e | 5996 | # ifdef _LIBC |
48afdd44 RM |
5997 | /* Make these definitions weak in libc, so POSIX programs can redefine |
5998 | these names if they don't use our functions, and still use | |
5999 | regcomp/regexec below without link errors. */ | |
6000 | weak_function | |
0b32bf0e | 6001 | # endif |
fa9a63c5 RM |
6002 | re_comp (s) |
6003 | const char *s; | |
6004 | { | |
6005 | reg_errcode_t ret; | |
5e69f11e | 6006 | |
fa9a63c5 RM |
6007 | if (!s) |
6008 | { | |
6009 | if (!re_comp_buf.buffer) | |
0b32bf0e | 6010 | /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */ |
a60198e5 | 6011 | return (char *) gettext ("No previous regular expression"); |
fa9a63c5 RM |
6012 | return 0; |
6013 | } | |
6014 | ||
6015 | if (!re_comp_buf.buffer) | |
6016 | { | |
6017 | re_comp_buf.buffer = (unsigned char *) malloc (200); | |
6018 | if (re_comp_buf.buffer == NULL) | |
0b32bf0e SM |
6019 | /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */ |
6020 | return (char *) gettext (re_error_msgid[(int) REG_ESPACE]); | |
fa9a63c5 RM |
6021 | re_comp_buf.allocated = 200; |
6022 | ||
6023 | re_comp_buf.fastmap = (char *) malloc (1 << BYTEWIDTH); | |
6024 | if (re_comp_buf.fastmap == NULL) | |
a60198e5 SM |
6025 | /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */ |
6026 | return (char *) gettext (re_error_msgid[(int) REG_ESPACE]); | |
fa9a63c5 RM |
6027 | } |
6028 | ||
6029 | /* Since `re_exec' always passes NULL for the `regs' argument, we | |
6030 | don't need to initialize the pattern buffer fields which affect it. */ | |
6031 | ||
fa9a63c5 | 6032 | ret = regex_compile (s, strlen (s), re_syntax_options, &re_comp_buf); |
5e69f11e | 6033 | |
fa9a63c5 RM |
6034 | if (!ret) |
6035 | return NULL; | |
6036 | ||
6037 | /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */ | |
6038 | return (char *) gettext (re_error_msgid[(int) ret]); | |
6039 | } | |
6040 | ||
6041 | ||
6042 | int | |
0b32bf0e | 6043 | # ifdef _LIBC |
48afdd44 | 6044 | weak_function |
0b32bf0e | 6045 | # endif |
fa9a63c5 RM |
6046 | re_exec (s) |
6047 | const char *s; | |
6048 | { | |
6049 | const int len = strlen (s); | |
6050 | return | |
6051 | 0 <= re_search (&re_comp_buf, s, len, 0, len, (struct re_registers *) 0); | |
6052 | } | |
6053 | #endif /* _REGEX_RE_COMP */ | |
6054 | \f | |
6055 | /* POSIX.2 functions. Don't define these for Emacs. */ | |
6056 | ||
6057 | #ifndef emacs | |
6058 | ||
6059 | /* regcomp takes a regular expression as a string and compiles it. | |
6060 | ||
b18215fc | 6061 | PREG is a regex_t *. We do not expect any fields to be initialized, |
fa9a63c5 RM |
6062 | since POSIX says we shouldn't. Thus, we set |
6063 | ||
6064 | `buffer' to the compiled pattern; | |
6065 | `used' to the length of the compiled pattern; | |
6066 | `syntax' to RE_SYNTAX_POSIX_EXTENDED if the | |
6067 | REG_EXTENDED bit in CFLAGS is set; otherwise, to | |
6068 | RE_SYNTAX_POSIX_BASIC; | |
c0f9ea08 SM |
6069 | `fastmap' to an allocated space for the fastmap; |
6070 | `fastmap_accurate' to zero; | |
fa9a63c5 RM |
6071 | `re_nsub' to the number of subexpressions in PATTERN. |
6072 | ||
6073 | PATTERN is the address of the pattern string. | |
6074 | ||
6075 | CFLAGS is a series of bits which affect compilation. | |
6076 | ||
6077 | If REG_EXTENDED is set, we use POSIX extended syntax; otherwise, we | |
6078 | use POSIX basic syntax. | |
6079 | ||
6080 | If REG_NEWLINE is set, then . and [^...] don't match newline. | |
6081 | Also, regexec will try a match beginning after every newline. | |
6082 | ||
6083 | If REG_ICASE is set, then we considers upper- and lowercase | |
6084 | versions of letters to be equivalent when matching. | |
6085 | ||
6086 | If REG_NOSUB is set, then when PREG is passed to regexec, that | |
6087 | routine will report only success or failure, and nothing about the | |
6088 | registers. | |
6089 | ||
b18215fc | 6090 | It returns 0 if it succeeds, nonzero if it doesn't. (See regex.h for |
fa9a63c5 RM |
6091 | the return codes and their meanings.) */ |
6092 | ||
6093 | int | |
6094 | regcomp (preg, pattern, cflags) | |
ada30c0e SM |
6095 | regex_t *__restrict preg; |
6096 | const char *__restrict pattern; | |
fa9a63c5 RM |
6097 | int cflags; |
6098 | { | |
6099 | reg_errcode_t ret; | |
4bb91c68 | 6100 | reg_syntax_t syntax |
fa9a63c5 RM |
6101 | = (cflags & REG_EXTENDED) ? |
6102 | RE_SYNTAX_POSIX_EXTENDED : RE_SYNTAX_POSIX_BASIC; | |
6103 | ||
6104 | /* regex_compile will allocate the space for the compiled pattern. */ | |
6105 | preg->buffer = 0; | |
6106 | preg->allocated = 0; | |
6107 | preg->used = 0; | |
5e69f11e | 6108 | |
c0f9ea08 SM |
6109 | /* Try to allocate space for the fastmap. */ |
6110 | preg->fastmap = (char *) malloc (1 << BYTEWIDTH); | |
5e69f11e | 6111 | |
fa9a63c5 RM |
6112 | if (cflags & REG_ICASE) |
6113 | { | |
6114 | unsigned i; | |
5e69f11e | 6115 | |
6676cb1c RS |
6116 | preg->translate |
6117 | = (RE_TRANSLATE_TYPE) malloc (CHAR_SET_SIZE | |
6118 | * sizeof (*(RE_TRANSLATE_TYPE)0)); | |
fa9a63c5 | 6119 | if (preg->translate == NULL) |
0b32bf0e | 6120 | return (int) REG_ESPACE; |
fa9a63c5 RM |
6121 | |
6122 | /* Map uppercase characters to corresponding lowercase ones. */ | |
6123 | for (i = 0; i < CHAR_SET_SIZE; i++) | |
4bb91c68 | 6124 | preg->translate[i] = ISUPPER (i) ? TOLOWER (i) : i; |
fa9a63c5 RM |
6125 | } |
6126 | else | |
6127 | preg->translate = NULL; | |
6128 | ||
6129 | /* If REG_NEWLINE is set, newlines are treated differently. */ | |
6130 | if (cflags & REG_NEWLINE) | |
6131 | { /* REG_NEWLINE implies neither . nor [^...] match newline. */ | |
6132 | syntax &= ~RE_DOT_NEWLINE; | |
6133 | syntax |= RE_HAT_LISTS_NOT_NEWLINE; | |
fa9a63c5 RM |
6134 | } |
6135 | else | |
c0f9ea08 | 6136 | syntax |= RE_NO_NEWLINE_ANCHOR; |
fa9a63c5 RM |
6137 | |
6138 | preg->no_sub = !!(cflags & REG_NOSUB); | |
6139 | ||
5e69f11e | 6140 | /* POSIX says a null character in the pattern terminates it, so we |
fa9a63c5 | 6141 | can use strlen here in compiling the pattern. */ |
4bb91c68 | 6142 | ret = regex_compile ((re_char*) pattern, strlen (pattern), syntax, preg); |
5e69f11e | 6143 | |
fa9a63c5 RM |
6144 | /* POSIX doesn't distinguish between an unmatched open-group and an |
6145 | unmatched close-group: both are REG_EPAREN. */ | |
c0f9ea08 SM |
6146 | if (ret == REG_ERPAREN) |
6147 | ret = REG_EPAREN; | |
6148 | ||
6149 | if (ret == REG_NOERROR && preg->fastmap) | |
6150 | { /* Compute the fastmap now, since regexec cannot modify the pattern | |
6151 | buffer. */ | |
6152 | re_compile_fastmap (preg); | |
6153 | if (preg->can_be_null) | |
6154 | { /* The fastmap can't be used anyway. */ | |
6155 | free (preg->fastmap); | |
6156 | preg->fastmap = NULL; | |
6157 | } | |
6158 | } | |
fa9a63c5 RM |
6159 | return (int) ret; |
6160 | } | |
c0f9ea08 | 6161 | WEAK_ALIAS (__regcomp, regcomp) |
fa9a63c5 RM |
6162 | |
6163 | ||
6164 | /* regexec searches for a given pattern, specified by PREG, in the | |
6165 | string STRING. | |
5e69f11e | 6166 | |
fa9a63c5 | 6167 | If NMATCH is zero or REG_NOSUB was set in the cflags argument to |
b18215fc | 6168 | `regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at |
fa9a63c5 RM |
6169 | least NMATCH elements, and we set them to the offsets of the |
6170 | corresponding matched substrings. | |
5e69f11e | 6171 | |
fa9a63c5 RM |
6172 | EFLAGS specifies `execution flags' which affect matching: if |
6173 | REG_NOTBOL is set, then ^ does not match at the beginning of the | |
6174 | string; if REG_NOTEOL is set, then $ does not match at the end. | |
5e69f11e | 6175 | |
fa9a63c5 RM |
6176 | We return 0 if we find a match and REG_NOMATCH if not. */ |
6177 | ||
6178 | int | |
6179 | regexec (preg, string, nmatch, pmatch, eflags) | |
ada30c0e SM |
6180 | const regex_t *__restrict preg; |
6181 | const char *__restrict string; | |
5e69f11e RM |
6182 | size_t nmatch; |
6183 | regmatch_t pmatch[]; | |
fa9a63c5 RM |
6184 | int eflags; |
6185 | { | |
6186 | int ret; | |
6187 | struct re_registers regs; | |
6188 | regex_t private_preg; | |
6189 | int len = strlen (string); | |
c0f9ea08 | 6190 | boolean want_reg_info = !preg->no_sub && nmatch > 0 && pmatch; |
fa9a63c5 RM |
6191 | |
6192 | private_preg = *preg; | |
5e69f11e | 6193 | |
fa9a63c5 RM |
6194 | private_preg.not_bol = !!(eflags & REG_NOTBOL); |
6195 | private_preg.not_eol = !!(eflags & REG_NOTEOL); | |
5e69f11e | 6196 | |
fa9a63c5 RM |
6197 | /* The user has told us exactly how many registers to return |
6198 | information about, via `nmatch'. We have to pass that on to the | |
b18215fc | 6199 | matching routines. */ |
fa9a63c5 | 6200 | private_preg.regs_allocated = REGS_FIXED; |
5e69f11e | 6201 | |
fa9a63c5 RM |
6202 | if (want_reg_info) |
6203 | { | |
6204 | regs.num_regs = nmatch; | |
4bb91c68 SM |
6205 | regs.start = TALLOC (nmatch * 2, regoff_t); |
6206 | if (regs.start == NULL) | |
0b32bf0e | 6207 | return (int) REG_NOMATCH; |
4bb91c68 | 6208 | regs.end = regs.start + nmatch; |
fa9a63c5 RM |
6209 | } |
6210 | ||
c0f9ea08 SM |
6211 | /* Instead of using not_eol to implement REG_NOTEOL, we could simply |
6212 | pass (&private_preg, string, len + 1, 0, len, ...) pretending the string | |
6213 | was a little bit longer but still only matching the real part. | |
6214 | This works because the `endline' will check for a '\n' and will find a | |
6215 | '\0', correctly deciding that this is not the end of a line. | |
6216 | But it doesn't work out so nicely for REG_NOTBOL, since we don't have | |
6217 | a convenient '\0' there. For all we know, the string could be preceded | |
6218 | by '\n' which would throw things off. */ | |
6219 | ||
fa9a63c5 RM |
6220 | /* Perform the searching operation. */ |
6221 | ret = re_search (&private_preg, string, len, | |
0b32bf0e SM |
6222 | /* start: */ 0, /* range: */ len, |
6223 | want_reg_info ? ®s : (struct re_registers *) 0); | |
5e69f11e | 6224 | |
fa9a63c5 RM |
6225 | /* Copy the register information to the POSIX structure. */ |
6226 | if (want_reg_info) | |
6227 | { | |
6228 | if (ret >= 0) | |
0b32bf0e SM |
6229 | { |
6230 | unsigned r; | |
fa9a63c5 | 6231 | |
0b32bf0e SM |
6232 | for (r = 0; r < nmatch; r++) |
6233 | { | |
6234 | pmatch[r].rm_so = regs.start[r]; | |
6235 | pmatch[r].rm_eo = regs.end[r]; | |
6236 | } | |
6237 | } | |
fa9a63c5 | 6238 | |
b18215fc | 6239 | /* If we needed the temporary register info, free the space now. */ |
fa9a63c5 | 6240 | free (regs.start); |
fa9a63c5 RM |
6241 | } |
6242 | ||
6243 | /* We want zero return to mean success, unlike `re_search'. */ | |
6244 | return ret >= 0 ? (int) REG_NOERROR : (int) REG_NOMATCH; | |
6245 | } | |
c0f9ea08 | 6246 | WEAK_ALIAS (__regexec, regexec) |
fa9a63c5 RM |
6247 | |
6248 | ||
6249 | /* Returns a message corresponding to an error code, ERRCODE, returned | |
6250 | from either regcomp or regexec. We don't use PREG here. */ | |
6251 | ||
6252 | size_t | |
6253 | regerror (errcode, preg, errbuf, errbuf_size) | |
6254 | int errcode; | |
6255 | const regex_t *preg; | |
6256 | char *errbuf; | |
6257 | size_t errbuf_size; | |
6258 | { | |
6259 | const char *msg; | |
6260 | size_t msg_size; | |
6261 | ||
6262 | if (errcode < 0 | |
6263 | || errcode >= (sizeof (re_error_msgid) / sizeof (re_error_msgid[0]))) | |
5e69f11e | 6264 | /* Only error codes returned by the rest of the code should be passed |
b18215fc | 6265 | to this routine. If we are given anything else, or if other regex |
fa9a63c5 RM |
6266 | code generates an invalid error code, then the program has a bug. |
6267 | Dump core so we can fix it. */ | |
6268 | abort (); | |
6269 | ||
6270 | msg = gettext (re_error_msgid[errcode]); | |
6271 | ||
6272 | msg_size = strlen (msg) + 1; /* Includes the null. */ | |
5e69f11e | 6273 | |
fa9a63c5 RM |
6274 | if (errbuf_size != 0) |
6275 | { | |
6276 | if (msg_size > errbuf_size) | |
0b32bf0e SM |
6277 | { |
6278 | strncpy (errbuf, msg, errbuf_size - 1); | |
6279 | errbuf[errbuf_size - 1] = 0; | |
6280 | } | |
fa9a63c5 | 6281 | else |
0b32bf0e | 6282 | strcpy (errbuf, msg); |
fa9a63c5 RM |
6283 | } |
6284 | ||
6285 | return msg_size; | |
6286 | } | |
c0f9ea08 | 6287 | WEAK_ALIAS (__regerror, regerror) |
fa9a63c5 RM |
6288 | |
6289 | ||
6290 | /* Free dynamically allocated space used by PREG. */ | |
6291 | ||
6292 | void | |
6293 | regfree (preg) | |
6294 | regex_t *preg; | |
6295 | { | |
6296 | if (preg->buffer != NULL) | |
6297 | free (preg->buffer); | |
6298 | preg->buffer = NULL; | |
5e69f11e | 6299 | |
fa9a63c5 RM |
6300 | preg->allocated = 0; |
6301 | preg->used = 0; | |
6302 | ||
6303 | if (preg->fastmap != NULL) | |
6304 | free (preg->fastmap); | |
6305 | preg->fastmap = NULL; | |
6306 | preg->fastmap_accurate = 0; | |
6307 | ||
6308 | if (preg->translate != NULL) | |
6309 | free (preg->translate); | |
6310 | preg->translate = NULL; | |
6311 | } | |
c0f9ea08 | 6312 | WEAK_ALIAS (__regfree, regfree) |
fa9a63c5 RM |
6313 | |
6314 | #endif /* not emacs */ |