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 | |
505bde11 | 22 | /* TODO: |
505bde11 | 23 | - structure the opcode space into opcode+flag. |
dc1e502d | 24 | - merge with glibc's regex.[ch]. |
01618498 | 25 | - replace (succeed_n + jump_n + set_number_at) with something that doesn't |
6dcf2d0e SM |
26 | need to modify the compiled regexp so that re_match can be reentrant. |
27 | - get rid of on_failure_jump_smart by doing the optimization in re_comp | |
28 | rather than at run-time, so that re_match can be reentrant. | |
01618498 | 29 | */ |
505bde11 | 30 | |
fa9a63c5 | 31 | /* AIX requires this to be the first thing in the file. */ |
0b32bf0e | 32 | #if defined _AIX && !defined REGEX_MALLOC |
fa9a63c5 RM |
33 | #pragma alloca |
34 | #endif | |
35 | ||
68d96f02 | 36 | #undef _GNU_SOURCE |
fa9a63c5 RM |
37 | #define _GNU_SOURCE |
38 | ||
39 | #ifdef HAVE_CONFIG_H | |
0b32bf0e | 40 | # include <config.h> |
fa9a63c5 RM |
41 | #endif |
42 | ||
4bb91c68 SM |
43 | #if defined STDC_HEADERS && !defined emacs |
44 | # include <stddef.h> | |
45 | #else | |
46 | /* We need this for `regex.h', and perhaps for the Emacs include files. */ | |
47 | # include <sys/types.h> | |
48 | #endif | |
fa9a63c5 | 49 | |
14473664 SM |
50 | /* Whether to use ISO C Amendment 1 wide char functions. |
51 | Those should not be used for Emacs since it uses its own. */ | |
52 | #define WIDE_CHAR_SUPPORT \ | |
a0ad02f7 | 53 | (defined _LIBC || HAVE_WCTYPE_H && HAVE_WCHAR_H && HAVE_BTOWC && !emacs) |
14473664 SM |
54 | |
55 | /* For platform which support the ISO C amendement 1 functionality we | |
56 | support user defined character classes. */ | |
a0ad02f7 | 57 | #if WIDE_CHAR_SUPPORT |
14473664 SM |
58 | /* Solaris 2.5 has a bug: <wchar.h> must be included before <wctype.h>. */ |
59 | # include <wchar.h> | |
60 | # include <wctype.h> | |
61 | #endif | |
62 | ||
c0f9ea08 SM |
63 | #ifdef _LIBC |
64 | /* We have to keep the namespace clean. */ | |
65 | # define regfree(preg) __regfree (preg) | |
66 | # define regexec(pr, st, nm, pm, ef) __regexec (pr, st, nm, pm, ef) | |
67 | # define regcomp(preg, pattern, cflags) __regcomp (preg, pattern, cflags) | |
68 | # define regerror(errcode, preg, errbuf, errbuf_size) \ | |
69 | __regerror(errcode, preg, errbuf, errbuf_size) | |
70 | # define re_set_registers(bu, re, nu, st, en) \ | |
71 | __re_set_registers (bu, re, nu, st, en) | |
72 | # define re_match_2(bufp, string1, size1, string2, size2, pos, regs, stop) \ | |
73 | __re_match_2 (bufp, string1, size1, string2, size2, pos, regs, stop) | |
74 | # define re_match(bufp, string, size, pos, regs) \ | |
75 | __re_match (bufp, string, size, pos, regs) | |
76 | # define re_search(bufp, string, size, startpos, range, regs) \ | |
77 | __re_search (bufp, string, size, startpos, range, regs) | |
78 | # define re_compile_pattern(pattern, length, bufp) \ | |
79 | __re_compile_pattern (pattern, length, bufp) | |
80 | # define re_set_syntax(syntax) __re_set_syntax (syntax) | |
81 | # define re_search_2(bufp, st1, s1, st2, s2, startpos, range, regs, stop) \ | |
82 | __re_search_2 (bufp, st1, s1, st2, s2, startpos, range, regs, stop) | |
83 | # define re_compile_fastmap(bufp) __re_compile_fastmap (bufp) | |
84 | ||
14473664 SM |
85 | /* Make sure we call libc's function even if the user overrides them. */ |
86 | # define btowc __btowc | |
87 | # define iswctype __iswctype | |
88 | # define wctype __wctype | |
89 | ||
c0f9ea08 SM |
90 | # define WEAK_ALIAS(a,b) weak_alias (a, b) |
91 | ||
92 | /* We are also using some library internals. */ | |
93 | # include <locale/localeinfo.h> | |
94 | # include <locale/elem-hash.h> | |
95 | # include <langinfo.h> | |
96 | #else | |
97 | # define WEAK_ALIAS(a,b) | |
98 | #endif | |
99 | ||
4bb91c68 | 100 | /* This is for other GNU distributions with internationalized messages. */ |
0b32bf0e | 101 | #if HAVE_LIBINTL_H || defined _LIBC |
fa9a63c5 RM |
102 | # include <libintl.h> |
103 | #else | |
104 | # define gettext(msgid) (msgid) | |
105 | #endif | |
106 | ||
5e69f11e RM |
107 | #ifndef gettext_noop |
108 | /* This define is so xgettext can find the internationalizable | |
109 | strings. */ | |
0b32bf0e | 110 | # define gettext_noop(String) String |
5e69f11e RM |
111 | #endif |
112 | ||
fa9a63c5 RM |
113 | /* The `emacs' switch turns on certain matching commands |
114 | that make sense only in Emacs. */ | |
115 | #ifdef emacs | |
116 | ||
0b32bf0e SM |
117 | # include "lisp.h" |
118 | # include "buffer.h" | |
b18215fc RS |
119 | |
120 | /* Make syntax table lookup grant data in gl_state. */ | |
0b32bf0e | 121 | # define SYNTAX_ENTRY_VIA_PROPERTY |
b18215fc | 122 | |
0b32bf0e SM |
123 | # include "syntax.h" |
124 | # include "charset.h" | |
125 | # include "category.h" | |
fa9a63c5 | 126 | |
0b32bf0e SM |
127 | # define malloc xmalloc |
128 | # define realloc xrealloc | |
129 | # define free xfree | |
9abbd165 | 130 | |
0b32bf0e SM |
131 | /* Converts the pointer to the char to BEG-based offset from the start. */ |
132 | # define PTR_TO_OFFSET(d) POS_AS_IN_BUFFER (POINTER_TO_OFFSET (d)) | |
133 | # define POS_AS_IN_BUFFER(p) ((p) + (NILP (re_match_object) || BUFFERP (re_match_object))) | |
134 | ||
135 | # define RE_MULTIBYTE_P(bufp) ((bufp)->multibyte) | |
136 | # define RE_STRING_CHAR(p, s) \ | |
4e8a9132 | 137 | (multibyte ? (STRING_CHAR (p, s)) : (*(p))) |
0b32bf0e | 138 | # define RE_STRING_CHAR_AND_LENGTH(p, s, len) \ |
2d1675e4 SM |
139 | (multibyte ? (STRING_CHAR_AND_LENGTH (p, s, len)) : ((len) = 1, *(p))) |
140 | ||
141 | /* Set C a (possibly multibyte) character before P. P points into a | |
142 | string which is the virtual concatenation of STR1 (which ends at | |
143 | END1) or STR2 (which ends at END2). */ | |
0b32bf0e | 144 | # define GET_CHAR_BEFORE_2(c, p, str1, end1, str2, end2) \ |
2d1675e4 SM |
145 | do { \ |
146 | if (multibyte) \ | |
147 | { \ | |
0b32bf0e SM |
148 | re_char *dtemp = (p) == (str2) ? (end1) : (p); \ |
149 | re_char *dlimit = ((p) > (str2) && (p) <= (end2)) ? (str2) : (str1); \ | |
150 | while (dtemp-- > dlimit && !CHAR_HEAD_P (*dtemp)); \ | |
151 | c = STRING_CHAR (dtemp, (p) - dtemp); \ | |
2d1675e4 SM |
152 | } \ |
153 | else \ | |
154 | (c = ((p) == (str2) ? (end1) : (p))[-1]); \ | |
155 | } while (0) | |
156 | ||
4e8a9132 | 157 | |
fa9a63c5 RM |
158 | #else /* not emacs */ |
159 | ||
160 | /* If we are not linking with Emacs proper, | |
161 | we can't use the relocating allocator | |
162 | even if config.h says that we can. */ | |
0b32bf0e | 163 | # undef REL_ALLOC |
fa9a63c5 | 164 | |
0b32bf0e SM |
165 | # if defined STDC_HEADERS || defined _LIBC |
166 | # include <stdlib.h> | |
167 | # else | |
fa9a63c5 RM |
168 | char *malloc (); |
169 | char *realloc (); | |
0b32bf0e | 170 | # endif |
fa9a63c5 | 171 | |
9e4ecb26 | 172 | /* When used in Emacs's lib-src, we need to get bzero and bcopy somehow. |
4bb91c68 | 173 | If nothing else has been done, use the method below. */ |
0b32bf0e SM |
174 | # ifdef INHIBIT_STRING_HEADER |
175 | # if !(defined HAVE_BZERO && defined HAVE_BCOPY) | |
176 | # if !defined bzero && !defined bcopy | |
177 | # undef INHIBIT_STRING_HEADER | |
178 | # endif | |
179 | # endif | |
180 | # endif | |
9e4ecb26 | 181 | |
4bb91c68 | 182 | /* This is the normal way of making sure we have memcpy, memcmp and bzero. |
9e4ecb26 KH |
183 | This is used in most programs--a few other programs avoid this |
184 | by defining INHIBIT_STRING_HEADER. */ | |
0b32bf0e SM |
185 | # ifndef INHIBIT_STRING_HEADER |
186 | # if defined HAVE_STRING_H || defined STDC_HEADERS || defined _LIBC | |
187 | # include <string.h> | |
0b32bf0e | 188 | # ifndef bzero |
4bb91c68 SM |
189 | # ifndef _LIBC |
190 | # define bzero(s, n) (memset (s, '\0', n), (s)) | |
191 | # else | |
192 | # define bzero(s, n) __bzero (s, n) | |
193 | # endif | |
0b32bf0e SM |
194 | # endif |
195 | # else | |
196 | # include <strings.h> | |
4bb91c68 SM |
197 | # ifndef memcmp |
198 | # define memcmp(s1, s2, n) bcmp (s1, s2, n) | |
199 | # endif | |
200 | # ifndef memcpy | |
201 | # define memcpy(d, s, n) (bcopy (s, d, n), (d)) | |
202 | # endif | |
0b32bf0e SM |
203 | # endif |
204 | # endif | |
fa9a63c5 RM |
205 | |
206 | /* Define the syntax stuff for \<, \>, etc. */ | |
207 | ||
990b2375 SM |
208 | /* Sword must be nonzero for the wordchar pattern commands in re_match_2. */ |
209 | enum syntaxcode { Swhitespace = 0, Sword = 1 }; | |
fa9a63c5 | 210 | |
0b32bf0e SM |
211 | # ifdef SWITCH_ENUM_BUG |
212 | # define SWITCH_ENUM_CAST(x) ((int)(x)) | |
213 | # else | |
214 | # define SWITCH_ENUM_CAST(x) (x) | |
215 | # endif | |
fa9a63c5 | 216 | |
e934739e | 217 | /* Dummy macros for non-Emacs environments. */ |
0b32bf0e SM |
218 | # define BASE_LEADING_CODE_P(c) (0) |
219 | # define CHAR_CHARSET(c) 0 | |
220 | # define CHARSET_LEADING_CODE_BASE(c) 0 | |
221 | # define MAX_MULTIBYTE_LENGTH 1 | |
222 | # define RE_MULTIBYTE_P(x) 0 | |
223 | # define WORD_BOUNDARY_P(c1, c2) (0) | |
224 | # define CHAR_HEAD_P(p) (1) | |
225 | # define SINGLE_BYTE_CHAR_P(c) (1) | |
226 | # define SAME_CHARSET_P(c1, c2) (1) | |
227 | # define MULTIBYTE_FORM_LENGTH(p, s) (1) | |
228 | # define STRING_CHAR(p, s) (*(p)) | |
229 | # define RE_STRING_CHAR STRING_CHAR | |
230 | # define CHAR_STRING(c, s) (*(s) = (c), 1) | |
231 | # define STRING_CHAR_AND_LENGTH(p, s, actual_len) ((actual_len) = 1, *(p)) | |
232 | # define RE_STRING_CHAR_AND_LENGTH STRING_CHAR_AND_LENGTH | |
233 | # define GET_CHAR_BEFORE_2(c, p, str1, end1, str2, end2) \ | |
b18215fc | 234 | (c = ((p) == (str2) ? *((end1) - 1) : *((p) - 1))) |
0b32bf0e | 235 | # define MAKE_CHAR(charset, c1, c2) (c1) |
fa9a63c5 | 236 | #endif /* not emacs */ |
4e8a9132 SM |
237 | |
238 | #ifndef RE_TRANSLATE | |
0b32bf0e SM |
239 | # define RE_TRANSLATE(TBL, C) ((unsigned char)(TBL)[C]) |
240 | # define RE_TRANSLATE_P(TBL) (TBL) | |
4e8a9132 | 241 | #endif |
fa9a63c5 RM |
242 | \f |
243 | /* Get the interface, including the syntax bits. */ | |
244 | #include "regex.h" | |
245 | ||
f71b19b6 DL |
246 | /* isalpha etc. are used for the character classes. */ |
247 | #include <ctype.h> | |
fa9a63c5 | 248 | |
f71b19b6 | 249 | #ifdef emacs |
fa9a63c5 | 250 | |
f71b19b6 | 251 | /* 1 if C is an ASCII character. */ |
0b32bf0e | 252 | # define IS_REAL_ASCII(c) ((c) < 0200) |
fa9a63c5 | 253 | |
f71b19b6 | 254 | /* 1 if C is a unibyte character. */ |
0b32bf0e | 255 | # define ISUNIBYTE(c) (SINGLE_BYTE_CHAR_P ((c))) |
96cc36cc | 256 | |
f71b19b6 | 257 | /* The Emacs definitions should not be directly affected by locales. */ |
96cc36cc | 258 | |
f71b19b6 | 259 | /* In Emacs, these are only used for single-byte characters. */ |
0b32bf0e SM |
260 | # define ISDIGIT(c) ((c) >= '0' && (c) <= '9') |
261 | # define ISCNTRL(c) ((c) < ' ') | |
262 | # define ISXDIGIT(c) (((c) >= '0' && (c) <= '9') \ | |
f71b19b6 DL |
263 | || ((c) >= 'a' && (c) <= 'f') \ |
264 | || ((c) >= 'A' && (c) <= 'F')) | |
96cc36cc RS |
265 | |
266 | /* This is only used for single-byte characters. */ | |
0b32bf0e | 267 | # define ISBLANK(c) ((c) == ' ' || (c) == '\t') |
96cc36cc RS |
268 | |
269 | /* The rest must handle multibyte characters. */ | |
270 | ||
0b32bf0e | 271 | # define ISGRAPH(c) (SINGLE_BYTE_CHAR_P (c) \ |
f71b19b6 | 272 | ? (c) > ' ' && !((c) >= 0177 && (c) <= 0237) \ |
96cc36cc RS |
273 | : 1) |
274 | ||
14473664 | 275 | # define ISPRINT(c) (SINGLE_BYTE_CHAR_P (c) \ |
f71b19b6 | 276 | ? (c) >= ' ' && !((c) >= 0177 && (c) <= 0237) \ |
96cc36cc RS |
277 | : 1) |
278 | ||
0b32bf0e | 279 | # define ISALNUM(c) (IS_REAL_ASCII (c) \ |
f71b19b6 DL |
280 | ? (((c) >= 'a' && (c) <= 'z') \ |
281 | || ((c) >= 'A' && (c) <= 'Z') \ | |
282 | || ((c) >= '0' && (c) <= '9')) \ | |
96cc36cc RS |
283 | : SYNTAX (c) == Sword) |
284 | ||
0b32bf0e | 285 | # define ISALPHA(c) (IS_REAL_ASCII (c) \ |
f71b19b6 DL |
286 | ? (((c) >= 'a' && (c) <= 'z') \ |
287 | || ((c) >= 'A' && (c) <= 'Z')) \ | |
96cc36cc RS |
288 | : SYNTAX (c) == Sword) |
289 | ||
0b32bf0e | 290 | # define ISLOWER(c) (LOWERCASEP (c)) |
96cc36cc | 291 | |
0b32bf0e | 292 | # define ISPUNCT(c) (IS_REAL_ASCII (c) \ |
f71b19b6 DL |
293 | ? ((c) > ' ' && (c) < 0177 \ |
294 | && !(((c) >= 'a' && (c) <= 'z') \ | |
4bb91c68 SM |
295 | || ((c) >= 'A' && (c) <= 'Z') \ |
296 | || ((c) >= '0' && (c) <= '9'))) \ | |
96cc36cc RS |
297 | : SYNTAX (c) != Sword) |
298 | ||
0b32bf0e | 299 | # define ISSPACE(c) (SYNTAX (c) == Swhitespace) |
96cc36cc | 300 | |
0b32bf0e | 301 | # define ISUPPER(c) (UPPERCASEP (c)) |
96cc36cc | 302 | |
0b32bf0e | 303 | # define ISWORD(c) (SYNTAX (c) == Sword) |
96cc36cc RS |
304 | |
305 | #else /* not emacs */ | |
306 | ||
f71b19b6 DL |
307 | /* Jim Meyering writes: |
308 | ||
309 | "... Some ctype macros are valid only for character codes that | |
310 | isascii says are ASCII (SGI's IRIX-4.0.5 is one such system --when | |
311 | using /bin/cc or gcc but without giving an ansi option). So, all | |
4bb91c68 | 312 | ctype uses should be through macros like ISPRINT... If |
f71b19b6 DL |
313 | STDC_HEADERS is defined, then autoconf has verified that the ctype |
314 | macros don't need to be guarded with references to isascii. ... | |
315 | Defining isascii to 1 should let any compiler worth its salt | |
4bb91c68 SM |
316 | eliminate the && through constant folding." |
317 | Solaris defines some of these symbols so we must undefine them first. */ | |
f71b19b6 | 318 | |
4bb91c68 | 319 | # undef ISASCII |
0b32bf0e SM |
320 | # if defined STDC_HEADERS || (!defined isascii && !defined HAVE_ISASCII) |
321 | # define ISASCII(c) 1 | |
322 | # else | |
323 | # define ISASCII(c) isascii(c) | |
324 | # endif | |
f71b19b6 DL |
325 | |
326 | /* 1 if C is an ASCII character. */ | |
0b32bf0e | 327 | # define IS_REAL_ASCII(c) ((c) < 0200) |
f71b19b6 DL |
328 | |
329 | /* This distinction is not meaningful, except in Emacs. */ | |
0b32bf0e SM |
330 | # define ISUNIBYTE(c) 1 |
331 | ||
332 | # ifdef isblank | |
333 | # define ISBLANK(c) (ISASCII (c) && isblank (c)) | |
334 | # else | |
335 | # define ISBLANK(c) ((c) == ' ' || (c) == '\t') | |
336 | # endif | |
337 | # ifdef isgraph | |
338 | # define ISGRAPH(c) (ISASCII (c) && isgraph (c)) | |
339 | # else | |
340 | # define ISGRAPH(c) (ISASCII (c) && isprint (c) && !isspace (c)) | |
341 | # endif | |
342 | ||
4bb91c68 | 343 | # undef ISPRINT |
0b32bf0e SM |
344 | # define ISPRINT(c) (ISASCII (c) && isprint (c)) |
345 | # define ISDIGIT(c) (ISASCII (c) && isdigit (c)) | |
346 | # define ISALNUM(c) (ISASCII (c) && isalnum (c)) | |
347 | # define ISALPHA(c) (ISASCII (c) && isalpha (c)) | |
348 | # define ISCNTRL(c) (ISASCII (c) && iscntrl (c)) | |
349 | # define ISLOWER(c) (ISASCII (c) && islower (c)) | |
350 | # define ISPUNCT(c) (ISASCII (c) && ispunct (c)) | |
351 | # define ISSPACE(c) (ISASCII (c) && isspace (c)) | |
352 | # define ISUPPER(c) (ISASCII (c) && isupper (c)) | |
353 | # define ISXDIGIT(c) (ISASCII (c) && isxdigit (c)) | |
354 | ||
355 | # define ISWORD(c) ISALPHA(c) | |
356 | ||
4bb91c68 SM |
357 | # ifdef _tolower |
358 | # define TOLOWER(c) _tolower(c) | |
359 | # else | |
360 | # define TOLOWER(c) tolower(c) | |
361 | # endif | |
362 | ||
363 | /* How many characters in the character set. */ | |
364 | # define CHAR_SET_SIZE 256 | |
365 | ||
0b32bf0e | 366 | # ifdef SYNTAX_TABLE |
f71b19b6 | 367 | |
0b32bf0e | 368 | extern char *re_syntax_table; |
f71b19b6 | 369 | |
0b32bf0e SM |
370 | # else /* not SYNTAX_TABLE */ |
371 | ||
0b32bf0e SM |
372 | static char re_syntax_table[CHAR_SET_SIZE]; |
373 | ||
374 | static void | |
375 | init_syntax_once () | |
376 | { | |
377 | register int c; | |
378 | static int done = 0; | |
379 | ||
380 | if (done) | |
381 | return; | |
382 | ||
383 | bzero (re_syntax_table, sizeof re_syntax_table); | |
384 | ||
4bb91c68 SM |
385 | for (c = 0; c < CHAR_SET_SIZE; ++c) |
386 | if (ISALNUM (c)) | |
387 | re_syntax_table[c] = Sword; | |
fa9a63c5 | 388 | |
0b32bf0e | 389 | re_syntax_table['_'] = Sword; |
fa9a63c5 | 390 | |
0b32bf0e SM |
391 | done = 1; |
392 | } | |
393 | ||
394 | # endif /* not SYNTAX_TABLE */ | |
96cc36cc | 395 | |
4bb91c68 SM |
396 | # define SYNTAX(c) re_syntax_table[(c)] |
397 | ||
96cc36cc RS |
398 | #endif /* not emacs */ |
399 | \f | |
fa9a63c5 | 400 | #ifndef NULL |
0b32bf0e | 401 | # define NULL (void *)0 |
fa9a63c5 RM |
402 | #endif |
403 | ||
404 | /* We remove any previous definition of `SIGN_EXTEND_CHAR', | |
405 | since ours (we hope) works properly with all combinations of | |
406 | machines, compilers, `char' and `unsigned char' argument types. | |
4bb91c68 | 407 | (Per Bothner suggested the basic approach.) */ |
fa9a63c5 RM |
408 | #undef SIGN_EXTEND_CHAR |
409 | #if __STDC__ | |
0b32bf0e | 410 | # define SIGN_EXTEND_CHAR(c) ((signed char) (c)) |
fa9a63c5 RM |
411 | #else /* not __STDC__ */ |
412 | /* As in Harbison and Steele. */ | |
0b32bf0e | 413 | # define SIGN_EXTEND_CHAR(c) ((((unsigned char) (c)) ^ 128) - 128) |
fa9a63c5 RM |
414 | #endif |
415 | \f | |
416 | /* Should we use malloc or alloca? If REGEX_MALLOC is not defined, we | |
417 | use `alloca' instead of `malloc'. This is because using malloc in | |
418 | re_search* or re_match* could cause memory leaks when C-g is used in | |
419 | Emacs; also, malloc is slower and causes storage fragmentation. On | |
5e69f11e RM |
420 | the other hand, malloc is more portable, and easier to debug. |
421 | ||
fa9a63c5 RM |
422 | Because we sometimes use alloca, some routines have to be macros, |
423 | not functions -- `alloca'-allocated space disappears at the end of the | |
424 | function it is called in. */ | |
425 | ||
426 | #ifdef REGEX_MALLOC | |
427 | ||
0b32bf0e SM |
428 | # define REGEX_ALLOCATE malloc |
429 | # define REGEX_REALLOCATE(source, osize, nsize) realloc (source, nsize) | |
430 | # define REGEX_FREE free | |
fa9a63c5 RM |
431 | |
432 | #else /* not REGEX_MALLOC */ | |
433 | ||
434 | /* Emacs already defines alloca, sometimes. */ | |
0b32bf0e | 435 | # ifndef alloca |
fa9a63c5 RM |
436 | |
437 | /* Make alloca work the best possible way. */ | |
0b32bf0e SM |
438 | # ifdef __GNUC__ |
439 | # define alloca __builtin_alloca | |
440 | # else /* not __GNUC__ */ | |
441 | # if HAVE_ALLOCA_H | |
442 | # include <alloca.h> | |
443 | # endif /* HAVE_ALLOCA_H */ | |
444 | # endif /* not __GNUC__ */ | |
fa9a63c5 | 445 | |
0b32bf0e | 446 | # endif /* not alloca */ |
fa9a63c5 | 447 | |
0b32bf0e | 448 | # define REGEX_ALLOCATE alloca |
fa9a63c5 RM |
449 | |
450 | /* Assumes a `char *destination' variable. */ | |
0b32bf0e | 451 | # define REGEX_REALLOCATE(source, osize, nsize) \ |
fa9a63c5 | 452 | (destination = (char *) alloca (nsize), \ |
4bb91c68 | 453 | memcpy (destination, source, osize)) |
fa9a63c5 RM |
454 | |
455 | /* No need to do anything to free, after alloca. */ | |
0b32bf0e | 456 | # define REGEX_FREE(arg) ((void)0) /* Do nothing! But inhibit gcc warning. */ |
fa9a63c5 RM |
457 | |
458 | #endif /* not REGEX_MALLOC */ | |
459 | ||
460 | /* Define how to allocate the failure stack. */ | |
461 | ||
0b32bf0e | 462 | #if defined REL_ALLOC && defined REGEX_MALLOC |
4297555e | 463 | |
0b32bf0e | 464 | # define REGEX_ALLOCATE_STACK(size) \ |
fa9a63c5 | 465 | r_alloc (&failure_stack_ptr, (size)) |
0b32bf0e | 466 | # define REGEX_REALLOCATE_STACK(source, osize, nsize) \ |
fa9a63c5 | 467 | r_re_alloc (&failure_stack_ptr, (nsize)) |
0b32bf0e | 468 | # define REGEX_FREE_STACK(ptr) \ |
fa9a63c5 RM |
469 | r_alloc_free (&failure_stack_ptr) |
470 | ||
4297555e | 471 | #else /* not using relocating allocator */ |
fa9a63c5 | 472 | |
0b32bf0e | 473 | # ifdef REGEX_MALLOC |
fa9a63c5 | 474 | |
0b32bf0e SM |
475 | # define REGEX_ALLOCATE_STACK malloc |
476 | # define REGEX_REALLOCATE_STACK(source, osize, nsize) realloc (source, nsize) | |
477 | # define REGEX_FREE_STACK free | |
fa9a63c5 | 478 | |
0b32bf0e | 479 | # else /* not REGEX_MALLOC */ |
fa9a63c5 | 480 | |
0b32bf0e | 481 | # define REGEX_ALLOCATE_STACK alloca |
fa9a63c5 | 482 | |
0b32bf0e | 483 | # define REGEX_REALLOCATE_STACK(source, osize, nsize) \ |
fa9a63c5 | 484 | REGEX_REALLOCATE (source, osize, nsize) |
25fe55af | 485 | /* No need to explicitly free anything. */ |
0b32bf0e | 486 | # define REGEX_FREE_STACK(arg) ((void)0) |
fa9a63c5 | 487 | |
0b32bf0e | 488 | # endif /* not REGEX_MALLOC */ |
4297555e | 489 | #endif /* not using relocating allocator */ |
fa9a63c5 RM |
490 | |
491 | ||
492 | /* True if `size1' is non-NULL and PTR is pointing anywhere inside | |
493 | `string1' or just past its end. This works if PTR is NULL, which is | |
494 | a good thing. */ | |
25fe55af | 495 | #define FIRST_STRING_P(ptr) \ |
fa9a63c5 RM |
496 | (size1 && string1 <= (ptr) && (ptr) <= string1 + size1) |
497 | ||
498 | /* (Re)Allocate N items of type T using malloc, or fail. */ | |
499 | #define TALLOC(n, t) ((t *) malloc ((n) * sizeof (t))) | |
500 | #define RETALLOC(addr, n, t) ((addr) = (t *) realloc (addr, (n) * sizeof (t))) | |
501 | #define RETALLOC_IF(addr, n, t) \ | |
502 | if (addr) RETALLOC((addr), (n), t); else (addr) = TALLOC ((n), t) | |
503 | #define REGEX_TALLOC(n, t) ((t *) REGEX_ALLOCATE ((n) * sizeof (t))) | |
504 | ||
4bb91c68 | 505 | #define BYTEWIDTH 8 /* In bits. */ |
fa9a63c5 RM |
506 | |
507 | #define STREQ(s1, s2) ((strcmp (s1, s2) == 0)) | |
508 | ||
509 | #undef MAX | |
510 | #undef MIN | |
511 | #define MAX(a, b) ((a) > (b) ? (a) : (b)) | |
512 | #define MIN(a, b) ((a) < (b) ? (a) : (b)) | |
513 | ||
66f0296e SM |
514 | /* Type of source-pattern and string chars. */ |
515 | typedef const unsigned char re_char; | |
516 | ||
fa9a63c5 RM |
517 | typedef char boolean; |
518 | #define false 0 | |
519 | #define true 1 | |
520 | ||
4bb91c68 SM |
521 | static int re_match_2_internal _RE_ARGS ((struct re_pattern_buffer *bufp, |
522 | re_char *string1, int size1, | |
523 | re_char *string2, int size2, | |
524 | int pos, | |
525 | struct re_registers *regs, | |
526 | int stop)); | |
fa9a63c5 RM |
527 | \f |
528 | /* These are the command codes that appear in compiled regular | |
4bb91c68 | 529 | expressions. Some opcodes are followed by argument bytes. A |
fa9a63c5 RM |
530 | command code can specify any interpretation whatsoever for its |
531 | arguments. Zero bytes may appear in the compiled regular expression. */ | |
532 | ||
533 | typedef enum | |
534 | { | |
535 | no_op = 0, | |
536 | ||
4bb91c68 | 537 | /* Succeed right away--no more backtracking. */ |
fa9a63c5 RM |
538 | succeed, |
539 | ||
25fe55af | 540 | /* Followed by one byte giving n, then by n literal bytes. */ |
fa9a63c5 RM |
541 | exactn, |
542 | ||
25fe55af | 543 | /* Matches any (more or less) character. */ |
fa9a63c5 RM |
544 | anychar, |
545 | ||
25fe55af RS |
546 | /* Matches any one char belonging to specified set. First |
547 | following byte is number of bitmap bytes. Then come bytes | |
548 | for a bitmap saying which chars are in. Bits in each byte | |
549 | are ordered low-bit-first. A character is in the set if its | |
550 | bit is 1. A character too large to have a bit in the map is | |
96cc36cc RS |
551 | automatically not in the set. |
552 | ||
553 | If the length byte has the 0x80 bit set, then that stuff | |
554 | is followed by a range table: | |
555 | 2 bytes of flags for character sets (low 8 bits, high 8 bits) | |
0b32bf0e | 556 | See RANGE_TABLE_WORK_BITS below. |
01618498 | 557 | 2 bytes, the number of pairs that follow (upto 32767) |
96cc36cc | 558 | pairs, each 2 multibyte characters, |
0b32bf0e | 559 | each multibyte character represented as 3 bytes. */ |
fa9a63c5 RM |
560 | charset, |
561 | ||
25fe55af | 562 | /* Same parameters as charset, but match any character that is |
4bb91c68 | 563 | not one of those specified. */ |
fa9a63c5 RM |
564 | charset_not, |
565 | ||
25fe55af RS |
566 | /* Start remembering the text that is matched, for storing in a |
567 | register. Followed by one byte with the register number, in | |
568 | the range 0 to one less than the pattern buffer's re_nsub | |
505bde11 | 569 | field. */ |
fa9a63c5 RM |
570 | start_memory, |
571 | ||
25fe55af RS |
572 | /* Stop remembering the text that is matched and store it in a |
573 | memory register. Followed by one byte with the register | |
574 | number, in the range 0 to one less than `re_nsub' in the | |
505bde11 | 575 | pattern buffer. */ |
fa9a63c5 RM |
576 | stop_memory, |
577 | ||
25fe55af | 578 | /* Match a duplicate of something remembered. Followed by one |
4bb91c68 | 579 | byte containing the register number. */ |
fa9a63c5 RM |
580 | duplicate, |
581 | ||
25fe55af | 582 | /* Fail unless at beginning of line. */ |
fa9a63c5 RM |
583 | begline, |
584 | ||
4bb91c68 | 585 | /* Fail unless at end of line. */ |
fa9a63c5 RM |
586 | endline, |
587 | ||
25fe55af RS |
588 | /* Succeeds if at beginning of buffer (if emacs) or at beginning |
589 | of string to be matched (if not). */ | |
fa9a63c5 RM |
590 | begbuf, |
591 | ||
25fe55af | 592 | /* Analogously, for end of buffer/string. */ |
fa9a63c5 | 593 | endbuf, |
5e69f11e | 594 | |
25fe55af | 595 | /* Followed by two byte relative address to which to jump. */ |
5e69f11e | 596 | jump, |
fa9a63c5 | 597 | |
25fe55af RS |
598 | /* Followed by two-byte relative address of place to resume at |
599 | in case of failure. */ | |
fa9a63c5 | 600 | on_failure_jump, |
5e69f11e | 601 | |
25fe55af RS |
602 | /* Like on_failure_jump, but pushes a placeholder instead of the |
603 | current string position when executed. */ | |
fa9a63c5 | 604 | on_failure_keep_string_jump, |
5e69f11e | 605 | |
505bde11 SM |
606 | /* Just like `on_failure_jump', except that it checks that we |
607 | don't get stuck in an infinite loop (matching an empty string | |
608 | indefinitely). */ | |
609 | on_failure_jump_loop, | |
610 | ||
0683b6fa SM |
611 | /* Just like `on_failure_jump_loop', except that it checks for |
612 | a different kind of loop (the kind that shows up with non-greedy | |
613 | operators). This operation has to be immediately preceded | |
614 | by a `no_op'. */ | |
615 | on_failure_jump_nastyloop, | |
616 | ||
0b32bf0e | 617 | /* A smart `on_failure_jump' used for greedy * and + operators. |
505bde11 SM |
618 | It analyses the loop before which it is put and if the |
619 | loop does not require backtracking, it changes itself to | |
4e8a9132 SM |
620 | `on_failure_keep_string_jump' and short-circuits the loop, |
621 | else it just defaults to changing itself into `on_failure_jump'. | |
622 | It assumes that it is pointing to just past a `jump'. */ | |
505bde11 | 623 | on_failure_jump_smart, |
fa9a63c5 | 624 | |
25fe55af | 625 | /* Followed by two-byte relative address and two-byte number n. |
ed0767d8 SM |
626 | After matching N times, jump to the address upon failure. |
627 | Does not work if N starts at 0: use on_failure_jump_loop | |
628 | instead. */ | |
fa9a63c5 RM |
629 | succeed_n, |
630 | ||
25fe55af RS |
631 | /* Followed by two-byte relative address, and two-byte number n. |
632 | Jump to the address N times, then fail. */ | |
fa9a63c5 RM |
633 | jump_n, |
634 | ||
25fe55af RS |
635 | /* Set the following two-byte relative address to the |
636 | subsequent two-byte number. The address *includes* the two | |
637 | bytes of number. */ | |
fa9a63c5 RM |
638 | set_number_at, |
639 | ||
fa9a63c5 RM |
640 | wordbeg, /* Succeeds if at word beginning. */ |
641 | wordend, /* Succeeds if at word end. */ | |
642 | ||
643 | wordbound, /* Succeeds if at a word boundary. */ | |
1fb352e0 | 644 | notwordbound, /* Succeeds if not at a word boundary. */ |
fa9a63c5 RM |
645 | |
646 | /* Matches any character whose syntax is specified. Followed by | |
25fe55af | 647 | a byte which contains a syntax code, e.g., Sword. */ |
fa9a63c5 RM |
648 | syntaxspec, |
649 | ||
650 | /* Matches any character whose syntax is not that specified. */ | |
1fb352e0 SM |
651 | notsyntaxspec |
652 | ||
653 | #ifdef emacs | |
654 | ,before_dot, /* Succeeds if before point. */ | |
655 | at_dot, /* Succeeds if at point. */ | |
656 | after_dot, /* Succeeds if after point. */ | |
b18215fc RS |
657 | |
658 | /* Matches any character whose category-set contains the specified | |
25fe55af RS |
659 | category. The operator is followed by a byte which contains a |
660 | category code (mnemonic ASCII character). */ | |
b18215fc RS |
661 | categoryspec, |
662 | ||
663 | /* Matches any character whose category-set does not contain the | |
664 | specified category. The operator is followed by a byte which | |
665 | contains the category code (mnemonic ASCII character). */ | |
666 | notcategoryspec | |
fa9a63c5 RM |
667 | #endif /* emacs */ |
668 | } re_opcode_t; | |
669 | \f | |
670 | /* Common operations on the compiled pattern. */ | |
671 | ||
672 | /* Store NUMBER in two contiguous bytes starting at DESTINATION. */ | |
673 | ||
674 | #define STORE_NUMBER(destination, number) \ | |
675 | do { \ | |
676 | (destination)[0] = (number) & 0377; \ | |
677 | (destination)[1] = (number) >> 8; \ | |
678 | } while (0) | |
679 | ||
680 | /* Same as STORE_NUMBER, except increment DESTINATION to | |
681 | the byte after where the number is stored. Therefore, DESTINATION | |
682 | must be an lvalue. */ | |
683 | ||
684 | #define STORE_NUMBER_AND_INCR(destination, number) \ | |
685 | do { \ | |
686 | STORE_NUMBER (destination, number); \ | |
687 | (destination) += 2; \ | |
688 | } while (0) | |
689 | ||
690 | /* Put into DESTINATION a number stored in two contiguous bytes starting | |
691 | at SOURCE. */ | |
692 | ||
693 | #define EXTRACT_NUMBER(destination, source) \ | |
694 | do { \ | |
695 | (destination) = *(source) & 0377; \ | |
696 | (destination) += SIGN_EXTEND_CHAR (*((source) + 1)) << 8; \ | |
697 | } while (0) | |
698 | ||
699 | #ifdef DEBUG | |
4bb91c68 | 700 | static void extract_number _RE_ARGS ((int *dest, re_char *source)); |
fa9a63c5 RM |
701 | static void |
702 | extract_number (dest, source) | |
703 | int *dest; | |
01618498 | 704 | re_char *source; |
fa9a63c5 | 705 | { |
5e69f11e | 706 | int temp = SIGN_EXTEND_CHAR (*(source + 1)); |
fa9a63c5 RM |
707 | *dest = *source & 0377; |
708 | *dest += temp << 8; | |
709 | } | |
710 | ||
4bb91c68 | 711 | # ifndef EXTRACT_MACROS /* To debug the macros. */ |
0b32bf0e SM |
712 | # undef EXTRACT_NUMBER |
713 | # define EXTRACT_NUMBER(dest, src) extract_number (&dest, src) | |
714 | # endif /* not EXTRACT_MACROS */ | |
fa9a63c5 RM |
715 | |
716 | #endif /* DEBUG */ | |
717 | ||
718 | /* Same as EXTRACT_NUMBER, except increment SOURCE to after the number. | |
719 | SOURCE must be an lvalue. */ | |
720 | ||
721 | #define EXTRACT_NUMBER_AND_INCR(destination, source) \ | |
722 | do { \ | |
723 | EXTRACT_NUMBER (destination, source); \ | |
25fe55af | 724 | (source) += 2; \ |
fa9a63c5 RM |
725 | } while (0) |
726 | ||
727 | #ifdef DEBUG | |
4bb91c68 SM |
728 | static void extract_number_and_incr _RE_ARGS ((int *destination, |
729 | re_char **source)); | |
fa9a63c5 RM |
730 | static void |
731 | extract_number_and_incr (destination, source) | |
732 | int *destination; | |
01618498 | 733 | re_char **source; |
5e69f11e | 734 | { |
fa9a63c5 RM |
735 | extract_number (destination, *source); |
736 | *source += 2; | |
737 | } | |
738 | ||
0b32bf0e SM |
739 | # ifndef EXTRACT_MACROS |
740 | # undef EXTRACT_NUMBER_AND_INCR | |
741 | # define EXTRACT_NUMBER_AND_INCR(dest, src) \ | |
fa9a63c5 | 742 | extract_number_and_incr (&dest, &src) |
0b32bf0e | 743 | # endif /* not EXTRACT_MACROS */ |
fa9a63c5 RM |
744 | |
745 | #endif /* DEBUG */ | |
746 | \f | |
b18215fc RS |
747 | /* Store a multibyte character in three contiguous bytes starting |
748 | DESTINATION, and increment DESTINATION to the byte after where the | |
25fe55af | 749 | character is stored. Therefore, DESTINATION must be an lvalue. */ |
b18215fc RS |
750 | |
751 | #define STORE_CHARACTER_AND_INCR(destination, character) \ | |
752 | do { \ | |
753 | (destination)[0] = (character) & 0377; \ | |
754 | (destination)[1] = ((character) >> 8) & 0377; \ | |
755 | (destination)[2] = (character) >> 16; \ | |
756 | (destination) += 3; \ | |
757 | } while (0) | |
758 | ||
759 | /* Put into DESTINATION a character stored in three contiguous bytes | |
25fe55af | 760 | starting at SOURCE. */ |
b18215fc RS |
761 | |
762 | #define EXTRACT_CHARACTER(destination, source) \ | |
763 | do { \ | |
764 | (destination) = ((source)[0] \ | |
765 | | ((source)[1] << 8) \ | |
766 | | ((source)[2] << 16)); \ | |
767 | } while (0) | |
768 | ||
769 | ||
770 | /* Macros for charset. */ | |
771 | ||
772 | /* Size of bitmap of charset P in bytes. P is a start of charset, | |
773 | i.e. *P is (re_opcode_t) charset or (re_opcode_t) charset_not. */ | |
774 | #define CHARSET_BITMAP_SIZE(p) ((p)[1] & 0x7F) | |
775 | ||
776 | /* Nonzero if charset P has range table. */ | |
25fe55af | 777 | #define CHARSET_RANGE_TABLE_EXISTS_P(p) ((p)[1] & 0x80) |
b18215fc RS |
778 | |
779 | /* Return the address of range table of charset P. But not the start | |
780 | of table itself, but the before where the number of ranges is | |
96cc36cc RS |
781 | stored. `2 +' means to skip re_opcode_t and size of bitmap, |
782 | and the 2 bytes of flags at the start of the range table. */ | |
783 | #define CHARSET_RANGE_TABLE(p) (&(p)[4 + CHARSET_BITMAP_SIZE (p)]) | |
784 | ||
785 | /* Extract the bit flags that start a range table. */ | |
786 | #define CHARSET_RANGE_TABLE_BITS(p) \ | |
787 | ((p)[2 + CHARSET_BITMAP_SIZE (p)] \ | |
788 | + (p)[3 + CHARSET_BITMAP_SIZE (p)] * 0x100) | |
b18215fc RS |
789 | |
790 | /* Test if C is listed in the bitmap of charset P. */ | |
791 | #define CHARSET_LOOKUP_BITMAP(p, c) \ | |
792 | ((c) < CHARSET_BITMAP_SIZE (p) * BYTEWIDTH \ | |
793 | && (p)[2 + (c) / BYTEWIDTH] & (1 << ((c) % BYTEWIDTH))) | |
794 | ||
795 | /* Return the address of end of RANGE_TABLE. COUNT is number of | |
25fe55af RS |
796 | ranges (which is a pair of (start, end)) in the RANGE_TABLE. `* 2' |
797 | is start of range and end of range. `* 3' is size of each start | |
b18215fc RS |
798 | and end. */ |
799 | #define CHARSET_RANGE_TABLE_END(range_table, count) \ | |
800 | ((range_table) + (count) * 2 * 3) | |
801 | ||
25fe55af | 802 | /* Test if C is in RANGE_TABLE. A flag NOT is negated if C is in. |
b18215fc RS |
803 | COUNT is number of ranges in RANGE_TABLE. */ |
804 | #define CHARSET_LOOKUP_RANGE_TABLE_RAW(not, c, range_table, count) \ | |
805 | do \ | |
806 | { \ | |
01618498 SM |
807 | re_wchar_t range_start, range_end; \ |
808 | re_char *p; \ | |
809 | re_char *range_table_end \ | |
b18215fc RS |
810 | = CHARSET_RANGE_TABLE_END ((range_table), (count)); \ |
811 | \ | |
812 | for (p = (range_table); p < range_table_end; p += 2 * 3) \ | |
813 | { \ | |
814 | EXTRACT_CHARACTER (range_start, p); \ | |
815 | EXTRACT_CHARACTER (range_end, p + 3); \ | |
816 | \ | |
817 | if (range_start <= (c) && (c) <= range_end) \ | |
818 | { \ | |
819 | (not) = !(not); \ | |
820 | break; \ | |
821 | } \ | |
822 | } \ | |
823 | } \ | |
824 | while (0) | |
825 | ||
826 | /* Test if C is in range table of CHARSET. The flag NOT is negated if | |
827 | C is listed in it. */ | |
828 | #define CHARSET_LOOKUP_RANGE_TABLE(not, c, charset) \ | |
829 | do \ | |
830 | { \ | |
831 | /* Number of ranges in range table. */ \ | |
832 | int count; \ | |
01618498 SM |
833 | re_char *range_table = CHARSET_RANGE_TABLE (charset); \ |
834 | \ | |
b18215fc RS |
835 | EXTRACT_NUMBER_AND_INCR (count, range_table); \ |
836 | CHARSET_LOOKUP_RANGE_TABLE_RAW ((not), (c), range_table, count); \ | |
837 | } \ | |
838 | while (0) | |
839 | \f | |
fa9a63c5 RM |
840 | /* If DEBUG is defined, Regex prints many voluminous messages about what |
841 | it is doing (if the variable `debug' is nonzero). If linked with the | |
842 | main program in `iregex.c', you can enter patterns and strings | |
843 | interactively. And if linked with the main program in `main.c' and | |
4bb91c68 | 844 | the other test files, you can run the already-written tests. */ |
fa9a63c5 RM |
845 | |
846 | #ifdef DEBUG | |
847 | ||
848 | /* We use standard I/O for debugging. */ | |
0b32bf0e | 849 | # include <stdio.h> |
fa9a63c5 RM |
850 | |
851 | /* It is useful to test things that ``must'' be true when debugging. */ | |
0b32bf0e | 852 | # include <assert.h> |
fa9a63c5 | 853 | |
99633e97 | 854 | static int debug = -100000; |
fa9a63c5 | 855 | |
0b32bf0e SM |
856 | # define DEBUG_STATEMENT(e) e |
857 | # define DEBUG_PRINT1(x) if (debug > 0) printf (x) | |
858 | # define DEBUG_PRINT2(x1, x2) if (debug > 0) printf (x1, x2) | |
859 | # define DEBUG_PRINT3(x1, x2, x3) if (debug > 0) printf (x1, x2, x3) | |
860 | # define DEBUG_PRINT4(x1, x2, x3, x4) if (debug > 0) printf (x1, x2, x3, x4) | |
861 | # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) \ | |
99633e97 | 862 | if (debug > 0) print_partial_compiled_pattern (s, e) |
0b32bf0e | 863 | # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) \ |
99633e97 | 864 | if (debug > 0) print_double_string (w, s1, sz1, s2, sz2) |
fa9a63c5 RM |
865 | |
866 | ||
867 | /* Print the fastmap in human-readable form. */ | |
868 | ||
869 | void | |
870 | print_fastmap (fastmap) | |
871 | char *fastmap; | |
872 | { | |
873 | unsigned was_a_range = 0; | |
5e69f11e RM |
874 | unsigned i = 0; |
875 | ||
fa9a63c5 RM |
876 | while (i < (1 << BYTEWIDTH)) |
877 | { | |
878 | if (fastmap[i++]) | |
879 | { | |
880 | was_a_range = 0; | |
25fe55af RS |
881 | putchar (i - 1); |
882 | while (i < (1 << BYTEWIDTH) && fastmap[i]) | |
883 | { | |
884 | was_a_range = 1; | |
885 | i++; | |
886 | } | |
fa9a63c5 | 887 | if (was_a_range) |
25fe55af RS |
888 | { |
889 | printf ("-"); | |
890 | putchar (i - 1); | |
891 | } | |
892 | } | |
fa9a63c5 | 893 | } |
5e69f11e | 894 | putchar ('\n'); |
fa9a63c5 RM |
895 | } |
896 | ||
897 | ||
898 | /* Print a compiled pattern string in human-readable form, starting at | |
899 | the START pointer into it and ending just before the pointer END. */ | |
900 | ||
901 | void | |
902 | print_partial_compiled_pattern (start, end) | |
01618498 SM |
903 | re_char *start; |
904 | re_char *end; | |
fa9a63c5 RM |
905 | { |
906 | int mcnt, mcnt2; | |
01618498 SM |
907 | re_char *p = start; |
908 | re_char *pend = end; | |
fa9a63c5 RM |
909 | |
910 | if (start == NULL) | |
911 | { | |
912 | printf ("(null)\n"); | |
913 | return; | |
914 | } | |
5e69f11e | 915 | |
fa9a63c5 RM |
916 | /* Loop over pattern commands. */ |
917 | while (p < pend) | |
918 | { | |
919 | printf ("%d:\t", p - start); | |
920 | ||
921 | switch ((re_opcode_t) *p++) | |
922 | { | |
25fe55af RS |
923 | case no_op: |
924 | printf ("/no_op"); | |
925 | break; | |
fa9a63c5 | 926 | |
99633e97 SM |
927 | case succeed: |
928 | printf ("/succeed"); | |
929 | break; | |
930 | ||
fa9a63c5 RM |
931 | case exactn: |
932 | mcnt = *p++; | |
25fe55af RS |
933 | printf ("/exactn/%d", mcnt); |
934 | do | |
fa9a63c5 | 935 | { |
25fe55af | 936 | putchar ('/'); |
fa9a63c5 | 937 | putchar (*p++); |
25fe55af RS |
938 | } |
939 | while (--mcnt); | |
940 | break; | |
fa9a63c5 RM |
941 | |
942 | case start_memory: | |
505bde11 | 943 | printf ("/start_memory/%d", *p++); |
25fe55af | 944 | break; |
fa9a63c5 RM |
945 | |
946 | case stop_memory: | |
505bde11 | 947 | printf ("/stop_memory/%d", *p++); |
25fe55af | 948 | break; |
fa9a63c5 RM |
949 | |
950 | case duplicate: | |
951 | printf ("/duplicate/%d", *p++); | |
952 | break; | |
953 | ||
954 | case anychar: | |
955 | printf ("/anychar"); | |
956 | break; | |
957 | ||
958 | case charset: | |
25fe55af RS |
959 | case charset_not: |
960 | { | |
961 | register int c, last = -100; | |
fa9a63c5 | 962 | register int in_range = 0; |
99633e97 SM |
963 | int length = CHARSET_BITMAP_SIZE (p - 1); |
964 | int has_range_table = CHARSET_RANGE_TABLE_EXISTS_P (p - 1); | |
fa9a63c5 RM |
965 | |
966 | printf ("/charset [%s", | |
25fe55af | 967 | (re_opcode_t) *(p - 1) == charset_not ? "^" : ""); |
5e69f11e | 968 | |
25fe55af | 969 | assert (p + *p < pend); |
fa9a63c5 | 970 | |
25fe55af | 971 | for (c = 0; c < 256; c++) |
96cc36cc | 972 | if (c / 8 < length |
fa9a63c5 RM |
973 | && (p[1 + (c/8)] & (1 << (c % 8)))) |
974 | { | |
975 | /* Are we starting a range? */ | |
976 | if (last + 1 == c && ! in_range) | |
977 | { | |
978 | putchar ('-'); | |
979 | in_range = 1; | |
980 | } | |
981 | /* Have we broken a range? */ | |
982 | else if (last + 1 != c && in_range) | |
96cc36cc | 983 | { |
fa9a63c5 RM |
984 | putchar (last); |
985 | in_range = 0; | |
986 | } | |
5e69f11e | 987 | |
fa9a63c5 RM |
988 | if (! in_range) |
989 | putchar (c); | |
990 | ||
991 | last = c; | |
25fe55af | 992 | } |
fa9a63c5 RM |
993 | |
994 | if (in_range) | |
995 | putchar (last); | |
996 | ||
997 | putchar (']'); | |
998 | ||
99633e97 | 999 | p += 1 + length; |
96cc36cc | 1000 | |
96cc36cc | 1001 | if (has_range_table) |
99633e97 SM |
1002 | { |
1003 | int count; | |
1004 | printf ("has-range-table"); | |
1005 | ||
1006 | /* ??? Should print the range table; for now, just skip it. */ | |
1007 | p += 2; /* skip range table bits */ | |
1008 | EXTRACT_NUMBER_AND_INCR (count, p); | |
1009 | p = CHARSET_RANGE_TABLE_END (p, count); | |
1010 | } | |
fa9a63c5 RM |
1011 | } |
1012 | break; | |
1013 | ||
1014 | case begline: | |
1015 | printf ("/begline"); | |
25fe55af | 1016 | break; |
fa9a63c5 RM |
1017 | |
1018 | case endline: | |
25fe55af RS |
1019 | printf ("/endline"); |
1020 | break; | |
fa9a63c5 RM |
1021 | |
1022 | case on_failure_jump: | |
25fe55af RS |
1023 | extract_number_and_incr (&mcnt, &p); |
1024 | printf ("/on_failure_jump to %d", p + mcnt - start); | |
1025 | break; | |
fa9a63c5 RM |
1026 | |
1027 | case on_failure_keep_string_jump: | |
25fe55af RS |
1028 | extract_number_and_incr (&mcnt, &p); |
1029 | printf ("/on_failure_keep_string_jump to %d", p + mcnt - start); | |
1030 | break; | |
fa9a63c5 | 1031 | |
0683b6fa SM |
1032 | case on_failure_jump_nastyloop: |
1033 | extract_number_and_incr (&mcnt, &p); | |
1034 | printf ("/on_failure_jump_nastyloop to %d", p + mcnt - start); | |
1035 | break; | |
1036 | ||
505bde11 | 1037 | case on_failure_jump_loop: |
fa9a63c5 | 1038 | extract_number_and_incr (&mcnt, &p); |
505bde11 | 1039 | printf ("/on_failure_jump_loop to %d", p + mcnt - start); |
5e69f11e RM |
1040 | break; |
1041 | ||
505bde11 | 1042 | case on_failure_jump_smart: |
fa9a63c5 | 1043 | extract_number_and_incr (&mcnt, &p); |
505bde11 | 1044 | printf ("/on_failure_jump_smart to %d", p + mcnt - start); |
5e69f11e RM |
1045 | break; |
1046 | ||
25fe55af | 1047 | case jump: |
fa9a63c5 | 1048 | extract_number_and_incr (&mcnt, &p); |
25fe55af | 1049 | printf ("/jump to %d", p + mcnt - start); |
fa9a63c5 RM |
1050 | break; |
1051 | ||
25fe55af RS |
1052 | case succeed_n: |
1053 | extract_number_and_incr (&mcnt, &p); | |
1054 | extract_number_and_incr (&mcnt2, &p); | |
99633e97 | 1055 | printf ("/succeed_n to %d, %d times", p - 2 + mcnt - start, mcnt2); |
25fe55af | 1056 | break; |
5e69f11e | 1057 | |
25fe55af RS |
1058 | case jump_n: |
1059 | extract_number_and_incr (&mcnt, &p); | |
1060 | extract_number_and_incr (&mcnt2, &p); | |
99633e97 | 1061 | printf ("/jump_n to %d, %d times", p - 2 + mcnt - start, mcnt2); |
25fe55af | 1062 | break; |
5e69f11e | 1063 | |
25fe55af RS |
1064 | case set_number_at: |
1065 | extract_number_and_incr (&mcnt, &p); | |
1066 | extract_number_and_incr (&mcnt2, &p); | |
99633e97 | 1067 | printf ("/set_number_at location %d to %d", p - 2 + mcnt - start, mcnt2); |
25fe55af | 1068 | break; |
5e69f11e | 1069 | |
25fe55af | 1070 | case wordbound: |
fa9a63c5 RM |
1071 | printf ("/wordbound"); |
1072 | break; | |
1073 | ||
1074 | case notwordbound: | |
1075 | printf ("/notwordbound"); | |
25fe55af | 1076 | break; |
fa9a63c5 RM |
1077 | |
1078 | case wordbeg: | |
1079 | printf ("/wordbeg"); | |
1080 | break; | |
5e69f11e | 1081 | |
fa9a63c5 RM |
1082 | case wordend: |
1083 | printf ("/wordend"); | |
5e69f11e | 1084 | |
1fb352e0 SM |
1085 | case syntaxspec: |
1086 | printf ("/syntaxspec"); | |
1087 | mcnt = *p++; | |
1088 | printf ("/%d", mcnt); | |
1089 | break; | |
1090 | ||
1091 | case notsyntaxspec: | |
1092 | printf ("/notsyntaxspec"); | |
1093 | mcnt = *p++; | |
1094 | printf ("/%d", mcnt); | |
1095 | break; | |
1096 | ||
0b32bf0e | 1097 | # ifdef emacs |
fa9a63c5 RM |
1098 | case before_dot: |
1099 | printf ("/before_dot"); | |
25fe55af | 1100 | break; |
fa9a63c5 RM |
1101 | |
1102 | case at_dot: | |
1103 | printf ("/at_dot"); | |
25fe55af | 1104 | break; |
fa9a63c5 RM |
1105 | |
1106 | case after_dot: | |
1107 | printf ("/after_dot"); | |
25fe55af | 1108 | break; |
fa9a63c5 | 1109 | |
1fb352e0 SM |
1110 | case categoryspec: |
1111 | printf ("/categoryspec"); | |
fa9a63c5 RM |
1112 | mcnt = *p++; |
1113 | printf ("/%d", mcnt); | |
25fe55af | 1114 | break; |
5e69f11e | 1115 | |
1fb352e0 SM |
1116 | case notcategoryspec: |
1117 | printf ("/notcategoryspec"); | |
fa9a63c5 RM |
1118 | mcnt = *p++; |
1119 | printf ("/%d", mcnt); | |
1120 | break; | |
0b32bf0e | 1121 | # endif /* emacs */ |
fa9a63c5 | 1122 | |
fa9a63c5 RM |
1123 | case begbuf: |
1124 | printf ("/begbuf"); | |
25fe55af | 1125 | break; |
fa9a63c5 RM |
1126 | |
1127 | case endbuf: | |
1128 | printf ("/endbuf"); | |
25fe55af | 1129 | break; |
fa9a63c5 | 1130 | |
25fe55af RS |
1131 | default: |
1132 | printf ("?%d", *(p-1)); | |
fa9a63c5 RM |
1133 | } |
1134 | ||
1135 | putchar ('\n'); | |
1136 | } | |
1137 | ||
1138 | printf ("%d:\tend of pattern.\n", p - start); | |
1139 | } | |
1140 | ||
1141 | ||
1142 | void | |
1143 | print_compiled_pattern (bufp) | |
1144 | struct re_pattern_buffer *bufp; | |
1145 | { | |
01618498 | 1146 | re_char *buffer = bufp->buffer; |
fa9a63c5 RM |
1147 | |
1148 | print_partial_compiled_pattern (buffer, buffer + bufp->used); | |
4bb91c68 SM |
1149 | printf ("%ld bytes used/%ld bytes allocated.\n", |
1150 | bufp->used, bufp->allocated); | |
fa9a63c5 RM |
1151 | |
1152 | if (bufp->fastmap_accurate && bufp->fastmap) | |
1153 | { | |
1154 | printf ("fastmap: "); | |
1155 | print_fastmap (bufp->fastmap); | |
1156 | } | |
1157 | ||
1158 | printf ("re_nsub: %d\t", bufp->re_nsub); | |
1159 | printf ("regs_alloc: %d\t", bufp->regs_allocated); | |
1160 | printf ("can_be_null: %d\t", bufp->can_be_null); | |
fa9a63c5 RM |
1161 | printf ("no_sub: %d\t", bufp->no_sub); |
1162 | printf ("not_bol: %d\t", bufp->not_bol); | |
1163 | printf ("not_eol: %d\t", bufp->not_eol); | |
4bb91c68 | 1164 | printf ("syntax: %lx\n", bufp->syntax); |
505bde11 | 1165 | fflush (stdout); |
fa9a63c5 RM |
1166 | /* Perhaps we should print the translate table? */ |
1167 | } | |
1168 | ||
1169 | ||
1170 | void | |
1171 | print_double_string (where, string1, size1, string2, size2) | |
66f0296e SM |
1172 | re_char *where; |
1173 | re_char *string1; | |
1174 | re_char *string2; | |
fa9a63c5 RM |
1175 | int size1; |
1176 | int size2; | |
1177 | { | |
4bb91c68 | 1178 | int this_char; |
5e69f11e | 1179 | |
fa9a63c5 RM |
1180 | if (where == NULL) |
1181 | printf ("(null)"); | |
1182 | else | |
1183 | { | |
1184 | if (FIRST_STRING_P (where)) | |
25fe55af RS |
1185 | { |
1186 | for (this_char = where - string1; this_char < size1; this_char++) | |
1187 | putchar (string1[this_char]); | |
fa9a63c5 | 1188 | |
25fe55af RS |
1189 | where = string2; |
1190 | } | |
fa9a63c5 RM |
1191 | |
1192 | for (this_char = where - string2; this_char < size2; this_char++) | |
25fe55af | 1193 | putchar (string2[this_char]); |
fa9a63c5 RM |
1194 | } |
1195 | } | |
1196 | ||
1197 | #else /* not DEBUG */ | |
1198 | ||
0b32bf0e SM |
1199 | # undef assert |
1200 | # define assert(e) | |
fa9a63c5 | 1201 | |
0b32bf0e SM |
1202 | # define DEBUG_STATEMENT(e) |
1203 | # define DEBUG_PRINT1(x) | |
1204 | # define DEBUG_PRINT2(x1, x2) | |
1205 | # define DEBUG_PRINT3(x1, x2, x3) | |
1206 | # define DEBUG_PRINT4(x1, x2, x3, x4) | |
1207 | # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) | |
1208 | # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) | |
fa9a63c5 RM |
1209 | |
1210 | #endif /* not DEBUG */ | |
1211 | \f | |
1212 | /* Set by `re_set_syntax' to the current regexp syntax to recognize. Can | |
1213 | also be assigned to arbitrarily: each pattern buffer stores its own | |
1214 | syntax, so it can be changed between regex compilations. */ | |
1215 | /* This has no initializer because initialized variables in Emacs | |
1216 | become read-only after dumping. */ | |
1217 | reg_syntax_t re_syntax_options; | |
1218 | ||
1219 | ||
1220 | /* Specify the precise syntax of regexps for compilation. This provides | |
1221 | for compatibility for various utilities which historically have | |
1222 | different, incompatible syntaxes. | |
1223 | ||
1224 | The argument SYNTAX is a bit mask comprised of the various bits | |
4bb91c68 | 1225 | defined in regex.h. We return the old syntax. */ |
fa9a63c5 RM |
1226 | |
1227 | reg_syntax_t | |
1228 | re_set_syntax (syntax) | |
1229 | reg_syntax_t syntax; | |
1230 | { | |
1231 | reg_syntax_t ret = re_syntax_options; | |
5e69f11e | 1232 | |
fa9a63c5 RM |
1233 | re_syntax_options = syntax; |
1234 | return ret; | |
1235 | } | |
c0f9ea08 | 1236 | WEAK_ALIAS (__re_set_syntax, re_set_syntax) |
fa9a63c5 RM |
1237 | \f |
1238 | /* This table gives an error message for each of the error codes listed | |
4bb91c68 | 1239 | in regex.h. Obviously the order here has to be same as there. |
fa9a63c5 | 1240 | POSIX doesn't require that we do anything for REG_NOERROR, |
4bb91c68 | 1241 | but why not be nice? */ |
fa9a63c5 RM |
1242 | |
1243 | static const char *re_error_msgid[] = | |
5e69f11e RM |
1244 | { |
1245 | gettext_noop ("Success"), /* REG_NOERROR */ | |
1246 | gettext_noop ("No match"), /* REG_NOMATCH */ | |
1247 | gettext_noop ("Invalid regular expression"), /* REG_BADPAT */ | |
1248 | gettext_noop ("Invalid collation character"), /* REG_ECOLLATE */ | |
1249 | gettext_noop ("Invalid character class name"), /* REG_ECTYPE */ | |
1250 | gettext_noop ("Trailing backslash"), /* REG_EESCAPE */ | |
1251 | gettext_noop ("Invalid back reference"), /* REG_ESUBREG */ | |
1252 | gettext_noop ("Unmatched [ or [^"), /* REG_EBRACK */ | |
1253 | gettext_noop ("Unmatched ( or \\("), /* REG_EPAREN */ | |
1254 | gettext_noop ("Unmatched \\{"), /* REG_EBRACE */ | |
1255 | gettext_noop ("Invalid content of \\{\\}"), /* REG_BADBR */ | |
1256 | gettext_noop ("Invalid range end"), /* REG_ERANGE */ | |
1257 | gettext_noop ("Memory exhausted"), /* REG_ESPACE */ | |
1258 | gettext_noop ("Invalid preceding regular expression"), /* REG_BADRPT */ | |
1259 | gettext_noop ("Premature end of regular expression"), /* REG_EEND */ | |
1260 | gettext_noop ("Regular expression too big"), /* REG_ESIZE */ | |
1261 | gettext_noop ("Unmatched ) or \\)"), /* REG_ERPAREN */ | |
fa9a63c5 RM |
1262 | }; |
1263 | \f | |
4bb91c68 | 1264 | /* Avoiding alloca during matching, to placate r_alloc. */ |
fa9a63c5 RM |
1265 | |
1266 | /* Define MATCH_MAY_ALLOCATE unless we need to make sure that the | |
1267 | searching and matching functions should not call alloca. On some | |
1268 | systems, alloca is implemented in terms of malloc, and if we're | |
1269 | using the relocating allocator routines, then malloc could cause a | |
1270 | relocation, which might (if the strings being searched are in the | |
1271 | ralloc heap) shift the data out from underneath the regexp | |
1272 | routines. | |
1273 | ||
5e69f11e | 1274 | Here's another reason to avoid allocation: Emacs |
fa9a63c5 RM |
1275 | processes input from X in a signal handler; processing X input may |
1276 | call malloc; if input arrives while a matching routine is calling | |
1277 | malloc, then we're scrod. But Emacs can't just block input while | |
1278 | calling matching routines; then we don't notice interrupts when | |
1279 | they come in. So, Emacs blocks input around all regexp calls | |
1280 | except the matching calls, which it leaves unprotected, in the | |
1281 | faith that they will not malloc. */ | |
1282 | ||
1283 | /* Normally, this is fine. */ | |
1284 | #define MATCH_MAY_ALLOCATE | |
1285 | ||
1286 | /* When using GNU C, we are not REALLY using the C alloca, no matter | |
1287 | what config.h may say. So don't take precautions for it. */ | |
1288 | #ifdef __GNUC__ | |
0b32bf0e | 1289 | # undef C_ALLOCA |
fa9a63c5 RM |
1290 | #endif |
1291 | ||
1292 | /* The match routines may not allocate if (1) they would do it with malloc | |
1293 | and (2) it's not safe for them to use malloc. | |
1294 | Note that if REL_ALLOC is defined, matching would not use malloc for the | |
1295 | failure stack, but we would still use it for the register vectors; | |
4bb91c68 | 1296 | so REL_ALLOC should not affect this. */ |
0b32bf0e SM |
1297 | #if (defined C_ALLOCA || defined REGEX_MALLOC) && defined emacs |
1298 | # undef MATCH_MAY_ALLOCATE | |
fa9a63c5 RM |
1299 | #endif |
1300 | ||
1301 | \f | |
1302 | /* Failure stack declarations and macros; both re_compile_fastmap and | |
1303 | re_match_2 use a failure stack. These have to be macros because of | |
1304 | REGEX_ALLOCATE_STACK. */ | |
5e69f11e | 1305 | |
fa9a63c5 | 1306 | |
320a2a73 | 1307 | /* Approximate number of failure points for which to initially allocate space |
fa9a63c5 RM |
1308 | when matching. If this number is exceeded, we allocate more |
1309 | space, so it is not a hard limit. */ | |
1310 | #ifndef INIT_FAILURE_ALLOC | |
0b32bf0e | 1311 | # define INIT_FAILURE_ALLOC 20 |
fa9a63c5 RM |
1312 | #endif |
1313 | ||
1314 | /* Roughly the maximum number of failure points on the stack. Would be | |
320a2a73 | 1315 | exactly that if always used TYPICAL_FAILURE_SIZE items each time we failed. |
fa9a63c5 | 1316 | This is a variable only so users of regex can assign to it; we never |
c0f9ea08 SM |
1317 | change it ourselves. */ |
1318 | # if defined MATCH_MAY_ALLOCATE | |
1319 | /* Note that 4400 was enough to cause a crash on Alpha OSF/1, | |
320a2a73 KH |
1320 | whose default stack limit is 2mb. In order for a larger |
1321 | value to work reliably, you have to try to make it accord | |
1322 | with the process stack limit. */ | |
c0f9ea08 SM |
1323 | size_t re_max_failures = 40000; |
1324 | # else | |
1325 | size_t re_max_failures = 4000; | |
1326 | # endif | |
fa9a63c5 RM |
1327 | |
1328 | union fail_stack_elt | |
1329 | { | |
01618498 | 1330 | re_char *pointer; |
c0f9ea08 SM |
1331 | /* This should be the biggest `int' that's no bigger than a pointer. */ |
1332 | long integer; | |
fa9a63c5 RM |
1333 | }; |
1334 | ||
1335 | typedef union fail_stack_elt fail_stack_elt_t; | |
1336 | ||
1337 | typedef struct | |
1338 | { | |
1339 | fail_stack_elt_t *stack; | |
c0f9ea08 SM |
1340 | size_t size; |
1341 | size_t avail; /* Offset of next open position. */ | |
1342 | size_t frame; /* Offset of the cur constructed frame. */ | |
fa9a63c5 RM |
1343 | } fail_stack_type; |
1344 | ||
505bde11 | 1345 | #define FAIL_STACK_EMPTY() (fail_stack.frame == 0) |
fa9a63c5 RM |
1346 | #define FAIL_STACK_FULL() (fail_stack.avail == fail_stack.size) |
1347 | ||
1348 | ||
1349 | /* Define macros to initialize and free the failure stack. | |
1350 | Do `return -2' if the alloc fails. */ | |
1351 | ||
1352 | #ifdef MATCH_MAY_ALLOCATE | |
0b32bf0e | 1353 | # define INIT_FAIL_STACK() \ |
fa9a63c5 RM |
1354 | do { \ |
1355 | fail_stack.stack = (fail_stack_elt_t *) \ | |
320a2a73 KH |
1356 | REGEX_ALLOCATE_STACK (INIT_FAILURE_ALLOC * TYPICAL_FAILURE_SIZE \ |
1357 | * sizeof (fail_stack_elt_t)); \ | |
fa9a63c5 RM |
1358 | \ |
1359 | if (fail_stack.stack == NULL) \ | |
1360 | return -2; \ | |
1361 | \ | |
1362 | fail_stack.size = INIT_FAILURE_ALLOC; \ | |
1363 | fail_stack.avail = 0; \ | |
505bde11 | 1364 | fail_stack.frame = 0; \ |
fa9a63c5 RM |
1365 | } while (0) |
1366 | ||
0b32bf0e | 1367 | # define RESET_FAIL_STACK() REGEX_FREE_STACK (fail_stack.stack) |
fa9a63c5 | 1368 | #else |
0b32bf0e | 1369 | # define INIT_FAIL_STACK() \ |
fa9a63c5 RM |
1370 | do { \ |
1371 | fail_stack.avail = 0; \ | |
505bde11 | 1372 | fail_stack.frame = 0; \ |
fa9a63c5 RM |
1373 | } while (0) |
1374 | ||
0b32bf0e | 1375 | # define RESET_FAIL_STACK() ((void)0) |
fa9a63c5 RM |
1376 | #endif |
1377 | ||
1378 | ||
320a2a73 KH |
1379 | /* Double the size of FAIL_STACK, up to a limit |
1380 | which allows approximately `re_max_failures' items. | |
fa9a63c5 RM |
1381 | |
1382 | Return 1 if succeeds, and 0 if either ran out of memory | |
5e69f11e RM |
1383 | allocating space for it or it was already too large. |
1384 | ||
4bb91c68 | 1385 | REGEX_REALLOCATE_STACK requires `destination' be declared. */ |
fa9a63c5 | 1386 | |
320a2a73 KH |
1387 | /* Factor to increase the failure stack size by |
1388 | when we increase it. | |
1389 | This used to be 2, but 2 was too wasteful | |
1390 | because the old discarded stacks added up to as much space | |
1391 | were as ultimate, maximum-size stack. */ | |
1392 | #define FAIL_STACK_GROWTH_FACTOR 4 | |
1393 | ||
1394 | #define GROW_FAIL_STACK(fail_stack) \ | |
eead07d6 KH |
1395 | (((fail_stack).size * sizeof (fail_stack_elt_t) \ |
1396 | >= re_max_failures * TYPICAL_FAILURE_SIZE) \ | |
fa9a63c5 | 1397 | ? 0 \ |
320a2a73 KH |
1398 | : ((fail_stack).stack \ |
1399 | = (fail_stack_elt_t *) \ | |
25fe55af RS |
1400 | REGEX_REALLOCATE_STACK ((fail_stack).stack, \ |
1401 | (fail_stack).size * sizeof (fail_stack_elt_t), \ | |
320a2a73 KH |
1402 | MIN (re_max_failures * TYPICAL_FAILURE_SIZE, \ |
1403 | ((fail_stack).size * sizeof (fail_stack_elt_t) \ | |
1404 | * FAIL_STACK_GROWTH_FACTOR))), \ | |
fa9a63c5 RM |
1405 | \ |
1406 | (fail_stack).stack == NULL \ | |
1407 | ? 0 \ | |
6453db45 KH |
1408 | : ((fail_stack).size \ |
1409 | = (MIN (re_max_failures * TYPICAL_FAILURE_SIZE, \ | |
1410 | ((fail_stack).size * sizeof (fail_stack_elt_t) \ | |
1411 | * FAIL_STACK_GROWTH_FACTOR)) \ | |
1412 | / sizeof (fail_stack_elt_t)), \ | |
25fe55af | 1413 | 1))) |
fa9a63c5 RM |
1414 | |
1415 | ||
fa9a63c5 RM |
1416 | /* Push a pointer value onto the failure stack. |
1417 | Assumes the variable `fail_stack'. Probably should only | |
4bb91c68 | 1418 | be called from within `PUSH_FAILURE_POINT'. */ |
fa9a63c5 | 1419 | #define PUSH_FAILURE_POINTER(item) \ |
01618498 | 1420 | fail_stack.stack[fail_stack.avail++].pointer = (item) |
fa9a63c5 RM |
1421 | |
1422 | /* This pushes an integer-valued item onto the failure stack. | |
1423 | Assumes the variable `fail_stack'. Probably should only | |
4bb91c68 | 1424 | be called from within `PUSH_FAILURE_POINT'. */ |
fa9a63c5 RM |
1425 | #define PUSH_FAILURE_INT(item) \ |
1426 | fail_stack.stack[fail_stack.avail++].integer = (item) | |
1427 | ||
1428 | /* Push a fail_stack_elt_t value onto the failure stack. | |
1429 | Assumes the variable `fail_stack'. Probably should only | |
4bb91c68 | 1430 | be called from within `PUSH_FAILURE_POINT'. */ |
fa9a63c5 RM |
1431 | #define PUSH_FAILURE_ELT(item) \ |
1432 | fail_stack.stack[fail_stack.avail++] = (item) | |
1433 | ||
1434 | /* These three POP... operations complement the three PUSH... operations. | |
1435 | All assume that `fail_stack' is nonempty. */ | |
1436 | #define POP_FAILURE_POINTER() fail_stack.stack[--fail_stack.avail].pointer | |
1437 | #define POP_FAILURE_INT() fail_stack.stack[--fail_stack.avail].integer | |
1438 | #define POP_FAILURE_ELT() fail_stack.stack[--fail_stack.avail] | |
1439 | ||
505bde11 SM |
1440 | /* Individual items aside from the registers. */ |
1441 | #define NUM_NONREG_ITEMS 3 | |
1442 | ||
1443 | /* Used to examine the stack (to detect infinite loops). */ | |
1444 | #define FAILURE_PAT(h) fail_stack.stack[(h) - 1].pointer | |
66f0296e | 1445 | #define FAILURE_STR(h) (fail_stack.stack[(h) - 2].pointer) |
505bde11 SM |
1446 | #define NEXT_FAILURE_HANDLE(h) fail_stack.stack[(h) - 3].integer |
1447 | #define TOP_FAILURE_HANDLE() fail_stack.frame | |
fa9a63c5 RM |
1448 | |
1449 | ||
505bde11 SM |
1450 | #define ENSURE_FAIL_STACK(space) \ |
1451 | while (REMAINING_AVAIL_SLOTS <= space) { \ | |
1452 | if (!GROW_FAIL_STACK (fail_stack)) \ | |
1453 | return -2; \ | |
1454 | DEBUG_PRINT2 ("\n Doubled stack; size now: %d\n", (fail_stack).size);\ | |
1455 | DEBUG_PRINT2 (" slots available: %d\n", REMAINING_AVAIL_SLOTS);\ | |
1456 | } | |
1457 | ||
1458 | /* Push register NUM onto the stack. */ | |
1459 | #define PUSH_FAILURE_REG(num) \ | |
1460 | do { \ | |
1461 | char *destination; \ | |
1462 | ENSURE_FAIL_STACK(3); \ | |
1463 | DEBUG_PRINT4 (" Push reg %d (spanning %p -> %p)\n", \ | |
1464 | num, regstart[num], regend[num]); \ | |
1465 | PUSH_FAILURE_POINTER (regstart[num]); \ | |
1466 | PUSH_FAILURE_POINTER (regend[num]); \ | |
1467 | PUSH_FAILURE_INT (num); \ | |
1468 | } while (0) | |
1469 | ||
01618498 SM |
1470 | /* Change the counter's value to VAL, but make sure that it will |
1471 | be reset when backtracking. */ | |
1472 | #define PUSH_NUMBER(ptr,val) \ | |
dc1e502d SM |
1473 | do { \ |
1474 | char *destination; \ | |
1475 | int c; \ | |
1476 | ENSURE_FAIL_STACK(3); \ | |
1477 | EXTRACT_NUMBER (c, ptr); \ | |
01618498 | 1478 | DEBUG_PRINT4 (" Push number %p = %d -> %d\n", ptr, c, val); \ |
dc1e502d SM |
1479 | PUSH_FAILURE_INT (c); \ |
1480 | PUSH_FAILURE_POINTER (ptr); \ | |
1481 | PUSH_FAILURE_INT (-1); \ | |
01618498 | 1482 | STORE_NUMBER (ptr, val); \ |
dc1e502d SM |
1483 | } while (0) |
1484 | ||
505bde11 | 1485 | /* Pop a saved register off the stack. */ |
dc1e502d | 1486 | #define POP_FAILURE_REG_OR_COUNT() \ |
505bde11 SM |
1487 | do { \ |
1488 | int reg = POP_FAILURE_INT (); \ | |
dc1e502d SM |
1489 | if (reg == -1) \ |
1490 | { \ | |
1491 | /* It's a counter. */ \ | |
6dcf2d0e SM |
1492 | /* Here, we discard `const', making re_match non-reentrant. */ \ |
1493 | unsigned char *ptr = (unsigned char*) POP_FAILURE_POINTER (); \ | |
dc1e502d SM |
1494 | reg = POP_FAILURE_INT (); \ |
1495 | STORE_NUMBER (ptr, reg); \ | |
1496 | DEBUG_PRINT3 (" Pop counter %p = %d\n", ptr, reg); \ | |
1497 | } \ | |
1498 | else \ | |
1499 | { \ | |
1500 | regend[reg] = POP_FAILURE_POINTER (); \ | |
1501 | regstart[reg] = POP_FAILURE_POINTER (); \ | |
1502 | DEBUG_PRINT4 (" Pop reg %d (spanning %p -> %p)\n", \ | |
1503 | reg, regstart[reg], regend[reg]); \ | |
1504 | } \ | |
505bde11 SM |
1505 | } while (0) |
1506 | ||
1507 | /* Check that we are not stuck in an infinite loop. */ | |
1508 | #define CHECK_INFINITE_LOOP(pat_cur, string_place) \ | |
1509 | do { \ | |
1510 | int failure = TOP_FAILURE_HANDLE(); \ | |
1511 | /* Check for infinite matching loops */ \ | |
1512 | while (failure > 0 && \ | |
1513 | (FAILURE_STR (failure) == string_place \ | |
1514 | || FAILURE_STR (failure) == NULL)) \ | |
1515 | { \ | |
1516 | assert (FAILURE_PAT (failure) >= bufp->buffer \ | |
66f0296e | 1517 | && FAILURE_PAT (failure) <= bufp->buffer + bufp->used); \ |
505bde11 SM |
1518 | if (FAILURE_PAT (failure) == pat_cur) \ |
1519 | goto fail; \ | |
66f0296e | 1520 | DEBUG_PRINT2 (" Other pattern: %p\n", FAILURE_PAT (failure)); \ |
505bde11 SM |
1521 | failure = NEXT_FAILURE_HANDLE(failure); \ |
1522 | } \ | |
1523 | DEBUG_PRINT2 (" Other string: %p\n", FAILURE_STR (failure)); \ | |
1524 | } while (0) | |
1525 | ||
fa9a63c5 | 1526 | /* Push the information about the state we will need |
5e69f11e RM |
1527 | if we ever fail back to it. |
1528 | ||
505bde11 | 1529 | Requires variables fail_stack, regstart, regend and |
320a2a73 | 1530 | num_regs be declared. GROW_FAIL_STACK requires `destination' be |
fa9a63c5 | 1531 | declared. |
5e69f11e | 1532 | |
fa9a63c5 RM |
1533 | Does `return FAILURE_CODE' if runs out of memory. */ |
1534 | ||
505bde11 SM |
1535 | #define PUSH_FAILURE_POINT(pattern, string_place) \ |
1536 | do { \ | |
1537 | char *destination; \ | |
1538 | /* Must be int, so when we don't save any registers, the arithmetic \ | |
1539 | of 0 + -1 isn't done as unsigned. */ \ | |
1540 | \ | |
505bde11 | 1541 | DEBUG_STATEMENT (nfailure_points_pushed++); \ |
4bb91c68 | 1542 | DEBUG_PRINT1 ("\nPUSH_FAILURE_POINT:\n"); \ |
505bde11 SM |
1543 | DEBUG_PRINT2 (" Before push, next avail: %d\n", (fail_stack).avail); \ |
1544 | DEBUG_PRINT2 (" size: %d\n", (fail_stack).size);\ | |
1545 | \ | |
1546 | ENSURE_FAIL_STACK (NUM_NONREG_ITEMS); \ | |
1547 | \ | |
1548 | DEBUG_PRINT1 ("\n"); \ | |
1549 | \ | |
1550 | DEBUG_PRINT2 (" Push frame index: %d\n", fail_stack.frame); \ | |
1551 | PUSH_FAILURE_INT (fail_stack.frame); \ | |
1552 | \ | |
1553 | DEBUG_PRINT2 (" Push string %p: `", string_place); \ | |
1554 | DEBUG_PRINT_DOUBLE_STRING (string_place, string1, size1, string2, size2);\ | |
1555 | DEBUG_PRINT1 ("'\n"); \ | |
1556 | PUSH_FAILURE_POINTER (string_place); \ | |
1557 | \ | |
1558 | DEBUG_PRINT2 (" Push pattern %p: ", pattern); \ | |
1559 | DEBUG_PRINT_COMPILED_PATTERN (bufp, pattern, pend); \ | |
1560 | PUSH_FAILURE_POINTER (pattern); \ | |
1561 | \ | |
1562 | /* Close the frame by moving the frame pointer past it. */ \ | |
1563 | fail_stack.frame = fail_stack.avail; \ | |
1564 | } while (0) | |
fa9a63c5 | 1565 | |
320a2a73 KH |
1566 | /* Estimate the size of data pushed by a typical failure stack entry. |
1567 | An estimate is all we need, because all we use this for | |
1568 | is to choose a limit for how big to make the failure stack. */ | |
1569 | ||
1570 | #define TYPICAL_FAILURE_SIZE 20 | |
fa9a63c5 | 1571 | |
fa9a63c5 RM |
1572 | /* How many items can still be added to the stack without overflowing it. */ |
1573 | #define REMAINING_AVAIL_SLOTS ((fail_stack).size - (fail_stack).avail) | |
1574 | ||
1575 | ||
1576 | /* Pops what PUSH_FAIL_STACK pushes. | |
1577 | ||
1578 | We restore into the parameters, all of which should be lvalues: | |
1579 | STR -- the saved data position. | |
1580 | PAT -- the saved pattern position. | |
fa9a63c5 | 1581 | REGSTART, REGEND -- arrays of string positions. |
5e69f11e | 1582 | |
fa9a63c5 | 1583 | Also assumes the variables `fail_stack' and (if debugging), `bufp', |
25fe55af | 1584 | `pend', `string1', `size1', `string2', and `size2'. */ |
fa9a63c5 | 1585 | |
505bde11 SM |
1586 | #define POP_FAILURE_POINT(str, pat) \ |
1587 | do { \ | |
fa9a63c5 RM |
1588 | assert (!FAIL_STACK_EMPTY ()); \ |
1589 | \ | |
1590 | /* Remove failure points and point to how many regs pushed. */ \ | |
1591 | DEBUG_PRINT1 ("POP_FAILURE_POINT:\n"); \ | |
1592 | DEBUG_PRINT2 (" Before pop, next avail: %d\n", fail_stack.avail); \ | |
25fe55af | 1593 | DEBUG_PRINT2 (" size: %d\n", fail_stack.size); \ |
fa9a63c5 | 1594 | \ |
505bde11 SM |
1595 | /* Pop the saved registers. */ \ |
1596 | while (fail_stack.frame < fail_stack.avail) \ | |
dc1e502d | 1597 | POP_FAILURE_REG_OR_COUNT (); \ |
fa9a63c5 | 1598 | \ |
01618498 | 1599 | pat = POP_FAILURE_POINTER (); \ |
505bde11 SM |
1600 | DEBUG_PRINT2 (" Popping pattern %p: ", pat); \ |
1601 | DEBUG_PRINT_COMPILED_PATTERN (bufp, pat, pend); \ | |
fa9a63c5 RM |
1602 | \ |
1603 | /* If the saved string location is NULL, it came from an \ | |
1604 | on_failure_keep_string_jump opcode, and we want to throw away the \ | |
1605 | saved NULL, thus retaining our current position in the string. */ \ | |
01618498 | 1606 | str = POP_FAILURE_POINTER (); \ |
505bde11 | 1607 | DEBUG_PRINT2 (" Popping string %p: `", str); \ |
fa9a63c5 RM |
1608 | DEBUG_PRINT_DOUBLE_STRING (str, string1, size1, string2, size2); \ |
1609 | DEBUG_PRINT1 ("'\n"); \ | |
1610 | \ | |
505bde11 SM |
1611 | fail_stack.frame = POP_FAILURE_INT (); \ |
1612 | DEBUG_PRINT2 (" Popping frame index: %d\n", fail_stack.frame); \ | |
fa9a63c5 | 1613 | \ |
505bde11 SM |
1614 | assert (fail_stack.avail >= 0); \ |
1615 | assert (fail_stack.frame <= fail_stack.avail); \ | |
fa9a63c5 | 1616 | \ |
fa9a63c5 | 1617 | DEBUG_STATEMENT (nfailure_points_popped++); \ |
505bde11 | 1618 | } while (0) /* POP_FAILURE_POINT */ |
fa9a63c5 RM |
1619 | |
1620 | ||
1621 | \f | |
fa9a63c5 | 1622 | /* Registers are set to a sentinel when they haven't yet matched. */ |
4bb91c68 | 1623 | #define REG_UNSET(e) ((e) == NULL) |
fa9a63c5 RM |
1624 | \f |
1625 | /* Subroutine declarations and macros for regex_compile. */ | |
1626 | ||
4bb91c68 SM |
1627 | static reg_errcode_t regex_compile _RE_ARGS ((re_char *pattern, size_t size, |
1628 | reg_syntax_t syntax, | |
1629 | struct re_pattern_buffer *bufp)); | |
1630 | static void store_op1 _RE_ARGS ((re_opcode_t op, unsigned char *loc, int arg)); | |
1631 | static void store_op2 _RE_ARGS ((re_opcode_t op, unsigned char *loc, | |
1632 | int arg1, int arg2)); | |
1633 | static void insert_op1 _RE_ARGS ((re_opcode_t op, unsigned char *loc, | |
1634 | int arg, unsigned char *end)); | |
1635 | static void insert_op2 _RE_ARGS ((re_opcode_t op, unsigned char *loc, | |
1636 | int arg1, int arg2, unsigned char *end)); | |
01618498 SM |
1637 | static boolean at_begline_loc_p _RE_ARGS ((re_char *pattern, |
1638 | re_char *p, | |
4bb91c68 | 1639 | reg_syntax_t syntax)); |
01618498 SM |
1640 | static boolean at_endline_loc_p _RE_ARGS ((re_char *p, |
1641 | re_char *pend, | |
4bb91c68 | 1642 | reg_syntax_t syntax)); |
01618498 SM |
1643 | static re_char *skip_one_char _RE_ARGS ((re_char *p)); |
1644 | static int analyse_first _RE_ARGS ((re_char *p, re_char *pend, | |
4bb91c68 | 1645 | char *fastmap, const int multibyte)); |
fa9a63c5 | 1646 | |
5e69f11e | 1647 | /* Fetch the next character in the uncompiled pattern---translating it |
01618498 | 1648 | if necessary. */ |
fa9a63c5 | 1649 | #define PATFETCH(c) \ |
99633e97 SM |
1650 | do { \ |
1651 | PATFETCH_RAW (c); \ | |
2d1675e4 | 1652 | c = TRANSLATE (c); \ |
fa9a63c5 RM |
1653 | } while (0) |
1654 | ||
1655 | /* Fetch the next character in the uncompiled pattern, with no | |
4bb91c68 | 1656 | translation. */ |
fa9a63c5 | 1657 | #define PATFETCH_RAW(c) \ |
2d1675e4 SM |
1658 | do { \ |
1659 | int len; \ | |
1660 | if (p == pend) return REG_EEND; \ | |
1661 | c = RE_STRING_CHAR_AND_LENGTH (p, pend - p, len); \ | |
1662 | p += len; \ | |
fa9a63c5 RM |
1663 | } while (0) |
1664 | ||
fa9a63c5 RM |
1665 | |
1666 | /* If `translate' is non-null, return translate[D], else just D. We | |
1667 | cast the subscript to translate because some data is declared as | |
1668 | `char *', to avoid warnings when a string constant is passed. But | |
1669 | when we use a character as a subscript we must make it unsigned. */ | |
6676cb1c | 1670 | #ifndef TRANSLATE |
0b32bf0e | 1671 | # define TRANSLATE(d) \ |
66f0296e | 1672 | (RE_TRANSLATE_P (translate) ? RE_TRANSLATE (translate, (d)) : (d)) |
6676cb1c | 1673 | #endif |
fa9a63c5 RM |
1674 | |
1675 | ||
1676 | /* Macros for outputting the compiled pattern into `buffer'. */ | |
1677 | ||
1678 | /* If the buffer isn't allocated when it comes in, use this. */ | |
1679 | #define INIT_BUF_SIZE 32 | |
1680 | ||
4bb91c68 | 1681 | /* Make sure we have at least N more bytes of space in buffer. */ |
fa9a63c5 | 1682 | #define GET_BUFFER_SPACE(n) \ |
01618498 | 1683 | while ((size_t) (b - bufp->buffer + (n)) > bufp->allocated) \ |
fa9a63c5 RM |
1684 | EXTEND_BUFFER () |
1685 | ||
1686 | /* Make sure we have one more byte of buffer space and then add C to it. */ | |
1687 | #define BUF_PUSH(c) \ | |
1688 | do { \ | |
1689 | GET_BUFFER_SPACE (1); \ | |
1690 | *b++ = (unsigned char) (c); \ | |
1691 | } while (0) | |
1692 | ||
1693 | ||
1694 | /* Ensure we have two more bytes of buffer space and then append C1 and C2. */ | |
1695 | #define BUF_PUSH_2(c1, c2) \ | |
1696 | do { \ | |
1697 | GET_BUFFER_SPACE (2); \ | |
1698 | *b++ = (unsigned char) (c1); \ | |
1699 | *b++ = (unsigned char) (c2); \ | |
1700 | } while (0) | |
1701 | ||
1702 | ||
4bb91c68 | 1703 | /* As with BUF_PUSH_2, except for three bytes. */ |
fa9a63c5 RM |
1704 | #define BUF_PUSH_3(c1, c2, c3) \ |
1705 | do { \ | |
1706 | GET_BUFFER_SPACE (3); \ | |
1707 | *b++ = (unsigned char) (c1); \ | |
1708 | *b++ = (unsigned char) (c2); \ | |
1709 | *b++ = (unsigned char) (c3); \ | |
1710 | } while (0) | |
1711 | ||
1712 | ||
1713 | /* Store a jump with opcode OP at LOC to location TO. We store a | |
4bb91c68 | 1714 | relative address offset by the three bytes the jump itself occupies. */ |
fa9a63c5 RM |
1715 | #define STORE_JUMP(op, loc, to) \ |
1716 | store_op1 (op, loc, (to) - (loc) - 3) | |
1717 | ||
1718 | /* Likewise, for a two-argument jump. */ | |
1719 | #define STORE_JUMP2(op, loc, to, arg) \ | |
1720 | store_op2 (op, loc, (to) - (loc) - 3, arg) | |
1721 | ||
4bb91c68 | 1722 | /* Like `STORE_JUMP', but for inserting. Assume `b' is the buffer end. */ |
fa9a63c5 RM |
1723 | #define INSERT_JUMP(op, loc, to) \ |
1724 | insert_op1 (op, loc, (to) - (loc) - 3, b) | |
1725 | ||
1726 | /* Like `STORE_JUMP2', but for inserting. Assume `b' is the buffer end. */ | |
1727 | #define INSERT_JUMP2(op, loc, to, arg) \ | |
1728 | insert_op2 (op, loc, (to) - (loc) - 3, arg, b) | |
1729 | ||
1730 | ||
1731 | /* This is not an arbitrary limit: the arguments which represent offsets | |
4bb91c68 | 1732 | into the pattern are two bytes long. So if 2^16 bytes turns out to |
fa9a63c5 | 1733 | be too small, many things would have to change. */ |
4bb91c68 SM |
1734 | /* Any other compiler which, like MSC, has allocation limit below 2^16 |
1735 | bytes will have to use approach similar to what was done below for | |
1736 | MSC and drop MAX_BUF_SIZE a bit. Otherwise you may end up | |
1737 | reallocating to 0 bytes. Such thing is not going to work too well. | |
1738 | You have been warned!! */ | |
1739 | #if defined _MSC_VER && !defined WIN32 | |
1740 | /* Microsoft C 16-bit versions limit malloc to approx 65512 bytes. */ | |
1741 | # define MAX_BUF_SIZE 65500L | |
1742 | #else | |
1743 | # define MAX_BUF_SIZE (1L << 16) | |
1744 | #endif | |
fa9a63c5 RM |
1745 | |
1746 | /* Extend the buffer by twice its current size via realloc and | |
1747 | reset the pointers that pointed into the old block to point to the | |
1748 | correct places in the new one. If extending the buffer results in it | |
4bb91c68 SM |
1749 | being larger than MAX_BUF_SIZE, then flag memory exhausted. */ |
1750 | #if __BOUNDED_POINTERS__ | |
1751 | # define SET_HIGH_BOUND(P) (__ptrhigh (P) = __ptrlow (P) + bufp->allocated) | |
1752 | # define MOVE_BUFFER_POINTER(P) \ | |
1753 | (__ptrlow (P) += incr, SET_HIGH_BOUND (P), __ptrvalue (P) += incr) | |
1754 | # define ELSE_EXTEND_BUFFER_HIGH_BOUND \ | |
1755 | else \ | |
1756 | { \ | |
1757 | SET_HIGH_BOUND (b); \ | |
1758 | SET_HIGH_BOUND (begalt); \ | |
1759 | if (fixup_alt_jump) \ | |
1760 | SET_HIGH_BOUND (fixup_alt_jump); \ | |
1761 | if (laststart) \ | |
1762 | SET_HIGH_BOUND (laststart); \ | |
1763 | if (pending_exact) \ | |
1764 | SET_HIGH_BOUND (pending_exact); \ | |
1765 | } | |
1766 | #else | |
1767 | # define MOVE_BUFFER_POINTER(P) (P) += incr | |
1768 | # define ELSE_EXTEND_BUFFER_HIGH_BOUND | |
1769 | #endif | |
fa9a63c5 | 1770 | #define EXTEND_BUFFER() \ |
25fe55af | 1771 | do { \ |
01618498 | 1772 | re_char *old_buffer = bufp->buffer; \ |
25fe55af | 1773 | if (bufp->allocated == MAX_BUF_SIZE) \ |
fa9a63c5 RM |
1774 | return REG_ESIZE; \ |
1775 | bufp->allocated <<= 1; \ | |
1776 | if (bufp->allocated > MAX_BUF_SIZE) \ | |
25fe55af | 1777 | bufp->allocated = MAX_BUF_SIZE; \ |
01618498 | 1778 | RETALLOC (bufp->buffer, bufp->allocated, unsigned char); \ |
fa9a63c5 RM |
1779 | if (bufp->buffer == NULL) \ |
1780 | return REG_ESPACE; \ | |
1781 | /* If the buffer moved, move all the pointers into it. */ \ | |
1782 | if (old_buffer != bufp->buffer) \ | |
1783 | { \ | |
4bb91c68 SM |
1784 | int incr = bufp->buffer - old_buffer; \ |
1785 | MOVE_BUFFER_POINTER (b); \ | |
1786 | MOVE_BUFFER_POINTER (begalt); \ | |
25fe55af | 1787 | if (fixup_alt_jump) \ |
4bb91c68 | 1788 | MOVE_BUFFER_POINTER (fixup_alt_jump); \ |
25fe55af | 1789 | if (laststart) \ |
4bb91c68 | 1790 | MOVE_BUFFER_POINTER (laststart); \ |
25fe55af | 1791 | if (pending_exact) \ |
4bb91c68 | 1792 | MOVE_BUFFER_POINTER (pending_exact); \ |
fa9a63c5 | 1793 | } \ |
4bb91c68 | 1794 | ELSE_EXTEND_BUFFER_HIGH_BOUND \ |
fa9a63c5 RM |
1795 | } while (0) |
1796 | ||
1797 | ||
1798 | /* Since we have one byte reserved for the register number argument to | |
1799 | {start,stop}_memory, the maximum number of groups we can report | |
1800 | things about is what fits in that byte. */ | |
1801 | #define MAX_REGNUM 255 | |
1802 | ||
1803 | /* But patterns can have more than `MAX_REGNUM' registers. We just | |
1804 | ignore the excess. */ | |
1805 | typedef unsigned regnum_t; | |
1806 | ||
1807 | ||
1808 | /* Macros for the compile stack. */ | |
1809 | ||
1810 | /* Since offsets can go either forwards or backwards, this type needs to | |
4bb91c68 SM |
1811 | be able to hold values from -(MAX_BUF_SIZE - 1) to MAX_BUF_SIZE - 1. */ |
1812 | /* int may be not enough when sizeof(int) == 2. */ | |
1813 | typedef long pattern_offset_t; | |
fa9a63c5 RM |
1814 | |
1815 | typedef struct | |
1816 | { | |
1817 | pattern_offset_t begalt_offset; | |
1818 | pattern_offset_t fixup_alt_jump; | |
5e69f11e | 1819 | pattern_offset_t laststart_offset; |
fa9a63c5 RM |
1820 | regnum_t regnum; |
1821 | } compile_stack_elt_t; | |
1822 | ||
1823 | ||
1824 | typedef struct | |
1825 | { | |
1826 | compile_stack_elt_t *stack; | |
1827 | unsigned size; | |
1828 | unsigned avail; /* Offset of next open position. */ | |
1829 | } compile_stack_type; | |
1830 | ||
1831 | ||
1832 | #define INIT_COMPILE_STACK_SIZE 32 | |
1833 | ||
1834 | #define COMPILE_STACK_EMPTY (compile_stack.avail == 0) | |
1835 | #define COMPILE_STACK_FULL (compile_stack.avail == compile_stack.size) | |
1836 | ||
4bb91c68 | 1837 | /* The next available element. */ |
fa9a63c5 RM |
1838 | #define COMPILE_STACK_TOP (compile_stack.stack[compile_stack.avail]) |
1839 | ||
1840 | ||
b18215fc RS |
1841 | /* Structure to manage work area for range table. */ |
1842 | struct range_table_work_area | |
1843 | { | |
1844 | int *table; /* actual work area. */ | |
1845 | int allocated; /* allocated size for work area in bytes. */ | |
25fe55af | 1846 | int used; /* actually used size in words. */ |
96cc36cc | 1847 | int bits; /* flag to record character classes */ |
b18215fc RS |
1848 | }; |
1849 | ||
1850 | /* Make sure that WORK_AREA can hold more N multibyte characters. */ | |
1851 | #define EXTEND_RANGE_TABLE_WORK_AREA(work_area, n) \ | |
1852 | do { \ | |
1853 | if (((work_area).used + (n)) * sizeof (int) > (work_area).allocated) \ | |
1854 | { \ | |
1855 | (work_area).allocated += 16 * sizeof (int); \ | |
1856 | if ((work_area).table) \ | |
1857 | (work_area).table \ | |
1858 | = (int *) realloc ((work_area).table, (work_area).allocated); \ | |
1859 | else \ | |
1860 | (work_area).table \ | |
1861 | = (int *) malloc ((work_area).allocated); \ | |
1862 | if ((work_area).table == 0) \ | |
1863 | FREE_STACK_RETURN (REG_ESPACE); \ | |
1864 | } \ | |
1865 | } while (0) | |
1866 | ||
96cc36cc RS |
1867 | #define SET_RANGE_TABLE_WORK_AREA_BIT(work_area, bit) \ |
1868 | (work_area).bits |= (bit) | |
1869 | ||
14473664 SM |
1870 | /* Bits used to implement the multibyte-part of the various character classes |
1871 | such as [:alnum:] in a charset's range table. */ | |
1872 | #define BIT_WORD 0x1 | |
1873 | #define BIT_LOWER 0x2 | |
1874 | #define BIT_PUNCT 0x4 | |
1875 | #define BIT_SPACE 0x8 | |
1876 | #define BIT_UPPER 0x10 | |
1877 | #define BIT_MULTIBYTE 0x20 | |
96cc36cc | 1878 | |
b18215fc RS |
1879 | /* Set a range (RANGE_START, RANGE_END) to WORK_AREA. */ |
1880 | #define SET_RANGE_TABLE_WORK_AREA(work_area, range_start, range_end) \ | |
1881 | do { \ | |
1882 | EXTEND_RANGE_TABLE_WORK_AREA ((work_area), 2); \ | |
1883 | (work_area).table[(work_area).used++] = (range_start); \ | |
1884 | (work_area).table[(work_area).used++] = (range_end); \ | |
1885 | } while (0) | |
1886 | ||
25fe55af | 1887 | /* Free allocated memory for WORK_AREA. */ |
b18215fc RS |
1888 | #define FREE_RANGE_TABLE_WORK_AREA(work_area) \ |
1889 | do { \ | |
1890 | if ((work_area).table) \ | |
1891 | free ((work_area).table); \ | |
1892 | } while (0) | |
1893 | ||
96cc36cc | 1894 | #define CLEAR_RANGE_TABLE_WORK_USED(work_area) ((work_area).used = 0, (work_area).bits = 0) |
b18215fc | 1895 | #define RANGE_TABLE_WORK_USED(work_area) ((work_area).used) |
96cc36cc | 1896 | #define RANGE_TABLE_WORK_BITS(work_area) ((work_area).bits) |
b18215fc RS |
1897 | #define RANGE_TABLE_WORK_ELT(work_area, i) ((work_area).table[i]) |
1898 | ||
1899 | ||
fa9a63c5 | 1900 | /* Set the bit for character C in a list. */ |
01618498 | 1901 | #define SET_LIST_BIT(c) (b[((c)) / BYTEWIDTH] |= 1 << ((c) % BYTEWIDTH)) |
fa9a63c5 RM |
1902 | |
1903 | ||
1904 | /* Get the next unsigned number in the uncompiled pattern. */ | |
25fe55af | 1905 | #define GET_UNSIGNED_NUMBER(num) \ |
99633e97 | 1906 | do { if (p != pend) \ |
fa9a63c5 | 1907 | { \ |
25fe55af | 1908 | PATFETCH (c); \ |
4bb91c68 | 1909 | while ('0' <= c && c <= '9') \ |
25fe55af RS |
1910 | { \ |
1911 | if (num < 0) \ | |
1912 | num = 0; \ | |
1913 | num = num * 10 + c - '0'; \ | |
1914 | if (p == pend) \ | |
1915 | break; \ | |
1916 | PATFETCH (c); \ | |
1917 | } \ | |
1918 | } \ | |
99633e97 | 1919 | } while (0) |
fa9a63c5 | 1920 | |
a0ad02f7 | 1921 | #if WIDE_CHAR_SUPPORT |
14473664 SM |
1922 | /* The GNU C library provides support for user-defined character classes |
1923 | and the functions from ISO C amendement 1. */ | |
1924 | # ifdef CHARCLASS_NAME_MAX | |
1925 | # define CHAR_CLASS_MAX_LENGTH CHARCLASS_NAME_MAX | |
1926 | # else | |
1927 | /* This shouldn't happen but some implementation might still have this | |
1928 | problem. Use a reasonable default value. */ | |
1929 | # define CHAR_CLASS_MAX_LENGTH 256 | |
1930 | # endif | |
1931 | typedef wctype_t re_wctype_t; | |
01618498 | 1932 | typedef wchar_t re_wchar_t; |
14473664 SM |
1933 | # define re_wctype wctype |
1934 | # define re_iswctype iswctype | |
1935 | # define re_wctype_to_bit(cc) 0 | |
1936 | #else | |
1937 | # define CHAR_CLASS_MAX_LENGTH 9 /* Namely, `multibyte'. */ | |
1938 | # define btowc(c) c | |
1939 | ||
01618498 | 1940 | /* Character classes. */ |
14473664 SM |
1941 | typedef enum { RECC_ERROR = 0, |
1942 | RECC_ALNUM, RECC_ALPHA, RECC_WORD, | |
1943 | RECC_GRAPH, RECC_PRINT, | |
1944 | RECC_LOWER, RECC_UPPER, | |
1945 | RECC_PUNCT, RECC_CNTRL, | |
1946 | RECC_DIGIT, RECC_XDIGIT, | |
1947 | RECC_BLANK, RECC_SPACE, | |
1948 | RECC_MULTIBYTE, RECC_NONASCII, | |
1949 | RECC_ASCII, RECC_UNIBYTE | |
1950 | } re_wctype_t; | |
1951 | ||
01618498 SM |
1952 | typedef int re_wchar_t; |
1953 | ||
14473664 SM |
1954 | /* Map a string to the char class it names (if any). */ |
1955 | static re_wctype_t | |
1956 | re_wctype (string) | |
01618498 | 1957 | re_char *string; |
14473664 SM |
1958 | { |
1959 | if (STREQ (string, "alnum")) return RECC_ALNUM; | |
1960 | else if (STREQ (string, "alpha")) return RECC_ALPHA; | |
1961 | else if (STREQ (string, "word")) return RECC_WORD; | |
1962 | else if (STREQ (string, "ascii")) return RECC_ASCII; | |
1963 | else if (STREQ (string, "nonascii")) return RECC_NONASCII; | |
1964 | else if (STREQ (string, "graph")) return RECC_GRAPH; | |
1965 | else if (STREQ (string, "lower")) return RECC_LOWER; | |
1966 | else if (STREQ (string, "print")) return RECC_PRINT; | |
1967 | else if (STREQ (string, "punct")) return RECC_PUNCT; | |
1968 | else if (STREQ (string, "space")) return RECC_SPACE; | |
1969 | else if (STREQ (string, "upper")) return RECC_UPPER; | |
1970 | else if (STREQ (string, "unibyte")) return RECC_UNIBYTE; | |
1971 | else if (STREQ (string, "multibyte")) return RECC_MULTIBYTE; | |
1972 | else if (STREQ (string, "digit")) return RECC_DIGIT; | |
1973 | else if (STREQ (string, "xdigit")) return RECC_XDIGIT; | |
1974 | else if (STREQ (string, "cntrl")) return RECC_CNTRL; | |
1975 | else if (STREQ (string, "blank")) return RECC_BLANK; | |
1976 | else return 0; | |
1977 | } | |
1978 | ||
1979 | /* True iff CH is in the char class CC. */ | |
1980 | static boolean | |
1981 | re_iswctype (ch, cc) | |
1982 | int ch; | |
1983 | re_wctype_t cc; | |
1984 | { | |
1985 | switch (cc) | |
1986 | { | |
0cdd06f8 SM |
1987 | case RECC_ALNUM: return ISALNUM (ch); |
1988 | case RECC_ALPHA: return ISALPHA (ch); | |
1989 | case RECC_BLANK: return ISBLANK (ch); | |
1990 | case RECC_CNTRL: return ISCNTRL (ch); | |
1991 | case RECC_DIGIT: return ISDIGIT (ch); | |
1992 | case RECC_GRAPH: return ISGRAPH (ch); | |
1993 | case RECC_LOWER: return ISLOWER (ch); | |
1994 | case RECC_PRINT: return ISPRINT (ch); | |
1995 | case RECC_PUNCT: return ISPUNCT (ch); | |
1996 | case RECC_SPACE: return ISSPACE (ch); | |
1997 | case RECC_UPPER: return ISUPPER (ch); | |
1998 | case RECC_XDIGIT: return ISXDIGIT (ch); | |
1999 | case RECC_ASCII: return IS_REAL_ASCII (ch); | |
2000 | case RECC_NONASCII: return !IS_REAL_ASCII (ch); | |
2001 | case RECC_UNIBYTE: return ISUNIBYTE (ch); | |
2002 | case RECC_MULTIBYTE: return !ISUNIBYTE (ch); | |
2003 | case RECC_WORD: return ISWORD (ch); | |
2004 | case RECC_ERROR: return false; | |
2005 | default: | |
2006 | abort(); | |
14473664 SM |
2007 | } |
2008 | } | |
fa9a63c5 | 2009 | |
14473664 SM |
2010 | /* Return a bit-pattern to use in the range-table bits to match multibyte |
2011 | chars of class CC. */ | |
2012 | static int | |
2013 | re_wctype_to_bit (cc) | |
2014 | re_wctype_t cc; | |
2015 | { | |
2016 | switch (cc) | |
2017 | { | |
2018 | case RECC_NONASCII: case RECC_PRINT: case RECC_GRAPH: | |
0cdd06f8 SM |
2019 | case RECC_MULTIBYTE: return BIT_MULTIBYTE; |
2020 | case RECC_ALPHA: case RECC_ALNUM: case RECC_WORD: return BIT_WORD; | |
2021 | case RECC_LOWER: return BIT_LOWER; | |
2022 | case RECC_UPPER: return BIT_UPPER; | |
2023 | case RECC_PUNCT: return BIT_PUNCT; | |
2024 | case RECC_SPACE: return BIT_SPACE; | |
14473664 | 2025 | case RECC_ASCII: case RECC_DIGIT: case RECC_XDIGIT: case RECC_CNTRL: |
0cdd06f8 SM |
2026 | case RECC_BLANK: case RECC_UNIBYTE: case RECC_ERROR: return 0; |
2027 | default: | |
2028 | abort(); | |
14473664 SM |
2029 | } |
2030 | } | |
2031 | #endif | |
99633e97 | 2032 | |
5b370c2b AI |
2033 | /* Explicit quit checking is only used on NTemacs. */ |
2034 | #if defined WINDOWSNT && defined emacs && defined QUIT | |
2035 | extern int immediate_quit; | |
2036 | # define IMMEDIATE_QUIT_CHECK \ | |
2037 | do { \ | |
2038 | if (immediate_quit) QUIT; \ | |
2039 | } while (0) | |
2040 | #else | |
01618498 | 2041 | # define IMMEDIATE_QUIT_CHECK ((void)0) |
99633e97 | 2042 | #endif |
fa9a63c5 RM |
2043 | \f |
2044 | #ifndef MATCH_MAY_ALLOCATE | |
2045 | ||
2046 | /* If we cannot allocate large objects within re_match_2_internal, | |
2047 | we make the fail stack and register vectors global. | |
2048 | The fail stack, we grow to the maximum size when a regexp | |
2049 | is compiled. | |
2050 | The register vectors, we adjust in size each time we | |
2051 | compile a regexp, according to the number of registers it needs. */ | |
2052 | ||
2053 | static fail_stack_type fail_stack; | |
2054 | ||
2055 | /* Size with which the following vectors are currently allocated. | |
2056 | That is so we can make them bigger as needed, | |
4bb91c68 | 2057 | but never make them smaller. */ |
fa9a63c5 RM |
2058 | static int regs_allocated_size; |
2059 | ||
66f0296e SM |
2060 | static re_char ** regstart, ** regend; |
2061 | static re_char **best_regstart, **best_regend; | |
fa9a63c5 RM |
2062 | |
2063 | /* Make the register vectors big enough for NUM_REGS registers, | |
4bb91c68 | 2064 | but don't make them smaller. */ |
fa9a63c5 RM |
2065 | |
2066 | static | |
2067 | regex_grow_registers (num_regs) | |
2068 | int num_regs; | |
2069 | { | |
2070 | if (num_regs > regs_allocated_size) | |
2071 | { | |
66f0296e SM |
2072 | RETALLOC_IF (regstart, num_regs, re_char *); |
2073 | RETALLOC_IF (regend, num_regs, re_char *); | |
2074 | RETALLOC_IF (best_regstart, num_regs, re_char *); | |
2075 | RETALLOC_IF (best_regend, num_regs, re_char *); | |
fa9a63c5 RM |
2076 | |
2077 | regs_allocated_size = num_regs; | |
2078 | } | |
2079 | } | |
2080 | ||
2081 | #endif /* not MATCH_MAY_ALLOCATE */ | |
2082 | \f | |
99633e97 SM |
2083 | static boolean group_in_compile_stack _RE_ARGS ((compile_stack_type |
2084 | compile_stack, | |
2085 | regnum_t regnum)); | |
2086 | ||
fa9a63c5 RM |
2087 | /* `regex_compile' compiles PATTERN (of length SIZE) according to SYNTAX. |
2088 | Returns one of error codes defined in `regex.h', or zero for success. | |
2089 | ||
2090 | Assumes the `allocated' (and perhaps `buffer') and `translate' | |
2091 | fields are set in BUFP on entry. | |
2092 | ||
2093 | If it succeeds, results are put in BUFP (if it returns an error, the | |
2094 | contents of BUFP are undefined): | |
2095 | `buffer' is the compiled pattern; | |
2096 | `syntax' is set to SYNTAX; | |
2097 | `used' is set to the length of the compiled pattern; | |
2098 | `fastmap_accurate' is zero; | |
2099 | `re_nsub' is the number of subexpressions in PATTERN; | |
2100 | `not_bol' and `not_eol' are zero; | |
5e69f11e | 2101 | |
c0f9ea08 | 2102 | The `fastmap' field is neither examined nor set. */ |
fa9a63c5 | 2103 | |
505bde11 SM |
2104 | /* Insert the `jump' from the end of last alternative to "here". |
2105 | The space for the jump has already been allocated. */ | |
2106 | #define FIXUP_ALT_JUMP() \ | |
2107 | do { \ | |
2108 | if (fixup_alt_jump) \ | |
2109 | STORE_JUMP (jump, fixup_alt_jump, b); \ | |
2110 | } while (0) | |
2111 | ||
2112 | ||
fa9a63c5 RM |
2113 | /* Return, freeing storage we allocated. */ |
2114 | #define FREE_STACK_RETURN(value) \ | |
b18215fc RS |
2115 | do { \ |
2116 | FREE_RANGE_TABLE_WORK_AREA (range_table_work); \ | |
2117 | free (compile_stack.stack); \ | |
2118 | return value; \ | |
2119 | } while (0) | |
fa9a63c5 RM |
2120 | |
2121 | static reg_errcode_t | |
2122 | regex_compile (pattern, size, syntax, bufp) | |
66f0296e | 2123 | re_char *pattern; |
4bb91c68 | 2124 | size_t size; |
fa9a63c5 RM |
2125 | reg_syntax_t syntax; |
2126 | struct re_pattern_buffer *bufp; | |
2127 | { | |
01618498 SM |
2128 | /* We fetch characters from PATTERN here. */ |
2129 | register re_wchar_t c, c1; | |
5e69f11e | 2130 | |
fa9a63c5 | 2131 | /* A random temporary spot in PATTERN. */ |
66f0296e | 2132 | re_char *p1; |
fa9a63c5 RM |
2133 | |
2134 | /* Points to the end of the buffer, where we should append. */ | |
2135 | register unsigned char *b; | |
5e69f11e | 2136 | |
fa9a63c5 RM |
2137 | /* Keeps track of unclosed groups. */ |
2138 | compile_stack_type compile_stack; | |
2139 | ||
2140 | /* Points to the current (ending) position in the pattern. */ | |
22336245 RS |
2141 | #ifdef AIX |
2142 | /* `const' makes AIX compiler fail. */ | |
66f0296e | 2143 | unsigned char *p = pattern; |
22336245 | 2144 | #else |
66f0296e | 2145 | re_char *p = pattern; |
22336245 | 2146 | #endif |
66f0296e | 2147 | re_char *pend = pattern + size; |
5e69f11e | 2148 | |
fa9a63c5 | 2149 | /* How to translate the characters in the pattern. */ |
6676cb1c | 2150 | RE_TRANSLATE_TYPE translate = bufp->translate; |
fa9a63c5 RM |
2151 | |
2152 | /* Address of the count-byte of the most recently inserted `exactn' | |
2153 | command. This makes it possible to tell if a new exact-match | |
2154 | character can be added to that command or if the character requires | |
2155 | a new `exactn' command. */ | |
2156 | unsigned char *pending_exact = 0; | |
2157 | ||
2158 | /* Address of start of the most recently finished expression. | |
2159 | This tells, e.g., postfix * where to find the start of its | |
2160 | operand. Reset at the beginning of groups and alternatives. */ | |
2161 | unsigned char *laststart = 0; | |
2162 | ||
2163 | /* Address of beginning of regexp, or inside of last group. */ | |
2164 | unsigned char *begalt; | |
2165 | ||
2166 | /* Place in the uncompiled pattern (i.e., the {) to | |
2167 | which to go back if the interval is invalid. */ | |
66f0296e | 2168 | re_char *beg_interval; |
5e69f11e | 2169 | |
fa9a63c5 | 2170 | /* Address of the place where a forward jump should go to the end of |
25fe55af | 2171 | the containing expression. Each alternative of an `or' -- except the |
fa9a63c5 RM |
2172 | last -- ends with a forward jump of this sort. */ |
2173 | unsigned char *fixup_alt_jump = 0; | |
2174 | ||
2175 | /* Counts open-groups as they are encountered. Remembered for the | |
2176 | matching close-group on the compile stack, so the same register | |
2177 | number is put in the stop_memory as the start_memory. */ | |
2178 | regnum_t regnum = 0; | |
2179 | ||
b18215fc RS |
2180 | /* Work area for range table of charset. */ |
2181 | struct range_table_work_area range_table_work; | |
2182 | ||
2d1675e4 SM |
2183 | /* If the object matched can contain multibyte characters. */ |
2184 | const boolean multibyte = RE_MULTIBYTE_P (bufp); | |
2185 | ||
fa9a63c5 | 2186 | #ifdef DEBUG |
99633e97 | 2187 | debug++; |
fa9a63c5 | 2188 | DEBUG_PRINT1 ("\nCompiling pattern: "); |
99633e97 | 2189 | if (debug > 0) |
fa9a63c5 RM |
2190 | { |
2191 | unsigned debug_count; | |
5e69f11e | 2192 | |
fa9a63c5 | 2193 | for (debug_count = 0; debug_count < size; debug_count++) |
25fe55af | 2194 | putchar (pattern[debug_count]); |
fa9a63c5 RM |
2195 | putchar ('\n'); |
2196 | } | |
2197 | #endif /* DEBUG */ | |
2198 | ||
2199 | /* Initialize the compile stack. */ | |
2200 | compile_stack.stack = TALLOC (INIT_COMPILE_STACK_SIZE, compile_stack_elt_t); | |
2201 | if (compile_stack.stack == NULL) | |
2202 | return REG_ESPACE; | |
2203 | ||
2204 | compile_stack.size = INIT_COMPILE_STACK_SIZE; | |
2205 | compile_stack.avail = 0; | |
2206 | ||
b18215fc RS |
2207 | range_table_work.table = 0; |
2208 | range_table_work.allocated = 0; | |
2209 | ||
fa9a63c5 RM |
2210 | /* Initialize the pattern buffer. */ |
2211 | bufp->syntax = syntax; | |
2212 | bufp->fastmap_accurate = 0; | |
2213 | bufp->not_bol = bufp->not_eol = 0; | |
2214 | ||
2215 | /* Set `used' to zero, so that if we return an error, the pattern | |
2216 | printer (for debugging) will think there's no pattern. We reset it | |
2217 | at the end. */ | |
2218 | bufp->used = 0; | |
5e69f11e | 2219 | |
fa9a63c5 | 2220 | /* Always count groups, whether or not bufp->no_sub is set. */ |
5e69f11e | 2221 | bufp->re_nsub = 0; |
fa9a63c5 | 2222 | |
0b32bf0e | 2223 | #if !defined emacs && !defined SYNTAX_TABLE |
fa9a63c5 RM |
2224 | /* Initialize the syntax table. */ |
2225 | init_syntax_once (); | |
2226 | #endif | |
2227 | ||
2228 | if (bufp->allocated == 0) | |
2229 | { | |
2230 | if (bufp->buffer) | |
2231 | { /* If zero allocated, but buffer is non-null, try to realloc | |
25fe55af RS |
2232 | enough space. This loses if buffer's address is bogus, but |
2233 | that is the user's responsibility. */ | |
2234 | RETALLOC (bufp->buffer, INIT_BUF_SIZE, unsigned char); | |
2235 | } | |
fa9a63c5 | 2236 | else |
25fe55af RS |
2237 | { /* Caller did not allocate a buffer. Do it for them. */ |
2238 | bufp->buffer = TALLOC (INIT_BUF_SIZE, unsigned char); | |
2239 | } | |
fa9a63c5 RM |
2240 | if (!bufp->buffer) FREE_STACK_RETURN (REG_ESPACE); |
2241 | ||
2242 | bufp->allocated = INIT_BUF_SIZE; | |
2243 | } | |
2244 | ||
2245 | begalt = b = bufp->buffer; | |
2246 | ||
2247 | /* Loop through the uncompiled pattern until we're at the end. */ | |
2248 | while (p != pend) | |
2249 | { | |
2250 | PATFETCH (c); | |
2251 | ||
2252 | switch (c) | |
25fe55af RS |
2253 | { |
2254 | case '^': | |
2255 | { | |
2256 | if ( /* If at start of pattern, it's an operator. */ | |
2257 | p == pattern + 1 | |
2258 | /* If context independent, it's an operator. */ | |
2259 | || syntax & RE_CONTEXT_INDEP_ANCHORS | |
2260 | /* Otherwise, depends on what's come before. */ | |
2261 | || at_begline_loc_p (pattern, p, syntax)) | |
c0f9ea08 | 2262 | BUF_PUSH ((syntax & RE_NO_NEWLINE_ANCHOR) ? begbuf : begline); |
25fe55af RS |
2263 | else |
2264 | goto normal_char; | |
2265 | } | |
2266 | break; | |
2267 | ||
2268 | ||
2269 | case '$': | |
2270 | { | |
2271 | if ( /* If at end of pattern, it's an operator. */ | |
2272 | p == pend | |
2273 | /* If context independent, it's an operator. */ | |
2274 | || syntax & RE_CONTEXT_INDEP_ANCHORS | |
2275 | /* Otherwise, depends on what's next. */ | |
2276 | || at_endline_loc_p (p, pend, syntax)) | |
c0f9ea08 | 2277 | BUF_PUSH ((syntax & RE_NO_NEWLINE_ANCHOR) ? endbuf : endline); |
25fe55af RS |
2278 | else |
2279 | goto normal_char; | |
2280 | } | |
2281 | break; | |
fa9a63c5 RM |
2282 | |
2283 | ||
2284 | case '+': | |
25fe55af RS |
2285 | case '?': |
2286 | if ((syntax & RE_BK_PLUS_QM) | |
2287 | || (syntax & RE_LIMITED_OPS)) | |
2288 | goto normal_char; | |
2289 | handle_plus: | |
2290 | case '*': | |
2291 | /* If there is no previous pattern... */ | |
2292 | if (!laststart) | |
2293 | { | |
2294 | if (syntax & RE_CONTEXT_INVALID_OPS) | |
2295 | FREE_STACK_RETURN (REG_BADRPT); | |
2296 | else if (!(syntax & RE_CONTEXT_INDEP_OPS)) | |
2297 | goto normal_char; | |
2298 | } | |
2299 | ||
2300 | { | |
25fe55af | 2301 | /* 1 means zero (many) matches is allowed. */ |
66f0296e SM |
2302 | boolean zero_times_ok = 0, many_times_ok = 0; |
2303 | boolean greedy = 1; | |
25fe55af RS |
2304 | |
2305 | /* If there is a sequence of repetition chars, collapse it | |
2306 | down to just one (the right one). We can't combine | |
2307 | interval operators with these because of, e.g., `a{2}*', | |
2308 | which should only match an even number of `a's. */ | |
2309 | ||
2310 | for (;;) | |
2311 | { | |
0b32bf0e | 2312 | if ((syntax & RE_FRUGAL) |
1c8c6d39 DL |
2313 | && c == '?' && (zero_times_ok || many_times_ok)) |
2314 | greedy = 0; | |
2315 | else | |
2316 | { | |
2317 | zero_times_ok |= c != '+'; | |
2318 | many_times_ok |= c != '?'; | |
2319 | } | |
25fe55af RS |
2320 | |
2321 | if (p == pend) | |
2322 | break; | |
ed0767d8 SM |
2323 | else if (*p == '*' |
2324 | || (!(syntax & RE_BK_PLUS_QM) | |
2325 | && (*p == '+' || *p == '?'))) | |
25fe55af | 2326 | ; |
ed0767d8 | 2327 | else if (syntax & RE_BK_PLUS_QM && *p == '\\') |
25fe55af | 2328 | { |
ed0767d8 SM |
2329 | if (p+1 == pend) |
2330 | FREE_STACK_RETURN (REG_EESCAPE); | |
2331 | if (p[1] == '+' || p[1] == '?') | |
2332 | PATFETCH (c); /* Gobble up the backslash. */ | |
2333 | else | |
2334 | break; | |
25fe55af RS |
2335 | } |
2336 | else | |
ed0767d8 | 2337 | break; |
25fe55af | 2338 | /* If we get here, we found another repeat character. */ |
ed0767d8 SM |
2339 | PATFETCH (c); |
2340 | } | |
25fe55af RS |
2341 | |
2342 | /* Star, etc. applied to an empty pattern is equivalent | |
2343 | to an empty pattern. */ | |
4e8a9132 | 2344 | if (!laststart || laststart == b) |
25fe55af RS |
2345 | break; |
2346 | ||
2347 | /* Now we know whether or not zero matches is allowed | |
2348 | and also whether or not two or more matches is allowed. */ | |
1c8c6d39 DL |
2349 | if (greedy) |
2350 | { | |
99633e97 | 2351 | if (many_times_ok) |
4e8a9132 SM |
2352 | { |
2353 | boolean simple = skip_one_char (laststart) == b; | |
2354 | unsigned int startoffset = 0; | |
f6a3f532 | 2355 | re_opcode_t ofj = |
01618498 | 2356 | /* Check if the loop can match the empty string. */ |
f6a3f532 SM |
2357 | (simple || !analyse_first (laststart, b, NULL, 0)) ? |
2358 | on_failure_jump : on_failure_jump_loop; | |
4e8a9132 SM |
2359 | assert (skip_one_char (laststart) <= b); |
2360 | ||
2361 | if (!zero_times_ok && simple) | |
2362 | { /* Since simple * loops can be made faster by using | |
2363 | on_failure_keep_string_jump, we turn simple P+ | |
2364 | into PP* if P is simple. */ | |
2365 | unsigned char *p1, *p2; | |
2366 | startoffset = b - laststart; | |
2367 | GET_BUFFER_SPACE (startoffset); | |
2368 | p1 = b; p2 = laststart; | |
2369 | while (p2 < p1) | |
2370 | *b++ = *p2++; | |
2371 | zero_times_ok = 1; | |
99633e97 | 2372 | } |
4e8a9132 SM |
2373 | |
2374 | GET_BUFFER_SPACE (6); | |
2375 | if (!zero_times_ok) | |
2376 | /* A + loop. */ | |
f6a3f532 | 2377 | STORE_JUMP (ofj, b, b + 6); |
99633e97 | 2378 | else |
4e8a9132 SM |
2379 | /* Simple * loops can use on_failure_keep_string_jump |
2380 | depending on what follows. But since we don't know | |
2381 | that yet, we leave the decision up to | |
2382 | on_failure_jump_smart. */ | |
f6a3f532 | 2383 | INSERT_JUMP (simple ? on_failure_jump_smart : ofj, |
4e8a9132 | 2384 | laststart + startoffset, b + 6); |
99633e97 | 2385 | b += 3; |
4e8a9132 | 2386 | STORE_JUMP (jump, b, laststart + startoffset); |
99633e97 SM |
2387 | b += 3; |
2388 | } | |
2389 | else | |
2390 | { | |
4e8a9132 SM |
2391 | /* A simple ? pattern. */ |
2392 | assert (zero_times_ok); | |
2393 | GET_BUFFER_SPACE (3); | |
2394 | INSERT_JUMP (on_failure_jump, laststart, b + 3); | |
99633e97 SM |
2395 | b += 3; |
2396 | } | |
1c8c6d39 DL |
2397 | } |
2398 | else /* not greedy */ | |
2399 | { /* I wish the greedy and non-greedy cases could be merged. */ | |
2400 | ||
0683b6fa | 2401 | GET_BUFFER_SPACE (7); /* We might use less. */ |
1c8c6d39 DL |
2402 | if (many_times_ok) |
2403 | { | |
f6a3f532 SM |
2404 | boolean emptyp = analyse_first (laststart, b, NULL, 0); |
2405 | ||
505bde11 | 2406 | /* The non-greedy multiple match looks like a repeat..until: |
1c8c6d39 | 2407 | we only need a conditional jump at the end of the loop */ |
f6a3f532 SM |
2408 | if (emptyp) BUF_PUSH (no_op); |
2409 | STORE_JUMP (emptyp ? on_failure_jump_nastyloop | |
2410 | : on_failure_jump, b, laststart); | |
1c8c6d39 DL |
2411 | b += 3; |
2412 | if (zero_times_ok) | |
2413 | { | |
2414 | /* The repeat...until naturally matches one or more. | |
2415 | To also match zero times, we need to first jump to | |
2416 | the end of the loop (its conditional jump). */ | |
1c8c6d39 DL |
2417 | INSERT_JUMP (jump, laststart, b); |
2418 | b += 3; | |
2419 | } | |
2420 | } | |
2421 | else | |
2422 | { | |
2423 | /* non-greedy a?? */ | |
1c8c6d39 DL |
2424 | INSERT_JUMP (jump, laststart, b + 3); |
2425 | b += 3; | |
2426 | INSERT_JUMP (on_failure_jump, laststart, laststart + 6); | |
2427 | b += 3; | |
2428 | } | |
2429 | } | |
2430 | } | |
4e8a9132 | 2431 | pending_exact = 0; |
fa9a63c5 RM |
2432 | break; |
2433 | ||
2434 | ||
2435 | case '.': | |
25fe55af RS |
2436 | laststart = b; |
2437 | BUF_PUSH (anychar); | |
2438 | break; | |
fa9a63c5 RM |
2439 | |
2440 | ||
25fe55af RS |
2441 | case '[': |
2442 | { | |
b18215fc | 2443 | CLEAR_RANGE_TABLE_WORK_USED (range_table_work); |
fa9a63c5 | 2444 | |
25fe55af | 2445 | if (p == pend) FREE_STACK_RETURN (REG_EBRACK); |
fa9a63c5 | 2446 | |
25fe55af RS |
2447 | /* Ensure that we have enough space to push a charset: the |
2448 | opcode, the length count, and the bitset; 34 bytes in all. */ | |
fa9a63c5 RM |
2449 | GET_BUFFER_SPACE (34); |
2450 | ||
25fe55af | 2451 | laststart = b; |
e318085a | 2452 | |
25fe55af RS |
2453 | /* We test `*p == '^' twice, instead of using an if |
2454 | statement, so we only need one BUF_PUSH. */ | |
2455 | BUF_PUSH (*p == '^' ? charset_not : charset); | |
2456 | if (*p == '^') | |
2457 | p++; | |
e318085a | 2458 | |
25fe55af RS |
2459 | /* Remember the first position in the bracket expression. */ |
2460 | p1 = p; | |
e318085a | 2461 | |
25fe55af RS |
2462 | /* Push the number of bytes in the bitmap. */ |
2463 | BUF_PUSH ((1 << BYTEWIDTH) / BYTEWIDTH); | |
e318085a | 2464 | |
25fe55af RS |
2465 | /* Clear the whole map. */ |
2466 | bzero (b, (1 << BYTEWIDTH) / BYTEWIDTH); | |
e318085a | 2467 | |
25fe55af RS |
2468 | /* charset_not matches newline according to a syntax bit. */ |
2469 | if ((re_opcode_t) b[-2] == charset_not | |
2470 | && (syntax & RE_HAT_LISTS_NOT_NEWLINE)) | |
2471 | SET_LIST_BIT ('\n'); | |
fa9a63c5 | 2472 | |
25fe55af RS |
2473 | /* Read in characters and ranges, setting map bits. */ |
2474 | for (;;) | |
2475 | { | |
b18215fc | 2476 | boolean escaped_char = false; |
2d1675e4 | 2477 | const unsigned char *p2 = p; |
e318085a | 2478 | |
25fe55af | 2479 | if (p == pend) FREE_STACK_RETURN (REG_EBRACK); |
e318085a | 2480 | |
25fe55af | 2481 | PATFETCH (c); |
e318085a | 2482 | |
25fe55af RS |
2483 | /* \ might escape characters inside [...] and [^...]. */ |
2484 | if ((syntax & RE_BACKSLASH_ESCAPE_IN_LISTS) && c == '\\') | |
2485 | { | |
2486 | if (p == pend) FREE_STACK_RETURN (REG_EESCAPE); | |
e318085a RS |
2487 | |
2488 | PATFETCH (c); | |
b18215fc | 2489 | escaped_char = true; |
25fe55af | 2490 | } |
b18215fc RS |
2491 | else |
2492 | { | |
657fcfbd RS |
2493 | /* Could be the end of the bracket expression. If it's |
2494 | not (i.e., when the bracket expression is `[]' so | |
2495 | far), the ']' character bit gets set way below. */ | |
2d1675e4 | 2496 | if (c == ']' && p2 != p1) |
657fcfbd | 2497 | break; |
25fe55af | 2498 | } |
b18215fc | 2499 | |
b18215fc RS |
2500 | /* What should we do for the character which is |
2501 | greater than 0x7F, but not BASE_LEADING_CODE_P? | |
2502 | XXX */ | |
2503 | ||
25fe55af RS |
2504 | /* See if we're at the beginning of a possible character |
2505 | class. */ | |
b18215fc | 2506 | |
2d1675e4 SM |
2507 | if (!escaped_char && |
2508 | syntax & RE_CHAR_CLASSES && c == '[' && *p == ':') | |
657fcfbd RS |
2509 | { |
2510 | /* Leave room for the null. */ | |
14473664 | 2511 | unsigned char str[CHAR_CLASS_MAX_LENGTH + 1]; |
ed0767d8 | 2512 | const unsigned char *class_beg; |
b18215fc | 2513 | |
25fe55af RS |
2514 | PATFETCH (c); |
2515 | c1 = 0; | |
ed0767d8 | 2516 | class_beg = p; |
b18215fc | 2517 | |
25fe55af RS |
2518 | /* If pattern is `[[:'. */ |
2519 | if (p == pend) FREE_STACK_RETURN (REG_EBRACK); | |
b18215fc | 2520 | |
25fe55af RS |
2521 | for (;;) |
2522 | { | |
14473664 SM |
2523 | PATFETCH (c); |
2524 | if ((c == ':' && *p == ']') || p == pend) | |
2525 | break; | |
2526 | if (c1 < CHAR_CLASS_MAX_LENGTH) | |
2527 | str[c1++] = c; | |
2528 | else | |
2529 | /* This is in any case an invalid class name. */ | |
2530 | str[0] = '\0'; | |
25fe55af RS |
2531 | } |
2532 | str[c1] = '\0'; | |
b18215fc RS |
2533 | |
2534 | /* If isn't a word bracketed by `[:' and `:]': | |
2535 | undo the ending character, the letters, and | |
2536 | leave the leading `:' and `[' (but set bits for | |
2537 | them). */ | |
25fe55af RS |
2538 | if (c == ':' && *p == ']') |
2539 | { | |
2540 | int ch; | |
14473664 SM |
2541 | re_wctype_t cc; |
2542 | ||
2543 | cc = re_wctype (str); | |
2544 | ||
2545 | if (cc == 0) | |
fa9a63c5 RM |
2546 | FREE_STACK_RETURN (REG_ECTYPE); |
2547 | ||
14473664 SM |
2548 | /* Throw away the ] at the end of the character |
2549 | class. */ | |
2550 | PATFETCH (c); | |
fa9a63c5 | 2551 | |
14473664 | 2552 | if (p == pend) FREE_STACK_RETURN (REG_EBRACK); |
fa9a63c5 | 2553 | |
96cc36cc RS |
2554 | /* Most character classes in a multibyte match |
2555 | just set a flag. Exceptions are is_blank, | |
2556 | is_digit, is_cntrl, and is_xdigit, since | |
2557 | they can only match ASCII characters. We | |
14473664 SM |
2558 | don't need to handle them for multibyte. |
2559 | They are distinguished by a negative wctype. */ | |
96cc36cc | 2560 | |
2d1675e4 | 2561 | if (multibyte) |
14473664 SM |
2562 | SET_RANGE_TABLE_WORK_AREA_BIT (range_table_work, |
2563 | re_wctype_to_bit (cc)); | |
96cc36cc | 2564 | |
14473664 | 2565 | for (ch = 0; ch < 1 << BYTEWIDTH; ++ch) |
25fe55af | 2566 | { |
7ae68633 | 2567 | int translated = TRANSLATE (ch); |
14473664 | 2568 | if (re_iswctype (btowc (ch), cc)) |
96cc36cc | 2569 | SET_LIST_BIT (translated); |
25fe55af | 2570 | } |
b18215fc RS |
2571 | |
2572 | /* Repeat the loop. */ | |
2573 | continue; | |
25fe55af RS |
2574 | } |
2575 | else | |
2576 | { | |
ed0767d8 SM |
2577 | /* Go back to right after the "[:". */ |
2578 | p = class_beg; | |
25fe55af | 2579 | SET_LIST_BIT ('['); |
b18215fc RS |
2580 | |
2581 | /* Because the `:' may starts the range, we | |
2582 | can't simply set bit and repeat the loop. | |
25fe55af | 2583 | Instead, just set it to C and handle below. */ |
b18215fc | 2584 | c = ':'; |
25fe55af RS |
2585 | } |
2586 | } | |
b18215fc RS |
2587 | |
2588 | if (p < pend && p[0] == '-' && p[1] != ']') | |
2589 | { | |
2590 | ||
2591 | /* Discard the `-'. */ | |
2592 | PATFETCH (c1); | |
2593 | ||
2594 | /* Fetch the character which ends the range. */ | |
2595 | PATFETCH (c1); | |
b18215fc | 2596 | |
b54f61ed | 2597 | if (SINGLE_BYTE_CHAR_P (c)) |
e934739e | 2598 | { |
b54f61ed KH |
2599 | if (! SINGLE_BYTE_CHAR_P (c1)) |
2600 | { | |
3ff2446d KH |
2601 | /* Handle a range starting with a |
2602 | character of less than 256, and ending | |
2603 | with a character of not less than 256. | |
2604 | Split that into two ranges, the low one | |
2605 | ending at 0377, and the high one | |
2606 | starting at the smallest character in | |
2607 | the charset of C1 and ending at C1. */ | |
b54f61ed KH |
2608 | int charset = CHAR_CHARSET (c1); |
2609 | int c2 = MAKE_CHAR (charset, 0, 0); | |
2610 | ||
2611 | SET_RANGE_TABLE_WORK_AREA (range_table_work, | |
2612 | c2, c1); | |
333526e0 | 2613 | c1 = 0377; |
b54f61ed | 2614 | } |
e934739e RS |
2615 | } |
2616 | else if (!SAME_CHARSET_P (c, c1)) | |
b18215fc | 2617 | FREE_STACK_RETURN (REG_ERANGE); |
e318085a | 2618 | } |
25fe55af | 2619 | else |
b18215fc RS |
2620 | /* Range from C to C. */ |
2621 | c1 = c; | |
2622 | ||
2623 | /* Set the range ... */ | |
2624 | if (SINGLE_BYTE_CHAR_P (c)) | |
2625 | /* ... into bitmap. */ | |
25fe55af | 2626 | { |
01618498 | 2627 | re_wchar_t this_char; |
b18215fc RS |
2628 | int range_start = c, range_end = c1; |
2629 | ||
2630 | /* If the start is after the end, the range is empty. */ | |
2631 | if (range_start > range_end) | |
2632 | { | |
2633 | if (syntax & RE_NO_EMPTY_RANGES) | |
2634 | FREE_STACK_RETURN (REG_ERANGE); | |
2635 | /* Else, repeat the loop. */ | |
2636 | } | |
2637 | else | |
2638 | { | |
2639 | for (this_char = range_start; this_char <= range_end; | |
2640 | this_char++) | |
2641 | SET_LIST_BIT (TRANSLATE (this_char)); | |
e934739e | 2642 | } |
25fe55af | 2643 | } |
e318085a | 2644 | else |
b18215fc RS |
2645 | /* ... into range table. */ |
2646 | SET_RANGE_TABLE_WORK_AREA (range_table_work, c, c1); | |
e318085a RS |
2647 | } |
2648 | ||
25fe55af RS |
2649 | /* Discard any (non)matching list bytes that are all 0 at the |
2650 | end of the map. Decrease the map-length byte too. */ | |
2651 | while ((int) b[-1] > 0 && b[b[-1] - 1] == 0) | |
2652 | b[-1]--; | |
2653 | b += b[-1]; | |
fa9a63c5 | 2654 | |
96cc36cc RS |
2655 | /* Build real range table from work area. */ |
2656 | if (RANGE_TABLE_WORK_USED (range_table_work) | |
2657 | || RANGE_TABLE_WORK_BITS (range_table_work)) | |
b18215fc RS |
2658 | { |
2659 | int i; | |
2660 | int used = RANGE_TABLE_WORK_USED (range_table_work); | |
fa9a63c5 | 2661 | |
b18215fc | 2662 | /* Allocate space for COUNT + RANGE_TABLE. Needs two |
96cc36cc RS |
2663 | bytes for flags, two for COUNT, and three bytes for |
2664 | each character. */ | |
2665 | GET_BUFFER_SPACE (4 + used * 3); | |
fa9a63c5 | 2666 | |
b18215fc RS |
2667 | /* Indicate the existence of range table. */ |
2668 | laststart[1] |= 0x80; | |
fa9a63c5 | 2669 | |
96cc36cc RS |
2670 | /* Store the character class flag bits into the range table. |
2671 | If not in emacs, these flag bits are always 0. */ | |
2672 | *b++ = RANGE_TABLE_WORK_BITS (range_table_work) & 0xff; | |
2673 | *b++ = RANGE_TABLE_WORK_BITS (range_table_work) >> 8; | |
2674 | ||
b18215fc RS |
2675 | STORE_NUMBER_AND_INCR (b, used / 2); |
2676 | for (i = 0; i < used; i++) | |
2677 | STORE_CHARACTER_AND_INCR | |
2678 | (b, RANGE_TABLE_WORK_ELT (range_table_work, i)); | |
2679 | } | |
25fe55af RS |
2680 | } |
2681 | break; | |
fa9a63c5 RM |
2682 | |
2683 | ||
b18215fc | 2684 | case '(': |
25fe55af RS |
2685 | if (syntax & RE_NO_BK_PARENS) |
2686 | goto handle_open; | |
2687 | else | |
2688 | goto normal_char; | |
fa9a63c5 RM |
2689 | |
2690 | ||
25fe55af RS |
2691 | case ')': |
2692 | if (syntax & RE_NO_BK_PARENS) | |
2693 | goto handle_close; | |
2694 | else | |
2695 | goto normal_char; | |
e318085a RS |
2696 | |
2697 | ||
25fe55af RS |
2698 | case '\n': |
2699 | if (syntax & RE_NEWLINE_ALT) | |
2700 | goto handle_alt; | |
2701 | else | |
2702 | goto normal_char; | |
e318085a RS |
2703 | |
2704 | ||
b18215fc | 2705 | case '|': |
25fe55af RS |
2706 | if (syntax & RE_NO_BK_VBAR) |
2707 | goto handle_alt; | |
2708 | else | |
2709 | goto normal_char; | |
2710 | ||
2711 | ||
2712 | case '{': | |
2713 | if (syntax & RE_INTERVALS && syntax & RE_NO_BK_BRACES) | |
2714 | goto handle_interval; | |
2715 | else | |
2716 | goto normal_char; | |
2717 | ||
2718 | ||
2719 | case '\\': | |
2720 | if (p == pend) FREE_STACK_RETURN (REG_EESCAPE); | |
2721 | ||
2722 | /* Do not translate the character after the \, so that we can | |
2723 | distinguish, e.g., \B from \b, even if we normally would | |
2724 | translate, e.g., B to b. */ | |
2725 | PATFETCH_RAW (c); | |
2726 | ||
2727 | switch (c) | |
2728 | { | |
2729 | case '(': | |
2730 | if (syntax & RE_NO_BK_PARENS) | |
2731 | goto normal_backslash; | |
2732 | ||
2733 | handle_open: | |
505bde11 SM |
2734 | { |
2735 | int shy = 0; | |
2736 | if (p+1 < pend) | |
2737 | { | |
2738 | /* Look for a special (?...) construct */ | |
ed0767d8 | 2739 | if ((syntax & RE_SHY_GROUPS) && *p == '?') |
505bde11 | 2740 | { |
ed0767d8 | 2741 | PATFETCH (c); /* Gobble up the '?'. */ |
505bde11 SM |
2742 | PATFETCH (c); |
2743 | switch (c) | |
2744 | { | |
2745 | case ':': shy = 1; break; | |
2746 | default: | |
2747 | /* Only (?:...) is supported right now. */ | |
2748 | FREE_STACK_RETURN (REG_BADPAT); | |
2749 | } | |
2750 | } | |
505bde11 SM |
2751 | } |
2752 | ||
2753 | if (!shy) | |
2754 | { | |
2755 | bufp->re_nsub++; | |
2756 | regnum++; | |
2757 | } | |
25fe55af | 2758 | |
99633e97 SM |
2759 | if (COMPILE_STACK_FULL) |
2760 | { | |
2761 | RETALLOC (compile_stack.stack, compile_stack.size << 1, | |
2762 | compile_stack_elt_t); | |
2763 | if (compile_stack.stack == NULL) return REG_ESPACE; | |
25fe55af | 2764 | |
99633e97 SM |
2765 | compile_stack.size <<= 1; |
2766 | } | |
25fe55af | 2767 | |
99633e97 SM |
2768 | /* These are the values to restore when we hit end of this |
2769 | group. They are all relative offsets, so that if the | |
2770 | whole pattern moves because of realloc, they will still | |
2771 | be valid. */ | |
2772 | COMPILE_STACK_TOP.begalt_offset = begalt - bufp->buffer; | |
2773 | COMPILE_STACK_TOP.fixup_alt_jump | |
2774 | = fixup_alt_jump ? fixup_alt_jump - bufp->buffer + 1 : 0; | |
2775 | COMPILE_STACK_TOP.laststart_offset = b - bufp->buffer; | |
2776 | COMPILE_STACK_TOP.regnum = shy ? -regnum : regnum; | |
2777 | ||
2778 | /* Do not push a | |
2779 | start_memory for groups beyond the last one we can | |
2780 | represent in the compiled pattern. */ | |
2781 | if (regnum <= MAX_REGNUM && !shy) | |
2782 | BUF_PUSH_2 (start_memory, regnum); | |
2783 | ||
2784 | compile_stack.avail++; | |
2785 | ||
2786 | fixup_alt_jump = 0; | |
2787 | laststart = 0; | |
2788 | begalt = b; | |
2789 | /* If we've reached MAX_REGNUM groups, then this open | |
2790 | won't actually generate any code, so we'll have to | |
2791 | clear pending_exact explicitly. */ | |
2792 | pending_exact = 0; | |
2793 | break; | |
505bde11 | 2794 | } |
25fe55af RS |
2795 | |
2796 | case ')': | |
2797 | if (syntax & RE_NO_BK_PARENS) goto normal_backslash; | |
2798 | ||
2799 | if (COMPILE_STACK_EMPTY) | |
505bde11 SM |
2800 | { |
2801 | if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD) | |
2802 | goto normal_backslash; | |
2803 | else | |
2804 | FREE_STACK_RETURN (REG_ERPAREN); | |
2805 | } | |
25fe55af RS |
2806 | |
2807 | handle_close: | |
505bde11 | 2808 | FIXUP_ALT_JUMP (); |
25fe55af RS |
2809 | |
2810 | /* See similar code for backslashed left paren above. */ | |
2811 | if (COMPILE_STACK_EMPTY) | |
505bde11 SM |
2812 | { |
2813 | if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD) | |
2814 | goto normal_char; | |
2815 | else | |
2816 | FREE_STACK_RETURN (REG_ERPAREN); | |
2817 | } | |
25fe55af RS |
2818 | |
2819 | /* Since we just checked for an empty stack above, this | |
2820 | ``can't happen''. */ | |
2821 | assert (compile_stack.avail != 0); | |
2822 | { | |
2823 | /* We don't just want to restore into `regnum', because | |
2824 | later groups should continue to be numbered higher, | |
2825 | as in `(ab)c(de)' -- the second group is #2. */ | |
2826 | regnum_t this_group_regnum; | |
2827 | ||
2828 | compile_stack.avail--; | |
2829 | begalt = bufp->buffer + COMPILE_STACK_TOP.begalt_offset; | |
2830 | fixup_alt_jump | |
2831 | = COMPILE_STACK_TOP.fixup_alt_jump | |
2832 | ? bufp->buffer + COMPILE_STACK_TOP.fixup_alt_jump - 1 | |
2833 | : 0; | |
2834 | laststart = bufp->buffer + COMPILE_STACK_TOP.laststart_offset; | |
2835 | this_group_regnum = COMPILE_STACK_TOP.regnum; | |
b18215fc RS |
2836 | /* If we've reached MAX_REGNUM groups, then this open |
2837 | won't actually generate any code, so we'll have to | |
2838 | clear pending_exact explicitly. */ | |
2839 | pending_exact = 0; | |
e318085a | 2840 | |
25fe55af RS |
2841 | /* We're at the end of the group, so now we know how many |
2842 | groups were inside this one. */ | |
505bde11 SM |
2843 | if (this_group_regnum <= MAX_REGNUM && this_group_regnum > 0) |
2844 | BUF_PUSH_2 (stop_memory, this_group_regnum); | |
25fe55af RS |
2845 | } |
2846 | break; | |
2847 | ||
2848 | ||
2849 | case '|': /* `\|'. */ | |
2850 | if (syntax & RE_LIMITED_OPS || syntax & RE_NO_BK_VBAR) | |
2851 | goto normal_backslash; | |
2852 | handle_alt: | |
2853 | if (syntax & RE_LIMITED_OPS) | |
2854 | goto normal_char; | |
2855 | ||
2856 | /* Insert before the previous alternative a jump which | |
2857 | jumps to this alternative if the former fails. */ | |
2858 | GET_BUFFER_SPACE (3); | |
2859 | INSERT_JUMP (on_failure_jump, begalt, b + 6); | |
2860 | pending_exact = 0; | |
2861 | b += 3; | |
2862 | ||
2863 | /* The alternative before this one has a jump after it | |
2864 | which gets executed if it gets matched. Adjust that | |
2865 | jump so it will jump to this alternative's analogous | |
2866 | jump (put in below, which in turn will jump to the next | |
2867 | (if any) alternative's such jump, etc.). The last such | |
2868 | jump jumps to the correct final destination. A picture: | |
2869 | _____ _____ | |
2870 | | | | | | |
2871 | | v | v | |
2872 | a | b | c | |
2873 | ||
2874 | If we are at `b', then fixup_alt_jump right now points to a | |
2875 | three-byte space after `a'. We'll put in the jump, set | |
2876 | fixup_alt_jump to right after `b', and leave behind three | |
2877 | bytes which we'll fill in when we get to after `c'. */ | |
2878 | ||
505bde11 | 2879 | FIXUP_ALT_JUMP (); |
25fe55af RS |
2880 | |
2881 | /* Mark and leave space for a jump after this alternative, | |
2882 | to be filled in later either by next alternative or | |
2883 | when know we're at the end of a series of alternatives. */ | |
2884 | fixup_alt_jump = b; | |
2885 | GET_BUFFER_SPACE (3); | |
2886 | b += 3; | |
2887 | ||
2888 | laststart = 0; | |
2889 | begalt = b; | |
2890 | break; | |
2891 | ||
2892 | ||
2893 | case '{': | |
2894 | /* If \{ is a literal. */ | |
2895 | if (!(syntax & RE_INTERVALS) | |
2896 | /* If we're at `\{' and it's not the open-interval | |
2897 | operator. */ | |
4bb91c68 | 2898 | || (syntax & RE_NO_BK_BRACES)) |
25fe55af RS |
2899 | goto normal_backslash; |
2900 | ||
2901 | handle_interval: | |
2902 | { | |
2903 | /* If got here, then the syntax allows intervals. */ | |
2904 | ||
2905 | /* At least (most) this many matches must be made. */ | |
99633e97 | 2906 | int lower_bound = 0, upper_bound = -1; |
25fe55af | 2907 | |
ed0767d8 | 2908 | beg_interval = p; |
25fe55af RS |
2909 | |
2910 | if (p == pend) | |
4bb91c68 | 2911 | FREE_STACK_RETURN (REG_EBRACE); |
25fe55af RS |
2912 | |
2913 | GET_UNSIGNED_NUMBER (lower_bound); | |
2914 | ||
2915 | if (c == ',') | |
ed0767d8 | 2916 | GET_UNSIGNED_NUMBER (upper_bound); |
25fe55af RS |
2917 | else |
2918 | /* Interval such as `{1}' => match exactly once. */ | |
2919 | upper_bound = lower_bound; | |
2920 | ||
2921 | if (lower_bound < 0 || upper_bound > RE_DUP_MAX | |
ed0767d8 | 2922 | || (upper_bound >= 0 && lower_bound > upper_bound)) |
4bb91c68 | 2923 | FREE_STACK_RETURN (REG_BADBR); |
25fe55af RS |
2924 | |
2925 | if (!(syntax & RE_NO_BK_BRACES)) | |
2926 | { | |
4bb91c68 SM |
2927 | if (c != '\\') |
2928 | FREE_STACK_RETURN (REG_BADBR); | |
25fe55af RS |
2929 | |
2930 | PATFETCH (c); | |
2931 | } | |
2932 | ||
2933 | if (c != '}') | |
4bb91c68 | 2934 | FREE_STACK_RETURN (REG_BADBR); |
25fe55af RS |
2935 | |
2936 | /* We just parsed a valid interval. */ | |
2937 | ||
2938 | /* If it's invalid to have no preceding re. */ | |
2939 | if (!laststart) | |
2940 | { | |
2941 | if (syntax & RE_CONTEXT_INVALID_OPS) | |
2942 | FREE_STACK_RETURN (REG_BADRPT); | |
2943 | else if (syntax & RE_CONTEXT_INDEP_OPS) | |
2944 | laststart = b; | |
2945 | else | |
2946 | goto unfetch_interval; | |
2947 | } | |
2948 | ||
25fe55af | 2949 | if (upper_bound == 0) |
ed0767d8 SM |
2950 | /* If the upper bound is zero, just drop the sub pattern |
2951 | altogether. */ | |
2952 | b = laststart; | |
2953 | else if (lower_bound == 1 && upper_bound == 1) | |
2954 | /* Just match it once: nothing to do here. */ | |
2955 | ; | |
25fe55af RS |
2956 | |
2957 | /* Otherwise, we have a nontrivial interval. When | |
2958 | we're all done, the pattern will look like: | |
2959 | set_number_at <jump count> <upper bound> | |
2960 | set_number_at <succeed_n count> <lower bound> | |
2961 | succeed_n <after jump addr> <succeed_n count> | |
2962 | <body of loop> | |
2963 | jump_n <succeed_n addr> <jump count> | |
2964 | (The upper bound and `jump_n' are omitted if | |
2965 | `upper_bound' is 1, though.) */ | |
2966 | else | |
2967 | { /* If the upper bound is > 1, we need to insert | |
2968 | more at the end of the loop. */ | |
ed0767d8 SM |
2969 | unsigned int nbytes = (upper_bound < 0 ? 3 |
2970 | : upper_bound > 1 ? 5 : 0); | |
2971 | unsigned int startoffset = 0; | |
2972 | ||
2973 | GET_BUFFER_SPACE (20); /* We might use less. */ | |
2974 | ||
2975 | if (lower_bound == 0) | |
2976 | { | |
2977 | /* A succeed_n that starts with 0 is really a | |
2978 | a simple on_failure_jump_loop. */ | |
2979 | INSERT_JUMP (on_failure_jump_loop, laststart, | |
2980 | b + 3 + nbytes); | |
2981 | b += 3; | |
2982 | } | |
2983 | else | |
2984 | { | |
2985 | /* Initialize lower bound of the `succeed_n', even | |
2986 | though it will be set during matching by its | |
2987 | attendant `set_number_at' (inserted next), | |
2988 | because `re_compile_fastmap' needs to know. | |
2989 | Jump to the `jump_n' we might insert below. */ | |
2990 | INSERT_JUMP2 (succeed_n, laststart, | |
2991 | b + 5 + nbytes, | |
2992 | lower_bound); | |
2993 | b += 5; | |
2994 | ||
2995 | /* Code to initialize the lower bound. Insert | |
2996 | before the `succeed_n'. The `5' is the last two | |
2997 | bytes of this `set_number_at', plus 3 bytes of | |
2998 | the following `succeed_n'. */ | |
2999 | insert_op2 (set_number_at, laststart, 5, lower_bound, b); | |
3000 | b += 5; | |
3001 | startoffset += 5; | |
3002 | } | |
3003 | ||
3004 | if (upper_bound < 0) | |
3005 | { | |
3006 | /* A negative upper bound stands for infinity, | |
3007 | in which case it degenerates to a plain jump. */ | |
3008 | STORE_JUMP (jump, b, laststart + startoffset); | |
3009 | b += 3; | |
3010 | } | |
3011 | else if (upper_bound > 1) | |
25fe55af RS |
3012 | { /* More than one repetition is allowed, so |
3013 | append a backward jump to the `succeed_n' | |
3014 | that starts this interval. | |
3015 | ||
3016 | When we've reached this during matching, | |
3017 | we'll have matched the interval once, so | |
3018 | jump back only `upper_bound - 1' times. */ | |
ed0767d8 | 3019 | STORE_JUMP2 (jump_n, b, laststart + startoffset, |
25fe55af RS |
3020 | upper_bound - 1); |
3021 | b += 5; | |
3022 | ||
3023 | /* The location we want to set is the second | |
3024 | parameter of the `jump_n'; that is `b-2' as | |
3025 | an absolute address. `laststart' will be | |
3026 | the `set_number_at' we're about to insert; | |
3027 | `laststart+3' the number to set, the source | |
3028 | for the relative address. But we are | |
3029 | inserting into the middle of the pattern -- | |
3030 | so everything is getting moved up by 5. | |
3031 | Conclusion: (b - 2) - (laststart + 3) + 5, | |
3032 | i.e., b - laststart. | |
3033 | ||
3034 | We insert this at the beginning of the loop | |
3035 | so that if we fail during matching, we'll | |
3036 | reinitialize the bounds. */ | |
3037 | insert_op2 (set_number_at, laststart, b - laststart, | |
3038 | upper_bound - 1, b); | |
3039 | b += 5; | |
3040 | } | |
3041 | } | |
3042 | pending_exact = 0; | |
3043 | beg_interval = NULL; | |
3044 | } | |
3045 | break; | |
3046 | ||
3047 | unfetch_interval: | |
3048 | /* If an invalid interval, match the characters as literals. */ | |
3049 | assert (beg_interval); | |
3050 | p = beg_interval; | |
3051 | beg_interval = NULL; | |
3052 | ||
3053 | /* normal_char and normal_backslash need `c'. */ | |
ed0767d8 | 3054 | c = '{'; |
25fe55af RS |
3055 | |
3056 | if (!(syntax & RE_NO_BK_BRACES)) | |
3057 | { | |
ed0767d8 SM |
3058 | assert (p > pattern && p[-1] == '\\'); |
3059 | goto normal_backslash; | |
25fe55af | 3060 | } |
ed0767d8 SM |
3061 | else |
3062 | goto normal_char; | |
e318085a | 3063 | |
b18215fc | 3064 | #ifdef emacs |
25fe55af RS |
3065 | /* There is no way to specify the before_dot and after_dot |
3066 | operators. rms says this is ok. --karl */ | |
3067 | case '=': | |
3068 | BUF_PUSH (at_dot); | |
3069 | break; | |
3070 | ||
3071 | case 's': | |
3072 | laststart = b; | |
3073 | PATFETCH (c); | |
3074 | BUF_PUSH_2 (syntaxspec, syntax_spec_code[c]); | |
3075 | break; | |
3076 | ||
3077 | case 'S': | |
3078 | laststart = b; | |
3079 | PATFETCH (c); | |
3080 | BUF_PUSH_2 (notsyntaxspec, syntax_spec_code[c]); | |
3081 | break; | |
b18215fc RS |
3082 | |
3083 | case 'c': | |
3084 | laststart = b; | |
3085 | PATFETCH_RAW (c); | |
3086 | BUF_PUSH_2 (categoryspec, c); | |
3087 | break; | |
e318085a | 3088 | |
b18215fc RS |
3089 | case 'C': |
3090 | laststart = b; | |
3091 | PATFETCH_RAW (c); | |
3092 | BUF_PUSH_2 (notcategoryspec, c); | |
3093 | break; | |
3094 | #endif /* emacs */ | |
e318085a | 3095 | |
e318085a | 3096 | |
25fe55af | 3097 | case 'w': |
4bb91c68 SM |
3098 | if (syntax & RE_NO_GNU_OPS) |
3099 | goto normal_char; | |
25fe55af | 3100 | laststart = b; |
1fb352e0 | 3101 | BUF_PUSH_2 (syntaxspec, Sword); |
25fe55af | 3102 | break; |
e318085a | 3103 | |
e318085a | 3104 | |
25fe55af | 3105 | case 'W': |
4bb91c68 SM |
3106 | if (syntax & RE_NO_GNU_OPS) |
3107 | goto normal_char; | |
25fe55af | 3108 | laststart = b; |
1fb352e0 | 3109 | BUF_PUSH_2 (notsyntaxspec, Sword); |
25fe55af | 3110 | break; |
e318085a RS |
3111 | |
3112 | ||
25fe55af | 3113 | case '<': |
4bb91c68 SM |
3114 | if (syntax & RE_NO_GNU_OPS) |
3115 | goto normal_char; | |
25fe55af RS |
3116 | BUF_PUSH (wordbeg); |
3117 | break; | |
e318085a | 3118 | |
25fe55af | 3119 | case '>': |
4bb91c68 SM |
3120 | if (syntax & RE_NO_GNU_OPS) |
3121 | goto normal_char; | |
25fe55af RS |
3122 | BUF_PUSH (wordend); |
3123 | break; | |
e318085a | 3124 | |
25fe55af | 3125 | case 'b': |
4bb91c68 SM |
3126 | if (syntax & RE_NO_GNU_OPS) |
3127 | goto normal_char; | |
25fe55af RS |
3128 | BUF_PUSH (wordbound); |
3129 | break; | |
e318085a | 3130 | |
25fe55af | 3131 | case 'B': |
4bb91c68 SM |
3132 | if (syntax & RE_NO_GNU_OPS) |
3133 | goto normal_char; | |
25fe55af RS |
3134 | BUF_PUSH (notwordbound); |
3135 | break; | |
fa9a63c5 | 3136 | |
25fe55af | 3137 | case '`': |
4bb91c68 SM |
3138 | if (syntax & RE_NO_GNU_OPS) |
3139 | goto normal_char; | |
25fe55af RS |
3140 | BUF_PUSH (begbuf); |
3141 | break; | |
e318085a | 3142 | |
25fe55af | 3143 | case '\'': |
4bb91c68 SM |
3144 | if (syntax & RE_NO_GNU_OPS) |
3145 | goto normal_char; | |
25fe55af RS |
3146 | BUF_PUSH (endbuf); |
3147 | break; | |
e318085a | 3148 | |
25fe55af RS |
3149 | case '1': case '2': case '3': case '4': case '5': |
3150 | case '6': case '7': case '8': case '9': | |
0cdd06f8 SM |
3151 | { |
3152 | regnum_t reg; | |
e318085a | 3153 | |
0cdd06f8 SM |
3154 | if (syntax & RE_NO_BK_REFS) |
3155 | goto normal_backslash; | |
e318085a | 3156 | |
0cdd06f8 | 3157 | reg = c - '0'; |
e318085a | 3158 | |
0cdd06f8 SM |
3159 | /* Can't back reference to a subexpression before its end. */ |
3160 | if (reg > regnum || group_in_compile_stack (compile_stack, reg)) | |
3161 | FREE_STACK_RETURN (REG_ESUBREG); | |
e318085a | 3162 | |
0cdd06f8 SM |
3163 | laststart = b; |
3164 | BUF_PUSH_2 (duplicate, reg); | |
3165 | } | |
25fe55af | 3166 | break; |
e318085a | 3167 | |
e318085a | 3168 | |
25fe55af RS |
3169 | case '+': |
3170 | case '?': | |
3171 | if (syntax & RE_BK_PLUS_QM) | |
3172 | goto handle_plus; | |
3173 | else | |
3174 | goto normal_backslash; | |
3175 | ||
3176 | default: | |
3177 | normal_backslash: | |
3178 | /* You might think it would be useful for \ to mean | |
3179 | not to translate; but if we don't translate it | |
4bb91c68 | 3180 | it will never match anything. */ |
25fe55af RS |
3181 | c = TRANSLATE (c); |
3182 | goto normal_char; | |
3183 | } | |
3184 | break; | |
fa9a63c5 RM |
3185 | |
3186 | ||
3187 | default: | |
25fe55af | 3188 | /* Expects the character in `c'. */ |
fa9a63c5 RM |
3189 | normal_char: |
3190 | /* If no exactn currently being built. */ | |
25fe55af | 3191 | if (!pending_exact |
fa9a63c5 | 3192 | |
25fe55af RS |
3193 | /* If last exactn not at current position. */ |
3194 | || pending_exact + *pending_exact + 1 != b | |
5e69f11e | 3195 | |
25fe55af | 3196 | /* We have only one byte following the exactn for the count. */ |
2d1675e4 | 3197 | || *pending_exact >= (1 << BYTEWIDTH) - MAX_MULTIBYTE_LENGTH |
fa9a63c5 | 3198 | |
25fe55af | 3199 | /* If followed by a repetition operator. */ |
9d99031f | 3200 | || (p != pend && (*p == '*' || *p == '^')) |
fa9a63c5 | 3201 | || ((syntax & RE_BK_PLUS_QM) |
9d99031f RS |
3202 | ? p + 1 < pend && *p == '\\' && (p[1] == '+' || p[1] == '?') |
3203 | : p != pend && (*p == '+' || *p == '?')) | |
fa9a63c5 | 3204 | || ((syntax & RE_INTERVALS) |
25fe55af | 3205 | && ((syntax & RE_NO_BK_BRACES) |
9d99031f RS |
3206 | ? p != pend && *p == '{' |
3207 | : p + 1 < pend && p[0] == '\\' && p[1] == '{'))) | |
fa9a63c5 RM |
3208 | { |
3209 | /* Start building a new exactn. */ | |
5e69f11e | 3210 | |
25fe55af | 3211 | laststart = b; |
fa9a63c5 RM |
3212 | |
3213 | BUF_PUSH_2 (exactn, 0); | |
3214 | pending_exact = b - 1; | |
25fe55af | 3215 | } |
5e69f11e | 3216 | |
2d1675e4 SM |
3217 | GET_BUFFER_SPACE (MAX_MULTIBYTE_LENGTH); |
3218 | { | |
e0277a47 KH |
3219 | int len; |
3220 | ||
3221 | if (multibyte) | |
3222 | len = CHAR_STRING (c, b); | |
3223 | else | |
3224 | *b = c, len = 1; | |
2d1675e4 SM |
3225 | b += len; |
3226 | (*pending_exact) += len; | |
3227 | } | |
3228 | ||
fa9a63c5 | 3229 | break; |
25fe55af | 3230 | } /* switch (c) */ |
fa9a63c5 RM |
3231 | } /* while p != pend */ |
3232 | ||
5e69f11e | 3233 | |
fa9a63c5 | 3234 | /* Through the pattern now. */ |
5e69f11e | 3235 | |
505bde11 | 3236 | FIXUP_ALT_JUMP (); |
fa9a63c5 | 3237 | |
5e69f11e | 3238 | if (!COMPILE_STACK_EMPTY) |
fa9a63c5 RM |
3239 | FREE_STACK_RETURN (REG_EPAREN); |
3240 | ||
3241 | /* If we don't want backtracking, force success | |
3242 | the first time we reach the end of the compiled pattern. */ | |
3243 | if (syntax & RE_NO_POSIX_BACKTRACKING) | |
3244 | BUF_PUSH (succeed); | |
3245 | ||
3246 | free (compile_stack.stack); | |
3247 | ||
3248 | /* We have succeeded; set the length of the buffer. */ | |
3249 | bufp->used = b - bufp->buffer; | |
3250 | ||
3251 | #ifdef DEBUG | |
99633e97 | 3252 | if (debug > 0) |
fa9a63c5 | 3253 | { |
505bde11 | 3254 | re_compile_fastmap (bufp); |
fa9a63c5 RM |
3255 | DEBUG_PRINT1 ("\nCompiled pattern: \n"); |
3256 | print_compiled_pattern (bufp); | |
3257 | } | |
99633e97 | 3258 | debug--; |
fa9a63c5 RM |
3259 | #endif /* DEBUG */ |
3260 | ||
3261 | #ifndef MATCH_MAY_ALLOCATE | |
3262 | /* Initialize the failure stack to the largest possible stack. This | |
3263 | isn't necessary unless we're trying to avoid calling alloca in | |
3264 | the search and match routines. */ | |
3265 | { | |
3266 | int num_regs = bufp->re_nsub + 1; | |
3267 | ||
320a2a73 | 3268 | if (fail_stack.size < re_max_failures * TYPICAL_FAILURE_SIZE) |
fa9a63c5 | 3269 | { |
a26f4ccd | 3270 | fail_stack.size = re_max_failures * TYPICAL_FAILURE_SIZE; |
fa9a63c5 | 3271 | |
fa9a63c5 RM |
3272 | if (! fail_stack.stack) |
3273 | fail_stack.stack | |
5e69f11e | 3274 | = (fail_stack_elt_t *) malloc (fail_stack.size |
fa9a63c5 RM |
3275 | * sizeof (fail_stack_elt_t)); |
3276 | else | |
3277 | fail_stack.stack | |
3278 | = (fail_stack_elt_t *) realloc (fail_stack.stack, | |
3279 | (fail_stack.size | |
3280 | * sizeof (fail_stack_elt_t))); | |
fa9a63c5 RM |
3281 | } |
3282 | ||
3283 | regex_grow_registers (num_regs); | |
3284 | } | |
3285 | #endif /* not MATCH_MAY_ALLOCATE */ | |
3286 | ||
3287 | return REG_NOERROR; | |
3288 | } /* regex_compile */ | |
3289 | \f | |
3290 | /* Subroutines for `regex_compile'. */ | |
3291 | ||
25fe55af | 3292 | /* Store OP at LOC followed by two-byte integer parameter ARG. */ |
fa9a63c5 RM |
3293 | |
3294 | static void | |
3295 | store_op1 (op, loc, arg) | |
3296 | re_opcode_t op; | |
3297 | unsigned char *loc; | |
3298 | int arg; | |
3299 | { | |
3300 | *loc = (unsigned char) op; | |
3301 | STORE_NUMBER (loc + 1, arg); | |
3302 | } | |
3303 | ||
3304 | ||
3305 | /* Like `store_op1', but for two two-byte parameters ARG1 and ARG2. */ | |
3306 | ||
3307 | static void | |
3308 | store_op2 (op, loc, arg1, arg2) | |
3309 | re_opcode_t op; | |
3310 | unsigned char *loc; | |
3311 | int arg1, arg2; | |
3312 | { | |
3313 | *loc = (unsigned char) op; | |
3314 | STORE_NUMBER (loc + 1, arg1); | |
3315 | STORE_NUMBER (loc + 3, arg2); | |
3316 | } | |
3317 | ||
3318 | ||
3319 | /* Copy the bytes from LOC to END to open up three bytes of space at LOC | |
3320 | for OP followed by two-byte integer parameter ARG. */ | |
3321 | ||
3322 | static void | |
3323 | insert_op1 (op, loc, arg, end) | |
3324 | re_opcode_t op; | |
3325 | unsigned char *loc; | |
3326 | int arg; | |
5e69f11e | 3327 | unsigned char *end; |
fa9a63c5 RM |
3328 | { |
3329 | register unsigned char *pfrom = end; | |
3330 | register unsigned char *pto = end + 3; | |
3331 | ||
3332 | while (pfrom != loc) | |
3333 | *--pto = *--pfrom; | |
5e69f11e | 3334 | |
fa9a63c5 RM |
3335 | store_op1 (op, loc, arg); |
3336 | } | |
3337 | ||
3338 | ||
3339 | /* Like `insert_op1', but for two two-byte parameters ARG1 and ARG2. */ | |
3340 | ||
3341 | static void | |
3342 | insert_op2 (op, loc, arg1, arg2, end) | |
3343 | re_opcode_t op; | |
3344 | unsigned char *loc; | |
3345 | int arg1, arg2; | |
5e69f11e | 3346 | unsigned char *end; |
fa9a63c5 RM |
3347 | { |
3348 | register unsigned char *pfrom = end; | |
3349 | register unsigned char *pto = end + 5; | |
3350 | ||
3351 | while (pfrom != loc) | |
3352 | *--pto = *--pfrom; | |
5e69f11e | 3353 | |
fa9a63c5 RM |
3354 | store_op2 (op, loc, arg1, arg2); |
3355 | } | |
3356 | ||
3357 | ||
3358 | /* P points to just after a ^ in PATTERN. Return true if that ^ comes | |
3359 | after an alternative or a begin-subexpression. We assume there is at | |
3360 | least one character before the ^. */ | |
3361 | ||
3362 | static boolean | |
3363 | at_begline_loc_p (pattern, p, syntax) | |
01618498 | 3364 | re_char *pattern, *p; |
fa9a63c5 RM |
3365 | reg_syntax_t syntax; |
3366 | { | |
01618498 | 3367 | re_char *prev = p - 2; |
fa9a63c5 | 3368 | boolean prev_prev_backslash = prev > pattern && prev[-1] == '\\'; |
5e69f11e | 3369 | |
fa9a63c5 RM |
3370 | return |
3371 | /* After a subexpression? */ | |
3372 | (*prev == '(' && (syntax & RE_NO_BK_PARENS || prev_prev_backslash)) | |
25fe55af | 3373 | /* After an alternative? */ |
d2af47df SM |
3374 | || (*prev == '|' && (syntax & RE_NO_BK_VBAR || prev_prev_backslash)) |
3375 | /* After a shy subexpression? */ | |
3376 | || ((syntax & RE_SHY_GROUPS) && prev - 2 >= pattern | |
3377 | && prev[-1] == '?' && prev[-2] == '(' | |
3378 | && (syntax & RE_NO_BK_PARENS | |
3379 | || (prev - 3 >= pattern && prev[-3] == '\\'))); | |
fa9a63c5 RM |
3380 | } |
3381 | ||
3382 | ||
3383 | /* The dual of at_begline_loc_p. This one is for $. We assume there is | |
3384 | at least one character after the $, i.e., `P < PEND'. */ | |
3385 | ||
3386 | static boolean | |
3387 | at_endline_loc_p (p, pend, syntax) | |
01618498 | 3388 | re_char *p, *pend; |
99633e97 | 3389 | reg_syntax_t syntax; |
fa9a63c5 | 3390 | { |
01618498 | 3391 | re_char *next = p; |
fa9a63c5 | 3392 | boolean next_backslash = *next == '\\'; |
01618498 | 3393 | re_char *next_next = p + 1 < pend ? p + 1 : 0; |
5e69f11e | 3394 | |
fa9a63c5 RM |
3395 | return |
3396 | /* Before a subexpression? */ | |
3397 | (syntax & RE_NO_BK_PARENS ? *next == ')' | |
25fe55af | 3398 | : next_backslash && next_next && *next_next == ')') |
fa9a63c5 RM |
3399 | /* Before an alternative? */ |
3400 | || (syntax & RE_NO_BK_VBAR ? *next == '|' | |
25fe55af | 3401 | : next_backslash && next_next && *next_next == '|'); |
fa9a63c5 RM |
3402 | } |
3403 | ||
3404 | ||
5e69f11e | 3405 | /* Returns true if REGNUM is in one of COMPILE_STACK's elements and |
fa9a63c5 RM |
3406 | false if it's not. */ |
3407 | ||
3408 | static boolean | |
3409 | group_in_compile_stack (compile_stack, regnum) | |
3410 | compile_stack_type compile_stack; | |
3411 | regnum_t regnum; | |
3412 | { | |
3413 | int this_element; | |
3414 | ||
5e69f11e RM |
3415 | for (this_element = compile_stack.avail - 1; |
3416 | this_element >= 0; | |
fa9a63c5 RM |
3417 | this_element--) |
3418 | if (compile_stack.stack[this_element].regnum == regnum) | |
3419 | return true; | |
3420 | ||
3421 | return false; | |
3422 | } | |
fa9a63c5 | 3423 | \f |
f6a3f532 SM |
3424 | /* analyse_first. |
3425 | If fastmap is non-NULL, go through the pattern and fill fastmap | |
3426 | with all the possible leading chars. If fastmap is NULL, don't | |
3427 | bother filling it up (obviously) and only return whether the | |
3428 | pattern could potentially match the empty string. | |
3429 | ||
3430 | Return 1 if p..pend might match the empty string. | |
3431 | Return 0 if p..pend matches at least one char. | |
01618498 | 3432 | Return -1 if fastmap was not updated accurately. */ |
f6a3f532 SM |
3433 | |
3434 | static int | |
3435 | analyse_first (p, pend, fastmap, multibyte) | |
01618498 | 3436 | re_char *p, *pend; |
f6a3f532 SM |
3437 | char *fastmap; |
3438 | const int multibyte; | |
fa9a63c5 | 3439 | { |
505bde11 | 3440 | int j, k; |
1fb352e0 | 3441 | boolean not; |
fa9a63c5 | 3442 | |
b18215fc | 3443 | /* If all elements for base leading-codes in fastmap is set, this |
25fe55af | 3444 | flag is set true. */ |
b18215fc RS |
3445 | boolean match_any_multibyte_characters = false; |
3446 | ||
f6a3f532 | 3447 | assert (p); |
5e69f11e | 3448 | |
505bde11 SM |
3449 | /* The loop below works as follows: |
3450 | - It has a working-list kept in the PATTERN_STACK and which basically | |
3451 | starts by only containing a pointer to the first operation. | |
3452 | - If the opcode we're looking at is a match against some set of | |
3453 | chars, then we add those chars to the fastmap and go on to the | |
3454 | next work element from the worklist (done via `break'). | |
3455 | - If the opcode is a control operator on the other hand, we either | |
3456 | ignore it (if it's meaningless at this point, such as `start_memory') | |
3457 | or execute it (if it's a jump). If the jump has several destinations | |
3458 | (i.e. `on_failure_jump'), then we push the other destination onto the | |
3459 | worklist. | |
3460 | We guarantee termination by ignoring backward jumps (more or less), | |
3461 | so that `p' is monotonically increasing. More to the point, we | |
3462 | never set `p' (or push) anything `<= p1'. */ | |
3463 | ||
01618498 | 3464 | while (p < pend) |
fa9a63c5 | 3465 | { |
505bde11 SM |
3466 | /* `p1' is used as a marker of how far back a `on_failure_jump' |
3467 | can go without being ignored. It is normally equal to `p' | |
3468 | (which prevents any backward `on_failure_jump') except right | |
3469 | after a plain `jump', to allow patterns such as: | |
3470 | 0: jump 10 | |
3471 | 3..9: <body> | |
3472 | 10: on_failure_jump 3 | |
3473 | as used for the *? operator. */ | |
01618498 | 3474 | re_char *p1 = p; |
5e69f11e | 3475 | |
fa9a63c5 RM |
3476 | switch (SWITCH_ENUM_CAST ((re_opcode_t) *p++)) |
3477 | { | |
f6a3f532 | 3478 | case succeed: |
01618498 | 3479 | return 1; |
f6a3f532 | 3480 | continue; |
fa9a63c5 | 3481 | |
fa9a63c5 | 3482 | case duplicate: |
505bde11 SM |
3483 | /* If the first character has to match a backreference, that means |
3484 | that the group was empty (since it already matched). Since this | |
3485 | is the only case that interests us here, we can assume that the | |
3486 | backreference must match the empty string. */ | |
3487 | p++; | |
3488 | continue; | |
fa9a63c5 RM |
3489 | |
3490 | ||
3491 | /* Following are the cases which match a character. These end | |
25fe55af | 3492 | with `break'. */ |
fa9a63c5 RM |
3493 | |
3494 | case exactn: | |
e0277a47 KH |
3495 | if (fastmap) |
3496 | { | |
3497 | int c = RE_STRING_CHAR (p + 1, pend - p); | |
3498 | ||
3499 | if (SINGLE_BYTE_CHAR_P (c)) | |
3500 | fastmap[c] = 1; | |
3501 | else | |
3502 | fastmap[p[1]] = 1; | |
3503 | } | |
fa9a63c5 RM |
3504 | break; |
3505 | ||
3506 | ||
1fb352e0 SM |
3507 | case anychar: |
3508 | /* We could put all the chars except for \n (and maybe \0) | |
3509 | but we don't bother since it is generally not worth it. */ | |
f6a3f532 | 3510 | if (!fastmap) break; |
01618498 | 3511 | return -1; |
fa9a63c5 RM |
3512 | |
3513 | ||
b18215fc | 3514 | case charset_not: |
ba5c004d RS |
3515 | /* Chars beyond end of bitmap are possible matches. |
3516 | All the single-byte codes can occur in multibyte buffers. | |
3517 | So any that are not listed in the charset | |
3518 | are possible matches, even in multibyte buffers. */ | |
1fb352e0 | 3519 | if (!fastmap) break; |
b18215fc | 3520 | for (j = CHARSET_BITMAP_SIZE (&p[-1]) * BYTEWIDTH; |
1fb352e0 | 3521 | j < (1 << BYTEWIDTH); j++) |
b18215fc | 3522 | fastmap[j] = 1; |
1fb352e0 SM |
3523 | /* Fallthrough */ |
3524 | case charset: | |
3525 | if (!fastmap) break; | |
3526 | not = (re_opcode_t) *(p - 1) == charset_not; | |
b18215fc RS |
3527 | for (j = CHARSET_BITMAP_SIZE (&p[-1]) * BYTEWIDTH - 1, p++; |
3528 | j >= 0; j--) | |
1fb352e0 | 3529 | if (!!(p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH))) ^ not) |
b18215fc RS |
3530 | fastmap[j] = 1; |
3531 | ||
1fb352e0 SM |
3532 | if ((not && multibyte) |
3533 | /* Any character set can possibly contain a character | |
3534 | which doesn't match the specified set of characters. */ | |
3535 | || (CHARSET_RANGE_TABLE_EXISTS_P (&p[-2]) | |
3536 | && CHARSET_RANGE_TABLE_BITS (&p[-2]) != 0)) | |
3537 | /* If we can match a character class, we can match | |
3538 | any character set. */ | |
b18215fc RS |
3539 | { |
3540 | set_fastmap_for_multibyte_characters: | |
3541 | if (match_any_multibyte_characters == false) | |
3542 | { | |
3543 | for (j = 0x80; j < 0xA0; j++) /* XXX */ | |
3544 | if (BASE_LEADING_CODE_P (j)) | |
3545 | fastmap[j] = 1; | |
3546 | match_any_multibyte_characters = true; | |
3547 | } | |
3548 | } | |
b18215fc | 3549 | |
1fb352e0 SM |
3550 | else if (!not && CHARSET_RANGE_TABLE_EXISTS_P (&p[-2]) |
3551 | && match_any_multibyte_characters == false) | |
3552 | { | |
3553 | /* Set fastmap[I] 1 where I is a base leading code of each | |
3554 | multibyte character in the range table. */ | |
3555 | int c, count; | |
b18215fc | 3556 | |
1fb352e0 | 3557 | /* Make P points the range table. `+ 2' is to skip flag |
0b32bf0e | 3558 | bits for a character class. */ |
1fb352e0 | 3559 | p += CHARSET_BITMAP_SIZE (&p[-2]) + 2; |
b18215fc | 3560 | |
1fb352e0 SM |
3561 | /* Extract the number of ranges in range table into COUNT. */ |
3562 | EXTRACT_NUMBER_AND_INCR (count, p); | |
3563 | for (; count > 0; count--, p += 2 * 3) /* XXX */ | |
3564 | { | |
3565 | /* Extract the start of each range. */ | |
3566 | EXTRACT_CHARACTER (c, p); | |
3567 | j = CHAR_CHARSET (c); | |
3568 | fastmap[CHARSET_LEADING_CODE_BASE (j)] = 1; | |
3569 | } | |
3570 | } | |
b18215fc RS |
3571 | break; |
3572 | ||
1fb352e0 SM |
3573 | case syntaxspec: |
3574 | case notsyntaxspec: | |
3575 | if (!fastmap) break; | |
3576 | #ifndef emacs | |
3577 | not = (re_opcode_t)p[-1] == notsyntaxspec; | |
3578 | k = *p++; | |
3579 | for (j = 0; j < (1 << BYTEWIDTH); j++) | |
990b2375 | 3580 | if ((SYNTAX (j) == (enum syntaxcode) k) ^ not) |
b18215fc | 3581 | fastmap[j] = 1; |
b18215fc | 3582 | break; |
1fb352e0 | 3583 | #else /* emacs */ |
b18215fc RS |
3584 | /* This match depends on text properties. These end with |
3585 | aborting optimizations. */ | |
01618498 | 3586 | return -1; |
b18215fc RS |
3587 | |
3588 | case categoryspec: | |
b18215fc | 3589 | case notcategoryspec: |
1fb352e0 SM |
3590 | if (!fastmap) break; |
3591 | not = (re_opcode_t)p[-1] == notcategoryspec; | |
b18215fc | 3592 | k = *p++; |
1fb352e0 SM |
3593 | for (j = 0; j < (1 << BYTEWIDTH); j++) |
3594 | if ((CHAR_HAS_CATEGORY (j, k)) ^ not) | |
b18215fc RS |
3595 | fastmap[j] = 1; |
3596 | ||
1fb352e0 | 3597 | if (multibyte) |
b18215fc | 3598 | /* Any character set can possibly contain a character |
1fb352e0 | 3599 | whose category is K (or not). */ |
b18215fc RS |
3600 | goto set_fastmap_for_multibyte_characters; |
3601 | break; | |
3602 | ||
fa9a63c5 | 3603 | /* All cases after this match the empty string. These end with |
25fe55af | 3604 | `continue'. */ |
fa9a63c5 | 3605 | |
fa9a63c5 RM |
3606 | case before_dot: |
3607 | case at_dot: | |
3608 | case after_dot: | |
1fb352e0 | 3609 | #endif /* !emacs */ |
25fe55af RS |
3610 | case no_op: |
3611 | case begline: | |
3612 | case endline: | |
fa9a63c5 RM |
3613 | case begbuf: |
3614 | case endbuf: | |
3615 | case wordbound: | |
3616 | case notwordbound: | |
3617 | case wordbeg: | |
3618 | case wordend: | |
25fe55af | 3619 | continue; |
fa9a63c5 RM |
3620 | |
3621 | ||
fa9a63c5 | 3622 | case jump: |
25fe55af | 3623 | EXTRACT_NUMBER_AND_INCR (j, p); |
505bde11 SM |
3624 | if (j < 0) |
3625 | /* Backward jumps can only go back to code that we've already | |
3626 | visited. `re_compile' should make sure this is true. */ | |
3627 | break; | |
25fe55af | 3628 | p += j; |
505bde11 SM |
3629 | switch (SWITCH_ENUM_CAST ((re_opcode_t) *p)) |
3630 | { | |
3631 | case on_failure_jump: | |
3632 | case on_failure_keep_string_jump: | |
505bde11 | 3633 | case on_failure_jump_loop: |
0683b6fa | 3634 | case on_failure_jump_nastyloop: |
505bde11 SM |
3635 | case on_failure_jump_smart: |
3636 | p++; | |
3637 | break; | |
3638 | default: | |
3639 | continue; | |
3640 | }; | |
3641 | /* Keep `p1' to allow the `on_failure_jump' we are jumping to | |
3642 | to jump back to "just after here". */ | |
3643 | /* Fallthrough */ | |
fa9a63c5 | 3644 | |
25fe55af RS |
3645 | case on_failure_jump: |
3646 | case on_failure_keep_string_jump: | |
0683b6fa | 3647 | case on_failure_jump_nastyloop: |
505bde11 SM |
3648 | case on_failure_jump_loop: |
3649 | case on_failure_jump_smart: | |
25fe55af | 3650 | EXTRACT_NUMBER_AND_INCR (j, p); |
505bde11 | 3651 | if (p + j <= p1) |
ed0767d8 | 3652 | ; /* Backward jump to be ignored. */ |
01618498 SM |
3653 | else |
3654 | { /* We have to look down both arms. | |
3655 | We first go down the "straight" path so as to minimize | |
3656 | stack usage when going through alternatives. */ | |
3657 | int r = analyse_first (p, pend, fastmap, multibyte); | |
3658 | if (r) return r; | |
3659 | p += j; | |
3660 | } | |
25fe55af | 3661 | continue; |
fa9a63c5 RM |
3662 | |
3663 | ||
ed0767d8 SM |
3664 | case jump_n: |
3665 | /* This code simply does not properly handle forward jump_n. */ | |
3666 | DEBUG_STATEMENT (EXTRACT_NUMBER (j, p); assert (j < 0)); | |
3667 | p += 4; | |
3668 | /* jump_n can either jump or fall through. The (backward) jump | |
3669 | case has already been handled, so we only need to look at the | |
3670 | fallthrough case. */ | |
3671 | continue; | |
3672 | ||
fa9a63c5 | 3673 | case succeed_n: |
ed0767d8 SM |
3674 | /* If N == 0, it should be an on_failure_jump_loop instead. */ |
3675 | DEBUG_STATEMENT (EXTRACT_NUMBER (j, p + 2); assert (j > 0)); | |
3676 | p += 4; | |
3677 | /* We only care about one iteration of the loop, so we don't | |
3678 | need to consider the case where this behaves like an | |
3679 | on_failure_jump. */ | |
25fe55af | 3680 | continue; |
fa9a63c5 RM |
3681 | |
3682 | ||
3683 | case set_number_at: | |
25fe55af RS |
3684 | p += 4; |
3685 | continue; | |
fa9a63c5 RM |
3686 | |
3687 | ||
3688 | case start_memory: | |
25fe55af | 3689 | case stop_memory: |
505bde11 | 3690 | p += 1; |
fa9a63c5 RM |
3691 | continue; |
3692 | ||
3693 | ||
3694 | default: | |
25fe55af RS |
3695 | abort (); /* We have listed all the cases. */ |
3696 | } /* switch *p++ */ | |
fa9a63c5 RM |
3697 | |
3698 | /* Getting here means we have found the possible starting | |
25fe55af | 3699 | characters for one path of the pattern -- and that the empty |
01618498 SM |
3700 | string does not match. We need not follow this path further. */ |
3701 | return 0; | |
fa9a63c5 RM |
3702 | } /* while p */ |
3703 | ||
01618498 SM |
3704 | /* We reached the end without matching anything. */ |
3705 | return 1; | |
3706 | ||
f6a3f532 SM |
3707 | } /* analyse_first */ |
3708 | \f | |
3709 | /* re_compile_fastmap computes a ``fastmap'' for the compiled pattern in | |
3710 | BUFP. A fastmap records which of the (1 << BYTEWIDTH) possible | |
3711 | characters can start a string that matches the pattern. This fastmap | |
3712 | is used by re_search to skip quickly over impossible starting points. | |
3713 | ||
3714 | Character codes above (1 << BYTEWIDTH) are not represented in the | |
3715 | fastmap, but the leading codes are represented. Thus, the fastmap | |
3716 | indicates which character sets could start a match. | |
3717 | ||
3718 | The caller must supply the address of a (1 << BYTEWIDTH)-byte data | |
3719 | area as BUFP->fastmap. | |
3720 | ||
3721 | We set the `fastmap', `fastmap_accurate', and `can_be_null' fields in | |
3722 | the pattern buffer. | |
3723 | ||
3724 | Returns 0 if we succeed, -2 if an internal error. */ | |
3725 | ||
3726 | int | |
3727 | re_compile_fastmap (bufp) | |
3728 | struct re_pattern_buffer *bufp; | |
3729 | { | |
3730 | char *fastmap = bufp->fastmap; | |
3731 | int analysis; | |
3732 | ||
3733 | assert (fastmap && bufp->buffer); | |
3734 | ||
3735 | bzero (fastmap, 1 << BYTEWIDTH); /* Assume nothing's valid. */ | |
3736 | bufp->fastmap_accurate = 1; /* It will be when we're done. */ | |
3737 | ||
3738 | analysis = analyse_first (bufp->buffer, bufp->buffer + bufp->used, | |
2d1675e4 | 3739 | fastmap, RE_MULTIBYTE_P (bufp)); |
c0f9ea08 | 3740 | bufp->can_be_null = (analysis != 0); |
fa9a63c5 RM |
3741 | return 0; |
3742 | } /* re_compile_fastmap */ | |
3743 | \f | |
3744 | /* Set REGS to hold NUM_REGS registers, storing them in STARTS and | |
3745 | ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use | |
3746 | this memory for recording register information. STARTS and ENDS | |
3747 | must be allocated using the malloc library routine, and must each | |
3748 | be at least NUM_REGS * sizeof (regoff_t) bytes long. | |
3749 | ||
3750 | If NUM_REGS == 0, then subsequent matches should allocate their own | |
3751 | register data. | |
3752 | ||
3753 | Unless this function is called, the first search or match using | |
3754 | PATTERN_BUFFER will allocate its own register data, without | |
3755 | freeing the old data. */ | |
3756 | ||
3757 | void | |
3758 | re_set_registers (bufp, regs, num_regs, starts, ends) | |
3759 | struct re_pattern_buffer *bufp; | |
3760 | struct re_registers *regs; | |
3761 | unsigned num_regs; | |
3762 | regoff_t *starts, *ends; | |
3763 | { | |
3764 | if (num_regs) | |
3765 | { | |
3766 | bufp->regs_allocated = REGS_REALLOCATE; | |
3767 | regs->num_regs = num_regs; | |
3768 | regs->start = starts; | |
3769 | regs->end = ends; | |
3770 | } | |
3771 | else | |
3772 | { | |
3773 | bufp->regs_allocated = REGS_UNALLOCATED; | |
3774 | regs->num_regs = 0; | |
3775 | regs->start = regs->end = (regoff_t *) 0; | |
3776 | } | |
3777 | } | |
c0f9ea08 | 3778 | WEAK_ALIAS (__re_set_registers, re_set_registers) |
fa9a63c5 | 3779 | \f |
25fe55af | 3780 | /* Searching routines. */ |
fa9a63c5 RM |
3781 | |
3782 | /* Like re_search_2, below, but only one string is specified, and | |
3783 | doesn't let you say where to stop matching. */ | |
3784 | ||
3785 | int | |
3786 | re_search (bufp, string, size, startpos, range, regs) | |
3787 | struct re_pattern_buffer *bufp; | |
3788 | const char *string; | |
3789 | int size, startpos, range; | |
3790 | struct re_registers *regs; | |
3791 | { | |
5e69f11e | 3792 | return re_search_2 (bufp, NULL, 0, string, size, startpos, range, |
fa9a63c5 RM |
3793 | regs, size); |
3794 | } | |
c0f9ea08 | 3795 | WEAK_ALIAS (__re_search, re_search) |
fa9a63c5 | 3796 | |
b18215fc RS |
3797 | /* End address of virtual concatenation of string. */ |
3798 | #define STOP_ADDR_VSTRING(P) \ | |
3799 | (((P) >= size1 ? string2 + size2 : string1 + size1)) | |
3800 | ||
3801 | /* Address of POS in the concatenation of virtual string. */ | |
3802 | #define POS_ADDR_VSTRING(POS) \ | |
3803 | (((POS) >= size1 ? string2 - size1 : string1) + (POS)) | |
fa9a63c5 RM |
3804 | |
3805 | /* Using the compiled pattern in BUFP->buffer, first tries to match the | |
3806 | virtual concatenation of STRING1 and STRING2, starting first at index | |
3807 | STARTPOS, then at STARTPOS + 1, and so on. | |
5e69f11e | 3808 | |
fa9a63c5 | 3809 | STRING1 and STRING2 have length SIZE1 and SIZE2, respectively. |
5e69f11e | 3810 | |
fa9a63c5 RM |
3811 | RANGE is how far to scan while trying to match. RANGE = 0 means try |
3812 | only at STARTPOS; in general, the last start tried is STARTPOS + | |
3813 | RANGE. | |
5e69f11e | 3814 | |
fa9a63c5 RM |
3815 | In REGS, return the indices of the virtual concatenation of STRING1 |
3816 | and STRING2 that matched the entire BUFP->buffer and its contained | |
3817 | subexpressions. | |
5e69f11e | 3818 | |
fa9a63c5 RM |
3819 | Do not consider matching one past the index STOP in the virtual |
3820 | concatenation of STRING1 and STRING2. | |
3821 | ||
3822 | We return either the position in the strings at which the match was | |
3823 | found, -1 if no match, or -2 if error (such as failure | |
3824 | stack overflow). */ | |
3825 | ||
3826 | int | |
66f0296e | 3827 | re_search_2 (bufp, str1, size1, str2, size2, startpos, range, regs, stop) |
fa9a63c5 | 3828 | struct re_pattern_buffer *bufp; |
66f0296e | 3829 | const char *str1, *str2; |
fa9a63c5 RM |
3830 | int size1, size2; |
3831 | int startpos; | |
3832 | int range; | |
3833 | struct re_registers *regs; | |
3834 | int stop; | |
3835 | { | |
3836 | int val; | |
66f0296e SM |
3837 | re_char *string1 = (re_char*) str1; |
3838 | re_char *string2 = (re_char*) str2; | |
fa9a63c5 | 3839 | register char *fastmap = bufp->fastmap; |
6676cb1c | 3840 | register RE_TRANSLATE_TYPE translate = bufp->translate; |
fa9a63c5 RM |
3841 | int total_size = size1 + size2; |
3842 | int endpos = startpos + range; | |
c0f9ea08 | 3843 | boolean anchored_start; |
fa9a63c5 | 3844 | |
25fe55af | 3845 | /* Nonzero if we have to concern multibyte character. */ |
2d1675e4 | 3846 | const boolean multibyte = RE_MULTIBYTE_P (bufp); |
b18215fc | 3847 | |
fa9a63c5 RM |
3848 | /* Check for out-of-range STARTPOS. */ |
3849 | if (startpos < 0 || startpos > total_size) | |
3850 | return -1; | |
5e69f11e | 3851 | |
fa9a63c5 | 3852 | /* Fix up RANGE if it might eventually take us outside |
34597fa9 | 3853 | the virtual concatenation of STRING1 and STRING2. |
5e69f11e | 3854 | Make sure we won't move STARTPOS below 0 or above TOTAL_SIZE. */ |
34597fa9 RS |
3855 | if (endpos < 0) |
3856 | range = 0 - startpos; | |
fa9a63c5 RM |
3857 | else if (endpos > total_size) |
3858 | range = total_size - startpos; | |
3859 | ||
3860 | /* If the search isn't to be a backwards one, don't waste time in a | |
7b140fd7 | 3861 | search for a pattern anchored at beginning of buffer. */ |
fa9a63c5 RM |
3862 | if (bufp->used > 0 && (re_opcode_t) bufp->buffer[0] == begbuf && range > 0) |
3863 | { | |
3864 | if (startpos > 0) | |
3865 | return -1; | |
3866 | else | |
7b140fd7 | 3867 | range = 0; |
fa9a63c5 RM |
3868 | } |
3869 | ||
ae4788a8 RS |
3870 | #ifdef emacs |
3871 | /* In a forward search for something that starts with \=. | |
3872 | don't keep searching past point. */ | |
3873 | if (bufp->used > 0 && (re_opcode_t) bufp->buffer[0] == at_dot && range > 0) | |
3874 | { | |
7b140fd7 RS |
3875 | range = PT_BYTE - BEGV_BYTE - startpos; |
3876 | if (range < 0) | |
ae4788a8 RS |
3877 | return -1; |
3878 | } | |
3879 | #endif /* emacs */ | |
3880 | ||
fa9a63c5 RM |
3881 | /* Update the fastmap now if not correct already. */ |
3882 | if (fastmap && !bufp->fastmap_accurate) | |
01618498 | 3883 | re_compile_fastmap (bufp); |
5e69f11e | 3884 | |
c8499ba5 | 3885 | /* See whether the pattern is anchored. */ |
c0f9ea08 | 3886 | anchored_start = (bufp->buffer[0] == begline); |
c8499ba5 | 3887 | |
b18215fc | 3888 | #ifdef emacs |
cc9b4df2 KH |
3889 | gl_state.object = re_match_object; |
3890 | { | |
99633e97 | 3891 | int charpos = SYNTAX_TABLE_BYTE_TO_CHAR (POS_AS_IN_BUFFER (startpos)); |
cc9b4df2 KH |
3892 | |
3893 | SETUP_SYNTAX_TABLE_FOR_OBJECT (re_match_object, charpos, 1); | |
3894 | } | |
b18215fc RS |
3895 | #endif |
3896 | ||
fa9a63c5 RM |
3897 | /* Loop through the string, looking for a place to start matching. */ |
3898 | for (;;) | |
5e69f11e | 3899 | { |
c8499ba5 RS |
3900 | /* If the pattern is anchored, |
3901 | skip quickly past places we cannot match. | |
3902 | We don't bother to treat startpos == 0 specially | |
3903 | because that case doesn't repeat. */ | |
3904 | if (anchored_start && startpos > 0) | |
3905 | { | |
c0f9ea08 SM |
3906 | if (! ((startpos <= size1 ? string1[startpos - 1] |
3907 | : string2[startpos - size1 - 1]) | |
3908 | == '\n')) | |
c8499ba5 RS |
3909 | goto advance; |
3910 | } | |
3911 | ||
fa9a63c5 | 3912 | /* If a fastmap is supplied, skip quickly over characters that |
25fe55af RS |
3913 | cannot be the start of a match. If the pattern can match the |
3914 | null string, however, we don't need to skip characters; we want | |
3915 | the first null string. */ | |
fa9a63c5 RM |
3916 | if (fastmap && startpos < total_size && !bufp->can_be_null) |
3917 | { | |
66f0296e | 3918 | register re_char *d; |
01618498 | 3919 | register re_wchar_t buf_ch; |
e934739e RS |
3920 | |
3921 | d = POS_ADDR_VSTRING (startpos); | |
3922 | ||
25fe55af | 3923 | if (range > 0) /* Searching forwards. */ |
fa9a63c5 | 3924 | { |
fa9a63c5 RM |
3925 | register int lim = 0; |
3926 | int irange = range; | |
3927 | ||
25fe55af RS |
3928 | if (startpos < size1 && startpos + range >= size1) |
3929 | lim = range - (size1 - startpos); | |
fa9a63c5 | 3930 | |
25fe55af RS |
3931 | /* Written out as an if-else to avoid testing `translate' |
3932 | inside the loop. */ | |
28ae27ae AS |
3933 | if (RE_TRANSLATE_P (translate)) |
3934 | { | |
e934739e RS |
3935 | if (multibyte) |
3936 | while (range > lim) | |
3937 | { | |
3938 | int buf_charlen; | |
3939 | ||
3940 | buf_ch = STRING_CHAR_AND_LENGTH (d, range - lim, | |
3941 | buf_charlen); | |
3942 | ||
3943 | buf_ch = RE_TRANSLATE (translate, buf_ch); | |
3944 | if (buf_ch >= 0400 | |
3945 | || fastmap[buf_ch]) | |
3946 | break; | |
3947 | ||
3948 | range -= buf_charlen; | |
3949 | d += buf_charlen; | |
3950 | } | |
3951 | else | |
3952 | while (range > lim | |
66f0296e | 3953 | && !fastmap[RE_TRANSLATE (translate, *d)]) |
33c46939 RS |
3954 | { |
3955 | d++; | |
3956 | range--; | |
3957 | } | |
e934739e | 3958 | } |
fa9a63c5 | 3959 | else |
66f0296e | 3960 | while (range > lim && !fastmap[*d]) |
33c46939 RS |
3961 | { |
3962 | d++; | |
3963 | range--; | |
3964 | } | |
fa9a63c5 RM |
3965 | |
3966 | startpos += irange - range; | |
3967 | } | |
25fe55af | 3968 | else /* Searching backwards. */ |
fa9a63c5 | 3969 | { |
2d1675e4 SM |
3970 | int room = (startpos >= size1 |
3971 | ? size2 + size1 - startpos | |
3972 | : size1 - startpos); | |
3973 | buf_ch = RE_STRING_CHAR (d, room); | |
3974 | buf_ch = TRANSLATE (buf_ch); | |
fa9a63c5 | 3975 | |
e934739e RS |
3976 | if (! (buf_ch >= 0400 |
3977 | || fastmap[buf_ch])) | |
fa9a63c5 RM |
3978 | goto advance; |
3979 | } | |
3980 | } | |
3981 | ||
3982 | /* If can't match the null string, and that's all we have left, fail. */ | |
3983 | if (range >= 0 && startpos == total_size && fastmap | |
25fe55af | 3984 | && !bufp->can_be_null) |
fa9a63c5 RM |
3985 | return -1; |
3986 | ||
3987 | val = re_match_2_internal (bufp, string1, size1, string2, size2, | |
3988 | startpos, regs, stop); | |
3989 | #ifndef REGEX_MALLOC | |
0b32bf0e | 3990 | # ifdef C_ALLOCA |
fa9a63c5 | 3991 | alloca (0); |
0b32bf0e | 3992 | # endif |
fa9a63c5 RM |
3993 | #endif |
3994 | ||
3995 | if (val >= 0) | |
3996 | return startpos; | |
5e69f11e | 3997 | |
fa9a63c5 RM |
3998 | if (val == -2) |
3999 | return -2; | |
4000 | ||
4001 | advance: | |
5e69f11e | 4002 | if (!range) |
25fe55af | 4003 | break; |
5e69f11e | 4004 | else if (range > 0) |
25fe55af | 4005 | { |
b18215fc RS |
4006 | /* Update STARTPOS to the next character boundary. */ |
4007 | if (multibyte) | |
4008 | { | |
66f0296e SM |
4009 | re_char *p = POS_ADDR_VSTRING (startpos); |
4010 | re_char *pend = STOP_ADDR_VSTRING (startpos); | |
b18215fc RS |
4011 | int len = MULTIBYTE_FORM_LENGTH (p, pend - p); |
4012 | ||
4013 | range -= len; | |
4014 | if (range < 0) | |
4015 | break; | |
4016 | startpos += len; | |
4017 | } | |
4018 | else | |
4019 | { | |
b560c397 RS |
4020 | range--; |
4021 | startpos++; | |
4022 | } | |
e318085a | 4023 | } |
fa9a63c5 | 4024 | else |
25fe55af RS |
4025 | { |
4026 | range++; | |
4027 | startpos--; | |
b18215fc RS |
4028 | |
4029 | /* Update STARTPOS to the previous character boundary. */ | |
4030 | if (multibyte) | |
4031 | { | |
66f0296e | 4032 | re_char *p = POS_ADDR_VSTRING (startpos); |
b18215fc RS |
4033 | int len = 0; |
4034 | ||
4035 | /* Find the head of multibyte form. */ | |
5d967c7a | 4036 | while (!CHAR_HEAD_P (*p)) |
b18215fc RS |
4037 | p--, len++; |
4038 | ||
4039 | /* Adjust it. */ | |
4040 | #if 0 /* XXX */ | |
4041 | if (MULTIBYTE_FORM_LENGTH (p, len + 1) != (len + 1)) | |
4042 | ; | |
4043 | else | |
4044 | #endif | |
4045 | { | |
4046 | range += len; | |
4047 | if (range > 0) | |
4048 | break; | |
4049 | ||
4050 | startpos -= len; | |
4051 | } | |
4052 | } | |
25fe55af | 4053 | } |
fa9a63c5 RM |
4054 | } |
4055 | return -1; | |
4056 | } /* re_search_2 */ | |
c0f9ea08 | 4057 | WEAK_ALIAS (__re_search_2, re_search_2) |
fa9a63c5 RM |
4058 | \f |
4059 | /* Declarations and macros for re_match_2. */ | |
4060 | ||
2d1675e4 SM |
4061 | static int bcmp_translate _RE_ARGS((re_char *s1, re_char *s2, |
4062 | register int len, | |
4063 | RE_TRANSLATE_TYPE translate, | |
4064 | const int multibyte)); | |
fa9a63c5 RM |
4065 | |
4066 | /* This converts PTR, a pointer into one of the search strings `string1' | |
4067 | and `string2' into an offset from the beginning of that string. */ | |
4068 | #define POINTER_TO_OFFSET(ptr) \ | |
4069 | (FIRST_STRING_P (ptr) \ | |
4070 | ? ((regoff_t) ((ptr) - string1)) \ | |
4071 | : ((regoff_t) ((ptr) - string2 + size1))) | |
4072 | ||
fa9a63c5 | 4073 | /* Call before fetching a character with *d. This switches over to |
419d1c74 SM |
4074 | string2 if necessary. |
4075 | Check re_match_2_internal for a discussion of why end_match_2 might | |
4076 | not be within string2 (but be equal to end_match_1 instead). */ | |
fa9a63c5 | 4077 | #define PREFETCH() \ |
25fe55af | 4078 | while (d == dend) \ |
fa9a63c5 RM |
4079 | { \ |
4080 | /* End of string2 => fail. */ \ | |
25fe55af RS |
4081 | if (dend == end_match_2) \ |
4082 | goto fail; \ | |
4bb91c68 | 4083 | /* End of string1 => advance to string2. */ \ |
25fe55af | 4084 | d = string2; \ |
fa9a63c5 RM |
4085 | dend = end_match_2; \ |
4086 | } | |
4087 | ||
f1ad044f SM |
4088 | /* Call before fetching a char with *d if you already checked other limits. |
4089 | This is meant for use in lookahead operations like wordend, etc.. | |
4090 | where we might need to look at parts of the string that might be | |
4091 | outside of the LIMITs (i.e past `stop'). */ | |
4092 | #define PREFETCH_NOLIMIT() \ | |
4093 | if (d == end1) \ | |
4094 | { \ | |
4095 | d = string2; \ | |
4096 | dend = end_match_2; \ | |
4097 | } \ | |
fa9a63c5 RM |
4098 | |
4099 | /* Test if at very beginning or at very end of the virtual concatenation | |
25fe55af | 4100 | of `string1' and `string2'. If only one string, it's `string2'. */ |
fa9a63c5 | 4101 | #define AT_STRINGS_BEG(d) ((d) == (size1 ? string1 : string2) || !size2) |
5e69f11e | 4102 | #define AT_STRINGS_END(d) ((d) == end2) |
fa9a63c5 RM |
4103 | |
4104 | ||
4105 | /* Test if D points to a character which is word-constituent. We have | |
4106 | two special cases to check for: if past the end of string1, look at | |
4107 | the first character in string2; and if before the beginning of | |
4108 | string2, look at the last character in string1. */ | |
4109 | #define WORDCHAR_P(d) \ | |
4110 | (SYNTAX ((d) == end1 ? *string2 \ | |
25fe55af | 4111 | : (d) == string2 - 1 ? *(end1 - 1) : *(d)) \ |
fa9a63c5 RM |
4112 | == Sword) |
4113 | ||
9121ca40 | 4114 | /* Disabled due to a compiler bug -- see comment at case wordbound */ |
b18215fc RS |
4115 | |
4116 | /* The comment at case wordbound is following one, but we don't use | |
4117 | AT_WORD_BOUNDARY anymore to support multibyte form. | |
4118 | ||
4119 | The DEC Alpha C compiler 3.x generates incorrect code for the | |
25fe55af RS |
4120 | test WORDCHAR_P (d - 1) != WORDCHAR_P (d) in the expansion of |
4121 | AT_WORD_BOUNDARY, so this code is disabled. Expanding the | |
b18215fc RS |
4122 | macro and introducing temporary variables works around the bug. */ |
4123 | ||
9121ca40 | 4124 | #if 0 |
fa9a63c5 RM |
4125 | /* Test if the character before D and the one at D differ with respect |
4126 | to being word-constituent. */ | |
4127 | #define AT_WORD_BOUNDARY(d) \ | |
4128 | (AT_STRINGS_BEG (d) || AT_STRINGS_END (d) \ | |
4129 | || WORDCHAR_P (d - 1) != WORDCHAR_P (d)) | |
9121ca40 | 4130 | #endif |
fa9a63c5 RM |
4131 | |
4132 | /* Free everything we malloc. */ | |
4133 | #ifdef MATCH_MAY_ALLOCATE | |
0b32bf0e SM |
4134 | # define FREE_VAR(var) if (var) { REGEX_FREE (var); var = NULL; } else |
4135 | # define FREE_VARIABLES() \ | |
fa9a63c5 RM |
4136 | do { \ |
4137 | REGEX_FREE_STACK (fail_stack.stack); \ | |
4138 | FREE_VAR (regstart); \ | |
4139 | FREE_VAR (regend); \ | |
fa9a63c5 RM |
4140 | FREE_VAR (best_regstart); \ |
4141 | FREE_VAR (best_regend); \ | |
fa9a63c5 RM |
4142 | } while (0) |
4143 | #else | |
0b32bf0e | 4144 | # define FREE_VARIABLES() ((void)0) /* Do nothing! But inhibit gcc warning. */ |
fa9a63c5 RM |
4145 | #endif /* not MATCH_MAY_ALLOCATE */ |
4146 | ||
505bde11 SM |
4147 | \f |
4148 | /* Optimization routines. */ | |
4149 | ||
4e8a9132 SM |
4150 | /* If the operation is a match against one or more chars, |
4151 | return a pointer to the next operation, else return NULL. */ | |
01618498 | 4152 | static re_char * |
4e8a9132 | 4153 | skip_one_char (p) |
01618498 | 4154 | re_char *p; |
4e8a9132 SM |
4155 | { |
4156 | switch (SWITCH_ENUM_CAST (*p++)) | |
4157 | { | |
4158 | case anychar: | |
4159 | break; | |
4160 | ||
4161 | case exactn: | |
4162 | p += *p + 1; | |
4163 | break; | |
4164 | ||
4165 | case charset_not: | |
4166 | case charset: | |
4167 | if (CHARSET_RANGE_TABLE_EXISTS_P (p - 1)) | |
4168 | { | |
4169 | int mcnt; | |
4170 | p = CHARSET_RANGE_TABLE (p - 1); | |
4171 | EXTRACT_NUMBER_AND_INCR (mcnt, p); | |
4172 | p = CHARSET_RANGE_TABLE_END (p, mcnt); | |
4173 | } | |
4174 | else | |
4175 | p += 1 + CHARSET_BITMAP_SIZE (p - 1); | |
4176 | break; | |
4177 | ||
4e8a9132 SM |
4178 | case syntaxspec: |
4179 | case notsyntaxspec: | |
1fb352e0 | 4180 | #ifdef emacs |
4e8a9132 SM |
4181 | case categoryspec: |
4182 | case notcategoryspec: | |
4183 | #endif /* emacs */ | |
4184 | p++; | |
4185 | break; | |
4186 | ||
4187 | default: | |
4188 | p = NULL; | |
4189 | } | |
4190 | return p; | |
4191 | } | |
4192 | ||
4193 | ||
505bde11 SM |
4194 | /* Jump over non-matching operations. */ |
4195 | static unsigned char * | |
4e8a9132 | 4196 | skip_noops (p, pend) |
505bde11 | 4197 | unsigned char *p, *pend; |
505bde11 SM |
4198 | { |
4199 | int mcnt; | |
4200 | while (p < pend) | |
4201 | { | |
4202 | switch (SWITCH_ENUM_CAST ((re_opcode_t) *p)) | |
4203 | { | |
4204 | case start_memory: | |
505bde11 SM |
4205 | case stop_memory: |
4206 | p += 2; break; | |
4207 | case no_op: | |
4208 | p += 1; break; | |
4209 | case jump: | |
4210 | p += 1; | |
4211 | EXTRACT_NUMBER_AND_INCR (mcnt, p); | |
4212 | p += mcnt; | |
4213 | break; | |
4214 | default: | |
4215 | return p; | |
4216 | } | |
4217 | } | |
4218 | assert (p == pend); | |
4219 | return p; | |
4220 | } | |
4221 | ||
4222 | /* Non-zero if "p1 matches something" implies "p2 fails". */ | |
4223 | static int | |
4224 | mutually_exclusive_p (bufp, p1, p2) | |
4225 | struct re_pattern_buffer *bufp; | |
4226 | unsigned char *p1, *p2; | |
4227 | { | |
4e8a9132 | 4228 | re_opcode_t op2; |
2d1675e4 | 4229 | const boolean multibyte = RE_MULTIBYTE_P (bufp); |
505bde11 SM |
4230 | unsigned char *pend = bufp->buffer + bufp->used; |
4231 | ||
4e8a9132 | 4232 | assert (p1 >= bufp->buffer && p1 < pend |
505bde11 SM |
4233 | && p2 >= bufp->buffer && p2 <= pend); |
4234 | ||
4235 | /* Skip over open/close-group commands. | |
4236 | If what follows this loop is a ...+ construct, | |
4237 | look at what begins its body, since we will have to | |
4238 | match at least one of that. */ | |
4e8a9132 SM |
4239 | p2 = skip_noops (p2, pend); |
4240 | /* The same skip can be done for p1, except that this function | |
4241 | is only used in the case where p1 is a simple match operator. */ | |
4242 | /* p1 = skip_noops (p1, pend); */ | |
4243 | ||
4244 | assert (p1 >= bufp->buffer && p1 < pend | |
4245 | && p2 >= bufp->buffer && p2 <= pend); | |
4246 | ||
4247 | op2 = p2 == pend ? succeed : *p2; | |
4248 | ||
4249 | switch (SWITCH_ENUM_CAST (op2)) | |
505bde11 | 4250 | { |
4e8a9132 SM |
4251 | case succeed: |
4252 | case endbuf: | |
4253 | /* If we're at the end of the pattern, we can change. */ | |
4254 | if (skip_one_char (p1)) | |
505bde11 | 4255 | { |
505bde11 SM |
4256 | DEBUG_PRINT1 (" End of pattern: fast loop.\n"); |
4257 | return 1; | |
505bde11 | 4258 | } |
4e8a9132 SM |
4259 | break; |
4260 | ||
4261 | case endline: | |
4e8a9132 SM |
4262 | case exactn: |
4263 | { | |
01618498 | 4264 | register re_wchar_t c |
4e8a9132 | 4265 | = (re_opcode_t) *p2 == endline ? '\n' |
411e4203 | 4266 | : RE_STRING_CHAR (p2 + 2, pend - p2 - 2); |
505bde11 | 4267 | |
4e8a9132 SM |
4268 | if ((re_opcode_t) *p1 == exactn) |
4269 | { | |
4270 | if (c != RE_STRING_CHAR (p1 + 2, pend - p1 - 2)) | |
4271 | { | |
4272 | DEBUG_PRINT3 (" '%c' != '%c' => fast loop.\n", c, p1[2]); | |
4273 | return 1; | |
4274 | } | |
4275 | } | |
505bde11 | 4276 | |
4e8a9132 SM |
4277 | else if ((re_opcode_t) *p1 == charset |
4278 | || (re_opcode_t) *p1 == charset_not) | |
4279 | { | |
4280 | int not = (re_opcode_t) *p1 == charset_not; | |
505bde11 | 4281 | |
4e8a9132 SM |
4282 | /* Test if C is listed in charset (or charset_not) |
4283 | at `p1'. */ | |
4284 | if (SINGLE_BYTE_CHAR_P (c)) | |
4285 | { | |
4286 | if (c < CHARSET_BITMAP_SIZE (p1) * BYTEWIDTH | |
4287 | && p1[2 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH))) | |
4288 | not = !not; | |
4289 | } | |
4290 | else if (CHARSET_RANGE_TABLE_EXISTS_P (p1)) | |
4291 | CHARSET_LOOKUP_RANGE_TABLE (not, c, p1); | |
505bde11 | 4292 | |
4e8a9132 SM |
4293 | /* `not' is equal to 1 if c would match, which means |
4294 | that we can't change to pop_failure_jump. */ | |
4295 | if (!not) | |
4296 | { | |
4297 | DEBUG_PRINT1 (" No match => fast loop.\n"); | |
4298 | return 1; | |
4299 | } | |
4300 | } | |
4301 | else if ((re_opcode_t) *p1 == anychar | |
4302 | && c == '\n') | |
4303 | { | |
4304 | DEBUG_PRINT1 (" . != \\n => fast loop.\n"); | |
4305 | return 1; | |
4306 | } | |
4307 | } | |
4308 | break; | |
505bde11 | 4309 | |
4e8a9132 | 4310 | case charset: |
4e8a9132 SM |
4311 | { |
4312 | if ((re_opcode_t) *p1 == exactn) | |
4313 | /* Reuse the code above. */ | |
4314 | return mutually_exclusive_p (bufp, p2, p1); | |
505bde11 | 4315 | |
505bde11 SM |
4316 | /* It is hard to list up all the character in charset |
4317 | P2 if it includes multibyte character. Give up in | |
4318 | such case. */ | |
4319 | else if (!multibyte || !CHARSET_RANGE_TABLE_EXISTS_P (p2)) | |
4320 | { | |
4321 | /* Now, we are sure that P2 has no range table. | |
4322 | So, for the size of bitmap in P2, `p2[1]' is | |
4323 | enough. But P1 may have range table, so the | |
4324 | size of bitmap table of P1 is extracted by | |
4325 | using macro `CHARSET_BITMAP_SIZE'. | |
4326 | ||
4327 | Since we know that all the character listed in | |
4328 | P2 is ASCII, it is enough to test only bitmap | |
4329 | table of P1. */ | |
4330 | ||
411e4203 | 4331 | if ((re_opcode_t) *p1 == charset) |
505bde11 SM |
4332 | { |
4333 | int idx; | |
4334 | /* We win if the charset inside the loop | |
4335 | has no overlap with the one after the loop. */ | |
4336 | for (idx = 0; | |
4337 | (idx < (int) p2[1] | |
4338 | && idx < CHARSET_BITMAP_SIZE (p1)); | |
4339 | idx++) | |
4340 | if ((p2[2 + idx] & p1[2 + idx]) != 0) | |
4341 | break; | |
4342 | ||
4343 | if (idx == p2[1] | |
4344 | || idx == CHARSET_BITMAP_SIZE (p1)) | |
4345 | { | |
4346 | DEBUG_PRINT1 (" No match => fast loop.\n"); | |
4347 | return 1; | |
4348 | } | |
4349 | } | |
411e4203 | 4350 | else if ((re_opcode_t) *p1 == charset_not) |
505bde11 SM |
4351 | { |
4352 | int idx; | |
4353 | /* We win if the charset_not inside the loop lists | |
4354 | every character listed in the charset after. */ | |
4355 | for (idx = 0; idx < (int) p2[1]; idx++) | |
4356 | if (! (p2[2 + idx] == 0 | |
4357 | || (idx < CHARSET_BITMAP_SIZE (p1) | |
4358 | && ((p2[2 + idx] & ~ p1[2 + idx]) == 0)))) | |
4359 | break; | |
4360 | ||
4e8a9132 SM |
4361 | if (idx == p2[1]) |
4362 | { | |
4363 | DEBUG_PRINT1 (" No match => fast loop.\n"); | |
4364 | return 1; | |
4365 | } | |
4366 | } | |
4367 | } | |
4368 | } | |
609b757a | 4369 | break; |
4e8a9132 | 4370 | |
411e4203 SM |
4371 | case charset_not: |
4372 | switch (SWITCH_ENUM_CAST (*p1)) | |
4373 | { | |
4374 | case exactn: | |
4375 | case charset: | |
4376 | /* Reuse the code above. */ | |
4377 | return mutually_exclusive_p (bufp, p2, p1); | |
4378 | case charset_not: | |
4379 | /* When we have two charset_not, it's very unlikely that | |
4380 | they don't overlap. The union of the two sets of excluded | |
4381 | chars should cover all possible chars, which, as a matter of | |
4382 | fact, is virtually impossible in multibyte buffers. */ | |
4383 | ; | |
4384 | } | |
4385 | break; | |
4386 | ||
4e8a9132 SM |
4387 | case wordend: |
4388 | case notsyntaxspec: | |
4389 | return ((re_opcode_t) *p1 == syntaxspec | |
4390 | && p1[1] == (op2 == wordend ? Sword : p2[1])); | |
4391 | ||
4392 | case wordbeg: | |
4393 | case syntaxspec: | |
4394 | return ((re_opcode_t) *p1 == notsyntaxspec | |
4395 | && p1[1] == (op2 == wordend ? Sword : p2[1])); | |
4396 | ||
4397 | case wordbound: | |
4398 | return (((re_opcode_t) *p1 == notsyntaxspec | |
4399 | || (re_opcode_t) *p1 == syntaxspec) | |
4400 | && p1[1] == Sword); | |
4401 | ||
1fb352e0 | 4402 | #ifdef emacs |
4e8a9132 SM |
4403 | case categoryspec: |
4404 | return ((re_opcode_t) *p1 == notcategoryspec && p1[1] == p2[1]); | |
4405 | case notcategoryspec: | |
4406 | return ((re_opcode_t) *p1 == categoryspec && p1[1] == p2[1]); | |
4407 | #endif /* emacs */ | |
4408 | ||
4409 | default: | |
4410 | ; | |
505bde11 SM |
4411 | } |
4412 | ||
4413 | /* Safe default. */ | |
4414 | return 0; | |
4415 | } | |
4416 | ||
fa9a63c5 RM |
4417 | \f |
4418 | /* Matching routines. */ | |
4419 | ||
25fe55af | 4420 | #ifndef emacs /* Emacs never uses this. */ |
fa9a63c5 RM |
4421 | /* re_match is like re_match_2 except it takes only a single string. */ |
4422 | ||
4423 | int | |
4424 | re_match (bufp, string, size, pos, regs) | |
4425 | struct re_pattern_buffer *bufp; | |
4426 | const char *string; | |
4427 | int size, pos; | |
4428 | struct re_registers *regs; | |
4429 | { | |
4bb91c68 | 4430 | int result = re_match_2_internal (bufp, NULL, 0, (re_char*) string, size, |
fa9a63c5 | 4431 | pos, regs, size); |
0b32bf0e | 4432 | # if defined C_ALLOCA && !defined REGEX_MALLOC |
fa9a63c5 | 4433 | alloca (0); |
0b32bf0e | 4434 | # endif |
fa9a63c5 RM |
4435 | return result; |
4436 | } | |
c0f9ea08 | 4437 | WEAK_ALIAS (__re_match, re_match) |
fa9a63c5 RM |
4438 | #endif /* not emacs */ |
4439 | ||
b18215fc RS |
4440 | #ifdef emacs |
4441 | /* In Emacs, this is the string or buffer in which we | |
25fe55af | 4442 | are matching. It is used for looking up syntax properties. */ |
b18215fc RS |
4443 | Lisp_Object re_match_object; |
4444 | #endif | |
fa9a63c5 RM |
4445 | |
4446 | /* re_match_2 matches the compiled pattern in BUFP against the | |
4447 | the (virtual) concatenation of STRING1 and STRING2 (of length SIZE1 | |
4448 | and SIZE2, respectively). We start matching at POS, and stop | |
4449 | matching at STOP. | |
5e69f11e | 4450 | |
fa9a63c5 | 4451 | If REGS is non-null and the `no_sub' field of BUFP is nonzero, we |
25fe55af | 4452 | store offsets for the substring each group matched in REGS. See the |
fa9a63c5 RM |
4453 | documentation for exactly how many groups we fill. |
4454 | ||
4455 | We return -1 if no match, -2 if an internal error (such as the | |
25fe55af | 4456 | failure stack overflowing). Otherwise, we return the length of the |
fa9a63c5 RM |
4457 | matched substring. */ |
4458 | ||
4459 | int | |
4460 | re_match_2 (bufp, string1, size1, string2, size2, pos, regs, stop) | |
4461 | struct re_pattern_buffer *bufp; | |
4462 | const char *string1, *string2; | |
4463 | int size1, size2; | |
4464 | int pos; | |
4465 | struct re_registers *regs; | |
4466 | int stop; | |
4467 | { | |
b18215fc | 4468 | int result; |
25fe55af | 4469 | |
b18215fc | 4470 | #ifdef emacs |
cc9b4df2 KH |
4471 | int charpos; |
4472 | gl_state.object = re_match_object; | |
99633e97 | 4473 | charpos = SYNTAX_TABLE_BYTE_TO_CHAR (POS_AS_IN_BUFFER (pos)); |
cc9b4df2 | 4474 | SETUP_SYNTAX_TABLE_FOR_OBJECT (re_match_object, charpos, 1); |
b18215fc RS |
4475 | #endif |
4476 | ||
4bb91c68 SM |
4477 | result = re_match_2_internal (bufp, (re_char*) string1, size1, |
4478 | (re_char*) string2, size2, | |
cc9b4df2 | 4479 | pos, regs, stop); |
0b32bf0e | 4480 | #if defined C_ALLOCA && !defined REGEX_MALLOC |
fa9a63c5 | 4481 | alloca (0); |
a60198e5 | 4482 | #endif |
fa9a63c5 RM |
4483 | return result; |
4484 | } | |
c0f9ea08 | 4485 | WEAK_ALIAS (__re_match_2, re_match_2) |
fa9a63c5 RM |
4486 | |
4487 | /* This is a separate function so that we can force an alloca cleanup | |
25fe55af | 4488 | afterwards. */ |
fa9a63c5 RM |
4489 | static int |
4490 | re_match_2_internal (bufp, string1, size1, string2, size2, pos, regs, stop) | |
4491 | struct re_pattern_buffer *bufp; | |
66f0296e | 4492 | re_char *string1, *string2; |
fa9a63c5 RM |
4493 | int size1, size2; |
4494 | int pos; | |
4495 | struct re_registers *regs; | |
4496 | int stop; | |
4497 | { | |
4498 | /* General temporaries. */ | |
4499 | int mcnt; | |
01618498 | 4500 | size_t reg; |
66f0296e | 4501 | boolean not; |
fa9a63c5 RM |
4502 | |
4503 | /* Just past the end of the corresponding string. */ | |
66f0296e | 4504 | re_char *end1, *end2; |
fa9a63c5 RM |
4505 | |
4506 | /* Pointers into string1 and string2, just past the last characters in | |
25fe55af | 4507 | each to consider matching. */ |
66f0296e | 4508 | re_char *end_match_1, *end_match_2; |
fa9a63c5 RM |
4509 | |
4510 | /* Where we are in the data, and the end of the current string. */ | |
66f0296e | 4511 | re_char *d, *dend; |
5e69f11e | 4512 | |
99633e97 SM |
4513 | /* Used sometimes to remember where we were before starting matching |
4514 | an operator so that we can go back in case of failure. This "atomic" | |
4515 | behavior of matching opcodes is indispensable to the correctness | |
4516 | of the on_failure_keep_string_jump optimization. */ | |
4517 | re_char *dfail; | |
4518 | ||
fa9a63c5 | 4519 | /* Where we are in the pattern, and the end of the pattern. */ |
01618498 SM |
4520 | re_char *p = bufp->buffer; |
4521 | re_char *pend = p + bufp->used; | |
fa9a63c5 | 4522 | |
25fe55af | 4523 | /* We use this to map every character in the string. */ |
6676cb1c | 4524 | RE_TRANSLATE_TYPE translate = bufp->translate; |
fa9a63c5 | 4525 | |
25fe55af | 4526 | /* Nonzero if we have to concern multibyte character. */ |
2d1675e4 | 4527 | const boolean multibyte = RE_MULTIBYTE_P (bufp); |
b18215fc | 4528 | |
fa9a63c5 RM |
4529 | /* Failure point stack. Each place that can handle a failure further |
4530 | down the line pushes a failure point on this stack. It consists of | |
505bde11 | 4531 | regstart, and regend for all registers corresponding to |
fa9a63c5 RM |
4532 | the subexpressions we're currently inside, plus the number of such |
4533 | registers, and, finally, two char *'s. The first char * is where | |
4534 | to resume scanning the pattern; the second one is where to resume | |
505bde11 | 4535 | scanning the strings. */ |
25fe55af | 4536 | #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */ |
fa9a63c5 RM |
4537 | fail_stack_type fail_stack; |
4538 | #endif | |
4539 | #ifdef DEBUG | |
fa9a63c5 RM |
4540 | unsigned nfailure_points_pushed = 0, nfailure_points_popped = 0; |
4541 | #endif | |
4542 | ||
0b32bf0e | 4543 | #if defined REL_ALLOC && defined REGEX_MALLOC |
fa9a63c5 RM |
4544 | /* This holds the pointer to the failure stack, when |
4545 | it is allocated relocatably. */ | |
4546 | fail_stack_elt_t *failure_stack_ptr; | |
99633e97 | 4547 | #endif |
fa9a63c5 RM |
4548 | |
4549 | /* We fill all the registers internally, independent of what we | |
25fe55af | 4550 | return, for use in backreferences. The number here includes |
fa9a63c5 | 4551 | an element for register zero. */ |
4bb91c68 | 4552 | size_t num_regs = bufp->re_nsub + 1; |
5e69f11e | 4553 | |
fa9a63c5 RM |
4554 | /* Information on the contents of registers. These are pointers into |
4555 | the input strings; they record just what was matched (on this | |
4556 | attempt) by a subexpression part of the pattern, that is, the | |
4557 | regnum-th regstart pointer points to where in the pattern we began | |
4558 | matching and the regnum-th regend points to right after where we | |
4559 | stopped matching the regnum-th subexpression. (The zeroth register | |
4560 | keeps track of what the whole pattern matches.) */ | |
4561 | #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */ | |
66f0296e | 4562 | re_char **regstart, **regend; |
fa9a63c5 RM |
4563 | #endif |
4564 | ||
fa9a63c5 | 4565 | /* The following record the register info as found in the above |
5e69f11e | 4566 | variables when we find a match better than any we've seen before. |
fa9a63c5 RM |
4567 | This happens as we backtrack through the failure points, which in |
4568 | turn happens only if we have not yet matched the entire string. */ | |
4569 | unsigned best_regs_set = false; | |
4570 | #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */ | |
66f0296e | 4571 | re_char **best_regstart, **best_regend; |
fa9a63c5 | 4572 | #endif |
5e69f11e | 4573 | |
fa9a63c5 RM |
4574 | /* Logically, this is `best_regend[0]'. But we don't want to have to |
4575 | allocate space for that if we're not allocating space for anything | |
25fe55af | 4576 | else (see below). Also, we never need info about register 0 for |
fa9a63c5 RM |
4577 | any of the other register vectors, and it seems rather a kludge to |
4578 | treat `best_regend' differently than the rest. So we keep track of | |
4579 | the end of the best match so far in a separate variable. We | |
4580 | initialize this to NULL so that when we backtrack the first time | |
4581 | and need to test it, it's not garbage. */ | |
66f0296e | 4582 | re_char *match_end = NULL; |
fa9a63c5 | 4583 | |
fa9a63c5 RM |
4584 | #ifdef DEBUG |
4585 | /* Counts the total number of registers pushed. */ | |
5e69f11e | 4586 | unsigned num_regs_pushed = 0; |
fa9a63c5 RM |
4587 | #endif |
4588 | ||
4589 | DEBUG_PRINT1 ("\n\nEntering re_match_2.\n"); | |
5e69f11e | 4590 | |
fa9a63c5 | 4591 | INIT_FAIL_STACK (); |
5e69f11e | 4592 | |
fa9a63c5 RM |
4593 | #ifdef MATCH_MAY_ALLOCATE |
4594 | /* Do not bother to initialize all the register variables if there are | |
4595 | no groups in the pattern, as it takes a fair amount of time. If | |
4596 | there are groups, we include space for register 0 (the whole | |
4597 | pattern), even though we never use it, since it simplifies the | |
4598 | array indexing. We should fix this. */ | |
4599 | if (bufp->re_nsub) | |
4600 | { | |
66f0296e SM |
4601 | regstart = REGEX_TALLOC (num_regs, re_char *); |
4602 | regend = REGEX_TALLOC (num_regs, re_char *); | |
4603 | best_regstart = REGEX_TALLOC (num_regs, re_char *); | |
4604 | best_regend = REGEX_TALLOC (num_regs, re_char *); | |
fa9a63c5 | 4605 | |
505bde11 | 4606 | if (!(regstart && regend && best_regstart && best_regend)) |
25fe55af RS |
4607 | { |
4608 | FREE_VARIABLES (); | |
4609 | return -2; | |
4610 | } | |
fa9a63c5 RM |
4611 | } |
4612 | else | |
4613 | { | |
4614 | /* We must initialize all our variables to NULL, so that | |
25fe55af | 4615 | `FREE_VARIABLES' doesn't try to free them. */ |
505bde11 | 4616 | regstart = regend = best_regstart = best_regend = NULL; |
fa9a63c5 RM |
4617 | } |
4618 | #endif /* MATCH_MAY_ALLOCATE */ | |
4619 | ||
4620 | /* The starting position is bogus. */ | |
4621 | if (pos < 0 || pos > size1 + size2) | |
4622 | { | |
4623 | FREE_VARIABLES (); | |
4624 | return -1; | |
4625 | } | |
5e69f11e | 4626 | |
fa9a63c5 RM |
4627 | /* Initialize subexpression text positions to -1 to mark ones that no |
4628 | start_memory/stop_memory has been seen for. Also initialize the | |
4629 | register information struct. */ | |
01618498 SM |
4630 | for (reg = 1; reg < num_regs; reg++) |
4631 | regstart[reg] = regend[reg] = NULL; | |
99633e97 | 4632 | |
fa9a63c5 | 4633 | /* We move `string1' into `string2' if the latter's empty -- but not if |
25fe55af | 4634 | `string1' is null. */ |
fa9a63c5 RM |
4635 | if (size2 == 0 && string1 != NULL) |
4636 | { | |
4637 | string2 = string1; | |
4638 | size2 = size1; | |
4639 | string1 = 0; | |
4640 | size1 = 0; | |
4641 | } | |
4642 | end1 = string1 + size1; | |
4643 | end2 = string2 + size2; | |
4644 | ||
5e69f11e | 4645 | /* `p' scans through the pattern as `d' scans through the data. |
fa9a63c5 RM |
4646 | `dend' is the end of the input string that `d' points within. `d' |
4647 | is advanced into the following input string whenever necessary, but | |
4648 | this happens before fetching; therefore, at the beginning of the | |
4649 | loop, `d' can be pointing at the end of a string, but it cannot | |
4650 | equal `string2'. */ | |
419d1c74 | 4651 | if (pos >= size1) |
fa9a63c5 | 4652 | { |
419d1c74 SM |
4653 | /* Only match within string2. */ |
4654 | d = string2 + pos - size1; | |
4655 | dend = end_match_2 = string2 + stop - size1; | |
4656 | end_match_1 = end1; /* Just to give it a value. */ | |
fa9a63c5 RM |
4657 | } |
4658 | else | |
4659 | { | |
f1ad044f | 4660 | if (stop < size1) |
419d1c74 SM |
4661 | { |
4662 | /* Only match within string1. */ | |
4663 | end_match_1 = string1 + stop; | |
4664 | /* BEWARE! | |
4665 | When we reach end_match_1, PREFETCH normally switches to string2. | |
4666 | But in the present case, this means that just doing a PREFETCH | |
4667 | makes us jump from `stop' to `gap' within the string. | |
4668 | What we really want here is for the search to stop as | |
4669 | soon as we hit end_match_1. That's why we set end_match_2 | |
4670 | to end_match_1 (since PREFETCH fails as soon as we hit | |
4671 | end_match_2). */ | |
4672 | end_match_2 = end_match_1; | |
4673 | } | |
4674 | else | |
f1ad044f SM |
4675 | { /* It's important to use this code when stop == size so that |
4676 | moving `d' from end1 to string2 will not prevent the d == dend | |
4677 | check from catching the end of string. */ | |
419d1c74 SM |
4678 | end_match_1 = end1; |
4679 | end_match_2 = string2 + stop - size1; | |
4680 | } | |
4681 | d = string1 + pos; | |
4682 | dend = end_match_1; | |
fa9a63c5 RM |
4683 | } |
4684 | ||
4685 | DEBUG_PRINT1 ("The compiled pattern is: "); | |
4686 | DEBUG_PRINT_COMPILED_PATTERN (bufp, p, pend); | |
4687 | DEBUG_PRINT1 ("The string to match is: `"); | |
4688 | DEBUG_PRINT_DOUBLE_STRING (d, string1, size1, string2, size2); | |
4689 | DEBUG_PRINT1 ("'\n"); | |
5e69f11e | 4690 | |
25fe55af | 4691 | /* This loops over pattern commands. It exits by returning from the |
fa9a63c5 RM |
4692 | function if the match is complete, or it drops through if the match |
4693 | fails at this starting point in the input data. */ | |
4694 | for (;;) | |
4695 | { | |
505bde11 | 4696 | DEBUG_PRINT2 ("\n%p: ", p); |
fa9a63c5 RM |
4697 | |
4698 | if (p == pend) | |
4699 | { /* End of pattern means we might have succeeded. */ | |
25fe55af | 4700 | DEBUG_PRINT1 ("end of pattern ... "); |
5e69f11e | 4701 | |
fa9a63c5 | 4702 | /* If we haven't matched the entire string, and we want the |
25fe55af RS |
4703 | longest match, try backtracking. */ |
4704 | if (d != end_match_2) | |
fa9a63c5 RM |
4705 | { |
4706 | /* 1 if this match ends in the same string (string1 or string2) | |
4707 | as the best previous match. */ | |
5e69f11e | 4708 | boolean same_str_p = (FIRST_STRING_P (match_end) |
99633e97 | 4709 | == FIRST_STRING_P (d)); |
fa9a63c5 RM |
4710 | /* 1 if this match is the best seen so far. */ |
4711 | boolean best_match_p; | |
4712 | ||
4713 | /* AIX compiler got confused when this was combined | |
25fe55af | 4714 | with the previous declaration. */ |
fa9a63c5 RM |
4715 | if (same_str_p) |
4716 | best_match_p = d > match_end; | |
4717 | else | |
99633e97 | 4718 | best_match_p = !FIRST_STRING_P (d); |
fa9a63c5 | 4719 | |
25fe55af RS |
4720 | DEBUG_PRINT1 ("backtracking.\n"); |
4721 | ||
4722 | if (!FAIL_STACK_EMPTY ()) | |
4723 | { /* More failure points to try. */ | |
4724 | ||
4725 | /* If exceeds best match so far, save it. */ | |
4726 | if (!best_regs_set || best_match_p) | |
4727 | { | |
4728 | best_regs_set = true; | |
4729 | match_end = d; | |
4730 | ||
4731 | DEBUG_PRINT1 ("\nSAVING match as best so far.\n"); | |
4732 | ||
01618498 | 4733 | for (reg = 1; reg < num_regs; reg++) |
25fe55af | 4734 | { |
01618498 SM |
4735 | best_regstart[reg] = regstart[reg]; |
4736 | best_regend[reg] = regend[reg]; | |
25fe55af RS |
4737 | } |
4738 | } | |
4739 | goto fail; | |
4740 | } | |
4741 | ||
4742 | /* If no failure points, don't restore garbage. And if | |
4743 | last match is real best match, don't restore second | |
4744 | best one. */ | |
4745 | else if (best_regs_set && !best_match_p) | |
4746 | { | |
4747 | restore_best_regs: | |
4748 | /* Restore best match. It may happen that `dend == | |
4749 | end_match_1' while the restored d is in string2. | |
4750 | For example, the pattern `x.*y.*z' against the | |
4751 | strings `x-' and `y-z-', if the two strings are | |
4752 | not consecutive in memory. */ | |
4753 | DEBUG_PRINT1 ("Restoring best registers.\n"); | |
4754 | ||
4755 | d = match_end; | |
4756 | dend = ((d >= string1 && d <= end1) | |
4757 | ? end_match_1 : end_match_2); | |
fa9a63c5 | 4758 | |
01618498 | 4759 | for (reg = 1; reg < num_regs; reg++) |
fa9a63c5 | 4760 | { |
01618498 SM |
4761 | regstart[reg] = best_regstart[reg]; |
4762 | regend[reg] = best_regend[reg]; | |
fa9a63c5 | 4763 | } |
25fe55af RS |
4764 | } |
4765 | } /* d != end_match_2 */ | |
fa9a63c5 RM |
4766 | |
4767 | succeed_label: | |
25fe55af | 4768 | DEBUG_PRINT1 ("Accepting match.\n"); |
fa9a63c5 | 4769 | |
25fe55af RS |
4770 | /* If caller wants register contents data back, do it. */ |
4771 | if (regs && !bufp->no_sub) | |
fa9a63c5 | 4772 | { |
25fe55af RS |
4773 | /* Have the register data arrays been allocated? */ |
4774 | if (bufp->regs_allocated == REGS_UNALLOCATED) | |
4775 | { /* No. So allocate them with malloc. We need one | |
4776 | extra element beyond `num_regs' for the `-1' marker | |
4777 | GNU code uses. */ | |
4778 | regs->num_regs = MAX (RE_NREGS, num_regs + 1); | |
4779 | regs->start = TALLOC (regs->num_regs, regoff_t); | |
4780 | regs->end = TALLOC (regs->num_regs, regoff_t); | |
4781 | if (regs->start == NULL || regs->end == NULL) | |
fa9a63c5 RM |
4782 | { |
4783 | FREE_VARIABLES (); | |
4784 | return -2; | |
4785 | } | |
25fe55af RS |
4786 | bufp->regs_allocated = REGS_REALLOCATE; |
4787 | } | |
4788 | else if (bufp->regs_allocated == REGS_REALLOCATE) | |
4789 | { /* Yes. If we need more elements than were already | |
4790 | allocated, reallocate them. If we need fewer, just | |
4791 | leave it alone. */ | |
4792 | if (regs->num_regs < num_regs + 1) | |
4793 | { | |
4794 | regs->num_regs = num_regs + 1; | |
4795 | RETALLOC (regs->start, regs->num_regs, regoff_t); | |
4796 | RETALLOC (regs->end, regs->num_regs, regoff_t); | |
4797 | if (regs->start == NULL || regs->end == NULL) | |
fa9a63c5 RM |
4798 | { |
4799 | FREE_VARIABLES (); | |
4800 | return -2; | |
4801 | } | |
25fe55af RS |
4802 | } |
4803 | } | |
4804 | else | |
fa9a63c5 RM |
4805 | { |
4806 | /* These braces fend off a "empty body in an else-statement" | |
25fe55af | 4807 | warning under GCC when assert expands to nothing. */ |
fa9a63c5 RM |
4808 | assert (bufp->regs_allocated == REGS_FIXED); |
4809 | } | |
4810 | ||
25fe55af RS |
4811 | /* Convert the pointer data in `regstart' and `regend' to |
4812 | indices. Register zero has to be set differently, | |
4813 | since we haven't kept track of any info for it. */ | |
4814 | if (regs->num_regs > 0) | |
4815 | { | |
4816 | regs->start[0] = pos; | |
99633e97 | 4817 | regs->end[0] = POINTER_TO_OFFSET (d); |
25fe55af | 4818 | } |
5e69f11e | 4819 | |
25fe55af RS |
4820 | /* Go through the first `min (num_regs, regs->num_regs)' |
4821 | registers, since that is all we initialized. */ | |
01618498 | 4822 | for (reg = 1; reg < MIN (num_regs, regs->num_regs); reg++) |
fa9a63c5 | 4823 | { |
01618498 SM |
4824 | if (REG_UNSET (regstart[reg]) || REG_UNSET (regend[reg])) |
4825 | regs->start[reg] = regs->end[reg] = -1; | |
25fe55af RS |
4826 | else |
4827 | { | |
01618498 SM |
4828 | regs->start[reg] |
4829 | = (regoff_t) POINTER_TO_OFFSET (regstart[reg]); | |
4830 | regs->end[reg] | |
4831 | = (regoff_t) POINTER_TO_OFFSET (regend[reg]); | |
25fe55af | 4832 | } |
fa9a63c5 | 4833 | } |
5e69f11e | 4834 | |
25fe55af RS |
4835 | /* If the regs structure we return has more elements than |
4836 | were in the pattern, set the extra elements to -1. If | |
4837 | we (re)allocated the registers, this is the case, | |
4838 | because we always allocate enough to have at least one | |
4839 | -1 at the end. */ | |
01618498 SM |
4840 | for (reg = num_regs; reg < regs->num_regs; reg++) |
4841 | regs->start[reg] = regs->end[reg] = -1; | |
fa9a63c5 RM |
4842 | } /* regs && !bufp->no_sub */ |
4843 | ||
25fe55af RS |
4844 | DEBUG_PRINT4 ("%u failure points pushed, %u popped (%u remain).\n", |
4845 | nfailure_points_pushed, nfailure_points_popped, | |
4846 | nfailure_points_pushed - nfailure_points_popped); | |
4847 | DEBUG_PRINT2 ("%u registers pushed.\n", num_regs_pushed); | |
fa9a63c5 | 4848 | |
99633e97 | 4849 | mcnt = POINTER_TO_OFFSET (d) - pos; |
fa9a63c5 | 4850 | |
25fe55af | 4851 | DEBUG_PRINT2 ("Returning %d from re_match_2.\n", mcnt); |
fa9a63c5 | 4852 | |
25fe55af RS |
4853 | FREE_VARIABLES (); |
4854 | return mcnt; | |
4855 | } | |
fa9a63c5 | 4856 | |
25fe55af | 4857 | /* Otherwise match next pattern command. */ |
fa9a63c5 RM |
4858 | switch (SWITCH_ENUM_CAST ((re_opcode_t) *p++)) |
4859 | { | |
25fe55af RS |
4860 | /* Ignore these. Used to ignore the n of succeed_n's which |
4861 | currently have n == 0. */ | |
4862 | case no_op: | |
4863 | DEBUG_PRINT1 ("EXECUTING no_op.\n"); | |
4864 | break; | |
fa9a63c5 RM |
4865 | |
4866 | case succeed: | |
25fe55af | 4867 | DEBUG_PRINT1 ("EXECUTING succeed.\n"); |
fa9a63c5 RM |
4868 | goto succeed_label; |
4869 | ||
25fe55af RS |
4870 | /* Match the next n pattern characters exactly. The following |
4871 | byte in the pattern defines n, and the n bytes after that | |
4872 | are the characters to match. */ | |
fa9a63c5 RM |
4873 | case exactn: |
4874 | mcnt = *p++; | |
25fe55af | 4875 | DEBUG_PRINT2 ("EXECUTING exactn %d.\n", mcnt); |
fa9a63c5 | 4876 | |
99633e97 SM |
4877 | /* Remember the start point to rollback upon failure. */ |
4878 | dfail = d; | |
4879 | ||
25fe55af RS |
4880 | /* This is written out as an if-else so we don't waste time |
4881 | testing `translate' inside the loop. */ | |
28703c16 | 4882 | if (RE_TRANSLATE_P (translate)) |
fa9a63c5 | 4883 | { |
e934739e RS |
4884 | if (multibyte) |
4885 | do | |
4886 | { | |
4887 | int pat_charlen, buf_charlen; | |
e71b1971 | 4888 | unsigned int pat_ch, buf_ch; |
e934739e RS |
4889 | |
4890 | PREFETCH (); | |
4891 | pat_ch = STRING_CHAR_AND_LENGTH (p, pend - p, pat_charlen); | |
4892 | buf_ch = STRING_CHAR_AND_LENGTH (d, dend - d, buf_charlen); | |
4893 | ||
4894 | if (RE_TRANSLATE (translate, buf_ch) | |
4895 | != pat_ch) | |
99633e97 SM |
4896 | { |
4897 | d = dfail; | |
4898 | goto fail; | |
4899 | } | |
e934739e RS |
4900 | |
4901 | p += pat_charlen; | |
4902 | d += buf_charlen; | |
4903 | mcnt -= pat_charlen; | |
4904 | } | |
4905 | while (mcnt > 0); | |
4906 | else | |
e934739e RS |
4907 | do |
4908 | { | |
4909 | PREFETCH (); | |
66f0296e | 4910 | if (RE_TRANSLATE (translate, *d) != *p++) |
99633e97 SM |
4911 | { |
4912 | d = dfail; | |
4913 | goto fail; | |
4914 | } | |
33c46939 | 4915 | d++; |
e934739e RS |
4916 | } |
4917 | while (--mcnt); | |
fa9a63c5 RM |
4918 | } |
4919 | else | |
4920 | { | |
4921 | do | |
4922 | { | |
4923 | PREFETCH (); | |
99633e97 SM |
4924 | if (*d++ != *p++) |
4925 | { | |
4926 | d = dfail; | |
4927 | goto fail; | |
4928 | } | |
fa9a63c5 RM |
4929 | } |
4930 | while (--mcnt); | |
4931 | } | |
25fe55af | 4932 | break; |
fa9a63c5 RM |
4933 | |
4934 | ||
25fe55af | 4935 | /* Match any character except possibly a newline or a null. */ |
fa9a63c5 | 4936 | case anychar: |
e934739e RS |
4937 | { |
4938 | int buf_charlen; | |
01618498 | 4939 | re_wchar_t buf_ch; |
fa9a63c5 | 4940 | |
e934739e | 4941 | DEBUG_PRINT1 ("EXECUTING anychar.\n"); |
fa9a63c5 | 4942 | |
e934739e | 4943 | PREFETCH (); |
2d1675e4 | 4944 | buf_ch = RE_STRING_CHAR_AND_LENGTH (d, dend - d, buf_charlen); |
e934739e RS |
4945 | buf_ch = TRANSLATE (buf_ch); |
4946 | ||
4947 | if ((!(bufp->syntax & RE_DOT_NEWLINE) | |
4948 | && buf_ch == '\n') | |
4949 | || ((bufp->syntax & RE_DOT_NOT_NULL) | |
4950 | && buf_ch == '\000')) | |
4951 | goto fail; | |
4952 | ||
e934739e RS |
4953 | DEBUG_PRINT2 (" Matched `%d'.\n", *d); |
4954 | d += buf_charlen; | |
4955 | } | |
fa9a63c5 RM |
4956 | break; |
4957 | ||
4958 | ||
4959 | case charset: | |
4960 | case charset_not: | |
4961 | { | |
b18215fc | 4962 | register unsigned int c; |
fa9a63c5 | 4963 | boolean not = (re_opcode_t) *(p - 1) == charset_not; |
b18215fc RS |
4964 | int len; |
4965 | ||
4966 | /* Start of actual range_table, or end of bitmap if there is no | |
4967 | range table. */ | |
01618498 | 4968 | re_char *range_table; |
b18215fc | 4969 | |
96cc36cc | 4970 | /* Nonzero if there is a range table. */ |
b18215fc RS |
4971 | int range_table_exists; |
4972 | ||
96cc36cc RS |
4973 | /* Number of ranges of range table. This is not included |
4974 | in the initial byte-length of the command. */ | |
4975 | int count = 0; | |
fa9a63c5 | 4976 | |
25fe55af | 4977 | DEBUG_PRINT2 ("EXECUTING charset%s.\n", not ? "_not" : ""); |
fa9a63c5 | 4978 | |
b18215fc | 4979 | range_table_exists = CHARSET_RANGE_TABLE_EXISTS_P (&p[-1]); |
96cc36cc | 4980 | |
b18215fc | 4981 | if (range_table_exists) |
96cc36cc RS |
4982 | { |
4983 | range_table = CHARSET_RANGE_TABLE (&p[-1]); /* Past the bitmap. */ | |
4984 | EXTRACT_NUMBER_AND_INCR (count, range_table); | |
4985 | } | |
b18215fc | 4986 | |
2d1675e4 SM |
4987 | PREFETCH (); |
4988 | c = RE_STRING_CHAR_AND_LENGTH (d, dend - d, len); | |
4989 | c = TRANSLATE (c); /* The character to match. */ | |
b18215fc RS |
4990 | |
4991 | if (SINGLE_BYTE_CHAR_P (c)) | |
4992 | { /* Lookup bitmap. */ | |
b18215fc RS |
4993 | /* Cast to `unsigned' instead of `unsigned char' in |
4994 | case the bit list is a full 32 bytes long. */ | |
4995 | if (c < (unsigned) (CHARSET_BITMAP_SIZE (&p[-1]) * BYTEWIDTH) | |
96cc36cc RS |
4996 | && p[1 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH))) |
4997 | not = !not; | |
b18215fc | 4998 | } |
96cc36cc | 4999 | #ifdef emacs |
b18215fc | 5000 | else if (range_table_exists) |
96cc36cc RS |
5001 | { |
5002 | int class_bits = CHARSET_RANGE_TABLE_BITS (&p[-1]); | |
5003 | ||
14473664 SM |
5004 | if ( (class_bits & BIT_LOWER && ISLOWER (c)) |
5005 | | (class_bits & BIT_MULTIBYTE) | |
96cc36cc RS |
5006 | | (class_bits & BIT_PUNCT && ISPUNCT (c)) |
5007 | | (class_bits & BIT_SPACE && ISSPACE (c)) | |
5008 | | (class_bits & BIT_UPPER && ISUPPER (c)) | |
5009 | | (class_bits & BIT_WORD && ISWORD (c))) | |
5010 | not = !not; | |
5011 | else | |
5012 | CHARSET_LOOKUP_RANGE_TABLE_RAW (not, c, range_table, count); | |
5013 | } | |
5014 | #endif /* emacs */ | |
fa9a63c5 | 5015 | |
96cc36cc RS |
5016 | if (range_table_exists) |
5017 | p = CHARSET_RANGE_TABLE_END (range_table, count); | |
5018 | else | |
5019 | p += CHARSET_BITMAP_SIZE (&p[-1]) + 1; | |
fa9a63c5 RM |
5020 | |
5021 | if (!not) goto fail; | |
5e69f11e | 5022 | |
b18215fc | 5023 | d += len; |
fa9a63c5 RM |
5024 | break; |
5025 | } | |
5026 | ||
5027 | ||
25fe55af | 5028 | /* The beginning of a group is represented by start_memory. |
505bde11 | 5029 | The argument is the register number. The text |
25fe55af RS |
5030 | matched within the group is recorded (in the internal |
5031 | registers data structure) under the register number. */ | |
5032 | case start_memory: | |
505bde11 SM |
5033 | DEBUG_PRINT2 ("EXECUTING start_memory %d:\n", *p); |
5034 | ||
5035 | /* In case we need to undo this operation (via backtracking). */ | |
5036 | PUSH_FAILURE_REG ((unsigned int)*p); | |
fa9a63c5 | 5037 | |
25fe55af | 5038 | regstart[*p] = d; |
4bb91c68 | 5039 | regend[*p] = NULL; /* probably unnecessary. -sm */ |
fa9a63c5 RM |
5040 | DEBUG_PRINT2 (" regstart: %d\n", POINTER_TO_OFFSET (regstart[*p])); |
5041 | ||
25fe55af | 5042 | /* Move past the register number and inner group count. */ |
505bde11 | 5043 | p += 1; |
25fe55af | 5044 | break; |
fa9a63c5 RM |
5045 | |
5046 | ||
25fe55af | 5047 | /* The stop_memory opcode represents the end of a group. Its |
505bde11 | 5048 | argument is the same as start_memory's: the register number. */ |
fa9a63c5 | 5049 | case stop_memory: |
505bde11 SM |
5050 | DEBUG_PRINT2 ("EXECUTING stop_memory %d:\n", *p); |
5051 | ||
5052 | assert (!REG_UNSET (regstart[*p])); | |
5053 | /* Strictly speaking, there should be code such as: | |
5054 | ||
0b32bf0e | 5055 | assert (REG_UNSET (regend[*p])); |
505bde11 SM |
5056 | PUSH_FAILURE_REGSTOP ((unsigned int)*p); |
5057 | ||
5058 | But the only info to be pushed is regend[*p] and it is known to | |
5059 | be UNSET, so there really isn't anything to push. | |
5060 | Not pushing anything, on the other hand deprives us from the | |
5061 | guarantee that regend[*p] is UNSET since undoing this operation | |
5062 | will not reset its value properly. This is not important since | |
5063 | the value will only be read on the next start_memory or at | |
5064 | the very end and both events can only happen if this stop_memory | |
5065 | is *not* undone. */ | |
fa9a63c5 | 5066 | |
25fe55af | 5067 | regend[*p] = d; |
fa9a63c5 RM |
5068 | DEBUG_PRINT2 (" regend: %d\n", POINTER_TO_OFFSET (regend[*p])); |
5069 | ||
25fe55af | 5070 | /* Move past the register number and the inner group count. */ |
505bde11 | 5071 | p += 1; |
25fe55af | 5072 | break; |
fa9a63c5 RM |
5073 | |
5074 | ||
5075 | /* \<digit> has been turned into a `duplicate' command which is | |
25fe55af RS |
5076 | followed by the numeric value of <digit> as the register number. */ |
5077 | case duplicate: | |
fa9a63c5 | 5078 | { |
66f0296e | 5079 | register re_char *d2, *dend2; |
25fe55af | 5080 | int regno = *p++; /* Get which register to match against. */ |
fa9a63c5 RM |
5081 | DEBUG_PRINT2 ("EXECUTING duplicate %d.\n", regno); |
5082 | ||
25fe55af RS |
5083 | /* Can't back reference a group which we've never matched. */ |
5084 | if (REG_UNSET (regstart[regno]) || REG_UNSET (regend[regno])) | |
5085 | goto fail; | |
5e69f11e | 5086 | |
25fe55af RS |
5087 | /* Where in input to try to start matching. */ |
5088 | d2 = regstart[regno]; | |
5e69f11e | 5089 | |
99633e97 SM |
5090 | /* Remember the start point to rollback upon failure. */ |
5091 | dfail = d; | |
5092 | ||
25fe55af RS |
5093 | /* Where to stop matching; if both the place to start and |
5094 | the place to stop matching are in the same string, then | |
5095 | set to the place to stop, otherwise, for now have to use | |
5096 | the end of the first string. */ | |
fa9a63c5 | 5097 | |
25fe55af | 5098 | dend2 = ((FIRST_STRING_P (regstart[regno]) |
fa9a63c5 RM |
5099 | == FIRST_STRING_P (regend[regno])) |
5100 | ? regend[regno] : end_match_1); | |
5101 | for (;;) | |
5102 | { | |
5103 | /* If necessary, advance to next segment in register | |
25fe55af | 5104 | contents. */ |
fa9a63c5 RM |
5105 | while (d2 == dend2) |
5106 | { | |
5107 | if (dend2 == end_match_2) break; | |
5108 | if (dend2 == regend[regno]) break; | |
5109 | ||
25fe55af RS |
5110 | /* End of string1 => advance to string2. */ |
5111 | d2 = string2; | |
5112 | dend2 = regend[regno]; | |
fa9a63c5 RM |
5113 | } |
5114 | /* At end of register contents => success */ | |
5115 | if (d2 == dend2) break; | |
5116 | ||
5117 | /* If necessary, advance to next segment in data. */ | |
5118 | PREFETCH (); | |
5119 | ||
5120 | /* How many characters left in this segment to match. */ | |
5121 | mcnt = dend - d; | |
5e69f11e | 5122 | |
fa9a63c5 | 5123 | /* Want how many consecutive characters we can match in |
25fe55af RS |
5124 | one shot, so, if necessary, adjust the count. */ |
5125 | if (mcnt > dend2 - d2) | |
fa9a63c5 | 5126 | mcnt = dend2 - d2; |
5e69f11e | 5127 | |
fa9a63c5 | 5128 | /* Compare that many; failure if mismatch, else move |
25fe55af | 5129 | past them. */ |
28703c16 | 5130 | if (RE_TRANSLATE_P (translate) |
2d1675e4 | 5131 | ? bcmp_translate (d, d2, mcnt, translate, multibyte) |
4bb91c68 | 5132 | : memcmp (d, d2, mcnt)) |
99633e97 SM |
5133 | { |
5134 | d = dfail; | |
5135 | goto fail; | |
5136 | } | |
fa9a63c5 | 5137 | d += mcnt, d2 += mcnt; |
fa9a63c5 RM |
5138 | } |
5139 | } | |
5140 | break; | |
5141 | ||
5142 | ||
25fe55af | 5143 | /* begline matches the empty string at the beginning of the string |
c0f9ea08 | 5144 | (unless `not_bol' is set in `bufp'), and after newlines. */ |
fa9a63c5 | 5145 | case begline: |
25fe55af | 5146 | DEBUG_PRINT1 ("EXECUTING begline.\n"); |
5e69f11e | 5147 | |
25fe55af RS |
5148 | if (AT_STRINGS_BEG (d)) |
5149 | { | |
5150 | if (!bufp->not_bol) break; | |
5151 | } | |
419d1c74 | 5152 | else |
25fe55af | 5153 | { |
419d1c74 SM |
5154 | unsigned char c; |
5155 | GET_CHAR_BEFORE_2 (c, d, string1, end1, string2, end2); | |
c0f9ea08 | 5156 | if (c == '\n') |
419d1c74 | 5157 | break; |
25fe55af RS |
5158 | } |
5159 | /* In all other cases, we fail. */ | |
5160 | goto fail; | |
fa9a63c5 RM |
5161 | |
5162 | ||
25fe55af | 5163 | /* endline is the dual of begline. */ |
fa9a63c5 | 5164 | case endline: |
25fe55af | 5165 | DEBUG_PRINT1 ("EXECUTING endline.\n"); |
fa9a63c5 | 5166 | |
25fe55af RS |
5167 | if (AT_STRINGS_END (d)) |
5168 | { | |
5169 | if (!bufp->not_eol) break; | |
5170 | } | |
f1ad044f | 5171 | else |
25fe55af | 5172 | { |
f1ad044f | 5173 | PREFETCH_NOLIMIT (); |
c0f9ea08 | 5174 | if (*d == '\n') |
f1ad044f | 5175 | break; |
25fe55af RS |
5176 | } |
5177 | goto fail; | |
fa9a63c5 RM |
5178 | |
5179 | ||
5180 | /* Match at the very beginning of the data. */ | |
25fe55af RS |
5181 | case begbuf: |
5182 | DEBUG_PRINT1 ("EXECUTING begbuf.\n"); | |
5183 | if (AT_STRINGS_BEG (d)) | |
5184 | break; | |
5185 | goto fail; | |
fa9a63c5 RM |
5186 | |
5187 | ||
5188 | /* Match at the very end of the data. */ | |
25fe55af RS |
5189 | case endbuf: |
5190 | DEBUG_PRINT1 ("EXECUTING endbuf.\n"); | |
fa9a63c5 RM |
5191 | if (AT_STRINGS_END (d)) |
5192 | break; | |
25fe55af | 5193 | goto fail; |
5e69f11e | 5194 | |
5e69f11e | 5195 | |
25fe55af RS |
5196 | /* on_failure_keep_string_jump is used to optimize `.*\n'. It |
5197 | pushes NULL as the value for the string on the stack. Then | |
505bde11 | 5198 | `POP_FAILURE_POINT' will keep the current value for the |
25fe55af RS |
5199 | string, instead of restoring it. To see why, consider |
5200 | matching `foo\nbar' against `.*\n'. The .* matches the foo; | |
5201 | then the . fails against the \n. But the next thing we want | |
5202 | to do is match the \n against the \n; if we restored the | |
5203 | string value, we would be back at the foo. | |
5204 | ||
5205 | Because this is used only in specific cases, we don't need to | |
5206 | check all the things that `on_failure_jump' does, to make | |
5207 | sure the right things get saved on the stack. Hence we don't | |
5208 | share its code. The only reason to push anything on the | |
5209 | stack at all is that otherwise we would have to change | |
5210 | `anychar's code to do something besides goto fail in this | |
5211 | case; that seems worse than this. */ | |
5212 | case on_failure_keep_string_jump: | |
505bde11 SM |
5213 | EXTRACT_NUMBER_AND_INCR (mcnt, p); |
5214 | DEBUG_PRINT3 ("EXECUTING on_failure_keep_string_jump %d (to %p):\n", | |
5215 | mcnt, p + mcnt); | |
fa9a63c5 | 5216 | |
505bde11 SM |
5217 | PUSH_FAILURE_POINT (p - 3, NULL); |
5218 | break; | |
5219 | ||
0683b6fa SM |
5220 | /* A nasty loop is introduced by the non-greedy *? and +?. |
5221 | With such loops, the stack only ever contains one failure point | |
5222 | at a time, so that a plain on_failure_jump_loop kind of | |
5223 | cycle detection cannot work. Worse yet, such a detection | |
5224 | can not only fail to detect a cycle, but it can also wrongly | |
5225 | detect a cycle (between different instantiations of the same | |
5226 | loop. | |
5227 | So the method used for those nasty loops is a little different: | |
5228 | We use a special cycle-detection-stack-frame which is pushed | |
5229 | when the on_failure_jump_nastyloop failure-point is *popped*. | |
5230 | This special frame thus marks the beginning of one iteration | |
5231 | through the loop and we can hence easily check right here | |
5232 | whether something matched between the beginning and the end of | |
5233 | the loop. */ | |
5234 | case on_failure_jump_nastyloop: | |
5235 | EXTRACT_NUMBER_AND_INCR (mcnt, p); | |
5236 | DEBUG_PRINT3 ("EXECUTING on_failure_jump_nastyloop %d (to %p):\n", | |
5237 | mcnt, p + mcnt); | |
5238 | ||
5239 | assert ((re_opcode_t)p[-4] == no_op); | |
5240 | CHECK_INFINITE_LOOP (p - 4, d); | |
5241 | PUSH_FAILURE_POINT (p - 3, d); | |
5242 | break; | |
5243 | ||
505bde11 | 5244 | |
4e8a9132 SM |
5245 | /* Simple loop detecting on_failure_jump: just check on the |
5246 | failure stack if the same spot was already hit earlier. */ | |
505bde11 SM |
5247 | case on_failure_jump_loop: |
5248 | on_failure: | |
5249 | EXTRACT_NUMBER_AND_INCR (mcnt, p); | |
5250 | DEBUG_PRINT3 ("EXECUTING on_failure_jump_loop %d (to %p):\n", | |
5251 | mcnt, p + mcnt); | |
5252 | ||
5253 | CHECK_INFINITE_LOOP (p - 3, d); | |
5254 | PUSH_FAILURE_POINT (p - 3, d); | |
25fe55af | 5255 | break; |
fa9a63c5 RM |
5256 | |
5257 | ||
5258 | /* Uses of on_failure_jump: | |
5e69f11e | 5259 | |
25fe55af RS |
5260 | Each alternative starts with an on_failure_jump that points |
5261 | to the beginning of the next alternative. Each alternative | |
5262 | except the last ends with a jump that in effect jumps past | |
5263 | the rest of the alternatives. (They really jump to the | |
5264 | ending jump of the following alternative, because tensioning | |
5265 | these jumps is a hassle.) | |
fa9a63c5 | 5266 | |
25fe55af RS |
5267 | Repeats start with an on_failure_jump that points past both |
5268 | the repetition text and either the following jump or | |
5269 | pop_failure_jump back to this on_failure_jump. */ | |
fa9a63c5 | 5270 | case on_failure_jump: |
5b370c2b | 5271 | IMMEDIATE_QUIT_CHECK; |
25fe55af | 5272 | EXTRACT_NUMBER_AND_INCR (mcnt, p); |
505bde11 SM |
5273 | DEBUG_PRINT3 ("EXECUTING on_failure_jump %d (to %p):\n", |
5274 | mcnt, p + mcnt); | |
25fe55af | 5275 | |
505bde11 | 5276 | PUSH_FAILURE_POINT (p -3, d); |
25fe55af RS |
5277 | break; |
5278 | ||
4e8a9132 | 5279 | /* This operation is used for greedy *. |
505bde11 SM |
5280 | Compare the beginning of the repeat with what in the |
5281 | pattern follows its end. If we can establish that there | |
5282 | is nothing that they would both match, i.e., that we | |
5283 | would have to backtrack because of (as in, e.g., `a*a') | |
5284 | then we can use a non-backtracking loop based on | |
4e8a9132 | 5285 | on_failure_keep_string_jump instead of on_failure_jump. */ |
505bde11 | 5286 | case on_failure_jump_smart: |
5b370c2b | 5287 | IMMEDIATE_QUIT_CHECK; |
25fe55af | 5288 | EXTRACT_NUMBER_AND_INCR (mcnt, p); |
505bde11 SM |
5289 | DEBUG_PRINT3 ("EXECUTING on_failure_jump_smart %d (to %p).\n", |
5290 | mcnt, p + mcnt); | |
25fe55af | 5291 | { |
01618498 | 5292 | re_char *p1 = p; /* Next operation. */ |
6dcf2d0e SM |
5293 | /* Here, we discard `const', making re_match non-reentrant. */ |
5294 | unsigned char *p2 = (unsigned char*) p + mcnt; /* Jump dest. */ | |
5295 | unsigned char *p3 = (unsigned char*) p - 3; /* opcode location. */ | |
fa9a63c5 | 5296 | |
505bde11 SM |
5297 | p -= 3; /* Reset so that we will re-execute the |
5298 | instruction once it's been changed. */ | |
fa9a63c5 | 5299 | |
4e8a9132 SM |
5300 | EXTRACT_NUMBER (mcnt, p2 - 2); |
5301 | ||
5302 | /* Ensure this is a indeed the trivial kind of loop | |
5303 | we are expecting. */ | |
5304 | assert (skip_one_char (p1) == p2 - 3); | |
5305 | assert ((re_opcode_t) p2[-3] == jump && p2 + mcnt == p); | |
99633e97 | 5306 | DEBUG_STATEMENT (debug += 2); |
505bde11 | 5307 | if (mutually_exclusive_p (bufp, p1, p2)) |
fa9a63c5 | 5308 | { |
505bde11 | 5309 | /* Use a fast `on_failure_keep_string_jump' loop. */ |
4e8a9132 | 5310 | DEBUG_PRINT1 (" smart exclusive => fast loop.\n"); |
01618498 | 5311 | *p3 = (unsigned char) on_failure_keep_string_jump; |
4e8a9132 | 5312 | STORE_NUMBER (p2 - 2, mcnt + 3); |
25fe55af | 5313 | } |
505bde11 | 5314 | else |
fa9a63c5 | 5315 | { |
505bde11 SM |
5316 | /* Default to a safe `on_failure_jump' loop. */ |
5317 | DEBUG_PRINT1 (" smart default => slow loop.\n"); | |
01618498 | 5318 | *p3 = (unsigned char) on_failure_jump; |
fa9a63c5 | 5319 | } |
99633e97 | 5320 | DEBUG_STATEMENT (debug -= 2); |
25fe55af | 5321 | } |
505bde11 | 5322 | break; |
25fe55af RS |
5323 | |
5324 | /* Unconditionally jump (without popping any failure points). */ | |
5325 | case jump: | |
fa9a63c5 | 5326 | unconditional_jump: |
5b370c2b | 5327 | IMMEDIATE_QUIT_CHECK; |
fa9a63c5 | 5328 | EXTRACT_NUMBER_AND_INCR (mcnt, p); /* Get the amount to jump. */ |
25fe55af RS |
5329 | DEBUG_PRINT2 ("EXECUTING jump %d ", mcnt); |
5330 | p += mcnt; /* Do the jump. */ | |
505bde11 | 5331 | DEBUG_PRINT2 ("(to %p).\n", p); |
25fe55af RS |
5332 | break; |
5333 | ||
5334 | ||
25fe55af RS |
5335 | /* Have to succeed matching what follows at least n times. |
5336 | After that, handle like `on_failure_jump'. */ | |
5337 | case succeed_n: | |
01618498 | 5338 | /* Signedness doesn't matter since we only compare MCNT to 0. */ |
25fe55af RS |
5339 | EXTRACT_NUMBER (mcnt, p + 2); |
5340 | DEBUG_PRINT2 ("EXECUTING succeed_n %d.\n", mcnt); | |
5e69f11e | 5341 | |
dc1e502d SM |
5342 | /* Originally, mcnt is how many times we HAVE to succeed. */ |
5343 | if (mcnt != 0) | |
25fe55af | 5344 | { |
6dcf2d0e SM |
5345 | /* Here, we discard `const', making re_match non-reentrant. */ |
5346 | unsigned char *p2 = (unsigned char*) p + 2; /* counter loc. */ | |
dc1e502d | 5347 | mcnt--; |
01618498 SM |
5348 | p += 4; |
5349 | PUSH_NUMBER (p2, mcnt); | |
25fe55af | 5350 | } |
dc1e502d SM |
5351 | else |
5352 | /* The two bytes encoding mcnt == 0 are two no_op opcodes. */ | |
5353 | goto on_failure; | |
25fe55af RS |
5354 | break; |
5355 | ||
5356 | case jump_n: | |
01618498 | 5357 | /* Signedness doesn't matter since we only compare MCNT to 0. */ |
25fe55af RS |
5358 | EXTRACT_NUMBER (mcnt, p + 2); |
5359 | DEBUG_PRINT2 ("EXECUTING jump_n %d.\n", mcnt); | |
5360 | ||
5361 | /* Originally, this is how many times we CAN jump. */ | |
dc1e502d | 5362 | if (mcnt != 0) |
25fe55af | 5363 | { |
6dcf2d0e SM |
5364 | /* Here, we discard `const', making re_match non-reentrant. */ |
5365 | unsigned char *p2 = (unsigned char*) p + 2; /* counter loc. */ | |
dc1e502d | 5366 | mcnt--; |
01618498 | 5367 | PUSH_NUMBER (p2, mcnt); |
dc1e502d | 5368 | goto unconditional_jump; |
25fe55af RS |
5369 | } |
5370 | /* If don't have to jump any more, skip over the rest of command. */ | |
5e69f11e RM |
5371 | else |
5372 | p += 4; | |
25fe55af | 5373 | break; |
5e69f11e | 5374 | |
fa9a63c5 RM |
5375 | case set_number_at: |
5376 | { | |
01618498 | 5377 | unsigned char *p2; /* Location of the counter. */ |
25fe55af | 5378 | DEBUG_PRINT1 ("EXECUTING set_number_at.\n"); |
fa9a63c5 | 5379 | |
25fe55af | 5380 | EXTRACT_NUMBER_AND_INCR (mcnt, p); |
6dcf2d0e SM |
5381 | /* Here, we discard `const', making re_match non-reentrant. */ |
5382 | p2 = (unsigned char*) p + mcnt; | |
01618498 | 5383 | /* Signedness doesn't matter since we only copy MCNT's bits . */ |
25fe55af | 5384 | EXTRACT_NUMBER_AND_INCR (mcnt, p); |
01618498 SM |
5385 | DEBUG_PRINT3 (" Setting %p to %d.\n", p2, mcnt); |
5386 | PUSH_NUMBER (p2, mcnt); | |
25fe55af RS |
5387 | break; |
5388 | } | |
9121ca40 KH |
5389 | |
5390 | case wordbound: | |
66f0296e SM |
5391 | case notwordbound: |
5392 | not = (re_opcode_t) *(p - 1) == notwordbound; | |
5393 | DEBUG_PRINT2 ("EXECUTING %swordbound.\n", not?"not":""); | |
fa9a63c5 | 5394 | |
99633e97 | 5395 | /* We SUCCEED (or FAIL) in one of the following cases: */ |
9121ca40 | 5396 | |
b18215fc | 5397 | /* Case 1: D is at the beginning or the end of string. */ |
9121ca40 | 5398 | if (AT_STRINGS_BEG (d) || AT_STRINGS_END (d)) |
66f0296e | 5399 | not = !not; |
b18215fc RS |
5400 | else |
5401 | { | |
5402 | /* C1 is the character before D, S1 is the syntax of C1, C2 | |
5403 | is the character at D, and S2 is the syntax of C2. */ | |
01618498 SM |
5404 | re_wchar_t c1, c2; |
5405 | int s1, s2; | |
b18215fc | 5406 | #ifdef emacs |
2d1675e4 SM |
5407 | int offset = PTR_TO_OFFSET (d - 1); |
5408 | int charpos = SYNTAX_TABLE_BYTE_TO_CHAR (offset); | |
5d967c7a | 5409 | UPDATE_SYNTAX_TABLE (charpos); |
25fe55af | 5410 | #endif |
66f0296e | 5411 | GET_CHAR_BEFORE_2 (c1, d, string1, end1, string2, end2); |
b18215fc RS |
5412 | s1 = SYNTAX (c1); |
5413 | #ifdef emacs | |
5d967c7a | 5414 | UPDATE_SYNTAX_TABLE_FORWARD (charpos + 1); |
25fe55af | 5415 | #endif |
f1ad044f | 5416 | PREFETCH_NOLIMIT (); |
2d1675e4 | 5417 | c2 = RE_STRING_CHAR (d, dend - d); |
b18215fc RS |
5418 | s2 = SYNTAX (c2); |
5419 | ||
5420 | if (/* Case 2: Only one of S1 and S2 is Sword. */ | |
5421 | ((s1 == Sword) != (s2 == Sword)) | |
5422 | /* Case 3: Both of S1 and S2 are Sword, and macro | |
25fe55af | 5423 | WORD_BOUNDARY_P (C1, C2) returns nonzero. */ |
b18215fc | 5424 | || ((s1 == Sword) && WORD_BOUNDARY_P (c1, c2))) |
66f0296e SM |
5425 | not = !not; |
5426 | } | |
5427 | if (not) | |
9121ca40 | 5428 | break; |
b18215fc | 5429 | else |
9121ca40 | 5430 | goto fail; |
fa9a63c5 RM |
5431 | |
5432 | case wordbeg: | |
25fe55af | 5433 | DEBUG_PRINT1 ("EXECUTING wordbeg.\n"); |
fa9a63c5 | 5434 | |
b18215fc RS |
5435 | /* We FAIL in one of the following cases: */ |
5436 | ||
25fe55af | 5437 | /* Case 1: D is at the end of string. */ |
b18215fc | 5438 | if (AT_STRINGS_END (d)) |
99633e97 | 5439 | goto fail; |
b18215fc RS |
5440 | else |
5441 | { | |
5442 | /* C1 is the character before D, S1 is the syntax of C1, C2 | |
5443 | is the character at D, and S2 is the syntax of C2. */ | |
01618498 SM |
5444 | re_wchar_t c1, c2; |
5445 | int s1, s2; | |
fa9a63c5 | 5446 | #ifdef emacs |
2d1675e4 SM |
5447 | int offset = PTR_TO_OFFSET (d); |
5448 | int charpos = SYNTAX_TABLE_BYTE_TO_CHAR (offset); | |
92432794 | 5449 | UPDATE_SYNTAX_TABLE (charpos); |
25fe55af | 5450 | #endif |
99633e97 | 5451 | PREFETCH (); |
2d1675e4 | 5452 | c2 = RE_STRING_CHAR (d, dend - d); |
b18215fc | 5453 | s2 = SYNTAX (c2); |
25fe55af | 5454 | |
b18215fc RS |
5455 | /* Case 2: S2 is not Sword. */ |
5456 | if (s2 != Sword) | |
5457 | goto fail; | |
5458 | ||
5459 | /* Case 3: D is not at the beginning of string ... */ | |
5460 | if (!AT_STRINGS_BEG (d)) | |
5461 | { | |
5462 | GET_CHAR_BEFORE_2 (c1, d, string1, end1, string2, end2); | |
5463 | #ifdef emacs | |
5d967c7a | 5464 | UPDATE_SYNTAX_TABLE_BACKWARD (charpos - 1); |
25fe55af | 5465 | #endif |
b18215fc RS |
5466 | s1 = SYNTAX (c1); |
5467 | ||
5468 | /* ... and S1 is Sword, and WORD_BOUNDARY_P (C1, C2) | |
25fe55af | 5469 | returns 0. */ |
b18215fc RS |
5470 | if ((s1 == Sword) && !WORD_BOUNDARY_P (c1, c2)) |
5471 | goto fail; | |
5472 | } | |
5473 | } | |
e318085a RS |
5474 | break; |
5475 | ||
b18215fc | 5476 | case wordend: |
25fe55af | 5477 | DEBUG_PRINT1 ("EXECUTING wordend.\n"); |
b18215fc RS |
5478 | |
5479 | /* We FAIL in one of the following cases: */ | |
5480 | ||
5481 | /* Case 1: D is at the beginning of string. */ | |
5482 | if (AT_STRINGS_BEG (d)) | |
e318085a | 5483 | goto fail; |
b18215fc RS |
5484 | else |
5485 | { | |
5486 | /* C1 is the character before D, S1 is the syntax of C1, C2 | |
5487 | is the character at D, and S2 is the syntax of C2. */ | |
01618498 SM |
5488 | re_wchar_t c1, c2; |
5489 | int s1, s2; | |
5d967c7a | 5490 | #ifdef emacs |
2d1675e4 SM |
5491 | int offset = PTR_TO_OFFSET (d) - 1; |
5492 | int charpos = SYNTAX_TABLE_BYTE_TO_CHAR (offset); | |
92432794 | 5493 | UPDATE_SYNTAX_TABLE (charpos); |
5d967c7a | 5494 | #endif |
99633e97 | 5495 | GET_CHAR_BEFORE_2 (c1, d, string1, end1, string2, end2); |
b18215fc RS |
5496 | s1 = SYNTAX (c1); |
5497 | ||
5498 | /* Case 2: S1 is not Sword. */ | |
5499 | if (s1 != Sword) | |
5500 | goto fail; | |
5501 | ||
5502 | /* Case 3: D is not at the end of string ... */ | |
5503 | if (!AT_STRINGS_END (d)) | |
5504 | { | |
f1ad044f | 5505 | PREFETCH_NOLIMIT (); |
2d1675e4 | 5506 | c2 = RE_STRING_CHAR (d, dend - d); |
5d967c7a RS |
5507 | #ifdef emacs |
5508 | UPDATE_SYNTAX_TABLE_FORWARD (charpos); | |
5509 | #endif | |
b18215fc RS |
5510 | s2 = SYNTAX (c2); |
5511 | ||
5512 | /* ... and S2 is Sword, and WORD_BOUNDARY_P (C1, C2) | |
25fe55af | 5513 | returns 0. */ |
b18215fc | 5514 | if ((s2 == Sword) && !WORD_BOUNDARY_P (c1, c2)) |
25fe55af | 5515 | goto fail; |
b18215fc RS |
5516 | } |
5517 | } | |
e318085a RS |
5518 | break; |
5519 | ||
fa9a63c5 | 5520 | case syntaxspec: |
1fb352e0 SM |
5521 | case notsyntaxspec: |
5522 | not = (re_opcode_t) *(p - 1) == notsyntaxspec; | |
fa9a63c5 | 5523 | mcnt = *p++; |
1fb352e0 | 5524 | DEBUG_PRINT3 ("EXECUTING %ssyntaxspec %d.\n", not?"not":"", mcnt); |
fa9a63c5 | 5525 | PREFETCH (); |
b18215fc RS |
5526 | #ifdef emacs |
5527 | { | |
2d1675e4 SM |
5528 | int offset = PTR_TO_OFFSET (d); |
5529 | int pos1 = SYNTAX_TABLE_BYTE_TO_CHAR (offset); | |
b18215fc RS |
5530 | UPDATE_SYNTAX_TABLE (pos1); |
5531 | } | |
25fe55af | 5532 | #endif |
b18215fc | 5533 | { |
01618498 SM |
5534 | int len; |
5535 | re_wchar_t c; | |
b18215fc | 5536 | |
2d1675e4 | 5537 | c = RE_STRING_CHAR_AND_LENGTH (d, dend - d, len); |
b18215fc | 5538 | |
990b2375 | 5539 | if ((SYNTAX (c) != (enum syntaxcode) mcnt) ^ not) |
1fb352e0 | 5540 | goto fail; |
b18215fc RS |
5541 | d += len; |
5542 | } | |
fa9a63c5 RM |
5543 | break; |
5544 | ||
b18215fc | 5545 | #ifdef emacs |
1fb352e0 SM |
5546 | case before_dot: |
5547 | DEBUG_PRINT1 ("EXECUTING before_dot.\n"); | |
5548 | if (PTR_BYTE_POS (d) >= PT_BYTE) | |
fa9a63c5 | 5549 | goto fail; |
b18215fc RS |
5550 | break; |
5551 | ||
1fb352e0 SM |
5552 | case at_dot: |
5553 | DEBUG_PRINT1 ("EXECUTING at_dot.\n"); | |
5554 | if (PTR_BYTE_POS (d) != PT_BYTE) | |
5555 | goto fail; | |
5556 | break; | |
b18215fc | 5557 | |
1fb352e0 SM |
5558 | case after_dot: |
5559 | DEBUG_PRINT1 ("EXECUTING after_dot.\n"); | |
5560 | if (PTR_BYTE_POS (d) <= PT_BYTE) | |
5561 | goto fail; | |
e318085a | 5562 | break; |
fa9a63c5 | 5563 | |
1fb352e0 | 5564 | case categoryspec: |
b18215fc | 5565 | case notcategoryspec: |
1fb352e0 | 5566 | not = (re_opcode_t) *(p - 1) == notcategoryspec; |
b18215fc | 5567 | mcnt = *p++; |
1fb352e0 | 5568 | DEBUG_PRINT3 ("EXECUTING %scategoryspec %d.\n", not?"not":"", mcnt); |
b18215fc RS |
5569 | PREFETCH (); |
5570 | { | |
01618498 SM |
5571 | int len; |
5572 | re_wchar_t c; | |
5573 | ||
2d1675e4 | 5574 | c = RE_STRING_CHAR_AND_LENGTH (d, dend - d, len); |
b18215fc | 5575 | |
1fb352e0 | 5576 | if ((!CHAR_HAS_CATEGORY (c, mcnt)) ^ not) |
b18215fc RS |
5577 | goto fail; |
5578 | d += len; | |
5579 | } | |
fa9a63c5 | 5580 | break; |
5e69f11e | 5581 | |
1fb352e0 | 5582 | #endif /* emacs */ |
5e69f11e | 5583 | |
0b32bf0e SM |
5584 | default: |
5585 | abort (); | |
fa9a63c5 | 5586 | } |
b18215fc | 5587 | continue; /* Successfully executed one pattern command; keep going. */ |
fa9a63c5 RM |
5588 | |
5589 | ||
5590 | /* We goto here if a matching operation fails. */ | |
5591 | fail: | |
5b370c2b | 5592 | IMMEDIATE_QUIT_CHECK; |
fa9a63c5 | 5593 | if (!FAIL_STACK_EMPTY ()) |
505bde11 | 5594 | { |
01618498 | 5595 | re_char *str, *pat; |
505bde11 | 5596 | /* A restart point is known. Restore to that state. */ |
0b32bf0e SM |
5597 | DEBUG_PRINT1 ("\nFAIL:\n"); |
5598 | POP_FAILURE_POINT (str, pat); | |
505bde11 SM |
5599 | switch (SWITCH_ENUM_CAST ((re_opcode_t) *pat++)) |
5600 | { | |
5601 | case on_failure_keep_string_jump: | |
5602 | assert (str == NULL); | |
5603 | goto continue_failure_jump; | |
5604 | ||
0683b6fa SM |
5605 | case on_failure_jump_nastyloop: |
5606 | assert ((re_opcode_t)pat[-2] == no_op); | |
5607 | PUSH_FAILURE_POINT (pat - 2, str); | |
5608 | /* Fallthrough */ | |
5609 | ||
505bde11 SM |
5610 | case on_failure_jump_loop: |
5611 | case on_failure_jump: | |
5612 | case succeed_n: | |
5613 | d = str; | |
5614 | continue_failure_jump: | |
5615 | EXTRACT_NUMBER_AND_INCR (mcnt, pat); | |
5616 | p = pat + mcnt; | |
5617 | break; | |
b18215fc | 5618 | |
0683b6fa SM |
5619 | case no_op: |
5620 | /* A special frame used for nastyloops. */ | |
5621 | goto fail; | |
5622 | ||
505bde11 SM |
5623 | default: |
5624 | abort(); | |
5625 | } | |
fa9a63c5 | 5626 | |
505bde11 | 5627 | assert (p >= bufp->buffer && p <= pend); |
b18215fc | 5628 | |
0b32bf0e | 5629 | if (d >= string1 && d <= end1) |
fa9a63c5 | 5630 | dend = end_match_1; |
0b32bf0e | 5631 | } |
fa9a63c5 | 5632 | else |
0b32bf0e | 5633 | break; /* Matching at this starting point really fails. */ |
fa9a63c5 RM |
5634 | } /* for (;;) */ |
5635 | ||
5636 | if (best_regs_set) | |
5637 | goto restore_best_regs; | |
5638 | ||
5639 | FREE_VARIABLES (); | |
5640 | ||
b18215fc | 5641 | return -1; /* Failure to match. */ |
fa9a63c5 RM |
5642 | } /* re_match_2 */ |
5643 | \f | |
5644 | /* Subroutine definitions for re_match_2. */ | |
5645 | ||
fa9a63c5 RM |
5646 | /* Return zero if TRANSLATE[S1] and TRANSLATE[S2] are identical for LEN |
5647 | bytes; nonzero otherwise. */ | |
5e69f11e | 5648 | |
fa9a63c5 | 5649 | static int |
2d1675e4 SM |
5650 | bcmp_translate (s1, s2, len, translate, multibyte) |
5651 | re_char *s1, *s2; | |
fa9a63c5 | 5652 | register int len; |
6676cb1c | 5653 | RE_TRANSLATE_TYPE translate; |
2d1675e4 | 5654 | const int multibyte; |
fa9a63c5 | 5655 | { |
2d1675e4 SM |
5656 | register re_char *p1 = s1, *p2 = s2; |
5657 | re_char *p1_end = s1 + len; | |
5658 | re_char *p2_end = s2 + len; | |
e934739e | 5659 | |
4bb91c68 SM |
5660 | /* FIXME: Checking both p1 and p2 presumes that the two strings might have |
5661 | different lengths, but relying on a single `len' would break this. -sm */ | |
5662 | while (p1 < p1_end && p2 < p2_end) | |
fa9a63c5 | 5663 | { |
e934739e | 5664 | int p1_charlen, p2_charlen; |
01618498 | 5665 | re_wchar_t p1_ch, p2_ch; |
e934739e | 5666 | |
2d1675e4 SM |
5667 | p1_ch = RE_STRING_CHAR_AND_LENGTH (p1, p1_end - p1, p1_charlen); |
5668 | p2_ch = RE_STRING_CHAR_AND_LENGTH (p2, p2_end - p2, p2_charlen); | |
e934739e RS |
5669 | |
5670 | if (RE_TRANSLATE (translate, p1_ch) | |
5671 | != RE_TRANSLATE (translate, p2_ch)) | |
bc192b5b | 5672 | return 1; |
e934739e RS |
5673 | |
5674 | p1 += p1_charlen, p2 += p2_charlen; | |
fa9a63c5 | 5675 | } |
e934739e RS |
5676 | |
5677 | if (p1 != p1_end || p2 != p2_end) | |
5678 | return 1; | |
5679 | ||
fa9a63c5 RM |
5680 | return 0; |
5681 | } | |
5682 | \f | |
5683 | /* Entry points for GNU code. */ | |
5684 | ||
5685 | /* re_compile_pattern is the GNU regular expression compiler: it | |
5686 | compiles PATTERN (of length SIZE) and puts the result in BUFP. | |
5687 | Returns 0 if the pattern was valid, otherwise an error string. | |
5e69f11e | 5688 | |
fa9a63c5 RM |
5689 | Assumes the `allocated' (and perhaps `buffer') and `translate' fields |
5690 | are set in BUFP on entry. | |
5e69f11e | 5691 | |
b18215fc | 5692 | We call regex_compile to do the actual compilation. */ |
fa9a63c5 RM |
5693 | |
5694 | const char * | |
5695 | re_compile_pattern (pattern, length, bufp) | |
5696 | const char *pattern; | |
0b32bf0e | 5697 | size_t length; |
fa9a63c5 RM |
5698 | struct re_pattern_buffer *bufp; |
5699 | { | |
5700 | reg_errcode_t ret; | |
5e69f11e | 5701 | |
fa9a63c5 RM |
5702 | /* GNU code is written to assume at least RE_NREGS registers will be set |
5703 | (and at least one extra will be -1). */ | |
5704 | bufp->regs_allocated = REGS_UNALLOCATED; | |
5e69f11e | 5705 | |
fa9a63c5 RM |
5706 | /* And GNU code determines whether or not to get register information |
5707 | by passing null for the REGS argument to re_match, etc., not by | |
5708 | setting no_sub. */ | |
5709 | bufp->no_sub = 0; | |
5e69f11e | 5710 | |
4bb91c68 | 5711 | ret = regex_compile ((re_char*) pattern, length, re_syntax_options, bufp); |
fa9a63c5 RM |
5712 | |
5713 | if (!ret) | |
5714 | return NULL; | |
5715 | return gettext (re_error_msgid[(int) ret]); | |
5e69f11e | 5716 | } |
c0f9ea08 | 5717 | WEAK_ALIAS (__re_compile_pattern, re_compile_pattern) |
fa9a63c5 | 5718 | \f |
b18215fc RS |
5719 | /* Entry points compatible with 4.2 BSD regex library. We don't define |
5720 | them unless specifically requested. */ | |
fa9a63c5 | 5721 | |
0b32bf0e | 5722 | #if defined _REGEX_RE_COMP || defined _LIBC |
fa9a63c5 RM |
5723 | |
5724 | /* BSD has one and only one pattern buffer. */ | |
5725 | static struct re_pattern_buffer re_comp_buf; | |
5726 | ||
5727 | char * | |
0b32bf0e | 5728 | # ifdef _LIBC |
48afdd44 RM |
5729 | /* Make these definitions weak in libc, so POSIX programs can redefine |
5730 | these names if they don't use our functions, and still use | |
5731 | regcomp/regexec below without link errors. */ | |
5732 | weak_function | |
0b32bf0e | 5733 | # endif |
fa9a63c5 RM |
5734 | re_comp (s) |
5735 | const char *s; | |
5736 | { | |
5737 | reg_errcode_t ret; | |
5e69f11e | 5738 | |
fa9a63c5 RM |
5739 | if (!s) |
5740 | { | |
5741 | if (!re_comp_buf.buffer) | |
0b32bf0e | 5742 | /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */ |
a60198e5 | 5743 | return (char *) gettext ("No previous regular expression"); |
fa9a63c5 RM |
5744 | return 0; |
5745 | } | |
5746 | ||
5747 | if (!re_comp_buf.buffer) | |
5748 | { | |
5749 | re_comp_buf.buffer = (unsigned char *) malloc (200); | |
5750 | if (re_comp_buf.buffer == NULL) | |
0b32bf0e SM |
5751 | /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */ |
5752 | return (char *) gettext (re_error_msgid[(int) REG_ESPACE]); | |
fa9a63c5 RM |
5753 | re_comp_buf.allocated = 200; |
5754 | ||
5755 | re_comp_buf.fastmap = (char *) malloc (1 << BYTEWIDTH); | |
5756 | if (re_comp_buf.fastmap == NULL) | |
a60198e5 SM |
5757 | /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */ |
5758 | return (char *) gettext (re_error_msgid[(int) REG_ESPACE]); | |
fa9a63c5 RM |
5759 | } |
5760 | ||
5761 | /* Since `re_exec' always passes NULL for the `regs' argument, we | |
5762 | don't need to initialize the pattern buffer fields which affect it. */ | |
5763 | ||
fa9a63c5 | 5764 | ret = regex_compile (s, strlen (s), re_syntax_options, &re_comp_buf); |
5e69f11e | 5765 | |
fa9a63c5 RM |
5766 | if (!ret) |
5767 | return NULL; | |
5768 | ||
5769 | /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */ | |
5770 | return (char *) gettext (re_error_msgid[(int) ret]); | |
5771 | } | |
5772 | ||
5773 | ||
5774 | int | |
0b32bf0e | 5775 | # ifdef _LIBC |
48afdd44 | 5776 | weak_function |
0b32bf0e | 5777 | # endif |
fa9a63c5 RM |
5778 | re_exec (s) |
5779 | const char *s; | |
5780 | { | |
5781 | const int len = strlen (s); | |
5782 | return | |
5783 | 0 <= re_search (&re_comp_buf, s, len, 0, len, (struct re_registers *) 0); | |
5784 | } | |
5785 | #endif /* _REGEX_RE_COMP */ | |
5786 | \f | |
5787 | /* POSIX.2 functions. Don't define these for Emacs. */ | |
5788 | ||
5789 | #ifndef emacs | |
5790 | ||
5791 | /* regcomp takes a regular expression as a string and compiles it. | |
5792 | ||
b18215fc | 5793 | PREG is a regex_t *. We do not expect any fields to be initialized, |
fa9a63c5 RM |
5794 | since POSIX says we shouldn't. Thus, we set |
5795 | ||
5796 | `buffer' to the compiled pattern; | |
5797 | `used' to the length of the compiled pattern; | |
5798 | `syntax' to RE_SYNTAX_POSIX_EXTENDED if the | |
5799 | REG_EXTENDED bit in CFLAGS is set; otherwise, to | |
5800 | RE_SYNTAX_POSIX_BASIC; | |
c0f9ea08 SM |
5801 | `fastmap' to an allocated space for the fastmap; |
5802 | `fastmap_accurate' to zero; | |
fa9a63c5 RM |
5803 | `re_nsub' to the number of subexpressions in PATTERN. |
5804 | ||
5805 | PATTERN is the address of the pattern string. | |
5806 | ||
5807 | CFLAGS is a series of bits which affect compilation. | |
5808 | ||
5809 | If REG_EXTENDED is set, we use POSIX extended syntax; otherwise, we | |
5810 | use POSIX basic syntax. | |
5811 | ||
5812 | If REG_NEWLINE is set, then . and [^...] don't match newline. | |
5813 | Also, regexec will try a match beginning after every newline. | |
5814 | ||
5815 | If REG_ICASE is set, then we considers upper- and lowercase | |
5816 | versions of letters to be equivalent when matching. | |
5817 | ||
5818 | If REG_NOSUB is set, then when PREG is passed to regexec, that | |
5819 | routine will report only success or failure, and nothing about the | |
5820 | registers. | |
5821 | ||
b18215fc | 5822 | It returns 0 if it succeeds, nonzero if it doesn't. (See regex.h for |
fa9a63c5 RM |
5823 | the return codes and their meanings.) */ |
5824 | ||
5825 | int | |
5826 | regcomp (preg, pattern, cflags) | |
5827 | regex_t *preg; | |
5e69f11e | 5828 | const char *pattern; |
fa9a63c5 RM |
5829 | int cflags; |
5830 | { | |
5831 | reg_errcode_t ret; | |
4bb91c68 | 5832 | reg_syntax_t syntax |
fa9a63c5 RM |
5833 | = (cflags & REG_EXTENDED) ? |
5834 | RE_SYNTAX_POSIX_EXTENDED : RE_SYNTAX_POSIX_BASIC; | |
5835 | ||
5836 | /* regex_compile will allocate the space for the compiled pattern. */ | |
5837 | preg->buffer = 0; | |
5838 | preg->allocated = 0; | |
5839 | preg->used = 0; | |
5e69f11e | 5840 | |
c0f9ea08 SM |
5841 | /* Try to allocate space for the fastmap. */ |
5842 | preg->fastmap = (char *) malloc (1 << BYTEWIDTH); | |
5e69f11e | 5843 | |
fa9a63c5 RM |
5844 | if (cflags & REG_ICASE) |
5845 | { | |
5846 | unsigned i; | |
5e69f11e | 5847 | |
6676cb1c RS |
5848 | preg->translate |
5849 | = (RE_TRANSLATE_TYPE) malloc (CHAR_SET_SIZE | |
5850 | * sizeof (*(RE_TRANSLATE_TYPE)0)); | |
fa9a63c5 | 5851 | if (preg->translate == NULL) |
0b32bf0e | 5852 | return (int) REG_ESPACE; |
fa9a63c5 RM |
5853 | |
5854 | /* Map uppercase characters to corresponding lowercase ones. */ | |
5855 | for (i = 0; i < CHAR_SET_SIZE; i++) | |
4bb91c68 | 5856 | preg->translate[i] = ISUPPER (i) ? TOLOWER (i) : i; |
fa9a63c5 RM |
5857 | } |
5858 | else | |
5859 | preg->translate = NULL; | |
5860 | ||
5861 | /* If REG_NEWLINE is set, newlines are treated differently. */ | |
5862 | if (cflags & REG_NEWLINE) | |
5863 | { /* REG_NEWLINE implies neither . nor [^...] match newline. */ | |
5864 | syntax &= ~RE_DOT_NEWLINE; | |
5865 | syntax |= RE_HAT_LISTS_NOT_NEWLINE; | |
fa9a63c5 RM |
5866 | } |
5867 | else | |
c0f9ea08 | 5868 | syntax |= RE_NO_NEWLINE_ANCHOR; |
fa9a63c5 RM |
5869 | |
5870 | preg->no_sub = !!(cflags & REG_NOSUB); | |
5871 | ||
5e69f11e | 5872 | /* POSIX says a null character in the pattern terminates it, so we |
fa9a63c5 | 5873 | can use strlen here in compiling the pattern. */ |
4bb91c68 | 5874 | ret = regex_compile ((re_char*) pattern, strlen (pattern), syntax, preg); |
5e69f11e | 5875 | |
fa9a63c5 RM |
5876 | /* POSIX doesn't distinguish between an unmatched open-group and an |
5877 | unmatched close-group: both are REG_EPAREN. */ | |
c0f9ea08 SM |
5878 | if (ret == REG_ERPAREN) |
5879 | ret = REG_EPAREN; | |
5880 | ||
5881 | if (ret == REG_NOERROR && preg->fastmap) | |
5882 | { /* Compute the fastmap now, since regexec cannot modify the pattern | |
5883 | buffer. */ | |
5884 | re_compile_fastmap (preg); | |
5885 | if (preg->can_be_null) | |
5886 | { /* The fastmap can't be used anyway. */ | |
5887 | free (preg->fastmap); | |
5888 | preg->fastmap = NULL; | |
5889 | } | |
5890 | } | |
fa9a63c5 RM |
5891 | return (int) ret; |
5892 | } | |
c0f9ea08 | 5893 | WEAK_ALIAS (__regcomp, regcomp) |
fa9a63c5 RM |
5894 | |
5895 | ||
5896 | /* regexec searches for a given pattern, specified by PREG, in the | |
5897 | string STRING. | |
5e69f11e | 5898 | |
fa9a63c5 | 5899 | If NMATCH is zero or REG_NOSUB was set in the cflags argument to |
b18215fc | 5900 | `regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at |
fa9a63c5 RM |
5901 | least NMATCH elements, and we set them to the offsets of the |
5902 | corresponding matched substrings. | |
5e69f11e | 5903 | |
fa9a63c5 RM |
5904 | EFLAGS specifies `execution flags' which affect matching: if |
5905 | REG_NOTBOL is set, then ^ does not match at the beginning of the | |
5906 | string; if REG_NOTEOL is set, then $ does not match at the end. | |
5e69f11e | 5907 | |
fa9a63c5 RM |
5908 | We return 0 if we find a match and REG_NOMATCH if not. */ |
5909 | ||
5910 | int | |
5911 | regexec (preg, string, nmatch, pmatch, eflags) | |
5912 | const regex_t *preg; | |
5e69f11e RM |
5913 | const char *string; |
5914 | size_t nmatch; | |
5915 | regmatch_t pmatch[]; | |
fa9a63c5 RM |
5916 | int eflags; |
5917 | { | |
5918 | int ret; | |
5919 | struct re_registers regs; | |
5920 | regex_t private_preg; | |
5921 | int len = strlen (string); | |
c0f9ea08 | 5922 | boolean want_reg_info = !preg->no_sub && nmatch > 0 && pmatch; |
fa9a63c5 RM |
5923 | |
5924 | private_preg = *preg; | |
5e69f11e | 5925 | |
fa9a63c5 RM |
5926 | private_preg.not_bol = !!(eflags & REG_NOTBOL); |
5927 | private_preg.not_eol = !!(eflags & REG_NOTEOL); | |
5e69f11e | 5928 | |
fa9a63c5 RM |
5929 | /* The user has told us exactly how many registers to return |
5930 | information about, via `nmatch'. We have to pass that on to the | |
b18215fc | 5931 | matching routines. */ |
fa9a63c5 | 5932 | private_preg.regs_allocated = REGS_FIXED; |
5e69f11e | 5933 | |
fa9a63c5 RM |
5934 | if (want_reg_info) |
5935 | { | |
5936 | regs.num_regs = nmatch; | |
4bb91c68 SM |
5937 | regs.start = TALLOC (nmatch * 2, regoff_t); |
5938 | if (regs.start == NULL) | |
0b32bf0e | 5939 | return (int) REG_NOMATCH; |
4bb91c68 | 5940 | regs.end = regs.start + nmatch; |
fa9a63c5 RM |
5941 | } |
5942 | ||
c0f9ea08 SM |
5943 | /* Instead of using not_eol to implement REG_NOTEOL, we could simply |
5944 | pass (&private_preg, string, len + 1, 0, len, ...) pretending the string | |
5945 | was a little bit longer but still only matching the real part. | |
5946 | This works because the `endline' will check for a '\n' and will find a | |
5947 | '\0', correctly deciding that this is not the end of a line. | |
5948 | But it doesn't work out so nicely for REG_NOTBOL, since we don't have | |
5949 | a convenient '\0' there. For all we know, the string could be preceded | |
5950 | by '\n' which would throw things off. */ | |
5951 | ||
fa9a63c5 RM |
5952 | /* Perform the searching operation. */ |
5953 | ret = re_search (&private_preg, string, len, | |
0b32bf0e SM |
5954 | /* start: */ 0, /* range: */ len, |
5955 | want_reg_info ? ®s : (struct re_registers *) 0); | |
5e69f11e | 5956 | |
fa9a63c5 RM |
5957 | /* Copy the register information to the POSIX structure. */ |
5958 | if (want_reg_info) | |
5959 | { | |
5960 | if (ret >= 0) | |
0b32bf0e SM |
5961 | { |
5962 | unsigned r; | |
fa9a63c5 | 5963 | |
0b32bf0e SM |
5964 | for (r = 0; r < nmatch; r++) |
5965 | { | |
5966 | pmatch[r].rm_so = regs.start[r]; | |
5967 | pmatch[r].rm_eo = regs.end[r]; | |
5968 | } | |
5969 | } | |
fa9a63c5 | 5970 | |
b18215fc | 5971 | /* If we needed the temporary register info, free the space now. */ |
fa9a63c5 | 5972 | free (regs.start); |
fa9a63c5 RM |
5973 | } |
5974 | ||
5975 | /* We want zero return to mean success, unlike `re_search'. */ | |
5976 | return ret >= 0 ? (int) REG_NOERROR : (int) REG_NOMATCH; | |
5977 | } | |
c0f9ea08 | 5978 | WEAK_ALIAS (__regexec, regexec) |
fa9a63c5 RM |
5979 | |
5980 | ||
5981 | /* Returns a message corresponding to an error code, ERRCODE, returned | |
5982 | from either regcomp or regexec. We don't use PREG here. */ | |
5983 | ||
5984 | size_t | |
5985 | regerror (errcode, preg, errbuf, errbuf_size) | |
5986 | int errcode; | |
5987 | const regex_t *preg; | |
5988 | char *errbuf; | |
5989 | size_t errbuf_size; | |
5990 | { | |
5991 | const char *msg; | |
5992 | size_t msg_size; | |
5993 | ||
5994 | if (errcode < 0 | |
5995 | || errcode >= (sizeof (re_error_msgid) / sizeof (re_error_msgid[0]))) | |
5e69f11e | 5996 | /* Only error codes returned by the rest of the code should be passed |
b18215fc | 5997 | to this routine. If we are given anything else, or if other regex |
fa9a63c5 RM |
5998 | code generates an invalid error code, then the program has a bug. |
5999 | Dump core so we can fix it. */ | |
6000 | abort (); | |
6001 | ||
6002 | msg = gettext (re_error_msgid[errcode]); | |
6003 | ||
6004 | msg_size = strlen (msg) + 1; /* Includes the null. */ | |
5e69f11e | 6005 | |
fa9a63c5 RM |
6006 | if (errbuf_size != 0) |
6007 | { | |
6008 | if (msg_size > errbuf_size) | |
0b32bf0e SM |
6009 | { |
6010 | strncpy (errbuf, msg, errbuf_size - 1); | |
6011 | errbuf[errbuf_size - 1] = 0; | |
6012 | } | |
fa9a63c5 | 6013 | else |
0b32bf0e | 6014 | strcpy (errbuf, msg); |
fa9a63c5 RM |
6015 | } |
6016 | ||
6017 | return msg_size; | |
6018 | } | |
c0f9ea08 | 6019 | WEAK_ALIAS (__regerror, regerror) |
fa9a63c5 RM |
6020 | |
6021 | ||
6022 | /* Free dynamically allocated space used by PREG. */ | |
6023 | ||
6024 | void | |
6025 | regfree (preg) | |
6026 | regex_t *preg; | |
6027 | { | |
6028 | if (preg->buffer != NULL) | |
6029 | free (preg->buffer); | |
6030 | preg->buffer = NULL; | |
5e69f11e | 6031 | |
fa9a63c5 RM |
6032 | preg->allocated = 0; |
6033 | preg->used = 0; | |
6034 | ||
6035 | if (preg->fastmap != NULL) | |
6036 | free (preg->fastmap); | |
6037 | preg->fastmap = NULL; | |
6038 | preg->fastmap_accurate = 0; | |
6039 | ||
6040 | if (preg->translate != NULL) | |
6041 | free (preg->translate); | |
6042 | preg->translate = NULL; | |
6043 | } | |
c0f9ea08 | 6044 | WEAK_ALIAS (__regfree, regfree) |
fa9a63c5 RM |
6045 | |
6046 | #endif /* not emacs */ |