Commit | Line | Data |
---|---|---|
e318085a | 1 | /* Extended regular expression matching and search library, version |
0b32bf0e | 2 | 0.12. (Implements POSIX draft P1003.2/D11.2, except for some of the |
bc78d348 KB |
3 | internationalization features.) |
4 | ||
0b5538bd TTN |
5 | Copyright (C) 1993, 1994, 1995, 1996, 1997, 1998, 1999, 2000, 2001, |
6 | 2002, 2003, 2004, 2005 Free Software Foundation, Inc. | |
bc78d348 | 7 | |
fa9a63c5 RM |
8 | This program is free software; you can redistribute it and/or modify |
9 | it under the terms of the GNU General Public License as published by | |
10 | the Free Software Foundation; either version 2, or (at your option) | |
11 | any later version. | |
12 | ||
13 | This program is distributed in the hope that it will be useful, | |
14 | but WITHOUT ANY WARRANTY; without even the implied warranty of | |
25fe55af | 15 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
fa9a63c5 RM |
16 | GNU General Public License for more details. |
17 | ||
18 | You should have received a copy of the GNU General Public License | |
19 | along with this program; if not, write to the Free Software | |
4fc5845f | 20 | Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, |
25fe55af | 21 | USA. */ |
fa9a63c5 | 22 | |
6df42991 | 23 | /* TODO: |
505bde11 | 24 | - structure the opcode space into opcode+flag. |
dc1e502d | 25 | - merge with glibc's regex.[ch]. |
01618498 | 26 | - replace (succeed_n + jump_n + set_number_at) with something that doesn't |
6dcf2d0e SM |
27 | need to modify the compiled regexp so that re_match can be reentrant. |
28 | - get rid of on_failure_jump_smart by doing the optimization in re_comp | |
29 | rather than at run-time, so that re_match can be reentrant. | |
01618498 | 30 | */ |
505bde11 | 31 | |
fa9a63c5 | 32 | /* AIX requires this to be the first thing in the file. */ |
0b32bf0e | 33 | #if defined _AIX && !defined REGEX_MALLOC |
fa9a63c5 RM |
34 | #pragma alloca |
35 | #endif | |
36 | ||
fa9a63c5 | 37 | #ifdef HAVE_CONFIG_H |
0b32bf0e | 38 | # include <config.h> |
fa9a63c5 RM |
39 | #endif |
40 | ||
4bb91c68 SM |
41 | #if defined STDC_HEADERS && !defined emacs |
42 | # include <stddef.h> | |
43 | #else | |
44 | /* We need this for `regex.h', and perhaps for the Emacs include files. */ | |
45 | # include <sys/types.h> | |
46 | #endif | |
fa9a63c5 | 47 | |
14473664 SM |
48 | /* Whether to use ISO C Amendment 1 wide char functions. |
49 | Those should not be used for Emacs since it uses its own. */ | |
5e5388f6 GM |
50 | #if defined _LIBC |
51 | #define WIDE_CHAR_SUPPORT 1 | |
52 | #else | |
14473664 | 53 | #define WIDE_CHAR_SUPPORT \ |
5e5388f6 GM |
54 | (HAVE_WCTYPE_H && HAVE_WCHAR_H && HAVE_BTOWC && !emacs) |
55 | #endif | |
14473664 SM |
56 | |
57 | /* For platform which support the ISO C amendement 1 functionality we | |
58 | support user defined character classes. */ | |
a0ad02f7 | 59 | #if WIDE_CHAR_SUPPORT |
14473664 SM |
60 | /* Solaris 2.5 has a bug: <wchar.h> must be included before <wctype.h>. */ |
61 | # include <wchar.h> | |
62 | # include <wctype.h> | |
63 | #endif | |
64 | ||
c0f9ea08 SM |
65 | #ifdef _LIBC |
66 | /* We have to keep the namespace clean. */ | |
67 | # define regfree(preg) __regfree (preg) | |
68 | # define regexec(pr, st, nm, pm, ef) __regexec (pr, st, nm, pm, ef) | |
69 | # define regcomp(preg, pattern, cflags) __regcomp (preg, pattern, cflags) | |
70 | # define regerror(errcode, preg, errbuf, errbuf_size) \ | |
71 | __regerror(errcode, preg, errbuf, errbuf_size) | |
72 | # define re_set_registers(bu, re, nu, st, en) \ | |
73 | __re_set_registers (bu, re, nu, st, en) | |
74 | # define re_match_2(bufp, string1, size1, string2, size2, pos, regs, stop) \ | |
75 | __re_match_2 (bufp, string1, size1, string2, size2, pos, regs, stop) | |
76 | # define re_match(bufp, string, size, pos, regs) \ | |
77 | __re_match (bufp, string, size, pos, regs) | |
78 | # define re_search(bufp, string, size, startpos, range, regs) \ | |
79 | __re_search (bufp, string, size, startpos, range, regs) | |
80 | # define re_compile_pattern(pattern, length, bufp) \ | |
81 | __re_compile_pattern (pattern, length, bufp) | |
82 | # define re_set_syntax(syntax) __re_set_syntax (syntax) | |
83 | # define re_search_2(bufp, st1, s1, st2, s2, startpos, range, regs, stop) \ | |
84 | __re_search_2 (bufp, st1, s1, st2, s2, startpos, range, regs, stop) | |
85 | # define re_compile_fastmap(bufp) __re_compile_fastmap (bufp) | |
86 | ||
14473664 SM |
87 | /* Make sure we call libc's function even if the user overrides them. */ |
88 | # define btowc __btowc | |
89 | # define iswctype __iswctype | |
90 | # define wctype __wctype | |
91 | ||
c0f9ea08 SM |
92 | # define WEAK_ALIAS(a,b) weak_alias (a, b) |
93 | ||
94 | /* We are also using some library internals. */ | |
95 | # include <locale/localeinfo.h> | |
96 | # include <locale/elem-hash.h> | |
97 | # include <langinfo.h> | |
98 | #else | |
99 | # define WEAK_ALIAS(a,b) | |
100 | #endif | |
101 | ||
4bb91c68 | 102 | /* This is for other GNU distributions with internationalized messages. */ |
0b32bf0e | 103 | #if HAVE_LIBINTL_H || defined _LIBC |
fa9a63c5 RM |
104 | # include <libintl.h> |
105 | #else | |
106 | # define gettext(msgid) (msgid) | |
107 | #endif | |
108 | ||
5e69f11e RM |
109 | #ifndef gettext_noop |
110 | /* This define is so xgettext can find the internationalizable | |
111 | strings. */ | |
0b32bf0e | 112 | # define gettext_noop(String) String |
5e69f11e RM |
113 | #endif |
114 | ||
fa9a63c5 RM |
115 | /* The `emacs' switch turns on certain matching commands |
116 | that make sense only in Emacs. */ | |
117 | #ifdef emacs | |
118 | ||
0b32bf0e SM |
119 | # include "lisp.h" |
120 | # include "buffer.h" | |
b18215fc RS |
121 | |
122 | /* Make syntax table lookup grant data in gl_state. */ | |
0b32bf0e | 123 | # define SYNTAX_ENTRY_VIA_PROPERTY |
b18215fc | 124 | |
0b32bf0e SM |
125 | # include "syntax.h" |
126 | # include "charset.h" | |
127 | # include "category.h" | |
fa9a63c5 | 128 | |
7689ef0b EZ |
129 | # ifdef malloc |
130 | # undef malloc | |
131 | # endif | |
0b32bf0e | 132 | # define malloc xmalloc |
7689ef0b EZ |
133 | # ifdef realloc |
134 | # undef realloc | |
135 | # endif | |
0b32bf0e | 136 | # define realloc xrealloc |
7689ef0b EZ |
137 | # ifdef free |
138 | # undef free | |
139 | # endif | |
0b32bf0e | 140 | # define free xfree |
9abbd165 | 141 | |
0b32bf0e SM |
142 | /* Converts the pointer to the char to BEG-based offset from the start. */ |
143 | # define PTR_TO_OFFSET(d) POS_AS_IN_BUFFER (POINTER_TO_OFFSET (d)) | |
144 | # define POS_AS_IN_BUFFER(p) ((p) + (NILP (re_match_object) || BUFFERP (re_match_object))) | |
145 | ||
146 | # define RE_MULTIBYTE_P(bufp) ((bufp)->multibyte) | |
147 | # define RE_STRING_CHAR(p, s) \ | |
4e8a9132 | 148 | (multibyte ? (STRING_CHAR (p, s)) : (*(p))) |
0b32bf0e | 149 | # define RE_STRING_CHAR_AND_LENGTH(p, s, len) \ |
2d1675e4 SM |
150 | (multibyte ? (STRING_CHAR_AND_LENGTH (p, s, len)) : ((len) = 1, *(p))) |
151 | ||
152 | /* Set C a (possibly multibyte) character before P. P points into a | |
153 | string which is the virtual concatenation of STR1 (which ends at | |
154 | END1) or STR2 (which ends at END2). */ | |
0b32bf0e | 155 | # define GET_CHAR_BEFORE_2(c, p, str1, end1, str2, end2) \ |
2d1675e4 SM |
156 | do { \ |
157 | if (multibyte) \ | |
158 | { \ | |
0b32bf0e SM |
159 | re_char *dtemp = (p) == (str2) ? (end1) : (p); \ |
160 | re_char *dlimit = ((p) > (str2) && (p) <= (end2)) ? (str2) : (str1); \ | |
70806df6 KH |
161 | re_char *d0 = dtemp; \ |
162 | PREV_CHAR_BOUNDARY (d0, dlimit); \ | |
163 | c = STRING_CHAR (d0, dtemp - d0); \ | |
2d1675e4 SM |
164 | } \ |
165 | else \ | |
166 | (c = ((p) == (str2) ? (end1) : (p))[-1]); \ | |
167 | } while (0) | |
168 | ||
4e8a9132 | 169 | |
fa9a63c5 RM |
170 | #else /* not emacs */ |
171 | ||
172 | /* If we are not linking with Emacs proper, | |
173 | we can't use the relocating allocator | |
174 | even if config.h says that we can. */ | |
0b32bf0e | 175 | # undef REL_ALLOC |
fa9a63c5 | 176 | |
0b32bf0e SM |
177 | # if defined STDC_HEADERS || defined _LIBC |
178 | # include <stdlib.h> | |
179 | # else | |
fa9a63c5 RM |
180 | char *malloc (); |
181 | char *realloc (); | |
0b32bf0e | 182 | # endif |
fa9a63c5 | 183 | |
9e4ecb26 | 184 | /* When used in Emacs's lib-src, we need to get bzero and bcopy somehow. |
4bb91c68 | 185 | If nothing else has been done, use the method below. */ |
0b32bf0e SM |
186 | # ifdef INHIBIT_STRING_HEADER |
187 | # if !(defined HAVE_BZERO && defined HAVE_BCOPY) | |
188 | # if !defined bzero && !defined bcopy | |
189 | # undef INHIBIT_STRING_HEADER | |
190 | # endif | |
191 | # endif | |
192 | # endif | |
9e4ecb26 | 193 | |
4bb91c68 | 194 | /* This is the normal way of making sure we have memcpy, memcmp and bzero. |
9e4ecb26 KH |
195 | This is used in most programs--a few other programs avoid this |
196 | by defining INHIBIT_STRING_HEADER. */ | |
0b32bf0e SM |
197 | # ifndef INHIBIT_STRING_HEADER |
198 | # if defined HAVE_STRING_H || defined STDC_HEADERS || defined _LIBC | |
199 | # include <string.h> | |
0b32bf0e | 200 | # ifndef bzero |
4bb91c68 SM |
201 | # ifndef _LIBC |
202 | # define bzero(s, n) (memset (s, '\0', n), (s)) | |
203 | # else | |
204 | # define bzero(s, n) __bzero (s, n) | |
205 | # endif | |
0b32bf0e SM |
206 | # endif |
207 | # else | |
208 | # include <strings.h> | |
4bb91c68 SM |
209 | # ifndef memcmp |
210 | # define memcmp(s1, s2, n) bcmp (s1, s2, n) | |
211 | # endif | |
212 | # ifndef memcpy | |
213 | # define memcpy(d, s, n) (bcopy (s, d, n), (d)) | |
214 | # endif | |
0b32bf0e SM |
215 | # endif |
216 | # endif | |
fa9a63c5 RM |
217 | |
218 | /* Define the syntax stuff for \<, \>, etc. */ | |
219 | ||
990b2375 | 220 | /* Sword must be nonzero for the wordchar pattern commands in re_match_2. */ |
669fa600 | 221 | enum syntaxcode { Swhitespace = 0, Sword = 1, Ssymbol = 2 }; |
fa9a63c5 | 222 | |
0b32bf0e SM |
223 | # ifdef SWITCH_ENUM_BUG |
224 | # define SWITCH_ENUM_CAST(x) ((int)(x)) | |
225 | # else | |
226 | # define SWITCH_ENUM_CAST(x) (x) | |
227 | # endif | |
fa9a63c5 | 228 | |
e934739e | 229 | /* Dummy macros for non-Emacs environments. */ |
0b32bf0e SM |
230 | # define BASE_LEADING_CODE_P(c) (0) |
231 | # define CHAR_CHARSET(c) 0 | |
232 | # define CHARSET_LEADING_CODE_BASE(c) 0 | |
233 | # define MAX_MULTIBYTE_LENGTH 1 | |
234 | # define RE_MULTIBYTE_P(x) 0 | |
235 | # define WORD_BOUNDARY_P(c1, c2) (0) | |
236 | # define CHAR_HEAD_P(p) (1) | |
237 | # define SINGLE_BYTE_CHAR_P(c) (1) | |
238 | # define SAME_CHARSET_P(c1, c2) (1) | |
239 | # define MULTIBYTE_FORM_LENGTH(p, s) (1) | |
70806df6 | 240 | # define PREV_CHAR_BOUNDARY(p, limit) ((p)--) |
0b32bf0e SM |
241 | # define STRING_CHAR(p, s) (*(p)) |
242 | # define RE_STRING_CHAR STRING_CHAR | |
243 | # define CHAR_STRING(c, s) (*(s) = (c), 1) | |
244 | # define STRING_CHAR_AND_LENGTH(p, s, actual_len) ((actual_len) = 1, *(p)) | |
245 | # define RE_STRING_CHAR_AND_LENGTH STRING_CHAR_AND_LENGTH | |
246 | # define GET_CHAR_BEFORE_2(c, p, str1, end1, str2, end2) \ | |
b18215fc | 247 | (c = ((p) == (str2) ? *((end1) - 1) : *((p) - 1))) |
0b32bf0e | 248 | # define MAKE_CHAR(charset, c1, c2) (c1) |
fa9a63c5 | 249 | #endif /* not emacs */ |
4e8a9132 SM |
250 | |
251 | #ifndef RE_TRANSLATE | |
0b32bf0e SM |
252 | # define RE_TRANSLATE(TBL, C) ((unsigned char)(TBL)[C]) |
253 | # define RE_TRANSLATE_P(TBL) (TBL) | |
4e8a9132 | 254 | #endif |
fa9a63c5 RM |
255 | \f |
256 | /* Get the interface, including the syntax bits. */ | |
257 | #include "regex.h" | |
258 | ||
f71b19b6 DL |
259 | /* isalpha etc. are used for the character classes. */ |
260 | #include <ctype.h> | |
fa9a63c5 | 261 | |
f71b19b6 | 262 | #ifdef emacs |
fa9a63c5 | 263 | |
f71b19b6 | 264 | /* 1 if C is an ASCII character. */ |
0b32bf0e | 265 | # define IS_REAL_ASCII(c) ((c) < 0200) |
fa9a63c5 | 266 | |
f71b19b6 | 267 | /* 1 if C is a unibyte character. */ |
0b32bf0e | 268 | # define ISUNIBYTE(c) (SINGLE_BYTE_CHAR_P ((c))) |
96cc36cc | 269 | |
f71b19b6 | 270 | /* The Emacs definitions should not be directly affected by locales. */ |
96cc36cc | 271 | |
f71b19b6 | 272 | /* In Emacs, these are only used for single-byte characters. */ |
0b32bf0e SM |
273 | # define ISDIGIT(c) ((c) >= '0' && (c) <= '9') |
274 | # define ISCNTRL(c) ((c) < ' ') | |
275 | # define ISXDIGIT(c) (((c) >= '0' && (c) <= '9') \ | |
f71b19b6 DL |
276 | || ((c) >= 'a' && (c) <= 'f') \ |
277 | || ((c) >= 'A' && (c) <= 'F')) | |
96cc36cc RS |
278 | |
279 | /* This is only used for single-byte characters. */ | |
0b32bf0e | 280 | # define ISBLANK(c) ((c) == ' ' || (c) == '\t') |
96cc36cc RS |
281 | |
282 | /* The rest must handle multibyte characters. */ | |
283 | ||
0b32bf0e | 284 | # define ISGRAPH(c) (SINGLE_BYTE_CHAR_P (c) \ |
f71b19b6 | 285 | ? (c) > ' ' && !((c) >= 0177 && (c) <= 0237) \ |
96cc36cc RS |
286 | : 1) |
287 | ||
14473664 | 288 | # define ISPRINT(c) (SINGLE_BYTE_CHAR_P (c) \ |
f71b19b6 | 289 | ? (c) >= ' ' && !((c) >= 0177 && (c) <= 0237) \ |
96cc36cc RS |
290 | : 1) |
291 | ||
0b32bf0e | 292 | # define ISALNUM(c) (IS_REAL_ASCII (c) \ |
f71b19b6 DL |
293 | ? (((c) >= 'a' && (c) <= 'z') \ |
294 | || ((c) >= 'A' && (c) <= 'Z') \ | |
295 | || ((c) >= '0' && (c) <= '9')) \ | |
96cc36cc RS |
296 | : SYNTAX (c) == Sword) |
297 | ||
0b32bf0e | 298 | # define ISALPHA(c) (IS_REAL_ASCII (c) \ |
f71b19b6 DL |
299 | ? (((c) >= 'a' && (c) <= 'z') \ |
300 | || ((c) >= 'A' && (c) <= 'Z')) \ | |
96cc36cc RS |
301 | : SYNTAX (c) == Sword) |
302 | ||
0b32bf0e | 303 | # define ISLOWER(c) (LOWERCASEP (c)) |
96cc36cc | 304 | |
0b32bf0e | 305 | # define ISPUNCT(c) (IS_REAL_ASCII (c) \ |
f71b19b6 DL |
306 | ? ((c) > ' ' && (c) < 0177 \ |
307 | && !(((c) >= 'a' && (c) <= 'z') \ | |
4bb91c68 SM |
308 | || ((c) >= 'A' && (c) <= 'Z') \ |
309 | || ((c) >= '0' && (c) <= '9'))) \ | |
96cc36cc RS |
310 | : SYNTAX (c) != Sword) |
311 | ||
0b32bf0e | 312 | # define ISSPACE(c) (SYNTAX (c) == Swhitespace) |
96cc36cc | 313 | |
0b32bf0e | 314 | # define ISUPPER(c) (UPPERCASEP (c)) |
96cc36cc | 315 | |
0b32bf0e | 316 | # define ISWORD(c) (SYNTAX (c) == Sword) |
96cc36cc RS |
317 | |
318 | #else /* not emacs */ | |
319 | ||
f71b19b6 DL |
320 | /* Jim Meyering writes: |
321 | ||
322 | "... Some ctype macros are valid only for character codes that | |
323 | isascii says are ASCII (SGI's IRIX-4.0.5 is one such system --when | |
324 | using /bin/cc or gcc but without giving an ansi option). So, all | |
4bb91c68 | 325 | ctype uses should be through macros like ISPRINT... If |
f71b19b6 DL |
326 | STDC_HEADERS is defined, then autoconf has verified that the ctype |
327 | macros don't need to be guarded with references to isascii. ... | |
328 | Defining isascii to 1 should let any compiler worth its salt | |
4bb91c68 SM |
329 | eliminate the && through constant folding." |
330 | Solaris defines some of these symbols so we must undefine them first. */ | |
f71b19b6 | 331 | |
4bb91c68 | 332 | # undef ISASCII |
0b32bf0e SM |
333 | # if defined STDC_HEADERS || (!defined isascii && !defined HAVE_ISASCII) |
334 | # define ISASCII(c) 1 | |
335 | # else | |
336 | # define ISASCII(c) isascii(c) | |
337 | # endif | |
f71b19b6 DL |
338 | |
339 | /* 1 if C is an ASCII character. */ | |
0b32bf0e | 340 | # define IS_REAL_ASCII(c) ((c) < 0200) |
f71b19b6 DL |
341 | |
342 | /* This distinction is not meaningful, except in Emacs. */ | |
0b32bf0e SM |
343 | # define ISUNIBYTE(c) 1 |
344 | ||
345 | # ifdef isblank | |
346 | # define ISBLANK(c) (ISASCII (c) && isblank (c)) | |
347 | # else | |
348 | # define ISBLANK(c) ((c) == ' ' || (c) == '\t') | |
349 | # endif | |
350 | # ifdef isgraph | |
351 | # define ISGRAPH(c) (ISASCII (c) && isgraph (c)) | |
352 | # else | |
353 | # define ISGRAPH(c) (ISASCII (c) && isprint (c) && !isspace (c)) | |
354 | # endif | |
355 | ||
4bb91c68 | 356 | # undef ISPRINT |
0b32bf0e SM |
357 | # define ISPRINT(c) (ISASCII (c) && isprint (c)) |
358 | # define ISDIGIT(c) (ISASCII (c) && isdigit (c)) | |
359 | # define ISALNUM(c) (ISASCII (c) && isalnum (c)) | |
360 | # define ISALPHA(c) (ISASCII (c) && isalpha (c)) | |
361 | # define ISCNTRL(c) (ISASCII (c) && iscntrl (c)) | |
362 | # define ISLOWER(c) (ISASCII (c) && islower (c)) | |
363 | # define ISPUNCT(c) (ISASCII (c) && ispunct (c)) | |
364 | # define ISSPACE(c) (ISASCII (c) && isspace (c)) | |
365 | # define ISUPPER(c) (ISASCII (c) && isupper (c)) | |
366 | # define ISXDIGIT(c) (ISASCII (c) && isxdigit (c)) | |
367 | ||
368 | # define ISWORD(c) ISALPHA(c) | |
369 | ||
4bb91c68 SM |
370 | # ifdef _tolower |
371 | # define TOLOWER(c) _tolower(c) | |
372 | # else | |
373 | # define TOLOWER(c) tolower(c) | |
374 | # endif | |
375 | ||
376 | /* How many characters in the character set. */ | |
377 | # define CHAR_SET_SIZE 256 | |
378 | ||
0b32bf0e | 379 | # ifdef SYNTAX_TABLE |
f71b19b6 | 380 | |
0b32bf0e | 381 | extern char *re_syntax_table; |
f71b19b6 | 382 | |
0b32bf0e SM |
383 | # else /* not SYNTAX_TABLE */ |
384 | ||
0b32bf0e SM |
385 | static char re_syntax_table[CHAR_SET_SIZE]; |
386 | ||
387 | static void | |
388 | init_syntax_once () | |
389 | { | |
390 | register int c; | |
391 | static int done = 0; | |
392 | ||
393 | if (done) | |
394 | return; | |
395 | ||
396 | bzero (re_syntax_table, sizeof re_syntax_table); | |
397 | ||
4bb91c68 SM |
398 | for (c = 0; c < CHAR_SET_SIZE; ++c) |
399 | if (ISALNUM (c)) | |
400 | re_syntax_table[c] = Sword; | |
fa9a63c5 | 401 | |
669fa600 | 402 | re_syntax_table['_'] = Ssymbol; |
fa9a63c5 | 403 | |
0b32bf0e SM |
404 | done = 1; |
405 | } | |
406 | ||
407 | # endif /* not SYNTAX_TABLE */ | |
96cc36cc | 408 | |
4bb91c68 SM |
409 | # define SYNTAX(c) re_syntax_table[(c)] |
410 | ||
96cc36cc RS |
411 | #endif /* not emacs */ |
412 | \f | |
fa9a63c5 | 413 | #ifndef NULL |
0b32bf0e | 414 | # define NULL (void *)0 |
fa9a63c5 RM |
415 | #endif |
416 | ||
417 | /* We remove any previous definition of `SIGN_EXTEND_CHAR', | |
418 | since ours (we hope) works properly with all combinations of | |
419 | machines, compilers, `char' and `unsigned char' argument types. | |
4bb91c68 | 420 | (Per Bothner suggested the basic approach.) */ |
fa9a63c5 RM |
421 | #undef SIGN_EXTEND_CHAR |
422 | #if __STDC__ | |
0b32bf0e | 423 | # define SIGN_EXTEND_CHAR(c) ((signed char) (c)) |
fa9a63c5 RM |
424 | #else /* not __STDC__ */ |
425 | /* As in Harbison and Steele. */ | |
0b32bf0e | 426 | # define SIGN_EXTEND_CHAR(c) ((((unsigned char) (c)) ^ 128) - 128) |
fa9a63c5 RM |
427 | #endif |
428 | \f | |
429 | /* Should we use malloc or alloca? If REGEX_MALLOC is not defined, we | |
430 | use `alloca' instead of `malloc'. This is because using malloc in | |
431 | re_search* or re_match* could cause memory leaks when C-g is used in | |
432 | Emacs; also, malloc is slower and causes storage fragmentation. On | |
5e69f11e RM |
433 | the other hand, malloc is more portable, and easier to debug. |
434 | ||
fa9a63c5 RM |
435 | Because we sometimes use alloca, some routines have to be macros, |
436 | not functions -- `alloca'-allocated space disappears at the end of the | |
437 | function it is called in. */ | |
438 | ||
439 | #ifdef REGEX_MALLOC | |
440 | ||
0b32bf0e SM |
441 | # define REGEX_ALLOCATE malloc |
442 | # define REGEX_REALLOCATE(source, osize, nsize) realloc (source, nsize) | |
443 | # define REGEX_FREE free | |
fa9a63c5 RM |
444 | |
445 | #else /* not REGEX_MALLOC */ | |
446 | ||
447 | /* Emacs already defines alloca, sometimes. */ | |
0b32bf0e | 448 | # ifndef alloca |
fa9a63c5 RM |
449 | |
450 | /* Make alloca work the best possible way. */ | |
0b32bf0e SM |
451 | # ifdef __GNUC__ |
452 | # define alloca __builtin_alloca | |
453 | # else /* not __GNUC__ */ | |
454 | # if HAVE_ALLOCA_H | |
455 | # include <alloca.h> | |
456 | # endif /* HAVE_ALLOCA_H */ | |
457 | # endif /* not __GNUC__ */ | |
fa9a63c5 | 458 | |
0b32bf0e | 459 | # endif /* not alloca */ |
fa9a63c5 | 460 | |
0b32bf0e | 461 | # define REGEX_ALLOCATE alloca |
fa9a63c5 RM |
462 | |
463 | /* Assumes a `char *destination' variable. */ | |
0b32bf0e | 464 | # define REGEX_REALLOCATE(source, osize, nsize) \ |
fa9a63c5 | 465 | (destination = (char *) alloca (nsize), \ |
4bb91c68 | 466 | memcpy (destination, source, osize)) |
fa9a63c5 RM |
467 | |
468 | /* No need to do anything to free, after alloca. */ | |
0b32bf0e | 469 | # define REGEX_FREE(arg) ((void)0) /* Do nothing! But inhibit gcc warning. */ |
fa9a63c5 RM |
470 | |
471 | #endif /* not REGEX_MALLOC */ | |
472 | ||
473 | /* Define how to allocate the failure stack. */ | |
474 | ||
0b32bf0e | 475 | #if defined REL_ALLOC && defined REGEX_MALLOC |
4297555e | 476 | |
0b32bf0e | 477 | # define REGEX_ALLOCATE_STACK(size) \ |
fa9a63c5 | 478 | r_alloc (&failure_stack_ptr, (size)) |
0b32bf0e | 479 | # define REGEX_REALLOCATE_STACK(source, osize, nsize) \ |
fa9a63c5 | 480 | r_re_alloc (&failure_stack_ptr, (nsize)) |
0b32bf0e | 481 | # define REGEX_FREE_STACK(ptr) \ |
fa9a63c5 RM |
482 | r_alloc_free (&failure_stack_ptr) |
483 | ||
4297555e | 484 | #else /* not using relocating allocator */ |
fa9a63c5 | 485 | |
0b32bf0e | 486 | # ifdef REGEX_MALLOC |
fa9a63c5 | 487 | |
0b32bf0e SM |
488 | # define REGEX_ALLOCATE_STACK malloc |
489 | # define REGEX_REALLOCATE_STACK(source, osize, nsize) realloc (source, nsize) | |
490 | # define REGEX_FREE_STACK free | |
fa9a63c5 | 491 | |
0b32bf0e | 492 | # else /* not REGEX_MALLOC */ |
fa9a63c5 | 493 | |
0b32bf0e | 494 | # define REGEX_ALLOCATE_STACK alloca |
fa9a63c5 | 495 | |
0b32bf0e | 496 | # define REGEX_REALLOCATE_STACK(source, osize, nsize) \ |
fa9a63c5 | 497 | REGEX_REALLOCATE (source, osize, nsize) |
25fe55af | 498 | /* No need to explicitly free anything. */ |
0b32bf0e | 499 | # define REGEX_FREE_STACK(arg) ((void)0) |
fa9a63c5 | 500 | |
0b32bf0e | 501 | # endif /* not REGEX_MALLOC */ |
4297555e | 502 | #endif /* not using relocating allocator */ |
fa9a63c5 RM |
503 | |
504 | ||
505 | /* True if `size1' is non-NULL and PTR is pointing anywhere inside | |
506 | `string1' or just past its end. This works if PTR is NULL, which is | |
507 | a good thing. */ | |
25fe55af | 508 | #define FIRST_STRING_P(ptr) \ |
fa9a63c5 RM |
509 | (size1 && string1 <= (ptr) && (ptr) <= string1 + size1) |
510 | ||
511 | /* (Re)Allocate N items of type T using malloc, or fail. */ | |
512 | #define TALLOC(n, t) ((t *) malloc ((n) * sizeof (t))) | |
513 | #define RETALLOC(addr, n, t) ((addr) = (t *) realloc (addr, (n) * sizeof (t))) | |
514 | #define RETALLOC_IF(addr, n, t) \ | |
515 | if (addr) RETALLOC((addr), (n), t); else (addr) = TALLOC ((n), t) | |
516 | #define REGEX_TALLOC(n, t) ((t *) REGEX_ALLOCATE ((n) * sizeof (t))) | |
517 | ||
4bb91c68 | 518 | #define BYTEWIDTH 8 /* In bits. */ |
fa9a63c5 RM |
519 | |
520 | #define STREQ(s1, s2) ((strcmp (s1, s2) == 0)) | |
521 | ||
522 | #undef MAX | |
523 | #undef MIN | |
524 | #define MAX(a, b) ((a) > (b) ? (a) : (b)) | |
525 | #define MIN(a, b) ((a) < (b) ? (a) : (b)) | |
526 | ||
66f0296e SM |
527 | /* Type of source-pattern and string chars. */ |
528 | typedef const unsigned char re_char; | |
529 | ||
fa9a63c5 RM |
530 | typedef char boolean; |
531 | #define false 0 | |
532 | #define true 1 | |
533 | ||
4bb91c68 SM |
534 | static int re_match_2_internal _RE_ARGS ((struct re_pattern_buffer *bufp, |
535 | re_char *string1, int size1, | |
536 | re_char *string2, int size2, | |
537 | int pos, | |
538 | struct re_registers *regs, | |
539 | int stop)); | |
fa9a63c5 RM |
540 | \f |
541 | /* These are the command codes that appear in compiled regular | |
4bb91c68 | 542 | expressions. Some opcodes are followed by argument bytes. A |
fa9a63c5 RM |
543 | command code can specify any interpretation whatsoever for its |
544 | arguments. Zero bytes may appear in the compiled regular expression. */ | |
545 | ||
546 | typedef enum | |
547 | { | |
548 | no_op = 0, | |
549 | ||
4bb91c68 | 550 | /* Succeed right away--no more backtracking. */ |
fa9a63c5 RM |
551 | succeed, |
552 | ||
25fe55af | 553 | /* Followed by one byte giving n, then by n literal bytes. */ |
fa9a63c5 RM |
554 | exactn, |
555 | ||
25fe55af | 556 | /* Matches any (more or less) character. */ |
fa9a63c5 RM |
557 | anychar, |
558 | ||
25fe55af RS |
559 | /* Matches any one char belonging to specified set. First |
560 | following byte is number of bitmap bytes. Then come bytes | |
561 | for a bitmap saying which chars are in. Bits in each byte | |
562 | are ordered low-bit-first. A character is in the set if its | |
563 | bit is 1. A character too large to have a bit in the map is | |
96cc36cc RS |
564 | automatically not in the set. |
565 | ||
566 | If the length byte has the 0x80 bit set, then that stuff | |
567 | is followed by a range table: | |
568 | 2 bytes of flags for character sets (low 8 bits, high 8 bits) | |
0b32bf0e | 569 | See RANGE_TABLE_WORK_BITS below. |
01618498 | 570 | 2 bytes, the number of pairs that follow (upto 32767) |
96cc36cc | 571 | pairs, each 2 multibyte characters, |
0b32bf0e | 572 | each multibyte character represented as 3 bytes. */ |
fa9a63c5 RM |
573 | charset, |
574 | ||
25fe55af | 575 | /* Same parameters as charset, but match any character that is |
4bb91c68 | 576 | not one of those specified. */ |
fa9a63c5 RM |
577 | charset_not, |
578 | ||
25fe55af RS |
579 | /* Start remembering the text that is matched, for storing in a |
580 | register. Followed by one byte with the register number, in | |
581 | the range 0 to one less than the pattern buffer's re_nsub | |
505bde11 | 582 | field. */ |
fa9a63c5 RM |
583 | start_memory, |
584 | ||
25fe55af RS |
585 | /* Stop remembering the text that is matched and store it in a |
586 | memory register. Followed by one byte with the register | |
587 | number, in the range 0 to one less than `re_nsub' in the | |
505bde11 | 588 | pattern buffer. */ |
fa9a63c5 RM |
589 | stop_memory, |
590 | ||
25fe55af | 591 | /* Match a duplicate of something remembered. Followed by one |
4bb91c68 | 592 | byte containing the register number. */ |
fa9a63c5 RM |
593 | duplicate, |
594 | ||
25fe55af | 595 | /* Fail unless at beginning of line. */ |
fa9a63c5 RM |
596 | begline, |
597 | ||
4bb91c68 | 598 | /* Fail unless at end of line. */ |
fa9a63c5 RM |
599 | endline, |
600 | ||
25fe55af RS |
601 | /* Succeeds if at beginning of buffer (if emacs) or at beginning |
602 | of string to be matched (if not). */ | |
fa9a63c5 RM |
603 | begbuf, |
604 | ||
25fe55af | 605 | /* Analogously, for end of buffer/string. */ |
fa9a63c5 | 606 | endbuf, |
5e69f11e | 607 | |
25fe55af | 608 | /* Followed by two byte relative address to which to jump. */ |
5e69f11e | 609 | jump, |
fa9a63c5 | 610 | |
25fe55af RS |
611 | /* Followed by two-byte relative address of place to resume at |
612 | in case of failure. */ | |
fa9a63c5 | 613 | on_failure_jump, |
5e69f11e | 614 | |
25fe55af RS |
615 | /* Like on_failure_jump, but pushes a placeholder instead of the |
616 | current string position when executed. */ | |
fa9a63c5 | 617 | on_failure_keep_string_jump, |
5e69f11e | 618 | |
505bde11 SM |
619 | /* Just like `on_failure_jump', except that it checks that we |
620 | don't get stuck in an infinite loop (matching an empty string | |
621 | indefinitely). */ | |
622 | on_failure_jump_loop, | |
623 | ||
0683b6fa SM |
624 | /* Just like `on_failure_jump_loop', except that it checks for |
625 | a different kind of loop (the kind that shows up with non-greedy | |
626 | operators). This operation has to be immediately preceded | |
627 | by a `no_op'. */ | |
628 | on_failure_jump_nastyloop, | |
629 | ||
0b32bf0e | 630 | /* A smart `on_failure_jump' used for greedy * and + operators. |
505bde11 SM |
631 | It analyses the loop before which it is put and if the |
632 | loop does not require backtracking, it changes itself to | |
4e8a9132 SM |
633 | `on_failure_keep_string_jump' and short-circuits the loop, |
634 | else it just defaults to changing itself into `on_failure_jump'. | |
635 | It assumes that it is pointing to just past a `jump'. */ | |
505bde11 | 636 | on_failure_jump_smart, |
fa9a63c5 | 637 | |
25fe55af | 638 | /* Followed by two-byte relative address and two-byte number n. |
ed0767d8 SM |
639 | After matching N times, jump to the address upon failure. |
640 | Does not work if N starts at 0: use on_failure_jump_loop | |
641 | instead. */ | |
fa9a63c5 RM |
642 | succeed_n, |
643 | ||
25fe55af RS |
644 | /* Followed by two-byte relative address, and two-byte number n. |
645 | Jump to the address N times, then fail. */ | |
fa9a63c5 RM |
646 | jump_n, |
647 | ||
25fe55af RS |
648 | /* Set the following two-byte relative address to the |
649 | subsequent two-byte number. The address *includes* the two | |
650 | bytes of number. */ | |
fa9a63c5 RM |
651 | set_number_at, |
652 | ||
fa9a63c5 RM |
653 | wordbeg, /* Succeeds if at word beginning. */ |
654 | wordend, /* Succeeds if at word end. */ | |
655 | ||
656 | wordbound, /* Succeeds if at a word boundary. */ | |
1fb352e0 | 657 | notwordbound, /* Succeeds if not at a word boundary. */ |
fa9a63c5 | 658 | |
669fa600 SM |
659 | symbeg, /* Succeeds if at symbol beginning. */ |
660 | symend, /* Succeeds if at symbol end. */ | |
661 | ||
fa9a63c5 | 662 | /* Matches any character whose syntax is specified. Followed by |
25fe55af | 663 | a byte which contains a syntax code, e.g., Sword. */ |
fa9a63c5 RM |
664 | syntaxspec, |
665 | ||
666 | /* Matches any character whose syntax is not that specified. */ | |
1fb352e0 SM |
667 | notsyntaxspec |
668 | ||
669 | #ifdef emacs | |
670 | ,before_dot, /* Succeeds if before point. */ | |
671 | at_dot, /* Succeeds if at point. */ | |
672 | after_dot, /* Succeeds if after point. */ | |
b18215fc RS |
673 | |
674 | /* Matches any character whose category-set contains the specified | |
25fe55af RS |
675 | category. The operator is followed by a byte which contains a |
676 | category code (mnemonic ASCII character). */ | |
b18215fc RS |
677 | categoryspec, |
678 | ||
679 | /* Matches any character whose category-set does not contain the | |
680 | specified category. The operator is followed by a byte which | |
681 | contains the category code (mnemonic ASCII character). */ | |
682 | notcategoryspec | |
fa9a63c5 RM |
683 | #endif /* emacs */ |
684 | } re_opcode_t; | |
685 | \f | |
686 | /* Common operations on the compiled pattern. */ | |
687 | ||
688 | /* Store NUMBER in two contiguous bytes starting at DESTINATION. */ | |
689 | ||
690 | #define STORE_NUMBER(destination, number) \ | |
691 | do { \ | |
692 | (destination)[0] = (number) & 0377; \ | |
693 | (destination)[1] = (number) >> 8; \ | |
694 | } while (0) | |
695 | ||
696 | /* Same as STORE_NUMBER, except increment DESTINATION to | |
697 | the byte after where the number is stored. Therefore, DESTINATION | |
698 | must be an lvalue. */ | |
699 | ||
700 | #define STORE_NUMBER_AND_INCR(destination, number) \ | |
701 | do { \ | |
702 | STORE_NUMBER (destination, number); \ | |
703 | (destination) += 2; \ | |
704 | } while (0) | |
705 | ||
706 | /* Put into DESTINATION a number stored in two contiguous bytes starting | |
707 | at SOURCE. */ | |
708 | ||
709 | #define EXTRACT_NUMBER(destination, source) \ | |
710 | do { \ | |
711 | (destination) = *(source) & 0377; \ | |
712 | (destination) += SIGN_EXTEND_CHAR (*((source) + 1)) << 8; \ | |
713 | } while (0) | |
714 | ||
715 | #ifdef DEBUG | |
4bb91c68 | 716 | static void extract_number _RE_ARGS ((int *dest, re_char *source)); |
fa9a63c5 RM |
717 | static void |
718 | extract_number (dest, source) | |
719 | int *dest; | |
01618498 | 720 | re_char *source; |
fa9a63c5 | 721 | { |
5e69f11e | 722 | int temp = SIGN_EXTEND_CHAR (*(source + 1)); |
fa9a63c5 RM |
723 | *dest = *source & 0377; |
724 | *dest += temp << 8; | |
725 | } | |
726 | ||
4bb91c68 | 727 | # ifndef EXTRACT_MACROS /* To debug the macros. */ |
0b32bf0e SM |
728 | # undef EXTRACT_NUMBER |
729 | # define EXTRACT_NUMBER(dest, src) extract_number (&dest, src) | |
730 | # endif /* not EXTRACT_MACROS */ | |
fa9a63c5 RM |
731 | |
732 | #endif /* DEBUG */ | |
733 | ||
734 | /* Same as EXTRACT_NUMBER, except increment SOURCE to after the number. | |
735 | SOURCE must be an lvalue. */ | |
736 | ||
737 | #define EXTRACT_NUMBER_AND_INCR(destination, source) \ | |
738 | do { \ | |
739 | EXTRACT_NUMBER (destination, source); \ | |
25fe55af | 740 | (source) += 2; \ |
fa9a63c5 RM |
741 | } while (0) |
742 | ||
743 | #ifdef DEBUG | |
4bb91c68 SM |
744 | static void extract_number_and_incr _RE_ARGS ((int *destination, |
745 | re_char **source)); | |
fa9a63c5 RM |
746 | static void |
747 | extract_number_and_incr (destination, source) | |
748 | int *destination; | |
01618498 | 749 | re_char **source; |
5e69f11e | 750 | { |
fa9a63c5 RM |
751 | extract_number (destination, *source); |
752 | *source += 2; | |
753 | } | |
754 | ||
0b32bf0e SM |
755 | # ifndef EXTRACT_MACROS |
756 | # undef EXTRACT_NUMBER_AND_INCR | |
757 | # define EXTRACT_NUMBER_AND_INCR(dest, src) \ | |
fa9a63c5 | 758 | extract_number_and_incr (&dest, &src) |
0b32bf0e | 759 | # endif /* not EXTRACT_MACROS */ |
fa9a63c5 RM |
760 | |
761 | #endif /* DEBUG */ | |
762 | \f | |
b18215fc RS |
763 | /* Store a multibyte character in three contiguous bytes starting |
764 | DESTINATION, and increment DESTINATION to the byte after where the | |
25fe55af | 765 | character is stored. Therefore, DESTINATION must be an lvalue. */ |
b18215fc RS |
766 | |
767 | #define STORE_CHARACTER_AND_INCR(destination, character) \ | |
768 | do { \ | |
769 | (destination)[0] = (character) & 0377; \ | |
770 | (destination)[1] = ((character) >> 8) & 0377; \ | |
771 | (destination)[2] = (character) >> 16; \ | |
772 | (destination) += 3; \ | |
773 | } while (0) | |
774 | ||
775 | /* Put into DESTINATION a character stored in three contiguous bytes | |
25fe55af | 776 | starting at SOURCE. */ |
b18215fc RS |
777 | |
778 | #define EXTRACT_CHARACTER(destination, source) \ | |
779 | do { \ | |
780 | (destination) = ((source)[0] \ | |
781 | | ((source)[1] << 8) \ | |
782 | | ((source)[2] << 16)); \ | |
783 | } while (0) | |
784 | ||
785 | ||
786 | /* Macros for charset. */ | |
787 | ||
788 | /* Size of bitmap of charset P in bytes. P is a start of charset, | |
789 | i.e. *P is (re_opcode_t) charset or (re_opcode_t) charset_not. */ | |
790 | #define CHARSET_BITMAP_SIZE(p) ((p)[1] & 0x7F) | |
791 | ||
792 | /* Nonzero if charset P has range table. */ | |
25fe55af | 793 | #define CHARSET_RANGE_TABLE_EXISTS_P(p) ((p)[1] & 0x80) |
b18215fc RS |
794 | |
795 | /* Return the address of range table of charset P. But not the start | |
796 | of table itself, but the before where the number of ranges is | |
96cc36cc RS |
797 | stored. `2 +' means to skip re_opcode_t and size of bitmap, |
798 | and the 2 bytes of flags at the start of the range table. */ | |
799 | #define CHARSET_RANGE_TABLE(p) (&(p)[4 + CHARSET_BITMAP_SIZE (p)]) | |
800 | ||
801 | /* Extract the bit flags that start a range table. */ | |
802 | #define CHARSET_RANGE_TABLE_BITS(p) \ | |
803 | ((p)[2 + CHARSET_BITMAP_SIZE (p)] \ | |
804 | + (p)[3 + CHARSET_BITMAP_SIZE (p)] * 0x100) | |
b18215fc RS |
805 | |
806 | /* Test if C is listed in the bitmap of charset P. */ | |
807 | #define CHARSET_LOOKUP_BITMAP(p, c) \ | |
808 | ((c) < CHARSET_BITMAP_SIZE (p) * BYTEWIDTH \ | |
809 | && (p)[2 + (c) / BYTEWIDTH] & (1 << ((c) % BYTEWIDTH))) | |
810 | ||
811 | /* Return the address of end of RANGE_TABLE. COUNT is number of | |
25fe55af RS |
812 | ranges (which is a pair of (start, end)) in the RANGE_TABLE. `* 2' |
813 | is start of range and end of range. `* 3' is size of each start | |
b18215fc RS |
814 | and end. */ |
815 | #define CHARSET_RANGE_TABLE_END(range_table, count) \ | |
816 | ((range_table) + (count) * 2 * 3) | |
817 | ||
25fe55af | 818 | /* Test if C is in RANGE_TABLE. A flag NOT is negated if C is in. |
b18215fc RS |
819 | COUNT is number of ranges in RANGE_TABLE. */ |
820 | #define CHARSET_LOOKUP_RANGE_TABLE_RAW(not, c, range_table, count) \ | |
821 | do \ | |
822 | { \ | |
01618498 SM |
823 | re_wchar_t range_start, range_end; \ |
824 | re_char *p; \ | |
825 | re_char *range_table_end \ | |
b18215fc RS |
826 | = CHARSET_RANGE_TABLE_END ((range_table), (count)); \ |
827 | \ | |
828 | for (p = (range_table); p < range_table_end; p += 2 * 3) \ | |
829 | { \ | |
830 | EXTRACT_CHARACTER (range_start, p); \ | |
831 | EXTRACT_CHARACTER (range_end, p + 3); \ | |
832 | \ | |
833 | if (range_start <= (c) && (c) <= range_end) \ | |
834 | { \ | |
835 | (not) = !(not); \ | |
836 | break; \ | |
837 | } \ | |
838 | } \ | |
839 | } \ | |
840 | while (0) | |
841 | ||
842 | /* Test if C is in range table of CHARSET. The flag NOT is negated if | |
843 | C is listed in it. */ | |
844 | #define CHARSET_LOOKUP_RANGE_TABLE(not, c, charset) \ | |
845 | do \ | |
846 | { \ | |
847 | /* Number of ranges in range table. */ \ | |
848 | int count; \ | |
01618498 SM |
849 | re_char *range_table = CHARSET_RANGE_TABLE (charset); \ |
850 | \ | |
b18215fc RS |
851 | EXTRACT_NUMBER_AND_INCR (count, range_table); \ |
852 | CHARSET_LOOKUP_RANGE_TABLE_RAW ((not), (c), range_table, count); \ | |
853 | } \ | |
854 | while (0) | |
855 | \f | |
fa9a63c5 RM |
856 | /* If DEBUG is defined, Regex prints many voluminous messages about what |
857 | it is doing (if the variable `debug' is nonzero). If linked with the | |
858 | main program in `iregex.c', you can enter patterns and strings | |
859 | interactively. And if linked with the main program in `main.c' and | |
4bb91c68 | 860 | the other test files, you can run the already-written tests. */ |
fa9a63c5 RM |
861 | |
862 | #ifdef DEBUG | |
863 | ||
864 | /* We use standard I/O for debugging. */ | |
0b32bf0e | 865 | # include <stdio.h> |
fa9a63c5 RM |
866 | |
867 | /* It is useful to test things that ``must'' be true when debugging. */ | |
0b32bf0e | 868 | # include <assert.h> |
fa9a63c5 | 869 | |
99633e97 | 870 | static int debug = -100000; |
fa9a63c5 | 871 | |
0b32bf0e SM |
872 | # define DEBUG_STATEMENT(e) e |
873 | # define DEBUG_PRINT1(x) if (debug > 0) printf (x) | |
874 | # define DEBUG_PRINT2(x1, x2) if (debug > 0) printf (x1, x2) | |
875 | # define DEBUG_PRINT3(x1, x2, x3) if (debug > 0) printf (x1, x2, x3) | |
876 | # define DEBUG_PRINT4(x1, x2, x3, x4) if (debug > 0) printf (x1, x2, x3, x4) | |
877 | # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) \ | |
99633e97 | 878 | if (debug > 0) print_partial_compiled_pattern (s, e) |
0b32bf0e | 879 | # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) \ |
99633e97 | 880 | if (debug > 0) print_double_string (w, s1, sz1, s2, sz2) |
fa9a63c5 RM |
881 | |
882 | ||
883 | /* Print the fastmap in human-readable form. */ | |
884 | ||
885 | void | |
886 | print_fastmap (fastmap) | |
887 | char *fastmap; | |
888 | { | |
889 | unsigned was_a_range = 0; | |
5e69f11e RM |
890 | unsigned i = 0; |
891 | ||
fa9a63c5 RM |
892 | while (i < (1 << BYTEWIDTH)) |
893 | { | |
894 | if (fastmap[i++]) | |
895 | { | |
896 | was_a_range = 0; | |
25fe55af RS |
897 | putchar (i - 1); |
898 | while (i < (1 << BYTEWIDTH) && fastmap[i]) | |
899 | { | |
900 | was_a_range = 1; | |
901 | i++; | |
902 | } | |
fa9a63c5 | 903 | if (was_a_range) |
25fe55af RS |
904 | { |
905 | printf ("-"); | |
906 | putchar (i - 1); | |
907 | } | |
908 | } | |
fa9a63c5 | 909 | } |
5e69f11e | 910 | putchar ('\n'); |
fa9a63c5 RM |
911 | } |
912 | ||
913 | ||
914 | /* Print a compiled pattern string in human-readable form, starting at | |
915 | the START pointer into it and ending just before the pointer END. */ | |
916 | ||
917 | void | |
918 | print_partial_compiled_pattern (start, end) | |
01618498 SM |
919 | re_char *start; |
920 | re_char *end; | |
fa9a63c5 RM |
921 | { |
922 | int mcnt, mcnt2; | |
01618498 SM |
923 | re_char *p = start; |
924 | re_char *pend = end; | |
fa9a63c5 RM |
925 | |
926 | if (start == NULL) | |
927 | { | |
a1a052df | 928 | fprintf (stderr, "(null)\n"); |
fa9a63c5 RM |
929 | return; |
930 | } | |
5e69f11e | 931 | |
fa9a63c5 RM |
932 | /* Loop over pattern commands. */ |
933 | while (p < pend) | |
934 | { | |
a1a052df | 935 | fprintf (stderr, "%d:\t", p - start); |
fa9a63c5 RM |
936 | |
937 | switch ((re_opcode_t) *p++) | |
938 | { | |
25fe55af | 939 | case no_op: |
a1a052df | 940 | fprintf (stderr, "/no_op"); |
25fe55af | 941 | break; |
fa9a63c5 | 942 | |
99633e97 | 943 | case succeed: |
a1a052df | 944 | fprintf (stderr, "/succeed"); |
99633e97 SM |
945 | break; |
946 | ||
fa9a63c5 RM |
947 | case exactn: |
948 | mcnt = *p++; | |
a1a052df | 949 | fprintf (stderr, "/exactn/%d", mcnt); |
25fe55af | 950 | do |
fa9a63c5 | 951 | { |
a1a052df | 952 | fprintf (stderr, "/%c", *p++); |
25fe55af RS |
953 | } |
954 | while (--mcnt); | |
955 | break; | |
fa9a63c5 RM |
956 | |
957 | case start_memory: | |
a1a052df | 958 | fprintf (stderr, "/start_memory/%d", *p++); |
25fe55af | 959 | break; |
fa9a63c5 RM |
960 | |
961 | case stop_memory: | |
a1a052df | 962 | fprintf (stderr, "/stop_memory/%d", *p++); |
25fe55af | 963 | break; |
fa9a63c5 RM |
964 | |
965 | case duplicate: | |
a1a052df | 966 | fprintf (stderr, "/duplicate/%d", *p++); |
fa9a63c5 RM |
967 | break; |
968 | ||
969 | case anychar: | |
a1a052df | 970 | fprintf (stderr, "/anychar"); |
fa9a63c5 RM |
971 | break; |
972 | ||
973 | case charset: | |
25fe55af RS |
974 | case charset_not: |
975 | { | |
976 | register int c, last = -100; | |
fa9a63c5 | 977 | register int in_range = 0; |
99633e97 SM |
978 | int length = CHARSET_BITMAP_SIZE (p - 1); |
979 | int has_range_table = CHARSET_RANGE_TABLE_EXISTS_P (p - 1); | |
fa9a63c5 | 980 | |
a1a052df | 981 | fprintf (stderr, "/charset [%s", |
275a3409 | 982 | (re_opcode_t) *(p - 1) == charset_not ? "^" : ""); |
5e69f11e | 983 | |
275a3409 RS |
984 | if (p + *p >= pend) |
985 | fprintf (stderr, " !extends past end of pattern! "); | |
fa9a63c5 | 986 | |
25fe55af | 987 | for (c = 0; c < 256; c++) |
96cc36cc | 988 | if (c / 8 < length |
fa9a63c5 RM |
989 | && (p[1 + (c/8)] & (1 << (c % 8)))) |
990 | { | |
991 | /* Are we starting a range? */ | |
992 | if (last + 1 == c && ! in_range) | |
993 | { | |
a1a052df | 994 | fprintf (stderr, "-"); |
fa9a63c5 RM |
995 | in_range = 1; |
996 | } | |
997 | /* Have we broken a range? */ | |
998 | else if (last + 1 != c && in_range) | |
96cc36cc | 999 | { |
a1a052df | 1000 | fprintf (stderr, "%c", last); |
fa9a63c5 RM |
1001 | in_range = 0; |
1002 | } | |
5e69f11e | 1003 | |
fa9a63c5 | 1004 | if (! in_range) |
a1a052df | 1005 | fprintf (stderr, "%c", c); |
fa9a63c5 RM |
1006 | |
1007 | last = c; | |
25fe55af | 1008 | } |
fa9a63c5 RM |
1009 | |
1010 | if (in_range) | |
a1a052df | 1011 | fprintf (stderr, "%c", last); |
fa9a63c5 | 1012 | |
a1a052df | 1013 | fprintf (stderr, "]"); |
fa9a63c5 | 1014 | |
99633e97 | 1015 | p += 1 + length; |
96cc36cc | 1016 | |
96cc36cc | 1017 | if (has_range_table) |
99633e97 SM |
1018 | { |
1019 | int count; | |
a1a052df | 1020 | fprintf (stderr, "has-range-table"); |
99633e97 SM |
1021 | |
1022 | /* ??? Should print the range table; for now, just skip it. */ | |
1023 | p += 2; /* skip range table bits */ | |
1024 | EXTRACT_NUMBER_AND_INCR (count, p); | |
1025 | p = CHARSET_RANGE_TABLE_END (p, count); | |
1026 | } | |
fa9a63c5 RM |
1027 | } |
1028 | break; | |
1029 | ||
1030 | case begline: | |
a1a052df | 1031 | fprintf (stderr, "/begline"); |
25fe55af | 1032 | break; |
fa9a63c5 RM |
1033 | |
1034 | case endline: | |
a1a052df | 1035 | fprintf (stderr, "/endline"); |
25fe55af | 1036 | break; |
fa9a63c5 RM |
1037 | |
1038 | case on_failure_jump: | |
25fe55af | 1039 | extract_number_and_incr (&mcnt, &p); |
a1a052df | 1040 | fprintf (stderr, "/on_failure_jump to %d", p + mcnt - start); |
25fe55af | 1041 | break; |
fa9a63c5 RM |
1042 | |
1043 | case on_failure_keep_string_jump: | |
25fe55af | 1044 | extract_number_and_incr (&mcnt, &p); |
a1a052df | 1045 | fprintf (stderr, "/on_failure_keep_string_jump to %d", p + mcnt - start); |
25fe55af | 1046 | break; |
fa9a63c5 | 1047 | |
0683b6fa SM |
1048 | case on_failure_jump_nastyloop: |
1049 | extract_number_and_incr (&mcnt, &p); | |
a1a052df | 1050 | fprintf (stderr, "/on_failure_jump_nastyloop to %d", p + mcnt - start); |
0683b6fa SM |
1051 | break; |
1052 | ||
505bde11 | 1053 | case on_failure_jump_loop: |
fa9a63c5 | 1054 | extract_number_and_incr (&mcnt, &p); |
a1a052df | 1055 | fprintf (stderr, "/on_failure_jump_loop to %d", p + mcnt - start); |
5e69f11e RM |
1056 | break; |
1057 | ||
505bde11 | 1058 | case on_failure_jump_smart: |
fa9a63c5 | 1059 | extract_number_and_incr (&mcnt, &p); |
a1a052df | 1060 | fprintf (stderr, "/on_failure_jump_smart to %d", p + mcnt - start); |
5e69f11e RM |
1061 | break; |
1062 | ||
25fe55af | 1063 | case jump: |
fa9a63c5 | 1064 | extract_number_and_incr (&mcnt, &p); |
a1a052df | 1065 | fprintf (stderr, "/jump to %d", p + mcnt - start); |
fa9a63c5 RM |
1066 | break; |
1067 | ||
25fe55af RS |
1068 | case succeed_n: |
1069 | extract_number_and_incr (&mcnt, &p); | |
1070 | extract_number_and_incr (&mcnt2, &p); | |
a1a052df | 1071 | fprintf (stderr, "/succeed_n to %d, %d times", p - 2 + mcnt - start, mcnt2); |
25fe55af | 1072 | break; |
5e69f11e | 1073 | |
25fe55af RS |
1074 | case jump_n: |
1075 | extract_number_and_incr (&mcnt, &p); | |
1076 | extract_number_and_incr (&mcnt2, &p); | |
a1a052df | 1077 | fprintf (stderr, "/jump_n to %d, %d times", p - 2 + mcnt - start, mcnt2); |
25fe55af | 1078 | break; |
5e69f11e | 1079 | |
25fe55af RS |
1080 | case set_number_at: |
1081 | extract_number_and_incr (&mcnt, &p); | |
1082 | extract_number_and_incr (&mcnt2, &p); | |
a1a052df | 1083 | fprintf (stderr, "/set_number_at location %d to %d", p - 2 + mcnt - start, mcnt2); |
25fe55af | 1084 | break; |
5e69f11e | 1085 | |
25fe55af | 1086 | case wordbound: |
a1a052df | 1087 | fprintf (stderr, "/wordbound"); |
fa9a63c5 RM |
1088 | break; |
1089 | ||
1090 | case notwordbound: | |
a1a052df | 1091 | fprintf (stderr, "/notwordbound"); |
25fe55af | 1092 | break; |
fa9a63c5 RM |
1093 | |
1094 | case wordbeg: | |
a1a052df | 1095 | fprintf (stderr, "/wordbeg"); |
fa9a63c5 | 1096 | break; |
5e69f11e | 1097 | |
fa9a63c5 | 1098 | case wordend: |
a1a052df | 1099 | fprintf (stderr, "/wordend"); |
e2543b02 | 1100 | break; |
5e69f11e | 1101 | |
669fa600 | 1102 | case symbeg: |
e2543b02 | 1103 | fprintf (stderr, "/symbeg"); |
669fa600 SM |
1104 | break; |
1105 | ||
1106 | case symend: | |
e2543b02 | 1107 | fprintf (stderr, "/symend"); |
669fa600 SM |
1108 | break; |
1109 | ||
1fb352e0 | 1110 | case syntaxspec: |
a1a052df | 1111 | fprintf (stderr, "/syntaxspec"); |
1fb352e0 | 1112 | mcnt = *p++; |
a1a052df | 1113 | fprintf (stderr, "/%d", mcnt); |
1fb352e0 SM |
1114 | break; |
1115 | ||
1116 | case notsyntaxspec: | |
a1a052df | 1117 | fprintf (stderr, "/notsyntaxspec"); |
1fb352e0 | 1118 | mcnt = *p++; |
a1a052df | 1119 | fprintf (stderr, "/%d", mcnt); |
1fb352e0 SM |
1120 | break; |
1121 | ||
0b32bf0e | 1122 | # ifdef emacs |
fa9a63c5 | 1123 | case before_dot: |
a1a052df | 1124 | fprintf (stderr, "/before_dot"); |
25fe55af | 1125 | break; |
fa9a63c5 RM |
1126 | |
1127 | case at_dot: | |
a1a052df | 1128 | fprintf (stderr, "/at_dot"); |
25fe55af | 1129 | break; |
fa9a63c5 RM |
1130 | |
1131 | case after_dot: | |
a1a052df | 1132 | fprintf (stderr, "/after_dot"); |
25fe55af | 1133 | break; |
fa9a63c5 | 1134 | |
1fb352e0 | 1135 | case categoryspec: |
a1a052df | 1136 | fprintf (stderr, "/categoryspec"); |
fa9a63c5 | 1137 | mcnt = *p++; |
a1a052df | 1138 | fprintf (stderr, "/%d", mcnt); |
25fe55af | 1139 | break; |
5e69f11e | 1140 | |
1fb352e0 | 1141 | case notcategoryspec: |
a1a052df | 1142 | fprintf (stderr, "/notcategoryspec"); |
fa9a63c5 | 1143 | mcnt = *p++; |
a1a052df | 1144 | fprintf (stderr, "/%d", mcnt); |
fa9a63c5 | 1145 | break; |
0b32bf0e | 1146 | # endif /* emacs */ |
fa9a63c5 | 1147 | |
fa9a63c5 | 1148 | case begbuf: |
a1a052df | 1149 | fprintf (stderr, "/begbuf"); |
25fe55af | 1150 | break; |
fa9a63c5 RM |
1151 | |
1152 | case endbuf: | |
a1a052df | 1153 | fprintf (stderr, "/endbuf"); |
25fe55af | 1154 | break; |
fa9a63c5 | 1155 | |
25fe55af | 1156 | default: |
a1a052df | 1157 | fprintf (stderr, "?%d", *(p-1)); |
fa9a63c5 RM |
1158 | } |
1159 | ||
a1a052df | 1160 | fprintf (stderr, "\n"); |
fa9a63c5 RM |
1161 | } |
1162 | ||
a1a052df | 1163 | fprintf (stderr, "%d:\tend of pattern.\n", p - start); |
fa9a63c5 RM |
1164 | } |
1165 | ||
1166 | ||
1167 | void | |
1168 | print_compiled_pattern (bufp) | |
1169 | struct re_pattern_buffer *bufp; | |
1170 | { | |
01618498 | 1171 | re_char *buffer = bufp->buffer; |
fa9a63c5 RM |
1172 | |
1173 | print_partial_compiled_pattern (buffer, buffer + bufp->used); | |
4bb91c68 SM |
1174 | printf ("%ld bytes used/%ld bytes allocated.\n", |
1175 | bufp->used, bufp->allocated); | |
fa9a63c5 RM |
1176 | |
1177 | if (bufp->fastmap_accurate && bufp->fastmap) | |
1178 | { | |
1179 | printf ("fastmap: "); | |
1180 | print_fastmap (bufp->fastmap); | |
1181 | } | |
1182 | ||
1183 | printf ("re_nsub: %d\t", bufp->re_nsub); | |
1184 | printf ("regs_alloc: %d\t", bufp->regs_allocated); | |
1185 | printf ("can_be_null: %d\t", bufp->can_be_null); | |
fa9a63c5 RM |
1186 | printf ("no_sub: %d\t", bufp->no_sub); |
1187 | printf ("not_bol: %d\t", bufp->not_bol); | |
1188 | printf ("not_eol: %d\t", bufp->not_eol); | |
4bb91c68 | 1189 | printf ("syntax: %lx\n", bufp->syntax); |
505bde11 | 1190 | fflush (stdout); |
fa9a63c5 RM |
1191 | /* Perhaps we should print the translate table? */ |
1192 | } | |
1193 | ||
1194 | ||
1195 | void | |
1196 | print_double_string (where, string1, size1, string2, size2) | |
66f0296e SM |
1197 | re_char *where; |
1198 | re_char *string1; | |
1199 | re_char *string2; | |
fa9a63c5 RM |
1200 | int size1; |
1201 | int size2; | |
1202 | { | |
4bb91c68 | 1203 | int this_char; |
5e69f11e | 1204 | |
fa9a63c5 RM |
1205 | if (where == NULL) |
1206 | printf ("(null)"); | |
1207 | else | |
1208 | { | |
1209 | if (FIRST_STRING_P (where)) | |
25fe55af RS |
1210 | { |
1211 | for (this_char = where - string1; this_char < size1; this_char++) | |
1212 | putchar (string1[this_char]); | |
fa9a63c5 | 1213 | |
25fe55af RS |
1214 | where = string2; |
1215 | } | |
fa9a63c5 RM |
1216 | |
1217 | for (this_char = where - string2; this_char < size2; this_char++) | |
25fe55af | 1218 | putchar (string2[this_char]); |
fa9a63c5 RM |
1219 | } |
1220 | } | |
1221 | ||
1222 | #else /* not DEBUG */ | |
1223 | ||
0b32bf0e SM |
1224 | # undef assert |
1225 | # define assert(e) | |
fa9a63c5 | 1226 | |
0b32bf0e SM |
1227 | # define DEBUG_STATEMENT(e) |
1228 | # define DEBUG_PRINT1(x) | |
1229 | # define DEBUG_PRINT2(x1, x2) | |
1230 | # define DEBUG_PRINT3(x1, x2, x3) | |
1231 | # define DEBUG_PRINT4(x1, x2, x3, x4) | |
1232 | # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) | |
1233 | # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) | |
fa9a63c5 RM |
1234 | |
1235 | #endif /* not DEBUG */ | |
1236 | \f | |
1237 | /* Set by `re_set_syntax' to the current regexp syntax to recognize. Can | |
1238 | also be assigned to arbitrarily: each pattern buffer stores its own | |
1239 | syntax, so it can be changed between regex compilations. */ | |
1240 | /* This has no initializer because initialized variables in Emacs | |
1241 | become read-only after dumping. */ | |
1242 | reg_syntax_t re_syntax_options; | |
1243 | ||
1244 | ||
1245 | /* Specify the precise syntax of regexps for compilation. This provides | |
1246 | for compatibility for various utilities which historically have | |
1247 | different, incompatible syntaxes. | |
1248 | ||
1249 | The argument SYNTAX is a bit mask comprised of the various bits | |
4bb91c68 | 1250 | defined in regex.h. We return the old syntax. */ |
fa9a63c5 RM |
1251 | |
1252 | reg_syntax_t | |
1253 | re_set_syntax (syntax) | |
f9b0fd99 | 1254 | reg_syntax_t syntax; |
fa9a63c5 RM |
1255 | { |
1256 | reg_syntax_t ret = re_syntax_options; | |
5e69f11e | 1257 | |
fa9a63c5 RM |
1258 | re_syntax_options = syntax; |
1259 | return ret; | |
1260 | } | |
c0f9ea08 | 1261 | WEAK_ALIAS (__re_set_syntax, re_set_syntax) |
f9b0fd99 RS |
1262 | |
1263 | /* Regexp to use to replace spaces, or NULL meaning don't. */ | |
1264 | static re_char *whitespace_regexp; | |
1265 | ||
1266 | void | |
1267 | re_set_whitespace_regexp (regexp) | |
1268 | re_char *regexp; | |
1269 | { | |
1270 | whitespace_regexp = regexp; | |
1271 | } | |
1272 | WEAK_ALIAS (__re_set_syntax, re_set_syntax) | |
fa9a63c5 RM |
1273 | \f |
1274 | /* This table gives an error message for each of the error codes listed | |
4bb91c68 | 1275 | in regex.h. Obviously the order here has to be same as there. |
fa9a63c5 | 1276 | POSIX doesn't require that we do anything for REG_NOERROR, |
4bb91c68 | 1277 | but why not be nice? */ |
fa9a63c5 RM |
1278 | |
1279 | static const char *re_error_msgid[] = | |
5e69f11e RM |
1280 | { |
1281 | gettext_noop ("Success"), /* REG_NOERROR */ | |
1282 | gettext_noop ("No match"), /* REG_NOMATCH */ | |
1283 | gettext_noop ("Invalid regular expression"), /* REG_BADPAT */ | |
1284 | gettext_noop ("Invalid collation character"), /* REG_ECOLLATE */ | |
1285 | gettext_noop ("Invalid character class name"), /* REG_ECTYPE */ | |
1286 | gettext_noop ("Trailing backslash"), /* REG_EESCAPE */ | |
1287 | gettext_noop ("Invalid back reference"), /* REG_ESUBREG */ | |
1288 | gettext_noop ("Unmatched [ or [^"), /* REG_EBRACK */ | |
1289 | gettext_noop ("Unmatched ( or \\("), /* REG_EPAREN */ | |
1290 | gettext_noop ("Unmatched \\{"), /* REG_EBRACE */ | |
1291 | gettext_noop ("Invalid content of \\{\\}"), /* REG_BADBR */ | |
1292 | gettext_noop ("Invalid range end"), /* REG_ERANGE */ | |
1293 | gettext_noop ("Memory exhausted"), /* REG_ESPACE */ | |
1294 | gettext_noop ("Invalid preceding regular expression"), /* REG_BADRPT */ | |
1295 | gettext_noop ("Premature end of regular expression"), /* REG_EEND */ | |
1296 | gettext_noop ("Regular expression too big"), /* REG_ESIZE */ | |
1297 | gettext_noop ("Unmatched ) or \\)"), /* REG_ERPAREN */ | |
b3e4c897 | 1298 | gettext_noop ("Range striding over charsets") /* REG_ERANGEX */ |
fa9a63c5 RM |
1299 | }; |
1300 | \f | |
4bb91c68 | 1301 | /* Avoiding alloca during matching, to placate r_alloc. */ |
fa9a63c5 RM |
1302 | |
1303 | /* Define MATCH_MAY_ALLOCATE unless we need to make sure that the | |
1304 | searching and matching functions should not call alloca. On some | |
1305 | systems, alloca is implemented in terms of malloc, and if we're | |
1306 | using the relocating allocator routines, then malloc could cause a | |
1307 | relocation, which might (if the strings being searched are in the | |
1308 | ralloc heap) shift the data out from underneath the regexp | |
1309 | routines. | |
1310 | ||
5e69f11e | 1311 | Here's another reason to avoid allocation: Emacs |
fa9a63c5 RM |
1312 | processes input from X in a signal handler; processing X input may |
1313 | call malloc; if input arrives while a matching routine is calling | |
1314 | malloc, then we're scrod. But Emacs can't just block input while | |
1315 | calling matching routines; then we don't notice interrupts when | |
1316 | they come in. So, Emacs blocks input around all regexp calls | |
1317 | except the matching calls, which it leaves unprotected, in the | |
1318 | faith that they will not malloc. */ | |
1319 | ||
1320 | /* Normally, this is fine. */ | |
1321 | #define MATCH_MAY_ALLOCATE | |
1322 | ||
1323 | /* When using GNU C, we are not REALLY using the C alloca, no matter | |
1324 | what config.h may say. So don't take precautions for it. */ | |
1325 | #ifdef __GNUC__ | |
0b32bf0e | 1326 | # undef C_ALLOCA |
fa9a63c5 RM |
1327 | #endif |
1328 | ||
1329 | /* The match routines may not allocate if (1) they would do it with malloc | |
1330 | and (2) it's not safe for them to use malloc. | |
1331 | Note that if REL_ALLOC is defined, matching would not use malloc for the | |
1332 | failure stack, but we would still use it for the register vectors; | |
4bb91c68 | 1333 | so REL_ALLOC should not affect this. */ |
0b32bf0e SM |
1334 | #if (defined C_ALLOCA || defined REGEX_MALLOC) && defined emacs |
1335 | # undef MATCH_MAY_ALLOCATE | |
fa9a63c5 RM |
1336 | #endif |
1337 | ||
1338 | \f | |
1339 | /* Failure stack declarations and macros; both re_compile_fastmap and | |
1340 | re_match_2 use a failure stack. These have to be macros because of | |
1341 | REGEX_ALLOCATE_STACK. */ | |
5e69f11e | 1342 | |
fa9a63c5 | 1343 | |
320a2a73 | 1344 | /* Approximate number of failure points for which to initially allocate space |
fa9a63c5 RM |
1345 | when matching. If this number is exceeded, we allocate more |
1346 | space, so it is not a hard limit. */ | |
1347 | #ifndef INIT_FAILURE_ALLOC | |
0b32bf0e | 1348 | # define INIT_FAILURE_ALLOC 20 |
fa9a63c5 RM |
1349 | #endif |
1350 | ||
1351 | /* Roughly the maximum number of failure points on the stack. Would be | |
320a2a73 | 1352 | exactly that if always used TYPICAL_FAILURE_SIZE items each time we failed. |
fa9a63c5 | 1353 | This is a variable only so users of regex can assign to it; we never |
ada30c0e SM |
1354 | change it ourselves. We always multiply it by TYPICAL_FAILURE_SIZE |
1355 | before using it, so it should probably be a byte-count instead. */ | |
c0f9ea08 SM |
1356 | # if defined MATCH_MAY_ALLOCATE |
1357 | /* Note that 4400 was enough to cause a crash on Alpha OSF/1, | |
320a2a73 KH |
1358 | whose default stack limit is 2mb. In order for a larger |
1359 | value to work reliably, you have to try to make it accord | |
1360 | with the process stack limit. */ | |
c0f9ea08 SM |
1361 | size_t re_max_failures = 40000; |
1362 | # else | |
1363 | size_t re_max_failures = 4000; | |
1364 | # endif | |
fa9a63c5 RM |
1365 | |
1366 | union fail_stack_elt | |
1367 | { | |
01618498 | 1368 | re_char *pointer; |
c0f9ea08 SM |
1369 | /* This should be the biggest `int' that's no bigger than a pointer. */ |
1370 | long integer; | |
fa9a63c5 RM |
1371 | }; |
1372 | ||
1373 | typedef union fail_stack_elt fail_stack_elt_t; | |
1374 | ||
1375 | typedef struct | |
1376 | { | |
1377 | fail_stack_elt_t *stack; | |
c0f9ea08 SM |
1378 | size_t size; |
1379 | size_t avail; /* Offset of next open position. */ | |
1380 | size_t frame; /* Offset of the cur constructed frame. */ | |
fa9a63c5 RM |
1381 | } fail_stack_type; |
1382 | ||
505bde11 | 1383 | #define FAIL_STACK_EMPTY() (fail_stack.frame == 0) |
fa9a63c5 RM |
1384 | #define FAIL_STACK_FULL() (fail_stack.avail == fail_stack.size) |
1385 | ||
1386 | ||
1387 | /* Define macros to initialize and free the failure stack. | |
1388 | Do `return -2' if the alloc fails. */ | |
1389 | ||
1390 | #ifdef MATCH_MAY_ALLOCATE | |
0b32bf0e | 1391 | # define INIT_FAIL_STACK() \ |
fa9a63c5 RM |
1392 | do { \ |
1393 | fail_stack.stack = (fail_stack_elt_t *) \ | |
320a2a73 KH |
1394 | REGEX_ALLOCATE_STACK (INIT_FAILURE_ALLOC * TYPICAL_FAILURE_SIZE \ |
1395 | * sizeof (fail_stack_elt_t)); \ | |
fa9a63c5 RM |
1396 | \ |
1397 | if (fail_stack.stack == NULL) \ | |
1398 | return -2; \ | |
1399 | \ | |
1400 | fail_stack.size = INIT_FAILURE_ALLOC; \ | |
1401 | fail_stack.avail = 0; \ | |
505bde11 | 1402 | fail_stack.frame = 0; \ |
fa9a63c5 RM |
1403 | } while (0) |
1404 | ||
0b32bf0e | 1405 | # define RESET_FAIL_STACK() REGEX_FREE_STACK (fail_stack.stack) |
fa9a63c5 | 1406 | #else |
0b32bf0e | 1407 | # define INIT_FAIL_STACK() \ |
fa9a63c5 RM |
1408 | do { \ |
1409 | fail_stack.avail = 0; \ | |
505bde11 | 1410 | fail_stack.frame = 0; \ |
fa9a63c5 RM |
1411 | } while (0) |
1412 | ||
0b32bf0e | 1413 | # define RESET_FAIL_STACK() ((void)0) |
fa9a63c5 RM |
1414 | #endif |
1415 | ||
1416 | ||
320a2a73 KH |
1417 | /* Double the size of FAIL_STACK, up to a limit |
1418 | which allows approximately `re_max_failures' items. | |
fa9a63c5 RM |
1419 | |
1420 | Return 1 if succeeds, and 0 if either ran out of memory | |
5e69f11e RM |
1421 | allocating space for it or it was already too large. |
1422 | ||
4bb91c68 | 1423 | REGEX_REALLOCATE_STACK requires `destination' be declared. */ |
fa9a63c5 | 1424 | |
320a2a73 KH |
1425 | /* Factor to increase the failure stack size by |
1426 | when we increase it. | |
1427 | This used to be 2, but 2 was too wasteful | |
1428 | because the old discarded stacks added up to as much space | |
1429 | were as ultimate, maximum-size stack. */ | |
1430 | #define FAIL_STACK_GROWTH_FACTOR 4 | |
1431 | ||
1432 | #define GROW_FAIL_STACK(fail_stack) \ | |
eead07d6 KH |
1433 | (((fail_stack).size * sizeof (fail_stack_elt_t) \ |
1434 | >= re_max_failures * TYPICAL_FAILURE_SIZE) \ | |
fa9a63c5 | 1435 | ? 0 \ |
320a2a73 KH |
1436 | : ((fail_stack).stack \ |
1437 | = (fail_stack_elt_t *) \ | |
25fe55af RS |
1438 | REGEX_REALLOCATE_STACK ((fail_stack).stack, \ |
1439 | (fail_stack).size * sizeof (fail_stack_elt_t), \ | |
320a2a73 KH |
1440 | MIN (re_max_failures * TYPICAL_FAILURE_SIZE, \ |
1441 | ((fail_stack).size * sizeof (fail_stack_elt_t) \ | |
1442 | * FAIL_STACK_GROWTH_FACTOR))), \ | |
fa9a63c5 RM |
1443 | \ |
1444 | (fail_stack).stack == NULL \ | |
1445 | ? 0 \ | |
6453db45 KH |
1446 | : ((fail_stack).size \ |
1447 | = (MIN (re_max_failures * TYPICAL_FAILURE_SIZE, \ | |
1448 | ((fail_stack).size * sizeof (fail_stack_elt_t) \ | |
1449 | * FAIL_STACK_GROWTH_FACTOR)) \ | |
1450 | / sizeof (fail_stack_elt_t)), \ | |
25fe55af | 1451 | 1))) |
fa9a63c5 RM |
1452 | |
1453 | ||
fa9a63c5 RM |
1454 | /* Push a pointer value onto the failure stack. |
1455 | Assumes the variable `fail_stack'. Probably should only | |
4bb91c68 | 1456 | be called from within `PUSH_FAILURE_POINT'. */ |
fa9a63c5 | 1457 | #define PUSH_FAILURE_POINTER(item) \ |
01618498 | 1458 | fail_stack.stack[fail_stack.avail++].pointer = (item) |
fa9a63c5 RM |
1459 | |
1460 | /* This pushes an integer-valued item onto the failure stack. | |
1461 | Assumes the variable `fail_stack'. Probably should only | |
4bb91c68 | 1462 | be called from within `PUSH_FAILURE_POINT'. */ |
fa9a63c5 RM |
1463 | #define PUSH_FAILURE_INT(item) \ |
1464 | fail_stack.stack[fail_stack.avail++].integer = (item) | |
1465 | ||
1466 | /* Push a fail_stack_elt_t value onto the failure stack. | |
1467 | Assumes the variable `fail_stack'. Probably should only | |
4bb91c68 | 1468 | be called from within `PUSH_FAILURE_POINT'. */ |
fa9a63c5 RM |
1469 | #define PUSH_FAILURE_ELT(item) \ |
1470 | fail_stack.stack[fail_stack.avail++] = (item) | |
1471 | ||
1472 | /* These three POP... operations complement the three PUSH... operations. | |
1473 | All assume that `fail_stack' is nonempty. */ | |
1474 | #define POP_FAILURE_POINTER() fail_stack.stack[--fail_stack.avail].pointer | |
1475 | #define POP_FAILURE_INT() fail_stack.stack[--fail_stack.avail].integer | |
1476 | #define POP_FAILURE_ELT() fail_stack.stack[--fail_stack.avail] | |
1477 | ||
505bde11 SM |
1478 | /* Individual items aside from the registers. */ |
1479 | #define NUM_NONREG_ITEMS 3 | |
1480 | ||
1481 | /* Used to examine the stack (to detect infinite loops). */ | |
1482 | #define FAILURE_PAT(h) fail_stack.stack[(h) - 1].pointer | |
66f0296e | 1483 | #define FAILURE_STR(h) (fail_stack.stack[(h) - 2].pointer) |
505bde11 SM |
1484 | #define NEXT_FAILURE_HANDLE(h) fail_stack.stack[(h) - 3].integer |
1485 | #define TOP_FAILURE_HANDLE() fail_stack.frame | |
fa9a63c5 RM |
1486 | |
1487 | ||
505bde11 SM |
1488 | #define ENSURE_FAIL_STACK(space) \ |
1489 | while (REMAINING_AVAIL_SLOTS <= space) { \ | |
1490 | if (!GROW_FAIL_STACK (fail_stack)) \ | |
1491 | return -2; \ | |
1492 | DEBUG_PRINT2 ("\n Doubled stack; size now: %d\n", (fail_stack).size);\ | |
1493 | DEBUG_PRINT2 (" slots available: %d\n", REMAINING_AVAIL_SLOTS);\ | |
1494 | } | |
1495 | ||
1496 | /* Push register NUM onto the stack. */ | |
1497 | #define PUSH_FAILURE_REG(num) \ | |
1498 | do { \ | |
1499 | char *destination; \ | |
1500 | ENSURE_FAIL_STACK(3); \ | |
1501 | DEBUG_PRINT4 (" Push reg %d (spanning %p -> %p)\n", \ | |
1502 | num, regstart[num], regend[num]); \ | |
1503 | PUSH_FAILURE_POINTER (regstart[num]); \ | |
1504 | PUSH_FAILURE_POINTER (regend[num]); \ | |
1505 | PUSH_FAILURE_INT (num); \ | |
1506 | } while (0) | |
1507 | ||
01618498 SM |
1508 | /* Change the counter's value to VAL, but make sure that it will |
1509 | be reset when backtracking. */ | |
1510 | #define PUSH_NUMBER(ptr,val) \ | |
dc1e502d SM |
1511 | do { \ |
1512 | char *destination; \ | |
1513 | int c; \ | |
1514 | ENSURE_FAIL_STACK(3); \ | |
1515 | EXTRACT_NUMBER (c, ptr); \ | |
01618498 | 1516 | DEBUG_PRINT4 (" Push number %p = %d -> %d\n", ptr, c, val); \ |
dc1e502d SM |
1517 | PUSH_FAILURE_INT (c); \ |
1518 | PUSH_FAILURE_POINTER (ptr); \ | |
1519 | PUSH_FAILURE_INT (-1); \ | |
01618498 | 1520 | STORE_NUMBER (ptr, val); \ |
dc1e502d SM |
1521 | } while (0) |
1522 | ||
505bde11 | 1523 | /* Pop a saved register off the stack. */ |
dc1e502d | 1524 | #define POP_FAILURE_REG_OR_COUNT() \ |
505bde11 SM |
1525 | do { \ |
1526 | int reg = POP_FAILURE_INT (); \ | |
dc1e502d SM |
1527 | if (reg == -1) \ |
1528 | { \ | |
1529 | /* It's a counter. */ \ | |
6dcf2d0e SM |
1530 | /* Here, we discard `const', making re_match non-reentrant. */ \ |
1531 | unsigned char *ptr = (unsigned char*) POP_FAILURE_POINTER (); \ | |
dc1e502d SM |
1532 | reg = POP_FAILURE_INT (); \ |
1533 | STORE_NUMBER (ptr, reg); \ | |
1534 | DEBUG_PRINT3 (" Pop counter %p = %d\n", ptr, reg); \ | |
1535 | } \ | |
1536 | else \ | |
1537 | { \ | |
1538 | regend[reg] = POP_FAILURE_POINTER (); \ | |
1539 | regstart[reg] = POP_FAILURE_POINTER (); \ | |
1540 | DEBUG_PRINT4 (" Pop reg %d (spanning %p -> %p)\n", \ | |
1541 | reg, regstart[reg], regend[reg]); \ | |
1542 | } \ | |
505bde11 SM |
1543 | } while (0) |
1544 | ||
1545 | /* Check that we are not stuck in an infinite loop. */ | |
1546 | #define CHECK_INFINITE_LOOP(pat_cur, string_place) \ | |
1547 | do { \ | |
f6df485f | 1548 | int failure = TOP_FAILURE_HANDLE (); \ |
505bde11 | 1549 | /* Check for infinite matching loops */ \ |
f6df485f RS |
1550 | while (failure > 0 \ |
1551 | && (FAILURE_STR (failure) == string_place \ | |
1552 | || FAILURE_STR (failure) == NULL)) \ | |
505bde11 SM |
1553 | { \ |
1554 | assert (FAILURE_PAT (failure) >= bufp->buffer \ | |
66f0296e | 1555 | && FAILURE_PAT (failure) <= bufp->buffer + bufp->used); \ |
505bde11 | 1556 | if (FAILURE_PAT (failure) == pat_cur) \ |
f6df485f | 1557 | { \ |
6df42991 SM |
1558 | cycle = 1; \ |
1559 | break; \ | |
f6df485f | 1560 | } \ |
66f0296e | 1561 | DEBUG_PRINT2 (" Other pattern: %p\n", FAILURE_PAT (failure)); \ |
505bde11 SM |
1562 | failure = NEXT_FAILURE_HANDLE(failure); \ |
1563 | } \ | |
1564 | DEBUG_PRINT2 (" Other string: %p\n", FAILURE_STR (failure)); \ | |
1565 | } while (0) | |
6df42991 | 1566 | |
fa9a63c5 | 1567 | /* Push the information about the state we will need |
5e69f11e RM |
1568 | if we ever fail back to it. |
1569 | ||
505bde11 | 1570 | Requires variables fail_stack, regstart, regend and |
320a2a73 | 1571 | num_regs be declared. GROW_FAIL_STACK requires `destination' be |
fa9a63c5 | 1572 | declared. |
5e69f11e | 1573 | |
fa9a63c5 RM |
1574 | Does `return FAILURE_CODE' if runs out of memory. */ |
1575 | ||
505bde11 SM |
1576 | #define PUSH_FAILURE_POINT(pattern, string_place) \ |
1577 | do { \ | |
1578 | char *destination; \ | |
1579 | /* Must be int, so when we don't save any registers, the arithmetic \ | |
1580 | of 0 + -1 isn't done as unsigned. */ \ | |
1581 | \ | |
505bde11 | 1582 | DEBUG_STATEMENT (nfailure_points_pushed++); \ |
4bb91c68 | 1583 | DEBUG_PRINT1 ("\nPUSH_FAILURE_POINT:\n"); \ |
505bde11 SM |
1584 | DEBUG_PRINT2 (" Before push, next avail: %d\n", (fail_stack).avail); \ |
1585 | DEBUG_PRINT2 (" size: %d\n", (fail_stack).size);\ | |
1586 | \ | |
1587 | ENSURE_FAIL_STACK (NUM_NONREG_ITEMS); \ | |
1588 | \ | |
1589 | DEBUG_PRINT1 ("\n"); \ | |
1590 | \ | |
1591 | DEBUG_PRINT2 (" Push frame index: %d\n", fail_stack.frame); \ | |
1592 | PUSH_FAILURE_INT (fail_stack.frame); \ | |
1593 | \ | |
1594 | DEBUG_PRINT2 (" Push string %p: `", string_place); \ | |
1595 | DEBUG_PRINT_DOUBLE_STRING (string_place, string1, size1, string2, size2);\ | |
1596 | DEBUG_PRINT1 ("'\n"); \ | |
1597 | PUSH_FAILURE_POINTER (string_place); \ | |
1598 | \ | |
1599 | DEBUG_PRINT2 (" Push pattern %p: ", pattern); \ | |
1600 | DEBUG_PRINT_COMPILED_PATTERN (bufp, pattern, pend); \ | |
1601 | PUSH_FAILURE_POINTER (pattern); \ | |
1602 | \ | |
1603 | /* Close the frame by moving the frame pointer past it. */ \ | |
1604 | fail_stack.frame = fail_stack.avail; \ | |
1605 | } while (0) | |
fa9a63c5 | 1606 | |
320a2a73 KH |
1607 | /* Estimate the size of data pushed by a typical failure stack entry. |
1608 | An estimate is all we need, because all we use this for | |
1609 | is to choose a limit for how big to make the failure stack. */ | |
ada30c0e | 1610 | /* BEWARE, the value `20' is hard-coded in emacs.c:main(). */ |
320a2a73 | 1611 | #define TYPICAL_FAILURE_SIZE 20 |
fa9a63c5 | 1612 | |
fa9a63c5 RM |
1613 | /* How many items can still be added to the stack without overflowing it. */ |
1614 | #define REMAINING_AVAIL_SLOTS ((fail_stack).size - (fail_stack).avail) | |
1615 | ||
1616 | ||
1617 | /* Pops what PUSH_FAIL_STACK pushes. | |
1618 | ||
1619 | We restore into the parameters, all of which should be lvalues: | |
1620 | STR -- the saved data position. | |
1621 | PAT -- the saved pattern position. | |
fa9a63c5 | 1622 | REGSTART, REGEND -- arrays of string positions. |
5e69f11e | 1623 | |
fa9a63c5 | 1624 | Also assumes the variables `fail_stack' and (if debugging), `bufp', |
25fe55af | 1625 | `pend', `string1', `size1', `string2', and `size2'. */ |
fa9a63c5 | 1626 | |
505bde11 SM |
1627 | #define POP_FAILURE_POINT(str, pat) \ |
1628 | do { \ | |
fa9a63c5 RM |
1629 | assert (!FAIL_STACK_EMPTY ()); \ |
1630 | \ | |
1631 | /* Remove failure points and point to how many regs pushed. */ \ | |
1632 | DEBUG_PRINT1 ("POP_FAILURE_POINT:\n"); \ | |
1633 | DEBUG_PRINT2 (" Before pop, next avail: %d\n", fail_stack.avail); \ | |
25fe55af | 1634 | DEBUG_PRINT2 (" size: %d\n", fail_stack.size); \ |
fa9a63c5 | 1635 | \ |
505bde11 SM |
1636 | /* Pop the saved registers. */ \ |
1637 | while (fail_stack.frame < fail_stack.avail) \ | |
dc1e502d | 1638 | POP_FAILURE_REG_OR_COUNT (); \ |
fa9a63c5 | 1639 | \ |
01618498 | 1640 | pat = POP_FAILURE_POINTER (); \ |
505bde11 SM |
1641 | DEBUG_PRINT2 (" Popping pattern %p: ", pat); \ |
1642 | DEBUG_PRINT_COMPILED_PATTERN (bufp, pat, pend); \ | |
fa9a63c5 RM |
1643 | \ |
1644 | /* If the saved string location is NULL, it came from an \ | |
1645 | on_failure_keep_string_jump opcode, and we want to throw away the \ | |
1646 | saved NULL, thus retaining our current position in the string. */ \ | |
01618498 | 1647 | str = POP_FAILURE_POINTER (); \ |
505bde11 | 1648 | DEBUG_PRINT2 (" Popping string %p: `", str); \ |
fa9a63c5 RM |
1649 | DEBUG_PRINT_DOUBLE_STRING (str, string1, size1, string2, size2); \ |
1650 | DEBUG_PRINT1 ("'\n"); \ | |
1651 | \ | |
505bde11 SM |
1652 | fail_stack.frame = POP_FAILURE_INT (); \ |
1653 | DEBUG_PRINT2 (" Popping frame index: %d\n", fail_stack.frame); \ | |
fa9a63c5 | 1654 | \ |
505bde11 SM |
1655 | assert (fail_stack.avail >= 0); \ |
1656 | assert (fail_stack.frame <= fail_stack.avail); \ | |
fa9a63c5 | 1657 | \ |
fa9a63c5 | 1658 | DEBUG_STATEMENT (nfailure_points_popped++); \ |
505bde11 | 1659 | } while (0) /* POP_FAILURE_POINT */ |
fa9a63c5 RM |
1660 | |
1661 | ||
1662 | \f | |
fa9a63c5 | 1663 | /* Registers are set to a sentinel when they haven't yet matched. */ |
4bb91c68 | 1664 | #define REG_UNSET(e) ((e) == NULL) |
fa9a63c5 RM |
1665 | \f |
1666 | /* Subroutine declarations and macros for regex_compile. */ | |
1667 | ||
4bb91c68 SM |
1668 | static reg_errcode_t regex_compile _RE_ARGS ((re_char *pattern, size_t size, |
1669 | reg_syntax_t syntax, | |
1670 | struct re_pattern_buffer *bufp)); | |
1671 | static void store_op1 _RE_ARGS ((re_opcode_t op, unsigned char *loc, int arg)); | |
1672 | static void store_op2 _RE_ARGS ((re_opcode_t op, unsigned char *loc, | |
1673 | int arg1, int arg2)); | |
1674 | static void insert_op1 _RE_ARGS ((re_opcode_t op, unsigned char *loc, | |
1675 | int arg, unsigned char *end)); | |
1676 | static void insert_op2 _RE_ARGS ((re_opcode_t op, unsigned char *loc, | |
1677 | int arg1, int arg2, unsigned char *end)); | |
01618498 SM |
1678 | static boolean at_begline_loc_p _RE_ARGS ((re_char *pattern, |
1679 | re_char *p, | |
4bb91c68 | 1680 | reg_syntax_t syntax)); |
01618498 SM |
1681 | static boolean at_endline_loc_p _RE_ARGS ((re_char *p, |
1682 | re_char *pend, | |
4bb91c68 | 1683 | reg_syntax_t syntax)); |
01618498 SM |
1684 | static re_char *skip_one_char _RE_ARGS ((re_char *p)); |
1685 | static int analyse_first _RE_ARGS ((re_char *p, re_char *pend, | |
4bb91c68 | 1686 | char *fastmap, const int multibyte)); |
fa9a63c5 | 1687 | |
fa9a63c5 | 1688 | /* Fetch the next character in the uncompiled pattern, with no |
4bb91c68 | 1689 | translation. */ |
36595814 | 1690 | #define PATFETCH(c) \ |
2d1675e4 SM |
1691 | do { \ |
1692 | int len; \ | |
1693 | if (p == pend) return REG_EEND; \ | |
1694 | c = RE_STRING_CHAR_AND_LENGTH (p, pend - p, len); \ | |
1695 | p += len; \ | |
fa9a63c5 RM |
1696 | } while (0) |
1697 | ||
fa9a63c5 RM |
1698 | |
1699 | /* If `translate' is non-null, return translate[D], else just D. We | |
1700 | cast the subscript to translate because some data is declared as | |
1701 | `char *', to avoid warnings when a string constant is passed. But | |
1702 | when we use a character as a subscript we must make it unsigned. */ | |
6676cb1c | 1703 | #ifndef TRANSLATE |
0b32bf0e | 1704 | # define TRANSLATE(d) \ |
66f0296e | 1705 | (RE_TRANSLATE_P (translate) ? RE_TRANSLATE (translate, (d)) : (d)) |
6676cb1c | 1706 | #endif |
fa9a63c5 RM |
1707 | |
1708 | ||
1709 | /* Macros for outputting the compiled pattern into `buffer'. */ | |
1710 | ||
1711 | /* If the buffer isn't allocated when it comes in, use this. */ | |
1712 | #define INIT_BUF_SIZE 32 | |
1713 | ||
4bb91c68 | 1714 | /* Make sure we have at least N more bytes of space in buffer. */ |
fa9a63c5 | 1715 | #define GET_BUFFER_SPACE(n) \ |
01618498 | 1716 | while ((size_t) (b - bufp->buffer + (n)) > bufp->allocated) \ |
fa9a63c5 RM |
1717 | EXTEND_BUFFER () |
1718 | ||
1719 | /* Make sure we have one more byte of buffer space and then add C to it. */ | |
1720 | #define BUF_PUSH(c) \ | |
1721 | do { \ | |
1722 | GET_BUFFER_SPACE (1); \ | |
1723 | *b++ = (unsigned char) (c); \ | |
1724 | } while (0) | |
1725 | ||
1726 | ||
1727 | /* Ensure we have two more bytes of buffer space and then append C1 and C2. */ | |
1728 | #define BUF_PUSH_2(c1, c2) \ | |
1729 | do { \ | |
1730 | GET_BUFFER_SPACE (2); \ | |
1731 | *b++ = (unsigned char) (c1); \ | |
1732 | *b++ = (unsigned char) (c2); \ | |
1733 | } while (0) | |
1734 | ||
1735 | ||
4bb91c68 | 1736 | /* As with BUF_PUSH_2, except for three bytes. */ |
fa9a63c5 RM |
1737 | #define BUF_PUSH_3(c1, c2, c3) \ |
1738 | do { \ | |
1739 | GET_BUFFER_SPACE (3); \ | |
1740 | *b++ = (unsigned char) (c1); \ | |
1741 | *b++ = (unsigned char) (c2); \ | |
1742 | *b++ = (unsigned char) (c3); \ | |
1743 | } while (0) | |
1744 | ||
1745 | ||
1746 | /* Store a jump with opcode OP at LOC to location TO. We store a | |
4bb91c68 | 1747 | relative address offset by the three bytes the jump itself occupies. */ |
fa9a63c5 RM |
1748 | #define STORE_JUMP(op, loc, to) \ |
1749 | store_op1 (op, loc, (to) - (loc) - 3) | |
1750 | ||
1751 | /* Likewise, for a two-argument jump. */ | |
1752 | #define STORE_JUMP2(op, loc, to, arg) \ | |
1753 | store_op2 (op, loc, (to) - (loc) - 3, arg) | |
1754 | ||
4bb91c68 | 1755 | /* Like `STORE_JUMP', but for inserting. Assume `b' is the buffer end. */ |
fa9a63c5 RM |
1756 | #define INSERT_JUMP(op, loc, to) \ |
1757 | insert_op1 (op, loc, (to) - (loc) - 3, b) | |
1758 | ||
1759 | /* Like `STORE_JUMP2', but for inserting. Assume `b' is the buffer end. */ | |
1760 | #define INSERT_JUMP2(op, loc, to, arg) \ | |
1761 | insert_op2 (op, loc, (to) - (loc) - 3, arg, b) | |
1762 | ||
1763 | ||
1764 | /* This is not an arbitrary limit: the arguments which represent offsets | |
275a3409 | 1765 | into the pattern are two bytes long. So if 2^15 bytes turns out to |
fa9a63c5 | 1766 | be too small, many things would have to change. */ |
275a3409 RS |
1767 | # define MAX_BUF_SIZE (1L << 15) |
1768 | ||
1769 | #if 0 /* This is when we thought it could be 2^16 bytes. */ | |
4bb91c68 SM |
1770 | /* Any other compiler which, like MSC, has allocation limit below 2^16 |
1771 | bytes will have to use approach similar to what was done below for | |
1772 | MSC and drop MAX_BUF_SIZE a bit. Otherwise you may end up | |
1773 | reallocating to 0 bytes. Such thing is not going to work too well. | |
1774 | You have been warned!! */ | |
1775 | #if defined _MSC_VER && !defined WIN32 | |
1776 | /* Microsoft C 16-bit versions limit malloc to approx 65512 bytes. */ | |
1777 | # define MAX_BUF_SIZE 65500L | |
1778 | #else | |
1779 | # define MAX_BUF_SIZE (1L << 16) | |
1780 | #endif | |
275a3409 | 1781 | #endif /* 0 */ |
fa9a63c5 RM |
1782 | |
1783 | /* Extend the buffer by twice its current size via realloc and | |
1784 | reset the pointers that pointed into the old block to point to the | |
1785 | correct places in the new one. If extending the buffer results in it | |
4bb91c68 SM |
1786 | being larger than MAX_BUF_SIZE, then flag memory exhausted. */ |
1787 | #if __BOUNDED_POINTERS__ | |
1788 | # define SET_HIGH_BOUND(P) (__ptrhigh (P) = __ptrlow (P) + bufp->allocated) | |
1789 | # define MOVE_BUFFER_POINTER(P) \ | |
1790 | (__ptrlow (P) += incr, SET_HIGH_BOUND (P), __ptrvalue (P) += incr) | |
1791 | # define ELSE_EXTEND_BUFFER_HIGH_BOUND \ | |
1792 | else \ | |
1793 | { \ | |
1794 | SET_HIGH_BOUND (b); \ | |
1795 | SET_HIGH_BOUND (begalt); \ | |
1796 | if (fixup_alt_jump) \ | |
1797 | SET_HIGH_BOUND (fixup_alt_jump); \ | |
1798 | if (laststart) \ | |
1799 | SET_HIGH_BOUND (laststart); \ | |
1800 | if (pending_exact) \ | |
1801 | SET_HIGH_BOUND (pending_exact); \ | |
1802 | } | |
1803 | #else | |
1804 | # define MOVE_BUFFER_POINTER(P) (P) += incr | |
1805 | # define ELSE_EXTEND_BUFFER_HIGH_BOUND | |
1806 | #endif | |
fa9a63c5 | 1807 | #define EXTEND_BUFFER() \ |
25fe55af | 1808 | do { \ |
01618498 | 1809 | re_char *old_buffer = bufp->buffer; \ |
25fe55af | 1810 | if (bufp->allocated == MAX_BUF_SIZE) \ |
fa9a63c5 RM |
1811 | return REG_ESIZE; \ |
1812 | bufp->allocated <<= 1; \ | |
1813 | if (bufp->allocated > MAX_BUF_SIZE) \ | |
25fe55af | 1814 | bufp->allocated = MAX_BUF_SIZE; \ |
01618498 | 1815 | RETALLOC (bufp->buffer, bufp->allocated, unsigned char); \ |
fa9a63c5 RM |
1816 | if (bufp->buffer == NULL) \ |
1817 | return REG_ESPACE; \ | |
1818 | /* If the buffer moved, move all the pointers into it. */ \ | |
1819 | if (old_buffer != bufp->buffer) \ | |
1820 | { \ | |
4bb91c68 SM |
1821 | int incr = bufp->buffer - old_buffer; \ |
1822 | MOVE_BUFFER_POINTER (b); \ | |
1823 | MOVE_BUFFER_POINTER (begalt); \ | |
25fe55af | 1824 | if (fixup_alt_jump) \ |
4bb91c68 | 1825 | MOVE_BUFFER_POINTER (fixup_alt_jump); \ |
25fe55af | 1826 | if (laststart) \ |
4bb91c68 | 1827 | MOVE_BUFFER_POINTER (laststart); \ |
25fe55af | 1828 | if (pending_exact) \ |
4bb91c68 | 1829 | MOVE_BUFFER_POINTER (pending_exact); \ |
fa9a63c5 | 1830 | } \ |
4bb91c68 | 1831 | ELSE_EXTEND_BUFFER_HIGH_BOUND \ |
fa9a63c5 RM |
1832 | } while (0) |
1833 | ||
1834 | ||
1835 | /* Since we have one byte reserved for the register number argument to | |
1836 | {start,stop}_memory, the maximum number of groups we can report | |
1837 | things about is what fits in that byte. */ | |
1838 | #define MAX_REGNUM 255 | |
1839 | ||
1840 | /* But patterns can have more than `MAX_REGNUM' registers. We just | |
1841 | ignore the excess. */ | |
098d42af | 1842 | typedef int regnum_t; |
fa9a63c5 RM |
1843 | |
1844 | ||
1845 | /* Macros for the compile stack. */ | |
1846 | ||
1847 | /* Since offsets can go either forwards or backwards, this type needs to | |
4bb91c68 SM |
1848 | be able to hold values from -(MAX_BUF_SIZE - 1) to MAX_BUF_SIZE - 1. */ |
1849 | /* int may be not enough when sizeof(int) == 2. */ | |
1850 | typedef long pattern_offset_t; | |
fa9a63c5 RM |
1851 | |
1852 | typedef struct | |
1853 | { | |
1854 | pattern_offset_t begalt_offset; | |
1855 | pattern_offset_t fixup_alt_jump; | |
5e69f11e | 1856 | pattern_offset_t laststart_offset; |
fa9a63c5 RM |
1857 | regnum_t regnum; |
1858 | } compile_stack_elt_t; | |
1859 | ||
1860 | ||
1861 | typedef struct | |
1862 | { | |
1863 | compile_stack_elt_t *stack; | |
1864 | unsigned size; | |
1865 | unsigned avail; /* Offset of next open position. */ | |
1866 | } compile_stack_type; | |
1867 | ||
1868 | ||
1869 | #define INIT_COMPILE_STACK_SIZE 32 | |
1870 | ||
1871 | #define COMPILE_STACK_EMPTY (compile_stack.avail == 0) | |
1872 | #define COMPILE_STACK_FULL (compile_stack.avail == compile_stack.size) | |
1873 | ||
4bb91c68 | 1874 | /* The next available element. */ |
fa9a63c5 RM |
1875 | #define COMPILE_STACK_TOP (compile_stack.stack[compile_stack.avail]) |
1876 | ||
77d11aec RS |
1877 | /* Explicit quit checking is only used on NTemacs. */ |
1878 | #if defined WINDOWSNT && defined emacs && defined QUIT | |
1879 | extern int immediate_quit; | |
1880 | # define IMMEDIATE_QUIT_CHECK \ | |
1881 | do { \ | |
1882 | if (immediate_quit) QUIT; \ | |
1883 | } while (0) | |
1884 | #else | |
1885 | # define IMMEDIATE_QUIT_CHECK ((void)0) | |
1886 | #endif | |
1887 | \f | |
b18215fc RS |
1888 | /* Structure to manage work area for range table. */ |
1889 | struct range_table_work_area | |
1890 | { | |
1891 | int *table; /* actual work area. */ | |
1892 | int allocated; /* allocated size for work area in bytes. */ | |
25fe55af | 1893 | int used; /* actually used size in words. */ |
96cc36cc | 1894 | int bits; /* flag to record character classes */ |
b18215fc RS |
1895 | }; |
1896 | ||
77d11aec RS |
1897 | /* Make sure that WORK_AREA can hold more N multibyte characters. |
1898 | This is used only in set_image_of_range and set_image_of_range_1. | |
1899 | It expects WORK_AREA to be a pointer. | |
1900 | If it can't get the space, it returns from the surrounding function. */ | |
1901 | ||
1902 | #define EXTEND_RANGE_TABLE(work_area, n) \ | |
1903 | do { \ | |
1904 | if (((work_area)->used + (n)) * sizeof (int) > (work_area)->allocated) \ | |
1905 | { \ | |
1906 | extend_range_table_work_area (work_area); \ | |
1907 | if ((work_area)->table == 0) \ | |
1908 | return (REG_ESPACE); \ | |
1909 | } \ | |
b18215fc RS |
1910 | } while (0) |
1911 | ||
96cc36cc RS |
1912 | #define SET_RANGE_TABLE_WORK_AREA_BIT(work_area, bit) \ |
1913 | (work_area).bits |= (bit) | |
1914 | ||
14473664 SM |
1915 | /* Bits used to implement the multibyte-part of the various character classes |
1916 | such as [:alnum:] in a charset's range table. */ | |
1917 | #define BIT_WORD 0x1 | |
1918 | #define BIT_LOWER 0x2 | |
1919 | #define BIT_PUNCT 0x4 | |
1920 | #define BIT_SPACE 0x8 | |
1921 | #define BIT_UPPER 0x10 | |
1922 | #define BIT_MULTIBYTE 0x20 | |
96cc36cc | 1923 | |
36595814 SM |
1924 | /* Set a range START..END to WORK_AREA. |
1925 | The range is passed through TRANSLATE, so START and END | |
1926 | should be untranslated. */ | |
77d11aec RS |
1927 | #define SET_RANGE_TABLE_WORK_AREA(work_area, start, end) \ |
1928 | do { \ | |
1929 | int tem; \ | |
1930 | tem = set_image_of_range (&work_area, start, end, translate); \ | |
1931 | if (tem > 0) \ | |
1932 | FREE_STACK_RETURN (tem); \ | |
b18215fc RS |
1933 | } while (0) |
1934 | ||
25fe55af | 1935 | /* Free allocated memory for WORK_AREA. */ |
b18215fc RS |
1936 | #define FREE_RANGE_TABLE_WORK_AREA(work_area) \ |
1937 | do { \ | |
1938 | if ((work_area).table) \ | |
1939 | free ((work_area).table); \ | |
1940 | } while (0) | |
1941 | ||
96cc36cc | 1942 | #define CLEAR_RANGE_TABLE_WORK_USED(work_area) ((work_area).used = 0, (work_area).bits = 0) |
b18215fc | 1943 | #define RANGE_TABLE_WORK_USED(work_area) ((work_area).used) |
96cc36cc | 1944 | #define RANGE_TABLE_WORK_BITS(work_area) ((work_area).bits) |
b18215fc | 1945 | #define RANGE_TABLE_WORK_ELT(work_area, i) ((work_area).table[i]) |
77d11aec | 1946 | \f |
b18215fc | 1947 | |
fa9a63c5 | 1948 | /* Set the bit for character C in a list. */ |
01618498 | 1949 | #define SET_LIST_BIT(c) (b[((c)) / BYTEWIDTH] |= 1 << ((c) % BYTEWIDTH)) |
fa9a63c5 RM |
1950 | |
1951 | ||
1952 | /* Get the next unsigned number in the uncompiled pattern. */ | |
25fe55af | 1953 | #define GET_UNSIGNED_NUMBER(num) \ |
c72b0edd SM |
1954 | do { \ |
1955 | if (p == pend) \ | |
1956 | FREE_STACK_RETURN (REG_EBRACE); \ | |
1957 | else \ | |
1958 | { \ | |
1959 | PATFETCH (c); \ | |
1960 | while ('0' <= c && c <= '9') \ | |
1961 | { \ | |
1962 | int prev; \ | |
1963 | if (num < 0) \ | |
1964 | num = 0; \ | |
1965 | prev = num; \ | |
1966 | num = num * 10 + c - '0'; \ | |
1967 | if (num / 10 != prev) \ | |
1968 | FREE_STACK_RETURN (REG_BADBR); \ | |
1969 | if (p == pend) \ | |
1970 | FREE_STACK_RETURN (REG_EBRACE); \ | |
1971 | PATFETCH (c); \ | |
1972 | } \ | |
1973 | } \ | |
1974 | } while (0) | |
77d11aec | 1975 | \f |
1fdab503 | 1976 | #if ! WIDE_CHAR_SUPPORT |
01618498 | 1977 | |
14473664 | 1978 | /* Map a string to the char class it names (if any). */ |
1fdab503 | 1979 | re_wctype_t |
ada30c0e SM |
1980 | re_wctype (str) |
1981 | re_char *str; | |
14473664 | 1982 | { |
ada30c0e | 1983 | const char *string = str; |
14473664 SM |
1984 | if (STREQ (string, "alnum")) return RECC_ALNUM; |
1985 | else if (STREQ (string, "alpha")) return RECC_ALPHA; | |
1986 | else if (STREQ (string, "word")) return RECC_WORD; | |
1987 | else if (STREQ (string, "ascii")) return RECC_ASCII; | |
1988 | else if (STREQ (string, "nonascii")) return RECC_NONASCII; | |
1989 | else if (STREQ (string, "graph")) return RECC_GRAPH; | |
1990 | else if (STREQ (string, "lower")) return RECC_LOWER; | |
1991 | else if (STREQ (string, "print")) return RECC_PRINT; | |
1992 | else if (STREQ (string, "punct")) return RECC_PUNCT; | |
1993 | else if (STREQ (string, "space")) return RECC_SPACE; | |
1994 | else if (STREQ (string, "upper")) return RECC_UPPER; | |
1995 | else if (STREQ (string, "unibyte")) return RECC_UNIBYTE; | |
1996 | else if (STREQ (string, "multibyte")) return RECC_MULTIBYTE; | |
1997 | else if (STREQ (string, "digit")) return RECC_DIGIT; | |
1998 | else if (STREQ (string, "xdigit")) return RECC_XDIGIT; | |
1999 | else if (STREQ (string, "cntrl")) return RECC_CNTRL; | |
2000 | else if (STREQ (string, "blank")) return RECC_BLANK; | |
2001 | else return 0; | |
2002 | } | |
2003 | ||
2004 | /* True iff CH is in the char class CC. */ | |
1fdab503 | 2005 | boolean |
14473664 SM |
2006 | re_iswctype (ch, cc) |
2007 | int ch; | |
2008 | re_wctype_t cc; | |
2009 | { | |
2010 | switch (cc) | |
2011 | { | |
0cdd06f8 SM |
2012 | case RECC_ALNUM: return ISALNUM (ch); |
2013 | case RECC_ALPHA: return ISALPHA (ch); | |
2014 | case RECC_BLANK: return ISBLANK (ch); | |
2015 | case RECC_CNTRL: return ISCNTRL (ch); | |
2016 | case RECC_DIGIT: return ISDIGIT (ch); | |
2017 | case RECC_GRAPH: return ISGRAPH (ch); | |
2018 | case RECC_LOWER: return ISLOWER (ch); | |
2019 | case RECC_PRINT: return ISPRINT (ch); | |
2020 | case RECC_PUNCT: return ISPUNCT (ch); | |
2021 | case RECC_SPACE: return ISSPACE (ch); | |
2022 | case RECC_UPPER: return ISUPPER (ch); | |
2023 | case RECC_XDIGIT: return ISXDIGIT (ch); | |
2024 | case RECC_ASCII: return IS_REAL_ASCII (ch); | |
2025 | case RECC_NONASCII: return !IS_REAL_ASCII (ch); | |
2026 | case RECC_UNIBYTE: return ISUNIBYTE (ch); | |
2027 | case RECC_MULTIBYTE: return !ISUNIBYTE (ch); | |
2028 | case RECC_WORD: return ISWORD (ch); | |
2029 | case RECC_ERROR: return false; | |
2030 | default: | |
2031 | abort(); | |
14473664 SM |
2032 | } |
2033 | } | |
fa9a63c5 | 2034 | |
14473664 SM |
2035 | /* Return a bit-pattern to use in the range-table bits to match multibyte |
2036 | chars of class CC. */ | |
2037 | static int | |
2038 | re_wctype_to_bit (cc) | |
2039 | re_wctype_t cc; | |
2040 | { | |
2041 | switch (cc) | |
2042 | { | |
2043 | case RECC_NONASCII: case RECC_PRINT: case RECC_GRAPH: | |
0cdd06f8 SM |
2044 | case RECC_MULTIBYTE: return BIT_MULTIBYTE; |
2045 | case RECC_ALPHA: case RECC_ALNUM: case RECC_WORD: return BIT_WORD; | |
2046 | case RECC_LOWER: return BIT_LOWER; | |
2047 | case RECC_UPPER: return BIT_UPPER; | |
2048 | case RECC_PUNCT: return BIT_PUNCT; | |
2049 | case RECC_SPACE: return BIT_SPACE; | |
14473664 | 2050 | case RECC_ASCII: case RECC_DIGIT: case RECC_XDIGIT: case RECC_CNTRL: |
0cdd06f8 SM |
2051 | case RECC_BLANK: case RECC_UNIBYTE: case RECC_ERROR: return 0; |
2052 | default: | |
2053 | abort(); | |
14473664 SM |
2054 | } |
2055 | } | |
2056 | #endif | |
77d11aec RS |
2057 | \f |
2058 | /* Filling in the work area of a range. */ | |
2059 | ||
2060 | /* Actually extend the space in WORK_AREA. */ | |
2061 | ||
2062 | static void | |
2063 | extend_range_table_work_area (work_area) | |
2064 | struct range_table_work_area *work_area; | |
177c0ea7 | 2065 | { |
77d11aec RS |
2066 | work_area->allocated += 16 * sizeof (int); |
2067 | if (work_area->table) | |
2068 | work_area->table | |
2069 | = (int *) realloc (work_area->table, work_area->allocated); | |
2070 | else | |
2071 | work_area->table | |
2072 | = (int *) malloc (work_area->allocated); | |
2073 | } | |
2074 | ||
2075 | #ifdef emacs | |
2076 | ||
2077 | /* Carefully find the ranges of codes that are equivalent | |
2078 | under case conversion to the range start..end when passed through | |
2079 | TRANSLATE. Handle the case where non-letters can come in between | |
2080 | two upper-case letters (which happens in Latin-1). | |
2081 | Also handle the case of groups of more than 2 case-equivalent chars. | |
2082 | ||
2083 | The basic method is to look at consecutive characters and see | |
2084 | if they can form a run that can be handled as one. | |
2085 | ||
2086 | Returns -1 if successful, REG_ESPACE if ran out of space. */ | |
2087 | ||
2088 | static int | |
2089 | set_image_of_range_1 (work_area, start, end, translate) | |
2090 | RE_TRANSLATE_TYPE translate; | |
2091 | struct range_table_work_area *work_area; | |
2092 | re_wchar_t start, end; | |
2093 | { | |
2094 | /* `one_case' indicates a character, or a run of characters, | |
2095 | each of which is an isolate (no case-equivalents). | |
2096 | This includes all ASCII non-letters. | |
2097 | ||
2098 | `two_case' indicates a character, or a run of characters, | |
2099 | each of which has two case-equivalent forms. | |
2100 | This includes all ASCII letters. | |
2101 | ||
2102 | `strange' indicates a character that has more than one | |
2103 | case-equivalent. */ | |
177c0ea7 | 2104 | |
77d11aec RS |
2105 | enum case_type {one_case, two_case, strange}; |
2106 | ||
2107 | /* Describe the run that is in progress, | |
2108 | which the next character can try to extend. | |
2109 | If run_type is strange, that means there really is no run. | |
2110 | If run_type is one_case, then run_start...run_end is the run. | |
2111 | If run_type is two_case, then the run is run_start...run_end, | |
2112 | and the case-equivalents end at run_eqv_end. */ | |
2113 | ||
2114 | enum case_type run_type = strange; | |
2115 | int run_start, run_end, run_eqv_end; | |
2116 | ||
2117 | Lisp_Object eqv_table; | |
2118 | ||
2119 | if (!RE_TRANSLATE_P (translate)) | |
2120 | { | |
b7c12565 | 2121 | EXTEND_RANGE_TABLE (work_area, 2); |
77d11aec RS |
2122 | work_area->table[work_area->used++] = (start); |
2123 | work_area->table[work_area->used++] = (end); | |
b7c12565 | 2124 | return -1; |
77d11aec RS |
2125 | } |
2126 | ||
2127 | eqv_table = XCHAR_TABLE (translate)->extras[2]; | |
99633e97 | 2128 | |
77d11aec RS |
2129 | for (; start <= end; start++) |
2130 | { | |
2131 | enum case_type this_type; | |
2132 | int eqv = RE_TRANSLATE (eqv_table, start); | |
2133 | int minchar, maxchar; | |
2134 | ||
2135 | /* Classify this character */ | |
2136 | if (eqv == start) | |
2137 | this_type = one_case; | |
2138 | else if (RE_TRANSLATE (eqv_table, eqv) == start) | |
2139 | this_type = two_case; | |
2140 | else | |
2141 | this_type = strange; | |
2142 | ||
2143 | if (start < eqv) | |
2144 | minchar = start, maxchar = eqv; | |
2145 | else | |
2146 | minchar = eqv, maxchar = start; | |
2147 | ||
2148 | /* Can this character extend the run in progress? */ | |
2149 | if (this_type == strange || this_type != run_type | |
2150 | || !(minchar == run_end + 1 | |
2151 | && (run_type == two_case | |
2152 | ? maxchar == run_eqv_end + 1 : 1))) | |
2153 | { | |
2154 | /* No, end the run. | |
2155 | Record each of its equivalent ranges. */ | |
2156 | if (run_type == one_case) | |
2157 | { | |
2158 | EXTEND_RANGE_TABLE (work_area, 2); | |
2159 | work_area->table[work_area->used++] = run_start; | |
2160 | work_area->table[work_area->used++] = run_end; | |
2161 | } | |
2162 | else if (run_type == two_case) | |
2163 | { | |
2164 | EXTEND_RANGE_TABLE (work_area, 4); | |
2165 | work_area->table[work_area->used++] = run_start; | |
2166 | work_area->table[work_area->used++] = run_end; | |
2167 | work_area->table[work_area->used++] | |
2168 | = RE_TRANSLATE (eqv_table, run_start); | |
2169 | work_area->table[work_area->used++] | |
2170 | = RE_TRANSLATE (eqv_table, run_end); | |
2171 | } | |
2172 | run_type = strange; | |
2173 | } | |
177c0ea7 | 2174 | |
77d11aec RS |
2175 | if (this_type == strange) |
2176 | { | |
2177 | /* For a strange character, add each of its equivalents, one | |
2178 | by one. Don't start a range. */ | |
2179 | do | |
2180 | { | |
2181 | EXTEND_RANGE_TABLE (work_area, 2); | |
2182 | work_area->table[work_area->used++] = eqv; | |
2183 | work_area->table[work_area->used++] = eqv; | |
2184 | eqv = RE_TRANSLATE (eqv_table, eqv); | |
2185 | } | |
2186 | while (eqv != start); | |
2187 | } | |
2188 | ||
2189 | /* Add this char to the run, or start a new run. */ | |
2190 | else if (run_type == strange) | |
2191 | { | |
2192 | /* Initialize a new range. */ | |
2193 | run_type = this_type; | |
2194 | run_start = start; | |
2195 | run_end = start; | |
2196 | run_eqv_end = RE_TRANSLATE (eqv_table, run_end); | |
2197 | } | |
2198 | else | |
2199 | { | |
2200 | /* Extend a running range. */ | |
2201 | run_end = minchar; | |
2202 | run_eqv_end = RE_TRANSLATE (eqv_table, run_end); | |
2203 | } | |
2204 | } | |
2205 | ||
2206 | /* If a run is still in progress at the end, finish it now | |
2207 | by recording its equivalent ranges. */ | |
2208 | if (run_type == one_case) | |
2209 | { | |
2210 | EXTEND_RANGE_TABLE (work_area, 2); | |
2211 | work_area->table[work_area->used++] = run_start; | |
2212 | work_area->table[work_area->used++] = run_end; | |
2213 | } | |
2214 | else if (run_type == two_case) | |
2215 | { | |
2216 | EXTEND_RANGE_TABLE (work_area, 4); | |
2217 | work_area->table[work_area->used++] = run_start; | |
2218 | work_area->table[work_area->used++] = run_end; | |
2219 | work_area->table[work_area->used++] | |
2220 | = RE_TRANSLATE (eqv_table, run_start); | |
2221 | work_area->table[work_area->used++] | |
2222 | = RE_TRANSLATE (eqv_table, run_end); | |
2223 | } | |
2224 | ||
2225 | return -1; | |
2226 | } | |
36595814 | 2227 | |
77d11aec | 2228 | #endif /* emacs */ |
36595814 | 2229 | |
b7c12565 | 2230 | /* Record the the image of the range start..end when passed through |
36595814 SM |
2231 | TRANSLATE. This is not necessarily TRANSLATE(start)..TRANSLATE(end) |
2232 | and is not even necessarily contiguous. | |
b7c12565 RS |
2233 | Normally we approximate it with the smallest contiguous range that contains |
2234 | all the chars we need. However, for Latin-1 we go to extra effort | |
2235 | to do a better job. | |
2236 | ||
2237 | This function is not called for ASCII ranges. | |
77d11aec RS |
2238 | |
2239 | Returns -1 if successful, REG_ESPACE if ran out of space. */ | |
2240 | ||
2241 | static int | |
36595814 SM |
2242 | set_image_of_range (work_area, start, end, translate) |
2243 | RE_TRANSLATE_TYPE translate; | |
2244 | struct range_table_work_area *work_area; | |
2245 | re_wchar_t start, end; | |
2246 | { | |
77d11aec RS |
2247 | re_wchar_t cmin, cmax; |
2248 | ||
2249 | #ifdef emacs | |
2250 | /* For Latin-1 ranges, use set_image_of_range_1 | |
2251 | to get proper handling of ranges that include letters and nonletters. | |
b7c12565 | 2252 | For a range that includes the whole of Latin-1, this is not necessary. |
77d11aec | 2253 | For other character sets, we don't bother to get this right. */ |
b7c12565 RS |
2254 | if (RE_TRANSLATE_P (translate) && start < 04400 |
2255 | && !(start < 04200 && end >= 04377)) | |
77d11aec | 2256 | { |
b7c12565 | 2257 | int newend; |
77d11aec | 2258 | int tem; |
b7c12565 RS |
2259 | newend = end; |
2260 | if (newend > 04377) | |
2261 | newend = 04377; | |
2262 | tem = set_image_of_range_1 (work_area, start, newend, translate); | |
77d11aec RS |
2263 | if (tem > 0) |
2264 | return tem; | |
2265 | ||
2266 | start = 04400; | |
2267 | if (end < 04400) | |
2268 | return -1; | |
2269 | } | |
2270 | #endif | |
2271 | ||
b7c12565 RS |
2272 | EXTEND_RANGE_TABLE (work_area, 2); |
2273 | work_area->table[work_area->used++] = (start); | |
2274 | work_area->table[work_area->used++] = (end); | |
2275 | ||
2276 | cmin = -1, cmax = -1; | |
77d11aec | 2277 | |
36595814 | 2278 | if (RE_TRANSLATE_P (translate)) |
b7c12565 RS |
2279 | { |
2280 | int ch; | |
77d11aec | 2281 | |
b7c12565 RS |
2282 | for (ch = start; ch <= end; ch++) |
2283 | { | |
2284 | re_wchar_t c = TRANSLATE (ch); | |
2285 | if (! (start <= c && c <= end)) | |
2286 | { | |
2287 | if (cmin == -1) | |
2288 | cmin = c, cmax = c; | |
2289 | else | |
2290 | { | |
2291 | cmin = MIN (cmin, c); | |
2292 | cmax = MAX (cmax, c); | |
2293 | } | |
2294 | } | |
2295 | } | |
2296 | ||
2297 | if (cmin != -1) | |
2298 | { | |
2299 | EXTEND_RANGE_TABLE (work_area, 2); | |
2300 | work_area->table[work_area->used++] = (cmin); | |
2301 | work_area->table[work_area->used++] = (cmax); | |
2302 | } | |
2303 | } | |
36595814 | 2304 | |
77d11aec RS |
2305 | return -1; |
2306 | } | |
fa9a63c5 RM |
2307 | \f |
2308 | #ifndef MATCH_MAY_ALLOCATE | |
2309 | ||
2310 | /* If we cannot allocate large objects within re_match_2_internal, | |
2311 | we make the fail stack and register vectors global. | |
2312 | The fail stack, we grow to the maximum size when a regexp | |
2313 | is compiled. | |
2314 | The register vectors, we adjust in size each time we | |
2315 | compile a regexp, according to the number of registers it needs. */ | |
2316 | ||
2317 | static fail_stack_type fail_stack; | |
2318 | ||
2319 | /* Size with which the following vectors are currently allocated. | |
2320 | That is so we can make them bigger as needed, | |
4bb91c68 | 2321 | but never make them smaller. */ |
fa9a63c5 RM |
2322 | static int regs_allocated_size; |
2323 | ||
66f0296e SM |
2324 | static re_char ** regstart, ** regend; |
2325 | static re_char **best_regstart, **best_regend; | |
fa9a63c5 RM |
2326 | |
2327 | /* Make the register vectors big enough for NUM_REGS registers, | |
4bb91c68 | 2328 | but don't make them smaller. */ |
fa9a63c5 RM |
2329 | |
2330 | static | |
2331 | regex_grow_registers (num_regs) | |
2332 | int num_regs; | |
2333 | { | |
2334 | if (num_regs > regs_allocated_size) | |
2335 | { | |
66f0296e SM |
2336 | RETALLOC_IF (regstart, num_regs, re_char *); |
2337 | RETALLOC_IF (regend, num_regs, re_char *); | |
2338 | RETALLOC_IF (best_regstart, num_regs, re_char *); | |
2339 | RETALLOC_IF (best_regend, num_regs, re_char *); | |
fa9a63c5 RM |
2340 | |
2341 | regs_allocated_size = num_regs; | |
2342 | } | |
2343 | } | |
2344 | ||
2345 | #endif /* not MATCH_MAY_ALLOCATE */ | |
2346 | \f | |
99633e97 SM |
2347 | static boolean group_in_compile_stack _RE_ARGS ((compile_stack_type |
2348 | compile_stack, | |
2349 | regnum_t regnum)); | |
2350 | ||
fa9a63c5 RM |
2351 | /* `regex_compile' compiles PATTERN (of length SIZE) according to SYNTAX. |
2352 | Returns one of error codes defined in `regex.h', or zero for success. | |
2353 | ||
2354 | Assumes the `allocated' (and perhaps `buffer') and `translate' | |
2355 | fields are set in BUFP on entry. | |
2356 | ||
2357 | If it succeeds, results are put in BUFP (if it returns an error, the | |
2358 | contents of BUFP are undefined): | |
2359 | `buffer' is the compiled pattern; | |
2360 | `syntax' is set to SYNTAX; | |
2361 | `used' is set to the length of the compiled pattern; | |
2362 | `fastmap_accurate' is zero; | |
2363 | `re_nsub' is the number of subexpressions in PATTERN; | |
2364 | `not_bol' and `not_eol' are zero; | |
5e69f11e | 2365 | |
c0f9ea08 | 2366 | The `fastmap' field is neither examined nor set. */ |
fa9a63c5 | 2367 | |
505bde11 SM |
2368 | /* Insert the `jump' from the end of last alternative to "here". |
2369 | The space for the jump has already been allocated. */ | |
2370 | #define FIXUP_ALT_JUMP() \ | |
2371 | do { \ | |
2372 | if (fixup_alt_jump) \ | |
2373 | STORE_JUMP (jump, fixup_alt_jump, b); \ | |
2374 | } while (0) | |
2375 | ||
2376 | ||
fa9a63c5 RM |
2377 | /* Return, freeing storage we allocated. */ |
2378 | #define FREE_STACK_RETURN(value) \ | |
b18215fc RS |
2379 | do { \ |
2380 | FREE_RANGE_TABLE_WORK_AREA (range_table_work); \ | |
2381 | free (compile_stack.stack); \ | |
2382 | return value; \ | |
2383 | } while (0) | |
fa9a63c5 RM |
2384 | |
2385 | static reg_errcode_t | |
2386 | regex_compile (pattern, size, syntax, bufp) | |
66f0296e | 2387 | re_char *pattern; |
4bb91c68 | 2388 | size_t size; |
fa9a63c5 RM |
2389 | reg_syntax_t syntax; |
2390 | struct re_pattern_buffer *bufp; | |
2391 | { | |
01618498 SM |
2392 | /* We fetch characters from PATTERN here. */ |
2393 | register re_wchar_t c, c1; | |
5e69f11e | 2394 | |
fa9a63c5 | 2395 | /* A random temporary spot in PATTERN. */ |
66f0296e | 2396 | re_char *p1; |
fa9a63c5 RM |
2397 | |
2398 | /* Points to the end of the buffer, where we should append. */ | |
2399 | register unsigned char *b; | |
5e69f11e | 2400 | |
fa9a63c5 RM |
2401 | /* Keeps track of unclosed groups. */ |
2402 | compile_stack_type compile_stack; | |
2403 | ||
2404 | /* Points to the current (ending) position in the pattern. */ | |
22336245 RS |
2405 | #ifdef AIX |
2406 | /* `const' makes AIX compiler fail. */ | |
66f0296e | 2407 | unsigned char *p = pattern; |
22336245 | 2408 | #else |
66f0296e | 2409 | re_char *p = pattern; |
22336245 | 2410 | #endif |
66f0296e | 2411 | re_char *pend = pattern + size; |
5e69f11e | 2412 | |
fa9a63c5 | 2413 | /* How to translate the characters in the pattern. */ |
6676cb1c | 2414 | RE_TRANSLATE_TYPE translate = bufp->translate; |
fa9a63c5 RM |
2415 | |
2416 | /* Address of the count-byte of the most recently inserted `exactn' | |
2417 | command. This makes it possible to tell if a new exact-match | |
2418 | character can be added to that command or if the character requires | |
2419 | a new `exactn' command. */ | |
2420 | unsigned char *pending_exact = 0; | |
2421 | ||
2422 | /* Address of start of the most recently finished expression. | |
2423 | This tells, e.g., postfix * where to find the start of its | |
2424 | operand. Reset at the beginning of groups and alternatives. */ | |
2425 | unsigned char *laststart = 0; | |
2426 | ||
2427 | /* Address of beginning of regexp, or inside of last group. */ | |
2428 | unsigned char *begalt; | |
2429 | ||
2430 | /* Place in the uncompiled pattern (i.e., the {) to | |
2431 | which to go back if the interval is invalid. */ | |
66f0296e | 2432 | re_char *beg_interval; |
5e69f11e | 2433 | |
fa9a63c5 | 2434 | /* Address of the place where a forward jump should go to the end of |
25fe55af | 2435 | the containing expression. Each alternative of an `or' -- except the |
fa9a63c5 RM |
2436 | last -- ends with a forward jump of this sort. */ |
2437 | unsigned char *fixup_alt_jump = 0; | |
2438 | ||
2439 | /* Counts open-groups as they are encountered. Remembered for the | |
2440 | matching close-group on the compile stack, so the same register | |
2441 | number is put in the stop_memory as the start_memory. */ | |
2442 | regnum_t regnum = 0; | |
2443 | ||
b18215fc RS |
2444 | /* Work area for range table of charset. */ |
2445 | struct range_table_work_area range_table_work; | |
2446 | ||
2d1675e4 SM |
2447 | /* If the object matched can contain multibyte characters. */ |
2448 | const boolean multibyte = RE_MULTIBYTE_P (bufp); | |
2449 | ||
f9b0fd99 RS |
2450 | /* Nonzero if we have pushed down into a subpattern. */ |
2451 | int in_subpattern = 0; | |
2452 | ||
2453 | /* These hold the values of p, pattern, and pend from the main | |
2454 | pattern when we have pushed into a subpattern. */ | |
2455 | re_char *main_p; | |
2456 | re_char *main_pattern; | |
2457 | re_char *main_pend; | |
2458 | ||
fa9a63c5 | 2459 | #ifdef DEBUG |
99633e97 | 2460 | debug++; |
fa9a63c5 | 2461 | DEBUG_PRINT1 ("\nCompiling pattern: "); |
99633e97 | 2462 | if (debug > 0) |
fa9a63c5 RM |
2463 | { |
2464 | unsigned debug_count; | |
5e69f11e | 2465 | |
fa9a63c5 | 2466 | for (debug_count = 0; debug_count < size; debug_count++) |
25fe55af | 2467 | putchar (pattern[debug_count]); |
fa9a63c5 RM |
2468 | putchar ('\n'); |
2469 | } | |
2470 | #endif /* DEBUG */ | |
2471 | ||
2472 | /* Initialize the compile stack. */ | |
2473 | compile_stack.stack = TALLOC (INIT_COMPILE_STACK_SIZE, compile_stack_elt_t); | |
2474 | if (compile_stack.stack == NULL) | |
2475 | return REG_ESPACE; | |
2476 | ||
2477 | compile_stack.size = INIT_COMPILE_STACK_SIZE; | |
2478 | compile_stack.avail = 0; | |
2479 | ||
b18215fc RS |
2480 | range_table_work.table = 0; |
2481 | range_table_work.allocated = 0; | |
2482 | ||
fa9a63c5 RM |
2483 | /* Initialize the pattern buffer. */ |
2484 | bufp->syntax = syntax; | |
2485 | bufp->fastmap_accurate = 0; | |
2486 | bufp->not_bol = bufp->not_eol = 0; | |
2487 | ||
2488 | /* Set `used' to zero, so that if we return an error, the pattern | |
2489 | printer (for debugging) will think there's no pattern. We reset it | |
2490 | at the end. */ | |
2491 | bufp->used = 0; | |
5e69f11e | 2492 | |
fa9a63c5 | 2493 | /* Always count groups, whether or not bufp->no_sub is set. */ |
5e69f11e | 2494 | bufp->re_nsub = 0; |
fa9a63c5 | 2495 | |
0b32bf0e | 2496 | #if !defined emacs && !defined SYNTAX_TABLE |
fa9a63c5 RM |
2497 | /* Initialize the syntax table. */ |
2498 | init_syntax_once (); | |
2499 | #endif | |
2500 | ||
2501 | if (bufp->allocated == 0) | |
2502 | { | |
2503 | if (bufp->buffer) | |
2504 | { /* If zero allocated, but buffer is non-null, try to realloc | |
25fe55af RS |
2505 | enough space. This loses if buffer's address is bogus, but |
2506 | that is the user's responsibility. */ | |
2507 | RETALLOC (bufp->buffer, INIT_BUF_SIZE, unsigned char); | |
2508 | } | |
fa9a63c5 | 2509 | else |
25fe55af RS |
2510 | { /* Caller did not allocate a buffer. Do it for them. */ |
2511 | bufp->buffer = TALLOC (INIT_BUF_SIZE, unsigned char); | |
2512 | } | |
fa9a63c5 RM |
2513 | if (!bufp->buffer) FREE_STACK_RETURN (REG_ESPACE); |
2514 | ||
2515 | bufp->allocated = INIT_BUF_SIZE; | |
2516 | } | |
2517 | ||
2518 | begalt = b = bufp->buffer; | |
2519 | ||
2520 | /* Loop through the uncompiled pattern until we're at the end. */ | |
f9b0fd99 | 2521 | while (1) |
fa9a63c5 | 2522 | { |
f9b0fd99 RS |
2523 | if (p == pend) |
2524 | { | |
2525 | /* If this is the end of an included regexp, | |
2526 | pop back to the main regexp and try again. */ | |
2527 | if (in_subpattern) | |
2528 | { | |
2529 | in_subpattern = 0; | |
2530 | pattern = main_pattern; | |
2531 | p = main_p; | |
2532 | pend = main_pend; | |
2533 | continue; | |
2534 | } | |
2535 | /* If this is the end of the main regexp, we are done. */ | |
2536 | break; | |
2537 | } | |
2538 | ||
fa9a63c5 RM |
2539 | PATFETCH (c); |
2540 | ||
2541 | switch (c) | |
25fe55af | 2542 | { |
f9b0fd99 RS |
2543 | case ' ': |
2544 | { | |
2545 | re_char *p1 = p; | |
2546 | ||
2547 | /* If there's no special whitespace regexp, treat | |
4fb680cd RS |
2548 | spaces normally. And don't try to do this recursively. */ |
2549 | if (!whitespace_regexp || in_subpattern) | |
f9b0fd99 RS |
2550 | goto normal_char; |
2551 | ||
2552 | /* Peek past following spaces. */ | |
2553 | while (p1 != pend) | |
2554 | { | |
2555 | if (*p1 != ' ') | |
2556 | break; | |
2557 | p1++; | |
2558 | } | |
2559 | /* If the spaces are followed by a repetition op, | |
2560 | treat them normally. */ | |
c721eee5 RS |
2561 | if (p1 != pend |
2562 | && (*p1 == '*' || *p1 == '+' || *p1 == '?' | |
f9b0fd99 RS |
2563 | || (*p1 == '\\' && p1 + 1 != pend && p1[1] == '{'))) |
2564 | goto normal_char; | |
2565 | ||
2566 | /* Replace the spaces with the whitespace regexp. */ | |
2567 | in_subpattern = 1; | |
2568 | main_p = p1; | |
2569 | main_pend = pend; | |
2570 | main_pattern = pattern; | |
2571 | p = pattern = whitespace_regexp; | |
2572 | pend = p + strlen (p); | |
2573 | break; | |
2574 | } | |
2575 | ||
25fe55af RS |
2576 | case '^': |
2577 | { | |
2578 | if ( /* If at start of pattern, it's an operator. */ | |
2579 | p == pattern + 1 | |
2580 | /* If context independent, it's an operator. */ | |
2581 | || syntax & RE_CONTEXT_INDEP_ANCHORS | |
2582 | /* Otherwise, depends on what's come before. */ | |
2583 | || at_begline_loc_p (pattern, p, syntax)) | |
c0f9ea08 | 2584 | BUF_PUSH ((syntax & RE_NO_NEWLINE_ANCHOR) ? begbuf : begline); |
25fe55af RS |
2585 | else |
2586 | goto normal_char; | |
2587 | } | |
2588 | break; | |
2589 | ||
2590 | ||
2591 | case '$': | |
2592 | { | |
2593 | if ( /* If at end of pattern, it's an operator. */ | |
2594 | p == pend | |
2595 | /* If context independent, it's an operator. */ | |
2596 | || syntax & RE_CONTEXT_INDEP_ANCHORS | |
2597 | /* Otherwise, depends on what's next. */ | |
2598 | || at_endline_loc_p (p, pend, syntax)) | |
c0f9ea08 | 2599 | BUF_PUSH ((syntax & RE_NO_NEWLINE_ANCHOR) ? endbuf : endline); |
25fe55af RS |
2600 | else |
2601 | goto normal_char; | |
2602 | } | |
2603 | break; | |
fa9a63c5 RM |
2604 | |
2605 | ||
2606 | case '+': | |
25fe55af RS |
2607 | case '?': |
2608 | if ((syntax & RE_BK_PLUS_QM) | |
2609 | || (syntax & RE_LIMITED_OPS)) | |
2610 | goto normal_char; | |
2611 | handle_plus: | |
2612 | case '*': | |
2613 | /* If there is no previous pattern... */ | |
2614 | if (!laststart) | |
2615 | { | |
2616 | if (syntax & RE_CONTEXT_INVALID_OPS) | |
2617 | FREE_STACK_RETURN (REG_BADRPT); | |
2618 | else if (!(syntax & RE_CONTEXT_INDEP_OPS)) | |
2619 | goto normal_char; | |
2620 | } | |
2621 | ||
2622 | { | |
25fe55af | 2623 | /* 1 means zero (many) matches is allowed. */ |
66f0296e SM |
2624 | boolean zero_times_ok = 0, many_times_ok = 0; |
2625 | boolean greedy = 1; | |
25fe55af RS |
2626 | |
2627 | /* If there is a sequence of repetition chars, collapse it | |
2628 | down to just one (the right one). We can't combine | |
2629 | interval operators with these because of, e.g., `a{2}*', | |
2630 | which should only match an even number of `a's. */ | |
2631 | ||
2632 | for (;;) | |
2633 | { | |
0b32bf0e | 2634 | if ((syntax & RE_FRUGAL) |
1c8c6d39 DL |
2635 | && c == '?' && (zero_times_ok || many_times_ok)) |
2636 | greedy = 0; | |
2637 | else | |
2638 | { | |
2639 | zero_times_ok |= c != '+'; | |
2640 | many_times_ok |= c != '?'; | |
2641 | } | |
25fe55af RS |
2642 | |
2643 | if (p == pend) | |
2644 | break; | |
ed0767d8 SM |
2645 | else if (*p == '*' |
2646 | || (!(syntax & RE_BK_PLUS_QM) | |
2647 | && (*p == '+' || *p == '?'))) | |
25fe55af | 2648 | ; |
ed0767d8 | 2649 | else if (syntax & RE_BK_PLUS_QM && *p == '\\') |
25fe55af | 2650 | { |
ed0767d8 SM |
2651 | if (p+1 == pend) |
2652 | FREE_STACK_RETURN (REG_EESCAPE); | |
2653 | if (p[1] == '+' || p[1] == '?') | |
2654 | PATFETCH (c); /* Gobble up the backslash. */ | |
2655 | else | |
2656 | break; | |
25fe55af RS |
2657 | } |
2658 | else | |
ed0767d8 | 2659 | break; |
25fe55af | 2660 | /* If we get here, we found another repeat character. */ |
ed0767d8 SM |
2661 | PATFETCH (c); |
2662 | } | |
25fe55af RS |
2663 | |
2664 | /* Star, etc. applied to an empty pattern is equivalent | |
2665 | to an empty pattern. */ | |
4e8a9132 | 2666 | if (!laststart || laststart == b) |
25fe55af RS |
2667 | break; |
2668 | ||
2669 | /* Now we know whether or not zero matches is allowed | |
2670 | and also whether or not two or more matches is allowed. */ | |
1c8c6d39 DL |
2671 | if (greedy) |
2672 | { | |
99633e97 | 2673 | if (many_times_ok) |
4e8a9132 SM |
2674 | { |
2675 | boolean simple = skip_one_char (laststart) == b; | |
2676 | unsigned int startoffset = 0; | |
f6a3f532 | 2677 | re_opcode_t ofj = |
01618498 | 2678 | /* Check if the loop can match the empty string. */ |
6df42991 SM |
2679 | (simple || !analyse_first (laststart, b, NULL, 0)) |
2680 | ? on_failure_jump : on_failure_jump_loop; | |
4e8a9132 | 2681 | assert (skip_one_char (laststart) <= b); |
177c0ea7 | 2682 | |
4e8a9132 SM |
2683 | if (!zero_times_ok && simple) |
2684 | { /* Since simple * loops can be made faster by using | |
2685 | on_failure_keep_string_jump, we turn simple P+ | |
2686 | into PP* if P is simple. */ | |
2687 | unsigned char *p1, *p2; | |
2688 | startoffset = b - laststart; | |
2689 | GET_BUFFER_SPACE (startoffset); | |
2690 | p1 = b; p2 = laststart; | |
2691 | while (p2 < p1) | |
2692 | *b++ = *p2++; | |
2693 | zero_times_ok = 1; | |
99633e97 | 2694 | } |
4e8a9132 SM |
2695 | |
2696 | GET_BUFFER_SPACE (6); | |
2697 | if (!zero_times_ok) | |
2698 | /* A + loop. */ | |
f6a3f532 | 2699 | STORE_JUMP (ofj, b, b + 6); |
99633e97 | 2700 | else |
4e8a9132 SM |
2701 | /* Simple * loops can use on_failure_keep_string_jump |
2702 | depending on what follows. But since we don't know | |
2703 | that yet, we leave the decision up to | |
2704 | on_failure_jump_smart. */ | |
f6a3f532 | 2705 | INSERT_JUMP (simple ? on_failure_jump_smart : ofj, |
4e8a9132 | 2706 | laststart + startoffset, b + 6); |
99633e97 | 2707 | b += 3; |
4e8a9132 | 2708 | STORE_JUMP (jump, b, laststart + startoffset); |
99633e97 SM |
2709 | b += 3; |
2710 | } | |
2711 | else | |
2712 | { | |
4e8a9132 SM |
2713 | /* A simple ? pattern. */ |
2714 | assert (zero_times_ok); | |
2715 | GET_BUFFER_SPACE (3); | |
2716 | INSERT_JUMP (on_failure_jump, laststart, b + 3); | |
99633e97 SM |
2717 | b += 3; |
2718 | } | |
1c8c6d39 DL |
2719 | } |
2720 | else /* not greedy */ | |
2721 | { /* I wish the greedy and non-greedy cases could be merged. */ | |
2722 | ||
0683b6fa | 2723 | GET_BUFFER_SPACE (7); /* We might use less. */ |
1c8c6d39 DL |
2724 | if (many_times_ok) |
2725 | { | |
f6a3f532 SM |
2726 | boolean emptyp = analyse_first (laststart, b, NULL, 0); |
2727 | ||
6df42991 SM |
2728 | /* The non-greedy multiple match looks like |
2729 | a repeat..until: we only need a conditional jump | |
2730 | at the end of the loop. */ | |
f6a3f532 SM |
2731 | if (emptyp) BUF_PUSH (no_op); |
2732 | STORE_JUMP (emptyp ? on_failure_jump_nastyloop | |
2733 | : on_failure_jump, b, laststart); | |
1c8c6d39 DL |
2734 | b += 3; |
2735 | if (zero_times_ok) | |
2736 | { | |
2737 | /* The repeat...until naturally matches one or more. | |
2738 | To also match zero times, we need to first jump to | |
6df42991 | 2739 | the end of the loop (its conditional jump). */ |
1c8c6d39 DL |
2740 | INSERT_JUMP (jump, laststart, b); |
2741 | b += 3; | |
2742 | } | |
2743 | } | |
2744 | else | |
2745 | { | |
2746 | /* non-greedy a?? */ | |
1c8c6d39 DL |
2747 | INSERT_JUMP (jump, laststart, b + 3); |
2748 | b += 3; | |
2749 | INSERT_JUMP (on_failure_jump, laststart, laststart + 6); | |
2750 | b += 3; | |
2751 | } | |
2752 | } | |
2753 | } | |
4e8a9132 | 2754 | pending_exact = 0; |
fa9a63c5 RM |
2755 | break; |
2756 | ||
2757 | ||
2758 | case '.': | |
25fe55af RS |
2759 | laststart = b; |
2760 | BUF_PUSH (anychar); | |
2761 | break; | |
fa9a63c5 RM |
2762 | |
2763 | ||
25fe55af RS |
2764 | case '[': |
2765 | { | |
b18215fc | 2766 | CLEAR_RANGE_TABLE_WORK_USED (range_table_work); |
fa9a63c5 | 2767 | |
25fe55af | 2768 | if (p == pend) FREE_STACK_RETURN (REG_EBRACK); |
fa9a63c5 | 2769 | |
25fe55af RS |
2770 | /* Ensure that we have enough space to push a charset: the |
2771 | opcode, the length count, and the bitset; 34 bytes in all. */ | |
fa9a63c5 RM |
2772 | GET_BUFFER_SPACE (34); |
2773 | ||
25fe55af | 2774 | laststart = b; |
e318085a | 2775 | |
25fe55af RS |
2776 | /* We test `*p == '^' twice, instead of using an if |
2777 | statement, so we only need one BUF_PUSH. */ | |
2778 | BUF_PUSH (*p == '^' ? charset_not : charset); | |
2779 | if (*p == '^') | |
2780 | p++; | |
e318085a | 2781 | |
25fe55af RS |
2782 | /* Remember the first position in the bracket expression. */ |
2783 | p1 = p; | |
e318085a | 2784 | |
25fe55af RS |
2785 | /* Push the number of bytes in the bitmap. */ |
2786 | BUF_PUSH ((1 << BYTEWIDTH) / BYTEWIDTH); | |
e318085a | 2787 | |
25fe55af RS |
2788 | /* Clear the whole map. */ |
2789 | bzero (b, (1 << BYTEWIDTH) / BYTEWIDTH); | |
e318085a | 2790 | |
25fe55af RS |
2791 | /* charset_not matches newline according to a syntax bit. */ |
2792 | if ((re_opcode_t) b[-2] == charset_not | |
2793 | && (syntax & RE_HAT_LISTS_NOT_NEWLINE)) | |
2794 | SET_LIST_BIT ('\n'); | |
fa9a63c5 | 2795 | |
25fe55af RS |
2796 | /* Read in characters and ranges, setting map bits. */ |
2797 | for (;;) | |
2798 | { | |
b18215fc | 2799 | boolean escaped_char = false; |
2d1675e4 | 2800 | const unsigned char *p2 = p; |
e318085a | 2801 | |
25fe55af | 2802 | if (p == pend) FREE_STACK_RETURN (REG_EBRACK); |
e318085a | 2803 | |
36595814 SM |
2804 | /* Don't translate yet. The range TRANSLATE(X..Y) cannot |
2805 | always be determined from TRANSLATE(X) and TRANSLATE(Y) | |
2806 | So the translation is done later in a loop. Example: | |
2807 | (let ((case-fold-search t)) (string-match "[A-_]" "A")) */ | |
25fe55af | 2808 | PATFETCH (c); |
e318085a | 2809 | |
25fe55af RS |
2810 | /* \ might escape characters inside [...] and [^...]. */ |
2811 | if ((syntax & RE_BACKSLASH_ESCAPE_IN_LISTS) && c == '\\') | |
2812 | { | |
2813 | if (p == pend) FREE_STACK_RETURN (REG_EESCAPE); | |
e318085a RS |
2814 | |
2815 | PATFETCH (c); | |
b18215fc | 2816 | escaped_char = true; |
25fe55af | 2817 | } |
b18215fc RS |
2818 | else |
2819 | { | |
657fcfbd RS |
2820 | /* Could be the end of the bracket expression. If it's |
2821 | not (i.e., when the bracket expression is `[]' so | |
2822 | far), the ']' character bit gets set way below. */ | |
2d1675e4 | 2823 | if (c == ']' && p2 != p1) |
657fcfbd | 2824 | break; |
25fe55af | 2825 | } |
b18215fc | 2826 | |
b18215fc RS |
2827 | /* What should we do for the character which is |
2828 | greater than 0x7F, but not BASE_LEADING_CODE_P? | |
2829 | XXX */ | |
2830 | ||
25fe55af RS |
2831 | /* See if we're at the beginning of a possible character |
2832 | class. */ | |
b18215fc | 2833 | |
2d1675e4 SM |
2834 | if (!escaped_char && |
2835 | syntax & RE_CHAR_CLASSES && c == '[' && *p == ':') | |
657fcfbd RS |
2836 | { |
2837 | /* Leave room for the null. */ | |
14473664 | 2838 | unsigned char str[CHAR_CLASS_MAX_LENGTH + 1]; |
ed0767d8 | 2839 | const unsigned char *class_beg; |
b18215fc | 2840 | |
25fe55af RS |
2841 | PATFETCH (c); |
2842 | c1 = 0; | |
ed0767d8 | 2843 | class_beg = p; |
b18215fc | 2844 | |
25fe55af RS |
2845 | /* If pattern is `[[:'. */ |
2846 | if (p == pend) FREE_STACK_RETURN (REG_EBRACK); | |
b18215fc | 2847 | |
25fe55af RS |
2848 | for (;;) |
2849 | { | |
14473664 SM |
2850 | PATFETCH (c); |
2851 | if ((c == ':' && *p == ']') || p == pend) | |
2852 | break; | |
2853 | if (c1 < CHAR_CLASS_MAX_LENGTH) | |
2854 | str[c1++] = c; | |
2855 | else | |
2856 | /* This is in any case an invalid class name. */ | |
2857 | str[0] = '\0'; | |
25fe55af RS |
2858 | } |
2859 | str[c1] = '\0'; | |
b18215fc RS |
2860 | |
2861 | /* If isn't a word bracketed by `[:' and `:]': | |
2862 | undo the ending character, the letters, and | |
2863 | leave the leading `:' and `[' (but set bits for | |
2864 | them). */ | |
25fe55af RS |
2865 | if (c == ':' && *p == ']') |
2866 | { | |
36595814 | 2867 | re_wchar_t ch; |
14473664 SM |
2868 | re_wctype_t cc; |
2869 | ||
2870 | cc = re_wctype (str); | |
2871 | ||
2872 | if (cc == 0) | |
fa9a63c5 RM |
2873 | FREE_STACK_RETURN (REG_ECTYPE); |
2874 | ||
14473664 SM |
2875 | /* Throw away the ] at the end of the character |
2876 | class. */ | |
2877 | PATFETCH (c); | |
fa9a63c5 | 2878 | |
14473664 | 2879 | if (p == pend) FREE_STACK_RETURN (REG_EBRACK); |
fa9a63c5 | 2880 | |
96cc36cc RS |
2881 | /* Most character classes in a multibyte match |
2882 | just set a flag. Exceptions are is_blank, | |
2883 | is_digit, is_cntrl, and is_xdigit, since | |
2884 | they can only match ASCII characters. We | |
14473664 SM |
2885 | don't need to handle them for multibyte. |
2886 | They are distinguished by a negative wctype. */ | |
96cc36cc | 2887 | |
2d1675e4 | 2888 | if (multibyte) |
14473664 SM |
2889 | SET_RANGE_TABLE_WORK_AREA_BIT (range_table_work, |
2890 | re_wctype_to_bit (cc)); | |
96cc36cc | 2891 | |
14473664 | 2892 | for (ch = 0; ch < 1 << BYTEWIDTH; ++ch) |
25fe55af | 2893 | { |
7ae68633 | 2894 | int translated = TRANSLATE (ch); |
14473664 | 2895 | if (re_iswctype (btowc (ch), cc)) |
96cc36cc | 2896 | SET_LIST_BIT (translated); |
25fe55af | 2897 | } |
b18215fc RS |
2898 | |
2899 | /* Repeat the loop. */ | |
2900 | continue; | |
25fe55af RS |
2901 | } |
2902 | else | |
2903 | { | |
ed0767d8 SM |
2904 | /* Go back to right after the "[:". */ |
2905 | p = class_beg; | |
25fe55af | 2906 | SET_LIST_BIT ('['); |
b18215fc RS |
2907 | |
2908 | /* Because the `:' may starts the range, we | |
2909 | can't simply set bit and repeat the loop. | |
25fe55af | 2910 | Instead, just set it to C and handle below. */ |
b18215fc | 2911 | c = ':'; |
25fe55af RS |
2912 | } |
2913 | } | |
b18215fc RS |
2914 | |
2915 | if (p < pend && p[0] == '-' && p[1] != ']') | |
2916 | { | |
2917 | ||
2918 | /* Discard the `-'. */ | |
2919 | PATFETCH (c1); | |
2920 | ||
2921 | /* Fetch the character which ends the range. */ | |
2922 | PATFETCH (c1); | |
b18215fc | 2923 | |
b54f61ed | 2924 | if (SINGLE_BYTE_CHAR_P (c)) |
e934739e | 2925 | { |
b54f61ed KH |
2926 | if (! SINGLE_BYTE_CHAR_P (c1)) |
2927 | { | |
3ff2446d KH |
2928 | /* Handle a range starting with a |
2929 | character of less than 256, and ending | |
2930 | with a character of not less than 256. | |
2931 | Split that into two ranges, the low one | |
2932 | ending at 0377, and the high one | |
2933 | starting at the smallest character in | |
2934 | the charset of C1 and ending at C1. */ | |
b54f61ed | 2935 | int charset = CHAR_CHARSET (c1); |
36595814 SM |
2936 | re_wchar_t c2 = MAKE_CHAR (charset, 0, 0); |
2937 | ||
b54f61ed KH |
2938 | SET_RANGE_TABLE_WORK_AREA (range_table_work, |
2939 | c2, c1); | |
333526e0 | 2940 | c1 = 0377; |
b54f61ed | 2941 | } |
e934739e RS |
2942 | } |
2943 | else if (!SAME_CHARSET_P (c, c1)) | |
b3e4c897 | 2944 | FREE_STACK_RETURN (REG_ERANGEX); |
e318085a | 2945 | } |
25fe55af | 2946 | else |
b18215fc RS |
2947 | /* Range from C to C. */ |
2948 | c1 = c; | |
2949 | ||
2950 | /* Set the range ... */ | |
2951 | if (SINGLE_BYTE_CHAR_P (c)) | |
2952 | /* ... into bitmap. */ | |
25fe55af | 2953 | { |
01618498 | 2954 | re_wchar_t this_char; |
36595814 | 2955 | re_wchar_t range_start = c, range_end = c1; |
b18215fc RS |
2956 | |
2957 | /* If the start is after the end, the range is empty. */ | |
2958 | if (range_start > range_end) | |
2959 | { | |
2960 | if (syntax & RE_NO_EMPTY_RANGES) | |
2961 | FREE_STACK_RETURN (REG_ERANGE); | |
2962 | /* Else, repeat the loop. */ | |
2963 | } | |
2964 | else | |
2965 | { | |
2966 | for (this_char = range_start; this_char <= range_end; | |
2967 | this_char++) | |
2968 | SET_LIST_BIT (TRANSLATE (this_char)); | |
e934739e | 2969 | } |
25fe55af | 2970 | } |
e318085a | 2971 | else |
b18215fc RS |
2972 | /* ... into range table. */ |
2973 | SET_RANGE_TABLE_WORK_AREA (range_table_work, c, c1); | |
e318085a RS |
2974 | } |
2975 | ||
25fe55af RS |
2976 | /* Discard any (non)matching list bytes that are all 0 at the |
2977 | end of the map. Decrease the map-length byte too. */ | |
2978 | while ((int) b[-1] > 0 && b[b[-1] - 1] == 0) | |
2979 | b[-1]--; | |
2980 | b += b[-1]; | |
fa9a63c5 | 2981 | |
96cc36cc RS |
2982 | /* Build real range table from work area. */ |
2983 | if (RANGE_TABLE_WORK_USED (range_table_work) | |
2984 | || RANGE_TABLE_WORK_BITS (range_table_work)) | |
b18215fc RS |
2985 | { |
2986 | int i; | |
2987 | int used = RANGE_TABLE_WORK_USED (range_table_work); | |
fa9a63c5 | 2988 | |
b18215fc | 2989 | /* Allocate space for COUNT + RANGE_TABLE. Needs two |
96cc36cc RS |
2990 | bytes for flags, two for COUNT, and three bytes for |
2991 | each character. */ | |
2992 | GET_BUFFER_SPACE (4 + used * 3); | |
fa9a63c5 | 2993 | |
b18215fc RS |
2994 | /* Indicate the existence of range table. */ |
2995 | laststart[1] |= 0x80; | |
fa9a63c5 | 2996 | |
96cc36cc RS |
2997 | /* Store the character class flag bits into the range table. |
2998 | If not in emacs, these flag bits are always 0. */ | |
2999 | *b++ = RANGE_TABLE_WORK_BITS (range_table_work) & 0xff; | |
3000 | *b++ = RANGE_TABLE_WORK_BITS (range_table_work) >> 8; | |
3001 | ||
b18215fc RS |
3002 | STORE_NUMBER_AND_INCR (b, used / 2); |
3003 | for (i = 0; i < used; i++) | |
3004 | STORE_CHARACTER_AND_INCR | |
3005 | (b, RANGE_TABLE_WORK_ELT (range_table_work, i)); | |
3006 | } | |
25fe55af RS |
3007 | } |
3008 | break; | |
fa9a63c5 RM |
3009 | |
3010 | ||
b18215fc | 3011 | case '(': |
25fe55af RS |
3012 | if (syntax & RE_NO_BK_PARENS) |
3013 | goto handle_open; | |
3014 | else | |
3015 | goto normal_char; | |
fa9a63c5 RM |
3016 | |
3017 | ||
25fe55af RS |
3018 | case ')': |
3019 | if (syntax & RE_NO_BK_PARENS) | |
3020 | goto handle_close; | |
3021 | else | |
3022 | goto normal_char; | |
e318085a RS |
3023 | |
3024 | ||
25fe55af RS |
3025 | case '\n': |
3026 | if (syntax & RE_NEWLINE_ALT) | |
3027 | goto handle_alt; | |
3028 | else | |
3029 | goto normal_char; | |
e318085a RS |
3030 | |
3031 | ||
b18215fc | 3032 | case '|': |
25fe55af RS |
3033 | if (syntax & RE_NO_BK_VBAR) |
3034 | goto handle_alt; | |
3035 | else | |
3036 | goto normal_char; | |
3037 | ||
3038 | ||
3039 | case '{': | |
3040 | if (syntax & RE_INTERVALS && syntax & RE_NO_BK_BRACES) | |
3041 | goto handle_interval; | |
3042 | else | |
3043 | goto normal_char; | |
3044 | ||
3045 | ||
3046 | case '\\': | |
3047 | if (p == pend) FREE_STACK_RETURN (REG_EESCAPE); | |
3048 | ||
3049 | /* Do not translate the character after the \, so that we can | |
3050 | distinguish, e.g., \B from \b, even if we normally would | |
3051 | translate, e.g., B to b. */ | |
36595814 | 3052 | PATFETCH (c); |
25fe55af RS |
3053 | |
3054 | switch (c) | |
3055 | { | |
3056 | case '(': | |
3057 | if (syntax & RE_NO_BK_PARENS) | |
3058 | goto normal_backslash; | |
3059 | ||
3060 | handle_open: | |
505bde11 SM |
3061 | { |
3062 | int shy = 0; | |
3063 | if (p+1 < pend) | |
3064 | { | |
3065 | /* Look for a special (?...) construct */ | |
ed0767d8 | 3066 | if ((syntax & RE_SHY_GROUPS) && *p == '?') |
505bde11 | 3067 | { |
ed0767d8 | 3068 | PATFETCH (c); /* Gobble up the '?'. */ |
505bde11 SM |
3069 | PATFETCH (c); |
3070 | switch (c) | |
3071 | { | |
3072 | case ':': shy = 1; break; | |
3073 | default: | |
3074 | /* Only (?:...) is supported right now. */ | |
3075 | FREE_STACK_RETURN (REG_BADPAT); | |
3076 | } | |
3077 | } | |
505bde11 SM |
3078 | } |
3079 | ||
3080 | if (!shy) | |
3081 | { | |
3082 | bufp->re_nsub++; | |
3083 | regnum++; | |
3084 | } | |
25fe55af | 3085 | |
99633e97 SM |
3086 | if (COMPILE_STACK_FULL) |
3087 | { | |
3088 | RETALLOC (compile_stack.stack, compile_stack.size << 1, | |
3089 | compile_stack_elt_t); | |
3090 | if (compile_stack.stack == NULL) return REG_ESPACE; | |
25fe55af | 3091 | |
99633e97 SM |
3092 | compile_stack.size <<= 1; |
3093 | } | |
25fe55af | 3094 | |
99633e97 SM |
3095 | /* These are the values to restore when we hit end of this |
3096 | group. They are all relative offsets, so that if the | |
3097 | whole pattern moves because of realloc, they will still | |
3098 | be valid. */ | |
3099 | COMPILE_STACK_TOP.begalt_offset = begalt - bufp->buffer; | |
3100 | COMPILE_STACK_TOP.fixup_alt_jump | |
3101 | = fixup_alt_jump ? fixup_alt_jump - bufp->buffer + 1 : 0; | |
3102 | COMPILE_STACK_TOP.laststart_offset = b - bufp->buffer; | |
3103 | COMPILE_STACK_TOP.regnum = shy ? -regnum : regnum; | |
3104 | ||
3105 | /* Do not push a | |
3106 | start_memory for groups beyond the last one we can | |
3107 | represent in the compiled pattern. */ | |
3108 | if (regnum <= MAX_REGNUM && !shy) | |
3109 | BUF_PUSH_2 (start_memory, regnum); | |
3110 | ||
3111 | compile_stack.avail++; | |
3112 | ||
3113 | fixup_alt_jump = 0; | |
3114 | laststart = 0; | |
3115 | begalt = b; | |
3116 | /* If we've reached MAX_REGNUM groups, then this open | |
3117 | won't actually generate any code, so we'll have to | |
3118 | clear pending_exact explicitly. */ | |
3119 | pending_exact = 0; | |
3120 | break; | |
505bde11 | 3121 | } |
25fe55af RS |
3122 | |
3123 | case ')': | |
3124 | if (syntax & RE_NO_BK_PARENS) goto normal_backslash; | |
3125 | ||
3126 | if (COMPILE_STACK_EMPTY) | |
505bde11 SM |
3127 | { |
3128 | if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD) | |
3129 | goto normal_backslash; | |
3130 | else | |
3131 | FREE_STACK_RETURN (REG_ERPAREN); | |
3132 | } | |
25fe55af RS |
3133 | |
3134 | handle_close: | |
505bde11 | 3135 | FIXUP_ALT_JUMP (); |
25fe55af RS |
3136 | |
3137 | /* See similar code for backslashed left paren above. */ | |
3138 | if (COMPILE_STACK_EMPTY) | |
505bde11 SM |
3139 | { |
3140 | if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD) | |
3141 | goto normal_char; | |
3142 | else | |
3143 | FREE_STACK_RETURN (REG_ERPAREN); | |
3144 | } | |
25fe55af RS |
3145 | |
3146 | /* Since we just checked for an empty stack above, this | |
3147 | ``can't happen''. */ | |
3148 | assert (compile_stack.avail != 0); | |
3149 | { | |
3150 | /* We don't just want to restore into `regnum', because | |
3151 | later groups should continue to be numbered higher, | |
3152 | as in `(ab)c(de)' -- the second group is #2. */ | |
3153 | regnum_t this_group_regnum; | |
3154 | ||
3155 | compile_stack.avail--; | |
3156 | begalt = bufp->buffer + COMPILE_STACK_TOP.begalt_offset; | |
3157 | fixup_alt_jump | |
3158 | = COMPILE_STACK_TOP.fixup_alt_jump | |
3159 | ? bufp->buffer + COMPILE_STACK_TOP.fixup_alt_jump - 1 | |
3160 | : 0; | |
3161 | laststart = bufp->buffer + COMPILE_STACK_TOP.laststart_offset; | |
3162 | this_group_regnum = COMPILE_STACK_TOP.regnum; | |
b18215fc RS |
3163 | /* If we've reached MAX_REGNUM groups, then this open |
3164 | won't actually generate any code, so we'll have to | |
3165 | clear pending_exact explicitly. */ | |
3166 | pending_exact = 0; | |
e318085a | 3167 | |
25fe55af RS |
3168 | /* We're at the end of the group, so now we know how many |
3169 | groups were inside this one. */ | |
505bde11 SM |
3170 | if (this_group_regnum <= MAX_REGNUM && this_group_regnum > 0) |
3171 | BUF_PUSH_2 (stop_memory, this_group_regnum); | |
25fe55af RS |
3172 | } |
3173 | break; | |
3174 | ||
3175 | ||
3176 | case '|': /* `\|'. */ | |
3177 | if (syntax & RE_LIMITED_OPS || syntax & RE_NO_BK_VBAR) | |
3178 | goto normal_backslash; | |
3179 | handle_alt: | |
3180 | if (syntax & RE_LIMITED_OPS) | |
3181 | goto normal_char; | |
3182 | ||
3183 | /* Insert before the previous alternative a jump which | |
3184 | jumps to this alternative if the former fails. */ | |
3185 | GET_BUFFER_SPACE (3); | |
3186 | INSERT_JUMP (on_failure_jump, begalt, b + 6); | |
3187 | pending_exact = 0; | |
3188 | b += 3; | |
3189 | ||
3190 | /* The alternative before this one has a jump after it | |
3191 | which gets executed if it gets matched. Adjust that | |
3192 | jump so it will jump to this alternative's analogous | |
3193 | jump (put in below, which in turn will jump to the next | |
3194 | (if any) alternative's such jump, etc.). The last such | |
3195 | jump jumps to the correct final destination. A picture: | |
3196 | _____ _____ | |
3197 | | | | | | |
3198 | | v | v | |
3199 | a | b | c | |
3200 | ||
3201 | If we are at `b', then fixup_alt_jump right now points to a | |
3202 | three-byte space after `a'. We'll put in the jump, set | |
3203 | fixup_alt_jump to right after `b', and leave behind three | |
3204 | bytes which we'll fill in when we get to after `c'. */ | |
3205 | ||
505bde11 | 3206 | FIXUP_ALT_JUMP (); |
25fe55af RS |
3207 | |
3208 | /* Mark and leave space for a jump after this alternative, | |
3209 | to be filled in later either by next alternative or | |
3210 | when know we're at the end of a series of alternatives. */ | |
3211 | fixup_alt_jump = b; | |
3212 | GET_BUFFER_SPACE (3); | |
3213 | b += 3; | |
3214 | ||
3215 | laststart = 0; | |
3216 | begalt = b; | |
3217 | break; | |
3218 | ||
3219 | ||
3220 | case '{': | |
3221 | /* If \{ is a literal. */ | |
3222 | if (!(syntax & RE_INTERVALS) | |
3223 | /* If we're at `\{' and it's not the open-interval | |
3224 | operator. */ | |
4bb91c68 | 3225 | || (syntax & RE_NO_BK_BRACES)) |
25fe55af RS |
3226 | goto normal_backslash; |
3227 | ||
3228 | handle_interval: | |
3229 | { | |
3230 | /* If got here, then the syntax allows intervals. */ | |
3231 | ||
3232 | /* At least (most) this many matches must be made. */ | |
99633e97 | 3233 | int lower_bound = 0, upper_bound = -1; |
25fe55af | 3234 | |
ed0767d8 | 3235 | beg_interval = p; |
25fe55af | 3236 | |
25fe55af RS |
3237 | GET_UNSIGNED_NUMBER (lower_bound); |
3238 | ||
3239 | if (c == ',') | |
ed0767d8 | 3240 | GET_UNSIGNED_NUMBER (upper_bound); |
25fe55af RS |
3241 | else |
3242 | /* Interval such as `{1}' => match exactly once. */ | |
3243 | upper_bound = lower_bound; | |
3244 | ||
3245 | if (lower_bound < 0 || upper_bound > RE_DUP_MAX | |
ed0767d8 | 3246 | || (upper_bound >= 0 && lower_bound > upper_bound)) |
4bb91c68 | 3247 | FREE_STACK_RETURN (REG_BADBR); |
25fe55af RS |
3248 | |
3249 | if (!(syntax & RE_NO_BK_BRACES)) | |
3250 | { | |
4bb91c68 SM |
3251 | if (c != '\\') |
3252 | FREE_STACK_RETURN (REG_BADBR); | |
c72b0edd SM |
3253 | if (p == pend) |
3254 | FREE_STACK_RETURN (REG_EESCAPE); | |
25fe55af RS |
3255 | PATFETCH (c); |
3256 | } | |
3257 | ||
3258 | if (c != '}') | |
4bb91c68 | 3259 | FREE_STACK_RETURN (REG_BADBR); |
25fe55af RS |
3260 | |
3261 | /* We just parsed a valid interval. */ | |
3262 | ||
3263 | /* If it's invalid to have no preceding re. */ | |
3264 | if (!laststart) | |
3265 | { | |
3266 | if (syntax & RE_CONTEXT_INVALID_OPS) | |
3267 | FREE_STACK_RETURN (REG_BADRPT); | |
3268 | else if (syntax & RE_CONTEXT_INDEP_OPS) | |
3269 | laststart = b; | |
3270 | else | |
3271 | goto unfetch_interval; | |
3272 | } | |
3273 | ||
6df42991 SM |
3274 | if (upper_bound == 0) |
3275 | /* If the upper bound is zero, just drop the sub pattern | |
3276 | altogether. */ | |
3277 | b = laststart; | |
3278 | else if (lower_bound == 1 && upper_bound == 1) | |
3279 | /* Just match it once: nothing to do here. */ | |
3280 | ; | |
3281 | ||
3282 | /* Otherwise, we have a nontrivial interval. When | |
3283 | we're all done, the pattern will look like: | |
3284 | set_number_at <jump count> <upper bound> | |
3285 | set_number_at <succeed_n count> <lower bound> | |
3286 | succeed_n <after jump addr> <succeed_n count> | |
3287 | <body of loop> | |
3288 | jump_n <succeed_n addr> <jump count> | |
3289 | (The upper bound and `jump_n' are omitted if | |
3290 | `upper_bound' is 1, though.) */ | |
3291 | else | |
3292 | { /* If the upper bound is > 1, we need to insert | |
3293 | more at the end of the loop. */ | |
3294 | unsigned int nbytes = (upper_bound < 0 ? 3 | |
3295 | : upper_bound > 1 ? 5 : 0); | |
3296 | unsigned int startoffset = 0; | |
3297 | ||
3298 | GET_BUFFER_SPACE (20); /* We might use less. */ | |
3299 | ||
3300 | if (lower_bound == 0) | |
3301 | { | |
3302 | /* A succeed_n that starts with 0 is really a | |
3303 | a simple on_failure_jump_loop. */ | |
3304 | INSERT_JUMP (on_failure_jump_loop, laststart, | |
3305 | b + 3 + nbytes); | |
3306 | b += 3; | |
3307 | } | |
3308 | else | |
3309 | { | |
3310 | /* Initialize lower bound of the `succeed_n', even | |
3311 | though it will be set during matching by its | |
3312 | attendant `set_number_at' (inserted next), | |
3313 | because `re_compile_fastmap' needs to know. | |
3314 | Jump to the `jump_n' we might insert below. */ | |
3315 | INSERT_JUMP2 (succeed_n, laststart, | |
3316 | b + 5 + nbytes, | |
3317 | lower_bound); | |
3318 | b += 5; | |
3319 | ||
3320 | /* Code to initialize the lower bound. Insert | |
3321 | before the `succeed_n'. The `5' is the last two | |
3322 | bytes of this `set_number_at', plus 3 bytes of | |
3323 | the following `succeed_n'. */ | |
3324 | insert_op2 (set_number_at, laststart, 5, lower_bound, b); | |
3325 | b += 5; | |
3326 | startoffset += 5; | |
3327 | } | |
3328 | ||
3329 | if (upper_bound < 0) | |
3330 | { | |
3331 | /* A negative upper bound stands for infinity, | |
3332 | in which case it degenerates to a plain jump. */ | |
3333 | STORE_JUMP (jump, b, laststart + startoffset); | |
3334 | b += 3; | |
3335 | } | |
3336 | else if (upper_bound > 1) | |
3337 | { /* More than one repetition is allowed, so | |
3338 | append a backward jump to the `succeed_n' | |
3339 | that starts this interval. | |
3340 | ||
3341 | When we've reached this during matching, | |
3342 | we'll have matched the interval once, so | |
3343 | jump back only `upper_bound - 1' times. */ | |
3344 | STORE_JUMP2 (jump_n, b, laststart + startoffset, | |
3345 | upper_bound - 1); | |
3346 | b += 5; | |
3347 | ||
3348 | /* The location we want to set is the second | |
3349 | parameter of the `jump_n'; that is `b-2' as | |
3350 | an absolute address. `laststart' will be | |
3351 | the `set_number_at' we're about to insert; | |
3352 | `laststart+3' the number to set, the source | |
3353 | for the relative address. But we are | |
3354 | inserting into the middle of the pattern -- | |
3355 | so everything is getting moved up by 5. | |
3356 | Conclusion: (b - 2) - (laststart + 3) + 5, | |
3357 | i.e., b - laststart. | |
3358 | ||
3359 | We insert this at the beginning of the loop | |
3360 | so that if we fail during matching, we'll | |
3361 | reinitialize the bounds. */ | |
3362 | insert_op2 (set_number_at, laststart, b - laststart, | |
3363 | upper_bound - 1, b); | |
3364 | b += 5; | |
3365 | } | |
3366 | } | |
25fe55af RS |
3367 | pending_exact = 0; |
3368 | beg_interval = NULL; | |
3369 | } | |
3370 | break; | |
3371 | ||
3372 | unfetch_interval: | |
3373 | /* If an invalid interval, match the characters as literals. */ | |
3374 | assert (beg_interval); | |
3375 | p = beg_interval; | |
3376 | beg_interval = NULL; | |
3377 | ||
3378 | /* normal_char and normal_backslash need `c'. */ | |
ed0767d8 | 3379 | c = '{'; |
25fe55af RS |
3380 | |
3381 | if (!(syntax & RE_NO_BK_BRACES)) | |
3382 | { | |
ed0767d8 SM |
3383 | assert (p > pattern && p[-1] == '\\'); |
3384 | goto normal_backslash; | |
25fe55af | 3385 | } |
ed0767d8 SM |
3386 | else |
3387 | goto normal_char; | |
e318085a | 3388 | |
b18215fc | 3389 | #ifdef emacs |
25fe55af RS |
3390 | /* There is no way to specify the before_dot and after_dot |
3391 | operators. rms says this is ok. --karl */ | |
3392 | case '=': | |
3393 | BUF_PUSH (at_dot); | |
3394 | break; | |
3395 | ||
3396 | case 's': | |
3397 | laststart = b; | |
3398 | PATFETCH (c); | |
3399 | BUF_PUSH_2 (syntaxspec, syntax_spec_code[c]); | |
3400 | break; | |
3401 | ||
3402 | case 'S': | |
3403 | laststart = b; | |
3404 | PATFETCH (c); | |
3405 | BUF_PUSH_2 (notsyntaxspec, syntax_spec_code[c]); | |
3406 | break; | |
b18215fc RS |
3407 | |
3408 | case 'c': | |
3409 | laststart = b; | |
36595814 | 3410 | PATFETCH (c); |
b18215fc RS |
3411 | BUF_PUSH_2 (categoryspec, c); |
3412 | break; | |
e318085a | 3413 | |
b18215fc RS |
3414 | case 'C': |
3415 | laststart = b; | |
36595814 | 3416 | PATFETCH (c); |
b18215fc RS |
3417 | BUF_PUSH_2 (notcategoryspec, c); |
3418 | break; | |
3419 | #endif /* emacs */ | |
e318085a | 3420 | |
e318085a | 3421 | |
25fe55af | 3422 | case 'w': |
4bb91c68 SM |
3423 | if (syntax & RE_NO_GNU_OPS) |
3424 | goto normal_char; | |
25fe55af | 3425 | laststart = b; |
1fb352e0 | 3426 | BUF_PUSH_2 (syntaxspec, Sword); |
25fe55af | 3427 | break; |
e318085a | 3428 | |
e318085a | 3429 | |
25fe55af | 3430 | case 'W': |
4bb91c68 SM |
3431 | if (syntax & RE_NO_GNU_OPS) |
3432 | goto normal_char; | |
25fe55af | 3433 | laststart = b; |
1fb352e0 | 3434 | BUF_PUSH_2 (notsyntaxspec, Sword); |
25fe55af | 3435 | break; |
e318085a RS |
3436 | |
3437 | ||
25fe55af | 3438 | case '<': |
4bb91c68 SM |
3439 | if (syntax & RE_NO_GNU_OPS) |
3440 | goto normal_char; | |
25fe55af RS |
3441 | BUF_PUSH (wordbeg); |
3442 | break; | |
e318085a | 3443 | |
25fe55af | 3444 | case '>': |
4bb91c68 SM |
3445 | if (syntax & RE_NO_GNU_OPS) |
3446 | goto normal_char; | |
25fe55af RS |
3447 | BUF_PUSH (wordend); |
3448 | break; | |
e318085a | 3449 | |
669fa600 SM |
3450 | case '_': |
3451 | if (syntax & RE_NO_GNU_OPS) | |
3452 | goto normal_char; | |
3453 | laststart = b; | |
3454 | PATFETCH (c); | |
3455 | if (c == '<') | |
3456 | BUF_PUSH (symbeg); | |
3457 | else if (c == '>') | |
3458 | BUF_PUSH (symend); | |
3459 | else | |
3460 | FREE_STACK_RETURN (REG_BADPAT); | |
3461 | break; | |
3462 | ||
25fe55af | 3463 | case 'b': |
4bb91c68 SM |
3464 | if (syntax & RE_NO_GNU_OPS) |
3465 | goto normal_char; | |
25fe55af RS |
3466 | BUF_PUSH (wordbound); |
3467 | break; | |
e318085a | 3468 | |
25fe55af | 3469 | case 'B': |
4bb91c68 SM |
3470 | if (syntax & RE_NO_GNU_OPS) |
3471 | goto normal_char; | |
25fe55af RS |
3472 | BUF_PUSH (notwordbound); |
3473 | break; | |
fa9a63c5 | 3474 | |
25fe55af | 3475 | case '`': |
4bb91c68 SM |
3476 | if (syntax & RE_NO_GNU_OPS) |
3477 | goto normal_char; | |
25fe55af RS |
3478 | BUF_PUSH (begbuf); |
3479 | break; | |
e318085a | 3480 | |
25fe55af | 3481 | case '\'': |
4bb91c68 SM |
3482 | if (syntax & RE_NO_GNU_OPS) |
3483 | goto normal_char; | |
25fe55af RS |
3484 | BUF_PUSH (endbuf); |
3485 | break; | |
e318085a | 3486 | |
25fe55af RS |
3487 | case '1': case '2': case '3': case '4': case '5': |
3488 | case '6': case '7': case '8': case '9': | |
0cdd06f8 SM |
3489 | { |
3490 | regnum_t reg; | |
e318085a | 3491 | |
0cdd06f8 SM |
3492 | if (syntax & RE_NO_BK_REFS) |
3493 | goto normal_backslash; | |
e318085a | 3494 | |
0cdd06f8 | 3495 | reg = c - '0'; |
e318085a | 3496 | |
0cdd06f8 SM |
3497 | /* Can't back reference to a subexpression before its end. */ |
3498 | if (reg > regnum || group_in_compile_stack (compile_stack, reg)) | |
3499 | FREE_STACK_RETURN (REG_ESUBREG); | |
e318085a | 3500 | |
0cdd06f8 SM |
3501 | laststart = b; |
3502 | BUF_PUSH_2 (duplicate, reg); | |
3503 | } | |
25fe55af | 3504 | break; |
e318085a | 3505 | |
e318085a | 3506 | |
25fe55af RS |
3507 | case '+': |
3508 | case '?': | |
3509 | if (syntax & RE_BK_PLUS_QM) | |
3510 | goto handle_plus; | |
3511 | else | |
3512 | goto normal_backslash; | |
3513 | ||
3514 | default: | |
3515 | normal_backslash: | |
3516 | /* You might think it would be useful for \ to mean | |
3517 | not to translate; but if we don't translate it | |
4bb91c68 | 3518 | it will never match anything. */ |
25fe55af RS |
3519 | goto normal_char; |
3520 | } | |
3521 | break; | |
fa9a63c5 RM |
3522 | |
3523 | ||
3524 | default: | |
25fe55af | 3525 | /* Expects the character in `c'. */ |
fa9a63c5 | 3526 | normal_char: |
36595814 | 3527 | /* If no exactn currently being built. */ |
25fe55af | 3528 | if (!pending_exact |
fa9a63c5 | 3529 | |
25fe55af RS |
3530 | /* If last exactn not at current position. */ |
3531 | || pending_exact + *pending_exact + 1 != b | |
5e69f11e | 3532 | |
25fe55af | 3533 | /* We have only one byte following the exactn for the count. */ |
2d1675e4 | 3534 | || *pending_exact >= (1 << BYTEWIDTH) - MAX_MULTIBYTE_LENGTH |
fa9a63c5 | 3535 | |
25fe55af | 3536 | /* If followed by a repetition operator. */ |
9d99031f | 3537 | || (p != pend && (*p == '*' || *p == '^')) |
fa9a63c5 | 3538 | || ((syntax & RE_BK_PLUS_QM) |
9d99031f RS |
3539 | ? p + 1 < pend && *p == '\\' && (p[1] == '+' || p[1] == '?') |
3540 | : p != pend && (*p == '+' || *p == '?')) | |
fa9a63c5 | 3541 | || ((syntax & RE_INTERVALS) |
25fe55af | 3542 | && ((syntax & RE_NO_BK_BRACES) |
9d99031f RS |
3543 | ? p != pend && *p == '{' |
3544 | : p + 1 < pend && p[0] == '\\' && p[1] == '{'))) | |
fa9a63c5 RM |
3545 | { |
3546 | /* Start building a new exactn. */ | |
5e69f11e | 3547 | |
25fe55af | 3548 | laststart = b; |
fa9a63c5 RM |
3549 | |
3550 | BUF_PUSH_2 (exactn, 0); | |
3551 | pending_exact = b - 1; | |
25fe55af | 3552 | } |
5e69f11e | 3553 | |
2d1675e4 SM |
3554 | GET_BUFFER_SPACE (MAX_MULTIBYTE_LENGTH); |
3555 | { | |
e0277a47 KH |
3556 | int len; |
3557 | ||
36595814 | 3558 | c = TRANSLATE (c); |
e0277a47 KH |
3559 | if (multibyte) |
3560 | len = CHAR_STRING (c, b); | |
3561 | else | |
3562 | *b = c, len = 1; | |
2d1675e4 SM |
3563 | b += len; |
3564 | (*pending_exact) += len; | |
3565 | } | |
3566 | ||
fa9a63c5 | 3567 | break; |
25fe55af | 3568 | } /* switch (c) */ |
fa9a63c5 RM |
3569 | } /* while p != pend */ |
3570 | ||
5e69f11e | 3571 | |
fa9a63c5 | 3572 | /* Through the pattern now. */ |
5e69f11e | 3573 | |
505bde11 | 3574 | FIXUP_ALT_JUMP (); |
fa9a63c5 | 3575 | |
5e69f11e | 3576 | if (!COMPILE_STACK_EMPTY) |
fa9a63c5 RM |
3577 | FREE_STACK_RETURN (REG_EPAREN); |
3578 | ||
3579 | /* If we don't want backtracking, force success | |
3580 | the first time we reach the end of the compiled pattern. */ | |
3581 | if (syntax & RE_NO_POSIX_BACKTRACKING) | |
3582 | BUF_PUSH (succeed); | |
3583 | ||
fa9a63c5 RM |
3584 | /* We have succeeded; set the length of the buffer. */ |
3585 | bufp->used = b - bufp->buffer; | |
3586 | ||
3587 | #ifdef DEBUG | |
99633e97 | 3588 | if (debug > 0) |
fa9a63c5 | 3589 | { |
505bde11 | 3590 | re_compile_fastmap (bufp); |
fa9a63c5 RM |
3591 | DEBUG_PRINT1 ("\nCompiled pattern: \n"); |
3592 | print_compiled_pattern (bufp); | |
3593 | } | |
99633e97 | 3594 | debug--; |
fa9a63c5 RM |
3595 | #endif /* DEBUG */ |
3596 | ||
3597 | #ifndef MATCH_MAY_ALLOCATE | |
3598 | /* Initialize the failure stack to the largest possible stack. This | |
3599 | isn't necessary unless we're trying to avoid calling alloca in | |
3600 | the search and match routines. */ | |
3601 | { | |
3602 | int num_regs = bufp->re_nsub + 1; | |
3603 | ||
320a2a73 | 3604 | if (fail_stack.size < re_max_failures * TYPICAL_FAILURE_SIZE) |
fa9a63c5 | 3605 | { |
a26f4ccd | 3606 | fail_stack.size = re_max_failures * TYPICAL_FAILURE_SIZE; |
fa9a63c5 | 3607 | |
fa9a63c5 RM |
3608 | if (! fail_stack.stack) |
3609 | fail_stack.stack | |
5e69f11e | 3610 | = (fail_stack_elt_t *) malloc (fail_stack.size |
fa9a63c5 RM |
3611 | * sizeof (fail_stack_elt_t)); |
3612 | else | |
3613 | fail_stack.stack | |
3614 | = (fail_stack_elt_t *) realloc (fail_stack.stack, | |
3615 | (fail_stack.size | |
3616 | * sizeof (fail_stack_elt_t))); | |
fa9a63c5 RM |
3617 | } |
3618 | ||
3619 | regex_grow_registers (num_regs); | |
3620 | } | |
3621 | #endif /* not MATCH_MAY_ALLOCATE */ | |
3622 | ||
7c67c9f6 | 3623 | FREE_STACK_RETURN (REG_NOERROR); |
fa9a63c5 RM |
3624 | } /* regex_compile */ |
3625 | \f | |
3626 | /* Subroutines for `regex_compile'. */ | |
3627 | ||
25fe55af | 3628 | /* Store OP at LOC followed by two-byte integer parameter ARG. */ |
fa9a63c5 RM |
3629 | |
3630 | static void | |
3631 | store_op1 (op, loc, arg) | |
3632 | re_opcode_t op; | |
3633 | unsigned char *loc; | |
3634 | int arg; | |
3635 | { | |
3636 | *loc = (unsigned char) op; | |
3637 | STORE_NUMBER (loc + 1, arg); | |
3638 | } | |
3639 | ||
3640 | ||
3641 | /* Like `store_op1', but for two two-byte parameters ARG1 and ARG2. */ | |
3642 | ||
3643 | static void | |
3644 | store_op2 (op, loc, arg1, arg2) | |
3645 | re_opcode_t op; | |
3646 | unsigned char *loc; | |
3647 | int arg1, arg2; | |
3648 | { | |
3649 | *loc = (unsigned char) op; | |
3650 | STORE_NUMBER (loc + 1, arg1); | |
3651 | STORE_NUMBER (loc + 3, arg2); | |
3652 | } | |
3653 | ||
3654 | ||
3655 | /* Copy the bytes from LOC to END to open up three bytes of space at LOC | |
3656 | for OP followed by two-byte integer parameter ARG. */ | |
3657 | ||
3658 | static void | |
3659 | insert_op1 (op, loc, arg, end) | |
3660 | re_opcode_t op; | |
3661 | unsigned char *loc; | |
3662 | int arg; | |
5e69f11e | 3663 | unsigned char *end; |
fa9a63c5 RM |
3664 | { |
3665 | register unsigned char *pfrom = end; | |
3666 | register unsigned char *pto = end + 3; | |
3667 | ||
3668 | while (pfrom != loc) | |
3669 | *--pto = *--pfrom; | |
5e69f11e | 3670 | |
fa9a63c5 RM |
3671 | store_op1 (op, loc, arg); |
3672 | } | |
3673 | ||
3674 | ||
3675 | /* Like `insert_op1', but for two two-byte parameters ARG1 and ARG2. */ | |
3676 | ||
3677 | static void | |
3678 | insert_op2 (op, loc, arg1, arg2, end) | |
3679 | re_opcode_t op; | |
3680 | unsigned char *loc; | |
3681 | int arg1, arg2; | |
5e69f11e | 3682 | unsigned char *end; |
fa9a63c5 RM |
3683 | { |
3684 | register unsigned char *pfrom = end; | |
3685 | register unsigned char *pto = end + 5; | |
3686 | ||
3687 | while (pfrom != loc) | |
3688 | *--pto = *--pfrom; | |
5e69f11e | 3689 | |
fa9a63c5 RM |
3690 | store_op2 (op, loc, arg1, arg2); |
3691 | } | |
3692 | ||
3693 | ||
3694 | /* P points to just after a ^ in PATTERN. Return true if that ^ comes | |
3695 | after an alternative or a begin-subexpression. We assume there is at | |
3696 | least one character before the ^. */ | |
3697 | ||
3698 | static boolean | |
3699 | at_begline_loc_p (pattern, p, syntax) | |
01618498 | 3700 | re_char *pattern, *p; |
fa9a63c5 RM |
3701 | reg_syntax_t syntax; |
3702 | { | |
01618498 | 3703 | re_char *prev = p - 2; |
fa9a63c5 | 3704 | boolean prev_prev_backslash = prev > pattern && prev[-1] == '\\'; |
5e69f11e | 3705 | |
fa9a63c5 RM |
3706 | return |
3707 | /* After a subexpression? */ | |
3708 | (*prev == '(' && (syntax & RE_NO_BK_PARENS || prev_prev_backslash)) | |
25fe55af | 3709 | /* After an alternative? */ |
d2af47df SM |
3710 | || (*prev == '|' && (syntax & RE_NO_BK_VBAR || prev_prev_backslash)) |
3711 | /* After a shy subexpression? */ | |
3712 | || ((syntax & RE_SHY_GROUPS) && prev - 2 >= pattern | |
3713 | && prev[-1] == '?' && prev[-2] == '(' | |
3714 | && (syntax & RE_NO_BK_PARENS | |
3715 | || (prev - 3 >= pattern && prev[-3] == '\\'))); | |
fa9a63c5 RM |
3716 | } |
3717 | ||
3718 | ||
3719 | /* The dual of at_begline_loc_p. This one is for $. We assume there is | |
3720 | at least one character after the $, i.e., `P < PEND'. */ | |
3721 | ||
3722 | static boolean | |
3723 | at_endline_loc_p (p, pend, syntax) | |
01618498 | 3724 | re_char *p, *pend; |
99633e97 | 3725 | reg_syntax_t syntax; |
fa9a63c5 | 3726 | { |
01618498 | 3727 | re_char *next = p; |
fa9a63c5 | 3728 | boolean next_backslash = *next == '\\'; |
01618498 | 3729 | re_char *next_next = p + 1 < pend ? p + 1 : 0; |
5e69f11e | 3730 | |
fa9a63c5 RM |
3731 | return |
3732 | /* Before a subexpression? */ | |
3733 | (syntax & RE_NO_BK_PARENS ? *next == ')' | |
25fe55af | 3734 | : next_backslash && next_next && *next_next == ')') |
fa9a63c5 RM |
3735 | /* Before an alternative? */ |
3736 | || (syntax & RE_NO_BK_VBAR ? *next == '|' | |
25fe55af | 3737 | : next_backslash && next_next && *next_next == '|'); |
fa9a63c5 RM |
3738 | } |
3739 | ||
3740 | ||
5e69f11e | 3741 | /* Returns true if REGNUM is in one of COMPILE_STACK's elements and |
fa9a63c5 RM |
3742 | false if it's not. */ |
3743 | ||
3744 | static boolean | |
3745 | group_in_compile_stack (compile_stack, regnum) | |
3746 | compile_stack_type compile_stack; | |
3747 | regnum_t regnum; | |
3748 | { | |
3749 | int this_element; | |
3750 | ||
5e69f11e RM |
3751 | for (this_element = compile_stack.avail - 1; |
3752 | this_element >= 0; | |
fa9a63c5 RM |
3753 | this_element--) |
3754 | if (compile_stack.stack[this_element].regnum == regnum) | |
3755 | return true; | |
3756 | ||
3757 | return false; | |
3758 | } | |
fa9a63c5 | 3759 | \f |
f6a3f532 SM |
3760 | /* analyse_first. |
3761 | If fastmap is non-NULL, go through the pattern and fill fastmap | |
3762 | with all the possible leading chars. If fastmap is NULL, don't | |
3763 | bother filling it up (obviously) and only return whether the | |
3764 | pattern could potentially match the empty string. | |
3765 | ||
3766 | Return 1 if p..pend might match the empty string. | |
3767 | Return 0 if p..pend matches at least one char. | |
01618498 | 3768 | Return -1 if fastmap was not updated accurately. */ |
f6a3f532 SM |
3769 | |
3770 | static int | |
3771 | analyse_first (p, pend, fastmap, multibyte) | |
01618498 | 3772 | re_char *p, *pend; |
f6a3f532 SM |
3773 | char *fastmap; |
3774 | const int multibyte; | |
fa9a63c5 | 3775 | { |
505bde11 | 3776 | int j, k; |
1fb352e0 | 3777 | boolean not; |
fa9a63c5 | 3778 | |
b18215fc | 3779 | /* If all elements for base leading-codes in fastmap is set, this |
25fe55af | 3780 | flag is set true. */ |
b18215fc RS |
3781 | boolean match_any_multibyte_characters = false; |
3782 | ||
f6a3f532 | 3783 | assert (p); |
5e69f11e | 3784 | |
505bde11 SM |
3785 | /* The loop below works as follows: |
3786 | - It has a working-list kept in the PATTERN_STACK and which basically | |
3787 | starts by only containing a pointer to the first operation. | |
3788 | - If the opcode we're looking at is a match against some set of | |
3789 | chars, then we add those chars to the fastmap and go on to the | |
3790 | next work element from the worklist (done via `break'). | |
3791 | - If the opcode is a control operator on the other hand, we either | |
3792 | ignore it (if it's meaningless at this point, such as `start_memory') | |
3793 | or execute it (if it's a jump). If the jump has several destinations | |
3794 | (i.e. `on_failure_jump'), then we push the other destination onto the | |
3795 | worklist. | |
3796 | We guarantee termination by ignoring backward jumps (more or less), | |
3797 | so that `p' is monotonically increasing. More to the point, we | |
3798 | never set `p' (or push) anything `<= p1'. */ | |
3799 | ||
01618498 | 3800 | while (p < pend) |
fa9a63c5 | 3801 | { |
505bde11 SM |
3802 | /* `p1' is used as a marker of how far back a `on_failure_jump' |
3803 | can go without being ignored. It is normally equal to `p' | |
3804 | (which prevents any backward `on_failure_jump') except right | |
3805 | after a plain `jump', to allow patterns such as: | |
3806 | 0: jump 10 | |
3807 | 3..9: <body> | |
3808 | 10: on_failure_jump 3 | |
3809 | as used for the *? operator. */ | |
01618498 | 3810 | re_char *p1 = p; |
5e69f11e | 3811 | |
fa9a63c5 RM |
3812 | switch (SWITCH_ENUM_CAST ((re_opcode_t) *p++)) |
3813 | { | |
f6a3f532 | 3814 | case succeed: |
01618498 | 3815 | return 1; |
f6a3f532 | 3816 | continue; |
fa9a63c5 | 3817 | |
fa9a63c5 | 3818 | case duplicate: |
505bde11 SM |
3819 | /* If the first character has to match a backreference, that means |
3820 | that the group was empty (since it already matched). Since this | |
3821 | is the only case that interests us here, we can assume that the | |
3822 | backreference must match the empty string. */ | |
3823 | p++; | |
3824 | continue; | |
fa9a63c5 RM |
3825 | |
3826 | ||
3827 | /* Following are the cases which match a character. These end | |
25fe55af | 3828 | with `break'. */ |
fa9a63c5 RM |
3829 | |
3830 | case exactn: | |
e0277a47 KH |
3831 | if (fastmap) |
3832 | { | |
3833 | int c = RE_STRING_CHAR (p + 1, pend - p); | |
3834 | ||
3835 | if (SINGLE_BYTE_CHAR_P (c)) | |
3836 | fastmap[c] = 1; | |
3837 | else | |
3838 | fastmap[p[1]] = 1; | |
3839 | } | |
fa9a63c5 RM |
3840 | break; |
3841 | ||
3842 | ||
1fb352e0 SM |
3843 | case anychar: |
3844 | /* We could put all the chars except for \n (and maybe \0) | |
3845 | but we don't bother since it is generally not worth it. */ | |
f6a3f532 | 3846 | if (!fastmap) break; |
01618498 | 3847 | return -1; |
fa9a63c5 RM |
3848 | |
3849 | ||
b18215fc | 3850 | case charset_not: |
ba5c004d RS |
3851 | /* Chars beyond end of bitmap are possible matches. |
3852 | All the single-byte codes can occur in multibyte buffers. | |
3853 | So any that are not listed in the charset | |
3854 | are possible matches, even in multibyte buffers. */ | |
1fb352e0 | 3855 | if (!fastmap) break; |
b18215fc | 3856 | for (j = CHARSET_BITMAP_SIZE (&p[-1]) * BYTEWIDTH; |
1fb352e0 | 3857 | j < (1 << BYTEWIDTH); j++) |
b18215fc | 3858 | fastmap[j] = 1; |
1fb352e0 SM |
3859 | /* Fallthrough */ |
3860 | case charset: | |
3861 | if (!fastmap) break; | |
3862 | not = (re_opcode_t) *(p - 1) == charset_not; | |
b18215fc RS |
3863 | for (j = CHARSET_BITMAP_SIZE (&p[-1]) * BYTEWIDTH - 1, p++; |
3864 | j >= 0; j--) | |
1fb352e0 | 3865 | if (!!(p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH))) ^ not) |
b18215fc RS |
3866 | fastmap[j] = 1; |
3867 | ||
1fb352e0 SM |
3868 | if ((not && multibyte) |
3869 | /* Any character set can possibly contain a character | |
3870 | which doesn't match the specified set of characters. */ | |
3871 | || (CHARSET_RANGE_TABLE_EXISTS_P (&p[-2]) | |
3872 | && CHARSET_RANGE_TABLE_BITS (&p[-2]) != 0)) | |
3873 | /* If we can match a character class, we can match | |
3874 | any character set. */ | |
b18215fc RS |
3875 | { |
3876 | set_fastmap_for_multibyte_characters: | |
3877 | if (match_any_multibyte_characters == false) | |
3878 | { | |
3879 | for (j = 0x80; j < 0xA0; j++) /* XXX */ | |
3880 | if (BASE_LEADING_CODE_P (j)) | |
3881 | fastmap[j] = 1; | |
3882 | match_any_multibyte_characters = true; | |
3883 | } | |
3884 | } | |
b18215fc | 3885 | |
1fb352e0 SM |
3886 | else if (!not && CHARSET_RANGE_TABLE_EXISTS_P (&p[-2]) |
3887 | && match_any_multibyte_characters == false) | |
3888 | { | |
3889 | /* Set fastmap[I] 1 where I is a base leading code of each | |
3890 | multibyte character in the range table. */ | |
3891 | int c, count; | |
b18215fc | 3892 | |
1fb352e0 | 3893 | /* Make P points the range table. `+ 2' is to skip flag |
0b32bf0e | 3894 | bits for a character class. */ |
1fb352e0 | 3895 | p += CHARSET_BITMAP_SIZE (&p[-2]) + 2; |
b18215fc | 3896 | |
1fb352e0 SM |
3897 | /* Extract the number of ranges in range table into COUNT. */ |
3898 | EXTRACT_NUMBER_AND_INCR (count, p); | |
3899 | for (; count > 0; count--, p += 2 * 3) /* XXX */ | |
3900 | { | |
3901 | /* Extract the start of each range. */ | |
3902 | EXTRACT_CHARACTER (c, p); | |
3903 | j = CHAR_CHARSET (c); | |
3904 | fastmap[CHARSET_LEADING_CODE_BASE (j)] = 1; | |
3905 | } | |
3906 | } | |
b18215fc RS |
3907 | break; |
3908 | ||
1fb352e0 SM |
3909 | case syntaxspec: |
3910 | case notsyntaxspec: | |
3911 | if (!fastmap) break; | |
3912 | #ifndef emacs | |
3913 | not = (re_opcode_t)p[-1] == notsyntaxspec; | |
3914 | k = *p++; | |
3915 | for (j = 0; j < (1 << BYTEWIDTH); j++) | |
990b2375 | 3916 | if ((SYNTAX (j) == (enum syntaxcode) k) ^ not) |
b18215fc | 3917 | fastmap[j] = 1; |
b18215fc | 3918 | break; |
1fb352e0 | 3919 | #else /* emacs */ |
b18215fc RS |
3920 | /* This match depends on text properties. These end with |
3921 | aborting optimizations. */ | |
01618498 | 3922 | return -1; |
b18215fc RS |
3923 | |
3924 | case categoryspec: | |
b18215fc | 3925 | case notcategoryspec: |
1fb352e0 SM |
3926 | if (!fastmap) break; |
3927 | not = (re_opcode_t)p[-1] == notcategoryspec; | |
b18215fc | 3928 | k = *p++; |
1fb352e0 SM |
3929 | for (j = 0; j < (1 << BYTEWIDTH); j++) |
3930 | if ((CHAR_HAS_CATEGORY (j, k)) ^ not) | |
b18215fc RS |
3931 | fastmap[j] = 1; |
3932 | ||
1fb352e0 | 3933 | if (multibyte) |
b18215fc | 3934 | /* Any character set can possibly contain a character |
1fb352e0 | 3935 | whose category is K (or not). */ |
b18215fc RS |
3936 | goto set_fastmap_for_multibyte_characters; |
3937 | break; | |
3938 | ||
fa9a63c5 | 3939 | /* All cases after this match the empty string. These end with |
25fe55af | 3940 | `continue'. */ |
fa9a63c5 | 3941 | |
fa9a63c5 RM |
3942 | case before_dot: |
3943 | case at_dot: | |
3944 | case after_dot: | |
1fb352e0 | 3945 | #endif /* !emacs */ |
25fe55af RS |
3946 | case no_op: |
3947 | case begline: | |
3948 | case endline: | |
fa9a63c5 RM |
3949 | case begbuf: |
3950 | case endbuf: | |
3951 | case wordbound: | |
3952 | case notwordbound: | |
3953 | case wordbeg: | |
3954 | case wordend: | |
669fa600 SM |
3955 | case symbeg: |
3956 | case symend: | |
25fe55af | 3957 | continue; |
fa9a63c5 RM |
3958 | |
3959 | ||
fa9a63c5 | 3960 | case jump: |
25fe55af | 3961 | EXTRACT_NUMBER_AND_INCR (j, p); |
505bde11 SM |
3962 | if (j < 0) |
3963 | /* Backward jumps can only go back to code that we've already | |
3964 | visited. `re_compile' should make sure this is true. */ | |
3965 | break; | |
25fe55af | 3966 | p += j; |
505bde11 SM |
3967 | switch (SWITCH_ENUM_CAST ((re_opcode_t) *p)) |
3968 | { | |
3969 | case on_failure_jump: | |
3970 | case on_failure_keep_string_jump: | |
505bde11 | 3971 | case on_failure_jump_loop: |
0683b6fa | 3972 | case on_failure_jump_nastyloop: |
505bde11 SM |
3973 | case on_failure_jump_smart: |
3974 | p++; | |
3975 | break; | |
3976 | default: | |
3977 | continue; | |
3978 | }; | |
3979 | /* Keep `p1' to allow the `on_failure_jump' we are jumping to | |
3980 | to jump back to "just after here". */ | |
3981 | /* Fallthrough */ | |
fa9a63c5 | 3982 | |
25fe55af RS |
3983 | case on_failure_jump: |
3984 | case on_failure_keep_string_jump: | |
0683b6fa | 3985 | case on_failure_jump_nastyloop: |
505bde11 SM |
3986 | case on_failure_jump_loop: |
3987 | case on_failure_jump_smart: | |
25fe55af | 3988 | EXTRACT_NUMBER_AND_INCR (j, p); |
505bde11 | 3989 | if (p + j <= p1) |
ed0767d8 | 3990 | ; /* Backward jump to be ignored. */ |
01618498 SM |
3991 | else |
3992 | { /* We have to look down both arms. | |
3993 | We first go down the "straight" path so as to minimize | |
3994 | stack usage when going through alternatives. */ | |
3995 | int r = analyse_first (p, pend, fastmap, multibyte); | |
3996 | if (r) return r; | |
3997 | p += j; | |
3998 | } | |
25fe55af | 3999 | continue; |
fa9a63c5 RM |
4000 | |
4001 | ||
ed0767d8 SM |
4002 | case jump_n: |
4003 | /* This code simply does not properly handle forward jump_n. */ | |
4004 | DEBUG_STATEMENT (EXTRACT_NUMBER (j, p); assert (j < 0)); | |
4005 | p += 4; | |
4006 | /* jump_n can either jump or fall through. The (backward) jump | |
4007 | case has already been handled, so we only need to look at the | |
4008 | fallthrough case. */ | |
4009 | continue; | |
177c0ea7 | 4010 | |
fa9a63c5 | 4011 | case succeed_n: |
ed0767d8 SM |
4012 | /* If N == 0, it should be an on_failure_jump_loop instead. */ |
4013 | DEBUG_STATEMENT (EXTRACT_NUMBER (j, p + 2); assert (j > 0)); | |
4014 | p += 4; | |
4015 | /* We only care about one iteration of the loop, so we don't | |
4016 | need to consider the case where this behaves like an | |
4017 | on_failure_jump. */ | |
25fe55af | 4018 | continue; |
fa9a63c5 RM |
4019 | |
4020 | ||
4021 | case set_number_at: | |
25fe55af RS |
4022 | p += 4; |
4023 | continue; | |
fa9a63c5 RM |
4024 | |
4025 | ||
4026 | case start_memory: | |
25fe55af | 4027 | case stop_memory: |
505bde11 | 4028 | p += 1; |
fa9a63c5 RM |
4029 | continue; |
4030 | ||
4031 | ||
4032 | default: | |
25fe55af RS |
4033 | abort (); /* We have listed all the cases. */ |
4034 | } /* switch *p++ */ | |
fa9a63c5 RM |
4035 | |
4036 | /* Getting here means we have found the possible starting | |
25fe55af | 4037 | characters for one path of the pattern -- and that the empty |
01618498 SM |
4038 | string does not match. We need not follow this path further. */ |
4039 | return 0; | |
fa9a63c5 RM |
4040 | } /* while p */ |
4041 | ||
01618498 SM |
4042 | /* We reached the end without matching anything. */ |
4043 | return 1; | |
4044 | ||
f6a3f532 SM |
4045 | } /* analyse_first */ |
4046 | \f | |
4047 | /* re_compile_fastmap computes a ``fastmap'' for the compiled pattern in | |
4048 | BUFP. A fastmap records which of the (1 << BYTEWIDTH) possible | |
4049 | characters can start a string that matches the pattern. This fastmap | |
4050 | is used by re_search to skip quickly over impossible starting points. | |
4051 | ||
4052 | Character codes above (1 << BYTEWIDTH) are not represented in the | |
4053 | fastmap, but the leading codes are represented. Thus, the fastmap | |
4054 | indicates which character sets could start a match. | |
4055 | ||
4056 | The caller must supply the address of a (1 << BYTEWIDTH)-byte data | |
4057 | area as BUFP->fastmap. | |
4058 | ||
4059 | We set the `fastmap', `fastmap_accurate', and `can_be_null' fields in | |
4060 | the pattern buffer. | |
4061 | ||
4062 | Returns 0 if we succeed, -2 if an internal error. */ | |
4063 | ||
4064 | int | |
4065 | re_compile_fastmap (bufp) | |
4066 | struct re_pattern_buffer *bufp; | |
4067 | { | |
4068 | char *fastmap = bufp->fastmap; | |
4069 | int analysis; | |
4070 | ||
4071 | assert (fastmap && bufp->buffer); | |
4072 | ||
4073 | bzero (fastmap, 1 << BYTEWIDTH); /* Assume nothing's valid. */ | |
4074 | bufp->fastmap_accurate = 1; /* It will be when we're done. */ | |
4075 | ||
4076 | analysis = analyse_first (bufp->buffer, bufp->buffer + bufp->used, | |
2d1675e4 | 4077 | fastmap, RE_MULTIBYTE_P (bufp)); |
c0f9ea08 | 4078 | bufp->can_be_null = (analysis != 0); |
fa9a63c5 RM |
4079 | return 0; |
4080 | } /* re_compile_fastmap */ | |
4081 | \f | |
4082 | /* Set REGS to hold NUM_REGS registers, storing them in STARTS and | |
4083 | ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use | |
4084 | this memory for recording register information. STARTS and ENDS | |
4085 | must be allocated using the malloc library routine, and must each | |
4086 | be at least NUM_REGS * sizeof (regoff_t) bytes long. | |
4087 | ||
4088 | If NUM_REGS == 0, then subsequent matches should allocate their own | |
4089 | register data. | |
4090 | ||
4091 | Unless this function is called, the first search or match using | |
4092 | PATTERN_BUFFER will allocate its own register data, without | |
4093 | freeing the old data. */ | |
4094 | ||
4095 | void | |
4096 | re_set_registers (bufp, regs, num_regs, starts, ends) | |
4097 | struct re_pattern_buffer *bufp; | |
4098 | struct re_registers *regs; | |
4099 | unsigned num_regs; | |
4100 | regoff_t *starts, *ends; | |
4101 | { | |
4102 | if (num_regs) | |
4103 | { | |
4104 | bufp->regs_allocated = REGS_REALLOCATE; | |
4105 | regs->num_regs = num_regs; | |
4106 | regs->start = starts; | |
4107 | regs->end = ends; | |
4108 | } | |
4109 | else | |
4110 | { | |
4111 | bufp->regs_allocated = REGS_UNALLOCATED; | |
4112 | regs->num_regs = 0; | |
4113 | regs->start = regs->end = (regoff_t *) 0; | |
4114 | } | |
4115 | } | |
c0f9ea08 | 4116 | WEAK_ALIAS (__re_set_registers, re_set_registers) |
fa9a63c5 | 4117 | \f |
25fe55af | 4118 | /* Searching routines. */ |
fa9a63c5 RM |
4119 | |
4120 | /* Like re_search_2, below, but only one string is specified, and | |
4121 | doesn't let you say where to stop matching. */ | |
4122 | ||
4123 | int | |
4124 | re_search (bufp, string, size, startpos, range, regs) | |
4125 | struct re_pattern_buffer *bufp; | |
4126 | const char *string; | |
4127 | int size, startpos, range; | |
4128 | struct re_registers *regs; | |
4129 | { | |
5e69f11e | 4130 | return re_search_2 (bufp, NULL, 0, string, size, startpos, range, |
fa9a63c5 RM |
4131 | regs, size); |
4132 | } | |
c0f9ea08 | 4133 | WEAK_ALIAS (__re_search, re_search) |
fa9a63c5 | 4134 | |
70806df6 KH |
4135 | /* Head address of virtual concatenation of string. */ |
4136 | #define HEAD_ADDR_VSTRING(P) \ | |
4137 | (((P) >= size1 ? string2 : string1)) | |
4138 | ||
b18215fc RS |
4139 | /* End address of virtual concatenation of string. */ |
4140 | #define STOP_ADDR_VSTRING(P) \ | |
4141 | (((P) >= size1 ? string2 + size2 : string1 + size1)) | |
4142 | ||
4143 | /* Address of POS in the concatenation of virtual string. */ | |
4144 | #define POS_ADDR_VSTRING(POS) \ | |
4145 | (((POS) >= size1 ? string2 - size1 : string1) + (POS)) | |
fa9a63c5 RM |
4146 | |
4147 | /* Using the compiled pattern in BUFP->buffer, first tries to match the | |
4148 | virtual concatenation of STRING1 and STRING2, starting first at index | |
4149 | STARTPOS, then at STARTPOS + 1, and so on. | |
5e69f11e | 4150 | |
fa9a63c5 | 4151 | STRING1 and STRING2 have length SIZE1 and SIZE2, respectively. |
5e69f11e | 4152 | |
fa9a63c5 RM |
4153 | RANGE is how far to scan while trying to match. RANGE = 0 means try |
4154 | only at STARTPOS; in general, the last start tried is STARTPOS + | |
4155 | RANGE. | |
5e69f11e | 4156 | |
fa9a63c5 RM |
4157 | In REGS, return the indices of the virtual concatenation of STRING1 |
4158 | and STRING2 that matched the entire BUFP->buffer and its contained | |
4159 | subexpressions. | |
5e69f11e | 4160 | |
fa9a63c5 RM |
4161 | Do not consider matching one past the index STOP in the virtual |
4162 | concatenation of STRING1 and STRING2. | |
4163 | ||
4164 | We return either the position in the strings at which the match was | |
4165 | found, -1 if no match, or -2 if error (such as failure | |
4166 | stack overflow). */ | |
4167 | ||
4168 | int | |
66f0296e | 4169 | re_search_2 (bufp, str1, size1, str2, size2, startpos, range, regs, stop) |
fa9a63c5 | 4170 | struct re_pattern_buffer *bufp; |
66f0296e | 4171 | const char *str1, *str2; |
fa9a63c5 RM |
4172 | int size1, size2; |
4173 | int startpos; | |
4174 | int range; | |
4175 | struct re_registers *regs; | |
4176 | int stop; | |
4177 | { | |
4178 | int val; | |
66f0296e SM |
4179 | re_char *string1 = (re_char*) str1; |
4180 | re_char *string2 = (re_char*) str2; | |
fa9a63c5 | 4181 | register char *fastmap = bufp->fastmap; |
6676cb1c | 4182 | register RE_TRANSLATE_TYPE translate = bufp->translate; |
fa9a63c5 RM |
4183 | int total_size = size1 + size2; |
4184 | int endpos = startpos + range; | |
c0f9ea08 | 4185 | boolean anchored_start; |
fa9a63c5 | 4186 | |
25fe55af | 4187 | /* Nonzero if we have to concern multibyte character. */ |
2d1675e4 | 4188 | const boolean multibyte = RE_MULTIBYTE_P (bufp); |
b18215fc | 4189 | |
fa9a63c5 RM |
4190 | /* Check for out-of-range STARTPOS. */ |
4191 | if (startpos < 0 || startpos > total_size) | |
4192 | return -1; | |
5e69f11e | 4193 | |
fa9a63c5 | 4194 | /* Fix up RANGE if it might eventually take us outside |
34597fa9 | 4195 | the virtual concatenation of STRING1 and STRING2. |
5e69f11e | 4196 | Make sure we won't move STARTPOS below 0 or above TOTAL_SIZE. */ |
34597fa9 RS |
4197 | if (endpos < 0) |
4198 | range = 0 - startpos; | |
fa9a63c5 RM |
4199 | else if (endpos > total_size) |
4200 | range = total_size - startpos; | |
4201 | ||
4202 | /* If the search isn't to be a backwards one, don't waste time in a | |
7b140fd7 | 4203 | search for a pattern anchored at beginning of buffer. */ |
fa9a63c5 RM |
4204 | if (bufp->used > 0 && (re_opcode_t) bufp->buffer[0] == begbuf && range > 0) |
4205 | { | |
4206 | if (startpos > 0) | |
4207 | return -1; | |
4208 | else | |
7b140fd7 | 4209 | range = 0; |
fa9a63c5 RM |
4210 | } |
4211 | ||
ae4788a8 RS |
4212 | #ifdef emacs |
4213 | /* In a forward search for something that starts with \=. | |
4214 | don't keep searching past point. */ | |
4215 | if (bufp->used > 0 && (re_opcode_t) bufp->buffer[0] == at_dot && range > 0) | |
4216 | { | |
7b140fd7 RS |
4217 | range = PT_BYTE - BEGV_BYTE - startpos; |
4218 | if (range < 0) | |
ae4788a8 RS |
4219 | return -1; |
4220 | } | |
4221 | #endif /* emacs */ | |
4222 | ||
fa9a63c5 RM |
4223 | /* Update the fastmap now if not correct already. */ |
4224 | if (fastmap && !bufp->fastmap_accurate) | |
01618498 | 4225 | re_compile_fastmap (bufp); |
5e69f11e | 4226 | |
c8499ba5 | 4227 | /* See whether the pattern is anchored. */ |
c0f9ea08 | 4228 | anchored_start = (bufp->buffer[0] == begline); |
c8499ba5 | 4229 | |
b18215fc | 4230 | #ifdef emacs |
cc9b4df2 KH |
4231 | gl_state.object = re_match_object; |
4232 | { | |
99633e97 | 4233 | int charpos = SYNTAX_TABLE_BYTE_TO_CHAR (POS_AS_IN_BUFFER (startpos)); |
cc9b4df2 KH |
4234 | |
4235 | SETUP_SYNTAX_TABLE_FOR_OBJECT (re_match_object, charpos, 1); | |
4236 | } | |
b18215fc RS |
4237 | #endif |
4238 | ||
fa9a63c5 RM |
4239 | /* Loop through the string, looking for a place to start matching. */ |
4240 | for (;;) | |
5e69f11e | 4241 | { |
c8499ba5 RS |
4242 | /* If the pattern is anchored, |
4243 | skip quickly past places we cannot match. | |
4244 | We don't bother to treat startpos == 0 specially | |
4245 | because that case doesn't repeat. */ | |
4246 | if (anchored_start && startpos > 0) | |
4247 | { | |
c0f9ea08 SM |
4248 | if (! ((startpos <= size1 ? string1[startpos - 1] |
4249 | : string2[startpos - size1 - 1]) | |
4250 | == '\n')) | |
c8499ba5 RS |
4251 | goto advance; |
4252 | } | |
4253 | ||
fa9a63c5 | 4254 | /* If a fastmap is supplied, skip quickly over characters that |
25fe55af RS |
4255 | cannot be the start of a match. If the pattern can match the |
4256 | null string, however, we don't need to skip characters; we want | |
4257 | the first null string. */ | |
fa9a63c5 RM |
4258 | if (fastmap && startpos < total_size && !bufp->can_be_null) |
4259 | { | |
66f0296e | 4260 | register re_char *d; |
01618498 | 4261 | register re_wchar_t buf_ch; |
e934739e RS |
4262 | |
4263 | d = POS_ADDR_VSTRING (startpos); | |
4264 | ||
25fe55af | 4265 | if (range > 0) /* Searching forwards. */ |
fa9a63c5 | 4266 | { |
fa9a63c5 RM |
4267 | register int lim = 0; |
4268 | int irange = range; | |
4269 | ||
25fe55af RS |
4270 | if (startpos < size1 && startpos + range >= size1) |
4271 | lim = range - (size1 - startpos); | |
fa9a63c5 | 4272 | |
25fe55af RS |
4273 | /* Written out as an if-else to avoid testing `translate' |
4274 | inside the loop. */ | |
28ae27ae AS |
4275 | if (RE_TRANSLATE_P (translate)) |
4276 | { | |
e934739e RS |
4277 | if (multibyte) |
4278 | while (range > lim) | |
4279 | { | |
4280 | int buf_charlen; | |
4281 | ||
4282 | buf_ch = STRING_CHAR_AND_LENGTH (d, range - lim, | |
4283 | buf_charlen); | |
4284 | ||
4285 | buf_ch = RE_TRANSLATE (translate, buf_ch); | |
4286 | if (buf_ch >= 0400 | |
4287 | || fastmap[buf_ch]) | |
4288 | break; | |
4289 | ||
4290 | range -= buf_charlen; | |
4291 | d += buf_charlen; | |
4292 | } | |
4293 | else | |
cf1982d9 EZ |
4294 | { |
4295 | /* Convert *d to integer to shut up GCC's | |
4296 | whining about comparison that is always | |
4297 | true. */ | |
4298 | int di = *d; | |
4299 | ||
4300 | while (range > lim | |
4301 | && !fastmap[RE_TRANSLATE (translate, di)]) | |
4302 | { | |
4303 | di = *(++d); | |
4304 | range--; | |
4305 | } | |
4306 | } | |
e934739e | 4307 | } |
fa9a63c5 | 4308 | else |
66f0296e | 4309 | while (range > lim && !fastmap[*d]) |
33c46939 RS |
4310 | { |
4311 | d++; | |
4312 | range--; | |
4313 | } | |
fa9a63c5 RM |
4314 | |
4315 | startpos += irange - range; | |
4316 | } | |
25fe55af | 4317 | else /* Searching backwards. */ |
fa9a63c5 | 4318 | { |
2d1675e4 SM |
4319 | int room = (startpos >= size1 |
4320 | ? size2 + size1 - startpos | |
4321 | : size1 - startpos); | |
4322 | buf_ch = RE_STRING_CHAR (d, room); | |
4323 | buf_ch = TRANSLATE (buf_ch); | |
fa9a63c5 | 4324 | |
e934739e RS |
4325 | if (! (buf_ch >= 0400 |
4326 | || fastmap[buf_ch])) | |
fa9a63c5 RM |
4327 | goto advance; |
4328 | } | |
4329 | } | |
4330 | ||
4331 | /* If can't match the null string, and that's all we have left, fail. */ | |
4332 | if (range >= 0 && startpos == total_size && fastmap | |
25fe55af | 4333 | && !bufp->can_be_null) |
fa9a63c5 RM |
4334 | return -1; |
4335 | ||
4336 | val = re_match_2_internal (bufp, string1, size1, string2, size2, | |
4337 | startpos, regs, stop); | |
4338 | #ifndef REGEX_MALLOC | |
0b32bf0e | 4339 | # ifdef C_ALLOCA |
fa9a63c5 | 4340 | alloca (0); |
0b32bf0e | 4341 | # endif |
fa9a63c5 RM |
4342 | #endif |
4343 | ||
4344 | if (val >= 0) | |
4345 | return startpos; | |
5e69f11e | 4346 | |
fa9a63c5 RM |
4347 | if (val == -2) |
4348 | return -2; | |
4349 | ||
4350 | advance: | |
5e69f11e | 4351 | if (!range) |
25fe55af | 4352 | break; |
5e69f11e | 4353 | else if (range > 0) |
25fe55af | 4354 | { |
b18215fc RS |
4355 | /* Update STARTPOS to the next character boundary. */ |
4356 | if (multibyte) | |
4357 | { | |
66f0296e SM |
4358 | re_char *p = POS_ADDR_VSTRING (startpos); |
4359 | re_char *pend = STOP_ADDR_VSTRING (startpos); | |
b18215fc RS |
4360 | int len = MULTIBYTE_FORM_LENGTH (p, pend - p); |
4361 | ||
4362 | range -= len; | |
4363 | if (range < 0) | |
4364 | break; | |
4365 | startpos += len; | |
4366 | } | |
4367 | else | |
4368 | { | |
b560c397 RS |
4369 | range--; |
4370 | startpos++; | |
4371 | } | |
e318085a | 4372 | } |
fa9a63c5 | 4373 | else |
25fe55af RS |
4374 | { |
4375 | range++; | |
4376 | startpos--; | |
b18215fc RS |
4377 | |
4378 | /* Update STARTPOS to the previous character boundary. */ | |
4379 | if (multibyte) | |
4380 | { | |
70806df6 KH |
4381 | re_char *p = POS_ADDR_VSTRING (startpos) + 1; |
4382 | re_char *p0 = p; | |
4383 | re_char *phead = HEAD_ADDR_VSTRING (startpos); | |
b18215fc RS |
4384 | |
4385 | /* Find the head of multibyte form. */ | |
70806df6 KH |
4386 | PREV_CHAR_BOUNDARY (p, phead); |
4387 | range += p0 - 1 - p; | |
4388 | if (range > 0) | |
4389 | break; | |
b18215fc | 4390 | |
70806df6 | 4391 | startpos -= p0 - 1 - p; |
b18215fc | 4392 | } |
25fe55af | 4393 | } |
fa9a63c5 RM |
4394 | } |
4395 | return -1; | |
4396 | } /* re_search_2 */ | |
c0f9ea08 | 4397 | WEAK_ALIAS (__re_search_2, re_search_2) |
fa9a63c5 RM |
4398 | \f |
4399 | /* Declarations and macros for re_match_2. */ | |
4400 | ||
2d1675e4 SM |
4401 | static int bcmp_translate _RE_ARGS((re_char *s1, re_char *s2, |
4402 | register int len, | |
4403 | RE_TRANSLATE_TYPE translate, | |
4404 | const int multibyte)); | |
fa9a63c5 RM |
4405 | |
4406 | /* This converts PTR, a pointer into one of the search strings `string1' | |
4407 | and `string2' into an offset from the beginning of that string. */ | |
4408 | #define POINTER_TO_OFFSET(ptr) \ | |
4409 | (FIRST_STRING_P (ptr) \ | |
4410 | ? ((regoff_t) ((ptr) - string1)) \ | |
4411 | : ((regoff_t) ((ptr) - string2 + size1))) | |
4412 | ||
fa9a63c5 | 4413 | /* Call before fetching a character with *d. This switches over to |
419d1c74 SM |
4414 | string2 if necessary. |
4415 | Check re_match_2_internal for a discussion of why end_match_2 might | |
4416 | not be within string2 (but be equal to end_match_1 instead). */ | |
fa9a63c5 | 4417 | #define PREFETCH() \ |
25fe55af | 4418 | while (d == dend) \ |
fa9a63c5 RM |
4419 | { \ |
4420 | /* End of string2 => fail. */ \ | |
25fe55af RS |
4421 | if (dend == end_match_2) \ |
4422 | goto fail; \ | |
4bb91c68 | 4423 | /* End of string1 => advance to string2. */ \ |
25fe55af | 4424 | d = string2; \ |
fa9a63c5 RM |
4425 | dend = end_match_2; \ |
4426 | } | |
4427 | ||
f1ad044f SM |
4428 | /* Call before fetching a char with *d if you already checked other limits. |
4429 | This is meant for use in lookahead operations like wordend, etc.. | |
4430 | where we might need to look at parts of the string that might be | |
4431 | outside of the LIMITs (i.e past `stop'). */ | |
4432 | #define PREFETCH_NOLIMIT() \ | |
4433 | if (d == end1) \ | |
4434 | { \ | |
4435 | d = string2; \ | |
4436 | dend = end_match_2; \ | |
4437 | } \ | |
fa9a63c5 RM |
4438 | |
4439 | /* Test if at very beginning or at very end of the virtual concatenation | |
25fe55af | 4440 | of `string1' and `string2'. If only one string, it's `string2'. */ |
fa9a63c5 | 4441 | #define AT_STRINGS_BEG(d) ((d) == (size1 ? string1 : string2) || !size2) |
5e69f11e | 4442 | #define AT_STRINGS_END(d) ((d) == end2) |
fa9a63c5 RM |
4443 | |
4444 | ||
4445 | /* Test if D points to a character which is word-constituent. We have | |
4446 | two special cases to check for: if past the end of string1, look at | |
4447 | the first character in string2; and if before the beginning of | |
4448 | string2, look at the last character in string1. */ | |
4449 | #define WORDCHAR_P(d) \ | |
4450 | (SYNTAX ((d) == end1 ? *string2 \ | |
25fe55af | 4451 | : (d) == string2 - 1 ? *(end1 - 1) : *(d)) \ |
fa9a63c5 RM |
4452 | == Sword) |
4453 | ||
9121ca40 | 4454 | /* Disabled due to a compiler bug -- see comment at case wordbound */ |
b18215fc RS |
4455 | |
4456 | /* The comment at case wordbound is following one, but we don't use | |
4457 | AT_WORD_BOUNDARY anymore to support multibyte form. | |
4458 | ||
4459 | The DEC Alpha C compiler 3.x generates incorrect code for the | |
25fe55af RS |
4460 | test WORDCHAR_P (d - 1) != WORDCHAR_P (d) in the expansion of |
4461 | AT_WORD_BOUNDARY, so this code is disabled. Expanding the | |
b18215fc RS |
4462 | macro and introducing temporary variables works around the bug. */ |
4463 | ||
9121ca40 | 4464 | #if 0 |
fa9a63c5 RM |
4465 | /* Test if the character before D and the one at D differ with respect |
4466 | to being word-constituent. */ | |
4467 | #define AT_WORD_BOUNDARY(d) \ | |
4468 | (AT_STRINGS_BEG (d) || AT_STRINGS_END (d) \ | |
4469 | || WORDCHAR_P (d - 1) != WORDCHAR_P (d)) | |
9121ca40 | 4470 | #endif |
fa9a63c5 RM |
4471 | |
4472 | /* Free everything we malloc. */ | |
4473 | #ifdef MATCH_MAY_ALLOCATE | |
0b32bf0e SM |
4474 | # define FREE_VAR(var) if (var) { REGEX_FREE (var); var = NULL; } else |
4475 | # define FREE_VARIABLES() \ | |
fa9a63c5 RM |
4476 | do { \ |
4477 | REGEX_FREE_STACK (fail_stack.stack); \ | |
4478 | FREE_VAR (regstart); \ | |
4479 | FREE_VAR (regend); \ | |
fa9a63c5 RM |
4480 | FREE_VAR (best_regstart); \ |
4481 | FREE_VAR (best_regend); \ | |
fa9a63c5 RM |
4482 | } while (0) |
4483 | #else | |
0b32bf0e | 4484 | # define FREE_VARIABLES() ((void)0) /* Do nothing! But inhibit gcc warning. */ |
fa9a63c5 RM |
4485 | #endif /* not MATCH_MAY_ALLOCATE */ |
4486 | ||
505bde11 SM |
4487 | \f |
4488 | /* Optimization routines. */ | |
4489 | ||
4e8a9132 SM |
4490 | /* If the operation is a match against one or more chars, |
4491 | return a pointer to the next operation, else return NULL. */ | |
01618498 | 4492 | static re_char * |
4e8a9132 | 4493 | skip_one_char (p) |
01618498 | 4494 | re_char *p; |
4e8a9132 SM |
4495 | { |
4496 | switch (SWITCH_ENUM_CAST (*p++)) | |
4497 | { | |
4498 | case anychar: | |
4499 | break; | |
177c0ea7 | 4500 | |
4e8a9132 SM |
4501 | case exactn: |
4502 | p += *p + 1; | |
4503 | break; | |
4504 | ||
4505 | case charset_not: | |
4506 | case charset: | |
4507 | if (CHARSET_RANGE_TABLE_EXISTS_P (p - 1)) | |
4508 | { | |
4509 | int mcnt; | |
4510 | p = CHARSET_RANGE_TABLE (p - 1); | |
4511 | EXTRACT_NUMBER_AND_INCR (mcnt, p); | |
4512 | p = CHARSET_RANGE_TABLE_END (p, mcnt); | |
4513 | } | |
4514 | else | |
4515 | p += 1 + CHARSET_BITMAP_SIZE (p - 1); | |
4516 | break; | |
177c0ea7 | 4517 | |
4e8a9132 SM |
4518 | case syntaxspec: |
4519 | case notsyntaxspec: | |
1fb352e0 | 4520 | #ifdef emacs |
4e8a9132 SM |
4521 | case categoryspec: |
4522 | case notcategoryspec: | |
4523 | #endif /* emacs */ | |
4524 | p++; | |
4525 | break; | |
4526 | ||
4527 | default: | |
4528 | p = NULL; | |
4529 | } | |
4530 | return p; | |
4531 | } | |
4532 | ||
4533 | ||
505bde11 | 4534 | /* Jump over non-matching operations. */ |
275a3409 | 4535 | static re_char * |
4e8a9132 | 4536 | skip_noops (p, pend) |
275a3409 | 4537 | re_char *p, *pend; |
505bde11 SM |
4538 | { |
4539 | int mcnt; | |
4540 | while (p < pend) | |
4541 | { | |
4542 | switch (SWITCH_ENUM_CAST ((re_opcode_t) *p)) | |
4543 | { | |
4544 | case start_memory: | |
505bde11 SM |
4545 | case stop_memory: |
4546 | p += 2; break; | |
4547 | case no_op: | |
4548 | p += 1; break; | |
4549 | case jump: | |
4550 | p += 1; | |
4551 | EXTRACT_NUMBER_AND_INCR (mcnt, p); | |
4552 | p += mcnt; | |
4553 | break; | |
4554 | default: | |
4555 | return p; | |
4556 | } | |
4557 | } | |
4558 | assert (p == pend); | |
4559 | return p; | |
4560 | } | |
4561 | ||
4562 | /* Non-zero if "p1 matches something" implies "p2 fails". */ | |
4563 | static int | |
4564 | mutually_exclusive_p (bufp, p1, p2) | |
4565 | struct re_pattern_buffer *bufp; | |
275a3409 | 4566 | re_char *p1, *p2; |
505bde11 | 4567 | { |
4e8a9132 | 4568 | re_opcode_t op2; |
2d1675e4 | 4569 | const boolean multibyte = RE_MULTIBYTE_P (bufp); |
505bde11 SM |
4570 | unsigned char *pend = bufp->buffer + bufp->used; |
4571 | ||
4e8a9132 | 4572 | assert (p1 >= bufp->buffer && p1 < pend |
505bde11 SM |
4573 | && p2 >= bufp->buffer && p2 <= pend); |
4574 | ||
4575 | /* Skip over open/close-group commands. | |
4576 | If what follows this loop is a ...+ construct, | |
4577 | look at what begins its body, since we will have to | |
4578 | match at least one of that. */ | |
4e8a9132 SM |
4579 | p2 = skip_noops (p2, pend); |
4580 | /* The same skip can be done for p1, except that this function | |
4581 | is only used in the case where p1 is a simple match operator. */ | |
4582 | /* p1 = skip_noops (p1, pend); */ | |
4583 | ||
4584 | assert (p1 >= bufp->buffer && p1 < pend | |
4585 | && p2 >= bufp->buffer && p2 <= pend); | |
4586 | ||
4587 | op2 = p2 == pend ? succeed : *p2; | |
4588 | ||
4589 | switch (SWITCH_ENUM_CAST (op2)) | |
505bde11 | 4590 | { |
4e8a9132 SM |
4591 | case succeed: |
4592 | case endbuf: | |
4593 | /* If we're at the end of the pattern, we can change. */ | |
4594 | if (skip_one_char (p1)) | |
505bde11 | 4595 | { |
505bde11 SM |
4596 | DEBUG_PRINT1 (" End of pattern: fast loop.\n"); |
4597 | return 1; | |
505bde11 | 4598 | } |
4e8a9132 | 4599 | break; |
177c0ea7 | 4600 | |
4e8a9132 | 4601 | case endline: |
4e8a9132 SM |
4602 | case exactn: |
4603 | { | |
01618498 | 4604 | register re_wchar_t c |
4e8a9132 | 4605 | = (re_opcode_t) *p2 == endline ? '\n' |
411e4203 | 4606 | : RE_STRING_CHAR (p2 + 2, pend - p2 - 2); |
505bde11 | 4607 | |
4e8a9132 SM |
4608 | if ((re_opcode_t) *p1 == exactn) |
4609 | { | |
4610 | if (c != RE_STRING_CHAR (p1 + 2, pend - p1 - 2)) | |
4611 | { | |
4612 | DEBUG_PRINT3 (" '%c' != '%c' => fast loop.\n", c, p1[2]); | |
4613 | return 1; | |
4614 | } | |
4615 | } | |
505bde11 | 4616 | |
4e8a9132 SM |
4617 | else if ((re_opcode_t) *p1 == charset |
4618 | || (re_opcode_t) *p1 == charset_not) | |
4619 | { | |
4620 | int not = (re_opcode_t) *p1 == charset_not; | |
505bde11 | 4621 | |
4e8a9132 SM |
4622 | /* Test if C is listed in charset (or charset_not) |
4623 | at `p1'. */ | |
4624 | if (SINGLE_BYTE_CHAR_P (c)) | |
4625 | { | |
4626 | if (c < CHARSET_BITMAP_SIZE (p1) * BYTEWIDTH | |
4627 | && p1[2 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH))) | |
4628 | not = !not; | |
4629 | } | |
4630 | else if (CHARSET_RANGE_TABLE_EXISTS_P (p1)) | |
4631 | CHARSET_LOOKUP_RANGE_TABLE (not, c, p1); | |
505bde11 | 4632 | |
4e8a9132 SM |
4633 | /* `not' is equal to 1 if c would match, which means |
4634 | that we can't change to pop_failure_jump. */ | |
4635 | if (!not) | |
4636 | { | |
4637 | DEBUG_PRINT1 (" No match => fast loop.\n"); | |
4638 | return 1; | |
4639 | } | |
4640 | } | |
4641 | else if ((re_opcode_t) *p1 == anychar | |
4642 | && c == '\n') | |
4643 | { | |
4644 | DEBUG_PRINT1 (" . != \\n => fast loop.\n"); | |
4645 | return 1; | |
4646 | } | |
4647 | } | |
4648 | break; | |
505bde11 | 4649 | |
4e8a9132 | 4650 | case charset: |
4e8a9132 SM |
4651 | { |
4652 | if ((re_opcode_t) *p1 == exactn) | |
4653 | /* Reuse the code above. */ | |
4654 | return mutually_exclusive_p (bufp, p2, p1); | |
505bde11 | 4655 | |
505bde11 SM |
4656 | /* It is hard to list up all the character in charset |
4657 | P2 if it includes multibyte character. Give up in | |
4658 | such case. */ | |
4659 | else if (!multibyte || !CHARSET_RANGE_TABLE_EXISTS_P (p2)) | |
4660 | { | |
4661 | /* Now, we are sure that P2 has no range table. | |
4662 | So, for the size of bitmap in P2, `p2[1]' is | |
4663 | enough. But P1 may have range table, so the | |
4664 | size of bitmap table of P1 is extracted by | |
4665 | using macro `CHARSET_BITMAP_SIZE'. | |
4666 | ||
4667 | Since we know that all the character listed in | |
4668 | P2 is ASCII, it is enough to test only bitmap | |
4669 | table of P1. */ | |
4670 | ||
411e4203 | 4671 | if ((re_opcode_t) *p1 == charset) |
505bde11 SM |
4672 | { |
4673 | int idx; | |
4674 | /* We win if the charset inside the loop | |
4675 | has no overlap with the one after the loop. */ | |
4676 | for (idx = 0; | |
4677 | (idx < (int) p2[1] | |
4678 | && idx < CHARSET_BITMAP_SIZE (p1)); | |
4679 | idx++) | |
4680 | if ((p2[2 + idx] & p1[2 + idx]) != 0) | |
4681 | break; | |
4682 | ||
4683 | if (idx == p2[1] | |
4684 | || idx == CHARSET_BITMAP_SIZE (p1)) | |
4685 | { | |
4686 | DEBUG_PRINT1 (" No match => fast loop.\n"); | |
4687 | return 1; | |
4688 | } | |
4689 | } | |
411e4203 | 4690 | else if ((re_opcode_t) *p1 == charset_not) |
505bde11 SM |
4691 | { |
4692 | int idx; | |
4693 | /* We win if the charset_not inside the loop lists | |
4694 | every character listed in the charset after. */ | |
4695 | for (idx = 0; idx < (int) p2[1]; idx++) | |
4696 | if (! (p2[2 + idx] == 0 | |
4697 | || (idx < CHARSET_BITMAP_SIZE (p1) | |
4698 | && ((p2[2 + idx] & ~ p1[2 + idx]) == 0)))) | |
4699 | break; | |
4700 | ||
4e8a9132 SM |
4701 | if (idx == p2[1]) |
4702 | { | |
4703 | DEBUG_PRINT1 (" No match => fast loop.\n"); | |
4704 | return 1; | |
4705 | } | |
4706 | } | |
4707 | } | |
4708 | } | |
609b757a | 4709 | break; |
177c0ea7 | 4710 | |
411e4203 SM |
4711 | case charset_not: |
4712 | switch (SWITCH_ENUM_CAST (*p1)) | |
4713 | { | |
4714 | case exactn: | |
4715 | case charset: | |
4716 | /* Reuse the code above. */ | |
4717 | return mutually_exclusive_p (bufp, p2, p1); | |
4718 | case charset_not: | |
4719 | /* When we have two charset_not, it's very unlikely that | |
4720 | they don't overlap. The union of the two sets of excluded | |
4721 | chars should cover all possible chars, which, as a matter of | |
4722 | fact, is virtually impossible in multibyte buffers. */ | |
36595814 | 4723 | break; |
411e4203 SM |
4724 | } |
4725 | break; | |
4726 | ||
4e8a9132 | 4727 | case wordend: |
669fa600 SM |
4728 | return ((re_opcode_t) *p1 == syntaxspec && p1[1] == Sword); |
4729 | case symend: | |
4e8a9132 | 4730 | return ((re_opcode_t) *p1 == syntaxspec |
669fa600 SM |
4731 | && (p1[1] == Ssymbol || p1[1] == Sword)); |
4732 | case notsyntaxspec: | |
4733 | return ((re_opcode_t) *p1 == syntaxspec && p1[1] == p2[1]); | |
4e8a9132 SM |
4734 | |
4735 | case wordbeg: | |
669fa600 SM |
4736 | return ((re_opcode_t) *p1 == notsyntaxspec && p1[1] == Sword); |
4737 | case symbeg: | |
4e8a9132 | 4738 | return ((re_opcode_t) *p1 == notsyntaxspec |
669fa600 SM |
4739 | && (p1[1] == Ssymbol || p1[1] == Sword)); |
4740 | case syntaxspec: | |
4741 | return ((re_opcode_t) *p1 == notsyntaxspec && p1[1] == p2[1]); | |
4e8a9132 SM |
4742 | |
4743 | case wordbound: | |
4744 | return (((re_opcode_t) *p1 == notsyntaxspec | |
4745 | || (re_opcode_t) *p1 == syntaxspec) | |
4746 | && p1[1] == Sword); | |
4747 | ||
1fb352e0 | 4748 | #ifdef emacs |
4e8a9132 SM |
4749 | case categoryspec: |
4750 | return ((re_opcode_t) *p1 == notcategoryspec && p1[1] == p2[1]); | |
4751 | case notcategoryspec: | |
4752 | return ((re_opcode_t) *p1 == categoryspec && p1[1] == p2[1]); | |
4753 | #endif /* emacs */ | |
4754 | ||
4755 | default: | |
4756 | ; | |
505bde11 SM |
4757 | } |
4758 | ||
4759 | /* Safe default. */ | |
4760 | return 0; | |
4761 | } | |
4762 | ||
fa9a63c5 RM |
4763 | \f |
4764 | /* Matching routines. */ | |
4765 | ||
25fe55af | 4766 | #ifndef emacs /* Emacs never uses this. */ |
fa9a63c5 RM |
4767 | /* re_match is like re_match_2 except it takes only a single string. */ |
4768 | ||
4769 | int | |
4770 | re_match (bufp, string, size, pos, regs) | |
4771 | struct re_pattern_buffer *bufp; | |
4772 | const char *string; | |
4773 | int size, pos; | |
4774 | struct re_registers *regs; | |
4775 | { | |
4bb91c68 | 4776 | int result = re_match_2_internal (bufp, NULL, 0, (re_char*) string, size, |
fa9a63c5 | 4777 | pos, regs, size); |
0b32bf0e | 4778 | # if defined C_ALLOCA && !defined REGEX_MALLOC |
fa9a63c5 | 4779 | alloca (0); |
0b32bf0e | 4780 | # endif |
fa9a63c5 RM |
4781 | return result; |
4782 | } | |
c0f9ea08 | 4783 | WEAK_ALIAS (__re_match, re_match) |
fa9a63c5 RM |
4784 | #endif /* not emacs */ |
4785 | ||
b18215fc RS |
4786 | #ifdef emacs |
4787 | /* In Emacs, this is the string or buffer in which we | |
25fe55af | 4788 | are matching. It is used for looking up syntax properties. */ |
b18215fc RS |
4789 | Lisp_Object re_match_object; |
4790 | #endif | |
fa9a63c5 RM |
4791 | |
4792 | /* re_match_2 matches the compiled pattern in BUFP against the | |
4793 | the (virtual) concatenation of STRING1 and STRING2 (of length SIZE1 | |
4794 | and SIZE2, respectively). We start matching at POS, and stop | |
4795 | matching at STOP. | |
5e69f11e | 4796 | |
fa9a63c5 | 4797 | If REGS is non-null and the `no_sub' field of BUFP is nonzero, we |
25fe55af | 4798 | store offsets for the substring each group matched in REGS. See the |
fa9a63c5 RM |
4799 | documentation for exactly how many groups we fill. |
4800 | ||
4801 | We return -1 if no match, -2 if an internal error (such as the | |
25fe55af | 4802 | failure stack overflowing). Otherwise, we return the length of the |
fa9a63c5 RM |
4803 | matched substring. */ |
4804 | ||
4805 | int | |
4806 | re_match_2 (bufp, string1, size1, string2, size2, pos, regs, stop) | |
4807 | struct re_pattern_buffer *bufp; | |
4808 | const char *string1, *string2; | |
4809 | int size1, size2; | |
4810 | int pos; | |
4811 | struct re_registers *regs; | |
4812 | int stop; | |
4813 | { | |
b18215fc | 4814 | int result; |
25fe55af | 4815 | |
b18215fc | 4816 | #ifdef emacs |
cc9b4df2 KH |
4817 | int charpos; |
4818 | gl_state.object = re_match_object; | |
99633e97 | 4819 | charpos = SYNTAX_TABLE_BYTE_TO_CHAR (POS_AS_IN_BUFFER (pos)); |
cc9b4df2 | 4820 | SETUP_SYNTAX_TABLE_FOR_OBJECT (re_match_object, charpos, 1); |
b18215fc RS |
4821 | #endif |
4822 | ||
4bb91c68 SM |
4823 | result = re_match_2_internal (bufp, (re_char*) string1, size1, |
4824 | (re_char*) string2, size2, | |
cc9b4df2 | 4825 | pos, regs, stop); |
0b32bf0e | 4826 | #if defined C_ALLOCA && !defined REGEX_MALLOC |
fa9a63c5 | 4827 | alloca (0); |
a60198e5 | 4828 | #endif |
fa9a63c5 RM |
4829 | return result; |
4830 | } | |
c0f9ea08 | 4831 | WEAK_ALIAS (__re_match_2, re_match_2) |
fa9a63c5 RM |
4832 | |
4833 | /* This is a separate function so that we can force an alloca cleanup | |
25fe55af | 4834 | afterwards. */ |
fa9a63c5 RM |
4835 | static int |
4836 | re_match_2_internal (bufp, string1, size1, string2, size2, pos, regs, stop) | |
4837 | struct re_pattern_buffer *bufp; | |
66f0296e | 4838 | re_char *string1, *string2; |
fa9a63c5 RM |
4839 | int size1, size2; |
4840 | int pos; | |
4841 | struct re_registers *regs; | |
4842 | int stop; | |
4843 | { | |
4844 | /* General temporaries. */ | |
4845 | int mcnt; | |
01618498 | 4846 | size_t reg; |
66f0296e | 4847 | boolean not; |
fa9a63c5 RM |
4848 | |
4849 | /* Just past the end of the corresponding string. */ | |
66f0296e | 4850 | re_char *end1, *end2; |
fa9a63c5 RM |
4851 | |
4852 | /* Pointers into string1 and string2, just past the last characters in | |
25fe55af | 4853 | each to consider matching. */ |
66f0296e | 4854 | re_char *end_match_1, *end_match_2; |
fa9a63c5 RM |
4855 | |
4856 | /* Where we are in the data, and the end of the current string. */ | |
66f0296e | 4857 | re_char *d, *dend; |
5e69f11e | 4858 | |
99633e97 SM |
4859 | /* Used sometimes to remember where we were before starting matching |
4860 | an operator so that we can go back in case of failure. This "atomic" | |
4861 | behavior of matching opcodes is indispensable to the correctness | |
4862 | of the on_failure_keep_string_jump optimization. */ | |
4863 | re_char *dfail; | |
4864 | ||
fa9a63c5 | 4865 | /* Where we are in the pattern, and the end of the pattern. */ |
01618498 SM |
4866 | re_char *p = bufp->buffer; |
4867 | re_char *pend = p + bufp->used; | |
fa9a63c5 | 4868 | |
25fe55af | 4869 | /* We use this to map every character in the string. */ |
6676cb1c | 4870 | RE_TRANSLATE_TYPE translate = bufp->translate; |
fa9a63c5 | 4871 | |
25fe55af | 4872 | /* Nonzero if we have to concern multibyte character. */ |
2d1675e4 | 4873 | const boolean multibyte = RE_MULTIBYTE_P (bufp); |
b18215fc | 4874 | |
fa9a63c5 RM |
4875 | /* Failure point stack. Each place that can handle a failure further |
4876 | down the line pushes a failure point on this stack. It consists of | |
505bde11 | 4877 | regstart, and regend for all registers corresponding to |
fa9a63c5 RM |
4878 | the subexpressions we're currently inside, plus the number of such |
4879 | registers, and, finally, two char *'s. The first char * is where | |
4880 | to resume scanning the pattern; the second one is where to resume | |
505bde11 | 4881 | scanning the strings. */ |
25fe55af | 4882 | #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */ |
fa9a63c5 RM |
4883 | fail_stack_type fail_stack; |
4884 | #endif | |
4885 | #ifdef DEBUG | |
fa9a63c5 RM |
4886 | unsigned nfailure_points_pushed = 0, nfailure_points_popped = 0; |
4887 | #endif | |
4888 | ||
0b32bf0e | 4889 | #if defined REL_ALLOC && defined REGEX_MALLOC |
fa9a63c5 RM |
4890 | /* This holds the pointer to the failure stack, when |
4891 | it is allocated relocatably. */ | |
4892 | fail_stack_elt_t *failure_stack_ptr; | |
99633e97 | 4893 | #endif |
fa9a63c5 RM |
4894 | |
4895 | /* We fill all the registers internally, independent of what we | |
25fe55af | 4896 | return, for use in backreferences. The number here includes |
fa9a63c5 | 4897 | an element for register zero. */ |
4bb91c68 | 4898 | size_t num_regs = bufp->re_nsub + 1; |
5e69f11e | 4899 | |
fa9a63c5 RM |
4900 | /* Information on the contents of registers. These are pointers into |
4901 | the input strings; they record just what was matched (on this | |
4902 | attempt) by a subexpression part of the pattern, that is, the | |
4903 | regnum-th regstart pointer points to where in the pattern we began | |
4904 | matching and the regnum-th regend points to right after where we | |
4905 | stopped matching the regnum-th subexpression. (The zeroth register | |
4906 | keeps track of what the whole pattern matches.) */ | |
4907 | #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */ | |
66f0296e | 4908 | re_char **regstart, **regend; |
fa9a63c5 RM |
4909 | #endif |
4910 | ||
fa9a63c5 | 4911 | /* The following record the register info as found in the above |
5e69f11e | 4912 | variables when we find a match better than any we've seen before. |
fa9a63c5 RM |
4913 | This happens as we backtrack through the failure points, which in |
4914 | turn happens only if we have not yet matched the entire string. */ | |
4915 | unsigned best_regs_set = false; | |
4916 | #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */ | |
66f0296e | 4917 | re_char **best_regstart, **best_regend; |
fa9a63c5 | 4918 | #endif |
5e69f11e | 4919 | |
fa9a63c5 RM |
4920 | /* Logically, this is `best_regend[0]'. But we don't want to have to |
4921 | allocate space for that if we're not allocating space for anything | |
25fe55af | 4922 | else (see below). Also, we never need info about register 0 for |
fa9a63c5 RM |
4923 | any of the other register vectors, and it seems rather a kludge to |
4924 | treat `best_regend' differently than the rest. So we keep track of | |
4925 | the end of the best match so far in a separate variable. We | |
4926 | initialize this to NULL so that when we backtrack the first time | |
4927 | and need to test it, it's not garbage. */ | |
66f0296e | 4928 | re_char *match_end = NULL; |
fa9a63c5 | 4929 | |
fa9a63c5 RM |
4930 | #ifdef DEBUG |
4931 | /* Counts the total number of registers pushed. */ | |
5e69f11e | 4932 | unsigned num_regs_pushed = 0; |
fa9a63c5 RM |
4933 | #endif |
4934 | ||
4935 | DEBUG_PRINT1 ("\n\nEntering re_match_2.\n"); | |
5e69f11e | 4936 | |
fa9a63c5 | 4937 | INIT_FAIL_STACK (); |
5e69f11e | 4938 | |
fa9a63c5 RM |
4939 | #ifdef MATCH_MAY_ALLOCATE |
4940 | /* Do not bother to initialize all the register variables if there are | |
4941 | no groups in the pattern, as it takes a fair amount of time. If | |
4942 | there are groups, we include space for register 0 (the whole | |
4943 | pattern), even though we never use it, since it simplifies the | |
4944 | array indexing. We should fix this. */ | |
4945 | if (bufp->re_nsub) | |
4946 | { | |
66f0296e SM |
4947 | regstart = REGEX_TALLOC (num_regs, re_char *); |
4948 | regend = REGEX_TALLOC (num_regs, re_char *); | |
4949 | best_regstart = REGEX_TALLOC (num_regs, re_char *); | |
4950 | best_regend = REGEX_TALLOC (num_regs, re_char *); | |
fa9a63c5 | 4951 | |
505bde11 | 4952 | if (!(regstart && regend && best_regstart && best_regend)) |
25fe55af RS |
4953 | { |
4954 | FREE_VARIABLES (); | |
4955 | return -2; | |
4956 | } | |
fa9a63c5 RM |
4957 | } |
4958 | else | |
4959 | { | |
4960 | /* We must initialize all our variables to NULL, so that | |
25fe55af | 4961 | `FREE_VARIABLES' doesn't try to free them. */ |
505bde11 | 4962 | regstart = regend = best_regstart = best_regend = NULL; |
fa9a63c5 RM |
4963 | } |
4964 | #endif /* MATCH_MAY_ALLOCATE */ | |
4965 | ||
4966 | /* The starting position is bogus. */ | |
4967 | if (pos < 0 || pos > size1 + size2) | |
4968 | { | |
4969 | FREE_VARIABLES (); | |
4970 | return -1; | |
4971 | } | |
5e69f11e | 4972 | |
fa9a63c5 RM |
4973 | /* Initialize subexpression text positions to -1 to mark ones that no |
4974 | start_memory/stop_memory has been seen for. Also initialize the | |
4975 | register information struct. */ | |
01618498 SM |
4976 | for (reg = 1; reg < num_regs; reg++) |
4977 | regstart[reg] = regend[reg] = NULL; | |
99633e97 | 4978 | |
fa9a63c5 | 4979 | /* We move `string1' into `string2' if the latter's empty -- but not if |
25fe55af | 4980 | `string1' is null. */ |
fa9a63c5 RM |
4981 | if (size2 == 0 && string1 != NULL) |
4982 | { | |
4983 | string2 = string1; | |
4984 | size2 = size1; | |
4985 | string1 = 0; | |
4986 | size1 = 0; | |
4987 | } | |
4988 | end1 = string1 + size1; | |
4989 | end2 = string2 + size2; | |
4990 | ||
5e69f11e | 4991 | /* `p' scans through the pattern as `d' scans through the data. |
fa9a63c5 RM |
4992 | `dend' is the end of the input string that `d' points within. `d' |
4993 | is advanced into the following input string whenever necessary, but | |
4994 | this happens before fetching; therefore, at the beginning of the | |
4995 | loop, `d' can be pointing at the end of a string, but it cannot | |
4996 | equal `string2'. */ | |
419d1c74 | 4997 | if (pos >= size1) |
fa9a63c5 | 4998 | { |
419d1c74 SM |
4999 | /* Only match within string2. */ |
5000 | d = string2 + pos - size1; | |
5001 | dend = end_match_2 = string2 + stop - size1; | |
5002 | end_match_1 = end1; /* Just to give it a value. */ | |
fa9a63c5 RM |
5003 | } |
5004 | else | |
5005 | { | |
f1ad044f | 5006 | if (stop < size1) |
419d1c74 SM |
5007 | { |
5008 | /* Only match within string1. */ | |
5009 | end_match_1 = string1 + stop; | |
5010 | /* BEWARE! | |
5011 | When we reach end_match_1, PREFETCH normally switches to string2. | |
5012 | But in the present case, this means that just doing a PREFETCH | |
5013 | makes us jump from `stop' to `gap' within the string. | |
5014 | What we really want here is for the search to stop as | |
5015 | soon as we hit end_match_1. That's why we set end_match_2 | |
5016 | to end_match_1 (since PREFETCH fails as soon as we hit | |
5017 | end_match_2). */ | |
5018 | end_match_2 = end_match_1; | |
5019 | } | |
5020 | else | |
f1ad044f SM |
5021 | { /* It's important to use this code when stop == size so that |
5022 | moving `d' from end1 to string2 will not prevent the d == dend | |
5023 | check from catching the end of string. */ | |
419d1c74 SM |
5024 | end_match_1 = end1; |
5025 | end_match_2 = string2 + stop - size1; | |
5026 | } | |
5027 | d = string1 + pos; | |
5028 | dend = end_match_1; | |
fa9a63c5 RM |
5029 | } |
5030 | ||
5031 | DEBUG_PRINT1 ("The compiled pattern is: "); | |
5032 | DEBUG_PRINT_COMPILED_PATTERN (bufp, p, pend); | |
5033 | DEBUG_PRINT1 ("The string to match is: `"); | |
5034 | DEBUG_PRINT_DOUBLE_STRING (d, string1, size1, string2, size2); | |
5035 | DEBUG_PRINT1 ("'\n"); | |
5e69f11e | 5036 | |
25fe55af | 5037 | /* This loops over pattern commands. It exits by returning from the |
fa9a63c5 RM |
5038 | function if the match is complete, or it drops through if the match |
5039 | fails at this starting point in the input data. */ | |
5040 | for (;;) | |
5041 | { | |
505bde11 | 5042 | DEBUG_PRINT2 ("\n%p: ", p); |
fa9a63c5 RM |
5043 | |
5044 | if (p == pend) | |
5045 | { /* End of pattern means we might have succeeded. */ | |
25fe55af | 5046 | DEBUG_PRINT1 ("end of pattern ... "); |
5e69f11e | 5047 | |
fa9a63c5 | 5048 | /* If we haven't matched the entire string, and we want the |
25fe55af RS |
5049 | longest match, try backtracking. */ |
5050 | if (d != end_match_2) | |
fa9a63c5 RM |
5051 | { |
5052 | /* 1 if this match ends in the same string (string1 or string2) | |
5053 | as the best previous match. */ | |
5e69f11e | 5054 | boolean same_str_p = (FIRST_STRING_P (match_end) |
99633e97 | 5055 | == FIRST_STRING_P (d)); |
fa9a63c5 RM |
5056 | /* 1 if this match is the best seen so far. */ |
5057 | boolean best_match_p; | |
5058 | ||
5059 | /* AIX compiler got confused when this was combined | |
25fe55af | 5060 | with the previous declaration. */ |
fa9a63c5 RM |
5061 | if (same_str_p) |
5062 | best_match_p = d > match_end; | |
5063 | else | |
99633e97 | 5064 | best_match_p = !FIRST_STRING_P (d); |
fa9a63c5 | 5065 | |
25fe55af RS |
5066 | DEBUG_PRINT1 ("backtracking.\n"); |
5067 | ||
5068 | if (!FAIL_STACK_EMPTY ()) | |
5069 | { /* More failure points to try. */ | |
5070 | ||
5071 | /* If exceeds best match so far, save it. */ | |
5072 | if (!best_regs_set || best_match_p) | |
5073 | { | |
5074 | best_regs_set = true; | |
5075 | match_end = d; | |
5076 | ||
5077 | DEBUG_PRINT1 ("\nSAVING match as best so far.\n"); | |
5078 | ||
01618498 | 5079 | for (reg = 1; reg < num_regs; reg++) |
25fe55af | 5080 | { |
01618498 SM |
5081 | best_regstart[reg] = regstart[reg]; |
5082 | best_regend[reg] = regend[reg]; | |
25fe55af RS |
5083 | } |
5084 | } | |
5085 | goto fail; | |
5086 | } | |
5087 | ||
5088 | /* If no failure points, don't restore garbage. And if | |
5089 | last match is real best match, don't restore second | |
5090 | best one. */ | |
5091 | else if (best_regs_set && !best_match_p) | |
5092 | { | |
5093 | restore_best_regs: | |
5094 | /* Restore best match. It may happen that `dend == | |
5095 | end_match_1' while the restored d is in string2. | |
5096 | For example, the pattern `x.*y.*z' against the | |
5097 | strings `x-' and `y-z-', if the two strings are | |
5098 | not consecutive in memory. */ | |
5099 | DEBUG_PRINT1 ("Restoring best registers.\n"); | |
5100 | ||
5101 | d = match_end; | |
5102 | dend = ((d >= string1 && d <= end1) | |
5103 | ? end_match_1 : end_match_2); | |
fa9a63c5 | 5104 | |
01618498 | 5105 | for (reg = 1; reg < num_regs; reg++) |
fa9a63c5 | 5106 | { |
01618498 SM |
5107 | regstart[reg] = best_regstart[reg]; |
5108 | regend[reg] = best_regend[reg]; | |
fa9a63c5 | 5109 | } |
25fe55af RS |
5110 | } |
5111 | } /* d != end_match_2 */ | |
fa9a63c5 RM |
5112 | |
5113 | succeed_label: | |
25fe55af | 5114 | DEBUG_PRINT1 ("Accepting match.\n"); |
fa9a63c5 | 5115 | |
25fe55af RS |
5116 | /* If caller wants register contents data back, do it. */ |
5117 | if (regs && !bufp->no_sub) | |
fa9a63c5 | 5118 | { |
25fe55af RS |
5119 | /* Have the register data arrays been allocated? */ |
5120 | if (bufp->regs_allocated == REGS_UNALLOCATED) | |
5121 | { /* No. So allocate them with malloc. We need one | |
5122 | extra element beyond `num_regs' for the `-1' marker | |
5123 | GNU code uses. */ | |
5124 | regs->num_regs = MAX (RE_NREGS, num_regs + 1); | |
5125 | regs->start = TALLOC (regs->num_regs, regoff_t); | |
5126 | regs->end = TALLOC (regs->num_regs, regoff_t); | |
5127 | if (regs->start == NULL || regs->end == NULL) | |
fa9a63c5 RM |
5128 | { |
5129 | FREE_VARIABLES (); | |
5130 | return -2; | |
5131 | } | |
25fe55af RS |
5132 | bufp->regs_allocated = REGS_REALLOCATE; |
5133 | } | |
5134 | else if (bufp->regs_allocated == REGS_REALLOCATE) | |
5135 | { /* Yes. If we need more elements than were already | |
5136 | allocated, reallocate them. If we need fewer, just | |
5137 | leave it alone. */ | |
5138 | if (regs->num_regs < num_regs + 1) | |
5139 | { | |
5140 | regs->num_regs = num_regs + 1; | |
5141 | RETALLOC (regs->start, regs->num_regs, regoff_t); | |
5142 | RETALLOC (regs->end, regs->num_regs, regoff_t); | |
5143 | if (regs->start == NULL || regs->end == NULL) | |
fa9a63c5 RM |
5144 | { |
5145 | FREE_VARIABLES (); | |
5146 | return -2; | |
5147 | } | |
25fe55af RS |
5148 | } |
5149 | } | |
5150 | else | |
fa9a63c5 RM |
5151 | { |
5152 | /* These braces fend off a "empty body in an else-statement" | |
25fe55af | 5153 | warning under GCC when assert expands to nothing. */ |
fa9a63c5 RM |
5154 | assert (bufp->regs_allocated == REGS_FIXED); |
5155 | } | |
5156 | ||
25fe55af RS |
5157 | /* Convert the pointer data in `regstart' and `regend' to |
5158 | indices. Register zero has to be set differently, | |
5159 | since we haven't kept track of any info for it. */ | |
5160 | if (regs->num_regs > 0) | |
5161 | { | |
5162 | regs->start[0] = pos; | |
99633e97 | 5163 | regs->end[0] = POINTER_TO_OFFSET (d); |
25fe55af | 5164 | } |
5e69f11e | 5165 | |
25fe55af RS |
5166 | /* Go through the first `min (num_regs, regs->num_regs)' |
5167 | registers, since that is all we initialized. */ | |
01618498 | 5168 | for (reg = 1; reg < MIN (num_regs, regs->num_regs); reg++) |
fa9a63c5 | 5169 | { |
01618498 SM |
5170 | if (REG_UNSET (regstart[reg]) || REG_UNSET (regend[reg])) |
5171 | regs->start[reg] = regs->end[reg] = -1; | |
25fe55af RS |
5172 | else |
5173 | { | |
01618498 SM |
5174 | regs->start[reg] |
5175 | = (regoff_t) POINTER_TO_OFFSET (regstart[reg]); | |
5176 | regs->end[reg] | |
5177 | = (regoff_t) POINTER_TO_OFFSET (regend[reg]); | |
25fe55af | 5178 | } |
fa9a63c5 | 5179 | } |
5e69f11e | 5180 | |
25fe55af RS |
5181 | /* If the regs structure we return has more elements than |
5182 | were in the pattern, set the extra elements to -1. If | |
5183 | we (re)allocated the registers, this is the case, | |
5184 | because we always allocate enough to have at least one | |
5185 | -1 at the end. */ | |
01618498 SM |
5186 | for (reg = num_regs; reg < regs->num_regs; reg++) |
5187 | regs->start[reg] = regs->end[reg] = -1; | |
fa9a63c5 RM |
5188 | } /* regs && !bufp->no_sub */ |
5189 | ||
25fe55af RS |
5190 | DEBUG_PRINT4 ("%u failure points pushed, %u popped (%u remain).\n", |
5191 | nfailure_points_pushed, nfailure_points_popped, | |
5192 | nfailure_points_pushed - nfailure_points_popped); | |
5193 | DEBUG_PRINT2 ("%u registers pushed.\n", num_regs_pushed); | |
fa9a63c5 | 5194 | |
99633e97 | 5195 | mcnt = POINTER_TO_OFFSET (d) - pos; |
fa9a63c5 | 5196 | |
25fe55af | 5197 | DEBUG_PRINT2 ("Returning %d from re_match_2.\n", mcnt); |
fa9a63c5 | 5198 | |
25fe55af RS |
5199 | FREE_VARIABLES (); |
5200 | return mcnt; | |
5201 | } | |
fa9a63c5 | 5202 | |
25fe55af | 5203 | /* Otherwise match next pattern command. */ |
fa9a63c5 RM |
5204 | switch (SWITCH_ENUM_CAST ((re_opcode_t) *p++)) |
5205 | { | |
25fe55af RS |
5206 | /* Ignore these. Used to ignore the n of succeed_n's which |
5207 | currently have n == 0. */ | |
5208 | case no_op: | |
5209 | DEBUG_PRINT1 ("EXECUTING no_op.\n"); | |
5210 | break; | |
fa9a63c5 RM |
5211 | |
5212 | case succeed: | |
25fe55af | 5213 | DEBUG_PRINT1 ("EXECUTING succeed.\n"); |
fa9a63c5 RM |
5214 | goto succeed_label; |
5215 | ||
25fe55af RS |
5216 | /* Match the next n pattern characters exactly. The following |
5217 | byte in the pattern defines n, and the n bytes after that | |
5218 | are the characters to match. */ | |
fa9a63c5 RM |
5219 | case exactn: |
5220 | mcnt = *p++; | |
25fe55af | 5221 | DEBUG_PRINT2 ("EXECUTING exactn %d.\n", mcnt); |
fa9a63c5 | 5222 | |
99633e97 SM |
5223 | /* Remember the start point to rollback upon failure. */ |
5224 | dfail = d; | |
5225 | ||
25fe55af RS |
5226 | /* This is written out as an if-else so we don't waste time |
5227 | testing `translate' inside the loop. */ | |
28703c16 | 5228 | if (RE_TRANSLATE_P (translate)) |
fa9a63c5 | 5229 | { |
e934739e RS |
5230 | if (multibyte) |
5231 | do | |
5232 | { | |
5233 | int pat_charlen, buf_charlen; | |
e71b1971 | 5234 | unsigned int pat_ch, buf_ch; |
e934739e RS |
5235 | |
5236 | PREFETCH (); | |
5237 | pat_ch = STRING_CHAR_AND_LENGTH (p, pend - p, pat_charlen); | |
5238 | buf_ch = STRING_CHAR_AND_LENGTH (d, dend - d, buf_charlen); | |
5239 | ||
5240 | if (RE_TRANSLATE (translate, buf_ch) | |
5241 | != pat_ch) | |
99633e97 SM |
5242 | { |
5243 | d = dfail; | |
5244 | goto fail; | |
5245 | } | |
e934739e RS |
5246 | |
5247 | p += pat_charlen; | |
5248 | d += buf_charlen; | |
5249 | mcnt -= pat_charlen; | |
5250 | } | |
5251 | while (mcnt > 0); | |
5252 | else | |
e934739e RS |
5253 | do |
5254 | { | |
cf1982d9 EZ |
5255 | /* Avoid compiler whining about comparison being |
5256 | always true. */ | |
5257 | int di; | |
5258 | ||
e934739e | 5259 | PREFETCH (); |
cf1982d9 EZ |
5260 | di = *d; |
5261 | if (RE_TRANSLATE (translate, di) != *p++) | |
99633e97 SM |
5262 | { |
5263 | d = dfail; | |
5264 | goto fail; | |
5265 | } | |
33c46939 | 5266 | d++; |
e934739e RS |
5267 | } |
5268 | while (--mcnt); | |
fa9a63c5 RM |
5269 | } |
5270 | else | |
5271 | { | |
5272 | do | |
5273 | { | |
5274 | PREFETCH (); | |
99633e97 SM |
5275 | if (*d++ != *p++) |
5276 | { | |
5277 | d = dfail; | |
5278 | goto fail; | |
5279 | } | |
fa9a63c5 RM |
5280 | } |
5281 | while (--mcnt); | |
5282 | } | |
25fe55af | 5283 | break; |
fa9a63c5 RM |
5284 | |
5285 | ||
25fe55af | 5286 | /* Match any character except possibly a newline or a null. */ |
fa9a63c5 | 5287 | case anychar: |
e934739e RS |
5288 | { |
5289 | int buf_charlen; | |
01618498 | 5290 | re_wchar_t buf_ch; |
fa9a63c5 | 5291 | |
e934739e | 5292 | DEBUG_PRINT1 ("EXECUTING anychar.\n"); |
fa9a63c5 | 5293 | |
e934739e | 5294 | PREFETCH (); |
2d1675e4 | 5295 | buf_ch = RE_STRING_CHAR_AND_LENGTH (d, dend - d, buf_charlen); |
e934739e RS |
5296 | buf_ch = TRANSLATE (buf_ch); |
5297 | ||
5298 | if ((!(bufp->syntax & RE_DOT_NEWLINE) | |
5299 | && buf_ch == '\n') | |
5300 | || ((bufp->syntax & RE_DOT_NOT_NULL) | |
5301 | && buf_ch == '\000')) | |
5302 | goto fail; | |
5303 | ||
e934739e RS |
5304 | DEBUG_PRINT2 (" Matched `%d'.\n", *d); |
5305 | d += buf_charlen; | |
5306 | } | |
fa9a63c5 RM |
5307 | break; |
5308 | ||
5309 | ||
5310 | case charset: | |
5311 | case charset_not: | |
5312 | { | |
b18215fc | 5313 | register unsigned int c; |
fa9a63c5 | 5314 | boolean not = (re_opcode_t) *(p - 1) == charset_not; |
b18215fc RS |
5315 | int len; |
5316 | ||
5317 | /* Start of actual range_table, or end of bitmap if there is no | |
5318 | range table. */ | |
01618498 | 5319 | re_char *range_table; |
b18215fc | 5320 | |
96cc36cc | 5321 | /* Nonzero if there is a range table. */ |
b18215fc RS |
5322 | int range_table_exists; |
5323 | ||
96cc36cc RS |
5324 | /* Number of ranges of range table. This is not included |
5325 | in the initial byte-length of the command. */ | |
5326 | int count = 0; | |
fa9a63c5 | 5327 | |
25fe55af | 5328 | DEBUG_PRINT2 ("EXECUTING charset%s.\n", not ? "_not" : ""); |
fa9a63c5 | 5329 | |
b18215fc | 5330 | range_table_exists = CHARSET_RANGE_TABLE_EXISTS_P (&p[-1]); |
96cc36cc | 5331 | |
b18215fc | 5332 | if (range_table_exists) |
96cc36cc RS |
5333 | { |
5334 | range_table = CHARSET_RANGE_TABLE (&p[-1]); /* Past the bitmap. */ | |
5335 | EXTRACT_NUMBER_AND_INCR (count, range_table); | |
5336 | } | |
b18215fc | 5337 | |
2d1675e4 SM |
5338 | PREFETCH (); |
5339 | c = RE_STRING_CHAR_AND_LENGTH (d, dend - d, len); | |
5340 | c = TRANSLATE (c); /* The character to match. */ | |
b18215fc RS |
5341 | |
5342 | if (SINGLE_BYTE_CHAR_P (c)) | |
5343 | { /* Lookup bitmap. */ | |
b18215fc RS |
5344 | /* Cast to `unsigned' instead of `unsigned char' in |
5345 | case the bit list is a full 32 bytes long. */ | |
5346 | if (c < (unsigned) (CHARSET_BITMAP_SIZE (&p[-1]) * BYTEWIDTH) | |
96cc36cc RS |
5347 | && p[1 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH))) |
5348 | not = !not; | |
b18215fc | 5349 | } |
96cc36cc | 5350 | #ifdef emacs |
b18215fc | 5351 | else if (range_table_exists) |
96cc36cc RS |
5352 | { |
5353 | int class_bits = CHARSET_RANGE_TABLE_BITS (&p[-1]); | |
5354 | ||
14473664 SM |
5355 | if ( (class_bits & BIT_LOWER && ISLOWER (c)) |
5356 | | (class_bits & BIT_MULTIBYTE) | |
96cc36cc RS |
5357 | | (class_bits & BIT_PUNCT && ISPUNCT (c)) |
5358 | | (class_bits & BIT_SPACE && ISSPACE (c)) | |
5359 | | (class_bits & BIT_UPPER && ISUPPER (c)) | |
5360 | | (class_bits & BIT_WORD && ISWORD (c))) | |
5361 | not = !not; | |
5362 | else | |
5363 | CHARSET_LOOKUP_RANGE_TABLE_RAW (not, c, range_table, count); | |
5364 | } | |
5365 | #endif /* emacs */ | |
fa9a63c5 | 5366 | |
96cc36cc RS |
5367 | if (range_table_exists) |
5368 | p = CHARSET_RANGE_TABLE_END (range_table, count); | |
5369 | else | |
5370 | p += CHARSET_BITMAP_SIZE (&p[-1]) + 1; | |
fa9a63c5 RM |
5371 | |
5372 | if (!not) goto fail; | |
5e69f11e | 5373 | |
b18215fc | 5374 | d += len; |
fa9a63c5 RM |
5375 | break; |
5376 | } | |
5377 | ||
5378 | ||
25fe55af | 5379 | /* The beginning of a group is represented by start_memory. |
505bde11 | 5380 | The argument is the register number. The text |
25fe55af RS |
5381 | matched within the group is recorded (in the internal |
5382 | registers data structure) under the register number. */ | |
5383 | case start_memory: | |
505bde11 SM |
5384 | DEBUG_PRINT2 ("EXECUTING start_memory %d:\n", *p); |
5385 | ||
5386 | /* In case we need to undo this operation (via backtracking). */ | |
5387 | PUSH_FAILURE_REG ((unsigned int)*p); | |
fa9a63c5 | 5388 | |
25fe55af | 5389 | regstart[*p] = d; |
4bb91c68 | 5390 | regend[*p] = NULL; /* probably unnecessary. -sm */ |
fa9a63c5 RM |
5391 | DEBUG_PRINT2 (" regstart: %d\n", POINTER_TO_OFFSET (regstart[*p])); |
5392 | ||
25fe55af | 5393 | /* Move past the register number and inner group count. */ |
505bde11 | 5394 | p += 1; |
25fe55af | 5395 | break; |
fa9a63c5 RM |
5396 | |
5397 | ||
25fe55af | 5398 | /* The stop_memory opcode represents the end of a group. Its |
505bde11 | 5399 | argument is the same as start_memory's: the register number. */ |
fa9a63c5 | 5400 | case stop_memory: |
505bde11 SM |
5401 | DEBUG_PRINT2 ("EXECUTING stop_memory %d:\n", *p); |
5402 | ||
5403 | assert (!REG_UNSET (regstart[*p])); | |
5404 | /* Strictly speaking, there should be code such as: | |
177c0ea7 | 5405 | |
0b32bf0e | 5406 | assert (REG_UNSET (regend[*p])); |
505bde11 SM |
5407 | PUSH_FAILURE_REGSTOP ((unsigned int)*p); |
5408 | ||
5409 | But the only info to be pushed is regend[*p] and it is known to | |
5410 | be UNSET, so there really isn't anything to push. | |
5411 | Not pushing anything, on the other hand deprives us from the | |
5412 | guarantee that regend[*p] is UNSET since undoing this operation | |
5413 | will not reset its value properly. This is not important since | |
5414 | the value will only be read on the next start_memory or at | |
5415 | the very end and both events can only happen if this stop_memory | |
5416 | is *not* undone. */ | |
fa9a63c5 | 5417 | |
25fe55af | 5418 | regend[*p] = d; |
fa9a63c5 RM |
5419 | DEBUG_PRINT2 (" regend: %d\n", POINTER_TO_OFFSET (regend[*p])); |
5420 | ||
25fe55af | 5421 | /* Move past the register number and the inner group count. */ |
505bde11 | 5422 | p += 1; |
25fe55af | 5423 | break; |
fa9a63c5 RM |
5424 | |
5425 | ||
5426 | /* \<digit> has been turned into a `duplicate' command which is | |
25fe55af RS |
5427 | followed by the numeric value of <digit> as the register number. */ |
5428 | case duplicate: | |
fa9a63c5 | 5429 | { |
66f0296e | 5430 | register re_char *d2, *dend2; |
25fe55af | 5431 | int regno = *p++; /* Get which register to match against. */ |
fa9a63c5 RM |
5432 | DEBUG_PRINT2 ("EXECUTING duplicate %d.\n", regno); |
5433 | ||
25fe55af RS |
5434 | /* Can't back reference a group which we've never matched. */ |
5435 | if (REG_UNSET (regstart[regno]) || REG_UNSET (regend[regno])) | |
5436 | goto fail; | |
5e69f11e | 5437 | |
25fe55af RS |
5438 | /* Where in input to try to start matching. */ |
5439 | d2 = regstart[regno]; | |
5e69f11e | 5440 | |
99633e97 SM |
5441 | /* Remember the start point to rollback upon failure. */ |
5442 | dfail = d; | |
5443 | ||
25fe55af RS |
5444 | /* Where to stop matching; if both the place to start and |
5445 | the place to stop matching are in the same string, then | |
5446 | set to the place to stop, otherwise, for now have to use | |
5447 | the end of the first string. */ | |
fa9a63c5 | 5448 | |
25fe55af | 5449 | dend2 = ((FIRST_STRING_P (regstart[regno]) |
fa9a63c5 RM |
5450 | == FIRST_STRING_P (regend[regno])) |
5451 | ? regend[regno] : end_match_1); | |
5452 | for (;;) | |
5453 | { | |
5454 | /* If necessary, advance to next segment in register | |
25fe55af | 5455 | contents. */ |
fa9a63c5 RM |
5456 | while (d2 == dend2) |
5457 | { | |
5458 | if (dend2 == end_match_2) break; | |
5459 | if (dend2 == regend[regno]) break; | |
5460 | ||
25fe55af RS |
5461 | /* End of string1 => advance to string2. */ |
5462 | d2 = string2; | |
5463 | dend2 = regend[regno]; | |
fa9a63c5 RM |
5464 | } |
5465 | /* At end of register contents => success */ | |
5466 | if (d2 == dend2) break; | |
5467 | ||
5468 | /* If necessary, advance to next segment in data. */ | |
5469 | PREFETCH (); | |
5470 | ||
5471 | /* How many characters left in this segment to match. */ | |
5472 | mcnt = dend - d; | |
5e69f11e | 5473 | |
fa9a63c5 | 5474 | /* Want how many consecutive characters we can match in |
25fe55af RS |
5475 | one shot, so, if necessary, adjust the count. */ |
5476 | if (mcnt > dend2 - d2) | |
fa9a63c5 | 5477 | mcnt = dend2 - d2; |
5e69f11e | 5478 | |
fa9a63c5 | 5479 | /* Compare that many; failure if mismatch, else move |
25fe55af | 5480 | past them. */ |
28703c16 | 5481 | if (RE_TRANSLATE_P (translate) |
2d1675e4 | 5482 | ? bcmp_translate (d, d2, mcnt, translate, multibyte) |
4bb91c68 | 5483 | : memcmp (d, d2, mcnt)) |
99633e97 SM |
5484 | { |
5485 | d = dfail; | |
5486 | goto fail; | |
5487 | } | |
fa9a63c5 | 5488 | d += mcnt, d2 += mcnt; |
fa9a63c5 RM |
5489 | } |
5490 | } | |
5491 | break; | |
5492 | ||
5493 | ||
25fe55af | 5494 | /* begline matches the empty string at the beginning of the string |
c0f9ea08 | 5495 | (unless `not_bol' is set in `bufp'), and after newlines. */ |
fa9a63c5 | 5496 | case begline: |
25fe55af | 5497 | DEBUG_PRINT1 ("EXECUTING begline.\n"); |
5e69f11e | 5498 | |
25fe55af RS |
5499 | if (AT_STRINGS_BEG (d)) |
5500 | { | |
5501 | if (!bufp->not_bol) break; | |
5502 | } | |
419d1c74 | 5503 | else |
25fe55af | 5504 | { |
419d1c74 SM |
5505 | unsigned char c; |
5506 | GET_CHAR_BEFORE_2 (c, d, string1, end1, string2, end2); | |
c0f9ea08 | 5507 | if (c == '\n') |
419d1c74 | 5508 | break; |
25fe55af RS |
5509 | } |
5510 | /* In all other cases, we fail. */ | |
5511 | goto fail; | |
fa9a63c5 RM |
5512 | |
5513 | ||
25fe55af | 5514 | /* endline is the dual of begline. */ |
fa9a63c5 | 5515 | case endline: |
25fe55af | 5516 | DEBUG_PRINT1 ("EXECUTING endline.\n"); |
fa9a63c5 | 5517 | |
25fe55af RS |
5518 | if (AT_STRINGS_END (d)) |
5519 | { | |
5520 | if (!bufp->not_eol) break; | |
5521 | } | |
f1ad044f | 5522 | else |
25fe55af | 5523 | { |
f1ad044f | 5524 | PREFETCH_NOLIMIT (); |
c0f9ea08 | 5525 | if (*d == '\n') |
f1ad044f | 5526 | break; |
25fe55af RS |
5527 | } |
5528 | goto fail; | |
fa9a63c5 RM |
5529 | |
5530 | ||
5531 | /* Match at the very beginning of the data. */ | |
25fe55af RS |
5532 | case begbuf: |
5533 | DEBUG_PRINT1 ("EXECUTING begbuf.\n"); | |
5534 | if (AT_STRINGS_BEG (d)) | |
5535 | break; | |
5536 | goto fail; | |
fa9a63c5 RM |
5537 | |
5538 | ||
5539 | /* Match at the very end of the data. */ | |
25fe55af RS |
5540 | case endbuf: |
5541 | DEBUG_PRINT1 ("EXECUTING endbuf.\n"); | |
fa9a63c5 RM |
5542 | if (AT_STRINGS_END (d)) |
5543 | break; | |
25fe55af | 5544 | goto fail; |
5e69f11e | 5545 | |
5e69f11e | 5546 | |
25fe55af RS |
5547 | /* on_failure_keep_string_jump is used to optimize `.*\n'. It |
5548 | pushes NULL as the value for the string on the stack. Then | |
505bde11 | 5549 | `POP_FAILURE_POINT' will keep the current value for the |
25fe55af RS |
5550 | string, instead of restoring it. To see why, consider |
5551 | matching `foo\nbar' against `.*\n'. The .* matches the foo; | |
5552 | then the . fails against the \n. But the next thing we want | |
5553 | to do is match the \n against the \n; if we restored the | |
5554 | string value, we would be back at the foo. | |
5555 | ||
5556 | Because this is used only in specific cases, we don't need to | |
5557 | check all the things that `on_failure_jump' does, to make | |
5558 | sure the right things get saved on the stack. Hence we don't | |
5559 | share its code. The only reason to push anything on the | |
5560 | stack at all is that otherwise we would have to change | |
5561 | `anychar's code to do something besides goto fail in this | |
5562 | case; that seems worse than this. */ | |
5563 | case on_failure_keep_string_jump: | |
505bde11 SM |
5564 | EXTRACT_NUMBER_AND_INCR (mcnt, p); |
5565 | DEBUG_PRINT3 ("EXECUTING on_failure_keep_string_jump %d (to %p):\n", | |
5566 | mcnt, p + mcnt); | |
fa9a63c5 | 5567 | |
505bde11 SM |
5568 | PUSH_FAILURE_POINT (p - 3, NULL); |
5569 | break; | |
5570 | ||
0683b6fa SM |
5571 | /* A nasty loop is introduced by the non-greedy *? and +?. |
5572 | With such loops, the stack only ever contains one failure point | |
5573 | at a time, so that a plain on_failure_jump_loop kind of | |
5574 | cycle detection cannot work. Worse yet, such a detection | |
5575 | can not only fail to detect a cycle, but it can also wrongly | |
5576 | detect a cycle (between different instantiations of the same | |
6df42991 | 5577 | loop). |
0683b6fa SM |
5578 | So the method used for those nasty loops is a little different: |
5579 | We use a special cycle-detection-stack-frame which is pushed | |
5580 | when the on_failure_jump_nastyloop failure-point is *popped*. | |
5581 | This special frame thus marks the beginning of one iteration | |
5582 | through the loop and we can hence easily check right here | |
5583 | whether something matched between the beginning and the end of | |
5584 | the loop. */ | |
5585 | case on_failure_jump_nastyloop: | |
5586 | EXTRACT_NUMBER_AND_INCR (mcnt, p); | |
5587 | DEBUG_PRINT3 ("EXECUTING on_failure_jump_nastyloop %d (to %p):\n", | |
5588 | mcnt, p + mcnt); | |
5589 | ||
5590 | assert ((re_opcode_t)p[-4] == no_op); | |
6df42991 SM |
5591 | { |
5592 | int cycle = 0; | |
5593 | CHECK_INFINITE_LOOP (p - 4, d); | |
5594 | if (!cycle) | |
5595 | /* If there's a cycle, just continue without pushing | |
5596 | this failure point. The failure point is the "try again" | |
5597 | option, which shouldn't be tried. | |
5598 | We want (x?)*?y\1z to match both xxyz and xxyxz. */ | |
5599 | PUSH_FAILURE_POINT (p - 3, d); | |
5600 | } | |
0683b6fa SM |
5601 | break; |
5602 | ||
4e8a9132 SM |
5603 | /* Simple loop detecting on_failure_jump: just check on the |
5604 | failure stack if the same spot was already hit earlier. */ | |
505bde11 SM |
5605 | case on_failure_jump_loop: |
5606 | on_failure: | |
5607 | EXTRACT_NUMBER_AND_INCR (mcnt, p); | |
5608 | DEBUG_PRINT3 ("EXECUTING on_failure_jump_loop %d (to %p):\n", | |
5609 | mcnt, p + mcnt); | |
6df42991 SM |
5610 | { |
5611 | int cycle = 0; | |
5612 | CHECK_INFINITE_LOOP (p - 3, d); | |
5613 | if (cycle) | |
5614 | /* If there's a cycle, get out of the loop, as if the matching | |
5615 | had failed. We used to just `goto fail' here, but that was | |
5616 | aborting the search a bit too early: we want to keep the | |
5617 | empty-loop-match and keep matching after the loop. | |
5618 | We want (x?)*y\1z to match both xxyz and xxyxz. */ | |
5619 | p += mcnt; | |
5620 | else | |
5621 | PUSH_FAILURE_POINT (p - 3, d); | |
5622 | } | |
25fe55af | 5623 | break; |
fa9a63c5 RM |
5624 | |
5625 | ||
5626 | /* Uses of on_failure_jump: | |
5e69f11e | 5627 | |
25fe55af RS |
5628 | Each alternative starts with an on_failure_jump that points |
5629 | to the beginning of the next alternative. Each alternative | |
5630 | except the last ends with a jump that in effect jumps past | |
5631 | the rest of the alternatives. (They really jump to the | |
5632 | ending jump of the following alternative, because tensioning | |
5633 | these jumps is a hassle.) | |
fa9a63c5 | 5634 | |
25fe55af RS |
5635 | Repeats start with an on_failure_jump that points past both |
5636 | the repetition text and either the following jump or | |
5637 | pop_failure_jump back to this on_failure_jump. */ | |
fa9a63c5 | 5638 | case on_failure_jump: |
5b370c2b | 5639 | IMMEDIATE_QUIT_CHECK; |
25fe55af | 5640 | EXTRACT_NUMBER_AND_INCR (mcnt, p); |
505bde11 SM |
5641 | DEBUG_PRINT3 ("EXECUTING on_failure_jump %d (to %p):\n", |
5642 | mcnt, p + mcnt); | |
25fe55af | 5643 | |
505bde11 | 5644 | PUSH_FAILURE_POINT (p -3, d); |
25fe55af RS |
5645 | break; |
5646 | ||
4e8a9132 | 5647 | /* This operation is used for greedy *. |
505bde11 SM |
5648 | Compare the beginning of the repeat with what in the |
5649 | pattern follows its end. If we can establish that there | |
5650 | is nothing that they would both match, i.e., that we | |
5651 | would have to backtrack because of (as in, e.g., `a*a') | |
5652 | then we can use a non-backtracking loop based on | |
4e8a9132 | 5653 | on_failure_keep_string_jump instead of on_failure_jump. */ |
505bde11 | 5654 | case on_failure_jump_smart: |
5b370c2b | 5655 | IMMEDIATE_QUIT_CHECK; |
25fe55af | 5656 | EXTRACT_NUMBER_AND_INCR (mcnt, p); |
505bde11 SM |
5657 | DEBUG_PRINT3 ("EXECUTING on_failure_jump_smart %d (to %p).\n", |
5658 | mcnt, p + mcnt); | |
25fe55af | 5659 | { |
01618498 | 5660 | re_char *p1 = p; /* Next operation. */ |
6dcf2d0e SM |
5661 | /* Here, we discard `const', making re_match non-reentrant. */ |
5662 | unsigned char *p2 = (unsigned char*) p + mcnt; /* Jump dest. */ | |
5663 | unsigned char *p3 = (unsigned char*) p - 3; /* opcode location. */ | |
fa9a63c5 | 5664 | |
505bde11 SM |
5665 | p -= 3; /* Reset so that we will re-execute the |
5666 | instruction once it's been changed. */ | |
fa9a63c5 | 5667 | |
4e8a9132 SM |
5668 | EXTRACT_NUMBER (mcnt, p2 - 2); |
5669 | ||
5670 | /* Ensure this is a indeed the trivial kind of loop | |
5671 | we are expecting. */ | |
5672 | assert (skip_one_char (p1) == p2 - 3); | |
5673 | assert ((re_opcode_t) p2[-3] == jump && p2 + mcnt == p); | |
99633e97 | 5674 | DEBUG_STATEMENT (debug += 2); |
505bde11 | 5675 | if (mutually_exclusive_p (bufp, p1, p2)) |
fa9a63c5 | 5676 | { |
505bde11 | 5677 | /* Use a fast `on_failure_keep_string_jump' loop. */ |
4e8a9132 | 5678 | DEBUG_PRINT1 (" smart exclusive => fast loop.\n"); |
01618498 | 5679 | *p3 = (unsigned char) on_failure_keep_string_jump; |
4e8a9132 | 5680 | STORE_NUMBER (p2 - 2, mcnt + 3); |
25fe55af | 5681 | } |
505bde11 | 5682 | else |
fa9a63c5 | 5683 | { |
505bde11 SM |
5684 | /* Default to a safe `on_failure_jump' loop. */ |
5685 | DEBUG_PRINT1 (" smart default => slow loop.\n"); | |
01618498 | 5686 | *p3 = (unsigned char) on_failure_jump; |
fa9a63c5 | 5687 | } |
99633e97 | 5688 | DEBUG_STATEMENT (debug -= 2); |
25fe55af | 5689 | } |
505bde11 | 5690 | break; |
25fe55af RS |
5691 | |
5692 | /* Unconditionally jump (without popping any failure points). */ | |
5693 | case jump: | |
fa9a63c5 | 5694 | unconditional_jump: |
5b370c2b | 5695 | IMMEDIATE_QUIT_CHECK; |
fa9a63c5 | 5696 | EXTRACT_NUMBER_AND_INCR (mcnt, p); /* Get the amount to jump. */ |
25fe55af RS |
5697 | DEBUG_PRINT2 ("EXECUTING jump %d ", mcnt); |
5698 | p += mcnt; /* Do the jump. */ | |
505bde11 | 5699 | DEBUG_PRINT2 ("(to %p).\n", p); |
25fe55af RS |
5700 | break; |
5701 | ||
5702 | ||
25fe55af RS |
5703 | /* Have to succeed matching what follows at least n times. |
5704 | After that, handle like `on_failure_jump'. */ | |
5705 | case succeed_n: | |
01618498 | 5706 | /* Signedness doesn't matter since we only compare MCNT to 0. */ |
25fe55af RS |
5707 | EXTRACT_NUMBER (mcnt, p + 2); |
5708 | DEBUG_PRINT2 ("EXECUTING succeed_n %d.\n", mcnt); | |
5e69f11e | 5709 | |
dc1e502d SM |
5710 | /* Originally, mcnt is how many times we HAVE to succeed. */ |
5711 | if (mcnt != 0) | |
25fe55af | 5712 | { |
6dcf2d0e SM |
5713 | /* Here, we discard `const', making re_match non-reentrant. */ |
5714 | unsigned char *p2 = (unsigned char*) p + 2; /* counter loc. */ | |
dc1e502d | 5715 | mcnt--; |
01618498 SM |
5716 | p += 4; |
5717 | PUSH_NUMBER (p2, mcnt); | |
25fe55af | 5718 | } |
dc1e502d SM |
5719 | else |
5720 | /* The two bytes encoding mcnt == 0 are two no_op opcodes. */ | |
5721 | goto on_failure; | |
25fe55af RS |
5722 | break; |
5723 | ||
5724 | case jump_n: | |
01618498 | 5725 | /* Signedness doesn't matter since we only compare MCNT to 0. */ |
25fe55af RS |
5726 | EXTRACT_NUMBER (mcnt, p + 2); |
5727 | DEBUG_PRINT2 ("EXECUTING jump_n %d.\n", mcnt); | |
5728 | ||
5729 | /* Originally, this is how many times we CAN jump. */ | |
dc1e502d | 5730 | if (mcnt != 0) |
25fe55af | 5731 | { |
6dcf2d0e SM |
5732 | /* Here, we discard `const', making re_match non-reentrant. */ |
5733 | unsigned char *p2 = (unsigned char*) p + 2; /* counter loc. */ | |
dc1e502d | 5734 | mcnt--; |
01618498 | 5735 | PUSH_NUMBER (p2, mcnt); |
dc1e502d | 5736 | goto unconditional_jump; |
25fe55af RS |
5737 | } |
5738 | /* If don't have to jump any more, skip over the rest of command. */ | |
5e69f11e RM |
5739 | else |
5740 | p += 4; | |
25fe55af | 5741 | break; |
5e69f11e | 5742 | |
fa9a63c5 RM |
5743 | case set_number_at: |
5744 | { | |
01618498 | 5745 | unsigned char *p2; /* Location of the counter. */ |
25fe55af | 5746 | DEBUG_PRINT1 ("EXECUTING set_number_at.\n"); |
fa9a63c5 | 5747 | |
25fe55af | 5748 | EXTRACT_NUMBER_AND_INCR (mcnt, p); |
6dcf2d0e SM |
5749 | /* Here, we discard `const', making re_match non-reentrant. */ |
5750 | p2 = (unsigned char*) p + mcnt; | |
01618498 | 5751 | /* Signedness doesn't matter since we only copy MCNT's bits . */ |
25fe55af | 5752 | EXTRACT_NUMBER_AND_INCR (mcnt, p); |
01618498 SM |
5753 | DEBUG_PRINT3 (" Setting %p to %d.\n", p2, mcnt); |
5754 | PUSH_NUMBER (p2, mcnt); | |
25fe55af RS |
5755 | break; |
5756 | } | |
9121ca40 KH |
5757 | |
5758 | case wordbound: | |
66f0296e SM |
5759 | case notwordbound: |
5760 | not = (re_opcode_t) *(p - 1) == notwordbound; | |
5761 | DEBUG_PRINT2 ("EXECUTING %swordbound.\n", not?"not":""); | |
fa9a63c5 | 5762 | |
99633e97 | 5763 | /* We SUCCEED (or FAIL) in one of the following cases: */ |
9121ca40 | 5764 | |
b18215fc | 5765 | /* Case 1: D is at the beginning or the end of string. */ |
9121ca40 | 5766 | if (AT_STRINGS_BEG (d) || AT_STRINGS_END (d)) |
66f0296e | 5767 | not = !not; |
b18215fc RS |
5768 | else |
5769 | { | |
5770 | /* C1 is the character before D, S1 is the syntax of C1, C2 | |
5771 | is the character at D, and S2 is the syntax of C2. */ | |
01618498 SM |
5772 | re_wchar_t c1, c2; |
5773 | int s1, s2; | |
b18215fc | 5774 | #ifdef emacs |
2d1675e4 SM |
5775 | int offset = PTR_TO_OFFSET (d - 1); |
5776 | int charpos = SYNTAX_TABLE_BYTE_TO_CHAR (offset); | |
5d967c7a | 5777 | UPDATE_SYNTAX_TABLE (charpos); |
25fe55af | 5778 | #endif |
66f0296e | 5779 | GET_CHAR_BEFORE_2 (c1, d, string1, end1, string2, end2); |
b18215fc RS |
5780 | s1 = SYNTAX (c1); |
5781 | #ifdef emacs | |
5d967c7a | 5782 | UPDATE_SYNTAX_TABLE_FORWARD (charpos + 1); |
25fe55af | 5783 | #endif |
f1ad044f | 5784 | PREFETCH_NOLIMIT (); |
2d1675e4 | 5785 | c2 = RE_STRING_CHAR (d, dend - d); |
b18215fc RS |
5786 | s2 = SYNTAX (c2); |
5787 | ||
5788 | if (/* Case 2: Only one of S1 and S2 is Sword. */ | |
5789 | ((s1 == Sword) != (s2 == Sword)) | |
5790 | /* Case 3: Both of S1 and S2 are Sword, and macro | |
25fe55af | 5791 | WORD_BOUNDARY_P (C1, C2) returns nonzero. */ |
b18215fc | 5792 | || ((s1 == Sword) && WORD_BOUNDARY_P (c1, c2))) |
66f0296e SM |
5793 | not = !not; |
5794 | } | |
5795 | if (not) | |
9121ca40 | 5796 | break; |
b18215fc | 5797 | else |
9121ca40 | 5798 | goto fail; |
fa9a63c5 RM |
5799 | |
5800 | case wordbeg: | |
25fe55af | 5801 | DEBUG_PRINT1 ("EXECUTING wordbeg.\n"); |
fa9a63c5 | 5802 | |
b18215fc RS |
5803 | /* We FAIL in one of the following cases: */ |
5804 | ||
25fe55af | 5805 | /* Case 1: D is at the end of string. */ |
b18215fc | 5806 | if (AT_STRINGS_END (d)) |
99633e97 | 5807 | goto fail; |
b18215fc RS |
5808 | else |
5809 | { | |
5810 | /* C1 is the character before D, S1 is the syntax of C1, C2 | |
5811 | is the character at D, and S2 is the syntax of C2. */ | |
01618498 SM |
5812 | re_wchar_t c1, c2; |
5813 | int s1, s2; | |
fa9a63c5 | 5814 | #ifdef emacs |
2d1675e4 SM |
5815 | int offset = PTR_TO_OFFSET (d); |
5816 | int charpos = SYNTAX_TABLE_BYTE_TO_CHAR (offset); | |
92432794 | 5817 | UPDATE_SYNTAX_TABLE (charpos); |
25fe55af | 5818 | #endif |
99633e97 | 5819 | PREFETCH (); |
2d1675e4 | 5820 | c2 = RE_STRING_CHAR (d, dend - d); |
b18215fc | 5821 | s2 = SYNTAX (c2); |
177c0ea7 | 5822 | |
b18215fc RS |
5823 | /* Case 2: S2 is not Sword. */ |
5824 | if (s2 != Sword) | |
5825 | goto fail; | |
5826 | ||
5827 | /* Case 3: D is not at the beginning of string ... */ | |
5828 | if (!AT_STRINGS_BEG (d)) | |
5829 | { | |
5830 | GET_CHAR_BEFORE_2 (c1, d, string1, end1, string2, end2); | |
5831 | #ifdef emacs | |
5d967c7a | 5832 | UPDATE_SYNTAX_TABLE_BACKWARD (charpos - 1); |
25fe55af | 5833 | #endif |
b18215fc RS |
5834 | s1 = SYNTAX (c1); |
5835 | ||
5836 | /* ... and S1 is Sword, and WORD_BOUNDARY_P (C1, C2) | |
25fe55af | 5837 | returns 0. */ |
b18215fc RS |
5838 | if ((s1 == Sword) && !WORD_BOUNDARY_P (c1, c2)) |
5839 | goto fail; | |
5840 | } | |
5841 | } | |
e318085a RS |
5842 | break; |
5843 | ||
b18215fc | 5844 | case wordend: |
25fe55af | 5845 | DEBUG_PRINT1 ("EXECUTING wordend.\n"); |
b18215fc RS |
5846 | |
5847 | /* We FAIL in one of the following cases: */ | |
5848 | ||
5849 | /* Case 1: D is at the beginning of string. */ | |
5850 | if (AT_STRINGS_BEG (d)) | |
e318085a | 5851 | goto fail; |
b18215fc RS |
5852 | else |
5853 | { | |
5854 | /* C1 is the character before D, S1 is the syntax of C1, C2 | |
5855 | is the character at D, and S2 is the syntax of C2. */ | |
01618498 SM |
5856 | re_wchar_t c1, c2; |
5857 | int s1, s2; | |
5d967c7a | 5858 | #ifdef emacs |
2d1675e4 SM |
5859 | int offset = PTR_TO_OFFSET (d) - 1; |
5860 | int charpos = SYNTAX_TABLE_BYTE_TO_CHAR (offset); | |
92432794 | 5861 | UPDATE_SYNTAX_TABLE (charpos); |
5d967c7a | 5862 | #endif |
99633e97 | 5863 | GET_CHAR_BEFORE_2 (c1, d, string1, end1, string2, end2); |
b18215fc RS |
5864 | s1 = SYNTAX (c1); |
5865 | ||
5866 | /* Case 2: S1 is not Sword. */ | |
5867 | if (s1 != Sword) | |
5868 | goto fail; | |
5869 | ||
5870 | /* Case 3: D is not at the end of string ... */ | |
5871 | if (!AT_STRINGS_END (d)) | |
5872 | { | |
f1ad044f | 5873 | PREFETCH_NOLIMIT (); |
2d1675e4 | 5874 | c2 = RE_STRING_CHAR (d, dend - d); |
5d967c7a | 5875 | #ifdef emacs |
134579f2 | 5876 | UPDATE_SYNTAX_TABLE_FORWARD (charpos + 1); |
5d967c7a | 5877 | #endif |
b18215fc RS |
5878 | s2 = SYNTAX (c2); |
5879 | ||
5880 | /* ... and S2 is Sword, and WORD_BOUNDARY_P (C1, C2) | |
25fe55af | 5881 | returns 0. */ |
b18215fc | 5882 | if ((s2 == Sword) && !WORD_BOUNDARY_P (c1, c2)) |
25fe55af | 5883 | goto fail; |
b18215fc RS |
5884 | } |
5885 | } | |
e318085a RS |
5886 | break; |
5887 | ||
669fa600 SM |
5888 | case symbeg: |
5889 | DEBUG_PRINT1 ("EXECUTING symbeg.\n"); | |
5890 | ||
5891 | /* We FAIL in one of the following cases: */ | |
5892 | ||
5893 | /* Case 1: D is at the end of string. */ | |
5894 | if (AT_STRINGS_END (d)) | |
5895 | goto fail; | |
5896 | else | |
5897 | { | |
5898 | /* C1 is the character before D, S1 is the syntax of C1, C2 | |
5899 | is the character at D, and S2 is the syntax of C2. */ | |
5900 | re_wchar_t c1, c2; | |
5901 | int s1, s2; | |
5902 | #ifdef emacs | |
5903 | int offset = PTR_TO_OFFSET (d); | |
5904 | int charpos = SYNTAX_TABLE_BYTE_TO_CHAR (offset); | |
5905 | UPDATE_SYNTAX_TABLE (charpos); | |
5906 | #endif | |
5907 | PREFETCH (); | |
5908 | c2 = RE_STRING_CHAR (d, dend - d); | |
5909 | s2 = SYNTAX (c2); | |
5910 | ||
5911 | /* Case 2: S2 is neither Sword nor Ssymbol. */ | |
5912 | if (s2 != Sword && s2 != Ssymbol) | |
5913 | goto fail; | |
5914 | ||
5915 | /* Case 3: D is not at the beginning of string ... */ | |
5916 | if (!AT_STRINGS_BEG (d)) | |
5917 | { | |
5918 | GET_CHAR_BEFORE_2 (c1, d, string1, end1, string2, end2); | |
5919 | #ifdef emacs | |
5920 | UPDATE_SYNTAX_TABLE_BACKWARD (charpos - 1); | |
5921 | #endif | |
5922 | s1 = SYNTAX (c1); | |
5923 | ||
5924 | /* ... and S1 is Sword or Ssymbol. */ | |
5925 | if (s1 == Sword || s1 == Ssymbol) | |
5926 | goto fail; | |
5927 | } | |
5928 | } | |
5929 | break; | |
5930 | ||
5931 | case symend: | |
5932 | DEBUG_PRINT1 ("EXECUTING symend.\n"); | |
5933 | ||
5934 | /* We FAIL in one of the following cases: */ | |
5935 | ||
5936 | /* Case 1: D is at the beginning of string. */ | |
5937 | if (AT_STRINGS_BEG (d)) | |
5938 | goto fail; | |
5939 | else | |
5940 | { | |
5941 | /* C1 is the character before D, S1 is the syntax of C1, C2 | |
5942 | is the character at D, and S2 is the syntax of C2. */ | |
5943 | re_wchar_t c1, c2; | |
5944 | int s1, s2; | |
5945 | #ifdef emacs | |
5946 | int offset = PTR_TO_OFFSET (d) - 1; | |
5947 | int charpos = SYNTAX_TABLE_BYTE_TO_CHAR (offset); | |
5948 | UPDATE_SYNTAX_TABLE (charpos); | |
5949 | #endif | |
5950 | GET_CHAR_BEFORE_2 (c1, d, string1, end1, string2, end2); | |
5951 | s1 = SYNTAX (c1); | |
5952 | ||
5953 | /* Case 2: S1 is neither Ssymbol nor Sword. */ | |
5954 | if (s1 != Sword && s1 != Ssymbol) | |
5955 | goto fail; | |
5956 | ||
5957 | /* Case 3: D is not at the end of string ... */ | |
5958 | if (!AT_STRINGS_END (d)) | |
5959 | { | |
5960 | PREFETCH_NOLIMIT (); | |
5961 | c2 = RE_STRING_CHAR (d, dend - d); | |
5962 | #ifdef emacs | |
134579f2 | 5963 | UPDATE_SYNTAX_TABLE_FORWARD (charpos + 1); |
669fa600 SM |
5964 | #endif |
5965 | s2 = SYNTAX (c2); | |
5966 | ||
5967 | /* ... and S2 is Sword or Ssymbol. */ | |
5968 | if (s2 == Sword || s2 == Ssymbol) | |
5969 | goto fail; | |
5970 | } | |
5971 | } | |
5972 | break; | |
5973 | ||
fa9a63c5 | 5974 | case syntaxspec: |
1fb352e0 SM |
5975 | case notsyntaxspec: |
5976 | not = (re_opcode_t) *(p - 1) == notsyntaxspec; | |
fa9a63c5 | 5977 | mcnt = *p++; |
1fb352e0 | 5978 | DEBUG_PRINT3 ("EXECUTING %ssyntaxspec %d.\n", not?"not":"", mcnt); |
fa9a63c5 | 5979 | PREFETCH (); |
b18215fc RS |
5980 | #ifdef emacs |
5981 | { | |
2d1675e4 SM |
5982 | int offset = PTR_TO_OFFSET (d); |
5983 | int pos1 = SYNTAX_TABLE_BYTE_TO_CHAR (offset); | |
b18215fc RS |
5984 | UPDATE_SYNTAX_TABLE (pos1); |
5985 | } | |
25fe55af | 5986 | #endif |
b18215fc | 5987 | { |
01618498 SM |
5988 | int len; |
5989 | re_wchar_t c; | |
b18215fc | 5990 | |
2d1675e4 | 5991 | c = RE_STRING_CHAR_AND_LENGTH (d, dend - d, len); |
b18215fc | 5992 | |
990b2375 | 5993 | if ((SYNTAX (c) != (enum syntaxcode) mcnt) ^ not) |
1fb352e0 | 5994 | goto fail; |
b18215fc RS |
5995 | d += len; |
5996 | } | |
fa9a63c5 RM |
5997 | break; |
5998 | ||
b18215fc | 5999 | #ifdef emacs |
1fb352e0 SM |
6000 | case before_dot: |
6001 | DEBUG_PRINT1 ("EXECUTING before_dot.\n"); | |
6002 | if (PTR_BYTE_POS (d) >= PT_BYTE) | |
fa9a63c5 | 6003 | goto fail; |
b18215fc RS |
6004 | break; |
6005 | ||
1fb352e0 SM |
6006 | case at_dot: |
6007 | DEBUG_PRINT1 ("EXECUTING at_dot.\n"); | |
6008 | if (PTR_BYTE_POS (d) != PT_BYTE) | |
6009 | goto fail; | |
6010 | break; | |
b18215fc | 6011 | |
1fb352e0 SM |
6012 | case after_dot: |
6013 | DEBUG_PRINT1 ("EXECUTING after_dot.\n"); | |
6014 | if (PTR_BYTE_POS (d) <= PT_BYTE) | |
6015 | goto fail; | |
e318085a | 6016 | break; |
fa9a63c5 | 6017 | |
1fb352e0 | 6018 | case categoryspec: |
b18215fc | 6019 | case notcategoryspec: |
1fb352e0 | 6020 | not = (re_opcode_t) *(p - 1) == notcategoryspec; |
b18215fc | 6021 | mcnt = *p++; |
1fb352e0 | 6022 | DEBUG_PRINT3 ("EXECUTING %scategoryspec %d.\n", not?"not":"", mcnt); |
b18215fc RS |
6023 | PREFETCH (); |
6024 | { | |
01618498 SM |
6025 | int len; |
6026 | re_wchar_t c; | |
6027 | ||
2d1675e4 | 6028 | c = RE_STRING_CHAR_AND_LENGTH (d, dend - d, len); |
b18215fc | 6029 | |
1fb352e0 | 6030 | if ((!CHAR_HAS_CATEGORY (c, mcnt)) ^ not) |
b18215fc RS |
6031 | goto fail; |
6032 | d += len; | |
6033 | } | |
fa9a63c5 | 6034 | break; |
5e69f11e | 6035 | |
1fb352e0 | 6036 | #endif /* emacs */ |
5e69f11e | 6037 | |
0b32bf0e SM |
6038 | default: |
6039 | abort (); | |
fa9a63c5 | 6040 | } |
b18215fc | 6041 | continue; /* Successfully executed one pattern command; keep going. */ |
fa9a63c5 RM |
6042 | |
6043 | ||
6044 | /* We goto here if a matching operation fails. */ | |
6045 | fail: | |
5b370c2b | 6046 | IMMEDIATE_QUIT_CHECK; |
fa9a63c5 | 6047 | if (!FAIL_STACK_EMPTY ()) |
505bde11 | 6048 | { |
01618498 | 6049 | re_char *str, *pat; |
505bde11 | 6050 | /* A restart point is known. Restore to that state. */ |
0b32bf0e SM |
6051 | DEBUG_PRINT1 ("\nFAIL:\n"); |
6052 | POP_FAILURE_POINT (str, pat); | |
505bde11 SM |
6053 | switch (SWITCH_ENUM_CAST ((re_opcode_t) *pat++)) |
6054 | { | |
6055 | case on_failure_keep_string_jump: | |
6056 | assert (str == NULL); | |
6057 | goto continue_failure_jump; | |
6058 | ||
0683b6fa SM |
6059 | case on_failure_jump_nastyloop: |
6060 | assert ((re_opcode_t)pat[-2] == no_op); | |
6061 | PUSH_FAILURE_POINT (pat - 2, str); | |
6062 | /* Fallthrough */ | |
6063 | ||
505bde11 SM |
6064 | case on_failure_jump_loop: |
6065 | case on_failure_jump: | |
6066 | case succeed_n: | |
6067 | d = str; | |
6068 | continue_failure_jump: | |
6069 | EXTRACT_NUMBER_AND_INCR (mcnt, pat); | |
6070 | p = pat + mcnt; | |
6071 | break; | |
b18215fc | 6072 | |
0683b6fa SM |
6073 | case no_op: |
6074 | /* A special frame used for nastyloops. */ | |
6075 | goto fail; | |
6076 | ||
505bde11 SM |
6077 | default: |
6078 | abort(); | |
6079 | } | |
fa9a63c5 | 6080 | |
505bde11 | 6081 | assert (p >= bufp->buffer && p <= pend); |
b18215fc | 6082 | |
0b32bf0e | 6083 | if (d >= string1 && d <= end1) |
fa9a63c5 | 6084 | dend = end_match_1; |
0b32bf0e | 6085 | } |
fa9a63c5 | 6086 | else |
0b32bf0e | 6087 | break; /* Matching at this starting point really fails. */ |
fa9a63c5 RM |
6088 | } /* for (;;) */ |
6089 | ||
6090 | if (best_regs_set) | |
6091 | goto restore_best_regs; | |
6092 | ||
6093 | FREE_VARIABLES (); | |
6094 | ||
b18215fc | 6095 | return -1; /* Failure to match. */ |
fa9a63c5 RM |
6096 | } /* re_match_2 */ |
6097 | \f | |
6098 | /* Subroutine definitions for re_match_2. */ | |
6099 | ||
fa9a63c5 RM |
6100 | /* Return zero if TRANSLATE[S1] and TRANSLATE[S2] are identical for LEN |
6101 | bytes; nonzero otherwise. */ | |
5e69f11e | 6102 | |
fa9a63c5 | 6103 | static int |
2d1675e4 SM |
6104 | bcmp_translate (s1, s2, len, translate, multibyte) |
6105 | re_char *s1, *s2; | |
fa9a63c5 | 6106 | register int len; |
6676cb1c | 6107 | RE_TRANSLATE_TYPE translate; |
2d1675e4 | 6108 | const int multibyte; |
fa9a63c5 | 6109 | { |
2d1675e4 SM |
6110 | register re_char *p1 = s1, *p2 = s2; |
6111 | re_char *p1_end = s1 + len; | |
6112 | re_char *p2_end = s2 + len; | |
e934739e | 6113 | |
4bb91c68 SM |
6114 | /* FIXME: Checking both p1 and p2 presumes that the two strings might have |
6115 | different lengths, but relying on a single `len' would break this. -sm */ | |
6116 | while (p1 < p1_end && p2 < p2_end) | |
fa9a63c5 | 6117 | { |
e934739e | 6118 | int p1_charlen, p2_charlen; |
01618498 | 6119 | re_wchar_t p1_ch, p2_ch; |
e934739e | 6120 | |
2d1675e4 SM |
6121 | p1_ch = RE_STRING_CHAR_AND_LENGTH (p1, p1_end - p1, p1_charlen); |
6122 | p2_ch = RE_STRING_CHAR_AND_LENGTH (p2, p2_end - p2, p2_charlen); | |
e934739e RS |
6123 | |
6124 | if (RE_TRANSLATE (translate, p1_ch) | |
6125 | != RE_TRANSLATE (translate, p2_ch)) | |
bc192b5b | 6126 | return 1; |
e934739e RS |
6127 | |
6128 | p1 += p1_charlen, p2 += p2_charlen; | |
fa9a63c5 | 6129 | } |
e934739e RS |
6130 | |
6131 | if (p1 != p1_end || p2 != p2_end) | |
6132 | return 1; | |
6133 | ||
fa9a63c5 RM |
6134 | return 0; |
6135 | } | |
6136 | \f | |
6137 | /* Entry points for GNU code. */ | |
6138 | ||
6139 | /* re_compile_pattern is the GNU regular expression compiler: it | |
6140 | compiles PATTERN (of length SIZE) and puts the result in BUFP. | |
6141 | Returns 0 if the pattern was valid, otherwise an error string. | |
5e69f11e | 6142 | |
fa9a63c5 RM |
6143 | Assumes the `allocated' (and perhaps `buffer') and `translate' fields |
6144 | are set in BUFP on entry. | |
5e69f11e | 6145 | |
b18215fc | 6146 | We call regex_compile to do the actual compilation. */ |
fa9a63c5 RM |
6147 | |
6148 | const char * | |
6149 | re_compile_pattern (pattern, length, bufp) | |
6150 | const char *pattern; | |
0b32bf0e | 6151 | size_t length; |
fa9a63c5 RM |
6152 | struct re_pattern_buffer *bufp; |
6153 | { | |
6154 | reg_errcode_t ret; | |
5e69f11e | 6155 | |
fa9a63c5 RM |
6156 | /* GNU code is written to assume at least RE_NREGS registers will be set |
6157 | (and at least one extra will be -1). */ | |
6158 | bufp->regs_allocated = REGS_UNALLOCATED; | |
5e69f11e | 6159 | |
fa9a63c5 RM |
6160 | /* And GNU code determines whether or not to get register information |
6161 | by passing null for the REGS argument to re_match, etc., not by | |
6162 | setting no_sub. */ | |
6163 | bufp->no_sub = 0; | |
5e69f11e | 6164 | |
4bb91c68 | 6165 | ret = regex_compile ((re_char*) pattern, length, re_syntax_options, bufp); |
fa9a63c5 RM |
6166 | |
6167 | if (!ret) | |
6168 | return NULL; | |
6169 | return gettext (re_error_msgid[(int) ret]); | |
5e69f11e | 6170 | } |
c0f9ea08 | 6171 | WEAK_ALIAS (__re_compile_pattern, re_compile_pattern) |
fa9a63c5 | 6172 | \f |
b18215fc RS |
6173 | /* Entry points compatible with 4.2 BSD regex library. We don't define |
6174 | them unless specifically requested. */ | |
fa9a63c5 | 6175 | |
0b32bf0e | 6176 | #if defined _REGEX_RE_COMP || defined _LIBC |
fa9a63c5 RM |
6177 | |
6178 | /* BSD has one and only one pattern buffer. */ | |
6179 | static struct re_pattern_buffer re_comp_buf; | |
6180 | ||
6181 | char * | |
0b32bf0e | 6182 | # ifdef _LIBC |
48afdd44 RM |
6183 | /* Make these definitions weak in libc, so POSIX programs can redefine |
6184 | these names if they don't use our functions, and still use | |
6185 | regcomp/regexec below without link errors. */ | |
6186 | weak_function | |
0b32bf0e | 6187 | # endif |
fa9a63c5 RM |
6188 | re_comp (s) |
6189 | const char *s; | |
6190 | { | |
6191 | reg_errcode_t ret; | |
5e69f11e | 6192 | |
fa9a63c5 RM |
6193 | if (!s) |
6194 | { | |
6195 | if (!re_comp_buf.buffer) | |
0b32bf0e | 6196 | /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */ |
a60198e5 | 6197 | return (char *) gettext ("No previous regular expression"); |
fa9a63c5 RM |
6198 | return 0; |
6199 | } | |
6200 | ||
6201 | if (!re_comp_buf.buffer) | |
6202 | { | |
6203 | re_comp_buf.buffer = (unsigned char *) malloc (200); | |
6204 | if (re_comp_buf.buffer == NULL) | |
0b32bf0e SM |
6205 | /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */ |
6206 | return (char *) gettext (re_error_msgid[(int) REG_ESPACE]); | |
fa9a63c5 RM |
6207 | re_comp_buf.allocated = 200; |
6208 | ||
6209 | re_comp_buf.fastmap = (char *) malloc (1 << BYTEWIDTH); | |
6210 | if (re_comp_buf.fastmap == NULL) | |
a60198e5 SM |
6211 | /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */ |
6212 | return (char *) gettext (re_error_msgid[(int) REG_ESPACE]); | |
fa9a63c5 RM |
6213 | } |
6214 | ||
6215 | /* Since `re_exec' always passes NULL for the `regs' argument, we | |
6216 | don't need to initialize the pattern buffer fields which affect it. */ | |
6217 | ||
fa9a63c5 | 6218 | ret = regex_compile (s, strlen (s), re_syntax_options, &re_comp_buf); |
5e69f11e | 6219 | |
fa9a63c5 RM |
6220 | if (!ret) |
6221 | return NULL; | |
6222 | ||
6223 | /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */ | |
6224 | return (char *) gettext (re_error_msgid[(int) ret]); | |
6225 | } | |
6226 | ||
6227 | ||
6228 | int | |
0b32bf0e | 6229 | # ifdef _LIBC |
48afdd44 | 6230 | weak_function |
0b32bf0e | 6231 | # endif |
fa9a63c5 RM |
6232 | re_exec (s) |
6233 | const char *s; | |
6234 | { | |
6235 | const int len = strlen (s); | |
6236 | return | |
6237 | 0 <= re_search (&re_comp_buf, s, len, 0, len, (struct re_registers *) 0); | |
6238 | } | |
6239 | #endif /* _REGEX_RE_COMP */ | |
6240 | \f | |
6241 | /* POSIX.2 functions. Don't define these for Emacs. */ | |
6242 | ||
6243 | #ifndef emacs | |
6244 | ||
6245 | /* regcomp takes a regular expression as a string and compiles it. | |
6246 | ||
b18215fc | 6247 | PREG is a regex_t *. We do not expect any fields to be initialized, |
fa9a63c5 RM |
6248 | since POSIX says we shouldn't. Thus, we set |
6249 | ||
6250 | `buffer' to the compiled pattern; | |
6251 | `used' to the length of the compiled pattern; | |
6252 | `syntax' to RE_SYNTAX_POSIX_EXTENDED if the | |
6253 | REG_EXTENDED bit in CFLAGS is set; otherwise, to | |
6254 | RE_SYNTAX_POSIX_BASIC; | |
c0f9ea08 SM |
6255 | `fastmap' to an allocated space for the fastmap; |
6256 | `fastmap_accurate' to zero; | |
fa9a63c5 RM |
6257 | `re_nsub' to the number of subexpressions in PATTERN. |
6258 | ||
6259 | PATTERN is the address of the pattern string. | |
6260 | ||
6261 | CFLAGS is a series of bits which affect compilation. | |
6262 | ||
6263 | If REG_EXTENDED is set, we use POSIX extended syntax; otherwise, we | |
6264 | use POSIX basic syntax. | |
6265 | ||
6266 | If REG_NEWLINE is set, then . and [^...] don't match newline. | |
6267 | Also, regexec will try a match beginning after every newline. | |
6268 | ||
6269 | If REG_ICASE is set, then we considers upper- and lowercase | |
6270 | versions of letters to be equivalent when matching. | |
6271 | ||
6272 | If REG_NOSUB is set, then when PREG is passed to regexec, that | |
6273 | routine will report only success or failure, and nothing about the | |
6274 | registers. | |
6275 | ||
b18215fc | 6276 | It returns 0 if it succeeds, nonzero if it doesn't. (See regex.h for |
fa9a63c5 RM |
6277 | the return codes and their meanings.) */ |
6278 | ||
6279 | int | |
6280 | regcomp (preg, pattern, cflags) | |
ada30c0e SM |
6281 | regex_t *__restrict preg; |
6282 | const char *__restrict pattern; | |
fa9a63c5 RM |
6283 | int cflags; |
6284 | { | |
6285 | reg_errcode_t ret; | |
4bb91c68 | 6286 | reg_syntax_t syntax |
fa9a63c5 RM |
6287 | = (cflags & REG_EXTENDED) ? |
6288 | RE_SYNTAX_POSIX_EXTENDED : RE_SYNTAX_POSIX_BASIC; | |
6289 | ||
6290 | /* regex_compile will allocate the space for the compiled pattern. */ | |
6291 | preg->buffer = 0; | |
6292 | preg->allocated = 0; | |
6293 | preg->used = 0; | |
5e69f11e | 6294 | |
c0f9ea08 SM |
6295 | /* Try to allocate space for the fastmap. */ |
6296 | preg->fastmap = (char *) malloc (1 << BYTEWIDTH); | |
5e69f11e | 6297 | |
fa9a63c5 RM |
6298 | if (cflags & REG_ICASE) |
6299 | { | |
6300 | unsigned i; | |
5e69f11e | 6301 | |
6676cb1c RS |
6302 | preg->translate |
6303 | = (RE_TRANSLATE_TYPE) malloc (CHAR_SET_SIZE | |
6304 | * sizeof (*(RE_TRANSLATE_TYPE)0)); | |
fa9a63c5 | 6305 | if (preg->translate == NULL) |
0b32bf0e | 6306 | return (int) REG_ESPACE; |
fa9a63c5 RM |
6307 | |
6308 | /* Map uppercase characters to corresponding lowercase ones. */ | |
6309 | for (i = 0; i < CHAR_SET_SIZE; i++) | |
4bb91c68 | 6310 | preg->translate[i] = ISUPPER (i) ? TOLOWER (i) : i; |
fa9a63c5 RM |
6311 | } |
6312 | else | |
6313 | preg->translate = NULL; | |
6314 | ||
6315 | /* If REG_NEWLINE is set, newlines are treated differently. */ | |
6316 | if (cflags & REG_NEWLINE) | |
6317 | { /* REG_NEWLINE implies neither . nor [^...] match newline. */ | |
6318 | syntax &= ~RE_DOT_NEWLINE; | |
6319 | syntax |= RE_HAT_LISTS_NOT_NEWLINE; | |
fa9a63c5 RM |
6320 | } |
6321 | else | |
c0f9ea08 | 6322 | syntax |= RE_NO_NEWLINE_ANCHOR; |
fa9a63c5 RM |
6323 | |
6324 | preg->no_sub = !!(cflags & REG_NOSUB); | |
6325 | ||
5e69f11e | 6326 | /* POSIX says a null character in the pattern terminates it, so we |
fa9a63c5 | 6327 | can use strlen here in compiling the pattern. */ |
4bb91c68 | 6328 | ret = regex_compile ((re_char*) pattern, strlen (pattern), syntax, preg); |
5e69f11e | 6329 | |
fa9a63c5 RM |
6330 | /* POSIX doesn't distinguish between an unmatched open-group and an |
6331 | unmatched close-group: both are REG_EPAREN. */ | |
c0f9ea08 SM |
6332 | if (ret == REG_ERPAREN) |
6333 | ret = REG_EPAREN; | |
6334 | ||
6335 | if (ret == REG_NOERROR && preg->fastmap) | |
6336 | { /* Compute the fastmap now, since regexec cannot modify the pattern | |
6337 | buffer. */ | |
6338 | re_compile_fastmap (preg); | |
6339 | if (preg->can_be_null) | |
6340 | { /* The fastmap can't be used anyway. */ | |
6341 | free (preg->fastmap); | |
6342 | preg->fastmap = NULL; | |
6343 | } | |
6344 | } | |
fa9a63c5 RM |
6345 | return (int) ret; |
6346 | } | |
c0f9ea08 | 6347 | WEAK_ALIAS (__regcomp, regcomp) |
fa9a63c5 RM |
6348 | |
6349 | ||
6350 | /* regexec searches for a given pattern, specified by PREG, in the | |
6351 | string STRING. | |
5e69f11e | 6352 | |
fa9a63c5 | 6353 | If NMATCH is zero or REG_NOSUB was set in the cflags argument to |
b18215fc | 6354 | `regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at |
fa9a63c5 RM |
6355 | least NMATCH elements, and we set them to the offsets of the |
6356 | corresponding matched substrings. | |
5e69f11e | 6357 | |
fa9a63c5 RM |
6358 | EFLAGS specifies `execution flags' which affect matching: if |
6359 | REG_NOTBOL is set, then ^ does not match at the beginning of the | |
6360 | string; if REG_NOTEOL is set, then $ does not match at the end. | |
5e69f11e | 6361 | |
fa9a63c5 RM |
6362 | We return 0 if we find a match and REG_NOMATCH if not. */ |
6363 | ||
6364 | int | |
6365 | regexec (preg, string, nmatch, pmatch, eflags) | |
ada30c0e SM |
6366 | const regex_t *__restrict preg; |
6367 | const char *__restrict string; | |
5e69f11e | 6368 | size_t nmatch; |
9f2dbe01 | 6369 | regmatch_t pmatch[__restrict_arr]; |
fa9a63c5 RM |
6370 | int eflags; |
6371 | { | |
6372 | int ret; | |
6373 | struct re_registers regs; | |
6374 | regex_t private_preg; | |
6375 | int len = strlen (string); | |
c0f9ea08 | 6376 | boolean want_reg_info = !preg->no_sub && nmatch > 0 && pmatch; |
fa9a63c5 RM |
6377 | |
6378 | private_preg = *preg; | |
5e69f11e | 6379 | |
fa9a63c5 RM |
6380 | private_preg.not_bol = !!(eflags & REG_NOTBOL); |
6381 | private_preg.not_eol = !!(eflags & REG_NOTEOL); | |
5e69f11e | 6382 | |
fa9a63c5 RM |
6383 | /* The user has told us exactly how many registers to return |
6384 | information about, via `nmatch'. We have to pass that on to the | |
b18215fc | 6385 | matching routines. */ |
fa9a63c5 | 6386 | private_preg.regs_allocated = REGS_FIXED; |
5e69f11e | 6387 | |
fa9a63c5 RM |
6388 | if (want_reg_info) |
6389 | { | |
6390 | regs.num_regs = nmatch; | |
4bb91c68 SM |
6391 | regs.start = TALLOC (nmatch * 2, regoff_t); |
6392 | if (regs.start == NULL) | |
0b32bf0e | 6393 | return (int) REG_NOMATCH; |
4bb91c68 | 6394 | regs.end = regs.start + nmatch; |
fa9a63c5 RM |
6395 | } |
6396 | ||
c0f9ea08 SM |
6397 | /* Instead of using not_eol to implement REG_NOTEOL, we could simply |
6398 | pass (&private_preg, string, len + 1, 0, len, ...) pretending the string | |
6399 | was a little bit longer but still only matching the real part. | |
6400 | This works because the `endline' will check for a '\n' and will find a | |
6401 | '\0', correctly deciding that this is not the end of a line. | |
6402 | But it doesn't work out so nicely for REG_NOTBOL, since we don't have | |
6403 | a convenient '\0' there. For all we know, the string could be preceded | |
6404 | by '\n' which would throw things off. */ | |
6405 | ||
fa9a63c5 RM |
6406 | /* Perform the searching operation. */ |
6407 | ret = re_search (&private_preg, string, len, | |
0b32bf0e SM |
6408 | /* start: */ 0, /* range: */ len, |
6409 | want_reg_info ? ®s : (struct re_registers *) 0); | |
5e69f11e | 6410 | |
fa9a63c5 RM |
6411 | /* Copy the register information to the POSIX structure. */ |
6412 | if (want_reg_info) | |
6413 | { | |
6414 | if (ret >= 0) | |
0b32bf0e SM |
6415 | { |
6416 | unsigned r; | |
fa9a63c5 | 6417 | |
0b32bf0e SM |
6418 | for (r = 0; r < nmatch; r++) |
6419 | { | |
6420 | pmatch[r].rm_so = regs.start[r]; | |
6421 | pmatch[r].rm_eo = regs.end[r]; | |
6422 | } | |
6423 | } | |
fa9a63c5 | 6424 | |
b18215fc | 6425 | /* If we needed the temporary register info, free the space now. */ |
fa9a63c5 | 6426 | free (regs.start); |
fa9a63c5 RM |
6427 | } |
6428 | ||
6429 | /* We want zero return to mean success, unlike `re_search'. */ | |
6430 | return ret >= 0 ? (int) REG_NOERROR : (int) REG_NOMATCH; | |
6431 | } | |
c0f9ea08 | 6432 | WEAK_ALIAS (__regexec, regexec) |
fa9a63c5 RM |
6433 | |
6434 | ||
6435 | /* Returns a message corresponding to an error code, ERRCODE, returned | |
6436 | from either regcomp or regexec. We don't use PREG here. */ | |
6437 | ||
6438 | size_t | |
6439 | regerror (errcode, preg, errbuf, errbuf_size) | |
6440 | int errcode; | |
6441 | const regex_t *preg; | |
6442 | char *errbuf; | |
6443 | size_t errbuf_size; | |
6444 | { | |
6445 | const char *msg; | |
6446 | size_t msg_size; | |
6447 | ||
6448 | if (errcode < 0 | |
6449 | || errcode >= (sizeof (re_error_msgid) / sizeof (re_error_msgid[0]))) | |
5e69f11e | 6450 | /* Only error codes returned by the rest of the code should be passed |
b18215fc | 6451 | to this routine. If we are given anything else, or if other regex |
fa9a63c5 RM |
6452 | code generates an invalid error code, then the program has a bug. |
6453 | Dump core so we can fix it. */ | |
6454 | abort (); | |
6455 | ||
6456 | msg = gettext (re_error_msgid[errcode]); | |
6457 | ||
6458 | msg_size = strlen (msg) + 1; /* Includes the null. */ | |
5e69f11e | 6459 | |
fa9a63c5 RM |
6460 | if (errbuf_size != 0) |
6461 | { | |
6462 | if (msg_size > errbuf_size) | |
0b32bf0e SM |
6463 | { |
6464 | strncpy (errbuf, msg, errbuf_size - 1); | |
6465 | errbuf[errbuf_size - 1] = 0; | |
6466 | } | |
fa9a63c5 | 6467 | else |
0b32bf0e | 6468 | strcpy (errbuf, msg); |
fa9a63c5 RM |
6469 | } |
6470 | ||
6471 | return msg_size; | |
6472 | } | |
c0f9ea08 | 6473 | WEAK_ALIAS (__regerror, regerror) |
fa9a63c5 RM |
6474 | |
6475 | ||
6476 | /* Free dynamically allocated space used by PREG. */ | |
6477 | ||
6478 | void | |
6479 | regfree (preg) | |
6480 | regex_t *preg; | |
6481 | { | |
6482 | if (preg->buffer != NULL) | |
6483 | free (preg->buffer); | |
6484 | preg->buffer = NULL; | |
5e69f11e | 6485 | |
fa9a63c5 RM |
6486 | preg->allocated = 0; |
6487 | preg->used = 0; | |
6488 | ||
6489 | if (preg->fastmap != NULL) | |
6490 | free (preg->fastmap); | |
6491 | preg->fastmap = NULL; | |
6492 | preg->fastmap_accurate = 0; | |
6493 | ||
6494 | if (preg->translate != NULL) | |
6495 | free (preg->translate); | |
6496 | preg->translate = NULL; | |
6497 | } | |
c0f9ea08 | 6498 | WEAK_ALIAS (__regfree, regfree) |
fa9a63c5 RM |
6499 | |
6500 | #endif /* not emacs */ | |
ab5796a9 MB |
6501 | |
6502 | /* arch-tag: 4ffd68ba-2a9e-435b-a21a-018990f9eeb2 | |
6503 | (do not change this comment) */ |