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 | ||
669fa600 | 5 | Copyright (C) 1993,94,95,96,97,98,99,2000,04 Free Software Foundation, Inc. |
bc78d348 | 6 | |
fa9a63c5 RM |
7 | This program is free software; you can redistribute it and/or modify |
8 | it under the terms of the GNU General Public License as published by | |
9 | the Free Software Foundation; either version 2, or (at your option) | |
10 | any later version. | |
11 | ||
12 | This program is distributed in the hope that it will be useful, | |
13 | but WITHOUT ANY WARRANTY; without even the implied warranty of | |
25fe55af | 14 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
fa9a63c5 RM |
15 | GNU General Public License for more details. |
16 | ||
17 | You should have received a copy of the GNU General Public License | |
18 | along with this program; if not, write to the Free Software | |
ba4a8e51 | 19 | Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, |
25fe55af | 20 | USA. */ |
fa9a63c5 | 21 | |
6df42991 | 22 | /* TODO: |
505bde11 | 23 | - structure the opcode space into opcode+flag. |
dc1e502d | 24 | - merge with glibc's regex.[ch]. |
01618498 | 25 | - replace (succeed_n + jump_n + set_number_at) with something that doesn't |
6dcf2d0e SM |
26 | need to modify the compiled regexp so that re_match can be reentrant. |
27 | - get rid of on_failure_jump_smart by doing the optimization in re_comp | |
28 | rather than at run-time, so that re_match can be reentrant. | |
01618498 | 29 | */ |
505bde11 | 30 | |
fa9a63c5 | 31 | /* AIX requires this to be the first thing in the file. */ |
0b32bf0e | 32 | #if defined _AIX && !defined REGEX_MALLOC |
fa9a63c5 RM |
33 | #pragma alloca |
34 | #endif | |
35 | ||
fa9a63c5 | 36 | #ifdef HAVE_CONFIG_H |
0b32bf0e | 37 | # include <config.h> |
fa9a63c5 RM |
38 | #endif |
39 | ||
4bb91c68 SM |
40 | #if defined STDC_HEADERS && !defined emacs |
41 | # include <stddef.h> | |
42 | #else | |
43 | /* We need this for `regex.h', and perhaps for the Emacs include files. */ | |
44 | # include <sys/types.h> | |
45 | #endif | |
fa9a63c5 | 46 | |
14473664 SM |
47 | /* Whether to use ISO C Amendment 1 wide char functions. |
48 | Those should not be used for Emacs since it uses its own. */ | |
5e5388f6 GM |
49 | #if defined _LIBC |
50 | #define WIDE_CHAR_SUPPORT 1 | |
51 | #else | |
14473664 | 52 | #define WIDE_CHAR_SUPPORT \ |
5e5388f6 GM |
53 | (HAVE_WCTYPE_H && HAVE_WCHAR_H && HAVE_BTOWC && !emacs) |
54 | #endif | |
14473664 SM |
55 | |
56 | /* For platform which support the ISO C amendement 1 functionality we | |
57 | support user defined character classes. */ | |
a0ad02f7 | 58 | #if WIDE_CHAR_SUPPORT |
14473664 SM |
59 | /* Solaris 2.5 has a bug: <wchar.h> must be included before <wctype.h>. */ |
60 | # include <wchar.h> | |
61 | # include <wctype.h> | |
62 | #endif | |
63 | ||
c0f9ea08 SM |
64 | #ifdef _LIBC |
65 | /* We have to keep the namespace clean. */ | |
66 | # define regfree(preg) __regfree (preg) | |
67 | # define regexec(pr, st, nm, pm, ef) __regexec (pr, st, nm, pm, ef) | |
68 | # define regcomp(preg, pattern, cflags) __regcomp (preg, pattern, cflags) | |
69 | # define regerror(errcode, preg, errbuf, errbuf_size) \ | |
70 | __regerror(errcode, preg, errbuf, errbuf_size) | |
71 | # define re_set_registers(bu, re, nu, st, en) \ | |
72 | __re_set_registers (bu, re, nu, st, en) | |
73 | # define re_match_2(bufp, string1, size1, string2, size2, pos, regs, stop) \ | |
74 | __re_match_2 (bufp, string1, size1, string2, size2, pos, regs, stop) | |
75 | # define re_match(bufp, string, size, pos, regs) \ | |
76 | __re_match (bufp, string, size, pos, regs) | |
77 | # define re_search(bufp, string, size, startpos, range, regs) \ | |
78 | __re_search (bufp, string, size, startpos, range, regs) | |
79 | # define re_compile_pattern(pattern, length, bufp) \ | |
80 | __re_compile_pattern (pattern, length, bufp) | |
81 | # define re_set_syntax(syntax) __re_set_syntax (syntax) | |
82 | # define re_search_2(bufp, st1, s1, st2, s2, startpos, range, regs, stop) \ | |
83 | __re_search_2 (bufp, st1, s1, st2, s2, startpos, range, regs, stop) | |
84 | # define re_compile_fastmap(bufp) __re_compile_fastmap (bufp) | |
85 | ||
14473664 SM |
86 | /* Make sure we call libc's function even if the user overrides them. */ |
87 | # define btowc __btowc | |
88 | # define iswctype __iswctype | |
89 | # define wctype __wctype | |
90 | ||
c0f9ea08 SM |
91 | # define WEAK_ALIAS(a,b) weak_alias (a, b) |
92 | ||
93 | /* We are also using some library internals. */ | |
94 | # include <locale/localeinfo.h> | |
95 | # include <locale/elem-hash.h> | |
96 | # include <langinfo.h> | |
97 | #else | |
98 | # define WEAK_ALIAS(a,b) | |
99 | #endif | |
100 | ||
4bb91c68 | 101 | /* This is for other GNU distributions with internationalized messages. */ |
0b32bf0e | 102 | #if HAVE_LIBINTL_H || defined _LIBC |
fa9a63c5 RM |
103 | # include <libintl.h> |
104 | #else | |
105 | # define gettext(msgid) (msgid) | |
106 | #endif | |
107 | ||
5e69f11e RM |
108 | #ifndef gettext_noop |
109 | /* This define is so xgettext can find the internationalizable | |
110 | strings. */ | |
0b32bf0e | 111 | # define gettext_noop(String) String |
5e69f11e RM |
112 | #endif |
113 | ||
fa9a63c5 RM |
114 | /* The `emacs' switch turns on certain matching commands |
115 | that make sense only in Emacs. */ | |
116 | #ifdef emacs | |
117 | ||
0b32bf0e SM |
118 | # include "lisp.h" |
119 | # include "buffer.h" | |
b18215fc RS |
120 | |
121 | /* Make syntax table lookup grant data in gl_state. */ | |
0b32bf0e | 122 | # define SYNTAX_ENTRY_VIA_PROPERTY |
b18215fc | 123 | |
0b32bf0e SM |
124 | # include "syntax.h" |
125 | # include "charset.h" | |
126 | # include "category.h" | |
fa9a63c5 | 127 | |
7689ef0b EZ |
128 | # ifdef malloc |
129 | # undef malloc | |
130 | # endif | |
0b32bf0e | 131 | # define malloc xmalloc |
7689ef0b EZ |
132 | # ifdef realloc |
133 | # undef realloc | |
134 | # endif | |
0b32bf0e | 135 | # define realloc xrealloc |
7689ef0b EZ |
136 | # ifdef free |
137 | # undef free | |
138 | # endif | |
0b32bf0e | 139 | # define free xfree |
9abbd165 | 140 | |
0b32bf0e SM |
141 | /* Converts the pointer to the char to BEG-based offset from the start. */ |
142 | # define PTR_TO_OFFSET(d) POS_AS_IN_BUFFER (POINTER_TO_OFFSET (d)) | |
143 | # define POS_AS_IN_BUFFER(p) ((p) + (NILP (re_match_object) || BUFFERP (re_match_object))) | |
144 | ||
145 | # define RE_MULTIBYTE_P(bufp) ((bufp)->multibyte) | |
146 | # define RE_STRING_CHAR(p, s) \ | |
4e8a9132 | 147 | (multibyte ? (STRING_CHAR (p, s)) : (*(p))) |
0b32bf0e | 148 | # define RE_STRING_CHAR_AND_LENGTH(p, s, len) \ |
2d1675e4 SM |
149 | (multibyte ? (STRING_CHAR_AND_LENGTH (p, s, len)) : ((len) = 1, *(p))) |
150 | ||
151 | /* Set C a (possibly multibyte) character before P. P points into a | |
152 | string which is the virtual concatenation of STR1 (which ends at | |
153 | END1) or STR2 (which ends at END2). */ | |
0b32bf0e | 154 | # define GET_CHAR_BEFORE_2(c, p, str1, end1, str2, end2) \ |
2d1675e4 SM |
155 | do { \ |
156 | if (multibyte) \ | |
157 | { \ | |
0b32bf0e SM |
158 | re_char *dtemp = (p) == (str2) ? (end1) : (p); \ |
159 | re_char *dlimit = ((p) > (str2) && (p) <= (end2)) ? (str2) : (str1); \ | |
70806df6 KH |
160 | re_char *d0 = dtemp; \ |
161 | PREV_CHAR_BOUNDARY (d0, dlimit); \ | |
162 | c = STRING_CHAR (d0, dtemp - d0); \ | |
2d1675e4 SM |
163 | } \ |
164 | else \ | |
165 | (c = ((p) == (str2) ? (end1) : (p))[-1]); \ | |
166 | } while (0) | |
167 | ||
4e8a9132 | 168 | |
fa9a63c5 RM |
169 | #else /* not emacs */ |
170 | ||
171 | /* If we are not linking with Emacs proper, | |
172 | we can't use the relocating allocator | |
173 | even if config.h says that we can. */ | |
0b32bf0e | 174 | # undef REL_ALLOC |
fa9a63c5 | 175 | |
0b32bf0e SM |
176 | # if defined STDC_HEADERS || defined _LIBC |
177 | # include <stdlib.h> | |
178 | # else | |
fa9a63c5 RM |
179 | char *malloc (); |
180 | char *realloc (); | |
0b32bf0e | 181 | # endif |
fa9a63c5 | 182 | |
9e4ecb26 | 183 | /* When used in Emacs's lib-src, we need to get bzero and bcopy somehow. |
4bb91c68 | 184 | If nothing else has been done, use the method below. */ |
0b32bf0e SM |
185 | # ifdef INHIBIT_STRING_HEADER |
186 | # if !(defined HAVE_BZERO && defined HAVE_BCOPY) | |
187 | # if !defined bzero && !defined bcopy | |
188 | # undef INHIBIT_STRING_HEADER | |
189 | # endif | |
190 | # endif | |
191 | # endif | |
9e4ecb26 | 192 | |
4bb91c68 | 193 | /* This is the normal way of making sure we have memcpy, memcmp and bzero. |
9e4ecb26 KH |
194 | This is used in most programs--a few other programs avoid this |
195 | by defining INHIBIT_STRING_HEADER. */ | |
0b32bf0e SM |
196 | # ifndef INHIBIT_STRING_HEADER |
197 | # if defined HAVE_STRING_H || defined STDC_HEADERS || defined _LIBC | |
198 | # include <string.h> | |
0b32bf0e | 199 | # ifndef bzero |
4bb91c68 SM |
200 | # ifndef _LIBC |
201 | # define bzero(s, n) (memset (s, '\0', n), (s)) | |
202 | # else | |
203 | # define bzero(s, n) __bzero (s, n) | |
204 | # endif | |
0b32bf0e SM |
205 | # endif |
206 | # else | |
207 | # include <strings.h> | |
4bb91c68 SM |
208 | # ifndef memcmp |
209 | # define memcmp(s1, s2, n) bcmp (s1, s2, n) | |
210 | # endif | |
211 | # ifndef memcpy | |
212 | # define memcpy(d, s, n) (bcopy (s, d, n), (d)) | |
213 | # endif | |
0b32bf0e SM |
214 | # endif |
215 | # endif | |
fa9a63c5 RM |
216 | |
217 | /* Define the syntax stuff for \<, \>, etc. */ | |
218 | ||
990b2375 | 219 | /* Sword must be nonzero for the wordchar pattern commands in re_match_2. */ |
669fa600 | 220 | enum syntaxcode { Swhitespace = 0, Sword = 1, Ssymbol = 2 }; |
fa9a63c5 | 221 | |
0b32bf0e SM |
222 | # ifdef SWITCH_ENUM_BUG |
223 | # define SWITCH_ENUM_CAST(x) ((int)(x)) | |
224 | # else | |
225 | # define SWITCH_ENUM_CAST(x) (x) | |
226 | # endif | |
fa9a63c5 | 227 | |
e934739e | 228 | /* Dummy macros for non-Emacs environments. */ |
0b32bf0e SM |
229 | # define BASE_LEADING_CODE_P(c) (0) |
230 | # define CHAR_CHARSET(c) 0 | |
231 | # define CHARSET_LEADING_CODE_BASE(c) 0 | |
232 | # define MAX_MULTIBYTE_LENGTH 1 | |
233 | # define RE_MULTIBYTE_P(x) 0 | |
234 | # define WORD_BOUNDARY_P(c1, c2) (0) | |
235 | # define CHAR_HEAD_P(p) (1) | |
236 | # define SINGLE_BYTE_CHAR_P(c) (1) | |
237 | # define SAME_CHARSET_P(c1, c2) (1) | |
238 | # define MULTIBYTE_FORM_LENGTH(p, s) (1) | |
70806df6 | 239 | # define PREV_CHAR_BOUNDARY(p, limit) ((p)--) |
0b32bf0e SM |
240 | # define STRING_CHAR(p, s) (*(p)) |
241 | # define RE_STRING_CHAR STRING_CHAR | |
242 | # define CHAR_STRING(c, s) (*(s) = (c), 1) | |
243 | # define STRING_CHAR_AND_LENGTH(p, s, actual_len) ((actual_len) = 1, *(p)) | |
244 | # define RE_STRING_CHAR_AND_LENGTH STRING_CHAR_AND_LENGTH | |
245 | # define GET_CHAR_BEFORE_2(c, p, str1, end1, str2, end2) \ | |
b18215fc | 246 | (c = ((p) == (str2) ? *((end1) - 1) : *((p) - 1))) |
0b32bf0e | 247 | # define MAKE_CHAR(charset, c1, c2) (c1) |
fa9a63c5 | 248 | #endif /* not emacs */ |
4e8a9132 SM |
249 | |
250 | #ifndef RE_TRANSLATE | |
0b32bf0e SM |
251 | # define RE_TRANSLATE(TBL, C) ((unsigned char)(TBL)[C]) |
252 | # define RE_TRANSLATE_P(TBL) (TBL) | |
4e8a9132 | 253 | #endif |
fa9a63c5 RM |
254 | \f |
255 | /* Get the interface, including the syntax bits. */ | |
256 | #include "regex.h" | |
257 | ||
f71b19b6 DL |
258 | /* isalpha etc. are used for the character classes. */ |
259 | #include <ctype.h> | |
fa9a63c5 | 260 | |
f71b19b6 | 261 | #ifdef emacs |
fa9a63c5 | 262 | |
f71b19b6 | 263 | /* 1 if C is an ASCII character. */ |
0b32bf0e | 264 | # define IS_REAL_ASCII(c) ((c) < 0200) |
fa9a63c5 | 265 | |
f71b19b6 | 266 | /* 1 if C is a unibyte character. */ |
0b32bf0e | 267 | # define ISUNIBYTE(c) (SINGLE_BYTE_CHAR_P ((c))) |
96cc36cc | 268 | |
f71b19b6 | 269 | /* The Emacs definitions should not be directly affected by locales. */ |
96cc36cc | 270 | |
f71b19b6 | 271 | /* In Emacs, these are only used for single-byte characters. */ |
0b32bf0e SM |
272 | # define ISDIGIT(c) ((c) >= '0' && (c) <= '9') |
273 | # define ISCNTRL(c) ((c) < ' ') | |
274 | # define ISXDIGIT(c) (((c) >= '0' && (c) <= '9') \ | |
f71b19b6 DL |
275 | || ((c) >= 'a' && (c) <= 'f') \ |
276 | || ((c) >= 'A' && (c) <= 'F')) | |
96cc36cc RS |
277 | |
278 | /* This is only used for single-byte characters. */ | |
0b32bf0e | 279 | # define ISBLANK(c) ((c) == ' ' || (c) == '\t') |
96cc36cc RS |
280 | |
281 | /* The rest must handle multibyte characters. */ | |
282 | ||
0b32bf0e | 283 | # define ISGRAPH(c) (SINGLE_BYTE_CHAR_P (c) \ |
f71b19b6 | 284 | ? (c) > ' ' && !((c) >= 0177 && (c) <= 0237) \ |
96cc36cc RS |
285 | : 1) |
286 | ||
14473664 | 287 | # define ISPRINT(c) (SINGLE_BYTE_CHAR_P (c) \ |
f71b19b6 | 288 | ? (c) >= ' ' && !((c) >= 0177 && (c) <= 0237) \ |
96cc36cc RS |
289 | : 1) |
290 | ||
0b32bf0e | 291 | # define ISALNUM(c) (IS_REAL_ASCII (c) \ |
f71b19b6 DL |
292 | ? (((c) >= 'a' && (c) <= 'z') \ |
293 | || ((c) >= 'A' && (c) <= 'Z') \ | |
294 | || ((c) >= '0' && (c) <= '9')) \ | |
96cc36cc RS |
295 | : SYNTAX (c) == Sword) |
296 | ||
0b32bf0e | 297 | # define ISALPHA(c) (IS_REAL_ASCII (c) \ |
f71b19b6 DL |
298 | ? (((c) >= 'a' && (c) <= 'z') \ |
299 | || ((c) >= 'A' && (c) <= 'Z')) \ | |
96cc36cc RS |
300 | : SYNTAX (c) == Sword) |
301 | ||
0b32bf0e | 302 | # define ISLOWER(c) (LOWERCASEP (c)) |
96cc36cc | 303 | |
0b32bf0e | 304 | # define ISPUNCT(c) (IS_REAL_ASCII (c) \ |
f71b19b6 DL |
305 | ? ((c) > ' ' && (c) < 0177 \ |
306 | && !(((c) >= 'a' && (c) <= 'z') \ | |
4bb91c68 SM |
307 | || ((c) >= 'A' && (c) <= 'Z') \ |
308 | || ((c) >= '0' && (c) <= '9'))) \ | |
96cc36cc RS |
309 | : SYNTAX (c) != Sword) |
310 | ||
0b32bf0e | 311 | # define ISSPACE(c) (SYNTAX (c) == Swhitespace) |
96cc36cc | 312 | |
0b32bf0e | 313 | # define ISUPPER(c) (UPPERCASEP (c)) |
96cc36cc | 314 | |
0b32bf0e | 315 | # define ISWORD(c) (SYNTAX (c) == Sword) |
96cc36cc RS |
316 | |
317 | #else /* not emacs */ | |
318 | ||
f71b19b6 DL |
319 | /* Jim Meyering writes: |
320 | ||
321 | "... Some ctype macros are valid only for character codes that | |
322 | isascii says are ASCII (SGI's IRIX-4.0.5 is one such system --when | |
323 | using /bin/cc or gcc but without giving an ansi option). So, all | |
4bb91c68 | 324 | ctype uses should be through macros like ISPRINT... If |
f71b19b6 DL |
325 | STDC_HEADERS is defined, then autoconf has verified that the ctype |
326 | macros don't need to be guarded with references to isascii. ... | |
327 | Defining isascii to 1 should let any compiler worth its salt | |
4bb91c68 SM |
328 | eliminate the && through constant folding." |
329 | Solaris defines some of these symbols so we must undefine them first. */ | |
f71b19b6 | 330 | |
4bb91c68 | 331 | # undef ISASCII |
0b32bf0e SM |
332 | # if defined STDC_HEADERS || (!defined isascii && !defined HAVE_ISASCII) |
333 | # define ISASCII(c) 1 | |
334 | # else | |
335 | # define ISASCII(c) isascii(c) | |
336 | # endif | |
f71b19b6 DL |
337 | |
338 | /* 1 if C is an ASCII character. */ | |
0b32bf0e | 339 | # define IS_REAL_ASCII(c) ((c) < 0200) |
f71b19b6 DL |
340 | |
341 | /* This distinction is not meaningful, except in Emacs. */ | |
0b32bf0e SM |
342 | # define ISUNIBYTE(c) 1 |
343 | ||
344 | # ifdef isblank | |
345 | # define ISBLANK(c) (ISASCII (c) && isblank (c)) | |
346 | # else | |
347 | # define ISBLANK(c) ((c) == ' ' || (c) == '\t') | |
348 | # endif | |
349 | # ifdef isgraph | |
350 | # define ISGRAPH(c) (ISASCII (c) && isgraph (c)) | |
351 | # else | |
352 | # define ISGRAPH(c) (ISASCII (c) && isprint (c) && !isspace (c)) | |
353 | # endif | |
354 | ||
4bb91c68 | 355 | # undef ISPRINT |
0b32bf0e SM |
356 | # define ISPRINT(c) (ISASCII (c) && isprint (c)) |
357 | # define ISDIGIT(c) (ISASCII (c) && isdigit (c)) | |
358 | # define ISALNUM(c) (ISASCII (c) && isalnum (c)) | |
359 | # define ISALPHA(c) (ISASCII (c) && isalpha (c)) | |
360 | # define ISCNTRL(c) (ISASCII (c) && iscntrl (c)) | |
361 | # define ISLOWER(c) (ISASCII (c) && islower (c)) | |
362 | # define ISPUNCT(c) (ISASCII (c) && ispunct (c)) | |
363 | # define ISSPACE(c) (ISASCII (c) && isspace (c)) | |
364 | # define ISUPPER(c) (ISASCII (c) && isupper (c)) | |
365 | # define ISXDIGIT(c) (ISASCII (c) && isxdigit (c)) | |
366 | ||
367 | # define ISWORD(c) ISALPHA(c) | |
368 | ||
4bb91c68 SM |
369 | # ifdef _tolower |
370 | # define TOLOWER(c) _tolower(c) | |
371 | # else | |
372 | # define TOLOWER(c) tolower(c) | |
373 | # endif | |
374 | ||
375 | /* How many characters in the character set. */ | |
376 | # define CHAR_SET_SIZE 256 | |
377 | ||
0b32bf0e | 378 | # ifdef SYNTAX_TABLE |
f71b19b6 | 379 | |
0b32bf0e | 380 | extern char *re_syntax_table; |
f71b19b6 | 381 | |
0b32bf0e SM |
382 | # else /* not SYNTAX_TABLE */ |
383 | ||
0b32bf0e SM |
384 | static char re_syntax_table[CHAR_SET_SIZE]; |
385 | ||
386 | static void | |
387 | init_syntax_once () | |
388 | { | |
389 | register int c; | |
390 | static int done = 0; | |
391 | ||
392 | if (done) | |
393 | return; | |
394 | ||
395 | bzero (re_syntax_table, sizeof re_syntax_table); | |
396 | ||
4bb91c68 SM |
397 | for (c = 0; c < CHAR_SET_SIZE; ++c) |
398 | if (ISALNUM (c)) | |
399 | re_syntax_table[c] = Sword; | |
fa9a63c5 | 400 | |
669fa600 | 401 | re_syntax_table['_'] = Ssymbol; |
fa9a63c5 | 402 | |
0b32bf0e SM |
403 | done = 1; |
404 | } | |
405 | ||
406 | # endif /* not SYNTAX_TABLE */ | |
96cc36cc | 407 | |
4bb91c68 SM |
408 | # define SYNTAX(c) re_syntax_table[(c)] |
409 | ||
96cc36cc RS |
410 | #endif /* not emacs */ |
411 | \f | |
fa9a63c5 | 412 | #ifndef NULL |
0b32bf0e | 413 | # define NULL (void *)0 |
fa9a63c5 RM |
414 | #endif |
415 | ||
416 | /* We remove any previous definition of `SIGN_EXTEND_CHAR', | |
417 | since ours (we hope) works properly with all combinations of | |
418 | machines, compilers, `char' and `unsigned char' argument types. | |
4bb91c68 | 419 | (Per Bothner suggested the basic approach.) */ |
fa9a63c5 RM |
420 | #undef SIGN_EXTEND_CHAR |
421 | #if __STDC__ | |
0b32bf0e | 422 | # define SIGN_EXTEND_CHAR(c) ((signed char) (c)) |
fa9a63c5 RM |
423 | #else /* not __STDC__ */ |
424 | /* As in Harbison and Steele. */ | |
0b32bf0e | 425 | # define SIGN_EXTEND_CHAR(c) ((((unsigned char) (c)) ^ 128) - 128) |
fa9a63c5 RM |
426 | #endif |
427 | \f | |
428 | /* Should we use malloc or alloca? If REGEX_MALLOC is not defined, we | |
429 | use `alloca' instead of `malloc'. This is because using malloc in | |
430 | re_search* or re_match* could cause memory leaks when C-g is used in | |
431 | Emacs; also, malloc is slower and causes storage fragmentation. On | |
5e69f11e RM |
432 | the other hand, malloc is more portable, and easier to debug. |
433 | ||
fa9a63c5 RM |
434 | Because we sometimes use alloca, some routines have to be macros, |
435 | not functions -- `alloca'-allocated space disappears at the end of the | |
436 | function it is called in. */ | |
437 | ||
438 | #ifdef REGEX_MALLOC | |
439 | ||
0b32bf0e SM |
440 | # define REGEX_ALLOCATE malloc |
441 | # define REGEX_REALLOCATE(source, osize, nsize) realloc (source, nsize) | |
442 | # define REGEX_FREE free | |
fa9a63c5 RM |
443 | |
444 | #else /* not REGEX_MALLOC */ | |
445 | ||
446 | /* Emacs already defines alloca, sometimes. */ | |
0b32bf0e | 447 | # ifndef alloca |
fa9a63c5 RM |
448 | |
449 | /* Make alloca work the best possible way. */ | |
0b32bf0e SM |
450 | # ifdef __GNUC__ |
451 | # define alloca __builtin_alloca | |
452 | # else /* not __GNUC__ */ | |
453 | # if HAVE_ALLOCA_H | |
454 | # include <alloca.h> | |
455 | # endif /* HAVE_ALLOCA_H */ | |
456 | # endif /* not __GNUC__ */ | |
fa9a63c5 | 457 | |
0b32bf0e | 458 | # endif /* not alloca */ |
fa9a63c5 | 459 | |
0b32bf0e | 460 | # define REGEX_ALLOCATE alloca |
fa9a63c5 RM |
461 | |
462 | /* Assumes a `char *destination' variable. */ | |
0b32bf0e | 463 | # define REGEX_REALLOCATE(source, osize, nsize) \ |
fa9a63c5 | 464 | (destination = (char *) alloca (nsize), \ |
4bb91c68 | 465 | memcpy (destination, source, osize)) |
fa9a63c5 RM |
466 | |
467 | /* No need to do anything to free, after alloca. */ | |
0b32bf0e | 468 | # define REGEX_FREE(arg) ((void)0) /* Do nothing! But inhibit gcc warning. */ |
fa9a63c5 RM |
469 | |
470 | #endif /* not REGEX_MALLOC */ | |
471 | ||
472 | /* Define how to allocate the failure stack. */ | |
473 | ||
0b32bf0e | 474 | #if defined REL_ALLOC && defined REGEX_MALLOC |
4297555e | 475 | |
0b32bf0e | 476 | # define REGEX_ALLOCATE_STACK(size) \ |
fa9a63c5 | 477 | r_alloc (&failure_stack_ptr, (size)) |
0b32bf0e | 478 | # define REGEX_REALLOCATE_STACK(source, osize, nsize) \ |
fa9a63c5 | 479 | r_re_alloc (&failure_stack_ptr, (nsize)) |
0b32bf0e | 480 | # define REGEX_FREE_STACK(ptr) \ |
fa9a63c5 RM |
481 | r_alloc_free (&failure_stack_ptr) |
482 | ||
4297555e | 483 | #else /* not using relocating allocator */ |
fa9a63c5 | 484 | |
0b32bf0e | 485 | # ifdef REGEX_MALLOC |
fa9a63c5 | 486 | |
0b32bf0e SM |
487 | # define REGEX_ALLOCATE_STACK malloc |
488 | # define REGEX_REALLOCATE_STACK(source, osize, nsize) realloc (source, nsize) | |
489 | # define REGEX_FREE_STACK free | |
fa9a63c5 | 490 | |
0b32bf0e | 491 | # else /* not REGEX_MALLOC */ |
fa9a63c5 | 492 | |
0b32bf0e | 493 | # define REGEX_ALLOCATE_STACK alloca |
fa9a63c5 | 494 | |
0b32bf0e | 495 | # define REGEX_REALLOCATE_STACK(source, osize, nsize) \ |
fa9a63c5 | 496 | REGEX_REALLOCATE (source, osize, nsize) |
25fe55af | 497 | /* No need to explicitly free anything. */ |
0b32bf0e | 498 | # define REGEX_FREE_STACK(arg) ((void)0) |
fa9a63c5 | 499 | |
0b32bf0e | 500 | # endif /* not REGEX_MALLOC */ |
4297555e | 501 | #endif /* not using relocating allocator */ |
fa9a63c5 RM |
502 | |
503 | ||
504 | /* True if `size1' is non-NULL and PTR is pointing anywhere inside | |
505 | `string1' or just past its end. This works if PTR is NULL, which is | |
506 | a good thing. */ | |
25fe55af | 507 | #define FIRST_STRING_P(ptr) \ |
fa9a63c5 RM |
508 | (size1 && string1 <= (ptr) && (ptr) <= string1 + size1) |
509 | ||
510 | /* (Re)Allocate N items of type T using malloc, or fail. */ | |
511 | #define TALLOC(n, t) ((t *) malloc ((n) * sizeof (t))) | |
512 | #define RETALLOC(addr, n, t) ((addr) = (t *) realloc (addr, (n) * sizeof (t))) | |
513 | #define RETALLOC_IF(addr, n, t) \ | |
514 | if (addr) RETALLOC((addr), (n), t); else (addr) = TALLOC ((n), t) | |
515 | #define REGEX_TALLOC(n, t) ((t *) REGEX_ALLOCATE ((n) * sizeof (t))) | |
516 | ||
4bb91c68 | 517 | #define BYTEWIDTH 8 /* In bits. */ |
fa9a63c5 RM |
518 | |
519 | #define STREQ(s1, s2) ((strcmp (s1, s2) == 0)) | |
520 | ||
521 | #undef MAX | |
522 | #undef MIN | |
523 | #define MAX(a, b) ((a) > (b) ? (a) : (b)) | |
524 | #define MIN(a, b) ((a) < (b) ? (a) : (b)) | |
525 | ||
66f0296e SM |
526 | /* Type of source-pattern and string chars. */ |
527 | typedef const unsigned char re_char; | |
528 | ||
fa9a63c5 RM |
529 | typedef char boolean; |
530 | #define false 0 | |
531 | #define true 1 | |
532 | ||
4bb91c68 SM |
533 | static int re_match_2_internal _RE_ARGS ((struct re_pattern_buffer *bufp, |
534 | re_char *string1, int size1, | |
535 | re_char *string2, int size2, | |
536 | int pos, | |
537 | struct re_registers *regs, | |
538 | int stop)); | |
fa9a63c5 RM |
539 | \f |
540 | /* These are the command codes that appear in compiled regular | |
4bb91c68 | 541 | expressions. Some opcodes are followed by argument bytes. A |
fa9a63c5 RM |
542 | command code can specify any interpretation whatsoever for its |
543 | arguments. Zero bytes may appear in the compiled regular expression. */ | |
544 | ||
545 | typedef enum | |
546 | { | |
547 | no_op = 0, | |
548 | ||
4bb91c68 | 549 | /* Succeed right away--no more backtracking. */ |
fa9a63c5 RM |
550 | succeed, |
551 | ||
25fe55af | 552 | /* Followed by one byte giving n, then by n literal bytes. */ |
fa9a63c5 RM |
553 | exactn, |
554 | ||
25fe55af | 555 | /* Matches any (more or less) character. */ |
fa9a63c5 RM |
556 | anychar, |
557 | ||
25fe55af RS |
558 | /* Matches any one char belonging to specified set. First |
559 | following byte is number of bitmap bytes. Then come bytes | |
560 | for a bitmap saying which chars are in. Bits in each byte | |
561 | are ordered low-bit-first. A character is in the set if its | |
562 | bit is 1. A character too large to have a bit in the map is | |
96cc36cc RS |
563 | automatically not in the set. |
564 | ||
565 | If the length byte has the 0x80 bit set, then that stuff | |
566 | is followed by a range table: | |
567 | 2 bytes of flags for character sets (low 8 bits, high 8 bits) | |
0b32bf0e | 568 | See RANGE_TABLE_WORK_BITS below. |
01618498 | 569 | 2 bytes, the number of pairs that follow (upto 32767) |
96cc36cc | 570 | pairs, each 2 multibyte characters, |
0b32bf0e | 571 | each multibyte character represented as 3 bytes. */ |
fa9a63c5 RM |
572 | charset, |
573 | ||
25fe55af | 574 | /* Same parameters as charset, but match any character that is |
4bb91c68 | 575 | not one of those specified. */ |
fa9a63c5 RM |
576 | charset_not, |
577 | ||
25fe55af RS |
578 | /* Start remembering the text that is matched, for storing in a |
579 | register. Followed by one byte with the register number, in | |
580 | the range 0 to one less than the pattern buffer's re_nsub | |
505bde11 | 581 | field. */ |
fa9a63c5 RM |
582 | start_memory, |
583 | ||
25fe55af RS |
584 | /* Stop remembering the text that is matched and store it in a |
585 | memory register. Followed by one byte with the register | |
586 | number, in the range 0 to one less than `re_nsub' in the | |
505bde11 | 587 | pattern buffer. */ |
fa9a63c5 RM |
588 | stop_memory, |
589 | ||
25fe55af | 590 | /* Match a duplicate of something remembered. Followed by one |
4bb91c68 | 591 | byte containing the register number. */ |
fa9a63c5 RM |
592 | duplicate, |
593 | ||
25fe55af | 594 | /* Fail unless at beginning of line. */ |
fa9a63c5 RM |
595 | begline, |
596 | ||
4bb91c68 | 597 | /* Fail unless at end of line. */ |
fa9a63c5 RM |
598 | endline, |
599 | ||
25fe55af RS |
600 | /* Succeeds if at beginning of buffer (if emacs) or at beginning |
601 | of string to be matched (if not). */ | |
fa9a63c5 RM |
602 | begbuf, |
603 | ||
25fe55af | 604 | /* Analogously, for end of buffer/string. */ |
fa9a63c5 | 605 | endbuf, |
5e69f11e | 606 | |
25fe55af | 607 | /* Followed by two byte relative address to which to jump. */ |
5e69f11e | 608 | jump, |
fa9a63c5 | 609 | |
25fe55af RS |
610 | /* Followed by two-byte relative address of place to resume at |
611 | in case of failure. */ | |
fa9a63c5 | 612 | on_failure_jump, |
5e69f11e | 613 | |
25fe55af RS |
614 | /* Like on_failure_jump, but pushes a placeholder instead of the |
615 | current string position when executed. */ | |
fa9a63c5 | 616 | on_failure_keep_string_jump, |
5e69f11e | 617 | |
505bde11 SM |
618 | /* Just like `on_failure_jump', except that it checks that we |
619 | don't get stuck in an infinite loop (matching an empty string | |
620 | indefinitely). */ | |
621 | on_failure_jump_loop, | |
622 | ||
0683b6fa SM |
623 | /* Just like `on_failure_jump_loop', except that it checks for |
624 | a different kind of loop (the kind that shows up with non-greedy | |
625 | operators). This operation has to be immediately preceded | |
626 | by a `no_op'. */ | |
627 | on_failure_jump_nastyloop, | |
628 | ||
0b32bf0e | 629 | /* A smart `on_failure_jump' used for greedy * and + operators. |
505bde11 SM |
630 | It analyses the loop before which it is put and if the |
631 | loop does not require backtracking, it changes itself to | |
4e8a9132 SM |
632 | `on_failure_keep_string_jump' and short-circuits the loop, |
633 | else it just defaults to changing itself into `on_failure_jump'. | |
634 | It assumes that it is pointing to just past a `jump'. */ | |
505bde11 | 635 | on_failure_jump_smart, |
fa9a63c5 | 636 | |
25fe55af | 637 | /* Followed by two-byte relative address and two-byte number n. |
ed0767d8 SM |
638 | After matching N times, jump to the address upon failure. |
639 | Does not work if N starts at 0: use on_failure_jump_loop | |
640 | instead. */ | |
fa9a63c5 RM |
641 | succeed_n, |
642 | ||
25fe55af RS |
643 | /* Followed by two-byte relative address, and two-byte number n. |
644 | Jump to the address N times, then fail. */ | |
fa9a63c5 RM |
645 | jump_n, |
646 | ||
25fe55af RS |
647 | /* Set the following two-byte relative address to the |
648 | subsequent two-byte number. The address *includes* the two | |
649 | bytes of number. */ | |
fa9a63c5 RM |
650 | set_number_at, |
651 | ||
fa9a63c5 RM |
652 | wordbeg, /* Succeeds if at word beginning. */ |
653 | wordend, /* Succeeds if at word end. */ | |
654 | ||
655 | wordbound, /* Succeeds if at a word boundary. */ | |
1fb352e0 | 656 | notwordbound, /* Succeeds if not at a word boundary. */ |
fa9a63c5 | 657 | |
669fa600 SM |
658 | symbeg, /* Succeeds if at symbol beginning. */ |
659 | symend, /* Succeeds if at symbol end. */ | |
660 | ||
fa9a63c5 | 661 | /* Matches any character whose syntax is specified. Followed by |
25fe55af | 662 | a byte which contains a syntax code, e.g., Sword. */ |
fa9a63c5 RM |
663 | syntaxspec, |
664 | ||
665 | /* Matches any character whose syntax is not that specified. */ | |
1fb352e0 SM |
666 | notsyntaxspec |
667 | ||
668 | #ifdef emacs | |
669 | ,before_dot, /* Succeeds if before point. */ | |
670 | at_dot, /* Succeeds if at point. */ | |
671 | after_dot, /* Succeeds if after point. */ | |
b18215fc RS |
672 | |
673 | /* Matches any character whose category-set contains the specified | |
25fe55af RS |
674 | category. The operator is followed by a byte which contains a |
675 | category code (mnemonic ASCII character). */ | |
b18215fc RS |
676 | categoryspec, |
677 | ||
678 | /* Matches any character whose category-set does not contain the | |
679 | specified category. The operator is followed by a byte which | |
680 | contains the category code (mnemonic ASCII character). */ | |
681 | notcategoryspec | |
fa9a63c5 RM |
682 | #endif /* emacs */ |
683 | } re_opcode_t; | |
684 | \f | |
685 | /* Common operations on the compiled pattern. */ | |
686 | ||
687 | /* Store NUMBER in two contiguous bytes starting at DESTINATION. */ | |
688 | ||
689 | #define STORE_NUMBER(destination, number) \ | |
690 | do { \ | |
691 | (destination)[0] = (number) & 0377; \ | |
692 | (destination)[1] = (number) >> 8; \ | |
693 | } while (0) | |
694 | ||
695 | /* Same as STORE_NUMBER, except increment DESTINATION to | |
696 | the byte after where the number is stored. Therefore, DESTINATION | |
697 | must be an lvalue. */ | |
698 | ||
699 | #define STORE_NUMBER_AND_INCR(destination, number) \ | |
700 | do { \ | |
701 | STORE_NUMBER (destination, number); \ | |
702 | (destination) += 2; \ | |
703 | } while (0) | |
704 | ||
705 | /* Put into DESTINATION a number stored in two contiguous bytes starting | |
706 | at SOURCE. */ | |
707 | ||
708 | #define EXTRACT_NUMBER(destination, source) \ | |
709 | do { \ | |
710 | (destination) = *(source) & 0377; \ | |
711 | (destination) += SIGN_EXTEND_CHAR (*((source) + 1)) << 8; \ | |
712 | } while (0) | |
713 | ||
714 | #ifdef DEBUG | |
4bb91c68 | 715 | static void extract_number _RE_ARGS ((int *dest, re_char *source)); |
fa9a63c5 RM |
716 | static void |
717 | extract_number (dest, source) | |
718 | int *dest; | |
01618498 | 719 | re_char *source; |
fa9a63c5 | 720 | { |
5e69f11e | 721 | int temp = SIGN_EXTEND_CHAR (*(source + 1)); |
fa9a63c5 RM |
722 | *dest = *source & 0377; |
723 | *dest += temp << 8; | |
724 | } | |
725 | ||
4bb91c68 | 726 | # ifndef EXTRACT_MACROS /* To debug the macros. */ |
0b32bf0e SM |
727 | # undef EXTRACT_NUMBER |
728 | # define EXTRACT_NUMBER(dest, src) extract_number (&dest, src) | |
729 | # endif /* not EXTRACT_MACROS */ | |
fa9a63c5 RM |
730 | |
731 | #endif /* DEBUG */ | |
732 | ||
733 | /* Same as EXTRACT_NUMBER, except increment SOURCE to after the number. | |
734 | SOURCE must be an lvalue. */ | |
735 | ||
736 | #define EXTRACT_NUMBER_AND_INCR(destination, source) \ | |
737 | do { \ | |
738 | EXTRACT_NUMBER (destination, source); \ | |
25fe55af | 739 | (source) += 2; \ |
fa9a63c5 RM |
740 | } while (0) |
741 | ||
742 | #ifdef DEBUG | |
4bb91c68 SM |
743 | static void extract_number_and_incr _RE_ARGS ((int *destination, |
744 | re_char **source)); | |
fa9a63c5 RM |
745 | static void |
746 | extract_number_and_incr (destination, source) | |
747 | int *destination; | |
01618498 | 748 | re_char **source; |
5e69f11e | 749 | { |
fa9a63c5 RM |
750 | extract_number (destination, *source); |
751 | *source += 2; | |
752 | } | |
753 | ||
0b32bf0e SM |
754 | # ifndef EXTRACT_MACROS |
755 | # undef EXTRACT_NUMBER_AND_INCR | |
756 | # define EXTRACT_NUMBER_AND_INCR(dest, src) \ | |
fa9a63c5 | 757 | extract_number_and_incr (&dest, &src) |
0b32bf0e | 758 | # endif /* not EXTRACT_MACROS */ |
fa9a63c5 RM |
759 | |
760 | #endif /* DEBUG */ | |
761 | \f | |
b18215fc RS |
762 | /* Store a multibyte character in three contiguous bytes starting |
763 | DESTINATION, and increment DESTINATION to the byte after where the | |
25fe55af | 764 | character is stored. Therefore, DESTINATION must be an lvalue. */ |
b18215fc RS |
765 | |
766 | #define STORE_CHARACTER_AND_INCR(destination, character) \ | |
767 | do { \ | |
768 | (destination)[0] = (character) & 0377; \ | |
769 | (destination)[1] = ((character) >> 8) & 0377; \ | |
770 | (destination)[2] = (character) >> 16; \ | |
771 | (destination) += 3; \ | |
772 | } while (0) | |
773 | ||
774 | /* Put into DESTINATION a character stored in three contiguous bytes | |
25fe55af | 775 | starting at SOURCE. */ |
b18215fc RS |
776 | |
777 | #define EXTRACT_CHARACTER(destination, source) \ | |
778 | do { \ | |
779 | (destination) = ((source)[0] \ | |
780 | | ((source)[1] << 8) \ | |
781 | | ((source)[2] << 16)); \ | |
782 | } while (0) | |
783 | ||
784 | ||
785 | /* Macros for charset. */ | |
786 | ||
787 | /* Size of bitmap of charset P in bytes. P is a start of charset, | |
788 | i.e. *P is (re_opcode_t) charset or (re_opcode_t) charset_not. */ | |
789 | #define CHARSET_BITMAP_SIZE(p) ((p)[1] & 0x7F) | |
790 | ||
791 | /* Nonzero if charset P has range table. */ | |
25fe55af | 792 | #define CHARSET_RANGE_TABLE_EXISTS_P(p) ((p)[1] & 0x80) |
b18215fc RS |
793 | |
794 | /* Return the address of range table of charset P. But not the start | |
795 | of table itself, but the before where the number of ranges is | |
96cc36cc RS |
796 | stored. `2 +' means to skip re_opcode_t and size of bitmap, |
797 | and the 2 bytes of flags at the start of the range table. */ | |
798 | #define CHARSET_RANGE_TABLE(p) (&(p)[4 + CHARSET_BITMAP_SIZE (p)]) | |
799 | ||
800 | /* Extract the bit flags that start a range table. */ | |
801 | #define CHARSET_RANGE_TABLE_BITS(p) \ | |
802 | ((p)[2 + CHARSET_BITMAP_SIZE (p)] \ | |
803 | + (p)[3 + CHARSET_BITMAP_SIZE (p)] * 0x100) | |
b18215fc RS |
804 | |
805 | /* Test if C is listed in the bitmap of charset P. */ | |
806 | #define CHARSET_LOOKUP_BITMAP(p, c) \ | |
807 | ((c) < CHARSET_BITMAP_SIZE (p) * BYTEWIDTH \ | |
808 | && (p)[2 + (c) / BYTEWIDTH] & (1 << ((c) % BYTEWIDTH))) | |
809 | ||
810 | /* Return the address of end of RANGE_TABLE. COUNT is number of | |
25fe55af RS |
811 | ranges (which is a pair of (start, end)) in the RANGE_TABLE. `* 2' |
812 | is start of range and end of range. `* 3' is size of each start | |
b18215fc RS |
813 | and end. */ |
814 | #define CHARSET_RANGE_TABLE_END(range_table, count) \ | |
815 | ((range_table) + (count) * 2 * 3) | |
816 | ||
25fe55af | 817 | /* Test if C is in RANGE_TABLE. A flag NOT is negated if C is in. |
b18215fc RS |
818 | COUNT is number of ranges in RANGE_TABLE. */ |
819 | #define CHARSET_LOOKUP_RANGE_TABLE_RAW(not, c, range_table, count) \ | |
820 | do \ | |
821 | { \ | |
01618498 SM |
822 | re_wchar_t range_start, range_end; \ |
823 | re_char *p; \ | |
824 | re_char *range_table_end \ | |
b18215fc RS |
825 | = CHARSET_RANGE_TABLE_END ((range_table), (count)); \ |
826 | \ | |
827 | for (p = (range_table); p < range_table_end; p += 2 * 3) \ | |
828 | { \ | |
829 | EXTRACT_CHARACTER (range_start, p); \ | |
830 | EXTRACT_CHARACTER (range_end, p + 3); \ | |
831 | \ | |
832 | if (range_start <= (c) && (c) <= range_end) \ | |
833 | { \ | |
834 | (not) = !(not); \ | |
835 | break; \ | |
836 | } \ | |
837 | } \ | |
838 | } \ | |
839 | while (0) | |
840 | ||
841 | /* Test if C is in range table of CHARSET. The flag NOT is negated if | |
842 | C is listed in it. */ | |
843 | #define CHARSET_LOOKUP_RANGE_TABLE(not, c, charset) \ | |
844 | do \ | |
845 | { \ | |
846 | /* Number of ranges in range table. */ \ | |
847 | int count; \ | |
01618498 SM |
848 | re_char *range_table = CHARSET_RANGE_TABLE (charset); \ |
849 | \ | |
b18215fc RS |
850 | EXTRACT_NUMBER_AND_INCR (count, range_table); \ |
851 | CHARSET_LOOKUP_RANGE_TABLE_RAW ((not), (c), range_table, count); \ | |
852 | } \ | |
853 | while (0) | |
854 | \f | |
fa9a63c5 RM |
855 | /* If DEBUG is defined, Regex prints many voluminous messages about what |
856 | it is doing (if the variable `debug' is nonzero). If linked with the | |
857 | main program in `iregex.c', you can enter patterns and strings | |
858 | interactively. And if linked with the main program in `main.c' and | |
4bb91c68 | 859 | the other test files, you can run the already-written tests. */ |
fa9a63c5 RM |
860 | |
861 | #ifdef DEBUG | |
862 | ||
863 | /* We use standard I/O for debugging. */ | |
0b32bf0e | 864 | # include <stdio.h> |
fa9a63c5 RM |
865 | |
866 | /* It is useful to test things that ``must'' be true when debugging. */ | |
0b32bf0e | 867 | # include <assert.h> |
fa9a63c5 | 868 | |
99633e97 | 869 | static int debug = -100000; |
fa9a63c5 | 870 | |
0b32bf0e SM |
871 | # define DEBUG_STATEMENT(e) e |
872 | # define DEBUG_PRINT1(x) if (debug > 0) printf (x) | |
873 | # define DEBUG_PRINT2(x1, x2) if (debug > 0) printf (x1, x2) | |
874 | # define DEBUG_PRINT3(x1, x2, x3) if (debug > 0) printf (x1, x2, x3) | |
875 | # define DEBUG_PRINT4(x1, x2, x3, x4) if (debug > 0) printf (x1, x2, x3, x4) | |
876 | # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) \ | |
99633e97 | 877 | if (debug > 0) print_partial_compiled_pattern (s, e) |
0b32bf0e | 878 | # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) \ |
99633e97 | 879 | if (debug > 0) print_double_string (w, s1, sz1, s2, sz2) |
fa9a63c5 RM |
880 | |
881 | ||
882 | /* Print the fastmap in human-readable form. */ | |
883 | ||
884 | void | |
885 | print_fastmap (fastmap) | |
886 | char *fastmap; | |
887 | { | |
888 | unsigned was_a_range = 0; | |
5e69f11e RM |
889 | unsigned i = 0; |
890 | ||
fa9a63c5 RM |
891 | while (i < (1 << BYTEWIDTH)) |
892 | { | |
893 | if (fastmap[i++]) | |
894 | { | |
895 | was_a_range = 0; | |
25fe55af RS |
896 | putchar (i - 1); |
897 | while (i < (1 << BYTEWIDTH) && fastmap[i]) | |
898 | { | |
899 | was_a_range = 1; | |
900 | i++; | |
901 | } | |
fa9a63c5 | 902 | if (was_a_range) |
25fe55af RS |
903 | { |
904 | printf ("-"); | |
905 | putchar (i - 1); | |
906 | } | |
907 | } | |
fa9a63c5 | 908 | } |
5e69f11e | 909 | putchar ('\n'); |
fa9a63c5 RM |
910 | } |
911 | ||
912 | ||
913 | /* Print a compiled pattern string in human-readable form, starting at | |
914 | the START pointer into it and ending just before the pointer END. */ | |
915 | ||
916 | void | |
917 | print_partial_compiled_pattern (start, end) | |
01618498 SM |
918 | re_char *start; |
919 | re_char *end; | |
fa9a63c5 RM |
920 | { |
921 | int mcnt, mcnt2; | |
01618498 SM |
922 | re_char *p = start; |
923 | re_char *pend = end; | |
fa9a63c5 RM |
924 | |
925 | if (start == NULL) | |
926 | { | |
a1a052df | 927 | fprintf (stderr, "(null)\n"); |
fa9a63c5 RM |
928 | return; |
929 | } | |
5e69f11e | 930 | |
fa9a63c5 RM |
931 | /* Loop over pattern commands. */ |
932 | while (p < pend) | |
933 | { | |
a1a052df | 934 | fprintf (stderr, "%d:\t", p - start); |
fa9a63c5 RM |
935 | |
936 | switch ((re_opcode_t) *p++) | |
937 | { | |
25fe55af | 938 | case no_op: |
a1a052df | 939 | fprintf (stderr, "/no_op"); |
25fe55af | 940 | break; |
fa9a63c5 | 941 | |
99633e97 | 942 | case succeed: |
a1a052df | 943 | fprintf (stderr, "/succeed"); |
99633e97 SM |
944 | break; |
945 | ||
fa9a63c5 RM |
946 | case exactn: |
947 | mcnt = *p++; | |
a1a052df | 948 | fprintf (stderr, "/exactn/%d", mcnt); |
25fe55af | 949 | do |
fa9a63c5 | 950 | { |
a1a052df | 951 | fprintf (stderr, "/%c", *p++); |
25fe55af RS |
952 | } |
953 | while (--mcnt); | |
954 | break; | |
fa9a63c5 RM |
955 | |
956 | case start_memory: | |
a1a052df | 957 | fprintf (stderr, "/start_memory/%d", *p++); |
25fe55af | 958 | break; |
fa9a63c5 RM |
959 | |
960 | case stop_memory: | |
a1a052df | 961 | fprintf (stderr, "/stop_memory/%d", *p++); |
25fe55af | 962 | break; |
fa9a63c5 RM |
963 | |
964 | case duplicate: | |
a1a052df | 965 | fprintf (stderr, "/duplicate/%d", *p++); |
fa9a63c5 RM |
966 | break; |
967 | ||
968 | case anychar: | |
a1a052df | 969 | fprintf (stderr, "/anychar"); |
fa9a63c5 RM |
970 | break; |
971 | ||
972 | case charset: | |
25fe55af RS |
973 | case charset_not: |
974 | { | |
975 | register int c, last = -100; | |
fa9a63c5 | 976 | register int in_range = 0; |
99633e97 SM |
977 | int length = CHARSET_BITMAP_SIZE (p - 1); |
978 | int has_range_table = CHARSET_RANGE_TABLE_EXISTS_P (p - 1); | |
fa9a63c5 | 979 | |
a1a052df | 980 | fprintf (stderr, "/charset [%s", |
275a3409 | 981 | (re_opcode_t) *(p - 1) == charset_not ? "^" : ""); |
5e69f11e | 982 | |
275a3409 RS |
983 | if (p + *p >= pend) |
984 | fprintf (stderr, " !extends past end of pattern! "); | |
fa9a63c5 | 985 | |
25fe55af | 986 | for (c = 0; c < 256; c++) |
96cc36cc | 987 | if (c / 8 < length |
fa9a63c5 RM |
988 | && (p[1 + (c/8)] & (1 << (c % 8)))) |
989 | { | |
990 | /* Are we starting a range? */ | |
991 | if (last + 1 == c && ! in_range) | |
992 | { | |
a1a052df | 993 | fprintf (stderr, "-"); |
fa9a63c5 RM |
994 | in_range = 1; |
995 | } | |
996 | /* Have we broken a range? */ | |
997 | else if (last + 1 != c && in_range) | |
96cc36cc | 998 | { |
a1a052df | 999 | fprintf (stderr, "%c", last); |
fa9a63c5 RM |
1000 | in_range = 0; |
1001 | } | |
5e69f11e | 1002 | |
fa9a63c5 | 1003 | if (! in_range) |
a1a052df | 1004 | fprintf (stderr, "%c", c); |
fa9a63c5 RM |
1005 | |
1006 | last = c; | |
25fe55af | 1007 | } |
fa9a63c5 RM |
1008 | |
1009 | if (in_range) | |
a1a052df | 1010 | fprintf (stderr, "%c", last); |
fa9a63c5 | 1011 | |
a1a052df | 1012 | fprintf (stderr, "]"); |
fa9a63c5 | 1013 | |
99633e97 | 1014 | p += 1 + length; |
96cc36cc | 1015 | |
96cc36cc | 1016 | if (has_range_table) |
99633e97 SM |
1017 | { |
1018 | int count; | |
a1a052df | 1019 | fprintf (stderr, "has-range-table"); |
99633e97 SM |
1020 | |
1021 | /* ??? Should print the range table; for now, just skip it. */ | |
1022 | p += 2; /* skip range table bits */ | |
1023 | EXTRACT_NUMBER_AND_INCR (count, p); | |
1024 | p = CHARSET_RANGE_TABLE_END (p, count); | |
1025 | } | |
fa9a63c5 RM |
1026 | } |
1027 | break; | |
1028 | ||
1029 | case begline: | |
a1a052df | 1030 | fprintf (stderr, "/begline"); |
25fe55af | 1031 | break; |
fa9a63c5 RM |
1032 | |
1033 | case endline: | |
a1a052df | 1034 | fprintf (stderr, "/endline"); |
25fe55af | 1035 | break; |
fa9a63c5 RM |
1036 | |
1037 | case on_failure_jump: | |
25fe55af | 1038 | extract_number_and_incr (&mcnt, &p); |
a1a052df | 1039 | fprintf (stderr, "/on_failure_jump to %d", p + mcnt - start); |
25fe55af | 1040 | break; |
fa9a63c5 RM |
1041 | |
1042 | case on_failure_keep_string_jump: | |
25fe55af | 1043 | extract_number_and_incr (&mcnt, &p); |
a1a052df | 1044 | fprintf (stderr, "/on_failure_keep_string_jump to %d", p + mcnt - start); |
25fe55af | 1045 | break; |
fa9a63c5 | 1046 | |
0683b6fa SM |
1047 | case on_failure_jump_nastyloop: |
1048 | extract_number_and_incr (&mcnt, &p); | |
a1a052df | 1049 | fprintf (stderr, "/on_failure_jump_nastyloop to %d", p + mcnt - start); |
0683b6fa SM |
1050 | break; |
1051 | ||
505bde11 | 1052 | case on_failure_jump_loop: |
fa9a63c5 | 1053 | extract_number_and_incr (&mcnt, &p); |
a1a052df | 1054 | fprintf (stderr, "/on_failure_jump_loop to %d", p + mcnt - start); |
5e69f11e RM |
1055 | break; |
1056 | ||
505bde11 | 1057 | case on_failure_jump_smart: |
fa9a63c5 | 1058 | extract_number_and_incr (&mcnt, &p); |
a1a052df | 1059 | fprintf (stderr, "/on_failure_jump_smart to %d", p + mcnt - start); |
5e69f11e RM |
1060 | break; |
1061 | ||
25fe55af | 1062 | case jump: |
fa9a63c5 | 1063 | extract_number_and_incr (&mcnt, &p); |
a1a052df | 1064 | fprintf (stderr, "/jump to %d", p + mcnt - start); |
fa9a63c5 RM |
1065 | break; |
1066 | ||
25fe55af RS |
1067 | case succeed_n: |
1068 | extract_number_and_incr (&mcnt, &p); | |
1069 | extract_number_and_incr (&mcnt2, &p); | |
a1a052df | 1070 | fprintf (stderr, "/succeed_n to %d, %d times", p - 2 + mcnt - start, mcnt2); |
25fe55af | 1071 | break; |
5e69f11e | 1072 | |
25fe55af RS |
1073 | case jump_n: |
1074 | extract_number_and_incr (&mcnt, &p); | |
1075 | extract_number_and_incr (&mcnt2, &p); | |
a1a052df | 1076 | fprintf (stderr, "/jump_n to %d, %d times", p - 2 + mcnt - start, mcnt2); |
25fe55af | 1077 | break; |
5e69f11e | 1078 | |
25fe55af RS |
1079 | case set_number_at: |
1080 | extract_number_and_incr (&mcnt, &p); | |
1081 | extract_number_and_incr (&mcnt2, &p); | |
a1a052df | 1082 | fprintf (stderr, "/set_number_at location %d to %d", p - 2 + mcnt - start, mcnt2); |
25fe55af | 1083 | break; |
5e69f11e | 1084 | |
25fe55af | 1085 | case wordbound: |
a1a052df | 1086 | fprintf (stderr, "/wordbound"); |
fa9a63c5 RM |
1087 | break; |
1088 | ||
1089 | case notwordbound: | |
a1a052df | 1090 | fprintf (stderr, "/notwordbound"); |
25fe55af | 1091 | break; |
fa9a63c5 RM |
1092 | |
1093 | case wordbeg: | |
a1a052df | 1094 | fprintf (stderr, "/wordbeg"); |
fa9a63c5 | 1095 | break; |
5e69f11e | 1096 | |
fa9a63c5 | 1097 | case wordend: |
a1a052df | 1098 | fprintf (stderr, "/wordend"); |
5e69f11e | 1099 | |
669fa600 SM |
1100 | case symbeg: |
1101 | printf ("/symbeg"); | |
1102 | break; | |
1103 | ||
1104 | case symend: | |
1105 | printf ("/symend"); | |
1106 | break; | |
1107 | ||
1fb352e0 | 1108 | case syntaxspec: |
a1a052df | 1109 | fprintf (stderr, "/syntaxspec"); |
1fb352e0 | 1110 | mcnt = *p++; |
a1a052df | 1111 | fprintf (stderr, "/%d", mcnt); |
1fb352e0 SM |
1112 | break; |
1113 | ||
1114 | case notsyntaxspec: | |
a1a052df | 1115 | fprintf (stderr, "/notsyntaxspec"); |
1fb352e0 | 1116 | mcnt = *p++; |
a1a052df | 1117 | fprintf (stderr, "/%d", mcnt); |
1fb352e0 SM |
1118 | break; |
1119 | ||
0b32bf0e | 1120 | # ifdef emacs |
fa9a63c5 | 1121 | case before_dot: |
a1a052df | 1122 | fprintf (stderr, "/before_dot"); |
25fe55af | 1123 | break; |
fa9a63c5 RM |
1124 | |
1125 | case at_dot: | |
a1a052df | 1126 | fprintf (stderr, "/at_dot"); |
25fe55af | 1127 | break; |
fa9a63c5 RM |
1128 | |
1129 | case after_dot: | |
a1a052df | 1130 | fprintf (stderr, "/after_dot"); |
25fe55af | 1131 | break; |
fa9a63c5 | 1132 | |
1fb352e0 | 1133 | case categoryspec: |
a1a052df | 1134 | fprintf (stderr, "/categoryspec"); |
fa9a63c5 | 1135 | mcnt = *p++; |
a1a052df | 1136 | fprintf (stderr, "/%d", mcnt); |
25fe55af | 1137 | break; |
5e69f11e | 1138 | |
1fb352e0 | 1139 | case notcategoryspec: |
a1a052df | 1140 | fprintf (stderr, "/notcategoryspec"); |
fa9a63c5 | 1141 | mcnt = *p++; |
a1a052df | 1142 | fprintf (stderr, "/%d", mcnt); |
fa9a63c5 | 1143 | break; |
0b32bf0e | 1144 | # endif /* emacs */ |
fa9a63c5 | 1145 | |
fa9a63c5 | 1146 | case begbuf: |
a1a052df | 1147 | fprintf (stderr, "/begbuf"); |
25fe55af | 1148 | break; |
fa9a63c5 RM |
1149 | |
1150 | case endbuf: | |
a1a052df | 1151 | fprintf (stderr, "/endbuf"); |
25fe55af | 1152 | break; |
fa9a63c5 | 1153 | |
25fe55af | 1154 | default: |
a1a052df | 1155 | fprintf (stderr, "?%d", *(p-1)); |
fa9a63c5 RM |
1156 | } |
1157 | ||
a1a052df | 1158 | fprintf (stderr, "\n"); |
fa9a63c5 RM |
1159 | } |
1160 | ||
a1a052df | 1161 | fprintf (stderr, "%d:\tend of pattern.\n", p - start); |
fa9a63c5 RM |
1162 | } |
1163 | ||
1164 | ||
1165 | void | |
1166 | print_compiled_pattern (bufp) | |
1167 | struct re_pattern_buffer *bufp; | |
1168 | { | |
01618498 | 1169 | re_char *buffer = bufp->buffer; |
fa9a63c5 RM |
1170 | |
1171 | print_partial_compiled_pattern (buffer, buffer + bufp->used); | |
4bb91c68 SM |
1172 | printf ("%ld bytes used/%ld bytes allocated.\n", |
1173 | bufp->used, bufp->allocated); | |
fa9a63c5 RM |
1174 | |
1175 | if (bufp->fastmap_accurate && bufp->fastmap) | |
1176 | { | |
1177 | printf ("fastmap: "); | |
1178 | print_fastmap (bufp->fastmap); | |
1179 | } | |
1180 | ||
1181 | printf ("re_nsub: %d\t", bufp->re_nsub); | |
1182 | printf ("regs_alloc: %d\t", bufp->regs_allocated); | |
1183 | printf ("can_be_null: %d\t", bufp->can_be_null); | |
fa9a63c5 RM |
1184 | printf ("no_sub: %d\t", bufp->no_sub); |
1185 | printf ("not_bol: %d\t", bufp->not_bol); | |
1186 | printf ("not_eol: %d\t", bufp->not_eol); | |
4bb91c68 | 1187 | printf ("syntax: %lx\n", bufp->syntax); |
505bde11 | 1188 | fflush (stdout); |
fa9a63c5 RM |
1189 | /* Perhaps we should print the translate table? */ |
1190 | } | |
1191 | ||
1192 | ||
1193 | void | |
1194 | print_double_string (where, string1, size1, string2, size2) | |
66f0296e SM |
1195 | re_char *where; |
1196 | re_char *string1; | |
1197 | re_char *string2; | |
fa9a63c5 RM |
1198 | int size1; |
1199 | int size2; | |
1200 | { | |
4bb91c68 | 1201 | int this_char; |
5e69f11e | 1202 | |
fa9a63c5 RM |
1203 | if (where == NULL) |
1204 | printf ("(null)"); | |
1205 | else | |
1206 | { | |
1207 | if (FIRST_STRING_P (where)) | |
25fe55af RS |
1208 | { |
1209 | for (this_char = where - string1; this_char < size1; this_char++) | |
1210 | putchar (string1[this_char]); | |
fa9a63c5 | 1211 | |
25fe55af RS |
1212 | where = string2; |
1213 | } | |
fa9a63c5 RM |
1214 | |
1215 | for (this_char = where - string2; this_char < size2; this_char++) | |
25fe55af | 1216 | putchar (string2[this_char]); |
fa9a63c5 RM |
1217 | } |
1218 | } | |
1219 | ||
1220 | #else /* not DEBUG */ | |
1221 | ||
0b32bf0e SM |
1222 | # undef assert |
1223 | # define assert(e) | |
fa9a63c5 | 1224 | |
0b32bf0e SM |
1225 | # define DEBUG_STATEMENT(e) |
1226 | # define DEBUG_PRINT1(x) | |
1227 | # define DEBUG_PRINT2(x1, x2) | |
1228 | # define DEBUG_PRINT3(x1, x2, x3) | |
1229 | # define DEBUG_PRINT4(x1, x2, x3, x4) | |
1230 | # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) | |
1231 | # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) | |
fa9a63c5 RM |
1232 | |
1233 | #endif /* not DEBUG */ | |
1234 | \f | |
1235 | /* Set by `re_set_syntax' to the current regexp syntax to recognize. Can | |
1236 | also be assigned to arbitrarily: each pattern buffer stores its own | |
1237 | syntax, so it can be changed between regex compilations. */ | |
1238 | /* This has no initializer because initialized variables in Emacs | |
1239 | become read-only after dumping. */ | |
1240 | reg_syntax_t re_syntax_options; | |
1241 | ||
1242 | ||
1243 | /* Specify the precise syntax of regexps for compilation. This provides | |
1244 | for compatibility for various utilities which historically have | |
1245 | different, incompatible syntaxes. | |
1246 | ||
1247 | The argument SYNTAX is a bit mask comprised of the various bits | |
4bb91c68 | 1248 | defined in regex.h. We return the old syntax. */ |
fa9a63c5 RM |
1249 | |
1250 | reg_syntax_t | |
1251 | re_set_syntax (syntax) | |
1252 | reg_syntax_t syntax; | |
1253 | { | |
1254 | reg_syntax_t ret = re_syntax_options; | |
5e69f11e | 1255 | |
fa9a63c5 RM |
1256 | re_syntax_options = syntax; |
1257 | return ret; | |
1258 | } | |
c0f9ea08 | 1259 | WEAK_ALIAS (__re_set_syntax, re_set_syntax) |
fa9a63c5 RM |
1260 | \f |
1261 | /* This table gives an error message for each of the error codes listed | |
4bb91c68 | 1262 | in regex.h. Obviously the order here has to be same as there. |
fa9a63c5 | 1263 | POSIX doesn't require that we do anything for REG_NOERROR, |
4bb91c68 | 1264 | but why not be nice? */ |
fa9a63c5 RM |
1265 | |
1266 | static const char *re_error_msgid[] = | |
5e69f11e RM |
1267 | { |
1268 | gettext_noop ("Success"), /* REG_NOERROR */ | |
1269 | gettext_noop ("No match"), /* REG_NOMATCH */ | |
1270 | gettext_noop ("Invalid regular expression"), /* REG_BADPAT */ | |
1271 | gettext_noop ("Invalid collation character"), /* REG_ECOLLATE */ | |
1272 | gettext_noop ("Invalid character class name"), /* REG_ECTYPE */ | |
1273 | gettext_noop ("Trailing backslash"), /* REG_EESCAPE */ | |
1274 | gettext_noop ("Invalid back reference"), /* REG_ESUBREG */ | |
1275 | gettext_noop ("Unmatched [ or [^"), /* REG_EBRACK */ | |
1276 | gettext_noop ("Unmatched ( or \\("), /* REG_EPAREN */ | |
1277 | gettext_noop ("Unmatched \\{"), /* REG_EBRACE */ | |
1278 | gettext_noop ("Invalid content of \\{\\}"), /* REG_BADBR */ | |
1279 | gettext_noop ("Invalid range end"), /* REG_ERANGE */ | |
1280 | gettext_noop ("Memory exhausted"), /* REG_ESPACE */ | |
1281 | gettext_noop ("Invalid preceding regular expression"), /* REG_BADRPT */ | |
1282 | gettext_noop ("Premature end of regular expression"), /* REG_EEND */ | |
1283 | gettext_noop ("Regular expression too big"), /* REG_ESIZE */ | |
1284 | gettext_noop ("Unmatched ) or \\)"), /* REG_ERPAREN */ | |
fa9a63c5 RM |
1285 | }; |
1286 | \f | |
4bb91c68 | 1287 | /* Avoiding alloca during matching, to placate r_alloc. */ |
fa9a63c5 RM |
1288 | |
1289 | /* Define MATCH_MAY_ALLOCATE unless we need to make sure that the | |
1290 | searching and matching functions should not call alloca. On some | |
1291 | systems, alloca is implemented in terms of malloc, and if we're | |
1292 | using the relocating allocator routines, then malloc could cause a | |
1293 | relocation, which might (if the strings being searched are in the | |
1294 | ralloc heap) shift the data out from underneath the regexp | |
1295 | routines. | |
1296 | ||
5e69f11e | 1297 | Here's another reason to avoid allocation: Emacs |
fa9a63c5 RM |
1298 | processes input from X in a signal handler; processing X input may |
1299 | call malloc; if input arrives while a matching routine is calling | |
1300 | malloc, then we're scrod. But Emacs can't just block input while | |
1301 | calling matching routines; then we don't notice interrupts when | |
1302 | they come in. So, Emacs blocks input around all regexp calls | |
1303 | except the matching calls, which it leaves unprotected, in the | |
1304 | faith that they will not malloc. */ | |
1305 | ||
1306 | /* Normally, this is fine. */ | |
1307 | #define MATCH_MAY_ALLOCATE | |
1308 | ||
1309 | /* When using GNU C, we are not REALLY using the C alloca, no matter | |
1310 | what config.h may say. So don't take precautions for it. */ | |
1311 | #ifdef __GNUC__ | |
0b32bf0e | 1312 | # undef C_ALLOCA |
fa9a63c5 RM |
1313 | #endif |
1314 | ||
1315 | /* The match routines may not allocate if (1) they would do it with malloc | |
1316 | and (2) it's not safe for them to use malloc. | |
1317 | Note that if REL_ALLOC is defined, matching would not use malloc for the | |
1318 | failure stack, but we would still use it for the register vectors; | |
4bb91c68 | 1319 | so REL_ALLOC should not affect this. */ |
0b32bf0e SM |
1320 | #if (defined C_ALLOCA || defined REGEX_MALLOC) && defined emacs |
1321 | # undef MATCH_MAY_ALLOCATE | |
fa9a63c5 RM |
1322 | #endif |
1323 | ||
1324 | \f | |
1325 | /* Failure stack declarations and macros; both re_compile_fastmap and | |
1326 | re_match_2 use a failure stack. These have to be macros because of | |
1327 | REGEX_ALLOCATE_STACK. */ | |
5e69f11e | 1328 | |
fa9a63c5 | 1329 | |
320a2a73 | 1330 | /* Approximate number of failure points for which to initially allocate space |
fa9a63c5 RM |
1331 | when matching. If this number is exceeded, we allocate more |
1332 | space, so it is not a hard limit. */ | |
1333 | #ifndef INIT_FAILURE_ALLOC | |
0b32bf0e | 1334 | # define INIT_FAILURE_ALLOC 20 |
fa9a63c5 RM |
1335 | #endif |
1336 | ||
1337 | /* Roughly the maximum number of failure points on the stack. Would be | |
320a2a73 | 1338 | exactly that if always used TYPICAL_FAILURE_SIZE items each time we failed. |
fa9a63c5 | 1339 | This is a variable only so users of regex can assign to it; we never |
ada30c0e SM |
1340 | change it ourselves. We always multiply it by TYPICAL_FAILURE_SIZE |
1341 | before using it, so it should probably be a byte-count instead. */ | |
c0f9ea08 SM |
1342 | # if defined MATCH_MAY_ALLOCATE |
1343 | /* Note that 4400 was enough to cause a crash on Alpha OSF/1, | |
320a2a73 KH |
1344 | whose default stack limit is 2mb. In order for a larger |
1345 | value to work reliably, you have to try to make it accord | |
1346 | with the process stack limit. */ | |
c0f9ea08 SM |
1347 | size_t re_max_failures = 40000; |
1348 | # else | |
1349 | size_t re_max_failures = 4000; | |
1350 | # endif | |
fa9a63c5 RM |
1351 | |
1352 | union fail_stack_elt | |
1353 | { | |
01618498 | 1354 | re_char *pointer; |
c0f9ea08 SM |
1355 | /* This should be the biggest `int' that's no bigger than a pointer. */ |
1356 | long integer; | |
fa9a63c5 RM |
1357 | }; |
1358 | ||
1359 | typedef union fail_stack_elt fail_stack_elt_t; | |
1360 | ||
1361 | typedef struct | |
1362 | { | |
1363 | fail_stack_elt_t *stack; | |
c0f9ea08 SM |
1364 | size_t size; |
1365 | size_t avail; /* Offset of next open position. */ | |
1366 | size_t frame; /* Offset of the cur constructed frame. */ | |
fa9a63c5 RM |
1367 | } fail_stack_type; |
1368 | ||
505bde11 | 1369 | #define FAIL_STACK_EMPTY() (fail_stack.frame == 0) |
fa9a63c5 RM |
1370 | #define FAIL_STACK_FULL() (fail_stack.avail == fail_stack.size) |
1371 | ||
1372 | ||
1373 | /* Define macros to initialize and free the failure stack. | |
1374 | Do `return -2' if the alloc fails. */ | |
1375 | ||
1376 | #ifdef MATCH_MAY_ALLOCATE | |
0b32bf0e | 1377 | # define INIT_FAIL_STACK() \ |
fa9a63c5 RM |
1378 | do { \ |
1379 | fail_stack.stack = (fail_stack_elt_t *) \ | |
320a2a73 KH |
1380 | REGEX_ALLOCATE_STACK (INIT_FAILURE_ALLOC * TYPICAL_FAILURE_SIZE \ |
1381 | * sizeof (fail_stack_elt_t)); \ | |
fa9a63c5 RM |
1382 | \ |
1383 | if (fail_stack.stack == NULL) \ | |
1384 | return -2; \ | |
1385 | \ | |
1386 | fail_stack.size = INIT_FAILURE_ALLOC; \ | |
1387 | fail_stack.avail = 0; \ | |
505bde11 | 1388 | fail_stack.frame = 0; \ |
fa9a63c5 RM |
1389 | } while (0) |
1390 | ||
0b32bf0e | 1391 | # define RESET_FAIL_STACK() REGEX_FREE_STACK (fail_stack.stack) |
fa9a63c5 | 1392 | #else |
0b32bf0e | 1393 | # define INIT_FAIL_STACK() \ |
fa9a63c5 RM |
1394 | do { \ |
1395 | fail_stack.avail = 0; \ | |
505bde11 | 1396 | fail_stack.frame = 0; \ |
fa9a63c5 RM |
1397 | } while (0) |
1398 | ||
0b32bf0e | 1399 | # define RESET_FAIL_STACK() ((void)0) |
fa9a63c5 RM |
1400 | #endif |
1401 | ||
1402 | ||
320a2a73 KH |
1403 | /* Double the size of FAIL_STACK, up to a limit |
1404 | which allows approximately `re_max_failures' items. | |
fa9a63c5 RM |
1405 | |
1406 | Return 1 if succeeds, and 0 if either ran out of memory | |
5e69f11e RM |
1407 | allocating space for it or it was already too large. |
1408 | ||
4bb91c68 | 1409 | REGEX_REALLOCATE_STACK requires `destination' be declared. */ |
fa9a63c5 | 1410 | |
320a2a73 KH |
1411 | /* Factor to increase the failure stack size by |
1412 | when we increase it. | |
1413 | This used to be 2, but 2 was too wasteful | |
1414 | because the old discarded stacks added up to as much space | |
1415 | were as ultimate, maximum-size stack. */ | |
1416 | #define FAIL_STACK_GROWTH_FACTOR 4 | |
1417 | ||
1418 | #define GROW_FAIL_STACK(fail_stack) \ | |
eead07d6 KH |
1419 | (((fail_stack).size * sizeof (fail_stack_elt_t) \ |
1420 | >= re_max_failures * TYPICAL_FAILURE_SIZE) \ | |
fa9a63c5 | 1421 | ? 0 \ |
320a2a73 KH |
1422 | : ((fail_stack).stack \ |
1423 | = (fail_stack_elt_t *) \ | |
25fe55af RS |
1424 | REGEX_REALLOCATE_STACK ((fail_stack).stack, \ |
1425 | (fail_stack).size * sizeof (fail_stack_elt_t), \ | |
320a2a73 KH |
1426 | MIN (re_max_failures * TYPICAL_FAILURE_SIZE, \ |
1427 | ((fail_stack).size * sizeof (fail_stack_elt_t) \ | |
1428 | * FAIL_STACK_GROWTH_FACTOR))), \ | |
fa9a63c5 RM |
1429 | \ |
1430 | (fail_stack).stack == NULL \ | |
1431 | ? 0 \ | |
6453db45 KH |
1432 | : ((fail_stack).size \ |
1433 | = (MIN (re_max_failures * TYPICAL_FAILURE_SIZE, \ | |
1434 | ((fail_stack).size * sizeof (fail_stack_elt_t) \ | |
1435 | * FAIL_STACK_GROWTH_FACTOR)) \ | |
1436 | / sizeof (fail_stack_elt_t)), \ | |
25fe55af | 1437 | 1))) |
fa9a63c5 RM |
1438 | |
1439 | ||
fa9a63c5 RM |
1440 | /* Push a pointer value onto the failure stack. |
1441 | Assumes the variable `fail_stack'. Probably should only | |
4bb91c68 | 1442 | be called from within `PUSH_FAILURE_POINT'. */ |
fa9a63c5 | 1443 | #define PUSH_FAILURE_POINTER(item) \ |
01618498 | 1444 | fail_stack.stack[fail_stack.avail++].pointer = (item) |
fa9a63c5 RM |
1445 | |
1446 | /* This pushes an integer-valued item onto the failure stack. | |
1447 | Assumes the variable `fail_stack'. Probably should only | |
4bb91c68 | 1448 | be called from within `PUSH_FAILURE_POINT'. */ |
fa9a63c5 RM |
1449 | #define PUSH_FAILURE_INT(item) \ |
1450 | fail_stack.stack[fail_stack.avail++].integer = (item) | |
1451 | ||
1452 | /* Push a fail_stack_elt_t value onto the failure stack. | |
1453 | Assumes the variable `fail_stack'. Probably should only | |
4bb91c68 | 1454 | be called from within `PUSH_FAILURE_POINT'. */ |
fa9a63c5 RM |
1455 | #define PUSH_FAILURE_ELT(item) \ |
1456 | fail_stack.stack[fail_stack.avail++] = (item) | |
1457 | ||
1458 | /* These three POP... operations complement the three PUSH... operations. | |
1459 | All assume that `fail_stack' is nonempty. */ | |
1460 | #define POP_FAILURE_POINTER() fail_stack.stack[--fail_stack.avail].pointer | |
1461 | #define POP_FAILURE_INT() fail_stack.stack[--fail_stack.avail].integer | |
1462 | #define POP_FAILURE_ELT() fail_stack.stack[--fail_stack.avail] | |
1463 | ||
505bde11 SM |
1464 | /* Individual items aside from the registers. */ |
1465 | #define NUM_NONREG_ITEMS 3 | |
1466 | ||
1467 | /* Used to examine the stack (to detect infinite loops). */ | |
1468 | #define FAILURE_PAT(h) fail_stack.stack[(h) - 1].pointer | |
66f0296e | 1469 | #define FAILURE_STR(h) (fail_stack.stack[(h) - 2].pointer) |
505bde11 SM |
1470 | #define NEXT_FAILURE_HANDLE(h) fail_stack.stack[(h) - 3].integer |
1471 | #define TOP_FAILURE_HANDLE() fail_stack.frame | |
fa9a63c5 RM |
1472 | |
1473 | ||
505bde11 SM |
1474 | #define ENSURE_FAIL_STACK(space) \ |
1475 | while (REMAINING_AVAIL_SLOTS <= space) { \ | |
1476 | if (!GROW_FAIL_STACK (fail_stack)) \ | |
1477 | return -2; \ | |
1478 | DEBUG_PRINT2 ("\n Doubled stack; size now: %d\n", (fail_stack).size);\ | |
1479 | DEBUG_PRINT2 (" slots available: %d\n", REMAINING_AVAIL_SLOTS);\ | |
1480 | } | |
1481 | ||
1482 | /* Push register NUM onto the stack. */ | |
1483 | #define PUSH_FAILURE_REG(num) \ | |
1484 | do { \ | |
1485 | char *destination; \ | |
1486 | ENSURE_FAIL_STACK(3); \ | |
1487 | DEBUG_PRINT4 (" Push reg %d (spanning %p -> %p)\n", \ | |
1488 | num, regstart[num], regend[num]); \ | |
1489 | PUSH_FAILURE_POINTER (regstart[num]); \ | |
1490 | PUSH_FAILURE_POINTER (regend[num]); \ | |
1491 | PUSH_FAILURE_INT (num); \ | |
1492 | } while (0) | |
1493 | ||
01618498 SM |
1494 | /* Change the counter's value to VAL, but make sure that it will |
1495 | be reset when backtracking. */ | |
1496 | #define PUSH_NUMBER(ptr,val) \ | |
dc1e502d SM |
1497 | do { \ |
1498 | char *destination; \ | |
1499 | int c; \ | |
1500 | ENSURE_FAIL_STACK(3); \ | |
1501 | EXTRACT_NUMBER (c, ptr); \ | |
01618498 | 1502 | DEBUG_PRINT4 (" Push number %p = %d -> %d\n", ptr, c, val); \ |
dc1e502d SM |
1503 | PUSH_FAILURE_INT (c); \ |
1504 | PUSH_FAILURE_POINTER (ptr); \ | |
1505 | PUSH_FAILURE_INT (-1); \ | |
01618498 | 1506 | STORE_NUMBER (ptr, val); \ |
dc1e502d SM |
1507 | } while (0) |
1508 | ||
505bde11 | 1509 | /* Pop a saved register off the stack. */ |
dc1e502d | 1510 | #define POP_FAILURE_REG_OR_COUNT() \ |
505bde11 SM |
1511 | do { \ |
1512 | int reg = POP_FAILURE_INT (); \ | |
dc1e502d SM |
1513 | if (reg == -1) \ |
1514 | { \ | |
1515 | /* It's a counter. */ \ | |
6dcf2d0e SM |
1516 | /* Here, we discard `const', making re_match non-reentrant. */ \ |
1517 | unsigned char *ptr = (unsigned char*) POP_FAILURE_POINTER (); \ | |
dc1e502d SM |
1518 | reg = POP_FAILURE_INT (); \ |
1519 | STORE_NUMBER (ptr, reg); \ | |
1520 | DEBUG_PRINT3 (" Pop counter %p = %d\n", ptr, reg); \ | |
1521 | } \ | |
1522 | else \ | |
1523 | { \ | |
1524 | regend[reg] = POP_FAILURE_POINTER (); \ | |
1525 | regstart[reg] = POP_FAILURE_POINTER (); \ | |
1526 | DEBUG_PRINT4 (" Pop reg %d (spanning %p -> %p)\n", \ | |
1527 | reg, regstart[reg], regend[reg]); \ | |
1528 | } \ | |
505bde11 SM |
1529 | } while (0) |
1530 | ||
1531 | /* Check that we are not stuck in an infinite loop. */ | |
1532 | #define CHECK_INFINITE_LOOP(pat_cur, string_place) \ | |
1533 | do { \ | |
f6df485f | 1534 | int failure = TOP_FAILURE_HANDLE (); \ |
505bde11 | 1535 | /* Check for infinite matching loops */ \ |
f6df485f RS |
1536 | while (failure > 0 \ |
1537 | && (FAILURE_STR (failure) == string_place \ | |
1538 | || FAILURE_STR (failure) == NULL)) \ | |
505bde11 SM |
1539 | { \ |
1540 | assert (FAILURE_PAT (failure) >= bufp->buffer \ | |
66f0296e | 1541 | && FAILURE_PAT (failure) <= bufp->buffer + bufp->used); \ |
505bde11 | 1542 | if (FAILURE_PAT (failure) == pat_cur) \ |
f6df485f | 1543 | { \ |
6df42991 SM |
1544 | cycle = 1; \ |
1545 | break; \ | |
f6df485f | 1546 | } \ |
66f0296e | 1547 | DEBUG_PRINT2 (" Other pattern: %p\n", FAILURE_PAT (failure)); \ |
505bde11 SM |
1548 | failure = NEXT_FAILURE_HANDLE(failure); \ |
1549 | } \ | |
1550 | DEBUG_PRINT2 (" Other string: %p\n", FAILURE_STR (failure)); \ | |
1551 | } while (0) | |
6df42991 | 1552 | |
fa9a63c5 | 1553 | /* Push the information about the state we will need |
5e69f11e RM |
1554 | if we ever fail back to it. |
1555 | ||
505bde11 | 1556 | Requires variables fail_stack, regstart, regend and |
320a2a73 | 1557 | num_regs be declared. GROW_FAIL_STACK requires `destination' be |
fa9a63c5 | 1558 | declared. |
5e69f11e | 1559 | |
fa9a63c5 RM |
1560 | Does `return FAILURE_CODE' if runs out of memory. */ |
1561 | ||
505bde11 SM |
1562 | #define PUSH_FAILURE_POINT(pattern, string_place) \ |
1563 | do { \ | |
1564 | char *destination; \ | |
1565 | /* Must be int, so when we don't save any registers, the arithmetic \ | |
1566 | of 0 + -1 isn't done as unsigned. */ \ | |
1567 | \ | |
505bde11 | 1568 | DEBUG_STATEMENT (nfailure_points_pushed++); \ |
4bb91c68 | 1569 | DEBUG_PRINT1 ("\nPUSH_FAILURE_POINT:\n"); \ |
505bde11 SM |
1570 | DEBUG_PRINT2 (" Before push, next avail: %d\n", (fail_stack).avail); \ |
1571 | DEBUG_PRINT2 (" size: %d\n", (fail_stack).size);\ | |
1572 | \ | |
1573 | ENSURE_FAIL_STACK (NUM_NONREG_ITEMS); \ | |
1574 | \ | |
1575 | DEBUG_PRINT1 ("\n"); \ | |
1576 | \ | |
1577 | DEBUG_PRINT2 (" Push frame index: %d\n", fail_stack.frame); \ | |
1578 | PUSH_FAILURE_INT (fail_stack.frame); \ | |
1579 | \ | |
1580 | DEBUG_PRINT2 (" Push string %p: `", string_place); \ | |
1581 | DEBUG_PRINT_DOUBLE_STRING (string_place, string1, size1, string2, size2);\ | |
1582 | DEBUG_PRINT1 ("'\n"); \ | |
1583 | PUSH_FAILURE_POINTER (string_place); \ | |
1584 | \ | |
1585 | DEBUG_PRINT2 (" Push pattern %p: ", pattern); \ | |
1586 | DEBUG_PRINT_COMPILED_PATTERN (bufp, pattern, pend); \ | |
1587 | PUSH_FAILURE_POINTER (pattern); \ | |
1588 | \ | |
1589 | /* Close the frame by moving the frame pointer past it. */ \ | |
1590 | fail_stack.frame = fail_stack.avail; \ | |
1591 | } while (0) | |
fa9a63c5 | 1592 | |
320a2a73 KH |
1593 | /* Estimate the size of data pushed by a typical failure stack entry. |
1594 | An estimate is all we need, because all we use this for | |
1595 | is to choose a limit for how big to make the failure stack. */ | |
ada30c0e | 1596 | /* BEWARE, the value `20' is hard-coded in emacs.c:main(). */ |
320a2a73 | 1597 | #define TYPICAL_FAILURE_SIZE 20 |
fa9a63c5 | 1598 | |
fa9a63c5 RM |
1599 | /* How many items can still be added to the stack without overflowing it. */ |
1600 | #define REMAINING_AVAIL_SLOTS ((fail_stack).size - (fail_stack).avail) | |
1601 | ||
1602 | ||
1603 | /* Pops what PUSH_FAIL_STACK pushes. | |
1604 | ||
1605 | We restore into the parameters, all of which should be lvalues: | |
1606 | STR -- the saved data position. | |
1607 | PAT -- the saved pattern position. | |
fa9a63c5 | 1608 | REGSTART, REGEND -- arrays of string positions. |
5e69f11e | 1609 | |
fa9a63c5 | 1610 | Also assumes the variables `fail_stack' and (if debugging), `bufp', |
25fe55af | 1611 | `pend', `string1', `size1', `string2', and `size2'. */ |
fa9a63c5 | 1612 | |
505bde11 SM |
1613 | #define POP_FAILURE_POINT(str, pat) \ |
1614 | do { \ | |
fa9a63c5 RM |
1615 | assert (!FAIL_STACK_EMPTY ()); \ |
1616 | \ | |
1617 | /* Remove failure points and point to how many regs pushed. */ \ | |
1618 | DEBUG_PRINT1 ("POP_FAILURE_POINT:\n"); \ | |
1619 | DEBUG_PRINT2 (" Before pop, next avail: %d\n", fail_stack.avail); \ | |
25fe55af | 1620 | DEBUG_PRINT2 (" size: %d\n", fail_stack.size); \ |
fa9a63c5 | 1621 | \ |
505bde11 SM |
1622 | /* Pop the saved registers. */ \ |
1623 | while (fail_stack.frame < fail_stack.avail) \ | |
dc1e502d | 1624 | POP_FAILURE_REG_OR_COUNT (); \ |
fa9a63c5 | 1625 | \ |
01618498 | 1626 | pat = POP_FAILURE_POINTER (); \ |
505bde11 SM |
1627 | DEBUG_PRINT2 (" Popping pattern %p: ", pat); \ |
1628 | DEBUG_PRINT_COMPILED_PATTERN (bufp, pat, pend); \ | |
fa9a63c5 RM |
1629 | \ |
1630 | /* If the saved string location is NULL, it came from an \ | |
1631 | on_failure_keep_string_jump opcode, and we want to throw away the \ | |
1632 | saved NULL, thus retaining our current position in the string. */ \ | |
01618498 | 1633 | str = POP_FAILURE_POINTER (); \ |
505bde11 | 1634 | DEBUG_PRINT2 (" Popping string %p: `", str); \ |
fa9a63c5 RM |
1635 | DEBUG_PRINT_DOUBLE_STRING (str, string1, size1, string2, size2); \ |
1636 | DEBUG_PRINT1 ("'\n"); \ | |
1637 | \ | |
505bde11 SM |
1638 | fail_stack.frame = POP_FAILURE_INT (); \ |
1639 | DEBUG_PRINT2 (" Popping frame index: %d\n", fail_stack.frame); \ | |
fa9a63c5 | 1640 | \ |
505bde11 SM |
1641 | assert (fail_stack.avail >= 0); \ |
1642 | assert (fail_stack.frame <= fail_stack.avail); \ | |
fa9a63c5 | 1643 | \ |
fa9a63c5 | 1644 | DEBUG_STATEMENT (nfailure_points_popped++); \ |
505bde11 | 1645 | } while (0) /* POP_FAILURE_POINT */ |
fa9a63c5 RM |
1646 | |
1647 | ||
1648 | \f | |
fa9a63c5 | 1649 | /* Registers are set to a sentinel when they haven't yet matched. */ |
4bb91c68 | 1650 | #define REG_UNSET(e) ((e) == NULL) |
fa9a63c5 RM |
1651 | \f |
1652 | /* Subroutine declarations and macros for regex_compile. */ | |
1653 | ||
4bb91c68 SM |
1654 | static reg_errcode_t regex_compile _RE_ARGS ((re_char *pattern, size_t size, |
1655 | reg_syntax_t syntax, | |
1656 | struct re_pattern_buffer *bufp)); | |
1657 | static void store_op1 _RE_ARGS ((re_opcode_t op, unsigned char *loc, int arg)); | |
1658 | static void store_op2 _RE_ARGS ((re_opcode_t op, unsigned char *loc, | |
1659 | int arg1, int arg2)); | |
1660 | static void insert_op1 _RE_ARGS ((re_opcode_t op, unsigned char *loc, | |
1661 | int arg, unsigned char *end)); | |
1662 | static void insert_op2 _RE_ARGS ((re_opcode_t op, unsigned char *loc, | |
1663 | int arg1, int arg2, unsigned char *end)); | |
01618498 SM |
1664 | static boolean at_begline_loc_p _RE_ARGS ((re_char *pattern, |
1665 | re_char *p, | |
4bb91c68 | 1666 | reg_syntax_t syntax)); |
01618498 SM |
1667 | static boolean at_endline_loc_p _RE_ARGS ((re_char *p, |
1668 | re_char *pend, | |
4bb91c68 | 1669 | reg_syntax_t syntax)); |
01618498 SM |
1670 | static re_char *skip_one_char _RE_ARGS ((re_char *p)); |
1671 | static int analyse_first _RE_ARGS ((re_char *p, re_char *pend, | |
4bb91c68 | 1672 | char *fastmap, const int multibyte)); |
fa9a63c5 | 1673 | |
fa9a63c5 | 1674 | /* Fetch the next character in the uncompiled pattern, with no |
4bb91c68 | 1675 | translation. */ |
36595814 | 1676 | #define PATFETCH(c) \ |
2d1675e4 SM |
1677 | do { \ |
1678 | int len; \ | |
1679 | if (p == pend) return REG_EEND; \ | |
1680 | c = RE_STRING_CHAR_AND_LENGTH (p, pend - p, len); \ | |
1681 | p += len; \ | |
fa9a63c5 RM |
1682 | } while (0) |
1683 | ||
fa9a63c5 RM |
1684 | |
1685 | /* If `translate' is non-null, return translate[D], else just D. We | |
1686 | cast the subscript to translate because some data is declared as | |
1687 | `char *', to avoid warnings when a string constant is passed. But | |
1688 | when we use a character as a subscript we must make it unsigned. */ | |
6676cb1c | 1689 | #ifndef TRANSLATE |
0b32bf0e | 1690 | # define TRANSLATE(d) \ |
66f0296e | 1691 | (RE_TRANSLATE_P (translate) ? RE_TRANSLATE (translate, (d)) : (d)) |
6676cb1c | 1692 | #endif |
fa9a63c5 RM |
1693 | |
1694 | ||
1695 | /* Macros for outputting the compiled pattern into `buffer'. */ | |
1696 | ||
1697 | /* If the buffer isn't allocated when it comes in, use this. */ | |
1698 | #define INIT_BUF_SIZE 32 | |
1699 | ||
4bb91c68 | 1700 | /* Make sure we have at least N more bytes of space in buffer. */ |
fa9a63c5 | 1701 | #define GET_BUFFER_SPACE(n) \ |
01618498 | 1702 | while ((size_t) (b - bufp->buffer + (n)) > bufp->allocated) \ |
fa9a63c5 RM |
1703 | EXTEND_BUFFER () |
1704 | ||
1705 | /* Make sure we have one more byte of buffer space and then add C to it. */ | |
1706 | #define BUF_PUSH(c) \ | |
1707 | do { \ | |
1708 | GET_BUFFER_SPACE (1); \ | |
1709 | *b++ = (unsigned char) (c); \ | |
1710 | } while (0) | |
1711 | ||
1712 | ||
1713 | /* Ensure we have two more bytes of buffer space and then append C1 and C2. */ | |
1714 | #define BUF_PUSH_2(c1, c2) \ | |
1715 | do { \ | |
1716 | GET_BUFFER_SPACE (2); \ | |
1717 | *b++ = (unsigned char) (c1); \ | |
1718 | *b++ = (unsigned char) (c2); \ | |
1719 | } while (0) | |
1720 | ||
1721 | ||
4bb91c68 | 1722 | /* As with BUF_PUSH_2, except for three bytes. */ |
fa9a63c5 RM |
1723 | #define BUF_PUSH_3(c1, c2, c3) \ |
1724 | do { \ | |
1725 | GET_BUFFER_SPACE (3); \ | |
1726 | *b++ = (unsigned char) (c1); \ | |
1727 | *b++ = (unsigned char) (c2); \ | |
1728 | *b++ = (unsigned char) (c3); \ | |
1729 | } while (0) | |
1730 | ||
1731 | ||
1732 | /* Store a jump with opcode OP at LOC to location TO. We store a | |
4bb91c68 | 1733 | relative address offset by the three bytes the jump itself occupies. */ |
fa9a63c5 RM |
1734 | #define STORE_JUMP(op, loc, to) \ |
1735 | store_op1 (op, loc, (to) - (loc) - 3) | |
1736 | ||
1737 | /* Likewise, for a two-argument jump. */ | |
1738 | #define STORE_JUMP2(op, loc, to, arg) \ | |
1739 | store_op2 (op, loc, (to) - (loc) - 3, arg) | |
1740 | ||
4bb91c68 | 1741 | /* Like `STORE_JUMP', but for inserting. Assume `b' is the buffer end. */ |
fa9a63c5 RM |
1742 | #define INSERT_JUMP(op, loc, to) \ |
1743 | insert_op1 (op, loc, (to) - (loc) - 3, b) | |
1744 | ||
1745 | /* Like `STORE_JUMP2', but for inserting. Assume `b' is the buffer end. */ | |
1746 | #define INSERT_JUMP2(op, loc, to, arg) \ | |
1747 | insert_op2 (op, loc, (to) - (loc) - 3, arg, b) | |
1748 | ||
1749 | ||
1750 | /* This is not an arbitrary limit: the arguments which represent offsets | |
275a3409 | 1751 | into the pattern are two bytes long. So if 2^15 bytes turns out to |
fa9a63c5 | 1752 | be too small, many things would have to change. */ |
275a3409 RS |
1753 | # define MAX_BUF_SIZE (1L << 15) |
1754 | ||
1755 | #if 0 /* This is when we thought it could be 2^16 bytes. */ | |
4bb91c68 SM |
1756 | /* Any other compiler which, like MSC, has allocation limit below 2^16 |
1757 | bytes will have to use approach similar to what was done below for | |
1758 | MSC and drop MAX_BUF_SIZE a bit. Otherwise you may end up | |
1759 | reallocating to 0 bytes. Such thing is not going to work too well. | |
1760 | You have been warned!! */ | |
1761 | #if defined _MSC_VER && !defined WIN32 | |
1762 | /* Microsoft C 16-bit versions limit malloc to approx 65512 bytes. */ | |
1763 | # define MAX_BUF_SIZE 65500L | |
1764 | #else | |
1765 | # define MAX_BUF_SIZE (1L << 16) | |
1766 | #endif | |
275a3409 | 1767 | #endif /* 0 */ |
fa9a63c5 RM |
1768 | |
1769 | /* Extend the buffer by twice its current size via realloc and | |
1770 | reset the pointers that pointed into the old block to point to the | |
1771 | correct places in the new one. If extending the buffer results in it | |
4bb91c68 SM |
1772 | being larger than MAX_BUF_SIZE, then flag memory exhausted. */ |
1773 | #if __BOUNDED_POINTERS__ | |
1774 | # define SET_HIGH_BOUND(P) (__ptrhigh (P) = __ptrlow (P) + bufp->allocated) | |
1775 | # define MOVE_BUFFER_POINTER(P) \ | |
1776 | (__ptrlow (P) += incr, SET_HIGH_BOUND (P), __ptrvalue (P) += incr) | |
1777 | # define ELSE_EXTEND_BUFFER_HIGH_BOUND \ | |
1778 | else \ | |
1779 | { \ | |
1780 | SET_HIGH_BOUND (b); \ | |
1781 | SET_HIGH_BOUND (begalt); \ | |
1782 | if (fixup_alt_jump) \ | |
1783 | SET_HIGH_BOUND (fixup_alt_jump); \ | |
1784 | if (laststart) \ | |
1785 | SET_HIGH_BOUND (laststart); \ | |
1786 | if (pending_exact) \ | |
1787 | SET_HIGH_BOUND (pending_exact); \ | |
1788 | } | |
1789 | #else | |
1790 | # define MOVE_BUFFER_POINTER(P) (P) += incr | |
1791 | # define ELSE_EXTEND_BUFFER_HIGH_BOUND | |
1792 | #endif | |
fa9a63c5 | 1793 | #define EXTEND_BUFFER() \ |
25fe55af | 1794 | do { \ |
01618498 | 1795 | re_char *old_buffer = bufp->buffer; \ |
25fe55af | 1796 | if (bufp->allocated == MAX_BUF_SIZE) \ |
fa9a63c5 RM |
1797 | return REG_ESIZE; \ |
1798 | bufp->allocated <<= 1; \ | |
1799 | if (bufp->allocated > MAX_BUF_SIZE) \ | |
25fe55af | 1800 | bufp->allocated = MAX_BUF_SIZE; \ |
01618498 | 1801 | RETALLOC (bufp->buffer, bufp->allocated, unsigned char); \ |
fa9a63c5 RM |
1802 | if (bufp->buffer == NULL) \ |
1803 | return REG_ESPACE; \ | |
1804 | /* If the buffer moved, move all the pointers into it. */ \ | |
1805 | if (old_buffer != bufp->buffer) \ | |
1806 | { \ | |
4bb91c68 SM |
1807 | int incr = bufp->buffer - old_buffer; \ |
1808 | MOVE_BUFFER_POINTER (b); \ | |
1809 | MOVE_BUFFER_POINTER (begalt); \ | |
25fe55af | 1810 | if (fixup_alt_jump) \ |
4bb91c68 | 1811 | MOVE_BUFFER_POINTER (fixup_alt_jump); \ |
25fe55af | 1812 | if (laststart) \ |
4bb91c68 | 1813 | MOVE_BUFFER_POINTER (laststart); \ |
25fe55af | 1814 | if (pending_exact) \ |
4bb91c68 | 1815 | MOVE_BUFFER_POINTER (pending_exact); \ |
fa9a63c5 | 1816 | } \ |
4bb91c68 | 1817 | ELSE_EXTEND_BUFFER_HIGH_BOUND \ |
fa9a63c5 RM |
1818 | } while (0) |
1819 | ||
1820 | ||
1821 | /* Since we have one byte reserved for the register number argument to | |
1822 | {start,stop}_memory, the maximum number of groups we can report | |
1823 | things about is what fits in that byte. */ | |
1824 | #define MAX_REGNUM 255 | |
1825 | ||
1826 | /* But patterns can have more than `MAX_REGNUM' registers. We just | |
1827 | ignore the excess. */ | |
098d42af | 1828 | typedef int regnum_t; |
fa9a63c5 RM |
1829 | |
1830 | ||
1831 | /* Macros for the compile stack. */ | |
1832 | ||
1833 | /* Since offsets can go either forwards or backwards, this type needs to | |
4bb91c68 SM |
1834 | be able to hold values from -(MAX_BUF_SIZE - 1) to MAX_BUF_SIZE - 1. */ |
1835 | /* int may be not enough when sizeof(int) == 2. */ | |
1836 | typedef long pattern_offset_t; | |
fa9a63c5 RM |
1837 | |
1838 | typedef struct | |
1839 | { | |
1840 | pattern_offset_t begalt_offset; | |
1841 | pattern_offset_t fixup_alt_jump; | |
5e69f11e | 1842 | pattern_offset_t laststart_offset; |
fa9a63c5 RM |
1843 | regnum_t regnum; |
1844 | } compile_stack_elt_t; | |
1845 | ||
1846 | ||
1847 | typedef struct | |
1848 | { | |
1849 | compile_stack_elt_t *stack; | |
1850 | unsigned size; | |
1851 | unsigned avail; /* Offset of next open position. */ | |
1852 | } compile_stack_type; | |
1853 | ||
1854 | ||
1855 | #define INIT_COMPILE_STACK_SIZE 32 | |
1856 | ||
1857 | #define COMPILE_STACK_EMPTY (compile_stack.avail == 0) | |
1858 | #define COMPILE_STACK_FULL (compile_stack.avail == compile_stack.size) | |
1859 | ||
4bb91c68 | 1860 | /* The next available element. */ |
fa9a63c5 RM |
1861 | #define COMPILE_STACK_TOP (compile_stack.stack[compile_stack.avail]) |
1862 | ||
77d11aec RS |
1863 | /* Explicit quit checking is only used on NTemacs. */ |
1864 | #if defined WINDOWSNT && defined emacs && defined QUIT | |
1865 | extern int immediate_quit; | |
1866 | # define IMMEDIATE_QUIT_CHECK \ | |
1867 | do { \ | |
1868 | if (immediate_quit) QUIT; \ | |
1869 | } while (0) | |
1870 | #else | |
1871 | # define IMMEDIATE_QUIT_CHECK ((void)0) | |
1872 | #endif | |
1873 | \f | |
b18215fc RS |
1874 | /* Structure to manage work area for range table. */ |
1875 | struct range_table_work_area | |
1876 | { | |
1877 | int *table; /* actual work area. */ | |
1878 | int allocated; /* allocated size for work area in bytes. */ | |
25fe55af | 1879 | int used; /* actually used size in words. */ |
96cc36cc | 1880 | int bits; /* flag to record character classes */ |
b18215fc RS |
1881 | }; |
1882 | ||
77d11aec RS |
1883 | /* Make sure that WORK_AREA can hold more N multibyte characters. |
1884 | This is used only in set_image_of_range and set_image_of_range_1. | |
1885 | It expects WORK_AREA to be a pointer. | |
1886 | If it can't get the space, it returns from the surrounding function. */ | |
1887 | ||
1888 | #define EXTEND_RANGE_TABLE(work_area, n) \ | |
1889 | do { \ | |
1890 | if (((work_area)->used + (n)) * sizeof (int) > (work_area)->allocated) \ | |
1891 | { \ | |
1892 | extend_range_table_work_area (work_area); \ | |
1893 | if ((work_area)->table == 0) \ | |
1894 | return (REG_ESPACE); \ | |
1895 | } \ | |
b18215fc RS |
1896 | } while (0) |
1897 | ||
96cc36cc RS |
1898 | #define SET_RANGE_TABLE_WORK_AREA_BIT(work_area, bit) \ |
1899 | (work_area).bits |= (bit) | |
1900 | ||
14473664 SM |
1901 | /* Bits used to implement the multibyte-part of the various character classes |
1902 | such as [:alnum:] in a charset's range table. */ | |
1903 | #define BIT_WORD 0x1 | |
1904 | #define BIT_LOWER 0x2 | |
1905 | #define BIT_PUNCT 0x4 | |
1906 | #define BIT_SPACE 0x8 | |
1907 | #define BIT_UPPER 0x10 | |
1908 | #define BIT_MULTIBYTE 0x20 | |
96cc36cc | 1909 | |
36595814 SM |
1910 | /* Set a range START..END to WORK_AREA. |
1911 | The range is passed through TRANSLATE, so START and END | |
1912 | should be untranslated. */ | |
77d11aec RS |
1913 | #define SET_RANGE_TABLE_WORK_AREA(work_area, start, end) \ |
1914 | do { \ | |
1915 | int tem; \ | |
1916 | tem = set_image_of_range (&work_area, start, end, translate); \ | |
1917 | if (tem > 0) \ | |
1918 | FREE_STACK_RETURN (tem); \ | |
b18215fc RS |
1919 | } while (0) |
1920 | ||
25fe55af | 1921 | /* Free allocated memory for WORK_AREA. */ |
b18215fc RS |
1922 | #define FREE_RANGE_TABLE_WORK_AREA(work_area) \ |
1923 | do { \ | |
1924 | if ((work_area).table) \ | |
1925 | free ((work_area).table); \ | |
1926 | } while (0) | |
1927 | ||
96cc36cc | 1928 | #define CLEAR_RANGE_TABLE_WORK_USED(work_area) ((work_area).used = 0, (work_area).bits = 0) |
b18215fc | 1929 | #define RANGE_TABLE_WORK_USED(work_area) ((work_area).used) |
96cc36cc | 1930 | #define RANGE_TABLE_WORK_BITS(work_area) ((work_area).bits) |
b18215fc | 1931 | #define RANGE_TABLE_WORK_ELT(work_area, i) ((work_area).table[i]) |
77d11aec | 1932 | \f |
b18215fc | 1933 | |
fa9a63c5 | 1934 | /* Set the bit for character C in a list. */ |
01618498 | 1935 | #define SET_LIST_BIT(c) (b[((c)) / BYTEWIDTH] |= 1 << ((c) % BYTEWIDTH)) |
fa9a63c5 RM |
1936 | |
1937 | ||
1938 | /* Get the next unsigned number in the uncompiled pattern. */ | |
25fe55af | 1939 | #define GET_UNSIGNED_NUMBER(num) \ |
99633e97 | 1940 | do { if (p != pend) \ |
fa9a63c5 | 1941 | { \ |
25fe55af | 1942 | PATFETCH (c); \ |
060cc8ff RS |
1943 | if (c == ' ') \ |
1944 | FREE_STACK_RETURN (REG_BADBR); \ | |
4bb91c68 | 1945 | while ('0' <= c && c <= '9') \ |
25fe55af | 1946 | { \ |
f6df485f | 1947 | int prev; \ |
25fe55af | 1948 | if (num < 0) \ |
f6df485f RS |
1949 | num = 0; \ |
1950 | prev = num; \ | |
25fe55af | 1951 | num = num * 10 + c - '0'; \ |
f6df485f RS |
1952 | if (num / 10 != prev) \ |
1953 | FREE_STACK_RETURN (REG_BADBR); \ | |
25fe55af | 1954 | if (p == pend) \ |
f6df485f | 1955 | break; \ |
25fe55af RS |
1956 | PATFETCH (c); \ |
1957 | } \ | |
060cc8ff RS |
1958 | if (c == ' ') \ |
1959 | FREE_STACK_RETURN (REG_BADBR); \ | |
25fe55af | 1960 | } \ |
99633e97 | 1961 | } while (0) |
77d11aec | 1962 | \f |
a0ad02f7 | 1963 | #if WIDE_CHAR_SUPPORT |
14473664 SM |
1964 | /* The GNU C library provides support for user-defined character classes |
1965 | and the functions from ISO C amendement 1. */ | |
1966 | # ifdef CHARCLASS_NAME_MAX | |
1967 | # define CHAR_CLASS_MAX_LENGTH CHARCLASS_NAME_MAX | |
1968 | # else | |
1969 | /* This shouldn't happen but some implementation might still have this | |
1970 | problem. Use a reasonable default value. */ | |
1971 | # define CHAR_CLASS_MAX_LENGTH 256 | |
1972 | # endif | |
1973 | typedef wctype_t re_wctype_t; | |
01618498 | 1974 | typedef wchar_t re_wchar_t; |
14473664 SM |
1975 | # define re_wctype wctype |
1976 | # define re_iswctype iswctype | |
1977 | # define re_wctype_to_bit(cc) 0 | |
1978 | #else | |
1979 | # define CHAR_CLASS_MAX_LENGTH 9 /* Namely, `multibyte'. */ | |
1980 | # define btowc(c) c | |
1981 | ||
01618498 | 1982 | /* Character classes. */ |
14473664 SM |
1983 | typedef enum { RECC_ERROR = 0, |
1984 | RECC_ALNUM, RECC_ALPHA, RECC_WORD, | |
1985 | RECC_GRAPH, RECC_PRINT, | |
1986 | RECC_LOWER, RECC_UPPER, | |
1987 | RECC_PUNCT, RECC_CNTRL, | |
1988 | RECC_DIGIT, RECC_XDIGIT, | |
1989 | RECC_BLANK, RECC_SPACE, | |
1990 | RECC_MULTIBYTE, RECC_NONASCII, | |
1991 | RECC_ASCII, RECC_UNIBYTE | |
1992 | } re_wctype_t; | |
1993 | ||
01618498 SM |
1994 | typedef int re_wchar_t; |
1995 | ||
14473664 SM |
1996 | /* Map a string to the char class it names (if any). */ |
1997 | static re_wctype_t | |
ada30c0e SM |
1998 | re_wctype (str) |
1999 | re_char *str; | |
14473664 | 2000 | { |
ada30c0e | 2001 | const char *string = str; |
14473664 SM |
2002 | if (STREQ (string, "alnum")) return RECC_ALNUM; |
2003 | else if (STREQ (string, "alpha")) return RECC_ALPHA; | |
2004 | else if (STREQ (string, "word")) return RECC_WORD; | |
2005 | else if (STREQ (string, "ascii")) return RECC_ASCII; | |
2006 | else if (STREQ (string, "nonascii")) return RECC_NONASCII; | |
2007 | else if (STREQ (string, "graph")) return RECC_GRAPH; | |
2008 | else if (STREQ (string, "lower")) return RECC_LOWER; | |
2009 | else if (STREQ (string, "print")) return RECC_PRINT; | |
2010 | else if (STREQ (string, "punct")) return RECC_PUNCT; | |
2011 | else if (STREQ (string, "space")) return RECC_SPACE; | |
2012 | else if (STREQ (string, "upper")) return RECC_UPPER; | |
2013 | else if (STREQ (string, "unibyte")) return RECC_UNIBYTE; | |
2014 | else if (STREQ (string, "multibyte")) return RECC_MULTIBYTE; | |
2015 | else if (STREQ (string, "digit")) return RECC_DIGIT; | |
2016 | else if (STREQ (string, "xdigit")) return RECC_XDIGIT; | |
2017 | else if (STREQ (string, "cntrl")) return RECC_CNTRL; | |
2018 | else if (STREQ (string, "blank")) return RECC_BLANK; | |
2019 | else return 0; | |
2020 | } | |
2021 | ||
2022 | /* True iff CH is in the char class CC. */ | |
2023 | static boolean | |
2024 | re_iswctype (ch, cc) | |
2025 | int ch; | |
2026 | re_wctype_t cc; | |
2027 | { | |
2028 | switch (cc) | |
2029 | { | |
0cdd06f8 SM |
2030 | case RECC_ALNUM: return ISALNUM (ch); |
2031 | case RECC_ALPHA: return ISALPHA (ch); | |
2032 | case RECC_BLANK: return ISBLANK (ch); | |
2033 | case RECC_CNTRL: return ISCNTRL (ch); | |
2034 | case RECC_DIGIT: return ISDIGIT (ch); | |
2035 | case RECC_GRAPH: return ISGRAPH (ch); | |
2036 | case RECC_LOWER: return ISLOWER (ch); | |
2037 | case RECC_PRINT: return ISPRINT (ch); | |
2038 | case RECC_PUNCT: return ISPUNCT (ch); | |
2039 | case RECC_SPACE: return ISSPACE (ch); | |
2040 | case RECC_UPPER: return ISUPPER (ch); | |
2041 | case RECC_XDIGIT: return ISXDIGIT (ch); | |
2042 | case RECC_ASCII: return IS_REAL_ASCII (ch); | |
2043 | case RECC_NONASCII: return !IS_REAL_ASCII (ch); | |
2044 | case RECC_UNIBYTE: return ISUNIBYTE (ch); | |
2045 | case RECC_MULTIBYTE: return !ISUNIBYTE (ch); | |
2046 | case RECC_WORD: return ISWORD (ch); | |
2047 | case RECC_ERROR: return false; | |
2048 | default: | |
2049 | abort(); | |
14473664 SM |
2050 | } |
2051 | } | |
fa9a63c5 | 2052 | |
14473664 SM |
2053 | /* Return a bit-pattern to use in the range-table bits to match multibyte |
2054 | chars of class CC. */ | |
2055 | static int | |
2056 | re_wctype_to_bit (cc) | |
2057 | re_wctype_t cc; | |
2058 | { | |
2059 | switch (cc) | |
2060 | { | |
2061 | case RECC_NONASCII: case RECC_PRINT: case RECC_GRAPH: | |
0cdd06f8 SM |
2062 | case RECC_MULTIBYTE: return BIT_MULTIBYTE; |
2063 | case RECC_ALPHA: case RECC_ALNUM: case RECC_WORD: return BIT_WORD; | |
2064 | case RECC_LOWER: return BIT_LOWER; | |
2065 | case RECC_UPPER: return BIT_UPPER; | |
2066 | case RECC_PUNCT: return BIT_PUNCT; | |
2067 | case RECC_SPACE: return BIT_SPACE; | |
14473664 | 2068 | case RECC_ASCII: case RECC_DIGIT: case RECC_XDIGIT: case RECC_CNTRL: |
0cdd06f8 SM |
2069 | case RECC_BLANK: case RECC_UNIBYTE: case RECC_ERROR: return 0; |
2070 | default: | |
2071 | abort(); | |
14473664 SM |
2072 | } |
2073 | } | |
2074 | #endif | |
77d11aec RS |
2075 | \f |
2076 | /* Filling in the work area of a range. */ | |
2077 | ||
2078 | /* Actually extend the space in WORK_AREA. */ | |
2079 | ||
2080 | static void | |
2081 | extend_range_table_work_area (work_area) | |
2082 | struct range_table_work_area *work_area; | |
177c0ea7 | 2083 | { |
77d11aec RS |
2084 | work_area->allocated += 16 * sizeof (int); |
2085 | if (work_area->table) | |
2086 | work_area->table | |
2087 | = (int *) realloc (work_area->table, work_area->allocated); | |
2088 | else | |
2089 | work_area->table | |
2090 | = (int *) malloc (work_area->allocated); | |
2091 | } | |
2092 | ||
2093 | #ifdef emacs | |
2094 | ||
2095 | /* Carefully find the ranges of codes that are equivalent | |
2096 | under case conversion to the range start..end when passed through | |
2097 | TRANSLATE. Handle the case where non-letters can come in between | |
2098 | two upper-case letters (which happens in Latin-1). | |
2099 | Also handle the case of groups of more than 2 case-equivalent chars. | |
2100 | ||
2101 | The basic method is to look at consecutive characters and see | |
2102 | if they can form a run that can be handled as one. | |
2103 | ||
2104 | Returns -1 if successful, REG_ESPACE if ran out of space. */ | |
2105 | ||
2106 | static int | |
2107 | set_image_of_range_1 (work_area, start, end, translate) | |
2108 | RE_TRANSLATE_TYPE translate; | |
2109 | struct range_table_work_area *work_area; | |
2110 | re_wchar_t start, end; | |
2111 | { | |
2112 | /* `one_case' indicates a character, or a run of characters, | |
2113 | each of which is an isolate (no case-equivalents). | |
2114 | This includes all ASCII non-letters. | |
2115 | ||
2116 | `two_case' indicates a character, or a run of characters, | |
2117 | each of which has two case-equivalent forms. | |
2118 | This includes all ASCII letters. | |
2119 | ||
2120 | `strange' indicates a character that has more than one | |
2121 | case-equivalent. */ | |
177c0ea7 | 2122 | |
77d11aec RS |
2123 | enum case_type {one_case, two_case, strange}; |
2124 | ||
2125 | /* Describe the run that is in progress, | |
2126 | which the next character can try to extend. | |
2127 | If run_type is strange, that means there really is no run. | |
2128 | If run_type is one_case, then run_start...run_end is the run. | |
2129 | If run_type is two_case, then the run is run_start...run_end, | |
2130 | and the case-equivalents end at run_eqv_end. */ | |
2131 | ||
2132 | enum case_type run_type = strange; | |
2133 | int run_start, run_end, run_eqv_end; | |
2134 | ||
2135 | Lisp_Object eqv_table; | |
2136 | ||
2137 | if (!RE_TRANSLATE_P (translate)) | |
2138 | { | |
b7c12565 | 2139 | EXTEND_RANGE_TABLE (work_area, 2); |
77d11aec RS |
2140 | work_area->table[work_area->used++] = (start); |
2141 | work_area->table[work_area->used++] = (end); | |
b7c12565 | 2142 | return -1; |
77d11aec RS |
2143 | } |
2144 | ||
2145 | eqv_table = XCHAR_TABLE (translate)->extras[2]; | |
99633e97 | 2146 | |
77d11aec RS |
2147 | for (; start <= end; start++) |
2148 | { | |
2149 | enum case_type this_type; | |
2150 | int eqv = RE_TRANSLATE (eqv_table, start); | |
2151 | int minchar, maxchar; | |
2152 | ||
2153 | /* Classify this character */ | |
2154 | if (eqv == start) | |
2155 | this_type = one_case; | |
2156 | else if (RE_TRANSLATE (eqv_table, eqv) == start) | |
2157 | this_type = two_case; | |
2158 | else | |
2159 | this_type = strange; | |
2160 | ||
2161 | if (start < eqv) | |
2162 | minchar = start, maxchar = eqv; | |
2163 | else | |
2164 | minchar = eqv, maxchar = start; | |
2165 | ||
2166 | /* Can this character extend the run in progress? */ | |
2167 | if (this_type == strange || this_type != run_type | |
2168 | || !(minchar == run_end + 1 | |
2169 | && (run_type == two_case | |
2170 | ? maxchar == run_eqv_end + 1 : 1))) | |
2171 | { | |
2172 | /* No, end the run. | |
2173 | Record each of its equivalent ranges. */ | |
2174 | if (run_type == one_case) | |
2175 | { | |
2176 | EXTEND_RANGE_TABLE (work_area, 2); | |
2177 | work_area->table[work_area->used++] = run_start; | |
2178 | work_area->table[work_area->used++] = run_end; | |
2179 | } | |
2180 | else if (run_type == two_case) | |
2181 | { | |
2182 | EXTEND_RANGE_TABLE (work_area, 4); | |
2183 | work_area->table[work_area->used++] = run_start; | |
2184 | work_area->table[work_area->used++] = run_end; | |
2185 | work_area->table[work_area->used++] | |
2186 | = RE_TRANSLATE (eqv_table, run_start); | |
2187 | work_area->table[work_area->used++] | |
2188 | = RE_TRANSLATE (eqv_table, run_end); | |
2189 | } | |
2190 | run_type = strange; | |
2191 | } | |
177c0ea7 | 2192 | |
77d11aec RS |
2193 | if (this_type == strange) |
2194 | { | |
2195 | /* For a strange character, add each of its equivalents, one | |
2196 | by one. Don't start a range. */ | |
2197 | do | |
2198 | { | |
2199 | EXTEND_RANGE_TABLE (work_area, 2); | |
2200 | work_area->table[work_area->used++] = eqv; | |
2201 | work_area->table[work_area->used++] = eqv; | |
2202 | eqv = RE_TRANSLATE (eqv_table, eqv); | |
2203 | } | |
2204 | while (eqv != start); | |
2205 | } | |
2206 | ||
2207 | /* Add this char to the run, or start a new run. */ | |
2208 | else if (run_type == strange) | |
2209 | { | |
2210 | /* Initialize a new range. */ | |
2211 | run_type = this_type; | |
2212 | run_start = start; | |
2213 | run_end = start; | |
2214 | run_eqv_end = RE_TRANSLATE (eqv_table, run_end); | |
2215 | } | |
2216 | else | |
2217 | { | |
2218 | /* Extend a running range. */ | |
2219 | run_end = minchar; | |
2220 | run_eqv_end = RE_TRANSLATE (eqv_table, run_end); | |
2221 | } | |
2222 | } | |
2223 | ||
2224 | /* If a run is still in progress at the end, finish it now | |
2225 | by recording its equivalent ranges. */ | |
2226 | if (run_type == one_case) | |
2227 | { | |
2228 | EXTEND_RANGE_TABLE (work_area, 2); | |
2229 | work_area->table[work_area->used++] = run_start; | |
2230 | work_area->table[work_area->used++] = run_end; | |
2231 | } | |
2232 | else if (run_type == two_case) | |
2233 | { | |
2234 | EXTEND_RANGE_TABLE (work_area, 4); | |
2235 | work_area->table[work_area->used++] = run_start; | |
2236 | work_area->table[work_area->used++] = run_end; | |
2237 | work_area->table[work_area->used++] | |
2238 | = RE_TRANSLATE (eqv_table, run_start); | |
2239 | work_area->table[work_area->used++] | |
2240 | = RE_TRANSLATE (eqv_table, run_end); | |
2241 | } | |
2242 | ||
2243 | return -1; | |
2244 | } | |
36595814 | 2245 | |
77d11aec | 2246 | #endif /* emacs */ |
36595814 | 2247 | |
b7c12565 | 2248 | /* Record the the image of the range start..end when passed through |
36595814 SM |
2249 | TRANSLATE. This is not necessarily TRANSLATE(start)..TRANSLATE(end) |
2250 | and is not even necessarily contiguous. | |
b7c12565 RS |
2251 | Normally we approximate it with the smallest contiguous range that contains |
2252 | all the chars we need. However, for Latin-1 we go to extra effort | |
2253 | to do a better job. | |
2254 | ||
2255 | This function is not called for ASCII ranges. | |
77d11aec RS |
2256 | |
2257 | Returns -1 if successful, REG_ESPACE if ran out of space. */ | |
2258 | ||
2259 | static int | |
36595814 SM |
2260 | set_image_of_range (work_area, start, end, translate) |
2261 | RE_TRANSLATE_TYPE translate; | |
2262 | struct range_table_work_area *work_area; | |
2263 | re_wchar_t start, end; | |
2264 | { | |
77d11aec RS |
2265 | re_wchar_t cmin, cmax; |
2266 | ||
2267 | #ifdef emacs | |
2268 | /* For Latin-1 ranges, use set_image_of_range_1 | |
2269 | to get proper handling of ranges that include letters and nonletters. | |
b7c12565 | 2270 | For a range that includes the whole of Latin-1, this is not necessary. |
77d11aec | 2271 | For other character sets, we don't bother to get this right. */ |
b7c12565 RS |
2272 | if (RE_TRANSLATE_P (translate) && start < 04400 |
2273 | && !(start < 04200 && end >= 04377)) | |
77d11aec | 2274 | { |
b7c12565 | 2275 | int newend; |
77d11aec | 2276 | int tem; |
b7c12565 RS |
2277 | newend = end; |
2278 | if (newend > 04377) | |
2279 | newend = 04377; | |
2280 | tem = set_image_of_range_1 (work_area, start, newend, translate); | |
77d11aec RS |
2281 | if (tem > 0) |
2282 | return tem; | |
2283 | ||
2284 | start = 04400; | |
2285 | if (end < 04400) | |
2286 | return -1; | |
2287 | } | |
2288 | #endif | |
2289 | ||
b7c12565 RS |
2290 | EXTEND_RANGE_TABLE (work_area, 2); |
2291 | work_area->table[work_area->used++] = (start); | |
2292 | work_area->table[work_area->used++] = (end); | |
2293 | ||
2294 | cmin = -1, cmax = -1; | |
77d11aec | 2295 | |
36595814 | 2296 | if (RE_TRANSLATE_P (translate)) |
b7c12565 RS |
2297 | { |
2298 | int ch; | |
77d11aec | 2299 | |
b7c12565 RS |
2300 | for (ch = start; ch <= end; ch++) |
2301 | { | |
2302 | re_wchar_t c = TRANSLATE (ch); | |
2303 | if (! (start <= c && c <= end)) | |
2304 | { | |
2305 | if (cmin == -1) | |
2306 | cmin = c, cmax = c; | |
2307 | else | |
2308 | { | |
2309 | cmin = MIN (cmin, c); | |
2310 | cmax = MAX (cmax, c); | |
2311 | } | |
2312 | } | |
2313 | } | |
2314 | ||
2315 | if (cmin != -1) | |
2316 | { | |
2317 | EXTEND_RANGE_TABLE (work_area, 2); | |
2318 | work_area->table[work_area->used++] = (cmin); | |
2319 | work_area->table[work_area->used++] = (cmax); | |
2320 | } | |
2321 | } | |
36595814 | 2322 | |
77d11aec RS |
2323 | return -1; |
2324 | } | |
fa9a63c5 RM |
2325 | \f |
2326 | #ifndef MATCH_MAY_ALLOCATE | |
2327 | ||
2328 | /* If we cannot allocate large objects within re_match_2_internal, | |
2329 | we make the fail stack and register vectors global. | |
2330 | The fail stack, we grow to the maximum size when a regexp | |
2331 | is compiled. | |
2332 | The register vectors, we adjust in size each time we | |
2333 | compile a regexp, according to the number of registers it needs. */ | |
2334 | ||
2335 | static fail_stack_type fail_stack; | |
2336 | ||
2337 | /* Size with which the following vectors are currently allocated. | |
2338 | That is so we can make them bigger as needed, | |
4bb91c68 | 2339 | but never make them smaller. */ |
fa9a63c5 RM |
2340 | static int regs_allocated_size; |
2341 | ||
66f0296e SM |
2342 | static re_char ** regstart, ** regend; |
2343 | static re_char **best_regstart, **best_regend; | |
fa9a63c5 RM |
2344 | |
2345 | /* Make the register vectors big enough for NUM_REGS registers, | |
4bb91c68 | 2346 | but don't make them smaller. */ |
fa9a63c5 RM |
2347 | |
2348 | static | |
2349 | regex_grow_registers (num_regs) | |
2350 | int num_regs; | |
2351 | { | |
2352 | if (num_regs > regs_allocated_size) | |
2353 | { | |
66f0296e SM |
2354 | RETALLOC_IF (regstart, num_regs, re_char *); |
2355 | RETALLOC_IF (regend, num_regs, re_char *); | |
2356 | RETALLOC_IF (best_regstart, num_regs, re_char *); | |
2357 | RETALLOC_IF (best_regend, num_regs, re_char *); | |
fa9a63c5 RM |
2358 | |
2359 | regs_allocated_size = num_regs; | |
2360 | } | |
2361 | } | |
2362 | ||
2363 | #endif /* not MATCH_MAY_ALLOCATE */ | |
2364 | \f | |
99633e97 SM |
2365 | static boolean group_in_compile_stack _RE_ARGS ((compile_stack_type |
2366 | compile_stack, | |
2367 | regnum_t regnum)); | |
2368 | ||
fa9a63c5 RM |
2369 | /* `regex_compile' compiles PATTERN (of length SIZE) according to SYNTAX. |
2370 | Returns one of error codes defined in `regex.h', or zero for success. | |
2371 | ||
2372 | Assumes the `allocated' (and perhaps `buffer') and `translate' | |
2373 | fields are set in BUFP on entry. | |
2374 | ||
2375 | If it succeeds, results are put in BUFP (if it returns an error, the | |
2376 | contents of BUFP are undefined): | |
2377 | `buffer' is the compiled pattern; | |
2378 | `syntax' is set to SYNTAX; | |
2379 | `used' is set to the length of the compiled pattern; | |
2380 | `fastmap_accurate' is zero; | |
2381 | `re_nsub' is the number of subexpressions in PATTERN; | |
2382 | `not_bol' and `not_eol' are zero; | |
5e69f11e | 2383 | |
c0f9ea08 | 2384 | The `fastmap' field is neither examined nor set. */ |
fa9a63c5 | 2385 | |
505bde11 SM |
2386 | /* Insert the `jump' from the end of last alternative to "here". |
2387 | The space for the jump has already been allocated. */ | |
2388 | #define FIXUP_ALT_JUMP() \ | |
2389 | do { \ | |
2390 | if (fixup_alt_jump) \ | |
2391 | STORE_JUMP (jump, fixup_alt_jump, b); \ | |
2392 | } while (0) | |
2393 | ||
2394 | ||
fa9a63c5 RM |
2395 | /* Return, freeing storage we allocated. */ |
2396 | #define FREE_STACK_RETURN(value) \ | |
b18215fc RS |
2397 | do { \ |
2398 | FREE_RANGE_TABLE_WORK_AREA (range_table_work); \ | |
2399 | free (compile_stack.stack); \ | |
2400 | return value; \ | |
2401 | } while (0) | |
fa9a63c5 RM |
2402 | |
2403 | static reg_errcode_t | |
2404 | regex_compile (pattern, size, syntax, bufp) | |
66f0296e | 2405 | re_char *pattern; |
4bb91c68 | 2406 | size_t size; |
fa9a63c5 RM |
2407 | reg_syntax_t syntax; |
2408 | struct re_pattern_buffer *bufp; | |
2409 | { | |
01618498 SM |
2410 | /* We fetch characters from PATTERN here. */ |
2411 | register re_wchar_t c, c1; | |
5e69f11e | 2412 | |
fa9a63c5 | 2413 | /* A random temporary spot in PATTERN. */ |
66f0296e | 2414 | re_char *p1; |
fa9a63c5 RM |
2415 | |
2416 | /* Points to the end of the buffer, where we should append. */ | |
2417 | register unsigned char *b; | |
5e69f11e | 2418 | |
fa9a63c5 RM |
2419 | /* Keeps track of unclosed groups. */ |
2420 | compile_stack_type compile_stack; | |
2421 | ||
2422 | /* Points to the current (ending) position in the pattern. */ | |
22336245 RS |
2423 | #ifdef AIX |
2424 | /* `const' makes AIX compiler fail. */ | |
66f0296e | 2425 | unsigned char *p = pattern; |
22336245 | 2426 | #else |
66f0296e | 2427 | re_char *p = pattern; |
22336245 | 2428 | #endif |
66f0296e | 2429 | re_char *pend = pattern + size; |
5e69f11e | 2430 | |
fa9a63c5 | 2431 | /* How to translate the characters in the pattern. */ |
6676cb1c | 2432 | RE_TRANSLATE_TYPE translate = bufp->translate; |
fa9a63c5 RM |
2433 | |
2434 | /* Address of the count-byte of the most recently inserted `exactn' | |
2435 | command. This makes it possible to tell if a new exact-match | |
2436 | character can be added to that command or if the character requires | |
2437 | a new `exactn' command. */ | |
2438 | unsigned char *pending_exact = 0; | |
2439 | ||
2440 | /* Address of start of the most recently finished expression. | |
2441 | This tells, e.g., postfix * where to find the start of its | |
2442 | operand. Reset at the beginning of groups and alternatives. */ | |
2443 | unsigned char *laststart = 0; | |
2444 | ||
2445 | /* Address of beginning of regexp, or inside of last group. */ | |
2446 | unsigned char *begalt; | |
2447 | ||
2448 | /* Place in the uncompiled pattern (i.e., the {) to | |
2449 | which to go back if the interval is invalid. */ | |
66f0296e | 2450 | re_char *beg_interval; |
5e69f11e | 2451 | |
fa9a63c5 | 2452 | /* Address of the place where a forward jump should go to the end of |
25fe55af | 2453 | the containing expression. Each alternative of an `or' -- except the |
fa9a63c5 RM |
2454 | last -- ends with a forward jump of this sort. */ |
2455 | unsigned char *fixup_alt_jump = 0; | |
2456 | ||
2457 | /* Counts open-groups as they are encountered. Remembered for the | |
2458 | matching close-group on the compile stack, so the same register | |
2459 | number is put in the stop_memory as the start_memory. */ | |
2460 | regnum_t regnum = 0; | |
2461 | ||
b18215fc RS |
2462 | /* Work area for range table of charset. */ |
2463 | struct range_table_work_area range_table_work; | |
2464 | ||
2d1675e4 SM |
2465 | /* If the object matched can contain multibyte characters. */ |
2466 | const boolean multibyte = RE_MULTIBYTE_P (bufp); | |
2467 | ||
fa9a63c5 | 2468 | #ifdef DEBUG |
99633e97 | 2469 | debug++; |
fa9a63c5 | 2470 | DEBUG_PRINT1 ("\nCompiling pattern: "); |
99633e97 | 2471 | if (debug > 0) |
fa9a63c5 RM |
2472 | { |
2473 | unsigned debug_count; | |
5e69f11e | 2474 | |
fa9a63c5 | 2475 | for (debug_count = 0; debug_count < size; debug_count++) |
25fe55af | 2476 | putchar (pattern[debug_count]); |
fa9a63c5 RM |
2477 | putchar ('\n'); |
2478 | } | |
2479 | #endif /* DEBUG */ | |
2480 | ||
2481 | /* Initialize the compile stack. */ | |
2482 | compile_stack.stack = TALLOC (INIT_COMPILE_STACK_SIZE, compile_stack_elt_t); | |
2483 | if (compile_stack.stack == NULL) | |
2484 | return REG_ESPACE; | |
2485 | ||
2486 | compile_stack.size = INIT_COMPILE_STACK_SIZE; | |
2487 | compile_stack.avail = 0; | |
2488 | ||
b18215fc RS |
2489 | range_table_work.table = 0; |
2490 | range_table_work.allocated = 0; | |
2491 | ||
fa9a63c5 RM |
2492 | /* Initialize the pattern buffer. */ |
2493 | bufp->syntax = syntax; | |
2494 | bufp->fastmap_accurate = 0; | |
2495 | bufp->not_bol = bufp->not_eol = 0; | |
2496 | ||
2497 | /* Set `used' to zero, so that if we return an error, the pattern | |
2498 | printer (for debugging) will think there's no pattern. We reset it | |
2499 | at the end. */ | |
2500 | bufp->used = 0; | |
5e69f11e | 2501 | |
fa9a63c5 | 2502 | /* Always count groups, whether or not bufp->no_sub is set. */ |
5e69f11e | 2503 | bufp->re_nsub = 0; |
fa9a63c5 | 2504 | |
0b32bf0e | 2505 | #if !defined emacs && !defined SYNTAX_TABLE |
fa9a63c5 RM |
2506 | /* Initialize the syntax table. */ |
2507 | init_syntax_once (); | |
2508 | #endif | |
2509 | ||
2510 | if (bufp->allocated == 0) | |
2511 | { | |
2512 | if (bufp->buffer) | |
2513 | { /* If zero allocated, but buffer is non-null, try to realloc | |
25fe55af RS |
2514 | enough space. This loses if buffer's address is bogus, but |
2515 | that is the user's responsibility. */ | |
2516 | RETALLOC (bufp->buffer, INIT_BUF_SIZE, unsigned char); | |
2517 | } | |
fa9a63c5 | 2518 | else |
25fe55af RS |
2519 | { /* Caller did not allocate a buffer. Do it for them. */ |
2520 | bufp->buffer = TALLOC (INIT_BUF_SIZE, unsigned char); | |
2521 | } | |
fa9a63c5 RM |
2522 | if (!bufp->buffer) FREE_STACK_RETURN (REG_ESPACE); |
2523 | ||
2524 | bufp->allocated = INIT_BUF_SIZE; | |
2525 | } | |
2526 | ||
2527 | begalt = b = bufp->buffer; | |
2528 | ||
2529 | /* Loop through the uncompiled pattern until we're at the end. */ | |
2530 | while (p != pend) | |
2531 | { | |
2532 | PATFETCH (c); | |
2533 | ||
2534 | switch (c) | |
25fe55af RS |
2535 | { |
2536 | case '^': | |
2537 | { | |
2538 | if ( /* If at start of pattern, it's an operator. */ | |
2539 | p == pattern + 1 | |
2540 | /* If context independent, it's an operator. */ | |
2541 | || syntax & RE_CONTEXT_INDEP_ANCHORS | |
2542 | /* Otherwise, depends on what's come before. */ | |
2543 | || at_begline_loc_p (pattern, p, syntax)) | |
c0f9ea08 | 2544 | BUF_PUSH ((syntax & RE_NO_NEWLINE_ANCHOR) ? begbuf : begline); |
25fe55af RS |
2545 | else |
2546 | goto normal_char; | |
2547 | } | |
2548 | break; | |
2549 | ||
2550 | ||
2551 | case '$': | |
2552 | { | |
2553 | if ( /* If at end of pattern, it's an operator. */ | |
2554 | p == pend | |
2555 | /* If context independent, it's an operator. */ | |
2556 | || syntax & RE_CONTEXT_INDEP_ANCHORS | |
2557 | /* Otherwise, depends on what's next. */ | |
2558 | || at_endline_loc_p (p, pend, syntax)) | |
c0f9ea08 | 2559 | BUF_PUSH ((syntax & RE_NO_NEWLINE_ANCHOR) ? endbuf : endline); |
25fe55af RS |
2560 | else |
2561 | goto normal_char; | |
2562 | } | |
2563 | break; | |
fa9a63c5 RM |
2564 | |
2565 | ||
2566 | case '+': | |
25fe55af RS |
2567 | case '?': |
2568 | if ((syntax & RE_BK_PLUS_QM) | |
2569 | || (syntax & RE_LIMITED_OPS)) | |
2570 | goto normal_char; | |
2571 | handle_plus: | |
2572 | case '*': | |
2573 | /* If there is no previous pattern... */ | |
2574 | if (!laststart) | |
2575 | { | |
2576 | if (syntax & RE_CONTEXT_INVALID_OPS) | |
2577 | FREE_STACK_RETURN (REG_BADRPT); | |
2578 | else if (!(syntax & RE_CONTEXT_INDEP_OPS)) | |
2579 | goto normal_char; | |
2580 | } | |
2581 | ||
2582 | { | |
25fe55af | 2583 | /* 1 means zero (many) matches is allowed. */ |
66f0296e SM |
2584 | boolean zero_times_ok = 0, many_times_ok = 0; |
2585 | boolean greedy = 1; | |
25fe55af RS |
2586 | |
2587 | /* If there is a sequence of repetition chars, collapse it | |
2588 | down to just one (the right one). We can't combine | |
2589 | interval operators with these because of, e.g., `a{2}*', | |
2590 | which should only match an even number of `a's. */ | |
2591 | ||
2592 | for (;;) | |
2593 | { | |
0b32bf0e | 2594 | if ((syntax & RE_FRUGAL) |
1c8c6d39 DL |
2595 | && c == '?' && (zero_times_ok || many_times_ok)) |
2596 | greedy = 0; | |
2597 | else | |
2598 | { | |
2599 | zero_times_ok |= c != '+'; | |
2600 | many_times_ok |= c != '?'; | |
2601 | } | |
25fe55af RS |
2602 | |
2603 | if (p == pend) | |
2604 | break; | |
ed0767d8 SM |
2605 | else if (*p == '*' |
2606 | || (!(syntax & RE_BK_PLUS_QM) | |
2607 | && (*p == '+' || *p == '?'))) | |
25fe55af | 2608 | ; |
ed0767d8 | 2609 | else if (syntax & RE_BK_PLUS_QM && *p == '\\') |
25fe55af | 2610 | { |
ed0767d8 SM |
2611 | if (p+1 == pend) |
2612 | FREE_STACK_RETURN (REG_EESCAPE); | |
2613 | if (p[1] == '+' || p[1] == '?') | |
2614 | PATFETCH (c); /* Gobble up the backslash. */ | |
2615 | else | |
2616 | break; | |
25fe55af RS |
2617 | } |
2618 | else | |
ed0767d8 | 2619 | break; |
25fe55af | 2620 | /* If we get here, we found another repeat character. */ |
ed0767d8 SM |
2621 | PATFETCH (c); |
2622 | } | |
25fe55af RS |
2623 | |
2624 | /* Star, etc. applied to an empty pattern is equivalent | |
2625 | to an empty pattern. */ | |
4e8a9132 | 2626 | if (!laststart || laststart == b) |
25fe55af RS |
2627 | break; |
2628 | ||
2629 | /* Now we know whether or not zero matches is allowed | |
2630 | and also whether or not two or more matches is allowed. */ | |
1c8c6d39 DL |
2631 | if (greedy) |
2632 | { | |
99633e97 | 2633 | if (many_times_ok) |
4e8a9132 SM |
2634 | { |
2635 | boolean simple = skip_one_char (laststart) == b; | |
2636 | unsigned int startoffset = 0; | |
f6a3f532 | 2637 | re_opcode_t ofj = |
01618498 | 2638 | /* Check if the loop can match the empty string. */ |
6df42991 SM |
2639 | (simple || !analyse_first (laststart, b, NULL, 0)) |
2640 | ? on_failure_jump : on_failure_jump_loop; | |
4e8a9132 | 2641 | assert (skip_one_char (laststart) <= b); |
177c0ea7 | 2642 | |
4e8a9132 SM |
2643 | if (!zero_times_ok && simple) |
2644 | { /* Since simple * loops can be made faster by using | |
2645 | on_failure_keep_string_jump, we turn simple P+ | |
2646 | into PP* if P is simple. */ | |
2647 | unsigned char *p1, *p2; | |
2648 | startoffset = b - laststart; | |
2649 | GET_BUFFER_SPACE (startoffset); | |
2650 | p1 = b; p2 = laststart; | |
2651 | while (p2 < p1) | |
2652 | *b++ = *p2++; | |
2653 | zero_times_ok = 1; | |
99633e97 | 2654 | } |
4e8a9132 SM |
2655 | |
2656 | GET_BUFFER_SPACE (6); | |
2657 | if (!zero_times_ok) | |
2658 | /* A + loop. */ | |
f6a3f532 | 2659 | STORE_JUMP (ofj, b, b + 6); |
99633e97 | 2660 | else |
4e8a9132 SM |
2661 | /* Simple * loops can use on_failure_keep_string_jump |
2662 | depending on what follows. But since we don't know | |
2663 | that yet, we leave the decision up to | |
2664 | on_failure_jump_smart. */ | |
f6a3f532 | 2665 | INSERT_JUMP (simple ? on_failure_jump_smart : ofj, |
4e8a9132 | 2666 | laststart + startoffset, b + 6); |
99633e97 | 2667 | b += 3; |
4e8a9132 | 2668 | STORE_JUMP (jump, b, laststart + startoffset); |
99633e97 SM |
2669 | b += 3; |
2670 | } | |
2671 | else | |
2672 | { | |
4e8a9132 SM |
2673 | /* A simple ? pattern. */ |
2674 | assert (zero_times_ok); | |
2675 | GET_BUFFER_SPACE (3); | |
2676 | INSERT_JUMP (on_failure_jump, laststart, b + 3); | |
99633e97 SM |
2677 | b += 3; |
2678 | } | |
1c8c6d39 DL |
2679 | } |
2680 | else /* not greedy */ | |
2681 | { /* I wish the greedy and non-greedy cases could be merged. */ | |
2682 | ||
0683b6fa | 2683 | GET_BUFFER_SPACE (7); /* We might use less. */ |
1c8c6d39 DL |
2684 | if (many_times_ok) |
2685 | { | |
f6a3f532 SM |
2686 | boolean emptyp = analyse_first (laststart, b, NULL, 0); |
2687 | ||
6df42991 SM |
2688 | /* The non-greedy multiple match looks like |
2689 | a repeat..until: we only need a conditional jump | |
2690 | at the end of the loop. */ | |
f6a3f532 SM |
2691 | if (emptyp) BUF_PUSH (no_op); |
2692 | STORE_JUMP (emptyp ? on_failure_jump_nastyloop | |
2693 | : on_failure_jump, b, laststart); | |
1c8c6d39 DL |
2694 | b += 3; |
2695 | if (zero_times_ok) | |
2696 | { | |
2697 | /* The repeat...until naturally matches one or more. | |
2698 | To also match zero times, we need to first jump to | |
6df42991 | 2699 | the end of the loop (its conditional jump). */ |
1c8c6d39 DL |
2700 | INSERT_JUMP (jump, laststart, b); |
2701 | b += 3; | |
2702 | } | |
2703 | } | |
2704 | else | |
2705 | { | |
2706 | /* non-greedy a?? */ | |
1c8c6d39 DL |
2707 | INSERT_JUMP (jump, laststart, b + 3); |
2708 | b += 3; | |
2709 | INSERT_JUMP (on_failure_jump, laststart, laststart + 6); | |
2710 | b += 3; | |
2711 | } | |
2712 | } | |
2713 | } | |
4e8a9132 | 2714 | pending_exact = 0; |
fa9a63c5 RM |
2715 | break; |
2716 | ||
2717 | ||
2718 | case '.': | |
25fe55af RS |
2719 | laststart = b; |
2720 | BUF_PUSH (anychar); | |
2721 | break; | |
fa9a63c5 RM |
2722 | |
2723 | ||
25fe55af RS |
2724 | case '[': |
2725 | { | |
b18215fc | 2726 | CLEAR_RANGE_TABLE_WORK_USED (range_table_work); |
fa9a63c5 | 2727 | |
25fe55af | 2728 | if (p == pend) FREE_STACK_RETURN (REG_EBRACK); |
fa9a63c5 | 2729 | |
25fe55af RS |
2730 | /* Ensure that we have enough space to push a charset: the |
2731 | opcode, the length count, and the bitset; 34 bytes in all. */ | |
fa9a63c5 RM |
2732 | GET_BUFFER_SPACE (34); |
2733 | ||
25fe55af | 2734 | laststart = b; |
e318085a | 2735 | |
25fe55af RS |
2736 | /* We test `*p == '^' twice, instead of using an if |
2737 | statement, so we only need one BUF_PUSH. */ | |
2738 | BUF_PUSH (*p == '^' ? charset_not : charset); | |
2739 | if (*p == '^') | |
2740 | p++; | |
e318085a | 2741 | |
25fe55af RS |
2742 | /* Remember the first position in the bracket expression. */ |
2743 | p1 = p; | |
e318085a | 2744 | |
25fe55af RS |
2745 | /* Push the number of bytes in the bitmap. */ |
2746 | BUF_PUSH ((1 << BYTEWIDTH) / BYTEWIDTH); | |
e318085a | 2747 | |
25fe55af RS |
2748 | /* Clear the whole map. */ |
2749 | bzero (b, (1 << BYTEWIDTH) / BYTEWIDTH); | |
e318085a | 2750 | |
25fe55af RS |
2751 | /* charset_not matches newline according to a syntax bit. */ |
2752 | if ((re_opcode_t) b[-2] == charset_not | |
2753 | && (syntax & RE_HAT_LISTS_NOT_NEWLINE)) | |
2754 | SET_LIST_BIT ('\n'); | |
fa9a63c5 | 2755 | |
25fe55af RS |
2756 | /* Read in characters and ranges, setting map bits. */ |
2757 | for (;;) | |
2758 | { | |
b18215fc | 2759 | boolean escaped_char = false; |
2d1675e4 | 2760 | const unsigned char *p2 = p; |
e318085a | 2761 | |
25fe55af | 2762 | if (p == pend) FREE_STACK_RETURN (REG_EBRACK); |
e318085a | 2763 | |
36595814 SM |
2764 | /* Don't translate yet. The range TRANSLATE(X..Y) cannot |
2765 | always be determined from TRANSLATE(X) and TRANSLATE(Y) | |
2766 | So the translation is done later in a loop. Example: | |
2767 | (let ((case-fold-search t)) (string-match "[A-_]" "A")) */ | |
25fe55af | 2768 | PATFETCH (c); |
e318085a | 2769 | |
25fe55af RS |
2770 | /* \ might escape characters inside [...] and [^...]. */ |
2771 | if ((syntax & RE_BACKSLASH_ESCAPE_IN_LISTS) && c == '\\') | |
2772 | { | |
2773 | if (p == pend) FREE_STACK_RETURN (REG_EESCAPE); | |
e318085a RS |
2774 | |
2775 | PATFETCH (c); | |
b18215fc | 2776 | escaped_char = true; |
25fe55af | 2777 | } |
b18215fc RS |
2778 | else |
2779 | { | |
657fcfbd RS |
2780 | /* Could be the end of the bracket expression. If it's |
2781 | not (i.e., when the bracket expression is `[]' so | |
2782 | far), the ']' character bit gets set way below. */ | |
2d1675e4 | 2783 | if (c == ']' && p2 != p1) |
657fcfbd | 2784 | break; |
25fe55af | 2785 | } |
b18215fc | 2786 | |
b18215fc RS |
2787 | /* What should we do for the character which is |
2788 | greater than 0x7F, but not BASE_LEADING_CODE_P? | |
2789 | XXX */ | |
2790 | ||
25fe55af RS |
2791 | /* See if we're at the beginning of a possible character |
2792 | class. */ | |
b18215fc | 2793 | |
2d1675e4 SM |
2794 | if (!escaped_char && |
2795 | syntax & RE_CHAR_CLASSES && c == '[' && *p == ':') | |
657fcfbd RS |
2796 | { |
2797 | /* Leave room for the null. */ | |
14473664 | 2798 | unsigned char str[CHAR_CLASS_MAX_LENGTH + 1]; |
ed0767d8 | 2799 | const unsigned char *class_beg; |
b18215fc | 2800 | |
25fe55af RS |
2801 | PATFETCH (c); |
2802 | c1 = 0; | |
ed0767d8 | 2803 | class_beg = p; |
b18215fc | 2804 | |
25fe55af RS |
2805 | /* If pattern is `[[:'. */ |
2806 | if (p == pend) FREE_STACK_RETURN (REG_EBRACK); | |
b18215fc | 2807 | |
25fe55af RS |
2808 | for (;;) |
2809 | { | |
14473664 SM |
2810 | PATFETCH (c); |
2811 | if ((c == ':' && *p == ']') || p == pend) | |
2812 | break; | |
2813 | if (c1 < CHAR_CLASS_MAX_LENGTH) | |
2814 | str[c1++] = c; | |
2815 | else | |
2816 | /* This is in any case an invalid class name. */ | |
2817 | str[0] = '\0'; | |
25fe55af RS |
2818 | } |
2819 | str[c1] = '\0'; | |
b18215fc RS |
2820 | |
2821 | /* If isn't a word bracketed by `[:' and `:]': | |
2822 | undo the ending character, the letters, and | |
2823 | leave the leading `:' and `[' (but set bits for | |
2824 | them). */ | |
25fe55af RS |
2825 | if (c == ':' && *p == ']') |
2826 | { | |
36595814 | 2827 | re_wchar_t ch; |
14473664 SM |
2828 | re_wctype_t cc; |
2829 | ||
2830 | cc = re_wctype (str); | |
2831 | ||
2832 | if (cc == 0) | |
fa9a63c5 RM |
2833 | FREE_STACK_RETURN (REG_ECTYPE); |
2834 | ||
14473664 SM |
2835 | /* Throw away the ] at the end of the character |
2836 | class. */ | |
2837 | PATFETCH (c); | |
fa9a63c5 | 2838 | |
14473664 | 2839 | if (p == pend) FREE_STACK_RETURN (REG_EBRACK); |
fa9a63c5 | 2840 | |
96cc36cc RS |
2841 | /* Most character classes in a multibyte match |
2842 | just set a flag. Exceptions are is_blank, | |
2843 | is_digit, is_cntrl, and is_xdigit, since | |
2844 | they can only match ASCII characters. We | |
14473664 SM |
2845 | don't need to handle them for multibyte. |
2846 | They are distinguished by a negative wctype. */ | |
96cc36cc | 2847 | |
2d1675e4 | 2848 | if (multibyte) |
14473664 SM |
2849 | SET_RANGE_TABLE_WORK_AREA_BIT (range_table_work, |
2850 | re_wctype_to_bit (cc)); | |
96cc36cc | 2851 | |
14473664 | 2852 | for (ch = 0; ch < 1 << BYTEWIDTH; ++ch) |
25fe55af | 2853 | { |
7ae68633 | 2854 | int translated = TRANSLATE (ch); |
14473664 | 2855 | if (re_iswctype (btowc (ch), cc)) |
96cc36cc | 2856 | SET_LIST_BIT (translated); |
25fe55af | 2857 | } |
b18215fc RS |
2858 | |
2859 | /* Repeat the loop. */ | |
2860 | continue; | |
25fe55af RS |
2861 | } |
2862 | else | |
2863 | { | |
ed0767d8 SM |
2864 | /* Go back to right after the "[:". */ |
2865 | p = class_beg; | |
25fe55af | 2866 | SET_LIST_BIT ('['); |
b18215fc RS |
2867 | |
2868 | /* Because the `:' may starts the range, we | |
2869 | can't simply set bit and repeat the loop. | |
25fe55af | 2870 | Instead, just set it to C and handle below. */ |
b18215fc | 2871 | c = ':'; |
25fe55af RS |
2872 | } |
2873 | } | |
b18215fc RS |
2874 | |
2875 | if (p < pend && p[0] == '-' && p[1] != ']') | |
2876 | { | |
2877 | ||
2878 | /* Discard the `-'. */ | |
2879 | PATFETCH (c1); | |
2880 | ||
2881 | /* Fetch the character which ends the range. */ | |
2882 | PATFETCH (c1); | |
b18215fc | 2883 | |
b54f61ed | 2884 | if (SINGLE_BYTE_CHAR_P (c)) |
e934739e | 2885 | { |
b54f61ed KH |
2886 | if (! SINGLE_BYTE_CHAR_P (c1)) |
2887 | { | |
3ff2446d KH |
2888 | /* Handle a range starting with a |
2889 | character of less than 256, and ending | |
2890 | with a character of not less than 256. | |
2891 | Split that into two ranges, the low one | |
2892 | ending at 0377, and the high one | |
2893 | starting at the smallest character in | |
2894 | the charset of C1 and ending at C1. */ | |
b54f61ed | 2895 | int charset = CHAR_CHARSET (c1); |
36595814 SM |
2896 | re_wchar_t c2 = MAKE_CHAR (charset, 0, 0); |
2897 | ||
b54f61ed KH |
2898 | SET_RANGE_TABLE_WORK_AREA (range_table_work, |
2899 | c2, c1); | |
333526e0 | 2900 | c1 = 0377; |
b54f61ed | 2901 | } |
e934739e RS |
2902 | } |
2903 | else if (!SAME_CHARSET_P (c, c1)) | |
b18215fc | 2904 | FREE_STACK_RETURN (REG_ERANGE); |
e318085a | 2905 | } |
25fe55af | 2906 | else |
b18215fc RS |
2907 | /* Range from C to C. */ |
2908 | c1 = c; | |
2909 | ||
2910 | /* Set the range ... */ | |
2911 | if (SINGLE_BYTE_CHAR_P (c)) | |
2912 | /* ... into bitmap. */ | |
25fe55af | 2913 | { |
01618498 | 2914 | re_wchar_t this_char; |
36595814 | 2915 | re_wchar_t range_start = c, range_end = c1; |
b18215fc RS |
2916 | |
2917 | /* If the start is after the end, the range is empty. */ | |
2918 | if (range_start > range_end) | |
2919 | { | |
2920 | if (syntax & RE_NO_EMPTY_RANGES) | |
2921 | FREE_STACK_RETURN (REG_ERANGE); | |
2922 | /* Else, repeat the loop. */ | |
2923 | } | |
2924 | else | |
2925 | { | |
2926 | for (this_char = range_start; this_char <= range_end; | |
2927 | this_char++) | |
2928 | SET_LIST_BIT (TRANSLATE (this_char)); | |
e934739e | 2929 | } |
25fe55af | 2930 | } |
e318085a | 2931 | else |
b18215fc RS |
2932 | /* ... into range table. */ |
2933 | SET_RANGE_TABLE_WORK_AREA (range_table_work, c, c1); | |
e318085a RS |
2934 | } |
2935 | ||
25fe55af RS |
2936 | /* Discard any (non)matching list bytes that are all 0 at the |
2937 | end of the map. Decrease the map-length byte too. */ | |
2938 | while ((int) b[-1] > 0 && b[b[-1] - 1] == 0) | |
2939 | b[-1]--; | |
2940 | b += b[-1]; | |
fa9a63c5 | 2941 | |
96cc36cc RS |
2942 | /* Build real range table from work area. */ |
2943 | if (RANGE_TABLE_WORK_USED (range_table_work) | |
2944 | || RANGE_TABLE_WORK_BITS (range_table_work)) | |
b18215fc RS |
2945 | { |
2946 | int i; | |
2947 | int used = RANGE_TABLE_WORK_USED (range_table_work); | |
fa9a63c5 | 2948 | |
b18215fc | 2949 | /* Allocate space for COUNT + RANGE_TABLE. Needs two |
96cc36cc RS |
2950 | bytes for flags, two for COUNT, and three bytes for |
2951 | each character. */ | |
2952 | GET_BUFFER_SPACE (4 + used * 3); | |
fa9a63c5 | 2953 | |
b18215fc RS |
2954 | /* Indicate the existence of range table. */ |
2955 | laststart[1] |= 0x80; | |
fa9a63c5 | 2956 | |
96cc36cc RS |
2957 | /* Store the character class flag bits into the range table. |
2958 | If not in emacs, these flag bits are always 0. */ | |
2959 | *b++ = RANGE_TABLE_WORK_BITS (range_table_work) & 0xff; | |
2960 | *b++ = RANGE_TABLE_WORK_BITS (range_table_work) >> 8; | |
2961 | ||
b18215fc RS |
2962 | STORE_NUMBER_AND_INCR (b, used / 2); |
2963 | for (i = 0; i < used; i++) | |
2964 | STORE_CHARACTER_AND_INCR | |
2965 | (b, RANGE_TABLE_WORK_ELT (range_table_work, i)); | |
2966 | } | |
25fe55af RS |
2967 | } |
2968 | break; | |
fa9a63c5 RM |
2969 | |
2970 | ||
b18215fc | 2971 | case '(': |
25fe55af RS |
2972 | if (syntax & RE_NO_BK_PARENS) |
2973 | goto handle_open; | |
2974 | else | |
2975 | goto normal_char; | |
fa9a63c5 RM |
2976 | |
2977 | ||
25fe55af RS |
2978 | case ')': |
2979 | if (syntax & RE_NO_BK_PARENS) | |
2980 | goto handle_close; | |
2981 | else | |
2982 | goto normal_char; | |
e318085a RS |
2983 | |
2984 | ||
25fe55af RS |
2985 | case '\n': |
2986 | if (syntax & RE_NEWLINE_ALT) | |
2987 | goto handle_alt; | |
2988 | else | |
2989 | goto normal_char; | |
e318085a RS |
2990 | |
2991 | ||
b18215fc | 2992 | case '|': |
25fe55af RS |
2993 | if (syntax & RE_NO_BK_VBAR) |
2994 | goto handle_alt; | |
2995 | else | |
2996 | goto normal_char; | |
2997 | ||
2998 | ||
2999 | case '{': | |
3000 | if (syntax & RE_INTERVALS && syntax & RE_NO_BK_BRACES) | |
3001 | goto handle_interval; | |
3002 | else | |
3003 | goto normal_char; | |
3004 | ||
3005 | ||
3006 | case '\\': | |
3007 | if (p == pend) FREE_STACK_RETURN (REG_EESCAPE); | |
3008 | ||
3009 | /* Do not translate the character after the \, so that we can | |
3010 | distinguish, e.g., \B from \b, even if we normally would | |
3011 | translate, e.g., B to b. */ | |
36595814 | 3012 | PATFETCH (c); |
25fe55af RS |
3013 | |
3014 | switch (c) | |
3015 | { | |
3016 | case '(': | |
3017 | if (syntax & RE_NO_BK_PARENS) | |
3018 | goto normal_backslash; | |
3019 | ||
3020 | handle_open: | |
505bde11 SM |
3021 | { |
3022 | int shy = 0; | |
3023 | if (p+1 < pend) | |
3024 | { | |
3025 | /* Look for a special (?...) construct */ | |
ed0767d8 | 3026 | if ((syntax & RE_SHY_GROUPS) && *p == '?') |
505bde11 | 3027 | { |
ed0767d8 | 3028 | PATFETCH (c); /* Gobble up the '?'. */ |
505bde11 SM |
3029 | PATFETCH (c); |
3030 | switch (c) | |
3031 | { | |
3032 | case ':': shy = 1; break; | |
3033 | default: | |
3034 | /* Only (?:...) is supported right now. */ | |
3035 | FREE_STACK_RETURN (REG_BADPAT); | |
3036 | } | |
3037 | } | |
505bde11 SM |
3038 | } |
3039 | ||
3040 | if (!shy) | |
3041 | { | |
3042 | bufp->re_nsub++; | |
3043 | regnum++; | |
3044 | } | |
25fe55af | 3045 | |
99633e97 SM |
3046 | if (COMPILE_STACK_FULL) |
3047 | { | |
3048 | RETALLOC (compile_stack.stack, compile_stack.size << 1, | |
3049 | compile_stack_elt_t); | |
3050 | if (compile_stack.stack == NULL) return REG_ESPACE; | |
25fe55af | 3051 | |
99633e97 SM |
3052 | compile_stack.size <<= 1; |
3053 | } | |
25fe55af | 3054 | |
99633e97 SM |
3055 | /* These are the values to restore when we hit end of this |
3056 | group. They are all relative offsets, so that if the | |
3057 | whole pattern moves because of realloc, they will still | |
3058 | be valid. */ | |
3059 | COMPILE_STACK_TOP.begalt_offset = begalt - bufp->buffer; | |
3060 | COMPILE_STACK_TOP.fixup_alt_jump | |
3061 | = fixup_alt_jump ? fixup_alt_jump - bufp->buffer + 1 : 0; | |
3062 | COMPILE_STACK_TOP.laststart_offset = b - bufp->buffer; | |
3063 | COMPILE_STACK_TOP.regnum = shy ? -regnum : regnum; | |
3064 | ||
3065 | /* Do not push a | |
3066 | start_memory for groups beyond the last one we can | |
3067 | represent in the compiled pattern. */ | |
3068 | if (regnum <= MAX_REGNUM && !shy) | |
3069 | BUF_PUSH_2 (start_memory, regnum); | |
3070 | ||
3071 | compile_stack.avail++; | |
3072 | ||
3073 | fixup_alt_jump = 0; | |
3074 | laststart = 0; | |
3075 | begalt = b; | |
3076 | /* If we've reached MAX_REGNUM groups, then this open | |
3077 | won't actually generate any code, so we'll have to | |
3078 | clear pending_exact explicitly. */ | |
3079 | pending_exact = 0; | |
3080 | break; | |
505bde11 | 3081 | } |
25fe55af RS |
3082 | |
3083 | case ')': | |
3084 | if (syntax & RE_NO_BK_PARENS) goto normal_backslash; | |
3085 | ||
3086 | if (COMPILE_STACK_EMPTY) | |
505bde11 SM |
3087 | { |
3088 | if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD) | |
3089 | goto normal_backslash; | |
3090 | else | |
3091 | FREE_STACK_RETURN (REG_ERPAREN); | |
3092 | } | |
25fe55af RS |
3093 | |
3094 | handle_close: | |
505bde11 | 3095 | FIXUP_ALT_JUMP (); |
25fe55af RS |
3096 | |
3097 | /* See similar code for backslashed left paren above. */ | |
3098 | if (COMPILE_STACK_EMPTY) | |
505bde11 SM |
3099 | { |
3100 | if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD) | |
3101 | goto normal_char; | |
3102 | else | |
3103 | FREE_STACK_RETURN (REG_ERPAREN); | |
3104 | } | |
25fe55af RS |
3105 | |
3106 | /* Since we just checked for an empty stack above, this | |
3107 | ``can't happen''. */ | |
3108 | assert (compile_stack.avail != 0); | |
3109 | { | |
3110 | /* We don't just want to restore into `regnum', because | |
3111 | later groups should continue to be numbered higher, | |
3112 | as in `(ab)c(de)' -- the second group is #2. */ | |
3113 | regnum_t this_group_regnum; | |
3114 | ||
3115 | compile_stack.avail--; | |
3116 | begalt = bufp->buffer + COMPILE_STACK_TOP.begalt_offset; | |
3117 | fixup_alt_jump | |
3118 | = COMPILE_STACK_TOP.fixup_alt_jump | |
3119 | ? bufp->buffer + COMPILE_STACK_TOP.fixup_alt_jump - 1 | |
3120 | : 0; | |
3121 | laststart = bufp->buffer + COMPILE_STACK_TOP.laststart_offset; | |
3122 | this_group_regnum = COMPILE_STACK_TOP.regnum; | |
b18215fc RS |
3123 | /* If we've reached MAX_REGNUM groups, then this open |
3124 | won't actually generate any code, so we'll have to | |
3125 | clear pending_exact explicitly. */ | |
3126 | pending_exact = 0; | |
e318085a | 3127 | |
25fe55af RS |
3128 | /* We're at the end of the group, so now we know how many |
3129 | groups were inside this one. */ | |
505bde11 SM |
3130 | if (this_group_regnum <= MAX_REGNUM && this_group_regnum > 0) |
3131 | BUF_PUSH_2 (stop_memory, this_group_regnum); | |
25fe55af RS |
3132 | } |
3133 | break; | |
3134 | ||
3135 | ||
3136 | case '|': /* `\|'. */ | |
3137 | if (syntax & RE_LIMITED_OPS || syntax & RE_NO_BK_VBAR) | |
3138 | goto normal_backslash; | |
3139 | handle_alt: | |
3140 | if (syntax & RE_LIMITED_OPS) | |
3141 | goto normal_char; | |
3142 | ||
3143 | /* Insert before the previous alternative a jump which | |
3144 | jumps to this alternative if the former fails. */ | |
3145 | GET_BUFFER_SPACE (3); | |
3146 | INSERT_JUMP (on_failure_jump, begalt, b + 6); | |
3147 | pending_exact = 0; | |
3148 | b += 3; | |
3149 | ||
3150 | /* The alternative before this one has a jump after it | |
3151 | which gets executed if it gets matched. Adjust that | |
3152 | jump so it will jump to this alternative's analogous | |
3153 | jump (put in below, which in turn will jump to the next | |
3154 | (if any) alternative's such jump, etc.). The last such | |
3155 | jump jumps to the correct final destination. A picture: | |
3156 | _____ _____ | |
3157 | | | | | | |
3158 | | v | v | |
3159 | a | b | c | |
3160 | ||
3161 | If we are at `b', then fixup_alt_jump right now points to a | |
3162 | three-byte space after `a'. We'll put in the jump, set | |
3163 | fixup_alt_jump to right after `b', and leave behind three | |
3164 | bytes which we'll fill in when we get to after `c'. */ | |
3165 | ||
505bde11 | 3166 | FIXUP_ALT_JUMP (); |
25fe55af RS |
3167 | |
3168 | /* Mark and leave space for a jump after this alternative, | |
3169 | to be filled in later either by next alternative or | |
3170 | when know we're at the end of a series of alternatives. */ | |
3171 | fixup_alt_jump = b; | |
3172 | GET_BUFFER_SPACE (3); | |
3173 | b += 3; | |
3174 | ||
3175 | laststart = 0; | |
3176 | begalt = b; | |
3177 | break; | |
3178 | ||
3179 | ||
3180 | case '{': | |
3181 | /* If \{ is a literal. */ | |
3182 | if (!(syntax & RE_INTERVALS) | |
3183 | /* If we're at `\{' and it's not the open-interval | |
3184 | operator. */ | |
4bb91c68 | 3185 | || (syntax & RE_NO_BK_BRACES)) |
25fe55af RS |
3186 | goto normal_backslash; |
3187 | ||
3188 | handle_interval: | |
3189 | { | |
3190 | /* If got here, then the syntax allows intervals. */ | |
3191 | ||
3192 | /* At least (most) this many matches must be made. */ | |
99633e97 | 3193 | int lower_bound = 0, upper_bound = -1; |
25fe55af | 3194 | |
ed0767d8 | 3195 | beg_interval = p; |
25fe55af RS |
3196 | |
3197 | if (p == pend) | |
4bb91c68 | 3198 | FREE_STACK_RETURN (REG_EBRACE); |
25fe55af RS |
3199 | |
3200 | GET_UNSIGNED_NUMBER (lower_bound); | |
3201 | ||
3202 | if (c == ',') | |
ed0767d8 | 3203 | GET_UNSIGNED_NUMBER (upper_bound); |
25fe55af RS |
3204 | else |
3205 | /* Interval such as `{1}' => match exactly once. */ | |
3206 | upper_bound = lower_bound; | |
3207 | ||
3208 | if (lower_bound < 0 || upper_bound > RE_DUP_MAX | |
ed0767d8 | 3209 | || (upper_bound >= 0 && lower_bound > upper_bound)) |
4bb91c68 | 3210 | FREE_STACK_RETURN (REG_BADBR); |
25fe55af RS |
3211 | |
3212 | if (!(syntax & RE_NO_BK_BRACES)) | |
3213 | { | |
4bb91c68 SM |
3214 | if (c != '\\') |
3215 | FREE_STACK_RETURN (REG_BADBR); | |
25fe55af RS |
3216 | |
3217 | PATFETCH (c); | |
3218 | } | |
3219 | ||
3220 | if (c != '}') | |
4bb91c68 | 3221 | FREE_STACK_RETURN (REG_BADBR); |
25fe55af RS |
3222 | |
3223 | /* We just parsed a valid interval. */ | |
3224 | ||
3225 | /* If it's invalid to have no preceding re. */ | |
3226 | if (!laststart) | |
3227 | { | |
3228 | if (syntax & RE_CONTEXT_INVALID_OPS) | |
3229 | FREE_STACK_RETURN (REG_BADRPT); | |
3230 | else if (syntax & RE_CONTEXT_INDEP_OPS) | |
3231 | laststart = b; | |
3232 | else | |
3233 | goto unfetch_interval; | |
3234 | } | |
3235 | ||
6df42991 SM |
3236 | if (upper_bound == 0) |
3237 | /* If the upper bound is zero, just drop the sub pattern | |
3238 | altogether. */ | |
3239 | b = laststart; | |
3240 | else if (lower_bound == 1 && upper_bound == 1) | |
3241 | /* Just match it once: nothing to do here. */ | |
3242 | ; | |
3243 | ||
3244 | /* Otherwise, we have a nontrivial interval. When | |
3245 | we're all done, the pattern will look like: | |
3246 | set_number_at <jump count> <upper bound> | |
3247 | set_number_at <succeed_n count> <lower bound> | |
3248 | succeed_n <after jump addr> <succeed_n count> | |
3249 | <body of loop> | |
3250 | jump_n <succeed_n addr> <jump count> | |
3251 | (The upper bound and `jump_n' are omitted if | |
3252 | `upper_bound' is 1, though.) */ | |
3253 | else | |
3254 | { /* If the upper bound is > 1, we need to insert | |
3255 | more at the end of the loop. */ | |
3256 | unsigned int nbytes = (upper_bound < 0 ? 3 | |
3257 | : upper_bound > 1 ? 5 : 0); | |
3258 | unsigned int startoffset = 0; | |
3259 | ||
3260 | GET_BUFFER_SPACE (20); /* We might use less. */ | |
3261 | ||
3262 | if (lower_bound == 0) | |
3263 | { | |
3264 | /* A succeed_n that starts with 0 is really a | |
3265 | a simple on_failure_jump_loop. */ | |
3266 | INSERT_JUMP (on_failure_jump_loop, laststart, | |
3267 | b + 3 + nbytes); | |
3268 | b += 3; | |
3269 | } | |
3270 | else | |
3271 | { | |
3272 | /* Initialize lower bound of the `succeed_n', even | |
3273 | though it will be set during matching by its | |
3274 | attendant `set_number_at' (inserted next), | |
3275 | because `re_compile_fastmap' needs to know. | |
3276 | Jump to the `jump_n' we might insert below. */ | |
3277 | INSERT_JUMP2 (succeed_n, laststart, | |
3278 | b + 5 + nbytes, | |
3279 | lower_bound); | |
3280 | b += 5; | |
3281 | ||
3282 | /* Code to initialize the lower bound. Insert | |
3283 | before the `succeed_n'. The `5' is the last two | |
3284 | bytes of this `set_number_at', plus 3 bytes of | |
3285 | the following `succeed_n'. */ | |
3286 | insert_op2 (set_number_at, laststart, 5, lower_bound, b); | |
3287 | b += 5; | |
3288 | startoffset += 5; | |
3289 | } | |
3290 | ||
3291 | if (upper_bound < 0) | |
3292 | { | |
3293 | /* A negative upper bound stands for infinity, | |
3294 | in which case it degenerates to a plain jump. */ | |
3295 | STORE_JUMP (jump, b, laststart + startoffset); | |
3296 | b += 3; | |
3297 | } | |
3298 | else if (upper_bound > 1) | |
3299 | { /* More than one repetition is allowed, so | |
3300 | append a backward jump to the `succeed_n' | |
3301 | that starts this interval. | |
3302 | ||
3303 | When we've reached this during matching, | |
3304 | we'll have matched the interval once, so | |
3305 | jump back only `upper_bound - 1' times. */ | |
3306 | STORE_JUMP2 (jump_n, b, laststart + startoffset, | |
3307 | upper_bound - 1); | |
3308 | b += 5; | |
3309 | ||
3310 | /* The location we want to set is the second | |
3311 | parameter of the `jump_n'; that is `b-2' as | |
3312 | an absolute address. `laststart' will be | |
3313 | the `set_number_at' we're about to insert; | |
3314 | `laststart+3' the number to set, the source | |
3315 | for the relative address. But we are | |
3316 | inserting into the middle of the pattern -- | |
3317 | so everything is getting moved up by 5. | |
3318 | Conclusion: (b - 2) - (laststart + 3) + 5, | |
3319 | i.e., b - laststart. | |
3320 | ||
3321 | We insert this at the beginning of the loop | |
3322 | so that if we fail during matching, we'll | |
3323 | reinitialize the bounds. */ | |
3324 | insert_op2 (set_number_at, laststart, b - laststart, | |
3325 | upper_bound - 1, b); | |
3326 | b += 5; | |
3327 | } | |
3328 | } | |
25fe55af RS |
3329 | pending_exact = 0; |
3330 | beg_interval = NULL; | |
3331 | } | |
3332 | break; | |
3333 | ||
3334 | unfetch_interval: | |
3335 | /* If an invalid interval, match the characters as literals. */ | |
3336 | assert (beg_interval); | |
3337 | p = beg_interval; | |
3338 | beg_interval = NULL; | |
3339 | ||
3340 | /* normal_char and normal_backslash need `c'. */ | |
ed0767d8 | 3341 | c = '{'; |
25fe55af RS |
3342 | |
3343 | if (!(syntax & RE_NO_BK_BRACES)) | |
3344 | { | |
ed0767d8 SM |
3345 | assert (p > pattern && p[-1] == '\\'); |
3346 | goto normal_backslash; | |
25fe55af | 3347 | } |
ed0767d8 SM |
3348 | else |
3349 | goto normal_char; | |
e318085a | 3350 | |
b18215fc | 3351 | #ifdef emacs |
25fe55af RS |
3352 | /* There is no way to specify the before_dot and after_dot |
3353 | operators. rms says this is ok. --karl */ | |
3354 | case '=': | |
3355 | BUF_PUSH (at_dot); | |
3356 | break; | |
3357 | ||
3358 | case 's': | |
3359 | laststart = b; | |
3360 | PATFETCH (c); | |
3361 | BUF_PUSH_2 (syntaxspec, syntax_spec_code[c]); | |
3362 | break; | |
3363 | ||
3364 | case 'S': | |
3365 | laststart = b; | |
3366 | PATFETCH (c); | |
3367 | BUF_PUSH_2 (notsyntaxspec, syntax_spec_code[c]); | |
3368 | break; | |
b18215fc RS |
3369 | |
3370 | case 'c': | |
3371 | laststart = b; | |
36595814 | 3372 | PATFETCH (c); |
b18215fc RS |
3373 | BUF_PUSH_2 (categoryspec, c); |
3374 | break; | |
e318085a | 3375 | |
b18215fc RS |
3376 | case 'C': |
3377 | laststart = b; | |
36595814 | 3378 | PATFETCH (c); |
b18215fc RS |
3379 | BUF_PUSH_2 (notcategoryspec, c); |
3380 | break; | |
3381 | #endif /* emacs */ | |
e318085a | 3382 | |
e318085a | 3383 | |
25fe55af | 3384 | case 'w': |
4bb91c68 SM |
3385 | if (syntax & RE_NO_GNU_OPS) |
3386 | goto normal_char; | |
25fe55af | 3387 | laststart = b; |
1fb352e0 | 3388 | BUF_PUSH_2 (syntaxspec, Sword); |
25fe55af | 3389 | break; |
e318085a | 3390 | |
e318085a | 3391 | |
25fe55af | 3392 | case 'W': |
4bb91c68 SM |
3393 | if (syntax & RE_NO_GNU_OPS) |
3394 | goto normal_char; | |
25fe55af | 3395 | laststart = b; |
1fb352e0 | 3396 | BUF_PUSH_2 (notsyntaxspec, Sword); |
25fe55af | 3397 | break; |
e318085a RS |
3398 | |
3399 | ||
25fe55af | 3400 | case '<': |
4bb91c68 SM |
3401 | if (syntax & RE_NO_GNU_OPS) |
3402 | goto normal_char; | |
25fe55af RS |
3403 | BUF_PUSH (wordbeg); |
3404 | break; | |
e318085a | 3405 | |
25fe55af | 3406 | case '>': |
4bb91c68 SM |
3407 | if (syntax & RE_NO_GNU_OPS) |
3408 | goto normal_char; | |
25fe55af RS |
3409 | BUF_PUSH (wordend); |
3410 | break; | |
e318085a | 3411 | |
669fa600 SM |
3412 | case '_': |
3413 | if (syntax & RE_NO_GNU_OPS) | |
3414 | goto normal_char; | |
3415 | laststart = b; | |
3416 | PATFETCH (c); | |
3417 | if (c == '<') | |
3418 | BUF_PUSH (symbeg); | |
3419 | else if (c == '>') | |
3420 | BUF_PUSH (symend); | |
3421 | else | |
3422 | FREE_STACK_RETURN (REG_BADPAT); | |
3423 | break; | |
3424 | ||
25fe55af | 3425 | case 'b': |
4bb91c68 SM |
3426 | if (syntax & RE_NO_GNU_OPS) |
3427 | goto normal_char; | |
25fe55af RS |
3428 | BUF_PUSH (wordbound); |
3429 | break; | |
e318085a | 3430 | |
25fe55af | 3431 | case 'B': |
4bb91c68 SM |
3432 | if (syntax & RE_NO_GNU_OPS) |
3433 | goto normal_char; | |
25fe55af RS |
3434 | BUF_PUSH (notwordbound); |
3435 | break; | |
fa9a63c5 | 3436 | |
25fe55af | 3437 | case '`': |
4bb91c68 SM |
3438 | if (syntax & RE_NO_GNU_OPS) |
3439 | goto normal_char; | |
25fe55af RS |
3440 | BUF_PUSH (begbuf); |
3441 | break; | |
e318085a | 3442 | |
25fe55af | 3443 | case '\'': |
4bb91c68 SM |
3444 | if (syntax & RE_NO_GNU_OPS) |
3445 | goto normal_char; | |
25fe55af RS |
3446 | BUF_PUSH (endbuf); |
3447 | break; | |
e318085a | 3448 | |
25fe55af RS |
3449 | case '1': case '2': case '3': case '4': case '5': |
3450 | case '6': case '7': case '8': case '9': | |
0cdd06f8 SM |
3451 | { |
3452 | regnum_t reg; | |
e318085a | 3453 | |
0cdd06f8 SM |
3454 | if (syntax & RE_NO_BK_REFS) |
3455 | goto normal_backslash; | |
e318085a | 3456 | |
0cdd06f8 | 3457 | reg = c - '0'; |
e318085a | 3458 | |
0cdd06f8 SM |
3459 | /* Can't back reference to a subexpression before its end. */ |
3460 | if (reg > regnum || group_in_compile_stack (compile_stack, reg)) | |
3461 | FREE_STACK_RETURN (REG_ESUBREG); | |
e318085a | 3462 | |
0cdd06f8 SM |
3463 | laststart = b; |
3464 | BUF_PUSH_2 (duplicate, reg); | |
3465 | } | |
25fe55af | 3466 | break; |
e318085a | 3467 | |
e318085a | 3468 | |
25fe55af RS |
3469 | case '+': |
3470 | case '?': | |
3471 | if (syntax & RE_BK_PLUS_QM) | |
3472 | goto handle_plus; | |
3473 | else | |
3474 | goto normal_backslash; | |
3475 | ||
3476 | default: | |
3477 | normal_backslash: | |
3478 | /* You might think it would be useful for \ to mean | |
3479 | not to translate; but if we don't translate it | |
4bb91c68 | 3480 | it will never match anything. */ |
25fe55af RS |
3481 | goto normal_char; |
3482 | } | |
3483 | break; | |
fa9a63c5 RM |
3484 | |
3485 | ||
3486 | default: | |
25fe55af | 3487 | /* Expects the character in `c'. */ |
fa9a63c5 | 3488 | normal_char: |
36595814 | 3489 | /* If no exactn currently being built. */ |
25fe55af | 3490 | if (!pending_exact |
fa9a63c5 | 3491 | |
25fe55af RS |
3492 | /* If last exactn not at current position. */ |
3493 | || pending_exact + *pending_exact + 1 != b | |
5e69f11e | 3494 | |
25fe55af | 3495 | /* We have only one byte following the exactn for the count. */ |
2d1675e4 | 3496 | || *pending_exact >= (1 << BYTEWIDTH) - MAX_MULTIBYTE_LENGTH |
fa9a63c5 | 3497 | |
25fe55af | 3498 | /* If followed by a repetition operator. */ |
9d99031f | 3499 | || (p != pend && (*p == '*' || *p == '^')) |
fa9a63c5 | 3500 | || ((syntax & RE_BK_PLUS_QM) |
9d99031f RS |
3501 | ? p + 1 < pend && *p == '\\' && (p[1] == '+' || p[1] == '?') |
3502 | : p != pend && (*p == '+' || *p == '?')) | |
fa9a63c5 | 3503 | || ((syntax & RE_INTERVALS) |
25fe55af | 3504 | && ((syntax & RE_NO_BK_BRACES) |
9d99031f RS |
3505 | ? p != pend && *p == '{' |
3506 | : p + 1 < pend && p[0] == '\\' && p[1] == '{'))) | |
fa9a63c5 RM |
3507 | { |
3508 | /* Start building a new exactn. */ | |
5e69f11e | 3509 | |
25fe55af | 3510 | laststart = b; |
fa9a63c5 RM |
3511 | |
3512 | BUF_PUSH_2 (exactn, 0); | |
3513 | pending_exact = b - 1; | |
25fe55af | 3514 | } |
5e69f11e | 3515 | |
2d1675e4 SM |
3516 | GET_BUFFER_SPACE (MAX_MULTIBYTE_LENGTH); |
3517 | { | |
e0277a47 KH |
3518 | int len; |
3519 | ||
36595814 | 3520 | c = TRANSLATE (c); |
e0277a47 KH |
3521 | if (multibyte) |
3522 | len = CHAR_STRING (c, b); | |
3523 | else | |
3524 | *b = c, len = 1; | |
2d1675e4 SM |
3525 | b += len; |
3526 | (*pending_exact) += len; | |
3527 | } | |
3528 | ||
fa9a63c5 | 3529 | break; |
25fe55af | 3530 | } /* switch (c) */ |
fa9a63c5 RM |
3531 | } /* while p != pend */ |
3532 | ||
5e69f11e | 3533 | |
fa9a63c5 | 3534 | /* Through the pattern now. */ |
5e69f11e | 3535 | |
505bde11 | 3536 | FIXUP_ALT_JUMP (); |
fa9a63c5 | 3537 | |
5e69f11e | 3538 | if (!COMPILE_STACK_EMPTY) |
fa9a63c5 RM |
3539 | FREE_STACK_RETURN (REG_EPAREN); |
3540 | ||
3541 | /* If we don't want backtracking, force success | |
3542 | the first time we reach the end of the compiled pattern. */ | |
3543 | if (syntax & RE_NO_POSIX_BACKTRACKING) | |
3544 | BUF_PUSH (succeed); | |
3545 | ||
fa9a63c5 RM |
3546 | /* We have succeeded; set the length of the buffer. */ |
3547 | bufp->used = b - bufp->buffer; | |
3548 | ||
3549 | #ifdef DEBUG | |
99633e97 | 3550 | if (debug > 0) |
fa9a63c5 | 3551 | { |
505bde11 | 3552 | re_compile_fastmap (bufp); |
fa9a63c5 RM |
3553 | DEBUG_PRINT1 ("\nCompiled pattern: \n"); |
3554 | print_compiled_pattern (bufp); | |
3555 | } | |
99633e97 | 3556 | debug--; |
fa9a63c5 RM |
3557 | #endif /* DEBUG */ |
3558 | ||
3559 | #ifndef MATCH_MAY_ALLOCATE | |
3560 | /* Initialize the failure stack to the largest possible stack. This | |
3561 | isn't necessary unless we're trying to avoid calling alloca in | |
3562 | the search and match routines. */ | |
3563 | { | |
3564 | int num_regs = bufp->re_nsub + 1; | |
3565 | ||
320a2a73 | 3566 | if (fail_stack.size < re_max_failures * TYPICAL_FAILURE_SIZE) |
fa9a63c5 | 3567 | { |
a26f4ccd | 3568 | fail_stack.size = re_max_failures * TYPICAL_FAILURE_SIZE; |
fa9a63c5 | 3569 | |
fa9a63c5 RM |
3570 | if (! fail_stack.stack) |
3571 | fail_stack.stack | |
5e69f11e | 3572 | = (fail_stack_elt_t *) malloc (fail_stack.size |
fa9a63c5 RM |
3573 | * sizeof (fail_stack_elt_t)); |
3574 | else | |
3575 | fail_stack.stack | |
3576 | = (fail_stack_elt_t *) realloc (fail_stack.stack, | |
3577 | (fail_stack.size | |
3578 | * sizeof (fail_stack_elt_t))); | |
fa9a63c5 RM |
3579 | } |
3580 | ||
3581 | regex_grow_registers (num_regs); | |
3582 | } | |
3583 | #endif /* not MATCH_MAY_ALLOCATE */ | |
3584 | ||
7c67c9f6 | 3585 | FREE_STACK_RETURN (REG_NOERROR); |
fa9a63c5 RM |
3586 | } /* regex_compile */ |
3587 | \f | |
3588 | /* Subroutines for `regex_compile'. */ | |
3589 | ||
25fe55af | 3590 | /* Store OP at LOC followed by two-byte integer parameter ARG. */ |
fa9a63c5 RM |
3591 | |
3592 | static void | |
3593 | store_op1 (op, loc, arg) | |
3594 | re_opcode_t op; | |
3595 | unsigned char *loc; | |
3596 | int arg; | |
3597 | { | |
3598 | *loc = (unsigned char) op; | |
3599 | STORE_NUMBER (loc + 1, arg); | |
3600 | } | |
3601 | ||
3602 | ||
3603 | /* Like `store_op1', but for two two-byte parameters ARG1 and ARG2. */ | |
3604 | ||
3605 | static void | |
3606 | store_op2 (op, loc, arg1, arg2) | |
3607 | re_opcode_t op; | |
3608 | unsigned char *loc; | |
3609 | int arg1, arg2; | |
3610 | { | |
3611 | *loc = (unsigned char) op; | |
3612 | STORE_NUMBER (loc + 1, arg1); | |
3613 | STORE_NUMBER (loc + 3, arg2); | |
3614 | } | |
3615 | ||
3616 | ||
3617 | /* Copy the bytes from LOC to END to open up three bytes of space at LOC | |
3618 | for OP followed by two-byte integer parameter ARG. */ | |
3619 | ||
3620 | static void | |
3621 | insert_op1 (op, loc, arg, end) | |
3622 | re_opcode_t op; | |
3623 | unsigned char *loc; | |
3624 | int arg; | |
5e69f11e | 3625 | unsigned char *end; |
fa9a63c5 RM |
3626 | { |
3627 | register unsigned char *pfrom = end; | |
3628 | register unsigned char *pto = end + 3; | |
3629 | ||
3630 | while (pfrom != loc) | |
3631 | *--pto = *--pfrom; | |
5e69f11e | 3632 | |
fa9a63c5 RM |
3633 | store_op1 (op, loc, arg); |
3634 | } | |
3635 | ||
3636 | ||
3637 | /* Like `insert_op1', but for two two-byte parameters ARG1 and ARG2. */ | |
3638 | ||
3639 | static void | |
3640 | insert_op2 (op, loc, arg1, arg2, end) | |
3641 | re_opcode_t op; | |
3642 | unsigned char *loc; | |
3643 | int arg1, arg2; | |
5e69f11e | 3644 | unsigned char *end; |
fa9a63c5 RM |
3645 | { |
3646 | register unsigned char *pfrom = end; | |
3647 | register unsigned char *pto = end + 5; | |
3648 | ||
3649 | while (pfrom != loc) | |
3650 | *--pto = *--pfrom; | |
5e69f11e | 3651 | |
fa9a63c5 RM |
3652 | store_op2 (op, loc, arg1, arg2); |
3653 | } | |
3654 | ||
3655 | ||
3656 | /* P points to just after a ^ in PATTERN. Return true if that ^ comes | |
3657 | after an alternative or a begin-subexpression. We assume there is at | |
3658 | least one character before the ^. */ | |
3659 | ||
3660 | static boolean | |
3661 | at_begline_loc_p (pattern, p, syntax) | |
01618498 | 3662 | re_char *pattern, *p; |
fa9a63c5 RM |
3663 | reg_syntax_t syntax; |
3664 | { | |
01618498 | 3665 | re_char *prev = p - 2; |
fa9a63c5 | 3666 | boolean prev_prev_backslash = prev > pattern && prev[-1] == '\\'; |
5e69f11e | 3667 | |
fa9a63c5 RM |
3668 | return |
3669 | /* After a subexpression? */ | |
3670 | (*prev == '(' && (syntax & RE_NO_BK_PARENS || prev_prev_backslash)) | |
25fe55af | 3671 | /* After an alternative? */ |
d2af47df SM |
3672 | || (*prev == '|' && (syntax & RE_NO_BK_VBAR || prev_prev_backslash)) |
3673 | /* After a shy subexpression? */ | |
3674 | || ((syntax & RE_SHY_GROUPS) && prev - 2 >= pattern | |
3675 | && prev[-1] == '?' && prev[-2] == '(' | |
3676 | && (syntax & RE_NO_BK_PARENS | |
3677 | || (prev - 3 >= pattern && prev[-3] == '\\'))); | |
fa9a63c5 RM |
3678 | } |
3679 | ||
3680 | ||
3681 | /* The dual of at_begline_loc_p. This one is for $. We assume there is | |
3682 | at least one character after the $, i.e., `P < PEND'. */ | |
3683 | ||
3684 | static boolean | |
3685 | at_endline_loc_p (p, pend, syntax) | |
01618498 | 3686 | re_char *p, *pend; |
99633e97 | 3687 | reg_syntax_t syntax; |
fa9a63c5 | 3688 | { |
01618498 | 3689 | re_char *next = p; |
fa9a63c5 | 3690 | boolean next_backslash = *next == '\\'; |
01618498 | 3691 | re_char *next_next = p + 1 < pend ? p + 1 : 0; |
5e69f11e | 3692 | |
fa9a63c5 RM |
3693 | return |
3694 | /* Before a subexpression? */ | |
3695 | (syntax & RE_NO_BK_PARENS ? *next == ')' | |
25fe55af | 3696 | : next_backslash && next_next && *next_next == ')') |
fa9a63c5 RM |
3697 | /* Before an alternative? */ |
3698 | || (syntax & RE_NO_BK_VBAR ? *next == '|' | |
25fe55af | 3699 | : next_backslash && next_next && *next_next == '|'); |
fa9a63c5 RM |
3700 | } |
3701 | ||
3702 | ||
5e69f11e | 3703 | /* Returns true if REGNUM is in one of COMPILE_STACK's elements and |
fa9a63c5 RM |
3704 | false if it's not. */ |
3705 | ||
3706 | static boolean | |
3707 | group_in_compile_stack (compile_stack, regnum) | |
3708 | compile_stack_type compile_stack; | |
3709 | regnum_t regnum; | |
3710 | { | |
3711 | int this_element; | |
3712 | ||
5e69f11e RM |
3713 | for (this_element = compile_stack.avail - 1; |
3714 | this_element >= 0; | |
fa9a63c5 RM |
3715 | this_element--) |
3716 | if (compile_stack.stack[this_element].regnum == regnum) | |
3717 | return true; | |
3718 | ||
3719 | return false; | |
3720 | } | |
fa9a63c5 | 3721 | \f |
f6a3f532 SM |
3722 | /* analyse_first. |
3723 | If fastmap is non-NULL, go through the pattern and fill fastmap | |
3724 | with all the possible leading chars. If fastmap is NULL, don't | |
3725 | bother filling it up (obviously) and only return whether the | |
3726 | pattern could potentially match the empty string. | |
3727 | ||
3728 | Return 1 if p..pend might match the empty string. | |
3729 | Return 0 if p..pend matches at least one char. | |
01618498 | 3730 | Return -1 if fastmap was not updated accurately. */ |
f6a3f532 SM |
3731 | |
3732 | static int | |
3733 | analyse_first (p, pend, fastmap, multibyte) | |
01618498 | 3734 | re_char *p, *pend; |
f6a3f532 SM |
3735 | char *fastmap; |
3736 | const int multibyte; | |
fa9a63c5 | 3737 | { |
505bde11 | 3738 | int j, k; |
1fb352e0 | 3739 | boolean not; |
fa9a63c5 | 3740 | |
b18215fc | 3741 | /* If all elements for base leading-codes in fastmap is set, this |
25fe55af | 3742 | flag is set true. */ |
b18215fc RS |
3743 | boolean match_any_multibyte_characters = false; |
3744 | ||
f6a3f532 | 3745 | assert (p); |
5e69f11e | 3746 | |
505bde11 SM |
3747 | /* The loop below works as follows: |
3748 | - It has a working-list kept in the PATTERN_STACK and which basically | |
3749 | starts by only containing a pointer to the first operation. | |
3750 | - If the opcode we're looking at is a match against some set of | |
3751 | chars, then we add those chars to the fastmap and go on to the | |
3752 | next work element from the worklist (done via `break'). | |
3753 | - If the opcode is a control operator on the other hand, we either | |
3754 | ignore it (if it's meaningless at this point, such as `start_memory') | |
3755 | or execute it (if it's a jump). If the jump has several destinations | |
3756 | (i.e. `on_failure_jump'), then we push the other destination onto the | |
3757 | worklist. | |
3758 | We guarantee termination by ignoring backward jumps (more or less), | |
3759 | so that `p' is monotonically increasing. More to the point, we | |
3760 | never set `p' (or push) anything `<= p1'. */ | |
3761 | ||
01618498 | 3762 | while (p < pend) |
fa9a63c5 | 3763 | { |
505bde11 SM |
3764 | /* `p1' is used as a marker of how far back a `on_failure_jump' |
3765 | can go without being ignored. It is normally equal to `p' | |
3766 | (which prevents any backward `on_failure_jump') except right | |
3767 | after a plain `jump', to allow patterns such as: | |
3768 | 0: jump 10 | |
3769 | 3..9: <body> | |
3770 | 10: on_failure_jump 3 | |
3771 | as used for the *? operator. */ | |
01618498 | 3772 | re_char *p1 = p; |
5e69f11e | 3773 | |
fa9a63c5 RM |
3774 | switch (SWITCH_ENUM_CAST ((re_opcode_t) *p++)) |
3775 | { | |
f6a3f532 | 3776 | case succeed: |
01618498 | 3777 | return 1; |
f6a3f532 | 3778 | continue; |
fa9a63c5 | 3779 | |
fa9a63c5 | 3780 | case duplicate: |
505bde11 SM |
3781 | /* If the first character has to match a backreference, that means |
3782 | that the group was empty (since it already matched). Since this | |
3783 | is the only case that interests us here, we can assume that the | |
3784 | backreference must match the empty string. */ | |
3785 | p++; | |
3786 | continue; | |
fa9a63c5 RM |
3787 | |
3788 | ||
3789 | /* Following are the cases which match a character. These end | |
25fe55af | 3790 | with `break'. */ |
fa9a63c5 RM |
3791 | |
3792 | case exactn: | |
e0277a47 KH |
3793 | if (fastmap) |
3794 | { | |
3795 | int c = RE_STRING_CHAR (p + 1, pend - p); | |
3796 | ||
3797 | if (SINGLE_BYTE_CHAR_P (c)) | |
3798 | fastmap[c] = 1; | |
3799 | else | |
3800 | fastmap[p[1]] = 1; | |
3801 | } | |
fa9a63c5 RM |
3802 | break; |
3803 | ||
3804 | ||
1fb352e0 SM |
3805 | case anychar: |
3806 | /* We could put all the chars except for \n (and maybe \0) | |
3807 | but we don't bother since it is generally not worth it. */ | |
f6a3f532 | 3808 | if (!fastmap) break; |
01618498 | 3809 | return -1; |
fa9a63c5 RM |
3810 | |
3811 | ||
b18215fc | 3812 | case charset_not: |
ba5c004d RS |
3813 | /* Chars beyond end of bitmap are possible matches. |
3814 | All the single-byte codes can occur in multibyte buffers. | |
3815 | So any that are not listed in the charset | |
3816 | are possible matches, even in multibyte buffers. */ | |
1fb352e0 | 3817 | if (!fastmap) break; |
b18215fc | 3818 | for (j = CHARSET_BITMAP_SIZE (&p[-1]) * BYTEWIDTH; |
1fb352e0 | 3819 | j < (1 << BYTEWIDTH); j++) |
b18215fc | 3820 | fastmap[j] = 1; |
1fb352e0 SM |
3821 | /* Fallthrough */ |
3822 | case charset: | |
3823 | if (!fastmap) break; | |
3824 | not = (re_opcode_t) *(p - 1) == charset_not; | |
b18215fc RS |
3825 | for (j = CHARSET_BITMAP_SIZE (&p[-1]) * BYTEWIDTH - 1, p++; |
3826 | j >= 0; j--) | |
1fb352e0 | 3827 | if (!!(p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH))) ^ not) |
b18215fc RS |
3828 | fastmap[j] = 1; |
3829 | ||
1fb352e0 SM |
3830 | if ((not && multibyte) |
3831 | /* Any character set can possibly contain a character | |
3832 | which doesn't match the specified set of characters. */ | |
3833 | || (CHARSET_RANGE_TABLE_EXISTS_P (&p[-2]) | |
3834 | && CHARSET_RANGE_TABLE_BITS (&p[-2]) != 0)) | |
3835 | /* If we can match a character class, we can match | |
3836 | any character set. */ | |
b18215fc RS |
3837 | { |
3838 | set_fastmap_for_multibyte_characters: | |
3839 | if (match_any_multibyte_characters == false) | |
3840 | { | |
3841 | for (j = 0x80; j < 0xA0; j++) /* XXX */ | |
3842 | if (BASE_LEADING_CODE_P (j)) | |
3843 | fastmap[j] = 1; | |
3844 | match_any_multibyte_characters = true; | |
3845 | } | |
3846 | } | |
b18215fc | 3847 | |
1fb352e0 SM |
3848 | else if (!not && CHARSET_RANGE_TABLE_EXISTS_P (&p[-2]) |
3849 | && match_any_multibyte_characters == false) | |
3850 | { | |
3851 | /* Set fastmap[I] 1 where I is a base leading code of each | |
3852 | multibyte character in the range table. */ | |
3853 | int c, count; | |
b18215fc | 3854 | |
1fb352e0 | 3855 | /* Make P points the range table. `+ 2' is to skip flag |
0b32bf0e | 3856 | bits for a character class. */ |
1fb352e0 | 3857 | p += CHARSET_BITMAP_SIZE (&p[-2]) + 2; |
b18215fc | 3858 | |
1fb352e0 SM |
3859 | /* Extract the number of ranges in range table into COUNT. */ |
3860 | EXTRACT_NUMBER_AND_INCR (count, p); | |
3861 | for (; count > 0; count--, p += 2 * 3) /* XXX */ | |
3862 | { | |
3863 | /* Extract the start of each range. */ | |
3864 | EXTRACT_CHARACTER (c, p); | |
3865 | j = CHAR_CHARSET (c); | |
3866 | fastmap[CHARSET_LEADING_CODE_BASE (j)] = 1; | |
3867 | } | |
3868 | } | |
b18215fc RS |
3869 | break; |
3870 | ||
1fb352e0 SM |
3871 | case syntaxspec: |
3872 | case notsyntaxspec: | |
3873 | if (!fastmap) break; | |
3874 | #ifndef emacs | |
3875 | not = (re_opcode_t)p[-1] == notsyntaxspec; | |
3876 | k = *p++; | |
3877 | for (j = 0; j < (1 << BYTEWIDTH); j++) | |
990b2375 | 3878 | if ((SYNTAX (j) == (enum syntaxcode) k) ^ not) |
b18215fc | 3879 | fastmap[j] = 1; |
b18215fc | 3880 | break; |
1fb352e0 | 3881 | #else /* emacs */ |
b18215fc RS |
3882 | /* This match depends on text properties. These end with |
3883 | aborting optimizations. */ | |
01618498 | 3884 | return -1; |
b18215fc RS |
3885 | |
3886 | case categoryspec: | |
b18215fc | 3887 | case notcategoryspec: |
1fb352e0 SM |
3888 | if (!fastmap) break; |
3889 | not = (re_opcode_t)p[-1] == notcategoryspec; | |
b18215fc | 3890 | k = *p++; |
1fb352e0 SM |
3891 | for (j = 0; j < (1 << BYTEWIDTH); j++) |
3892 | if ((CHAR_HAS_CATEGORY (j, k)) ^ not) | |
b18215fc RS |
3893 | fastmap[j] = 1; |
3894 | ||
1fb352e0 | 3895 | if (multibyte) |
b18215fc | 3896 | /* Any character set can possibly contain a character |
1fb352e0 | 3897 | whose category is K (or not). */ |
b18215fc RS |
3898 | goto set_fastmap_for_multibyte_characters; |
3899 | break; | |
3900 | ||
fa9a63c5 | 3901 | /* All cases after this match the empty string. These end with |
25fe55af | 3902 | `continue'. */ |
fa9a63c5 | 3903 | |
fa9a63c5 RM |
3904 | case before_dot: |
3905 | case at_dot: | |
3906 | case after_dot: | |
1fb352e0 | 3907 | #endif /* !emacs */ |
25fe55af RS |
3908 | case no_op: |
3909 | case begline: | |
3910 | case endline: | |
fa9a63c5 RM |
3911 | case begbuf: |
3912 | case endbuf: | |
3913 | case wordbound: | |
3914 | case notwordbound: | |
3915 | case wordbeg: | |
3916 | case wordend: | |
669fa600 SM |
3917 | case symbeg: |
3918 | case symend: | |
25fe55af | 3919 | continue; |
fa9a63c5 RM |
3920 | |
3921 | ||
fa9a63c5 | 3922 | case jump: |
25fe55af | 3923 | EXTRACT_NUMBER_AND_INCR (j, p); |
505bde11 SM |
3924 | if (j < 0) |
3925 | /* Backward jumps can only go back to code that we've already | |
3926 | visited. `re_compile' should make sure this is true. */ | |
3927 | break; | |
25fe55af | 3928 | p += j; |
505bde11 SM |
3929 | switch (SWITCH_ENUM_CAST ((re_opcode_t) *p)) |
3930 | { | |
3931 | case on_failure_jump: | |
3932 | case on_failure_keep_string_jump: | |
505bde11 | 3933 | case on_failure_jump_loop: |
0683b6fa | 3934 | case on_failure_jump_nastyloop: |
505bde11 SM |
3935 | case on_failure_jump_smart: |
3936 | p++; | |
3937 | break; | |
3938 | default: | |
3939 | continue; | |
3940 | }; | |
3941 | /* Keep `p1' to allow the `on_failure_jump' we are jumping to | |
3942 | to jump back to "just after here". */ | |
3943 | /* Fallthrough */ | |
fa9a63c5 | 3944 | |
25fe55af RS |
3945 | case on_failure_jump: |
3946 | case on_failure_keep_string_jump: | |
0683b6fa | 3947 | case on_failure_jump_nastyloop: |
505bde11 SM |
3948 | case on_failure_jump_loop: |
3949 | case on_failure_jump_smart: | |
25fe55af | 3950 | EXTRACT_NUMBER_AND_INCR (j, p); |
505bde11 | 3951 | if (p + j <= p1) |
ed0767d8 | 3952 | ; /* Backward jump to be ignored. */ |
01618498 SM |
3953 | else |
3954 | { /* We have to look down both arms. | |
3955 | We first go down the "straight" path so as to minimize | |
3956 | stack usage when going through alternatives. */ | |
3957 | int r = analyse_first (p, pend, fastmap, multibyte); | |
3958 | if (r) return r; | |
3959 | p += j; | |
3960 | } | |
25fe55af | 3961 | continue; |
fa9a63c5 RM |
3962 | |
3963 | ||
ed0767d8 SM |
3964 | case jump_n: |
3965 | /* This code simply does not properly handle forward jump_n. */ | |
3966 | DEBUG_STATEMENT (EXTRACT_NUMBER (j, p); assert (j < 0)); | |
3967 | p += 4; | |
3968 | /* jump_n can either jump or fall through. The (backward) jump | |
3969 | case has already been handled, so we only need to look at the | |
3970 | fallthrough case. */ | |
3971 | continue; | |
177c0ea7 | 3972 | |
fa9a63c5 | 3973 | case succeed_n: |
ed0767d8 SM |
3974 | /* If N == 0, it should be an on_failure_jump_loop instead. */ |
3975 | DEBUG_STATEMENT (EXTRACT_NUMBER (j, p + 2); assert (j > 0)); | |
3976 | p += 4; | |
3977 | /* We only care about one iteration of the loop, so we don't | |
3978 | need to consider the case where this behaves like an | |
3979 | on_failure_jump. */ | |
25fe55af | 3980 | continue; |
fa9a63c5 RM |
3981 | |
3982 | ||
3983 | case set_number_at: | |
25fe55af RS |
3984 | p += 4; |
3985 | continue; | |
fa9a63c5 RM |
3986 | |
3987 | ||
3988 | case start_memory: | |
25fe55af | 3989 | case stop_memory: |
505bde11 | 3990 | p += 1; |
fa9a63c5 RM |
3991 | continue; |
3992 | ||
3993 | ||
3994 | default: | |
25fe55af RS |
3995 | abort (); /* We have listed all the cases. */ |
3996 | } /* switch *p++ */ | |
fa9a63c5 RM |
3997 | |
3998 | /* Getting here means we have found the possible starting | |
25fe55af | 3999 | characters for one path of the pattern -- and that the empty |
01618498 SM |
4000 | string does not match. We need not follow this path further. */ |
4001 | return 0; | |
fa9a63c5 RM |
4002 | } /* while p */ |
4003 | ||
01618498 SM |
4004 | /* We reached the end without matching anything. */ |
4005 | return 1; | |
4006 | ||
f6a3f532 SM |
4007 | } /* analyse_first */ |
4008 | \f | |
4009 | /* re_compile_fastmap computes a ``fastmap'' for the compiled pattern in | |
4010 | BUFP. A fastmap records which of the (1 << BYTEWIDTH) possible | |
4011 | characters can start a string that matches the pattern. This fastmap | |
4012 | is used by re_search to skip quickly over impossible starting points. | |
4013 | ||
4014 | Character codes above (1 << BYTEWIDTH) are not represented in the | |
4015 | fastmap, but the leading codes are represented. Thus, the fastmap | |
4016 | indicates which character sets could start a match. | |
4017 | ||
4018 | The caller must supply the address of a (1 << BYTEWIDTH)-byte data | |
4019 | area as BUFP->fastmap. | |
4020 | ||
4021 | We set the `fastmap', `fastmap_accurate', and `can_be_null' fields in | |
4022 | the pattern buffer. | |
4023 | ||
4024 | Returns 0 if we succeed, -2 if an internal error. */ | |
4025 | ||
4026 | int | |
4027 | re_compile_fastmap (bufp) | |
4028 | struct re_pattern_buffer *bufp; | |
4029 | { | |
4030 | char *fastmap = bufp->fastmap; | |
4031 | int analysis; | |
4032 | ||
4033 | assert (fastmap && bufp->buffer); | |
4034 | ||
4035 | bzero (fastmap, 1 << BYTEWIDTH); /* Assume nothing's valid. */ | |
4036 | bufp->fastmap_accurate = 1; /* It will be when we're done. */ | |
4037 | ||
4038 | analysis = analyse_first (bufp->buffer, bufp->buffer + bufp->used, | |
2d1675e4 | 4039 | fastmap, RE_MULTIBYTE_P (bufp)); |
c0f9ea08 | 4040 | bufp->can_be_null = (analysis != 0); |
fa9a63c5 RM |
4041 | return 0; |
4042 | } /* re_compile_fastmap */ | |
4043 | \f | |
4044 | /* Set REGS to hold NUM_REGS registers, storing them in STARTS and | |
4045 | ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use | |
4046 | this memory for recording register information. STARTS and ENDS | |
4047 | must be allocated using the malloc library routine, and must each | |
4048 | be at least NUM_REGS * sizeof (regoff_t) bytes long. | |
4049 | ||
4050 | If NUM_REGS == 0, then subsequent matches should allocate their own | |
4051 | register data. | |
4052 | ||
4053 | Unless this function is called, the first search or match using | |
4054 | PATTERN_BUFFER will allocate its own register data, without | |
4055 | freeing the old data. */ | |
4056 | ||
4057 | void | |
4058 | re_set_registers (bufp, regs, num_regs, starts, ends) | |
4059 | struct re_pattern_buffer *bufp; | |
4060 | struct re_registers *regs; | |
4061 | unsigned num_regs; | |
4062 | regoff_t *starts, *ends; | |
4063 | { | |
4064 | if (num_regs) | |
4065 | { | |
4066 | bufp->regs_allocated = REGS_REALLOCATE; | |
4067 | regs->num_regs = num_regs; | |
4068 | regs->start = starts; | |
4069 | regs->end = ends; | |
4070 | } | |
4071 | else | |
4072 | { | |
4073 | bufp->regs_allocated = REGS_UNALLOCATED; | |
4074 | regs->num_regs = 0; | |
4075 | regs->start = regs->end = (regoff_t *) 0; | |
4076 | } | |
4077 | } | |
c0f9ea08 | 4078 | WEAK_ALIAS (__re_set_registers, re_set_registers) |
fa9a63c5 | 4079 | \f |
25fe55af | 4080 | /* Searching routines. */ |
fa9a63c5 RM |
4081 | |
4082 | /* Like re_search_2, below, but only one string is specified, and | |
4083 | doesn't let you say where to stop matching. */ | |
4084 | ||
4085 | int | |
4086 | re_search (bufp, string, size, startpos, range, regs) | |
4087 | struct re_pattern_buffer *bufp; | |
4088 | const char *string; | |
4089 | int size, startpos, range; | |
4090 | struct re_registers *regs; | |
4091 | { | |
5e69f11e | 4092 | return re_search_2 (bufp, NULL, 0, string, size, startpos, range, |
fa9a63c5 RM |
4093 | regs, size); |
4094 | } | |
c0f9ea08 | 4095 | WEAK_ALIAS (__re_search, re_search) |
fa9a63c5 | 4096 | |
70806df6 KH |
4097 | /* Head address of virtual concatenation of string. */ |
4098 | #define HEAD_ADDR_VSTRING(P) \ | |
4099 | (((P) >= size1 ? string2 : string1)) | |
4100 | ||
b18215fc RS |
4101 | /* End address of virtual concatenation of string. */ |
4102 | #define STOP_ADDR_VSTRING(P) \ | |
4103 | (((P) >= size1 ? string2 + size2 : string1 + size1)) | |
4104 | ||
4105 | /* Address of POS in the concatenation of virtual string. */ | |
4106 | #define POS_ADDR_VSTRING(POS) \ | |
4107 | (((POS) >= size1 ? string2 - size1 : string1) + (POS)) | |
fa9a63c5 RM |
4108 | |
4109 | /* Using the compiled pattern in BUFP->buffer, first tries to match the | |
4110 | virtual concatenation of STRING1 and STRING2, starting first at index | |
4111 | STARTPOS, then at STARTPOS + 1, and so on. | |
5e69f11e | 4112 | |
fa9a63c5 | 4113 | STRING1 and STRING2 have length SIZE1 and SIZE2, respectively. |
5e69f11e | 4114 | |
fa9a63c5 RM |
4115 | RANGE is how far to scan while trying to match. RANGE = 0 means try |
4116 | only at STARTPOS; in general, the last start tried is STARTPOS + | |
4117 | RANGE. | |
5e69f11e | 4118 | |
fa9a63c5 RM |
4119 | In REGS, return the indices of the virtual concatenation of STRING1 |
4120 | and STRING2 that matched the entire BUFP->buffer and its contained | |
4121 | subexpressions. | |
5e69f11e | 4122 | |
fa9a63c5 RM |
4123 | Do not consider matching one past the index STOP in the virtual |
4124 | concatenation of STRING1 and STRING2. | |
4125 | ||
4126 | We return either the position in the strings at which the match was | |
4127 | found, -1 if no match, or -2 if error (such as failure | |
4128 | stack overflow). */ | |
4129 | ||
4130 | int | |
66f0296e | 4131 | re_search_2 (bufp, str1, size1, str2, size2, startpos, range, regs, stop) |
fa9a63c5 | 4132 | struct re_pattern_buffer *bufp; |
66f0296e | 4133 | const char *str1, *str2; |
fa9a63c5 RM |
4134 | int size1, size2; |
4135 | int startpos; | |
4136 | int range; | |
4137 | struct re_registers *regs; | |
4138 | int stop; | |
4139 | { | |
4140 | int val; | |
66f0296e SM |
4141 | re_char *string1 = (re_char*) str1; |
4142 | re_char *string2 = (re_char*) str2; | |
fa9a63c5 | 4143 | register char *fastmap = bufp->fastmap; |
6676cb1c | 4144 | register RE_TRANSLATE_TYPE translate = bufp->translate; |
fa9a63c5 RM |
4145 | int total_size = size1 + size2; |
4146 | int endpos = startpos + range; | |
c0f9ea08 | 4147 | boolean anchored_start; |
fa9a63c5 | 4148 | |
25fe55af | 4149 | /* Nonzero if we have to concern multibyte character. */ |
2d1675e4 | 4150 | const boolean multibyte = RE_MULTIBYTE_P (bufp); |
b18215fc | 4151 | |
fa9a63c5 RM |
4152 | /* Check for out-of-range STARTPOS. */ |
4153 | if (startpos < 0 || startpos > total_size) | |
4154 | return -1; | |
5e69f11e | 4155 | |
fa9a63c5 | 4156 | /* Fix up RANGE if it might eventually take us outside |
34597fa9 | 4157 | the virtual concatenation of STRING1 and STRING2. |
5e69f11e | 4158 | Make sure we won't move STARTPOS below 0 or above TOTAL_SIZE. */ |
34597fa9 RS |
4159 | if (endpos < 0) |
4160 | range = 0 - startpos; | |
fa9a63c5 RM |
4161 | else if (endpos > total_size) |
4162 | range = total_size - startpos; | |
4163 | ||
4164 | /* If the search isn't to be a backwards one, don't waste time in a | |
7b140fd7 | 4165 | search for a pattern anchored at beginning of buffer. */ |
fa9a63c5 RM |
4166 | if (bufp->used > 0 && (re_opcode_t) bufp->buffer[0] == begbuf && range > 0) |
4167 | { | |
4168 | if (startpos > 0) | |
4169 | return -1; | |
4170 | else | |
7b140fd7 | 4171 | range = 0; |
fa9a63c5 RM |
4172 | } |
4173 | ||
ae4788a8 RS |
4174 | #ifdef emacs |
4175 | /* In a forward search for something that starts with \=. | |
4176 | don't keep searching past point. */ | |
4177 | if (bufp->used > 0 && (re_opcode_t) bufp->buffer[0] == at_dot && range > 0) | |
4178 | { | |
7b140fd7 RS |
4179 | range = PT_BYTE - BEGV_BYTE - startpos; |
4180 | if (range < 0) | |
ae4788a8 RS |
4181 | return -1; |
4182 | } | |
4183 | #endif /* emacs */ | |
4184 | ||
fa9a63c5 RM |
4185 | /* Update the fastmap now if not correct already. */ |
4186 | if (fastmap && !bufp->fastmap_accurate) | |
01618498 | 4187 | re_compile_fastmap (bufp); |
5e69f11e | 4188 | |
c8499ba5 | 4189 | /* See whether the pattern is anchored. */ |
c0f9ea08 | 4190 | anchored_start = (bufp->buffer[0] == begline); |
c8499ba5 | 4191 | |
b18215fc | 4192 | #ifdef emacs |
cc9b4df2 KH |
4193 | gl_state.object = re_match_object; |
4194 | { | |
99633e97 | 4195 | int charpos = SYNTAX_TABLE_BYTE_TO_CHAR (POS_AS_IN_BUFFER (startpos)); |
cc9b4df2 KH |
4196 | |
4197 | SETUP_SYNTAX_TABLE_FOR_OBJECT (re_match_object, charpos, 1); | |
4198 | } | |
b18215fc RS |
4199 | #endif |
4200 | ||
fa9a63c5 RM |
4201 | /* Loop through the string, looking for a place to start matching. */ |
4202 | for (;;) | |
5e69f11e | 4203 | { |
c8499ba5 RS |
4204 | /* If the pattern is anchored, |
4205 | skip quickly past places we cannot match. | |
4206 | We don't bother to treat startpos == 0 specially | |
4207 | because that case doesn't repeat. */ | |
4208 | if (anchored_start && startpos > 0) | |
4209 | { | |
c0f9ea08 SM |
4210 | if (! ((startpos <= size1 ? string1[startpos - 1] |
4211 | : string2[startpos - size1 - 1]) | |
4212 | == '\n')) | |
c8499ba5 RS |
4213 | goto advance; |
4214 | } | |
4215 | ||
fa9a63c5 | 4216 | /* If a fastmap is supplied, skip quickly over characters that |
25fe55af RS |
4217 | cannot be the start of a match. If the pattern can match the |
4218 | null string, however, we don't need to skip characters; we want | |
4219 | the first null string. */ | |
fa9a63c5 RM |
4220 | if (fastmap && startpos < total_size && !bufp->can_be_null) |
4221 | { | |
66f0296e | 4222 | register re_char *d; |
01618498 | 4223 | register re_wchar_t buf_ch; |
e934739e RS |
4224 | |
4225 | d = POS_ADDR_VSTRING (startpos); | |
4226 | ||
25fe55af | 4227 | if (range > 0) /* Searching forwards. */ |
fa9a63c5 | 4228 | { |
fa9a63c5 RM |
4229 | register int lim = 0; |
4230 | int irange = range; | |
4231 | ||
25fe55af RS |
4232 | if (startpos < size1 && startpos + range >= size1) |
4233 | lim = range - (size1 - startpos); | |
fa9a63c5 | 4234 | |
25fe55af RS |
4235 | /* Written out as an if-else to avoid testing `translate' |
4236 | inside the loop. */ | |
28ae27ae AS |
4237 | if (RE_TRANSLATE_P (translate)) |
4238 | { | |
e934739e RS |
4239 | if (multibyte) |
4240 | while (range > lim) | |
4241 | { | |
4242 | int buf_charlen; | |
4243 | ||
4244 | buf_ch = STRING_CHAR_AND_LENGTH (d, range - lim, | |
4245 | buf_charlen); | |
4246 | ||
4247 | buf_ch = RE_TRANSLATE (translate, buf_ch); | |
4248 | if (buf_ch >= 0400 | |
4249 | || fastmap[buf_ch]) | |
4250 | break; | |
4251 | ||
4252 | range -= buf_charlen; | |
4253 | d += buf_charlen; | |
4254 | } | |
4255 | else | |
4256 | while (range > lim | |
66f0296e | 4257 | && !fastmap[RE_TRANSLATE (translate, *d)]) |
33c46939 RS |
4258 | { |
4259 | d++; | |
4260 | range--; | |
4261 | } | |
e934739e | 4262 | } |
fa9a63c5 | 4263 | else |
66f0296e | 4264 | while (range > lim && !fastmap[*d]) |
33c46939 RS |
4265 | { |
4266 | d++; | |
4267 | range--; | |
4268 | } | |
fa9a63c5 RM |
4269 | |
4270 | startpos += irange - range; | |
4271 | } | |
25fe55af | 4272 | else /* Searching backwards. */ |
fa9a63c5 | 4273 | { |
2d1675e4 SM |
4274 | int room = (startpos >= size1 |
4275 | ? size2 + size1 - startpos | |
4276 | : size1 - startpos); | |
4277 | buf_ch = RE_STRING_CHAR (d, room); | |
4278 | buf_ch = TRANSLATE (buf_ch); | |
fa9a63c5 | 4279 | |
e934739e RS |
4280 | if (! (buf_ch >= 0400 |
4281 | || fastmap[buf_ch])) | |
fa9a63c5 RM |
4282 | goto advance; |
4283 | } | |
4284 | } | |
4285 | ||
4286 | /* If can't match the null string, and that's all we have left, fail. */ | |
4287 | if (range >= 0 && startpos == total_size && fastmap | |
25fe55af | 4288 | && !bufp->can_be_null) |
fa9a63c5 RM |
4289 | return -1; |
4290 | ||
4291 | val = re_match_2_internal (bufp, string1, size1, string2, size2, | |
4292 | startpos, regs, stop); | |
4293 | #ifndef REGEX_MALLOC | |
0b32bf0e | 4294 | # ifdef C_ALLOCA |
fa9a63c5 | 4295 | alloca (0); |
0b32bf0e | 4296 | # endif |
fa9a63c5 RM |
4297 | #endif |
4298 | ||
4299 | if (val >= 0) | |
4300 | return startpos; | |
5e69f11e | 4301 | |
fa9a63c5 RM |
4302 | if (val == -2) |
4303 | return -2; | |
4304 | ||
4305 | advance: | |
5e69f11e | 4306 | if (!range) |
25fe55af | 4307 | break; |
5e69f11e | 4308 | else if (range > 0) |
25fe55af | 4309 | { |
b18215fc RS |
4310 | /* Update STARTPOS to the next character boundary. */ |
4311 | if (multibyte) | |
4312 | { | |
66f0296e SM |
4313 | re_char *p = POS_ADDR_VSTRING (startpos); |
4314 | re_char *pend = STOP_ADDR_VSTRING (startpos); | |
b18215fc RS |
4315 | int len = MULTIBYTE_FORM_LENGTH (p, pend - p); |
4316 | ||
4317 | range -= len; | |
4318 | if (range < 0) | |
4319 | break; | |
4320 | startpos += len; | |
4321 | } | |
4322 | else | |
4323 | { | |
b560c397 RS |
4324 | range--; |
4325 | startpos++; | |
4326 | } | |
e318085a | 4327 | } |
fa9a63c5 | 4328 | else |
25fe55af RS |
4329 | { |
4330 | range++; | |
4331 | startpos--; | |
b18215fc RS |
4332 | |
4333 | /* Update STARTPOS to the previous character boundary. */ | |
4334 | if (multibyte) | |
4335 | { | |
70806df6 KH |
4336 | re_char *p = POS_ADDR_VSTRING (startpos) + 1; |
4337 | re_char *p0 = p; | |
4338 | re_char *phead = HEAD_ADDR_VSTRING (startpos); | |
b18215fc RS |
4339 | |
4340 | /* Find the head of multibyte form. */ | |
70806df6 KH |
4341 | PREV_CHAR_BOUNDARY (p, phead); |
4342 | range += p0 - 1 - p; | |
4343 | if (range > 0) | |
4344 | break; | |
b18215fc | 4345 | |
70806df6 | 4346 | startpos -= p0 - 1 - p; |
b18215fc | 4347 | } |
25fe55af | 4348 | } |
fa9a63c5 RM |
4349 | } |
4350 | return -1; | |
4351 | } /* re_search_2 */ | |
c0f9ea08 | 4352 | WEAK_ALIAS (__re_search_2, re_search_2) |
fa9a63c5 RM |
4353 | \f |
4354 | /* Declarations and macros for re_match_2. */ | |
4355 | ||
2d1675e4 SM |
4356 | static int bcmp_translate _RE_ARGS((re_char *s1, re_char *s2, |
4357 | register int len, | |
4358 | RE_TRANSLATE_TYPE translate, | |
4359 | const int multibyte)); | |
fa9a63c5 RM |
4360 | |
4361 | /* This converts PTR, a pointer into one of the search strings `string1' | |
4362 | and `string2' into an offset from the beginning of that string. */ | |
4363 | #define POINTER_TO_OFFSET(ptr) \ | |
4364 | (FIRST_STRING_P (ptr) \ | |
4365 | ? ((regoff_t) ((ptr) - string1)) \ | |
4366 | : ((regoff_t) ((ptr) - string2 + size1))) | |
4367 | ||
fa9a63c5 | 4368 | /* Call before fetching a character with *d. This switches over to |
419d1c74 SM |
4369 | string2 if necessary. |
4370 | Check re_match_2_internal for a discussion of why end_match_2 might | |
4371 | not be within string2 (but be equal to end_match_1 instead). */ | |
fa9a63c5 | 4372 | #define PREFETCH() \ |
25fe55af | 4373 | while (d == dend) \ |
fa9a63c5 RM |
4374 | { \ |
4375 | /* End of string2 => fail. */ \ | |
25fe55af RS |
4376 | if (dend == end_match_2) \ |
4377 | goto fail; \ | |
4bb91c68 | 4378 | /* End of string1 => advance to string2. */ \ |
25fe55af | 4379 | d = string2; \ |
fa9a63c5 RM |
4380 | dend = end_match_2; \ |
4381 | } | |
4382 | ||
f1ad044f SM |
4383 | /* Call before fetching a char with *d if you already checked other limits. |
4384 | This is meant for use in lookahead operations like wordend, etc.. | |
4385 | where we might need to look at parts of the string that might be | |
4386 | outside of the LIMITs (i.e past `stop'). */ | |
4387 | #define PREFETCH_NOLIMIT() \ | |
4388 | if (d == end1) \ | |
4389 | { \ | |
4390 | d = string2; \ | |
4391 | dend = end_match_2; \ | |
4392 | } \ | |
fa9a63c5 RM |
4393 | |
4394 | /* Test if at very beginning or at very end of the virtual concatenation | |
25fe55af | 4395 | of `string1' and `string2'. If only one string, it's `string2'. */ |
fa9a63c5 | 4396 | #define AT_STRINGS_BEG(d) ((d) == (size1 ? string1 : string2) || !size2) |
5e69f11e | 4397 | #define AT_STRINGS_END(d) ((d) == end2) |
fa9a63c5 RM |
4398 | |
4399 | ||
4400 | /* Test if D points to a character which is word-constituent. We have | |
4401 | two special cases to check for: if past the end of string1, look at | |
4402 | the first character in string2; and if before the beginning of | |
4403 | string2, look at the last character in string1. */ | |
4404 | #define WORDCHAR_P(d) \ | |
4405 | (SYNTAX ((d) == end1 ? *string2 \ | |
25fe55af | 4406 | : (d) == string2 - 1 ? *(end1 - 1) : *(d)) \ |
fa9a63c5 RM |
4407 | == Sword) |
4408 | ||
9121ca40 | 4409 | /* Disabled due to a compiler bug -- see comment at case wordbound */ |
b18215fc RS |
4410 | |
4411 | /* The comment at case wordbound is following one, but we don't use | |
4412 | AT_WORD_BOUNDARY anymore to support multibyte form. | |
4413 | ||
4414 | The DEC Alpha C compiler 3.x generates incorrect code for the | |
25fe55af RS |
4415 | test WORDCHAR_P (d - 1) != WORDCHAR_P (d) in the expansion of |
4416 | AT_WORD_BOUNDARY, so this code is disabled. Expanding the | |
b18215fc RS |
4417 | macro and introducing temporary variables works around the bug. */ |
4418 | ||
9121ca40 | 4419 | #if 0 |
fa9a63c5 RM |
4420 | /* Test if the character before D and the one at D differ with respect |
4421 | to being word-constituent. */ | |
4422 | #define AT_WORD_BOUNDARY(d) \ | |
4423 | (AT_STRINGS_BEG (d) || AT_STRINGS_END (d) \ | |
4424 | || WORDCHAR_P (d - 1) != WORDCHAR_P (d)) | |
9121ca40 | 4425 | #endif |
fa9a63c5 RM |
4426 | |
4427 | /* Free everything we malloc. */ | |
4428 | #ifdef MATCH_MAY_ALLOCATE | |
0b32bf0e SM |
4429 | # define FREE_VAR(var) if (var) { REGEX_FREE (var); var = NULL; } else |
4430 | # define FREE_VARIABLES() \ | |
fa9a63c5 RM |
4431 | do { \ |
4432 | REGEX_FREE_STACK (fail_stack.stack); \ | |
4433 | FREE_VAR (regstart); \ | |
4434 | FREE_VAR (regend); \ | |
fa9a63c5 RM |
4435 | FREE_VAR (best_regstart); \ |
4436 | FREE_VAR (best_regend); \ | |
fa9a63c5 RM |
4437 | } while (0) |
4438 | #else | |
0b32bf0e | 4439 | # define FREE_VARIABLES() ((void)0) /* Do nothing! But inhibit gcc warning. */ |
fa9a63c5 RM |
4440 | #endif /* not MATCH_MAY_ALLOCATE */ |
4441 | ||
505bde11 SM |
4442 | \f |
4443 | /* Optimization routines. */ | |
4444 | ||
4e8a9132 SM |
4445 | /* If the operation is a match against one or more chars, |
4446 | return a pointer to the next operation, else return NULL. */ | |
01618498 | 4447 | static re_char * |
4e8a9132 | 4448 | skip_one_char (p) |
01618498 | 4449 | re_char *p; |
4e8a9132 SM |
4450 | { |
4451 | switch (SWITCH_ENUM_CAST (*p++)) | |
4452 | { | |
4453 | case anychar: | |
4454 | break; | |
177c0ea7 | 4455 | |
4e8a9132 SM |
4456 | case exactn: |
4457 | p += *p + 1; | |
4458 | break; | |
4459 | ||
4460 | case charset_not: | |
4461 | case charset: | |
4462 | if (CHARSET_RANGE_TABLE_EXISTS_P (p - 1)) | |
4463 | { | |
4464 | int mcnt; | |
4465 | p = CHARSET_RANGE_TABLE (p - 1); | |
4466 | EXTRACT_NUMBER_AND_INCR (mcnt, p); | |
4467 | p = CHARSET_RANGE_TABLE_END (p, mcnt); | |
4468 | } | |
4469 | else | |
4470 | p += 1 + CHARSET_BITMAP_SIZE (p - 1); | |
4471 | break; | |
177c0ea7 | 4472 | |
4e8a9132 SM |
4473 | case syntaxspec: |
4474 | case notsyntaxspec: | |
1fb352e0 | 4475 | #ifdef emacs |
4e8a9132 SM |
4476 | case categoryspec: |
4477 | case notcategoryspec: | |
4478 | #endif /* emacs */ | |
4479 | p++; | |
4480 | break; | |
4481 | ||
4482 | default: | |
4483 | p = NULL; | |
4484 | } | |
4485 | return p; | |
4486 | } | |
4487 | ||
4488 | ||
505bde11 | 4489 | /* Jump over non-matching operations. */ |
275a3409 | 4490 | static re_char * |
4e8a9132 | 4491 | skip_noops (p, pend) |
275a3409 | 4492 | re_char *p, *pend; |
505bde11 SM |
4493 | { |
4494 | int mcnt; | |
4495 | while (p < pend) | |
4496 | { | |
4497 | switch (SWITCH_ENUM_CAST ((re_opcode_t) *p)) | |
4498 | { | |
4499 | case start_memory: | |
505bde11 SM |
4500 | case stop_memory: |
4501 | p += 2; break; | |
4502 | case no_op: | |
4503 | p += 1; break; | |
4504 | case jump: | |
4505 | p += 1; | |
4506 | EXTRACT_NUMBER_AND_INCR (mcnt, p); | |
4507 | p += mcnt; | |
4508 | break; | |
4509 | default: | |
4510 | return p; | |
4511 | } | |
4512 | } | |
4513 | assert (p == pend); | |
4514 | return p; | |
4515 | } | |
4516 | ||
4517 | /* Non-zero if "p1 matches something" implies "p2 fails". */ | |
4518 | static int | |
4519 | mutually_exclusive_p (bufp, p1, p2) | |
4520 | struct re_pattern_buffer *bufp; | |
275a3409 | 4521 | re_char *p1, *p2; |
505bde11 | 4522 | { |
4e8a9132 | 4523 | re_opcode_t op2; |
2d1675e4 | 4524 | const boolean multibyte = RE_MULTIBYTE_P (bufp); |
505bde11 SM |
4525 | unsigned char *pend = bufp->buffer + bufp->used; |
4526 | ||
4e8a9132 | 4527 | assert (p1 >= bufp->buffer && p1 < pend |
505bde11 SM |
4528 | && p2 >= bufp->buffer && p2 <= pend); |
4529 | ||
4530 | /* Skip over open/close-group commands. | |
4531 | If what follows this loop is a ...+ construct, | |
4532 | look at what begins its body, since we will have to | |
4533 | match at least one of that. */ | |
4e8a9132 SM |
4534 | p2 = skip_noops (p2, pend); |
4535 | /* The same skip can be done for p1, except that this function | |
4536 | is only used in the case where p1 is a simple match operator. */ | |
4537 | /* p1 = skip_noops (p1, pend); */ | |
4538 | ||
4539 | assert (p1 >= bufp->buffer && p1 < pend | |
4540 | && p2 >= bufp->buffer && p2 <= pend); | |
4541 | ||
4542 | op2 = p2 == pend ? succeed : *p2; | |
4543 | ||
4544 | switch (SWITCH_ENUM_CAST (op2)) | |
505bde11 | 4545 | { |
4e8a9132 SM |
4546 | case succeed: |
4547 | case endbuf: | |
4548 | /* If we're at the end of the pattern, we can change. */ | |
4549 | if (skip_one_char (p1)) | |
505bde11 | 4550 | { |
505bde11 SM |
4551 | DEBUG_PRINT1 (" End of pattern: fast loop.\n"); |
4552 | return 1; | |
505bde11 | 4553 | } |
4e8a9132 | 4554 | break; |
177c0ea7 | 4555 | |
4e8a9132 | 4556 | case endline: |
4e8a9132 SM |
4557 | case exactn: |
4558 | { | |
01618498 | 4559 | register re_wchar_t c |
4e8a9132 | 4560 | = (re_opcode_t) *p2 == endline ? '\n' |
411e4203 | 4561 | : RE_STRING_CHAR (p2 + 2, pend - p2 - 2); |
505bde11 | 4562 | |
4e8a9132 SM |
4563 | if ((re_opcode_t) *p1 == exactn) |
4564 | { | |
4565 | if (c != RE_STRING_CHAR (p1 + 2, pend - p1 - 2)) | |
4566 | { | |
4567 | DEBUG_PRINT3 (" '%c' != '%c' => fast loop.\n", c, p1[2]); | |
4568 | return 1; | |
4569 | } | |
4570 | } | |
505bde11 | 4571 | |
4e8a9132 SM |
4572 | else if ((re_opcode_t) *p1 == charset |
4573 | || (re_opcode_t) *p1 == charset_not) | |
4574 | { | |
4575 | int not = (re_opcode_t) *p1 == charset_not; | |
505bde11 | 4576 | |
4e8a9132 SM |
4577 | /* Test if C is listed in charset (or charset_not) |
4578 | at `p1'. */ | |
4579 | if (SINGLE_BYTE_CHAR_P (c)) | |
4580 | { | |
4581 | if (c < CHARSET_BITMAP_SIZE (p1) * BYTEWIDTH | |
4582 | && p1[2 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH))) | |
4583 | not = !not; | |
4584 | } | |
4585 | else if (CHARSET_RANGE_TABLE_EXISTS_P (p1)) | |
4586 | CHARSET_LOOKUP_RANGE_TABLE (not, c, p1); | |
505bde11 | 4587 | |
4e8a9132 SM |
4588 | /* `not' is equal to 1 if c would match, which means |
4589 | that we can't change to pop_failure_jump. */ | |
4590 | if (!not) | |
4591 | { | |
4592 | DEBUG_PRINT1 (" No match => fast loop.\n"); | |
4593 | return 1; | |
4594 | } | |
4595 | } | |
4596 | else if ((re_opcode_t) *p1 == anychar | |
4597 | && c == '\n') | |
4598 | { | |
4599 | DEBUG_PRINT1 (" . != \\n => fast loop.\n"); | |
4600 | return 1; | |
4601 | } | |
4602 | } | |
4603 | break; | |
505bde11 | 4604 | |
4e8a9132 | 4605 | case charset: |
4e8a9132 SM |
4606 | { |
4607 | if ((re_opcode_t) *p1 == exactn) | |
4608 | /* Reuse the code above. */ | |
4609 | return mutually_exclusive_p (bufp, p2, p1); | |
505bde11 | 4610 | |
505bde11 SM |
4611 | /* It is hard to list up all the character in charset |
4612 | P2 if it includes multibyte character. Give up in | |
4613 | such case. */ | |
4614 | else if (!multibyte || !CHARSET_RANGE_TABLE_EXISTS_P (p2)) | |
4615 | { | |
4616 | /* Now, we are sure that P2 has no range table. | |
4617 | So, for the size of bitmap in P2, `p2[1]' is | |
4618 | enough. But P1 may have range table, so the | |
4619 | size of bitmap table of P1 is extracted by | |
4620 | using macro `CHARSET_BITMAP_SIZE'. | |
4621 | ||
4622 | Since we know that all the character listed in | |
4623 | P2 is ASCII, it is enough to test only bitmap | |
4624 | table of P1. */ | |
4625 | ||
411e4203 | 4626 | if ((re_opcode_t) *p1 == charset) |
505bde11 SM |
4627 | { |
4628 | int idx; | |
4629 | /* We win if the charset inside the loop | |
4630 | has no overlap with the one after the loop. */ | |
4631 | for (idx = 0; | |
4632 | (idx < (int) p2[1] | |
4633 | && idx < CHARSET_BITMAP_SIZE (p1)); | |
4634 | idx++) | |
4635 | if ((p2[2 + idx] & p1[2 + idx]) != 0) | |
4636 | break; | |
4637 | ||
4638 | if (idx == p2[1] | |
4639 | || idx == CHARSET_BITMAP_SIZE (p1)) | |
4640 | { | |
4641 | DEBUG_PRINT1 (" No match => fast loop.\n"); | |
4642 | return 1; | |
4643 | } | |
4644 | } | |
411e4203 | 4645 | else if ((re_opcode_t) *p1 == charset_not) |
505bde11 SM |
4646 | { |
4647 | int idx; | |
4648 | /* We win if the charset_not inside the loop lists | |
4649 | every character listed in the charset after. */ | |
4650 | for (idx = 0; idx < (int) p2[1]; idx++) | |
4651 | if (! (p2[2 + idx] == 0 | |
4652 | || (idx < CHARSET_BITMAP_SIZE (p1) | |
4653 | && ((p2[2 + idx] & ~ p1[2 + idx]) == 0)))) | |
4654 | break; | |
4655 | ||
4e8a9132 SM |
4656 | if (idx == p2[1]) |
4657 | { | |
4658 | DEBUG_PRINT1 (" No match => fast loop.\n"); | |
4659 | return 1; | |
4660 | } | |
4661 | } | |
4662 | } | |
4663 | } | |
609b757a | 4664 | break; |
177c0ea7 | 4665 | |
411e4203 SM |
4666 | case charset_not: |
4667 | switch (SWITCH_ENUM_CAST (*p1)) | |
4668 | { | |
4669 | case exactn: | |
4670 | case charset: | |
4671 | /* Reuse the code above. */ | |
4672 | return mutually_exclusive_p (bufp, p2, p1); | |
4673 | case charset_not: | |
4674 | /* When we have two charset_not, it's very unlikely that | |
4675 | they don't overlap. The union of the two sets of excluded | |
4676 | chars should cover all possible chars, which, as a matter of | |
4677 | fact, is virtually impossible in multibyte buffers. */ | |
36595814 | 4678 | break; |
411e4203 SM |
4679 | } |
4680 | break; | |
4681 | ||
4e8a9132 | 4682 | case wordend: |
669fa600 SM |
4683 | return ((re_opcode_t) *p1 == syntaxspec && p1[1] == Sword); |
4684 | case symend: | |
4e8a9132 | 4685 | return ((re_opcode_t) *p1 == syntaxspec |
669fa600 SM |
4686 | && (p1[1] == Ssymbol || p1[1] == Sword)); |
4687 | case notsyntaxspec: | |
4688 | return ((re_opcode_t) *p1 == syntaxspec && p1[1] == p2[1]); | |
4e8a9132 SM |
4689 | |
4690 | case wordbeg: | |
669fa600 SM |
4691 | return ((re_opcode_t) *p1 == notsyntaxspec && p1[1] == Sword); |
4692 | case symbeg: | |
4e8a9132 | 4693 | return ((re_opcode_t) *p1 == notsyntaxspec |
669fa600 SM |
4694 | && (p1[1] == Ssymbol || p1[1] == Sword)); |
4695 | case syntaxspec: | |
4696 | return ((re_opcode_t) *p1 == notsyntaxspec && p1[1] == p2[1]); | |
4e8a9132 SM |
4697 | |
4698 | case wordbound: | |
4699 | return (((re_opcode_t) *p1 == notsyntaxspec | |
4700 | || (re_opcode_t) *p1 == syntaxspec) | |
4701 | && p1[1] == Sword); | |
4702 | ||
1fb352e0 | 4703 | #ifdef emacs |
4e8a9132 SM |
4704 | case categoryspec: |
4705 | return ((re_opcode_t) *p1 == notcategoryspec && p1[1] == p2[1]); | |
4706 | case notcategoryspec: | |
4707 | return ((re_opcode_t) *p1 == categoryspec && p1[1] == p2[1]); | |
4708 | #endif /* emacs */ | |
4709 | ||
4710 | default: | |
4711 | ; | |
505bde11 SM |
4712 | } |
4713 | ||
4714 | /* Safe default. */ | |
4715 | return 0; | |
4716 | } | |
4717 | ||
fa9a63c5 RM |
4718 | \f |
4719 | /* Matching routines. */ | |
4720 | ||
25fe55af | 4721 | #ifndef emacs /* Emacs never uses this. */ |
fa9a63c5 RM |
4722 | /* re_match is like re_match_2 except it takes only a single string. */ |
4723 | ||
4724 | int | |
4725 | re_match (bufp, string, size, pos, regs) | |
4726 | struct re_pattern_buffer *bufp; | |
4727 | const char *string; | |
4728 | int size, pos; | |
4729 | struct re_registers *regs; | |
4730 | { | |
4bb91c68 | 4731 | int result = re_match_2_internal (bufp, NULL, 0, (re_char*) string, size, |
fa9a63c5 | 4732 | pos, regs, size); |
0b32bf0e | 4733 | # if defined C_ALLOCA && !defined REGEX_MALLOC |
fa9a63c5 | 4734 | alloca (0); |
0b32bf0e | 4735 | # endif |
fa9a63c5 RM |
4736 | return result; |
4737 | } | |
c0f9ea08 | 4738 | WEAK_ALIAS (__re_match, re_match) |
fa9a63c5 RM |
4739 | #endif /* not emacs */ |
4740 | ||
b18215fc RS |
4741 | #ifdef emacs |
4742 | /* In Emacs, this is the string or buffer in which we | |
25fe55af | 4743 | are matching. It is used for looking up syntax properties. */ |
b18215fc RS |
4744 | Lisp_Object re_match_object; |
4745 | #endif | |
fa9a63c5 RM |
4746 | |
4747 | /* re_match_2 matches the compiled pattern in BUFP against the | |
4748 | the (virtual) concatenation of STRING1 and STRING2 (of length SIZE1 | |
4749 | and SIZE2, respectively). We start matching at POS, and stop | |
4750 | matching at STOP. | |
5e69f11e | 4751 | |
fa9a63c5 | 4752 | If REGS is non-null and the `no_sub' field of BUFP is nonzero, we |
25fe55af | 4753 | store offsets for the substring each group matched in REGS. See the |
fa9a63c5 RM |
4754 | documentation for exactly how many groups we fill. |
4755 | ||
4756 | We return -1 if no match, -2 if an internal error (such as the | |
25fe55af | 4757 | failure stack overflowing). Otherwise, we return the length of the |
fa9a63c5 RM |
4758 | matched substring. */ |
4759 | ||
4760 | int | |
4761 | re_match_2 (bufp, string1, size1, string2, size2, pos, regs, stop) | |
4762 | struct re_pattern_buffer *bufp; | |
4763 | const char *string1, *string2; | |
4764 | int size1, size2; | |
4765 | int pos; | |
4766 | struct re_registers *regs; | |
4767 | int stop; | |
4768 | { | |
b18215fc | 4769 | int result; |
25fe55af | 4770 | |
b18215fc | 4771 | #ifdef emacs |
cc9b4df2 KH |
4772 | int charpos; |
4773 | gl_state.object = re_match_object; | |
99633e97 | 4774 | charpos = SYNTAX_TABLE_BYTE_TO_CHAR (POS_AS_IN_BUFFER (pos)); |
cc9b4df2 | 4775 | SETUP_SYNTAX_TABLE_FOR_OBJECT (re_match_object, charpos, 1); |
b18215fc RS |
4776 | #endif |
4777 | ||
4bb91c68 SM |
4778 | result = re_match_2_internal (bufp, (re_char*) string1, size1, |
4779 | (re_char*) string2, size2, | |
cc9b4df2 | 4780 | pos, regs, stop); |
0b32bf0e | 4781 | #if defined C_ALLOCA && !defined REGEX_MALLOC |
fa9a63c5 | 4782 | alloca (0); |
a60198e5 | 4783 | #endif |
fa9a63c5 RM |
4784 | return result; |
4785 | } | |
c0f9ea08 | 4786 | WEAK_ALIAS (__re_match_2, re_match_2) |
fa9a63c5 RM |
4787 | |
4788 | /* This is a separate function so that we can force an alloca cleanup | |
25fe55af | 4789 | afterwards. */ |
fa9a63c5 RM |
4790 | static int |
4791 | re_match_2_internal (bufp, string1, size1, string2, size2, pos, regs, stop) | |
4792 | struct re_pattern_buffer *bufp; | |
66f0296e | 4793 | re_char *string1, *string2; |
fa9a63c5 RM |
4794 | int size1, size2; |
4795 | int pos; | |
4796 | struct re_registers *regs; | |
4797 | int stop; | |
4798 | { | |
4799 | /* General temporaries. */ | |
4800 | int mcnt; | |
01618498 | 4801 | size_t reg; |
66f0296e | 4802 | boolean not; |
fa9a63c5 RM |
4803 | |
4804 | /* Just past the end of the corresponding string. */ | |
66f0296e | 4805 | re_char *end1, *end2; |
fa9a63c5 RM |
4806 | |
4807 | /* Pointers into string1 and string2, just past the last characters in | |
25fe55af | 4808 | each to consider matching. */ |
66f0296e | 4809 | re_char *end_match_1, *end_match_2; |
fa9a63c5 RM |
4810 | |
4811 | /* Where we are in the data, and the end of the current string. */ | |
66f0296e | 4812 | re_char *d, *dend; |
5e69f11e | 4813 | |
99633e97 SM |
4814 | /* Used sometimes to remember where we were before starting matching |
4815 | an operator so that we can go back in case of failure. This "atomic" | |
4816 | behavior of matching opcodes is indispensable to the correctness | |
4817 | of the on_failure_keep_string_jump optimization. */ | |
4818 | re_char *dfail; | |
4819 | ||
fa9a63c5 | 4820 | /* Where we are in the pattern, and the end of the pattern. */ |
01618498 SM |
4821 | re_char *p = bufp->buffer; |
4822 | re_char *pend = p + bufp->used; | |
fa9a63c5 | 4823 | |
25fe55af | 4824 | /* We use this to map every character in the string. */ |
6676cb1c | 4825 | RE_TRANSLATE_TYPE translate = bufp->translate; |
fa9a63c5 | 4826 | |
25fe55af | 4827 | /* Nonzero if we have to concern multibyte character. */ |
2d1675e4 | 4828 | const boolean multibyte = RE_MULTIBYTE_P (bufp); |
b18215fc | 4829 | |
fa9a63c5 RM |
4830 | /* Failure point stack. Each place that can handle a failure further |
4831 | down the line pushes a failure point on this stack. It consists of | |
505bde11 | 4832 | regstart, and regend for all registers corresponding to |
fa9a63c5 RM |
4833 | the subexpressions we're currently inside, plus the number of such |
4834 | registers, and, finally, two char *'s. The first char * is where | |
4835 | to resume scanning the pattern; the second one is where to resume | |
505bde11 | 4836 | scanning the strings. */ |
25fe55af | 4837 | #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */ |
fa9a63c5 RM |
4838 | fail_stack_type fail_stack; |
4839 | #endif | |
4840 | #ifdef DEBUG | |
fa9a63c5 RM |
4841 | unsigned nfailure_points_pushed = 0, nfailure_points_popped = 0; |
4842 | #endif | |
4843 | ||
0b32bf0e | 4844 | #if defined REL_ALLOC && defined REGEX_MALLOC |
fa9a63c5 RM |
4845 | /* This holds the pointer to the failure stack, when |
4846 | it is allocated relocatably. */ | |
4847 | fail_stack_elt_t *failure_stack_ptr; | |
99633e97 | 4848 | #endif |
fa9a63c5 RM |
4849 | |
4850 | /* We fill all the registers internally, independent of what we | |
25fe55af | 4851 | return, for use in backreferences. The number here includes |
fa9a63c5 | 4852 | an element for register zero. */ |
4bb91c68 | 4853 | size_t num_regs = bufp->re_nsub + 1; |
5e69f11e | 4854 | |
fa9a63c5 RM |
4855 | /* Information on the contents of registers. These are pointers into |
4856 | the input strings; they record just what was matched (on this | |
4857 | attempt) by a subexpression part of the pattern, that is, the | |
4858 | regnum-th regstart pointer points to where in the pattern we began | |
4859 | matching and the regnum-th regend points to right after where we | |
4860 | stopped matching the regnum-th subexpression. (The zeroth register | |
4861 | keeps track of what the whole pattern matches.) */ | |
4862 | #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */ | |
66f0296e | 4863 | re_char **regstart, **regend; |
fa9a63c5 RM |
4864 | #endif |
4865 | ||
fa9a63c5 | 4866 | /* The following record the register info as found in the above |
5e69f11e | 4867 | variables when we find a match better than any we've seen before. |
fa9a63c5 RM |
4868 | This happens as we backtrack through the failure points, which in |
4869 | turn happens only if we have not yet matched the entire string. */ | |
4870 | unsigned best_regs_set = false; | |
4871 | #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */ | |
66f0296e | 4872 | re_char **best_regstart, **best_regend; |
fa9a63c5 | 4873 | #endif |
5e69f11e | 4874 | |
fa9a63c5 RM |
4875 | /* Logically, this is `best_regend[0]'. But we don't want to have to |
4876 | allocate space for that if we're not allocating space for anything | |
25fe55af | 4877 | else (see below). Also, we never need info about register 0 for |
fa9a63c5 RM |
4878 | any of the other register vectors, and it seems rather a kludge to |
4879 | treat `best_regend' differently than the rest. So we keep track of | |
4880 | the end of the best match so far in a separate variable. We | |
4881 | initialize this to NULL so that when we backtrack the first time | |
4882 | and need to test it, it's not garbage. */ | |
66f0296e | 4883 | re_char *match_end = NULL; |
fa9a63c5 | 4884 | |
fa9a63c5 RM |
4885 | #ifdef DEBUG |
4886 | /* Counts the total number of registers pushed. */ | |
5e69f11e | 4887 | unsigned num_regs_pushed = 0; |
fa9a63c5 RM |
4888 | #endif |
4889 | ||
4890 | DEBUG_PRINT1 ("\n\nEntering re_match_2.\n"); | |
5e69f11e | 4891 | |
fa9a63c5 | 4892 | INIT_FAIL_STACK (); |
5e69f11e | 4893 | |
fa9a63c5 RM |
4894 | #ifdef MATCH_MAY_ALLOCATE |
4895 | /* Do not bother to initialize all the register variables if there are | |
4896 | no groups in the pattern, as it takes a fair amount of time. If | |
4897 | there are groups, we include space for register 0 (the whole | |
4898 | pattern), even though we never use it, since it simplifies the | |
4899 | array indexing. We should fix this. */ | |
4900 | if (bufp->re_nsub) | |
4901 | { | |
66f0296e SM |
4902 | regstart = REGEX_TALLOC (num_regs, re_char *); |
4903 | regend = REGEX_TALLOC (num_regs, re_char *); | |
4904 | best_regstart = REGEX_TALLOC (num_regs, re_char *); | |
4905 | best_regend = REGEX_TALLOC (num_regs, re_char *); | |
fa9a63c5 | 4906 | |
505bde11 | 4907 | if (!(regstart && regend && best_regstart && best_regend)) |
25fe55af RS |
4908 | { |
4909 | FREE_VARIABLES (); | |
4910 | return -2; | |
4911 | } | |
fa9a63c5 RM |
4912 | } |
4913 | else | |
4914 | { | |
4915 | /* We must initialize all our variables to NULL, so that | |
25fe55af | 4916 | `FREE_VARIABLES' doesn't try to free them. */ |
505bde11 | 4917 | regstart = regend = best_regstart = best_regend = NULL; |
fa9a63c5 RM |
4918 | } |
4919 | #endif /* MATCH_MAY_ALLOCATE */ | |
4920 | ||
4921 | /* The starting position is bogus. */ | |
4922 | if (pos < 0 || pos > size1 + size2) | |
4923 | { | |
4924 | FREE_VARIABLES (); | |
4925 | return -1; | |
4926 | } | |
5e69f11e | 4927 | |
fa9a63c5 RM |
4928 | /* Initialize subexpression text positions to -1 to mark ones that no |
4929 | start_memory/stop_memory has been seen for. Also initialize the | |
4930 | register information struct. */ | |
01618498 SM |
4931 | for (reg = 1; reg < num_regs; reg++) |
4932 | regstart[reg] = regend[reg] = NULL; | |
99633e97 | 4933 | |
fa9a63c5 | 4934 | /* We move `string1' into `string2' if the latter's empty -- but not if |
25fe55af | 4935 | `string1' is null. */ |
fa9a63c5 RM |
4936 | if (size2 == 0 && string1 != NULL) |
4937 | { | |
4938 | string2 = string1; | |
4939 | size2 = size1; | |
4940 | string1 = 0; | |
4941 | size1 = 0; | |
4942 | } | |
4943 | end1 = string1 + size1; | |
4944 | end2 = string2 + size2; | |
4945 | ||
5e69f11e | 4946 | /* `p' scans through the pattern as `d' scans through the data. |
fa9a63c5 RM |
4947 | `dend' is the end of the input string that `d' points within. `d' |
4948 | is advanced into the following input string whenever necessary, but | |
4949 | this happens before fetching; therefore, at the beginning of the | |
4950 | loop, `d' can be pointing at the end of a string, but it cannot | |
4951 | equal `string2'. */ | |
419d1c74 | 4952 | if (pos >= size1) |
fa9a63c5 | 4953 | { |
419d1c74 SM |
4954 | /* Only match within string2. */ |
4955 | d = string2 + pos - size1; | |
4956 | dend = end_match_2 = string2 + stop - size1; | |
4957 | end_match_1 = end1; /* Just to give it a value. */ | |
fa9a63c5 RM |
4958 | } |
4959 | else | |
4960 | { | |
f1ad044f | 4961 | if (stop < size1) |
419d1c74 SM |
4962 | { |
4963 | /* Only match within string1. */ | |
4964 | end_match_1 = string1 + stop; | |
4965 | /* BEWARE! | |
4966 | When we reach end_match_1, PREFETCH normally switches to string2. | |
4967 | But in the present case, this means that just doing a PREFETCH | |
4968 | makes us jump from `stop' to `gap' within the string. | |
4969 | What we really want here is for the search to stop as | |
4970 | soon as we hit end_match_1. That's why we set end_match_2 | |
4971 | to end_match_1 (since PREFETCH fails as soon as we hit | |
4972 | end_match_2). */ | |
4973 | end_match_2 = end_match_1; | |
4974 | } | |
4975 | else | |
f1ad044f SM |
4976 | { /* It's important to use this code when stop == size so that |
4977 | moving `d' from end1 to string2 will not prevent the d == dend | |
4978 | check from catching the end of string. */ | |
419d1c74 SM |
4979 | end_match_1 = end1; |
4980 | end_match_2 = string2 + stop - size1; | |
4981 | } | |
4982 | d = string1 + pos; | |
4983 | dend = end_match_1; | |
fa9a63c5 RM |
4984 | } |
4985 | ||
4986 | DEBUG_PRINT1 ("The compiled pattern is: "); | |
4987 | DEBUG_PRINT_COMPILED_PATTERN (bufp, p, pend); | |
4988 | DEBUG_PRINT1 ("The string to match is: `"); | |
4989 | DEBUG_PRINT_DOUBLE_STRING (d, string1, size1, string2, size2); | |
4990 | DEBUG_PRINT1 ("'\n"); | |
5e69f11e | 4991 | |
25fe55af | 4992 | /* This loops over pattern commands. It exits by returning from the |
fa9a63c5 RM |
4993 | function if the match is complete, or it drops through if the match |
4994 | fails at this starting point in the input data. */ | |
4995 | for (;;) | |
4996 | { | |
505bde11 | 4997 | DEBUG_PRINT2 ("\n%p: ", p); |
fa9a63c5 RM |
4998 | |
4999 | if (p == pend) | |
5000 | { /* End of pattern means we might have succeeded. */ | |
25fe55af | 5001 | DEBUG_PRINT1 ("end of pattern ... "); |
5e69f11e | 5002 | |
fa9a63c5 | 5003 | /* If we haven't matched the entire string, and we want the |
25fe55af RS |
5004 | longest match, try backtracking. */ |
5005 | if (d != end_match_2) | |
fa9a63c5 RM |
5006 | { |
5007 | /* 1 if this match ends in the same string (string1 or string2) | |
5008 | as the best previous match. */ | |
5e69f11e | 5009 | boolean same_str_p = (FIRST_STRING_P (match_end) |
99633e97 | 5010 | == FIRST_STRING_P (d)); |
fa9a63c5 RM |
5011 | /* 1 if this match is the best seen so far. */ |
5012 | boolean best_match_p; | |
5013 | ||
5014 | /* AIX compiler got confused when this was combined | |
25fe55af | 5015 | with the previous declaration. */ |
fa9a63c5 RM |
5016 | if (same_str_p) |
5017 | best_match_p = d > match_end; | |
5018 | else | |
99633e97 | 5019 | best_match_p = !FIRST_STRING_P (d); |
fa9a63c5 | 5020 | |
25fe55af RS |
5021 | DEBUG_PRINT1 ("backtracking.\n"); |
5022 | ||
5023 | if (!FAIL_STACK_EMPTY ()) | |
5024 | { /* More failure points to try. */ | |
5025 | ||
5026 | /* If exceeds best match so far, save it. */ | |
5027 | if (!best_regs_set || best_match_p) | |
5028 | { | |
5029 | best_regs_set = true; | |
5030 | match_end = d; | |
5031 | ||
5032 | DEBUG_PRINT1 ("\nSAVING match as best so far.\n"); | |
5033 | ||
01618498 | 5034 | for (reg = 1; reg < num_regs; reg++) |
25fe55af | 5035 | { |
01618498 SM |
5036 | best_regstart[reg] = regstart[reg]; |
5037 | best_regend[reg] = regend[reg]; | |
25fe55af RS |
5038 | } |
5039 | } | |
5040 | goto fail; | |
5041 | } | |
5042 | ||
5043 | /* If no failure points, don't restore garbage. And if | |
5044 | last match is real best match, don't restore second | |
5045 | best one. */ | |
5046 | else if (best_regs_set && !best_match_p) | |
5047 | { | |
5048 | restore_best_regs: | |
5049 | /* Restore best match. It may happen that `dend == | |
5050 | end_match_1' while the restored d is in string2. | |
5051 | For example, the pattern `x.*y.*z' against the | |
5052 | strings `x-' and `y-z-', if the two strings are | |
5053 | not consecutive in memory. */ | |
5054 | DEBUG_PRINT1 ("Restoring best registers.\n"); | |
5055 | ||
5056 | d = match_end; | |
5057 | dend = ((d >= string1 && d <= end1) | |
5058 | ? end_match_1 : end_match_2); | |
fa9a63c5 | 5059 | |
01618498 | 5060 | for (reg = 1; reg < num_regs; reg++) |
fa9a63c5 | 5061 | { |
01618498 SM |
5062 | regstart[reg] = best_regstart[reg]; |
5063 | regend[reg] = best_regend[reg]; | |
fa9a63c5 | 5064 | } |
25fe55af RS |
5065 | } |
5066 | } /* d != end_match_2 */ | |
fa9a63c5 RM |
5067 | |
5068 | succeed_label: | |
25fe55af | 5069 | DEBUG_PRINT1 ("Accepting match.\n"); |
fa9a63c5 | 5070 | |
25fe55af RS |
5071 | /* If caller wants register contents data back, do it. */ |
5072 | if (regs && !bufp->no_sub) | |
fa9a63c5 | 5073 | { |
25fe55af RS |
5074 | /* Have the register data arrays been allocated? */ |
5075 | if (bufp->regs_allocated == REGS_UNALLOCATED) | |
5076 | { /* No. So allocate them with malloc. We need one | |
5077 | extra element beyond `num_regs' for the `-1' marker | |
5078 | GNU code uses. */ | |
5079 | regs->num_regs = MAX (RE_NREGS, num_regs + 1); | |
5080 | regs->start = TALLOC (regs->num_regs, regoff_t); | |
5081 | regs->end = TALLOC (regs->num_regs, regoff_t); | |
5082 | if (regs->start == NULL || regs->end == NULL) | |
fa9a63c5 RM |
5083 | { |
5084 | FREE_VARIABLES (); | |
5085 | return -2; | |
5086 | } | |
25fe55af RS |
5087 | bufp->regs_allocated = REGS_REALLOCATE; |
5088 | } | |
5089 | else if (bufp->regs_allocated == REGS_REALLOCATE) | |
5090 | { /* Yes. If we need more elements than were already | |
5091 | allocated, reallocate them. If we need fewer, just | |
5092 | leave it alone. */ | |
5093 | if (regs->num_regs < num_regs + 1) | |
5094 | { | |
5095 | regs->num_regs = num_regs + 1; | |
5096 | RETALLOC (regs->start, regs->num_regs, regoff_t); | |
5097 | RETALLOC (regs->end, regs->num_regs, regoff_t); | |
5098 | if (regs->start == NULL || regs->end == NULL) | |
fa9a63c5 RM |
5099 | { |
5100 | FREE_VARIABLES (); | |
5101 | return -2; | |
5102 | } | |
25fe55af RS |
5103 | } |
5104 | } | |
5105 | else | |
fa9a63c5 RM |
5106 | { |
5107 | /* These braces fend off a "empty body in an else-statement" | |
25fe55af | 5108 | warning under GCC when assert expands to nothing. */ |
fa9a63c5 RM |
5109 | assert (bufp->regs_allocated == REGS_FIXED); |
5110 | } | |
5111 | ||
25fe55af RS |
5112 | /* Convert the pointer data in `regstart' and `regend' to |
5113 | indices. Register zero has to be set differently, | |
5114 | since we haven't kept track of any info for it. */ | |
5115 | if (regs->num_regs > 0) | |
5116 | { | |
5117 | regs->start[0] = pos; | |
99633e97 | 5118 | regs->end[0] = POINTER_TO_OFFSET (d); |
25fe55af | 5119 | } |
5e69f11e | 5120 | |
25fe55af RS |
5121 | /* Go through the first `min (num_regs, regs->num_regs)' |
5122 | registers, since that is all we initialized. */ | |
01618498 | 5123 | for (reg = 1; reg < MIN (num_regs, regs->num_regs); reg++) |
fa9a63c5 | 5124 | { |
01618498 SM |
5125 | if (REG_UNSET (regstart[reg]) || REG_UNSET (regend[reg])) |
5126 | regs->start[reg] = regs->end[reg] = -1; | |
25fe55af RS |
5127 | else |
5128 | { | |
01618498 SM |
5129 | regs->start[reg] |
5130 | = (regoff_t) POINTER_TO_OFFSET (regstart[reg]); | |
5131 | regs->end[reg] | |
5132 | = (regoff_t) POINTER_TO_OFFSET (regend[reg]); | |
25fe55af | 5133 | } |
fa9a63c5 | 5134 | } |
5e69f11e | 5135 | |
25fe55af RS |
5136 | /* If the regs structure we return has more elements than |
5137 | were in the pattern, set the extra elements to -1. If | |
5138 | we (re)allocated the registers, this is the case, | |
5139 | because we always allocate enough to have at least one | |
5140 | -1 at the end. */ | |
01618498 SM |
5141 | for (reg = num_regs; reg < regs->num_regs; reg++) |
5142 | regs->start[reg] = regs->end[reg] = -1; | |
fa9a63c5 RM |
5143 | } /* regs && !bufp->no_sub */ |
5144 | ||
25fe55af RS |
5145 | DEBUG_PRINT4 ("%u failure points pushed, %u popped (%u remain).\n", |
5146 | nfailure_points_pushed, nfailure_points_popped, | |
5147 | nfailure_points_pushed - nfailure_points_popped); | |
5148 | DEBUG_PRINT2 ("%u registers pushed.\n", num_regs_pushed); | |
fa9a63c5 | 5149 | |
99633e97 | 5150 | mcnt = POINTER_TO_OFFSET (d) - pos; |
fa9a63c5 | 5151 | |
25fe55af | 5152 | DEBUG_PRINT2 ("Returning %d from re_match_2.\n", mcnt); |
fa9a63c5 | 5153 | |
25fe55af RS |
5154 | FREE_VARIABLES (); |
5155 | return mcnt; | |
5156 | } | |
fa9a63c5 | 5157 | |
25fe55af | 5158 | /* Otherwise match next pattern command. */ |
fa9a63c5 RM |
5159 | switch (SWITCH_ENUM_CAST ((re_opcode_t) *p++)) |
5160 | { | |
25fe55af RS |
5161 | /* Ignore these. Used to ignore the n of succeed_n's which |
5162 | currently have n == 0. */ | |
5163 | case no_op: | |
5164 | DEBUG_PRINT1 ("EXECUTING no_op.\n"); | |
5165 | break; | |
fa9a63c5 RM |
5166 | |
5167 | case succeed: | |
25fe55af | 5168 | DEBUG_PRINT1 ("EXECUTING succeed.\n"); |
fa9a63c5 RM |
5169 | goto succeed_label; |
5170 | ||
25fe55af RS |
5171 | /* Match the next n pattern characters exactly. The following |
5172 | byte in the pattern defines n, and the n bytes after that | |
5173 | are the characters to match. */ | |
fa9a63c5 RM |
5174 | case exactn: |
5175 | mcnt = *p++; | |
25fe55af | 5176 | DEBUG_PRINT2 ("EXECUTING exactn %d.\n", mcnt); |
fa9a63c5 | 5177 | |
99633e97 SM |
5178 | /* Remember the start point to rollback upon failure. */ |
5179 | dfail = d; | |
5180 | ||
25fe55af RS |
5181 | /* This is written out as an if-else so we don't waste time |
5182 | testing `translate' inside the loop. */ | |
28703c16 | 5183 | if (RE_TRANSLATE_P (translate)) |
fa9a63c5 | 5184 | { |
e934739e RS |
5185 | if (multibyte) |
5186 | do | |
5187 | { | |
5188 | int pat_charlen, buf_charlen; | |
e71b1971 | 5189 | unsigned int pat_ch, buf_ch; |
e934739e RS |
5190 | |
5191 | PREFETCH (); | |
5192 | pat_ch = STRING_CHAR_AND_LENGTH (p, pend - p, pat_charlen); | |
5193 | buf_ch = STRING_CHAR_AND_LENGTH (d, dend - d, buf_charlen); | |
5194 | ||
5195 | if (RE_TRANSLATE (translate, buf_ch) | |
5196 | != pat_ch) | |
99633e97 SM |
5197 | { |
5198 | d = dfail; | |
5199 | goto fail; | |
5200 | } | |
e934739e RS |
5201 | |
5202 | p += pat_charlen; | |
5203 | d += buf_charlen; | |
5204 | mcnt -= pat_charlen; | |
5205 | } | |
5206 | while (mcnt > 0); | |
5207 | else | |
e934739e RS |
5208 | do |
5209 | { | |
5210 | PREFETCH (); | |
66f0296e | 5211 | if (RE_TRANSLATE (translate, *d) != *p++) |
99633e97 SM |
5212 | { |
5213 | d = dfail; | |
5214 | goto fail; | |
5215 | } | |
33c46939 | 5216 | d++; |
e934739e RS |
5217 | } |
5218 | while (--mcnt); | |
fa9a63c5 RM |
5219 | } |
5220 | else | |
5221 | { | |
5222 | do | |
5223 | { | |
5224 | PREFETCH (); | |
99633e97 SM |
5225 | if (*d++ != *p++) |
5226 | { | |
5227 | d = dfail; | |
5228 | goto fail; | |
5229 | } | |
fa9a63c5 RM |
5230 | } |
5231 | while (--mcnt); | |
5232 | } | |
25fe55af | 5233 | break; |
fa9a63c5 RM |
5234 | |
5235 | ||
25fe55af | 5236 | /* Match any character except possibly a newline or a null. */ |
fa9a63c5 | 5237 | case anychar: |
e934739e RS |
5238 | { |
5239 | int buf_charlen; | |
01618498 | 5240 | re_wchar_t buf_ch; |
fa9a63c5 | 5241 | |
e934739e | 5242 | DEBUG_PRINT1 ("EXECUTING anychar.\n"); |
fa9a63c5 | 5243 | |
e934739e | 5244 | PREFETCH (); |
2d1675e4 | 5245 | buf_ch = RE_STRING_CHAR_AND_LENGTH (d, dend - d, buf_charlen); |
e934739e RS |
5246 | buf_ch = TRANSLATE (buf_ch); |
5247 | ||
5248 | if ((!(bufp->syntax & RE_DOT_NEWLINE) | |
5249 | && buf_ch == '\n') | |
5250 | || ((bufp->syntax & RE_DOT_NOT_NULL) | |
5251 | && buf_ch == '\000')) | |
5252 | goto fail; | |
5253 | ||
e934739e RS |
5254 | DEBUG_PRINT2 (" Matched `%d'.\n", *d); |
5255 | d += buf_charlen; | |
5256 | } | |
fa9a63c5 RM |
5257 | break; |
5258 | ||
5259 | ||
5260 | case charset: | |
5261 | case charset_not: | |
5262 | { | |
b18215fc | 5263 | register unsigned int c; |
fa9a63c5 | 5264 | boolean not = (re_opcode_t) *(p - 1) == charset_not; |
b18215fc RS |
5265 | int len; |
5266 | ||
5267 | /* Start of actual range_table, or end of bitmap if there is no | |
5268 | range table. */ | |
01618498 | 5269 | re_char *range_table; |
b18215fc | 5270 | |
96cc36cc | 5271 | /* Nonzero if there is a range table. */ |
b18215fc RS |
5272 | int range_table_exists; |
5273 | ||
96cc36cc RS |
5274 | /* Number of ranges of range table. This is not included |
5275 | in the initial byte-length of the command. */ | |
5276 | int count = 0; | |
fa9a63c5 | 5277 | |
25fe55af | 5278 | DEBUG_PRINT2 ("EXECUTING charset%s.\n", not ? "_not" : ""); |
fa9a63c5 | 5279 | |
b18215fc | 5280 | range_table_exists = CHARSET_RANGE_TABLE_EXISTS_P (&p[-1]); |
96cc36cc | 5281 | |
b18215fc | 5282 | if (range_table_exists) |
96cc36cc RS |
5283 | { |
5284 | range_table = CHARSET_RANGE_TABLE (&p[-1]); /* Past the bitmap. */ | |
5285 | EXTRACT_NUMBER_AND_INCR (count, range_table); | |
5286 | } | |
b18215fc | 5287 | |
2d1675e4 SM |
5288 | PREFETCH (); |
5289 | c = RE_STRING_CHAR_AND_LENGTH (d, dend - d, len); | |
5290 | c = TRANSLATE (c); /* The character to match. */ | |
b18215fc RS |
5291 | |
5292 | if (SINGLE_BYTE_CHAR_P (c)) | |
5293 | { /* Lookup bitmap. */ | |
b18215fc RS |
5294 | /* Cast to `unsigned' instead of `unsigned char' in |
5295 | case the bit list is a full 32 bytes long. */ | |
5296 | if (c < (unsigned) (CHARSET_BITMAP_SIZE (&p[-1]) * BYTEWIDTH) | |
96cc36cc RS |
5297 | && p[1 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH))) |
5298 | not = !not; | |
b18215fc | 5299 | } |
96cc36cc | 5300 | #ifdef emacs |
b18215fc | 5301 | else if (range_table_exists) |
96cc36cc RS |
5302 | { |
5303 | int class_bits = CHARSET_RANGE_TABLE_BITS (&p[-1]); | |
5304 | ||
14473664 SM |
5305 | if ( (class_bits & BIT_LOWER && ISLOWER (c)) |
5306 | | (class_bits & BIT_MULTIBYTE) | |
96cc36cc RS |
5307 | | (class_bits & BIT_PUNCT && ISPUNCT (c)) |
5308 | | (class_bits & BIT_SPACE && ISSPACE (c)) | |
5309 | | (class_bits & BIT_UPPER && ISUPPER (c)) | |
5310 | | (class_bits & BIT_WORD && ISWORD (c))) | |
5311 | not = !not; | |
5312 | else | |
5313 | CHARSET_LOOKUP_RANGE_TABLE_RAW (not, c, range_table, count); | |
5314 | } | |
5315 | #endif /* emacs */ | |
fa9a63c5 | 5316 | |
96cc36cc RS |
5317 | if (range_table_exists) |
5318 | p = CHARSET_RANGE_TABLE_END (range_table, count); | |
5319 | else | |
5320 | p += CHARSET_BITMAP_SIZE (&p[-1]) + 1; | |
fa9a63c5 RM |
5321 | |
5322 | if (!not) goto fail; | |
5e69f11e | 5323 | |
b18215fc | 5324 | d += len; |
fa9a63c5 RM |
5325 | break; |
5326 | } | |
5327 | ||
5328 | ||
25fe55af | 5329 | /* The beginning of a group is represented by start_memory. |
505bde11 | 5330 | The argument is the register number. The text |
25fe55af RS |
5331 | matched within the group is recorded (in the internal |
5332 | registers data structure) under the register number. */ | |
5333 | case start_memory: | |
505bde11 SM |
5334 | DEBUG_PRINT2 ("EXECUTING start_memory %d:\n", *p); |
5335 | ||
5336 | /* In case we need to undo this operation (via backtracking). */ | |
5337 | PUSH_FAILURE_REG ((unsigned int)*p); | |
fa9a63c5 | 5338 | |
25fe55af | 5339 | regstart[*p] = d; |
4bb91c68 | 5340 | regend[*p] = NULL; /* probably unnecessary. -sm */ |
fa9a63c5 RM |
5341 | DEBUG_PRINT2 (" regstart: %d\n", POINTER_TO_OFFSET (regstart[*p])); |
5342 | ||
25fe55af | 5343 | /* Move past the register number and inner group count. */ |
505bde11 | 5344 | p += 1; |
25fe55af | 5345 | break; |
fa9a63c5 RM |
5346 | |
5347 | ||
25fe55af | 5348 | /* The stop_memory opcode represents the end of a group. Its |
505bde11 | 5349 | argument is the same as start_memory's: the register number. */ |
fa9a63c5 | 5350 | case stop_memory: |
505bde11 SM |
5351 | DEBUG_PRINT2 ("EXECUTING stop_memory %d:\n", *p); |
5352 | ||
5353 | assert (!REG_UNSET (regstart[*p])); | |
5354 | /* Strictly speaking, there should be code such as: | |
177c0ea7 | 5355 | |
0b32bf0e | 5356 | assert (REG_UNSET (regend[*p])); |
505bde11 SM |
5357 | PUSH_FAILURE_REGSTOP ((unsigned int)*p); |
5358 | ||
5359 | But the only info to be pushed is regend[*p] and it is known to | |
5360 | be UNSET, so there really isn't anything to push. | |
5361 | Not pushing anything, on the other hand deprives us from the | |
5362 | guarantee that regend[*p] is UNSET since undoing this operation | |
5363 | will not reset its value properly. This is not important since | |
5364 | the value will only be read on the next start_memory or at | |
5365 | the very end and both events can only happen if this stop_memory | |
5366 | is *not* undone. */ | |
fa9a63c5 | 5367 | |
25fe55af | 5368 | regend[*p] = d; |
fa9a63c5 RM |
5369 | DEBUG_PRINT2 (" regend: %d\n", POINTER_TO_OFFSET (regend[*p])); |
5370 | ||
25fe55af | 5371 | /* Move past the register number and the inner group count. */ |
505bde11 | 5372 | p += 1; |
25fe55af | 5373 | break; |
fa9a63c5 RM |
5374 | |
5375 | ||
5376 | /* \<digit> has been turned into a `duplicate' command which is | |
25fe55af RS |
5377 | followed by the numeric value of <digit> as the register number. */ |
5378 | case duplicate: | |
fa9a63c5 | 5379 | { |
66f0296e | 5380 | register re_char *d2, *dend2; |
25fe55af | 5381 | int regno = *p++; /* Get which register to match against. */ |
fa9a63c5 RM |
5382 | DEBUG_PRINT2 ("EXECUTING duplicate %d.\n", regno); |
5383 | ||
25fe55af RS |
5384 | /* Can't back reference a group which we've never matched. */ |
5385 | if (REG_UNSET (regstart[regno]) || REG_UNSET (regend[regno])) | |
5386 | goto fail; | |
5e69f11e | 5387 | |
25fe55af RS |
5388 | /* Where in input to try to start matching. */ |
5389 | d2 = regstart[regno]; | |
5e69f11e | 5390 | |
99633e97 SM |
5391 | /* Remember the start point to rollback upon failure. */ |
5392 | dfail = d; | |
5393 | ||
25fe55af RS |
5394 | /* Where to stop matching; if both the place to start and |
5395 | the place to stop matching are in the same string, then | |
5396 | set to the place to stop, otherwise, for now have to use | |
5397 | the end of the first string. */ | |
fa9a63c5 | 5398 | |
25fe55af | 5399 | dend2 = ((FIRST_STRING_P (regstart[regno]) |
fa9a63c5 RM |
5400 | == FIRST_STRING_P (regend[regno])) |
5401 | ? regend[regno] : end_match_1); | |
5402 | for (;;) | |
5403 | { | |
5404 | /* If necessary, advance to next segment in register | |
25fe55af | 5405 | contents. */ |
fa9a63c5 RM |
5406 | while (d2 == dend2) |
5407 | { | |
5408 | if (dend2 == end_match_2) break; | |
5409 | if (dend2 == regend[regno]) break; | |
5410 | ||
25fe55af RS |
5411 | /* End of string1 => advance to string2. */ |
5412 | d2 = string2; | |
5413 | dend2 = regend[regno]; | |
fa9a63c5 RM |
5414 | } |
5415 | /* At end of register contents => success */ | |
5416 | if (d2 == dend2) break; | |
5417 | ||
5418 | /* If necessary, advance to next segment in data. */ | |
5419 | PREFETCH (); | |
5420 | ||
5421 | /* How many characters left in this segment to match. */ | |
5422 | mcnt = dend - d; | |
5e69f11e | 5423 | |
fa9a63c5 | 5424 | /* Want how many consecutive characters we can match in |
25fe55af RS |
5425 | one shot, so, if necessary, adjust the count. */ |
5426 | if (mcnt > dend2 - d2) | |
fa9a63c5 | 5427 | mcnt = dend2 - d2; |
5e69f11e | 5428 | |
fa9a63c5 | 5429 | /* Compare that many; failure if mismatch, else move |
25fe55af | 5430 | past them. */ |
28703c16 | 5431 | if (RE_TRANSLATE_P (translate) |
2d1675e4 | 5432 | ? bcmp_translate (d, d2, mcnt, translate, multibyte) |
4bb91c68 | 5433 | : memcmp (d, d2, mcnt)) |
99633e97 SM |
5434 | { |
5435 | d = dfail; | |
5436 | goto fail; | |
5437 | } | |
fa9a63c5 | 5438 | d += mcnt, d2 += mcnt; |
fa9a63c5 RM |
5439 | } |
5440 | } | |
5441 | break; | |
5442 | ||
5443 | ||
25fe55af | 5444 | /* begline matches the empty string at the beginning of the string |
c0f9ea08 | 5445 | (unless `not_bol' is set in `bufp'), and after newlines. */ |
fa9a63c5 | 5446 | case begline: |
25fe55af | 5447 | DEBUG_PRINT1 ("EXECUTING begline.\n"); |
5e69f11e | 5448 | |
25fe55af RS |
5449 | if (AT_STRINGS_BEG (d)) |
5450 | { | |
5451 | if (!bufp->not_bol) break; | |
5452 | } | |
419d1c74 | 5453 | else |
25fe55af | 5454 | { |
419d1c74 SM |
5455 | unsigned char c; |
5456 | GET_CHAR_BEFORE_2 (c, d, string1, end1, string2, end2); | |
c0f9ea08 | 5457 | if (c == '\n') |
419d1c74 | 5458 | break; |
25fe55af RS |
5459 | } |
5460 | /* In all other cases, we fail. */ | |
5461 | goto fail; | |
fa9a63c5 RM |
5462 | |
5463 | ||
25fe55af | 5464 | /* endline is the dual of begline. */ |
fa9a63c5 | 5465 | case endline: |
25fe55af | 5466 | DEBUG_PRINT1 ("EXECUTING endline.\n"); |
fa9a63c5 | 5467 | |
25fe55af RS |
5468 | if (AT_STRINGS_END (d)) |
5469 | { | |
5470 | if (!bufp->not_eol) break; | |
5471 | } | |
f1ad044f | 5472 | else |
25fe55af | 5473 | { |
f1ad044f | 5474 | PREFETCH_NOLIMIT (); |
c0f9ea08 | 5475 | if (*d == '\n') |
f1ad044f | 5476 | break; |
25fe55af RS |
5477 | } |
5478 | goto fail; | |
fa9a63c5 RM |
5479 | |
5480 | ||
5481 | /* Match at the very beginning of the data. */ | |
25fe55af RS |
5482 | case begbuf: |
5483 | DEBUG_PRINT1 ("EXECUTING begbuf.\n"); | |
5484 | if (AT_STRINGS_BEG (d)) | |
5485 | break; | |
5486 | goto fail; | |
fa9a63c5 RM |
5487 | |
5488 | ||
5489 | /* Match at the very end of the data. */ | |
25fe55af RS |
5490 | case endbuf: |
5491 | DEBUG_PRINT1 ("EXECUTING endbuf.\n"); | |
fa9a63c5 RM |
5492 | if (AT_STRINGS_END (d)) |
5493 | break; | |
25fe55af | 5494 | goto fail; |
5e69f11e | 5495 | |
5e69f11e | 5496 | |
25fe55af RS |
5497 | /* on_failure_keep_string_jump is used to optimize `.*\n'. It |
5498 | pushes NULL as the value for the string on the stack. Then | |
505bde11 | 5499 | `POP_FAILURE_POINT' will keep the current value for the |
25fe55af RS |
5500 | string, instead of restoring it. To see why, consider |
5501 | matching `foo\nbar' against `.*\n'. The .* matches the foo; | |
5502 | then the . fails against the \n. But the next thing we want | |
5503 | to do is match the \n against the \n; if we restored the | |
5504 | string value, we would be back at the foo. | |
5505 | ||
5506 | Because this is used only in specific cases, we don't need to | |
5507 | check all the things that `on_failure_jump' does, to make | |
5508 | sure the right things get saved on the stack. Hence we don't | |
5509 | share its code. The only reason to push anything on the | |
5510 | stack at all is that otherwise we would have to change | |
5511 | `anychar's code to do something besides goto fail in this | |
5512 | case; that seems worse than this. */ | |
5513 | case on_failure_keep_string_jump: | |
505bde11 SM |
5514 | EXTRACT_NUMBER_AND_INCR (mcnt, p); |
5515 | DEBUG_PRINT3 ("EXECUTING on_failure_keep_string_jump %d (to %p):\n", | |
5516 | mcnt, p + mcnt); | |
fa9a63c5 | 5517 | |
505bde11 SM |
5518 | PUSH_FAILURE_POINT (p - 3, NULL); |
5519 | break; | |
5520 | ||
0683b6fa SM |
5521 | /* A nasty loop is introduced by the non-greedy *? and +?. |
5522 | With such loops, the stack only ever contains one failure point | |
5523 | at a time, so that a plain on_failure_jump_loop kind of | |
5524 | cycle detection cannot work. Worse yet, such a detection | |
5525 | can not only fail to detect a cycle, but it can also wrongly | |
5526 | detect a cycle (between different instantiations of the same | |
6df42991 | 5527 | loop). |
0683b6fa SM |
5528 | So the method used for those nasty loops is a little different: |
5529 | We use a special cycle-detection-stack-frame which is pushed | |
5530 | when the on_failure_jump_nastyloop failure-point is *popped*. | |
5531 | This special frame thus marks the beginning of one iteration | |
5532 | through the loop and we can hence easily check right here | |
5533 | whether something matched between the beginning and the end of | |
5534 | the loop. */ | |
5535 | case on_failure_jump_nastyloop: | |
5536 | EXTRACT_NUMBER_AND_INCR (mcnt, p); | |
5537 | DEBUG_PRINT3 ("EXECUTING on_failure_jump_nastyloop %d (to %p):\n", | |
5538 | mcnt, p + mcnt); | |
5539 | ||
5540 | assert ((re_opcode_t)p[-4] == no_op); | |
6df42991 SM |
5541 | { |
5542 | int cycle = 0; | |
5543 | CHECK_INFINITE_LOOP (p - 4, d); | |
5544 | if (!cycle) | |
5545 | /* If there's a cycle, just continue without pushing | |
5546 | this failure point. The failure point is the "try again" | |
5547 | option, which shouldn't be tried. | |
5548 | We want (x?)*?y\1z to match both xxyz and xxyxz. */ | |
5549 | PUSH_FAILURE_POINT (p - 3, d); | |
5550 | } | |
0683b6fa SM |
5551 | break; |
5552 | ||
4e8a9132 SM |
5553 | /* Simple loop detecting on_failure_jump: just check on the |
5554 | failure stack if the same spot was already hit earlier. */ | |
505bde11 SM |
5555 | case on_failure_jump_loop: |
5556 | on_failure: | |
5557 | EXTRACT_NUMBER_AND_INCR (mcnt, p); | |
5558 | DEBUG_PRINT3 ("EXECUTING on_failure_jump_loop %d (to %p):\n", | |
5559 | mcnt, p + mcnt); | |
6df42991 SM |
5560 | { |
5561 | int cycle = 0; | |
5562 | CHECK_INFINITE_LOOP (p - 3, d); | |
5563 | if (cycle) | |
5564 | /* If there's a cycle, get out of the loop, as if the matching | |
5565 | had failed. We used to just `goto fail' here, but that was | |
5566 | aborting the search a bit too early: we want to keep the | |
5567 | empty-loop-match and keep matching after the loop. | |
5568 | We want (x?)*y\1z to match both xxyz and xxyxz. */ | |
5569 | p += mcnt; | |
5570 | else | |
5571 | PUSH_FAILURE_POINT (p - 3, d); | |
5572 | } | |
25fe55af | 5573 | break; |
fa9a63c5 RM |
5574 | |
5575 | ||
5576 | /* Uses of on_failure_jump: | |
5e69f11e | 5577 | |
25fe55af RS |
5578 | Each alternative starts with an on_failure_jump that points |
5579 | to the beginning of the next alternative. Each alternative | |
5580 | except the last ends with a jump that in effect jumps past | |
5581 | the rest of the alternatives. (They really jump to the | |
5582 | ending jump of the following alternative, because tensioning | |
5583 | these jumps is a hassle.) | |
fa9a63c5 | 5584 | |
25fe55af RS |
5585 | Repeats start with an on_failure_jump that points past both |
5586 | the repetition text and either the following jump or | |
5587 | pop_failure_jump back to this on_failure_jump. */ | |
fa9a63c5 | 5588 | case on_failure_jump: |
5b370c2b | 5589 | IMMEDIATE_QUIT_CHECK; |
25fe55af | 5590 | EXTRACT_NUMBER_AND_INCR (mcnt, p); |
505bde11 SM |
5591 | DEBUG_PRINT3 ("EXECUTING on_failure_jump %d (to %p):\n", |
5592 | mcnt, p + mcnt); | |
25fe55af | 5593 | |
505bde11 | 5594 | PUSH_FAILURE_POINT (p -3, d); |
25fe55af RS |
5595 | break; |
5596 | ||
4e8a9132 | 5597 | /* This operation is used for greedy *. |
505bde11 SM |
5598 | Compare the beginning of the repeat with what in the |
5599 | pattern follows its end. If we can establish that there | |
5600 | is nothing that they would both match, i.e., that we | |
5601 | would have to backtrack because of (as in, e.g., `a*a') | |
5602 | then we can use a non-backtracking loop based on | |
4e8a9132 | 5603 | on_failure_keep_string_jump instead of on_failure_jump. */ |
505bde11 | 5604 | case on_failure_jump_smart: |
5b370c2b | 5605 | IMMEDIATE_QUIT_CHECK; |
25fe55af | 5606 | EXTRACT_NUMBER_AND_INCR (mcnt, p); |
505bde11 SM |
5607 | DEBUG_PRINT3 ("EXECUTING on_failure_jump_smart %d (to %p).\n", |
5608 | mcnt, p + mcnt); | |
25fe55af | 5609 | { |
01618498 | 5610 | re_char *p1 = p; /* Next operation. */ |
6dcf2d0e SM |
5611 | /* Here, we discard `const', making re_match non-reentrant. */ |
5612 | unsigned char *p2 = (unsigned char*) p + mcnt; /* Jump dest. */ | |
5613 | unsigned char *p3 = (unsigned char*) p - 3; /* opcode location. */ | |
fa9a63c5 | 5614 | |
505bde11 SM |
5615 | p -= 3; /* Reset so that we will re-execute the |
5616 | instruction once it's been changed. */ | |
fa9a63c5 | 5617 | |
4e8a9132 SM |
5618 | EXTRACT_NUMBER (mcnt, p2 - 2); |
5619 | ||
5620 | /* Ensure this is a indeed the trivial kind of loop | |
5621 | we are expecting. */ | |
5622 | assert (skip_one_char (p1) == p2 - 3); | |
5623 | assert ((re_opcode_t) p2[-3] == jump && p2 + mcnt == p); | |
99633e97 | 5624 | DEBUG_STATEMENT (debug += 2); |
505bde11 | 5625 | if (mutually_exclusive_p (bufp, p1, p2)) |
fa9a63c5 | 5626 | { |
505bde11 | 5627 | /* Use a fast `on_failure_keep_string_jump' loop. */ |
4e8a9132 | 5628 | DEBUG_PRINT1 (" smart exclusive => fast loop.\n"); |
01618498 | 5629 | *p3 = (unsigned char) on_failure_keep_string_jump; |
4e8a9132 | 5630 | STORE_NUMBER (p2 - 2, mcnt + 3); |
25fe55af | 5631 | } |
505bde11 | 5632 | else |
fa9a63c5 | 5633 | { |
505bde11 SM |
5634 | /* Default to a safe `on_failure_jump' loop. */ |
5635 | DEBUG_PRINT1 (" smart default => slow loop.\n"); | |
01618498 | 5636 | *p3 = (unsigned char) on_failure_jump; |
fa9a63c5 | 5637 | } |
99633e97 | 5638 | DEBUG_STATEMENT (debug -= 2); |
25fe55af | 5639 | } |
505bde11 | 5640 | break; |
25fe55af RS |
5641 | |
5642 | /* Unconditionally jump (without popping any failure points). */ | |
5643 | case jump: | |
fa9a63c5 | 5644 | unconditional_jump: |
5b370c2b | 5645 | IMMEDIATE_QUIT_CHECK; |
fa9a63c5 | 5646 | EXTRACT_NUMBER_AND_INCR (mcnt, p); /* Get the amount to jump. */ |
25fe55af RS |
5647 | DEBUG_PRINT2 ("EXECUTING jump %d ", mcnt); |
5648 | p += mcnt; /* Do the jump. */ | |
505bde11 | 5649 | DEBUG_PRINT2 ("(to %p).\n", p); |
25fe55af RS |
5650 | break; |
5651 | ||
5652 | ||
25fe55af RS |
5653 | /* Have to succeed matching what follows at least n times. |
5654 | After that, handle like `on_failure_jump'. */ | |
5655 | case succeed_n: | |
01618498 | 5656 | /* Signedness doesn't matter since we only compare MCNT to 0. */ |
25fe55af RS |
5657 | EXTRACT_NUMBER (mcnt, p + 2); |
5658 | DEBUG_PRINT2 ("EXECUTING succeed_n %d.\n", mcnt); | |
5e69f11e | 5659 | |
dc1e502d SM |
5660 | /* Originally, mcnt is how many times we HAVE to succeed. */ |
5661 | if (mcnt != 0) | |
25fe55af | 5662 | { |
6dcf2d0e SM |
5663 | /* Here, we discard `const', making re_match non-reentrant. */ |
5664 | unsigned char *p2 = (unsigned char*) p + 2; /* counter loc. */ | |
dc1e502d | 5665 | mcnt--; |
01618498 SM |
5666 | p += 4; |
5667 | PUSH_NUMBER (p2, mcnt); | |
25fe55af | 5668 | } |
dc1e502d SM |
5669 | else |
5670 | /* The two bytes encoding mcnt == 0 are two no_op opcodes. */ | |
5671 | goto on_failure; | |
25fe55af RS |
5672 | break; |
5673 | ||
5674 | case jump_n: | |
01618498 | 5675 | /* Signedness doesn't matter since we only compare MCNT to 0. */ |
25fe55af RS |
5676 | EXTRACT_NUMBER (mcnt, p + 2); |
5677 | DEBUG_PRINT2 ("EXECUTING jump_n %d.\n", mcnt); | |
5678 | ||
5679 | /* Originally, this is how many times we CAN jump. */ | |
dc1e502d | 5680 | if (mcnt != 0) |
25fe55af | 5681 | { |
6dcf2d0e SM |
5682 | /* Here, we discard `const', making re_match non-reentrant. */ |
5683 | unsigned char *p2 = (unsigned char*) p + 2; /* counter loc. */ | |
dc1e502d | 5684 | mcnt--; |
01618498 | 5685 | PUSH_NUMBER (p2, mcnt); |
dc1e502d | 5686 | goto unconditional_jump; |
25fe55af RS |
5687 | } |
5688 | /* If don't have to jump any more, skip over the rest of command. */ | |
5e69f11e RM |
5689 | else |
5690 | p += 4; | |
25fe55af | 5691 | break; |
5e69f11e | 5692 | |
fa9a63c5 RM |
5693 | case set_number_at: |
5694 | { | |
01618498 | 5695 | unsigned char *p2; /* Location of the counter. */ |
25fe55af | 5696 | DEBUG_PRINT1 ("EXECUTING set_number_at.\n"); |
fa9a63c5 | 5697 | |
25fe55af | 5698 | EXTRACT_NUMBER_AND_INCR (mcnt, p); |
6dcf2d0e SM |
5699 | /* Here, we discard `const', making re_match non-reentrant. */ |
5700 | p2 = (unsigned char*) p + mcnt; | |
01618498 | 5701 | /* Signedness doesn't matter since we only copy MCNT's bits . */ |
25fe55af | 5702 | EXTRACT_NUMBER_AND_INCR (mcnt, p); |
01618498 SM |
5703 | DEBUG_PRINT3 (" Setting %p to %d.\n", p2, mcnt); |
5704 | PUSH_NUMBER (p2, mcnt); | |
25fe55af RS |
5705 | break; |
5706 | } | |
9121ca40 KH |
5707 | |
5708 | case wordbound: | |
66f0296e SM |
5709 | case notwordbound: |
5710 | not = (re_opcode_t) *(p - 1) == notwordbound; | |
5711 | DEBUG_PRINT2 ("EXECUTING %swordbound.\n", not?"not":""); | |
fa9a63c5 | 5712 | |
99633e97 | 5713 | /* We SUCCEED (or FAIL) in one of the following cases: */ |
9121ca40 | 5714 | |
b18215fc | 5715 | /* Case 1: D is at the beginning or the end of string. */ |
9121ca40 | 5716 | if (AT_STRINGS_BEG (d) || AT_STRINGS_END (d)) |
66f0296e | 5717 | not = !not; |
b18215fc RS |
5718 | else |
5719 | { | |
5720 | /* C1 is the character before D, S1 is the syntax of C1, C2 | |
5721 | is the character at D, and S2 is the syntax of C2. */ | |
01618498 SM |
5722 | re_wchar_t c1, c2; |
5723 | int s1, s2; | |
b18215fc | 5724 | #ifdef emacs |
2d1675e4 SM |
5725 | int offset = PTR_TO_OFFSET (d - 1); |
5726 | int charpos = SYNTAX_TABLE_BYTE_TO_CHAR (offset); | |
5d967c7a | 5727 | UPDATE_SYNTAX_TABLE (charpos); |
25fe55af | 5728 | #endif |
66f0296e | 5729 | GET_CHAR_BEFORE_2 (c1, d, string1, end1, string2, end2); |
b18215fc RS |
5730 | s1 = SYNTAX (c1); |
5731 | #ifdef emacs | |
5d967c7a | 5732 | UPDATE_SYNTAX_TABLE_FORWARD (charpos + 1); |
25fe55af | 5733 | #endif |
f1ad044f | 5734 | PREFETCH_NOLIMIT (); |
2d1675e4 | 5735 | c2 = RE_STRING_CHAR (d, dend - d); |
b18215fc RS |
5736 | s2 = SYNTAX (c2); |
5737 | ||
5738 | if (/* Case 2: Only one of S1 and S2 is Sword. */ | |
5739 | ((s1 == Sword) != (s2 == Sword)) | |
5740 | /* Case 3: Both of S1 and S2 are Sword, and macro | |
25fe55af | 5741 | WORD_BOUNDARY_P (C1, C2) returns nonzero. */ |
b18215fc | 5742 | || ((s1 == Sword) && WORD_BOUNDARY_P (c1, c2))) |
66f0296e SM |
5743 | not = !not; |
5744 | } | |
5745 | if (not) | |
9121ca40 | 5746 | break; |
b18215fc | 5747 | else |
9121ca40 | 5748 | goto fail; |
fa9a63c5 RM |
5749 | |
5750 | case wordbeg: | |
25fe55af | 5751 | DEBUG_PRINT1 ("EXECUTING wordbeg.\n"); |
fa9a63c5 | 5752 | |
b18215fc RS |
5753 | /* We FAIL in one of the following cases: */ |
5754 | ||
25fe55af | 5755 | /* Case 1: D is at the end of string. */ |
b18215fc | 5756 | if (AT_STRINGS_END (d)) |
99633e97 | 5757 | goto fail; |
b18215fc RS |
5758 | else |
5759 | { | |
5760 | /* C1 is the character before D, S1 is the syntax of C1, C2 | |
5761 | is the character at D, and S2 is the syntax of C2. */ | |
01618498 SM |
5762 | re_wchar_t c1, c2; |
5763 | int s1, s2; | |
fa9a63c5 | 5764 | #ifdef emacs |
2d1675e4 SM |
5765 | int offset = PTR_TO_OFFSET (d); |
5766 | int charpos = SYNTAX_TABLE_BYTE_TO_CHAR (offset); | |
92432794 | 5767 | UPDATE_SYNTAX_TABLE (charpos); |
25fe55af | 5768 | #endif |
99633e97 | 5769 | PREFETCH (); |
2d1675e4 | 5770 | c2 = RE_STRING_CHAR (d, dend - d); |
b18215fc | 5771 | s2 = SYNTAX (c2); |
177c0ea7 | 5772 | |
b18215fc RS |
5773 | /* Case 2: S2 is not Sword. */ |
5774 | if (s2 != Sword) | |
5775 | goto fail; | |
5776 | ||
5777 | /* Case 3: D is not at the beginning of string ... */ | |
5778 | if (!AT_STRINGS_BEG (d)) | |
5779 | { | |
5780 | GET_CHAR_BEFORE_2 (c1, d, string1, end1, string2, end2); | |
5781 | #ifdef emacs | |
5d967c7a | 5782 | UPDATE_SYNTAX_TABLE_BACKWARD (charpos - 1); |
25fe55af | 5783 | #endif |
b18215fc RS |
5784 | s1 = SYNTAX (c1); |
5785 | ||
5786 | /* ... and S1 is Sword, and WORD_BOUNDARY_P (C1, C2) | |
25fe55af | 5787 | returns 0. */ |
b18215fc RS |
5788 | if ((s1 == Sword) && !WORD_BOUNDARY_P (c1, c2)) |
5789 | goto fail; | |
5790 | } | |
5791 | } | |
e318085a RS |
5792 | break; |
5793 | ||
b18215fc | 5794 | case wordend: |
25fe55af | 5795 | DEBUG_PRINT1 ("EXECUTING wordend.\n"); |
b18215fc RS |
5796 | |
5797 | /* We FAIL in one of the following cases: */ | |
5798 | ||
5799 | /* Case 1: D is at the beginning of string. */ | |
5800 | if (AT_STRINGS_BEG (d)) | |
e318085a | 5801 | goto fail; |
b18215fc RS |
5802 | else |
5803 | { | |
5804 | /* C1 is the character before D, S1 is the syntax of C1, C2 | |
5805 | is the character at D, and S2 is the syntax of C2. */ | |
01618498 SM |
5806 | re_wchar_t c1, c2; |
5807 | int s1, s2; | |
5d967c7a | 5808 | #ifdef emacs |
2d1675e4 SM |
5809 | int offset = PTR_TO_OFFSET (d) - 1; |
5810 | int charpos = SYNTAX_TABLE_BYTE_TO_CHAR (offset); | |
92432794 | 5811 | UPDATE_SYNTAX_TABLE (charpos); |
5d967c7a | 5812 | #endif |
99633e97 | 5813 | GET_CHAR_BEFORE_2 (c1, d, string1, end1, string2, end2); |
b18215fc RS |
5814 | s1 = SYNTAX (c1); |
5815 | ||
5816 | /* Case 2: S1 is not Sword. */ | |
5817 | if (s1 != Sword) | |
5818 | goto fail; | |
5819 | ||
5820 | /* Case 3: D is not at the end of string ... */ | |
5821 | if (!AT_STRINGS_END (d)) | |
5822 | { | |
f1ad044f | 5823 | PREFETCH_NOLIMIT (); |
2d1675e4 | 5824 | c2 = RE_STRING_CHAR (d, dend - d); |
5d967c7a RS |
5825 | #ifdef emacs |
5826 | UPDATE_SYNTAX_TABLE_FORWARD (charpos); | |
5827 | #endif | |
b18215fc RS |
5828 | s2 = SYNTAX (c2); |
5829 | ||
5830 | /* ... and S2 is Sword, and WORD_BOUNDARY_P (C1, C2) | |
25fe55af | 5831 | returns 0. */ |
b18215fc | 5832 | if ((s2 == Sword) && !WORD_BOUNDARY_P (c1, c2)) |
25fe55af | 5833 | goto fail; |
b18215fc RS |
5834 | } |
5835 | } | |
e318085a RS |
5836 | break; |
5837 | ||
669fa600 SM |
5838 | case symbeg: |
5839 | DEBUG_PRINT1 ("EXECUTING symbeg.\n"); | |
5840 | ||
5841 | /* We FAIL in one of the following cases: */ | |
5842 | ||
5843 | /* Case 1: D is at the end of string. */ | |
5844 | if (AT_STRINGS_END (d)) | |
5845 | goto fail; | |
5846 | else | |
5847 | { | |
5848 | /* C1 is the character before D, S1 is the syntax of C1, C2 | |
5849 | is the character at D, and S2 is the syntax of C2. */ | |
5850 | re_wchar_t c1, c2; | |
5851 | int s1, s2; | |
5852 | #ifdef emacs | |
5853 | int offset = PTR_TO_OFFSET (d); | |
5854 | int charpos = SYNTAX_TABLE_BYTE_TO_CHAR (offset); | |
5855 | UPDATE_SYNTAX_TABLE (charpos); | |
5856 | #endif | |
5857 | PREFETCH (); | |
5858 | c2 = RE_STRING_CHAR (d, dend - d); | |
5859 | s2 = SYNTAX (c2); | |
5860 | ||
5861 | /* Case 2: S2 is neither Sword nor Ssymbol. */ | |
5862 | if (s2 != Sword && s2 != Ssymbol) | |
5863 | goto fail; | |
5864 | ||
5865 | /* Case 3: D is not at the beginning of string ... */ | |
5866 | if (!AT_STRINGS_BEG (d)) | |
5867 | { | |
5868 | GET_CHAR_BEFORE_2 (c1, d, string1, end1, string2, end2); | |
5869 | #ifdef emacs | |
5870 | UPDATE_SYNTAX_TABLE_BACKWARD (charpos - 1); | |
5871 | #endif | |
5872 | s1 = SYNTAX (c1); | |
5873 | ||
5874 | /* ... and S1 is Sword or Ssymbol. */ | |
5875 | if (s1 == Sword || s1 == Ssymbol) | |
5876 | goto fail; | |
5877 | } | |
5878 | } | |
5879 | break; | |
5880 | ||
5881 | case symend: | |
5882 | DEBUG_PRINT1 ("EXECUTING symend.\n"); | |
5883 | ||
5884 | /* We FAIL in one of the following cases: */ | |
5885 | ||
5886 | /* Case 1: D is at the beginning of string. */ | |
5887 | if (AT_STRINGS_BEG (d)) | |
5888 | goto fail; | |
5889 | else | |
5890 | { | |
5891 | /* C1 is the character before D, S1 is the syntax of C1, C2 | |
5892 | is the character at D, and S2 is the syntax of C2. */ | |
5893 | re_wchar_t c1, c2; | |
5894 | int s1, s2; | |
5895 | #ifdef emacs | |
5896 | int offset = PTR_TO_OFFSET (d) - 1; | |
5897 | int charpos = SYNTAX_TABLE_BYTE_TO_CHAR (offset); | |
5898 | UPDATE_SYNTAX_TABLE (charpos); | |
5899 | #endif | |
5900 | GET_CHAR_BEFORE_2 (c1, d, string1, end1, string2, end2); | |
5901 | s1 = SYNTAX (c1); | |
5902 | ||
5903 | /* Case 2: S1 is neither Ssymbol nor Sword. */ | |
5904 | if (s1 != Sword && s1 != Ssymbol) | |
5905 | goto fail; | |
5906 | ||
5907 | /* Case 3: D is not at the end of string ... */ | |
5908 | if (!AT_STRINGS_END (d)) | |
5909 | { | |
5910 | PREFETCH_NOLIMIT (); | |
5911 | c2 = RE_STRING_CHAR (d, dend - d); | |
5912 | #ifdef emacs | |
5913 | UPDATE_SYNTAX_TABLE_FORWARD (charpos); | |
5914 | #endif | |
5915 | s2 = SYNTAX (c2); | |
5916 | ||
5917 | /* ... and S2 is Sword or Ssymbol. */ | |
5918 | if (s2 == Sword || s2 == Ssymbol) | |
5919 | goto fail; | |
5920 | } | |
5921 | } | |
5922 | break; | |
5923 | ||
fa9a63c5 | 5924 | case syntaxspec: |
1fb352e0 SM |
5925 | case notsyntaxspec: |
5926 | not = (re_opcode_t) *(p - 1) == notsyntaxspec; | |
fa9a63c5 | 5927 | mcnt = *p++; |
1fb352e0 | 5928 | DEBUG_PRINT3 ("EXECUTING %ssyntaxspec %d.\n", not?"not":"", mcnt); |
fa9a63c5 | 5929 | PREFETCH (); |
b18215fc RS |
5930 | #ifdef emacs |
5931 | { | |
2d1675e4 SM |
5932 | int offset = PTR_TO_OFFSET (d); |
5933 | int pos1 = SYNTAX_TABLE_BYTE_TO_CHAR (offset); | |
b18215fc RS |
5934 | UPDATE_SYNTAX_TABLE (pos1); |
5935 | } | |
25fe55af | 5936 | #endif |
b18215fc | 5937 | { |
01618498 SM |
5938 | int len; |
5939 | re_wchar_t c; | |
b18215fc | 5940 | |
2d1675e4 | 5941 | c = RE_STRING_CHAR_AND_LENGTH (d, dend - d, len); |
b18215fc | 5942 | |
990b2375 | 5943 | if ((SYNTAX (c) != (enum syntaxcode) mcnt) ^ not) |
1fb352e0 | 5944 | goto fail; |
b18215fc RS |
5945 | d += len; |
5946 | } | |
fa9a63c5 RM |
5947 | break; |
5948 | ||
b18215fc | 5949 | #ifdef emacs |
1fb352e0 SM |
5950 | case before_dot: |
5951 | DEBUG_PRINT1 ("EXECUTING before_dot.\n"); | |
5952 | if (PTR_BYTE_POS (d) >= PT_BYTE) | |
fa9a63c5 | 5953 | goto fail; |
b18215fc RS |
5954 | break; |
5955 | ||
1fb352e0 SM |
5956 | case at_dot: |
5957 | DEBUG_PRINT1 ("EXECUTING at_dot.\n"); | |
5958 | if (PTR_BYTE_POS (d) != PT_BYTE) | |
5959 | goto fail; | |
5960 | break; | |
b18215fc | 5961 | |
1fb352e0 SM |
5962 | case after_dot: |
5963 | DEBUG_PRINT1 ("EXECUTING after_dot.\n"); | |
5964 | if (PTR_BYTE_POS (d) <= PT_BYTE) | |
5965 | goto fail; | |
e318085a | 5966 | break; |
fa9a63c5 | 5967 | |
1fb352e0 | 5968 | case categoryspec: |
b18215fc | 5969 | case notcategoryspec: |
1fb352e0 | 5970 | not = (re_opcode_t) *(p - 1) == notcategoryspec; |
b18215fc | 5971 | mcnt = *p++; |
1fb352e0 | 5972 | DEBUG_PRINT3 ("EXECUTING %scategoryspec %d.\n", not?"not":"", mcnt); |
b18215fc RS |
5973 | PREFETCH (); |
5974 | { | |
01618498 SM |
5975 | int len; |
5976 | re_wchar_t c; | |
5977 | ||
2d1675e4 | 5978 | c = RE_STRING_CHAR_AND_LENGTH (d, dend - d, len); |
b18215fc | 5979 | |
1fb352e0 | 5980 | if ((!CHAR_HAS_CATEGORY (c, mcnt)) ^ not) |
b18215fc RS |
5981 | goto fail; |
5982 | d += len; | |
5983 | } | |
fa9a63c5 | 5984 | break; |
5e69f11e | 5985 | |
1fb352e0 | 5986 | #endif /* emacs */ |
5e69f11e | 5987 | |
0b32bf0e SM |
5988 | default: |
5989 | abort (); | |
fa9a63c5 | 5990 | } |
b18215fc | 5991 | continue; /* Successfully executed one pattern command; keep going. */ |
fa9a63c5 RM |
5992 | |
5993 | ||
5994 | /* We goto here if a matching operation fails. */ | |
5995 | fail: | |
5b370c2b | 5996 | IMMEDIATE_QUIT_CHECK; |
fa9a63c5 | 5997 | if (!FAIL_STACK_EMPTY ()) |
505bde11 | 5998 | { |
01618498 | 5999 | re_char *str, *pat; |
505bde11 | 6000 | /* A restart point is known. Restore to that state. */ |
0b32bf0e SM |
6001 | DEBUG_PRINT1 ("\nFAIL:\n"); |
6002 | POP_FAILURE_POINT (str, pat); | |
505bde11 SM |
6003 | switch (SWITCH_ENUM_CAST ((re_opcode_t) *pat++)) |
6004 | { | |
6005 | case on_failure_keep_string_jump: | |
6006 | assert (str == NULL); | |
6007 | goto continue_failure_jump; | |
6008 | ||
0683b6fa SM |
6009 | case on_failure_jump_nastyloop: |
6010 | assert ((re_opcode_t)pat[-2] == no_op); | |
6011 | PUSH_FAILURE_POINT (pat - 2, str); | |
6012 | /* Fallthrough */ | |
6013 | ||
505bde11 SM |
6014 | case on_failure_jump_loop: |
6015 | case on_failure_jump: | |
6016 | case succeed_n: | |
6017 | d = str; | |
6018 | continue_failure_jump: | |
6019 | EXTRACT_NUMBER_AND_INCR (mcnt, pat); | |
6020 | p = pat + mcnt; | |
6021 | break; | |
b18215fc | 6022 | |
0683b6fa SM |
6023 | case no_op: |
6024 | /* A special frame used for nastyloops. */ | |
6025 | goto fail; | |
6026 | ||
505bde11 SM |
6027 | default: |
6028 | abort(); | |
6029 | } | |
fa9a63c5 | 6030 | |
505bde11 | 6031 | assert (p >= bufp->buffer && p <= pend); |
b18215fc | 6032 | |
0b32bf0e | 6033 | if (d >= string1 && d <= end1) |
fa9a63c5 | 6034 | dend = end_match_1; |
0b32bf0e | 6035 | } |
fa9a63c5 | 6036 | else |
0b32bf0e | 6037 | break; /* Matching at this starting point really fails. */ |
fa9a63c5 RM |
6038 | } /* for (;;) */ |
6039 | ||
6040 | if (best_regs_set) | |
6041 | goto restore_best_regs; | |
6042 | ||
6043 | FREE_VARIABLES (); | |
6044 | ||
b18215fc | 6045 | return -1; /* Failure to match. */ |
fa9a63c5 RM |
6046 | } /* re_match_2 */ |
6047 | \f | |
6048 | /* Subroutine definitions for re_match_2. */ | |
6049 | ||
fa9a63c5 RM |
6050 | /* Return zero if TRANSLATE[S1] and TRANSLATE[S2] are identical for LEN |
6051 | bytes; nonzero otherwise. */ | |
5e69f11e | 6052 | |
fa9a63c5 | 6053 | static int |
2d1675e4 SM |
6054 | bcmp_translate (s1, s2, len, translate, multibyte) |
6055 | re_char *s1, *s2; | |
fa9a63c5 | 6056 | register int len; |
6676cb1c | 6057 | RE_TRANSLATE_TYPE translate; |
2d1675e4 | 6058 | const int multibyte; |
fa9a63c5 | 6059 | { |
2d1675e4 SM |
6060 | register re_char *p1 = s1, *p2 = s2; |
6061 | re_char *p1_end = s1 + len; | |
6062 | re_char *p2_end = s2 + len; | |
e934739e | 6063 | |
4bb91c68 SM |
6064 | /* FIXME: Checking both p1 and p2 presumes that the two strings might have |
6065 | different lengths, but relying on a single `len' would break this. -sm */ | |
6066 | while (p1 < p1_end && p2 < p2_end) | |
fa9a63c5 | 6067 | { |
e934739e | 6068 | int p1_charlen, p2_charlen; |
01618498 | 6069 | re_wchar_t p1_ch, p2_ch; |
e934739e | 6070 | |
2d1675e4 SM |
6071 | p1_ch = RE_STRING_CHAR_AND_LENGTH (p1, p1_end - p1, p1_charlen); |
6072 | p2_ch = RE_STRING_CHAR_AND_LENGTH (p2, p2_end - p2, p2_charlen); | |
e934739e RS |
6073 | |
6074 | if (RE_TRANSLATE (translate, p1_ch) | |
6075 | != RE_TRANSLATE (translate, p2_ch)) | |
bc192b5b | 6076 | return 1; |
e934739e RS |
6077 | |
6078 | p1 += p1_charlen, p2 += p2_charlen; | |
fa9a63c5 | 6079 | } |
e934739e RS |
6080 | |
6081 | if (p1 != p1_end || p2 != p2_end) | |
6082 | return 1; | |
6083 | ||
fa9a63c5 RM |
6084 | return 0; |
6085 | } | |
6086 | \f | |
6087 | /* Entry points for GNU code. */ | |
6088 | ||
6089 | /* re_compile_pattern is the GNU regular expression compiler: it | |
6090 | compiles PATTERN (of length SIZE) and puts the result in BUFP. | |
6091 | Returns 0 if the pattern was valid, otherwise an error string. | |
5e69f11e | 6092 | |
fa9a63c5 RM |
6093 | Assumes the `allocated' (and perhaps `buffer') and `translate' fields |
6094 | are set in BUFP on entry. | |
5e69f11e | 6095 | |
b18215fc | 6096 | We call regex_compile to do the actual compilation. */ |
fa9a63c5 RM |
6097 | |
6098 | const char * | |
6099 | re_compile_pattern (pattern, length, bufp) | |
6100 | const char *pattern; | |
0b32bf0e | 6101 | size_t length; |
fa9a63c5 RM |
6102 | struct re_pattern_buffer *bufp; |
6103 | { | |
6104 | reg_errcode_t ret; | |
5e69f11e | 6105 | |
fa9a63c5 RM |
6106 | /* GNU code is written to assume at least RE_NREGS registers will be set |
6107 | (and at least one extra will be -1). */ | |
6108 | bufp->regs_allocated = REGS_UNALLOCATED; | |
5e69f11e | 6109 | |
fa9a63c5 RM |
6110 | /* And GNU code determines whether or not to get register information |
6111 | by passing null for the REGS argument to re_match, etc., not by | |
6112 | setting no_sub. */ | |
6113 | bufp->no_sub = 0; | |
5e69f11e | 6114 | |
4bb91c68 | 6115 | ret = regex_compile ((re_char*) pattern, length, re_syntax_options, bufp); |
fa9a63c5 RM |
6116 | |
6117 | if (!ret) | |
6118 | return NULL; | |
6119 | return gettext (re_error_msgid[(int) ret]); | |
5e69f11e | 6120 | } |
c0f9ea08 | 6121 | WEAK_ALIAS (__re_compile_pattern, re_compile_pattern) |
fa9a63c5 | 6122 | \f |
b18215fc RS |
6123 | /* Entry points compatible with 4.2 BSD regex library. We don't define |
6124 | them unless specifically requested. */ | |
fa9a63c5 | 6125 | |
0b32bf0e | 6126 | #if defined _REGEX_RE_COMP || defined _LIBC |
fa9a63c5 RM |
6127 | |
6128 | /* BSD has one and only one pattern buffer. */ | |
6129 | static struct re_pattern_buffer re_comp_buf; | |
6130 | ||
6131 | char * | |
0b32bf0e | 6132 | # ifdef _LIBC |
48afdd44 RM |
6133 | /* Make these definitions weak in libc, so POSIX programs can redefine |
6134 | these names if they don't use our functions, and still use | |
6135 | regcomp/regexec below without link errors. */ | |
6136 | weak_function | |
0b32bf0e | 6137 | # endif |
fa9a63c5 RM |
6138 | re_comp (s) |
6139 | const char *s; | |
6140 | { | |
6141 | reg_errcode_t ret; | |
5e69f11e | 6142 | |
fa9a63c5 RM |
6143 | if (!s) |
6144 | { | |
6145 | if (!re_comp_buf.buffer) | |
0b32bf0e | 6146 | /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */ |
a60198e5 | 6147 | return (char *) gettext ("No previous regular expression"); |
fa9a63c5 RM |
6148 | return 0; |
6149 | } | |
6150 | ||
6151 | if (!re_comp_buf.buffer) | |
6152 | { | |
6153 | re_comp_buf.buffer = (unsigned char *) malloc (200); | |
6154 | if (re_comp_buf.buffer == NULL) | |
0b32bf0e SM |
6155 | /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */ |
6156 | return (char *) gettext (re_error_msgid[(int) REG_ESPACE]); | |
fa9a63c5 RM |
6157 | re_comp_buf.allocated = 200; |
6158 | ||
6159 | re_comp_buf.fastmap = (char *) malloc (1 << BYTEWIDTH); | |
6160 | if (re_comp_buf.fastmap == NULL) | |
a60198e5 SM |
6161 | /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */ |
6162 | return (char *) gettext (re_error_msgid[(int) REG_ESPACE]); | |
fa9a63c5 RM |
6163 | } |
6164 | ||
6165 | /* Since `re_exec' always passes NULL for the `regs' argument, we | |
6166 | don't need to initialize the pattern buffer fields which affect it. */ | |
6167 | ||
fa9a63c5 | 6168 | ret = regex_compile (s, strlen (s), re_syntax_options, &re_comp_buf); |
5e69f11e | 6169 | |
fa9a63c5 RM |
6170 | if (!ret) |
6171 | return NULL; | |
6172 | ||
6173 | /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */ | |
6174 | return (char *) gettext (re_error_msgid[(int) ret]); | |
6175 | } | |
6176 | ||
6177 | ||
6178 | int | |
0b32bf0e | 6179 | # ifdef _LIBC |
48afdd44 | 6180 | weak_function |
0b32bf0e | 6181 | # endif |
fa9a63c5 RM |
6182 | re_exec (s) |
6183 | const char *s; | |
6184 | { | |
6185 | const int len = strlen (s); | |
6186 | return | |
6187 | 0 <= re_search (&re_comp_buf, s, len, 0, len, (struct re_registers *) 0); | |
6188 | } | |
6189 | #endif /* _REGEX_RE_COMP */ | |
6190 | \f | |
6191 | /* POSIX.2 functions. Don't define these for Emacs. */ | |
6192 | ||
6193 | #ifndef emacs | |
6194 | ||
6195 | /* regcomp takes a regular expression as a string and compiles it. | |
6196 | ||
b18215fc | 6197 | PREG is a regex_t *. We do not expect any fields to be initialized, |
fa9a63c5 RM |
6198 | since POSIX says we shouldn't. Thus, we set |
6199 | ||
6200 | `buffer' to the compiled pattern; | |
6201 | `used' to the length of the compiled pattern; | |
6202 | `syntax' to RE_SYNTAX_POSIX_EXTENDED if the | |
6203 | REG_EXTENDED bit in CFLAGS is set; otherwise, to | |
6204 | RE_SYNTAX_POSIX_BASIC; | |
c0f9ea08 SM |
6205 | `fastmap' to an allocated space for the fastmap; |
6206 | `fastmap_accurate' to zero; | |
fa9a63c5 RM |
6207 | `re_nsub' to the number of subexpressions in PATTERN. |
6208 | ||
6209 | PATTERN is the address of the pattern string. | |
6210 | ||
6211 | CFLAGS is a series of bits which affect compilation. | |
6212 | ||
6213 | If REG_EXTENDED is set, we use POSIX extended syntax; otherwise, we | |
6214 | use POSIX basic syntax. | |
6215 | ||
6216 | If REG_NEWLINE is set, then . and [^...] don't match newline. | |
6217 | Also, regexec will try a match beginning after every newline. | |
6218 | ||
6219 | If REG_ICASE is set, then we considers upper- and lowercase | |
6220 | versions of letters to be equivalent when matching. | |
6221 | ||
6222 | If REG_NOSUB is set, then when PREG is passed to regexec, that | |
6223 | routine will report only success or failure, and nothing about the | |
6224 | registers. | |
6225 | ||
b18215fc | 6226 | It returns 0 if it succeeds, nonzero if it doesn't. (See regex.h for |
fa9a63c5 RM |
6227 | the return codes and their meanings.) */ |
6228 | ||
6229 | int | |
6230 | regcomp (preg, pattern, cflags) | |
ada30c0e SM |
6231 | regex_t *__restrict preg; |
6232 | const char *__restrict pattern; | |
fa9a63c5 RM |
6233 | int cflags; |
6234 | { | |
6235 | reg_errcode_t ret; | |
4bb91c68 | 6236 | reg_syntax_t syntax |
fa9a63c5 RM |
6237 | = (cflags & REG_EXTENDED) ? |
6238 | RE_SYNTAX_POSIX_EXTENDED : RE_SYNTAX_POSIX_BASIC; | |
6239 | ||
6240 | /* regex_compile will allocate the space for the compiled pattern. */ | |
6241 | preg->buffer = 0; | |
6242 | preg->allocated = 0; | |
6243 | preg->used = 0; | |
5e69f11e | 6244 | |
c0f9ea08 SM |
6245 | /* Try to allocate space for the fastmap. */ |
6246 | preg->fastmap = (char *) malloc (1 << BYTEWIDTH); | |
5e69f11e | 6247 | |
fa9a63c5 RM |
6248 | if (cflags & REG_ICASE) |
6249 | { | |
6250 | unsigned i; | |
5e69f11e | 6251 | |
6676cb1c RS |
6252 | preg->translate |
6253 | = (RE_TRANSLATE_TYPE) malloc (CHAR_SET_SIZE | |
6254 | * sizeof (*(RE_TRANSLATE_TYPE)0)); | |
fa9a63c5 | 6255 | if (preg->translate == NULL) |
0b32bf0e | 6256 | return (int) REG_ESPACE; |
fa9a63c5 RM |
6257 | |
6258 | /* Map uppercase characters to corresponding lowercase ones. */ | |
6259 | for (i = 0; i < CHAR_SET_SIZE; i++) | |
4bb91c68 | 6260 | preg->translate[i] = ISUPPER (i) ? TOLOWER (i) : i; |
fa9a63c5 RM |
6261 | } |
6262 | else | |
6263 | preg->translate = NULL; | |
6264 | ||
6265 | /* If REG_NEWLINE is set, newlines are treated differently. */ | |
6266 | if (cflags & REG_NEWLINE) | |
6267 | { /* REG_NEWLINE implies neither . nor [^...] match newline. */ | |
6268 | syntax &= ~RE_DOT_NEWLINE; | |
6269 | syntax |= RE_HAT_LISTS_NOT_NEWLINE; | |
fa9a63c5 RM |
6270 | } |
6271 | else | |
c0f9ea08 | 6272 | syntax |= RE_NO_NEWLINE_ANCHOR; |
fa9a63c5 RM |
6273 | |
6274 | preg->no_sub = !!(cflags & REG_NOSUB); | |
6275 | ||
5e69f11e | 6276 | /* POSIX says a null character in the pattern terminates it, so we |
fa9a63c5 | 6277 | can use strlen here in compiling the pattern. */ |
4bb91c68 | 6278 | ret = regex_compile ((re_char*) pattern, strlen (pattern), syntax, preg); |
5e69f11e | 6279 | |
fa9a63c5 RM |
6280 | /* POSIX doesn't distinguish between an unmatched open-group and an |
6281 | unmatched close-group: both are REG_EPAREN. */ | |
c0f9ea08 SM |
6282 | if (ret == REG_ERPAREN) |
6283 | ret = REG_EPAREN; | |
6284 | ||
6285 | if (ret == REG_NOERROR && preg->fastmap) | |
6286 | { /* Compute the fastmap now, since regexec cannot modify the pattern | |
6287 | buffer. */ | |
6288 | re_compile_fastmap (preg); | |
6289 | if (preg->can_be_null) | |
6290 | { /* The fastmap can't be used anyway. */ | |
6291 | free (preg->fastmap); | |
6292 | preg->fastmap = NULL; | |
6293 | } | |
6294 | } | |
fa9a63c5 RM |
6295 | return (int) ret; |
6296 | } | |
c0f9ea08 | 6297 | WEAK_ALIAS (__regcomp, regcomp) |
fa9a63c5 RM |
6298 | |
6299 | ||
6300 | /* regexec searches for a given pattern, specified by PREG, in the | |
6301 | string STRING. | |
5e69f11e | 6302 | |
fa9a63c5 | 6303 | If NMATCH is zero or REG_NOSUB was set in the cflags argument to |
b18215fc | 6304 | `regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at |
fa9a63c5 RM |
6305 | least NMATCH elements, and we set them to the offsets of the |
6306 | corresponding matched substrings. | |
5e69f11e | 6307 | |
fa9a63c5 RM |
6308 | EFLAGS specifies `execution flags' which affect matching: if |
6309 | REG_NOTBOL is set, then ^ does not match at the beginning of the | |
6310 | string; if REG_NOTEOL is set, then $ does not match at the end. | |
5e69f11e | 6311 | |
fa9a63c5 RM |
6312 | We return 0 if we find a match and REG_NOMATCH if not. */ |
6313 | ||
6314 | int | |
6315 | regexec (preg, string, nmatch, pmatch, eflags) | |
ada30c0e SM |
6316 | const regex_t *__restrict preg; |
6317 | const char *__restrict string; | |
5e69f11e | 6318 | size_t nmatch; |
9f2dbe01 | 6319 | regmatch_t pmatch[__restrict_arr]; |
fa9a63c5 RM |
6320 | int eflags; |
6321 | { | |
6322 | int ret; | |
6323 | struct re_registers regs; | |
6324 | regex_t private_preg; | |
6325 | int len = strlen (string); | |
c0f9ea08 | 6326 | boolean want_reg_info = !preg->no_sub && nmatch > 0 && pmatch; |
fa9a63c5 RM |
6327 | |
6328 | private_preg = *preg; | |
5e69f11e | 6329 | |
fa9a63c5 RM |
6330 | private_preg.not_bol = !!(eflags & REG_NOTBOL); |
6331 | private_preg.not_eol = !!(eflags & REG_NOTEOL); | |
5e69f11e | 6332 | |
fa9a63c5 RM |
6333 | /* The user has told us exactly how many registers to return |
6334 | information about, via `nmatch'. We have to pass that on to the | |
b18215fc | 6335 | matching routines. */ |
fa9a63c5 | 6336 | private_preg.regs_allocated = REGS_FIXED; |
5e69f11e | 6337 | |
fa9a63c5 RM |
6338 | if (want_reg_info) |
6339 | { | |
6340 | regs.num_regs = nmatch; | |
4bb91c68 SM |
6341 | regs.start = TALLOC (nmatch * 2, regoff_t); |
6342 | if (regs.start == NULL) | |
0b32bf0e | 6343 | return (int) REG_NOMATCH; |
4bb91c68 | 6344 | regs.end = regs.start + nmatch; |
fa9a63c5 RM |
6345 | } |
6346 | ||
c0f9ea08 SM |
6347 | /* Instead of using not_eol to implement REG_NOTEOL, we could simply |
6348 | pass (&private_preg, string, len + 1, 0, len, ...) pretending the string | |
6349 | was a little bit longer but still only matching the real part. | |
6350 | This works because the `endline' will check for a '\n' and will find a | |
6351 | '\0', correctly deciding that this is not the end of a line. | |
6352 | But it doesn't work out so nicely for REG_NOTBOL, since we don't have | |
6353 | a convenient '\0' there. For all we know, the string could be preceded | |
6354 | by '\n' which would throw things off. */ | |
6355 | ||
fa9a63c5 RM |
6356 | /* Perform the searching operation. */ |
6357 | ret = re_search (&private_preg, string, len, | |
0b32bf0e SM |
6358 | /* start: */ 0, /* range: */ len, |
6359 | want_reg_info ? ®s : (struct re_registers *) 0); | |
5e69f11e | 6360 | |
fa9a63c5 RM |
6361 | /* Copy the register information to the POSIX structure. */ |
6362 | if (want_reg_info) | |
6363 | { | |
6364 | if (ret >= 0) | |
0b32bf0e SM |
6365 | { |
6366 | unsigned r; | |
fa9a63c5 | 6367 | |
0b32bf0e SM |
6368 | for (r = 0; r < nmatch; r++) |
6369 | { | |
6370 | pmatch[r].rm_so = regs.start[r]; | |
6371 | pmatch[r].rm_eo = regs.end[r]; | |
6372 | } | |
6373 | } | |
fa9a63c5 | 6374 | |
b18215fc | 6375 | /* If we needed the temporary register info, free the space now. */ |
fa9a63c5 | 6376 | free (regs.start); |
fa9a63c5 RM |
6377 | } |
6378 | ||
6379 | /* We want zero return to mean success, unlike `re_search'. */ | |
6380 | return ret >= 0 ? (int) REG_NOERROR : (int) REG_NOMATCH; | |
6381 | } | |
c0f9ea08 | 6382 | WEAK_ALIAS (__regexec, regexec) |
fa9a63c5 RM |
6383 | |
6384 | ||
6385 | /* Returns a message corresponding to an error code, ERRCODE, returned | |
6386 | from either regcomp or regexec. We don't use PREG here. */ | |
6387 | ||
6388 | size_t | |
6389 | regerror (errcode, preg, errbuf, errbuf_size) | |
6390 | int errcode; | |
6391 | const regex_t *preg; | |
6392 | char *errbuf; | |
6393 | size_t errbuf_size; | |
6394 | { | |
6395 | const char *msg; | |
6396 | size_t msg_size; | |
6397 | ||
6398 | if (errcode < 0 | |
6399 | || errcode >= (sizeof (re_error_msgid) / sizeof (re_error_msgid[0]))) | |
5e69f11e | 6400 | /* Only error codes returned by the rest of the code should be passed |
b18215fc | 6401 | to this routine. If we are given anything else, or if other regex |
fa9a63c5 RM |
6402 | code generates an invalid error code, then the program has a bug. |
6403 | Dump core so we can fix it. */ | |
6404 | abort (); | |
6405 | ||
6406 | msg = gettext (re_error_msgid[errcode]); | |
6407 | ||
6408 | msg_size = strlen (msg) + 1; /* Includes the null. */ | |
5e69f11e | 6409 | |
fa9a63c5 RM |
6410 | if (errbuf_size != 0) |
6411 | { | |
6412 | if (msg_size > errbuf_size) | |
0b32bf0e SM |
6413 | { |
6414 | strncpy (errbuf, msg, errbuf_size - 1); | |
6415 | errbuf[errbuf_size - 1] = 0; | |
6416 | } | |
fa9a63c5 | 6417 | else |
0b32bf0e | 6418 | strcpy (errbuf, msg); |
fa9a63c5 RM |
6419 | } |
6420 | ||
6421 | return msg_size; | |
6422 | } | |
c0f9ea08 | 6423 | WEAK_ALIAS (__regerror, regerror) |
fa9a63c5 RM |
6424 | |
6425 | ||
6426 | /* Free dynamically allocated space used by PREG. */ | |
6427 | ||
6428 | void | |
6429 | regfree (preg) | |
6430 | regex_t *preg; | |
6431 | { | |
6432 | if (preg->buffer != NULL) | |
6433 | free (preg->buffer); | |
6434 | preg->buffer = NULL; | |
5e69f11e | 6435 | |
fa9a63c5 RM |
6436 | preg->allocated = 0; |
6437 | preg->used = 0; | |
6438 | ||
6439 | if (preg->fastmap != NULL) | |
6440 | free (preg->fastmap); | |
6441 | preg->fastmap = NULL; | |
6442 | preg->fastmap_accurate = 0; | |
6443 | ||
6444 | if (preg->translate != NULL) | |
6445 | free (preg->translate); | |
6446 | preg->translate = NULL; | |
6447 | } | |
c0f9ea08 | 6448 | WEAK_ALIAS (__regfree, regfree) |
fa9a63c5 RM |
6449 | |
6450 | #endif /* not emacs */ | |
ab5796a9 MB |
6451 | |
6452 | /* arch-tag: 4ffd68ba-2a9e-435b-a21a-018990f9eeb2 | |
6453 | (do not change this comment) */ |