HCoop / bpt / emacs.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
(struct glyph_string) [USE_FONT_BACKEND]: New
[bpt/emacs.git] / src / regex.c
1 /* Extended regular expression matching and search library, version
2 0.12. (Implements POSIX draft P1003.2/D11.2, except for some of the
3 internationalization features.)
4
5 Copyright (C) 1993, 1994, 1995, 1996, 1997, 1998, 1999, 2000, 2001,
6 2002, 2003, 2004, 2005, 2006 Free Software Foundation, Inc.
7
8 This program is free software; you can redistribute it and/or modify
9 it under the terms of the GNU General Public License as published by
10 the Free Software Foundation; either version 2, or (at your option)
11 any later version.
12
13 This program is distributed in the hope that it will be useful,
14 but WITHOUT ANY WARRANTY; without even the implied warranty of
15 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
16 GNU General Public License for more details.
17
18 You should have received a copy of the GNU General Public License
19 along with this program; if not, write to the Free Software
20 Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301,
21 USA. */
22
23 /* TODO:
24 - structure the opcode space into opcode+flag.
25 - merge with glibc's regex.[ch].
26 - replace (succeed_n + jump_n + set_number_at) with something that doesn't
27 need to modify the compiled regexp so that re_match can be reentrant.
28 - get rid of on_failure_jump_smart by doing the optimization in re_comp
29 rather than at run-time, so that re_match can be reentrant.
30 */
31
32 /* AIX requires this to be the first thing in the file. */
33 #if defined _AIX && !defined REGEX_MALLOC
34 #pragma alloca
35 #endif
36
37 #ifdef HAVE_CONFIG_H
38 # include <config.h>
39 #endif
40
41 #if defined STDC_HEADERS && !defined emacs
42 # include <stddef.h>
43 #else
44 /* We need this for `regex.h', and perhaps for the Emacs include files. */
45 # include <sys/types.h>
46 #endif
47
48 /* Whether to use ISO C Amendment 1 wide char functions.
49 Those should not be used for Emacs since it uses its own. */
50 #if defined _LIBC
51 #define WIDE_CHAR_SUPPORT 1
52 #else
53 #define WIDE_CHAR_SUPPORT \
54 (HAVE_WCTYPE_H && HAVE_WCHAR_H && HAVE_BTOWC && !emacs)
55 #endif
56
57 /* For platform which support the ISO C amendement 1 functionality we
58 support user defined character classes. */
59 #if WIDE_CHAR_SUPPORT
60 /* Solaris 2.5 has a bug: <wchar.h> must be included before <wctype.h>. */
61 # include <wchar.h>
62 # include <wctype.h>
63 #endif
64
65 #ifdef _LIBC
66 /* We have to keep the namespace clean. */
67 # define regfree(preg) __regfree (preg)
68 # define regexec(pr, st, nm, pm, ef) __regexec (pr, st, nm, pm, ef)
69 # define regcomp(preg, pattern, cflags) __regcomp (preg, pattern, cflags)
70 # define regerror(errcode, preg, errbuf, errbuf_size) \
71 __regerror(errcode, preg, errbuf, errbuf_size)
72 # define re_set_registers(bu, re, nu, st, en) \
73 __re_set_registers (bu, re, nu, st, en)
74 # define re_match_2(bufp, string1, size1, string2, size2, pos, regs, stop) \
75 __re_match_2 (bufp, string1, size1, string2, size2, pos, regs, stop)
76 # define re_match(bufp, string, size, pos, regs) \
77 __re_match (bufp, string, size, pos, regs)
78 # define re_search(bufp, string, size, startpos, range, regs) \
79 __re_search (bufp, string, size, startpos, range, regs)
80 # define re_compile_pattern(pattern, length, bufp) \
81 __re_compile_pattern (pattern, length, bufp)
82 # define re_set_syntax(syntax) __re_set_syntax (syntax)
83 # define re_search_2(bufp, st1, s1, st2, s2, startpos, range, regs, stop) \
84 __re_search_2 (bufp, st1, s1, st2, s2, startpos, range, regs, stop)
85 # define re_compile_fastmap(bufp) __re_compile_fastmap (bufp)
86
87 /* Make sure we call libc's function even if the user overrides them. */
88 # define btowc __btowc
89 # define iswctype __iswctype
90 # define wctype __wctype
91
92 # define WEAK_ALIAS(a,b) weak_alias (a, b)
93
94 /* We are also using some library internals. */
95 # include <locale/localeinfo.h>
96 # include <locale/elem-hash.h>
97 # include <langinfo.h>
98 #else
99 # define WEAK_ALIAS(a,b)
100 #endif
101
102 /* This is for other GNU distributions with internationalized messages. */
103 #if HAVE_LIBINTL_H || defined _LIBC
104 # include <libintl.h>
105 #else
106 # define gettext(msgid) (msgid)
107 #endif
108
109 #ifndef gettext_noop
110 /* This define is so xgettext can find the internationalizable
111 strings. */
112 # define gettext_noop(String) String
113 #endif
114
115 /* The `emacs' switch turns on certain matching commands
116 that make sense only in Emacs. */
117 #ifdef emacs
118
119 # include "lisp.h"
120 # include "buffer.h"
121
122 /* Make syntax table lookup grant data in gl_state. */
123 # define SYNTAX_ENTRY_VIA_PROPERTY
124
125 # include "syntax.h"
126 # include "character.h"
127 # include "category.h"
128
129 # ifdef malloc
130 # undef malloc
131 # endif
132 # define malloc xmalloc
133 # ifdef realloc
134 # undef realloc
135 # endif
136 # define realloc xrealloc
137 # ifdef free
138 # undef free
139 # endif
140 # define free xfree
141
142 /* Converts the pointer to the char to BEG-based offset from the start. */
143 # define PTR_TO_OFFSET(d) POS_AS_IN_BUFFER (POINTER_TO_OFFSET (d))
144 # define POS_AS_IN_BUFFER(p) ((p) + (NILP (re_match_object) || BUFFERP (re_match_object)))
145
146 # define RE_MULTIBYTE_P(bufp) ((bufp)->multibyte)
147 # define RE_TARGET_MULTIBYTE_P(bufp) ((bufp)->target_multibyte)
148 # define RE_STRING_CHAR(p, s) \
149 (multibyte ? (STRING_CHAR (p, s)) : (*(p)))
150 # define RE_STRING_CHAR_AND_LENGTH(p, s, len) \
151 (multibyte ? (STRING_CHAR_AND_LENGTH (p, s, len)) : ((len) = 1, *(p)))
152
153 /* Set C a (possibly converted to multibyte) character before P. P
154 points into a string which is the virtual concatenation of STR1
155 (which ends at END1) or STR2 (which ends at END2). */
156 # define GET_CHAR_BEFORE_2(c, p, str1, end1, str2, end2) \
157 do { \
158 if (multibyte) \
159 { \
160 re_char *dtemp = (p) == (str2) ? (end1) : (p); \
161 re_char *dlimit = ((p) > (str2) && (p) <= (end2)) ? (str2) : (str1); \
162 while (dtemp-- > dlimit && !CHAR_HEAD_P (*dtemp)); \
163 c = STRING_CHAR (dtemp, (p) - dtemp); \
164 } \
165 else \
166 { \
167 (c = ((p) == (str2) ? (end1) : (p))[-1]); \
168 MAKE_CHAR_MULTIBYTE (c); \
169 } \
170 } while (0)
171
172 /* Set C a (possibly converted to multibyte) character at P, and set
173 LEN to the byte length of that character. */
174 # define GET_CHAR_AFTER(c, p, len) \
175 do { \
176 if (multibyte) \
177 c = STRING_CHAR_AND_LENGTH (p, 0, len); \
178 else \
179 { \
180 c = *p; \
181 len = 1; \
182 MAKE_CHAR_MULTIBYTE (c); \
183 } \
184 } while (0)
185
186 #else /* not emacs */
187
188 /* If we are not linking with Emacs proper,
189 we can't use the relocating allocator
190 even if config.h says that we can. */
191 # undef REL_ALLOC
192
193 # if defined STDC_HEADERS || defined _LIBC
194 # include <stdlib.h>
195 # else
196 char *malloc ();
197 char *realloc ();
198 # endif
199
200 /* When used in Emacs's lib-src, we need xmalloc and xrealloc. */
201
202 void *
203 xmalloc (size)
204 size_t size;
205 {
206 register void *val;
207 val = (void *) malloc (size);
208 if (!val && size)
209 {
210 write (2, "virtual memory exhausted\n", 25);
211 exit (1);
212 }
213 return val;
214 }
215
216 void *
217 xrealloc (block, size)
218 void *block;
219 size_t size;
220 {
221 register void *val;
222 /* We must call malloc explicitly when BLOCK is 0, since some
223 reallocs don't do this. */
224 if (! block)
225 val = (void *) malloc (size);
226 else
227 val = (void *) realloc (block, size);
228 if (!val && size)
229 {
230 write (2, "virtual memory exhausted\n", 25);
231 exit (1);
232 }
233 return val;
234 }
235
236 # ifdef malloc
237 # undef malloc
238 # endif
239 # define malloc xmalloc
240 # ifdef realloc
241 # undef realloc
242 # endif
243 # define realloc xrealloc
244
245 /* When used in Emacs's lib-src, we need to get bzero and bcopy somehow.
246 If nothing else has been done, use the method below. */
247 # ifdef INHIBIT_STRING_HEADER
248 # if !(defined HAVE_BZERO && defined HAVE_BCOPY)
249 # if !defined bzero && !defined bcopy
250 # undef INHIBIT_STRING_HEADER
251 # endif
252 # endif
253 # endif
254
255 /* This is the normal way of making sure we have memcpy, memcmp and bzero.
256 This is used in most programs--a few other programs avoid this
257 by defining INHIBIT_STRING_HEADER. */
258 # ifndef INHIBIT_STRING_HEADER
259 # if defined HAVE_STRING_H || defined STDC_HEADERS || defined _LIBC
260 # include <string.h>
261 # ifndef bzero
262 # ifndef _LIBC
263 # define bzero(s, n) (memset (s, '\0', n), (s))
264 # else
265 # define bzero(s, n) __bzero (s, n)
266 # endif
267 # endif
268 # else
269 # include <strings.h>
270 # ifndef memcmp
271 # define memcmp(s1, s2, n) bcmp (s1, s2, n)
272 # endif
273 # ifndef memcpy
274 # define memcpy(d, s, n) (bcopy (s, d, n), (d))
275 # endif
276 # endif
277 # endif
278
279 /* Define the syntax stuff for \<, \>, etc. */
280
281 /* Sword must be nonzero for the wordchar pattern commands in re_match_2. */
282 enum syntaxcode { Swhitespace = 0, Sword = 1, Ssymbol = 2 };
283
284 # ifdef SWITCH_ENUM_BUG
285 # define SWITCH_ENUM_CAST(x) ((int)(x))
286 # else
287 # define SWITCH_ENUM_CAST(x) (x)
288 # endif
289
290 /* Dummy macros for non-Emacs environments. */
291 # define BASE_LEADING_CODE_P(c) (0)
292 # define CHAR_CHARSET(c) 0
293 # define CHARSET_LEADING_CODE_BASE(c) 0
294 # define MAX_MULTIBYTE_LENGTH 1
295 # define RE_MULTIBYTE_P(x) 0
296 # define RE_TARGET_MULTIBYTE_P(x) 0
297 # define WORD_BOUNDARY_P(c1, c2) (0)
298 # define CHAR_HEAD_P(p) (1)
299 # define SINGLE_BYTE_CHAR_P(c) (1)
300 # define SAME_CHARSET_P(c1, c2) (1)
301 # define MULTIBYTE_FORM_LENGTH(p, s) (1)
302 # define PREV_CHAR_BOUNDARY(p, limit) ((p)--)
303 # define STRING_CHAR(p, s) (*(p))
304 # define RE_STRING_CHAR STRING_CHAR
305 # define CHAR_STRING(c, s) (*(s) = (c), 1)
306 # define STRING_CHAR_AND_LENGTH(p, s, actual_len) ((actual_len) = 1, *(p))
307 # define RE_STRING_CHAR_AND_LENGTH STRING_CHAR_AND_LENGTH
308 # define GET_CHAR_BEFORE_2(c, p, str1, end1, str2, end2) \
309 (c = ((p) == (str2) ? *((end1) - 1) : *((p) - 1)))
310 # define GET_CHAR_AFTER(c, p, len) \
311 (c = *p, len = 1)
312 # define MAKE_CHAR(charset, c1, c2) (c1)
313 # define BYTE8_TO_CHAR(c) (c)
314 # define CHAR_BYTE8_P(c) (0)
315 # define MAKE_CHAR_MULTIBYTE(c) (c)
316 # define MAKE_CHAR_UNIBYTE(c) (c)
317 # define CHAR_LEADING_CODE(c) (c)
318
319 #endif /* not emacs */
320
321 #ifndef RE_TRANSLATE
322 # define RE_TRANSLATE(TBL, C) ((unsigned char)(TBL)[C])
323 # define RE_TRANSLATE_P(TBL) (TBL)
324 #endif
325 \f
326 /* Get the interface, including the syntax bits. */
327 #include "regex.h"
328
329 /* isalpha etc. are used for the character classes. */
330 #include <ctype.h>
331
332 #ifdef emacs
333
334 /* 1 if C is an ASCII character. */
335 # define IS_REAL_ASCII(c) ((c) < 0200)
336
337 /* 1 if C is a unibyte character. */
338 # define ISUNIBYTE(c) (SINGLE_BYTE_CHAR_P ((c)))
339
340 /* The Emacs definitions should not be directly affected by locales. */
341
342 /* In Emacs, these are only used for single-byte characters. */
343 # define ISDIGIT(c) ((c) >= '0' && (c) <= '9')
344 # define ISCNTRL(c) ((c) < ' ')
345 # define ISXDIGIT(c) (((c) >= '0' && (c) <= '9') \
346 || ((c) >= 'a' && (c) <= 'f') \
347 || ((c) >= 'A' && (c) <= 'F'))
348
349 /* This is only used for single-byte characters. */
350 # define ISBLANK(c) ((c) == ' ' || (c) == '\t')
351
352 /* The rest must handle multibyte characters. */
353
354 # define ISGRAPH(c) (SINGLE_BYTE_CHAR_P (c) \
355 ? (c) > ' ' && !((c) >= 0177 && (c) <= 0237) \
356 : 1)
357
358 # define ISPRINT(c) (SINGLE_BYTE_CHAR_P (c) \
359 ? (c) >= ' ' && !((c) >= 0177 && (c) <= 0237) \
360 : 1)
361
362 # define ISALNUM(c) (IS_REAL_ASCII (c) \
363 ? (((c) >= 'a' && (c) <= 'z') \
364 || ((c) >= 'A' && (c) <= 'Z') \
365 || ((c) >= '0' && (c) <= '9')) \
366 : SYNTAX (c) == Sword)
367
368 # define ISALPHA(c) (IS_REAL_ASCII (c) \
369 ? (((c) >= 'a' && (c) <= 'z') \
370 || ((c) >= 'A' && (c) <= 'Z')) \
371 : SYNTAX (c) == Sword)
372
373 # define ISLOWER(c) (LOWERCASEP (c))
374
375 # define ISPUNCT(c) (IS_REAL_ASCII (c) \
376 ? ((c) > ' ' && (c) < 0177 \
377 && !(((c) >= 'a' && (c) <= 'z') \
378 || ((c) >= 'A' && (c) <= 'Z') \
379 || ((c) >= '0' && (c) <= '9'))) \
380 : SYNTAX (c) != Sword)
381
382 # define ISSPACE(c) (SYNTAX (c) == Swhitespace)
383
384 # define ISUPPER(c) (UPPERCASEP (c))
385
386 # define ISWORD(c) (SYNTAX (c) == Sword)
387
388 #else /* not emacs */
389
390 /* Jim Meyering writes:
391
392 "... Some ctype macros are valid only for character codes that
393 isascii says are ASCII (SGI's IRIX-4.0.5 is one such system --when
394 using /bin/cc or gcc but without giving an ansi option). So, all
395 ctype uses should be through macros like ISPRINT... If
396 STDC_HEADERS is defined, then autoconf has verified that the ctype
397 macros don't need to be guarded with references to isascii. ...
398 Defining isascii to 1 should let any compiler worth its salt
399 eliminate the && through constant folding."
400 Solaris defines some of these symbols so we must undefine them first. */
401
402 # undef ISASCII
403 # if defined STDC_HEADERS || (!defined isascii && !defined HAVE_ISASCII)
404 # define ISASCII(c) 1
405 # else
406 # define ISASCII(c) isascii(c)
407 # endif
408
409 /* 1 if C is an ASCII character. */
410 # define IS_REAL_ASCII(c) ((c) < 0200)
411
412 /* This distinction is not meaningful, except in Emacs. */
413 # define ISUNIBYTE(c) 1
414
415 # ifdef isblank
416 # define ISBLANK(c) (ISASCII (c) && isblank (c))
417 # else
418 # define ISBLANK(c) ((c) == ' ' || (c) == '\t')
419 # endif
420 # ifdef isgraph
421 # define ISGRAPH(c) (ISASCII (c) && isgraph (c))
422 # else
423 # define ISGRAPH(c) (ISASCII (c) && isprint (c) && !isspace (c))
424 # endif
425
426 # undef ISPRINT
427 # define ISPRINT(c) (ISASCII (c) && isprint (c))
428 # define ISDIGIT(c) (ISASCII (c) && isdigit (c))
429 # define ISALNUM(c) (ISASCII (c) && isalnum (c))
430 # define ISALPHA(c) (ISASCII (c) && isalpha (c))
431 # define ISCNTRL(c) (ISASCII (c) && iscntrl (c))
432 # define ISLOWER(c) (ISASCII (c) && islower (c))
433 # define ISPUNCT(c) (ISASCII (c) && ispunct (c))
434 # define ISSPACE(c) (ISASCII (c) && isspace (c))
435 # define ISUPPER(c) (ISASCII (c) && isupper (c))
436 # define ISXDIGIT(c) (ISASCII (c) && isxdigit (c))
437
438 # define ISWORD(c) ISALPHA(c)
439
440 # ifdef _tolower
441 # define TOLOWER(c) _tolower(c)
442 # else
443 # define TOLOWER(c) tolower(c)
444 # endif
445
446 /* How many characters in the character set. */
447 # define CHAR_SET_SIZE 256
448
449 # ifdef SYNTAX_TABLE
450
451 extern char *re_syntax_table;
452
453 # else /* not SYNTAX_TABLE */
454
455 static char re_syntax_table[CHAR_SET_SIZE];
456
457 static void
458 init_syntax_once ()
459 {
460 register int c;
461 static int done = 0;
462
463 if (done)
464 return;
465
466 bzero (re_syntax_table, sizeof re_syntax_table);
467
468 for (c = 0; c < CHAR_SET_SIZE; ++c)
469 if (ISALNUM (c))
470 re_syntax_table[c] = Sword;
471
472 re_syntax_table['_'] = Ssymbol;
473
474 done = 1;
475 }
476
477 # endif /* not SYNTAX_TABLE */
478
479 # define SYNTAX(c) re_syntax_table[(c)]
480
481 #endif /* not emacs */
482 \f
483 #ifndef NULL
484 # define NULL (void *)0
485 #endif
486
487 /* We remove any previous definition of `SIGN_EXTEND_CHAR',
488 since ours (we hope) works properly with all combinations of
489 machines, compilers, `char' and `unsigned char' argument types.
490 (Per Bothner suggested the basic approach.) */
491 #undef SIGN_EXTEND_CHAR
492 #if __STDC__
493 # define SIGN_EXTEND_CHAR(c) ((signed char) (c))
494 #else /* not __STDC__ */
495 /* As in Harbison and Steele. */
496 # define SIGN_EXTEND_CHAR(c) ((((unsigned char) (c)) ^ 128) - 128)
497 #endif
498 \f
499 /* Should we use malloc or alloca? If REGEX_MALLOC is not defined, we
500 use `alloca' instead of `malloc'. This is because using malloc in
501 re_search* or re_match* could cause memory leaks when C-g is used in
502 Emacs; also, malloc is slower and causes storage fragmentation. On
503 the other hand, malloc is more portable, and easier to debug.
504
505 Because we sometimes use alloca, some routines have to be macros,
506 not functions -- `alloca'-allocated space disappears at the end of the
507 function it is called in. */
508
509 #ifdef REGEX_MALLOC
510
511 # define REGEX_ALLOCATE malloc
512 # define REGEX_REALLOCATE(source, osize, nsize) realloc (source, nsize)
513 # define REGEX_FREE free
514
515 #else /* not REGEX_MALLOC */
516
517 /* Emacs already defines alloca, sometimes. */
518 # ifndef alloca
519
520 /* Make alloca work the best possible way. */
521 # ifdef __GNUC__
522 # define alloca __builtin_alloca
523 # else /* not __GNUC__ */
524 # ifdef HAVE_ALLOCA_H
525 # include <alloca.h>
526 # endif /* HAVE_ALLOCA_H */
527 # endif /* not __GNUC__ */
528
529 # endif /* not alloca */
530
531 # define REGEX_ALLOCATE alloca
532
533 /* Assumes a `char *destination' variable. */
534 # define REGEX_REALLOCATE(source, osize, nsize) \
535 (destination = (char *) alloca (nsize), \
536 memcpy (destination, source, osize))
537
538 /* No need to do anything to free, after alloca. */
539 # define REGEX_FREE(arg) ((void)0) /* Do nothing! But inhibit gcc warning. */
540
541 #endif /* not REGEX_MALLOC */
542
543 /* Define how to allocate the failure stack. */
544
545 #if defined REL_ALLOC && defined REGEX_MALLOC
546
547 # define REGEX_ALLOCATE_STACK(size) \
548 r_alloc (&failure_stack_ptr, (size))
549 # define REGEX_REALLOCATE_STACK(source, osize, nsize) \
550 r_re_alloc (&failure_stack_ptr, (nsize))
551 # define REGEX_FREE_STACK(ptr) \
552 r_alloc_free (&failure_stack_ptr)
553
554 #else /* not using relocating allocator */
555
556 # ifdef REGEX_MALLOC
557
558 # define REGEX_ALLOCATE_STACK malloc
559 # define REGEX_REALLOCATE_STACK(source, osize, nsize) realloc (source, nsize)
560 # define REGEX_FREE_STACK free
561
562 # else /* not REGEX_MALLOC */
563
564 # define REGEX_ALLOCATE_STACK alloca
565
566 # define REGEX_REALLOCATE_STACK(source, osize, nsize) \
567 REGEX_REALLOCATE (source, osize, nsize)
568 /* No need to explicitly free anything. */
569 # define REGEX_FREE_STACK(arg) ((void)0)
570
571 # endif /* not REGEX_MALLOC */
572 #endif /* not using relocating allocator */
573
574
575 /* True if `size1' is non-NULL and PTR is pointing anywhere inside
576 `string1' or just past its end. This works if PTR is NULL, which is
577 a good thing. */
578 #define FIRST_STRING_P(ptr) \
579 (size1 && string1 <= (ptr) && (ptr) <= string1 + size1)
580
581 /* (Re)Allocate N items of type T using malloc, or fail. */
582 #define TALLOC(n, t) ((t *) malloc ((n) * sizeof (t)))
583 #define RETALLOC(addr, n, t) ((addr) = (t *) realloc (addr, (n) * sizeof (t)))
584 #define RETALLOC_IF(addr, n, t) \
585 if (addr) RETALLOC((addr), (n), t); else (addr) = TALLOC ((n), t)
586 #define REGEX_TALLOC(n, t) ((t *) REGEX_ALLOCATE ((n) * sizeof (t)))
587
588 #define BYTEWIDTH 8 /* In bits. */
589
590 #define STREQ(s1, s2) ((strcmp (s1, s2) == 0))
591
592 #undef MAX
593 #undef MIN
594 #define MAX(a, b) ((a) > (b) ? (a) : (b))
595 #define MIN(a, b) ((a) < (b) ? (a) : (b))
596
597 /* Type of source-pattern and string chars. */
598 typedef const unsigned char re_char;
599
600 typedef char boolean;
601 #define false 0
602 #define true 1
603
604 static int re_match_2_internal _RE_ARGS ((struct re_pattern_buffer *bufp,
605 re_char *string1, int size1,
606 re_char *string2, int size2,
607 int pos,
608 struct re_registers *regs,
609 int stop));
610 \f
611 /* These are the command codes that appear in compiled regular
612 expressions. Some opcodes are followed by argument bytes. A
613 command code can specify any interpretation whatsoever for its
614 arguments. Zero bytes may appear in the compiled regular expression. */
615
616 typedef enum
617 {
618 no_op = 0,
619
620 /* Succeed right away--no more backtracking. */
621 succeed,
622
623 /* Followed by one byte giving n, then by n literal bytes. */
624 exactn,
625
626 /* Matches any (more or less) character. */
627 anychar,
628
629 /* Matches any one char belonging to specified set. First
630 following byte is number of bitmap bytes. Then come bytes
631 for a bitmap saying which chars are in. Bits in each byte
632 are ordered low-bit-first. A character is in the set if its
633 bit is 1. A character too large to have a bit in the map is
634 automatically not in the set.
635
636 If the length byte has the 0x80 bit set, then that stuff
637 is followed by a range table:
638 2 bytes of flags for character sets (low 8 bits, high 8 bits)
639 See RANGE_TABLE_WORK_BITS below.
640 2 bytes, the number of pairs that follow (upto 32767)
641 pairs, each 2 multibyte characters,
642 each multibyte character represented as 3 bytes. */
643 charset,
644
645 /* Same parameters as charset, but match any character that is
646 not one of those specified. */
647 charset_not,
648
649 /* Start remembering the text that is matched, for storing in a
650 register. Followed by one byte with the register number, in
651 the range 0 to one less than the pattern buffer's re_nsub
652 field. */
653 start_memory,
654
655 /* Stop remembering the text that is matched and store it in a
656 memory register. Followed by one byte with the register
657 number, in the range 0 to one less than `re_nsub' in the
658 pattern buffer. */
659 stop_memory,
660
661 /* Match a duplicate of something remembered. Followed by one
662 byte containing the register number. */
663 duplicate,
664
665 /* Fail unless at beginning of line. */
666 begline,
667
668 /* Fail unless at end of line. */
669 endline,
670
671 /* Succeeds if at beginning of buffer (if emacs) or at beginning
672 of string to be matched (if not). */
673 begbuf,
674
675 /* Analogously, for end of buffer/string. */
676 endbuf,
677
678 /* Followed by two byte relative address to which to jump. */
679 jump,
680
681 /* Followed by two-byte relative address of place to resume at
682 in case of failure. */
683 on_failure_jump,
684
685 /* Like on_failure_jump, but pushes a placeholder instead of the
686 current string position when executed. */
687 on_failure_keep_string_jump,
688
689 /* Just like `on_failure_jump', except that it checks that we
690 don't get stuck in an infinite loop (matching an empty string
691 indefinitely). */
692 on_failure_jump_loop,
693
694 /* Just like `on_failure_jump_loop', except that it checks for
695 a different kind of loop (the kind that shows up with non-greedy
696 operators). This operation has to be immediately preceded
697 by a `no_op'. */
698 on_failure_jump_nastyloop,
699
700 /* A smart `on_failure_jump' used for greedy * and + operators.
701 It analyses the loop before which it is put and if the
702 loop does not require backtracking, it changes itself to
703 `on_failure_keep_string_jump' and short-circuits the loop,
704 else it just defaults to changing itself into `on_failure_jump'.
705 It assumes that it is pointing to just past a `jump'. */
706 on_failure_jump_smart,
707
708 /* Followed by two-byte relative address and two-byte number n.
709 After matching N times, jump to the address upon failure.
710 Does not work if N starts at 0: use on_failure_jump_loop
711 instead. */
712 succeed_n,
713
714 /* Followed by two-byte relative address, and two-byte number n.
715 Jump to the address N times, then fail. */
716 jump_n,
717
718 /* Set the following two-byte relative address to the
719 subsequent two-byte number. The address *includes* the two
720 bytes of number. */
721 set_number_at,
722
723 wordbeg, /* Succeeds if at word beginning. */
724 wordend, /* Succeeds if at word end. */
725
726 wordbound, /* Succeeds if at a word boundary. */
727 notwordbound, /* Succeeds if not at a word boundary. */
728
729 symbeg, /* Succeeds if at symbol beginning. */
730 symend, /* Succeeds if at symbol end. */
731
732 /* Matches any character whose syntax is specified. Followed by
733 a byte which contains a syntax code, e.g., Sword. */
734 syntaxspec,
735
736 /* Matches any character whose syntax is not that specified. */
737 notsyntaxspec
738
739 #ifdef emacs
740 ,before_dot, /* Succeeds if before point. */
741 at_dot, /* Succeeds if at point. */
742 after_dot, /* Succeeds if after point. */
743
744 /* Matches any character whose category-set contains the specified
745 category. The operator is followed by a byte which contains a
746 category code (mnemonic ASCII character). */
747 categoryspec,
748
749 /* Matches any character whose category-set does not contain the
750 specified category. The operator is followed by a byte which
751 contains the category code (mnemonic ASCII character). */
752 notcategoryspec
753 #endif /* emacs */
754 } re_opcode_t;
755 \f
756 /* Common operations on the compiled pattern. */
757
758 /* Store NUMBER in two contiguous bytes starting at DESTINATION. */
759
760 #define STORE_NUMBER(destination, number) \
761 do { \
762 (destination)[0] = (number) & 0377; \
763 (destination)[1] = (number) >> 8; \
764 } while (0)
765
766 /* Same as STORE_NUMBER, except increment DESTINATION to
767 the byte after where the number is stored. Therefore, DESTINATION
768 must be an lvalue. */
769
770 #define STORE_NUMBER_AND_INCR(destination, number) \
771 do { \
772 STORE_NUMBER (destination, number); \
773 (destination) += 2; \
774 } while (0)
775
776 /* Put into DESTINATION a number stored in two contiguous bytes starting
777 at SOURCE. */
778
779 #define EXTRACT_NUMBER(destination, source) \
780 do { \
781 (destination) = *(source) & 0377; \
782 (destination) += SIGN_EXTEND_CHAR (*((source) + 1)) << 8; \
783 } while (0)
784
785 #ifdef DEBUG
786 static void extract_number _RE_ARGS ((int *dest, re_char *source));
787 static void
788 extract_number (dest, source)
789 int *dest;
790 re_char *source;
791 {
792 int temp = SIGN_EXTEND_CHAR (*(source + 1));
793 *dest = *source & 0377;
794 *dest += temp << 8;
795 }
796
797 # ifndef EXTRACT_MACROS /* To debug the macros. */
798 # undef EXTRACT_NUMBER
799 # define EXTRACT_NUMBER(dest, src) extract_number (&dest, src)
800 # endif /* not EXTRACT_MACROS */
801
802 #endif /* DEBUG */
803
804 /* Same as EXTRACT_NUMBER, except increment SOURCE to after the number.
805 SOURCE must be an lvalue. */
806
807 #define EXTRACT_NUMBER_AND_INCR(destination, source) \
808 do { \
809 EXTRACT_NUMBER (destination, source); \
810 (source) += 2; \
811 } while (0)
812
813 #ifdef DEBUG
814 static void extract_number_and_incr _RE_ARGS ((int *destination,
815 re_char **source));
816 static void
817 extract_number_and_incr (destination, source)
818 int *destination;
819 re_char **source;
820 {
821 extract_number (destination, *source);
822 *source += 2;
823 }
824
825 # ifndef EXTRACT_MACROS
826 # undef EXTRACT_NUMBER_AND_INCR
827 # define EXTRACT_NUMBER_AND_INCR(dest, src) \
828 extract_number_and_incr (&dest, &src)
829 # endif /* not EXTRACT_MACROS */
830
831 #endif /* DEBUG */
832 \f
833 /* Store a multibyte character in three contiguous bytes starting
834 DESTINATION, and increment DESTINATION to the byte after where the
835 character is stored. Therefore, DESTINATION must be an lvalue. */
836
837 #define STORE_CHARACTER_AND_INCR(destination, character) \
838 do { \
839 (destination)[0] = (character) & 0377; \
840 (destination)[1] = ((character) >> 8) & 0377; \
841 (destination)[2] = (character) >> 16; \
842 (destination) += 3; \
843 } while (0)
844
845 /* Put into DESTINATION a character stored in three contiguous bytes
846 starting at SOURCE. */
847
848 #define EXTRACT_CHARACTER(destination, source) \
849 do { \
850 (destination) = ((source)[0] \
851 | ((source)[1] << 8) \
852 | ((source)[2] << 16)); \
853 } while (0)
854
855
856 /* Macros for charset. */
857
858 /* Size of bitmap of charset P in bytes. P is a start of charset,
859 i.e. *P is (re_opcode_t) charset or (re_opcode_t) charset_not. */
860 #define CHARSET_BITMAP_SIZE(p) ((p)[1] & 0x7F)
861
862 /* Nonzero if charset P has range table. */
863 #define CHARSET_RANGE_TABLE_EXISTS_P(p) ((p)[1] & 0x80)
864
865 /* Return the address of range table of charset P. But not the start
866 of table itself, but the before where the number of ranges is
867 stored. `2 +' means to skip re_opcode_t and size of bitmap,
868 and the 2 bytes of flags at the start of the range table. */
869 #define CHARSET_RANGE_TABLE(p) (&(p)[4 + CHARSET_BITMAP_SIZE (p)])
870
871 /* Extract the bit flags that start a range table. */
872 #define CHARSET_RANGE_TABLE_BITS(p) \
873 ((p)[2 + CHARSET_BITMAP_SIZE (p)] \
874 + (p)[3 + CHARSET_BITMAP_SIZE (p)] * 0x100)
875
876 /* Test if C is listed in the bitmap of charset P. */
877 #define CHARSET_LOOKUP_BITMAP(p, c) \
878 ((c) < CHARSET_BITMAP_SIZE (p) * BYTEWIDTH \
879 && (p)[2 + (c) / BYTEWIDTH] & (1 << ((c) % BYTEWIDTH)))
880
881 /* Return the address of end of RANGE_TABLE. COUNT is number of
882 ranges (which is a pair of (start, end)) in the RANGE_TABLE. `* 2'
883 is start of range and end of range. `* 3' is size of each start
884 and end. */
885 #define CHARSET_RANGE_TABLE_END(range_table, count) \
886 ((range_table) + (count) * 2 * 3)
887
888 /* Test if C is in RANGE_TABLE. A flag NOT is negated if C is in.
889 COUNT is number of ranges in RANGE_TABLE. */
890 #define CHARSET_LOOKUP_RANGE_TABLE_RAW(not, c, range_table, count) \
891 do \
892 { \
893 re_wchar_t range_start, range_end; \
894 re_char *p; \
895 re_char *range_table_end \
896 = CHARSET_RANGE_TABLE_END ((range_table), (count)); \
897 \
898 for (p = (range_table); p < range_table_end; p += 2 * 3) \
899 { \
900 EXTRACT_CHARACTER (range_start, p); \
901 EXTRACT_CHARACTER (range_end, p + 3); \
902 \
903 if (range_start <= (c) && (c) <= range_end) \
904 { \
905 (not) = !(not); \
906 break; \
907 } \
908 } \
909 } \
910 while (0)
911
912 /* Test if C is in range table of CHARSET. The flag NOT is negated if
913 C is listed in it. */
914 #define CHARSET_LOOKUP_RANGE_TABLE(not, c, charset) \
915 do \
916 { \
917 /* Number of ranges in range table. */ \
918 int count; \
919 re_char *range_table = CHARSET_RANGE_TABLE (charset); \
920 \
921 EXTRACT_NUMBER_AND_INCR (count, range_table); \
922 CHARSET_LOOKUP_RANGE_TABLE_RAW ((not), (c), range_table, count); \
923 } \
924 while (0)
925 \f
926 /* If DEBUG is defined, Regex prints many voluminous messages about what
927 it is doing (if the variable `debug' is nonzero). If linked with the
928 main program in `iregex.c', you can enter patterns and strings
929 interactively. And if linked with the main program in `main.c' and
930 the other test files, you can run the already-written tests. */
931
932 #ifdef DEBUG
933
934 /* We use standard I/O for debugging. */
935 # include <stdio.h>
936
937 /* It is useful to test things that ``must'' be true when debugging. */
938 # include <assert.h>
939
940 static int debug = -100000;
941
942 # define DEBUG_STATEMENT(e) e
943 # define DEBUG_PRINT1(x) if (debug > 0) printf (x)
944 # define DEBUG_PRINT2(x1, x2) if (debug > 0) printf (x1, x2)
945 # define DEBUG_PRINT3(x1, x2, x3) if (debug > 0) printf (x1, x2, x3)
946 # define DEBUG_PRINT4(x1, x2, x3, x4) if (debug > 0) printf (x1, x2, x3, x4)
947 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) \
948 if (debug > 0) print_partial_compiled_pattern (s, e)
949 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) \
950 if (debug > 0) print_double_string (w, s1, sz1, s2, sz2)
951
952
953 /* Print the fastmap in human-readable form. */
954
955 void
956 print_fastmap (fastmap)
957 char *fastmap;
958 {
959 unsigned was_a_range = 0;
960 unsigned i = 0;
961
962 while (i < (1 << BYTEWIDTH))
963 {
964 if (fastmap[i++])
965 {
966 was_a_range = 0;
967 putchar (i - 1);
968 while (i < (1 << BYTEWIDTH) && fastmap[i])
969 {
970 was_a_range = 1;
971 i++;
972 }
973 if (was_a_range)
974 {
975 printf ("-");
976 putchar (i - 1);
977 }
978 }
979 }
980 putchar ('\n');
981 }
982
983
984 /* Print a compiled pattern string in human-readable form, starting at
985 the START pointer into it and ending just before the pointer END. */
986
987 void
988 print_partial_compiled_pattern (start, end)
989 re_char *start;
990 re_char *end;
991 {
992 int mcnt, mcnt2;
993 re_char *p = start;
994 re_char *pend = end;
995
996 if (start == NULL)
997 {
998 fprintf (stderr, "(null)\n");
999 return;
1000 }
1001
1002 /* Loop over pattern commands. */
1003 while (p < pend)
1004 {
1005 fprintf (stderr, "%d:\t", p - start);
1006
1007 switch ((re_opcode_t) *p++)
1008 {
1009 case no_op:
1010 fprintf (stderr, "/no_op");
1011 break;
1012
1013 case succeed:
1014 fprintf (stderr, "/succeed");
1015 break;
1016
1017 case exactn:
1018 mcnt = *p++;
1019 fprintf (stderr, "/exactn/%d", mcnt);
1020 do
1021 {
1022 fprintf (stderr, "/%c", *p++);
1023 }
1024 while (--mcnt);
1025 break;
1026
1027 case start_memory:
1028 fprintf (stderr, "/start_memory/%d", *p++);
1029 break;
1030
1031 case stop_memory:
1032 fprintf (stderr, "/stop_memory/%d", *p++);
1033 break;
1034
1035 case duplicate:
1036 fprintf (stderr, "/duplicate/%d", *p++);
1037 break;
1038
1039 case anychar:
1040 fprintf (stderr, "/anychar");
1041 break;
1042
1043 case charset:
1044 case charset_not:
1045 {
1046 register int c, last = -100;
1047 register int in_range = 0;
1048 int length = CHARSET_BITMAP_SIZE (p - 1);
1049 int has_range_table = CHARSET_RANGE_TABLE_EXISTS_P (p - 1);
1050
1051 fprintf (stderr, "/charset [%s",
1052 (re_opcode_t) *(p - 1) == charset_not ? "^" : "");
1053
1054 if (p + *p >= pend)
1055 fprintf (stderr, " !extends past end of pattern! ");
1056
1057 for (c = 0; c < 256; c++)
1058 if (c / 8 < length
1059 && (p[1 + (c/8)] & (1 << (c % 8))))
1060 {
1061 /* Are we starting a range? */
1062 if (last + 1 == c && ! in_range)
1063 {
1064 fprintf (stderr, "-");
1065 in_range = 1;
1066 }
1067 /* Have we broken a range? */
1068 else if (last + 1 != c && in_range)
1069 {
1070 fprintf (stderr, "%c", last);
1071 in_range = 0;
1072 }
1073
1074 if (! in_range)
1075 fprintf (stderr, "%c", c);
1076
1077 last = c;
1078 }
1079
1080 if (in_range)
1081 fprintf (stderr, "%c", last);
1082
1083 fprintf (stderr, "]");
1084
1085 p += 1 + length;
1086
1087 if (has_range_table)
1088 {
1089 int count;
1090 fprintf (stderr, "has-range-table");
1091
1092 /* ??? Should print the range table; for now, just skip it. */
1093 p += 2; /* skip range table bits */
1094 EXTRACT_NUMBER_AND_INCR (count, p);
1095 p = CHARSET_RANGE_TABLE_END (p, count);
1096 }
1097 }
1098 break;
1099
1100 case begline:
1101 fprintf (stderr, "/begline");
1102 break;
1103
1104 case endline:
1105 fprintf (stderr, "/endline");
1106 break;
1107
1108 case on_failure_jump:
1109 extract_number_and_incr (&mcnt, &p);
1110 fprintf (stderr, "/on_failure_jump to %d", p + mcnt - start);
1111 break;
1112
1113 case on_failure_keep_string_jump:
1114 extract_number_and_incr (&mcnt, &p);
1115 fprintf (stderr, "/on_failure_keep_string_jump to %d", p + mcnt - start);
1116 break;
1117
1118 case on_failure_jump_nastyloop:
1119 extract_number_and_incr (&mcnt, &p);
1120 fprintf (stderr, "/on_failure_jump_nastyloop to %d", p + mcnt - start);
1121 break;
1122
1123 case on_failure_jump_loop:
1124 extract_number_and_incr (&mcnt, &p);
1125 fprintf (stderr, "/on_failure_jump_loop to %d", p + mcnt - start);
1126 break;
1127
1128 case on_failure_jump_smart:
1129 extract_number_and_incr (&mcnt, &p);
1130 fprintf (stderr, "/on_failure_jump_smart to %d", p + mcnt - start);
1131 break;
1132
1133 case jump:
1134 extract_number_and_incr (&mcnt, &p);
1135 fprintf (stderr, "/jump to %d", p + mcnt - start);
1136 break;
1137
1138 case succeed_n:
1139 extract_number_and_incr (&mcnt, &p);
1140 extract_number_and_incr (&mcnt2, &p);
1141 fprintf (stderr, "/succeed_n to %d, %d times", p - 2 + mcnt - start, mcnt2);
1142 break;
1143
1144 case jump_n:
1145 extract_number_and_incr (&mcnt, &p);
1146 extract_number_and_incr (&mcnt2, &p);
1147 fprintf (stderr, "/jump_n to %d, %d times", p - 2 + mcnt - start, mcnt2);
1148 break;
1149
1150 case set_number_at:
1151 extract_number_and_incr (&mcnt, &p);
1152 extract_number_and_incr (&mcnt2, &p);
1153 fprintf (stderr, "/set_number_at location %d to %d", p - 2 + mcnt - start, mcnt2);
1154 break;
1155
1156 case wordbound:
1157 fprintf (stderr, "/wordbound");
1158 break;
1159
1160 case notwordbound:
1161 fprintf (stderr, "/notwordbound");
1162 break;
1163
1164 case wordbeg:
1165 fprintf (stderr, "/wordbeg");
1166 break;
1167
1168 case wordend:
1169 fprintf (stderr, "/wordend");
1170 break;
1171
1172 case symbeg:
1173 fprintf (stderr, "/symbeg");
1174 break;
1175
1176 case symend:
1177 fprintf (stderr, "/symend");
1178 break;
1179
1180 case syntaxspec:
1181 fprintf (stderr, "/syntaxspec");
1182 mcnt = *p++;
1183 fprintf (stderr, "/%d", mcnt);
1184 break;
1185
1186 case notsyntaxspec:
1187 fprintf (stderr, "/notsyntaxspec");
1188 mcnt = *p++;
1189 fprintf (stderr, "/%d", mcnt);
1190 break;
1191
1192 # ifdef emacs
1193 case before_dot:
1194 fprintf (stderr, "/before_dot");
1195 break;
1196
1197 case at_dot:
1198 fprintf (stderr, "/at_dot");
1199 break;
1200
1201 case after_dot:
1202 fprintf (stderr, "/after_dot");
1203 break;
1204
1205 case categoryspec:
1206 fprintf (stderr, "/categoryspec");
1207 mcnt = *p++;
1208 fprintf (stderr, "/%d", mcnt);
1209 break;
1210
1211 case notcategoryspec:
1212 fprintf (stderr, "/notcategoryspec");
1213 mcnt = *p++;
1214 fprintf (stderr, "/%d", mcnt);
1215 break;
1216 # endif /* emacs */
1217
1218 case begbuf:
1219 fprintf (stderr, "/begbuf");
1220 break;
1221
1222 case endbuf:
1223 fprintf (stderr, "/endbuf");
1224 break;
1225
1226 default:
1227 fprintf (stderr, "?%d", *(p-1));
1228 }
1229
1230 fprintf (stderr, "\n");
1231 }
1232
1233 fprintf (stderr, "%d:\tend of pattern.\n", p - start);
1234 }
1235
1236
1237 void
1238 print_compiled_pattern (bufp)
1239 struct re_pattern_buffer *bufp;
1240 {
1241 re_char *buffer = bufp->buffer;
1242
1243 print_partial_compiled_pattern (buffer, buffer + bufp->used);
1244 printf ("%ld bytes used/%ld bytes allocated.\n",
1245 bufp->used, bufp->allocated);
1246
1247 if (bufp->fastmap_accurate && bufp->fastmap)
1248 {
1249 printf ("fastmap: ");
1250 print_fastmap (bufp->fastmap);
1251 }
1252
1253 printf ("re_nsub: %d\t", bufp->re_nsub);
1254 printf ("regs_alloc: %d\t", bufp->regs_allocated);
1255 printf ("can_be_null: %d\t", bufp->can_be_null);
1256 printf ("no_sub: %d\t", bufp->no_sub);
1257 printf ("not_bol: %d\t", bufp->not_bol);
1258 printf ("not_eol: %d\t", bufp->not_eol);
1259 printf ("syntax: %lx\n", bufp->syntax);
1260 fflush (stdout);
1261 /* Perhaps we should print the translate table? */
1262 }
1263
1264
1265 void
1266 print_double_string (where, string1, size1, string2, size2)
1267 re_char *where;
1268 re_char *string1;
1269 re_char *string2;
1270 int size1;
1271 int size2;
1272 {
1273 int this_char;
1274
1275 if (where == NULL)
1276 printf ("(null)");
1277 else
1278 {
1279 if (FIRST_STRING_P (where))
1280 {
1281 for (this_char = where - string1; this_char < size1; this_char++)
1282 putchar (string1[this_char]);
1283
1284 where = string2;
1285 }
1286
1287 for (this_char = where - string2; this_char < size2; this_char++)
1288 putchar (string2[this_char]);
1289 }
1290 }
1291
1292 #else /* not DEBUG */
1293
1294 # undef assert
1295 # define assert(e)
1296
1297 # define DEBUG_STATEMENT(e)
1298 # define DEBUG_PRINT1(x)
1299 # define DEBUG_PRINT2(x1, x2)
1300 # define DEBUG_PRINT3(x1, x2, x3)
1301 # define DEBUG_PRINT4(x1, x2, x3, x4)
1302 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e)
1303 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2)
1304
1305 #endif /* not DEBUG */
1306 \f
1307 /* Set by `re_set_syntax' to the current regexp syntax to recognize. Can
1308 also be assigned to arbitrarily: each pattern buffer stores its own
1309 syntax, so it can be changed between regex compilations. */
1310 /* This has no initializer because initialized variables in Emacs
1311 become read-only after dumping. */
1312 reg_syntax_t re_syntax_options;
1313
1314
1315 /* Specify the precise syntax of regexps for compilation. This provides
1316 for compatibility for various utilities which historically have
1317 different, incompatible syntaxes.
1318
1319 The argument SYNTAX is a bit mask comprised of the various bits
1320 defined in regex.h. We return the old syntax. */
1321
1322 reg_syntax_t
1323 re_set_syntax (syntax)
1324 reg_syntax_t syntax;
1325 {
1326 reg_syntax_t ret = re_syntax_options;
1327
1328 re_syntax_options = syntax;
1329 return ret;
1330 }
1331 WEAK_ALIAS (__re_set_syntax, re_set_syntax)
1332
1333 /* Regexp to use to replace spaces, or NULL meaning don't. */
1334 static re_char *whitespace_regexp;
1335
1336 void
1337 re_set_whitespace_regexp (regexp)
1338 const char *regexp;
1339 {
1340 whitespace_regexp = (re_char *) regexp;
1341 }
1342 WEAK_ALIAS (__re_set_syntax, re_set_syntax)
1343 \f
1344 /* This table gives an error message for each of the error codes listed
1345 in regex.h. Obviously the order here has to be same as there.
1346 POSIX doesn't require that we do anything for REG_NOERROR,
1347 but why not be nice? */
1348
1349 static const char *re_error_msgid[] =
1350 {
1351 gettext_noop ("Success"), /* REG_NOERROR */
1352 gettext_noop ("No match"), /* REG_NOMATCH */
1353 gettext_noop ("Invalid regular expression"), /* REG_BADPAT */
1354 gettext_noop ("Invalid collation character"), /* REG_ECOLLATE */
1355 gettext_noop ("Invalid character class name"), /* REG_ECTYPE */
1356 gettext_noop ("Trailing backslash"), /* REG_EESCAPE */
1357 gettext_noop ("Invalid back reference"), /* REG_ESUBREG */
1358 gettext_noop ("Unmatched [ or [^"), /* REG_EBRACK */
1359 gettext_noop ("Unmatched ( or \\("), /* REG_EPAREN */
1360 gettext_noop ("Unmatched \\{"), /* REG_EBRACE */
1361 gettext_noop ("Invalid content of \\{\\}"), /* REG_BADBR */
1362 gettext_noop ("Invalid range end"), /* REG_ERANGE */
1363 gettext_noop ("Memory exhausted"), /* REG_ESPACE */
1364 gettext_noop ("Invalid preceding regular expression"), /* REG_BADRPT */
1365 gettext_noop ("Premature end of regular expression"), /* REG_EEND */
1366 gettext_noop ("Regular expression too big"), /* REG_ESIZE */
1367 gettext_noop ("Unmatched ) or \\)"), /* REG_ERPAREN */
1368 gettext_noop ("Range striding over charsets") /* REG_ERANGEX */
1369 };
1370 \f
1371 /* Avoiding alloca during matching, to placate r_alloc. */
1372
1373 /* Define MATCH_MAY_ALLOCATE unless we need to make sure that the
1374 searching and matching functions should not call alloca. On some
1375 systems, alloca is implemented in terms of malloc, and if we're
1376 using the relocating allocator routines, then malloc could cause a
1377 relocation, which might (if the strings being searched are in the
1378 ralloc heap) shift the data out from underneath the regexp
1379 routines.
1380
1381 Here's another reason to avoid allocation: Emacs
1382 processes input from X in a signal handler; processing X input may
1383 call malloc; if input arrives while a matching routine is calling
1384 malloc, then we're scrod. But Emacs can't just block input while
1385 calling matching routines; then we don't notice interrupts when
1386 they come in. So, Emacs blocks input around all regexp calls
1387 except the matching calls, which it leaves unprotected, in the
1388 faith that they will not malloc. */
1389
1390 /* Normally, this is fine. */
1391 #define MATCH_MAY_ALLOCATE
1392
1393 /* When using GNU C, we are not REALLY using the C alloca, no matter
1394 what config.h may say. So don't take precautions for it. */
1395 #ifdef __GNUC__
1396 # undef C_ALLOCA
1397 #endif
1398
1399 /* The match routines may not allocate if (1) they would do it with malloc
1400 and (2) it's not safe for them to use malloc.
1401 Note that if REL_ALLOC is defined, matching would not use malloc for the
1402 failure stack, but we would still use it for the register vectors;
1403 so REL_ALLOC should not affect this. */
1404 #if (defined C_ALLOCA || defined REGEX_MALLOC) && defined emacs
1405 # undef MATCH_MAY_ALLOCATE
1406 #endif
1407
1408 \f
1409 /* Failure stack declarations and macros; both re_compile_fastmap and
1410 re_match_2 use a failure stack. These have to be macros because of
1411 REGEX_ALLOCATE_STACK. */
1412
1413
1414 /* Approximate number of failure points for which to initially allocate space
1415 when matching. If this number is exceeded, we allocate more
1416 space, so it is not a hard limit. */
1417 #ifndef INIT_FAILURE_ALLOC
1418 # define INIT_FAILURE_ALLOC 20
1419 #endif
1420
1421 /* Roughly the maximum number of failure points on the stack. Would be
1422 exactly that if always used TYPICAL_FAILURE_SIZE items each time we failed.
1423 This is a variable only so users of regex can assign to it; we never
1424 change it ourselves. We always multiply it by TYPICAL_FAILURE_SIZE
1425 before using it, so it should probably be a byte-count instead. */
1426 # if defined MATCH_MAY_ALLOCATE
1427 /* Note that 4400 was enough to cause a crash on Alpha OSF/1,
1428 whose default stack limit is 2mb. In order for a larger
1429 value to work reliably, you have to try to make it accord
1430 with the process stack limit. */
1431 size_t re_max_failures = 40000;
1432 # else
1433 size_t re_max_failures = 4000;
1434 # endif
1435
1436 union fail_stack_elt
1437 {
1438 re_char *pointer;
1439 /* This should be the biggest `int' that's no bigger than a pointer. */
1440 long integer;
1441 };
1442
1443 typedef union fail_stack_elt fail_stack_elt_t;
1444
1445 typedef struct
1446 {
1447 fail_stack_elt_t *stack;
1448 size_t size;
1449 size_t avail; /* Offset of next open position. */
1450 size_t frame; /* Offset of the cur constructed frame. */
1451 } fail_stack_type;
1452
1453 #define FAIL_STACK_EMPTY() (fail_stack.frame == 0)
1454 #define FAIL_STACK_FULL() (fail_stack.avail == fail_stack.size)
1455
1456
1457 /* Define macros to initialize and free the failure stack.
1458 Do `return -2' if the alloc fails. */
1459
1460 #ifdef MATCH_MAY_ALLOCATE
1461 # define INIT_FAIL_STACK() \
1462 do { \
1463 fail_stack.stack = (fail_stack_elt_t *) \
1464 REGEX_ALLOCATE_STACK (INIT_FAILURE_ALLOC * TYPICAL_FAILURE_SIZE \
1465 * sizeof (fail_stack_elt_t)); \
1466 \
1467 if (fail_stack.stack == NULL) \
1468 return -2; \
1469 \
1470 fail_stack.size = INIT_FAILURE_ALLOC; \
1471 fail_stack.avail = 0; \
1472 fail_stack.frame = 0; \
1473 } while (0)
1474
1475 # define RESET_FAIL_STACK() REGEX_FREE_STACK (fail_stack.stack)
1476 #else
1477 # define INIT_FAIL_STACK() \
1478 do { \
1479 fail_stack.avail = 0; \
1480 fail_stack.frame = 0; \
1481 } while (0)
1482
1483 # define RESET_FAIL_STACK() ((void)0)
1484 #endif
1485
1486
1487 /* Double the size of FAIL_STACK, up to a limit
1488 which allows approximately `re_max_failures' items.
1489
1490 Return 1 if succeeds, and 0 if either ran out of memory
1491 allocating space for it or it was already too large.
1492
1493 REGEX_REALLOCATE_STACK requires `destination' be declared. */
1494
1495 /* Factor to increase the failure stack size by
1496 when we increase it.
1497 This used to be 2, but 2 was too wasteful
1498 because the old discarded stacks added up to as much space
1499 were as ultimate, maximum-size stack. */
1500 #define FAIL_STACK_GROWTH_FACTOR 4
1501
1502 #define GROW_FAIL_STACK(fail_stack) \
1503 (((fail_stack).size * sizeof (fail_stack_elt_t) \
1504 >= re_max_failures * TYPICAL_FAILURE_SIZE) \
1505 ? 0 \
1506 : ((fail_stack).stack \
1507 = (fail_stack_elt_t *) \
1508 REGEX_REALLOCATE_STACK ((fail_stack).stack, \
1509 (fail_stack).size * sizeof (fail_stack_elt_t), \
1510 MIN (re_max_failures * TYPICAL_FAILURE_SIZE, \
1511 ((fail_stack).size * sizeof (fail_stack_elt_t) \
1512 * FAIL_STACK_GROWTH_FACTOR))), \
1513 \
1514 (fail_stack).stack == NULL \
1515 ? 0 \
1516 : ((fail_stack).size \
1517 = (MIN (re_max_failures * TYPICAL_FAILURE_SIZE, \
1518 ((fail_stack).size * sizeof (fail_stack_elt_t) \
1519 * FAIL_STACK_GROWTH_FACTOR)) \
1520 / sizeof (fail_stack_elt_t)), \
1521 1)))
1522
1523
1524 /* Push a pointer value onto the failure stack.
1525 Assumes the variable `fail_stack'. Probably should only
1526 be called from within `PUSH_FAILURE_POINT'. */
1527 #define PUSH_FAILURE_POINTER(item) \
1528 fail_stack.stack[fail_stack.avail++].pointer = (item)
1529
1530 /* This pushes an integer-valued item onto the failure stack.
1531 Assumes the variable `fail_stack'. Probably should only
1532 be called from within `PUSH_FAILURE_POINT'. */
1533 #define PUSH_FAILURE_INT(item) \
1534 fail_stack.stack[fail_stack.avail++].integer = (item)
1535
1536 /* Push a fail_stack_elt_t value onto the failure stack.
1537 Assumes the variable `fail_stack'. Probably should only
1538 be called from within `PUSH_FAILURE_POINT'. */
1539 #define PUSH_FAILURE_ELT(item) \
1540 fail_stack.stack[fail_stack.avail++] = (item)
1541
1542 /* These three POP... operations complement the three PUSH... operations.
1543 All assume that `fail_stack' is nonempty. */
1544 #define POP_FAILURE_POINTER() fail_stack.stack[--fail_stack.avail].pointer
1545 #define POP_FAILURE_INT() fail_stack.stack[--fail_stack.avail].integer
1546 #define POP_FAILURE_ELT() fail_stack.stack[--fail_stack.avail]
1547
1548 /* Individual items aside from the registers. */
1549 #define NUM_NONREG_ITEMS 3
1550
1551 /* Used to examine the stack (to detect infinite loops). */
1552 #define FAILURE_PAT(h) fail_stack.stack[(h) - 1].pointer
1553 #define FAILURE_STR(h) (fail_stack.stack[(h) - 2].pointer)
1554 #define NEXT_FAILURE_HANDLE(h) fail_stack.stack[(h) - 3].integer
1555 #define TOP_FAILURE_HANDLE() fail_stack.frame
1556
1557
1558 #define ENSURE_FAIL_STACK(space) \
1559 while (REMAINING_AVAIL_SLOTS <= space) { \
1560 if (!GROW_FAIL_STACK (fail_stack)) \
1561 return -2; \
1562 DEBUG_PRINT2 ("\n Doubled stack; size now: %d\n", (fail_stack).size);\
1563 DEBUG_PRINT2 (" slots available: %d\n", REMAINING_AVAIL_SLOTS);\
1564 }
1565
1566 /* Push register NUM onto the stack. */
1567 #define PUSH_FAILURE_REG(num) \
1568 do { \
1569 char *destination; \
1570 ENSURE_FAIL_STACK(3); \
1571 DEBUG_PRINT4 (" Push reg %d (spanning %p -> %p)\n", \
1572 num, regstart[num], regend[num]); \
1573 PUSH_FAILURE_POINTER (regstart[num]); \
1574 PUSH_FAILURE_POINTER (regend[num]); \
1575 PUSH_FAILURE_INT (num); \
1576 } while (0)
1577
1578 /* Change the counter's value to VAL, but make sure that it will
1579 be reset when backtracking. */
1580 #define PUSH_NUMBER(ptr,val) \
1581 do { \
1582 char *destination; \
1583 int c; \
1584 ENSURE_FAIL_STACK(3); \
1585 EXTRACT_NUMBER (c, ptr); \
1586 DEBUG_PRINT4 (" Push number %p = %d -> %d\n", ptr, c, val); \
1587 PUSH_FAILURE_INT (c); \
1588 PUSH_FAILURE_POINTER (ptr); \
1589 PUSH_FAILURE_INT (-1); \
1590 STORE_NUMBER (ptr, val); \
1591 } while (0)
1592
1593 /* Pop a saved register off the stack. */
1594 #define POP_FAILURE_REG_OR_COUNT() \
1595 do { \
1596 int reg = POP_FAILURE_INT (); \
1597 if (reg == -1) \
1598 { \
1599 /* It's a counter. */ \
1600 /* Here, we discard `const', making re_match non-reentrant. */ \
1601 unsigned char *ptr = (unsigned char*) POP_FAILURE_POINTER (); \
1602 reg = POP_FAILURE_INT (); \
1603 STORE_NUMBER (ptr, reg); \
1604 DEBUG_PRINT3 (" Pop counter %p = %d\n", ptr, reg); \
1605 } \
1606 else \
1607 { \
1608 regend[reg] = POP_FAILURE_POINTER (); \
1609 regstart[reg] = POP_FAILURE_POINTER (); \
1610 DEBUG_PRINT4 (" Pop reg %d (spanning %p -> %p)\n", \
1611 reg, regstart[reg], regend[reg]); \
1612 } \
1613 } while (0)
1614
1615 /* Check that we are not stuck in an infinite loop. */
1616 #define CHECK_INFINITE_LOOP(pat_cur, string_place) \
1617 do { \
1618 int failure = TOP_FAILURE_HANDLE (); \
1619 /* Check for infinite matching loops */ \
1620 while (failure > 0 \
1621 && (FAILURE_STR (failure) == string_place \
1622 || FAILURE_STR (failure) == NULL)) \
1623 { \
1624 assert (FAILURE_PAT (failure) >= bufp->buffer \
1625 && FAILURE_PAT (failure) <= bufp->buffer + bufp->used); \
1626 if (FAILURE_PAT (failure) == pat_cur) \
1627 { \
1628 cycle = 1; \
1629 break; \
1630 } \
1631 DEBUG_PRINT2 (" Other pattern: %p\n", FAILURE_PAT (failure)); \
1632 failure = NEXT_FAILURE_HANDLE(failure); \
1633 } \
1634 DEBUG_PRINT2 (" Other string: %p\n", FAILURE_STR (failure)); \
1635 } while (0)
1636
1637 /* Push the information about the state we will need
1638 if we ever fail back to it.
1639
1640 Requires variables fail_stack, regstart, regend and
1641 num_regs be declared. GROW_FAIL_STACK requires `destination' be
1642 declared.
1643
1644 Does `return FAILURE_CODE' if runs out of memory. */
1645
1646 #define PUSH_FAILURE_POINT(pattern, string_place) \
1647 do { \
1648 char *destination; \
1649 /* Must be int, so when we don't save any registers, the arithmetic \
1650 of 0 + -1 isn't done as unsigned. */ \
1651 \
1652 DEBUG_STATEMENT (nfailure_points_pushed++); \
1653 DEBUG_PRINT1 ("\nPUSH_FAILURE_POINT:\n"); \
1654 DEBUG_PRINT2 (" Before push, next avail: %d\n", (fail_stack).avail); \
1655 DEBUG_PRINT2 (" size: %d\n", (fail_stack).size);\
1656 \
1657 ENSURE_FAIL_STACK (NUM_NONREG_ITEMS); \
1658 \
1659 DEBUG_PRINT1 ("\n"); \
1660 \
1661 DEBUG_PRINT2 (" Push frame index: %d\n", fail_stack.frame); \
1662 PUSH_FAILURE_INT (fail_stack.frame); \
1663 \
1664 DEBUG_PRINT2 (" Push string %p: `", string_place); \
1665 DEBUG_PRINT_DOUBLE_STRING (string_place, string1, size1, string2, size2);\
1666 DEBUG_PRINT1 ("'\n"); \
1667 PUSH_FAILURE_POINTER (string_place); \
1668 \
1669 DEBUG_PRINT2 (" Push pattern %p: ", pattern); \
1670 DEBUG_PRINT_COMPILED_PATTERN (bufp, pattern, pend); \
1671 PUSH_FAILURE_POINTER (pattern); \
1672 \
1673 /* Close the frame by moving the frame pointer past it. */ \
1674 fail_stack.frame = fail_stack.avail; \
1675 } while (0)
1676
1677 /* Estimate the size of data pushed by a typical failure stack entry.
1678 An estimate is all we need, because all we use this for
1679 is to choose a limit for how big to make the failure stack. */
1680 /* BEWARE, the value `20' is hard-coded in emacs.c:main(). */
1681 #define TYPICAL_FAILURE_SIZE 20
1682
1683 /* How many items can still be added to the stack without overflowing it. */
1684 #define REMAINING_AVAIL_SLOTS ((fail_stack).size - (fail_stack).avail)
1685
1686
1687 /* Pops what PUSH_FAIL_STACK pushes.
1688
1689 We restore into the parameters, all of which should be lvalues:
1690 STR -- the saved data position.
1691 PAT -- the saved pattern position.
1692 REGSTART, REGEND -- arrays of string positions.
1693
1694 Also assumes the variables `fail_stack' and (if debugging), `bufp',
1695 `pend', `string1', `size1', `string2', and `size2'. */
1696
1697 #define POP_FAILURE_POINT(str, pat) \
1698 do { \
1699 assert (!FAIL_STACK_EMPTY ()); \
1700 \
1701 /* Remove failure points and point to how many regs pushed. */ \
1702 DEBUG_PRINT1 ("POP_FAILURE_POINT:\n"); \
1703 DEBUG_PRINT2 (" Before pop, next avail: %d\n", fail_stack.avail); \
1704 DEBUG_PRINT2 (" size: %d\n", fail_stack.size); \
1705 \
1706 /* Pop the saved registers. */ \
1707 while (fail_stack.frame < fail_stack.avail) \
1708 POP_FAILURE_REG_OR_COUNT (); \
1709 \
1710 pat = POP_FAILURE_POINTER (); \
1711 DEBUG_PRINT2 (" Popping pattern %p: ", pat); \
1712 DEBUG_PRINT_COMPILED_PATTERN (bufp, pat, pend); \
1713 \
1714 /* If the saved string location is NULL, it came from an \
1715 on_failure_keep_string_jump opcode, and we want to throw away the \
1716 saved NULL, thus retaining our current position in the string. */ \
1717 str = POP_FAILURE_POINTER (); \
1718 DEBUG_PRINT2 (" Popping string %p: `", str); \
1719 DEBUG_PRINT_DOUBLE_STRING (str, string1, size1, string2, size2); \
1720 DEBUG_PRINT1 ("'\n"); \
1721 \
1722 fail_stack.frame = POP_FAILURE_INT (); \
1723 DEBUG_PRINT2 (" Popping frame index: %d\n", fail_stack.frame); \
1724 \
1725 assert (fail_stack.avail >= 0); \
1726 assert (fail_stack.frame <= fail_stack.avail); \
1727 \
1728 DEBUG_STATEMENT (nfailure_points_popped++); \
1729 } while (0) /* POP_FAILURE_POINT */
1730
1731
1732 \f
1733 /* Registers are set to a sentinel when they haven't yet matched. */
1734 #define REG_UNSET(e) ((e) == NULL)
1735 \f
1736 /* Subroutine declarations and macros for regex_compile. */
1737
1738 static reg_errcode_t regex_compile _RE_ARGS ((re_char *pattern, size_t size,
1739 reg_syntax_t syntax,
1740 struct re_pattern_buffer *bufp));
1741 static void store_op1 _RE_ARGS ((re_opcode_t op, unsigned char *loc, int arg));
1742 static void store_op2 _RE_ARGS ((re_opcode_t op, unsigned char *loc,
1743 int arg1, int arg2));
1744 static void insert_op1 _RE_ARGS ((re_opcode_t op, unsigned char *loc,
1745 int arg, unsigned char *end));
1746 static void insert_op2 _RE_ARGS ((re_opcode_t op, unsigned char *loc,
1747 int arg1, int arg2, unsigned char *end));
1748 static boolean at_begline_loc_p _RE_ARGS ((re_char *pattern,
1749 re_char *p,
1750 reg_syntax_t syntax));
1751 static boolean at_endline_loc_p _RE_ARGS ((re_char *p,
1752 re_char *pend,
1753 reg_syntax_t syntax));
1754 static re_char *skip_one_char _RE_ARGS ((re_char *p));
1755 static int analyse_first _RE_ARGS ((re_char *p, re_char *pend,
1756 char *fastmap, const int multibyte));
1757
1758 /* Fetch the next character in the uncompiled pattern, with no
1759 translation. */
1760 #define PATFETCH(c) \
1761 do { \
1762 int len; \
1763 if (p == pend) return REG_EEND; \
1764 c = RE_STRING_CHAR_AND_LENGTH (p, pend - p, len); \
1765 p += len; \
1766 } while (0)
1767
1768
1769 /* If `translate' is non-null, return translate[D], else just D. We
1770 cast the subscript to translate because some data is declared as
1771 `char *', to avoid warnings when a string constant is passed. But
1772 when we use a character as a subscript we must make it unsigned. */
1773 #ifndef TRANSLATE
1774 # define TRANSLATE(d) \
1775 (RE_TRANSLATE_P (translate) ? RE_TRANSLATE (translate, (d)) : (d))
1776 #endif
1777
1778
1779 /* Macros for outputting the compiled pattern into `buffer'. */
1780
1781 /* If the buffer isn't allocated when it comes in, use this. */
1782 #define INIT_BUF_SIZE 32
1783
1784 /* Make sure we have at least N more bytes of space in buffer. */
1785 #define GET_BUFFER_SPACE(n) \
1786 while ((size_t) (b - bufp->buffer + (n)) > bufp->allocated) \
1787 EXTEND_BUFFER ()
1788
1789 /* Make sure we have one more byte of buffer space and then add C to it. */
1790 #define BUF_PUSH(c) \
1791 do { \
1792 GET_BUFFER_SPACE (1); \
1793 *b++ = (unsigned char) (c); \
1794 } while (0)
1795
1796
1797 /* Ensure we have two more bytes of buffer space and then append C1 and C2. */
1798 #define BUF_PUSH_2(c1, c2) \
1799 do { \
1800 GET_BUFFER_SPACE (2); \
1801 *b++ = (unsigned char) (c1); \
1802 *b++ = (unsigned char) (c2); \
1803 } while (0)
1804
1805
1806 /* As with BUF_PUSH_2, except for three bytes. */
1807 #define BUF_PUSH_3(c1, c2, c3) \
1808 do { \
1809 GET_BUFFER_SPACE (3); \
1810 *b++ = (unsigned char) (c1); \
1811 *b++ = (unsigned char) (c2); \
1812 *b++ = (unsigned char) (c3); \
1813 } while (0)
1814
1815
1816 /* Store a jump with opcode OP at LOC to location TO. We store a
1817 relative address offset by the three bytes the jump itself occupies. */
1818 #define STORE_JUMP(op, loc, to) \
1819 store_op1 (op, loc, (to) - (loc) - 3)
1820
1821 /* Likewise, for a two-argument jump. */
1822 #define STORE_JUMP2(op, loc, to, arg) \
1823 store_op2 (op, loc, (to) - (loc) - 3, arg)
1824
1825 /* Like `STORE_JUMP', but for inserting. Assume `b' is the buffer end. */
1826 #define INSERT_JUMP(op, loc, to) \
1827 insert_op1 (op, loc, (to) - (loc) - 3, b)
1828
1829 /* Like `STORE_JUMP2', but for inserting. Assume `b' is the buffer end. */
1830 #define INSERT_JUMP2(op, loc, to, arg) \
1831 insert_op2 (op, loc, (to) - (loc) - 3, arg, b)
1832
1833
1834 /* This is not an arbitrary limit: the arguments which represent offsets
1835 into the pattern are two bytes long. So if 2^15 bytes turns out to
1836 be too small, many things would have to change. */
1837 # define MAX_BUF_SIZE (1L << 15)
1838
1839 #if 0 /* This is when we thought it could be 2^16 bytes. */
1840 /* Any other compiler which, like MSC, has allocation limit below 2^16
1841 bytes will have to use approach similar to what was done below for
1842 MSC and drop MAX_BUF_SIZE a bit. Otherwise you may end up
1843 reallocating to 0 bytes. Such thing is not going to work too well.
1844 You have been warned!! */
1845 #if defined _MSC_VER && !defined WIN32
1846 /* Microsoft C 16-bit versions limit malloc to approx 65512 bytes. */
1847 # define MAX_BUF_SIZE 65500L
1848 #else
1849 # define MAX_BUF_SIZE (1L << 16)
1850 #endif
1851 #endif /* 0 */
1852
1853 /* Extend the buffer by twice its current size via realloc and
1854 reset the pointers that pointed into the old block to point to the
1855 correct places in the new one. If extending the buffer results in it
1856 being larger than MAX_BUF_SIZE, then flag memory exhausted. */
1857 #if __BOUNDED_POINTERS__
1858 # define SET_HIGH_BOUND(P) (__ptrhigh (P) = __ptrlow (P) + bufp->allocated)
1859 # define MOVE_BUFFER_POINTER(P) \
1860 (__ptrlow (P) += incr, SET_HIGH_BOUND (P), __ptrvalue (P) += incr)
1861 # define ELSE_EXTEND_BUFFER_HIGH_BOUND \
1862 else \
1863 { \
1864 SET_HIGH_BOUND (b); \
1865 SET_HIGH_BOUND (begalt); \
1866 if (fixup_alt_jump) \
1867 SET_HIGH_BOUND (fixup_alt_jump); \
1868 if (laststart) \
1869 SET_HIGH_BOUND (laststart); \
1870 if (pending_exact) \
1871 SET_HIGH_BOUND (pending_exact); \
1872 }
1873 #else
1874 # define MOVE_BUFFER_POINTER(P) (P) += incr
1875 # define ELSE_EXTEND_BUFFER_HIGH_BOUND
1876 #endif
1877 #define EXTEND_BUFFER() \
1878 do { \
1879 re_char *old_buffer = bufp->buffer; \
1880 if (bufp->allocated == MAX_BUF_SIZE) \
1881 return REG_ESIZE; \
1882 bufp->allocated <<= 1; \
1883 if (bufp->allocated > MAX_BUF_SIZE) \
1884 bufp->allocated = MAX_BUF_SIZE; \
1885 RETALLOC (bufp->buffer, bufp->allocated, unsigned char); \
1886 if (bufp->buffer == NULL) \
1887 return REG_ESPACE; \
1888 /* If the buffer moved, move all the pointers into it. */ \
1889 if (old_buffer != bufp->buffer) \
1890 { \
1891 int incr = bufp->buffer - old_buffer; \
1892 MOVE_BUFFER_POINTER (b); \
1893 MOVE_BUFFER_POINTER (begalt); \
1894 if (fixup_alt_jump) \
1895 MOVE_BUFFER_POINTER (fixup_alt_jump); \
1896 if (laststart) \
1897 MOVE_BUFFER_POINTER (laststart); \
1898 if (pending_exact) \
1899 MOVE_BUFFER_POINTER (pending_exact); \
1900 } \
1901 ELSE_EXTEND_BUFFER_HIGH_BOUND \
1902 } while (0)
1903
1904
1905 /* Since we have one byte reserved for the register number argument to
1906 {start,stop}_memory, the maximum number of groups we can report
1907 things about is what fits in that byte. */
1908 #define MAX_REGNUM 255
1909
1910 /* But patterns can have more than `MAX_REGNUM' registers. We just
1911 ignore the excess. */
1912 typedef int regnum_t;
1913
1914
1915 /* Macros for the compile stack. */
1916
1917 /* Since offsets can go either forwards or backwards, this type needs to
1918 be able to hold values from -(MAX_BUF_SIZE - 1) to MAX_BUF_SIZE - 1. */
1919 /* int may be not enough when sizeof(int) == 2. */
1920 typedef long pattern_offset_t;
1921
1922 typedef struct
1923 {
1924 pattern_offset_t begalt_offset;
1925 pattern_offset_t fixup_alt_jump;
1926 pattern_offset_t laststart_offset;
1927 regnum_t regnum;
1928 } compile_stack_elt_t;
1929
1930
1931 typedef struct
1932 {
1933 compile_stack_elt_t *stack;
1934 unsigned size;
1935 unsigned avail; /* Offset of next open position. */
1936 } compile_stack_type;
1937
1938
1939 #define INIT_COMPILE_STACK_SIZE 32
1940
1941 #define COMPILE_STACK_EMPTY (compile_stack.avail == 0)
1942 #define COMPILE_STACK_FULL (compile_stack.avail == compile_stack.size)
1943
1944 /* The next available element. */
1945 #define COMPILE_STACK_TOP (compile_stack.stack[compile_stack.avail])
1946
1947 /* Explicit quit checking is only used on NTemacs and whenever we
1948 use polling to process input events. */
1949 #if defined emacs && (defined WINDOWSNT || defined SYNC_INPUT) && defined QUIT
1950 extern int immediate_quit;
1951 # define IMMEDIATE_QUIT_CHECK \
1952 do { \
1953 if (immediate_quit) QUIT; \
1954 } while (0)
1955 #else
1956 # define IMMEDIATE_QUIT_CHECK ((void)0)
1957 #endif
1958 \f
1959 /* Structure to manage work area for range table. */
1960 struct range_table_work_area
1961 {
1962 int *table; /* actual work area. */
1963 int allocated; /* allocated size for work area in bytes. */
1964 int used; /* actually used size in words. */
1965 int bits; /* flag to record character classes */
1966 };
1967
1968 /* Make sure that WORK_AREA can hold more N multibyte characters.
1969 This is used only in set_image_of_range and set_image_of_range_1.
1970 It expects WORK_AREA to be a pointer.
1971 If it can't get the space, it returns from the surrounding function. */
1972
1973 #define EXTEND_RANGE_TABLE(work_area, n) \
1974 do { \
1975 if (((work_area).used + (n)) * sizeof (int) > (work_area).allocated) \
1976 { \
1977 extend_range_table_work_area (&work_area); \
1978 if ((work_area).table == 0) \
1979 return (REG_ESPACE); \
1980 } \
1981 } while (0)
1982
1983 #define SET_RANGE_TABLE_WORK_AREA_BIT(work_area, bit) \
1984 (work_area).bits |= (bit)
1985
1986 /* Bits used to implement the multibyte-part of the various character classes
1987 such as [:alnum:] in a charset's range table. */
1988 #define BIT_WORD 0x1
1989 #define BIT_LOWER 0x2
1990 #define BIT_PUNCT 0x4
1991 #define BIT_SPACE 0x8
1992 #define BIT_UPPER 0x10
1993 #define BIT_MULTIBYTE 0x20
1994
1995 /* Set a range (RANGE_START, RANGE_END) to WORK_AREA. */
1996 #define SET_RANGE_TABLE_WORK_AREA(work_area, range_start, range_end) \
1997 do { \
1998 EXTEND_RANGE_TABLE ((work_area), 2); \
1999 (work_area).table[(work_area).used++] = (range_start); \
2000 (work_area).table[(work_area).used++] = (range_end); \
2001 } while (0)
2002
2003 /* Free allocated memory for WORK_AREA. */
2004 #define FREE_RANGE_TABLE_WORK_AREA(work_area) \
2005 do { \
2006 if ((work_area).table) \
2007 free ((work_area).table); \
2008 } while (0)
2009
2010 #define CLEAR_RANGE_TABLE_WORK_USED(work_area) ((work_area).used = 0, (work_area).bits = 0)
2011 #define RANGE_TABLE_WORK_USED(work_area) ((work_area).used)
2012 #define RANGE_TABLE_WORK_BITS(work_area) ((work_area).bits)
2013 #define RANGE_TABLE_WORK_ELT(work_area, i) ((work_area).table[i])
2014 \f
2015
2016 /* Set the bit for character C in a list. */
2017 #define SET_LIST_BIT(c) (b[((c)) / BYTEWIDTH] |= 1 << ((c) % BYTEWIDTH))
2018
2019
2020 #ifdef emacs
2021
2022 /* Store characters in the rage range C0 to C1 in WORK_AREA while
2023 translating them and paying attention to the continuity of
2024 translated characters.
2025
2026 Implementation note: It is better to implement this fairly big
2027 macro by a function, but it's not that easy because macros called
2028 in this macro assume various local variables already declared. */
2029
2030 #define SETUP_MULTIBYTE_RANGE(work_area, c0, c1) \
2031 do { \
2032 re_wchar_t c, t, t_last; \
2033 int n; \
2034 \
2035 c = (c0); \
2036 t_last = multibyte ? TRANSLATE (c) : TRANSLATE (MAKE_CHAR_MULTIBYTE (c)); \
2037 for (c++, n = 1; c <= (c1); c++, n++) \
2038 { \
2039 t = multibyte ? TRANSLATE (c) : TRANSLATE (MAKE_CHAR_MULTIBYTE (c)); \
2040 if (t_last + n == t) \
2041 continue; \
2042 SET_RANGE_TABLE_WORK_AREA ((work_area), t_last, t_last + n - 1); \
2043 t_last = t; \
2044 n = 0; \
2045 } \
2046 if (n > 0) \
2047 SET_RANGE_TABLE_WORK_AREA ((work_area), t_last, t_last + n - 1); \
2048 } while (0)
2049
2050 #endif /* emacs */
2051
2052 /* Get the next unsigned number in the uncompiled pattern. */
2053 #define GET_UNSIGNED_NUMBER(num) \
2054 do { \
2055 if (p == pend) \
2056 FREE_STACK_RETURN (REG_EBRACE); \
2057 else \
2058 { \
2059 PATFETCH (c); \
2060 while ('0' <= c && c <= '9') \
2061 { \
2062 int prev; \
2063 if (num < 0) \
2064 num = 0; \
2065 prev = num; \
2066 num = num * 10 + c - '0'; \
2067 if (num / 10 != prev) \
2068 FREE_STACK_RETURN (REG_BADBR); \
2069 if (p == pend) \
2070 FREE_STACK_RETURN (REG_EBRACE); \
2071 PATFETCH (c); \
2072 } \
2073 } \
2074 } while (0)
2075 \f
2076 #if ! WIDE_CHAR_SUPPORT
2077
2078 /* Map a string to the char class it names (if any). */
2079 re_wctype_t
2080 re_wctype (str)
2081 re_char *str;
2082 {
2083 const char *string = str;
2084 if (STREQ (string, "alnum")) return RECC_ALNUM;
2085 else if (STREQ (string, "alpha")) return RECC_ALPHA;
2086 else if (STREQ (string, "word")) return RECC_WORD;
2087 else if (STREQ (string, "ascii")) return RECC_ASCII;
2088 else if (STREQ (string, "nonascii")) return RECC_NONASCII;
2089 else if (STREQ (string, "graph")) return RECC_GRAPH;
2090 else if (STREQ (string, "lower")) return RECC_LOWER;
2091 else if (STREQ (string, "print")) return RECC_PRINT;
2092 else if (STREQ (string, "punct")) return RECC_PUNCT;
2093 else if (STREQ (string, "space")) return RECC_SPACE;
2094 else if (STREQ (string, "upper")) return RECC_UPPER;
2095 else if (STREQ (string, "unibyte")) return RECC_UNIBYTE;
2096 else if (STREQ (string, "multibyte")) return RECC_MULTIBYTE;
2097 else if (STREQ (string, "digit")) return RECC_DIGIT;
2098 else if (STREQ (string, "xdigit")) return RECC_XDIGIT;
2099 else if (STREQ (string, "cntrl")) return RECC_CNTRL;
2100 else if (STREQ (string, "blank")) return RECC_BLANK;
2101 else return 0;
2102 }
2103
2104 /* True iff CH is in the char class CC. */
2105 boolean
2106 re_iswctype (ch, cc)
2107 int ch;
2108 re_wctype_t cc;
2109 {
2110 switch (cc)
2111 {
2112 case RECC_ALNUM: return ISALNUM (ch);
2113 case RECC_ALPHA: return ISALPHA (ch);
2114 case RECC_BLANK: return ISBLANK (ch);
2115 case RECC_CNTRL: return ISCNTRL (ch);
2116 case RECC_DIGIT: return ISDIGIT (ch);
2117 case RECC_GRAPH: return ISGRAPH (ch);
2118 case RECC_LOWER: return ISLOWER (ch);
2119 case RECC_PRINT: return ISPRINT (ch);
2120 case RECC_PUNCT: return ISPUNCT (ch);
2121 case RECC_SPACE: return ISSPACE (ch);
2122 case RECC_UPPER: return ISUPPER (ch);
2123 case RECC_XDIGIT: return ISXDIGIT (ch);
2124 case RECC_ASCII: return IS_REAL_ASCII (ch);
2125 case RECC_NONASCII: return !IS_REAL_ASCII (ch);
2126 case RECC_UNIBYTE: return ISUNIBYTE (ch);
2127 case RECC_MULTIBYTE: return !ISUNIBYTE (ch);
2128 case RECC_WORD: return ISWORD (ch);
2129 case RECC_ERROR: return false;
2130 default:
2131 abort();
2132 }
2133 }
2134
2135 /* Return a bit-pattern to use in the range-table bits to match multibyte
2136 chars of class CC. */
2137 static int
2138 re_wctype_to_bit (cc)
2139 re_wctype_t cc;
2140 {
2141 switch (cc)
2142 {
2143 case RECC_NONASCII: case RECC_PRINT: case RECC_GRAPH:
2144 case RECC_MULTIBYTE: return BIT_MULTIBYTE;
2145 case RECC_ALPHA: case RECC_ALNUM: case RECC_WORD: return BIT_WORD;
2146 case RECC_LOWER: return BIT_LOWER;
2147 case RECC_UPPER: return BIT_UPPER;
2148 case RECC_PUNCT: return BIT_PUNCT;
2149 case RECC_SPACE: return BIT_SPACE;
2150 case RECC_ASCII: case RECC_DIGIT: case RECC_XDIGIT: case RECC_CNTRL:
2151 case RECC_BLANK: case RECC_UNIBYTE: case RECC_ERROR: return 0;
2152 default:
2153 abort();
2154 }
2155 }
2156 #endif
2157 \f
2158 /* Filling in the work area of a range. */
2159
2160 /* Actually extend the space in WORK_AREA. */
2161
2162 static void
2163 extend_range_table_work_area (work_area)
2164 struct range_table_work_area *work_area;
2165 {
2166 work_area->allocated += 16 * sizeof (int);
2167 if (work_area->table)
2168 work_area->table
2169 = (int *) realloc (work_area->table, work_area->allocated);
2170 else
2171 work_area->table
2172 = (int *) malloc (work_area->allocated);
2173 }
2174
2175 #if 0
2176 #ifdef emacs
2177
2178 /* Carefully find the ranges of codes that are equivalent
2179 under case conversion to the range start..end when passed through
2180 TRANSLATE. Handle the case where non-letters can come in between
2181 two upper-case letters (which happens in Latin-1).
2182 Also handle the case of groups of more than 2 case-equivalent chars.
2183
2184 The basic method is to look at consecutive characters and see
2185 if they can form a run that can be handled as one.
2186
2187 Returns -1 if successful, REG_ESPACE if ran out of space. */
2188
2189 static int
2190 set_image_of_range_1 (work_area, start, end, translate)
2191 RE_TRANSLATE_TYPE translate;
2192 struct range_table_work_area *work_area;
2193 re_wchar_t start, end;
2194 {
2195 /* `one_case' indicates a character, or a run of characters,
2196 each of which is an isolate (no case-equivalents).
2197 This includes all ASCII non-letters.
2198
2199 `two_case' indicates a character, or a run of characters,
2200 each of which has two case-equivalent forms.
2201 This includes all ASCII letters.
2202
2203 `strange' indicates a character that has more than one
2204 case-equivalent. */
2205
2206 enum case_type {one_case, two_case, strange};
2207
2208 /* Describe the run that is in progress,
2209 which the next character can try to extend.
2210 If run_type is strange, that means there really is no run.
2211 If run_type is one_case, then run_start...run_end is the run.
2212 If run_type is two_case, then the run is run_start...run_end,
2213 and the case-equivalents end at run_eqv_end. */
2214
2215 enum case_type run_type = strange;
2216 int run_start, run_end, run_eqv_end;
2217
2218 Lisp_Object eqv_table;
2219
2220 if (!RE_TRANSLATE_P (translate))
2221 {
2222 EXTEND_RANGE_TABLE (work_area, 2);
2223 work_area->table[work_area->used++] = (start);
2224 work_area->table[work_area->used++] = (end);
2225 return -1;
2226 }
2227
2228 eqv_table = XCHAR_TABLE (translate)->extras[2];
2229
2230 for (; start <= end; start++)
2231 {
2232 enum case_type this_type;
2233 int eqv = RE_TRANSLATE (eqv_table, start);
2234 int minchar, maxchar;
2235
2236 /* Classify this character */
2237 if (eqv == start)
2238 this_type = one_case;
2239 else if (RE_TRANSLATE (eqv_table, eqv) == start)
2240 this_type = two_case;
2241 else
2242 this_type = strange;
2243
2244 if (start < eqv)
2245 minchar = start, maxchar = eqv;
2246 else
2247 minchar = eqv, maxchar = start;
2248
2249 /* Can this character extend the run in progress? */
2250 if (this_type == strange || this_type != run_type
2251 || !(minchar == run_end + 1
2252 && (run_type == two_case
2253 ? maxchar == run_eqv_end + 1 : 1)))
2254 {
2255 /* No, end the run.
2256 Record each of its equivalent ranges. */
2257 if (run_type == one_case)
2258 {
2259 EXTEND_RANGE_TABLE (work_area, 2);
2260 work_area->table[work_area->used++] = run_start;
2261 work_area->table[work_area->used++] = run_end;
2262 }
2263 else if (run_type == two_case)
2264 {
2265 EXTEND_RANGE_TABLE (work_area, 4);
2266 work_area->table[work_area->used++] = run_start;
2267 work_area->table[work_area->used++] = run_end;
2268 work_area->table[work_area->used++]
2269 = RE_TRANSLATE (eqv_table, run_start);
2270 work_area->table[work_area->used++]
2271 = RE_TRANSLATE (eqv_table, run_end);
2272 }
2273 run_type = strange;
2274 }
2275
2276 if (this_type == strange)
2277 {
2278 /* For a strange character, add each of its equivalents, one
2279 by one. Don't start a range. */
2280 do
2281 {
2282 EXTEND_RANGE_TABLE (work_area, 2);
2283 work_area->table[work_area->used++] = eqv;
2284 work_area->table[work_area->used++] = eqv;
2285 eqv = RE_TRANSLATE (eqv_table, eqv);
2286 }
2287 while (eqv != start);
2288 }
2289
2290 /* Add this char to the run, or start a new run. */
2291 else if (run_type == strange)
2292 {
2293 /* Initialize a new range. */
2294 run_type = this_type;
2295 run_start = start;
2296 run_end = start;
2297 run_eqv_end = RE_TRANSLATE (eqv_table, run_end);
2298 }
2299 else
2300 {
2301 /* Extend a running range. */
2302 run_end = minchar;
2303 run_eqv_end = RE_TRANSLATE (eqv_table, run_end);
2304 }
2305 }
2306
2307 /* If a run is still in progress at the end, finish it now
2308 by recording its equivalent ranges. */
2309 if (run_type == one_case)
2310 {
2311 EXTEND_RANGE_TABLE (work_area, 2);
2312 work_area->table[work_area->used++] = run_start;
2313 work_area->table[work_area->used++] = run_end;
2314 }
2315 else if (run_type == two_case)
2316 {
2317 EXTEND_RANGE_TABLE (work_area, 4);
2318 work_area->table[work_area->used++] = run_start;
2319 work_area->table[work_area->used++] = run_end;
2320 work_area->table[work_area->used++]
2321 = RE_TRANSLATE (eqv_table, run_start);
2322 work_area->table[work_area->used++]
2323 = RE_TRANSLATE (eqv_table, run_end);
2324 }
2325
2326 return -1;
2327 }
2328
2329 #endif /* emacs */
2330
2331 /* Record the the image of the range start..end when passed through
2332 TRANSLATE. This is not necessarily TRANSLATE(start)..TRANSLATE(end)
2333 and is not even necessarily contiguous.
2334 Normally we approximate it with the smallest contiguous range that contains
2335 all the chars we need. However, for Latin-1 we go to extra effort
2336 to do a better job.
2337
2338 This function is not called for ASCII ranges.
2339
2340 Returns -1 if successful, REG_ESPACE if ran out of space. */
2341
2342 static int
2343 set_image_of_range (work_area, start, end, translate)
2344 RE_TRANSLATE_TYPE translate;
2345 struct range_table_work_area *work_area;
2346 re_wchar_t start, end;
2347 {
2348 re_wchar_t cmin, cmax;
2349
2350 #ifdef emacs
2351 /* For Latin-1 ranges, use set_image_of_range_1
2352 to get proper handling of ranges that include letters and nonletters.
2353 For a range that includes the whole of Latin-1, this is not necessary.
2354 For other character sets, we don't bother to get this right. */
2355 if (RE_TRANSLATE_P (translate) && start < 04400
2356 && !(start < 04200 && end >= 04377))
2357 {
2358 int newend;
2359 int tem;
2360 newend = end;
2361 if (newend > 04377)
2362 newend = 04377;
2363 tem = set_image_of_range_1 (work_area, start, newend, translate);
2364 if (tem > 0)
2365 return tem;
2366
2367 start = 04400;
2368 if (end < 04400)
2369 return -1;
2370 }
2371 #endif
2372
2373 EXTEND_RANGE_TABLE (work_area, 2);
2374 work_area->table[work_area->used++] = (start);
2375 work_area->table[work_area->used++] = (end);
2376
2377 cmin = -1, cmax = -1;
2378
2379 if (RE_TRANSLATE_P (translate))
2380 {
2381 int ch;
2382
2383 for (ch = start; ch <= end; ch++)
2384 {
2385 re_wchar_t c = TRANSLATE (ch);
2386 if (! (start <= c && c <= end))
2387 {
2388 if (cmin == -1)
2389 cmin = c, cmax = c;
2390 else
2391 {
2392 cmin = MIN (cmin, c);
2393 cmax = MAX (cmax, c);
2394 }
2395 }
2396 }
2397
2398 if (cmin != -1)
2399 {
2400 EXTEND_RANGE_TABLE (work_area, 2);
2401 work_area->table[work_area->used++] = (cmin);
2402 work_area->table[work_area->used++] = (cmax);
2403 }
2404 }
2405
2406 return -1;
2407 }
2408 #endif /* 0 */
2409 \f
2410 #ifndef MATCH_MAY_ALLOCATE
2411
2412 /* If we cannot allocate large objects within re_match_2_internal,
2413 we make the fail stack and register vectors global.
2414 The fail stack, we grow to the maximum size when a regexp
2415 is compiled.
2416 The register vectors, we adjust in size each time we
2417 compile a regexp, according to the number of registers it needs. */
2418
2419 static fail_stack_type fail_stack;
2420
2421 /* Size with which the following vectors are currently allocated.
2422 That is so we can make them bigger as needed,
2423 but never make them smaller. */
2424 static int regs_allocated_size;
2425
2426 static re_char ** regstart, ** regend;
2427 static re_char **best_regstart, **best_regend;
2428
2429 /* Make the register vectors big enough for NUM_REGS registers,
2430 but don't make them smaller. */
2431
2432 static
2433 regex_grow_registers (num_regs)
2434 int num_regs;
2435 {
2436 if (num_regs > regs_allocated_size)
2437 {
2438 RETALLOC_IF (regstart, num_regs, re_char *);
2439 RETALLOC_IF (regend, num_regs, re_char *);
2440 RETALLOC_IF (best_regstart, num_regs, re_char *);
2441 RETALLOC_IF (best_regend, num_regs, re_char *);
2442
2443 regs_allocated_size = num_regs;
2444 }
2445 }
2446
2447 #endif /* not MATCH_MAY_ALLOCATE */
2448 \f
2449 static boolean group_in_compile_stack _RE_ARGS ((compile_stack_type
2450 compile_stack,
2451 regnum_t regnum));
2452
2453 /* `regex_compile' compiles PATTERN (of length SIZE) according to SYNTAX.
2454 Returns one of error codes defined in `regex.h', or zero for success.
2455
2456 Assumes the `allocated' (and perhaps `buffer') and `translate'
2457 fields are set in BUFP on entry.
2458
2459 If it succeeds, results are put in BUFP (if it returns an error, the
2460 contents of BUFP are undefined):
2461 `buffer' is the compiled pattern;
2462 `syntax' is set to SYNTAX;
2463 `used' is set to the length of the compiled pattern;
2464 `fastmap_accurate' is zero;
2465 `re_nsub' is the number of subexpressions in PATTERN;
2466 `not_bol' and `not_eol' are zero;
2467
2468 The `fastmap' field is neither examined nor set. */
2469
2470 /* Insert the `jump' from the end of last alternative to "here".
2471 The space for the jump has already been allocated. */
2472 #define FIXUP_ALT_JUMP() \
2473 do { \
2474 if (fixup_alt_jump) \
2475 STORE_JUMP (jump, fixup_alt_jump, b); \
2476 } while (0)
2477
2478
2479 /* Return, freeing storage we allocated. */
2480 #define FREE_STACK_RETURN(value) \
2481 do { \
2482 FREE_RANGE_TABLE_WORK_AREA (range_table_work); \
2483 free (compile_stack.stack); \
2484 return value; \
2485 } while (0)
2486
2487 static reg_errcode_t
2488 regex_compile (pattern, size, syntax, bufp)
2489 re_char *pattern;
2490 size_t size;
2491 reg_syntax_t syntax;
2492 struct re_pattern_buffer *bufp;
2493 {
2494 /* We fetch characters from PATTERN here. */
2495 register re_wchar_t c, c1;
2496
2497 /* A random temporary spot in PATTERN. */
2498 re_char *p1;
2499
2500 /* Points to the end of the buffer, where we should append. */
2501 register unsigned char *b;
2502
2503 /* Keeps track of unclosed groups. */
2504 compile_stack_type compile_stack;
2505
2506 /* Points to the current (ending) position in the pattern. */
2507 #ifdef AIX
2508 /* `const' makes AIX compiler fail. */
2509 unsigned char *p = pattern;
2510 #else
2511 re_char *p = pattern;
2512 #endif
2513 re_char *pend = pattern + size;
2514
2515 /* How to translate the characters in the pattern. */
2516 RE_TRANSLATE_TYPE translate = bufp->translate;
2517
2518 /* Address of the count-byte of the most recently inserted `exactn'
2519 command. This makes it possible to tell if a new exact-match
2520 character can be added to that command or if the character requires
2521 a new `exactn' command. */
2522 unsigned char *pending_exact = 0;
2523
2524 /* Address of start of the most recently finished expression.
2525 This tells, e.g., postfix * where to find the start of its
2526 operand. Reset at the beginning of groups and alternatives. */
2527 unsigned char *laststart = 0;
2528
2529 /* Address of beginning of regexp, or inside of last group. */
2530 unsigned char *begalt;
2531
2532 /* Place in the uncompiled pattern (i.e., the {) to
2533 which to go back if the interval is invalid. */
2534 re_char *beg_interval;
2535
2536 /* Address of the place where a forward jump should go to the end of
2537 the containing expression. Each alternative of an `or' -- except the
2538 last -- ends with a forward jump of this sort. */
2539 unsigned char *fixup_alt_jump = 0;
2540
2541 /* Counts open-groups as they are encountered. Remembered for the
2542 matching close-group on the compile stack, so the same register
2543 number is put in the stop_memory as the start_memory. */
2544 regnum_t regnum = 0;
2545
2546 /* Work area for range table of charset. */
2547 struct range_table_work_area range_table_work;
2548
2549 /* If the object matched can contain multibyte characters. */
2550 const boolean multibyte = RE_MULTIBYTE_P (bufp);
2551
2552 /* If a target of matching can contain multibyte characters. */
2553 const boolean target_multibyte = RE_TARGET_MULTIBYTE_P (bufp);
2554
2555 /* Nonzero if we have pushed down into a subpattern. */
2556 int in_subpattern = 0;
2557
2558 /* These hold the values of p, pattern, and pend from the main
2559 pattern when we have pushed into a subpattern. */
2560 re_char *main_p;
2561 re_char *main_pattern;
2562 re_char *main_pend;
2563
2564 #ifdef DEBUG
2565 debug++;
2566 DEBUG_PRINT1 ("\nCompiling pattern: ");
2567 if (debug > 0)
2568 {
2569 unsigned debug_count;
2570
2571 for (debug_count = 0; debug_count < size; debug_count++)
2572 putchar (pattern[debug_count]);
2573 putchar ('\n');
2574 }
2575 #endif /* DEBUG */
2576
2577 /* Initialize the compile stack. */
2578 compile_stack.stack = TALLOC (INIT_COMPILE_STACK_SIZE, compile_stack_elt_t);
2579 if (compile_stack.stack == NULL)
2580 return REG_ESPACE;
2581
2582 compile_stack.size = INIT_COMPILE_STACK_SIZE;
2583 compile_stack.avail = 0;
2584
2585 range_table_work.table = 0;
2586 range_table_work.allocated = 0;
2587
2588 /* Initialize the pattern buffer. */
2589 bufp->syntax = syntax;
2590 bufp->fastmap_accurate = 0;
2591 bufp->not_bol = bufp->not_eol = 0;
2592
2593 /* Set `used' to zero, so that if we return an error, the pattern
2594 printer (for debugging) will think there's no pattern. We reset it
2595 at the end. */
2596 bufp->used = 0;
2597
2598 /* Always count groups, whether or not bufp->no_sub is set. */
2599 bufp->re_nsub = 0;
2600
2601 #if !defined emacs && !defined SYNTAX_TABLE
2602 /* Initialize the syntax table. */
2603 init_syntax_once ();
2604 #endif
2605
2606 if (bufp->allocated == 0)
2607 {
2608 if (bufp->buffer)
2609 { /* If zero allocated, but buffer is non-null, try to realloc
2610 enough space. This loses if buffer's address is bogus, but
2611 that is the user's responsibility. */
2612 RETALLOC (bufp->buffer, INIT_BUF_SIZE, unsigned char);
2613 }
2614 else
2615 { /* Caller did not allocate a buffer. Do it for them. */
2616 bufp->buffer = TALLOC (INIT_BUF_SIZE, unsigned char);
2617 }
2618 if (!bufp->buffer) FREE_STACK_RETURN (REG_ESPACE);
2619
2620 bufp->allocated = INIT_BUF_SIZE;
2621 }
2622
2623 begalt = b = bufp->buffer;
2624
2625 /* Loop through the uncompiled pattern until we're at the end. */
2626 while (1)
2627 {
2628 if (p == pend)
2629 {
2630 /* If this is the end of an included regexp,
2631 pop back to the main regexp and try again. */
2632 if (in_subpattern)
2633 {
2634 in_subpattern = 0;
2635 pattern = main_pattern;
2636 p = main_p;
2637 pend = main_pend;
2638 continue;
2639 }
2640 /* If this is the end of the main regexp, we are done. */
2641 break;
2642 }
2643
2644 PATFETCH (c);
2645
2646 switch (c)
2647 {
2648 case ' ':
2649 {
2650 re_char *p1 = p;
2651
2652 /* If there's no special whitespace regexp, treat
2653 spaces normally. And don't try to do this recursively. */
2654 if (!whitespace_regexp || in_subpattern)
2655 goto normal_char;
2656
2657 /* Peek past following spaces. */
2658 while (p1 != pend)
2659 {
2660 if (*p1 != ' ')
2661 break;
2662 p1++;
2663 }
2664 /* If the spaces are followed by a repetition op,
2665 treat them normally. */
2666 if (p1 != pend
2667 && (*p1 == '*' || *p1 == '+' || *p1 == '?'
2668 || (*p1 == '\\' && p1 + 1 != pend && p1[1] == '{')))
2669 goto normal_char;
2670
2671 /* Replace the spaces with the whitespace regexp. */
2672 in_subpattern = 1;
2673 main_p = p1;
2674 main_pend = pend;
2675 main_pattern = pattern;
2676 p = pattern = whitespace_regexp;
2677 pend = p + strlen (p);
2678 break;
2679 }
2680
2681 case '^':
2682 {
2683 if ( /* If at start of pattern, it's an operator. */
2684 p == pattern + 1
2685 /* If context independent, it's an operator. */
2686 || syntax & RE_CONTEXT_INDEP_ANCHORS
2687 /* Otherwise, depends on what's come before. */
2688 || at_begline_loc_p (pattern, p, syntax))
2689 BUF_PUSH ((syntax & RE_NO_NEWLINE_ANCHOR) ? begbuf : begline);
2690 else
2691 goto normal_char;
2692 }
2693 break;
2694
2695
2696 case '$':
2697 {
2698 if ( /* If at end of pattern, it's an operator. */
2699 p == pend
2700 /* If context independent, it's an operator. */
2701 || syntax & RE_CONTEXT_INDEP_ANCHORS
2702 /* Otherwise, depends on what's next. */
2703 || at_endline_loc_p (p, pend, syntax))
2704 BUF_PUSH ((syntax & RE_NO_NEWLINE_ANCHOR) ? endbuf : endline);
2705 else
2706 goto normal_char;
2707 }
2708 break;
2709
2710
2711 case '+':
2712 case '?':
2713 if ((syntax & RE_BK_PLUS_QM)
2714 || (syntax & RE_LIMITED_OPS))
2715 goto normal_char;
2716 handle_plus:
2717 case '*':
2718 /* If there is no previous pattern... */
2719 if (!laststart)
2720 {
2721 if (syntax & RE_CONTEXT_INVALID_OPS)
2722 FREE_STACK_RETURN (REG_BADRPT);
2723 else if (!(syntax & RE_CONTEXT_INDEP_OPS))
2724 goto normal_char;
2725 }
2726
2727 {
2728 /* 1 means zero (many) matches is allowed. */
2729 boolean zero_times_ok = 0, many_times_ok = 0;
2730 boolean greedy = 1;
2731
2732 /* If there is a sequence of repetition chars, collapse it
2733 down to just one (the right one). We can't combine
2734 interval operators with these because of, e.g., `a{2}*',
2735 which should only match an even number of `a's. */
2736
2737 for (;;)
2738 {
2739 if ((syntax & RE_FRUGAL)
2740 && c == '?' && (zero_times_ok || many_times_ok))
2741 greedy = 0;
2742 else
2743 {
2744 zero_times_ok |= c != '+';
2745 many_times_ok |= c != '?';
2746 }
2747
2748 if (p == pend)
2749 break;
2750 else if (*p == '*'
2751 || (!(syntax & RE_BK_PLUS_QM)
2752 && (*p == '+' || *p == '?')))
2753 ;
2754 else if (syntax & RE_BK_PLUS_QM && *p == '\\')
2755 {
2756 if (p+1 == pend)
2757 FREE_STACK_RETURN (REG_EESCAPE);
2758 if (p[1] == '+' || p[1] == '?')
2759 PATFETCH (c); /* Gobble up the backslash. */
2760 else
2761 break;
2762 }
2763 else
2764 break;
2765 /* If we get here, we found another repeat character. */
2766 PATFETCH (c);
2767 }
2768
2769 /* Star, etc. applied to an empty pattern is equivalent
2770 to an empty pattern. */
2771 if (!laststart || laststart == b)
2772 break;
2773
2774 /* Now we know whether or not zero matches is allowed
2775 and also whether or not two or more matches is allowed. */
2776 if (greedy)
2777 {
2778 if (many_times_ok)
2779 {
2780 boolean simple = skip_one_char (laststart) == b;
2781 unsigned int startoffset = 0;
2782 re_opcode_t ofj =
2783 /* Check if the loop can match the empty string. */
2784 (simple || !analyse_first (laststart, b, NULL, 0))
2785 ? on_failure_jump : on_failure_jump_loop;
2786 assert (skip_one_char (laststart) <= b);
2787
2788 if (!zero_times_ok && simple)
2789 { /* Since simple * loops can be made faster by using
2790 on_failure_keep_string_jump, we turn simple P+
2791 into PP* if P is simple. */
2792 unsigned char *p1, *p2;
2793 startoffset = b - laststart;
2794 GET_BUFFER_SPACE (startoffset);
2795 p1 = b; p2 = laststart;
2796 while (p2 < p1)
2797 *b++ = *p2++;
2798 zero_times_ok = 1;
2799 }
2800
2801 GET_BUFFER_SPACE (6);
2802 if (!zero_times_ok)
2803 /* A + loop. */
2804 STORE_JUMP (ofj, b, b + 6);
2805 else
2806 /* Simple * loops can use on_failure_keep_string_jump
2807 depending on what follows. But since we don't know
2808 that yet, we leave the decision up to
2809 on_failure_jump_smart. */
2810 INSERT_JUMP (simple ? on_failure_jump_smart : ofj,
2811 laststart + startoffset, b + 6);
2812 b += 3;
2813 STORE_JUMP (jump, b, laststart + startoffset);
2814 b += 3;
2815 }
2816 else
2817 {
2818 /* A simple ? pattern. */
2819 assert (zero_times_ok);
2820 GET_BUFFER_SPACE (3);
2821 INSERT_JUMP (on_failure_jump, laststart, b + 3);
2822 b += 3;
2823 }
2824 }
2825 else /* not greedy */
2826 { /* I wish the greedy and non-greedy cases could be merged. */
2827
2828 GET_BUFFER_SPACE (7); /* We might use less. */
2829 if (many_times_ok)
2830 {
2831 boolean emptyp = analyse_first (laststart, b, NULL, 0);
2832
2833 /* The non-greedy multiple match looks like
2834 a repeat..until: we only need a conditional jump
2835 at the end of the loop. */
2836 if (emptyp) BUF_PUSH (no_op);
2837 STORE_JUMP (emptyp ? on_failure_jump_nastyloop
2838 : on_failure_jump, b, laststart);
2839 b += 3;
2840 if (zero_times_ok)
2841 {
2842 /* The repeat...until naturally matches one or more.
2843 To also match zero times, we need to first jump to
2844 the end of the loop (its conditional jump). */
2845 INSERT_JUMP (jump, laststart, b);
2846 b += 3;
2847 }
2848 }
2849 else
2850 {
2851 /* non-greedy a?? */
2852 INSERT_JUMP (jump, laststart, b + 3);
2853 b += 3;
2854 INSERT_JUMP (on_failure_jump, laststart, laststart + 6);
2855 b += 3;
2856 }
2857 }
2858 }
2859 pending_exact = 0;
2860 break;
2861
2862
2863 case '.':
2864 laststart = b;
2865 BUF_PUSH (anychar);
2866 break;
2867
2868
2869 case '[':
2870 {
2871 CLEAR_RANGE_TABLE_WORK_USED (range_table_work);
2872
2873 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2874
2875 /* Ensure that we have enough space to push a charset: the
2876 opcode, the length count, and the bitset; 34 bytes in all. */
2877 GET_BUFFER_SPACE (34);
2878
2879 laststart = b;
2880
2881 /* We test `*p == '^' twice, instead of using an if
2882 statement, so we only need one BUF_PUSH. */
2883 BUF_PUSH (*p == '^' ? charset_not : charset);
2884 if (*p == '^')
2885 p++;
2886
2887 /* Remember the first position in the bracket expression. */
2888 p1 = p;
2889
2890 /* Push the number of bytes in the bitmap. */
2891 BUF_PUSH ((1 << BYTEWIDTH) / BYTEWIDTH);
2892
2893 /* Clear the whole map. */
2894 bzero (b, (1 << BYTEWIDTH) / BYTEWIDTH);
2895
2896 /* charset_not matches newline according to a syntax bit. */
2897 if ((re_opcode_t) b[-2] == charset_not
2898 && (syntax & RE_HAT_LISTS_NOT_NEWLINE))
2899 SET_LIST_BIT ('\n');
2900
2901 /* Read in characters and ranges, setting map bits. */
2902 for (;;)
2903 {
2904 boolean escaped_char = false;
2905 const unsigned char *p2 = p;
2906
2907 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2908
2909 /* Don't translate yet. The range TRANSLATE(X..Y) cannot
2910 always be determined from TRANSLATE(X) and TRANSLATE(Y)
2911 So the translation is done later in a loop. Example:
2912 (let ((case-fold-search t)) (string-match "[A-_]" "A")) */
2913 PATFETCH (c);
2914
2915 /* \ might escape characters inside [...] and [^...]. */
2916 if ((syntax & RE_BACKSLASH_ESCAPE_IN_LISTS) && c == '\\')
2917 {
2918 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
2919
2920 PATFETCH (c);
2921 escaped_char = true;
2922 }
2923 else
2924 {
2925 /* Could be the end of the bracket expression. If it's
2926 not (i.e., when the bracket expression is `[]' so
2927 far), the ']' character bit gets set way below. */
2928 if (c == ']' && p2 != p1)
2929 break;
2930 }
2931
2932 /* See if we're at the beginning of a possible character
2933 class. */
2934
2935 if (!escaped_char &&
2936 syntax & RE_CHAR_CLASSES && c == '[' && *p == ':')
2937 {
2938 /* Leave room for the null. */
2939 unsigned char str[CHAR_CLASS_MAX_LENGTH + 1];
2940 const unsigned char *class_beg;
2941
2942 PATFETCH (c);
2943 c1 = 0;
2944 class_beg = p;
2945
2946 /* If pattern is `[[:'. */
2947 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2948
2949 for (;;)
2950 {
2951 PATFETCH (c);
2952 if ((c == ':' && *p == ']') || p == pend)
2953 break;
2954 if (c1 < CHAR_CLASS_MAX_LENGTH)
2955 str[c1++] = c;
2956 else
2957 /* This is in any case an invalid class name. */
2958 str[0] = '\0';
2959 }
2960 str[c1] = '\0';
2961
2962 /* If isn't a word bracketed by `[:' and `:]':
2963 undo the ending character, the letters, and
2964 leave the leading `:' and `[' (but set bits for
2965 them). */
2966 if (c == ':' && *p == ']')
2967 {
2968 re_wchar_t ch;
2969 re_wctype_t cc;
2970 int limit;
2971
2972 cc = re_wctype (str);
2973
2974 if (cc == 0)
2975 FREE_STACK_RETURN (REG_ECTYPE);
2976
2977 /* Throw away the ] at the end of the character
2978 class. */
2979 PATFETCH (c);
2980
2981 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2982
2983 /* Most character classes in a multibyte match
2984 just set a flag. Exceptions are is_blank,
2985 is_digit, is_cntrl, and is_xdigit, since
2986 they can only match ASCII characters. We
2987 don't need to handle them for multibyte.
2988 They are distinguished by a negative wctype. */
2989
2990 for (ch = 0; ch < 128; ++ch)
2991 if (re_iswctype (btowc (ch), cc))
2992 {
2993 c = TRANSLATE (ch);
2994 SET_LIST_BIT (c);
2995 }
2996
2997 if (target_multibyte)
2998 {
2999 SET_RANGE_TABLE_WORK_AREA_BIT
3000 (range_table_work, re_wctype_to_bit (cc));
3001 }
3002 else
3003 {
3004 for (ch = 0; ch < (1 << BYTEWIDTH); ++ch)
3005 {
3006 c = ch;
3007 MAKE_CHAR_MULTIBYTE (c);
3008 if (re_iswctype (btowc (c), cc))
3009 {
3010 c = TRANSLATE (c);
3011 MAKE_CHAR_UNIBYTE (c);
3012 SET_LIST_BIT (c);
3013 }
3014 }
3015 }
3016
3017 /* Repeat the loop. */
3018 continue;
3019 }
3020 else
3021 {
3022 /* Go back to right after the "[:". */
3023 p = class_beg;
3024 SET_LIST_BIT ('[');
3025
3026 /* Because the `:' may starts the range, we
3027 can't simply set bit and repeat the loop.
3028 Instead, just set it to C and handle below. */
3029 c = ':';
3030 }
3031 }
3032
3033 if (p < pend && p[0] == '-' && p[1] != ']')
3034 {
3035
3036 /* Discard the `-'. */
3037 PATFETCH (c1);
3038
3039 /* Fetch the character which ends the range. */
3040 PATFETCH (c1);
3041 if (c > c1)
3042 {
3043 if (syntax & RE_NO_EMPTY_RANGES)
3044 FREE_STACK_RETURN (REG_ERANGEX);
3045 /* Else, repeat the loop. */
3046 }
3047 }
3048 else
3049 /* Range from C to C. */
3050 c1 = c;
3051
3052 #ifndef emacs
3053 c = TRANSLATE (c);
3054 c1 = TRANSLATE (c1);
3055 /* Set the range into bitmap */
3056 for (; c <= c1; c++)
3057 SET_LIST_BIT (TRANSLATE (c));
3058 #else /* not emacs */
3059 if (target_multibyte)
3060 {
3061 if (c1 >= 128)
3062 {
3063 re_wchar_t c0 = MAX (c, 128);
3064
3065 SETUP_MULTIBYTE_RANGE (range_table_work, c0, c1);
3066 c1 = 127;
3067 }
3068 for (; c <= c1; c++)
3069 SET_LIST_BIT (TRANSLATE (c));
3070 }
3071 else
3072 {
3073 re_wchar_t c0;
3074
3075 for (; c <= c1; c++)
3076 {
3077 c0 = c;
3078 if (! multibyte)
3079 MAKE_CHAR_MULTIBYTE (c0);
3080 c0 = TRANSLATE (c0);
3081 MAKE_CHAR_UNIBYTE (c0);
3082 SET_LIST_BIT (c0);
3083 }
3084 }
3085 #endif /* not emacs */
3086 }
3087
3088 /* Discard any (non)matching list bytes that are all 0 at the
3089 end of the map. Decrease the map-length byte too. */
3090 while ((int) b[-1] > 0 && b[b[-1] - 1] == 0)
3091 b[-1]--;
3092 b += b[-1];
3093
3094 /* Build real range table from work area. */
3095 if (RANGE_TABLE_WORK_USED (range_table_work)
3096 || RANGE_TABLE_WORK_BITS (range_table_work))
3097 {
3098 int i;
3099 int used = RANGE_TABLE_WORK_USED (range_table_work);
3100
3101 /* Allocate space for COUNT + RANGE_TABLE. Needs two
3102 bytes for flags, two for COUNT, and three bytes for
3103 each character. */
3104 GET_BUFFER_SPACE (4 + used * 3);
3105
3106 /* Indicate the existence of range table. */
3107 laststart[1] |= 0x80;
3108
3109 /* Store the character class flag bits into the range table.
3110 If not in emacs, these flag bits are always 0. */
3111 *b++ = RANGE_TABLE_WORK_BITS (range_table_work) & 0xff;
3112 *b++ = RANGE_TABLE_WORK_BITS (range_table_work) >> 8;
3113
3114 STORE_NUMBER_AND_INCR (b, used / 2);
3115 for (i = 0; i < used; i++)
3116 STORE_CHARACTER_AND_INCR
3117 (b, RANGE_TABLE_WORK_ELT (range_table_work, i));
3118 }
3119 }
3120 break;
3121
3122
3123 case '(':
3124 if (syntax & RE_NO_BK_PARENS)
3125 goto handle_open;
3126 else
3127 goto normal_char;
3128
3129
3130 case ')':
3131 if (syntax & RE_NO_BK_PARENS)
3132 goto handle_close;
3133 else
3134 goto normal_char;
3135
3136
3137 case '\n':
3138 if (syntax & RE_NEWLINE_ALT)
3139 goto handle_alt;
3140 else
3141 goto normal_char;
3142
3143
3144 case '|':
3145 if (syntax & RE_NO_BK_VBAR)
3146 goto handle_alt;
3147 else
3148 goto normal_char;
3149
3150
3151 case '{':
3152 if (syntax & RE_INTERVALS && syntax & RE_NO_BK_BRACES)
3153 goto handle_interval;
3154 else
3155 goto normal_char;
3156
3157
3158 case '\\':
3159 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
3160
3161 /* Do not translate the character after the \, so that we can
3162 distinguish, e.g., \B from \b, even if we normally would
3163 translate, e.g., B to b. */
3164 PATFETCH (c);
3165
3166 switch (c)
3167 {
3168 case '(':
3169 if (syntax & RE_NO_BK_PARENS)
3170 goto normal_backslash;
3171
3172 handle_open:
3173 {
3174 int shy = 0;
3175 if (p+1 < pend)
3176 {
3177 /* Look for a special (?...) construct */
3178 if ((syntax & RE_SHY_GROUPS) && *p == '?')
3179 {
3180 PATFETCH (c); /* Gobble up the '?'. */
3181 PATFETCH (c);
3182 switch (c)
3183 {
3184 case ':': shy = 1; break;
3185 default:
3186 /* Only (?:...) is supported right now. */
3187 FREE_STACK_RETURN (REG_BADPAT);
3188 }
3189 }
3190 }
3191
3192 if (!shy)
3193 {
3194 bufp->re_nsub++;
3195 regnum++;
3196 }
3197
3198 if (COMPILE_STACK_FULL)
3199 {
3200 RETALLOC (compile_stack.stack, compile_stack.size << 1,
3201 compile_stack_elt_t);
3202 if (compile_stack.stack == NULL) return REG_ESPACE;
3203
3204 compile_stack.size <<= 1;
3205 }
3206
3207 /* These are the values to restore when we hit end of this
3208 group. They are all relative offsets, so that if the
3209 whole pattern moves because of realloc, they will still
3210 be valid. */
3211 COMPILE_STACK_TOP.begalt_offset = begalt - bufp->buffer;
3212 COMPILE_STACK_TOP.fixup_alt_jump
3213 = fixup_alt_jump ? fixup_alt_jump - bufp->buffer + 1 : 0;
3214 COMPILE_STACK_TOP.laststart_offset = b - bufp->buffer;
3215 COMPILE_STACK_TOP.regnum = shy ? -regnum : regnum;
3216
3217 /* Do not push a
3218 start_memory for groups beyond the last one we can
3219 represent in the compiled pattern. */
3220 if (regnum <= MAX_REGNUM && !shy)
3221 BUF_PUSH_2 (start_memory, regnum);
3222
3223 compile_stack.avail++;
3224
3225 fixup_alt_jump = 0;
3226 laststart = 0;
3227 begalt = b;
3228 /* If we've reached MAX_REGNUM groups, then this open
3229 won't actually generate any code, so we'll have to
3230 clear pending_exact explicitly. */
3231 pending_exact = 0;
3232 break;
3233 }
3234
3235 case ')':
3236 if (syntax & RE_NO_BK_PARENS) goto normal_backslash;
3237
3238 if (COMPILE_STACK_EMPTY)
3239 {
3240 if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD)
3241 goto normal_backslash;
3242 else
3243 FREE_STACK_RETURN (REG_ERPAREN);
3244 }
3245
3246 handle_close:
3247 FIXUP_ALT_JUMP ();
3248
3249 /* See similar code for backslashed left paren above. */
3250 if (COMPILE_STACK_EMPTY)
3251 {
3252 if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD)
3253 goto normal_char;
3254 else
3255 FREE_STACK_RETURN (REG_ERPAREN);
3256 }
3257
3258 /* Since we just checked for an empty stack above, this
3259 ``can't happen''. */
3260 assert (compile_stack.avail != 0);
3261 {
3262 /* We don't just want to restore into `regnum', because
3263 later groups should continue to be numbered higher,
3264 as in `(ab)c(de)' -- the second group is #2. */
3265 regnum_t this_group_regnum;
3266
3267 compile_stack.avail--;
3268 begalt = bufp->buffer + COMPILE_STACK_TOP.begalt_offset;
3269 fixup_alt_jump
3270 = COMPILE_STACK_TOP.fixup_alt_jump
3271 ? bufp->buffer + COMPILE_STACK_TOP.fixup_alt_jump - 1
3272 : 0;
3273 laststart = bufp->buffer + COMPILE_STACK_TOP.laststart_offset;
3274 this_group_regnum = COMPILE_STACK_TOP.regnum;
3275 /* If we've reached MAX_REGNUM groups, then this open
3276 won't actually generate any code, so we'll have to
3277 clear pending_exact explicitly. */
3278 pending_exact = 0;
3279
3280 /* We're at the end of the group, so now we know how many
3281 groups were inside this one. */
3282 if (this_group_regnum <= MAX_REGNUM && this_group_regnum > 0)
3283 BUF_PUSH_2 (stop_memory, this_group_regnum);
3284 }
3285 break;
3286
3287
3288 case '|': /* `\|'. */
3289 if (syntax & RE_LIMITED_OPS || syntax & RE_NO_BK_VBAR)
3290 goto normal_backslash;
3291 handle_alt:
3292 if (syntax & RE_LIMITED_OPS)
3293 goto normal_char;
3294
3295 /* Insert before the previous alternative a jump which
3296 jumps to this alternative if the former fails. */
3297 GET_BUFFER_SPACE (3);
3298 INSERT_JUMP (on_failure_jump, begalt, b + 6);
3299 pending_exact = 0;
3300 b += 3;
3301
3302 /* The alternative before this one has a jump after it
3303 which gets executed if it gets matched. Adjust that
3304 jump so it will jump to this alternative's analogous
3305 jump (put in below, which in turn will jump to the next
3306 (if any) alternative's such jump, etc.). The last such
3307 jump jumps to the correct final destination. A picture:
3308 _____ _____
3309 | | | |
3310 | v | v
3311 a | b | c
3312
3313 If we are at `b', then fixup_alt_jump right now points to a
3314 three-byte space after `a'. We'll put in the jump, set
3315 fixup_alt_jump to right after `b', and leave behind three
3316 bytes which we'll fill in when we get to after `c'. */
3317
3318 FIXUP_ALT_JUMP ();
3319
3320 /* Mark and leave space for a jump after this alternative,
3321 to be filled in later either by next alternative or
3322 when know we're at the end of a series of alternatives. */
3323 fixup_alt_jump = b;
3324 GET_BUFFER_SPACE (3);
3325 b += 3;
3326
3327 laststart = 0;
3328 begalt = b;
3329 break;
3330
3331
3332 case '{':
3333 /* If \{ is a literal. */
3334 if (!(syntax & RE_INTERVALS)
3335 /* If we're at `\{' and it's not the open-interval
3336 operator. */
3337 || (syntax & RE_NO_BK_BRACES))
3338 goto normal_backslash;
3339
3340 handle_interval:
3341 {
3342 /* If got here, then the syntax allows intervals. */
3343
3344 /* At least (most) this many matches must be made. */
3345 int lower_bound = 0, upper_bound = -1;
3346
3347 beg_interval = p;
3348
3349 GET_UNSIGNED_NUMBER (lower_bound);
3350
3351 if (c == ',')
3352 GET_UNSIGNED_NUMBER (upper_bound);
3353 else
3354 /* Interval such as `{1}' => match exactly once. */
3355 upper_bound = lower_bound;
3356
3357 if (lower_bound < 0 || upper_bound > RE_DUP_MAX
3358 || (upper_bound >= 0 && lower_bound > upper_bound))
3359 FREE_STACK_RETURN (REG_BADBR);
3360
3361 if (!(syntax & RE_NO_BK_BRACES))
3362 {
3363 if (c != '\\')
3364 FREE_STACK_RETURN (REG_BADBR);
3365 if (p == pend)
3366 FREE_STACK_RETURN (REG_EESCAPE);
3367 PATFETCH (c);
3368 }
3369
3370 if (c != '}')
3371 FREE_STACK_RETURN (REG_BADBR);
3372
3373 /* We just parsed a valid interval. */
3374
3375 /* If it's invalid to have no preceding re. */
3376 if (!laststart)
3377 {
3378 if (syntax & RE_CONTEXT_INVALID_OPS)
3379 FREE_STACK_RETURN (REG_BADRPT);
3380 else if (syntax & RE_CONTEXT_INDEP_OPS)
3381 laststart = b;
3382 else
3383 goto unfetch_interval;
3384 }
3385
3386 if (upper_bound == 0)
3387 /* If the upper bound is zero, just drop the sub pattern
3388 altogether. */
3389 b = laststart;
3390 else if (lower_bound == 1 && upper_bound == 1)
3391 /* Just match it once: nothing to do here. */
3392 ;
3393
3394 /* Otherwise, we have a nontrivial interval. When
3395 we're all done, the pattern will look like:
3396 set_number_at <jump count> <upper bound>
3397 set_number_at <succeed_n count> <lower bound>
3398 succeed_n <after jump addr> <succeed_n count>
3399 <body of loop>
3400 jump_n <succeed_n addr> <jump count>
3401 (The upper bound and `jump_n' are omitted if
3402 `upper_bound' is 1, though.) */
3403 else
3404 { /* If the upper bound is > 1, we need to insert
3405 more at the end of the loop. */
3406 unsigned int nbytes = (upper_bound < 0 ? 3
3407 : upper_bound > 1 ? 5 : 0);
3408 unsigned int startoffset = 0;
3409
3410 GET_BUFFER_SPACE (20); /* We might use less. */
3411
3412 if (lower_bound == 0)
3413 {
3414 /* A succeed_n that starts with 0 is really a
3415 a simple on_failure_jump_loop. */
3416 INSERT_JUMP (on_failure_jump_loop, laststart,
3417 b + 3 + nbytes);
3418 b += 3;
3419 }
3420 else
3421 {
3422 /* Initialize lower bound of the `succeed_n', even
3423 though it will be set during matching by its
3424 attendant `set_number_at' (inserted next),
3425 because `re_compile_fastmap' needs to know.
3426 Jump to the `jump_n' we might insert below. */
3427 INSERT_JUMP2 (succeed_n, laststart,
3428 b + 5 + nbytes,
3429 lower_bound);
3430 b += 5;
3431
3432 /* Code to initialize the lower bound. Insert
3433 before the `succeed_n'. The `5' is the last two
3434 bytes of this `set_number_at', plus 3 bytes of
3435 the following `succeed_n'. */
3436 insert_op2 (set_number_at, laststart, 5, lower_bound, b);
3437 b += 5;
3438 startoffset += 5;
3439 }
3440
3441 if (upper_bound < 0)
3442 {
3443 /* A negative upper bound stands for infinity,
3444 in which case it degenerates to a plain jump. */
3445 STORE_JUMP (jump, b, laststart + startoffset);
3446 b += 3;
3447 }
3448 else if (upper_bound > 1)
3449 { /* More than one repetition is allowed, so
3450 append a backward jump to the `succeed_n'
3451 that starts this interval.
3452
3453 When we've reached this during matching,
3454 we'll have matched the interval once, so
3455 jump back only `upper_bound - 1' times. */
3456 STORE_JUMP2 (jump_n, b, laststart + startoffset,
3457 upper_bound - 1);
3458 b += 5;
3459
3460 /* The location we want to set is the second
3461 parameter of the `jump_n'; that is `b-2' as
3462 an absolute address. `laststart' will be
3463 the `set_number_at' we're about to insert;
3464 `laststart+3' the number to set, the source
3465 for the relative address. But we are
3466 inserting into the middle of the pattern --
3467 so everything is getting moved up by 5.
3468 Conclusion: (b - 2) - (laststart + 3) + 5,
3469 i.e., b - laststart.
3470
3471 We insert this at the beginning of the loop
3472 so that if we fail during matching, we'll
3473 reinitialize the bounds. */
3474 insert_op2 (set_number_at, laststart, b - laststart,
3475 upper_bound - 1, b);
3476 b += 5;
3477 }
3478 }
3479 pending_exact = 0;
3480 beg_interval = NULL;
3481 }
3482 break;
3483
3484 unfetch_interval:
3485 /* If an invalid interval, match the characters as literals. */
3486 assert (beg_interval);
3487 p = beg_interval;
3488 beg_interval = NULL;
3489
3490 /* normal_char and normal_backslash need `c'. */
3491 c = '{';
3492
3493 if (!(syntax & RE_NO_BK_BRACES))
3494 {
3495 assert (p > pattern && p[-1] == '\\');
3496 goto normal_backslash;
3497 }
3498 else
3499 goto normal_char;
3500
3501 #ifdef emacs
3502 /* There is no way to specify the before_dot and after_dot
3503 operators. rms says this is ok. --karl */
3504 case '=':
3505 BUF_PUSH (at_dot);
3506 break;
3507
3508 case 's':
3509 laststart = b;
3510 PATFETCH (c);
3511 BUF_PUSH_2 (syntaxspec, syntax_spec_code[c]);
3512 break;
3513
3514 case 'S':
3515 laststart = b;
3516 PATFETCH (c);
3517 BUF_PUSH_2 (notsyntaxspec, syntax_spec_code[c]);
3518 break;
3519
3520 case 'c':
3521 laststart = b;
3522 PATFETCH (c);
3523 BUF_PUSH_2 (categoryspec, c);
3524 break;
3525
3526 case 'C':
3527 laststart = b;
3528 PATFETCH (c);
3529 BUF_PUSH_2 (notcategoryspec, c);
3530 break;
3531 #endif /* emacs */
3532
3533
3534 case 'w':
3535 if (syntax & RE_NO_GNU_OPS)
3536 goto normal_char;
3537 laststart = b;
3538 BUF_PUSH_2 (syntaxspec, Sword);
3539 break;
3540
3541
3542 case 'W':
3543 if (syntax & RE_NO_GNU_OPS)
3544 goto normal_char;
3545 laststart = b;
3546 BUF_PUSH_2 (notsyntaxspec, Sword);
3547 break;
3548
3549
3550 case '<':
3551 if (syntax & RE_NO_GNU_OPS)
3552 goto normal_char;
3553 BUF_PUSH (wordbeg);
3554 break;
3555
3556 case '>':
3557 if (syntax & RE_NO_GNU_OPS)
3558 goto normal_char;
3559 BUF_PUSH (wordend);
3560 break;
3561
3562 case '_':
3563 if (syntax & RE_NO_GNU_OPS)
3564 goto normal_char;
3565 laststart = b;
3566 PATFETCH (c);
3567 if (c == '<')
3568 BUF_PUSH (symbeg);
3569 else if (c == '>')
3570 BUF_PUSH (symend);
3571 else
3572 FREE_STACK_RETURN (REG_BADPAT);
3573 break;
3574
3575 case 'b':
3576 if (syntax & RE_NO_GNU_OPS)
3577 goto normal_char;
3578 BUF_PUSH (wordbound);
3579 break;
3580
3581 case 'B':
3582 if (syntax & RE_NO_GNU_OPS)
3583 goto normal_char;
3584 BUF_PUSH (notwordbound);
3585 break;
3586
3587 case '`':
3588 if (syntax & RE_NO_GNU_OPS)
3589 goto normal_char;
3590 BUF_PUSH (begbuf);
3591 break;
3592
3593 case '\'':
3594 if (syntax & RE_NO_GNU_OPS)
3595 goto normal_char;
3596 BUF_PUSH (endbuf);
3597 break;
3598
3599 case '1': case '2': case '3': case '4': case '5':
3600 case '6': case '7': case '8': case '9':
3601 {
3602 regnum_t reg;
3603
3604 if (syntax & RE_NO_BK_REFS)
3605 goto normal_backslash;
3606
3607 reg = c - '0';
3608
3609 /* Can't back reference to a subexpression before its end. */
3610 if (reg > regnum || group_in_compile_stack (compile_stack, reg))
3611 FREE_STACK_RETURN (REG_ESUBREG);
3612
3613 laststart = b;
3614 BUF_PUSH_2 (duplicate, reg);
3615 }
3616 break;
3617
3618
3619 case '+':
3620 case '?':
3621 if (syntax & RE_BK_PLUS_QM)
3622 goto handle_plus;
3623 else
3624 goto normal_backslash;
3625
3626 default:
3627 normal_backslash:
3628 /* You might think it would be useful for \ to mean
3629 not to translate; but if we don't translate it
3630 it will never match anything. */
3631 goto normal_char;
3632 }
3633 break;
3634
3635
3636 default:
3637 /* Expects the character in `c'. */
3638 normal_char:
3639 /* If no exactn currently being built. */
3640 if (!pending_exact
3641
3642 /* If last exactn not at current position. */
3643 || pending_exact + *pending_exact + 1 != b
3644
3645 /* We have only one byte following the exactn for the count. */
3646 || *pending_exact >= (1 << BYTEWIDTH) - MAX_MULTIBYTE_LENGTH
3647
3648 /* If followed by a repetition operator. */
3649 || (p != pend && (*p == '*' || *p == '^'))
3650 || ((syntax & RE_BK_PLUS_QM)
3651 ? p + 1 < pend && *p == '\\' && (p[1] == '+' || p[1] == '?')
3652 : p != pend && (*p == '+' || *p == '?'))
3653 || ((syntax & RE_INTERVALS)
3654 && ((syntax & RE_NO_BK_BRACES)
3655 ? p != pend && *p == '{'
3656 : p + 1 < pend && p[0] == '\\' && p[1] == '{')))
3657 {
3658 /* Start building a new exactn. */
3659
3660 laststart = b;
3661
3662 BUF_PUSH_2 (exactn, 0);
3663 pending_exact = b - 1;
3664 }
3665
3666 GET_BUFFER_SPACE (MAX_MULTIBYTE_LENGTH);
3667 {
3668 int len;
3669
3670 if (! multibyte)
3671 MAKE_CHAR_MULTIBYTE (c);
3672 c = TRANSLATE (c);
3673 if (target_multibyte)
3674 {
3675 len = CHAR_STRING (c, b);
3676 b += len;
3677 }
3678 else
3679 {
3680 MAKE_CHAR_UNIBYTE (c);
3681 *b++ = c;
3682 len = 1;
3683 }
3684 (*pending_exact) += len;
3685 }
3686
3687 break;
3688 } /* switch (c) */
3689 } /* while p != pend */
3690
3691
3692 /* Through the pattern now. */
3693
3694 FIXUP_ALT_JUMP ();
3695
3696 if (!COMPILE_STACK_EMPTY)
3697 FREE_STACK_RETURN (REG_EPAREN);
3698
3699 /* If we don't want backtracking, force success
3700 the first time we reach the end of the compiled pattern. */
3701 if (syntax & RE_NO_POSIX_BACKTRACKING)
3702 BUF_PUSH (succeed);
3703
3704 /* We have succeeded; set the length of the buffer. */
3705 bufp->used = b - bufp->buffer;
3706
3707 #ifdef emacs
3708 /* Now the buffer is adjusted for the multibyteness of a target. */
3709 bufp->multibyte = bufp->target_multibyte;
3710 #endif
3711
3712 #ifdef DEBUG
3713 if (debug > 0)
3714 {
3715 re_compile_fastmap (bufp);
3716 DEBUG_PRINT1 ("\nCompiled pattern: \n");
3717 print_compiled_pattern (bufp);
3718 }
3719 debug--;
3720 #endif /* DEBUG */
3721
3722 #ifndef MATCH_MAY_ALLOCATE
3723 /* Initialize the failure stack to the largest possible stack. This
3724 isn't necessary unless we're trying to avoid calling alloca in
3725 the search and match routines. */
3726 {
3727 int num_regs = bufp->re_nsub + 1;
3728
3729 if (fail_stack.size < re_max_failures * TYPICAL_FAILURE_SIZE)
3730 {
3731 fail_stack.size = re_max_failures * TYPICAL_FAILURE_SIZE;
3732
3733 if (! fail_stack.stack)
3734 fail_stack.stack
3735 = (fail_stack_elt_t *) malloc (fail_stack.size
3736 * sizeof (fail_stack_elt_t));
3737 else
3738 fail_stack.stack
3739 = (fail_stack_elt_t *) realloc (fail_stack.stack,
3740 (fail_stack.size
3741 * sizeof (fail_stack_elt_t)));
3742 }
3743
3744 regex_grow_registers (num_regs);
3745 }
3746 #endif /* not MATCH_MAY_ALLOCATE */
3747
3748 FREE_STACK_RETURN (REG_NOERROR);
3749 } /* regex_compile */
3750 \f
3751 /* Subroutines for `regex_compile'. */
3752
3753 /* Store OP at LOC followed by two-byte integer parameter ARG. */
3754
3755 static void
3756 store_op1 (op, loc, arg)
3757 re_opcode_t op;
3758 unsigned char *loc;
3759 int arg;
3760 {
3761 *loc = (unsigned char) op;
3762 STORE_NUMBER (loc + 1, arg);
3763 }
3764
3765
3766 /* Like `store_op1', but for two two-byte parameters ARG1 and ARG2. */
3767
3768 static void
3769 store_op2 (op, loc, arg1, arg2)
3770 re_opcode_t op;
3771 unsigned char *loc;
3772 int arg1, arg2;
3773 {
3774 *loc = (unsigned char) op;
3775 STORE_NUMBER (loc + 1, arg1);
3776 STORE_NUMBER (loc + 3, arg2);
3777 }
3778
3779
3780 /* Copy the bytes from LOC to END to open up three bytes of space at LOC
3781 for OP followed by two-byte integer parameter ARG. */
3782
3783 static void
3784 insert_op1 (op, loc, arg, end)
3785 re_opcode_t op;
3786 unsigned char *loc;
3787 int arg;
3788 unsigned char *end;
3789 {
3790 register unsigned char *pfrom = end;
3791 register unsigned char *pto = end + 3;
3792
3793 while (pfrom != loc)
3794 *--pto = *--pfrom;
3795
3796 store_op1 (op, loc, arg);
3797 }
3798
3799
3800 /* Like `insert_op1', but for two two-byte parameters ARG1 and ARG2. */
3801
3802 static void
3803 insert_op2 (op, loc, arg1, arg2, end)
3804 re_opcode_t op;
3805 unsigned char *loc;
3806 int arg1, arg2;
3807 unsigned char *end;
3808 {
3809 register unsigned char *pfrom = end;
3810 register unsigned char *pto = end + 5;
3811
3812 while (pfrom != loc)
3813 *--pto = *--pfrom;
3814
3815 store_op2 (op, loc, arg1, arg2);
3816 }
3817
3818
3819 /* P points to just after a ^ in PATTERN. Return true if that ^ comes
3820 after an alternative or a begin-subexpression. We assume there is at
3821 least one character before the ^. */
3822
3823 static boolean
3824 at_begline_loc_p (pattern, p, syntax)
3825 re_char *pattern, *p;
3826 reg_syntax_t syntax;
3827 {
3828 re_char *prev = p - 2;
3829 boolean prev_prev_backslash = prev > pattern && prev[-1] == '\\';
3830
3831 return
3832 /* After a subexpression? */
3833 (*prev == '(' && (syntax & RE_NO_BK_PARENS || prev_prev_backslash))
3834 /* After an alternative? */
3835 || (*prev == '|' && (syntax & RE_NO_BK_VBAR || prev_prev_backslash))
3836 /* After a shy subexpression? */
3837 || ((syntax & RE_SHY_GROUPS) && prev - 2 >= pattern
3838 && prev[-1] == '?' && prev[-2] == '('
3839 && (syntax & RE_NO_BK_PARENS
3840 || (prev - 3 >= pattern && prev[-3] == '\\')));
3841 }
3842
3843
3844 /* The dual of at_begline_loc_p. This one is for $. We assume there is
3845 at least one character after the $, i.e., `P < PEND'. */
3846
3847 static boolean
3848 at_endline_loc_p (p, pend, syntax)
3849 re_char *p, *pend;
3850 reg_syntax_t syntax;
3851 {
3852 re_char *next = p;
3853 boolean next_backslash = *next == '\\';
3854 re_char *next_next = p + 1 < pend ? p + 1 : 0;
3855
3856 return
3857 /* Before a subexpression? */
3858 (syntax & RE_NO_BK_PARENS ? *next == ')'
3859 : next_backslash && next_next && *next_next == ')')
3860 /* Before an alternative? */
3861 || (syntax & RE_NO_BK_VBAR ? *next == '|'
3862 : next_backslash && next_next && *next_next == '|');
3863 }
3864
3865
3866 /* Returns true if REGNUM is in one of COMPILE_STACK's elements and
3867 false if it's not. */
3868
3869 static boolean
3870 group_in_compile_stack (compile_stack, regnum)
3871 compile_stack_type compile_stack;
3872 regnum_t regnum;
3873 {
3874 int this_element;
3875
3876 for (this_element = compile_stack.avail - 1;
3877 this_element >= 0;
3878 this_element--)
3879 if (compile_stack.stack[this_element].regnum == regnum)
3880 return true;
3881
3882 return false;
3883 }
3884 \f
3885 /* analyse_first.
3886 If fastmap is non-NULL, go through the pattern and fill fastmap
3887 with all the possible leading chars. If fastmap is NULL, don't
3888 bother filling it up (obviously) and only return whether the
3889 pattern could potentially match the empty string.
3890
3891 Return 1 if p..pend might match the empty string.
3892 Return 0 if p..pend matches at least one char.
3893 Return -1 if fastmap was not updated accurately. */
3894
3895 static int
3896 analyse_first (p, pend, fastmap, multibyte)
3897 re_char *p, *pend;
3898 char *fastmap;
3899 const int multibyte;
3900 {
3901 int j, k;
3902 boolean not;
3903
3904 /* If all elements for base leading-codes in fastmap is set, this
3905 flag is set true. */
3906 boolean match_any_multibyte_characters = false;
3907
3908 assert (p);
3909
3910 /* The loop below works as follows:
3911 - It has a working-list kept in the PATTERN_STACK and which basically
3912 starts by only containing a pointer to the first operation.
3913 - If the opcode we're looking at is a match against some set of
3914 chars, then we add those chars to the fastmap and go on to the
3915 next work element from the worklist (done via `break').
3916 - If the opcode is a control operator on the other hand, we either
3917 ignore it (if it's meaningless at this point, such as `start_memory')
3918 or execute it (if it's a jump). If the jump has several destinations
3919 (i.e. `on_failure_jump'), then we push the other destination onto the
3920 worklist.
3921 We guarantee termination by ignoring backward jumps (more or less),
3922 so that `p' is monotonically increasing. More to the point, we
3923 never set `p' (or push) anything `<= p1'. */
3924
3925 while (p < pend)
3926 {
3927 /* `p1' is used as a marker of how far back a `on_failure_jump'
3928 can go without being ignored. It is normally equal to `p'
3929 (which prevents any backward `on_failure_jump') except right
3930 after a plain `jump', to allow patterns such as:
3931 0: jump 10
3932 3..9: <body>
3933 10: on_failure_jump 3
3934 as used for the *? operator. */
3935 re_char *p1 = p;
3936
3937 switch (SWITCH_ENUM_CAST ((re_opcode_t) *p++))
3938 {
3939 case succeed:
3940 return 1;
3941 continue;
3942
3943 case duplicate:
3944 /* If the first character has to match a backreference, that means
3945 that the group was empty (since it already matched). Since this
3946 is the only case that interests us here, we can assume that the
3947 backreference must match the empty string. */
3948 p++;
3949 continue;
3950
3951
3952 /* Following are the cases which match a character. These end
3953 with `break'. */
3954
3955 case exactn:
3956 if (fastmap)
3957 /* If multibyte is nonzero, the first byte of each
3958 character is an ASCII or a leading code. Otherwise,
3959 each byte is a character. Thus, this works in both
3960 cases. */
3961 fastmap[p[1]] = 1;
3962 break;
3963
3964
3965 case anychar:
3966 /* We could put all the chars except for \n (and maybe \0)
3967 but we don't bother since it is generally not worth it. */
3968 if (!fastmap) break;
3969 return -1;
3970
3971
3972 case charset_not:
3973 if (!fastmap) break;
3974 {
3975 /* Chars beyond end of bitmap are possible matches. */
3976 /* In a multibyte case, the bitmap is used only for ASCII
3977 characters. */
3978 int limit = multibyte ? 128 : (1 << BYTEWIDTH);
3979
3980 for (j = CHARSET_BITMAP_SIZE (&p[-1]) * BYTEWIDTH;
3981 j < limit; j++)
3982 fastmap[j] = 1;
3983 }
3984
3985 /* Fallthrough */
3986 case charset:
3987 if (!fastmap) break;
3988 not = (re_opcode_t) *(p - 1) == charset_not;
3989 for (j = CHARSET_BITMAP_SIZE (&p[-1]) * BYTEWIDTH - 1, p++;
3990 j >= 0; j--)
3991 if (!!(p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH))) ^ not)
3992 fastmap[j] = 1;
3993
3994 if ((not && multibyte)
3995 /* Any leading code can possibly start a character
3996 which doesn't match the specified set of characters. */
3997 || (CHARSET_RANGE_TABLE_EXISTS_P (&p[-2])
3998 && CHARSET_RANGE_TABLE_BITS (&p[-2]) != 0))
3999 /* If we can match a character class, we can match
4000 any multibyte characters. */
4001 {
4002 if (match_any_multibyte_characters == false)
4003 {
4004 for (j = 0x80; j < (1 << BYTEWIDTH); j++)
4005 fastmap[j] = 1;
4006 match_any_multibyte_characters = true;
4007 }
4008 }
4009
4010 else if (!not && CHARSET_RANGE_TABLE_EXISTS_P (&p[-2])
4011 && match_any_multibyte_characters == false)
4012 {
4013 /* Set fastmap[I] to 1 where I is a leading code of each
4014 multibyte characer in the range table. */
4015 int c, count;
4016 unsigned char lc1, lc2;
4017
4018 /* Make P points the range table. `+ 2' is to skip flag
4019 bits for a character class. */
4020 p += CHARSET_BITMAP_SIZE (&p[-2]) + 2;
4021
4022 /* Extract the number of ranges in range table into COUNT. */
4023 EXTRACT_NUMBER_AND_INCR (count, p);
4024 for (; count > 0; count--, p += 2 * 3) /* XXX */
4025 {
4026 /* Extract the start and end of each range. */
4027 EXTRACT_CHARACTER (c, p);
4028 lc1 = CHAR_LEADING_CODE (c);
4029 p += 3;
4030 EXTRACT_CHARACTER (c, p);
4031 lc2 = CHAR_LEADING_CODE (c);
4032 for (j = lc1; j <= lc2; j++)
4033 fastmap[j] = 1;
4034 }
4035 }
4036 break;
4037
4038 case syntaxspec:
4039 case notsyntaxspec:
4040 if (!fastmap) break;
4041 #ifndef emacs
4042 not = (re_opcode_t)p[-1] == notsyntaxspec;
4043 k = *p++;
4044 for (j = 0; j < (1 << BYTEWIDTH); j++)
4045 if ((SYNTAX (j) == (enum syntaxcode) k) ^ not)
4046 fastmap[j] = 1;
4047 break;
4048 #else /* emacs */
4049 /* This match depends on text properties. These end with
4050 aborting optimizations. */
4051 return -1;
4052
4053 case categoryspec:
4054 case notcategoryspec:
4055 if (!fastmap) break;
4056 not = (re_opcode_t)p[-1] == notcategoryspec;
4057 k = *p++;
4058 for (j = (multibyte ? 127 : (1 << BYTEWIDTH)); j >= 0; j--)
4059 if ((CHAR_HAS_CATEGORY (j, k)) ^ not)
4060 fastmap[j] = 1;
4061
4062 if (multibyte)
4063 {
4064 /* Any character set can possibly contain a character
4065 whose category is K (or not). */
4066 if (match_any_multibyte_characters == false)
4067 {
4068 for (j = 0x80; j < (1 << BYTEWIDTH); j++)
4069 fastmap[j] = 1;
4070 match_any_multibyte_characters = true;
4071 }
4072 }
4073 break;
4074
4075 /* All cases after this match the empty string. These end with
4076 `continue'. */
4077
4078 case before_dot:
4079 case at_dot:
4080 case after_dot:
4081 #endif /* !emacs */
4082 case no_op:
4083 case begline:
4084 case endline:
4085 case begbuf:
4086 case endbuf:
4087 case wordbound:
4088 case notwordbound:
4089 case wordbeg:
4090 case wordend:
4091 case symbeg:
4092 case symend:
4093 continue;
4094
4095
4096 case jump:
4097 EXTRACT_NUMBER_AND_INCR (j, p);
4098 if (j < 0)
4099 /* Backward jumps can only go back to code that we've already
4100 visited. `re_compile' should make sure this is true. */
4101 break;
4102 p += j;
4103 switch (SWITCH_ENUM_CAST ((re_opcode_t) *p))
4104 {
4105 case on_failure_jump:
4106 case on_failure_keep_string_jump:
4107 case on_failure_jump_loop:
4108 case on_failure_jump_nastyloop:
4109 case on_failure_jump_smart:
4110 p++;
4111 break;
4112 default:
4113 continue;
4114 };
4115 /* Keep `p1' to allow the `on_failure_jump' we are jumping to
4116 to jump back to "just after here". */
4117 /* Fallthrough */
4118
4119 case on_failure_jump:
4120 case on_failure_keep_string_jump:
4121 case on_failure_jump_nastyloop:
4122 case on_failure_jump_loop:
4123 case on_failure_jump_smart:
4124 EXTRACT_NUMBER_AND_INCR (j, p);
4125 if (p + j <= p1)
4126 ; /* Backward jump to be ignored. */
4127 else
4128 { /* We have to look down both arms.
4129 We first go down the "straight" path so as to minimize
4130 stack usage when going through alternatives. */
4131 int r = analyse_first (p, pend, fastmap, multibyte);
4132 if (r) return r;
4133 p += j;
4134 }
4135 continue;
4136
4137
4138 case jump_n:
4139 /* This code simply does not properly handle forward jump_n. */
4140 DEBUG_STATEMENT (EXTRACT_NUMBER (j, p); assert (j < 0));
4141 p += 4;
4142 /* jump_n can either jump or fall through. The (backward) jump
4143 case has already been handled, so we only need to look at the
4144 fallthrough case. */
4145 continue;
4146
4147 case succeed_n:
4148 /* If N == 0, it should be an on_failure_jump_loop instead. */
4149 DEBUG_STATEMENT (EXTRACT_NUMBER (j, p + 2); assert (j > 0));
4150 p += 4;
4151 /* We only care about one iteration of the loop, so we don't
4152 need to consider the case where this behaves like an
4153 on_failure_jump. */
4154 continue;
4155
4156
4157 case set_number_at:
4158 p += 4;
4159 continue;
4160
4161
4162 case start_memory:
4163 case stop_memory:
4164 p += 1;
4165 continue;
4166
4167
4168 default:
4169 abort (); /* We have listed all the cases. */
4170 } /* switch *p++ */
4171
4172 /* Getting here means we have found the possible starting
4173 characters for one path of the pattern -- and that the empty
4174 string does not match. We need not follow this path further. */
4175 return 0;
4176 } /* while p */
4177
4178 /* We reached the end without matching anything. */
4179 return 1;
4180
4181 } /* analyse_first */
4182 \f
4183 /* re_compile_fastmap computes a ``fastmap'' for the compiled pattern in
4184 BUFP. A fastmap records which of the (1 << BYTEWIDTH) possible
4185 characters can start a string that matches the pattern. This fastmap
4186 is used by re_search to skip quickly over impossible starting points.
4187
4188 Character codes above (1 << BYTEWIDTH) are not represented in the
4189 fastmap, but the leading codes are represented. Thus, the fastmap
4190 indicates which character sets could start a match.
4191
4192 The caller must supply the address of a (1 << BYTEWIDTH)-byte data
4193 area as BUFP->fastmap.
4194
4195 We set the `fastmap', `fastmap_accurate', and `can_be_null' fields in
4196 the pattern buffer.
4197
4198 Returns 0 if we succeed, -2 if an internal error. */
4199
4200 int
4201 re_compile_fastmap (bufp)
4202 struct re_pattern_buffer *bufp;
4203 {
4204 char *fastmap = bufp->fastmap;
4205 int analysis;
4206
4207 assert (fastmap && bufp->buffer);
4208
4209 bzero (fastmap, 1 << BYTEWIDTH); /* Assume nothing's valid. */
4210 bufp->fastmap_accurate = 1; /* It will be when we're done. */
4211
4212 analysis = analyse_first (bufp->buffer, bufp->buffer + bufp->used,
4213 fastmap, RE_MULTIBYTE_P (bufp));
4214 bufp->can_be_null = (analysis != 0);
4215 return 0;
4216 } /* re_compile_fastmap */
4217 \f
4218 /* Set REGS to hold NUM_REGS registers, storing them in STARTS and
4219 ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use
4220 this memory for recording register information. STARTS and ENDS
4221 must be allocated using the malloc library routine, and must each
4222 be at least NUM_REGS * sizeof (regoff_t) bytes long.
4223
4224 If NUM_REGS == 0, then subsequent matches should allocate their own
4225 register data.
4226
4227 Unless this function is called, the first search or match using
4228 PATTERN_BUFFER will allocate its own register data, without
4229 freeing the old data. */
4230
4231 void
4232 re_set_registers (bufp, regs, num_regs, starts, ends)
4233 struct re_pattern_buffer *bufp;
4234 struct re_registers *regs;
4235 unsigned num_regs;
4236 regoff_t *starts, *ends;
4237 {
4238 if (num_regs)
4239 {
4240 bufp->regs_allocated = REGS_REALLOCATE;
4241 regs->num_regs = num_regs;
4242 regs->start = starts;
4243 regs->end = ends;
4244 }
4245 else
4246 {
4247 bufp->regs_allocated = REGS_UNALLOCATED;
4248 regs->num_regs = 0;
4249 regs->start = regs->end = (regoff_t *) 0;
4250 }
4251 }
4252 WEAK_ALIAS (__re_set_registers, re_set_registers)
4253 \f
4254 /* Searching routines. */
4255
4256 /* Like re_search_2, below, but only one string is specified, and
4257 doesn't let you say where to stop matching. */
4258
4259 int
4260 re_search (bufp, string, size, startpos, range, regs)
4261 struct re_pattern_buffer *bufp;
4262 const char *string;
4263 int size, startpos, range;
4264 struct re_registers *regs;
4265 {
4266 return re_search_2 (bufp, NULL, 0, string, size, startpos, range,
4267 regs, size);
4268 }
4269 WEAK_ALIAS (__re_search, re_search)
4270
4271 /* Head address of virtual concatenation of string. */
4272 #define HEAD_ADDR_VSTRING(P) \
4273 (((P) >= size1 ? string2 : string1))
4274
4275 /* End address of virtual concatenation of string. */
4276 #define STOP_ADDR_VSTRING(P) \
4277 (((P) >= size1 ? string2 + size2 : string1 + size1))
4278
4279 /* Address of POS in the concatenation of virtual string. */
4280 #define POS_ADDR_VSTRING(POS) \
4281 (((POS) >= size1 ? string2 - size1 : string1) + (POS))
4282
4283 /* Using the compiled pattern in BUFP->buffer, first tries to match the
4284 virtual concatenation of STRING1 and STRING2, starting first at index
4285 STARTPOS, then at STARTPOS + 1, and so on.
4286
4287 STRING1 and STRING2 have length SIZE1 and SIZE2, respectively.
4288
4289 RANGE is how far to scan while trying to match. RANGE = 0 means try
4290 only at STARTPOS; in general, the last start tried is STARTPOS +
4291 RANGE.
4292
4293 In REGS, return the indices of the virtual concatenation of STRING1
4294 and STRING2 that matched the entire BUFP->buffer and its contained
4295 subexpressions.
4296
4297 Do not consider matching one past the index STOP in the virtual
4298 concatenation of STRING1 and STRING2.
4299
4300 We return either the position in the strings at which the match was
4301 found, -1 if no match, or -2 if error (such as failure
4302 stack overflow). */
4303
4304 int
4305 re_search_2 (bufp, str1, size1, str2, size2, startpos, range, regs, stop)
4306 struct re_pattern_buffer *bufp;
4307 const char *str1, *str2;
4308 int size1, size2;
4309 int startpos;
4310 int range;
4311 struct re_registers *regs;
4312 int stop;
4313 {
4314 int val;
4315 re_char *string1 = (re_char*) str1;
4316 re_char *string2 = (re_char*) str2;
4317 register char *fastmap = bufp->fastmap;
4318 register RE_TRANSLATE_TYPE translate = bufp->translate;
4319 int total_size = size1 + size2;
4320 int endpos = startpos + range;
4321 boolean anchored_start;
4322 /* Nonzero if BUFP is setup for multibyte characters. We are sure
4323 that it is the same as RE_TARGET_MULTIBYTE_P (bufp). */
4324 const boolean multibyte = RE_MULTIBYTE_P (bufp);
4325
4326 /* Check for out-of-range STARTPOS. */
4327 if (startpos < 0 || startpos > total_size)
4328 return -1;
4329
4330 /* Fix up RANGE if it might eventually take us outside
4331 the virtual concatenation of STRING1 and STRING2.
4332 Make sure we won't move STARTPOS below 0 or above TOTAL_SIZE. */
4333 if (endpos < 0)
4334 range = 0 - startpos;
4335 else if (endpos > total_size)
4336 range = total_size - startpos;
4337
4338 /* If the search isn't to be a backwards one, don't waste time in a
4339 search for a pattern anchored at beginning of buffer. */
4340 if (bufp->used > 0 && (re_opcode_t) bufp->buffer[0] == begbuf && range > 0)
4341 {
4342 if (startpos > 0)
4343 return -1;
4344 else
4345 range = 0;
4346 }
4347
4348 #ifdef emacs
4349 /* In a forward search for something that starts with \=.
4350 don't keep searching past point. */
4351 if (bufp->used > 0 && (re_opcode_t) bufp->buffer[0] == at_dot && range > 0)
4352 {
4353 range = PT_BYTE - BEGV_BYTE - startpos;
4354 if (range < 0)
4355 return -1;
4356 }
4357 #endif /* emacs */
4358
4359 /* Update the fastmap now if not correct already. */
4360 if (fastmap && !bufp->fastmap_accurate)
4361 re_compile_fastmap (bufp);
4362
4363 /* See whether the pattern is anchored. */
4364 anchored_start = (bufp->buffer[0] == begline);
4365
4366 #ifdef emacs
4367 gl_state.object = re_match_object;
4368 {
4369 int charpos = SYNTAX_TABLE_BYTE_TO_CHAR (POS_AS_IN_BUFFER (startpos));
4370
4371 SETUP_SYNTAX_TABLE_FOR_OBJECT (re_match_object, charpos, 1);
4372 }
4373 #endif
4374
4375 /* Loop through the string, looking for a place to start matching. */
4376 for (;;)
4377 {
4378 /* If the pattern is anchored,
4379 skip quickly past places we cannot match.
4380 We don't bother to treat startpos == 0 specially
4381 because that case doesn't repeat. */
4382 if (anchored_start && startpos > 0)
4383 {
4384 if (! ((startpos <= size1 ? string1[startpos - 1]
4385 : string2[startpos - size1 - 1])
4386 == '\n'))
4387 goto advance;
4388 }
4389
4390 /* If a fastmap is supplied, skip quickly over characters that
4391 cannot be the start of a match. If the pattern can match the
4392 null string, however, we don't need to skip characters; we want
4393 the first null string. */
4394 if (fastmap && startpos < total_size && !bufp->can_be_null)
4395 {
4396 register re_char *d;
4397 register re_wchar_t buf_ch;
4398
4399 d = POS_ADDR_VSTRING (startpos);
4400
4401 if (range > 0) /* Searching forwards. */
4402 {
4403 register int lim = 0;
4404 int irange = range;
4405
4406 if (startpos < size1 && startpos + range >= size1)
4407 lim = range - (size1 - startpos);
4408
4409 /* Written out as an if-else to avoid testing `translate'
4410 inside the loop. */
4411 if (RE_TRANSLATE_P (translate))
4412 {
4413 if (multibyte)
4414 while (range > lim)
4415 {
4416 int buf_charlen;
4417
4418 buf_ch = STRING_CHAR_AND_LENGTH (d, range - lim,
4419 buf_charlen);
4420 buf_ch = RE_TRANSLATE (translate, buf_ch);
4421 if (fastmap[CHAR_LEADING_CODE (buf_ch)])
4422 break;
4423
4424 range -= buf_charlen;
4425 d += buf_charlen;
4426 }
4427 else
4428 while (range > lim)
4429 {
4430 buf_ch = *d;
4431 MAKE_CHAR_MULTIBYTE (buf_ch);
4432 buf_ch = RE_TRANSLATE (translate, buf_ch);
4433 MAKE_CHAR_UNIBYTE (buf_ch);
4434 if (fastmap[buf_ch])
4435 break;
4436 d++;
4437 range--;
4438 }
4439 }
4440 else
4441 {
4442 if (multibyte)
4443 while (range > lim)
4444 {
4445 int buf_charlen;
4446
4447 buf_ch = STRING_CHAR_AND_LENGTH (d, range - lim,
4448 buf_charlen);
4449 if (fastmap[CHAR_LEADING_CODE (buf_ch)])
4450 break;
4451 range -= buf_charlen;
4452 d += buf_charlen;
4453 }
4454 else
4455 while (range > lim && !fastmap[*d])
4456 {
4457 d++;
4458 range--;
4459 }
4460 }
4461 startpos += irange - range;
4462 }
4463 else /* Searching backwards. */
4464 {
4465 int room = (startpos >= size1
4466 ? size2 + size1 - startpos
4467 : size1 - startpos);
4468 if (multibyte)
4469 {
4470 buf_ch = STRING_CHAR (d, room);
4471 buf_ch = TRANSLATE (buf_ch);
4472 if (! fastmap[CHAR_LEADING_CODE (buf_ch)])
4473 goto advance;
4474 }
4475 else
4476 {
4477 if (! fastmap[TRANSLATE (*d)])
4478 goto advance;
4479 }
4480 }
4481 }
4482
4483 /* If can't match the null string, and that's all we have left, fail. */
4484 if (range >= 0 && startpos == total_size && fastmap
4485 && !bufp->can_be_null)
4486 return -1;
4487
4488 val = re_match_2_internal (bufp, string1, size1, string2, size2,
4489 startpos, regs, stop);
4490 #ifndef REGEX_MALLOC
4491 # ifdef C_ALLOCA
4492 alloca (0);
4493 # endif
4494 #endif
4495
4496 if (val >= 0)
4497 return startpos;
4498
4499 if (val == -2)
4500 return -2;
4501
4502 advance:
4503 if (!range)
4504 break;
4505 else if (range > 0)
4506 {
4507 /* Update STARTPOS to the next character boundary. */
4508 if (multibyte)
4509 {
4510 re_char *p = POS_ADDR_VSTRING (startpos);
4511 re_char *pend = STOP_ADDR_VSTRING (startpos);
4512 int len = MULTIBYTE_FORM_LENGTH (p, pend - p);
4513
4514 range -= len;
4515 if (range < 0)
4516 break;
4517 startpos += len;
4518 }
4519 else
4520 {
4521 range--;
4522 startpos++;
4523 }
4524 }
4525 else
4526 {
4527 range++;
4528 startpos--;
4529
4530 /* Update STARTPOS to the previous character boundary. */
4531 if (multibyte)
4532 {
4533 re_char *p = POS_ADDR_VSTRING (startpos) + 1;
4534 re_char *p0 = p;
4535 re_char *phead = HEAD_ADDR_VSTRING (startpos);
4536
4537 /* Find the head of multibyte form. */
4538 PREV_CHAR_BOUNDARY (p, phead);
4539 range += p0 - 1 - p;
4540 if (range > 0)
4541 break;
4542
4543 startpos -= p0 - 1 - p;
4544 }
4545 }
4546 }
4547 return -1;
4548 } /* re_search_2 */
4549 WEAK_ALIAS (__re_search_2, re_search_2)
4550 \f
4551 /* Declarations and macros for re_match_2. */
4552
4553 static int bcmp_translate _RE_ARGS((re_char *s1, re_char *s2,
4554 register int len,
4555 RE_TRANSLATE_TYPE translate,
4556 const int multibyte));
4557
4558 /* This converts PTR, a pointer into one of the search strings `string1'
4559 and `string2' into an offset from the beginning of that string. */
4560 #define POINTER_TO_OFFSET(ptr) \
4561 (FIRST_STRING_P (ptr) \
4562 ? ((regoff_t) ((ptr) - string1)) \
4563 : ((regoff_t) ((ptr) - string2 + size1)))
4564
4565 /* Call before fetching a character with *d. This switches over to
4566 string2 if necessary.
4567 Check re_match_2_internal for a discussion of why end_match_2 might
4568 not be within string2 (but be equal to end_match_1 instead). */
4569 #define PREFETCH() \
4570 while (d == dend) \
4571 { \
4572 /* End of string2 => fail. */ \
4573 if (dend == end_match_2) \
4574 goto fail; \
4575 /* End of string1 => advance to string2. */ \
4576 d = string2; \
4577 dend = end_match_2; \
4578 }
4579
4580 /* Call before fetching a char with *d if you already checked other limits.
4581 This is meant for use in lookahead operations like wordend, etc..
4582 where we might need to look at parts of the string that might be
4583 outside of the LIMITs (i.e past `stop'). */
4584 #define PREFETCH_NOLIMIT() \
4585 if (d == end1) \
4586 { \
4587 d = string2; \
4588 dend = end_match_2; \
4589 } \
4590
4591 /* Test if at very beginning or at very end of the virtual concatenation
4592 of `string1' and `string2'. If only one string, it's `string2'. */
4593 #define AT_STRINGS_BEG(d) ((d) == (size1 ? string1 : string2) || !size2)
4594 #define AT_STRINGS_END(d) ((d) == end2)
4595
4596
4597 /* Test if D points to a character which is word-constituent. We have
4598 two special cases to check for: if past the end of string1, look at
4599 the first character in string2; and if before the beginning of
4600 string2, look at the last character in string1. */
4601 #define WORDCHAR_P(d) \
4602 (SYNTAX ((d) == end1 ? *string2 \
4603 : (d) == string2 - 1 ? *(end1 - 1) : *(d)) \
4604 == Sword)
4605
4606 /* Disabled due to a compiler bug -- see comment at case wordbound */
4607
4608 /* The comment at case wordbound is following one, but we don't use
4609 AT_WORD_BOUNDARY anymore to support multibyte form.
4610
4611 The DEC Alpha C compiler 3.x generates incorrect code for the
4612 test WORDCHAR_P (d - 1) != WORDCHAR_P (d) in the expansion of
4613 AT_WORD_BOUNDARY, so this code is disabled. Expanding the
4614 macro and introducing temporary variables works around the bug. */
4615
4616 #if 0
4617 /* Test if the character before D and the one at D differ with respect
4618 to being word-constituent. */
4619 #define AT_WORD_BOUNDARY(d) \
4620 (AT_STRINGS_BEG (d) || AT_STRINGS_END (d) \
4621 || WORDCHAR_P (d - 1) != WORDCHAR_P (d))
4622 #endif
4623
4624 /* Free everything we malloc. */
4625 #ifdef MATCH_MAY_ALLOCATE
4626 # define FREE_VAR(var) if (var) { REGEX_FREE (var); var = NULL; } else
4627 # define FREE_VARIABLES() \
4628 do { \
4629 REGEX_FREE_STACK (fail_stack.stack); \
4630 FREE_VAR (regstart); \
4631 FREE_VAR (regend); \
4632 FREE_VAR (best_regstart); \
4633 FREE_VAR (best_regend); \
4634 } while (0)
4635 #else
4636 # define FREE_VARIABLES() ((void)0) /* Do nothing! But inhibit gcc warning. */
4637 #endif /* not MATCH_MAY_ALLOCATE */
4638
4639 \f
4640 /* Optimization routines. */
4641
4642 /* If the operation is a match against one or more chars,
4643 return a pointer to the next operation, else return NULL. */
4644 static re_char *
4645 skip_one_char (p)
4646 re_char *p;
4647 {
4648 switch (SWITCH_ENUM_CAST (*p++))
4649 {
4650 case anychar:
4651 break;
4652
4653 case exactn:
4654 p += *p + 1;
4655 break;
4656
4657 case charset_not:
4658 case charset:
4659 if (CHARSET_RANGE_TABLE_EXISTS_P (p - 1))
4660 {
4661 int mcnt;
4662 p = CHARSET_RANGE_TABLE (p - 1);
4663 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4664 p = CHARSET_RANGE_TABLE_END (p, mcnt);
4665 }
4666 else
4667 p += 1 + CHARSET_BITMAP_SIZE (p - 1);
4668 break;
4669
4670 case syntaxspec:
4671 case notsyntaxspec:
4672 #ifdef emacs
4673 case categoryspec:
4674 case notcategoryspec:
4675 #endif /* emacs */
4676 p++;
4677 break;
4678
4679 default:
4680 p = NULL;
4681 }
4682 return p;
4683 }
4684
4685
4686 /* Jump over non-matching operations. */
4687 static re_char *
4688 skip_noops (p, pend)
4689 re_char *p, *pend;
4690 {
4691 int mcnt;
4692 while (p < pend)
4693 {
4694 switch (SWITCH_ENUM_CAST ((re_opcode_t) *p))
4695 {
4696 case start_memory:
4697 case stop_memory:
4698 p += 2; break;
4699 case no_op:
4700 p += 1; break;
4701 case jump:
4702 p += 1;
4703 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4704 p += mcnt;
4705 break;
4706 default:
4707 return p;
4708 }
4709 }
4710 assert (p == pend);
4711 return p;
4712 }
4713
4714 /* Non-zero if "p1 matches something" implies "p2 fails". */
4715 static int
4716 mutually_exclusive_p (bufp, p1, p2)
4717 struct re_pattern_buffer *bufp;
4718 re_char *p1, *p2;
4719 {
4720 re_opcode_t op2;
4721 const boolean multibyte = RE_MULTIBYTE_P (bufp);
4722 unsigned char *pend = bufp->buffer + bufp->used;
4723
4724 assert (p1 >= bufp->buffer && p1 < pend
4725 && p2 >= bufp->buffer && p2 <= pend);
4726
4727 /* Skip over open/close-group commands.
4728 If what follows this loop is a ...+ construct,
4729 look at what begins its body, since we will have to
4730 match at least one of that. */
4731 p2 = skip_noops (p2, pend);
4732 /* The same skip can be done for p1, except that this function
4733 is only used in the case where p1 is a simple match operator. */
4734 /* p1 = skip_noops (p1, pend); */
4735
4736 assert (p1 >= bufp->buffer && p1 < pend
4737 && p2 >= bufp->buffer && p2 <= pend);
4738
4739 op2 = p2 == pend ? succeed : *p2;
4740
4741 switch (SWITCH_ENUM_CAST (op2))
4742 {
4743 case succeed:
4744 case endbuf:
4745 /* If we're at the end of the pattern, we can change. */
4746 if (skip_one_char (p1))
4747 {
4748 DEBUG_PRINT1 (" End of pattern: fast loop.\n");
4749 return 1;
4750 }
4751 break;
4752
4753 case endline:
4754 case exactn:
4755 {
4756 register re_wchar_t c
4757 = (re_opcode_t) *p2 == endline ? '\n'
4758 : RE_STRING_CHAR (p2 + 2, pend - p2 - 2);
4759
4760 if ((re_opcode_t) *p1 == exactn)
4761 {
4762 if (c != RE_STRING_CHAR (p1 + 2, pend - p1 - 2))
4763 {
4764 DEBUG_PRINT3 (" '%c' != '%c' => fast loop.\n", c, p1[2]);
4765 return 1;
4766 }
4767 }
4768
4769 else if ((re_opcode_t) *p1 == charset
4770 || (re_opcode_t) *p1 == charset_not)
4771 {
4772 int not = (re_opcode_t) *p1 == charset_not;
4773
4774 /* Test if C is listed in charset (or charset_not)
4775 at `p1'. */
4776 if (! multibyte || IS_REAL_ASCII (c))
4777 {
4778 if (c < CHARSET_BITMAP_SIZE (p1) * BYTEWIDTH
4779 && p1[2 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH)))
4780 not = !not;
4781 }
4782 else if (CHARSET_RANGE_TABLE_EXISTS_P (p1))
4783 CHARSET_LOOKUP_RANGE_TABLE (not, c, p1);
4784
4785 /* `not' is equal to 1 if c would match, which means
4786 that we can't change to pop_failure_jump. */
4787 if (!not)
4788 {
4789 DEBUG_PRINT1 (" No match => fast loop.\n");
4790 return 1;
4791 }
4792 }
4793 else if ((re_opcode_t) *p1 == anychar
4794 && c == '\n')
4795 {
4796 DEBUG_PRINT1 (" . != \\n => fast loop.\n");
4797 return 1;
4798 }
4799 }
4800 break;
4801
4802 case charset:
4803 {
4804 if ((re_opcode_t) *p1 == exactn)
4805 /* Reuse the code above. */
4806 return mutually_exclusive_p (bufp, p2, p1);
4807
4808 /* It is hard to list up all the character in charset
4809 P2 if it includes multibyte character. Give up in
4810 such case. */
4811 else if (!multibyte || !CHARSET_RANGE_TABLE_EXISTS_P (p2))
4812 {
4813 /* Now, we are sure that P2 has no range table.
4814 So, for the size of bitmap in P2, `p2[1]' is
4815 enough. But P1 may have range table, so the
4816 size of bitmap table of P1 is extracted by
4817 using macro `CHARSET_BITMAP_SIZE'.
4818
4819 In a multibyte case, we know that all the character
4820 listed in P2 is ASCII. In a unibyte case, P1 has only a
4821 bitmap table. So, in both cases, it is enough to test
4822 only the bitmap table of P1. */
4823
4824 if ((re_opcode_t) *p1 == charset)
4825 {
4826 int idx;
4827 /* We win if the charset inside the loop
4828 has no overlap with the one after the loop. */
4829 for (idx = 0;
4830 (idx < (int) p2[1]
4831 && idx < CHARSET_BITMAP_SIZE (p1));
4832 idx++)
4833 if ((p2[2 + idx] & p1[2 + idx]) != 0)
4834 break;
4835
4836 if (idx == p2[1]
4837 || idx == CHARSET_BITMAP_SIZE (p1))
4838 {
4839 DEBUG_PRINT1 (" No match => fast loop.\n");
4840 return 1;
4841 }
4842 }
4843 else if ((re_opcode_t) *p1 == charset_not)
4844 {
4845 int idx;
4846 /* We win if the charset_not inside the loop lists
4847 every character listed in the charset after. */
4848 for (idx = 0; idx < (int) p2[1]; idx++)
4849 if (! (p2[2 + idx] == 0
4850 || (idx < CHARSET_BITMAP_SIZE (p1)
4851 && ((p2[2 + idx] & ~ p1[2 + idx]) == 0))))
4852 break;
4853
4854 if (idx == p2[1])
4855 {
4856 DEBUG_PRINT1 (" No match => fast loop.\n");
4857 return 1;
4858 }
4859 }
4860 }
4861 }
4862 break;
4863
4864 case charset_not:
4865 switch (SWITCH_ENUM_CAST (*p1))
4866 {
4867 case exactn:
4868 case charset:
4869 /* Reuse the code above. */
4870 return mutually_exclusive_p (bufp, p2, p1);
4871 case charset_not:
4872 /* When we have two charset_not, it's very unlikely that
4873 they don't overlap. The union of the two sets of excluded
4874 chars should cover all possible chars, which, as a matter of
4875 fact, is virtually impossible in multibyte buffers. */
4876 break;
4877 }
4878 break;
4879
4880 case wordend:
4881 return ((re_opcode_t) *p1 == syntaxspec && p1[1] == Sword);
4882 case symend:
4883 return ((re_opcode_t) *p1 == syntaxspec
4884 && (p1[1] == Ssymbol || p1[1] == Sword));
4885 case notsyntaxspec:
4886 return ((re_opcode_t) *p1 == syntaxspec && p1[1] == p2[1]);
4887
4888 case wordbeg:
4889 return ((re_opcode_t) *p1 == notsyntaxspec && p1[1] == Sword);
4890 case symbeg:
4891 return ((re_opcode_t) *p1 == notsyntaxspec
4892 && (p1[1] == Ssymbol || p1[1] == Sword));
4893 case syntaxspec:
4894 return ((re_opcode_t) *p1 == notsyntaxspec && p1[1] == p2[1]);
4895
4896 case wordbound:
4897 return (((re_opcode_t) *p1 == notsyntaxspec
4898 || (re_opcode_t) *p1 == syntaxspec)
4899 && p1[1] == Sword);
4900
4901 #ifdef emacs
4902 case categoryspec:
4903 return ((re_opcode_t) *p1 == notcategoryspec && p1[1] == p2[1]);
4904 case notcategoryspec:
4905 return ((re_opcode_t) *p1 == categoryspec && p1[1] == p2[1]);
4906 #endif /* emacs */
4907
4908 default:
4909 ;
4910 }
4911
4912 /* Safe default. */
4913 return 0;
4914 }
4915
4916 \f
4917 /* Matching routines. */
4918
4919 #ifndef emacs /* Emacs never uses this. */
4920 /* re_match is like re_match_2 except it takes only a single string. */
4921
4922 int
4923 re_match (bufp, string, size, pos, regs)
4924 struct re_pattern_buffer *bufp;
4925 const char *string;
4926 int size, pos;
4927 struct re_registers *regs;
4928 {
4929 int result = re_match_2_internal (bufp, NULL, 0, (re_char*) string, size,
4930 pos, regs, size);
4931 # if defined C_ALLOCA && !defined REGEX_MALLOC
4932 alloca (0);
4933 # endif
4934 return result;
4935 }
4936 WEAK_ALIAS (__re_match, re_match)
4937 #endif /* not emacs */
4938
4939 #ifdef emacs
4940 /* In Emacs, this is the string or buffer in which we
4941 are matching. It is used for looking up syntax properties. */
4942 Lisp_Object re_match_object;
4943 #endif
4944
4945 /* re_match_2 matches the compiled pattern in BUFP against the
4946 the (virtual) concatenation of STRING1 and STRING2 (of length SIZE1
4947 and SIZE2, respectively). We start matching at POS, and stop
4948 matching at STOP.
4949
4950 If REGS is non-null and the `no_sub' field of BUFP is nonzero, we
4951 store offsets for the substring each group matched in REGS. See the
4952 documentation for exactly how many groups we fill.
4953
4954 We return -1 if no match, -2 if an internal error (such as the
4955 failure stack overflowing). Otherwise, we return the length of the
4956 matched substring. */
4957
4958 int
4959 re_match_2 (bufp, string1, size1, string2, size2, pos, regs, stop)
4960 struct re_pattern_buffer *bufp;
4961 const char *string1, *string2;
4962 int size1, size2;
4963 int pos;
4964 struct re_registers *regs;
4965 int stop;
4966 {
4967 int result;
4968
4969 #ifdef emacs
4970 int charpos;
4971 gl_state.object = re_match_object;
4972 charpos = SYNTAX_TABLE_BYTE_TO_CHAR (POS_AS_IN_BUFFER (pos));
4973 SETUP_SYNTAX_TABLE_FOR_OBJECT (re_match_object, charpos, 1);
4974 #endif
4975
4976 result = re_match_2_internal (bufp, (re_char*) string1, size1,
4977 (re_char*) string2, size2,
4978 pos, regs, stop);
4979 #if defined C_ALLOCA && !defined REGEX_MALLOC
4980 alloca (0);
4981 #endif
4982 return result;
4983 }
4984 WEAK_ALIAS (__re_match_2, re_match_2)
4985
4986 #ifdef emacs
4987 #define TRANSLATE_VIA_MULTIBYTE(c) \
4988 do { \
4989 if (multibyte) \
4990 (c) = TRANSLATE (c); \
4991 else \
4992 { \
4993 MAKE_CHAR_MULTIBYTE (c); \
4994 (c) = TRANSLATE (c); \
4995 MAKE_CHAR_UNIBYTE (c); \
4996 } \
4997 } while (0)
4998
4999 #else
5000 #define TRANSLATE_VIA_MULTIBYTE(c) ((c) = TRANSLATE (c))
5001 #endif
5002
5003
5004 /* This is a separate function so that we can force an alloca cleanup
5005 afterwards. */
5006 static int
5007 re_match_2_internal (bufp, string1, size1, string2, size2, pos, regs, stop)
5008 struct re_pattern_buffer *bufp;
5009 re_char *string1, *string2;
5010 int size1, size2;
5011 int pos;
5012 struct re_registers *regs;
5013 int stop;
5014 {
5015 /* General temporaries. */
5016 int mcnt;
5017 size_t reg;
5018 boolean not;
5019
5020 /* Just past the end of the corresponding string. */
5021 re_char *end1, *end2;
5022
5023 /* Pointers into string1 and string2, just past the last characters in
5024 each to consider matching. */
5025 re_char *end_match_1, *end_match_2;
5026
5027 /* Where we are in the data, and the end of the current string. */
5028 re_char *d, *dend;
5029
5030 /* Used sometimes to remember where we were before starting matching
5031 an operator so that we can go back in case of failure. This "atomic"
5032 behavior of matching opcodes is indispensable to the correctness
5033 of the on_failure_keep_string_jump optimization. */
5034 re_char *dfail;
5035
5036 /* Where we are in the pattern, and the end of the pattern. */
5037 re_char *p = bufp->buffer;
5038 re_char *pend = p + bufp->used;
5039
5040 /* We use this to map every character in the string. */
5041 RE_TRANSLATE_TYPE translate = bufp->translate;
5042
5043 /* Nonzero if BUFP is setup for multibyte characters. We are sure
5044 that it is the same as RE_TARGET_MULTIBYTE_P (bufp). */
5045 const boolean multibyte = RE_MULTIBYTE_P (bufp);
5046
5047 /* Failure point stack. Each place that can handle a failure further
5048 down the line pushes a failure point on this stack. It consists of
5049 regstart, and regend for all registers corresponding to
5050 the subexpressions we're currently inside, plus the number of such
5051 registers, and, finally, two char *'s. The first char * is where
5052 to resume scanning the pattern; the second one is where to resume
5053 scanning the strings. */
5054 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
5055 fail_stack_type fail_stack;
5056 #endif
5057 #ifdef DEBUG
5058 unsigned nfailure_points_pushed = 0, nfailure_points_popped = 0;
5059 #endif
5060
5061 #if defined REL_ALLOC && defined REGEX_MALLOC
5062 /* This holds the pointer to the failure stack, when
5063 it is allocated relocatably. */
5064 fail_stack_elt_t *failure_stack_ptr;
5065 #endif
5066
5067 /* We fill all the registers internally, independent of what we
5068 return, for use in backreferences. The number here includes
5069 an element for register zero. */
5070 size_t num_regs = bufp->re_nsub + 1;
5071
5072 /* Information on the contents of registers. These are pointers into
5073 the input strings; they record just what was matched (on this
5074 attempt) by a subexpression part of the pattern, that is, the
5075 regnum-th regstart pointer points to where in the pattern we began
5076 matching and the regnum-th regend points to right after where we
5077 stopped matching the regnum-th subexpression. (The zeroth register
5078 keeps track of what the whole pattern matches.) */
5079 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
5080 re_char **regstart, **regend;
5081 #endif
5082
5083 /* The following record the register info as found in the above
5084 variables when we find a match better than any we've seen before.
5085 This happens as we backtrack through the failure points, which in
5086 turn happens only if we have not yet matched the entire string. */
5087 unsigned best_regs_set = false;
5088 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
5089 re_char **best_regstart, **best_regend;
5090 #endif
5091
5092 /* Logically, this is `best_regend[0]'. But we don't want to have to
5093 allocate space for that if we're not allocating space for anything
5094 else (see below). Also, we never need info about register 0 for
5095 any of the other register vectors, and it seems rather a kludge to
5096 treat `best_regend' differently than the rest. So we keep track of
5097 the end of the best match so far in a separate variable. We
5098 initialize this to NULL so that when we backtrack the first time
5099 and need to test it, it's not garbage. */
5100 re_char *match_end = NULL;
5101
5102 #ifdef DEBUG
5103 /* Counts the total number of registers pushed. */
5104 unsigned num_regs_pushed = 0;
5105 #endif
5106
5107 DEBUG_PRINT1 ("\n\nEntering re_match_2.\n");
5108
5109 INIT_FAIL_STACK ();
5110
5111 #ifdef MATCH_MAY_ALLOCATE
5112 /* Do not bother to initialize all the register variables if there are
5113 no groups in the pattern, as it takes a fair amount of time. If
5114 there are groups, we include space for register 0 (the whole
5115 pattern), even though we never use it, since it simplifies the
5116 array indexing. We should fix this. */
5117 if (bufp->re_nsub)
5118 {
5119 regstart = REGEX_TALLOC (num_regs, re_char *);
5120 regend = REGEX_TALLOC (num_regs, re_char *);
5121 best_regstart = REGEX_TALLOC (num_regs, re_char *);
5122 best_regend = REGEX_TALLOC (num_regs, re_char *);
5123
5124 if (!(regstart && regend && best_regstart && best_regend))
5125 {
5126 FREE_VARIABLES ();
5127 return -2;
5128 }
5129 }
5130 else
5131 {
5132 /* We must initialize all our variables to NULL, so that
5133 `FREE_VARIABLES' doesn't try to free them. */
5134 regstart = regend = best_regstart = best_regend = NULL;
5135 }
5136 #endif /* MATCH_MAY_ALLOCATE */
5137
5138 /* The starting position is bogus. */
5139 if (pos < 0 || pos > size1 + size2)
5140 {
5141 FREE_VARIABLES ();
5142 return -1;
5143 }
5144
5145 /* Initialize subexpression text positions to -1 to mark ones that no
5146 start_memory/stop_memory has been seen for. Also initialize the
5147 register information struct. */
5148 for (reg = 1; reg < num_regs; reg++)
5149 regstart[reg] = regend[reg] = NULL;
5150
5151 /* We move `string1' into `string2' if the latter's empty -- but not if
5152 `string1' is null. */
5153 if (size2 == 0 && string1 != NULL)
5154 {
5155 string2 = string1;
5156 size2 = size1;
5157 string1 = 0;
5158 size1 = 0;
5159 }
5160 end1 = string1 + size1;
5161 end2 = string2 + size2;
5162
5163 /* `p' scans through the pattern as `d' scans through the data.
5164 `dend' is the end of the input string that `d' points within. `d'
5165 is advanced into the following input string whenever necessary, but
5166 this happens before fetching; therefore, at the beginning of the
5167 loop, `d' can be pointing at the end of a string, but it cannot
5168 equal `string2'. */
5169 if (pos >= size1)
5170 {
5171 /* Only match within string2. */
5172 d = string2 + pos - size1;
5173 dend = end_match_2 = string2 + stop - size1;
5174 end_match_1 = end1; /* Just to give it a value. */
5175 }
5176 else
5177 {
5178 if (stop < size1)
5179 {
5180 /* Only match within string1. */
5181 end_match_1 = string1 + stop;
5182 /* BEWARE!
5183 When we reach end_match_1, PREFETCH normally switches to string2.
5184 But in the present case, this means that just doing a PREFETCH
5185 makes us jump from `stop' to `gap' within the string.
5186 What we really want here is for the search to stop as
5187 soon as we hit end_match_1. That's why we set end_match_2
5188 to end_match_1 (since PREFETCH fails as soon as we hit
5189 end_match_2). */
5190 end_match_2 = end_match_1;
5191 }
5192 else
5193 { /* It's important to use this code when stop == size so that
5194 moving `d' from end1 to string2 will not prevent the d == dend
5195 check from catching the end of string. */
5196 end_match_1 = end1;
5197 end_match_2 = string2 + stop - size1;
5198 }
5199 d = string1 + pos;
5200 dend = end_match_1;
5201 }
5202
5203 DEBUG_PRINT1 ("The compiled pattern is: ");
5204 DEBUG_PRINT_COMPILED_PATTERN (bufp, p, pend);
5205 DEBUG_PRINT1 ("The string to match is: `");
5206 DEBUG_PRINT_DOUBLE_STRING (d, string1, size1, string2, size2);
5207 DEBUG_PRINT1 ("'\n");
5208
5209 /* This loops over pattern commands. It exits by returning from the
5210 function if the match is complete, or it drops through if the match
5211 fails at this starting point in the input data. */
5212 for (;;)
5213 {
5214 DEBUG_PRINT2 ("\n%p: ", p);
5215
5216 if (p == pend)
5217 { /* End of pattern means we might have succeeded. */
5218 DEBUG_PRINT1 ("end of pattern ... ");
5219
5220 /* If we haven't matched the entire string, and we want the
5221 longest match, try backtracking. */
5222 if (d != end_match_2)
5223 {
5224 /* 1 if this match ends in the same string (string1 or string2)
5225 as the best previous match. */
5226 boolean same_str_p = (FIRST_STRING_P (match_end)
5227 == FIRST_STRING_P (d));
5228 /* 1 if this match is the best seen so far. */
5229 boolean best_match_p;
5230
5231 /* AIX compiler got confused when this was combined
5232 with the previous declaration. */
5233 if (same_str_p)
5234 best_match_p = d > match_end;
5235 else
5236 best_match_p = !FIRST_STRING_P (d);
5237
5238 DEBUG_PRINT1 ("backtracking.\n");
5239
5240 if (!FAIL_STACK_EMPTY ())
5241 { /* More failure points to try. */
5242
5243 /* If exceeds best match so far, save it. */
5244 if (!best_regs_set || best_match_p)
5245 {
5246 best_regs_set = true;
5247 match_end = d;
5248
5249 DEBUG_PRINT1 ("\nSAVING match as best so far.\n");
5250
5251 for (reg = 1; reg < num_regs; reg++)
5252 {
5253 best_regstart[reg] = regstart[reg];
5254 best_regend[reg] = regend[reg];
5255 }
5256 }
5257 goto fail;
5258 }
5259
5260 /* If no failure points, don't restore garbage. And if
5261 last match is real best match, don't restore second
5262 best one. */
5263 else if (best_regs_set && !best_match_p)
5264 {
5265 restore_best_regs:
5266 /* Restore best match. It may happen that `dend ==
5267 end_match_1' while the restored d is in string2.
5268 For example, the pattern `x.*y.*z' against the
5269 strings `x-' and `y-z-', if the two strings are
5270 not consecutive in memory. */
5271 DEBUG_PRINT1 ("Restoring best registers.\n");
5272
5273 d = match_end;
5274 dend = ((d >= string1 && d <= end1)
5275 ? end_match_1 : end_match_2);
5276
5277 for (reg = 1; reg < num_regs; reg++)
5278 {
5279 regstart[reg] = best_regstart[reg];
5280 regend[reg] = best_regend[reg];
5281 }
5282 }
5283 } /* d != end_match_2 */
5284
5285 succeed_label:
5286 DEBUG_PRINT1 ("Accepting match.\n");
5287
5288 /* If caller wants register contents data back, do it. */
5289 if (regs && !bufp->no_sub)
5290 {
5291 /* Have the register data arrays been allocated? */
5292 if (bufp->regs_allocated == REGS_UNALLOCATED)
5293 { /* No. So allocate them with malloc. We need one
5294 extra element beyond `num_regs' for the `-1' marker
5295 GNU code uses. */
5296 regs->num_regs = MAX (RE_NREGS, num_regs + 1);
5297 regs->start = TALLOC (regs->num_regs, regoff_t);
5298 regs->end = TALLOC (regs->num_regs, regoff_t);
5299 if (regs->start == NULL || regs->end == NULL)
5300 {
5301 FREE_VARIABLES ();
5302 return -2;
5303 }
5304 bufp->regs_allocated = REGS_REALLOCATE;
5305 }
5306 else if (bufp->regs_allocated == REGS_REALLOCATE)
5307 { /* Yes. If we need more elements than were already
5308 allocated, reallocate them. If we need fewer, just
5309 leave it alone. */
5310 if (regs->num_regs < num_regs + 1)
5311 {
5312 regs->num_regs = num_regs + 1;
5313 RETALLOC (regs->start, regs->num_regs, regoff_t);
5314 RETALLOC (regs->end, regs->num_regs, regoff_t);
5315 if (regs->start == NULL || regs->end == NULL)
5316 {
5317 FREE_VARIABLES ();
5318 return -2;
5319 }
5320 }
5321 }
5322 else
5323 {
5324 /* These braces fend off a "empty body in an else-statement"
5325 warning under GCC when assert expands to nothing. */
5326 assert (bufp->regs_allocated == REGS_FIXED);
5327 }
5328
5329 /* Convert the pointer data in `regstart' and `regend' to
5330 indices. Register zero has to be set differently,
5331 since we haven't kept track of any info for it. */
5332 if (regs->num_regs > 0)
5333 {
5334 regs->start[0] = pos;
5335 regs->end[0] = POINTER_TO_OFFSET (d);
5336 }
5337
5338 /* Go through the first `min (num_regs, regs->num_regs)'
5339 registers, since that is all we initialized. */
5340 for (reg = 1; reg < MIN (num_regs, regs->num_regs); reg++)
5341 {
5342 if (REG_UNSET (regstart[reg]) || REG_UNSET (regend[reg]))
5343 regs->start[reg] = regs->end[reg] = -1;
5344 else
5345 {
5346 regs->start[reg]
5347 = (regoff_t) POINTER_TO_OFFSET (regstart[reg]);
5348 regs->end[reg]
5349 = (regoff_t) POINTER_TO_OFFSET (regend[reg]);
5350 }
5351 }
5352
5353 /* If the regs structure we return has more elements than
5354 were in the pattern, set the extra elements to -1. If
5355 we (re)allocated the registers, this is the case,
5356 because we always allocate enough to have at least one
5357 -1 at the end. */
5358 for (reg = num_regs; reg < regs->num_regs; reg++)
5359 regs->start[reg] = regs->end[reg] = -1;
5360 } /* regs && !bufp->no_sub */
5361
5362 DEBUG_PRINT4 ("%u failure points pushed, %u popped (%u remain).\n",
5363 nfailure_points_pushed, nfailure_points_popped,
5364 nfailure_points_pushed - nfailure_points_popped);
5365 DEBUG_PRINT2 ("%u registers pushed.\n", num_regs_pushed);
5366
5367 mcnt = POINTER_TO_OFFSET (d) - pos;
5368
5369 DEBUG_PRINT2 ("Returning %d from re_match_2.\n", mcnt);
5370
5371 FREE_VARIABLES ();
5372 return mcnt;
5373 }
5374
5375 /* Otherwise match next pattern command. */
5376 switch (SWITCH_ENUM_CAST ((re_opcode_t) *p++))
5377 {
5378 /* Ignore these. Used to ignore the n of succeed_n's which
5379 currently have n == 0. */
5380 case no_op:
5381 DEBUG_PRINT1 ("EXECUTING no_op.\n");
5382 break;
5383
5384 case succeed:
5385 DEBUG_PRINT1 ("EXECUTING succeed.\n");
5386 goto succeed_label;
5387
5388 /* Match the next n pattern characters exactly. The following
5389 byte in the pattern defines n, and the n bytes after that
5390 are the characters to match. */
5391 case exactn:
5392 mcnt = *p++;
5393 DEBUG_PRINT2 ("EXECUTING exactn %d.\n", mcnt);
5394
5395 /* Remember the start point to rollback upon failure. */
5396 dfail = d;
5397
5398 #ifndef emacs
5399 /* This is written out as an if-else so we don't waste time
5400 testing `translate' inside the loop. */
5401 if (RE_TRANSLATE_P (translate))
5402 do
5403 {
5404 PREFETCH ();
5405 if (RE_TRANSLATE (translate, *d) != *p++)
5406 {
5407 d = dfail;
5408 goto fail;
5409 }
5410 d++;
5411 }
5412 while (--mcnt);
5413 else
5414 do
5415 {
5416 PREFETCH ();
5417 if (*d++ != *p++)
5418 {
5419 d = dfail;
5420 goto fail;
5421 }
5422 }
5423 while (--mcnt);
5424 #else /* emacs */
5425 /* The cost of testing `translate' is comparatively small. */
5426 if (multibyte)
5427 do
5428 {
5429 int pat_charlen, buf_charlen;
5430 unsigned int pat_ch, buf_ch;
5431
5432 PREFETCH ();
5433 pat_ch = STRING_CHAR_AND_LENGTH (p, pend - p, pat_charlen);
5434 buf_ch = STRING_CHAR_AND_LENGTH (d, dend - d, buf_charlen);
5435
5436 if (TRANSLATE (buf_ch) != pat_ch)
5437 {
5438 d = dfail;
5439 goto fail;
5440 }
5441
5442 p += pat_charlen;
5443 d += buf_charlen;
5444 mcnt -= pat_charlen;
5445 }
5446 while (mcnt > 0);
5447 else
5448 do
5449 {
5450 unsigned int buf_ch;
5451
5452 PREFETCH ();
5453 buf_ch = *d++;
5454 TRANSLATE_VIA_MULTIBYTE (buf_ch);
5455 if (buf_ch != *p++)
5456 {
5457 d = dfail;
5458 goto fail;
5459 }
5460 }
5461 while (--mcnt);
5462 #endif
5463 break;
5464
5465
5466 /* Match any character except possibly a newline or a null. */
5467 case anychar:
5468 {
5469 int buf_charlen;
5470 re_wchar_t buf_ch;
5471
5472 DEBUG_PRINT1 ("EXECUTING anychar.\n");
5473
5474 PREFETCH ();
5475 buf_ch = RE_STRING_CHAR_AND_LENGTH (d, dend - d, buf_charlen);
5476 buf_ch = TRANSLATE (buf_ch);
5477
5478 if ((!(bufp->syntax & RE_DOT_NEWLINE)
5479 && buf_ch == '\n')
5480 || ((bufp->syntax & RE_DOT_NOT_NULL)
5481 && buf_ch == '\000'))
5482 goto fail;
5483
5484 DEBUG_PRINT2 (" Matched `%d'.\n", *d);
5485 d += buf_charlen;
5486 }
5487 break;
5488
5489
5490 case charset:
5491 case charset_not:
5492 {
5493 register unsigned int c;
5494 boolean not = (re_opcode_t) *(p - 1) == charset_not;
5495 int len;
5496
5497 /* Start of actual range_table, or end of bitmap if there is no
5498 range table. */
5499 re_char *range_table;
5500
5501 /* Nonzero if there is a range table. */
5502 int range_table_exists;
5503
5504 /* Number of ranges of range table. This is not included
5505 in the initial byte-length of the command. */
5506 int count = 0;
5507
5508 DEBUG_PRINT2 ("EXECUTING charset%s.\n", not ? "_not" : "");
5509
5510 range_table_exists = CHARSET_RANGE_TABLE_EXISTS_P (&p[-1]);
5511
5512 if (range_table_exists)
5513 {
5514 range_table = CHARSET_RANGE_TABLE (&p[-1]); /* Past the bitmap. */
5515 EXTRACT_NUMBER_AND_INCR (count, range_table);
5516 }
5517
5518 PREFETCH ();
5519 c = RE_STRING_CHAR_AND_LENGTH (d, dend - d, len);
5520 TRANSLATE_VIA_MULTIBYTE (c); /* The character to match. */
5521
5522 if (! multibyte || IS_REAL_ASCII (c))
5523 { /* Lookup bitmap. */
5524 /* Cast to `unsigned' instead of `unsigned char' in
5525 case the bit list is a full 32 bytes long. */
5526 if (c < (unsigned) (CHARSET_BITMAP_SIZE (&p[-1]) * BYTEWIDTH)
5527 && p[1 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH)))
5528 not = !not;
5529 }
5530 #ifdef emacs
5531 else if (range_table_exists)
5532 {
5533 int class_bits = CHARSET_RANGE_TABLE_BITS (&p[-1]);
5534
5535 if ( (class_bits & BIT_LOWER && ISLOWER (c))
5536 | (class_bits & BIT_MULTIBYTE)
5537 | (class_bits & BIT_PUNCT && ISPUNCT (c))
5538 | (class_bits & BIT_SPACE && ISSPACE (c))
5539 | (class_bits & BIT_UPPER && ISUPPER (c))
5540 | (class_bits & BIT_WORD && ISWORD (c)))
5541 not = !not;
5542 else
5543 CHARSET_LOOKUP_RANGE_TABLE_RAW (not, c, range_table, count);
5544 }
5545 #endif /* emacs */
5546
5547 if (range_table_exists)
5548 p = CHARSET_RANGE_TABLE_END (range_table, count);
5549 else
5550 p += CHARSET_BITMAP_SIZE (&p[-1]) + 1;
5551
5552 if (!not) goto fail;
5553
5554 d += len;
5555 break;
5556 }
5557
5558
5559 /* The beginning of a group is represented by start_memory.
5560 The argument is the register number. The text
5561 matched within the group is recorded (in the internal
5562 registers data structure) under the register number. */
5563 case start_memory:
5564 DEBUG_PRINT2 ("EXECUTING start_memory %d:\n", *p);
5565
5566 /* In case we need to undo this operation (via backtracking). */
5567 PUSH_FAILURE_REG ((unsigned int)*p);
5568
5569 regstart[*p] = d;
5570 regend[*p] = NULL; /* probably unnecessary. -sm */
5571 DEBUG_PRINT2 (" regstart: %d\n", POINTER_TO_OFFSET (regstart[*p]));
5572
5573 /* Move past the register number and inner group count. */
5574 p += 1;
5575 break;
5576
5577
5578 /* The stop_memory opcode represents the end of a group. Its
5579 argument is the same as start_memory's: the register number. */
5580 case stop_memory:
5581 DEBUG_PRINT2 ("EXECUTING stop_memory %d:\n", *p);
5582
5583 assert (!REG_UNSET (regstart[*p]));
5584 /* Strictly speaking, there should be code such as:
5585
5586 assert (REG_UNSET (regend[*p]));
5587 PUSH_FAILURE_REGSTOP ((unsigned int)*p);
5588
5589 But the only info to be pushed is regend[*p] and it is known to
5590 be UNSET, so there really isn't anything to push.
5591 Not pushing anything, on the other hand deprives us from the
5592 guarantee that regend[*p] is UNSET since undoing this operation
5593 will not reset its value properly. This is not important since
5594 the value will only be read on the next start_memory or at
5595 the very end and both events can only happen if this stop_memory
5596 is *not* undone. */
5597
5598 regend[*p] = d;
5599 DEBUG_PRINT2 (" regend: %d\n", POINTER_TO_OFFSET (regend[*p]));
5600
5601 /* Move past the register number and the inner group count. */
5602 p += 1;
5603 break;
5604
5605
5606 /* \<digit> has been turned into a `duplicate' command which is
5607 followed by the numeric value of <digit> as the register number. */
5608 case duplicate:
5609 {
5610 register re_char *d2, *dend2;
5611 int regno = *p++; /* Get which register to match against. */
5612 DEBUG_PRINT2 ("EXECUTING duplicate %d.\n", regno);
5613
5614 /* Can't back reference a group which we've never matched. */
5615 if (REG_UNSET (regstart[regno]) || REG_UNSET (regend[regno]))
5616 goto fail;
5617
5618 /* Where in input to try to start matching. */
5619 d2 = regstart[regno];
5620
5621 /* Remember the start point to rollback upon failure. */
5622 dfail = d;
5623
5624 /* Where to stop matching; if both the place to start and
5625 the place to stop matching are in the same string, then
5626 set to the place to stop, otherwise, for now have to use
5627 the end of the first string. */
5628
5629 dend2 = ((FIRST_STRING_P (regstart[regno])
5630 == FIRST_STRING_P (regend[regno]))
5631 ? regend[regno] : end_match_1);
5632 for (;;)
5633 {
5634 /* If necessary, advance to next segment in register
5635 contents. */
5636 while (d2 == dend2)
5637 {
5638 if (dend2 == end_match_2) break;
5639 if (dend2 == regend[regno]) break;
5640
5641 /* End of string1 => advance to string2. */
5642 d2 = string2;
5643 dend2 = regend[regno];
5644 }
5645 /* At end of register contents => success */
5646 if (d2 == dend2) break;
5647
5648 /* If necessary, advance to next segment in data. */
5649 PREFETCH ();
5650
5651 /* How many characters left in this segment to match. */
5652 mcnt = dend - d;
5653
5654 /* Want how many consecutive characters we can match in
5655 one shot, so, if necessary, adjust the count. */
5656 if (mcnt > dend2 - d2)
5657 mcnt = dend2 - d2;
5658
5659 /* Compare that many; failure if mismatch, else move
5660 past them. */
5661 if (RE_TRANSLATE_P (translate)
5662 ? bcmp_translate (d, d2, mcnt, translate, multibyte)
5663 : memcmp (d, d2, mcnt))
5664 {
5665 d = dfail;
5666 goto fail;
5667 }
5668 d += mcnt, d2 += mcnt;
5669 }
5670 }
5671 break;
5672
5673
5674 /* begline matches the empty string at the beginning of the string
5675 (unless `not_bol' is set in `bufp'), and after newlines. */
5676 case begline:
5677 DEBUG_PRINT1 ("EXECUTING begline.\n");
5678
5679 if (AT_STRINGS_BEG (d))
5680 {
5681 if (!bufp->not_bol) break;
5682 }
5683 else
5684 {
5685 unsigned c;
5686 GET_CHAR_BEFORE_2 (c, d, string1, end1, string2, end2);
5687 if (c == '\n')
5688 break;
5689 }
5690 /* In all other cases, we fail. */
5691 goto fail;
5692
5693
5694 /* endline is the dual of begline. */
5695 case endline:
5696 DEBUG_PRINT1 ("EXECUTING endline.\n");
5697
5698 if (AT_STRINGS_END (d))
5699 {
5700 if (!bufp->not_eol) break;
5701 }
5702 else
5703 {
5704 PREFETCH_NOLIMIT ();
5705 if (*d == '\n')
5706 break;
5707 }
5708 goto fail;
5709
5710
5711 /* Match at the very beginning of the data. */
5712 case begbuf:
5713 DEBUG_PRINT1 ("EXECUTING begbuf.\n");
5714 if (AT_STRINGS_BEG (d))
5715 break;
5716 goto fail;
5717
5718
5719 /* Match at the very end of the data. */
5720 case endbuf:
5721 DEBUG_PRINT1 ("EXECUTING endbuf.\n");
5722 if (AT_STRINGS_END (d))
5723 break;
5724 goto fail;
5725
5726
5727 /* on_failure_keep_string_jump is used to optimize `.*\n'. It
5728 pushes NULL as the value for the string on the stack. Then
5729 `POP_FAILURE_POINT' will keep the current value for the
5730 string, instead of restoring it. To see why, consider
5731 matching `foo\nbar' against `.*\n'. The .* matches the foo;
5732 then the . fails against the \n. But the next thing we want
5733 to do is match the \n against the \n; if we restored the
5734 string value, we would be back at the foo.
5735
5736 Because this is used only in specific cases, we don't need to
5737 check all the things that `on_failure_jump' does, to make
5738 sure the right things get saved on the stack. Hence we don't
5739 share its code. The only reason to push anything on the
5740 stack at all is that otherwise we would have to change
5741 `anychar's code to do something besides goto fail in this
5742 case; that seems worse than this. */
5743 case on_failure_keep_string_jump:
5744 EXTRACT_NUMBER_AND_INCR (mcnt, p);
5745 DEBUG_PRINT3 ("EXECUTING on_failure_keep_string_jump %d (to %p):\n",
5746 mcnt, p + mcnt);
5747
5748 PUSH_FAILURE_POINT (p - 3, NULL);
5749 break;
5750
5751 /* A nasty loop is introduced by the non-greedy *? and +?.
5752 With such loops, the stack only ever contains one failure point
5753 at a time, so that a plain on_failure_jump_loop kind of
5754 cycle detection cannot work. Worse yet, such a detection
5755 can not only fail to detect a cycle, but it can also wrongly
5756 detect a cycle (between different instantiations of the same
5757 loop).
5758 So the method used for those nasty loops is a little different:
5759 We use a special cycle-detection-stack-frame which is pushed
5760 when the on_failure_jump_nastyloop failure-point is *popped*.
5761 This special frame thus marks the beginning of one iteration
5762 through the loop and we can hence easily check right here
5763 whether something matched between the beginning and the end of
5764 the loop. */
5765 case on_failure_jump_nastyloop:
5766 EXTRACT_NUMBER_AND_INCR (mcnt, p);
5767 DEBUG_PRINT3 ("EXECUTING on_failure_jump_nastyloop %d (to %p):\n",
5768 mcnt, p + mcnt);
5769
5770 assert ((re_opcode_t)p[-4] == no_op);
5771 {
5772 int cycle = 0;
5773 CHECK_INFINITE_LOOP (p - 4, d);
5774 if (!cycle)
5775 /* If there's a cycle, just continue without pushing
5776 this failure point. The failure point is the "try again"
5777 option, which shouldn't be tried.
5778 We want (x?)*?y\1z to match both xxyz and xxyxz. */
5779 PUSH_FAILURE_POINT (p - 3, d);
5780 }
5781 break;
5782
5783 /* Simple loop detecting on_failure_jump: just check on the
5784 failure stack if the same spot was already hit earlier. */
5785 case on_failure_jump_loop:
5786 on_failure:
5787 EXTRACT_NUMBER_AND_INCR (mcnt, p);
5788 DEBUG_PRINT3 ("EXECUTING on_failure_jump_loop %d (to %p):\n",
5789 mcnt, p + mcnt);
5790 {
5791 int cycle = 0;
5792 CHECK_INFINITE_LOOP (p - 3, d);
5793 if (cycle)
5794 /* If there's a cycle, get out of the loop, as if the matching
5795 had failed. We used to just `goto fail' here, but that was
5796 aborting the search a bit too early: we want to keep the
5797 empty-loop-match and keep matching after the loop.
5798 We want (x?)*y\1z to match both xxyz and xxyxz. */
5799 p += mcnt;
5800 else
5801 PUSH_FAILURE_POINT (p - 3, d);
5802 }
5803 break;
5804
5805
5806 /* Uses of on_failure_jump:
5807
5808 Each alternative starts with an on_failure_jump that points
5809 to the beginning of the next alternative. Each alternative
5810 except the last ends with a jump that in effect jumps past
5811 the rest of the alternatives. (They really jump to the
5812 ending jump of the following alternative, because tensioning
5813 these jumps is a hassle.)
5814
5815 Repeats start with an on_failure_jump that points past both
5816 the repetition text and either the following jump or
5817 pop_failure_jump back to this on_failure_jump. */
5818 case on_failure_jump:
5819 EXTRACT_NUMBER_AND_INCR (mcnt, p);
5820 DEBUG_PRINT3 ("EXECUTING on_failure_jump %d (to %p):\n",
5821 mcnt, p + mcnt);
5822
5823 PUSH_FAILURE_POINT (p -3, d);
5824 break;
5825
5826 /* This operation is used for greedy *.
5827 Compare the beginning of the repeat with what in the
5828 pattern follows its end. If we can establish that there
5829 is nothing that they would both match, i.e., that we
5830 would have to backtrack because of (as in, e.g., `a*a')
5831 then we can use a non-backtracking loop based on
5832 on_failure_keep_string_jump instead of on_failure_jump. */
5833 case on_failure_jump_smart:
5834 EXTRACT_NUMBER_AND_INCR (mcnt, p);
5835 DEBUG_PRINT3 ("EXECUTING on_failure_jump_smart %d (to %p).\n",
5836 mcnt, p + mcnt);
5837 {
5838 re_char *p1 = p; /* Next operation. */
5839 /* Here, we discard `const', making re_match non-reentrant. */
5840 unsigned char *p2 = (unsigned char*) p + mcnt; /* Jump dest. */
5841 unsigned char *p3 = (unsigned char*) p - 3; /* opcode location. */
5842
5843 p -= 3; /* Reset so that we will re-execute the
5844 instruction once it's been changed. */
5845
5846 EXTRACT_NUMBER (mcnt, p2 - 2);
5847
5848 /* Ensure this is a indeed the trivial kind of loop
5849 we are expecting. */
5850 assert (skip_one_char (p1) == p2 - 3);
5851 assert ((re_opcode_t) p2[-3] == jump && p2 + mcnt == p);
5852 DEBUG_STATEMENT (debug += 2);
5853 if (mutually_exclusive_p (bufp, p1, p2))
5854 {
5855 /* Use a fast `on_failure_keep_string_jump' loop. */
5856 DEBUG_PRINT1 (" smart exclusive => fast loop.\n");
5857 *p3 = (unsigned char) on_failure_keep_string_jump;
5858 STORE_NUMBER (p2 - 2, mcnt + 3);
5859 }
5860 else
5861 {
5862 /* Default to a safe `on_failure_jump' loop. */
5863 DEBUG_PRINT1 (" smart default => slow loop.\n");
5864 *p3 = (unsigned char) on_failure_jump;
5865 }
5866 DEBUG_STATEMENT (debug -= 2);
5867 }
5868 break;
5869
5870 /* Unconditionally jump (without popping any failure points). */
5871 case jump:
5872 unconditional_jump:
5873 IMMEDIATE_QUIT_CHECK;
5874 EXTRACT_NUMBER_AND_INCR (mcnt, p); /* Get the amount to jump. */
5875 DEBUG_PRINT2 ("EXECUTING jump %d ", mcnt);
5876 p += mcnt; /* Do the jump. */
5877 DEBUG_PRINT2 ("(to %p).\n", p);
5878 break;
5879
5880
5881 /* Have to succeed matching what follows at least n times.
5882 After that, handle like `on_failure_jump'. */
5883 case succeed_n:
5884 /* Signedness doesn't matter since we only compare MCNT to 0. */
5885 EXTRACT_NUMBER (mcnt, p + 2);
5886 DEBUG_PRINT2 ("EXECUTING succeed_n %d.\n", mcnt);
5887
5888 /* Originally, mcnt is how many times we HAVE to succeed. */
5889 if (mcnt != 0)
5890 {
5891 /* Here, we discard `const', making re_match non-reentrant. */
5892 unsigned char *p2 = (unsigned char*) p + 2; /* counter loc. */
5893 mcnt--;
5894 p += 4;
5895 PUSH_NUMBER (p2, mcnt);
5896 }
5897 else
5898 /* The two bytes encoding mcnt == 0 are two no_op opcodes. */
5899 goto on_failure;
5900 break;
5901
5902 case jump_n:
5903 /* Signedness doesn't matter since we only compare MCNT to 0. */
5904 EXTRACT_NUMBER (mcnt, p + 2);
5905 DEBUG_PRINT2 ("EXECUTING jump_n %d.\n", mcnt);
5906
5907 /* Originally, this is how many times we CAN jump. */
5908 if (mcnt != 0)
5909 {
5910 /* Here, we discard `const', making re_match non-reentrant. */
5911 unsigned char *p2 = (unsigned char*) p + 2; /* counter loc. */
5912 mcnt--;
5913 PUSH_NUMBER (p2, mcnt);
5914 goto unconditional_jump;
5915 }
5916 /* If don't have to jump any more, skip over the rest of command. */
5917 else
5918 p += 4;
5919 break;
5920
5921 case set_number_at:
5922 {
5923 unsigned char *p2; /* Location of the counter. */
5924 DEBUG_PRINT1 ("EXECUTING set_number_at.\n");
5925
5926 EXTRACT_NUMBER_AND_INCR (mcnt, p);
5927 /* Here, we discard `const', making re_match non-reentrant. */
5928 p2 = (unsigned char*) p + mcnt;
5929 /* Signedness doesn't matter since we only copy MCNT's bits . */
5930 EXTRACT_NUMBER_AND_INCR (mcnt, p);
5931 DEBUG_PRINT3 (" Setting %p to %d.\n", p2, mcnt);
5932 PUSH_NUMBER (p2, mcnt);
5933 break;
5934 }
5935
5936 case wordbound:
5937 case notwordbound:
5938 not = (re_opcode_t) *(p - 1) == notwordbound;
5939 DEBUG_PRINT2 ("EXECUTING %swordbound.\n", not?"not":"");
5940
5941 /* We SUCCEED (or FAIL) in one of the following cases: */
5942
5943 /* Case 1: D is at the beginning or the end of string. */
5944 if (AT_STRINGS_BEG (d) || AT_STRINGS_END (d))
5945 not = !not;
5946 else
5947 {
5948 /* C1 is the character before D, S1 is the syntax of C1, C2
5949 is the character at D, and S2 is the syntax of C2. */
5950 re_wchar_t c1, c2;
5951 int s1, s2;
5952 int dummy;
5953 #ifdef emacs
5954 int offset = PTR_TO_OFFSET (d - 1);
5955 int charpos = SYNTAX_TABLE_BYTE_TO_CHAR (offset);
5956 UPDATE_SYNTAX_TABLE (charpos);
5957 #endif
5958 GET_CHAR_BEFORE_2 (c1, d, string1, end1, string2, end2);
5959 s1 = SYNTAX (c1);
5960 #ifdef emacs
5961 UPDATE_SYNTAX_TABLE_FORWARD (charpos + 1);
5962 #endif
5963 PREFETCH_NOLIMIT ();
5964 GET_CHAR_AFTER (c2, d, dummy);
5965 s2 = SYNTAX (c2);
5966
5967 if (/* Case 2: Only one of S1 and S2 is Sword. */
5968 ((s1 == Sword) != (s2 == Sword))
5969 /* Case 3: Both of S1 and S2 are Sword, and macro
5970 WORD_BOUNDARY_P (C1, C2) returns nonzero. */
5971 || ((s1 == Sword) && WORD_BOUNDARY_P (c1, c2)))
5972 not = !not;
5973 }
5974 if (not)
5975 break;
5976 else
5977 goto fail;
5978
5979 case wordbeg:
5980 DEBUG_PRINT1 ("EXECUTING wordbeg.\n");
5981
5982 /* We FAIL in one of the following cases: */
5983
5984 /* Case 1: D is at the end of string. */
5985 if (AT_STRINGS_END (d))
5986 goto fail;
5987 else
5988 {
5989 /* C1 is the character before D, S1 is the syntax of C1, C2
5990 is the character at D, and S2 is the syntax of C2. */
5991 re_wchar_t c1, c2;
5992 int s1, s2;
5993 int dummy;
5994 #ifdef emacs
5995 int offset = PTR_TO_OFFSET (d);
5996 int charpos = SYNTAX_TABLE_BYTE_TO_CHAR (offset);
5997 UPDATE_SYNTAX_TABLE (charpos);
5998 #endif
5999 PREFETCH ();
6000 GET_CHAR_AFTER (c2, d, dummy);
6001 s2 = SYNTAX (c2);
6002
6003 /* Case 2: S2 is not Sword. */
6004 if (s2 != Sword)
6005 goto fail;
6006
6007 /* Case 3: D is not at the beginning of string ... */
6008 if (!AT_STRINGS_BEG (d))
6009 {
6010 GET_CHAR_BEFORE_2 (c1, d, string1, end1, string2, end2);
6011 #ifdef emacs
6012 UPDATE_SYNTAX_TABLE_BACKWARD (charpos - 1);
6013 #endif
6014 s1 = SYNTAX (c1);
6015
6016 /* ... and S1 is Sword, and WORD_BOUNDARY_P (C1, C2)
6017 returns 0. */
6018 if ((s1 == Sword) && !WORD_BOUNDARY_P (c1, c2))
6019 goto fail;
6020 }
6021 }
6022 break;
6023
6024 case wordend:
6025 DEBUG_PRINT1 ("EXECUTING wordend.\n");
6026
6027 /* We FAIL in one of the following cases: */
6028
6029 /* Case 1: D is at the beginning of string. */
6030 if (AT_STRINGS_BEG (d))
6031 goto fail;
6032 else
6033 {
6034 /* C1 is the character before D, S1 is the syntax of C1, C2
6035 is the character at D, and S2 is the syntax of C2. */
6036 re_wchar_t c1, c2;
6037 int s1, s2;
6038 int dummy;
6039 #ifdef emacs
6040 int offset = PTR_TO_OFFSET (d) - 1;
6041 int charpos = SYNTAX_TABLE_BYTE_TO_CHAR (offset);
6042 UPDATE_SYNTAX_TABLE (charpos);
6043 #endif
6044 GET_CHAR_BEFORE_2 (c1, d, string1, end1, string2, end2);
6045 s1 = SYNTAX (c1);
6046
6047 /* Case 2: S1 is not Sword. */
6048 if (s1 != Sword)
6049 goto fail;
6050
6051 /* Case 3: D is not at the end of string ... */
6052 if (!AT_STRINGS_END (d))
6053 {
6054 PREFETCH_NOLIMIT ();
6055 GET_CHAR_AFTER (c2, d, dummy);
6056 #ifdef emacs
6057 UPDATE_SYNTAX_TABLE_FORWARD (charpos);
6058 #endif
6059 s2 = SYNTAX (c2);
6060
6061 /* ... and S2 is Sword, and WORD_BOUNDARY_P (C1, C2)
6062 returns 0. */
6063 if ((s2 == Sword) && !WORD_BOUNDARY_P (c1, c2))
6064 goto fail;
6065 }
6066 }
6067 break;
6068
6069 case symbeg:
6070 DEBUG_PRINT1 ("EXECUTING symbeg.\n");
6071
6072 /* We FAIL in one of the following cases: */
6073
6074 /* Case 1: D is at the end of string. */
6075 if (AT_STRINGS_END (d))
6076 goto fail;
6077 else
6078 {
6079 /* C1 is the character before D, S1 is the syntax of C1, C2
6080 is the character at D, and S2 is the syntax of C2. */
6081 re_wchar_t c1, c2;
6082 int s1, s2;
6083 #ifdef emacs
6084 int offset = PTR_TO_OFFSET (d);
6085 int charpos = SYNTAX_TABLE_BYTE_TO_CHAR (offset);
6086 UPDATE_SYNTAX_TABLE (charpos);
6087 #endif
6088 PREFETCH ();
6089 c2 = RE_STRING_CHAR (d, dend - d);
6090 s2 = SYNTAX (c2);
6091
6092 /* Case 2: S2 is neither Sword nor Ssymbol. */
6093 if (s2 != Sword && s2 != Ssymbol)
6094 goto fail;
6095
6096 /* Case 3: D is not at the beginning of string ... */
6097 if (!AT_STRINGS_BEG (d))
6098 {
6099 GET_CHAR_BEFORE_2 (c1, d, string1, end1, string2, end2);
6100 #ifdef emacs
6101 UPDATE_SYNTAX_TABLE_BACKWARD (charpos - 1);
6102 #endif
6103 s1 = SYNTAX (c1);
6104
6105 /* ... and S1 is Sword or Ssymbol. */
6106 if (s1 == Sword || s1 == Ssymbol)
6107 goto fail;
6108 }
6109 }
6110 break;
6111
6112 case symend:
6113 DEBUG_PRINT1 ("EXECUTING symend.\n");
6114
6115 /* We FAIL in one of the following cases: */
6116
6117 /* Case 1: D is at the beginning of string. */
6118 if (AT_STRINGS_BEG (d))
6119 goto fail;
6120 else
6121 {
6122 /* C1 is the character before D, S1 is the syntax of C1, C2
6123 is the character at D, and S2 is the syntax of C2. */
6124 re_wchar_t c1, c2;
6125 int s1, s2;
6126 #ifdef emacs
6127 int offset = PTR_TO_OFFSET (d) - 1;
6128 int charpos = SYNTAX_TABLE_BYTE_TO_CHAR (offset);
6129 UPDATE_SYNTAX_TABLE (charpos);
6130 #endif
6131 GET_CHAR_BEFORE_2 (c1, d, string1, end1, string2, end2);
6132 s1 = SYNTAX (c1);
6133
6134 /* Case 2: S1 is neither Ssymbol nor Sword. */
6135 if (s1 != Sword && s1 != Ssymbol)
6136 goto fail;
6137
6138 /* Case 3: D is not at the end of string ... */
6139 if (!AT_STRINGS_END (d))
6140 {
6141 PREFETCH_NOLIMIT ();
6142 c2 = RE_STRING_CHAR (d, dend - d);
6143 #ifdef emacs
6144 UPDATE_SYNTAX_TABLE_FORWARD (charpos + 1);
6145 #endif
6146 s2 = SYNTAX (c2);
6147
6148 /* ... and S2 is Sword or Ssymbol. */
6149 if (s2 == Sword || s2 == Ssymbol)
6150 goto fail;
6151 }
6152 }
6153 break;
6154
6155 case syntaxspec:
6156 case notsyntaxspec:
6157 not = (re_opcode_t) *(p - 1) == notsyntaxspec;
6158 mcnt = *p++;
6159 DEBUG_PRINT3 ("EXECUTING %ssyntaxspec %d.\n", not?"not":"", mcnt);
6160 PREFETCH ();
6161 #ifdef emacs
6162 {
6163 int offset = PTR_TO_OFFSET (d);
6164 int pos1 = SYNTAX_TABLE_BYTE_TO_CHAR (offset);
6165 UPDATE_SYNTAX_TABLE (pos1);
6166 }
6167 #endif
6168 {
6169 int len;
6170 re_wchar_t c;
6171
6172 GET_CHAR_AFTER (c, d, len);
6173 if ((SYNTAX (c) != (enum syntaxcode) mcnt) ^ not)
6174 goto fail;
6175 d += len;
6176 }
6177 break;
6178
6179 #ifdef emacs
6180 case before_dot:
6181 DEBUG_PRINT1 ("EXECUTING before_dot.\n");
6182 if (PTR_BYTE_POS (d) >= PT_BYTE)
6183 goto fail;
6184 break;
6185
6186 case at_dot:
6187 DEBUG_PRINT1 ("EXECUTING at_dot.\n");
6188 if (PTR_BYTE_POS (d) != PT_BYTE)
6189 goto fail;
6190 break;
6191
6192 case after_dot:
6193 DEBUG_PRINT1 ("EXECUTING after_dot.\n");
6194 if (PTR_BYTE_POS (d) <= PT_BYTE)
6195 goto fail;
6196 break;
6197
6198 case categoryspec:
6199 case notcategoryspec:
6200 not = (re_opcode_t) *(p - 1) == notcategoryspec;
6201 mcnt = *p++;
6202 DEBUG_PRINT3 ("EXECUTING %scategoryspec %d.\n", not?"not":"", mcnt);
6203 PREFETCH ();
6204 {
6205 int len;
6206 re_wchar_t c;
6207
6208 GET_CHAR_AFTER (c, d, len);
6209 if ((!CHAR_HAS_CATEGORY (c, mcnt)) ^ not)
6210 goto fail;
6211 d += len;
6212 }
6213 break;
6214
6215 #endif /* emacs */
6216
6217 default:
6218 abort ();
6219 }
6220 continue; /* Successfully executed one pattern command; keep going. */
6221
6222
6223 /* We goto here if a matching operation fails. */
6224 fail:
6225 IMMEDIATE_QUIT_CHECK;
6226 if (!FAIL_STACK_EMPTY ())
6227 {
6228 re_char *str, *pat;
6229 /* A restart point is known. Restore to that state. */
6230 DEBUG_PRINT1 ("\nFAIL:\n");
6231 POP_FAILURE_POINT (str, pat);
6232 switch (SWITCH_ENUM_CAST ((re_opcode_t) *pat++))
6233 {
6234 case on_failure_keep_string_jump:
6235 assert (str == NULL);
6236 goto continue_failure_jump;
6237
6238 case on_failure_jump_nastyloop:
6239 assert ((re_opcode_t)pat[-2] == no_op);
6240 PUSH_FAILURE_POINT (pat - 2, str);
6241 /* Fallthrough */
6242
6243 case on_failure_jump_loop:
6244 case on_failure_jump:
6245 case succeed_n:
6246 d = str;
6247 continue_failure_jump:
6248 EXTRACT_NUMBER_AND_INCR (mcnt, pat);
6249 p = pat + mcnt;
6250 break;
6251
6252 case no_op:
6253 /* A special frame used for nastyloops. */
6254 goto fail;
6255
6256 default:
6257 abort();
6258 }
6259
6260 assert (p >= bufp->buffer && p <= pend);
6261
6262 if (d >= string1 && d <= end1)
6263 dend = end_match_1;
6264 }
6265 else
6266 break; /* Matching at this starting point really fails. */
6267 } /* for (;;) */
6268
6269 if (best_regs_set)
6270 goto restore_best_regs;
6271
6272 FREE_VARIABLES ();
6273
6274 return -1; /* Failure to match. */
6275 } /* re_match_2 */
6276 \f
6277 /* Subroutine definitions for re_match_2. */
6278
6279 /* Return zero if TRANSLATE[S1] and TRANSLATE[S2] are identical for LEN
6280 bytes; nonzero otherwise. */
6281
6282 static int
6283 bcmp_translate (s1, s2, len, translate, multibyte)
6284 re_char *s1, *s2;
6285 register int len;
6286 RE_TRANSLATE_TYPE translate;
6287 const int multibyte;
6288 {
6289 register re_char *p1 = s1, *p2 = s2;
6290 re_char *p1_end = s1 + len;
6291 re_char *p2_end = s2 + len;
6292
6293 /* FIXME: Checking both p1 and p2 presumes that the two strings might have
6294 different lengths, but relying on a single `len' would break this. -sm */
6295 while (p1 < p1_end && p2 < p2_end)
6296 {
6297 int p1_charlen, p2_charlen;
6298 re_wchar_t p1_ch, p2_ch;
6299
6300 GET_CHAR_AFTER (p1_ch, p1, p1_charlen);
6301 GET_CHAR_AFTER (p2_ch, p2, p2_charlen);
6302
6303 if (RE_TRANSLATE (translate, p1_ch)
6304 != RE_TRANSLATE (translate, p2_ch))
6305 return 1;
6306
6307 p1 += p1_charlen, p2 += p2_charlen;
6308 }
6309
6310 if (p1 != p1_end || p2 != p2_end)
6311 return 1;
6312
6313 return 0;
6314 }
6315 \f
6316 /* Entry points for GNU code. */
6317
6318 /* re_compile_pattern is the GNU regular expression compiler: it
6319 compiles PATTERN (of length SIZE) and puts the result in BUFP.
6320 Returns 0 if the pattern was valid, otherwise an error string.
6321
6322 Assumes the `allocated' (and perhaps `buffer') and `translate' fields
6323 are set in BUFP on entry.
6324
6325 We call regex_compile to do the actual compilation. */
6326
6327 const char *
6328 re_compile_pattern (pattern, length, bufp)
6329 const char *pattern;
6330 size_t length;
6331 struct re_pattern_buffer *bufp;
6332 {
6333 reg_errcode_t ret;
6334
6335 /* GNU code is written to assume at least RE_NREGS registers will be set
6336 (and at least one extra will be -1). */
6337 bufp->regs_allocated = REGS_UNALLOCATED;
6338
6339 /* And GNU code determines whether or not to get register information
6340 by passing null for the REGS argument to re_match, etc., not by
6341 setting no_sub. */
6342 bufp->no_sub = 0;
6343
6344 ret = regex_compile ((re_char*) pattern, length, re_syntax_options, bufp);
6345
6346 if (!ret)
6347 return NULL;
6348 return gettext (re_error_msgid[(int) ret]);
6349 }
6350 WEAK_ALIAS (__re_compile_pattern, re_compile_pattern)
6351 \f
6352 /* Entry points compatible with 4.2 BSD regex library. We don't define
6353 them unless specifically requested. */
6354
6355 #if defined _REGEX_RE_COMP || defined _LIBC
6356
6357 /* BSD has one and only one pattern buffer. */
6358 static struct re_pattern_buffer re_comp_buf;
6359
6360 char *
6361 # ifdef _LIBC
6362 /* Make these definitions weak in libc, so POSIX programs can redefine
6363 these names if they don't use our functions, and still use
6364 regcomp/regexec below without link errors. */
6365 weak_function
6366 # endif
6367 re_comp (s)
6368 const char *s;
6369 {
6370 reg_errcode_t ret;
6371
6372 if (!s)
6373 {
6374 if (!re_comp_buf.buffer)
6375 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6376 return (char *) gettext ("No previous regular expression");
6377 return 0;
6378 }
6379
6380 if (!re_comp_buf.buffer)
6381 {
6382 re_comp_buf.buffer = (unsigned char *) malloc (200);
6383 if (re_comp_buf.buffer == NULL)
6384 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6385 return (char *) gettext (re_error_msgid[(int) REG_ESPACE]);
6386 re_comp_buf.allocated = 200;
6387
6388 re_comp_buf.fastmap = (char *) malloc (1 << BYTEWIDTH);
6389 if (re_comp_buf.fastmap == NULL)
6390 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6391 return (char *) gettext (re_error_msgid[(int) REG_ESPACE]);
6392 }
6393
6394 /* Since `re_exec' always passes NULL for the `regs' argument, we
6395 don't need to initialize the pattern buffer fields which affect it. */
6396
6397 ret = regex_compile (s, strlen (s), re_syntax_options, &re_comp_buf);
6398
6399 if (!ret)
6400 return NULL;
6401
6402 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6403 return (char *) gettext (re_error_msgid[(int) ret]);
6404 }
6405
6406
6407 int
6408 # ifdef _LIBC
6409 weak_function
6410 # endif
6411 re_exec (s)
6412 const char *s;
6413 {
6414 const int len = strlen (s);
6415 return
6416 0 <= re_search (&re_comp_buf, s, len, 0, len, (struct re_registers *) 0);
6417 }
6418 #endif /* _REGEX_RE_COMP */
6419 \f
6420 /* POSIX.2 functions. Don't define these for Emacs. */
6421
6422 #ifndef emacs
6423
6424 /* regcomp takes a regular expression as a string and compiles it.
6425
6426 PREG is a regex_t *. We do not expect any fields to be initialized,
6427 since POSIX says we shouldn't. Thus, we set
6428
6429 `buffer' to the compiled pattern;
6430 `used' to the length of the compiled pattern;
6431 `syntax' to RE_SYNTAX_POSIX_EXTENDED if the
6432 REG_EXTENDED bit in CFLAGS is set; otherwise, to
6433 RE_SYNTAX_POSIX_BASIC;
6434 `fastmap' to an allocated space for the fastmap;
6435 `fastmap_accurate' to zero;
6436 `re_nsub' to the number of subexpressions in PATTERN.
6437
6438 PATTERN is the address of the pattern string.
6439
6440 CFLAGS is a series of bits which affect compilation.
6441
6442 If REG_EXTENDED is set, we use POSIX extended syntax; otherwise, we
6443 use POSIX basic syntax.
6444
6445 If REG_NEWLINE is set, then . and [^...] don't match newline.
6446 Also, regexec will try a match beginning after every newline.
6447
6448 If REG_ICASE is set, then we considers upper- and lowercase
6449 versions of letters to be equivalent when matching.
6450
6451 If REG_NOSUB is set, then when PREG is passed to regexec, that
6452 routine will report only success or failure, and nothing about the
6453 registers.
6454
6455 It returns 0 if it succeeds, nonzero if it doesn't. (See regex.h for
6456 the return codes and their meanings.) */
6457
6458 int
6459 regcomp (preg, pattern, cflags)
6460 regex_t *__restrict preg;
6461 const char *__restrict pattern;
6462 int cflags;
6463 {
6464 reg_errcode_t ret;
6465 reg_syntax_t syntax
6466 = (cflags & REG_EXTENDED) ?
6467 RE_SYNTAX_POSIX_EXTENDED : RE_SYNTAX_POSIX_BASIC;
6468
6469 /* regex_compile will allocate the space for the compiled pattern. */
6470 preg->buffer = 0;
6471 preg->allocated = 0;
6472 preg->used = 0;
6473
6474 /* Try to allocate space for the fastmap. */
6475 preg->fastmap = (char *) malloc (1 << BYTEWIDTH);
6476
6477 if (cflags & REG_ICASE)
6478 {
6479 unsigned i;
6480
6481 preg->translate
6482 = (RE_TRANSLATE_TYPE) malloc (CHAR_SET_SIZE
6483 * sizeof (*(RE_TRANSLATE_TYPE)0));
6484 if (preg->translate == NULL)
6485 return (int) REG_ESPACE;
6486
6487 /* Map uppercase characters to corresponding lowercase ones. */
6488 for (i = 0; i < CHAR_SET_SIZE; i++)
6489 preg->translate[i] = ISUPPER (i) ? TOLOWER (i) : i;
6490 }
6491 else
6492 preg->translate = NULL;
6493
6494 /* If REG_NEWLINE is set, newlines are treated differently. */
6495 if (cflags & REG_NEWLINE)
6496 { /* REG_NEWLINE implies neither . nor [^...] match newline. */
6497 syntax &= ~RE_DOT_NEWLINE;
6498 syntax |= RE_HAT_LISTS_NOT_NEWLINE;
6499 }
6500 else
6501 syntax |= RE_NO_NEWLINE_ANCHOR;
6502
6503 preg->no_sub = !!(cflags & REG_NOSUB);
6504
6505 /* POSIX says a null character in the pattern terminates it, so we
6506 can use strlen here in compiling the pattern. */
6507 ret = regex_compile ((re_char*) pattern, strlen (pattern), syntax, preg);
6508
6509 /* POSIX doesn't distinguish between an unmatched open-group and an
6510 unmatched close-group: both are REG_EPAREN. */
6511 if (ret == REG_ERPAREN)
6512 ret = REG_EPAREN;
6513
6514 if (ret == REG_NOERROR && preg->fastmap)
6515 { /* Compute the fastmap now, since regexec cannot modify the pattern
6516 buffer. */
6517 re_compile_fastmap (preg);
6518 if (preg->can_be_null)
6519 { /* The fastmap can't be used anyway. */
6520 free (preg->fastmap);
6521 preg->fastmap = NULL;
6522 }
6523 }
6524 return (int) ret;
6525 }
6526 WEAK_ALIAS (__regcomp, regcomp)
6527
6528
6529 /* regexec searches for a given pattern, specified by PREG, in the
6530 string STRING.
6531
6532 If NMATCH is zero or REG_NOSUB was set in the cflags argument to
6533 `regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at
6534 least NMATCH elements, and we set them to the offsets of the
6535 corresponding matched substrings.
6536
6537 EFLAGS specifies `execution flags' which affect matching: if
6538 REG_NOTBOL is set, then ^ does not match at the beginning of the
6539 string; if REG_NOTEOL is set, then $ does not match at the end.
6540
6541 We return 0 if we find a match and REG_NOMATCH if not. */
6542
6543 int
6544 regexec (preg, string, nmatch, pmatch, eflags)
6545 const regex_t *__restrict preg;
6546 const char *__restrict string;
6547 size_t nmatch;
6548 regmatch_t pmatch[__restrict_arr];
6549 int eflags;
6550 {
6551 int ret;
6552 struct re_registers regs;
6553 regex_t private_preg;
6554 int len = strlen (string);
6555 boolean want_reg_info = !preg->no_sub && nmatch > 0 && pmatch;
6556
6557 private_preg = *preg;
6558
6559 private_preg.not_bol = !!(eflags & REG_NOTBOL);
6560 private_preg.not_eol = !!(eflags & REG_NOTEOL);
6561
6562 /* The user has told us exactly how many registers to return
6563 information about, via `nmatch'. We have to pass that on to the
6564 matching routines. */
6565 private_preg.regs_allocated = REGS_FIXED;
6566
6567 if (want_reg_info)
6568 {
6569 regs.num_regs = nmatch;
6570 regs.start = TALLOC (nmatch * 2, regoff_t);
6571 if (regs.start == NULL)
6572 return (int) REG_NOMATCH;
6573 regs.end = regs.start + nmatch;
6574 }
6575
6576 /* Instead of using not_eol to implement REG_NOTEOL, we could simply
6577 pass (&private_preg, string, len + 1, 0, len, ...) pretending the string
6578 was a little bit longer but still only matching the real part.
6579 This works because the `endline' will check for a '\n' and will find a
6580 '\0', correctly deciding that this is not the end of a line.
6581 But it doesn't work out so nicely for REG_NOTBOL, since we don't have
6582 a convenient '\0' there. For all we know, the string could be preceded
6583 by '\n' which would throw things off. */
6584
6585 /* Perform the searching operation. */
6586 ret = re_search (&private_preg, string, len,
6587 /* start: */ 0, /* range: */ len,
6588 want_reg_info ? &regs : (struct re_registers *) 0);
6589
6590 /* Copy the register information to the POSIX structure. */
6591 if (want_reg_info)
6592 {
6593 if (ret >= 0)
6594 {
6595 unsigned r;
6596
6597 for (r = 0; r < nmatch; r++)
6598 {
6599 pmatch[r].rm_so = regs.start[r];
6600 pmatch[r].rm_eo = regs.end[r];
6601 }
6602 }
6603
6604 /* If we needed the temporary register info, free the space now. */
6605 free (regs.start);
6606 }
6607
6608 /* We want zero return to mean success, unlike `re_search'. */
6609 return ret >= 0 ? (int) REG_NOERROR : (int) REG_NOMATCH;
6610 }
6611 WEAK_ALIAS (__regexec, regexec)
6612
6613
6614 /* Returns a message corresponding to an error code, ERRCODE, returned
6615 from either regcomp or regexec. We don't use PREG here. */
6616
6617 size_t
6618 regerror (errcode, preg, errbuf, errbuf_size)
6619 int errcode;
6620 const regex_t *preg;
6621 char *errbuf;
6622 size_t errbuf_size;
6623 {
6624 const char *msg;
6625 size_t msg_size;
6626
6627 if (errcode < 0
6628 || errcode >= (sizeof (re_error_msgid) / sizeof (re_error_msgid[0])))
6629 /* Only error codes returned by the rest of the code should be passed
6630 to this routine. If we are given anything else, or if other regex
6631 code generates an invalid error code, then the program has a bug.
6632 Dump core so we can fix it. */
6633 abort ();
6634
6635 msg = gettext (re_error_msgid[errcode]);
6636
6637 msg_size = strlen (msg) + 1; /* Includes the null. */
6638
6639 if (errbuf_size != 0)
6640 {
6641 if (msg_size > errbuf_size)
6642 {
6643 strncpy (errbuf, msg, errbuf_size - 1);
6644 errbuf[errbuf_size - 1] = 0;
6645 }
6646 else
6647 strcpy (errbuf, msg);
6648 }
6649
6650 return msg_size;
6651 }
6652 WEAK_ALIAS (__regerror, regerror)
6653
6654
6655 /* Free dynamically allocated space used by PREG. */
6656
6657 void
6658 regfree (preg)
6659 regex_t *preg;
6660 {
6661 if (preg->buffer != NULL)
6662 free (preg->buffer);
6663 preg->buffer = NULL;
6664
6665 preg->allocated = 0;
6666 preg->used = 0;
6667
6668 if (preg->fastmap != NULL)
6669 free (preg->fastmap);
6670 preg->fastmap = NULL;
6671 preg->fastmap_accurate = 0;
6672
6673 if (preg->translate != NULL)
6674 free (preg->translate);
6675 preg->translate = NULL;
6676 }
6677 WEAK_ALIAS (__regfree, regfree)
6678
6679 #endif /* not emacs */
6680
6681 /* arch-tag: 4ffd68ba-2a9e-435b-a21a-018990f9eeb2
6682 (do not change this comment) */
Unnamed repository; edit this file 'description' to name the repository.