HCoop / bpt / emacs.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Merge unicode branch
[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, 2007
7 Free Software Foundation, Inc.
8
9 This program is free software; you can redistribute it and/or modify
10 it under the terms of the GNU General Public License as published by
11 the Free Software Foundation; either version 3, or (at your option)
12 any later version.
13
14 This program is distributed in the hope that it will be useful,
15 but WITHOUT ANY WARRANTY; without even the implied warranty of
16 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
17 GNU General Public License for more details.
18
19 You should have received a copy of the GNU General Public License
20 along with this program; if not, write to the Free Software
21 Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301,
22 USA. */
23
24 /* TODO:
25 - structure the opcode space into opcode+flag.
26 - merge with glibc's regex.[ch].
27 - replace (succeed_n + jump_n + set_number_at) with something that doesn't
28 need to modify the compiled regexp so that re_match can be reentrant.
29 - get rid of on_failure_jump_smart by doing the optimization in re_comp
30 rather than at run-time, so that re_match can be reentrant.
31 */
32
33 /* AIX requires this to be the first thing in the file. */
34 #if defined _AIX && !defined REGEX_MALLOC
35 #pragma alloca
36 #endif
37
38 #ifdef HAVE_CONFIG_H
39 # include <config.h>
40 #endif
41
42 #if defined STDC_HEADERS && !defined emacs
43 # include <stddef.h>
44 #else
45 /* We need this for `regex.h', and perhaps for the Emacs include files. */
46 # include <sys/types.h>
47 #endif
48
49 /* Whether to use ISO C Amendment 1 wide char functions.
50 Those should not be used for Emacs since it uses its own. */
51 #if defined _LIBC
52 #define WIDE_CHAR_SUPPORT 1
53 #else
54 #define WIDE_CHAR_SUPPORT \
55 (HAVE_WCTYPE_H && HAVE_WCHAR_H && HAVE_BTOWC && !emacs)
56 #endif
57
58 /* For platform which support the ISO C amendement 1 functionality we
59 support user defined character classes. */
60 #if WIDE_CHAR_SUPPORT
61 /* Solaris 2.5 has a bug: <wchar.h> must be included before <wctype.h>. */
62 # include <wchar.h>
63 # include <wctype.h>
64 #endif
65
66 #ifdef _LIBC
67 /* We have to keep the namespace clean. */
68 # define regfree(preg) __regfree (preg)
69 # define regexec(pr, st, nm, pm, ef) __regexec (pr, st, nm, pm, ef)
70 # define regcomp(preg, pattern, cflags) __regcomp (preg, pattern, cflags)
71 # define regerror(err_code, preg, errbuf, errbuf_size) \
72 __regerror(err_code, preg, errbuf, errbuf_size)
73 # define re_set_registers(bu, re, nu, st, en) \
74 __re_set_registers (bu, re, nu, st, en)
75 # define re_match_2(bufp, string1, size1, string2, size2, pos, regs, stop) \
76 __re_match_2 (bufp, string1, size1, string2, size2, pos, regs, stop)
77 # define re_match(bufp, string, size, pos, regs) \
78 __re_match (bufp, string, size, pos, regs)
79 # define re_search(bufp, string, size, startpos, range, regs) \
80 __re_search (bufp, string, size, startpos, range, regs)
81 # define re_compile_pattern(pattern, length, bufp) \
82 __re_compile_pattern (pattern, length, bufp)
83 # define re_set_syntax(syntax) __re_set_syntax (syntax)
84 # define re_search_2(bufp, st1, s1, st2, s2, startpos, range, regs, stop) \
85 __re_search_2 (bufp, st1, s1, st2, s2, startpos, range, regs, stop)
86 # define re_compile_fastmap(bufp) __re_compile_fastmap (bufp)
87
88 /* Make sure we call libc's function even if the user overrides them. */
89 # define btowc __btowc
90 # define iswctype __iswctype
91 # define wctype __wctype
92
93 # define WEAK_ALIAS(a,b) weak_alias (a, b)
94
95 /* We are also using some library internals. */
96 # include <locale/localeinfo.h>
97 # include <locale/elem-hash.h>
98 # include <langinfo.h>
99 #else
100 # define WEAK_ALIAS(a,b)
101 #endif
102
103 /* This is for other GNU distributions with internationalized messages. */
104 #if HAVE_LIBINTL_H || defined _LIBC
105 # include <libintl.h>
106 #else
107 # define gettext(msgid) (msgid)
108 #endif
109
110 #ifndef gettext_noop
111 /* This define is so xgettext can find the internationalizable
112 strings. */
113 # define gettext_noop(String) String
114 #endif
115
116 /* The `emacs' switch turns on certain matching commands
117 that make sense only in Emacs. */
118 #ifdef emacs
119
120 # include "lisp.h"
121 # include "buffer.h"
122
123 /* Make syntax table lookup grant data in gl_state. */
124 # define SYNTAX_ENTRY_VIA_PROPERTY
125
126 # include "syntax.h"
127 # include "character.h"
128 # include "category.h"
129
130 # ifdef malloc
131 # undef malloc
132 # endif
133 # define malloc xmalloc
134 # ifdef realloc
135 # undef realloc
136 # endif
137 # define realloc xrealloc
138 # ifdef free
139 # undef free
140 # endif
141 # define free xfree
142
143 /* Converts the pointer to the char to BEG-based offset from the start. */
144 # define PTR_TO_OFFSET(d) POS_AS_IN_BUFFER (POINTER_TO_OFFSET (d))
145 # define POS_AS_IN_BUFFER(p) ((p) + (NILP (re_match_object) || BUFFERP (re_match_object)))
146
147 # define RE_MULTIBYTE_P(bufp) ((bufp)->multibyte)
148 # define RE_TARGET_MULTIBYTE_P(bufp) ((bufp)->target_multibyte)
149 # define RE_STRING_CHAR(p, s, multibyte) \
150 (multibyte ? (STRING_CHAR (p, s)) : (*(p)))
151 # define RE_STRING_CHAR_AND_LENGTH(p, s, len, multibyte) \
152 (multibyte ? (STRING_CHAR_AND_LENGTH (p, s, len)) : ((len) = 1, *(p)))
153
154 # define RE_CHAR_TO_MULTIBYTE(c) unibyte_to_multibyte_table[(c)]
155
156 # define RE_CHAR_TO_UNIBYTE(c) \
157 (ASCII_CHAR_P (c) ? (c) \
158 : CHAR_BYTE8_P (c) ? CHAR_TO_BYTE8 (c) \
159 : multibyte_char_to_unibyte_safe (c))
160
161 /* Set C a (possibly converted to multibyte) character before P. P
162 points into a string which is the virtual concatenation of STR1
163 (which ends at END1) or STR2 (which ends at END2). */
164 # define GET_CHAR_BEFORE_2(c, p, str1, end1, str2, end2) \
165 do { \
166 if (target_multibyte) \
167 { \
168 re_char *dtemp = (p) == (str2) ? (end1) : (p); \
169 re_char *dlimit = ((p) > (str2) && (p) <= (end2)) ? (str2) : (str1); \
170 while (dtemp-- > dlimit && !CHAR_HEAD_P (*dtemp)); \
171 c = STRING_CHAR (dtemp, (p) - dtemp); \
172 } \
173 else \
174 { \
175 (c = ((p) == (str2) ? (end1) : (p))[-1]); \
176 (c) = RE_CHAR_TO_MULTIBYTE (c); \
177 } \
178 } while (0)
179
180 /* Set C a (possibly converted to multibyte) character at P, and set
181 LEN to the byte length of that character. */
182 # define GET_CHAR_AFTER(c, p, len) \
183 do { \
184 if (target_multibyte) \
185 (c) = STRING_CHAR_AND_LENGTH (p, 0, len); \
186 else \
187 { \
188 (c) = *p; \
189 len = 1; \
190 (c) = RE_CHAR_TO_MULTIBYTE (c); \
191 } \
192 } while (0)
193
194 #else /* not emacs */
195
196 /* If we are not linking with Emacs proper,
197 we can't use the relocating allocator
198 even if config.h says that we can. */
199 # undef REL_ALLOC
200
201 # if defined STDC_HEADERS || defined _LIBC
202 # include <stdlib.h>
203 # else
204 char *malloc ();
205 char *realloc ();
206 # endif
207
208 /* When used in Emacs's lib-src, we need xmalloc and xrealloc. */
209
210 void *
211 xmalloc (size)
212 size_t size;
213 {
214 register void *val;
215 val = (void *) malloc (size);
216 if (!val && size)
217 {
218 write (2, "virtual memory exhausted\n", 25);
219 exit (1);
220 }
221 return val;
222 }
223
224 void *
225 xrealloc (block, size)
226 void *block;
227 size_t size;
228 {
229 register void *val;
230 /* We must call malloc explicitly when BLOCK is 0, since some
231 reallocs don't do this. */
232 if (! block)
233 val = (void *) malloc (size);
234 else
235 val = (void *) realloc (block, size);
236 if (!val && size)
237 {
238 write (2, "virtual memory exhausted\n", 25);
239 exit (1);
240 }
241 return val;
242 }
243
244 # ifdef malloc
245 # undef malloc
246 # endif
247 # define malloc xmalloc
248 # ifdef realloc
249 # undef realloc
250 # endif
251 # define realloc xrealloc
252
253 /* When used in Emacs's lib-src, we need to get bzero and bcopy somehow.
254 If nothing else has been done, use the method below. */
255 # ifdef INHIBIT_STRING_HEADER
256 # if !(defined HAVE_BZERO && defined HAVE_BCOPY)
257 # if !defined bzero && !defined bcopy
258 # undef INHIBIT_STRING_HEADER
259 # endif
260 # endif
261 # endif
262
263 /* This is the normal way of making sure we have memcpy, memcmp and bzero.
264 This is used in most programs--a few other programs avoid this
265 by defining INHIBIT_STRING_HEADER. */
266 # ifndef INHIBIT_STRING_HEADER
267 # if defined HAVE_STRING_H || defined STDC_HEADERS || defined _LIBC
268 # include <string.h>
269 # ifndef bzero
270 # ifndef _LIBC
271 # define bzero(s, n) (memset (s, '\0', n), (s))
272 # else
273 # define bzero(s, n) __bzero (s, n)
274 # endif
275 # endif
276 # else
277 # include <strings.h>
278 # ifndef memcmp
279 # define memcmp(s1, s2, n) bcmp (s1, s2, n)
280 # endif
281 # ifndef memcpy
282 # define memcpy(d, s, n) (bcopy (s, d, n), (d))
283 # endif
284 # endif
285 # endif
286
287 /* Define the syntax stuff for \<, \>, etc. */
288
289 /* Sword must be nonzero for the wordchar pattern commands in re_match_2. */
290 enum syntaxcode { Swhitespace = 0, Sword = 1, Ssymbol = 2 };
291
292 # ifdef SWITCH_ENUM_BUG
293 # define SWITCH_ENUM_CAST(x) ((int)(x))
294 # else
295 # define SWITCH_ENUM_CAST(x) (x)
296 # endif
297
298 /* Dummy macros for non-Emacs environments. */
299 # define BASE_LEADING_CODE_P(c) (0)
300 # define CHAR_CHARSET(c) 0
301 # define CHARSET_LEADING_CODE_BASE(c) 0
302 # define MAX_MULTIBYTE_LENGTH 1
303 # define RE_MULTIBYTE_P(x) 0
304 # define RE_TARGET_MULTIBYTE_P(x) 0
305 # define WORD_BOUNDARY_P(c1, c2) (0)
306 # define CHAR_HEAD_P(p) (1)
307 # define SINGLE_BYTE_CHAR_P(c) (1)
308 # define SAME_CHARSET_P(c1, c2) (1)
309 # define MULTIBYTE_FORM_LENGTH(p, s) (1)
310 # define PREV_CHAR_BOUNDARY(p, limit) ((p)--)
311 # define STRING_CHAR(p, s) (*(p))
312 # define RE_STRING_CHAR(p, s, multibyte) STRING_CHAR ((p), (s))
313 # define CHAR_STRING(c, s) (*(s) = (c), 1)
314 # define STRING_CHAR_AND_LENGTH(p, s, actual_len) ((actual_len) = 1, *(p))
315 # define RE_STRING_CHAR_AND_LENGTH(p, s, len, multibyte) STRING_CHAR_AND_LENGTH ((p), (s), (len))
316 # define RE_CHAR_TO_MULTIBYTE(c) (c)
317 # define RE_CHAR_TO_UNIBYTE(c) (c)
318 # define GET_CHAR_BEFORE_2(c, p, str1, end1, str2, end2) \
319 (c = ((p) == (str2) ? *((end1) - 1) : *((p) - 1)))
320 # define GET_CHAR_AFTER(c, p, len) \
321 (c = *p, len = 1)
322 # define MAKE_CHAR(charset, c1, c2) (c1)
323 # define BYTE8_TO_CHAR(c) (c)
324 # define CHAR_BYTE8_P(c) (0)
325 # define CHAR_LEADING_CODE(c) (c)
326
327 #endif /* not emacs */
328
329 #ifndef RE_TRANSLATE
330 # define RE_TRANSLATE(TBL, C) ((unsigned char)(TBL)[C])
331 # define RE_TRANSLATE_P(TBL) (TBL)
332 #endif
333 \f
334 /* Get the interface, including the syntax bits. */
335 #include "regex.h"
336
337 /* isalpha etc. are used for the character classes. */
338 #include <ctype.h>
339
340 #ifdef emacs
341
342 /* 1 if C is an ASCII character. */
343 # define IS_REAL_ASCII(c) ((c) < 0200)
344
345 /* 1 if C is a unibyte character. */
346 # define ISUNIBYTE(c) (SINGLE_BYTE_CHAR_P ((c)))
347
348 /* The Emacs definitions should not be directly affected by locales. */
349
350 /* In Emacs, these are only used for single-byte characters. */
351 # define ISDIGIT(c) ((c) >= '0' && (c) <= '9')
352 # define ISCNTRL(c) ((c) < ' ')
353 # define ISXDIGIT(c) (((c) >= '0' && (c) <= '9') \
354 || ((c) >= 'a' && (c) <= 'f') \
355 || ((c) >= 'A' && (c) <= 'F'))
356
357 /* This is only used for single-byte characters. */
358 # define ISBLANK(c) ((c) == ' ' || (c) == '\t')
359
360 /* The rest must handle multibyte characters. */
361
362 # define ISGRAPH(c) (SINGLE_BYTE_CHAR_P (c) \
363 ? (c) > ' ' && !((c) >= 0177 && (c) <= 0237) \
364 : 1)
365
366 # define ISPRINT(c) (SINGLE_BYTE_CHAR_P (c) \
367 ? (c) >= ' ' && !((c) >= 0177 && (c) <= 0237) \
368 : 1)
369
370 # define ISALNUM(c) (IS_REAL_ASCII (c) \
371 ? (((c) >= 'a' && (c) <= 'z') \
372 || ((c) >= 'A' && (c) <= 'Z') \
373 || ((c) >= '0' && (c) <= '9')) \
374 : SYNTAX (c) == Sword)
375
376 # define ISALPHA(c) (IS_REAL_ASCII (c) \
377 ? (((c) >= 'a' && (c) <= 'z') \
378 || ((c) >= 'A' && (c) <= 'Z')) \
379 : SYNTAX (c) == Sword)
380
381 # define ISLOWER(c) (LOWERCASEP (c))
382
383 # define ISPUNCT(c) (IS_REAL_ASCII (c) \
384 ? ((c) > ' ' && (c) < 0177 \
385 && !(((c) >= 'a' && (c) <= 'z') \
386 || ((c) >= 'A' && (c) <= 'Z') \
387 || ((c) >= '0' && (c) <= '9'))) \
388 : SYNTAX (c) != Sword)
389
390 # define ISSPACE(c) (SYNTAX (c) == Swhitespace)
391
392 # define ISUPPER(c) (UPPERCASEP (c))
393
394 # define ISWORD(c) (SYNTAX (c) == Sword)
395
396 #else /* not emacs */
397
398 /* Jim Meyering writes:
399
400 "... Some ctype macros are valid only for character codes that
401 isascii says are ASCII (SGI's IRIX-4.0.5 is one such system --when
402 using /bin/cc or gcc but without giving an ansi option). So, all
403 ctype uses should be through macros like ISPRINT... If
404 STDC_HEADERS is defined, then autoconf has verified that the ctype
405 macros don't need to be guarded with references to isascii. ...
406 Defining isascii to 1 should let any compiler worth its salt
407 eliminate the && through constant folding."
408 Solaris defines some of these symbols so we must undefine them first. */
409
410 # undef ISASCII
411 # if defined STDC_HEADERS || (!defined isascii && !defined HAVE_ISASCII)
412 # define ISASCII(c) 1
413 # else
414 # define ISASCII(c) isascii(c)
415 # endif
416
417 /* 1 if C is an ASCII character. */
418 # define IS_REAL_ASCII(c) ((c) < 0200)
419
420 /* This distinction is not meaningful, except in Emacs. */
421 # define ISUNIBYTE(c) 1
422
423 # ifdef isblank
424 # define ISBLANK(c) (ISASCII (c) && isblank (c))
425 # else
426 # define ISBLANK(c) ((c) == ' ' || (c) == '\t')
427 # endif
428 # ifdef isgraph
429 # define ISGRAPH(c) (ISASCII (c) && isgraph (c))
430 # else
431 # define ISGRAPH(c) (ISASCII (c) && isprint (c) && !isspace (c))
432 # endif
433
434 # undef ISPRINT
435 # define ISPRINT(c) (ISASCII (c) && isprint (c))
436 # define ISDIGIT(c) (ISASCII (c) && isdigit (c))
437 # define ISALNUM(c) (ISASCII (c) && isalnum (c))
438 # define ISALPHA(c) (ISASCII (c) && isalpha (c))
439 # define ISCNTRL(c) (ISASCII (c) && iscntrl (c))
440 # define ISLOWER(c) (ISASCII (c) && islower (c))
441 # define ISPUNCT(c) (ISASCII (c) && ispunct (c))
442 # define ISSPACE(c) (ISASCII (c) && isspace (c))
443 # define ISUPPER(c) (ISASCII (c) && isupper (c))
444 # define ISXDIGIT(c) (ISASCII (c) && isxdigit (c))
445
446 # define ISWORD(c) ISALPHA(c)
447
448 # ifdef _tolower
449 # define TOLOWER(c) _tolower(c)
450 # else
451 # define TOLOWER(c) tolower(c)
452 # endif
453
454 /* How many characters in the character set. */
455 # define CHAR_SET_SIZE 256
456
457 # ifdef SYNTAX_TABLE
458
459 extern char *re_syntax_table;
460
461 # else /* not SYNTAX_TABLE */
462
463 static char re_syntax_table[CHAR_SET_SIZE];
464
465 static void
466 init_syntax_once ()
467 {
468 register int c;
469 static int done = 0;
470
471 if (done)
472 return;
473
474 bzero (re_syntax_table, sizeof re_syntax_table);
475
476 for (c = 0; c < CHAR_SET_SIZE; ++c)
477 if (ISALNUM (c))
478 re_syntax_table[c] = Sword;
479
480 re_syntax_table['_'] = Ssymbol;
481
482 done = 1;
483 }
484
485 # endif /* not SYNTAX_TABLE */
486
487 # define SYNTAX(c) re_syntax_table[(c)]
488
489 #endif /* not emacs */
490 \f
491 #ifndef NULL
492 # define NULL (void *)0
493 #endif
494
495 /* We remove any previous definition of `SIGN_EXTEND_CHAR',
496 since ours (we hope) works properly with all combinations of
497 machines, compilers, `char' and `unsigned char' argument types.
498 (Per Bothner suggested the basic approach.) */
499 #undef SIGN_EXTEND_CHAR
500 #if __STDC__
501 # define SIGN_EXTEND_CHAR(c) ((signed char) (c))
502 #else /* not __STDC__ */
503 /* As in Harbison and Steele. */
504 # define SIGN_EXTEND_CHAR(c) ((((unsigned char) (c)) ^ 128) - 128)
505 #endif
506 \f
507 /* Should we use malloc or alloca? If REGEX_MALLOC is not defined, we
508 use `alloca' instead of `malloc'. This is because using malloc in
509 re_search* or re_match* could cause memory leaks when C-g is used in
510 Emacs; also, malloc is slower and causes storage fragmentation. On
511 the other hand, malloc is more portable, and easier to debug.
512
513 Because we sometimes use alloca, some routines have to be macros,
514 not functions -- `alloca'-allocated space disappears at the end of the
515 function it is called in. */
516
517 #ifdef REGEX_MALLOC
518
519 # define REGEX_ALLOCATE malloc
520 # define REGEX_REALLOCATE(source, osize, nsize) realloc (source, nsize)
521 # define REGEX_FREE free
522
523 #else /* not REGEX_MALLOC */
524
525 /* Emacs already defines alloca, sometimes. */
526 # ifndef alloca
527
528 /* Make alloca work the best possible way. */
529 # ifdef __GNUC__
530 # define alloca __builtin_alloca
531 # else /* not __GNUC__ */
532 # ifdef HAVE_ALLOCA_H
533 # include <alloca.h>
534 # endif /* HAVE_ALLOCA_H */
535 # endif /* not __GNUC__ */
536
537 # endif /* not alloca */
538
539 # define REGEX_ALLOCATE alloca
540
541 /* Assumes a `char *destination' variable. */
542 # define REGEX_REALLOCATE(source, osize, nsize) \
543 (destination = (char *) alloca (nsize), \
544 memcpy (destination, source, osize))
545
546 /* No need to do anything to free, after alloca. */
547 # define REGEX_FREE(arg) ((void)0) /* Do nothing! But inhibit gcc warning. */
548
549 #endif /* not REGEX_MALLOC */
550
551 /* Define how to allocate the failure stack. */
552
553 #if defined REL_ALLOC && defined REGEX_MALLOC
554
555 # define REGEX_ALLOCATE_STACK(size) \
556 r_alloc (&failure_stack_ptr, (size))
557 # define REGEX_REALLOCATE_STACK(source, osize, nsize) \
558 r_re_alloc (&failure_stack_ptr, (nsize))
559 # define REGEX_FREE_STACK(ptr) \
560 r_alloc_free (&failure_stack_ptr)
561
562 #else /* not using relocating allocator */
563
564 # ifdef REGEX_MALLOC
565
566 # define REGEX_ALLOCATE_STACK malloc
567 # define REGEX_REALLOCATE_STACK(source, osize, nsize) realloc (source, nsize)
568 # define REGEX_FREE_STACK free
569
570 # else /* not REGEX_MALLOC */
571
572 # define REGEX_ALLOCATE_STACK alloca
573
574 # define REGEX_REALLOCATE_STACK(source, osize, nsize) \
575 REGEX_REALLOCATE (source, osize, nsize)
576 /* No need to explicitly free anything. */
577 # define REGEX_FREE_STACK(arg) ((void)0)
578
579 # endif /* not REGEX_MALLOC */
580 #endif /* not using relocating allocator */
581
582
583 /* True if `size1' is non-NULL and PTR is pointing anywhere inside
584 `string1' or just past its end. This works if PTR is NULL, which is
585 a good thing. */
586 #define FIRST_STRING_P(ptr) \
587 (size1 && string1 <= (ptr) && (ptr) <= string1 + size1)
588
589 /* (Re)Allocate N items of type T using malloc, or fail. */
590 #define TALLOC(n, t) ((t *) malloc ((n) * sizeof (t)))
591 #define RETALLOC(addr, n, t) ((addr) = (t *) realloc (addr, (n) * sizeof (t)))
592 #define RETALLOC_IF(addr, n, t) \
593 if (addr) RETALLOC((addr), (n), t); else (addr) = TALLOC ((n), t)
594 #define REGEX_TALLOC(n, t) ((t *) REGEX_ALLOCATE ((n) * sizeof (t)))
595
596 #define BYTEWIDTH 8 /* In bits. */
597
598 #define STREQ(s1, s2) ((strcmp (s1, s2) == 0))
599
600 #undef MAX
601 #undef MIN
602 #define MAX(a, b) ((a) > (b) ? (a) : (b))
603 #define MIN(a, b) ((a) < (b) ? (a) : (b))
604
605 /* Type of source-pattern and string chars. */
606 typedef const unsigned char re_char;
607
608 typedef char boolean;
609 #define false 0
610 #define true 1
611
612 static int re_match_2_internal _RE_ARGS ((struct re_pattern_buffer *bufp,
613 re_char *string1, int size1,
614 re_char *string2, int size2,
615 int pos,
616 struct re_registers *regs,
617 int stop));
618 \f
619 /* These are the command codes that appear in compiled regular
620 expressions. Some opcodes are followed by argument bytes. A
621 command code can specify any interpretation whatsoever for its
622 arguments. Zero bytes may appear in the compiled regular expression. */
623
624 typedef enum
625 {
626 no_op = 0,
627
628 /* Succeed right away--no more backtracking. */
629 succeed,
630
631 /* Followed by one byte giving n, then by n literal bytes. */
632 exactn,
633
634 /* Matches any (more or less) character. */
635 anychar,
636
637 /* Matches any one char belonging to specified set. First
638 following byte is number of bitmap bytes. Then come bytes
639 for a bitmap saying which chars are in. Bits in each byte
640 are ordered low-bit-first. A character is in the set if its
641 bit is 1. A character too large to have a bit in the map is
642 automatically not in the set.
643
644 If the length byte has the 0x80 bit set, then that stuff
645 is followed by a range table:
646 2 bytes of flags for character sets (low 8 bits, high 8 bits)
647 See RANGE_TABLE_WORK_BITS below.
648 2 bytes, the number of pairs that follow (upto 32767)
649 pairs, each 2 multibyte characters,
650 each multibyte character represented as 3 bytes. */
651 charset,
652
653 /* Same parameters as charset, but match any character that is
654 not one of those specified. */
655 charset_not,
656
657 /* Start remembering the text that is matched, for storing in a
658 register. Followed by one byte with the register number, in
659 the range 0 to one less than the pattern buffer's re_nsub
660 field. */
661 start_memory,
662
663 /* Stop remembering the text that is matched and store it in a
664 memory register. Followed by one byte with the register
665 number, in the range 0 to one less than `re_nsub' in the
666 pattern buffer. */
667 stop_memory,
668
669 /* Match a duplicate of something remembered. Followed by one
670 byte containing the register number. */
671 duplicate,
672
673 /* Fail unless at beginning of line. */
674 begline,
675
676 /* Fail unless at end of line. */
677 endline,
678
679 /* Succeeds if at beginning of buffer (if emacs) or at beginning
680 of string to be matched (if not). */
681 begbuf,
682
683 /* Analogously, for end of buffer/string. */
684 endbuf,
685
686 /* Followed by two byte relative address to which to jump. */
687 jump,
688
689 /* Followed by two-byte relative address of place to resume at
690 in case of failure. */
691 on_failure_jump,
692
693 /* Like on_failure_jump, but pushes a placeholder instead of the
694 current string position when executed. */
695 on_failure_keep_string_jump,
696
697 /* Just like `on_failure_jump', except that it checks that we
698 don't get stuck in an infinite loop (matching an empty string
699 indefinitely). */
700 on_failure_jump_loop,
701
702 /* Just like `on_failure_jump_loop', except that it checks for
703 a different kind of loop (the kind that shows up with non-greedy
704 operators). This operation has to be immediately preceded
705 by a `no_op'. */
706 on_failure_jump_nastyloop,
707
708 /* A smart `on_failure_jump' used for greedy * and + operators.
709 It analyses the loop before which it is put and if the
710 loop does not require backtracking, it changes itself to
711 `on_failure_keep_string_jump' and short-circuits the loop,
712 else it just defaults to changing itself into `on_failure_jump'.
713 It assumes that it is pointing to just past a `jump'. */
714 on_failure_jump_smart,
715
716 /* Followed by two-byte relative address and two-byte number n.
717 After matching N times, jump to the address upon failure.
718 Does not work if N starts at 0: use on_failure_jump_loop
719 instead. */
720 succeed_n,
721
722 /* Followed by two-byte relative address, and two-byte number n.
723 Jump to the address N times, then fail. */
724 jump_n,
725
726 /* Set the following two-byte relative address to the
727 subsequent two-byte number. The address *includes* the two
728 bytes of number. */
729 set_number_at,
730
731 wordbeg, /* Succeeds if at word beginning. */
732 wordend, /* Succeeds if at word end. */
733
734 wordbound, /* Succeeds if at a word boundary. */
735 notwordbound, /* Succeeds if not at a word boundary. */
736
737 symbeg, /* Succeeds if at symbol beginning. */
738 symend, /* Succeeds if at symbol end. */
739
740 /* Matches any character whose syntax is specified. Followed by
741 a byte which contains a syntax code, e.g., Sword. */
742 syntaxspec,
743
744 /* Matches any character whose syntax is not that specified. */
745 notsyntaxspec
746
747 #ifdef emacs
748 ,before_dot, /* Succeeds if before point. */
749 at_dot, /* Succeeds if at point. */
750 after_dot, /* Succeeds if after point. */
751
752 /* Matches any character whose category-set contains the specified
753 category. The operator is followed by a byte which contains a
754 category code (mnemonic ASCII character). */
755 categoryspec,
756
757 /* Matches any character whose category-set does not contain the
758 specified category. The operator is followed by a byte which
759 contains the category code (mnemonic ASCII character). */
760 notcategoryspec
761 #endif /* emacs */
762 } re_opcode_t;
763 \f
764 /* Common operations on the compiled pattern. */
765
766 /* Store NUMBER in two contiguous bytes starting at DESTINATION. */
767
768 #define STORE_NUMBER(destination, number) \
769 do { \
770 (destination)[0] = (number) & 0377; \
771 (destination)[1] = (number) >> 8; \
772 } while (0)
773
774 /* Same as STORE_NUMBER, except increment DESTINATION to
775 the byte after where the number is stored. Therefore, DESTINATION
776 must be an lvalue. */
777
778 #define STORE_NUMBER_AND_INCR(destination, number) \
779 do { \
780 STORE_NUMBER (destination, number); \
781 (destination) += 2; \
782 } while (0)
783
784 /* Put into DESTINATION a number stored in two contiguous bytes starting
785 at SOURCE. */
786
787 #define EXTRACT_NUMBER(destination, source) \
788 do { \
789 (destination) = *(source) & 0377; \
790 (destination) += SIGN_EXTEND_CHAR (*((source) + 1)) << 8; \
791 } while (0)
792
793 #ifdef DEBUG
794 static void extract_number _RE_ARGS ((int *dest, re_char *source));
795 static void
796 extract_number (dest, source)
797 int *dest;
798 re_char *source;
799 {
800 int temp = SIGN_EXTEND_CHAR (*(source + 1));
801 *dest = *source & 0377;
802 *dest += temp << 8;
803 }
804
805 # ifndef EXTRACT_MACROS /* To debug the macros. */
806 # undef EXTRACT_NUMBER
807 # define EXTRACT_NUMBER(dest, src) extract_number (&dest, src)
808 # endif /* not EXTRACT_MACROS */
809
810 #endif /* DEBUG */
811
812 /* Same as EXTRACT_NUMBER, except increment SOURCE to after the number.
813 SOURCE must be an lvalue. */
814
815 #define EXTRACT_NUMBER_AND_INCR(destination, source) \
816 do { \
817 EXTRACT_NUMBER (destination, source); \
818 (source) += 2; \
819 } while (0)
820
821 #ifdef DEBUG
822 static void extract_number_and_incr _RE_ARGS ((int *destination,
823 re_char **source));
824 static void
825 extract_number_and_incr (destination, source)
826 int *destination;
827 re_char **source;
828 {
829 extract_number (destination, *source);
830 *source += 2;
831 }
832
833 # ifndef EXTRACT_MACROS
834 # undef EXTRACT_NUMBER_AND_INCR
835 # define EXTRACT_NUMBER_AND_INCR(dest, src) \
836 extract_number_and_incr (&dest, &src)
837 # endif /* not EXTRACT_MACROS */
838
839 #endif /* DEBUG */
840 \f
841 /* Store a multibyte character in three contiguous bytes starting
842 DESTINATION, and increment DESTINATION to the byte after where the
843 character is stored. Therefore, DESTINATION must be an lvalue. */
844
845 #define STORE_CHARACTER_AND_INCR(destination, character) \
846 do { \
847 (destination)[0] = (character) & 0377; \
848 (destination)[1] = ((character) >> 8) & 0377; \
849 (destination)[2] = (character) >> 16; \
850 (destination) += 3; \
851 } while (0)
852
853 /* Put into DESTINATION a character stored in three contiguous bytes
854 starting at SOURCE. */
855
856 #define EXTRACT_CHARACTER(destination, source) \
857 do { \
858 (destination) = ((source)[0] \
859 | ((source)[1] << 8) \
860 | ((source)[2] << 16)); \
861 } while (0)
862
863
864 /* Macros for charset. */
865
866 /* Size of bitmap of charset P in bytes. P is a start of charset,
867 i.e. *P is (re_opcode_t) charset or (re_opcode_t) charset_not. */
868 #define CHARSET_BITMAP_SIZE(p) ((p)[1] & 0x7F)
869
870 /* Nonzero if charset P has range table. */
871 #define CHARSET_RANGE_TABLE_EXISTS_P(p) ((p)[1] & 0x80)
872
873 /* Return the address of range table of charset P. But not the start
874 of table itself, but the before where the number of ranges is
875 stored. `2 +' means to skip re_opcode_t and size of bitmap,
876 and the 2 bytes of flags at the start of the range table. */
877 #define CHARSET_RANGE_TABLE(p) (&(p)[4 + CHARSET_BITMAP_SIZE (p)])
878
879 /* Extract the bit flags that start a range table. */
880 #define CHARSET_RANGE_TABLE_BITS(p) \
881 ((p)[2 + CHARSET_BITMAP_SIZE (p)] \
882 + (p)[3 + CHARSET_BITMAP_SIZE (p)] * 0x100)
883
884 /* Test if C is listed in the bitmap of charset P. */
885 #define CHARSET_LOOKUP_BITMAP(p, c) \
886 ((c) < CHARSET_BITMAP_SIZE (p) * BYTEWIDTH \
887 && (p)[2 + (c) / BYTEWIDTH] & (1 << ((c) % BYTEWIDTH)))
888
889 /* Return the address of end of RANGE_TABLE. COUNT is number of
890 ranges (which is a pair of (start, end)) in the RANGE_TABLE. `* 2'
891 is start of range and end of range. `* 3' is size of each start
892 and end. */
893 #define CHARSET_RANGE_TABLE_END(range_table, count) \
894 ((range_table) + (count) * 2 * 3)
895
896 /* Test if C is in RANGE_TABLE. A flag NOT is negated if C is in.
897 COUNT is number of ranges in RANGE_TABLE. */
898 #define CHARSET_LOOKUP_RANGE_TABLE_RAW(not, c, range_table, count) \
899 do \
900 { \
901 re_wchar_t range_start, range_end; \
902 re_char *p; \
903 re_char *range_table_end \
904 = CHARSET_RANGE_TABLE_END ((range_table), (count)); \
905 \
906 for (p = (range_table); p < range_table_end; p += 2 * 3) \
907 { \
908 EXTRACT_CHARACTER (range_start, p); \
909 EXTRACT_CHARACTER (range_end, p + 3); \
910 \
911 if (range_start <= (c) && (c) <= range_end) \
912 { \
913 (not) = !(not); \
914 break; \
915 } \
916 } \
917 } \
918 while (0)
919
920 /* Test if C is in range table of CHARSET. The flag NOT is negated if
921 C is listed in it. */
922 #define CHARSET_LOOKUP_RANGE_TABLE(not, c, charset) \
923 do \
924 { \
925 /* Number of ranges in range table. */ \
926 int count; \
927 re_char *range_table = CHARSET_RANGE_TABLE (charset); \
928 \
929 EXTRACT_NUMBER_AND_INCR (count, range_table); \
930 CHARSET_LOOKUP_RANGE_TABLE_RAW ((not), (c), range_table, count); \
931 } \
932 while (0)
933 \f
934 /* If DEBUG is defined, Regex prints many voluminous messages about what
935 it is doing (if the variable `debug' is nonzero). If linked with the
936 main program in `iregex.c', you can enter patterns and strings
937 interactively. And if linked with the main program in `main.c' and
938 the other test files, you can run the already-written tests. */
939
940 #ifdef DEBUG
941
942 /* We use standard I/O for debugging. */
943 # include <stdio.h>
944
945 /* It is useful to test things that ``must'' be true when debugging. */
946 # include <assert.h>
947
948 static int debug = -100000;
949
950 # define DEBUG_STATEMENT(e) e
951 # define DEBUG_PRINT1(x) if (debug > 0) printf (x)
952 # define DEBUG_PRINT2(x1, x2) if (debug > 0) printf (x1, x2)
953 # define DEBUG_PRINT3(x1, x2, x3) if (debug > 0) printf (x1, x2, x3)
954 # define DEBUG_PRINT4(x1, x2, x3, x4) if (debug > 0) printf (x1, x2, x3, x4)
955 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) \
956 if (debug > 0) print_partial_compiled_pattern (s, e)
957 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) \
958 if (debug > 0) print_double_string (w, s1, sz1, s2, sz2)
959
960
961 /* Print the fastmap in human-readable form. */
962
963 void
964 print_fastmap (fastmap)
965 char *fastmap;
966 {
967 unsigned was_a_range = 0;
968 unsigned i = 0;
969
970 while (i < (1 << BYTEWIDTH))
971 {
972 if (fastmap[i++])
973 {
974 was_a_range = 0;
975 putchar (i - 1);
976 while (i < (1 << BYTEWIDTH) && fastmap[i])
977 {
978 was_a_range = 1;
979 i++;
980 }
981 if (was_a_range)
982 {
983 printf ("-");
984 putchar (i - 1);
985 }
986 }
987 }
988 putchar ('\n');
989 }
990
991
992 /* Print a compiled pattern string in human-readable form, starting at
993 the START pointer into it and ending just before the pointer END. */
994
995 void
996 print_partial_compiled_pattern (start, end)
997 re_char *start;
998 re_char *end;
999 {
1000 int mcnt, mcnt2;
1001 re_char *p = start;
1002 re_char *pend = end;
1003
1004 if (start == NULL)
1005 {
1006 fprintf (stderr, "(null)\n");
1007 return;
1008 }
1009
1010 /* Loop over pattern commands. */
1011 while (p < pend)
1012 {
1013 fprintf (stderr, "%d:\t", p - start);
1014
1015 switch ((re_opcode_t) *p++)
1016 {
1017 case no_op:
1018 fprintf (stderr, "/no_op");
1019 break;
1020
1021 case succeed:
1022 fprintf (stderr, "/succeed");
1023 break;
1024
1025 case exactn:
1026 mcnt = *p++;
1027 fprintf (stderr, "/exactn/%d", mcnt);
1028 do
1029 {
1030 fprintf (stderr, "/%c", *p++);
1031 }
1032 while (--mcnt);
1033 break;
1034
1035 case start_memory:
1036 fprintf (stderr, "/start_memory/%d", *p++);
1037 break;
1038
1039 case stop_memory:
1040 fprintf (stderr, "/stop_memory/%d", *p++);
1041 break;
1042
1043 case duplicate:
1044 fprintf (stderr, "/duplicate/%d", *p++);
1045 break;
1046
1047 case anychar:
1048 fprintf (stderr, "/anychar");
1049 break;
1050
1051 case charset:
1052 case charset_not:
1053 {
1054 register int c, last = -100;
1055 register int in_range = 0;
1056 int length = CHARSET_BITMAP_SIZE (p - 1);
1057 int has_range_table = CHARSET_RANGE_TABLE_EXISTS_P (p - 1);
1058
1059 fprintf (stderr, "/charset [%s",
1060 (re_opcode_t) *(p - 1) == charset_not ? "^" : "");
1061
1062 if (p + *p >= pend)
1063 fprintf (stderr, " !extends past end of pattern! ");
1064
1065 for (c = 0; c < 256; c++)
1066 if (c / 8 < length
1067 && (p[1 + (c/8)] & (1 << (c % 8))))
1068 {
1069 /* Are we starting a range? */
1070 if (last + 1 == c && ! in_range)
1071 {
1072 fprintf (stderr, "-");
1073 in_range = 1;
1074 }
1075 /* Have we broken a range? */
1076 else if (last + 1 != c && in_range)
1077 {
1078 fprintf (stderr, "%c", last);
1079 in_range = 0;
1080 }
1081
1082 if (! in_range)
1083 fprintf (stderr, "%c", c);
1084
1085 last = c;
1086 }
1087
1088 if (in_range)
1089 fprintf (stderr, "%c", last);
1090
1091 fprintf (stderr, "]");
1092
1093 p += 1 + length;
1094
1095 if (has_range_table)
1096 {
1097 int count;
1098 fprintf (stderr, "has-range-table");
1099
1100 /* ??? Should print the range table; for now, just skip it. */
1101 p += 2; /* skip range table bits */
1102 EXTRACT_NUMBER_AND_INCR (count, p);
1103 p = CHARSET_RANGE_TABLE_END (p, count);
1104 }
1105 }
1106 break;
1107
1108 case begline:
1109 fprintf (stderr, "/begline");
1110 break;
1111
1112 case endline:
1113 fprintf (stderr, "/endline");
1114 break;
1115
1116 case on_failure_jump:
1117 extract_number_and_incr (&mcnt, &p);
1118 fprintf (stderr, "/on_failure_jump to %d", p + mcnt - start);
1119 break;
1120
1121 case on_failure_keep_string_jump:
1122 extract_number_and_incr (&mcnt, &p);
1123 fprintf (stderr, "/on_failure_keep_string_jump to %d", p + mcnt - start);
1124 break;
1125
1126 case on_failure_jump_nastyloop:
1127 extract_number_and_incr (&mcnt, &p);
1128 fprintf (stderr, "/on_failure_jump_nastyloop to %d", p + mcnt - start);
1129 break;
1130
1131 case on_failure_jump_loop:
1132 extract_number_and_incr (&mcnt, &p);
1133 fprintf (stderr, "/on_failure_jump_loop to %d", p + mcnt - start);
1134 break;
1135
1136 case on_failure_jump_smart:
1137 extract_number_and_incr (&mcnt, &p);
1138 fprintf (stderr, "/on_failure_jump_smart to %d", p + mcnt - start);
1139 break;
1140
1141 case jump:
1142 extract_number_and_incr (&mcnt, &p);
1143 fprintf (stderr, "/jump to %d", p + mcnt - start);
1144 break;
1145
1146 case succeed_n:
1147 extract_number_and_incr (&mcnt, &p);
1148 extract_number_and_incr (&mcnt2, &p);
1149 fprintf (stderr, "/succeed_n to %d, %d times", p - 2 + mcnt - start, mcnt2);
1150 break;
1151
1152 case jump_n:
1153 extract_number_and_incr (&mcnt, &p);
1154 extract_number_and_incr (&mcnt2, &p);
1155 fprintf (stderr, "/jump_n to %d, %d times", p - 2 + mcnt - start, mcnt2);
1156 break;
1157
1158 case set_number_at:
1159 extract_number_and_incr (&mcnt, &p);
1160 extract_number_and_incr (&mcnt2, &p);
1161 fprintf (stderr, "/set_number_at location %d to %d", p - 2 + mcnt - start, mcnt2);
1162 break;
1163
1164 case wordbound:
1165 fprintf (stderr, "/wordbound");
1166 break;
1167
1168 case notwordbound:
1169 fprintf (stderr, "/notwordbound");
1170 break;
1171
1172 case wordbeg:
1173 fprintf (stderr, "/wordbeg");
1174 break;
1175
1176 case wordend:
1177 fprintf (stderr, "/wordend");
1178 break;
1179
1180 case symbeg:
1181 fprintf (stderr, "/symbeg");
1182 break;
1183
1184 case symend:
1185 fprintf (stderr, "/symend");
1186 break;
1187
1188 case syntaxspec:
1189 fprintf (stderr, "/syntaxspec");
1190 mcnt = *p++;
1191 fprintf (stderr, "/%d", mcnt);
1192 break;
1193
1194 case notsyntaxspec:
1195 fprintf (stderr, "/notsyntaxspec");
1196 mcnt = *p++;
1197 fprintf (stderr, "/%d", mcnt);
1198 break;
1199
1200 # ifdef emacs
1201 case before_dot:
1202 fprintf (stderr, "/before_dot");
1203 break;
1204
1205 case at_dot:
1206 fprintf (stderr, "/at_dot");
1207 break;
1208
1209 case after_dot:
1210 fprintf (stderr, "/after_dot");
1211 break;
1212
1213 case categoryspec:
1214 fprintf (stderr, "/categoryspec");
1215 mcnt = *p++;
1216 fprintf (stderr, "/%d", mcnt);
1217 break;
1218
1219 case notcategoryspec:
1220 fprintf (stderr, "/notcategoryspec");
1221 mcnt = *p++;
1222 fprintf (stderr, "/%d", mcnt);
1223 break;
1224 # endif /* emacs */
1225
1226 case begbuf:
1227 fprintf (stderr, "/begbuf");
1228 break;
1229
1230 case endbuf:
1231 fprintf (stderr, "/endbuf");
1232 break;
1233
1234 default:
1235 fprintf (stderr, "?%d", *(p-1));
1236 }
1237
1238 fprintf (stderr, "\n");
1239 }
1240
1241 fprintf (stderr, "%d:\tend of pattern.\n", p - start);
1242 }
1243
1244
1245 void
1246 print_compiled_pattern (bufp)
1247 struct re_pattern_buffer *bufp;
1248 {
1249 re_char *buffer = bufp->buffer;
1250
1251 print_partial_compiled_pattern (buffer, buffer + bufp->used);
1252 printf ("%ld bytes used/%ld bytes allocated.\n",
1253 bufp->used, bufp->allocated);
1254
1255 if (bufp->fastmap_accurate && bufp->fastmap)
1256 {
1257 printf ("fastmap: ");
1258 print_fastmap (bufp->fastmap);
1259 }
1260
1261 printf ("re_nsub: %d\t", bufp->re_nsub);
1262 printf ("regs_alloc: %d\t", bufp->regs_allocated);
1263 printf ("can_be_null: %d\t", bufp->can_be_null);
1264 printf ("no_sub: %d\t", bufp->no_sub);
1265 printf ("not_bol: %d\t", bufp->not_bol);
1266 printf ("not_eol: %d\t", bufp->not_eol);
1267 printf ("syntax: %lx\n", bufp->syntax);
1268 fflush (stdout);
1269 /* Perhaps we should print the translate table? */
1270 }
1271
1272
1273 void
1274 print_double_string (where, string1, size1, string2, size2)
1275 re_char *where;
1276 re_char *string1;
1277 re_char *string2;
1278 int size1;
1279 int size2;
1280 {
1281 int this_char;
1282
1283 if (where == NULL)
1284 printf ("(null)");
1285 else
1286 {
1287 if (FIRST_STRING_P (where))
1288 {
1289 for (this_char = where - string1; this_char < size1; this_char++)
1290 putchar (string1[this_char]);
1291
1292 where = string2;
1293 }
1294
1295 for (this_char = where - string2; this_char < size2; this_char++)
1296 putchar (string2[this_char]);
1297 }
1298 }
1299
1300 #else /* not DEBUG */
1301
1302 # undef assert
1303 # define assert(e)
1304
1305 # define DEBUG_STATEMENT(e)
1306 # define DEBUG_PRINT1(x)
1307 # define DEBUG_PRINT2(x1, x2)
1308 # define DEBUG_PRINT3(x1, x2, x3)
1309 # define DEBUG_PRINT4(x1, x2, x3, x4)
1310 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e)
1311 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2)
1312
1313 #endif /* not DEBUG */
1314 \f
1315 /* Set by `re_set_syntax' to the current regexp syntax to recognize. Can
1316 also be assigned to arbitrarily: each pattern buffer stores its own
1317 syntax, so it can be changed between regex compilations. */
1318 /* This has no initializer because initialized variables in Emacs
1319 become read-only after dumping. */
1320 reg_syntax_t re_syntax_options;
1321
1322
1323 /* Specify the precise syntax of regexps for compilation. This provides
1324 for compatibility for various utilities which historically have
1325 different, incompatible syntaxes.
1326
1327 The argument SYNTAX is a bit mask comprised of the various bits
1328 defined in regex.h. We return the old syntax. */
1329
1330 reg_syntax_t
1331 re_set_syntax (syntax)
1332 reg_syntax_t syntax;
1333 {
1334 reg_syntax_t ret = re_syntax_options;
1335
1336 re_syntax_options = syntax;
1337 return ret;
1338 }
1339 WEAK_ALIAS (__re_set_syntax, re_set_syntax)
1340
1341 /* Regexp to use to replace spaces, or NULL meaning don't. */
1342 static re_char *whitespace_regexp;
1343
1344 void
1345 re_set_whitespace_regexp (regexp)
1346 const char *regexp;
1347 {
1348 whitespace_regexp = (re_char *) regexp;
1349 }
1350 WEAK_ALIAS (__re_set_syntax, re_set_syntax)
1351 \f
1352 /* This table gives an error message for each of the error codes listed
1353 in regex.h. Obviously the order here has to be same as there.
1354 POSIX doesn't require that we do anything for REG_NOERROR,
1355 but why not be nice? */
1356
1357 static const char *re_error_msgid[] =
1358 {
1359 gettext_noop ("Success"), /* REG_NOERROR */
1360 gettext_noop ("No match"), /* REG_NOMATCH */
1361 gettext_noop ("Invalid regular expression"), /* REG_BADPAT */
1362 gettext_noop ("Invalid collation character"), /* REG_ECOLLATE */
1363 gettext_noop ("Invalid character class name"), /* REG_ECTYPE */
1364 gettext_noop ("Trailing backslash"), /* REG_EESCAPE */
1365 gettext_noop ("Invalid back reference"), /* REG_ESUBREG */
1366 gettext_noop ("Unmatched [ or [^"), /* REG_EBRACK */
1367 gettext_noop ("Unmatched ( or \\("), /* REG_EPAREN */
1368 gettext_noop ("Unmatched \\{"), /* REG_EBRACE */
1369 gettext_noop ("Invalid content of \\{\\}"), /* REG_BADBR */
1370 gettext_noop ("Invalid range end"), /* REG_ERANGE */
1371 gettext_noop ("Memory exhausted"), /* REG_ESPACE */
1372 gettext_noop ("Invalid preceding regular expression"), /* REG_BADRPT */
1373 gettext_noop ("Premature end of regular expression"), /* REG_EEND */
1374 gettext_noop ("Regular expression too big"), /* REG_ESIZE */
1375 gettext_noop ("Unmatched ) or \\)"), /* REG_ERPAREN */
1376 gettext_noop ("Range striding over charsets") /* REG_ERANGEX */
1377 };
1378 \f
1379 /* Avoiding alloca during matching, to placate r_alloc. */
1380
1381 /* Define MATCH_MAY_ALLOCATE unless we need to make sure that the
1382 searching and matching functions should not call alloca. On some
1383 systems, alloca is implemented in terms of malloc, and if we're
1384 using the relocating allocator routines, then malloc could cause a
1385 relocation, which might (if the strings being searched are in the
1386 ralloc heap) shift the data out from underneath the regexp
1387 routines.
1388
1389 Here's another reason to avoid allocation: Emacs
1390 processes input from X in a signal handler; processing X input may
1391 call malloc; if input arrives while a matching routine is calling
1392 malloc, then we're scrod. But Emacs can't just block input while
1393 calling matching routines; then we don't notice interrupts when
1394 they come in. So, Emacs blocks input around all regexp calls
1395 except the matching calls, which it leaves unprotected, in the
1396 faith that they will not malloc. */
1397
1398 /* Normally, this is fine. */
1399 #define MATCH_MAY_ALLOCATE
1400
1401 /* The match routines may not allocate if (1) they would do it with malloc
1402 and (2) it's not safe for them to use malloc.
1403 Note that if REL_ALLOC is defined, matching would not use malloc for the
1404 failure stack, but we would still use it for the register vectors;
1405 so REL_ALLOC should not affect this. */
1406 #if defined REGEX_MALLOC && defined emacs
1407 # undef MATCH_MAY_ALLOCATE
1408 #endif
1409
1410 \f
1411 /* Failure stack declarations and macros; both re_compile_fastmap and
1412 re_match_2 use a failure stack. These have to be macros because of
1413 REGEX_ALLOCATE_STACK. */
1414
1415
1416 /* Approximate number of failure points for which to initially allocate space
1417 when matching. If this number is exceeded, we allocate more
1418 space, so it is not a hard limit. */
1419 #ifndef INIT_FAILURE_ALLOC
1420 # define INIT_FAILURE_ALLOC 20
1421 #endif
1422
1423 /* Roughly the maximum number of failure points on the stack. Would be
1424 exactly that if always used TYPICAL_FAILURE_SIZE items each time we failed.
1425 This is a variable only so users of regex can assign to it; we never
1426 change it ourselves. We always multiply it by TYPICAL_FAILURE_SIZE
1427 before using it, so it should probably be a byte-count instead. */
1428 # if defined MATCH_MAY_ALLOCATE
1429 /* Note that 4400 was enough to cause a crash on Alpha OSF/1,
1430 whose default stack limit is 2mb. In order for a larger
1431 value to work reliably, you have to try to make it accord
1432 with the process stack limit. */
1433 size_t re_max_failures = 40000;
1434 # else
1435 size_t re_max_failures = 4000;
1436 # endif
1437
1438 union fail_stack_elt
1439 {
1440 re_char *pointer;
1441 /* This should be the biggest `int' that's no bigger than a pointer. */
1442 long integer;
1443 };
1444
1445 typedef union fail_stack_elt fail_stack_elt_t;
1446
1447 typedef struct
1448 {
1449 fail_stack_elt_t *stack;
1450 size_t size;
1451 size_t avail; /* Offset of next open position. */
1452 size_t frame; /* Offset of the cur constructed frame. */
1453 } fail_stack_type;
1454
1455 #define FAIL_STACK_EMPTY() (fail_stack.frame == 0)
1456 #define FAIL_STACK_FULL() (fail_stack.avail == fail_stack.size)
1457
1458
1459 /* Define macros to initialize and free the failure stack.
1460 Do `return -2' if the alloc fails. */
1461
1462 #ifdef MATCH_MAY_ALLOCATE
1463 # define INIT_FAIL_STACK() \
1464 do { \
1465 fail_stack.stack = (fail_stack_elt_t *) \
1466 REGEX_ALLOCATE_STACK (INIT_FAILURE_ALLOC * TYPICAL_FAILURE_SIZE \
1467 * sizeof (fail_stack_elt_t)); \
1468 \
1469 if (fail_stack.stack == NULL) \
1470 return -2; \
1471 \
1472 fail_stack.size = INIT_FAILURE_ALLOC; \
1473 fail_stack.avail = 0; \
1474 fail_stack.frame = 0; \
1475 } while (0)
1476
1477 # define RESET_FAIL_STACK() REGEX_FREE_STACK (fail_stack.stack)
1478 #else
1479 # define INIT_FAIL_STACK() \
1480 do { \
1481 fail_stack.avail = 0; \
1482 fail_stack.frame = 0; \
1483 } while (0)
1484
1485 # define RESET_FAIL_STACK() ((void)0)
1486 #endif
1487
1488
1489 /* Double the size of FAIL_STACK, up to a limit
1490 which allows approximately `re_max_failures' items.
1491
1492 Return 1 if succeeds, and 0 if either ran out of memory
1493 allocating space for it or it was already too large.
1494
1495 REGEX_REALLOCATE_STACK requires `destination' be declared. */
1496
1497 /* Factor to increase the failure stack size by
1498 when we increase it.
1499 This used to be 2, but 2 was too wasteful
1500 because the old discarded stacks added up to as much space
1501 were as ultimate, maximum-size stack. */
1502 #define FAIL_STACK_GROWTH_FACTOR 4
1503
1504 #define GROW_FAIL_STACK(fail_stack) \
1505 (((fail_stack).size * sizeof (fail_stack_elt_t) \
1506 >= re_max_failures * TYPICAL_FAILURE_SIZE) \
1507 ? 0 \
1508 : ((fail_stack).stack \
1509 = (fail_stack_elt_t *) \
1510 REGEX_REALLOCATE_STACK ((fail_stack).stack, \
1511 (fail_stack).size * sizeof (fail_stack_elt_t), \
1512 MIN (re_max_failures * TYPICAL_FAILURE_SIZE, \
1513 ((fail_stack).size * sizeof (fail_stack_elt_t) \
1514 * FAIL_STACK_GROWTH_FACTOR))), \
1515 \
1516 (fail_stack).stack == NULL \
1517 ? 0 \
1518 : ((fail_stack).size \
1519 = (MIN (re_max_failures * TYPICAL_FAILURE_SIZE, \
1520 ((fail_stack).size * sizeof (fail_stack_elt_t) \
1521 * FAIL_STACK_GROWTH_FACTOR)) \
1522 / sizeof (fail_stack_elt_t)), \
1523 1)))
1524
1525
1526 /* Push a pointer value onto the failure stack.
1527 Assumes the variable `fail_stack'. Probably should only
1528 be called from within `PUSH_FAILURE_POINT'. */
1529 #define PUSH_FAILURE_POINTER(item) \
1530 fail_stack.stack[fail_stack.avail++].pointer = (item)
1531
1532 /* This pushes an integer-valued item onto the failure stack.
1533 Assumes the variable `fail_stack'. Probably should only
1534 be called from within `PUSH_FAILURE_POINT'. */
1535 #define PUSH_FAILURE_INT(item) \
1536 fail_stack.stack[fail_stack.avail++].integer = (item)
1537
1538 /* Push a fail_stack_elt_t value onto the failure stack.
1539 Assumes the variable `fail_stack'. Probably should only
1540 be called from within `PUSH_FAILURE_POINT'. */
1541 #define PUSH_FAILURE_ELT(item) \
1542 fail_stack.stack[fail_stack.avail++] = (item)
1543
1544 /* These three POP... operations complement the three PUSH... operations.
1545 All assume that `fail_stack' is nonempty. */
1546 #define POP_FAILURE_POINTER() fail_stack.stack[--fail_stack.avail].pointer
1547 #define POP_FAILURE_INT() fail_stack.stack[--fail_stack.avail].integer
1548 #define POP_FAILURE_ELT() fail_stack.stack[--fail_stack.avail]
1549
1550 /* Individual items aside from the registers. */
1551 #define NUM_NONREG_ITEMS 3
1552
1553 /* Used to examine the stack (to detect infinite loops). */
1554 #define FAILURE_PAT(h) fail_stack.stack[(h) - 1].pointer
1555 #define FAILURE_STR(h) (fail_stack.stack[(h) - 2].pointer)
1556 #define NEXT_FAILURE_HANDLE(h) fail_stack.stack[(h) - 3].integer
1557 #define TOP_FAILURE_HANDLE() fail_stack.frame
1558
1559
1560 #define ENSURE_FAIL_STACK(space) \
1561 while (REMAINING_AVAIL_SLOTS <= space) { \
1562 if (!GROW_FAIL_STACK (fail_stack)) \
1563 return -2; \
1564 DEBUG_PRINT2 ("\n Doubled stack; size now: %d\n", (fail_stack).size);\
1565 DEBUG_PRINT2 (" slots available: %d\n", REMAINING_AVAIL_SLOTS);\
1566 }
1567
1568 /* Push register NUM onto the stack. */
1569 #define PUSH_FAILURE_REG(num) \
1570 do { \
1571 char *destination; \
1572 ENSURE_FAIL_STACK(3); \
1573 DEBUG_PRINT4 (" Push reg %d (spanning %p -> %p)\n", \
1574 num, regstart[num], regend[num]); \
1575 PUSH_FAILURE_POINTER (regstart[num]); \
1576 PUSH_FAILURE_POINTER (regend[num]); \
1577 PUSH_FAILURE_INT (num); \
1578 } while (0)
1579
1580 /* Change the counter's value to VAL, but make sure that it will
1581 be reset when backtracking. */
1582 #define PUSH_NUMBER(ptr,val) \
1583 do { \
1584 char *destination; \
1585 int c; \
1586 ENSURE_FAIL_STACK(3); \
1587 EXTRACT_NUMBER (c, ptr); \
1588 DEBUG_PRINT4 (" Push number %p = %d -> %d\n", ptr, c, val); \
1589 PUSH_FAILURE_INT (c); \
1590 PUSH_FAILURE_POINTER (ptr); \
1591 PUSH_FAILURE_INT (-1); \
1592 STORE_NUMBER (ptr, val); \
1593 } while (0)
1594
1595 /* Pop a saved register off the stack. */
1596 #define POP_FAILURE_REG_OR_COUNT() \
1597 do { \
1598 int reg = POP_FAILURE_INT (); \
1599 if (reg == -1) \
1600 { \
1601 /* It's a counter. */ \
1602 /* Here, we discard `const', making re_match non-reentrant. */ \
1603 unsigned char *ptr = (unsigned char*) POP_FAILURE_POINTER (); \
1604 reg = POP_FAILURE_INT (); \
1605 STORE_NUMBER (ptr, reg); \
1606 DEBUG_PRINT3 (" Pop counter %p = %d\n", ptr, reg); \
1607 } \
1608 else \
1609 { \
1610 regend[reg] = POP_FAILURE_POINTER (); \
1611 regstart[reg] = POP_FAILURE_POINTER (); \
1612 DEBUG_PRINT4 (" Pop reg %d (spanning %p -> %p)\n", \
1613 reg, regstart[reg], regend[reg]); \
1614 } \
1615 } while (0)
1616
1617 /* Check that we are not stuck in an infinite loop. */
1618 #define CHECK_INFINITE_LOOP(pat_cur, string_place) \
1619 do { \
1620 int failure = TOP_FAILURE_HANDLE (); \
1621 /* Check for infinite matching loops */ \
1622 while (failure > 0 \
1623 && (FAILURE_STR (failure) == string_place \
1624 || FAILURE_STR (failure) == NULL)) \
1625 { \
1626 assert (FAILURE_PAT (failure) >= bufp->buffer \
1627 && FAILURE_PAT (failure) <= bufp->buffer + bufp->used); \
1628 if (FAILURE_PAT (failure) == pat_cur) \
1629 { \
1630 cycle = 1; \
1631 break; \
1632 } \
1633 DEBUG_PRINT2 (" Other pattern: %p\n", FAILURE_PAT (failure)); \
1634 failure = NEXT_FAILURE_HANDLE(failure); \
1635 } \
1636 DEBUG_PRINT2 (" Other string: %p\n", FAILURE_STR (failure)); \
1637 } while (0)
1638
1639 /* Push the information about the state we will need
1640 if we ever fail back to it.
1641
1642 Requires variables fail_stack, regstart, regend and
1643 num_regs be declared. GROW_FAIL_STACK requires `destination' be
1644 declared.
1645
1646 Does `return FAILURE_CODE' if runs out of memory. */
1647
1648 #define PUSH_FAILURE_POINT(pattern, string_place) \
1649 do { \
1650 char *destination; \
1651 /* Must be int, so when we don't save any registers, the arithmetic \
1652 of 0 + -1 isn't done as unsigned. */ \
1653 \
1654 DEBUG_STATEMENT (nfailure_points_pushed++); \
1655 DEBUG_PRINT1 ("\nPUSH_FAILURE_POINT:\n"); \
1656 DEBUG_PRINT2 (" Before push, next avail: %d\n", (fail_stack).avail); \
1657 DEBUG_PRINT2 (" size: %d\n", (fail_stack).size);\
1658 \
1659 ENSURE_FAIL_STACK (NUM_NONREG_ITEMS); \
1660 \
1661 DEBUG_PRINT1 ("\n"); \
1662 \
1663 DEBUG_PRINT2 (" Push frame index: %d\n", fail_stack.frame); \
1664 PUSH_FAILURE_INT (fail_stack.frame); \
1665 \
1666 DEBUG_PRINT2 (" Push string %p: `", string_place); \
1667 DEBUG_PRINT_DOUBLE_STRING (string_place, string1, size1, string2, size2);\
1668 DEBUG_PRINT1 ("'\n"); \
1669 PUSH_FAILURE_POINTER (string_place); \
1670 \
1671 DEBUG_PRINT2 (" Push pattern %p: ", pattern); \
1672 DEBUG_PRINT_COMPILED_PATTERN (bufp, pattern, pend); \
1673 PUSH_FAILURE_POINTER (pattern); \
1674 \
1675 /* Close the frame by moving the frame pointer past it. */ \
1676 fail_stack.frame = fail_stack.avail; \
1677 } while (0)
1678
1679 /* Estimate the size of data pushed by a typical failure stack entry.
1680 An estimate is all we need, because all we use this for
1681 is to choose a limit for how big to make the failure stack. */
1682 /* BEWARE, the value `20' is hard-coded in emacs.c:main(). */
1683 #define TYPICAL_FAILURE_SIZE 20
1684
1685 /* How many items can still be added to the stack without overflowing it. */
1686 #define REMAINING_AVAIL_SLOTS ((fail_stack).size - (fail_stack).avail)
1687
1688
1689 /* Pops what PUSH_FAIL_STACK pushes.
1690
1691 We restore into the parameters, all of which should be lvalues:
1692 STR -- the saved data position.
1693 PAT -- the saved pattern position.
1694 REGSTART, REGEND -- arrays of string positions.
1695
1696 Also assumes the variables `fail_stack' and (if debugging), `bufp',
1697 `pend', `string1', `size1', `string2', and `size2'. */
1698
1699 #define POP_FAILURE_POINT(str, pat) \
1700 do { \
1701 assert (!FAIL_STACK_EMPTY ()); \
1702 \
1703 /* Remove failure points and point to how many regs pushed. */ \
1704 DEBUG_PRINT1 ("POP_FAILURE_POINT:\n"); \
1705 DEBUG_PRINT2 (" Before pop, next avail: %d\n", fail_stack.avail); \
1706 DEBUG_PRINT2 (" size: %d\n", fail_stack.size); \
1707 \
1708 /* Pop the saved registers. */ \
1709 while (fail_stack.frame < fail_stack.avail) \
1710 POP_FAILURE_REG_OR_COUNT (); \
1711 \
1712 pat = POP_FAILURE_POINTER (); \
1713 DEBUG_PRINT2 (" Popping pattern %p: ", pat); \
1714 DEBUG_PRINT_COMPILED_PATTERN (bufp, pat, pend); \
1715 \
1716 /* If the saved string location is NULL, it came from an \
1717 on_failure_keep_string_jump opcode, and we want to throw away the \
1718 saved NULL, thus retaining our current position in the string. */ \
1719 str = POP_FAILURE_POINTER (); \
1720 DEBUG_PRINT2 (" Popping string %p: `", str); \
1721 DEBUG_PRINT_DOUBLE_STRING (str, string1, size1, string2, size2); \
1722 DEBUG_PRINT1 ("'\n"); \
1723 \
1724 fail_stack.frame = POP_FAILURE_INT (); \
1725 DEBUG_PRINT2 (" Popping frame index: %d\n", fail_stack.frame); \
1726 \
1727 assert (fail_stack.avail >= 0); \
1728 assert (fail_stack.frame <= fail_stack.avail); \
1729 \
1730 DEBUG_STATEMENT (nfailure_points_popped++); \
1731 } while (0) /* POP_FAILURE_POINT */
1732
1733
1734 \f
1735 /* Registers are set to a sentinel when they haven't yet matched. */
1736 #define REG_UNSET(e) ((e) == NULL)
1737 \f
1738 /* Subroutine declarations and macros for regex_compile. */
1739
1740 static reg_errcode_t regex_compile _RE_ARGS ((re_char *pattern, size_t size,
1741 reg_syntax_t syntax,
1742 struct re_pattern_buffer *bufp));
1743 static void store_op1 _RE_ARGS ((re_opcode_t op, unsigned char *loc, int arg));
1744 static void store_op2 _RE_ARGS ((re_opcode_t op, unsigned char *loc,
1745 int arg1, int arg2));
1746 static void insert_op1 _RE_ARGS ((re_opcode_t op, unsigned char *loc,
1747 int arg, unsigned char *end));
1748 static void insert_op2 _RE_ARGS ((re_opcode_t op, unsigned char *loc,
1749 int arg1, int arg2, unsigned char *end));
1750 static boolean at_begline_loc_p _RE_ARGS ((re_char *pattern,
1751 re_char *p,
1752 reg_syntax_t syntax));
1753 static boolean at_endline_loc_p _RE_ARGS ((re_char *p,
1754 re_char *pend,
1755 reg_syntax_t syntax));
1756 static re_char *skip_one_char _RE_ARGS ((re_char *p));
1757 static int analyse_first _RE_ARGS ((re_char *p, re_char *pend,
1758 char *fastmap, const int multibyte));
1759
1760 /* Fetch the next character in the uncompiled pattern, with no
1761 translation. */
1762 #define PATFETCH(c) \
1763 do { \
1764 int len; \
1765 if (p == pend) return REG_EEND; \
1766 c = RE_STRING_CHAR_AND_LENGTH (p, pend - p, len, multibyte); \
1767 p += len; \
1768 } while (0)
1769
1770
1771 /* If `translate' is non-null, return translate[D], else just D. We
1772 cast the subscript to translate because some data is declared as
1773 `char *', to avoid warnings when a string constant is passed. But
1774 when we use a character as a subscript we must make it unsigned. */
1775 #ifndef TRANSLATE
1776 # define TRANSLATE(d) \
1777 (RE_TRANSLATE_P (translate) ? RE_TRANSLATE (translate, (d)) : (d))
1778 #endif
1779
1780
1781 /* Macros for outputting the compiled pattern into `buffer'. */
1782
1783 /* If the buffer isn't allocated when it comes in, use this. */
1784 #define INIT_BUF_SIZE 32
1785
1786 /* Make sure we have at least N more bytes of space in buffer. */
1787 #define GET_BUFFER_SPACE(n) \
1788 while ((size_t) (b - bufp->buffer + (n)) > bufp->allocated) \
1789 EXTEND_BUFFER ()
1790
1791 /* Make sure we have one more byte of buffer space and then add C to it. */
1792 #define BUF_PUSH(c) \
1793 do { \
1794 GET_BUFFER_SPACE (1); \
1795 *b++ = (unsigned char) (c); \
1796 } while (0)
1797
1798
1799 /* Ensure we have two more bytes of buffer space and then append C1 and C2. */
1800 #define BUF_PUSH_2(c1, c2) \
1801 do { \
1802 GET_BUFFER_SPACE (2); \
1803 *b++ = (unsigned char) (c1); \
1804 *b++ = (unsigned char) (c2); \
1805 } while (0)
1806
1807
1808 /* As with BUF_PUSH_2, except for three bytes. */
1809 #define BUF_PUSH_3(c1, c2, c3) \
1810 do { \
1811 GET_BUFFER_SPACE (3); \
1812 *b++ = (unsigned char) (c1); \
1813 *b++ = (unsigned char) (c2); \
1814 *b++ = (unsigned char) (c3); \
1815 } while (0)
1816
1817
1818 /* Store a jump with opcode OP at LOC to location TO. We store a
1819 relative address offset by the three bytes the jump itself occupies. */
1820 #define STORE_JUMP(op, loc, to) \
1821 store_op1 (op, loc, (to) - (loc) - 3)
1822
1823 /* Likewise, for a two-argument jump. */
1824 #define STORE_JUMP2(op, loc, to, arg) \
1825 store_op2 (op, loc, (to) - (loc) - 3, arg)
1826
1827 /* Like `STORE_JUMP', but for inserting. Assume `b' is the buffer end. */
1828 #define INSERT_JUMP(op, loc, to) \
1829 insert_op1 (op, loc, (to) - (loc) - 3, b)
1830
1831 /* Like `STORE_JUMP2', but for inserting. Assume `b' is the buffer end. */
1832 #define INSERT_JUMP2(op, loc, to, arg) \
1833 insert_op2 (op, loc, (to) - (loc) - 3, arg, b)
1834
1835
1836 /* This is not an arbitrary limit: the arguments which represent offsets
1837 into the pattern are two bytes long. So if 2^15 bytes turns out to
1838 be too small, many things would have to change. */
1839 # define MAX_BUF_SIZE (1L << 15)
1840
1841 #if 0 /* This is when we thought it could be 2^16 bytes. */
1842 /* Any other compiler which, like MSC, has allocation limit below 2^16
1843 bytes will have to use approach similar to what was done below for
1844 MSC and drop MAX_BUF_SIZE a bit. Otherwise you may end up
1845 reallocating to 0 bytes. Such thing is not going to work too well.
1846 You have been warned!! */
1847 #if defined _MSC_VER && !defined WIN32
1848 /* Microsoft C 16-bit versions limit malloc to approx 65512 bytes. */
1849 # define MAX_BUF_SIZE 65500L
1850 #else
1851 # define MAX_BUF_SIZE (1L << 16)
1852 #endif
1853 #endif /* 0 */
1854
1855 /* Extend the buffer by twice its current size via realloc and
1856 reset the pointers that pointed into the old block to point to the
1857 correct places in the new one. If extending the buffer results in it
1858 being larger than MAX_BUF_SIZE, then flag memory exhausted. */
1859 #if __BOUNDED_POINTERS__
1860 # define SET_HIGH_BOUND(P) (__ptrhigh (P) = __ptrlow (P) + bufp->allocated)
1861 # define MOVE_BUFFER_POINTER(P) \
1862 (__ptrlow (P) += incr, SET_HIGH_BOUND (P), __ptrvalue (P) += incr)
1863 # define ELSE_EXTEND_BUFFER_HIGH_BOUND \
1864 else \
1865 { \
1866 SET_HIGH_BOUND (b); \
1867 SET_HIGH_BOUND (begalt); \
1868 if (fixup_alt_jump) \
1869 SET_HIGH_BOUND (fixup_alt_jump); \
1870 if (laststart) \
1871 SET_HIGH_BOUND (laststart); \
1872 if (pending_exact) \
1873 SET_HIGH_BOUND (pending_exact); \
1874 }
1875 #else
1876 # define MOVE_BUFFER_POINTER(P) (P) += incr
1877 # define ELSE_EXTEND_BUFFER_HIGH_BOUND
1878 #endif
1879 #define EXTEND_BUFFER() \
1880 do { \
1881 re_char *old_buffer = bufp->buffer; \
1882 if (bufp->allocated == MAX_BUF_SIZE) \
1883 return REG_ESIZE; \
1884 bufp->allocated <<= 1; \
1885 if (bufp->allocated > MAX_BUF_SIZE) \
1886 bufp->allocated = MAX_BUF_SIZE; \
1887 RETALLOC (bufp->buffer, bufp->allocated, unsigned char); \
1888 if (bufp->buffer == NULL) \
1889 return REG_ESPACE; \
1890 /* If the buffer moved, move all the pointers into it. */ \
1891 if (old_buffer != bufp->buffer) \
1892 { \
1893 int incr = bufp->buffer - old_buffer; \
1894 MOVE_BUFFER_POINTER (b); \
1895 MOVE_BUFFER_POINTER (begalt); \
1896 if (fixup_alt_jump) \
1897 MOVE_BUFFER_POINTER (fixup_alt_jump); \
1898 if (laststart) \
1899 MOVE_BUFFER_POINTER (laststart); \
1900 if (pending_exact) \
1901 MOVE_BUFFER_POINTER (pending_exact); \
1902 } \
1903 ELSE_EXTEND_BUFFER_HIGH_BOUND \
1904 } while (0)
1905
1906
1907 /* Since we have one byte reserved for the register number argument to
1908 {start,stop}_memory, the maximum number of groups we can report
1909 things about is what fits in that byte. */
1910 #define MAX_REGNUM 255
1911
1912 /* But patterns can have more than `MAX_REGNUM' registers. We just
1913 ignore the excess. */
1914 typedef int regnum_t;
1915
1916
1917 /* Macros for the compile stack. */
1918
1919 /* Since offsets can go either forwards or backwards, this type needs to
1920 be able to hold values from -(MAX_BUF_SIZE - 1) to MAX_BUF_SIZE - 1. */
1921 /* int may be not enough when sizeof(int) == 2. */
1922 typedef long pattern_offset_t;
1923
1924 typedef struct
1925 {
1926 pattern_offset_t begalt_offset;
1927 pattern_offset_t fixup_alt_jump;
1928 pattern_offset_t laststart_offset;
1929 regnum_t regnum;
1930 } compile_stack_elt_t;
1931
1932
1933 typedef struct
1934 {
1935 compile_stack_elt_t *stack;
1936 unsigned size;
1937 unsigned avail; /* Offset of next open position. */
1938 } compile_stack_type;
1939
1940
1941 #define INIT_COMPILE_STACK_SIZE 32
1942
1943 #define COMPILE_STACK_EMPTY (compile_stack.avail == 0)
1944 #define COMPILE_STACK_FULL (compile_stack.avail == compile_stack.size)
1945
1946 /* The next available element. */
1947 #define COMPILE_STACK_TOP (compile_stack.stack[compile_stack.avail])
1948
1949 /* Explicit quit checking is only used on NTemacs and whenever we
1950 use polling to process input events. */
1951 #if defined emacs && (defined WINDOWSNT || defined SYNC_INPUT) && defined QUIT
1952 extern int immediate_quit;
1953 # define IMMEDIATE_QUIT_CHECK \
1954 do { \
1955 if (immediate_quit) QUIT; \
1956 } while (0)
1957 #else
1958 # define IMMEDIATE_QUIT_CHECK ((void)0)
1959 #endif
1960 \f
1961 /* Structure to manage work area for range table. */
1962 struct range_table_work_area
1963 {
1964 int *table; /* actual work area. */
1965 int allocated; /* allocated size for work area in bytes. */
1966 int used; /* actually used size in words. */
1967 int bits; /* flag to record character classes */
1968 };
1969
1970 /* Make sure that WORK_AREA can hold more N multibyte characters.
1971 This is used only in set_image_of_range and set_image_of_range_1.
1972 It expects WORK_AREA to be a pointer.
1973 If it can't get the space, it returns from the surrounding function. */
1974
1975 #define EXTEND_RANGE_TABLE(work_area, n) \
1976 do { \
1977 if (((work_area).used + (n)) * sizeof (int) > (work_area).allocated) \
1978 { \
1979 extend_range_table_work_area (&work_area); \
1980 if ((work_area).table == 0) \
1981 return (REG_ESPACE); \
1982 } \
1983 } while (0)
1984
1985 #define SET_RANGE_TABLE_WORK_AREA_BIT(work_area, bit) \
1986 (work_area).bits |= (bit)
1987
1988 /* Bits used to implement the multibyte-part of the various character classes
1989 such as [:alnum:] in a charset's range table. */
1990 #define BIT_WORD 0x1
1991 #define BIT_LOWER 0x2
1992 #define BIT_PUNCT 0x4
1993 #define BIT_SPACE 0x8
1994 #define BIT_UPPER 0x10
1995 #define BIT_MULTIBYTE 0x20
1996
1997 /* Set a range (RANGE_START, RANGE_END) to WORK_AREA. */
1998 #define SET_RANGE_TABLE_WORK_AREA(work_area, range_start, range_end) \
1999 do { \
2000 EXTEND_RANGE_TABLE ((work_area), 2); \
2001 (work_area).table[(work_area).used++] = (range_start); \
2002 (work_area).table[(work_area).used++] = (range_end); \
2003 } while (0)
2004
2005 /* Free allocated memory for WORK_AREA. */
2006 #define FREE_RANGE_TABLE_WORK_AREA(work_area) \
2007 do { \
2008 if ((work_area).table) \
2009 free ((work_area).table); \
2010 } while (0)
2011
2012 #define CLEAR_RANGE_TABLE_WORK_USED(work_area) ((work_area).used = 0, (work_area).bits = 0)
2013 #define RANGE_TABLE_WORK_USED(work_area) ((work_area).used)
2014 #define RANGE_TABLE_WORK_BITS(work_area) ((work_area).bits)
2015 #define RANGE_TABLE_WORK_ELT(work_area, i) ((work_area).table[i])
2016 \f
2017
2018 /* Set the bit for character C in a list. */
2019 #define SET_LIST_BIT(c) (b[((c)) / BYTEWIDTH] |= 1 << ((c) % BYTEWIDTH))
2020
2021
2022 #ifdef emacs
2023
2024 /* Store characters in the range FROM to TO in the bitmap at B (for
2025 ASCII and unibyte characters) and WORK_AREA (for multibyte
2026 characters) while translating them and paying attention to the
2027 continuity of translated characters.
2028
2029 Implementation note: It is better to implement these fairly big
2030 macros by a function, but it's not that easy because macros called
2031 in this macro assume various local variables already declared. */
2032
2033 /* Both FROM and TO are ASCII characters. */
2034
2035 #define SETUP_ASCII_RANGE(work_area, FROM, TO) \
2036 do { \
2037 int C0, C1; \
2038 \
2039 for (C0 = (FROM); C0 <= (TO); C0++) \
2040 { \
2041 C1 = TRANSLATE (C0); \
2042 if (! ASCII_CHAR_P (C1)) \
2043 { \
2044 SET_RANGE_TABLE_WORK_AREA ((work_area), C1, C1); \
2045 if ((C1 = RE_CHAR_TO_UNIBYTE (C1)) < 0) \
2046 C1 = C0; \
2047 } \
2048 SET_LIST_BIT (C1); \
2049 } \
2050 } while (0)
2051
2052
2053 /* Both FROM and TO are unibyte characters (0x80..0xFF). */
2054
2055 #define SETUP_UNIBYTE_RANGE(work_area, FROM, TO) \
2056 do { \
2057 int C0, C1, C2, I; \
2058 int USED = RANGE_TABLE_WORK_USED (work_area); \
2059 \
2060 for (C0 = (FROM); C0 <= (TO); C0++) \
2061 { \
2062 C1 = RE_CHAR_TO_MULTIBYTE (C0); \
2063 if (CHAR_BYTE8_P (C1)) \
2064 SET_LIST_BIT (C0); \
2065 else \
2066 { \
2067 C2 = TRANSLATE (C1); \
2068 if (C2 == C1 \
2069 || (C1 = RE_CHAR_TO_UNIBYTE (C2)) < 0) \
2070 C1 = C0; \
2071 SET_LIST_BIT (C1); \
2072 for (I = RANGE_TABLE_WORK_USED (work_area) - 2; I >= USED; I -= 2) \
2073 { \
2074 int from = RANGE_TABLE_WORK_ELT (work_area, I); \
2075 int to = RANGE_TABLE_WORK_ELT (work_area, I + 1); \
2076 \
2077 if (C2 >= from - 1 && C2 <= to + 1) \
2078 { \
2079 if (C2 == from - 1) \
2080 RANGE_TABLE_WORK_ELT (work_area, I)--; \
2081 else if (C2 == to + 1) \
2082 RANGE_TABLE_WORK_ELT (work_area, I + 1)++; \
2083 break; \
2084 } \
2085 } \
2086 if (I < USED) \
2087 SET_RANGE_TABLE_WORK_AREA ((work_area), C2, C2); \
2088 } \
2089 } \
2090 } while (0)
2091
2092
2093 /* Both FROM and TO are mulitbyte characters. */
2094
2095 #define SETUP_MULTIBYTE_RANGE(work_area, FROM, TO) \
2096 do { \
2097 int C0, C1, C2, I, USED = RANGE_TABLE_WORK_USED (work_area); \
2098 \
2099 SET_RANGE_TABLE_WORK_AREA ((work_area), (FROM), (TO)); \
2100 for (C0 = (FROM); C0 <= (TO); C0++) \
2101 { \
2102 C1 = TRANSLATE (C0); \
2103 if ((C2 = RE_CHAR_TO_UNIBYTE (C1)) >= 0 \
2104 || (C1 != C0 && (C2 = RE_CHAR_TO_UNIBYTE (C0)) >= 0)) \
2105 SET_LIST_BIT (C2); \
2106 if (C1 >= (FROM) && C1 <= (TO)) \
2107 continue; \
2108 for (I = RANGE_TABLE_WORK_USED (work_area) - 2; I >= USED; I -= 2) \
2109 { \
2110 int from = RANGE_TABLE_WORK_ELT (work_area, I); \
2111 int to = RANGE_TABLE_WORK_ELT (work_area, I + 1); \
2112 \
2113 if (C1 >= from - 1 && C1 <= to + 1) \
2114 { \
2115 if (C1 == from - 1) \
2116 RANGE_TABLE_WORK_ELT (work_area, I)--; \
2117 else if (C1 == to + 1) \
2118 RANGE_TABLE_WORK_ELT (work_area, I + 1)++; \
2119 break; \
2120 } \
2121 } \
2122 if (I < USED) \
2123 SET_RANGE_TABLE_WORK_AREA ((work_area), C1, C1); \
2124 } \
2125 } while (0)
2126
2127 #endif /* emacs */
2128
2129 /* Get the next unsigned number in the uncompiled pattern. */
2130 #define GET_UNSIGNED_NUMBER(num) \
2131 do { \
2132 if (p == pend) \
2133 FREE_STACK_RETURN (REG_EBRACE); \
2134 else \
2135 { \
2136 PATFETCH (c); \
2137 while ('0' <= c && c <= '9') \
2138 { \
2139 int prev; \
2140 if (num < 0) \
2141 num = 0; \
2142 prev = num; \
2143 num = num * 10 + c - '0'; \
2144 if (num / 10 != prev) \
2145 FREE_STACK_RETURN (REG_BADBR); \
2146 if (p == pend) \
2147 FREE_STACK_RETURN (REG_EBRACE); \
2148 PATFETCH (c); \
2149 } \
2150 } \
2151 } while (0)
2152 \f
2153 #if ! WIDE_CHAR_SUPPORT
2154
2155 /* Map a string to the char class it names (if any). */
2156 re_wctype_t
2157 re_wctype (str)
2158 re_char *str;
2159 {
2160 const char *string = str;
2161 if (STREQ (string, "alnum")) return RECC_ALNUM;
2162 else if (STREQ (string, "alpha")) return RECC_ALPHA;
2163 else if (STREQ (string, "word")) return RECC_WORD;
2164 else if (STREQ (string, "ascii")) return RECC_ASCII;
2165 else if (STREQ (string, "nonascii")) return RECC_NONASCII;
2166 else if (STREQ (string, "graph")) return RECC_GRAPH;
2167 else if (STREQ (string, "lower")) return RECC_LOWER;
2168 else if (STREQ (string, "print")) return RECC_PRINT;
2169 else if (STREQ (string, "punct")) return RECC_PUNCT;
2170 else if (STREQ (string, "space")) return RECC_SPACE;
2171 else if (STREQ (string, "upper")) return RECC_UPPER;
2172 else if (STREQ (string, "unibyte")) return RECC_UNIBYTE;
2173 else if (STREQ (string, "multibyte")) return RECC_MULTIBYTE;
2174 else if (STREQ (string, "digit")) return RECC_DIGIT;
2175 else if (STREQ (string, "xdigit")) return RECC_XDIGIT;
2176 else if (STREQ (string, "cntrl")) return RECC_CNTRL;
2177 else if (STREQ (string, "blank")) return RECC_BLANK;
2178 else return 0;
2179 }
2180
2181 /* True if CH is in the char class CC. */
2182 boolean
2183 re_iswctype (ch, cc)
2184 int ch;
2185 re_wctype_t cc;
2186 {
2187 switch (cc)
2188 {
2189 case RECC_ALNUM: return ISALNUM (ch);
2190 case RECC_ALPHA: return ISALPHA (ch);
2191 case RECC_BLANK: return ISBLANK (ch);
2192 case RECC_CNTRL: return ISCNTRL (ch);
2193 case RECC_DIGIT: return ISDIGIT (ch);
2194 case RECC_GRAPH: return ISGRAPH (ch);
2195 case RECC_LOWER: return ISLOWER (ch);
2196 case RECC_PRINT: return ISPRINT (ch);
2197 case RECC_PUNCT: return ISPUNCT (ch);
2198 case RECC_SPACE: return ISSPACE (ch);
2199 case RECC_UPPER: return ISUPPER (ch);
2200 case RECC_XDIGIT: return ISXDIGIT (ch);
2201 case RECC_ASCII: return IS_REAL_ASCII (ch);
2202 case RECC_NONASCII: return !IS_REAL_ASCII (ch);
2203 case RECC_UNIBYTE: return ISUNIBYTE (ch);
2204 case RECC_MULTIBYTE: return !ISUNIBYTE (ch);
2205 case RECC_WORD: return ISWORD (ch);
2206 case RECC_ERROR: return false;
2207 default:
2208 abort();
2209 }
2210 }
2211
2212 /* Return a bit-pattern to use in the range-table bits to match multibyte
2213 chars of class CC. */
2214 static int
2215 re_wctype_to_bit (cc)
2216 re_wctype_t cc;
2217 {
2218 switch (cc)
2219 {
2220 case RECC_NONASCII: case RECC_PRINT: case RECC_GRAPH:
2221 case RECC_MULTIBYTE: return BIT_MULTIBYTE;
2222 case RECC_ALPHA: case RECC_ALNUM: case RECC_WORD: return BIT_WORD;
2223 case RECC_LOWER: return BIT_LOWER;
2224 case RECC_UPPER: return BIT_UPPER;
2225 case RECC_PUNCT: return BIT_PUNCT;
2226 case RECC_SPACE: return BIT_SPACE;
2227 case RECC_ASCII: case RECC_DIGIT: case RECC_XDIGIT: case RECC_CNTRL:
2228 case RECC_BLANK: case RECC_UNIBYTE: case RECC_ERROR: return 0;
2229 default:
2230 abort();
2231 }
2232 }
2233 #endif
2234 \f
2235 /* Filling in the work area of a range. */
2236
2237 /* Actually extend the space in WORK_AREA. */
2238
2239 static void
2240 extend_range_table_work_area (work_area)
2241 struct range_table_work_area *work_area;
2242 {
2243 work_area->allocated += 16 * sizeof (int);
2244 if (work_area->table)
2245 work_area->table
2246 = (int *) realloc (work_area->table, work_area->allocated);
2247 else
2248 work_area->table
2249 = (int *) malloc (work_area->allocated);
2250 }
2251
2252 #if 0
2253 #ifdef emacs
2254
2255 /* Carefully find the ranges of codes that are equivalent
2256 under case conversion to the range start..end when passed through
2257 TRANSLATE. Handle the case where non-letters can come in between
2258 two upper-case letters (which happens in Latin-1).
2259 Also handle the case of groups of more than 2 case-equivalent chars.
2260
2261 The basic method is to look at consecutive characters and see
2262 if they can form a run that can be handled as one.
2263
2264 Returns -1 if successful, REG_ESPACE if ran out of space. */
2265
2266 static int
2267 set_image_of_range_1 (work_area, start, end, translate)
2268 RE_TRANSLATE_TYPE translate;
2269 struct range_table_work_area *work_area;
2270 re_wchar_t start, end;
2271 {
2272 /* `one_case' indicates a character, or a run of characters,
2273 each of which is an isolate (no case-equivalents).
2274 This includes all ASCII non-letters.
2275
2276 `two_case' indicates a character, or a run of characters,
2277 each of which has two case-equivalent forms.
2278 This includes all ASCII letters.
2279
2280 `strange' indicates a character that has more than one
2281 case-equivalent. */
2282
2283 enum case_type {one_case, two_case, strange};
2284
2285 /* Describe the run that is in progress,
2286 which the next character can try to extend.
2287 If run_type is strange, that means there really is no run.
2288 If run_type is one_case, then run_start...run_end is the run.
2289 If run_type is two_case, then the run is run_start...run_end,
2290 and the case-equivalents end at run_eqv_end. */
2291
2292 enum case_type run_type = strange;
2293 int run_start, run_end, run_eqv_end;
2294
2295 Lisp_Object eqv_table;
2296
2297 if (!RE_TRANSLATE_P (translate))
2298 {
2299 EXTEND_RANGE_TABLE (work_area, 2);
2300 work_area->table[work_area->used++] = (start);
2301 work_area->table[work_area->used++] = (end);
2302 return -1;
2303 }
2304
2305 eqv_table = XCHAR_TABLE (translate)->extras[2];
2306
2307 for (; start <= end; start++)
2308 {
2309 enum case_type this_type;
2310 int eqv = RE_TRANSLATE (eqv_table, start);
2311 int minchar, maxchar;
2312
2313 /* Classify this character */
2314 if (eqv == start)
2315 this_type = one_case;
2316 else if (RE_TRANSLATE (eqv_table, eqv) == start)
2317 this_type = two_case;
2318 else
2319 this_type = strange;
2320
2321 if (start < eqv)
2322 minchar = start, maxchar = eqv;
2323 else
2324 minchar = eqv, maxchar = start;
2325
2326 /* Can this character extend the run in progress? */
2327 if (this_type == strange || this_type != run_type
2328 || !(minchar == run_end + 1
2329 && (run_type == two_case
2330 ? maxchar == run_eqv_end + 1 : 1)))
2331 {
2332 /* No, end the run.
2333 Record each of its equivalent ranges. */
2334 if (run_type == one_case)
2335 {
2336 EXTEND_RANGE_TABLE (work_area, 2);
2337 work_area->table[work_area->used++] = run_start;
2338 work_area->table[work_area->used++] = run_end;
2339 }
2340 else if (run_type == two_case)
2341 {
2342 EXTEND_RANGE_TABLE (work_area, 4);
2343 work_area->table[work_area->used++] = run_start;
2344 work_area->table[work_area->used++] = run_end;
2345 work_area->table[work_area->used++]
2346 = RE_TRANSLATE (eqv_table, run_start);
2347 work_area->table[work_area->used++]
2348 = RE_TRANSLATE (eqv_table, run_end);
2349 }
2350 run_type = strange;
2351 }
2352
2353 if (this_type == strange)
2354 {
2355 /* For a strange character, add each of its equivalents, one
2356 by one. Don't start a range. */
2357 do
2358 {
2359 EXTEND_RANGE_TABLE (work_area, 2);
2360 work_area->table[work_area->used++] = eqv;
2361 work_area->table[work_area->used++] = eqv;
2362 eqv = RE_TRANSLATE (eqv_table, eqv);
2363 }
2364 while (eqv != start);
2365 }
2366
2367 /* Add this char to the run, or start a new run. */
2368 else if (run_type == strange)
2369 {
2370 /* Initialize a new range. */
2371 run_type = this_type;
2372 run_start = start;
2373 run_end = start;
2374 run_eqv_end = RE_TRANSLATE (eqv_table, run_end);
2375 }
2376 else
2377 {
2378 /* Extend a running range. */
2379 run_end = minchar;
2380 run_eqv_end = RE_TRANSLATE (eqv_table, run_end);
2381 }
2382 }
2383
2384 /* If a run is still in progress at the end, finish it now
2385 by recording its equivalent ranges. */
2386 if (run_type == one_case)
2387 {
2388 EXTEND_RANGE_TABLE (work_area, 2);
2389 work_area->table[work_area->used++] = run_start;
2390 work_area->table[work_area->used++] = run_end;
2391 }
2392 else if (run_type == two_case)
2393 {
2394 EXTEND_RANGE_TABLE (work_area, 4);
2395 work_area->table[work_area->used++] = run_start;
2396 work_area->table[work_area->used++] = run_end;
2397 work_area->table[work_area->used++]
2398 = RE_TRANSLATE (eqv_table, run_start);
2399 work_area->table[work_area->used++]
2400 = RE_TRANSLATE (eqv_table, run_end);
2401 }
2402
2403 return -1;
2404 }
2405
2406 #endif /* emacs */
2407
2408 /* Record the the image of the range start..end when passed through
2409 TRANSLATE. This is not necessarily TRANSLATE(start)..TRANSLATE(end)
2410 and is not even necessarily contiguous.
2411 Normally we approximate it with the smallest contiguous range that contains
2412 all the chars we need. However, for Latin-1 we go to extra effort
2413 to do a better job.
2414
2415 This function is not called for ASCII ranges.
2416
2417 Returns -1 if successful, REG_ESPACE if ran out of space. */
2418
2419 static int
2420 set_image_of_range (work_area, start, end, translate)
2421 RE_TRANSLATE_TYPE translate;
2422 struct range_table_work_area *work_area;
2423 re_wchar_t start, end;
2424 {
2425 re_wchar_t cmin, cmax;
2426
2427 #ifdef emacs
2428 /* For Latin-1 ranges, use set_image_of_range_1
2429 to get proper handling of ranges that include letters and nonletters.
2430 For a range that includes the whole of Latin-1, this is not necessary.
2431 For other character sets, we don't bother to get this right. */
2432 if (RE_TRANSLATE_P (translate) && start < 04400
2433 && !(start < 04200 && end >= 04377))
2434 {
2435 int newend;
2436 int tem;
2437 newend = end;
2438 if (newend > 04377)
2439 newend = 04377;
2440 tem = set_image_of_range_1 (work_area, start, newend, translate);
2441 if (tem > 0)
2442 return tem;
2443
2444 start = 04400;
2445 if (end < 04400)
2446 return -1;
2447 }
2448 #endif
2449
2450 EXTEND_RANGE_TABLE (work_area, 2);
2451 work_area->table[work_area->used++] = (start);
2452 work_area->table[work_area->used++] = (end);
2453
2454 cmin = -1, cmax = -1;
2455
2456 if (RE_TRANSLATE_P (translate))
2457 {
2458 int ch;
2459
2460 for (ch = start; ch <= end; ch++)
2461 {
2462 re_wchar_t c = TRANSLATE (ch);
2463 if (! (start <= c && c <= end))
2464 {
2465 if (cmin == -1)
2466 cmin = c, cmax = c;
2467 else
2468 {
2469 cmin = MIN (cmin, c);
2470 cmax = MAX (cmax, c);
2471 }
2472 }
2473 }
2474
2475 if (cmin != -1)
2476 {
2477 EXTEND_RANGE_TABLE (work_area, 2);
2478 work_area->table[work_area->used++] = (cmin);
2479 work_area->table[work_area->used++] = (cmax);
2480 }
2481 }
2482
2483 return -1;
2484 }
2485 #endif /* 0 */
2486 \f
2487 #ifndef MATCH_MAY_ALLOCATE
2488
2489 /* If we cannot allocate large objects within re_match_2_internal,
2490 we make the fail stack and register vectors global.
2491 The fail stack, we grow to the maximum size when a regexp
2492 is compiled.
2493 The register vectors, we adjust in size each time we
2494 compile a regexp, according to the number of registers it needs. */
2495
2496 static fail_stack_type fail_stack;
2497
2498 /* Size with which the following vectors are currently allocated.
2499 That is so we can make them bigger as needed,
2500 but never make them smaller. */
2501 static int regs_allocated_size;
2502
2503 static re_char ** regstart, ** regend;
2504 static re_char **best_regstart, **best_regend;
2505
2506 /* Make the register vectors big enough for NUM_REGS registers,
2507 but don't make them smaller. */
2508
2509 static
2510 regex_grow_registers (num_regs)
2511 int num_regs;
2512 {
2513 if (num_regs > regs_allocated_size)
2514 {
2515 RETALLOC_IF (regstart, num_regs, re_char *);
2516 RETALLOC_IF (regend, num_regs, re_char *);
2517 RETALLOC_IF (best_regstart, num_regs, re_char *);
2518 RETALLOC_IF (best_regend, num_regs, re_char *);
2519
2520 regs_allocated_size = num_regs;
2521 }
2522 }
2523
2524 #endif /* not MATCH_MAY_ALLOCATE */
2525 \f
2526 static boolean group_in_compile_stack _RE_ARGS ((compile_stack_type
2527 compile_stack,
2528 regnum_t regnum));
2529
2530 /* `regex_compile' compiles PATTERN (of length SIZE) according to SYNTAX.
2531 Returns one of error codes defined in `regex.h', or zero for success.
2532
2533 Assumes the `allocated' (and perhaps `buffer') and `translate'
2534 fields are set in BUFP on entry.
2535
2536 If it succeeds, results are put in BUFP (if it returns an error, the
2537 contents of BUFP are undefined):
2538 `buffer' is the compiled pattern;
2539 `syntax' is set to SYNTAX;
2540 `used' is set to the length of the compiled pattern;
2541 `fastmap_accurate' is zero;
2542 `re_nsub' is the number of subexpressions in PATTERN;
2543 `not_bol' and `not_eol' are zero;
2544
2545 The `fastmap' field is neither examined nor set. */
2546
2547 /* Insert the `jump' from the end of last alternative to "here".
2548 The space for the jump has already been allocated. */
2549 #define FIXUP_ALT_JUMP() \
2550 do { \
2551 if (fixup_alt_jump) \
2552 STORE_JUMP (jump, fixup_alt_jump, b); \
2553 } while (0)
2554
2555
2556 /* Return, freeing storage we allocated. */
2557 #define FREE_STACK_RETURN(value) \
2558 do { \
2559 FREE_RANGE_TABLE_WORK_AREA (range_table_work); \
2560 free (compile_stack.stack); \
2561 return value; \
2562 } while (0)
2563
2564 static reg_errcode_t
2565 regex_compile (pattern, size, syntax, bufp)
2566 re_char *pattern;
2567 size_t size;
2568 reg_syntax_t syntax;
2569 struct re_pattern_buffer *bufp;
2570 {
2571 /* We fetch characters from PATTERN here. */
2572 register re_wchar_t c, c1;
2573
2574 /* A random temporary spot in PATTERN. */
2575 re_char *p1;
2576
2577 /* Points to the end of the buffer, where we should append. */
2578 register unsigned char *b;
2579
2580 /* Keeps track of unclosed groups. */
2581 compile_stack_type compile_stack;
2582
2583 /* Points to the current (ending) position in the pattern. */
2584 #ifdef AIX
2585 /* `const' makes AIX compiler fail. */
2586 unsigned char *p = pattern;
2587 #else
2588 re_char *p = pattern;
2589 #endif
2590 re_char *pend = pattern + size;
2591
2592 /* How to translate the characters in the pattern. */
2593 RE_TRANSLATE_TYPE translate = bufp->translate;
2594
2595 /* Address of the count-byte of the most recently inserted `exactn'
2596 command. This makes it possible to tell if a new exact-match
2597 character can be added to that command or if the character requires
2598 a new `exactn' command. */
2599 unsigned char *pending_exact = 0;
2600
2601 /* Address of start of the most recently finished expression.
2602 This tells, e.g., postfix * where to find the start of its
2603 operand. Reset at the beginning of groups and alternatives. */
2604 unsigned char *laststart = 0;
2605
2606 /* Address of beginning of regexp, or inside of last group. */
2607 unsigned char *begalt;
2608
2609 /* Place in the uncompiled pattern (i.e., the {) to
2610 which to go back if the interval is invalid. */
2611 re_char *beg_interval;
2612
2613 /* Address of the place where a forward jump should go to the end of
2614 the containing expression. Each alternative of an `or' -- except the
2615 last -- ends with a forward jump of this sort. */
2616 unsigned char *fixup_alt_jump = 0;
2617
2618 /* Work area for range table of charset. */
2619 struct range_table_work_area range_table_work;
2620
2621 /* If the object matched can contain multibyte characters. */
2622 const boolean multibyte = RE_MULTIBYTE_P (bufp);
2623
2624 /* If a target of matching can contain multibyte characters. */
2625 const boolean target_multibyte = RE_TARGET_MULTIBYTE_P (bufp);
2626
2627 /* Nonzero if we have pushed down into a subpattern. */
2628 int in_subpattern = 0;
2629
2630 /* These hold the values of p, pattern, and pend from the main
2631 pattern when we have pushed into a subpattern. */
2632 re_char *main_p;
2633 re_char *main_pattern;
2634 re_char *main_pend;
2635
2636 #ifdef DEBUG
2637 debug++;
2638 DEBUG_PRINT1 ("\nCompiling pattern: ");
2639 if (debug > 0)
2640 {
2641 unsigned debug_count;
2642
2643 for (debug_count = 0; debug_count < size; debug_count++)
2644 putchar (pattern[debug_count]);
2645 putchar ('\n');
2646 }
2647 #endif /* DEBUG */
2648
2649 /* Initialize the compile stack. */
2650 compile_stack.stack = TALLOC (INIT_COMPILE_STACK_SIZE, compile_stack_elt_t);
2651 if (compile_stack.stack == NULL)
2652 return REG_ESPACE;
2653
2654 compile_stack.size = INIT_COMPILE_STACK_SIZE;
2655 compile_stack.avail = 0;
2656
2657 range_table_work.table = 0;
2658 range_table_work.allocated = 0;
2659
2660 /* Initialize the pattern buffer. */
2661 bufp->syntax = syntax;
2662 bufp->fastmap_accurate = 0;
2663 bufp->not_bol = bufp->not_eol = 0;
2664 bufp->used_syntax = 0;
2665
2666 /* Set `used' to zero, so that if we return an error, the pattern
2667 printer (for debugging) will think there's no pattern. We reset it
2668 at the end. */
2669 bufp->used = 0;
2670
2671 /* Always count groups, whether or not bufp->no_sub is set. */
2672 bufp->re_nsub = 0;
2673
2674 #if !defined emacs && !defined SYNTAX_TABLE
2675 /* Initialize the syntax table. */
2676 init_syntax_once ();
2677 #endif
2678
2679 if (bufp->allocated == 0)
2680 {
2681 if (bufp->buffer)
2682 { /* If zero allocated, but buffer is non-null, try to realloc
2683 enough space. This loses if buffer's address is bogus, but
2684 that is the user's responsibility. */
2685 RETALLOC (bufp->buffer, INIT_BUF_SIZE, unsigned char);
2686 }
2687 else
2688 { /* Caller did not allocate a buffer. Do it for them. */
2689 bufp->buffer = TALLOC (INIT_BUF_SIZE, unsigned char);
2690 }
2691 if (!bufp->buffer) FREE_STACK_RETURN (REG_ESPACE);
2692
2693 bufp->allocated = INIT_BUF_SIZE;
2694 }
2695
2696 begalt = b = bufp->buffer;
2697
2698 /* Loop through the uncompiled pattern until we're at the end. */
2699 while (1)
2700 {
2701 if (p == pend)
2702 {
2703 /* If this is the end of an included regexp,
2704 pop back to the main regexp and try again. */
2705 if (in_subpattern)
2706 {
2707 in_subpattern = 0;
2708 pattern = main_pattern;
2709 p = main_p;
2710 pend = main_pend;
2711 continue;
2712 }
2713 /* If this is the end of the main regexp, we are done. */
2714 break;
2715 }
2716
2717 PATFETCH (c);
2718
2719 switch (c)
2720 {
2721 case ' ':
2722 {
2723 re_char *p1 = p;
2724
2725 /* If there's no special whitespace regexp, treat
2726 spaces normally. And don't try to do this recursively. */
2727 if (!whitespace_regexp || in_subpattern)
2728 goto normal_char;
2729
2730 /* Peek past following spaces. */
2731 while (p1 != pend)
2732 {
2733 if (*p1 != ' ')
2734 break;
2735 p1++;
2736 }
2737 /* If the spaces are followed by a repetition op,
2738 treat them normally. */
2739 if (p1 != pend
2740 && (*p1 == '*' || *p1 == '+' || *p1 == '?'
2741 || (*p1 == '\\' && p1 + 1 != pend && p1[1] == '{')))
2742 goto normal_char;
2743
2744 /* Replace the spaces with the whitespace regexp. */
2745 in_subpattern = 1;
2746 main_p = p1;
2747 main_pend = pend;
2748 main_pattern = pattern;
2749 p = pattern = whitespace_regexp;
2750 pend = p + strlen (p);
2751 break;
2752 }
2753
2754 case '^':
2755 {
2756 if ( /* If at start of pattern, it's an operator. */
2757 p == pattern + 1
2758 /* If context independent, it's an operator. */
2759 || syntax & RE_CONTEXT_INDEP_ANCHORS
2760 /* Otherwise, depends on what's come before. */
2761 || at_begline_loc_p (pattern, p, syntax))
2762 BUF_PUSH ((syntax & RE_NO_NEWLINE_ANCHOR) ? begbuf : begline);
2763 else
2764 goto normal_char;
2765 }
2766 break;
2767
2768
2769 case '$':
2770 {
2771 if ( /* If at end of pattern, it's an operator. */
2772 p == pend
2773 /* If context independent, it's an operator. */
2774 || syntax & RE_CONTEXT_INDEP_ANCHORS
2775 /* Otherwise, depends on what's next. */
2776 || at_endline_loc_p (p, pend, syntax))
2777 BUF_PUSH ((syntax & RE_NO_NEWLINE_ANCHOR) ? endbuf : endline);
2778 else
2779 goto normal_char;
2780 }
2781 break;
2782
2783
2784 case '+':
2785 case '?':
2786 if ((syntax & RE_BK_PLUS_QM)
2787 || (syntax & RE_LIMITED_OPS))
2788 goto normal_char;
2789 handle_plus:
2790 case '*':
2791 /* If there is no previous pattern... */
2792 if (!laststart)
2793 {
2794 if (syntax & RE_CONTEXT_INVALID_OPS)
2795 FREE_STACK_RETURN (REG_BADRPT);
2796 else if (!(syntax & RE_CONTEXT_INDEP_OPS))
2797 goto normal_char;
2798 }
2799
2800 {
2801 /* 1 means zero (many) matches is allowed. */
2802 boolean zero_times_ok = 0, many_times_ok = 0;
2803 boolean greedy = 1;
2804
2805 /* If there is a sequence of repetition chars, collapse it
2806 down to just one (the right one). We can't combine
2807 interval operators with these because of, e.g., `a{2}*',
2808 which should only match an even number of `a's. */
2809
2810 for (;;)
2811 {
2812 if ((syntax & RE_FRUGAL)
2813 && c == '?' && (zero_times_ok || many_times_ok))
2814 greedy = 0;
2815 else
2816 {
2817 zero_times_ok |= c != '+';
2818 many_times_ok |= c != '?';
2819 }
2820
2821 if (p == pend)
2822 break;
2823 else if (*p == '*'
2824 || (!(syntax & RE_BK_PLUS_QM)
2825 && (*p == '+' || *p == '?')))
2826 ;
2827 else if (syntax & RE_BK_PLUS_QM && *p == '\\')
2828 {
2829 if (p+1 == pend)
2830 FREE_STACK_RETURN (REG_EESCAPE);
2831 if (p[1] == '+' || p[1] == '?')
2832 PATFETCH (c); /* Gobble up the backslash. */
2833 else
2834 break;
2835 }
2836 else
2837 break;
2838 /* If we get here, we found another repeat character. */
2839 PATFETCH (c);
2840 }
2841
2842 /* Star, etc. applied to an empty pattern is equivalent
2843 to an empty pattern. */
2844 if (!laststart || laststart == b)
2845 break;
2846
2847 /* Now we know whether or not zero matches is allowed
2848 and also whether or not two or more matches is allowed. */
2849 if (greedy)
2850 {
2851 if (many_times_ok)
2852 {
2853 boolean simple = skip_one_char (laststart) == b;
2854 unsigned int startoffset = 0;
2855 re_opcode_t ofj =
2856 /* Check if the loop can match the empty string. */
2857 (simple || !analyse_first (laststart, b, NULL, 0))
2858 ? on_failure_jump : on_failure_jump_loop;
2859 assert (skip_one_char (laststart) <= b);
2860
2861 if (!zero_times_ok && simple)
2862 { /* Since simple * loops can be made faster by using
2863 on_failure_keep_string_jump, we turn simple P+
2864 into PP* if P is simple. */
2865 unsigned char *p1, *p2;
2866 startoffset = b - laststart;
2867 GET_BUFFER_SPACE (startoffset);
2868 p1 = b; p2 = laststart;
2869 while (p2 < p1)
2870 *b++ = *p2++;
2871 zero_times_ok = 1;
2872 }
2873
2874 GET_BUFFER_SPACE (6);
2875 if (!zero_times_ok)
2876 /* A + loop. */
2877 STORE_JUMP (ofj, b, b + 6);
2878 else
2879 /* Simple * loops can use on_failure_keep_string_jump
2880 depending on what follows. But since we don't know
2881 that yet, we leave the decision up to
2882 on_failure_jump_smart. */
2883 INSERT_JUMP (simple ? on_failure_jump_smart : ofj,
2884 laststart + startoffset, b + 6);
2885 b += 3;
2886 STORE_JUMP (jump, b, laststart + startoffset);
2887 b += 3;
2888 }
2889 else
2890 {
2891 /* A simple ? pattern. */
2892 assert (zero_times_ok);
2893 GET_BUFFER_SPACE (3);
2894 INSERT_JUMP (on_failure_jump, laststart, b + 3);
2895 b += 3;
2896 }
2897 }
2898 else /* not greedy */
2899 { /* I wish the greedy and non-greedy cases could be merged. */
2900
2901 GET_BUFFER_SPACE (7); /* We might use less. */
2902 if (many_times_ok)
2903 {
2904 boolean emptyp = analyse_first (laststart, b, NULL, 0);
2905
2906 /* The non-greedy multiple match looks like
2907 a repeat..until: we only need a conditional jump
2908 at the end of the loop. */
2909 if (emptyp) BUF_PUSH (no_op);
2910 STORE_JUMP (emptyp ? on_failure_jump_nastyloop
2911 : on_failure_jump, b, laststart);
2912 b += 3;
2913 if (zero_times_ok)
2914 {
2915 /* The repeat...until naturally matches one or more.
2916 To also match zero times, we need to first jump to
2917 the end of the loop (its conditional jump). */
2918 INSERT_JUMP (jump, laststart, b);
2919 b += 3;
2920 }
2921 }
2922 else
2923 {
2924 /* non-greedy a?? */
2925 INSERT_JUMP (jump, laststart, b + 3);
2926 b += 3;
2927 INSERT_JUMP (on_failure_jump, laststart, laststart + 6);
2928 b += 3;
2929 }
2930 }
2931 }
2932 pending_exact = 0;
2933 break;
2934
2935
2936 case '.':
2937 laststart = b;
2938 BUF_PUSH (anychar);
2939 break;
2940
2941
2942 case '[':
2943 {
2944 CLEAR_RANGE_TABLE_WORK_USED (range_table_work);
2945
2946 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2947
2948 /* Ensure that we have enough space to push a charset: the
2949 opcode, the length count, and the bitset; 34 bytes in all. */
2950 GET_BUFFER_SPACE (34);
2951
2952 laststart = b;
2953
2954 /* We test `*p == '^' twice, instead of using an if
2955 statement, so we only need one BUF_PUSH. */
2956 BUF_PUSH (*p == '^' ? charset_not : charset);
2957 if (*p == '^')
2958 p++;
2959
2960 /* Remember the first position in the bracket expression. */
2961 p1 = p;
2962
2963 /* Push the number of bytes in the bitmap. */
2964 BUF_PUSH ((1 << BYTEWIDTH) / BYTEWIDTH);
2965
2966 /* Clear the whole map. */
2967 bzero (b, (1 << BYTEWIDTH) / BYTEWIDTH);
2968
2969 /* charset_not matches newline according to a syntax bit. */
2970 if ((re_opcode_t) b[-2] == charset_not
2971 && (syntax & RE_HAT_LISTS_NOT_NEWLINE))
2972 SET_LIST_BIT ('\n');
2973
2974 /* Read in characters and ranges, setting map bits. */
2975 for (;;)
2976 {
2977 boolean escaped_char = false;
2978 const unsigned char *p2 = p;
2979 re_wchar_t ch, c2;
2980
2981 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2982
2983 /* Don't translate yet. The range TRANSLATE(X..Y) cannot
2984 always be determined from TRANSLATE(X) and TRANSLATE(Y)
2985 So the translation is done later in a loop. Example:
2986 (let ((case-fold-search t)) (string-match "[A-_]" "A")) */
2987 PATFETCH (c);
2988
2989 /* \ might escape characters inside [...] and [^...]. */
2990 if ((syntax & RE_BACKSLASH_ESCAPE_IN_LISTS) && c == '\\')
2991 {
2992 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
2993
2994 PATFETCH (c);
2995 escaped_char = true;
2996 }
2997 else
2998 {
2999 /* Could be the end of the bracket expression. If it's
3000 not (i.e., when the bracket expression is `[]' so
3001 far), the ']' character bit gets set way below. */
3002 if (c == ']' && p2 != p1)
3003 break;
3004 }
3005
3006 /* See if we're at the beginning of a possible character
3007 class. */
3008
3009 if (!escaped_char &&
3010 syntax & RE_CHAR_CLASSES && c == '[' && *p == ':')
3011 {
3012 /* Leave room for the null. */
3013 unsigned char str[CHAR_CLASS_MAX_LENGTH + 1];
3014 const unsigned char *class_beg;
3015
3016 PATFETCH (c);
3017 c1 = 0;
3018 class_beg = p;
3019
3020 /* If pattern is `[[:'. */
3021 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
3022
3023 for (;;)
3024 {
3025 PATFETCH (c);
3026 if ((c == ':' && *p == ']') || p == pend)
3027 break;
3028 if (c1 < CHAR_CLASS_MAX_LENGTH)
3029 str[c1++] = c;
3030 else
3031 /* This is in any case an invalid class name. */
3032 str[0] = '\0';
3033 }
3034 str[c1] = '\0';
3035
3036 /* If isn't a word bracketed by `[:' and `:]':
3037 undo the ending character, the letters, and
3038 leave the leading `:' and `[' (but set bits for
3039 them). */
3040 if (c == ':' && *p == ']')
3041 {
3042 re_wctype_t cc;
3043 int limit;
3044
3045 cc = re_wctype (str);
3046
3047 if (cc == 0)
3048 FREE_STACK_RETURN (REG_ECTYPE);
3049
3050 /* Throw away the ] at the end of the character
3051 class. */
3052 PATFETCH (c);
3053
3054 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
3055
3056 #ifndef emacs
3057 for (ch = 0; ch < (1 << BYTEWIDTH); ++ch)
3058 if (re_iswctype (btowc (ch), cc))
3059 {
3060 c = TRANSLATE (ch);
3061 if (c < (1 << BYTEWIDTH))
3062 SET_LIST_BIT (c);
3063 }
3064 #else /* emacs */
3065 /* Most character classes in a multibyte match
3066 just set a flag. Exceptions are is_blank,
3067 is_digit, is_cntrl, and is_xdigit, since
3068 they can only match ASCII characters. We
3069 don't need to handle them for multibyte.
3070 They are distinguished by a negative wctype. */
3071
3072 for (ch = 0; ch < 256; ++ch)
3073 {
3074 c = RE_CHAR_TO_MULTIBYTE (ch);
3075 if (! CHAR_BYTE8_P (c)
3076 && re_iswctype (c, cc))
3077 {
3078 SET_LIST_BIT (ch);
3079 c1 = TRANSLATE (c);
3080 if (c1 == c)
3081 continue;
3082 if (ASCII_CHAR_P (c1))
3083 SET_LIST_BIT (c1);
3084 else if ((c1 = RE_CHAR_TO_UNIBYTE (c1)) >= 0)
3085 SET_LIST_BIT (c1);
3086 }
3087 }
3088 SET_RANGE_TABLE_WORK_AREA_BIT
3089 (range_table_work, re_wctype_to_bit (cc));
3090 #endif /* emacs */
3091 /* In most cases the matching rule for char classes
3092 only uses the syntax table for multibyte chars,
3093 so that the content of the syntax-table it is not
3094 hardcoded in the range_table. SPACE and WORD are
3095 the two exceptions. */
3096 if ((1 << cc) & ((1 << RECC_SPACE) | (1 << RECC_WORD)))
3097 bufp->used_syntax = 1;
3098
3099 /* Repeat the loop. */
3100 continue;
3101 }
3102 else
3103 {
3104 /* Go back to right after the "[:". */
3105 p = class_beg;
3106 SET_LIST_BIT ('[');
3107
3108 /* Because the `:' may starts the range, we
3109 can't simply set bit and repeat the loop.
3110 Instead, just set it to C and handle below. */
3111 c = ':';
3112 }
3113 }
3114
3115 if (p < pend && p[0] == '-' && p[1] != ']')
3116 {
3117
3118 /* Discard the `-'. */
3119 PATFETCH (c1);
3120
3121 /* Fetch the character which ends the range. */
3122 PATFETCH (c1);
3123 #ifdef emacs
3124 if (CHAR_BYTE8_P (c1)
3125 && ! ASCII_CHAR_P (c) && ! CHAR_BYTE8_P (c))
3126 /* Treat the range from a multibyte character to
3127 raw-byte character as empty. */
3128 c = c1 + 1;
3129 #endif /* emacs */
3130 }
3131 else
3132 /* Range from C to C. */
3133 c1 = c;
3134
3135 if (c > c1)
3136 {
3137 if (syntax & RE_NO_EMPTY_RANGES)
3138 FREE_STACK_RETURN (REG_ERANGEX);
3139 /* Else, repeat the loop. */
3140 }
3141 else
3142 {
3143 #ifndef emacs
3144 /* Set the range into bitmap */
3145 for (; c <= c1; c++)
3146 {
3147 ch = TRANSLATE (c);
3148 if (ch < (1 << BYTEWIDTH))
3149 SET_LIST_BIT (ch);
3150 }
3151 #else /* emacs */
3152 if (c < 128)
3153 {
3154 ch = MIN (127, c1);
3155 SETUP_ASCII_RANGE (range_table_work, c, ch);
3156 c = ch + 1;
3157 if (CHAR_BYTE8_P (c1))
3158 c = BYTE8_TO_CHAR (128);
3159 }
3160 if (c <= c1)
3161 {
3162 if (CHAR_BYTE8_P (c))
3163 {
3164 c = CHAR_TO_BYTE8 (c);
3165 c1 = CHAR_TO_BYTE8 (c1);
3166 for (; c <= c1; c++)
3167 SET_LIST_BIT (c);
3168 }
3169 else if (multibyte)
3170 {
3171 SETUP_MULTIBYTE_RANGE (range_table_work, c, c1);
3172 }
3173 else
3174 {
3175 SETUP_UNIBYTE_RANGE (range_table_work, c, c1);
3176 }
3177 }
3178 #endif /* emacs */
3179 }
3180 }
3181
3182 /* Discard any (non)matching list bytes that are all 0 at the
3183 end of the map. Decrease the map-length byte too. */
3184 while ((int) b[-1] > 0 && b[b[-1] - 1] == 0)
3185 b[-1]--;
3186 b += b[-1];
3187
3188 /* Build real range table from work area. */
3189 if (RANGE_TABLE_WORK_USED (range_table_work)
3190 || RANGE_TABLE_WORK_BITS (range_table_work))
3191 {
3192 int i;
3193 int used = RANGE_TABLE_WORK_USED (range_table_work);
3194
3195 /* Allocate space for COUNT + RANGE_TABLE. Needs two
3196 bytes for flags, two for COUNT, and three bytes for
3197 each character. */
3198 GET_BUFFER_SPACE (4 + used * 3);
3199
3200 /* Indicate the existence of range table. */
3201 laststart[1] |= 0x80;
3202
3203 /* Store the character class flag bits into the range table.
3204 If not in emacs, these flag bits are always 0. */
3205 *b++ = RANGE_TABLE_WORK_BITS (range_table_work) & 0xff;
3206 *b++ = RANGE_TABLE_WORK_BITS (range_table_work) >> 8;
3207
3208 STORE_NUMBER_AND_INCR (b, used / 2);
3209 for (i = 0; i < used; i++)
3210 STORE_CHARACTER_AND_INCR
3211 (b, RANGE_TABLE_WORK_ELT (range_table_work, i));
3212 }
3213 }
3214 break;
3215
3216
3217 case '(':
3218 if (syntax & RE_NO_BK_PARENS)
3219 goto handle_open;
3220 else
3221 goto normal_char;
3222
3223
3224 case ')':
3225 if (syntax & RE_NO_BK_PARENS)
3226 goto handle_close;
3227 else
3228 goto normal_char;
3229
3230
3231 case '\n':
3232 if (syntax & RE_NEWLINE_ALT)
3233 goto handle_alt;
3234 else
3235 goto normal_char;
3236
3237
3238 case '|':
3239 if (syntax & RE_NO_BK_VBAR)
3240 goto handle_alt;
3241 else
3242 goto normal_char;
3243
3244
3245 case '{':
3246 if (syntax & RE_INTERVALS && syntax & RE_NO_BK_BRACES)
3247 goto handle_interval;
3248 else
3249 goto normal_char;
3250
3251
3252 case '\\':
3253 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
3254
3255 /* Do not translate the character after the \, so that we can
3256 distinguish, e.g., \B from \b, even if we normally would
3257 translate, e.g., B to b. */
3258 PATFETCH (c);
3259
3260 switch (c)
3261 {
3262 case '(':
3263 if (syntax & RE_NO_BK_PARENS)
3264 goto normal_backslash;
3265
3266 handle_open:
3267 {
3268 int shy = 0;
3269 regnum_t regnum = 0;
3270 if (p+1 < pend)
3271 {
3272 /* Look for a special (?...) construct */
3273 if ((syntax & RE_SHY_GROUPS) && *p == '?')
3274 {
3275 PATFETCH (c); /* Gobble up the '?'. */
3276 while (!shy)
3277 {
3278 PATFETCH (c);
3279 switch (c)
3280 {
3281 case ':': shy = 1; break;
3282 case '0':
3283 /* An explicitly specified regnum must start
3284 with non-0. */
3285 if (regnum == 0)
3286 FREE_STACK_RETURN (REG_BADPAT);
3287 case '1': case '2': case '3': case '4':
3288 case '5': case '6': case '7': case '8': case '9':
3289 regnum = 10*regnum + (c - '0'); break;
3290 default:
3291 /* Only (?:...) is supported right now. */
3292 FREE_STACK_RETURN (REG_BADPAT);
3293 }
3294 }
3295 }
3296 }
3297
3298 if (!shy)
3299 regnum = ++bufp->re_nsub;
3300 else if (regnum)
3301 { /* It's actually not shy, but explicitly numbered. */
3302 shy = 0;
3303 if (regnum > bufp->re_nsub)
3304 bufp->re_nsub = regnum;
3305 else if (regnum > bufp->re_nsub
3306 /* Ideally, we'd want to check that the specified
3307 group can't have matched (i.e. all subgroups
3308 using the same regnum are in other branches of
3309 OR patterns), but we don't currently keep track
3310 of enough info to do that easily. */
3311 || group_in_compile_stack (compile_stack, regnum))
3312 FREE_STACK_RETURN (REG_BADPAT);
3313 }
3314 else
3315 /* It's really shy. */
3316 regnum = - bufp->re_nsub;
3317
3318 if (COMPILE_STACK_FULL)
3319 {
3320 RETALLOC (compile_stack.stack, compile_stack.size << 1,
3321 compile_stack_elt_t);
3322 if (compile_stack.stack == NULL) return REG_ESPACE;
3323
3324 compile_stack.size <<= 1;
3325 }
3326
3327 /* These are the values to restore when we hit end of this
3328 group. They are all relative offsets, so that if the
3329 whole pattern moves because of realloc, they will still
3330 be valid. */
3331 COMPILE_STACK_TOP.begalt_offset = begalt - bufp->buffer;
3332 COMPILE_STACK_TOP.fixup_alt_jump
3333 = fixup_alt_jump ? fixup_alt_jump - bufp->buffer + 1 : 0;
3334 COMPILE_STACK_TOP.laststart_offset = b - bufp->buffer;
3335 COMPILE_STACK_TOP.regnum = regnum;
3336
3337 /* Do not push a start_memory for groups beyond the last one
3338 we can represent in the compiled pattern. */
3339 if (regnum <= MAX_REGNUM && regnum > 0)
3340 BUF_PUSH_2 (start_memory, regnum);
3341
3342 compile_stack.avail++;
3343
3344 fixup_alt_jump = 0;
3345 laststart = 0;
3346 begalt = b;
3347 /* If we've reached MAX_REGNUM groups, then this open
3348 won't actually generate any code, so we'll have to
3349 clear pending_exact explicitly. */
3350 pending_exact = 0;
3351 break;
3352 }
3353
3354 case ')':
3355 if (syntax & RE_NO_BK_PARENS) goto normal_backslash;
3356
3357 if (COMPILE_STACK_EMPTY)
3358 {
3359 if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD)
3360 goto normal_backslash;
3361 else
3362 FREE_STACK_RETURN (REG_ERPAREN);
3363 }
3364
3365 handle_close:
3366 FIXUP_ALT_JUMP ();
3367
3368 /* See similar code for backslashed left paren above. */
3369 if (COMPILE_STACK_EMPTY)
3370 {
3371 if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD)
3372 goto normal_char;
3373 else
3374 FREE_STACK_RETURN (REG_ERPAREN);
3375 }
3376
3377 /* Since we just checked for an empty stack above, this
3378 ``can't happen''. */
3379 assert (compile_stack.avail != 0);
3380 {
3381 /* We don't just want to restore into `regnum', because
3382 later groups should continue to be numbered higher,
3383 as in `(ab)c(de)' -- the second group is #2. */
3384 regnum_t regnum;
3385
3386 compile_stack.avail--;
3387 begalt = bufp->buffer + COMPILE_STACK_TOP.begalt_offset;
3388 fixup_alt_jump
3389 = COMPILE_STACK_TOP.fixup_alt_jump
3390 ? bufp->buffer + COMPILE_STACK_TOP.fixup_alt_jump - 1
3391 : 0;
3392 laststart = bufp->buffer + COMPILE_STACK_TOP.laststart_offset;
3393 regnum = COMPILE_STACK_TOP.regnum;
3394 /* If we've reached MAX_REGNUM groups, then this open
3395 won't actually generate any code, so we'll have to
3396 clear pending_exact explicitly. */
3397 pending_exact = 0;
3398
3399 /* We're at the end of the group, so now we know how many
3400 groups were inside this one. */
3401 if (regnum <= MAX_REGNUM && regnum > 0)
3402 BUF_PUSH_2 (stop_memory, regnum);
3403 }
3404 break;
3405
3406
3407 case '|': /* `\|'. */
3408 if (syntax & RE_LIMITED_OPS || syntax & RE_NO_BK_VBAR)
3409 goto normal_backslash;
3410 handle_alt:
3411 if (syntax & RE_LIMITED_OPS)
3412 goto normal_char;
3413
3414 /* Insert before the previous alternative a jump which
3415 jumps to this alternative if the former fails. */
3416 GET_BUFFER_SPACE (3);
3417 INSERT_JUMP (on_failure_jump, begalt, b + 6);
3418 pending_exact = 0;
3419 b += 3;
3420
3421 /* The alternative before this one has a jump after it
3422 which gets executed if it gets matched. Adjust that
3423 jump so it will jump to this alternative's analogous
3424 jump (put in below, which in turn will jump to the next
3425 (if any) alternative's such jump, etc.). The last such
3426 jump jumps to the correct final destination. A picture:
3427 _____ _____
3428 | | | |
3429 | v | v
3430 a | b | c
3431
3432 If we are at `b', then fixup_alt_jump right now points to a
3433 three-byte space after `a'. We'll put in the jump, set
3434 fixup_alt_jump to right after `b', and leave behind three
3435 bytes which we'll fill in when we get to after `c'. */
3436
3437 FIXUP_ALT_JUMP ();
3438
3439 /* Mark and leave space for a jump after this alternative,
3440 to be filled in later either by next alternative or
3441 when know we're at the end of a series of alternatives. */
3442 fixup_alt_jump = b;
3443 GET_BUFFER_SPACE (3);
3444 b += 3;
3445
3446 laststart = 0;
3447 begalt = b;
3448 break;
3449
3450
3451 case '{':
3452 /* If \{ is a literal. */
3453 if (!(syntax & RE_INTERVALS)
3454 /* If we're at `\{' and it's not the open-interval
3455 operator. */
3456 || (syntax & RE_NO_BK_BRACES))
3457 goto normal_backslash;
3458
3459 handle_interval:
3460 {
3461 /* If got here, then the syntax allows intervals. */
3462
3463 /* At least (most) this many matches must be made. */
3464 int lower_bound = 0, upper_bound = -1;
3465
3466 beg_interval = p;
3467
3468 GET_UNSIGNED_NUMBER (lower_bound);
3469
3470 if (c == ',')
3471 GET_UNSIGNED_NUMBER (upper_bound);
3472 else
3473 /* Interval such as `{1}' => match exactly once. */
3474 upper_bound = lower_bound;
3475
3476 if (lower_bound < 0 || upper_bound > RE_DUP_MAX
3477 || (upper_bound >= 0 && lower_bound > upper_bound))
3478 FREE_STACK_RETURN (REG_BADBR);
3479
3480 if (!(syntax & RE_NO_BK_BRACES))
3481 {
3482 if (c != '\\')
3483 FREE_STACK_RETURN (REG_BADBR);
3484 if (p == pend)
3485 FREE_STACK_RETURN (REG_EESCAPE);
3486 PATFETCH (c);
3487 }
3488
3489 if (c != '}')
3490 FREE_STACK_RETURN (REG_BADBR);
3491
3492 /* We just parsed a valid interval. */
3493
3494 /* If it's invalid to have no preceding re. */
3495 if (!laststart)
3496 {
3497 if (syntax & RE_CONTEXT_INVALID_OPS)
3498 FREE_STACK_RETURN (REG_BADRPT);
3499 else if (syntax & RE_CONTEXT_INDEP_OPS)
3500 laststart = b;
3501 else
3502 goto unfetch_interval;
3503 }
3504
3505 if (upper_bound == 0)
3506 /* If the upper bound is zero, just drop the sub pattern
3507 altogether. */
3508 b = laststart;
3509 else if (lower_bound == 1 && upper_bound == 1)
3510 /* Just match it once: nothing to do here. */
3511 ;
3512
3513 /* Otherwise, we have a nontrivial interval. When
3514 we're all done, the pattern will look like:
3515 set_number_at <jump count> <upper bound>
3516 set_number_at <succeed_n count> <lower bound>
3517 succeed_n <after jump addr> <succeed_n count>
3518 <body of loop>
3519 jump_n <succeed_n addr> <jump count>
3520 (The upper bound and `jump_n' are omitted if
3521 `upper_bound' is 1, though.) */
3522 else
3523 { /* If the upper bound is > 1, we need to insert
3524 more at the end of the loop. */
3525 unsigned int nbytes = (upper_bound < 0 ? 3
3526 : upper_bound > 1 ? 5 : 0);
3527 unsigned int startoffset = 0;
3528
3529 GET_BUFFER_SPACE (20); /* We might use less. */
3530
3531 if (lower_bound == 0)
3532 {
3533 /* A succeed_n that starts with 0 is really a
3534 a simple on_failure_jump_loop. */
3535 INSERT_JUMP (on_failure_jump_loop, laststart,
3536 b + 3 + nbytes);
3537 b += 3;
3538 }
3539 else
3540 {
3541 /* Initialize lower bound of the `succeed_n', even
3542 though it will be set during matching by its
3543 attendant `set_number_at' (inserted next),
3544 because `re_compile_fastmap' needs to know.
3545 Jump to the `jump_n' we might insert below. */
3546 INSERT_JUMP2 (succeed_n, laststart,
3547 b + 5 + nbytes,
3548 lower_bound);
3549 b += 5;
3550
3551 /* Code to initialize the lower bound. Insert
3552 before the `succeed_n'. The `5' is the last two
3553 bytes of this `set_number_at', plus 3 bytes of
3554 the following `succeed_n'. */
3555 insert_op2 (set_number_at, laststart, 5, lower_bound, b);
3556 b += 5;
3557 startoffset += 5;
3558 }
3559
3560 if (upper_bound < 0)
3561 {
3562 /* A negative upper bound stands for infinity,
3563 in which case it degenerates to a plain jump. */
3564 STORE_JUMP (jump, b, laststart + startoffset);
3565 b += 3;
3566 }
3567 else if (upper_bound > 1)
3568 { /* More than one repetition is allowed, so
3569 append a backward jump to the `succeed_n'
3570 that starts this interval.
3571
3572 When we've reached this during matching,
3573 we'll have matched the interval once, so
3574 jump back only `upper_bound - 1' times. */
3575 STORE_JUMP2 (jump_n, b, laststart + startoffset,
3576 upper_bound - 1);
3577 b += 5;
3578
3579 /* The location we want to set is the second
3580 parameter of the `jump_n'; that is `b-2' as
3581 an absolute address. `laststart' will be
3582 the `set_number_at' we're about to insert;
3583 `laststart+3' the number to set, the source
3584 for the relative address. But we are
3585 inserting into the middle of the pattern --
3586 so everything is getting moved up by 5.
3587 Conclusion: (b - 2) - (laststart + 3) + 5,
3588 i.e., b - laststart.
3589
3590 We insert this at the beginning of the loop
3591 so that if we fail during matching, we'll
3592 reinitialize the bounds. */
3593 insert_op2 (set_number_at, laststart, b - laststart,
3594 upper_bound - 1, b);
3595 b += 5;
3596 }
3597 }
3598 pending_exact = 0;
3599 beg_interval = NULL;
3600 }
3601 break;
3602
3603 unfetch_interval:
3604 /* If an invalid interval, match the characters as literals. */
3605 assert (beg_interval);
3606 p = beg_interval;
3607 beg_interval = NULL;
3608
3609 /* normal_char and normal_backslash need `c'. */
3610 c = '{';
3611
3612 if (!(syntax & RE_NO_BK_BRACES))
3613 {
3614 assert (p > pattern && p[-1] == '\\');
3615 goto normal_backslash;
3616 }
3617 else
3618 goto normal_char;
3619
3620 #ifdef emacs
3621 /* There is no way to specify the before_dot and after_dot
3622 operators. rms says this is ok. --karl */
3623 case '=':
3624 BUF_PUSH (at_dot);
3625 break;
3626
3627 case 's':
3628 laststart = b;
3629 PATFETCH (c);
3630 BUF_PUSH_2 (syntaxspec, syntax_spec_code[c]);
3631 break;
3632
3633 case 'S':
3634 laststart = b;
3635 PATFETCH (c);
3636 BUF_PUSH_2 (notsyntaxspec, syntax_spec_code[c]);
3637 break;
3638
3639 case 'c':
3640 laststart = b;
3641 PATFETCH (c);
3642 BUF_PUSH_2 (categoryspec, c);
3643 break;
3644
3645 case 'C':
3646 laststart = b;
3647 PATFETCH (c);
3648 BUF_PUSH_2 (notcategoryspec, c);
3649 break;
3650 #endif /* emacs */
3651
3652
3653 case 'w':
3654 if (syntax & RE_NO_GNU_OPS)
3655 goto normal_char;
3656 laststart = b;
3657 BUF_PUSH_2 (syntaxspec, Sword);
3658 break;
3659
3660
3661 case 'W':
3662 if (syntax & RE_NO_GNU_OPS)
3663 goto normal_char;
3664 laststart = b;
3665 BUF_PUSH_2 (notsyntaxspec, Sword);
3666 break;
3667
3668
3669 case '<':
3670 if (syntax & RE_NO_GNU_OPS)
3671 goto normal_char;
3672 BUF_PUSH (wordbeg);
3673 break;
3674
3675 case '>':
3676 if (syntax & RE_NO_GNU_OPS)
3677 goto normal_char;
3678 BUF_PUSH (wordend);
3679 break;
3680
3681 case '_':
3682 if (syntax & RE_NO_GNU_OPS)
3683 goto normal_char;
3684 laststart = b;
3685 PATFETCH (c);
3686 if (c == '<')
3687 BUF_PUSH (symbeg);
3688 else if (c == '>')
3689 BUF_PUSH (symend);
3690 else
3691 FREE_STACK_RETURN (REG_BADPAT);
3692 break;
3693
3694 case 'b':
3695 if (syntax & RE_NO_GNU_OPS)
3696 goto normal_char;
3697 BUF_PUSH (wordbound);
3698 break;
3699
3700 case 'B':
3701 if (syntax & RE_NO_GNU_OPS)
3702 goto normal_char;
3703 BUF_PUSH (notwordbound);
3704 break;
3705
3706 case '`':
3707 if (syntax & RE_NO_GNU_OPS)
3708 goto normal_char;
3709 BUF_PUSH (begbuf);
3710 break;
3711
3712 case '\'':
3713 if (syntax & RE_NO_GNU_OPS)
3714 goto normal_char;
3715 BUF_PUSH (endbuf);
3716 break;
3717
3718 case '1': case '2': case '3': case '4': case '5':
3719 case '6': case '7': case '8': case '9':
3720 {
3721 regnum_t reg;
3722
3723 if (syntax & RE_NO_BK_REFS)
3724 goto normal_backslash;
3725
3726 reg = c - '0';
3727
3728 if (reg > bufp->re_nsub || reg < 1
3729 /* Can't back reference to a subexp before its end. */
3730 || group_in_compile_stack (compile_stack, reg))
3731 FREE_STACK_RETURN (REG_ESUBREG);
3732
3733 laststart = b;
3734 BUF_PUSH_2 (duplicate, reg);
3735 }
3736 break;
3737
3738
3739 case '+':
3740 case '?':
3741 if (syntax & RE_BK_PLUS_QM)
3742 goto handle_plus;
3743 else
3744 goto normal_backslash;
3745
3746 default:
3747 normal_backslash:
3748 /* You might think it would be useful for \ to mean
3749 not to translate; but if we don't translate it
3750 it will never match anything. */
3751 goto normal_char;
3752 }
3753 break;
3754
3755
3756 default:
3757 /* Expects the character in `c'. */
3758 normal_char:
3759 /* If no exactn currently being built. */
3760 if (!pending_exact
3761
3762 /* If last exactn not at current position. */
3763 || pending_exact + *pending_exact + 1 != b
3764
3765 /* We have only one byte following the exactn for the count. */
3766 || *pending_exact >= (1 << BYTEWIDTH) - MAX_MULTIBYTE_LENGTH
3767
3768 /* If followed by a repetition operator. */
3769 || (p != pend && (*p == '*' || *p == '^'))
3770 || ((syntax & RE_BK_PLUS_QM)
3771 ? p + 1 < pend && *p == '\\' && (p[1] == '+' || p[1] == '?')
3772 : p != pend && (*p == '+' || *p == '?'))
3773 || ((syntax & RE_INTERVALS)
3774 && ((syntax & RE_NO_BK_BRACES)
3775 ? p != pend && *p == '{'
3776 : p + 1 < pend && p[0] == '\\' && p[1] == '{')))
3777 {
3778 /* Start building a new exactn. */
3779
3780 laststart = b;
3781
3782 BUF_PUSH_2 (exactn, 0);
3783 pending_exact = b - 1;
3784 }
3785
3786 GET_BUFFER_SPACE (MAX_MULTIBYTE_LENGTH);
3787 {
3788 int len;
3789
3790 if (multibyte)
3791 {
3792 c = TRANSLATE (c);
3793 len = CHAR_STRING (c, b);
3794 b += len;
3795 }
3796 else
3797 {
3798 c1 = RE_CHAR_TO_MULTIBYTE (c);
3799 if (! CHAR_BYTE8_P (c1))
3800 {
3801 re_wchar_t c2 = TRANSLATE (c1);
3802
3803 if (c1 != c2 && (c1 = RE_CHAR_TO_UNIBYTE (c2)) >= 0)
3804 c = c1;
3805 }
3806 *b++ = c;
3807 len = 1;
3808 }
3809 (*pending_exact) += len;
3810 }
3811
3812 break;
3813 } /* switch (c) */
3814 } /* while p != pend */
3815
3816
3817 /* Through the pattern now. */
3818
3819 FIXUP_ALT_JUMP ();
3820
3821 if (!COMPILE_STACK_EMPTY)
3822 FREE_STACK_RETURN (REG_EPAREN);
3823
3824 /* If we don't want backtracking, force success
3825 the first time we reach the end of the compiled pattern. */
3826 if (syntax & RE_NO_POSIX_BACKTRACKING)
3827 BUF_PUSH (succeed);
3828
3829 /* We have succeeded; set the length of the buffer. */
3830 bufp->used = b - bufp->buffer;
3831
3832 #ifdef DEBUG
3833 if (debug > 0)
3834 {
3835 re_compile_fastmap (bufp);
3836 DEBUG_PRINT1 ("\nCompiled pattern: \n");
3837 print_compiled_pattern (bufp);
3838 }
3839 debug--;
3840 #endif /* DEBUG */
3841
3842 #ifndef MATCH_MAY_ALLOCATE
3843 /* Initialize the failure stack to the largest possible stack. This
3844 isn't necessary unless we're trying to avoid calling alloca in
3845 the search and match routines. */
3846 {
3847 int num_regs = bufp->re_nsub + 1;
3848
3849 if (fail_stack.size < re_max_failures * TYPICAL_FAILURE_SIZE)
3850 {
3851 fail_stack.size = re_max_failures * TYPICAL_FAILURE_SIZE;
3852
3853 if (! fail_stack.stack)
3854 fail_stack.stack
3855 = (fail_stack_elt_t *) malloc (fail_stack.size
3856 * sizeof (fail_stack_elt_t));
3857 else
3858 fail_stack.stack
3859 = (fail_stack_elt_t *) realloc (fail_stack.stack,
3860 (fail_stack.size
3861 * sizeof (fail_stack_elt_t)));
3862 }
3863
3864 regex_grow_registers (num_regs);
3865 }
3866 #endif /* not MATCH_MAY_ALLOCATE */
3867
3868 FREE_STACK_RETURN (REG_NOERROR);
3869 } /* regex_compile */
3870 \f
3871 /* Subroutines for `regex_compile'. */
3872
3873 /* Store OP at LOC followed by two-byte integer parameter ARG. */
3874
3875 static void
3876 store_op1 (op, loc, arg)
3877 re_opcode_t op;
3878 unsigned char *loc;
3879 int arg;
3880 {
3881 *loc = (unsigned char) op;
3882 STORE_NUMBER (loc + 1, arg);
3883 }
3884
3885
3886 /* Like `store_op1', but for two two-byte parameters ARG1 and ARG2. */
3887
3888 static void
3889 store_op2 (op, loc, arg1, arg2)
3890 re_opcode_t op;
3891 unsigned char *loc;
3892 int arg1, arg2;
3893 {
3894 *loc = (unsigned char) op;
3895 STORE_NUMBER (loc + 1, arg1);
3896 STORE_NUMBER (loc + 3, arg2);
3897 }
3898
3899
3900 /* Copy the bytes from LOC to END to open up three bytes of space at LOC
3901 for OP followed by two-byte integer parameter ARG. */
3902
3903 static void
3904 insert_op1 (op, loc, arg, end)
3905 re_opcode_t op;
3906 unsigned char *loc;
3907 int arg;
3908 unsigned char *end;
3909 {
3910 register unsigned char *pfrom = end;
3911 register unsigned char *pto = end + 3;
3912
3913 while (pfrom != loc)
3914 *--pto = *--pfrom;
3915
3916 store_op1 (op, loc, arg);
3917 }
3918
3919
3920 /* Like `insert_op1', but for two two-byte parameters ARG1 and ARG2. */
3921
3922 static void
3923 insert_op2 (op, loc, arg1, arg2, end)
3924 re_opcode_t op;
3925 unsigned char *loc;
3926 int arg1, arg2;
3927 unsigned char *end;
3928 {
3929 register unsigned char *pfrom = end;
3930 register unsigned char *pto = end + 5;
3931
3932 while (pfrom != loc)
3933 *--pto = *--pfrom;
3934
3935 store_op2 (op, loc, arg1, arg2);
3936 }
3937
3938
3939 /* P points to just after a ^ in PATTERN. Return true if that ^ comes
3940 after an alternative or a begin-subexpression. We assume there is at
3941 least one character before the ^. */
3942
3943 static boolean
3944 at_begline_loc_p (pattern, p, syntax)
3945 re_char *pattern, *p;
3946 reg_syntax_t syntax;
3947 {
3948 re_char *prev = p - 2;
3949 boolean prev_prev_backslash = prev > pattern && prev[-1] == '\\';
3950
3951 return
3952 /* After a subexpression? */
3953 (*prev == '(' && (syntax & RE_NO_BK_PARENS || prev_prev_backslash))
3954 /* After an alternative? */
3955 || (*prev == '|' && (syntax & RE_NO_BK_VBAR || prev_prev_backslash))
3956 /* After a shy subexpression? */
3957 || ((syntax & RE_SHY_GROUPS) && prev - 2 >= pattern
3958 && prev[-1] == '?' && prev[-2] == '('
3959 && (syntax & RE_NO_BK_PARENS
3960 || (prev - 3 >= pattern && prev[-3] == '\\')));
3961 }
3962
3963
3964 /* The dual of at_begline_loc_p. This one is for $. We assume there is
3965 at least one character after the $, i.e., `P < PEND'. */
3966
3967 static boolean
3968 at_endline_loc_p (p, pend, syntax)
3969 re_char *p, *pend;
3970 reg_syntax_t syntax;
3971 {
3972 re_char *next = p;
3973 boolean next_backslash = *next == '\\';
3974 re_char *next_next = p + 1 < pend ? p + 1 : 0;
3975
3976 return
3977 /* Before a subexpression? */
3978 (syntax & RE_NO_BK_PARENS ? *next == ')'
3979 : next_backslash && next_next && *next_next == ')')
3980 /* Before an alternative? */
3981 || (syntax & RE_NO_BK_VBAR ? *next == '|'
3982 : next_backslash && next_next && *next_next == '|');
3983 }
3984
3985
3986 /* Returns true if REGNUM is in one of COMPILE_STACK's elements and
3987 false if it's not. */
3988
3989 static boolean
3990 group_in_compile_stack (compile_stack, regnum)
3991 compile_stack_type compile_stack;
3992 regnum_t regnum;
3993 {
3994 int this_element;
3995
3996 for (this_element = compile_stack.avail - 1;
3997 this_element >= 0;
3998 this_element--)
3999 if (compile_stack.stack[this_element].regnum == regnum)
4000 return true;
4001
4002 return false;
4003 }
4004 \f
4005 /* analyse_first.
4006 If fastmap is non-NULL, go through the pattern and fill fastmap
4007 with all the possible leading chars. If fastmap is NULL, don't
4008 bother filling it up (obviously) and only return whether the
4009 pattern could potentially match the empty string.
4010
4011 Return 1 if p..pend might match the empty string.
4012 Return 0 if p..pend matches at least one char.
4013 Return -1 if fastmap was not updated accurately. */
4014
4015 static int
4016 analyse_first (p, pend, fastmap, multibyte)
4017 re_char *p, *pend;
4018 char *fastmap;
4019 const int multibyte;
4020 {
4021 int j, k;
4022 boolean not;
4023
4024 /* If all elements for base leading-codes in fastmap is set, this
4025 flag is set true. */
4026 boolean match_any_multibyte_characters = false;
4027
4028 assert (p);
4029
4030 /* The loop below works as follows:
4031 - It has a working-list kept in the PATTERN_STACK and which basically
4032 starts by only containing a pointer to the first operation.
4033 - If the opcode we're looking at is a match against some set of
4034 chars, then we add those chars to the fastmap and go on to the
4035 next work element from the worklist (done via `break').
4036 - If the opcode is a control operator on the other hand, we either
4037 ignore it (if it's meaningless at this point, such as `start_memory')
4038 or execute it (if it's a jump). If the jump has several destinations
4039 (i.e. `on_failure_jump'), then we push the other destination onto the
4040 worklist.
4041 We guarantee termination by ignoring backward jumps (more or less),
4042 so that `p' is monotonically increasing. More to the point, we
4043 never set `p' (or push) anything `<= p1'. */
4044
4045 while (p < pend)
4046 {
4047 /* `p1' is used as a marker of how far back a `on_failure_jump'
4048 can go without being ignored. It is normally equal to `p'
4049 (which prevents any backward `on_failure_jump') except right
4050 after a plain `jump', to allow patterns such as:
4051 0: jump 10
4052 3..9: <body>
4053 10: on_failure_jump 3
4054 as used for the *? operator. */
4055 re_char *p1 = p;
4056
4057 switch (SWITCH_ENUM_CAST ((re_opcode_t) *p++))
4058 {
4059 case succeed:
4060 return 1;
4061 continue;
4062
4063 case duplicate:
4064 /* If the first character has to match a backreference, that means
4065 that the group was empty (since it already matched). Since this
4066 is the only case that interests us here, we can assume that the
4067 backreference must match the empty string. */
4068 p++;
4069 continue;
4070
4071
4072 /* Following are the cases which match a character. These end
4073 with `break'. */
4074
4075 case exactn:
4076 if (fastmap)
4077 {
4078 /* If multibyte is nonzero, the first byte of each
4079 character is an ASCII or a leading code. Otherwise,
4080 each byte is a character. Thus, this works in both
4081 cases. */
4082 fastmap[p[1]] = 1;
4083 if (! multibyte)
4084 {
4085 /* For the case of matching this unibyte regex
4086 against multibyte, we must set a leading code of
4087 the corresponding multibyte character. */
4088 int c = RE_CHAR_TO_MULTIBYTE (p[1]);
4089
4090 if (! CHAR_BYTE8_P (c))
4091 fastmap[CHAR_LEADING_CODE (c)] = 1;
4092 }
4093 }
4094 break;
4095
4096
4097 case anychar:
4098 /* We could put all the chars except for \n (and maybe \0)
4099 but we don't bother since it is generally not worth it. */
4100 if (!fastmap) break;
4101 return -1;
4102
4103
4104 case charset_not:
4105 if (!fastmap) break;
4106 {
4107 /* Chars beyond end of bitmap are possible matches. */
4108 for (j = CHARSET_BITMAP_SIZE (&p[-1]) * BYTEWIDTH;
4109 j < (1 << BYTEWIDTH); j++)
4110 fastmap[j] = 1;
4111 }
4112
4113 /* Fallthrough */
4114 case charset:
4115 if (!fastmap) break;
4116 not = (re_opcode_t) *(p - 1) == charset_not;
4117 for (j = CHARSET_BITMAP_SIZE (&p[-1]) * BYTEWIDTH - 1, p++;
4118 j >= 0; j--)
4119 if (!!(p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH))) ^ not)
4120 fastmap[j] = 1;
4121
4122 #ifdef emacs
4123 if (/* Any leading code can possibly start a character
4124 which doesn't match the specified set of characters. */
4125 not
4126 ||
4127 /* If we can match a character class, we can match any
4128 multibyte characters. */
4129 (CHARSET_RANGE_TABLE_EXISTS_P (&p[-2])
4130 && CHARSET_RANGE_TABLE_BITS (&p[-2]) != 0))
4131
4132 {
4133 if (match_any_multibyte_characters == false)
4134 {
4135 for (j = MIN_MULTIBYTE_LEADING_CODE;
4136 j <= MAX_MULTIBYTE_LEADING_CODE; j++)
4137 fastmap[j] = 1;
4138 match_any_multibyte_characters = true;
4139 }
4140 }
4141
4142 else if (!not && CHARSET_RANGE_TABLE_EXISTS_P (&p[-2])
4143 && match_any_multibyte_characters == false)
4144 {
4145 /* Set fastmap[I] to 1 where I is a leading code of each
4146 multibyte characer in the range table. */
4147 int c, count;
4148 unsigned char lc1, lc2;
4149
4150 /* Make P points the range table. `+ 2' is to skip flag
4151 bits for a character class. */
4152 p += CHARSET_BITMAP_SIZE (&p[-2]) + 2;
4153
4154 /* Extract the number of ranges in range table into COUNT. */
4155 EXTRACT_NUMBER_AND_INCR (count, p);
4156 for (; count > 0; count--, p += 3)
4157 {
4158 /* Extract the start and end of each range. */
4159 EXTRACT_CHARACTER (c, p);
4160 lc1 = CHAR_LEADING_CODE (c);
4161 p += 3;
4162 EXTRACT_CHARACTER (c, p);
4163 lc2 = CHAR_LEADING_CODE (c);
4164 for (j = lc1; j <= lc2; j++)
4165 fastmap[j] = 1;
4166 }
4167 }
4168 #endif
4169 break;
4170
4171 case syntaxspec:
4172 case notsyntaxspec:
4173 if (!fastmap) break;
4174 #ifndef emacs
4175 not = (re_opcode_t)p[-1] == notsyntaxspec;
4176 k = *p++;
4177 for (j = 0; j < (1 << BYTEWIDTH); j++)
4178 if ((SYNTAX (j) == (enum syntaxcode) k) ^ not)
4179 fastmap[j] = 1;
4180 break;
4181 #else /* emacs */
4182 /* This match depends on text properties. These end with
4183 aborting optimizations. */
4184 return -1;
4185
4186 case categoryspec:
4187 case notcategoryspec:
4188 if (!fastmap) break;
4189 not = (re_opcode_t)p[-1] == notcategoryspec;
4190 k = *p++;
4191 for (j = (1 << BYTEWIDTH); j >= 0; j--)
4192 if ((CHAR_HAS_CATEGORY (j, k)) ^ not)
4193 fastmap[j] = 1;
4194
4195 /* Any leading code can possibly start a character which
4196 has or doesn't has the specified category. */
4197 if (match_any_multibyte_characters == false)
4198 {
4199 for (j = MIN_MULTIBYTE_LEADING_CODE;
4200 j <= MAX_MULTIBYTE_LEADING_CODE; j++)
4201 fastmap[j] = 1;
4202 match_any_multibyte_characters = true;
4203 }
4204 break;
4205
4206 /* All cases after this match the empty string. These end with
4207 `continue'. */
4208
4209 case before_dot:
4210 case at_dot:
4211 case after_dot:
4212 #endif /* !emacs */
4213 case no_op:
4214 case begline:
4215 case endline:
4216 case begbuf:
4217 case endbuf:
4218 case wordbound:
4219 case notwordbound:
4220 case wordbeg:
4221 case wordend:
4222 case symbeg:
4223 case symend:
4224 continue;
4225
4226
4227 case jump:
4228 EXTRACT_NUMBER_AND_INCR (j, p);
4229 if (j < 0)
4230 /* Backward jumps can only go back to code that we've already
4231 visited. `re_compile' should make sure this is true. */
4232 break;
4233 p += j;
4234 switch (SWITCH_ENUM_CAST ((re_opcode_t) *p))
4235 {
4236 case on_failure_jump:
4237 case on_failure_keep_string_jump:
4238 case on_failure_jump_loop:
4239 case on_failure_jump_nastyloop:
4240 case on_failure_jump_smart:
4241 p++;
4242 break;
4243 default:
4244 continue;
4245 };
4246 /* Keep `p1' to allow the `on_failure_jump' we are jumping to
4247 to jump back to "just after here". */
4248 /* Fallthrough */
4249
4250 case on_failure_jump:
4251 case on_failure_keep_string_jump:
4252 case on_failure_jump_nastyloop:
4253 case on_failure_jump_loop:
4254 case on_failure_jump_smart:
4255 EXTRACT_NUMBER_AND_INCR (j, p);
4256 if (p + j <= p1)
4257 ; /* Backward jump to be ignored. */
4258 else
4259 { /* We have to look down both arms.
4260 We first go down the "straight" path so as to minimize
4261 stack usage when going through alternatives. */
4262 int r = analyse_first (p, pend, fastmap, multibyte);
4263 if (r) return r;
4264 p += j;
4265 }
4266 continue;
4267
4268
4269 case jump_n:
4270 /* This code simply does not properly handle forward jump_n. */
4271 DEBUG_STATEMENT (EXTRACT_NUMBER (j, p); assert (j < 0));
4272 p += 4;
4273 /* jump_n can either jump or fall through. The (backward) jump
4274 case has already been handled, so we only need to look at the
4275 fallthrough case. */
4276 continue;
4277
4278 case succeed_n:
4279 /* If N == 0, it should be an on_failure_jump_loop instead. */
4280 DEBUG_STATEMENT (EXTRACT_NUMBER (j, p + 2); assert (j > 0));
4281 p += 4;
4282 /* We only care about one iteration of the loop, so we don't
4283 need to consider the case where this behaves like an
4284 on_failure_jump. */
4285 continue;
4286
4287
4288 case set_number_at:
4289 p += 4;
4290 continue;
4291
4292
4293 case start_memory:
4294 case stop_memory:
4295 p += 1;
4296 continue;
4297
4298
4299 default:
4300 abort (); /* We have listed all the cases. */
4301 } /* switch *p++ */
4302
4303 /* Getting here means we have found the possible starting
4304 characters for one path of the pattern -- and that the empty
4305 string does not match. We need not follow this path further. */
4306 return 0;
4307 } /* while p */
4308
4309 /* We reached the end without matching anything. */
4310 return 1;
4311
4312 } /* analyse_first */
4313 \f
4314 /* re_compile_fastmap computes a ``fastmap'' for the compiled pattern in
4315 BUFP. A fastmap records which of the (1 << BYTEWIDTH) possible
4316 characters can start a string that matches the pattern. This fastmap
4317 is used by re_search to skip quickly over impossible starting points.
4318
4319 Character codes above (1 << BYTEWIDTH) are not represented in the
4320 fastmap, but the leading codes are represented. Thus, the fastmap
4321 indicates which character sets could start a match.
4322
4323 The caller must supply the address of a (1 << BYTEWIDTH)-byte data
4324 area as BUFP->fastmap.
4325
4326 We set the `fastmap', `fastmap_accurate', and `can_be_null' fields in
4327 the pattern buffer.
4328
4329 Returns 0 if we succeed, -2 if an internal error. */
4330
4331 int
4332 re_compile_fastmap (bufp)
4333 struct re_pattern_buffer *bufp;
4334 {
4335 char *fastmap = bufp->fastmap;
4336 int analysis;
4337
4338 assert (fastmap && bufp->buffer);
4339
4340 bzero (fastmap, 1 << BYTEWIDTH); /* Assume nothing's valid. */
4341 bufp->fastmap_accurate = 1; /* It will be when we're done. */
4342
4343 analysis = analyse_first (bufp->buffer, bufp->buffer + bufp->used,
4344 fastmap, RE_MULTIBYTE_P (bufp));
4345 bufp->can_be_null = (analysis != 0);
4346 return 0;
4347 } /* re_compile_fastmap */
4348 \f
4349 /* Set REGS to hold NUM_REGS registers, storing them in STARTS and
4350 ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use
4351 this memory for recording register information. STARTS and ENDS
4352 must be allocated using the malloc library routine, and must each
4353 be at least NUM_REGS * sizeof (regoff_t) bytes long.
4354
4355 If NUM_REGS == 0, then subsequent matches should allocate their own
4356 register data.
4357
4358 Unless this function is called, the first search or match using
4359 PATTERN_BUFFER will allocate its own register data, without
4360 freeing the old data. */
4361
4362 void
4363 re_set_registers (bufp, regs, num_regs, starts, ends)
4364 struct re_pattern_buffer *bufp;
4365 struct re_registers *regs;
4366 unsigned num_regs;
4367 regoff_t *starts, *ends;
4368 {
4369 if (num_regs)
4370 {
4371 bufp->regs_allocated = REGS_REALLOCATE;
4372 regs->num_regs = num_regs;
4373 regs->start = starts;
4374 regs->end = ends;
4375 }
4376 else
4377 {
4378 bufp->regs_allocated = REGS_UNALLOCATED;
4379 regs->num_regs = 0;
4380 regs->start = regs->end = (regoff_t *) 0;
4381 }
4382 }
4383 WEAK_ALIAS (__re_set_registers, re_set_registers)
4384 \f
4385 /* Searching routines. */
4386
4387 /* Like re_search_2, below, but only one string is specified, and
4388 doesn't let you say where to stop matching. */
4389
4390 int
4391 re_search (bufp, string, size, startpos, range, regs)
4392 struct re_pattern_buffer *bufp;
4393 const char *string;
4394 int size, startpos, range;
4395 struct re_registers *regs;
4396 {
4397 return re_search_2 (bufp, NULL, 0, string, size, startpos, range,
4398 regs, size);
4399 }
4400 WEAK_ALIAS (__re_search, re_search)
4401
4402 /* Head address of virtual concatenation of string. */
4403 #define HEAD_ADDR_VSTRING(P) \
4404 (((P) >= size1 ? string2 : string1))
4405
4406 /* End address of virtual concatenation of string. */
4407 #define STOP_ADDR_VSTRING(P) \
4408 (((P) >= size1 ? string2 + size2 : string1 + size1))
4409
4410 /* Address of POS in the concatenation of virtual string. */
4411 #define POS_ADDR_VSTRING(POS) \
4412 (((POS) >= size1 ? string2 - size1 : string1) + (POS))
4413
4414 /* Using the compiled pattern in BUFP->buffer, first tries to match the
4415 virtual concatenation of STRING1 and STRING2, starting first at index
4416 STARTPOS, then at STARTPOS + 1, and so on.
4417
4418 STRING1 and STRING2 have length SIZE1 and SIZE2, respectively.
4419
4420 RANGE is how far to scan while trying to match. RANGE = 0 means try
4421 only at STARTPOS; in general, the last start tried is STARTPOS +
4422 RANGE.
4423
4424 In REGS, return the indices of the virtual concatenation of STRING1
4425 and STRING2 that matched the entire BUFP->buffer and its contained
4426 subexpressions.
4427
4428 Do not consider matching one past the index STOP in the virtual
4429 concatenation of STRING1 and STRING2.
4430
4431 We return either the position in the strings at which the match was
4432 found, -1 if no match, or -2 if error (such as failure
4433 stack overflow). */
4434
4435 int
4436 re_search_2 (bufp, str1, size1, str2, size2, startpos, range, regs, stop)
4437 struct re_pattern_buffer *bufp;
4438 const char *str1, *str2;
4439 int size1, size2;
4440 int startpos;
4441 int range;
4442 struct re_registers *regs;
4443 int stop;
4444 {
4445 int val;
4446 re_char *string1 = (re_char*) str1;
4447 re_char *string2 = (re_char*) str2;
4448 register char *fastmap = bufp->fastmap;
4449 register RE_TRANSLATE_TYPE translate = bufp->translate;
4450 int total_size = size1 + size2;
4451 int endpos = startpos + range;
4452 boolean anchored_start;
4453 /* Nonzero if we are searching multibyte string. */
4454 const boolean multibyte = RE_TARGET_MULTIBYTE_P (bufp);
4455
4456 /* Check for out-of-range STARTPOS. */
4457 if (startpos < 0 || startpos > total_size)
4458 return -1;
4459
4460 /* Fix up RANGE if it might eventually take us outside
4461 the virtual concatenation of STRING1 and STRING2.
4462 Make sure we won't move STARTPOS below 0 or above TOTAL_SIZE. */
4463 if (endpos < 0)
4464 range = 0 - startpos;
4465 else if (endpos > total_size)
4466 range = total_size - startpos;
4467
4468 /* If the search isn't to be a backwards one, don't waste time in a
4469 search for a pattern anchored at beginning of buffer. */
4470 if (bufp->used > 0 && (re_opcode_t) bufp->buffer[0] == begbuf && range > 0)
4471 {
4472 if (startpos > 0)
4473 return -1;
4474 else
4475 range = 0;
4476 }
4477
4478 #ifdef emacs
4479 /* In a forward search for something that starts with \=.
4480 don't keep searching past point. */
4481 if (bufp->used > 0 && (re_opcode_t) bufp->buffer[0] == at_dot && range > 0)
4482 {
4483 range = PT_BYTE - BEGV_BYTE - startpos;
4484 if (range < 0)
4485 return -1;
4486 }
4487 #endif /* emacs */
4488
4489 /* Update the fastmap now if not correct already. */
4490 if (fastmap && !bufp->fastmap_accurate)
4491 re_compile_fastmap (bufp);
4492
4493 /* See whether the pattern is anchored. */
4494 anchored_start = (bufp->buffer[0] == begline);
4495
4496 #ifdef emacs
4497 gl_state.object = re_match_object;
4498 {
4499 int charpos = SYNTAX_TABLE_BYTE_TO_CHAR (POS_AS_IN_BUFFER (startpos));
4500
4501 SETUP_SYNTAX_TABLE_FOR_OBJECT (re_match_object, charpos, 1);
4502 }
4503 #endif
4504
4505 /* Loop through the string, looking for a place to start matching. */
4506 for (;;)
4507 {
4508 /* If the pattern is anchored,
4509 skip quickly past places we cannot match.
4510 We don't bother to treat startpos == 0 specially
4511 because that case doesn't repeat. */
4512 if (anchored_start && startpos > 0)
4513 {
4514 if (! ((startpos <= size1 ? string1[startpos - 1]
4515 : string2[startpos - size1 - 1])
4516 == '\n'))
4517 goto advance;
4518 }
4519
4520 /* If a fastmap is supplied, skip quickly over characters that
4521 cannot be the start of a match. If the pattern can match the
4522 null string, however, we don't need to skip characters; we want
4523 the first null string. */
4524 if (fastmap && startpos < total_size && !bufp->can_be_null)
4525 {
4526 register re_char *d;
4527 register re_wchar_t buf_ch;
4528
4529 d = POS_ADDR_VSTRING (startpos);
4530
4531 if (range > 0) /* Searching forwards. */
4532 {
4533 register int lim = 0;
4534 int irange = range;
4535
4536 if (startpos < size1 && startpos + range >= size1)
4537 lim = range - (size1 - startpos);
4538
4539 /* Written out as an if-else to avoid testing `translate'
4540 inside the loop. */
4541 if (RE_TRANSLATE_P (translate))
4542 {
4543 if (multibyte)
4544 while (range > lim)
4545 {
4546 int buf_charlen;
4547
4548 buf_ch = STRING_CHAR_AND_LENGTH (d, range - lim,
4549 buf_charlen);
4550 buf_ch = RE_TRANSLATE (translate, buf_ch);
4551 if (fastmap[CHAR_LEADING_CODE (buf_ch)])
4552 break;
4553
4554 range -= buf_charlen;
4555 d += buf_charlen;
4556 }
4557 else
4558 while (range > lim)
4559 {
4560 register re_wchar_t ch, translated;
4561
4562 buf_ch = *d;
4563 ch = RE_CHAR_TO_MULTIBYTE (buf_ch);
4564 translated = RE_TRANSLATE (translate, ch);
4565 if (translated != ch
4566 && (ch = RE_CHAR_TO_UNIBYTE (translated)) >= 0)
4567 buf_ch = ch;
4568 if (fastmap[buf_ch])
4569 break;
4570 d++;
4571 range--;
4572 }
4573 }
4574 else
4575 {
4576 if (multibyte)
4577 while (range > lim)
4578 {
4579 int buf_charlen;
4580
4581 buf_ch = STRING_CHAR_AND_LENGTH (d, range - lim,
4582 buf_charlen);
4583 if (fastmap[CHAR_LEADING_CODE (buf_ch)])
4584 break;
4585 range -= buf_charlen;
4586 d += buf_charlen;
4587 }
4588 else
4589 while (range > lim && !fastmap[*d])
4590 {
4591 d++;
4592 range--;
4593 }
4594 }
4595 startpos += irange - range;
4596 }
4597 else /* Searching backwards. */
4598 {
4599 int room = (startpos >= size1
4600 ? size2 + size1 - startpos
4601 : size1 - startpos);
4602 if (multibyte)
4603 {
4604 buf_ch = STRING_CHAR (d, room);
4605 buf_ch = TRANSLATE (buf_ch);
4606 if (! fastmap[CHAR_LEADING_CODE (buf_ch)])
4607 goto advance;
4608 }
4609 else
4610 {
4611 register re_wchar_t ch, translated;
4612
4613 buf_ch = *d;
4614 ch = RE_CHAR_TO_MULTIBYTE (buf_ch);
4615 translated = TRANSLATE (ch);
4616 if (translated != ch
4617 && (ch = RE_CHAR_TO_UNIBYTE (translated)) >= 0)
4618 buf_ch = ch;
4619 if (! fastmap[TRANSLATE (buf_ch)])
4620 goto advance;
4621 }
4622 }
4623 }
4624
4625 /* If can't match the null string, and that's all we have left, fail. */
4626 if (range >= 0 && startpos == total_size && fastmap
4627 && !bufp->can_be_null)
4628 return -1;
4629
4630 val = re_match_2_internal (bufp, string1, size1, string2, size2,
4631 startpos, regs, stop);
4632
4633 if (val >= 0)
4634 return startpos;
4635
4636 if (val == -2)
4637 return -2;
4638
4639 advance:
4640 if (!range)
4641 break;
4642 else if (range > 0)
4643 {
4644 /* Update STARTPOS to the next character boundary. */
4645 if (multibyte)
4646 {
4647 re_char *p = POS_ADDR_VSTRING (startpos);
4648 re_char *pend = STOP_ADDR_VSTRING (startpos);
4649 int len = MULTIBYTE_FORM_LENGTH (p, pend - p);
4650
4651 range -= len;
4652 if (range < 0)
4653 break;
4654 startpos += len;
4655 }
4656 else
4657 {
4658 range--;
4659 startpos++;
4660 }
4661 }
4662 else
4663 {
4664 range++;
4665 startpos--;
4666
4667 /* Update STARTPOS to the previous character boundary. */
4668 if (multibyte)
4669 {
4670 re_char *p = POS_ADDR_VSTRING (startpos) + 1;
4671 re_char *p0 = p;
4672 re_char *phead = HEAD_ADDR_VSTRING (startpos);
4673
4674 /* Find the head of multibyte form. */
4675 PREV_CHAR_BOUNDARY (p, phead);
4676 range += p0 - 1 - p;
4677 if (range > 0)
4678 break;
4679
4680 startpos -= p0 - 1 - p;
4681 }
4682 }
4683 }
4684 return -1;
4685 } /* re_search_2 */
4686 WEAK_ALIAS (__re_search_2, re_search_2)
4687 \f
4688 /* Declarations and macros for re_match_2. */
4689
4690 static int bcmp_translate _RE_ARGS((re_char *s1, re_char *s2,
4691 register int len,
4692 RE_TRANSLATE_TYPE translate,
4693 const int multibyte));
4694
4695 /* This converts PTR, a pointer into one of the search strings `string1'
4696 and `string2' into an offset from the beginning of that string. */
4697 #define POINTER_TO_OFFSET(ptr) \
4698 (FIRST_STRING_P (ptr) \
4699 ? ((regoff_t) ((ptr) - string1)) \
4700 : ((regoff_t) ((ptr) - string2 + size1)))
4701
4702 /* Call before fetching a character with *d. This switches over to
4703 string2 if necessary.
4704 Check re_match_2_internal for a discussion of why end_match_2 might
4705 not be within string2 (but be equal to end_match_1 instead). */
4706 #define PREFETCH() \
4707 while (d == dend) \
4708 { \
4709 /* End of string2 => fail. */ \
4710 if (dend == end_match_2) \
4711 goto fail; \
4712 /* End of string1 => advance to string2. */ \
4713 d = string2; \
4714 dend = end_match_2; \
4715 }
4716
4717 /* Call before fetching a char with *d if you already checked other limits.
4718 This is meant for use in lookahead operations like wordend, etc..
4719 where we might need to look at parts of the string that might be
4720 outside of the LIMITs (i.e past `stop'). */
4721 #define PREFETCH_NOLIMIT() \
4722 if (d == end1) \
4723 { \
4724 d = string2; \
4725 dend = end_match_2; \
4726 } \
4727
4728 /* Test if at very beginning or at very end of the virtual concatenation
4729 of `string1' and `string2'. If only one string, it's `string2'. */
4730 #define AT_STRINGS_BEG(d) ((d) == (size1 ? string1 : string2) || !size2)
4731 #define AT_STRINGS_END(d) ((d) == end2)
4732
4733
4734 /* Test if D points to a character which is word-constituent. We have
4735 two special cases to check for: if past the end of string1, look at
4736 the first character in string2; and if before the beginning of
4737 string2, look at the last character in string1. */
4738 #define WORDCHAR_P(d) \
4739 (SYNTAX ((d) == end1 ? *string2 \
4740 : (d) == string2 - 1 ? *(end1 - 1) : *(d)) \
4741 == Sword)
4742
4743 /* Disabled due to a compiler bug -- see comment at case wordbound */
4744
4745 /* The comment at case wordbound is following one, but we don't use
4746 AT_WORD_BOUNDARY anymore to support multibyte form.
4747
4748 The DEC Alpha C compiler 3.x generates incorrect code for the
4749 test WORDCHAR_P (d - 1) != WORDCHAR_P (d) in the expansion of
4750 AT_WORD_BOUNDARY, so this code is disabled. Expanding the
4751 macro and introducing temporary variables works around the bug. */
4752
4753 #if 0
4754 /* Test if the character before D and the one at D differ with respect
4755 to being word-constituent. */
4756 #define AT_WORD_BOUNDARY(d) \
4757 (AT_STRINGS_BEG (d) || AT_STRINGS_END (d) \
4758 || WORDCHAR_P (d - 1) != WORDCHAR_P (d))
4759 #endif
4760
4761 /* Free everything we malloc. */
4762 #ifdef MATCH_MAY_ALLOCATE
4763 # define FREE_VAR(var) if (var) { REGEX_FREE (var); var = NULL; } else
4764 # define FREE_VARIABLES() \
4765 do { \
4766 REGEX_FREE_STACK (fail_stack.stack); \
4767 FREE_VAR (regstart); \
4768 FREE_VAR (regend); \
4769 FREE_VAR (best_regstart); \
4770 FREE_VAR (best_regend); \
4771 } while (0)
4772 #else
4773 # define FREE_VARIABLES() ((void)0) /* Do nothing! But inhibit gcc warning. */
4774 #endif /* not MATCH_MAY_ALLOCATE */
4775
4776 \f
4777 /* Optimization routines. */
4778
4779 /* If the operation is a match against one or more chars,
4780 return a pointer to the next operation, else return NULL. */
4781 static re_char *
4782 skip_one_char (p)
4783 re_char *p;
4784 {
4785 switch (SWITCH_ENUM_CAST (*p++))
4786 {
4787 case anychar:
4788 break;
4789
4790 case exactn:
4791 p += *p + 1;
4792 break;
4793
4794 case charset_not:
4795 case charset:
4796 if (CHARSET_RANGE_TABLE_EXISTS_P (p - 1))
4797 {
4798 int mcnt;
4799 p = CHARSET_RANGE_TABLE (p - 1);
4800 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4801 p = CHARSET_RANGE_TABLE_END (p, mcnt);
4802 }
4803 else
4804 p += 1 + CHARSET_BITMAP_SIZE (p - 1);
4805 break;
4806
4807 case syntaxspec:
4808 case notsyntaxspec:
4809 #ifdef emacs
4810 case categoryspec:
4811 case notcategoryspec:
4812 #endif /* emacs */
4813 p++;
4814 break;
4815
4816 default:
4817 p = NULL;
4818 }
4819 return p;
4820 }
4821
4822
4823 /* Jump over non-matching operations. */
4824 static re_char *
4825 skip_noops (p, pend)
4826 re_char *p, *pend;
4827 {
4828 int mcnt;
4829 while (p < pend)
4830 {
4831 switch (SWITCH_ENUM_CAST ((re_opcode_t) *p))
4832 {
4833 case start_memory:
4834 case stop_memory:
4835 p += 2; break;
4836 case no_op:
4837 p += 1; break;
4838 case jump:
4839 p += 1;
4840 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4841 p += mcnt;
4842 break;
4843 default:
4844 return p;
4845 }
4846 }
4847 assert (p == pend);
4848 return p;
4849 }
4850
4851 /* Non-zero if "p1 matches something" implies "p2 fails". */
4852 static int
4853 mutually_exclusive_p (bufp, p1, p2)
4854 struct re_pattern_buffer *bufp;
4855 re_char *p1, *p2;
4856 {
4857 re_opcode_t op2;
4858 const boolean multibyte = RE_MULTIBYTE_P (bufp);
4859 unsigned char *pend = bufp->buffer + bufp->used;
4860
4861 assert (p1 >= bufp->buffer && p1 < pend
4862 && p2 >= bufp->buffer && p2 <= pend);
4863
4864 /* Skip over open/close-group commands.
4865 If what follows this loop is a ...+ construct,
4866 look at what begins its body, since we will have to
4867 match at least one of that. */
4868 p2 = skip_noops (p2, pend);
4869 /* The same skip can be done for p1, except that this function
4870 is only used in the case where p1 is a simple match operator. */
4871 /* p1 = skip_noops (p1, pend); */
4872
4873 assert (p1 >= bufp->buffer && p1 < pend
4874 && p2 >= bufp->buffer && p2 <= pend);
4875
4876 op2 = p2 == pend ? succeed : *p2;
4877
4878 switch (SWITCH_ENUM_CAST (op2))
4879 {
4880 case succeed:
4881 case endbuf:
4882 /* If we're at the end of the pattern, we can change. */
4883 if (skip_one_char (p1))
4884 {
4885 DEBUG_PRINT1 (" End of pattern: fast loop.\n");
4886 return 1;
4887 }
4888 break;
4889
4890 case endline:
4891 case exactn:
4892 {
4893 register re_wchar_t c
4894 = (re_opcode_t) *p2 == endline ? '\n'
4895 : RE_STRING_CHAR (p2 + 2, pend - p2 - 2, multibyte);
4896
4897 if ((re_opcode_t) *p1 == exactn)
4898 {
4899 if (c != RE_STRING_CHAR (p1 + 2, pend - p1 - 2, multibyte))
4900 {
4901 DEBUG_PRINT3 (" '%c' != '%c' => fast loop.\n", c, p1[2]);
4902 return 1;
4903 }
4904 }
4905
4906 else if ((re_opcode_t) *p1 == charset
4907 || (re_opcode_t) *p1 == charset_not)
4908 {
4909 int not = (re_opcode_t) *p1 == charset_not;
4910
4911 /* Test if C is listed in charset (or charset_not)
4912 at `p1'. */
4913 if (! multibyte || IS_REAL_ASCII (c))
4914 {
4915 if (c < CHARSET_BITMAP_SIZE (p1) * BYTEWIDTH
4916 && p1[2 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH)))
4917 not = !not;
4918 }
4919 else if (CHARSET_RANGE_TABLE_EXISTS_P (p1))
4920 CHARSET_LOOKUP_RANGE_TABLE (not, c, p1);
4921
4922 /* `not' is equal to 1 if c would match, which means
4923 that we can't change to pop_failure_jump. */
4924 if (!not)
4925 {
4926 DEBUG_PRINT1 (" No match => fast loop.\n");
4927 return 1;
4928 }
4929 }
4930 else if ((re_opcode_t) *p1 == anychar
4931 && c == '\n')
4932 {
4933 DEBUG_PRINT1 (" . != \\n => fast loop.\n");
4934 return 1;
4935 }
4936 }
4937 break;
4938
4939 case charset:
4940 {
4941 if ((re_opcode_t) *p1 == exactn)
4942 /* Reuse the code above. */
4943 return mutually_exclusive_p (bufp, p2, p1);
4944
4945 /* It is hard to list up all the character in charset
4946 P2 if it includes multibyte character. Give up in
4947 such case. */
4948 else if (!multibyte || !CHARSET_RANGE_TABLE_EXISTS_P (p2))
4949 {
4950 /* Now, we are sure that P2 has no range table.
4951 So, for the size of bitmap in P2, `p2[1]' is
4952 enough. But P1 may have range table, so the
4953 size of bitmap table of P1 is extracted by
4954 using macro `CHARSET_BITMAP_SIZE'.
4955
4956 In a multibyte case, we know that all the character
4957 listed in P2 is ASCII. In a unibyte case, P1 has only a
4958 bitmap table. So, in both cases, it is enough to test
4959 only the bitmap table of P1. */
4960
4961 if ((re_opcode_t) *p1 == charset)
4962 {
4963 int idx;
4964 /* We win if the charset inside the loop
4965 has no overlap with the one after the loop. */
4966 for (idx = 0;
4967 (idx < (int) p2[1]
4968 && idx < CHARSET_BITMAP_SIZE (p1));
4969 idx++)
4970 if ((p2[2 + idx] & p1[2 + idx]) != 0)
4971 break;
4972
4973 if (idx == p2[1]
4974 || idx == CHARSET_BITMAP_SIZE (p1))
4975 {
4976 DEBUG_PRINT1 (" No match => fast loop.\n");
4977 return 1;
4978 }
4979 }
4980 else if ((re_opcode_t) *p1 == charset_not)
4981 {
4982 int idx;
4983 /* We win if the charset_not inside the loop lists
4984 every character listed in the charset after. */
4985 for (idx = 0; idx < (int) p2[1]; idx++)
4986 if (! (p2[2 + idx] == 0
4987 || (idx < CHARSET_BITMAP_SIZE (p1)
4988 && ((p2[2 + idx] & ~ p1[2 + idx]) == 0))))
4989 break;
4990
4991 if (idx == p2[1])
4992 {
4993 DEBUG_PRINT1 (" No match => fast loop.\n");
4994 return 1;
4995 }
4996 }
4997 }
4998 }
4999 break;
5000
5001 case charset_not:
5002 switch (SWITCH_ENUM_CAST (*p1))
5003 {
5004 case exactn:
5005 case charset:
5006 /* Reuse the code above. */
5007 return mutually_exclusive_p (bufp, p2, p1);
5008 case charset_not:
5009 /* When we have two charset_not, it's very unlikely that
5010 they don't overlap. The union of the two sets of excluded
5011 chars should cover all possible chars, which, as a matter of
5012 fact, is virtually impossible in multibyte buffers. */
5013 break;
5014 }
5015 break;
5016
5017 case wordend:
5018 return ((re_opcode_t) *p1 == syntaxspec && p1[1] == Sword);
5019 case symend:
5020 return ((re_opcode_t) *p1 == syntaxspec
5021 && (p1[1] == Ssymbol || p1[1] == Sword));
5022 case notsyntaxspec:
5023 return ((re_opcode_t) *p1 == syntaxspec && p1[1] == p2[1]);
5024
5025 case wordbeg:
5026 return ((re_opcode_t) *p1 == notsyntaxspec && p1[1] == Sword);
5027 case symbeg:
5028 return ((re_opcode_t) *p1 == notsyntaxspec
5029 && (p1[1] == Ssymbol || p1[1] == Sword));
5030 case syntaxspec:
5031 return ((re_opcode_t) *p1 == notsyntaxspec && p1[1] == p2[1]);
5032
5033 case wordbound:
5034 return (((re_opcode_t) *p1 == notsyntaxspec
5035 || (re_opcode_t) *p1 == syntaxspec)
5036 && p1[1] == Sword);
5037
5038 #ifdef emacs
5039 case categoryspec:
5040 return ((re_opcode_t) *p1 == notcategoryspec && p1[1] == p2[1]);
5041 case notcategoryspec:
5042 return ((re_opcode_t) *p1 == categoryspec && p1[1] == p2[1]);
5043 #endif /* emacs */
5044
5045 default:
5046 ;
5047 }
5048
5049 /* Safe default. */
5050 return 0;
5051 }
5052
5053 \f
5054 /* Matching routines. */
5055
5056 #ifndef emacs /* Emacs never uses this. */
5057 /* re_match is like re_match_2 except it takes only a single string. */
5058
5059 int
5060 re_match (bufp, string, size, pos, regs)
5061 struct re_pattern_buffer *bufp;
5062 const char *string;
5063 int size, pos;
5064 struct re_registers *regs;
5065 {
5066 int result = re_match_2_internal (bufp, NULL, 0, (re_char*) string, size,
5067 pos, regs, size);
5068 return result;
5069 }
5070 WEAK_ALIAS (__re_match, re_match)
5071 #endif /* not emacs */
5072
5073 #ifdef emacs
5074 /* In Emacs, this is the string or buffer in which we
5075 are matching. It is used for looking up syntax properties. */
5076 Lisp_Object re_match_object;
5077 #endif
5078
5079 /* re_match_2 matches the compiled pattern in BUFP against the
5080 the (virtual) concatenation of STRING1 and STRING2 (of length SIZE1
5081 and SIZE2, respectively). We start matching at POS, and stop
5082 matching at STOP.
5083
5084 If REGS is non-null and the `no_sub' field of BUFP is nonzero, we
5085 store offsets for the substring each group matched in REGS. See the
5086 documentation for exactly how many groups we fill.
5087
5088 We return -1 if no match, -2 if an internal error (such as the
5089 failure stack overflowing). Otherwise, we return the length of the
5090 matched substring. */
5091
5092 int
5093 re_match_2 (bufp, string1, size1, string2, size2, pos, regs, stop)
5094 struct re_pattern_buffer *bufp;
5095 const char *string1, *string2;
5096 int size1, size2;
5097 int pos;
5098 struct re_registers *regs;
5099 int stop;
5100 {
5101 int result;
5102
5103 #ifdef emacs
5104 int charpos;
5105 gl_state.object = re_match_object;
5106 charpos = SYNTAX_TABLE_BYTE_TO_CHAR (POS_AS_IN_BUFFER (pos));
5107 SETUP_SYNTAX_TABLE_FOR_OBJECT (re_match_object, charpos, 1);
5108 #endif
5109
5110 result = re_match_2_internal (bufp, (re_char*) string1, size1,
5111 (re_char*) string2, size2,
5112 pos, regs, stop);
5113 return result;
5114 }
5115 WEAK_ALIAS (__re_match_2, re_match_2)
5116
5117
5118 /* This is a separate function so that we can force an alloca cleanup
5119 afterwards. */
5120 static int
5121 re_match_2_internal (bufp, string1, size1, string2, size2, pos, regs, stop)
5122 struct re_pattern_buffer *bufp;
5123 re_char *string1, *string2;
5124 int size1, size2;
5125 int pos;
5126 struct re_registers *regs;
5127 int stop;
5128 {
5129 /* General temporaries. */
5130 int mcnt;
5131 size_t reg;
5132 boolean not;
5133
5134 /* Just past the end of the corresponding string. */
5135 re_char *end1, *end2;
5136
5137 /* Pointers into string1 and string2, just past the last characters in
5138 each to consider matching. */
5139 re_char *end_match_1, *end_match_2;
5140
5141 /* Where we are in the data, and the end of the current string. */
5142 re_char *d, *dend;
5143
5144 /* Used sometimes to remember where we were before starting matching
5145 an operator so that we can go back in case of failure. This "atomic"
5146 behavior of matching opcodes is indispensable to the correctness
5147 of the on_failure_keep_string_jump optimization. */
5148 re_char *dfail;
5149
5150 /* Where we are in the pattern, and the end of the pattern. */
5151 re_char *p = bufp->buffer;
5152 re_char *pend = p + bufp->used;
5153
5154 /* We use this to map every character in the string. */
5155 RE_TRANSLATE_TYPE translate = bufp->translate;
5156
5157 /* Nonzero if BUFP is setup from a multibyte regex. */
5158 const boolean multibyte = RE_MULTIBYTE_P (bufp);
5159
5160 /* Nonzero if STRING1/STRING2 are multibyte. */
5161 const boolean target_multibyte = RE_TARGET_MULTIBYTE_P (bufp);
5162
5163 /* Failure point stack. Each place that can handle a failure further
5164 down the line pushes a failure point on this stack. It consists of
5165 regstart, and regend for all registers corresponding to
5166 the subexpressions we're currently inside, plus the number of such
5167 registers, and, finally, two char *'s. The first char * is where
5168 to resume scanning the pattern; the second one is where to resume
5169 scanning the strings. */
5170 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
5171 fail_stack_type fail_stack;
5172 #endif
5173 #ifdef DEBUG
5174 unsigned nfailure_points_pushed = 0, nfailure_points_popped = 0;
5175 #endif
5176
5177 #if defined REL_ALLOC && defined REGEX_MALLOC
5178 /* This holds the pointer to the failure stack, when
5179 it is allocated relocatably. */
5180 fail_stack_elt_t *failure_stack_ptr;
5181 #endif
5182
5183 /* We fill all the registers internally, independent of what we
5184 return, for use in backreferences. The number here includes
5185 an element for register zero. */
5186 size_t num_regs = bufp->re_nsub + 1;
5187
5188 /* Information on the contents of registers. These are pointers into
5189 the input strings; they record just what was matched (on this
5190 attempt) by a subexpression part of the pattern, that is, the
5191 regnum-th regstart pointer points to where in the pattern we began
5192 matching and the regnum-th regend points to right after where we
5193 stopped matching the regnum-th subexpression. (The zeroth register
5194 keeps track of what the whole pattern matches.) */
5195 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
5196 re_char **regstart, **regend;
5197 #endif
5198
5199 /* The following record the register info as found in the above
5200 variables when we find a match better than any we've seen before.
5201 This happens as we backtrack through the failure points, which in
5202 turn happens only if we have not yet matched the entire string. */
5203 unsigned best_regs_set = false;
5204 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
5205 re_char **best_regstart, **best_regend;
5206 #endif
5207
5208 /* Logically, this is `best_regend[0]'. But we don't want to have to
5209 allocate space for that if we're not allocating space for anything
5210 else (see below). Also, we never need info about register 0 for
5211 any of the other register vectors, and it seems rather a kludge to
5212 treat `best_regend' differently than the rest. So we keep track of
5213 the end of the best match so far in a separate variable. We
5214 initialize this to NULL so that when we backtrack the first time
5215 and need to test it, it's not garbage. */
5216 re_char *match_end = NULL;
5217
5218 #ifdef DEBUG
5219 /* Counts the total number of registers pushed. */
5220 unsigned num_regs_pushed = 0;
5221 #endif
5222
5223 DEBUG_PRINT1 ("\n\nEntering re_match_2.\n");
5224
5225 INIT_FAIL_STACK ();
5226
5227 #ifdef MATCH_MAY_ALLOCATE
5228 /* Do not bother to initialize all the register variables if there are
5229 no groups in the pattern, as it takes a fair amount of time. If
5230 there are groups, we include space for register 0 (the whole
5231 pattern), even though we never use it, since it simplifies the
5232 array indexing. We should fix this. */
5233 if (bufp->re_nsub)
5234 {
5235 regstart = REGEX_TALLOC (num_regs, re_char *);
5236 regend = REGEX_TALLOC (num_regs, re_char *);
5237 best_regstart = REGEX_TALLOC (num_regs, re_char *);
5238 best_regend = REGEX_TALLOC (num_regs, re_char *);
5239
5240 if (!(regstart && regend && best_regstart && best_regend))
5241 {
5242 FREE_VARIABLES ();
5243 return -2;
5244 }
5245 }
5246 else
5247 {
5248 /* We must initialize all our variables to NULL, so that
5249 `FREE_VARIABLES' doesn't try to free them. */
5250 regstart = regend = best_regstart = best_regend = NULL;
5251 }
5252 #endif /* MATCH_MAY_ALLOCATE */
5253
5254 /* The starting position is bogus. */
5255 if (pos < 0 || pos > size1 + size2)
5256 {
5257 FREE_VARIABLES ();
5258 return -1;
5259 }
5260
5261 /* Initialize subexpression text positions to -1 to mark ones that no
5262 start_memory/stop_memory has been seen for. Also initialize the
5263 register information struct. */
5264 for (reg = 1; reg < num_regs; reg++)
5265 regstart[reg] = regend[reg] = NULL;
5266
5267 /* We move `string1' into `string2' if the latter's empty -- but not if
5268 `string1' is null. */
5269 if (size2 == 0 && string1 != NULL)
5270 {
5271 string2 = string1;
5272 size2 = size1;
5273 string1 = 0;
5274 size1 = 0;
5275 }
5276 end1 = string1 + size1;
5277 end2 = string2 + size2;
5278
5279 /* `p' scans through the pattern as `d' scans through the data.
5280 `dend' is the end of the input string that `d' points within. `d'
5281 is advanced into the following input string whenever necessary, but
5282 this happens before fetching; therefore, at the beginning of the
5283 loop, `d' can be pointing at the end of a string, but it cannot
5284 equal `string2'. */
5285 if (pos >= size1)
5286 {
5287 /* Only match within string2. */
5288 d = string2 + pos - size1;
5289 dend = end_match_2 = string2 + stop - size1;
5290 end_match_1 = end1; /* Just to give it a value. */
5291 }
5292 else
5293 {
5294 if (stop < size1)
5295 {
5296 /* Only match within string1. */
5297 end_match_1 = string1 + stop;
5298 /* BEWARE!
5299 When we reach end_match_1, PREFETCH normally switches to string2.
5300 But in the present case, this means that just doing a PREFETCH
5301 makes us jump from `stop' to `gap' within the string.
5302 What we really want here is for the search to stop as
5303 soon as we hit end_match_1. That's why we set end_match_2
5304 to end_match_1 (since PREFETCH fails as soon as we hit
5305 end_match_2). */
5306 end_match_2 = end_match_1;
5307 }
5308 else
5309 { /* It's important to use this code when stop == size so that
5310 moving `d' from end1 to string2 will not prevent the d == dend
5311 check from catching the end of string. */
5312 end_match_1 = end1;
5313 end_match_2 = string2 + stop - size1;
5314 }
5315 d = string1 + pos;
5316 dend = end_match_1;
5317 }
5318
5319 DEBUG_PRINT1 ("The compiled pattern is: ");
5320 DEBUG_PRINT_COMPILED_PATTERN (bufp, p, pend);
5321 DEBUG_PRINT1 ("The string to match is: `");
5322 DEBUG_PRINT_DOUBLE_STRING (d, string1, size1, string2, size2);
5323 DEBUG_PRINT1 ("'\n");
5324
5325 /* This loops over pattern commands. It exits by returning from the
5326 function if the match is complete, or it drops through if the match
5327 fails at this starting point in the input data. */
5328 for (;;)
5329 {
5330 DEBUG_PRINT2 ("\n%p: ", p);
5331
5332 if (p == pend)
5333 { /* End of pattern means we might have succeeded. */
5334 DEBUG_PRINT1 ("end of pattern ... ");
5335
5336 /* If we haven't matched the entire string, and we want the
5337 longest match, try backtracking. */
5338 if (d != end_match_2)
5339 {
5340 /* 1 if this match ends in the same string (string1 or string2)
5341 as the best previous match. */
5342 boolean same_str_p = (FIRST_STRING_P (match_end)
5343 == FIRST_STRING_P (d));
5344 /* 1 if this match is the best seen so far. */
5345 boolean best_match_p;
5346
5347 /* AIX compiler got confused when this was combined
5348 with the previous declaration. */
5349 if (same_str_p)
5350 best_match_p = d > match_end;
5351 else
5352 best_match_p = !FIRST_STRING_P (d);
5353
5354 DEBUG_PRINT1 ("backtracking.\n");
5355
5356 if (!FAIL_STACK_EMPTY ())
5357 { /* More failure points to try. */
5358
5359 /* If exceeds best match so far, save it. */
5360 if (!best_regs_set || best_match_p)
5361 {
5362 best_regs_set = true;
5363 match_end = d;
5364
5365 DEBUG_PRINT1 ("\nSAVING match as best so far.\n");
5366
5367 for (reg = 1; reg < num_regs; reg++)
5368 {
5369 best_regstart[reg] = regstart[reg];
5370 best_regend[reg] = regend[reg];
5371 }
5372 }
5373 goto fail;
5374 }
5375
5376 /* If no failure points, don't restore garbage. And if
5377 last match is real best match, don't restore second
5378 best one. */
5379 else if (best_regs_set && !best_match_p)
5380 {
5381 restore_best_regs:
5382 /* Restore best match. It may happen that `dend ==
5383 end_match_1' while the restored d is in string2.
5384 For example, the pattern `x.*y.*z' against the
5385 strings `x-' and `y-z-', if the two strings are
5386 not consecutive in memory. */
5387 DEBUG_PRINT1 ("Restoring best registers.\n");
5388
5389 d = match_end;
5390 dend = ((d >= string1 && d <= end1)
5391 ? end_match_1 : end_match_2);
5392
5393 for (reg = 1; reg < num_regs; reg++)
5394 {
5395 regstart[reg] = best_regstart[reg];
5396 regend[reg] = best_regend[reg];
5397 }
5398 }
5399 } /* d != end_match_2 */
5400
5401 succeed_label:
5402 DEBUG_PRINT1 ("Accepting match.\n");
5403
5404 /* If caller wants register contents data back, do it. */
5405 if (regs && !bufp->no_sub)
5406 {
5407 /* Have the register data arrays been allocated? */
5408 if (bufp->regs_allocated == REGS_UNALLOCATED)
5409 { /* No. So allocate them with malloc. We need one
5410 extra element beyond `num_regs' for the `-1' marker
5411 GNU code uses. */
5412 regs->num_regs = MAX (RE_NREGS, num_regs + 1);
5413 regs->start = TALLOC (regs->num_regs, regoff_t);
5414 regs->end = TALLOC (regs->num_regs, regoff_t);
5415 if (regs->start == NULL || regs->end == NULL)
5416 {
5417 FREE_VARIABLES ();
5418 return -2;
5419 }
5420 bufp->regs_allocated = REGS_REALLOCATE;
5421 }
5422 else if (bufp->regs_allocated == REGS_REALLOCATE)
5423 { /* Yes. If we need more elements than were already
5424 allocated, reallocate them. If we need fewer, just
5425 leave it alone. */
5426 if (regs->num_regs < num_regs + 1)
5427 {
5428 regs->num_regs = num_regs + 1;
5429 RETALLOC (regs->start, regs->num_regs, regoff_t);
5430 RETALLOC (regs->end, regs->num_regs, regoff_t);
5431 if (regs->start == NULL || regs->end == NULL)
5432 {
5433 FREE_VARIABLES ();
5434 return -2;
5435 }
5436 }
5437 }
5438 else
5439 {
5440 /* These braces fend off a "empty body in an else-statement"
5441 warning under GCC when assert expands to nothing. */
5442 assert (bufp->regs_allocated == REGS_FIXED);
5443 }
5444
5445 /* Convert the pointer data in `regstart' and `regend' to
5446 indices. Register zero has to be set differently,
5447 since we haven't kept track of any info for it. */
5448 if (regs->num_regs > 0)
5449 {
5450 regs->start[0] = pos;
5451 regs->end[0] = POINTER_TO_OFFSET (d);
5452 }
5453
5454 /* Go through the first `min (num_regs, regs->num_regs)'
5455 registers, since that is all we initialized. */
5456 for (reg = 1; reg < MIN (num_regs, regs->num_regs); reg++)
5457 {
5458 if (REG_UNSET (regstart[reg]) || REG_UNSET (regend[reg]))
5459 regs->start[reg] = regs->end[reg] = -1;
5460 else
5461 {
5462 regs->start[reg]
5463 = (regoff_t) POINTER_TO_OFFSET (regstart[reg]);
5464 regs->end[reg]
5465 = (regoff_t) POINTER_TO_OFFSET (regend[reg]);
5466 }
5467 }
5468
5469 /* If the regs structure we return has more elements than
5470 were in the pattern, set the extra elements to -1. If
5471 we (re)allocated the registers, this is the case,
5472 because we always allocate enough to have at least one
5473 -1 at the end. */
5474 for (reg = num_regs; reg < regs->num_regs; reg++)
5475 regs->start[reg] = regs->end[reg] = -1;
5476 } /* regs && !bufp->no_sub */
5477
5478 DEBUG_PRINT4 ("%u failure points pushed, %u popped (%u remain).\n",
5479 nfailure_points_pushed, nfailure_points_popped,
5480 nfailure_points_pushed - nfailure_points_popped);
5481 DEBUG_PRINT2 ("%u registers pushed.\n", num_regs_pushed);
5482
5483 mcnt = POINTER_TO_OFFSET (d) - pos;
5484
5485 DEBUG_PRINT2 ("Returning %d from re_match_2.\n", mcnt);
5486
5487 FREE_VARIABLES ();
5488 return mcnt;
5489 }
5490
5491 /* Otherwise match next pattern command. */
5492 switch (SWITCH_ENUM_CAST ((re_opcode_t) *p++))
5493 {
5494 /* Ignore these. Used to ignore the n of succeed_n's which
5495 currently have n == 0. */
5496 case no_op:
5497 DEBUG_PRINT1 ("EXECUTING no_op.\n");
5498 break;
5499
5500 case succeed:
5501 DEBUG_PRINT1 ("EXECUTING succeed.\n");
5502 goto succeed_label;
5503
5504 /* Match the next n pattern characters exactly. The following
5505 byte in the pattern defines n, and the n bytes after that
5506 are the characters to match. */
5507 case exactn:
5508 mcnt = *p++;
5509 DEBUG_PRINT2 ("EXECUTING exactn %d.\n", mcnt);
5510
5511 /* Remember the start point to rollback upon failure. */
5512 dfail = d;
5513
5514 #ifndef emacs
5515 /* This is written out as an if-else so we don't waste time
5516 testing `translate' inside the loop. */
5517 if (RE_TRANSLATE_P (translate))
5518 do
5519 {
5520 PREFETCH ();
5521 if (RE_TRANSLATE (translate, *d) != *p++)
5522 {
5523 d = dfail;
5524 goto fail;
5525 }
5526 d++;
5527 }
5528 while (--mcnt);
5529 else
5530 do
5531 {
5532 PREFETCH ();
5533 if (*d++ != *p++)
5534 {
5535 d = dfail;
5536 goto fail;
5537 }
5538 }
5539 while (--mcnt);
5540 #else /* emacs */
5541 /* The cost of testing `translate' is comparatively small. */
5542 if (target_multibyte)
5543 do
5544 {
5545 int pat_charlen, buf_charlen;
5546 int pat_ch, buf_ch;
5547
5548 PREFETCH ();
5549 if (multibyte)
5550 pat_ch = STRING_CHAR_AND_LENGTH (p, pend - p, pat_charlen);
5551 else
5552 {
5553 pat_ch = RE_CHAR_TO_MULTIBYTE (*p);
5554 pat_charlen = 1;
5555 }
5556 buf_ch = STRING_CHAR_AND_LENGTH (d, dend - d, buf_charlen);
5557
5558 if (TRANSLATE (buf_ch) != pat_ch)
5559 {
5560 d = dfail;
5561 goto fail;
5562 }
5563
5564 p += pat_charlen;
5565 d += buf_charlen;
5566 mcnt -= pat_charlen;
5567 }
5568 while (mcnt > 0);
5569 else
5570 do
5571 {
5572 int pat_charlen, buf_charlen;
5573 int pat_ch, buf_ch;
5574
5575 PREFETCH ();
5576 if (multibyte)
5577 {
5578 pat_ch = STRING_CHAR_AND_LENGTH (p, pend - p, pat_charlen);
5579 if (CHAR_BYTE8_P (pat_ch))
5580 pat_ch = CHAR_TO_BYTE8 (pat_ch);
5581 else
5582 pat_ch = RE_CHAR_TO_UNIBYTE (pat_ch);
5583 }
5584 else
5585 {
5586 pat_ch = *p;
5587 pat_charlen = 1;
5588 }
5589 buf_ch = RE_CHAR_TO_MULTIBYTE (*d);
5590 if (! CHAR_BYTE8_P (buf_ch))
5591 {
5592 buf_ch = TRANSLATE (buf_ch);
5593 buf_ch = RE_CHAR_TO_UNIBYTE (buf_ch);
5594 if (buf_ch < 0)
5595 buf_ch = *d;
5596 }
5597 if (buf_ch != pat_ch)
5598 {
5599 d = dfail;
5600 goto fail;
5601 }
5602 p += pat_charlen;
5603 d++;
5604 }
5605 while (--mcnt);
5606 #endif
5607 break;
5608
5609
5610 /* Match any character except possibly a newline or a null. */
5611 case anychar:
5612 {
5613 int buf_charlen;
5614 re_wchar_t buf_ch;
5615
5616 DEBUG_PRINT1 ("EXECUTING anychar.\n");
5617
5618 PREFETCH ();
5619 buf_ch = RE_STRING_CHAR_AND_LENGTH (d, dend - d, buf_charlen,
5620 target_multibyte);
5621 buf_ch = TRANSLATE (buf_ch);
5622
5623 if ((!(bufp->syntax & RE_DOT_NEWLINE)
5624 && buf_ch == '\n')
5625 || ((bufp->syntax & RE_DOT_NOT_NULL)
5626 && buf_ch == '\000'))
5627 goto fail;
5628
5629 DEBUG_PRINT2 (" Matched `%d'.\n", *d);
5630 d += buf_charlen;
5631 }
5632 break;
5633
5634
5635 case charset:
5636 case charset_not:
5637 {
5638 register unsigned int c;
5639 boolean not = (re_opcode_t) *(p - 1) == charset_not;
5640 int len;
5641
5642 /* Start of actual range_table, or end of bitmap if there is no
5643 range table. */
5644 re_char *range_table;
5645
5646 /* Nonzero if there is a range table. */
5647 int range_table_exists;
5648
5649 /* Number of ranges of range table. This is not included
5650 in the initial byte-length of the command. */
5651 int count = 0;
5652
5653 DEBUG_PRINT2 ("EXECUTING charset%s.\n", not ? "_not" : "");
5654
5655 range_table_exists = CHARSET_RANGE_TABLE_EXISTS_P (&p[-1]);
5656
5657 if (range_table_exists)
5658 {
5659 range_table = CHARSET_RANGE_TABLE (&p[-1]); /* Past the bitmap. */
5660 EXTRACT_NUMBER_AND_INCR (count, range_table);
5661 }
5662
5663 PREFETCH ();
5664 c = RE_STRING_CHAR_AND_LENGTH (d, dend - d, len, target_multibyte);
5665 if (target_multibyte)
5666 {
5667 int c1;
5668
5669 c = TRANSLATE (c);
5670 c1 = RE_CHAR_TO_UNIBYTE (c);
5671 if (c1 >= 0)
5672 c = c1;
5673 }
5674 else
5675 {
5676 int c1 = RE_CHAR_TO_MULTIBYTE (c);
5677
5678 if (! CHAR_BYTE8_P (c1))
5679 {
5680 c1 = TRANSLATE (c1);
5681 c1 = RE_CHAR_TO_UNIBYTE (c1);
5682 if (c1 >= 0)
5683 c = c1;
5684 }
5685 }
5686
5687 if (c < (1 << BYTEWIDTH))
5688 { /* Lookup bitmap. */
5689 /* Cast to `unsigned' instead of `unsigned char' in
5690 case the bit list is a full 32 bytes long. */
5691 if (c < (unsigned) (CHARSET_BITMAP_SIZE (&p[-1]) * BYTEWIDTH)
5692 && p[1 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH)))
5693 not = !not;
5694 }
5695 #ifdef emacs
5696 else if (range_table_exists)
5697 {
5698 int class_bits = CHARSET_RANGE_TABLE_BITS (&p[-1]);
5699
5700 if ( (class_bits & BIT_LOWER && ISLOWER (c))
5701 | (class_bits & BIT_MULTIBYTE)
5702 | (class_bits & BIT_PUNCT && ISPUNCT (c))
5703 | (class_bits & BIT_SPACE && ISSPACE (c))
5704 | (class_bits & BIT_UPPER && ISUPPER (c))
5705 | (class_bits & BIT_WORD && ISWORD (c)))
5706 not = !not;
5707 else
5708 CHARSET_LOOKUP_RANGE_TABLE_RAW (not, c, range_table, count);
5709 }
5710 #endif /* emacs */
5711
5712 if (range_table_exists)
5713 p = CHARSET_RANGE_TABLE_END (range_table, count);
5714 else
5715 p += CHARSET_BITMAP_SIZE (&p[-1]) + 1;
5716
5717 if (!not) goto fail;
5718
5719 d += len;
5720 break;
5721 }
5722
5723
5724 /* The beginning of a group is represented by start_memory.
5725 The argument is the register number. The text
5726 matched within the group is recorded (in the internal
5727 registers data structure) under the register number. */
5728 case start_memory:
5729 DEBUG_PRINT2 ("EXECUTING start_memory %d:\n", *p);
5730
5731 /* In case we need to undo this operation (via backtracking). */
5732 PUSH_FAILURE_REG ((unsigned int)*p);
5733
5734 regstart[*p] = d;
5735 regend[*p] = NULL; /* probably unnecessary. -sm */
5736 DEBUG_PRINT2 (" regstart: %d\n", POINTER_TO_OFFSET (regstart[*p]));
5737
5738 /* Move past the register number and inner group count. */
5739 p += 1;
5740 break;
5741
5742
5743 /* The stop_memory opcode represents the end of a group. Its
5744 argument is the same as start_memory's: the register number. */
5745 case stop_memory:
5746 DEBUG_PRINT2 ("EXECUTING stop_memory %d:\n", *p);
5747
5748 assert (!REG_UNSET (regstart[*p]));
5749 /* Strictly speaking, there should be code such as:
5750
5751 assert (REG_UNSET (regend[*p]));
5752 PUSH_FAILURE_REGSTOP ((unsigned int)*p);
5753
5754 But the only info to be pushed is regend[*p] and it is known to
5755 be UNSET, so there really isn't anything to push.
5756 Not pushing anything, on the other hand deprives us from the
5757 guarantee that regend[*p] is UNSET since undoing this operation
5758 will not reset its value properly. This is not important since
5759 the value will only be read on the next start_memory or at
5760 the very end and both events can only happen if this stop_memory
5761 is *not* undone. */
5762
5763 regend[*p] = d;
5764 DEBUG_PRINT2 (" regend: %d\n", POINTER_TO_OFFSET (regend[*p]));
5765
5766 /* Move past the register number and the inner group count. */
5767 p += 1;
5768 break;
5769
5770
5771 /* \<digit> has been turned into a `duplicate' command which is
5772 followed by the numeric value of <digit> as the register number. */
5773 case duplicate:
5774 {
5775 register re_char *d2, *dend2;
5776 int regno = *p++; /* Get which register to match against. */
5777 DEBUG_PRINT2 ("EXECUTING duplicate %d.\n", regno);
5778
5779 /* Can't back reference a group which we've never matched. */
5780 if (REG_UNSET (regstart[regno]) || REG_UNSET (regend[regno]))
5781 goto fail;
5782
5783 /* Where in input to try to start matching. */
5784 d2 = regstart[regno];
5785
5786 /* Remember the start point to rollback upon failure. */
5787 dfail = d;
5788
5789 /* Where to stop matching; if both the place to start and
5790 the place to stop matching are in the same string, then
5791 set to the place to stop, otherwise, for now have to use
5792 the end of the first string. */
5793
5794 dend2 = ((FIRST_STRING_P (regstart[regno])
5795 == FIRST_STRING_P (regend[regno]))
5796 ? regend[regno] : end_match_1);
5797 for (;;)
5798 {
5799 /* If necessary, advance to next segment in register
5800 contents. */
5801 while (d2 == dend2)
5802 {
5803 if (dend2 == end_match_2) break;
5804 if (dend2 == regend[regno]) break;
5805
5806 /* End of string1 => advance to string2. */
5807 d2 = string2;
5808 dend2 = regend[regno];
5809 }
5810 /* At end of register contents => success */
5811 if (d2 == dend2) break;
5812
5813 /* If necessary, advance to next segment in data. */
5814 PREFETCH ();
5815
5816 /* How many characters left in this segment to match. */
5817 mcnt = dend - d;
5818
5819 /* Want how many consecutive characters we can match in
5820 one shot, so, if necessary, adjust the count. */
5821 if (mcnt > dend2 - d2)
5822 mcnt = dend2 - d2;
5823
5824 /* Compare that many; failure if mismatch, else move
5825 past them. */
5826 if (RE_TRANSLATE_P (translate)
5827 ? bcmp_translate (d, d2, mcnt, translate, target_multibyte)
5828 : memcmp (d, d2, mcnt))
5829 {
5830 d = dfail;
5831 goto fail;
5832 }
5833 d += mcnt, d2 += mcnt;
5834 }
5835 }
5836 break;
5837
5838
5839 /* begline matches the empty string at the beginning of the string
5840 (unless `not_bol' is set in `bufp'), and after newlines. */
5841 case begline:
5842 DEBUG_PRINT1 ("EXECUTING begline.\n");
5843
5844 if (AT_STRINGS_BEG (d))
5845 {
5846 if (!bufp->not_bol) break;
5847 }
5848 else
5849 {
5850 unsigned c;
5851 GET_CHAR_BEFORE_2 (c, d, string1, end1, string2, end2);
5852 if (c == '\n')
5853 break;
5854 }
5855 /* In all other cases, we fail. */
5856 goto fail;
5857
5858
5859 /* endline is the dual of begline. */
5860 case endline:
5861 DEBUG_PRINT1 ("EXECUTING endline.\n");
5862
5863 if (AT_STRINGS_END (d))
5864 {
5865 if (!bufp->not_eol) break;
5866 }
5867 else
5868 {
5869 PREFETCH_NOLIMIT ();
5870 if (*d == '\n')
5871 break;
5872 }
5873 goto fail;
5874
5875
5876 /* Match at the very beginning of the data. */
5877 case begbuf:
5878 DEBUG_PRINT1 ("EXECUTING begbuf.\n");
5879 if (AT_STRINGS_BEG (d))
5880 break;
5881 goto fail;
5882
5883
5884 /* Match at the very end of the data. */
5885 case endbuf:
5886 DEBUG_PRINT1 ("EXECUTING endbuf.\n");
5887 if (AT_STRINGS_END (d))
5888 break;
5889 goto fail;
5890
5891
5892 /* on_failure_keep_string_jump is used to optimize `.*\n'. It
5893 pushes NULL as the value for the string on the stack. Then
5894 `POP_FAILURE_POINT' will keep the current value for the
5895 string, instead of restoring it. To see why, consider
5896 matching `foo\nbar' against `.*\n'. The .* matches the foo;
5897 then the . fails against the \n. But the next thing we want
5898 to do is match the \n against the \n; if we restored the
5899 string value, we would be back at the foo.
5900
5901 Because this is used only in specific cases, we don't need to
5902 check all the things that `on_failure_jump' does, to make
5903 sure the right things get saved on the stack. Hence we don't
5904 share its code. The only reason to push anything on the
5905 stack at all is that otherwise we would have to change
5906 `anychar's code to do something besides goto fail in this
5907 case; that seems worse than this. */
5908 case on_failure_keep_string_jump:
5909 EXTRACT_NUMBER_AND_INCR (mcnt, p);
5910 DEBUG_PRINT3 ("EXECUTING on_failure_keep_string_jump %d (to %p):\n",
5911 mcnt, p + mcnt);
5912
5913 PUSH_FAILURE_POINT (p - 3, NULL);
5914 break;
5915
5916 /* A nasty loop is introduced by the non-greedy *? and +?.
5917 With such loops, the stack only ever contains one failure point
5918 at a time, so that a plain on_failure_jump_loop kind of
5919 cycle detection cannot work. Worse yet, such a detection
5920 can not only fail to detect a cycle, but it can also wrongly
5921 detect a cycle (between different instantiations of the same
5922 loop).
5923 So the method used for those nasty loops is a little different:
5924 We use a special cycle-detection-stack-frame which is pushed
5925 when the on_failure_jump_nastyloop failure-point is *popped*.
5926 This special frame thus marks the beginning of one iteration
5927 through the loop and we can hence easily check right here
5928 whether something matched between the beginning and the end of
5929 the loop. */
5930 case on_failure_jump_nastyloop:
5931 EXTRACT_NUMBER_AND_INCR (mcnt, p);
5932 DEBUG_PRINT3 ("EXECUTING on_failure_jump_nastyloop %d (to %p):\n",
5933 mcnt, p + mcnt);
5934
5935 assert ((re_opcode_t)p[-4] == no_op);
5936 {
5937 int cycle = 0;
5938 CHECK_INFINITE_LOOP (p - 4, d);
5939 if (!cycle)
5940 /* If there's a cycle, just continue without pushing
5941 this failure point. The failure point is the "try again"
5942 option, which shouldn't be tried.
5943 We want (x?)*?y\1z to match both xxyz and xxyxz. */
5944 PUSH_FAILURE_POINT (p - 3, d);
5945 }
5946 break;
5947
5948 /* Simple loop detecting on_failure_jump: just check on the
5949 failure stack if the same spot was already hit earlier. */
5950 case on_failure_jump_loop:
5951 on_failure:
5952 EXTRACT_NUMBER_AND_INCR (mcnt, p);
5953 DEBUG_PRINT3 ("EXECUTING on_failure_jump_loop %d (to %p):\n",
5954 mcnt, p + mcnt);
5955 {
5956 int cycle = 0;
5957 CHECK_INFINITE_LOOP (p - 3, d);
5958 if (cycle)
5959 /* If there's a cycle, get out of the loop, as if the matching
5960 had failed. We used to just `goto fail' here, but that was
5961 aborting the search a bit too early: we want to keep the
5962 empty-loop-match and keep matching after the loop.
5963 We want (x?)*y\1z to match both xxyz and xxyxz. */
5964 p += mcnt;
5965 else
5966 PUSH_FAILURE_POINT (p - 3, d);
5967 }
5968 break;
5969
5970
5971 /* Uses of on_failure_jump:
5972
5973 Each alternative starts with an on_failure_jump that points
5974 to the beginning of the next alternative. Each alternative
5975 except the last ends with a jump that in effect jumps past
5976 the rest of the alternatives. (They really jump to the
5977 ending jump of the following alternative, because tensioning
5978 these jumps is a hassle.)
5979
5980 Repeats start with an on_failure_jump that points past both
5981 the repetition text and either the following jump or
5982 pop_failure_jump back to this on_failure_jump. */
5983 case on_failure_jump:
5984 EXTRACT_NUMBER_AND_INCR (mcnt, p);
5985 DEBUG_PRINT3 ("EXECUTING on_failure_jump %d (to %p):\n",
5986 mcnt, p + mcnt);
5987
5988 PUSH_FAILURE_POINT (p -3, d);
5989 break;
5990
5991 /* This operation is used for greedy *.
5992 Compare the beginning of the repeat with what in the
5993 pattern follows its end. If we can establish that there
5994 is nothing that they would both match, i.e., that we
5995 would have to backtrack because of (as in, e.g., `a*a')
5996 then we can use a non-backtracking loop based on
5997 on_failure_keep_string_jump instead of on_failure_jump. */
5998 case on_failure_jump_smart:
5999 EXTRACT_NUMBER_AND_INCR (mcnt, p);
6000 DEBUG_PRINT3 ("EXECUTING on_failure_jump_smart %d (to %p).\n",
6001 mcnt, p + mcnt);
6002 {
6003 re_char *p1 = p; /* Next operation. */
6004 /* Here, we discard `const', making re_match non-reentrant. */
6005 unsigned char *p2 = (unsigned char*) p + mcnt; /* Jump dest. */
6006 unsigned char *p3 = (unsigned char*) p - 3; /* opcode location. */
6007
6008 p -= 3; /* Reset so that we will re-execute the
6009 instruction once it's been changed. */
6010
6011 EXTRACT_NUMBER (mcnt, p2 - 2);
6012
6013 /* Ensure this is a indeed the trivial kind of loop
6014 we are expecting. */
6015 assert (skip_one_char (p1) == p2 - 3);
6016 assert ((re_opcode_t) p2[-3] == jump && p2 + mcnt == p);
6017 DEBUG_STATEMENT (debug += 2);
6018 if (mutually_exclusive_p (bufp, p1, p2))
6019 {
6020 /* Use a fast `on_failure_keep_string_jump' loop. */
6021 DEBUG_PRINT1 (" smart exclusive => fast loop.\n");
6022 *p3 = (unsigned char) on_failure_keep_string_jump;
6023 STORE_NUMBER (p2 - 2, mcnt + 3);
6024 }
6025 else
6026 {
6027 /* Default to a safe `on_failure_jump' loop. */
6028 DEBUG_PRINT1 (" smart default => slow loop.\n");
6029 *p3 = (unsigned char) on_failure_jump;
6030 }
6031 DEBUG_STATEMENT (debug -= 2);
6032 }
6033 break;
6034
6035 /* Unconditionally jump (without popping any failure points). */
6036 case jump:
6037 unconditional_jump:
6038 IMMEDIATE_QUIT_CHECK;
6039 EXTRACT_NUMBER_AND_INCR (mcnt, p); /* Get the amount to jump. */
6040 DEBUG_PRINT2 ("EXECUTING jump %d ", mcnt);
6041 p += mcnt; /* Do the jump. */
6042 DEBUG_PRINT2 ("(to %p).\n", p);
6043 break;
6044
6045
6046 /* Have to succeed matching what follows at least n times.
6047 After that, handle like `on_failure_jump'. */
6048 case succeed_n:
6049 /* Signedness doesn't matter since we only compare MCNT to 0. */
6050 EXTRACT_NUMBER (mcnt, p + 2);
6051 DEBUG_PRINT2 ("EXECUTING succeed_n %d.\n", mcnt);
6052
6053 /* Originally, mcnt is how many times we HAVE to succeed. */
6054 if (mcnt != 0)
6055 {
6056 /* Here, we discard `const', making re_match non-reentrant. */
6057 unsigned char *p2 = (unsigned char*) p + 2; /* counter loc. */
6058 mcnt--;
6059 p += 4;
6060 PUSH_NUMBER (p2, mcnt);
6061 }
6062 else
6063 /* The two bytes encoding mcnt == 0 are two no_op opcodes. */
6064 goto on_failure;
6065 break;
6066
6067 case jump_n:
6068 /* Signedness doesn't matter since we only compare MCNT to 0. */
6069 EXTRACT_NUMBER (mcnt, p + 2);
6070 DEBUG_PRINT2 ("EXECUTING jump_n %d.\n", mcnt);
6071
6072 /* Originally, this is how many times we CAN jump. */
6073 if (mcnt != 0)
6074 {
6075 /* Here, we discard `const', making re_match non-reentrant. */
6076 unsigned char *p2 = (unsigned char*) p + 2; /* counter loc. */
6077 mcnt--;
6078 PUSH_NUMBER (p2, mcnt);
6079 goto unconditional_jump;
6080 }
6081 /* If don't have to jump any more, skip over the rest of command. */
6082 else
6083 p += 4;
6084 break;
6085
6086 case set_number_at:
6087 {
6088 unsigned char *p2; /* Location of the counter. */
6089 DEBUG_PRINT1 ("EXECUTING set_number_at.\n");
6090
6091 EXTRACT_NUMBER_AND_INCR (mcnt, p);
6092 /* Here, we discard `const', making re_match non-reentrant. */
6093 p2 = (unsigned char*) p + mcnt;
6094 /* Signedness doesn't matter since we only copy MCNT's bits . */
6095 EXTRACT_NUMBER_AND_INCR (mcnt, p);
6096 DEBUG_PRINT3 (" Setting %p to %d.\n", p2, mcnt);
6097 PUSH_NUMBER (p2, mcnt);
6098 break;
6099 }
6100
6101 case wordbound:
6102 case notwordbound:
6103 not = (re_opcode_t) *(p - 1) == notwordbound;
6104 DEBUG_PRINT2 ("EXECUTING %swordbound.\n", not?"not":"");
6105
6106 /* We SUCCEED (or FAIL) in one of the following cases: */
6107
6108 /* Case 1: D is at the beginning or the end of string. */
6109 if (AT_STRINGS_BEG (d) || AT_STRINGS_END (d))
6110 not = !not;
6111 else
6112 {
6113 /* C1 is the character before D, S1 is the syntax of C1, C2
6114 is the character at D, and S2 is the syntax of C2. */
6115 re_wchar_t c1, c2;
6116 int s1, s2;
6117 int dummy;
6118 #ifdef emacs
6119 int offset = PTR_TO_OFFSET (d - 1);
6120 int charpos = SYNTAX_TABLE_BYTE_TO_CHAR (offset);
6121 UPDATE_SYNTAX_TABLE (charpos);
6122 #endif
6123 GET_CHAR_BEFORE_2 (c1, d, string1, end1, string2, end2);
6124 s1 = SYNTAX (c1);
6125 #ifdef emacs
6126 UPDATE_SYNTAX_TABLE_FORWARD (charpos + 1);
6127 #endif
6128 PREFETCH_NOLIMIT ();
6129 GET_CHAR_AFTER (c2, d, dummy);
6130 s2 = SYNTAX (c2);
6131
6132 if (/* Case 2: Only one of S1 and S2 is Sword. */
6133 ((s1 == Sword) != (s2 == Sword))
6134 /* Case 3: Both of S1 and S2 are Sword, and macro
6135 WORD_BOUNDARY_P (C1, C2) returns nonzero. */
6136 || ((s1 == Sword) && WORD_BOUNDARY_P (c1, c2)))
6137 not = !not;
6138 }
6139 if (not)
6140 break;
6141 else
6142 goto fail;
6143
6144 case wordbeg:
6145 DEBUG_PRINT1 ("EXECUTING wordbeg.\n");
6146
6147 /* We FAIL in one of the following cases: */
6148
6149 /* Case 1: D is at the end of string. */
6150 if (AT_STRINGS_END (d))
6151 goto fail;
6152 else
6153 {
6154 /* C1 is the character before D, S1 is the syntax of C1, C2
6155 is the character at D, and S2 is the syntax of C2. */
6156 re_wchar_t c1, c2;
6157 int s1, s2;
6158 int dummy;
6159 #ifdef emacs
6160 int offset = PTR_TO_OFFSET (d);
6161 int charpos = SYNTAX_TABLE_BYTE_TO_CHAR (offset);
6162 UPDATE_SYNTAX_TABLE (charpos);
6163 #endif
6164 PREFETCH ();
6165 GET_CHAR_AFTER (c2, d, dummy);
6166 s2 = SYNTAX (c2);
6167
6168 /* Case 2: S2 is not Sword. */
6169 if (s2 != Sword)
6170 goto fail;
6171
6172 /* Case 3: D is not at the beginning of string ... */
6173 if (!AT_STRINGS_BEG (d))
6174 {
6175 GET_CHAR_BEFORE_2 (c1, d, string1, end1, string2, end2);
6176 #ifdef emacs
6177 UPDATE_SYNTAX_TABLE_BACKWARD (charpos - 1);
6178 #endif
6179 s1 = SYNTAX (c1);
6180
6181 /* ... and S1 is Sword, and WORD_BOUNDARY_P (C1, C2)
6182 returns 0. */
6183 if ((s1 == Sword) && !WORD_BOUNDARY_P (c1, c2))
6184 goto fail;
6185 }
6186 }
6187 break;
6188
6189 case wordend:
6190 DEBUG_PRINT1 ("EXECUTING wordend.\n");
6191
6192 /* We FAIL in one of the following cases: */
6193
6194 /* Case 1: D is at the beginning of string. */
6195 if (AT_STRINGS_BEG (d))
6196 goto fail;
6197 else
6198 {
6199 /* C1 is the character before D, S1 is the syntax of C1, C2
6200 is the character at D, and S2 is the syntax of C2. */
6201 re_wchar_t c1, c2;
6202 int s1, s2;
6203 int dummy;
6204 #ifdef emacs
6205 int offset = PTR_TO_OFFSET (d) - 1;
6206 int charpos = SYNTAX_TABLE_BYTE_TO_CHAR (offset);
6207 UPDATE_SYNTAX_TABLE (charpos);
6208 #endif
6209 GET_CHAR_BEFORE_2 (c1, d, string1, end1, string2, end2);
6210 s1 = SYNTAX (c1);
6211
6212 /* Case 2: S1 is not Sword. */
6213 if (s1 != Sword)
6214 goto fail;
6215
6216 /* Case 3: D is not at the end of string ... */
6217 if (!AT_STRINGS_END (d))
6218 {
6219 PREFETCH_NOLIMIT ();
6220 GET_CHAR_AFTER (c2, d, dummy);
6221 #ifdef emacs
6222 UPDATE_SYNTAX_TABLE_FORWARD (charpos);
6223 #endif
6224 s2 = SYNTAX (c2);
6225
6226 /* ... and S2 is Sword, and WORD_BOUNDARY_P (C1, C2)
6227 returns 0. */
6228 if ((s2 == Sword) && !WORD_BOUNDARY_P (c1, c2))
6229 goto fail;
6230 }
6231 }
6232 break;
6233
6234 case symbeg:
6235 DEBUG_PRINT1 ("EXECUTING symbeg.\n");
6236
6237 /* We FAIL in one of the following cases: */
6238
6239 /* Case 1: D is at the end of string. */
6240 if (AT_STRINGS_END (d))
6241 goto fail;
6242 else
6243 {
6244 /* C1 is the character before D, S1 is the syntax of C1, C2
6245 is the character at D, and S2 is the syntax of C2. */
6246 re_wchar_t c1, c2;
6247 int s1, s2;
6248 #ifdef emacs
6249 int offset = PTR_TO_OFFSET (d);
6250 int charpos = SYNTAX_TABLE_BYTE_TO_CHAR (offset);
6251 UPDATE_SYNTAX_TABLE (charpos);
6252 #endif
6253 PREFETCH ();
6254 c2 = RE_STRING_CHAR (d, dend - d, target_multibyte);
6255 s2 = SYNTAX (c2);
6256
6257 /* Case 2: S2 is neither Sword nor Ssymbol. */
6258 if (s2 != Sword && s2 != Ssymbol)
6259 goto fail;
6260
6261 /* Case 3: D is not at the beginning of string ... */
6262 if (!AT_STRINGS_BEG (d))
6263 {
6264 GET_CHAR_BEFORE_2 (c1, d, string1, end1, string2, end2);
6265 #ifdef emacs
6266 UPDATE_SYNTAX_TABLE_BACKWARD (charpos - 1);
6267 #endif
6268 s1 = SYNTAX (c1);
6269
6270 /* ... and S1 is Sword or Ssymbol. */
6271 if (s1 == Sword || s1 == Ssymbol)
6272 goto fail;
6273 }
6274 }
6275 break;
6276
6277 case symend:
6278 DEBUG_PRINT1 ("EXECUTING symend.\n");
6279
6280 /* We FAIL in one of the following cases: */
6281
6282 /* Case 1: D is at the beginning of string. */
6283 if (AT_STRINGS_BEG (d))
6284 goto fail;
6285 else
6286 {
6287 /* C1 is the character before D, S1 is the syntax of C1, C2
6288 is the character at D, and S2 is the syntax of C2. */
6289 re_wchar_t c1, c2;
6290 int s1, s2;
6291 #ifdef emacs
6292 int offset = PTR_TO_OFFSET (d) - 1;
6293 int charpos = SYNTAX_TABLE_BYTE_TO_CHAR (offset);
6294 UPDATE_SYNTAX_TABLE (charpos);
6295 #endif
6296 GET_CHAR_BEFORE_2 (c1, d, string1, end1, string2, end2);
6297 s1 = SYNTAX (c1);
6298
6299 /* Case 2: S1 is neither Ssymbol nor Sword. */
6300 if (s1 != Sword && s1 != Ssymbol)
6301 goto fail;
6302
6303 /* Case 3: D is not at the end of string ... */
6304 if (!AT_STRINGS_END (d))
6305 {
6306 PREFETCH_NOLIMIT ();
6307 c2 = RE_STRING_CHAR (d, dend - d, target_multibyte);
6308 #ifdef emacs
6309 UPDATE_SYNTAX_TABLE_FORWARD (charpos + 1);
6310 #endif
6311 s2 = SYNTAX (c2);
6312
6313 /* ... and S2 is Sword or Ssymbol. */
6314 if (s2 == Sword || s2 == Ssymbol)
6315 goto fail;
6316 }
6317 }
6318 break;
6319
6320 case syntaxspec:
6321 case notsyntaxspec:
6322 not = (re_opcode_t) *(p - 1) == notsyntaxspec;
6323 mcnt = *p++;
6324 DEBUG_PRINT3 ("EXECUTING %ssyntaxspec %d.\n", not?"not":"", mcnt);
6325 PREFETCH ();
6326 #ifdef emacs
6327 {
6328 int offset = PTR_TO_OFFSET (d);
6329 int pos1 = SYNTAX_TABLE_BYTE_TO_CHAR (offset);
6330 UPDATE_SYNTAX_TABLE (pos1);
6331 }
6332 #endif
6333 {
6334 int len;
6335 re_wchar_t c;
6336
6337 GET_CHAR_AFTER (c, d, len);
6338 if ((SYNTAX (c) != (enum syntaxcode) mcnt) ^ not)
6339 goto fail;
6340 d += len;
6341 }
6342 break;
6343
6344 #ifdef emacs
6345 case before_dot:
6346 DEBUG_PRINT1 ("EXECUTING before_dot.\n");
6347 if (PTR_BYTE_POS (d) >= PT_BYTE)
6348 goto fail;
6349 break;
6350
6351 case at_dot:
6352 DEBUG_PRINT1 ("EXECUTING at_dot.\n");
6353 if (PTR_BYTE_POS (d) != PT_BYTE)
6354 goto fail;
6355 break;
6356
6357 case after_dot:
6358 DEBUG_PRINT1 ("EXECUTING after_dot.\n");
6359 if (PTR_BYTE_POS (d) <= PT_BYTE)
6360 goto fail;
6361 break;
6362
6363 case categoryspec:
6364 case notcategoryspec:
6365 not = (re_opcode_t) *(p - 1) == notcategoryspec;
6366 mcnt = *p++;
6367 DEBUG_PRINT3 ("EXECUTING %scategoryspec %d.\n", not?"not":"", mcnt);
6368 PREFETCH ();
6369 {
6370 int len;
6371 re_wchar_t c;
6372
6373 GET_CHAR_AFTER (c, d, len);
6374 if ((!CHAR_HAS_CATEGORY (c, mcnt)) ^ not)
6375 goto fail;
6376 d += len;
6377 }
6378 break;
6379
6380 #endif /* emacs */
6381
6382 default:
6383 abort ();
6384 }
6385 continue; /* Successfully executed one pattern command; keep going. */
6386
6387
6388 /* We goto here if a matching operation fails. */
6389 fail:
6390 IMMEDIATE_QUIT_CHECK;
6391 if (!FAIL_STACK_EMPTY ())
6392 {
6393 re_char *str, *pat;
6394 /* A restart point is known. Restore to that state. */
6395 DEBUG_PRINT1 ("\nFAIL:\n");
6396 POP_FAILURE_POINT (str, pat);
6397 switch (SWITCH_ENUM_CAST ((re_opcode_t) *pat++))
6398 {
6399 case on_failure_keep_string_jump:
6400 assert (str == NULL);
6401 goto continue_failure_jump;
6402
6403 case on_failure_jump_nastyloop:
6404 assert ((re_opcode_t)pat[-2] == no_op);
6405 PUSH_FAILURE_POINT (pat - 2, str);
6406 /* Fallthrough */
6407
6408 case on_failure_jump_loop:
6409 case on_failure_jump:
6410 case succeed_n:
6411 d = str;
6412 continue_failure_jump:
6413 EXTRACT_NUMBER_AND_INCR (mcnt, pat);
6414 p = pat + mcnt;
6415 break;
6416
6417 case no_op:
6418 /* A special frame used for nastyloops. */
6419 goto fail;
6420
6421 default:
6422 abort();
6423 }
6424
6425 assert (p >= bufp->buffer && p <= pend);
6426
6427 if (d >= string1 && d <= end1)
6428 dend = end_match_1;
6429 }
6430 else
6431 break; /* Matching at this starting point really fails. */
6432 } /* for (;;) */
6433
6434 if (best_regs_set)
6435 goto restore_best_regs;
6436
6437 FREE_VARIABLES ();
6438
6439 return -1; /* Failure to match. */
6440 } /* re_match_2 */
6441 \f
6442 /* Subroutine definitions for re_match_2. */
6443
6444 /* Return zero if TRANSLATE[S1] and TRANSLATE[S2] are identical for LEN
6445 bytes; nonzero otherwise. */
6446
6447 static int
6448 bcmp_translate (s1, s2, len, translate, target_multibyte)
6449 re_char *s1, *s2;
6450 register int len;
6451 RE_TRANSLATE_TYPE translate;
6452 const int target_multibyte;
6453 {
6454 register re_char *p1 = s1, *p2 = s2;
6455 re_char *p1_end = s1 + len;
6456 re_char *p2_end = s2 + len;
6457
6458 /* FIXME: Checking both p1 and p2 presumes that the two strings might have
6459 different lengths, but relying on a single `len' would break this. -sm */
6460 while (p1 < p1_end && p2 < p2_end)
6461 {
6462 int p1_charlen, p2_charlen;
6463 re_wchar_t p1_ch, p2_ch;
6464
6465 GET_CHAR_AFTER (p1_ch, p1, p1_charlen);
6466 GET_CHAR_AFTER (p2_ch, p2, p2_charlen);
6467
6468 if (RE_TRANSLATE (translate, p1_ch)
6469 != RE_TRANSLATE (translate, p2_ch))
6470 return 1;
6471
6472 p1 += p1_charlen, p2 += p2_charlen;
6473 }
6474
6475 if (p1 != p1_end || p2 != p2_end)
6476 return 1;
6477
6478 return 0;
6479 }
6480 \f
6481 /* Entry points for GNU code. */
6482
6483 /* re_compile_pattern is the GNU regular expression compiler: it
6484 compiles PATTERN (of length SIZE) and puts the result in BUFP.
6485 Returns 0 if the pattern was valid, otherwise an error string.
6486
6487 Assumes the `allocated' (and perhaps `buffer') and `translate' fields
6488 are set in BUFP on entry.
6489
6490 We call regex_compile to do the actual compilation. */
6491
6492 const char *
6493 re_compile_pattern (pattern, length, bufp)
6494 const char *pattern;
6495 size_t length;
6496 struct re_pattern_buffer *bufp;
6497 {
6498 reg_errcode_t ret;
6499
6500 #ifdef emacs
6501 gl_state.current_syntax_table = current_buffer->syntax_table;
6502 #endif
6503
6504 /* GNU code is written to assume at least RE_NREGS registers will be set
6505 (and at least one extra will be -1). */
6506 bufp->regs_allocated = REGS_UNALLOCATED;
6507
6508 /* And GNU code determines whether or not to get register information
6509 by passing null for the REGS argument to re_match, etc., not by
6510 setting no_sub. */
6511 bufp->no_sub = 0;
6512
6513 ret = regex_compile ((re_char*) pattern, length, re_syntax_options, bufp);
6514
6515 if (!ret)
6516 return NULL;
6517 return gettext (re_error_msgid[(int) ret]);
6518 }
6519 WEAK_ALIAS (__re_compile_pattern, re_compile_pattern)
6520 \f
6521 /* Entry points compatible with 4.2 BSD regex library. We don't define
6522 them unless specifically requested. */
6523
6524 #if defined _REGEX_RE_COMP || defined _LIBC
6525
6526 /* BSD has one and only one pattern buffer. */
6527 static struct re_pattern_buffer re_comp_buf;
6528
6529 char *
6530 # ifdef _LIBC
6531 /* Make these definitions weak in libc, so POSIX programs can redefine
6532 these names if they don't use our functions, and still use
6533 regcomp/regexec below without link errors. */
6534 weak_function
6535 # endif
6536 re_comp (s)
6537 const char *s;
6538 {
6539 reg_errcode_t ret;
6540
6541 if (!s)
6542 {
6543 if (!re_comp_buf.buffer)
6544 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6545 return (char *) gettext ("No previous regular expression");
6546 return 0;
6547 }
6548
6549 if (!re_comp_buf.buffer)
6550 {
6551 re_comp_buf.buffer = (unsigned char *) malloc (200);
6552 if (re_comp_buf.buffer == NULL)
6553 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6554 return (char *) gettext (re_error_msgid[(int) REG_ESPACE]);
6555 re_comp_buf.allocated = 200;
6556
6557 re_comp_buf.fastmap = (char *) malloc (1 << BYTEWIDTH);
6558 if (re_comp_buf.fastmap == NULL)
6559 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6560 return (char *) gettext (re_error_msgid[(int) REG_ESPACE]);
6561 }
6562
6563 /* Since `re_exec' always passes NULL for the `regs' argument, we
6564 don't need to initialize the pattern buffer fields which affect it. */
6565
6566 ret = regex_compile (s, strlen (s), re_syntax_options, &re_comp_buf);
6567
6568 if (!ret)
6569 return NULL;
6570
6571 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6572 return (char *) gettext (re_error_msgid[(int) ret]);
6573 }
6574
6575
6576 int
6577 # ifdef _LIBC
6578 weak_function
6579 # endif
6580 re_exec (s)
6581 const char *s;
6582 {
6583 const int len = strlen (s);
6584 return
6585 0 <= re_search (&re_comp_buf, s, len, 0, len, (struct re_registers *) 0);
6586 }
6587 #endif /* _REGEX_RE_COMP */
6588 \f
6589 /* POSIX.2 functions. Don't define these for Emacs. */
6590
6591 #ifndef emacs
6592
6593 /* regcomp takes a regular expression as a string and compiles it.
6594
6595 PREG is a regex_t *. We do not expect any fields to be initialized,
6596 since POSIX says we shouldn't. Thus, we set
6597
6598 `buffer' to the compiled pattern;
6599 `used' to the length of the compiled pattern;
6600 `syntax' to RE_SYNTAX_POSIX_EXTENDED if the
6601 REG_EXTENDED bit in CFLAGS is set; otherwise, to
6602 RE_SYNTAX_POSIX_BASIC;
6603 `fastmap' to an allocated space for the fastmap;
6604 `fastmap_accurate' to zero;
6605 `re_nsub' to the number of subexpressions in PATTERN.
6606
6607 PATTERN is the address of the pattern string.
6608
6609 CFLAGS is a series of bits which affect compilation.
6610
6611 If REG_EXTENDED is set, we use POSIX extended syntax; otherwise, we
6612 use POSIX basic syntax.
6613
6614 If REG_NEWLINE is set, then . and [^...] don't match newline.
6615 Also, regexec will try a match beginning after every newline.
6616
6617 If REG_ICASE is set, then we considers upper- and lowercase
6618 versions of letters to be equivalent when matching.
6619
6620 If REG_NOSUB is set, then when PREG is passed to regexec, that
6621 routine will report only success or failure, and nothing about the
6622 registers.
6623
6624 It returns 0 if it succeeds, nonzero if it doesn't. (See regex.h for
6625 the return codes and their meanings.) */
6626
6627 int
6628 regcomp (preg, pattern, cflags)
6629 regex_t *__restrict preg;
6630 const char *__restrict pattern;
6631 int cflags;
6632 {
6633 reg_errcode_t ret;
6634 reg_syntax_t syntax
6635 = (cflags & REG_EXTENDED) ?
6636 RE_SYNTAX_POSIX_EXTENDED : RE_SYNTAX_POSIX_BASIC;
6637
6638 /* regex_compile will allocate the space for the compiled pattern. */
6639 preg->buffer = 0;
6640 preg->allocated = 0;
6641 preg->used = 0;
6642
6643 /* Try to allocate space for the fastmap. */
6644 preg->fastmap = (char *) malloc (1 << BYTEWIDTH);
6645
6646 if (cflags & REG_ICASE)
6647 {
6648 unsigned i;
6649
6650 preg->translate
6651 = (RE_TRANSLATE_TYPE) malloc (CHAR_SET_SIZE
6652 * sizeof (*(RE_TRANSLATE_TYPE)0));
6653 if (preg->translate == NULL)
6654 return (int) REG_ESPACE;
6655
6656 /* Map uppercase characters to corresponding lowercase ones. */
6657 for (i = 0; i < CHAR_SET_SIZE; i++)
6658 preg->translate[i] = ISUPPER (i) ? TOLOWER (i) : i;
6659 }
6660 else
6661 preg->translate = NULL;
6662
6663 /* If REG_NEWLINE is set, newlines are treated differently. */
6664 if (cflags & REG_NEWLINE)
6665 { /* REG_NEWLINE implies neither . nor [^...] match newline. */
6666 syntax &= ~RE_DOT_NEWLINE;
6667 syntax |= RE_HAT_LISTS_NOT_NEWLINE;
6668 }
6669 else
6670 syntax |= RE_NO_NEWLINE_ANCHOR;
6671
6672 preg->no_sub = !!(cflags & REG_NOSUB);
6673
6674 /* POSIX says a null character in the pattern terminates it, so we
6675 can use strlen here in compiling the pattern. */
6676 ret = regex_compile ((re_char*) pattern, strlen (pattern), syntax, preg);
6677
6678 /* POSIX doesn't distinguish between an unmatched open-group and an
6679 unmatched close-group: both are REG_EPAREN. */
6680 if (ret == REG_ERPAREN)
6681 ret = REG_EPAREN;
6682
6683 if (ret == REG_NOERROR && preg->fastmap)
6684 { /* Compute the fastmap now, since regexec cannot modify the pattern
6685 buffer. */
6686 re_compile_fastmap (preg);
6687 if (preg->can_be_null)
6688 { /* The fastmap can't be used anyway. */
6689 free (preg->fastmap);
6690 preg->fastmap = NULL;
6691 }
6692 }
6693 return (int) ret;
6694 }
6695 WEAK_ALIAS (__regcomp, regcomp)
6696
6697
6698 /* regexec searches for a given pattern, specified by PREG, in the
6699 string STRING.
6700
6701 If NMATCH is zero or REG_NOSUB was set in the cflags argument to
6702 `regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at
6703 least NMATCH elements, and we set them to the offsets of the
6704 corresponding matched substrings.
6705
6706 EFLAGS specifies `execution flags' which affect matching: if
6707 REG_NOTBOL is set, then ^ does not match at the beginning of the
6708 string; if REG_NOTEOL is set, then $ does not match at the end.
6709
6710 We return 0 if we find a match and REG_NOMATCH if not. */
6711
6712 int
6713 regexec (preg, string, nmatch, pmatch, eflags)
6714 const regex_t *__restrict preg;
6715 const char *__restrict string;
6716 size_t nmatch;
6717 regmatch_t pmatch[__restrict_arr];
6718 int eflags;
6719 {
6720 int ret;
6721 struct re_registers regs;
6722 regex_t private_preg;
6723 int len = strlen (string);
6724 boolean want_reg_info = !preg->no_sub && nmatch > 0 && pmatch;
6725
6726 private_preg = *preg;
6727
6728 private_preg.not_bol = !!(eflags & REG_NOTBOL);
6729 private_preg.not_eol = !!(eflags & REG_NOTEOL);
6730
6731 /* The user has told us exactly how many registers to return
6732 information about, via `nmatch'. We have to pass that on to the
6733 matching routines. */
6734 private_preg.regs_allocated = REGS_FIXED;
6735
6736 if (want_reg_info)
6737 {
6738 regs.num_regs = nmatch;
6739 regs.start = TALLOC (nmatch * 2, regoff_t);
6740 if (regs.start == NULL)
6741 return (int) REG_NOMATCH;
6742 regs.end = regs.start + nmatch;
6743 }
6744
6745 /* Instead of using not_eol to implement REG_NOTEOL, we could simply
6746 pass (&private_preg, string, len + 1, 0, len, ...) pretending the string
6747 was a little bit longer but still only matching the real part.
6748 This works because the `endline' will check for a '\n' and will find a
6749 '\0', correctly deciding that this is not the end of a line.
6750 But it doesn't work out so nicely for REG_NOTBOL, since we don't have
6751 a convenient '\0' there. For all we know, the string could be preceded
6752 by '\n' which would throw things off. */
6753
6754 /* Perform the searching operation. */
6755 ret = re_search (&private_preg, string, len,
6756 /* start: */ 0, /* range: */ len,
6757 want_reg_info ? &regs : (struct re_registers *) 0);
6758
6759 /* Copy the register information to the POSIX structure. */
6760 if (want_reg_info)
6761 {
6762 if (ret >= 0)
6763 {
6764 unsigned r;
6765
6766 for (r = 0; r < nmatch; r++)
6767 {
6768 pmatch[r].rm_so = regs.start[r];
6769 pmatch[r].rm_eo = regs.end[r];
6770 }
6771 }
6772
6773 /* If we needed the temporary register info, free the space now. */
6774 free (regs.start);
6775 }
6776
6777 /* We want zero return to mean success, unlike `re_search'. */
6778 return ret >= 0 ? (int) REG_NOERROR : (int) REG_NOMATCH;
6779 }
6780 WEAK_ALIAS (__regexec, regexec)
6781
6782
6783 /* Returns a message corresponding to an error code, ERR_CODE, returned
6784 from either regcomp or regexec. We don't use PREG here.
6785
6786 ERR_CODE was previously called ERRCODE, but that name causes an
6787 error with msvc8 compiler. */
6788
6789 size_t
6790 regerror (err_code, preg, errbuf, errbuf_size)
6791 int err_code;
6792 const regex_t *preg;
6793 char *errbuf;
6794 size_t errbuf_size;
6795 {
6796 const char *msg;
6797 size_t msg_size;
6798
6799 if (err_code < 0
6800 || err_code >= (sizeof (re_error_msgid) / sizeof (re_error_msgid[0])))
6801 /* Only error codes returned by the rest of the code should be passed
6802 to this routine. If we are given anything else, or if other regex
6803 code generates an invalid error code, then the program has a bug.
6804 Dump core so we can fix it. */
6805 abort ();
6806
6807 msg = gettext (re_error_msgid[err_code]);
6808
6809 msg_size = strlen (msg) + 1; /* Includes the null. */
6810
6811 if (errbuf_size != 0)
6812 {
6813 if (msg_size > errbuf_size)
6814 {
6815 strncpy (errbuf, msg, errbuf_size - 1);
6816 errbuf[errbuf_size - 1] = 0;
6817 }
6818 else
6819 strcpy (errbuf, msg);
6820 }
6821
6822 return msg_size;
6823 }
6824 WEAK_ALIAS (__regerror, regerror)
6825
6826
6827 /* Free dynamically allocated space used by PREG. */
6828
6829 void
6830 regfree (preg)
6831 regex_t *preg;
6832 {
6833 if (preg->buffer != NULL)
6834 free (preg->buffer);
6835 preg->buffer = NULL;
6836
6837 preg->allocated = 0;
6838 preg->used = 0;
6839
6840 if (preg->fastmap != NULL)
6841 free (preg->fastmap);
6842 preg->fastmap = NULL;
6843 preg->fastmap_accurate = 0;
6844
6845 if (preg->translate != NULL)
6846 free (preg->translate);
6847 preg->translate = NULL;
6848 }
6849 WEAK_ALIAS (__regfree, regfree)
6850
6851 #endif /* not emacs */
6852
6853 /* arch-tag: 4ffd68ba-2a9e-435b-a21a-018990f9eeb2
6854 (do not change this comment) */
Unnamed repository; edit this file 'description' to name the repository.