HCoop / bpt / emacs.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
(syms_of_ntproc) <w32-get-true-file-attributes>: Doc fix.
[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, 2008
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 # define SWITCH_ENUM_CAST(x) (x)
293
294 /* Dummy macros for non-Emacs environments. */
295 # define BASE_LEADING_CODE_P(c) (0)
296 # define CHAR_CHARSET(c) 0
297 # define CHARSET_LEADING_CODE_BASE(c) 0
298 # define MAX_MULTIBYTE_LENGTH 1
299 # define RE_MULTIBYTE_P(x) 0
300 # define RE_TARGET_MULTIBYTE_P(x) 0
301 # define WORD_BOUNDARY_P(c1, c2) (0)
302 # define CHAR_HEAD_P(p) (1)
303 # define SINGLE_BYTE_CHAR_P(c) (1)
304 # define SAME_CHARSET_P(c1, c2) (1)
305 # define MULTIBYTE_FORM_LENGTH(p, s) (1)
306 # define PREV_CHAR_BOUNDARY(p, limit) ((p)--)
307 # define STRING_CHAR(p, s) (*(p))
308 # define RE_STRING_CHAR(p, s, multibyte) STRING_CHAR ((p), (s))
309 # define CHAR_STRING(c, s) (*(s) = (c), 1)
310 # define STRING_CHAR_AND_LENGTH(p, s, actual_len) ((actual_len) = 1, *(p))
311 # define RE_STRING_CHAR_AND_LENGTH(p, s, len, multibyte) STRING_CHAR_AND_LENGTH ((p), (s), (len))
312 # define RE_CHAR_TO_MULTIBYTE(c) (c)
313 # define RE_CHAR_TO_UNIBYTE(c) (c)
314 # define GET_CHAR_BEFORE_2(c, p, str1, end1, str2, end2) \
315 (c = ((p) == (str2) ? *((end1) - 1) : *((p) - 1)))
316 # define GET_CHAR_AFTER(c, p, len) \
317 (c = *p, len = 1)
318 # define MAKE_CHAR(charset, c1, c2) (c1)
319 # define BYTE8_TO_CHAR(c) (c)
320 # define CHAR_BYTE8_P(c) (0)
321 # define CHAR_LEADING_CODE(c) (c)
322
323 #endif /* not emacs */
324
325 #ifndef RE_TRANSLATE
326 # define RE_TRANSLATE(TBL, C) ((unsigned char)(TBL)[C])
327 # define RE_TRANSLATE_P(TBL) (TBL)
328 #endif
329 \f
330 /* Get the interface, including the syntax bits. */
331 #include "regex.h"
332
333 /* isalpha etc. are used for the character classes. */
334 #include <ctype.h>
335
336 #ifdef emacs
337
338 /* 1 if C is an ASCII character. */
339 # define IS_REAL_ASCII(c) ((c) < 0200)
340
341 /* 1 if C is a unibyte character. */
342 # define ISUNIBYTE(c) (SINGLE_BYTE_CHAR_P ((c)))
343
344 /* The Emacs definitions should not be directly affected by locales. */
345
346 /* In Emacs, these are only used for single-byte characters. */
347 # define ISDIGIT(c) ((c) >= '0' && (c) <= '9')
348 # define ISCNTRL(c) ((c) < ' ')
349 # define ISXDIGIT(c) (((c) >= '0' && (c) <= '9') \
350 || ((c) >= 'a' && (c) <= 'f') \
351 || ((c) >= 'A' && (c) <= 'F'))
352
353 /* This is only used for single-byte characters. */
354 # define ISBLANK(c) ((c) == ' ' || (c) == '\t')
355
356 /* The rest must handle multibyte characters. */
357
358 # define ISGRAPH(c) (SINGLE_BYTE_CHAR_P (c) \
359 ? (c) > ' ' && !((c) >= 0177 && (c) <= 0237) \
360 : 1)
361
362 # define ISPRINT(c) (SINGLE_BYTE_CHAR_P (c) \
363 ? (c) >= ' ' && !((c) >= 0177 && (c) <= 0237) \
364 : 1)
365
366 # define ISALNUM(c) (IS_REAL_ASCII (c) \
367 ? (((c) >= 'a' && (c) <= 'z') \
368 || ((c) >= 'A' && (c) <= 'Z') \
369 || ((c) >= '0' && (c) <= '9')) \
370 : SYNTAX (c) == Sword)
371
372 # define ISALPHA(c) (IS_REAL_ASCII (c) \
373 ? (((c) >= 'a' && (c) <= 'z') \
374 || ((c) >= 'A' && (c) <= 'Z')) \
375 : SYNTAX (c) == Sword)
376
377 # define ISLOWER(c) (LOWERCASEP (c))
378
379 # define ISPUNCT(c) (IS_REAL_ASCII (c) \
380 ? ((c) > ' ' && (c) < 0177 \
381 && !(((c) >= 'a' && (c) <= 'z') \
382 || ((c) >= 'A' && (c) <= 'Z') \
383 || ((c) >= '0' && (c) <= '9'))) \
384 : SYNTAX (c) != Sword)
385
386 # define ISSPACE(c) (SYNTAX (c) == Swhitespace)
387
388 # define ISUPPER(c) (UPPERCASEP (c))
389
390 # define ISWORD(c) (SYNTAX (c) == Sword)
391
392 #else /* not emacs */
393
394 /* Jim Meyering writes:
395
396 "... Some ctype macros are valid only for character codes that
397 isascii says are ASCII (SGI's IRIX-4.0.5 is one such system --when
398 using /bin/cc or gcc but without giving an ansi option). So, all
399 ctype uses should be through macros like ISPRINT... If
400 STDC_HEADERS is defined, then autoconf has verified that the ctype
401 macros don't need to be guarded with references to isascii. ...
402 Defining isascii to 1 should let any compiler worth its salt
403 eliminate the && through constant folding."
404 Solaris defines some of these symbols so we must undefine them first. */
405
406 # undef ISASCII
407 # if defined STDC_HEADERS || (!defined isascii && !defined HAVE_ISASCII)
408 # define ISASCII(c) 1
409 # else
410 # define ISASCII(c) isascii(c)
411 # endif
412
413 /* 1 if C is an ASCII character. */
414 # define IS_REAL_ASCII(c) ((c) < 0200)
415
416 /* This distinction is not meaningful, except in Emacs. */
417 # define ISUNIBYTE(c) 1
418
419 # ifdef isblank
420 # define ISBLANK(c) (ISASCII (c) && isblank (c))
421 # else
422 # define ISBLANK(c) ((c) == ' ' || (c) == '\t')
423 # endif
424 # ifdef isgraph
425 # define ISGRAPH(c) (ISASCII (c) && isgraph (c))
426 # else
427 # define ISGRAPH(c) (ISASCII (c) && isprint (c) && !isspace (c))
428 # endif
429
430 # undef ISPRINT
431 # define ISPRINT(c) (ISASCII (c) && isprint (c))
432 # define ISDIGIT(c) (ISASCII (c) && isdigit (c))
433 # define ISALNUM(c) (ISASCII (c) && isalnum (c))
434 # define ISALPHA(c) (ISASCII (c) && isalpha (c))
435 # define ISCNTRL(c) (ISASCII (c) && iscntrl (c))
436 # define ISLOWER(c) (ISASCII (c) && islower (c))
437 # define ISPUNCT(c) (ISASCII (c) && ispunct (c))
438 # define ISSPACE(c) (ISASCII (c) && isspace (c))
439 # define ISUPPER(c) (ISASCII (c) && isupper (c))
440 # define ISXDIGIT(c) (ISASCII (c) && isxdigit (c))
441
442 # define ISWORD(c) ISALPHA(c)
443
444 # ifdef _tolower
445 # define TOLOWER(c) _tolower(c)
446 # else
447 # define TOLOWER(c) tolower(c)
448 # endif
449
450 /* How many characters in the character set. */
451 # define CHAR_SET_SIZE 256
452
453 # ifdef SYNTAX_TABLE
454
455 extern char *re_syntax_table;
456
457 # else /* not SYNTAX_TABLE */
458
459 static char re_syntax_table[CHAR_SET_SIZE];
460
461 static void
462 init_syntax_once ()
463 {
464 register int c;
465 static int done = 0;
466
467 if (done)
468 return;
469
470 bzero (re_syntax_table, sizeof re_syntax_table);
471
472 for (c = 0; c < CHAR_SET_SIZE; ++c)
473 if (ISALNUM (c))
474 re_syntax_table[c] = Sword;
475
476 re_syntax_table['_'] = Ssymbol;
477
478 done = 1;
479 }
480
481 # endif /* not SYNTAX_TABLE */
482
483 # define SYNTAX(c) re_syntax_table[(c)]
484
485 #endif /* not emacs */
486 \f
487 #ifndef NULL
488 # define NULL (void *)0
489 #endif
490
491 /* We remove any previous definition of `SIGN_EXTEND_CHAR',
492 since ours (we hope) works properly with all combinations of
493 machines, compilers, `char' and `unsigned char' argument types.
494 (Per Bothner suggested the basic approach.) */
495 #undef SIGN_EXTEND_CHAR
496 #if __STDC__
497 # define SIGN_EXTEND_CHAR(c) ((signed char) (c))
498 #else /* not __STDC__ */
499 /* As in Harbison and Steele. */
500 # define SIGN_EXTEND_CHAR(c) ((((unsigned char) (c)) ^ 128) - 128)
501 #endif
502 \f
503 /* Should we use malloc or alloca? If REGEX_MALLOC is not defined, we
504 use `alloca' instead of `malloc'. This is because using malloc in
505 re_search* or re_match* could cause memory leaks when C-g is used in
506 Emacs; also, malloc is slower and causes storage fragmentation. On
507 the other hand, malloc is more portable, and easier to debug.
508
509 Because we sometimes use alloca, some routines have to be macros,
510 not functions -- `alloca'-allocated space disappears at the end of the
511 function it is called in. */
512
513 #ifdef REGEX_MALLOC
514
515 # define REGEX_ALLOCATE malloc
516 # define REGEX_REALLOCATE(source, osize, nsize) realloc (source, nsize)
517 # define REGEX_FREE free
518
519 #else /* not REGEX_MALLOC */
520
521 /* Emacs already defines alloca, sometimes. */
522 # ifndef alloca
523
524 /* Make alloca work the best possible way. */
525 # ifdef __GNUC__
526 # define alloca __builtin_alloca
527 # else /* not __GNUC__ */
528 # ifdef HAVE_ALLOCA_H
529 # include <alloca.h>
530 # endif /* HAVE_ALLOCA_H */
531 # endif /* not __GNUC__ */
532
533 # endif /* not alloca */
534
535 # define REGEX_ALLOCATE alloca
536
537 /* Assumes a `char *destination' variable. */
538 # define REGEX_REALLOCATE(source, osize, nsize) \
539 (destination = (char *) alloca (nsize), \
540 memcpy (destination, source, osize))
541
542 /* No need to do anything to free, after alloca. */
543 # define REGEX_FREE(arg) ((void)0) /* Do nothing! But inhibit gcc warning. */
544
545 #endif /* not REGEX_MALLOC */
546
547 /* Define how to allocate the failure stack. */
548
549 #if defined REL_ALLOC && defined REGEX_MALLOC
550
551 # define REGEX_ALLOCATE_STACK(size) \
552 r_alloc (&failure_stack_ptr, (size))
553 # define REGEX_REALLOCATE_STACK(source, osize, nsize) \
554 r_re_alloc (&failure_stack_ptr, (nsize))
555 # define REGEX_FREE_STACK(ptr) \
556 r_alloc_free (&failure_stack_ptr)
557
558 #else /* not using relocating allocator */
559
560 # ifdef REGEX_MALLOC
561
562 # define REGEX_ALLOCATE_STACK malloc
563 # define REGEX_REALLOCATE_STACK(source, osize, nsize) realloc (source, nsize)
564 # define REGEX_FREE_STACK free
565
566 # else /* not REGEX_MALLOC */
567
568 # define REGEX_ALLOCATE_STACK alloca
569
570 # define REGEX_REALLOCATE_STACK(source, osize, nsize) \
571 REGEX_REALLOCATE (source, osize, nsize)
572 /* No need to explicitly free anything. */
573 # define REGEX_FREE_STACK(arg) ((void)0)
574
575 # endif /* not REGEX_MALLOC */
576 #endif /* not using relocating allocator */
577
578
579 /* True if `size1' is non-NULL and PTR is pointing anywhere inside
580 `string1' or just past its end. This works if PTR is NULL, which is
581 a good thing. */
582 #define FIRST_STRING_P(ptr) \
583 (size1 && string1 <= (ptr) && (ptr) <= string1 + size1)
584
585 /* (Re)Allocate N items of type T using malloc, or fail. */
586 #define TALLOC(n, t) ((t *) malloc ((n) * sizeof (t)))
587 #define RETALLOC(addr, n, t) ((addr) = (t *) realloc (addr, (n) * sizeof (t)))
588 #define RETALLOC_IF(addr, n, t) \
589 if (addr) RETALLOC((addr), (n), t); else (addr) = TALLOC ((n), t)
590 #define REGEX_TALLOC(n, t) ((t *) REGEX_ALLOCATE ((n) * sizeof (t)))
591
592 #define BYTEWIDTH 8 /* In bits. */
593
594 #define STREQ(s1, s2) ((strcmp (s1, s2) == 0))
595
596 #undef MAX
597 #undef MIN
598 #define MAX(a, b) ((a) > (b) ? (a) : (b))
599 #define MIN(a, b) ((a) < (b) ? (a) : (b))
600
601 /* Type of source-pattern and string chars. */
602 typedef const unsigned char re_char;
603
604 typedef char boolean;
605 #define false 0
606 #define true 1
607
608 static int re_match_2_internal _RE_ARGS ((struct re_pattern_buffer *bufp,
609 re_char *string1, int size1,
610 re_char *string2, int size2,
611 int pos,
612 struct re_registers *regs,
613 int stop));
614 \f
615 /* These are the command codes that appear in compiled regular
616 expressions. Some opcodes are followed by argument bytes. A
617 command code can specify any interpretation whatsoever for its
618 arguments. Zero bytes may appear in the compiled regular expression. */
619
620 typedef enum
621 {
622 no_op = 0,
623
624 /* Succeed right away--no more backtracking. */
625 succeed,
626
627 /* Followed by one byte giving n, then by n literal bytes. */
628 exactn,
629
630 /* Matches any (more or less) character. */
631 anychar,
632
633 /* Matches any one char belonging to specified set. First
634 following byte is number of bitmap bytes. Then come bytes
635 for a bitmap saying which chars are in. Bits in each byte
636 are ordered low-bit-first. A character is in the set if its
637 bit is 1. A character too large to have a bit in the map is
638 automatically not in the set.
639
640 If the length byte has the 0x80 bit set, then that stuff
641 is followed by a range table:
642 2 bytes of flags for character sets (low 8 bits, high 8 bits)
643 See RANGE_TABLE_WORK_BITS below.
644 2 bytes, the number of pairs that follow (upto 32767)
645 pairs, each 2 multibyte characters,
646 each multibyte character represented as 3 bytes. */
647 charset,
648
649 /* Same parameters as charset, but match any character that is
650 not one of those specified. */
651 charset_not,
652
653 /* Start remembering the text that is matched, for storing in a
654 register. Followed by one byte with the register number, in
655 the range 0 to one less than the pattern buffer's re_nsub
656 field. */
657 start_memory,
658
659 /* Stop remembering the text that is matched and store it in a
660 memory register. Followed by one byte with the register
661 number, in the range 0 to one less than `re_nsub' in the
662 pattern buffer. */
663 stop_memory,
664
665 /* Match a duplicate of something remembered. Followed by one
666 byte containing the register number. */
667 duplicate,
668
669 /* Fail unless at beginning of line. */
670 begline,
671
672 /* Fail unless at end of line. */
673 endline,
674
675 /* Succeeds if at beginning of buffer (if emacs) or at beginning
676 of string to be matched (if not). */
677 begbuf,
678
679 /* Analogously, for end of buffer/string. */
680 endbuf,
681
682 /* Followed by two byte relative address to which to jump. */
683 jump,
684
685 /* Followed by two-byte relative address of place to resume at
686 in case of failure. */
687 on_failure_jump,
688
689 /* Like on_failure_jump, but pushes a placeholder instead of the
690 current string position when executed. */
691 on_failure_keep_string_jump,
692
693 /* Just like `on_failure_jump', except that it checks that we
694 don't get stuck in an infinite loop (matching an empty string
695 indefinitely). */
696 on_failure_jump_loop,
697
698 /* Just like `on_failure_jump_loop', except that it checks for
699 a different kind of loop (the kind that shows up with non-greedy
700 operators). This operation has to be immediately preceded
701 by a `no_op'. */
702 on_failure_jump_nastyloop,
703
704 /* A smart `on_failure_jump' used for greedy * and + operators.
705 It analyses the loop before which it is put and if the
706 loop does not require backtracking, it changes itself to
707 `on_failure_keep_string_jump' and short-circuits the loop,
708 else it just defaults to changing itself into `on_failure_jump'.
709 It assumes that it is pointing to just past a `jump'. */
710 on_failure_jump_smart,
711
712 /* Followed by two-byte relative address and two-byte number n.
713 After matching N times, jump to the address upon failure.
714 Does not work if N starts at 0: use on_failure_jump_loop
715 instead. */
716 succeed_n,
717
718 /* Followed by two-byte relative address, and two-byte number n.
719 Jump to the address N times, then fail. */
720 jump_n,
721
722 /* Set the following two-byte relative address to the
723 subsequent two-byte number. The address *includes* the two
724 bytes of number. */
725 set_number_at,
726
727 wordbeg, /* Succeeds if at word beginning. */
728 wordend, /* Succeeds if at word end. */
729
730 wordbound, /* Succeeds if at a word boundary. */
731 notwordbound, /* Succeeds if not at a word boundary. */
732
733 symbeg, /* Succeeds if at symbol beginning. */
734 symend, /* Succeeds if at symbol end. */
735
736 /* Matches any character whose syntax is specified. Followed by
737 a byte which contains a syntax code, e.g., Sword. */
738 syntaxspec,
739
740 /* Matches any character whose syntax is not that specified. */
741 notsyntaxspec
742
743 #ifdef emacs
744 ,before_dot, /* Succeeds if before point. */
745 at_dot, /* Succeeds if at point. */
746 after_dot, /* Succeeds if after point. */
747
748 /* Matches any character whose category-set contains the specified
749 category. The operator is followed by a byte which contains a
750 category code (mnemonic ASCII character). */
751 categoryspec,
752
753 /* Matches any character whose category-set does not contain the
754 specified category. The operator is followed by a byte which
755 contains the category code (mnemonic ASCII character). */
756 notcategoryspec
757 #endif /* emacs */
758 } re_opcode_t;
759 \f
760 /* Common operations on the compiled pattern. */
761
762 /* Store NUMBER in two contiguous bytes starting at DESTINATION. */
763
764 #define STORE_NUMBER(destination, number) \
765 do { \
766 (destination)[0] = (number) & 0377; \
767 (destination)[1] = (number) >> 8; \
768 } while (0)
769
770 /* Same as STORE_NUMBER, except increment DESTINATION to
771 the byte after where the number is stored. Therefore, DESTINATION
772 must be an lvalue. */
773
774 #define STORE_NUMBER_AND_INCR(destination, number) \
775 do { \
776 STORE_NUMBER (destination, number); \
777 (destination) += 2; \
778 } while (0)
779
780 /* Put into DESTINATION a number stored in two contiguous bytes starting
781 at SOURCE. */
782
783 #define EXTRACT_NUMBER(destination, source) \
784 do { \
785 (destination) = *(source) & 0377; \
786 (destination) += SIGN_EXTEND_CHAR (*((source) + 1)) << 8; \
787 } while (0)
788
789 #ifdef DEBUG
790 static void extract_number _RE_ARGS ((int *dest, re_char *source));
791 static void
792 extract_number (dest, source)
793 int *dest;
794 re_char *source;
795 {
796 int temp = SIGN_EXTEND_CHAR (*(source + 1));
797 *dest = *source & 0377;
798 *dest += temp << 8;
799 }
800
801 # ifndef EXTRACT_MACROS /* To debug the macros. */
802 # undef EXTRACT_NUMBER
803 # define EXTRACT_NUMBER(dest, src) extract_number (&dest, src)
804 # endif /* not EXTRACT_MACROS */
805
806 #endif /* DEBUG */
807
808 /* Same as EXTRACT_NUMBER, except increment SOURCE to after the number.
809 SOURCE must be an lvalue. */
810
811 #define EXTRACT_NUMBER_AND_INCR(destination, source) \
812 do { \
813 EXTRACT_NUMBER (destination, source); \
814 (source) += 2; \
815 } while (0)
816
817 #ifdef DEBUG
818 static void extract_number_and_incr _RE_ARGS ((int *destination,
819 re_char **source));
820 static void
821 extract_number_and_incr (destination, source)
822 int *destination;
823 re_char **source;
824 {
825 extract_number (destination, *source);
826 *source += 2;
827 }
828
829 # ifndef EXTRACT_MACROS
830 # undef EXTRACT_NUMBER_AND_INCR
831 # define EXTRACT_NUMBER_AND_INCR(dest, src) \
832 extract_number_and_incr (&dest, &src)
833 # endif /* not EXTRACT_MACROS */
834
835 #endif /* DEBUG */
836 \f
837 /* Store a multibyte character in three contiguous bytes starting
838 DESTINATION, and increment DESTINATION to the byte after where the
839 character is stored. Therefore, DESTINATION must be an lvalue. */
840
841 #define STORE_CHARACTER_AND_INCR(destination, character) \
842 do { \
843 (destination)[0] = (character) & 0377; \
844 (destination)[1] = ((character) >> 8) & 0377; \
845 (destination)[2] = (character) >> 16; \
846 (destination) += 3; \
847 } while (0)
848
849 /* Put into DESTINATION a character stored in three contiguous bytes
850 starting at SOURCE. */
851
852 #define EXTRACT_CHARACTER(destination, source) \
853 do { \
854 (destination) = ((source)[0] \
855 | ((source)[1] << 8) \
856 | ((source)[2] << 16)); \
857 } while (0)
858
859
860 /* Macros for charset. */
861
862 /* Size of bitmap of charset P in bytes. P is a start of charset,
863 i.e. *P is (re_opcode_t) charset or (re_opcode_t) charset_not. */
864 #define CHARSET_BITMAP_SIZE(p) ((p)[1] & 0x7F)
865
866 /* Nonzero if charset P has range table. */
867 #define CHARSET_RANGE_TABLE_EXISTS_P(p) ((p)[1] & 0x80)
868
869 /* Return the address of range table of charset P. But not the start
870 of table itself, but the before where the number of ranges is
871 stored. `2 +' means to skip re_opcode_t and size of bitmap,
872 and the 2 bytes of flags at the start of the range table. */
873 #define CHARSET_RANGE_TABLE(p) (&(p)[4 + CHARSET_BITMAP_SIZE (p)])
874
875 /* Extract the bit flags that start a range table. */
876 #define CHARSET_RANGE_TABLE_BITS(p) \
877 ((p)[2 + CHARSET_BITMAP_SIZE (p)] \
878 + (p)[3 + CHARSET_BITMAP_SIZE (p)] * 0x100)
879
880 /* Test if C is listed in the bitmap of charset P. */
881 #define CHARSET_LOOKUP_BITMAP(p, c) \
882 ((c) < CHARSET_BITMAP_SIZE (p) * BYTEWIDTH \
883 && (p)[2 + (c) / BYTEWIDTH] & (1 << ((c) % BYTEWIDTH)))
884
885 /* Return the address of end of RANGE_TABLE. COUNT is number of
886 ranges (which is a pair of (start, end)) in the RANGE_TABLE. `* 2'
887 is start of range and end of range. `* 3' is size of each start
888 and end. */
889 #define CHARSET_RANGE_TABLE_END(range_table, count) \
890 ((range_table) + (count) * 2 * 3)
891
892 /* Test if C is in RANGE_TABLE. A flag NOT is negated if C is in.
893 COUNT is number of ranges in RANGE_TABLE. */
894 #define CHARSET_LOOKUP_RANGE_TABLE_RAW(not, c, range_table, count) \
895 do \
896 { \
897 re_wchar_t range_start, range_end; \
898 re_char *p; \
899 re_char *range_table_end \
900 = CHARSET_RANGE_TABLE_END ((range_table), (count)); \
901 \
902 for (p = (range_table); p < range_table_end; p += 2 * 3) \
903 { \
904 EXTRACT_CHARACTER (range_start, p); \
905 EXTRACT_CHARACTER (range_end, p + 3); \
906 \
907 if (range_start <= (c) && (c) <= range_end) \
908 { \
909 (not) = !(not); \
910 break; \
911 } \
912 } \
913 } \
914 while (0)
915
916 /* Test if C is in range table of CHARSET. The flag NOT is negated if
917 C is listed in it. */
918 #define CHARSET_LOOKUP_RANGE_TABLE(not, c, charset) \
919 do \
920 { \
921 /* Number of ranges in range table. */ \
922 int count; \
923 re_char *range_table = CHARSET_RANGE_TABLE (charset); \
924 \
925 EXTRACT_NUMBER_AND_INCR (count, range_table); \
926 CHARSET_LOOKUP_RANGE_TABLE_RAW ((not), (c), range_table, count); \
927 } \
928 while (0)
929 \f
930 /* If DEBUG is defined, Regex prints many voluminous messages about what
931 it is doing (if the variable `debug' is nonzero). If linked with the
932 main program in `iregex.c', you can enter patterns and strings
933 interactively. And if linked with the main program in `main.c' and
934 the other test files, you can run the already-written tests. */
935
936 #ifdef DEBUG
937
938 /* We use standard I/O for debugging. */
939 # include <stdio.h>
940
941 /* It is useful to test things that ``must'' be true when debugging. */
942 # include <assert.h>
943
944 static int debug = -100000;
945
946 # define DEBUG_STATEMENT(e) e
947 # define DEBUG_PRINT1(x) if (debug > 0) printf (x)
948 # define DEBUG_PRINT2(x1, x2) if (debug > 0) printf (x1, x2)
949 # define DEBUG_PRINT3(x1, x2, x3) if (debug > 0) printf (x1, x2, x3)
950 # define DEBUG_PRINT4(x1, x2, x3, x4) if (debug > 0) printf (x1, x2, x3, x4)
951 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) \
952 if (debug > 0) print_partial_compiled_pattern (s, e)
953 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) \
954 if (debug > 0) print_double_string (w, s1, sz1, s2, sz2)
955
956
957 /* Print the fastmap in human-readable form. */
958
959 void
960 print_fastmap (fastmap)
961 char *fastmap;
962 {
963 unsigned was_a_range = 0;
964 unsigned i = 0;
965
966 while (i < (1 << BYTEWIDTH))
967 {
968 if (fastmap[i++])
969 {
970 was_a_range = 0;
971 putchar (i - 1);
972 while (i < (1 << BYTEWIDTH) && fastmap[i])
973 {
974 was_a_range = 1;
975 i++;
976 }
977 if (was_a_range)
978 {
979 printf ("-");
980 putchar (i - 1);
981 }
982 }
983 }
984 putchar ('\n');
985 }
986
987
988 /* Print a compiled pattern string in human-readable form, starting at
989 the START pointer into it and ending just before the pointer END. */
990
991 void
992 print_partial_compiled_pattern (start, end)
993 re_char *start;
994 re_char *end;
995 {
996 int mcnt, mcnt2;
997 re_char *p = start;
998 re_char *pend = end;
999
1000 if (start == NULL)
1001 {
1002 fprintf (stderr, "(null)\n");
1003 return;
1004 }
1005
1006 /* Loop over pattern commands. */
1007 while (p < pend)
1008 {
1009 fprintf (stderr, "%d:\t", p - start);
1010
1011 switch ((re_opcode_t) *p++)
1012 {
1013 case no_op:
1014 fprintf (stderr, "/no_op");
1015 break;
1016
1017 case succeed:
1018 fprintf (stderr, "/succeed");
1019 break;
1020
1021 case exactn:
1022 mcnt = *p++;
1023 fprintf (stderr, "/exactn/%d", mcnt);
1024 do
1025 {
1026 fprintf (stderr, "/%c", *p++);
1027 }
1028 while (--mcnt);
1029 break;
1030
1031 case start_memory:
1032 fprintf (stderr, "/start_memory/%d", *p++);
1033 break;
1034
1035 case stop_memory:
1036 fprintf (stderr, "/stop_memory/%d", *p++);
1037 break;
1038
1039 case duplicate:
1040 fprintf (stderr, "/duplicate/%d", *p++);
1041 break;
1042
1043 case anychar:
1044 fprintf (stderr, "/anychar");
1045 break;
1046
1047 case charset:
1048 case charset_not:
1049 {
1050 register int c, last = -100;
1051 register int in_range = 0;
1052 int length = CHARSET_BITMAP_SIZE (p - 1);
1053 int has_range_table = CHARSET_RANGE_TABLE_EXISTS_P (p - 1);
1054
1055 fprintf (stderr, "/charset [%s",
1056 (re_opcode_t) *(p - 1) == charset_not ? "^" : "");
1057
1058 if (p + *p >= pend)
1059 fprintf (stderr, " !extends past end of pattern! ");
1060
1061 for (c = 0; c < 256; c++)
1062 if (c / 8 < length
1063 && (p[1 + (c/8)] & (1 << (c % 8))))
1064 {
1065 /* Are we starting a range? */
1066 if (last + 1 == c && ! in_range)
1067 {
1068 fprintf (stderr, "-");
1069 in_range = 1;
1070 }
1071 /* Have we broken a range? */
1072 else if (last + 1 != c && in_range)
1073 {
1074 fprintf (stderr, "%c", last);
1075 in_range = 0;
1076 }
1077
1078 if (! in_range)
1079 fprintf (stderr, "%c", c);
1080
1081 last = c;
1082 }
1083
1084 if (in_range)
1085 fprintf (stderr, "%c", last);
1086
1087 fprintf (stderr, "]");
1088
1089 p += 1 + length;
1090
1091 if (has_range_table)
1092 {
1093 int count;
1094 fprintf (stderr, "has-range-table");
1095
1096 /* ??? Should print the range table; for now, just skip it. */
1097 p += 2; /* skip range table bits */
1098 EXTRACT_NUMBER_AND_INCR (count, p);
1099 p = CHARSET_RANGE_TABLE_END (p, count);
1100 }
1101 }
1102 break;
1103
1104 case begline:
1105 fprintf (stderr, "/begline");
1106 break;
1107
1108 case endline:
1109 fprintf (stderr, "/endline");
1110 break;
1111
1112 case on_failure_jump:
1113 extract_number_and_incr (&mcnt, &p);
1114 fprintf (stderr, "/on_failure_jump to %d", p + mcnt - start);
1115 break;
1116
1117 case on_failure_keep_string_jump:
1118 extract_number_and_incr (&mcnt, &p);
1119 fprintf (stderr, "/on_failure_keep_string_jump to %d", p + mcnt - start);
1120 break;
1121
1122 case on_failure_jump_nastyloop:
1123 extract_number_and_incr (&mcnt, &p);
1124 fprintf (stderr, "/on_failure_jump_nastyloop to %d", p + mcnt - start);
1125 break;
1126
1127 case on_failure_jump_loop:
1128 extract_number_and_incr (&mcnt, &p);
1129 fprintf (stderr, "/on_failure_jump_loop to %d", p + mcnt - start);
1130 break;
1131
1132 case on_failure_jump_smart:
1133 extract_number_and_incr (&mcnt, &p);
1134 fprintf (stderr, "/on_failure_jump_smart to %d", p + mcnt - start);
1135 break;
1136
1137 case jump:
1138 extract_number_and_incr (&mcnt, &p);
1139 fprintf (stderr, "/jump to %d", p + mcnt - start);
1140 break;
1141
1142 case succeed_n:
1143 extract_number_and_incr (&mcnt, &p);
1144 extract_number_and_incr (&mcnt2, &p);
1145 fprintf (stderr, "/succeed_n to %d, %d times", p - 2 + mcnt - start, mcnt2);
1146 break;
1147
1148 case jump_n:
1149 extract_number_and_incr (&mcnt, &p);
1150 extract_number_and_incr (&mcnt2, &p);
1151 fprintf (stderr, "/jump_n to %d, %d times", p - 2 + mcnt - start, mcnt2);
1152 break;
1153
1154 case set_number_at:
1155 extract_number_and_incr (&mcnt, &p);
1156 extract_number_and_incr (&mcnt2, &p);
1157 fprintf (stderr, "/set_number_at location %d to %d", p - 2 + mcnt - start, mcnt2);
1158 break;
1159
1160 case wordbound:
1161 fprintf (stderr, "/wordbound");
1162 break;
1163
1164 case notwordbound:
1165 fprintf (stderr, "/notwordbound");
1166 break;
1167
1168 case wordbeg:
1169 fprintf (stderr, "/wordbeg");
1170 break;
1171
1172 case wordend:
1173 fprintf (stderr, "/wordend");
1174 break;
1175
1176 case symbeg:
1177 fprintf (stderr, "/symbeg");
1178 break;
1179
1180 case symend:
1181 fprintf (stderr, "/symend");
1182 break;
1183
1184 case syntaxspec:
1185 fprintf (stderr, "/syntaxspec");
1186 mcnt = *p++;
1187 fprintf (stderr, "/%d", mcnt);
1188 break;
1189
1190 case notsyntaxspec:
1191 fprintf (stderr, "/notsyntaxspec");
1192 mcnt = *p++;
1193 fprintf (stderr, "/%d", mcnt);
1194 break;
1195
1196 # ifdef emacs
1197 case before_dot:
1198 fprintf (stderr, "/before_dot");
1199 break;
1200
1201 case at_dot:
1202 fprintf (stderr, "/at_dot");
1203 break;
1204
1205 case after_dot:
1206 fprintf (stderr, "/after_dot");
1207 break;
1208
1209 case categoryspec:
1210 fprintf (stderr, "/categoryspec");
1211 mcnt = *p++;
1212 fprintf (stderr, "/%d", mcnt);
1213 break;
1214
1215 case notcategoryspec:
1216 fprintf (stderr, "/notcategoryspec");
1217 mcnt = *p++;
1218 fprintf (stderr, "/%d", mcnt);
1219 break;
1220 # endif /* emacs */
1221
1222 case begbuf:
1223 fprintf (stderr, "/begbuf");
1224 break;
1225
1226 case endbuf:
1227 fprintf (stderr, "/endbuf");
1228 break;
1229
1230 default:
1231 fprintf (stderr, "?%d", *(p-1));
1232 }
1233
1234 fprintf (stderr, "\n");
1235 }
1236
1237 fprintf (stderr, "%d:\tend of pattern.\n", p - start);
1238 }
1239
1240
1241 void
1242 print_compiled_pattern (bufp)
1243 struct re_pattern_buffer *bufp;
1244 {
1245 re_char *buffer = bufp->buffer;
1246
1247 print_partial_compiled_pattern (buffer, buffer + bufp->used);
1248 printf ("%ld bytes used/%ld bytes allocated.\n",
1249 bufp->used, bufp->allocated);
1250
1251 if (bufp->fastmap_accurate && bufp->fastmap)
1252 {
1253 printf ("fastmap: ");
1254 print_fastmap (bufp->fastmap);
1255 }
1256
1257 printf ("re_nsub: %d\t", bufp->re_nsub);
1258 printf ("regs_alloc: %d\t", bufp->regs_allocated);
1259 printf ("can_be_null: %d\t", bufp->can_be_null);
1260 printf ("no_sub: %d\t", bufp->no_sub);
1261 printf ("not_bol: %d\t", bufp->not_bol);
1262 printf ("not_eol: %d\t", bufp->not_eol);
1263 printf ("syntax: %lx\n", bufp->syntax);
1264 fflush (stdout);
1265 /* Perhaps we should print the translate table? */
1266 }
1267
1268
1269 void
1270 print_double_string (where, string1, size1, string2, size2)
1271 re_char *where;
1272 re_char *string1;
1273 re_char *string2;
1274 int size1;
1275 int size2;
1276 {
1277 int this_char;
1278
1279 if (where == NULL)
1280 printf ("(null)");
1281 else
1282 {
1283 if (FIRST_STRING_P (where))
1284 {
1285 for (this_char = where - string1; this_char < size1; this_char++)
1286 putchar (string1[this_char]);
1287
1288 where = string2;
1289 }
1290
1291 for (this_char = where - string2; this_char < size2; this_char++)
1292 putchar (string2[this_char]);
1293 }
1294 }
1295
1296 #else /* not DEBUG */
1297
1298 # undef assert
1299 # define assert(e)
1300
1301 # define DEBUG_STATEMENT(e)
1302 # define DEBUG_PRINT1(x)
1303 # define DEBUG_PRINT2(x1, x2)
1304 # define DEBUG_PRINT3(x1, x2, x3)
1305 # define DEBUG_PRINT4(x1, x2, x3, x4)
1306 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e)
1307 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2)
1308
1309 #endif /* not DEBUG */
1310 \f
1311 /* Set by `re_set_syntax' to the current regexp syntax to recognize. Can
1312 also be assigned to arbitrarily: each pattern buffer stores its own
1313 syntax, so it can be changed between regex compilations. */
1314 /* This has no initializer because initialized variables in Emacs
1315 become read-only after dumping. */
1316 reg_syntax_t re_syntax_options;
1317
1318
1319 /* Specify the precise syntax of regexps for compilation. This provides
1320 for compatibility for various utilities which historically have
1321 different, incompatible syntaxes.
1322
1323 The argument SYNTAX is a bit mask comprised of the various bits
1324 defined in regex.h. We return the old syntax. */
1325
1326 reg_syntax_t
1327 re_set_syntax (syntax)
1328 reg_syntax_t syntax;
1329 {
1330 reg_syntax_t ret = re_syntax_options;
1331
1332 re_syntax_options = syntax;
1333 return ret;
1334 }
1335 WEAK_ALIAS (__re_set_syntax, re_set_syntax)
1336
1337 /* Regexp to use to replace spaces, or NULL meaning don't. */
1338 static re_char *whitespace_regexp;
1339
1340 void
1341 re_set_whitespace_regexp (regexp)
1342 const char *regexp;
1343 {
1344 whitespace_regexp = (re_char *) regexp;
1345 }
1346 WEAK_ALIAS (__re_set_syntax, re_set_syntax)
1347 \f
1348 /* This table gives an error message for each of the error codes listed
1349 in regex.h. Obviously the order here has to be same as there.
1350 POSIX doesn't require that we do anything for REG_NOERROR,
1351 but why not be nice? */
1352
1353 static const char *re_error_msgid[] =
1354 {
1355 gettext_noop ("Success"), /* REG_NOERROR */
1356 gettext_noop ("No match"), /* REG_NOMATCH */
1357 gettext_noop ("Invalid regular expression"), /* REG_BADPAT */
1358 gettext_noop ("Invalid collation character"), /* REG_ECOLLATE */
1359 gettext_noop ("Invalid character class name"), /* REG_ECTYPE */
1360 gettext_noop ("Trailing backslash"), /* REG_EESCAPE */
1361 gettext_noop ("Invalid back reference"), /* REG_ESUBREG */
1362 gettext_noop ("Unmatched [ or [^"), /* REG_EBRACK */
1363 gettext_noop ("Unmatched ( or \\("), /* REG_EPAREN */
1364 gettext_noop ("Unmatched \\{"), /* REG_EBRACE */
1365 gettext_noop ("Invalid content of \\{\\}"), /* REG_BADBR */
1366 gettext_noop ("Invalid range end"), /* REG_ERANGE */
1367 gettext_noop ("Memory exhausted"), /* REG_ESPACE */
1368 gettext_noop ("Invalid preceding regular expression"), /* REG_BADRPT */
1369 gettext_noop ("Premature end of regular expression"), /* REG_EEND */
1370 gettext_noop ("Regular expression too big"), /* REG_ESIZE */
1371 gettext_noop ("Unmatched ) or \\)"), /* REG_ERPAREN */
1372 gettext_noop ("Range striding over charsets") /* REG_ERANGEX */
1373 };
1374 \f
1375 /* Avoiding alloca during matching, to placate r_alloc. */
1376
1377 /* Define MATCH_MAY_ALLOCATE unless we need to make sure that the
1378 searching and matching functions should not call alloca. On some
1379 systems, alloca is implemented in terms of malloc, and if we're
1380 using the relocating allocator routines, then malloc could cause a
1381 relocation, which might (if the strings being searched are in the
1382 ralloc heap) shift the data out from underneath the regexp
1383 routines.
1384
1385 Here's another reason to avoid allocation: Emacs
1386 processes input from X in a signal handler; processing X input may
1387 call malloc; if input arrives while a matching routine is calling
1388 malloc, then we're scrod. But Emacs can't just block input while
1389 calling matching routines; then we don't notice interrupts when
1390 they come in. So, Emacs blocks input around all regexp calls
1391 except the matching calls, which it leaves unprotected, in the
1392 faith that they will not malloc. */
1393
1394 /* Normally, this is fine. */
1395 #define MATCH_MAY_ALLOCATE
1396
1397 /* The match routines may not allocate if (1) they would do it with malloc
1398 and (2) it's not safe for them to use malloc.
1399 Note that if REL_ALLOC is defined, matching would not use malloc for the
1400 failure stack, but we would still use it for the register vectors;
1401 so REL_ALLOC should not affect this. */
1402 #if defined REGEX_MALLOC && defined emacs
1403 # undef MATCH_MAY_ALLOCATE
1404 #endif
1405
1406 \f
1407 /* Failure stack declarations and macros; both re_compile_fastmap and
1408 re_match_2 use a failure stack. These have to be macros because of
1409 REGEX_ALLOCATE_STACK. */
1410
1411
1412 /* Approximate number of failure points for which to initially allocate space
1413 when matching. If this number is exceeded, we allocate more
1414 space, so it is not a hard limit. */
1415 #ifndef INIT_FAILURE_ALLOC
1416 # define INIT_FAILURE_ALLOC 20
1417 #endif
1418
1419 /* Roughly the maximum number of failure points on the stack. Would be
1420 exactly that if always used TYPICAL_FAILURE_SIZE items each time we failed.
1421 This is a variable only so users of regex can assign to it; we never
1422 change it ourselves. We always multiply it by TYPICAL_FAILURE_SIZE
1423 before using it, so it should probably be a byte-count instead. */
1424 # if defined MATCH_MAY_ALLOCATE
1425 /* Note that 4400 was enough to cause a crash on Alpha OSF/1,
1426 whose default stack limit is 2mb. In order for a larger
1427 value to work reliably, you have to try to make it accord
1428 with the process stack limit. */
1429 size_t re_max_failures = 40000;
1430 # else
1431 size_t re_max_failures = 4000;
1432 # endif
1433
1434 union fail_stack_elt
1435 {
1436 re_char *pointer;
1437 /* This should be the biggest `int' that's no bigger than a pointer. */
1438 long integer;
1439 };
1440
1441 typedef union fail_stack_elt fail_stack_elt_t;
1442
1443 typedef struct
1444 {
1445 fail_stack_elt_t *stack;
1446 size_t size;
1447 size_t avail; /* Offset of next open position. */
1448 size_t frame; /* Offset of the cur constructed frame. */
1449 } fail_stack_type;
1450
1451 #define FAIL_STACK_EMPTY() (fail_stack.frame == 0)
1452 #define FAIL_STACK_FULL() (fail_stack.avail == fail_stack.size)
1453
1454
1455 /* Define macros to initialize and free the failure stack.
1456 Do `return -2' if the alloc fails. */
1457
1458 #ifdef MATCH_MAY_ALLOCATE
1459 # define INIT_FAIL_STACK() \
1460 do { \
1461 fail_stack.stack = (fail_stack_elt_t *) \
1462 REGEX_ALLOCATE_STACK (INIT_FAILURE_ALLOC * TYPICAL_FAILURE_SIZE \
1463 * sizeof (fail_stack_elt_t)); \
1464 \
1465 if (fail_stack.stack == NULL) \
1466 return -2; \
1467 \
1468 fail_stack.size = INIT_FAILURE_ALLOC; \
1469 fail_stack.avail = 0; \
1470 fail_stack.frame = 0; \
1471 } while (0)
1472
1473 # define RESET_FAIL_STACK() REGEX_FREE_STACK (fail_stack.stack)
1474 #else
1475 # define INIT_FAIL_STACK() \
1476 do { \
1477 fail_stack.avail = 0; \
1478 fail_stack.frame = 0; \
1479 } while (0)
1480
1481 # define RESET_FAIL_STACK() ((void)0)
1482 #endif
1483
1484
1485 /* Double the size of FAIL_STACK, up to a limit
1486 which allows approximately `re_max_failures' items.
1487
1488 Return 1 if succeeds, and 0 if either ran out of memory
1489 allocating space for it or it was already too large.
1490
1491 REGEX_REALLOCATE_STACK requires `destination' be declared. */
1492
1493 /* Factor to increase the failure stack size by
1494 when we increase it.
1495 This used to be 2, but 2 was too wasteful
1496 because the old discarded stacks added up to as much space
1497 were as ultimate, maximum-size stack. */
1498 #define FAIL_STACK_GROWTH_FACTOR 4
1499
1500 #define GROW_FAIL_STACK(fail_stack) \
1501 (((fail_stack).size * sizeof (fail_stack_elt_t) \
1502 >= re_max_failures * TYPICAL_FAILURE_SIZE) \
1503 ? 0 \
1504 : ((fail_stack).stack \
1505 = (fail_stack_elt_t *) \
1506 REGEX_REALLOCATE_STACK ((fail_stack).stack, \
1507 (fail_stack).size * sizeof (fail_stack_elt_t), \
1508 MIN (re_max_failures * TYPICAL_FAILURE_SIZE, \
1509 ((fail_stack).size * sizeof (fail_stack_elt_t) \
1510 * FAIL_STACK_GROWTH_FACTOR))), \
1511 \
1512 (fail_stack).stack == NULL \
1513 ? 0 \
1514 : ((fail_stack).size \
1515 = (MIN (re_max_failures * TYPICAL_FAILURE_SIZE, \
1516 ((fail_stack).size * sizeof (fail_stack_elt_t) \
1517 * FAIL_STACK_GROWTH_FACTOR)) \
1518 / sizeof (fail_stack_elt_t)), \
1519 1)))
1520
1521
1522 /* Push a pointer value onto the failure stack.
1523 Assumes the variable `fail_stack'. Probably should only
1524 be called from within `PUSH_FAILURE_POINT'. */
1525 #define PUSH_FAILURE_POINTER(item) \
1526 fail_stack.stack[fail_stack.avail++].pointer = (item)
1527
1528 /* This pushes an integer-valued item onto the failure stack.
1529 Assumes the variable `fail_stack'. Probably should only
1530 be called from within `PUSH_FAILURE_POINT'. */
1531 #define PUSH_FAILURE_INT(item) \
1532 fail_stack.stack[fail_stack.avail++].integer = (item)
1533
1534 /* Push a fail_stack_elt_t value onto the failure stack.
1535 Assumes the variable `fail_stack'. Probably should only
1536 be called from within `PUSH_FAILURE_POINT'. */
1537 #define PUSH_FAILURE_ELT(item) \
1538 fail_stack.stack[fail_stack.avail++] = (item)
1539
1540 /* These three POP... operations complement the three PUSH... operations.
1541 All assume that `fail_stack' is nonempty. */
1542 #define POP_FAILURE_POINTER() fail_stack.stack[--fail_stack.avail].pointer
1543 #define POP_FAILURE_INT() fail_stack.stack[--fail_stack.avail].integer
1544 #define POP_FAILURE_ELT() fail_stack.stack[--fail_stack.avail]
1545
1546 /* Individual items aside from the registers. */
1547 #define NUM_NONREG_ITEMS 3
1548
1549 /* Used to examine the stack (to detect infinite loops). */
1550 #define FAILURE_PAT(h) fail_stack.stack[(h) - 1].pointer
1551 #define FAILURE_STR(h) (fail_stack.stack[(h) - 2].pointer)
1552 #define NEXT_FAILURE_HANDLE(h) fail_stack.stack[(h) - 3].integer
1553 #define TOP_FAILURE_HANDLE() fail_stack.frame
1554
1555
1556 #define ENSURE_FAIL_STACK(space) \
1557 while (REMAINING_AVAIL_SLOTS <= space) { \
1558 if (!GROW_FAIL_STACK (fail_stack)) \
1559 return -2; \
1560 DEBUG_PRINT2 ("\n Doubled stack; size now: %d\n", (fail_stack).size);\
1561 DEBUG_PRINT2 (" slots available: %d\n", REMAINING_AVAIL_SLOTS);\
1562 }
1563
1564 /* Push register NUM onto the stack. */
1565 #define PUSH_FAILURE_REG(num) \
1566 do { \
1567 char *destination; \
1568 ENSURE_FAIL_STACK(3); \
1569 DEBUG_PRINT4 (" Push reg %d (spanning %p -> %p)\n", \
1570 num, regstart[num], regend[num]); \
1571 PUSH_FAILURE_POINTER (regstart[num]); \
1572 PUSH_FAILURE_POINTER (regend[num]); \
1573 PUSH_FAILURE_INT (num); \
1574 } while (0)
1575
1576 /* Change the counter's value to VAL, but make sure that it will
1577 be reset when backtracking. */
1578 #define PUSH_NUMBER(ptr,val) \
1579 do { \
1580 char *destination; \
1581 int c; \
1582 ENSURE_FAIL_STACK(3); \
1583 EXTRACT_NUMBER (c, ptr); \
1584 DEBUG_PRINT4 (" Push number %p = %d -> %d\n", ptr, c, val); \
1585 PUSH_FAILURE_INT (c); \
1586 PUSH_FAILURE_POINTER (ptr); \
1587 PUSH_FAILURE_INT (-1); \
1588 STORE_NUMBER (ptr, val); \
1589 } while (0)
1590
1591 /* Pop a saved register off the stack. */
1592 #define POP_FAILURE_REG_OR_COUNT() \
1593 do { \
1594 int reg = POP_FAILURE_INT (); \
1595 if (reg == -1) \
1596 { \
1597 /* It's a counter. */ \
1598 /* Here, we discard `const', making re_match non-reentrant. */ \
1599 unsigned char *ptr = (unsigned char*) POP_FAILURE_POINTER (); \
1600 reg = POP_FAILURE_INT (); \
1601 STORE_NUMBER (ptr, reg); \
1602 DEBUG_PRINT3 (" Pop counter %p = %d\n", ptr, reg); \
1603 } \
1604 else \
1605 { \
1606 regend[reg] = POP_FAILURE_POINTER (); \
1607 regstart[reg] = POP_FAILURE_POINTER (); \
1608 DEBUG_PRINT4 (" Pop reg %d (spanning %p -> %p)\n", \
1609 reg, regstart[reg], regend[reg]); \
1610 } \
1611 } while (0)
1612
1613 /* Check that we are not stuck in an infinite loop. */
1614 #define CHECK_INFINITE_LOOP(pat_cur, string_place) \
1615 do { \
1616 int failure = TOP_FAILURE_HANDLE (); \
1617 /* Check for infinite matching loops */ \
1618 while (failure > 0 \
1619 && (FAILURE_STR (failure) == string_place \
1620 || FAILURE_STR (failure) == NULL)) \
1621 { \
1622 assert (FAILURE_PAT (failure) >= bufp->buffer \
1623 && FAILURE_PAT (failure) <= bufp->buffer + bufp->used); \
1624 if (FAILURE_PAT (failure) == pat_cur) \
1625 { \
1626 cycle = 1; \
1627 break; \
1628 } \
1629 DEBUG_PRINT2 (" Other pattern: %p\n", FAILURE_PAT (failure)); \
1630 failure = NEXT_FAILURE_HANDLE(failure); \
1631 } \
1632 DEBUG_PRINT2 (" Other string: %p\n", FAILURE_STR (failure)); \
1633 } while (0)
1634
1635 /* Push the information about the state we will need
1636 if we ever fail back to it.
1637
1638 Requires variables fail_stack, regstart, regend and
1639 num_regs be declared. GROW_FAIL_STACK requires `destination' be
1640 declared.
1641
1642 Does `return FAILURE_CODE' if runs out of memory. */
1643
1644 #define PUSH_FAILURE_POINT(pattern, string_place) \
1645 do { \
1646 char *destination; \
1647 /* Must be int, so when we don't save any registers, the arithmetic \
1648 of 0 + -1 isn't done as unsigned. */ \
1649 \
1650 DEBUG_STATEMENT (nfailure_points_pushed++); \
1651 DEBUG_PRINT1 ("\nPUSH_FAILURE_POINT:\n"); \
1652 DEBUG_PRINT2 (" Before push, next avail: %d\n", (fail_stack).avail); \
1653 DEBUG_PRINT2 (" size: %d\n", (fail_stack).size);\
1654 \
1655 ENSURE_FAIL_STACK (NUM_NONREG_ITEMS); \
1656 \
1657 DEBUG_PRINT1 ("\n"); \
1658 \
1659 DEBUG_PRINT2 (" Push frame index: %d\n", fail_stack.frame); \
1660 PUSH_FAILURE_INT (fail_stack.frame); \
1661 \
1662 DEBUG_PRINT2 (" Push string %p: `", string_place); \
1663 DEBUG_PRINT_DOUBLE_STRING (string_place, string1, size1, string2, size2);\
1664 DEBUG_PRINT1 ("'\n"); \
1665 PUSH_FAILURE_POINTER (string_place); \
1666 \
1667 DEBUG_PRINT2 (" Push pattern %p: ", pattern); \
1668 DEBUG_PRINT_COMPILED_PATTERN (bufp, pattern, pend); \
1669 PUSH_FAILURE_POINTER (pattern); \
1670 \
1671 /* Close the frame by moving the frame pointer past it. */ \
1672 fail_stack.frame = fail_stack.avail; \
1673 } while (0)
1674
1675 /* Estimate the size of data pushed by a typical failure stack entry.
1676 An estimate is all we need, because all we use this for
1677 is to choose a limit for how big to make the failure stack. */
1678 /* BEWARE, the value `20' is hard-coded in emacs.c:main(). */
1679 #define TYPICAL_FAILURE_SIZE 20
1680
1681 /* How many items can still be added to the stack without overflowing it. */
1682 #define REMAINING_AVAIL_SLOTS ((fail_stack).size - (fail_stack).avail)
1683
1684
1685 /* Pops what PUSH_FAIL_STACK pushes.
1686
1687 We restore into the parameters, all of which should be lvalues:
1688 STR -- the saved data position.
1689 PAT -- the saved pattern position.
1690 REGSTART, REGEND -- arrays of string positions.
1691
1692 Also assumes the variables `fail_stack' and (if debugging), `bufp',
1693 `pend', `string1', `size1', `string2', and `size2'. */
1694
1695 #define POP_FAILURE_POINT(str, pat) \
1696 do { \
1697 assert (!FAIL_STACK_EMPTY ()); \
1698 \
1699 /* Remove failure points and point to how many regs pushed. */ \
1700 DEBUG_PRINT1 ("POP_FAILURE_POINT:\n"); \
1701 DEBUG_PRINT2 (" Before pop, next avail: %d\n", fail_stack.avail); \
1702 DEBUG_PRINT2 (" size: %d\n", fail_stack.size); \
1703 \
1704 /* Pop the saved registers. */ \
1705 while (fail_stack.frame < fail_stack.avail) \
1706 POP_FAILURE_REG_OR_COUNT (); \
1707 \
1708 pat = POP_FAILURE_POINTER (); \
1709 DEBUG_PRINT2 (" Popping pattern %p: ", pat); \
1710 DEBUG_PRINT_COMPILED_PATTERN (bufp, pat, pend); \
1711 \
1712 /* If the saved string location is NULL, it came from an \
1713 on_failure_keep_string_jump opcode, and we want to throw away the \
1714 saved NULL, thus retaining our current position in the string. */ \
1715 str = POP_FAILURE_POINTER (); \
1716 DEBUG_PRINT2 (" Popping string %p: `", str); \
1717 DEBUG_PRINT_DOUBLE_STRING (str, string1, size1, string2, size2); \
1718 DEBUG_PRINT1 ("'\n"); \
1719 \
1720 fail_stack.frame = POP_FAILURE_INT (); \
1721 DEBUG_PRINT2 (" Popping frame index: %d\n", fail_stack.frame); \
1722 \
1723 assert (fail_stack.avail >= 0); \
1724 assert (fail_stack.frame <= fail_stack.avail); \
1725 \
1726 DEBUG_STATEMENT (nfailure_points_popped++); \
1727 } while (0) /* POP_FAILURE_POINT */
1728
1729
1730 \f
1731 /* Registers are set to a sentinel when they haven't yet matched. */
1732 #define REG_UNSET(e) ((e) == NULL)
1733 \f
1734 /* Subroutine declarations and macros for regex_compile. */
1735
1736 static reg_errcode_t regex_compile _RE_ARGS ((re_char *pattern, size_t size,
1737 reg_syntax_t syntax,
1738 struct re_pattern_buffer *bufp));
1739 static void store_op1 _RE_ARGS ((re_opcode_t op, unsigned char *loc, int arg));
1740 static void store_op2 _RE_ARGS ((re_opcode_t op, unsigned char *loc,
1741 int arg1, int arg2));
1742 static void insert_op1 _RE_ARGS ((re_opcode_t op, unsigned char *loc,
1743 int arg, unsigned char *end));
1744 static void insert_op2 _RE_ARGS ((re_opcode_t op, unsigned char *loc,
1745 int arg1, int arg2, unsigned char *end));
1746 static boolean at_begline_loc_p _RE_ARGS ((re_char *pattern,
1747 re_char *p,
1748 reg_syntax_t syntax));
1749 static boolean at_endline_loc_p _RE_ARGS ((re_char *p,
1750 re_char *pend,
1751 reg_syntax_t syntax));
1752 static re_char *skip_one_char _RE_ARGS ((re_char *p));
1753 static int analyse_first _RE_ARGS ((re_char *p, re_char *pend,
1754 char *fastmap, const int multibyte));
1755
1756 /* Fetch the next character in the uncompiled pattern, with no
1757 translation. */
1758 #define PATFETCH(c) \
1759 do { \
1760 int len; \
1761 if (p == pend) return REG_EEND; \
1762 c = RE_STRING_CHAR_AND_LENGTH (p, pend - p, len, multibyte); \
1763 p += len; \
1764 } while (0)
1765
1766
1767 /* If `translate' is non-null, return translate[D], else just D. We
1768 cast the subscript to translate because some data is declared as
1769 `char *', to avoid warnings when a string constant is passed. But
1770 when we use a character as a subscript we must make it unsigned. */
1771 #ifndef TRANSLATE
1772 # define TRANSLATE(d) \
1773 (RE_TRANSLATE_P (translate) ? RE_TRANSLATE (translate, (d)) : (d))
1774 #endif
1775
1776
1777 /* Macros for outputting the compiled pattern into `buffer'. */
1778
1779 /* If the buffer isn't allocated when it comes in, use this. */
1780 #define INIT_BUF_SIZE 32
1781
1782 /* Make sure we have at least N more bytes of space in buffer. */
1783 #define GET_BUFFER_SPACE(n) \
1784 while ((size_t) (b - bufp->buffer + (n)) > bufp->allocated) \
1785 EXTEND_BUFFER ()
1786
1787 /* Make sure we have one more byte of buffer space and then add C to it. */
1788 #define BUF_PUSH(c) \
1789 do { \
1790 GET_BUFFER_SPACE (1); \
1791 *b++ = (unsigned char) (c); \
1792 } while (0)
1793
1794
1795 /* Ensure we have two more bytes of buffer space and then append C1 and C2. */
1796 #define BUF_PUSH_2(c1, c2) \
1797 do { \
1798 GET_BUFFER_SPACE (2); \
1799 *b++ = (unsigned char) (c1); \
1800 *b++ = (unsigned char) (c2); \
1801 } while (0)
1802
1803
1804 /* As with BUF_PUSH_2, except for three bytes. */
1805 #define BUF_PUSH_3(c1, c2, c3) \
1806 do { \
1807 GET_BUFFER_SPACE (3); \
1808 *b++ = (unsigned char) (c1); \
1809 *b++ = (unsigned char) (c2); \
1810 *b++ = (unsigned char) (c3); \
1811 } while (0)
1812
1813
1814 /* Store a jump with opcode OP at LOC to location TO. We store a
1815 relative address offset by the three bytes the jump itself occupies. */
1816 #define STORE_JUMP(op, loc, to) \
1817 store_op1 (op, loc, (to) - (loc) - 3)
1818
1819 /* Likewise, for a two-argument jump. */
1820 #define STORE_JUMP2(op, loc, to, arg) \
1821 store_op2 (op, loc, (to) - (loc) - 3, arg)
1822
1823 /* Like `STORE_JUMP', but for inserting. Assume `b' is the buffer end. */
1824 #define INSERT_JUMP(op, loc, to) \
1825 insert_op1 (op, loc, (to) - (loc) - 3, b)
1826
1827 /* Like `STORE_JUMP2', but for inserting. Assume `b' is the buffer end. */
1828 #define INSERT_JUMP2(op, loc, to, arg) \
1829 insert_op2 (op, loc, (to) - (loc) - 3, arg, b)
1830
1831
1832 /* This is not an arbitrary limit: the arguments which represent offsets
1833 into the pattern are two bytes long. So if 2^15 bytes turns out to
1834 be too small, many things would have to change. */
1835 # define MAX_BUF_SIZE (1L << 15)
1836
1837 #if 0 /* This is when we thought it could be 2^16 bytes. */
1838 /* Any other compiler which, like MSC, has allocation limit below 2^16
1839 bytes will have to use approach similar to what was done below for
1840 MSC and drop MAX_BUF_SIZE a bit. Otherwise you may end up
1841 reallocating to 0 bytes. Such thing is not going to work too well.
1842 You have been warned!! */
1843 #if defined _MSC_VER && !defined WIN32
1844 /* Microsoft C 16-bit versions limit malloc to approx 65512 bytes. */
1845 # define MAX_BUF_SIZE 65500L
1846 #else
1847 # define MAX_BUF_SIZE (1L << 16)
1848 #endif
1849 #endif /* 0 */
1850
1851 /* Extend the buffer by twice its current size via realloc and
1852 reset the pointers that pointed into the old block to point to the
1853 correct places in the new one. If extending the buffer results in it
1854 being larger than MAX_BUF_SIZE, then flag memory exhausted. */
1855 #if __BOUNDED_POINTERS__
1856 # define SET_HIGH_BOUND(P) (__ptrhigh (P) = __ptrlow (P) + bufp->allocated)
1857 # define MOVE_BUFFER_POINTER(P) \
1858 (__ptrlow (P) = new_buffer + (__ptrlow (P) - old_buffer), \
1859 SET_HIGH_BOUND (P), \
1860 __ptrvalue (P) = new_buffer + (__ptrvalue (P) - old_buffer))
1861 # define ELSE_EXTEND_BUFFER_HIGH_BOUND \
1862 else \
1863 { \
1864 SET_HIGH_BOUND (b); \
1865 SET_HIGH_BOUND (begalt); \
1866 if (fixup_alt_jump) \
1867 SET_HIGH_BOUND (fixup_alt_jump); \
1868 if (laststart) \
1869 SET_HIGH_BOUND (laststart); \
1870 if (pending_exact) \
1871 SET_HIGH_BOUND (pending_exact); \
1872 }
1873 #else
1874 # define MOVE_BUFFER_POINTER(P) ((P) = new_buffer + ((P) - old_buffer))
1875 # define ELSE_EXTEND_BUFFER_HIGH_BOUND
1876 #endif
1877 #define EXTEND_BUFFER() \
1878 do { \
1879 unsigned char *old_buffer = bufp->buffer; \
1880 if (bufp->allocated == MAX_BUF_SIZE) \
1881 return REG_ESIZE; \
1882 bufp->allocated <<= 1; \
1883 if (bufp->allocated > MAX_BUF_SIZE) \
1884 bufp->allocated = MAX_BUF_SIZE; \
1885 RETALLOC (bufp->buffer, bufp->allocated, unsigned char); \
1886 if (bufp->buffer == NULL) \
1887 return REG_ESPACE; \
1888 /* If the buffer moved, move all the pointers into it. */ \
1889 if (old_buffer != bufp->buffer) \
1890 { \
1891 unsigned char *new_buffer = bufp->buffer; \
1892 MOVE_BUFFER_POINTER (b); \
1893 MOVE_BUFFER_POINTER (begalt); \
1894 if (fixup_alt_jump) \
1895 MOVE_BUFFER_POINTER (fixup_alt_jump); \
1896 if (laststart) \
1897 MOVE_BUFFER_POINTER (laststart); \
1898 if (pending_exact) \
1899 MOVE_BUFFER_POINTER (pending_exact); \
1900 } \
1901 ELSE_EXTEND_BUFFER_HIGH_BOUND \
1902 } while (0)
1903
1904
1905 /* Since we have one byte reserved for the register number argument to
1906 {start,stop}_memory, the maximum number of groups we can report
1907 things about is what fits in that byte. */
1908 #define MAX_REGNUM 255
1909
1910 /* But patterns can have more than `MAX_REGNUM' registers. We just
1911 ignore the excess. */
1912 typedef int regnum_t;
1913
1914
1915 /* Macros for the compile stack. */
1916
1917 /* Since offsets can go either forwards or backwards, this type needs to
1918 be able to hold values from -(MAX_BUF_SIZE - 1) to MAX_BUF_SIZE - 1. */
1919 /* int may be not enough when sizeof(int) == 2. */
1920 typedef long pattern_offset_t;
1921
1922 typedef struct
1923 {
1924 pattern_offset_t begalt_offset;
1925 pattern_offset_t fixup_alt_jump;
1926 pattern_offset_t laststart_offset;
1927 regnum_t regnum;
1928 } compile_stack_elt_t;
1929
1930
1931 typedef struct
1932 {
1933 compile_stack_elt_t *stack;
1934 unsigned size;
1935 unsigned avail; /* Offset of next open position. */
1936 } compile_stack_type;
1937
1938
1939 #define INIT_COMPILE_STACK_SIZE 32
1940
1941 #define COMPILE_STACK_EMPTY (compile_stack.avail == 0)
1942 #define COMPILE_STACK_FULL (compile_stack.avail == compile_stack.size)
1943
1944 /* The next available element. */
1945 #define COMPILE_STACK_TOP (compile_stack.stack[compile_stack.avail])
1946
1947 /* Explicit quit checking is only used on NTemacs and whenever we
1948 use polling to process input events. */
1949 #if defined emacs && (defined WINDOWSNT || defined SYNC_INPUT) && defined QUIT
1950 extern int immediate_quit;
1951 # define IMMEDIATE_QUIT_CHECK \
1952 do { \
1953 if (immediate_quit) QUIT; \
1954 } while (0)
1955 #else
1956 # define IMMEDIATE_QUIT_CHECK ((void)0)
1957 #endif
1958 \f
1959 /* Structure to manage work area for range table. */
1960 struct range_table_work_area
1961 {
1962 int *table; /* actual work area. */
1963 int allocated; /* allocated size for work area in bytes. */
1964 int used; /* actually used size in words. */
1965 int bits; /* flag to record character classes */
1966 };
1967
1968 /* Make sure that WORK_AREA can hold more N multibyte characters.
1969 This is used only in set_image_of_range and set_image_of_range_1.
1970 It expects WORK_AREA to be a pointer.
1971 If it can't get the space, it returns from the surrounding function. */
1972
1973 #define EXTEND_RANGE_TABLE(work_area, n) \
1974 do { \
1975 if (((work_area).used + (n)) * sizeof (int) > (work_area).allocated) \
1976 { \
1977 extend_range_table_work_area (&work_area); \
1978 if ((work_area).table == 0) \
1979 return (REG_ESPACE); \
1980 } \
1981 } while (0)
1982
1983 #define SET_RANGE_TABLE_WORK_AREA_BIT(work_area, bit) \
1984 (work_area).bits |= (bit)
1985
1986 /* Bits used to implement the multibyte-part of the various character classes
1987 such as [:alnum:] in a charset's range table. */
1988 #define BIT_WORD 0x1
1989 #define BIT_LOWER 0x2
1990 #define BIT_PUNCT 0x4
1991 #define BIT_SPACE 0x8
1992 #define BIT_UPPER 0x10
1993 #define BIT_MULTIBYTE 0x20
1994
1995 /* Set a range (RANGE_START, RANGE_END) to WORK_AREA. */
1996 #define SET_RANGE_TABLE_WORK_AREA(work_area, range_start, range_end) \
1997 do { \
1998 EXTEND_RANGE_TABLE ((work_area), 2); \
1999 (work_area).table[(work_area).used++] = (range_start); \
2000 (work_area).table[(work_area).used++] = (range_end); \
2001 } while (0)
2002
2003 /* Free allocated memory for WORK_AREA. */
2004 #define FREE_RANGE_TABLE_WORK_AREA(work_area) \
2005 do { \
2006 if ((work_area).table) \
2007 free ((work_area).table); \
2008 } while (0)
2009
2010 #define CLEAR_RANGE_TABLE_WORK_USED(work_area) ((work_area).used = 0, (work_area).bits = 0)
2011 #define RANGE_TABLE_WORK_USED(work_area) ((work_area).used)
2012 #define RANGE_TABLE_WORK_BITS(work_area) ((work_area).bits)
2013 #define RANGE_TABLE_WORK_ELT(work_area, i) ((work_area).table[i])
2014 \f
2015
2016 /* Set the bit for character C in a list. */
2017 #define SET_LIST_BIT(c) (b[((c)) / BYTEWIDTH] |= 1 << ((c) % BYTEWIDTH))
2018
2019
2020 #ifdef emacs
2021
2022 /* Store characters in the range FROM to TO in the bitmap at B (for
2023 ASCII and unibyte characters) and WORK_AREA (for multibyte
2024 characters) while translating them and paying attention to the
2025 continuity of translated characters.
2026
2027 Implementation note: It is better to implement these fairly big
2028 macros by a function, but it's not that easy because macros called
2029 in this macro assume various local variables already declared. */
2030
2031 /* Both FROM and TO are ASCII characters. */
2032
2033 #define SETUP_ASCII_RANGE(work_area, FROM, TO) \
2034 do { \
2035 int C0, C1; \
2036 \
2037 for (C0 = (FROM); C0 <= (TO); C0++) \
2038 { \
2039 C1 = TRANSLATE (C0); \
2040 if (! ASCII_CHAR_P (C1)) \
2041 { \
2042 SET_RANGE_TABLE_WORK_AREA ((work_area), C1, C1); \
2043 if ((C1 = RE_CHAR_TO_UNIBYTE (C1)) < 0) \
2044 C1 = C0; \
2045 } \
2046 SET_LIST_BIT (C1); \
2047 } \
2048 } while (0)
2049
2050
2051 /* Both FROM and TO are unibyte characters (0x80..0xFF). */
2052
2053 #define SETUP_UNIBYTE_RANGE(work_area, FROM, TO) \
2054 do { \
2055 int C0, C1, C2, I; \
2056 int USED = RANGE_TABLE_WORK_USED (work_area); \
2057 \
2058 for (C0 = (FROM); C0 <= (TO); C0++) \
2059 { \
2060 C1 = RE_CHAR_TO_MULTIBYTE (C0); \
2061 if (CHAR_BYTE8_P (C1)) \
2062 SET_LIST_BIT (C0); \
2063 else \
2064 { \
2065 C2 = TRANSLATE (C1); \
2066 if (C2 == C1 \
2067 || (C1 = RE_CHAR_TO_UNIBYTE (C2)) < 0) \
2068 C1 = C0; \
2069 SET_LIST_BIT (C1); \
2070 for (I = RANGE_TABLE_WORK_USED (work_area) - 2; I >= USED; I -= 2) \
2071 { \
2072 int from = RANGE_TABLE_WORK_ELT (work_area, I); \
2073 int to = RANGE_TABLE_WORK_ELT (work_area, I + 1); \
2074 \
2075 if (C2 >= from - 1 && C2 <= to + 1) \
2076 { \
2077 if (C2 == from - 1) \
2078 RANGE_TABLE_WORK_ELT (work_area, I)--; \
2079 else if (C2 == to + 1) \
2080 RANGE_TABLE_WORK_ELT (work_area, I + 1)++; \
2081 break; \
2082 } \
2083 } \
2084 if (I < USED) \
2085 SET_RANGE_TABLE_WORK_AREA ((work_area), C2, C2); \
2086 } \
2087 } \
2088 } while (0)
2089
2090
2091 /* Both FROM and TO are mulitbyte characters. */
2092
2093 #define SETUP_MULTIBYTE_RANGE(work_area, FROM, TO) \
2094 do { \
2095 int C0, C1, C2, I, USED = RANGE_TABLE_WORK_USED (work_area); \
2096 \
2097 SET_RANGE_TABLE_WORK_AREA ((work_area), (FROM), (TO)); \
2098 for (C0 = (FROM); C0 <= (TO); C0++) \
2099 { \
2100 C1 = TRANSLATE (C0); \
2101 if ((C2 = RE_CHAR_TO_UNIBYTE (C1)) >= 0 \
2102 || (C1 != C0 && (C2 = RE_CHAR_TO_UNIBYTE (C0)) >= 0)) \
2103 SET_LIST_BIT (C2); \
2104 if (C1 >= (FROM) && C1 <= (TO)) \
2105 continue; \
2106 for (I = RANGE_TABLE_WORK_USED (work_area) - 2; I >= USED; I -= 2) \
2107 { \
2108 int from = RANGE_TABLE_WORK_ELT (work_area, I); \
2109 int to = RANGE_TABLE_WORK_ELT (work_area, I + 1); \
2110 \
2111 if (C1 >= from - 1 && C1 <= to + 1) \
2112 { \
2113 if (C1 == from - 1) \
2114 RANGE_TABLE_WORK_ELT (work_area, I)--; \
2115 else if (C1 == to + 1) \
2116 RANGE_TABLE_WORK_ELT (work_area, I + 1)++; \
2117 break; \
2118 } \
2119 } \
2120 if (I < USED) \
2121 SET_RANGE_TABLE_WORK_AREA ((work_area), C1, C1); \
2122 } \
2123 } while (0)
2124
2125 #endif /* emacs */
2126
2127 /* Get the next unsigned number in the uncompiled pattern. */
2128 #define GET_UNSIGNED_NUMBER(num) \
2129 do { \
2130 if (p == pend) \
2131 FREE_STACK_RETURN (REG_EBRACE); \
2132 else \
2133 { \
2134 PATFETCH (c); \
2135 while ('0' <= c && c <= '9') \
2136 { \
2137 int prev; \
2138 if (num < 0) \
2139 num = 0; \
2140 prev = num; \
2141 num = num * 10 + c - '0'; \
2142 if (num / 10 != prev) \
2143 FREE_STACK_RETURN (REG_BADBR); \
2144 if (p == pend) \
2145 FREE_STACK_RETURN (REG_EBRACE); \
2146 PATFETCH (c); \
2147 } \
2148 } \
2149 } while (0)
2150 \f
2151 #if ! WIDE_CHAR_SUPPORT
2152
2153 /* Map a string to the char class it names (if any). */
2154 re_wctype_t
2155 re_wctype (str)
2156 re_char *str;
2157 {
2158 const char *string = str;
2159 if (STREQ (string, "alnum")) return RECC_ALNUM;
2160 else if (STREQ (string, "alpha")) return RECC_ALPHA;
2161 else if (STREQ (string, "word")) return RECC_WORD;
2162 else if (STREQ (string, "ascii")) return RECC_ASCII;
2163 else if (STREQ (string, "nonascii")) return RECC_NONASCII;
2164 else if (STREQ (string, "graph")) return RECC_GRAPH;
2165 else if (STREQ (string, "lower")) return RECC_LOWER;
2166 else if (STREQ (string, "print")) return RECC_PRINT;
2167 else if (STREQ (string, "punct")) return RECC_PUNCT;
2168 else if (STREQ (string, "space")) return RECC_SPACE;
2169 else if (STREQ (string, "upper")) return RECC_UPPER;
2170 else if (STREQ (string, "unibyte")) return RECC_UNIBYTE;
2171 else if (STREQ (string, "multibyte")) return RECC_MULTIBYTE;
2172 else if (STREQ (string, "digit")) return RECC_DIGIT;
2173 else if (STREQ (string, "xdigit")) return RECC_XDIGIT;
2174 else if (STREQ (string, "cntrl")) return RECC_CNTRL;
2175 else if (STREQ (string, "blank")) return RECC_BLANK;
2176 else return 0;
2177 }
2178
2179 /* True if CH is in the char class CC. */
2180 boolean
2181 re_iswctype (ch, cc)
2182 int ch;
2183 re_wctype_t cc;
2184 {
2185 switch (cc)
2186 {
2187 case RECC_ALNUM: return ISALNUM (ch);
2188 case RECC_ALPHA: return ISALPHA (ch);
2189 case RECC_BLANK: return ISBLANK (ch);
2190 case RECC_CNTRL: return ISCNTRL (ch);
2191 case RECC_DIGIT: return ISDIGIT (ch);
2192 case RECC_GRAPH: return ISGRAPH (ch);
2193 case RECC_LOWER: return ISLOWER (ch);
2194 case RECC_PRINT: return ISPRINT (ch);
2195 case RECC_PUNCT: return ISPUNCT (ch);
2196 case RECC_SPACE: return ISSPACE (ch);
2197 case RECC_UPPER: return ISUPPER (ch);
2198 case RECC_XDIGIT: return ISXDIGIT (ch);
2199 case RECC_ASCII: return IS_REAL_ASCII (ch);
2200 case RECC_NONASCII: return !IS_REAL_ASCII (ch);
2201 case RECC_UNIBYTE: return ISUNIBYTE (ch);
2202 case RECC_MULTIBYTE: return !ISUNIBYTE (ch);
2203 case RECC_WORD: return ISWORD (ch);
2204 case RECC_ERROR: return false;
2205 default:
2206 abort();
2207 }
2208 }
2209
2210 /* Return a bit-pattern to use in the range-table bits to match multibyte
2211 chars of class CC. */
2212 static int
2213 re_wctype_to_bit (cc)
2214 re_wctype_t cc;
2215 {
2216 switch (cc)
2217 {
2218 case RECC_NONASCII: case RECC_PRINT: case RECC_GRAPH:
2219 case RECC_MULTIBYTE: return BIT_MULTIBYTE;
2220 case RECC_ALPHA: case RECC_ALNUM: case RECC_WORD: return BIT_WORD;
2221 case RECC_LOWER: return BIT_LOWER;
2222 case RECC_UPPER: return BIT_UPPER;
2223 case RECC_PUNCT: return BIT_PUNCT;
2224 case RECC_SPACE: return BIT_SPACE;
2225 case RECC_ASCII: case RECC_DIGIT: case RECC_XDIGIT: case RECC_CNTRL:
2226 case RECC_BLANK: case RECC_UNIBYTE: case RECC_ERROR: return 0;
2227 default:
2228 abort();
2229 }
2230 }
2231 #endif
2232 \f
2233 /* Filling in the work area of a range. */
2234
2235 /* Actually extend the space in WORK_AREA. */
2236
2237 static void
2238 extend_range_table_work_area (work_area)
2239 struct range_table_work_area *work_area;
2240 {
2241 work_area->allocated += 16 * sizeof (int);
2242 if (work_area->table)
2243 work_area->table
2244 = (int *) realloc (work_area->table, work_area->allocated);
2245 else
2246 work_area->table
2247 = (int *) malloc (work_area->allocated);
2248 }
2249
2250 #if 0
2251 #ifdef emacs
2252
2253 /* Carefully find the ranges of codes that are equivalent
2254 under case conversion to the range start..end when passed through
2255 TRANSLATE. Handle the case where non-letters can come in between
2256 two upper-case letters (which happens in Latin-1).
2257 Also handle the case of groups of more than 2 case-equivalent chars.
2258
2259 The basic method is to look at consecutive characters and see
2260 if they can form a run that can be handled as one.
2261
2262 Returns -1 if successful, REG_ESPACE if ran out of space. */
2263
2264 static int
2265 set_image_of_range_1 (work_area, start, end, translate)
2266 RE_TRANSLATE_TYPE translate;
2267 struct range_table_work_area *work_area;
2268 re_wchar_t start, end;
2269 {
2270 /* `one_case' indicates a character, or a run of characters,
2271 each of which is an isolate (no case-equivalents).
2272 This includes all ASCII non-letters.
2273
2274 `two_case' indicates a character, or a run of characters,
2275 each of which has two case-equivalent forms.
2276 This includes all ASCII letters.
2277
2278 `strange' indicates a character that has more than one
2279 case-equivalent. */
2280
2281 enum case_type {one_case, two_case, strange};
2282
2283 /* Describe the run that is in progress,
2284 which the next character can try to extend.
2285 If run_type is strange, that means there really is no run.
2286 If run_type is one_case, then run_start...run_end is the run.
2287 If run_type is two_case, then the run is run_start...run_end,
2288 and the case-equivalents end at run_eqv_end. */
2289
2290 enum case_type run_type = strange;
2291 int run_start, run_end, run_eqv_end;
2292
2293 Lisp_Object eqv_table;
2294
2295 if (!RE_TRANSLATE_P (translate))
2296 {
2297 EXTEND_RANGE_TABLE (work_area, 2);
2298 work_area->table[work_area->used++] = (start);
2299 work_area->table[work_area->used++] = (end);
2300 return -1;
2301 }
2302
2303 eqv_table = XCHAR_TABLE (translate)->extras[2];
2304
2305 for (; start <= end; start++)
2306 {
2307 enum case_type this_type;
2308 int eqv = RE_TRANSLATE (eqv_table, start);
2309 int minchar, maxchar;
2310
2311 /* Classify this character */
2312 if (eqv == start)
2313 this_type = one_case;
2314 else if (RE_TRANSLATE (eqv_table, eqv) == start)
2315 this_type = two_case;
2316 else
2317 this_type = strange;
2318
2319 if (start < eqv)
2320 minchar = start, maxchar = eqv;
2321 else
2322 minchar = eqv, maxchar = start;
2323
2324 /* Can this character extend the run in progress? */
2325 if (this_type == strange || this_type != run_type
2326 || !(minchar == run_end + 1
2327 && (run_type == two_case
2328 ? maxchar == run_eqv_end + 1 : 1)))
2329 {
2330 /* No, end the run.
2331 Record each of its equivalent ranges. */
2332 if (run_type == one_case)
2333 {
2334 EXTEND_RANGE_TABLE (work_area, 2);
2335 work_area->table[work_area->used++] = run_start;
2336 work_area->table[work_area->used++] = run_end;
2337 }
2338 else if (run_type == two_case)
2339 {
2340 EXTEND_RANGE_TABLE (work_area, 4);
2341 work_area->table[work_area->used++] = run_start;
2342 work_area->table[work_area->used++] = run_end;
2343 work_area->table[work_area->used++]
2344 = RE_TRANSLATE (eqv_table, run_start);
2345 work_area->table[work_area->used++]
2346 = RE_TRANSLATE (eqv_table, run_end);
2347 }
2348 run_type = strange;
2349 }
2350
2351 if (this_type == strange)
2352 {
2353 /* For a strange character, add each of its equivalents, one
2354 by one. Don't start a range. */
2355 do
2356 {
2357 EXTEND_RANGE_TABLE (work_area, 2);
2358 work_area->table[work_area->used++] = eqv;
2359 work_area->table[work_area->used++] = eqv;
2360 eqv = RE_TRANSLATE (eqv_table, eqv);
2361 }
2362 while (eqv != start);
2363 }
2364
2365 /* Add this char to the run, or start a new run. */
2366 else if (run_type == strange)
2367 {
2368 /* Initialize a new range. */
2369 run_type = this_type;
2370 run_start = start;
2371 run_end = start;
2372 run_eqv_end = RE_TRANSLATE (eqv_table, run_end);
2373 }
2374 else
2375 {
2376 /* Extend a running range. */
2377 run_end = minchar;
2378 run_eqv_end = RE_TRANSLATE (eqv_table, run_end);
2379 }
2380 }
2381
2382 /* If a run is still in progress at the end, finish it now
2383 by recording its equivalent ranges. */
2384 if (run_type == one_case)
2385 {
2386 EXTEND_RANGE_TABLE (work_area, 2);
2387 work_area->table[work_area->used++] = run_start;
2388 work_area->table[work_area->used++] = run_end;
2389 }
2390 else if (run_type == two_case)
2391 {
2392 EXTEND_RANGE_TABLE (work_area, 4);
2393 work_area->table[work_area->used++] = run_start;
2394 work_area->table[work_area->used++] = run_end;
2395 work_area->table[work_area->used++]
2396 = RE_TRANSLATE (eqv_table, run_start);
2397 work_area->table[work_area->used++]
2398 = RE_TRANSLATE (eqv_table, run_end);
2399 }
2400
2401 return -1;
2402 }
2403
2404 #endif /* emacs */
2405
2406 /* Record the the image of the range start..end when passed through
2407 TRANSLATE. This is not necessarily TRANSLATE(start)..TRANSLATE(end)
2408 and is not even necessarily contiguous.
2409 Normally we approximate it with the smallest contiguous range that contains
2410 all the chars we need. However, for Latin-1 we go to extra effort
2411 to do a better job.
2412
2413 This function is not called for ASCII ranges.
2414
2415 Returns -1 if successful, REG_ESPACE if ran out of space. */
2416
2417 static int
2418 set_image_of_range (work_area, start, end, translate)
2419 RE_TRANSLATE_TYPE translate;
2420 struct range_table_work_area *work_area;
2421 re_wchar_t start, end;
2422 {
2423 re_wchar_t cmin, cmax;
2424
2425 #ifdef emacs
2426 /* For Latin-1 ranges, use set_image_of_range_1
2427 to get proper handling of ranges that include letters and nonletters.
2428 For a range that includes the whole of Latin-1, this is not necessary.
2429 For other character sets, we don't bother to get this right. */
2430 if (RE_TRANSLATE_P (translate) && start < 04400
2431 && !(start < 04200 && end >= 04377))
2432 {
2433 int newend;
2434 int tem;
2435 newend = end;
2436 if (newend > 04377)
2437 newend = 04377;
2438 tem = set_image_of_range_1 (work_area, start, newend, translate);
2439 if (tem > 0)
2440 return tem;
2441
2442 start = 04400;
2443 if (end < 04400)
2444 return -1;
2445 }
2446 #endif
2447
2448 EXTEND_RANGE_TABLE (work_area, 2);
2449 work_area->table[work_area->used++] = (start);
2450 work_area->table[work_area->used++] = (end);
2451
2452 cmin = -1, cmax = -1;
2453
2454 if (RE_TRANSLATE_P (translate))
2455 {
2456 int ch;
2457
2458 for (ch = start; ch <= end; ch++)
2459 {
2460 re_wchar_t c = TRANSLATE (ch);
2461 if (! (start <= c && c <= end))
2462 {
2463 if (cmin == -1)
2464 cmin = c, cmax = c;
2465 else
2466 {
2467 cmin = MIN (cmin, c);
2468 cmax = MAX (cmax, c);
2469 }
2470 }
2471 }
2472
2473 if (cmin != -1)
2474 {
2475 EXTEND_RANGE_TABLE (work_area, 2);
2476 work_area->table[work_area->used++] = (cmin);
2477 work_area->table[work_area->used++] = (cmax);
2478 }
2479 }
2480
2481 return -1;
2482 }
2483 #endif /* 0 */
2484 \f
2485 #ifndef MATCH_MAY_ALLOCATE
2486
2487 /* If we cannot allocate large objects within re_match_2_internal,
2488 we make the fail stack and register vectors global.
2489 The fail stack, we grow to the maximum size when a regexp
2490 is compiled.
2491 The register vectors, we adjust in size each time we
2492 compile a regexp, according to the number of registers it needs. */
2493
2494 static fail_stack_type fail_stack;
2495
2496 /* Size with which the following vectors are currently allocated.
2497 That is so we can make them bigger as needed,
2498 but never make them smaller. */
2499 static int regs_allocated_size;
2500
2501 static re_char ** regstart, ** regend;
2502 static re_char **best_regstart, **best_regend;
2503
2504 /* Make the register vectors big enough for NUM_REGS registers,
2505 but don't make them smaller. */
2506
2507 static
2508 regex_grow_registers (num_regs)
2509 int num_regs;
2510 {
2511 if (num_regs > regs_allocated_size)
2512 {
2513 RETALLOC_IF (regstart, num_regs, re_char *);
2514 RETALLOC_IF (regend, num_regs, re_char *);
2515 RETALLOC_IF (best_regstart, num_regs, re_char *);
2516 RETALLOC_IF (best_regend, num_regs, re_char *);
2517
2518 regs_allocated_size = num_regs;
2519 }
2520 }
2521
2522 #endif /* not MATCH_MAY_ALLOCATE */
2523 \f
2524 static boolean group_in_compile_stack _RE_ARGS ((compile_stack_type
2525 compile_stack,
2526 regnum_t regnum));
2527
2528 /* `regex_compile' compiles PATTERN (of length SIZE) according to SYNTAX.
2529 Returns one of error codes defined in `regex.h', or zero for success.
2530
2531 Assumes the `allocated' (and perhaps `buffer') and `translate'
2532 fields are set in BUFP on entry.
2533
2534 If it succeeds, results are put in BUFP (if it returns an error, the
2535 contents of BUFP are undefined):
2536 `buffer' is the compiled pattern;
2537 `syntax' is set to SYNTAX;
2538 `used' is set to the length of the compiled pattern;
2539 `fastmap_accurate' is zero;
2540 `re_nsub' is the number of subexpressions in PATTERN;
2541 `not_bol' and `not_eol' are zero;
2542
2543 The `fastmap' field is neither examined nor set. */
2544
2545 /* Insert the `jump' from the end of last alternative to "here".
2546 The space for the jump has already been allocated. */
2547 #define FIXUP_ALT_JUMP() \
2548 do { \
2549 if (fixup_alt_jump) \
2550 STORE_JUMP (jump, fixup_alt_jump, b); \
2551 } while (0)
2552
2553
2554 /* Return, freeing storage we allocated. */
2555 #define FREE_STACK_RETURN(value) \
2556 do { \
2557 FREE_RANGE_TABLE_WORK_AREA (range_table_work); \
2558 free (compile_stack.stack); \
2559 return value; \
2560 } while (0)
2561
2562 static reg_errcode_t
2563 regex_compile (pattern, size, syntax, bufp)
2564 re_char *pattern;
2565 size_t size;
2566 reg_syntax_t syntax;
2567 struct re_pattern_buffer *bufp;
2568 {
2569 /* We fetch characters from PATTERN here. */
2570 register re_wchar_t c, c1;
2571
2572 /* A random temporary spot in PATTERN. */
2573 re_char *p1;
2574
2575 /* Points to the end of the buffer, where we should append. */
2576 register unsigned char *b;
2577
2578 /* Keeps track of unclosed groups. */
2579 compile_stack_type compile_stack;
2580
2581 /* Points to the current (ending) position in the pattern. */
2582 #ifdef AIX
2583 /* `const' makes AIX compiler fail. */
2584 unsigned char *p = pattern;
2585 #else
2586 re_char *p = pattern;
2587 #endif
2588 re_char *pend = pattern + size;
2589
2590 /* How to translate the characters in the pattern. */
2591 RE_TRANSLATE_TYPE translate = bufp->translate;
2592
2593 /* Address of the count-byte of the most recently inserted `exactn'
2594 command. This makes it possible to tell if a new exact-match
2595 character can be added to that command or if the character requires
2596 a new `exactn' command. */
2597 unsigned char *pending_exact = 0;
2598
2599 /* Address of start of the most recently finished expression.
2600 This tells, e.g., postfix * where to find the start of its
2601 operand. Reset at the beginning of groups and alternatives. */
2602 unsigned char *laststart = 0;
2603
2604 /* Address of beginning of regexp, or inside of last group. */
2605 unsigned char *begalt;
2606
2607 /* Place in the uncompiled pattern (i.e., the {) to
2608 which to go back if the interval is invalid. */
2609 re_char *beg_interval;
2610
2611 /* Address of the place where a forward jump should go to the end of
2612 the containing expression. Each alternative of an `or' -- except the
2613 last -- ends with a forward jump of this sort. */
2614 unsigned char *fixup_alt_jump = 0;
2615
2616 /* Work area for range table of charset. */
2617 struct range_table_work_area range_table_work;
2618
2619 /* If the object matched can contain multibyte characters. */
2620 const boolean multibyte = RE_MULTIBYTE_P (bufp);
2621
2622 /* If a target of matching can contain multibyte characters. */
2623 const boolean target_multibyte = RE_TARGET_MULTIBYTE_P (bufp);
2624
2625 /* Nonzero if we have pushed down into a subpattern. */
2626 int in_subpattern = 0;
2627
2628 /* These hold the values of p, pattern, and pend from the main
2629 pattern when we have pushed into a subpattern. */
2630 re_char *main_p;
2631 re_char *main_pattern;
2632 re_char *main_pend;
2633
2634 #ifdef DEBUG
2635 debug++;
2636 DEBUG_PRINT1 ("\nCompiling pattern: ");
2637 if (debug > 0)
2638 {
2639 unsigned debug_count;
2640
2641 for (debug_count = 0; debug_count < size; debug_count++)
2642 putchar (pattern[debug_count]);
2643 putchar ('\n');
2644 }
2645 #endif /* DEBUG */
2646
2647 /* Initialize the compile stack. */
2648 compile_stack.stack = TALLOC (INIT_COMPILE_STACK_SIZE, compile_stack_elt_t);
2649 if (compile_stack.stack == NULL)
2650 return REG_ESPACE;
2651
2652 compile_stack.size = INIT_COMPILE_STACK_SIZE;
2653 compile_stack.avail = 0;
2654
2655 range_table_work.table = 0;
2656 range_table_work.allocated = 0;
2657
2658 /* Initialize the pattern buffer. */
2659 bufp->syntax = syntax;
2660 bufp->fastmap_accurate = 0;
2661 bufp->not_bol = bufp->not_eol = 0;
2662 bufp->used_syntax = 0;
2663
2664 /* Set `used' to zero, so that if we return an error, the pattern
2665 printer (for debugging) will think there's no pattern. We reset it
2666 at the end. */
2667 bufp->used = 0;
2668
2669 /* Always count groups, whether or not bufp->no_sub is set. */
2670 bufp->re_nsub = 0;
2671
2672 #if !defined emacs && !defined SYNTAX_TABLE
2673 /* Initialize the syntax table. */
2674 init_syntax_once ();
2675 #endif
2676
2677 if (bufp->allocated == 0)
2678 {
2679 if (bufp->buffer)
2680 { /* If zero allocated, but buffer is non-null, try to realloc
2681 enough space. This loses if buffer's address is bogus, but
2682 that is the user's responsibility. */
2683 RETALLOC (bufp->buffer, INIT_BUF_SIZE, unsigned char);
2684 }
2685 else
2686 { /* Caller did not allocate a buffer. Do it for them. */
2687 bufp->buffer = TALLOC (INIT_BUF_SIZE, unsigned char);
2688 }
2689 if (!bufp->buffer) FREE_STACK_RETURN (REG_ESPACE);
2690
2691 bufp->allocated = INIT_BUF_SIZE;
2692 }
2693
2694 begalt = b = bufp->buffer;
2695
2696 /* Loop through the uncompiled pattern until we're at the end. */
2697 while (1)
2698 {
2699 if (p == pend)
2700 {
2701 /* If this is the end of an included regexp,
2702 pop back to the main regexp and try again. */
2703 if (in_subpattern)
2704 {
2705 in_subpattern = 0;
2706 pattern = main_pattern;
2707 p = main_p;
2708 pend = main_pend;
2709 continue;
2710 }
2711 /* If this is the end of the main regexp, we are done. */
2712 break;
2713 }
2714
2715 PATFETCH (c);
2716
2717 switch (c)
2718 {
2719 case ' ':
2720 {
2721 re_char *p1 = p;
2722
2723 /* If there's no special whitespace regexp, treat
2724 spaces normally. And don't try to do this recursively. */
2725 if (!whitespace_regexp || in_subpattern)
2726 goto normal_char;
2727
2728 /* Peek past following spaces. */
2729 while (p1 != pend)
2730 {
2731 if (*p1 != ' ')
2732 break;
2733 p1++;
2734 }
2735 /* If the spaces are followed by a repetition op,
2736 treat them normally. */
2737 if (p1 != pend
2738 && (*p1 == '*' || *p1 == '+' || *p1 == '?'
2739 || (*p1 == '\\' && p1 + 1 != pend && p1[1] == '{')))
2740 goto normal_char;
2741
2742 /* Replace the spaces with the whitespace regexp. */
2743 in_subpattern = 1;
2744 main_p = p1;
2745 main_pend = pend;
2746 main_pattern = pattern;
2747 p = pattern = whitespace_regexp;
2748 pend = p + strlen (p);
2749 break;
2750 }
2751
2752 case '^':
2753 {
2754 if ( /* If at start of pattern, it's an operator. */
2755 p == pattern + 1
2756 /* If context independent, it's an operator. */
2757 || syntax & RE_CONTEXT_INDEP_ANCHORS
2758 /* Otherwise, depends on what's come before. */
2759 || at_begline_loc_p (pattern, p, syntax))
2760 BUF_PUSH ((syntax & RE_NO_NEWLINE_ANCHOR) ? begbuf : begline);
2761 else
2762 goto normal_char;
2763 }
2764 break;
2765
2766
2767 case '$':
2768 {
2769 if ( /* If at end of pattern, it's an operator. */
2770 p == pend
2771 /* If context independent, it's an operator. */
2772 || syntax & RE_CONTEXT_INDEP_ANCHORS
2773 /* Otherwise, depends on what's next. */
2774 || at_endline_loc_p (p, pend, syntax))
2775 BUF_PUSH ((syntax & RE_NO_NEWLINE_ANCHOR) ? endbuf : endline);
2776 else
2777 goto normal_char;
2778 }
2779 break;
2780
2781
2782 case '+':
2783 case '?':
2784 if ((syntax & RE_BK_PLUS_QM)
2785 || (syntax & RE_LIMITED_OPS))
2786 goto normal_char;
2787 handle_plus:
2788 case '*':
2789 /* If there is no previous pattern... */
2790 if (!laststart)
2791 {
2792 if (syntax & RE_CONTEXT_INVALID_OPS)
2793 FREE_STACK_RETURN (REG_BADRPT);
2794 else if (!(syntax & RE_CONTEXT_INDEP_OPS))
2795 goto normal_char;
2796 }
2797
2798 {
2799 /* 1 means zero (many) matches is allowed. */
2800 boolean zero_times_ok = 0, many_times_ok = 0;
2801 boolean greedy = 1;
2802
2803 /* If there is a sequence of repetition chars, collapse it
2804 down to just one (the right one). We can't combine
2805 interval operators with these because of, e.g., `a{2}*',
2806 which should only match an even number of `a's. */
2807
2808 for (;;)
2809 {
2810 if ((syntax & RE_FRUGAL)
2811 && c == '?' && (zero_times_ok || many_times_ok))
2812 greedy = 0;
2813 else
2814 {
2815 zero_times_ok |= c != '+';
2816 many_times_ok |= c != '?';
2817 }
2818
2819 if (p == pend)
2820 break;
2821 else if (*p == '*'
2822 || (!(syntax & RE_BK_PLUS_QM)
2823 && (*p == '+' || *p == '?')))
2824 ;
2825 else if (syntax & RE_BK_PLUS_QM && *p == '\\')
2826 {
2827 if (p+1 == pend)
2828 FREE_STACK_RETURN (REG_EESCAPE);
2829 if (p[1] == '+' || p[1] == '?')
2830 PATFETCH (c); /* Gobble up the backslash. */
2831 else
2832 break;
2833 }
2834 else
2835 break;
2836 /* If we get here, we found another repeat character. */
2837 PATFETCH (c);
2838 }
2839
2840 /* Star, etc. applied to an empty pattern is equivalent
2841 to an empty pattern. */
2842 if (!laststart || laststart == b)
2843 break;
2844
2845 /* Now we know whether or not zero matches is allowed
2846 and also whether or not two or more matches is allowed. */
2847 if (greedy)
2848 {
2849 if (many_times_ok)
2850 {
2851 boolean simple = skip_one_char (laststart) == b;
2852 unsigned int startoffset = 0;
2853 re_opcode_t ofj =
2854 /* Check if the loop can match the empty string. */
2855 (simple || !analyse_first (laststart, b, NULL, 0))
2856 ? on_failure_jump : on_failure_jump_loop;
2857 assert (skip_one_char (laststart) <= b);
2858
2859 if (!zero_times_ok && simple)
2860 { /* Since simple * loops can be made faster by using
2861 on_failure_keep_string_jump, we turn simple P+
2862 into PP* if P is simple. */
2863 unsigned char *p1, *p2;
2864 startoffset = b - laststart;
2865 GET_BUFFER_SPACE (startoffset);
2866 p1 = b; p2 = laststart;
2867 while (p2 < p1)
2868 *b++ = *p2++;
2869 zero_times_ok = 1;
2870 }
2871
2872 GET_BUFFER_SPACE (6);
2873 if (!zero_times_ok)
2874 /* A + loop. */
2875 STORE_JUMP (ofj, b, b + 6);
2876 else
2877 /* Simple * loops can use on_failure_keep_string_jump
2878 depending on what follows. But since we don't know
2879 that yet, we leave the decision up to
2880 on_failure_jump_smart. */
2881 INSERT_JUMP (simple ? on_failure_jump_smart : ofj,
2882 laststart + startoffset, b + 6);
2883 b += 3;
2884 STORE_JUMP (jump, b, laststart + startoffset);
2885 b += 3;
2886 }
2887 else
2888 {
2889 /* A simple ? pattern. */
2890 assert (zero_times_ok);
2891 GET_BUFFER_SPACE (3);
2892 INSERT_JUMP (on_failure_jump, laststart, b + 3);
2893 b += 3;
2894 }
2895 }
2896 else /* not greedy */
2897 { /* I wish the greedy and non-greedy cases could be merged. */
2898
2899 GET_BUFFER_SPACE (7); /* We might use less. */
2900 if (many_times_ok)
2901 {
2902 boolean emptyp = analyse_first (laststart, b, NULL, 0);
2903
2904 /* The non-greedy multiple match looks like
2905 a repeat..until: we only need a conditional jump
2906 at the end of the loop. */
2907 if (emptyp) BUF_PUSH (no_op);
2908 STORE_JUMP (emptyp ? on_failure_jump_nastyloop
2909 : on_failure_jump, b, laststart);
2910 b += 3;
2911 if (zero_times_ok)
2912 {
2913 /* The repeat...until naturally matches one or more.
2914 To also match zero times, we need to first jump to
2915 the end of the loop (its conditional jump). */
2916 INSERT_JUMP (jump, laststart, b);
2917 b += 3;
2918 }
2919 }
2920 else
2921 {
2922 /* non-greedy a?? */
2923 INSERT_JUMP (jump, laststart, b + 3);
2924 b += 3;
2925 INSERT_JUMP (on_failure_jump, laststart, laststart + 6);
2926 b += 3;
2927 }
2928 }
2929 }
2930 pending_exact = 0;
2931 break;
2932
2933
2934 case '.':
2935 laststart = b;
2936 BUF_PUSH (anychar);
2937 break;
2938
2939
2940 case '[':
2941 {
2942 CLEAR_RANGE_TABLE_WORK_USED (range_table_work);
2943
2944 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2945
2946 /* Ensure that we have enough space to push a charset: the
2947 opcode, the length count, and the bitset; 34 bytes in all. */
2948 GET_BUFFER_SPACE (34);
2949
2950 laststart = b;
2951
2952 /* We test `*p == '^' twice, instead of using an if
2953 statement, so we only need one BUF_PUSH. */
2954 BUF_PUSH (*p == '^' ? charset_not : charset);
2955 if (*p == '^')
2956 p++;
2957
2958 /* Remember the first position in the bracket expression. */
2959 p1 = p;
2960
2961 /* Push the number of bytes in the bitmap. */
2962 BUF_PUSH ((1 << BYTEWIDTH) / BYTEWIDTH);
2963
2964 /* Clear the whole map. */
2965 bzero (b, (1 << BYTEWIDTH) / BYTEWIDTH);
2966
2967 /* charset_not matches newline according to a syntax bit. */
2968 if ((re_opcode_t) b[-2] == charset_not
2969 && (syntax & RE_HAT_LISTS_NOT_NEWLINE))
2970 SET_LIST_BIT ('\n');
2971
2972 /* Read in characters and ranges, setting map bits. */
2973 for (;;)
2974 {
2975 boolean escaped_char = false;
2976 const unsigned char *p2 = p;
2977 re_wchar_t ch, c2;
2978
2979 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2980
2981 /* Don't translate yet. The range TRANSLATE(X..Y) cannot
2982 always be determined from TRANSLATE(X) and TRANSLATE(Y)
2983 So the translation is done later in a loop. Example:
2984 (let ((case-fold-search t)) (string-match "[A-_]" "A")) */
2985 PATFETCH (c);
2986
2987 /* \ might escape characters inside [...] and [^...]. */
2988 if ((syntax & RE_BACKSLASH_ESCAPE_IN_LISTS) && c == '\\')
2989 {
2990 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
2991
2992 PATFETCH (c);
2993 escaped_char = true;
2994 }
2995 else
2996 {
2997 /* Could be the end of the bracket expression. If it's
2998 not (i.e., when the bracket expression is `[]' so
2999 far), the ']' character bit gets set way below. */
3000 if (c == ']' && p2 != p1)
3001 break;
3002 }
3003
3004 /* See if we're at the beginning of a possible character
3005 class. */
3006
3007 if (!escaped_char &&
3008 syntax & RE_CHAR_CLASSES && c == '[' && *p == ':')
3009 {
3010 /* Leave room for the null. */
3011 unsigned char str[CHAR_CLASS_MAX_LENGTH + 1];
3012 const unsigned char *class_beg;
3013
3014 PATFETCH (c);
3015 c1 = 0;
3016 class_beg = p;
3017
3018 /* If pattern is `[[:'. */
3019 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
3020
3021 for (;;)
3022 {
3023 PATFETCH (c);
3024 if ((c == ':' && *p == ']') || p == pend)
3025 break;
3026 if (c1 < CHAR_CLASS_MAX_LENGTH)
3027 str[c1++] = c;
3028 else
3029 /* This is in any case an invalid class name. */
3030 str[0] = '\0';
3031 }
3032 str[c1] = '\0';
3033
3034 /* If isn't a word bracketed by `[:' and `:]':
3035 undo the ending character, the letters, and
3036 leave the leading `:' and `[' (but set bits for
3037 them). */
3038 if (c == ':' && *p == ']')
3039 {
3040 re_wctype_t cc;
3041 int limit;
3042
3043 cc = re_wctype (str);
3044
3045 if (cc == 0)
3046 FREE_STACK_RETURN (REG_ECTYPE);
3047
3048 /* Throw away the ] at the end of the character
3049 class. */
3050 PATFETCH (c);
3051
3052 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
3053
3054 #ifndef emacs
3055 for (ch = 0; ch < (1 << BYTEWIDTH); ++ch)
3056 if (re_iswctype (btowc (ch), cc))
3057 {
3058 c = TRANSLATE (ch);
3059 if (c < (1 << BYTEWIDTH))
3060 SET_LIST_BIT (c);
3061 }
3062 #else /* emacs */
3063 /* Most character classes in a multibyte match
3064 just set a flag. Exceptions are is_blank,
3065 is_digit, is_cntrl, and is_xdigit, since
3066 they can only match ASCII characters. We
3067 don't need to handle them for multibyte.
3068 They are distinguished by a negative wctype. */
3069
3070 for (ch = 0; ch < 256; ++ch)
3071 {
3072 c = RE_CHAR_TO_MULTIBYTE (ch);
3073 if (! CHAR_BYTE8_P (c)
3074 && re_iswctype (c, cc))
3075 {
3076 SET_LIST_BIT (ch);
3077 c1 = TRANSLATE (c);
3078 if (c1 == c)
3079 continue;
3080 if (ASCII_CHAR_P (c1))
3081 SET_LIST_BIT (c1);
3082 else if ((c1 = RE_CHAR_TO_UNIBYTE (c1)) >= 0)
3083 SET_LIST_BIT (c1);
3084 }
3085 }
3086 SET_RANGE_TABLE_WORK_AREA_BIT
3087 (range_table_work, re_wctype_to_bit (cc));
3088 #endif /* emacs */
3089 /* In most cases the matching rule for char classes
3090 only uses the syntax table for multibyte chars,
3091 so that the content of the syntax-table it is not
3092 hardcoded in the range_table. SPACE and WORD are
3093 the two exceptions. */
3094 if ((1 << cc) & ((1 << RECC_SPACE) | (1 << RECC_WORD)))
3095 bufp->used_syntax = 1;
3096
3097 /* Repeat the loop. */
3098 continue;
3099 }
3100 else
3101 {
3102 /* Go back to right after the "[:". */
3103 p = class_beg;
3104 SET_LIST_BIT ('[');
3105
3106 /* Because the `:' may starts the range, we
3107 can't simply set bit and repeat the loop.
3108 Instead, just set it to C and handle below. */
3109 c = ':';
3110 }
3111 }
3112
3113 if (p < pend && p[0] == '-' && p[1] != ']')
3114 {
3115
3116 /* Discard the `-'. */
3117 PATFETCH (c1);
3118
3119 /* Fetch the character which ends the range. */
3120 PATFETCH (c1);
3121 #ifdef emacs
3122 if (CHAR_BYTE8_P (c1)
3123 && ! ASCII_CHAR_P (c) && ! CHAR_BYTE8_P (c))
3124 /* Treat the range from a multibyte character to
3125 raw-byte character as empty. */
3126 c = c1 + 1;
3127 #endif /* emacs */
3128 }
3129 else
3130 /* Range from C to C. */
3131 c1 = c;
3132
3133 if (c > c1)
3134 {
3135 if (syntax & RE_NO_EMPTY_RANGES)
3136 FREE_STACK_RETURN (REG_ERANGEX);
3137 /* Else, repeat the loop. */
3138 }
3139 else
3140 {
3141 #ifndef emacs
3142 /* Set the range into bitmap */
3143 for (; c <= c1; c++)
3144 {
3145 ch = TRANSLATE (c);
3146 if (ch < (1 << BYTEWIDTH))
3147 SET_LIST_BIT (ch);
3148 }
3149 #else /* emacs */
3150 if (c < 128)
3151 {
3152 ch = MIN (127, c1);
3153 SETUP_ASCII_RANGE (range_table_work, c, ch);
3154 c = ch + 1;
3155 if (CHAR_BYTE8_P (c1))
3156 c = BYTE8_TO_CHAR (128);
3157 }
3158 if (c <= c1)
3159 {
3160 if (CHAR_BYTE8_P (c))
3161 {
3162 c = CHAR_TO_BYTE8 (c);
3163 c1 = CHAR_TO_BYTE8 (c1);
3164 for (; c <= c1; c++)
3165 SET_LIST_BIT (c);
3166 }
3167 else if (multibyte)
3168 {
3169 SETUP_MULTIBYTE_RANGE (range_table_work, c, c1);
3170 }
3171 else
3172 {
3173 SETUP_UNIBYTE_RANGE (range_table_work, c, c1);
3174 }
3175 }
3176 #endif /* emacs */
3177 }
3178 }
3179
3180 /* Discard any (non)matching list bytes that are all 0 at the
3181 end of the map. Decrease the map-length byte too. */
3182 while ((int) b[-1] > 0 && b[b[-1] - 1] == 0)
3183 b[-1]--;
3184 b += b[-1];
3185
3186 /* Build real range table from work area. */
3187 if (RANGE_TABLE_WORK_USED (range_table_work)
3188 || RANGE_TABLE_WORK_BITS (range_table_work))
3189 {
3190 int i;
3191 int used = RANGE_TABLE_WORK_USED (range_table_work);
3192
3193 /* Allocate space for COUNT + RANGE_TABLE. Needs two
3194 bytes for flags, two for COUNT, and three bytes for
3195 each character. */
3196 GET_BUFFER_SPACE (4 + used * 3);
3197
3198 /* Indicate the existence of range table. */
3199 laststart[1] |= 0x80;
3200
3201 /* Store the character class flag bits into the range table.
3202 If not in emacs, these flag bits are always 0. */
3203 *b++ = RANGE_TABLE_WORK_BITS (range_table_work) & 0xff;
3204 *b++ = RANGE_TABLE_WORK_BITS (range_table_work) >> 8;
3205
3206 STORE_NUMBER_AND_INCR (b, used / 2);
3207 for (i = 0; i < used; i++)
3208 STORE_CHARACTER_AND_INCR
3209 (b, RANGE_TABLE_WORK_ELT (range_table_work, i));
3210 }
3211 }
3212 break;
3213
3214
3215 case '(':
3216 if (syntax & RE_NO_BK_PARENS)
3217 goto handle_open;
3218 else
3219 goto normal_char;
3220
3221
3222 case ')':
3223 if (syntax & RE_NO_BK_PARENS)
3224 goto handle_close;
3225 else
3226 goto normal_char;
3227
3228
3229 case '\n':
3230 if (syntax & RE_NEWLINE_ALT)
3231 goto handle_alt;
3232 else
3233 goto normal_char;
3234
3235
3236 case '|':
3237 if (syntax & RE_NO_BK_VBAR)
3238 goto handle_alt;
3239 else
3240 goto normal_char;
3241
3242
3243 case '{':
3244 if (syntax & RE_INTERVALS && syntax & RE_NO_BK_BRACES)
3245 goto handle_interval;
3246 else
3247 goto normal_char;
3248
3249
3250 case '\\':
3251 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
3252
3253 /* Do not translate the character after the \, so that we can
3254 distinguish, e.g., \B from \b, even if we normally would
3255 translate, e.g., B to b. */
3256 PATFETCH (c);
3257
3258 switch (c)
3259 {
3260 case '(':
3261 if (syntax & RE_NO_BK_PARENS)
3262 goto normal_backslash;
3263
3264 handle_open:
3265 {
3266 int shy = 0;
3267 regnum_t regnum = 0;
3268 if (p+1 < pend)
3269 {
3270 /* Look for a special (?...) construct */
3271 if ((syntax & RE_SHY_GROUPS) && *p == '?')
3272 {
3273 PATFETCH (c); /* Gobble up the '?'. */
3274 while (!shy)
3275 {
3276 PATFETCH (c);
3277 switch (c)
3278 {
3279 case ':': shy = 1; break;
3280 case '0':
3281 /* An explicitly specified regnum must start
3282 with non-0. */
3283 if (regnum == 0)
3284 FREE_STACK_RETURN (REG_BADPAT);
3285 case '1': case '2': case '3': case '4':
3286 case '5': case '6': case '7': case '8': case '9':
3287 regnum = 10*regnum + (c - '0'); break;
3288 default:
3289 /* Only (?:...) is supported right now. */
3290 FREE_STACK_RETURN (REG_BADPAT);
3291 }
3292 }
3293 }
3294 }
3295
3296 if (!shy)
3297 regnum = ++bufp->re_nsub;
3298 else if (regnum)
3299 { /* It's actually not shy, but explicitly numbered. */
3300 shy = 0;
3301 if (regnum > bufp->re_nsub)
3302 bufp->re_nsub = regnum;
3303 else if (regnum > bufp->re_nsub
3304 /* Ideally, we'd want to check that the specified
3305 group can't have matched (i.e. all subgroups
3306 using the same regnum are in other branches of
3307 OR patterns), but we don't currently keep track
3308 of enough info to do that easily. */
3309 || group_in_compile_stack (compile_stack, regnum))
3310 FREE_STACK_RETURN (REG_BADPAT);
3311 }
3312 else
3313 /* It's really shy. */
3314 regnum = - bufp->re_nsub;
3315
3316 if (COMPILE_STACK_FULL)
3317 {
3318 RETALLOC (compile_stack.stack, compile_stack.size << 1,
3319 compile_stack_elt_t);
3320 if (compile_stack.stack == NULL) return REG_ESPACE;
3321
3322 compile_stack.size <<= 1;
3323 }
3324
3325 /* These are the values to restore when we hit end of this
3326 group. They are all relative offsets, so that if the
3327 whole pattern moves because of realloc, they will still
3328 be valid. */
3329 COMPILE_STACK_TOP.begalt_offset = begalt - bufp->buffer;
3330 COMPILE_STACK_TOP.fixup_alt_jump
3331 = fixup_alt_jump ? fixup_alt_jump - bufp->buffer + 1 : 0;
3332 COMPILE_STACK_TOP.laststart_offset = b - bufp->buffer;
3333 COMPILE_STACK_TOP.regnum = regnum;
3334
3335 /* Do not push a start_memory for groups beyond the last one
3336 we can represent in the compiled pattern. */
3337 if (regnum <= MAX_REGNUM && regnum > 0)
3338 BUF_PUSH_2 (start_memory, regnum);
3339
3340 compile_stack.avail++;
3341
3342 fixup_alt_jump = 0;
3343 laststart = 0;
3344 begalt = b;
3345 /* If we've reached MAX_REGNUM groups, then this open
3346 won't actually generate any code, so we'll have to
3347 clear pending_exact explicitly. */
3348 pending_exact = 0;
3349 break;
3350 }
3351
3352 case ')':
3353 if (syntax & RE_NO_BK_PARENS) goto normal_backslash;
3354
3355 if (COMPILE_STACK_EMPTY)
3356 {
3357 if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD)
3358 goto normal_backslash;
3359 else
3360 FREE_STACK_RETURN (REG_ERPAREN);
3361 }
3362
3363 handle_close:
3364 FIXUP_ALT_JUMP ();
3365
3366 /* See similar code for backslashed left paren above. */
3367 if (COMPILE_STACK_EMPTY)
3368 {
3369 if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD)
3370 goto normal_char;
3371 else
3372 FREE_STACK_RETURN (REG_ERPAREN);
3373 }
3374
3375 /* Since we just checked for an empty stack above, this
3376 ``can't happen''. */
3377 assert (compile_stack.avail != 0);
3378 {
3379 /* We don't just want to restore into `regnum', because
3380 later groups should continue to be numbered higher,
3381 as in `(ab)c(de)' -- the second group is #2. */
3382 regnum_t regnum;
3383
3384 compile_stack.avail--;
3385 begalt = bufp->buffer + COMPILE_STACK_TOP.begalt_offset;
3386 fixup_alt_jump
3387 = COMPILE_STACK_TOP.fixup_alt_jump
3388 ? bufp->buffer + COMPILE_STACK_TOP.fixup_alt_jump - 1
3389 : 0;
3390 laststart = bufp->buffer + COMPILE_STACK_TOP.laststart_offset;
3391 regnum = COMPILE_STACK_TOP.regnum;
3392 /* If we've reached MAX_REGNUM groups, then this open
3393 won't actually generate any code, so we'll have to
3394 clear pending_exact explicitly. */
3395 pending_exact = 0;
3396
3397 /* We're at the end of the group, so now we know how many
3398 groups were inside this one. */
3399 if (regnum <= MAX_REGNUM && regnum > 0)
3400 BUF_PUSH_2 (stop_memory, regnum);
3401 }
3402 break;
3403
3404
3405 case '|': /* `\|'. */
3406 if (syntax & RE_LIMITED_OPS || syntax & RE_NO_BK_VBAR)
3407 goto normal_backslash;
3408 handle_alt:
3409 if (syntax & RE_LIMITED_OPS)
3410 goto normal_char;
3411
3412 /* Insert before the previous alternative a jump which
3413 jumps to this alternative if the former fails. */
3414 GET_BUFFER_SPACE (3);
3415 INSERT_JUMP (on_failure_jump, begalt, b + 6);
3416 pending_exact = 0;
3417 b += 3;
3418
3419 /* The alternative before this one has a jump after it
3420 which gets executed if it gets matched. Adjust that
3421 jump so it will jump to this alternative's analogous
3422 jump (put in below, which in turn will jump to the next
3423 (if any) alternative's such jump, etc.). The last such
3424 jump jumps to the correct final destination. A picture:
3425 _____ _____
3426 | | | |
3427 | v | v
3428 a | b | c
3429
3430 If we are at `b', then fixup_alt_jump right now points to a
3431 three-byte space after `a'. We'll put in the jump, set
3432 fixup_alt_jump to right after `b', and leave behind three
3433 bytes which we'll fill in when we get to after `c'. */
3434
3435 FIXUP_ALT_JUMP ();
3436
3437 /* Mark and leave space for a jump after this alternative,
3438 to be filled in later either by next alternative or
3439 when know we're at the end of a series of alternatives. */
3440 fixup_alt_jump = b;
3441 GET_BUFFER_SPACE (3);
3442 b += 3;
3443
3444 laststart = 0;
3445 begalt = b;
3446 break;
3447
3448
3449 case '{':
3450 /* If \{ is a literal. */
3451 if (!(syntax & RE_INTERVALS)
3452 /* If we're at `\{' and it's not the open-interval
3453 operator. */
3454 || (syntax & RE_NO_BK_BRACES))
3455 goto normal_backslash;
3456
3457 handle_interval:
3458 {
3459 /* If got here, then the syntax allows intervals. */
3460
3461 /* At least (most) this many matches must be made. */
3462 int lower_bound = 0, upper_bound = -1;
3463
3464 beg_interval = p;
3465
3466 GET_UNSIGNED_NUMBER (lower_bound);
3467
3468 if (c == ',')
3469 GET_UNSIGNED_NUMBER (upper_bound);
3470 else
3471 /* Interval such as `{1}' => match exactly once. */
3472 upper_bound = lower_bound;
3473
3474 if (lower_bound < 0 || upper_bound > RE_DUP_MAX
3475 || (upper_bound >= 0 && lower_bound > upper_bound))
3476 FREE_STACK_RETURN (REG_BADBR);
3477
3478 if (!(syntax & RE_NO_BK_BRACES))
3479 {
3480 if (c != '\\')
3481 FREE_STACK_RETURN (REG_BADBR);
3482 if (p == pend)
3483 FREE_STACK_RETURN (REG_EESCAPE);
3484 PATFETCH (c);
3485 }
3486
3487 if (c != '}')
3488 FREE_STACK_RETURN (REG_BADBR);
3489
3490 /* We just parsed a valid interval. */
3491
3492 /* If it's invalid to have no preceding re. */
3493 if (!laststart)
3494 {
3495 if (syntax & RE_CONTEXT_INVALID_OPS)
3496 FREE_STACK_RETURN (REG_BADRPT);
3497 else if (syntax & RE_CONTEXT_INDEP_OPS)
3498 laststart = b;
3499 else
3500 goto unfetch_interval;
3501 }
3502
3503 if (upper_bound == 0)
3504 /* If the upper bound is zero, just drop the sub pattern
3505 altogether. */
3506 b = laststart;
3507 else if (lower_bound == 1 && upper_bound == 1)
3508 /* Just match it once: nothing to do here. */
3509 ;
3510
3511 /* Otherwise, we have a nontrivial interval. When
3512 we're all done, the pattern will look like:
3513 set_number_at <jump count> <upper bound>
3514 set_number_at <succeed_n count> <lower bound>
3515 succeed_n <after jump addr> <succeed_n count>
3516 <body of loop>
3517 jump_n <succeed_n addr> <jump count>
3518 (The upper bound and `jump_n' are omitted if
3519 `upper_bound' is 1, though.) */
3520 else
3521 { /* If the upper bound is > 1, we need to insert
3522 more at the end of the loop. */
3523 unsigned int nbytes = (upper_bound < 0 ? 3
3524 : upper_bound > 1 ? 5 : 0);
3525 unsigned int startoffset = 0;
3526
3527 GET_BUFFER_SPACE (20); /* We might use less. */
3528
3529 if (lower_bound == 0)
3530 {
3531 /* A succeed_n that starts with 0 is really a
3532 a simple on_failure_jump_loop. */
3533 INSERT_JUMP (on_failure_jump_loop, laststart,
3534 b + 3 + nbytes);
3535 b += 3;
3536 }
3537 else
3538 {
3539 /* Initialize lower bound of the `succeed_n', even
3540 though it will be set during matching by its
3541 attendant `set_number_at' (inserted next),
3542 because `re_compile_fastmap' needs to know.
3543 Jump to the `jump_n' we might insert below. */
3544 INSERT_JUMP2 (succeed_n, laststart,
3545 b + 5 + nbytes,
3546 lower_bound);
3547 b += 5;
3548
3549 /* Code to initialize the lower bound. Insert
3550 before the `succeed_n'. The `5' is the last two
3551 bytes of this `set_number_at', plus 3 bytes of
3552 the following `succeed_n'. */
3553 insert_op2 (set_number_at, laststart, 5, lower_bound, b);
3554 b += 5;
3555 startoffset += 5;
3556 }
3557
3558 if (upper_bound < 0)
3559 {
3560 /* A negative upper bound stands for infinity,
3561 in which case it degenerates to a plain jump. */
3562 STORE_JUMP (jump, b, laststart + startoffset);
3563 b += 3;
3564 }
3565 else if (upper_bound > 1)
3566 { /* More than one repetition is allowed, so
3567 append a backward jump to the `succeed_n'
3568 that starts this interval.
3569
3570 When we've reached this during matching,
3571 we'll have matched the interval once, so
3572 jump back only `upper_bound - 1' times. */
3573 STORE_JUMP2 (jump_n, b, laststart + startoffset,
3574 upper_bound - 1);
3575 b += 5;
3576
3577 /* The location we want to set is the second
3578 parameter of the `jump_n'; that is `b-2' as
3579 an absolute address. `laststart' will be
3580 the `set_number_at' we're about to insert;
3581 `laststart+3' the number to set, the source
3582 for the relative address. But we are
3583 inserting into the middle of the pattern --
3584 so everything is getting moved up by 5.
3585 Conclusion: (b - 2) - (laststart + 3) + 5,
3586 i.e., b - laststart.
3587
3588 We insert this at the beginning of the loop
3589 so that if we fail during matching, we'll
3590 reinitialize the bounds. */
3591 insert_op2 (set_number_at, laststart, b - laststart,
3592 upper_bound - 1, b);
3593 b += 5;
3594 }
3595 }
3596 pending_exact = 0;
3597 beg_interval = NULL;
3598 }
3599 break;
3600
3601 unfetch_interval:
3602 /* If an invalid interval, match the characters as literals. */
3603 assert (beg_interval);
3604 p = beg_interval;
3605 beg_interval = NULL;
3606
3607 /* normal_char and normal_backslash need `c'. */
3608 c = '{';
3609
3610 if (!(syntax & RE_NO_BK_BRACES))
3611 {
3612 assert (p > pattern && p[-1] == '\\');
3613 goto normal_backslash;
3614 }
3615 else
3616 goto normal_char;
3617
3618 #ifdef emacs
3619 /* There is no way to specify the before_dot and after_dot
3620 operators. rms says this is ok. --karl */
3621 case '=':
3622 BUF_PUSH (at_dot);
3623 break;
3624
3625 case 's':
3626 laststart = b;
3627 PATFETCH (c);
3628 BUF_PUSH_2 (syntaxspec, syntax_spec_code[c]);
3629 break;
3630
3631 case 'S':
3632 laststart = b;
3633 PATFETCH (c);
3634 BUF_PUSH_2 (notsyntaxspec, syntax_spec_code[c]);
3635 break;
3636
3637 case 'c':
3638 laststart = b;
3639 PATFETCH (c);
3640 BUF_PUSH_2 (categoryspec, c);
3641 break;
3642
3643 case 'C':
3644 laststart = b;
3645 PATFETCH (c);
3646 BUF_PUSH_2 (notcategoryspec, c);
3647 break;
3648 #endif /* emacs */
3649
3650
3651 case 'w':
3652 if (syntax & RE_NO_GNU_OPS)
3653 goto normal_char;
3654 laststart = b;
3655 BUF_PUSH_2 (syntaxspec, Sword);
3656 break;
3657
3658
3659 case 'W':
3660 if (syntax & RE_NO_GNU_OPS)
3661 goto normal_char;
3662 laststart = b;
3663 BUF_PUSH_2 (notsyntaxspec, Sword);
3664 break;
3665
3666
3667 case '<':
3668 if (syntax & RE_NO_GNU_OPS)
3669 goto normal_char;
3670 BUF_PUSH (wordbeg);
3671 break;
3672
3673 case '>':
3674 if (syntax & RE_NO_GNU_OPS)
3675 goto normal_char;
3676 BUF_PUSH (wordend);
3677 break;
3678
3679 case '_':
3680 if (syntax & RE_NO_GNU_OPS)
3681 goto normal_char;
3682 laststart = b;
3683 PATFETCH (c);
3684 if (c == '<')
3685 BUF_PUSH (symbeg);
3686 else if (c == '>')
3687 BUF_PUSH (symend);
3688 else
3689 FREE_STACK_RETURN (REG_BADPAT);
3690 break;
3691
3692 case 'b':
3693 if (syntax & RE_NO_GNU_OPS)
3694 goto normal_char;
3695 BUF_PUSH (wordbound);
3696 break;
3697
3698 case 'B':
3699 if (syntax & RE_NO_GNU_OPS)
3700 goto normal_char;
3701 BUF_PUSH (notwordbound);
3702 break;
3703
3704 case '`':
3705 if (syntax & RE_NO_GNU_OPS)
3706 goto normal_char;
3707 BUF_PUSH (begbuf);
3708 break;
3709
3710 case '\'':
3711 if (syntax & RE_NO_GNU_OPS)
3712 goto normal_char;
3713 BUF_PUSH (endbuf);
3714 break;
3715
3716 case '1': case '2': case '3': case '4': case '5':
3717 case '6': case '7': case '8': case '9':
3718 {
3719 regnum_t reg;
3720
3721 if (syntax & RE_NO_BK_REFS)
3722 goto normal_backslash;
3723
3724 reg = c - '0';
3725
3726 if (reg > bufp->re_nsub || reg < 1
3727 /* Can't back reference to a subexp before its end. */
3728 || group_in_compile_stack (compile_stack, reg))
3729 FREE_STACK_RETURN (REG_ESUBREG);
3730
3731 laststart = b;
3732 BUF_PUSH_2 (duplicate, reg);
3733 }
3734 break;
3735
3736
3737 case '+':
3738 case '?':
3739 if (syntax & RE_BK_PLUS_QM)
3740 goto handle_plus;
3741 else
3742 goto normal_backslash;
3743
3744 default:
3745 normal_backslash:
3746 /* You might think it would be useful for \ to mean
3747 not to translate; but if we don't translate it
3748 it will never match anything. */
3749 goto normal_char;
3750 }
3751 break;
3752
3753
3754 default:
3755 /* Expects the character in `c'. */
3756 normal_char:
3757 /* If no exactn currently being built. */
3758 if (!pending_exact
3759
3760 /* If last exactn not at current position. */
3761 || pending_exact + *pending_exact + 1 != b
3762
3763 /* We have only one byte following the exactn for the count. */
3764 || *pending_exact >= (1 << BYTEWIDTH) - MAX_MULTIBYTE_LENGTH
3765
3766 /* If followed by a repetition operator. */
3767 || (p != pend && (*p == '*' || *p == '^'))
3768 || ((syntax & RE_BK_PLUS_QM)
3769 ? p + 1 < pend && *p == '\\' && (p[1] == '+' || p[1] == '?')
3770 : p != pend && (*p == '+' || *p == '?'))
3771 || ((syntax & RE_INTERVALS)
3772 && ((syntax & RE_NO_BK_BRACES)
3773 ? p != pend && *p == '{'
3774 : p + 1 < pend && p[0] == '\\' && p[1] == '{')))
3775 {
3776 /* Start building a new exactn. */
3777
3778 laststart = b;
3779
3780 BUF_PUSH_2 (exactn, 0);
3781 pending_exact = b - 1;
3782 }
3783
3784 GET_BUFFER_SPACE (MAX_MULTIBYTE_LENGTH);
3785 {
3786 int len;
3787
3788 if (multibyte)
3789 {
3790 c = TRANSLATE (c);
3791 len = CHAR_STRING (c, b);
3792 b += len;
3793 }
3794 else
3795 {
3796 c1 = RE_CHAR_TO_MULTIBYTE (c);
3797 if (! CHAR_BYTE8_P (c1))
3798 {
3799 re_wchar_t c2 = TRANSLATE (c1);
3800
3801 if (c1 != c2 && (c1 = RE_CHAR_TO_UNIBYTE (c2)) >= 0)
3802 c = c1;
3803 }
3804 *b++ = c;
3805 len = 1;
3806 }
3807 (*pending_exact) += len;
3808 }
3809
3810 break;
3811 } /* switch (c) */
3812 } /* while p != pend */
3813
3814
3815 /* Through the pattern now. */
3816
3817 FIXUP_ALT_JUMP ();
3818
3819 if (!COMPILE_STACK_EMPTY)
3820 FREE_STACK_RETURN (REG_EPAREN);
3821
3822 /* If we don't want backtracking, force success
3823 the first time we reach the end of the compiled pattern. */
3824 if (syntax & RE_NO_POSIX_BACKTRACKING)
3825 BUF_PUSH (succeed);
3826
3827 /* We have succeeded; set the length of the buffer. */
3828 bufp->used = b - bufp->buffer;
3829
3830 #ifdef DEBUG
3831 if (debug > 0)
3832 {
3833 re_compile_fastmap (bufp);
3834 DEBUG_PRINT1 ("\nCompiled pattern: \n");
3835 print_compiled_pattern (bufp);
3836 }
3837 debug--;
3838 #endif /* DEBUG */
3839
3840 #ifndef MATCH_MAY_ALLOCATE
3841 /* Initialize the failure stack to the largest possible stack. This
3842 isn't necessary unless we're trying to avoid calling alloca in
3843 the search and match routines. */
3844 {
3845 int num_regs = bufp->re_nsub + 1;
3846
3847 if (fail_stack.size < re_max_failures * TYPICAL_FAILURE_SIZE)
3848 {
3849 fail_stack.size = re_max_failures * TYPICAL_FAILURE_SIZE;
3850
3851 if (! fail_stack.stack)
3852 fail_stack.stack
3853 = (fail_stack_elt_t *) malloc (fail_stack.size
3854 * sizeof (fail_stack_elt_t));
3855 else
3856 fail_stack.stack
3857 = (fail_stack_elt_t *) realloc (fail_stack.stack,
3858 (fail_stack.size
3859 * sizeof (fail_stack_elt_t)));
3860 }
3861
3862 regex_grow_registers (num_regs);
3863 }
3864 #endif /* not MATCH_MAY_ALLOCATE */
3865
3866 FREE_STACK_RETURN (REG_NOERROR);
3867 } /* regex_compile */
3868 \f
3869 /* Subroutines for `regex_compile'. */
3870
3871 /* Store OP at LOC followed by two-byte integer parameter ARG. */
3872
3873 static void
3874 store_op1 (op, loc, arg)
3875 re_opcode_t op;
3876 unsigned char *loc;
3877 int arg;
3878 {
3879 *loc = (unsigned char) op;
3880 STORE_NUMBER (loc + 1, arg);
3881 }
3882
3883
3884 /* Like `store_op1', but for two two-byte parameters ARG1 and ARG2. */
3885
3886 static void
3887 store_op2 (op, loc, arg1, arg2)
3888 re_opcode_t op;
3889 unsigned char *loc;
3890 int arg1, arg2;
3891 {
3892 *loc = (unsigned char) op;
3893 STORE_NUMBER (loc + 1, arg1);
3894 STORE_NUMBER (loc + 3, arg2);
3895 }
3896
3897
3898 /* Copy the bytes from LOC to END to open up three bytes of space at LOC
3899 for OP followed by two-byte integer parameter ARG. */
3900
3901 static void
3902 insert_op1 (op, loc, arg, end)
3903 re_opcode_t op;
3904 unsigned char *loc;
3905 int arg;
3906 unsigned char *end;
3907 {
3908 register unsigned char *pfrom = end;
3909 register unsigned char *pto = end + 3;
3910
3911 while (pfrom != loc)
3912 *--pto = *--pfrom;
3913
3914 store_op1 (op, loc, arg);
3915 }
3916
3917
3918 /* Like `insert_op1', but for two two-byte parameters ARG1 and ARG2. */
3919
3920 static void
3921 insert_op2 (op, loc, arg1, arg2, end)
3922 re_opcode_t op;
3923 unsigned char *loc;
3924 int arg1, arg2;
3925 unsigned char *end;
3926 {
3927 register unsigned char *pfrom = end;
3928 register unsigned char *pto = end + 5;
3929
3930 while (pfrom != loc)
3931 *--pto = *--pfrom;
3932
3933 store_op2 (op, loc, arg1, arg2);
3934 }
3935
3936
3937 /* P points to just after a ^ in PATTERN. Return true if that ^ comes
3938 after an alternative or a begin-subexpression. We assume there is at
3939 least one character before the ^. */
3940
3941 static boolean
3942 at_begline_loc_p (pattern, p, syntax)
3943 re_char *pattern, *p;
3944 reg_syntax_t syntax;
3945 {
3946 re_char *prev = p - 2;
3947 boolean prev_prev_backslash = prev > pattern && prev[-1] == '\\';
3948
3949 return
3950 /* After a subexpression? */
3951 (*prev == '(' && (syntax & RE_NO_BK_PARENS || prev_prev_backslash))
3952 /* After an alternative? */
3953 || (*prev == '|' && (syntax & RE_NO_BK_VBAR || prev_prev_backslash))
3954 /* After a shy subexpression? */
3955 || ((syntax & RE_SHY_GROUPS) && prev - 2 >= pattern
3956 && prev[-1] == '?' && prev[-2] == '('
3957 && (syntax & RE_NO_BK_PARENS
3958 || (prev - 3 >= pattern && prev[-3] == '\\')));
3959 }
3960
3961
3962 /* The dual of at_begline_loc_p. This one is for $. We assume there is
3963 at least one character after the $, i.e., `P < PEND'. */
3964
3965 static boolean
3966 at_endline_loc_p (p, pend, syntax)
3967 re_char *p, *pend;
3968 reg_syntax_t syntax;
3969 {
3970 re_char *next = p;
3971 boolean next_backslash = *next == '\\';
3972 re_char *next_next = p + 1 < pend ? p + 1 : 0;
3973
3974 return
3975 /* Before a subexpression? */
3976 (syntax & RE_NO_BK_PARENS ? *next == ')'
3977 : next_backslash && next_next && *next_next == ')')
3978 /* Before an alternative? */
3979 || (syntax & RE_NO_BK_VBAR ? *next == '|'
3980 : next_backslash && next_next && *next_next == '|');
3981 }
3982
3983
3984 /* Returns true if REGNUM is in one of COMPILE_STACK's elements and
3985 false if it's not. */
3986
3987 static boolean
3988 group_in_compile_stack (compile_stack, regnum)
3989 compile_stack_type compile_stack;
3990 regnum_t regnum;
3991 {
3992 int this_element;
3993
3994 for (this_element = compile_stack.avail - 1;
3995 this_element >= 0;
3996 this_element--)
3997 if (compile_stack.stack[this_element].regnum == regnum)
3998 return true;
3999
4000 return false;
4001 }
4002 \f
4003 /* analyse_first.
4004 If fastmap is non-NULL, go through the pattern and fill fastmap
4005 with all the possible leading chars. If fastmap is NULL, don't
4006 bother filling it up (obviously) and only return whether the
4007 pattern could potentially match the empty string.
4008
4009 Return 1 if p..pend might match the empty string.
4010 Return 0 if p..pend matches at least one char.
4011 Return -1 if fastmap was not updated accurately. */
4012
4013 static int
4014 analyse_first (p, pend, fastmap, multibyte)
4015 re_char *p, *pend;
4016 char *fastmap;
4017 const int multibyte;
4018 {
4019 int j, k;
4020 boolean not;
4021
4022 /* If all elements for base leading-codes in fastmap is set, this
4023 flag is set true. */
4024 boolean match_any_multibyte_characters = false;
4025
4026 assert (p);
4027
4028 /* The loop below works as follows:
4029 - It has a working-list kept in the PATTERN_STACK and which basically
4030 starts by only containing a pointer to the first operation.
4031 - If the opcode we're looking at is a match against some set of
4032 chars, then we add those chars to the fastmap and go on to the
4033 next work element from the worklist (done via `break').
4034 - If the opcode is a control operator on the other hand, we either
4035 ignore it (if it's meaningless at this point, such as `start_memory')
4036 or execute it (if it's a jump). If the jump has several destinations
4037 (i.e. `on_failure_jump'), then we push the other destination onto the
4038 worklist.
4039 We guarantee termination by ignoring backward jumps (more or less),
4040 so that `p' is monotonically increasing. More to the point, we
4041 never set `p' (or push) anything `<= p1'. */
4042
4043 while (p < pend)
4044 {
4045 /* `p1' is used as a marker of how far back a `on_failure_jump'
4046 can go without being ignored. It is normally equal to `p'
4047 (which prevents any backward `on_failure_jump') except right
4048 after a plain `jump', to allow patterns such as:
4049 0: jump 10
4050 3..9: <body>
4051 10: on_failure_jump 3
4052 as used for the *? operator. */
4053 re_char *p1 = p;
4054
4055 switch (SWITCH_ENUM_CAST ((re_opcode_t) *p++))
4056 {
4057 case succeed:
4058 return 1;
4059 continue;
4060
4061 case duplicate:
4062 /* If the first character has to match a backreference, that means
4063 that the group was empty (since it already matched). Since this
4064 is the only case that interests us here, we can assume that the
4065 backreference must match the empty string. */
4066 p++;
4067 continue;
4068
4069
4070 /* Following are the cases which match a character. These end
4071 with `break'. */
4072
4073 case exactn:
4074 if (fastmap)
4075 {
4076 /* If multibyte is nonzero, the first byte of each
4077 character is an ASCII or a leading code. Otherwise,
4078 each byte is a character. Thus, this works in both
4079 cases. */
4080 fastmap[p[1]] = 1;
4081 if (! multibyte)
4082 {
4083 /* For the case of matching this unibyte regex
4084 against multibyte, we must set a leading code of
4085 the corresponding multibyte character. */
4086 int c = RE_CHAR_TO_MULTIBYTE (p[1]);
4087
4088 if (! CHAR_BYTE8_P (c))
4089 fastmap[CHAR_LEADING_CODE (c)] = 1;
4090 }
4091 }
4092 break;
4093
4094
4095 case anychar:
4096 /* We could put all the chars except for \n (and maybe \0)
4097 but we don't bother since it is generally not worth it. */
4098 if (!fastmap) break;
4099 return -1;
4100
4101
4102 case charset_not:
4103 if (!fastmap) break;
4104 {
4105 /* Chars beyond end of bitmap are possible matches. */
4106 for (j = CHARSET_BITMAP_SIZE (&p[-1]) * BYTEWIDTH;
4107 j < (1 << BYTEWIDTH); j++)
4108 fastmap[j] = 1;
4109 }
4110
4111 /* Fallthrough */
4112 case charset:
4113 if (!fastmap) break;
4114 not = (re_opcode_t) *(p - 1) == charset_not;
4115 for (j = CHARSET_BITMAP_SIZE (&p[-1]) * BYTEWIDTH - 1, p++;
4116 j >= 0; j--)
4117 if (!!(p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH))) ^ not)
4118 fastmap[j] = 1;
4119
4120 #ifdef emacs
4121 if (/* Any leading code can possibly start a character
4122 which doesn't match the specified set of characters. */
4123 not
4124 ||
4125 /* If we can match a character class, we can match any
4126 multibyte characters. */
4127 (CHARSET_RANGE_TABLE_EXISTS_P (&p[-2])
4128 && CHARSET_RANGE_TABLE_BITS (&p[-2]) != 0))
4129
4130 {
4131 if (match_any_multibyte_characters == false)
4132 {
4133 for (j = MIN_MULTIBYTE_LEADING_CODE;
4134 j <= MAX_MULTIBYTE_LEADING_CODE; j++)
4135 fastmap[j] = 1;
4136 match_any_multibyte_characters = true;
4137 }
4138 }
4139
4140 else if (!not && CHARSET_RANGE_TABLE_EXISTS_P (&p[-2])
4141 && match_any_multibyte_characters == false)
4142 {
4143 /* Set fastmap[I] to 1 where I is a leading code of each
4144 multibyte characer in the range table. */
4145 int c, count;
4146 unsigned char lc1, lc2;
4147
4148 /* Make P points the range table. `+ 2' is to skip flag
4149 bits for a character class. */
4150 p += CHARSET_BITMAP_SIZE (&p[-2]) + 2;
4151
4152 /* Extract the number of ranges in range table into COUNT. */
4153 EXTRACT_NUMBER_AND_INCR (count, p);
4154 for (; count > 0; count--, p += 3)
4155 {
4156 /* Extract the start and end of each range. */
4157 EXTRACT_CHARACTER (c, p);
4158 lc1 = CHAR_LEADING_CODE (c);
4159 p += 3;
4160 EXTRACT_CHARACTER (c, p);
4161 lc2 = CHAR_LEADING_CODE (c);
4162 for (j = lc1; j <= lc2; j++)
4163 fastmap[j] = 1;
4164 }
4165 }
4166 #endif
4167 break;
4168
4169 case syntaxspec:
4170 case notsyntaxspec:
4171 if (!fastmap) break;
4172 #ifndef emacs
4173 not = (re_opcode_t)p[-1] == notsyntaxspec;
4174 k = *p++;
4175 for (j = 0; j < (1 << BYTEWIDTH); j++)
4176 if ((SYNTAX (j) == (enum syntaxcode) k) ^ not)
4177 fastmap[j] = 1;
4178 break;
4179 #else /* emacs */
4180 /* This match depends on text properties. These end with
4181 aborting optimizations. */
4182 return -1;
4183
4184 case categoryspec:
4185 case notcategoryspec:
4186 if (!fastmap) break;
4187 not = (re_opcode_t)p[-1] == notcategoryspec;
4188 k = *p++;
4189 for (j = (1 << BYTEWIDTH); j >= 0; j--)
4190 if ((CHAR_HAS_CATEGORY (j, k)) ^ not)
4191 fastmap[j] = 1;
4192
4193 /* Any leading code can possibly start a character which
4194 has or doesn't has the specified category. */
4195 if (match_any_multibyte_characters == false)
4196 {
4197 for (j = MIN_MULTIBYTE_LEADING_CODE;
4198 j <= MAX_MULTIBYTE_LEADING_CODE; j++)
4199 fastmap[j] = 1;
4200 match_any_multibyte_characters = true;
4201 }
4202 break;
4203
4204 /* All cases after this match the empty string. These end with
4205 `continue'. */
4206
4207 case before_dot:
4208 case at_dot:
4209 case after_dot:
4210 #endif /* !emacs */
4211 case no_op:
4212 case begline:
4213 case endline:
4214 case begbuf:
4215 case endbuf:
4216 case wordbound:
4217 case notwordbound:
4218 case wordbeg:
4219 case wordend:
4220 case symbeg:
4221 case symend:
4222 continue;
4223
4224
4225 case jump:
4226 EXTRACT_NUMBER_AND_INCR (j, p);
4227 if (j < 0)
4228 /* Backward jumps can only go back to code that we've already
4229 visited. `re_compile' should make sure this is true. */
4230 break;
4231 p += j;
4232 switch (SWITCH_ENUM_CAST ((re_opcode_t) *p))
4233 {
4234 case on_failure_jump:
4235 case on_failure_keep_string_jump:
4236 case on_failure_jump_loop:
4237 case on_failure_jump_nastyloop:
4238 case on_failure_jump_smart:
4239 p++;
4240 break;
4241 default:
4242 continue;
4243 };
4244 /* Keep `p1' to allow the `on_failure_jump' we are jumping to
4245 to jump back to "just after here". */
4246 /* Fallthrough */
4247
4248 case on_failure_jump:
4249 case on_failure_keep_string_jump:
4250 case on_failure_jump_nastyloop:
4251 case on_failure_jump_loop:
4252 case on_failure_jump_smart:
4253 EXTRACT_NUMBER_AND_INCR (j, p);
4254 if (p + j <= p1)
4255 ; /* Backward jump to be ignored. */
4256 else
4257 { /* We have to look down both arms.
4258 We first go down the "straight" path so as to minimize
4259 stack usage when going through alternatives. */
4260 int r = analyse_first (p, pend, fastmap, multibyte);
4261 if (r) return r;
4262 p += j;
4263 }
4264 continue;
4265
4266
4267 case jump_n:
4268 /* This code simply does not properly handle forward jump_n. */
4269 DEBUG_STATEMENT (EXTRACT_NUMBER (j, p); assert (j < 0));
4270 p += 4;
4271 /* jump_n can either jump or fall through. The (backward) jump
4272 case has already been handled, so we only need to look at the
4273 fallthrough case. */
4274 continue;
4275
4276 case succeed_n:
4277 /* If N == 0, it should be an on_failure_jump_loop instead. */
4278 DEBUG_STATEMENT (EXTRACT_NUMBER (j, p + 2); assert (j > 0));
4279 p += 4;
4280 /* We only care about one iteration of the loop, so we don't
4281 need to consider the case where this behaves like an
4282 on_failure_jump. */
4283 continue;
4284
4285
4286 case set_number_at:
4287 p += 4;
4288 continue;
4289
4290
4291 case start_memory:
4292 case stop_memory:
4293 p += 1;
4294 continue;
4295
4296
4297 default:
4298 abort (); /* We have listed all the cases. */
4299 } /* switch *p++ */
4300
4301 /* Getting here means we have found the possible starting
4302 characters for one path of the pattern -- and that the empty
4303 string does not match. We need not follow this path further. */
4304 return 0;
4305 } /* while p */
4306
4307 /* We reached the end without matching anything. */
4308 return 1;
4309
4310 } /* analyse_first */
4311 \f
4312 /* re_compile_fastmap computes a ``fastmap'' for the compiled pattern in
4313 BUFP. A fastmap records which of the (1 << BYTEWIDTH) possible
4314 characters can start a string that matches the pattern. This fastmap
4315 is used by re_search to skip quickly over impossible starting points.
4316
4317 Character codes above (1 << BYTEWIDTH) are not represented in the
4318 fastmap, but the leading codes are represented. Thus, the fastmap
4319 indicates which character sets could start a match.
4320
4321 The caller must supply the address of a (1 << BYTEWIDTH)-byte data
4322 area as BUFP->fastmap.
4323
4324 We set the `fastmap', `fastmap_accurate', and `can_be_null' fields in
4325 the pattern buffer.
4326
4327 Returns 0 if we succeed, -2 if an internal error. */
4328
4329 int
4330 re_compile_fastmap (bufp)
4331 struct re_pattern_buffer *bufp;
4332 {
4333 char *fastmap = bufp->fastmap;
4334 int analysis;
4335
4336 assert (fastmap && bufp->buffer);
4337
4338 bzero (fastmap, 1 << BYTEWIDTH); /* Assume nothing's valid. */
4339 bufp->fastmap_accurate = 1; /* It will be when we're done. */
4340
4341 analysis = analyse_first (bufp->buffer, bufp->buffer + bufp->used,
4342 fastmap, RE_MULTIBYTE_P (bufp));
4343 bufp->can_be_null = (analysis != 0);
4344 return 0;
4345 } /* re_compile_fastmap */
4346 \f
4347 /* Set REGS to hold NUM_REGS registers, storing them in STARTS and
4348 ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use
4349 this memory for recording register information. STARTS and ENDS
4350 must be allocated using the malloc library routine, and must each
4351 be at least NUM_REGS * sizeof (regoff_t) bytes long.
4352
4353 If NUM_REGS == 0, then subsequent matches should allocate their own
4354 register data.
4355
4356 Unless this function is called, the first search or match using
4357 PATTERN_BUFFER will allocate its own register data, without
4358 freeing the old data. */
4359
4360 void
4361 re_set_registers (bufp, regs, num_regs, starts, ends)
4362 struct re_pattern_buffer *bufp;
4363 struct re_registers *regs;
4364 unsigned num_regs;
4365 regoff_t *starts, *ends;
4366 {
4367 if (num_regs)
4368 {
4369 bufp->regs_allocated = REGS_REALLOCATE;
4370 regs->num_regs = num_regs;
4371 regs->start = starts;
4372 regs->end = ends;
4373 }
4374 else
4375 {
4376 bufp->regs_allocated = REGS_UNALLOCATED;
4377 regs->num_regs = 0;
4378 regs->start = regs->end = (regoff_t *) 0;
4379 }
4380 }
4381 WEAK_ALIAS (__re_set_registers, re_set_registers)
4382 \f
4383 /* Searching routines. */
4384
4385 /* Like re_search_2, below, but only one string is specified, and
4386 doesn't let you say where to stop matching. */
4387
4388 int
4389 re_search (bufp, string, size, startpos, range, regs)
4390 struct re_pattern_buffer *bufp;
4391 const char *string;
4392 int size, startpos, range;
4393 struct re_registers *regs;
4394 {
4395 return re_search_2 (bufp, NULL, 0, string, size, startpos, range,
4396 regs, size);
4397 }
4398 WEAK_ALIAS (__re_search, re_search)
4399
4400 /* Head address of virtual concatenation of string. */
4401 #define HEAD_ADDR_VSTRING(P) \
4402 (((P) >= size1 ? string2 : string1))
4403
4404 /* End address of virtual concatenation of string. */
4405 #define STOP_ADDR_VSTRING(P) \
4406 (((P) >= size1 ? string2 + size2 : string1 + size1))
4407
4408 /* Address of POS in the concatenation of virtual string. */
4409 #define POS_ADDR_VSTRING(POS) \
4410 (((POS) >= size1 ? string2 - size1 : string1) + (POS))
4411
4412 /* Using the compiled pattern in BUFP->buffer, first tries to match the
4413 virtual concatenation of STRING1 and STRING2, starting first at index
4414 STARTPOS, then at STARTPOS + 1, and so on.
4415
4416 STRING1 and STRING2 have length SIZE1 and SIZE2, respectively.
4417
4418 RANGE is how far to scan while trying to match. RANGE = 0 means try
4419 only at STARTPOS; in general, the last start tried is STARTPOS +
4420 RANGE.
4421
4422 In REGS, return the indices of the virtual concatenation of STRING1
4423 and STRING2 that matched the entire BUFP->buffer and its contained
4424 subexpressions.
4425
4426 Do not consider matching one past the index STOP in the virtual
4427 concatenation of STRING1 and STRING2.
4428
4429 We return either the position in the strings at which the match was
4430 found, -1 if no match, or -2 if error (such as failure
4431 stack overflow). */
4432
4433 int
4434 re_search_2 (bufp, str1, size1, str2, size2, startpos, range, regs, stop)
4435 struct re_pattern_buffer *bufp;
4436 const char *str1, *str2;
4437 int size1, size2;
4438 int startpos;
4439 int range;
4440 struct re_registers *regs;
4441 int stop;
4442 {
4443 int val;
4444 re_char *string1 = (re_char*) str1;
4445 re_char *string2 = (re_char*) str2;
4446 register char *fastmap = bufp->fastmap;
4447 register RE_TRANSLATE_TYPE translate = bufp->translate;
4448 int total_size = size1 + size2;
4449 int endpos = startpos + range;
4450 boolean anchored_start;
4451 /* Nonzero if we are searching multibyte string. */
4452 const boolean multibyte = RE_TARGET_MULTIBYTE_P (bufp);
4453
4454 /* Check for out-of-range STARTPOS. */
4455 if (startpos < 0 || startpos > total_size)
4456 return -1;
4457
4458 /* Fix up RANGE if it might eventually take us outside
4459 the virtual concatenation of STRING1 and STRING2.
4460 Make sure we won't move STARTPOS below 0 or above TOTAL_SIZE. */
4461 if (endpos < 0)
4462 range = 0 - startpos;
4463 else if (endpos > total_size)
4464 range = total_size - startpos;
4465
4466 /* If the search isn't to be a backwards one, don't waste time in a
4467 search for a pattern anchored at beginning of buffer. */
4468 if (bufp->used > 0 && (re_opcode_t) bufp->buffer[0] == begbuf && range > 0)
4469 {
4470 if (startpos > 0)
4471 return -1;
4472 else
4473 range = 0;
4474 }
4475
4476 #ifdef emacs
4477 /* In a forward search for something that starts with \=.
4478 don't keep searching past point. */
4479 if (bufp->used > 0 && (re_opcode_t) bufp->buffer[0] == at_dot && range > 0)
4480 {
4481 range = PT_BYTE - BEGV_BYTE - startpos;
4482 if (range < 0)
4483 return -1;
4484 }
4485 #endif /* emacs */
4486
4487 /* Update the fastmap now if not correct already. */
4488 if (fastmap && !bufp->fastmap_accurate)
4489 re_compile_fastmap (bufp);
4490
4491 /* See whether the pattern is anchored. */
4492 anchored_start = (bufp->buffer[0] == begline);
4493
4494 #ifdef emacs
4495 gl_state.object = re_match_object;
4496 {
4497 int charpos = SYNTAX_TABLE_BYTE_TO_CHAR (POS_AS_IN_BUFFER (startpos));
4498
4499 SETUP_SYNTAX_TABLE_FOR_OBJECT (re_match_object, charpos, 1);
4500 }
4501 #endif
4502
4503 /* Loop through the string, looking for a place to start matching. */
4504 for (;;)
4505 {
4506 /* If the pattern is anchored,
4507 skip quickly past places we cannot match.
4508 We don't bother to treat startpos == 0 specially
4509 because that case doesn't repeat. */
4510 if (anchored_start && startpos > 0)
4511 {
4512 if (! ((startpos <= size1 ? string1[startpos - 1]
4513 : string2[startpos - size1 - 1])
4514 == '\n'))
4515 goto advance;
4516 }
4517
4518 /* If a fastmap is supplied, skip quickly over characters that
4519 cannot be the start of a match. If the pattern can match the
4520 null string, however, we don't need to skip characters; we want
4521 the first null string. */
4522 if (fastmap && startpos < total_size && !bufp->can_be_null)
4523 {
4524 register re_char *d;
4525 register re_wchar_t buf_ch;
4526
4527 d = POS_ADDR_VSTRING (startpos);
4528
4529 if (range > 0) /* Searching forwards. */
4530 {
4531 register int lim = 0;
4532 int irange = range;
4533
4534 if (startpos < size1 && startpos + range >= size1)
4535 lim = range - (size1 - startpos);
4536
4537 /* Written out as an if-else to avoid testing `translate'
4538 inside the loop. */
4539 if (RE_TRANSLATE_P (translate))
4540 {
4541 if (multibyte)
4542 while (range > lim)
4543 {
4544 int buf_charlen;
4545
4546 buf_ch = STRING_CHAR_AND_LENGTH (d, range - lim,
4547 buf_charlen);
4548 buf_ch = RE_TRANSLATE (translate, buf_ch);
4549 if (fastmap[CHAR_LEADING_CODE (buf_ch)])
4550 break;
4551
4552 range -= buf_charlen;
4553 d += buf_charlen;
4554 }
4555 else
4556 while (range > lim)
4557 {
4558 register re_wchar_t ch, translated;
4559
4560 buf_ch = *d;
4561 ch = RE_CHAR_TO_MULTIBYTE (buf_ch);
4562 translated = RE_TRANSLATE (translate, ch);
4563 if (translated != ch
4564 && (ch = RE_CHAR_TO_UNIBYTE (translated)) >= 0)
4565 buf_ch = ch;
4566 if (fastmap[buf_ch])
4567 break;
4568 d++;
4569 range--;
4570 }
4571 }
4572 else
4573 {
4574 if (multibyte)
4575 while (range > lim)
4576 {
4577 int buf_charlen;
4578
4579 buf_ch = STRING_CHAR_AND_LENGTH (d, range - lim,
4580 buf_charlen);
4581 if (fastmap[CHAR_LEADING_CODE (buf_ch)])
4582 break;
4583 range -= buf_charlen;
4584 d += buf_charlen;
4585 }
4586 else
4587 while (range > lim && !fastmap[*d])
4588 {
4589 d++;
4590 range--;
4591 }
4592 }
4593 startpos += irange - range;
4594 }
4595 else /* Searching backwards. */
4596 {
4597 int room = (startpos >= size1
4598 ? size2 + size1 - startpos
4599 : size1 - startpos);
4600 if (multibyte)
4601 {
4602 buf_ch = STRING_CHAR (d, room);
4603 buf_ch = TRANSLATE (buf_ch);
4604 if (! fastmap[CHAR_LEADING_CODE (buf_ch)])
4605 goto advance;
4606 }
4607 else
4608 {
4609 register re_wchar_t ch, translated;
4610
4611 buf_ch = *d;
4612 ch = RE_CHAR_TO_MULTIBYTE (buf_ch);
4613 translated = TRANSLATE (ch);
4614 if (translated != ch
4615 && (ch = RE_CHAR_TO_UNIBYTE (translated)) >= 0)
4616 buf_ch = ch;
4617 if (! fastmap[TRANSLATE (buf_ch)])
4618 goto advance;
4619 }
4620 }
4621 }
4622
4623 /* If can't match the null string, and that's all we have left, fail. */
4624 if (range >= 0 && startpos == total_size && fastmap
4625 && !bufp->can_be_null)
4626 return -1;
4627
4628 val = re_match_2_internal (bufp, string1, size1, string2, size2,
4629 startpos, regs, stop);
4630
4631 if (val >= 0)
4632 return startpos;
4633
4634 if (val == -2)
4635 return -2;
4636
4637 advance:
4638 if (!range)
4639 break;
4640 else if (range > 0)
4641 {
4642 /* Update STARTPOS to the next character boundary. */
4643 if (multibyte)
4644 {
4645 re_char *p = POS_ADDR_VSTRING (startpos);
4646 re_char *pend = STOP_ADDR_VSTRING (startpos);
4647 int len = MULTIBYTE_FORM_LENGTH (p, pend - p);
4648
4649 range -= len;
4650 if (range < 0)
4651 break;
4652 startpos += len;
4653 }
4654 else
4655 {
4656 range--;
4657 startpos++;
4658 }
4659 }
4660 else
4661 {
4662 range++;
4663 startpos--;
4664
4665 /* Update STARTPOS to the previous character boundary. */
4666 if (multibyte)
4667 {
4668 re_char *p = POS_ADDR_VSTRING (startpos) + 1;
4669 re_char *p0 = p;
4670 re_char *phead = HEAD_ADDR_VSTRING (startpos);
4671
4672 /* Find the head of multibyte form. */
4673 PREV_CHAR_BOUNDARY (p, phead);
4674 range += p0 - 1 - p;
4675 if (range > 0)
4676 break;
4677
4678 startpos -= p0 - 1 - p;
4679 }
4680 }
4681 }
4682 return -1;
4683 } /* re_search_2 */
4684 WEAK_ALIAS (__re_search_2, re_search_2)
4685 \f
4686 /* Declarations and macros for re_match_2. */
4687
4688 static int bcmp_translate _RE_ARGS((re_char *s1, re_char *s2,
4689 register int len,
4690 RE_TRANSLATE_TYPE translate,
4691 const int multibyte));
4692
4693 /* This converts PTR, a pointer into one of the search strings `string1'
4694 and `string2' into an offset from the beginning of that string. */
4695 #define POINTER_TO_OFFSET(ptr) \
4696 (FIRST_STRING_P (ptr) \
4697 ? ((regoff_t) ((ptr) - string1)) \
4698 : ((regoff_t) ((ptr) - string2 + size1)))
4699
4700 /* Call before fetching a character with *d. This switches over to
4701 string2 if necessary.
4702 Check re_match_2_internal for a discussion of why end_match_2 might
4703 not be within string2 (but be equal to end_match_1 instead). */
4704 #define PREFETCH() \
4705 while (d == dend) \
4706 { \
4707 /* End of string2 => fail. */ \
4708 if (dend == end_match_2) \
4709 goto fail; \
4710 /* End of string1 => advance to string2. */ \
4711 d = string2; \
4712 dend = end_match_2; \
4713 }
4714
4715 /* Call before fetching a char with *d if you already checked other limits.
4716 This is meant for use in lookahead operations like wordend, etc..
4717 where we might need to look at parts of the string that might be
4718 outside of the LIMITs (i.e past `stop'). */
4719 #define PREFETCH_NOLIMIT() \
4720 if (d == end1) \
4721 { \
4722 d = string2; \
4723 dend = end_match_2; \
4724 } \
4725
4726 /* Test if at very beginning or at very end of the virtual concatenation
4727 of `string1' and `string2'. If only one string, it's `string2'. */
4728 #define AT_STRINGS_BEG(d) ((d) == (size1 ? string1 : string2) || !size2)
4729 #define AT_STRINGS_END(d) ((d) == end2)
4730
4731
4732 /* Test if D points to a character which is word-constituent. We have
4733 two special cases to check for: if past the end of string1, look at
4734 the first character in string2; and if before the beginning of
4735 string2, look at the last character in string1. */
4736 #define WORDCHAR_P(d) \
4737 (SYNTAX ((d) == end1 ? *string2 \
4738 : (d) == string2 - 1 ? *(end1 - 1) : *(d)) \
4739 == Sword)
4740
4741 /* Disabled due to a compiler bug -- see comment at case wordbound */
4742
4743 /* The comment at case wordbound is following one, but we don't use
4744 AT_WORD_BOUNDARY anymore to support multibyte form.
4745
4746 The DEC Alpha C compiler 3.x generates incorrect code for the
4747 test WORDCHAR_P (d - 1) != WORDCHAR_P (d) in the expansion of
4748 AT_WORD_BOUNDARY, so this code is disabled. Expanding the
4749 macro and introducing temporary variables works around the bug. */
4750
4751 #if 0
4752 /* Test if the character before D and the one at D differ with respect
4753 to being word-constituent. */
4754 #define AT_WORD_BOUNDARY(d) \
4755 (AT_STRINGS_BEG (d) || AT_STRINGS_END (d) \
4756 || WORDCHAR_P (d - 1) != WORDCHAR_P (d))
4757 #endif
4758
4759 /* Free everything we malloc. */
4760 #ifdef MATCH_MAY_ALLOCATE
4761 # define FREE_VAR(var) if (var) { REGEX_FREE (var); var = NULL; } else
4762 # define FREE_VARIABLES() \
4763 do { \
4764 REGEX_FREE_STACK (fail_stack.stack); \
4765 FREE_VAR (regstart); \
4766 FREE_VAR (regend); \
4767 FREE_VAR (best_regstart); \
4768 FREE_VAR (best_regend); \
4769 } while (0)
4770 #else
4771 # define FREE_VARIABLES() ((void)0) /* Do nothing! But inhibit gcc warning. */
4772 #endif /* not MATCH_MAY_ALLOCATE */
4773
4774 \f
4775 /* Optimization routines. */
4776
4777 /* If the operation is a match against one or more chars,
4778 return a pointer to the next operation, else return NULL. */
4779 static re_char *
4780 skip_one_char (p)
4781 re_char *p;
4782 {
4783 switch (SWITCH_ENUM_CAST (*p++))
4784 {
4785 case anychar:
4786 break;
4787
4788 case exactn:
4789 p += *p + 1;
4790 break;
4791
4792 case charset_not:
4793 case charset:
4794 if (CHARSET_RANGE_TABLE_EXISTS_P (p - 1))
4795 {
4796 int mcnt;
4797 p = CHARSET_RANGE_TABLE (p - 1);
4798 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4799 p = CHARSET_RANGE_TABLE_END (p, mcnt);
4800 }
4801 else
4802 p += 1 + CHARSET_BITMAP_SIZE (p - 1);
4803 break;
4804
4805 case syntaxspec:
4806 case notsyntaxspec:
4807 #ifdef emacs
4808 case categoryspec:
4809 case notcategoryspec:
4810 #endif /* emacs */
4811 p++;
4812 break;
4813
4814 default:
4815 p = NULL;
4816 }
4817 return p;
4818 }
4819
4820
4821 /* Jump over non-matching operations. */
4822 static re_char *
4823 skip_noops (p, pend)
4824 re_char *p, *pend;
4825 {
4826 int mcnt;
4827 while (p < pend)
4828 {
4829 switch (SWITCH_ENUM_CAST ((re_opcode_t) *p))
4830 {
4831 case start_memory:
4832 case stop_memory:
4833 p += 2; break;
4834 case no_op:
4835 p += 1; break;
4836 case jump:
4837 p += 1;
4838 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4839 p += mcnt;
4840 break;
4841 default:
4842 return p;
4843 }
4844 }
4845 assert (p == pend);
4846 return p;
4847 }
4848
4849 /* Non-zero if "p1 matches something" implies "p2 fails". */
4850 static int
4851 mutually_exclusive_p (bufp, p1, p2)
4852 struct re_pattern_buffer *bufp;
4853 re_char *p1, *p2;
4854 {
4855 re_opcode_t op2;
4856 const boolean multibyte = RE_MULTIBYTE_P (bufp);
4857 unsigned char *pend = bufp->buffer + bufp->used;
4858
4859 assert (p1 >= bufp->buffer && p1 < pend
4860 && p2 >= bufp->buffer && p2 <= pend);
4861
4862 /* Skip over open/close-group commands.
4863 If what follows this loop is a ...+ construct,
4864 look at what begins its body, since we will have to
4865 match at least one of that. */
4866 p2 = skip_noops (p2, pend);
4867 /* The same skip can be done for p1, except that this function
4868 is only used in the case where p1 is a simple match operator. */
4869 /* p1 = skip_noops (p1, pend); */
4870
4871 assert (p1 >= bufp->buffer && p1 < pend
4872 && p2 >= bufp->buffer && p2 <= pend);
4873
4874 op2 = p2 == pend ? succeed : *p2;
4875
4876 switch (SWITCH_ENUM_CAST (op2))
4877 {
4878 case succeed:
4879 case endbuf:
4880 /* If we're at the end of the pattern, we can change. */
4881 if (skip_one_char (p1))
4882 {
4883 DEBUG_PRINT1 (" End of pattern: fast loop.\n");
4884 return 1;
4885 }
4886 break;
4887
4888 case endline:
4889 case exactn:
4890 {
4891 register re_wchar_t c
4892 = (re_opcode_t) *p2 == endline ? '\n'
4893 : RE_STRING_CHAR (p2 + 2, pend - p2 - 2, multibyte);
4894
4895 if ((re_opcode_t) *p1 == exactn)
4896 {
4897 if (c != RE_STRING_CHAR (p1 + 2, pend - p1 - 2, multibyte))
4898 {
4899 DEBUG_PRINT3 (" '%c' != '%c' => fast loop.\n", c, p1[2]);
4900 return 1;
4901 }
4902 }
4903
4904 else if ((re_opcode_t) *p1 == charset
4905 || (re_opcode_t) *p1 == charset_not)
4906 {
4907 int not = (re_opcode_t) *p1 == charset_not;
4908
4909 /* Test if C is listed in charset (or charset_not)
4910 at `p1'. */
4911 if (! multibyte || IS_REAL_ASCII (c))
4912 {
4913 if (c < CHARSET_BITMAP_SIZE (p1) * BYTEWIDTH
4914 && p1[2 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH)))
4915 not = !not;
4916 }
4917 else if (CHARSET_RANGE_TABLE_EXISTS_P (p1))
4918 CHARSET_LOOKUP_RANGE_TABLE (not, c, p1);
4919
4920 /* `not' is equal to 1 if c would match, which means
4921 that we can't change to pop_failure_jump. */
4922 if (!not)
4923 {
4924 DEBUG_PRINT1 (" No match => fast loop.\n");
4925 return 1;
4926 }
4927 }
4928 else if ((re_opcode_t) *p1 == anychar
4929 && c == '\n')
4930 {
4931 DEBUG_PRINT1 (" . != \\n => fast loop.\n");
4932 return 1;
4933 }
4934 }
4935 break;
4936
4937 case charset:
4938 {
4939 if ((re_opcode_t) *p1 == exactn)
4940 /* Reuse the code above. */
4941 return mutually_exclusive_p (bufp, p2, p1);
4942
4943 /* It is hard to list up all the character in charset
4944 P2 if it includes multibyte character. Give up in
4945 such case. */
4946 else if (!multibyte || !CHARSET_RANGE_TABLE_EXISTS_P (p2))
4947 {
4948 /* Now, we are sure that P2 has no range table.
4949 So, for the size of bitmap in P2, `p2[1]' is
4950 enough. But P1 may have range table, so the
4951 size of bitmap table of P1 is extracted by
4952 using macro `CHARSET_BITMAP_SIZE'.
4953
4954 In a multibyte case, we know that all the character
4955 listed in P2 is ASCII. In a unibyte case, P1 has only a
4956 bitmap table. So, in both cases, it is enough to test
4957 only the bitmap table of P1. */
4958
4959 if ((re_opcode_t) *p1 == charset)
4960 {
4961 int idx;
4962 /* We win if the charset inside the loop
4963 has no overlap with the one after the loop. */
4964 for (idx = 0;
4965 (idx < (int) p2[1]
4966 && idx < CHARSET_BITMAP_SIZE (p1));
4967 idx++)
4968 if ((p2[2 + idx] & p1[2 + idx]) != 0)
4969 break;
4970
4971 if (idx == p2[1]
4972 || idx == CHARSET_BITMAP_SIZE (p1))
4973 {
4974 DEBUG_PRINT1 (" No match => fast loop.\n");
4975 return 1;
4976 }
4977 }
4978 else if ((re_opcode_t) *p1 == charset_not)
4979 {
4980 int idx;
4981 /* We win if the charset_not inside the loop lists
4982 every character listed in the charset after. */
4983 for (idx = 0; idx < (int) p2[1]; idx++)
4984 if (! (p2[2 + idx] == 0
4985 || (idx < CHARSET_BITMAP_SIZE (p1)
4986 && ((p2[2 + idx] & ~ p1[2 + idx]) == 0))))
4987 break;
4988
4989 if (idx == p2[1])
4990 {
4991 DEBUG_PRINT1 (" No match => fast loop.\n");
4992 return 1;
4993 }
4994 }
4995 }
4996 }
4997 break;
4998
4999 case charset_not:
5000 switch (SWITCH_ENUM_CAST (*p1))
5001 {
5002 case exactn:
5003 case charset:
5004 /* Reuse the code above. */
5005 return mutually_exclusive_p (bufp, p2, p1);
5006 case charset_not:
5007 /* When we have two charset_not, it's very unlikely that
5008 they don't overlap. The union of the two sets of excluded
5009 chars should cover all possible chars, which, as a matter of
5010 fact, is virtually impossible in multibyte buffers. */
5011 break;
5012 }
5013 break;
5014
5015 case wordend:
5016 return ((re_opcode_t) *p1 == syntaxspec && p1[1] == Sword);
5017 case symend:
5018 return ((re_opcode_t) *p1 == syntaxspec
5019 && (p1[1] == Ssymbol || p1[1] == Sword));
5020 case notsyntaxspec:
5021 return ((re_opcode_t) *p1 == syntaxspec && p1[1] == p2[1]);
5022
5023 case wordbeg:
5024 return ((re_opcode_t) *p1 == notsyntaxspec && p1[1] == Sword);
5025 case symbeg:
5026 return ((re_opcode_t) *p1 == notsyntaxspec
5027 && (p1[1] == Ssymbol || p1[1] == Sword));
5028 case syntaxspec:
5029 return ((re_opcode_t) *p1 == notsyntaxspec && p1[1] == p2[1]);
5030
5031 case wordbound:
5032 return (((re_opcode_t) *p1 == notsyntaxspec
5033 || (re_opcode_t) *p1 == syntaxspec)
5034 && p1[1] == Sword);
5035
5036 #ifdef emacs
5037 case categoryspec:
5038 return ((re_opcode_t) *p1 == notcategoryspec && p1[1] == p2[1]);
5039 case notcategoryspec:
5040 return ((re_opcode_t) *p1 == categoryspec && p1[1] == p2[1]);
5041 #endif /* emacs */
5042
5043 default:
5044 ;
5045 }
5046
5047 /* Safe default. */
5048 return 0;
5049 }
5050
5051 \f
5052 /* Matching routines. */
5053
5054 #ifndef emacs /* Emacs never uses this. */
5055 /* re_match is like re_match_2 except it takes only a single string. */
5056
5057 int
5058 re_match (bufp, string, size, pos, regs)
5059 struct re_pattern_buffer *bufp;
5060 const char *string;
5061 int size, pos;
5062 struct re_registers *regs;
5063 {
5064 int result = re_match_2_internal (bufp, NULL, 0, (re_char*) string, size,
5065 pos, regs, size);
5066 return result;
5067 }
5068 WEAK_ALIAS (__re_match, re_match)
5069 #endif /* not emacs */
5070
5071 #ifdef emacs
5072 /* In Emacs, this is the string or buffer in which we
5073 are matching. It is used for looking up syntax properties. */
5074 Lisp_Object re_match_object;
5075 #endif
5076
5077 /* re_match_2 matches the compiled pattern in BUFP against the
5078 the (virtual) concatenation of STRING1 and STRING2 (of length SIZE1
5079 and SIZE2, respectively). We start matching at POS, and stop
5080 matching at STOP.
5081
5082 If REGS is non-null and the `no_sub' field of BUFP is nonzero, we
5083 store offsets for the substring each group matched in REGS. See the
5084 documentation for exactly how many groups we fill.
5085
5086 We return -1 if no match, -2 if an internal error (such as the
5087 failure stack overflowing). Otherwise, we return the length of the
5088 matched substring. */
5089
5090 int
5091 re_match_2 (bufp, string1, size1, string2, size2, pos, regs, stop)
5092 struct re_pattern_buffer *bufp;
5093 const char *string1, *string2;
5094 int size1, size2;
5095 int pos;
5096 struct re_registers *regs;
5097 int stop;
5098 {
5099 int result;
5100
5101 #ifdef emacs
5102 int charpos;
5103 gl_state.object = re_match_object;
5104 charpos = SYNTAX_TABLE_BYTE_TO_CHAR (POS_AS_IN_BUFFER (pos));
5105 SETUP_SYNTAX_TABLE_FOR_OBJECT (re_match_object, charpos, 1);
5106 #endif
5107
5108 result = re_match_2_internal (bufp, (re_char*) string1, size1,
5109 (re_char*) string2, size2,
5110 pos, regs, stop);
5111 return result;
5112 }
5113 WEAK_ALIAS (__re_match_2, re_match_2)
5114
5115
5116 /* This is a separate function so that we can force an alloca cleanup
5117 afterwards. */
5118 static int
5119 re_match_2_internal (bufp, string1, size1, string2, size2, pos, regs, stop)
5120 struct re_pattern_buffer *bufp;
5121 re_char *string1, *string2;
5122 int size1, size2;
5123 int pos;
5124 struct re_registers *regs;
5125 int stop;
5126 {
5127 /* General temporaries. */
5128 int mcnt;
5129 size_t reg;
5130 boolean not;
5131
5132 /* Just past the end of the corresponding string. */
5133 re_char *end1, *end2;
5134
5135 /* Pointers into string1 and string2, just past the last characters in
5136 each to consider matching. */
5137 re_char *end_match_1, *end_match_2;
5138
5139 /* Where we are in the data, and the end of the current string. */
5140 re_char *d, *dend;
5141
5142 /* Used sometimes to remember where we were before starting matching
5143 an operator so that we can go back in case of failure. This "atomic"
5144 behavior of matching opcodes is indispensable to the correctness
5145 of the on_failure_keep_string_jump optimization. */
5146 re_char *dfail;
5147
5148 /* Where we are in the pattern, and the end of the pattern. */
5149 re_char *p = bufp->buffer;
5150 re_char *pend = p + bufp->used;
5151
5152 /* We use this to map every character in the string. */
5153 RE_TRANSLATE_TYPE translate = bufp->translate;
5154
5155 /* Nonzero if BUFP is setup from a multibyte regex. */
5156 const boolean multibyte = RE_MULTIBYTE_P (bufp);
5157
5158 /* Nonzero if STRING1/STRING2 are multibyte. */
5159 const boolean target_multibyte = RE_TARGET_MULTIBYTE_P (bufp);
5160
5161 /* Failure point stack. Each place that can handle a failure further
5162 down the line pushes a failure point on this stack. It consists of
5163 regstart, and regend for all registers corresponding to
5164 the subexpressions we're currently inside, plus the number of such
5165 registers, and, finally, two char *'s. The first char * is where
5166 to resume scanning the pattern; the second one is where to resume
5167 scanning the strings. */
5168 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
5169 fail_stack_type fail_stack;
5170 #endif
5171 #ifdef DEBUG
5172 unsigned nfailure_points_pushed = 0, nfailure_points_popped = 0;
5173 #endif
5174
5175 #if defined REL_ALLOC && defined REGEX_MALLOC
5176 /* This holds the pointer to the failure stack, when
5177 it is allocated relocatably. */
5178 fail_stack_elt_t *failure_stack_ptr;
5179 #endif
5180
5181 /* We fill all the registers internally, independent of what we
5182 return, for use in backreferences. The number here includes
5183 an element for register zero. */
5184 size_t num_regs = bufp->re_nsub + 1;
5185
5186 /* Information on the contents of registers. These are pointers into
5187 the input strings; they record just what was matched (on this
5188 attempt) by a subexpression part of the pattern, that is, the
5189 regnum-th regstart pointer points to where in the pattern we began
5190 matching and the regnum-th regend points to right after where we
5191 stopped matching the regnum-th subexpression. (The zeroth register
5192 keeps track of what the whole pattern matches.) */
5193 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
5194 re_char **regstart, **regend;
5195 #endif
5196
5197 /* The following record the register info as found in the above
5198 variables when we find a match better than any we've seen before.
5199 This happens as we backtrack through the failure points, which in
5200 turn happens only if we have not yet matched the entire string. */
5201 unsigned best_regs_set = false;
5202 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
5203 re_char **best_regstart, **best_regend;
5204 #endif
5205
5206 /* Logically, this is `best_regend[0]'. But we don't want to have to
5207 allocate space for that if we're not allocating space for anything
5208 else (see below). Also, we never need info about register 0 for
5209 any of the other register vectors, and it seems rather a kludge to
5210 treat `best_regend' differently than the rest. So we keep track of
5211 the end of the best match so far in a separate variable. We
5212 initialize this to NULL so that when we backtrack the first time
5213 and need to test it, it's not garbage. */
5214 re_char *match_end = NULL;
5215
5216 #ifdef DEBUG
5217 /* Counts the total number of registers pushed. */
5218 unsigned num_regs_pushed = 0;
5219 #endif
5220
5221 DEBUG_PRINT1 ("\n\nEntering re_match_2.\n");
5222
5223 INIT_FAIL_STACK ();
5224
5225 #ifdef MATCH_MAY_ALLOCATE
5226 /* Do not bother to initialize all the register variables if there are
5227 no groups in the pattern, as it takes a fair amount of time. If
5228 there are groups, we include space for register 0 (the whole
5229 pattern), even though we never use it, since it simplifies the
5230 array indexing. We should fix this. */
5231 if (bufp->re_nsub)
5232 {
5233 regstart = REGEX_TALLOC (num_regs, re_char *);
5234 regend = REGEX_TALLOC (num_regs, re_char *);
5235 best_regstart = REGEX_TALLOC (num_regs, re_char *);
5236 best_regend = REGEX_TALLOC (num_regs, re_char *);
5237
5238 if (!(regstart && regend && best_regstart && best_regend))
5239 {
5240 FREE_VARIABLES ();
5241 return -2;
5242 }
5243 }
5244 else
5245 {
5246 /* We must initialize all our variables to NULL, so that
5247 `FREE_VARIABLES' doesn't try to free them. */
5248 regstart = regend = best_regstart = best_regend = NULL;
5249 }
5250 #endif /* MATCH_MAY_ALLOCATE */
5251
5252 /* The starting position is bogus. */
5253 if (pos < 0 || pos > size1 + size2)
5254 {
5255 FREE_VARIABLES ();
5256 return -1;
5257 }
5258
5259 /* Initialize subexpression text positions to -1 to mark ones that no
5260 start_memory/stop_memory has been seen for. Also initialize the
5261 register information struct. */
5262 for (reg = 1; reg < num_regs; reg++)
5263 regstart[reg] = regend[reg] = NULL;
5264
5265 /* We move `string1' into `string2' if the latter's empty -- but not if
5266 `string1' is null. */
5267 if (size2 == 0 && string1 != NULL)
5268 {
5269 string2 = string1;
5270 size2 = size1;
5271 string1 = 0;
5272 size1 = 0;
5273 }
5274 end1 = string1 + size1;
5275 end2 = string2 + size2;
5276
5277 /* `p' scans through the pattern as `d' scans through the data.
5278 `dend' is the end of the input string that `d' points within. `d'
5279 is advanced into the following input string whenever necessary, but
5280 this happens before fetching; therefore, at the beginning of the
5281 loop, `d' can be pointing at the end of a string, but it cannot
5282 equal `string2'. */
5283 if (pos >= size1)
5284 {
5285 /* Only match within string2. */
5286 d = string2 + pos - size1;
5287 dend = end_match_2 = string2 + stop - size1;
5288 end_match_1 = end1; /* Just to give it a value. */
5289 }
5290 else
5291 {
5292 if (stop < size1)
5293 {
5294 /* Only match within string1. */
5295 end_match_1 = string1 + stop;
5296 /* BEWARE!
5297 When we reach end_match_1, PREFETCH normally switches to string2.
5298 But in the present case, this means that just doing a PREFETCH
5299 makes us jump from `stop' to `gap' within the string.
5300 What we really want here is for the search to stop as
5301 soon as we hit end_match_1. That's why we set end_match_2
5302 to end_match_1 (since PREFETCH fails as soon as we hit
5303 end_match_2). */
5304 end_match_2 = end_match_1;
5305 }
5306 else
5307 { /* It's important to use this code when stop == size so that
5308 moving `d' from end1 to string2 will not prevent the d == dend
5309 check from catching the end of string. */
5310 end_match_1 = end1;
5311 end_match_2 = string2 + stop - size1;
5312 }
5313 d = string1 + pos;
5314 dend = end_match_1;
5315 }
5316
5317 DEBUG_PRINT1 ("The compiled pattern is: ");
5318 DEBUG_PRINT_COMPILED_PATTERN (bufp, p, pend);
5319 DEBUG_PRINT1 ("The string to match is: `");
5320 DEBUG_PRINT_DOUBLE_STRING (d, string1, size1, string2, size2);
5321 DEBUG_PRINT1 ("'\n");
5322
5323 /* This loops over pattern commands. It exits by returning from the
5324 function if the match is complete, or it drops through if the match
5325 fails at this starting point in the input data. */
5326 for (;;)
5327 {
5328 DEBUG_PRINT2 ("\n%p: ", p);
5329
5330 if (p == pend)
5331 { /* End of pattern means we might have succeeded. */
5332 DEBUG_PRINT1 ("end of pattern ... ");
5333
5334 /* If we haven't matched the entire string, and we want the
5335 longest match, try backtracking. */
5336 if (d != end_match_2)
5337 {
5338 /* 1 if this match ends in the same string (string1 or string2)
5339 as the best previous match. */
5340 boolean same_str_p = (FIRST_STRING_P (match_end)
5341 == FIRST_STRING_P (d));
5342 /* 1 if this match is the best seen so far. */
5343 boolean best_match_p;
5344
5345 /* AIX compiler got confused when this was combined
5346 with the previous declaration. */
5347 if (same_str_p)
5348 best_match_p = d > match_end;
5349 else
5350 best_match_p = !FIRST_STRING_P (d);
5351
5352 DEBUG_PRINT1 ("backtracking.\n");
5353
5354 if (!FAIL_STACK_EMPTY ())
5355 { /* More failure points to try. */
5356
5357 /* If exceeds best match so far, save it. */
5358 if (!best_regs_set || best_match_p)
5359 {
5360 best_regs_set = true;
5361 match_end = d;
5362
5363 DEBUG_PRINT1 ("\nSAVING match as best so far.\n");
5364
5365 for (reg = 1; reg < num_regs; reg++)
5366 {
5367 best_regstart[reg] = regstart[reg];
5368 best_regend[reg] = regend[reg];
5369 }
5370 }
5371 goto fail;
5372 }
5373
5374 /* If no failure points, don't restore garbage. And if
5375 last match is real best match, don't restore second
5376 best one. */
5377 else if (best_regs_set && !best_match_p)
5378 {
5379 restore_best_regs:
5380 /* Restore best match. It may happen that `dend ==
5381 end_match_1' while the restored d is in string2.
5382 For example, the pattern `x.*y.*z' against the
5383 strings `x-' and `y-z-', if the two strings are
5384 not consecutive in memory. */
5385 DEBUG_PRINT1 ("Restoring best registers.\n");
5386
5387 d = match_end;
5388 dend = ((d >= string1 && d <= end1)
5389 ? end_match_1 : end_match_2);
5390
5391 for (reg = 1; reg < num_regs; reg++)
5392 {
5393 regstart[reg] = best_regstart[reg];
5394 regend[reg] = best_regend[reg];
5395 }
5396 }
5397 } /* d != end_match_2 */
5398
5399 succeed_label:
5400 DEBUG_PRINT1 ("Accepting match.\n");
5401
5402 /* If caller wants register contents data back, do it. */
5403 if (regs && !bufp->no_sub)
5404 {
5405 /* Have the register data arrays been allocated? */
5406 if (bufp->regs_allocated == REGS_UNALLOCATED)
5407 { /* No. So allocate them with malloc. We need one
5408 extra element beyond `num_regs' for the `-1' marker
5409 GNU code uses. */
5410 regs->num_regs = MAX (RE_NREGS, num_regs + 1);
5411 regs->start = TALLOC (regs->num_regs, regoff_t);
5412 regs->end = TALLOC (regs->num_regs, regoff_t);
5413 if (regs->start == NULL || regs->end == NULL)
5414 {
5415 FREE_VARIABLES ();
5416 return -2;
5417 }
5418 bufp->regs_allocated = REGS_REALLOCATE;
5419 }
5420 else if (bufp->regs_allocated == REGS_REALLOCATE)
5421 { /* Yes. If we need more elements than were already
5422 allocated, reallocate them. If we need fewer, just
5423 leave it alone. */
5424 if (regs->num_regs < num_regs + 1)
5425 {
5426 regs->num_regs = num_regs + 1;
5427 RETALLOC (regs->start, regs->num_regs, regoff_t);
5428 RETALLOC (regs->end, regs->num_regs, regoff_t);
5429 if (regs->start == NULL || regs->end == NULL)
5430 {
5431 FREE_VARIABLES ();
5432 return -2;
5433 }
5434 }
5435 }
5436 else
5437 {
5438 /* These braces fend off a "empty body in an else-statement"
5439 warning under GCC when assert expands to nothing. */
5440 assert (bufp->regs_allocated == REGS_FIXED);
5441 }
5442
5443 /* Convert the pointer data in `regstart' and `regend' to
5444 indices. Register zero has to be set differently,
5445 since we haven't kept track of any info for it. */
5446 if (regs->num_regs > 0)
5447 {
5448 regs->start[0] = pos;
5449 regs->end[0] = POINTER_TO_OFFSET (d);
5450 }
5451
5452 /* Go through the first `min (num_regs, regs->num_regs)'
5453 registers, since that is all we initialized. */
5454 for (reg = 1; reg < MIN (num_regs, regs->num_regs); reg++)
5455 {
5456 if (REG_UNSET (regstart[reg]) || REG_UNSET (regend[reg]))
5457 regs->start[reg] = regs->end[reg] = -1;
5458 else
5459 {
5460 regs->start[reg]
5461 = (regoff_t) POINTER_TO_OFFSET (regstart[reg]);
5462 regs->end[reg]
5463 = (regoff_t) POINTER_TO_OFFSET (regend[reg]);
5464 }
5465 }
5466
5467 /* If the regs structure we return has more elements than
5468 were in the pattern, set the extra elements to -1. If
5469 we (re)allocated the registers, this is the case,
5470 because we always allocate enough to have at least one
5471 -1 at the end. */
5472 for (reg = num_regs; reg < regs->num_regs; reg++)
5473 regs->start[reg] = regs->end[reg] = -1;
5474 } /* regs && !bufp->no_sub */
5475
5476 DEBUG_PRINT4 ("%u failure points pushed, %u popped (%u remain).\n",
5477 nfailure_points_pushed, nfailure_points_popped,
5478 nfailure_points_pushed - nfailure_points_popped);
5479 DEBUG_PRINT2 ("%u registers pushed.\n", num_regs_pushed);
5480
5481 mcnt = POINTER_TO_OFFSET (d) - pos;
5482
5483 DEBUG_PRINT2 ("Returning %d from re_match_2.\n", mcnt);
5484
5485 FREE_VARIABLES ();
5486 return mcnt;
5487 }
5488
5489 /* Otherwise match next pattern command. */
5490 switch (SWITCH_ENUM_CAST ((re_opcode_t) *p++))
5491 {
5492 /* Ignore these. Used to ignore the n of succeed_n's which
5493 currently have n == 0. */
5494 case no_op:
5495 DEBUG_PRINT1 ("EXECUTING no_op.\n");
5496 break;
5497
5498 case succeed:
5499 DEBUG_PRINT1 ("EXECUTING succeed.\n");
5500 goto succeed_label;
5501
5502 /* Match the next n pattern characters exactly. The following
5503 byte in the pattern defines n, and the n bytes after that
5504 are the characters to match. */
5505 case exactn:
5506 mcnt = *p++;
5507 DEBUG_PRINT2 ("EXECUTING exactn %d.\n", mcnt);
5508
5509 /* Remember the start point to rollback upon failure. */
5510 dfail = d;
5511
5512 #ifndef emacs
5513 /* This is written out as an if-else so we don't waste time
5514 testing `translate' inside the loop. */
5515 if (RE_TRANSLATE_P (translate))
5516 do
5517 {
5518 PREFETCH ();
5519 if (RE_TRANSLATE (translate, *d) != *p++)
5520 {
5521 d = dfail;
5522 goto fail;
5523 }
5524 d++;
5525 }
5526 while (--mcnt);
5527 else
5528 do
5529 {
5530 PREFETCH ();
5531 if (*d++ != *p++)
5532 {
5533 d = dfail;
5534 goto fail;
5535 }
5536 }
5537 while (--mcnt);
5538 #else /* emacs */
5539 /* The cost of testing `translate' is comparatively small. */
5540 if (target_multibyte)
5541 do
5542 {
5543 int pat_charlen, buf_charlen;
5544 int pat_ch, buf_ch;
5545
5546 PREFETCH ();
5547 if (multibyte)
5548 pat_ch = STRING_CHAR_AND_LENGTH (p, pend - p, pat_charlen);
5549 else
5550 {
5551 pat_ch = RE_CHAR_TO_MULTIBYTE (*p);
5552 pat_charlen = 1;
5553 }
5554 buf_ch = STRING_CHAR_AND_LENGTH (d, dend - d, buf_charlen);
5555
5556 if (TRANSLATE (buf_ch) != pat_ch)
5557 {
5558 d = dfail;
5559 goto fail;
5560 }
5561
5562 p += pat_charlen;
5563 d += buf_charlen;
5564 mcnt -= pat_charlen;
5565 }
5566 while (mcnt > 0);
5567 else
5568 do
5569 {
5570 int pat_charlen, buf_charlen;
5571 int pat_ch, buf_ch;
5572
5573 PREFETCH ();
5574 if (multibyte)
5575 {
5576 pat_ch = STRING_CHAR_AND_LENGTH (p, pend - p, pat_charlen);
5577 if (CHAR_BYTE8_P (pat_ch))
5578 pat_ch = CHAR_TO_BYTE8 (pat_ch);
5579 else
5580 pat_ch = RE_CHAR_TO_UNIBYTE (pat_ch);
5581 }
5582 else
5583 {
5584 pat_ch = *p;
5585 pat_charlen = 1;
5586 }
5587 buf_ch = RE_CHAR_TO_MULTIBYTE (*d);
5588 if (! CHAR_BYTE8_P (buf_ch))
5589 {
5590 buf_ch = TRANSLATE (buf_ch);
5591 buf_ch = RE_CHAR_TO_UNIBYTE (buf_ch);
5592 if (buf_ch < 0)
5593 buf_ch = *d;
5594 }
5595 else
5596 buf_ch = *d;
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 /* Whether matching against a unibyte character. */
5654 boolean unibyte_char = false;
5655
5656 DEBUG_PRINT2 ("EXECUTING charset%s.\n", not ? "_not" : "");
5657
5658 range_table_exists = CHARSET_RANGE_TABLE_EXISTS_P (&p[-1]);
5659
5660 if (range_table_exists)
5661 {
5662 range_table = CHARSET_RANGE_TABLE (&p[-1]); /* Past the bitmap. */
5663 EXTRACT_NUMBER_AND_INCR (count, range_table);
5664 }
5665
5666 PREFETCH ();
5667 c = RE_STRING_CHAR_AND_LENGTH (d, dend - d, len, target_multibyte);
5668 if (target_multibyte)
5669 {
5670 int c1;
5671
5672 c = TRANSLATE (c);
5673 c1 = RE_CHAR_TO_UNIBYTE (c);
5674 if (c1 >= 0)
5675 {
5676 unibyte_char = true;
5677 c = c1;
5678 }
5679 }
5680 else
5681 {
5682 int c1 = RE_CHAR_TO_MULTIBYTE (c);
5683
5684 if (! CHAR_BYTE8_P (c1))
5685 {
5686 c1 = TRANSLATE (c1);
5687 c1 = RE_CHAR_TO_UNIBYTE (c1);
5688 if (c1 >= 0)
5689 {
5690 unibyte_char = true;
5691 c = c1;
5692 }
5693 }
5694 else
5695 unibyte_char = true;
5696 }
5697
5698 if (unibyte_char && c < (1 << BYTEWIDTH))
5699 { /* Lookup bitmap. */
5700 /* Cast to `unsigned' instead of `unsigned char' in
5701 case the bit list is a full 32 bytes long. */
5702 if (c < (unsigned) (CHARSET_BITMAP_SIZE (&p[-1]) * BYTEWIDTH)
5703 && p[1 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH)))
5704 not = !not;
5705 }
5706 #ifdef emacs
5707 else if (range_table_exists)
5708 {
5709 int class_bits = CHARSET_RANGE_TABLE_BITS (&p[-1]);
5710
5711 if ( (class_bits & BIT_LOWER && ISLOWER (c))
5712 | (class_bits & BIT_MULTIBYTE)
5713 | (class_bits & BIT_PUNCT && ISPUNCT (c))
5714 | (class_bits & BIT_SPACE && ISSPACE (c))
5715 | (class_bits & BIT_UPPER && ISUPPER (c))
5716 | (class_bits & BIT_WORD && ISWORD (c)))
5717 not = !not;
5718 else
5719 CHARSET_LOOKUP_RANGE_TABLE_RAW (not, c, range_table, count);
5720 }
5721 #endif /* emacs */
5722
5723 if (range_table_exists)
5724 p = CHARSET_RANGE_TABLE_END (range_table, count);
5725 else
5726 p += CHARSET_BITMAP_SIZE (&p[-1]) + 1;
5727
5728 if (!not) goto fail;
5729
5730 d += len;
5731 break;
5732 }
5733
5734
5735 /* The beginning of a group is represented by start_memory.
5736 The argument is the register number. The text
5737 matched within the group is recorded (in the internal
5738 registers data structure) under the register number. */
5739 case start_memory:
5740 DEBUG_PRINT2 ("EXECUTING start_memory %d:\n", *p);
5741
5742 /* In case we need to undo this operation (via backtracking). */
5743 PUSH_FAILURE_REG ((unsigned int)*p);
5744
5745 regstart[*p] = d;
5746 regend[*p] = NULL; /* probably unnecessary. -sm */
5747 DEBUG_PRINT2 (" regstart: %d\n", POINTER_TO_OFFSET (regstart[*p]));
5748
5749 /* Move past the register number and inner group count. */
5750 p += 1;
5751 break;
5752
5753
5754 /* The stop_memory opcode represents the end of a group. Its
5755 argument is the same as start_memory's: the register number. */
5756 case stop_memory:
5757 DEBUG_PRINT2 ("EXECUTING stop_memory %d:\n", *p);
5758
5759 assert (!REG_UNSET (regstart[*p]));
5760 /* Strictly speaking, there should be code such as:
5761
5762 assert (REG_UNSET (regend[*p]));
5763 PUSH_FAILURE_REGSTOP ((unsigned int)*p);
5764
5765 But the only info to be pushed is regend[*p] and it is known to
5766 be UNSET, so there really isn't anything to push.
5767 Not pushing anything, on the other hand deprives us from the
5768 guarantee that regend[*p] is UNSET since undoing this operation
5769 will not reset its value properly. This is not important since
5770 the value will only be read on the next start_memory or at
5771 the very end and both events can only happen if this stop_memory
5772 is *not* undone. */
5773
5774 regend[*p] = d;
5775 DEBUG_PRINT2 (" regend: %d\n", POINTER_TO_OFFSET (regend[*p]));
5776
5777 /* Move past the register number and the inner group count. */
5778 p += 1;
5779 break;
5780
5781
5782 /* \<digit> has been turned into a `duplicate' command which is
5783 followed by the numeric value of <digit> as the register number. */
5784 case duplicate:
5785 {
5786 register re_char *d2, *dend2;
5787 int regno = *p++; /* Get which register to match against. */
5788 DEBUG_PRINT2 ("EXECUTING duplicate %d.\n", regno);
5789
5790 /* Can't back reference a group which we've never matched. */
5791 if (REG_UNSET (regstart[regno]) || REG_UNSET (regend[regno]))
5792 goto fail;
5793
5794 /* Where in input to try to start matching. */
5795 d2 = regstart[regno];
5796
5797 /* Remember the start point to rollback upon failure. */
5798 dfail = d;
5799
5800 /* Where to stop matching; if both the place to start and
5801 the place to stop matching are in the same string, then
5802 set to the place to stop, otherwise, for now have to use
5803 the end of the first string. */
5804
5805 dend2 = ((FIRST_STRING_P (regstart[regno])
5806 == FIRST_STRING_P (regend[regno]))
5807 ? regend[regno] : end_match_1);
5808 for (;;)
5809 {
5810 /* If necessary, advance to next segment in register
5811 contents. */
5812 while (d2 == dend2)
5813 {
5814 if (dend2 == end_match_2) break;
5815 if (dend2 == regend[regno]) break;
5816
5817 /* End of string1 => advance to string2. */
5818 d2 = string2;
5819 dend2 = regend[regno];
5820 }
5821 /* At end of register contents => success */
5822 if (d2 == dend2) break;
5823
5824 /* If necessary, advance to next segment in data. */
5825 PREFETCH ();
5826
5827 /* How many characters left in this segment to match. */
5828 mcnt = dend - d;
5829
5830 /* Want how many consecutive characters we can match in
5831 one shot, so, if necessary, adjust the count. */
5832 if (mcnt > dend2 - d2)
5833 mcnt = dend2 - d2;
5834
5835 /* Compare that many; failure if mismatch, else move
5836 past them. */
5837 if (RE_TRANSLATE_P (translate)
5838 ? bcmp_translate (d, d2, mcnt, translate, target_multibyte)
5839 : memcmp (d, d2, mcnt))
5840 {
5841 d = dfail;
5842 goto fail;
5843 }
5844 d += mcnt, d2 += mcnt;
5845 }
5846 }
5847 break;
5848
5849
5850 /* begline matches the empty string at the beginning of the string
5851 (unless `not_bol' is set in `bufp'), and after newlines. */
5852 case begline:
5853 DEBUG_PRINT1 ("EXECUTING begline.\n");
5854
5855 if (AT_STRINGS_BEG (d))
5856 {
5857 if (!bufp->not_bol) break;
5858 }
5859 else
5860 {
5861 unsigned c;
5862 GET_CHAR_BEFORE_2 (c, d, string1, end1, string2, end2);
5863 if (c == '\n')
5864 break;
5865 }
5866 /* In all other cases, we fail. */
5867 goto fail;
5868
5869
5870 /* endline is the dual of begline. */
5871 case endline:
5872 DEBUG_PRINT1 ("EXECUTING endline.\n");
5873
5874 if (AT_STRINGS_END (d))
5875 {
5876 if (!bufp->not_eol) break;
5877 }
5878 else
5879 {
5880 PREFETCH_NOLIMIT ();
5881 if (*d == '\n')
5882 break;
5883 }
5884 goto fail;
5885
5886
5887 /* Match at the very beginning of the data. */
5888 case begbuf:
5889 DEBUG_PRINT1 ("EXECUTING begbuf.\n");
5890 if (AT_STRINGS_BEG (d))
5891 break;
5892 goto fail;
5893
5894
5895 /* Match at the very end of the data. */
5896 case endbuf:
5897 DEBUG_PRINT1 ("EXECUTING endbuf.\n");
5898 if (AT_STRINGS_END (d))
5899 break;
5900 goto fail;
5901
5902
5903 /* on_failure_keep_string_jump is used to optimize `.*\n'. It
5904 pushes NULL as the value for the string on the stack. Then
5905 `POP_FAILURE_POINT' will keep the current value for the
5906 string, instead of restoring it. To see why, consider
5907 matching `foo\nbar' against `.*\n'. The .* matches the foo;
5908 then the . fails against the \n. But the next thing we want
5909 to do is match the \n against the \n; if we restored the
5910 string value, we would be back at the foo.
5911
5912 Because this is used only in specific cases, we don't need to
5913 check all the things that `on_failure_jump' does, to make
5914 sure the right things get saved on the stack. Hence we don't
5915 share its code. The only reason to push anything on the
5916 stack at all is that otherwise we would have to change
5917 `anychar's code to do something besides goto fail in this
5918 case; that seems worse than this. */
5919 case on_failure_keep_string_jump:
5920 EXTRACT_NUMBER_AND_INCR (mcnt, p);
5921 DEBUG_PRINT3 ("EXECUTING on_failure_keep_string_jump %d (to %p):\n",
5922 mcnt, p + mcnt);
5923
5924 PUSH_FAILURE_POINT (p - 3, NULL);
5925 break;
5926
5927 /* A nasty loop is introduced by the non-greedy *? and +?.
5928 With such loops, the stack only ever contains one failure point
5929 at a time, so that a plain on_failure_jump_loop kind of
5930 cycle detection cannot work. Worse yet, such a detection
5931 can not only fail to detect a cycle, but it can also wrongly
5932 detect a cycle (between different instantiations of the same
5933 loop).
5934 So the method used for those nasty loops is a little different:
5935 We use a special cycle-detection-stack-frame which is pushed
5936 when the on_failure_jump_nastyloop failure-point is *popped*.
5937 This special frame thus marks the beginning of one iteration
5938 through the loop and we can hence easily check right here
5939 whether something matched between the beginning and the end of
5940 the loop. */
5941 case on_failure_jump_nastyloop:
5942 EXTRACT_NUMBER_AND_INCR (mcnt, p);
5943 DEBUG_PRINT3 ("EXECUTING on_failure_jump_nastyloop %d (to %p):\n",
5944 mcnt, p + mcnt);
5945
5946 assert ((re_opcode_t)p[-4] == no_op);
5947 {
5948 int cycle = 0;
5949 CHECK_INFINITE_LOOP (p - 4, d);
5950 if (!cycle)
5951 /* If there's a cycle, just continue without pushing
5952 this failure point. The failure point is the "try again"
5953 option, which shouldn't be tried.
5954 We want (x?)*?y\1z to match both xxyz and xxyxz. */
5955 PUSH_FAILURE_POINT (p - 3, d);
5956 }
5957 break;
5958
5959 /* Simple loop detecting on_failure_jump: just check on the
5960 failure stack if the same spot was already hit earlier. */
5961 case on_failure_jump_loop:
5962 on_failure:
5963 EXTRACT_NUMBER_AND_INCR (mcnt, p);
5964 DEBUG_PRINT3 ("EXECUTING on_failure_jump_loop %d (to %p):\n",
5965 mcnt, p + mcnt);
5966 {
5967 int cycle = 0;
5968 CHECK_INFINITE_LOOP (p - 3, d);
5969 if (cycle)
5970 /* If there's a cycle, get out of the loop, as if the matching
5971 had failed. We used to just `goto fail' here, but that was
5972 aborting the search a bit too early: we want to keep the
5973 empty-loop-match and keep matching after the loop.
5974 We want (x?)*y\1z to match both xxyz and xxyxz. */
5975 p += mcnt;
5976 else
5977 PUSH_FAILURE_POINT (p - 3, d);
5978 }
5979 break;
5980
5981
5982 /* Uses of on_failure_jump:
5983
5984 Each alternative starts with an on_failure_jump that points
5985 to the beginning of the next alternative. Each alternative
5986 except the last ends with a jump that in effect jumps past
5987 the rest of the alternatives. (They really jump to the
5988 ending jump of the following alternative, because tensioning
5989 these jumps is a hassle.)
5990
5991 Repeats start with an on_failure_jump that points past both
5992 the repetition text and either the following jump or
5993 pop_failure_jump back to this on_failure_jump. */
5994 case on_failure_jump:
5995 EXTRACT_NUMBER_AND_INCR (mcnt, p);
5996 DEBUG_PRINT3 ("EXECUTING on_failure_jump %d (to %p):\n",
5997 mcnt, p + mcnt);
5998
5999 PUSH_FAILURE_POINT (p -3, d);
6000 break;
6001
6002 /* This operation is used for greedy *.
6003 Compare the beginning of the repeat with what in the
6004 pattern follows its end. If we can establish that there
6005 is nothing that they would both match, i.e., that we
6006 would have to backtrack because of (as in, e.g., `a*a')
6007 then we can use a non-backtracking loop based on
6008 on_failure_keep_string_jump instead of on_failure_jump. */
6009 case on_failure_jump_smart:
6010 EXTRACT_NUMBER_AND_INCR (mcnt, p);
6011 DEBUG_PRINT3 ("EXECUTING on_failure_jump_smart %d (to %p).\n",
6012 mcnt, p + mcnt);
6013 {
6014 re_char *p1 = p; /* Next operation. */
6015 /* Here, we discard `const', making re_match non-reentrant. */
6016 unsigned char *p2 = (unsigned char*) p + mcnt; /* Jump dest. */
6017 unsigned char *p3 = (unsigned char*) p - 3; /* opcode location. */
6018
6019 p -= 3; /* Reset so that we will re-execute the
6020 instruction once it's been changed. */
6021
6022 EXTRACT_NUMBER (mcnt, p2 - 2);
6023
6024 /* Ensure this is a indeed the trivial kind of loop
6025 we are expecting. */
6026 assert (skip_one_char (p1) == p2 - 3);
6027 assert ((re_opcode_t) p2[-3] == jump && p2 + mcnt == p);
6028 DEBUG_STATEMENT (debug += 2);
6029 if (mutually_exclusive_p (bufp, p1, p2))
6030 {
6031 /* Use a fast `on_failure_keep_string_jump' loop. */
6032 DEBUG_PRINT1 (" smart exclusive => fast loop.\n");
6033 *p3 = (unsigned char) on_failure_keep_string_jump;
6034 STORE_NUMBER (p2 - 2, mcnt + 3);
6035 }
6036 else
6037 {
6038 /* Default to a safe `on_failure_jump' loop. */
6039 DEBUG_PRINT1 (" smart default => slow loop.\n");
6040 *p3 = (unsigned char) on_failure_jump;
6041 }
6042 DEBUG_STATEMENT (debug -= 2);
6043 }
6044 break;
6045
6046 /* Unconditionally jump (without popping any failure points). */
6047 case jump:
6048 unconditional_jump:
6049 IMMEDIATE_QUIT_CHECK;
6050 EXTRACT_NUMBER_AND_INCR (mcnt, p); /* Get the amount to jump. */
6051 DEBUG_PRINT2 ("EXECUTING jump %d ", mcnt);
6052 p += mcnt; /* Do the jump. */
6053 DEBUG_PRINT2 ("(to %p).\n", p);
6054 break;
6055
6056
6057 /* Have to succeed matching what follows at least n times.
6058 After that, handle like `on_failure_jump'. */
6059 case succeed_n:
6060 /* Signedness doesn't matter since we only compare MCNT to 0. */
6061 EXTRACT_NUMBER (mcnt, p + 2);
6062 DEBUG_PRINT2 ("EXECUTING succeed_n %d.\n", mcnt);
6063
6064 /* Originally, mcnt is how many times we HAVE to succeed. */
6065 if (mcnt != 0)
6066 {
6067 /* Here, we discard `const', making re_match non-reentrant. */
6068 unsigned char *p2 = (unsigned char*) p + 2; /* counter loc. */
6069 mcnt--;
6070 p += 4;
6071 PUSH_NUMBER (p2, mcnt);
6072 }
6073 else
6074 /* The two bytes encoding mcnt == 0 are two no_op opcodes. */
6075 goto on_failure;
6076 break;
6077
6078 case jump_n:
6079 /* Signedness doesn't matter since we only compare MCNT to 0. */
6080 EXTRACT_NUMBER (mcnt, p + 2);
6081 DEBUG_PRINT2 ("EXECUTING jump_n %d.\n", mcnt);
6082
6083 /* Originally, this is how many times we CAN jump. */
6084 if (mcnt != 0)
6085 {
6086 /* Here, we discard `const', making re_match non-reentrant. */
6087 unsigned char *p2 = (unsigned char*) p + 2; /* counter loc. */
6088 mcnt--;
6089 PUSH_NUMBER (p2, mcnt);
6090 goto unconditional_jump;
6091 }
6092 /* If don't have to jump any more, skip over the rest of command. */
6093 else
6094 p += 4;
6095 break;
6096
6097 case set_number_at:
6098 {
6099 unsigned char *p2; /* Location of the counter. */
6100 DEBUG_PRINT1 ("EXECUTING set_number_at.\n");
6101
6102 EXTRACT_NUMBER_AND_INCR (mcnt, p);
6103 /* Here, we discard `const', making re_match non-reentrant. */
6104 p2 = (unsigned char*) p + mcnt;
6105 /* Signedness doesn't matter since we only copy MCNT's bits . */
6106 EXTRACT_NUMBER_AND_INCR (mcnt, p);
6107 DEBUG_PRINT3 (" Setting %p to %d.\n", p2, mcnt);
6108 PUSH_NUMBER (p2, mcnt);
6109 break;
6110 }
6111
6112 case wordbound:
6113 case notwordbound:
6114 not = (re_opcode_t) *(p - 1) == notwordbound;
6115 DEBUG_PRINT2 ("EXECUTING %swordbound.\n", not?"not":"");
6116
6117 /* We SUCCEED (or FAIL) in one of the following cases: */
6118
6119 /* Case 1: D is at the beginning or the end of string. */
6120 if (AT_STRINGS_BEG (d) || AT_STRINGS_END (d))
6121 not = !not;
6122 else
6123 {
6124 /* C1 is the character before D, S1 is the syntax of C1, C2
6125 is the character at D, and S2 is the syntax of C2. */
6126 re_wchar_t c1, c2;
6127 int s1, s2;
6128 int dummy;
6129 #ifdef emacs
6130 int offset = PTR_TO_OFFSET (d - 1);
6131 int charpos = SYNTAX_TABLE_BYTE_TO_CHAR (offset);
6132 UPDATE_SYNTAX_TABLE (charpos);
6133 #endif
6134 GET_CHAR_BEFORE_2 (c1, d, string1, end1, string2, end2);
6135 s1 = SYNTAX (c1);
6136 #ifdef emacs
6137 UPDATE_SYNTAX_TABLE_FORWARD (charpos + 1);
6138 #endif
6139 PREFETCH_NOLIMIT ();
6140 GET_CHAR_AFTER (c2, d, dummy);
6141 s2 = SYNTAX (c2);
6142
6143 if (/* Case 2: Only one of S1 and S2 is Sword. */
6144 ((s1 == Sword) != (s2 == Sword))
6145 /* Case 3: Both of S1 and S2 are Sword, and macro
6146 WORD_BOUNDARY_P (C1, C2) returns nonzero. */
6147 || ((s1 == Sword) && WORD_BOUNDARY_P (c1, c2)))
6148 not = !not;
6149 }
6150 if (not)
6151 break;
6152 else
6153 goto fail;
6154
6155 case wordbeg:
6156 DEBUG_PRINT1 ("EXECUTING wordbeg.\n");
6157
6158 /* We FAIL in one of the following cases: */
6159
6160 /* Case 1: D is at the end of string. */
6161 if (AT_STRINGS_END (d))
6162 goto fail;
6163 else
6164 {
6165 /* C1 is the character before D, S1 is the syntax of C1, C2
6166 is the character at D, and S2 is the syntax of C2. */
6167 re_wchar_t c1, c2;
6168 int s1, s2;
6169 int dummy;
6170 #ifdef emacs
6171 int offset = PTR_TO_OFFSET (d);
6172 int charpos = SYNTAX_TABLE_BYTE_TO_CHAR (offset);
6173 UPDATE_SYNTAX_TABLE (charpos);
6174 #endif
6175 PREFETCH ();
6176 GET_CHAR_AFTER (c2, d, dummy);
6177 s2 = SYNTAX (c2);
6178
6179 /* Case 2: S2 is not Sword. */
6180 if (s2 != Sword)
6181 goto fail;
6182
6183 /* Case 3: D is not at the beginning of string ... */
6184 if (!AT_STRINGS_BEG (d))
6185 {
6186 GET_CHAR_BEFORE_2 (c1, d, string1, end1, string2, end2);
6187 #ifdef emacs
6188 UPDATE_SYNTAX_TABLE_BACKWARD (charpos - 1);
6189 #endif
6190 s1 = SYNTAX (c1);
6191
6192 /* ... and S1 is Sword, and WORD_BOUNDARY_P (C1, C2)
6193 returns 0. */
6194 if ((s1 == Sword) && !WORD_BOUNDARY_P (c1, c2))
6195 goto fail;
6196 }
6197 }
6198 break;
6199
6200 case wordend:
6201 DEBUG_PRINT1 ("EXECUTING wordend.\n");
6202
6203 /* We FAIL in one of the following cases: */
6204
6205 /* Case 1: D is at the beginning of string. */
6206 if (AT_STRINGS_BEG (d))
6207 goto fail;
6208 else
6209 {
6210 /* C1 is the character before D, S1 is the syntax of C1, C2
6211 is the character at D, and S2 is the syntax of C2. */
6212 re_wchar_t c1, c2;
6213 int s1, s2;
6214 int dummy;
6215 #ifdef emacs
6216 int offset = PTR_TO_OFFSET (d) - 1;
6217 int charpos = SYNTAX_TABLE_BYTE_TO_CHAR (offset);
6218 UPDATE_SYNTAX_TABLE (charpos);
6219 #endif
6220 GET_CHAR_BEFORE_2 (c1, d, string1, end1, string2, end2);
6221 s1 = SYNTAX (c1);
6222
6223 /* Case 2: S1 is not Sword. */
6224 if (s1 != Sword)
6225 goto fail;
6226
6227 /* Case 3: D is not at the end of string ... */
6228 if (!AT_STRINGS_END (d))
6229 {
6230 PREFETCH_NOLIMIT ();
6231 GET_CHAR_AFTER (c2, d, dummy);
6232 #ifdef emacs
6233 UPDATE_SYNTAX_TABLE_FORWARD (charpos);
6234 #endif
6235 s2 = SYNTAX (c2);
6236
6237 /* ... and S2 is Sword, and WORD_BOUNDARY_P (C1, C2)
6238 returns 0. */
6239 if ((s2 == Sword) && !WORD_BOUNDARY_P (c1, c2))
6240 goto fail;
6241 }
6242 }
6243 break;
6244
6245 case symbeg:
6246 DEBUG_PRINT1 ("EXECUTING symbeg.\n");
6247
6248 /* We FAIL in one of the following cases: */
6249
6250 /* Case 1: D is at the end of string. */
6251 if (AT_STRINGS_END (d))
6252 goto fail;
6253 else
6254 {
6255 /* C1 is the character before D, S1 is the syntax of C1, C2
6256 is the character at D, and S2 is the syntax of C2. */
6257 re_wchar_t c1, c2;
6258 int s1, s2;
6259 #ifdef emacs
6260 int offset = PTR_TO_OFFSET (d);
6261 int charpos = SYNTAX_TABLE_BYTE_TO_CHAR (offset);
6262 UPDATE_SYNTAX_TABLE (charpos);
6263 #endif
6264 PREFETCH ();
6265 c2 = RE_STRING_CHAR (d, dend - d, target_multibyte);
6266 s2 = SYNTAX (c2);
6267
6268 /* Case 2: S2 is neither Sword nor Ssymbol. */
6269 if (s2 != Sword && s2 != Ssymbol)
6270 goto fail;
6271
6272 /* Case 3: D is not at the beginning of string ... */
6273 if (!AT_STRINGS_BEG (d))
6274 {
6275 GET_CHAR_BEFORE_2 (c1, d, string1, end1, string2, end2);
6276 #ifdef emacs
6277 UPDATE_SYNTAX_TABLE_BACKWARD (charpos - 1);
6278 #endif
6279 s1 = SYNTAX (c1);
6280
6281 /* ... and S1 is Sword or Ssymbol. */
6282 if (s1 == Sword || s1 == Ssymbol)
6283 goto fail;
6284 }
6285 }
6286 break;
6287
6288 case symend:
6289 DEBUG_PRINT1 ("EXECUTING symend.\n");
6290
6291 /* We FAIL in one of the following cases: */
6292
6293 /* Case 1: D is at the beginning of string. */
6294 if (AT_STRINGS_BEG (d))
6295 goto fail;
6296 else
6297 {
6298 /* C1 is the character before D, S1 is the syntax of C1, C2
6299 is the character at D, and S2 is the syntax of C2. */
6300 re_wchar_t c1, c2;
6301 int s1, s2;
6302 #ifdef emacs
6303 int offset = PTR_TO_OFFSET (d) - 1;
6304 int charpos = SYNTAX_TABLE_BYTE_TO_CHAR (offset);
6305 UPDATE_SYNTAX_TABLE (charpos);
6306 #endif
6307 GET_CHAR_BEFORE_2 (c1, d, string1, end1, string2, end2);
6308 s1 = SYNTAX (c1);
6309
6310 /* Case 2: S1 is neither Ssymbol nor Sword. */
6311 if (s1 != Sword && s1 != Ssymbol)
6312 goto fail;
6313
6314 /* Case 3: D is not at the end of string ... */
6315 if (!AT_STRINGS_END (d))
6316 {
6317 PREFETCH_NOLIMIT ();
6318 c2 = RE_STRING_CHAR (d, dend - d, target_multibyte);
6319 #ifdef emacs
6320 UPDATE_SYNTAX_TABLE_FORWARD (charpos + 1);
6321 #endif
6322 s2 = SYNTAX (c2);
6323
6324 /* ... and S2 is Sword or Ssymbol. */
6325 if (s2 == Sword || s2 == Ssymbol)
6326 goto fail;
6327 }
6328 }
6329 break;
6330
6331 case syntaxspec:
6332 case notsyntaxspec:
6333 not = (re_opcode_t) *(p - 1) == notsyntaxspec;
6334 mcnt = *p++;
6335 DEBUG_PRINT3 ("EXECUTING %ssyntaxspec %d.\n", not?"not":"", mcnt);
6336 PREFETCH ();
6337 #ifdef emacs
6338 {
6339 int offset = PTR_TO_OFFSET (d);
6340 int pos1 = SYNTAX_TABLE_BYTE_TO_CHAR (offset);
6341 UPDATE_SYNTAX_TABLE (pos1);
6342 }
6343 #endif
6344 {
6345 int len;
6346 re_wchar_t c;
6347
6348 GET_CHAR_AFTER (c, d, len);
6349 if ((SYNTAX (c) != (enum syntaxcode) mcnt) ^ not)
6350 goto fail;
6351 d += len;
6352 }
6353 break;
6354
6355 #ifdef emacs
6356 case before_dot:
6357 DEBUG_PRINT1 ("EXECUTING before_dot.\n");
6358 if (PTR_BYTE_POS (d) >= PT_BYTE)
6359 goto fail;
6360 break;
6361
6362 case at_dot:
6363 DEBUG_PRINT1 ("EXECUTING at_dot.\n");
6364 if (PTR_BYTE_POS (d) != PT_BYTE)
6365 goto fail;
6366 break;
6367
6368 case after_dot:
6369 DEBUG_PRINT1 ("EXECUTING after_dot.\n");
6370 if (PTR_BYTE_POS (d) <= PT_BYTE)
6371 goto fail;
6372 break;
6373
6374 case categoryspec:
6375 case notcategoryspec:
6376 not = (re_opcode_t) *(p - 1) == notcategoryspec;
6377 mcnt = *p++;
6378 DEBUG_PRINT3 ("EXECUTING %scategoryspec %d.\n", not?"not":"", mcnt);
6379 PREFETCH ();
6380 {
6381 int len;
6382 re_wchar_t c;
6383
6384 GET_CHAR_AFTER (c, d, len);
6385 if ((!CHAR_HAS_CATEGORY (c, mcnt)) ^ not)
6386 goto fail;
6387 d += len;
6388 }
6389 break;
6390
6391 #endif /* emacs */
6392
6393 default:
6394 abort ();
6395 }
6396 continue; /* Successfully executed one pattern command; keep going. */
6397
6398
6399 /* We goto here if a matching operation fails. */
6400 fail:
6401 IMMEDIATE_QUIT_CHECK;
6402 if (!FAIL_STACK_EMPTY ())
6403 {
6404 re_char *str, *pat;
6405 /* A restart point is known. Restore to that state. */
6406 DEBUG_PRINT1 ("\nFAIL:\n");
6407 POP_FAILURE_POINT (str, pat);
6408 switch (SWITCH_ENUM_CAST ((re_opcode_t) *pat++))
6409 {
6410 case on_failure_keep_string_jump:
6411 assert (str == NULL);
6412 goto continue_failure_jump;
6413
6414 case on_failure_jump_nastyloop:
6415 assert ((re_opcode_t)pat[-2] == no_op);
6416 PUSH_FAILURE_POINT (pat - 2, str);
6417 /* Fallthrough */
6418
6419 case on_failure_jump_loop:
6420 case on_failure_jump:
6421 case succeed_n:
6422 d = str;
6423 continue_failure_jump:
6424 EXTRACT_NUMBER_AND_INCR (mcnt, pat);
6425 p = pat + mcnt;
6426 break;
6427
6428 case no_op:
6429 /* A special frame used for nastyloops. */
6430 goto fail;
6431
6432 default:
6433 abort();
6434 }
6435
6436 assert (p >= bufp->buffer && p <= pend);
6437
6438 if (d >= string1 && d <= end1)
6439 dend = end_match_1;
6440 }
6441 else
6442 break; /* Matching at this starting point really fails. */
6443 } /* for (;;) */
6444
6445 if (best_regs_set)
6446 goto restore_best_regs;
6447
6448 FREE_VARIABLES ();
6449
6450 return -1; /* Failure to match. */
6451 } /* re_match_2 */
6452 \f
6453 /* Subroutine definitions for re_match_2. */
6454
6455 /* Return zero if TRANSLATE[S1] and TRANSLATE[S2] are identical for LEN
6456 bytes; nonzero otherwise. */
6457
6458 static int
6459 bcmp_translate (s1, s2, len, translate, target_multibyte)
6460 re_char *s1, *s2;
6461 register int len;
6462 RE_TRANSLATE_TYPE translate;
6463 const int target_multibyte;
6464 {
6465 register re_char *p1 = s1, *p2 = s2;
6466 re_char *p1_end = s1 + len;
6467 re_char *p2_end = s2 + len;
6468
6469 /* FIXME: Checking both p1 and p2 presumes that the two strings might have
6470 different lengths, but relying on a single `len' would break this. -sm */
6471 while (p1 < p1_end && p2 < p2_end)
6472 {
6473 int p1_charlen, p2_charlen;
6474 re_wchar_t p1_ch, p2_ch;
6475
6476 GET_CHAR_AFTER (p1_ch, p1, p1_charlen);
6477 GET_CHAR_AFTER (p2_ch, p2, p2_charlen);
6478
6479 if (RE_TRANSLATE (translate, p1_ch)
6480 != RE_TRANSLATE (translate, p2_ch))
6481 return 1;
6482
6483 p1 += p1_charlen, p2 += p2_charlen;
6484 }
6485
6486 if (p1 != p1_end || p2 != p2_end)
6487 return 1;
6488
6489 return 0;
6490 }
6491 \f
6492 /* Entry points for GNU code. */
6493
6494 /* re_compile_pattern is the GNU regular expression compiler: it
6495 compiles PATTERN (of length SIZE) and puts the result in BUFP.
6496 Returns 0 if the pattern was valid, otherwise an error string.
6497
6498 Assumes the `allocated' (and perhaps `buffer') and `translate' fields
6499 are set in BUFP on entry.
6500
6501 We call regex_compile to do the actual compilation. */
6502
6503 const char *
6504 re_compile_pattern (pattern, length, bufp)
6505 const char *pattern;
6506 size_t length;
6507 struct re_pattern_buffer *bufp;
6508 {
6509 reg_errcode_t ret;
6510
6511 #ifdef emacs
6512 gl_state.current_syntax_table = current_buffer->syntax_table;
6513 #endif
6514
6515 /* GNU code is written to assume at least RE_NREGS registers will be set
6516 (and at least one extra will be -1). */
6517 bufp->regs_allocated = REGS_UNALLOCATED;
6518
6519 /* And GNU code determines whether or not to get register information
6520 by passing null for the REGS argument to re_match, etc., not by
6521 setting no_sub. */
6522 bufp->no_sub = 0;
6523
6524 ret = regex_compile ((re_char*) pattern, length, re_syntax_options, bufp);
6525
6526 if (!ret)
6527 return NULL;
6528 return gettext (re_error_msgid[(int) ret]);
6529 }
6530 WEAK_ALIAS (__re_compile_pattern, re_compile_pattern)
6531 \f
6532 /* Entry points compatible with 4.2 BSD regex library. We don't define
6533 them unless specifically requested. */
6534
6535 #if defined _REGEX_RE_COMP || defined _LIBC
6536
6537 /* BSD has one and only one pattern buffer. */
6538 static struct re_pattern_buffer re_comp_buf;
6539
6540 char *
6541 # ifdef _LIBC
6542 /* Make these definitions weak in libc, so POSIX programs can redefine
6543 these names if they don't use our functions, and still use
6544 regcomp/regexec below without link errors. */
6545 weak_function
6546 # endif
6547 re_comp (s)
6548 const char *s;
6549 {
6550 reg_errcode_t ret;
6551
6552 if (!s)
6553 {
6554 if (!re_comp_buf.buffer)
6555 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6556 return (char *) gettext ("No previous regular expression");
6557 return 0;
6558 }
6559
6560 if (!re_comp_buf.buffer)
6561 {
6562 re_comp_buf.buffer = (unsigned char *) malloc (200);
6563 if (re_comp_buf.buffer == NULL)
6564 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6565 return (char *) gettext (re_error_msgid[(int) REG_ESPACE]);
6566 re_comp_buf.allocated = 200;
6567
6568 re_comp_buf.fastmap = (char *) malloc (1 << BYTEWIDTH);
6569 if (re_comp_buf.fastmap == NULL)
6570 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6571 return (char *) gettext (re_error_msgid[(int) REG_ESPACE]);
6572 }
6573
6574 /* Since `re_exec' always passes NULL for the `regs' argument, we
6575 don't need to initialize the pattern buffer fields which affect it. */
6576
6577 ret = regex_compile (s, strlen (s), re_syntax_options, &re_comp_buf);
6578
6579 if (!ret)
6580 return NULL;
6581
6582 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6583 return (char *) gettext (re_error_msgid[(int) ret]);
6584 }
6585
6586
6587 int
6588 # ifdef _LIBC
6589 weak_function
6590 # endif
6591 re_exec (s)
6592 const char *s;
6593 {
6594 const int len = strlen (s);
6595 return
6596 0 <= re_search (&re_comp_buf, s, len, 0, len, (struct re_registers *) 0);
6597 }
6598 #endif /* _REGEX_RE_COMP */
6599 \f
6600 /* POSIX.2 functions. Don't define these for Emacs. */
6601
6602 #ifndef emacs
6603
6604 /* regcomp takes a regular expression as a string and compiles it.
6605
6606 PREG is a regex_t *. We do not expect any fields to be initialized,
6607 since POSIX says we shouldn't. Thus, we set
6608
6609 `buffer' to the compiled pattern;
6610 `used' to the length of the compiled pattern;
6611 `syntax' to RE_SYNTAX_POSIX_EXTENDED if the
6612 REG_EXTENDED bit in CFLAGS is set; otherwise, to
6613 RE_SYNTAX_POSIX_BASIC;
6614 `fastmap' to an allocated space for the fastmap;
6615 `fastmap_accurate' to zero;
6616 `re_nsub' to the number of subexpressions in PATTERN.
6617
6618 PATTERN is the address of the pattern string.
6619
6620 CFLAGS is a series of bits which affect compilation.
6621
6622 If REG_EXTENDED is set, we use POSIX extended syntax; otherwise, we
6623 use POSIX basic syntax.
6624
6625 If REG_NEWLINE is set, then . and [^...] don't match newline.
6626 Also, regexec will try a match beginning after every newline.
6627
6628 If REG_ICASE is set, then we considers upper- and lowercase
6629 versions of letters to be equivalent when matching.
6630
6631 If REG_NOSUB is set, then when PREG is passed to regexec, that
6632 routine will report only success or failure, and nothing about the
6633 registers.
6634
6635 It returns 0 if it succeeds, nonzero if it doesn't. (See regex.h for
6636 the return codes and their meanings.) */
6637
6638 int
6639 regcomp (preg, pattern, cflags)
6640 regex_t *__restrict preg;
6641 const char *__restrict pattern;
6642 int cflags;
6643 {
6644 reg_errcode_t ret;
6645 reg_syntax_t syntax
6646 = (cflags & REG_EXTENDED) ?
6647 RE_SYNTAX_POSIX_EXTENDED : RE_SYNTAX_POSIX_BASIC;
6648
6649 /* regex_compile will allocate the space for the compiled pattern. */
6650 preg->buffer = 0;
6651 preg->allocated = 0;
6652 preg->used = 0;
6653
6654 /* Try to allocate space for the fastmap. */
6655 preg->fastmap = (char *) malloc (1 << BYTEWIDTH);
6656
6657 if (cflags & REG_ICASE)
6658 {
6659 unsigned i;
6660
6661 preg->translate
6662 = (RE_TRANSLATE_TYPE) malloc (CHAR_SET_SIZE
6663 * sizeof (*(RE_TRANSLATE_TYPE)0));
6664 if (preg->translate == NULL)
6665 return (int) REG_ESPACE;
6666
6667 /* Map uppercase characters to corresponding lowercase ones. */
6668 for (i = 0; i < CHAR_SET_SIZE; i++)
6669 preg->translate[i] = ISUPPER (i) ? TOLOWER (i) : i;
6670 }
6671 else
6672 preg->translate = NULL;
6673
6674 /* If REG_NEWLINE is set, newlines are treated differently. */
6675 if (cflags & REG_NEWLINE)
6676 { /* REG_NEWLINE implies neither . nor [^...] match newline. */
6677 syntax &= ~RE_DOT_NEWLINE;
6678 syntax |= RE_HAT_LISTS_NOT_NEWLINE;
6679 }
6680 else
6681 syntax |= RE_NO_NEWLINE_ANCHOR;
6682
6683 preg->no_sub = !!(cflags & REG_NOSUB);
6684
6685 /* POSIX says a null character in the pattern terminates it, so we
6686 can use strlen here in compiling the pattern. */
6687 ret = regex_compile ((re_char*) pattern, strlen (pattern), syntax, preg);
6688
6689 /* POSIX doesn't distinguish between an unmatched open-group and an
6690 unmatched close-group: both are REG_EPAREN. */
6691 if (ret == REG_ERPAREN)
6692 ret = REG_EPAREN;
6693
6694 if (ret == REG_NOERROR && preg->fastmap)
6695 { /* Compute the fastmap now, since regexec cannot modify the pattern
6696 buffer. */
6697 re_compile_fastmap (preg);
6698 if (preg->can_be_null)
6699 { /* The fastmap can't be used anyway. */
6700 free (preg->fastmap);
6701 preg->fastmap = NULL;
6702 }
6703 }
6704 return (int) ret;
6705 }
6706 WEAK_ALIAS (__regcomp, regcomp)
6707
6708
6709 /* regexec searches for a given pattern, specified by PREG, in the
6710 string STRING.
6711
6712 If NMATCH is zero or REG_NOSUB was set in the cflags argument to
6713 `regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at
6714 least NMATCH elements, and we set them to the offsets of the
6715 corresponding matched substrings.
6716
6717 EFLAGS specifies `execution flags' which affect matching: if
6718 REG_NOTBOL is set, then ^ does not match at the beginning of the
6719 string; if REG_NOTEOL is set, then $ does not match at the end.
6720
6721 We return 0 if we find a match and REG_NOMATCH if not. */
6722
6723 int
6724 regexec (preg, string, nmatch, pmatch, eflags)
6725 const regex_t *__restrict preg;
6726 const char *__restrict string;
6727 size_t nmatch;
6728 regmatch_t pmatch[__restrict_arr];
6729 int eflags;
6730 {
6731 int ret;
6732 struct re_registers regs;
6733 regex_t private_preg;
6734 int len = strlen (string);
6735 boolean want_reg_info = !preg->no_sub && nmatch > 0 && pmatch;
6736
6737 private_preg = *preg;
6738
6739 private_preg.not_bol = !!(eflags & REG_NOTBOL);
6740 private_preg.not_eol = !!(eflags & REG_NOTEOL);
6741
6742 /* The user has told us exactly how many registers to return
6743 information about, via `nmatch'. We have to pass that on to the
6744 matching routines. */
6745 private_preg.regs_allocated = REGS_FIXED;
6746
6747 if (want_reg_info)
6748 {
6749 regs.num_regs = nmatch;
6750 regs.start = TALLOC (nmatch * 2, regoff_t);
6751 if (regs.start == NULL)
6752 return (int) REG_NOMATCH;
6753 regs.end = regs.start + nmatch;
6754 }
6755
6756 /* Instead of using not_eol to implement REG_NOTEOL, we could simply
6757 pass (&private_preg, string, len + 1, 0, len, ...) pretending the string
6758 was a little bit longer but still only matching the real part.
6759 This works because the `endline' will check for a '\n' and will find a
6760 '\0', correctly deciding that this is not the end of a line.
6761 But it doesn't work out so nicely for REG_NOTBOL, since we don't have
6762 a convenient '\0' there. For all we know, the string could be preceded
6763 by '\n' which would throw things off. */
6764
6765 /* Perform the searching operation. */
6766 ret = re_search (&private_preg, string, len,
6767 /* start: */ 0, /* range: */ len,
6768 want_reg_info ? &regs : (struct re_registers *) 0);
6769
6770 /* Copy the register information to the POSIX structure. */
6771 if (want_reg_info)
6772 {
6773 if (ret >= 0)
6774 {
6775 unsigned r;
6776
6777 for (r = 0; r < nmatch; r++)
6778 {
6779 pmatch[r].rm_so = regs.start[r];
6780 pmatch[r].rm_eo = regs.end[r];
6781 }
6782 }
6783
6784 /* If we needed the temporary register info, free the space now. */
6785 free (regs.start);
6786 }
6787
6788 /* We want zero return to mean success, unlike `re_search'. */
6789 return ret >= 0 ? (int) REG_NOERROR : (int) REG_NOMATCH;
6790 }
6791 WEAK_ALIAS (__regexec, regexec)
6792
6793
6794 /* Returns a message corresponding to an error code, ERR_CODE, returned
6795 from either regcomp or regexec. We don't use PREG here.
6796
6797 ERR_CODE was previously called ERRCODE, but that name causes an
6798 error with msvc8 compiler. */
6799
6800 size_t
6801 regerror (err_code, preg, errbuf, errbuf_size)
6802 int err_code;
6803 const regex_t *preg;
6804 char *errbuf;
6805 size_t errbuf_size;
6806 {
6807 const char *msg;
6808 size_t msg_size;
6809
6810 if (err_code < 0
6811 || err_code >= (sizeof (re_error_msgid) / sizeof (re_error_msgid[0])))
6812 /* Only error codes returned by the rest of the code should be passed
6813 to this routine. If we are given anything else, or if other regex
6814 code generates an invalid error code, then the program has a bug.
6815 Dump core so we can fix it. */
6816 abort ();
6817
6818 msg = gettext (re_error_msgid[err_code]);
6819
6820 msg_size = strlen (msg) + 1; /* Includes the null. */
6821
6822 if (errbuf_size != 0)
6823 {
6824 if (msg_size > errbuf_size)
6825 {
6826 strncpy (errbuf, msg, errbuf_size - 1);
6827 errbuf[errbuf_size - 1] = 0;
6828 }
6829 else
6830 strcpy (errbuf, msg);
6831 }
6832
6833 return msg_size;
6834 }
6835 WEAK_ALIAS (__regerror, regerror)
6836
6837
6838 /* Free dynamically allocated space used by PREG. */
6839
6840 void
6841 regfree (preg)
6842 regex_t *preg;
6843 {
6844 if (preg->buffer != NULL)
6845 free (preg->buffer);
6846 preg->buffer = NULL;
6847
6848 preg->allocated = 0;
6849 preg->used = 0;
6850
6851 if (preg->fastmap != NULL)
6852 free (preg->fastmap);
6853 preg->fastmap = NULL;
6854 preg->fastmap_accurate = 0;
6855
6856 if (preg->translate != NULL)
6857 free (preg->translate);
6858 preg->translate = NULL;
6859 }
6860 WEAK_ALIAS (__regfree, regfree)
6861
6862 #endif /* not emacs */
6863
6864 /* arch-tag: 4ffd68ba-2a9e-435b-a21a-018990f9eeb2
6865 (do not change this comment) */
Unnamed repository; edit this file 'description' to name the repository.