?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
(calendar-chinese-from-absolute, calendar-chinese-date-string): Expand
[emacs.git] / src / regex.c
blob1400a7daa77e7c8e1bc77b8b6cbf5840ee9f78ed
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) += incr, SET_HIGH_BOUND (P), __ptrvalue (P) += incr)
1859 # define ELSE_EXTEND_BUFFER_HIGH_BOUND \
1860 else \
1861 { \
1862 SET_HIGH_BOUND (b); \
1863 SET_HIGH_BOUND (begalt); \
1864 if (fixup_alt_jump) \
1865 SET_HIGH_BOUND (fixup_alt_jump); \
1866 if (laststart) \
1867 SET_HIGH_BOUND (laststart); \
1868 if (pending_exact) \
1869 SET_HIGH_BOUND (pending_exact); \
1870 }
1871 #else
1872 # define MOVE_BUFFER_POINTER(P) (P) += incr
1873 # define ELSE_EXTEND_BUFFER_HIGH_BOUND
1874 #endif
1875 #define EXTEND_BUFFER() \
1876 do { \
1877 re_char *old_buffer = bufp->buffer; \
1878 if (bufp->allocated == MAX_BUF_SIZE) \
1879 return REG_ESIZE; \
1880 bufp->allocated <<= 1; \
1881 if (bufp->allocated > MAX_BUF_SIZE) \
1882 bufp->allocated = MAX_BUF_SIZE; \
1883 RETALLOC (bufp->buffer, bufp->allocated, unsigned char); \
1884 if (bufp->buffer == NULL) \
1885 return REG_ESPACE; \
1886 /* If the buffer moved, move all the pointers into it. */ \
1887 if (old_buffer != bufp->buffer) \
1888 { \
1889 int incr = bufp->buffer - old_buffer; \
1890 MOVE_BUFFER_POINTER (b); \
1891 MOVE_BUFFER_POINTER (begalt); \
1892 if (fixup_alt_jump) \
1893 MOVE_BUFFER_POINTER (fixup_alt_jump); \
1894 if (laststart) \
1895 MOVE_BUFFER_POINTER (laststart); \
1896 if (pending_exact) \
1897 MOVE_BUFFER_POINTER (pending_exact); \
1898 } \
1899 ELSE_EXTEND_BUFFER_HIGH_BOUND \
1900 } while (0)
1901
1902
1903 /* Since we have one byte reserved for the register number argument to
1904 {start,stop}_memory, the maximum number of groups we can report
1905 things about is what fits in that byte. */
1906 #define MAX_REGNUM 255
1907
1908 /* But patterns can have more than `MAX_REGNUM' registers. We just
1909 ignore the excess. */
1910 typedef int regnum_t;
1911
1912
1913 /* Macros for the compile stack. */
1914
1915 /* Since offsets can go either forwards or backwards, this type needs to
1916 be able to hold values from -(MAX_BUF_SIZE - 1) to MAX_BUF_SIZE - 1. */
1917 /* int may be not enough when sizeof(int) == 2. */
1918 typedef long pattern_offset_t;
1919
1920 typedef struct
1921 {
1922 pattern_offset_t begalt_offset;
1923 pattern_offset_t fixup_alt_jump;
1924 pattern_offset_t laststart_offset;
1925 regnum_t regnum;
1926 } compile_stack_elt_t;
1927
1928
1929 typedef struct
1930 {
1931 compile_stack_elt_t *stack;
1932 unsigned size;
1933 unsigned avail; /* Offset of next open position. */
1934 } compile_stack_type;
1935
1936
1937 #define INIT_COMPILE_STACK_SIZE 32
1938
1939 #define COMPILE_STACK_EMPTY (compile_stack.avail == 0)
1940 #define COMPILE_STACK_FULL (compile_stack.avail == compile_stack.size)
1941
1942 /* The next available element. */
1943 #define COMPILE_STACK_TOP (compile_stack.stack[compile_stack.avail])
1944
1945 /* Explicit quit checking is only used on NTemacs and whenever we
1946 use polling to process input events. */
1947 #if defined emacs && (defined WINDOWSNT || defined SYNC_INPUT) && defined QUIT
1948 extern int immediate_quit;
1949 # define IMMEDIATE_QUIT_CHECK \
1950 do { \
1951 if (immediate_quit) QUIT; \
1952 } while (0)
1953 #else
1954 # define IMMEDIATE_QUIT_CHECK ((void)0)
1955 #endif
1956 \f
1957 /* Structure to manage work area for range table. */
1958 struct range_table_work_area
1959 {
1960 int *table; /* actual work area. */
1961 int allocated; /* allocated size for work area in bytes. */
1962 int used; /* actually used size in words. */
1963 int bits; /* flag to record character classes */
1964 };
1965
1966 /* Make sure that WORK_AREA can hold more N multibyte characters.
1967 This is used only in set_image_of_range and set_image_of_range_1.
1968 It expects WORK_AREA to be a pointer.
1969 If it can't get the space, it returns from the surrounding function. */
1970
1971 #define EXTEND_RANGE_TABLE(work_area, n) \
1972 do { \
1973 if (((work_area).used + (n)) * sizeof (int) > (work_area).allocated) \
1974 { \
1975 extend_range_table_work_area (&work_area); \
1976 if ((work_area).table == 0) \
1977 return (REG_ESPACE); \
1978 } \
1979 } while (0)
1980
1981 #define SET_RANGE_TABLE_WORK_AREA_BIT(work_area, bit) \
1982 (work_area).bits |= (bit)
1983
1984 /* Bits used to implement the multibyte-part of the various character classes
1985 such as [:alnum:] in a charset's range table. */
1986 #define BIT_WORD 0x1
1987 #define BIT_LOWER 0x2
1988 #define BIT_PUNCT 0x4
1989 #define BIT_SPACE 0x8
1990 #define BIT_UPPER 0x10
1991 #define BIT_MULTIBYTE 0x20
1992
1993 /* Set a range (RANGE_START, RANGE_END) to WORK_AREA. */
1994 #define SET_RANGE_TABLE_WORK_AREA(work_area, range_start, range_end) \
1995 do { \
1996 EXTEND_RANGE_TABLE ((work_area), 2); \
1997 (work_area).table[(work_area).used++] = (range_start); \
1998 (work_area).table[(work_area).used++] = (range_end); \
1999 } while (0)
2000
2001 /* Free allocated memory for WORK_AREA. */
2002 #define FREE_RANGE_TABLE_WORK_AREA(work_area) \
2003 do { \
2004 if ((work_area).table) \
2005 free ((work_area).table); \
2006 } while (0)
2007
2008 #define CLEAR_RANGE_TABLE_WORK_USED(work_area) ((work_area).used = 0, (work_area).bits = 0)
2009 #define RANGE_TABLE_WORK_USED(work_area) ((work_area).used)
2010 #define RANGE_TABLE_WORK_BITS(work_area) ((work_area).bits)
2011 #define RANGE_TABLE_WORK_ELT(work_area, i) ((work_area).table[i])
2012 \f
2013
2014 /* Set the bit for character C in a list. */
2015 #define SET_LIST_BIT(c) (b[((c)) / BYTEWIDTH] |= 1 << ((c) % BYTEWIDTH))
2016
2017
2018 #ifdef emacs
2019
2020 /* Store characters in the range FROM to TO in the bitmap at B (for
2021 ASCII and unibyte characters) and WORK_AREA (for multibyte
2022 characters) while translating them and paying attention to the
2023 continuity of translated characters.
2024
2025 Implementation note: It is better to implement these fairly big
2026 macros by a function, but it's not that easy because macros called
2027 in this macro assume various local variables already declared. */
2028
2029 /* Both FROM and TO are ASCII characters. */
2030
2031 #define SETUP_ASCII_RANGE(work_area, FROM, TO) \
2032 do { \
2033 int C0, C1; \
2034 \
2035 for (C0 = (FROM); C0 <= (TO); C0++) \
2036 { \
2037 C1 = TRANSLATE (C0); \
2038 if (! ASCII_CHAR_P (C1)) \
2039 { \
2040 SET_RANGE_TABLE_WORK_AREA ((work_area), C1, C1); \
2041 if ((C1 = RE_CHAR_TO_UNIBYTE (C1)) < 0) \
2042 C1 = C0; \
2043 } \
2044 SET_LIST_BIT (C1); \
2045 } \
2046 } while (0)
2047
2048
2049 /* Both FROM and TO are unibyte characters (0x80..0xFF). */
2050
2051 #define SETUP_UNIBYTE_RANGE(work_area, FROM, TO) \
2052 do { \
2053 int C0, C1, C2, I; \
2054 int USED = RANGE_TABLE_WORK_USED (work_area); \
2055 \
2056 for (C0 = (FROM); C0 <= (TO); C0++) \
2057 { \
2058 C1 = RE_CHAR_TO_MULTIBYTE (C0); \
2059 if (CHAR_BYTE8_P (C1)) \
2060 SET_LIST_BIT (C0); \
2061 else \
2062 { \
2063 C2 = TRANSLATE (C1); \
2064 if (C2 == C1 \
2065 || (C1 = RE_CHAR_TO_UNIBYTE (C2)) < 0) \
2066 C1 = C0; \
2067 SET_LIST_BIT (C1); \
2068 for (I = RANGE_TABLE_WORK_USED (work_area) - 2; I >= USED; I -= 2) \
2069 { \
2070 int from = RANGE_TABLE_WORK_ELT (work_area, I); \
2071 int to = RANGE_TABLE_WORK_ELT (work_area, I + 1); \
2072 \
2073 if (C2 >= from - 1 && C2 <= to + 1) \
2074 { \
2075 if (C2 == from - 1) \
2076 RANGE_TABLE_WORK_ELT (work_area, I)--; \
2077 else if (C2 == to + 1) \
2078 RANGE_TABLE_WORK_ELT (work_area, I + 1)++; \
2079 break; \
2080 } \
2081 } \
2082 if (I < USED) \
2083 SET_RANGE_TABLE_WORK_AREA ((work_area), C2, C2); \
2084 } \
2085 } \
2086 } while (0)
2087
2088
2089 /* Both FROM and TO are mulitbyte characters. */
2090
2091 #define SETUP_MULTIBYTE_RANGE(work_area, FROM, TO) \
2092 do { \
2093 int C0, C1, C2, I, USED = RANGE_TABLE_WORK_USED (work_area); \
2094 \
2095 SET_RANGE_TABLE_WORK_AREA ((work_area), (FROM), (TO)); \
2096 for (C0 = (FROM); C0 <= (TO); C0++) \
2097 { \
2098 C1 = TRANSLATE (C0); \
2099 if ((C2 = RE_CHAR_TO_UNIBYTE (C1)) >= 0 \
2100 || (C1 != C0 && (C2 = RE_CHAR_TO_UNIBYTE (C0)) >= 0)) \
2101 SET_LIST_BIT (C2); \
2102 if (C1 >= (FROM) && C1 <= (TO)) \
2103 continue; \
2104 for (I = RANGE_TABLE_WORK_USED (work_area) - 2; I >= USED; I -= 2) \
2105 { \
2106 int from = RANGE_TABLE_WORK_ELT (work_area, I); \
2107 int to = RANGE_TABLE_WORK_ELT (work_area, I + 1); \
2108 \
2109 if (C1 >= from - 1 && C1 <= to + 1) \
2110 { \
2111 if (C1 == from - 1) \
2112 RANGE_TABLE_WORK_ELT (work_area, I)--; \
2113 else if (C1 == to + 1) \
2114 RANGE_TABLE_WORK_ELT (work_area, I + 1)++; \
2115 break; \
2116 } \
2117 } \
2118 if (I < USED) \
2119 SET_RANGE_TABLE_WORK_AREA ((work_area), C1, C1); \
2120 } \
2121 } while (0)
2122
2123 #endif /* emacs */
2124
2125 /* Get the next unsigned number in the uncompiled pattern. */
2126 #define GET_UNSIGNED_NUMBER(num) \
2127 do { \
2128 if (p == pend) \
2129 FREE_STACK_RETURN (REG_EBRACE); \
2130 else \
2131 { \
2132 PATFETCH (c); \
2133 while ('0' <= c && c <= '9') \
2134 { \
2135 int prev; \
2136 if (num < 0) \
2137 num = 0; \
2138 prev = num; \
2139 num = num * 10 + c - '0'; \
2140 if (num / 10 != prev) \
2141 FREE_STACK_RETURN (REG_BADBR); \
2142 if (p == pend) \
2143 FREE_STACK_RETURN (REG_EBRACE); \
2144 PATFETCH (c); \
2145 } \
2146 } \
2147 } while (0)
2148 \f
2149 #if ! WIDE_CHAR_SUPPORT
2150
2151 /* Map a string to the char class it names (if any). */
2152 re_wctype_t
2153 re_wctype (str)
2154 re_char *str;
2155 {
2156 const char *string = str;
2157 if (STREQ (string, "alnum")) return RECC_ALNUM;
2158 else if (STREQ (string, "alpha")) return RECC_ALPHA;
2159 else if (STREQ (string, "word")) return RECC_WORD;
2160 else if (STREQ (string, "ascii")) return RECC_ASCII;
2161 else if (STREQ (string, "nonascii")) return RECC_NONASCII;
2162 else if (STREQ (string, "graph")) return RECC_GRAPH;
2163 else if (STREQ (string, "lower")) return RECC_LOWER;
2164 else if (STREQ (string, "print")) return RECC_PRINT;
2165 else if (STREQ (string, "punct")) return RECC_PUNCT;
2166 else if (STREQ (string, "space")) return RECC_SPACE;
2167 else if (STREQ (string, "upper")) return RECC_UPPER;
2168 else if (STREQ (string, "unibyte")) return RECC_UNIBYTE;
2169 else if (STREQ (string, "multibyte")) return RECC_MULTIBYTE;
2170 else if (STREQ (string, "digit")) return RECC_DIGIT;
2171 else if (STREQ (string, "xdigit")) return RECC_XDIGIT;
2172 else if (STREQ (string, "cntrl")) return RECC_CNTRL;
2173 else if (STREQ (string, "blank")) return RECC_BLANK;
2174 else return 0;
2175 }
2176
2177 /* True if CH is in the char class CC. */
2178 boolean
2179 re_iswctype (ch, cc)
2180 int ch;
2181 re_wctype_t cc;
2182 {
2183 switch (cc)
2184 {
2185 case RECC_ALNUM: return ISALNUM (ch);
2186 case RECC_ALPHA: return ISALPHA (ch);
2187 case RECC_BLANK: return ISBLANK (ch);
2188 case RECC_CNTRL: return ISCNTRL (ch);
2189 case RECC_DIGIT: return ISDIGIT (ch);
2190 case RECC_GRAPH: return ISGRAPH (ch);
2191 case RECC_LOWER: return ISLOWER (ch);
2192 case RECC_PRINT: return ISPRINT (ch);
2193 case RECC_PUNCT: return ISPUNCT (ch);
2194 case RECC_SPACE: return ISSPACE (ch);
2195 case RECC_UPPER: return ISUPPER (ch);
2196 case RECC_XDIGIT: return ISXDIGIT (ch);
2197 case RECC_ASCII: return IS_REAL_ASCII (ch);
2198 case RECC_NONASCII: return !IS_REAL_ASCII (ch);
2199 case RECC_UNIBYTE: return ISUNIBYTE (ch);
2200 case RECC_MULTIBYTE: return !ISUNIBYTE (ch);
2201 case RECC_WORD: return ISWORD (ch);
2202 case RECC_ERROR: return false;
2203 default:
2204 abort();
2205 }
2206 }
2207
2208 /* Return a bit-pattern to use in the range-table bits to match multibyte
2209 chars of class CC. */
2210 static int
2211 re_wctype_to_bit (cc)
2212 re_wctype_t cc;
2213 {
2214 switch (cc)
2215 {
2216 case RECC_NONASCII: case RECC_PRINT: case RECC_GRAPH:
2217 case RECC_MULTIBYTE: return BIT_MULTIBYTE;
2218 case RECC_ALPHA: case RECC_ALNUM: case RECC_WORD: return BIT_WORD;
2219 case RECC_LOWER: return BIT_LOWER;
2220 case RECC_UPPER: return BIT_UPPER;
2221 case RECC_PUNCT: return BIT_PUNCT;
2222 case RECC_SPACE: return BIT_SPACE;
2223 case RECC_ASCII: case RECC_DIGIT: case RECC_XDIGIT: case RECC_CNTRL:
2224 case RECC_BLANK: case RECC_UNIBYTE: case RECC_ERROR: return 0;
2225 default:
2226 abort();
2227 }
2228 }
2229 #endif
2230 \f
2231 /* Filling in the work area of a range. */
2232
2233 /* Actually extend the space in WORK_AREA. */
2234
2235 static void
2236 extend_range_table_work_area (work_area)
2237 struct range_table_work_area *work_area;
2238 {
2239 work_area->allocated += 16 * sizeof (int);
2240 if (work_area->table)
2241 work_area->table
2242 = (int *) realloc (work_area->table, work_area->allocated);
2243 else
2244 work_area->table
2245 = (int *) malloc (work_area->allocated);
2246 }
2247
2248 #if 0
2249 #ifdef emacs
2250
2251 /* Carefully find the ranges of codes that are equivalent
2252 under case conversion to the range start..end when passed through
2253 TRANSLATE. Handle the case where non-letters can come in between
2254 two upper-case letters (which happens in Latin-1).
2255 Also handle the case of groups of more than 2 case-equivalent chars.
2256
2257 The basic method is to look at consecutive characters and see
2258 if they can form a run that can be handled as one.
2259
2260 Returns -1 if successful, REG_ESPACE if ran out of space. */
2261
2262 static int
2263 set_image_of_range_1 (work_area, start, end, translate)
2264 RE_TRANSLATE_TYPE translate;
2265 struct range_table_work_area *work_area;
2266 re_wchar_t start, end;
2267 {
2268 /* `one_case' indicates a character, or a run of characters,
2269 each of which is an isolate (no case-equivalents).
2270 This includes all ASCII non-letters.
2271
2272 `two_case' indicates a character, or a run of characters,
2273 each of which has two case-equivalent forms.
2274 This includes all ASCII letters.
2275
2276 `strange' indicates a character that has more than one
2277 case-equivalent. */
2278
2279 enum case_type {one_case, two_case, strange};
2280
2281 /* Describe the run that is in progress,
2282 which the next character can try to extend.
2283 If run_type is strange, that means there really is no run.
2284 If run_type is one_case, then run_start...run_end is the run.
2285 If run_type is two_case, then the run is run_start...run_end,
2286 and the case-equivalents end at run_eqv_end. */
2287
2288 enum case_type run_type = strange;
2289 int run_start, run_end, run_eqv_end;
2290
2291 Lisp_Object eqv_table;
2292
2293 if (!RE_TRANSLATE_P (translate))
2294 {
2295 EXTEND_RANGE_TABLE (work_area, 2);
2296 work_area->table[work_area->used++] = (start);
2297 work_area->table[work_area->used++] = (end);
2298 return -1;
2299 }
2300
2301 eqv_table = XCHAR_TABLE (translate)->extras[2];
2302
2303 for (; start <= end; start++)
2304 {
2305 enum case_type this_type;
2306 int eqv = RE_TRANSLATE (eqv_table, start);
2307 int minchar, maxchar;
2308
2309 /* Classify this character */
2310 if (eqv == start)
2311 this_type = one_case;
2312 else if (RE_TRANSLATE (eqv_table, eqv) == start)
2313 this_type = two_case;
2314 else
2315 this_type = strange;
2316
2317 if (start < eqv)
2318 minchar = start, maxchar = eqv;
2319 else
2320 minchar = eqv, maxchar = start;
2321
2322 /* Can this character extend the run in progress? */
2323 if (this_type == strange || this_type != run_type
2324 || !(minchar == run_end + 1
2325 && (run_type == two_case
2326 ? maxchar == run_eqv_end + 1 : 1)))
2327 {
2328 /* No, end the run.
2329 Record each of its equivalent ranges. */
2330 if (run_type == one_case)
2331 {
2332 EXTEND_RANGE_TABLE (work_area, 2);
2333 work_area->table[work_area->used++] = run_start;
2334 work_area->table[work_area->used++] = run_end;
2335 }
2336 else if (run_type == two_case)
2337 {
2338 EXTEND_RANGE_TABLE (work_area, 4);
2339 work_area->table[work_area->used++] = run_start;
2340 work_area->table[work_area->used++] = run_end;
2341 work_area->table[work_area->used++]
2342 = RE_TRANSLATE (eqv_table, run_start);
2343 work_area->table[work_area->used++]
2344 = RE_TRANSLATE (eqv_table, run_end);
2345 }
2346 run_type = strange;
2347 }
2348
2349 if (this_type == strange)
2350 {
2351 /* For a strange character, add each of its equivalents, one
2352 by one. Don't start a range. */
2353 do
2354 {
2355 EXTEND_RANGE_TABLE (work_area, 2);
2356 work_area->table[work_area->used++] = eqv;
2357 work_area->table[work_area->used++] = eqv;
2358 eqv = RE_TRANSLATE (eqv_table, eqv);
2359 }
2360 while (eqv != start);
2361 }
2362
2363 /* Add this char to the run, or start a new run. */
2364 else if (run_type == strange)
2365 {
2366 /* Initialize a new range. */
2367 run_type = this_type;
2368 run_start = start;
2369 run_end = start;
2370 run_eqv_end = RE_TRANSLATE (eqv_table, run_end);
2371 }
2372 else
2373 {
2374 /* Extend a running range. */
2375 run_end = minchar;
2376 run_eqv_end = RE_TRANSLATE (eqv_table, run_end);
2377 }
2378 }
2379
2380 /* If a run is still in progress at the end, finish it now
2381 by recording its equivalent ranges. */
2382 if (run_type == one_case)
2383 {
2384 EXTEND_RANGE_TABLE (work_area, 2);
2385 work_area->table[work_area->used++] = run_start;
2386 work_area->table[work_area->used++] = run_end;
2387 }
2388 else if (run_type == two_case)
2389 {
2390 EXTEND_RANGE_TABLE (work_area, 4);
2391 work_area->table[work_area->used++] = run_start;
2392 work_area->table[work_area->used++] = run_end;
2393 work_area->table[work_area->used++]
2394 = RE_TRANSLATE (eqv_table, run_start);
2395 work_area->table[work_area->used++]
2396 = RE_TRANSLATE (eqv_table, run_end);
2397 }
2398
2399 return -1;
2400 }
2401
2402 #endif /* emacs */
2403
2404 /* Record the the image of the range start..end when passed through
2405 TRANSLATE. This is not necessarily TRANSLATE(start)..TRANSLATE(end)
2406 and is not even necessarily contiguous.
2407 Normally we approximate it with the smallest contiguous range that contains
2408 all the chars we need. However, for Latin-1 we go to extra effort
2409 to do a better job.
2410
2411 This function is not called for ASCII ranges.
2412
2413 Returns -1 if successful, REG_ESPACE if ran out of space. */
2414
2415 static int
2416 set_image_of_range (work_area, start, end, translate)
2417 RE_TRANSLATE_TYPE translate;
2418 struct range_table_work_area *work_area;
2419 re_wchar_t start, end;
2420 {
2421 re_wchar_t cmin, cmax;
2422
2423 #ifdef emacs
2424 /* For Latin-1 ranges, use set_image_of_range_1
2425 to get proper handling of ranges that include letters and nonletters.
2426 For a range that includes the whole of Latin-1, this is not necessary.
2427 For other character sets, we don't bother to get this right. */
2428 if (RE_TRANSLATE_P (translate) && start < 04400
2429 && !(start < 04200 && end >= 04377))
2430 {
2431 int newend;
2432 int tem;
2433 newend = end;
2434 if (newend > 04377)
2435 newend = 04377;
2436 tem = set_image_of_range_1 (work_area, start, newend, translate);
2437 if (tem > 0)
2438 return tem;
2439
2440 start = 04400;
2441 if (end < 04400)
2442 return -1;
2443 }
2444 #endif
2445
2446 EXTEND_RANGE_TABLE (work_area, 2);
2447 work_area->table[work_area->used++] = (start);
2448 work_area->table[work_area->used++] = (end);
2449
2450 cmin = -1, cmax = -1;
2451
2452 if (RE_TRANSLATE_P (translate))
2453 {
2454 int ch;
2455
2456 for (ch = start; ch <= end; ch++)
2457 {
2458 re_wchar_t c = TRANSLATE (ch);
2459 if (! (start <= c && c <= end))
2460 {
2461 if (cmin == -1)
2462 cmin = c, cmax = c;
2463 else
2464 {
2465 cmin = MIN (cmin, c);
2466 cmax = MAX (cmax, c);
2467 }
2468 }
2469 }
2470
2471 if (cmin != -1)
2472 {
2473 EXTEND_RANGE_TABLE (work_area, 2);
2474 work_area->table[work_area->used++] = (cmin);
2475 work_area->table[work_area->used++] = (cmax);
2476 }
2477 }
2478
2479 return -1;
2480 }
2481 #endif /* 0 */
2482 \f
2483 #ifndef MATCH_MAY_ALLOCATE
2484
2485 /* If we cannot allocate large objects within re_match_2_internal,
2486 we make the fail stack and register vectors global.
2487 The fail stack, we grow to the maximum size when a regexp
2488 is compiled.
2489 The register vectors, we adjust in size each time we
2490 compile a regexp, according to the number of registers it needs. */
2491
2492 static fail_stack_type fail_stack;
2493
2494 /* Size with which the following vectors are currently allocated.
2495 That is so we can make them bigger as needed,
2496 but never make them smaller. */
2497 static int regs_allocated_size;
2498
2499 static re_char ** regstart, ** regend;
2500 static re_char **best_regstart, **best_regend;
2501
2502 /* Make the register vectors big enough for NUM_REGS registers,
2503 but don't make them smaller. */
2504
2505 static
2506 regex_grow_registers (num_regs)
2507 int num_regs;
2508 {
2509 if (num_regs > regs_allocated_size)
2510 {
2511 RETALLOC_IF (regstart, num_regs, re_char *);
2512 RETALLOC_IF (regend, num_regs, re_char *);
2513 RETALLOC_IF (best_regstart, num_regs, re_char *);
2514 RETALLOC_IF (best_regend, num_regs, re_char *);
2515
2516 regs_allocated_size = num_regs;
2517 }
2518 }
2519
2520 #endif /* not MATCH_MAY_ALLOCATE */
2521 \f
2522 static boolean group_in_compile_stack _RE_ARGS ((compile_stack_type
2523 compile_stack,
2524 regnum_t regnum));
2525
2526 /* `regex_compile' compiles PATTERN (of length SIZE) according to SYNTAX.
2527 Returns one of error codes defined in `regex.h', or zero for success.
2528
2529 Assumes the `allocated' (and perhaps `buffer') and `translate'
2530 fields are set in BUFP on entry.
2531
2532 If it succeeds, results are put in BUFP (if it returns an error, the
2533 contents of BUFP are undefined):
2534 `buffer' is the compiled pattern;
2535 `syntax' is set to SYNTAX;
2536 `used' is set to the length of the compiled pattern;
2537 `fastmap_accurate' is zero;
2538 `re_nsub' is the number of subexpressions in PATTERN;
2539 `not_bol' and `not_eol' are zero;
2540
2541 The `fastmap' field is neither examined nor set. */
2542
2543 /* Insert the `jump' from the end of last alternative to "here".
2544 The space for the jump has already been allocated. */
2545 #define FIXUP_ALT_JUMP() \
2546 do { \
2547 if (fixup_alt_jump) \
2548 STORE_JUMP (jump, fixup_alt_jump, b); \
2549 } while (0)
2550
2551
2552 /* Return, freeing storage we allocated. */
2553 #define FREE_STACK_RETURN(value) \
2554 do { \
2555 FREE_RANGE_TABLE_WORK_AREA (range_table_work); \
2556 free (compile_stack.stack); \
2557 return value; \
2558 } while (0)
2559
2560 static reg_errcode_t
2561 regex_compile (pattern, size, syntax, bufp)
2562 re_char *pattern;
2563 size_t size;
2564 reg_syntax_t syntax;
2565 struct re_pattern_buffer *bufp;
2566 {
2567 /* We fetch characters from PATTERN here. */
2568 register re_wchar_t c, c1;
2569
2570 /* A random temporary spot in PATTERN. */
2571 re_char *p1;
2572
2573 /* Points to the end of the buffer, where we should append. */
2574 register unsigned char *b;
2575
2576 /* Keeps track of unclosed groups. */
2577 compile_stack_type compile_stack;
2578
2579 /* Points to the current (ending) position in the pattern. */
2580 #ifdef AIX
2581 /* `const' makes AIX compiler fail. */
2582 unsigned char *p = pattern;
2583 #else
2584 re_char *p = pattern;
2585 #endif
2586 re_char *pend = pattern + size;
2587
2588 /* How to translate the characters in the pattern. */
2589 RE_TRANSLATE_TYPE translate = bufp->translate;
2590
2591 /* Address of the count-byte of the most recently inserted `exactn'
2592 command. This makes it possible to tell if a new exact-match
2593 character can be added to that command or if the character requires
2594 a new `exactn' command. */
2595 unsigned char *pending_exact = 0;
2596
2597 /* Address of start of the most recently finished expression.
2598 This tells, e.g., postfix * where to find the start of its
2599 operand. Reset at the beginning of groups and alternatives. */
2600 unsigned char *laststart = 0;
2601
2602 /* Address of beginning of regexp, or inside of last group. */
2603 unsigned char *begalt;
2604
2605 /* Place in the uncompiled pattern (i.e., the {) to
2606 which to go back if the interval is invalid. */
2607 re_char *beg_interval;
2608
2609 /* Address of the place where a forward jump should go to the end of
2610 the containing expression. Each alternative of an `or' -- except the
2611 last -- ends with a forward jump of this sort. */
2612 unsigned char *fixup_alt_jump = 0;
2613
2614 /* Work area for range table of charset. */
2615 struct range_table_work_area range_table_work;
2616
2617 /* If the object matched can contain multibyte characters. */
2618 const boolean multibyte = RE_MULTIBYTE_P (bufp);
2619
2620 /* If a target of matching can contain multibyte characters. */
2621 const boolean target_multibyte = RE_TARGET_MULTIBYTE_P (bufp);
2622
2623 /* Nonzero if we have pushed down into a subpattern. */
2624 int in_subpattern = 0;
2625
2626 /* These hold the values of p, pattern, and pend from the main
2627 pattern when we have pushed into a subpattern. */
2628 re_char *main_p;
2629 re_char *main_pattern;
2630 re_char *main_pend;
2631
2632 #ifdef DEBUG
2633 debug++;
2634 DEBUG_PRINT1 ("\nCompiling pattern: ");
2635 if (debug > 0)
2636 {
2637 unsigned debug_count;
2638
2639 for (debug_count = 0; debug_count < size; debug_count++)
2640 putchar (pattern[debug_count]);
2641 putchar ('\n');
2642 }
2643 #endif /* DEBUG */
2644
2645 /* Initialize the compile stack. */
2646 compile_stack.stack = TALLOC (INIT_COMPILE_STACK_SIZE, compile_stack_elt_t);
2647 if (compile_stack.stack == NULL)
2648 return REG_ESPACE;
2649
2650 compile_stack.size = INIT_COMPILE_STACK_SIZE;
2651 compile_stack.avail = 0;
2652
2653 range_table_work.table = 0;
2654 range_table_work.allocated = 0;
2655
2656 /* Initialize the pattern buffer. */
2657 bufp->syntax = syntax;
2658 bufp->fastmap_accurate = 0;
2659 bufp->not_bol = bufp->not_eol = 0;
2660 bufp->used_syntax = 0;
2661
2662 /* Set `used' to zero, so that if we return an error, the pattern
2663 printer (for debugging) will think there's no pattern. We reset it
2664 at the end. */
2665 bufp->used = 0;
2666
2667 /* Always count groups, whether or not bufp->no_sub is set. */
2668 bufp->re_nsub = 0;
2669
2670 #if !defined emacs && !defined SYNTAX_TABLE
2671 /* Initialize the syntax table. */
2672 init_syntax_once ();
2673 #endif
2674
2675 if (bufp->allocated == 0)
2676 {
2677 if (bufp->buffer)
2678 { /* If zero allocated, but buffer is non-null, try to realloc
2679 enough space. This loses if buffer's address is bogus, but
2680 that is the user's responsibility. */
2681 RETALLOC (bufp->buffer, INIT_BUF_SIZE, unsigned char);
2682 }
2683 else
2684 { /* Caller did not allocate a buffer. Do it for them. */
2685 bufp->buffer = TALLOC (INIT_BUF_SIZE, unsigned char);
2686 }
2687 if (!bufp->buffer) FREE_STACK_RETURN (REG_ESPACE);
2688
2689 bufp->allocated = INIT_BUF_SIZE;
2690 }
2691
2692 begalt = b = bufp->buffer;
2693
2694 /* Loop through the uncompiled pattern until we're at the end. */
2695 while (1)
2696 {
2697 if (p == pend)
2698 {
2699 /* If this is the end of an included regexp,
2700 pop back to the main regexp and try again. */
2701 if (in_subpattern)
2702 {
2703 in_subpattern = 0;
2704 pattern = main_pattern;
2705 p = main_p;
2706 pend = main_pend;
2707 continue;
2708 }
2709 /* If this is the end of the main regexp, we are done. */
2710 break;
2711 }
2712
2713 PATFETCH (c);
2714
2715 switch (c)
2716 {
2717 case ' ':
2718 {
2719 re_char *p1 = p;
2720
2721 /* If there's no special whitespace regexp, treat
2722 spaces normally. And don't try to do this recursively. */
2723 if (!whitespace_regexp || in_subpattern)
2724 goto normal_char;
2725
2726 /* Peek past following spaces. */
2727 while (p1 != pend)
2728 {
2729 if (*p1 != ' ')
2730 break;
2731 p1++;
2732 }
2733 /* If the spaces are followed by a repetition op,
2734 treat them normally. */
2735 if (p1 != pend
2736 && (*p1 == '*' || *p1 == '+' || *p1 == '?'
2737 || (*p1 == '\\' && p1 + 1 != pend && p1[1] == '{')))
2738 goto normal_char;
2739
2740 /* Replace the spaces with the whitespace regexp. */
2741 in_subpattern = 1;
2742 main_p = p1;
2743 main_pend = pend;
2744 main_pattern = pattern;
2745 p = pattern = whitespace_regexp;
2746 pend = p + strlen (p);
2747 break;
2748 }
2749
2750 case '^':
2751 {
2752 if ( /* If at start of pattern, it's an operator. */
2753 p == pattern + 1
2754 /* If context independent, it's an operator. */
2755 || syntax & RE_CONTEXT_INDEP_ANCHORS
2756 /* Otherwise, depends on what's come before. */
2757 || at_begline_loc_p (pattern, p, syntax))
2758 BUF_PUSH ((syntax & RE_NO_NEWLINE_ANCHOR) ? begbuf : begline);
2759 else
2760 goto normal_char;
2761 }
2762 break;
2763
2764
2765 case '$':
2766 {
2767 if ( /* If at end of pattern, it's an operator. */
2768 p == pend
2769 /* If context independent, it's an operator. */
2770 || syntax & RE_CONTEXT_INDEP_ANCHORS
2771 /* Otherwise, depends on what's next. */
2772 || at_endline_loc_p (p, pend, syntax))
2773 BUF_PUSH ((syntax & RE_NO_NEWLINE_ANCHOR) ? endbuf : endline);
2774 else
2775 goto normal_char;
2776 }
2777 break;
2778
2779
2780 case '+':
2781 case '?':
2782 if ((syntax & RE_BK_PLUS_QM)
2783 || (syntax & RE_LIMITED_OPS))
2784 goto normal_char;
2785 handle_plus:
2786 case '*':
2787 /* If there is no previous pattern... */
2788 if (!laststart)
2789 {
2790 if (syntax & RE_CONTEXT_INVALID_OPS)
2791 FREE_STACK_RETURN (REG_BADRPT);
2792 else if (!(syntax & RE_CONTEXT_INDEP_OPS))
2793 goto normal_char;
2794 }
2795
2796 {
2797 /* 1 means zero (many) matches is allowed. */
2798 boolean zero_times_ok = 0, many_times_ok = 0;
2799 boolean greedy = 1;
2800
2801 /* If there is a sequence of repetition chars, collapse it
2802 down to just one (the right one). We can't combine
2803 interval operators with these because of, e.g., `a{2}*',
2804 which should only match an even number of `a's. */
2805
2806 for (;;)
2807 {
2808 if ((syntax & RE_FRUGAL)
2809 && c == '?' && (zero_times_ok || many_times_ok))
2810 greedy = 0;
2811 else
2812 {
2813 zero_times_ok |= c != '+';
2814 many_times_ok |= c != '?';
2815 }
2816
2817 if (p == pend)
2818 break;
2819 else if (*p == '*'
2820 || (!(syntax & RE_BK_PLUS_QM)
2821 && (*p == '+' || *p == '?')))
2822 ;
2823 else if (syntax & RE_BK_PLUS_QM && *p == '\\')
2824 {
2825 if (p+1 == pend)
2826 FREE_STACK_RETURN (REG_EESCAPE);
2827 if (p[1] == '+' || p[1] == '?')
2828 PATFETCH (c); /* Gobble up the backslash. */
2829 else
2830 break;
2831 }
2832 else
2833 break;
2834 /* If we get here, we found another repeat character. */
2835 PATFETCH (c);
2836 }
2837
2838 /* Star, etc. applied to an empty pattern is equivalent
2839 to an empty pattern. */
2840 if (!laststart || laststart == b)
2841 break;
2842
2843 /* Now we know whether or not zero matches is allowed
2844 and also whether or not two or more matches is allowed. */
2845 if (greedy)
2846 {
2847 if (many_times_ok)
2848 {
2849 boolean simple = skip_one_char (laststart) == b;
2850 unsigned int startoffset = 0;
2851 re_opcode_t ofj =
2852 /* Check if the loop can match the empty string. */
2853 (simple || !analyse_first (laststart, b, NULL, 0))
2854 ? on_failure_jump : on_failure_jump_loop;
2855 assert (skip_one_char (laststart) <= b);
2856
2857 if (!zero_times_ok && simple)
2858 { /* Since simple * loops can be made faster by using
2859 on_failure_keep_string_jump, we turn simple P+
2860 into PP* if P is simple. */
2861 unsigned char *p1, *p2;
2862 startoffset = b - laststart;
2863 GET_BUFFER_SPACE (startoffset);
2864 p1 = b; p2 = laststart;
2865 while (p2 < p1)
2866 *b++ = *p2++;
2867 zero_times_ok = 1;
2868 }
2869
2870 GET_BUFFER_SPACE (6);
2871 if (!zero_times_ok)
2872 /* A + loop. */
2873 STORE_JUMP (ofj, b, b + 6);
2874 else
2875 /* Simple * loops can use on_failure_keep_string_jump
2876 depending on what follows. But since we don't know
2877 that yet, we leave the decision up to
2878 on_failure_jump_smart. */
2879 INSERT_JUMP (simple ? on_failure_jump_smart : ofj,
2880 laststart + startoffset, b + 6);
2881 b += 3;
2882 STORE_JUMP (jump, b, laststart + startoffset);
2883 b += 3;
2884 }
2885 else
2886 {
2887 /* A simple ? pattern. */
2888 assert (zero_times_ok);
2889 GET_BUFFER_SPACE (3);
2890 INSERT_JUMP (on_failure_jump, laststart, b + 3);
2891 b += 3;
2892 }
2893 }
2894 else /* not greedy */
2895 { /* I wish the greedy and non-greedy cases could be merged. */
2896
2897 GET_BUFFER_SPACE (7); /* We might use less. */
2898 if (many_times_ok)
2899 {
2900 boolean emptyp = analyse_first (laststart, b, NULL, 0);
2901
2902 /* The non-greedy multiple match looks like
2903 a repeat..until: we only need a conditional jump
2904 at the end of the loop. */
2905 if (emptyp) BUF_PUSH (no_op);
2906 STORE_JUMP (emptyp ? on_failure_jump_nastyloop
2907 : on_failure_jump, b, laststart);
2908 b += 3;
2909 if (zero_times_ok)
2910 {
2911 /* The repeat...until naturally matches one or more.
2912 To also match zero times, we need to first jump to
2913 the end of the loop (its conditional jump). */
2914 INSERT_JUMP (jump, laststart, b);
2915 b += 3;
2916 }
2917 }
2918 else
2919 {
2920 /* non-greedy a?? */
2921 INSERT_JUMP (jump, laststart, b + 3);
2922 b += 3;
2923 INSERT_JUMP (on_failure_jump, laststart, laststart + 6);
2924 b += 3;
2925 }
2926 }
2927 }
2928 pending_exact = 0;
2929 break;
2930
2931
2932 case '.':
2933 laststart = b;
2934 BUF_PUSH (anychar);
2935 break;
2936
2937
2938 case '[':
2939 {
2940 CLEAR_RANGE_TABLE_WORK_USED (range_table_work);
2941
2942 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2943
2944 /* Ensure that we have enough space to push a charset: the
2945 opcode, the length count, and the bitset; 34 bytes in all. */
2946 GET_BUFFER_SPACE (34);
2947
2948 laststart = b;
2949
2950 /* We test `*p == '^' twice, instead of using an if
2951 statement, so we only need one BUF_PUSH. */
2952 BUF_PUSH (*p == '^' ? charset_not : charset);
2953 if (*p == '^')
2954 p++;
2955
2956 /* Remember the first position in the bracket expression. */
2957 p1 = p;
2958
2959 /* Push the number of bytes in the bitmap. */
2960 BUF_PUSH ((1 << BYTEWIDTH) / BYTEWIDTH);
2961
2962 /* Clear the whole map. */
2963 bzero (b, (1 << BYTEWIDTH) / BYTEWIDTH);
2964
2965 /* charset_not matches newline according to a syntax bit. */
2966 if ((re_opcode_t) b[-2] == charset_not
2967 && (syntax & RE_HAT_LISTS_NOT_NEWLINE))
2968 SET_LIST_BIT ('\n');
2969
2970 /* Read in characters and ranges, setting map bits. */
2971 for (;;)
2972 {
2973 boolean escaped_char = false;
2974 const unsigned char *p2 = p;
2975 re_wchar_t ch, c2;
2976
2977 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2978
2979 /* Don't translate yet. The range TRANSLATE(X..Y) cannot
2980 always be determined from TRANSLATE(X) and TRANSLATE(Y)
2981 So the translation is done later in a loop. Example:
2982 (let ((case-fold-search t)) (string-match "[A-_]" "A")) */
2983 PATFETCH (c);
2984
2985 /* \ might escape characters inside [...] and [^...]. */
2986 if ((syntax & RE_BACKSLASH_ESCAPE_IN_LISTS) && c == '\\')
2987 {
2988 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
2989
2990 PATFETCH (c);
2991 escaped_char = true;
2992 }
2993 else
2994 {
2995 /* Could be the end of the bracket expression. If it's
2996 not (i.e., when the bracket expression is `[]' so
2997 far), the ']' character bit gets set way below. */
2998 if (c == ']' && p2 != p1)
2999 break;
3000 }
3001
3002 /* See if we're at the beginning of a possible character
3003 class. */
3004
3005 if (!escaped_char &&
3006 syntax & RE_CHAR_CLASSES && c == '[' && *p == ':')
3007 {
3008 /* Leave room for the null. */
3009 unsigned char str[CHAR_CLASS_MAX_LENGTH + 1];
3010 const unsigned char *class_beg;
3011
3012 PATFETCH (c);
3013 c1 = 0;
3014 class_beg = p;
3015
3016 /* If pattern is `[[:'. */
3017 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
3018
3019 for (;;)
3020 {
3021 PATFETCH (c);
3022 if ((c == ':' && *p == ']') || p == pend)
3023 break;
3024 if (c1 < CHAR_CLASS_MAX_LENGTH)
3025 str[c1++] = c;
3026 else
3027 /* This is in any case an invalid class name. */
3028 str[0] = '\0';
3029 }
3030 str[c1] = '\0';
3031
3032 /* If isn't a word bracketed by `[:' and `:]':
3033 undo the ending character, the letters, and
3034 leave the leading `:' and `[' (but set bits for
3035 them). */
3036 if (c == ':' && *p == ']')
3037 {
3038 re_wctype_t cc;
3039 int limit;
3040
3041 cc = re_wctype (str);
3042
3043 if (cc == 0)
3044 FREE_STACK_RETURN (REG_ECTYPE);
3045
3046 /* Throw away the ] at the end of the character
3047 class. */
3048 PATFETCH (c);
3049
3050 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
3051
3052 #ifndef emacs
3053 for (ch = 0; ch < (1 << BYTEWIDTH); ++ch)
3054 if (re_iswctype (btowc (ch), cc))
3055 {
3056 c = TRANSLATE (ch);
3057 if (c < (1 << BYTEWIDTH))
3058 SET_LIST_BIT (c);
3059 }
3060 #else /* emacs */
3061 /* Most character classes in a multibyte match
3062 just set a flag. Exceptions are is_blank,
3063 is_digit, is_cntrl, and is_xdigit, since
3064 they can only match ASCII characters. We
3065 don't need to handle them for multibyte.
3066 They are distinguished by a negative wctype. */
3067
3068 for (ch = 0; ch < 256; ++ch)
3069 {
3070 c = RE_CHAR_TO_MULTIBYTE (ch);
3071 if (! CHAR_BYTE8_P (c)
3072 && re_iswctype (c, cc))
3073 {
3074 SET_LIST_BIT (ch);
3075 c1 = TRANSLATE (c);
3076 if (c1 == c)
3077 continue;
3078 if (ASCII_CHAR_P (c1))
3079 SET_LIST_BIT (c1);
3080 else if ((c1 = RE_CHAR_TO_UNIBYTE (c1)) >= 0)
3081 SET_LIST_BIT (c1);
3082 }
3083 }
3084 SET_RANGE_TABLE_WORK_AREA_BIT
3085 (range_table_work, re_wctype_to_bit (cc));
3086 #endif /* emacs */
3087 /* In most cases the matching rule for char classes
3088 only uses the syntax table for multibyte chars,
3089 so that the content of the syntax-table it is not
3090 hardcoded in the range_table. SPACE and WORD are
3091 the two exceptions. */
3092 if ((1 << cc) & ((1 << RECC_SPACE) | (1 << RECC_WORD)))
3093 bufp->used_syntax = 1;
3094
3095 /* Repeat the loop. */
3096 continue;
3097 }
3098 else
3099 {
3100 /* Go back to right after the "[:". */
3101 p = class_beg;
3102 SET_LIST_BIT ('[');
3103
3104 /* Because the `:' may starts the range, we
3105 can't simply set bit and repeat the loop.
3106 Instead, just set it to C and handle below. */
3107 c = ':';
3108 }
3109 }
3110
3111 if (p < pend && p[0] == '-' && p[1] != ']')
3112 {
3113
3114 /* Discard the `-'. */
3115 PATFETCH (c1);
3116
3117 /* Fetch the character which ends the range. */
3118 PATFETCH (c1);
3119 #ifdef emacs
3120 if (CHAR_BYTE8_P (c1)
3121 && ! ASCII_CHAR_P (c) && ! CHAR_BYTE8_P (c))
3122 /* Treat the range from a multibyte character to
3123 raw-byte character as empty. */
3124 c = c1 + 1;
3125 #endif /* emacs */
3126 }
3127 else
3128 /* Range from C to C. */
3129 c1 = c;
3130
3131 if (c > c1)
3132 {
3133 if (syntax & RE_NO_EMPTY_RANGES)
3134 FREE_STACK_RETURN (REG_ERANGEX);
3135 /* Else, repeat the loop. */
3136 }
3137 else
3138 {
3139 #ifndef emacs
3140 /* Set the range into bitmap */
3141 for (; c <= c1; c++)
3142 {
3143 ch = TRANSLATE (c);
3144 if (ch < (1 << BYTEWIDTH))
3145 SET_LIST_BIT (ch);
3146 }
3147 #else /* emacs */
3148 if (c < 128)
3149 {
3150 ch = MIN (127, c1);
3151 SETUP_ASCII_RANGE (range_table_work, c, ch);
3152 c = ch + 1;
3153 if (CHAR_BYTE8_P (c1))
3154 c = BYTE8_TO_CHAR (128);
3155 }
3156 if (c <= c1)
3157 {
3158 if (CHAR_BYTE8_P (c))
3159 {
3160 c = CHAR_TO_BYTE8 (c);
3161 c1 = CHAR_TO_BYTE8 (c1);
3162 for (; c <= c1; c++)
3163 SET_LIST_BIT (c);
3164 }
3165 else if (multibyte)
3166 {
3167 SETUP_MULTIBYTE_RANGE (range_table_work, c, c1);
3168 }
3169 else
3170 {
3171 SETUP_UNIBYTE_RANGE (range_table_work, c, c1);
3172 }
3173 }
3174 #endif /* emacs */
3175 }
3176 }
3177
3178 /* Discard any (non)matching list bytes that are all 0 at the
3179 end of the map. Decrease the map-length byte too. */
3180 while ((int) b[-1] > 0 && b[b[-1] - 1] == 0)
3181 b[-1]--;
3182 b += b[-1];
3183
3184 /* Build real range table from work area. */
3185 if (RANGE_TABLE_WORK_USED (range_table_work)
3186 || RANGE_TABLE_WORK_BITS (range_table_work))
3187 {
3188 int i;
3189 int used = RANGE_TABLE_WORK_USED (range_table_work);
3190
3191 /* Allocate space for COUNT + RANGE_TABLE. Needs two
3192 bytes for flags, two for COUNT, and three bytes for
3193 each character. */
3194 GET_BUFFER_SPACE (4 + used * 3);
3195
3196 /* Indicate the existence of range table. */
3197 laststart[1] |= 0x80;
3198
3199 /* Store the character class flag bits into the range table.
3200 If not in emacs, these flag bits are always 0. */
3201 *b++ = RANGE_TABLE_WORK_BITS (range_table_work) & 0xff;
3202 *b++ = RANGE_TABLE_WORK_BITS (range_table_work) >> 8;
3203
3204 STORE_NUMBER_AND_INCR (b, used / 2);
3205 for (i = 0; i < used; i++)
3206 STORE_CHARACTER_AND_INCR
3207 (b, RANGE_TABLE_WORK_ELT (range_table_work, i));
3208 }
3209 }
3210 break;
3211
3212
3213 case '(':
3214 if (syntax & RE_NO_BK_PARENS)
3215 goto handle_open;
3216 else
3217 goto normal_char;
3218
3219
3220 case ')':
3221 if (syntax & RE_NO_BK_PARENS)
3222 goto handle_close;
3223 else
3224 goto normal_char;
3225
3226
3227 case '\n':
3228 if (syntax & RE_NEWLINE_ALT)
3229 goto handle_alt;
3230 else
3231 goto normal_char;
3232
3233
3234 case '|':
3235 if (syntax & RE_NO_BK_VBAR)
3236 goto handle_alt;
3237 else
3238 goto normal_char;
3239
3240
3241 case '{':
3242 if (syntax & RE_INTERVALS && syntax & RE_NO_BK_BRACES)
3243 goto handle_interval;
3244 else
3245 goto normal_char;
3246
3247
3248 case '\\':
3249 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
3250
3251 /* Do not translate the character after the \, so that we can
3252 distinguish, e.g., \B from \b, even if we normally would
3253 translate, e.g., B to b. */
3254 PATFETCH (c);
3255
3256 switch (c)
3257 {
3258 case '(':
3259 if (syntax & RE_NO_BK_PARENS)
3260 goto normal_backslash;
3261
3262 handle_open:
3263 {
3264 int shy = 0;
3265 regnum_t regnum = 0;
3266 if (p+1 < pend)
3267 {
3268 /* Look for a special (?...) construct */
3269 if ((syntax & RE_SHY_GROUPS) && *p == '?')
3270 {
3271 PATFETCH (c); /* Gobble up the '?'. */
3272 while (!shy)
3273 {
3274 PATFETCH (c);
3275 switch (c)
3276 {
3277 case ':': shy = 1; break;
3278 case '0':
3279 /* An explicitly specified regnum must start
3280 with non-0. */
3281 if (regnum == 0)
3282 FREE_STACK_RETURN (REG_BADPAT);
3283 case '1': case '2': case '3': case '4':
3284 case '5': case '6': case '7': case '8': case '9':
3285 regnum = 10*regnum + (c - '0'); break;
3286 default:
3287 /* Only (?:...) is supported right now. */
3288 FREE_STACK_RETURN (REG_BADPAT);
3289 }
3290 }
3291 }
3292 }
3293
3294 if (!shy)
3295 regnum = ++bufp->re_nsub;
3296 else if (regnum)
3297 { /* It's actually not shy, but explicitly numbered. */
3298 shy = 0;
3299 if (regnum > bufp->re_nsub)
3300 bufp->re_nsub = regnum;
3301 else if (regnum > bufp->re_nsub
3302 /* Ideally, we'd want to check that the specified
3303 group can't have matched (i.e. all subgroups
3304 using the same regnum are in other branches of
3305 OR patterns), but we don't currently keep track
3306 of enough info to do that easily. */
3307 || group_in_compile_stack (compile_stack, regnum))
3308 FREE_STACK_RETURN (REG_BADPAT);
3309 }
3310 else
3311 /* It's really shy. */
3312 regnum = - bufp->re_nsub;
3313
3314 if (COMPILE_STACK_FULL)
3315 {
3316 RETALLOC (compile_stack.stack, compile_stack.size << 1,
3317 compile_stack_elt_t);
3318 if (compile_stack.stack == NULL) return REG_ESPACE;
3319
3320 compile_stack.size <<= 1;
3321 }
3322
3323 /* These are the values to restore when we hit end of this
3324 group. They are all relative offsets, so that if the
3325 whole pattern moves because of realloc, they will still
3326 be valid. */
3327 COMPILE_STACK_TOP.begalt_offset = begalt - bufp->buffer;
3328 COMPILE_STACK_TOP.fixup_alt_jump
3329 = fixup_alt_jump ? fixup_alt_jump - bufp->buffer + 1 : 0;
3330 COMPILE_STACK_TOP.laststart_offset = b - bufp->buffer;
3331 COMPILE_STACK_TOP.regnum = regnum;
3332
3333 /* Do not push a start_memory for groups beyond the last one
3334 we can represent in the compiled pattern. */
3335 if (regnum <= MAX_REGNUM && regnum > 0)
3336 BUF_PUSH_2 (start_memory, regnum);
3337
3338 compile_stack.avail++;
3339
3340 fixup_alt_jump = 0;
3341 laststart = 0;
3342 begalt = b;
3343 /* If we've reached MAX_REGNUM groups, then this open
3344 won't actually generate any code, so we'll have to
3345 clear pending_exact explicitly. */
3346 pending_exact = 0;
3347 break;
3348 }
3349
3350 case ')':
3351 if (syntax & RE_NO_BK_PARENS) goto normal_backslash;
3352
3353 if (COMPILE_STACK_EMPTY)
3354 {
3355 if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD)
3356 goto normal_backslash;
3357 else
3358 FREE_STACK_RETURN (REG_ERPAREN);
3359 }
3360
3361 handle_close:
3362 FIXUP_ALT_JUMP ();
3363
3364 /* See similar code for backslashed left paren above. */
3365 if (COMPILE_STACK_EMPTY)
3366 {
3367 if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD)
3368 goto normal_char;
3369 else
3370 FREE_STACK_RETURN (REG_ERPAREN);
3371 }
3372
3373 /* Since we just checked for an empty stack above, this
3374 ``can't happen''. */
3375 assert (compile_stack.avail != 0);
3376 {
3377 /* We don't just want to restore into `regnum', because
3378 later groups should continue to be numbered higher,
3379 as in `(ab)c(de)' -- the second group is #2. */
3380 regnum_t regnum;
3381
3382 compile_stack.avail--;
3383 begalt = bufp->buffer + COMPILE_STACK_TOP.begalt_offset;
3384 fixup_alt_jump
3385 = COMPILE_STACK_TOP.fixup_alt_jump
3386 ? bufp->buffer + COMPILE_STACK_TOP.fixup_alt_jump - 1
3387 : 0;
3388 laststart = bufp->buffer + COMPILE_STACK_TOP.laststart_offset;
3389 regnum = COMPILE_STACK_TOP.regnum;
3390 /* If we've reached MAX_REGNUM groups, then this open
3391 won't actually generate any code, so we'll have to
3392 clear pending_exact explicitly. */
3393 pending_exact = 0;
3394
3395 /* We're at the end of the group, so now we know how many
3396 groups were inside this one. */
3397 if (regnum <= MAX_REGNUM && regnum > 0)
3398 BUF_PUSH_2 (stop_memory, regnum);
3399 }
3400 break;
3401
3402
3403 case '|': /* `\|'. */
3404 if (syntax & RE_LIMITED_OPS || syntax & RE_NO_BK_VBAR)
3405 goto normal_backslash;
3406 handle_alt:
3407 if (syntax & RE_LIMITED_OPS)
3408 goto normal_char;
3409
3410 /* Insert before the previous alternative a jump which
3411 jumps to this alternative if the former fails. */
3412 GET_BUFFER_SPACE (3);
3413 INSERT_JUMP (on_failure_jump, begalt, b + 6);
3414 pending_exact = 0;
3415 b += 3;
3416
3417 /* The alternative before this one has a jump after it
3418 which gets executed if it gets matched. Adjust that
3419 jump so it will jump to this alternative's analogous
3420 jump (put in below, which in turn will jump to the next
3421 (if any) alternative's such jump, etc.). The last such
3422 jump jumps to the correct final destination. A picture:
3423 _____ _____
3424 | | | |
3425 | v | v
3426 a | b | c
3427
3428 If we are at `b', then fixup_alt_jump right now points to a
3429 three-byte space after `a'. We'll put in the jump, set
3430 fixup_alt_jump to right after `b', and leave behind three
3431 bytes which we'll fill in when we get to after `c'. */
3432
3433 FIXUP_ALT_JUMP ();
3434
3435 /* Mark and leave space for a jump after this alternative,
3436 to be filled in later either by next alternative or
3437 when know we're at the end of a series of alternatives. */
3438 fixup_alt_jump = b;
3439 GET_BUFFER_SPACE (3);
3440 b += 3;
3441
3442 laststart = 0;
3443 begalt = b;
3444 break;
3445
3446
3447 case '{':
3448 /* If \{ is a literal. */
3449 if (!(syntax & RE_INTERVALS)
3450 /* If we're at `\{' and it's not the open-interval
3451 operator. */
3452 || (syntax & RE_NO_BK_BRACES))
3453 goto normal_backslash;
3454
3455 handle_interval:
3456 {
3457 /* If got here, then the syntax allows intervals. */
3458
3459 /* At least (most) this many matches must be made. */
3460 int lower_bound = 0, upper_bound = -1;
3461
3462 beg_interval = p;
3463
3464 GET_UNSIGNED_NUMBER (lower_bound);
3465
3466 if (c == ',')
3467 GET_UNSIGNED_NUMBER (upper_bound);
3468 else
3469 /* Interval such as `{1}' => match exactly once. */
3470 upper_bound = lower_bound;
3471
3472 if (lower_bound < 0 || upper_bound > RE_DUP_MAX
3473 || (upper_bound >= 0 && lower_bound > upper_bound))
3474 FREE_STACK_RETURN (REG_BADBR);
3475
3476 if (!(syntax & RE_NO_BK_BRACES))
3477 {
3478 if (c != '\\')
3479 FREE_STACK_RETURN (REG_BADBR);
3480 if (p == pend)
3481 FREE_STACK_RETURN (REG_EESCAPE);
3482 PATFETCH (c);
3483 }
3484
3485 if (c != '}')
3486 FREE_STACK_RETURN (REG_BADBR);
3487
3488 /* We just parsed a valid interval. */
3489
3490 /* If it's invalid to have no preceding re. */
3491 if (!laststart)
3492 {
3493 if (syntax & RE_CONTEXT_INVALID_OPS)
3494 FREE_STACK_RETURN (REG_BADRPT);
3495 else if (syntax & RE_CONTEXT_INDEP_OPS)
3496 laststart = b;
3497 else
3498 goto unfetch_interval;
3499 }
3500
3501 if (upper_bound == 0)
3502 /* If the upper bound is zero, just drop the sub pattern
3503 altogether. */
3504 b = laststart;
3505 else if (lower_bound == 1 && upper_bound == 1)
3506 /* Just match it once: nothing to do here. */
3507 ;
3508
3509 /* Otherwise, we have a nontrivial interval. When
3510 we're all done, the pattern will look like:
3511 set_number_at <jump count> <upper bound>
3512 set_number_at <succeed_n count> <lower bound>
3513 succeed_n <after jump addr> <succeed_n count>
3514 <body of loop>
3515 jump_n <succeed_n addr> <jump count>
3516 (The upper bound and `jump_n' are omitted if
3517 `upper_bound' is 1, though.) */
3518 else
3519 { /* If the upper bound is > 1, we need to insert
3520 more at the end of the loop. */
3521 unsigned int nbytes = (upper_bound < 0 ? 3
3522 : upper_bound > 1 ? 5 : 0);
3523 unsigned int startoffset = 0;
3524
3525 GET_BUFFER_SPACE (20); /* We might use less. */
3526
3527 if (lower_bound == 0)
3528 {
3529 /* A succeed_n that starts with 0 is really a
3530 a simple on_failure_jump_loop. */
3531 INSERT_JUMP (on_failure_jump_loop, laststart,
3532 b + 3 + nbytes);
3533 b += 3;
3534 }
3535 else
3536 {
3537 /* Initialize lower bound of the `succeed_n', even
3538 though it will be set during matching by its
3539 attendant `set_number_at' (inserted next),
3540 because `re_compile_fastmap' needs to know.
3541 Jump to the `jump_n' we might insert below. */
3542 INSERT_JUMP2 (succeed_n, laststart,
3543 b + 5 + nbytes,
3544 lower_bound);
3545 b += 5;
3546
3547 /* Code to initialize the lower bound. Insert
3548 before the `succeed_n'. The `5' is the last two
3549 bytes of this `set_number_at', plus 3 bytes of
3550 the following `succeed_n'. */
3551 insert_op2 (set_number_at, laststart, 5, lower_bound, b);
3552 b += 5;
3553 startoffset += 5;
3554 }
3555
3556 if (upper_bound < 0)
3557 {
3558 /* A negative upper bound stands for infinity,
3559 in which case it degenerates to a plain jump. */
3560 STORE_JUMP (jump, b, laststart + startoffset);
3561 b += 3;
3562 }
3563 else if (upper_bound > 1)
3564 { /* More than one repetition is allowed, so
3565 append a backward jump to the `succeed_n'
3566 that starts this interval.
3567
3568 When we've reached this during matching,
3569 we'll have matched the interval once, so
3570 jump back only `upper_bound - 1' times. */
3571 STORE_JUMP2 (jump_n, b, laststart + startoffset,
3572 upper_bound - 1);
3573 b += 5;
3574
3575 /* The location we want to set is the second
3576 parameter of the `jump_n'; that is `b-2' as
3577 an absolute address. `laststart' will be
3578 the `set_number_at' we're about to insert;
3579 `laststart+3' the number to set, the source
3580 for the relative address. But we are
3581 inserting into the middle of the pattern --
3582 so everything is getting moved up by 5.
3583 Conclusion: (b - 2) - (laststart + 3) + 5,
3584 i.e., b - laststart.
3585
3586 We insert this at the beginning of the loop
3587 so that if we fail during matching, we'll
3588 reinitialize the bounds. */
3589 insert_op2 (set_number_at, laststart, b - laststart,
3590 upper_bound - 1, b);
3591 b += 5;
3592 }
3593 }
3594 pending_exact = 0;
3595 beg_interval = NULL;
3596 }
3597 break;
3598
3599 unfetch_interval:
3600 /* If an invalid interval, match the characters as literals. */
3601 assert (beg_interval);
3602 p = beg_interval;
3603 beg_interval = NULL;
3604
3605 /* normal_char and normal_backslash need `c'. */
3606 c = '{';
3607
3608 if (!(syntax & RE_NO_BK_BRACES))
3609 {
3610 assert (p > pattern && p[-1] == '\\');
3611 goto normal_backslash;
3612 }
3613 else
3614 goto normal_char;
3615
3616 #ifdef emacs
3617 /* There is no way to specify the before_dot and after_dot
3618 operators. rms says this is ok. --karl */
3619 case '=':
3620 BUF_PUSH (at_dot);
3621 break;
3622
3623 case 's':
3624 laststart = b;
3625 PATFETCH (c);
3626 BUF_PUSH_2 (syntaxspec, syntax_spec_code[c]);
3627 break;
3628
3629 case 'S':
3630 laststart = b;
3631 PATFETCH (c);
3632 BUF_PUSH_2 (notsyntaxspec, syntax_spec_code[c]);
3633 break;
3634
3635 case 'c':
3636 laststart = b;
3637 PATFETCH (c);
3638 BUF_PUSH_2 (categoryspec, c);
3639 break;
3640
3641 case 'C':
3642 laststart = b;
3643 PATFETCH (c);
3644 BUF_PUSH_2 (notcategoryspec, c);
3645 break;
3646 #endif /* emacs */
3647
3648
3649 case 'w':
3650 if (syntax & RE_NO_GNU_OPS)
3651 goto normal_char;
3652 laststart = b;
3653 BUF_PUSH_2 (syntaxspec, Sword);
3654 break;
3655
3656
3657 case 'W':
3658 if (syntax & RE_NO_GNU_OPS)
3659 goto normal_char;
3660 laststart = b;
3661 BUF_PUSH_2 (notsyntaxspec, Sword);
3662 break;
3663
3664
3665 case '<':
3666 if (syntax & RE_NO_GNU_OPS)
3667 goto normal_char;
3668 BUF_PUSH (wordbeg);
3669 break;
3670
3671 case '>':
3672 if (syntax & RE_NO_GNU_OPS)
3673 goto normal_char;
3674 BUF_PUSH (wordend);
3675 break;
3676
3677 case '_':
3678 if (syntax & RE_NO_GNU_OPS)
3679 goto normal_char;
3680 laststart = b;
3681 PATFETCH (c);
3682 if (c == '<')
3683 BUF_PUSH (symbeg);
3684 else if (c == '>')
3685 BUF_PUSH (symend);
3686 else
3687 FREE_STACK_RETURN (REG_BADPAT);
3688 break;
3689
3690 case 'b':
3691 if (syntax & RE_NO_GNU_OPS)
3692 goto normal_char;
3693 BUF_PUSH (wordbound);
3694 break;
3695
3696 case 'B':
3697 if (syntax & RE_NO_GNU_OPS)
3698 goto normal_char;
3699 BUF_PUSH (notwordbound);
3700 break;
3701
3702 case '`':
3703 if (syntax & RE_NO_GNU_OPS)
3704 goto normal_char;
3705 BUF_PUSH (begbuf);
3706 break;
3707
3708 case '\'':
3709 if (syntax & RE_NO_GNU_OPS)
3710 goto normal_char;
3711 BUF_PUSH (endbuf);
3712 break;
3713
3714 case '1': case '2': case '3': case '4': case '5':
3715 case '6': case '7': case '8': case '9':
3716 {
3717 regnum_t reg;
3718
3719 if (syntax & RE_NO_BK_REFS)
3720 goto normal_backslash;
3721
3722 reg = c - '0';
3723
3724 if (reg > bufp->re_nsub || reg < 1
3725 /* Can't back reference to a subexp before its end. */
3726 || group_in_compile_stack (compile_stack, reg))
3727 FREE_STACK_RETURN (REG_ESUBREG);
3728
3729 laststart = b;
3730 BUF_PUSH_2 (duplicate, reg);
3731 }
3732 break;
3733
3734
3735 case '+':
3736 case '?':
3737 if (syntax & RE_BK_PLUS_QM)
3738 goto handle_plus;
3739 else
3740 goto normal_backslash;
3741
3742 default:
3743 normal_backslash:
3744 /* You might think it would be useful for \ to mean
3745 not to translate; but if we don't translate it
3746 it will never match anything. */
3747 goto normal_char;
3748 }
3749 break;
3750
3751
3752 default:
3753 /* Expects the character in `c'. */
3754 normal_char:
3755 /* If no exactn currently being built. */
3756 if (!pending_exact
3757
3758 /* If last exactn not at current position. */
3759 || pending_exact + *pending_exact + 1 != b
3760
3761 /* We have only one byte following the exactn for the count. */
3762 || *pending_exact >= (1 << BYTEWIDTH) - MAX_MULTIBYTE_LENGTH
3763
3764 /* If followed by a repetition operator. */
3765 || (p != pend && (*p == '*' || *p == '^'))
3766 || ((syntax & RE_BK_PLUS_QM)
3767 ? p + 1 < pend && *p == '\\' && (p[1] == '+' || p[1] == '?')
3768 : p != pend && (*p == '+' || *p == '?'))
3769 || ((syntax & RE_INTERVALS)
3770 && ((syntax & RE_NO_BK_BRACES)
3771 ? p != pend && *p == '{'
3772 : p + 1 < pend && p[0] == '\\' && p[1] == '{')))
3773 {
3774 /* Start building a new exactn. */
3775
3776 laststart = b;
3777
3778 BUF_PUSH_2 (exactn, 0);
3779 pending_exact = b - 1;
3780 }
3781
3782 GET_BUFFER_SPACE (MAX_MULTIBYTE_LENGTH);
3783 {
3784 int len;
3785
3786 if (multibyte)
3787 {
3788 c = TRANSLATE (c);
3789 len = CHAR_STRING (c, b);
3790 b += len;
3791 }
3792 else
3793 {
3794 c1 = RE_CHAR_TO_MULTIBYTE (c);
3795 if (! CHAR_BYTE8_P (c1))
3796 {
3797 re_wchar_t c2 = TRANSLATE (c1);
3798
3799 if (c1 != c2 && (c1 = RE_CHAR_TO_UNIBYTE (c2)) >= 0)
3800 c = c1;
3801 }
3802 *b++ = c;
3803 len = 1;
3804 }
3805 (*pending_exact) += len;
3806 }
3807
3808 break;
3809 } /* switch (c) */
3810 } /* while p != pend */
3811
3812
3813 /* Through the pattern now. */
3814
3815 FIXUP_ALT_JUMP ();
3816
3817 if (!COMPILE_STACK_EMPTY)
3818 FREE_STACK_RETURN (REG_EPAREN);
3819
3820 /* If we don't want backtracking, force success
3821 the first time we reach the end of the compiled pattern. */
3822 if (syntax & RE_NO_POSIX_BACKTRACKING)
3823 BUF_PUSH (succeed);
3824
3825 /* We have succeeded; set the length of the buffer. */
3826 bufp->used = b - bufp->buffer;
3827
3828 #ifdef DEBUG
3829 if (debug > 0)
3830 {
3831 re_compile_fastmap (bufp);
3832 DEBUG_PRINT1 ("\nCompiled pattern: \n");
3833 print_compiled_pattern (bufp);
3834 }
3835 debug--;
3836 #endif /* DEBUG */
3837
3838 #ifndef MATCH_MAY_ALLOCATE
3839 /* Initialize the failure stack to the largest possible stack. This
3840 isn't necessary unless we're trying to avoid calling alloca in
3841 the search and match routines. */
3842 {
3843 int num_regs = bufp->re_nsub + 1;
3844
3845 if (fail_stack.size < re_max_failures * TYPICAL_FAILURE_SIZE)
3846 {
3847 fail_stack.size = re_max_failures * TYPICAL_FAILURE_SIZE;
3848
3849 if (! fail_stack.stack)
3850 fail_stack.stack
3851 = (fail_stack_elt_t *) malloc (fail_stack.size
3852 * sizeof (fail_stack_elt_t));
3853 else
3854 fail_stack.stack
3855 = (fail_stack_elt_t *) realloc (fail_stack.stack,
3856 (fail_stack.size
3857 * sizeof (fail_stack_elt_t)));
3858 }
3859
3860 regex_grow_registers (num_regs);
3861 }
3862 #endif /* not MATCH_MAY_ALLOCATE */
3863
3864 FREE_STACK_RETURN (REG_NOERROR);
3865 } /* regex_compile */
3866 \f
3867 /* Subroutines for `regex_compile'. */
3868
3869 /* Store OP at LOC followed by two-byte integer parameter ARG. */
3870
3871 static void
3872 store_op1 (op, loc, arg)
3873 re_opcode_t op;
3874 unsigned char *loc;
3875 int arg;
3876 {
3877 *loc = (unsigned char) op;
3878 STORE_NUMBER (loc + 1, arg);
3879 }
3880
3881
3882 /* Like `store_op1', but for two two-byte parameters ARG1 and ARG2. */
3883
3884 static void
3885 store_op2 (op, loc, arg1, arg2)
3886 re_opcode_t op;
3887 unsigned char *loc;
3888 int arg1, arg2;
3889 {
3890 *loc = (unsigned char) op;
3891 STORE_NUMBER (loc + 1, arg1);
3892 STORE_NUMBER (loc + 3, arg2);
3893 }
3894
3895
3896 /* Copy the bytes from LOC to END to open up three bytes of space at LOC
3897 for OP followed by two-byte integer parameter ARG. */
3898
3899 static void
3900 insert_op1 (op, loc, arg, end)
3901 re_opcode_t op;
3902 unsigned char *loc;
3903 int arg;
3904 unsigned char *end;
3905 {
3906 register unsigned char *pfrom = end;
3907 register unsigned char *pto = end + 3;
3908
3909 while (pfrom != loc)
3910 *--pto = *--pfrom;
3911
3912 store_op1 (op, loc, arg);
3913 }
3914
3915
3916 /* Like `insert_op1', but for two two-byte parameters ARG1 and ARG2. */
3917
3918 static void
3919 insert_op2 (op, loc, arg1, arg2, end)
3920 re_opcode_t op;
3921 unsigned char *loc;
3922 int arg1, arg2;
3923 unsigned char *end;
3924 {
3925 register unsigned char *pfrom = end;
3926 register unsigned char *pto = end + 5;
3927
3928 while (pfrom != loc)
3929 *--pto = *--pfrom;
3930
3931 store_op2 (op, loc, arg1, arg2);
3932 }
3933
3934
3935 /* P points to just after a ^ in PATTERN. Return true if that ^ comes
3936 after an alternative or a begin-subexpression. We assume there is at
3937 least one character before the ^. */
3938
3939 static boolean
3940 at_begline_loc_p (pattern, p, syntax)
3941 re_char *pattern, *p;
3942 reg_syntax_t syntax;
3943 {
3944 re_char *prev = p - 2;
3945 boolean prev_prev_backslash = prev > pattern && prev[-1] == '\\';
3946
3947 return
3948 /* After a subexpression? */
3949 (*prev == '(' && (syntax & RE_NO_BK_PARENS || prev_prev_backslash))
3950 /* After an alternative? */
3951 || (*prev == '|' && (syntax & RE_NO_BK_VBAR || prev_prev_backslash))
3952 /* After a shy subexpression? */
3953 || ((syntax & RE_SHY_GROUPS) && prev - 2 >= pattern
3954 && prev[-1] == '?' && prev[-2] == '('
3955 && (syntax & RE_NO_BK_PARENS
3956 || (prev - 3 >= pattern && prev[-3] == '\\')));
3957 }
3958
3959
3960 /* The dual of at_begline_loc_p. This one is for $. We assume there is
3961 at least one character after the $, i.e., `P < PEND'. */
3962
3963 static boolean
3964 at_endline_loc_p (p, pend, syntax)
3965 re_char *p, *pend;
3966 reg_syntax_t syntax;
3967 {
3968 re_char *next = p;
3969 boolean next_backslash = *next == '\\';
3970 re_char *next_next = p + 1 < pend ? p + 1 : 0;
3971
3972 return
3973 /* Before a subexpression? */
3974 (syntax & RE_NO_BK_PARENS ? *next == ')'
3975 : next_backslash && next_next && *next_next == ')')
3976 /* Before an alternative? */
3977 || (syntax & RE_NO_BK_VBAR ? *next == '|'
3978 : next_backslash && next_next && *next_next == '|');
3979 }
3980
3981
3982 /* Returns true if REGNUM is in one of COMPILE_STACK's elements and
3983 false if it's not. */
3984
3985 static boolean
3986 group_in_compile_stack (compile_stack, regnum)
3987 compile_stack_type compile_stack;
3988 regnum_t regnum;
3989 {
3990 int this_element;
3991
3992 for (this_element = compile_stack.avail - 1;
3993 this_element >= 0;
3994 this_element--)
3995 if (compile_stack.stack[this_element].regnum == regnum)
3996 return true;
3997
3998 return false;
3999 }
4000 \f
4001 /* analyse_first.
4002 If fastmap is non-NULL, go through the pattern and fill fastmap
4003 with all the possible leading chars. If fastmap is NULL, don't
4004 bother filling it up (obviously) and only return whether the
4005 pattern could potentially match the empty string.
4006
4007 Return 1 if p..pend might match the empty string.
4008 Return 0 if p..pend matches at least one char.
4009 Return -1 if fastmap was not updated accurately. */
4010
4011 static int
4012 analyse_first (p, pend, fastmap, multibyte)
4013 re_char *p, *pend;
4014 char *fastmap;
4015 const int multibyte;
4016 {
4017 int j, k;
4018 boolean not;
4019
4020 /* If all elements for base leading-codes in fastmap is set, this
4021 flag is set true. */
4022 boolean match_any_multibyte_characters = false;
4023
4024 assert (p);
4025
4026 /* The loop below works as follows:
4027 - It has a working-list kept in the PATTERN_STACK and which basically
4028 starts by only containing a pointer to the first operation.
4029 - If the opcode we're looking at is a match against some set of
4030 chars, then we add those chars to the fastmap and go on to the
4031 next work element from the worklist (done via `break').
4032 - If the opcode is a control operator on the other hand, we either
4033 ignore it (if it's meaningless at this point, such as `start_memory')
4034 or execute it (if it's a jump). If the jump has several destinations
4035 (i.e. `on_failure_jump'), then we push the other destination onto the
4036 worklist.
4037 We guarantee termination by ignoring backward jumps (more or less),
4038 so that `p' is monotonically increasing. More to the point, we
4039 never set `p' (or push) anything `<= p1'. */
4040
4041 while (p < pend)
4042 {
4043 /* `p1' is used as a marker of how far back a `on_failure_jump'
4044 can go without being ignored. It is normally equal to `p'
4045 (which prevents any backward `on_failure_jump') except right
4046 after a plain `jump', to allow patterns such as:
4047 0: jump 10
4048 3..9: <body>
4049 10: on_failure_jump 3
4050 as used for the *? operator. */
4051 re_char *p1 = p;
4052
4053 switch (SWITCH_ENUM_CAST ((re_opcode_t) *p++))
4054 {
4055 case succeed:
4056 return 1;
4057 continue;
4058
4059 case duplicate:
4060 /* If the first character has to match a backreference, that means
4061 that the group was empty (since it already matched). Since this
4062 is the only case that interests us here, we can assume that the
4063 backreference must match the empty string. */
4064 p++;
4065 continue;
4066
4067
4068 /* Following are the cases which match a character. These end
4069 with `break'. */
4070
4071 case exactn:
4072 if (fastmap)
4073 {
4074 /* If multibyte is nonzero, the first byte of each
4075 character is an ASCII or a leading code. Otherwise,
4076 each byte is a character. Thus, this works in both
4077 cases. */
4078 fastmap[p[1]] = 1;
4079 if (! multibyte)
4080 {
4081 /* For the case of matching this unibyte regex
4082 against multibyte, we must set a leading code of
4083 the corresponding multibyte character. */
4084 int c = RE_CHAR_TO_MULTIBYTE (p[1]);
4085
4086 if (! CHAR_BYTE8_P (c))
4087 fastmap[CHAR_LEADING_CODE (c)] = 1;
4088 }
4089 }
4090 break;
4091
4092
4093 case anychar:
4094 /* We could put all the chars except for \n (and maybe \0)
4095 but we don't bother since it is generally not worth it. */
4096 if (!fastmap) break;
4097 return -1;
4098
4099
4100 case charset_not:
4101 if (!fastmap) break;
4102 {
4103 /* Chars beyond end of bitmap are possible matches. */
4104 for (j = CHARSET_BITMAP_SIZE (&p[-1]) * BYTEWIDTH;
4105 j < (1 << BYTEWIDTH); j++)
4106 fastmap[j] = 1;
4107 }
4108
4109 /* Fallthrough */
4110 case charset:
4111 if (!fastmap) break;
4112 not = (re_opcode_t) *(p - 1) == charset_not;
4113 for (j = CHARSET_BITMAP_SIZE (&p[-1]) * BYTEWIDTH - 1, p++;
4114 j >= 0; j--)
4115 if (!!(p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH))) ^ not)
4116 fastmap[j] = 1;
4117
4118 #ifdef emacs
4119 if (/* Any leading code can possibly start a character
4120 which doesn't match the specified set of characters. */
4121 not
4122 ||
4123 /* If we can match a character class, we can match any
4124 multibyte characters. */
4125 (CHARSET_RANGE_TABLE_EXISTS_P (&p[-2])
4126 && CHARSET_RANGE_TABLE_BITS (&p[-2]) != 0))
4127
4128 {
4129 if (match_any_multibyte_characters == false)
4130 {
4131 for (j = MIN_MULTIBYTE_LEADING_CODE;
4132 j <= MAX_MULTIBYTE_LEADING_CODE; j++)
4133 fastmap[j] = 1;
4134 match_any_multibyte_characters = true;
4135 }
4136 }
4137
4138 else if (!not && CHARSET_RANGE_TABLE_EXISTS_P (&p[-2])
4139 && match_any_multibyte_characters == false)
4140 {
4141 /* Set fastmap[I] to 1 where I is a leading code of each
4142 multibyte characer in the range table. */
4143 int c, count;
4144 unsigned char lc1, lc2;
4145
4146 /* Make P points the range table. `+ 2' is to skip flag
4147 bits for a character class. */
4148 p += CHARSET_BITMAP_SIZE (&p[-2]) + 2;
4149
4150 /* Extract the number of ranges in range table into COUNT. */
4151 EXTRACT_NUMBER_AND_INCR (count, p);
4152 for (; count > 0; count--, p += 3)
4153 {
4154 /* Extract the start and end of each range. */
4155 EXTRACT_CHARACTER (c, p);
4156 lc1 = CHAR_LEADING_CODE (c);
4157 p += 3;
4158 EXTRACT_CHARACTER (c, p);
4159 lc2 = CHAR_LEADING_CODE (c);
4160 for (j = lc1; j <= lc2; j++)
4161 fastmap[j] = 1;
4162 }
4163 }
4164 #endif
4165 break;
4166
4167 case syntaxspec:
4168 case notsyntaxspec:
4169 if (!fastmap) break;
4170 #ifndef emacs
4171 not = (re_opcode_t)p[-1] == notsyntaxspec;
4172 k = *p++;
4173 for (j = 0; j < (1 << BYTEWIDTH); j++)
4174 if ((SYNTAX (j) == (enum syntaxcode) k) ^ not)
4175 fastmap[j] = 1;
4176 break;
4177 #else /* emacs */
4178 /* This match depends on text properties. These end with
4179 aborting optimizations. */
4180 return -1;
4181
4182 case categoryspec:
4183 case notcategoryspec:
4184 if (!fastmap) break;
4185 not = (re_opcode_t)p[-1] == notcategoryspec;
4186 k = *p++;
4187 for (j = (1 << BYTEWIDTH); j >= 0; j--)
4188 if ((CHAR_HAS_CATEGORY (j, k)) ^ not)
4189 fastmap[j] = 1;
4190
4191 /* Any leading code can possibly start a character which
4192 has or doesn't has the specified category. */
4193 if (match_any_multibyte_characters == false)
4194 {
4195 for (j = MIN_MULTIBYTE_LEADING_CODE;
4196 j <= MAX_MULTIBYTE_LEADING_CODE; j++)
4197 fastmap[j] = 1;
4198 match_any_multibyte_characters = true;
4199 }
4200 break;
4201
4202 /* All cases after this match the empty string. These end with
4203 `continue'. */
4204
4205 case before_dot:
4206 case at_dot:
4207 case after_dot:
4208 #endif /* !emacs */
4209 case no_op:
4210 case begline:
4211 case endline:
4212 case begbuf:
4213 case endbuf:
4214 case wordbound:
4215 case notwordbound:
4216 case wordbeg:
4217 case wordend:
4218 case symbeg:
4219 case symend:
4220 continue;
4221
4222
4223 case jump:
4224 EXTRACT_NUMBER_AND_INCR (j, p);
4225 if (j < 0)
4226 /* Backward jumps can only go back to code that we've already
4227 visited. `re_compile' should make sure this is true. */
4228 break;
4229 p += j;
4230 switch (SWITCH_ENUM_CAST ((re_opcode_t) *p))
4231 {
4232 case on_failure_jump:
4233 case on_failure_keep_string_jump:
4234 case on_failure_jump_loop:
4235 case on_failure_jump_nastyloop:
4236 case on_failure_jump_smart:
4237 p++;
4238 break;
4239 default:
4240 continue;
4241 };
4242 /* Keep `p1' to allow the `on_failure_jump' we are jumping to
4243 to jump back to "just after here". */
4244 /* Fallthrough */
4245
4246 case on_failure_jump:
4247 case on_failure_keep_string_jump:
4248 case on_failure_jump_nastyloop:
4249 case on_failure_jump_loop:
4250 case on_failure_jump_smart:
4251 EXTRACT_NUMBER_AND_INCR (j, p);
4252 if (p + j <= p1)
4253 ; /* Backward jump to be ignored. */
4254 else
4255 { /* We have to look down both arms.
4256 We first go down the "straight" path so as to minimize
4257 stack usage when going through alternatives. */
4258 int r = analyse_first (p, pend, fastmap, multibyte);
4259 if (r) return r;
4260 p += j;
4261 }
4262 continue;
4263
4264
4265 case jump_n:
4266 /* This code simply does not properly handle forward jump_n. */
4267 DEBUG_STATEMENT (EXTRACT_NUMBER (j, p); assert (j < 0));
4268 p += 4;
4269 /* jump_n can either jump or fall through. The (backward) jump
4270 case has already been handled, so we only need to look at the
4271 fallthrough case. */
4272 continue;
4273
4274 case succeed_n:
4275 /* If N == 0, it should be an on_failure_jump_loop instead. */
4276 DEBUG_STATEMENT (EXTRACT_NUMBER (j, p + 2); assert (j > 0));
4277 p += 4;
4278 /* We only care about one iteration of the loop, so we don't
4279 need to consider the case where this behaves like an
4280 on_failure_jump. */
4281 continue;
4282
4283
4284 case set_number_at:
4285 p += 4;
4286 continue;
4287
4288
4289 case start_memory:
4290 case stop_memory:
4291 p += 1;
4292 continue;
4293
4294
4295 default:
4296 abort (); /* We have listed all the cases. */
4297 } /* switch *p++ */
4298
4299 /* Getting here means we have found the possible starting
4300 characters for one path of the pattern -- and that the empty
4301 string does not match. We need not follow this path further. */
4302 return 0;
4303 } /* while p */
4304
4305 /* We reached the end without matching anything. */
4306 return 1;
4307
4308 } /* analyse_first */
4309 \f
4310 /* re_compile_fastmap computes a ``fastmap'' for the compiled pattern in
4311 BUFP. A fastmap records which of the (1 << BYTEWIDTH) possible
4312 characters can start a string that matches the pattern. This fastmap
4313 is used by re_search to skip quickly over impossible starting points.
4314
4315 Character codes above (1 << BYTEWIDTH) are not represented in the
4316 fastmap, but the leading codes are represented. Thus, the fastmap
4317 indicates which character sets could start a match.
4318
4319 The caller must supply the address of a (1 << BYTEWIDTH)-byte data
4320 area as BUFP->fastmap.
4321
4322 We set the `fastmap', `fastmap_accurate', and `can_be_null' fields in
4323 the pattern buffer.
4324
4325 Returns 0 if we succeed, -2 if an internal error. */
4326
4327 int
4328 re_compile_fastmap (bufp)
4329 struct re_pattern_buffer *bufp;
4330 {
4331 char *fastmap = bufp->fastmap;
4332 int analysis;
4333
4334 assert (fastmap && bufp->buffer);
4335
4336 bzero (fastmap, 1 << BYTEWIDTH); /* Assume nothing's valid. */
4337 bufp->fastmap_accurate = 1; /* It will be when we're done. */
4338
4339 analysis = analyse_first (bufp->buffer, bufp->buffer + bufp->used,
4340 fastmap, RE_MULTIBYTE_P (bufp));
4341 bufp->can_be_null = (analysis != 0);
4342 return 0;
4343 } /* re_compile_fastmap */
4344 \f
4345 /* Set REGS to hold NUM_REGS registers, storing them in STARTS and
4346 ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use
4347 this memory for recording register information. STARTS and ENDS
4348 must be allocated using the malloc library routine, and must each
4349 be at least NUM_REGS * sizeof (regoff_t) bytes long.
4350
4351 If NUM_REGS == 0, then subsequent matches should allocate their own
4352 register data.
4353
4354 Unless this function is called, the first search or match using
4355 PATTERN_BUFFER will allocate its own register data, without
4356 freeing the old data. */
4357
4358 void
4359 re_set_registers (bufp, regs, num_regs, starts, ends)
4360 struct re_pattern_buffer *bufp;
4361 struct re_registers *regs;
4362 unsigned num_regs;
4363 regoff_t *starts, *ends;
4364 {
4365 if (num_regs)
4366 {
4367 bufp->regs_allocated = REGS_REALLOCATE;
4368 regs->num_regs = num_regs;
4369 regs->start = starts;
4370 regs->end = ends;
4371 }
4372 else
4373 {
4374 bufp->regs_allocated = REGS_UNALLOCATED;
4375 regs->num_regs = 0;
4376 regs->start = regs->end = (regoff_t *) 0;
4377 }
4378 }
4379 WEAK_ALIAS (__re_set_registers, re_set_registers)
4380 \f
4381 /* Searching routines. */
4382
4383 /* Like re_search_2, below, but only one string is specified, and
4384 doesn't let you say where to stop matching. */
4385
4386 int
4387 re_search (bufp, string, size, startpos, range, regs)
4388 struct re_pattern_buffer *bufp;
4389 const char *string;
4390 int size, startpos, range;
4391 struct re_registers *regs;
4392 {
4393 return re_search_2 (bufp, NULL, 0, string, size, startpos, range,
4394 regs, size);
4395 }
4396 WEAK_ALIAS (__re_search, re_search)
4397
4398 /* Head address of virtual concatenation of string. */
4399 #define HEAD_ADDR_VSTRING(P) \
4400 (((P) >= size1 ? string2 : string1))
4401
4402 /* End address of virtual concatenation of string. */
4403 #define STOP_ADDR_VSTRING(P) \
4404 (((P) >= size1 ? string2 + size2 : string1 + size1))
4405
4406 /* Address of POS in the concatenation of virtual string. */
4407 #define POS_ADDR_VSTRING(POS) \
4408 (((POS) >= size1 ? string2 - size1 : string1) + (POS))
4409
4410 /* Using the compiled pattern in BUFP->buffer, first tries to match the
4411 virtual concatenation of STRING1 and STRING2, starting first at index
4412 STARTPOS, then at STARTPOS + 1, and so on.
4413
4414 STRING1 and STRING2 have length SIZE1 and SIZE2, respectively.
4415
4416 RANGE is how far to scan while trying to match. RANGE = 0 means try
4417 only at STARTPOS; in general, the last start tried is STARTPOS +
4418 RANGE.
4419
4420 In REGS, return the indices of the virtual concatenation of STRING1
4421 and STRING2 that matched the entire BUFP->buffer and its contained
4422 subexpressions.
4423
4424 Do not consider matching one past the index STOP in the virtual
4425 concatenation of STRING1 and STRING2.
4426
4427 We return either the position in the strings at which the match was
4428 found, -1 if no match, or -2 if error (such as failure
4429 stack overflow). */
4430
4431 int
4432 re_search_2 (bufp, str1, size1, str2, size2, startpos, range, regs, stop)
4433 struct re_pattern_buffer *bufp;
4434 const char *str1, *str2;
4435 int size1, size2;
4436 int startpos;
4437 int range;
4438 struct re_registers *regs;
4439 int stop;
4440 {
4441 int val;
4442 re_char *string1 = (re_char*) str1;
4443 re_char *string2 = (re_char*) str2;
4444 register char *fastmap = bufp->fastmap;
4445 register RE_TRANSLATE_TYPE translate = bufp->translate;
4446 int total_size = size1 + size2;
4447 int endpos = startpos + range;
4448 boolean anchored_start;
4449 /* Nonzero if we are searching multibyte string. */
4450 const boolean multibyte = RE_TARGET_MULTIBYTE_P (bufp);
4451
4452 /* Check for out-of-range STARTPOS. */
4453 if (startpos < 0 || startpos > total_size)
4454 return -1;
4455
4456 /* Fix up RANGE if it might eventually take us outside
4457 the virtual concatenation of STRING1 and STRING2.
4458 Make sure we won't move STARTPOS below 0 or above TOTAL_SIZE. */
4459 if (endpos < 0)
4460 range = 0 - startpos;
4461 else if (endpos > total_size)
4462 range = total_size - startpos;
4463
4464 /* If the search isn't to be a backwards one, don't waste time in a
4465 search for a pattern anchored at beginning of buffer. */
4466 if (bufp->used > 0 && (re_opcode_t) bufp->buffer[0] == begbuf && range > 0)
4467 {
4468 if (startpos > 0)
4469 return -1;
4470 else
4471 range = 0;
4472 }
4473
4474 #ifdef emacs
4475 /* In a forward search for something that starts with \=.
4476 don't keep searching past point. */
4477 if (bufp->used > 0 && (re_opcode_t) bufp->buffer[0] == at_dot && range > 0)
4478 {
4479 range = PT_BYTE - BEGV_BYTE - startpos;
4480 if (range < 0)
4481 return -1;
4482 }
4483 #endif /* emacs */
4484
4485 /* Update the fastmap now if not correct already. */
4486 if (fastmap && !bufp->fastmap_accurate)
4487 re_compile_fastmap (bufp);
4488
4489 /* See whether the pattern is anchored. */
4490 anchored_start = (bufp->buffer[0] == begline);
4491
4492 #ifdef emacs
4493 gl_state.object = re_match_object;
4494 {
4495 int charpos = SYNTAX_TABLE_BYTE_TO_CHAR (POS_AS_IN_BUFFER (startpos));
4496
4497 SETUP_SYNTAX_TABLE_FOR_OBJECT (re_match_object, charpos, 1);
4498 }
4499 #endif
4500
4501 /* Loop through the string, looking for a place to start matching. */
4502 for (;;)
4503 {
4504 /* If the pattern is anchored,
4505 skip quickly past places we cannot match.
4506 We don't bother to treat startpos == 0 specially
4507 because that case doesn't repeat. */
4508 if (anchored_start && startpos > 0)
4509 {
4510 if (! ((startpos <= size1 ? string1[startpos - 1]
4511 : string2[startpos - size1 - 1])
4512 == '\n'))
4513 goto advance;
4514 }
4515
4516 /* If a fastmap is supplied, skip quickly over characters that
4517 cannot be the start of a match. If the pattern can match the
4518 null string, however, we don't need to skip characters; we want
4519 the first null string. */
4520 if (fastmap && startpos < total_size && !bufp->can_be_null)
4521 {
4522 register re_char *d;
4523 register re_wchar_t buf_ch;
4524
4525 d = POS_ADDR_VSTRING (startpos);
4526
4527 if (range > 0) /* Searching forwards. */
4528 {
4529 register int lim = 0;
4530 int irange = range;
4531
4532 if (startpos < size1 && startpos + range >= size1)
4533 lim = range - (size1 - startpos);
4534
4535 /* Written out as an if-else to avoid testing `translate'
4536 inside the loop. */
4537 if (RE_TRANSLATE_P (translate))
4538 {
4539 if (multibyte)
4540 while (range > lim)
4541 {
4542 int buf_charlen;
4543
4544 buf_ch = STRING_CHAR_AND_LENGTH (d, range - lim,
4545 buf_charlen);
4546 buf_ch = RE_TRANSLATE (translate, buf_ch);
4547 if (fastmap[CHAR_LEADING_CODE (buf_ch)])
4548 break;
4549
4550 range -= buf_charlen;
4551 d += buf_charlen;
4552 }
4553 else
4554 while (range > lim)
4555 {
4556 register re_wchar_t ch, translated;
4557
4558 buf_ch = *d;
4559 ch = RE_CHAR_TO_MULTIBYTE (buf_ch);
4560 translated = RE_TRANSLATE (translate, ch);
4561 if (translated != ch
4562 && (ch = RE_CHAR_TO_UNIBYTE (translated)) >= 0)
4563 buf_ch = ch;
4564 if (fastmap[buf_ch])
4565 break;
4566 d++;
4567 range--;
4568 }
4569 }
4570 else
4571 {
4572 if (multibyte)
4573 while (range > lim)
4574 {
4575 int buf_charlen;
4576
4577 buf_ch = STRING_CHAR_AND_LENGTH (d, range - lim,
4578 buf_charlen);
4579 if (fastmap[CHAR_LEADING_CODE (buf_ch)])
4580 break;
4581 range -= buf_charlen;
4582 d += buf_charlen;
4583 }
4584 else
4585 while (range > lim && !fastmap[*d])
4586 {
4587 d++;
4588 range--;
4589 }
4590 }
4591 startpos += irange - range;
4592 }
4593 else /* Searching backwards. */
4594 {
4595 int room = (startpos >= size1
4596 ? size2 + size1 - startpos
4597 : size1 - startpos);
4598 if (multibyte)
4599 {
4600 buf_ch = STRING_CHAR (d, room);
4601 buf_ch = TRANSLATE (buf_ch);
4602 if (! fastmap[CHAR_LEADING_CODE (buf_ch)])
4603 goto advance;
4604 }
4605 else
4606 {
4607 register re_wchar_t ch, translated;
4608
4609 buf_ch = *d;
4610 ch = RE_CHAR_TO_MULTIBYTE (buf_ch);
4611 translated = TRANSLATE (ch);
4612 if (translated != ch
4613 && (ch = RE_CHAR_TO_UNIBYTE (translated)) >= 0)
4614 buf_ch = ch;
4615 if (! fastmap[TRANSLATE (buf_ch)])
4616 goto advance;
4617 }
4618 }
4619 }
4620
4621 /* If can't match the null string, and that's all we have left, fail. */
4622 if (range >= 0 && startpos == total_size && fastmap
4623 && !bufp->can_be_null)
4624 return -1;
4625
4626 val = re_match_2_internal (bufp, string1, size1, string2, size2,
4627 startpos, regs, stop);
4628
4629 if (val >= 0)
4630 return startpos;
4631
4632 if (val == -2)
4633 return -2;
4634
4635 advance:
4636 if (!range)
4637 break;
4638 else if (range > 0)
4639 {
4640 /* Update STARTPOS to the next character boundary. */
4641 if (multibyte)
4642 {
4643 re_char *p = POS_ADDR_VSTRING (startpos);
4644 re_char *pend = STOP_ADDR_VSTRING (startpos);
4645 int len = MULTIBYTE_FORM_LENGTH (p, pend - p);
4646
4647 range -= len;
4648 if (range < 0)
4649 break;
4650 startpos += len;
4651 }
4652 else
4653 {
4654 range--;
4655 startpos++;
4656 }
4657 }
4658 else
4659 {
4660 range++;
4661 startpos--;
4662
4663 /* Update STARTPOS to the previous character boundary. */
4664 if (multibyte)
4665 {
4666 re_char *p = POS_ADDR_VSTRING (startpos) + 1;
4667 re_char *p0 = p;
4668 re_char *phead = HEAD_ADDR_VSTRING (startpos);
4669
4670 /* Find the head of multibyte form. */
4671 PREV_CHAR_BOUNDARY (p, phead);
4672 range += p0 - 1 - p;
4673 if (range > 0)
4674 break;
4675
4676 startpos -= p0 - 1 - p;
4677 }
4678 }
4679 }
4680 return -1;
4681 } /* re_search_2 */
4682 WEAK_ALIAS (__re_search_2, re_search_2)
4683 \f
4684 /* Declarations and macros for re_match_2. */
4685
4686 static int bcmp_translate _RE_ARGS((re_char *s1, re_char *s2,
4687 register int len,
4688 RE_TRANSLATE_TYPE translate,
4689 const int multibyte));
4690
4691 /* This converts PTR, a pointer into one of the search strings `string1'
4692 and `string2' into an offset from the beginning of that string. */
4693 #define POINTER_TO_OFFSET(ptr) \
4694 (FIRST_STRING_P (ptr) \
4695 ? ((regoff_t) ((ptr) - string1)) \
4696 : ((regoff_t) ((ptr) - string2 + size1)))
4697
4698 /* Call before fetching a character with *d. This switches over to
4699 string2 if necessary.
4700 Check re_match_2_internal for a discussion of why end_match_2 might
4701 not be within string2 (but be equal to end_match_1 instead). */
4702 #define PREFETCH() \
4703 while (d == dend) \
4704 { \
4705 /* End of string2 => fail. */ \
4706 if (dend == end_match_2) \
4707 goto fail; \
4708 /* End of string1 => advance to string2. */ \
4709 d = string2; \
4710 dend = end_match_2; \
4711 }
4712
4713 /* Call before fetching a char with *d if you already checked other limits.
4714 This is meant for use in lookahead operations like wordend, etc..
4715 where we might need to look at parts of the string that might be
4716 outside of the LIMITs (i.e past `stop'). */
4717 #define PREFETCH_NOLIMIT() \
4718 if (d == end1) \
4719 { \
4720 d = string2; \
4721 dend = end_match_2; \
4722 } \
4723
4724 /* Test if at very beginning or at very end of the virtual concatenation
4725 of `string1' and `string2'. If only one string, it's `string2'. */
4726 #define AT_STRINGS_BEG(d) ((d) == (size1 ? string1 : string2) || !size2)
4727 #define AT_STRINGS_END(d) ((d) == end2)
4728
4729
4730 /* Test if D points to a character which is word-constituent. We have
4731 two special cases to check for: if past the end of string1, look at
4732 the first character in string2; and if before the beginning of
4733 string2, look at the last character in string1. */
4734 #define WORDCHAR_P(d) \
4735 (SYNTAX ((d) == end1 ? *string2 \
4736 : (d) == string2 - 1 ? *(end1 - 1) : *(d)) \
4737 == Sword)
4738
4739 /* Disabled due to a compiler bug -- see comment at case wordbound */
4740
4741 /* The comment at case wordbound is following one, but we don't use
4742 AT_WORD_BOUNDARY anymore to support multibyte form.
4743
4744 The DEC Alpha C compiler 3.x generates incorrect code for the
4745 test WORDCHAR_P (d - 1) != WORDCHAR_P (d) in the expansion of
4746 AT_WORD_BOUNDARY, so this code is disabled. Expanding the
4747 macro and introducing temporary variables works around the bug. */
4748
4749 #if 0
4750 /* Test if the character before D and the one at D differ with respect
4751 to being word-constituent. */
4752 #define AT_WORD_BOUNDARY(d) \
4753 (AT_STRINGS_BEG (d) || AT_STRINGS_END (d) \
4754 || WORDCHAR_P (d - 1) != WORDCHAR_P (d))
4755 #endif
4756
4757 /* Free everything we malloc. */
4758 #ifdef MATCH_MAY_ALLOCATE
4759 # define FREE_VAR(var) if (var) { REGEX_FREE (var); var = NULL; } else
4760 # define FREE_VARIABLES() \
4761 do { \
4762 REGEX_FREE_STACK (fail_stack.stack); \
4763 FREE_VAR (regstart); \
4764 FREE_VAR (regend); \
4765 FREE_VAR (best_regstart); \
4766 FREE_VAR (best_regend); \
4767 } while (0)
4768 #else
4769 # define FREE_VARIABLES() ((void)0) /* Do nothing! But inhibit gcc warning. */
4770 #endif /* not MATCH_MAY_ALLOCATE */
4771
4772 \f
4773 /* Optimization routines. */
4774
4775 /* If the operation is a match against one or more chars,
4776 return a pointer to the next operation, else return NULL. */
4777 static re_char *
4778 skip_one_char (p)
4779 re_char *p;
4780 {
4781 switch (SWITCH_ENUM_CAST (*p++))
4782 {
4783 case anychar:
4784 break;
4785
4786 case exactn:
4787 p += *p + 1;
4788 break;
4789
4790 case charset_not:
4791 case charset:
4792 if (CHARSET_RANGE_TABLE_EXISTS_P (p - 1))
4793 {
4794 int mcnt;
4795 p = CHARSET_RANGE_TABLE (p - 1);
4796 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4797 p = CHARSET_RANGE_TABLE_END (p, mcnt);
4798 }
4799 else
4800 p += 1 + CHARSET_BITMAP_SIZE (p - 1);
4801 break;
4802
4803 case syntaxspec:
4804 case notsyntaxspec:
4805 #ifdef emacs
4806 case categoryspec:
4807 case notcategoryspec:
4808 #endif /* emacs */
4809 p++;
4810 break;
4811
4812 default:
4813 p = NULL;
4814 }
4815 return p;
4816 }
4817
4818
4819 /* Jump over non-matching operations. */
4820 static re_char *
4821 skip_noops (p, pend)
4822 re_char *p, *pend;
4823 {
4824 int mcnt;
4825 while (p < pend)
4826 {
4827 switch (SWITCH_ENUM_CAST ((re_opcode_t) *p))
4828 {
4829 case start_memory:
4830 case stop_memory:
4831 p += 2; break;
4832 case no_op:
4833 p += 1; break;
4834 case jump:
4835 p += 1;
4836 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4837 p += mcnt;
4838 break;
4839 default:
4840 return p;
4841 }
4842 }
4843 assert (p == pend);
4844 return p;
4845 }
4846
4847 /* Non-zero if "p1 matches something" implies "p2 fails". */
4848 static int
4849 mutually_exclusive_p (bufp, p1, p2)
4850 struct re_pattern_buffer *bufp;
4851 re_char *p1, *p2;
4852 {
4853 re_opcode_t op2;
4854 const boolean multibyte = RE_MULTIBYTE_P (bufp);
4855 unsigned char *pend = bufp->buffer + bufp->used;
4856
4857 assert (p1 >= bufp->buffer && p1 < pend
4858 && p2 >= bufp->buffer && p2 <= pend);
4859
4860 /* Skip over open/close-group commands.
4861 If what follows this loop is a ...+ construct,
4862 look at what begins its body, since we will have to
4863 match at least one of that. */
4864 p2 = skip_noops (p2, pend);
4865 /* The same skip can be done for p1, except that this function
4866 is only used in the case where p1 is a simple match operator. */
4867 /* p1 = skip_noops (p1, pend); */
4868
4869 assert (p1 >= bufp->buffer && p1 < pend
4870 && p2 >= bufp->buffer && p2 <= pend);
4871
4872 op2 = p2 == pend ? succeed : *p2;
4873
4874 switch (SWITCH_ENUM_CAST (op2))
4875 {
4876 case succeed:
4877 case endbuf:
4878 /* If we're at the end of the pattern, we can change. */
4879 if (skip_one_char (p1))
4880 {
4881 DEBUG_PRINT1 (" End of pattern: fast loop.\n");
4882 return 1;
4883 }
4884 break;
4885
4886 case endline:
4887 case exactn:
4888 {
4889 register re_wchar_t c
4890 = (re_opcode_t) *p2 == endline ? '\n'
4891 : RE_STRING_CHAR (p2 + 2, pend - p2 - 2, multibyte);
4892
4893 if ((re_opcode_t) *p1 == exactn)
4894 {
4895 if (c != RE_STRING_CHAR (p1 + 2, pend - p1 - 2, multibyte))
4896 {
4897 DEBUG_PRINT3 (" '%c' != '%c' => fast loop.\n", c, p1[2]);
4898 return 1;
4899 }
4900 }
4901
4902 else if ((re_opcode_t) *p1 == charset
4903 || (re_opcode_t) *p1 == charset_not)
4904 {
4905 int not = (re_opcode_t) *p1 == charset_not;
4906
4907 /* Test if C is listed in charset (or charset_not)
4908 at `p1'. */
4909 if (! multibyte || IS_REAL_ASCII (c))
4910 {
4911 if (c < CHARSET_BITMAP_SIZE (p1) * BYTEWIDTH
4912 && p1[2 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH)))
4913 not = !not;
4914 }
4915 else if (CHARSET_RANGE_TABLE_EXISTS_P (p1))
4916 CHARSET_LOOKUP_RANGE_TABLE (not, c, p1);
4917
4918 /* `not' is equal to 1 if c would match, which means
4919 that we can't change to pop_failure_jump. */
4920 if (!not)
4921 {
4922 DEBUG_PRINT1 (" No match => fast loop.\n");
4923 return 1;
4924 }
4925 }
4926 else if ((re_opcode_t) *p1 == anychar
4927 && c == '\n')
4928 {
4929 DEBUG_PRINT1 (" . != \\n => fast loop.\n");
4930 return 1;
4931 }
4932 }
4933 break;
4934
4935 case charset:
4936 {
4937 if ((re_opcode_t) *p1 == exactn)
4938 /* Reuse the code above. */
4939 return mutually_exclusive_p (bufp, p2, p1);
4940
4941 /* It is hard to list up all the character in charset
4942 P2 if it includes multibyte character. Give up in
4943 such case. */
4944 else if (!multibyte || !CHARSET_RANGE_TABLE_EXISTS_P (p2))
4945 {
4946 /* Now, we are sure that P2 has no range table.
4947 So, for the size of bitmap in P2, `p2[1]' is
4948 enough. But P1 may have range table, so the
4949 size of bitmap table of P1 is extracted by
4950 using macro `CHARSET_BITMAP_SIZE'.
4951
4952 In a multibyte case, we know that all the character
4953 listed in P2 is ASCII. In a unibyte case, P1 has only a
4954 bitmap table. So, in both cases, it is enough to test
4955 only the bitmap table of P1. */
4956
4957 if ((re_opcode_t) *p1 == charset)
4958 {
4959 int idx;
4960 /* We win if the charset inside the loop
4961 has no overlap with the one after the loop. */
4962 for (idx = 0;
4963 (idx < (int) p2[1]
4964 && idx < CHARSET_BITMAP_SIZE (p1));
4965 idx++)
4966 if ((p2[2 + idx] & p1[2 + idx]) != 0)
4967 break;
4968
4969 if (idx == p2[1]
4970 || idx == CHARSET_BITMAP_SIZE (p1))
4971 {
4972 DEBUG_PRINT1 (" No match => fast loop.\n");
4973 return 1;
4974 }
4975 }
4976 else if ((re_opcode_t) *p1 == charset_not)
4977 {
4978 int idx;
4979 /* We win if the charset_not inside the loop lists
4980 every character listed in the charset after. */
4981 for (idx = 0; idx < (int) p2[1]; idx++)
4982 if (! (p2[2 + idx] == 0
4983 || (idx < CHARSET_BITMAP_SIZE (p1)
4984 && ((p2[2 + idx] & ~ p1[2 + idx]) == 0))))
4985 break;
4986
4987 if (idx == p2[1])
4988 {
4989 DEBUG_PRINT1 (" No match => fast loop.\n");
4990 return 1;
4991 }
4992 }
4993 }
4994 }
4995 break;
4996
4997 case charset_not:
4998 switch (SWITCH_ENUM_CAST (*p1))
4999 {
5000 case exactn:
5001 case charset:
5002 /* Reuse the code above. */
5003 return mutually_exclusive_p (bufp, p2, p1);
5004 case charset_not:
5005 /* When we have two charset_not, it's very unlikely that
5006 they don't overlap. The union of the two sets of excluded
5007 chars should cover all possible chars, which, as a matter of
5008 fact, is virtually impossible in multibyte buffers. */
5009 break;
5010 }
5011 break;
5012
5013 case wordend:
5014 return ((re_opcode_t) *p1 == syntaxspec && p1[1] == Sword);
5015 case symend:
5016 return ((re_opcode_t) *p1 == syntaxspec
5017 && (p1[1] == Ssymbol || p1[1] == Sword));
5018 case notsyntaxspec:
5019 return ((re_opcode_t) *p1 == syntaxspec && p1[1] == p2[1]);
5020
5021 case wordbeg:
5022 return ((re_opcode_t) *p1 == notsyntaxspec && p1[1] == Sword);
5023 case symbeg:
5024 return ((re_opcode_t) *p1 == notsyntaxspec
5025 && (p1[1] == Ssymbol || p1[1] == Sword));
5026 case syntaxspec:
5027 return ((re_opcode_t) *p1 == notsyntaxspec && p1[1] == p2[1]);
5028
5029 case wordbound:
5030 return (((re_opcode_t) *p1 == notsyntaxspec
5031 || (re_opcode_t) *p1 == syntaxspec)
5032 && p1[1] == Sword);
5033
5034 #ifdef emacs
5035 case categoryspec:
5036 return ((re_opcode_t) *p1 == notcategoryspec && p1[1] == p2[1]);
5037 case notcategoryspec:
5038 return ((re_opcode_t) *p1 == categoryspec && p1[1] == p2[1]);
5039 #endif /* emacs */
5040
5041 default:
5042 ;
5043 }
5044
5045 /* Safe default. */
5046 return 0;
5047 }
5048
5049 \f
5050 /* Matching routines. */
5051
5052 #ifndef emacs /* Emacs never uses this. */
5053 /* re_match is like re_match_2 except it takes only a single string. */
5054
5055 int
5056 re_match (bufp, string, size, pos, regs)
5057 struct re_pattern_buffer *bufp;
5058 const char *string;
5059 int size, pos;
5060 struct re_registers *regs;
5061 {
5062 int result = re_match_2_internal (bufp, NULL, 0, (re_char*) string, size,
5063 pos, regs, size);
5064 return result;
5065 }
5066 WEAK_ALIAS (__re_match, re_match)
5067 #endif /* not emacs */
5068
5069 #ifdef emacs
5070 /* In Emacs, this is the string or buffer in which we
5071 are matching. It is used for looking up syntax properties. */
5072 Lisp_Object re_match_object;
5073 #endif
5074
5075 /* re_match_2 matches the compiled pattern in BUFP against the
5076 the (virtual) concatenation of STRING1 and STRING2 (of length SIZE1
5077 and SIZE2, respectively). We start matching at POS, and stop
5078 matching at STOP.
5079
5080 If REGS is non-null and the `no_sub' field of BUFP is nonzero, we
5081 store offsets for the substring each group matched in REGS. See the
5082 documentation for exactly how many groups we fill.
5083
5084 We return -1 if no match, -2 if an internal error (such as the
5085 failure stack overflowing). Otherwise, we return the length of the
5086 matched substring. */
5087
5088 int
5089 re_match_2 (bufp, string1, size1, string2, size2, pos, regs, stop)
5090 struct re_pattern_buffer *bufp;
5091 const char *string1, *string2;
5092 int size1, size2;
5093 int pos;
5094 struct re_registers *regs;
5095 int stop;
5096 {
5097 int result;
5098
5099 #ifdef emacs
5100 int charpos;
5101 gl_state.object = re_match_object;
5102 charpos = SYNTAX_TABLE_BYTE_TO_CHAR (POS_AS_IN_BUFFER (pos));
5103 SETUP_SYNTAX_TABLE_FOR_OBJECT (re_match_object, charpos, 1);
5104 #endif
5105
5106 result = re_match_2_internal (bufp, (re_char*) string1, size1,
5107 (re_char*) string2, size2,
5108 pos, regs, stop);
5109 return result;
5110 }
5111 WEAK_ALIAS (__re_match_2, re_match_2)
5112
5113
5114 /* This is a separate function so that we can force an alloca cleanup
5115 afterwards. */
5116 static int
5117 re_match_2_internal (bufp, string1, size1, string2, size2, pos, regs, stop)
5118 struct re_pattern_buffer *bufp;
5119 re_char *string1, *string2;
5120 int size1, size2;
5121 int pos;
5122 struct re_registers *regs;
5123 int stop;
5124 {
5125 /* General temporaries. */
5126 int mcnt;
5127 size_t reg;
5128 boolean not;
5129
5130 /* Just past the end of the corresponding string. */
5131 re_char *end1, *end2;
5132
5133 /* Pointers into string1 and string2, just past the last characters in
5134 each to consider matching. */
5135 re_char *end_match_1, *end_match_2;
5136
5137 /* Where we are in the data, and the end of the current string. */
5138 re_char *d, *dend;
5139
5140 /* Used sometimes to remember where we were before starting matching
5141 an operator so that we can go back in case of failure. This "atomic"
5142 behavior of matching opcodes is indispensable to the correctness
5143 of the on_failure_keep_string_jump optimization. */
5144 re_char *dfail;
5145
5146 /* Where we are in the pattern, and the end of the pattern. */
5147 re_char *p = bufp->buffer;
5148 re_char *pend = p + bufp->used;
5149
5150 /* We use this to map every character in the string. */
5151 RE_TRANSLATE_TYPE translate = bufp->translate;
5152
5153 /* Nonzero if BUFP is setup from a multibyte regex. */
5154 const boolean multibyte = RE_MULTIBYTE_P (bufp);
5155
5156 /* Nonzero if STRING1/STRING2 are multibyte. */
5157 const boolean target_multibyte = RE_TARGET_MULTIBYTE_P (bufp);
5158
5159 /* Failure point stack. Each place that can handle a failure further
5160 down the line pushes a failure point on this stack. It consists of
5161 regstart, and regend for all registers corresponding to
5162 the subexpressions we're currently inside, plus the number of such
5163 registers, and, finally, two char *'s. The first char * is where
5164 to resume scanning the pattern; the second one is where to resume
5165 scanning the strings. */
5166 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
5167 fail_stack_type fail_stack;
5168 #endif
5169 #ifdef DEBUG
5170 unsigned nfailure_points_pushed = 0, nfailure_points_popped = 0;
5171 #endif
5172
5173 #if defined REL_ALLOC && defined REGEX_MALLOC
5174 /* This holds the pointer to the failure stack, when
5175 it is allocated relocatably. */
5176 fail_stack_elt_t *failure_stack_ptr;
5177 #endif
5178
5179 /* We fill all the registers internally, independent of what we
5180 return, for use in backreferences. The number here includes
5181 an element for register zero. */
5182 size_t num_regs = bufp->re_nsub + 1;
5183
5184 /* Information on the contents of registers. These are pointers into
5185 the input strings; they record just what was matched (on this
5186 attempt) by a subexpression part of the pattern, that is, the
5187 regnum-th regstart pointer points to where in the pattern we began
5188 matching and the regnum-th regend points to right after where we
5189 stopped matching the regnum-th subexpression. (The zeroth register
5190 keeps track of what the whole pattern matches.) */
5191 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
5192 re_char **regstart, **regend;
5193 #endif
5194
5195 /* The following record the register info as found in the above
5196 variables when we find a match better than any we've seen before.
5197 This happens as we backtrack through the failure points, which in
5198 turn happens only if we have not yet matched the entire string. */
5199 unsigned best_regs_set = false;
5200 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
5201 re_char **best_regstart, **best_regend;
5202 #endif
5203
5204 /* Logically, this is `best_regend[0]'. But we don't want to have to
5205 allocate space for that if we're not allocating space for anything
5206 else (see below). Also, we never need info about register 0 for
5207 any of the other register vectors, and it seems rather a kludge to
5208 treat `best_regend' differently than the rest. So we keep track of
5209 the end of the best match so far in a separate variable. We
5210 initialize this to NULL so that when we backtrack the first time
5211 and need to test it, it's not garbage. */
5212 re_char *match_end = NULL;
5213
5214 #ifdef DEBUG
5215 /* Counts the total number of registers pushed. */
5216 unsigned num_regs_pushed = 0;
5217 #endif
5218
5219 DEBUG_PRINT1 ("\n\nEntering re_match_2.\n");
5220
5221 INIT_FAIL_STACK ();
5222
5223 #ifdef MATCH_MAY_ALLOCATE
5224 /* Do not bother to initialize all the register variables if there are
5225 no groups in the pattern, as it takes a fair amount of time. If
5226 there are groups, we include space for register 0 (the whole
5227 pattern), even though we never use it, since it simplifies the
5228 array indexing. We should fix this. */
5229 if (bufp->re_nsub)
5230 {
5231 regstart = REGEX_TALLOC (num_regs, re_char *);
5232 regend = REGEX_TALLOC (num_regs, re_char *);
5233 best_regstart = REGEX_TALLOC (num_regs, re_char *);
5234 best_regend = REGEX_TALLOC (num_regs, re_char *);
5235
5236 if (!(regstart && regend && best_regstart && best_regend))
5237 {
5238 FREE_VARIABLES ();
5239 return -2;
5240 }
5241 }
5242 else
5243 {
5244 /* We must initialize all our variables to NULL, so that
5245 `FREE_VARIABLES' doesn't try to free them. */
5246 regstart = regend = best_regstart = best_regend = NULL;
5247 }
5248 #endif /* MATCH_MAY_ALLOCATE */
5249
5250 /* The starting position is bogus. */
5251 if (pos < 0 || pos > size1 + size2)
5252 {
5253 FREE_VARIABLES ();
5254 return -1;
5255 }
5256
5257 /* Initialize subexpression text positions to -1 to mark ones that no
5258 start_memory/stop_memory has been seen for. Also initialize the
5259 register information struct. */
5260 for (reg = 1; reg < num_regs; reg++)
5261 regstart[reg] = regend[reg] = NULL;
5262
5263 /* We move `string1' into `string2' if the latter's empty -- but not if
5264 `string1' is null. */
5265 if (size2 == 0 && string1 != NULL)
5266 {
5267 string2 = string1;
5268 size2 = size1;
5269 string1 = 0;
5270 size1 = 0;
5271 }
5272 end1 = string1 + size1;
5273 end2 = string2 + size2;
5274
5275 /* `p' scans through the pattern as `d' scans through the data.
5276 `dend' is the end of the input string that `d' points within. `d'
5277 is advanced into the following input string whenever necessary, but
5278 this happens before fetching; therefore, at the beginning of the
5279 loop, `d' can be pointing at the end of a string, but it cannot
5280 equal `string2'. */
5281 if (pos >= size1)
5282 {
5283 /* Only match within string2. */
5284 d = string2 + pos - size1;
5285 dend = end_match_2 = string2 + stop - size1;
5286 end_match_1 = end1; /* Just to give it a value. */
5287 }
5288 else
5289 {
5290 if (stop < size1)
5291 {
5292 /* Only match within string1. */
5293 end_match_1 = string1 + stop;
5294 /* BEWARE!
5295 When we reach end_match_1, PREFETCH normally switches to string2.
5296 But in the present case, this means that just doing a PREFETCH
5297 makes us jump from `stop' to `gap' within the string.
5298 What we really want here is for the search to stop as
5299 soon as we hit end_match_1. That's why we set end_match_2
5300 to end_match_1 (since PREFETCH fails as soon as we hit
5301 end_match_2). */
5302 end_match_2 = end_match_1;
5303 }
5304 else
5305 { /* It's important to use this code when stop == size so that
5306 moving `d' from end1 to string2 will not prevent the d == dend
5307 check from catching the end of string. */
5308 end_match_1 = end1;
5309 end_match_2 = string2 + stop - size1;
5310 }
5311 d = string1 + pos;
5312 dend = end_match_1;
5313 }
5314
5315 DEBUG_PRINT1 ("The compiled pattern is: ");
5316 DEBUG_PRINT_COMPILED_PATTERN (bufp, p, pend);
5317 DEBUG_PRINT1 ("The string to match is: `");
5318 DEBUG_PRINT_DOUBLE_STRING (d, string1, size1, string2, size2);
5319 DEBUG_PRINT1 ("'\n");
5320
5321 /* This loops over pattern commands. It exits by returning from the
5322 function if the match is complete, or it drops through if the match
5323 fails at this starting point in the input data. */
5324 for (;;)
5325 {
5326 DEBUG_PRINT2 ("\n%p: ", p);
5327
5328 if (p == pend)
5329 { /* End of pattern means we might have succeeded. */
5330 DEBUG_PRINT1 ("end of pattern ... ");
5331
5332 /* If we haven't matched the entire string, and we want the
5333 longest match, try backtracking. */
5334 if (d != end_match_2)
5335 {
5336 /* 1 if this match ends in the same string (string1 or string2)
5337 as the best previous match. */
5338 boolean same_str_p = (FIRST_STRING_P (match_end)
5339 == FIRST_STRING_P (d));
5340 /* 1 if this match is the best seen so far. */
5341 boolean best_match_p;
5342
5343 /* AIX compiler got confused when this was combined
5344 with the previous declaration. */
5345 if (same_str_p)
5346 best_match_p = d > match_end;
5347 else
5348 best_match_p = !FIRST_STRING_P (d);
5349
5350 DEBUG_PRINT1 ("backtracking.\n");
5351
5352 if (!FAIL_STACK_EMPTY ())
5353 { /* More failure points to try. */
5354
5355 /* If exceeds best match so far, save it. */
5356 if (!best_regs_set || best_match_p)
5357 {
5358 best_regs_set = true;
5359 match_end = d;
5360
5361 DEBUG_PRINT1 ("\nSAVING match as best so far.\n");
5362
5363 for (reg = 1; reg < num_regs; reg++)
5364 {
5365 best_regstart[reg] = regstart[reg];
5366 best_regend[reg] = regend[reg];
5367 }
5368 }
5369 goto fail;
5370 }
5371
5372 /* If no failure points, don't restore garbage. And if
5373 last match is real best match, don't restore second
5374 best one. */
5375 else if (best_regs_set && !best_match_p)
5376 {
5377 restore_best_regs:
5378 /* Restore best match. It may happen that `dend ==
5379 end_match_1' while the restored d is in string2.
5380 For example, the pattern `x.*y.*z' against the
5381 strings `x-' and `y-z-', if the two strings are
5382 not consecutive in memory. */
5383 DEBUG_PRINT1 ("Restoring best registers.\n");
5384
5385 d = match_end;
5386 dend = ((d >= string1 && d <= end1)
5387 ? end_match_1 : end_match_2);
5388
5389 for (reg = 1; reg < num_regs; reg++)
5390 {
5391 regstart[reg] = best_regstart[reg];
5392 regend[reg] = best_regend[reg];
5393 }
5394 }
5395 } /* d != end_match_2 */
5396
5397 succeed_label:
5398 DEBUG_PRINT1 ("Accepting match.\n");
5399
5400 /* If caller wants register contents data back, do it. */
5401 if (regs && !bufp->no_sub)
5402 {
5403 /* Have the register data arrays been allocated? */
5404 if (bufp->regs_allocated == REGS_UNALLOCATED)
5405 { /* No. So allocate them with malloc. We need one
5406 extra element beyond `num_regs' for the `-1' marker
5407 GNU code uses. */
5408 regs->num_regs = MAX (RE_NREGS, num_regs + 1);
5409 regs->start = TALLOC (regs->num_regs, regoff_t);
5410 regs->end = TALLOC (regs->num_regs, regoff_t);
5411 if (regs->start == NULL || regs->end == NULL)
5412 {
5413 FREE_VARIABLES ();
5414 return -2;
5415 }
5416 bufp->regs_allocated = REGS_REALLOCATE;
5417 }
5418 else if (bufp->regs_allocated == REGS_REALLOCATE)
5419 { /* Yes. If we need more elements than were already
5420 allocated, reallocate them. If we need fewer, just
5421 leave it alone. */
5422 if (regs->num_regs < num_regs + 1)
5423 {
5424 regs->num_regs = num_regs + 1;
5425 RETALLOC (regs->start, regs->num_regs, regoff_t);
5426 RETALLOC (regs->end, regs->num_regs, regoff_t);
5427 if (regs->start == NULL || regs->end == NULL)
5428 {
5429 FREE_VARIABLES ();
5430 return -2;
5431 }
5432 }
5433 }
5434 else
5435 {
5436 /* These braces fend off a "empty body in an else-statement"
5437 warning under GCC when assert expands to nothing. */
5438 assert (bufp->regs_allocated == REGS_FIXED);
5439 }
5440
5441 /* Convert the pointer data in `regstart' and `regend' to
5442 indices. Register zero has to be set differently,
5443 since we haven't kept track of any info for it. */
5444 if (regs->num_regs > 0)
5445 {
5446 regs->start[0] = pos;
5447 regs->end[0] = POINTER_TO_OFFSET (d);
5448 }
5449
5450 /* Go through the first `min (num_regs, regs->num_regs)'
5451 registers, since that is all we initialized. */
5452 for (reg = 1; reg < MIN (num_regs, regs->num_regs); reg++)
5453 {
5454 if (REG_UNSET (regstart[reg]) || REG_UNSET (regend[reg]))
5455 regs->start[reg] = regs->end[reg] = -1;
5456 else
5457 {
5458 regs->start[reg]
5459 = (regoff_t) POINTER_TO_OFFSET (regstart[reg]);
5460 regs->end[reg]
5461 = (regoff_t) POINTER_TO_OFFSET (regend[reg]);
5462 }
5463 }
5464
5465 /* If the regs structure we return has more elements than
5466 were in the pattern, set the extra elements to -1. If
5467 we (re)allocated the registers, this is the case,
5468 because we always allocate enough to have at least one
5469 -1 at the end. */
5470 for (reg = num_regs; reg < regs->num_regs; reg++)
5471 regs->start[reg] = regs->end[reg] = -1;
5472 } /* regs && !bufp->no_sub */
5473
5474 DEBUG_PRINT4 ("%u failure points pushed, %u popped (%u remain).\n",
5475 nfailure_points_pushed, nfailure_points_popped,
5476 nfailure_points_pushed - nfailure_points_popped);
5477 DEBUG_PRINT2 ("%u registers pushed.\n", num_regs_pushed);
5478
5479 mcnt = POINTER_TO_OFFSET (d) - pos;
5480
5481 DEBUG_PRINT2 ("Returning %d from re_match_2.\n", mcnt);
5482
5483 FREE_VARIABLES ();
5484 return mcnt;
5485 }
5486
5487 /* Otherwise match next pattern command. */
5488 switch (SWITCH_ENUM_CAST ((re_opcode_t) *p++))
5489 {
5490 /* Ignore these. Used to ignore the n of succeed_n's which
5491 currently have n == 0. */
5492 case no_op:
5493 DEBUG_PRINT1 ("EXECUTING no_op.\n");
5494 break;
5495
5496 case succeed:
5497 DEBUG_PRINT1 ("EXECUTING succeed.\n");
5498 goto succeed_label;
5499
5500 /* Match the next n pattern characters exactly. The following
5501 byte in the pattern defines n, and the n bytes after that
5502 are the characters to match. */
5503 case exactn:
5504 mcnt = *p++;
5505 DEBUG_PRINT2 ("EXECUTING exactn %d.\n", mcnt);
5506
5507 /* Remember the start point to rollback upon failure. */
5508 dfail = d;
5509
5510 #ifndef emacs
5511 /* This is written out as an if-else so we don't waste time
5512 testing `translate' inside the loop. */
5513 if (RE_TRANSLATE_P (translate))
5514 do
5515 {
5516 PREFETCH ();
5517 if (RE_TRANSLATE (translate, *d) != *p++)
5518 {
5519 d = dfail;
5520 goto fail;
5521 }
5522 d++;
5523 }
5524 while (--mcnt);
5525 else
5526 do
5527 {
5528 PREFETCH ();
5529 if (*d++ != *p++)
5530 {
5531 d = dfail;
5532 goto fail;
5533 }
5534 }
5535 while (--mcnt);
5536 #else /* emacs */
5537 /* The cost of testing `translate' is comparatively small. */
5538 if (target_multibyte)
5539 do
5540 {
5541 int pat_charlen, buf_charlen;
5542 int pat_ch, buf_ch;
5543
5544 PREFETCH ();
5545 if (multibyte)
5546 pat_ch = STRING_CHAR_AND_LENGTH (p, pend - p, pat_charlen);
5547 else
5548 {
5549 pat_ch = RE_CHAR_TO_MULTIBYTE (*p);
5550 pat_charlen = 1;
5551 }
5552 buf_ch = STRING_CHAR_AND_LENGTH (d, dend - d, buf_charlen);
5553
5554 if (TRANSLATE (buf_ch) != pat_ch)
5555 {
5556 d = dfail;
5557 goto fail;
5558 }
5559
5560 p += pat_charlen;
5561 d += buf_charlen;
5562 mcnt -= pat_charlen;
5563 }
5564 while (mcnt > 0);
5565 else
5566 do
5567 {
5568 int pat_charlen, buf_charlen;
5569 int pat_ch, buf_ch;
5570
5571 PREFETCH ();
5572 if (multibyte)
5573 {
5574 pat_ch = STRING_CHAR_AND_LENGTH (p, pend - p, pat_charlen);
5575 if (CHAR_BYTE8_P (pat_ch))
5576 pat_ch = CHAR_TO_BYTE8 (pat_ch);
5577 else
5578 pat_ch = RE_CHAR_TO_UNIBYTE (pat_ch);
5579 }
5580 else
5581 {
5582 pat_ch = *p;
5583 pat_charlen = 1;
5584 }
5585 buf_ch = RE_CHAR_TO_MULTIBYTE (*d);
5586 if (! CHAR_BYTE8_P (buf_ch))
5587 {
5588 buf_ch = TRANSLATE (buf_ch);
5589 buf_ch = RE_CHAR_TO_UNIBYTE (buf_ch);
5590 if (buf_ch < 0)
5591 buf_ch = *d;
5592 }
5593 else
5594 buf_ch = *d;
5595 if (buf_ch != pat_ch)
5596 {
5597 d = dfail;
5598 goto fail;
5599 }
5600 p += pat_charlen;
5601 d++;
5602 }
5603 while (--mcnt);
5604 #endif
5605 break;
5606
5607
5608 /* Match any character except possibly a newline or a null. */
5609 case anychar:
5610 {
5611 int buf_charlen;
5612 re_wchar_t buf_ch;
5613
5614 DEBUG_PRINT1 ("EXECUTING anychar.\n");
5615
5616 PREFETCH ();
5617 buf_ch = RE_STRING_CHAR_AND_LENGTH (d, dend - d, buf_charlen,
5618 target_multibyte);
5619 buf_ch = TRANSLATE (buf_ch);
5620
5621 if ((!(bufp->syntax & RE_DOT_NEWLINE)
5622 && buf_ch == '\n')
5623 || ((bufp->syntax & RE_DOT_NOT_NULL)
5624 && buf_ch == '\000'))
5625 goto fail;
5626
5627 DEBUG_PRINT2 (" Matched `%d'.\n", *d);
5628 d += buf_charlen;
5629 }
5630 break;
5631
5632
5633 case charset:
5634 case charset_not:
5635 {
5636 register unsigned int c;
5637 boolean not = (re_opcode_t) *(p - 1) == charset_not;
5638 int len;
5639
5640 /* Start of actual range_table, or end of bitmap if there is no
5641 range table. */
5642 re_char *range_table;
5643
5644 /* Nonzero if there is a range table. */
5645 int range_table_exists;
5646
5647 /* Number of ranges of range table. This is not included
5648 in the initial byte-length of the command. */
5649 int count = 0;
5650
5651 DEBUG_PRINT2 ("EXECUTING charset%s.\n", not ? "_not" : "");
5652
5653 range_table_exists = CHARSET_RANGE_TABLE_EXISTS_P (&p[-1]);
5654
5655 if (range_table_exists)
5656 {
5657 range_table = CHARSET_RANGE_TABLE (&p[-1]); /* Past the bitmap. */
5658 EXTRACT_NUMBER_AND_INCR (count, range_table);
5659 }
5660
5661 PREFETCH ();
5662 c = RE_STRING_CHAR_AND_LENGTH (d, dend - d, len, target_multibyte);
5663 if (target_multibyte)
5664 {
5665 int c1;
5666
5667 c = TRANSLATE (c);
5668 c1 = RE_CHAR_TO_UNIBYTE (c);
5669 if (c1 >= 0)
5670 c = c1;
5671 }
5672 else
5673 {
5674 int c1 = RE_CHAR_TO_MULTIBYTE (c);
5675
5676 if (! CHAR_BYTE8_P (c1))
5677 {
5678 c1 = TRANSLATE (c1);
5679 c1 = RE_CHAR_TO_UNIBYTE (c1);
5680 if (c1 >= 0)
5681 c = c1;
5682 }
5683 }
5684
5685 if (c < (1 << BYTEWIDTH))
5686 { /* Lookup bitmap. */
5687 /* Cast to `unsigned' instead of `unsigned char' in
5688 case the bit list is a full 32 bytes long. */
5689 if (c < (unsigned) (CHARSET_BITMAP_SIZE (&p[-1]) * BYTEWIDTH)
5690 && p[1 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH)))
5691 not = !not;
5692 }
5693 #ifdef emacs
5694 else if (range_table_exists)
5695 {
5696 int class_bits = CHARSET_RANGE_TABLE_BITS (&p[-1]);
5697
5698 if ( (class_bits & BIT_LOWER && ISLOWER (c))
5699 | (class_bits & BIT_MULTIBYTE)
5700 | (class_bits & BIT_PUNCT && ISPUNCT (c))
5701 | (class_bits & BIT_SPACE && ISSPACE (c))
5702 | (class_bits & BIT_UPPER && ISUPPER (c))
5703 | (class_bits & BIT_WORD && ISWORD (c)))
5704 not = !not;
5705 else
5706 CHARSET_LOOKUP_RANGE_TABLE_RAW (not, c, range_table, count);
5707 }
5708 #endif /* emacs */
5709
5710 if (range_table_exists)
5711 p = CHARSET_RANGE_TABLE_END (range_table, count);
5712 else
5713 p += CHARSET_BITMAP_SIZE (&p[-1]) + 1;
5714
5715 if (!not) goto fail;
5716
5717 d += len;
5718 break;
5719 }
5720
5721
5722 /* The beginning of a group is represented by start_memory.
5723 The argument is the register number. The text
5724 matched within the group is recorded (in the internal
5725 registers data structure) under the register number. */
5726 case start_memory:
5727 DEBUG_PRINT2 ("EXECUTING start_memory %d:\n", *p);
5728
5729 /* In case we need to undo this operation (via backtracking). */
5730 PUSH_FAILURE_REG ((unsigned int)*p);
5731
5732 regstart[*p] = d;
5733 regend[*p] = NULL; /* probably unnecessary. -sm */
5734 DEBUG_PRINT2 (" regstart: %d\n", POINTER_TO_OFFSET (regstart[*p]));
5735
5736 /* Move past the register number and inner group count. */
5737 p += 1;
5738 break;
5739
5740
5741 /* The stop_memory opcode represents the end of a group. Its
5742 argument is the same as start_memory's: the register number. */
5743 case stop_memory:
5744 DEBUG_PRINT2 ("EXECUTING stop_memory %d:\n", *p);
5745
5746 assert (!REG_UNSET (regstart[*p]));
5747 /* Strictly speaking, there should be code such as:
5748
5749 assert (REG_UNSET (regend[*p]));
5750 PUSH_FAILURE_REGSTOP ((unsigned int)*p);
5751
5752 But the only info to be pushed is regend[*p] and it is known to
5753 be UNSET, so there really isn't anything to push.
5754 Not pushing anything, on the other hand deprives us from the
5755 guarantee that regend[*p] is UNSET since undoing this operation
5756 will not reset its value properly. This is not important since
5757 the value will only be read on the next start_memory or at
5758 the very end and both events can only happen if this stop_memory
5759 is *not* undone. */
5760
5761 regend[*p] = d;
5762 DEBUG_PRINT2 (" regend: %d\n", POINTER_TO_OFFSET (regend[*p]));
5763
5764 /* Move past the register number and the inner group count. */
5765 p += 1;
5766 break;
5767
5768
5769 /* \<digit> has been turned into a `duplicate' command which is
5770 followed by the numeric value of <digit> as the register number. */
5771 case duplicate:
5772 {
5773 register re_char *d2, *dend2;
5774 int regno = *p++; /* Get which register to match against. */
5775 DEBUG_PRINT2 ("EXECUTING duplicate %d.\n", regno);
5776
5777 /* Can't back reference a group which we've never matched. */
5778 if (REG_UNSET (regstart[regno]) || REG_UNSET (regend[regno]))
5779 goto fail;
5780
5781 /* Where in input to try to start matching. */
5782 d2 = regstart[regno];
5783
5784 /* Remember the start point to rollback upon failure. */
5785 dfail = d;
5786
5787 /* Where to stop matching; if both the place to start and
5788 the place to stop matching are in the same string, then
5789 set to the place to stop, otherwise, for now have to use
5790 the end of the first string. */
5791
5792 dend2 = ((FIRST_STRING_P (regstart[regno])
5793 == FIRST_STRING_P (regend[regno]))
5794 ? regend[regno] : end_match_1);
5795 for (;;)
5796 {
5797 /* If necessary, advance to next segment in register
5798 contents. */
5799 while (d2 == dend2)
5800 {
5801 if (dend2 == end_match_2) break;
5802 if (dend2 == regend[regno]) break;
5803
5804 /* End of string1 => advance to string2. */
5805 d2 = string2;
5806 dend2 = regend[regno];
5807 }
5808 /* At end of register contents => success */
5809 if (d2 == dend2) break;
5810
5811 /* If necessary, advance to next segment in data. */
5812 PREFETCH ();
5813
5814 /* How many characters left in this segment to match. */
5815 mcnt = dend - d;
5816
5817 /* Want how many consecutive characters we can match in
5818 one shot, so, if necessary, adjust the count. */
5819 if (mcnt > dend2 - d2)
5820 mcnt = dend2 - d2;
5821
5822 /* Compare that many; failure if mismatch, else move
5823 past them. */
5824 if (RE_TRANSLATE_P (translate)
5825 ? bcmp_translate (d, d2, mcnt, translate, target_multibyte)
5826 : memcmp (d, d2, mcnt))
5827 {
5828 d = dfail;
5829 goto fail;
5830 }
5831 d += mcnt, d2 += mcnt;
5832 }
5833 }
5834 break;
5835
5836
5837 /* begline matches the empty string at the beginning of the string
5838 (unless `not_bol' is set in `bufp'), and after newlines. */
5839 case begline:
5840 DEBUG_PRINT1 ("EXECUTING begline.\n");
5841
5842 if (AT_STRINGS_BEG (d))
5843 {
5844 if (!bufp->not_bol) break;
5845 }
5846 else
5847 {
5848 unsigned c;
5849 GET_CHAR_BEFORE_2 (c, d, string1, end1, string2, end2);
5850 if (c == '\n')
5851 break;
5852 }
5853 /* In all other cases, we fail. */
5854 goto fail;
5855
5856
5857 /* endline is the dual of begline. */
5858 case endline:
5859 DEBUG_PRINT1 ("EXECUTING endline.\n");
5860
5861 if (AT_STRINGS_END (d))
5862 {
5863 if (!bufp->not_eol) break;
5864 }
5865 else
5866 {
5867 PREFETCH_NOLIMIT ();
5868 if (*d == '\n')
5869 break;
5870 }
5871 goto fail;
5872
5873
5874 /* Match at the very beginning of the data. */
5875 case begbuf:
5876 DEBUG_PRINT1 ("EXECUTING begbuf.\n");
5877 if (AT_STRINGS_BEG (d))
5878 break;
5879 goto fail;
5880
5881
5882 /* Match at the very end of the data. */
5883 case endbuf:
5884 DEBUG_PRINT1 ("EXECUTING endbuf.\n");
5885 if (AT_STRINGS_END (d))
5886 break;
5887 goto fail;
5888
5889
5890 /* on_failure_keep_string_jump is used to optimize `.*\n'. It
5891 pushes NULL as the value for the string on the stack. Then
5892 `POP_FAILURE_POINT' will keep the current value for the
5893 string, instead of restoring it. To see why, consider
5894 matching `foo\nbar' against `.*\n'. The .* matches the foo;
5895 then the . fails against the \n. But the next thing we want
5896 to do is match the \n against the \n; if we restored the
5897 string value, we would be back at the foo.
5898
5899 Because this is used only in specific cases, we don't need to
5900 check all the things that `on_failure_jump' does, to make
5901 sure the right things get saved on the stack. Hence we don't
5902 share its code. The only reason to push anything on the
5903 stack at all is that otherwise we would have to change
5904 `anychar's code to do something besides goto fail in this
5905 case; that seems worse than this. */
5906 case on_failure_keep_string_jump:
5907 EXTRACT_NUMBER_AND_INCR (mcnt, p);
5908 DEBUG_PRINT3 ("EXECUTING on_failure_keep_string_jump %d (to %p):\n",
5909 mcnt, p + mcnt);
5910
5911 PUSH_FAILURE_POINT (p - 3, NULL);
5912 break;
5913
5914 /* A nasty loop is introduced by the non-greedy *? and +?.
5915 With such loops, the stack only ever contains one failure point
5916 at a time, so that a plain on_failure_jump_loop kind of
5917 cycle detection cannot work. Worse yet, such a detection
5918 can not only fail to detect a cycle, but it can also wrongly
5919 detect a cycle (between different instantiations of the same
5920 loop).
5921 So the method used for those nasty loops is a little different:
5922 We use a special cycle-detection-stack-frame which is pushed
5923 when the on_failure_jump_nastyloop failure-point is *popped*.
5924 This special frame thus marks the beginning of one iteration
5925 through the loop and we can hence easily check right here
5926 whether something matched between the beginning and the end of
5927 the loop. */
5928 case on_failure_jump_nastyloop:
5929 EXTRACT_NUMBER_AND_INCR (mcnt, p);
5930 DEBUG_PRINT3 ("EXECUTING on_failure_jump_nastyloop %d (to %p):\n",
5931 mcnt, p + mcnt);
5932
5933 assert ((re_opcode_t)p[-4] == no_op);
5934 {
5935 int cycle = 0;
5936 CHECK_INFINITE_LOOP (p - 4, d);
5937 if (!cycle)
5938 /* If there's a cycle, just continue without pushing
5939 this failure point. The failure point is the "try again"
5940 option, which shouldn't be tried.
5941 We want (x?)*?y\1z to match both xxyz and xxyxz. */
5942 PUSH_FAILURE_POINT (p - 3, d);
5943 }
5944 break;
5945
5946 /* Simple loop detecting on_failure_jump: just check on the
5947 failure stack if the same spot was already hit earlier. */
5948 case on_failure_jump_loop:
5949 on_failure:
5950 EXTRACT_NUMBER_AND_INCR (mcnt, p);
5951 DEBUG_PRINT3 ("EXECUTING on_failure_jump_loop %d (to %p):\n",
5952 mcnt, p + mcnt);
5953 {
5954 int cycle = 0;
5955 CHECK_INFINITE_LOOP (p - 3, d);
5956 if (cycle)
5957 /* If there's a cycle, get out of the loop, as if the matching
5958 had failed. We used to just `goto fail' here, but that was
5959 aborting the search a bit too early: we want to keep the
5960 empty-loop-match and keep matching after the loop.
5961 We want (x?)*y\1z to match both xxyz and xxyxz. */
5962 p += mcnt;
5963 else
5964 PUSH_FAILURE_POINT (p - 3, d);
5965 }
5966 break;
5967
5968
5969 /* Uses of on_failure_jump:
5970
5971 Each alternative starts with an on_failure_jump that points
5972 to the beginning of the next alternative. Each alternative
5973 except the last ends with a jump that in effect jumps past
5974 the rest of the alternatives. (They really jump to the
5975 ending jump of the following alternative, because tensioning
5976 these jumps is a hassle.)
5977
5978 Repeats start with an on_failure_jump that points past both
5979 the repetition text and either the following jump or
5980 pop_failure_jump back to this on_failure_jump. */
5981 case on_failure_jump:
5982 EXTRACT_NUMBER_AND_INCR (mcnt, p);
5983 DEBUG_PRINT3 ("EXECUTING on_failure_jump %d (to %p):\n",
5984 mcnt, p + mcnt);
5985
5986 PUSH_FAILURE_POINT (p -3, d);
5987 break;
5988
5989 /* This operation is used for greedy *.
5990 Compare the beginning of the repeat with what in the
5991 pattern follows its end. If we can establish that there
5992 is nothing that they would both match, i.e., that we
5993 would have to backtrack because of (as in, e.g., `a*a')
5994 then we can use a non-backtracking loop based on
5995 on_failure_keep_string_jump instead of on_failure_jump. */
5996 case on_failure_jump_smart:
5997 EXTRACT_NUMBER_AND_INCR (mcnt, p);
5998 DEBUG_PRINT3 ("EXECUTING on_failure_jump_smart %d (to %p).\n",
5999 mcnt, p + mcnt);
6000 {
6001 re_char *p1 = p; /* Next operation. */
6002 /* Here, we discard `const', making re_match non-reentrant. */
6003 unsigned char *p2 = (unsigned char*) p + mcnt; /* Jump dest. */
6004 unsigned char *p3 = (unsigned char*) p - 3; /* opcode location. */
6005
6006 p -= 3; /* Reset so that we will re-execute the
6007 instruction once it's been changed. */
6008
6009 EXTRACT_NUMBER (mcnt, p2 - 2);
6010
6011 /* Ensure this is a indeed the trivial kind of loop
6012 we are expecting. */
6013 assert (skip_one_char (p1) == p2 - 3);
6014 assert ((re_opcode_t) p2[-3] == jump && p2 + mcnt == p);
6015 DEBUG_STATEMENT (debug += 2);
6016 if (mutually_exclusive_p (bufp, p1, p2))
6017 {
6018 /* Use a fast `on_failure_keep_string_jump' loop. */
6019 DEBUG_PRINT1 (" smart exclusive => fast loop.\n");
6020 *p3 = (unsigned char) on_failure_keep_string_jump;
6021 STORE_NUMBER (p2 - 2, mcnt + 3);
6022 }
6023 else
6024 {
6025 /* Default to a safe `on_failure_jump' loop. */
6026 DEBUG_PRINT1 (" smart default => slow loop.\n");
6027 *p3 = (unsigned char) on_failure_jump;
6028 }
6029 DEBUG_STATEMENT (debug -= 2);
6030 }
6031 break;
6032
6033 /* Unconditionally jump (without popping any failure points). */
6034 case jump:
6035 unconditional_jump:
6036 IMMEDIATE_QUIT_CHECK;
6037 EXTRACT_NUMBER_AND_INCR (mcnt, p); /* Get the amount to jump. */
6038 DEBUG_PRINT2 ("EXECUTING jump %d ", mcnt);
6039 p += mcnt; /* Do the jump. */
6040 DEBUG_PRINT2 ("(to %p).\n", p);
6041 break;
6042
6043
6044 /* Have to succeed matching what follows at least n times.
6045 After that, handle like `on_failure_jump'. */
6046 case succeed_n:
6047 /* Signedness doesn't matter since we only compare MCNT to 0. */
6048 EXTRACT_NUMBER (mcnt, p + 2);
6049 DEBUG_PRINT2 ("EXECUTING succeed_n %d.\n", mcnt);
6050
6051 /* Originally, mcnt is how many times we HAVE to succeed. */
6052 if (mcnt != 0)
6053 {
6054 /* Here, we discard `const', making re_match non-reentrant. */
6055 unsigned char *p2 = (unsigned char*) p + 2; /* counter loc. */
6056 mcnt--;
6057 p += 4;
6058 PUSH_NUMBER (p2, mcnt);
6059 }
6060 else
6061 /* The two bytes encoding mcnt == 0 are two no_op opcodes. */
6062 goto on_failure;
6063 break;
6064
6065 case jump_n:
6066 /* Signedness doesn't matter since we only compare MCNT to 0. */
6067 EXTRACT_NUMBER (mcnt, p + 2);
6068 DEBUG_PRINT2 ("EXECUTING jump_n %d.\n", mcnt);
6069
6070 /* Originally, this is how many times we CAN jump. */
6071 if (mcnt != 0)
6072 {
6073 /* Here, we discard `const', making re_match non-reentrant. */
6074 unsigned char *p2 = (unsigned char*) p + 2; /* counter loc. */
6075 mcnt--;
6076 PUSH_NUMBER (p2, mcnt);
6077 goto unconditional_jump;
6078 }
6079 /* If don't have to jump any more, skip over the rest of command. */
6080 else
6081 p += 4;
6082 break;
6083
6084 case set_number_at:
6085 {
6086 unsigned char *p2; /* Location of the counter. */
6087 DEBUG_PRINT1 ("EXECUTING set_number_at.\n");
6088
6089 EXTRACT_NUMBER_AND_INCR (mcnt, p);
6090 /* Here, we discard `const', making re_match non-reentrant. */
6091 p2 = (unsigned char*) p + mcnt;
6092 /* Signedness doesn't matter since we only copy MCNT's bits . */
6093 EXTRACT_NUMBER_AND_INCR (mcnt, p);
6094 DEBUG_PRINT3 (" Setting %p to %d.\n", p2, mcnt);
6095 PUSH_NUMBER (p2, mcnt);
6096 break;
6097 }
6098
6099 case wordbound:
6100 case notwordbound:
6101 not = (re_opcode_t) *(p - 1) == notwordbound;
6102 DEBUG_PRINT2 ("EXECUTING %swordbound.\n", not?"not":"");
6103
6104 /* We SUCCEED (or FAIL) in one of the following cases: */
6105
6106 /* Case 1: D is at the beginning or the end of string. */
6107 if (AT_STRINGS_BEG (d) || AT_STRINGS_END (d))
6108 not = !not;
6109 else
6110 {
6111 /* C1 is the character before D, S1 is the syntax of C1, C2
6112 is the character at D, and S2 is the syntax of C2. */
6113 re_wchar_t c1, c2;
6114 int s1, s2;
6115 int dummy;
6116 #ifdef emacs
6117 int offset = PTR_TO_OFFSET (d - 1);
6118 int charpos = SYNTAX_TABLE_BYTE_TO_CHAR (offset);
6119 UPDATE_SYNTAX_TABLE (charpos);
6120 #endif
6121 GET_CHAR_BEFORE_2 (c1, d, string1, end1, string2, end2);
6122 s1 = SYNTAX (c1);
6123 #ifdef emacs
6124 UPDATE_SYNTAX_TABLE_FORWARD (charpos + 1);
6125 #endif
6126 PREFETCH_NOLIMIT ();
6127 GET_CHAR_AFTER (c2, d, dummy);
6128 s2 = SYNTAX (c2);
6129
6130 if (/* Case 2: Only one of S1 and S2 is Sword. */
6131 ((s1 == Sword) != (s2 == Sword))
6132 /* Case 3: Both of S1 and S2 are Sword, and macro
6133 WORD_BOUNDARY_P (C1, C2) returns nonzero. */
6134 || ((s1 == Sword) && WORD_BOUNDARY_P (c1, c2)))
6135 not = !not;
6136 }
6137 if (not)
6138 break;
6139 else
6140 goto fail;
6141
6142 case wordbeg:
6143 DEBUG_PRINT1 ("EXECUTING wordbeg.\n");
6144
6145 /* We FAIL in one of the following cases: */
6146
6147 /* Case 1: D is at the end of string. */
6148 if (AT_STRINGS_END (d))
6149 goto fail;
6150 else
6151 {
6152 /* C1 is the character before D, S1 is the syntax of C1, C2
6153 is the character at D, and S2 is the syntax of C2. */
6154 re_wchar_t c1, c2;
6155 int s1, s2;
6156 int dummy;
6157 #ifdef emacs
6158 int offset = PTR_TO_OFFSET (d);
6159 int charpos = SYNTAX_TABLE_BYTE_TO_CHAR (offset);
6160 UPDATE_SYNTAX_TABLE (charpos);
6161 #endif
6162 PREFETCH ();
6163 GET_CHAR_AFTER (c2, d, dummy);
6164 s2 = SYNTAX (c2);
6165
6166 /* Case 2: S2 is not Sword. */
6167 if (s2 != Sword)
6168 goto fail;
6169
6170 /* Case 3: D is not at the beginning of string ... */
6171 if (!AT_STRINGS_BEG (d))
6172 {
6173 GET_CHAR_BEFORE_2 (c1, d, string1, end1, string2, end2);
6174 #ifdef emacs
6175 UPDATE_SYNTAX_TABLE_BACKWARD (charpos - 1);
6176 #endif
6177 s1 = SYNTAX (c1);
6178
6179 /* ... and S1 is Sword, and WORD_BOUNDARY_P (C1, C2)
6180 returns 0. */
6181 if ((s1 == Sword) && !WORD_BOUNDARY_P (c1, c2))
6182 goto fail;
6183 }
6184 }
6185 break;
6186
6187 case wordend:
6188 DEBUG_PRINT1 ("EXECUTING wordend.\n");
6189
6190 /* We FAIL in one of the following cases: */
6191
6192 /* Case 1: D is at the beginning of string. */
6193 if (AT_STRINGS_BEG (d))
6194 goto fail;
6195 else
6196 {
6197 /* C1 is the character before D, S1 is the syntax of C1, C2
6198 is the character at D, and S2 is the syntax of C2. */
6199 re_wchar_t c1, c2;
6200 int s1, s2;
6201 int dummy;
6202 #ifdef emacs
6203 int offset = PTR_TO_OFFSET (d) - 1;
6204 int charpos = SYNTAX_TABLE_BYTE_TO_CHAR (offset);
6205 UPDATE_SYNTAX_TABLE (charpos);
6206 #endif
6207 GET_CHAR_BEFORE_2 (c1, d, string1, end1, string2, end2);
6208 s1 = SYNTAX (c1);
6209
6210 /* Case 2: S1 is not Sword. */
6211 if (s1 != Sword)
6212 goto fail;
6213
6214 /* Case 3: D is not at the end of string ... */
6215 if (!AT_STRINGS_END (d))
6216 {
6217 PREFETCH_NOLIMIT ();
6218 GET_CHAR_AFTER (c2, d, dummy);
6219 #ifdef emacs
6220 UPDATE_SYNTAX_TABLE_FORWARD (charpos);
6221 #endif
6222 s2 = SYNTAX (c2);
6223
6224 /* ... and S2 is Sword, and WORD_BOUNDARY_P (C1, C2)
6225 returns 0. */
6226 if ((s2 == Sword) && !WORD_BOUNDARY_P (c1, c2))
6227 goto fail;
6228 }
6229 }
6230 break;
6231
6232 case symbeg:
6233 DEBUG_PRINT1 ("EXECUTING symbeg.\n");
6234
6235 /* We FAIL in one of the following cases: */
6236
6237 /* Case 1: D is at the end of string. */
6238 if (AT_STRINGS_END (d))
6239 goto fail;
6240 else
6241 {
6242 /* C1 is the character before D, S1 is the syntax of C1, C2
6243 is the character at D, and S2 is the syntax of C2. */
6244 re_wchar_t c1, c2;
6245 int s1, s2;
6246 #ifdef emacs
6247 int offset = PTR_TO_OFFSET (d);
6248 int charpos = SYNTAX_TABLE_BYTE_TO_CHAR (offset);
6249 UPDATE_SYNTAX_TABLE (charpos);
6250 #endif
6251 PREFETCH ();
6252 c2 = RE_STRING_CHAR (d, dend - d, target_multibyte);
6253 s2 = SYNTAX (c2);
6254
6255 /* Case 2: S2 is neither Sword nor Ssymbol. */
6256 if (s2 != Sword && s2 != Ssymbol)
6257 goto fail;
6258
6259 /* Case 3: D is not at the beginning of string ... */
6260 if (!AT_STRINGS_BEG (d))
6261 {
6262 GET_CHAR_BEFORE_2 (c1, d, string1, end1, string2, end2);
6263 #ifdef emacs
6264 UPDATE_SYNTAX_TABLE_BACKWARD (charpos - 1);
6265 #endif
6266 s1 = SYNTAX (c1);
6267
6268 /* ... and S1 is Sword or Ssymbol. */
6269 if (s1 == Sword || s1 == Ssymbol)
6270 goto fail;
6271 }
6272 }
6273 break;
6274
6275 case symend:
6276 DEBUG_PRINT1 ("EXECUTING symend.\n");
6277
6278 /* We FAIL in one of the following cases: */
6279
6280 /* Case 1: D is at the beginning of string. */
6281 if (AT_STRINGS_BEG (d))
6282 goto fail;
6283 else
6284 {
6285 /* C1 is the character before D, S1 is the syntax of C1, C2
6286 is the character at D, and S2 is the syntax of C2. */
6287 re_wchar_t c1, c2;
6288 int s1, s2;
6289 #ifdef emacs
6290 int offset = PTR_TO_OFFSET (d) - 1;
6291 int charpos = SYNTAX_TABLE_BYTE_TO_CHAR (offset);
6292 UPDATE_SYNTAX_TABLE (charpos);
6293 #endif
6294 GET_CHAR_BEFORE_2 (c1, d, string1, end1, string2, end2);
6295 s1 = SYNTAX (c1);
6296
6297 /* Case 2: S1 is neither Ssymbol nor Sword. */
6298 if (s1 != Sword && s1 != Ssymbol)
6299 goto fail;
6300
6301 /* Case 3: D is not at the end of string ... */
6302 if (!AT_STRINGS_END (d))
6303 {
6304 PREFETCH_NOLIMIT ();
6305 c2 = RE_STRING_CHAR (d, dend - d, target_multibyte);
6306 #ifdef emacs
6307 UPDATE_SYNTAX_TABLE_FORWARD (charpos + 1);
6308 #endif
6309 s2 = SYNTAX (c2);
6310
6311 /* ... and S2 is Sword or Ssymbol. */
6312 if (s2 == Sword || s2 == Ssymbol)
6313 goto fail;
6314 }
6315 }
6316 break;
6317
6318 case syntaxspec:
6319 case notsyntaxspec:
6320 not = (re_opcode_t) *(p - 1) == notsyntaxspec;
6321 mcnt = *p++;
6322 DEBUG_PRINT3 ("EXECUTING %ssyntaxspec %d.\n", not?"not":"", mcnt);
6323 PREFETCH ();
6324 #ifdef emacs
6325 {
6326 int offset = PTR_TO_OFFSET (d);
6327 int pos1 = SYNTAX_TABLE_BYTE_TO_CHAR (offset);
6328 UPDATE_SYNTAX_TABLE (pos1);
6329 }
6330 #endif
6331 {
6332 int len;
6333 re_wchar_t c;
6334
6335 GET_CHAR_AFTER (c, d, len);
6336 if ((SYNTAX (c) != (enum syntaxcode) mcnt) ^ not)
6337 goto fail;
6338 d += len;
6339 }
6340 break;
6341
6342 #ifdef emacs
6343 case before_dot:
6344 DEBUG_PRINT1 ("EXECUTING before_dot.\n");
6345 if (PTR_BYTE_POS (d) >= PT_BYTE)
6346 goto fail;
6347 break;
6348
6349 case at_dot:
6350 DEBUG_PRINT1 ("EXECUTING at_dot.\n");
6351 if (PTR_BYTE_POS (d) != PT_BYTE)
6352 goto fail;
6353 break;
6354
6355 case after_dot:
6356 DEBUG_PRINT1 ("EXECUTING after_dot.\n");
6357 if (PTR_BYTE_POS (d) <= PT_BYTE)
6358 goto fail;
6359 break;
6360
6361 case categoryspec:
6362 case notcategoryspec:
6363 not = (re_opcode_t) *(p - 1) == notcategoryspec;
6364 mcnt = *p++;
6365 DEBUG_PRINT3 ("EXECUTING %scategoryspec %d.\n", not?"not":"", mcnt);
6366 PREFETCH ();
6367 {
6368 int len;
6369 re_wchar_t c;
6370
6371 GET_CHAR_AFTER (c, d, len);
6372 if ((!CHAR_HAS_CATEGORY (c, mcnt)) ^ not)
6373 goto fail;
6374 d += len;
6375 }
6376 break;
6377
6378 #endif /* emacs */
6379
6380 default:
6381 abort ();
6382 }
6383 continue; /* Successfully executed one pattern command; keep going. */
6384
6385
6386 /* We goto here if a matching operation fails. */
6387 fail:
6388 IMMEDIATE_QUIT_CHECK;
6389 if (!FAIL_STACK_EMPTY ())
6390 {
6391 re_char *str, *pat;
6392 /* A restart point is known. Restore to that state. */
6393 DEBUG_PRINT1 ("\nFAIL:\n");
6394 POP_FAILURE_POINT (str, pat);
6395 switch (SWITCH_ENUM_CAST ((re_opcode_t) *pat++))
6396 {
6397 case on_failure_keep_string_jump:
6398 assert (str == NULL);
6399 goto continue_failure_jump;
6400
6401 case on_failure_jump_nastyloop:
6402 assert ((re_opcode_t)pat[-2] == no_op);
6403 PUSH_FAILURE_POINT (pat - 2, str);
6404 /* Fallthrough */
6405
6406 case on_failure_jump_loop:
6407 case on_failure_jump:
6408 case succeed_n:
6409 d = str;
6410 continue_failure_jump:
6411 EXTRACT_NUMBER_AND_INCR (mcnt, pat);
6412 p = pat + mcnt;
6413 break;
6414
6415 case no_op:
6416 /* A special frame used for nastyloops. */
6417 goto fail;
6418
6419 default:
6420 abort();
6421 }
6422
6423 assert (p >= bufp->buffer && p <= pend);
6424
6425 if (d >= string1 && d <= end1)
6426 dend = end_match_1;
6427 }
6428 else
6429 break; /* Matching at this starting point really fails. */
6430 } /* for (;;) */
6431
6432 if (best_regs_set)
6433 goto restore_best_regs;
6434
6435 FREE_VARIABLES ();
6436
6437 return -1; /* Failure to match. */
6438 } /* re_match_2 */
6439 \f
6440 /* Subroutine definitions for re_match_2. */
6441
6442 /* Return zero if TRANSLATE[S1] and TRANSLATE[S2] are identical for LEN
6443 bytes; nonzero otherwise. */
6444
6445 static int
6446 bcmp_translate (s1, s2, len, translate, target_multibyte)
6447 re_char *s1, *s2;
6448 register int len;
6449 RE_TRANSLATE_TYPE translate;
6450 const int target_multibyte;
6451 {
6452 register re_char *p1 = s1, *p2 = s2;
6453 re_char *p1_end = s1 + len;
6454 re_char *p2_end = s2 + len;
6455
6456 /* FIXME: Checking both p1 and p2 presumes that the two strings might have
6457 different lengths, but relying on a single `len' would break this. -sm */
6458 while (p1 < p1_end && p2 < p2_end)
6459 {
6460 int p1_charlen, p2_charlen;
6461 re_wchar_t p1_ch, p2_ch;
6462
6463 GET_CHAR_AFTER (p1_ch, p1, p1_charlen);
6464 GET_CHAR_AFTER (p2_ch, p2, p2_charlen);
6465
6466 if (RE_TRANSLATE (translate, p1_ch)
6467 != RE_TRANSLATE (translate, p2_ch))
6468 return 1;
6469
6470 p1 += p1_charlen, p2 += p2_charlen;
6471 }
6472
6473 if (p1 != p1_end || p2 != p2_end)
6474 return 1;
6475
6476 return 0;
6477 }
6478 \f
6479 /* Entry points for GNU code. */
6480
6481 /* re_compile_pattern is the GNU regular expression compiler: it
6482 compiles PATTERN (of length SIZE) and puts the result in BUFP.
6483 Returns 0 if the pattern was valid, otherwise an error string.
6484
6485 Assumes the `allocated' (and perhaps `buffer') and `translate' fields
6486 are set in BUFP on entry.
6487
6488 We call regex_compile to do the actual compilation. */
6489
6490 const char *
6491 re_compile_pattern (pattern, length, bufp)
6492 const char *pattern;
6493 size_t length;
6494 struct re_pattern_buffer *bufp;
6495 {
6496 reg_errcode_t ret;
6497
6498 #ifdef emacs
6499 gl_state.current_syntax_table = current_buffer->syntax_table;
6500 #endif
6501
6502 /* GNU code is written to assume at least RE_NREGS registers will be set
6503 (and at least one extra will be -1). */
6504 bufp->regs_allocated = REGS_UNALLOCATED;
6505
6506 /* And GNU code determines whether or not to get register information
6507 by passing null for the REGS argument to re_match, etc., not by
6508 setting no_sub. */
6509 bufp->no_sub = 0;
6510
6511 ret = regex_compile ((re_char*) pattern, length, re_syntax_options, bufp);
6512
6513 if (!ret)
6514 return NULL;
6515 return gettext (re_error_msgid[(int) ret]);
6516 }
6517 WEAK_ALIAS (__re_compile_pattern, re_compile_pattern)
6518 \f
6519 /* Entry points compatible with 4.2 BSD regex library. We don't define
6520 them unless specifically requested. */
6521
6522 #if defined _REGEX_RE_COMP || defined _LIBC
6523
6524 /* BSD has one and only one pattern buffer. */
6525 static struct re_pattern_buffer re_comp_buf;
6526
6527 char *
6528 # ifdef _LIBC
6529 /* Make these definitions weak in libc, so POSIX programs can redefine
6530 these names if they don't use our functions, and still use
6531 regcomp/regexec below without link errors. */
6532 weak_function
6533 # endif
6534 re_comp (s)
6535 const char *s;
6536 {
6537 reg_errcode_t ret;
6538
6539 if (!s)
6540 {
6541 if (!re_comp_buf.buffer)
6542 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6543 return (char *) gettext ("No previous regular expression");
6544 return 0;
6545 }
6546
6547 if (!re_comp_buf.buffer)
6548 {
6549 re_comp_buf.buffer = (unsigned char *) malloc (200);
6550 if (re_comp_buf.buffer == NULL)
6551 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6552 return (char *) gettext (re_error_msgid[(int) REG_ESPACE]);
6553 re_comp_buf.allocated = 200;
6554
6555 re_comp_buf.fastmap = (char *) malloc (1 << BYTEWIDTH);
6556 if (re_comp_buf.fastmap == NULL)
6557 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6558 return (char *) gettext (re_error_msgid[(int) REG_ESPACE]);
6559 }
6560
6561 /* Since `re_exec' always passes NULL for the `regs' argument, we
6562 don't need to initialize the pattern buffer fields which affect it. */
6563
6564 ret = regex_compile (s, strlen (s), re_syntax_options, &re_comp_buf);
6565
6566 if (!ret)
6567 return NULL;
6568
6569 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6570 return (char *) gettext (re_error_msgid[(int) ret]);
6571 }
6572
6573
6574 int
6575 # ifdef _LIBC
6576 weak_function
6577 # endif
6578 re_exec (s)
6579 const char *s;
6580 {
6581 const int len = strlen (s);
6582 return
6583 0 <= re_search (&re_comp_buf, s, len, 0, len, (struct re_registers *) 0);
6584 }
6585 #endif /* _REGEX_RE_COMP */
6586 \f
6587 /* POSIX.2 functions. Don't define these for Emacs. */
6588
6589 #ifndef emacs
6590
6591 /* regcomp takes a regular expression as a string and compiles it.
6592
6593 PREG is a regex_t *. We do not expect any fields to be initialized,
6594 since POSIX says we shouldn't. Thus, we set
6595
6596 `buffer' to the compiled pattern;
6597 `used' to the length of the compiled pattern;
6598 `syntax' to RE_SYNTAX_POSIX_EXTENDED if the
6599 REG_EXTENDED bit in CFLAGS is set; otherwise, to
6600 RE_SYNTAX_POSIX_BASIC;
6601 `fastmap' to an allocated space for the fastmap;
6602 `fastmap_accurate' to zero;
6603 `re_nsub' to the number of subexpressions in PATTERN.
6604
6605 PATTERN is the address of the pattern string.
6606
6607 CFLAGS is a series of bits which affect compilation.
6608
6609 If REG_EXTENDED is set, we use POSIX extended syntax; otherwise, we
6610 use POSIX basic syntax.
6611
6612 If REG_NEWLINE is set, then . and [^...] don't match newline.
6613 Also, regexec will try a match beginning after every newline.
6614
6615 If REG_ICASE is set, then we considers upper- and lowercase
6616 versions of letters to be equivalent when matching.
6617
6618 If REG_NOSUB is set, then when PREG is passed to regexec, that
6619 routine will report only success or failure, and nothing about the
6620 registers.
6621
6622 It returns 0 if it succeeds, nonzero if it doesn't. (See regex.h for
6623 the return codes and their meanings.) */
6624
6625 int
6626 regcomp (preg, pattern, cflags)
6627 regex_t *__restrict preg;
6628 const char *__restrict pattern;
6629 int cflags;
6630 {
6631 reg_errcode_t ret;
6632 reg_syntax_t syntax
6633 = (cflags & REG_EXTENDED) ?
6634 RE_SYNTAX_POSIX_EXTENDED : RE_SYNTAX_POSIX_BASIC;
6635
6636 /* regex_compile will allocate the space for the compiled pattern. */
6637 preg->buffer = 0;
6638 preg->allocated = 0;
6639 preg->used = 0;
6640
6641 /* Try to allocate space for the fastmap. */
6642 preg->fastmap = (char *) malloc (1 << BYTEWIDTH);
6643
6644 if (cflags & REG_ICASE)
6645 {
6646 unsigned i;
6647
6648 preg->translate
6649 = (RE_TRANSLATE_TYPE) malloc (CHAR_SET_SIZE
6650 * sizeof (*(RE_TRANSLATE_TYPE)0));
6651 if (preg->translate == NULL)
6652 return (int) REG_ESPACE;
6653
6654 /* Map uppercase characters to corresponding lowercase ones. */
6655 for (i = 0; i < CHAR_SET_SIZE; i++)
6656 preg->translate[i] = ISUPPER (i) ? TOLOWER (i) : i;
6657 }
6658 else
6659 preg->translate = NULL;
6660
6661 /* If REG_NEWLINE is set, newlines are treated differently. */
6662 if (cflags & REG_NEWLINE)
6663 { /* REG_NEWLINE implies neither . nor [^...] match newline. */
6664 syntax &= ~RE_DOT_NEWLINE;
6665 syntax |= RE_HAT_LISTS_NOT_NEWLINE;
6666 }
6667 else
6668 syntax |= RE_NO_NEWLINE_ANCHOR;
6669
6670 preg->no_sub = !!(cflags & REG_NOSUB);
6671
6672 /* POSIX says a null character in the pattern terminates it, so we
6673 can use strlen here in compiling the pattern. */
6674 ret = regex_compile ((re_char*) pattern, strlen (pattern), syntax, preg);
6675
6676 /* POSIX doesn't distinguish between an unmatched open-group and an
6677 unmatched close-group: both are REG_EPAREN. */
6678 if (ret == REG_ERPAREN)
6679 ret = REG_EPAREN;
6680
6681 if (ret == REG_NOERROR && preg->fastmap)
6682 { /* Compute the fastmap now, since regexec cannot modify the pattern
6683 buffer. */
6684 re_compile_fastmap (preg);
6685 if (preg->can_be_null)
6686 { /* The fastmap can't be used anyway. */
6687 free (preg->fastmap);
6688 preg->fastmap = NULL;
6689 }
6690 }
6691 return (int) ret;
6692 }
6693 WEAK_ALIAS (__regcomp, regcomp)
6694
6695
6696 /* regexec searches for a given pattern, specified by PREG, in the
6697 string STRING.
6698
6699 If NMATCH is zero or REG_NOSUB was set in the cflags argument to
6700 `regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at
6701 least NMATCH elements, and we set them to the offsets of the
6702 corresponding matched substrings.
6703
6704 EFLAGS specifies `execution flags' which affect matching: if
6705 REG_NOTBOL is set, then ^ does not match at the beginning of the
6706 string; if REG_NOTEOL is set, then $ does not match at the end.
6707
6708 We return 0 if we find a match and REG_NOMATCH if not. */
6709
6710 int
6711 regexec (preg, string, nmatch, pmatch, eflags)
6712 const regex_t *__restrict preg;
6713 const char *__restrict string;
6714 size_t nmatch;
6715 regmatch_t pmatch[__restrict_arr];
6716 int eflags;
6717 {
6718 int ret;
6719 struct re_registers regs;
6720 regex_t private_preg;
6721 int len = strlen (string);
6722 boolean want_reg_info = !preg->no_sub && nmatch > 0 && pmatch;
6723
6724 private_preg = *preg;
6725
6726 private_preg.not_bol = !!(eflags & REG_NOTBOL);
6727 private_preg.not_eol = !!(eflags & REG_NOTEOL);
6728
6729 /* The user has told us exactly how many registers to return
6730 information about, via `nmatch'. We have to pass that on to the
6731 matching routines. */
6732 private_preg.regs_allocated = REGS_FIXED;
6733
6734 if (want_reg_info)
6735 {
6736 regs.num_regs = nmatch;
6737 regs.start = TALLOC (nmatch * 2, regoff_t);
6738 if (regs.start == NULL)
6739 return (int) REG_NOMATCH;
6740 regs.end = regs.start + nmatch;
6741 }
6742
6743 /* Instead of using not_eol to implement REG_NOTEOL, we could simply
6744 pass (&private_preg, string, len + 1, 0, len, ...) pretending the string
6745 was a little bit longer but still only matching the real part.
6746 This works because the `endline' will check for a '\n' and will find a
6747 '\0', correctly deciding that this is not the end of a line.
6748 But it doesn't work out so nicely for REG_NOTBOL, since we don't have
6749 a convenient '\0' there. For all we know, the string could be preceded
6750 by '\n' which would throw things off. */
6751
6752 /* Perform the searching operation. */
6753 ret = re_search (&private_preg, string, len,
6754 /* start: */ 0, /* range: */ len,
6755 want_reg_info ? &regs : (struct re_registers *) 0);
6756
6757 /* Copy the register information to the POSIX structure. */
6758 if (want_reg_info)
6759 {
6760 if (ret >= 0)
6761 {
6762 unsigned r;
6763
6764 for (r = 0; r < nmatch; r++)
6765 {
6766 pmatch[r].rm_so = regs.start[r];
6767 pmatch[r].rm_eo = regs.end[r];
6768 }
6769 }
6770
6771 /* If we needed the temporary register info, free the space now. */
6772 free (regs.start);
6773 }
6774
6775 /* We want zero return to mean success, unlike `re_search'. */
6776 return ret >= 0 ? (int) REG_NOERROR : (int) REG_NOMATCH;
6777 }
6778 WEAK_ALIAS (__regexec, regexec)
6779
6780
6781 /* Returns a message corresponding to an error code, ERR_CODE, returned
6782 from either regcomp or regexec. We don't use PREG here.
6783
6784 ERR_CODE was previously called ERRCODE, but that name causes an
6785 error with msvc8 compiler. */
6786
6787 size_t
6788 regerror (err_code, preg, errbuf, errbuf_size)
6789 int err_code;
6790 const regex_t *preg;
6791 char *errbuf;
6792 size_t errbuf_size;
6793 {
6794 const char *msg;
6795 size_t msg_size;
6796
6797 if (err_code < 0
6798 || err_code >= (sizeof (re_error_msgid) / sizeof (re_error_msgid[0])))
6799 /* Only error codes returned by the rest of the code should be passed
6800 to this routine. If we are given anything else, or if other regex
6801 code generates an invalid error code, then the program has a bug.
6802 Dump core so we can fix it. */
6803 abort ();
6804
6805 msg = gettext (re_error_msgid[err_code]);
6806
6807 msg_size = strlen (msg) + 1; /* Includes the null. */
6808
6809 if (errbuf_size != 0)
6810 {
6811 if (msg_size > errbuf_size)
6812 {
6813 strncpy (errbuf, msg, errbuf_size - 1);
6814 errbuf[errbuf_size - 1] = 0;
6815 }
6816 else
6817 strcpy (errbuf, msg);
6818 }
6819
6820 return msg_size;
6821 }
6822 WEAK_ALIAS (__regerror, regerror)
6823
6824
6825 /* Free dynamically allocated space used by PREG. */
6826
6827 void
6828 regfree (preg)
6829 regex_t *preg;
6830 {
6831 if (preg->buffer != NULL)
6832 free (preg->buffer);
6833 preg->buffer = NULL;
6834
6835 preg->allocated = 0;
6836 preg->used = 0;
6837
6838 if (preg->fastmap != NULL)
6839 free (preg->fastmap);
6840 preg->fastmap = NULL;
6841 preg->fastmap_accurate = 0;
6842
6843 if (preg->translate != NULL)
6844 free (preg->translate);
6845 preg->translate = NULL;
6846 }
6847 WEAK_ALIAS (__regfree, regfree)
6848
6849 #endif /* not emacs */
6850
6851 /* arch-tag: 4ffd68ba-2a9e-435b-a21a-018990f9eeb2
6852 (do not change this comment) */
Emacs repository mirror