?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
(rmail-convert-babyl-to-mbox): Bind coding-system-for-read to `raw-text', and
[emacs.git] / src / regex.c
blob91259dc77942602eaebaf6fb3dcc186467faf806
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, 2009
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) CHAR_TO_BYTE_SAFE (c)
157
158 /* Set C a (possibly converted to multibyte) character before P. P
159 points into a string which is the virtual concatenation of STR1
160 (which ends at END1) or STR2 (which ends at END2). */
161 # define GET_CHAR_BEFORE_2(c, p, str1, end1, str2, end2) \
162 do { \
163 if (target_multibyte) \
164 { \
165 re_char *dtemp = (p) == (str2) ? (end1) : (p); \
166 re_char *dlimit = ((p) > (str2) && (p) <= (end2)) ? (str2) : (str1); \
167 while (dtemp-- > dlimit && !CHAR_HEAD_P (*dtemp)); \
168 c = STRING_CHAR (dtemp, (p) - dtemp); \
169 } \
170 else \
171 { \
172 (c = ((p) == (str2) ? (end1) : (p))[-1]); \
173 (c) = RE_CHAR_TO_MULTIBYTE (c); \
174 } \
175 } while (0)
176
177 /* Set C a (possibly converted to multibyte) character at P, and set
178 LEN to the byte length of that character. */
179 # define GET_CHAR_AFTER(c, p, len) \
180 do { \
181 if (target_multibyte) \
182 (c) = STRING_CHAR_AND_LENGTH (p, 0, len); \
183 else \
184 { \
185 (c) = *p; \
186 len = 1; \
187 (c) = RE_CHAR_TO_MULTIBYTE (c); \
188 } \
189 } while (0)
190
191 #else /* not emacs */
192
193 /* If we are not linking with Emacs proper,
194 we can't use the relocating allocator
195 even if config.h says that we can. */
196 # undef REL_ALLOC
197
198 # if defined STDC_HEADERS || defined _LIBC
199 # include <stdlib.h>
200 # else
201 char *malloc ();
202 char *realloc ();
203 # endif
204
205 /* When used in Emacs's lib-src, we need xmalloc and xrealloc. */
206
207 void *
208 xmalloc (size)
209 size_t size;
210 {
211 register void *val;
212 val = (void *) malloc (size);
213 if (!val && size)
214 {
215 write (2, "virtual memory exhausted\n", 25);
216 exit (1);
217 }
218 return val;
219 }
220
221 void *
222 xrealloc (block, size)
223 void *block;
224 size_t size;
225 {
226 register void *val;
227 /* We must call malloc explicitly when BLOCK is 0, since some
228 reallocs don't do this. */
229 if (! block)
230 val = (void *) malloc (size);
231 else
232 val = (void *) realloc (block, size);
233 if (!val && size)
234 {
235 write (2, "virtual memory exhausted\n", 25);
236 exit (1);
237 }
238 return val;
239 }
240
241 # ifdef malloc
242 # undef malloc
243 # endif
244 # define malloc xmalloc
245 # ifdef realloc
246 # undef realloc
247 # endif
248 # define realloc xrealloc
249
250 /* When used in Emacs's lib-src, we need to get bzero and bcopy somehow.
251 If nothing else has been done, use the method below. */
252 # ifdef INHIBIT_STRING_HEADER
253 # if !(defined HAVE_BZERO && defined HAVE_BCOPY)
254 # if !defined bzero && !defined bcopy
255 # undef INHIBIT_STRING_HEADER
256 # endif
257 # endif
258 # endif
259
260 /* This is the normal way of making sure we have memcpy, memcmp and bzero.
261 This is used in most programs--a few other programs avoid this
262 by defining INHIBIT_STRING_HEADER. */
263 # ifndef INHIBIT_STRING_HEADER
264 # if defined HAVE_STRING_H || defined STDC_HEADERS || defined _LIBC
265 # include <string.h>
266 # ifndef bzero
267 # ifndef _LIBC
268 # define bzero(s, n) (memset (s, '\0', n), (s))
269 # else
270 # define bzero(s, n) __bzero (s, n)
271 # endif
272 # endif
273 # else
274 # include <strings.h>
275 # ifndef memcmp
276 # define memcmp(s1, s2, n) bcmp (s1, s2, n)
277 # endif
278 # ifndef memcpy
279 # define memcpy(d, s, n) (bcopy (s, d, n), (d))
280 # endif
281 # endif
282 # endif
283
284 /* Define the syntax stuff for \<, \>, etc. */
285
286 /* Sword must be nonzero for the wordchar pattern commands in re_match_2. */
287 enum syntaxcode { Swhitespace = 0, Sword = 1, Ssymbol = 2 };
288
289 # define SWITCH_ENUM_CAST(x) (x)
290
291 /* Dummy macros for non-Emacs environments. */
292 # define BASE_LEADING_CODE_P(c) (0)
293 # define CHAR_CHARSET(c) 0
294 # define CHARSET_LEADING_CODE_BASE(c) 0
295 # define MAX_MULTIBYTE_LENGTH 1
296 # define RE_MULTIBYTE_P(x) 0
297 # define RE_TARGET_MULTIBYTE_P(x) 0
298 # define WORD_BOUNDARY_P(c1, c2) (0)
299 # define CHAR_HEAD_P(p) (1)
300 # define SINGLE_BYTE_CHAR_P(c) (1)
301 # define SAME_CHARSET_P(c1, c2) (1)
302 # define MULTIBYTE_FORM_LENGTH(p, s) (1)
303 # define PREV_CHAR_BOUNDARY(p, limit) ((p)--)
304 # define STRING_CHAR(p, s) (*(p))
305 # define RE_STRING_CHAR(p, s, multibyte) STRING_CHAR ((p), (s))
306 # define CHAR_STRING(c, s) (*(s) = (c), 1)
307 # define STRING_CHAR_AND_LENGTH(p, s, actual_len) ((actual_len) = 1, *(p))
308 # define RE_STRING_CHAR_AND_LENGTH(p, s, len, multibyte) STRING_CHAR_AND_LENGTH ((p), (s), (len))
309 # define RE_CHAR_TO_MULTIBYTE(c) (c)
310 # define RE_CHAR_TO_UNIBYTE(c) (c)
311 # define GET_CHAR_BEFORE_2(c, p, str1, end1, str2, end2) \
312 (c = ((p) == (str2) ? *((end1) - 1) : *((p) - 1)))
313 # define GET_CHAR_AFTER(c, p, len) \
314 (c = *p, len = 1)
315 # define MAKE_CHAR(charset, c1, c2) (c1)
316 # define BYTE8_TO_CHAR(c) (c)
317 # define CHAR_BYTE8_P(c) (0)
318 # define CHAR_LEADING_CODE(c) (c)
319
320 #endif /* not emacs */
321
322 #ifndef RE_TRANSLATE
323 # define RE_TRANSLATE(TBL, C) ((unsigned char)(TBL)[C])
324 # define RE_TRANSLATE_P(TBL) (TBL)
325 #endif
326 \f
327 /* Get the interface, including the syntax bits. */
328 #include "regex.h"
329
330 /* isalpha etc. are used for the character classes. */
331 #include <ctype.h>
332
333 #ifdef emacs
334
335 /* 1 if C is an ASCII character. */
336 # define IS_REAL_ASCII(c) ((c) < 0200)
337
338 /* 1 if C is a unibyte character. */
339 # define ISUNIBYTE(c) (SINGLE_BYTE_CHAR_P ((c)))
340
341 /* The Emacs definitions should not be directly affected by locales. */
342
343 /* In Emacs, these are only used for single-byte characters. */
344 # define ISDIGIT(c) ((c) >= '0' && (c) <= '9')
345 # define ISCNTRL(c) ((c) < ' ')
346 # define ISXDIGIT(c) (((c) >= '0' && (c) <= '9') \
347 || ((c) >= 'a' && (c) <= 'f') \
348 || ((c) >= 'A' && (c) <= 'F'))
349
350 /* This is only used for single-byte characters. */
351 # define ISBLANK(c) ((c) == ' ' || (c) == '\t')
352
353 /* The rest must handle multibyte characters. */
354
355 # define ISGRAPH(c) (SINGLE_BYTE_CHAR_P (c) \
356 ? (c) > ' ' && !((c) >= 0177 && (c) <= 0237) \
357 : 1)
358
359 # define ISPRINT(c) (SINGLE_BYTE_CHAR_P (c) \
360 ? (c) >= ' ' && !((c) >= 0177 && (c) <= 0237) \
361 : 1)
362
363 # define ISALNUM(c) (IS_REAL_ASCII (c) \
364 ? (((c) >= 'a' && (c) <= 'z') \
365 || ((c) >= 'A' && (c) <= 'Z') \
366 || ((c) >= '0' && (c) <= '9')) \
367 : SYNTAX (c) == Sword)
368
369 # define ISALPHA(c) (IS_REAL_ASCII (c) \
370 ? (((c) >= 'a' && (c) <= 'z') \
371 || ((c) >= 'A' && (c) <= 'Z')) \
372 : SYNTAX (c) == Sword)
373
374 # define ISLOWER(c) (LOWERCASEP (c))
375
376 # define ISPUNCT(c) (IS_REAL_ASCII (c) \
377 ? ((c) > ' ' && (c) < 0177 \
378 && !(((c) >= 'a' && (c) <= 'z') \
379 || ((c) >= 'A' && (c) <= 'Z') \
380 || ((c) >= '0' && (c) <= '9'))) \
381 : SYNTAX (c) != Sword)
382
383 # define ISSPACE(c) (SYNTAX (c) == Swhitespace)
384
385 # define ISUPPER(c) (UPPERCASEP (c))
386
387 # define ISWORD(c) (SYNTAX (c) == Sword)
388
389 #else /* not emacs */
390
391 /* Jim Meyering writes:
392
393 "... Some ctype macros are valid only for character codes that
394 isascii says are ASCII (SGI's IRIX-4.0.5 is one such system --when
395 using /bin/cc or gcc but without giving an ansi option). So, all
396 ctype uses should be through macros like ISPRINT... If
397 STDC_HEADERS is defined, then autoconf has verified that the ctype
398 macros don't need to be guarded with references to isascii. ...
399 Defining isascii to 1 should let any compiler worth its salt
400 eliminate the && through constant folding."
401 Solaris defines some of these symbols so we must undefine them first. */
402
403 # undef ISASCII
404 # if defined STDC_HEADERS || (!defined isascii && !defined HAVE_ISASCII)
405 # define ISASCII(c) 1
406 # else
407 # define ISASCII(c) isascii(c)
408 # endif
409
410 /* 1 if C is an ASCII character. */
411 # define IS_REAL_ASCII(c) ((c) < 0200)
412
413 /* This distinction is not meaningful, except in Emacs. */
414 # define ISUNIBYTE(c) 1
415
416 # ifdef isblank
417 # define ISBLANK(c) (ISASCII (c) && isblank (c))
418 # else
419 # define ISBLANK(c) ((c) == ' ' || (c) == '\t')
420 # endif
421 # ifdef isgraph
422 # define ISGRAPH(c) (ISASCII (c) && isgraph (c))
423 # else
424 # define ISGRAPH(c) (ISASCII (c) && isprint (c) && !isspace (c))
425 # endif
426
427 # undef ISPRINT
428 # define ISPRINT(c) (ISASCII (c) && isprint (c))
429 # define ISDIGIT(c) (ISASCII (c) && isdigit (c))
430 # define ISALNUM(c) (ISASCII (c) && isalnum (c))
431 # define ISALPHA(c) (ISASCII (c) && isalpha (c))
432 # define ISCNTRL(c) (ISASCII (c) && iscntrl (c))
433 # define ISLOWER(c) (ISASCII (c) && islower (c))
434 # define ISPUNCT(c) (ISASCII (c) && ispunct (c))
435 # define ISSPACE(c) (ISASCII (c) && isspace (c))
436 # define ISUPPER(c) (ISASCII (c) && isupper (c))
437 # define ISXDIGIT(c) (ISASCII (c) && isxdigit (c))
438
439 # define ISWORD(c) ISALPHA(c)
440
441 # ifdef _tolower
442 # define TOLOWER(c) _tolower(c)
443 # else
444 # define TOLOWER(c) tolower(c)
445 # endif
446
447 /* How many characters in the character set. */
448 # define CHAR_SET_SIZE 256
449
450 # ifdef SYNTAX_TABLE
451
452 extern char *re_syntax_table;
453
454 # else /* not SYNTAX_TABLE */
455
456 static char re_syntax_table[CHAR_SET_SIZE];
457
458 static void
459 init_syntax_once ()
460 {
461 register int c;
462 static int done = 0;
463
464 if (done)
465 return;
466
467 bzero (re_syntax_table, sizeof re_syntax_table);
468
469 for (c = 0; c < CHAR_SET_SIZE; ++c)
470 if (ISALNUM (c))
471 re_syntax_table[c] = Sword;
472
473 re_syntax_table['_'] = Ssymbol;
474
475 done = 1;
476 }
477
478 # endif /* not SYNTAX_TABLE */
479
480 # define SYNTAX(c) re_syntax_table[(c)]
481
482 #endif /* not emacs */
483 \f
484 #ifndef NULL
485 # define NULL (void *)0
486 #endif
487
488 /* We remove any previous definition of `SIGN_EXTEND_CHAR',
489 since ours (we hope) works properly with all combinations of
490 machines, compilers, `char' and `unsigned char' argument types.
491 (Per Bothner suggested the basic approach.) */
492 #undef SIGN_EXTEND_CHAR
493 #if __STDC__
494 # define SIGN_EXTEND_CHAR(c) ((signed char) (c))
495 #else /* not __STDC__ */
496 /* As in Harbison and Steele. */
497 # define SIGN_EXTEND_CHAR(c) ((((unsigned char) (c)) ^ 128) - 128)
498 #endif
499 \f
500 /* Should we use malloc or alloca? If REGEX_MALLOC is not defined, we
501 use `alloca' instead of `malloc'. This is because using malloc in
502 re_search* or re_match* could cause memory leaks when C-g is used in
503 Emacs; also, malloc is slower and causes storage fragmentation. On
504 the other hand, malloc is more portable, and easier to debug.
505
506 Because we sometimes use alloca, some routines have to be macros,
507 not functions -- `alloca'-allocated space disappears at the end of the
508 function it is called in. */
509
510 #ifdef REGEX_MALLOC
511
512 # define REGEX_ALLOCATE malloc
513 # define REGEX_REALLOCATE(source, osize, nsize) realloc (source, nsize)
514 # define REGEX_FREE free
515
516 #else /* not REGEX_MALLOC */
517
518 /* Emacs already defines alloca, sometimes. */
519 # ifndef alloca
520
521 /* Make alloca work the best possible way. */
522 # ifdef __GNUC__
523 # define alloca __builtin_alloca
524 # else /* not __GNUC__ */
525 # ifdef HAVE_ALLOCA_H
526 # include <alloca.h>
527 # endif /* HAVE_ALLOCA_H */
528 # endif /* not __GNUC__ */
529
530 # endif /* not alloca */
531
532 # define REGEX_ALLOCATE alloca
533
534 /* Assumes a `char *destination' variable. */
535 # define REGEX_REALLOCATE(source, osize, nsize) \
536 (destination = (char *) alloca (nsize), \
537 memcpy (destination, source, osize))
538
539 /* No need to do anything to free, after alloca. */
540 # define REGEX_FREE(arg) ((void)0) /* Do nothing! But inhibit gcc warning. */
541
542 #endif /* not REGEX_MALLOC */
543
544 /* Define how to allocate the failure stack. */
545
546 #if defined REL_ALLOC && defined REGEX_MALLOC
547
548 # define REGEX_ALLOCATE_STACK(size) \
549 r_alloc (&failure_stack_ptr, (size))
550 # define REGEX_REALLOCATE_STACK(source, osize, nsize) \
551 r_re_alloc (&failure_stack_ptr, (nsize))
552 # define REGEX_FREE_STACK(ptr) \
553 r_alloc_free (&failure_stack_ptr)
554
555 #else /* not using relocating allocator */
556
557 # ifdef REGEX_MALLOC
558
559 # define REGEX_ALLOCATE_STACK malloc
560 # define REGEX_REALLOCATE_STACK(source, osize, nsize) realloc (source, nsize)
561 # define REGEX_FREE_STACK free
562
563 # else /* not REGEX_MALLOC */
564
565 # define REGEX_ALLOCATE_STACK alloca
566
567 # define REGEX_REALLOCATE_STACK(source, osize, nsize) \
568 REGEX_REALLOCATE (source, osize, nsize)
569 /* No need to explicitly free anything. */
570 # define REGEX_FREE_STACK(arg) ((void)0)
571
572 # endif /* not REGEX_MALLOC */
573 #endif /* not using relocating allocator */
574
575
576 /* True if `size1' is non-NULL and PTR is pointing anywhere inside
577 `string1' or just past its end. This works if PTR is NULL, which is
578 a good thing. */
579 #define FIRST_STRING_P(ptr) \
580 (size1 && string1 <= (ptr) && (ptr) <= string1 + size1)
581
582 /* (Re)Allocate N items of type T using malloc, or fail. */
583 #define TALLOC(n, t) ((t *) malloc ((n) * sizeof (t)))
584 #define RETALLOC(addr, n, t) ((addr) = (t *) realloc (addr, (n) * sizeof (t)))
585 #define RETALLOC_IF(addr, n, t) \
586 if (addr) RETALLOC((addr), (n), t); else (addr) = TALLOC ((n), t)
587 #define REGEX_TALLOC(n, t) ((t *) REGEX_ALLOCATE ((n) * sizeof (t)))
588
589 #define BYTEWIDTH 8 /* In bits. */
590
591 #define STREQ(s1, s2) ((strcmp (s1, s2) == 0))
592
593 #undef MAX
594 #undef MIN
595 #define MAX(a, b) ((a) > (b) ? (a) : (b))
596 #define MIN(a, b) ((a) < (b) ? (a) : (b))
597
598 /* Type of source-pattern and string chars. */
599 typedef const unsigned char re_char;
600
601 typedef char boolean;
602 #define false 0
603 #define true 1
604
605 static int re_match_2_internal _RE_ARGS ((struct re_pattern_buffer *bufp,
606 re_char *string1, int size1,
607 re_char *string2, int size2,
608 int pos,
609 struct re_registers *regs,
610 int stop));
611 \f
612 /* These are the command codes that appear in compiled regular
613 expressions. Some opcodes are followed by argument bytes. A
614 command code can specify any interpretation whatsoever for its
615 arguments. Zero bytes may appear in the compiled regular expression. */
616
617 typedef enum
618 {
619 no_op = 0,
620
621 /* Succeed right away--no more backtracking. */
622 succeed,
623
624 /* Followed by one byte giving n, then by n literal bytes. */
625 exactn,
626
627 /* Matches any (more or less) character. */
628 anychar,
629
630 /* Matches any one char belonging to specified set. First
631 following byte is number of bitmap bytes. Then come bytes
632 for a bitmap saying which chars are in. Bits in each byte
633 are ordered low-bit-first. A character is in the set if its
634 bit is 1. A character too large to have a bit in the map is
635 automatically not in the set.
636
637 If the length byte has the 0x80 bit set, then that stuff
638 is followed by a range table:
639 2 bytes of flags for character sets (low 8 bits, high 8 bits)
640 See RANGE_TABLE_WORK_BITS below.
641 2 bytes, the number of pairs that follow (upto 32767)
642 pairs, each 2 multibyte characters,
643 each multibyte character represented as 3 bytes. */
644 charset,
645
646 /* Same parameters as charset, but match any character that is
647 not one of those specified. */
648 charset_not,
649
650 /* Start remembering the text that is matched, for storing in a
651 register. Followed by one byte with the register number, in
652 the range 0 to one less than the pattern buffer's re_nsub
653 field. */
654 start_memory,
655
656 /* Stop remembering the text that is matched and store it in a
657 memory register. Followed by one byte with the register
658 number, in the range 0 to one less than `re_nsub' in the
659 pattern buffer. */
660 stop_memory,
661
662 /* Match a duplicate of something remembered. Followed by one
663 byte containing the register number. */
664 duplicate,
665
666 /* Fail unless at beginning of line. */
667 begline,
668
669 /* Fail unless at end of line. */
670 endline,
671
672 /* Succeeds if at beginning of buffer (if emacs) or at beginning
673 of string to be matched (if not). */
674 begbuf,
675
676 /* Analogously, for end of buffer/string. */
677 endbuf,
678
679 /* Followed by two byte relative address to which to jump. */
680 jump,
681
682 /* Followed by two-byte relative address of place to resume at
683 in case of failure. */
684 on_failure_jump,
685
686 /* Like on_failure_jump, but pushes a placeholder instead of the
687 current string position when executed. */
688 on_failure_keep_string_jump,
689
690 /* Just like `on_failure_jump', except that it checks that we
691 don't get stuck in an infinite loop (matching an empty string
692 indefinitely). */
693 on_failure_jump_loop,
694
695 /* Just like `on_failure_jump_loop', except that it checks for
696 a different kind of loop (the kind that shows up with non-greedy
697 operators). This operation has to be immediately preceded
698 by a `no_op'. */
699 on_failure_jump_nastyloop,
700
701 /* A smart `on_failure_jump' used for greedy * and + operators.
702 It analyses the loop before which it is put and if the
703 loop does not require backtracking, it changes itself to
704 `on_failure_keep_string_jump' and short-circuits the loop,
705 else it just defaults to changing itself into `on_failure_jump'.
706 It assumes that it is pointing to just past a `jump'. */
707 on_failure_jump_smart,
708
709 /* Followed by two-byte relative address and two-byte number n.
710 After matching N times, jump to the address upon failure.
711 Does not work if N starts at 0: use on_failure_jump_loop
712 instead. */
713 succeed_n,
714
715 /* Followed by two-byte relative address, and two-byte number n.
716 Jump to the address N times, then fail. */
717 jump_n,
718
719 /* Set the following two-byte relative address to the
720 subsequent two-byte number. The address *includes* the two
721 bytes of number. */
722 set_number_at,
723
724 wordbeg, /* Succeeds if at word beginning. */
725 wordend, /* Succeeds if at word end. */
726
727 wordbound, /* Succeeds if at a word boundary. */
728 notwordbound, /* Succeeds if not at a word boundary. */
729
730 symbeg, /* Succeeds if at symbol beginning. */
731 symend, /* Succeeds if at symbol end. */
732
733 /* Matches any character whose syntax is specified. Followed by
734 a byte which contains a syntax code, e.g., Sword. */
735 syntaxspec,
736
737 /* Matches any character whose syntax is not that specified. */
738 notsyntaxspec
739
740 #ifdef emacs
741 ,before_dot, /* Succeeds if before point. */
742 at_dot, /* Succeeds if at point. */
743 after_dot, /* Succeeds if after point. */
744
745 /* Matches any character whose category-set contains the specified
746 category. The operator is followed by a byte which contains a
747 category code (mnemonic ASCII character). */
748 categoryspec,
749
750 /* Matches any character whose category-set does not contain the
751 specified category. The operator is followed by a byte which
752 contains the category code (mnemonic ASCII character). */
753 notcategoryspec
754 #endif /* emacs */
755 } re_opcode_t;
756 \f
757 /* Common operations on the compiled pattern. */
758
759 /* Store NUMBER in two contiguous bytes starting at DESTINATION. */
760
761 #define STORE_NUMBER(destination, number) \
762 do { \
763 (destination)[0] = (number) & 0377; \
764 (destination)[1] = (number) >> 8; \
765 } while (0)
766
767 /* Same as STORE_NUMBER, except increment DESTINATION to
768 the byte after where the number is stored. Therefore, DESTINATION
769 must be an lvalue. */
770
771 #define STORE_NUMBER_AND_INCR(destination, number) \
772 do { \
773 STORE_NUMBER (destination, number); \
774 (destination) += 2; \
775 } while (0)
776
777 /* Put into DESTINATION a number stored in two contiguous bytes starting
778 at SOURCE. */
779
780 #define EXTRACT_NUMBER(destination, source) \
781 do { \
782 (destination) = *(source) & 0377; \
783 (destination) += SIGN_EXTEND_CHAR (*((source) + 1)) << 8; \
784 } while (0)
785
786 #ifdef DEBUG
787 static void extract_number _RE_ARGS ((int *dest, re_char *source));
788 static void
789 extract_number (dest, source)
790 int *dest;
791 re_char *source;
792 {
793 int temp = SIGN_EXTEND_CHAR (*(source + 1));
794 *dest = *source & 0377;
795 *dest += temp << 8;
796 }
797
798 # ifndef EXTRACT_MACROS /* To debug the macros. */
799 # undef EXTRACT_NUMBER
800 # define EXTRACT_NUMBER(dest, src) extract_number (&dest, src)
801 # endif /* not EXTRACT_MACROS */
802
803 #endif /* DEBUG */
804
805 /* Same as EXTRACT_NUMBER, except increment SOURCE to after the number.
806 SOURCE must be an lvalue. */
807
808 #define EXTRACT_NUMBER_AND_INCR(destination, source) \
809 do { \
810 EXTRACT_NUMBER (destination, source); \
811 (source) += 2; \
812 } while (0)
813
814 #ifdef DEBUG
815 static void extract_number_and_incr _RE_ARGS ((int *destination,
816 re_char **source));
817 static void
818 extract_number_and_incr (destination, source)
819 int *destination;
820 re_char **source;
821 {
822 extract_number (destination, *source);
823 *source += 2;
824 }
825
826 # ifndef EXTRACT_MACROS
827 # undef EXTRACT_NUMBER_AND_INCR
828 # define EXTRACT_NUMBER_AND_INCR(dest, src) \
829 extract_number_and_incr (&dest, &src)
830 # endif /* not EXTRACT_MACROS */
831
832 #endif /* DEBUG */
833 \f
834 /* Store a multibyte character in three contiguous bytes starting
835 DESTINATION, and increment DESTINATION to the byte after where the
836 character is stored. Therefore, DESTINATION must be an lvalue. */
837
838 #define STORE_CHARACTER_AND_INCR(destination, character) \
839 do { \
840 (destination)[0] = (character) & 0377; \
841 (destination)[1] = ((character) >> 8) & 0377; \
842 (destination)[2] = (character) >> 16; \
843 (destination) += 3; \
844 } while (0)
845
846 /* Put into DESTINATION a character stored in three contiguous bytes
847 starting at SOURCE. */
848
849 #define EXTRACT_CHARACTER(destination, source) \
850 do { \
851 (destination) = ((source)[0] \
852 | ((source)[1] << 8) \
853 | ((source)[2] << 16)); \
854 } while (0)
855
856
857 /* Macros for charset. */
858
859 /* Size of bitmap of charset P in bytes. P is a start of charset,
860 i.e. *P is (re_opcode_t) charset or (re_opcode_t) charset_not. */
861 #define CHARSET_BITMAP_SIZE(p) ((p)[1] & 0x7F)
862
863 /* Nonzero if charset P has range table. */
864 #define CHARSET_RANGE_TABLE_EXISTS_P(p) ((p)[1] & 0x80)
865
866 /* Return the address of range table of charset P. But not the start
867 of table itself, but the before where the number of ranges is
868 stored. `2 +' means to skip re_opcode_t and size of bitmap,
869 and the 2 bytes of flags at the start of the range table. */
870 #define CHARSET_RANGE_TABLE(p) (&(p)[4 + CHARSET_BITMAP_SIZE (p)])
871
872 /* Extract the bit flags that start a range table. */
873 #define CHARSET_RANGE_TABLE_BITS(p) \
874 ((p)[2 + CHARSET_BITMAP_SIZE (p)] \
875 + (p)[3 + CHARSET_BITMAP_SIZE (p)] * 0x100)
876
877 /* Test if C is listed in the bitmap of charset P. */
878 #define CHARSET_LOOKUP_BITMAP(p, c) \
879 ((c) < CHARSET_BITMAP_SIZE (p) * BYTEWIDTH \
880 && (p)[2 + (c) / BYTEWIDTH] & (1 << ((c) % BYTEWIDTH)))
881
882 /* Return the address of end of RANGE_TABLE. COUNT is number of
883 ranges (which is a pair of (start, end)) in the RANGE_TABLE. `* 2'
884 is start of range and end of range. `* 3' is size of each start
885 and end. */
886 #define CHARSET_RANGE_TABLE_END(range_table, count) \
887 ((range_table) + (count) * 2 * 3)
888
889 /* Test if C is in RANGE_TABLE. A flag NOT is negated if C is in.
890 COUNT is number of ranges in RANGE_TABLE. */
891 #define CHARSET_LOOKUP_RANGE_TABLE_RAW(not, c, range_table, count) \
892 do \
893 { \
894 re_wchar_t range_start, range_end; \
895 re_char *p; \
896 re_char *range_table_end \
897 = CHARSET_RANGE_TABLE_END ((range_table), (count)); \
898 \
899 for (p = (range_table); p < range_table_end; p += 2 * 3) \
900 { \
901 EXTRACT_CHARACTER (range_start, p); \
902 EXTRACT_CHARACTER (range_end, p + 3); \
903 \
904 if (range_start <= (c) && (c) <= range_end) \
905 { \
906 (not) = !(not); \
907 break; \
908 } \
909 } \
910 } \
911 while (0)
912
913 /* Test if C is in range table of CHARSET. The flag NOT is negated if
914 C is listed in it. */
915 #define CHARSET_LOOKUP_RANGE_TABLE(not, c, charset) \
916 do \
917 { \
918 /* Number of ranges in range table. */ \
919 int count; \
920 re_char *range_table = CHARSET_RANGE_TABLE (charset); \
921 \
922 EXTRACT_NUMBER_AND_INCR (count, range_table); \
923 CHARSET_LOOKUP_RANGE_TABLE_RAW ((not), (c), range_table, count); \
924 } \
925 while (0)
926 \f
927 /* If DEBUG is defined, Regex prints many voluminous messages about what
928 it is doing (if the variable `debug' is nonzero). If linked with the
929 main program in `iregex.c', you can enter patterns and strings
930 interactively. And if linked with the main program in `main.c' and
931 the other test files, you can run the already-written tests. */
932
933 #ifdef DEBUG
934
935 /* We use standard I/O for debugging. */
936 # include <stdio.h>
937
938 /* It is useful to test things that ``must'' be true when debugging. */
939 # include <assert.h>
940
941 static int debug = -100000;
942
943 # define DEBUG_STATEMENT(e) e
944 # define DEBUG_PRINT1(x) if (debug > 0) printf (x)
945 # define DEBUG_PRINT2(x1, x2) if (debug > 0) printf (x1, x2)
946 # define DEBUG_PRINT3(x1, x2, x3) if (debug > 0) printf (x1, x2, x3)
947 # define DEBUG_PRINT4(x1, x2, x3, x4) if (debug > 0) printf (x1, x2, x3, x4)
948 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) \
949 if (debug > 0) print_partial_compiled_pattern (s, e)
950 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) \
951 if (debug > 0) print_double_string (w, s1, sz1, s2, sz2)
952
953
954 /* Print the fastmap in human-readable form. */
955
956 void
957 print_fastmap (fastmap)
958 char *fastmap;
959 {
960 unsigned was_a_range = 0;
961 unsigned i = 0;
962
963 while (i < (1 << BYTEWIDTH))
964 {
965 if (fastmap[i++])
966 {
967 was_a_range = 0;
968 putchar (i - 1);
969 while (i < (1 << BYTEWIDTH) && fastmap[i])
970 {
971 was_a_range = 1;
972 i++;
973 }
974 if (was_a_range)
975 {
976 printf ("-");
977 putchar (i - 1);
978 }
979 }
980 }
981 putchar ('\n');
982 }
983
984
985 /* Print a compiled pattern string in human-readable form, starting at
986 the START pointer into it and ending just before the pointer END. */
987
988 void
989 print_partial_compiled_pattern (start, end)
990 re_char *start;
991 re_char *end;
992 {
993 int mcnt, mcnt2;
994 re_char *p = start;
995 re_char *pend = end;
996
997 if (start == NULL)
998 {
999 fprintf (stderr, "(null)\n");
1000 return;
1001 }
1002
1003 /* Loop over pattern commands. */
1004 while (p < pend)
1005 {
1006 fprintf (stderr, "%d:\t", p - start);
1007
1008 switch ((re_opcode_t) *p++)
1009 {
1010 case no_op:
1011 fprintf (stderr, "/no_op");
1012 break;
1013
1014 case succeed:
1015 fprintf (stderr, "/succeed");
1016 break;
1017
1018 case exactn:
1019 mcnt = *p++;
1020 fprintf (stderr, "/exactn/%d", mcnt);
1021 do
1022 {
1023 fprintf (stderr, "/%c", *p++);
1024 }
1025 while (--mcnt);
1026 break;
1027
1028 case start_memory:
1029 fprintf (stderr, "/start_memory/%d", *p++);
1030 break;
1031
1032 case stop_memory:
1033 fprintf (stderr, "/stop_memory/%d", *p++);
1034 break;
1035
1036 case duplicate:
1037 fprintf (stderr, "/duplicate/%d", *p++);
1038 break;
1039
1040 case anychar:
1041 fprintf (stderr, "/anychar");
1042 break;
1043
1044 case charset:
1045 case charset_not:
1046 {
1047 register int c, last = -100;
1048 register int in_range = 0;
1049 int length = CHARSET_BITMAP_SIZE (p - 1);
1050 int has_range_table = CHARSET_RANGE_TABLE_EXISTS_P (p - 1);
1051
1052 fprintf (stderr, "/charset [%s",
1053 (re_opcode_t) *(p - 1) == charset_not ? "^" : "");
1054
1055 if (p + *p >= pend)
1056 fprintf (stderr, " !extends past end of pattern! ");
1057
1058 for (c = 0; c < 256; c++)
1059 if (c / 8 < length
1060 && (p[1 + (c/8)] & (1 << (c % 8))))
1061 {
1062 /* Are we starting a range? */
1063 if (last + 1 == c && ! in_range)
1064 {
1065 fprintf (stderr, "-");
1066 in_range = 1;
1067 }
1068 /* Have we broken a range? */
1069 else if (last + 1 != c && in_range)
1070 {
1071 fprintf (stderr, "%c", last);
1072 in_range = 0;
1073 }
1074
1075 if (! in_range)
1076 fprintf (stderr, "%c", c);
1077
1078 last = c;
1079 }
1080
1081 if (in_range)
1082 fprintf (stderr, "%c", last);
1083
1084 fprintf (stderr, "]");
1085
1086 p += 1 + length;
1087
1088 if (has_range_table)
1089 {
1090 int count;
1091 fprintf (stderr, "has-range-table");
1092
1093 /* ??? Should print the range table; for now, just skip it. */
1094 p += 2; /* skip range table bits */
1095 EXTRACT_NUMBER_AND_INCR (count, p);
1096 p = CHARSET_RANGE_TABLE_END (p, count);
1097 }
1098 }
1099 break;
1100
1101 case begline:
1102 fprintf (stderr, "/begline");
1103 break;
1104
1105 case endline:
1106 fprintf (stderr, "/endline");
1107 break;
1108
1109 case on_failure_jump:
1110 extract_number_and_incr (&mcnt, &p);
1111 fprintf (stderr, "/on_failure_jump to %d", p + mcnt - start);
1112 break;
1113
1114 case on_failure_keep_string_jump:
1115 extract_number_and_incr (&mcnt, &p);
1116 fprintf (stderr, "/on_failure_keep_string_jump to %d", p + mcnt - start);
1117 break;
1118
1119 case on_failure_jump_nastyloop:
1120 extract_number_and_incr (&mcnt, &p);
1121 fprintf (stderr, "/on_failure_jump_nastyloop to %d", p + mcnt - start);
1122 break;
1123
1124 case on_failure_jump_loop:
1125 extract_number_and_incr (&mcnt, &p);
1126 fprintf (stderr, "/on_failure_jump_loop to %d", p + mcnt - start);
1127 break;
1128
1129 case on_failure_jump_smart:
1130 extract_number_and_incr (&mcnt, &p);
1131 fprintf (stderr, "/on_failure_jump_smart to %d", p + mcnt - start);
1132 break;
1133
1134 case jump:
1135 extract_number_and_incr (&mcnt, &p);
1136 fprintf (stderr, "/jump to %d", p + mcnt - start);
1137 break;
1138
1139 case succeed_n:
1140 extract_number_and_incr (&mcnt, &p);
1141 extract_number_and_incr (&mcnt2, &p);
1142 fprintf (stderr, "/succeed_n to %d, %d times", p - 2 + mcnt - start, mcnt2);
1143 break;
1144
1145 case jump_n:
1146 extract_number_and_incr (&mcnt, &p);
1147 extract_number_and_incr (&mcnt2, &p);
1148 fprintf (stderr, "/jump_n to %d, %d times", p - 2 + mcnt - start, mcnt2);
1149 break;
1150
1151 case set_number_at:
1152 extract_number_and_incr (&mcnt, &p);
1153 extract_number_and_incr (&mcnt2, &p);
1154 fprintf (stderr, "/set_number_at location %d to %d", p - 2 + mcnt - start, mcnt2);
1155 break;
1156
1157 case wordbound:
1158 fprintf (stderr, "/wordbound");
1159 break;
1160
1161 case notwordbound:
1162 fprintf (stderr, "/notwordbound");
1163 break;
1164
1165 case wordbeg:
1166 fprintf (stderr, "/wordbeg");
1167 break;
1168
1169 case wordend:
1170 fprintf (stderr, "/wordend");
1171 break;
1172
1173 case symbeg:
1174 fprintf (stderr, "/symbeg");
1175 break;
1176
1177 case symend:
1178 fprintf (stderr, "/symend");
1179 break;
1180
1181 case syntaxspec:
1182 fprintf (stderr, "/syntaxspec");
1183 mcnt = *p++;
1184 fprintf (stderr, "/%d", mcnt);
1185 break;
1186
1187 case notsyntaxspec:
1188 fprintf (stderr, "/notsyntaxspec");
1189 mcnt = *p++;
1190 fprintf (stderr, "/%d", mcnt);
1191 break;
1192
1193 # ifdef emacs
1194 case before_dot:
1195 fprintf (stderr, "/before_dot");
1196 break;
1197
1198 case at_dot:
1199 fprintf (stderr, "/at_dot");
1200 break;
1201
1202 case after_dot:
1203 fprintf (stderr, "/after_dot");
1204 break;
1205
1206 case categoryspec:
1207 fprintf (stderr, "/categoryspec");
1208 mcnt = *p++;
1209 fprintf (stderr, "/%d", mcnt);
1210 break;
1211
1212 case notcategoryspec:
1213 fprintf (stderr, "/notcategoryspec");
1214 mcnt = *p++;
1215 fprintf (stderr, "/%d", mcnt);
1216 break;
1217 # endif /* emacs */
1218
1219 case begbuf:
1220 fprintf (stderr, "/begbuf");
1221 break;
1222
1223 case endbuf:
1224 fprintf (stderr, "/endbuf");
1225 break;
1226
1227 default:
1228 fprintf (stderr, "?%d", *(p-1));
1229 }
1230
1231 fprintf (stderr, "\n");
1232 }
1233
1234 fprintf (stderr, "%d:\tend of pattern.\n", p - start);
1235 }
1236
1237
1238 void
1239 print_compiled_pattern (bufp)
1240 struct re_pattern_buffer *bufp;
1241 {
1242 re_char *buffer = bufp->buffer;
1243
1244 print_partial_compiled_pattern (buffer, buffer + bufp->used);
1245 printf ("%ld bytes used/%ld bytes allocated.\n",
1246 bufp->used, bufp->allocated);
1247
1248 if (bufp->fastmap_accurate && bufp->fastmap)
1249 {
1250 printf ("fastmap: ");
1251 print_fastmap (bufp->fastmap);
1252 }
1253
1254 printf ("re_nsub: %d\t", bufp->re_nsub);
1255 printf ("regs_alloc: %d\t", bufp->regs_allocated);
1256 printf ("can_be_null: %d\t", bufp->can_be_null);
1257 printf ("no_sub: %d\t", bufp->no_sub);
1258 printf ("not_bol: %d\t", bufp->not_bol);
1259 printf ("not_eol: %d\t", bufp->not_eol);
1260 printf ("syntax: %lx\n", bufp->syntax);
1261 fflush (stdout);
1262 /* Perhaps we should print the translate table? */
1263 }
1264
1265
1266 void
1267 print_double_string (where, string1, size1, string2, size2)
1268 re_char *where;
1269 re_char *string1;
1270 re_char *string2;
1271 int size1;
1272 int size2;
1273 {
1274 int this_char;
1275
1276 if (where == NULL)
1277 printf ("(null)");
1278 else
1279 {
1280 if (FIRST_STRING_P (where))
1281 {
1282 for (this_char = where - string1; this_char < size1; this_char++)
1283 putchar (string1[this_char]);
1284
1285 where = string2;
1286 }
1287
1288 for (this_char = where - string2; this_char < size2; this_char++)
1289 putchar (string2[this_char]);
1290 }
1291 }
1292
1293 #else /* not DEBUG */
1294
1295 # undef assert
1296 # define assert(e)
1297
1298 # define DEBUG_STATEMENT(e)
1299 # define DEBUG_PRINT1(x)
1300 # define DEBUG_PRINT2(x1, x2)
1301 # define DEBUG_PRINT3(x1, x2, x3)
1302 # define DEBUG_PRINT4(x1, x2, x3, x4)
1303 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e)
1304 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2)
1305
1306 #endif /* not DEBUG */
1307 \f
1308 /* Set by `re_set_syntax' to the current regexp syntax to recognize. Can
1309 also be assigned to arbitrarily: each pattern buffer stores its own
1310 syntax, so it can be changed between regex compilations. */
1311 /* This has no initializer because initialized variables in Emacs
1312 become read-only after dumping. */
1313 reg_syntax_t re_syntax_options;
1314
1315
1316 /* Specify the precise syntax of regexps for compilation. This provides
1317 for compatibility for various utilities which historically have
1318 different, incompatible syntaxes.
1319
1320 The argument SYNTAX is a bit mask comprised of the various bits
1321 defined in regex.h. We return the old syntax. */
1322
1323 reg_syntax_t
1324 re_set_syntax (syntax)
1325 reg_syntax_t syntax;
1326 {
1327 reg_syntax_t ret = re_syntax_options;
1328
1329 re_syntax_options = syntax;
1330 return ret;
1331 }
1332 WEAK_ALIAS (__re_set_syntax, re_set_syntax)
1333
1334 /* Regexp to use to replace spaces, or NULL meaning don't. */
1335 static re_char *whitespace_regexp;
1336
1337 void
1338 re_set_whitespace_regexp (regexp)
1339 const char *regexp;
1340 {
1341 whitespace_regexp = (re_char *) regexp;
1342 }
1343 WEAK_ALIAS (__re_set_syntax, re_set_syntax)
1344 \f
1345 /* This table gives an error message for each of the error codes listed
1346 in regex.h. Obviously the order here has to be same as there.
1347 POSIX doesn't require that we do anything for REG_NOERROR,
1348 but why not be nice? */
1349
1350 static const char *re_error_msgid[] =
1351 {
1352 gettext_noop ("Success"), /* REG_NOERROR */
1353 gettext_noop ("No match"), /* REG_NOMATCH */
1354 gettext_noop ("Invalid regular expression"), /* REG_BADPAT */
1355 gettext_noop ("Invalid collation character"), /* REG_ECOLLATE */
1356 gettext_noop ("Invalid character class name"), /* REG_ECTYPE */
1357 gettext_noop ("Trailing backslash"), /* REG_EESCAPE */
1358 gettext_noop ("Invalid back reference"), /* REG_ESUBREG */
1359 gettext_noop ("Unmatched [ or [^"), /* REG_EBRACK */
1360 gettext_noop ("Unmatched ( or \\("), /* REG_EPAREN */
1361 gettext_noop ("Unmatched \\{"), /* REG_EBRACE */
1362 gettext_noop ("Invalid content of \\{\\}"), /* REG_BADBR */
1363 gettext_noop ("Invalid range end"), /* REG_ERANGE */
1364 gettext_noop ("Memory exhausted"), /* REG_ESPACE */
1365 gettext_noop ("Invalid preceding regular expression"), /* REG_BADRPT */
1366 gettext_noop ("Premature end of regular expression"), /* REG_EEND */
1367 gettext_noop ("Regular expression too big"), /* REG_ESIZE */
1368 gettext_noop ("Unmatched ) or \\)"), /* REG_ERPAREN */
1369 gettext_noop ("Range striding over charsets") /* REG_ERANGEX */
1370 };
1371 \f
1372 /* Avoiding alloca during matching, to placate r_alloc. */
1373
1374 /* Define MATCH_MAY_ALLOCATE unless we need to make sure that the
1375 searching and matching functions should not call alloca. On some
1376 systems, alloca is implemented in terms of malloc, and if we're
1377 using the relocating allocator routines, then malloc could cause a
1378 relocation, which might (if the strings being searched are in the
1379 ralloc heap) shift the data out from underneath the regexp
1380 routines.
1381
1382 Here's another reason to avoid allocation: Emacs
1383 processes input from X in a signal handler; processing X input may
1384 call malloc; if input arrives while a matching routine is calling
1385 malloc, then we're scrod. But Emacs can't just block input while
1386 calling matching routines; then we don't notice interrupts when
1387 they come in. So, Emacs blocks input around all regexp calls
1388 except the matching calls, which it leaves unprotected, in the
1389 faith that they will not malloc. */
1390
1391 /* Normally, this is fine. */
1392 #define MATCH_MAY_ALLOCATE
1393
1394 /* The match routines may not allocate if (1) they would do it with malloc
1395 and (2) it's not safe for them to use malloc.
1396 Note that if REL_ALLOC is defined, matching would not use malloc for the
1397 failure stack, but we would still use it for the register vectors;
1398 so REL_ALLOC should not affect this. */
1399 #if defined REGEX_MALLOC && defined emacs
1400 # undef MATCH_MAY_ALLOCATE
1401 #endif
1402
1403 \f
1404 /* Failure stack declarations and macros; both re_compile_fastmap and
1405 re_match_2 use a failure stack. These have to be macros because of
1406 REGEX_ALLOCATE_STACK. */
1407
1408
1409 /* Approximate number of failure points for which to initially allocate space
1410 when matching. If this number is exceeded, we allocate more
1411 space, so it is not a hard limit. */
1412 #ifndef INIT_FAILURE_ALLOC
1413 # define INIT_FAILURE_ALLOC 20
1414 #endif
1415
1416 /* Roughly the maximum number of failure points on the stack. Would be
1417 exactly that if always used TYPICAL_FAILURE_SIZE items each time we failed.
1418 This is a variable only so users of regex can assign to it; we never
1419 change it ourselves. We always multiply it by TYPICAL_FAILURE_SIZE
1420 before using it, so it should probably be a byte-count instead. */
1421 # if defined MATCH_MAY_ALLOCATE
1422 /* Note that 4400 was enough to cause a crash on Alpha OSF/1,
1423 whose default stack limit is 2mb. In order for a larger
1424 value to work reliably, you have to try to make it accord
1425 with the process stack limit. */
1426 size_t re_max_failures = 40000;
1427 # else
1428 size_t re_max_failures = 4000;
1429 # endif
1430
1431 union fail_stack_elt
1432 {
1433 re_char *pointer;
1434 /* This should be the biggest `int' that's no bigger than a pointer. */
1435 long integer;
1436 };
1437
1438 typedef union fail_stack_elt fail_stack_elt_t;
1439
1440 typedef struct
1441 {
1442 fail_stack_elt_t *stack;
1443 size_t size;
1444 size_t avail; /* Offset of next open position. */
1445 size_t frame; /* Offset of the cur constructed frame. */
1446 } fail_stack_type;
1447
1448 #define FAIL_STACK_EMPTY() (fail_stack.frame == 0)
1449 #define FAIL_STACK_FULL() (fail_stack.avail == fail_stack.size)
1450
1451
1452 /* Define macros to initialize and free the failure stack.
1453 Do `return -2' if the alloc fails. */
1454
1455 #ifdef MATCH_MAY_ALLOCATE
1456 # define INIT_FAIL_STACK() \
1457 do { \
1458 fail_stack.stack = (fail_stack_elt_t *) \
1459 REGEX_ALLOCATE_STACK (INIT_FAILURE_ALLOC * TYPICAL_FAILURE_SIZE \
1460 * sizeof (fail_stack_elt_t)); \
1461 \
1462 if (fail_stack.stack == NULL) \
1463 return -2; \
1464 \
1465 fail_stack.size = INIT_FAILURE_ALLOC; \
1466 fail_stack.avail = 0; \
1467 fail_stack.frame = 0; \
1468 } while (0)
1469
1470 # define RESET_FAIL_STACK() REGEX_FREE_STACK (fail_stack.stack)
1471 #else
1472 # define INIT_FAIL_STACK() \
1473 do { \
1474 fail_stack.avail = 0; \
1475 fail_stack.frame = 0; \
1476 } while (0)
1477
1478 # define RESET_FAIL_STACK() ((void)0)
1479 #endif
1480
1481
1482 /* Double the size of FAIL_STACK, up to a limit
1483 which allows approximately `re_max_failures' items.
1484
1485 Return 1 if succeeds, and 0 if either ran out of memory
1486 allocating space for it or it was already too large.
1487
1488 REGEX_REALLOCATE_STACK requires `destination' be declared. */
1489
1490 /* Factor to increase the failure stack size by
1491 when we increase it.
1492 This used to be 2, but 2 was too wasteful
1493 because the old discarded stacks added up to as much space
1494 were as ultimate, maximum-size stack. */
1495 #define FAIL_STACK_GROWTH_FACTOR 4
1496
1497 #define GROW_FAIL_STACK(fail_stack) \
1498 (((fail_stack).size * sizeof (fail_stack_elt_t) \
1499 >= re_max_failures * TYPICAL_FAILURE_SIZE) \
1500 ? 0 \
1501 : ((fail_stack).stack \
1502 = (fail_stack_elt_t *) \
1503 REGEX_REALLOCATE_STACK ((fail_stack).stack, \
1504 (fail_stack).size * sizeof (fail_stack_elt_t), \
1505 MIN (re_max_failures * TYPICAL_FAILURE_SIZE, \
1506 ((fail_stack).size * sizeof (fail_stack_elt_t) \
1507 * FAIL_STACK_GROWTH_FACTOR))), \
1508 \
1509 (fail_stack).stack == NULL \
1510 ? 0 \
1511 : ((fail_stack).size \
1512 = (MIN (re_max_failures * TYPICAL_FAILURE_SIZE, \
1513 ((fail_stack).size * sizeof (fail_stack_elt_t) \
1514 * FAIL_STACK_GROWTH_FACTOR)) \
1515 / sizeof (fail_stack_elt_t)), \
1516 1)))
1517
1518
1519 /* Push a pointer value onto the failure stack.
1520 Assumes the variable `fail_stack'. Probably should only
1521 be called from within `PUSH_FAILURE_POINT'. */
1522 #define PUSH_FAILURE_POINTER(item) \
1523 fail_stack.stack[fail_stack.avail++].pointer = (item)
1524
1525 /* This pushes an integer-valued item onto the failure stack.
1526 Assumes the variable `fail_stack'. Probably should only
1527 be called from within `PUSH_FAILURE_POINT'. */
1528 #define PUSH_FAILURE_INT(item) \
1529 fail_stack.stack[fail_stack.avail++].integer = (item)
1530
1531 /* Push a fail_stack_elt_t value onto the failure stack.
1532 Assumes the variable `fail_stack'. Probably should only
1533 be called from within `PUSH_FAILURE_POINT'. */
1534 #define PUSH_FAILURE_ELT(item) \
1535 fail_stack.stack[fail_stack.avail++] = (item)
1536
1537 /* These three POP... operations complement the three PUSH... operations.
1538 All assume that `fail_stack' is nonempty. */
1539 #define POP_FAILURE_POINTER() fail_stack.stack[--fail_stack.avail].pointer
1540 #define POP_FAILURE_INT() fail_stack.stack[--fail_stack.avail].integer
1541 #define POP_FAILURE_ELT() fail_stack.stack[--fail_stack.avail]
1542
1543 /* Individual items aside from the registers. */
1544 #define NUM_NONREG_ITEMS 3
1545
1546 /* Used to examine the stack (to detect infinite loops). */
1547 #define FAILURE_PAT(h) fail_stack.stack[(h) - 1].pointer
1548 #define FAILURE_STR(h) (fail_stack.stack[(h) - 2].pointer)
1549 #define NEXT_FAILURE_HANDLE(h) fail_stack.stack[(h) - 3].integer
1550 #define TOP_FAILURE_HANDLE() fail_stack.frame
1551
1552
1553 #define ENSURE_FAIL_STACK(space) \
1554 while (REMAINING_AVAIL_SLOTS <= space) { \
1555 if (!GROW_FAIL_STACK (fail_stack)) \
1556 return -2; \
1557 DEBUG_PRINT2 ("\n Doubled stack; size now: %d\n", (fail_stack).size);\
1558 DEBUG_PRINT2 (" slots available: %d\n", REMAINING_AVAIL_SLOTS);\
1559 }
1560
1561 /* Push register NUM onto the stack. */
1562 #define PUSH_FAILURE_REG(num) \
1563 do { \
1564 char *destination; \
1565 ENSURE_FAIL_STACK(3); \
1566 DEBUG_PRINT4 (" Push reg %d (spanning %p -> %p)\n", \
1567 num, regstart[num], regend[num]); \
1568 PUSH_FAILURE_POINTER (regstart[num]); \
1569 PUSH_FAILURE_POINTER (regend[num]); \
1570 PUSH_FAILURE_INT (num); \
1571 } while (0)
1572
1573 /* Change the counter's value to VAL, but make sure that it will
1574 be reset when backtracking. */
1575 #define PUSH_NUMBER(ptr,val) \
1576 do { \
1577 char *destination; \
1578 int c; \
1579 ENSURE_FAIL_STACK(3); \
1580 EXTRACT_NUMBER (c, ptr); \
1581 DEBUG_PRINT4 (" Push number %p = %d -> %d\n", ptr, c, val); \
1582 PUSH_FAILURE_INT (c); \
1583 PUSH_FAILURE_POINTER (ptr); \
1584 PUSH_FAILURE_INT (-1); \
1585 STORE_NUMBER (ptr, val); \
1586 } while (0)
1587
1588 /* Pop a saved register off the stack. */
1589 #define POP_FAILURE_REG_OR_COUNT() \
1590 do { \
1591 int reg = POP_FAILURE_INT (); \
1592 if (reg == -1) \
1593 { \
1594 /* It's a counter. */ \
1595 /* Here, we discard `const', making re_match non-reentrant. */ \
1596 unsigned char *ptr = (unsigned char*) POP_FAILURE_POINTER (); \
1597 reg = POP_FAILURE_INT (); \
1598 STORE_NUMBER (ptr, reg); \
1599 DEBUG_PRINT3 (" Pop counter %p = %d\n", ptr, reg); \
1600 } \
1601 else \
1602 { \
1603 regend[reg] = POP_FAILURE_POINTER (); \
1604 regstart[reg] = POP_FAILURE_POINTER (); \
1605 DEBUG_PRINT4 (" Pop reg %d (spanning %p -> %p)\n", \
1606 reg, regstart[reg], regend[reg]); \
1607 } \
1608 } while (0)
1609
1610 /* Check that we are not stuck in an infinite loop. */
1611 #define CHECK_INFINITE_LOOP(pat_cur, string_place) \
1612 do { \
1613 int failure = TOP_FAILURE_HANDLE (); \
1614 /* Check for infinite matching loops */ \
1615 while (failure > 0 \
1616 && (FAILURE_STR (failure) == string_place \
1617 || FAILURE_STR (failure) == NULL)) \
1618 { \
1619 assert (FAILURE_PAT (failure) >= bufp->buffer \
1620 && FAILURE_PAT (failure) <= bufp->buffer + bufp->used); \
1621 if (FAILURE_PAT (failure) == pat_cur) \
1622 { \
1623 cycle = 1; \
1624 break; \
1625 } \
1626 DEBUG_PRINT2 (" Other pattern: %p\n", FAILURE_PAT (failure)); \
1627 failure = NEXT_FAILURE_HANDLE(failure); \
1628 } \
1629 DEBUG_PRINT2 (" Other string: %p\n", FAILURE_STR (failure)); \
1630 } while (0)
1631
1632 /* Push the information about the state we will need
1633 if we ever fail back to it.
1634
1635 Requires variables fail_stack, regstart, regend and
1636 num_regs be declared. GROW_FAIL_STACK requires `destination' be
1637 declared.
1638
1639 Does `return FAILURE_CODE' if runs out of memory. */
1640
1641 #define PUSH_FAILURE_POINT(pattern, string_place) \
1642 do { \
1643 char *destination; \
1644 /* Must be int, so when we don't save any registers, the arithmetic \
1645 of 0 + -1 isn't done as unsigned. */ \
1646 \
1647 DEBUG_STATEMENT (nfailure_points_pushed++); \
1648 DEBUG_PRINT1 ("\nPUSH_FAILURE_POINT:\n"); \
1649 DEBUG_PRINT2 (" Before push, next avail: %d\n", (fail_stack).avail); \
1650 DEBUG_PRINT2 (" size: %d\n", (fail_stack).size);\
1651 \
1652 ENSURE_FAIL_STACK (NUM_NONREG_ITEMS); \
1653 \
1654 DEBUG_PRINT1 ("\n"); \
1655 \
1656 DEBUG_PRINT2 (" Push frame index: %d\n", fail_stack.frame); \
1657 PUSH_FAILURE_INT (fail_stack.frame); \
1658 \
1659 DEBUG_PRINT2 (" Push string %p: `", string_place); \
1660 DEBUG_PRINT_DOUBLE_STRING (string_place, string1, size1, string2, size2);\
1661 DEBUG_PRINT1 ("'\n"); \
1662 PUSH_FAILURE_POINTER (string_place); \
1663 \
1664 DEBUG_PRINT2 (" Push pattern %p: ", pattern); \
1665 DEBUG_PRINT_COMPILED_PATTERN (bufp, pattern, pend); \
1666 PUSH_FAILURE_POINTER (pattern); \
1667 \
1668 /* Close the frame by moving the frame pointer past it. */ \
1669 fail_stack.frame = fail_stack.avail; \
1670 } while (0)
1671
1672 /* Estimate the size of data pushed by a typical failure stack entry.
1673 An estimate is all we need, because all we use this for
1674 is to choose a limit for how big to make the failure stack. */
1675 /* BEWARE, the value `20' is hard-coded in emacs.c:main(). */
1676 #define TYPICAL_FAILURE_SIZE 20
1677
1678 /* How many items can still be added to the stack without overflowing it. */
1679 #define REMAINING_AVAIL_SLOTS ((fail_stack).size - (fail_stack).avail)
1680
1681
1682 /* Pops what PUSH_FAIL_STACK pushes.
1683
1684 We restore into the parameters, all of which should be lvalues:
1685 STR -- the saved data position.
1686 PAT -- the saved pattern position.
1687 REGSTART, REGEND -- arrays of string positions.
1688
1689 Also assumes the variables `fail_stack' and (if debugging), `bufp',
1690 `pend', `string1', `size1', `string2', and `size2'. */
1691
1692 #define POP_FAILURE_POINT(str, pat) \
1693 do { \
1694 assert (!FAIL_STACK_EMPTY ()); \
1695 \
1696 /* Remove failure points and point to how many regs pushed. */ \
1697 DEBUG_PRINT1 ("POP_FAILURE_POINT:\n"); \
1698 DEBUG_PRINT2 (" Before pop, next avail: %d\n", fail_stack.avail); \
1699 DEBUG_PRINT2 (" size: %d\n", fail_stack.size); \
1700 \
1701 /* Pop the saved registers. */ \
1702 while (fail_stack.frame < fail_stack.avail) \
1703 POP_FAILURE_REG_OR_COUNT (); \
1704 \
1705 pat = POP_FAILURE_POINTER (); \
1706 DEBUG_PRINT2 (" Popping pattern %p: ", pat); \
1707 DEBUG_PRINT_COMPILED_PATTERN (bufp, pat, pend); \
1708 \
1709 /* If the saved string location is NULL, it came from an \
1710 on_failure_keep_string_jump opcode, and we want to throw away the \
1711 saved NULL, thus retaining our current position in the string. */ \
1712 str = POP_FAILURE_POINTER (); \
1713 DEBUG_PRINT2 (" Popping string %p: `", str); \
1714 DEBUG_PRINT_DOUBLE_STRING (str, string1, size1, string2, size2); \
1715 DEBUG_PRINT1 ("'\n"); \
1716 \
1717 fail_stack.frame = POP_FAILURE_INT (); \
1718 DEBUG_PRINT2 (" Popping frame index: %d\n", fail_stack.frame); \
1719 \
1720 assert (fail_stack.avail >= 0); \
1721 assert (fail_stack.frame <= fail_stack.avail); \
1722 \
1723 DEBUG_STATEMENT (nfailure_points_popped++); \
1724 } while (0) /* POP_FAILURE_POINT */
1725
1726
1727 \f
1728 /* Registers are set to a sentinel when they haven't yet matched. */
1729 #define REG_UNSET(e) ((e) == NULL)
1730 \f
1731 /* Subroutine declarations and macros for regex_compile. */
1732
1733 static reg_errcode_t regex_compile _RE_ARGS ((re_char *pattern, size_t size,
1734 reg_syntax_t syntax,
1735 struct re_pattern_buffer *bufp));
1736 static void store_op1 _RE_ARGS ((re_opcode_t op, unsigned char *loc, int arg));
1737 static void store_op2 _RE_ARGS ((re_opcode_t op, unsigned char *loc,
1738 int arg1, int arg2));
1739 static void insert_op1 _RE_ARGS ((re_opcode_t op, unsigned char *loc,
1740 int arg, unsigned char *end));
1741 static void insert_op2 _RE_ARGS ((re_opcode_t op, unsigned char *loc,
1742 int arg1, int arg2, unsigned char *end));
1743 static boolean at_begline_loc_p _RE_ARGS ((re_char *pattern,
1744 re_char *p,
1745 reg_syntax_t syntax));
1746 static boolean at_endline_loc_p _RE_ARGS ((re_char *p,
1747 re_char *pend,
1748 reg_syntax_t syntax));
1749 static re_char *skip_one_char _RE_ARGS ((re_char *p));
1750 static int analyse_first _RE_ARGS ((re_char *p, re_char *pend,
1751 char *fastmap, const int multibyte));
1752
1753 /* Fetch the next character in the uncompiled pattern, with no
1754 translation. */
1755 #define PATFETCH(c) \
1756 do { \
1757 int len; \
1758 if (p == pend) return REG_EEND; \
1759 c = RE_STRING_CHAR_AND_LENGTH (p, pend - p, len, multibyte); \
1760 p += len; \
1761 } while (0)
1762
1763
1764 /* If `translate' is non-null, return translate[D], else just D. We
1765 cast the subscript to translate because some data is declared as
1766 `char *', to avoid warnings when a string constant is passed. But
1767 when we use a character as a subscript we must make it unsigned. */
1768 #ifndef TRANSLATE
1769 # define TRANSLATE(d) \
1770 (RE_TRANSLATE_P (translate) ? RE_TRANSLATE (translate, (d)) : (d))
1771 #endif
1772
1773
1774 /* Macros for outputting the compiled pattern into `buffer'. */
1775
1776 /* If the buffer isn't allocated when it comes in, use this. */
1777 #define INIT_BUF_SIZE 32
1778
1779 /* Make sure we have at least N more bytes of space in buffer. */
1780 #define GET_BUFFER_SPACE(n) \
1781 while ((size_t) (b - bufp->buffer + (n)) > bufp->allocated) \
1782 EXTEND_BUFFER ()
1783
1784 /* Make sure we have one more byte of buffer space and then add C to it. */
1785 #define BUF_PUSH(c) \
1786 do { \
1787 GET_BUFFER_SPACE (1); \
1788 *b++ = (unsigned char) (c); \
1789 } while (0)
1790
1791
1792 /* Ensure we have two more bytes of buffer space and then append C1 and C2. */
1793 #define BUF_PUSH_2(c1, c2) \
1794 do { \
1795 GET_BUFFER_SPACE (2); \
1796 *b++ = (unsigned char) (c1); \
1797 *b++ = (unsigned char) (c2); \
1798 } while (0)
1799
1800
1801 /* As with BUF_PUSH_2, except for three bytes. */
1802 #define BUF_PUSH_3(c1, c2, c3) \
1803 do { \
1804 GET_BUFFER_SPACE (3); \
1805 *b++ = (unsigned char) (c1); \
1806 *b++ = (unsigned char) (c2); \
1807 *b++ = (unsigned char) (c3); \
1808 } while (0)
1809
1810
1811 /* Store a jump with opcode OP at LOC to location TO. We store a
1812 relative address offset by the three bytes the jump itself occupies. */
1813 #define STORE_JUMP(op, loc, to) \
1814 store_op1 (op, loc, (to) - (loc) - 3)
1815
1816 /* Likewise, for a two-argument jump. */
1817 #define STORE_JUMP2(op, loc, to, arg) \
1818 store_op2 (op, loc, (to) - (loc) - 3, arg)
1819
1820 /* Like `STORE_JUMP', but for inserting. Assume `b' is the buffer end. */
1821 #define INSERT_JUMP(op, loc, to) \
1822 insert_op1 (op, loc, (to) - (loc) - 3, b)
1823
1824 /* Like `STORE_JUMP2', but for inserting. Assume `b' is the buffer end. */
1825 #define INSERT_JUMP2(op, loc, to, arg) \
1826 insert_op2 (op, loc, (to) - (loc) - 3, arg, b)
1827
1828
1829 /* This is not an arbitrary limit: the arguments which represent offsets
1830 into the pattern are two bytes long. So if 2^15 bytes turns out to
1831 be too small, many things would have to change. */
1832 # define MAX_BUF_SIZE (1L << 15)
1833
1834 #if 0 /* This is when we thought it could be 2^16 bytes. */
1835 /* Any other compiler which, like MSC, has allocation limit below 2^16
1836 bytes will have to use approach similar to what was done below for
1837 MSC and drop MAX_BUF_SIZE a bit. Otherwise you may end up
1838 reallocating to 0 bytes. Such thing is not going to work too well.
1839 You have been warned!! */
1840 #if defined _MSC_VER && !defined WIN32
1841 /* Microsoft C 16-bit versions limit malloc to approx 65512 bytes. */
1842 # define MAX_BUF_SIZE 65500L
1843 #else
1844 # define MAX_BUF_SIZE (1L << 16)
1845 #endif
1846 #endif /* 0 */
1847
1848 /* Extend the buffer by twice its current size via realloc and
1849 reset the pointers that pointed into the old block to point to the
1850 correct places in the new one. If extending the buffer results in it
1851 being larger than MAX_BUF_SIZE, then flag memory exhausted. */
1852 #if __BOUNDED_POINTERS__
1853 # define SET_HIGH_BOUND(P) (__ptrhigh (P) = __ptrlow (P) + bufp->allocated)
1854 # define MOVE_BUFFER_POINTER(P) \
1855 (__ptrlow (P) = new_buffer + (__ptrlow (P) - old_buffer), \
1856 SET_HIGH_BOUND (P), \
1857 __ptrvalue (P) = new_buffer + (__ptrvalue (P) - old_buffer))
1858 # define ELSE_EXTEND_BUFFER_HIGH_BOUND \
1859 else \
1860 { \
1861 SET_HIGH_BOUND (b); \
1862 SET_HIGH_BOUND (begalt); \
1863 if (fixup_alt_jump) \
1864 SET_HIGH_BOUND (fixup_alt_jump); \
1865 if (laststart) \
1866 SET_HIGH_BOUND (laststart); \
1867 if (pending_exact) \
1868 SET_HIGH_BOUND (pending_exact); \
1869 }
1870 #else
1871 # define MOVE_BUFFER_POINTER(P) ((P) = new_buffer + ((P) - old_buffer))
1872 # define ELSE_EXTEND_BUFFER_HIGH_BOUND
1873 #endif
1874 #define EXTEND_BUFFER() \
1875 do { \
1876 unsigned char *old_buffer = bufp->buffer; \
1877 if (bufp->allocated == MAX_BUF_SIZE) \
1878 return REG_ESIZE; \
1879 bufp->allocated <<= 1; \
1880 if (bufp->allocated > MAX_BUF_SIZE) \
1881 bufp->allocated = MAX_BUF_SIZE; \
1882 RETALLOC (bufp->buffer, bufp->allocated, unsigned char); \
1883 if (bufp->buffer == NULL) \
1884 return REG_ESPACE; \
1885 /* If the buffer moved, move all the pointers into it. */ \
1886 if (old_buffer != bufp->buffer) \
1887 { \
1888 unsigned char *new_buffer = bufp->buffer; \
1889 MOVE_BUFFER_POINTER (b); \
1890 MOVE_BUFFER_POINTER (begalt); \
1891 if (fixup_alt_jump) \
1892 MOVE_BUFFER_POINTER (fixup_alt_jump); \
1893 if (laststart) \
1894 MOVE_BUFFER_POINTER (laststart); \
1895 if (pending_exact) \
1896 MOVE_BUFFER_POINTER (pending_exact); \
1897 } \
1898 ELSE_EXTEND_BUFFER_HIGH_BOUND \
1899 } while (0)
1900
1901
1902 /* Since we have one byte reserved for the register number argument to
1903 {start,stop}_memory, the maximum number of groups we can report
1904 things about is what fits in that byte. */
1905 #define MAX_REGNUM 255
1906
1907 /* But patterns can have more than `MAX_REGNUM' registers. We just
1908 ignore the excess. */
1909 typedef int regnum_t;
1910
1911
1912 /* Macros for the compile stack. */
1913
1914 /* Since offsets can go either forwards or backwards, this type needs to
1915 be able to hold values from -(MAX_BUF_SIZE - 1) to MAX_BUF_SIZE - 1. */
1916 /* int may be not enough when sizeof(int) == 2. */
1917 typedef long pattern_offset_t;
1918
1919 typedef struct
1920 {
1921 pattern_offset_t begalt_offset;
1922 pattern_offset_t fixup_alt_jump;
1923 pattern_offset_t laststart_offset;
1924 regnum_t regnum;
1925 } compile_stack_elt_t;
1926
1927
1928 typedef struct
1929 {
1930 compile_stack_elt_t *stack;
1931 unsigned size;
1932 unsigned avail; /* Offset of next open position. */
1933 } compile_stack_type;
1934
1935
1936 #define INIT_COMPILE_STACK_SIZE 32
1937
1938 #define COMPILE_STACK_EMPTY (compile_stack.avail == 0)
1939 #define COMPILE_STACK_FULL (compile_stack.avail == compile_stack.size)
1940
1941 /* The next available element. */
1942 #define COMPILE_STACK_TOP (compile_stack.stack[compile_stack.avail])
1943
1944 /* Explicit quit checking is only used on NTemacs and whenever we
1945 use polling to process input events. */
1946 #if defined emacs && (defined WINDOWSNT || defined SYNC_INPUT) && defined QUIT
1947 extern int immediate_quit;
1948 # define IMMEDIATE_QUIT_CHECK \
1949 do { \
1950 if (immediate_quit) QUIT; \
1951 } while (0)
1952 #else
1953 # define IMMEDIATE_QUIT_CHECK ((void)0)
1954 #endif
1955 \f
1956 /* Structure to manage work area for range table. */
1957 struct range_table_work_area
1958 {
1959 int *table; /* actual work area. */
1960 int allocated; /* allocated size for work area in bytes. */
1961 int used; /* actually used size in words. */
1962 int bits; /* flag to record character classes */
1963 };
1964
1965 /* Make sure that WORK_AREA can hold more N multibyte characters.
1966 This is used only in set_image_of_range and set_image_of_range_1.
1967 It expects WORK_AREA to be a pointer.
1968 If it can't get the space, it returns from the surrounding function. */
1969
1970 #define EXTEND_RANGE_TABLE(work_area, n) \
1971 do { \
1972 if (((work_area).used + (n)) * sizeof (int) > (work_area).allocated) \
1973 { \
1974 extend_range_table_work_area (&work_area); \
1975 if ((work_area).table == 0) \
1976 return (REG_ESPACE); \
1977 } \
1978 } while (0)
1979
1980 #define SET_RANGE_TABLE_WORK_AREA_BIT(work_area, bit) \
1981 (work_area).bits |= (bit)
1982
1983 /* Bits used to implement the multibyte-part of the various character classes
1984 such as [:alnum:] in a charset's range table. */
1985 #define BIT_WORD 0x1
1986 #define BIT_LOWER 0x2
1987 #define BIT_PUNCT 0x4
1988 #define BIT_SPACE 0x8
1989 #define BIT_UPPER 0x10
1990 #define BIT_MULTIBYTE 0x20
1991
1992 /* Set a range (RANGE_START, RANGE_END) to WORK_AREA. */
1993 #define SET_RANGE_TABLE_WORK_AREA(work_area, range_start, range_end) \
1994 do { \
1995 EXTEND_RANGE_TABLE ((work_area), 2); \
1996 (work_area).table[(work_area).used++] = (range_start); \
1997 (work_area).table[(work_area).used++] = (range_end); \
1998 } while (0)
1999
2000 /* Free allocated memory for WORK_AREA. */
2001 #define FREE_RANGE_TABLE_WORK_AREA(work_area) \
2002 do { \
2003 if ((work_area).table) \
2004 free ((work_area).table); \
2005 } while (0)
2006
2007 #define CLEAR_RANGE_TABLE_WORK_USED(work_area) ((work_area).used = 0, (work_area).bits = 0)
2008 #define RANGE_TABLE_WORK_USED(work_area) ((work_area).used)
2009 #define RANGE_TABLE_WORK_BITS(work_area) ((work_area).bits)
2010 #define RANGE_TABLE_WORK_ELT(work_area, i) ((work_area).table[i])
2011 \f
2012
2013 /* Set the bit for character C in a list. */
2014 #define SET_LIST_BIT(c) (b[((c)) / BYTEWIDTH] |= 1 << ((c) % BYTEWIDTH))
2015
2016
2017 #ifdef emacs
2018
2019 /* Store characters in the range FROM to TO in the bitmap at B (for
2020 ASCII and unibyte characters) and WORK_AREA (for multibyte
2021 characters) while translating them and paying attention to the
2022 continuity of translated characters.
2023
2024 Implementation note: It is better to implement these fairly big
2025 macros by a function, but it's not that easy because macros called
2026 in this macro assume various local variables already declared. */
2027
2028 /* Both FROM and TO are ASCII characters. */
2029
2030 #define SETUP_ASCII_RANGE(work_area, FROM, TO) \
2031 do { \
2032 int C0, C1; \
2033 \
2034 for (C0 = (FROM); C0 <= (TO); C0++) \
2035 { \
2036 C1 = TRANSLATE (C0); \
2037 if (! ASCII_CHAR_P (C1)) \
2038 { \
2039 SET_RANGE_TABLE_WORK_AREA ((work_area), C1, C1); \
2040 if ((C1 = RE_CHAR_TO_UNIBYTE (C1)) < 0) \
2041 C1 = C0; \
2042 } \
2043 SET_LIST_BIT (C1); \
2044 } \
2045 } while (0)
2046
2047
2048 /* Both FROM and TO are unibyte characters (0x80..0xFF). */
2049
2050 #define SETUP_UNIBYTE_RANGE(work_area, FROM, TO) \
2051 do { \
2052 int C0, C1, C2, I; \
2053 int USED = RANGE_TABLE_WORK_USED (work_area); \
2054 \
2055 for (C0 = (FROM); C0 <= (TO); C0++) \
2056 { \
2057 C1 = RE_CHAR_TO_MULTIBYTE (C0); \
2058 if (CHAR_BYTE8_P (C1)) \
2059 SET_LIST_BIT (C0); \
2060 else \
2061 { \
2062 C2 = TRANSLATE (C1); \
2063 if (C2 == C1 \
2064 || (C1 = RE_CHAR_TO_UNIBYTE (C2)) < 0) \
2065 C1 = C0; \
2066 SET_LIST_BIT (C1); \
2067 for (I = RANGE_TABLE_WORK_USED (work_area) - 2; I >= USED; I -= 2) \
2068 { \
2069 int from = RANGE_TABLE_WORK_ELT (work_area, I); \
2070 int to = RANGE_TABLE_WORK_ELT (work_area, I + 1); \
2071 \
2072 if (C2 >= from - 1 && C2 <= to + 1) \
2073 { \
2074 if (C2 == from - 1) \
2075 RANGE_TABLE_WORK_ELT (work_area, I)--; \
2076 else if (C2 == to + 1) \
2077 RANGE_TABLE_WORK_ELT (work_area, I + 1)++; \
2078 break; \
2079 } \
2080 } \
2081 if (I < USED) \
2082 SET_RANGE_TABLE_WORK_AREA ((work_area), C2, C2); \
2083 } \
2084 } \
2085 } while (0)
2086
2087
2088 /* Both FROM and TO are mulitbyte characters. */
2089
2090 #define SETUP_MULTIBYTE_RANGE(work_area, FROM, TO) \
2091 do { \
2092 int C0, C1, C2, I, USED = RANGE_TABLE_WORK_USED (work_area); \
2093 \
2094 SET_RANGE_TABLE_WORK_AREA ((work_area), (FROM), (TO)); \
2095 for (C0 = (FROM); C0 <= (TO); C0++) \
2096 { \
2097 C1 = TRANSLATE (C0); \
2098 if ((C2 = RE_CHAR_TO_UNIBYTE (C1)) >= 0 \
2099 || (C1 != C0 && (C2 = RE_CHAR_TO_UNIBYTE (C0)) >= 0)) \
2100 SET_LIST_BIT (C2); \
2101 if (C1 >= (FROM) && C1 <= (TO)) \
2102 continue; \
2103 for (I = RANGE_TABLE_WORK_USED (work_area) - 2; I >= USED; I -= 2) \
2104 { \
2105 int from = RANGE_TABLE_WORK_ELT (work_area, I); \
2106 int to = RANGE_TABLE_WORK_ELT (work_area, I + 1); \
2107 \
2108 if (C1 >= from - 1 && C1 <= to + 1) \
2109 { \
2110 if (C1 == from - 1) \
2111 RANGE_TABLE_WORK_ELT (work_area, I)--; \
2112 else if (C1 == to + 1) \
2113 RANGE_TABLE_WORK_ELT (work_area, I + 1)++; \
2114 break; \
2115 } \
2116 } \
2117 if (I < USED) \
2118 SET_RANGE_TABLE_WORK_AREA ((work_area), C1, C1); \
2119 } \
2120 } while (0)
2121
2122 #endif /* emacs */
2123
2124 /* Get the next unsigned number in the uncompiled pattern. */
2125 #define GET_UNSIGNED_NUMBER(num) \
2126 do { \
2127 if (p == pend) \
2128 FREE_STACK_RETURN (REG_EBRACE); \
2129 else \
2130 { \
2131 PATFETCH (c); \
2132 while ('0' <= c && c <= '9') \
2133 { \
2134 int prev; \
2135 if (num < 0) \
2136 num = 0; \
2137 prev = num; \
2138 num = num * 10 + c - '0'; \
2139 if (num / 10 != prev) \
2140 FREE_STACK_RETURN (REG_BADBR); \
2141 if (p == pend) \
2142 FREE_STACK_RETURN (REG_EBRACE); \
2143 PATFETCH (c); \
2144 } \
2145 } \
2146 } while (0)
2147 \f
2148 #if ! WIDE_CHAR_SUPPORT
2149
2150 /* Map a string to the char class it names (if any). */
2151 re_wctype_t
2152 re_wctype (str)
2153 re_char *str;
2154 {
2155 const char *string = str;
2156 if (STREQ (string, "alnum")) return RECC_ALNUM;
2157 else if (STREQ (string, "alpha")) return RECC_ALPHA;
2158 else if (STREQ (string, "word")) return RECC_WORD;
2159 else if (STREQ (string, "ascii")) return RECC_ASCII;
2160 else if (STREQ (string, "nonascii")) return RECC_NONASCII;
2161 else if (STREQ (string, "graph")) return RECC_GRAPH;
2162 else if (STREQ (string, "lower")) return RECC_LOWER;
2163 else if (STREQ (string, "print")) return RECC_PRINT;
2164 else if (STREQ (string, "punct")) return RECC_PUNCT;
2165 else if (STREQ (string, "space")) return RECC_SPACE;
2166 else if (STREQ (string, "upper")) return RECC_UPPER;
2167 else if (STREQ (string, "unibyte")) return RECC_UNIBYTE;
2168 else if (STREQ (string, "multibyte")) return RECC_MULTIBYTE;
2169 else if (STREQ (string, "digit")) return RECC_DIGIT;
2170 else if (STREQ (string, "xdigit")) return RECC_XDIGIT;
2171 else if (STREQ (string, "cntrl")) return RECC_CNTRL;
2172 else if (STREQ (string, "blank")) return RECC_BLANK;
2173 else return 0;
2174 }
2175
2176 /* True if CH is in the char class CC. */
2177 boolean
2178 re_iswctype (ch, cc)
2179 int ch;
2180 re_wctype_t cc;
2181 {
2182 switch (cc)
2183 {
2184 case RECC_ALNUM: return ISALNUM (ch);
2185 case RECC_ALPHA: return ISALPHA (ch);
2186 case RECC_BLANK: return ISBLANK (ch);
2187 case RECC_CNTRL: return ISCNTRL (ch);
2188 case RECC_DIGIT: return ISDIGIT (ch);
2189 case RECC_GRAPH: return ISGRAPH (ch);
2190 case RECC_LOWER: return ISLOWER (ch);
2191 case RECC_PRINT: return ISPRINT (ch);
2192 case RECC_PUNCT: return ISPUNCT (ch);
2193 case RECC_SPACE: return ISSPACE (ch);
2194 case RECC_UPPER: return ISUPPER (ch);
2195 case RECC_XDIGIT: return ISXDIGIT (ch);
2196 case RECC_ASCII: return IS_REAL_ASCII (ch);
2197 case RECC_NONASCII: return !IS_REAL_ASCII (ch);
2198 case RECC_UNIBYTE: return ISUNIBYTE (ch);
2199 case RECC_MULTIBYTE: return !ISUNIBYTE (ch);
2200 case RECC_WORD: return ISWORD (ch);
2201 case RECC_ERROR: return false;
2202 default:
2203 abort();
2204 }
2205 }
2206
2207 /* Return a bit-pattern to use in the range-table bits to match multibyte
2208 chars of class CC. */
2209 static int
2210 re_wctype_to_bit (cc)
2211 re_wctype_t cc;
2212 {
2213 switch (cc)
2214 {
2215 case RECC_NONASCII: case RECC_PRINT: case RECC_GRAPH:
2216 case RECC_MULTIBYTE: return BIT_MULTIBYTE;
2217 case RECC_ALPHA: case RECC_ALNUM: case RECC_WORD: return BIT_WORD;
2218 case RECC_LOWER: return BIT_LOWER;
2219 case RECC_UPPER: return BIT_UPPER;
2220 case RECC_PUNCT: return BIT_PUNCT;
2221 case RECC_SPACE: return BIT_SPACE;
2222 case RECC_ASCII: case RECC_DIGIT: case RECC_XDIGIT: case RECC_CNTRL:
2223 case RECC_BLANK: case RECC_UNIBYTE: case RECC_ERROR: return 0;
2224 default:
2225 abort();
2226 }
2227 }
2228 #endif
2229 \f
2230 /* Filling in the work area of a range. */
2231
2232 /* Actually extend the space in WORK_AREA. */
2233
2234 static void
2235 extend_range_table_work_area (work_area)
2236 struct range_table_work_area *work_area;
2237 {
2238 work_area->allocated += 16 * sizeof (int);
2239 if (work_area->table)
2240 work_area->table
2241 = (int *) realloc (work_area->table, work_area->allocated);
2242 else
2243 work_area->table
2244 = (int *) malloc (work_area->allocated);
2245 }
2246
2247 #if 0
2248 #ifdef emacs
2249
2250 /* Carefully find the ranges of codes that are equivalent
2251 under case conversion to the range start..end when passed through
2252 TRANSLATE. Handle the case where non-letters can come in between
2253 two upper-case letters (which happens in Latin-1).
2254 Also handle the case of groups of more than 2 case-equivalent chars.
2255
2256 The basic method is to look at consecutive characters and see
2257 if they can form a run that can be handled as one.
2258
2259 Returns -1 if successful, REG_ESPACE if ran out of space. */
2260
2261 static int
2262 set_image_of_range_1 (work_area, start, end, translate)
2263 RE_TRANSLATE_TYPE translate;
2264 struct range_table_work_area *work_area;
2265 re_wchar_t start, end;
2266 {
2267 /* `one_case' indicates a character, or a run of characters,
2268 each of which is an isolate (no case-equivalents).
2269 This includes all ASCII non-letters.
2270
2271 `two_case' indicates a character, or a run of characters,
2272 each of which has two case-equivalent forms.
2273 This includes all ASCII letters.
2274
2275 `strange' indicates a character that has more than one
2276 case-equivalent. */
2277
2278 enum case_type {one_case, two_case, strange};
2279
2280 /* Describe the run that is in progress,
2281 which the next character can try to extend.
2282 If run_type is strange, that means there really is no run.
2283 If run_type is one_case, then run_start...run_end is the run.
2284 If run_type is two_case, then the run is run_start...run_end,
2285 and the case-equivalents end at run_eqv_end. */
2286
2287 enum case_type run_type = strange;
2288 int run_start, run_end, run_eqv_end;
2289
2290 Lisp_Object eqv_table;
2291
2292 if (!RE_TRANSLATE_P (translate))
2293 {
2294 EXTEND_RANGE_TABLE (work_area, 2);
2295 work_area->table[work_area->used++] = (start);
2296 work_area->table[work_area->used++] = (end);
2297 return -1;
2298 }
2299
2300 eqv_table = XCHAR_TABLE (translate)->extras[2];
2301
2302 for (; start <= end; start++)
2303 {
2304 enum case_type this_type;
2305 int eqv = RE_TRANSLATE (eqv_table, start);
2306 int minchar, maxchar;
2307
2308 /* Classify this character */
2309 if (eqv == start)
2310 this_type = one_case;
2311 else if (RE_TRANSLATE (eqv_table, eqv) == start)
2312 this_type = two_case;
2313 else
2314 this_type = strange;
2315
2316 if (start < eqv)
2317 minchar = start, maxchar = eqv;
2318 else
2319 minchar = eqv, maxchar = start;
2320
2321 /* Can this character extend the run in progress? */
2322 if (this_type == strange || this_type != run_type
2323 || !(minchar == run_end + 1
2324 && (run_type == two_case
2325 ? maxchar == run_eqv_end + 1 : 1)))
2326 {
2327 /* No, end the run.
2328 Record each of its equivalent ranges. */
2329 if (run_type == one_case)
2330 {
2331 EXTEND_RANGE_TABLE (work_area, 2);
2332 work_area->table[work_area->used++] = run_start;
2333 work_area->table[work_area->used++] = run_end;
2334 }
2335 else if (run_type == two_case)
2336 {
2337 EXTEND_RANGE_TABLE (work_area, 4);
2338 work_area->table[work_area->used++] = run_start;
2339 work_area->table[work_area->used++] = run_end;
2340 work_area->table[work_area->used++]
2341 = RE_TRANSLATE (eqv_table, run_start);
2342 work_area->table[work_area->used++]
2343 = RE_TRANSLATE (eqv_table, run_end);
2344 }
2345 run_type = strange;
2346 }
2347
2348 if (this_type == strange)
2349 {
2350 /* For a strange character, add each of its equivalents, one
2351 by one. Don't start a range. */
2352 do
2353 {
2354 EXTEND_RANGE_TABLE (work_area, 2);
2355 work_area->table[work_area->used++] = eqv;
2356 work_area->table[work_area->used++] = eqv;
2357 eqv = RE_TRANSLATE (eqv_table, eqv);
2358 }
2359 while (eqv != start);
2360 }
2361
2362 /* Add this char to the run, or start a new run. */
2363 else if (run_type == strange)
2364 {
2365 /* Initialize a new range. */
2366 run_type = this_type;
2367 run_start = start;
2368 run_end = start;
2369 run_eqv_end = RE_TRANSLATE (eqv_table, run_end);
2370 }
2371 else
2372 {
2373 /* Extend a running range. */
2374 run_end = minchar;
2375 run_eqv_end = RE_TRANSLATE (eqv_table, run_end);
2376 }
2377 }
2378
2379 /* If a run is still in progress at the end, finish it now
2380 by recording its equivalent ranges. */
2381 if (run_type == one_case)
2382 {
2383 EXTEND_RANGE_TABLE (work_area, 2);
2384 work_area->table[work_area->used++] = run_start;
2385 work_area->table[work_area->used++] = run_end;
2386 }
2387 else if (run_type == two_case)
2388 {
2389 EXTEND_RANGE_TABLE (work_area, 4);
2390 work_area->table[work_area->used++] = run_start;
2391 work_area->table[work_area->used++] = run_end;
2392 work_area->table[work_area->used++]
2393 = RE_TRANSLATE (eqv_table, run_start);
2394 work_area->table[work_area->used++]
2395 = RE_TRANSLATE (eqv_table, run_end);
2396 }
2397
2398 return -1;
2399 }
2400
2401 #endif /* emacs */
2402
2403 /* Record the the image of the range start..end when passed through
2404 TRANSLATE. This is not necessarily TRANSLATE(start)..TRANSLATE(end)
2405 and is not even necessarily contiguous.
2406 Normally we approximate it with the smallest contiguous range that contains
2407 all the chars we need. However, for Latin-1 we go to extra effort
2408 to do a better job.
2409
2410 This function is not called for ASCII ranges.
2411
2412 Returns -1 if successful, REG_ESPACE if ran out of space. */
2413
2414 static int
2415 set_image_of_range (work_area, start, end, translate)
2416 RE_TRANSLATE_TYPE translate;
2417 struct range_table_work_area *work_area;
2418 re_wchar_t start, end;
2419 {
2420 re_wchar_t cmin, cmax;
2421
2422 #ifdef emacs
2423 /* For Latin-1 ranges, use set_image_of_range_1
2424 to get proper handling of ranges that include letters and nonletters.
2425 For a range that includes the whole of Latin-1, this is not necessary.
2426 For other character sets, we don't bother to get this right. */
2427 if (RE_TRANSLATE_P (translate) && start < 04400
2428 && !(start < 04200 && end >= 04377))
2429 {
2430 int newend;
2431 int tem;
2432 newend = end;
2433 if (newend > 04377)
2434 newend = 04377;
2435 tem = set_image_of_range_1 (work_area, start, newend, translate);
2436 if (tem > 0)
2437 return tem;
2438
2439 start = 04400;
2440 if (end < 04400)
2441 return -1;
2442 }
2443 #endif
2444
2445 EXTEND_RANGE_TABLE (work_area, 2);
2446 work_area->table[work_area->used++] = (start);
2447 work_area->table[work_area->used++] = (end);
2448
2449 cmin = -1, cmax = -1;
2450
2451 if (RE_TRANSLATE_P (translate))
2452 {
2453 int ch;
2454
2455 for (ch = start; ch <= end; ch++)
2456 {
2457 re_wchar_t c = TRANSLATE (ch);
2458 if (! (start <= c && c <= end))
2459 {
2460 if (cmin == -1)
2461 cmin = c, cmax = c;
2462 else
2463 {
2464 cmin = MIN (cmin, c);
2465 cmax = MAX (cmax, c);
2466 }
2467 }
2468 }
2469
2470 if (cmin != -1)
2471 {
2472 EXTEND_RANGE_TABLE (work_area, 2);
2473 work_area->table[work_area->used++] = (cmin);
2474 work_area->table[work_area->used++] = (cmax);
2475 }
2476 }
2477
2478 return -1;
2479 }
2480 #endif /* 0 */
2481 \f
2482 #ifndef MATCH_MAY_ALLOCATE
2483
2484 /* If we cannot allocate large objects within re_match_2_internal,
2485 we make the fail stack and register vectors global.
2486 The fail stack, we grow to the maximum size when a regexp
2487 is compiled.
2488 The register vectors, we adjust in size each time we
2489 compile a regexp, according to the number of registers it needs. */
2490
2491 static fail_stack_type fail_stack;
2492
2493 /* Size with which the following vectors are currently allocated.
2494 That is so we can make them bigger as needed,
2495 but never make them smaller. */
2496 static int regs_allocated_size;
2497
2498 static re_char ** regstart, ** regend;
2499 static re_char **best_regstart, **best_regend;
2500
2501 /* Make the register vectors big enough for NUM_REGS registers,
2502 but don't make them smaller. */
2503
2504 static
2505 regex_grow_registers (num_regs)
2506 int num_regs;
2507 {
2508 if (num_regs > regs_allocated_size)
2509 {
2510 RETALLOC_IF (regstart, num_regs, re_char *);
2511 RETALLOC_IF (regend, num_regs, re_char *);
2512 RETALLOC_IF (best_regstart, num_regs, re_char *);
2513 RETALLOC_IF (best_regend, num_regs, re_char *);
2514
2515 regs_allocated_size = num_regs;
2516 }
2517 }
2518
2519 #endif /* not MATCH_MAY_ALLOCATE */
2520 \f
2521 static boolean group_in_compile_stack _RE_ARGS ((compile_stack_type
2522 compile_stack,
2523 regnum_t regnum));
2524
2525 /* `regex_compile' compiles PATTERN (of length SIZE) according to SYNTAX.
2526 Returns one of error codes defined in `regex.h', or zero for success.
2527
2528 Assumes the `allocated' (and perhaps `buffer') and `translate'
2529 fields are set in BUFP on entry.
2530
2531 If it succeeds, results are put in BUFP (if it returns an error, the
2532 contents of BUFP are undefined):
2533 `buffer' is the compiled pattern;
2534 `syntax' is set to SYNTAX;
2535 `used' is set to the length of the compiled pattern;
2536 `fastmap_accurate' is zero;
2537 `re_nsub' is the number of subexpressions in PATTERN;
2538 `not_bol' and `not_eol' are zero;
2539
2540 The `fastmap' field is neither examined nor set. */
2541
2542 /* Insert the `jump' from the end of last alternative to "here".
2543 The space for the jump has already been allocated. */
2544 #define FIXUP_ALT_JUMP() \
2545 do { \
2546 if (fixup_alt_jump) \
2547 STORE_JUMP (jump, fixup_alt_jump, b); \
2548 } while (0)
2549
2550
2551 /* Return, freeing storage we allocated. */
2552 #define FREE_STACK_RETURN(value) \
2553 do { \
2554 FREE_RANGE_TABLE_WORK_AREA (range_table_work); \
2555 free (compile_stack.stack); \
2556 return value; \
2557 } while (0)
2558
2559 static reg_errcode_t
2560 regex_compile (pattern, size, syntax, bufp)
2561 re_char *pattern;
2562 size_t size;
2563 reg_syntax_t syntax;
2564 struct re_pattern_buffer *bufp;
2565 {
2566 /* We fetch characters from PATTERN here. */
2567 register re_wchar_t c, c1;
2568
2569 /* A random temporary spot in PATTERN. */
2570 re_char *p1;
2571
2572 /* Points to the end of the buffer, where we should append. */
2573 register unsigned char *b;
2574
2575 /* Keeps track of unclosed groups. */
2576 compile_stack_type compile_stack;
2577
2578 /* Points to the current (ending) position in the pattern. */
2579 #ifdef AIX
2580 /* `const' makes AIX compiler fail. */
2581 unsigned char *p = pattern;
2582 #else
2583 re_char *p = pattern;
2584 #endif
2585 re_char *pend = pattern + size;
2586
2587 /* How to translate the characters in the pattern. */
2588 RE_TRANSLATE_TYPE translate = bufp->translate;
2589
2590 /* Address of the count-byte of the most recently inserted `exactn'
2591 command. This makes it possible to tell if a new exact-match
2592 character can be added to that command or if the character requires
2593 a new `exactn' command. */
2594 unsigned char *pending_exact = 0;
2595
2596 /* Address of start of the most recently finished expression.
2597 This tells, e.g., postfix * where to find the start of its
2598 operand. Reset at the beginning of groups and alternatives. */
2599 unsigned char *laststart = 0;
2600
2601 /* Address of beginning of regexp, or inside of last group. */
2602 unsigned char *begalt;
2603
2604 /* Place in the uncompiled pattern (i.e., the {) to
2605 which to go back if the interval is invalid. */
2606 re_char *beg_interval;
2607
2608 /* Address of the place where a forward jump should go to the end of
2609 the containing expression. Each alternative of an `or' -- except the
2610 last -- ends with a forward jump of this sort. */
2611 unsigned char *fixup_alt_jump = 0;
2612
2613 /* Work area for range table of charset. */
2614 struct range_table_work_area range_table_work;
2615
2616 /* If the object matched can contain multibyte characters. */
2617 const boolean multibyte = RE_MULTIBYTE_P (bufp);
2618
2619 /* If a target of matching can contain multibyte characters. */
2620 const boolean target_multibyte = RE_TARGET_MULTIBYTE_P (bufp);
2621
2622 /* Nonzero if we have pushed down into a subpattern. */
2623 int in_subpattern = 0;
2624
2625 /* These hold the values of p, pattern, and pend from the main
2626 pattern when we have pushed into a subpattern. */
2627 re_char *main_p;
2628 re_char *main_pattern;
2629 re_char *main_pend;
2630
2631 #ifdef DEBUG
2632 debug++;
2633 DEBUG_PRINT1 ("\nCompiling pattern: ");
2634 if (debug > 0)
2635 {
2636 unsigned debug_count;
2637
2638 for (debug_count = 0; debug_count < size; debug_count++)
2639 putchar (pattern[debug_count]);
2640 putchar ('\n');
2641 }
2642 #endif /* DEBUG */
2643
2644 /* Initialize the compile stack. */
2645 compile_stack.stack = TALLOC (INIT_COMPILE_STACK_SIZE, compile_stack_elt_t);
2646 if (compile_stack.stack == NULL)
2647 return REG_ESPACE;
2648
2649 compile_stack.size = INIT_COMPILE_STACK_SIZE;
2650 compile_stack.avail = 0;
2651
2652 range_table_work.table = 0;
2653 range_table_work.allocated = 0;
2654
2655 /* Initialize the pattern buffer. */
2656 bufp->syntax = syntax;
2657 bufp->fastmap_accurate = 0;
2658 bufp->not_bol = bufp->not_eol = 0;
2659 bufp->used_syntax = 0;
2660
2661 /* Set `used' to zero, so that if we return an error, the pattern
2662 printer (for debugging) will think there's no pattern. We reset it
2663 at the end. */
2664 bufp->used = 0;
2665
2666 /* Always count groups, whether or not bufp->no_sub is set. */
2667 bufp->re_nsub = 0;
2668
2669 #if !defined emacs && !defined SYNTAX_TABLE
2670 /* Initialize the syntax table. */
2671 init_syntax_once ();
2672 #endif
2673
2674 if (bufp->allocated == 0)
2675 {
2676 if (bufp->buffer)
2677 { /* If zero allocated, but buffer is non-null, try to realloc
2678 enough space. This loses if buffer's address is bogus, but
2679 that is the user's responsibility. */
2680 RETALLOC (bufp->buffer, INIT_BUF_SIZE, unsigned char);
2681 }
2682 else
2683 { /* Caller did not allocate a buffer. Do it for them. */
2684 bufp->buffer = TALLOC (INIT_BUF_SIZE, unsigned char);
2685 }
2686 if (!bufp->buffer) FREE_STACK_RETURN (REG_ESPACE);
2687
2688 bufp->allocated = INIT_BUF_SIZE;
2689 }
2690
2691 begalt = b = bufp->buffer;
2692
2693 /* Loop through the uncompiled pattern until we're at the end. */
2694 while (1)
2695 {
2696 if (p == pend)
2697 {
2698 /* If this is the end of an included regexp,
2699 pop back to the main regexp and try again. */
2700 if (in_subpattern)
2701 {
2702 in_subpattern = 0;
2703 pattern = main_pattern;
2704 p = main_p;
2705 pend = main_pend;
2706 continue;
2707 }
2708 /* If this is the end of the main regexp, we are done. */
2709 break;
2710 }
2711
2712 PATFETCH (c);
2713
2714 switch (c)
2715 {
2716 case ' ':
2717 {
2718 re_char *p1 = p;
2719
2720 /* If there's no special whitespace regexp, treat
2721 spaces normally. And don't try to do this recursively. */
2722 if (!whitespace_regexp || in_subpattern)
2723 goto normal_char;
2724
2725 /* Peek past following spaces. */
2726 while (p1 != pend)
2727 {
2728 if (*p1 != ' ')
2729 break;
2730 p1++;
2731 }
2732 /* If the spaces are followed by a repetition op,
2733 treat them normally. */
2734 if (p1 != pend
2735 && (*p1 == '*' || *p1 == '+' || *p1 == '?'
2736 || (*p1 == '\\' && p1 + 1 != pend && p1[1] == '{')))
2737 goto normal_char;
2738
2739 /* Replace the spaces with the whitespace regexp. */
2740 in_subpattern = 1;
2741 main_p = p1;
2742 main_pend = pend;
2743 main_pattern = pattern;
2744 p = pattern = whitespace_regexp;
2745 pend = p + strlen (p);
2746 break;
2747 }
2748
2749 case '^':
2750 {
2751 if ( /* If at start of pattern, it's an operator. */
2752 p == pattern + 1
2753 /* If context independent, it's an operator. */
2754 || syntax & RE_CONTEXT_INDEP_ANCHORS
2755 /* Otherwise, depends on what's come before. */
2756 || at_begline_loc_p (pattern, p, syntax))
2757 BUF_PUSH ((syntax & RE_NO_NEWLINE_ANCHOR) ? begbuf : begline);
2758 else
2759 goto normal_char;
2760 }
2761 break;
2762
2763
2764 case '$':
2765 {
2766 if ( /* If at end of pattern, it's an operator. */
2767 p == pend
2768 /* If context independent, it's an operator. */
2769 || syntax & RE_CONTEXT_INDEP_ANCHORS
2770 /* Otherwise, depends on what's next. */
2771 || at_endline_loc_p (p, pend, syntax))
2772 BUF_PUSH ((syntax & RE_NO_NEWLINE_ANCHOR) ? endbuf : endline);
2773 else
2774 goto normal_char;
2775 }
2776 break;
2777
2778
2779 case '+':
2780 case '?':
2781 if ((syntax & RE_BK_PLUS_QM)
2782 || (syntax & RE_LIMITED_OPS))
2783 goto normal_char;
2784 handle_plus:
2785 case '*':
2786 /* If there is no previous pattern... */
2787 if (!laststart)
2788 {
2789 if (syntax & RE_CONTEXT_INVALID_OPS)
2790 FREE_STACK_RETURN (REG_BADRPT);
2791 else if (!(syntax & RE_CONTEXT_INDEP_OPS))
2792 goto normal_char;
2793 }
2794
2795 {
2796 /* 1 means zero (many) matches is allowed. */
2797 boolean zero_times_ok = 0, many_times_ok = 0;
2798 boolean greedy = 1;
2799
2800 /* If there is a sequence of repetition chars, collapse it
2801 down to just one (the right one). We can't combine
2802 interval operators with these because of, e.g., `a{2}*',
2803 which should only match an even number of `a's. */
2804
2805 for (;;)
2806 {
2807 if ((syntax & RE_FRUGAL)
2808 && c == '?' && (zero_times_ok || many_times_ok))
2809 greedy = 0;
2810 else
2811 {
2812 zero_times_ok |= c != '+';
2813 many_times_ok |= c != '?';
2814 }
2815
2816 if (p == pend)
2817 break;
2818 else if (*p == '*'
2819 || (!(syntax & RE_BK_PLUS_QM)
2820 && (*p == '+' || *p == '?')))
2821 ;
2822 else if (syntax & RE_BK_PLUS_QM && *p == '\\')
2823 {
2824 if (p+1 == pend)
2825 FREE_STACK_RETURN (REG_EESCAPE);
2826 if (p[1] == '+' || p[1] == '?')
2827 PATFETCH (c); /* Gobble up the backslash. */
2828 else
2829 break;
2830 }
2831 else
2832 break;
2833 /* If we get here, we found another repeat character. */
2834 PATFETCH (c);
2835 }
2836
2837 /* Star, etc. applied to an empty pattern is equivalent
2838 to an empty pattern. */
2839 if (!laststart || laststart == b)
2840 break;
2841
2842 /* Now we know whether or not zero matches is allowed
2843 and also whether or not two or more matches is allowed. */
2844 if (greedy)
2845 {
2846 if (many_times_ok)
2847 {
2848 boolean simple = skip_one_char (laststart) == b;
2849 unsigned int startoffset = 0;
2850 re_opcode_t ofj =
2851 /* Check if the loop can match the empty string. */
2852 (simple || !analyse_first (laststart, b, NULL, 0))
2853 ? on_failure_jump : on_failure_jump_loop;
2854 assert (skip_one_char (laststart) <= b);
2855
2856 if (!zero_times_ok && simple)
2857 { /* Since simple * loops can be made faster by using
2858 on_failure_keep_string_jump, we turn simple P+
2859 into PP* if P is simple. */
2860 unsigned char *p1, *p2;
2861 startoffset = b - laststart;
2862 GET_BUFFER_SPACE (startoffset);
2863 p1 = b; p2 = laststart;
2864 while (p2 < p1)
2865 *b++ = *p2++;
2866 zero_times_ok = 1;
2867 }
2868
2869 GET_BUFFER_SPACE (6);
2870 if (!zero_times_ok)
2871 /* A + loop. */
2872 STORE_JUMP (ofj, b, b + 6);
2873 else
2874 /* Simple * loops can use on_failure_keep_string_jump
2875 depending on what follows. But since we don't know
2876 that yet, we leave the decision up to
2877 on_failure_jump_smart. */
2878 INSERT_JUMP (simple ? on_failure_jump_smart : ofj,
2879 laststart + startoffset, b + 6);
2880 b += 3;
2881 STORE_JUMP (jump, b, laststart + startoffset);
2882 b += 3;
2883 }
2884 else
2885 {
2886 /* A simple ? pattern. */
2887 assert (zero_times_ok);
2888 GET_BUFFER_SPACE (3);
2889 INSERT_JUMP (on_failure_jump, laststart, b + 3);
2890 b += 3;
2891 }
2892 }
2893 else /* not greedy */
2894 { /* I wish the greedy and non-greedy cases could be merged. */
2895
2896 GET_BUFFER_SPACE (7); /* We might use less. */
2897 if (many_times_ok)
2898 {
2899 boolean emptyp = analyse_first (laststart, b, NULL, 0);
2900
2901 /* The non-greedy multiple match looks like
2902 a repeat..until: we only need a conditional jump
2903 at the end of the loop. */
2904 if (emptyp) BUF_PUSH (no_op);
2905 STORE_JUMP (emptyp ? on_failure_jump_nastyloop
2906 : on_failure_jump, b, laststart);
2907 b += 3;
2908 if (zero_times_ok)
2909 {
2910 /* The repeat...until naturally matches one or more.
2911 To also match zero times, we need to first jump to
2912 the end of the loop (its conditional jump). */
2913 INSERT_JUMP (jump, laststart, b);
2914 b += 3;
2915 }
2916 }
2917 else
2918 {
2919 /* non-greedy a?? */
2920 INSERT_JUMP (jump, laststart, b + 3);
2921 b += 3;
2922 INSERT_JUMP (on_failure_jump, laststart, laststart + 6);
2923 b += 3;
2924 }
2925 }
2926 }
2927 pending_exact = 0;
2928 break;
2929
2930
2931 case '.':
2932 laststart = b;
2933 BUF_PUSH (anychar);
2934 break;
2935
2936
2937 case '[':
2938 {
2939 CLEAR_RANGE_TABLE_WORK_USED (range_table_work);
2940
2941 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2942
2943 /* Ensure that we have enough space to push a charset: the
2944 opcode, the length count, and the bitset; 34 bytes in all. */
2945 GET_BUFFER_SPACE (34);
2946
2947 laststart = b;
2948
2949 /* We test `*p == '^' twice, instead of using an if
2950 statement, so we only need one BUF_PUSH. */
2951 BUF_PUSH (*p == '^' ? charset_not : charset);
2952 if (*p == '^')
2953 p++;
2954
2955 /* Remember the first position in the bracket expression. */
2956 p1 = p;
2957
2958 /* Push the number of bytes in the bitmap. */
2959 BUF_PUSH ((1 << BYTEWIDTH) / BYTEWIDTH);
2960
2961 /* Clear the whole map. */
2962 bzero (b, (1 << BYTEWIDTH) / BYTEWIDTH);
2963
2964 /* charset_not matches newline according to a syntax bit. */
2965 if ((re_opcode_t) b[-2] == charset_not
2966 && (syntax & RE_HAT_LISTS_NOT_NEWLINE))
2967 SET_LIST_BIT ('\n');
2968
2969 /* Read in characters and ranges, setting map bits. */
2970 for (;;)
2971 {
2972 boolean escaped_char = false;
2973 const unsigned char *p2 = p;
2974 re_wchar_t ch, c2;
2975
2976 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2977
2978 /* Don't translate yet. The range TRANSLATE(X..Y) cannot
2979 always be determined from TRANSLATE(X) and TRANSLATE(Y)
2980 So the translation is done later in a loop. Example:
2981 (let ((case-fold-search t)) (string-match "[A-_]" "A")) */
2982 PATFETCH (c);
2983
2984 /* \ might escape characters inside [...] and [^...]. */
2985 if ((syntax & RE_BACKSLASH_ESCAPE_IN_LISTS) && c == '\\')
2986 {
2987 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
2988
2989 PATFETCH (c);
2990 escaped_char = true;
2991 }
2992 else
2993 {
2994 /* Could be the end of the bracket expression. If it's
2995 not (i.e., when the bracket expression is `[]' so
2996 far), the ']' character bit gets set way below. */
2997 if (c == ']' && p2 != p1)
2998 break;
2999 }
3000
3001 /* See if we're at the beginning of a possible character
3002 class. */
3003
3004 if (!escaped_char &&
3005 syntax & RE_CHAR_CLASSES && c == '[' && *p == ':')
3006 {
3007 /* Leave room for the null. */
3008 unsigned char str[CHAR_CLASS_MAX_LENGTH + 1];
3009 const unsigned char *class_beg;
3010
3011 PATFETCH (c);
3012 c1 = 0;
3013 class_beg = p;
3014
3015 /* If pattern is `[[:'. */
3016 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
3017
3018 for (;;)
3019 {
3020 PATFETCH (c);
3021 if ((c == ':' && *p == ']') || p == pend)
3022 break;
3023 if (c1 < CHAR_CLASS_MAX_LENGTH)
3024 str[c1++] = c;
3025 else
3026 /* This is in any case an invalid class name. */
3027 str[0] = '\0';
3028 }
3029 str[c1] = '\0';
3030
3031 /* If isn't a word bracketed by `[:' and `:]':
3032 undo the ending character, the letters, and
3033 leave the leading `:' and `[' (but set bits for
3034 them). */
3035 if (c == ':' && *p == ']')
3036 {
3037 re_wctype_t cc;
3038 int limit;
3039
3040 cc = re_wctype (str);
3041
3042 if (cc == 0)
3043 FREE_STACK_RETURN (REG_ECTYPE);
3044
3045 /* Throw away the ] at the end of the character
3046 class. */
3047 PATFETCH (c);
3048
3049 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
3050
3051 #ifndef emacs
3052 for (ch = 0; ch < (1 << BYTEWIDTH); ++ch)
3053 if (re_iswctype (btowc (ch), cc))
3054 {
3055 c = TRANSLATE (ch);
3056 if (c < (1 << BYTEWIDTH))
3057 SET_LIST_BIT (c);
3058 }
3059 #else /* emacs */
3060 /* Most character classes in a multibyte match
3061 just set a flag. Exceptions are is_blank,
3062 is_digit, is_cntrl, and is_xdigit, since
3063 they can only match ASCII characters. We
3064 don't need to handle them for multibyte.
3065 They are distinguished by a negative wctype. */
3066
3067 for (ch = 0; ch < 256; ++ch)
3068 {
3069 c = RE_CHAR_TO_MULTIBYTE (ch);
3070 if (! CHAR_BYTE8_P (c)
3071 && re_iswctype (c, cc))
3072 {
3073 SET_LIST_BIT (ch);
3074 c1 = TRANSLATE (c);
3075 if (c1 == c)
3076 continue;
3077 if (ASCII_CHAR_P (c1))
3078 SET_LIST_BIT (c1);
3079 else if ((c1 = RE_CHAR_TO_UNIBYTE (c1)) >= 0)
3080 SET_LIST_BIT (c1);
3081 }
3082 }
3083 SET_RANGE_TABLE_WORK_AREA_BIT
3084 (range_table_work, re_wctype_to_bit (cc));
3085 #endif /* emacs */
3086 /* In most cases the matching rule for char classes
3087 only uses the syntax table for multibyte chars,
3088 so that the content of the syntax-table it is not
3089 hardcoded in the range_table. SPACE and WORD are
3090 the two exceptions. */
3091 if ((1 << cc) & ((1 << RECC_SPACE) | (1 << RECC_WORD)))
3092 bufp->used_syntax = 1;
3093
3094 /* Repeat the loop. */
3095 continue;
3096 }
3097 else
3098 {
3099 /* Go back to right after the "[:". */
3100 p = class_beg;
3101 SET_LIST_BIT ('[');
3102
3103 /* Because the `:' may starts the range, we
3104 can't simply set bit and repeat the loop.
3105 Instead, just set it to C and handle below. */
3106 c = ':';
3107 }
3108 }
3109
3110 if (p < pend && p[0] == '-' && p[1] != ']')
3111 {
3112
3113 /* Discard the `-'. */
3114 PATFETCH (c1);
3115
3116 /* Fetch the character which ends the range. */
3117 PATFETCH (c1);
3118 #ifdef emacs
3119 if (CHAR_BYTE8_P (c1)
3120 && ! ASCII_CHAR_P (c) && ! CHAR_BYTE8_P (c))
3121 /* Treat the range from a multibyte character to
3122 raw-byte character as empty. */
3123 c = c1 + 1;
3124 #endif /* emacs */
3125 }
3126 else
3127 /* Range from C to C. */
3128 c1 = c;
3129
3130 if (c > c1)
3131 {
3132 if (syntax & RE_NO_EMPTY_RANGES)
3133 FREE_STACK_RETURN (REG_ERANGEX);
3134 /* Else, repeat the loop. */
3135 }
3136 else
3137 {
3138 #ifndef emacs
3139 /* Set the range into bitmap */
3140 for (; c <= c1; c++)
3141 {
3142 ch = TRANSLATE (c);
3143 if (ch < (1 << BYTEWIDTH))
3144 SET_LIST_BIT (ch);
3145 }
3146 #else /* emacs */
3147 if (c < 128)
3148 {
3149 ch = MIN (127, c1);
3150 SETUP_ASCII_RANGE (range_table_work, c, ch);
3151 c = ch + 1;
3152 if (CHAR_BYTE8_P (c1))
3153 c = BYTE8_TO_CHAR (128);
3154 }
3155 if (c <= c1)
3156 {
3157 if (CHAR_BYTE8_P (c))
3158 {
3159 c = CHAR_TO_BYTE8 (c);
3160 c1 = CHAR_TO_BYTE8 (c1);
3161 for (; c <= c1; c++)
3162 SET_LIST_BIT (c);
3163 }
3164 else if (multibyte)
3165 {
3166 SETUP_MULTIBYTE_RANGE (range_table_work, c, c1);
3167 }
3168 else
3169 {
3170 SETUP_UNIBYTE_RANGE (range_table_work, c, c1);
3171 }
3172 }
3173 #endif /* emacs */
3174 }
3175 }
3176
3177 /* Discard any (non)matching list bytes that are all 0 at the
3178 end of the map. Decrease the map-length byte too. */
3179 while ((int) b[-1] > 0 && b[b[-1] - 1] == 0)
3180 b[-1]--;
3181 b += b[-1];
3182
3183 /* Build real range table from work area. */
3184 if (RANGE_TABLE_WORK_USED (range_table_work)
3185 || RANGE_TABLE_WORK_BITS (range_table_work))
3186 {
3187 int i;
3188 int used = RANGE_TABLE_WORK_USED (range_table_work);
3189
3190 /* Allocate space for COUNT + RANGE_TABLE. Needs two
3191 bytes for flags, two for COUNT, and three bytes for
3192 each character. */
3193 GET_BUFFER_SPACE (4 + used * 3);
3194
3195 /* Indicate the existence of range table. */
3196 laststart[1] |= 0x80;
3197
3198 /* Store the character class flag bits into the range table.
3199 If not in emacs, these flag bits are always 0. */
3200 *b++ = RANGE_TABLE_WORK_BITS (range_table_work) & 0xff;
3201 *b++ = RANGE_TABLE_WORK_BITS (range_table_work) >> 8;
3202
3203 STORE_NUMBER_AND_INCR (b, used / 2);
3204 for (i = 0; i < used; i++)
3205 STORE_CHARACTER_AND_INCR
3206 (b, RANGE_TABLE_WORK_ELT (range_table_work, i));
3207 }
3208 }
3209 break;
3210
3211
3212 case '(':
3213 if (syntax & RE_NO_BK_PARENS)
3214 goto handle_open;
3215 else
3216 goto normal_char;
3217
3218
3219 case ')':
3220 if (syntax & RE_NO_BK_PARENS)
3221 goto handle_close;
3222 else
3223 goto normal_char;
3224
3225
3226 case '\n':
3227 if (syntax & RE_NEWLINE_ALT)
3228 goto handle_alt;
3229 else
3230 goto normal_char;
3231
3232
3233 case '|':
3234 if (syntax & RE_NO_BK_VBAR)
3235 goto handle_alt;
3236 else
3237 goto normal_char;
3238
3239
3240 case '{':
3241 if (syntax & RE_INTERVALS && syntax & RE_NO_BK_BRACES)
3242 goto handle_interval;
3243 else
3244 goto normal_char;
3245
3246
3247 case '\\':
3248 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
3249
3250 /* Do not translate the character after the \, so that we can
3251 distinguish, e.g., \B from \b, even if we normally would
3252 translate, e.g., B to b. */
3253 PATFETCH (c);
3254
3255 switch (c)
3256 {
3257 case '(':
3258 if (syntax & RE_NO_BK_PARENS)
3259 goto normal_backslash;
3260
3261 handle_open:
3262 {
3263 int shy = 0;
3264 regnum_t regnum = 0;
3265 if (p+1 < pend)
3266 {
3267 /* Look for a special (?...) construct */
3268 if ((syntax & RE_SHY_GROUPS) && *p == '?')
3269 {
3270 PATFETCH (c); /* Gobble up the '?'. */
3271 while (!shy)
3272 {
3273 PATFETCH (c);
3274 switch (c)
3275 {
3276 case ':': shy = 1; break;
3277 case '0':
3278 /* An explicitly specified regnum must start
3279 with non-0. */
3280 if (regnum == 0)
3281 FREE_STACK_RETURN (REG_BADPAT);
3282 case '1': case '2': case '3': case '4':
3283 case '5': case '6': case '7': case '8': case '9':
3284 regnum = 10*regnum + (c - '0'); break;
3285 default:
3286 /* Only (?:...) is supported right now. */
3287 FREE_STACK_RETURN (REG_BADPAT);
3288 }
3289 }
3290 }
3291 }
3292
3293 if (!shy)
3294 regnum = ++bufp->re_nsub;
3295 else if (regnum)
3296 { /* It's actually not shy, but explicitly numbered. */
3297 shy = 0;
3298 if (regnum > bufp->re_nsub)
3299 bufp->re_nsub = regnum;
3300 else if (regnum > bufp->re_nsub
3301 /* Ideally, we'd want to check that the specified
3302 group can't have matched (i.e. all subgroups
3303 using the same regnum are in other branches of
3304 OR patterns), but we don't currently keep track
3305 of enough info to do that easily. */
3306 || group_in_compile_stack (compile_stack, regnum))
3307 FREE_STACK_RETURN (REG_BADPAT);
3308 }
3309 else
3310 /* It's really shy. */
3311 regnum = - bufp->re_nsub;
3312
3313 if (COMPILE_STACK_FULL)
3314 {
3315 RETALLOC (compile_stack.stack, compile_stack.size << 1,
3316 compile_stack_elt_t);
3317 if (compile_stack.stack == NULL) return REG_ESPACE;
3318
3319 compile_stack.size <<= 1;
3320 }
3321
3322 /* These are the values to restore when we hit end of this
3323 group. They are all relative offsets, so that if the
3324 whole pattern moves because of realloc, they will still
3325 be valid. */
3326 COMPILE_STACK_TOP.begalt_offset = begalt - bufp->buffer;
3327 COMPILE_STACK_TOP.fixup_alt_jump
3328 = fixup_alt_jump ? fixup_alt_jump - bufp->buffer + 1 : 0;
3329 COMPILE_STACK_TOP.laststart_offset = b - bufp->buffer;
3330 COMPILE_STACK_TOP.regnum = regnum;
3331
3332 /* Do not push a start_memory for groups beyond the last one
3333 we can represent in the compiled pattern. */
3334 if (regnum <= MAX_REGNUM && regnum > 0)
3335 BUF_PUSH_2 (start_memory, regnum);
3336
3337 compile_stack.avail++;
3338
3339 fixup_alt_jump = 0;
3340 laststart = 0;
3341 begalt = b;
3342 /* If we've reached MAX_REGNUM groups, then this open
3343 won't actually generate any code, so we'll have to
3344 clear pending_exact explicitly. */
3345 pending_exact = 0;
3346 break;
3347 }
3348
3349 case ')':
3350 if (syntax & RE_NO_BK_PARENS) goto normal_backslash;
3351
3352 if (COMPILE_STACK_EMPTY)
3353 {
3354 if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD)
3355 goto normal_backslash;
3356 else
3357 FREE_STACK_RETURN (REG_ERPAREN);
3358 }
3359
3360 handle_close:
3361 FIXUP_ALT_JUMP ();
3362
3363 /* See similar code for backslashed left paren above. */
3364 if (COMPILE_STACK_EMPTY)
3365 {
3366 if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD)
3367 goto normal_char;
3368 else
3369 FREE_STACK_RETURN (REG_ERPAREN);
3370 }
3371
3372 /* Since we just checked for an empty stack above, this
3373 ``can't happen''. */
3374 assert (compile_stack.avail != 0);
3375 {
3376 /* We don't just want to restore into `regnum', because
3377 later groups should continue to be numbered higher,
3378 as in `(ab)c(de)' -- the second group is #2. */
3379 regnum_t regnum;
3380
3381 compile_stack.avail--;
3382 begalt = bufp->buffer + COMPILE_STACK_TOP.begalt_offset;
3383 fixup_alt_jump
3384 = COMPILE_STACK_TOP.fixup_alt_jump
3385 ? bufp->buffer + COMPILE_STACK_TOP.fixup_alt_jump - 1
3386 : 0;
3387 laststart = bufp->buffer + COMPILE_STACK_TOP.laststart_offset;
3388 regnum = COMPILE_STACK_TOP.regnum;
3389 /* If we've reached MAX_REGNUM groups, then this open
3390 won't actually generate any code, so we'll have to
3391 clear pending_exact explicitly. */
3392 pending_exact = 0;
3393
3394 /* We're at the end of the group, so now we know how many
3395 groups were inside this one. */
3396 if (regnum <= MAX_REGNUM && regnum > 0)
3397 BUF_PUSH_2 (stop_memory, regnum);
3398 }
3399 break;
3400
3401
3402 case '|': /* `\|'. */
3403 if (syntax & RE_LIMITED_OPS || syntax & RE_NO_BK_VBAR)
3404 goto normal_backslash;
3405 handle_alt:
3406 if (syntax & RE_LIMITED_OPS)
3407 goto normal_char;
3408
3409 /* Insert before the previous alternative a jump which
3410 jumps to this alternative if the former fails. */
3411 GET_BUFFER_SPACE (3);
3412 INSERT_JUMP (on_failure_jump, begalt, b + 6);
3413 pending_exact = 0;
3414 b += 3;
3415
3416 /* The alternative before this one has a jump after it
3417 which gets executed if it gets matched. Adjust that
3418 jump so it will jump to this alternative's analogous
3419 jump (put in below, which in turn will jump to the next
3420 (if any) alternative's such jump, etc.). The last such
3421 jump jumps to the correct final destination. A picture:
3422 _____ _____
3423 | | | |
3424 | v | v
3425 a | b | c
3426
3427 If we are at `b', then fixup_alt_jump right now points to a
3428 three-byte space after `a'. We'll put in the jump, set
3429 fixup_alt_jump to right after `b', and leave behind three
3430 bytes which we'll fill in when we get to after `c'. */
3431
3432 FIXUP_ALT_JUMP ();
3433
3434 /* Mark and leave space for a jump after this alternative,
3435 to be filled in later either by next alternative or
3436 when know we're at the end of a series of alternatives. */
3437 fixup_alt_jump = b;
3438 GET_BUFFER_SPACE (3);
3439 b += 3;
3440
3441 laststart = 0;
3442 begalt = b;
3443 break;
3444
3445
3446 case '{':
3447 /* If \{ is a literal. */
3448 if (!(syntax & RE_INTERVALS)
3449 /* If we're at `\{' and it's not the open-interval
3450 operator. */
3451 || (syntax & RE_NO_BK_BRACES))
3452 goto normal_backslash;
3453
3454 handle_interval:
3455 {
3456 /* If got here, then the syntax allows intervals. */
3457
3458 /* At least (most) this many matches must be made. */
3459 int lower_bound = 0, upper_bound = -1;
3460
3461 beg_interval = p;
3462
3463 GET_UNSIGNED_NUMBER (lower_bound);
3464
3465 if (c == ',')
3466 GET_UNSIGNED_NUMBER (upper_bound);
3467 else
3468 /* Interval such as `{1}' => match exactly once. */
3469 upper_bound = lower_bound;
3470
3471 if (lower_bound < 0 || upper_bound > RE_DUP_MAX
3472 || (upper_bound >= 0 && lower_bound > upper_bound))
3473 FREE_STACK_RETURN (REG_BADBR);
3474
3475 if (!(syntax & RE_NO_BK_BRACES))
3476 {
3477 if (c != '\\')
3478 FREE_STACK_RETURN (REG_BADBR);
3479 if (p == pend)
3480 FREE_STACK_RETURN (REG_EESCAPE);
3481 PATFETCH (c);
3482 }
3483
3484 if (c != '}')
3485 FREE_STACK_RETURN (REG_BADBR);
3486
3487 /* We just parsed a valid interval. */
3488
3489 /* If it's invalid to have no preceding re. */
3490 if (!laststart)
3491 {
3492 if (syntax & RE_CONTEXT_INVALID_OPS)
3493 FREE_STACK_RETURN (REG_BADRPT);
3494 else if (syntax & RE_CONTEXT_INDEP_OPS)
3495 laststart = b;
3496 else
3497 goto unfetch_interval;
3498 }
3499
3500 if (upper_bound == 0)
3501 /* If the upper bound is zero, just drop the sub pattern
3502 altogether. */
3503 b = laststart;
3504 else if (lower_bound == 1 && upper_bound == 1)
3505 /* Just match it once: nothing to do here. */
3506 ;
3507
3508 /* Otherwise, we have a nontrivial interval. When
3509 we're all done, the pattern will look like:
3510 set_number_at <jump count> <upper bound>
3511 set_number_at <succeed_n count> <lower bound>
3512 succeed_n <after jump addr> <succeed_n count>
3513 <body of loop>
3514 jump_n <succeed_n addr> <jump count>
3515 (The upper bound and `jump_n' are omitted if
3516 `upper_bound' is 1, though.) */
3517 else
3518 { /* If the upper bound is > 1, we need to insert
3519 more at the end of the loop. */
3520 unsigned int nbytes = (upper_bound < 0 ? 3
3521 : upper_bound > 1 ? 5 : 0);
3522 unsigned int startoffset = 0;
3523
3524 GET_BUFFER_SPACE (20); /* We might use less. */
3525
3526 if (lower_bound == 0)
3527 {
3528 /* A succeed_n that starts with 0 is really a
3529 a simple on_failure_jump_loop. */
3530 INSERT_JUMP (on_failure_jump_loop, laststart,
3531 b + 3 + nbytes);
3532 b += 3;
3533 }
3534 else
3535 {
3536 /* Initialize lower bound of the `succeed_n', even
3537 though it will be set during matching by its
3538 attendant `set_number_at' (inserted next),
3539 because `re_compile_fastmap' needs to know.
3540 Jump to the `jump_n' we might insert below. */
3541 INSERT_JUMP2 (succeed_n, laststart,
3542 b + 5 + nbytes,
3543 lower_bound);
3544 b += 5;
3545
3546 /* Code to initialize the lower bound. Insert
3547 before the `succeed_n'. The `5' is the last two
3548 bytes of this `set_number_at', plus 3 bytes of
3549 the following `succeed_n'. */
3550 insert_op2 (set_number_at, laststart, 5, lower_bound, b);
3551 b += 5;
3552 startoffset += 5;
3553 }
3554
3555 if (upper_bound < 0)
3556 {
3557 /* A negative upper bound stands for infinity,
3558 in which case it degenerates to a plain jump. */
3559 STORE_JUMP (jump, b, laststart + startoffset);
3560 b += 3;
3561 }
3562 else if (upper_bound > 1)
3563 { /* More than one repetition is allowed, so
3564 append a backward jump to the `succeed_n'
3565 that starts this interval.
3566
3567 When we've reached this during matching,
3568 we'll have matched the interval once, so
3569 jump back only `upper_bound - 1' times. */
3570 STORE_JUMP2 (jump_n, b, laststart + startoffset,
3571 upper_bound - 1);
3572 b += 5;
3573
3574 /* The location we want to set is the second
3575 parameter of the `jump_n'; that is `b-2' as
3576 an absolute address. `laststart' will be
3577 the `set_number_at' we're about to insert;
3578 `laststart+3' the number to set, the source
3579 for the relative address. But we are
3580 inserting into the middle of the pattern --
3581 so everything is getting moved up by 5.
3582 Conclusion: (b - 2) - (laststart + 3) + 5,
3583 i.e., b - laststart.
3584
3585 We insert this at the beginning of the loop
3586 so that if we fail during matching, we'll
3587 reinitialize the bounds. */
3588 insert_op2 (set_number_at, laststart, b - laststart,
3589 upper_bound - 1, b);
3590 b += 5;
3591 }
3592 }
3593 pending_exact = 0;
3594 beg_interval = NULL;
3595 }
3596 break;
3597
3598 unfetch_interval:
3599 /* If an invalid interval, match the characters as literals. */
3600 assert (beg_interval);
3601 p = beg_interval;
3602 beg_interval = NULL;
3603
3604 /* normal_char and normal_backslash need `c'. */
3605 c = '{';
3606
3607 if (!(syntax & RE_NO_BK_BRACES))
3608 {
3609 assert (p > pattern && p[-1] == '\\');
3610 goto normal_backslash;
3611 }
3612 else
3613 goto normal_char;
3614
3615 #ifdef emacs
3616 /* There is no way to specify the before_dot and after_dot
3617 operators. rms says this is ok. --karl */
3618 case '=':
3619 BUF_PUSH (at_dot);
3620 break;
3621
3622 case 's':
3623 laststart = b;
3624 PATFETCH (c);
3625 BUF_PUSH_2 (syntaxspec, syntax_spec_code[c]);
3626 break;
3627
3628 case 'S':
3629 laststart = b;
3630 PATFETCH (c);
3631 BUF_PUSH_2 (notsyntaxspec, syntax_spec_code[c]);
3632 break;
3633
3634 case 'c':
3635 laststart = b;
3636 PATFETCH (c);
3637 BUF_PUSH_2 (categoryspec, c);
3638 break;
3639
3640 case 'C':
3641 laststart = b;
3642 PATFETCH (c);
3643 BUF_PUSH_2 (notcategoryspec, c);
3644 break;
3645 #endif /* emacs */
3646
3647
3648 case 'w':
3649 if (syntax & RE_NO_GNU_OPS)
3650 goto normal_char;
3651 laststart = b;
3652 BUF_PUSH_2 (syntaxspec, Sword);
3653 break;
3654
3655
3656 case 'W':
3657 if (syntax & RE_NO_GNU_OPS)
3658 goto normal_char;
3659 laststart = b;
3660 BUF_PUSH_2 (notsyntaxspec, Sword);
3661 break;
3662
3663
3664 case '<':
3665 if (syntax & RE_NO_GNU_OPS)
3666 goto normal_char;
3667 BUF_PUSH (wordbeg);
3668 break;
3669
3670 case '>':
3671 if (syntax & RE_NO_GNU_OPS)
3672 goto normal_char;
3673 BUF_PUSH (wordend);
3674 break;
3675
3676 case '_':
3677 if (syntax & RE_NO_GNU_OPS)
3678 goto normal_char;
3679 laststart = b;
3680 PATFETCH (c);
3681 if (c == '<')
3682 BUF_PUSH (symbeg);
3683 else if (c == '>')
3684 BUF_PUSH (symend);
3685 else
3686 FREE_STACK_RETURN (REG_BADPAT);
3687 break;
3688
3689 case 'b':
3690 if (syntax & RE_NO_GNU_OPS)
3691 goto normal_char;
3692 BUF_PUSH (wordbound);
3693 break;
3694
3695 case 'B':
3696 if (syntax & RE_NO_GNU_OPS)
3697 goto normal_char;
3698 BUF_PUSH (notwordbound);
3699 break;
3700
3701 case '`':
3702 if (syntax & RE_NO_GNU_OPS)
3703 goto normal_char;
3704 BUF_PUSH (begbuf);
3705 break;
3706
3707 case '\'':
3708 if (syntax & RE_NO_GNU_OPS)
3709 goto normal_char;
3710 BUF_PUSH (endbuf);
3711 break;
3712
3713 case '1': case '2': case '3': case '4': case '5':
3714 case '6': case '7': case '8': case '9':
3715 {
3716 regnum_t reg;
3717
3718 if (syntax & RE_NO_BK_REFS)
3719 goto normal_backslash;
3720
3721 reg = c - '0';
3722
3723 if (reg > bufp->re_nsub || reg < 1
3724 /* Can't back reference to a subexp before its end. */
3725 || group_in_compile_stack (compile_stack, reg))
3726 FREE_STACK_RETURN (REG_ESUBREG);
3727
3728 laststart = b;
3729 BUF_PUSH_2 (duplicate, reg);
3730 }
3731 break;
3732
3733
3734 case '+':
3735 case '?':
3736 if (syntax & RE_BK_PLUS_QM)
3737 goto handle_plus;
3738 else
3739 goto normal_backslash;
3740
3741 default:
3742 normal_backslash:
3743 /* You might think it would be useful for \ to mean
3744 not to translate; but if we don't translate it
3745 it will never match anything. */
3746 goto normal_char;
3747 }
3748 break;
3749
3750
3751 default:
3752 /* Expects the character in `c'. */
3753 normal_char:
3754 /* If no exactn currently being built. */
3755 if (!pending_exact
3756
3757 /* If last exactn not at current position. */
3758 || pending_exact + *pending_exact + 1 != b
3759
3760 /* We have only one byte following the exactn for the count. */
3761 || *pending_exact >= (1 << BYTEWIDTH) - MAX_MULTIBYTE_LENGTH
3762
3763 /* If followed by a repetition operator. */
3764 || (p != pend && (*p == '*' || *p == '^'))
3765 || ((syntax & RE_BK_PLUS_QM)
3766 ? p + 1 < pend && *p == '\\' && (p[1] == '+' || p[1] == '?')
3767 : p != pend && (*p == '+' || *p == '?'))
3768 || ((syntax & RE_INTERVALS)
3769 && ((syntax & RE_NO_BK_BRACES)
3770 ? p != pend && *p == '{'
3771 : p + 1 < pend && p[0] == '\\' && p[1] == '{')))
3772 {
3773 /* Start building a new exactn. */
3774
3775 laststart = b;
3776
3777 BUF_PUSH_2 (exactn, 0);
3778 pending_exact = b - 1;
3779 }
3780
3781 GET_BUFFER_SPACE (MAX_MULTIBYTE_LENGTH);
3782 {
3783 int len;
3784
3785 if (multibyte)
3786 {
3787 c = TRANSLATE (c);
3788 len = CHAR_STRING (c, b);
3789 b += len;
3790 }
3791 else
3792 {
3793 c1 = RE_CHAR_TO_MULTIBYTE (c);
3794 if (! CHAR_BYTE8_P (c1))
3795 {
3796 re_wchar_t c2 = TRANSLATE (c1);
3797
3798 if (c1 != c2 && (c1 = RE_CHAR_TO_UNIBYTE (c2)) >= 0)
3799 c = c1;
3800 }
3801 *b++ = c;
3802 len = 1;
3803 }
3804 (*pending_exact) += len;
3805 }
3806
3807 break;
3808 } /* switch (c) */
3809 } /* while p != pend */
3810
3811
3812 /* Through the pattern now. */
3813
3814 FIXUP_ALT_JUMP ();
3815
3816 if (!COMPILE_STACK_EMPTY)
3817 FREE_STACK_RETURN (REG_EPAREN);
3818
3819 /* If we don't want backtracking, force success
3820 the first time we reach the end of the compiled pattern. */
3821 if (syntax & RE_NO_POSIX_BACKTRACKING)
3822 BUF_PUSH (succeed);
3823
3824 /* We have succeeded; set the length of the buffer. */
3825 bufp->used = b - bufp->buffer;
3826
3827 #ifdef DEBUG
3828 if (debug > 0)
3829 {
3830 re_compile_fastmap (bufp);
3831 DEBUG_PRINT1 ("\nCompiled pattern: \n");
3832 print_compiled_pattern (bufp);
3833 }
3834 debug--;
3835 #endif /* DEBUG */
3836
3837 #ifndef MATCH_MAY_ALLOCATE
3838 /* Initialize the failure stack to the largest possible stack. This
3839 isn't necessary unless we're trying to avoid calling alloca in
3840 the search and match routines. */
3841 {
3842 int num_regs = bufp->re_nsub + 1;
3843
3844 if (fail_stack.size < re_max_failures * TYPICAL_FAILURE_SIZE)
3845 {
3846 fail_stack.size = re_max_failures * TYPICAL_FAILURE_SIZE;
3847
3848 if (! fail_stack.stack)
3849 fail_stack.stack
3850 = (fail_stack_elt_t *) malloc (fail_stack.size
3851 * sizeof (fail_stack_elt_t));
3852 else
3853 fail_stack.stack
3854 = (fail_stack_elt_t *) realloc (fail_stack.stack,
3855 (fail_stack.size
3856 * sizeof (fail_stack_elt_t)));
3857 }
3858
3859 regex_grow_registers (num_regs);
3860 }
3861 #endif /* not MATCH_MAY_ALLOCATE */
3862
3863 FREE_STACK_RETURN (REG_NOERROR);
3864 } /* regex_compile */
3865 \f
3866 /* Subroutines for `regex_compile'. */
3867
3868 /* Store OP at LOC followed by two-byte integer parameter ARG. */
3869
3870 static void
3871 store_op1 (op, loc, arg)
3872 re_opcode_t op;
3873 unsigned char *loc;
3874 int arg;
3875 {
3876 *loc = (unsigned char) op;
3877 STORE_NUMBER (loc + 1, arg);
3878 }
3879
3880
3881 /* Like `store_op1', but for two two-byte parameters ARG1 and ARG2. */
3882
3883 static void
3884 store_op2 (op, loc, arg1, arg2)
3885 re_opcode_t op;
3886 unsigned char *loc;
3887 int arg1, arg2;
3888 {
3889 *loc = (unsigned char) op;
3890 STORE_NUMBER (loc + 1, arg1);
3891 STORE_NUMBER (loc + 3, arg2);
3892 }
3893
3894
3895 /* Copy the bytes from LOC to END to open up three bytes of space at LOC
3896 for OP followed by two-byte integer parameter ARG. */
3897
3898 static void
3899 insert_op1 (op, loc, arg, end)
3900 re_opcode_t op;
3901 unsigned char *loc;
3902 int arg;
3903 unsigned char *end;
3904 {
3905 register unsigned char *pfrom = end;
3906 register unsigned char *pto = end + 3;
3907
3908 while (pfrom != loc)
3909 *--pto = *--pfrom;
3910
3911 store_op1 (op, loc, arg);
3912 }
3913
3914
3915 /* Like `insert_op1', but for two two-byte parameters ARG1 and ARG2. */
3916
3917 static void
3918 insert_op2 (op, loc, arg1, arg2, end)
3919 re_opcode_t op;
3920 unsigned char *loc;
3921 int arg1, arg2;
3922 unsigned char *end;
3923 {
3924 register unsigned char *pfrom = end;
3925 register unsigned char *pto = end + 5;
3926
3927 while (pfrom != loc)
3928 *--pto = *--pfrom;
3929
3930 store_op2 (op, loc, arg1, arg2);
3931 }
3932
3933
3934 /* P points to just after a ^ in PATTERN. Return true if that ^ comes
3935 after an alternative or a begin-subexpression. We assume there is at
3936 least one character before the ^. */
3937
3938 static boolean
3939 at_begline_loc_p (pattern, p, syntax)
3940 re_char *pattern, *p;
3941 reg_syntax_t syntax;
3942 {
3943 re_char *prev = p - 2;
3944 boolean prev_prev_backslash = prev > pattern && prev[-1] == '\\';
3945
3946 return
3947 /* After a subexpression? */
3948 (*prev == '(' && (syntax & RE_NO_BK_PARENS || prev_prev_backslash))
3949 /* After an alternative? */
3950 || (*prev == '|' && (syntax & RE_NO_BK_VBAR || prev_prev_backslash))
3951 /* After a shy subexpression? */
3952 || ((syntax & RE_SHY_GROUPS) && prev - 2 >= pattern
3953 && prev[-1] == '?' && prev[-2] == '('
3954 && (syntax & RE_NO_BK_PARENS
3955 || (prev - 3 >= pattern && prev[-3] == '\\')));
3956 }
3957
3958
3959 /* The dual of at_begline_loc_p. This one is for $. We assume there is
3960 at least one character after the $, i.e., `P < PEND'. */
3961
3962 static boolean
3963 at_endline_loc_p (p, pend, syntax)
3964 re_char *p, *pend;
3965 reg_syntax_t syntax;
3966 {
3967 re_char *next = p;
3968 boolean next_backslash = *next == '\\';
3969 re_char *next_next = p + 1 < pend ? p + 1 : 0;
3970
3971 return
3972 /* Before a subexpression? */
3973 (syntax & RE_NO_BK_PARENS ? *next == ')'
3974 : next_backslash && next_next && *next_next == ')')
3975 /* Before an alternative? */
3976 || (syntax & RE_NO_BK_VBAR ? *next == '|'
3977 : next_backslash && next_next && *next_next == '|');
3978 }
3979
3980
3981 /* Returns true if REGNUM is in one of COMPILE_STACK's elements and
3982 false if it's not. */
3983
3984 static boolean
3985 group_in_compile_stack (compile_stack, regnum)
3986 compile_stack_type compile_stack;
3987 regnum_t regnum;
3988 {
3989 int this_element;
3990
3991 for (this_element = compile_stack.avail - 1;
3992 this_element >= 0;
3993 this_element--)
3994 if (compile_stack.stack[this_element].regnum == regnum)
3995 return true;
3996
3997 return false;
3998 }
3999 \f
4000 /* analyse_first.
4001 If fastmap is non-NULL, go through the pattern and fill fastmap
4002 with all the possible leading chars. If fastmap is NULL, don't
4003 bother filling it up (obviously) and only return whether the
4004 pattern could potentially match the empty string.
4005
4006 Return 1 if p..pend might match the empty string.
4007 Return 0 if p..pend matches at least one char.
4008 Return -1 if fastmap was not updated accurately. */
4009
4010 static int
4011 analyse_first (p, pend, fastmap, multibyte)
4012 re_char *p, *pend;
4013 char *fastmap;
4014 const int multibyte;
4015 {
4016 int j, k;
4017 boolean not;
4018
4019 /* If all elements for base leading-codes in fastmap is set, this
4020 flag is set true. */
4021 boolean match_any_multibyte_characters = false;
4022
4023 assert (p);
4024
4025 /* The loop below works as follows:
4026 - It has a working-list kept in the PATTERN_STACK and which basically
4027 starts by only containing a pointer to the first operation.
4028 - If the opcode we're looking at is a match against some set of
4029 chars, then we add those chars to the fastmap and go on to the
4030 next work element from the worklist (done via `break').
4031 - If the opcode is a control operator on the other hand, we either
4032 ignore it (if it's meaningless at this point, such as `start_memory')
4033 or execute it (if it's a jump). If the jump has several destinations
4034 (i.e. `on_failure_jump'), then we push the other destination onto the
4035 worklist.
4036 We guarantee termination by ignoring backward jumps (more or less),
4037 so that `p' is monotonically increasing. More to the point, we
4038 never set `p' (or push) anything `<= p1'. */
4039
4040 while (p < pend)
4041 {
4042 /* `p1' is used as a marker of how far back a `on_failure_jump'
4043 can go without being ignored. It is normally equal to `p'
4044 (which prevents any backward `on_failure_jump') except right
4045 after a plain `jump', to allow patterns such as:
4046 0: jump 10
4047 3..9: <body>
4048 10: on_failure_jump 3
4049 as used for the *? operator. */
4050 re_char *p1 = p;
4051
4052 switch (SWITCH_ENUM_CAST ((re_opcode_t) *p++))
4053 {
4054 case succeed:
4055 return 1;
4056 continue;
4057
4058 case duplicate:
4059 /* If the first character has to match a backreference, that means
4060 that the group was empty (since it already matched). Since this
4061 is the only case that interests us here, we can assume that the
4062 backreference must match the empty string. */
4063 p++;
4064 continue;
4065
4066
4067 /* Following are the cases which match a character. These end
4068 with `break'. */
4069
4070 case exactn:
4071 if (fastmap)
4072 {
4073 /* If multibyte is nonzero, the first byte of each
4074 character is an ASCII or a leading code. Otherwise,
4075 each byte is a character. Thus, this works in both
4076 cases. */
4077 fastmap[p[1]] = 1;
4078 if (! multibyte)
4079 {
4080 /* For the case of matching this unibyte regex
4081 against multibyte, we must set a leading code of
4082 the corresponding multibyte character. */
4083 int c = RE_CHAR_TO_MULTIBYTE (p[1]);
4084
4085 if (! CHAR_BYTE8_P (c))
4086 fastmap[CHAR_LEADING_CODE (c)] = 1;
4087 }
4088 }
4089 break;
4090
4091
4092 case anychar:
4093 /* We could put all the chars except for \n (and maybe \0)
4094 but we don't bother since it is generally not worth it. */
4095 if (!fastmap) break;
4096 return -1;
4097
4098
4099 case charset_not:
4100 if (!fastmap) break;
4101 {
4102 /* Chars beyond end of bitmap are possible matches. */
4103 for (j = CHARSET_BITMAP_SIZE (&p[-1]) * BYTEWIDTH;
4104 j < (1 << BYTEWIDTH); j++)
4105 fastmap[j] = 1;
4106 }
4107
4108 /* Fallthrough */
4109 case charset:
4110 if (!fastmap) break;
4111 not = (re_opcode_t) *(p - 1) == charset_not;
4112 for (j = CHARSET_BITMAP_SIZE (&p[-1]) * BYTEWIDTH - 1, p++;
4113 j >= 0; j--)
4114 if (!!(p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH))) ^ not)
4115 fastmap[j] = 1;
4116
4117 #ifdef emacs
4118 if (/* Any leading code can possibly start a character
4119 which doesn't match the specified set of characters. */
4120 not
4121 ||
4122 /* If we can match a character class, we can match any
4123 multibyte characters. */
4124 (CHARSET_RANGE_TABLE_EXISTS_P (&p[-2])
4125 && CHARSET_RANGE_TABLE_BITS (&p[-2]) != 0))
4126
4127 {
4128 if (match_any_multibyte_characters == false)
4129 {
4130 for (j = MIN_MULTIBYTE_LEADING_CODE;
4131 j <= MAX_MULTIBYTE_LEADING_CODE; j++)
4132 fastmap[j] = 1;
4133 match_any_multibyte_characters = true;
4134 }
4135 }
4136
4137 else if (!not && CHARSET_RANGE_TABLE_EXISTS_P (&p[-2])
4138 && match_any_multibyte_characters == false)
4139 {
4140 /* Set fastmap[I] to 1 where I is a leading code of each
4141 multibyte characer in the range table. */
4142 int c, count;
4143 unsigned char lc1, lc2;
4144
4145 /* Make P points the range table. `+ 2' is to skip flag
4146 bits for a character class. */
4147 p += CHARSET_BITMAP_SIZE (&p[-2]) + 2;
4148
4149 /* Extract the number of ranges in range table into COUNT. */
4150 EXTRACT_NUMBER_AND_INCR (count, p);
4151 for (; count > 0; count--, p += 3)
4152 {
4153 /* Extract the start and end of each range. */
4154 EXTRACT_CHARACTER (c, p);
4155 lc1 = CHAR_LEADING_CODE (c);
4156 p += 3;
4157 EXTRACT_CHARACTER (c, p);
4158 lc2 = CHAR_LEADING_CODE (c);
4159 for (j = lc1; j <= lc2; j++)
4160 fastmap[j] = 1;
4161 }
4162 }
4163 #endif
4164 break;
4165
4166 case syntaxspec:
4167 case notsyntaxspec:
4168 if (!fastmap) break;
4169 #ifndef emacs
4170 not = (re_opcode_t)p[-1] == notsyntaxspec;
4171 k = *p++;
4172 for (j = 0; j < (1 << BYTEWIDTH); j++)
4173 if ((SYNTAX (j) == (enum syntaxcode) k) ^ not)
4174 fastmap[j] = 1;
4175 break;
4176 #else /* emacs */
4177 /* This match depends on text properties. These end with
4178 aborting optimizations. */
4179 return -1;
4180
4181 case categoryspec:
4182 case notcategoryspec:
4183 if (!fastmap) break;
4184 not = (re_opcode_t)p[-1] == notcategoryspec;
4185 k = *p++;
4186 for (j = (1 << BYTEWIDTH); j >= 0; j--)
4187 if ((CHAR_HAS_CATEGORY (j, k)) ^ not)
4188 fastmap[j] = 1;
4189
4190 /* Any leading code can possibly start a character which
4191 has or doesn't has the specified category. */
4192 if (match_any_multibyte_characters == false)
4193 {
4194 for (j = MIN_MULTIBYTE_LEADING_CODE;
4195 j <= MAX_MULTIBYTE_LEADING_CODE; j++)
4196 fastmap[j] = 1;
4197 match_any_multibyte_characters = true;
4198 }
4199 break;
4200
4201 /* All cases after this match the empty string. These end with
4202 `continue'. */
4203
4204 case before_dot:
4205 case at_dot:
4206 case after_dot:
4207 #endif /* !emacs */
4208 case no_op:
4209 case begline:
4210 case endline:
4211 case begbuf:
4212 case endbuf:
4213 case wordbound:
4214 case notwordbound:
4215 case wordbeg:
4216 case wordend:
4217 case symbeg:
4218 case symend:
4219 continue;
4220
4221
4222 case jump:
4223 EXTRACT_NUMBER_AND_INCR (j, p);
4224 if (j < 0)
4225 /* Backward jumps can only go back to code that we've already
4226 visited. `re_compile' should make sure this is true. */
4227 break;
4228 p += j;
4229 switch (SWITCH_ENUM_CAST ((re_opcode_t) *p))
4230 {
4231 case on_failure_jump:
4232 case on_failure_keep_string_jump:
4233 case on_failure_jump_loop:
4234 case on_failure_jump_nastyloop:
4235 case on_failure_jump_smart:
4236 p++;
4237 break;
4238 default:
4239 continue;
4240 };
4241 /* Keep `p1' to allow the `on_failure_jump' we are jumping to
4242 to jump back to "just after here". */
4243 /* Fallthrough */
4244
4245 case on_failure_jump:
4246 case on_failure_keep_string_jump:
4247 case on_failure_jump_nastyloop:
4248 case on_failure_jump_loop:
4249 case on_failure_jump_smart:
4250 EXTRACT_NUMBER_AND_INCR (j, p);
4251 if (p + j <= p1)
4252 ; /* Backward jump to be ignored. */
4253 else
4254 { /* We have to look down both arms.
4255 We first go down the "straight" path so as to minimize
4256 stack usage when going through alternatives. */
4257 int r = analyse_first (p, pend, fastmap, multibyte);
4258 if (r) return r;
4259 p += j;
4260 }
4261 continue;
4262
4263
4264 case jump_n:
4265 /* This code simply does not properly handle forward jump_n. */
4266 DEBUG_STATEMENT (EXTRACT_NUMBER (j, p); assert (j < 0));
4267 p += 4;
4268 /* jump_n can either jump or fall through. The (backward) jump
4269 case has already been handled, so we only need to look at the
4270 fallthrough case. */
4271 continue;
4272
4273 case succeed_n:
4274 /* If N == 0, it should be an on_failure_jump_loop instead. */
4275 DEBUG_STATEMENT (EXTRACT_NUMBER (j, p + 2); assert (j > 0));
4276 p += 4;
4277 /* We only care about one iteration of the loop, so we don't
4278 need to consider the case where this behaves like an
4279 on_failure_jump. */
4280 continue;
4281
4282
4283 case set_number_at:
4284 p += 4;
4285 continue;
4286
4287
4288 case start_memory:
4289 case stop_memory:
4290 p += 1;
4291 continue;
4292
4293
4294 default:
4295 abort (); /* We have listed all the cases. */
4296 } /* switch *p++ */
4297
4298 /* Getting here means we have found the possible starting
4299 characters for one path of the pattern -- and that the empty
4300 string does not match. We need not follow this path further. */
4301 return 0;
4302 } /* while p */
4303
4304 /* We reached the end without matching anything. */
4305 return 1;
4306
4307 } /* analyse_first */
4308 \f
4309 /* re_compile_fastmap computes a ``fastmap'' for the compiled pattern in
4310 BUFP. A fastmap records which of the (1 << BYTEWIDTH) possible
4311 characters can start a string that matches the pattern. This fastmap
4312 is used by re_search to skip quickly over impossible starting points.
4313
4314 Character codes above (1 << BYTEWIDTH) are not represented in the
4315 fastmap, but the leading codes are represented. Thus, the fastmap
4316 indicates which character sets could start a match.
4317
4318 The caller must supply the address of a (1 << BYTEWIDTH)-byte data
4319 area as BUFP->fastmap.
4320
4321 We set the `fastmap', `fastmap_accurate', and `can_be_null' fields in
4322 the pattern buffer.
4323
4324 Returns 0 if we succeed, -2 if an internal error. */
4325
4326 int
4327 re_compile_fastmap (bufp)
4328 struct re_pattern_buffer *bufp;
4329 {
4330 char *fastmap = bufp->fastmap;
4331 int analysis;
4332
4333 assert (fastmap && bufp->buffer);
4334
4335 bzero (fastmap, 1 << BYTEWIDTH); /* Assume nothing's valid. */
4336 bufp->fastmap_accurate = 1; /* It will be when we're done. */
4337
4338 analysis = analyse_first (bufp->buffer, bufp->buffer + bufp->used,
4339 fastmap, RE_MULTIBYTE_P (bufp));
4340 bufp->can_be_null = (analysis != 0);
4341 return 0;
4342 } /* re_compile_fastmap */
4343 \f
4344 /* Set REGS to hold NUM_REGS registers, storing them in STARTS and
4345 ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use
4346 this memory for recording register information. STARTS and ENDS
4347 must be allocated using the malloc library routine, and must each
4348 be at least NUM_REGS * sizeof (regoff_t) bytes long.
4349
4350 If NUM_REGS == 0, then subsequent matches should allocate their own
4351 register data.
4352
4353 Unless this function is called, the first search or match using
4354 PATTERN_BUFFER will allocate its own register data, without
4355 freeing the old data. */
4356
4357 void
4358 re_set_registers (bufp, regs, num_regs, starts, ends)
4359 struct re_pattern_buffer *bufp;
4360 struct re_registers *regs;
4361 unsigned num_regs;
4362 regoff_t *starts, *ends;
4363 {
4364 if (num_regs)
4365 {
4366 bufp->regs_allocated = REGS_REALLOCATE;
4367 regs->num_regs = num_regs;
4368 regs->start = starts;
4369 regs->end = ends;
4370 }
4371 else
4372 {
4373 bufp->regs_allocated = REGS_UNALLOCATED;
4374 regs->num_regs = 0;
4375 regs->start = regs->end = (regoff_t *) 0;
4376 }
4377 }
4378 WEAK_ALIAS (__re_set_registers, re_set_registers)
4379 \f
4380 /* Searching routines. */
4381
4382 /* Like re_search_2, below, but only one string is specified, and
4383 doesn't let you say where to stop matching. */
4384
4385 int
4386 re_search (bufp, string, size, startpos, range, regs)
4387 struct re_pattern_buffer *bufp;
4388 const char *string;
4389 int size, startpos, range;
4390 struct re_registers *regs;
4391 {
4392 return re_search_2 (bufp, NULL, 0, string, size, startpos, range,
4393 regs, size);
4394 }
4395 WEAK_ALIAS (__re_search, re_search)
4396
4397 /* Head address of virtual concatenation of string. */
4398 #define HEAD_ADDR_VSTRING(P) \
4399 (((P) >= size1 ? string2 : string1))
4400
4401 /* End address of virtual concatenation of string. */
4402 #define STOP_ADDR_VSTRING(P) \
4403 (((P) >= size1 ? string2 + size2 : string1 + size1))
4404
4405 /* Address of POS in the concatenation of virtual string. */
4406 #define POS_ADDR_VSTRING(POS) \
4407 (((POS) >= size1 ? string2 - size1 : string1) + (POS))
4408
4409 /* Using the compiled pattern in BUFP->buffer, first tries to match the
4410 virtual concatenation of STRING1 and STRING2, starting first at index
4411 STARTPOS, then at STARTPOS + 1, and so on.
4412
4413 STRING1 and STRING2 have length SIZE1 and SIZE2, respectively.
4414
4415 RANGE is how far to scan while trying to match. RANGE = 0 means try
4416 only at STARTPOS; in general, the last start tried is STARTPOS +
4417 RANGE.
4418
4419 In REGS, return the indices of the virtual concatenation of STRING1
4420 and STRING2 that matched the entire BUFP->buffer and its contained
4421 subexpressions.
4422
4423 Do not consider matching one past the index STOP in the virtual
4424 concatenation of STRING1 and STRING2.
4425
4426 We return either the position in the strings at which the match was
4427 found, -1 if no match, or -2 if error (such as failure
4428 stack overflow). */
4429
4430 int
4431 re_search_2 (bufp, str1, size1, str2, size2, startpos, range, regs, stop)
4432 struct re_pattern_buffer *bufp;
4433 const char *str1, *str2;
4434 int size1, size2;
4435 int startpos;
4436 int range;
4437 struct re_registers *regs;
4438 int stop;
4439 {
4440 int val;
4441 re_char *string1 = (re_char*) str1;
4442 re_char *string2 = (re_char*) str2;
4443 register char *fastmap = bufp->fastmap;
4444 register RE_TRANSLATE_TYPE translate = bufp->translate;
4445 int total_size = size1 + size2;
4446 int endpos = startpos + range;
4447 boolean anchored_start;
4448 /* Nonzero if we are searching multibyte string. */
4449 const boolean multibyte = RE_TARGET_MULTIBYTE_P (bufp);
4450
4451 /* Check for out-of-range STARTPOS. */
4452 if (startpos < 0 || startpos > total_size)
4453 return -1;
4454
4455 /* Fix up RANGE if it might eventually take us outside
4456 the virtual concatenation of STRING1 and STRING2.
4457 Make sure we won't move STARTPOS below 0 or above TOTAL_SIZE. */
4458 if (endpos < 0)
4459 range = 0 - startpos;
4460 else if (endpos > total_size)
4461 range = total_size - startpos;
4462
4463 /* If the search isn't to be a backwards one, don't waste time in a
4464 search for a pattern anchored at beginning of buffer. */
4465 if (bufp->used > 0 && (re_opcode_t) bufp->buffer[0] == begbuf && range > 0)
4466 {
4467 if (startpos > 0)
4468 return -1;
4469 else
4470 range = 0;
4471 }
4472
4473 #ifdef emacs
4474 /* In a forward search for something that starts with \=.
4475 don't keep searching past point. */
4476 if (bufp->used > 0 && (re_opcode_t) bufp->buffer[0] == at_dot && range > 0)
4477 {
4478 range = PT_BYTE - BEGV_BYTE - startpos;
4479 if (range < 0)
4480 return -1;
4481 }
4482 #endif /* emacs */
4483
4484 /* Update the fastmap now if not correct already. */
4485 if (fastmap && !bufp->fastmap_accurate)
4486 re_compile_fastmap (bufp);
4487
4488 /* See whether the pattern is anchored. */
4489 anchored_start = (bufp->buffer[0] == begline);
4490
4491 #ifdef emacs
4492 gl_state.object = re_match_object;
4493 {
4494 int charpos = SYNTAX_TABLE_BYTE_TO_CHAR (POS_AS_IN_BUFFER (startpos));
4495
4496 SETUP_SYNTAX_TABLE_FOR_OBJECT (re_match_object, charpos, 1);
4497 }
4498 #endif
4499
4500 /* Loop through the string, looking for a place to start matching. */
4501 for (;;)
4502 {
4503 /* If the pattern is anchored,
4504 skip quickly past places we cannot match.
4505 We don't bother to treat startpos == 0 specially
4506 because that case doesn't repeat. */
4507 if (anchored_start && startpos > 0)
4508 {
4509 if (! ((startpos <= size1 ? string1[startpos - 1]
4510 : string2[startpos - size1 - 1])
4511 == '\n'))
4512 goto advance;
4513 }
4514
4515 /* If a fastmap is supplied, skip quickly over characters that
4516 cannot be the start of a match. If the pattern can match the
4517 null string, however, we don't need to skip characters; we want
4518 the first null string. */
4519 if (fastmap && startpos < total_size && !bufp->can_be_null)
4520 {
4521 register re_char *d;
4522 register re_wchar_t buf_ch;
4523
4524 d = POS_ADDR_VSTRING (startpos);
4525
4526 if (range > 0) /* Searching forwards. */
4527 {
4528 register int lim = 0;
4529 int irange = range;
4530
4531 if (startpos < size1 && startpos + range >= size1)
4532 lim = range - (size1 - startpos);
4533
4534 /* Written out as an if-else to avoid testing `translate'
4535 inside the loop. */
4536 if (RE_TRANSLATE_P (translate))
4537 {
4538 if (multibyte)
4539 while (range > lim)
4540 {
4541 int buf_charlen;
4542
4543 buf_ch = STRING_CHAR_AND_LENGTH (d, range - lim,
4544 buf_charlen);
4545 buf_ch = RE_TRANSLATE (translate, buf_ch);
4546 if (fastmap[CHAR_LEADING_CODE (buf_ch)])
4547 break;
4548
4549 range -= buf_charlen;
4550 d += buf_charlen;
4551 }
4552 else
4553 while (range > lim)
4554 {
4555 register re_wchar_t ch, translated;
4556
4557 buf_ch = *d;
4558 ch = RE_CHAR_TO_MULTIBYTE (buf_ch);
4559 translated = RE_TRANSLATE (translate, ch);
4560 if (translated != ch
4561 && (ch = RE_CHAR_TO_UNIBYTE (translated)) >= 0)
4562 buf_ch = ch;
4563 if (fastmap[buf_ch])
4564 break;
4565 d++;
4566 range--;
4567 }
4568 }
4569 else
4570 {
4571 if (multibyte)
4572 while (range > lim)
4573 {
4574 int buf_charlen;
4575
4576 buf_ch = STRING_CHAR_AND_LENGTH (d, range - lim,
4577 buf_charlen);
4578 if (fastmap[CHAR_LEADING_CODE (buf_ch)])
4579 break;
4580 range -= buf_charlen;
4581 d += buf_charlen;
4582 }
4583 else
4584 while (range > lim && !fastmap[*d])
4585 {
4586 d++;
4587 range--;
4588 }
4589 }
4590 startpos += irange - range;
4591 }
4592 else /* Searching backwards. */
4593 {
4594 int room = (startpos >= size1
4595 ? size2 + size1 - startpos
4596 : size1 - startpos);
4597 if (multibyte)
4598 {
4599 buf_ch = STRING_CHAR (d, room);
4600 buf_ch = TRANSLATE (buf_ch);
4601 if (! fastmap[CHAR_LEADING_CODE (buf_ch)])
4602 goto advance;
4603 }
4604 else
4605 {
4606 register re_wchar_t ch, translated;
4607
4608 buf_ch = *d;
4609 ch = RE_CHAR_TO_MULTIBYTE (buf_ch);
4610 translated = TRANSLATE (ch);
4611 if (translated != ch
4612 && (ch = RE_CHAR_TO_UNIBYTE (translated)) >= 0)
4613 buf_ch = ch;
4614 if (! fastmap[TRANSLATE (buf_ch)])
4615 goto advance;
4616 }
4617 }
4618 }
4619
4620 /* If can't match the null string, and that's all we have left, fail. */
4621 if (range >= 0 && startpos == total_size && fastmap
4622 && !bufp->can_be_null)
4623 return -1;
4624
4625 val = re_match_2_internal (bufp, string1, size1, string2, size2,
4626 startpos, regs, stop);
4627
4628 if (val >= 0)
4629 return startpos;
4630
4631 if (val == -2)
4632 return -2;
4633
4634 advance:
4635 if (!range)
4636 break;
4637 else if (range > 0)
4638 {
4639 /* Update STARTPOS to the next character boundary. */
4640 if (multibyte)
4641 {
4642 re_char *p = POS_ADDR_VSTRING (startpos);
4643 re_char *pend = STOP_ADDR_VSTRING (startpos);
4644 int len = MULTIBYTE_FORM_LENGTH (p, pend - p);
4645
4646 range -= len;
4647 if (range < 0)
4648 break;
4649 startpos += len;
4650 }
4651 else
4652 {
4653 range--;
4654 startpos++;
4655 }
4656 }
4657 else
4658 {
4659 range++;
4660 startpos--;
4661
4662 /* Update STARTPOS to the previous character boundary. */
4663 if (multibyte)
4664 {
4665 re_char *p = POS_ADDR_VSTRING (startpos) + 1;
4666 re_char *p0 = p;
4667 re_char *phead = HEAD_ADDR_VSTRING (startpos);
4668
4669 /* Find the head of multibyte form. */
4670 PREV_CHAR_BOUNDARY (p, phead);
4671 range += p0 - 1 - p;
4672 if (range > 0)
4673 break;
4674
4675 startpos -= p0 - 1 - p;
4676 }
4677 }
4678 }
4679 return -1;
4680 } /* re_search_2 */
4681 WEAK_ALIAS (__re_search_2, re_search_2)
4682 \f
4683 /* Declarations and macros for re_match_2. */
4684
4685 static int bcmp_translate _RE_ARGS((re_char *s1, re_char *s2,
4686 register int len,
4687 RE_TRANSLATE_TYPE translate,
4688 const int multibyte));
4689
4690 /* This converts PTR, a pointer into one of the search strings `string1'
4691 and `string2' into an offset from the beginning of that string. */
4692 #define POINTER_TO_OFFSET(ptr) \
4693 (FIRST_STRING_P (ptr) \
4694 ? ((regoff_t) ((ptr) - string1)) \
4695 : ((regoff_t) ((ptr) - string2 + size1)))
4696
4697 /* Call before fetching a character with *d. This switches over to
4698 string2 if necessary.
4699 Check re_match_2_internal for a discussion of why end_match_2 might
4700 not be within string2 (but be equal to end_match_1 instead). */
4701 #define PREFETCH() \
4702 while (d == dend) \
4703 { \
4704 /* End of string2 => fail. */ \
4705 if (dend == end_match_2) \
4706 goto fail; \
4707 /* End of string1 => advance to string2. */ \
4708 d = string2; \
4709 dend = end_match_2; \
4710 }
4711
4712 /* Call before fetching a char with *d if you already checked other limits.
4713 This is meant for use in lookahead operations like wordend, etc..
4714 where we might need to look at parts of the string that might be
4715 outside of the LIMITs (i.e past `stop'). */
4716 #define PREFETCH_NOLIMIT() \
4717 if (d == end1) \
4718 { \
4719 d = string2; \
4720 dend = end_match_2; \
4721 } \
4722
4723 /* Test if at very beginning or at very end of the virtual concatenation
4724 of `string1' and `string2'. If only one string, it's `string2'. */
4725 #define AT_STRINGS_BEG(d) ((d) == (size1 ? string1 : string2) || !size2)
4726 #define AT_STRINGS_END(d) ((d) == end2)
4727
4728
4729 /* Test if D points to a character which is word-constituent. We have
4730 two special cases to check for: if past the end of string1, look at
4731 the first character in string2; and if before the beginning of
4732 string2, look at the last character in string1. */
4733 #define WORDCHAR_P(d) \
4734 (SYNTAX ((d) == end1 ? *string2 \
4735 : (d) == string2 - 1 ? *(end1 - 1) : *(d)) \
4736 == Sword)
4737
4738 /* Disabled due to a compiler bug -- see comment at case wordbound */
4739
4740 /* The comment at case wordbound is following one, but we don't use
4741 AT_WORD_BOUNDARY anymore to support multibyte form.
4742
4743 The DEC Alpha C compiler 3.x generates incorrect code for the
4744 test WORDCHAR_P (d - 1) != WORDCHAR_P (d) in the expansion of
4745 AT_WORD_BOUNDARY, so this code is disabled. Expanding the
4746 macro and introducing temporary variables works around the bug. */
4747
4748 #if 0
4749 /* Test if the character before D and the one at D differ with respect
4750 to being word-constituent. */
4751 #define AT_WORD_BOUNDARY(d) \
4752 (AT_STRINGS_BEG (d) || AT_STRINGS_END (d) \
4753 || WORDCHAR_P (d - 1) != WORDCHAR_P (d))
4754 #endif
4755
4756 /* Free everything we malloc. */
4757 #ifdef MATCH_MAY_ALLOCATE
4758 # define FREE_VAR(var) if (var) { REGEX_FREE (var); var = NULL; } else
4759 # define FREE_VARIABLES() \
4760 do { \
4761 REGEX_FREE_STACK (fail_stack.stack); \
4762 FREE_VAR (regstart); \
4763 FREE_VAR (regend); \
4764 FREE_VAR (best_regstart); \
4765 FREE_VAR (best_regend); \
4766 } while (0)
4767 #else
4768 # define FREE_VARIABLES() ((void)0) /* Do nothing! But inhibit gcc warning. */
4769 #endif /* not MATCH_MAY_ALLOCATE */
4770
4771 \f
4772 /* Optimization routines. */
4773
4774 /* If the operation is a match against one or more chars,
4775 return a pointer to the next operation, else return NULL. */
4776 static re_char *
4777 skip_one_char (p)
4778 re_char *p;
4779 {
4780 switch (SWITCH_ENUM_CAST (*p++))
4781 {
4782 case anychar:
4783 break;
4784
4785 case exactn:
4786 p += *p + 1;
4787 break;
4788
4789 case charset_not:
4790 case charset:
4791 if (CHARSET_RANGE_TABLE_EXISTS_P (p - 1))
4792 {
4793 int mcnt;
4794 p = CHARSET_RANGE_TABLE (p - 1);
4795 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4796 p = CHARSET_RANGE_TABLE_END (p, mcnt);
4797 }
4798 else
4799 p += 1 + CHARSET_BITMAP_SIZE (p - 1);
4800 break;
4801
4802 case syntaxspec:
4803 case notsyntaxspec:
4804 #ifdef emacs
4805 case categoryspec:
4806 case notcategoryspec:
4807 #endif /* emacs */
4808 p++;
4809 break;
4810
4811 default:
4812 p = NULL;
4813 }
4814 return p;
4815 }
4816
4817
4818 /* Jump over non-matching operations. */
4819 static re_char *
4820 skip_noops (p, pend)
4821 re_char *p, *pend;
4822 {
4823 int mcnt;
4824 while (p < pend)
4825 {
4826 switch (SWITCH_ENUM_CAST ((re_opcode_t) *p))
4827 {
4828 case start_memory:
4829 case stop_memory:
4830 p += 2; break;
4831 case no_op:
4832 p += 1; break;
4833 case jump:
4834 p += 1;
4835 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4836 p += mcnt;
4837 break;
4838 default:
4839 return p;
4840 }
4841 }
4842 assert (p == pend);
4843 return p;
4844 }
4845
4846 /* Non-zero if "p1 matches something" implies "p2 fails". */
4847 static int
4848 mutually_exclusive_p (bufp, p1, p2)
4849 struct re_pattern_buffer *bufp;
4850 re_char *p1, *p2;
4851 {
4852 re_opcode_t op2;
4853 const boolean multibyte = RE_MULTIBYTE_P (bufp);
4854 unsigned char *pend = bufp->buffer + bufp->used;
4855
4856 assert (p1 >= bufp->buffer && p1 < pend
4857 && p2 >= bufp->buffer && p2 <= pend);
4858
4859 /* Skip over open/close-group commands.
4860 If what follows this loop is a ...+ construct,
4861 look at what begins its body, since we will have to
4862 match at least one of that. */
4863 p2 = skip_noops (p2, pend);
4864 /* The same skip can be done for p1, except that this function
4865 is only used in the case where p1 is a simple match operator. */
4866 /* p1 = skip_noops (p1, pend); */
4867
4868 assert (p1 >= bufp->buffer && p1 < pend
4869 && p2 >= bufp->buffer && p2 <= pend);
4870
4871 op2 = p2 == pend ? succeed : *p2;
4872
4873 switch (SWITCH_ENUM_CAST (op2))
4874 {
4875 case succeed:
4876 case endbuf:
4877 /* If we're at the end of the pattern, we can change. */
4878 if (skip_one_char (p1))
4879 {
4880 DEBUG_PRINT1 (" End of pattern: fast loop.\n");
4881 return 1;
4882 }
4883 break;
4884
4885 case endline:
4886 case exactn:
4887 {
4888 register re_wchar_t c
4889 = (re_opcode_t) *p2 == endline ? '\n'
4890 : RE_STRING_CHAR (p2 + 2, pend - p2 - 2, multibyte);
4891
4892 if ((re_opcode_t) *p1 == exactn)
4893 {
4894 if (c != RE_STRING_CHAR (p1 + 2, pend - p1 - 2, multibyte))
4895 {
4896 DEBUG_PRINT3 (" '%c' != '%c' => fast loop.\n", c, p1[2]);
4897 return 1;
4898 }
4899 }
4900
4901 else if ((re_opcode_t) *p1 == charset
4902 || (re_opcode_t) *p1 == charset_not)
4903 {
4904 int not = (re_opcode_t) *p1 == charset_not;
4905
4906 /* Test if C is listed in charset (or charset_not)
4907 at `p1'. */
4908 if (! multibyte || IS_REAL_ASCII (c))
4909 {
4910 if (c < CHARSET_BITMAP_SIZE (p1) * BYTEWIDTH
4911 && p1[2 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH)))
4912 not = !not;
4913 }
4914 else if (CHARSET_RANGE_TABLE_EXISTS_P (p1))
4915 CHARSET_LOOKUP_RANGE_TABLE (not, c, p1);
4916
4917 /* `not' is equal to 1 if c would match, which means
4918 that we can't change to pop_failure_jump. */
4919 if (!not)
4920 {
4921 DEBUG_PRINT1 (" No match => fast loop.\n");
4922 return 1;
4923 }
4924 }
4925 else if ((re_opcode_t) *p1 == anychar
4926 && c == '\n')
4927 {
4928 DEBUG_PRINT1 (" . != \\n => fast loop.\n");
4929 return 1;
4930 }
4931 }
4932 break;
4933
4934 case charset:
4935 {
4936 if ((re_opcode_t) *p1 == exactn)
4937 /* Reuse the code above. */
4938 return mutually_exclusive_p (bufp, p2, p1);
4939
4940 /* It is hard to list up all the character in charset
4941 P2 if it includes multibyte character. Give up in
4942 such case. */
4943 else if (!multibyte || !CHARSET_RANGE_TABLE_EXISTS_P (p2))
4944 {
4945 /* Now, we are sure that P2 has no range table.
4946 So, for the size of bitmap in P2, `p2[1]' is
4947 enough. But P1 may have range table, so the
4948 size of bitmap table of P1 is extracted by
4949 using macro `CHARSET_BITMAP_SIZE'.
4950
4951 In a multibyte case, we know that all the character
4952 listed in P2 is ASCII. In a unibyte case, P1 has only a
4953 bitmap table. So, in both cases, it is enough to test
4954 only the bitmap table of P1. */
4955
4956 if ((re_opcode_t) *p1 == charset)
4957 {
4958 int idx;
4959 /* We win if the charset inside the loop
4960 has no overlap with the one after the loop. */
4961 for (idx = 0;
4962 (idx < (int) p2[1]
4963 && idx < CHARSET_BITMAP_SIZE (p1));
4964 idx++)
4965 if ((p2[2 + idx] & p1[2 + idx]) != 0)
4966 break;
4967
4968 if (idx == p2[1]
4969 || idx == CHARSET_BITMAP_SIZE (p1))
4970 {
4971 DEBUG_PRINT1 (" No match => fast loop.\n");
4972 return 1;
4973 }
4974 }
4975 else if ((re_opcode_t) *p1 == charset_not)
4976 {
4977 int idx;
4978 /* We win if the charset_not inside the loop lists
4979 every character listed in the charset after. */
4980 for (idx = 0; idx < (int) p2[1]; idx++)
4981 if (! (p2[2 + idx] == 0
4982 || (idx < CHARSET_BITMAP_SIZE (p1)
4983 && ((p2[2 + idx] & ~ p1[2 + idx]) == 0))))
4984 break;
4985
4986 if (idx == p2[1])
4987 {
4988 DEBUG_PRINT1 (" No match => fast loop.\n");
4989 return 1;
4990 }
4991 }
4992 }
4993 }
4994 break;
4995
4996 case charset_not:
4997 switch (SWITCH_ENUM_CAST (*p1))
4998 {
4999 case exactn:
5000 case charset:
5001 /* Reuse the code above. */
5002 return mutually_exclusive_p (bufp, p2, p1);
5003 case charset_not:
5004 /* When we have two charset_not, it's very unlikely that
5005 they don't overlap. The union of the two sets of excluded
5006 chars should cover all possible chars, which, as a matter of
5007 fact, is virtually impossible in multibyte buffers. */
5008 break;
5009 }
5010 break;
5011
5012 case wordend:
5013 return ((re_opcode_t) *p1 == syntaxspec && p1[1] == Sword);
5014 case symend:
5015 return ((re_opcode_t) *p1 == syntaxspec
5016 && (p1[1] == Ssymbol || p1[1] == Sword));
5017 case notsyntaxspec:
5018 return ((re_opcode_t) *p1 == syntaxspec && p1[1] == p2[1]);
5019
5020 case wordbeg:
5021 return ((re_opcode_t) *p1 == notsyntaxspec && p1[1] == Sword);
5022 case symbeg:
5023 return ((re_opcode_t) *p1 == notsyntaxspec
5024 && (p1[1] == Ssymbol || p1[1] == Sword));
5025 case syntaxspec:
5026 return ((re_opcode_t) *p1 == notsyntaxspec && p1[1] == p2[1]);
5027
5028 case wordbound:
5029 return (((re_opcode_t) *p1 == notsyntaxspec
5030 || (re_opcode_t) *p1 == syntaxspec)
5031 && p1[1] == Sword);
5032
5033 #ifdef emacs
5034 case categoryspec:
5035 return ((re_opcode_t) *p1 == notcategoryspec && p1[1] == p2[1]);
5036 case notcategoryspec:
5037 return ((re_opcode_t) *p1 == categoryspec && p1[1] == p2[1]);
5038 #endif /* emacs */
5039
5040 default:
5041 ;
5042 }
5043
5044 /* Safe default. */
5045 return 0;
5046 }
5047
5048 \f
5049 /* Matching routines. */
5050
5051 #ifndef emacs /* Emacs never uses this. */
5052 /* re_match is like re_match_2 except it takes only a single string. */
5053
5054 int
5055 re_match (bufp, string, size, pos, regs)
5056 struct re_pattern_buffer *bufp;
5057 const char *string;
5058 int size, pos;
5059 struct re_registers *regs;
5060 {
5061 int result = re_match_2_internal (bufp, NULL, 0, (re_char*) string, size,
5062 pos, regs, size);
5063 return result;
5064 }
5065 WEAK_ALIAS (__re_match, re_match)
5066 #endif /* not emacs */
5067
5068 #ifdef emacs
5069 /* In Emacs, this is the string or buffer in which we
5070 are matching. It is used for looking up syntax properties. */
5071 Lisp_Object re_match_object;
5072 #endif
5073
5074 /* re_match_2 matches the compiled pattern in BUFP against the
5075 the (virtual) concatenation of STRING1 and STRING2 (of length SIZE1
5076 and SIZE2, respectively). We start matching at POS, and stop
5077 matching at STOP.
5078
5079 If REGS is non-null and the `no_sub' field of BUFP is nonzero, we
5080 store offsets for the substring each group matched in REGS. See the
5081 documentation for exactly how many groups we fill.
5082
5083 We return -1 if no match, -2 if an internal error (such as the
5084 failure stack overflowing). Otherwise, we return the length of the
5085 matched substring. */
5086
5087 int
5088 re_match_2 (bufp, string1, size1, string2, size2, pos, regs, stop)
5089 struct re_pattern_buffer *bufp;
5090 const char *string1, *string2;
5091 int size1, size2;
5092 int pos;
5093 struct re_registers *regs;
5094 int stop;
5095 {
5096 int result;
5097
5098 #ifdef emacs
5099 int charpos;
5100 gl_state.object = re_match_object;
5101 charpos = SYNTAX_TABLE_BYTE_TO_CHAR (POS_AS_IN_BUFFER (pos));
5102 SETUP_SYNTAX_TABLE_FOR_OBJECT (re_match_object, charpos, 1);
5103 #endif
5104
5105 result = re_match_2_internal (bufp, (re_char*) string1, size1,
5106 (re_char*) string2, size2,
5107 pos, regs, stop);
5108 return result;
5109 }
5110 WEAK_ALIAS (__re_match_2, re_match_2)
5111
5112
5113 /* This is a separate function so that we can force an alloca cleanup
5114 afterwards. */
5115 static int
5116 re_match_2_internal (bufp, string1, size1, string2, size2, pos, regs, stop)
5117 struct re_pattern_buffer *bufp;
5118 re_char *string1, *string2;
5119 int size1, size2;
5120 int pos;
5121 struct re_registers *regs;
5122 int stop;
5123 {
5124 /* General temporaries. */
5125 int mcnt;
5126 size_t reg;
5127 boolean not;
5128
5129 /* Just past the end of the corresponding string. */
5130 re_char *end1, *end2;
5131
5132 /* Pointers into string1 and string2, just past the last characters in
5133 each to consider matching. */
5134 re_char *end_match_1, *end_match_2;
5135
5136 /* Where we are in the data, and the end of the current string. */
5137 re_char *d, *dend;
5138
5139 /* Used sometimes to remember where we were before starting matching
5140 an operator so that we can go back in case of failure. This "atomic"
5141 behavior of matching opcodes is indispensable to the correctness
5142 of the on_failure_keep_string_jump optimization. */
5143 re_char *dfail;
5144
5145 /* Where we are in the pattern, and the end of the pattern. */
5146 re_char *p = bufp->buffer;
5147 re_char *pend = p + bufp->used;
5148
5149 /* We use this to map every character in the string. */
5150 RE_TRANSLATE_TYPE translate = bufp->translate;
5151
5152 /* Nonzero if BUFP is setup from a multibyte regex. */
5153 const boolean multibyte = RE_MULTIBYTE_P (bufp);
5154
5155 /* Nonzero if STRING1/STRING2 are multibyte. */
5156 const boolean target_multibyte = RE_TARGET_MULTIBYTE_P (bufp);
5157
5158 /* Failure point stack. Each place that can handle a failure further
5159 down the line pushes a failure point on this stack. It consists of
5160 regstart, and regend for all registers corresponding to
5161 the subexpressions we're currently inside, plus the number of such
5162 registers, and, finally, two char *'s. The first char * is where
5163 to resume scanning the pattern; the second one is where to resume
5164 scanning the strings. */
5165 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
5166 fail_stack_type fail_stack;
5167 #endif
5168 #ifdef DEBUG
5169 unsigned nfailure_points_pushed = 0, nfailure_points_popped = 0;
5170 #endif
5171
5172 #if defined REL_ALLOC && defined REGEX_MALLOC
5173 /* This holds the pointer to the failure stack, when
5174 it is allocated relocatably. */
5175 fail_stack_elt_t *failure_stack_ptr;
5176 #endif
5177
5178 /* We fill all the registers internally, independent of what we
5179 return, for use in backreferences. The number here includes
5180 an element for register zero. */
5181 size_t num_regs = bufp->re_nsub + 1;
5182
5183 /* Information on the contents of registers. These are pointers into
5184 the input strings; they record just what was matched (on this
5185 attempt) by a subexpression part of the pattern, that is, the
5186 regnum-th regstart pointer points to where in the pattern we began
5187 matching and the regnum-th regend points to right after where we
5188 stopped matching the regnum-th subexpression. (The zeroth register
5189 keeps track of what the whole pattern matches.) */
5190 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
5191 re_char **regstart, **regend;
5192 #endif
5193
5194 /* The following record the register info as found in the above
5195 variables when we find a match better than any we've seen before.
5196 This happens as we backtrack through the failure points, which in
5197 turn happens only if we have not yet matched the entire string. */
5198 unsigned best_regs_set = false;
5199 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
5200 re_char **best_regstart, **best_regend;
5201 #endif
5202
5203 /* Logically, this is `best_regend[0]'. But we don't want to have to
5204 allocate space for that if we're not allocating space for anything
5205 else (see below). Also, we never need info about register 0 for
5206 any of the other register vectors, and it seems rather a kludge to
5207 treat `best_regend' differently than the rest. So we keep track of
5208 the end of the best match so far in a separate variable. We
5209 initialize this to NULL so that when we backtrack the first time
5210 and need to test it, it's not garbage. */
5211 re_char *match_end = NULL;
5212
5213 #ifdef DEBUG
5214 /* Counts the total number of registers pushed. */
5215 unsigned num_regs_pushed = 0;
5216 #endif
5217
5218 DEBUG_PRINT1 ("\n\nEntering re_match_2.\n");
5219
5220 INIT_FAIL_STACK ();
5221
5222 #ifdef MATCH_MAY_ALLOCATE
5223 /* Do not bother to initialize all the register variables if there are
5224 no groups in the pattern, as it takes a fair amount of time. If
5225 there are groups, we include space for register 0 (the whole
5226 pattern), even though we never use it, since it simplifies the
5227 array indexing. We should fix this. */
5228 if (bufp->re_nsub)
5229 {
5230 regstart = REGEX_TALLOC (num_regs, re_char *);
5231 regend = REGEX_TALLOC (num_regs, re_char *);
5232 best_regstart = REGEX_TALLOC (num_regs, re_char *);
5233 best_regend = REGEX_TALLOC (num_regs, re_char *);
5234
5235 if (!(regstart && regend && best_regstart && best_regend))
5236 {
5237 FREE_VARIABLES ();
5238 return -2;
5239 }
5240 }
5241 else
5242 {
5243 /* We must initialize all our variables to NULL, so that
5244 `FREE_VARIABLES' doesn't try to free them. */
5245 regstart = regend = best_regstart = best_regend = NULL;
5246 }
5247 #endif /* MATCH_MAY_ALLOCATE */
5248
5249 /* The starting position is bogus. */
5250 if (pos < 0 || pos > size1 + size2)
5251 {
5252 FREE_VARIABLES ();
5253 return -1;
5254 }
5255
5256 /* Initialize subexpression text positions to -1 to mark ones that no
5257 start_memory/stop_memory has been seen for. Also initialize the
5258 register information struct. */
5259 for (reg = 1; reg < num_regs; reg++)
5260 regstart[reg] = regend[reg] = NULL;
5261
5262 /* We move `string1' into `string2' if the latter's empty -- but not if
5263 `string1' is null. */
5264 if (size2 == 0 && string1 != NULL)
5265 {
5266 string2 = string1;
5267 size2 = size1;
5268 string1 = 0;
5269 size1 = 0;
5270 }
5271 end1 = string1 + size1;
5272 end2 = string2 + size2;
5273
5274 /* `p' scans through the pattern as `d' scans through the data.
5275 `dend' is the end of the input string that `d' points within. `d'
5276 is advanced into the following input string whenever necessary, but
5277 this happens before fetching; therefore, at the beginning of the
5278 loop, `d' can be pointing at the end of a string, but it cannot
5279 equal `string2'. */
5280 if (pos >= size1)
5281 {
5282 /* Only match within string2. */
5283 d = string2 + pos - size1;
5284 dend = end_match_2 = string2 + stop - size1;
5285 end_match_1 = end1; /* Just to give it a value. */
5286 }
5287 else
5288 {
5289 if (stop < size1)
5290 {
5291 /* Only match within string1. */
5292 end_match_1 = string1 + stop;
5293 /* BEWARE!
5294 When we reach end_match_1, PREFETCH normally switches to string2.
5295 But in the present case, this means that just doing a PREFETCH
5296 makes us jump from `stop' to `gap' within the string.
5297 What we really want here is for the search to stop as
5298 soon as we hit end_match_1. That's why we set end_match_2
5299 to end_match_1 (since PREFETCH fails as soon as we hit
5300 end_match_2). */
5301 end_match_2 = end_match_1;
5302 }
5303 else
5304 { /* It's important to use this code when stop == size so that
5305 moving `d' from end1 to string2 will not prevent the d == dend
5306 check from catching the end of string. */
5307 end_match_1 = end1;
5308 end_match_2 = string2 + stop - size1;
5309 }
5310 d = string1 + pos;
5311 dend = end_match_1;
5312 }
5313
5314 DEBUG_PRINT1 ("The compiled pattern is: ");
5315 DEBUG_PRINT_COMPILED_PATTERN (bufp, p, pend);
5316 DEBUG_PRINT1 ("The string to match is: `");
5317 DEBUG_PRINT_DOUBLE_STRING (d, string1, size1, string2, size2);
5318 DEBUG_PRINT1 ("'\n");
5319
5320 /* This loops over pattern commands. It exits by returning from the
5321 function if the match is complete, or it drops through if the match
5322 fails at this starting point in the input data. */
5323 for (;;)
5324 {
5325 DEBUG_PRINT2 ("\n%p: ", p);
5326
5327 if (p == pend)
5328 { /* End of pattern means we might have succeeded. */
5329 DEBUG_PRINT1 ("end of pattern ... ");
5330
5331 /* If we haven't matched the entire string, and we want the
5332 longest match, try backtracking. */
5333 if (d != end_match_2)
5334 {
5335 /* 1 if this match ends in the same string (string1 or string2)
5336 as the best previous match. */
5337 boolean same_str_p = (FIRST_STRING_P (match_end)
5338 == FIRST_STRING_P (d));
5339 /* 1 if this match is the best seen so far. */
5340 boolean best_match_p;
5341
5342 /* AIX compiler got confused when this was combined
5343 with the previous declaration. */
5344 if (same_str_p)
5345 best_match_p = d > match_end;
5346 else
5347 best_match_p = !FIRST_STRING_P (d);
5348
5349 DEBUG_PRINT1 ("backtracking.\n");
5350
5351 if (!FAIL_STACK_EMPTY ())
5352 { /* More failure points to try. */
5353
5354 /* If exceeds best match so far, save it. */
5355 if (!best_regs_set || best_match_p)
5356 {
5357 best_regs_set = true;
5358 match_end = d;
5359
5360 DEBUG_PRINT1 ("\nSAVING match as best so far.\n");
5361
5362 for (reg = 1; reg < num_regs; reg++)
5363 {
5364 best_regstart[reg] = regstart[reg];
5365 best_regend[reg] = regend[reg];
5366 }
5367 }
5368 goto fail;
5369 }
5370
5371 /* If no failure points, don't restore garbage. And if
5372 last match is real best match, don't restore second
5373 best one. */
5374 else if (best_regs_set && !best_match_p)
5375 {
5376 restore_best_regs:
5377 /* Restore best match. It may happen that `dend ==
5378 end_match_1' while the restored d is in string2.
5379 For example, the pattern `x.*y.*z' against the
5380 strings `x-' and `y-z-', if the two strings are
5381 not consecutive in memory. */
5382 DEBUG_PRINT1 ("Restoring best registers.\n");
5383
5384 d = match_end;
5385 dend = ((d >= string1 && d <= end1)
5386 ? end_match_1 : end_match_2);
5387
5388 for (reg = 1; reg < num_regs; reg++)
5389 {
5390 regstart[reg] = best_regstart[reg];
5391 regend[reg] = best_regend[reg];
5392 }
5393 }
5394 } /* d != end_match_2 */
5395
5396 succeed_label:
5397 DEBUG_PRINT1 ("Accepting match.\n");
5398
5399 /* If caller wants register contents data back, do it. */
5400 if (regs && !bufp->no_sub)
5401 {
5402 /* Have the register data arrays been allocated? */
5403 if (bufp->regs_allocated == REGS_UNALLOCATED)
5404 { /* No. So allocate them with malloc. We need one
5405 extra element beyond `num_regs' for the `-1' marker
5406 GNU code uses. */
5407 regs->num_regs = MAX (RE_NREGS, num_regs + 1);
5408 regs->start = TALLOC (regs->num_regs, regoff_t);
5409 regs->end = TALLOC (regs->num_regs, regoff_t);
5410 if (regs->start == NULL || regs->end == NULL)
5411 {
5412 FREE_VARIABLES ();
5413 return -2;
5414 }
5415 bufp->regs_allocated = REGS_REALLOCATE;
5416 }
5417 else if (bufp->regs_allocated == REGS_REALLOCATE)
5418 { /* Yes. If we need more elements than were already
5419 allocated, reallocate them. If we need fewer, just
5420 leave it alone. */
5421 if (regs->num_regs < num_regs + 1)
5422 {
5423 regs->num_regs = num_regs + 1;
5424 RETALLOC (regs->start, regs->num_regs, regoff_t);
5425 RETALLOC (regs->end, regs->num_regs, regoff_t);
5426 if (regs->start == NULL || regs->end == NULL)
5427 {
5428 FREE_VARIABLES ();
5429 return -2;
5430 }
5431 }
5432 }
5433 else
5434 {
5435 /* These braces fend off a "empty body in an else-statement"
5436 warning under GCC when assert expands to nothing. */
5437 assert (bufp->regs_allocated == REGS_FIXED);
5438 }
5439
5440 /* Convert the pointer data in `regstart' and `regend' to
5441 indices. Register zero has to be set differently,
5442 since we haven't kept track of any info for it. */
5443 if (regs->num_regs > 0)
5444 {
5445 regs->start[0] = pos;
5446 regs->end[0] = POINTER_TO_OFFSET (d);
5447 }
5448
5449 /* Go through the first `min (num_regs, regs->num_regs)'
5450 registers, since that is all we initialized. */
5451 for (reg = 1; reg < MIN (num_regs, regs->num_regs); reg++)
5452 {
5453 if (REG_UNSET (regstart[reg]) || REG_UNSET (regend[reg]))
5454 regs->start[reg] = regs->end[reg] = -1;
5455 else
5456 {
5457 regs->start[reg]
5458 = (regoff_t) POINTER_TO_OFFSET (regstart[reg]);
5459 regs->end[reg]
5460 = (regoff_t) POINTER_TO_OFFSET (regend[reg]);
5461 }
5462 }
5463
5464 /* If the regs structure we return has more elements than
5465 were in the pattern, set the extra elements to -1. If
5466 we (re)allocated the registers, this is the case,
5467 because we always allocate enough to have at least one
5468 -1 at the end. */
5469 for (reg = num_regs; reg < regs->num_regs; reg++)
5470 regs->start[reg] = regs->end[reg] = -1;
5471 } /* regs && !bufp->no_sub */
5472
5473 DEBUG_PRINT4 ("%u failure points pushed, %u popped (%u remain).\n",
5474 nfailure_points_pushed, nfailure_points_popped,
5475 nfailure_points_pushed - nfailure_points_popped);
5476 DEBUG_PRINT2 ("%u registers pushed.\n", num_regs_pushed);
5477
5478 mcnt = POINTER_TO_OFFSET (d) - pos;
5479
5480 DEBUG_PRINT2 ("Returning %d from re_match_2.\n", mcnt);
5481
5482 FREE_VARIABLES ();
5483 return mcnt;
5484 }
5485
5486 /* Otherwise match next pattern command. */
5487 switch (SWITCH_ENUM_CAST ((re_opcode_t) *p++))
5488 {
5489 /* Ignore these. Used to ignore the n of succeed_n's which
5490 currently have n == 0. */
5491 case no_op:
5492 DEBUG_PRINT1 ("EXECUTING no_op.\n");
5493 break;
5494
5495 case succeed:
5496 DEBUG_PRINT1 ("EXECUTING succeed.\n");
5497 goto succeed_label;
5498
5499 /* Match the next n pattern characters exactly. The following
5500 byte in the pattern defines n, and the n bytes after that
5501 are the characters to match. */
5502 case exactn:
5503 mcnt = *p++;
5504 DEBUG_PRINT2 ("EXECUTING exactn %d.\n", mcnt);
5505
5506 /* Remember the start point to rollback upon failure. */
5507 dfail = d;
5508
5509 #ifndef emacs
5510 /* This is written out as an if-else so we don't waste time
5511 testing `translate' inside the loop. */
5512 if (RE_TRANSLATE_P (translate))
5513 do
5514 {
5515 PREFETCH ();
5516 if (RE_TRANSLATE (translate, *d) != *p++)
5517 {
5518 d = dfail;
5519 goto fail;
5520 }
5521 d++;
5522 }
5523 while (--mcnt);
5524 else
5525 do
5526 {
5527 PREFETCH ();
5528 if (*d++ != *p++)
5529 {
5530 d = dfail;
5531 goto fail;
5532 }
5533 }
5534 while (--mcnt);
5535 #else /* emacs */
5536 /* The cost of testing `translate' is comparatively small. */
5537 if (target_multibyte)
5538 do
5539 {
5540 int pat_charlen, buf_charlen;
5541 int pat_ch, buf_ch;
5542
5543 PREFETCH ();
5544 if (multibyte)
5545 pat_ch = STRING_CHAR_AND_LENGTH (p, pend - p, pat_charlen);
5546 else
5547 {
5548 pat_ch = RE_CHAR_TO_MULTIBYTE (*p);
5549 pat_charlen = 1;
5550 }
5551 buf_ch = STRING_CHAR_AND_LENGTH (d, dend - d, buf_charlen);
5552
5553 if (TRANSLATE (buf_ch) != pat_ch)
5554 {
5555 d = dfail;
5556 goto fail;
5557 }
5558
5559 p += pat_charlen;
5560 d += buf_charlen;
5561 mcnt -= pat_charlen;
5562 }
5563 while (mcnt > 0);
5564 else
5565 do
5566 {
5567 int pat_charlen, buf_charlen;
5568 int pat_ch, buf_ch;
5569
5570 PREFETCH ();
5571 if (multibyte)
5572 {
5573 pat_ch = STRING_CHAR_AND_LENGTH (p, pend - p, pat_charlen);
5574 pat_ch = RE_CHAR_TO_UNIBYTE (pat_ch);
5575 }
5576 else
5577 {
5578 pat_ch = *p;
5579 pat_charlen = 1;
5580 }
5581 buf_ch = RE_CHAR_TO_MULTIBYTE (*d);
5582 if (! CHAR_BYTE8_P (buf_ch))
5583 {
5584 buf_ch = TRANSLATE (buf_ch);
5585 buf_ch = RE_CHAR_TO_UNIBYTE (buf_ch);
5586 if (buf_ch < 0)
5587 buf_ch = *d;
5588 }
5589 else
5590 buf_ch = *d;
5591 if (buf_ch != pat_ch)
5592 {
5593 d = dfail;
5594 goto fail;
5595 }
5596 p += pat_charlen;
5597 d++;
5598 }
5599 while (--mcnt);
5600 #endif
5601 break;
5602
5603
5604 /* Match any character except possibly a newline or a null. */
5605 case anychar:
5606 {
5607 int buf_charlen;
5608 re_wchar_t buf_ch;
5609
5610 DEBUG_PRINT1 ("EXECUTING anychar.\n");
5611
5612 PREFETCH ();
5613 buf_ch = RE_STRING_CHAR_AND_LENGTH (d, dend - d, buf_charlen,
5614 target_multibyte);
5615 buf_ch = TRANSLATE (buf_ch);
5616
5617 if ((!(bufp->syntax & RE_DOT_NEWLINE)
5618 && buf_ch == '\n')
5619 || ((bufp->syntax & RE_DOT_NOT_NULL)
5620 && buf_ch == '\000'))
5621 goto fail;
5622
5623 DEBUG_PRINT2 (" Matched `%d'.\n", *d);
5624 d += buf_charlen;
5625 }
5626 break;
5627
5628
5629 case charset:
5630 case charset_not:
5631 {
5632 register unsigned int c;
5633 boolean not = (re_opcode_t) *(p - 1) == charset_not;
5634 int len;
5635
5636 /* Start of actual range_table, or end of bitmap if there is no
5637 range table. */
5638 re_char *range_table;
5639
5640 /* Nonzero if there is a range table. */
5641 int range_table_exists;
5642
5643 /* Number of ranges of range table. This is not included
5644 in the initial byte-length of the command. */
5645 int count = 0;
5646
5647 /* Whether matching against a unibyte character. */
5648 boolean unibyte_char = false;
5649
5650 DEBUG_PRINT2 ("EXECUTING charset%s.\n", not ? "_not" : "");
5651
5652 range_table_exists = CHARSET_RANGE_TABLE_EXISTS_P (&p[-1]);
5653
5654 if (range_table_exists)
5655 {
5656 range_table = CHARSET_RANGE_TABLE (&p[-1]); /* Past the bitmap. */
5657 EXTRACT_NUMBER_AND_INCR (count, range_table);
5658 }
5659
5660 PREFETCH ();
5661 c = RE_STRING_CHAR_AND_LENGTH (d, dend - d, len, target_multibyte);
5662 if (target_multibyte)
5663 {
5664 int c1;
5665
5666 c = TRANSLATE (c);
5667 c1 = RE_CHAR_TO_UNIBYTE (c);
5668 if (c1 >= 0)
5669 {
5670 unibyte_char = true;
5671 c = c1;
5672 }
5673 }
5674 else
5675 {
5676 int c1 = RE_CHAR_TO_MULTIBYTE (c);
5677
5678 if (! CHAR_BYTE8_P (c1))
5679 {
5680 c1 = TRANSLATE (c1);
5681 c1 = RE_CHAR_TO_UNIBYTE (c1);
5682 if (c1 >= 0)
5683 {
5684 unibyte_char = true;
5685 c = c1;
5686 }
5687 }
5688 else
5689 unibyte_char = true;
5690 }
5691
5692 if (unibyte_char && c < (1 << BYTEWIDTH))
5693 { /* Lookup bitmap. */
5694 /* Cast to `unsigned' instead of `unsigned char' in
5695 case the bit list is a full 32 bytes long. */
5696 if (c < (unsigned) (CHARSET_BITMAP_SIZE (&p[-1]) * BYTEWIDTH)
5697 && p[1 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH)))
5698 not = !not;
5699 }
5700 #ifdef emacs
5701 else if (range_table_exists)
5702 {
5703 int class_bits = CHARSET_RANGE_TABLE_BITS (&p[-1]);
5704
5705 if ( (class_bits & BIT_LOWER && ISLOWER (c))
5706 | (class_bits & BIT_MULTIBYTE)
5707 | (class_bits & BIT_PUNCT && ISPUNCT (c))
5708 | (class_bits & BIT_SPACE && ISSPACE (c))
5709 | (class_bits & BIT_UPPER && ISUPPER (c))
5710 | (class_bits & BIT_WORD && ISWORD (c)))
5711 not = !not;
5712 else
5713 CHARSET_LOOKUP_RANGE_TABLE_RAW (not, c, range_table, count);
5714 }
5715 #endif /* emacs */
5716
5717 if (range_table_exists)
5718 p = CHARSET_RANGE_TABLE_END (range_table, count);
5719 else
5720 p += CHARSET_BITMAP_SIZE (&p[-1]) + 1;
5721
5722 if (!not) goto fail;
5723
5724 d += len;
5725 break;
5726 }
5727
5728
5729 /* The beginning of a group is represented by start_memory.
5730 The argument is the register number. The text
5731 matched within the group is recorded (in the internal
5732 registers data structure) under the register number. */
5733 case start_memory:
5734 DEBUG_PRINT2 ("EXECUTING start_memory %d:\n", *p);
5735
5736 /* In case we need to undo this operation (via backtracking). */
5737 PUSH_FAILURE_REG ((unsigned int)*p);
5738
5739 regstart[*p] = d;
5740 regend[*p] = NULL; /* probably unnecessary. -sm */
5741 DEBUG_PRINT2 (" regstart: %d\n", POINTER_TO_OFFSET (regstart[*p]));
5742
5743 /* Move past the register number and inner group count. */
5744 p += 1;
5745 break;
5746
5747
5748 /* The stop_memory opcode represents the end of a group. Its
5749 argument is the same as start_memory's: the register number. */
5750 case stop_memory:
5751 DEBUG_PRINT2 ("EXECUTING stop_memory %d:\n", *p);
5752
5753 assert (!REG_UNSET (regstart[*p]));
5754 /* Strictly speaking, there should be code such as:
5755
5756 assert (REG_UNSET (regend[*p]));
5757 PUSH_FAILURE_REGSTOP ((unsigned int)*p);
5758
5759 But the only info to be pushed is regend[*p] and it is known to
5760 be UNSET, so there really isn't anything to push.
5761 Not pushing anything, on the other hand deprives us from the
5762 guarantee that regend[*p] is UNSET since undoing this operation
5763 will not reset its value properly. This is not important since
5764 the value will only be read on the next start_memory or at
5765 the very end and both events can only happen if this stop_memory
5766 is *not* undone. */
5767
5768 regend[*p] = d;
5769 DEBUG_PRINT2 (" regend: %d\n", POINTER_TO_OFFSET (regend[*p]));
5770
5771 /* Move past the register number and the inner group count. */
5772 p += 1;
5773 break;
5774
5775
5776 /* \<digit> has been turned into a `duplicate' command which is
5777 followed by the numeric value of <digit> as the register number. */
5778 case duplicate:
5779 {
5780 register re_char *d2, *dend2;
5781 int regno = *p++; /* Get which register to match against. */
5782 DEBUG_PRINT2 ("EXECUTING duplicate %d.\n", regno);
5783
5784 /* Can't back reference a group which we've never matched. */
5785 if (REG_UNSET (regstart[regno]) || REG_UNSET (regend[regno]))
5786 goto fail;
5787
5788 /* Where in input to try to start matching. */
5789 d2 = regstart[regno];
5790
5791 /* Remember the start point to rollback upon failure. */
5792 dfail = d;
5793
5794 /* Where to stop matching; if both the place to start and
5795 the place to stop matching are in the same string, then
5796 set to the place to stop, otherwise, for now have to use
5797 the end of the first string. */
5798
5799 dend2 = ((FIRST_STRING_P (regstart[regno])
5800 == FIRST_STRING_P (regend[regno]))
5801 ? regend[regno] : end_match_1);
5802 for (;;)
5803 {
5804 /* If necessary, advance to next segment in register
5805 contents. */
5806 while (d2 == dend2)
5807 {
5808 if (dend2 == end_match_2) break;
5809 if (dend2 == regend[regno]) break;
5810
5811 /* End of string1 => advance to string2. */
5812 d2 = string2;
5813 dend2 = regend[regno];
5814 }
5815 /* At end of register contents => success */
5816 if (d2 == dend2) break;
5817
5818 /* If necessary, advance to next segment in data. */
5819 PREFETCH ();
5820
5821 /* How many characters left in this segment to match. */
5822 mcnt = dend - d;
5823
5824 /* Want how many consecutive characters we can match in
5825 one shot, so, if necessary, adjust the count. */
5826 if (mcnt > dend2 - d2)
5827 mcnt = dend2 - d2;
5828
5829 /* Compare that many; failure if mismatch, else move
5830 past them. */
5831 if (RE_TRANSLATE_P (translate)
5832 ? bcmp_translate (d, d2, mcnt, translate, target_multibyte)
5833 : memcmp (d, d2, mcnt))
5834 {
5835 d = dfail;
5836 goto fail;
5837 }
5838 d += mcnt, d2 += mcnt;
5839 }
5840 }
5841 break;
5842
5843
5844 /* begline matches the empty string at the beginning of the string
5845 (unless `not_bol' is set in `bufp'), and after newlines. */
5846 case begline:
5847 DEBUG_PRINT1 ("EXECUTING begline.\n");
5848
5849 if (AT_STRINGS_BEG (d))
5850 {
5851 if (!bufp->not_bol) break;
5852 }
5853 else
5854 {
5855 unsigned c;
5856 GET_CHAR_BEFORE_2 (c, d, string1, end1, string2, end2);
5857 if (c == '\n')
5858 break;
5859 }
5860 /* In all other cases, we fail. */
5861 goto fail;
5862
5863
5864 /* endline is the dual of begline. */
5865 case endline:
5866 DEBUG_PRINT1 ("EXECUTING endline.\n");
5867
5868 if (AT_STRINGS_END (d))
5869 {
5870 if (!bufp->not_eol) break;
5871 }
5872 else
5873 {
5874 PREFETCH_NOLIMIT ();
5875 if (*d == '\n')
5876 break;
5877 }
5878 goto fail;
5879
5880
5881 /* Match at the very beginning of the data. */
5882 case begbuf:
5883 DEBUG_PRINT1 ("EXECUTING begbuf.\n");
5884 if (AT_STRINGS_BEG (d))
5885 break;
5886 goto fail;
5887
5888
5889 /* Match at the very end of the data. */
5890 case endbuf:
5891 DEBUG_PRINT1 ("EXECUTING endbuf.\n");
5892 if (AT_STRINGS_END (d))
5893 break;
5894 goto fail;
5895
5896
5897 /* on_failure_keep_string_jump is used to optimize `.*\n'. It
5898 pushes NULL as the value for the string on the stack. Then
5899 `POP_FAILURE_POINT' will keep the current value for the
5900 string, instead of restoring it. To see why, consider
5901 matching `foo\nbar' against `.*\n'. The .* matches the foo;
5902 then the . fails against the \n. But the next thing we want
5903 to do is match the \n against the \n; if we restored the
5904 string value, we would be back at the foo.
5905
5906 Because this is used only in specific cases, we don't need to
5907 check all the things that `on_failure_jump' does, to make
5908 sure the right things get saved on the stack. Hence we don't
5909 share its code. The only reason to push anything on the
5910 stack at all is that otherwise we would have to change
5911 `anychar's code to do something besides goto fail in this
5912 case; that seems worse than this. */
5913 case on_failure_keep_string_jump:
5914 EXTRACT_NUMBER_AND_INCR (mcnt, p);
5915 DEBUG_PRINT3 ("EXECUTING on_failure_keep_string_jump %d (to %p):\n",
5916 mcnt, p + mcnt);
5917
5918 PUSH_FAILURE_POINT (p - 3, NULL);
5919 break;
5920
5921 /* A nasty loop is introduced by the non-greedy *? and +?.
5922 With such loops, the stack only ever contains one failure point
5923 at a time, so that a plain on_failure_jump_loop kind of
5924 cycle detection cannot work. Worse yet, such a detection
5925 can not only fail to detect a cycle, but it can also wrongly
5926 detect a cycle (between different instantiations of the same
5927 loop).
5928 So the method used for those nasty loops is a little different:
5929 We use a special cycle-detection-stack-frame which is pushed
5930 when the on_failure_jump_nastyloop failure-point is *popped*.
5931 This special frame thus marks the beginning of one iteration
5932 through the loop and we can hence easily check right here
5933 whether something matched between the beginning and the end of
5934 the loop. */
5935 case on_failure_jump_nastyloop:
5936 EXTRACT_NUMBER_AND_INCR (mcnt, p);
5937 DEBUG_PRINT3 ("EXECUTING on_failure_jump_nastyloop %d (to %p):\n",
5938 mcnt, p + mcnt);
5939
5940 assert ((re_opcode_t)p[-4] == no_op);
5941 {
5942 int cycle = 0;
5943 CHECK_INFINITE_LOOP (p - 4, d);
5944 if (!cycle)
5945 /* If there's a cycle, just continue without pushing
5946 this failure point. The failure point is the "try again"
5947 option, which shouldn't be tried.
5948 We want (x?)*?y\1z to match both xxyz and xxyxz. */
5949 PUSH_FAILURE_POINT (p - 3, d);
5950 }
5951 break;
5952
5953 /* Simple loop detecting on_failure_jump: just check on the
5954 failure stack if the same spot was already hit earlier. */
5955 case on_failure_jump_loop:
5956 on_failure:
5957 EXTRACT_NUMBER_AND_INCR (mcnt, p);
5958 DEBUG_PRINT3 ("EXECUTING on_failure_jump_loop %d (to %p):\n",
5959 mcnt, p + mcnt);
5960 {
5961 int cycle = 0;
5962 CHECK_INFINITE_LOOP (p - 3, d);
5963 if (cycle)
5964 /* If there's a cycle, get out of the loop, as if the matching
5965 had failed. We used to just `goto fail' here, but that was
5966 aborting the search a bit too early: we want to keep the
5967 empty-loop-match and keep matching after the loop.
5968 We want (x?)*y\1z to match both xxyz and xxyxz. */
5969 p += mcnt;
5970 else
5971 PUSH_FAILURE_POINT (p - 3, d);
5972 }
5973 break;
5974
5975
5976 /* Uses of on_failure_jump:
5977
5978 Each alternative starts with an on_failure_jump that points
5979 to the beginning of the next alternative. Each alternative
5980 except the last ends with a jump that in effect jumps past
5981 the rest of the alternatives. (They really jump to the
5982 ending jump of the following alternative, because tensioning
5983 these jumps is a hassle.)
5984
5985 Repeats start with an on_failure_jump that points past both
5986 the repetition text and either the following jump or
5987 pop_failure_jump back to this on_failure_jump. */
5988 case on_failure_jump:
5989 EXTRACT_NUMBER_AND_INCR (mcnt, p);
5990 DEBUG_PRINT3 ("EXECUTING on_failure_jump %d (to %p):\n",
5991 mcnt, p + mcnt);
5992
5993 PUSH_FAILURE_POINT (p -3, d);
5994 break;
5995
5996 /* This operation is used for greedy *.
5997 Compare the beginning of the repeat with what in the
5998 pattern follows its end. If we can establish that there
5999 is nothing that they would both match, i.e., that we
6000 would have to backtrack because of (as in, e.g., `a*a')
6001 then we can use a non-backtracking loop based on
6002 on_failure_keep_string_jump instead of on_failure_jump. */
6003 case on_failure_jump_smart:
6004 EXTRACT_NUMBER_AND_INCR (mcnt, p);
6005 DEBUG_PRINT3 ("EXECUTING on_failure_jump_smart %d (to %p).\n",
6006 mcnt, p + mcnt);
6007 {
6008 re_char *p1 = p; /* Next operation. */
6009 /* Here, we discard `const', making re_match non-reentrant. */
6010 unsigned char *p2 = (unsigned char*) p + mcnt; /* Jump dest. */
6011 unsigned char *p3 = (unsigned char*) p - 3; /* opcode location. */
6012
6013 p -= 3; /* Reset so that we will re-execute the
6014 instruction once it's been changed. */
6015
6016 EXTRACT_NUMBER (mcnt, p2 - 2);
6017
6018 /* Ensure this is a indeed the trivial kind of loop
6019 we are expecting. */
6020 assert (skip_one_char (p1) == p2 - 3);
6021 assert ((re_opcode_t) p2[-3] == jump && p2 + mcnt == p);
6022 DEBUG_STATEMENT (debug += 2);
6023 if (mutually_exclusive_p (bufp, p1, p2))
6024 {
6025 /* Use a fast `on_failure_keep_string_jump' loop. */
6026 DEBUG_PRINT1 (" smart exclusive => fast loop.\n");
6027 *p3 = (unsigned char) on_failure_keep_string_jump;
6028 STORE_NUMBER (p2 - 2, mcnt + 3);
6029 }
6030 else
6031 {
6032 /* Default to a safe `on_failure_jump' loop. */
6033 DEBUG_PRINT1 (" smart default => slow loop.\n");
6034 *p3 = (unsigned char) on_failure_jump;
6035 }
6036 DEBUG_STATEMENT (debug -= 2);
6037 }
6038 break;
6039
6040 /* Unconditionally jump (without popping any failure points). */
6041 case jump:
6042 unconditional_jump:
6043 IMMEDIATE_QUIT_CHECK;
6044 EXTRACT_NUMBER_AND_INCR (mcnt, p); /* Get the amount to jump. */
6045 DEBUG_PRINT2 ("EXECUTING jump %d ", mcnt);
6046 p += mcnt; /* Do the jump. */
6047 DEBUG_PRINT2 ("(to %p).\n", p);
6048 break;
6049
6050
6051 /* Have to succeed matching what follows at least n times.
6052 After that, handle like `on_failure_jump'. */
6053 case succeed_n:
6054 /* Signedness doesn't matter since we only compare MCNT to 0. */
6055 EXTRACT_NUMBER (mcnt, p + 2);
6056 DEBUG_PRINT2 ("EXECUTING succeed_n %d.\n", mcnt);
6057
6058 /* Originally, mcnt is how many times we HAVE to succeed. */
6059 if (mcnt != 0)
6060 {
6061 /* Here, we discard `const', making re_match non-reentrant. */
6062 unsigned char *p2 = (unsigned char*) p + 2; /* counter loc. */
6063 mcnt--;
6064 p += 4;
6065 PUSH_NUMBER (p2, mcnt);
6066 }
6067 else
6068 /* The two bytes encoding mcnt == 0 are two no_op opcodes. */
6069 goto on_failure;
6070 break;
6071
6072 case jump_n:
6073 /* Signedness doesn't matter since we only compare MCNT to 0. */
6074 EXTRACT_NUMBER (mcnt, p + 2);
6075 DEBUG_PRINT2 ("EXECUTING jump_n %d.\n", mcnt);
6076
6077 /* Originally, this is how many times we CAN jump. */
6078 if (mcnt != 0)
6079 {
6080 /* Here, we discard `const', making re_match non-reentrant. */
6081 unsigned char *p2 = (unsigned char*) p + 2; /* counter loc. */
6082 mcnt--;
6083 PUSH_NUMBER (p2, mcnt);
6084 goto unconditional_jump;
6085 }
6086 /* If don't have to jump any more, skip over the rest of command. */
6087 else
6088 p += 4;
6089 break;
6090
6091 case set_number_at:
6092 {
6093 unsigned char *p2; /* Location of the counter. */
6094 DEBUG_PRINT1 ("EXECUTING set_number_at.\n");
6095
6096 EXTRACT_NUMBER_AND_INCR (mcnt, p);
6097 /* Here, we discard `const', making re_match non-reentrant. */
6098 p2 = (unsigned char*) p + mcnt;
6099 /* Signedness doesn't matter since we only copy MCNT's bits . */
6100 EXTRACT_NUMBER_AND_INCR (mcnt, p);
6101 DEBUG_PRINT3 (" Setting %p to %d.\n", p2, mcnt);
6102 PUSH_NUMBER (p2, mcnt);
6103 break;
6104 }
6105
6106 case wordbound:
6107 case notwordbound:
6108 not = (re_opcode_t) *(p - 1) == notwordbound;
6109 DEBUG_PRINT2 ("EXECUTING %swordbound.\n", not?"not":"");
6110
6111 /* We SUCCEED (or FAIL) in one of the following cases: */
6112
6113 /* Case 1: D is at the beginning or the end of string. */
6114 if (AT_STRINGS_BEG (d) || AT_STRINGS_END (d))
6115 not = !not;
6116 else
6117 {
6118 /* C1 is the character before D, S1 is the syntax of C1, C2
6119 is the character at D, and S2 is the syntax of C2. */
6120 re_wchar_t c1, c2;
6121 int s1, s2;
6122 int dummy;
6123 #ifdef emacs
6124 int offset = PTR_TO_OFFSET (d - 1);
6125 int charpos = SYNTAX_TABLE_BYTE_TO_CHAR (offset);
6126 UPDATE_SYNTAX_TABLE (charpos);
6127 #endif
6128 GET_CHAR_BEFORE_2 (c1, d, string1, end1, string2, end2);
6129 s1 = SYNTAX (c1);
6130 #ifdef emacs
6131 UPDATE_SYNTAX_TABLE_FORWARD (charpos + 1);
6132 #endif
6133 PREFETCH_NOLIMIT ();
6134 GET_CHAR_AFTER (c2, d, dummy);
6135 s2 = SYNTAX (c2);
6136
6137 if (/* Case 2: Only one of S1 and S2 is Sword. */
6138 ((s1 == Sword) != (s2 == Sword))
6139 /* Case 3: Both of S1 and S2 are Sword, and macro
6140 WORD_BOUNDARY_P (C1, C2) returns nonzero. */
6141 || ((s1 == Sword) && WORD_BOUNDARY_P (c1, c2)))
6142 not = !not;
6143 }
6144 if (not)
6145 break;
6146 else
6147 goto fail;
6148
6149 case wordbeg:
6150 DEBUG_PRINT1 ("EXECUTING wordbeg.\n");
6151
6152 /* We FAIL in one of the following cases: */
6153
6154 /* Case 1: D is at the end of string. */
6155 if (AT_STRINGS_END (d))
6156 goto fail;
6157 else
6158 {
6159 /* C1 is the character before D, S1 is the syntax of C1, C2
6160 is the character at D, and S2 is the syntax of C2. */
6161 re_wchar_t c1, c2;
6162 int s1, s2;
6163 int dummy;
6164 #ifdef emacs
6165 int offset = PTR_TO_OFFSET (d);
6166 int charpos = SYNTAX_TABLE_BYTE_TO_CHAR (offset);
6167 UPDATE_SYNTAX_TABLE (charpos);
6168 #endif
6169 PREFETCH ();
6170 GET_CHAR_AFTER (c2, d, dummy);
6171 s2 = SYNTAX (c2);
6172
6173 /* Case 2: S2 is not Sword. */
6174 if (s2 != Sword)
6175 goto fail;
6176
6177 /* Case 3: D is not at the beginning of string ... */
6178 if (!AT_STRINGS_BEG (d))
6179 {
6180 GET_CHAR_BEFORE_2 (c1, d, string1, end1, string2, end2);
6181 #ifdef emacs
6182 UPDATE_SYNTAX_TABLE_BACKWARD (charpos - 1);
6183 #endif
6184 s1 = SYNTAX (c1);
6185
6186 /* ... and S1 is Sword, and WORD_BOUNDARY_P (C1, C2)
6187 returns 0. */
6188 if ((s1 == Sword) && !WORD_BOUNDARY_P (c1, c2))
6189 goto fail;
6190 }
6191 }
6192 break;
6193
6194 case wordend:
6195 DEBUG_PRINT1 ("EXECUTING wordend.\n");
6196
6197 /* We FAIL in one of the following cases: */
6198
6199 /* Case 1: D is at the beginning of string. */
6200 if (AT_STRINGS_BEG (d))
6201 goto fail;
6202 else
6203 {
6204 /* C1 is the character before D, S1 is the syntax of C1, C2
6205 is the character at D, and S2 is the syntax of C2. */
6206 re_wchar_t c1, c2;
6207 int s1, s2;
6208 int dummy;
6209 #ifdef emacs
6210 int offset = PTR_TO_OFFSET (d) - 1;
6211 int charpos = SYNTAX_TABLE_BYTE_TO_CHAR (offset);
6212 UPDATE_SYNTAX_TABLE (charpos);
6213 #endif
6214 GET_CHAR_BEFORE_2 (c1, d, string1, end1, string2, end2);
6215 s1 = SYNTAX (c1);
6216
6217 /* Case 2: S1 is not Sword. */
6218 if (s1 != Sword)
6219 goto fail;
6220
6221 /* Case 3: D is not at the end of string ... */
6222 if (!AT_STRINGS_END (d))
6223 {
6224 PREFETCH_NOLIMIT ();
6225 GET_CHAR_AFTER (c2, d, dummy);
6226 #ifdef emacs
6227 UPDATE_SYNTAX_TABLE_FORWARD (charpos);
6228 #endif
6229 s2 = SYNTAX (c2);
6230
6231 /* ... and S2 is Sword, and WORD_BOUNDARY_P (C1, C2)
6232 returns 0. */
6233 if ((s2 == Sword) && !WORD_BOUNDARY_P (c1, c2))
6234 goto fail;
6235 }
6236 }
6237 break;
6238
6239 case symbeg:
6240 DEBUG_PRINT1 ("EXECUTING symbeg.\n");
6241
6242 /* We FAIL in one of the following cases: */
6243
6244 /* Case 1: D is at the end of string. */
6245 if (AT_STRINGS_END (d))
6246 goto fail;
6247 else
6248 {
6249 /* C1 is the character before D, S1 is the syntax of C1, C2
6250 is the character at D, and S2 is the syntax of C2. */
6251 re_wchar_t c1, c2;
6252 int s1, s2;
6253 #ifdef emacs
6254 int offset = PTR_TO_OFFSET (d);
6255 int charpos = SYNTAX_TABLE_BYTE_TO_CHAR (offset);
6256 UPDATE_SYNTAX_TABLE (charpos);
6257 #endif
6258 PREFETCH ();
6259 c2 = RE_STRING_CHAR (d, dend - d, target_multibyte);
6260 s2 = SYNTAX (c2);
6261
6262 /* Case 2: S2 is neither Sword nor Ssymbol. */
6263 if (s2 != Sword && s2 != Ssymbol)
6264 goto fail;
6265
6266 /* Case 3: D is not at the beginning of string ... */
6267 if (!AT_STRINGS_BEG (d))
6268 {
6269 GET_CHAR_BEFORE_2 (c1, d, string1, end1, string2, end2);
6270 #ifdef emacs
6271 UPDATE_SYNTAX_TABLE_BACKWARD (charpos - 1);
6272 #endif
6273 s1 = SYNTAX (c1);
6274
6275 /* ... and S1 is Sword or Ssymbol. */
6276 if (s1 == Sword || s1 == Ssymbol)
6277 goto fail;
6278 }
6279 }
6280 break;
6281
6282 case symend:
6283 DEBUG_PRINT1 ("EXECUTING symend.\n");
6284
6285 /* We FAIL in one of the following cases: */
6286
6287 /* Case 1: D is at the beginning of string. */
6288 if (AT_STRINGS_BEG (d))
6289 goto fail;
6290 else
6291 {
6292 /* C1 is the character before D, S1 is the syntax of C1, C2
6293 is the character at D, and S2 is the syntax of C2. */
6294 re_wchar_t c1, c2;
6295 int s1, s2;
6296 #ifdef emacs
6297 int offset = PTR_TO_OFFSET (d) - 1;
6298 int charpos = SYNTAX_TABLE_BYTE_TO_CHAR (offset);
6299 UPDATE_SYNTAX_TABLE (charpos);
6300 #endif
6301 GET_CHAR_BEFORE_2 (c1, d, string1, end1, string2, end2);
6302 s1 = SYNTAX (c1);
6303
6304 /* Case 2: S1 is neither Ssymbol nor Sword. */
6305 if (s1 != Sword && s1 != Ssymbol)
6306 goto fail;
6307
6308 /* Case 3: D is not at the end of string ... */
6309 if (!AT_STRINGS_END (d))
6310 {
6311 PREFETCH_NOLIMIT ();
6312 c2 = RE_STRING_CHAR (d, dend - d, target_multibyte);
6313 #ifdef emacs
6314 UPDATE_SYNTAX_TABLE_FORWARD (charpos + 1);
6315 #endif
6316 s2 = SYNTAX (c2);
6317
6318 /* ... and S2 is Sword or Ssymbol. */
6319 if (s2 == Sword || s2 == Ssymbol)
6320 goto fail;
6321 }
6322 }
6323 break;
6324
6325 case syntaxspec:
6326 case notsyntaxspec:
6327 not = (re_opcode_t) *(p - 1) == notsyntaxspec;
6328 mcnt = *p++;
6329 DEBUG_PRINT3 ("EXECUTING %ssyntaxspec %d.\n", not?"not":"", mcnt);
6330 PREFETCH ();
6331 #ifdef emacs
6332 {
6333 int offset = PTR_TO_OFFSET (d);
6334 int pos1 = SYNTAX_TABLE_BYTE_TO_CHAR (offset);
6335 UPDATE_SYNTAX_TABLE (pos1);
6336 }
6337 #endif
6338 {
6339 int len;
6340 re_wchar_t c;
6341
6342 GET_CHAR_AFTER (c, d, len);
6343 if ((SYNTAX (c) != (enum syntaxcode) mcnt) ^ not)
6344 goto fail;
6345 d += len;
6346 }
6347 break;
6348
6349 #ifdef emacs
6350 case before_dot:
6351 DEBUG_PRINT1 ("EXECUTING before_dot.\n");
6352 if (PTR_BYTE_POS (d) >= PT_BYTE)
6353 goto fail;
6354 break;
6355
6356 case at_dot:
6357 DEBUG_PRINT1 ("EXECUTING at_dot.\n");
6358 if (PTR_BYTE_POS (d) != PT_BYTE)
6359 goto fail;
6360 break;
6361
6362 case after_dot:
6363 DEBUG_PRINT1 ("EXECUTING after_dot.\n");
6364 if (PTR_BYTE_POS (d) <= PT_BYTE)
6365 goto fail;
6366 break;
6367
6368 case categoryspec:
6369 case notcategoryspec:
6370 not = (re_opcode_t) *(p - 1) == notcategoryspec;
6371 mcnt = *p++;
6372 DEBUG_PRINT3 ("EXECUTING %scategoryspec %d.\n", not?"not":"", mcnt);
6373 PREFETCH ();
6374 {
6375 int len;
6376 re_wchar_t c;
6377
6378 GET_CHAR_AFTER (c, d, len);
6379 if ((!CHAR_HAS_CATEGORY (c, mcnt)) ^ not)
6380 goto fail;
6381 d += len;
6382 }
6383 break;
6384
6385 #endif /* emacs */
6386
6387 default:
6388 abort ();
6389 }
6390 continue; /* Successfully executed one pattern command; keep going. */
6391
6392
6393 /* We goto here if a matching operation fails. */
6394 fail:
6395 IMMEDIATE_QUIT_CHECK;
6396 if (!FAIL_STACK_EMPTY ())
6397 {
6398 re_char *str, *pat;
6399 /* A restart point is known. Restore to that state. */
6400 DEBUG_PRINT1 ("\nFAIL:\n");
6401 POP_FAILURE_POINT (str, pat);
6402 switch (SWITCH_ENUM_CAST ((re_opcode_t) *pat++))
6403 {
6404 case on_failure_keep_string_jump:
6405 assert (str == NULL);
6406 goto continue_failure_jump;
6407
6408 case on_failure_jump_nastyloop:
6409 assert ((re_opcode_t)pat[-2] == no_op);
6410 PUSH_FAILURE_POINT (pat - 2, str);
6411 /* Fallthrough */
6412
6413 case on_failure_jump_loop:
6414 case on_failure_jump:
6415 case succeed_n:
6416 d = str;
6417 continue_failure_jump:
6418 EXTRACT_NUMBER_AND_INCR (mcnt, pat);
6419 p = pat + mcnt;
6420 break;
6421
6422 case no_op:
6423 /* A special frame used for nastyloops. */
6424 goto fail;
6425
6426 default:
6427 abort();
6428 }
6429
6430 assert (p >= bufp->buffer && p <= pend);
6431
6432 if (d >= string1 && d <= end1)
6433 dend = end_match_1;
6434 }
6435 else
6436 break; /* Matching at this starting point really fails. */
6437 } /* for (;;) */
6438
6439 if (best_regs_set)
6440 goto restore_best_regs;
6441
6442 FREE_VARIABLES ();
6443
6444 return -1; /* Failure to match. */
6445 } /* re_match_2 */
6446 \f
6447 /* Subroutine definitions for re_match_2. */
6448
6449 /* Return zero if TRANSLATE[S1] and TRANSLATE[S2] are identical for LEN
6450 bytes; nonzero otherwise. */
6451
6452 static int
6453 bcmp_translate (s1, s2, len, translate, target_multibyte)
6454 re_char *s1, *s2;
6455 register int len;
6456 RE_TRANSLATE_TYPE translate;
6457 const int target_multibyte;
6458 {
6459 register re_char *p1 = s1, *p2 = s2;
6460 re_char *p1_end = s1 + len;
6461 re_char *p2_end = s2 + len;
6462
6463 /* FIXME: Checking both p1 and p2 presumes that the two strings might have
6464 different lengths, but relying on a single `len' would break this. -sm */
6465 while (p1 < p1_end && p2 < p2_end)
6466 {
6467 int p1_charlen, p2_charlen;
6468 re_wchar_t p1_ch, p2_ch;
6469
6470 GET_CHAR_AFTER (p1_ch, p1, p1_charlen);
6471 GET_CHAR_AFTER (p2_ch, p2, p2_charlen);
6472
6473 if (RE_TRANSLATE (translate, p1_ch)
6474 != RE_TRANSLATE (translate, p2_ch))
6475 return 1;
6476
6477 p1 += p1_charlen, p2 += p2_charlen;
6478 }
6479
6480 if (p1 != p1_end || p2 != p2_end)
6481 return 1;
6482
6483 return 0;
6484 }
6485 \f
6486 /* Entry points for GNU code. */
6487
6488 /* re_compile_pattern is the GNU regular expression compiler: it
6489 compiles PATTERN (of length SIZE) and puts the result in BUFP.
6490 Returns 0 if the pattern was valid, otherwise an error string.
6491
6492 Assumes the `allocated' (and perhaps `buffer') and `translate' fields
6493 are set in BUFP on entry.
6494
6495 We call regex_compile to do the actual compilation. */
6496
6497 const char *
6498 re_compile_pattern (pattern, length, bufp)
6499 const char *pattern;
6500 size_t length;
6501 struct re_pattern_buffer *bufp;
6502 {
6503 reg_errcode_t ret;
6504
6505 #ifdef emacs
6506 gl_state.current_syntax_table = current_buffer->syntax_table;
6507 #endif
6508
6509 /* GNU code is written to assume at least RE_NREGS registers will be set
6510 (and at least one extra will be -1). */
6511 bufp->regs_allocated = REGS_UNALLOCATED;
6512
6513 /* And GNU code determines whether or not to get register information
6514 by passing null for the REGS argument to re_match, etc., not by
6515 setting no_sub. */
6516 bufp->no_sub = 0;
6517
6518 ret = regex_compile ((re_char*) pattern, length, re_syntax_options, bufp);
6519
6520 if (!ret)
6521 return NULL;
6522 return gettext (re_error_msgid[(int) ret]);
6523 }
6524 WEAK_ALIAS (__re_compile_pattern, re_compile_pattern)
6525 \f
6526 /* Entry points compatible with 4.2 BSD regex library. We don't define
6527 them unless specifically requested. */
6528
6529 #if defined _REGEX_RE_COMP || defined _LIBC
6530
6531 /* BSD has one and only one pattern buffer. */
6532 static struct re_pattern_buffer re_comp_buf;
6533
6534 char *
6535 # ifdef _LIBC
6536 /* Make these definitions weak in libc, so POSIX programs can redefine
6537 these names if they don't use our functions, and still use
6538 regcomp/regexec below without link errors. */
6539 weak_function
6540 # endif
6541 re_comp (s)
6542 const char *s;
6543 {
6544 reg_errcode_t ret;
6545
6546 if (!s)
6547 {
6548 if (!re_comp_buf.buffer)
6549 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6550 return (char *) gettext ("No previous regular expression");
6551 return 0;
6552 }
6553
6554 if (!re_comp_buf.buffer)
6555 {
6556 re_comp_buf.buffer = (unsigned char *) malloc (200);
6557 if (re_comp_buf.buffer == NULL)
6558 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6559 return (char *) gettext (re_error_msgid[(int) REG_ESPACE]);
6560 re_comp_buf.allocated = 200;
6561
6562 re_comp_buf.fastmap = (char *) malloc (1 << BYTEWIDTH);
6563 if (re_comp_buf.fastmap == NULL)
6564 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6565 return (char *) gettext (re_error_msgid[(int) REG_ESPACE]);
6566 }
6567
6568 /* Since `re_exec' always passes NULL for the `regs' argument, we
6569 don't need to initialize the pattern buffer fields which affect it. */
6570
6571 ret = regex_compile (s, strlen (s), re_syntax_options, &re_comp_buf);
6572
6573 if (!ret)
6574 return NULL;
6575
6576 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6577 return (char *) gettext (re_error_msgid[(int) ret]);
6578 }
6579
6580
6581 int
6582 # ifdef _LIBC
6583 weak_function
6584 # endif
6585 re_exec (s)
6586 const char *s;
6587 {
6588 const int len = strlen (s);
6589 return
6590 0 <= re_search (&re_comp_buf, s, len, 0, len, (struct re_registers *) 0);
6591 }
6592 #endif /* _REGEX_RE_COMP */
6593 \f
6594 /* POSIX.2 functions. Don't define these for Emacs. */
6595
6596 #ifndef emacs
6597
6598 /* regcomp takes a regular expression as a string and compiles it.
6599
6600 PREG is a regex_t *. We do not expect any fields to be initialized,
6601 since POSIX says we shouldn't. Thus, we set
6602
6603 `buffer' to the compiled pattern;
6604 `used' to the length of the compiled pattern;
6605 `syntax' to RE_SYNTAX_POSIX_EXTENDED if the
6606 REG_EXTENDED bit in CFLAGS is set; otherwise, to
6607 RE_SYNTAX_POSIX_BASIC;
6608 `fastmap' to an allocated space for the fastmap;
6609 `fastmap_accurate' to zero;
6610 `re_nsub' to the number of subexpressions in PATTERN.
6611
6612 PATTERN is the address of the pattern string.
6613
6614 CFLAGS is a series of bits which affect compilation.
6615
6616 If REG_EXTENDED is set, we use POSIX extended syntax; otherwise, we
6617 use POSIX basic syntax.
6618
6619 If REG_NEWLINE is set, then . and [^...] don't match newline.
6620 Also, regexec will try a match beginning after every newline.
6621
6622 If REG_ICASE is set, then we considers upper- and lowercase
6623 versions of letters to be equivalent when matching.
6624
6625 If REG_NOSUB is set, then when PREG is passed to regexec, that
6626 routine will report only success or failure, and nothing about the
6627 registers.
6628
6629 It returns 0 if it succeeds, nonzero if it doesn't. (See regex.h for
6630 the return codes and their meanings.) */
6631
6632 int
6633 regcomp (preg, pattern, cflags)
6634 regex_t *__restrict preg;
6635 const char *__restrict pattern;
6636 int cflags;
6637 {
6638 reg_errcode_t ret;
6639 reg_syntax_t syntax
6640 = (cflags & REG_EXTENDED) ?
6641 RE_SYNTAX_POSIX_EXTENDED : RE_SYNTAX_POSIX_BASIC;
6642
6643 /* regex_compile will allocate the space for the compiled pattern. */
6644 preg->buffer = 0;
6645 preg->allocated = 0;
6646 preg->used = 0;
6647
6648 /* Try to allocate space for the fastmap. */
6649 preg->fastmap = (char *) malloc (1 << BYTEWIDTH);
6650
6651 if (cflags & REG_ICASE)
6652 {
6653 unsigned i;
6654
6655 preg->translate
6656 = (RE_TRANSLATE_TYPE) malloc (CHAR_SET_SIZE
6657 * sizeof (*(RE_TRANSLATE_TYPE)0));
6658 if (preg->translate == NULL)
6659 return (int) REG_ESPACE;
6660
6661 /* Map uppercase characters to corresponding lowercase ones. */
6662 for (i = 0; i < CHAR_SET_SIZE; i++)
6663 preg->translate[i] = ISUPPER (i) ? TOLOWER (i) : i;
6664 }
6665 else
6666 preg->translate = NULL;
6667
6668 /* If REG_NEWLINE is set, newlines are treated differently. */
6669 if (cflags & REG_NEWLINE)
6670 { /* REG_NEWLINE implies neither . nor [^...] match newline. */
6671 syntax &= ~RE_DOT_NEWLINE;
6672 syntax |= RE_HAT_LISTS_NOT_NEWLINE;
6673 }
6674 else
6675 syntax |= RE_NO_NEWLINE_ANCHOR;
6676
6677 preg->no_sub = !!(cflags & REG_NOSUB);
6678
6679 /* POSIX says a null character in the pattern terminates it, so we
6680 can use strlen here in compiling the pattern. */
6681 ret = regex_compile ((re_char*) pattern, strlen (pattern), syntax, preg);
6682
6683 /* POSIX doesn't distinguish between an unmatched open-group and an
6684 unmatched close-group: both are REG_EPAREN. */
6685 if (ret == REG_ERPAREN)
6686 ret = REG_EPAREN;
6687
6688 if (ret == REG_NOERROR && preg->fastmap)
6689 { /* Compute the fastmap now, since regexec cannot modify the pattern
6690 buffer. */
6691 re_compile_fastmap (preg);
6692 if (preg->can_be_null)
6693 { /* The fastmap can't be used anyway. */
6694 free (preg->fastmap);
6695 preg->fastmap = NULL;
6696 }
6697 }
6698 return (int) ret;
6699 }
6700 WEAK_ALIAS (__regcomp, regcomp)
6701
6702
6703 /* regexec searches for a given pattern, specified by PREG, in the
6704 string STRING.
6705
6706 If NMATCH is zero or REG_NOSUB was set in the cflags argument to
6707 `regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at
6708 least NMATCH elements, and we set them to the offsets of the
6709 corresponding matched substrings.
6710
6711 EFLAGS specifies `execution flags' which affect matching: if
6712 REG_NOTBOL is set, then ^ does not match at the beginning of the
6713 string; if REG_NOTEOL is set, then $ does not match at the end.
6714
6715 We return 0 if we find a match and REG_NOMATCH if not. */
6716
6717 int
6718 regexec (preg, string, nmatch, pmatch, eflags)
6719 const regex_t *__restrict preg;
6720 const char *__restrict string;
6721 size_t nmatch;
6722 regmatch_t pmatch[__restrict_arr];
6723 int eflags;
6724 {
6725 int ret;
6726 struct re_registers regs;
6727 regex_t private_preg;
6728 int len = strlen (string);
6729 boolean want_reg_info = !preg->no_sub && nmatch > 0 && pmatch;
6730
6731 private_preg = *preg;
6732
6733 private_preg.not_bol = !!(eflags & REG_NOTBOL);
6734 private_preg.not_eol = !!(eflags & REG_NOTEOL);
6735
6736 /* The user has told us exactly how many registers to return
6737 information about, via `nmatch'. We have to pass that on to the
6738 matching routines. */
6739 private_preg.regs_allocated = REGS_FIXED;
6740
6741 if (want_reg_info)
6742 {
6743 regs.num_regs = nmatch;
6744 regs.start = TALLOC (nmatch * 2, regoff_t);
6745 if (regs.start == NULL)
6746 return (int) REG_NOMATCH;
6747 regs.end = regs.start + nmatch;
6748 }
6749
6750 /* Instead of using not_eol to implement REG_NOTEOL, we could simply
6751 pass (&private_preg, string, len + 1, 0, len, ...) pretending the string
6752 was a little bit longer but still only matching the real part.
6753 This works because the `endline' will check for a '\n' and will find a
6754 '\0', correctly deciding that this is not the end of a line.
6755 But it doesn't work out so nicely for REG_NOTBOL, since we don't have
6756 a convenient '\0' there. For all we know, the string could be preceded
6757 by '\n' which would throw things off. */
6758
6759 /* Perform the searching operation. */
6760 ret = re_search (&private_preg, string, len,
6761 /* start: */ 0, /* range: */ len,
6762 want_reg_info ? &regs : (struct re_registers *) 0);
6763
6764 /* Copy the register information to the POSIX structure. */
6765 if (want_reg_info)
6766 {
6767 if (ret >= 0)
6768 {
6769 unsigned r;
6770
6771 for (r = 0; r < nmatch; r++)
6772 {
6773 pmatch[r].rm_so = regs.start[r];
6774 pmatch[r].rm_eo = regs.end[r];
6775 }
6776 }
6777
6778 /* If we needed the temporary register info, free the space now. */
6779 free (regs.start);
6780 }
6781
6782 /* We want zero return to mean success, unlike `re_search'. */
6783 return ret >= 0 ? (int) REG_NOERROR : (int) REG_NOMATCH;
6784 }
6785 WEAK_ALIAS (__regexec, regexec)
6786
6787
6788 /* Returns a message corresponding to an error code, ERR_CODE, returned
6789 from either regcomp or regexec. We don't use PREG here.
6790
6791 ERR_CODE was previously called ERRCODE, but that name causes an
6792 error with msvc8 compiler. */
6793
6794 size_t
6795 regerror (err_code, preg, errbuf, errbuf_size)
6796 int err_code;
6797 const regex_t *preg;
6798 char *errbuf;
6799 size_t errbuf_size;
6800 {
6801 const char *msg;
6802 size_t msg_size;
6803
6804 if (err_code < 0
6805 || err_code >= (sizeof (re_error_msgid) / sizeof (re_error_msgid[0])))
6806 /* Only error codes returned by the rest of the code should be passed
6807 to this routine. If we are given anything else, or if other regex
6808 code generates an invalid error code, then the program has a bug.
6809 Dump core so we can fix it. */
6810 abort ();
6811
6812 msg = gettext (re_error_msgid[err_code]);
6813
6814 msg_size = strlen (msg) + 1; /* Includes the null. */
6815
6816 if (errbuf_size != 0)
6817 {
6818 if (msg_size > errbuf_size)
6819 {
6820 strncpy (errbuf, msg, errbuf_size - 1);
6821 errbuf[errbuf_size - 1] = 0;
6822 }
6823 else
6824 strcpy (errbuf, msg);
6825 }
6826
6827 return msg_size;
6828 }
6829 WEAK_ALIAS (__regerror, regerror)
6830
6831
6832 /* Free dynamically allocated space used by PREG. */
6833
6834 void
6835 regfree (preg)
6836 regex_t *preg;
6837 {
6838 free (preg->buffer);
6839 preg->buffer = NULL;
6840
6841 preg->allocated = 0;
6842 preg->used = 0;
6843
6844 free (preg->fastmap);
6845 preg->fastmap = NULL;
6846 preg->fastmap_accurate = 0;
6847
6848 free (preg->translate);
6849 preg->translate = NULL;
6850 }
6851 WEAK_ALIAS (__regfree, regfree)
6852
6853 #endif /* not emacs */
6854
6855 /* arch-tag: 4ffd68ba-2a9e-435b-a21a-018990f9eeb2
6856 (do not change this comment) */
Emacs repository mirror