?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
* character.h (STRING_CHAR, STRING_CHAR_AND_LENGTH): Remove
[emacs.git] / src / regex.c
blobb3e1430fa31514f65103212d881a510ab55dfbde
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 <setjmp.h>
121 # include "lisp.h"
122 # include "buffer.h"
123
124 /* Make syntax table lookup grant data in gl_state. */
125 # define SYNTAX_ENTRY_VIA_PROPERTY
126
127 # include "syntax.h"
128 # include "character.h"
129 # include "category.h"
130
131 # ifdef malloc
132 # undef malloc
133 # endif
134 # define malloc xmalloc
135 # ifdef realloc
136 # undef realloc
137 # endif
138 # define realloc xrealloc
139 # ifdef free
140 # undef free
141 # endif
142 # define free xfree
143
144 /* Converts the pointer to the char to BEG-based offset from the start. */
145 # define PTR_TO_OFFSET(d) POS_AS_IN_BUFFER (POINTER_TO_OFFSET (d))
146 # define POS_AS_IN_BUFFER(p) ((p) + (NILP (re_match_object) || BUFFERP (re_match_object)))
147
148 # define RE_MULTIBYTE_P(bufp) ((bufp)->multibyte)
149 # define RE_TARGET_MULTIBYTE_P(bufp) ((bufp)->target_multibyte)
150 # define RE_STRING_CHAR(p, multibyte) \
151 (multibyte ? (STRING_CHAR (p)) : (*(p)))
152 # define RE_STRING_CHAR_AND_LENGTH(p, len, multibyte) \
153 (multibyte ? (STRING_CHAR_AND_LENGTH (p, len)) : ((len) = 1, *(p)))
154
155 # define RE_CHAR_TO_MULTIBYTE(c) UNIBYTE_TO_CHAR (c)
156
157 # define RE_CHAR_TO_UNIBYTE(c) CHAR_TO_BYTE_SAFE (c)
158
159 /* Set C a (possibly converted to multibyte) character before P. P
160 points into a string which is the virtual concatenation of STR1
161 (which ends at END1) or STR2 (which ends at END2). */
162 # define GET_CHAR_BEFORE_2(c, p, str1, end1, str2, end2) \
163 do { \
164 if (target_multibyte) \
165 { \
166 re_char *dtemp = (p) == (str2) ? (end1) : (p); \
167 re_char *dlimit = ((p) > (str2) && (p) <= (end2)) ? (str2) : (str1); \
168 while (dtemp-- > dlimit && !CHAR_HEAD_P (*dtemp)); \
169 c = STRING_CHAR (dtemp); \
170 } \
171 else \
172 { \
173 (c = ((p) == (str2) ? (end1) : (p))[-1]); \
174 (c) = RE_CHAR_TO_MULTIBYTE (c); \
175 } \
176 } while (0)
177
178 /* Set C a (possibly converted to multibyte) character at P, and set
179 LEN to the byte length of that character. */
180 # define GET_CHAR_AFTER(c, p, len) \
181 do { \
182 if (target_multibyte) \
183 (c) = STRING_CHAR_AND_LENGTH (p, len); \
184 else \
185 { \
186 (c) = *p; \
187 len = 1; \
188 (c) = RE_CHAR_TO_MULTIBYTE (c); \
189 } \
190 } while (0)
191
192 #else /* not emacs */
193
194 /* If we are not linking with Emacs proper,
195 we can't use the relocating allocator
196 even if config.h says that we can. */
197 # undef REL_ALLOC
198
199 # if defined STDC_HEADERS || defined _LIBC
200 # include <stdlib.h>
201 # else
202 char *malloc ();
203 char *realloc ();
204 # endif
205
206 /* When used in Emacs's lib-src, we need xmalloc and xrealloc. */
207
208 void *
209 xmalloc (size)
210 size_t size;
211 {
212 register void *val;
213 val = (void *) malloc (size);
214 if (!val && size)
215 {
216 write (2, "virtual memory exhausted\n", 25);
217 exit (1);
218 }
219 return val;
220 }
221
222 void *
223 xrealloc (block, size)
224 void *block;
225 size_t size;
226 {
227 register void *val;
228 /* We must call malloc explicitly when BLOCK is 0, since some
229 reallocs don't do this. */
230 if (! block)
231 val = (void *) malloc (size);
232 else
233 val = (void *) realloc (block, size);
234 if (!val && size)
235 {
236 write (2, "virtual memory exhausted\n", 25);
237 exit (1);
238 }
239 return val;
240 }
241
242 # ifdef malloc
243 # undef malloc
244 # endif
245 # define malloc xmalloc
246 # ifdef realloc
247 # undef realloc
248 # endif
249 # define realloc xrealloc
250
251 /* When used in Emacs's lib-src, we need to get bzero and bcopy somehow.
252 If nothing else has been done, use the method below. */
253 # ifdef INHIBIT_STRING_HEADER
254 # if !(defined HAVE_BZERO && defined HAVE_BCOPY)
255 # if !defined bzero && !defined bcopy
256 # undef INHIBIT_STRING_HEADER
257 # endif
258 # endif
259 # endif
260
261 /* This is the normal way of making sure we have memcpy, memcmp and bzero.
262 This is used in most programs--a few other programs avoid this
263 by defining INHIBIT_STRING_HEADER. */
264 # ifndef INHIBIT_STRING_HEADER
265 # if defined HAVE_STRING_H || defined STDC_HEADERS || defined _LIBC
266 # include <string.h>
267 # ifndef bzero
268 # ifndef _LIBC
269 # define bzero(s, n) (memset (s, '\0', n), (s))
270 # else
271 # define bzero(s, n) __bzero (s, n)
272 # endif
273 # endif
274 # else
275 # include <strings.h>
276 # ifndef memcmp
277 # define memcmp(s1, s2, n) bcmp (s1, s2, n)
278 # endif
279 # ifndef memcpy
280 # define memcpy(d, s, n) (bcopy (s, d, n), (d))
281 # endif
282 # endif
283 # endif
284
285 /* Define the syntax stuff for \<, \>, etc. */
286
287 /* Sword must be nonzero for the wordchar pattern commands in re_match_2. */
288 enum syntaxcode { Swhitespace = 0, Sword = 1, Ssymbol = 2 };
289
290 # define SWITCH_ENUM_CAST(x) (x)
291
292 /* Dummy macros for non-Emacs environments. */
293 # define BASE_LEADING_CODE_P(c) (0)
294 # define CHAR_CHARSET(c) 0
295 # define CHARSET_LEADING_CODE_BASE(c) 0
296 # define MAX_MULTIBYTE_LENGTH 1
297 # define RE_MULTIBYTE_P(x) 0
298 # define RE_TARGET_MULTIBYTE_P(x) 0
299 # define WORD_BOUNDARY_P(c1, c2) (0)
300 # define CHAR_HEAD_P(p) (1)
301 # define SINGLE_BYTE_CHAR_P(c) (1)
302 # define SAME_CHARSET_P(c1, c2) (1)
303 # define MULTIBYTE_FORM_LENGTH(p, s) (1)
304 # define PREV_CHAR_BOUNDARY(p, limit) ((p)--)
305 # define STRING_CHAR(p) (*(p))
306 # define RE_STRING_CHAR(p, multibyte) STRING_CHAR (p)
307 # define CHAR_STRING(c, s) (*(s) = (c), 1)
308 # define STRING_CHAR_AND_LENGTH(p, actual_len) ((actual_len) = 1, *(p))
309 # define RE_STRING_CHAR_AND_LENGTH(p, len, multibyte) STRING_CHAR_AND_LENGTH (p, len)
310 # define RE_CHAR_TO_MULTIBYTE(c) (c)
311 # define RE_CHAR_TO_UNIBYTE(c) (c)
312 # define GET_CHAR_BEFORE_2(c, p, str1, end1, str2, end2) \
313 (c = ((p) == (str2) ? *((end1) - 1) : *((p) - 1)))
314 # define GET_CHAR_AFTER(c, p, len) \
315 (c = *p, len = 1)
316 # define MAKE_CHAR(charset, c1, c2) (c1)
317 # define BYTE8_TO_CHAR(c) (c)
318 # define CHAR_BYTE8_P(c) (0)
319 # define CHAR_LEADING_CODE(c) (c)
320
321 #endif /* not emacs */
322
323 #ifndef RE_TRANSLATE
324 # define RE_TRANSLATE(TBL, C) ((unsigned char)(TBL)[C])
325 # define RE_TRANSLATE_P(TBL) (TBL)
326 #endif
327 \f
328 /* Get the interface, including the syntax bits. */
329 #include "regex.h"
330
331 /* isalpha etc. are used for the character classes. */
332 #include <ctype.h>
333
334 #ifdef emacs
335
336 /* 1 if C is an ASCII character. */
337 # define IS_REAL_ASCII(c) ((c) < 0200)
338
339 /* 1 if C is a unibyte character. */
340 # define ISUNIBYTE(c) (SINGLE_BYTE_CHAR_P ((c)))
341
342 /* The Emacs definitions should not be directly affected by locales. */
343
344 /* In Emacs, these are only used for single-byte characters. */
345 # define ISDIGIT(c) ((c) >= '0' && (c) <= '9')
346 # define ISCNTRL(c) ((c) < ' ')
347 # define ISXDIGIT(c) (((c) >= '0' && (c) <= '9') \
348 || ((c) >= 'a' && (c) <= 'f') \
349 || ((c) >= 'A' && (c) <= 'F'))
350
351 /* This is only used for single-byte characters. */
352 # define ISBLANK(c) ((c) == ' ' || (c) == '\t')
353
354 /* The rest must handle multibyte characters. */
355
356 # define ISGRAPH(c) (SINGLE_BYTE_CHAR_P (c) \
357 ? (c) > ' ' && !((c) >= 0177 && (c) <= 0237) \
358 : 1)
359
360 # define ISPRINT(c) (SINGLE_BYTE_CHAR_P (c) \
361 ? (c) >= ' ' && !((c) >= 0177 && (c) <= 0237) \
362 : 1)
363
364 # define ISALNUM(c) (IS_REAL_ASCII (c) \
365 ? (((c) >= 'a' && (c) <= 'z') \
366 || ((c) >= 'A' && (c) <= 'Z') \
367 || ((c) >= '0' && (c) <= '9')) \
368 : SYNTAX (c) == Sword)
369
370 # define ISALPHA(c) (IS_REAL_ASCII (c) \
371 ? (((c) >= 'a' && (c) <= 'z') \
372 || ((c) >= 'A' && (c) <= 'Z')) \
373 : SYNTAX (c) == Sword)
374
375 # define ISLOWER(c) (LOWERCASEP (c))
376
377 # define ISPUNCT(c) (IS_REAL_ASCII (c) \
378 ? ((c) > ' ' && (c) < 0177 \
379 && !(((c) >= 'a' && (c) <= 'z') \
380 || ((c) >= 'A' && (c) <= 'Z') \
381 || ((c) >= '0' && (c) <= '9'))) \
382 : SYNTAX (c) != Sword)
383
384 # define ISSPACE(c) (SYNTAX (c) == Swhitespace)
385
386 # define ISUPPER(c) (UPPERCASEP (c))
387
388 # define ISWORD(c) (SYNTAX (c) == Sword)
389
390 #else /* not emacs */
391
392 /* Jim Meyering writes:
393
394 "... Some ctype macros are valid only for character codes that
395 isascii says are ASCII (SGI's IRIX-4.0.5 is one such system --when
396 using /bin/cc or gcc but without giving an ansi option). So, all
397 ctype uses should be through macros like ISPRINT... If
398 STDC_HEADERS is defined, then autoconf has verified that the ctype
399 macros don't need to be guarded with references to isascii. ...
400 Defining isascii to 1 should let any compiler worth its salt
401 eliminate the && through constant folding."
402 Solaris defines some of these symbols so we must undefine them first. */
403
404 # undef ISASCII
405 # if defined STDC_HEADERS || (!defined isascii && !defined HAVE_ISASCII)
406 # define ISASCII(c) 1
407 # else
408 # define ISASCII(c) isascii(c)
409 # endif
410
411 /* 1 if C is an ASCII character. */
412 # define IS_REAL_ASCII(c) ((c) < 0200)
413
414 /* This distinction is not meaningful, except in Emacs. */
415 # define ISUNIBYTE(c) 1
416
417 # ifdef isblank
418 # define ISBLANK(c) (ISASCII (c) && isblank (c))
419 # else
420 # define ISBLANK(c) ((c) == ' ' || (c) == '\t')
421 # endif
422 # ifdef isgraph
423 # define ISGRAPH(c) (ISASCII (c) && isgraph (c))
424 # else
425 # define ISGRAPH(c) (ISASCII (c) && isprint (c) && !isspace (c))
426 # endif
427
428 # undef ISPRINT
429 # define ISPRINT(c) (ISASCII (c) && isprint (c))
430 # define ISDIGIT(c) (ISASCII (c) && isdigit (c))
431 # define ISALNUM(c) (ISASCII (c) && isalnum (c))
432 # define ISALPHA(c) (ISASCII (c) && isalpha (c))
433 # define ISCNTRL(c) (ISASCII (c) && iscntrl (c))
434 # define ISLOWER(c) (ISASCII (c) && islower (c))
435 # define ISPUNCT(c) (ISASCII (c) && ispunct (c))
436 # define ISSPACE(c) (ISASCII (c) && isspace (c))
437 # define ISUPPER(c) (ISASCII (c) && isupper (c))
438 # define ISXDIGIT(c) (ISASCII (c) && isxdigit (c))
439
440 # define ISWORD(c) ISALPHA(c)
441
442 # ifdef _tolower
443 # define TOLOWER(c) _tolower(c)
444 # else
445 # define TOLOWER(c) tolower(c)
446 # endif
447
448 /* How many characters in the character set. */
449 # define CHAR_SET_SIZE 256
450
451 # ifdef SYNTAX_TABLE
452
453 extern char *re_syntax_table;
454
455 # else /* not SYNTAX_TABLE */
456
457 static char re_syntax_table[CHAR_SET_SIZE];
458
459 static void
460 init_syntax_once ()
461 {
462 register int c;
463 static int done = 0;
464
465 if (done)
466 return;
467
468 bzero (re_syntax_table, sizeof re_syntax_table);
469
470 for (c = 0; c < CHAR_SET_SIZE; ++c)
471 if (ISALNUM (c))
472 re_syntax_table[c] = Sword;
473
474 re_syntax_table['_'] = Ssymbol;
475
476 done = 1;
477 }
478
479 # endif /* not SYNTAX_TABLE */
480
481 # define SYNTAX(c) re_syntax_table[(c)]
482
483 #endif /* not emacs */
484 \f
485 #ifndef NULL
486 # define NULL (void *)0
487 #endif
488
489 /* We remove any previous definition of `SIGN_EXTEND_CHAR',
490 since ours (we hope) works properly with all combinations of
491 machines, compilers, `char' and `unsigned char' argument types.
492 (Per Bothner suggested the basic approach.) */
493 #undef SIGN_EXTEND_CHAR
494 #if __STDC__
495 # define SIGN_EXTEND_CHAR(c) ((signed char) (c))
496 #else /* not __STDC__ */
497 /* As in Harbison and Steele. */
498 # define SIGN_EXTEND_CHAR(c) ((((unsigned char) (c)) ^ 128) - 128)
499 #endif
500 \f
501 /* Should we use malloc or alloca? If REGEX_MALLOC is not defined, we
502 use `alloca' instead of `malloc'. This is because using malloc in
503 re_search* or re_match* could cause memory leaks when C-g is used in
504 Emacs; also, malloc is slower and causes storage fragmentation. On
505 the other hand, malloc is more portable, and easier to debug.
506
507 Because we sometimes use alloca, some routines have to be macros,
508 not functions -- `alloca'-allocated space disappears at the end of the
509 function it is called in. */
510
511 #ifdef REGEX_MALLOC
512
513 # define REGEX_ALLOCATE malloc
514 # define REGEX_REALLOCATE(source, osize, nsize) realloc (source, nsize)
515 # define REGEX_FREE free
516
517 #else /* not REGEX_MALLOC */
518
519 /* Emacs already defines alloca, sometimes. */
520 # ifndef alloca
521
522 /* Make alloca work the best possible way. */
523 # ifdef __GNUC__
524 # define alloca __builtin_alloca
525 # else /* not __GNUC__ */
526 # ifdef HAVE_ALLOCA_H
527 # include <alloca.h>
528 # endif /* HAVE_ALLOCA_H */
529 # endif /* not __GNUC__ */
530
531 # endif /* not alloca */
532
533 # define REGEX_ALLOCATE alloca
534
535 /* Assumes a `char *destination' variable. */
536 # define REGEX_REALLOCATE(source, osize, nsize) \
537 (destination = (char *) alloca (nsize), \
538 memcpy (destination, source, osize))
539
540 /* No need to do anything to free, after alloca. */
541 # define REGEX_FREE(arg) ((void)0) /* Do nothing! But inhibit gcc warning. */
542
543 #endif /* not REGEX_MALLOC */
544
545 /* Define how to allocate the failure stack. */
546
547 #if defined REL_ALLOC && defined REGEX_MALLOC
548
549 # define REGEX_ALLOCATE_STACK(size) \
550 r_alloc (&failure_stack_ptr, (size))
551 # define REGEX_REALLOCATE_STACK(source, osize, nsize) \
552 r_re_alloc (&failure_stack_ptr, (nsize))
553 # define REGEX_FREE_STACK(ptr) \
554 r_alloc_free (&failure_stack_ptr)
555
556 #else /* not using relocating allocator */
557
558 # ifdef REGEX_MALLOC
559
560 # define REGEX_ALLOCATE_STACK malloc
561 # define REGEX_REALLOCATE_STACK(source, osize, nsize) realloc (source, nsize)
562 # define REGEX_FREE_STACK free
563
564 # else /* not REGEX_MALLOC */
565
566 # define REGEX_ALLOCATE_STACK alloca
567
568 # define REGEX_REALLOCATE_STACK(source, osize, nsize) \
569 REGEX_REALLOCATE (source, osize, nsize)
570 /* No need to explicitly free anything. */
571 # define REGEX_FREE_STACK(arg) ((void)0)
572
573 # endif /* not REGEX_MALLOC */
574 #endif /* not using relocating allocator */
575
576
577 /* True if `size1' is non-NULL and PTR is pointing anywhere inside
578 `string1' or just past its end. This works if PTR is NULL, which is
579 a good thing. */
580 #define FIRST_STRING_P(ptr) \
581 (size1 && string1 <= (ptr) && (ptr) <= string1 + size1)
582
583 /* (Re)Allocate N items of type T using malloc, or fail. */
584 #define TALLOC(n, t) ((t *) malloc ((n) * sizeof (t)))
585 #define RETALLOC(addr, n, t) ((addr) = (t *) realloc (addr, (n) * sizeof (t)))
586 #define RETALLOC_IF(addr, n, t) \
587 if (addr) RETALLOC((addr), (n), t); else (addr) = TALLOC ((n), t)
588 #define REGEX_TALLOC(n, t) ((t *) REGEX_ALLOCATE ((n) * sizeof (t)))
589
590 #define BYTEWIDTH 8 /* In bits. */
591
592 #define STREQ(s1, s2) ((strcmp (s1, s2) == 0))
593
594 #undef MAX
595 #undef MIN
596 #define MAX(a, b) ((a) > (b) ? (a) : (b))
597 #define MIN(a, b) ((a) < (b) ? (a) : (b))
598
599 /* Type of source-pattern and string chars. */
600 typedef const unsigned char re_char;
601
602 typedef char boolean;
603 #define false 0
604 #define true 1
605
606 static int re_match_2_internal _RE_ARGS ((struct re_pattern_buffer *bufp,
607 re_char *string1, int size1,
608 re_char *string2, int size2,
609 int pos,
610 struct re_registers *regs,
611 int stop));
612 \f
613 /* These are the command codes that appear in compiled regular
614 expressions. Some opcodes are followed by argument bytes. A
615 command code can specify any interpretation whatsoever for its
616 arguments. Zero bytes may appear in the compiled regular expression. */
617
618 typedef enum
619 {
620 no_op = 0,
621
622 /* Succeed right away--no more backtracking. */
623 succeed,
624
625 /* Followed by one byte giving n, then by n literal bytes. */
626 exactn,
627
628 /* Matches any (more or less) character. */
629 anychar,
630
631 /* Matches any one char belonging to specified set. First
632 following byte is number of bitmap bytes. Then come bytes
633 for a bitmap saying which chars are in. Bits in each byte
634 are ordered low-bit-first. A character is in the set if its
635 bit is 1. A character too large to have a bit in the map is
636 automatically not in the set.
637
638 If the length byte has the 0x80 bit set, then that stuff
639 is followed by a range table:
640 2 bytes of flags for character sets (low 8 bits, high 8 bits)
641 See RANGE_TABLE_WORK_BITS below.
642 2 bytes, the number of pairs that follow (upto 32767)
643 pairs, each 2 multibyte characters,
644 each multibyte character represented as 3 bytes. */
645 charset,
646
647 /* Same parameters as charset, but match any character that is
648 not one of those specified. */
649 charset_not,
650
651 /* Start remembering the text that is matched, for storing in a
652 register. Followed by one byte with the register number, in
653 the range 0 to one less than the pattern buffer's re_nsub
654 field. */
655 start_memory,
656
657 /* Stop remembering the text that is matched and store it in a
658 memory register. Followed by one byte with the register
659 number, in the range 0 to one less than `re_nsub' in the
660 pattern buffer. */
661 stop_memory,
662
663 /* Match a duplicate of something remembered. Followed by one
664 byte containing the register number. */
665 duplicate,
666
667 /* Fail unless at beginning of line. */
668 begline,
669
670 /* Fail unless at end of line. */
671 endline,
672
673 /* Succeeds if at beginning of buffer (if emacs) or at beginning
674 of string to be matched (if not). */
675 begbuf,
676
677 /* Analogously, for end of buffer/string. */
678 endbuf,
679
680 /* Followed by two byte relative address to which to jump. */
681 jump,
682
683 /* Followed by two-byte relative address of place to resume at
684 in case of failure. */
685 on_failure_jump,
686
687 /* Like on_failure_jump, but pushes a placeholder instead of the
688 current string position when executed. */
689 on_failure_keep_string_jump,
690
691 /* Just like `on_failure_jump', except that it checks that we
692 don't get stuck in an infinite loop (matching an empty string
693 indefinitely). */
694 on_failure_jump_loop,
695
696 /* Just like `on_failure_jump_loop', except that it checks for
697 a different kind of loop (the kind that shows up with non-greedy
698 operators). This operation has to be immediately preceded
699 by a `no_op'. */
700 on_failure_jump_nastyloop,
701
702 /* A smart `on_failure_jump' used for greedy * and + operators.
703 It analyses the loop before which it is put and if the
704 loop does not require backtracking, it changes itself to
705 `on_failure_keep_string_jump' and short-circuits the loop,
706 else it just defaults to changing itself into `on_failure_jump'.
707 It assumes that it is pointing to just past a `jump'. */
708 on_failure_jump_smart,
709
710 /* Followed by two-byte relative address and two-byte number n.
711 After matching N times, jump to the address upon failure.
712 Does not work if N starts at 0: use on_failure_jump_loop
713 instead. */
714 succeed_n,
715
716 /* Followed by two-byte relative address, and two-byte number n.
717 Jump to the address N times, then fail. */
718 jump_n,
719
720 /* Set the following two-byte relative address to the
721 subsequent two-byte number. The address *includes* the two
722 bytes of number. */
723 set_number_at,
724
725 wordbeg, /* Succeeds if at word beginning. */
726 wordend, /* Succeeds if at word end. */
727
728 wordbound, /* Succeeds if at a word boundary. */
729 notwordbound, /* Succeeds if not at a word boundary. */
730
731 symbeg, /* Succeeds if at symbol beginning. */
732 symend, /* Succeeds if at symbol end. */
733
734 /* Matches any character whose syntax is specified. Followed by
735 a byte which contains a syntax code, e.g., Sword. */
736 syntaxspec,
737
738 /* Matches any character whose syntax is not that specified. */
739 notsyntaxspec
740
741 #ifdef emacs
742 ,before_dot, /* Succeeds if before point. */
743 at_dot, /* Succeeds if at point. */
744 after_dot, /* Succeeds if after point. */
745
746 /* Matches any character whose category-set contains the specified
747 category. The operator is followed by a byte which contains a
748 category code (mnemonic ASCII character). */
749 categoryspec,
750
751 /* Matches any character whose category-set does not contain the
752 specified category. The operator is followed by a byte which
753 contains the category code (mnemonic ASCII character). */
754 notcategoryspec
755 #endif /* emacs */
756 } re_opcode_t;
757 \f
758 /* Common operations on the compiled pattern. */
759
760 /* Store NUMBER in two contiguous bytes starting at DESTINATION. */
761
762 #define STORE_NUMBER(destination, number) \
763 do { \
764 (destination)[0] = (number) & 0377; \
765 (destination)[1] = (number) >> 8; \
766 } while (0)
767
768 /* Same as STORE_NUMBER, except increment DESTINATION to
769 the byte after where the number is stored. Therefore, DESTINATION
770 must be an lvalue. */
771
772 #define STORE_NUMBER_AND_INCR(destination, number) \
773 do { \
774 STORE_NUMBER (destination, number); \
775 (destination) += 2; \
776 } while (0)
777
778 /* Put into DESTINATION a number stored in two contiguous bytes starting
779 at SOURCE. */
780
781 #define EXTRACT_NUMBER(destination, source) \
782 do { \
783 (destination) = *(source) & 0377; \
784 (destination) += SIGN_EXTEND_CHAR (*((source) + 1)) << 8; \
785 } while (0)
786
787 #ifdef DEBUG
788 static void extract_number _RE_ARGS ((int *dest, re_char *source));
789 static void
790 extract_number (dest, source)
791 int *dest;
792 re_char *source;
793 {
794 int temp = SIGN_EXTEND_CHAR (*(source + 1));
795 *dest = *source & 0377;
796 *dest += temp << 8;
797 }
798
799 # ifndef EXTRACT_MACROS /* To debug the macros. */
800 # undef EXTRACT_NUMBER
801 # define EXTRACT_NUMBER(dest, src) extract_number (&dest, src)
802 # endif /* not EXTRACT_MACROS */
803
804 #endif /* DEBUG */
805
806 /* Same as EXTRACT_NUMBER, except increment SOURCE to after the number.
807 SOURCE must be an lvalue. */
808
809 #define EXTRACT_NUMBER_AND_INCR(destination, source) \
810 do { \
811 EXTRACT_NUMBER (destination, source); \
812 (source) += 2; \
813 } while (0)
814
815 #ifdef DEBUG
816 static void extract_number_and_incr _RE_ARGS ((int *destination,
817 re_char **source));
818 static void
819 extract_number_and_incr (destination, source)
820 int *destination;
821 re_char **source;
822 {
823 extract_number (destination, *source);
824 *source += 2;
825 }
826
827 # ifndef EXTRACT_MACROS
828 # undef EXTRACT_NUMBER_AND_INCR
829 # define EXTRACT_NUMBER_AND_INCR(dest, src) \
830 extract_number_and_incr (&dest, &src)
831 # endif /* not EXTRACT_MACROS */
832
833 #endif /* DEBUG */
834 \f
835 /* Store a multibyte character in three contiguous bytes starting
836 DESTINATION, and increment DESTINATION to the byte after where the
837 character is stored. Therefore, DESTINATION must be an lvalue. */
838
839 #define STORE_CHARACTER_AND_INCR(destination, character) \
840 do { \
841 (destination)[0] = (character) & 0377; \
842 (destination)[1] = ((character) >> 8) & 0377; \
843 (destination)[2] = (character) >> 16; \
844 (destination) += 3; \
845 } while (0)
846
847 /* Put into DESTINATION a character stored in three contiguous bytes
848 starting at SOURCE. */
849
850 #define EXTRACT_CHARACTER(destination, source) \
851 do { \
852 (destination) = ((source)[0] \
853 | ((source)[1] << 8) \
854 | ((source)[2] << 16)); \
855 } while (0)
856
857
858 /* Macros for charset. */
859
860 /* Size of bitmap of charset P in bytes. P is a start of charset,
861 i.e. *P is (re_opcode_t) charset or (re_opcode_t) charset_not. */
862 #define CHARSET_BITMAP_SIZE(p) ((p)[1] & 0x7F)
863
864 /* Nonzero if charset P has range table. */
865 #define CHARSET_RANGE_TABLE_EXISTS_P(p) ((p)[1] & 0x80)
866
867 /* Return the address of range table of charset P. But not the start
868 of table itself, but the before where the number of ranges is
869 stored. `2 +' means to skip re_opcode_t and size of bitmap,
870 and the 2 bytes of flags at the start of the range table. */
871 #define CHARSET_RANGE_TABLE(p) (&(p)[4 + CHARSET_BITMAP_SIZE (p)])
872
873 /* Extract the bit flags that start a range table. */
874 #define CHARSET_RANGE_TABLE_BITS(p) \
875 ((p)[2 + CHARSET_BITMAP_SIZE (p)] \
876 + (p)[3 + CHARSET_BITMAP_SIZE (p)] * 0x100)
877
878 /* Test if C is listed in the bitmap of charset P. */
879 #define CHARSET_LOOKUP_BITMAP(p, c) \
880 ((c) < CHARSET_BITMAP_SIZE (p) * BYTEWIDTH \
881 && (p)[2 + (c) / BYTEWIDTH] & (1 << ((c) % BYTEWIDTH)))
882
883 /* Return the address of end of RANGE_TABLE. COUNT is number of
884 ranges (which is a pair of (start, end)) in the RANGE_TABLE. `* 2'
885 is start of range and end of range. `* 3' is size of each start
886 and end. */
887 #define CHARSET_RANGE_TABLE_END(range_table, count) \
888 ((range_table) + (count) * 2 * 3)
889
890 /* Test if C is in RANGE_TABLE. A flag NOT is negated if C is in.
891 COUNT is number of ranges in RANGE_TABLE. */
892 #define CHARSET_LOOKUP_RANGE_TABLE_RAW(not, c, range_table, count) \
893 do \
894 { \
895 re_wchar_t range_start, range_end; \
896 re_char *p; \
897 re_char *range_table_end \
898 = CHARSET_RANGE_TABLE_END ((range_table), (count)); \
899 \
900 for (p = (range_table); p < range_table_end; p += 2 * 3) \
901 { \
902 EXTRACT_CHARACTER (range_start, p); \
903 EXTRACT_CHARACTER (range_end, p + 3); \
904 \
905 if (range_start <= (c) && (c) <= range_end) \
906 { \
907 (not) = !(not); \
908 break; \
909 } \
910 } \
911 } \
912 while (0)
913
914 /* Test if C is in range table of CHARSET. The flag NOT is negated if
915 C is listed in it. */
916 #define CHARSET_LOOKUP_RANGE_TABLE(not, c, charset) \
917 do \
918 { \
919 /* Number of ranges in range table. */ \
920 int count; \
921 re_char *range_table = CHARSET_RANGE_TABLE (charset); \
922 \
923 EXTRACT_NUMBER_AND_INCR (count, range_table); \
924 CHARSET_LOOKUP_RANGE_TABLE_RAW ((not), (c), range_table, count); \
925 } \
926 while (0)
927 \f
928 /* If DEBUG is defined, Regex prints many voluminous messages about what
929 it is doing (if the variable `debug' is nonzero). If linked with the
930 main program in `iregex.c', you can enter patterns and strings
931 interactively. And if linked with the main program in `main.c' and
932 the other test files, you can run the already-written tests. */
933
934 #ifdef DEBUG
935
936 /* We use standard I/O for debugging. */
937 # include <stdio.h>
938
939 /* It is useful to test things that ``must'' be true when debugging. */
940 # include <assert.h>
941
942 static int debug = -100000;
943
944 # define DEBUG_STATEMENT(e) e
945 # define DEBUG_PRINT1(x) if (debug > 0) printf (x)
946 # define DEBUG_PRINT2(x1, x2) if (debug > 0) printf (x1, x2)
947 # define DEBUG_PRINT3(x1, x2, x3) if (debug > 0) printf (x1, x2, x3)
948 # define DEBUG_PRINT4(x1, x2, x3, x4) if (debug > 0) printf (x1, x2, x3, x4)
949 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) \
950 if (debug > 0) print_partial_compiled_pattern (s, e)
951 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) \
952 if (debug > 0) print_double_string (w, s1, sz1, s2, sz2)
953
954
955 /* Print the fastmap in human-readable form. */
956
957 void
958 print_fastmap (fastmap)
959 char *fastmap;
960 {
961 unsigned was_a_range = 0;
962 unsigned i = 0;
963
964 while (i < (1 << BYTEWIDTH))
965 {
966 if (fastmap[i++])
967 {
968 was_a_range = 0;
969 putchar (i - 1);
970 while (i < (1 << BYTEWIDTH) && fastmap[i])
971 {
972 was_a_range = 1;
973 i++;
974 }
975 if (was_a_range)
976 {
977 printf ("-");
978 putchar (i - 1);
979 }
980 }
981 }
982 putchar ('\n');
983 }
984
985
986 /* Print a compiled pattern string in human-readable form, starting at
987 the START pointer into it and ending just before the pointer END. */
988
989 void
990 print_partial_compiled_pattern (start, end)
991 re_char *start;
992 re_char *end;
993 {
994 int mcnt, mcnt2;
995 re_char *p = start;
996 re_char *pend = end;
997
998 if (start == NULL)
999 {
1000 fprintf (stderr, "(null)\n");
1001 return;
1002 }
1003
1004 /* Loop over pattern commands. */
1005 while (p < pend)
1006 {
1007 fprintf (stderr, "%d:\t", p - start);
1008
1009 switch ((re_opcode_t) *p++)
1010 {
1011 case no_op:
1012 fprintf (stderr, "/no_op");
1013 break;
1014
1015 case succeed:
1016 fprintf (stderr, "/succeed");
1017 break;
1018
1019 case exactn:
1020 mcnt = *p++;
1021 fprintf (stderr, "/exactn/%d", mcnt);
1022 do
1023 {
1024 fprintf (stderr, "/%c", *p++);
1025 }
1026 while (--mcnt);
1027 break;
1028
1029 case start_memory:
1030 fprintf (stderr, "/start_memory/%d", *p++);
1031 break;
1032
1033 case stop_memory:
1034 fprintf (stderr, "/stop_memory/%d", *p++);
1035 break;
1036
1037 case duplicate:
1038 fprintf (stderr, "/duplicate/%d", *p++);
1039 break;
1040
1041 case anychar:
1042 fprintf (stderr, "/anychar");
1043 break;
1044
1045 case charset:
1046 case charset_not:
1047 {
1048 register int c, last = -100;
1049 register int in_range = 0;
1050 int length = CHARSET_BITMAP_SIZE (p - 1);
1051 int has_range_table = CHARSET_RANGE_TABLE_EXISTS_P (p - 1);
1052
1053 fprintf (stderr, "/charset [%s",
1054 (re_opcode_t) *(p - 1) == charset_not ? "^" : "");
1055
1056 if (p + *p >= pend)
1057 fprintf (stderr, " !extends past end of pattern! ");
1058
1059 for (c = 0; c < 256; c++)
1060 if (c / 8 < length
1061 && (p[1 + (c/8)] & (1 << (c % 8))))
1062 {
1063 /* Are we starting a range? */
1064 if (last + 1 == c && ! in_range)
1065 {
1066 fprintf (stderr, "-");
1067 in_range = 1;
1068 }
1069 /* Have we broken a range? */
1070 else if (last + 1 != c && in_range)
1071 {
1072 fprintf (stderr, "%c", last);
1073 in_range = 0;
1074 }
1075
1076 if (! in_range)
1077 fprintf (stderr, "%c", c);
1078
1079 last = c;
1080 }
1081
1082 if (in_range)
1083 fprintf (stderr, "%c", last);
1084
1085 fprintf (stderr, "]");
1086
1087 p += 1 + length;
1088
1089 if (has_range_table)
1090 {
1091 int count;
1092 fprintf (stderr, "has-range-table");
1093
1094 /* ??? Should print the range table; for now, just skip it. */
1095 p += 2; /* skip range table bits */
1096 EXTRACT_NUMBER_AND_INCR (count, p);
1097 p = CHARSET_RANGE_TABLE_END (p, count);
1098 }
1099 }
1100 break;
1101
1102 case begline:
1103 fprintf (stderr, "/begline");
1104 break;
1105
1106 case endline:
1107 fprintf (stderr, "/endline");
1108 break;
1109
1110 case on_failure_jump:
1111 extract_number_and_incr (&mcnt, &p);
1112 fprintf (stderr, "/on_failure_jump to %d", p + mcnt - start);
1113 break;
1114
1115 case on_failure_keep_string_jump:
1116 extract_number_and_incr (&mcnt, &p);
1117 fprintf (stderr, "/on_failure_keep_string_jump to %d", p + mcnt - start);
1118 break;
1119
1120 case on_failure_jump_nastyloop:
1121 extract_number_and_incr (&mcnt, &p);
1122 fprintf (stderr, "/on_failure_jump_nastyloop to %d", p + mcnt - start);
1123 break;
1124
1125 case on_failure_jump_loop:
1126 extract_number_and_incr (&mcnt, &p);
1127 fprintf (stderr, "/on_failure_jump_loop to %d", p + mcnt - start);
1128 break;
1129
1130 case on_failure_jump_smart:
1131 extract_number_and_incr (&mcnt, &p);
1132 fprintf (stderr, "/on_failure_jump_smart to %d", p + mcnt - start);
1133 break;
1134
1135 case jump:
1136 extract_number_and_incr (&mcnt, &p);
1137 fprintf (stderr, "/jump to %d", p + mcnt - start);
1138 break;
1139
1140 case succeed_n:
1141 extract_number_and_incr (&mcnt, &p);
1142 extract_number_and_incr (&mcnt2, &p);
1143 fprintf (stderr, "/succeed_n to %d, %d times", p - 2 + mcnt - start, mcnt2);
1144 break;
1145
1146 case jump_n:
1147 extract_number_and_incr (&mcnt, &p);
1148 extract_number_and_incr (&mcnt2, &p);
1149 fprintf (stderr, "/jump_n to %d, %d times", p - 2 + mcnt - start, mcnt2);
1150 break;
1151
1152 case set_number_at:
1153 extract_number_and_incr (&mcnt, &p);
1154 extract_number_and_incr (&mcnt2, &p);
1155 fprintf (stderr, "/set_number_at location %d to %d", p - 2 + mcnt - start, mcnt2);
1156 break;
1157
1158 case wordbound:
1159 fprintf (stderr, "/wordbound");
1160 break;
1161
1162 case notwordbound:
1163 fprintf (stderr, "/notwordbound");
1164 break;
1165
1166 case wordbeg:
1167 fprintf (stderr, "/wordbeg");
1168 break;
1169
1170 case wordend:
1171 fprintf (stderr, "/wordend");
1172 break;
1173
1174 case symbeg:
1175 fprintf (stderr, "/symbeg");
1176 break;
1177
1178 case symend:
1179 fprintf (stderr, "/symend");
1180 break;
1181
1182 case syntaxspec:
1183 fprintf (stderr, "/syntaxspec");
1184 mcnt = *p++;
1185 fprintf (stderr, "/%d", mcnt);
1186 break;
1187
1188 case notsyntaxspec:
1189 fprintf (stderr, "/notsyntaxspec");
1190 mcnt = *p++;
1191 fprintf (stderr, "/%d", mcnt);
1192 break;
1193
1194 # ifdef emacs
1195 case before_dot:
1196 fprintf (stderr, "/before_dot");
1197 break;
1198
1199 case at_dot:
1200 fprintf (stderr, "/at_dot");
1201 break;
1202
1203 case after_dot:
1204 fprintf (stderr, "/after_dot");
1205 break;
1206
1207 case categoryspec:
1208 fprintf (stderr, "/categoryspec");
1209 mcnt = *p++;
1210 fprintf (stderr, "/%d", mcnt);
1211 break;
1212
1213 case notcategoryspec:
1214 fprintf (stderr, "/notcategoryspec");
1215 mcnt = *p++;
1216 fprintf (stderr, "/%d", mcnt);
1217 break;
1218 # endif /* emacs */
1219
1220 case begbuf:
1221 fprintf (stderr, "/begbuf");
1222 break;
1223
1224 case endbuf:
1225 fprintf (stderr, "/endbuf");
1226 break;
1227
1228 default:
1229 fprintf (stderr, "?%d", *(p-1));
1230 }
1231
1232 fprintf (stderr, "\n");
1233 }
1234
1235 fprintf (stderr, "%d:\tend of pattern.\n", p - start);
1236 }
1237
1238
1239 void
1240 print_compiled_pattern (bufp)
1241 struct re_pattern_buffer *bufp;
1242 {
1243 re_char *buffer = bufp->buffer;
1244
1245 print_partial_compiled_pattern (buffer, buffer + bufp->used);
1246 printf ("%ld bytes used/%ld bytes allocated.\n",
1247 bufp->used, bufp->allocated);
1248
1249 if (bufp->fastmap_accurate && bufp->fastmap)
1250 {
1251 printf ("fastmap: ");
1252 print_fastmap (bufp->fastmap);
1253 }
1254
1255 printf ("re_nsub: %d\t", bufp->re_nsub);
1256 printf ("regs_alloc: %d\t", bufp->regs_allocated);
1257 printf ("can_be_null: %d\t", bufp->can_be_null);
1258 printf ("no_sub: %d\t", bufp->no_sub);
1259 printf ("not_bol: %d\t", bufp->not_bol);
1260 printf ("not_eol: %d\t", bufp->not_eol);
1261 printf ("syntax: %lx\n", bufp->syntax);
1262 fflush (stdout);
1263 /* Perhaps we should print the translate table? */
1264 }
1265
1266
1267 void
1268 print_double_string (where, string1, size1, string2, size2)
1269 re_char *where;
1270 re_char *string1;
1271 re_char *string2;
1272 int size1;
1273 int size2;
1274 {
1275 int this_char;
1276
1277 if (where == NULL)
1278 printf ("(null)");
1279 else
1280 {
1281 if (FIRST_STRING_P (where))
1282 {
1283 for (this_char = where - string1; this_char < size1; this_char++)
1284 putchar (string1[this_char]);
1285
1286 where = string2;
1287 }
1288
1289 for (this_char = where - string2; this_char < size2; this_char++)
1290 putchar (string2[this_char]);
1291 }
1292 }
1293
1294 #else /* not DEBUG */
1295
1296 # undef assert
1297 # define assert(e)
1298
1299 # define DEBUG_STATEMENT(e)
1300 # define DEBUG_PRINT1(x)
1301 # define DEBUG_PRINT2(x1, x2)
1302 # define DEBUG_PRINT3(x1, x2, x3)
1303 # define DEBUG_PRINT4(x1, x2, x3, x4)
1304 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e)
1305 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2)
1306
1307 #endif /* not DEBUG */
1308 \f
1309 /* Set by `re_set_syntax' to the current regexp syntax to recognize. Can
1310 also be assigned to arbitrarily: each pattern buffer stores its own
1311 syntax, so it can be changed between regex compilations. */
1312 /* This has no initializer because initialized variables in Emacs
1313 become read-only after dumping. */
1314 reg_syntax_t re_syntax_options;
1315
1316
1317 /* Specify the precise syntax of regexps for compilation. This provides
1318 for compatibility for various utilities which historically have
1319 different, incompatible syntaxes.
1320
1321 The argument SYNTAX is a bit mask comprised of the various bits
1322 defined in regex.h. We return the old syntax. */
1323
1324 reg_syntax_t
1325 re_set_syntax (syntax)
1326 reg_syntax_t syntax;
1327 {
1328 reg_syntax_t ret = re_syntax_options;
1329
1330 re_syntax_options = syntax;
1331 return ret;
1332 }
1333 WEAK_ALIAS (__re_set_syntax, re_set_syntax)
1334
1335 /* Regexp to use to replace spaces, or NULL meaning don't. */
1336 static re_char *whitespace_regexp;
1337
1338 void
1339 re_set_whitespace_regexp (regexp)
1340 const char *regexp;
1341 {
1342 whitespace_regexp = (re_char *) regexp;
1343 }
1344 WEAK_ALIAS (__re_set_syntax, re_set_syntax)
1345 \f
1346 /* This table gives an error message for each of the error codes listed
1347 in regex.h. Obviously the order here has to be same as there.
1348 POSIX doesn't require that we do anything for REG_NOERROR,
1349 but why not be nice? */
1350
1351 static const char *re_error_msgid[] =
1352 {
1353 gettext_noop ("Success"), /* REG_NOERROR */
1354 gettext_noop ("No match"), /* REG_NOMATCH */
1355 gettext_noop ("Invalid regular expression"), /* REG_BADPAT */
1356 gettext_noop ("Invalid collation character"), /* REG_ECOLLATE */
1357 gettext_noop ("Invalid character class name"), /* REG_ECTYPE */
1358 gettext_noop ("Trailing backslash"), /* REG_EESCAPE */
1359 gettext_noop ("Invalid back reference"), /* REG_ESUBREG */
1360 gettext_noop ("Unmatched [ or [^"), /* REG_EBRACK */
1361 gettext_noop ("Unmatched ( or \\("), /* REG_EPAREN */
1362 gettext_noop ("Unmatched \\{"), /* REG_EBRACE */
1363 gettext_noop ("Invalid content of \\{\\}"), /* REG_BADBR */
1364 gettext_noop ("Invalid range end"), /* REG_ERANGE */
1365 gettext_noop ("Memory exhausted"), /* REG_ESPACE */
1366 gettext_noop ("Invalid preceding regular expression"), /* REG_BADRPT */
1367 gettext_noop ("Premature end of regular expression"), /* REG_EEND */
1368 gettext_noop ("Regular expression too big"), /* REG_ESIZE */
1369 gettext_noop ("Unmatched ) or \\)"), /* REG_ERPAREN */
1370 gettext_noop ("Range striding over charsets") /* REG_ERANGEX */
1371 };
1372 \f
1373 /* Avoiding alloca during matching, to placate r_alloc. */
1374
1375 /* Define MATCH_MAY_ALLOCATE unless we need to make sure that the
1376 searching and matching functions should not call alloca. On some
1377 systems, alloca is implemented in terms of malloc, and if we're
1378 using the relocating allocator routines, then malloc could cause a
1379 relocation, which might (if the strings being searched are in the
1380 ralloc heap) shift the data out from underneath the regexp
1381 routines.
1382
1383 Here's another reason to avoid allocation: Emacs
1384 processes input from X in a signal handler; processing X input may
1385 call malloc; if input arrives while a matching routine is calling
1386 malloc, then we're scrod. But Emacs can't just block input while
1387 calling matching routines; then we don't notice interrupts when
1388 they come in. So, Emacs blocks input around all regexp calls
1389 except the matching calls, which it leaves unprotected, in the
1390 faith that they will not malloc. */
1391
1392 /* Normally, this is fine. */
1393 #define MATCH_MAY_ALLOCATE
1394
1395 /* The match routines may not allocate if (1) they would do it with malloc
1396 and (2) it's not safe for them to use malloc.
1397 Note that if REL_ALLOC is defined, matching would not use malloc for the
1398 failure stack, but we would still use it for the register vectors;
1399 so REL_ALLOC should not affect this. */
1400 #if defined REGEX_MALLOC && defined emacs
1401 # undef MATCH_MAY_ALLOCATE
1402 #endif
1403
1404 \f
1405 /* Failure stack declarations and macros; both re_compile_fastmap and
1406 re_match_2 use a failure stack. These have to be macros because of
1407 REGEX_ALLOCATE_STACK. */
1408
1409
1410 /* Approximate number of failure points for which to initially allocate space
1411 when matching. If this number is exceeded, we allocate more
1412 space, so it is not a hard limit. */
1413 #ifndef INIT_FAILURE_ALLOC
1414 # define INIT_FAILURE_ALLOC 20
1415 #endif
1416
1417 /* Roughly the maximum number of failure points on the stack. Would be
1418 exactly that if always used TYPICAL_FAILURE_SIZE items each time we failed.
1419 This is a variable only so users of regex can assign to it; we never
1420 change it ourselves. We always multiply it by TYPICAL_FAILURE_SIZE
1421 before using it, so it should probably be a byte-count instead. */
1422 # if defined MATCH_MAY_ALLOCATE
1423 /* Note that 4400 was enough to cause a crash on Alpha OSF/1,
1424 whose default stack limit is 2mb. In order for a larger
1425 value to work reliably, you have to try to make it accord
1426 with the process stack limit. */
1427 size_t re_max_failures = 40000;
1428 # else
1429 size_t re_max_failures = 4000;
1430 # endif
1431
1432 union fail_stack_elt
1433 {
1434 re_char *pointer;
1435 /* This should be the biggest `int' that's no bigger than a pointer. */
1436 long integer;
1437 };
1438
1439 typedef union fail_stack_elt fail_stack_elt_t;
1440
1441 typedef struct
1442 {
1443 fail_stack_elt_t *stack;
1444 size_t size;
1445 size_t avail; /* Offset of next open position. */
1446 size_t frame; /* Offset of the cur constructed frame. */
1447 } fail_stack_type;
1448
1449 #define FAIL_STACK_EMPTY() (fail_stack.frame == 0)
1450 #define FAIL_STACK_FULL() (fail_stack.avail == fail_stack.size)
1451
1452
1453 /* Define macros to initialize and free the failure stack.
1454 Do `return -2' if the alloc fails. */
1455
1456 #ifdef MATCH_MAY_ALLOCATE
1457 # define INIT_FAIL_STACK() \
1458 do { \
1459 fail_stack.stack = (fail_stack_elt_t *) \
1460 REGEX_ALLOCATE_STACK (INIT_FAILURE_ALLOC * TYPICAL_FAILURE_SIZE \
1461 * sizeof (fail_stack_elt_t)); \
1462 \
1463 if (fail_stack.stack == NULL) \
1464 return -2; \
1465 \
1466 fail_stack.size = INIT_FAILURE_ALLOC; \
1467 fail_stack.avail = 0; \
1468 fail_stack.frame = 0; \
1469 } while (0)
1470
1471 # define RESET_FAIL_STACK() REGEX_FREE_STACK (fail_stack.stack)
1472 #else
1473 # define INIT_FAIL_STACK() \
1474 do { \
1475 fail_stack.avail = 0; \
1476 fail_stack.frame = 0; \
1477 } while (0)
1478
1479 # define RESET_FAIL_STACK() ((void)0)
1480 #endif
1481
1482
1483 /* Double the size of FAIL_STACK, up to a limit
1484 which allows approximately `re_max_failures' items.
1485
1486 Return 1 if succeeds, and 0 if either ran out of memory
1487 allocating space for it or it was already too large.
1488
1489 REGEX_REALLOCATE_STACK requires `destination' be declared. */
1490
1491 /* Factor to increase the failure stack size by
1492 when we increase it.
1493 This used to be 2, but 2 was too wasteful
1494 because the old discarded stacks added up to as much space
1495 were as ultimate, maximum-size stack. */
1496 #define FAIL_STACK_GROWTH_FACTOR 4
1497
1498 #define GROW_FAIL_STACK(fail_stack) \
1499 (((fail_stack).size * sizeof (fail_stack_elt_t) \
1500 >= re_max_failures * TYPICAL_FAILURE_SIZE) \
1501 ? 0 \
1502 : ((fail_stack).stack \
1503 = (fail_stack_elt_t *) \
1504 REGEX_REALLOCATE_STACK ((fail_stack).stack, \
1505 (fail_stack).size * sizeof (fail_stack_elt_t), \
1506 MIN (re_max_failures * TYPICAL_FAILURE_SIZE, \
1507 ((fail_stack).size * sizeof (fail_stack_elt_t) \
1508 * FAIL_STACK_GROWTH_FACTOR))), \
1509 \
1510 (fail_stack).stack == NULL \
1511 ? 0 \
1512 : ((fail_stack).size \
1513 = (MIN (re_max_failures * TYPICAL_FAILURE_SIZE, \
1514 ((fail_stack).size * sizeof (fail_stack_elt_t) \
1515 * FAIL_STACK_GROWTH_FACTOR)) \
1516 / sizeof (fail_stack_elt_t)), \
1517 1)))
1518
1519
1520 /* Push a pointer value onto the failure stack.
1521 Assumes the variable `fail_stack'. Probably should only
1522 be called from within `PUSH_FAILURE_POINT'. */
1523 #define PUSH_FAILURE_POINTER(item) \
1524 fail_stack.stack[fail_stack.avail++].pointer = (item)
1525
1526 /* This pushes an integer-valued item onto the failure stack.
1527 Assumes the variable `fail_stack'. Probably should only
1528 be called from within `PUSH_FAILURE_POINT'. */
1529 #define PUSH_FAILURE_INT(item) \
1530 fail_stack.stack[fail_stack.avail++].integer = (item)
1531
1532 /* Push a fail_stack_elt_t value onto the failure stack.
1533 Assumes the variable `fail_stack'. Probably should only
1534 be called from within `PUSH_FAILURE_POINT'. */
1535 #define PUSH_FAILURE_ELT(item) \
1536 fail_stack.stack[fail_stack.avail++] = (item)
1537
1538 /* These three POP... operations complement the three PUSH... operations.
1539 All assume that `fail_stack' is nonempty. */
1540 #define POP_FAILURE_POINTER() fail_stack.stack[--fail_stack.avail].pointer
1541 #define POP_FAILURE_INT() fail_stack.stack[--fail_stack.avail].integer
1542 #define POP_FAILURE_ELT() fail_stack.stack[--fail_stack.avail]
1543
1544 /* Individual items aside from the registers. */
1545 #define NUM_NONREG_ITEMS 3
1546
1547 /* Used to examine the stack (to detect infinite loops). */
1548 #define FAILURE_PAT(h) fail_stack.stack[(h) - 1].pointer
1549 #define FAILURE_STR(h) (fail_stack.stack[(h) - 2].pointer)
1550 #define NEXT_FAILURE_HANDLE(h) fail_stack.stack[(h) - 3].integer
1551 #define TOP_FAILURE_HANDLE() fail_stack.frame
1552
1553
1554 #define ENSURE_FAIL_STACK(space) \
1555 while (REMAINING_AVAIL_SLOTS <= space) { \
1556 if (!GROW_FAIL_STACK (fail_stack)) \
1557 return -2; \
1558 DEBUG_PRINT2 ("\n Doubled stack; size now: %d\n", (fail_stack).size);\
1559 DEBUG_PRINT2 (" slots available: %d\n", REMAINING_AVAIL_SLOTS);\
1560 }
1561
1562 /* Push register NUM onto the stack. */
1563 #define PUSH_FAILURE_REG(num) \
1564 do { \
1565 char *destination; \
1566 ENSURE_FAIL_STACK(3); \
1567 DEBUG_PRINT4 (" Push reg %d (spanning %p -> %p)\n", \
1568 num, regstart[num], regend[num]); \
1569 PUSH_FAILURE_POINTER (regstart[num]); \
1570 PUSH_FAILURE_POINTER (regend[num]); \
1571 PUSH_FAILURE_INT (num); \
1572 } while (0)
1573
1574 /* Change the counter's value to VAL, but make sure that it will
1575 be reset when backtracking. */
1576 #define PUSH_NUMBER(ptr,val) \
1577 do { \
1578 char *destination; \
1579 int c; \
1580 ENSURE_FAIL_STACK(3); \
1581 EXTRACT_NUMBER (c, ptr); \
1582 DEBUG_PRINT4 (" Push number %p = %d -> %d\n", ptr, c, val); \
1583 PUSH_FAILURE_INT (c); \
1584 PUSH_FAILURE_POINTER (ptr); \
1585 PUSH_FAILURE_INT (-1); \
1586 STORE_NUMBER (ptr, val); \
1587 } while (0)
1588
1589 /* Pop a saved register off the stack. */
1590 #define POP_FAILURE_REG_OR_COUNT() \
1591 do { \
1592 int reg = POP_FAILURE_INT (); \
1593 if (reg == -1) \
1594 { \
1595 /* It's a counter. */ \
1596 /* Here, we discard `const', making re_match non-reentrant. */ \
1597 unsigned char *ptr = (unsigned char*) POP_FAILURE_POINTER (); \
1598 reg = POP_FAILURE_INT (); \
1599 STORE_NUMBER (ptr, reg); \
1600 DEBUG_PRINT3 (" Pop counter %p = %d\n", ptr, reg); \
1601 } \
1602 else \
1603 { \
1604 regend[reg] = POP_FAILURE_POINTER (); \
1605 regstart[reg] = POP_FAILURE_POINTER (); \
1606 DEBUG_PRINT4 (" Pop reg %d (spanning %p -> %p)\n", \
1607 reg, regstart[reg], regend[reg]); \
1608 } \
1609 } while (0)
1610
1611 /* Check that we are not stuck in an infinite loop. */
1612 #define CHECK_INFINITE_LOOP(pat_cur, string_place) \
1613 do { \
1614 int failure = TOP_FAILURE_HANDLE (); \
1615 /* Check for infinite matching loops */ \
1616 while (failure > 0 \
1617 && (FAILURE_STR (failure) == string_place \
1618 || FAILURE_STR (failure) == NULL)) \
1619 { \
1620 assert (FAILURE_PAT (failure) >= bufp->buffer \
1621 && FAILURE_PAT (failure) <= bufp->buffer + bufp->used); \
1622 if (FAILURE_PAT (failure) == pat_cur) \
1623 { \
1624 cycle = 1; \
1625 break; \
1626 } \
1627 DEBUG_PRINT2 (" Other pattern: %p\n", FAILURE_PAT (failure)); \
1628 failure = NEXT_FAILURE_HANDLE(failure); \
1629 } \
1630 DEBUG_PRINT2 (" Other string: %p\n", FAILURE_STR (failure)); \
1631 } while (0)
1632
1633 /* Push the information about the state we will need
1634 if we ever fail back to it.
1635
1636 Requires variables fail_stack, regstart, regend and
1637 num_regs be declared. GROW_FAIL_STACK requires `destination' be
1638 declared.
1639
1640 Does `return FAILURE_CODE' if runs out of memory. */
1641
1642 #define PUSH_FAILURE_POINT(pattern, string_place) \
1643 do { \
1644 char *destination; \
1645 /* Must be int, so when we don't save any registers, the arithmetic \
1646 of 0 + -1 isn't done as unsigned. */ \
1647 \
1648 DEBUG_STATEMENT (nfailure_points_pushed++); \
1649 DEBUG_PRINT1 ("\nPUSH_FAILURE_POINT:\n"); \
1650 DEBUG_PRINT2 (" Before push, next avail: %d\n", (fail_stack).avail); \
1651 DEBUG_PRINT2 (" size: %d\n", (fail_stack).size);\
1652 \
1653 ENSURE_FAIL_STACK (NUM_NONREG_ITEMS); \
1654 \
1655 DEBUG_PRINT1 ("\n"); \
1656 \
1657 DEBUG_PRINT2 (" Push frame index: %d\n", fail_stack.frame); \
1658 PUSH_FAILURE_INT (fail_stack.frame); \
1659 \
1660 DEBUG_PRINT2 (" Push string %p: `", string_place); \
1661 DEBUG_PRINT_DOUBLE_STRING (string_place, string1, size1, string2, size2);\
1662 DEBUG_PRINT1 ("'\n"); \
1663 PUSH_FAILURE_POINTER (string_place); \
1664 \
1665 DEBUG_PRINT2 (" Push pattern %p: ", pattern); \
1666 DEBUG_PRINT_COMPILED_PATTERN (bufp, pattern, pend); \
1667 PUSH_FAILURE_POINTER (pattern); \
1668 \
1669 /* Close the frame by moving the frame pointer past it. */ \
1670 fail_stack.frame = fail_stack.avail; \
1671 } while (0)
1672
1673 /* Estimate the size of data pushed by a typical failure stack entry.
1674 An estimate is all we need, because all we use this for
1675 is to choose a limit for how big to make the failure stack. */
1676 /* BEWARE, the value `20' is hard-coded in emacs.c:main(). */
1677 #define TYPICAL_FAILURE_SIZE 20
1678
1679 /* How many items can still be added to the stack without overflowing it. */
1680 #define REMAINING_AVAIL_SLOTS ((fail_stack).size - (fail_stack).avail)
1681
1682
1683 /* Pops what PUSH_FAIL_STACK pushes.
1684
1685 We restore into the parameters, all of which should be lvalues:
1686 STR -- the saved data position.
1687 PAT -- the saved pattern position.
1688 REGSTART, REGEND -- arrays of string positions.
1689
1690 Also assumes the variables `fail_stack' and (if debugging), `bufp',
1691 `pend', `string1', `size1', `string2', and `size2'. */
1692
1693 #define POP_FAILURE_POINT(str, pat) \
1694 do { \
1695 assert (!FAIL_STACK_EMPTY ()); \
1696 \
1697 /* Remove failure points and point to how many regs pushed. */ \
1698 DEBUG_PRINT1 ("POP_FAILURE_POINT:\n"); \
1699 DEBUG_PRINT2 (" Before pop, next avail: %d\n", fail_stack.avail); \
1700 DEBUG_PRINT2 (" size: %d\n", fail_stack.size); \
1701 \
1702 /* Pop the saved registers. */ \
1703 while (fail_stack.frame < fail_stack.avail) \
1704 POP_FAILURE_REG_OR_COUNT (); \
1705 \
1706 pat = POP_FAILURE_POINTER (); \
1707 DEBUG_PRINT2 (" Popping pattern %p: ", pat); \
1708 DEBUG_PRINT_COMPILED_PATTERN (bufp, pat, pend); \
1709 \
1710 /* If the saved string location is NULL, it came from an \
1711 on_failure_keep_string_jump opcode, and we want to throw away the \
1712 saved NULL, thus retaining our current position in the string. */ \
1713 str = POP_FAILURE_POINTER (); \
1714 DEBUG_PRINT2 (" Popping string %p: `", str); \
1715 DEBUG_PRINT_DOUBLE_STRING (str, string1, size1, string2, size2); \
1716 DEBUG_PRINT1 ("'\n"); \
1717 \
1718 fail_stack.frame = POP_FAILURE_INT (); \
1719 DEBUG_PRINT2 (" Popping frame index: %d\n", fail_stack.frame); \
1720 \
1721 assert (fail_stack.avail >= 0); \
1722 assert (fail_stack.frame <= fail_stack.avail); \
1723 \
1724 DEBUG_STATEMENT (nfailure_points_popped++); \
1725 } while (0) /* POP_FAILURE_POINT */
1726
1727
1728 \f
1729 /* Registers are set to a sentinel when they haven't yet matched. */
1730 #define REG_UNSET(e) ((e) == NULL)
1731 \f
1732 /* Subroutine declarations and macros for regex_compile. */
1733
1734 static reg_errcode_t regex_compile _RE_ARGS ((re_char *pattern, size_t size,
1735 reg_syntax_t syntax,
1736 struct re_pattern_buffer *bufp));
1737 static void store_op1 _RE_ARGS ((re_opcode_t op, unsigned char *loc, int arg));
1738 static void store_op2 _RE_ARGS ((re_opcode_t op, unsigned char *loc,
1739 int arg1, int arg2));
1740 static void insert_op1 _RE_ARGS ((re_opcode_t op, unsigned char *loc,
1741 int arg, unsigned char *end));
1742 static void insert_op2 _RE_ARGS ((re_opcode_t op, unsigned char *loc,
1743 int arg1, int arg2, unsigned char *end));
1744 static boolean at_begline_loc_p _RE_ARGS ((re_char *pattern,
1745 re_char *p,
1746 reg_syntax_t syntax));
1747 static boolean at_endline_loc_p _RE_ARGS ((re_char *p,
1748 re_char *pend,
1749 reg_syntax_t syntax));
1750 static re_char *skip_one_char _RE_ARGS ((re_char *p));
1751 static int analyse_first _RE_ARGS ((re_char *p, re_char *pend,
1752 char *fastmap, const int multibyte));
1753
1754 /* Fetch the next character in the uncompiled pattern, with no
1755 translation. */
1756 #define PATFETCH(c) \
1757 do { \
1758 int len; \
1759 if (p == pend) return REG_EEND; \
1760 c = RE_STRING_CHAR_AND_LENGTH (p, len, multibyte); \
1761 p += len; \
1762 } while (0)
1763
1764
1765 /* If `translate' is non-null, return translate[D], else just D. We
1766 cast the subscript to translate because some data is declared as
1767 `char *', to avoid warnings when a string constant is passed. But
1768 when we use a character as a subscript we must make it unsigned. */
1769 #ifndef TRANSLATE
1770 # define TRANSLATE(d) \
1771 (RE_TRANSLATE_P (translate) ? RE_TRANSLATE (translate, (d)) : (d))
1772 #endif
1773
1774
1775 /* Macros for outputting the compiled pattern into `buffer'. */
1776
1777 /* If the buffer isn't allocated when it comes in, use this. */
1778 #define INIT_BUF_SIZE 32
1779
1780 /* Make sure we have at least N more bytes of space in buffer. */
1781 #define GET_BUFFER_SPACE(n) \
1782 while ((size_t) (b - bufp->buffer + (n)) > bufp->allocated) \
1783 EXTEND_BUFFER ()
1784
1785 /* Make sure we have one more byte of buffer space and then add C to it. */
1786 #define BUF_PUSH(c) \
1787 do { \
1788 GET_BUFFER_SPACE (1); \
1789 *b++ = (unsigned char) (c); \
1790 } while (0)
1791
1792
1793 /* Ensure we have two more bytes of buffer space and then append C1 and C2. */
1794 #define BUF_PUSH_2(c1, c2) \
1795 do { \
1796 GET_BUFFER_SPACE (2); \
1797 *b++ = (unsigned char) (c1); \
1798 *b++ = (unsigned char) (c2); \
1799 } while (0)
1800
1801
1802 /* As with BUF_PUSH_2, except for three bytes. */
1803 #define BUF_PUSH_3(c1, c2, c3) \
1804 do { \
1805 GET_BUFFER_SPACE (3); \
1806 *b++ = (unsigned char) (c1); \
1807 *b++ = (unsigned char) (c2); \
1808 *b++ = (unsigned char) (c3); \
1809 } while (0)
1810
1811
1812 /* Store a jump with opcode OP at LOC to location TO. We store a
1813 relative address offset by the three bytes the jump itself occupies. */
1814 #define STORE_JUMP(op, loc, to) \
1815 store_op1 (op, loc, (to) - (loc) - 3)
1816
1817 /* Likewise, for a two-argument jump. */
1818 #define STORE_JUMP2(op, loc, to, arg) \
1819 store_op2 (op, loc, (to) - (loc) - 3, arg)
1820
1821 /* Like `STORE_JUMP', but for inserting. Assume `b' is the buffer end. */
1822 #define INSERT_JUMP(op, loc, to) \
1823 insert_op1 (op, loc, (to) - (loc) - 3, b)
1824
1825 /* Like `STORE_JUMP2', but for inserting. Assume `b' is the buffer end. */
1826 #define INSERT_JUMP2(op, loc, to, arg) \
1827 insert_op2 (op, loc, (to) - (loc) - 3, arg, b)
1828
1829
1830 /* This is not an arbitrary limit: the arguments which represent offsets
1831 into the pattern are two bytes long. So if 2^15 bytes turns out to
1832 be too small, many things would have to change. */
1833 # define MAX_BUF_SIZE (1L << 15)
1834
1835 #if 0 /* This is when we thought it could be 2^16 bytes. */
1836 /* Any other compiler which, like MSC, has allocation limit below 2^16
1837 bytes will have to use approach similar to what was done below for
1838 MSC and drop MAX_BUF_SIZE a bit. Otherwise you may end up
1839 reallocating to 0 bytes. Such thing is not going to work too well.
1840 You have been warned!! */
1841 #if defined _MSC_VER && !defined WIN32
1842 /* Microsoft C 16-bit versions limit malloc to approx 65512 bytes. */
1843 # define MAX_BUF_SIZE 65500L
1844 #else
1845 # define MAX_BUF_SIZE (1L << 16)
1846 #endif
1847 #endif /* 0 */
1848
1849 /* Extend the buffer by twice its current size via realloc and
1850 reset the pointers that pointed into the old block to point to the
1851 correct places in the new one. If extending the buffer results in it
1852 being larger than MAX_BUF_SIZE, then flag memory exhausted. */
1853 #if __BOUNDED_POINTERS__
1854 # define SET_HIGH_BOUND(P) (__ptrhigh (P) = __ptrlow (P) + bufp->allocated)
1855 # define MOVE_BUFFER_POINTER(P) \
1856 (__ptrlow (P) = new_buffer + (__ptrlow (P) - old_buffer), \
1857 SET_HIGH_BOUND (P), \
1858 __ptrvalue (P) = new_buffer + (__ptrvalue (P) - old_buffer))
1859 # define ELSE_EXTEND_BUFFER_HIGH_BOUND \
1860 else \
1861 { \
1862 SET_HIGH_BOUND (b); \
1863 SET_HIGH_BOUND (begalt); \
1864 if (fixup_alt_jump) \
1865 SET_HIGH_BOUND (fixup_alt_jump); \
1866 if (laststart) \
1867 SET_HIGH_BOUND (laststart); \
1868 if (pending_exact) \
1869 SET_HIGH_BOUND (pending_exact); \
1870 }
1871 #else
1872 # define MOVE_BUFFER_POINTER(P) ((P) = new_buffer + ((P) - old_buffer))
1873 # define ELSE_EXTEND_BUFFER_HIGH_BOUND
1874 #endif
1875 #define EXTEND_BUFFER() \
1876 do { \
1877 unsigned char *old_buffer = bufp->buffer; \
1878 if (bufp->allocated == MAX_BUF_SIZE) \
1879 return REG_ESIZE; \
1880 bufp->allocated <<= 1; \
1881 if (bufp->allocated > MAX_BUF_SIZE) \
1882 bufp->allocated = MAX_BUF_SIZE; \
1883 RETALLOC (bufp->buffer, bufp->allocated, unsigned char); \
1884 if (bufp->buffer == NULL) \
1885 return REG_ESPACE; \
1886 /* If the buffer moved, move all the pointers into it. */ \
1887 if (old_buffer != bufp->buffer) \
1888 { \
1889 unsigned char *new_buffer = bufp->buffer; \
1890 MOVE_BUFFER_POINTER (b); \
1891 MOVE_BUFFER_POINTER (begalt); \
1892 if (fixup_alt_jump) \
1893 MOVE_BUFFER_POINTER (fixup_alt_jump); \
1894 if (laststart) \
1895 MOVE_BUFFER_POINTER (laststart); \
1896 if (pending_exact) \
1897 MOVE_BUFFER_POINTER (pending_exact); \
1898 } \
1899 ELSE_EXTEND_BUFFER_HIGH_BOUND \
1900 } while (0)
1901
1902
1903 /* Since we have one byte reserved for the register number argument to
1904 {start,stop}_memory, the maximum number of groups we can report
1905 things about is what fits in that byte. */
1906 #define MAX_REGNUM 255
1907
1908 /* But patterns can have more than `MAX_REGNUM' registers. We just
1909 ignore the excess. */
1910 typedef int regnum_t;
1911
1912
1913 /* Macros for the compile stack. */
1914
1915 /* Since offsets can go either forwards or backwards, this type needs to
1916 be able to hold values from -(MAX_BUF_SIZE - 1) to MAX_BUF_SIZE - 1. */
1917 /* int may be not enough when sizeof(int) == 2. */
1918 typedef long pattern_offset_t;
1919
1920 typedef struct
1921 {
1922 pattern_offset_t begalt_offset;
1923 pattern_offset_t fixup_alt_jump;
1924 pattern_offset_t laststart_offset;
1925 regnum_t regnum;
1926 } compile_stack_elt_t;
1927
1928
1929 typedef struct
1930 {
1931 compile_stack_elt_t *stack;
1932 unsigned size;
1933 unsigned avail; /* Offset of next open position. */
1934 } compile_stack_type;
1935
1936
1937 #define INIT_COMPILE_STACK_SIZE 32
1938
1939 #define COMPILE_STACK_EMPTY (compile_stack.avail == 0)
1940 #define COMPILE_STACK_FULL (compile_stack.avail == compile_stack.size)
1941
1942 /* The next available element. */
1943 #define COMPILE_STACK_TOP (compile_stack.stack[compile_stack.avail])
1944
1945 /* Explicit quit checking is only used on NTemacs and whenever we
1946 use polling to process input events. */
1947 #if defined emacs && (defined WINDOWSNT || defined SYNC_INPUT) && defined QUIT
1948 extern int immediate_quit;
1949 # define IMMEDIATE_QUIT_CHECK \
1950 do { \
1951 if (immediate_quit) QUIT; \
1952 } while (0)
1953 #else
1954 # define IMMEDIATE_QUIT_CHECK ((void)0)
1955 #endif
1956 \f
1957 /* Structure to manage work area for range table. */
1958 struct range_table_work_area
1959 {
1960 int *table; /* actual work area. */
1961 int allocated; /* allocated size for work area in bytes. */
1962 int used; /* actually used size in words. */
1963 int bits; /* flag to record character classes */
1964 };
1965
1966 /* Make sure that WORK_AREA can hold more N multibyte characters.
1967 This is used only in set_image_of_range and set_image_of_range_1.
1968 It expects WORK_AREA to be a pointer.
1969 If it can't get the space, it returns from the surrounding function. */
1970
1971 #define EXTEND_RANGE_TABLE(work_area, n) \
1972 do { \
1973 if (((work_area).used + (n)) * sizeof (int) > (work_area).allocated) \
1974 { \
1975 extend_range_table_work_area (&work_area); \
1976 if ((work_area).table == 0) \
1977 return (REG_ESPACE); \
1978 } \
1979 } while (0)
1980
1981 #define SET_RANGE_TABLE_WORK_AREA_BIT(work_area, bit) \
1982 (work_area).bits |= (bit)
1983
1984 /* Bits used to implement the multibyte-part of the various character classes
1985 such as [:alnum:] in a charset's range table. */
1986 #define BIT_WORD 0x1
1987 #define BIT_LOWER 0x2
1988 #define BIT_PUNCT 0x4
1989 #define BIT_SPACE 0x8
1990 #define BIT_UPPER 0x10
1991 #define BIT_MULTIBYTE 0x20
1992
1993 /* Set a range (RANGE_START, RANGE_END) to WORK_AREA. */
1994 #define SET_RANGE_TABLE_WORK_AREA(work_area, range_start, range_end) \
1995 do { \
1996 EXTEND_RANGE_TABLE ((work_area), 2); \
1997 (work_area).table[(work_area).used++] = (range_start); \
1998 (work_area).table[(work_area).used++] = (range_end); \
1999 } while (0)
2000
2001 /* Free allocated memory for WORK_AREA. */
2002 #define FREE_RANGE_TABLE_WORK_AREA(work_area) \
2003 do { \
2004 if ((work_area).table) \
2005 free ((work_area).table); \
2006 } while (0)
2007
2008 #define CLEAR_RANGE_TABLE_WORK_USED(work_area) ((work_area).used = 0, (work_area).bits = 0)
2009 #define RANGE_TABLE_WORK_USED(work_area) ((work_area).used)
2010 #define RANGE_TABLE_WORK_BITS(work_area) ((work_area).bits)
2011 #define RANGE_TABLE_WORK_ELT(work_area, i) ((work_area).table[i])
2012 \f
2013
2014 /* Set the bit for character C in a list. */
2015 #define SET_LIST_BIT(c) (b[((c)) / BYTEWIDTH] |= 1 << ((c) % BYTEWIDTH))
2016
2017
2018 #ifdef emacs
2019
2020 /* Store characters in the range FROM to TO in the bitmap at B (for
2021 ASCII and unibyte characters) and WORK_AREA (for multibyte
2022 characters) while translating them and paying attention to the
2023 continuity of translated characters.
2024
2025 Implementation note: It is better to implement these fairly big
2026 macros by a function, but it's not that easy because macros called
2027 in this macro assume various local variables already declared. */
2028
2029 /* Both FROM and TO are ASCII characters. */
2030
2031 #define SETUP_ASCII_RANGE(work_area, FROM, TO) \
2032 do { \
2033 int C0, C1; \
2034 \
2035 for (C0 = (FROM); C0 <= (TO); C0++) \
2036 { \
2037 C1 = TRANSLATE (C0); \
2038 if (! ASCII_CHAR_P (C1)) \
2039 { \
2040 SET_RANGE_TABLE_WORK_AREA ((work_area), C1, C1); \
2041 if ((C1 = RE_CHAR_TO_UNIBYTE (C1)) < 0) \
2042 C1 = C0; \
2043 } \
2044 SET_LIST_BIT (C1); \
2045 } \
2046 } while (0)
2047
2048
2049 /* Both FROM and TO are unibyte characters (0x80..0xFF). */
2050
2051 #define SETUP_UNIBYTE_RANGE(work_area, FROM, TO) \
2052 do { \
2053 int C0, C1, C2, I; \
2054 int USED = RANGE_TABLE_WORK_USED (work_area); \
2055 \
2056 for (C0 = (FROM); C0 <= (TO); C0++) \
2057 { \
2058 C1 = RE_CHAR_TO_MULTIBYTE (C0); \
2059 if (CHAR_BYTE8_P (C1)) \
2060 SET_LIST_BIT (C0); \
2061 else \
2062 { \
2063 C2 = TRANSLATE (C1); \
2064 if (C2 == C1 \
2065 || (C1 = RE_CHAR_TO_UNIBYTE (C2)) < 0) \
2066 C1 = C0; \
2067 SET_LIST_BIT (C1); \
2068 for (I = RANGE_TABLE_WORK_USED (work_area) - 2; I >= USED; I -= 2) \
2069 { \
2070 int from = RANGE_TABLE_WORK_ELT (work_area, I); \
2071 int to = RANGE_TABLE_WORK_ELT (work_area, I + 1); \
2072 \
2073 if (C2 >= from - 1 && C2 <= to + 1) \
2074 { \
2075 if (C2 == from - 1) \
2076 RANGE_TABLE_WORK_ELT (work_area, I)--; \
2077 else if (C2 == to + 1) \
2078 RANGE_TABLE_WORK_ELT (work_area, I + 1)++; \
2079 break; \
2080 } \
2081 } \
2082 if (I < USED) \
2083 SET_RANGE_TABLE_WORK_AREA ((work_area), C2, C2); \
2084 } \
2085 } \
2086 } while (0)
2087
2088
2089 /* Both FROM and TO are mulitbyte characters. */
2090
2091 #define SETUP_MULTIBYTE_RANGE(work_area, FROM, TO) \
2092 do { \
2093 int C0, C1, C2, I, USED = RANGE_TABLE_WORK_USED (work_area); \
2094 \
2095 SET_RANGE_TABLE_WORK_AREA ((work_area), (FROM), (TO)); \
2096 for (C0 = (FROM); C0 <= (TO); C0++) \
2097 { \
2098 C1 = TRANSLATE (C0); \
2099 if ((C2 = RE_CHAR_TO_UNIBYTE (C1)) >= 0 \
2100 || (C1 != C0 && (C2 = RE_CHAR_TO_UNIBYTE (C0)) >= 0)) \
2101 SET_LIST_BIT (C2); \
2102 if (C1 >= (FROM) && C1 <= (TO)) \
2103 continue; \
2104 for (I = RANGE_TABLE_WORK_USED (work_area) - 2; I >= USED; I -= 2) \
2105 { \
2106 int from = RANGE_TABLE_WORK_ELT (work_area, I); \
2107 int to = RANGE_TABLE_WORK_ELT (work_area, I + 1); \
2108 \
2109 if (C1 >= from - 1 && C1 <= to + 1) \
2110 { \
2111 if (C1 == from - 1) \
2112 RANGE_TABLE_WORK_ELT (work_area, I)--; \
2113 else if (C1 == to + 1) \
2114 RANGE_TABLE_WORK_ELT (work_area, I + 1)++; \
2115 break; \
2116 } \
2117 } \
2118 if (I < USED) \
2119 SET_RANGE_TABLE_WORK_AREA ((work_area), C1, C1); \
2120 } \
2121 } while (0)
2122
2123 #endif /* emacs */
2124
2125 /* Get the next unsigned number in the uncompiled pattern. */
2126 #define GET_UNSIGNED_NUMBER(num) \
2127 do { \
2128 if (p == pend) \
2129 FREE_STACK_RETURN (REG_EBRACE); \
2130 else \
2131 { \
2132 PATFETCH (c); \
2133 while ('0' <= c && c <= '9') \
2134 { \
2135 int prev; \
2136 if (num < 0) \
2137 num = 0; \
2138 prev = num; \
2139 num = num * 10 + c - '0'; \
2140 if (num / 10 != prev) \
2141 FREE_STACK_RETURN (REG_BADBR); \
2142 if (p == pend) \
2143 FREE_STACK_RETURN (REG_EBRACE); \
2144 PATFETCH (c); \
2145 } \
2146 } \
2147 } while (0)
2148 \f
2149 #if ! WIDE_CHAR_SUPPORT
2150
2151 /* Map a string to the char class it names (if any). */
2152 re_wctype_t
2153 re_wctype (str)
2154 re_char *str;
2155 {
2156 const char *string = str;
2157 if (STREQ (string, "alnum")) return RECC_ALNUM;
2158 else if (STREQ (string, "alpha")) return RECC_ALPHA;
2159 else if (STREQ (string, "word")) return RECC_WORD;
2160 else if (STREQ (string, "ascii")) return RECC_ASCII;
2161 else if (STREQ (string, "nonascii")) return RECC_NONASCII;
2162 else if (STREQ (string, "graph")) return RECC_GRAPH;
2163 else if (STREQ (string, "lower")) return RECC_LOWER;
2164 else if (STREQ (string, "print")) return RECC_PRINT;
2165 else if (STREQ (string, "punct")) return RECC_PUNCT;
2166 else if (STREQ (string, "space")) return RECC_SPACE;
2167 else if (STREQ (string, "upper")) return RECC_UPPER;
2168 else if (STREQ (string, "unibyte")) return RECC_UNIBYTE;
2169 else if (STREQ (string, "multibyte")) return RECC_MULTIBYTE;
2170 else if (STREQ (string, "digit")) return RECC_DIGIT;
2171 else if (STREQ (string, "xdigit")) return RECC_XDIGIT;
2172 else if (STREQ (string, "cntrl")) return RECC_CNTRL;
2173 else if (STREQ (string, "blank")) return RECC_BLANK;
2174 else return 0;
2175 }
2176
2177 /* True if CH is in the char class CC. */
2178 boolean
2179 re_iswctype (ch, cc)
2180 int ch;
2181 re_wctype_t cc;
2182 {
2183 switch (cc)
2184 {
2185 case RECC_ALNUM: return ISALNUM (ch);
2186 case RECC_ALPHA: return ISALPHA (ch);
2187 case RECC_BLANK: return ISBLANK (ch);
2188 case RECC_CNTRL: return ISCNTRL (ch);
2189 case RECC_DIGIT: return ISDIGIT (ch);
2190 case RECC_GRAPH: return ISGRAPH (ch);
2191 case RECC_LOWER: return ISLOWER (ch);
2192 case RECC_PRINT: return ISPRINT (ch);
2193 case RECC_PUNCT: return ISPUNCT (ch);
2194 case RECC_SPACE: return ISSPACE (ch);
2195 case RECC_UPPER: return ISUPPER (ch);
2196 case RECC_XDIGIT: return ISXDIGIT (ch);
2197 case RECC_ASCII: return IS_REAL_ASCII (ch);
2198 case RECC_NONASCII: return !IS_REAL_ASCII (ch);
2199 case RECC_UNIBYTE: return ISUNIBYTE (ch);
2200 case RECC_MULTIBYTE: return !ISUNIBYTE (ch);
2201 case RECC_WORD: return ISWORD (ch);
2202 case RECC_ERROR: return false;
2203 default:
2204 abort();
2205 }
2206 }
2207
2208 /* Return a bit-pattern to use in the range-table bits to match multibyte
2209 chars of class CC. */
2210 static int
2211 re_wctype_to_bit (cc)
2212 re_wctype_t cc;
2213 {
2214 switch (cc)
2215 {
2216 case RECC_NONASCII: case RECC_PRINT: case RECC_GRAPH:
2217 case RECC_MULTIBYTE: return BIT_MULTIBYTE;
2218 case RECC_ALPHA: case RECC_ALNUM: case RECC_WORD: return BIT_WORD;
2219 case RECC_LOWER: return BIT_LOWER;
2220 case RECC_UPPER: return BIT_UPPER;
2221 case RECC_PUNCT: return BIT_PUNCT;
2222 case RECC_SPACE: return BIT_SPACE;
2223 case RECC_ASCII: case RECC_DIGIT: case RECC_XDIGIT: case RECC_CNTRL:
2224 case RECC_BLANK: case RECC_UNIBYTE: case RECC_ERROR: return 0;
2225 default:
2226 abort();
2227 }
2228 }
2229 #endif
2230 \f
2231 /* Filling in the work area of a range. */
2232
2233 /* Actually extend the space in WORK_AREA. */
2234
2235 static void
2236 extend_range_table_work_area (work_area)
2237 struct range_table_work_area *work_area;
2238 {
2239 work_area->allocated += 16 * sizeof (int);
2240 if (work_area->table)
2241 work_area->table
2242 = (int *) realloc (work_area->table, work_area->allocated);
2243 else
2244 work_area->table
2245 = (int *) malloc (work_area->allocated);
2246 }
2247
2248 #if 0
2249 #ifdef emacs
2250
2251 /* Carefully find the ranges of codes that are equivalent
2252 under case conversion to the range start..end when passed through
2253 TRANSLATE. Handle the case where non-letters can come in between
2254 two upper-case letters (which happens in Latin-1).
2255 Also handle the case of groups of more than 2 case-equivalent chars.
2256
2257 The basic method is to look at consecutive characters and see
2258 if they can form a run that can be handled as one.
2259
2260 Returns -1 if successful, REG_ESPACE if ran out of space. */
2261
2262 static int
2263 set_image_of_range_1 (work_area, start, end, translate)
2264 RE_TRANSLATE_TYPE translate;
2265 struct range_table_work_area *work_area;
2266 re_wchar_t start, end;
2267 {
2268 /* `one_case' indicates a character, or a run of characters,
2269 each of which is an isolate (no case-equivalents).
2270 This includes all ASCII non-letters.
2271
2272 `two_case' indicates a character, or a run of characters,
2273 each of which has two case-equivalent forms.
2274 This includes all ASCII letters.
2275
2276 `strange' indicates a character that has more than one
2277 case-equivalent. */
2278
2279 enum case_type {one_case, two_case, strange};
2280
2281 /* Describe the run that is in progress,
2282 which the next character can try to extend.
2283 If run_type is strange, that means there really is no run.
2284 If run_type is one_case, then run_start...run_end is the run.
2285 If run_type is two_case, then the run is run_start...run_end,
2286 and the case-equivalents end at run_eqv_end. */
2287
2288 enum case_type run_type = strange;
2289 int run_start, run_end, run_eqv_end;
2290
2291 Lisp_Object eqv_table;
2292
2293 if (!RE_TRANSLATE_P (translate))
2294 {
2295 EXTEND_RANGE_TABLE (work_area, 2);
2296 work_area->table[work_area->used++] = (start);
2297 work_area->table[work_area->used++] = (end);
2298 return -1;
2299 }
2300
2301 eqv_table = XCHAR_TABLE (translate)->extras[2];
2302
2303 for (; start <= end; start++)
2304 {
2305 enum case_type this_type;
2306 int eqv = RE_TRANSLATE (eqv_table, start);
2307 int minchar, maxchar;
2308
2309 /* Classify this character */
2310 if (eqv == start)
2311 this_type = one_case;
2312 else if (RE_TRANSLATE (eqv_table, eqv) == start)
2313 this_type = two_case;
2314 else
2315 this_type = strange;
2316
2317 if (start < eqv)
2318 minchar = start, maxchar = eqv;
2319 else
2320 minchar = eqv, maxchar = start;
2321
2322 /* Can this character extend the run in progress? */
2323 if (this_type == strange || this_type != run_type
2324 || !(minchar == run_end + 1
2325 && (run_type == two_case
2326 ? maxchar == run_eqv_end + 1 : 1)))
2327 {
2328 /* No, end the run.
2329 Record each of its equivalent ranges. */
2330 if (run_type == one_case)
2331 {
2332 EXTEND_RANGE_TABLE (work_area, 2);
2333 work_area->table[work_area->used++] = run_start;
2334 work_area->table[work_area->used++] = run_end;
2335 }
2336 else if (run_type == two_case)
2337 {
2338 EXTEND_RANGE_TABLE (work_area, 4);
2339 work_area->table[work_area->used++] = run_start;
2340 work_area->table[work_area->used++] = run_end;
2341 work_area->table[work_area->used++]
2342 = RE_TRANSLATE (eqv_table, run_start);
2343 work_area->table[work_area->used++]
2344 = RE_TRANSLATE (eqv_table, run_end);
2345 }
2346 run_type = strange;
2347 }
2348
2349 if (this_type == strange)
2350 {
2351 /* For a strange character, add each of its equivalents, one
2352 by one. Don't start a range. */
2353 do
2354 {
2355 EXTEND_RANGE_TABLE (work_area, 2);
2356 work_area->table[work_area->used++] = eqv;
2357 work_area->table[work_area->used++] = eqv;
2358 eqv = RE_TRANSLATE (eqv_table, eqv);
2359 }
2360 while (eqv != start);
2361 }
2362
2363 /* Add this char to the run, or start a new run. */
2364 else if (run_type == strange)
2365 {
2366 /* Initialize a new range. */
2367 run_type = this_type;
2368 run_start = start;
2369 run_end = start;
2370 run_eqv_end = RE_TRANSLATE (eqv_table, run_end);
2371 }
2372 else
2373 {
2374 /* Extend a running range. */
2375 run_end = minchar;
2376 run_eqv_end = RE_TRANSLATE (eqv_table, run_end);
2377 }
2378 }
2379
2380 /* If a run is still in progress at the end, finish it now
2381 by recording its equivalent ranges. */
2382 if (run_type == one_case)
2383 {
2384 EXTEND_RANGE_TABLE (work_area, 2);
2385 work_area->table[work_area->used++] = run_start;
2386 work_area->table[work_area->used++] = run_end;
2387 }
2388 else if (run_type == two_case)
2389 {
2390 EXTEND_RANGE_TABLE (work_area, 4);
2391 work_area->table[work_area->used++] = run_start;
2392 work_area->table[work_area->used++] = run_end;
2393 work_area->table[work_area->used++]
2394 = RE_TRANSLATE (eqv_table, run_start);
2395 work_area->table[work_area->used++]
2396 = RE_TRANSLATE (eqv_table, run_end);
2397 }
2398
2399 return -1;
2400 }
2401
2402 #endif /* emacs */
2403
2404 /* Record the image of the range start..end when passed through
2405 TRANSLATE. This is not necessarily TRANSLATE(start)..TRANSLATE(end)
2406 and is not even necessarily contiguous.
2407 Normally we approximate it with the smallest contiguous range that contains
2408 all the chars we need. However, for Latin-1 we go to extra effort
2409 to do a better job.
2410
2411 This function is not called for ASCII ranges.
2412
2413 Returns -1 if successful, REG_ESPACE if ran out of space. */
2414
2415 static int
2416 set_image_of_range (work_area, start, end, translate)
2417 RE_TRANSLATE_TYPE translate;
2418 struct range_table_work_area *work_area;
2419 re_wchar_t start, end;
2420 {
2421 re_wchar_t cmin, cmax;
2422
2423 #ifdef emacs
2424 /* For Latin-1 ranges, use set_image_of_range_1
2425 to get proper handling of ranges that include letters and nonletters.
2426 For a range that includes the whole of Latin-1, this is not necessary.
2427 For other character sets, we don't bother to get this right. */
2428 if (RE_TRANSLATE_P (translate) && start < 04400
2429 && !(start < 04200 && end >= 04377))
2430 {
2431 int newend;
2432 int tem;
2433 newend = end;
2434 if (newend > 04377)
2435 newend = 04377;
2436 tem = set_image_of_range_1 (work_area, start, newend, translate);
2437 if (tem > 0)
2438 return tem;
2439
2440 start = 04400;
2441 if (end < 04400)
2442 return -1;
2443 }
2444 #endif
2445
2446 EXTEND_RANGE_TABLE (work_area, 2);
2447 work_area->table[work_area->used++] = (start);
2448 work_area->table[work_area->used++] = (end);
2449
2450 cmin = -1, cmax = -1;
2451
2452 if (RE_TRANSLATE_P (translate))
2453 {
2454 int ch;
2455
2456 for (ch = start; ch <= end; ch++)
2457 {
2458 re_wchar_t c = TRANSLATE (ch);
2459 if (! (start <= c && c <= end))
2460 {
2461 if (cmin == -1)
2462 cmin = c, cmax = c;
2463 else
2464 {
2465 cmin = MIN (cmin, c);
2466 cmax = MAX (cmax, c);
2467 }
2468 }
2469 }
2470
2471 if (cmin != -1)
2472 {
2473 EXTEND_RANGE_TABLE (work_area, 2);
2474 work_area->table[work_area->used++] = (cmin);
2475 work_area->table[work_area->used++] = (cmax);
2476 }
2477 }
2478
2479 return -1;
2480 }
2481 #endif /* 0 */
2482 \f
2483 #ifndef MATCH_MAY_ALLOCATE
2484
2485 /* If we cannot allocate large objects within re_match_2_internal,
2486 we make the fail stack and register vectors global.
2487 The fail stack, we grow to the maximum size when a regexp
2488 is compiled.
2489 The register vectors, we adjust in size each time we
2490 compile a regexp, according to the number of registers it needs. */
2491
2492 static fail_stack_type fail_stack;
2493
2494 /* Size with which the following vectors are currently allocated.
2495 That is so we can make them bigger as needed,
2496 but never make them smaller. */
2497 static int regs_allocated_size;
2498
2499 static re_char ** regstart, ** regend;
2500 static re_char **best_regstart, **best_regend;
2501
2502 /* Make the register vectors big enough for NUM_REGS registers,
2503 but don't make them smaller. */
2504
2505 static
2506 regex_grow_registers (num_regs)
2507 int num_regs;
2508 {
2509 if (num_regs > regs_allocated_size)
2510 {
2511 RETALLOC_IF (regstart, num_regs, re_char *);
2512 RETALLOC_IF (regend, num_regs, re_char *);
2513 RETALLOC_IF (best_regstart, num_regs, re_char *);
2514 RETALLOC_IF (best_regend, num_regs, re_char *);
2515
2516 regs_allocated_size = num_regs;
2517 }
2518 }
2519
2520 #endif /* not MATCH_MAY_ALLOCATE */
2521 \f
2522 static boolean group_in_compile_stack _RE_ARGS ((compile_stack_type
2523 compile_stack,
2524 regnum_t regnum));
2525
2526 /* `regex_compile' compiles PATTERN (of length SIZE) according to SYNTAX.
2527 Returns one of error codes defined in `regex.h', or zero for success.
2528
2529 Assumes the `allocated' (and perhaps `buffer') and `translate'
2530 fields are set in BUFP on entry.
2531
2532 If it succeeds, results are put in BUFP (if it returns an error, the
2533 contents of BUFP are undefined):
2534 `buffer' is the compiled pattern;
2535 `syntax' is set to SYNTAX;
2536 `used' is set to the length of the compiled pattern;
2537 `fastmap_accurate' is zero;
2538 `re_nsub' is the number of subexpressions in PATTERN;
2539 `not_bol' and `not_eol' are zero;
2540
2541 The `fastmap' field is neither examined nor set. */
2542
2543 /* Insert the `jump' from the end of last alternative to "here".
2544 The space for the jump has already been allocated. */
2545 #define FIXUP_ALT_JUMP() \
2546 do { \
2547 if (fixup_alt_jump) \
2548 STORE_JUMP (jump, fixup_alt_jump, b); \
2549 } while (0)
2550
2551
2552 /* Return, freeing storage we allocated. */
2553 #define FREE_STACK_RETURN(value) \
2554 do { \
2555 FREE_RANGE_TABLE_WORK_AREA (range_table_work); \
2556 free (compile_stack.stack); \
2557 return value; \
2558 } while (0)
2559
2560 static reg_errcode_t
2561 regex_compile (pattern, size, syntax, bufp)
2562 re_char *pattern;
2563 size_t size;
2564 reg_syntax_t syntax;
2565 struct re_pattern_buffer *bufp;
2566 {
2567 /* We fetch characters from PATTERN here. */
2568 register re_wchar_t c, c1;
2569
2570 /* A random temporary spot in PATTERN. */
2571 re_char *p1;
2572
2573 /* Points to the end of the buffer, where we should append. */
2574 register unsigned char *b;
2575
2576 /* Keeps track of unclosed groups. */
2577 compile_stack_type compile_stack;
2578
2579 /* Points to the current (ending) position in the pattern. */
2580 #ifdef AIX
2581 /* `const' makes AIX compiler fail. */
2582 unsigned char *p = pattern;
2583 #else
2584 re_char *p = pattern;
2585 #endif
2586 re_char *pend = pattern + size;
2587
2588 /* How to translate the characters in the pattern. */
2589 RE_TRANSLATE_TYPE translate = bufp->translate;
2590
2591 /* Address of the count-byte of the most recently inserted `exactn'
2592 command. This makes it possible to tell if a new exact-match
2593 character can be added to that command or if the character requires
2594 a new `exactn' command. */
2595 unsigned char *pending_exact = 0;
2596
2597 /* Address of start of the most recently finished expression.
2598 This tells, e.g., postfix * where to find the start of its
2599 operand. Reset at the beginning of groups and alternatives. */
2600 unsigned char *laststart = 0;
2601
2602 /* Address of beginning of regexp, or inside of last group. */
2603 unsigned char *begalt;
2604
2605 /* Place in the uncompiled pattern (i.e., the {) to
2606 which to go back if the interval is invalid. */
2607 re_char *beg_interval;
2608
2609 /* Address of the place where a forward jump should go to the end of
2610 the containing expression. Each alternative of an `or' -- except the
2611 last -- ends with a forward jump of this sort. */
2612 unsigned char *fixup_alt_jump = 0;
2613
2614 /* Work area for range table of charset. */
2615 struct range_table_work_area range_table_work;
2616
2617 /* If the object matched can contain multibyte characters. */
2618 const boolean multibyte = RE_MULTIBYTE_P (bufp);
2619
2620 /* If a target of matching can contain multibyte characters. */
2621 const boolean target_multibyte = RE_TARGET_MULTIBYTE_P (bufp);
2622
2623 /* Nonzero if we have pushed down into a subpattern. */
2624 int in_subpattern = 0;
2625
2626 /* These hold the values of p, pattern, and pend from the main
2627 pattern when we have pushed into a subpattern. */
2628 re_char *main_p;
2629 re_char *main_pattern;
2630 re_char *main_pend;
2631
2632 #ifdef DEBUG
2633 debug++;
2634 DEBUG_PRINT1 ("\nCompiling pattern: ");
2635 if (debug > 0)
2636 {
2637 unsigned debug_count;
2638
2639 for (debug_count = 0; debug_count < size; debug_count++)
2640 putchar (pattern[debug_count]);
2641 putchar ('\n');
2642 }
2643 #endif /* DEBUG */
2644
2645 /* Initialize the compile stack. */
2646 compile_stack.stack = TALLOC (INIT_COMPILE_STACK_SIZE, compile_stack_elt_t);
2647 if (compile_stack.stack == NULL)
2648 return REG_ESPACE;
2649
2650 compile_stack.size = INIT_COMPILE_STACK_SIZE;
2651 compile_stack.avail = 0;
2652
2653 range_table_work.table = 0;
2654 range_table_work.allocated = 0;
2655
2656 /* Initialize the pattern buffer. */
2657 bufp->syntax = syntax;
2658 bufp->fastmap_accurate = 0;
2659 bufp->not_bol = bufp->not_eol = 0;
2660 bufp->used_syntax = 0;
2661
2662 /* Set `used' to zero, so that if we return an error, the pattern
2663 printer (for debugging) will think there's no pattern. We reset it
2664 at the end. */
2665 bufp->used = 0;
2666
2667 /* Always count groups, whether or not bufp->no_sub is set. */
2668 bufp->re_nsub = 0;
2669
2670 #if !defined emacs && !defined SYNTAX_TABLE
2671 /* Initialize the syntax table. */
2672 init_syntax_once ();
2673 #endif
2674
2675 if (bufp->allocated == 0)
2676 {
2677 if (bufp->buffer)
2678 { /* If zero allocated, but buffer is non-null, try to realloc
2679 enough space. This loses if buffer's address is bogus, but
2680 that is the user's responsibility. */
2681 RETALLOC (bufp->buffer, INIT_BUF_SIZE, unsigned char);
2682 }
2683 else
2684 { /* Caller did not allocate a buffer. Do it for them. */
2685 bufp->buffer = TALLOC (INIT_BUF_SIZE, unsigned char);
2686 }
2687 if (!bufp->buffer) FREE_STACK_RETURN (REG_ESPACE);
2688
2689 bufp->allocated = INIT_BUF_SIZE;
2690 }
2691
2692 begalt = b = bufp->buffer;
2693
2694 /* Loop through the uncompiled pattern until we're at the end. */
2695 while (1)
2696 {
2697 if (p == pend)
2698 {
2699 /* If this is the end of an included regexp,
2700 pop back to the main regexp and try again. */
2701 if (in_subpattern)
2702 {
2703 in_subpattern = 0;
2704 pattern = main_pattern;
2705 p = main_p;
2706 pend = main_pend;
2707 continue;
2708 }
2709 /* If this is the end of the main regexp, we are done. */
2710 break;
2711 }
2712
2713 PATFETCH (c);
2714
2715 switch (c)
2716 {
2717 case ' ':
2718 {
2719 re_char *p1 = p;
2720
2721 /* If there's no special whitespace regexp, treat
2722 spaces normally. And don't try to do this recursively. */
2723 if (!whitespace_regexp || in_subpattern)
2724 goto normal_char;
2725
2726 /* Peek past following spaces. */
2727 while (p1 != pend)
2728 {
2729 if (*p1 != ' ')
2730 break;
2731 p1++;
2732 }
2733 /* If the spaces are followed by a repetition op,
2734 treat them normally. */
2735 if (p1 != pend
2736 && (*p1 == '*' || *p1 == '+' || *p1 == '?'
2737 || (*p1 == '\\' && p1 + 1 != pend && p1[1] == '{')))
2738 goto normal_char;
2739
2740 /* Replace the spaces with the whitespace regexp. */
2741 in_subpattern = 1;
2742 main_p = p1;
2743 main_pend = pend;
2744 main_pattern = pattern;
2745 p = pattern = whitespace_regexp;
2746 pend = p + strlen (p);
2747 break;
2748 }
2749
2750 case '^':
2751 {
2752 if ( /* If at start of pattern, it's an operator. */
2753 p == pattern + 1
2754 /* If context independent, it's an operator. */
2755 || syntax & RE_CONTEXT_INDEP_ANCHORS
2756 /* Otherwise, depends on what's come before. */
2757 || at_begline_loc_p (pattern, p, syntax))
2758 BUF_PUSH ((syntax & RE_NO_NEWLINE_ANCHOR) ? begbuf : begline);
2759 else
2760 goto normal_char;
2761 }
2762 break;
2763
2764
2765 case '$':
2766 {
2767 if ( /* If at end of pattern, it's an operator. */
2768 p == pend
2769 /* If context independent, it's an operator. */
2770 || syntax & RE_CONTEXT_INDEP_ANCHORS
2771 /* Otherwise, depends on what's next. */
2772 || at_endline_loc_p (p, pend, syntax))
2773 BUF_PUSH ((syntax & RE_NO_NEWLINE_ANCHOR) ? endbuf : endline);
2774 else
2775 goto normal_char;
2776 }
2777 break;
2778
2779
2780 case '+':
2781 case '?':
2782 if ((syntax & RE_BK_PLUS_QM)
2783 || (syntax & RE_LIMITED_OPS))
2784 goto normal_char;
2785 handle_plus:
2786 case '*':
2787 /* If there is no previous pattern... */
2788 if (!laststart)
2789 {
2790 if (syntax & RE_CONTEXT_INVALID_OPS)
2791 FREE_STACK_RETURN (REG_BADRPT);
2792 else if (!(syntax & RE_CONTEXT_INDEP_OPS))
2793 goto normal_char;
2794 }
2795
2796 {
2797 /* 1 means zero (many) matches is allowed. */
2798 boolean zero_times_ok = 0, many_times_ok = 0;
2799 boolean greedy = 1;
2800
2801 /* If there is a sequence of repetition chars, collapse it
2802 down to just one (the right one). We can't combine
2803 interval operators with these because of, e.g., `a{2}*',
2804 which should only match an even number of `a's. */
2805
2806 for (;;)
2807 {
2808 if ((syntax & RE_FRUGAL)
2809 && c == '?' && (zero_times_ok || many_times_ok))
2810 greedy = 0;
2811 else
2812 {
2813 zero_times_ok |= c != '+';
2814 many_times_ok |= c != '?';
2815 }
2816
2817 if (p == pend)
2818 break;
2819 else if (*p == '*'
2820 || (!(syntax & RE_BK_PLUS_QM)
2821 && (*p == '+' || *p == '?')))
2822 ;
2823 else if (syntax & RE_BK_PLUS_QM && *p == '\\')
2824 {
2825 if (p+1 == pend)
2826 FREE_STACK_RETURN (REG_EESCAPE);
2827 if (p[1] == '+' || p[1] == '?')
2828 PATFETCH (c); /* Gobble up the backslash. */
2829 else
2830 break;
2831 }
2832 else
2833 break;
2834 /* If we get here, we found another repeat character. */
2835 PATFETCH (c);
2836 }
2837
2838 /* Star, etc. applied to an empty pattern is equivalent
2839 to an empty pattern. */
2840 if (!laststart || laststart == b)
2841 break;
2842
2843 /* Now we know whether or not zero matches is allowed
2844 and also whether or not two or more matches is allowed. */
2845 if (greedy)
2846 {
2847 if (many_times_ok)
2848 {
2849 boolean simple = skip_one_char (laststart) == b;
2850 unsigned int startoffset = 0;
2851 re_opcode_t ofj =
2852 /* Check if the loop can match the empty string. */
2853 (simple || !analyse_first (laststart, b, NULL, 0))
2854 ? on_failure_jump : on_failure_jump_loop;
2855 assert (skip_one_char (laststart) <= b);
2856
2857 if (!zero_times_ok && simple)
2858 { /* Since simple * loops can be made faster by using
2859 on_failure_keep_string_jump, we turn simple P+
2860 into PP* if P is simple. */
2861 unsigned char *p1, *p2;
2862 startoffset = b - laststart;
2863 GET_BUFFER_SPACE (startoffset);
2864 p1 = b; p2 = laststart;
2865 while (p2 < p1)
2866 *b++ = *p2++;
2867 zero_times_ok = 1;
2868 }
2869
2870 GET_BUFFER_SPACE (6);
2871 if (!zero_times_ok)
2872 /* A + loop. */
2873 STORE_JUMP (ofj, b, b + 6);
2874 else
2875 /* Simple * loops can use on_failure_keep_string_jump
2876 depending on what follows. But since we don't know
2877 that yet, we leave the decision up to
2878 on_failure_jump_smart. */
2879 INSERT_JUMP (simple ? on_failure_jump_smart : ofj,
2880 laststart + startoffset, b + 6);
2881 b += 3;
2882 STORE_JUMP (jump, b, laststart + startoffset);
2883 b += 3;
2884 }
2885 else
2886 {
2887 /* A simple ? pattern. */
2888 assert (zero_times_ok);
2889 GET_BUFFER_SPACE (3);
2890 INSERT_JUMP (on_failure_jump, laststart, b + 3);
2891 b += 3;
2892 }
2893 }
2894 else /* not greedy */
2895 { /* I wish the greedy and non-greedy cases could be merged. */
2896
2897 GET_BUFFER_SPACE (7); /* We might use less. */
2898 if (many_times_ok)
2899 {
2900 boolean emptyp = analyse_first (laststart, b, NULL, 0);
2901
2902 /* The non-greedy multiple match looks like
2903 a repeat..until: we only need a conditional jump
2904 at the end of the loop. */
2905 if (emptyp) BUF_PUSH (no_op);
2906 STORE_JUMP (emptyp ? on_failure_jump_nastyloop
2907 : on_failure_jump, b, laststart);
2908 b += 3;
2909 if (zero_times_ok)
2910 {
2911 /* The repeat...until naturally matches one or more.
2912 To also match zero times, we need to first jump to
2913 the end of the loop (its conditional jump). */
2914 INSERT_JUMP (jump, laststart, b);
2915 b += 3;
2916 }
2917 }
2918 else
2919 {
2920 /* non-greedy a?? */
2921 INSERT_JUMP (jump, laststart, b + 3);
2922 b += 3;
2923 INSERT_JUMP (on_failure_jump, laststart, laststart + 6);
2924 b += 3;
2925 }
2926 }
2927 }
2928 pending_exact = 0;
2929 break;
2930
2931
2932 case '.':
2933 laststart = b;
2934 BUF_PUSH (anychar);
2935 break;
2936
2937
2938 case '[':
2939 {
2940 CLEAR_RANGE_TABLE_WORK_USED (range_table_work);
2941
2942 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2943
2944 /* Ensure that we have enough space to push a charset: the
2945 opcode, the length count, and the bitset; 34 bytes in all. */
2946 GET_BUFFER_SPACE (34);
2947
2948 laststart = b;
2949
2950 /* We test `*p == '^' twice, instead of using an if
2951 statement, so we only need one BUF_PUSH. */
2952 BUF_PUSH (*p == '^' ? charset_not : charset);
2953 if (*p == '^')
2954 p++;
2955
2956 /* Remember the first position in the bracket expression. */
2957 p1 = p;
2958
2959 /* Push the number of bytes in the bitmap. */
2960 BUF_PUSH ((1 << BYTEWIDTH) / BYTEWIDTH);
2961
2962 /* Clear the whole map. */
2963 bzero (b, (1 << BYTEWIDTH) / BYTEWIDTH);
2964
2965 /* charset_not matches newline according to a syntax bit. */
2966 if ((re_opcode_t) b[-2] == charset_not
2967 && (syntax & RE_HAT_LISTS_NOT_NEWLINE))
2968 SET_LIST_BIT ('\n');
2969
2970 /* Read in characters and ranges, setting map bits. */
2971 for (;;)
2972 {
2973 boolean escaped_char = false;
2974 const unsigned char *p2 = p;
2975 re_wchar_t ch, c2;
2976
2977 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2978
2979 /* Don't translate yet. The range TRANSLATE(X..Y) cannot
2980 always be determined from TRANSLATE(X) and TRANSLATE(Y)
2981 So the translation is done later in a loop. Example:
2982 (let ((case-fold-search t)) (string-match "[A-_]" "A")) */
2983 PATFETCH (c);
2984
2985 /* \ might escape characters inside [...] and [^...]. */
2986 if ((syntax & RE_BACKSLASH_ESCAPE_IN_LISTS) && c == '\\')
2987 {
2988 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
2989
2990 PATFETCH (c);
2991 escaped_char = true;
2992 }
2993 else
2994 {
2995 /* Could be the end of the bracket expression. If it's
2996 not (i.e., when the bracket expression is `[]' so
2997 far), the ']' character bit gets set way below. */
2998 if (c == ']' && p2 != p1)
2999 break;
3000 }
3001
3002 /* See if we're at the beginning of a possible character
3003 class. */
3004
3005 if (!escaped_char &&
3006 syntax & RE_CHAR_CLASSES && c == '[' && *p == ':')
3007 {
3008 /* Leave room for the null. */
3009 unsigned char str[CHAR_CLASS_MAX_LENGTH + 1];
3010 const unsigned char *class_beg;
3011
3012 PATFETCH (c);
3013 c1 = 0;
3014 class_beg = p;
3015
3016 /* If pattern is `[[:'. */
3017 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
3018
3019 for (;;)
3020 {
3021 PATFETCH (c);
3022 if ((c == ':' && *p == ']') || p == pend)
3023 break;
3024 if (c1 < CHAR_CLASS_MAX_LENGTH)
3025 str[c1++] = c;
3026 else
3027 /* This is in any case an invalid class name. */
3028 str[0] = '\0';
3029 }
3030 str[c1] = '\0';
3031
3032 /* If isn't a word bracketed by `[:' and `:]':
3033 undo the ending character, the letters, and
3034 leave the leading `:' and `[' (but set bits for
3035 them). */
3036 if (c == ':' && *p == ']')
3037 {
3038 re_wctype_t cc;
3039 int limit;
3040
3041 cc = re_wctype (str);
3042
3043 if (cc == 0)
3044 FREE_STACK_RETURN (REG_ECTYPE);
3045
3046 /* Throw away the ] at the end of the character
3047 class. */
3048 PATFETCH (c);
3049
3050 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
3051
3052 #ifndef emacs
3053 for (ch = 0; ch < (1 << BYTEWIDTH); ++ch)
3054 if (re_iswctype (btowc (ch), cc))
3055 {
3056 c = TRANSLATE (ch);
3057 if (c < (1 << BYTEWIDTH))
3058 SET_LIST_BIT (c);
3059 }
3060 #else /* emacs */
3061 /* Most character classes in a multibyte match
3062 just set a flag. Exceptions are is_blank,
3063 is_digit, is_cntrl, and is_xdigit, since
3064 they can only match ASCII characters. We
3065 don't need to handle them for multibyte.
3066 They are distinguished by a negative wctype. */
3067
3068 for (ch = 0; ch < 256; ++ch)
3069 {
3070 c = RE_CHAR_TO_MULTIBYTE (ch);
3071 if (! CHAR_BYTE8_P (c)
3072 && re_iswctype (c, cc))
3073 {
3074 SET_LIST_BIT (ch);
3075 c1 = TRANSLATE (c);
3076 if (c1 == c)
3077 continue;
3078 if (ASCII_CHAR_P (c1))
3079 SET_LIST_BIT (c1);
3080 else if ((c1 = RE_CHAR_TO_UNIBYTE (c1)) >= 0)
3081 SET_LIST_BIT (c1);
3082 }
3083 }
3084 SET_RANGE_TABLE_WORK_AREA_BIT
3085 (range_table_work, re_wctype_to_bit (cc));
3086 #endif /* emacs */
3087 /* In most cases the matching rule for char classes
3088 only uses the syntax table for multibyte chars,
3089 so that the content of the syntax-table it is not
3090 hardcoded in the range_table. SPACE and WORD are
3091 the two exceptions. */
3092 if ((1 << cc) & ((1 << RECC_SPACE) | (1 << RECC_WORD)))
3093 bufp->used_syntax = 1;
3094
3095 /* Repeat the loop. */
3096 continue;
3097 }
3098 else
3099 {
3100 /* Go back to right after the "[:". */
3101 p = class_beg;
3102 SET_LIST_BIT ('[');
3103
3104 /* Because the `:' may starts the range, we
3105 can't simply set bit and repeat the loop.
3106 Instead, just set it to C and handle below. */
3107 c = ':';
3108 }
3109 }
3110
3111 if (p < pend && p[0] == '-' && p[1] != ']')
3112 {
3113
3114 /* Discard the `-'. */
3115 PATFETCH (c1);
3116
3117 /* Fetch the character which ends the range. */
3118 PATFETCH (c1);
3119 #ifdef emacs
3120 if (CHAR_BYTE8_P (c1)
3121 && ! ASCII_CHAR_P (c) && ! CHAR_BYTE8_P (c))
3122 /* Treat the range from a multibyte character to
3123 raw-byte character as empty. */
3124 c = c1 + 1;
3125 #endif /* emacs */
3126 }
3127 else
3128 /* Range from C to C. */
3129 c1 = c;
3130
3131 if (c > c1)
3132 {
3133 if (syntax & RE_NO_EMPTY_RANGES)
3134 FREE_STACK_RETURN (REG_ERANGEX);
3135 /* Else, repeat the loop. */
3136 }
3137 else
3138 {
3139 #ifndef emacs
3140 /* Set the range into bitmap */
3141 for (; c <= c1; c++)
3142 {
3143 ch = TRANSLATE (c);
3144 if (ch < (1 << BYTEWIDTH))
3145 SET_LIST_BIT (ch);
3146 }
3147 #else /* emacs */
3148 if (c < 128)
3149 {
3150 ch = MIN (127, c1);
3151 SETUP_ASCII_RANGE (range_table_work, c, ch);
3152 c = ch + 1;
3153 if (CHAR_BYTE8_P (c1))
3154 c = BYTE8_TO_CHAR (128);
3155 }
3156 if (c <= c1)
3157 {
3158 if (CHAR_BYTE8_P (c))
3159 {
3160 c = CHAR_TO_BYTE8 (c);
3161 c1 = CHAR_TO_BYTE8 (c1);
3162 for (; c <= c1; c++)
3163 SET_LIST_BIT (c);
3164 }
3165 else if (multibyte)
3166 {
3167 SETUP_MULTIBYTE_RANGE (range_table_work, c, c1);
3168 }
3169 else
3170 {
3171 SETUP_UNIBYTE_RANGE (range_table_work, c, c1);
3172 }
3173 }
3174 #endif /* emacs */
3175 }
3176 }
3177
3178 /* Discard any (non)matching list bytes that are all 0 at the
3179 end of the map. Decrease the map-length byte too. */
3180 while ((int) b[-1] > 0 && b[b[-1] - 1] == 0)
3181 b[-1]--;
3182 b += b[-1];
3183
3184 /* Build real range table from work area. */
3185 if (RANGE_TABLE_WORK_USED (range_table_work)
3186 || RANGE_TABLE_WORK_BITS (range_table_work))
3187 {
3188 int i;
3189 int used = RANGE_TABLE_WORK_USED (range_table_work);
3190
3191 /* Allocate space for COUNT + RANGE_TABLE. Needs two
3192 bytes for flags, two for COUNT, and three bytes for
3193 each character. */
3194 GET_BUFFER_SPACE (4 + used * 3);
3195
3196 /* Indicate the existence of range table. */
3197 laststart[1] |= 0x80;
3198
3199 /* Store the character class flag bits into the range table.
3200 If not in emacs, these flag bits are always 0. */
3201 *b++ = RANGE_TABLE_WORK_BITS (range_table_work) & 0xff;
3202 *b++ = RANGE_TABLE_WORK_BITS (range_table_work) >> 8;
3203
3204 STORE_NUMBER_AND_INCR (b, used / 2);
3205 for (i = 0; i < used; i++)
3206 STORE_CHARACTER_AND_INCR
3207 (b, RANGE_TABLE_WORK_ELT (range_table_work, i));
3208 }
3209 }
3210 break;
3211
3212
3213 case '(':
3214 if (syntax & RE_NO_BK_PARENS)
3215 goto handle_open;
3216 else
3217 goto normal_char;
3218
3219
3220 case ')':
3221 if (syntax & RE_NO_BK_PARENS)
3222 goto handle_close;
3223 else
3224 goto normal_char;
3225
3226
3227 case '\n':
3228 if (syntax & RE_NEWLINE_ALT)
3229 goto handle_alt;
3230 else
3231 goto normal_char;
3232
3233
3234 case '|':
3235 if (syntax & RE_NO_BK_VBAR)
3236 goto handle_alt;
3237 else
3238 goto normal_char;
3239
3240
3241 case '{':
3242 if (syntax & RE_INTERVALS && syntax & RE_NO_BK_BRACES)
3243 goto handle_interval;
3244 else
3245 goto normal_char;
3246
3247
3248 case '\\':
3249 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
3250
3251 /* Do not translate the character after the \, so that we can
3252 distinguish, e.g., \B from \b, even if we normally would
3253 translate, e.g., B to b. */
3254 PATFETCH (c);
3255
3256 switch (c)
3257 {
3258 case '(':
3259 if (syntax & RE_NO_BK_PARENS)
3260 goto normal_backslash;
3261
3262 handle_open:
3263 {
3264 int shy = 0;
3265 regnum_t regnum = 0;
3266 if (p+1 < pend)
3267 {
3268 /* Look for a special (?...) construct */
3269 if ((syntax & RE_SHY_GROUPS) && *p == '?')
3270 {
3271 PATFETCH (c); /* Gobble up the '?'. */
3272 while (!shy)
3273 {
3274 PATFETCH (c);
3275 switch (c)
3276 {
3277 case ':': shy = 1; break;
3278 case '0':
3279 /* An explicitly specified regnum must start
3280 with non-0. */
3281 if (regnum == 0)
3282 FREE_STACK_RETURN (REG_BADPAT);
3283 case '1': case '2': case '3': case '4':
3284 case '5': case '6': case '7': case '8': case '9':
3285 regnum = 10*regnum + (c - '0'); break;
3286 default:
3287 /* Only (?:...) is supported right now. */
3288 FREE_STACK_RETURN (REG_BADPAT);
3289 }
3290 }
3291 }
3292 }
3293
3294 if (!shy)
3295 regnum = ++bufp->re_nsub;
3296 else if (regnum)
3297 { /* It's actually not shy, but explicitly numbered. */
3298 shy = 0;
3299 if (regnum > bufp->re_nsub)
3300 bufp->re_nsub = regnum;
3301 else if (regnum > bufp->re_nsub
3302 /* Ideally, we'd want to check that the specified
3303 group can't have matched (i.e. all subgroups
3304 using the same regnum are in other branches of
3305 OR patterns), but we don't currently keep track
3306 of enough info to do that easily. */
3307 || group_in_compile_stack (compile_stack, regnum))
3308 FREE_STACK_RETURN (REG_BADPAT);
3309 }
3310 else
3311 /* It's really shy. */
3312 regnum = - bufp->re_nsub;
3313
3314 if (COMPILE_STACK_FULL)
3315 {
3316 RETALLOC (compile_stack.stack, compile_stack.size << 1,
3317 compile_stack_elt_t);
3318 if (compile_stack.stack == NULL) return REG_ESPACE;
3319
3320 compile_stack.size <<= 1;
3321 }
3322
3323 /* These are the values to restore when we hit end of this
3324 group. They are all relative offsets, so that if the
3325 whole pattern moves because of realloc, they will still
3326 be valid. */
3327 COMPILE_STACK_TOP.begalt_offset = begalt - bufp->buffer;
3328 COMPILE_STACK_TOP.fixup_alt_jump
3329 = fixup_alt_jump ? fixup_alt_jump - bufp->buffer + 1 : 0;
3330 COMPILE_STACK_TOP.laststart_offset = b - bufp->buffer;
3331 COMPILE_STACK_TOP.regnum = regnum;
3332
3333 /* Do not push a start_memory for groups beyond the last one
3334 we can represent in the compiled pattern. */
3335 if (regnum <= MAX_REGNUM && regnum > 0)
3336 BUF_PUSH_2 (start_memory, regnum);
3337
3338 compile_stack.avail++;
3339
3340 fixup_alt_jump = 0;
3341 laststart = 0;
3342 begalt = b;
3343 /* If we've reached MAX_REGNUM groups, then this open
3344 won't actually generate any code, so we'll have to
3345 clear pending_exact explicitly. */
3346 pending_exact = 0;
3347 break;
3348 }
3349
3350 case ')':
3351 if (syntax & RE_NO_BK_PARENS) goto normal_backslash;
3352
3353 if (COMPILE_STACK_EMPTY)
3354 {
3355 if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD)
3356 goto normal_backslash;
3357 else
3358 FREE_STACK_RETURN (REG_ERPAREN);
3359 }
3360
3361 handle_close:
3362 FIXUP_ALT_JUMP ();
3363
3364 /* See similar code for backslashed left paren above. */
3365 if (COMPILE_STACK_EMPTY)
3366 {
3367 if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD)
3368 goto normal_char;
3369 else
3370 FREE_STACK_RETURN (REG_ERPAREN);
3371 }
3372
3373 /* Since we just checked for an empty stack above, this
3374 ``can't happen''. */
3375 assert (compile_stack.avail != 0);
3376 {
3377 /* We don't just want to restore into `regnum', because
3378 later groups should continue to be numbered higher,
3379 as in `(ab)c(de)' -- the second group is #2. */
3380 regnum_t regnum;
3381
3382 compile_stack.avail--;
3383 begalt = bufp->buffer + COMPILE_STACK_TOP.begalt_offset;
3384 fixup_alt_jump
3385 = COMPILE_STACK_TOP.fixup_alt_jump
3386 ? bufp->buffer + COMPILE_STACK_TOP.fixup_alt_jump - 1
3387 : 0;
3388 laststart = bufp->buffer + COMPILE_STACK_TOP.laststart_offset;
3389 regnum = COMPILE_STACK_TOP.regnum;
3390 /* If we've reached MAX_REGNUM groups, then this open
3391 won't actually generate any code, so we'll have to
3392 clear pending_exact explicitly. */
3393 pending_exact = 0;
3394
3395 /* We're at the end of the group, so now we know how many
3396 groups were inside this one. */
3397 if (regnum <= MAX_REGNUM && regnum > 0)
3398 BUF_PUSH_2 (stop_memory, regnum);
3399 }
3400 break;
3401
3402
3403 case '|': /* `\|'. */
3404 if (syntax & RE_LIMITED_OPS || syntax & RE_NO_BK_VBAR)
3405 goto normal_backslash;
3406 handle_alt:
3407 if (syntax & RE_LIMITED_OPS)
3408 goto normal_char;
3409
3410 /* Insert before the previous alternative a jump which
3411 jumps to this alternative if the former fails. */
3412 GET_BUFFER_SPACE (3);
3413 INSERT_JUMP (on_failure_jump, begalt, b + 6);
3414 pending_exact = 0;
3415 b += 3;
3416
3417 /* The alternative before this one has a jump after it
3418 which gets executed if it gets matched. Adjust that
3419 jump so it will jump to this alternative's analogous
3420 jump (put in below, which in turn will jump to the next
3421 (if any) alternative's such jump, etc.). The last such
3422 jump jumps to the correct final destination. A picture:
3423 _____ _____
3424 | | | |
3425 | v | v
3426 a | b | c
3427
3428 If we are at `b', then fixup_alt_jump right now points to a
3429 three-byte space after `a'. We'll put in the jump, set
3430 fixup_alt_jump to right after `b', and leave behind three
3431 bytes which we'll fill in when we get to after `c'. */
3432
3433 FIXUP_ALT_JUMP ();
3434
3435 /* Mark and leave space for a jump after this alternative,
3436 to be filled in later either by next alternative or
3437 when know we're at the end of a series of alternatives. */
3438 fixup_alt_jump = b;
3439 GET_BUFFER_SPACE (3);
3440 b += 3;
3441
3442 laststart = 0;
3443 begalt = b;
3444 break;
3445
3446
3447 case '{':
3448 /* If \{ is a literal. */
3449 if (!(syntax & RE_INTERVALS)
3450 /* If we're at `\{' and it's not the open-interval
3451 operator. */
3452 || (syntax & RE_NO_BK_BRACES))
3453 goto normal_backslash;
3454
3455 handle_interval:
3456 {
3457 /* If got here, then the syntax allows intervals. */
3458
3459 /* At least (most) this many matches must be made. */
3460 int lower_bound = 0, upper_bound = -1;
3461
3462 beg_interval = p;
3463
3464 GET_UNSIGNED_NUMBER (lower_bound);
3465
3466 if (c == ',')
3467 GET_UNSIGNED_NUMBER (upper_bound);
3468 else
3469 /* Interval such as `{1}' => match exactly once. */
3470 upper_bound = lower_bound;
3471
3472 if (lower_bound < 0 || upper_bound > RE_DUP_MAX
3473 || (upper_bound >= 0 && lower_bound > upper_bound))
3474 FREE_STACK_RETURN (REG_BADBR);
3475
3476 if (!(syntax & RE_NO_BK_BRACES))
3477 {
3478 if (c != '\\')
3479 FREE_STACK_RETURN (REG_BADBR);
3480 if (p == pend)
3481 FREE_STACK_RETURN (REG_EESCAPE);
3482 PATFETCH (c);
3483 }
3484
3485 if (c != '}')
3486 FREE_STACK_RETURN (REG_BADBR);
3487
3488 /* We just parsed a valid interval. */
3489
3490 /* If it's invalid to have no preceding re. */
3491 if (!laststart)
3492 {
3493 if (syntax & RE_CONTEXT_INVALID_OPS)
3494 FREE_STACK_RETURN (REG_BADRPT);
3495 else if (syntax & RE_CONTEXT_INDEP_OPS)
3496 laststart = b;
3497 else
3498 goto unfetch_interval;
3499 }
3500
3501 if (upper_bound == 0)
3502 /* If the upper bound is zero, just drop the sub pattern
3503 altogether. */
3504 b = laststart;
3505 else if (lower_bound == 1 && upper_bound == 1)
3506 /* Just match it once: nothing to do here. */
3507 ;
3508
3509 /* Otherwise, we have a nontrivial interval. When
3510 we're all done, the pattern will look like:
3511 set_number_at <jump count> <upper bound>
3512 set_number_at <succeed_n count> <lower bound>
3513 succeed_n <after jump addr> <succeed_n count>
3514 <body of loop>
3515 jump_n <succeed_n addr> <jump count>
3516 (The upper bound and `jump_n' are omitted if
3517 `upper_bound' is 1, though.) */
3518 else
3519 { /* If the upper bound is > 1, we need to insert
3520 more at the end of the loop. */
3521 unsigned int nbytes = (upper_bound < 0 ? 3
3522 : upper_bound > 1 ? 5 : 0);
3523 unsigned int startoffset = 0;
3524
3525 GET_BUFFER_SPACE (20); /* We might use less. */
3526
3527 if (lower_bound == 0)
3528 {
3529 /* A succeed_n that starts with 0 is really a
3530 a simple on_failure_jump_loop. */
3531 INSERT_JUMP (on_failure_jump_loop, laststart,
3532 b + 3 + nbytes);
3533 b += 3;
3534 }
3535 else
3536 {
3537 /* Initialize lower bound of the `succeed_n', even
3538 though it will be set during matching by its
3539 attendant `set_number_at' (inserted next),
3540 because `re_compile_fastmap' needs to know.
3541 Jump to the `jump_n' we might insert below. */
3542 INSERT_JUMP2 (succeed_n, laststart,
3543 b + 5 + nbytes,
3544 lower_bound);
3545 b += 5;
3546
3547 /* Code to initialize the lower bound. Insert
3548 before the `succeed_n'. The `5' is the last two
3549 bytes of this `set_number_at', plus 3 bytes of
3550 the following `succeed_n'. */
3551 insert_op2 (set_number_at, laststart, 5, lower_bound, b);
3552 b += 5;
3553 startoffset += 5;
3554 }
3555
3556 if (upper_bound < 0)
3557 {
3558 /* A negative upper bound stands for infinity,
3559 in which case it degenerates to a plain jump. */
3560 STORE_JUMP (jump, b, laststart + startoffset);
3561 b += 3;
3562 }
3563 else if (upper_bound > 1)
3564 { /* More than one repetition is allowed, so
3565 append a backward jump to the `succeed_n'
3566 that starts this interval.
3567
3568 When we've reached this during matching,
3569 we'll have matched the interval once, so
3570 jump back only `upper_bound - 1' times. */
3571 STORE_JUMP2 (jump_n, b, laststart + startoffset,
3572 upper_bound - 1);
3573 b += 5;
3574
3575 /* The location we want to set is the second
3576 parameter of the `jump_n'; that is `b-2' as
3577 an absolute address. `laststart' will be
3578 the `set_number_at' we're about to insert;
3579 `laststart+3' the number to set, the source
3580 for the relative address. But we are
3581 inserting into the middle of the pattern --
3582 so everything is getting moved up by 5.
3583 Conclusion: (b - 2) - (laststart + 3) + 5,
3584 i.e., b - laststart.
3585
3586 We insert this at the beginning of the loop
3587 so that if we fail during matching, we'll
3588 reinitialize the bounds. */
3589 insert_op2 (set_number_at, laststart, b - laststart,
3590 upper_bound - 1, b);
3591 b += 5;
3592 }
3593 }
3594 pending_exact = 0;
3595 beg_interval = NULL;
3596 }
3597 break;
3598
3599 unfetch_interval:
3600 /* If an invalid interval, match the characters as literals. */
3601 assert (beg_interval);
3602 p = beg_interval;
3603 beg_interval = NULL;
3604
3605 /* normal_char and normal_backslash need `c'. */
3606 c = '{';
3607
3608 if (!(syntax & RE_NO_BK_BRACES))
3609 {
3610 assert (p > pattern && p[-1] == '\\');
3611 goto normal_backslash;
3612 }
3613 else
3614 goto normal_char;
3615
3616 #ifdef emacs
3617 /* There is no way to specify the before_dot and after_dot
3618 operators. rms says this is ok. --karl */
3619 case '=':
3620 BUF_PUSH (at_dot);
3621 break;
3622
3623 case 's':
3624 laststart = b;
3625 PATFETCH (c);
3626 BUF_PUSH_2 (syntaxspec, syntax_spec_code[c]);
3627 break;
3628
3629 case 'S':
3630 laststart = b;
3631 PATFETCH (c);
3632 BUF_PUSH_2 (notsyntaxspec, syntax_spec_code[c]);
3633 break;
3634
3635 case 'c':
3636 laststart = b;
3637 PATFETCH (c);
3638 BUF_PUSH_2 (categoryspec, c);
3639 break;
3640
3641 case 'C':
3642 laststart = b;
3643 PATFETCH (c);
3644 BUF_PUSH_2 (notcategoryspec, c);
3645 break;
3646 #endif /* emacs */
3647
3648
3649 case 'w':
3650 if (syntax & RE_NO_GNU_OPS)
3651 goto normal_char;
3652 laststart = b;
3653 BUF_PUSH_2 (syntaxspec, Sword);
3654 break;
3655
3656
3657 case 'W':
3658 if (syntax & RE_NO_GNU_OPS)
3659 goto normal_char;
3660 laststart = b;
3661 BUF_PUSH_2 (notsyntaxspec, Sword);
3662 break;
3663
3664
3665 case '<':
3666 if (syntax & RE_NO_GNU_OPS)
3667 goto normal_char;
3668 BUF_PUSH (wordbeg);
3669 break;
3670
3671 case '>':
3672 if (syntax & RE_NO_GNU_OPS)
3673 goto normal_char;
3674 BUF_PUSH (wordend);
3675 break;
3676
3677 case '_':
3678 if (syntax & RE_NO_GNU_OPS)
3679 goto normal_char;
3680 laststart = b;
3681 PATFETCH (c);
3682 if (c == '<')
3683 BUF_PUSH (symbeg);
3684 else if (c == '>')
3685 BUF_PUSH (symend);
3686 else
3687 FREE_STACK_RETURN (REG_BADPAT);
3688 break;
3689
3690 case 'b':
3691 if (syntax & RE_NO_GNU_OPS)
3692 goto normal_char;
3693 BUF_PUSH (wordbound);
3694 break;
3695
3696 case 'B':
3697 if (syntax & RE_NO_GNU_OPS)
3698 goto normal_char;
3699 BUF_PUSH (notwordbound);
3700 break;
3701
3702 case '`':
3703 if (syntax & RE_NO_GNU_OPS)
3704 goto normal_char;
3705 BUF_PUSH (begbuf);
3706 break;
3707
3708 case '\'':
3709 if (syntax & RE_NO_GNU_OPS)
3710 goto normal_char;
3711 BUF_PUSH (endbuf);
3712 break;
3713
3714 case '1': case '2': case '3': case '4': case '5':
3715 case '6': case '7': case '8': case '9':
3716 {
3717 regnum_t reg;
3718
3719 if (syntax & RE_NO_BK_REFS)
3720 goto normal_backslash;
3721
3722 reg = c - '0';
3723
3724 if (reg > bufp->re_nsub || reg < 1
3725 /* Can't back reference to a subexp before its end. */
3726 || group_in_compile_stack (compile_stack, reg))
3727 FREE_STACK_RETURN (REG_ESUBREG);
3728
3729 laststart = b;
3730 BUF_PUSH_2 (duplicate, reg);
3731 }
3732 break;
3733
3734
3735 case '+':
3736 case '?':
3737 if (syntax & RE_BK_PLUS_QM)
3738 goto handle_plus;
3739 else
3740 goto normal_backslash;
3741
3742 default:
3743 normal_backslash:
3744 /* You might think it would be useful for \ to mean
3745 not to translate; but if we don't translate it
3746 it will never match anything. */
3747 goto normal_char;
3748 }
3749 break;
3750
3751
3752 default:
3753 /* Expects the character in `c'. */
3754 normal_char:
3755 /* If no exactn currently being built. */
3756 if (!pending_exact
3757
3758 /* If last exactn not at current position. */
3759 || pending_exact + *pending_exact + 1 != b
3760
3761 /* We have only one byte following the exactn for the count. */
3762 || *pending_exact >= (1 << BYTEWIDTH) - MAX_MULTIBYTE_LENGTH
3763
3764 /* If followed by a repetition operator. */
3765 || (p != pend && (*p == '*' || *p == '^'))
3766 || ((syntax & RE_BK_PLUS_QM)
3767 ? p + 1 < pend && *p == '\\' && (p[1] == '+' || p[1] == '?')
3768 : p != pend && (*p == '+' || *p == '?'))
3769 || ((syntax & RE_INTERVALS)
3770 && ((syntax & RE_NO_BK_BRACES)
3771 ? p != pend && *p == '{'
3772 : p + 1 < pend && p[0] == '\\' && p[1] == '{')))
3773 {
3774 /* Start building a new exactn. */
3775
3776 laststart = b;
3777
3778 BUF_PUSH_2 (exactn, 0);
3779 pending_exact = b - 1;
3780 }
3781
3782 GET_BUFFER_SPACE (MAX_MULTIBYTE_LENGTH);
3783 {
3784 int len;
3785
3786 if (multibyte)
3787 {
3788 c = TRANSLATE (c);
3789 len = CHAR_STRING (c, b);
3790 b += len;
3791 }
3792 else
3793 {
3794 c1 = RE_CHAR_TO_MULTIBYTE (c);
3795 if (! CHAR_BYTE8_P (c1))
3796 {
3797 re_wchar_t c2 = TRANSLATE (c1);
3798
3799 if (c1 != c2 && (c1 = RE_CHAR_TO_UNIBYTE (c2)) >= 0)
3800 c = c1;
3801 }
3802 *b++ = c;
3803 len = 1;
3804 }
3805 (*pending_exact) += len;
3806 }
3807
3808 break;
3809 } /* switch (c) */
3810 } /* while p != pend */
3811
3812
3813 /* Through the pattern now. */
3814
3815 FIXUP_ALT_JUMP ();
3816
3817 if (!COMPILE_STACK_EMPTY)
3818 FREE_STACK_RETURN (REG_EPAREN);
3819
3820 /* If we don't want backtracking, force success
3821 the first time we reach the end of the compiled pattern. */
3822 if (syntax & RE_NO_POSIX_BACKTRACKING)
3823 BUF_PUSH (succeed);
3824
3825 /* We have succeeded; set the length of the buffer. */
3826 bufp->used = b - bufp->buffer;
3827
3828 #ifdef DEBUG
3829 if (debug > 0)
3830 {
3831 re_compile_fastmap (bufp);
3832 DEBUG_PRINT1 ("\nCompiled pattern: \n");
3833 print_compiled_pattern (bufp);
3834 }
3835 debug--;
3836 #endif /* DEBUG */
3837
3838 #ifndef MATCH_MAY_ALLOCATE
3839 /* Initialize the failure stack to the largest possible stack. This
3840 isn't necessary unless we're trying to avoid calling alloca in
3841 the search and match routines. */
3842 {
3843 int num_regs = bufp->re_nsub + 1;
3844
3845 if (fail_stack.size < re_max_failures * TYPICAL_FAILURE_SIZE)
3846 {
3847 fail_stack.size = re_max_failures * TYPICAL_FAILURE_SIZE;
3848
3849 if (! fail_stack.stack)
3850 fail_stack.stack
3851 = (fail_stack_elt_t *) malloc (fail_stack.size
3852 * sizeof (fail_stack_elt_t));
3853 else
3854 fail_stack.stack
3855 = (fail_stack_elt_t *) realloc (fail_stack.stack,
3856 (fail_stack.size
3857 * sizeof (fail_stack_elt_t)));
3858 }
3859
3860 regex_grow_registers (num_regs);
3861 }
3862 #endif /* not MATCH_MAY_ALLOCATE */
3863
3864 FREE_STACK_RETURN (REG_NOERROR);
3865 } /* regex_compile */
3866 \f
3867 /* Subroutines for `regex_compile'. */
3868
3869 /* Store OP at LOC followed by two-byte integer parameter ARG. */
3870
3871 static void
3872 store_op1 (op, loc, arg)
3873 re_opcode_t op;
3874 unsigned char *loc;
3875 int arg;
3876 {
3877 *loc = (unsigned char) op;
3878 STORE_NUMBER (loc + 1, arg);
3879 }
3880
3881
3882 /* Like `store_op1', but for two two-byte parameters ARG1 and ARG2. */
3883
3884 static void
3885 store_op2 (op, loc, arg1, arg2)
3886 re_opcode_t op;
3887 unsigned char *loc;
3888 int arg1, arg2;
3889 {
3890 *loc = (unsigned char) op;
3891 STORE_NUMBER (loc + 1, arg1);
3892 STORE_NUMBER (loc + 3, arg2);
3893 }
3894
3895
3896 /* Copy the bytes from LOC to END to open up three bytes of space at LOC
3897 for OP followed by two-byte integer parameter ARG. */
3898
3899 static void
3900 insert_op1 (op, loc, arg, end)
3901 re_opcode_t op;
3902 unsigned char *loc;
3903 int arg;
3904 unsigned char *end;
3905 {
3906 register unsigned char *pfrom = end;
3907 register unsigned char *pto = end + 3;
3908
3909 while (pfrom != loc)
3910 *--pto = *--pfrom;
3911
3912 store_op1 (op, loc, arg);
3913 }
3914
3915
3916 /* Like `insert_op1', but for two two-byte parameters ARG1 and ARG2. */
3917
3918 static void
3919 insert_op2 (op, loc, arg1, arg2, end)
3920 re_opcode_t op;
3921 unsigned char *loc;
3922 int arg1, arg2;
3923 unsigned char *end;
3924 {
3925 register unsigned char *pfrom = end;
3926 register unsigned char *pto = end + 5;
3927
3928 while (pfrom != loc)
3929 *--pto = *--pfrom;
3930
3931 store_op2 (op, loc, arg1, arg2);
3932 }
3933
3934
3935 /* P points to just after a ^ in PATTERN. Return true if that ^ comes
3936 after an alternative or a begin-subexpression. We assume there is at
3937 least one character before the ^. */
3938
3939 static boolean
3940 at_begline_loc_p (pattern, p, syntax)
3941 re_char *pattern, *p;
3942 reg_syntax_t syntax;
3943 {
3944 re_char *prev = p - 2;
3945 boolean prev_prev_backslash = prev > pattern && prev[-1] == '\\';
3946
3947 return
3948 /* After a subexpression? */
3949 (*prev == '(' && (syntax & RE_NO_BK_PARENS || prev_prev_backslash))
3950 /* After an alternative? */
3951 || (*prev == '|' && (syntax & RE_NO_BK_VBAR || prev_prev_backslash))
3952 /* After a shy subexpression? */
3953 || ((syntax & RE_SHY_GROUPS) && prev - 2 >= pattern
3954 && prev[-1] == '?' && prev[-2] == '('
3955 && (syntax & RE_NO_BK_PARENS
3956 || (prev - 3 >= pattern && prev[-3] == '\\')));
3957 }
3958
3959
3960 /* The dual of at_begline_loc_p. This one is for $. We assume there is
3961 at least one character after the $, i.e., `P < PEND'. */
3962
3963 static boolean
3964 at_endline_loc_p (p, pend, syntax)
3965 re_char *p, *pend;
3966 reg_syntax_t syntax;
3967 {
3968 re_char *next = p;
3969 boolean next_backslash = *next == '\\';
3970 re_char *next_next = p + 1 < pend ? p + 1 : 0;
3971
3972 return
3973 /* Before a subexpression? */
3974 (syntax & RE_NO_BK_PARENS ? *next == ')'
3975 : next_backslash && next_next && *next_next == ')')
3976 /* Before an alternative? */
3977 || (syntax & RE_NO_BK_VBAR ? *next == '|'
3978 : next_backslash && next_next && *next_next == '|');
3979 }
3980
3981
3982 /* Returns true if REGNUM is in one of COMPILE_STACK's elements and
3983 false if it's not. */
3984
3985 static boolean
3986 group_in_compile_stack (compile_stack, regnum)
3987 compile_stack_type compile_stack;
3988 regnum_t regnum;
3989 {
3990 int this_element;
3991
3992 for (this_element = compile_stack.avail - 1;
3993 this_element >= 0;
3994 this_element--)
3995 if (compile_stack.stack[this_element].regnum == regnum)
3996 return true;
3997
3998 return false;
3999 }
4000 \f
4001 /* analyse_first.
4002 If fastmap is non-NULL, go through the pattern and fill fastmap
4003 with all the possible leading chars. If fastmap is NULL, don't
4004 bother filling it up (obviously) and only return whether the
4005 pattern could potentially match the empty string.
4006
4007 Return 1 if p..pend might match the empty string.
4008 Return 0 if p..pend matches at least one char.
4009 Return -1 if fastmap was not updated accurately. */
4010
4011 static int
4012 analyse_first (p, pend, fastmap, multibyte)
4013 re_char *p, *pend;
4014 char *fastmap;
4015 const int multibyte;
4016 {
4017 int j, k;
4018 boolean not;
4019
4020 /* If all elements for base leading-codes in fastmap is set, this
4021 flag is set true. */
4022 boolean match_any_multibyte_characters = false;
4023
4024 assert (p);
4025
4026 /* The loop below works as follows:
4027 - It has a working-list kept in the PATTERN_STACK and which basically
4028 starts by only containing a pointer to the first operation.
4029 - If the opcode we're looking at is a match against some set of
4030 chars, then we add those chars to the fastmap and go on to the
4031 next work element from the worklist (done via `break').
4032 - If the opcode is a control operator on the other hand, we either
4033 ignore it (if it's meaningless at this point, such as `start_memory')
4034 or execute it (if it's a jump). If the jump has several destinations
4035 (i.e. `on_failure_jump'), then we push the other destination onto the
4036 worklist.
4037 We guarantee termination by ignoring backward jumps (more or less),
4038 so that `p' is monotonically increasing. More to the point, we
4039 never set `p' (or push) anything `<= p1'. */
4040
4041 while (p < pend)
4042 {
4043 /* `p1' is used as a marker of how far back a `on_failure_jump'
4044 can go without being ignored. It is normally equal to `p'
4045 (which prevents any backward `on_failure_jump') except right
4046 after a plain `jump', to allow patterns such as:
4047 0: jump 10
4048 3..9: <body>
4049 10: on_failure_jump 3
4050 as used for the *? operator. */
4051 re_char *p1 = p;
4052
4053 switch (SWITCH_ENUM_CAST ((re_opcode_t) *p++))
4054 {
4055 case succeed:
4056 return 1;
4057 continue;
4058
4059 case duplicate:
4060 /* If the first character has to match a backreference, that means
4061 that the group was empty (since it already matched). Since this
4062 is the only case that interests us here, we can assume that the
4063 backreference must match the empty string. */
4064 p++;
4065 continue;
4066
4067
4068 /* Following are the cases which match a character. These end
4069 with `break'. */
4070
4071 case exactn:
4072 if (fastmap)
4073 {
4074 /* If multibyte is nonzero, the first byte of each
4075 character is an ASCII or a leading code. Otherwise,
4076 each byte is a character. Thus, this works in both
4077 cases. */
4078 fastmap[p[1]] = 1;
4079 if (! multibyte)
4080 {
4081 /* For the case of matching this unibyte regex
4082 against multibyte, we must set a leading code of
4083 the corresponding multibyte character. */
4084 int c = RE_CHAR_TO_MULTIBYTE (p[1]);
4085
4086 if (! CHAR_BYTE8_P (c))
4087 fastmap[CHAR_LEADING_CODE (c)] = 1;
4088 }
4089 }
4090 break;
4091
4092
4093 case anychar:
4094 /* We could put all the chars except for \n (and maybe \0)
4095 but we don't bother since it is generally not worth it. */
4096 if (!fastmap) break;
4097 return -1;
4098
4099
4100 case charset_not:
4101 if (!fastmap) break;
4102 {
4103 /* Chars beyond end of bitmap are possible matches. */
4104 for (j = CHARSET_BITMAP_SIZE (&p[-1]) * BYTEWIDTH;
4105 j < (1 << BYTEWIDTH); j++)
4106 fastmap[j] = 1;
4107 }
4108
4109 /* Fallthrough */
4110 case charset:
4111 if (!fastmap) break;
4112 not = (re_opcode_t) *(p - 1) == charset_not;
4113 for (j = CHARSET_BITMAP_SIZE (&p[-1]) * BYTEWIDTH - 1, p++;
4114 j >= 0; j--)
4115 if (!!(p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH))) ^ not)
4116 fastmap[j] = 1;
4117
4118 #ifdef emacs
4119 if (/* Any leading code can possibly start a character
4120 which doesn't match the specified set of characters. */
4121 not
4122 ||
4123 /* If we can match a character class, we can match any
4124 multibyte characters. */
4125 (CHARSET_RANGE_TABLE_EXISTS_P (&p[-2])
4126 && CHARSET_RANGE_TABLE_BITS (&p[-2]) != 0))
4127
4128 {
4129 if (match_any_multibyte_characters == false)
4130 {
4131 for (j = MIN_MULTIBYTE_LEADING_CODE;
4132 j <= MAX_MULTIBYTE_LEADING_CODE; j++)
4133 fastmap[j] = 1;
4134 match_any_multibyte_characters = true;
4135 }
4136 }
4137
4138 else if (!not && CHARSET_RANGE_TABLE_EXISTS_P (&p[-2])
4139 && match_any_multibyte_characters == false)
4140 {
4141 /* Set fastmap[I] to 1 where I is a leading code of each
4142 multibyte characer in the range table. */
4143 int c, count;
4144 unsigned char lc1, lc2;
4145
4146 /* Make P points the range table. `+ 2' is to skip flag
4147 bits for a character class. */
4148 p += CHARSET_BITMAP_SIZE (&p[-2]) + 2;
4149
4150 /* Extract the number of ranges in range table into COUNT. */
4151 EXTRACT_NUMBER_AND_INCR (count, p);
4152 for (; count > 0; count--, p += 3)
4153 {
4154 /* Extract the start and end of each range. */
4155 EXTRACT_CHARACTER (c, p);
4156 lc1 = CHAR_LEADING_CODE (c);
4157 p += 3;
4158 EXTRACT_CHARACTER (c, p);
4159 lc2 = CHAR_LEADING_CODE (c);
4160 for (j = lc1; j <= lc2; j++)
4161 fastmap[j] = 1;
4162 }
4163 }
4164 #endif
4165 break;
4166
4167 case syntaxspec:
4168 case notsyntaxspec:
4169 if (!fastmap) break;
4170 #ifndef emacs
4171 not = (re_opcode_t)p[-1] == notsyntaxspec;
4172 k = *p++;
4173 for (j = 0; j < (1 << BYTEWIDTH); j++)
4174 if ((SYNTAX (j) == (enum syntaxcode) k) ^ not)
4175 fastmap[j] = 1;
4176 break;
4177 #else /* emacs */
4178 /* This match depends on text properties. These end with
4179 aborting optimizations. */
4180 return -1;
4181
4182 case categoryspec:
4183 case notcategoryspec:
4184 if (!fastmap) break;
4185 not = (re_opcode_t)p[-1] == notcategoryspec;
4186 k = *p++;
4187 for (j = (1 << BYTEWIDTH); j >= 0; j--)
4188 if ((CHAR_HAS_CATEGORY (j, k)) ^ not)
4189 fastmap[j] = 1;
4190
4191 /* Any leading code can possibly start a character which
4192 has or doesn't has the specified category. */
4193 if (match_any_multibyte_characters == false)
4194 {
4195 for (j = MIN_MULTIBYTE_LEADING_CODE;
4196 j <= MAX_MULTIBYTE_LEADING_CODE; j++)
4197 fastmap[j] = 1;
4198 match_any_multibyte_characters = true;
4199 }
4200 break;
4201
4202 /* All cases after this match the empty string. These end with
4203 `continue'. */
4204
4205 case before_dot:
4206 case at_dot:
4207 case after_dot:
4208 #endif /* !emacs */
4209 case no_op:
4210 case begline:
4211 case endline:
4212 case begbuf:
4213 case endbuf:
4214 case wordbound:
4215 case notwordbound:
4216 case wordbeg:
4217 case wordend:
4218 case symbeg:
4219 case symend:
4220 continue;
4221
4222
4223 case jump:
4224 EXTRACT_NUMBER_AND_INCR (j, p);
4225 if (j < 0)
4226 /* Backward jumps can only go back to code that we've already
4227 visited. `re_compile' should make sure this is true. */
4228 break;
4229 p += j;
4230 switch (SWITCH_ENUM_CAST ((re_opcode_t) *p))
4231 {
4232 case on_failure_jump:
4233 case on_failure_keep_string_jump:
4234 case on_failure_jump_loop:
4235 case on_failure_jump_nastyloop:
4236 case on_failure_jump_smart:
4237 p++;
4238 break;
4239 default:
4240 continue;
4241 };
4242 /* Keep `p1' to allow the `on_failure_jump' we are jumping to
4243 to jump back to "just after here". */
4244 /* Fallthrough */
4245
4246 case on_failure_jump:
4247 case on_failure_keep_string_jump:
4248 case on_failure_jump_nastyloop:
4249 case on_failure_jump_loop:
4250 case on_failure_jump_smart:
4251 EXTRACT_NUMBER_AND_INCR (j, p);
4252 if (p + j <= p1)
4253 ; /* Backward jump to be ignored. */
4254 else
4255 { /* We have to look down both arms.
4256 We first go down the "straight" path so as to minimize
4257 stack usage when going through alternatives. */
4258 int r = analyse_first (p, pend, fastmap, multibyte);
4259 if (r) return r;
4260 p += j;
4261 }
4262 continue;
4263
4264
4265 case jump_n:
4266 /* This code simply does not properly handle forward jump_n. */
4267 DEBUG_STATEMENT (EXTRACT_NUMBER (j, p); assert (j < 0));
4268 p += 4;
4269 /* jump_n can either jump or fall through. The (backward) jump
4270 case has already been handled, so we only need to look at the
4271 fallthrough case. */
4272 continue;
4273
4274 case succeed_n:
4275 /* If N == 0, it should be an on_failure_jump_loop instead. */
4276 DEBUG_STATEMENT (EXTRACT_NUMBER (j, p + 2); assert (j > 0));
4277 p += 4;
4278 /* We only care about one iteration of the loop, so we don't
4279 need to consider the case where this behaves like an
4280 on_failure_jump. */
4281 continue;
4282
4283
4284 case set_number_at:
4285 p += 4;
4286 continue;
4287
4288
4289 case start_memory:
4290 case stop_memory:
4291 p += 1;
4292 continue;
4293
4294
4295 default:
4296 abort (); /* We have listed all the cases. */
4297 } /* switch *p++ */
4298
4299 /* Getting here means we have found the possible starting
4300 characters for one path of the pattern -- and that the empty
4301 string does not match. We need not follow this path further. */
4302 return 0;
4303 } /* while p */
4304
4305 /* We reached the end without matching anything. */
4306 return 1;
4307
4308 } /* analyse_first */
4309 \f
4310 /* re_compile_fastmap computes a ``fastmap'' for the compiled pattern in
4311 BUFP. A fastmap records which of the (1 << BYTEWIDTH) possible
4312 characters can start a string that matches the pattern. This fastmap
4313 is used by re_search to skip quickly over impossible starting points.
4314
4315 Character codes above (1 << BYTEWIDTH) are not represented in the
4316 fastmap, but the leading codes are represented. Thus, the fastmap
4317 indicates which character sets could start a match.
4318
4319 The caller must supply the address of a (1 << BYTEWIDTH)-byte data
4320 area as BUFP->fastmap.
4321
4322 We set the `fastmap', `fastmap_accurate', and `can_be_null' fields in
4323 the pattern buffer.
4324
4325 Returns 0 if we succeed, -2 if an internal error. */
4326
4327 int
4328 re_compile_fastmap (bufp)
4329 struct re_pattern_buffer *bufp;
4330 {
4331 char *fastmap = bufp->fastmap;
4332 int analysis;
4333
4334 assert (fastmap && bufp->buffer);
4335
4336 bzero (fastmap, 1 << BYTEWIDTH); /* Assume nothing's valid. */
4337 bufp->fastmap_accurate = 1; /* It will be when we're done. */
4338
4339 analysis = analyse_first (bufp->buffer, bufp->buffer + bufp->used,
4340 fastmap, RE_MULTIBYTE_P (bufp));
4341 bufp->can_be_null = (analysis != 0);
4342 return 0;
4343 } /* re_compile_fastmap */
4344 \f
4345 /* Set REGS to hold NUM_REGS registers, storing them in STARTS and
4346 ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use
4347 this memory for recording register information. STARTS and ENDS
4348 must be allocated using the malloc library routine, and must each
4349 be at least NUM_REGS * sizeof (regoff_t) bytes long.
4350
4351 If NUM_REGS == 0, then subsequent matches should allocate their own
4352 register data.
4353
4354 Unless this function is called, the first search or match using
4355 PATTERN_BUFFER will allocate its own register data, without
4356 freeing the old data. */
4357
4358 void
4359 re_set_registers (bufp, regs, num_regs, starts, ends)
4360 struct re_pattern_buffer *bufp;
4361 struct re_registers *regs;
4362 unsigned num_regs;
4363 regoff_t *starts, *ends;
4364 {
4365 if (num_regs)
4366 {
4367 bufp->regs_allocated = REGS_REALLOCATE;
4368 regs->num_regs = num_regs;
4369 regs->start = starts;
4370 regs->end = ends;
4371 }
4372 else
4373 {
4374 bufp->regs_allocated = REGS_UNALLOCATED;
4375 regs->num_regs = 0;
4376 regs->start = regs->end = (regoff_t *) 0;
4377 }
4378 }
4379 WEAK_ALIAS (__re_set_registers, re_set_registers)
4380 \f
4381 /* Searching routines. */
4382
4383 /* Like re_search_2, below, but only one string is specified, and
4384 doesn't let you say where to stop matching. */
4385
4386 int
4387 re_search (bufp, string, size, startpos, range, regs)
4388 struct re_pattern_buffer *bufp;
4389 const char *string;
4390 int size, startpos, range;
4391 struct re_registers *regs;
4392 {
4393 return re_search_2 (bufp, NULL, 0, string, size, startpos, range,
4394 regs, size);
4395 }
4396 WEAK_ALIAS (__re_search, re_search)
4397
4398 /* Head address of virtual concatenation of string. */
4399 #define HEAD_ADDR_VSTRING(P) \
4400 (((P) >= size1 ? string2 : string1))
4401
4402 /* End address of virtual concatenation of string. */
4403 #define STOP_ADDR_VSTRING(P) \
4404 (((P) >= size1 ? string2 + size2 : string1 + size1))
4405
4406 /* Address of POS in the concatenation of virtual string. */
4407 #define POS_ADDR_VSTRING(POS) \
4408 (((POS) >= size1 ? string2 - size1 : string1) + (POS))
4409
4410 /* Using the compiled pattern in BUFP->buffer, first tries to match the
4411 virtual concatenation of STRING1 and STRING2, starting first at index
4412 STARTPOS, then at STARTPOS + 1, and so on.
4413
4414 STRING1 and STRING2 have length SIZE1 and SIZE2, respectively.
4415
4416 RANGE is how far to scan while trying to match. RANGE = 0 means try
4417 only at STARTPOS; in general, the last start tried is STARTPOS +
4418 RANGE.
4419
4420 In REGS, return the indices of the virtual concatenation of STRING1
4421 and STRING2 that matched the entire BUFP->buffer and its contained
4422 subexpressions.
4423
4424 Do not consider matching one past the index STOP in the virtual
4425 concatenation of STRING1 and STRING2.
4426
4427 We return either the position in the strings at which the match was
4428 found, -1 if no match, or -2 if error (such as failure
4429 stack overflow). */
4430
4431 int
4432 re_search_2 (bufp, str1, size1, str2, size2, startpos, range, regs, stop)
4433 struct re_pattern_buffer *bufp;
4434 const char *str1, *str2;
4435 int size1, size2;
4436 int startpos;
4437 int range;
4438 struct re_registers *regs;
4439 int stop;
4440 {
4441 int val;
4442 re_char *string1 = (re_char*) str1;
4443 re_char *string2 = (re_char*) str2;
4444 register char *fastmap = bufp->fastmap;
4445 register RE_TRANSLATE_TYPE translate = bufp->translate;
4446 int total_size = size1 + size2;
4447 int endpos = startpos + range;
4448 boolean anchored_start;
4449 /* Nonzero if we are searching multibyte string. */
4450 const boolean multibyte = RE_TARGET_MULTIBYTE_P (bufp);
4451
4452 /* Check for out-of-range STARTPOS. */
4453 if (startpos < 0 || startpos > total_size)
4454 return -1;
4455
4456 /* Fix up RANGE if it might eventually take us outside
4457 the virtual concatenation of STRING1 and STRING2.
4458 Make sure we won't move STARTPOS below 0 or above TOTAL_SIZE. */
4459 if (endpos < 0)
4460 range = 0 - startpos;
4461 else if (endpos > total_size)
4462 range = total_size - startpos;
4463
4464 /* If the search isn't to be a backwards one, don't waste time in a
4465 search for a pattern anchored at beginning of buffer. */
4466 if (bufp->used > 0 && (re_opcode_t) bufp->buffer[0] == begbuf && range > 0)
4467 {
4468 if (startpos > 0)
4469 return -1;
4470 else
4471 range = 0;
4472 }
4473
4474 #ifdef emacs
4475 /* In a forward search for something that starts with \=.
4476 don't keep searching past point. */
4477 if (bufp->used > 0 && (re_opcode_t) bufp->buffer[0] == at_dot && range > 0)
4478 {
4479 range = PT_BYTE - BEGV_BYTE - startpos;
4480 if (range < 0)
4481 return -1;
4482 }
4483 #endif /* emacs */
4484
4485 /* Update the fastmap now if not correct already. */
4486 if (fastmap && !bufp->fastmap_accurate)
4487 re_compile_fastmap (bufp);
4488
4489 /* See whether the pattern is anchored. */
4490 anchored_start = (bufp->buffer[0] == begline);
4491
4492 #ifdef emacs
4493 gl_state.object = re_match_object;
4494 {
4495 int charpos = SYNTAX_TABLE_BYTE_TO_CHAR (POS_AS_IN_BUFFER (startpos));
4496
4497 SETUP_SYNTAX_TABLE_FOR_OBJECT (re_match_object, charpos, 1);
4498 }
4499 #endif
4500
4501 /* Loop through the string, looking for a place to start matching. */
4502 for (;;)
4503 {
4504 /* If the pattern is anchored,
4505 skip quickly past places we cannot match.
4506 We don't bother to treat startpos == 0 specially
4507 because that case doesn't repeat. */
4508 if (anchored_start && startpos > 0)
4509 {
4510 if (! ((startpos <= size1 ? string1[startpos - 1]
4511 : string2[startpos - size1 - 1])
4512 == '\n'))
4513 goto advance;
4514 }
4515
4516 /* If a fastmap is supplied, skip quickly over characters that
4517 cannot be the start of a match. If the pattern can match the
4518 null string, however, we don't need to skip characters; we want
4519 the first null string. */
4520 if (fastmap && startpos < total_size && !bufp->can_be_null)
4521 {
4522 register re_char *d;
4523 register re_wchar_t buf_ch;
4524
4525 d = POS_ADDR_VSTRING (startpos);
4526
4527 if (range > 0) /* Searching forwards. */
4528 {
4529 register int lim = 0;
4530 int irange = range;
4531
4532 if (startpos < size1 && startpos + range >= size1)
4533 lim = range - (size1 - startpos);
4534
4535 /* Written out as an if-else to avoid testing `translate'
4536 inside the loop. */
4537 if (RE_TRANSLATE_P (translate))
4538 {
4539 if (multibyte)
4540 while (range > lim)
4541 {
4542 int buf_charlen;
4543
4544 buf_ch = STRING_CHAR_AND_LENGTH (d, 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, buf_charlen);
4577 if (fastmap[CHAR_LEADING_CODE (buf_ch)])
4578 break;
4579 range -= buf_charlen;
4580 d += buf_charlen;
4581 }
4582 else
4583 while (range > lim && !fastmap[*d])
4584 {
4585 d++;
4586 range--;
4587 }
4588 }
4589 startpos += irange - range;
4590 }
4591 else /* Searching backwards. */
4592 {
4593 if (multibyte)
4594 {
4595 buf_ch = STRING_CHAR (d);
4596 buf_ch = TRANSLATE (buf_ch);
4597 if (! fastmap[CHAR_LEADING_CODE (buf_ch)])
4598 goto advance;
4599 }
4600 else
4601 {
4602 register re_wchar_t ch, translated;
4603
4604 buf_ch = *d;
4605 ch = RE_CHAR_TO_MULTIBYTE (buf_ch);
4606 translated = TRANSLATE (ch);
4607 if (translated != ch
4608 && (ch = RE_CHAR_TO_UNIBYTE (translated)) >= 0)
4609 buf_ch = ch;
4610 if (! fastmap[TRANSLATE (buf_ch)])
4611 goto advance;
4612 }
4613 }
4614 }
4615
4616 /* If can't match the null string, and that's all we have left, fail. */
4617 if (range >= 0 && startpos == total_size && fastmap
4618 && !bufp->can_be_null)
4619 return -1;
4620
4621 val = re_match_2_internal (bufp, string1, size1, string2, size2,
4622 startpos, regs, stop);
4623
4624 if (val >= 0)
4625 return startpos;
4626
4627 if (val == -2)
4628 return -2;
4629
4630 advance:
4631 if (!range)
4632 break;
4633 else if (range > 0)
4634 {
4635 /* Update STARTPOS to the next character boundary. */
4636 if (multibyte)
4637 {
4638 re_char *p = POS_ADDR_VSTRING (startpos);
4639 re_char *pend = STOP_ADDR_VSTRING (startpos);
4640 int len = MULTIBYTE_FORM_LENGTH (p, pend - p);
4641
4642 range -= len;
4643 if (range < 0)
4644 break;
4645 startpos += len;
4646 }
4647 else
4648 {
4649 range--;
4650 startpos++;
4651 }
4652 }
4653 else
4654 {
4655 range++;
4656 startpos--;
4657
4658 /* Update STARTPOS to the previous character boundary. */
4659 if (multibyte)
4660 {
4661 re_char *p = POS_ADDR_VSTRING (startpos) + 1;
4662 re_char *p0 = p;
4663 re_char *phead = HEAD_ADDR_VSTRING (startpos);
4664
4665 /* Find the head of multibyte form. */
4666 PREV_CHAR_BOUNDARY (p, phead);
4667 range += p0 - 1 - p;
4668 if (range > 0)
4669 break;
4670
4671 startpos -= p0 - 1 - p;
4672 }
4673 }
4674 }
4675 return -1;
4676 } /* re_search_2 */
4677 WEAK_ALIAS (__re_search_2, re_search_2)
4678 \f
4679 /* Declarations and macros for re_match_2. */
4680
4681 static int bcmp_translate _RE_ARGS((re_char *s1, re_char *s2,
4682 register int len,
4683 RE_TRANSLATE_TYPE translate,
4684 const int multibyte));
4685
4686 /* This converts PTR, a pointer into one of the search strings `string1'
4687 and `string2' into an offset from the beginning of that string. */
4688 #define POINTER_TO_OFFSET(ptr) \
4689 (FIRST_STRING_P (ptr) \
4690 ? ((regoff_t) ((ptr) - string1)) \
4691 : ((regoff_t) ((ptr) - string2 + size1)))
4692
4693 /* Call before fetching a character with *d. This switches over to
4694 string2 if necessary.
4695 Check re_match_2_internal for a discussion of why end_match_2 might
4696 not be within string2 (but be equal to end_match_1 instead). */
4697 #define PREFETCH() \
4698 while (d == dend) \
4699 { \
4700 /* End of string2 => fail. */ \
4701 if (dend == end_match_2) \
4702 goto fail; \
4703 /* End of string1 => advance to string2. */ \
4704 d = string2; \
4705 dend = end_match_2; \
4706 }
4707
4708 /* Call before fetching a char with *d if you already checked other limits.
4709 This is meant for use in lookahead operations like wordend, etc..
4710 where we might need to look at parts of the string that might be
4711 outside of the LIMITs (i.e past `stop'). */
4712 #define PREFETCH_NOLIMIT() \
4713 if (d == end1) \
4714 { \
4715 d = string2; \
4716 dend = end_match_2; \
4717 } \
4718
4719 /* Test if at very beginning or at very end of the virtual concatenation
4720 of `string1' and `string2'. If only one string, it's `string2'. */
4721 #define AT_STRINGS_BEG(d) ((d) == (size1 ? string1 : string2) || !size2)
4722 #define AT_STRINGS_END(d) ((d) == end2)
4723
4724
4725 /* Test if D points to a character which is word-constituent. We have
4726 two special cases to check for: if past the end of string1, look at
4727 the first character in string2; and if before the beginning of
4728 string2, look at the last character in string1. */
4729 #define WORDCHAR_P(d) \
4730 (SYNTAX ((d) == end1 ? *string2 \
4731 : (d) == string2 - 1 ? *(end1 - 1) : *(d)) \
4732 == Sword)
4733
4734 /* Disabled due to a compiler bug -- see comment at case wordbound */
4735
4736 /* The comment at case wordbound is following one, but we don't use
4737 AT_WORD_BOUNDARY anymore to support multibyte form.
4738
4739 The DEC Alpha C compiler 3.x generates incorrect code for the
4740 test WORDCHAR_P (d - 1) != WORDCHAR_P (d) in the expansion of
4741 AT_WORD_BOUNDARY, so this code is disabled. Expanding the
4742 macro and introducing temporary variables works around the bug. */
4743
4744 #if 0
4745 /* Test if the character before D and the one at D differ with respect
4746 to being word-constituent. */
4747 #define AT_WORD_BOUNDARY(d) \
4748 (AT_STRINGS_BEG (d) || AT_STRINGS_END (d) \
4749 || WORDCHAR_P (d - 1) != WORDCHAR_P (d))
4750 #endif
4751
4752 /* Free everything we malloc. */
4753 #ifdef MATCH_MAY_ALLOCATE
4754 # define FREE_VAR(var) if (var) { REGEX_FREE (var); var = NULL; } else
4755 # define FREE_VARIABLES() \
4756 do { \
4757 REGEX_FREE_STACK (fail_stack.stack); \
4758 FREE_VAR (regstart); \
4759 FREE_VAR (regend); \
4760 FREE_VAR (best_regstart); \
4761 FREE_VAR (best_regend); \
4762 } while (0)
4763 #else
4764 # define FREE_VARIABLES() ((void)0) /* Do nothing! But inhibit gcc warning. */
4765 #endif /* not MATCH_MAY_ALLOCATE */
4766
4767 \f
4768 /* Optimization routines. */
4769
4770 /* If the operation is a match against one or more chars,
4771 return a pointer to the next operation, else return NULL. */
4772 static re_char *
4773 skip_one_char (p)
4774 re_char *p;
4775 {
4776 switch (SWITCH_ENUM_CAST (*p++))
4777 {
4778 case anychar:
4779 break;
4780
4781 case exactn:
4782 p += *p + 1;
4783 break;
4784
4785 case charset_not:
4786 case charset:
4787 if (CHARSET_RANGE_TABLE_EXISTS_P (p - 1))
4788 {
4789 int mcnt;
4790 p = CHARSET_RANGE_TABLE (p - 1);
4791 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4792 p = CHARSET_RANGE_TABLE_END (p, mcnt);
4793 }
4794 else
4795 p += 1 + CHARSET_BITMAP_SIZE (p - 1);
4796 break;
4797
4798 case syntaxspec:
4799 case notsyntaxspec:
4800 #ifdef emacs
4801 case categoryspec:
4802 case notcategoryspec:
4803 #endif /* emacs */
4804 p++;
4805 break;
4806
4807 default:
4808 p = NULL;
4809 }
4810 return p;
4811 }
4812
4813
4814 /* Jump over non-matching operations. */
4815 static re_char *
4816 skip_noops (p, pend)
4817 re_char *p, *pend;
4818 {
4819 int mcnt;
4820 while (p < pend)
4821 {
4822 switch (SWITCH_ENUM_CAST ((re_opcode_t) *p))
4823 {
4824 case start_memory:
4825 case stop_memory:
4826 p += 2; break;
4827 case no_op:
4828 p += 1; break;
4829 case jump:
4830 p += 1;
4831 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4832 p += mcnt;
4833 break;
4834 default:
4835 return p;
4836 }
4837 }
4838 assert (p == pend);
4839 return p;
4840 }
4841
4842 /* Non-zero if "p1 matches something" implies "p2 fails". */
4843 static int
4844 mutually_exclusive_p (bufp, p1, p2)
4845 struct re_pattern_buffer *bufp;
4846 re_char *p1, *p2;
4847 {
4848 re_opcode_t op2;
4849 const boolean multibyte = RE_MULTIBYTE_P (bufp);
4850 unsigned char *pend = bufp->buffer + bufp->used;
4851
4852 assert (p1 >= bufp->buffer && p1 < pend
4853 && p2 >= bufp->buffer && p2 <= pend);
4854
4855 /* Skip over open/close-group commands.
4856 If what follows this loop is a ...+ construct,
4857 look at what begins its body, since we will have to
4858 match at least one of that. */
4859 p2 = skip_noops (p2, pend);
4860 /* The same skip can be done for p1, except that this function
4861 is only used in the case where p1 is a simple match operator. */
4862 /* p1 = skip_noops (p1, pend); */
4863
4864 assert (p1 >= bufp->buffer && p1 < pend
4865 && p2 >= bufp->buffer && p2 <= pend);
4866
4867 op2 = p2 == pend ? succeed : *p2;
4868
4869 switch (SWITCH_ENUM_CAST (op2))
4870 {
4871 case succeed:
4872 case endbuf:
4873 /* If we're at the end of the pattern, we can change. */
4874 if (skip_one_char (p1))
4875 {
4876 DEBUG_PRINT1 (" End of pattern: fast loop.\n");
4877 return 1;
4878 }
4879 break;
4880
4881 case endline:
4882 case exactn:
4883 {
4884 register re_wchar_t c
4885 = (re_opcode_t) *p2 == endline ? '\n'
4886 : RE_STRING_CHAR (p2 + 2, multibyte);
4887
4888 if ((re_opcode_t) *p1 == exactn)
4889 {
4890 if (c != RE_STRING_CHAR (p1 + 2, multibyte))
4891 {
4892 DEBUG_PRINT3 (" '%c' != '%c' => fast loop.\n", c, p1[2]);
4893 return 1;
4894 }
4895 }
4896
4897 else if ((re_opcode_t) *p1 == charset
4898 || (re_opcode_t) *p1 == charset_not)
4899 {
4900 int not = (re_opcode_t) *p1 == charset_not;
4901
4902 /* Test if C is listed in charset (or charset_not)
4903 at `p1'. */
4904 if (! multibyte || IS_REAL_ASCII (c))
4905 {
4906 if (c < CHARSET_BITMAP_SIZE (p1) * BYTEWIDTH
4907 && p1[2 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH)))
4908 not = !not;
4909 }
4910 else if (CHARSET_RANGE_TABLE_EXISTS_P (p1))
4911 CHARSET_LOOKUP_RANGE_TABLE (not, c, p1);
4912
4913 /* `not' is equal to 1 if c would match, which means
4914 that we can't change to pop_failure_jump. */
4915 if (!not)
4916 {
4917 DEBUG_PRINT1 (" No match => fast loop.\n");
4918 return 1;
4919 }
4920 }
4921 else if ((re_opcode_t) *p1 == anychar
4922 && c == '\n')
4923 {
4924 DEBUG_PRINT1 (" . != \\n => fast loop.\n");
4925 return 1;
4926 }
4927 }
4928 break;
4929
4930 case charset:
4931 {
4932 if ((re_opcode_t) *p1 == exactn)
4933 /* Reuse the code above. */
4934 return mutually_exclusive_p (bufp, p2, p1);
4935
4936 /* It is hard to list up all the character in charset
4937 P2 if it includes multibyte character. Give up in
4938 such case. */
4939 else if (!multibyte || !CHARSET_RANGE_TABLE_EXISTS_P (p2))
4940 {
4941 /* Now, we are sure that P2 has no range table.
4942 So, for the size of bitmap in P2, `p2[1]' is
4943 enough. But P1 may have range table, so the
4944 size of bitmap table of P1 is extracted by
4945 using macro `CHARSET_BITMAP_SIZE'.
4946
4947 In a multibyte case, we know that all the character
4948 listed in P2 is ASCII. In a unibyte case, P1 has only a
4949 bitmap table. So, in both cases, it is enough to test
4950 only the bitmap table of P1. */
4951
4952 if ((re_opcode_t) *p1 == charset)
4953 {
4954 int idx;
4955 /* We win if the charset inside the loop
4956 has no overlap with the one after the loop. */
4957 for (idx = 0;
4958 (idx < (int) p2[1]
4959 && idx < CHARSET_BITMAP_SIZE (p1));
4960 idx++)
4961 if ((p2[2 + idx] & p1[2 + idx]) != 0)
4962 break;
4963
4964 if (idx == p2[1]
4965 || idx == CHARSET_BITMAP_SIZE (p1))
4966 {
4967 DEBUG_PRINT1 (" No match => fast loop.\n");
4968 return 1;
4969 }
4970 }
4971 else if ((re_opcode_t) *p1 == charset_not)
4972 {
4973 int idx;
4974 /* We win if the charset_not inside the loop lists
4975 every character listed in the charset after. */
4976 for (idx = 0; idx < (int) p2[1]; idx++)
4977 if (! (p2[2 + idx] == 0
4978 || (idx < CHARSET_BITMAP_SIZE (p1)
4979 && ((p2[2 + idx] & ~ p1[2 + idx]) == 0))))
4980 break;
4981
4982 if (idx == p2[1])
4983 {
4984 DEBUG_PRINT1 (" No match => fast loop.\n");
4985 return 1;
4986 }
4987 }
4988 }
4989 }
4990 break;
4991
4992 case charset_not:
4993 switch (SWITCH_ENUM_CAST (*p1))
4994 {
4995 case exactn:
4996 case charset:
4997 /* Reuse the code above. */
4998 return mutually_exclusive_p (bufp, p2, p1);
4999 case charset_not:
5000 /* When we have two charset_not, it's very unlikely that
5001 they don't overlap. The union of the two sets of excluded
5002 chars should cover all possible chars, which, as a matter of
5003 fact, is virtually impossible in multibyte buffers. */
5004 break;
5005 }
5006 break;
5007
5008 case wordend:
5009 return ((re_opcode_t) *p1 == syntaxspec && p1[1] == Sword);
5010 case symend:
5011 return ((re_opcode_t) *p1 == syntaxspec
5012 && (p1[1] == Ssymbol || p1[1] == Sword));
5013 case notsyntaxspec:
5014 return ((re_opcode_t) *p1 == syntaxspec && p1[1] == p2[1]);
5015
5016 case wordbeg:
5017 return ((re_opcode_t) *p1 == notsyntaxspec && p1[1] == Sword);
5018 case symbeg:
5019 return ((re_opcode_t) *p1 == notsyntaxspec
5020 && (p1[1] == Ssymbol || p1[1] == Sword));
5021 case syntaxspec:
5022 return ((re_opcode_t) *p1 == notsyntaxspec && p1[1] == p2[1]);
5023
5024 case wordbound:
5025 return (((re_opcode_t) *p1 == notsyntaxspec
5026 || (re_opcode_t) *p1 == syntaxspec)
5027 && p1[1] == Sword);
5028
5029 #ifdef emacs
5030 case categoryspec:
5031 return ((re_opcode_t) *p1 == notcategoryspec && p1[1] == p2[1]);
5032 case notcategoryspec:
5033 return ((re_opcode_t) *p1 == categoryspec && p1[1] == p2[1]);
5034 #endif /* emacs */
5035
5036 default:
5037 ;
5038 }
5039
5040 /* Safe default. */
5041 return 0;
5042 }
5043
5044 \f
5045 /* Matching routines. */
5046
5047 #ifndef emacs /* Emacs never uses this. */
5048 /* re_match is like re_match_2 except it takes only a single string. */
5049
5050 int
5051 re_match (bufp, string, size, pos, regs)
5052 struct re_pattern_buffer *bufp;
5053 const char *string;
5054 int size, pos;
5055 struct re_registers *regs;
5056 {
5057 int result = re_match_2_internal (bufp, NULL, 0, (re_char*) string, size,
5058 pos, regs, size);
5059 return result;
5060 }
5061 WEAK_ALIAS (__re_match, re_match)
5062 #endif /* not emacs */
5063
5064 #ifdef emacs
5065 /* In Emacs, this is the string or buffer in which we
5066 are matching. It is used for looking up syntax properties. */
5067 Lisp_Object re_match_object;
5068 #endif
5069
5070 /* re_match_2 matches the compiled pattern in BUFP against the
5071 the (virtual) concatenation of STRING1 and STRING2 (of length SIZE1
5072 and SIZE2, respectively). We start matching at POS, and stop
5073 matching at STOP.
5074
5075 If REGS is non-null and the `no_sub' field of BUFP is nonzero, we
5076 store offsets for the substring each group matched in REGS. See the
5077 documentation for exactly how many groups we fill.
5078
5079 We return -1 if no match, -2 if an internal error (such as the
5080 failure stack overflowing). Otherwise, we return the length of the
5081 matched substring. */
5082
5083 int
5084 re_match_2 (bufp, string1, size1, string2, size2, pos, regs, stop)
5085 struct re_pattern_buffer *bufp;
5086 const char *string1, *string2;
5087 int size1, size2;
5088 int pos;
5089 struct re_registers *regs;
5090 int stop;
5091 {
5092 int result;
5093
5094 #ifdef emacs
5095 int charpos;
5096 gl_state.object = re_match_object;
5097 charpos = SYNTAX_TABLE_BYTE_TO_CHAR (POS_AS_IN_BUFFER (pos));
5098 SETUP_SYNTAX_TABLE_FOR_OBJECT (re_match_object, charpos, 1);
5099 #endif
5100
5101 result = re_match_2_internal (bufp, (re_char*) string1, size1,
5102 (re_char*) string2, size2,
5103 pos, regs, stop);
5104 return result;
5105 }
5106 WEAK_ALIAS (__re_match_2, re_match_2)
5107
5108
5109 /* This is a separate function so that we can force an alloca cleanup
5110 afterwards. */
5111 static int
5112 re_match_2_internal (bufp, string1, size1, string2, size2, pos, regs, stop)
5113 struct re_pattern_buffer *bufp;
5114 re_char *string1, *string2;
5115 int size1, size2;
5116 int pos;
5117 struct re_registers *regs;
5118 int stop;
5119 {
5120 /* General temporaries. */
5121 int mcnt;
5122 size_t reg;
5123 boolean not;
5124
5125 /* Just past the end of the corresponding string. */
5126 re_char *end1, *end2;
5127
5128 /* Pointers into string1 and string2, just past the last characters in
5129 each to consider matching. */
5130 re_char *end_match_1, *end_match_2;
5131
5132 /* Where we are in the data, and the end of the current string. */
5133 re_char *d, *dend;
5134
5135 /* Used sometimes to remember where we were before starting matching
5136 an operator so that we can go back in case of failure. This "atomic"
5137 behavior of matching opcodes is indispensable to the correctness
5138 of the on_failure_keep_string_jump optimization. */
5139 re_char *dfail;
5140
5141 /* Where we are in the pattern, and the end of the pattern. */
5142 re_char *p = bufp->buffer;
5143 re_char *pend = p + bufp->used;
5144
5145 /* We use this to map every character in the string. */
5146 RE_TRANSLATE_TYPE translate = bufp->translate;
5147
5148 /* Nonzero if BUFP is setup from a multibyte regex. */
5149 const boolean multibyte = RE_MULTIBYTE_P (bufp);
5150
5151 /* Nonzero if STRING1/STRING2 are multibyte. */
5152 const boolean target_multibyte = RE_TARGET_MULTIBYTE_P (bufp);
5153
5154 /* Failure point stack. Each place that can handle a failure further
5155 down the line pushes a failure point on this stack. It consists of
5156 regstart, and regend for all registers corresponding to
5157 the subexpressions we're currently inside, plus the number of such
5158 registers, and, finally, two char *'s. The first char * is where
5159 to resume scanning the pattern; the second one is where to resume
5160 scanning the strings. */
5161 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
5162 fail_stack_type fail_stack;
5163 #endif
5164 #ifdef DEBUG
5165 unsigned nfailure_points_pushed = 0, nfailure_points_popped = 0;
5166 #endif
5167
5168 #if defined REL_ALLOC && defined REGEX_MALLOC
5169 /* This holds the pointer to the failure stack, when
5170 it is allocated relocatably. */
5171 fail_stack_elt_t *failure_stack_ptr;
5172 #endif
5173
5174 /* We fill all the registers internally, independent of what we
5175 return, for use in backreferences. The number here includes
5176 an element for register zero. */
5177 size_t num_regs = bufp->re_nsub + 1;
5178
5179 /* Information on the contents of registers. These are pointers into
5180 the input strings; they record just what was matched (on this
5181 attempt) by a subexpression part of the pattern, that is, the
5182 regnum-th regstart pointer points to where in the pattern we began
5183 matching and the regnum-th regend points to right after where we
5184 stopped matching the regnum-th subexpression. (The zeroth register
5185 keeps track of what the whole pattern matches.) */
5186 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
5187 re_char **regstart, **regend;
5188 #endif
5189
5190 /* The following record the register info as found in the above
5191 variables when we find a match better than any we've seen before.
5192 This happens as we backtrack through the failure points, which in
5193 turn happens only if we have not yet matched the entire string. */
5194 unsigned best_regs_set = false;
5195 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
5196 re_char **best_regstart, **best_regend;
5197 #endif
5198
5199 /* Logically, this is `best_regend[0]'. But we don't want to have to
5200 allocate space for that if we're not allocating space for anything
5201 else (see below). Also, we never need info about register 0 for
5202 any of the other register vectors, and it seems rather a kludge to
5203 treat `best_regend' differently than the rest. So we keep track of
5204 the end of the best match so far in a separate variable. We
5205 initialize this to NULL so that when we backtrack the first time
5206 and need to test it, it's not garbage. */
5207 re_char *match_end = NULL;
5208
5209 #ifdef DEBUG
5210 /* Counts the total number of registers pushed. */
5211 unsigned num_regs_pushed = 0;
5212 #endif
5213
5214 DEBUG_PRINT1 ("\n\nEntering re_match_2.\n");
5215
5216 INIT_FAIL_STACK ();
5217
5218 #ifdef MATCH_MAY_ALLOCATE
5219 /* Do not bother to initialize all the register variables if there are
5220 no groups in the pattern, as it takes a fair amount of time. If
5221 there are groups, we include space for register 0 (the whole
5222 pattern), even though we never use it, since it simplifies the
5223 array indexing. We should fix this. */
5224 if (bufp->re_nsub)
5225 {
5226 regstart = REGEX_TALLOC (num_regs, re_char *);
5227 regend = REGEX_TALLOC (num_regs, re_char *);
5228 best_regstart = REGEX_TALLOC (num_regs, re_char *);
5229 best_regend = REGEX_TALLOC (num_regs, re_char *);
5230
5231 if (!(regstart && regend && best_regstart && best_regend))
5232 {
5233 FREE_VARIABLES ();
5234 return -2;
5235 }
5236 }
5237 else
5238 {
5239 /* We must initialize all our variables to NULL, so that
5240 `FREE_VARIABLES' doesn't try to free them. */
5241 regstart = regend = best_regstart = best_regend = NULL;
5242 }
5243 #endif /* MATCH_MAY_ALLOCATE */
5244
5245 /* The starting position is bogus. */
5246 if (pos < 0 || pos > size1 + size2)
5247 {
5248 FREE_VARIABLES ();
5249 return -1;
5250 }
5251
5252 /* Initialize subexpression text positions to -1 to mark ones that no
5253 start_memory/stop_memory has been seen for. Also initialize the
5254 register information struct. */
5255 for (reg = 1; reg < num_regs; reg++)
5256 regstart[reg] = regend[reg] = NULL;
5257
5258 /* We move `string1' into `string2' if the latter's empty -- but not if
5259 `string1' is null. */
5260 if (size2 == 0 && string1 != NULL)
5261 {
5262 string2 = string1;
5263 size2 = size1;
5264 string1 = 0;
5265 size1 = 0;
5266 }
5267 end1 = string1 + size1;
5268 end2 = string2 + size2;
5269
5270 /* `p' scans through the pattern as `d' scans through the data.
5271 `dend' is the end of the input string that `d' points within. `d'
5272 is advanced into the following input string whenever necessary, but
5273 this happens before fetching; therefore, at the beginning of the
5274 loop, `d' can be pointing at the end of a string, but it cannot
5275 equal `string2'. */
5276 if (pos >= size1)
5277 {
5278 /* Only match within string2. */
5279 d = string2 + pos - size1;
5280 dend = end_match_2 = string2 + stop - size1;
5281 end_match_1 = end1; /* Just to give it a value. */
5282 }
5283 else
5284 {
5285 if (stop < size1)
5286 {
5287 /* Only match within string1. */
5288 end_match_1 = string1 + stop;
5289 /* BEWARE!
5290 When we reach end_match_1, PREFETCH normally switches to string2.
5291 But in the present case, this means that just doing a PREFETCH
5292 makes us jump from `stop' to `gap' within the string.
5293 What we really want here is for the search to stop as
5294 soon as we hit end_match_1. That's why we set end_match_2
5295 to end_match_1 (since PREFETCH fails as soon as we hit
5296 end_match_2). */
5297 end_match_2 = end_match_1;
5298 }
5299 else
5300 { /* It's important to use this code when stop == size so that
5301 moving `d' from end1 to string2 will not prevent the d == dend
5302 check from catching the end of string. */
5303 end_match_1 = end1;
5304 end_match_2 = string2 + stop - size1;
5305 }
5306 d = string1 + pos;
5307 dend = end_match_1;
5308 }
5309
5310 DEBUG_PRINT1 ("The compiled pattern is: ");
5311 DEBUG_PRINT_COMPILED_PATTERN (bufp, p, pend);
5312 DEBUG_PRINT1 ("The string to match is: `");
5313 DEBUG_PRINT_DOUBLE_STRING (d, string1, size1, string2, size2);
5314 DEBUG_PRINT1 ("'\n");
5315
5316 /* This loops over pattern commands. It exits by returning from the
5317 function if the match is complete, or it drops through if the match
5318 fails at this starting point in the input data. */
5319 for (;;)
5320 {
5321 DEBUG_PRINT2 ("\n%p: ", p);
5322
5323 if (p == pend)
5324 { /* End of pattern means we might have succeeded. */
5325 DEBUG_PRINT1 ("end of pattern ... ");
5326
5327 /* If we haven't matched the entire string, and we want the
5328 longest match, try backtracking. */
5329 if (d != end_match_2)
5330 {
5331 /* 1 if this match ends in the same string (string1 or string2)
5332 as the best previous match. */
5333 boolean same_str_p = (FIRST_STRING_P (match_end)
5334 == FIRST_STRING_P (d));
5335 /* 1 if this match is the best seen so far. */
5336 boolean best_match_p;
5337
5338 /* AIX compiler got confused when this was combined
5339 with the previous declaration. */
5340 if (same_str_p)
5341 best_match_p = d > match_end;
5342 else
5343 best_match_p = !FIRST_STRING_P (d);
5344
5345 DEBUG_PRINT1 ("backtracking.\n");
5346
5347 if (!FAIL_STACK_EMPTY ())
5348 { /* More failure points to try. */
5349
5350 /* If exceeds best match so far, save it. */
5351 if (!best_regs_set || best_match_p)
5352 {
5353 best_regs_set = true;
5354 match_end = d;
5355
5356 DEBUG_PRINT1 ("\nSAVING match as best so far.\n");
5357
5358 for (reg = 1; reg < num_regs; reg++)
5359 {
5360 best_regstart[reg] = regstart[reg];
5361 best_regend[reg] = regend[reg];
5362 }
5363 }
5364 goto fail;
5365 }
5366
5367 /* If no failure points, don't restore garbage. And if
5368 last match is real best match, don't restore second
5369 best one. */
5370 else if (best_regs_set && !best_match_p)
5371 {
5372 restore_best_regs:
5373 /* Restore best match. It may happen that `dend ==
5374 end_match_1' while the restored d is in string2.
5375 For example, the pattern `x.*y.*z' against the
5376 strings `x-' and `y-z-', if the two strings are
5377 not consecutive in memory. */
5378 DEBUG_PRINT1 ("Restoring best registers.\n");
5379
5380 d = match_end;
5381 dend = ((d >= string1 && d <= end1)
5382 ? end_match_1 : end_match_2);
5383
5384 for (reg = 1; reg < num_regs; reg++)
5385 {
5386 regstart[reg] = best_regstart[reg];
5387 regend[reg] = best_regend[reg];
5388 }
5389 }
5390 } /* d != end_match_2 */
5391
5392 succeed_label:
5393 DEBUG_PRINT1 ("Accepting match.\n");
5394
5395 /* If caller wants register contents data back, do it. */
5396 if (regs && !bufp->no_sub)
5397 {
5398 /* Have the register data arrays been allocated? */
5399 if (bufp->regs_allocated == REGS_UNALLOCATED)
5400 { /* No. So allocate them with malloc. We need one
5401 extra element beyond `num_regs' for the `-1' marker
5402 GNU code uses. */
5403 regs->num_regs = MAX (RE_NREGS, num_regs + 1);
5404 regs->start = TALLOC (regs->num_regs, regoff_t);
5405 regs->end = TALLOC (regs->num_regs, regoff_t);
5406 if (regs->start == NULL || regs->end == NULL)
5407 {
5408 FREE_VARIABLES ();
5409 return -2;
5410 }
5411 bufp->regs_allocated = REGS_REALLOCATE;
5412 }
5413 else if (bufp->regs_allocated == REGS_REALLOCATE)
5414 { /* Yes. If we need more elements than were already
5415 allocated, reallocate them. If we need fewer, just
5416 leave it alone. */
5417 if (regs->num_regs < num_regs + 1)
5418 {
5419 regs->num_regs = num_regs + 1;
5420 RETALLOC (regs->start, regs->num_regs, regoff_t);
5421 RETALLOC (regs->end, regs->num_regs, regoff_t);
5422 if (regs->start == NULL || regs->end == NULL)
5423 {
5424 FREE_VARIABLES ();
5425 return -2;
5426 }
5427 }
5428 }
5429 else
5430 {
5431 /* These braces fend off a "empty body in an else-statement"
5432 warning under GCC when assert expands to nothing. */
5433 assert (bufp->regs_allocated == REGS_FIXED);
5434 }
5435
5436 /* Convert the pointer data in `regstart' and `regend' to
5437 indices. Register zero has to be set differently,
5438 since we haven't kept track of any info for it. */
5439 if (regs->num_regs > 0)
5440 {
5441 regs->start[0] = pos;
5442 regs->end[0] = POINTER_TO_OFFSET (d);
5443 }
5444
5445 /* Go through the first `min (num_regs, regs->num_regs)'
5446 registers, since that is all we initialized. */
5447 for (reg = 1; reg < MIN (num_regs, regs->num_regs); reg++)
5448 {
5449 if (REG_UNSET (regstart[reg]) || REG_UNSET (regend[reg]))
5450 regs->start[reg] = regs->end[reg] = -1;
5451 else
5452 {
5453 regs->start[reg]
5454 = (regoff_t) POINTER_TO_OFFSET (regstart[reg]);
5455 regs->end[reg]
5456 = (regoff_t) POINTER_TO_OFFSET (regend[reg]);
5457 }
5458 }
5459
5460 /* If the regs structure we return has more elements than
5461 were in the pattern, set the extra elements to -1. If
5462 we (re)allocated the registers, this is the case,
5463 because we always allocate enough to have at least one
5464 -1 at the end. */
5465 for (reg = num_regs; reg < regs->num_regs; reg++)
5466 regs->start[reg] = regs->end[reg] = -1;
5467 } /* regs && !bufp->no_sub */
5468
5469 DEBUG_PRINT4 ("%u failure points pushed, %u popped (%u remain).\n",
5470 nfailure_points_pushed, nfailure_points_popped,
5471 nfailure_points_pushed - nfailure_points_popped);
5472 DEBUG_PRINT2 ("%u registers pushed.\n", num_regs_pushed);
5473
5474 mcnt = POINTER_TO_OFFSET (d) - pos;
5475
5476 DEBUG_PRINT2 ("Returning %d from re_match_2.\n", mcnt);
5477
5478 FREE_VARIABLES ();
5479 return mcnt;
5480 }
5481
5482 /* Otherwise match next pattern command. */
5483 switch (SWITCH_ENUM_CAST ((re_opcode_t) *p++))
5484 {
5485 /* Ignore these. Used to ignore the n of succeed_n's which
5486 currently have n == 0. */
5487 case no_op:
5488 DEBUG_PRINT1 ("EXECUTING no_op.\n");
5489 break;
5490
5491 case succeed:
5492 DEBUG_PRINT1 ("EXECUTING succeed.\n");
5493 goto succeed_label;
5494
5495 /* Match the next n pattern characters exactly. The following
5496 byte in the pattern defines n, and the n bytes after that
5497 are the characters to match. */
5498 case exactn:
5499 mcnt = *p++;
5500 DEBUG_PRINT2 ("EXECUTING exactn %d.\n", mcnt);
5501
5502 /* Remember the start point to rollback upon failure. */
5503 dfail = d;
5504
5505 #ifndef emacs
5506 /* This is written out as an if-else so we don't waste time
5507 testing `translate' inside the loop. */
5508 if (RE_TRANSLATE_P (translate))
5509 do
5510 {
5511 PREFETCH ();
5512 if (RE_TRANSLATE (translate, *d) != *p++)
5513 {
5514 d = dfail;
5515 goto fail;
5516 }
5517 d++;
5518 }
5519 while (--mcnt);
5520 else
5521 do
5522 {
5523 PREFETCH ();
5524 if (*d++ != *p++)
5525 {
5526 d = dfail;
5527 goto fail;
5528 }
5529 }
5530 while (--mcnt);
5531 #else /* emacs */
5532 /* The cost of testing `translate' is comparatively small. */
5533 if (target_multibyte)
5534 do
5535 {
5536 int pat_charlen, buf_charlen;
5537 int pat_ch, buf_ch;
5538
5539 PREFETCH ();
5540 if (multibyte)
5541 pat_ch = STRING_CHAR_AND_LENGTH (p, pat_charlen);
5542 else
5543 {
5544 pat_ch = RE_CHAR_TO_MULTIBYTE (*p);
5545 pat_charlen = 1;
5546 }
5547 buf_ch = STRING_CHAR_AND_LENGTH (d, buf_charlen);
5548
5549 if (TRANSLATE (buf_ch) != pat_ch)
5550 {
5551 d = dfail;
5552 goto fail;
5553 }
5554
5555 p += pat_charlen;
5556 d += buf_charlen;
5557 mcnt -= pat_charlen;
5558 }
5559 while (mcnt > 0);
5560 else
5561 do
5562 {
5563 int pat_charlen, buf_charlen;
5564 int pat_ch, buf_ch;
5565
5566 PREFETCH ();
5567 if (multibyte)
5568 {
5569 pat_ch = STRING_CHAR_AND_LENGTH (p, pat_charlen);
5570 pat_ch = RE_CHAR_TO_UNIBYTE (pat_ch);
5571 }
5572 else
5573 {
5574 pat_ch = *p;
5575 pat_charlen = 1;
5576 }
5577 buf_ch = RE_CHAR_TO_MULTIBYTE (*d);
5578 if (! CHAR_BYTE8_P (buf_ch))
5579 {
5580 buf_ch = TRANSLATE (buf_ch);
5581 buf_ch = RE_CHAR_TO_UNIBYTE (buf_ch);
5582 if (buf_ch < 0)
5583 buf_ch = *d;
5584 }
5585 else
5586 buf_ch = *d;
5587 if (buf_ch != pat_ch)
5588 {
5589 d = dfail;
5590 goto fail;
5591 }
5592 p += pat_charlen;
5593 d++;
5594 }
5595 while (--mcnt);
5596 #endif
5597 break;
5598
5599
5600 /* Match any character except possibly a newline or a null. */
5601 case anychar:
5602 {
5603 int buf_charlen;
5604 re_wchar_t buf_ch;
5605
5606 DEBUG_PRINT1 ("EXECUTING anychar.\n");
5607
5608 PREFETCH ();
5609 buf_ch = RE_STRING_CHAR_AND_LENGTH (d, buf_charlen,
5610 target_multibyte);
5611 buf_ch = TRANSLATE (buf_ch);
5612
5613 if ((!(bufp->syntax & RE_DOT_NEWLINE)
5614 && buf_ch == '\n')
5615 || ((bufp->syntax & RE_DOT_NOT_NULL)
5616 && buf_ch == '\000'))
5617 goto fail;
5618
5619 DEBUG_PRINT2 (" Matched `%d'.\n", *d);
5620 d += buf_charlen;
5621 }
5622 break;
5623
5624
5625 case charset:
5626 case charset_not:
5627 {
5628 register unsigned int c;
5629 boolean not = (re_opcode_t) *(p - 1) == charset_not;
5630 int len;
5631
5632 /* Start of actual range_table, or end of bitmap if there is no
5633 range table. */
5634 re_char *range_table;
5635
5636 /* Nonzero if there is a range table. */
5637 int range_table_exists;
5638
5639 /* Number of ranges of range table. This is not included
5640 in the initial byte-length of the command. */
5641 int count = 0;
5642
5643 /* Whether matching against a unibyte character. */
5644 boolean unibyte_char = false;
5645
5646 DEBUG_PRINT2 ("EXECUTING charset%s.\n", not ? "_not" : "");
5647
5648 range_table_exists = CHARSET_RANGE_TABLE_EXISTS_P (&p[-1]);
5649
5650 if (range_table_exists)
5651 {
5652 range_table = CHARSET_RANGE_TABLE (&p[-1]); /* Past the bitmap. */
5653 EXTRACT_NUMBER_AND_INCR (count, range_table);
5654 }
5655
5656 PREFETCH ();
5657 c = RE_STRING_CHAR_AND_LENGTH (d, len, target_multibyte);
5658 if (target_multibyte)
5659 {
5660 int c1;
5661
5662 c = TRANSLATE (c);
5663 c1 = RE_CHAR_TO_UNIBYTE (c);
5664 if (c1 >= 0)
5665 {
5666 unibyte_char = true;
5667 c = c1;
5668 }
5669 }
5670 else
5671 {
5672 int c1 = RE_CHAR_TO_MULTIBYTE (c);
5673
5674 if (! CHAR_BYTE8_P (c1))
5675 {
5676 c1 = TRANSLATE (c1);
5677 c1 = RE_CHAR_TO_UNIBYTE (c1);
5678 if (c1 >= 0)
5679 {
5680 unibyte_char = true;
5681 c = c1;
5682 }
5683 }
5684 else
5685 unibyte_char = true;
5686 }
5687
5688 if (unibyte_char && c < (1 << BYTEWIDTH))
5689 { /* Lookup bitmap. */
5690 /* Cast to `unsigned' instead of `unsigned char' in
5691 case the bit list is a full 32 bytes long. */
5692 if (c < (unsigned) (CHARSET_BITMAP_SIZE (&p[-1]) * BYTEWIDTH)
5693 && p[1 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH)))
5694 not = !not;
5695 }
5696 #ifdef emacs
5697 else if (range_table_exists)
5698 {
5699 int class_bits = CHARSET_RANGE_TABLE_BITS (&p[-1]);
5700
5701 if ( (class_bits & BIT_LOWER && ISLOWER (c))
5702 | (class_bits & BIT_MULTIBYTE)
5703 | (class_bits & BIT_PUNCT && ISPUNCT (c))
5704 | (class_bits & BIT_SPACE && ISSPACE (c))
5705 | (class_bits & BIT_UPPER && ISUPPER (c))
5706 | (class_bits & BIT_WORD && ISWORD (c)))
5707 not = !not;
5708 else
5709 CHARSET_LOOKUP_RANGE_TABLE_RAW (not, c, range_table, count);
5710 }
5711 #endif /* emacs */
5712
5713 if (range_table_exists)
5714 p = CHARSET_RANGE_TABLE_END (range_table, count);
5715 else
5716 p += CHARSET_BITMAP_SIZE (&p[-1]) + 1;
5717
5718 if (!not) goto fail;
5719
5720 d += len;
5721 break;
5722 }
5723
5724
5725 /* The beginning of a group is represented by start_memory.
5726 The argument is the register number. The text
5727 matched within the group is recorded (in the internal
5728 registers data structure) under the register number. */
5729 case start_memory:
5730 DEBUG_PRINT2 ("EXECUTING start_memory %d:\n", *p);
5731
5732 /* In case we need to undo this operation (via backtracking). */
5733 PUSH_FAILURE_REG ((unsigned int)*p);
5734
5735 regstart[*p] = d;
5736 regend[*p] = NULL; /* probably unnecessary. -sm */
5737 DEBUG_PRINT2 (" regstart: %d\n", POINTER_TO_OFFSET (regstart[*p]));
5738
5739 /* Move past the register number and inner group count. */
5740 p += 1;
5741 break;
5742
5743
5744 /* The stop_memory opcode represents the end of a group. Its
5745 argument is the same as start_memory's: the register number. */
5746 case stop_memory:
5747 DEBUG_PRINT2 ("EXECUTING stop_memory %d:\n", *p);
5748
5749 assert (!REG_UNSET (regstart[*p]));
5750 /* Strictly speaking, there should be code such as:
5751
5752 assert (REG_UNSET (regend[*p]));
5753 PUSH_FAILURE_REGSTOP ((unsigned int)*p);
5754
5755 But the only info to be pushed is regend[*p] and it is known to
5756 be UNSET, so there really isn't anything to push.
5757 Not pushing anything, on the other hand deprives us from the
5758 guarantee that regend[*p] is UNSET since undoing this operation
5759 will not reset its value properly. This is not important since
5760 the value will only be read on the next start_memory or at
5761 the very end and both events can only happen if this stop_memory
5762 is *not* undone. */
5763
5764 regend[*p] = d;
5765 DEBUG_PRINT2 (" regend: %d\n", POINTER_TO_OFFSET (regend[*p]));
5766
5767 /* Move past the register number and the inner group count. */
5768 p += 1;
5769 break;
5770
5771
5772 /* \<digit> has been turned into a `duplicate' command which is
5773 followed by the numeric value of <digit> as the register number. */
5774 case duplicate:
5775 {
5776 register re_char *d2, *dend2;
5777 int regno = *p++; /* Get which register to match against. */
5778 DEBUG_PRINT2 ("EXECUTING duplicate %d.\n", regno);
5779
5780 /* Can't back reference a group which we've never matched. */
5781 if (REG_UNSET (regstart[regno]) || REG_UNSET (regend[regno]))
5782 goto fail;
5783
5784 /* Where in input to try to start matching. */
5785 d2 = regstart[regno];
5786
5787 /* Remember the start point to rollback upon failure. */
5788 dfail = d;
5789
5790 /* Where to stop matching; if both the place to start and
5791 the place to stop matching are in the same string, then
5792 set to the place to stop, otherwise, for now have to use
5793 the end of the first string. */
5794
5795 dend2 = ((FIRST_STRING_P (regstart[regno])
5796 == FIRST_STRING_P (regend[regno]))
5797 ? regend[regno] : end_match_1);
5798 for (;;)
5799 {
5800 /* If necessary, advance to next segment in register
5801 contents. */
5802 while (d2 == dend2)
5803 {
5804 if (dend2 == end_match_2) break;
5805 if (dend2 == regend[regno]) break;
5806
5807 /* End of string1 => advance to string2. */
5808 d2 = string2;
5809 dend2 = regend[regno];
5810 }
5811 /* At end of register contents => success */
5812 if (d2 == dend2) break;
5813
5814 /* If necessary, advance to next segment in data. */
5815 PREFETCH ();
5816
5817 /* How many characters left in this segment to match. */
5818 mcnt = dend - d;
5819
5820 /* Want how many consecutive characters we can match in
5821 one shot, so, if necessary, adjust the count. */
5822 if (mcnt > dend2 - d2)
5823 mcnt = dend2 - d2;
5824
5825 /* Compare that many; failure if mismatch, else move
5826 past them. */
5827 if (RE_TRANSLATE_P (translate)
5828 ? bcmp_translate (d, d2, mcnt, translate, target_multibyte)
5829 : memcmp (d, d2, mcnt))
5830 {
5831 d = dfail;
5832 goto fail;
5833 }
5834 d += mcnt, d2 += mcnt;
5835 }
5836 }
5837 break;
5838
5839
5840 /* begline matches the empty string at the beginning of the string
5841 (unless `not_bol' is set in `bufp'), and after newlines. */
5842 case begline:
5843 DEBUG_PRINT1 ("EXECUTING begline.\n");
5844
5845 if (AT_STRINGS_BEG (d))
5846 {
5847 if (!bufp->not_bol) break;
5848 }
5849 else
5850 {
5851 unsigned c;
5852 GET_CHAR_BEFORE_2 (c, d, string1, end1, string2, end2);
5853 if (c == '\n')
5854 break;
5855 }
5856 /* In all other cases, we fail. */
5857 goto fail;
5858
5859
5860 /* endline is the dual of begline. */
5861 case endline:
5862 DEBUG_PRINT1 ("EXECUTING endline.\n");
5863
5864 if (AT_STRINGS_END (d))
5865 {
5866 if (!bufp->not_eol) break;
5867 }
5868 else
5869 {
5870 PREFETCH_NOLIMIT ();
5871 if (*d == '\n')
5872 break;
5873 }
5874 goto fail;
5875
5876
5877 /* Match at the very beginning of the data. */
5878 case begbuf:
5879 DEBUG_PRINT1 ("EXECUTING begbuf.\n");
5880 if (AT_STRINGS_BEG (d))
5881 break;
5882 goto fail;
5883
5884
5885 /* Match at the very end of the data. */
5886 case endbuf:
5887 DEBUG_PRINT1 ("EXECUTING endbuf.\n");
5888 if (AT_STRINGS_END (d))
5889 break;
5890 goto fail;
5891
5892
5893 /* on_failure_keep_string_jump is used to optimize `.*\n'. It
5894 pushes NULL as the value for the string on the stack. Then
5895 `POP_FAILURE_POINT' will keep the current value for the
5896 string, instead of restoring it. To see why, consider
5897 matching `foo\nbar' against `.*\n'. The .* matches the foo;
5898 then the . fails against the \n. But the next thing we want
5899 to do is match the \n against the \n; if we restored the
5900 string value, we would be back at the foo.
5901
5902 Because this is used only in specific cases, we don't need to
5903 check all the things that `on_failure_jump' does, to make
5904 sure the right things get saved on the stack. Hence we don't
5905 share its code. The only reason to push anything on the
5906 stack at all is that otherwise we would have to change
5907 `anychar's code to do something besides goto fail in this
5908 case; that seems worse than this. */
5909 case on_failure_keep_string_jump:
5910 EXTRACT_NUMBER_AND_INCR (mcnt, p);
5911 DEBUG_PRINT3 ("EXECUTING on_failure_keep_string_jump %d (to %p):\n",
5912 mcnt, p + mcnt);
5913
5914 PUSH_FAILURE_POINT (p - 3, NULL);
5915 break;
5916
5917 /* A nasty loop is introduced by the non-greedy *? and +?.
5918 With such loops, the stack only ever contains one failure point
5919 at a time, so that a plain on_failure_jump_loop kind of
5920 cycle detection cannot work. Worse yet, such a detection
5921 can not only fail to detect a cycle, but it can also wrongly
5922 detect a cycle (between different instantiations of the same
5923 loop).
5924 So the method used for those nasty loops is a little different:
5925 We use a special cycle-detection-stack-frame which is pushed
5926 when the on_failure_jump_nastyloop failure-point is *popped*.
5927 This special frame thus marks the beginning of one iteration
5928 through the loop and we can hence easily check right here
5929 whether something matched between the beginning and the end of
5930 the loop. */
5931 case on_failure_jump_nastyloop:
5932 EXTRACT_NUMBER_AND_INCR (mcnt, p);
5933 DEBUG_PRINT3 ("EXECUTING on_failure_jump_nastyloop %d (to %p):\n",
5934 mcnt, p + mcnt);
5935
5936 assert ((re_opcode_t)p[-4] == no_op);
5937 {
5938 int cycle = 0;
5939 CHECK_INFINITE_LOOP (p - 4, d);
5940 if (!cycle)
5941 /* If there's a cycle, just continue without pushing
5942 this failure point. The failure point is the "try again"
5943 option, which shouldn't be tried.
5944 We want (x?)*?y\1z to match both xxyz and xxyxz. */
5945 PUSH_FAILURE_POINT (p - 3, d);
5946 }
5947 break;
5948
5949 /* Simple loop detecting on_failure_jump: just check on the
5950 failure stack if the same spot was already hit earlier. */
5951 case on_failure_jump_loop:
5952 on_failure:
5953 EXTRACT_NUMBER_AND_INCR (mcnt, p);
5954 DEBUG_PRINT3 ("EXECUTING on_failure_jump_loop %d (to %p):\n",
5955 mcnt, p + mcnt);
5956 {
5957 int cycle = 0;
5958 CHECK_INFINITE_LOOP (p - 3, d);
5959 if (cycle)
5960 /* If there's a cycle, get out of the loop, as if the matching
5961 had failed. We used to just `goto fail' here, but that was
5962 aborting the search a bit too early: we want to keep the
5963 empty-loop-match and keep matching after the loop.
5964 We want (x?)*y\1z to match both xxyz and xxyxz. */
5965 p += mcnt;
5966 else
5967 PUSH_FAILURE_POINT (p - 3, d);
5968 }
5969 break;
5970
5971
5972 /* Uses of on_failure_jump:
5973
5974 Each alternative starts with an on_failure_jump that points
5975 to the beginning of the next alternative. Each alternative
5976 except the last ends with a jump that in effect jumps past
5977 the rest of the alternatives. (They really jump to the
5978 ending jump of the following alternative, because tensioning
5979 these jumps is a hassle.)
5980
5981 Repeats start with an on_failure_jump that points past both
5982 the repetition text and either the following jump or
5983 pop_failure_jump back to this on_failure_jump. */
5984 case on_failure_jump:
5985 EXTRACT_NUMBER_AND_INCR (mcnt, p);
5986 DEBUG_PRINT3 ("EXECUTING on_failure_jump %d (to %p):\n",
5987 mcnt, p + mcnt);
5988
5989 PUSH_FAILURE_POINT (p -3, d);
5990 break;
5991
5992 /* This operation is used for greedy *.
5993 Compare the beginning of the repeat with what in the
5994 pattern follows its end. If we can establish that there
5995 is nothing that they would both match, i.e., that we
5996 would have to backtrack because of (as in, e.g., `a*a')
5997 then we can use a non-backtracking loop based on
5998 on_failure_keep_string_jump instead of on_failure_jump. */
5999 case on_failure_jump_smart:
6000 EXTRACT_NUMBER_AND_INCR (mcnt, p);
6001 DEBUG_PRINT3 ("EXECUTING on_failure_jump_smart %d (to %p).\n",
6002 mcnt, p + mcnt);
6003 {
6004 re_char *p1 = p; /* Next operation. */
6005 /* Here, we discard `const', making re_match non-reentrant. */
6006 unsigned char *p2 = (unsigned char*) p + mcnt; /* Jump dest. */
6007 unsigned char *p3 = (unsigned char*) p - 3; /* opcode location. */
6008
6009 p -= 3; /* Reset so that we will re-execute the
6010 instruction once it's been changed. */
6011
6012 EXTRACT_NUMBER (mcnt, p2 - 2);
6013
6014 /* Ensure this is a indeed the trivial kind of loop
6015 we are expecting. */
6016 assert (skip_one_char (p1) == p2 - 3);
6017 assert ((re_opcode_t) p2[-3] == jump && p2 + mcnt == p);
6018 DEBUG_STATEMENT (debug += 2);
6019 if (mutually_exclusive_p (bufp, p1, p2))
6020 {
6021 /* Use a fast `on_failure_keep_string_jump' loop. */
6022 DEBUG_PRINT1 (" smart exclusive => fast loop.\n");
6023 *p3 = (unsigned char) on_failure_keep_string_jump;
6024 STORE_NUMBER (p2 - 2, mcnt + 3);
6025 }
6026 else
6027 {
6028 /* Default to a safe `on_failure_jump' loop. */
6029 DEBUG_PRINT1 (" smart default => slow loop.\n");
6030 *p3 = (unsigned char) on_failure_jump;
6031 }
6032 DEBUG_STATEMENT (debug -= 2);
6033 }
6034 break;
6035
6036 /* Unconditionally jump (without popping any failure points). */
6037 case jump:
6038 unconditional_jump:
6039 IMMEDIATE_QUIT_CHECK;
6040 EXTRACT_NUMBER_AND_INCR (mcnt, p); /* Get the amount to jump. */
6041 DEBUG_PRINT2 ("EXECUTING jump %d ", mcnt);
6042 p += mcnt; /* Do the jump. */
6043 DEBUG_PRINT2 ("(to %p).\n", p);
6044 break;
6045
6046
6047 /* Have to succeed matching what follows at least n times.
6048 After that, handle like `on_failure_jump'. */
6049 case succeed_n:
6050 /* Signedness doesn't matter since we only compare MCNT to 0. */
6051 EXTRACT_NUMBER (mcnt, p + 2);
6052 DEBUG_PRINT2 ("EXECUTING succeed_n %d.\n", mcnt);
6053
6054 /* Originally, mcnt is how many times we HAVE to succeed. */
6055 if (mcnt != 0)
6056 {
6057 /* Here, we discard `const', making re_match non-reentrant. */
6058 unsigned char *p2 = (unsigned char*) p + 2; /* counter loc. */
6059 mcnt--;
6060 p += 4;
6061 PUSH_NUMBER (p2, mcnt);
6062 }
6063 else
6064 /* The two bytes encoding mcnt == 0 are two no_op opcodes. */
6065 goto on_failure;
6066 break;
6067
6068 case jump_n:
6069 /* Signedness doesn't matter since we only compare MCNT to 0. */
6070 EXTRACT_NUMBER (mcnt, p + 2);
6071 DEBUG_PRINT2 ("EXECUTING jump_n %d.\n", mcnt);
6072
6073 /* Originally, this is how many times we CAN jump. */
6074 if (mcnt != 0)
6075 {
6076 /* Here, we discard `const', making re_match non-reentrant. */
6077 unsigned char *p2 = (unsigned char*) p + 2; /* counter loc. */
6078 mcnt--;
6079 PUSH_NUMBER (p2, mcnt);
6080 goto unconditional_jump;
6081 }
6082 /* If don't have to jump any more, skip over the rest of command. */
6083 else
6084 p += 4;
6085 break;
6086
6087 case set_number_at:
6088 {
6089 unsigned char *p2; /* Location of the counter. */
6090 DEBUG_PRINT1 ("EXECUTING set_number_at.\n");
6091
6092 EXTRACT_NUMBER_AND_INCR (mcnt, p);
6093 /* Here, we discard `const', making re_match non-reentrant. */
6094 p2 = (unsigned char*) p + mcnt;
6095 /* Signedness doesn't matter since we only copy MCNT's bits . */
6096 EXTRACT_NUMBER_AND_INCR (mcnt, p);
6097 DEBUG_PRINT3 (" Setting %p to %d.\n", p2, mcnt);
6098 PUSH_NUMBER (p2, mcnt);
6099 break;
6100 }
6101
6102 case wordbound:
6103 case notwordbound:
6104 not = (re_opcode_t) *(p - 1) == notwordbound;
6105 DEBUG_PRINT2 ("EXECUTING %swordbound.\n", not?"not":"");
6106
6107 /* We SUCCEED (or FAIL) in one of the following cases: */
6108
6109 /* Case 1: D is at the beginning or the end of string. */
6110 if (AT_STRINGS_BEG (d) || AT_STRINGS_END (d))
6111 not = !not;
6112 else
6113 {
6114 /* C1 is the character before D, S1 is the syntax of C1, C2
6115 is the character at D, and S2 is the syntax of C2. */
6116 re_wchar_t c1, c2;
6117 int s1, s2;
6118 int dummy;
6119 #ifdef emacs
6120 int offset = PTR_TO_OFFSET (d - 1);
6121 int charpos = SYNTAX_TABLE_BYTE_TO_CHAR (offset);
6122 UPDATE_SYNTAX_TABLE (charpos);
6123 #endif
6124 GET_CHAR_BEFORE_2 (c1, d, string1, end1, string2, end2);
6125 s1 = SYNTAX (c1);
6126 #ifdef emacs
6127 UPDATE_SYNTAX_TABLE_FORWARD (charpos + 1);
6128 #endif
6129 PREFETCH_NOLIMIT ();
6130 GET_CHAR_AFTER (c2, d, dummy);
6131 s2 = SYNTAX (c2);
6132
6133 if (/* Case 2: Only one of S1 and S2 is Sword. */
6134 ((s1 == Sword) != (s2 == Sword))
6135 /* Case 3: Both of S1 and S2 are Sword, and macro
6136 WORD_BOUNDARY_P (C1, C2) returns nonzero. */
6137 || ((s1 == Sword) && WORD_BOUNDARY_P (c1, c2)))
6138 not = !not;
6139 }
6140 if (not)
6141 break;
6142 else
6143 goto fail;
6144
6145 case wordbeg:
6146 DEBUG_PRINT1 ("EXECUTING wordbeg.\n");
6147
6148 /* We FAIL in one of the following cases: */
6149
6150 /* Case 1: D is at the end of string. */
6151 if (AT_STRINGS_END (d))
6152 goto fail;
6153 else
6154 {
6155 /* C1 is the character before D, S1 is the syntax of C1, C2
6156 is the character at D, and S2 is the syntax of C2. */
6157 re_wchar_t c1, c2;
6158 int s1, s2;
6159 int dummy;
6160 #ifdef emacs
6161 int offset = PTR_TO_OFFSET (d);
6162 int charpos = SYNTAX_TABLE_BYTE_TO_CHAR (offset);
6163 UPDATE_SYNTAX_TABLE (charpos);
6164 #endif
6165 PREFETCH ();
6166 GET_CHAR_AFTER (c2, d, dummy);
6167 s2 = SYNTAX (c2);
6168
6169 /* Case 2: S2 is not Sword. */
6170 if (s2 != Sword)
6171 goto fail;
6172
6173 /* Case 3: D is not at the beginning of string ... */
6174 if (!AT_STRINGS_BEG (d))
6175 {
6176 GET_CHAR_BEFORE_2 (c1, d, string1, end1, string2, end2);
6177 #ifdef emacs
6178 UPDATE_SYNTAX_TABLE_BACKWARD (charpos - 1);
6179 #endif
6180 s1 = SYNTAX (c1);
6181
6182 /* ... and S1 is Sword, and WORD_BOUNDARY_P (C1, C2)
6183 returns 0. */
6184 if ((s1 == Sword) && !WORD_BOUNDARY_P (c1, c2))
6185 goto fail;
6186 }
6187 }
6188 break;
6189
6190 case wordend:
6191 DEBUG_PRINT1 ("EXECUTING wordend.\n");
6192
6193 /* We FAIL in one of the following cases: */
6194
6195 /* Case 1: D is at the beginning of string. */
6196 if (AT_STRINGS_BEG (d))
6197 goto fail;
6198 else
6199 {
6200 /* C1 is the character before D, S1 is the syntax of C1, C2
6201 is the character at D, and S2 is the syntax of C2. */
6202 re_wchar_t c1, c2;
6203 int s1, s2;
6204 int dummy;
6205 #ifdef emacs
6206 int offset = PTR_TO_OFFSET (d) - 1;
6207 int charpos = SYNTAX_TABLE_BYTE_TO_CHAR (offset);
6208 UPDATE_SYNTAX_TABLE (charpos);
6209 #endif
6210 GET_CHAR_BEFORE_2 (c1, d, string1, end1, string2, end2);
6211 s1 = SYNTAX (c1);
6212
6213 /* Case 2: S1 is not Sword. */
6214 if (s1 != Sword)
6215 goto fail;
6216
6217 /* Case 3: D is not at the end of string ... */
6218 if (!AT_STRINGS_END (d))
6219 {
6220 PREFETCH_NOLIMIT ();
6221 GET_CHAR_AFTER (c2, d, dummy);
6222 #ifdef emacs
6223 UPDATE_SYNTAX_TABLE_FORWARD (charpos);
6224 #endif
6225 s2 = SYNTAX (c2);
6226
6227 /* ... and S2 is Sword, and WORD_BOUNDARY_P (C1, C2)
6228 returns 0. */
6229 if ((s2 == Sword) && !WORD_BOUNDARY_P (c1, c2))
6230 goto fail;
6231 }
6232 }
6233 break;
6234
6235 case symbeg:
6236 DEBUG_PRINT1 ("EXECUTING symbeg.\n");
6237
6238 /* We FAIL in one of the following cases: */
6239
6240 /* Case 1: D is at the end of string. */
6241 if (AT_STRINGS_END (d))
6242 goto fail;
6243 else
6244 {
6245 /* C1 is the character before D, S1 is the syntax of C1, C2
6246 is the character at D, and S2 is the syntax of C2. */
6247 re_wchar_t c1, c2;
6248 int s1, s2;
6249 #ifdef emacs
6250 int offset = PTR_TO_OFFSET (d);
6251 int charpos = SYNTAX_TABLE_BYTE_TO_CHAR (offset);
6252 UPDATE_SYNTAX_TABLE (charpos);
6253 #endif
6254 PREFETCH ();
6255 c2 = RE_STRING_CHAR (d, target_multibyte);
6256 s2 = SYNTAX (c2);
6257
6258 /* Case 2: S2 is neither Sword nor Ssymbol. */
6259 if (s2 != Sword && s2 != Ssymbol)
6260 goto fail;
6261
6262 /* Case 3: D is not at the beginning of string ... */
6263 if (!AT_STRINGS_BEG (d))
6264 {
6265 GET_CHAR_BEFORE_2 (c1, d, string1, end1, string2, end2);
6266 #ifdef emacs
6267 UPDATE_SYNTAX_TABLE_BACKWARD (charpos - 1);
6268 #endif
6269 s1 = SYNTAX (c1);
6270
6271 /* ... and S1 is Sword or Ssymbol. */
6272 if (s1 == Sword || s1 == Ssymbol)
6273 goto fail;
6274 }
6275 }
6276 break;
6277
6278 case symend:
6279 DEBUG_PRINT1 ("EXECUTING symend.\n");
6280
6281 /* We FAIL in one of the following cases: */
6282
6283 /* Case 1: D is at the beginning of string. */
6284 if (AT_STRINGS_BEG (d))
6285 goto fail;
6286 else
6287 {
6288 /* C1 is the character before D, S1 is the syntax of C1, C2
6289 is the character at D, and S2 is the syntax of C2. */
6290 re_wchar_t c1, c2;
6291 int s1, s2;
6292 #ifdef emacs
6293 int offset = PTR_TO_OFFSET (d) - 1;
6294 int charpos = SYNTAX_TABLE_BYTE_TO_CHAR (offset);
6295 UPDATE_SYNTAX_TABLE (charpos);
6296 #endif
6297 GET_CHAR_BEFORE_2 (c1, d, string1, end1, string2, end2);
6298 s1 = SYNTAX (c1);
6299
6300 /* Case 2: S1 is neither Ssymbol nor Sword. */
6301 if (s1 != Sword && s1 != Ssymbol)
6302 goto fail;
6303
6304 /* Case 3: D is not at the end of string ... */
6305 if (!AT_STRINGS_END (d))
6306 {
6307 PREFETCH_NOLIMIT ();
6308 c2 = RE_STRING_CHAR (d, target_multibyte);
6309 #ifdef emacs
6310 UPDATE_SYNTAX_TABLE_FORWARD (charpos + 1);
6311 #endif
6312 s2 = SYNTAX (c2);
6313
6314 /* ... and S2 is Sword or Ssymbol. */
6315 if (s2 == Sword || s2 == Ssymbol)
6316 goto fail;
6317 }
6318 }
6319 break;
6320
6321 case syntaxspec:
6322 case notsyntaxspec:
6323 not = (re_opcode_t) *(p - 1) == notsyntaxspec;
6324 mcnt = *p++;
6325 DEBUG_PRINT3 ("EXECUTING %ssyntaxspec %d.\n", not?"not":"", mcnt);
6326 PREFETCH ();
6327 #ifdef emacs
6328 {
6329 int offset = PTR_TO_OFFSET (d);
6330 int pos1 = SYNTAX_TABLE_BYTE_TO_CHAR (offset);
6331 UPDATE_SYNTAX_TABLE (pos1);
6332 }
6333 #endif
6334 {
6335 int len;
6336 re_wchar_t c;
6337
6338 GET_CHAR_AFTER (c, d, len);
6339 if ((SYNTAX (c) != (enum syntaxcode) mcnt) ^ not)
6340 goto fail;
6341 d += len;
6342 }
6343 break;
6344
6345 #ifdef emacs
6346 case before_dot:
6347 DEBUG_PRINT1 ("EXECUTING before_dot.\n");
6348 if (PTR_BYTE_POS (d) >= PT_BYTE)
6349 goto fail;
6350 break;
6351
6352 case at_dot:
6353 DEBUG_PRINT1 ("EXECUTING at_dot.\n");
6354 if (PTR_BYTE_POS (d) != PT_BYTE)
6355 goto fail;
6356 break;
6357
6358 case after_dot:
6359 DEBUG_PRINT1 ("EXECUTING after_dot.\n");
6360 if (PTR_BYTE_POS (d) <= PT_BYTE)
6361 goto fail;
6362 break;
6363
6364 case categoryspec:
6365 case notcategoryspec:
6366 not = (re_opcode_t) *(p - 1) == notcategoryspec;
6367 mcnt = *p++;
6368 DEBUG_PRINT3 ("EXECUTING %scategoryspec %d.\n", not?"not":"", mcnt);
6369 PREFETCH ();
6370 {
6371 int len;
6372 re_wchar_t c;
6373
6374 GET_CHAR_AFTER (c, d, len);
6375 if ((!CHAR_HAS_CATEGORY (c, mcnt)) ^ not)
6376 goto fail;
6377 d += len;
6378 }
6379 break;
6380
6381 #endif /* emacs */
6382
6383 default:
6384 abort ();
6385 }
6386 continue; /* Successfully executed one pattern command; keep going. */
6387
6388
6389 /* We goto here if a matching operation fails. */
6390 fail:
6391 IMMEDIATE_QUIT_CHECK;
6392 if (!FAIL_STACK_EMPTY ())
6393 {
6394 re_char *str, *pat;
6395 /* A restart point is known. Restore to that state. */
6396 DEBUG_PRINT1 ("\nFAIL:\n");
6397 POP_FAILURE_POINT (str, pat);
6398 switch (SWITCH_ENUM_CAST ((re_opcode_t) *pat++))
6399 {
6400 case on_failure_keep_string_jump:
6401 assert (str == NULL);
6402 goto continue_failure_jump;
6403
6404 case on_failure_jump_nastyloop:
6405 assert ((re_opcode_t)pat[-2] == no_op);
6406 PUSH_FAILURE_POINT (pat - 2, str);
6407 /* Fallthrough */
6408
6409 case on_failure_jump_loop:
6410 case on_failure_jump:
6411 case succeed_n:
6412 d = str;
6413 continue_failure_jump:
6414 EXTRACT_NUMBER_AND_INCR (mcnt, pat);
6415 p = pat + mcnt;
6416 break;
6417
6418 case no_op:
6419 /* A special frame used for nastyloops. */
6420 goto fail;
6421
6422 default:
6423 abort();
6424 }
6425
6426 assert (p >= bufp->buffer && p <= pend);
6427
6428 if (d >= string1 && d <= end1)
6429 dend = end_match_1;
6430 }
6431 else
6432 break; /* Matching at this starting point really fails. */
6433 } /* for (;;) */
6434
6435 if (best_regs_set)
6436 goto restore_best_regs;
6437
6438 FREE_VARIABLES ();
6439
6440 return -1; /* Failure to match. */
6441 } /* re_match_2 */
6442 \f
6443 /* Subroutine definitions for re_match_2. */
6444
6445 /* Return zero if TRANSLATE[S1] and TRANSLATE[S2] are identical for LEN
6446 bytes; nonzero otherwise. */
6447
6448 static int
6449 bcmp_translate (s1, s2, len, translate, target_multibyte)
6450 re_char *s1, *s2;
6451 register int len;
6452 RE_TRANSLATE_TYPE translate;
6453 const int target_multibyte;
6454 {
6455 register re_char *p1 = s1, *p2 = s2;
6456 re_char *p1_end = s1 + len;
6457 re_char *p2_end = s2 + len;
6458
6459 /* FIXME: Checking both p1 and p2 presumes that the two strings might have
6460 different lengths, but relying on a single `len' would break this. -sm */
6461 while (p1 < p1_end && p2 < p2_end)
6462 {
6463 int p1_charlen, p2_charlen;
6464 re_wchar_t p1_ch, p2_ch;
6465
6466 GET_CHAR_AFTER (p1_ch, p1, p1_charlen);
6467 GET_CHAR_AFTER (p2_ch, p2, p2_charlen);
6468
6469 if (RE_TRANSLATE (translate, p1_ch)
6470 != RE_TRANSLATE (translate, p2_ch))
6471 return 1;
6472
6473 p1 += p1_charlen, p2 += p2_charlen;
6474 }
6475
6476 if (p1 != p1_end || p2 != p2_end)
6477 return 1;
6478
6479 return 0;
6480 }
6481 \f
6482 /* Entry points for GNU code. */
6483
6484 /* re_compile_pattern is the GNU regular expression compiler: it
6485 compiles PATTERN (of length SIZE) and puts the result in BUFP.
6486 Returns 0 if the pattern was valid, otherwise an error string.
6487
6488 Assumes the `allocated' (and perhaps `buffer') and `translate' fields
6489 are set in BUFP on entry.
6490
6491 We call regex_compile to do the actual compilation. */
6492
6493 const char *
6494 re_compile_pattern (pattern, length, bufp)
6495 const char *pattern;
6496 size_t length;
6497 struct re_pattern_buffer *bufp;
6498 {
6499 reg_errcode_t ret;
6500
6501 #ifdef emacs
6502 gl_state.current_syntax_table = current_buffer->syntax_table;
6503 #endif
6504
6505 /* GNU code is written to assume at least RE_NREGS registers will be set
6506 (and at least one extra will be -1). */
6507 bufp->regs_allocated = REGS_UNALLOCATED;
6508
6509 /* And GNU code determines whether or not to get register information
6510 by passing null for the REGS argument to re_match, etc., not by
6511 setting no_sub. */
6512 bufp->no_sub = 0;
6513
6514 ret = regex_compile ((re_char*) pattern, length, re_syntax_options, bufp);
6515
6516 if (!ret)
6517 return NULL;
6518 return gettext (re_error_msgid[(int) ret]);
6519 }
6520 WEAK_ALIAS (__re_compile_pattern, re_compile_pattern)
6521 \f
6522 /* Entry points compatible with 4.2 BSD regex library. We don't define
6523 them unless specifically requested. */
6524
6525 #if defined _REGEX_RE_COMP || defined _LIBC
6526
6527 /* BSD has one and only one pattern buffer. */
6528 static struct re_pattern_buffer re_comp_buf;
6529
6530 char *
6531 # ifdef _LIBC
6532 /* Make these definitions weak in libc, so POSIX programs can redefine
6533 these names if they don't use our functions, and still use
6534 regcomp/regexec below without link errors. */
6535 weak_function
6536 # endif
6537 re_comp (s)
6538 const char *s;
6539 {
6540 reg_errcode_t ret;
6541
6542 if (!s)
6543 {
6544 if (!re_comp_buf.buffer)
6545 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6546 return (char *) gettext ("No previous regular expression");
6547 return 0;
6548 }
6549
6550 if (!re_comp_buf.buffer)
6551 {
6552 re_comp_buf.buffer = (unsigned char *) malloc (200);
6553 if (re_comp_buf.buffer == NULL)
6554 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6555 return (char *) gettext (re_error_msgid[(int) REG_ESPACE]);
6556 re_comp_buf.allocated = 200;
6557
6558 re_comp_buf.fastmap = (char *) malloc (1 << BYTEWIDTH);
6559 if (re_comp_buf.fastmap == NULL)
6560 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6561 return (char *) gettext (re_error_msgid[(int) REG_ESPACE]);
6562 }
6563
6564 /* Since `re_exec' always passes NULL for the `regs' argument, we
6565 don't need to initialize the pattern buffer fields which affect it. */
6566
6567 ret = regex_compile (s, strlen (s), re_syntax_options, &re_comp_buf);
6568
6569 if (!ret)
6570 return NULL;
6571
6572 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6573 return (char *) gettext (re_error_msgid[(int) ret]);
6574 }
6575
6576
6577 int
6578 # ifdef _LIBC
6579 weak_function
6580 # endif
6581 re_exec (s)
6582 const char *s;
6583 {
6584 const int len = strlen (s);
6585 return
6586 0 <= re_search (&re_comp_buf, s, len, 0, len, (struct re_registers *) 0);
6587 }
6588 #endif /* _REGEX_RE_COMP */
6589 \f
6590 /* POSIX.2 functions. Don't define these for Emacs. */
6591
6592 #ifndef emacs
6593
6594 /* regcomp takes a regular expression as a string and compiles it.
6595
6596 PREG is a regex_t *. We do not expect any fields to be initialized,
6597 since POSIX says we shouldn't. Thus, we set
6598
6599 `buffer' to the compiled pattern;
6600 `used' to the length of the compiled pattern;
6601 `syntax' to RE_SYNTAX_POSIX_EXTENDED if the
6602 REG_EXTENDED bit in CFLAGS is set; otherwise, to
6603 RE_SYNTAX_POSIX_BASIC;
6604 `fastmap' to an allocated space for the fastmap;
6605 `fastmap_accurate' to zero;
6606 `re_nsub' to the number of subexpressions in PATTERN.
6607
6608 PATTERN is the address of the pattern string.
6609
6610 CFLAGS is a series of bits which affect compilation.
6611
6612 If REG_EXTENDED is set, we use POSIX extended syntax; otherwise, we
6613 use POSIX basic syntax.
6614
6615 If REG_NEWLINE is set, then . and [^...] don't match newline.
6616 Also, regexec will try a match beginning after every newline.
6617
6618 If REG_ICASE is set, then we considers upper- and lowercase
6619 versions of letters to be equivalent when matching.
6620
6621 If REG_NOSUB is set, then when PREG is passed to regexec, that
6622 routine will report only success or failure, and nothing about the
6623 registers.
6624
6625 It returns 0 if it succeeds, nonzero if it doesn't. (See regex.h for
6626 the return codes and their meanings.) */
6627
6628 int
6629 regcomp (preg, pattern, cflags)
6630 regex_t *__restrict preg;
6631 const char *__restrict pattern;
6632 int cflags;
6633 {
6634 reg_errcode_t ret;
6635 reg_syntax_t syntax
6636 = (cflags & REG_EXTENDED) ?
6637 RE_SYNTAX_POSIX_EXTENDED : RE_SYNTAX_POSIX_BASIC;
6638
6639 /* regex_compile will allocate the space for the compiled pattern. */
6640 preg->buffer = 0;
6641 preg->allocated = 0;
6642 preg->used = 0;
6643
6644 /* Try to allocate space for the fastmap. */
6645 preg->fastmap = (char *) malloc (1 << BYTEWIDTH);
6646
6647 if (cflags & REG_ICASE)
6648 {
6649 unsigned i;
6650
6651 preg->translate
6652 = (RE_TRANSLATE_TYPE) malloc (CHAR_SET_SIZE
6653 * sizeof (*(RE_TRANSLATE_TYPE)0));
6654 if (preg->translate == NULL)
6655 return (int) REG_ESPACE;
6656
6657 /* Map uppercase characters to corresponding lowercase ones. */
6658 for (i = 0; i < CHAR_SET_SIZE; i++)
6659 preg->translate[i] = ISUPPER (i) ? TOLOWER (i) : i;
6660 }
6661 else
6662 preg->translate = NULL;
6663
6664 /* If REG_NEWLINE is set, newlines are treated differently. */
6665 if (cflags & REG_NEWLINE)
6666 { /* REG_NEWLINE implies neither . nor [^...] match newline. */
6667 syntax &= ~RE_DOT_NEWLINE;
6668 syntax |= RE_HAT_LISTS_NOT_NEWLINE;
6669 }
6670 else
6671 syntax |= RE_NO_NEWLINE_ANCHOR;
6672
6673 preg->no_sub = !!(cflags & REG_NOSUB);
6674
6675 /* POSIX says a null character in the pattern terminates it, so we
6676 can use strlen here in compiling the pattern. */
6677 ret = regex_compile ((re_char*) pattern, strlen (pattern), syntax, preg);
6678
6679 /* POSIX doesn't distinguish between an unmatched open-group and an
6680 unmatched close-group: both are REG_EPAREN. */
6681 if (ret == REG_ERPAREN)
6682 ret = REG_EPAREN;
6683
6684 if (ret == REG_NOERROR && preg->fastmap)
6685 { /* Compute the fastmap now, since regexec cannot modify the pattern
6686 buffer. */
6687 re_compile_fastmap (preg);
6688 if (preg->can_be_null)
6689 { /* The fastmap can't be used anyway. */
6690 free (preg->fastmap);
6691 preg->fastmap = NULL;
6692 }
6693 }
6694 return (int) ret;
6695 }
6696 WEAK_ALIAS (__regcomp, regcomp)
6697
6698
6699 /* regexec searches for a given pattern, specified by PREG, in the
6700 string STRING.
6701
6702 If NMATCH is zero or REG_NOSUB was set in the cflags argument to
6703 `regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at
6704 least NMATCH elements, and we set them to the offsets of the
6705 corresponding matched substrings.
6706
6707 EFLAGS specifies `execution flags' which affect matching: if
6708 REG_NOTBOL is set, then ^ does not match at the beginning of the
6709 string; if REG_NOTEOL is set, then $ does not match at the end.
6710
6711 We return 0 if we find a match and REG_NOMATCH if not. */
6712
6713 int
6714 regexec (preg, string, nmatch, pmatch, eflags)
6715 const regex_t *__restrict preg;
6716 const char *__restrict string;
6717 size_t nmatch;
6718 regmatch_t pmatch[__restrict_arr];
6719 int eflags;
6720 {
6721 int ret;
6722 struct re_registers regs;
6723 regex_t private_preg;
6724 int len = strlen (string);
6725 boolean want_reg_info = !preg->no_sub && nmatch > 0 && pmatch;
6726
6727 private_preg = *preg;
6728
6729 private_preg.not_bol = !!(eflags & REG_NOTBOL);
6730 private_preg.not_eol = !!(eflags & REG_NOTEOL);
6731
6732 /* The user has told us exactly how many registers to return
6733 information about, via `nmatch'. We have to pass that on to the
6734 matching routines. */
6735 private_preg.regs_allocated = REGS_FIXED;
6736
6737 if (want_reg_info)
6738 {
6739 regs.num_regs = nmatch;
6740 regs.start = TALLOC (nmatch * 2, regoff_t);
6741 if (regs.start == NULL)
6742 return (int) REG_NOMATCH;
6743 regs.end = regs.start + nmatch;
6744 }
6745
6746 /* Instead of using not_eol to implement REG_NOTEOL, we could simply
6747 pass (&private_preg, string, len + 1, 0, len, ...) pretending the string
6748 was a little bit longer but still only matching the real part.
6749 This works because the `endline' will check for a '\n' and will find a
6750 '\0', correctly deciding that this is not the end of a line.
6751 But it doesn't work out so nicely for REG_NOTBOL, since we don't have
6752 a convenient '\0' there. For all we know, the string could be preceded
6753 by '\n' which would throw things off. */
6754
6755 /* Perform the searching operation. */
6756 ret = re_search (&private_preg, string, len,
6757 /* start: */ 0, /* range: */ len,
6758 want_reg_info ? &regs : (struct re_registers *) 0);
6759
6760 /* Copy the register information to the POSIX structure. */
6761 if (want_reg_info)
6762 {
6763 if (ret >= 0)
6764 {
6765 unsigned r;
6766
6767 for (r = 0; r < nmatch; r++)
6768 {
6769 pmatch[r].rm_so = regs.start[r];
6770 pmatch[r].rm_eo = regs.end[r];
6771 }
6772 }
6773
6774 /* If we needed the temporary register info, free the space now. */
6775 free (regs.start);
6776 }
6777
6778 /* We want zero return to mean success, unlike `re_search'. */
6779 return ret >= 0 ? (int) REG_NOERROR : (int) REG_NOMATCH;
6780 }
6781 WEAK_ALIAS (__regexec, regexec)
6782
6783
6784 /* Returns a message corresponding to an error code, ERR_CODE, returned
6785 from either regcomp or regexec. We don't use PREG here.
6786
6787 ERR_CODE was previously called ERRCODE, but that name causes an
6788 error with msvc8 compiler. */
6789
6790 size_t
6791 regerror (err_code, preg, errbuf, errbuf_size)
6792 int err_code;
6793 const regex_t *preg;
6794 char *errbuf;
6795 size_t errbuf_size;
6796 {
6797 const char *msg;
6798 size_t msg_size;
6799
6800 if (err_code < 0
6801 || err_code >= (sizeof (re_error_msgid) / sizeof (re_error_msgid[0])))
6802 /* Only error codes returned by the rest of the code should be passed
6803 to this routine. If we are given anything else, or if other regex
6804 code generates an invalid error code, then the program has a bug.
6805 Dump core so we can fix it. */
6806 abort ();
6807
6808 msg = gettext (re_error_msgid[err_code]);
6809
6810 msg_size = strlen (msg) + 1; /* Includes the null. */
6811
6812 if (errbuf_size != 0)
6813 {
6814 if (msg_size > errbuf_size)
6815 {
6816 strncpy (errbuf, msg, errbuf_size - 1);
6817 errbuf[errbuf_size - 1] = 0;
6818 }
6819 else
6820 strcpy (errbuf, msg);
6821 }
6822
6823 return msg_size;
6824 }
6825 WEAK_ALIAS (__regerror, regerror)
6826
6827
6828 /* Free dynamically allocated space used by PREG. */
6829
6830 void
6831 regfree (preg)
6832 regex_t *preg;
6833 {
6834 free (preg->buffer);
6835 preg->buffer = NULL;
6836
6837 preg->allocated = 0;
6838 preg->used = 0;
6839
6840 free (preg->fastmap);
6841 preg->fastmap = NULL;
6842 preg->fastmap_accurate = 0;
6843
6844 free (preg->translate);
6845 preg->translate = NULL;
6846 }
6847 WEAK_ALIAS (__regfree, regfree)
6848
6849 #endif /* not emacs */
6850
6851 /* arch-tag: 4ffd68ba-2a9e-435b-a21a-018990f9eeb2
6852 (do not change this comment) */
Emacs repository mirror