?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Auto-commit of generated files.
[emacs.git] / src / regex.c
blob4ab98bbf0984435adba01eda7c21a25addb13792
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) 1993-2013 Free Software Foundation, Inc.
6
7 This program is free software; you can redistribute it and/or modify
8 it under the terms of the GNU General Public License as published by
9 the Free Software Foundation; either version 3, or (at your option)
10 any later version.
11
12 This program is distributed in the hope that it will be useful,
13 but WITHOUT ANY WARRANTY; without even the implied warranty of
14 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 GNU General Public License for more details.
16
17 You should have received a copy of the GNU General Public License
18 along with this program. If not, see <http://www.gnu.org/licenses/>. */
19
20 /* TODO:
21 - structure the opcode space into opcode+flag.
22 - merge with glibc's regex.[ch].
23 - replace (succeed_n + jump_n + set_number_at) with something that doesn't
24 need to modify the compiled regexp so that re_match can be reentrant.
25 - get rid of on_failure_jump_smart by doing the optimization in re_comp
26 rather than at run-time, so that re_match can be reentrant.
27 */
28
29 /* AIX requires this to be the first thing in the file. */
30 #if defined _AIX && !defined REGEX_MALLOC
31 #pragma alloca
32 #endif
33
34 /* Ignore some GCC warnings for now. This section should go away
35 once the Emacs and Gnulib regex code is merged. */
36 #if 4 < __GNUC__ + (5 <= __GNUC_MINOR__) || defined __clang__
37 # pragma GCC diagnostic ignored "-Wstrict-overflow"
38 # ifndef emacs
39 # pragma GCC diagnostic ignored "-Wunused-function"
40 # pragma GCC diagnostic ignored "-Wunused-macros"
41 # pragma GCC diagnostic ignored "-Wunused-result"
42 # pragma GCC diagnostic ignored "-Wunused-variable"
43 # endif
44 #endif
45
46 #if 4 < __GNUC__ + (5 <= __GNUC_MINOR__) && ! defined __clang__
47 # pragma GCC diagnostic ignored "-Wunused-but-set-variable"
48 #endif
49
50 #include <config.h>
51
52 #include <stddef.h>
53
54 #ifdef emacs
55 /* We need this for `regex.h', and perhaps for the Emacs include files. */
56 # include <sys/types.h>
57 #endif
58
59 /* Whether to use ISO C Amendment 1 wide char functions.
60 Those should not be used for Emacs since it uses its own. */
61 #if defined _LIBC
62 #define WIDE_CHAR_SUPPORT 1
63 #else
64 #define WIDE_CHAR_SUPPORT \
65 (HAVE_WCTYPE_H && HAVE_WCHAR_H && HAVE_BTOWC && !emacs)
66 #endif
67
68 /* For platform which support the ISO C amendment 1 functionality we
69 support user defined character classes. */
70 #if WIDE_CHAR_SUPPORT
71 /* Solaris 2.5 has a bug: <wchar.h> must be included before <wctype.h>. */
72 # include <wchar.h>
73 # include <wctype.h>
74 #endif
75
76 #ifdef _LIBC
77 /* We have to keep the namespace clean. */
78 # define regfree(preg) __regfree (preg)
79 # define regexec(pr, st, nm, pm, ef) __regexec (pr, st, nm, pm, ef)
80 # define regcomp(preg, pattern, cflags) __regcomp (preg, pattern, cflags)
81 # define regerror(err_code, preg, errbuf, errbuf_size) \
82 __regerror (err_code, preg, errbuf, errbuf_size)
83 # define re_set_registers(bu, re, nu, st, en) \
84 __re_set_registers (bu, re, nu, st, en)
85 # define re_match_2(bufp, string1, size1, string2, size2, pos, regs, stop) \
86 __re_match_2 (bufp, string1, size1, string2, size2, pos, regs, stop)
87 # define re_match(bufp, string, size, pos, regs) \
88 __re_match (bufp, string, size, pos, regs)
89 # define re_search(bufp, string, size, startpos, range, regs) \
90 __re_search (bufp, string, size, startpos, range, regs)
91 # define re_compile_pattern(pattern, length, bufp) \
92 __re_compile_pattern (pattern, length, bufp)
93 # define re_set_syntax(syntax) __re_set_syntax (syntax)
94 # define re_search_2(bufp, st1, s1, st2, s2, startpos, range, regs, stop) \
95 __re_search_2 (bufp, st1, s1, st2, s2, startpos, range, regs, stop)
96 # define re_compile_fastmap(bufp) __re_compile_fastmap (bufp)
97
98 /* Make sure we call libc's function even if the user overrides them. */
99 # define btowc __btowc
100 # define iswctype __iswctype
101 # define wctype __wctype
102
103 # define WEAK_ALIAS(a,b) weak_alias (a, b)
104
105 /* We are also using some library internals. */
106 # include <locale/localeinfo.h>
107 # include <locale/elem-hash.h>
108 # include <langinfo.h>
109 #else
110 # define WEAK_ALIAS(a,b)
111 #endif
112
113 /* This is for other GNU distributions with internationalized messages. */
114 #if HAVE_LIBINTL_H || defined _LIBC
115 # include <libintl.h>
116 #else
117 # define gettext(msgid) (msgid)
118 #endif
119
120 #ifndef gettext_noop
121 /* This define is so xgettext can find the internationalizable
122 strings. */
123 # define gettext_noop(String) String
124 #endif
125
126 /* The `emacs' switch turns on certain matching commands
127 that make sense only in Emacs. */
128 #ifdef emacs
129
130 # include "lisp.h"
131 # include "character.h"
132 # include "buffer.h"
133
134 # include "syntax.h"
135 # include "category.h"
136
137 /* Make syntax table lookup grant data in gl_state. */
138 # define SYNTAX(c) syntax_property (c, 1)
139
140 # ifdef malloc
141 # undef malloc
142 # endif
143 # define malloc xmalloc
144 # ifdef realloc
145 # undef realloc
146 # endif
147 # define realloc xrealloc
148 # ifdef free
149 # undef free
150 # endif
151 # define free xfree
152
153 /* Converts the pointer to the char to BEG-based offset from the start. */
154 # define PTR_TO_OFFSET(d) POS_AS_IN_BUFFER (POINTER_TO_OFFSET (d))
155 # define POS_AS_IN_BUFFER(p) ((p) + (NILP (re_match_object) || BUFFERP (re_match_object)))
156
157 # define RE_MULTIBYTE_P(bufp) ((bufp)->multibyte)
158 # define RE_TARGET_MULTIBYTE_P(bufp) ((bufp)->target_multibyte)
159 # define RE_STRING_CHAR(p, multibyte) \
160 (multibyte ? (STRING_CHAR (p)) : (*(p)))
161 # define RE_STRING_CHAR_AND_LENGTH(p, len, multibyte) \
162 (multibyte ? (STRING_CHAR_AND_LENGTH (p, len)) : ((len) = 1, *(p)))
163
164 # define RE_CHAR_TO_MULTIBYTE(c) UNIBYTE_TO_CHAR (c)
165
166 # define RE_CHAR_TO_UNIBYTE(c) CHAR_TO_BYTE_SAFE (c)
167
168 /* Set C a (possibly converted to multibyte) character before P. P
169 points into a string which is the virtual concatenation of STR1
170 (which ends at END1) or STR2 (which ends at END2). */
171 # define GET_CHAR_BEFORE_2(c, p, str1, end1, str2, end2) \
172 do { \
173 if (target_multibyte) \
174 { \
175 re_char *dtemp = (p) == (str2) ? (end1) : (p); \
176 re_char *dlimit = ((p) > (str2) && (p) <= (end2)) ? (str2) : (str1); \
177 while (dtemp-- > dlimit && !CHAR_HEAD_P (*dtemp)); \
178 c = STRING_CHAR (dtemp); \
179 } \
180 else \
181 { \
182 (c = ((p) == (str2) ? (end1) : (p))[-1]); \
183 (c) = RE_CHAR_TO_MULTIBYTE (c); \
184 } \
185 } while (0)
186
187 /* Set C a (possibly converted to multibyte) character at P, and set
188 LEN to the byte length of that character. */
189 # define GET_CHAR_AFTER(c, p, len) \
190 do { \
191 if (target_multibyte) \
192 (c) = STRING_CHAR_AND_LENGTH (p, len); \
193 else \
194 { \
195 (c) = *p; \
196 len = 1; \
197 (c) = RE_CHAR_TO_MULTIBYTE (c); \
198 } \
199 } while (0)
200
201 #else /* not emacs */
202
203 /* If we are not linking with Emacs proper,
204 we can't use the relocating allocator
205 even if config.h says that we can. */
206 # undef REL_ALLOC
207
208 # include <unistd.h>
209
210 /* When used in Emacs's lib-src, we need xmalloc and xrealloc. */
211
212 static void *
213 xmalloc (size_t size)
214 {
215 void *val = malloc (size);
216 if (!val && size)
217 {
218 write (2, "virtual memory exhausted\n", 25);
219 exit (1);
220 }
221 return val;
222 }
223
224 static void *
225 xrealloc (void *block, size_t size)
226 {
227 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 = malloc (size);
232 else
233 val = 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 # include <stdbool.h>
252 # include <string.h>
253
254 /* Define the syntax stuff for \<, \>, etc. */
255
256 /* Sword must be nonzero for the wordchar pattern commands in re_match_2. */
257 enum syntaxcode { Swhitespace = 0, Sword = 1, Ssymbol = 2 };
258
259 /* Dummy macros for non-Emacs environments. */
260 # define MAX_MULTIBYTE_LENGTH 1
261 # define RE_MULTIBYTE_P(x) 0
262 # define RE_TARGET_MULTIBYTE_P(x) 0
263 # define WORD_BOUNDARY_P(c1, c2) (0)
264 # define BYTES_BY_CHAR_HEAD(p) (1)
265 # define PREV_CHAR_BOUNDARY(p, limit) ((p)--)
266 # define STRING_CHAR(p) (*(p))
267 # define RE_STRING_CHAR(p, multibyte) STRING_CHAR (p)
268 # define CHAR_STRING(c, s) (*(s) = (c), 1)
269 # define STRING_CHAR_AND_LENGTH(p, actual_len) ((actual_len) = 1, *(p))
270 # define RE_STRING_CHAR_AND_LENGTH(p, len, multibyte) STRING_CHAR_AND_LENGTH (p, len)
271 # define RE_CHAR_TO_MULTIBYTE(c) (c)
272 # define RE_CHAR_TO_UNIBYTE(c) (c)
273 # define GET_CHAR_BEFORE_2(c, p, str1, end1, str2, end2) \
274 (c = ((p) == (str2) ? *((end1) - 1) : *((p) - 1)))
275 # define GET_CHAR_AFTER(c, p, len) \
276 (c = *p, len = 1)
277 # define CHAR_BYTE8_P(c) (0)
278 # define CHAR_LEADING_CODE(c) (c)
279
280 #endif /* not emacs */
281
282 #ifndef RE_TRANSLATE
283 # define RE_TRANSLATE(TBL, C) ((unsigned char)(TBL)[C])
284 # define RE_TRANSLATE_P(TBL) (TBL)
285 #endif
286 \f
287 /* Get the interface, including the syntax bits. */
288 #include "regex.h"
289
290 /* isalpha etc. are used for the character classes. */
291 #include <ctype.h>
292
293 #ifdef emacs
294
295 /* 1 if C is an ASCII character. */
296 # define IS_REAL_ASCII(c) ((c) < 0200)
297
298 /* 1 if C is a unibyte character. */
299 # define ISUNIBYTE(c) (SINGLE_BYTE_CHAR_P ((c)))
300
301 /* The Emacs definitions should not be directly affected by locales. */
302
303 /* In Emacs, these are only used for single-byte characters. */
304 # define ISDIGIT(c) ((c) >= '0' && (c) <= '9')
305 # define ISCNTRL(c) ((c) < ' ')
306 # define ISXDIGIT(c) (((c) >= '0' && (c) <= '9') \
307 || ((c) >= 'a' && (c) <= 'f') \
308 || ((c) >= 'A' && (c) <= 'F'))
309
310 /* This is only used for single-byte characters. */
311 # define ISBLANK(c) ((c) == ' ' || (c) == '\t')
312
313 /* The rest must handle multibyte characters. */
314
315 # define ISGRAPH(c) (SINGLE_BYTE_CHAR_P (c) \
316 ? (c) > ' ' && !((c) >= 0177 && (c) <= 0237) \
317 : 1)
318
319 # define ISPRINT(c) (SINGLE_BYTE_CHAR_P (c) \
320 ? (c) >= ' ' && !((c) >= 0177 && (c) <= 0237) \
321 : 1)
322
323 # define ISALNUM(c) (IS_REAL_ASCII (c) \
324 ? (((c) >= 'a' && (c) <= 'z') \
325 || ((c) >= 'A' && (c) <= 'Z') \
326 || ((c) >= '0' && (c) <= '9')) \
327 : SYNTAX (c) == Sword)
328
329 # define ISALPHA(c) (IS_REAL_ASCII (c) \
330 ? (((c) >= 'a' && (c) <= 'z') \
331 || ((c) >= 'A' && (c) <= 'Z')) \
332 : SYNTAX (c) == Sword)
333
334 # define ISLOWER(c) lowercasep (c)
335
336 # define ISPUNCT(c) (IS_REAL_ASCII (c) \
337 ? ((c) > ' ' && (c) < 0177 \
338 && !(((c) >= 'a' && (c) <= 'z') \
339 || ((c) >= 'A' && (c) <= 'Z') \
340 || ((c) >= '0' && (c) <= '9'))) \
341 : SYNTAX (c) != Sword)
342
343 # define ISSPACE(c) (SYNTAX (c) == Swhitespace)
344
345 # define ISUPPER(c) uppercasep (c)
346
347 # define ISWORD(c) (SYNTAX (c) == Sword)
348
349 #else /* not emacs */
350
351 /* 1 if C is an ASCII character. */
352 # define IS_REAL_ASCII(c) ((c) < 0200)
353
354 /* This distinction is not meaningful, except in Emacs. */
355 # define ISUNIBYTE(c) 1
356
357 # ifdef isblank
358 # define ISBLANK(c) isblank (c)
359 # else
360 # define ISBLANK(c) ((c) == ' ' || (c) == '\t')
361 # endif
362 # ifdef isgraph
363 # define ISGRAPH(c) isgraph (c)
364 # else
365 # define ISGRAPH(c) (isprint (c) && !isspace (c))
366 # endif
367
368 /* Solaris defines ISPRINT so we must undefine it first. */
369 # undef ISPRINT
370 # define ISPRINT(c) isprint (c)
371 # define ISDIGIT(c) isdigit (c)
372 # define ISALNUM(c) isalnum (c)
373 # define ISALPHA(c) isalpha (c)
374 # define ISCNTRL(c) iscntrl (c)
375 # define ISLOWER(c) islower (c)
376 # define ISPUNCT(c) ispunct (c)
377 # define ISSPACE(c) isspace (c)
378 # define ISUPPER(c) isupper (c)
379 # define ISXDIGIT(c) isxdigit (c)
380
381 # define ISWORD(c) ISALPHA (c)
382
383 # ifdef _tolower
384 # define TOLOWER(c) _tolower (c)
385 # else
386 # define TOLOWER(c) tolower (c)
387 # endif
388
389 /* How many characters in the character set. */
390 # define CHAR_SET_SIZE 256
391
392 # ifdef SYNTAX_TABLE
393
394 extern char *re_syntax_table;
395
396 # else /* not SYNTAX_TABLE */
397
398 static char re_syntax_table[CHAR_SET_SIZE];
399
400 static void
401 init_syntax_once (void)
402 {
403 register int c;
404 static int done = 0;
405
406 if (done)
407 return;
408
409 memset (re_syntax_table, 0, sizeof re_syntax_table);
410
411 for (c = 0; c < CHAR_SET_SIZE; ++c)
412 if (ISALNUM (c))
413 re_syntax_table[c] = Sword;
414
415 re_syntax_table['_'] = Ssymbol;
416
417 done = 1;
418 }
419
420 # endif /* not SYNTAX_TABLE */
421
422 # define SYNTAX(c) re_syntax_table[(c)]
423
424 #endif /* not emacs */
425 \f
426 #define SIGN_EXTEND_CHAR(c) ((signed char) (c))
427 \f
428 /* Should we use malloc or alloca? If REGEX_MALLOC is not defined, we
429 use `alloca' instead of `malloc'. This is because using malloc in
430 re_search* or re_match* could cause memory leaks when C-g is used in
431 Emacs; also, malloc is slower and causes storage fragmentation. On
432 the other hand, malloc is more portable, and easier to debug.
433
434 Because we sometimes use alloca, some routines have to be macros,
435 not functions -- `alloca'-allocated space disappears at the end of the
436 function it is called in. */
437
438 #ifdef REGEX_MALLOC
439
440 # define REGEX_ALLOCATE malloc
441 # define REGEX_REALLOCATE(source, osize, nsize) realloc (source, nsize)
442 # define REGEX_FREE free
443
444 #else /* not REGEX_MALLOC */
445
446 /* Emacs already defines alloca, sometimes. */
447 # ifndef alloca
448
449 /* Make alloca work the best possible way. */
450 # ifdef __GNUC__
451 # define alloca __builtin_alloca
452 # else /* not __GNUC__ */
453 # ifdef HAVE_ALLOCA_H
454 # include <alloca.h>
455 # endif /* HAVE_ALLOCA_H */
456 # endif /* not __GNUC__ */
457
458 # endif /* not alloca */
459
460 # define REGEX_ALLOCATE alloca
461
462 /* Assumes a `char *destination' variable. */
463 # define REGEX_REALLOCATE(source, osize, nsize) \
464 (destination = alloca (nsize), \
465 memcpy (destination, source, osize))
466
467 /* No need to do anything to free, after alloca. */
468 # define REGEX_FREE(arg) ((void)0) /* Do nothing! But inhibit gcc warning. */
469
470 #endif /* not REGEX_MALLOC */
471
472 /* Define how to allocate the failure stack. */
473
474 #if defined REL_ALLOC && defined REGEX_MALLOC
475
476 # define REGEX_ALLOCATE_STACK(size) \
477 r_alloc (&failure_stack_ptr, (size))
478 # define REGEX_REALLOCATE_STACK(source, osize, nsize) \
479 r_re_alloc (&failure_stack_ptr, (nsize))
480 # define REGEX_FREE_STACK(ptr) \
481 r_alloc_free (&failure_stack_ptr)
482
483 #else /* not using relocating allocator */
484
485 # ifdef REGEX_MALLOC
486
487 # define REGEX_ALLOCATE_STACK malloc
488 # define REGEX_REALLOCATE_STACK(source, osize, nsize) realloc (source, nsize)
489 # define REGEX_FREE_STACK free
490
491 # else /* not REGEX_MALLOC */
492
493 # define REGEX_ALLOCATE_STACK alloca
494
495 # define REGEX_REALLOCATE_STACK(source, osize, nsize) \
496 REGEX_REALLOCATE (source, osize, nsize)
497 /* No need to explicitly free anything. */
498 # define REGEX_FREE_STACK(arg) ((void)0)
499
500 # endif /* not REGEX_MALLOC */
501 #endif /* not using relocating allocator */
502
503
504 /* True if `size1' is non-NULL and PTR is pointing anywhere inside
505 `string1' or just past its end. This works if PTR is NULL, which is
506 a good thing. */
507 #define FIRST_STRING_P(ptr) \
508 (size1 && string1 <= (ptr) && (ptr) <= string1 + size1)
509
510 /* (Re)Allocate N items of type T using malloc, or fail. */
511 #define TALLOC(n, t) ((t *) malloc ((n) * sizeof (t)))
512 #define RETALLOC(addr, n, t) ((addr) = (t *) realloc (addr, (n) * sizeof (t)))
513 #define REGEX_TALLOC(n, t) ((t *) REGEX_ALLOCATE ((n) * sizeof (t)))
514
515 #define BYTEWIDTH 8 /* In bits. */
516
517 #define STREQ(s1, s2) ((strcmp (s1, s2) == 0))
518
519 #undef MAX
520 #undef MIN
521 #define MAX(a, b) ((a) > (b) ? (a) : (b))
522 #define MIN(a, b) ((a) < (b) ? (a) : (b))
523
524 /* Type of source-pattern and string chars. */
525 #ifdef _MSC_VER
526 typedef unsigned char re_char;
527 typedef const re_char const_re_char;
528 #else
529 typedef const unsigned char re_char;
530 typedef re_char const_re_char;
531 #endif
532
533 typedef char boolean;
534
535 static regoff_t re_match_2_internal (struct re_pattern_buffer *bufp,
536 re_char *string1, size_t size1,
537 re_char *string2, size_t size2,
538 ssize_t pos,
539 struct re_registers *regs,
540 ssize_t stop);
541 \f
542 /* These are the command codes that appear in compiled regular
543 expressions. Some opcodes are followed by argument bytes. A
544 command code can specify any interpretation whatsoever for its
545 arguments. Zero bytes may appear in the compiled regular expression. */
546
547 typedef enum
548 {
549 no_op = 0,
550
551 /* Succeed right away--no more backtracking. */
552 succeed,
553
554 /* Followed by one byte giving n, then by n literal bytes. */
555 exactn,
556
557 /* Matches any (more or less) character. */
558 anychar,
559
560 /* Matches any one char belonging to specified set. First
561 following byte is number of bitmap bytes. Then come bytes
562 for a bitmap saying which chars are in. Bits in each byte
563 are ordered low-bit-first. A character is in the set if its
564 bit is 1. A character too large to have a bit in the map is
565 automatically not in the set.
566
567 If the length byte has the 0x80 bit set, then that stuff
568 is followed by a range table:
569 2 bytes of flags for character sets (low 8 bits, high 8 bits)
570 See RANGE_TABLE_WORK_BITS below.
571 2 bytes, the number of pairs that follow (upto 32767)
572 pairs, each 2 multibyte characters,
573 each multibyte character represented as 3 bytes. */
574 charset,
575
576 /* Same parameters as charset, but match any character that is
577 not one of those specified. */
578 charset_not,
579
580 /* Start remembering the text that is matched, for storing in a
581 register. Followed by one byte with the register number, in
582 the range 0 to one less than the pattern buffer's re_nsub
583 field. */
584 start_memory,
585
586 /* Stop remembering the text that is matched and store it in a
587 memory register. Followed by one byte with the register
588 number, in the range 0 to one less than `re_nsub' in the
589 pattern buffer. */
590 stop_memory,
591
592 /* Match a duplicate of something remembered. Followed by one
593 byte containing the register number. */
594 duplicate,
595
596 /* Fail unless at beginning of line. */
597 begline,
598
599 /* Fail unless at end of line. */
600 endline,
601
602 /* Succeeds if at beginning of buffer (if emacs) or at beginning
603 of string to be matched (if not). */
604 begbuf,
605
606 /* Analogously, for end of buffer/string. */
607 endbuf,
608
609 /* Followed by two byte relative address to which to jump. */
610 jump,
611
612 /* Followed by two-byte relative address of place to resume at
613 in case of failure. */
614 on_failure_jump,
615
616 /* Like on_failure_jump, but pushes a placeholder instead of the
617 current string position when executed. */
618 on_failure_keep_string_jump,
619
620 /* Just like `on_failure_jump', except that it checks that we
621 don't get stuck in an infinite loop (matching an empty string
622 indefinitely). */
623 on_failure_jump_loop,
624
625 /* Just like `on_failure_jump_loop', except that it checks for
626 a different kind of loop (the kind that shows up with non-greedy
627 operators). This operation has to be immediately preceded
628 by a `no_op'. */
629 on_failure_jump_nastyloop,
630
631 /* A smart `on_failure_jump' used for greedy * and + operators.
632 It analyzes the loop before which it is put and if the
633 loop does not require backtracking, it changes itself to
634 `on_failure_keep_string_jump' and short-circuits the loop,
635 else it just defaults to changing itself into `on_failure_jump'.
636 It assumes that it is pointing to just past a `jump'. */
637 on_failure_jump_smart,
638
639 /* Followed by two-byte relative address and two-byte number n.
640 After matching N times, jump to the address upon failure.
641 Does not work if N starts at 0: use on_failure_jump_loop
642 instead. */
643 succeed_n,
644
645 /* Followed by two-byte relative address, and two-byte number n.
646 Jump to the address N times, then fail. */
647 jump_n,
648
649 /* Set the following two-byte relative address to the
650 subsequent two-byte number. The address *includes* the two
651 bytes of number. */
652 set_number_at,
653
654 wordbeg, /* Succeeds if at word beginning. */
655 wordend, /* Succeeds if at word end. */
656
657 wordbound, /* Succeeds if at a word boundary. */
658 notwordbound, /* Succeeds if not at a word boundary. */
659
660 symbeg, /* Succeeds if at symbol beginning. */
661 symend, /* Succeeds if at symbol end. */
662
663 /* Matches any character whose syntax is specified. Followed by
664 a byte which contains a syntax code, e.g., Sword. */
665 syntaxspec,
666
667 /* Matches any character whose syntax is not that specified. */
668 notsyntaxspec
669
670 #ifdef emacs
671 ,before_dot, /* Succeeds if before point. */
672 at_dot, /* Succeeds if at point. */
673 after_dot, /* Succeeds if after point. */
674
675 /* Matches any character whose category-set contains the specified
676 category. The operator is followed by a byte which contains a
677 category code (mnemonic ASCII character). */
678 categoryspec,
679
680 /* Matches any character whose category-set does not contain the
681 specified category. The operator is followed by a byte which
682 contains the category code (mnemonic ASCII character). */
683 notcategoryspec
684 #endif /* emacs */
685 } re_opcode_t;
686 \f
687 /* Common operations on the compiled pattern. */
688
689 /* Store NUMBER in two contiguous bytes starting at DESTINATION. */
690
691 #define STORE_NUMBER(destination, number) \
692 do { \
693 (destination)[0] = (number) & 0377; \
694 (destination)[1] = (number) >> 8; \
695 } while (0)
696
697 /* Same as STORE_NUMBER, except increment DESTINATION to
698 the byte after where the number is stored. Therefore, DESTINATION
699 must be an lvalue. */
700
701 #define STORE_NUMBER_AND_INCR(destination, number) \
702 do { \
703 STORE_NUMBER (destination, number); \
704 (destination) += 2; \
705 } while (0)
706
707 /* Put into DESTINATION a number stored in two contiguous bytes starting
708 at SOURCE. */
709
710 #define EXTRACT_NUMBER(destination, source) \
711 ((destination) = extract_number (source))
712
713 static int
714 extract_number (re_char *source)
715 {
716 return (SIGN_EXTEND_CHAR (source[1]) << 8) + source[0];
717 }
718
719 /* Same as EXTRACT_NUMBER, except increment SOURCE to after the number.
720 SOURCE must be an lvalue. */
721
722 #define EXTRACT_NUMBER_AND_INCR(destination, source) \
723 ((destination) = extract_number_and_incr (&source))
724
725 static int
726 extract_number_and_incr (re_char **source)
727 {
728 int num = extract_number (*source);
729 *source += 2;
730 return num;
731 }
732 \f
733 /* Store a multibyte character in three contiguous bytes starting
734 DESTINATION, and increment DESTINATION to the byte after where the
735 character is stored. Therefore, DESTINATION must be an lvalue. */
736
737 #define STORE_CHARACTER_AND_INCR(destination, character) \
738 do { \
739 (destination)[0] = (character) & 0377; \
740 (destination)[1] = ((character) >> 8) & 0377; \
741 (destination)[2] = (character) >> 16; \
742 (destination) += 3; \
743 } while (0)
744
745 /* Put into DESTINATION a character stored in three contiguous bytes
746 starting at SOURCE. */
747
748 #define EXTRACT_CHARACTER(destination, source) \
749 do { \
750 (destination) = ((source)[0] \
751 | ((source)[1] << 8) \
752 | ((source)[2] << 16)); \
753 } while (0)
754
755
756 /* Macros for charset. */
757
758 /* Size of bitmap of charset P in bytes. P is a start of charset,
759 i.e. *P is (re_opcode_t) charset or (re_opcode_t) charset_not. */
760 #define CHARSET_BITMAP_SIZE(p) ((p)[1] & 0x7F)
761
762 /* Nonzero if charset P has range table. */
763 #define CHARSET_RANGE_TABLE_EXISTS_P(p) ((p)[1] & 0x80)
764
765 /* Return the address of range table of charset P. But not the start
766 of table itself, but the before where the number of ranges is
767 stored. `2 +' means to skip re_opcode_t and size of bitmap,
768 and the 2 bytes of flags at the start of the range table. */
769 #define CHARSET_RANGE_TABLE(p) (&(p)[4 + CHARSET_BITMAP_SIZE (p)])
770
771 #ifdef emacs
772 /* Extract the bit flags that start a range table. */
773 #define CHARSET_RANGE_TABLE_BITS(p) \
774 ((p)[2 + CHARSET_BITMAP_SIZE (p)] \
775 + (p)[3 + CHARSET_BITMAP_SIZE (p)] * 0x100)
776 #endif
777
778 /* Return the address of end of RANGE_TABLE. COUNT is number of
779 ranges (which is a pair of (start, end)) in the RANGE_TABLE. `* 2'
780 is start of range and end of range. `* 3' is size of each start
781 and end. */
782 #define CHARSET_RANGE_TABLE_END(range_table, count) \
783 ((range_table) + (count) * 2 * 3)
784
785 /* Test if C is in RANGE_TABLE. A flag NOT is negated if C is in.
786 COUNT is number of ranges in RANGE_TABLE. */
787 #define CHARSET_LOOKUP_RANGE_TABLE_RAW(not, c, range_table, count) \
788 do \
789 { \
790 re_wchar_t range_start, range_end; \
791 re_char *rtp; \
792 re_char *range_table_end \
793 = CHARSET_RANGE_TABLE_END ((range_table), (count)); \
794 \
795 for (rtp = (range_table); rtp < range_table_end; rtp += 2 * 3) \
796 { \
797 EXTRACT_CHARACTER (range_start, rtp); \
798 EXTRACT_CHARACTER (range_end, rtp + 3); \
799 \
800 if (range_start <= (c) && (c) <= range_end) \
801 { \
802 (not) = !(not); \
803 break; \
804 } \
805 } \
806 } \
807 while (0)
808
809 /* Test if C is in range table of CHARSET. The flag NOT is negated if
810 C is listed in it. */
811 #define CHARSET_LOOKUP_RANGE_TABLE(not, c, charset) \
812 do \
813 { \
814 /* Number of ranges in range table. */ \
815 int count; \
816 re_char *range_table = CHARSET_RANGE_TABLE (charset); \
817 \
818 EXTRACT_NUMBER_AND_INCR (count, range_table); \
819 CHARSET_LOOKUP_RANGE_TABLE_RAW ((not), (c), range_table, count); \
820 } \
821 while (0)
822 \f
823 /* If DEBUG is defined, Regex prints many voluminous messages about what
824 it is doing (if the variable `debug' is nonzero). If linked with the
825 main program in `iregex.c', you can enter patterns and strings
826 interactively. And if linked with the main program in `main.c' and
827 the other test files, you can run the already-written tests. */
828
829 #ifdef DEBUG
830
831 /* We use standard I/O for debugging. */
832 # include <stdio.h>
833
834 /* It is useful to test things that ``must'' be true when debugging. */
835 # include <assert.h>
836
837 static int debug = -100000;
838
839 # define DEBUG_STATEMENT(e) e
840 # define DEBUG_PRINT(...) if (debug > 0) printf (__VA_ARGS__)
841 # define DEBUG_COMPILES_ARGUMENTS
842 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) \
843 if (debug > 0) print_partial_compiled_pattern (s, e)
844 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) \
845 if (debug > 0) print_double_string (w, s1, sz1, s2, sz2)
846
847
848 /* Print the fastmap in human-readable form. */
849
850 static void
851 print_fastmap (char *fastmap)
852 {
853 unsigned was_a_range = 0;
854 unsigned i = 0;
855
856 while (i < (1 << BYTEWIDTH))
857 {
858 if (fastmap[i++])
859 {
860 was_a_range = 0;
861 putchar (i - 1);
862 while (i < (1 << BYTEWIDTH) && fastmap[i])
863 {
864 was_a_range = 1;
865 i++;
866 }
867 if (was_a_range)
868 {
869 printf ("-");
870 putchar (i - 1);
871 }
872 }
873 }
874 putchar ('\n');
875 }
876
877
878 /* Print a compiled pattern string in human-readable form, starting at
879 the START pointer into it and ending just before the pointer END. */
880
881 static void
882 print_partial_compiled_pattern (re_char *start, re_char *end)
883 {
884 int mcnt, mcnt2;
885 re_char *p = start;
886 re_char *pend = end;
887
888 if (start == NULL)
889 {
890 fprintf (stderr, "(null)\n");
891 return;
892 }
893
894 /* Loop over pattern commands. */
895 while (p < pend)
896 {
897 fprintf (stderr, "%td:\t", p - start);
898
899 switch ((re_opcode_t) *p++)
900 {
901 case no_op:
902 fprintf (stderr, "/no_op");
903 break;
904
905 case succeed:
906 fprintf (stderr, "/succeed");
907 break;
908
909 case exactn:
910 mcnt = *p++;
911 fprintf (stderr, "/exactn/%d", mcnt);
912 do
913 {
914 fprintf (stderr, "/%c", *p++);
915 }
916 while (--mcnt);
917 break;
918
919 case start_memory:
920 fprintf (stderr, "/start_memory/%d", *p++);
921 break;
922
923 case stop_memory:
924 fprintf (stderr, "/stop_memory/%d", *p++);
925 break;
926
927 case duplicate:
928 fprintf (stderr, "/duplicate/%d", *p++);
929 break;
930
931 case anychar:
932 fprintf (stderr, "/anychar");
933 break;
934
935 case charset:
936 case charset_not:
937 {
938 register int c, last = -100;
939 register int in_range = 0;
940 int length = CHARSET_BITMAP_SIZE (p - 1);
941 int has_range_table = CHARSET_RANGE_TABLE_EXISTS_P (p - 1);
942
943 fprintf (stderr, "/charset [%s",
944 (re_opcode_t) *(p - 1) == charset_not ? "^" : "");
945
946 if (p + *p >= pend)
947 fprintf (stderr, " !extends past end of pattern! ");
948
949 for (c = 0; c < 256; c++)
950 if (c / 8 < length
951 && (p[1 + (c/8)] & (1 << (c % 8))))
952 {
953 /* Are we starting a range? */
954 if (last + 1 == c && ! in_range)
955 {
956 fprintf (stderr, "-");
957 in_range = 1;
958 }
959 /* Have we broken a range? */
960 else if (last + 1 != c && in_range)
961 {
962 fprintf (stderr, "%c", last);
963 in_range = 0;
964 }
965
966 if (! in_range)
967 fprintf (stderr, "%c", c);
968
969 last = c;
970 }
971
972 if (in_range)
973 fprintf (stderr, "%c", last);
974
975 fprintf (stderr, "]");
976
977 p += 1 + length;
978
979 if (has_range_table)
980 {
981 int count;
982 fprintf (stderr, "has-range-table");
983
984 /* ??? Should print the range table; for now, just skip it. */
985 p += 2; /* skip range table bits */
986 EXTRACT_NUMBER_AND_INCR (count, p);
987 p = CHARSET_RANGE_TABLE_END (p, count);
988 }
989 }
990 break;
991
992 case begline:
993 fprintf (stderr, "/begline");
994 break;
995
996 case endline:
997 fprintf (stderr, "/endline");
998 break;
999
1000 case on_failure_jump:
1001 EXTRACT_NUMBER_AND_INCR (mcnt, p);
1002 fprintf (stderr, "/on_failure_jump to %td", p + mcnt - start);
1003 break;
1004
1005 case on_failure_keep_string_jump:
1006 EXTRACT_NUMBER_AND_INCR (mcnt, p);
1007 fprintf (stderr, "/on_failure_keep_string_jump to %td",
1008 p + mcnt - start);
1009 break;
1010
1011 case on_failure_jump_nastyloop:
1012 EXTRACT_NUMBER_AND_INCR (mcnt, p);
1013 fprintf (stderr, "/on_failure_jump_nastyloop to %td",
1014 p + mcnt - start);
1015 break;
1016
1017 case on_failure_jump_loop:
1018 EXTRACT_NUMBER_AND_INCR (mcnt, p);
1019 fprintf (stderr, "/on_failure_jump_loop to %td",
1020 p + mcnt - start);
1021 break;
1022
1023 case on_failure_jump_smart:
1024 EXTRACT_NUMBER_AND_INCR (mcnt, p);
1025 fprintf (stderr, "/on_failure_jump_smart to %td",
1026 p + mcnt - start);
1027 break;
1028
1029 case jump:
1030 EXTRACT_NUMBER_AND_INCR (mcnt, p);
1031 fprintf (stderr, "/jump to %td", p + mcnt - start);
1032 break;
1033
1034 case succeed_n:
1035 EXTRACT_NUMBER_AND_INCR (mcnt, p);
1036 EXTRACT_NUMBER_AND_INCR (mcnt2, p);
1037 fprintf (stderr, "/succeed_n to %td, %d times",
1038 p - 2 + mcnt - start, mcnt2);
1039 break;
1040
1041 case jump_n:
1042 EXTRACT_NUMBER_AND_INCR (mcnt, p);
1043 EXTRACT_NUMBER_AND_INCR (mcnt2, p);
1044 fprintf (stderr, "/jump_n to %td, %d times",
1045 p - 2 + mcnt - start, mcnt2);
1046 break;
1047
1048 case set_number_at:
1049 EXTRACT_NUMBER_AND_INCR (mcnt, p);
1050 EXTRACT_NUMBER_AND_INCR (mcnt2, p);
1051 fprintf (stderr, "/set_number_at location %td to %d",
1052 p - 2 + mcnt - start, mcnt2);
1053 break;
1054
1055 case wordbound:
1056 fprintf (stderr, "/wordbound");
1057 break;
1058
1059 case notwordbound:
1060 fprintf (stderr, "/notwordbound");
1061 break;
1062
1063 case wordbeg:
1064 fprintf (stderr, "/wordbeg");
1065 break;
1066
1067 case wordend:
1068 fprintf (stderr, "/wordend");
1069 break;
1070
1071 case symbeg:
1072 fprintf (stderr, "/symbeg");
1073 break;
1074
1075 case symend:
1076 fprintf (stderr, "/symend");
1077 break;
1078
1079 case syntaxspec:
1080 fprintf (stderr, "/syntaxspec");
1081 mcnt = *p++;
1082 fprintf (stderr, "/%d", mcnt);
1083 break;
1084
1085 case notsyntaxspec:
1086 fprintf (stderr, "/notsyntaxspec");
1087 mcnt = *p++;
1088 fprintf (stderr, "/%d", mcnt);
1089 break;
1090
1091 # ifdef emacs
1092 case before_dot:
1093 fprintf (stderr, "/before_dot");
1094 break;
1095
1096 case at_dot:
1097 fprintf (stderr, "/at_dot");
1098 break;
1099
1100 case after_dot:
1101 fprintf (stderr, "/after_dot");
1102 break;
1103
1104 case categoryspec:
1105 fprintf (stderr, "/categoryspec");
1106 mcnt = *p++;
1107 fprintf (stderr, "/%d", mcnt);
1108 break;
1109
1110 case notcategoryspec:
1111 fprintf (stderr, "/notcategoryspec");
1112 mcnt = *p++;
1113 fprintf (stderr, "/%d", mcnt);
1114 break;
1115 # endif /* emacs */
1116
1117 case begbuf:
1118 fprintf (stderr, "/begbuf");
1119 break;
1120
1121 case endbuf:
1122 fprintf (stderr, "/endbuf");
1123 break;
1124
1125 default:
1126 fprintf (stderr, "?%d", *(p-1));
1127 }
1128
1129 fprintf (stderr, "\n");
1130 }
1131
1132 fprintf (stderr, "%td:\tend of pattern.\n", p - start);
1133 }
1134
1135
1136 static void
1137 print_compiled_pattern (struct re_pattern_buffer *bufp)
1138 {
1139 re_char *buffer = bufp->buffer;
1140
1141 print_partial_compiled_pattern (buffer, buffer + bufp->used);
1142 printf ("%ld bytes used/%ld bytes allocated.\n",
1143 bufp->used, bufp->allocated);
1144
1145 if (bufp->fastmap_accurate && bufp->fastmap)
1146 {
1147 printf ("fastmap: ");
1148 print_fastmap (bufp->fastmap);
1149 }
1150
1151 printf ("re_nsub: %zu\t", bufp->re_nsub);
1152 printf ("regs_alloc: %d\t", bufp->regs_allocated);
1153 printf ("can_be_null: %d\t", bufp->can_be_null);
1154 printf ("no_sub: %d\t", bufp->no_sub);
1155 printf ("not_bol: %d\t", bufp->not_bol);
1156 printf ("not_eol: %d\t", bufp->not_eol);
1157 printf ("syntax: %lx\n", bufp->syntax);
1158 fflush (stdout);
1159 /* Perhaps we should print the translate table? */
1160 }
1161
1162
1163 static void
1164 print_double_string (re_char *where, re_char *string1, ssize_t size1,
1165 re_char *string2, ssize_t size2)
1166 {
1167 ssize_t this_char;
1168
1169 if (where == NULL)
1170 printf ("(null)");
1171 else
1172 {
1173 if (FIRST_STRING_P (where))
1174 {
1175 for (this_char = where - string1; this_char < size1; this_char++)
1176 putchar (string1[this_char]);
1177
1178 where = string2;
1179 }
1180
1181 for (this_char = where - string2; this_char < size2; this_char++)
1182 putchar (string2[this_char]);
1183 }
1184 }
1185
1186 #else /* not DEBUG */
1187
1188 # undef assert
1189 # define assert(e)
1190
1191 # define DEBUG_STATEMENT(e)
1192 # if __STDC_VERSION__ < 199901L
1193 # define DEBUG_COMPILES_ARGUMENTS
1194 # define DEBUG_PRINT /* 'DEBUG_PRINT (x, y)' discards X and Y. */ (void)
1195 # else
1196 # define DEBUG_PRINT(...)
1197 # endif
1198 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e)
1199 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2)
1200
1201 #endif /* not DEBUG */
1202 \f
1203 /* Use this to suppress gcc's `...may be used before initialized' warnings. */
1204 #ifdef lint
1205 # define IF_LINT(Code) Code
1206 #else
1207 # define IF_LINT(Code) /* empty */
1208 #endif
1209 \f
1210 /* Set by `re_set_syntax' to the current regexp syntax to recognize. Can
1211 also be assigned to arbitrarily: each pattern buffer stores its own
1212 syntax, so it can be changed between regex compilations. */
1213 /* This has no initializer because initialized variables in Emacs
1214 become read-only after dumping. */
1215 reg_syntax_t re_syntax_options;
1216
1217
1218 /* Specify the precise syntax of regexps for compilation. This provides
1219 for compatibility for various utilities which historically have
1220 different, incompatible syntaxes.
1221
1222 The argument SYNTAX is a bit mask comprised of the various bits
1223 defined in regex.h. We return the old syntax. */
1224
1225 reg_syntax_t
1226 re_set_syntax (reg_syntax_t syntax)
1227 {
1228 reg_syntax_t ret = re_syntax_options;
1229
1230 re_syntax_options = syntax;
1231 return ret;
1232 }
1233 WEAK_ALIAS (__re_set_syntax, re_set_syntax)
1234
1235 /* Regexp to use to replace spaces, or NULL meaning don't. */
1236 static const_re_char *whitespace_regexp;
1237
1238 void
1239 re_set_whitespace_regexp (const char *regexp)
1240 {
1241 whitespace_regexp = (const_re_char *) regexp;
1242 }
1243 WEAK_ALIAS (__re_set_syntax, re_set_syntax)
1244 \f
1245 /* This table gives an error message for each of the error codes listed
1246 in regex.h. Obviously the order here has to be same as there.
1247 POSIX doesn't require that we do anything for REG_NOERROR,
1248 but why not be nice? */
1249
1250 static const char *re_error_msgid[] =
1251 {
1252 gettext_noop ("Success"), /* REG_NOERROR */
1253 gettext_noop ("No match"), /* REG_NOMATCH */
1254 gettext_noop ("Invalid regular expression"), /* REG_BADPAT */
1255 gettext_noop ("Invalid collation character"), /* REG_ECOLLATE */
1256 gettext_noop ("Invalid character class name"), /* REG_ECTYPE */
1257 gettext_noop ("Trailing backslash"), /* REG_EESCAPE */
1258 gettext_noop ("Invalid back reference"), /* REG_ESUBREG */
1259 gettext_noop ("Unmatched [ or [^"), /* REG_EBRACK */
1260 gettext_noop ("Unmatched ( or \\("), /* REG_EPAREN */
1261 gettext_noop ("Unmatched \\{"), /* REG_EBRACE */
1262 gettext_noop ("Invalid content of \\{\\}"), /* REG_BADBR */
1263 gettext_noop ("Invalid range end"), /* REG_ERANGE */
1264 gettext_noop ("Memory exhausted"), /* REG_ESPACE */
1265 gettext_noop ("Invalid preceding regular expression"), /* REG_BADRPT */
1266 gettext_noop ("Premature end of regular expression"), /* REG_EEND */
1267 gettext_noop ("Regular expression too big"), /* REG_ESIZE */
1268 gettext_noop ("Unmatched ) or \\)"), /* REG_ERPAREN */
1269 gettext_noop ("Range striding over charsets") /* REG_ERANGEX */
1270 };
1271 \f
1272 /* Avoiding alloca during matching, to placate r_alloc. */
1273
1274 /* Define MATCH_MAY_ALLOCATE unless we need to make sure that the
1275 searching and matching functions should not call alloca. On some
1276 systems, alloca is implemented in terms of malloc, and if we're
1277 using the relocating allocator routines, then malloc could cause a
1278 relocation, which might (if the strings being searched are in the
1279 ralloc heap) shift the data out from underneath the regexp
1280 routines.
1281
1282 Here's another reason to avoid allocation: Emacs
1283 processes input from X in a signal handler; processing X input may
1284 call malloc; if input arrives while a matching routine is calling
1285 malloc, then we're scrod. But Emacs can't just block input while
1286 calling matching routines; then we don't notice interrupts when
1287 they come in. So, Emacs blocks input around all regexp calls
1288 except the matching calls, which it leaves unprotected, in the
1289 faith that they will not malloc. */
1290
1291 /* Normally, this is fine. */
1292 #define MATCH_MAY_ALLOCATE
1293
1294 /* The match routines may not allocate if (1) they would do it with malloc
1295 and (2) it's not safe for them to use malloc.
1296 Note that if REL_ALLOC is defined, matching would not use malloc for the
1297 failure stack, but we would still use it for the register vectors;
1298 so REL_ALLOC should not affect this. */
1299 #if defined REGEX_MALLOC && defined emacs
1300 # undef MATCH_MAY_ALLOCATE
1301 #endif
1302
1303 \f
1304 /* Failure stack declarations and macros; both re_compile_fastmap and
1305 re_match_2 use a failure stack. These have to be macros because of
1306 REGEX_ALLOCATE_STACK. */
1307
1308
1309 /* Approximate number of failure points for which to initially allocate space
1310 when matching. If this number is exceeded, we allocate more
1311 space, so it is not a hard limit. */
1312 #ifndef INIT_FAILURE_ALLOC
1313 # define INIT_FAILURE_ALLOC 20
1314 #endif
1315
1316 /* Roughly the maximum number of failure points on the stack. Would be
1317 exactly that if always used TYPICAL_FAILURE_SIZE items each time we failed.
1318 This is a variable only so users of regex can assign to it; we never
1319 change it ourselves. We always multiply it by TYPICAL_FAILURE_SIZE
1320 before using it, so it should probably be a byte-count instead. */
1321 # if defined MATCH_MAY_ALLOCATE
1322 /* Note that 4400 was enough to cause a crash on Alpha OSF/1,
1323 whose default stack limit is 2mb. In order for a larger
1324 value to work reliably, you have to try to make it accord
1325 with the process stack limit. */
1326 size_t re_max_failures = 40000;
1327 # else
1328 size_t re_max_failures = 4000;
1329 # endif
1330
1331 union fail_stack_elt
1332 {
1333 re_char *pointer;
1334 /* This should be the biggest `int' that's no bigger than a pointer. */
1335 long integer;
1336 };
1337
1338 typedef union fail_stack_elt fail_stack_elt_t;
1339
1340 typedef struct
1341 {
1342 fail_stack_elt_t *stack;
1343 size_t size;
1344 size_t avail; /* Offset of next open position. */
1345 size_t frame; /* Offset of the cur constructed frame. */
1346 } fail_stack_type;
1347
1348 #define FAIL_STACK_EMPTY() (fail_stack.frame == 0)
1349
1350
1351 /* Define macros to initialize and free the failure stack.
1352 Do `return -2' if the alloc fails. */
1353
1354 #ifdef MATCH_MAY_ALLOCATE
1355 # define INIT_FAIL_STACK() \
1356 do { \
1357 fail_stack.stack = \
1358 REGEX_ALLOCATE_STACK (INIT_FAILURE_ALLOC * TYPICAL_FAILURE_SIZE \
1359 * sizeof (fail_stack_elt_t)); \
1360 \
1361 if (fail_stack.stack == NULL) \
1362 return -2; \
1363 \
1364 fail_stack.size = INIT_FAILURE_ALLOC; \
1365 fail_stack.avail = 0; \
1366 fail_stack.frame = 0; \
1367 } while (0)
1368 #else
1369 # define INIT_FAIL_STACK() \
1370 do { \
1371 fail_stack.avail = 0; \
1372 fail_stack.frame = 0; \
1373 } while (0)
1374
1375 # define RETALLOC_IF(addr, n, t) \
1376 if (addr) RETALLOC((addr), (n), t); else (addr) = TALLOC ((n), t)
1377 #endif
1378
1379
1380 /* Double the size of FAIL_STACK, up to a limit
1381 which allows approximately `re_max_failures' items.
1382
1383 Return 1 if succeeds, and 0 if either ran out of memory
1384 allocating space for it or it was already too large.
1385
1386 REGEX_REALLOCATE_STACK requires `destination' be declared. */
1387
1388 /* Factor to increase the failure stack size by
1389 when we increase it.
1390 This used to be 2, but 2 was too wasteful
1391 because the old discarded stacks added up to as much space
1392 were as ultimate, maximum-size stack. */
1393 #define FAIL_STACK_GROWTH_FACTOR 4
1394
1395 #define GROW_FAIL_STACK(fail_stack) \
1396 (((fail_stack).size * sizeof (fail_stack_elt_t) \
1397 >= re_max_failures * TYPICAL_FAILURE_SIZE) \
1398 ? 0 \
1399 : ((fail_stack).stack \
1400 = REGEX_REALLOCATE_STACK ((fail_stack).stack, \
1401 (fail_stack).size * sizeof (fail_stack_elt_t), \
1402 MIN (re_max_failures * TYPICAL_FAILURE_SIZE, \
1403 ((fail_stack).size * sizeof (fail_stack_elt_t) \
1404 * FAIL_STACK_GROWTH_FACTOR))), \
1405 \
1406 (fail_stack).stack == NULL \
1407 ? 0 \
1408 : ((fail_stack).size \
1409 = (MIN (re_max_failures * TYPICAL_FAILURE_SIZE, \
1410 ((fail_stack).size * sizeof (fail_stack_elt_t) \
1411 * FAIL_STACK_GROWTH_FACTOR)) \
1412 / sizeof (fail_stack_elt_t)), \
1413 1)))
1414
1415
1416 /* Push a pointer value onto the failure stack.
1417 Assumes the variable `fail_stack'. Probably should only
1418 be called from within `PUSH_FAILURE_POINT'. */
1419 #define PUSH_FAILURE_POINTER(item) \
1420 fail_stack.stack[fail_stack.avail++].pointer = (item)
1421
1422 /* This pushes an integer-valued item onto the failure stack.
1423 Assumes the variable `fail_stack'. Probably should only
1424 be called from within `PUSH_FAILURE_POINT'. */
1425 #define PUSH_FAILURE_INT(item) \
1426 fail_stack.stack[fail_stack.avail++].integer = (item)
1427
1428 /* These POP... operations complement the PUSH... operations.
1429 All assume that `fail_stack' is nonempty. */
1430 #define POP_FAILURE_POINTER() fail_stack.stack[--fail_stack.avail].pointer
1431 #define POP_FAILURE_INT() fail_stack.stack[--fail_stack.avail].integer
1432
1433 /* Individual items aside from the registers. */
1434 #define NUM_NONREG_ITEMS 3
1435
1436 /* Used to examine the stack (to detect infinite loops). */
1437 #define FAILURE_PAT(h) fail_stack.stack[(h) - 1].pointer
1438 #define FAILURE_STR(h) (fail_stack.stack[(h) - 2].pointer)
1439 #define NEXT_FAILURE_HANDLE(h) fail_stack.stack[(h) - 3].integer
1440 #define TOP_FAILURE_HANDLE() fail_stack.frame
1441
1442
1443 #define ENSURE_FAIL_STACK(space) \
1444 while (REMAINING_AVAIL_SLOTS <= space) { \
1445 if (!GROW_FAIL_STACK (fail_stack)) \
1446 return -2; \
1447 DEBUG_PRINT ("\n Doubled stack; size now: %zd\n", (fail_stack).size);\
1448 DEBUG_PRINT (" slots available: %zd\n", REMAINING_AVAIL_SLOTS);\
1449 }
1450
1451 /* Push register NUM onto the stack. */
1452 #define PUSH_FAILURE_REG(num) \
1453 do { \
1454 char *destination; \
1455 long n = num; \
1456 ENSURE_FAIL_STACK(3); \
1457 DEBUG_PRINT (" Push reg %ld (spanning %p -> %p)\n", \
1458 n, regstart[n], regend[n]); \
1459 PUSH_FAILURE_POINTER (regstart[n]); \
1460 PUSH_FAILURE_POINTER (regend[n]); \
1461 PUSH_FAILURE_INT (n); \
1462 } while (0)
1463
1464 /* Change the counter's value to VAL, but make sure that it will
1465 be reset when backtracking. */
1466 #define PUSH_NUMBER(ptr,val) \
1467 do { \
1468 char *destination; \
1469 int c; \
1470 ENSURE_FAIL_STACK(3); \
1471 EXTRACT_NUMBER (c, ptr); \
1472 DEBUG_PRINT (" Push number %p = %d -> %d\n", ptr, c, val); \
1473 PUSH_FAILURE_INT (c); \
1474 PUSH_FAILURE_POINTER (ptr); \
1475 PUSH_FAILURE_INT (-1); \
1476 STORE_NUMBER (ptr, val); \
1477 } while (0)
1478
1479 /* Pop a saved register off the stack. */
1480 #define POP_FAILURE_REG_OR_COUNT() \
1481 do { \
1482 long pfreg = POP_FAILURE_INT (); \
1483 if (pfreg == -1) \
1484 { \
1485 /* It's a counter. */ \
1486 /* Here, we discard `const', making re_match non-reentrant. */ \
1487 unsigned char *ptr = (unsigned char*) POP_FAILURE_POINTER (); \
1488 pfreg = POP_FAILURE_INT (); \
1489 STORE_NUMBER (ptr, pfreg); \
1490 DEBUG_PRINT (" Pop counter %p = %ld\n", ptr, pfreg); \
1491 } \
1492 else \
1493 { \
1494 regend[pfreg] = POP_FAILURE_POINTER (); \
1495 regstart[pfreg] = POP_FAILURE_POINTER (); \
1496 DEBUG_PRINT (" Pop reg %ld (spanning %p -> %p)\n", \
1497 pfreg, regstart[pfreg], regend[pfreg]); \
1498 } \
1499 } while (0)
1500
1501 /* Check that we are not stuck in an infinite loop. */
1502 #define CHECK_INFINITE_LOOP(pat_cur, string_place) \
1503 do { \
1504 ssize_t failure = TOP_FAILURE_HANDLE (); \
1505 /* Check for infinite matching loops */ \
1506 while (failure > 0 \
1507 && (FAILURE_STR (failure) == string_place \
1508 || FAILURE_STR (failure) == NULL)) \
1509 { \
1510 assert (FAILURE_PAT (failure) >= bufp->buffer \
1511 && FAILURE_PAT (failure) <= bufp->buffer + bufp->used); \
1512 if (FAILURE_PAT (failure) == pat_cur) \
1513 { \
1514 cycle = 1; \
1515 break; \
1516 } \
1517 DEBUG_PRINT (" Other pattern: %p\n", FAILURE_PAT (failure)); \
1518 failure = NEXT_FAILURE_HANDLE(failure); \
1519 } \
1520 DEBUG_PRINT (" Other string: %p\n", FAILURE_STR (failure)); \
1521 } while (0)
1522
1523 /* Push the information about the state we will need
1524 if we ever fail back to it.
1525
1526 Requires variables fail_stack, regstart, regend and
1527 num_regs be declared. GROW_FAIL_STACK requires `destination' be
1528 declared.
1529
1530 Does `return FAILURE_CODE' if runs out of memory. */
1531
1532 #define PUSH_FAILURE_POINT(pattern, string_place) \
1533 do { \
1534 char *destination; \
1535 /* Must be int, so when we don't save any registers, the arithmetic \
1536 of 0 + -1 isn't done as unsigned. */ \
1537 \
1538 DEBUG_STATEMENT (nfailure_points_pushed++); \
1539 DEBUG_PRINT ("\nPUSH_FAILURE_POINT:\n"); \
1540 DEBUG_PRINT (" Before push, next avail: %zd\n", (fail_stack).avail); \
1541 DEBUG_PRINT (" size: %zd\n", (fail_stack).size);\
1542 \
1543 ENSURE_FAIL_STACK (NUM_NONREG_ITEMS); \
1544 \
1545 DEBUG_PRINT ("\n"); \
1546 \
1547 DEBUG_PRINT (" Push frame index: %zd\n", fail_stack.frame); \
1548 PUSH_FAILURE_INT (fail_stack.frame); \
1549 \
1550 DEBUG_PRINT (" Push string %p: `", string_place); \
1551 DEBUG_PRINT_DOUBLE_STRING (string_place, string1, size1, string2, size2);\
1552 DEBUG_PRINT ("'\n"); \
1553 PUSH_FAILURE_POINTER (string_place); \
1554 \
1555 DEBUG_PRINT (" Push pattern %p: ", pattern); \
1556 DEBUG_PRINT_COMPILED_PATTERN (bufp, pattern, pend); \
1557 PUSH_FAILURE_POINTER (pattern); \
1558 \
1559 /* Close the frame by moving the frame pointer past it. */ \
1560 fail_stack.frame = fail_stack.avail; \
1561 } while (0)
1562
1563 /* Estimate the size of data pushed by a typical failure stack entry.
1564 An estimate is all we need, because all we use this for
1565 is to choose a limit for how big to make the failure stack. */
1566 /* BEWARE, the value `20' is hard-coded in emacs.c:main(). */
1567 #define TYPICAL_FAILURE_SIZE 20
1568
1569 /* How many items can still be added to the stack without overflowing it. */
1570 #define REMAINING_AVAIL_SLOTS ((fail_stack).size - (fail_stack).avail)
1571
1572
1573 /* Pops what PUSH_FAIL_STACK pushes.
1574
1575 We restore into the parameters, all of which should be lvalues:
1576 STR -- the saved data position.
1577 PAT -- the saved pattern position.
1578 REGSTART, REGEND -- arrays of string positions.
1579
1580 Also assumes the variables `fail_stack' and (if debugging), `bufp',
1581 `pend', `string1', `size1', `string2', and `size2'. */
1582
1583 #define POP_FAILURE_POINT(str, pat) \
1584 do { \
1585 assert (!FAIL_STACK_EMPTY ()); \
1586 \
1587 /* Remove failure points and point to how many regs pushed. */ \
1588 DEBUG_PRINT ("POP_FAILURE_POINT:\n"); \
1589 DEBUG_PRINT (" Before pop, next avail: %zd\n", fail_stack.avail); \
1590 DEBUG_PRINT (" size: %zd\n", fail_stack.size); \
1591 \
1592 /* Pop the saved registers. */ \
1593 while (fail_stack.frame < fail_stack.avail) \
1594 POP_FAILURE_REG_OR_COUNT (); \
1595 \
1596 pat = POP_FAILURE_POINTER (); \
1597 DEBUG_PRINT (" Popping pattern %p: ", pat); \
1598 DEBUG_PRINT_COMPILED_PATTERN (bufp, pat, pend); \
1599 \
1600 /* If the saved string location is NULL, it came from an \
1601 on_failure_keep_string_jump opcode, and we want to throw away the \
1602 saved NULL, thus retaining our current position in the string. */ \
1603 str = POP_FAILURE_POINTER (); \
1604 DEBUG_PRINT (" Popping string %p: `", str); \
1605 DEBUG_PRINT_DOUBLE_STRING (str, string1, size1, string2, size2); \
1606 DEBUG_PRINT ("'\n"); \
1607 \
1608 fail_stack.frame = POP_FAILURE_INT (); \
1609 DEBUG_PRINT (" Popping frame index: %zd\n", fail_stack.frame); \
1610 \
1611 assert (fail_stack.avail >= 0); \
1612 assert (fail_stack.frame <= fail_stack.avail); \
1613 \
1614 DEBUG_STATEMENT (nfailure_points_popped++); \
1615 } while (0) /* POP_FAILURE_POINT */
1616
1617
1618 \f
1619 /* Registers are set to a sentinel when they haven't yet matched. */
1620 #define REG_UNSET(e) ((e) == NULL)
1621 \f
1622 /* Subroutine declarations and macros for regex_compile. */
1623
1624 static reg_errcode_t regex_compile (re_char *pattern, size_t size,
1625 reg_syntax_t syntax,
1626 struct re_pattern_buffer *bufp);
1627 static void store_op1 (re_opcode_t op, unsigned char *loc, int arg);
1628 static void store_op2 (re_opcode_t op, unsigned char *loc, int arg1, int arg2);
1629 static void insert_op1 (re_opcode_t op, unsigned char *loc,
1630 int arg, unsigned char *end);
1631 static void insert_op2 (re_opcode_t op, unsigned char *loc,
1632 int arg1, int arg2, unsigned char *end);
1633 static boolean at_begline_loc_p (re_char *pattern, re_char *p,
1634 reg_syntax_t syntax);
1635 static boolean at_endline_loc_p (re_char *p, re_char *pend,
1636 reg_syntax_t syntax);
1637 static re_char *skip_one_char (re_char *p);
1638 static int analyse_first (re_char *p, re_char *pend,
1639 char *fastmap, const int multibyte);
1640
1641 /* Fetch the next character in the uncompiled pattern, with no
1642 translation. */
1643 #define PATFETCH(c) \
1644 do { \
1645 int len; \
1646 if (p == pend) return REG_EEND; \
1647 c = RE_STRING_CHAR_AND_LENGTH (p, len, multibyte); \
1648 p += len; \
1649 } while (0)
1650
1651
1652 /* If `translate' is non-null, return translate[D], else just D. We
1653 cast the subscript to translate because some data is declared as
1654 `char *', to avoid warnings when a string constant is passed. But
1655 when we use a character as a subscript we must make it unsigned. */
1656 #ifndef TRANSLATE
1657 # define TRANSLATE(d) \
1658 (RE_TRANSLATE_P (translate) ? RE_TRANSLATE (translate, (d)) : (d))
1659 #endif
1660
1661
1662 /* Macros for outputting the compiled pattern into `buffer'. */
1663
1664 /* If the buffer isn't allocated when it comes in, use this. */
1665 #define INIT_BUF_SIZE 32
1666
1667 /* Make sure we have at least N more bytes of space in buffer. */
1668 #define GET_BUFFER_SPACE(n) \
1669 while ((size_t) (b - bufp->buffer + (n)) > bufp->allocated) \
1670 EXTEND_BUFFER ()
1671
1672 /* Make sure we have one more byte of buffer space and then add C to it. */
1673 #define BUF_PUSH(c) \
1674 do { \
1675 GET_BUFFER_SPACE (1); \
1676 *b++ = (unsigned char) (c); \
1677 } while (0)
1678
1679
1680 /* Ensure we have two more bytes of buffer space and then append C1 and C2. */
1681 #define BUF_PUSH_2(c1, c2) \
1682 do { \
1683 GET_BUFFER_SPACE (2); \
1684 *b++ = (unsigned char) (c1); \
1685 *b++ = (unsigned char) (c2); \
1686 } while (0)
1687
1688
1689 /* Store a jump with opcode OP at LOC to location TO. We store a
1690 relative address offset by the three bytes the jump itself occupies. */
1691 #define STORE_JUMP(op, loc, to) \
1692 store_op1 (op, loc, (to) - (loc) - 3)
1693
1694 /* Likewise, for a two-argument jump. */
1695 #define STORE_JUMP2(op, loc, to, arg) \
1696 store_op2 (op, loc, (to) - (loc) - 3, arg)
1697
1698 /* Like `STORE_JUMP', but for inserting. Assume `b' is the buffer end. */
1699 #define INSERT_JUMP(op, loc, to) \
1700 insert_op1 (op, loc, (to) - (loc) - 3, b)
1701
1702 /* Like `STORE_JUMP2', but for inserting. Assume `b' is the buffer end. */
1703 #define INSERT_JUMP2(op, loc, to, arg) \
1704 insert_op2 (op, loc, (to) - (loc) - 3, arg, b)
1705
1706
1707 /* This is not an arbitrary limit: the arguments which represent offsets
1708 into the pattern are two bytes long. So if 2^15 bytes turns out to
1709 be too small, many things would have to change. */
1710 # define MAX_BUF_SIZE (1L << 15)
1711
1712 /* Extend the buffer by twice its current size via realloc and
1713 reset the pointers that pointed into the old block to point to the
1714 correct places in the new one. If extending the buffer results in it
1715 being larger than MAX_BUF_SIZE, then flag memory exhausted. */
1716 #if __BOUNDED_POINTERS__
1717 # define SET_HIGH_BOUND(P) (__ptrhigh (P) = __ptrlow (P) + bufp->allocated)
1718 # define MOVE_BUFFER_POINTER(P) \
1719 (__ptrlow (P) = new_buffer + (__ptrlow (P) - old_buffer), \
1720 SET_HIGH_BOUND (P), \
1721 __ptrvalue (P) = new_buffer + (__ptrvalue (P) - old_buffer))
1722 # define ELSE_EXTEND_BUFFER_HIGH_BOUND \
1723 else \
1724 { \
1725 SET_HIGH_BOUND (b); \
1726 SET_HIGH_BOUND (begalt); \
1727 if (fixup_alt_jump) \
1728 SET_HIGH_BOUND (fixup_alt_jump); \
1729 if (laststart) \
1730 SET_HIGH_BOUND (laststart); \
1731 if (pending_exact) \
1732 SET_HIGH_BOUND (pending_exact); \
1733 }
1734 #else
1735 # define MOVE_BUFFER_POINTER(P) ((P) = new_buffer + ((P) - old_buffer))
1736 # define ELSE_EXTEND_BUFFER_HIGH_BOUND
1737 #endif
1738 #define EXTEND_BUFFER() \
1739 do { \
1740 unsigned char *old_buffer = bufp->buffer; \
1741 if (bufp->allocated == MAX_BUF_SIZE) \
1742 return REG_ESIZE; \
1743 bufp->allocated <<= 1; \
1744 if (bufp->allocated > MAX_BUF_SIZE) \
1745 bufp->allocated = MAX_BUF_SIZE; \
1746 RETALLOC (bufp->buffer, bufp->allocated, unsigned char); \
1747 if (bufp->buffer == NULL) \
1748 return REG_ESPACE; \
1749 /* If the buffer moved, move all the pointers into it. */ \
1750 if (old_buffer != bufp->buffer) \
1751 { \
1752 unsigned char *new_buffer = bufp->buffer; \
1753 MOVE_BUFFER_POINTER (b); \
1754 MOVE_BUFFER_POINTER (begalt); \
1755 if (fixup_alt_jump) \
1756 MOVE_BUFFER_POINTER (fixup_alt_jump); \
1757 if (laststart) \
1758 MOVE_BUFFER_POINTER (laststart); \
1759 if (pending_exact) \
1760 MOVE_BUFFER_POINTER (pending_exact); \
1761 } \
1762 ELSE_EXTEND_BUFFER_HIGH_BOUND \
1763 } while (0)
1764
1765
1766 /* Since we have one byte reserved for the register number argument to
1767 {start,stop}_memory, the maximum number of groups we can report
1768 things about is what fits in that byte. */
1769 #define MAX_REGNUM 255
1770
1771 /* But patterns can have more than `MAX_REGNUM' registers. We just
1772 ignore the excess. */
1773 typedef int regnum_t;
1774
1775
1776 /* Macros for the compile stack. */
1777
1778 /* Since offsets can go either forwards or backwards, this type needs to
1779 be able to hold values from -(MAX_BUF_SIZE - 1) to MAX_BUF_SIZE - 1. */
1780 /* int may be not enough when sizeof(int) == 2. */
1781 typedef long pattern_offset_t;
1782
1783 typedef struct
1784 {
1785 pattern_offset_t begalt_offset;
1786 pattern_offset_t fixup_alt_jump;
1787 pattern_offset_t laststart_offset;
1788 regnum_t regnum;
1789 } compile_stack_elt_t;
1790
1791
1792 typedef struct
1793 {
1794 compile_stack_elt_t *stack;
1795 size_t size;
1796 size_t avail; /* Offset of next open position. */
1797 } compile_stack_type;
1798
1799
1800 #define INIT_COMPILE_STACK_SIZE 32
1801
1802 #define COMPILE_STACK_EMPTY (compile_stack.avail == 0)
1803 #define COMPILE_STACK_FULL (compile_stack.avail == compile_stack.size)
1804
1805 /* The next available element. */
1806 #define COMPILE_STACK_TOP (compile_stack.stack[compile_stack.avail])
1807
1808 /* Explicit quit checking is needed for Emacs, which uses polling to
1809 process input events. */
1810 #ifdef emacs
1811 # define IMMEDIATE_QUIT_CHECK \
1812 do { \
1813 if (immediate_quit) QUIT; \
1814 } while (0)
1815 #else
1816 # define IMMEDIATE_QUIT_CHECK ((void)0)
1817 #endif
1818 \f
1819 /* Structure to manage work area for range table. */
1820 struct range_table_work_area
1821 {
1822 int *table; /* actual work area. */
1823 int allocated; /* allocated size for work area in bytes. */
1824 int used; /* actually used size in words. */
1825 int bits; /* flag to record character classes */
1826 };
1827
1828 #ifdef emacs
1829
1830 /* Make sure that WORK_AREA can hold more N multibyte characters.
1831 This is used only in set_image_of_range and set_image_of_range_1.
1832 It expects WORK_AREA to be a pointer.
1833 If it can't get the space, it returns from the surrounding function. */
1834
1835 #define EXTEND_RANGE_TABLE(work_area, n) \
1836 do { \
1837 if (((work_area).used + (n)) * sizeof (int) > (work_area).allocated) \
1838 { \
1839 extend_range_table_work_area (&work_area); \
1840 if ((work_area).table == 0) \
1841 return (REG_ESPACE); \
1842 } \
1843 } while (0)
1844
1845 #define SET_RANGE_TABLE_WORK_AREA_BIT(work_area, bit) \
1846 (work_area).bits |= (bit)
1847
1848 /* Set a range (RANGE_START, RANGE_END) to WORK_AREA. */
1849 #define SET_RANGE_TABLE_WORK_AREA(work_area, range_start, range_end) \
1850 do { \
1851 EXTEND_RANGE_TABLE ((work_area), 2); \
1852 (work_area).table[(work_area).used++] = (range_start); \
1853 (work_area).table[(work_area).used++] = (range_end); \
1854 } while (0)
1855
1856 #endif /* emacs */
1857
1858 /* Free allocated memory for WORK_AREA. */
1859 #define FREE_RANGE_TABLE_WORK_AREA(work_area) \
1860 do { \
1861 if ((work_area).table) \
1862 free ((work_area).table); \
1863 } while (0)
1864
1865 #define CLEAR_RANGE_TABLE_WORK_USED(work_area) ((work_area).used = 0, (work_area).bits = 0)
1866 #define RANGE_TABLE_WORK_USED(work_area) ((work_area).used)
1867 #define RANGE_TABLE_WORK_BITS(work_area) ((work_area).bits)
1868 #define RANGE_TABLE_WORK_ELT(work_area, i) ((work_area).table[i])
1869
1870 /* Bits used to implement the multibyte-part of the various character classes
1871 such as [:alnum:] in a charset's range table. */
1872 #define BIT_WORD 0x1
1873 #define BIT_LOWER 0x2
1874 #define BIT_PUNCT 0x4
1875 #define BIT_SPACE 0x8
1876 #define BIT_UPPER 0x10
1877 #define BIT_MULTIBYTE 0x20
1878 \f
1879
1880 /* Set the bit for character C in a list. */
1881 #define SET_LIST_BIT(c) (b[((c)) / BYTEWIDTH] |= 1 << ((c) % BYTEWIDTH))
1882
1883
1884 #ifdef emacs
1885
1886 /* Store characters in the range FROM to TO in the bitmap at B (for
1887 ASCII and unibyte characters) and WORK_AREA (for multibyte
1888 characters) while translating them and paying attention to the
1889 continuity of translated characters.
1890
1891 Implementation note: It is better to implement these fairly big
1892 macros by a function, but it's not that easy because macros called
1893 in this macro assume various local variables already declared. */
1894
1895 /* Both FROM and TO are ASCII characters. */
1896
1897 #define SETUP_ASCII_RANGE(work_area, FROM, TO) \
1898 do { \
1899 int C0, C1; \
1900 \
1901 for (C0 = (FROM); C0 <= (TO); C0++) \
1902 { \
1903 C1 = TRANSLATE (C0); \
1904 if (! ASCII_CHAR_P (C1)) \
1905 { \
1906 SET_RANGE_TABLE_WORK_AREA ((work_area), C1, C1); \
1907 if ((C1 = RE_CHAR_TO_UNIBYTE (C1)) < 0) \
1908 C1 = C0; \
1909 } \
1910 SET_LIST_BIT (C1); \
1911 } \
1912 } while (0)
1913
1914
1915 /* Both FROM and TO are unibyte characters (0x80..0xFF). */
1916
1917 #define SETUP_UNIBYTE_RANGE(work_area, FROM, TO) \
1918 do { \
1919 int C0, C1, C2, I; \
1920 int USED = RANGE_TABLE_WORK_USED (work_area); \
1921 \
1922 for (C0 = (FROM); C0 <= (TO); C0++) \
1923 { \
1924 C1 = RE_CHAR_TO_MULTIBYTE (C0); \
1925 if (CHAR_BYTE8_P (C1)) \
1926 SET_LIST_BIT (C0); \
1927 else \
1928 { \
1929 C2 = TRANSLATE (C1); \
1930 if (C2 == C1 \
1931 || (C1 = RE_CHAR_TO_UNIBYTE (C2)) < 0) \
1932 C1 = C0; \
1933 SET_LIST_BIT (C1); \
1934 for (I = RANGE_TABLE_WORK_USED (work_area) - 2; I >= USED; I -= 2) \
1935 { \
1936 int from = RANGE_TABLE_WORK_ELT (work_area, I); \
1937 int to = RANGE_TABLE_WORK_ELT (work_area, I + 1); \
1938 \
1939 if (C2 >= from - 1 && C2 <= to + 1) \
1940 { \
1941 if (C2 == from - 1) \
1942 RANGE_TABLE_WORK_ELT (work_area, I)--; \
1943 else if (C2 == to + 1) \
1944 RANGE_TABLE_WORK_ELT (work_area, I + 1)++; \
1945 break; \
1946 } \
1947 } \
1948 if (I < USED) \
1949 SET_RANGE_TABLE_WORK_AREA ((work_area), C2, C2); \
1950 } \
1951 } \
1952 } while (0)
1953
1954
1955 /* Both FROM and TO are multibyte characters. */
1956
1957 #define SETUP_MULTIBYTE_RANGE(work_area, FROM, TO) \
1958 do { \
1959 int C0, C1, C2, I, USED = RANGE_TABLE_WORK_USED (work_area); \
1960 \
1961 SET_RANGE_TABLE_WORK_AREA ((work_area), (FROM), (TO)); \
1962 for (C0 = (FROM); C0 <= (TO); C0++) \
1963 { \
1964 C1 = TRANSLATE (C0); \
1965 if ((C2 = RE_CHAR_TO_UNIBYTE (C1)) >= 0 \
1966 || (C1 != C0 && (C2 = RE_CHAR_TO_UNIBYTE (C0)) >= 0)) \
1967 SET_LIST_BIT (C2); \
1968 if (C1 >= (FROM) && C1 <= (TO)) \
1969 continue; \
1970 for (I = RANGE_TABLE_WORK_USED (work_area) - 2; I >= USED; I -= 2) \
1971 { \
1972 int from = RANGE_TABLE_WORK_ELT (work_area, I); \
1973 int to = RANGE_TABLE_WORK_ELT (work_area, I + 1); \
1974 \
1975 if (C1 >= from - 1 && C1 <= to + 1) \
1976 { \
1977 if (C1 == from - 1) \
1978 RANGE_TABLE_WORK_ELT (work_area, I)--; \
1979 else if (C1 == to + 1) \
1980 RANGE_TABLE_WORK_ELT (work_area, I + 1)++; \
1981 break; \
1982 } \
1983 } \
1984 if (I < USED) \
1985 SET_RANGE_TABLE_WORK_AREA ((work_area), C1, C1); \
1986 } \
1987 } while (0)
1988
1989 #endif /* emacs */
1990
1991 /* Get the next unsigned number in the uncompiled pattern. */
1992 #define GET_UNSIGNED_NUMBER(num) \
1993 do { \
1994 if (p == pend) \
1995 FREE_STACK_RETURN (REG_EBRACE); \
1996 else \
1997 { \
1998 PATFETCH (c); \
1999 while ('0' <= c && c <= '9') \
2000 { \
2001 int prev; \
2002 if (num < 0) \
2003 num = 0; \
2004 prev = num; \
2005 num = num * 10 + c - '0'; \
2006 if (num / 10 != prev) \
2007 FREE_STACK_RETURN (REG_BADBR); \
2008 if (p == pend) \
2009 FREE_STACK_RETURN (REG_EBRACE); \
2010 PATFETCH (c); \
2011 } \
2012 } \
2013 } while (0)
2014 \f
2015 #if ! WIDE_CHAR_SUPPORT
2016
2017 /* Map a string to the char class it names (if any). */
2018 re_wctype_t
2019 re_wctype (const_re_char *str)
2020 {
2021 const char *string = (const char *) str;
2022 if (STREQ (string, "alnum")) return RECC_ALNUM;
2023 else if (STREQ (string, "alpha")) return RECC_ALPHA;
2024 else if (STREQ (string, "word")) return RECC_WORD;
2025 else if (STREQ (string, "ascii")) return RECC_ASCII;
2026 else if (STREQ (string, "nonascii")) return RECC_NONASCII;
2027 else if (STREQ (string, "graph")) return RECC_GRAPH;
2028 else if (STREQ (string, "lower")) return RECC_LOWER;
2029 else if (STREQ (string, "print")) return RECC_PRINT;
2030 else if (STREQ (string, "punct")) return RECC_PUNCT;
2031 else if (STREQ (string, "space")) return RECC_SPACE;
2032 else if (STREQ (string, "upper")) return RECC_UPPER;
2033 else if (STREQ (string, "unibyte")) return RECC_UNIBYTE;
2034 else if (STREQ (string, "multibyte")) return RECC_MULTIBYTE;
2035 else if (STREQ (string, "digit")) return RECC_DIGIT;
2036 else if (STREQ (string, "xdigit")) return RECC_XDIGIT;
2037 else if (STREQ (string, "cntrl")) return RECC_CNTRL;
2038 else if (STREQ (string, "blank")) return RECC_BLANK;
2039 else return 0;
2040 }
2041
2042 /* True if CH is in the char class CC. */
2043 boolean
2044 re_iswctype (int ch, re_wctype_t cc)
2045 {
2046 switch (cc)
2047 {
2048 case RECC_ALNUM: return ISALNUM (ch) != 0;
2049 case RECC_ALPHA: return ISALPHA (ch) != 0;
2050 case RECC_BLANK: return ISBLANK (ch) != 0;
2051 case RECC_CNTRL: return ISCNTRL (ch) != 0;
2052 case RECC_DIGIT: return ISDIGIT (ch) != 0;
2053 case RECC_GRAPH: return ISGRAPH (ch) != 0;
2054 case RECC_LOWER: return ISLOWER (ch) != 0;
2055 case RECC_PRINT: return ISPRINT (ch) != 0;
2056 case RECC_PUNCT: return ISPUNCT (ch) != 0;
2057 case RECC_SPACE: return ISSPACE (ch) != 0;
2058 case RECC_UPPER: return ISUPPER (ch) != 0;
2059 case RECC_XDIGIT: return ISXDIGIT (ch) != 0;
2060 case RECC_ASCII: return IS_REAL_ASCII (ch) != 0;
2061 case RECC_NONASCII: return !IS_REAL_ASCII (ch);
2062 case RECC_UNIBYTE: return ISUNIBYTE (ch) != 0;
2063 case RECC_MULTIBYTE: return !ISUNIBYTE (ch);
2064 case RECC_WORD: return ISWORD (ch) != 0;
2065 case RECC_ERROR: return false;
2066 default:
2067 abort ();
2068 }
2069 }
2070
2071 /* Return a bit-pattern to use in the range-table bits to match multibyte
2072 chars of class CC. */
2073 static int
2074 re_wctype_to_bit (re_wctype_t cc)
2075 {
2076 switch (cc)
2077 {
2078 case RECC_NONASCII: case RECC_PRINT: case RECC_GRAPH:
2079 case RECC_MULTIBYTE: return BIT_MULTIBYTE;
2080 case RECC_ALPHA: case RECC_ALNUM: case RECC_WORD: return BIT_WORD;
2081 case RECC_LOWER: return BIT_LOWER;
2082 case RECC_UPPER: return BIT_UPPER;
2083 case RECC_PUNCT: return BIT_PUNCT;
2084 case RECC_SPACE: return BIT_SPACE;
2085 case RECC_ASCII: case RECC_DIGIT: case RECC_XDIGIT: case RECC_CNTRL:
2086 case RECC_BLANK: case RECC_UNIBYTE: case RECC_ERROR: return 0;
2087 default:
2088 abort ();
2089 }
2090 }
2091 #endif
2092 \f
2093 /* Filling in the work area of a range. */
2094
2095 /* Actually extend the space in WORK_AREA. */
2096
2097 static void
2098 extend_range_table_work_area (struct range_table_work_area *work_area)
2099 {
2100 work_area->allocated += 16 * sizeof (int);
2101 work_area->table = realloc (work_area->table, work_area->allocated);
2102 }
2103
2104 #if 0
2105 #ifdef emacs
2106
2107 /* Carefully find the ranges of codes that are equivalent
2108 under case conversion to the range start..end when passed through
2109 TRANSLATE. Handle the case where non-letters can come in between
2110 two upper-case letters (which happens in Latin-1).
2111 Also handle the case of groups of more than 2 case-equivalent chars.
2112
2113 The basic method is to look at consecutive characters and see
2114 if they can form a run that can be handled as one.
2115
2116 Returns -1 if successful, REG_ESPACE if ran out of space. */
2117
2118 static int
2119 set_image_of_range_1 (struct range_table_work_area *work_area,
2120 re_wchar_t start, re_wchar_t end,
2121 RE_TRANSLATE_TYPE translate)
2122 {
2123 /* `one_case' indicates a character, or a run of characters,
2124 each of which is an isolate (no case-equivalents).
2125 This includes all ASCII non-letters.
2126
2127 `two_case' indicates a character, or a run of characters,
2128 each of which has two case-equivalent forms.
2129 This includes all ASCII letters.
2130
2131 `strange' indicates a character that has more than one
2132 case-equivalent. */
2133
2134 enum case_type {one_case, two_case, strange};
2135
2136 /* Describe the run that is in progress,
2137 which the next character can try to extend.
2138 If run_type is strange, that means there really is no run.
2139 If run_type is one_case, then run_start...run_end is the run.
2140 If run_type is two_case, then the run is run_start...run_end,
2141 and the case-equivalents end at run_eqv_end. */
2142
2143 enum case_type run_type = strange;
2144 int run_start, run_end, run_eqv_end;
2145
2146 Lisp_Object eqv_table;
2147
2148 if (!RE_TRANSLATE_P (translate))
2149 {
2150 EXTEND_RANGE_TABLE (work_area, 2);
2151 work_area->table[work_area->used++] = (start);
2152 work_area->table[work_area->used++] = (end);
2153 return -1;
2154 }
2155
2156 eqv_table = XCHAR_TABLE (translate)->extras[2];
2157
2158 for (; start <= end; start++)
2159 {
2160 enum case_type this_type;
2161 int eqv = RE_TRANSLATE (eqv_table, start);
2162 int minchar, maxchar;
2163
2164 /* Classify this character */
2165 if (eqv == start)
2166 this_type = one_case;
2167 else if (RE_TRANSLATE (eqv_table, eqv) == start)
2168 this_type = two_case;
2169 else
2170 this_type = strange;
2171
2172 if (start < eqv)
2173 minchar = start, maxchar = eqv;
2174 else
2175 minchar = eqv, maxchar = start;
2176
2177 /* Can this character extend the run in progress? */
2178 if (this_type == strange || this_type != run_type
2179 || !(minchar == run_end + 1
2180 && (run_type == two_case
2181 ? maxchar == run_eqv_end + 1 : 1)))
2182 {
2183 /* No, end the run.
2184 Record each of its equivalent ranges. */
2185 if (run_type == one_case)
2186 {
2187 EXTEND_RANGE_TABLE (work_area, 2);
2188 work_area->table[work_area->used++] = run_start;
2189 work_area->table[work_area->used++] = run_end;
2190 }
2191 else if (run_type == two_case)
2192 {
2193 EXTEND_RANGE_TABLE (work_area, 4);
2194 work_area->table[work_area->used++] = run_start;
2195 work_area->table[work_area->used++] = run_end;
2196 work_area->table[work_area->used++]
2197 = RE_TRANSLATE (eqv_table, run_start);
2198 work_area->table[work_area->used++]
2199 = RE_TRANSLATE (eqv_table, run_end);
2200 }
2201 run_type = strange;
2202 }
2203
2204 if (this_type == strange)
2205 {
2206 /* For a strange character, add each of its equivalents, one
2207 by one. Don't start a range. */
2208 do
2209 {
2210 EXTEND_RANGE_TABLE (work_area, 2);
2211 work_area->table[work_area->used++] = eqv;
2212 work_area->table[work_area->used++] = eqv;
2213 eqv = RE_TRANSLATE (eqv_table, eqv);
2214 }
2215 while (eqv != start);
2216 }
2217
2218 /* Add this char to the run, or start a new run. */
2219 else if (run_type == strange)
2220 {
2221 /* Initialize a new range. */
2222 run_type = this_type;
2223 run_start = start;
2224 run_end = start;
2225 run_eqv_end = RE_TRANSLATE (eqv_table, run_end);
2226 }
2227 else
2228 {
2229 /* Extend a running range. */
2230 run_end = minchar;
2231 run_eqv_end = RE_TRANSLATE (eqv_table, run_end);
2232 }
2233 }
2234
2235 /* If a run is still in progress at the end, finish it now
2236 by recording its equivalent ranges. */
2237 if (run_type == one_case)
2238 {
2239 EXTEND_RANGE_TABLE (work_area, 2);
2240 work_area->table[work_area->used++] = run_start;
2241 work_area->table[work_area->used++] = run_end;
2242 }
2243 else if (run_type == two_case)
2244 {
2245 EXTEND_RANGE_TABLE (work_area, 4);
2246 work_area->table[work_area->used++] = run_start;
2247 work_area->table[work_area->used++] = run_end;
2248 work_area->table[work_area->used++]
2249 = RE_TRANSLATE (eqv_table, run_start);
2250 work_area->table[work_area->used++]
2251 = RE_TRANSLATE (eqv_table, run_end);
2252 }
2253
2254 return -1;
2255 }
2256
2257 #endif /* emacs */
2258
2259 /* Record the image of the range start..end when passed through
2260 TRANSLATE. This is not necessarily TRANSLATE(start)..TRANSLATE(end)
2261 and is not even necessarily contiguous.
2262 Normally we approximate it with the smallest contiguous range that contains
2263 all the chars we need. However, for Latin-1 we go to extra effort
2264 to do a better job.
2265
2266 This function is not called for ASCII ranges.
2267
2268 Returns -1 if successful, REG_ESPACE if ran out of space. */
2269
2270 static int
2271 set_image_of_range (struct range_table_work_area *work_area,
2272 re_wchar_t start, re_wchar_t end,
2273 RE_TRANSLATE_TYPE translate)
2274 {
2275 re_wchar_t cmin, cmax;
2276
2277 #ifdef emacs
2278 /* For Latin-1 ranges, use set_image_of_range_1
2279 to get proper handling of ranges that include letters and nonletters.
2280 For a range that includes the whole of Latin-1, this is not necessary.
2281 For other character sets, we don't bother to get this right. */
2282 if (RE_TRANSLATE_P (translate) && start < 04400
2283 && !(start < 04200 && end >= 04377))
2284 {
2285 int newend;
2286 int tem;
2287 newend = end;
2288 if (newend > 04377)
2289 newend = 04377;
2290 tem = set_image_of_range_1 (work_area, start, newend, translate);
2291 if (tem > 0)
2292 return tem;
2293
2294 start = 04400;
2295 if (end < 04400)
2296 return -1;
2297 }
2298 #endif
2299
2300 EXTEND_RANGE_TABLE (work_area, 2);
2301 work_area->table[work_area->used++] = (start);
2302 work_area->table[work_area->used++] = (end);
2303
2304 cmin = -1, cmax = -1;
2305
2306 if (RE_TRANSLATE_P (translate))
2307 {
2308 int ch;
2309
2310 for (ch = start; ch <= end; ch++)
2311 {
2312 re_wchar_t c = TRANSLATE (ch);
2313 if (! (start <= c && c <= end))
2314 {
2315 if (cmin == -1)
2316 cmin = c, cmax = c;
2317 else
2318 {
2319 cmin = MIN (cmin, c);
2320 cmax = MAX (cmax, c);
2321 }
2322 }
2323 }
2324
2325 if (cmin != -1)
2326 {
2327 EXTEND_RANGE_TABLE (work_area, 2);
2328 work_area->table[work_area->used++] = (cmin);
2329 work_area->table[work_area->used++] = (cmax);
2330 }
2331 }
2332
2333 return -1;
2334 }
2335 #endif /* 0 */
2336 \f
2337 #ifndef MATCH_MAY_ALLOCATE
2338
2339 /* If we cannot allocate large objects within re_match_2_internal,
2340 we make the fail stack and register vectors global.
2341 The fail stack, we grow to the maximum size when a regexp
2342 is compiled.
2343 The register vectors, we adjust in size each time we
2344 compile a regexp, according to the number of registers it needs. */
2345
2346 static fail_stack_type fail_stack;
2347
2348 /* Size with which the following vectors are currently allocated.
2349 That is so we can make them bigger as needed,
2350 but never make them smaller. */
2351 static int regs_allocated_size;
2352
2353 static re_char ** regstart, ** regend;
2354 static re_char **best_regstart, **best_regend;
2355
2356 /* Make the register vectors big enough for NUM_REGS registers,
2357 but don't make them smaller. */
2358
2359 static
2360 regex_grow_registers (int num_regs)
2361 {
2362 if (num_regs > regs_allocated_size)
2363 {
2364 RETALLOC_IF (regstart, num_regs, re_char *);
2365 RETALLOC_IF (regend, num_regs, re_char *);
2366 RETALLOC_IF (best_regstart, num_regs, re_char *);
2367 RETALLOC_IF (best_regend, num_regs, re_char *);
2368
2369 regs_allocated_size = num_regs;
2370 }
2371 }
2372
2373 #endif /* not MATCH_MAY_ALLOCATE */
2374 \f
2375 static boolean group_in_compile_stack (compile_stack_type compile_stack,
2376 regnum_t regnum);
2377
2378 /* `regex_compile' compiles PATTERN (of length SIZE) according to SYNTAX.
2379 Returns one of error codes defined in `regex.h', or zero for success.
2380
2381 Assumes the `allocated' (and perhaps `buffer') and `translate'
2382 fields are set in BUFP on entry.
2383
2384 If it succeeds, results are put in BUFP (if it returns an error, the
2385 contents of BUFP are undefined):
2386 `buffer' is the compiled pattern;
2387 `syntax' is set to SYNTAX;
2388 `used' is set to the length of the compiled pattern;
2389 `fastmap_accurate' is zero;
2390 `re_nsub' is the number of subexpressions in PATTERN;
2391 `not_bol' and `not_eol' are zero;
2392
2393 The `fastmap' field is neither examined nor set. */
2394
2395 /* Insert the `jump' from the end of last alternative to "here".
2396 The space for the jump has already been allocated. */
2397 #define FIXUP_ALT_JUMP() \
2398 do { \
2399 if (fixup_alt_jump) \
2400 STORE_JUMP (jump, fixup_alt_jump, b); \
2401 } while (0)
2402
2403
2404 /* Return, freeing storage we allocated. */
2405 #define FREE_STACK_RETURN(value) \
2406 do { \
2407 FREE_RANGE_TABLE_WORK_AREA (range_table_work); \
2408 free (compile_stack.stack); \
2409 return value; \
2410 } while (0)
2411
2412 static reg_errcode_t
2413 regex_compile (const_re_char *pattern, size_t size, reg_syntax_t syntax,
2414 struct re_pattern_buffer *bufp)
2415 {
2416 /* We fetch characters from PATTERN here. */
2417 register re_wchar_t c, c1;
2418
2419 /* Points to the end of the buffer, where we should append. */
2420 register unsigned char *b;
2421
2422 /* Keeps track of unclosed groups. */
2423 compile_stack_type compile_stack;
2424
2425 /* Points to the current (ending) position in the pattern. */
2426 #ifdef AIX
2427 /* `const' makes AIX compiler fail. */
2428 unsigned char *p = pattern;
2429 #else
2430 re_char *p = pattern;
2431 #endif
2432 re_char *pend = pattern + size;
2433
2434 /* How to translate the characters in the pattern. */
2435 RE_TRANSLATE_TYPE translate = bufp->translate;
2436
2437 /* Address of the count-byte of the most recently inserted `exactn'
2438 command. This makes it possible to tell if a new exact-match
2439 character can be added to that command or if the character requires
2440 a new `exactn' command. */
2441 unsigned char *pending_exact = 0;
2442
2443 /* Address of start of the most recently finished expression.
2444 This tells, e.g., postfix * where to find the start of its
2445 operand. Reset at the beginning of groups and alternatives. */
2446 unsigned char *laststart = 0;
2447
2448 /* Address of beginning of regexp, or inside of last group. */
2449 unsigned char *begalt;
2450
2451 /* Place in the uncompiled pattern (i.e., the {) to
2452 which to go back if the interval is invalid. */
2453 re_char *beg_interval;
2454
2455 /* Address of the place where a forward jump should go to the end of
2456 the containing expression. Each alternative of an `or' -- except the
2457 last -- ends with a forward jump of this sort. */
2458 unsigned char *fixup_alt_jump = 0;
2459
2460 /* Work area for range table of charset. */
2461 struct range_table_work_area range_table_work;
2462
2463 /* If the object matched can contain multibyte characters. */
2464 const boolean multibyte = RE_MULTIBYTE_P (bufp);
2465
2466 /* Nonzero if we have pushed down into a subpattern. */
2467 int in_subpattern = 0;
2468
2469 /* These hold the values of p, pattern, and pend from the main
2470 pattern when we have pushed into a subpattern. */
2471 re_char *main_p IF_LINT (= NULL);
2472 re_char *main_pattern IF_LINT (= NULL);
2473 re_char *main_pend IF_LINT (= NULL);
2474
2475 #ifdef DEBUG
2476 debug++;
2477 DEBUG_PRINT ("\nCompiling pattern: ");
2478 if (debug > 0)
2479 {
2480 unsigned debug_count;
2481
2482 for (debug_count = 0; debug_count < size; debug_count++)
2483 putchar (pattern[debug_count]);
2484 putchar ('\n');
2485 }
2486 #endif /* DEBUG */
2487
2488 /* Initialize the compile stack. */
2489 compile_stack.stack = TALLOC (INIT_COMPILE_STACK_SIZE, compile_stack_elt_t);
2490 if (compile_stack.stack == NULL)
2491 return REG_ESPACE;
2492
2493 compile_stack.size = INIT_COMPILE_STACK_SIZE;
2494 compile_stack.avail = 0;
2495
2496 range_table_work.table = 0;
2497 range_table_work.allocated = 0;
2498
2499 /* Initialize the pattern buffer. */
2500 bufp->syntax = syntax;
2501 bufp->fastmap_accurate = 0;
2502 bufp->not_bol = bufp->not_eol = 0;
2503 bufp->used_syntax = 0;
2504
2505 /* Set `used' to zero, so that if we return an error, the pattern
2506 printer (for debugging) will think there's no pattern. We reset it
2507 at the end. */
2508 bufp->used = 0;
2509
2510 /* Always count groups, whether or not bufp->no_sub is set. */
2511 bufp->re_nsub = 0;
2512
2513 #if !defined emacs && !defined SYNTAX_TABLE
2514 /* Initialize the syntax table. */
2515 init_syntax_once ();
2516 #endif
2517
2518 if (bufp->allocated == 0)
2519 {
2520 if (bufp->buffer)
2521 { /* If zero allocated, but buffer is non-null, try to realloc
2522 enough space. This loses if buffer's address is bogus, but
2523 that is the user's responsibility. */
2524 RETALLOC (bufp->buffer, INIT_BUF_SIZE, unsigned char);
2525 }
2526 else
2527 { /* Caller did not allocate a buffer. Do it for them. */
2528 bufp->buffer = TALLOC (INIT_BUF_SIZE, unsigned char);
2529 }
2530 if (!bufp->buffer) FREE_STACK_RETURN (REG_ESPACE);
2531
2532 bufp->allocated = INIT_BUF_SIZE;
2533 }
2534
2535 begalt = b = bufp->buffer;
2536
2537 /* Loop through the uncompiled pattern until we're at the end. */
2538 while (1)
2539 {
2540 if (p == pend)
2541 {
2542 /* If this is the end of an included regexp,
2543 pop back to the main regexp and try again. */
2544 if (in_subpattern)
2545 {
2546 in_subpattern = 0;
2547 pattern = main_pattern;
2548 p = main_p;
2549 pend = main_pend;
2550 continue;
2551 }
2552 /* If this is the end of the main regexp, we are done. */
2553 break;
2554 }
2555
2556 PATFETCH (c);
2557
2558 switch (c)
2559 {
2560 case ' ':
2561 {
2562 re_char *p1 = p;
2563
2564 /* If there's no special whitespace regexp, treat
2565 spaces normally. And don't try to do this recursively. */
2566 if (!whitespace_regexp || in_subpattern)
2567 goto normal_char;
2568
2569 /* Peek past following spaces. */
2570 while (p1 != pend)
2571 {
2572 if (*p1 != ' ')
2573 break;
2574 p1++;
2575 }
2576 /* If the spaces are followed by a repetition op,
2577 treat them normally. */
2578 if (p1 != pend
2579 && (*p1 == '*' || *p1 == '+' || *p1 == '?'
2580 || (*p1 == '\\' && p1 + 1 != pend && p1[1] == '{')))
2581 goto normal_char;
2582
2583 /* Replace the spaces with the whitespace regexp. */
2584 in_subpattern = 1;
2585 main_p = p1;
2586 main_pend = pend;
2587 main_pattern = pattern;
2588 p = pattern = whitespace_regexp;
2589 pend = p + strlen ((const char *) p);
2590 break;
2591 }
2592
2593 case '^':
2594 {
2595 if ( /* If at start of pattern, it's an operator. */
2596 p == pattern + 1
2597 /* If context independent, it's an operator. */
2598 || syntax & RE_CONTEXT_INDEP_ANCHORS
2599 /* Otherwise, depends on what's come before. */
2600 || at_begline_loc_p (pattern, p, syntax))
2601 BUF_PUSH ((syntax & RE_NO_NEWLINE_ANCHOR) ? begbuf : begline);
2602 else
2603 goto normal_char;
2604 }
2605 break;
2606
2607
2608 case '$':
2609 {
2610 if ( /* If at end of pattern, it's an operator. */
2611 p == pend
2612 /* If context independent, it's an operator. */
2613 || syntax & RE_CONTEXT_INDEP_ANCHORS
2614 /* Otherwise, depends on what's next. */
2615 || at_endline_loc_p (p, pend, syntax))
2616 BUF_PUSH ((syntax & RE_NO_NEWLINE_ANCHOR) ? endbuf : endline);
2617 else
2618 goto normal_char;
2619 }
2620 break;
2621
2622
2623 case '+':
2624 case '?':
2625 if ((syntax & RE_BK_PLUS_QM)
2626 || (syntax & RE_LIMITED_OPS))
2627 goto normal_char;
2628 handle_plus:
2629 case '*':
2630 /* If there is no previous pattern... */
2631 if (!laststart)
2632 {
2633 if (syntax & RE_CONTEXT_INVALID_OPS)
2634 FREE_STACK_RETURN (REG_BADRPT);
2635 else if (!(syntax & RE_CONTEXT_INDEP_OPS))
2636 goto normal_char;
2637 }
2638
2639 {
2640 /* 1 means zero (many) matches is allowed. */
2641 boolean zero_times_ok = 0, many_times_ok = 0;
2642 boolean greedy = 1;
2643
2644 /* If there is a sequence of repetition chars, collapse it
2645 down to just one (the right one). We can't combine
2646 interval operators with these because of, e.g., `a{2}*',
2647 which should only match an even number of `a's. */
2648
2649 for (;;)
2650 {
2651 if ((syntax & RE_FRUGAL)
2652 && c == '?' && (zero_times_ok || many_times_ok))
2653 greedy = 0;
2654 else
2655 {
2656 zero_times_ok |= c != '+';
2657 many_times_ok |= c != '?';
2658 }
2659
2660 if (p == pend)
2661 break;
2662 else if (*p == '*'
2663 || (!(syntax & RE_BK_PLUS_QM)
2664 && (*p == '+' || *p == '?')))
2665 ;
2666 else if (syntax & RE_BK_PLUS_QM && *p == '\\')
2667 {
2668 if (p+1 == pend)
2669 FREE_STACK_RETURN (REG_EESCAPE);
2670 if (p[1] == '+' || p[1] == '?')
2671 PATFETCH (c); /* Gobble up the backslash. */
2672 else
2673 break;
2674 }
2675 else
2676 break;
2677 /* If we get here, we found another repeat character. */
2678 PATFETCH (c);
2679 }
2680
2681 /* Star, etc. applied to an empty pattern is equivalent
2682 to an empty pattern. */
2683 if (!laststart || laststart == b)
2684 break;
2685
2686 /* Now we know whether or not zero matches is allowed
2687 and also whether or not two or more matches is allowed. */
2688 if (greedy)
2689 {
2690 if (many_times_ok)
2691 {
2692 boolean simple = skip_one_char (laststart) == b;
2693 size_t startoffset = 0;
2694 re_opcode_t ofj =
2695 /* Check if the loop can match the empty string. */
2696 (simple || !analyse_first (laststart, b, NULL, 0))
2697 ? on_failure_jump : on_failure_jump_loop;
2698 assert (skip_one_char (laststart) <= b);
2699
2700 if (!zero_times_ok && simple)
2701 { /* Since simple * loops can be made faster by using
2702 on_failure_keep_string_jump, we turn simple P+
2703 into PP* if P is simple. */
2704 unsigned char *p1, *p2;
2705 startoffset = b - laststart;
2706 GET_BUFFER_SPACE (startoffset);
2707 p1 = b; p2 = laststart;
2708 while (p2 < p1)
2709 *b++ = *p2++;
2710 zero_times_ok = 1;
2711 }
2712
2713 GET_BUFFER_SPACE (6);
2714 if (!zero_times_ok)
2715 /* A + loop. */
2716 STORE_JUMP (ofj, b, b + 6);
2717 else
2718 /* Simple * loops can use on_failure_keep_string_jump
2719 depending on what follows. But since we don't know
2720 that yet, we leave the decision up to
2721 on_failure_jump_smart. */
2722 INSERT_JUMP (simple ? on_failure_jump_smart : ofj,
2723 laststart + startoffset, b + 6);
2724 b += 3;
2725 STORE_JUMP (jump, b, laststart + startoffset);
2726 b += 3;
2727 }
2728 else
2729 {
2730 /* A simple ? pattern. */
2731 assert (zero_times_ok);
2732 GET_BUFFER_SPACE (3);
2733 INSERT_JUMP (on_failure_jump, laststart, b + 3);
2734 b += 3;
2735 }
2736 }
2737 else /* not greedy */
2738 { /* I wish the greedy and non-greedy cases could be merged. */
2739
2740 GET_BUFFER_SPACE (7); /* We might use less. */
2741 if (many_times_ok)
2742 {
2743 boolean emptyp = analyse_first (laststart, b, NULL, 0);
2744
2745 /* The non-greedy multiple match looks like
2746 a repeat..until: we only need a conditional jump
2747 at the end of the loop. */
2748 if (emptyp) BUF_PUSH (no_op);
2749 STORE_JUMP (emptyp ? on_failure_jump_nastyloop
2750 : on_failure_jump, b, laststart);
2751 b += 3;
2752 if (zero_times_ok)
2753 {
2754 /* The repeat...until naturally matches one or more.
2755 To also match zero times, we need to first jump to
2756 the end of the loop (its conditional jump). */
2757 INSERT_JUMP (jump, laststart, b);
2758 b += 3;
2759 }
2760 }
2761 else
2762 {
2763 /* non-greedy a?? */
2764 INSERT_JUMP (jump, laststart, b + 3);
2765 b += 3;
2766 INSERT_JUMP (on_failure_jump, laststart, laststart + 6);
2767 b += 3;
2768 }
2769 }
2770 }
2771 pending_exact = 0;
2772 break;
2773
2774
2775 case '.':
2776 laststart = b;
2777 BUF_PUSH (anychar);
2778 break;
2779
2780
2781 case '[':
2782 {
2783 re_char *p1;
2784
2785 CLEAR_RANGE_TABLE_WORK_USED (range_table_work);
2786
2787 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2788
2789 /* Ensure that we have enough space to push a charset: the
2790 opcode, the length count, and the bitset; 34 bytes in all. */
2791 GET_BUFFER_SPACE (34);
2792
2793 laststart = b;
2794
2795 /* We test `*p == '^' twice, instead of using an if
2796 statement, so we only need one BUF_PUSH. */
2797 BUF_PUSH (*p == '^' ? charset_not : charset);
2798 if (*p == '^')
2799 p++;
2800
2801 /* Remember the first position in the bracket expression. */
2802 p1 = p;
2803
2804 /* Push the number of bytes in the bitmap. */
2805 BUF_PUSH ((1 << BYTEWIDTH) / BYTEWIDTH);
2806
2807 /* Clear the whole map. */
2808 memset (b, 0, (1 << BYTEWIDTH) / BYTEWIDTH);
2809
2810 /* charset_not matches newline according to a syntax bit. */
2811 if ((re_opcode_t) b[-2] == charset_not
2812 && (syntax & RE_HAT_LISTS_NOT_NEWLINE))
2813 SET_LIST_BIT ('\n');
2814
2815 /* Read in characters and ranges, setting map bits. */
2816 for (;;)
2817 {
2818 boolean escaped_char = false;
2819 const unsigned char *p2 = p;
2820 re_wchar_t ch;
2821
2822 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2823
2824 /* Don't translate yet. The range TRANSLATE(X..Y) cannot
2825 always be determined from TRANSLATE(X) and TRANSLATE(Y)
2826 So the translation is done later in a loop. Example:
2827 (let ((case-fold-search t)) (string-match "[A-_]" "A")) */
2828 PATFETCH (c);
2829
2830 /* \ might escape characters inside [...] and [^...]. */
2831 if ((syntax & RE_BACKSLASH_ESCAPE_IN_LISTS) && c == '\\')
2832 {
2833 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
2834
2835 PATFETCH (c);
2836 escaped_char = true;
2837 }
2838 else
2839 {
2840 /* Could be the end of the bracket expression. If it's
2841 not (i.e., when the bracket expression is `[]' so
2842 far), the ']' character bit gets set way below. */
2843 if (c == ']' && p2 != p1)
2844 break;
2845 }
2846
2847 /* See if we're at the beginning of a possible character
2848 class. */
2849
2850 if (!escaped_char &&
2851 syntax & RE_CHAR_CLASSES && c == '[' && *p == ':')
2852 {
2853 /* Leave room for the null. */
2854 unsigned char str[CHAR_CLASS_MAX_LENGTH + 1];
2855 const unsigned char *class_beg;
2856
2857 PATFETCH (c);
2858 c1 = 0;
2859 class_beg = p;
2860
2861 /* If pattern is `[[:'. */
2862 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2863
2864 for (;;)
2865 {
2866 PATFETCH (c);
2867 if ((c == ':' && *p == ']') || p == pend)
2868 break;
2869 if (c1 < CHAR_CLASS_MAX_LENGTH)
2870 str[c1++] = c;
2871 else
2872 /* This is in any case an invalid class name. */
2873 str[0] = '\0';
2874 }
2875 str[c1] = '\0';
2876
2877 /* If isn't a word bracketed by `[:' and `:]':
2878 undo the ending character, the letters, and
2879 leave the leading `:' and `[' (but set bits for
2880 them). */
2881 if (c == ':' && *p == ']')
2882 {
2883 re_wctype_t cc = re_wctype (str);
2884
2885 if (cc == 0)
2886 FREE_STACK_RETURN (REG_ECTYPE);
2887
2888 /* Throw away the ] at the end of the character
2889 class. */
2890 PATFETCH (c);
2891
2892 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2893
2894 #ifndef emacs
2895 for (ch = 0; ch < (1 << BYTEWIDTH); ++ch)
2896 if (re_iswctype (btowc (ch), cc))
2897 {
2898 c = TRANSLATE (ch);
2899 if (c < (1 << BYTEWIDTH))
2900 SET_LIST_BIT (c);
2901 }
2902 #else /* emacs */
2903 /* Most character classes in a multibyte match
2904 just set a flag. Exceptions are is_blank,
2905 is_digit, is_cntrl, and is_xdigit, since
2906 they can only match ASCII characters. We
2907 don't need to handle them for multibyte.
2908 They are distinguished by a negative wctype. */
2909
2910 /* Setup the gl_state object to its buffer-defined
2911 value. This hardcodes the buffer-global
2912 syntax-table for ASCII chars, while the other chars
2913 will obey syntax-table properties. It's not ideal,
2914 but it's the way it's been done until now. */
2915 SETUP_BUFFER_SYNTAX_TABLE ();
2916
2917 for (ch = 0; ch < 256; ++ch)
2918 {
2919 c = RE_CHAR_TO_MULTIBYTE (ch);
2920 if (! CHAR_BYTE8_P (c)
2921 && re_iswctype (c, cc))
2922 {
2923 SET_LIST_BIT (ch);
2924 c1 = TRANSLATE (c);
2925 if (c1 == c)
2926 continue;
2927 if (ASCII_CHAR_P (c1))
2928 SET_LIST_BIT (c1);
2929 else if ((c1 = RE_CHAR_TO_UNIBYTE (c1)) >= 0)
2930 SET_LIST_BIT (c1);
2931 }
2932 }
2933 SET_RANGE_TABLE_WORK_AREA_BIT
2934 (range_table_work, re_wctype_to_bit (cc));
2935 #endif /* emacs */
2936 /* In most cases the matching rule for char classes
2937 only uses the syntax table for multibyte chars,
2938 so that the content of the syntax-table it is not
2939 hardcoded in the range_table. SPACE and WORD are
2940 the two exceptions. */
2941 if ((1 << cc) & ((1 << RECC_SPACE) | (1 << RECC_WORD)))
2942 bufp->used_syntax = 1;
2943
2944 /* Repeat the loop. */
2945 continue;
2946 }
2947 else
2948 {
2949 /* Go back to right after the "[:". */
2950 p = class_beg;
2951 SET_LIST_BIT ('[');
2952
2953 /* Because the `:' may starts the range, we
2954 can't simply set bit and repeat the loop.
2955 Instead, just set it to C and handle below. */
2956 c = ':';
2957 }
2958 }
2959
2960 if (p < pend && p[0] == '-' && p[1] != ']')
2961 {
2962
2963 /* Discard the `-'. */
2964 PATFETCH (c1);
2965
2966 /* Fetch the character which ends the range. */
2967 PATFETCH (c1);
2968 #ifdef emacs
2969 if (CHAR_BYTE8_P (c1)
2970 && ! ASCII_CHAR_P (c) && ! CHAR_BYTE8_P (c))
2971 /* Treat the range from a multibyte character to
2972 raw-byte character as empty. */
2973 c = c1 + 1;
2974 #endif /* emacs */
2975 }
2976 else
2977 /* Range from C to C. */
2978 c1 = c;
2979
2980 if (c > c1)
2981 {
2982 if (syntax & RE_NO_EMPTY_RANGES)
2983 FREE_STACK_RETURN (REG_ERANGEX);
2984 /* Else, repeat the loop. */
2985 }
2986 else
2987 {
2988 #ifndef emacs
2989 /* Set the range into bitmap */
2990 for (; c <= c1; c++)
2991 {
2992 ch = TRANSLATE (c);
2993 if (ch < (1 << BYTEWIDTH))
2994 SET_LIST_BIT (ch);
2995 }
2996 #else /* emacs */
2997 if (c < 128)
2998 {
2999 ch = MIN (127, c1);
3000 SETUP_ASCII_RANGE (range_table_work, c, ch);
3001 c = ch + 1;
3002 if (CHAR_BYTE8_P (c1))
3003 c = BYTE8_TO_CHAR (128);
3004 }
3005 if (c <= c1)
3006 {
3007 if (CHAR_BYTE8_P (c))
3008 {
3009 c = CHAR_TO_BYTE8 (c);
3010 c1 = CHAR_TO_BYTE8 (c1);
3011 for (; c <= c1; c++)
3012 SET_LIST_BIT (c);
3013 }
3014 else if (multibyte)
3015 {
3016 SETUP_MULTIBYTE_RANGE (range_table_work, c, c1);
3017 }
3018 else
3019 {
3020 SETUP_UNIBYTE_RANGE (range_table_work, c, c1);
3021 }
3022 }
3023 #endif /* emacs */
3024 }
3025 }
3026
3027 /* Discard any (non)matching list bytes that are all 0 at the
3028 end of the map. Decrease the map-length byte too. */
3029 while ((int) b[-1] > 0 && b[b[-1] - 1] == 0)
3030 b[-1]--;
3031 b += b[-1];
3032
3033 /* Build real range table from work area. */
3034 if (RANGE_TABLE_WORK_USED (range_table_work)
3035 || RANGE_TABLE_WORK_BITS (range_table_work))
3036 {
3037 int i;
3038 int used = RANGE_TABLE_WORK_USED (range_table_work);
3039
3040 /* Allocate space for COUNT + RANGE_TABLE. Needs two
3041 bytes for flags, two for COUNT, and three bytes for
3042 each character. */
3043 GET_BUFFER_SPACE (4 + used * 3);
3044
3045 /* Indicate the existence of range table. */
3046 laststart[1] |= 0x80;
3047
3048 /* Store the character class flag bits into the range table.
3049 If not in emacs, these flag bits are always 0. */
3050 *b++ = RANGE_TABLE_WORK_BITS (range_table_work) & 0xff;
3051 *b++ = RANGE_TABLE_WORK_BITS (range_table_work) >> 8;
3052
3053 STORE_NUMBER_AND_INCR (b, used / 2);
3054 for (i = 0; i < used; i++)
3055 STORE_CHARACTER_AND_INCR
3056 (b, RANGE_TABLE_WORK_ELT (range_table_work, i));
3057 }
3058 }
3059 break;
3060
3061
3062 case '(':
3063 if (syntax & RE_NO_BK_PARENS)
3064 goto handle_open;
3065 else
3066 goto normal_char;
3067
3068
3069 case ')':
3070 if (syntax & RE_NO_BK_PARENS)
3071 goto handle_close;
3072 else
3073 goto normal_char;
3074
3075
3076 case '\n':
3077 if (syntax & RE_NEWLINE_ALT)
3078 goto handle_alt;
3079 else
3080 goto normal_char;
3081
3082
3083 case '|':
3084 if (syntax & RE_NO_BK_VBAR)
3085 goto handle_alt;
3086 else
3087 goto normal_char;
3088
3089
3090 case '{':
3091 if (syntax & RE_INTERVALS && syntax & RE_NO_BK_BRACES)
3092 goto handle_interval;
3093 else
3094 goto normal_char;
3095
3096
3097 case '\\':
3098 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
3099
3100 /* Do not translate the character after the \, so that we can
3101 distinguish, e.g., \B from \b, even if we normally would
3102 translate, e.g., B to b. */
3103 PATFETCH (c);
3104
3105 switch (c)
3106 {
3107 case '(':
3108 if (syntax & RE_NO_BK_PARENS)
3109 goto normal_backslash;
3110
3111 handle_open:
3112 {
3113 int shy = 0;
3114 regnum_t regnum = 0;
3115 if (p+1 < pend)
3116 {
3117 /* Look for a special (?...) construct */
3118 if ((syntax & RE_SHY_GROUPS) && *p == '?')
3119 {
3120 PATFETCH (c); /* Gobble up the '?'. */
3121 while (!shy)
3122 {
3123 PATFETCH (c);
3124 switch (c)
3125 {
3126 case ':': shy = 1; break;
3127 case '0':
3128 /* An explicitly specified regnum must start
3129 with non-0. */
3130 if (regnum == 0)
3131 FREE_STACK_RETURN (REG_BADPAT);
3132 case '1': case '2': case '3': case '4':
3133 case '5': case '6': case '7': case '8': case '9':
3134 regnum = 10*regnum + (c - '0'); break;
3135 default:
3136 /* Only (?:...) is supported right now. */
3137 FREE_STACK_RETURN (REG_BADPAT);
3138 }
3139 }
3140 }
3141 }
3142
3143 if (!shy)
3144 regnum = ++bufp->re_nsub;
3145 else if (regnum)
3146 { /* It's actually not shy, but explicitly numbered. */
3147 shy = 0;
3148 if (regnum > bufp->re_nsub)
3149 bufp->re_nsub = regnum;
3150 else if (regnum > bufp->re_nsub
3151 /* Ideally, we'd want to check that the specified
3152 group can't have matched (i.e. all subgroups
3153 using the same regnum are in other branches of
3154 OR patterns), but we don't currently keep track
3155 of enough info to do that easily. */
3156 || group_in_compile_stack (compile_stack, regnum))
3157 FREE_STACK_RETURN (REG_BADPAT);
3158 }
3159 else
3160 /* It's really shy. */
3161 regnum = - bufp->re_nsub;
3162
3163 if (COMPILE_STACK_FULL)
3164 {
3165 RETALLOC (compile_stack.stack, compile_stack.size << 1,
3166 compile_stack_elt_t);
3167 if (compile_stack.stack == NULL) return REG_ESPACE;
3168
3169 compile_stack.size <<= 1;
3170 }
3171
3172 /* These are the values to restore when we hit end of this
3173 group. They are all relative offsets, so that if the
3174 whole pattern moves because of realloc, they will still
3175 be valid. */
3176 COMPILE_STACK_TOP.begalt_offset = begalt - bufp->buffer;
3177 COMPILE_STACK_TOP.fixup_alt_jump
3178 = fixup_alt_jump ? fixup_alt_jump - bufp->buffer + 1 : 0;
3179 COMPILE_STACK_TOP.laststart_offset = b - bufp->buffer;
3180 COMPILE_STACK_TOP.regnum = regnum;
3181
3182 /* Do not push a start_memory for groups beyond the last one
3183 we can represent in the compiled pattern. */
3184 if (regnum <= MAX_REGNUM && regnum > 0)
3185 BUF_PUSH_2 (start_memory, regnum);
3186
3187 compile_stack.avail++;
3188
3189 fixup_alt_jump = 0;
3190 laststart = 0;
3191 begalt = b;
3192 /* If we've reached MAX_REGNUM groups, then this open
3193 won't actually generate any code, so we'll have to
3194 clear pending_exact explicitly. */
3195 pending_exact = 0;
3196 break;
3197 }
3198
3199 case ')':
3200 if (syntax & RE_NO_BK_PARENS) goto normal_backslash;
3201
3202 if (COMPILE_STACK_EMPTY)
3203 {
3204 if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD)
3205 goto normal_backslash;
3206 else
3207 FREE_STACK_RETURN (REG_ERPAREN);
3208 }
3209
3210 handle_close:
3211 FIXUP_ALT_JUMP ();
3212
3213 /* See similar code for backslashed left paren above. */
3214 if (COMPILE_STACK_EMPTY)
3215 {
3216 if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD)
3217 goto normal_char;
3218 else
3219 FREE_STACK_RETURN (REG_ERPAREN);
3220 }
3221
3222 /* Since we just checked for an empty stack above, this
3223 ``can't happen''. */
3224 assert (compile_stack.avail != 0);
3225 {
3226 /* We don't just want to restore into `regnum', because
3227 later groups should continue to be numbered higher,
3228 as in `(ab)c(de)' -- the second group is #2. */
3229 regnum_t regnum;
3230
3231 compile_stack.avail--;
3232 begalt = bufp->buffer + COMPILE_STACK_TOP.begalt_offset;
3233 fixup_alt_jump
3234 = COMPILE_STACK_TOP.fixup_alt_jump
3235 ? bufp->buffer + COMPILE_STACK_TOP.fixup_alt_jump - 1
3236 : 0;
3237 laststart = bufp->buffer + COMPILE_STACK_TOP.laststart_offset;
3238 regnum = COMPILE_STACK_TOP.regnum;
3239 /* If we've reached MAX_REGNUM groups, then this open
3240 won't actually generate any code, so we'll have to
3241 clear pending_exact explicitly. */
3242 pending_exact = 0;
3243
3244 /* We're at the end of the group, so now we know how many
3245 groups were inside this one. */
3246 if (regnum <= MAX_REGNUM && regnum > 0)
3247 BUF_PUSH_2 (stop_memory, regnum);
3248 }
3249 break;
3250
3251
3252 case '|': /* `\|'. */
3253 if (syntax & RE_LIMITED_OPS || syntax & RE_NO_BK_VBAR)
3254 goto normal_backslash;
3255 handle_alt:
3256 if (syntax & RE_LIMITED_OPS)
3257 goto normal_char;
3258
3259 /* Insert before the previous alternative a jump which
3260 jumps to this alternative if the former fails. */
3261 GET_BUFFER_SPACE (3);
3262 INSERT_JUMP (on_failure_jump, begalt, b + 6);
3263 pending_exact = 0;
3264 b += 3;
3265
3266 /* The alternative before this one has a jump after it
3267 which gets executed if it gets matched. Adjust that
3268 jump so it will jump to this alternative's analogous
3269 jump (put in below, which in turn will jump to the next
3270 (if any) alternative's such jump, etc.). The last such
3271 jump jumps to the correct final destination. A picture:
3272 _____ _____
3273 | | | |
3274 | v | v
3275 a | b | c
3276
3277 If we are at `b', then fixup_alt_jump right now points to a
3278 three-byte space after `a'. We'll put in the jump, set
3279 fixup_alt_jump to right after `b', and leave behind three
3280 bytes which we'll fill in when we get to after `c'. */
3281
3282 FIXUP_ALT_JUMP ();
3283
3284 /* Mark and leave space for a jump after this alternative,
3285 to be filled in later either by next alternative or
3286 when know we're at the end of a series of alternatives. */
3287 fixup_alt_jump = b;
3288 GET_BUFFER_SPACE (3);
3289 b += 3;
3290
3291 laststart = 0;
3292 begalt = b;
3293 break;
3294
3295
3296 case '{':
3297 /* If \{ is a literal. */
3298 if (!(syntax & RE_INTERVALS)
3299 /* If we're at `\{' and it's not the open-interval
3300 operator. */
3301 || (syntax & RE_NO_BK_BRACES))
3302 goto normal_backslash;
3303
3304 handle_interval:
3305 {
3306 /* If got here, then the syntax allows intervals. */
3307
3308 /* At least (most) this many matches must be made. */
3309 int lower_bound = 0, upper_bound = -1;
3310
3311 beg_interval = p;
3312
3313 GET_UNSIGNED_NUMBER (lower_bound);
3314
3315 if (c == ',')
3316 GET_UNSIGNED_NUMBER (upper_bound);
3317 else
3318 /* Interval such as `{1}' => match exactly once. */
3319 upper_bound = lower_bound;
3320
3321 if (lower_bound < 0 || upper_bound > RE_DUP_MAX
3322 || (upper_bound >= 0 && lower_bound > upper_bound))
3323 FREE_STACK_RETURN (REG_BADBR);
3324
3325 if (!(syntax & RE_NO_BK_BRACES))
3326 {
3327 if (c != '\\')
3328 FREE_STACK_RETURN (REG_BADBR);
3329 if (p == pend)
3330 FREE_STACK_RETURN (REG_EESCAPE);
3331 PATFETCH (c);
3332 }
3333
3334 if (c != '}')
3335 FREE_STACK_RETURN (REG_BADBR);
3336
3337 /* We just parsed a valid interval. */
3338
3339 /* If it's invalid to have no preceding re. */
3340 if (!laststart)
3341 {
3342 if (syntax & RE_CONTEXT_INVALID_OPS)
3343 FREE_STACK_RETURN (REG_BADRPT);
3344 else if (syntax & RE_CONTEXT_INDEP_OPS)
3345 laststart = b;
3346 else
3347 goto unfetch_interval;
3348 }
3349
3350 if (upper_bound == 0)
3351 /* If the upper bound is zero, just drop the sub pattern
3352 altogether. */
3353 b = laststart;
3354 else if (lower_bound == 1 && upper_bound == 1)
3355 /* Just match it once: nothing to do here. */
3356 ;
3357
3358 /* Otherwise, we have a nontrivial interval. When
3359 we're all done, the pattern will look like:
3360 set_number_at <jump count> <upper bound>
3361 set_number_at <succeed_n count> <lower bound>
3362 succeed_n <after jump addr> <succeed_n count>
3363 <body of loop>
3364 jump_n <succeed_n addr> <jump count>
3365 (The upper bound and `jump_n' are omitted if
3366 `upper_bound' is 1, though.) */
3367 else
3368 { /* If the upper bound is > 1, we need to insert
3369 more at the end of the loop. */
3370 unsigned int nbytes = (upper_bound < 0 ? 3
3371 : upper_bound > 1 ? 5 : 0);
3372 unsigned int startoffset = 0;
3373
3374 GET_BUFFER_SPACE (20); /* We might use less. */
3375
3376 if (lower_bound == 0)
3377 {
3378 /* A succeed_n that starts with 0 is really a
3379 a simple on_failure_jump_loop. */
3380 INSERT_JUMP (on_failure_jump_loop, laststart,
3381 b + 3 + nbytes);
3382 b += 3;
3383 }
3384 else
3385 {
3386 /* Initialize lower bound of the `succeed_n', even
3387 though it will be set during matching by its
3388 attendant `set_number_at' (inserted next),
3389 because `re_compile_fastmap' needs to know.
3390 Jump to the `jump_n' we might insert below. */
3391 INSERT_JUMP2 (succeed_n, laststart,
3392 b + 5 + nbytes,
3393 lower_bound);
3394 b += 5;
3395
3396 /* Code to initialize the lower bound. Insert
3397 before the `succeed_n'. The `5' is the last two
3398 bytes of this `set_number_at', plus 3 bytes of
3399 the following `succeed_n'. */
3400 insert_op2 (set_number_at, laststart, 5, lower_bound, b);
3401 b += 5;
3402 startoffset += 5;
3403 }
3404
3405 if (upper_bound < 0)
3406 {
3407 /* A negative upper bound stands for infinity,
3408 in which case it degenerates to a plain jump. */
3409 STORE_JUMP (jump, b, laststart + startoffset);
3410 b += 3;
3411 }
3412 else if (upper_bound > 1)
3413 { /* More than one repetition is allowed, so
3414 append a backward jump to the `succeed_n'
3415 that starts this interval.
3416
3417 When we've reached this during matching,
3418 we'll have matched the interval once, so
3419 jump back only `upper_bound - 1' times. */
3420 STORE_JUMP2 (jump_n, b, laststart + startoffset,
3421 upper_bound - 1);
3422 b += 5;
3423
3424 /* The location we want to set is the second
3425 parameter of the `jump_n'; that is `b-2' as
3426 an absolute address. `laststart' will be
3427 the `set_number_at' we're about to insert;
3428 `laststart+3' the number to set, the source
3429 for the relative address. But we are
3430 inserting into the middle of the pattern --
3431 so everything is getting moved up by 5.
3432 Conclusion: (b - 2) - (laststart + 3) + 5,
3433 i.e., b - laststart.
3434
3435 We insert this at the beginning of the loop
3436 so that if we fail during matching, we'll
3437 reinitialize the bounds. */
3438 insert_op2 (set_number_at, laststart, b - laststart,
3439 upper_bound - 1, b);
3440 b += 5;
3441 }
3442 }
3443 pending_exact = 0;
3444 beg_interval = NULL;
3445 }
3446 break;
3447
3448 unfetch_interval:
3449 /* If an invalid interval, match the characters as literals. */
3450 assert (beg_interval);
3451 p = beg_interval;
3452 beg_interval = NULL;
3453
3454 /* normal_char and normal_backslash need `c'. */
3455 c = '{';
3456
3457 if (!(syntax & RE_NO_BK_BRACES))
3458 {
3459 assert (p > pattern && p[-1] == '\\');
3460 goto normal_backslash;
3461 }
3462 else
3463 goto normal_char;
3464
3465 #ifdef emacs
3466 /* There is no way to specify the before_dot and after_dot
3467 operators. rms says this is ok. --karl */
3468 case '=':
3469 laststart = b;
3470 BUF_PUSH (at_dot);
3471 break;
3472
3473 case 's':
3474 laststart = b;
3475 PATFETCH (c);
3476 BUF_PUSH_2 (syntaxspec, syntax_spec_code[c]);
3477 break;
3478
3479 case 'S':
3480 laststart = b;
3481 PATFETCH (c);
3482 BUF_PUSH_2 (notsyntaxspec, syntax_spec_code[c]);
3483 break;
3484
3485 case 'c':
3486 laststart = b;
3487 PATFETCH (c);
3488 BUF_PUSH_2 (categoryspec, c);
3489 break;
3490
3491 case 'C':
3492 laststart = b;
3493 PATFETCH (c);
3494 BUF_PUSH_2 (notcategoryspec, c);
3495 break;
3496 #endif /* emacs */
3497
3498
3499 case 'w':
3500 if (syntax & RE_NO_GNU_OPS)
3501 goto normal_char;
3502 laststart = b;
3503 BUF_PUSH_2 (syntaxspec, Sword);
3504 break;
3505
3506
3507 case 'W':
3508 if (syntax & RE_NO_GNU_OPS)
3509 goto normal_char;
3510 laststart = b;
3511 BUF_PUSH_2 (notsyntaxspec, Sword);
3512 break;
3513
3514
3515 case '<':
3516 if (syntax & RE_NO_GNU_OPS)
3517 goto normal_char;
3518 laststart = b;
3519 BUF_PUSH (wordbeg);
3520 break;
3521
3522 case '>':
3523 if (syntax & RE_NO_GNU_OPS)
3524 goto normal_char;
3525 laststart = b;
3526 BUF_PUSH (wordend);
3527 break;
3528
3529 case '_':
3530 if (syntax & RE_NO_GNU_OPS)
3531 goto normal_char;
3532 laststart = b;
3533 PATFETCH (c);
3534 if (c == '<')
3535 BUF_PUSH (symbeg);
3536 else if (c == '>')
3537 BUF_PUSH (symend);
3538 else
3539 FREE_STACK_RETURN (REG_BADPAT);
3540 break;
3541
3542 case 'b':
3543 if (syntax & RE_NO_GNU_OPS)
3544 goto normal_char;
3545 BUF_PUSH (wordbound);
3546 break;
3547
3548 case 'B':
3549 if (syntax & RE_NO_GNU_OPS)
3550 goto normal_char;
3551 BUF_PUSH (notwordbound);
3552 break;
3553
3554 case '`':
3555 if (syntax & RE_NO_GNU_OPS)
3556 goto normal_char;
3557 BUF_PUSH (begbuf);
3558 break;
3559
3560 case '\'':
3561 if (syntax & RE_NO_GNU_OPS)
3562 goto normal_char;
3563 BUF_PUSH (endbuf);
3564 break;
3565
3566 case '1': case '2': case '3': case '4': case '5':
3567 case '6': case '7': case '8': case '9':
3568 {
3569 regnum_t reg;
3570
3571 if (syntax & RE_NO_BK_REFS)
3572 goto normal_backslash;
3573
3574 reg = c - '0';
3575
3576 if (reg > bufp->re_nsub || reg < 1
3577 /* Can't back reference to a subexp before its end. */
3578 || group_in_compile_stack (compile_stack, reg))
3579 FREE_STACK_RETURN (REG_ESUBREG);
3580
3581 laststart = b;
3582 BUF_PUSH_2 (duplicate, reg);
3583 }
3584 break;
3585
3586
3587 case '+':
3588 case '?':
3589 if (syntax & RE_BK_PLUS_QM)
3590 goto handle_plus;
3591 else
3592 goto normal_backslash;
3593
3594 default:
3595 normal_backslash:
3596 /* You might think it would be useful for \ to mean
3597 not to translate; but if we don't translate it
3598 it will never match anything. */
3599 goto normal_char;
3600 }
3601 break;
3602
3603
3604 default:
3605 /* Expects the character in `c'. */
3606 normal_char:
3607 /* If no exactn currently being built. */
3608 if (!pending_exact
3609
3610 /* If last exactn not at current position. */
3611 || pending_exact + *pending_exact + 1 != b
3612
3613 /* We have only one byte following the exactn for the count. */
3614 || *pending_exact >= (1 << BYTEWIDTH) - MAX_MULTIBYTE_LENGTH
3615
3616 /* If followed by a repetition operator. */
3617 || (p != pend && (*p == '*' || *p == '^'))
3618 || ((syntax & RE_BK_PLUS_QM)
3619 ? p + 1 < pend && *p == '\\' && (p[1] == '+' || p[1] == '?')
3620 : p != pend && (*p == '+' || *p == '?'))
3621 || ((syntax & RE_INTERVALS)
3622 && ((syntax & RE_NO_BK_BRACES)
3623 ? p != pend && *p == '{'
3624 : p + 1 < pend && p[0] == '\\' && p[1] == '{')))
3625 {
3626 /* Start building a new exactn. */
3627
3628 laststart = b;
3629
3630 BUF_PUSH_2 (exactn, 0);
3631 pending_exact = b - 1;
3632 }
3633
3634 GET_BUFFER_SPACE (MAX_MULTIBYTE_LENGTH);
3635 {
3636 int len;
3637
3638 if (multibyte)
3639 {
3640 c = TRANSLATE (c);
3641 len = CHAR_STRING (c, b);
3642 b += len;
3643 }
3644 else
3645 {
3646 c1 = RE_CHAR_TO_MULTIBYTE (c);
3647 if (! CHAR_BYTE8_P (c1))
3648 {
3649 re_wchar_t c2 = TRANSLATE (c1);
3650
3651 if (c1 != c2 && (c1 = RE_CHAR_TO_UNIBYTE (c2)) >= 0)
3652 c = c1;
3653 }
3654 *b++ = c;
3655 len = 1;
3656 }
3657 (*pending_exact) += len;
3658 }
3659
3660 break;
3661 } /* switch (c) */
3662 } /* while p != pend */
3663
3664
3665 /* Through the pattern now. */
3666
3667 FIXUP_ALT_JUMP ();
3668
3669 if (!COMPILE_STACK_EMPTY)
3670 FREE_STACK_RETURN (REG_EPAREN);
3671
3672 /* If we don't want backtracking, force success
3673 the first time we reach the end of the compiled pattern. */
3674 if (syntax & RE_NO_POSIX_BACKTRACKING)
3675 BUF_PUSH (succeed);
3676
3677 /* We have succeeded; set the length of the buffer. */
3678 bufp->used = b - bufp->buffer;
3679
3680 #ifdef DEBUG
3681 if (debug > 0)
3682 {
3683 re_compile_fastmap (bufp);
3684 DEBUG_PRINT ("\nCompiled pattern: \n");
3685 print_compiled_pattern (bufp);
3686 }
3687 debug--;
3688 #endif /* DEBUG */
3689
3690 #ifndef MATCH_MAY_ALLOCATE
3691 /* Initialize the failure stack to the largest possible stack. This
3692 isn't necessary unless we're trying to avoid calling alloca in
3693 the search and match routines. */
3694 {
3695 int num_regs = bufp->re_nsub + 1;
3696
3697 if (fail_stack.size < re_max_failures * TYPICAL_FAILURE_SIZE)
3698 {
3699 fail_stack.size = re_max_failures * TYPICAL_FAILURE_SIZE;
3700 falk_stack.stack = realloc (fail_stack.stack,
3701 fail_stack.size * sizeof *falk_stack.stack);
3702 }
3703
3704 regex_grow_registers (num_regs);
3705 }
3706 #endif /* not MATCH_MAY_ALLOCATE */
3707
3708 FREE_STACK_RETURN (REG_NOERROR);
3709 } /* regex_compile */
3710 \f
3711 /* Subroutines for `regex_compile'. */
3712
3713 /* Store OP at LOC followed by two-byte integer parameter ARG. */
3714
3715 static void
3716 store_op1 (re_opcode_t op, unsigned char *loc, int arg)
3717 {
3718 *loc = (unsigned char) op;
3719 STORE_NUMBER (loc + 1, arg);
3720 }
3721
3722
3723 /* Like `store_op1', but for two two-byte parameters ARG1 and ARG2. */
3724
3725 static void
3726 store_op2 (re_opcode_t op, unsigned char *loc, int arg1, int arg2)
3727 {
3728 *loc = (unsigned char) op;
3729 STORE_NUMBER (loc + 1, arg1);
3730 STORE_NUMBER (loc + 3, arg2);
3731 }
3732
3733
3734 /* Copy the bytes from LOC to END to open up three bytes of space at LOC
3735 for OP followed by two-byte integer parameter ARG. */
3736
3737 static void
3738 insert_op1 (re_opcode_t op, unsigned char *loc, int arg, unsigned char *end)
3739 {
3740 register unsigned char *pfrom = end;
3741 register unsigned char *pto = end + 3;
3742
3743 while (pfrom != loc)
3744 *--pto = *--pfrom;
3745
3746 store_op1 (op, loc, arg);
3747 }
3748
3749
3750 /* Like `insert_op1', but for two two-byte parameters ARG1 and ARG2. */
3751
3752 static void
3753 insert_op2 (re_opcode_t op, unsigned char *loc, int arg1, int arg2, unsigned char *end)
3754 {
3755 register unsigned char *pfrom = end;
3756 register unsigned char *pto = end + 5;
3757
3758 while (pfrom != loc)
3759 *--pto = *--pfrom;
3760
3761 store_op2 (op, loc, arg1, arg2);
3762 }
3763
3764
3765 /* P points to just after a ^ in PATTERN. Return true if that ^ comes
3766 after an alternative or a begin-subexpression. We assume there is at
3767 least one character before the ^. */
3768
3769 static boolean
3770 at_begline_loc_p (const_re_char *pattern, const_re_char *p, reg_syntax_t syntax)
3771 {
3772 re_char *prev = p - 2;
3773 boolean odd_backslashes;
3774
3775 /* After a subexpression? */
3776 if (*prev == '(')
3777 odd_backslashes = (syntax & RE_NO_BK_PARENS) == 0;
3778
3779 /* After an alternative? */
3780 else if (*prev == '|')
3781 odd_backslashes = (syntax & RE_NO_BK_VBAR) == 0;
3782
3783 /* After a shy subexpression? */
3784 else if (*prev == ':' && (syntax & RE_SHY_GROUPS))
3785 {
3786 /* Skip over optional regnum. */
3787 while (prev - 1 >= pattern && prev[-1] >= '0' && prev[-1] <= '9')
3788 --prev;
3789
3790 if (!(prev - 2 >= pattern
3791 && prev[-1] == '?' && prev[-2] == '('))
3792 return false;
3793 prev -= 2;
3794 odd_backslashes = (syntax & RE_NO_BK_PARENS) == 0;
3795 }
3796 else
3797 return false;
3798
3799 /* Count the number of preceding backslashes. */
3800 p = prev;
3801 while (prev - 1 >= pattern && prev[-1] == '\\')
3802 --prev;
3803 return (p - prev) & odd_backslashes;
3804 }
3805
3806
3807 /* The dual of at_begline_loc_p. This one is for $. We assume there is
3808 at least one character after the $, i.e., `P < PEND'. */
3809
3810 static boolean
3811 at_endline_loc_p (const_re_char *p, const_re_char *pend, reg_syntax_t syntax)
3812 {
3813 re_char *next = p;
3814 boolean next_backslash = *next == '\\';
3815 re_char *next_next = p + 1 < pend ? p + 1 : 0;
3816
3817 return
3818 /* Before a subexpression? */
3819 (syntax & RE_NO_BK_PARENS ? *next == ')'
3820 : next_backslash && next_next && *next_next == ')')
3821 /* Before an alternative? */
3822 || (syntax & RE_NO_BK_VBAR ? *next == '|'
3823 : next_backslash && next_next && *next_next == '|');
3824 }
3825
3826
3827 /* Returns true if REGNUM is in one of COMPILE_STACK's elements and
3828 false if it's not. */
3829
3830 static boolean
3831 group_in_compile_stack (compile_stack_type compile_stack, regnum_t regnum)
3832 {
3833 ssize_t this_element;
3834
3835 for (this_element = compile_stack.avail - 1;
3836 this_element >= 0;
3837 this_element--)
3838 if (compile_stack.stack[this_element].regnum == regnum)
3839 return true;
3840
3841 return false;
3842 }
3843 \f
3844 /* analyse_first.
3845 If fastmap is non-NULL, go through the pattern and fill fastmap
3846 with all the possible leading chars. If fastmap is NULL, don't
3847 bother filling it up (obviously) and only return whether the
3848 pattern could potentially match the empty string.
3849
3850 Return 1 if p..pend might match the empty string.
3851 Return 0 if p..pend matches at least one char.
3852 Return -1 if fastmap was not updated accurately. */
3853
3854 static int
3855 analyse_first (const_re_char *p, const_re_char *pend, char *fastmap,
3856 const int multibyte)
3857 {
3858 int j, k;
3859 boolean not;
3860
3861 /* If all elements for base leading-codes in fastmap is set, this
3862 flag is set true. */
3863 boolean match_any_multibyte_characters = false;
3864
3865 assert (p);
3866
3867 /* The loop below works as follows:
3868 - It has a working-list kept in the PATTERN_STACK and which basically
3869 starts by only containing a pointer to the first operation.
3870 - If the opcode we're looking at is a match against some set of
3871 chars, then we add those chars to the fastmap and go on to the
3872 next work element from the worklist (done via `break').
3873 - If the opcode is a control operator on the other hand, we either
3874 ignore it (if it's meaningless at this point, such as `start_memory')
3875 or execute it (if it's a jump). If the jump has several destinations
3876 (i.e. `on_failure_jump'), then we push the other destination onto the
3877 worklist.
3878 We guarantee termination by ignoring backward jumps (more or less),
3879 so that `p' is monotonically increasing. More to the point, we
3880 never set `p' (or push) anything `<= p1'. */
3881
3882 while (p < pend)
3883 {
3884 /* `p1' is used as a marker of how far back a `on_failure_jump'
3885 can go without being ignored. It is normally equal to `p'
3886 (which prevents any backward `on_failure_jump') except right
3887 after a plain `jump', to allow patterns such as:
3888 0: jump 10
3889 3..9: <body>
3890 10: on_failure_jump 3
3891 as used for the *? operator. */
3892 re_char *p1 = p;
3893
3894 switch (*p++)
3895 {
3896 case succeed:
3897 return 1;
3898
3899 case duplicate:
3900 /* If the first character has to match a backreference, that means
3901 that the group was empty (since it already matched). Since this
3902 is the only case that interests us here, we can assume that the
3903 backreference must match the empty string. */
3904 p++;
3905 continue;
3906
3907
3908 /* Following are the cases which match a character. These end
3909 with `break'. */
3910
3911 case exactn:
3912 if (fastmap)
3913 {
3914 /* If multibyte is nonzero, the first byte of each
3915 character is an ASCII or a leading code. Otherwise,
3916 each byte is a character. Thus, this works in both
3917 cases. */
3918 fastmap[p[1]] = 1;
3919 if (! multibyte)
3920 {
3921 /* For the case of matching this unibyte regex
3922 against multibyte, we must set a leading code of
3923 the corresponding multibyte character. */
3924 int c = RE_CHAR_TO_MULTIBYTE (p[1]);
3925
3926 fastmap[CHAR_LEADING_CODE (c)] = 1;
3927 }
3928 }
3929 break;
3930
3931
3932 case anychar:
3933 /* We could put all the chars except for \n (and maybe \0)
3934 but we don't bother since it is generally not worth it. */
3935 if (!fastmap) break;
3936 return -1;
3937
3938
3939 case charset_not:
3940 if (!fastmap) break;
3941 {
3942 /* Chars beyond end of bitmap are possible matches. */
3943 for (j = CHARSET_BITMAP_SIZE (&p[-1]) * BYTEWIDTH;
3944 j < (1 << BYTEWIDTH); j++)
3945 fastmap[j] = 1;
3946 }
3947
3948 /* Fallthrough */
3949 case charset:
3950 if (!fastmap) break;
3951 not = (re_opcode_t) *(p - 1) == charset_not;
3952 for (j = CHARSET_BITMAP_SIZE (&p[-1]) * BYTEWIDTH - 1, p++;
3953 j >= 0; j--)
3954 if (!!(p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH))) ^ not)
3955 fastmap[j] = 1;
3956
3957 #ifdef emacs
3958 if (/* Any leading code can possibly start a character
3959 which doesn't match the specified set of characters. */
3960 not
3961 ||
3962 /* If we can match a character class, we can match any
3963 multibyte characters. */
3964 (CHARSET_RANGE_TABLE_EXISTS_P (&p[-2])
3965 && CHARSET_RANGE_TABLE_BITS (&p[-2]) != 0))
3966
3967 {
3968 if (match_any_multibyte_characters == false)
3969 {
3970 for (j = MIN_MULTIBYTE_LEADING_CODE;
3971 j <= MAX_MULTIBYTE_LEADING_CODE; j++)
3972 fastmap[j] = 1;
3973 match_any_multibyte_characters = true;
3974 }
3975 }
3976
3977 else if (!not && CHARSET_RANGE_TABLE_EXISTS_P (&p[-2])
3978 && match_any_multibyte_characters == false)
3979 {
3980 /* Set fastmap[I] to 1 where I is a leading code of each
3981 multibyte character in the range table. */
3982 int c, count;
3983 unsigned char lc1, lc2;
3984
3985 /* Make P points the range table. `+ 2' is to skip flag
3986 bits for a character class. */
3987 p += CHARSET_BITMAP_SIZE (&p[-2]) + 2;
3988
3989 /* Extract the number of ranges in range table into COUNT. */
3990 EXTRACT_NUMBER_AND_INCR (count, p);
3991 for (; count > 0; count--, p += 3)
3992 {
3993 /* Extract the start and end of each range. */
3994 EXTRACT_CHARACTER (c, p);
3995 lc1 = CHAR_LEADING_CODE (c);
3996 p += 3;
3997 EXTRACT_CHARACTER (c, p);
3998 lc2 = CHAR_LEADING_CODE (c);
3999 for (j = lc1; j <= lc2; j++)
4000 fastmap[j] = 1;
4001 }
4002 }
4003 #endif
4004 break;
4005
4006 case syntaxspec:
4007 case notsyntaxspec:
4008 if (!fastmap) break;
4009 #ifndef emacs
4010 not = (re_opcode_t)p[-1] == notsyntaxspec;
4011 k = *p++;
4012 for (j = 0; j < (1 << BYTEWIDTH); j++)
4013 if ((SYNTAX (j) == (enum syntaxcode) k) ^ not)
4014 fastmap[j] = 1;
4015 break;
4016 #else /* emacs */
4017 /* This match depends on text properties. These end with
4018 aborting optimizations. */
4019 return -1;
4020
4021 case categoryspec:
4022 case notcategoryspec:
4023 if (!fastmap) break;
4024 not = (re_opcode_t)p[-1] == notcategoryspec;
4025 k = *p++;
4026 for (j = (1 << BYTEWIDTH); j >= 0; j--)
4027 if ((CHAR_HAS_CATEGORY (j, k)) ^ not)
4028 fastmap[j] = 1;
4029
4030 /* Any leading code can possibly start a character which
4031 has or doesn't has the specified category. */
4032 if (match_any_multibyte_characters == false)
4033 {
4034 for (j = MIN_MULTIBYTE_LEADING_CODE;
4035 j <= MAX_MULTIBYTE_LEADING_CODE; j++)
4036 fastmap[j] = 1;
4037 match_any_multibyte_characters = true;
4038 }
4039 break;
4040
4041 /* All cases after this match the empty string. These end with
4042 `continue'. */
4043
4044 case before_dot:
4045 case at_dot:
4046 case after_dot:
4047 #endif /* !emacs */
4048 case no_op:
4049 case begline:
4050 case endline:
4051 case begbuf:
4052 case endbuf:
4053 case wordbound:
4054 case notwordbound:
4055 case wordbeg:
4056 case wordend:
4057 case symbeg:
4058 case symend:
4059 continue;
4060
4061
4062 case jump:
4063 EXTRACT_NUMBER_AND_INCR (j, p);
4064 if (j < 0)
4065 /* Backward jumps can only go back to code that we've already
4066 visited. `re_compile' should make sure this is true. */
4067 break;
4068 p += j;
4069 switch (*p)
4070 {
4071 case on_failure_jump:
4072 case on_failure_keep_string_jump:
4073 case on_failure_jump_loop:
4074 case on_failure_jump_nastyloop:
4075 case on_failure_jump_smart:
4076 p++;
4077 break;
4078 default:
4079 continue;
4080 };
4081 /* Keep `p1' to allow the `on_failure_jump' we are jumping to
4082 to jump back to "just after here". */
4083 /* Fallthrough */
4084
4085 case on_failure_jump:
4086 case on_failure_keep_string_jump:
4087 case on_failure_jump_nastyloop:
4088 case on_failure_jump_loop:
4089 case on_failure_jump_smart:
4090 EXTRACT_NUMBER_AND_INCR (j, p);
4091 if (p + j <= p1)
4092 ; /* Backward jump to be ignored. */
4093 else
4094 { /* We have to look down both arms.
4095 We first go down the "straight" path so as to minimize
4096 stack usage when going through alternatives. */
4097 int r = analyse_first (p, pend, fastmap, multibyte);
4098 if (r) return r;
4099 p += j;
4100 }
4101 continue;
4102
4103
4104 case jump_n:
4105 /* This code simply does not properly handle forward jump_n. */
4106 DEBUG_STATEMENT (EXTRACT_NUMBER (j, p); assert (j < 0));
4107 p += 4;
4108 /* jump_n can either jump or fall through. The (backward) jump
4109 case has already been handled, so we only need to look at the
4110 fallthrough case. */
4111 continue;
4112
4113 case succeed_n:
4114 /* If N == 0, it should be an on_failure_jump_loop instead. */
4115 DEBUG_STATEMENT (EXTRACT_NUMBER (j, p + 2); assert (j > 0));
4116 p += 4;
4117 /* We only care about one iteration of the loop, so we don't
4118 need to consider the case where this behaves like an
4119 on_failure_jump. */
4120 continue;
4121
4122
4123 case set_number_at:
4124 p += 4;
4125 continue;
4126
4127
4128 case start_memory:
4129 case stop_memory:
4130 p += 1;
4131 continue;
4132
4133
4134 default:
4135 abort (); /* We have listed all the cases. */
4136 } /* switch *p++ */
4137
4138 /* Getting here means we have found the possible starting
4139 characters for one path of the pattern -- and that the empty
4140 string does not match. We need not follow this path further. */
4141 return 0;
4142 } /* while p */
4143
4144 /* We reached the end without matching anything. */
4145 return 1;
4146
4147 } /* analyse_first */
4148 \f
4149 /* re_compile_fastmap computes a ``fastmap'' for the compiled pattern in
4150 BUFP. A fastmap records which of the (1 << BYTEWIDTH) possible
4151 characters can start a string that matches the pattern. This fastmap
4152 is used by re_search to skip quickly over impossible starting points.
4153
4154 Character codes above (1 << BYTEWIDTH) are not represented in the
4155 fastmap, but the leading codes are represented. Thus, the fastmap
4156 indicates which character sets could start a match.
4157
4158 The caller must supply the address of a (1 << BYTEWIDTH)-byte data
4159 area as BUFP->fastmap.
4160
4161 We set the `fastmap', `fastmap_accurate', and `can_be_null' fields in
4162 the pattern buffer.
4163
4164 Returns 0 if we succeed, -2 if an internal error. */
4165
4166 int
4167 re_compile_fastmap (struct re_pattern_buffer *bufp)
4168 {
4169 char *fastmap = bufp->fastmap;
4170 int analysis;
4171
4172 assert (fastmap && bufp->buffer);
4173
4174 memset (fastmap, 0, 1 << BYTEWIDTH); /* Assume nothing's valid. */
4175 bufp->fastmap_accurate = 1; /* It will be when we're done. */
4176
4177 analysis = analyse_first (bufp->buffer, bufp->buffer + bufp->used,
4178 fastmap, RE_MULTIBYTE_P (bufp));
4179 bufp->can_be_null = (analysis != 0);
4180 return 0;
4181 } /* re_compile_fastmap */
4182 \f
4183 /* Set REGS to hold NUM_REGS registers, storing them in STARTS and
4184 ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use
4185 this memory for recording register information. STARTS and ENDS
4186 must be allocated using the malloc library routine, and must each
4187 be at least NUM_REGS * sizeof (regoff_t) bytes long.
4188
4189 If NUM_REGS == 0, then subsequent matches should allocate their own
4190 register data.
4191
4192 Unless this function is called, the first search or match using
4193 PATTERN_BUFFER will allocate its own register data, without
4194 freeing the old data. */
4195
4196 void
4197 re_set_registers (struct re_pattern_buffer *bufp, struct re_registers *regs, unsigned int num_regs, regoff_t *starts, regoff_t *ends)
4198 {
4199 if (num_regs)
4200 {
4201 bufp->regs_allocated = REGS_REALLOCATE;
4202 regs->num_regs = num_regs;
4203 regs->start = starts;
4204 regs->end = ends;
4205 }
4206 else
4207 {
4208 bufp->regs_allocated = REGS_UNALLOCATED;
4209 regs->num_regs = 0;
4210 regs->start = regs->end = 0;
4211 }
4212 }
4213 WEAK_ALIAS (__re_set_registers, re_set_registers)
4214 \f
4215 /* Searching routines. */
4216
4217 /* Like re_search_2, below, but only one string is specified, and
4218 doesn't let you say where to stop matching. */
4219
4220 regoff_t
4221 re_search (struct re_pattern_buffer *bufp, const char *string, size_t size,
4222 ssize_t startpos, ssize_t range, struct re_registers *regs)
4223 {
4224 return re_search_2 (bufp, NULL, 0, string, size, startpos, range,
4225 regs, size);
4226 }
4227 WEAK_ALIAS (__re_search, re_search)
4228
4229 /* Head address of virtual concatenation of string. */
4230 #define HEAD_ADDR_VSTRING(P) \
4231 (((P) >= size1 ? string2 : string1))
4232
4233 /* Address of POS in the concatenation of virtual string. */
4234 #define POS_ADDR_VSTRING(POS) \
4235 (((POS) >= size1 ? string2 - size1 : string1) + (POS))
4236
4237 /* Using the compiled pattern in BUFP->buffer, first tries to match the
4238 virtual concatenation of STRING1 and STRING2, starting first at index
4239 STARTPOS, then at STARTPOS + 1, and so on.
4240
4241 STRING1 and STRING2 have length SIZE1 and SIZE2, respectively.
4242
4243 RANGE is how far to scan while trying to match. RANGE = 0 means try
4244 only at STARTPOS; in general, the last start tried is STARTPOS +
4245 RANGE.
4246
4247 In REGS, return the indices of the virtual concatenation of STRING1
4248 and STRING2 that matched the entire BUFP->buffer and its contained
4249 subexpressions.
4250
4251 Do not consider matching one past the index STOP in the virtual
4252 concatenation of STRING1 and STRING2.
4253
4254 We return either the position in the strings at which the match was
4255 found, -1 if no match, or -2 if error (such as failure
4256 stack overflow). */
4257
4258 regoff_t
4259 re_search_2 (struct re_pattern_buffer *bufp, const char *str1, size_t size1,
4260 const char *str2, size_t size2, ssize_t startpos, ssize_t range,
4261 struct re_registers *regs, ssize_t stop)
4262 {
4263 regoff_t val;
4264 re_char *string1 = (re_char*) str1;
4265 re_char *string2 = (re_char*) str2;
4266 register char *fastmap = bufp->fastmap;
4267 register RE_TRANSLATE_TYPE translate = bufp->translate;
4268 size_t total_size = size1 + size2;
4269 ssize_t endpos = startpos + range;
4270 boolean anchored_start;
4271 /* Nonzero if we are searching multibyte string. */
4272 const boolean multibyte = RE_TARGET_MULTIBYTE_P (bufp);
4273
4274 /* Check for out-of-range STARTPOS. */
4275 if (startpos < 0 || startpos > total_size)
4276 return -1;
4277
4278 /* Fix up RANGE if it might eventually take us outside
4279 the virtual concatenation of STRING1 and STRING2.
4280 Make sure we won't move STARTPOS below 0 or above TOTAL_SIZE. */
4281 if (endpos < 0)
4282 range = 0 - startpos;
4283 else if (endpos > total_size)
4284 range = total_size - startpos;
4285
4286 /* If the search isn't to be a backwards one, don't waste time in a
4287 search for a pattern anchored at beginning of buffer. */
4288 if (bufp->used > 0 && (re_opcode_t) bufp->buffer[0] == begbuf && range > 0)
4289 {
4290 if (startpos > 0)
4291 return -1;
4292 else
4293 range = 0;
4294 }
4295
4296 #ifdef emacs
4297 /* In a forward search for something that starts with \=.
4298 don't keep searching past point. */
4299 if (bufp->used > 0 && (re_opcode_t) bufp->buffer[0] == at_dot && range > 0)
4300 {
4301 range = PT_BYTE - BEGV_BYTE - startpos;
4302 if (range < 0)
4303 return -1;
4304 }
4305 #endif /* emacs */
4306
4307 /* Update the fastmap now if not correct already. */
4308 if (fastmap && !bufp->fastmap_accurate)
4309 re_compile_fastmap (bufp);
4310
4311 /* See whether the pattern is anchored. */
4312 anchored_start = (bufp->buffer[0] == begline);
4313
4314 #ifdef emacs
4315 gl_state.object = re_match_object; /* Used by SYNTAX_TABLE_BYTE_TO_CHAR. */
4316 {
4317 ssize_t charpos = SYNTAX_TABLE_BYTE_TO_CHAR (POS_AS_IN_BUFFER (startpos));
4318
4319 SETUP_SYNTAX_TABLE_FOR_OBJECT (re_match_object, charpos, 1);
4320 }
4321 #endif
4322
4323 /* Loop through the string, looking for a place to start matching. */
4324 for (;;)
4325 {
4326 /* If the pattern is anchored,
4327 skip quickly past places we cannot match.
4328 We don't bother to treat startpos == 0 specially
4329 because that case doesn't repeat. */
4330 if (anchored_start && startpos > 0)
4331 {
4332 if (! ((startpos <= size1 ? string1[startpos - 1]
4333 : string2[startpos - size1 - 1])
4334 == '\n'))
4335 goto advance;
4336 }
4337
4338 /* If a fastmap is supplied, skip quickly over characters that
4339 cannot be the start of a match. If the pattern can match the
4340 null string, however, we don't need to skip characters; we want
4341 the first null string. */
4342 if (fastmap && startpos < total_size && !bufp->can_be_null)
4343 {
4344 register re_char *d;
4345 register re_wchar_t buf_ch;
4346
4347 d = POS_ADDR_VSTRING (startpos);
4348
4349 if (range > 0) /* Searching forwards. */
4350 {
4351 register int lim = 0;
4352 ssize_t irange = range;
4353
4354 if (startpos < size1 && startpos + range >= size1)
4355 lim = range - (size1 - startpos);
4356
4357 /* Written out as an if-else to avoid testing `translate'
4358 inside the loop. */
4359 if (RE_TRANSLATE_P (translate))
4360 {
4361 if (multibyte)
4362 while (range > lim)
4363 {
4364 int buf_charlen;
4365
4366 buf_ch = STRING_CHAR_AND_LENGTH (d, buf_charlen);
4367 buf_ch = RE_TRANSLATE (translate, buf_ch);
4368 if (fastmap[CHAR_LEADING_CODE (buf_ch)])
4369 break;
4370
4371 range -= buf_charlen;
4372 d += buf_charlen;
4373 }
4374 else
4375 while (range > lim)
4376 {
4377 register re_wchar_t ch, translated;
4378
4379 buf_ch = *d;
4380 ch = RE_CHAR_TO_MULTIBYTE (buf_ch);
4381 translated = RE_TRANSLATE (translate, ch);
4382 if (translated != ch
4383 && (ch = RE_CHAR_TO_UNIBYTE (translated)) >= 0)
4384 buf_ch = ch;
4385 if (fastmap[buf_ch])
4386 break;
4387 d++;
4388 range--;
4389 }
4390 }
4391 else
4392 {
4393 if (multibyte)
4394 while (range > lim)
4395 {
4396 int buf_charlen;
4397
4398 buf_ch = STRING_CHAR_AND_LENGTH (d, buf_charlen);
4399 if (fastmap[CHAR_LEADING_CODE (buf_ch)])
4400 break;
4401 range -= buf_charlen;
4402 d += buf_charlen;
4403 }
4404 else
4405 while (range > lim && !fastmap[*d])
4406 {
4407 d++;
4408 range--;
4409 }
4410 }
4411 startpos += irange - range;
4412 }
4413 else /* Searching backwards. */
4414 {
4415 if (multibyte)
4416 {
4417 buf_ch = STRING_CHAR (d);
4418 buf_ch = TRANSLATE (buf_ch);
4419 if (! fastmap[CHAR_LEADING_CODE (buf_ch)])
4420 goto advance;
4421 }
4422 else
4423 {
4424 register re_wchar_t ch, translated;
4425
4426 buf_ch = *d;
4427 ch = RE_CHAR_TO_MULTIBYTE (buf_ch);
4428 translated = TRANSLATE (ch);
4429 if (translated != ch
4430 && (ch = RE_CHAR_TO_UNIBYTE (translated)) >= 0)
4431 buf_ch = ch;
4432 if (! fastmap[TRANSLATE (buf_ch)])
4433 goto advance;
4434 }
4435 }
4436 }
4437
4438 /* If can't match the null string, and that's all we have left, fail. */
4439 if (range >= 0 && startpos == total_size && fastmap
4440 && !bufp->can_be_null)
4441 return -1;
4442
4443 val = re_match_2_internal (bufp, string1, size1, string2, size2,
4444 startpos, regs, stop);
4445
4446 if (val >= 0)
4447 return startpos;
4448
4449 if (val == -2)
4450 return -2;
4451
4452 advance:
4453 if (!range)
4454 break;
4455 else if (range > 0)
4456 {
4457 /* Update STARTPOS to the next character boundary. */
4458 if (multibyte)
4459 {
4460 re_char *p = POS_ADDR_VSTRING (startpos);
4461 int len = BYTES_BY_CHAR_HEAD (*p);
4462
4463 range -= len;
4464 if (range < 0)
4465 break;
4466 startpos += len;
4467 }
4468 else
4469 {
4470 range--;
4471 startpos++;
4472 }
4473 }
4474 else
4475 {
4476 range++;
4477 startpos--;
4478
4479 /* Update STARTPOS to the previous character boundary. */
4480 if (multibyte)
4481 {
4482 re_char *p = POS_ADDR_VSTRING (startpos) + 1;
4483 re_char *p0 = p;
4484 re_char *phead = HEAD_ADDR_VSTRING (startpos);
4485
4486 /* Find the head of multibyte form. */
4487 PREV_CHAR_BOUNDARY (p, phead);
4488 range += p0 - 1 - p;
4489 if (range > 0)
4490 break;
4491
4492 startpos -= p0 - 1 - p;
4493 }
4494 }
4495 }
4496 return -1;
4497 } /* re_search_2 */
4498 WEAK_ALIAS (__re_search_2, re_search_2)
4499 \f
4500 /* Declarations and macros for re_match_2. */
4501
4502 static int bcmp_translate (re_char *s1, re_char *s2,
4503 register ssize_t len,
4504 RE_TRANSLATE_TYPE translate,
4505 const int multibyte);
4506
4507 /* This converts PTR, a pointer into one of the search strings `string1'
4508 and `string2' into an offset from the beginning of that string. */
4509 #define POINTER_TO_OFFSET(ptr) \
4510 (FIRST_STRING_P (ptr) \
4511 ? (ptr) - string1 \
4512 : (ptr) - string2 + (ptrdiff_t) size1)
4513
4514 /* Call before fetching a character with *d. This switches over to
4515 string2 if necessary.
4516 Check re_match_2_internal for a discussion of why end_match_2 might
4517 not be within string2 (but be equal to end_match_1 instead). */
4518 #define PREFETCH() \
4519 while (d == dend) \
4520 { \
4521 /* End of string2 => fail. */ \
4522 if (dend == end_match_2) \
4523 goto fail; \
4524 /* End of string1 => advance to string2. */ \
4525 d = string2; \
4526 dend = end_match_2; \
4527 }
4528
4529 /* Call before fetching a char with *d if you already checked other limits.
4530 This is meant for use in lookahead operations like wordend, etc..
4531 where we might need to look at parts of the string that might be
4532 outside of the LIMITs (i.e past `stop'). */
4533 #define PREFETCH_NOLIMIT() \
4534 if (d == end1) \
4535 { \
4536 d = string2; \
4537 dend = end_match_2; \
4538 } \
4539
4540 /* Test if at very beginning or at very end of the virtual concatenation
4541 of `string1' and `string2'. If only one string, it's `string2'. */
4542 #define AT_STRINGS_BEG(d) ((d) == (size1 ? string1 : string2) || !size2)
4543 #define AT_STRINGS_END(d) ((d) == end2)
4544
4545 /* Disabled due to a compiler bug -- see comment at case wordbound */
4546
4547 /* The comment at case wordbound is following one, but we don't use
4548 AT_WORD_BOUNDARY anymore to support multibyte form.
4549
4550 The DEC Alpha C compiler 3.x generates incorrect code for the
4551 test WORDCHAR_P (d - 1) != WORDCHAR_P (d) in the expansion of
4552 AT_WORD_BOUNDARY, so this code is disabled. Expanding the
4553 macro and introducing temporary variables works around the bug. */
4554
4555 #if 0
4556 /* Test if D points to a character which is word-constituent. We have
4557 two special cases to check for: if past the end of string1, look at
4558 the first character in string2; and if before the beginning of
4559 string2, look at the last character in string1. */
4560 #define WORDCHAR_P(d) \
4561 (SYNTAX ((d) == end1 ? *string2 \
4562 : (d) == string2 - 1 ? *(end1 - 1) : *(d)) \
4563 == Sword)
4564
4565 /* Test if the character before D and the one at D differ with respect
4566 to being word-constituent. */
4567 #define AT_WORD_BOUNDARY(d) \
4568 (AT_STRINGS_BEG (d) || AT_STRINGS_END (d) \
4569 || WORDCHAR_P (d - 1) != WORDCHAR_P (d))
4570 #endif
4571
4572 /* Free everything we malloc. */
4573 #ifdef MATCH_MAY_ALLOCATE
4574 # define FREE_VAR(var) \
4575 do { \
4576 if (var) \
4577 { \
4578 REGEX_FREE (var); \
4579 var = NULL; \
4580 } \
4581 } while (0)
4582 # define FREE_VARIABLES() \
4583 do { \
4584 REGEX_FREE_STACK (fail_stack.stack); \
4585 FREE_VAR (regstart); \
4586 FREE_VAR (regend); \
4587 FREE_VAR (best_regstart); \
4588 FREE_VAR (best_regend); \
4589 } while (0)
4590 #else
4591 # define FREE_VARIABLES() ((void)0) /* Do nothing! But inhibit gcc warning. */
4592 #endif /* not MATCH_MAY_ALLOCATE */
4593
4594 \f
4595 /* Optimization routines. */
4596
4597 /* If the operation is a match against one or more chars,
4598 return a pointer to the next operation, else return NULL. */
4599 static re_char *
4600 skip_one_char (const_re_char *p)
4601 {
4602 switch (*p++)
4603 {
4604 case anychar:
4605 break;
4606
4607 case exactn:
4608 p += *p + 1;
4609 break;
4610
4611 case charset_not:
4612 case charset:
4613 if (CHARSET_RANGE_TABLE_EXISTS_P (p - 1))
4614 {
4615 int mcnt;
4616 p = CHARSET_RANGE_TABLE (p - 1);
4617 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4618 p = CHARSET_RANGE_TABLE_END (p, mcnt);
4619 }
4620 else
4621 p += 1 + CHARSET_BITMAP_SIZE (p - 1);
4622 break;
4623
4624 case syntaxspec:
4625 case notsyntaxspec:
4626 #ifdef emacs
4627 case categoryspec:
4628 case notcategoryspec:
4629 #endif /* emacs */
4630 p++;
4631 break;
4632
4633 default:
4634 p = NULL;
4635 }
4636 return p;
4637 }
4638
4639
4640 /* Jump over non-matching operations. */
4641 static re_char *
4642 skip_noops (const_re_char *p, const_re_char *pend)
4643 {
4644 int mcnt;
4645 while (p < pend)
4646 {
4647 switch (*p)
4648 {
4649 case start_memory:
4650 case stop_memory:
4651 p += 2; break;
4652 case no_op:
4653 p += 1; break;
4654 case jump:
4655 p += 1;
4656 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4657 p += mcnt;
4658 break;
4659 default:
4660 return p;
4661 }
4662 }
4663 assert (p == pend);
4664 return p;
4665 }
4666
4667 /* Non-zero if "p1 matches something" implies "p2 fails". */
4668 static int
4669 mutually_exclusive_p (struct re_pattern_buffer *bufp, const_re_char *p1,
4670 const_re_char *p2)
4671 {
4672 re_opcode_t op2;
4673 const boolean multibyte = RE_MULTIBYTE_P (bufp);
4674 unsigned char *pend = bufp->buffer + bufp->used;
4675
4676 assert (p1 >= bufp->buffer && p1 < pend
4677 && p2 >= bufp->buffer && p2 <= pend);
4678
4679 /* Skip over open/close-group commands.
4680 If what follows this loop is a ...+ construct,
4681 look at what begins its body, since we will have to
4682 match at least one of that. */
4683 p2 = skip_noops (p2, pend);
4684 /* The same skip can be done for p1, except that this function
4685 is only used in the case where p1 is a simple match operator. */
4686 /* p1 = skip_noops (p1, pend); */
4687
4688 assert (p1 >= bufp->buffer && p1 < pend
4689 && p2 >= bufp->buffer && p2 <= pend);
4690
4691 op2 = p2 == pend ? succeed : *p2;
4692
4693 switch (op2)
4694 {
4695 case succeed:
4696 case endbuf:
4697 /* If we're at the end of the pattern, we can change. */
4698 if (skip_one_char (p1))
4699 {
4700 DEBUG_PRINT (" End of pattern: fast loop.\n");
4701 return 1;
4702 }
4703 break;
4704
4705 case endline:
4706 case exactn:
4707 {
4708 register re_wchar_t c
4709 = (re_opcode_t) *p2 == endline ? '\n'
4710 : RE_STRING_CHAR (p2 + 2, multibyte);
4711
4712 if ((re_opcode_t) *p1 == exactn)
4713 {
4714 if (c != RE_STRING_CHAR (p1 + 2, multibyte))
4715 {
4716 DEBUG_PRINT (" '%c' != '%c' => fast loop.\n", c, p1[2]);
4717 return 1;
4718 }
4719 }
4720
4721 else if ((re_opcode_t) *p1 == charset
4722 || (re_opcode_t) *p1 == charset_not)
4723 {
4724 int not = (re_opcode_t) *p1 == charset_not;
4725
4726 /* Test if C is listed in charset (or charset_not)
4727 at `p1'. */
4728 if (! multibyte || IS_REAL_ASCII (c))
4729 {
4730 if (c < CHARSET_BITMAP_SIZE (p1) * BYTEWIDTH
4731 && p1[2 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH)))
4732 not = !not;
4733 }
4734 else if (CHARSET_RANGE_TABLE_EXISTS_P (p1))
4735 CHARSET_LOOKUP_RANGE_TABLE (not, c, p1);
4736
4737 /* `not' is equal to 1 if c would match, which means
4738 that we can't change to pop_failure_jump. */
4739 if (!not)
4740 {
4741 DEBUG_PRINT (" No match => fast loop.\n");
4742 return 1;
4743 }
4744 }
4745 else if ((re_opcode_t) *p1 == anychar
4746 && c == '\n')
4747 {
4748 DEBUG_PRINT (" . != \\n => fast loop.\n");
4749 return 1;
4750 }
4751 }
4752 break;
4753
4754 case charset:
4755 {
4756 if ((re_opcode_t) *p1 == exactn)
4757 /* Reuse the code above. */
4758 return mutually_exclusive_p (bufp, p2, p1);
4759
4760 /* It is hard to list up all the character in charset
4761 P2 if it includes multibyte character. Give up in
4762 such case. */
4763 else if (!multibyte || !CHARSET_RANGE_TABLE_EXISTS_P (p2))
4764 {
4765 /* Now, we are sure that P2 has no range table.
4766 So, for the size of bitmap in P2, `p2[1]' is
4767 enough. But P1 may have range table, so the
4768 size of bitmap table of P1 is extracted by
4769 using macro `CHARSET_BITMAP_SIZE'.
4770
4771 In a multibyte case, we know that all the character
4772 listed in P2 is ASCII. In a unibyte case, P1 has only a
4773 bitmap table. So, in both cases, it is enough to test
4774 only the bitmap table of P1. */
4775
4776 if ((re_opcode_t) *p1 == charset)
4777 {
4778 int idx;
4779 /* We win if the charset inside the loop
4780 has no overlap with the one after the loop. */
4781 for (idx = 0;
4782 (idx < (int) p2[1]
4783 && idx < CHARSET_BITMAP_SIZE (p1));
4784 idx++)
4785 if ((p2[2 + idx] & p1[2 + idx]) != 0)
4786 break;
4787
4788 if (idx == p2[1]
4789 || idx == CHARSET_BITMAP_SIZE (p1))
4790 {
4791 DEBUG_PRINT (" No match => fast loop.\n");
4792 return 1;
4793 }
4794 }
4795 else if ((re_opcode_t) *p1 == charset_not)
4796 {
4797 int idx;
4798 /* We win if the charset_not inside the loop lists
4799 every character listed in the charset after. */
4800 for (idx = 0; idx < (int) p2[1]; idx++)
4801 if (! (p2[2 + idx] == 0
4802 || (idx < CHARSET_BITMAP_SIZE (p1)
4803 && ((p2[2 + idx] & ~ p1[2 + idx]) == 0))))
4804 break;
4805
4806 if (idx == p2[1])
4807 {
4808 DEBUG_PRINT (" No match => fast loop.\n");
4809 return 1;
4810 }
4811 }
4812 }
4813 }
4814 break;
4815
4816 case charset_not:
4817 switch (*p1)
4818 {
4819 case exactn:
4820 case charset:
4821 /* Reuse the code above. */
4822 return mutually_exclusive_p (bufp, p2, p1);
4823 case charset_not:
4824 /* When we have two charset_not, it's very unlikely that
4825 they don't overlap. The union of the two sets of excluded
4826 chars should cover all possible chars, which, as a matter of
4827 fact, is virtually impossible in multibyte buffers. */
4828 break;
4829 }
4830 break;
4831
4832 case wordend:
4833 return ((re_opcode_t) *p1 == syntaxspec && p1[1] == Sword);
4834 case symend:
4835 return ((re_opcode_t) *p1 == syntaxspec
4836 && (p1[1] == Ssymbol || p1[1] == Sword));
4837 case notsyntaxspec:
4838 return ((re_opcode_t) *p1 == syntaxspec && p1[1] == p2[1]);
4839
4840 case wordbeg:
4841 return ((re_opcode_t) *p1 == notsyntaxspec && p1[1] == Sword);
4842 case symbeg:
4843 return ((re_opcode_t) *p1 == notsyntaxspec
4844 && (p1[1] == Ssymbol || p1[1] == Sword));
4845 case syntaxspec:
4846 return ((re_opcode_t) *p1 == notsyntaxspec && p1[1] == p2[1]);
4847
4848 case wordbound:
4849 return (((re_opcode_t) *p1 == notsyntaxspec
4850 || (re_opcode_t) *p1 == syntaxspec)
4851 && p1[1] == Sword);
4852
4853 #ifdef emacs
4854 case categoryspec:
4855 return ((re_opcode_t) *p1 == notcategoryspec && p1[1] == p2[1]);
4856 case notcategoryspec:
4857 return ((re_opcode_t) *p1 == categoryspec && p1[1] == p2[1]);
4858 #endif /* emacs */
4859
4860 default:
4861 ;
4862 }
4863
4864 /* Safe default. */
4865 return 0;
4866 }
4867
4868 \f
4869 /* Matching routines. */
4870
4871 #ifndef emacs /* Emacs never uses this. */
4872 /* re_match is like re_match_2 except it takes only a single string. */
4873
4874 regoff_t
4875 re_match (struct re_pattern_buffer *bufp, const char *string,
4876 size_t size, ssize_t pos, struct re_registers *regs)
4877 {
4878 regoff_t result = re_match_2_internal (bufp, NULL, 0, (re_char*) string,
4879 size, pos, regs, size);
4880 return result;
4881 }
4882 WEAK_ALIAS (__re_match, re_match)
4883 #endif /* not emacs */
4884
4885 #ifdef emacs
4886 /* In Emacs, this is the string or buffer in which we
4887 are matching. It is used for looking up syntax properties. */
4888 Lisp_Object re_match_object;
4889 #endif
4890
4891 /* re_match_2 matches the compiled pattern in BUFP against the
4892 the (virtual) concatenation of STRING1 and STRING2 (of length SIZE1
4893 and SIZE2, respectively). We start matching at POS, and stop
4894 matching at STOP.
4895
4896 If REGS is non-null and the `no_sub' field of BUFP is nonzero, we
4897 store offsets for the substring each group matched in REGS. See the
4898 documentation for exactly how many groups we fill.
4899
4900 We return -1 if no match, -2 if an internal error (such as the
4901 failure stack overflowing). Otherwise, we return the length of the
4902 matched substring. */
4903
4904 regoff_t
4905 re_match_2 (struct re_pattern_buffer *bufp, const char *string1,
4906 size_t size1, const char *string2, size_t size2, ssize_t pos,
4907 struct re_registers *regs, ssize_t stop)
4908 {
4909 regoff_t result;
4910
4911 #ifdef emacs
4912 ssize_t charpos;
4913 gl_state.object = re_match_object; /* Used by SYNTAX_TABLE_BYTE_TO_CHAR. */
4914 charpos = SYNTAX_TABLE_BYTE_TO_CHAR (POS_AS_IN_BUFFER (pos));
4915 SETUP_SYNTAX_TABLE_FOR_OBJECT (re_match_object, charpos, 1);
4916 #endif
4917
4918 result = re_match_2_internal (bufp, (re_char*) string1, size1,
4919 (re_char*) string2, size2,
4920 pos, regs, stop);
4921 return result;
4922 }
4923 WEAK_ALIAS (__re_match_2, re_match_2)
4924
4925
4926 /* This is a separate function so that we can force an alloca cleanup
4927 afterwards. */
4928 static regoff_t
4929 re_match_2_internal (struct re_pattern_buffer *bufp, const_re_char *string1,
4930 size_t size1, const_re_char *string2, size_t size2,
4931 ssize_t pos, struct re_registers *regs, ssize_t stop)
4932 {
4933 /* General temporaries. */
4934 int mcnt;
4935 size_t reg;
4936
4937 /* Just past the end of the corresponding string. */
4938 re_char *end1, *end2;
4939
4940 /* Pointers into string1 and string2, just past the last characters in
4941 each to consider matching. */
4942 re_char *end_match_1, *end_match_2;
4943
4944 /* Where we are in the data, and the end of the current string. */
4945 re_char *d, *dend;
4946
4947 /* Used sometimes to remember where we were before starting matching
4948 an operator so that we can go back in case of failure. This "atomic"
4949 behavior of matching opcodes is indispensable to the correctness
4950 of the on_failure_keep_string_jump optimization. */
4951 re_char *dfail;
4952
4953 /* Where we are in the pattern, and the end of the pattern. */
4954 re_char *p = bufp->buffer;
4955 re_char *pend = p + bufp->used;
4956
4957 /* We use this to map every character in the string. */
4958 RE_TRANSLATE_TYPE translate = bufp->translate;
4959
4960 /* Nonzero if BUFP is setup from a multibyte regex. */
4961 const boolean multibyte = RE_MULTIBYTE_P (bufp);
4962
4963 /* Nonzero if STRING1/STRING2 are multibyte. */
4964 const boolean target_multibyte = RE_TARGET_MULTIBYTE_P (bufp);
4965
4966 /* Failure point stack. Each place that can handle a failure further
4967 down the line pushes a failure point on this stack. It consists of
4968 regstart, and regend for all registers corresponding to
4969 the subexpressions we're currently inside, plus the number of such
4970 registers, and, finally, two char *'s. The first char * is where
4971 to resume scanning the pattern; the second one is where to resume
4972 scanning the strings. */
4973 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
4974 fail_stack_type fail_stack;
4975 #endif
4976 #ifdef DEBUG_COMPILES_ARGUMENTS
4977 unsigned nfailure_points_pushed = 0, nfailure_points_popped = 0;
4978 #endif
4979
4980 #if defined REL_ALLOC && defined REGEX_MALLOC
4981 /* This holds the pointer to the failure stack, when
4982 it is allocated relocatably. */
4983 fail_stack_elt_t *failure_stack_ptr;
4984 #endif
4985
4986 /* We fill all the registers internally, independent of what we
4987 return, for use in backreferences. The number here includes
4988 an element for register zero. */
4989 size_t num_regs = bufp->re_nsub + 1;
4990
4991 /* Information on the contents of registers. These are pointers into
4992 the input strings; they record just what was matched (on this
4993 attempt) by a subexpression part of the pattern, that is, the
4994 regnum-th regstart pointer points to where in the pattern we began
4995 matching and the regnum-th regend points to right after where we
4996 stopped matching the regnum-th subexpression. (The zeroth register
4997 keeps track of what the whole pattern matches.) */
4998 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
4999 re_char **regstart, **regend;
5000 #endif
5001
5002 /* The following record the register info as found in the above
5003 variables when we find a match better than any we've seen before.
5004 This happens as we backtrack through the failure points, which in
5005 turn happens only if we have not yet matched the entire string. */
5006 unsigned best_regs_set = false;
5007 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
5008 re_char **best_regstart, **best_regend;
5009 #endif
5010
5011 /* Logically, this is `best_regend[0]'. But we don't want to have to
5012 allocate space for that if we're not allocating space for anything
5013 else (see below). Also, we never need info about register 0 for
5014 any of the other register vectors, and it seems rather a kludge to
5015 treat `best_regend' differently than the rest. So we keep track of
5016 the end of the best match so far in a separate variable. We
5017 initialize this to NULL so that when we backtrack the first time
5018 and need to test it, it's not garbage. */
5019 re_char *match_end = NULL;
5020
5021 #ifdef DEBUG_COMPILES_ARGUMENTS
5022 /* Counts the total number of registers pushed. */
5023 unsigned num_regs_pushed = 0;
5024 #endif
5025
5026 DEBUG_PRINT ("\n\nEntering re_match_2.\n");
5027
5028 INIT_FAIL_STACK ();
5029
5030 #ifdef MATCH_MAY_ALLOCATE
5031 /* Do not bother to initialize all the register variables if there are
5032 no groups in the pattern, as it takes a fair amount of time. If
5033 there are groups, we include space for register 0 (the whole
5034 pattern), even though we never use it, since it simplifies the
5035 array indexing. We should fix this. */
5036 if (bufp->re_nsub)
5037 {
5038 regstart = REGEX_TALLOC (num_regs, re_char *);
5039 regend = REGEX_TALLOC (num_regs, re_char *);
5040 best_regstart = REGEX_TALLOC (num_regs, re_char *);
5041 best_regend = REGEX_TALLOC (num_regs, re_char *);
5042
5043 if (!(regstart && regend && best_regstart && best_regend))
5044 {
5045 FREE_VARIABLES ();
5046 return -2;
5047 }
5048 }
5049 else
5050 {
5051 /* We must initialize all our variables to NULL, so that
5052 `FREE_VARIABLES' doesn't try to free them. */
5053 regstart = regend = best_regstart = best_regend = NULL;
5054 }
5055 #endif /* MATCH_MAY_ALLOCATE */
5056
5057 /* The starting position is bogus. */
5058 if (pos < 0 || pos > size1 + size2)
5059 {
5060 FREE_VARIABLES ();
5061 return -1;
5062 }
5063
5064 /* Initialize subexpression text positions to -1 to mark ones that no
5065 start_memory/stop_memory has been seen for. Also initialize the
5066 register information struct. */
5067 for (reg = 1; reg < num_regs; reg++)
5068 regstart[reg] = regend[reg] = NULL;
5069
5070 /* We move `string1' into `string2' if the latter's empty -- but not if
5071 `string1' is null. */
5072 if (size2 == 0 && string1 != NULL)
5073 {
5074 string2 = string1;
5075 size2 = size1;
5076 string1 = 0;
5077 size1 = 0;
5078 }
5079 end1 = string1 + size1;
5080 end2 = string2 + size2;
5081
5082 /* `p' scans through the pattern as `d' scans through the data.
5083 `dend' is the end of the input string that `d' points within. `d'
5084 is advanced into the following input string whenever necessary, but
5085 this happens before fetching; therefore, at the beginning of the
5086 loop, `d' can be pointing at the end of a string, but it cannot
5087 equal `string2'. */
5088 if (pos >= size1)
5089 {
5090 /* Only match within string2. */
5091 d = string2 + pos - size1;
5092 dend = end_match_2 = string2 + stop - size1;
5093 end_match_1 = end1; /* Just to give it a value. */
5094 }
5095 else
5096 {
5097 if (stop < size1)
5098 {
5099 /* Only match within string1. */
5100 end_match_1 = string1 + stop;
5101 /* BEWARE!
5102 When we reach end_match_1, PREFETCH normally switches to string2.
5103 But in the present case, this means that just doing a PREFETCH
5104 makes us jump from `stop' to `gap' within the string.
5105 What we really want here is for the search to stop as
5106 soon as we hit end_match_1. That's why we set end_match_2
5107 to end_match_1 (since PREFETCH fails as soon as we hit
5108 end_match_2). */
5109 end_match_2 = end_match_1;
5110 }
5111 else
5112 { /* It's important to use this code when stop == size so that
5113 moving `d' from end1 to string2 will not prevent the d == dend
5114 check from catching the end of string. */
5115 end_match_1 = end1;
5116 end_match_2 = string2 + stop - size1;
5117 }
5118 d = string1 + pos;
5119 dend = end_match_1;
5120 }
5121
5122 DEBUG_PRINT ("The compiled pattern is: ");
5123 DEBUG_PRINT_COMPILED_PATTERN (bufp, p, pend);
5124 DEBUG_PRINT ("The string to match is: `");
5125 DEBUG_PRINT_DOUBLE_STRING (d, string1, size1, string2, size2);
5126 DEBUG_PRINT ("'\n");
5127
5128 /* This loops over pattern commands. It exits by returning from the
5129 function if the match is complete, or it drops through if the match
5130 fails at this starting point in the input data. */
5131 for (;;)
5132 {
5133 DEBUG_PRINT ("\n%p: ", p);
5134
5135 if (p == pend)
5136 {
5137 ptrdiff_t dcnt;
5138
5139 /* End of pattern means we might have succeeded. */
5140 DEBUG_PRINT ("end of pattern ... ");
5141
5142 /* If we haven't matched the entire string, and we want the
5143 longest match, try backtracking. */
5144 if (d != end_match_2)
5145 {
5146 /* 1 if this match ends in the same string (string1 or string2)
5147 as the best previous match. */
5148 boolean same_str_p = (FIRST_STRING_P (match_end)
5149 == FIRST_STRING_P (d));
5150 /* 1 if this match is the best seen so far. */
5151 boolean best_match_p;
5152
5153 /* AIX compiler got confused when this was combined
5154 with the previous declaration. */
5155 if (same_str_p)
5156 best_match_p = d > match_end;
5157 else
5158 best_match_p = !FIRST_STRING_P (d);
5159
5160 DEBUG_PRINT ("backtracking.\n");
5161
5162 if (!FAIL_STACK_EMPTY ())
5163 { /* More failure points to try. */
5164
5165 /* If exceeds best match so far, save it. */
5166 if (!best_regs_set || best_match_p)
5167 {
5168 best_regs_set = true;
5169 match_end = d;
5170
5171 DEBUG_PRINT ("\nSAVING match as best so far.\n");
5172
5173 for (reg = 1; reg < num_regs; reg++)
5174 {
5175 best_regstart[reg] = regstart[reg];
5176 best_regend[reg] = regend[reg];
5177 }
5178 }
5179 goto fail;
5180 }
5181
5182 /* If no failure points, don't restore garbage. And if
5183 last match is real best match, don't restore second
5184 best one. */
5185 else if (best_regs_set && !best_match_p)
5186 {
5187 restore_best_regs:
5188 /* Restore best match. It may happen that `dend ==
5189 end_match_1' while the restored d is in string2.
5190 For example, the pattern `x.*y.*z' against the
5191 strings `x-' and `y-z-', if the two strings are
5192 not consecutive in memory. */
5193 DEBUG_PRINT ("Restoring best registers.\n");
5194
5195 d = match_end;
5196 dend = ((d >= string1 && d <= end1)
5197 ? end_match_1 : end_match_2);
5198
5199 for (reg = 1; reg < num_regs; reg++)
5200 {
5201 regstart[reg] = best_regstart[reg];
5202 regend[reg] = best_regend[reg];
5203 }
5204 }
5205 } /* d != end_match_2 */
5206
5207 succeed_label:
5208 DEBUG_PRINT ("Accepting match.\n");
5209
5210 /* If caller wants register contents data back, do it. */
5211 if (regs && !bufp->no_sub)
5212 {
5213 /* Have the register data arrays been allocated? */
5214 if (bufp->regs_allocated == REGS_UNALLOCATED)
5215 { /* No. So allocate them with malloc. We need one
5216 extra element beyond `num_regs' for the `-1' marker
5217 GNU code uses. */
5218 regs->num_regs = MAX (RE_NREGS, num_regs + 1);
5219 regs->start = TALLOC (regs->num_regs, regoff_t);
5220 regs->end = TALLOC (regs->num_regs, regoff_t);
5221 if (regs->start == NULL || regs->end == NULL)
5222 {
5223 FREE_VARIABLES ();
5224 return -2;
5225 }
5226 bufp->regs_allocated = REGS_REALLOCATE;
5227 }
5228 else if (bufp->regs_allocated == REGS_REALLOCATE)
5229 { /* Yes. If we need more elements than were already
5230 allocated, reallocate them. If we need fewer, just
5231 leave it alone. */
5232 if (regs->num_regs < num_regs + 1)
5233 {
5234 regs->num_regs = num_regs + 1;
5235 RETALLOC (regs->start, regs->num_regs, regoff_t);
5236 RETALLOC (regs->end, regs->num_regs, regoff_t);
5237 if (regs->start == NULL || regs->end == NULL)
5238 {
5239 FREE_VARIABLES ();
5240 return -2;
5241 }
5242 }
5243 }
5244 else
5245 {
5246 /* These braces fend off a "empty body in an else-statement"
5247 warning under GCC when assert expands to nothing. */
5248 assert (bufp->regs_allocated == REGS_FIXED);
5249 }
5250
5251 /* Convert the pointer data in `regstart' and `regend' to
5252 indices. Register zero has to be set differently,
5253 since we haven't kept track of any info for it. */
5254 if (regs->num_regs > 0)
5255 {
5256 regs->start[0] = pos;
5257 regs->end[0] = POINTER_TO_OFFSET (d);
5258 }
5259
5260 /* Go through the first `min (num_regs, regs->num_regs)'
5261 registers, since that is all we initialized. */
5262 for (reg = 1; reg < MIN (num_regs, regs->num_regs); reg++)
5263 {
5264 if (REG_UNSET (regstart[reg]) || REG_UNSET (regend[reg]))
5265 regs->start[reg] = regs->end[reg] = -1;
5266 else
5267 {
5268 regs->start[reg] = POINTER_TO_OFFSET (regstart[reg]);
5269 regs->end[reg] = POINTER_TO_OFFSET (regend[reg]);
5270 }
5271 }
5272
5273 /* If the regs structure we return has more elements than
5274 were in the pattern, set the extra elements to -1. If
5275 we (re)allocated the registers, this is the case,
5276 because we always allocate enough to have at least one
5277 -1 at the end. */
5278 for (reg = num_regs; reg < regs->num_regs; reg++)
5279 regs->start[reg] = regs->end[reg] = -1;
5280 } /* regs && !bufp->no_sub */
5281
5282 DEBUG_PRINT ("%u failure points pushed, %u popped (%u remain).\n",
5283 nfailure_points_pushed, nfailure_points_popped,
5284 nfailure_points_pushed - nfailure_points_popped);
5285 DEBUG_PRINT ("%u registers pushed.\n", num_regs_pushed);
5286
5287 dcnt = POINTER_TO_OFFSET (d) - pos;
5288
5289 DEBUG_PRINT ("Returning %td from re_match_2.\n", dcnt);
5290
5291 FREE_VARIABLES ();
5292 return dcnt;
5293 }
5294
5295 /* Otherwise match next pattern command. */
5296 switch (*p++)
5297 {
5298 /* Ignore these. Used to ignore the n of succeed_n's which
5299 currently have n == 0. */
5300 case no_op:
5301 DEBUG_PRINT ("EXECUTING no_op.\n");
5302 break;
5303
5304 case succeed:
5305 DEBUG_PRINT ("EXECUTING succeed.\n");
5306 goto succeed_label;
5307
5308 /* Match the next n pattern characters exactly. The following
5309 byte in the pattern defines n, and the n bytes after that
5310 are the characters to match. */
5311 case exactn:
5312 mcnt = *p++;
5313 DEBUG_PRINT ("EXECUTING exactn %d.\n", mcnt);
5314
5315 /* Remember the start point to rollback upon failure. */
5316 dfail = d;
5317
5318 #ifndef emacs
5319 /* This is written out as an if-else so we don't waste time
5320 testing `translate' inside the loop. */
5321 if (RE_TRANSLATE_P (translate))
5322 do
5323 {
5324 PREFETCH ();
5325 if (RE_TRANSLATE (translate, *d) != *p++)
5326 {
5327 d = dfail;
5328 goto fail;
5329 }
5330 d++;
5331 }
5332 while (--mcnt);
5333 else
5334 do
5335 {
5336 PREFETCH ();
5337 if (*d++ != *p++)
5338 {
5339 d = dfail;
5340 goto fail;
5341 }
5342 }
5343 while (--mcnt);
5344 #else /* emacs */
5345 /* The cost of testing `translate' is comparatively small. */
5346 if (target_multibyte)
5347 do
5348 {
5349 int pat_charlen, buf_charlen;
5350 int pat_ch, buf_ch;
5351
5352 PREFETCH ();
5353 if (multibyte)
5354 pat_ch = STRING_CHAR_AND_LENGTH (p, pat_charlen);
5355 else
5356 {
5357 pat_ch = RE_CHAR_TO_MULTIBYTE (*p);
5358 pat_charlen = 1;
5359 }
5360 buf_ch = STRING_CHAR_AND_LENGTH (d, buf_charlen);
5361
5362 if (TRANSLATE (buf_ch) != pat_ch)
5363 {
5364 d = dfail;
5365 goto fail;
5366 }
5367
5368 p += pat_charlen;
5369 d += buf_charlen;
5370 mcnt -= pat_charlen;
5371 }
5372 while (mcnt > 0);
5373 else
5374 do
5375 {
5376 int pat_charlen;
5377 int pat_ch, buf_ch;
5378
5379 PREFETCH ();
5380 if (multibyte)
5381 {
5382 pat_ch = STRING_CHAR_AND_LENGTH (p, pat_charlen);
5383 pat_ch = RE_CHAR_TO_UNIBYTE (pat_ch);
5384 }
5385 else
5386 {
5387 pat_ch = *p;
5388 pat_charlen = 1;
5389 }
5390 buf_ch = RE_CHAR_TO_MULTIBYTE (*d);
5391 if (! CHAR_BYTE8_P (buf_ch))
5392 {
5393 buf_ch = TRANSLATE (buf_ch);
5394 buf_ch = RE_CHAR_TO_UNIBYTE (buf_ch);
5395 if (buf_ch < 0)
5396 buf_ch = *d;
5397 }
5398 else
5399 buf_ch = *d;
5400 if (buf_ch != pat_ch)
5401 {
5402 d = dfail;
5403 goto fail;
5404 }
5405 p += pat_charlen;
5406 d++;
5407 }
5408 while (--mcnt);
5409 #endif
5410 break;
5411
5412
5413 /* Match any character except possibly a newline or a null. */
5414 case anychar:
5415 {
5416 int buf_charlen;
5417 re_wchar_t buf_ch;
5418
5419 DEBUG_PRINT ("EXECUTING anychar.\n");
5420
5421 PREFETCH ();
5422 buf_ch = RE_STRING_CHAR_AND_LENGTH (d, buf_charlen,
5423 target_multibyte);
5424 buf_ch = TRANSLATE (buf_ch);
5425
5426 if ((!(bufp->syntax & RE_DOT_NEWLINE)
5427 && buf_ch == '\n')
5428 || ((bufp->syntax & RE_DOT_NOT_NULL)
5429 && buf_ch == '\000'))
5430 goto fail;
5431
5432 DEBUG_PRINT (" Matched `%d'.\n", *d);
5433 d += buf_charlen;
5434 }
5435 break;
5436
5437
5438 case charset:
5439 case charset_not:
5440 {
5441 register unsigned int c;
5442 boolean not = (re_opcode_t) *(p - 1) == charset_not;
5443 int len;
5444
5445 /* Start of actual range_table, or end of bitmap if there is no
5446 range table. */
5447 re_char *range_table IF_LINT (= NULL);
5448
5449 /* Nonzero if there is a range table. */
5450 int range_table_exists;
5451
5452 /* Number of ranges of range table. This is not included
5453 in the initial byte-length of the command. */
5454 int count = 0;
5455
5456 /* Whether matching against a unibyte character. */
5457 boolean unibyte_char = false;
5458
5459 DEBUG_PRINT ("EXECUTING charset%s.\n", not ? "_not" : "");
5460
5461 range_table_exists = CHARSET_RANGE_TABLE_EXISTS_P (&p[-1]);
5462
5463 if (range_table_exists)
5464 {
5465 range_table = CHARSET_RANGE_TABLE (&p[-1]); /* Past the bitmap. */
5466 EXTRACT_NUMBER_AND_INCR (count, range_table);
5467 }
5468
5469 PREFETCH ();
5470 c = RE_STRING_CHAR_AND_LENGTH (d, len, target_multibyte);
5471 if (target_multibyte)
5472 {
5473 int c1;
5474
5475 c = TRANSLATE (c);
5476 c1 = RE_CHAR_TO_UNIBYTE (c);
5477 if (c1 >= 0)
5478 {
5479 unibyte_char = true;
5480 c = c1;
5481 }
5482 }
5483 else
5484 {
5485 int c1 = RE_CHAR_TO_MULTIBYTE (c);
5486
5487 if (! CHAR_BYTE8_P (c1))
5488 {
5489 c1 = TRANSLATE (c1);
5490 c1 = RE_CHAR_TO_UNIBYTE (c1);
5491 if (c1 >= 0)
5492 {
5493 unibyte_char = true;
5494 c = c1;
5495 }
5496 }
5497 else
5498 unibyte_char = true;
5499 }
5500
5501 if (unibyte_char && c < (1 << BYTEWIDTH))
5502 { /* Lookup bitmap. */
5503 /* Cast to `unsigned' instead of `unsigned char' in
5504 case the bit list is a full 32 bytes long. */
5505 if (c < (unsigned) (CHARSET_BITMAP_SIZE (&p[-1]) * BYTEWIDTH)
5506 && p[1 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH)))
5507 not = !not;
5508 }
5509 #ifdef emacs
5510 else if (range_table_exists)
5511 {
5512 int class_bits = CHARSET_RANGE_TABLE_BITS (&p[-1]);
5513
5514 if ( (class_bits & BIT_LOWER && ISLOWER (c))
5515 | (class_bits & BIT_MULTIBYTE)
5516 | (class_bits & BIT_PUNCT && ISPUNCT (c))
5517 | (class_bits & BIT_SPACE && ISSPACE (c))
5518 | (class_bits & BIT_UPPER && ISUPPER (c))
5519 | (class_bits & BIT_WORD && ISWORD (c)))
5520 not = !not;
5521 else
5522 CHARSET_LOOKUP_RANGE_TABLE_RAW (not, c, range_table, count);
5523 }
5524 #endif /* emacs */
5525
5526 if (range_table_exists)
5527 p = CHARSET_RANGE_TABLE_END (range_table, count);
5528 else
5529 p += CHARSET_BITMAP_SIZE (&p[-1]) + 1;
5530
5531 if (!not) goto fail;
5532
5533 d += len;
5534 }
5535 break;
5536
5537
5538 /* The beginning of a group is represented by start_memory.
5539 The argument is the register number. The text
5540 matched within the group is recorded (in the internal
5541 registers data structure) under the register number. */
5542 case start_memory:
5543 DEBUG_PRINT ("EXECUTING start_memory %d:\n", *p);
5544
5545 /* In case we need to undo this operation (via backtracking). */
5546 PUSH_FAILURE_REG (*p);
5547
5548 regstart[*p] = d;
5549 regend[*p] = NULL; /* probably unnecessary. -sm */
5550 DEBUG_PRINT (" regstart: %td\n", POINTER_TO_OFFSET (regstart[*p]));
5551
5552 /* Move past the register number and inner group count. */
5553 p += 1;
5554 break;
5555
5556
5557 /* The stop_memory opcode represents the end of a group. Its
5558 argument is the same as start_memory's: the register number. */
5559 case stop_memory:
5560 DEBUG_PRINT ("EXECUTING stop_memory %d:\n", *p);
5561
5562 assert (!REG_UNSET (regstart[*p]));
5563 /* Strictly speaking, there should be code such as:
5564
5565 assert (REG_UNSET (regend[*p]));
5566 PUSH_FAILURE_REGSTOP ((unsigned int)*p);
5567
5568 But the only info to be pushed is regend[*p] and it is known to
5569 be UNSET, so there really isn't anything to push.
5570 Not pushing anything, on the other hand deprives us from the
5571 guarantee that regend[*p] is UNSET since undoing this operation
5572 will not reset its value properly. This is not important since
5573 the value will only be read on the next start_memory or at
5574 the very end and both events can only happen if this stop_memory
5575 is *not* undone. */
5576
5577 regend[*p] = d;
5578 DEBUG_PRINT (" regend: %td\n", POINTER_TO_OFFSET (regend[*p]));
5579
5580 /* Move past the register number and the inner group count. */
5581 p += 1;
5582 break;
5583
5584
5585 /* \<digit> has been turned into a `duplicate' command which is
5586 followed by the numeric value of <digit> as the register number. */
5587 case duplicate:
5588 {
5589 register re_char *d2, *dend2;
5590 int regno = *p++; /* Get which register to match against. */
5591 DEBUG_PRINT ("EXECUTING duplicate %d.\n", regno);
5592
5593 /* Can't back reference a group which we've never matched. */
5594 if (REG_UNSET (regstart[regno]) || REG_UNSET (regend[regno]))
5595 goto fail;
5596
5597 /* Where in input to try to start matching. */
5598 d2 = regstart[regno];
5599
5600 /* Remember the start point to rollback upon failure. */
5601 dfail = d;
5602
5603 /* Where to stop matching; if both the place to start and
5604 the place to stop matching are in the same string, then
5605 set to the place to stop, otherwise, for now have to use
5606 the end of the first string. */
5607
5608 dend2 = ((FIRST_STRING_P (regstart[regno])
5609 == FIRST_STRING_P (regend[regno]))
5610 ? regend[regno] : end_match_1);
5611 for (;;)
5612 {
5613 ptrdiff_t dcnt;
5614
5615 /* If necessary, advance to next segment in register
5616 contents. */
5617 while (d2 == dend2)
5618 {
5619 if (dend2 == end_match_2) break;
5620 if (dend2 == regend[regno]) break;
5621
5622 /* End of string1 => advance to string2. */
5623 d2 = string2;
5624 dend2 = regend[regno];
5625 }
5626 /* At end of register contents => success */
5627 if (d2 == dend2) break;
5628
5629 /* If necessary, advance to next segment in data. */
5630 PREFETCH ();
5631
5632 /* How many characters left in this segment to match. */
5633 dcnt = dend - d;
5634
5635 /* Want how many consecutive characters we can match in
5636 one shot, so, if necessary, adjust the count. */
5637 if (dcnt > dend2 - d2)
5638 dcnt = dend2 - d2;
5639
5640 /* Compare that many; failure if mismatch, else move
5641 past them. */
5642 if (RE_TRANSLATE_P (translate)
5643 ? bcmp_translate (d, d2, dcnt, translate, target_multibyte)
5644 : memcmp (d, d2, dcnt))
5645 {
5646 d = dfail;
5647 goto fail;
5648 }
5649 d += dcnt, d2 += dcnt;
5650 }
5651 }
5652 break;
5653
5654
5655 /* begline matches the empty string at the beginning of the string
5656 (unless `not_bol' is set in `bufp'), and after newlines. */
5657 case begline:
5658 DEBUG_PRINT ("EXECUTING begline.\n");
5659
5660 if (AT_STRINGS_BEG (d))
5661 {
5662 if (!bufp->not_bol) break;
5663 }
5664 else
5665 {
5666 unsigned c;
5667 GET_CHAR_BEFORE_2 (c, d, string1, end1, string2, end2);
5668 if (c == '\n')
5669 break;
5670 }
5671 /* In all other cases, we fail. */
5672 goto fail;
5673
5674
5675 /* endline is the dual of begline. */
5676 case endline:
5677 DEBUG_PRINT ("EXECUTING endline.\n");
5678
5679 if (AT_STRINGS_END (d))
5680 {
5681 if (!bufp->not_eol) break;
5682 }
5683 else
5684 {
5685 PREFETCH_NOLIMIT ();
5686 if (*d == '\n')
5687 break;
5688 }
5689 goto fail;
5690
5691
5692 /* Match at the very beginning of the data. */
5693 case begbuf:
5694 DEBUG_PRINT ("EXECUTING begbuf.\n");
5695 if (AT_STRINGS_BEG (d))
5696 break;
5697 goto fail;
5698
5699
5700 /* Match at the very end of the data. */
5701 case endbuf:
5702 DEBUG_PRINT ("EXECUTING endbuf.\n");
5703 if (AT_STRINGS_END (d))
5704 break;
5705 goto fail;
5706
5707
5708 /* on_failure_keep_string_jump is used to optimize `.*\n'. It
5709 pushes NULL as the value for the string on the stack. Then
5710 `POP_FAILURE_POINT' will keep the current value for the
5711 string, instead of restoring it. To see why, consider
5712 matching `foo\nbar' against `.*\n'. The .* matches the foo;
5713 then the . fails against the \n. But the next thing we want
5714 to do is match the \n against the \n; if we restored the
5715 string value, we would be back at the foo.
5716
5717 Because this is used only in specific cases, we don't need to
5718 check all the things that `on_failure_jump' does, to make
5719 sure the right things get saved on the stack. Hence we don't
5720 share its code. The only reason to push anything on the
5721 stack at all is that otherwise we would have to change
5722 `anychar's code to do something besides goto fail in this
5723 case; that seems worse than this. */
5724 case on_failure_keep_string_jump:
5725 EXTRACT_NUMBER_AND_INCR (mcnt, p);
5726 DEBUG_PRINT ("EXECUTING on_failure_keep_string_jump %d (to %p):\n",
5727 mcnt, p + mcnt);
5728
5729 PUSH_FAILURE_POINT (p - 3, NULL);
5730 break;
5731
5732 /* A nasty loop is introduced by the non-greedy *? and +?.
5733 With such loops, the stack only ever contains one failure point
5734 at a time, so that a plain on_failure_jump_loop kind of
5735 cycle detection cannot work. Worse yet, such a detection
5736 can not only fail to detect a cycle, but it can also wrongly
5737 detect a cycle (between different instantiations of the same
5738 loop).
5739 So the method used for those nasty loops is a little different:
5740 We use a special cycle-detection-stack-frame which is pushed
5741 when the on_failure_jump_nastyloop failure-point is *popped*.
5742 This special frame thus marks the beginning of one iteration
5743 through the loop and we can hence easily check right here
5744 whether something matched between the beginning and the end of
5745 the loop. */
5746 case on_failure_jump_nastyloop:
5747 EXTRACT_NUMBER_AND_INCR (mcnt, p);
5748 DEBUG_PRINT ("EXECUTING on_failure_jump_nastyloop %d (to %p):\n",
5749 mcnt, p + mcnt);
5750
5751 assert ((re_opcode_t)p[-4] == no_op);
5752 {
5753 int cycle = 0;
5754 CHECK_INFINITE_LOOP (p - 4, d);
5755 if (!cycle)
5756 /* If there's a cycle, just continue without pushing
5757 this failure point. The failure point is the "try again"
5758 option, which shouldn't be tried.
5759 We want (x?)*?y\1z to match both xxyz and xxyxz. */
5760 PUSH_FAILURE_POINT (p - 3, d);
5761 }
5762 break;
5763
5764 /* Simple loop detecting on_failure_jump: just check on the
5765 failure stack if the same spot was already hit earlier. */
5766 case on_failure_jump_loop:
5767 on_failure:
5768 EXTRACT_NUMBER_AND_INCR (mcnt, p);
5769 DEBUG_PRINT ("EXECUTING on_failure_jump_loop %d (to %p):\n",
5770 mcnt, p + mcnt);
5771 {
5772 int cycle = 0;
5773 CHECK_INFINITE_LOOP (p - 3, d);
5774 if (cycle)
5775 /* If there's a cycle, get out of the loop, as if the matching
5776 had failed. We used to just `goto fail' here, but that was
5777 aborting the search a bit too early: we want to keep the
5778 empty-loop-match and keep matching after the loop.
5779 We want (x?)*y\1z to match both xxyz and xxyxz. */
5780 p += mcnt;
5781 else
5782 PUSH_FAILURE_POINT (p - 3, d);
5783 }
5784 break;
5785
5786
5787 /* Uses of on_failure_jump:
5788
5789 Each alternative starts with an on_failure_jump that points
5790 to the beginning of the next alternative. Each alternative
5791 except the last ends with a jump that in effect jumps past
5792 the rest of the alternatives. (They really jump to the
5793 ending jump of the following alternative, because tensioning
5794 these jumps is a hassle.)
5795
5796 Repeats start with an on_failure_jump that points past both
5797 the repetition text and either the following jump or
5798 pop_failure_jump back to this on_failure_jump. */
5799 case on_failure_jump:
5800 EXTRACT_NUMBER_AND_INCR (mcnt, p);
5801 DEBUG_PRINT ("EXECUTING on_failure_jump %d (to %p):\n",
5802 mcnt, p + mcnt);
5803
5804 PUSH_FAILURE_POINT (p -3, d);
5805 break;
5806
5807 /* This operation is used for greedy *.
5808 Compare the beginning of the repeat with what in the
5809 pattern follows its end. If we can establish that there
5810 is nothing that they would both match, i.e., that we
5811 would have to backtrack because of (as in, e.g., `a*a')
5812 then we can use a non-backtracking loop based on
5813 on_failure_keep_string_jump instead of on_failure_jump. */
5814 case on_failure_jump_smart:
5815 EXTRACT_NUMBER_AND_INCR (mcnt, p);
5816 DEBUG_PRINT ("EXECUTING on_failure_jump_smart %d (to %p).\n",
5817 mcnt, p + mcnt);
5818 {
5819 re_char *p1 = p; /* Next operation. */
5820 /* Here, we discard `const', making re_match non-reentrant. */
5821 unsigned char *p2 = (unsigned char*) p + mcnt; /* Jump dest. */
5822 unsigned char *p3 = (unsigned char*) p - 3; /* opcode location. */
5823
5824 p -= 3; /* Reset so that we will re-execute the
5825 instruction once it's been changed. */
5826
5827 EXTRACT_NUMBER (mcnt, p2 - 2);
5828
5829 /* Ensure this is a indeed the trivial kind of loop
5830 we are expecting. */
5831 assert (skip_one_char (p1) == p2 - 3);
5832 assert ((re_opcode_t) p2[-3] == jump && p2 + mcnt == p);
5833 DEBUG_STATEMENT (debug += 2);
5834 if (mutually_exclusive_p (bufp, p1, p2))
5835 {
5836 /* Use a fast `on_failure_keep_string_jump' loop. */
5837 DEBUG_PRINT (" smart exclusive => fast loop.\n");
5838 *p3 = (unsigned char) on_failure_keep_string_jump;
5839 STORE_NUMBER (p2 - 2, mcnt + 3);
5840 }
5841 else
5842 {
5843 /* Default to a safe `on_failure_jump' loop. */
5844 DEBUG_PRINT (" smart default => slow loop.\n");
5845 *p3 = (unsigned char) on_failure_jump;
5846 }
5847 DEBUG_STATEMENT (debug -= 2);
5848 }
5849 break;
5850
5851 /* Unconditionally jump (without popping any failure points). */
5852 case jump:
5853 unconditional_jump:
5854 IMMEDIATE_QUIT_CHECK;
5855 EXTRACT_NUMBER_AND_INCR (mcnt, p); /* Get the amount to jump. */
5856 DEBUG_PRINT ("EXECUTING jump %d ", mcnt);
5857 p += mcnt; /* Do the jump. */
5858 DEBUG_PRINT ("(to %p).\n", p);
5859 break;
5860
5861
5862 /* Have to succeed matching what follows at least n times.
5863 After that, handle like `on_failure_jump'. */
5864 case succeed_n:
5865 /* Signedness doesn't matter since we only compare MCNT to 0. */
5866 EXTRACT_NUMBER (mcnt, p + 2);
5867 DEBUG_PRINT ("EXECUTING succeed_n %d.\n", mcnt);
5868
5869 /* Originally, mcnt is how many times we HAVE to succeed. */
5870 if (mcnt != 0)
5871 {
5872 /* Here, we discard `const', making re_match non-reentrant. */
5873 unsigned char *p2 = (unsigned char*) p + 2; /* counter loc. */
5874 mcnt--;
5875 p += 4;
5876 PUSH_NUMBER (p2, mcnt);
5877 }
5878 else
5879 /* The two bytes encoding mcnt == 0 are two no_op opcodes. */
5880 goto on_failure;
5881 break;
5882
5883 case jump_n:
5884 /* Signedness doesn't matter since we only compare MCNT to 0. */
5885 EXTRACT_NUMBER (mcnt, p + 2);
5886 DEBUG_PRINT ("EXECUTING jump_n %d.\n", mcnt);
5887
5888 /* Originally, this is how many times we CAN jump. */
5889 if (mcnt != 0)
5890 {
5891 /* Here, we discard `const', making re_match non-reentrant. */
5892 unsigned char *p2 = (unsigned char*) p + 2; /* counter loc. */
5893 mcnt--;
5894 PUSH_NUMBER (p2, mcnt);
5895 goto unconditional_jump;
5896 }
5897 /* If don't have to jump any more, skip over the rest of command. */
5898 else
5899 p += 4;
5900 break;
5901
5902 case set_number_at:
5903 {
5904 unsigned char *p2; /* Location of the counter. */
5905 DEBUG_PRINT ("EXECUTING set_number_at.\n");
5906
5907 EXTRACT_NUMBER_AND_INCR (mcnt, p);
5908 /* Here, we discard `const', making re_match non-reentrant. */
5909 p2 = (unsigned char*) p + mcnt;
5910 /* Signedness doesn't matter since we only copy MCNT's bits . */
5911 EXTRACT_NUMBER_AND_INCR (mcnt, p);
5912 DEBUG_PRINT (" Setting %p to %d.\n", p2, mcnt);
5913 PUSH_NUMBER (p2, mcnt);
5914 break;
5915 }
5916
5917 case wordbound:
5918 case notwordbound:
5919 {
5920 boolean not = (re_opcode_t) *(p - 1) == notwordbound;
5921 DEBUG_PRINT ("EXECUTING %swordbound.\n", not ? "not" : "");
5922
5923 /* We SUCCEED (or FAIL) in one of the following cases: */
5924
5925 /* Case 1: D is at the beginning or the end of string. */
5926 if (AT_STRINGS_BEG (d) || AT_STRINGS_END (d))
5927 not = !not;
5928 else
5929 {
5930 /* C1 is the character before D, S1 is the syntax of C1, C2
5931 is the character at D, and S2 is the syntax of C2. */
5932 re_wchar_t c1, c2;
5933 int s1, s2;
5934 int dummy;
5935 #ifdef emacs
5936 ssize_t offset = PTR_TO_OFFSET (d - 1);
5937 ssize_t charpos = SYNTAX_TABLE_BYTE_TO_CHAR (offset);
5938 UPDATE_SYNTAX_TABLE (charpos);
5939 #endif
5940 GET_CHAR_BEFORE_2 (c1, d, string1, end1, string2, end2);
5941 s1 = SYNTAX (c1);
5942 #ifdef emacs
5943 UPDATE_SYNTAX_TABLE_FORWARD (charpos + 1);
5944 #endif
5945 PREFETCH_NOLIMIT ();
5946 GET_CHAR_AFTER (c2, d, dummy);
5947 s2 = SYNTAX (c2);
5948
5949 if (/* Case 2: Only one of S1 and S2 is Sword. */
5950 ((s1 == Sword) != (s2 == Sword))
5951 /* Case 3: Both of S1 and S2 are Sword, and macro
5952 WORD_BOUNDARY_P (C1, C2) returns nonzero. */
5953 || ((s1 == Sword) && WORD_BOUNDARY_P (c1, c2)))
5954 not = !not;
5955 }
5956 if (not)
5957 break;
5958 else
5959 goto fail;
5960 }
5961
5962 case wordbeg:
5963 DEBUG_PRINT ("EXECUTING wordbeg.\n");
5964
5965 /* We FAIL in one of the following cases: */
5966
5967 /* Case 1: D is at the end of string. */
5968 if (AT_STRINGS_END (d))
5969 goto fail;
5970 else
5971 {
5972 /* C1 is the character before D, S1 is the syntax of C1, C2
5973 is the character at D, and S2 is the syntax of C2. */
5974 re_wchar_t c1, c2;
5975 int s1, s2;
5976 int dummy;
5977 #ifdef emacs
5978 ssize_t offset = PTR_TO_OFFSET (d);
5979 ssize_t charpos = SYNTAX_TABLE_BYTE_TO_CHAR (offset);
5980 UPDATE_SYNTAX_TABLE (charpos);
5981 #endif
5982 PREFETCH ();
5983 GET_CHAR_AFTER (c2, d, dummy);
5984 s2 = SYNTAX (c2);
5985
5986 /* Case 2: S2 is not Sword. */
5987 if (s2 != Sword)
5988 goto fail;
5989
5990 /* Case 3: D is not at the beginning of string ... */
5991 if (!AT_STRINGS_BEG (d))
5992 {
5993 GET_CHAR_BEFORE_2 (c1, d, string1, end1, string2, end2);
5994 #ifdef emacs
5995 UPDATE_SYNTAX_TABLE_BACKWARD (charpos - 1);
5996 #endif
5997 s1 = SYNTAX (c1);
5998
5999 /* ... and S1 is Sword, and WORD_BOUNDARY_P (C1, C2)
6000 returns 0. */
6001 if ((s1 == Sword) && !WORD_BOUNDARY_P (c1, c2))
6002 goto fail;
6003 }
6004 }
6005 break;
6006
6007 case wordend:
6008 DEBUG_PRINT ("EXECUTING wordend.\n");
6009
6010 /* We FAIL in one of the following cases: */
6011
6012 /* Case 1: D is at the beginning of string. */
6013 if (AT_STRINGS_BEG (d))
6014 goto fail;
6015 else
6016 {
6017 /* C1 is the character before D, S1 is the syntax of C1, C2
6018 is the character at D, and S2 is the syntax of C2. */
6019 re_wchar_t c1, c2;
6020 int s1, s2;
6021 int dummy;
6022 #ifdef emacs
6023 ssize_t offset = PTR_TO_OFFSET (d) - 1;
6024 ssize_t charpos = SYNTAX_TABLE_BYTE_TO_CHAR (offset);
6025 UPDATE_SYNTAX_TABLE (charpos);
6026 #endif
6027 GET_CHAR_BEFORE_2 (c1, d, string1, end1, string2, end2);
6028 s1 = SYNTAX (c1);
6029
6030 /* Case 2: S1 is not Sword. */
6031 if (s1 != Sword)
6032 goto fail;
6033
6034 /* Case 3: D is not at the end of string ... */
6035 if (!AT_STRINGS_END (d))
6036 {
6037 PREFETCH_NOLIMIT ();
6038 GET_CHAR_AFTER (c2, d, dummy);
6039 #ifdef emacs
6040 UPDATE_SYNTAX_TABLE_FORWARD (charpos);
6041 #endif
6042 s2 = SYNTAX (c2);
6043
6044 /* ... and S2 is Sword, and WORD_BOUNDARY_P (C1, C2)
6045 returns 0. */
6046 if ((s2 == Sword) && !WORD_BOUNDARY_P (c1, c2))
6047 goto fail;
6048 }
6049 }
6050 break;
6051
6052 case symbeg:
6053 DEBUG_PRINT ("EXECUTING symbeg.\n");
6054
6055 /* We FAIL in one of the following cases: */
6056
6057 /* Case 1: D is at the end of string. */
6058 if (AT_STRINGS_END (d))
6059 goto fail;
6060 else
6061 {
6062 /* C1 is the character before D, S1 is the syntax of C1, C2
6063 is the character at D, and S2 is the syntax of C2. */
6064 re_wchar_t c1, c2;
6065 int s1, s2;
6066 #ifdef emacs
6067 ssize_t offset = PTR_TO_OFFSET (d);
6068 ssize_t charpos = SYNTAX_TABLE_BYTE_TO_CHAR (offset);
6069 UPDATE_SYNTAX_TABLE (charpos);
6070 #endif
6071 PREFETCH ();
6072 c2 = RE_STRING_CHAR (d, target_multibyte);
6073 s2 = SYNTAX (c2);
6074
6075 /* Case 2: S2 is neither Sword nor Ssymbol. */
6076 if (s2 != Sword && s2 != Ssymbol)
6077 goto fail;
6078
6079 /* Case 3: D is not at the beginning of string ... */
6080 if (!AT_STRINGS_BEG (d))
6081 {
6082 GET_CHAR_BEFORE_2 (c1, d, string1, end1, string2, end2);
6083 #ifdef emacs
6084 UPDATE_SYNTAX_TABLE_BACKWARD (charpos - 1);
6085 #endif
6086 s1 = SYNTAX (c1);
6087
6088 /* ... and S1 is Sword or Ssymbol. */
6089 if (s1 == Sword || s1 == Ssymbol)
6090 goto fail;
6091 }
6092 }
6093 break;
6094
6095 case symend:
6096 DEBUG_PRINT ("EXECUTING symend.\n");
6097
6098 /* We FAIL in one of the following cases: */
6099
6100 /* Case 1: D is at the beginning of string. */
6101 if (AT_STRINGS_BEG (d))
6102 goto fail;
6103 else
6104 {
6105 /* C1 is the character before D, S1 is the syntax of C1, C2
6106 is the character at D, and S2 is the syntax of C2. */
6107 re_wchar_t c1, c2;
6108 int s1, s2;
6109 #ifdef emacs
6110 ssize_t offset = PTR_TO_OFFSET (d) - 1;
6111 ssize_t charpos = SYNTAX_TABLE_BYTE_TO_CHAR (offset);
6112 UPDATE_SYNTAX_TABLE (charpos);
6113 #endif
6114 GET_CHAR_BEFORE_2 (c1, d, string1, end1, string2, end2);
6115 s1 = SYNTAX (c1);
6116
6117 /* Case 2: S1 is neither Ssymbol nor Sword. */
6118 if (s1 != Sword && s1 != Ssymbol)
6119 goto fail;
6120
6121 /* Case 3: D is not at the end of string ... */
6122 if (!AT_STRINGS_END (d))
6123 {
6124 PREFETCH_NOLIMIT ();
6125 c2 = RE_STRING_CHAR (d, target_multibyte);
6126 #ifdef emacs
6127 UPDATE_SYNTAX_TABLE_FORWARD (charpos + 1);
6128 #endif
6129 s2 = SYNTAX (c2);
6130
6131 /* ... and S2 is Sword or Ssymbol. */
6132 if (s2 == Sword || s2 == Ssymbol)
6133 goto fail;
6134 }
6135 }
6136 break;
6137
6138 case syntaxspec:
6139 case notsyntaxspec:
6140 {
6141 boolean not = (re_opcode_t) *(p - 1) == notsyntaxspec;
6142 mcnt = *p++;
6143 DEBUG_PRINT ("EXECUTING %ssyntaxspec %d.\n", not ? "not" : "",
6144 mcnt);
6145 PREFETCH ();
6146 #ifdef emacs
6147 {
6148 ssize_t offset = PTR_TO_OFFSET (d);
6149 ssize_t pos1 = SYNTAX_TABLE_BYTE_TO_CHAR (offset);
6150 UPDATE_SYNTAX_TABLE (pos1);
6151 }
6152 #endif
6153 {
6154 int len;
6155 re_wchar_t c;
6156
6157 GET_CHAR_AFTER (c, d, len);
6158 if ((SYNTAX (c) != (enum syntaxcode) mcnt) ^ not)
6159 goto fail;
6160 d += len;
6161 }
6162 }
6163 break;
6164
6165 #ifdef emacs
6166 case before_dot:
6167 DEBUG_PRINT ("EXECUTING before_dot.\n");
6168 if (PTR_BYTE_POS (d) >= PT_BYTE)
6169 goto fail;
6170 break;
6171
6172 case at_dot:
6173 DEBUG_PRINT ("EXECUTING at_dot.\n");
6174 if (PTR_BYTE_POS (d) != PT_BYTE)
6175 goto fail;
6176 break;
6177
6178 case after_dot:
6179 DEBUG_PRINT ("EXECUTING after_dot.\n");
6180 if (PTR_BYTE_POS (d) <= PT_BYTE)
6181 goto fail;
6182 break;
6183
6184 case categoryspec:
6185 case notcategoryspec:
6186 {
6187 boolean not = (re_opcode_t) *(p - 1) == notcategoryspec;
6188 mcnt = *p++;
6189 DEBUG_PRINT ("EXECUTING %scategoryspec %d.\n",
6190 not ? "not" : "", mcnt);
6191 PREFETCH ();
6192
6193 {
6194 int len;
6195 re_wchar_t c;
6196 GET_CHAR_AFTER (c, d, len);
6197 if ((!CHAR_HAS_CATEGORY (c, mcnt)) ^ not)
6198 goto fail;
6199 d += len;
6200 }
6201 }
6202 break;
6203
6204 #endif /* emacs */
6205
6206 default:
6207 abort ();
6208 }
6209 continue; /* Successfully executed one pattern command; keep going. */
6210
6211
6212 /* We goto here if a matching operation fails. */
6213 fail:
6214 IMMEDIATE_QUIT_CHECK;
6215 if (!FAIL_STACK_EMPTY ())
6216 {
6217 re_char *str, *pat;
6218 /* A restart point is known. Restore to that state. */
6219 DEBUG_PRINT ("\nFAIL:\n");
6220 POP_FAILURE_POINT (str, pat);
6221 switch (*pat++)
6222 {
6223 case on_failure_keep_string_jump:
6224 assert (str == NULL);
6225 goto continue_failure_jump;
6226
6227 case on_failure_jump_nastyloop:
6228 assert ((re_opcode_t)pat[-2] == no_op);
6229 PUSH_FAILURE_POINT (pat - 2, str);
6230 /* Fallthrough */
6231
6232 case on_failure_jump_loop:
6233 case on_failure_jump:
6234 case succeed_n:
6235 d = str;
6236 continue_failure_jump:
6237 EXTRACT_NUMBER_AND_INCR (mcnt, pat);
6238 p = pat + mcnt;
6239 break;
6240
6241 case no_op:
6242 /* A special frame used for nastyloops. */
6243 goto fail;
6244
6245 default:
6246 abort ();
6247 }
6248
6249 assert (p >= bufp->buffer && p <= pend);
6250
6251 if (d >= string1 && d <= end1)
6252 dend = end_match_1;
6253 }
6254 else
6255 break; /* Matching at this starting point really fails. */
6256 } /* for (;;) */
6257
6258 if (best_regs_set)
6259 goto restore_best_regs;
6260
6261 FREE_VARIABLES ();
6262
6263 return -1; /* Failure to match. */
6264 }
6265 \f
6266 /* Subroutine definitions for re_match_2. */
6267
6268 /* Return zero if TRANSLATE[S1] and TRANSLATE[S2] are identical for LEN
6269 bytes; nonzero otherwise. */
6270
6271 static int
6272 bcmp_translate (const_re_char *s1, const_re_char *s2, register ssize_t len,
6273 RE_TRANSLATE_TYPE translate, const int target_multibyte)
6274 {
6275 register re_char *p1 = s1, *p2 = s2;
6276 re_char *p1_end = s1 + len;
6277 re_char *p2_end = s2 + len;
6278
6279 /* FIXME: Checking both p1 and p2 presumes that the two strings might have
6280 different lengths, but relying on a single `len' would break this. -sm */
6281 while (p1 < p1_end && p2 < p2_end)
6282 {
6283 int p1_charlen, p2_charlen;
6284 re_wchar_t p1_ch, p2_ch;
6285
6286 GET_CHAR_AFTER (p1_ch, p1, p1_charlen);
6287 GET_CHAR_AFTER (p2_ch, p2, p2_charlen);
6288
6289 if (RE_TRANSLATE (translate, p1_ch)
6290 != RE_TRANSLATE (translate, p2_ch))
6291 return 1;
6292
6293 p1 += p1_charlen, p2 += p2_charlen;
6294 }
6295
6296 if (p1 != p1_end || p2 != p2_end)
6297 return 1;
6298
6299 return 0;
6300 }
6301 \f
6302 /* Entry points for GNU code. */
6303
6304 /* re_compile_pattern is the GNU regular expression compiler: it
6305 compiles PATTERN (of length SIZE) and puts the result in BUFP.
6306 Returns 0 if the pattern was valid, otherwise an error string.
6307
6308 Assumes the `allocated' (and perhaps `buffer') and `translate' fields
6309 are set in BUFP on entry.
6310
6311 We call regex_compile to do the actual compilation. */
6312
6313 const char *
6314 re_compile_pattern (const char *pattern, size_t length,
6315 struct re_pattern_buffer *bufp)
6316 {
6317 reg_errcode_t ret;
6318
6319 /* GNU code is written to assume at least RE_NREGS registers will be set
6320 (and at least one extra will be -1). */
6321 bufp->regs_allocated = REGS_UNALLOCATED;
6322
6323 /* And GNU code determines whether or not to get register information
6324 by passing null for the REGS argument to re_match, etc., not by
6325 setting no_sub. */
6326 bufp->no_sub = 0;
6327
6328 ret = regex_compile ((re_char*) pattern, length, re_syntax_options, bufp);
6329
6330 if (!ret)
6331 return NULL;
6332 return gettext (re_error_msgid[(int) ret]);
6333 }
6334 WEAK_ALIAS (__re_compile_pattern, re_compile_pattern)
6335 \f
6336 /* Entry points compatible with 4.2 BSD regex library. We don't define
6337 them unless specifically requested. */
6338
6339 #if defined _REGEX_RE_COMP || defined _LIBC
6340
6341 /* BSD has one and only one pattern buffer. */
6342 static struct re_pattern_buffer re_comp_buf;
6343
6344 char *
6345 # ifdef _LIBC
6346 /* Make these definitions weak in libc, so POSIX programs can redefine
6347 these names if they don't use our functions, and still use
6348 regcomp/regexec below without link errors. */
6349 weak_function
6350 # endif
6351 re_comp (const char *s)
6352 {
6353 reg_errcode_t ret;
6354
6355 if (!s)
6356 {
6357 if (!re_comp_buf.buffer)
6358 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6359 return (char *) gettext ("No previous regular expression");
6360 return 0;
6361 }
6362
6363 if (!re_comp_buf.buffer)
6364 {
6365 re_comp_buf.buffer = malloc (200);
6366 if (re_comp_buf.buffer == NULL)
6367 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6368 return (char *) gettext (re_error_msgid[(int) REG_ESPACE]);
6369 re_comp_buf.allocated = 200;
6370
6371 re_comp_buf.fastmap = malloc (1 << BYTEWIDTH);
6372 if (re_comp_buf.fastmap == NULL)
6373 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6374 return (char *) gettext (re_error_msgid[(int) REG_ESPACE]);
6375 }
6376
6377 /* Since `re_exec' always passes NULL for the `regs' argument, we
6378 don't need to initialize the pattern buffer fields which affect it. */
6379
6380 ret = regex_compile (s, strlen (s), re_syntax_options, &re_comp_buf);
6381
6382 if (!ret)
6383 return NULL;
6384
6385 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6386 return (char *) gettext (re_error_msgid[(int) ret]);
6387 }
6388
6389
6390 int
6391 # ifdef _LIBC
6392 weak_function
6393 # endif
6394 re_exec (const char *s)
6395 {
6396 const size_t len = strlen (s);
6397 return re_search (&re_comp_buf, s, len, 0, len, 0) >= 0;
6398 }
6399 #endif /* _REGEX_RE_COMP */
6400 \f
6401 /* POSIX.2 functions. Don't define these for Emacs. */
6402
6403 #ifndef emacs
6404
6405 /* regcomp takes a regular expression as a string and compiles it.
6406
6407 PREG is a regex_t *. We do not expect any fields to be initialized,
6408 since POSIX says we shouldn't. Thus, we set
6409
6410 `buffer' to the compiled pattern;
6411 `used' to the length of the compiled pattern;
6412 `syntax' to RE_SYNTAX_POSIX_EXTENDED if the
6413 REG_EXTENDED bit in CFLAGS is set; otherwise, to
6414 RE_SYNTAX_POSIX_BASIC;
6415 `fastmap' to an allocated space for the fastmap;
6416 `fastmap_accurate' to zero;
6417 `re_nsub' to the number of subexpressions in PATTERN.
6418
6419 PATTERN is the address of the pattern string.
6420
6421 CFLAGS is a series of bits which affect compilation.
6422
6423 If REG_EXTENDED is set, we use POSIX extended syntax; otherwise, we
6424 use POSIX basic syntax.
6425
6426 If REG_NEWLINE is set, then . and [^...] don't match newline.
6427 Also, regexec will try a match beginning after every newline.
6428
6429 If REG_ICASE is set, then we considers upper- and lowercase
6430 versions of letters to be equivalent when matching.
6431
6432 If REG_NOSUB is set, then when PREG is passed to regexec, that
6433 routine will report only success or failure, and nothing about the
6434 registers.
6435
6436 It returns 0 if it succeeds, nonzero if it doesn't. (See regex.h for
6437 the return codes and their meanings.) */
6438
6439 reg_errcode_t
6440 regcomp (regex_t *_Restrict_ preg, const char *_Restrict_ pattern,
6441 int cflags)
6442 {
6443 reg_errcode_t ret;
6444 reg_syntax_t syntax
6445 = (cflags & REG_EXTENDED) ?
6446 RE_SYNTAX_POSIX_EXTENDED : RE_SYNTAX_POSIX_BASIC;
6447
6448 /* regex_compile will allocate the space for the compiled pattern. */
6449 preg->buffer = 0;
6450 preg->allocated = 0;
6451 preg->used = 0;
6452
6453 /* Try to allocate space for the fastmap. */
6454 preg->fastmap = malloc (1 << BYTEWIDTH);
6455
6456 if (cflags & REG_ICASE)
6457 {
6458 unsigned i;
6459
6460 preg->translate = malloc (CHAR_SET_SIZE * sizeof *preg->translate);
6461 if (preg->translate == NULL)
6462 return (int) REG_ESPACE;
6463
6464 /* Map uppercase characters to corresponding lowercase ones. */
6465 for (i = 0; i < CHAR_SET_SIZE; i++)
6466 preg->translate[i] = ISUPPER (i) ? TOLOWER (i) : i;
6467 }
6468 else
6469 preg->translate = NULL;
6470
6471 /* If REG_NEWLINE is set, newlines are treated differently. */
6472 if (cflags & REG_NEWLINE)
6473 { /* REG_NEWLINE implies neither . nor [^...] match newline. */
6474 syntax &= ~RE_DOT_NEWLINE;
6475 syntax |= RE_HAT_LISTS_NOT_NEWLINE;
6476 }
6477 else
6478 syntax |= RE_NO_NEWLINE_ANCHOR;
6479
6480 preg->no_sub = !!(cflags & REG_NOSUB);
6481
6482 /* POSIX says a null character in the pattern terminates it, so we
6483 can use strlen here in compiling the pattern. */
6484 ret = regex_compile ((re_char*) pattern, strlen (pattern), syntax, preg);
6485
6486 /* POSIX doesn't distinguish between an unmatched open-group and an
6487 unmatched close-group: both are REG_EPAREN. */
6488 if (ret == REG_ERPAREN)
6489 ret = REG_EPAREN;
6490
6491 if (ret == REG_NOERROR && preg->fastmap)
6492 { /* Compute the fastmap now, since regexec cannot modify the pattern
6493 buffer. */
6494 re_compile_fastmap (preg);
6495 if (preg->can_be_null)
6496 { /* The fastmap can't be used anyway. */
6497 free (preg->fastmap);
6498 preg->fastmap = NULL;
6499 }
6500 }
6501 return ret;
6502 }
6503 WEAK_ALIAS (__regcomp, regcomp)
6504
6505
6506 /* regexec searches for a given pattern, specified by PREG, in the
6507 string STRING.
6508
6509 If NMATCH is zero or REG_NOSUB was set in the cflags argument to
6510 `regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at
6511 least NMATCH elements, and we set them to the offsets of the
6512 corresponding matched substrings.
6513
6514 EFLAGS specifies `execution flags' which affect matching: if
6515 REG_NOTBOL is set, then ^ does not match at the beginning of the
6516 string; if REG_NOTEOL is set, then $ does not match at the end.
6517
6518 We return 0 if we find a match and REG_NOMATCH if not. */
6519
6520 reg_errcode_t
6521 regexec (const regex_t *_Restrict_ preg, const char *_Restrict_ string,
6522 size_t nmatch, regmatch_t pmatch[_Restrict_arr_], int eflags)
6523 {
6524 regoff_t ret;
6525 struct re_registers regs;
6526 regex_t private_preg;
6527 size_t len = strlen (string);
6528 boolean want_reg_info = !preg->no_sub && nmatch > 0 && pmatch;
6529
6530 private_preg = *preg;
6531
6532 private_preg.not_bol = !!(eflags & REG_NOTBOL);
6533 private_preg.not_eol = !!(eflags & REG_NOTEOL);
6534
6535 /* The user has told us exactly how many registers to return
6536 information about, via `nmatch'. We have to pass that on to the
6537 matching routines. */
6538 private_preg.regs_allocated = REGS_FIXED;
6539
6540 if (want_reg_info)
6541 {
6542 regs.num_regs = nmatch;
6543 regs.start = TALLOC (nmatch * 2, regoff_t);
6544 if (regs.start == NULL)
6545 return REG_NOMATCH;
6546 regs.end = regs.start + nmatch;
6547 }
6548
6549 /* Instead of using not_eol to implement REG_NOTEOL, we could simply
6550 pass (&private_preg, string, len + 1, 0, len, ...) pretending the string
6551 was a little bit longer but still only matching the real part.
6552 This works because the `endline' will check for a '\n' and will find a
6553 '\0', correctly deciding that this is not the end of a line.
6554 But it doesn't work out so nicely for REG_NOTBOL, since we don't have
6555 a convenient '\0' there. For all we know, the string could be preceded
6556 by '\n' which would throw things off. */
6557
6558 /* Perform the searching operation. */
6559 ret = re_search (&private_preg, string, len,
6560 /* start: */ 0, /* range: */ len,
6561 want_reg_info ? &regs : 0);
6562
6563 /* Copy the register information to the POSIX structure. */
6564 if (want_reg_info)
6565 {
6566 if (ret >= 0)
6567 {
6568 unsigned r;
6569
6570 for (r = 0; r < nmatch; r++)
6571 {
6572 pmatch[r].rm_so = regs.start[r];
6573 pmatch[r].rm_eo = regs.end[r];
6574 }
6575 }
6576
6577 /* If we needed the temporary register info, free the space now. */
6578 free (regs.start);
6579 }
6580
6581 /* We want zero return to mean success, unlike `re_search'. */
6582 return ret >= 0 ? REG_NOERROR : REG_NOMATCH;
6583 }
6584 WEAK_ALIAS (__regexec, regexec)
6585
6586
6587 /* Returns a message corresponding to an error code, ERR_CODE, returned
6588 from either regcomp or regexec. We don't use PREG here.
6589
6590 ERR_CODE was previously called ERRCODE, but that name causes an
6591 error with msvc8 compiler. */
6592
6593 size_t
6594 regerror (int err_code, const regex_t *preg, char *errbuf, size_t errbuf_size)
6595 {
6596 const char *msg;
6597 size_t msg_size;
6598
6599 if (err_code < 0
6600 || err_code >= (sizeof (re_error_msgid) / sizeof (re_error_msgid[0])))
6601 /* Only error codes returned by the rest of the code should be passed
6602 to this routine. If we are given anything else, or if other regex
6603 code generates an invalid error code, then the program has a bug.
6604 Dump core so we can fix it. */
6605 abort ();
6606
6607 msg = gettext (re_error_msgid[err_code]);
6608
6609 msg_size = strlen (msg) + 1; /* Includes the null. */
6610
6611 if (errbuf_size != 0)
6612 {
6613 if (msg_size > errbuf_size)
6614 {
6615 memcpy (errbuf, msg, errbuf_size - 1);
6616 errbuf[errbuf_size - 1] = 0;
6617 }
6618 else
6619 strcpy (errbuf, msg);
6620 }
6621
6622 return msg_size;
6623 }
6624 WEAK_ALIAS (__regerror, regerror)
6625
6626
6627 /* Free dynamically allocated space used by PREG. */
6628
6629 void
6630 regfree (regex_t *preg)
6631 {
6632 free (preg->buffer);
6633 preg->buffer = NULL;
6634
6635 preg->allocated = 0;
6636 preg->used = 0;
6637
6638 free (preg->fastmap);
6639 preg->fastmap = NULL;
6640 preg->fastmap_accurate = 0;
6641
6642 free (preg->translate);
6643 preg->translate = NULL;
6644 }
6645 WEAK_ALIAS (__regfree, regfree)
6646
6647 #endif /* not emacs */
Emacs repository mirror