?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
merge from trunk
[emacs.git] / src / regex.c
blob73a735cea655f91cdd53718dd132d5eb91518d3d
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 /* Make syntax table lookup grant data in gl_state. */
135 # define SYNTAX_ENTRY_VIA_PROPERTY
136
137 # include "syntax.h"
138 # include "category.h"
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 CHAR_CHARSET(c) 0
261 # define CHARSET_LEADING_CODE_BASE(c) 0
262 # define MAX_MULTIBYTE_LENGTH 1
263 # define RE_MULTIBYTE_P(x) 0
264 # define RE_TARGET_MULTIBYTE_P(x) 0
265 # define WORD_BOUNDARY_P(c1, c2) (0)
266 # define CHAR_HEAD_P(p) (1)
267 # define SINGLE_BYTE_CHAR_P(c) (1)
268 # define SAME_CHARSET_P(c1, c2) (1)
269 # define BYTES_BY_CHAR_HEAD(p) (1)
270 # define PREV_CHAR_BOUNDARY(p, limit) ((p)--)
271 # define STRING_CHAR(p) (*(p))
272 # define RE_STRING_CHAR(p, multibyte) STRING_CHAR (p)
273 # define CHAR_STRING(c, s) (*(s) = (c), 1)
274 # define STRING_CHAR_AND_LENGTH(p, actual_len) ((actual_len) = 1, *(p))
275 # define RE_STRING_CHAR_AND_LENGTH(p, len, multibyte) STRING_CHAR_AND_LENGTH (p, len)
276 # define RE_CHAR_TO_MULTIBYTE(c) (c)
277 # define RE_CHAR_TO_UNIBYTE(c) (c)
278 # define GET_CHAR_BEFORE_2(c, p, str1, end1, str2, end2) \
279 (c = ((p) == (str2) ? *((end1) - 1) : *((p) - 1)))
280 # define GET_CHAR_AFTER(c, p, len) \
281 (c = *p, len = 1)
282 # define MAKE_CHAR(charset, c1, c2) (c1)
283 # define BYTE8_TO_CHAR(c) (c)
284 # define CHAR_BYTE8_P(c) (0)
285 # define CHAR_LEADING_CODE(c) (c)
286
287 #endif /* not emacs */
288
289 #ifndef RE_TRANSLATE
290 # define RE_TRANSLATE(TBL, C) ((unsigned char)(TBL)[C])
291 # define RE_TRANSLATE_P(TBL) (TBL)
292 #endif
293 \f
294 /* Get the interface, including the syntax bits. */
295 #include "regex.h"
296
297 /* isalpha etc. are used for the character classes. */
298 #include <ctype.h>
299
300 #ifdef emacs
301
302 /* 1 if C is an ASCII character. */
303 # define IS_REAL_ASCII(c) ((c) < 0200)
304
305 /* 1 if C is a unibyte character. */
306 # define ISUNIBYTE(c) (SINGLE_BYTE_CHAR_P ((c)))
307
308 /* The Emacs definitions should not be directly affected by locales. */
309
310 /* In Emacs, these are only used for single-byte characters. */
311 # define ISDIGIT(c) ((c) >= '0' && (c) <= '9')
312 # define ISCNTRL(c) ((c) < ' ')
313 # define ISXDIGIT(c) (((c) >= '0' && (c) <= '9') \
314 || ((c) >= 'a' && (c) <= 'f') \
315 || ((c) >= 'A' && (c) <= 'F'))
316
317 /* This is only used for single-byte characters. */
318 # define ISBLANK(c) ((c) == ' ' || (c) == '\t')
319
320 /* The rest must handle multibyte characters. */
321
322 # define ISGRAPH(c) (SINGLE_BYTE_CHAR_P (c) \
323 ? (c) > ' ' && !((c) >= 0177 && (c) <= 0237) \
324 : 1)
325
326 # define ISPRINT(c) (SINGLE_BYTE_CHAR_P (c) \
327 ? (c) >= ' ' && !((c) >= 0177 && (c) <= 0237) \
328 : 1)
329
330 # define ISALNUM(c) (IS_REAL_ASCII (c) \
331 ? (((c) >= 'a' && (c) <= 'z') \
332 || ((c) >= 'A' && (c) <= 'Z') \
333 || ((c) >= '0' && (c) <= '9')) \
334 : SYNTAX (c) == Sword)
335
336 # define ISALPHA(c) (IS_REAL_ASCII (c) \
337 ? (((c) >= 'a' && (c) <= 'z') \
338 || ((c) >= 'A' && (c) <= 'Z')) \
339 : SYNTAX (c) == Sword)
340
341 # define ISLOWER(c) lowercasep (c)
342
343 # define ISPUNCT(c) (IS_REAL_ASCII (c) \
344 ? ((c) > ' ' && (c) < 0177 \
345 && !(((c) >= 'a' && (c) <= 'z') \
346 || ((c) >= 'A' && (c) <= 'Z') \
347 || ((c) >= '0' && (c) <= '9'))) \
348 : SYNTAX (c) != Sword)
349
350 # define ISSPACE(c) (SYNTAX (c) == Swhitespace)
351
352 # define ISUPPER(c) uppercasep (c)
353
354 # define ISWORD(c) (SYNTAX (c) == Sword)
355
356 #else /* not emacs */
357
358 /* 1 if C is an ASCII character. */
359 # define IS_REAL_ASCII(c) ((c) < 0200)
360
361 /* This distinction is not meaningful, except in Emacs. */
362 # define ISUNIBYTE(c) 1
363
364 # ifdef isblank
365 # define ISBLANK(c) isblank (c)
366 # else
367 # define ISBLANK(c) ((c) == ' ' || (c) == '\t')
368 # endif
369 # ifdef isgraph
370 # define ISGRAPH(c) isgraph (c)
371 # else
372 # define ISGRAPH(c) (isprint (c) && !isspace (c))
373 # endif
374
375 /* Solaris defines ISPRINT so we must undefine it first. */
376 # undef ISPRINT
377 # define ISPRINT(c) isprint (c)
378 # define ISDIGIT(c) isdigit (c)
379 # define ISALNUM(c) isalnum (c)
380 # define ISALPHA(c) isalpha (c)
381 # define ISCNTRL(c) iscntrl (c)
382 # define ISLOWER(c) islower (c)
383 # define ISPUNCT(c) ispunct (c)
384 # define ISSPACE(c) isspace (c)
385 # define ISUPPER(c) isupper (c)
386 # define ISXDIGIT(c) isxdigit (c)
387
388 # define ISWORD(c) ISALPHA (c)
389
390 # ifdef _tolower
391 # define TOLOWER(c) _tolower (c)
392 # else
393 # define TOLOWER(c) tolower (c)
394 # endif
395
396 /* How many characters in the character set. */
397 # define CHAR_SET_SIZE 256
398
399 # ifdef SYNTAX_TABLE
400
401 extern char *re_syntax_table;
402
403 # else /* not SYNTAX_TABLE */
404
405 static char re_syntax_table[CHAR_SET_SIZE];
406
407 static void
408 init_syntax_once (void)
409 {
410 register int c;
411 static int done = 0;
412
413 if (done)
414 return;
415
416 memset (re_syntax_table, 0, sizeof re_syntax_table);
417
418 for (c = 0; c < CHAR_SET_SIZE; ++c)
419 if (ISALNUM (c))
420 re_syntax_table[c] = Sword;
421
422 re_syntax_table['_'] = Ssymbol;
423
424 done = 1;
425 }
426
427 # endif /* not SYNTAX_TABLE */
428
429 # define SYNTAX(c) re_syntax_table[(c)]
430
431 #endif /* not emacs */
432 \f
433 #define SIGN_EXTEND_CHAR(c) ((signed char) (c))
434 \f
435 /* Should we use malloc or alloca? If REGEX_MALLOC is not defined, we
436 use `alloca' instead of `malloc'. This is because using malloc in
437 re_search* or re_match* could cause memory leaks when C-g is used in
438 Emacs; also, malloc is slower and causes storage fragmentation. On
439 the other hand, malloc is more portable, and easier to debug.
440
441 Because we sometimes use alloca, some routines have to be macros,
442 not functions -- `alloca'-allocated space disappears at the end of the
443 function it is called in. */
444
445 #ifdef REGEX_MALLOC
446
447 # define REGEX_ALLOCATE malloc
448 # define REGEX_REALLOCATE(source, osize, nsize) realloc (source, nsize)
449 # define REGEX_FREE free
450
451 #else /* not REGEX_MALLOC */
452
453 /* Emacs already defines alloca, sometimes. */
454 # ifndef alloca
455
456 /* Make alloca work the best possible way. */
457 # ifdef __GNUC__
458 # define alloca __builtin_alloca
459 # else /* not __GNUC__ */
460 # ifdef HAVE_ALLOCA_H
461 # include <alloca.h>
462 # endif /* HAVE_ALLOCA_H */
463 # endif /* not __GNUC__ */
464
465 # endif /* not alloca */
466
467 # define REGEX_ALLOCATE alloca
468
469 /* Assumes a `char *destination' variable. */
470 # define REGEX_REALLOCATE(source, osize, nsize) \
471 (destination = (char *) alloca (nsize), \
472 memcpy (destination, source, osize))
473
474 /* No need to do anything to free, after alloca. */
475 # define REGEX_FREE(arg) ((void)0) /* Do nothing! But inhibit gcc warning. */
476
477 #endif /* not REGEX_MALLOC */
478
479 /* Define how to allocate the failure stack. */
480
481 #if defined REL_ALLOC && defined REGEX_MALLOC
482
483 # define REGEX_ALLOCATE_STACK(size) \
484 r_alloc (&failure_stack_ptr, (size))
485 # define REGEX_REALLOCATE_STACK(source, osize, nsize) \
486 r_re_alloc (&failure_stack_ptr, (nsize))
487 # define REGEX_FREE_STACK(ptr) \
488 r_alloc_free (&failure_stack_ptr)
489
490 #else /* not using relocating allocator */
491
492 # ifdef REGEX_MALLOC
493
494 # define REGEX_ALLOCATE_STACK malloc
495 # define REGEX_REALLOCATE_STACK(source, osize, nsize) realloc (source, nsize)
496 # define REGEX_FREE_STACK free
497
498 # else /* not REGEX_MALLOC */
499
500 # define REGEX_ALLOCATE_STACK alloca
501
502 # define REGEX_REALLOCATE_STACK(source, osize, nsize) \
503 REGEX_REALLOCATE (source, osize, nsize)
504 /* No need to explicitly free anything. */
505 # define REGEX_FREE_STACK(arg) ((void)0)
506
507 # endif /* not REGEX_MALLOC */
508 #endif /* not using relocating allocator */
509
510
511 /* True if `size1' is non-NULL and PTR is pointing anywhere inside
512 `string1' or just past its end. This works if PTR is NULL, which is
513 a good thing. */
514 #define FIRST_STRING_P(ptr) \
515 (size1 && string1 <= (ptr) && (ptr) <= string1 + size1)
516
517 /* (Re)Allocate N items of type T using malloc, or fail. */
518 #define TALLOC(n, t) ((t *) malloc ((n) * sizeof (t)))
519 #define RETALLOC(addr, n, t) ((addr) = (t *) realloc (addr, (n) * sizeof (t)))
520 #define REGEX_TALLOC(n, t) ((t *) REGEX_ALLOCATE ((n) * sizeof (t)))
521
522 #define BYTEWIDTH 8 /* In bits. */
523
524 #define STREQ(s1, s2) ((strcmp (s1, s2) == 0))
525
526 #undef MAX
527 #undef MIN
528 #define MAX(a, b) ((a) > (b) ? (a) : (b))
529 #define MIN(a, b) ((a) < (b) ? (a) : (b))
530
531 /* Type of source-pattern and string chars. */
532 #ifdef _MSC_VER
533 typedef unsigned char re_char;
534 #else
535 typedef const unsigned char re_char;
536 #endif
537
538 typedef char boolean;
539
540 static regoff_t re_match_2_internal (struct re_pattern_buffer *bufp,
541 re_char *string1, size_t size1,
542 re_char *string2, size_t size2,
543 ssize_t pos,
544 struct re_registers *regs,
545 ssize_t stop);
546 \f
547 /* These are the command codes that appear in compiled regular
548 expressions. Some opcodes are followed by argument bytes. A
549 command code can specify any interpretation whatsoever for its
550 arguments. Zero bytes may appear in the compiled regular expression. */
551
552 typedef enum
553 {
554 no_op = 0,
555
556 /* Succeed right away--no more backtracking. */
557 succeed,
558
559 /* Followed by one byte giving n, then by n literal bytes. */
560 exactn,
561
562 /* Matches any (more or less) character. */
563 anychar,
564
565 /* Matches any one char belonging to specified set. First
566 following byte is number of bitmap bytes. Then come bytes
567 for a bitmap saying which chars are in. Bits in each byte
568 are ordered low-bit-first. A character is in the set if its
569 bit is 1. A character too large to have a bit in the map is
570 automatically not in the set.
571
572 If the length byte has the 0x80 bit set, then that stuff
573 is followed by a range table:
574 2 bytes of flags for character sets (low 8 bits, high 8 bits)
575 See RANGE_TABLE_WORK_BITS below.
576 2 bytes, the number of pairs that follow (upto 32767)
577 pairs, each 2 multibyte characters,
578 each multibyte character represented as 3 bytes. */
579 charset,
580
581 /* Same parameters as charset, but match any character that is
582 not one of those specified. */
583 charset_not,
584
585 /* Start remembering the text that is matched, for storing in a
586 register. Followed by one byte with the register number, in
587 the range 0 to one less than the pattern buffer's re_nsub
588 field. */
589 start_memory,
590
591 /* Stop remembering the text that is matched and store it in a
592 memory register. Followed by one byte with the register
593 number, in the range 0 to one less than `re_nsub' in the
594 pattern buffer. */
595 stop_memory,
596
597 /* Match a duplicate of something remembered. Followed by one
598 byte containing the register number. */
599 duplicate,
600
601 /* Fail unless at beginning of line. */
602 begline,
603
604 /* Fail unless at end of line. */
605 endline,
606
607 /* Succeeds if at beginning of buffer (if emacs) or at beginning
608 of string to be matched (if not). */
609 begbuf,
610
611 /* Analogously, for end of buffer/string. */
612 endbuf,
613
614 /* Followed by two byte relative address to which to jump. */
615 jump,
616
617 /* Followed by two-byte relative address of place to resume at
618 in case of failure. */
619 on_failure_jump,
620
621 /* Like on_failure_jump, but pushes a placeholder instead of the
622 current string position when executed. */
623 on_failure_keep_string_jump,
624
625 /* Just like `on_failure_jump', except that it checks that we
626 don't get stuck in an infinite loop (matching an empty string
627 indefinitely). */
628 on_failure_jump_loop,
629
630 /* Just like `on_failure_jump_loop', except that it checks for
631 a different kind of loop (the kind that shows up with non-greedy
632 operators). This operation has to be immediately preceded
633 by a `no_op'. */
634 on_failure_jump_nastyloop,
635
636 /* A smart `on_failure_jump' used for greedy * and + operators.
637 It analyzes the loop before which it is put and if the
638 loop does not require backtracking, it changes itself to
639 `on_failure_keep_string_jump' and short-circuits the loop,
640 else it just defaults to changing itself into `on_failure_jump'.
641 It assumes that it is pointing to just past a `jump'. */
642 on_failure_jump_smart,
643
644 /* Followed by two-byte relative address and two-byte number n.
645 After matching N times, jump to the address upon failure.
646 Does not work if N starts at 0: use on_failure_jump_loop
647 instead. */
648 succeed_n,
649
650 /* Followed by two-byte relative address, and two-byte number n.
651 Jump to the address N times, then fail. */
652 jump_n,
653
654 /* Set the following two-byte relative address to the
655 subsequent two-byte number. The address *includes* the two
656 bytes of number. */
657 set_number_at,
658
659 wordbeg, /* Succeeds if at word beginning. */
660 wordend, /* Succeeds if at word end. */
661
662 wordbound, /* Succeeds if at a word boundary. */
663 notwordbound, /* Succeeds if not at a word boundary. */
664
665 symbeg, /* Succeeds if at symbol beginning. */
666 symend, /* Succeeds if at symbol end. */
667
668 /* Matches any character whose syntax is specified. Followed by
669 a byte which contains a syntax code, e.g., Sword. */
670 syntaxspec,
671
672 /* Matches any character whose syntax is not that specified. */
673 notsyntaxspec
674
675 #ifdef emacs
676 ,before_dot, /* Succeeds if before point. */
677 at_dot, /* Succeeds if at point. */
678 after_dot, /* Succeeds if after point. */
679
680 /* Matches any character whose category-set contains the specified
681 category. The operator is followed by a byte which contains a
682 category code (mnemonic ASCII character). */
683 categoryspec,
684
685 /* Matches any character whose category-set does not contain the
686 specified category. The operator is followed by a byte which
687 contains the category code (mnemonic ASCII character). */
688 notcategoryspec
689 #endif /* emacs */
690 } re_opcode_t;
691 \f
692 /* Common operations on the compiled pattern. */
693
694 /* Store NUMBER in two contiguous bytes starting at DESTINATION. */
695
696 #define STORE_NUMBER(destination, number) \
697 do { \
698 (destination)[0] = (number) & 0377; \
699 (destination)[1] = (number) >> 8; \
700 } while (0)
701
702 /* Same as STORE_NUMBER, except increment DESTINATION to
703 the byte after where the number is stored. Therefore, DESTINATION
704 must be an lvalue. */
705
706 #define STORE_NUMBER_AND_INCR(destination, number) \
707 do { \
708 STORE_NUMBER (destination, number); \
709 (destination) += 2; \
710 } while (0)
711
712 /* Put into DESTINATION a number stored in two contiguous bytes starting
713 at SOURCE. */
714
715 #define EXTRACT_NUMBER(destination, source) \
716 ((destination) = extract_number (source))
717
718 static int
719 extract_number (re_char *source)
720 {
721 return (SIGN_EXTEND_CHAR (source[1]) << 8) + source[0];
722 }
723
724 /* Same as EXTRACT_NUMBER, except increment SOURCE to after the number.
725 SOURCE must be an lvalue. */
726
727 #define EXTRACT_NUMBER_AND_INCR(destination, source) \
728 ((destination) = extract_number_and_incr (&source))
729
730 static int
731 extract_number_and_incr (re_char **source)
732 {
733 int num = extract_number (*source);
734 *source += 2;
735 return num;
736 }
737 \f
738 /* Store a multibyte character in three contiguous bytes starting
739 DESTINATION, and increment DESTINATION to the byte after where the
740 character is stored. Therefore, DESTINATION must be an lvalue. */
741
742 #define STORE_CHARACTER_AND_INCR(destination, character) \
743 do { \
744 (destination)[0] = (character) & 0377; \
745 (destination)[1] = ((character) >> 8) & 0377; \
746 (destination)[2] = (character) >> 16; \
747 (destination) += 3; \
748 } while (0)
749
750 /* Put into DESTINATION a character stored in three contiguous bytes
751 starting at SOURCE. */
752
753 #define EXTRACT_CHARACTER(destination, source) \
754 do { \
755 (destination) = ((source)[0] \
756 | ((source)[1] << 8) \
757 | ((source)[2] << 16)); \
758 } while (0)
759
760
761 /* Macros for charset. */
762
763 /* Size of bitmap of charset P in bytes. P is a start of charset,
764 i.e. *P is (re_opcode_t) charset or (re_opcode_t) charset_not. */
765 #define CHARSET_BITMAP_SIZE(p) ((p)[1] & 0x7F)
766
767 /* Nonzero if charset P has range table. */
768 #define CHARSET_RANGE_TABLE_EXISTS_P(p) ((p)[1] & 0x80)
769
770 /* Return the address of range table of charset P. But not the start
771 of table itself, but the before where the number of ranges is
772 stored. `2 +' means to skip re_opcode_t and size of bitmap,
773 and the 2 bytes of flags at the start of the range table. */
774 #define CHARSET_RANGE_TABLE(p) (&(p)[4 + CHARSET_BITMAP_SIZE (p)])
775
776 /* Extract the bit flags that start a range table. */
777 #define CHARSET_RANGE_TABLE_BITS(p) \
778 ((p)[2 + CHARSET_BITMAP_SIZE (p)] \
779 + (p)[3 + CHARSET_BITMAP_SIZE (p)] * 0x100)
780
781 /* Return the address of end of RANGE_TABLE. COUNT is number of
782 ranges (which is a pair of (start, end)) in the RANGE_TABLE. `* 2'
783 is start of range and end of range. `* 3' is size of each start
784 and end. */
785 #define CHARSET_RANGE_TABLE_END(range_table, count) \
786 ((range_table) + (count) * 2 * 3)
787
788 /* Test if C is in RANGE_TABLE. A flag NOT is negated if C is in.
789 COUNT is number of ranges in RANGE_TABLE. */
790 #define CHARSET_LOOKUP_RANGE_TABLE_RAW(not, c, range_table, count) \
791 do \
792 { \
793 re_wchar_t range_start, range_end; \
794 re_char *rtp; \
795 re_char *range_table_end \
796 = CHARSET_RANGE_TABLE_END ((range_table), (count)); \
797 \
798 for (rtp = (range_table); rtp < range_table_end; rtp += 2 * 3) \
799 { \
800 EXTRACT_CHARACTER (range_start, rtp); \
801 EXTRACT_CHARACTER (range_end, rtp + 3); \
802 \
803 if (range_start <= (c) && (c) <= range_end) \
804 { \
805 (not) = !(not); \
806 break; \
807 } \
808 } \
809 } \
810 while (0)
811
812 /* Test if C is in range table of CHARSET. The flag NOT is negated if
813 C is listed in it. */
814 #define CHARSET_LOOKUP_RANGE_TABLE(not, c, charset) \
815 do \
816 { \
817 /* Number of ranges in range table. */ \
818 int count; \
819 re_char *range_table = CHARSET_RANGE_TABLE (charset); \
820 \
821 EXTRACT_NUMBER_AND_INCR (count, range_table); \
822 CHARSET_LOOKUP_RANGE_TABLE_RAW ((not), (c), range_table, count); \
823 } \
824 while (0)
825 \f
826 /* If DEBUG is defined, Regex prints many voluminous messages about what
827 it is doing (if the variable `debug' is nonzero). If linked with the
828 main program in `iregex.c', you can enter patterns and strings
829 interactively. And if linked with the main program in `main.c' and
830 the other test files, you can run the already-written tests. */
831
832 #ifdef DEBUG
833
834 /* We use standard I/O for debugging. */
835 # include <stdio.h>
836
837 /* It is useful to test things that ``must'' be true when debugging. */
838 # include <assert.h>
839
840 static int debug = -100000;
841
842 # define DEBUG_STATEMENT(e) e
843 # define DEBUG_PRINT(...) if (debug > 0) printf (__VA_ARGS__)
844 # define DEBUG_COMPILES_ARGUMENTS
845 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) \
846 if (debug > 0) print_partial_compiled_pattern (s, e)
847 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) \
848 if (debug > 0) print_double_string (w, s1, sz1, s2, sz2)
849
850
851 /* Print the fastmap in human-readable form. */
852
853 static void
854 print_fastmap (char *fastmap)
855 {
856 unsigned was_a_range = 0;
857 unsigned i = 0;
858
859 while (i < (1 << BYTEWIDTH))
860 {
861 if (fastmap[i++])
862 {
863 was_a_range = 0;
864 putchar (i - 1);
865 while (i < (1 << BYTEWIDTH) && fastmap[i])
866 {
867 was_a_range = 1;
868 i++;
869 }
870 if (was_a_range)
871 {
872 printf ("-");
873 putchar (i - 1);
874 }
875 }
876 }
877 putchar ('\n');
878 }
879
880
881 /* Print a compiled pattern string in human-readable form, starting at
882 the START pointer into it and ending just before the pointer END. */
883
884 static void
885 print_partial_compiled_pattern (re_char *start, re_char *end)
886 {
887 int mcnt, mcnt2;
888 re_char *p = start;
889 re_char *pend = end;
890
891 if (start == NULL)
892 {
893 fprintf (stderr, "(null)\n");
894 return;
895 }
896
897 /* Loop over pattern commands. */
898 while (p < pend)
899 {
900 fprintf (stderr, "%td:\t", p - start);
901
902 switch ((re_opcode_t) *p++)
903 {
904 case no_op:
905 fprintf (stderr, "/no_op");
906 break;
907
908 case succeed:
909 fprintf (stderr, "/succeed");
910 break;
911
912 case exactn:
913 mcnt = *p++;
914 fprintf (stderr, "/exactn/%d", mcnt);
915 do
916 {
917 fprintf (stderr, "/%c", *p++);
918 }
919 while (--mcnt);
920 break;
921
922 case start_memory:
923 fprintf (stderr, "/start_memory/%d", *p++);
924 break;
925
926 case stop_memory:
927 fprintf (stderr, "/stop_memory/%d", *p++);
928 break;
929
930 case duplicate:
931 fprintf (stderr, "/duplicate/%d", *p++);
932 break;
933
934 case anychar:
935 fprintf (stderr, "/anychar");
936 break;
937
938 case charset:
939 case charset_not:
940 {
941 register int c, last = -100;
942 register int in_range = 0;
943 int length = CHARSET_BITMAP_SIZE (p - 1);
944 int has_range_table = CHARSET_RANGE_TABLE_EXISTS_P (p - 1);
945
946 fprintf (stderr, "/charset [%s",
947 (re_opcode_t) *(p - 1) == charset_not ? "^" : "");
948
949 if (p + *p >= pend)
950 fprintf (stderr, " !extends past end of pattern! ");
951
952 for (c = 0; c < 256; c++)
953 if (c / 8 < length
954 && (p[1 + (c/8)] & (1 << (c % 8))))
955 {
956 /* Are we starting a range? */
957 if (last + 1 == c && ! in_range)
958 {
959 fprintf (stderr, "-");
960 in_range = 1;
961 }
962 /* Have we broken a range? */
963 else if (last + 1 != c && in_range)
964 {
965 fprintf (stderr, "%c", last);
966 in_range = 0;
967 }
968
969 if (! in_range)
970 fprintf (stderr, "%c", c);
971
972 last = c;
973 }
974
975 if (in_range)
976 fprintf (stderr, "%c", last);
977
978 fprintf (stderr, "]");
979
980 p += 1 + length;
981
982 if (has_range_table)
983 {
984 int count;
985 fprintf (stderr, "has-range-table");
986
987 /* ??? Should print the range table; for now, just skip it. */
988 p += 2; /* skip range table bits */
989 EXTRACT_NUMBER_AND_INCR (count, p);
990 p = CHARSET_RANGE_TABLE_END (p, count);
991 }
992 }
993 break;
994
995 case begline:
996 fprintf (stderr, "/begline");
997 break;
998
999 case endline:
1000 fprintf (stderr, "/endline");
1001 break;
1002
1003 case on_failure_jump:
1004 EXTRACT_NUMBER_AND_INCR (mcnt, p);
1005 fprintf (stderr, "/on_failure_jump to %td", p + mcnt - start);
1006 break;
1007
1008 case on_failure_keep_string_jump:
1009 EXTRACT_NUMBER_AND_INCR (mcnt, p);
1010 fprintf (stderr, "/on_failure_keep_string_jump to %td",
1011 p + mcnt - start);
1012 break;
1013
1014 case on_failure_jump_nastyloop:
1015 EXTRACT_NUMBER_AND_INCR (mcnt, p);
1016 fprintf (stderr, "/on_failure_jump_nastyloop to %td",
1017 p + mcnt - start);
1018 break;
1019
1020 case on_failure_jump_loop:
1021 EXTRACT_NUMBER_AND_INCR (mcnt, p);
1022 fprintf (stderr, "/on_failure_jump_loop to %td",
1023 p + mcnt - start);
1024 break;
1025
1026 case on_failure_jump_smart:
1027 EXTRACT_NUMBER_AND_INCR (mcnt, p);
1028 fprintf (stderr, "/on_failure_jump_smart to %td",
1029 p + mcnt - start);
1030 break;
1031
1032 case jump:
1033 EXTRACT_NUMBER_AND_INCR (mcnt, p);
1034 fprintf (stderr, "/jump to %td", p + mcnt - start);
1035 break;
1036
1037 case succeed_n:
1038 EXTRACT_NUMBER_AND_INCR (mcnt, p);
1039 EXTRACT_NUMBER_AND_INCR (mcnt2, p);
1040 fprintf (stderr, "/succeed_n to %td, %d times",
1041 p - 2 + mcnt - start, mcnt2);
1042 break;
1043
1044 case jump_n:
1045 EXTRACT_NUMBER_AND_INCR (mcnt, p);
1046 EXTRACT_NUMBER_AND_INCR (mcnt2, p);
1047 fprintf (stderr, "/jump_n to %td, %d times",
1048 p - 2 + mcnt - start, mcnt2);
1049 break;
1050
1051 case set_number_at:
1052 EXTRACT_NUMBER_AND_INCR (mcnt, p);
1053 EXTRACT_NUMBER_AND_INCR (mcnt2, p);
1054 fprintf (stderr, "/set_number_at location %td to %d",
1055 p - 2 + mcnt - start, mcnt2);
1056 break;
1057
1058 case wordbound:
1059 fprintf (stderr, "/wordbound");
1060 break;
1061
1062 case notwordbound:
1063 fprintf (stderr, "/notwordbound");
1064 break;
1065
1066 case wordbeg:
1067 fprintf (stderr, "/wordbeg");
1068 break;
1069
1070 case wordend:
1071 fprintf (stderr, "/wordend");
1072 break;
1073
1074 case symbeg:
1075 fprintf (stderr, "/symbeg");
1076 break;
1077
1078 case symend:
1079 fprintf (stderr, "/symend");
1080 break;
1081
1082 case syntaxspec:
1083 fprintf (stderr, "/syntaxspec");
1084 mcnt = *p++;
1085 fprintf (stderr, "/%d", mcnt);
1086 break;
1087
1088 case notsyntaxspec:
1089 fprintf (stderr, "/notsyntaxspec");
1090 mcnt = *p++;
1091 fprintf (stderr, "/%d", mcnt);
1092 break;
1093
1094 # ifdef emacs
1095 case before_dot:
1096 fprintf (stderr, "/before_dot");
1097 break;
1098
1099 case at_dot:
1100 fprintf (stderr, "/at_dot");
1101 break;
1102
1103 case after_dot:
1104 fprintf (stderr, "/after_dot");
1105 break;
1106
1107 case categoryspec:
1108 fprintf (stderr, "/categoryspec");
1109 mcnt = *p++;
1110 fprintf (stderr, "/%d", mcnt);
1111 break;
1112
1113 case notcategoryspec:
1114 fprintf (stderr, "/notcategoryspec");
1115 mcnt = *p++;
1116 fprintf (stderr, "/%d", mcnt);
1117 break;
1118 # endif /* emacs */
1119
1120 case begbuf:
1121 fprintf (stderr, "/begbuf");
1122 break;
1123
1124 case endbuf:
1125 fprintf (stderr, "/endbuf");
1126 break;
1127
1128 default:
1129 fprintf (stderr, "?%d", *(p-1));
1130 }
1131
1132 fprintf (stderr, "\n");
1133 }
1134
1135 fprintf (stderr, "%td:\tend of pattern.\n", p - start);
1136 }
1137
1138
1139 static void
1140 print_compiled_pattern (struct re_pattern_buffer *bufp)
1141 {
1142 re_char *buffer = bufp->buffer;
1143
1144 print_partial_compiled_pattern (buffer, buffer + bufp->used);
1145 printf ("%ld bytes used/%ld bytes allocated.\n",
1146 bufp->used, bufp->allocated);
1147
1148 if (bufp->fastmap_accurate && bufp->fastmap)
1149 {
1150 printf ("fastmap: ");
1151 print_fastmap (bufp->fastmap);
1152 }
1153
1154 printf ("re_nsub: %zu\t", bufp->re_nsub);
1155 printf ("regs_alloc: %d\t", bufp->regs_allocated);
1156 printf ("can_be_null: %d\t", bufp->can_be_null);
1157 printf ("no_sub: %d\t", bufp->no_sub);
1158 printf ("not_bol: %d\t", bufp->not_bol);
1159 printf ("not_eol: %d\t", bufp->not_eol);
1160 printf ("syntax: %lx\n", bufp->syntax);
1161 fflush (stdout);
1162 /* Perhaps we should print the translate table? */
1163 }
1164
1165
1166 static void
1167 print_double_string (re_char *where, re_char *string1, ssize_t size1,
1168 re_char *string2, ssize_t size2)
1169 {
1170 ssize_t this_char;
1171
1172 if (where == NULL)
1173 printf ("(null)");
1174 else
1175 {
1176 if (FIRST_STRING_P (where))
1177 {
1178 for (this_char = where - string1; this_char < size1; this_char++)
1179 putchar (string1[this_char]);
1180
1181 where = string2;
1182 }
1183
1184 for (this_char = where - string2; this_char < size2; this_char++)
1185 putchar (string2[this_char]);
1186 }
1187 }
1188
1189 #else /* not DEBUG */
1190
1191 # undef assert
1192 # define assert(e)
1193
1194 # define DEBUG_STATEMENT(e)
1195 # if __STDC_VERSION__ < 199901L
1196 # define DEBUG_COMPILES_ARGUMENTS
1197 # define DEBUG_PRINT /* 'DEBUG_PRINT (x, y)' discards X and Y. */ (void)
1198 # else
1199 # define DEBUG_PRINT(...)
1200 # endif
1201 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e)
1202 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2)
1203
1204 #endif /* not DEBUG */
1205 \f
1206 /* Use this to suppress gcc's `...may be used before initialized' warnings. */
1207 #ifdef lint
1208 # define IF_LINT(Code) Code
1209 #else
1210 # define IF_LINT(Code) /* empty */
1211 #endif
1212 \f
1213 #ifndef emacs
1214 /* Set by `re_set_syntax' to the current regexp syntax to recognize. Can
1215 also be assigned to arbitrarily: each pattern buffer stores its own
1216 syntax, so it can be changed between regex compilations. */
1217 /* This has no initializer because initialized variables in Emacs
1218 become read-only after dumping. */
1219 reg_syntax_t re_syntax_options;
1220 #endif
1221
1222
1223 /* Specify the precise syntax of regexps for compilation. This provides
1224 for compatibility for various utilities which historically have
1225 different, incompatible syntaxes.
1226
1227 The argument SYNTAX is a bit mask comprised of the various bits
1228 defined in regex.h. We return the old syntax. */
1229
1230 reg_syntax_t
1231 re_set_syntax (reg_syntax_t syntax)
1232 {
1233 reg_syntax_t ret = re_syntax_options;
1234
1235 re_syntax_options = syntax;
1236 return ret;
1237 }
1238 WEAK_ALIAS (__re_set_syntax, re_set_syntax)
1239
1240 #ifndef emacs
1241 /* Regexp to use to replace spaces, or NULL meaning don't. */
1242 static re_char *whitespace_regexp;
1243 #endif
1244
1245 void
1246 re_set_whitespace_regexp (const char *regexp)
1247 {
1248 whitespace_regexp = (re_char *) regexp;
1249 }
1250 WEAK_ALIAS (__re_set_syntax, re_set_syntax)
1251 \f
1252 /* This table gives an error message for each of the error codes listed
1253 in regex.h. Obviously the order here has to be same as there.
1254 POSIX doesn't require that we do anything for REG_NOERROR,
1255 but why not be nice? */
1256
1257 static const char *re_error_msgid[] =
1258 {
1259 gettext_noop ("Success"), /* REG_NOERROR */
1260 gettext_noop ("No match"), /* REG_NOMATCH */
1261 gettext_noop ("Invalid regular expression"), /* REG_BADPAT */
1262 gettext_noop ("Invalid collation character"), /* REG_ECOLLATE */
1263 gettext_noop ("Invalid character class name"), /* REG_ECTYPE */
1264 gettext_noop ("Trailing backslash"), /* REG_EESCAPE */
1265 gettext_noop ("Invalid back reference"), /* REG_ESUBREG */
1266 gettext_noop ("Unmatched [ or [^"), /* REG_EBRACK */
1267 gettext_noop ("Unmatched ( or \\("), /* REG_EPAREN */
1268 gettext_noop ("Unmatched \\{"), /* REG_EBRACE */
1269 gettext_noop ("Invalid content of \\{\\}"), /* REG_BADBR */
1270 gettext_noop ("Invalid range end"), /* REG_ERANGE */
1271 gettext_noop ("Memory exhausted"), /* REG_ESPACE */
1272 gettext_noop ("Invalid preceding regular expression"), /* REG_BADRPT */
1273 gettext_noop ("Premature end of regular expression"), /* REG_EEND */
1274 gettext_noop ("Regular expression too big"), /* REG_ESIZE */
1275 gettext_noop ("Unmatched ) or \\)"), /* REG_ERPAREN */
1276 gettext_noop ("Range striding over charsets") /* REG_ERANGEX */
1277 };
1278 \f
1279 /* Avoiding alloca during matching, to placate r_alloc. */
1280
1281 /* Define MATCH_MAY_ALLOCATE unless we need to make sure that the
1282 searching and matching functions should not call alloca. On some
1283 systems, alloca is implemented in terms of malloc, and if we're
1284 using the relocating allocator routines, then malloc could cause a
1285 relocation, which might (if the strings being searched are in the
1286 ralloc heap) shift the data out from underneath the regexp
1287 routines.
1288
1289 Here's another reason to avoid allocation: Emacs
1290 processes input from X in a signal handler; processing X input may
1291 call malloc; if input arrives while a matching routine is calling
1292 malloc, then we're scrod. But Emacs can't just block input while
1293 calling matching routines; then we don't notice interrupts when
1294 they come in. So, Emacs blocks input around all regexp calls
1295 except the matching calls, which it leaves unprotected, in the
1296 faith that they will not malloc. */
1297
1298 /* Normally, this is fine. */
1299 #define MATCH_MAY_ALLOCATE
1300
1301 /* The match routines may not allocate if (1) they would do it with malloc
1302 and (2) it's not safe for them to use malloc.
1303 Note that if REL_ALLOC is defined, matching would not use malloc for the
1304 failure stack, but we would still use it for the register vectors;
1305 so REL_ALLOC should not affect this. */
1306 #if defined REGEX_MALLOC && defined emacs
1307 # undef MATCH_MAY_ALLOCATE
1308 #endif
1309
1310 \f
1311 /* Failure stack declarations and macros; both re_compile_fastmap and
1312 re_match_2 use a failure stack. These have to be macros because of
1313 REGEX_ALLOCATE_STACK. */
1314
1315
1316 /* Approximate number of failure points for which to initially allocate space
1317 when matching. If this number is exceeded, we allocate more
1318 space, so it is not a hard limit. */
1319 #ifndef INIT_FAILURE_ALLOC
1320 # define INIT_FAILURE_ALLOC 20
1321 #endif
1322
1323 /* Roughly the maximum number of failure points on the stack. Would be
1324 exactly that if always used TYPICAL_FAILURE_SIZE items each time we failed.
1325 This is a variable only so users of regex can assign to it; we never
1326 change it ourselves. We always multiply it by TYPICAL_FAILURE_SIZE
1327 before using it, so it should probably be a byte-count instead. */
1328 # if defined MATCH_MAY_ALLOCATE
1329 /* Note that 4400 was enough to cause a crash on Alpha OSF/1,
1330 whose default stack limit is 2mb. In order for a larger
1331 value to work reliably, you have to try to make it accord
1332 with the process stack limit. */
1333 size_t re_max_failures = 40000;
1334 # else
1335 size_t re_max_failures = 4000;
1336 # endif
1337
1338 union fail_stack_elt
1339 {
1340 re_char *pointer;
1341 /* This should be the biggest `int' that's no bigger than a pointer. */
1342 long integer;
1343 };
1344
1345 typedef union fail_stack_elt fail_stack_elt_t;
1346
1347 typedef struct
1348 {
1349 fail_stack_elt_t *stack;
1350 size_t size;
1351 size_t avail; /* Offset of next open position. */
1352 size_t frame; /* Offset of the cur constructed frame. */
1353 } fail_stack_type;
1354
1355 #define FAIL_STACK_EMPTY() (fail_stack.frame == 0)
1356
1357
1358 /* Define macros to initialize and free the failure stack.
1359 Do `return -2' if the alloc fails. */
1360
1361 #ifdef MATCH_MAY_ALLOCATE
1362 # define INIT_FAIL_STACK() \
1363 do { \
1364 fail_stack.stack = \
1365 REGEX_ALLOCATE_STACK (INIT_FAILURE_ALLOC * TYPICAL_FAILURE_SIZE \
1366 * sizeof (fail_stack_elt_t)); \
1367 \
1368 if (fail_stack.stack == NULL) \
1369 return -2; \
1370 \
1371 fail_stack.size = INIT_FAILURE_ALLOC; \
1372 fail_stack.avail = 0; \
1373 fail_stack.frame = 0; \
1374 } while (0)
1375 #else
1376 # define INIT_FAIL_STACK() \
1377 do { \
1378 fail_stack.avail = 0; \
1379 fail_stack.frame = 0; \
1380 } while (0)
1381
1382 # define RETALLOC_IF(addr, n, t) \
1383 if (addr) RETALLOC((addr), (n), t); else (addr) = TALLOC ((n), t)
1384 #endif
1385
1386
1387 /* Double the size of FAIL_STACK, up to a limit
1388 which allows approximately `re_max_failures' items.
1389
1390 Return 1 if succeeds, and 0 if either ran out of memory
1391 allocating space for it or it was already too large.
1392
1393 REGEX_REALLOCATE_STACK requires `destination' be declared. */
1394
1395 /* Factor to increase the failure stack size by
1396 when we increase it.
1397 This used to be 2, but 2 was too wasteful
1398 because the old discarded stacks added up to as much space
1399 were as ultimate, maximum-size stack. */
1400 #define FAIL_STACK_GROWTH_FACTOR 4
1401
1402 #define GROW_FAIL_STACK(fail_stack) \
1403 (((fail_stack).size * sizeof (fail_stack_elt_t) \
1404 >= re_max_failures * TYPICAL_FAILURE_SIZE) \
1405 ? 0 \
1406 : ((fail_stack).stack \
1407 = REGEX_REALLOCATE_STACK ((fail_stack).stack, \
1408 (fail_stack).size * sizeof (fail_stack_elt_t), \
1409 MIN (re_max_failures * TYPICAL_FAILURE_SIZE, \
1410 ((fail_stack).size * sizeof (fail_stack_elt_t) \
1411 * FAIL_STACK_GROWTH_FACTOR))), \
1412 \
1413 (fail_stack).stack == NULL \
1414 ? 0 \
1415 : ((fail_stack).size \
1416 = (MIN (re_max_failures * TYPICAL_FAILURE_SIZE, \
1417 ((fail_stack).size * sizeof (fail_stack_elt_t) \
1418 * FAIL_STACK_GROWTH_FACTOR)) \
1419 / sizeof (fail_stack_elt_t)), \
1420 1)))
1421
1422
1423 /* Push a pointer value onto the failure stack.
1424 Assumes the variable `fail_stack'. Probably should only
1425 be called from within `PUSH_FAILURE_POINT'. */
1426 #define PUSH_FAILURE_POINTER(item) \
1427 fail_stack.stack[fail_stack.avail++].pointer = (item)
1428
1429 /* This pushes an integer-valued item onto the failure stack.
1430 Assumes the variable `fail_stack'. Probably should only
1431 be called from within `PUSH_FAILURE_POINT'. */
1432 #define PUSH_FAILURE_INT(item) \
1433 fail_stack.stack[fail_stack.avail++].integer = (item)
1434
1435 /* These POP... operations complement the PUSH... operations.
1436 All assume that `fail_stack' is nonempty. */
1437 #define POP_FAILURE_POINTER() fail_stack.stack[--fail_stack.avail].pointer
1438 #define POP_FAILURE_INT() fail_stack.stack[--fail_stack.avail].integer
1439
1440 /* Individual items aside from the registers. */
1441 #define NUM_NONREG_ITEMS 3
1442
1443 /* Used to examine the stack (to detect infinite loops). */
1444 #define FAILURE_PAT(h) fail_stack.stack[(h) - 1].pointer
1445 #define FAILURE_STR(h) (fail_stack.stack[(h) - 2].pointer)
1446 #define NEXT_FAILURE_HANDLE(h) fail_stack.stack[(h) - 3].integer
1447 #define TOP_FAILURE_HANDLE() fail_stack.frame
1448
1449
1450 #define ENSURE_FAIL_STACK(space) \
1451 while (REMAINING_AVAIL_SLOTS <= space) { \
1452 if (!GROW_FAIL_STACK (fail_stack)) \
1453 return -2; \
1454 DEBUG_PRINT ("\n Doubled stack; size now: %zd\n", (fail_stack).size);\
1455 DEBUG_PRINT (" slots available: %zd\n", REMAINING_AVAIL_SLOTS);\
1456 }
1457
1458 /* Push register NUM onto the stack. */
1459 #define PUSH_FAILURE_REG(num) \
1460 do { \
1461 char *destination; \
1462 long n = num; \
1463 ENSURE_FAIL_STACK(3); \
1464 DEBUG_PRINT (" Push reg %ld (spanning %p -> %p)\n", \
1465 n, regstart[n], regend[n]); \
1466 PUSH_FAILURE_POINTER (regstart[n]); \
1467 PUSH_FAILURE_POINTER (regend[n]); \
1468 PUSH_FAILURE_INT (n); \
1469 } while (0)
1470
1471 /* Change the counter's value to VAL, but make sure that it will
1472 be reset when backtracking. */
1473 #define PUSH_NUMBER(ptr,val) \
1474 do { \
1475 char *destination; \
1476 int c; \
1477 ENSURE_FAIL_STACK(3); \
1478 EXTRACT_NUMBER (c, ptr); \
1479 DEBUG_PRINT (" Push number %p = %d -> %d\n", ptr, c, val); \
1480 PUSH_FAILURE_INT (c); \
1481 PUSH_FAILURE_POINTER (ptr); \
1482 PUSH_FAILURE_INT (-1); \
1483 STORE_NUMBER (ptr, val); \
1484 } while (0)
1485
1486 /* Pop a saved register off the stack. */
1487 #define POP_FAILURE_REG_OR_COUNT() \
1488 do { \
1489 long pfreg = POP_FAILURE_INT (); \
1490 if (pfreg == -1) \
1491 { \
1492 /* It's a counter. */ \
1493 /* Here, we discard `const', making re_match non-reentrant. */ \
1494 unsigned char *ptr = (unsigned char*) POP_FAILURE_POINTER (); \
1495 pfreg = POP_FAILURE_INT (); \
1496 STORE_NUMBER (ptr, pfreg); \
1497 DEBUG_PRINT (" Pop counter %p = %ld\n", ptr, pfreg); \
1498 } \
1499 else \
1500 { \
1501 regend[pfreg] = POP_FAILURE_POINTER (); \
1502 regstart[pfreg] = POP_FAILURE_POINTER (); \
1503 DEBUG_PRINT (" Pop reg %ld (spanning %p -> %p)\n", \
1504 pfreg, regstart[pfreg], regend[pfreg]); \
1505 } \
1506 } while (0)
1507
1508 /* Check that we are not stuck in an infinite loop. */
1509 #define CHECK_INFINITE_LOOP(pat_cur, string_place) \
1510 do { \
1511 ssize_t failure = TOP_FAILURE_HANDLE (); \
1512 /* Check for infinite matching loops */ \
1513 while (failure > 0 \
1514 && (FAILURE_STR (failure) == string_place \
1515 || FAILURE_STR (failure) == NULL)) \
1516 { \
1517 assert (FAILURE_PAT (failure) >= bufp->buffer \
1518 && FAILURE_PAT (failure) <= bufp->buffer + bufp->used); \
1519 if (FAILURE_PAT (failure) == pat_cur) \
1520 { \
1521 cycle = 1; \
1522 break; \
1523 } \
1524 DEBUG_PRINT (" Other pattern: %p\n", FAILURE_PAT (failure)); \
1525 failure = NEXT_FAILURE_HANDLE(failure); \
1526 } \
1527 DEBUG_PRINT (" Other string: %p\n", FAILURE_STR (failure)); \
1528 } while (0)
1529
1530 /* Push the information about the state we will need
1531 if we ever fail back to it.
1532
1533 Requires variables fail_stack, regstart, regend and
1534 num_regs be declared. GROW_FAIL_STACK requires `destination' be
1535 declared.
1536
1537 Does `return FAILURE_CODE' if runs out of memory. */
1538
1539 #define PUSH_FAILURE_POINT(pattern, string_place) \
1540 do { \
1541 char *destination; \
1542 /* Must be int, so when we don't save any registers, the arithmetic \
1543 of 0 + -1 isn't done as unsigned. */ \
1544 \
1545 DEBUG_STATEMENT (nfailure_points_pushed++); \
1546 DEBUG_PRINT ("\nPUSH_FAILURE_POINT:\n"); \
1547 DEBUG_PRINT (" Before push, next avail: %zd\n", (fail_stack).avail); \
1548 DEBUG_PRINT (" size: %zd\n", (fail_stack).size);\
1549 \
1550 ENSURE_FAIL_STACK (NUM_NONREG_ITEMS); \
1551 \
1552 DEBUG_PRINT ("\n"); \
1553 \
1554 DEBUG_PRINT (" Push frame index: %zd\n", fail_stack.frame); \
1555 PUSH_FAILURE_INT (fail_stack.frame); \
1556 \
1557 DEBUG_PRINT (" Push string %p: `", string_place); \
1558 DEBUG_PRINT_DOUBLE_STRING (string_place, string1, size1, string2, size2);\
1559 DEBUG_PRINT ("'\n"); \
1560 PUSH_FAILURE_POINTER (string_place); \
1561 \
1562 DEBUG_PRINT (" Push pattern %p: ", pattern); \
1563 DEBUG_PRINT_COMPILED_PATTERN (bufp, pattern, pend); \
1564 PUSH_FAILURE_POINTER (pattern); \
1565 \
1566 /* Close the frame by moving the frame pointer past it. */ \
1567 fail_stack.frame = fail_stack.avail; \
1568 } while (0)
1569
1570 /* Estimate the size of data pushed by a typical failure stack entry.
1571 An estimate is all we need, because all we use this for
1572 is to choose a limit for how big to make the failure stack. */
1573 /* BEWARE, the value `20' is hard-coded in emacs.c:main(). */
1574 #define TYPICAL_FAILURE_SIZE 20
1575
1576 /* How many items can still be added to the stack without overflowing it. */
1577 #define REMAINING_AVAIL_SLOTS ((fail_stack).size - (fail_stack).avail)
1578
1579
1580 /* Pops what PUSH_FAIL_STACK pushes.
1581
1582 We restore into the parameters, all of which should be lvalues:
1583 STR -- the saved data position.
1584 PAT -- the saved pattern position.
1585 REGSTART, REGEND -- arrays of string positions.
1586
1587 Also assumes the variables `fail_stack' and (if debugging), `bufp',
1588 `pend', `string1', `size1', `string2', and `size2'. */
1589
1590 #define POP_FAILURE_POINT(str, pat) \
1591 do { \
1592 assert (!FAIL_STACK_EMPTY ()); \
1593 \
1594 /* Remove failure points and point to how many regs pushed. */ \
1595 DEBUG_PRINT ("POP_FAILURE_POINT:\n"); \
1596 DEBUG_PRINT (" Before pop, next avail: %zd\n", fail_stack.avail); \
1597 DEBUG_PRINT (" size: %zd\n", fail_stack.size); \
1598 \
1599 /* Pop the saved registers. */ \
1600 while (fail_stack.frame < fail_stack.avail) \
1601 POP_FAILURE_REG_OR_COUNT (); \
1602 \
1603 pat = POP_FAILURE_POINTER (); \
1604 DEBUG_PRINT (" Popping pattern %p: ", pat); \
1605 DEBUG_PRINT_COMPILED_PATTERN (bufp, pat, pend); \
1606 \
1607 /* If the saved string location is NULL, it came from an \
1608 on_failure_keep_string_jump opcode, and we want to throw away the \
1609 saved NULL, thus retaining our current position in the string. */ \
1610 str = POP_FAILURE_POINTER (); \
1611 DEBUG_PRINT (" Popping string %p: `", str); \
1612 DEBUG_PRINT_DOUBLE_STRING (str, string1, size1, string2, size2); \
1613 DEBUG_PRINT ("'\n"); \
1614 \
1615 fail_stack.frame = POP_FAILURE_INT (); \
1616 DEBUG_PRINT (" Popping frame index: %zd\n", fail_stack.frame); \
1617 \
1618 assert (fail_stack.avail >= 0); \
1619 assert (fail_stack.frame <= fail_stack.avail); \
1620 \
1621 DEBUG_STATEMENT (nfailure_points_popped++); \
1622 } while (0) /* POP_FAILURE_POINT */
1623
1624
1625 \f
1626 /* Registers are set to a sentinel when they haven't yet matched. */
1627 #define REG_UNSET(e) ((e) == NULL)
1628 \f
1629 /* Subroutine declarations and macros for regex_compile. */
1630
1631 static reg_errcode_t regex_compile (re_char *pattern, size_t size,
1632 reg_syntax_t syntax,
1633 struct re_pattern_buffer *bufp);
1634 static void store_op1 (re_opcode_t op, unsigned char *loc, int arg);
1635 static void store_op2 (re_opcode_t op, unsigned char *loc, int arg1, int arg2);
1636 static void insert_op1 (re_opcode_t op, unsigned char *loc,
1637 int arg, unsigned char *end);
1638 static void insert_op2 (re_opcode_t op, unsigned char *loc,
1639 int arg1, int arg2, unsigned char *end);
1640 static boolean at_begline_loc_p (re_char *pattern, re_char *p,
1641 reg_syntax_t syntax);
1642 static boolean at_endline_loc_p (re_char *p, re_char *pend,
1643 reg_syntax_t syntax);
1644 static re_char *skip_one_char (re_char *p);
1645 static int analyse_first (re_char *p, re_char *pend,
1646 char *fastmap, const int multibyte);
1647
1648 /* Fetch the next character in the uncompiled pattern, with no
1649 translation. */
1650 #define PATFETCH(c) \
1651 do { \
1652 int len; \
1653 if (p == pend) return REG_EEND; \
1654 c = RE_STRING_CHAR_AND_LENGTH (p, len, multibyte); \
1655 p += len; \
1656 } while (0)
1657
1658
1659 /* If `translate' is non-null, return translate[D], else just D. We
1660 cast the subscript to translate because some data is declared as
1661 `char *', to avoid warnings when a string constant is passed. But
1662 when we use a character as a subscript we must make it unsigned. */
1663 #ifndef TRANSLATE
1664 # define TRANSLATE(d) \
1665 (RE_TRANSLATE_P (translate) ? RE_TRANSLATE (translate, (d)) : (d))
1666 #endif
1667
1668
1669 /* Macros for outputting the compiled pattern into `buffer'. */
1670
1671 /* If the buffer isn't allocated when it comes in, use this. */
1672 #define INIT_BUF_SIZE 32
1673
1674 /* Make sure we have at least N more bytes of space in buffer. */
1675 #define GET_BUFFER_SPACE(n) \
1676 while ((size_t) (b - bufp->buffer + (n)) > bufp->allocated) \
1677 EXTEND_BUFFER ()
1678
1679 /* Make sure we have one more byte of buffer space and then add C to it. */
1680 #define BUF_PUSH(c) \
1681 do { \
1682 GET_BUFFER_SPACE (1); \
1683 *b++ = (unsigned char) (c); \
1684 } while (0)
1685
1686
1687 /* Ensure we have two more bytes of buffer space and then append C1 and C2. */
1688 #define BUF_PUSH_2(c1, c2) \
1689 do { \
1690 GET_BUFFER_SPACE (2); \
1691 *b++ = (unsigned char) (c1); \
1692 *b++ = (unsigned char) (c2); \
1693 } while (0)
1694
1695
1696 /* Store a jump with opcode OP at LOC to location TO. We store a
1697 relative address offset by the three bytes the jump itself occupies. */
1698 #define STORE_JUMP(op, loc, to) \
1699 store_op1 (op, loc, (to) - (loc) - 3)
1700
1701 /* Likewise, for a two-argument jump. */
1702 #define STORE_JUMP2(op, loc, to, arg) \
1703 store_op2 (op, loc, (to) - (loc) - 3, arg)
1704
1705 /* Like `STORE_JUMP', but for inserting. Assume `b' is the buffer end. */
1706 #define INSERT_JUMP(op, loc, to) \
1707 insert_op1 (op, loc, (to) - (loc) - 3, b)
1708
1709 /* Like `STORE_JUMP2', but for inserting. Assume `b' is the buffer end. */
1710 #define INSERT_JUMP2(op, loc, to, arg) \
1711 insert_op2 (op, loc, (to) - (loc) - 3, arg, b)
1712
1713
1714 /* This is not an arbitrary limit: the arguments which represent offsets
1715 into the pattern are two bytes long. So if 2^15 bytes turns out to
1716 be too small, many things would have to change. */
1717 # define MAX_BUF_SIZE (1L << 15)
1718
1719 /* Extend the buffer by twice its current size via realloc and
1720 reset the pointers that pointed into the old block to point to the
1721 correct places in the new one. If extending the buffer results in it
1722 being larger than MAX_BUF_SIZE, then flag memory exhausted. */
1723 #if __BOUNDED_POINTERS__
1724 # define SET_HIGH_BOUND(P) (__ptrhigh (P) = __ptrlow (P) + bufp->allocated)
1725 # define MOVE_BUFFER_POINTER(P) \
1726 (__ptrlow (P) = new_buffer + (__ptrlow (P) - old_buffer), \
1727 SET_HIGH_BOUND (P), \
1728 __ptrvalue (P) = new_buffer + (__ptrvalue (P) - old_buffer))
1729 # define ELSE_EXTEND_BUFFER_HIGH_BOUND \
1730 else \
1731 { \
1732 SET_HIGH_BOUND (b); \
1733 SET_HIGH_BOUND (begalt); \
1734 if (fixup_alt_jump) \
1735 SET_HIGH_BOUND (fixup_alt_jump); \
1736 if (laststart) \
1737 SET_HIGH_BOUND (laststart); \
1738 if (pending_exact) \
1739 SET_HIGH_BOUND (pending_exact); \
1740 }
1741 #else
1742 # define MOVE_BUFFER_POINTER(P) ((P) = new_buffer + ((P) - old_buffer))
1743 # define ELSE_EXTEND_BUFFER_HIGH_BOUND
1744 #endif
1745 #define EXTEND_BUFFER() \
1746 do { \
1747 unsigned char *old_buffer = bufp->buffer; \
1748 if (bufp->allocated == MAX_BUF_SIZE) \
1749 return REG_ESIZE; \
1750 bufp->allocated <<= 1; \
1751 if (bufp->allocated > MAX_BUF_SIZE) \
1752 bufp->allocated = MAX_BUF_SIZE; \
1753 RETALLOC (bufp->buffer, bufp->allocated, unsigned char); \
1754 if (bufp->buffer == NULL) \
1755 return REG_ESPACE; \
1756 /* If the buffer moved, move all the pointers into it. */ \
1757 if (old_buffer != bufp->buffer) \
1758 { \
1759 unsigned char *new_buffer = bufp->buffer; \
1760 MOVE_BUFFER_POINTER (b); \
1761 MOVE_BUFFER_POINTER (begalt); \
1762 if (fixup_alt_jump) \
1763 MOVE_BUFFER_POINTER (fixup_alt_jump); \
1764 if (laststart) \
1765 MOVE_BUFFER_POINTER (laststart); \
1766 if (pending_exact) \
1767 MOVE_BUFFER_POINTER (pending_exact); \
1768 } \
1769 ELSE_EXTEND_BUFFER_HIGH_BOUND \
1770 } while (0)
1771
1772
1773 /* Since we have one byte reserved for the register number argument to
1774 {start,stop}_memory, the maximum number of groups we can report
1775 things about is what fits in that byte. */
1776 #define MAX_REGNUM 255
1777
1778 /* But patterns can have more than `MAX_REGNUM' registers. We just
1779 ignore the excess. */
1780 typedef int regnum_t;
1781
1782
1783 /* Macros for the compile stack. */
1784
1785 /* Since offsets can go either forwards or backwards, this type needs to
1786 be able to hold values from -(MAX_BUF_SIZE - 1) to MAX_BUF_SIZE - 1. */
1787 /* int may be not enough when sizeof(int) == 2. */
1788 typedef long pattern_offset_t;
1789
1790 typedef struct
1791 {
1792 pattern_offset_t begalt_offset;
1793 pattern_offset_t fixup_alt_jump;
1794 pattern_offset_t laststart_offset;
1795 regnum_t regnum;
1796 } compile_stack_elt_t;
1797
1798
1799 typedef struct
1800 {
1801 compile_stack_elt_t *stack;
1802 size_t size;
1803 size_t avail; /* Offset of next open position. */
1804 } compile_stack_type;
1805
1806
1807 #define INIT_COMPILE_STACK_SIZE 32
1808
1809 #define COMPILE_STACK_EMPTY (compile_stack.avail == 0)
1810 #define COMPILE_STACK_FULL (compile_stack.avail == compile_stack.size)
1811
1812 /* The next available element. */
1813 #define COMPILE_STACK_TOP (compile_stack.stack[compile_stack.avail])
1814
1815 /* Explicit quit checking is needed for Emacs, which uses polling to
1816 process input events. */
1817 #ifdef emacs
1818 # define IMMEDIATE_QUIT_CHECK \
1819 do { \
1820 if (immediate_quit) QUIT; \
1821 } while (0)
1822 #else
1823 # define IMMEDIATE_QUIT_CHECK ((void)0)
1824 #endif
1825 \f
1826 /* Structure to manage work area for range table. */
1827 struct range_table_work_area
1828 {
1829 int *table; /* actual work area. */
1830 int allocated; /* allocated size for work area in bytes. */
1831 int used; /* actually used size in words. */
1832 int bits; /* flag to record character classes */
1833 };
1834
1835 /* Make sure that WORK_AREA can hold more N multibyte characters.
1836 This is used only in set_image_of_range and set_image_of_range_1.
1837 It expects WORK_AREA to be a pointer.
1838 If it can't get the space, it returns from the surrounding function. */
1839
1840 #define EXTEND_RANGE_TABLE(work_area, n) \
1841 do { \
1842 if (((work_area).used + (n)) * sizeof (int) > (work_area).allocated) \
1843 { \
1844 extend_range_table_work_area (&work_area); \
1845 if ((work_area).table == 0) \
1846 return (REG_ESPACE); \
1847 } \
1848 } while (0)
1849
1850 #define SET_RANGE_TABLE_WORK_AREA_BIT(work_area, bit) \
1851 (work_area).bits |= (bit)
1852
1853 /* Bits used to implement the multibyte-part of the various character classes
1854 such as [:alnum:] in a charset's range table. */
1855 #define BIT_WORD 0x1
1856 #define BIT_LOWER 0x2
1857 #define BIT_PUNCT 0x4
1858 #define BIT_SPACE 0x8
1859 #define BIT_UPPER 0x10
1860 #define BIT_MULTIBYTE 0x20
1861
1862 /* Set a range (RANGE_START, RANGE_END) to WORK_AREA. */
1863 #define SET_RANGE_TABLE_WORK_AREA(work_area, range_start, range_end) \
1864 do { \
1865 EXTEND_RANGE_TABLE ((work_area), 2); \
1866 (work_area).table[(work_area).used++] = (range_start); \
1867 (work_area).table[(work_area).used++] = (range_end); \
1868 } while (0)
1869
1870 /* Free allocated memory for WORK_AREA. */
1871 #define FREE_RANGE_TABLE_WORK_AREA(work_area) \
1872 do { \
1873 if ((work_area).table) \
1874 free ((work_area).table); \
1875 } while (0)
1876
1877 #define CLEAR_RANGE_TABLE_WORK_USED(work_area) ((work_area).used = 0, (work_area).bits = 0)
1878 #define RANGE_TABLE_WORK_USED(work_area) ((work_area).used)
1879 #define RANGE_TABLE_WORK_BITS(work_area) ((work_area).bits)
1880 #define RANGE_TABLE_WORK_ELT(work_area, i) ((work_area).table[i])
1881 \f
1882
1883 /* Set the bit for character C in a list. */
1884 #define SET_LIST_BIT(c) (b[((c)) / BYTEWIDTH] |= 1 << ((c) % BYTEWIDTH))
1885
1886
1887 #ifdef emacs
1888
1889 /* Store characters in the range FROM to TO in the bitmap at B (for
1890 ASCII and unibyte characters) and WORK_AREA (for multibyte
1891 characters) while translating them and paying attention to the
1892 continuity of translated characters.
1893
1894 Implementation note: It is better to implement these fairly big
1895 macros by a function, but it's not that easy because macros called
1896 in this macro assume various local variables already declared. */
1897
1898 /* Both FROM and TO are ASCII characters. */
1899
1900 #define SETUP_ASCII_RANGE(work_area, FROM, TO) \
1901 do { \
1902 int C0, C1; \
1903 \
1904 for (C0 = (FROM); C0 <= (TO); C0++) \
1905 { \
1906 C1 = TRANSLATE (C0); \
1907 if (! ASCII_CHAR_P (C1)) \
1908 { \
1909 SET_RANGE_TABLE_WORK_AREA ((work_area), C1, C1); \
1910 if ((C1 = RE_CHAR_TO_UNIBYTE (C1)) < 0) \
1911 C1 = C0; \
1912 } \
1913 SET_LIST_BIT (C1); \
1914 } \
1915 } while (0)
1916
1917
1918 /* Both FROM and TO are unibyte characters (0x80..0xFF). */
1919
1920 #define SETUP_UNIBYTE_RANGE(work_area, FROM, TO) \
1921 do { \
1922 int C0, C1, C2, I; \
1923 int USED = RANGE_TABLE_WORK_USED (work_area); \
1924 \
1925 for (C0 = (FROM); C0 <= (TO); C0++) \
1926 { \
1927 C1 = RE_CHAR_TO_MULTIBYTE (C0); \
1928 if (CHAR_BYTE8_P (C1)) \
1929 SET_LIST_BIT (C0); \
1930 else \
1931 { \
1932 C2 = TRANSLATE (C1); \
1933 if (C2 == C1 \
1934 || (C1 = RE_CHAR_TO_UNIBYTE (C2)) < 0) \
1935 C1 = C0; \
1936 SET_LIST_BIT (C1); \
1937 for (I = RANGE_TABLE_WORK_USED (work_area) - 2; I >= USED; I -= 2) \
1938 { \
1939 int from = RANGE_TABLE_WORK_ELT (work_area, I); \
1940 int to = RANGE_TABLE_WORK_ELT (work_area, I + 1); \
1941 \
1942 if (C2 >= from - 1 && C2 <= to + 1) \
1943 { \
1944 if (C2 == from - 1) \
1945 RANGE_TABLE_WORK_ELT (work_area, I)--; \
1946 else if (C2 == to + 1) \
1947 RANGE_TABLE_WORK_ELT (work_area, I + 1)++; \
1948 break; \
1949 } \
1950 } \
1951 if (I < USED) \
1952 SET_RANGE_TABLE_WORK_AREA ((work_area), C2, C2); \
1953 } \
1954 } \
1955 } while (0)
1956
1957
1958 /* Both FROM and TO are multibyte characters. */
1959
1960 #define SETUP_MULTIBYTE_RANGE(work_area, FROM, TO) \
1961 do { \
1962 int C0, C1, C2, I, USED = RANGE_TABLE_WORK_USED (work_area); \
1963 \
1964 SET_RANGE_TABLE_WORK_AREA ((work_area), (FROM), (TO)); \
1965 for (C0 = (FROM); C0 <= (TO); C0++) \
1966 { \
1967 C1 = TRANSLATE (C0); \
1968 if ((C2 = RE_CHAR_TO_UNIBYTE (C1)) >= 0 \
1969 || (C1 != C0 && (C2 = RE_CHAR_TO_UNIBYTE (C0)) >= 0)) \
1970 SET_LIST_BIT (C2); \
1971 if (C1 >= (FROM) && C1 <= (TO)) \
1972 continue; \
1973 for (I = RANGE_TABLE_WORK_USED (work_area) - 2; I >= USED; I -= 2) \
1974 { \
1975 int from = RANGE_TABLE_WORK_ELT (work_area, I); \
1976 int to = RANGE_TABLE_WORK_ELT (work_area, I + 1); \
1977 \
1978 if (C1 >= from - 1 && C1 <= to + 1) \
1979 { \
1980 if (C1 == from - 1) \
1981 RANGE_TABLE_WORK_ELT (work_area, I)--; \
1982 else if (C1 == to + 1) \
1983 RANGE_TABLE_WORK_ELT (work_area, I + 1)++; \
1984 break; \
1985 } \
1986 } \
1987 if (I < USED) \
1988 SET_RANGE_TABLE_WORK_AREA ((work_area), C1, C1); \
1989 } \
1990 } while (0)
1991
1992 #endif /* emacs */
1993
1994 /* Get the next unsigned number in the uncompiled pattern. */
1995 #define GET_UNSIGNED_NUMBER(num) \
1996 do { \
1997 if (p == pend) \
1998 FREE_STACK_RETURN (REG_EBRACE); \
1999 else \
2000 { \
2001 PATFETCH (c); \
2002 while ('0' <= c && c <= '9') \
2003 { \
2004 int prev; \
2005 if (num < 0) \
2006 num = 0; \
2007 prev = num; \
2008 num = num * 10 + c - '0'; \
2009 if (num / 10 != prev) \
2010 FREE_STACK_RETURN (REG_BADBR); \
2011 if (p == pend) \
2012 FREE_STACK_RETURN (REG_EBRACE); \
2013 PATFETCH (c); \
2014 } \
2015 } \
2016 } while (0)
2017 \f
2018 #if ! WIDE_CHAR_SUPPORT
2019
2020 /* Map a string to the char class it names (if any). */
2021 re_wctype_t
2022 re_wctype (const re_char *str)
2023 {
2024 const char *string = (const char *) str;
2025 if (STREQ (string, "alnum")) return RECC_ALNUM;
2026 else if (STREQ (string, "alpha")) return RECC_ALPHA;
2027 else if (STREQ (string, "word")) return RECC_WORD;
2028 else if (STREQ (string, "ascii")) return RECC_ASCII;
2029 else if (STREQ (string, "nonascii")) return RECC_NONASCII;
2030 else if (STREQ (string, "graph")) return RECC_GRAPH;
2031 else if (STREQ (string, "lower")) return RECC_LOWER;
2032 else if (STREQ (string, "print")) return RECC_PRINT;
2033 else if (STREQ (string, "punct")) return RECC_PUNCT;
2034 else if (STREQ (string, "space")) return RECC_SPACE;
2035 else if (STREQ (string, "upper")) return RECC_UPPER;
2036 else if (STREQ (string, "unibyte")) return RECC_UNIBYTE;
2037 else if (STREQ (string, "multibyte")) return RECC_MULTIBYTE;
2038 else if (STREQ (string, "digit")) return RECC_DIGIT;
2039 else if (STREQ (string, "xdigit")) return RECC_XDIGIT;
2040 else if (STREQ (string, "cntrl")) return RECC_CNTRL;
2041 else if (STREQ (string, "blank")) return RECC_BLANK;
2042 else return 0;
2043 }
2044
2045 /* True if CH is in the char class CC. */
2046 boolean
2047 re_iswctype (int ch, re_wctype_t cc)
2048 {
2049 switch (cc)
2050 {
2051 case RECC_ALNUM: return ISALNUM (ch) != 0;
2052 case RECC_ALPHA: return ISALPHA (ch) != 0;
2053 case RECC_BLANK: return ISBLANK (ch) != 0;
2054 case RECC_CNTRL: return ISCNTRL (ch) != 0;
2055 case RECC_DIGIT: return ISDIGIT (ch) != 0;
2056 case RECC_GRAPH: return ISGRAPH (ch) != 0;
2057 case RECC_LOWER: return ISLOWER (ch) != 0;
2058 case RECC_PRINT: return ISPRINT (ch) != 0;
2059 case RECC_PUNCT: return ISPUNCT (ch) != 0;
2060 case RECC_SPACE: return ISSPACE (ch) != 0;
2061 case RECC_UPPER: return ISUPPER (ch) != 0;
2062 case RECC_XDIGIT: return ISXDIGIT (ch) != 0;
2063 case RECC_ASCII: return IS_REAL_ASCII (ch) != 0;
2064 case RECC_NONASCII: return !IS_REAL_ASCII (ch);
2065 case RECC_UNIBYTE: return ISUNIBYTE (ch) != 0;
2066 case RECC_MULTIBYTE: return !ISUNIBYTE (ch);
2067 case RECC_WORD: return ISWORD (ch) != 0;
2068 case RECC_ERROR: return false;
2069 default:
2070 abort ();
2071 }
2072 }
2073
2074 /* Return a bit-pattern to use in the range-table bits to match multibyte
2075 chars of class CC. */
2076 static int
2077 re_wctype_to_bit (re_wctype_t cc)
2078 {
2079 switch (cc)
2080 {
2081 case RECC_NONASCII: case RECC_PRINT: case RECC_GRAPH:
2082 case RECC_MULTIBYTE: return BIT_MULTIBYTE;
2083 case RECC_ALPHA: case RECC_ALNUM: case RECC_WORD: return BIT_WORD;
2084 case RECC_LOWER: return BIT_LOWER;
2085 case RECC_UPPER: return BIT_UPPER;
2086 case RECC_PUNCT: return BIT_PUNCT;
2087 case RECC_SPACE: return BIT_SPACE;
2088 case RECC_ASCII: case RECC_DIGIT: case RECC_XDIGIT: case RECC_CNTRL:
2089 case RECC_BLANK: case RECC_UNIBYTE: case RECC_ERROR: return 0;
2090 default:
2091 abort ();
2092 }
2093 }
2094 #endif
2095 \f
2096 /* Filling in the work area of a range. */
2097
2098 /* Actually extend the space in WORK_AREA. */
2099
2100 static void
2101 extend_range_table_work_area (struct range_table_work_area *work_area)
2102 {
2103 work_area->allocated += 16 * sizeof (int);
2104 work_area->table = realloc (work_area->table, work_area->allocated);
2105 }
2106
2107 #if 0
2108 #ifdef emacs
2109
2110 /* Carefully find the ranges of codes that are equivalent
2111 under case conversion to the range start..end when passed through
2112 TRANSLATE. Handle the case where non-letters can come in between
2113 two upper-case letters (which happens in Latin-1).
2114 Also handle the case of groups of more than 2 case-equivalent chars.
2115
2116 The basic method is to look at consecutive characters and see
2117 if they can form a run that can be handled as one.
2118
2119 Returns -1 if successful, REG_ESPACE if ran out of space. */
2120
2121 static int
2122 set_image_of_range_1 (struct range_table_work_area *work_area,
2123 re_wchar_t start, re_wchar_t end,
2124 RE_TRANSLATE_TYPE translate)
2125 {
2126 /* `one_case' indicates a character, or a run of characters,
2127 each of which is an isolate (no case-equivalents).
2128 This includes all ASCII non-letters.
2129
2130 `two_case' indicates a character, or a run of characters,
2131 each of which has two case-equivalent forms.
2132 This includes all ASCII letters.
2133
2134 `strange' indicates a character that has more than one
2135 case-equivalent. */
2136
2137 enum case_type {one_case, two_case, strange};
2138
2139 /* Describe the run that is in progress,
2140 which the next character can try to extend.
2141 If run_type is strange, that means there really is no run.
2142 If run_type is one_case, then run_start...run_end is the run.
2143 If run_type is two_case, then the run is run_start...run_end,
2144 and the case-equivalents end at run_eqv_end. */
2145
2146 enum case_type run_type = strange;
2147 int run_start, run_end, run_eqv_end;
2148
2149 Lisp_Object eqv_table;
2150
2151 if (!RE_TRANSLATE_P (translate))
2152 {
2153 EXTEND_RANGE_TABLE (work_area, 2);
2154 work_area->table[work_area->used++] = (start);
2155 work_area->table[work_area->used++] = (end);
2156 return -1;
2157 }
2158
2159 eqv_table = XCHAR_TABLE (translate)->extras[2];
2160
2161 for (; start <= end; start++)
2162 {
2163 enum case_type this_type;
2164 int eqv = RE_TRANSLATE (eqv_table, start);
2165 int minchar, maxchar;
2166
2167 /* Classify this character */
2168 if (eqv == start)
2169 this_type = one_case;
2170 else if (RE_TRANSLATE (eqv_table, eqv) == start)
2171 this_type = two_case;
2172 else
2173 this_type = strange;
2174
2175 if (start < eqv)
2176 minchar = start, maxchar = eqv;
2177 else
2178 minchar = eqv, maxchar = start;
2179
2180 /* Can this character extend the run in progress? */
2181 if (this_type == strange || this_type != run_type
2182 || !(minchar == run_end + 1
2183 && (run_type == two_case
2184 ? maxchar == run_eqv_end + 1 : 1)))
2185 {
2186 /* No, end the run.
2187 Record each of its equivalent ranges. */
2188 if (run_type == one_case)
2189 {
2190 EXTEND_RANGE_TABLE (work_area, 2);
2191 work_area->table[work_area->used++] = run_start;
2192 work_area->table[work_area->used++] = run_end;
2193 }
2194 else if (run_type == two_case)
2195 {
2196 EXTEND_RANGE_TABLE (work_area, 4);
2197 work_area->table[work_area->used++] = run_start;
2198 work_area->table[work_area->used++] = run_end;
2199 work_area->table[work_area->used++]
2200 = RE_TRANSLATE (eqv_table, run_start);
2201 work_area->table[work_area->used++]
2202 = RE_TRANSLATE (eqv_table, run_end);
2203 }
2204 run_type = strange;
2205 }
2206
2207 if (this_type == strange)
2208 {
2209 /* For a strange character, add each of its equivalents, one
2210 by one. Don't start a range. */
2211 do
2212 {
2213 EXTEND_RANGE_TABLE (work_area, 2);
2214 work_area->table[work_area->used++] = eqv;
2215 work_area->table[work_area->used++] = eqv;
2216 eqv = RE_TRANSLATE (eqv_table, eqv);
2217 }
2218 while (eqv != start);
2219 }
2220
2221 /* Add this char to the run, or start a new run. */
2222 else if (run_type == strange)
2223 {
2224 /* Initialize a new range. */
2225 run_type = this_type;
2226 run_start = start;
2227 run_end = start;
2228 run_eqv_end = RE_TRANSLATE (eqv_table, run_end);
2229 }
2230 else
2231 {
2232 /* Extend a running range. */
2233 run_end = minchar;
2234 run_eqv_end = RE_TRANSLATE (eqv_table, run_end);
2235 }
2236 }
2237
2238 /* If a run is still in progress at the end, finish it now
2239 by recording its equivalent ranges. */
2240 if (run_type == one_case)
2241 {
2242 EXTEND_RANGE_TABLE (work_area, 2);
2243 work_area->table[work_area->used++] = run_start;
2244 work_area->table[work_area->used++] = run_end;
2245 }
2246 else if (run_type == two_case)
2247 {
2248 EXTEND_RANGE_TABLE (work_area, 4);
2249 work_area->table[work_area->used++] = run_start;
2250 work_area->table[work_area->used++] = run_end;
2251 work_area->table[work_area->used++]
2252 = RE_TRANSLATE (eqv_table, run_start);
2253 work_area->table[work_area->used++]
2254 = RE_TRANSLATE (eqv_table, run_end);
2255 }
2256
2257 return -1;
2258 }
2259
2260 #endif /* emacs */
2261
2262 /* Record the image of the range start..end when passed through
2263 TRANSLATE. This is not necessarily TRANSLATE(start)..TRANSLATE(end)
2264 and is not even necessarily contiguous.
2265 Normally we approximate it with the smallest contiguous range that contains
2266 all the chars we need. However, for Latin-1 we go to extra effort
2267 to do a better job.
2268
2269 This function is not called for ASCII ranges.
2270
2271 Returns -1 if successful, REG_ESPACE if ran out of space. */
2272
2273 static int
2274 set_image_of_range (struct range_table_work_area *work_area,
2275 re_wchar_t start, re_wchar_t end,
2276 RE_TRANSLATE_TYPE translate)
2277 {
2278 re_wchar_t cmin, cmax;
2279
2280 #ifdef emacs
2281 /* For Latin-1 ranges, use set_image_of_range_1
2282 to get proper handling of ranges that include letters and nonletters.
2283 For a range that includes the whole of Latin-1, this is not necessary.
2284 For other character sets, we don't bother to get this right. */
2285 if (RE_TRANSLATE_P (translate) && start < 04400
2286 && !(start < 04200 && end >= 04377))
2287 {
2288 int newend;
2289 int tem;
2290 newend = end;
2291 if (newend > 04377)
2292 newend = 04377;
2293 tem = set_image_of_range_1 (work_area, start, newend, translate);
2294 if (tem > 0)
2295 return tem;
2296
2297 start = 04400;
2298 if (end < 04400)
2299 return -1;
2300 }
2301 #endif
2302
2303 EXTEND_RANGE_TABLE (work_area, 2);
2304 work_area->table[work_area->used++] = (start);
2305 work_area->table[work_area->used++] = (end);
2306
2307 cmin = -1, cmax = -1;
2308
2309 if (RE_TRANSLATE_P (translate))
2310 {
2311 int ch;
2312
2313 for (ch = start; ch <= end; ch++)
2314 {
2315 re_wchar_t c = TRANSLATE (ch);
2316 if (! (start <= c && c <= end))
2317 {
2318 if (cmin == -1)
2319 cmin = c, cmax = c;
2320 else
2321 {
2322 cmin = MIN (cmin, c);
2323 cmax = MAX (cmax, c);
2324 }
2325 }
2326 }
2327
2328 if (cmin != -1)
2329 {
2330 EXTEND_RANGE_TABLE (work_area, 2);
2331 work_area->table[work_area->used++] = (cmin);
2332 work_area->table[work_area->used++] = (cmax);
2333 }
2334 }
2335
2336 return -1;
2337 }
2338 #endif /* 0 */
2339 \f
2340 #ifndef MATCH_MAY_ALLOCATE
2341
2342 /* If we cannot allocate large objects within re_match_2_internal,
2343 we make the fail stack and register vectors global.
2344 The fail stack, we grow to the maximum size when a regexp
2345 is compiled.
2346 The register vectors, we adjust in size each time we
2347 compile a regexp, according to the number of registers it needs. */
2348
2349 static fail_stack_type fail_stack;
2350
2351 /* Size with which the following vectors are currently allocated.
2352 That is so we can make them bigger as needed,
2353 but never make them smaller. */
2354 static int regs_allocated_size;
2355
2356 static re_char ** regstart, ** regend;
2357 static re_char **best_regstart, **best_regend;
2358
2359 /* Make the register vectors big enough for NUM_REGS registers,
2360 but don't make them smaller. */
2361
2362 static
2363 regex_grow_registers (int num_regs)
2364 {
2365 if (num_regs > regs_allocated_size)
2366 {
2367 RETALLOC_IF (regstart, num_regs, re_char *);
2368 RETALLOC_IF (regend, num_regs, re_char *);
2369 RETALLOC_IF (best_regstart, num_regs, re_char *);
2370 RETALLOC_IF (best_regend, num_regs, re_char *);
2371
2372 regs_allocated_size = num_regs;
2373 }
2374 }
2375
2376 #endif /* not MATCH_MAY_ALLOCATE */
2377 \f
2378 static boolean group_in_compile_stack (compile_stack_type compile_stack,
2379 regnum_t regnum);
2380
2381 /* `regex_compile' compiles PATTERN (of length SIZE) according to SYNTAX.
2382 Returns one of error codes defined in `regex.h', or zero for success.
2383
2384 Assumes the `allocated' (and perhaps `buffer') and `translate'
2385 fields are set in BUFP on entry.
2386
2387 If it succeeds, results are put in BUFP (if it returns an error, the
2388 contents of BUFP are undefined):
2389 `buffer' is the compiled pattern;
2390 `syntax' is set to SYNTAX;
2391 `used' is set to the length of the compiled pattern;
2392 `fastmap_accurate' is zero;
2393 `re_nsub' is the number of subexpressions in PATTERN;
2394 `not_bol' and `not_eol' are zero;
2395
2396 The `fastmap' field is neither examined nor set. */
2397
2398 /* Insert the `jump' from the end of last alternative to "here".
2399 The space for the jump has already been allocated. */
2400 #define FIXUP_ALT_JUMP() \
2401 do { \
2402 if (fixup_alt_jump) \
2403 STORE_JUMP (jump, fixup_alt_jump, b); \
2404 } while (0)
2405
2406
2407 /* Return, freeing storage we allocated. */
2408 #define FREE_STACK_RETURN(value) \
2409 do { \
2410 FREE_RANGE_TABLE_WORK_AREA (range_table_work); \
2411 free (compile_stack.stack); \
2412 return value; \
2413 } while (0)
2414
2415 static reg_errcode_t
2416 regex_compile (const re_char *pattern, size_t size, reg_syntax_t syntax, struct re_pattern_buffer *bufp)
2417 {
2418 /* We fetch characters from PATTERN here. */
2419 register re_wchar_t c, c1;
2420
2421 /* Points to the end of the buffer, where we should append. */
2422 register unsigned char *b;
2423
2424 /* Keeps track of unclosed groups. */
2425 compile_stack_type compile_stack;
2426
2427 /* Points to the current (ending) position in the pattern. */
2428 #ifdef AIX
2429 /* `const' makes AIX compiler fail. */
2430 unsigned char *p = pattern;
2431 #else
2432 re_char *p = pattern;
2433 #endif
2434 re_char *pend = pattern + size;
2435
2436 /* How to translate the characters in the pattern. */
2437 RE_TRANSLATE_TYPE translate = bufp->translate;
2438
2439 /* Address of the count-byte of the most recently inserted `exactn'
2440 command. This makes it possible to tell if a new exact-match
2441 character can be added to that command or if the character requires
2442 a new `exactn' command. */
2443 unsigned char *pending_exact = 0;
2444
2445 /* Address of start of the most recently finished expression.
2446 This tells, e.g., postfix * where to find the start of its
2447 operand. Reset at the beginning of groups and alternatives. */
2448 unsigned char *laststart = 0;
2449
2450 /* Address of beginning of regexp, or inside of last group. */
2451 unsigned char *begalt;
2452
2453 /* Place in the uncompiled pattern (i.e., the {) to
2454 which to go back if the interval is invalid. */
2455 re_char *beg_interval;
2456
2457 /* Address of the place where a forward jump should go to the end of
2458 the containing expression. Each alternative of an `or' -- except the
2459 last -- ends with a forward jump of this sort. */
2460 unsigned char *fixup_alt_jump = 0;
2461
2462 /* Work area for range table of charset. */
2463 struct range_table_work_area range_table_work;
2464
2465 /* If the object matched can contain multibyte characters. */
2466 const boolean multibyte = RE_MULTIBYTE_P (bufp);
2467
2468 /* Nonzero if we have pushed down into a subpattern. */
2469 int in_subpattern = 0;
2470
2471 /* These hold the values of p, pattern, and pend from the main
2472 pattern when we have pushed into a subpattern. */
2473 re_char *main_p IF_LINT (= NULL);
2474 re_char *main_pattern IF_LINT (= NULL);
2475 re_char *main_pend IF_LINT (= NULL);
2476
2477 #ifdef DEBUG
2478 debug++;
2479 DEBUG_PRINT ("\nCompiling pattern: ");
2480 if (debug > 0)
2481 {
2482 unsigned debug_count;
2483
2484 for (debug_count = 0; debug_count < size; debug_count++)
2485 putchar (pattern[debug_count]);
2486 putchar ('\n');
2487 }
2488 #endif /* DEBUG */
2489
2490 /* Initialize the compile stack. */
2491 compile_stack.stack = TALLOC (INIT_COMPILE_STACK_SIZE, compile_stack_elt_t);
2492 if (compile_stack.stack == NULL)
2493 return REG_ESPACE;
2494
2495 compile_stack.size = INIT_COMPILE_STACK_SIZE;
2496 compile_stack.avail = 0;
2497
2498 range_table_work.table = 0;
2499 range_table_work.allocated = 0;
2500
2501 /* Initialize the pattern buffer. */
2502 bufp->syntax = syntax;
2503 bufp->fastmap_accurate = 0;
2504 bufp->not_bol = bufp->not_eol = 0;
2505 bufp->used_syntax = 0;
2506
2507 /* Set `used' to zero, so that if we return an error, the pattern
2508 printer (for debugging) will think there's no pattern. We reset it
2509 at the end. */
2510 bufp->used = 0;
2511
2512 /* Always count groups, whether or not bufp->no_sub is set. */
2513 bufp->re_nsub = 0;
2514
2515 #if !defined emacs && !defined SYNTAX_TABLE
2516 /* Initialize the syntax table. */
2517 init_syntax_once ();
2518 #endif
2519
2520 if (bufp->allocated == 0)
2521 {
2522 if (bufp->buffer)
2523 { /* If zero allocated, but buffer is non-null, try to realloc
2524 enough space. This loses if buffer's address is bogus, but
2525 that is the user's responsibility. */
2526 RETALLOC (bufp->buffer, INIT_BUF_SIZE, unsigned char);
2527 }
2528 else
2529 { /* Caller did not allocate a buffer. Do it for them. */
2530 bufp->buffer = TALLOC (INIT_BUF_SIZE, unsigned char);
2531 }
2532 if (!bufp->buffer) FREE_STACK_RETURN (REG_ESPACE);
2533
2534 bufp->allocated = INIT_BUF_SIZE;
2535 }
2536
2537 begalt = b = bufp->buffer;
2538
2539 /* Loop through the uncompiled pattern until we're at the end. */
2540 while (1)
2541 {
2542 if (p == pend)
2543 {
2544 /* If this is the end of an included regexp,
2545 pop back to the main regexp and try again. */
2546 if (in_subpattern)
2547 {
2548 in_subpattern = 0;
2549 pattern = main_pattern;
2550 p = main_p;
2551 pend = main_pend;
2552 continue;
2553 }
2554 /* If this is the end of the main regexp, we are done. */
2555 break;
2556 }
2557
2558 PATFETCH (c);
2559
2560 switch (c)
2561 {
2562 case ' ':
2563 {
2564 re_char *p1 = p;
2565
2566 /* If there's no special whitespace regexp, treat
2567 spaces normally. And don't try to do this recursively. */
2568 if (!whitespace_regexp || in_subpattern)
2569 goto normal_char;
2570
2571 /* Peek past following spaces. */
2572 while (p1 != pend)
2573 {
2574 if (*p1 != ' ')
2575 break;
2576 p1++;
2577 }
2578 /* If the spaces are followed by a repetition op,
2579 treat them normally. */
2580 if (p1 != pend
2581 && (*p1 == '*' || *p1 == '+' || *p1 == '?'
2582 || (*p1 == '\\' && p1 + 1 != pend && p1[1] == '{')))
2583 goto normal_char;
2584
2585 /* Replace the spaces with the whitespace regexp. */
2586 in_subpattern = 1;
2587 main_p = p1;
2588 main_pend = pend;
2589 main_pattern = pattern;
2590 p = pattern = whitespace_regexp;
2591 pend = p + strlen ((const char *) p);
2592 break;
2593 }
2594
2595 case '^':
2596 {
2597 if ( /* If at start of pattern, it's an operator. */
2598 p == pattern + 1
2599 /* If context independent, it's an operator. */
2600 || syntax & RE_CONTEXT_INDEP_ANCHORS
2601 /* Otherwise, depends on what's come before. */
2602 || at_begline_loc_p (pattern, p, syntax))
2603 BUF_PUSH ((syntax & RE_NO_NEWLINE_ANCHOR) ? begbuf : begline);
2604 else
2605 goto normal_char;
2606 }
2607 break;
2608
2609
2610 case '$':
2611 {
2612 if ( /* If at end of pattern, it's an operator. */
2613 p == pend
2614 /* If context independent, it's an operator. */
2615 || syntax & RE_CONTEXT_INDEP_ANCHORS
2616 /* Otherwise, depends on what's next. */
2617 || at_endline_loc_p (p, pend, syntax))
2618 BUF_PUSH ((syntax & RE_NO_NEWLINE_ANCHOR) ? endbuf : endline);
2619 else
2620 goto normal_char;
2621 }
2622 break;
2623
2624
2625 case '+':
2626 case '?':
2627 if ((syntax & RE_BK_PLUS_QM)
2628 || (syntax & RE_LIMITED_OPS))
2629 goto normal_char;
2630 handle_plus:
2631 case '*':
2632 /* If there is no previous pattern... */
2633 if (!laststart)
2634 {
2635 if (syntax & RE_CONTEXT_INVALID_OPS)
2636 FREE_STACK_RETURN (REG_BADRPT);
2637 else if (!(syntax & RE_CONTEXT_INDEP_OPS))
2638 goto normal_char;
2639 }
2640
2641 {
2642 /* 1 means zero (many) matches is allowed. */
2643 boolean zero_times_ok = 0, many_times_ok = 0;
2644 boolean greedy = 1;
2645
2646 /* If there is a sequence of repetition chars, collapse it
2647 down to just one (the right one). We can't combine
2648 interval operators with these because of, e.g., `a{2}*',
2649 which should only match an even number of `a's. */
2650
2651 for (;;)
2652 {
2653 if ((syntax & RE_FRUGAL)
2654 && c == '?' && (zero_times_ok || many_times_ok))
2655 greedy = 0;
2656 else
2657 {
2658 zero_times_ok |= c != '+';
2659 many_times_ok |= c != '?';
2660 }
2661
2662 if (p == pend)
2663 break;
2664 else if (*p == '*'
2665 || (!(syntax & RE_BK_PLUS_QM)
2666 && (*p == '+' || *p == '?')))
2667 ;
2668 else if (syntax & RE_BK_PLUS_QM && *p == '\\')
2669 {
2670 if (p+1 == pend)
2671 FREE_STACK_RETURN (REG_EESCAPE);
2672 if (p[1] == '+' || p[1] == '?')
2673 PATFETCH (c); /* Gobble up the backslash. */
2674 else
2675 break;
2676 }
2677 else
2678 break;
2679 /* If we get here, we found another repeat character. */
2680 PATFETCH (c);
2681 }
2682
2683 /* Star, etc. applied to an empty pattern is equivalent
2684 to an empty pattern. */
2685 if (!laststart || laststart == b)
2686 break;
2687
2688 /* Now we know whether or not zero matches is allowed
2689 and also whether or not two or more matches is allowed. */
2690 if (greedy)
2691 {
2692 if (many_times_ok)
2693 {
2694 boolean simple = skip_one_char (laststart) == b;
2695 size_t startoffset = 0;
2696 re_opcode_t ofj =
2697 /* Check if the loop can match the empty string. */
2698 (simple || !analyse_first (laststart, b, NULL, 0))
2699 ? on_failure_jump : on_failure_jump_loop;
2700 assert (skip_one_char (laststart) <= b);
2701
2702 if (!zero_times_ok && simple)
2703 { /* Since simple * loops can be made faster by using
2704 on_failure_keep_string_jump, we turn simple P+
2705 into PP* if P is simple. */
2706 unsigned char *p1, *p2;
2707 startoffset = b - laststart;
2708 GET_BUFFER_SPACE (startoffset);
2709 p1 = b; p2 = laststart;
2710 while (p2 < p1)
2711 *b++ = *p2++;
2712 zero_times_ok = 1;
2713 }
2714
2715 GET_BUFFER_SPACE (6);
2716 if (!zero_times_ok)
2717 /* A + loop. */
2718 STORE_JUMP (ofj, b, b + 6);
2719 else
2720 /* Simple * loops can use on_failure_keep_string_jump
2721 depending on what follows. But since we don't know
2722 that yet, we leave the decision up to
2723 on_failure_jump_smart. */
2724 INSERT_JUMP (simple ? on_failure_jump_smart : ofj,
2725 laststart + startoffset, b + 6);
2726 b += 3;
2727 STORE_JUMP (jump, b, laststart + startoffset);
2728 b += 3;
2729 }
2730 else
2731 {
2732 /* A simple ? pattern. */
2733 assert (zero_times_ok);
2734 GET_BUFFER_SPACE (3);
2735 INSERT_JUMP (on_failure_jump, laststart, b + 3);
2736 b += 3;
2737 }
2738 }
2739 else /* not greedy */
2740 { /* I wish the greedy and non-greedy cases could be merged. */
2741
2742 GET_BUFFER_SPACE (7); /* We might use less. */
2743 if (many_times_ok)
2744 {
2745 boolean emptyp = analyse_first (laststart, b, NULL, 0);
2746
2747 /* The non-greedy multiple match looks like
2748 a repeat..until: we only need a conditional jump
2749 at the end of the loop. */
2750 if (emptyp) BUF_PUSH (no_op);
2751 STORE_JUMP (emptyp ? on_failure_jump_nastyloop
2752 : on_failure_jump, b, laststart);
2753 b += 3;
2754 if (zero_times_ok)
2755 {
2756 /* The repeat...until naturally matches one or more.
2757 To also match zero times, we need to first jump to
2758 the end of the loop (its conditional jump). */
2759 INSERT_JUMP (jump, laststart, b);
2760 b += 3;
2761 }
2762 }
2763 else
2764 {
2765 /* non-greedy a?? */
2766 INSERT_JUMP (jump, laststart, b + 3);
2767 b += 3;
2768 INSERT_JUMP (on_failure_jump, laststart, laststart + 6);
2769 b += 3;
2770 }
2771 }
2772 }
2773 pending_exact = 0;
2774 break;
2775
2776
2777 case '.':
2778 laststart = b;
2779 BUF_PUSH (anychar);
2780 break;
2781
2782
2783 case '[':
2784 {
2785 re_char *p1;
2786
2787 CLEAR_RANGE_TABLE_WORK_USED (range_table_work);
2788
2789 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2790
2791 /* Ensure that we have enough space to push a charset: the
2792 opcode, the length count, and the bitset; 34 bytes in all. */
2793 GET_BUFFER_SPACE (34);
2794
2795 laststart = b;
2796
2797 /* We test `*p == '^' twice, instead of using an if
2798 statement, so we only need one BUF_PUSH. */
2799 BUF_PUSH (*p == '^' ? charset_not : charset);
2800 if (*p == '^')
2801 p++;
2802
2803 /* Remember the first position in the bracket expression. */
2804 p1 = p;
2805
2806 /* Push the number of bytes in the bitmap. */
2807 BUF_PUSH ((1 << BYTEWIDTH) / BYTEWIDTH);
2808
2809 /* Clear the whole map. */
2810 memset (b, 0, (1 << BYTEWIDTH) / BYTEWIDTH);
2811
2812 /* charset_not matches newline according to a syntax bit. */
2813 if ((re_opcode_t) b[-2] == charset_not
2814 && (syntax & RE_HAT_LISTS_NOT_NEWLINE))
2815 SET_LIST_BIT ('\n');
2816
2817 /* Read in characters and ranges, setting map bits. */
2818 for (;;)
2819 {
2820 boolean escaped_char = false;
2821 const unsigned char *p2 = p;
2822 re_wchar_t ch;
2823
2824 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2825
2826 /* Don't translate yet. The range TRANSLATE(X..Y) cannot
2827 always be determined from TRANSLATE(X) and TRANSLATE(Y)
2828 So the translation is done later in a loop. Example:
2829 (let ((case-fold-search t)) (string-match "[A-_]" "A")) */
2830 PATFETCH (c);
2831
2832 /* \ might escape characters inside [...] and [^...]. */
2833 if ((syntax & RE_BACKSLASH_ESCAPE_IN_LISTS) && c == '\\')
2834 {
2835 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
2836
2837 PATFETCH (c);
2838 escaped_char = true;
2839 }
2840 else
2841 {
2842 /* Could be the end of the bracket expression. If it's
2843 not (i.e., when the bracket expression is `[]' so
2844 far), the ']' character bit gets set way below. */
2845 if (c == ']' && p2 != p1)
2846 break;
2847 }
2848
2849 /* See if we're at the beginning of a possible character
2850 class. */
2851
2852 if (!escaped_char &&
2853 syntax & RE_CHAR_CLASSES && c == '[' && *p == ':')
2854 {
2855 /* Leave room for the null. */
2856 unsigned char str[CHAR_CLASS_MAX_LENGTH + 1];
2857 const unsigned char *class_beg;
2858
2859 PATFETCH (c);
2860 c1 = 0;
2861 class_beg = p;
2862
2863 /* If pattern is `[[:'. */
2864 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2865
2866 for (;;)
2867 {
2868 PATFETCH (c);
2869 if ((c == ':' && *p == ']') || p == pend)
2870 break;
2871 if (c1 < CHAR_CLASS_MAX_LENGTH)
2872 str[c1++] = c;
2873 else
2874 /* This is in any case an invalid class name. */
2875 str[0] = '\0';
2876 }
2877 str[c1] = '\0';
2878
2879 /* If isn't a word bracketed by `[:' and `:]':
2880 undo the ending character, the letters, and
2881 leave the leading `:' and `[' (but set bits for
2882 them). */
2883 if (c == ':' && *p == ']')
2884 {
2885 re_wctype_t cc = re_wctype (str);
2886
2887 if (cc == 0)
2888 FREE_STACK_RETURN (REG_ECTYPE);
2889
2890 /* Throw away the ] at the end of the character
2891 class. */
2892 PATFETCH (c);
2893
2894 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2895
2896 #ifndef emacs
2897 for (ch = 0; ch < (1 << BYTEWIDTH); ++ch)
2898 if (re_iswctype (btowc (ch), cc))
2899 {
2900 c = TRANSLATE (ch);
2901 if (c < (1 << BYTEWIDTH))
2902 SET_LIST_BIT (c);
2903 }
2904 #else /* emacs */
2905 /* Most character classes in a multibyte match
2906 just set a flag. Exceptions are is_blank,
2907 is_digit, is_cntrl, and is_xdigit, since
2908 they can only match ASCII characters. We
2909 don't need to handle them for multibyte.
2910 They are distinguished by a negative wctype. */
2911
2912 /* Setup the gl_state object to its buffer-defined
2913 value. This hardcodes the buffer-global
2914 syntax-table for ASCII chars, while the other chars
2915 will obey syntax-table properties. It's not ideal,
2916 but it's the way it's been done until now. */
2917 SETUP_BUFFER_SYNTAX_TABLE ();
2918
2919 for (ch = 0; ch < 256; ++ch)
2920 {
2921 c = RE_CHAR_TO_MULTIBYTE (ch);
2922 if (! CHAR_BYTE8_P (c)
2923 && re_iswctype (c, cc))
2924 {
2925 SET_LIST_BIT (ch);
2926 c1 = TRANSLATE (c);
2927 if (c1 == c)
2928 continue;
2929 if (ASCII_CHAR_P (c1))
2930 SET_LIST_BIT (c1);
2931 else if ((c1 = RE_CHAR_TO_UNIBYTE (c1)) >= 0)
2932 SET_LIST_BIT (c1);
2933 }
2934 }
2935 SET_RANGE_TABLE_WORK_AREA_BIT
2936 (range_table_work, re_wctype_to_bit (cc));
2937 #endif /* emacs */
2938 /* In most cases the matching rule for char classes
2939 only uses the syntax table for multibyte chars,
2940 so that the content of the syntax-table it is not
2941 hardcoded in the range_table. SPACE and WORD are
2942 the two exceptions. */
2943 if ((1 << cc) & ((1 << RECC_SPACE) | (1 << RECC_WORD)))
2944 bufp->used_syntax = 1;
2945
2946 /* Repeat the loop. */
2947 continue;
2948 }
2949 else
2950 {
2951 /* Go back to right after the "[:". */
2952 p = class_beg;
2953 SET_LIST_BIT ('[');
2954
2955 /* Because the `:' may starts the range, we
2956 can't simply set bit and repeat the loop.
2957 Instead, just set it to C and handle below. */
2958 c = ':';
2959 }
2960 }
2961
2962 if (p < pend && p[0] == '-' && p[1] != ']')
2963 {
2964
2965 /* Discard the `-'. */
2966 PATFETCH (c1);
2967
2968 /* Fetch the character which ends the range. */
2969 PATFETCH (c1);
2970 #ifdef emacs
2971 if (CHAR_BYTE8_P (c1)
2972 && ! ASCII_CHAR_P (c) && ! CHAR_BYTE8_P (c))
2973 /* Treat the range from a multibyte character to
2974 raw-byte character as empty. */
2975 c = c1 + 1;
2976 #endif /* emacs */
2977 }
2978 else
2979 /* Range from C to C. */
2980 c1 = c;
2981
2982 if (c > c1)
2983 {
2984 if (syntax & RE_NO_EMPTY_RANGES)
2985 FREE_STACK_RETURN (REG_ERANGEX);
2986 /* Else, repeat the loop. */
2987 }
2988 else
2989 {
2990 #ifndef emacs
2991 /* Set the range into bitmap */
2992 for (; c <= c1; c++)
2993 {
2994 ch = TRANSLATE (c);
2995 if (ch < (1 << BYTEWIDTH))
2996 SET_LIST_BIT (ch);
2997 }
2998 #else /* emacs */
2999 if (c < 128)
3000 {
3001 ch = MIN (127, c1);
3002 SETUP_ASCII_RANGE (range_table_work, c, ch);
3003 c = ch + 1;
3004 if (CHAR_BYTE8_P (c1))
3005 c = BYTE8_TO_CHAR (128);
3006 }
3007 if (c <= c1)
3008 {
3009 if (CHAR_BYTE8_P (c))
3010 {
3011 c = CHAR_TO_BYTE8 (c);
3012 c1 = CHAR_TO_BYTE8 (c1);
3013 for (; c <= c1; c++)
3014 SET_LIST_BIT (c);
3015 }
3016 else if (multibyte)
3017 {
3018 SETUP_MULTIBYTE_RANGE (range_table_work, c, c1);
3019 }
3020 else
3021 {
3022 SETUP_UNIBYTE_RANGE (range_table_work, c, c1);
3023 }
3024 }
3025 #endif /* emacs */
3026 }
3027 }
3028
3029 /* Discard any (non)matching list bytes that are all 0 at the
3030 end of the map. Decrease the map-length byte too. */
3031 while ((int) b[-1] > 0 && b[b[-1] - 1] == 0)
3032 b[-1]--;
3033 b += b[-1];
3034
3035 /* Build real range table from work area. */
3036 if (RANGE_TABLE_WORK_USED (range_table_work)
3037 || RANGE_TABLE_WORK_BITS (range_table_work))
3038 {
3039 int i;
3040 int used = RANGE_TABLE_WORK_USED (range_table_work);
3041
3042 /* Allocate space for COUNT + RANGE_TABLE. Needs two
3043 bytes for flags, two for COUNT, and three bytes for
3044 each character. */
3045 GET_BUFFER_SPACE (4 + used * 3);
3046
3047 /* Indicate the existence of range table. */
3048 laststart[1] |= 0x80;
3049
3050 /* Store the character class flag bits into the range table.
3051 If not in emacs, these flag bits are always 0. */
3052 *b++ = RANGE_TABLE_WORK_BITS (range_table_work) & 0xff;
3053 *b++ = RANGE_TABLE_WORK_BITS (range_table_work) >> 8;
3054
3055 STORE_NUMBER_AND_INCR (b, used / 2);
3056 for (i = 0; i < used; i++)
3057 STORE_CHARACTER_AND_INCR
3058 (b, RANGE_TABLE_WORK_ELT (range_table_work, i));
3059 }
3060 }
3061 break;
3062
3063
3064 case '(':
3065 if (syntax & RE_NO_BK_PARENS)
3066 goto handle_open;
3067 else
3068 goto normal_char;
3069
3070
3071 case ')':
3072 if (syntax & RE_NO_BK_PARENS)
3073 goto handle_close;
3074 else
3075 goto normal_char;
3076
3077
3078 case '\n':
3079 if (syntax & RE_NEWLINE_ALT)
3080 goto handle_alt;
3081 else
3082 goto normal_char;
3083
3084
3085 case '|':
3086 if (syntax & RE_NO_BK_VBAR)
3087 goto handle_alt;
3088 else
3089 goto normal_char;
3090
3091
3092 case '{':
3093 if (syntax & RE_INTERVALS && syntax & RE_NO_BK_BRACES)
3094 goto handle_interval;
3095 else
3096 goto normal_char;
3097
3098
3099 case '\\':
3100 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
3101
3102 /* Do not translate the character after the \, so that we can
3103 distinguish, e.g., \B from \b, even if we normally would
3104 translate, e.g., B to b. */
3105 PATFETCH (c);
3106
3107 switch (c)
3108 {
3109 case '(':
3110 if (syntax & RE_NO_BK_PARENS)
3111 goto normal_backslash;
3112
3113 handle_open:
3114 {
3115 int shy = 0;
3116 regnum_t regnum = 0;
3117 if (p+1 < pend)
3118 {
3119 /* Look for a special (?...) construct */
3120 if ((syntax & RE_SHY_GROUPS) && *p == '?')
3121 {
3122 PATFETCH (c); /* Gobble up the '?'. */
3123 while (!shy)
3124 {
3125 PATFETCH (c);
3126 switch (c)
3127 {
3128 case ':': shy = 1; break;
3129 case '0':
3130 /* An explicitly specified regnum must start
3131 with non-0. */
3132 if (regnum == 0)
3133 FREE_STACK_RETURN (REG_BADPAT);
3134 case '1': case '2': case '3': case '4':
3135 case '5': case '6': case '7': case '8': case '9':
3136 regnum = 10*regnum + (c - '0'); break;
3137 default:
3138 /* Only (?:...) is supported right now. */
3139 FREE_STACK_RETURN (REG_BADPAT);
3140 }
3141 }
3142 }
3143 }
3144
3145 if (!shy)
3146 regnum = ++bufp->re_nsub;
3147 else if (regnum)
3148 { /* It's actually not shy, but explicitly numbered. */
3149 shy = 0;
3150 if (regnum > bufp->re_nsub)
3151 bufp->re_nsub = regnum;
3152 else if (regnum > bufp->re_nsub
3153 /* Ideally, we'd want to check that the specified
3154 group can't have matched (i.e. all subgroups
3155 using the same regnum are in other branches of
3156 OR patterns), but we don't currently keep track
3157 of enough info to do that easily. */
3158 || group_in_compile_stack (compile_stack, regnum))
3159 FREE_STACK_RETURN (REG_BADPAT);
3160 }
3161 else
3162 /* It's really shy. */
3163 regnum = - bufp->re_nsub;
3164
3165 if (COMPILE_STACK_FULL)
3166 {
3167 RETALLOC (compile_stack.stack, compile_stack.size << 1,
3168 compile_stack_elt_t);
3169 if (compile_stack.stack == NULL) return REG_ESPACE;
3170
3171 compile_stack.size <<= 1;
3172 }
3173
3174 /* These are the values to restore when we hit end of this
3175 group. They are all relative offsets, so that if the
3176 whole pattern moves because of realloc, they will still
3177 be valid. */
3178 COMPILE_STACK_TOP.begalt_offset = begalt - bufp->buffer;
3179 COMPILE_STACK_TOP.fixup_alt_jump
3180 = fixup_alt_jump ? fixup_alt_jump - bufp->buffer + 1 : 0;
3181 COMPILE_STACK_TOP.laststart_offset = b - bufp->buffer;
3182 COMPILE_STACK_TOP.regnum = regnum;
3183
3184 /* Do not push a start_memory for groups beyond the last one
3185 we can represent in the compiled pattern. */
3186 if (regnum <= MAX_REGNUM && regnum > 0)
3187 BUF_PUSH_2 (start_memory, regnum);
3188
3189 compile_stack.avail++;
3190
3191 fixup_alt_jump = 0;
3192 laststart = 0;
3193 begalt = b;
3194 /* If we've reached MAX_REGNUM groups, then this open
3195 won't actually generate any code, so we'll have to
3196 clear pending_exact explicitly. */
3197 pending_exact = 0;
3198 break;
3199 }
3200
3201 case ')':
3202 if (syntax & RE_NO_BK_PARENS) goto normal_backslash;
3203
3204 if (COMPILE_STACK_EMPTY)
3205 {
3206 if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD)
3207 goto normal_backslash;
3208 else
3209 FREE_STACK_RETURN (REG_ERPAREN);
3210 }
3211
3212 handle_close:
3213 FIXUP_ALT_JUMP ();
3214
3215 /* See similar code for backslashed left paren above. */
3216 if (COMPILE_STACK_EMPTY)
3217 {
3218 if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD)
3219 goto normal_char;
3220 else
3221 FREE_STACK_RETURN (REG_ERPAREN);
3222 }
3223
3224 /* Since we just checked for an empty stack above, this
3225 ``can't happen''. */
3226 assert (compile_stack.avail != 0);
3227 {
3228 /* We don't just want to restore into `regnum', because
3229 later groups should continue to be numbered higher,
3230 as in `(ab)c(de)' -- the second group is #2. */
3231 regnum_t regnum;
3232
3233 compile_stack.avail--;
3234 begalt = bufp->buffer + COMPILE_STACK_TOP.begalt_offset;
3235 fixup_alt_jump
3236 = COMPILE_STACK_TOP.fixup_alt_jump
3237 ? bufp->buffer + COMPILE_STACK_TOP.fixup_alt_jump - 1
3238 : 0;
3239 laststart = bufp->buffer + COMPILE_STACK_TOP.laststart_offset;
3240 regnum = COMPILE_STACK_TOP.regnum;
3241 /* If we've reached MAX_REGNUM groups, then this open
3242 won't actually generate any code, so we'll have to
3243 clear pending_exact explicitly. */
3244 pending_exact = 0;
3245
3246 /* We're at the end of the group, so now we know how many
3247 groups were inside this one. */
3248 if (regnum <= MAX_REGNUM && regnum > 0)
3249 BUF_PUSH_2 (stop_memory, regnum);
3250 }
3251 break;
3252
3253
3254 case '|': /* `\|'. */
3255 if (syntax & RE_LIMITED_OPS || syntax & RE_NO_BK_VBAR)
3256 goto normal_backslash;
3257 handle_alt:
3258 if (syntax & RE_LIMITED_OPS)
3259 goto normal_char;
3260
3261 /* Insert before the previous alternative a jump which
3262 jumps to this alternative if the former fails. */
3263 GET_BUFFER_SPACE (3);
3264 INSERT_JUMP (on_failure_jump, begalt, b + 6);
3265 pending_exact = 0;
3266 b += 3;
3267
3268 /* The alternative before this one has a jump after it
3269 which gets executed if it gets matched. Adjust that
3270 jump so it will jump to this alternative's analogous
3271 jump (put in below, which in turn will jump to the next
3272 (if any) alternative's such jump, etc.). The last such
3273 jump jumps to the correct final destination. A picture:
3274 _____ _____
3275 | | | |
3276 | v | v
3277 a | b | c
3278
3279 If we are at `b', then fixup_alt_jump right now points to a
3280 three-byte space after `a'. We'll put in the jump, set
3281 fixup_alt_jump to right after `b', and leave behind three
3282 bytes which we'll fill in when we get to after `c'. */
3283
3284 FIXUP_ALT_JUMP ();
3285
3286 /* Mark and leave space for a jump after this alternative,
3287 to be filled in later either by next alternative or
3288 when know we're at the end of a series of alternatives. */
3289 fixup_alt_jump = b;
3290 GET_BUFFER_SPACE (3);
3291 b += 3;
3292
3293 laststart = 0;
3294 begalt = b;
3295 break;
3296
3297
3298 case '{':
3299 /* If \{ is a literal. */
3300 if (!(syntax & RE_INTERVALS)
3301 /* If we're at `\{' and it's not the open-interval
3302 operator. */
3303 || (syntax & RE_NO_BK_BRACES))
3304 goto normal_backslash;
3305
3306 handle_interval:
3307 {
3308 /* If got here, then the syntax allows intervals. */
3309
3310 /* At least (most) this many matches must be made. */
3311 int lower_bound = 0, upper_bound = -1;
3312
3313 beg_interval = p;
3314
3315 GET_UNSIGNED_NUMBER (lower_bound);
3316
3317 if (c == ',')
3318 GET_UNSIGNED_NUMBER (upper_bound);
3319 else
3320 /* Interval such as `{1}' => match exactly once. */
3321 upper_bound = lower_bound;
3322
3323 if (lower_bound < 0 || upper_bound > RE_DUP_MAX
3324 || (upper_bound >= 0 && lower_bound > upper_bound))
3325 FREE_STACK_RETURN (REG_BADBR);
3326
3327 if (!(syntax & RE_NO_BK_BRACES))
3328 {
3329 if (c != '\\')
3330 FREE_STACK_RETURN (REG_BADBR);
3331 if (p == pend)
3332 FREE_STACK_RETURN (REG_EESCAPE);
3333 PATFETCH (c);
3334 }
3335
3336 if (c != '}')
3337 FREE_STACK_RETURN (REG_BADBR);
3338
3339 /* We just parsed a valid interval. */
3340
3341 /* If it's invalid to have no preceding re. */
3342 if (!laststart)
3343 {
3344 if (syntax & RE_CONTEXT_INVALID_OPS)
3345 FREE_STACK_RETURN (REG_BADRPT);
3346 else if (syntax & RE_CONTEXT_INDEP_OPS)
3347 laststart = b;
3348 else
3349 goto unfetch_interval;
3350 }
3351
3352 if (upper_bound == 0)
3353 /* If the upper bound is zero, just drop the sub pattern
3354 altogether. */
3355 b = laststart;
3356 else if (lower_bound == 1 && upper_bound == 1)
3357 /* Just match it once: nothing to do here. */
3358 ;
3359
3360 /* Otherwise, we have a nontrivial interval. When
3361 we're all done, the pattern will look like:
3362 set_number_at <jump count> <upper bound>
3363 set_number_at <succeed_n count> <lower bound>
3364 succeed_n <after jump addr> <succeed_n count>
3365 <body of loop>
3366 jump_n <succeed_n addr> <jump count>
3367 (The upper bound and `jump_n' are omitted if
3368 `upper_bound' is 1, though.) */
3369 else
3370 { /* If the upper bound is > 1, we need to insert
3371 more at the end of the loop. */
3372 unsigned int nbytes = (upper_bound < 0 ? 3
3373 : upper_bound > 1 ? 5 : 0);
3374 unsigned int startoffset = 0;
3375
3376 GET_BUFFER_SPACE (20); /* We might use less. */
3377
3378 if (lower_bound == 0)
3379 {
3380 /* A succeed_n that starts with 0 is really a
3381 a simple on_failure_jump_loop. */
3382 INSERT_JUMP (on_failure_jump_loop, laststart,
3383 b + 3 + nbytes);
3384 b += 3;
3385 }
3386 else
3387 {
3388 /* Initialize lower bound of the `succeed_n', even
3389 though it will be set during matching by its
3390 attendant `set_number_at' (inserted next),
3391 because `re_compile_fastmap' needs to know.
3392 Jump to the `jump_n' we might insert below. */
3393 INSERT_JUMP2 (succeed_n, laststart,
3394 b + 5 + nbytes,
3395 lower_bound);
3396 b += 5;
3397
3398 /* Code to initialize the lower bound. Insert
3399 before the `succeed_n'. The `5' is the last two
3400 bytes of this `set_number_at', plus 3 bytes of
3401 the following `succeed_n'. */
3402 insert_op2 (set_number_at, laststart, 5, lower_bound, b);
3403 b += 5;
3404 startoffset += 5;
3405 }
3406
3407 if (upper_bound < 0)
3408 {
3409 /* A negative upper bound stands for infinity,
3410 in which case it degenerates to a plain jump. */
3411 STORE_JUMP (jump, b, laststart + startoffset);
3412 b += 3;
3413 }
3414 else if (upper_bound > 1)
3415 { /* More than one repetition is allowed, so
3416 append a backward jump to the `succeed_n'
3417 that starts this interval.
3418
3419 When we've reached this during matching,
3420 we'll have matched the interval once, so
3421 jump back only `upper_bound - 1' times. */
3422 STORE_JUMP2 (jump_n, b, laststart + startoffset,
3423 upper_bound - 1);
3424 b += 5;
3425
3426 /* The location we want to set is the second
3427 parameter of the `jump_n'; that is `b-2' as
3428 an absolute address. `laststart' will be
3429 the `set_number_at' we're about to insert;
3430 `laststart+3' the number to set, the source
3431 for the relative address. But we are
3432 inserting into the middle of the pattern --
3433 so everything is getting moved up by 5.
3434 Conclusion: (b - 2) - (laststart + 3) + 5,
3435 i.e., b - laststart.
3436
3437 We insert this at the beginning of the loop
3438 so that if we fail during matching, we'll
3439 reinitialize the bounds. */
3440 insert_op2 (set_number_at, laststart, b - laststart,
3441 upper_bound - 1, b);
3442 b += 5;
3443 }
3444 }
3445 pending_exact = 0;
3446 beg_interval = NULL;
3447 }
3448 break;
3449
3450 unfetch_interval:
3451 /* If an invalid interval, match the characters as literals. */
3452 assert (beg_interval);
3453 p = beg_interval;
3454 beg_interval = NULL;
3455
3456 /* normal_char and normal_backslash need `c'. */
3457 c = '{';
3458
3459 if (!(syntax & RE_NO_BK_BRACES))
3460 {
3461 assert (p > pattern && p[-1] == '\\');
3462 goto normal_backslash;
3463 }
3464 else
3465 goto normal_char;
3466
3467 #ifdef emacs
3468 /* There is no way to specify the before_dot and after_dot
3469 operators. rms says this is ok. --karl */
3470 case '=':
3471 laststart = b;
3472 BUF_PUSH (at_dot);
3473 break;
3474
3475 case 's':
3476 laststart = b;
3477 PATFETCH (c);
3478 BUF_PUSH_2 (syntaxspec, syntax_spec_code[c]);
3479 break;
3480
3481 case 'S':
3482 laststart = b;
3483 PATFETCH (c);
3484 BUF_PUSH_2 (notsyntaxspec, syntax_spec_code[c]);
3485 break;
3486
3487 case 'c':
3488 laststart = b;
3489 PATFETCH (c);
3490 BUF_PUSH_2 (categoryspec, c);
3491 break;
3492
3493 case 'C':
3494 laststart = b;
3495 PATFETCH (c);
3496 BUF_PUSH_2 (notcategoryspec, c);
3497 break;
3498 #endif /* emacs */
3499
3500
3501 case 'w':
3502 if (syntax & RE_NO_GNU_OPS)
3503 goto normal_char;
3504 laststart = b;
3505 BUF_PUSH_2 (syntaxspec, Sword);
3506 break;
3507
3508
3509 case 'W':
3510 if (syntax & RE_NO_GNU_OPS)
3511 goto normal_char;
3512 laststart = b;
3513 BUF_PUSH_2 (notsyntaxspec, Sword);
3514 break;
3515
3516
3517 case '<':
3518 if (syntax & RE_NO_GNU_OPS)
3519 goto normal_char;
3520 laststart = b;
3521 BUF_PUSH (wordbeg);
3522 break;
3523
3524 case '>':
3525 if (syntax & RE_NO_GNU_OPS)
3526 goto normal_char;
3527 laststart = b;
3528 BUF_PUSH (wordend);
3529 break;
3530
3531 case '_':
3532 if (syntax & RE_NO_GNU_OPS)
3533 goto normal_char;
3534 laststart = b;
3535 PATFETCH (c);
3536 if (c == '<')
3537 BUF_PUSH (symbeg);
3538 else if (c == '>')
3539 BUF_PUSH (symend);
3540 else
3541 FREE_STACK_RETURN (REG_BADPAT);
3542 break;
3543
3544 case 'b':
3545 if (syntax & RE_NO_GNU_OPS)
3546 goto normal_char;
3547 BUF_PUSH (wordbound);
3548 break;
3549
3550 case 'B':
3551 if (syntax & RE_NO_GNU_OPS)
3552 goto normal_char;
3553 BUF_PUSH (notwordbound);
3554 break;
3555
3556 case '`':
3557 if (syntax & RE_NO_GNU_OPS)
3558 goto normal_char;
3559 BUF_PUSH (begbuf);
3560 break;
3561
3562 case '\'':
3563 if (syntax & RE_NO_GNU_OPS)
3564 goto normal_char;
3565 BUF_PUSH (endbuf);
3566 break;
3567
3568 case '1': case '2': case '3': case '4': case '5':
3569 case '6': case '7': case '8': case '9':
3570 {
3571 regnum_t reg;
3572
3573 if (syntax & RE_NO_BK_REFS)
3574 goto normal_backslash;
3575
3576 reg = c - '0';
3577
3578 if (reg > bufp->re_nsub || reg < 1
3579 /* Can't back reference to a subexp before its end. */
3580 || group_in_compile_stack (compile_stack, reg))
3581 FREE_STACK_RETURN (REG_ESUBREG);
3582
3583 laststart = b;
3584 BUF_PUSH_2 (duplicate, reg);
3585 }
3586 break;
3587
3588
3589 case '+':
3590 case '?':
3591 if (syntax & RE_BK_PLUS_QM)
3592 goto handle_plus;
3593 else
3594 goto normal_backslash;
3595
3596 default:
3597 normal_backslash:
3598 /* You might think it would be useful for \ to mean
3599 not to translate; but if we don't translate it
3600 it will never match anything. */
3601 goto normal_char;
3602 }
3603 break;
3604
3605
3606 default:
3607 /* Expects the character in `c'. */
3608 normal_char:
3609 /* If no exactn currently being built. */
3610 if (!pending_exact
3611
3612 /* If last exactn not at current position. */
3613 || pending_exact + *pending_exact + 1 != b
3614
3615 /* We have only one byte following the exactn for the count. */
3616 || *pending_exact >= (1 << BYTEWIDTH) - MAX_MULTIBYTE_LENGTH
3617
3618 /* If followed by a repetition operator. */
3619 || (p != pend && (*p == '*' || *p == '^'))
3620 || ((syntax & RE_BK_PLUS_QM)
3621 ? p + 1 < pend && *p == '\\' && (p[1] == '+' || p[1] == '?')
3622 : p != pend && (*p == '+' || *p == '?'))
3623 || ((syntax & RE_INTERVALS)
3624 && ((syntax & RE_NO_BK_BRACES)
3625 ? p != pend && *p == '{'
3626 : p + 1 < pend && p[0] == '\\' && p[1] == '{')))
3627 {
3628 /* Start building a new exactn. */
3629
3630 laststart = b;
3631
3632 BUF_PUSH_2 (exactn, 0);
3633 pending_exact = b - 1;
3634 }
3635
3636 GET_BUFFER_SPACE (MAX_MULTIBYTE_LENGTH);
3637 {
3638 int len;
3639
3640 if (multibyte)
3641 {
3642 c = TRANSLATE (c);
3643 len = CHAR_STRING (c, b);
3644 b += len;
3645 }
3646 else
3647 {
3648 c1 = RE_CHAR_TO_MULTIBYTE (c);
3649 if (! CHAR_BYTE8_P (c1))
3650 {
3651 re_wchar_t c2 = TRANSLATE (c1);
3652
3653 if (c1 != c2 && (c1 = RE_CHAR_TO_UNIBYTE (c2)) >= 0)
3654 c = c1;
3655 }
3656 *b++ = c;
3657 len = 1;
3658 }
3659 (*pending_exact) += len;
3660 }
3661
3662 break;
3663 } /* switch (c) */
3664 } /* while p != pend */
3665
3666
3667 /* Through the pattern now. */
3668
3669 FIXUP_ALT_JUMP ();
3670
3671 if (!COMPILE_STACK_EMPTY)
3672 FREE_STACK_RETURN (REG_EPAREN);
3673
3674 /* If we don't want backtracking, force success
3675 the first time we reach the end of the compiled pattern. */
3676 if (syntax & RE_NO_POSIX_BACKTRACKING)
3677 BUF_PUSH (succeed);
3678
3679 /* We have succeeded; set the length of the buffer. */
3680 bufp->used = b - bufp->buffer;
3681
3682 #ifdef DEBUG
3683 if (debug > 0)
3684 {
3685 re_compile_fastmap (bufp);
3686 DEBUG_PRINT ("\nCompiled pattern: \n");
3687 print_compiled_pattern (bufp);
3688 }
3689 debug--;
3690 #endif /* DEBUG */
3691
3692 #ifndef MATCH_MAY_ALLOCATE
3693 /* Initialize the failure stack to the largest possible stack. This
3694 isn't necessary unless we're trying to avoid calling alloca in
3695 the search and match routines. */
3696 {
3697 int num_regs = bufp->re_nsub + 1;
3698
3699 if (fail_stack.size < re_max_failures * TYPICAL_FAILURE_SIZE)
3700 {
3701 fail_stack.size = re_max_failures * TYPICAL_FAILURE_SIZE;
3702 falk_stack.stack = realloc (fail_stack.stack,
3703 fail_stack.size * sizeof *falk_stack.stack);
3704 }
3705
3706 regex_grow_registers (num_regs);
3707 }
3708 #endif /* not MATCH_MAY_ALLOCATE */
3709
3710 FREE_STACK_RETURN (REG_NOERROR);
3711 } /* regex_compile */
3712 \f
3713 /* Subroutines for `regex_compile'. */
3714
3715 /* Store OP at LOC followed by two-byte integer parameter ARG. */
3716
3717 static void
3718 store_op1 (re_opcode_t op, unsigned char *loc, int arg)
3719 {
3720 *loc = (unsigned char) op;
3721 STORE_NUMBER (loc + 1, arg);
3722 }
3723
3724
3725 /* Like `store_op1', but for two two-byte parameters ARG1 and ARG2. */
3726
3727 static void
3728 store_op2 (re_opcode_t op, unsigned char *loc, int arg1, int arg2)
3729 {
3730 *loc = (unsigned char) op;
3731 STORE_NUMBER (loc + 1, arg1);
3732 STORE_NUMBER (loc + 3, arg2);
3733 }
3734
3735
3736 /* Copy the bytes from LOC to END to open up three bytes of space at LOC
3737 for OP followed by two-byte integer parameter ARG. */
3738
3739 static void
3740 insert_op1 (re_opcode_t op, unsigned char *loc, int arg, unsigned char *end)
3741 {
3742 register unsigned char *pfrom = end;
3743 register unsigned char *pto = end + 3;
3744
3745 while (pfrom != loc)
3746 *--pto = *--pfrom;
3747
3748 store_op1 (op, loc, arg);
3749 }
3750
3751
3752 /* Like `insert_op1', but for two two-byte parameters ARG1 and ARG2. */
3753
3754 static void
3755 insert_op2 (re_opcode_t op, unsigned char *loc, int arg1, int arg2, unsigned char *end)
3756 {
3757 register unsigned char *pfrom = end;
3758 register unsigned char *pto = end + 5;
3759
3760 while (pfrom != loc)
3761 *--pto = *--pfrom;
3762
3763 store_op2 (op, loc, arg1, arg2);
3764 }
3765
3766
3767 /* P points to just after a ^ in PATTERN. Return true if that ^ comes
3768 after an alternative or a begin-subexpression. We assume there is at
3769 least one character before the ^. */
3770
3771 static boolean
3772 at_begline_loc_p (const re_char *pattern, const re_char *p, reg_syntax_t syntax)
3773 {
3774 re_char *prev = p - 2;
3775 boolean odd_backslashes;
3776
3777 /* After a subexpression? */
3778 if (*prev == '(')
3779 odd_backslashes = (syntax & RE_NO_BK_PARENS) == 0;
3780
3781 /* After an alternative? */
3782 else if (*prev == '|')
3783 odd_backslashes = (syntax & RE_NO_BK_VBAR) == 0;
3784
3785 /* After a shy subexpression? */
3786 else if (*prev == ':' && (syntax & RE_SHY_GROUPS))
3787 {
3788 /* Skip over optional regnum. */
3789 while (prev - 1 >= pattern && prev[-1] >= '0' && prev[-1] <= '9')
3790 --prev;
3791
3792 if (!(prev - 2 >= pattern
3793 && prev[-1] == '?' && prev[-2] == '('))
3794 return false;
3795 prev -= 2;
3796 odd_backslashes = (syntax & RE_NO_BK_PARENS) == 0;
3797 }
3798 else
3799 return false;
3800
3801 /* Count the number of preceding backslashes. */
3802 p = prev;
3803 while (prev - 1 >= pattern && prev[-1] == '\\')
3804 --prev;
3805 return (p - prev) & odd_backslashes;
3806 }
3807
3808
3809 /* The dual of at_begline_loc_p. This one is for $. We assume there is
3810 at least one character after the $, i.e., `P < PEND'. */
3811
3812 static boolean
3813 at_endline_loc_p (const re_char *p, const re_char *pend, reg_syntax_t syntax)
3814 {
3815 re_char *next = p;
3816 boolean next_backslash = *next == '\\';
3817 re_char *next_next = p + 1 < pend ? p + 1 : 0;
3818
3819 return
3820 /* Before a subexpression? */
3821 (syntax & RE_NO_BK_PARENS ? *next == ')'
3822 : next_backslash && next_next && *next_next == ')')
3823 /* Before an alternative? */
3824 || (syntax & RE_NO_BK_VBAR ? *next == '|'
3825 : next_backslash && next_next && *next_next == '|');
3826 }
3827
3828
3829 /* Returns true if REGNUM is in one of COMPILE_STACK's elements and
3830 false if it's not. */
3831
3832 static boolean
3833 group_in_compile_stack (compile_stack_type compile_stack, regnum_t regnum)
3834 {
3835 ssize_t this_element;
3836
3837 for (this_element = compile_stack.avail - 1;
3838 this_element >= 0;
3839 this_element--)
3840 if (compile_stack.stack[this_element].regnum == regnum)
3841 return true;
3842
3843 return false;
3844 }
3845 \f
3846 /* analyse_first.
3847 If fastmap is non-NULL, go through the pattern and fill fastmap
3848 with all the possible leading chars. If fastmap is NULL, don't
3849 bother filling it up (obviously) and only return whether the
3850 pattern could potentially match the empty string.
3851
3852 Return 1 if p..pend might match the empty string.
3853 Return 0 if p..pend matches at least one char.
3854 Return -1 if fastmap was not updated accurately. */
3855
3856 static int
3857 analyse_first (const re_char *p, const re_char *pend, char *fastmap, const int multibyte)
3858 {
3859 int j, k;
3860 boolean not;
3861
3862 /* If all elements for base leading-codes in fastmap is set, this
3863 flag is set true. */
3864 boolean match_any_multibyte_characters = false;
3865
3866 assert (p);
3867
3868 /* The loop below works as follows:
3869 - It has a working-list kept in the PATTERN_STACK and which basically
3870 starts by only containing a pointer to the first operation.
3871 - If the opcode we're looking at is a match against some set of
3872 chars, then we add those chars to the fastmap and go on to the
3873 next work element from the worklist (done via `break').
3874 - If the opcode is a control operator on the other hand, we either
3875 ignore it (if it's meaningless at this point, such as `start_memory')
3876 or execute it (if it's a jump). If the jump has several destinations
3877 (i.e. `on_failure_jump'), then we push the other destination onto the
3878 worklist.
3879 We guarantee termination by ignoring backward jumps (more or less),
3880 so that `p' is monotonically increasing. More to the point, we
3881 never set `p' (or push) anything `<= p1'. */
3882
3883 while (p < pend)
3884 {
3885 /* `p1' is used as a marker of how far back a `on_failure_jump'
3886 can go without being ignored. It is normally equal to `p'
3887 (which prevents any backward `on_failure_jump') except right
3888 after a plain `jump', to allow patterns such as:
3889 0: jump 10
3890 3..9: <body>
3891 10: on_failure_jump 3
3892 as used for the *? operator. */
3893 re_char *p1 = p;
3894
3895 switch (*p++)
3896 {
3897 case succeed:
3898 return 1;
3899
3900 case duplicate:
3901 /* If the first character has to match a backreference, that means
3902 that the group was empty (since it already matched). Since this
3903 is the only case that interests us here, we can assume that the
3904 backreference must match the empty string. */
3905 p++;
3906 continue;
3907
3908
3909 /* Following are the cases which match a character. These end
3910 with `break'. */
3911
3912 case exactn:
3913 if (fastmap)
3914 {
3915 /* If multibyte is nonzero, the first byte of each
3916 character is an ASCII or a leading code. Otherwise,
3917 each byte is a character. Thus, this works in both
3918 cases. */
3919 fastmap[p[1]] = 1;
3920 if (! multibyte)
3921 {
3922 /* For the case of matching this unibyte regex
3923 against multibyte, we must set a leading code of
3924 the corresponding multibyte character. */
3925 int c = RE_CHAR_TO_MULTIBYTE (p[1]);
3926
3927 fastmap[CHAR_LEADING_CODE (c)] = 1;
3928 }
3929 }
3930 break;
3931
3932
3933 case anychar:
3934 /* We could put all the chars except for \n (and maybe \0)
3935 but we don't bother since it is generally not worth it. */
3936 if (!fastmap) break;
3937 return -1;
3938
3939
3940 case charset_not:
3941 if (!fastmap) break;
3942 {
3943 /* Chars beyond end of bitmap are possible matches. */
3944 for (j = CHARSET_BITMAP_SIZE (&p[-1]) * BYTEWIDTH;
3945 j < (1 << BYTEWIDTH); j++)
3946 fastmap[j] = 1;
3947 }
3948
3949 /* Fallthrough */
3950 case charset:
3951 if (!fastmap) break;
3952 not = (re_opcode_t) *(p - 1) == charset_not;
3953 for (j = CHARSET_BITMAP_SIZE (&p[-1]) * BYTEWIDTH - 1, p++;
3954 j >= 0; j--)
3955 if (!!(p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH))) ^ not)
3956 fastmap[j] = 1;
3957
3958 #ifdef emacs
3959 if (/* Any leading code can possibly start a character
3960 which doesn't match the specified set of characters. */
3961 not
3962 ||
3963 /* If we can match a character class, we can match any
3964 multibyte characters. */
3965 (CHARSET_RANGE_TABLE_EXISTS_P (&p[-2])
3966 && CHARSET_RANGE_TABLE_BITS (&p[-2]) != 0))
3967
3968 {
3969 if (match_any_multibyte_characters == false)
3970 {
3971 for (j = MIN_MULTIBYTE_LEADING_CODE;
3972 j <= MAX_MULTIBYTE_LEADING_CODE; j++)
3973 fastmap[j] = 1;
3974 match_any_multibyte_characters = true;
3975 }
3976 }
3977
3978 else if (!not && CHARSET_RANGE_TABLE_EXISTS_P (&p[-2])
3979 && match_any_multibyte_characters == false)
3980 {
3981 /* Set fastmap[I] to 1 where I is a leading code of each
3982 multibyte character in the range table. */
3983 int c, count;
3984 unsigned char lc1, lc2;
3985
3986 /* Make P points the range table. `+ 2' is to skip flag
3987 bits for a character class. */
3988 p += CHARSET_BITMAP_SIZE (&p[-2]) + 2;
3989
3990 /* Extract the number of ranges in range table into COUNT. */
3991 EXTRACT_NUMBER_AND_INCR (count, p);
3992 for (; count > 0; count--, p += 3)
3993 {
3994 /* Extract the start and end of each range. */
3995 EXTRACT_CHARACTER (c, p);
3996 lc1 = CHAR_LEADING_CODE (c);
3997 p += 3;
3998 EXTRACT_CHARACTER (c, p);
3999 lc2 = CHAR_LEADING_CODE (c);
4000 for (j = lc1; j <= lc2; j++)
4001 fastmap[j] = 1;
4002 }
4003 }
4004 #endif
4005 break;
4006
4007 case syntaxspec:
4008 case notsyntaxspec:
4009 if (!fastmap) break;
4010 #ifndef emacs
4011 not = (re_opcode_t)p[-1] == notsyntaxspec;
4012 k = *p++;
4013 for (j = 0; j < (1 << BYTEWIDTH); j++)
4014 if ((SYNTAX (j) == (enum syntaxcode) k) ^ not)
4015 fastmap[j] = 1;
4016 break;
4017 #else /* emacs */
4018 /* This match depends on text properties. These end with
4019 aborting optimizations. */
4020 return -1;
4021
4022 case categoryspec:
4023 case notcategoryspec:
4024 if (!fastmap) break;
4025 not = (re_opcode_t)p[-1] == notcategoryspec;
4026 k = *p++;
4027 for (j = (1 << BYTEWIDTH); j >= 0; j--)
4028 if ((CHAR_HAS_CATEGORY (j, k)) ^ not)
4029 fastmap[j] = 1;
4030
4031 /* Any leading code can possibly start a character which
4032 has or doesn't has the specified category. */
4033 if (match_any_multibyte_characters == false)
4034 {
4035 for (j = MIN_MULTIBYTE_LEADING_CODE;
4036 j <= MAX_MULTIBYTE_LEADING_CODE; j++)
4037 fastmap[j] = 1;
4038 match_any_multibyte_characters = true;
4039 }
4040 break;
4041
4042 /* All cases after this match the empty string. These end with
4043 `continue'. */
4044
4045 case before_dot:
4046 case at_dot:
4047 case after_dot:
4048 #endif /* !emacs */
4049 case no_op:
4050 case begline:
4051 case endline:
4052 case begbuf:
4053 case endbuf:
4054 case wordbound:
4055 case notwordbound:
4056 case wordbeg:
4057 case wordend:
4058 case symbeg:
4059 case symend:
4060 continue;
4061
4062
4063 case jump:
4064 EXTRACT_NUMBER_AND_INCR (j, p);
4065 if (j < 0)
4066 /* Backward jumps can only go back to code that we've already
4067 visited. `re_compile' should make sure this is true. */
4068 break;
4069 p += j;
4070 switch (*p)
4071 {
4072 case on_failure_jump:
4073 case on_failure_keep_string_jump:
4074 case on_failure_jump_loop:
4075 case on_failure_jump_nastyloop:
4076 case on_failure_jump_smart:
4077 p++;
4078 break;
4079 default:
4080 continue;
4081 };
4082 /* Keep `p1' to allow the `on_failure_jump' we are jumping to
4083 to jump back to "just after here". */
4084 /* Fallthrough */
4085
4086 case on_failure_jump:
4087 case on_failure_keep_string_jump:
4088 case on_failure_jump_nastyloop:
4089 case on_failure_jump_loop:
4090 case on_failure_jump_smart:
4091 EXTRACT_NUMBER_AND_INCR (j, p);
4092 if (p + j <= p1)
4093 ; /* Backward jump to be ignored. */
4094 else
4095 { /* We have to look down both arms.
4096 We first go down the "straight" path so as to minimize
4097 stack usage when going through alternatives. */
4098 int r = analyse_first (p, pend, fastmap, multibyte);
4099 if (r) return r;
4100 p += j;
4101 }
4102 continue;
4103
4104
4105 case jump_n:
4106 /* This code simply does not properly handle forward jump_n. */
4107 DEBUG_STATEMENT (EXTRACT_NUMBER (j, p); assert (j < 0));
4108 p += 4;
4109 /* jump_n can either jump or fall through. The (backward) jump
4110 case has already been handled, so we only need to look at the
4111 fallthrough case. */
4112 continue;
4113
4114 case succeed_n:
4115 /* If N == 0, it should be an on_failure_jump_loop instead. */
4116 DEBUG_STATEMENT (EXTRACT_NUMBER (j, p + 2); assert (j > 0));
4117 p += 4;
4118 /* We only care about one iteration of the loop, so we don't
4119 need to consider the case where this behaves like an
4120 on_failure_jump. */
4121 continue;
4122
4123
4124 case set_number_at:
4125 p += 4;
4126 continue;
4127
4128
4129 case start_memory:
4130 case stop_memory:
4131 p += 1;
4132 continue;
4133
4134
4135 default:
4136 abort (); /* We have listed all the cases. */
4137 } /* switch *p++ */
4138
4139 /* Getting here means we have found the possible starting
4140 characters for one path of the pattern -- and that the empty
4141 string does not match. We need not follow this path further. */
4142 return 0;
4143 } /* while p */
4144
4145 /* We reached the end without matching anything. */
4146 return 1;
4147
4148 } /* analyse_first */
4149 \f
4150 /* re_compile_fastmap computes a ``fastmap'' for the compiled pattern in
4151 BUFP. A fastmap records which of the (1 << BYTEWIDTH) possible
4152 characters can start a string that matches the pattern. This fastmap
4153 is used by re_search to skip quickly over impossible starting points.
4154
4155 Character codes above (1 << BYTEWIDTH) are not represented in the
4156 fastmap, but the leading codes are represented. Thus, the fastmap
4157 indicates which character sets could start a match.
4158
4159 The caller must supply the address of a (1 << BYTEWIDTH)-byte data
4160 area as BUFP->fastmap.
4161
4162 We set the `fastmap', `fastmap_accurate', and `can_be_null' fields in
4163 the pattern buffer.
4164
4165 Returns 0 if we succeed, -2 if an internal error. */
4166
4167 int
4168 re_compile_fastmap (struct re_pattern_buffer *bufp)
4169 {
4170 char *fastmap = bufp->fastmap;
4171 int analysis;
4172
4173 assert (fastmap && bufp->buffer);
4174
4175 memset (fastmap, 0, 1 << BYTEWIDTH); /* Assume nothing's valid. */
4176 bufp->fastmap_accurate = 1; /* It will be when we're done. */
4177
4178 analysis = analyse_first (bufp->buffer, bufp->buffer + bufp->used,
4179 fastmap, RE_MULTIBYTE_P (bufp));
4180 bufp->can_be_null = (analysis != 0);
4181 return 0;
4182 } /* re_compile_fastmap */
4183 \f
4184 /* Set REGS to hold NUM_REGS registers, storing them in STARTS and
4185 ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use
4186 this memory for recording register information. STARTS and ENDS
4187 must be allocated using the malloc library routine, and must each
4188 be at least NUM_REGS * sizeof (regoff_t) bytes long.
4189
4190 If NUM_REGS == 0, then subsequent matches should allocate their own
4191 register data.
4192
4193 Unless this function is called, the first search or match using
4194 PATTERN_BUFFER will allocate its own register data, without
4195 freeing the old data. */
4196
4197 void
4198 re_set_registers (struct re_pattern_buffer *bufp, struct re_registers *regs, unsigned int num_regs, regoff_t *starts, regoff_t *ends)
4199 {
4200 if (num_regs)
4201 {
4202 bufp->regs_allocated = REGS_REALLOCATE;
4203 regs->num_regs = num_regs;
4204 regs->start = starts;
4205 regs->end = ends;
4206 }
4207 else
4208 {
4209 bufp->regs_allocated = REGS_UNALLOCATED;
4210 regs->num_regs = 0;
4211 regs->start = regs->end = (regoff_t *) 0;
4212 }
4213 }
4214 WEAK_ALIAS (__re_set_registers, re_set_registers)
4215 \f
4216 /* Searching routines. */
4217
4218 /* Like re_search_2, below, but only one string is specified, and
4219 doesn't let you say where to stop matching. */
4220
4221 regoff_t
4222 re_search (struct re_pattern_buffer *bufp, const char *string, size_t size,
4223 ssize_t startpos, ssize_t range, struct re_registers *regs)
4224 {
4225 return re_search_2 (bufp, NULL, 0, string, size, startpos, range,
4226 regs, size);
4227 }
4228 WEAK_ALIAS (__re_search, re_search)
4229
4230 /* Head address of virtual concatenation of string. */
4231 #define HEAD_ADDR_VSTRING(P) \
4232 (((P) >= size1 ? string2 : string1))
4233
4234 /* Address of POS in the concatenation of virtual string. */
4235 #define POS_ADDR_VSTRING(POS) \
4236 (((POS) >= size1 ? string2 - size1 : string1) + (POS))
4237
4238 /* Using the compiled pattern in BUFP->buffer, first tries to match the
4239 virtual concatenation of STRING1 and STRING2, starting first at index
4240 STARTPOS, then at STARTPOS + 1, and so on.
4241
4242 STRING1 and STRING2 have length SIZE1 and SIZE2, respectively.
4243
4244 RANGE is how far to scan while trying to match. RANGE = 0 means try
4245 only at STARTPOS; in general, the last start tried is STARTPOS +
4246 RANGE.
4247
4248 In REGS, return the indices of the virtual concatenation of STRING1
4249 and STRING2 that matched the entire BUFP->buffer and its contained
4250 subexpressions.
4251
4252 Do not consider matching one past the index STOP in the virtual
4253 concatenation of STRING1 and STRING2.
4254
4255 We return either the position in the strings at which the match was
4256 found, -1 if no match, or -2 if error (such as failure
4257 stack overflow). */
4258
4259 regoff_t
4260 re_search_2 (struct re_pattern_buffer *bufp, const char *str1, size_t size1,
4261 const char *str2, size_t size2, ssize_t startpos, ssize_t range,
4262 struct re_registers *regs, ssize_t stop)
4263 {
4264 regoff_t val;
4265 re_char *string1 = (re_char*) str1;
4266 re_char *string2 = (re_char*) str2;
4267 register char *fastmap = bufp->fastmap;
4268 register RE_TRANSLATE_TYPE translate = bufp->translate;
4269 size_t total_size = size1 + size2;
4270 ssize_t endpos = startpos + range;
4271 boolean anchored_start;
4272 /* Nonzero if we are searching multibyte string. */
4273 const boolean multibyte = RE_TARGET_MULTIBYTE_P (bufp);
4274
4275 /* Check for out-of-range STARTPOS. */
4276 if (startpos < 0 || startpos > total_size)
4277 return -1;
4278
4279 /* Fix up RANGE if it might eventually take us outside
4280 the virtual concatenation of STRING1 and STRING2.
4281 Make sure we won't move STARTPOS below 0 or above TOTAL_SIZE. */
4282 if (endpos < 0)
4283 range = 0 - startpos;
4284 else if (endpos > total_size)
4285 range = total_size - startpos;
4286
4287 /* If the search isn't to be a backwards one, don't waste time in a
4288 search for a pattern anchored at beginning of buffer. */
4289 if (bufp->used > 0 && (re_opcode_t) bufp->buffer[0] == begbuf && range > 0)
4290 {
4291 if (startpos > 0)
4292 return -1;
4293 else
4294 range = 0;
4295 }
4296
4297 #ifdef emacs
4298 /* In a forward search for something that starts with \=.
4299 don't keep searching past point. */
4300 if (bufp->used > 0 && (re_opcode_t) bufp->buffer[0] == at_dot && range > 0)
4301 {
4302 range = PT_BYTE - BEGV_BYTE - startpos;
4303 if (range < 0)
4304 return -1;
4305 }
4306 #endif /* emacs */
4307
4308 /* Update the fastmap now if not correct already. */
4309 if (fastmap && !bufp->fastmap_accurate)
4310 re_compile_fastmap (bufp);
4311
4312 /* See whether the pattern is anchored. */
4313 anchored_start = (bufp->buffer[0] == begline);
4314
4315 #ifdef emacs
4316 gl_state.object = re_match_object; /* Used by SYNTAX_TABLE_BYTE_TO_CHAR. */
4317 {
4318 ssize_t charpos = SYNTAX_TABLE_BYTE_TO_CHAR (POS_AS_IN_BUFFER (startpos));
4319
4320 SETUP_SYNTAX_TABLE_FOR_OBJECT (re_match_object, charpos, 1);
4321 }
4322 #endif
4323
4324 /* Loop through the string, looking for a place to start matching. */
4325 for (;;)
4326 {
4327 /* If the pattern is anchored,
4328 skip quickly past places we cannot match.
4329 We don't bother to treat startpos == 0 specially
4330 because that case doesn't repeat. */
4331 if (anchored_start && startpos > 0)
4332 {
4333 if (! ((startpos <= size1 ? string1[startpos - 1]
4334 : string2[startpos - size1 - 1])
4335 == '\n'))
4336 goto advance;
4337 }
4338
4339 /* If a fastmap is supplied, skip quickly over characters that
4340 cannot be the start of a match. If the pattern can match the
4341 null string, however, we don't need to skip characters; we want
4342 the first null string. */
4343 if (fastmap && startpos < total_size && !bufp->can_be_null)
4344 {
4345 register re_char *d;
4346 register re_wchar_t buf_ch;
4347
4348 d = POS_ADDR_VSTRING (startpos);
4349
4350 if (range > 0) /* Searching forwards. */
4351 {
4352 register int lim = 0;
4353 ssize_t irange = range;
4354
4355 if (startpos < size1 && startpos + range >= size1)
4356 lim = range - (size1 - startpos);
4357
4358 /* Written out as an if-else to avoid testing `translate'
4359 inside the loop. */
4360 if (RE_TRANSLATE_P (translate))
4361 {
4362 if (multibyte)
4363 while (range > lim)
4364 {
4365 int buf_charlen;
4366
4367 buf_ch = STRING_CHAR_AND_LENGTH (d, buf_charlen);
4368 buf_ch = RE_TRANSLATE (translate, buf_ch);
4369 if (fastmap[CHAR_LEADING_CODE (buf_ch)])
4370 break;
4371
4372 range -= buf_charlen;
4373 d += buf_charlen;
4374 }
4375 else
4376 while (range > lim)
4377 {
4378 register re_wchar_t ch, translated;
4379
4380 buf_ch = *d;
4381 ch = RE_CHAR_TO_MULTIBYTE (buf_ch);
4382 translated = RE_TRANSLATE (translate, ch);
4383 if (translated != ch
4384 && (ch = RE_CHAR_TO_UNIBYTE (translated)) >= 0)
4385 buf_ch = ch;
4386 if (fastmap[buf_ch])
4387 break;
4388 d++;
4389 range--;
4390 }
4391 }
4392 else
4393 {
4394 if (multibyte)
4395 while (range > lim)
4396 {
4397 int buf_charlen;
4398
4399 buf_ch = STRING_CHAR_AND_LENGTH (d, buf_charlen);
4400 if (fastmap[CHAR_LEADING_CODE (buf_ch)])
4401 break;
4402 range -= buf_charlen;
4403 d += buf_charlen;
4404 }
4405 else
4406 while (range > lim && !fastmap[*d])
4407 {
4408 d++;
4409 range--;
4410 }
4411 }
4412 startpos += irange - range;
4413 }
4414 else /* Searching backwards. */
4415 {
4416 if (multibyte)
4417 {
4418 buf_ch = STRING_CHAR (d);
4419 buf_ch = TRANSLATE (buf_ch);
4420 if (! fastmap[CHAR_LEADING_CODE (buf_ch)])
4421 goto advance;
4422 }
4423 else
4424 {
4425 register re_wchar_t ch, translated;
4426
4427 buf_ch = *d;
4428 ch = RE_CHAR_TO_MULTIBYTE (buf_ch);
4429 translated = TRANSLATE (ch);
4430 if (translated != ch
4431 && (ch = RE_CHAR_TO_UNIBYTE (translated)) >= 0)
4432 buf_ch = ch;
4433 if (! fastmap[TRANSLATE (buf_ch)])
4434 goto advance;
4435 }
4436 }
4437 }
4438
4439 /* If can't match the null string, and that's all we have left, fail. */
4440 if (range >= 0 && startpos == total_size && fastmap
4441 && !bufp->can_be_null)
4442 return -1;
4443
4444 val = re_match_2_internal (bufp, string1, size1, string2, size2,
4445 startpos, regs, stop);
4446
4447 if (val >= 0)
4448 return startpos;
4449
4450 if (val == -2)
4451 return -2;
4452
4453 advance:
4454 if (!range)
4455 break;
4456 else if (range > 0)
4457 {
4458 /* Update STARTPOS to the next character boundary. */
4459 if (multibyte)
4460 {
4461 re_char *p = POS_ADDR_VSTRING (startpos);
4462 int len = BYTES_BY_CHAR_HEAD (*p);
4463
4464 range -= len;
4465 if (range < 0)
4466 break;
4467 startpos += len;
4468 }
4469 else
4470 {
4471 range--;
4472 startpos++;
4473 }
4474 }
4475 else
4476 {
4477 range++;
4478 startpos--;
4479
4480 /* Update STARTPOS to the previous character boundary. */
4481 if (multibyte)
4482 {
4483 re_char *p = POS_ADDR_VSTRING (startpos) + 1;
4484 re_char *p0 = p;
4485 re_char *phead = HEAD_ADDR_VSTRING (startpos);
4486
4487 /* Find the head of multibyte form. */
4488 PREV_CHAR_BOUNDARY (p, phead);
4489 range += p0 - 1 - p;
4490 if (range > 0)
4491 break;
4492
4493 startpos -= p0 - 1 - p;
4494 }
4495 }
4496 }
4497 return -1;
4498 } /* re_search_2 */
4499 WEAK_ALIAS (__re_search_2, re_search_2)
4500 \f
4501 /* Declarations and macros for re_match_2. */
4502
4503 static int bcmp_translate (re_char *s1, re_char *s2,
4504 register ssize_t len,
4505 RE_TRANSLATE_TYPE translate,
4506 const int multibyte);
4507
4508 /* This converts PTR, a pointer into one of the search strings `string1'
4509 and `string2' into an offset from the beginning of that string. */
4510 #define POINTER_TO_OFFSET(ptr) \
4511 (FIRST_STRING_P (ptr) \
4512 ? (ptr) - string1 \
4513 : (ptr) - string2 + (ptrdiff_t) size1)
4514
4515 /* Call before fetching a character with *d. This switches over to
4516 string2 if necessary.
4517 Check re_match_2_internal for a discussion of why end_match_2 might
4518 not be within string2 (but be equal to end_match_1 instead). */
4519 #define PREFETCH() \
4520 while (d == dend) \
4521 { \
4522 /* End of string2 => fail. */ \
4523 if (dend == end_match_2) \
4524 goto fail; \
4525 /* End of string1 => advance to string2. */ \
4526 d = string2; \
4527 dend = end_match_2; \
4528 }
4529
4530 /* Call before fetching a char with *d if you already checked other limits.
4531 This is meant for use in lookahead operations like wordend, etc..
4532 where we might need to look at parts of the string that might be
4533 outside of the LIMITs (i.e past `stop'). */
4534 #define PREFETCH_NOLIMIT() \
4535 if (d == end1) \
4536 { \
4537 d = string2; \
4538 dend = end_match_2; \
4539 } \
4540
4541 /* Test if at very beginning or at very end of the virtual concatenation
4542 of `string1' and `string2'. If only one string, it's `string2'. */
4543 #define AT_STRINGS_BEG(d) ((d) == (size1 ? string1 : string2) || !size2)
4544 #define AT_STRINGS_END(d) ((d) == end2)
4545
4546 /* Disabled due to a compiler bug -- see comment at case wordbound */
4547
4548 /* The comment at case wordbound is following one, but we don't use
4549 AT_WORD_BOUNDARY anymore to support multibyte form.
4550
4551 The DEC Alpha C compiler 3.x generates incorrect code for the
4552 test WORDCHAR_P (d - 1) != WORDCHAR_P (d) in the expansion of
4553 AT_WORD_BOUNDARY, so this code is disabled. Expanding the
4554 macro and introducing temporary variables works around the bug. */
4555
4556 #if 0
4557 /* Test if D points to a character which is word-constituent. We have
4558 two special cases to check for: if past the end of string1, look at
4559 the first character in string2; and if before the beginning of
4560 string2, look at the last character in string1. */
4561 #define WORDCHAR_P(d) \
4562 (SYNTAX ((d) == end1 ? *string2 \
4563 : (d) == string2 - 1 ? *(end1 - 1) : *(d)) \
4564 == Sword)
4565
4566 /* Test if the character before D and the one at D differ with respect
4567 to being word-constituent. */
4568 #define AT_WORD_BOUNDARY(d) \
4569 (AT_STRINGS_BEG (d) || AT_STRINGS_END (d) \
4570 || WORDCHAR_P (d - 1) != WORDCHAR_P (d))
4571 #endif
4572
4573 /* Free everything we malloc. */
4574 #ifdef MATCH_MAY_ALLOCATE
4575 # define FREE_VAR(var) \
4576 do { \
4577 if (var) \
4578 { \
4579 REGEX_FREE (var); \
4580 var = NULL; \
4581 } \
4582 } while (0)
4583 # define FREE_VARIABLES() \
4584 do { \
4585 REGEX_FREE_STACK (fail_stack.stack); \
4586 FREE_VAR (regstart); \
4587 FREE_VAR (regend); \
4588 FREE_VAR (best_regstart); \
4589 FREE_VAR (best_regend); \
4590 } while (0)
4591 #else
4592 # define FREE_VARIABLES() ((void)0) /* Do nothing! But inhibit gcc warning. */
4593 #endif /* not MATCH_MAY_ALLOCATE */
4594
4595 \f
4596 /* Optimization routines. */
4597
4598 /* If the operation is a match against one or more chars,
4599 return a pointer to the next operation, else return NULL. */
4600 static re_char *
4601 skip_one_char (const re_char *p)
4602 {
4603 switch (*p++)
4604 {
4605 case anychar:
4606 break;
4607
4608 case exactn:
4609 p += *p + 1;
4610 break;
4611
4612 case charset_not:
4613 case charset:
4614 if (CHARSET_RANGE_TABLE_EXISTS_P (p - 1))
4615 {
4616 int mcnt;
4617 p = CHARSET_RANGE_TABLE (p - 1);
4618 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4619 p = CHARSET_RANGE_TABLE_END (p, mcnt);
4620 }
4621 else
4622 p += 1 + CHARSET_BITMAP_SIZE (p - 1);
4623 break;
4624
4625 case syntaxspec:
4626 case notsyntaxspec:
4627 #ifdef emacs
4628 case categoryspec:
4629 case notcategoryspec:
4630 #endif /* emacs */
4631 p++;
4632 break;
4633
4634 default:
4635 p = NULL;
4636 }
4637 return p;
4638 }
4639
4640
4641 /* Jump over non-matching operations. */
4642 static re_char *
4643 skip_noops (const re_char *p, const re_char *pend)
4644 {
4645 int mcnt;
4646 while (p < pend)
4647 {
4648 switch (*p)
4649 {
4650 case start_memory:
4651 case stop_memory:
4652 p += 2; break;
4653 case no_op:
4654 p += 1; break;
4655 case jump:
4656 p += 1;
4657 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4658 p += mcnt;
4659 break;
4660 default:
4661 return p;
4662 }
4663 }
4664 assert (p == pend);
4665 return p;
4666 }
4667
4668 /* Non-zero if "p1 matches something" implies "p2 fails". */
4669 static int
4670 mutually_exclusive_p (struct re_pattern_buffer *bufp, const re_char *p1, 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 /* re_match_2 matches the compiled pattern in BUFP against the
4886 the (virtual) concatenation of STRING1 and STRING2 (of length SIZE1
4887 and SIZE2, respectively). We start matching at POS, and stop
4888 matching at STOP.
4889
4890 If REGS is non-null and the `no_sub' field of BUFP is nonzero, we
4891 store offsets for the substring each group matched in REGS. See the
4892 documentation for exactly how many groups we fill.
4893
4894 We return -1 if no match, -2 if an internal error (such as the
4895 failure stack overflowing). Otherwise, we return the length of the
4896 matched substring. */
4897
4898 regoff_t
4899 re_match_2 (struct re_pattern_buffer *bufp, const char *string1,
4900 size_t size1, const char *string2, size_t size2, ssize_t pos,
4901 struct re_registers *regs, ssize_t stop)
4902 {
4903 regoff_t result;
4904
4905 #ifdef emacs
4906 ssize_t charpos;
4907 gl_state.object = re_match_object; /* Used by SYNTAX_TABLE_BYTE_TO_CHAR. */
4908 charpos = SYNTAX_TABLE_BYTE_TO_CHAR (POS_AS_IN_BUFFER (pos));
4909 SETUP_SYNTAX_TABLE_FOR_OBJECT (re_match_object, charpos, 1);
4910 #endif
4911
4912 result = re_match_2_internal (bufp, (re_char*) string1, size1,
4913 (re_char*) string2, size2,
4914 pos, regs, stop);
4915 return result;
4916 }
4917 WEAK_ALIAS (__re_match_2, re_match_2)
4918
4919
4920 /* This is a separate function so that we can force an alloca cleanup
4921 afterwards. */
4922 static regoff_t
4923 re_match_2_internal (struct re_pattern_buffer *bufp, const re_char *string1,
4924 size_t size1, const re_char *string2, size_t size2,
4925 ssize_t pos, struct re_registers *regs, ssize_t stop)
4926 {
4927 /* General temporaries. */
4928 int mcnt;
4929 size_t reg;
4930
4931 /* Just past the end of the corresponding string. */
4932 re_char *end1, *end2;
4933
4934 /* Pointers into string1 and string2, just past the last characters in
4935 each to consider matching. */
4936 re_char *end_match_1, *end_match_2;
4937
4938 /* Where we are in the data, and the end of the current string. */
4939 re_char *d, *dend;
4940
4941 /* Used sometimes to remember where we were before starting matching
4942 an operator so that we can go back in case of failure. This "atomic"
4943 behavior of matching opcodes is indispensable to the correctness
4944 of the on_failure_keep_string_jump optimization. */
4945 re_char *dfail;
4946
4947 /* Where we are in the pattern, and the end of the pattern. */
4948 re_char *p = bufp->buffer;
4949 re_char *pend = p + bufp->used;
4950
4951 /* We use this to map every character in the string. */
4952 RE_TRANSLATE_TYPE translate = bufp->translate;
4953
4954 /* Nonzero if BUFP is setup from a multibyte regex. */
4955 const boolean multibyte = RE_MULTIBYTE_P (bufp);
4956
4957 /* Nonzero if STRING1/STRING2 are multibyte. */
4958 const boolean target_multibyte = RE_TARGET_MULTIBYTE_P (bufp);
4959
4960 /* Failure point stack. Each place that can handle a failure further
4961 down the line pushes a failure point on this stack. It consists of
4962 regstart, and regend for all registers corresponding to
4963 the subexpressions we're currently inside, plus the number of such
4964 registers, and, finally, two char *'s. The first char * is where
4965 to resume scanning the pattern; the second one is where to resume
4966 scanning the strings. */
4967 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
4968 fail_stack_type fail_stack;
4969 #endif
4970 #ifdef DEBUG_COMPILES_ARGUMENTS
4971 unsigned nfailure_points_pushed = 0, nfailure_points_popped = 0;
4972 #endif
4973
4974 #if defined REL_ALLOC && defined REGEX_MALLOC
4975 /* This holds the pointer to the failure stack, when
4976 it is allocated relocatably. */
4977 fail_stack_elt_t *failure_stack_ptr;
4978 #endif
4979
4980 /* We fill all the registers internally, independent of what we
4981 return, for use in backreferences. The number here includes
4982 an element for register zero. */
4983 size_t num_regs = bufp->re_nsub + 1;
4984
4985 /* Information on the contents of registers. These are pointers into
4986 the input strings; they record just what was matched (on this
4987 attempt) by a subexpression part of the pattern, that is, the
4988 regnum-th regstart pointer points to where in the pattern we began
4989 matching and the regnum-th regend points to right after where we
4990 stopped matching the regnum-th subexpression. (The zeroth register
4991 keeps track of what the whole pattern matches.) */
4992 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
4993 re_char **regstart, **regend;
4994 #endif
4995
4996 /* The following record the register info as found in the above
4997 variables when we find a match better than any we've seen before.
4998 This happens as we backtrack through the failure points, which in
4999 turn happens only if we have not yet matched the entire string. */
5000 unsigned best_regs_set = false;
5001 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
5002 re_char **best_regstart, **best_regend;
5003 #endif
5004
5005 /* Logically, this is `best_regend[0]'. But we don't want to have to
5006 allocate space for that if we're not allocating space for anything
5007 else (see below). Also, we never need info about register 0 for
5008 any of the other register vectors, and it seems rather a kludge to
5009 treat `best_regend' differently than the rest. So we keep track of
5010 the end of the best match so far in a separate variable. We
5011 initialize this to NULL so that when we backtrack the first time
5012 and need to test it, it's not garbage. */
5013 re_char *match_end = NULL;
5014
5015 #ifdef DEBUG_COMPILES_ARGUMENTS
5016 /* Counts the total number of registers pushed. */
5017 unsigned num_regs_pushed = 0;
5018 #endif
5019
5020 DEBUG_PRINT ("\n\nEntering re_match_2.\n");
5021
5022 INIT_FAIL_STACK ();
5023
5024 #ifdef MATCH_MAY_ALLOCATE
5025 /* Do not bother to initialize all the register variables if there are
5026 no groups in the pattern, as it takes a fair amount of time. If
5027 there are groups, we include space for register 0 (the whole
5028 pattern), even though we never use it, since it simplifies the
5029 array indexing. We should fix this. */
5030 if (bufp->re_nsub)
5031 {
5032 regstart = REGEX_TALLOC (num_regs, re_char *);
5033 regend = REGEX_TALLOC (num_regs, re_char *);
5034 best_regstart = REGEX_TALLOC (num_regs, re_char *);
5035 best_regend = REGEX_TALLOC (num_regs, re_char *);
5036
5037 if (!(regstart && regend && best_regstart && best_regend))
5038 {
5039 FREE_VARIABLES ();
5040 return -2;
5041 }
5042 }
5043 else
5044 {
5045 /* We must initialize all our variables to NULL, so that
5046 `FREE_VARIABLES' doesn't try to free them. */
5047 regstart = regend = best_regstart = best_regend = NULL;
5048 }
5049 #endif /* MATCH_MAY_ALLOCATE */
5050
5051 /* The starting position is bogus. */
5052 if (pos < 0 || pos > size1 + size2)
5053 {
5054 FREE_VARIABLES ();
5055 return -1;
5056 }
5057
5058 /* Initialize subexpression text positions to -1 to mark ones that no
5059 start_memory/stop_memory has been seen for. Also initialize the
5060 register information struct. */
5061 for (reg = 1; reg < num_regs; reg++)
5062 regstart[reg] = regend[reg] = NULL;
5063
5064 /* We move `string1' into `string2' if the latter's empty -- but not if
5065 `string1' is null. */
5066 if (size2 == 0 && string1 != NULL)
5067 {
5068 string2 = string1;
5069 size2 = size1;
5070 string1 = 0;
5071 size1 = 0;
5072 }
5073 end1 = string1 + size1;
5074 end2 = string2 + size2;
5075
5076 /* `p' scans through the pattern as `d' scans through the data.
5077 `dend' is the end of the input string that `d' points within. `d'
5078 is advanced into the following input string whenever necessary, but
5079 this happens before fetching; therefore, at the beginning of the
5080 loop, `d' can be pointing at the end of a string, but it cannot
5081 equal `string2'. */
5082 if (pos >= size1)
5083 {
5084 /* Only match within string2. */
5085 d = string2 + pos - size1;
5086 dend = end_match_2 = string2 + stop - size1;
5087 end_match_1 = end1; /* Just to give it a value. */
5088 }
5089 else
5090 {
5091 if (stop < size1)
5092 {
5093 /* Only match within string1. */
5094 end_match_1 = string1 + stop;
5095 /* BEWARE!
5096 When we reach end_match_1, PREFETCH normally switches to string2.
5097 But in the present case, this means that just doing a PREFETCH
5098 makes us jump from `stop' to `gap' within the string.
5099 What we really want here is for the search to stop as
5100 soon as we hit end_match_1. That's why we set end_match_2
5101 to end_match_1 (since PREFETCH fails as soon as we hit
5102 end_match_2). */
5103 end_match_2 = end_match_1;
5104 }
5105 else
5106 { /* It's important to use this code when stop == size so that
5107 moving `d' from end1 to string2 will not prevent the d == dend
5108 check from catching the end of string. */
5109 end_match_1 = end1;
5110 end_match_2 = string2 + stop - size1;
5111 }
5112 d = string1 + pos;
5113 dend = end_match_1;
5114 }
5115
5116 DEBUG_PRINT ("The compiled pattern is: ");
5117 DEBUG_PRINT_COMPILED_PATTERN (bufp, p, pend);
5118 DEBUG_PRINT ("The string to match is: `");
5119 DEBUG_PRINT_DOUBLE_STRING (d, string1, size1, string2, size2);
5120 DEBUG_PRINT ("'\n");
5121
5122 /* This loops over pattern commands. It exits by returning from the
5123 function if the match is complete, or it drops through if the match
5124 fails at this starting point in the input data. */
5125 for (;;)
5126 {
5127 DEBUG_PRINT ("\n%p: ", p);
5128
5129 if (p == pend)
5130 {
5131 ptrdiff_t dcnt;
5132
5133 /* End of pattern means we might have succeeded. */
5134 DEBUG_PRINT ("end of pattern ... ");
5135
5136 /* If we haven't matched the entire string, and we want the
5137 longest match, try backtracking. */
5138 if (d != end_match_2)
5139 {
5140 /* 1 if this match ends in the same string (string1 or string2)
5141 as the best previous match. */
5142 boolean same_str_p = (FIRST_STRING_P (match_end)
5143 == FIRST_STRING_P (d));
5144 /* 1 if this match is the best seen so far. */
5145 boolean best_match_p;
5146
5147 /* AIX compiler got confused when this was combined
5148 with the previous declaration. */
5149 if (same_str_p)
5150 best_match_p = d > match_end;
5151 else
5152 best_match_p = !FIRST_STRING_P (d);
5153
5154 DEBUG_PRINT ("backtracking.\n");
5155
5156 if (!FAIL_STACK_EMPTY ())
5157 { /* More failure points to try. */
5158
5159 /* If exceeds best match so far, save it. */
5160 if (!best_regs_set || best_match_p)
5161 {
5162 best_regs_set = true;
5163 match_end = d;
5164
5165 DEBUG_PRINT ("\nSAVING match as best so far.\n");
5166
5167 for (reg = 1; reg < num_regs; reg++)
5168 {
5169 best_regstart[reg] = regstart[reg];
5170 best_regend[reg] = regend[reg];
5171 }
5172 }
5173 goto fail;
5174 }
5175
5176 /* If no failure points, don't restore garbage. And if
5177 last match is real best match, don't restore second
5178 best one. */
5179 else if (best_regs_set && !best_match_p)
5180 {
5181 restore_best_regs:
5182 /* Restore best match. It may happen that `dend ==
5183 end_match_1' while the restored d is in string2.
5184 For example, the pattern `x.*y.*z' against the
5185 strings `x-' and `y-z-', if the two strings are
5186 not consecutive in memory. */
5187 DEBUG_PRINT ("Restoring best registers.\n");
5188
5189 d = match_end;
5190 dend = ((d >= string1 && d <= end1)
5191 ? end_match_1 : end_match_2);
5192
5193 for (reg = 1; reg < num_regs; reg++)
5194 {
5195 regstart[reg] = best_regstart[reg];
5196 regend[reg] = best_regend[reg];
5197 }
5198 }
5199 } /* d != end_match_2 */
5200
5201 succeed_label:
5202 DEBUG_PRINT ("Accepting match.\n");
5203
5204 /* If caller wants register contents data back, do it. */
5205 if (regs && !bufp->no_sub)
5206 {
5207 /* Have the register data arrays been allocated? */
5208 if (bufp->regs_allocated == REGS_UNALLOCATED)
5209 { /* No. So allocate them with malloc. We need one
5210 extra element beyond `num_regs' for the `-1' marker
5211 GNU code uses. */
5212 regs->num_regs = MAX (RE_NREGS, num_regs + 1);
5213 regs->start = TALLOC (regs->num_regs, regoff_t);
5214 regs->end = TALLOC (regs->num_regs, regoff_t);
5215 if (regs->start == NULL || regs->end == NULL)
5216 {
5217 FREE_VARIABLES ();
5218 return -2;
5219 }
5220 bufp->regs_allocated = REGS_REALLOCATE;
5221 }
5222 else if (bufp->regs_allocated == REGS_REALLOCATE)
5223 { /* Yes. If we need more elements than were already
5224 allocated, reallocate them. If we need fewer, just
5225 leave it alone. */
5226 if (regs->num_regs < num_regs + 1)
5227 {
5228 regs->num_regs = num_regs + 1;
5229 RETALLOC (regs->start, regs->num_regs, regoff_t);
5230 RETALLOC (regs->end, regs->num_regs, regoff_t);
5231 if (regs->start == NULL || regs->end == NULL)
5232 {
5233 FREE_VARIABLES ();
5234 return -2;
5235 }
5236 }
5237 }
5238 else
5239 {
5240 /* These braces fend off a "empty body in an else-statement"
5241 warning under GCC when assert expands to nothing. */
5242 assert (bufp->regs_allocated == REGS_FIXED);
5243 }
5244
5245 /* Convert the pointer data in `regstart' and `regend' to
5246 indices. Register zero has to be set differently,
5247 since we haven't kept track of any info for it. */
5248 if (regs->num_regs > 0)
5249 {
5250 regs->start[0] = pos;
5251 regs->end[0] = POINTER_TO_OFFSET (d);
5252 }
5253
5254 /* Go through the first `min (num_regs, regs->num_regs)'
5255 registers, since that is all we initialized. */
5256 for (reg = 1; reg < MIN (num_regs, regs->num_regs); reg++)
5257 {
5258 if (REG_UNSET (regstart[reg]) || REG_UNSET (regend[reg]))
5259 regs->start[reg] = regs->end[reg] = -1;
5260 else
5261 {
5262 regs->start[reg] = POINTER_TO_OFFSET (regstart[reg]);
5263 regs->end[reg] = POINTER_TO_OFFSET (regend[reg]);
5264 }
5265 }
5266
5267 /* If the regs structure we return has more elements than
5268 were in the pattern, set the extra elements to -1. If
5269 we (re)allocated the registers, this is the case,
5270 because we always allocate enough to have at least one
5271 -1 at the end. */
5272 for (reg = num_regs; reg < regs->num_regs; reg++)
5273 regs->start[reg] = regs->end[reg] = -1;
5274 } /* regs && !bufp->no_sub */
5275
5276 DEBUG_PRINT ("%u failure points pushed, %u popped (%u remain).\n",
5277 nfailure_points_pushed, nfailure_points_popped,
5278 nfailure_points_pushed - nfailure_points_popped);
5279 DEBUG_PRINT ("%u registers pushed.\n", num_regs_pushed);
5280
5281 dcnt = POINTER_TO_OFFSET (d) - pos;
5282
5283 DEBUG_PRINT ("Returning %td from re_match_2.\n", dcnt);
5284
5285 FREE_VARIABLES ();
5286 return dcnt;
5287 }
5288
5289 /* Otherwise match next pattern command. */
5290 switch (*p++)
5291 {
5292 /* Ignore these. Used to ignore the n of succeed_n's which
5293 currently have n == 0. */
5294 case no_op:
5295 DEBUG_PRINT ("EXECUTING no_op.\n");
5296 break;
5297
5298 case succeed:
5299 DEBUG_PRINT ("EXECUTING succeed.\n");
5300 goto succeed_label;
5301
5302 /* Match the next n pattern characters exactly. The following
5303 byte in the pattern defines n, and the n bytes after that
5304 are the characters to match. */
5305 case exactn:
5306 mcnt = *p++;
5307 DEBUG_PRINT ("EXECUTING exactn %d.\n", mcnt);
5308
5309 /* Remember the start point to rollback upon failure. */
5310 dfail = d;
5311
5312 #ifndef emacs
5313 /* This is written out as an if-else so we don't waste time
5314 testing `translate' inside the loop. */
5315 if (RE_TRANSLATE_P (translate))
5316 do
5317 {
5318 PREFETCH ();
5319 if (RE_TRANSLATE (translate, *d) != *p++)
5320 {
5321 d = dfail;
5322 goto fail;
5323 }
5324 d++;
5325 }
5326 while (--mcnt);
5327 else
5328 do
5329 {
5330 PREFETCH ();
5331 if (*d++ != *p++)
5332 {
5333 d = dfail;
5334 goto fail;
5335 }
5336 }
5337 while (--mcnt);
5338 #else /* emacs */
5339 /* The cost of testing `translate' is comparatively small. */
5340 if (target_multibyte)
5341 do
5342 {
5343 int pat_charlen, buf_charlen;
5344 int pat_ch, buf_ch;
5345
5346 PREFETCH ();
5347 if (multibyte)
5348 pat_ch = STRING_CHAR_AND_LENGTH (p, pat_charlen);
5349 else
5350 {
5351 pat_ch = RE_CHAR_TO_MULTIBYTE (*p);
5352 pat_charlen = 1;
5353 }
5354 buf_ch = STRING_CHAR_AND_LENGTH (d, buf_charlen);
5355
5356 if (TRANSLATE (buf_ch) != pat_ch)
5357 {
5358 d = dfail;
5359 goto fail;
5360 }
5361
5362 p += pat_charlen;
5363 d += buf_charlen;
5364 mcnt -= pat_charlen;
5365 }
5366 while (mcnt > 0);
5367 else
5368 do
5369 {
5370 int pat_charlen;
5371 int pat_ch, buf_ch;
5372
5373 PREFETCH ();
5374 if (multibyte)
5375 {
5376 pat_ch = STRING_CHAR_AND_LENGTH (p, pat_charlen);
5377 pat_ch = RE_CHAR_TO_UNIBYTE (pat_ch);
5378 }
5379 else
5380 {
5381 pat_ch = *p;
5382 pat_charlen = 1;
5383 }
5384 buf_ch = RE_CHAR_TO_MULTIBYTE (*d);
5385 if (! CHAR_BYTE8_P (buf_ch))
5386 {
5387 buf_ch = TRANSLATE (buf_ch);
5388 buf_ch = RE_CHAR_TO_UNIBYTE (buf_ch);
5389 if (buf_ch < 0)
5390 buf_ch = *d;
5391 }
5392 else
5393 buf_ch = *d;
5394 if (buf_ch != pat_ch)
5395 {
5396 d = dfail;
5397 goto fail;
5398 }
5399 p += pat_charlen;
5400 d++;
5401 }
5402 while (--mcnt);
5403 #endif
5404 break;
5405
5406
5407 /* Match any character except possibly a newline or a null. */
5408 case anychar:
5409 {
5410 int buf_charlen;
5411 re_wchar_t buf_ch;
5412
5413 DEBUG_PRINT ("EXECUTING anychar.\n");
5414
5415 PREFETCH ();
5416 buf_ch = RE_STRING_CHAR_AND_LENGTH (d, buf_charlen,
5417 target_multibyte);
5418 buf_ch = TRANSLATE (buf_ch);
5419
5420 if ((!(bufp->syntax & RE_DOT_NEWLINE)
5421 && buf_ch == '\n')
5422 || ((bufp->syntax & RE_DOT_NOT_NULL)
5423 && buf_ch == '\000'))
5424 goto fail;
5425
5426 DEBUG_PRINT (" Matched `%d'.\n", *d);
5427 d += buf_charlen;
5428 }
5429 break;
5430
5431
5432 case charset:
5433 case charset_not:
5434 {
5435 register unsigned int c;
5436 boolean not = (re_opcode_t) *(p - 1) == charset_not;
5437 int len;
5438
5439 /* Start of actual range_table, or end of bitmap if there is no
5440 range table. */
5441 re_char *range_table IF_LINT (= NULL);
5442
5443 /* Nonzero if there is a range table. */
5444 int range_table_exists;
5445
5446 /* Number of ranges of range table. This is not included
5447 in the initial byte-length of the command. */
5448 int count = 0;
5449
5450 /* Whether matching against a unibyte character. */
5451 boolean unibyte_char = false;
5452
5453 DEBUG_PRINT ("EXECUTING charset%s.\n", not ? "_not" : "");
5454
5455 range_table_exists = CHARSET_RANGE_TABLE_EXISTS_P (&p[-1]);
5456
5457 if (range_table_exists)
5458 {
5459 range_table = CHARSET_RANGE_TABLE (&p[-1]); /* Past the bitmap. */
5460 EXTRACT_NUMBER_AND_INCR (count, range_table);
5461 }
5462
5463 PREFETCH ();
5464 c = RE_STRING_CHAR_AND_LENGTH (d, len, target_multibyte);
5465 if (target_multibyte)
5466 {
5467 int c1;
5468
5469 c = TRANSLATE (c);
5470 c1 = RE_CHAR_TO_UNIBYTE (c);
5471 if (c1 >= 0)
5472 {
5473 unibyte_char = true;
5474 c = c1;
5475 }
5476 }
5477 else
5478 {
5479 int c1 = RE_CHAR_TO_MULTIBYTE (c);
5480
5481 if (! CHAR_BYTE8_P (c1))
5482 {
5483 c1 = TRANSLATE (c1);
5484 c1 = RE_CHAR_TO_UNIBYTE (c1);
5485 if (c1 >= 0)
5486 {
5487 unibyte_char = true;
5488 c = c1;
5489 }
5490 }
5491 else
5492 unibyte_char = true;
5493 }
5494
5495 if (unibyte_char && c < (1 << BYTEWIDTH))
5496 { /* Lookup bitmap. */
5497 /* Cast to `unsigned' instead of `unsigned char' in
5498 case the bit list is a full 32 bytes long. */
5499 if (c < (unsigned) (CHARSET_BITMAP_SIZE (&p[-1]) * BYTEWIDTH)
5500 && p[1 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH)))
5501 not = !not;
5502 }
5503 #ifdef emacs
5504 else if (range_table_exists)
5505 {
5506 int class_bits = CHARSET_RANGE_TABLE_BITS (&p[-1]);
5507
5508 if ( (class_bits & BIT_LOWER && ISLOWER (c))
5509 | (class_bits & BIT_MULTIBYTE)
5510 | (class_bits & BIT_PUNCT && ISPUNCT (c))
5511 | (class_bits & BIT_SPACE && ISSPACE (c))
5512 | (class_bits & BIT_UPPER && ISUPPER (c))
5513 | (class_bits & BIT_WORD && ISWORD (c)))
5514 not = !not;
5515 else
5516 CHARSET_LOOKUP_RANGE_TABLE_RAW (not, c, range_table, count);
5517 }
5518 #endif /* emacs */
5519
5520 if (range_table_exists)
5521 p = CHARSET_RANGE_TABLE_END (range_table, count);
5522 else
5523 p += CHARSET_BITMAP_SIZE (&p[-1]) + 1;
5524
5525 if (!not) goto fail;
5526
5527 d += len;
5528 }
5529 break;
5530
5531
5532 /* The beginning of a group is represented by start_memory.
5533 The argument is the register number. The text
5534 matched within the group is recorded (in the internal
5535 registers data structure) under the register number. */
5536 case start_memory:
5537 DEBUG_PRINT ("EXECUTING start_memory %d:\n", *p);
5538
5539 /* In case we need to undo this operation (via backtracking). */
5540 PUSH_FAILURE_REG (*p);
5541
5542 regstart[*p] = d;
5543 regend[*p] = NULL; /* probably unnecessary. -sm */
5544 DEBUG_PRINT (" regstart: %td\n", POINTER_TO_OFFSET (regstart[*p]));
5545
5546 /* Move past the register number and inner group count. */
5547 p += 1;
5548 break;
5549
5550
5551 /* The stop_memory opcode represents the end of a group. Its
5552 argument is the same as start_memory's: the register number. */
5553 case stop_memory:
5554 DEBUG_PRINT ("EXECUTING stop_memory %d:\n", *p);
5555
5556 assert (!REG_UNSET (regstart[*p]));
5557 /* Strictly speaking, there should be code such as:
5558
5559 assert (REG_UNSET (regend[*p]));
5560 PUSH_FAILURE_REGSTOP ((unsigned int)*p);
5561
5562 But the only info to be pushed is regend[*p] and it is known to
5563 be UNSET, so there really isn't anything to push.
5564 Not pushing anything, on the other hand deprives us from the
5565 guarantee that regend[*p] is UNSET since undoing this operation
5566 will not reset its value properly. This is not important since
5567 the value will only be read on the next start_memory or at
5568 the very end and both events can only happen if this stop_memory
5569 is *not* undone. */
5570
5571 regend[*p] = d;
5572 DEBUG_PRINT (" regend: %td\n", POINTER_TO_OFFSET (regend[*p]));
5573
5574 /* Move past the register number and the inner group count. */
5575 p += 1;
5576 break;
5577
5578
5579 /* \<digit> has been turned into a `duplicate' command which is
5580 followed by the numeric value of <digit> as the register number. */
5581 case duplicate:
5582 {
5583 register re_char *d2, *dend2;
5584 int regno = *p++; /* Get which register to match against. */
5585 DEBUG_PRINT ("EXECUTING duplicate %d.\n", regno);
5586
5587 /* Can't back reference a group which we've never matched. */
5588 if (REG_UNSET (regstart[regno]) || REG_UNSET (regend[regno]))
5589 goto fail;
5590
5591 /* Where in input to try to start matching. */
5592 d2 = regstart[regno];
5593
5594 /* Remember the start point to rollback upon failure. */
5595 dfail = d;
5596
5597 /* Where to stop matching; if both the place to start and
5598 the place to stop matching are in the same string, then
5599 set to the place to stop, otherwise, for now have to use
5600 the end of the first string. */
5601
5602 dend2 = ((FIRST_STRING_P (regstart[regno])
5603 == FIRST_STRING_P (regend[regno]))
5604 ? regend[regno] : end_match_1);
5605 for (;;)
5606 {
5607 ptrdiff_t dcnt;
5608
5609 /* If necessary, advance to next segment in register
5610 contents. */
5611 while (d2 == dend2)
5612 {
5613 if (dend2 == end_match_2) break;
5614 if (dend2 == regend[regno]) break;
5615
5616 /* End of string1 => advance to string2. */
5617 d2 = string2;
5618 dend2 = regend[regno];
5619 }
5620 /* At end of register contents => success */
5621 if (d2 == dend2) break;
5622
5623 /* If necessary, advance to next segment in data. */
5624 PREFETCH ();
5625
5626 /* How many characters left in this segment to match. */
5627 dcnt = dend - d;
5628
5629 /* Want how many consecutive characters we can match in
5630 one shot, so, if necessary, adjust the count. */
5631 if (dcnt > dend2 - d2)
5632 dcnt = dend2 - d2;
5633
5634 /* Compare that many; failure if mismatch, else move
5635 past them. */
5636 if (RE_TRANSLATE_P (translate)
5637 ? bcmp_translate (d, d2, dcnt, translate, target_multibyte)
5638 : memcmp (d, d2, dcnt))
5639 {
5640 d = dfail;
5641 goto fail;
5642 }
5643 d += dcnt, d2 += dcnt;
5644 }
5645 }
5646 break;
5647
5648
5649 /* begline matches the empty string at the beginning of the string
5650 (unless `not_bol' is set in `bufp'), and after newlines. */
5651 case begline:
5652 DEBUG_PRINT ("EXECUTING begline.\n");
5653
5654 if (AT_STRINGS_BEG (d))
5655 {
5656 if (!bufp->not_bol) break;
5657 }
5658 else
5659 {
5660 unsigned c;
5661 GET_CHAR_BEFORE_2 (c, d, string1, end1, string2, end2);
5662 if (c == '\n')
5663 break;
5664 }
5665 /* In all other cases, we fail. */
5666 goto fail;
5667
5668
5669 /* endline is the dual of begline. */
5670 case endline:
5671 DEBUG_PRINT ("EXECUTING endline.\n");
5672
5673 if (AT_STRINGS_END (d))
5674 {
5675 if (!bufp->not_eol) break;
5676 }
5677 else
5678 {
5679 PREFETCH_NOLIMIT ();
5680 if (*d == '\n')
5681 break;
5682 }
5683 goto fail;
5684
5685
5686 /* Match at the very beginning of the data. */
5687 case begbuf:
5688 DEBUG_PRINT ("EXECUTING begbuf.\n");
5689 if (AT_STRINGS_BEG (d))
5690 break;
5691 goto fail;
5692
5693
5694 /* Match at the very end of the data. */
5695 case endbuf:
5696 DEBUG_PRINT ("EXECUTING endbuf.\n");
5697 if (AT_STRINGS_END (d))
5698 break;
5699 goto fail;
5700
5701
5702 /* on_failure_keep_string_jump is used to optimize `.*\n'. It
5703 pushes NULL as the value for the string on the stack. Then
5704 `POP_FAILURE_POINT' will keep the current value for the
5705 string, instead of restoring it. To see why, consider
5706 matching `foo\nbar' against `.*\n'. The .* matches the foo;
5707 then the . fails against the \n. But the next thing we want
5708 to do is match the \n against the \n; if we restored the
5709 string value, we would be back at the foo.
5710
5711 Because this is used only in specific cases, we don't need to
5712 check all the things that `on_failure_jump' does, to make
5713 sure the right things get saved on the stack. Hence we don't
5714 share its code. The only reason to push anything on the
5715 stack at all is that otherwise we would have to change
5716 `anychar's code to do something besides goto fail in this
5717 case; that seems worse than this. */
5718 case on_failure_keep_string_jump:
5719 EXTRACT_NUMBER_AND_INCR (mcnt, p);
5720 DEBUG_PRINT ("EXECUTING on_failure_keep_string_jump %d (to %p):\n",
5721 mcnt, p + mcnt);
5722
5723 PUSH_FAILURE_POINT (p - 3, NULL);
5724 break;
5725
5726 /* A nasty loop is introduced by the non-greedy *? and +?.
5727 With such loops, the stack only ever contains one failure point
5728 at a time, so that a plain on_failure_jump_loop kind of
5729 cycle detection cannot work. Worse yet, such a detection
5730 can not only fail to detect a cycle, but it can also wrongly
5731 detect a cycle (between different instantiations of the same
5732 loop).
5733 So the method used for those nasty loops is a little different:
5734 We use a special cycle-detection-stack-frame which is pushed
5735 when the on_failure_jump_nastyloop failure-point is *popped*.
5736 This special frame thus marks the beginning of one iteration
5737 through the loop and we can hence easily check right here
5738 whether something matched between the beginning and the end of
5739 the loop. */
5740 case on_failure_jump_nastyloop:
5741 EXTRACT_NUMBER_AND_INCR (mcnt, p);
5742 DEBUG_PRINT ("EXECUTING on_failure_jump_nastyloop %d (to %p):\n",
5743 mcnt, p + mcnt);
5744
5745 assert ((re_opcode_t)p[-4] == no_op);
5746 {
5747 int cycle = 0;
5748 CHECK_INFINITE_LOOP (p - 4, d);
5749 if (!cycle)
5750 /* If there's a cycle, just continue without pushing
5751 this failure point. The failure point is the "try again"
5752 option, which shouldn't be tried.
5753 We want (x?)*?y\1z to match both xxyz and xxyxz. */
5754 PUSH_FAILURE_POINT (p - 3, d);
5755 }
5756 break;
5757
5758 /* Simple loop detecting on_failure_jump: just check on the
5759 failure stack if the same spot was already hit earlier. */
5760 case on_failure_jump_loop:
5761 on_failure:
5762 EXTRACT_NUMBER_AND_INCR (mcnt, p);
5763 DEBUG_PRINT ("EXECUTING on_failure_jump_loop %d (to %p):\n",
5764 mcnt, p + mcnt);
5765 {
5766 int cycle = 0;
5767 CHECK_INFINITE_LOOP (p - 3, d);
5768 if (cycle)
5769 /* If there's a cycle, get out of the loop, as if the matching
5770 had failed. We used to just `goto fail' here, but that was
5771 aborting the search a bit too early: we want to keep the
5772 empty-loop-match and keep matching after the loop.
5773 We want (x?)*y\1z to match both xxyz and xxyxz. */
5774 p += mcnt;
5775 else
5776 PUSH_FAILURE_POINT (p - 3, d);
5777 }
5778 break;
5779
5780
5781 /* Uses of on_failure_jump:
5782
5783 Each alternative starts with an on_failure_jump that points
5784 to the beginning of the next alternative. Each alternative
5785 except the last ends with a jump that in effect jumps past
5786 the rest of the alternatives. (They really jump to the
5787 ending jump of the following alternative, because tensioning
5788 these jumps is a hassle.)
5789
5790 Repeats start with an on_failure_jump that points past both
5791 the repetition text and either the following jump or
5792 pop_failure_jump back to this on_failure_jump. */
5793 case on_failure_jump:
5794 EXTRACT_NUMBER_AND_INCR (mcnt, p);
5795 DEBUG_PRINT ("EXECUTING on_failure_jump %d (to %p):\n",
5796 mcnt, p + mcnt);
5797
5798 PUSH_FAILURE_POINT (p -3, d);
5799 break;
5800
5801 /* This operation is used for greedy *.
5802 Compare the beginning of the repeat with what in the
5803 pattern follows its end. If we can establish that there
5804 is nothing that they would both match, i.e., that we
5805 would have to backtrack because of (as in, e.g., `a*a')
5806 then we can use a non-backtracking loop based on
5807 on_failure_keep_string_jump instead of on_failure_jump. */
5808 case on_failure_jump_smart:
5809 EXTRACT_NUMBER_AND_INCR (mcnt, p);
5810 DEBUG_PRINT ("EXECUTING on_failure_jump_smart %d (to %p).\n",
5811 mcnt, p + mcnt);
5812 {
5813 re_char *p1 = p; /* Next operation. */
5814 /* Here, we discard `const', making re_match non-reentrant. */
5815 unsigned char *p2 = (unsigned char*) p + mcnt; /* Jump dest. */
5816 unsigned char *p3 = (unsigned char*) p - 3; /* opcode location. */
5817
5818 p -= 3; /* Reset so that we will re-execute the
5819 instruction once it's been changed. */
5820
5821 EXTRACT_NUMBER (mcnt, p2 - 2);
5822
5823 /* Ensure this is a indeed the trivial kind of loop
5824 we are expecting. */
5825 assert (skip_one_char (p1) == p2 - 3);
5826 assert ((re_opcode_t) p2[-3] == jump && p2 + mcnt == p);
5827 DEBUG_STATEMENT (debug += 2);
5828 if (mutually_exclusive_p (bufp, p1, p2))
5829 {
5830 /* Use a fast `on_failure_keep_string_jump' loop. */
5831 DEBUG_PRINT (" smart exclusive => fast loop.\n");
5832 *p3 = (unsigned char) on_failure_keep_string_jump;
5833 STORE_NUMBER (p2 - 2, mcnt + 3);
5834 }
5835 else
5836 {
5837 /* Default to a safe `on_failure_jump' loop. */
5838 DEBUG_PRINT (" smart default => slow loop.\n");
5839 *p3 = (unsigned char) on_failure_jump;
5840 }
5841 DEBUG_STATEMENT (debug -= 2);
5842 }
5843 break;
5844
5845 /* Unconditionally jump (without popping any failure points). */
5846 case jump:
5847 unconditional_jump:
5848 IMMEDIATE_QUIT_CHECK;
5849 EXTRACT_NUMBER_AND_INCR (mcnt, p); /* Get the amount to jump. */
5850 DEBUG_PRINT ("EXECUTING jump %d ", mcnt);
5851 p += mcnt; /* Do the jump. */
5852 DEBUG_PRINT ("(to %p).\n", p);
5853 break;
5854
5855
5856 /* Have to succeed matching what follows at least n times.
5857 After that, handle like `on_failure_jump'. */
5858 case succeed_n:
5859 /* Signedness doesn't matter since we only compare MCNT to 0. */
5860 EXTRACT_NUMBER (mcnt, p + 2);
5861 DEBUG_PRINT ("EXECUTING succeed_n %d.\n", mcnt);
5862
5863 /* Originally, mcnt is how many times we HAVE to succeed. */
5864 if (mcnt != 0)
5865 {
5866 /* Here, we discard `const', making re_match non-reentrant. */
5867 unsigned char *p2 = (unsigned char*) p + 2; /* counter loc. */
5868 mcnt--;
5869 p += 4;
5870 PUSH_NUMBER (p2, mcnt);
5871 }
5872 else
5873 /* The two bytes encoding mcnt == 0 are two no_op opcodes. */
5874 goto on_failure;
5875 break;
5876
5877 case jump_n:
5878 /* Signedness doesn't matter since we only compare MCNT to 0. */
5879 EXTRACT_NUMBER (mcnt, p + 2);
5880 DEBUG_PRINT ("EXECUTING jump_n %d.\n", mcnt);
5881
5882 /* Originally, this is how many times we CAN jump. */
5883 if (mcnt != 0)
5884 {
5885 /* Here, we discard `const', making re_match non-reentrant. */
5886 unsigned char *p2 = (unsigned char*) p + 2; /* counter loc. */
5887 mcnt--;
5888 PUSH_NUMBER (p2, mcnt);
5889 goto unconditional_jump;
5890 }
5891 /* If don't have to jump any more, skip over the rest of command. */
5892 else
5893 p += 4;
5894 break;
5895
5896 case set_number_at:
5897 {
5898 unsigned char *p2; /* Location of the counter. */
5899 DEBUG_PRINT ("EXECUTING set_number_at.\n");
5900
5901 EXTRACT_NUMBER_AND_INCR (mcnt, p);
5902 /* Here, we discard `const', making re_match non-reentrant. */
5903 p2 = (unsigned char*) p + mcnt;
5904 /* Signedness doesn't matter since we only copy MCNT's bits . */
5905 EXTRACT_NUMBER_AND_INCR (mcnt, p);
5906 DEBUG_PRINT (" Setting %p to %d.\n", p2, mcnt);
5907 PUSH_NUMBER (p2, mcnt);
5908 break;
5909 }
5910
5911 case wordbound:
5912 case notwordbound:
5913 {
5914 boolean not = (re_opcode_t) *(p - 1) == notwordbound;
5915 DEBUG_PRINT ("EXECUTING %swordbound.\n", not ? "not" : "");
5916
5917 /* We SUCCEED (or FAIL) in one of the following cases: */
5918
5919 /* Case 1: D is at the beginning or the end of string. */
5920 if (AT_STRINGS_BEG (d) || AT_STRINGS_END (d))
5921 not = !not;
5922 else
5923 {
5924 /* C1 is the character before D, S1 is the syntax of C1, C2
5925 is the character at D, and S2 is the syntax of C2. */
5926 re_wchar_t c1, c2;
5927 int s1, s2;
5928 int dummy;
5929 #ifdef emacs
5930 ssize_t offset = PTR_TO_OFFSET (d - 1);
5931 ssize_t charpos = SYNTAX_TABLE_BYTE_TO_CHAR (offset);
5932 UPDATE_SYNTAX_TABLE (charpos);
5933 #endif
5934 GET_CHAR_BEFORE_2 (c1, d, string1, end1, string2, end2);
5935 s1 = SYNTAX (c1);
5936 #ifdef emacs
5937 UPDATE_SYNTAX_TABLE_FORWARD (charpos + 1);
5938 #endif
5939 PREFETCH_NOLIMIT ();
5940 GET_CHAR_AFTER (c2, d, dummy);
5941 s2 = SYNTAX (c2);
5942
5943 if (/* Case 2: Only one of S1 and S2 is Sword. */
5944 ((s1 == Sword) != (s2 == Sword))
5945 /* Case 3: Both of S1 and S2 are Sword, and macro
5946 WORD_BOUNDARY_P (C1, C2) returns nonzero. */
5947 || ((s1 == Sword) && WORD_BOUNDARY_P (c1, c2)))
5948 not = !not;
5949 }
5950 if (not)
5951 break;
5952 else
5953 goto fail;
5954 }
5955
5956 case wordbeg:
5957 DEBUG_PRINT ("EXECUTING wordbeg.\n");
5958
5959 /* We FAIL in one of the following cases: */
5960
5961 /* Case 1: D is at the end of string. */
5962 if (AT_STRINGS_END (d))
5963 goto fail;
5964 else
5965 {
5966 /* C1 is the character before D, S1 is the syntax of C1, C2
5967 is the character at D, and S2 is the syntax of C2. */
5968 re_wchar_t c1, c2;
5969 int s1, s2;
5970 int dummy;
5971 #ifdef emacs
5972 ssize_t offset = PTR_TO_OFFSET (d);
5973 ssize_t charpos = SYNTAX_TABLE_BYTE_TO_CHAR (offset);
5974 UPDATE_SYNTAX_TABLE (charpos);
5975 #endif
5976 PREFETCH ();
5977 GET_CHAR_AFTER (c2, d, dummy);
5978 s2 = SYNTAX (c2);
5979
5980 /* Case 2: S2 is not Sword. */
5981 if (s2 != Sword)
5982 goto fail;
5983
5984 /* Case 3: D is not at the beginning of string ... */
5985 if (!AT_STRINGS_BEG (d))
5986 {
5987 GET_CHAR_BEFORE_2 (c1, d, string1, end1, string2, end2);
5988 #ifdef emacs
5989 UPDATE_SYNTAX_TABLE_BACKWARD (charpos - 1);
5990 #endif
5991 s1 = SYNTAX (c1);
5992
5993 /* ... and S1 is Sword, and WORD_BOUNDARY_P (C1, C2)
5994 returns 0. */
5995 if ((s1 == Sword) && !WORD_BOUNDARY_P (c1, c2))
5996 goto fail;
5997 }
5998 }
5999 break;
6000
6001 case wordend:
6002 DEBUG_PRINT ("EXECUTING wordend.\n");
6003
6004 /* We FAIL in one of the following cases: */
6005
6006 /* Case 1: D is at the beginning of string. */
6007 if (AT_STRINGS_BEG (d))
6008 goto fail;
6009 else
6010 {
6011 /* C1 is the character before D, S1 is the syntax of C1, C2
6012 is the character at D, and S2 is the syntax of C2. */
6013 re_wchar_t c1, c2;
6014 int s1, s2;
6015 int dummy;
6016 #ifdef emacs
6017 ssize_t offset = PTR_TO_OFFSET (d) - 1;
6018 ssize_t charpos = SYNTAX_TABLE_BYTE_TO_CHAR (offset);
6019 UPDATE_SYNTAX_TABLE (charpos);
6020 #endif
6021 GET_CHAR_BEFORE_2 (c1, d, string1, end1, string2, end2);
6022 s1 = SYNTAX (c1);
6023
6024 /* Case 2: S1 is not Sword. */
6025 if (s1 != Sword)
6026 goto fail;
6027
6028 /* Case 3: D is not at the end of string ... */
6029 if (!AT_STRINGS_END (d))
6030 {
6031 PREFETCH_NOLIMIT ();
6032 GET_CHAR_AFTER (c2, d, dummy);
6033 #ifdef emacs
6034 UPDATE_SYNTAX_TABLE_FORWARD (charpos);
6035 #endif
6036 s2 = SYNTAX (c2);
6037
6038 /* ... and S2 is Sword, and WORD_BOUNDARY_P (C1, C2)
6039 returns 0. */
6040 if ((s2 == Sword) && !WORD_BOUNDARY_P (c1, c2))
6041 goto fail;
6042 }
6043 }
6044 break;
6045
6046 case symbeg:
6047 DEBUG_PRINT ("EXECUTING symbeg.\n");
6048
6049 /* We FAIL in one of the following cases: */
6050
6051 /* Case 1: D is at the end of string. */
6052 if (AT_STRINGS_END (d))
6053 goto fail;
6054 else
6055 {
6056 /* C1 is the character before D, S1 is the syntax of C1, C2
6057 is the character at D, and S2 is the syntax of C2. */
6058 re_wchar_t c1, c2;
6059 int s1, s2;
6060 #ifdef emacs
6061 ssize_t offset = PTR_TO_OFFSET (d);
6062 ssize_t charpos = SYNTAX_TABLE_BYTE_TO_CHAR (offset);
6063 UPDATE_SYNTAX_TABLE (charpos);
6064 #endif
6065 PREFETCH ();
6066 c2 = RE_STRING_CHAR (d, target_multibyte);
6067 s2 = SYNTAX (c2);
6068
6069 /* Case 2: S2 is neither Sword nor Ssymbol. */
6070 if (s2 != Sword && s2 != Ssymbol)
6071 goto fail;
6072
6073 /* Case 3: D is not at the beginning of string ... */
6074 if (!AT_STRINGS_BEG (d))
6075 {
6076 GET_CHAR_BEFORE_2 (c1, d, string1, end1, string2, end2);
6077 #ifdef emacs
6078 UPDATE_SYNTAX_TABLE_BACKWARD (charpos - 1);
6079 #endif
6080 s1 = SYNTAX (c1);
6081
6082 /* ... and S1 is Sword or Ssymbol. */
6083 if (s1 == Sword || s1 == Ssymbol)
6084 goto fail;
6085 }
6086 }
6087 break;
6088
6089 case symend:
6090 DEBUG_PRINT ("EXECUTING symend.\n");
6091
6092 /* We FAIL in one of the following cases: */
6093
6094 /* Case 1: D is at the beginning of string. */
6095 if (AT_STRINGS_BEG (d))
6096 goto fail;
6097 else
6098 {
6099 /* C1 is the character before D, S1 is the syntax of C1, C2
6100 is the character at D, and S2 is the syntax of C2. */
6101 re_wchar_t c1, c2;
6102 int s1, s2;
6103 #ifdef emacs
6104 ssize_t offset = PTR_TO_OFFSET (d) - 1;
6105 ssize_t charpos = SYNTAX_TABLE_BYTE_TO_CHAR (offset);
6106 UPDATE_SYNTAX_TABLE (charpos);
6107 #endif
6108 GET_CHAR_BEFORE_2 (c1, d, string1, end1, string2, end2);
6109 s1 = SYNTAX (c1);
6110
6111 /* Case 2: S1 is neither Ssymbol nor Sword. */
6112 if (s1 != Sword && s1 != Ssymbol)
6113 goto fail;
6114
6115 /* Case 3: D is not at the end of string ... */
6116 if (!AT_STRINGS_END (d))
6117 {
6118 PREFETCH_NOLIMIT ();
6119 c2 = RE_STRING_CHAR (d, target_multibyte);
6120 #ifdef emacs
6121 UPDATE_SYNTAX_TABLE_FORWARD (charpos + 1);
6122 #endif
6123 s2 = SYNTAX (c2);
6124
6125 /* ... and S2 is Sword or Ssymbol. */
6126 if (s2 == Sword || s2 == Ssymbol)
6127 goto fail;
6128 }
6129 }
6130 break;
6131
6132 case syntaxspec:
6133 case notsyntaxspec:
6134 {
6135 boolean not = (re_opcode_t) *(p - 1) == notsyntaxspec;
6136 mcnt = *p++;
6137 DEBUG_PRINT ("EXECUTING %ssyntaxspec %d.\n", not ? "not" : "",
6138 mcnt);
6139 PREFETCH ();
6140 #ifdef emacs
6141 {
6142 ssize_t offset = PTR_TO_OFFSET (d);
6143 ssize_t pos1 = SYNTAX_TABLE_BYTE_TO_CHAR (offset);
6144 UPDATE_SYNTAX_TABLE (pos1);
6145 }
6146 #endif
6147 {
6148 int len;
6149 re_wchar_t c;
6150
6151 GET_CHAR_AFTER (c, d, len);
6152 if ((SYNTAX (c) != (enum syntaxcode) mcnt) ^ not)
6153 goto fail;
6154 d += len;
6155 }
6156 }
6157 break;
6158
6159 #ifdef emacs
6160 case before_dot:
6161 DEBUG_PRINT ("EXECUTING before_dot.\n");
6162 if (PTR_BYTE_POS (d) >= PT_BYTE)
6163 goto fail;
6164 break;
6165
6166 case at_dot:
6167 DEBUG_PRINT ("EXECUTING at_dot.\n");
6168 if (PTR_BYTE_POS (d) != PT_BYTE)
6169 goto fail;
6170 break;
6171
6172 case after_dot:
6173 DEBUG_PRINT ("EXECUTING after_dot.\n");
6174 if (PTR_BYTE_POS (d) <= PT_BYTE)
6175 goto fail;
6176 break;
6177
6178 case categoryspec:
6179 case notcategoryspec:
6180 {
6181 boolean not = (re_opcode_t) *(p - 1) == notcategoryspec;
6182 mcnt = *p++;
6183 DEBUG_PRINT ("EXECUTING %scategoryspec %d.\n",
6184 not ? "not" : "", mcnt);
6185 PREFETCH ();
6186
6187 {
6188 int len;
6189 re_wchar_t c;
6190 GET_CHAR_AFTER (c, d, len);
6191 if ((!CHAR_HAS_CATEGORY (c, mcnt)) ^ not)
6192 goto fail;
6193 d += len;
6194 }
6195 }
6196 break;
6197
6198 #endif /* emacs */
6199
6200 default:
6201 abort ();
6202 }
6203 continue; /* Successfully executed one pattern command; keep going. */
6204
6205
6206 /* We goto here if a matching operation fails. */
6207 fail:
6208 IMMEDIATE_QUIT_CHECK;
6209 if (!FAIL_STACK_EMPTY ())
6210 {
6211 re_char *str, *pat;
6212 /* A restart point is known. Restore to that state. */
6213 DEBUG_PRINT ("\nFAIL:\n");
6214 POP_FAILURE_POINT (str, pat);
6215 switch (*pat++)
6216 {
6217 case on_failure_keep_string_jump:
6218 assert (str == NULL);
6219 goto continue_failure_jump;
6220
6221 case on_failure_jump_nastyloop:
6222 assert ((re_opcode_t)pat[-2] == no_op);
6223 PUSH_FAILURE_POINT (pat - 2, str);
6224 /* Fallthrough */
6225
6226 case on_failure_jump_loop:
6227 case on_failure_jump:
6228 case succeed_n:
6229 d = str;
6230 continue_failure_jump:
6231 EXTRACT_NUMBER_AND_INCR (mcnt, pat);
6232 p = pat + mcnt;
6233 break;
6234
6235 case no_op:
6236 /* A special frame used for nastyloops. */
6237 goto fail;
6238
6239 default:
6240 abort ();
6241 }
6242
6243 assert (p >= bufp->buffer && p <= pend);
6244
6245 if (d >= string1 && d <= end1)
6246 dend = end_match_1;
6247 }
6248 else
6249 break; /* Matching at this starting point really fails. */
6250 } /* for (;;) */
6251
6252 if (best_regs_set)
6253 goto restore_best_regs;
6254
6255 FREE_VARIABLES ();
6256
6257 return -1; /* Failure to match. */
6258 }
6259 \f
6260 /* Subroutine definitions for re_match_2. */
6261
6262 /* Return zero if TRANSLATE[S1] and TRANSLATE[S2] are identical for LEN
6263 bytes; nonzero otherwise. */
6264
6265 static int
6266 bcmp_translate (const re_char *s1, const re_char *s2, register ssize_t len,
6267 RE_TRANSLATE_TYPE translate, const int target_multibyte)
6268 {
6269 register re_char *p1 = s1, *p2 = s2;
6270 re_char *p1_end = s1 + len;
6271 re_char *p2_end = s2 + len;
6272
6273 /* FIXME: Checking both p1 and p2 presumes that the two strings might have
6274 different lengths, but relying on a single `len' would break this. -sm */
6275 while (p1 < p1_end && p2 < p2_end)
6276 {
6277 int p1_charlen, p2_charlen;
6278 re_wchar_t p1_ch, p2_ch;
6279
6280 GET_CHAR_AFTER (p1_ch, p1, p1_charlen);
6281 GET_CHAR_AFTER (p2_ch, p2, p2_charlen);
6282
6283 if (RE_TRANSLATE (translate, p1_ch)
6284 != RE_TRANSLATE (translate, p2_ch))
6285 return 1;
6286
6287 p1 += p1_charlen, p2 += p2_charlen;
6288 }
6289
6290 if (p1 != p1_end || p2 != p2_end)
6291 return 1;
6292
6293 return 0;
6294 }
6295 \f
6296 /* Entry points for GNU code. */
6297
6298 /* re_compile_pattern is the GNU regular expression compiler: it
6299 compiles PATTERN (of length SIZE) and puts the result in BUFP.
6300 Returns 0 if the pattern was valid, otherwise an error string.
6301
6302 Assumes the `allocated' (and perhaps `buffer') and `translate' fields
6303 are set in BUFP on entry.
6304
6305 We call regex_compile to do the actual compilation. */
6306
6307 const char *
6308 re_compile_pattern (const char *pattern, size_t length,
6309 struct re_pattern_buffer *bufp)
6310 {
6311 reg_errcode_t ret;
6312
6313 /* GNU code is written to assume at least RE_NREGS registers will be set
6314 (and at least one extra will be -1). */
6315 bufp->regs_allocated = REGS_UNALLOCATED;
6316
6317 /* And GNU code determines whether or not to get register information
6318 by passing null for the REGS argument to re_match, etc., not by
6319 setting no_sub. */
6320 bufp->no_sub = 0;
6321
6322 ret = regex_compile ((re_char*) pattern, length, re_syntax_options, bufp);
6323
6324 if (!ret)
6325 return NULL;
6326 return gettext (re_error_msgid[(int) ret]);
6327 }
6328 WEAK_ALIAS (__re_compile_pattern, re_compile_pattern)
6329 \f
6330 /* Entry points compatible with 4.2 BSD regex library. We don't define
6331 them unless specifically requested. */
6332
6333 #if defined _REGEX_RE_COMP || defined _LIBC
6334
6335 /* BSD has one and only one pattern buffer. */
6336 static struct re_pattern_buffer re_comp_buf;
6337
6338 char *
6339 # ifdef _LIBC
6340 /* Make these definitions weak in libc, so POSIX programs can redefine
6341 these names if they don't use our functions, and still use
6342 regcomp/regexec below without link errors. */
6343 weak_function
6344 # endif
6345 re_comp (const char *s)
6346 {
6347 reg_errcode_t ret;
6348
6349 if (!s)
6350 {
6351 if (!re_comp_buf.buffer)
6352 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6353 return (char *) gettext ("No previous regular expression");
6354 return 0;
6355 }
6356
6357 if (!re_comp_buf.buffer)
6358 {
6359 re_comp_buf.buffer = malloc (200);
6360 if (re_comp_buf.buffer == NULL)
6361 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6362 return (char *) gettext (re_error_msgid[(int) REG_ESPACE]);
6363 re_comp_buf.allocated = 200;
6364
6365 re_comp_buf.fastmap = malloc (1 << BYTEWIDTH);
6366 if (re_comp_buf.fastmap == NULL)
6367 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6368 return (char *) gettext (re_error_msgid[(int) REG_ESPACE]);
6369 }
6370
6371 /* Since `re_exec' always passes NULL for the `regs' argument, we
6372 don't need to initialize the pattern buffer fields which affect it. */
6373
6374 ret = regex_compile (s, strlen (s), re_syntax_options, &re_comp_buf);
6375
6376 if (!ret)
6377 return NULL;
6378
6379 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6380 return (char *) gettext (re_error_msgid[(int) ret]);
6381 }
6382
6383
6384 int
6385 # ifdef _LIBC
6386 weak_function
6387 # endif
6388 re_exec (const char *s)
6389 {
6390 const size_t len = strlen (s);
6391 return (re_search (&re_comp_buf, s, len, 0, len, (struct re_registers *) 0)
6392 >= 0);
6393 }
6394 #endif /* _REGEX_RE_COMP */
6395 \f
6396 /* POSIX.2 functions. Don't define these for Emacs. */
6397
6398 #ifndef emacs
6399
6400 /* regcomp takes a regular expression as a string and compiles it.
6401
6402 PREG is a regex_t *. We do not expect any fields to be initialized,
6403 since POSIX says we shouldn't. Thus, we set
6404
6405 `buffer' to the compiled pattern;
6406 `used' to the length of the compiled pattern;
6407 `syntax' to RE_SYNTAX_POSIX_EXTENDED if the
6408 REG_EXTENDED bit in CFLAGS is set; otherwise, to
6409 RE_SYNTAX_POSIX_BASIC;
6410 `fastmap' to an allocated space for the fastmap;
6411 `fastmap_accurate' to zero;
6412 `re_nsub' to the number of subexpressions in PATTERN.
6413
6414 PATTERN is the address of the pattern string.
6415
6416 CFLAGS is a series of bits which affect compilation.
6417
6418 If REG_EXTENDED is set, we use POSIX extended syntax; otherwise, we
6419 use POSIX basic syntax.
6420
6421 If REG_NEWLINE is set, then . and [^...] don't match newline.
6422 Also, regexec will try a match beginning after every newline.
6423
6424 If REG_ICASE is set, then we considers upper- and lowercase
6425 versions of letters to be equivalent when matching.
6426
6427 If REG_NOSUB is set, then when PREG is passed to regexec, that
6428 routine will report only success or failure, and nothing about the
6429 registers.
6430
6431 It returns 0 if it succeeds, nonzero if it doesn't. (See regex.h for
6432 the return codes and their meanings.) */
6433
6434 reg_errcode_t
6435 regcomp (regex_t *__restrict preg, const char *__restrict pattern,
6436 int cflags)
6437 {
6438 reg_errcode_t ret;
6439 reg_syntax_t syntax
6440 = (cflags & REG_EXTENDED) ?
6441 RE_SYNTAX_POSIX_EXTENDED : RE_SYNTAX_POSIX_BASIC;
6442
6443 /* regex_compile will allocate the space for the compiled pattern. */
6444 preg->buffer = 0;
6445 preg->allocated = 0;
6446 preg->used = 0;
6447
6448 /* Try to allocate space for the fastmap. */
6449 preg->fastmap = malloc (1 << BYTEWIDTH);
6450
6451 if (cflags & REG_ICASE)
6452 {
6453 unsigned i;
6454
6455 preg->translate = malloc (CHAR_SET_SIZE * sizeof *preg->translate);
6456 if (preg->translate == NULL)
6457 return (int) REG_ESPACE;
6458
6459 /* Map uppercase characters to corresponding lowercase ones. */
6460 for (i = 0; i < CHAR_SET_SIZE; i++)
6461 preg->translate[i] = ISUPPER (i) ? TOLOWER (i) : i;
6462 }
6463 else
6464 preg->translate = NULL;
6465
6466 /* If REG_NEWLINE is set, newlines are treated differently. */
6467 if (cflags & REG_NEWLINE)
6468 { /* REG_NEWLINE implies neither . nor [^...] match newline. */
6469 syntax &= ~RE_DOT_NEWLINE;
6470 syntax |= RE_HAT_LISTS_NOT_NEWLINE;
6471 }
6472 else
6473 syntax |= RE_NO_NEWLINE_ANCHOR;
6474
6475 preg->no_sub = !!(cflags & REG_NOSUB);
6476
6477 /* POSIX says a null character in the pattern terminates it, so we
6478 can use strlen here in compiling the pattern. */
6479 ret = regex_compile ((re_char*) pattern, strlen (pattern), syntax, preg);
6480
6481 /* POSIX doesn't distinguish between an unmatched open-group and an
6482 unmatched close-group: both are REG_EPAREN. */
6483 if (ret == REG_ERPAREN)
6484 ret = REG_EPAREN;
6485
6486 if (ret == REG_NOERROR && preg->fastmap)
6487 { /* Compute the fastmap now, since regexec cannot modify the pattern
6488 buffer. */
6489 re_compile_fastmap (preg);
6490 if (preg->can_be_null)
6491 { /* The fastmap can't be used anyway. */
6492 free (preg->fastmap);
6493 preg->fastmap = NULL;
6494 }
6495 }
6496 return ret;
6497 }
6498 WEAK_ALIAS (__regcomp, regcomp)
6499
6500
6501 /* regexec searches for a given pattern, specified by PREG, in the
6502 string STRING.
6503
6504 If NMATCH is zero or REG_NOSUB was set in the cflags argument to
6505 `regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at
6506 least NMATCH elements, and we set them to the offsets of the
6507 corresponding matched substrings.
6508
6509 EFLAGS specifies `execution flags' which affect matching: if
6510 REG_NOTBOL is set, then ^ does not match at the beginning of the
6511 string; if REG_NOTEOL is set, then $ does not match at the end.
6512
6513 We return 0 if we find a match and REG_NOMATCH if not. */
6514
6515 reg_errcode_t
6516 regexec (const regex_t *__restrict preg, const char *__restrict string,
6517 size_t nmatch, regmatch_t pmatch[__restrict_arr], int eflags)
6518 {
6519 regoff_t ret;
6520 struct re_registers regs;
6521 regex_t private_preg;
6522 size_t len = strlen (string);
6523 boolean want_reg_info = !preg->no_sub && nmatch > 0 && pmatch;
6524
6525 private_preg = *preg;
6526
6527 private_preg.not_bol = !!(eflags & REG_NOTBOL);
6528 private_preg.not_eol = !!(eflags & REG_NOTEOL);
6529
6530 /* The user has told us exactly how many registers to return
6531 information about, via `nmatch'. We have to pass that on to the
6532 matching routines. */
6533 private_preg.regs_allocated = REGS_FIXED;
6534
6535 if (want_reg_info)
6536 {
6537 regs.num_regs = nmatch;
6538 regs.start = TALLOC (nmatch * 2, regoff_t);
6539 if (regs.start == NULL)
6540 return REG_NOMATCH;
6541 regs.end = regs.start + nmatch;
6542 }
6543
6544 /* Instead of using not_eol to implement REG_NOTEOL, we could simply
6545 pass (&private_preg, string, len + 1, 0, len, ...) pretending the string
6546 was a little bit longer but still only matching the real part.
6547 This works because the `endline' will check for a '\n' and will find a
6548 '\0', correctly deciding that this is not the end of a line.
6549 But it doesn't work out so nicely for REG_NOTBOL, since we don't have
6550 a convenient '\0' there. For all we know, the string could be preceded
6551 by '\n' which would throw things off. */
6552
6553 /* Perform the searching operation. */
6554 ret = re_search (&private_preg, string, len,
6555 /* start: */ 0, /* range: */ len,
6556 want_reg_info ? &regs : (struct re_registers *) 0);
6557
6558 /* Copy the register information to the POSIX structure. */
6559 if (want_reg_info)
6560 {
6561 if (ret >= 0)
6562 {
6563 unsigned r;
6564
6565 for (r = 0; r < nmatch; r++)
6566 {
6567 pmatch[r].rm_so = regs.start[r];
6568 pmatch[r].rm_eo = regs.end[r];
6569 }
6570 }
6571
6572 /* If we needed the temporary register info, free the space now. */
6573 free (regs.start);
6574 }
6575
6576 /* We want zero return to mean success, unlike `re_search'. */
6577 return ret >= 0 ? REG_NOERROR : REG_NOMATCH;
6578 }
6579 WEAK_ALIAS (__regexec, regexec)
6580
6581
6582 /* Returns a message corresponding to an error code, ERR_CODE, returned
6583 from either regcomp or regexec. We don't use PREG here.
6584
6585 ERR_CODE was previously called ERRCODE, but that name causes an
6586 error with msvc8 compiler. */
6587
6588 size_t
6589 regerror (int err_code, const regex_t *preg, char *errbuf, size_t errbuf_size)
6590 {
6591 const char *msg;
6592 size_t msg_size;
6593
6594 if (err_code < 0
6595 || err_code >= (sizeof (re_error_msgid) / sizeof (re_error_msgid[0])))
6596 /* Only error codes returned by the rest of the code should be passed
6597 to this routine. If we are given anything else, or if other regex
6598 code generates an invalid error code, then the program has a bug.
6599 Dump core so we can fix it. */
6600 abort ();
6601
6602 msg = gettext (re_error_msgid[err_code]);
6603
6604 msg_size = strlen (msg) + 1; /* Includes the null. */
6605
6606 if (errbuf_size != 0)
6607 {
6608 if (msg_size > errbuf_size)
6609 {
6610 memcpy (errbuf, msg, errbuf_size - 1);
6611 errbuf[errbuf_size - 1] = 0;
6612 }
6613 else
6614 strcpy (errbuf, msg);
6615 }
6616
6617 return msg_size;
6618 }
6619 WEAK_ALIAS (__regerror, regerror)
6620
6621
6622 /* Free dynamically allocated space used by PREG. */
6623
6624 void
6625 regfree (regex_t *preg)
6626 {
6627 free (preg->buffer);
6628 preg->buffer = NULL;
6629
6630 preg->allocated = 0;
6631 preg->used = 0;
6632
6633 free (preg->fastmap);
6634 preg->fastmap = NULL;
6635 preg->fastmap_accurate = 0;
6636
6637 free (preg->translate);
6638 preg->translate = NULL;
6639 }
6640 WEAK_ALIAS (__regfree, regfree)
6641
6642 #endif /* not emacs */
Emacs repository mirror