« back to all changes in this revision
Viewing changes to lib/regex.c
- browse files at revision 1
- compare with another revision
- download diff
- download tarball
- view history from revision 1
- Committer: Bazaar Package Importer
- Author(s): Santiago Vila
- Date: 2005-02-15 22:45:18 UTC
- Revision ID: james.westby@ubuntu.com-20050215224518-dw9ti3me00twpcmt
Tags: upstream-2.8.1
Import upstream version 2.8.1
- files added:
- config
- lib
- lib/posix
- m4
- man
- ms
- po
- src
added
removed
lib/regex.c
1
/* Extended regular expression matching and search library,
2
version 0.12.
3
(Implements POSIX draft P1003.2/D11.2, except for some of the
4
internationalization features.)
5
Copyright (C) 1993-1999, 2000, 2001 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 2, 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, write to the Free Software Foundation,
19
Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. */
20
21
/* AIX requires this to be the first thing in the file. */
22
#if defined _AIX && !defined REGEX_MALLOC
23
#pragma alloca
24
#endif
25
26
#undef _GNU_SOURCE
27
#define _GNU_SOURCE
28
29
#ifdef HAVE_CONFIG_H
30
# include <config.h>
31
#endif
32
33
#ifndef PARAMS
34
# if defined __GNUC__ || (defined __STDC__ && __STDC__)
35
# define PARAMS(args) args
36
# else
37
# define PARAMS(args) ()
38
# endif /* GCC. */
39
#endif /* Not PARAMS. */
40
41
#ifndef INSIDE_RECURSION
42
43
# if defined STDC_HEADERS && !defined emacs
44
# include <stddef.h>
45
# else
46
/* We need this for `regex.h', and perhaps for the Emacs include files. */
47
# include <sys/types.h>
48
# endif
49
50
# define WIDE_CHAR_SUPPORT (HAVE_WCTYPE_H && HAVE_WCHAR_H && HAVE_BTOWC)
51
52
/* For platform which support the ISO C amendement 1 functionality we
53
support user defined character classes. */
54
# if defined _LIBC || WIDE_CHAR_SUPPORT
55
/* Solaris 2.5 has a bug: <wchar.h> must be included before <wctype.h>. */
56
# include <wchar.h>
57
# include <wctype.h>
58
# endif
59
60
# ifdef _LIBC
61
/* We have to keep the namespace clean. */
62
# define regfree(preg) __regfree (preg)
63
# define regexec(pr, st, nm, pm, ef) __regexec (pr, st, nm, pm, ef)
64
# define regcomp(preg, pattern, cflags) __regcomp (preg, pattern, cflags)
65
# define regerror(errcode, preg, errbuf, errbuf_size) \
66
__regerror(errcode, preg, errbuf, errbuf_size)
67
# define re_set_registers(bu, re, nu, st, en) \
68
__re_set_registers (bu, re, nu, st, en)
69
# define re_match_2(bufp, string1, size1, string2, size2, pos, regs, stop) \
70
__re_match_2 (bufp, string1, size1, string2, size2, pos, regs, stop)
71
# define re_match(bufp, string, size, pos, regs) \
72
__re_match (bufp, string, size, pos, regs)
73
# define re_search(bufp, string, size, startpos, range, regs) \
74
__re_search (bufp, string, size, startpos, range, regs)
75
# define re_compile_pattern(pattern, length, bufp) \
76
__re_compile_pattern (pattern, length, bufp)
77
# define re_set_syntax(syntax) __re_set_syntax (syntax)
78
# define re_search_2(bufp, st1, s1, st2, s2, startpos, range, regs, stop) \
79
__re_search_2 (bufp, st1, s1, st2, s2, startpos, range, regs, stop)
80
# define re_compile_fastmap(bufp) __re_compile_fastmap (bufp)
81
82
# define btowc __btowc
83
# define iswctype __iswctype
84
# define mbrtowc __mbrtowc
85
# define wcslen __wcslen
86
# define wcscoll __wcscoll
87
# define wcrtomb __wcrtomb
88
89
/* We are also using some library internals. */
90
# include <locale/localeinfo.h>
91
# include <locale/elem-hash.h>
92
# include <langinfo.h>
93
# include <locale/coll-lookup.h>
94
# endif
95
96
/* This is for other GNU distributions with internationalized messages. */
97
# if (HAVE_LIBINTL_H && ENABLE_NLS) || defined _LIBC
98
# include <libintl.h>
99
# ifdef _LIBC
100
# undef gettext
101
# define gettext(msgid) __dcgettext ("libc", msgid, LC_MESSAGES)
102
# endif
103
# else
104
# define gettext(msgid) (msgid)
105
# endif
106
107
# ifndef gettext_noop
108
/* This define is so xgettext can find the internationalizable
109
strings. */
110
# define gettext_noop(String) String
111
# endif
112
113
/* Support for bounded pointers. */
114
# if !defined _LIBC && !defined __BOUNDED_POINTERS__
115
# define __bounded /* nothing */
116
# define __unbounded /* nothing */
117
# define __ptrvalue /* nothing */
118
# endif
119
120
/* The `emacs' switch turns on certain matching commands
121
that make sense only in Emacs. */
122
# ifdef emacs
123
124
# include "lisp.h"
125
# include "buffer.h"
126
# include "syntax.h"
127
128
# else /* not emacs */
129
130
/* If we are not linking with Emacs proper,
131
we can't use the relocating allocator
132
even if config.h says that we can. */
133
# undef REL_ALLOC
134
135
# if defined STDC_HEADERS || defined _LIBC
136
# include <stdlib.h>
137
# else
138
char *malloc ();
139
char *realloc ();
140
# endif
141
142
/* When used in Emacs's lib-src, we need to get bzero and bcopy somehow.
143
If nothing else has been done, use the method below. */
144
# ifdef INHIBIT_STRING_HEADER
145
# if !(defined HAVE_BZERO && defined HAVE_BCOPY)
146
# if !defined bzero && !defined bcopy
147
# undef INHIBIT_STRING_HEADER
148
# endif
149
# endif
150
# endif
151
152
/* This is the normal way of making sure we have a bcopy and a bzero.
153
This is used in most programs--a few other programs avoid this
154
by defining INHIBIT_STRING_HEADER. */
155
# ifndef INHIBIT_STRING_HEADER
156
# if defined HAVE_STRING_H || defined STDC_HEADERS || defined _LIBC
157
# include <string.h>
158
# ifndef bzero
159
# ifndef _LIBC
160
# define bzero(s, n) (memset (s, '\0', n), (s))
161
# else
162
# define bzero(s, n) __bzero (s, n)
163
# endif
164
# endif
165
# else
166
# include <strings.h>
167
# ifndef memcmp
168
# define memcmp(s1, s2, n) bcmp (s1, s2, n)
169
# endif
170
# ifndef memcpy
171
# define memcpy(d, s, n) (bcopy (s, d, n), (d))
172
# endif
173
# endif
174
# endif
175
176
/* Define the syntax stuff for \<, \>, etc. */
177
178
/* This must be nonzero for the wordchar and notwordchar pattern
179
commands in re_match_2. */
180
# ifndef Sword
181
# define Sword 1
182
# endif
183
184
# ifdef SWITCH_ENUM_BUG
185
# define SWITCH_ENUM_CAST(x) ((int)(x))
186
# else
187
# define SWITCH_ENUM_CAST(x) (x)
188
# endif
189
190
# endif /* not emacs */
191
192
# if defined _LIBC || HAVE_LIMITS_H
193
# include <limits.h>
194
# endif
195
196
# ifndef MB_LEN_MAX
197
# define MB_LEN_MAX 1
198
# endif
199
200
/* Get the interface, including the syntax bits. */
201
# include <regex.h>
202
203
/* isalpha etc. are used for the character classes. */
204
# include <ctype.h>
205
206
/* Jim Meyering writes:
207
208
"... Some ctype macros are valid only for character codes that
209
isascii says are ASCII (SGI's IRIX-4.0.5 is one such system --when
210
using /bin/cc or gcc but without giving an ansi option). So, all
211
ctype uses should be through macros like ISPRINT... If
212
STDC_HEADERS is defined, then autoconf has verified that the ctype
213
macros don't need to be guarded with references to isascii. ...
214
Defining isascii to 1 should let any compiler worth its salt
215
eliminate the && through constant folding."
216
Solaris defines some of these symbols so we must undefine them first. */
217
218
# if defined STDC_HEADERS || (!defined isascii && !defined HAVE_ISASCII)
219
# define IN_CTYPE_DOMAIN(c) 1
220
# else
221
# define IN_CTYPE_DOMAIN(c) isascii(c)
222
# endif
223
224
# ifdef isblank
225
# define ISBLANK(c) (IN_CTYPE_DOMAIN (c) && isblank (c))
226
# else
227
# define ISBLANK(c) ((c) == ' ' || (c) == '\t')
228
# endif
229
# ifdef isgraph
230
# define ISGRAPH(c) (IN_CTYPE_DOMAIN (c) && isgraph (c))
231
# else
232
# define ISGRAPH(c) (IN_CTYPE_DOMAIN (c) && isprint (c) && !isspace (c))
233
# endif
234
235
# undef ISPRINT
236
# define ISPRINT(c) (IN_CTYPE_DOMAIN (c) && isprint (c))
237
# define ISDIGIT(c) (IN_CTYPE_DOMAIN (c) && isdigit (c))
238
# define ISALNUM(c) (IN_CTYPE_DOMAIN (c) && isalnum (c))
239
# define ISALPHA(c) (IN_CTYPE_DOMAIN (c) && isalpha (c))
240
# define ISCNTRL(c) (IN_CTYPE_DOMAIN (c) && iscntrl (c))
241
# define ISLOWER(c) (IN_CTYPE_DOMAIN (c) && islower (c))
242
# define ISPUNCT(c) (IN_CTYPE_DOMAIN (c) && ispunct (c))
243
# define ISSPACE(c) (IN_CTYPE_DOMAIN (c) && isspace (c))
244
# define ISUPPER(c) (IN_CTYPE_DOMAIN (c) && isupper (c))
245
# define ISXDIGIT(c) (IN_CTYPE_DOMAIN (c) && isxdigit (c))
246
247
# ifdef _tolower
248
# define TOLOWER(c) _tolower(c)
249
# else
250
# define TOLOWER(c) tolower(c)
251
# endif
252
253
# ifndef NULL
254
# define NULL (void *)0
255
# endif
256
257
/* We remove any previous definition of `SIGN_EXTEND_CHAR',
258
since ours (we hope) works properly with all combinations of
259
machines, compilers, `char' and `unsigned char' argument types.
260
(Per Bothner suggested the basic approach.) */
261
# undef SIGN_EXTEND_CHAR
262
# if __STDC__
263
# define SIGN_EXTEND_CHAR(c) ((signed char) (c))
264
# else /* not __STDC__ */
265
/* As in Harbison and Steele. */
266
# define SIGN_EXTEND_CHAR(c) ((((unsigned char) (c)) ^ 128) - 128)
267
# endif
268
269
# ifndef emacs
270
/* How many characters in the character set. */
271
# define CHAR_SET_SIZE 256
272
273
# ifdef SYNTAX_TABLE
274
275
extern char *re_syntax_table;
276
277
# else /* not SYNTAX_TABLE */
278
279
static char re_syntax_table[CHAR_SET_SIZE];
280
281
static void init_syntax_once PARAMS ((void));
282
283
static void
284
init_syntax_once ()
285
{
286
register int c;
287
static int done = 0;
288
289
if (done)
290
return;
291
bzero (re_syntax_table, sizeof re_syntax_table);
292
293
for (c = 0; c < CHAR_SET_SIZE; ++c)
294
if (ISALNUM (c))
295
re_syntax_table[c] = Sword;
296
297
re_syntax_table['_'] = Sword;
298
299
done = 1;
300
}
301
302
# endif /* not SYNTAX_TABLE */
303
304
# define SYNTAX(c) re_syntax_table[(unsigned char) (c)]
305
306
# endif /* emacs */
307
308
/* Should we use malloc or alloca? If REGEX_MALLOC is not defined, we
309
use `alloca' instead of `malloc'. This is because using malloc in
310
re_search* or re_match* could cause memory leaks when C-g is used in
311
Emacs; also, malloc is slower and causes storage fragmentation. On
312
the other hand, malloc is more portable, and easier to debug.
313
314
Because we sometimes use alloca, some routines have to be macros,
315
not functions -- `alloca'-allocated space disappears at the end of the
316
function it is called in. */
317
318
# ifdef REGEX_MALLOC
319
320
# define REGEX_ALLOCATE malloc
321
# define REGEX_REALLOCATE(source, osize, nsize) realloc (source, nsize)
322
# define REGEX_FREE free
323
324
# else /* not REGEX_MALLOC */
325
326
/* Emacs already defines alloca, sometimes. */
327
# ifndef alloca
328
329
/* Make alloca work the best possible way. */
330
# ifdef __GNUC__
331
# define alloca __builtin_alloca
332
# else /* not __GNUC__ */
333
# if HAVE_ALLOCA_H
334
# include <alloca.h>
335
# endif /* HAVE_ALLOCA_H */
336
# endif /* not __GNUC__ */
337
338
# endif /* not alloca */
339
340
# define REGEX_ALLOCATE alloca
341
342
/* Assumes a `char *destination' variable. */
343
# define REGEX_REALLOCATE(source, osize, nsize) \
344
(destination = (char *) alloca (nsize), \
345
memcpy (destination, source, osize))
346
347
/* No need to do anything to free, after alloca. */
348
# define REGEX_FREE(arg) ((void)0) /* Do nothing! But inhibit gcc warning. */
349
350
# endif /* not REGEX_MALLOC */
351
352
/* Define how to allocate the failure stack. */
353
354
# if defined REL_ALLOC && defined REGEX_MALLOC
355
356
# define REGEX_ALLOCATE_STACK(size) \
357
r_alloc (&failure_stack_ptr, (size))
358
# define REGEX_REALLOCATE_STACK(source, osize, nsize) \
359
r_re_alloc (&failure_stack_ptr, (nsize))
360
# define REGEX_FREE_STACK(ptr) \
361
r_alloc_free (&failure_stack_ptr)
362
363
# else /* not using relocating allocator */
364
365
# ifdef REGEX_MALLOC
366
367
# define REGEX_ALLOCATE_STACK malloc
368
# define REGEX_REALLOCATE_STACK(source, osize, nsize) realloc (source, nsize)
369
# define REGEX_FREE_STACK free
370
371
# else /* not REGEX_MALLOC */
372
373
# define REGEX_ALLOCATE_STACK alloca
374
375
# define REGEX_REALLOCATE_STACK(source, osize, nsize) \
376
REGEX_REALLOCATE (source, osize, nsize)
377
/* No need to explicitly free anything. */
378
# define REGEX_FREE_STACK(arg)
379
380
# endif /* not REGEX_MALLOC */
381
# endif /* not using relocating allocator */
382
383
384
/* True if `size1' is non-NULL and PTR is pointing anywhere inside
385
`string1' or just past its end. This works if PTR is NULL, which is
386
a good thing. */
387
# define FIRST_STRING_P(ptr) \
388
(size1 && string1 <= (ptr) && (ptr) <= string1 + size1)
389
390
/* (Re)Allocate N items of type T using malloc, or fail. */
391
# define TALLOC(n, t) ((t *) malloc ((n) * sizeof (t)))
392
# define RETALLOC(addr, n, t) ((addr) = (t *) realloc (addr, (n) * sizeof (t)))
393
# define RETALLOC_IF(addr, n, t) \
394
if (addr) RETALLOC((addr), (n), t); else (addr) = TALLOC ((n), t)
395
# define REGEX_TALLOC(n, t) ((t *) REGEX_ALLOCATE ((n) * sizeof (t)))
396
397
# define BYTEWIDTH 8 /* In bits. */
398
399
# define STREQ(s1, s2) ((strcmp (s1, s2) == 0))
400
401
# undef MAX
402
# undef MIN
403
# define MAX(a, b) ((a) > (b) ? (a) : (b))
404
# define MIN(a, b) ((a) < (b) ? (a) : (b))
405
406
typedef char boolean;
407
# define false 0
408
# define true 1
409
410
static reg_errcode_t byte_regex_compile _RE_ARGS ((const char *pattern, size_t size,
411
reg_syntax_t syntax,
412
struct re_pattern_buffer *bufp));
413
414
static int byte_re_match_2_internal PARAMS ((struct re_pattern_buffer *bufp,
415
const char *string1, int size1,
416
const char *string2, int size2,
417
int pos,
418
struct re_registers *regs,
419
int stop));
420
static int byte_re_search_2 PARAMS ((struct re_pattern_buffer *bufp,
421
const char *string1, int size1,
422
const char *string2, int size2,
423
int startpos, int range,
424
struct re_registers *regs, int stop));
425
static int byte_re_compile_fastmap PARAMS ((struct re_pattern_buffer *bufp));
426
427
#ifdef MBS_SUPPORT
428
static reg_errcode_t wcs_regex_compile _RE_ARGS ((const char *pattern, size_t size,
429
reg_syntax_t syntax,
430
struct re_pattern_buffer *bufp));
431
432
433
static int wcs_re_match_2_internal PARAMS ((struct re_pattern_buffer *bufp,
434
const char *cstring1, int csize1,
435
const char *cstring2, int csize2,
436
int pos,
437
struct re_registers *regs,
438
int stop,
439
wchar_t *string1, int size1,
440
wchar_t *string2, int size2,
441
int *mbs_offset1, int *mbs_offset2));
442
static int wcs_re_search_2 PARAMS ((struct re_pattern_buffer *bufp,
443
const char *string1, int size1,
444
const char *string2, int size2,
445
int startpos, int range,
446
struct re_registers *regs, int stop));
447
static int wcs_re_compile_fastmap PARAMS ((struct re_pattern_buffer *bufp));
448
#endif
449
450
/* These are the command codes that appear in compiled regular
451
expressions. Some opcodes are followed by argument bytes. A
452
command code can specify any interpretation whatsoever for its
453
arguments. Zero bytes may appear in the compiled regular expression. */
454
455
typedef enum
456
{
457
no_op = 0,
458
459
/* Succeed right away--no more backtracking. */
460
succeed,
461
462
/* Followed by one byte giving n, then by n literal bytes. */
463
exactn,
464
465
# ifdef MBS_SUPPORT
466
/* Same as exactn, but contains binary data. */
467
exactn_bin,
468
# endif
469
470
/* Matches any (more or less) character. */
471
anychar,
472
473
/* Matches any one char belonging to specified set. First
474
following byte is number of bitmap bytes. Then come bytes
475
for a bitmap saying which chars are in. Bits in each byte
476
are ordered low-bit-first. A character is in the set if its
477
bit is 1. A character too large to have a bit in the map is
478
automatically not in the set. */
479
/* ifdef MBS_SUPPORT, following element is length of character
480
classes, length of collating symbols, length of equivalence
481
classes, length of character ranges, and length of characters.
482
Next, character class element, collating symbols elements,
483
equivalence class elements, range elements, and character
484
elements follow.
485
See regex_compile function. */
486
charset,
487
488
/* Same parameters as charset, but match any character that is
489
not one of those specified. */
490
charset_not,
491
492
/* Start remembering the text that is matched, for storing in a
493
register. Followed by one byte with the register number, in
494
the range 0 to one less than the pattern buffer's re_nsub
495
field. Then followed by one byte with the number of groups
496
inner to this one. (This last has to be part of the
497
start_memory only because we need it in the on_failure_jump
498
of re_match_2.) */
499
start_memory,
500
501
/* Stop remembering the text that is matched and store it in a
502
memory register. Followed by one byte with the register
503
number, in the range 0 to one less than `re_nsub' in the
504
pattern buffer, and one byte with the number of inner groups,
505
just like `start_memory'. (We need the number of inner
506
groups here because we don't have any easy way of finding the
507
corresponding start_memory when we're at a stop_memory.) */
508
stop_memory,
509
510
/* Match a duplicate of something remembered. Followed by one
511
byte containing the register number. */
512
duplicate,
513
514
/* Fail unless at beginning of line. */
515
begline,
516
517
/* Fail unless at end of line. */
518
endline,
519
520
/* Succeeds if at beginning of buffer (if emacs) or at beginning
521
of string to be matched (if not). */
522
begbuf,
523
524
/* Analogously, for end of buffer/string. */
525
endbuf,
526
527
/* Followed by two byte relative address to which to jump. */
528
jump,
529
530
/* Same as jump, but marks the end of an alternative. */
531
jump_past_alt,
532
533
/* Followed by two-byte relative address of place to resume at
534
in case of failure. */
535
/* ifdef MBS_SUPPORT, the size of address is 1. */
536
on_failure_jump,
537
538
/* Like on_failure_jump, but pushes a placeholder instead of the
539
current string position when executed. */
540
on_failure_keep_string_jump,
541
542
/* Throw away latest failure point and then jump to following
543
two-byte relative address. */
544
/* ifdef MBS_SUPPORT, the size of address is 1. */
545
pop_failure_jump,
546
547
/* Change to pop_failure_jump if know won't have to backtrack to
548
match; otherwise change to jump. This is used to jump
549
back to the beginning of a repeat. If what follows this jump
550
clearly won't match what the repeat does, such that we can be
551
sure that there is no use backtracking out of repetitions
552
already matched, then we change it to a pop_failure_jump.
553
Followed by two-byte address. */
554
/* ifdef MBS_SUPPORT, the size of address is 1. */
555
maybe_pop_jump,
556
557
/* Jump to following two-byte address, and push a dummy failure
558
point. This failure point will be thrown away if an attempt
559
is made to use it for a failure. A `+' construct makes this
560
before the first repeat. Also used as an intermediary kind
561
of jump when compiling an alternative. */
562
/* ifdef MBS_SUPPORT, the size of address is 1. */
563
dummy_failure_jump,
564
565
/* Push a dummy failure point and continue. Used at the end of
566
alternatives. */
567
push_dummy_failure,
568
569
/* Followed by two-byte relative address and two-byte number n.
570
After matching N times, jump to the address upon failure. */
571
/* ifdef MBS_SUPPORT, the size of address is 1. */
572
succeed_n,
573
574
/* Followed by two-byte relative address, and two-byte number n.
575
Jump to the address N times, then fail. */
576
/* ifdef MBS_SUPPORT, the size of address is 1. */
577
jump_n,
578
579
/* Set the following two-byte relative address to the
580
subsequent two-byte number. The address *includes* the two
581
bytes of number. */
582
/* ifdef MBS_SUPPORT, the size of address is 1. */
583
set_number_at,
584
585
wordchar, /* Matches any word-constituent character. */
586
notwordchar, /* Matches any char that is not a word-constituent. */
587
588
wordbeg, /* Succeeds if at word beginning. */
589
wordend, /* Succeeds if at word end. */
590
591
wordbound, /* Succeeds if at a word boundary. */
592
notwordbound /* Succeeds if not at a word boundary. */
593
594
# ifdef emacs
595
,before_dot, /* Succeeds if before point. */
596
at_dot, /* Succeeds if at point. */
597
after_dot, /* Succeeds if after point. */
598
599
/* Matches any character whose syntax is specified. Followed by
600
a byte which contains a syntax code, e.g., Sword. */
601
syntaxspec,
602
603
/* Matches any character whose syntax is not that specified. */
604
notsyntaxspec
605
# endif /* emacs */
606
} re_opcode_t;
607
#endif /* not INSIDE_RECURSION */
608
609
610
#ifdef BYTE
611
# define CHAR_T char
612
# define UCHAR_T unsigned char
613
# define COMPILED_BUFFER_VAR bufp->buffer
614
# define OFFSET_ADDRESS_SIZE 2
615
# define PREFIX(name) byte_##name
616
# define ARG_PREFIX(name) name
617
# define PUT_CHAR(c) putchar (c)
618
#else
619
# ifdef WCHAR
620
# define CHAR_T wchar_t
621
# define UCHAR_T wchar_t
622
# define COMPILED_BUFFER_VAR wc_buffer
623
# define OFFSET_ADDRESS_SIZE 1 /* the size which STORE_NUMBER macro use */
624
# define CHAR_CLASS_SIZE ((__alignof__(wctype_t)+sizeof(wctype_t))/sizeof(CHAR_T)+1)
625
# define PREFIX(name) wcs_##name
626
# define ARG_PREFIX(name) c##name
627
/* Should we use wide stream?? */
628
# define PUT_CHAR(c) printf ("%C", c);
629
# define TRUE 1
630
# define FALSE 0
631
# else
632
# ifdef MBS_SUPPORT
633
# define WCHAR
634
# define INSIDE_RECURSION
635
# include "regex.c"
636
# undef INSIDE_RECURSION
637
# endif
638
# define BYTE
639
# define INSIDE_RECURSION
640
# include "regex.c"
641
# undef INSIDE_RECURSION
642
# endif
643
#endif
644
#include "unlocked-io.h"
645
646
#ifdef INSIDE_RECURSION
647
/* Common operations on the compiled pattern. */
648
649
/* Store NUMBER in two contiguous bytes starting at DESTINATION. */
650
/* ifdef MBS_SUPPORT, we store NUMBER in 1 element. */
651
652
# ifdef WCHAR
653
# define STORE_NUMBER(destination, number) \
654
do { \
655
*(destination) = (UCHAR_T)(number); \
656
} while (0)
657
# else /* BYTE */
658
# define STORE_NUMBER(destination, number) \
659
do { \
660
(destination)[0] = (number) & 0377; \
661
(destination)[1] = (number) >> 8; \
662
} while (0)
663
# endif /* WCHAR */
664
665
/* Same as STORE_NUMBER, except increment DESTINATION to
666
the byte after where the number is stored. Therefore, DESTINATION
667
must be an lvalue. */
668
/* ifdef MBS_SUPPORT, we store NUMBER in 1 element. */
669
670
# define STORE_NUMBER_AND_INCR(destination, number) \
671
do { \
672
STORE_NUMBER (destination, number); \
673
(destination) += OFFSET_ADDRESS_SIZE; \
674
} while (0)
675
676
/* Put into DESTINATION a number stored in two contiguous bytes starting
677
at SOURCE. */
678
/* ifdef MBS_SUPPORT, we store NUMBER in 1 element. */
679
680
# ifdef WCHAR
681
# define EXTRACT_NUMBER(destination, source) \
682
do { \
683
(destination) = *(source); \
684
} while (0)
685
# else /* BYTE */
686
# define EXTRACT_NUMBER(destination, source) \
687
do { \
688
(destination) = *(source) & 0377; \
689
(destination) += SIGN_EXTEND_CHAR (*((source) + 1)) << 8; \
690
} while (0)
691
# endif
692
693
# ifdef DEBUG
694
static void PREFIX(extract_number) _RE_ARGS ((int *dest, UCHAR_T *source));
695
static void
696
PREFIX(extract_number) (dest, source)
697
int *dest;
698
UCHAR_T *source;
699
{
700
# ifdef WCHAR
701
*dest = *source;
702
# else /* BYTE */
703
int temp = SIGN_EXTEND_CHAR (*(source + 1));
704
*dest = *source & 0377;
705
*dest += temp << 8;
706
# endif
707
}
708
709
# ifndef EXTRACT_MACROS /* To debug the macros. */
710
# undef EXTRACT_NUMBER
711
# define EXTRACT_NUMBER(dest, src) PREFIX(extract_number) (&dest, src)
712
# endif /* not EXTRACT_MACROS */
713
714
# endif /* DEBUG */
715
716
/* Same as EXTRACT_NUMBER, except increment SOURCE to after the number.
717
SOURCE must be an lvalue. */
718
719
# define EXTRACT_NUMBER_AND_INCR(destination, source) \
720
do { \
721
EXTRACT_NUMBER (destination, source); \
722
(source) += OFFSET_ADDRESS_SIZE; \
723
} while (0)
724
725
# ifdef DEBUG
726
static void PREFIX(extract_number_and_incr) _RE_ARGS ((int *destination,
727
UCHAR_T **source));
728
static void
729
PREFIX(extract_number_and_incr) (destination, source)
730
int *destination;
731
UCHAR_T **source;
732
{
733
PREFIX(extract_number) (destination, *source);
734
*source += OFFSET_ADDRESS_SIZE;
735
}
736
737
# ifndef EXTRACT_MACROS
738
# undef EXTRACT_NUMBER_AND_INCR
739
# define EXTRACT_NUMBER_AND_INCR(dest, src) \
740
PREFIX(extract_number_and_incr) (&dest, &src)
741
# endif /* not EXTRACT_MACROS */
742
743
# endif /* DEBUG */
744
745
746
747
/* If DEBUG is defined, Regex prints many voluminous messages about what
748
it is doing (if the variable `debug' is nonzero). If linked with the
749
main program in `iregex.c', you can enter patterns and strings
750
interactively. And if linked with the main program in `main.c' and
751
the other test files, you can run the already-written tests. */
752
753
# ifdef DEBUG
754
755
# ifndef DEFINED_ONCE
756
757
/* We use standard I/O for debugging. */
758
# include <stdio.h>
759
760
/* It is useful to test things that ``must'' be true when debugging. */
761
# include <assert.h>
762
763
static int debug;
764
765
# define DEBUG_STATEMENT(e) e
766
# define DEBUG_PRINT1(x) if (debug) printf (x)
767
# define DEBUG_PRINT2(x1, x2) if (debug) printf (x1, x2)
768
# define DEBUG_PRINT3(x1, x2, x3) if (debug) printf (x1, x2, x3)
769
# define DEBUG_PRINT4(x1, x2, x3, x4) if (debug) printf (x1, x2, x3, x4)
770
# endif /* not DEFINED_ONCE */
771
772
# define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) \
773
if (debug) PREFIX(print_partial_compiled_pattern) (s, e)
774
# define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) \
775
if (debug) PREFIX(print_double_string) (w, s1, sz1, s2, sz2)
776
777
778
/* Print the fastmap in human-readable form. */
779
780
# ifndef DEFINED_ONCE
781
void
782
print_fastmap (fastmap)
783
char *fastmap;
784
{
785
unsigned was_a_range = 0;
786
unsigned i = 0;
787
788
while (i < (1 << BYTEWIDTH))
789
{
790
if (fastmap[i++])
791
{
792
was_a_range = 0;
793
putchar (i - 1);
794
while (i < (1 << BYTEWIDTH) && fastmap[i])
795
{
796
was_a_range = 1;
797
i++;
798
}
799
if (was_a_range)
800
{
801
printf ("-");
802
putchar (i - 1);
803
}
804
}
805
}
806
putchar ('\n');
807
}
808
# endif /* not DEFINED_ONCE */
809
810
811
/* Print a compiled pattern string in human-readable form, starting at
812
the START pointer into it and ending just before the pointer END. */
813
814
void
815
PREFIX(print_partial_compiled_pattern) (start, end)
816
UCHAR_T *start;
817
UCHAR_T *end;
818
{
819
int mcnt, mcnt2;
820
UCHAR_T *p1;
821
UCHAR_T *p = start;
822
UCHAR_T *pend = end;
823
824
if (start == NULL)
825
{
826
printf ("(null)\n");
827
return;
828
}
829
830
/* Loop over pattern commands. */
831
while (p < pend)
832
{
833
# ifdef _LIBC
834
printf ("%td:\t", p - start);
835
# else
836
printf ("%ld:\t", (long int) (p - start));
837
# endif
838
839
switch ((re_opcode_t) *p++)
840
{
841
case no_op:
842
printf ("/no_op");
843
break;
844
845
case exactn:
846
mcnt = *p++;
847
printf ("/exactn/%d", mcnt);
848
do
849
{
850
putchar ('/');
851
PUT_CHAR (*p++);
852
}
853
while (--mcnt);
854
break;
855
856
# ifdef MBS_SUPPORT
857
case exactn_bin:
858
mcnt = *p++;
859
printf ("/exactn_bin/%d", mcnt);
860
do
861
{
862
printf("/%lx", (long int) *p++);
863
}
864
while (--mcnt);
865
break;
866
# endif /* MBS_SUPPORT */
867
868
case start_memory:
869
mcnt = *p++;
870
printf ("/start_memory/%d/%ld", mcnt, (long int) *p++);
871
break;
872
873
case stop_memory:
874
mcnt = *p++;
875
printf ("/stop_memory/%d/%ld", mcnt, (long int) *p++);
876
break;
877
878
case duplicate:
879
printf ("/duplicate/%ld", (long int) *p++);
880
break;
881
882
case anychar:
883
printf ("/anychar");
884
break;
885
886
case charset:
887
case charset_not:
888
{
889
# ifdef WCHAR
890
int i, length;
891
wchar_t *workp = p;
892
printf ("/charset [%s",
893
(re_opcode_t) *(workp - 1) == charset_not ? "^" : "");
894
p += 5;
895
length = *workp++; /* the length of char_classes */
896
for (i=0 ; i<length ; i++)
897
printf("[:%lx:]", (long int) *p++);
898
length = *workp++; /* the length of collating_symbol */
899
for (i=0 ; i<length ;)
900
{
901
printf("[.");
902
while(*p != 0)
903
PUT_CHAR((i++,*p++));
904
i++,p++;
905
printf(".]");
906
}
907
length = *workp++; /* the length of equivalence_class */
908
for (i=0 ; i<length ;)
909
{
910
printf("[=");
911
while(*p != 0)
912
PUT_CHAR((i++,*p++));
913
i++,p++;
914
printf("=]");
915
}
916
length = *workp++; /* the length of char_range */
917
for (i=0 ; i<length ; i++)
918
{
919
wchar_t range_start = *p++;
920
wchar_t range_end = *p++;
921
printf("%C-%C", range_start, range_end);
922
}
923
length = *workp++; /* the length of char */
924
for (i=0 ; i<length ; i++)
925
printf("%C", *p++);
926
putchar (']');
927
# else
928
register int c, last = -100;
929
register int in_range = 0;
930
931
printf ("/charset [%s",
932
(re_opcode_t) *(p - 1) == charset_not ? "^" : "");
933
934
assert (p + *p < pend);
935
936
for (c = 0; c < 256; c++)
937
if (c / 8 < *p
938
&& (p[1 + (c/8)] & (1 << (c % 8))))
939
{
940
/* Are we starting a range? */
941
if (last + 1 == c && ! in_range)
942
{
943
putchar ('-');
944
in_range = 1;
945
}
946
/* Have we broken a range? */
947
else if (last + 1 != c && in_range)
948
{
949
putchar (last);
950
in_range = 0;
951
}
952
953
if (! in_range)
954
putchar (c);
955
956
last = c;
957
}
958
959
if (in_range)
960
putchar (last);
961
962
putchar (']');
963
964
p += 1 + *p;
965
# endif /* WCHAR */
966
}
967
break;
968
969
case begline:
970
printf ("/begline");
971
break;
972
973
case endline:
974
printf ("/endline");
975
break;
976
977
case on_failure_jump:
978
PREFIX(extract_number_and_incr) (&mcnt, &p);
979
# ifdef _LIBC
980
printf ("/on_failure_jump to %td", p + mcnt - start);
981
# else
982
printf ("/on_failure_jump to %ld", (long int) (p + mcnt - start));
983
# endif
984
break;
985
986
case on_failure_keep_string_jump:
987
PREFIX(extract_number_and_incr) (&mcnt, &p);
988
# ifdef _LIBC
989
printf ("/on_failure_keep_string_jump to %td", p + mcnt - start);
990
# else
991
printf ("/on_failure_keep_string_jump to %ld",
992
(long int) (p + mcnt - start));
993
# endif
994
break;
995
996
case dummy_failure_jump:
997
PREFIX(extract_number_and_incr) (&mcnt, &p);
998
# ifdef _LIBC
999
printf ("/dummy_failure_jump to %td", p + mcnt - start);
1000
# else
1001
printf ("/dummy_failure_jump to %ld", (long int) (p + mcnt - start));
1002
# endif
1003
break;
1004
1005
case push_dummy_failure:
1006
printf ("/push_dummy_failure");
1007
break;
1008
1009
case maybe_pop_jump:
1010
PREFIX(extract_number_and_incr) (&mcnt, &p);
1011
# ifdef _LIBC
1012
printf ("/maybe_pop_jump to %td", p + mcnt - start);
1013
# else
1014
printf ("/maybe_pop_jump to %ld", (long int) (p + mcnt - start));
1015
# endif
1016
break;
1017
1018
case pop_failure_jump:
1019
PREFIX(extract_number_and_incr) (&mcnt, &p);
1020
# ifdef _LIBC
1021
printf ("/pop_failure_jump to %td", p + mcnt - start);
1022
# else
1023
printf ("/pop_failure_jump to %ld", (long int) (p + mcnt - start));
1024
# endif
1025
break;
1026
1027
case jump_past_alt:
1028
PREFIX(extract_number_and_incr) (&mcnt, &p);
1029
# ifdef _LIBC
1030
printf ("/jump_past_alt to %td", p + mcnt - start);
1031
# else
1032
printf ("/jump_past_alt to %ld", (long int) (p + mcnt - start));
1033
# endif
1034
break;
1035
1036
case jump:
1037
PREFIX(extract_number_and_incr) (&mcnt, &p);
1038
# ifdef _LIBC
1039
printf ("/jump to %td", p + mcnt - start);
1040
# else
1041
printf ("/jump to %ld", (long int) (p + mcnt - start));
1042
# endif
1043
break;
1044
1045
case succeed_n:
1046
PREFIX(extract_number_and_incr) (&mcnt, &p);
1047
p1 = p + mcnt;
1048
PREFIX(extract_number_and_incr) (&mcnt2, &p);
1049
# ifdef _LIBC
1050
printf ("/succeed_n to %td, %d times", p1 - start, mcnt2);
1051
# else
1052
printf ("/succeed_n to %ld, %d times",
1053
(long int) (p1 - start), mcnt2);
1054
# endif
1055
break;
1056
1057
case jump_n:
1058
PREFIX(extract_number_and_incr) (&mcnt, &p);
1059
p1 = p + mcnt;
1060
PREFIX(extract_number_and_incr) (&mcnt2, &p);
1061
printf ("/jump_n to %d, %d times", p1 - start, mcnt2);
1062
break;
1063
1064
case set_number_at:
1065
PREFIX(extract_number_and_incr) (&mcnt, &p);
1066
p1 = p + mcnt;
1067
PREFIX(extract_number_and_incr) (&mcnt2, &p);
1068
# ifdef _LIBC
1069
printf ("/set_number_at location %td to %d", p1 - start, mcnt2);
1070
# else
1071
printf ("/set_number_at location %ld to %d",
1072
(long int) (p1 - start), mcnt2);
1073
# endif
1074
break;
1075
1076
case wordbound:
1077
printf ("/wordbound");
1078
break;
1079
1080
case notwordbound:
1081
printf ("/notwordbound");
1082
break;
1083
1084
case wordbeg:
1085
printf ("/wordbeg");
1086
break;
1087
1088
case wordend:
1089
printf ("/wordend");
1090
break;
1091
1092
# ifdef emacs
1093
case before_dot:
1094
printf ("/before_dot");
1095
break;
1096
1097
case at_dot:
1098
printf ("/at_dot");
1099
break;
1100
1101
case after_dot:
1102
printf ("/after_dot");
1103
break;
1104
1105
case syntaxspec:
1106
printf ("/syntaxspec");
1107
mcnt = *p++;
1108
printf ("/%d", mcnt);
1109
break;
1110
1111
case notsyntaxspec:
1112
printf ("/notsyntaxspec");
1113
mcnt = *p++;
1114
printf ("/%d", mcnt);
1115
break;
1116
# endif /* emacs */
1117
1118
case wordchar:
1119
printf ("/wordchar");
1120
break;
1121
1122
case notwordchar:
1123
printf ("/notwordchar");
1124
break;
1125
1126
case begbuf:
1127
printf ("/begbuf");
1128
break;
1129
1130
case endbuf:
1131
printf ("/endbuf");
1132
break;
1133
1134
default:
1135
printf ("?%ld", (long int) *(p-1));
1136
}
1137
1138
putchar ('\n');
1139
}
1140
1141
# ifdef _LIBC
1142
printf ("%td:\tend of pattern.\n", p - start);
1143
# else
1144
printf ("%ld:\tend of pattern.\n", (long int) (p - start));
1145
# endif
1146
}
1147
1148
1149
void
1150
PREFIX(print_compiled_pattern) (bufp)
1151
struct re_pattern_buffer *bufp;
1152
{
1153
UCHAR_T *buffer = (UCHAR_T*) bufp->buffer;
1154
1155
PREFIX(print_partial_compiled_pattern) (buffer, buffer
1156
+ bufp->used / sizeof(UCHAR_T));
1157
printf ("%ld bytes used/%ld bytes allocated.\n",
1158
bufp->used, bufp->allocated);
1159
1160
if (bufp->fastmap_accurate && bufp->fastmap)
1161
{
1162
printf ("fastmap: ");
1163
print_fastmap (bufp->fastmap);
1164
}
1165
1166
# ifdef _LIBC
1167
printf ("re_nsub: %Zd\t", bufp->re_nsub);
1168
# else
1169
printf ("re_nsub: %ld\t", (long int) bufp->re_nsub);
1170
# endif
1171
printf ("regs_alloc: %d\t", bufp->regs_allocated);
1172
printf ("can_be_null: %d\t", bufp->can_be_null);
1173
printf ("newline_anchor: %d\n", bufp->newline_anchor);
1174
printf ("no_sub: %d\t", bufp->no_sub);
1175
printf ("not_bol: %d\t", bufp->not_bol);
1176
printf ("not_eol: %d\t", bufp->not_eol);
1177
printf ("syntax: %lx\n", bufp->syntax);
1178
/* Perhaps we should print the translate table? */
1179
}
1180
1181
1182
void
1183
PREFIX(print_double_string) (where, string1, size1, string2, size2)
1184
const CHAR_T *where;
1185
const CHAR_T *string1;
1186
const CHAR_T *string2;
1187
int size1;
1188
int size2;
1189
{
1190
int this_char;
1191
1192
if (where == NULL)
1193
printf ("(null)");
1194
else
1195
{
1196
int cnt;
1197
1198
if (FIRST_STRING_P (where))
1199
{
1200
for (this_char = where - string1; this_char < size1; this_char++)
1201
PUT_CHAR (string1[this_char]);
1202
1203
where = string2;
1204
}
1205
1206
cnt = 0;
1207
for (this_char = where - string2; this_char < size2; this_char++)
1208
{
1209
PUT_CHAR (string2[this_char]);
1210
if (++cnt > 100)
1211
{
1212
fputs ("...", stdout);
1213
break;
1214
}
1215
}
1216
}
1217
}
1218
1219
# ifndef DEFINED_ONCE
1220
void
1221
printchar (c)
1222
int c;
1223
{
1224
putc (c, stderr);
1225
}
1226
# endif
1227
1228
# else /* not DEBUG */
1229
1230
# ifndef DEFINED_ONCE
1231
# undef assert
1232
# define assert(e)
1233
1234
# define DEBUG_STATEMENT(e)
1235
# define DEBUG_PRINT1(x)
1236
# define DEBUG_PRINT2(x1, x2)
1237
# define DEBUG_PRINT3(x1, x2, x3)
1238
# define DEBUG_PRINT4(x1, x2, x3, x4)
1239
# endif /* not DEFINED_ONCE */
1240
# define DEBUG_PRINT_COMPILED_PATTERN(p, s, e)
1241
# define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2)
1242
1243
# endif /* not DEBUG */
1244
1245
1246
1247
# ifdef WCHAR
1248
/* This convert a multibyte string to a wide character string.
1249
And write their correspondances to offset_buffer(see below)
1250
and write whether each wchar_t is binary data to is_binary.
1251
This assume invalid multibyte sequences as binary data.
1252
We assume offset_buffer and is_binary is already allocated
1253
enough space. */
1254
1255
static size_t convert_mbs_to_wcs (CHAR_T *dest, const unsigned char* src,
1256
size_t len, int *offset_buffer,
1257
char *is_binary);
1258
static size_t
1259
convert_mbs_to_wcs (dest, src, len, offset_buffer, is_binary)
1260
CHAR_T *dest;
1261
const unsigned char* src;
1262
size_t len; /* the length of multibyte string. */
1263
1264
/* It hold correspondances between src(char string) and
1265
dest(wchar_t string) for optimization.
1266
e.g. src = "xxxyzz"
1267
dest = {'X', 'Y', 'Z'}
1268
(each "xxx", "y" and "zz" represent one multibyte character
1269
corresponding to 'X', 'Y' and 'Z'.)
1270
offset_buffer = {0, 0+3("xxx"), 0+3+1("y"), 0+3+1+2("zz")}
1271
= {0, 3, 4, 6}
1272
*/
1273
int *offset_buffer;
1274
char *is_binary;
1275
{
1276
wchar_t *pdest = dest;
1277
const unsigned char *psrc = src;
1278
size_t wc_count = 0;
1279
1280
mbstate_t mbs;
1281
int i, consumed;
1282
size_t mb_remain = len;
1283
size_t mb_count = 0;
1284
1285
/* Initialize the conversion state. */
1286
memset (&mbs, 0, sizeof (mbstate_t));
1287
1288
offset_buffer[0] = 0;
1289
for( ; mb_remain > 0 ; ++wc_count, ++pdest, mb_remain -= consumed,
1290
psrc += consumed)
1291
{
1292
consumed = mbrtowc (pdest, psrc, mb_remain, &mbs);
1293
1294
if (consumed <= 0)
1295
/* failed to convert. maybe src contains binary data.
1296
So we consume 1 byte manualy. */
1297
{
1298
*pdest = *psrc;
1299
consumed = 1;
1300
is_binary[wc_count] = TRUE;
1301
}
1302
else
1303
is_binary[wc_count] = FALSE;
1304
/* In sjis encoding, we use yen sign as escape character in
1305
place of reverse solidus. So we convert 0x5c(yen sign in
1306
sjis) to not 0xa5(yen sign in UCS2) but 0x5c(reverse
1307
solidus in UCS2). */
1308
if (consumed == 1 && (int) *psrc == 0x5c && (int) *pdest == 0xa5)
1309
*pdest = (wchar_t) *psrc;
1310
1311
offset_buffer[wc_count + 1] = mb_count += consumed;
1312
}
1313
1314
/* Fill remain of the buffer with sentinel. */
1315
for (i = wc_count + 1 ; i <= len ; i++)
1316
offset_buffer[i] = mb_count + 1;
1317
1318
return wc_count;
1319
}
1320
1321
# endif /* WCHAR */
1322
1323
#else /* not INSIDE_RECURSION */
1324
1325
/* Set by `re_set_syntax' to the current regexp syntax to recognize. Can
1326
also be assigned to arbitrarily: each pattern buffer stores its own
1327
syntax, so it can be changed between regex compilations. */
1328
/* This has no initializer because initialized variables in Emacs
1329
become read-only after dumping. */
1330
reg_syntax_t re_syntax_options;
1331
1332
1333
/* Specify the precise syntax of regexps for compilation. This provides
1334
for compatibility for various utilities which historically have
1335
different, incompatible syntaxes.
1336
1337
The argument SYNTAX is a bit mask comprised of the various bits
1338
defined in regex.h. We return the old syntax. */
1339
1340
reg_syntax_t
1341
re_set_syntax (syntax)
1342
reg_syntax_t syntax;
1343
{
1344
reg_syntax_t ret = re_syntax_options;
1345
1346
re_syntax_options = syntax;
1347
# ifdef DEBUG
1348
if (syntax & RE_DEBUG)
1349
debug = 1;
1350
else if (debug) /* was on but now is not */
1351
debug = 0;
1352
# endif /* DEBUG */
1353
return ret;
1354
}
1355
# ifdef _LIBC
1356
weak_alias (__re_set_syntax, re_set_syntax)
1357
# endif
1358
1359
/* This table gives an error message for each of the error codes listed
1360
in regex.h. Obviously the order here has to be same as there.
1361
POSIX doesn't require that we do anything for REG_NOERROR,
1362
but why not be nice? */
1363
1364
static const char re_error_msgid[] =
1365
{
1366
# define REG_NOERROR_IDX 0
1367
gettext_noop ("Success") /* REG_NOERROR */
1368
"\0"
1369
# define REG_NOMATCH_IDX (REG_NOERROR_IDX + sizeof "Success")
1370
gettext_noop ("No match") /* REG_NOMATCH */
1371
"\0"
1372
# define REG_BADPAT_IDX (REG_NOMATCH_IDX + sizeof "No match")
1373
gettext_noop ("Invalid regular expression") /* REG_BADPAT */
1374
"\0"
1375
# define REG_ECOLLATE_IDX (REG_BADPAT_IDX + sizeof "Invalid regular expression")
1376
gettext_noop ("Invalid collation character") /* REG_ECOLLATE */
1377
"\0"
1378
# define REG_ECTYPE_IDX (REG_ECOLLATE_IDX + sizeof "Invalid collation character")
1379
gettext_noop ("Invalid character class name") /* REG_ECTYPE */
1380
"\0"
1381
# define REG_EESCAPE_IDX (REG_ECTYPE_IDX + sizeof "Invalid character class name")
1382
gettext_noop ("Trailing backslash") /* REG_EESCAPE */
1383
"\0"
1384
# define REG_ESUBREG_IDX (REG_EESCAPE_IDX + sizeof "Trailing backslash")
1385
gettext_noop ("Invalid back reference") /* REG_ESUBREG */
1386
"\0"
1387
# define REG_EBRACK_IDX (REG_ESUBREG_IDX + sizeof "Invalid back reference")
1388
gettext_noop ("Unmatched [ or [^") /* REG_EBRACK */
1389
"\0"
1390
# define REG_EPAREN_IDX (REG_EBRACK_IDX + sizeof "Unmatched [ or [^")
1391
gettext_noop ("Unmatched ( or \\(") /* REG_EPAREN */
1392
"\0"
1393
# define REG_EBRACE_IDX (REG_EPAREN_IDX + sizeof "Unmatched ( or \\(")
1394
gettext_noop ("Unmatched \\{") /* REG_EBRACE */
1395
"\0"
1396
# define REG_BADBR_IDX (REG_EBRACE_IDX + sizeof "Unmatched \\{")
1397
gettext_noop ("Invalid content of \\{\\}") /* REG_BADBR */
1398
"\0"
1399
# define REG_ERANGE_IDX (REG_BADBR_IDX + sizeof "Invalid content of \\{\\}")
1400
gettext_noop ("Invalid range end") /* REG_ERANGE */
1401
"\0"
1402
# define REG_ESPACE_IDX (REG_ERANGE_IDX + sizeof "Invalid range end")
1403
gettext_noop ("Memory exhausted") /* REG_ESPACE */
1404
"\0"
1405
# define REG_BADRPT_IDX (REG_ESPACE_IDX + sizeof "Memory exhausted")
1406
gettext_noop ("Invalid preceding regular expression") /* REG_BADRPT */
1407
"\0"
1408
# define REG_EEND_IDX (REG_BADRPT_IDX + sizeof "Invalid preceding regular expression")
1409
gettext_noop ("Premature end of regular expression") /* REG_EEND */
1410
"\0"
1411
# define REG_ESIZE_IDX (REG_EEND_IDX + sizeof "Premature end of regular expression")
1412
gettext_noop ("Regular expression too big") /* REG_ESIZE */
1413
"\0"
1414
# define REG_ERPAREN_IDX (REG_ESIZE_IDX + sizeof "Regular expression too big")
1415
gettext_noop ("Unmatched ) or \\)") /* REG_ERPAREN */
1416
};
1417
1418
static const size_t re_error_msgid_idx[] =
1419
{
1420
REG_NOERROR_IDX,
1421
REG_NOMATCH_IDX,
1422
REG_BADPAT_IDX,
1423
REG_ECOLLATE_IDX,
1424
REG_ECTYPE_IDX,
1425
REG_EESCAPE_IDX,
1426
REG_ESUBREG_IDX,
1427
REG_EBRACK_IDX,
1428
REG_EPAREN_IDX,
1429
REG_EBRACE_IDX,
1430
REG_BADBR_IDX,
1431
REG_ERANGE_IDX,
1432
REG_ESPACE_IDX,
1433
REG_BADRPT_IDX,
1434
REG_EEND_IDX,
1435
REG_ESIZE_IDX,
1436
REG_ERPAREN_IDX
1437
};
1438
1439
#endif /* INSIDE_RECURSION */
1440
1441
#ifndef DEFINED_ONCE
1442
/* Avoiding alloca during matching, to placate r_alloc. */
1443
1444
/* Define MATCH_MAY_ALLOCATE unless we need to make sure that the
1445
searching and matching functions should not call alloca. On some
1446
systems, alloca is implemented in terms of malloc, and if we're
1447
using the relocating allocator routines, then malloc could cause a
1448
relocation, which might (if the strings being searched are in the
1449
ralloc heap) shift the data out from underneath the regexp
1450
routines.
1451
1452
Here's another reason to avoid allocation: Emacs
1453
processes input from X in a signal handler; processing X input may
1454
call malloc; if input arrives while a matching routine is calling
1455
malloc, then we're scrod. But Emacs can't just block input while
1456
calling matching routines; then we don't notice interrupts when
1457
they come in. So, Emacs blocks input around all regexp calls
1458
except the matching calls, which it leaves unprotected, in the
1459
faith that they will not malloc. */
1460
1461
/* Normally, this is fine. */
1462
# define MATCH_MAY_ALLOCATE
1463
1464
/* When using GNU C, we are not REALLY using the C alloca, no matter
1465
what config.h may say. So don't take precautions for it. */
1466
# ifdef __GNUC__
1467
# undef C_ALLOCA
1468
# endif
1469
1470
/* The match routines may not allocate if (1) they would do it with malloc
1471
and (2) it's not safe for them to use malloc.
1472
Note that if REL_ALLOC is defined, matching would not use malloc for the
1473
failure stack, but we would still use it for the register vectors;
1474
so REL_ALLOC should not affect this. */
1475
# if (defined C_ALLOCA || defined REGEX_MALLOC) && defined emacs
1476
# undef MATCH_MAY_ALLOCATE
1477
# endif
1478
#endif /* not DEFINED_ONCE */
1479
1480
#ifdef INSIDE_RECURSION
1481
/* Failure stack declarations and macros; both re_compile_fastmap and
1482
re_match_2 use a failure stack. These have to be macros because of
1483
REGEX_ALLOCATE_STACK. */
1484
1485
1486
/* Number of failure points for which to initially allocate space
1487
when matching. If this number is exceeded, we allocate more
1488
space, so it is not a hard limit. */
1489
# ifndef INIT_FAILURE_ALLOC
1490
# define INIT_FAILURE_ALLOC 5
1491
# endif
1492
1493
/* Roughly the maximum number of failure points on the stack. Would be
1494
exactly that if always used MAX_FAILURE_ITEMS items each time we failed.
1495
This is a variable only so users of regex can assign to it; we never
1496
change it ourselves. */
1497
1498
# ifdef INT_IS_16BIT
1499
1500
# ifndef DEFINED_ONCE
1501
# if defined MATCH_MAY_ALLOCATE
1502
/* 4400 was enough to cause a crash on Alpha OSF/1,
1503
whose default stack limit is 2mb. */
1504
long int re_max_failures = 4000;
1505
# else
1506
long int re_max_failures = 2000;
1507
# endif
1508
# endif
1509
1510
union PREFIX(fail_stack_elt)
1511
{
1512
UCHAR_T *pointer;
1513
long int integer;
1514
};
1515
1516
typedef union PREFIX(fail_stack_elt) PREFIX(fail_stack_elt_t);
1517
1518
typedef struct
1519
{
1520
PREFIX(fail_stack_elt_t) *stack;
1521
unsigned long int size;
1522
unsigned long int avail; /* Offset of next open position. */
1523
} PREFIX(fail_stack_type);
1524
1525
# else /* not INT_IS_16BIT */
1526
1527
# ifndef DEFINED_ONCE
1528
# if defined MATCH_MAY_ALLOCATE
1529
/* 4400 was enough to cause a crash on Alpha OSF/1,
1530
whose default stack limit is 2mb. */
1531
int re_max_failures = 4000;
1532
# else
1533
int re_max_failures = 2000;
1534
# endif
1535
# endif
1536
1537
union PREFIX(fail_stack_elt)
1538
{
1539
UCHAR_T *pointer;
1540
int integer;
1541
};
1542
1543
typedef union PREFIX(fail_stack_elt) PREFIX(fail_stack_elt_t);
1544
1545
typedef struct
1546
{
1547
PREFIX(fail_stack_elt_t) *stack;
1548
unsigned size;
1549
unsigned avail; /* Offset of next open position. */
1550
} PREFIX(fail_stack_type);
1551
1552
# endif /* INT_IS_16BIT */
1553
1554
# ifndef DEFINED_ONCE
1555
# define FAIL_STACK_EMPTY() (fail_stack.avail == 0)
1556
# define FAIL_STACK_PTR_EMPTY() (fail_stack_ptr->avail == 0)
1557
# define FAIL_STACK_FULL() (fail_stack.avail == fail_stack.size)
1558
# endif
1559
1560
1561
/* Define macros to initialize and free the failure stack.
1562
Do `return -2' if the alloc fails. */
1563
1564
# ifdef MATCH_MAY_ALLOCATE
1565
# define INIT_FAIL_STACK() \
1566
do { \
1567
fail_stack.stack = (PREFIX(fail_stack_elt_t) *) \
1568
REGEX_ALLOCATE_STACK (INIT_FAILURE_ALLOC * sizeof (PREFIX(fail_stack_elt_t))); \
1569
\
1570
if (fail_stack.stack == NULL) \
1571
return -2; \
1572
\
1573
fail_stack.size = INIT_FAILURE_ALLOC; \
1574
fail_stack.avail = 0; \
1575
} while (0)
1576
1577
# define RESET_FAIL_STACK() REGEX_FREE_STACK (fail_stack.stack)
1578
# else
1579
# define INIT_FAIL_STACK() \
1580
do { \
1581
fail_stack.avail = 0; \
1582
} while (0)
1583
1584
# define RESET_FAIL_STACK()
1585
# endif
1586
1587
1588
/* Double the size of FAIL_STACK, up to approximately `re_max_failures' items.
1589
1590
Return 1 if succeeds, and 0 if either ran out of memory
1591
allocating space for it or it was already too large.
1592
1593
REGEX_REALLOCATE_STACK requires `destination' be declared. */
1594
1595
# define DOUBLE_FAIL_STACK(fail_stack) \
1596
((fail_stack).size > (unsigned) (re_max_failures * MAX_FAILURE_ITEMS) \
1597
? 0 \
1598
: ((fail_stack).stack = (PREFIX(fail_stack_elt_t) *) \
1599
REGEX_REALLOCATE_STACK ((fail_stack).stack, \
1600
(fail_stack).size * sizeof (PREFIX(fail_stack_elt_t)), \
1601
((fail_stack).size << 1) * sizeof (PREFIX(fail_stack_elt_t))),\
1602
\
1603
(fail_stack).stack == NULL \
1604
? 0 \
1605
: ((fail_stack).size <<= 1, \
1606
1)))
1607
1608
1609
/* Push pointer POINTER on FAIL_STACK.
1610
Return 1 if was able to do so and 0 if ran out of memory allocating
1611
space to do so. */
1612
# define PUSH_PATTERN_OP(POINTER, FAIL_STACK) \
1613
((FAIL_STACK_FULL () \
1614
&& !DOUBLE_FAIL_STACK (FAIL_STACK)) \
1615
? 0 \
1616
: ((FAIL_STACK).stack[(FAIL_STACK).avail++].pointer = POINTER, \
1617
1))
1618
1619
/* Push a pointer value onto the failure stack.
1620
Assumes the variable `fail_stack'. Probably should only
1621
be called from within `PUSH_FAILURE_POINT'. */
1622
# define PUSH_FAILURE_POINTER(item) \
1623
fail_stack.stack[fail_stack.avail++].pointer = (UCHAR_T *) (item)
1624
1625
/* This pushes an integer-valued item onto the failure stack.
1626
Assumes the variable `fail_stack'. Probably should only
1627
be called from within `PUSH_FAILURE_POINT'. */
1628
# define PUSH_FAILURE_INT(item) \
1629
fail_stack.stack[fail_stack.avail++].integer = (item)
1630
1631
/* Push a fail_stack_elt_t value onto the failure stack.
1632
Assumes the variable `fail_stack'. Probably should only
1633
be called from within `PUSH_FAILURE_POINT'. */
1634
# define PUSH_FAILURE_ELT(item) \
1635
fail_stack.stack[fail_stack.avail++] = (item)
1636
1637
/* These three POP... operations complement the three PUSH... operations.
1638
All assume that `fail_stack' is nonempty. */
1639
# define POP_FAILURE_POINTER() fail_stack.stack[--fail_stack.avail].pointer
1640
# define POP_FAILURE_INT() fail_stack.stack[--fail_stack.avail].integer
1641
# define POP_FAILURE_ELT() fail_stack.stack[--fail_stack.avail]
1642
1643
/* Used to omit pushing failure point id's when we're not debugging. */
1644
# ifdef DEBUG
1645
# define DEBUG_PUSH PUSH_FAILURE_INT
1646
# define DEBUG_POP(item_addr) *(item_addr) = POP_FAILURE_INT ()
1647
# else
1648
# define DEBUG_PUSH(item)
1649
# define DEBUG_POP(item_addr)
1650
# endif
1651
1652
1653
/* Push the information about the state we will need
1654
if we ever fail back to it.
1655
1656
Requires variables fail_stack, regstart, regend, reg_info, and
1657
num_regs_pushed be declared. DOUBLE_FAIL_STACK requires `destination'
1658
be declared.
1659
1660
Does `return FAILURE_CODE' if runs out of memory. */
1661
1662
# define PUSH_FAILURE_POINT(pattern_place, string_place, failure_code) \
1663
do { \
1664
char *destination; \
1665
/* Must be int, so when we don't save any registers, the arithmetic \
1666
of 0 + -1 isn't done as unsigned. */ \
1667
/* Can't be int, since there is not a shred of a guarantee that int \
1668
is wide enough to hold a value of something to which pointer can \
1669
be assigned */ \
1670
active_reg_t this_reg; \
1671
\
1672
DEBUG_STATEMENT (failure_id++); \
1673
DEBUG_STATEMENT (nfailure_points_pushed++); \
1674
DEBUG_PRINT2 ("\nPUSH_FAILURE_POINT #%u:\n", failure_id); \
1675
DEBUG_PRINT2 (" Before push, next avail: %d\n", (fail_stack).avail);\
1676
DEBUG_PRINT2 (" size: %d\n", (fail_stack).size);\
1677
\
1678
DEBUG_PRINT2 (" slots needed: %ld\n", NUM_FAILURE_ITEMS); \
1679
DEBUG_PRINT2 (" available: %d\n", REMAINING_AVAIL_SLOTS); \
1680
\
1681
/* Ensure we have enough space allocated for what we will push. */ \
1682
while (REMAINING_AVAIL_SLOTS < NUM_FAILURE_ITEMS) \
1683
{ \
1684
if (!DOUBLE_FAIL_STACK (fail_stack)) \
1685
return failure_code; \
1686
\
1687
DEBUG_PRINT2 ("\n Doubled stack; size now: %d\n", \
1688
(fail_stack).size); \
1689
DEBUG_PRINT2 (" slots available: %d\n", REMAINING_AVAIL_SLOTS);\
1690
} \
1691
\
1692
/* Push the info, starting with the registers. */ \
1693
DEBUG_PRINT1 ("\n"); \
1694
\
1695
if (1) \
1696
for (this_reg = lowest_active_reg; this_reg <= highest_active_reg; \
1697
this_reg++) \
1698
{ \
1699
DEBUG_PRINT2 (" Pushing reg: %lu\n", this_reg); \
1700
DEBUG_STATEMENT (num_regs_pushed++); \
1701
\
1702
DEBUG_PRINT2 (" start: %p\n", regstart[this_reg]); \
1703
PUSH_FAILURE_POINTER (regstart[this_reg]); \
1704
\
1705
DEBUG_PRINT2 (" end: %p\n", regend[this_reg]); \
1706
PUSH_FAILURE_POINTER (regend[this_reg]); \
1707
\
1708
DEBUG_PRINT2 (" info: %p\n ", \
1709
reg_info[this_reg].word.pointer); \
1710
DEBUG_PRINT2 (" match_null=%d", \
1711
REG_MATCH_NULL_STRING_P (reg_info[this_reg])); \
1712
DEBUG_PRINT2 (" active=%d", IS_ACTIVE (reg_info[this_reg])); \
1713
DEBUG_PRINT2 (" matched_something=%d", \
1714
MATCHED_SOMETHING (reg_info[this_reg])); \
1715
DEBUG_PRINT2 (" ever_matched=%d", \
1716
EVER_MATCHED_SOMETHING (reg_info[this_reg])); \
1717
DEBUG_PRINT1 ("\n"); \
1718
PUSH_FAILURE_ELT (reg_info[this_reg].word); \
1719
} \
1720
\
1721
DEBUG_PRINT2 (" Pushing low active reg: %ld\n", lowest_active_reg);\
1722
PUSH_FAILURE_INT (lowest_active_reg); \
1723
\
1724
DEBUG_PRINT2 (" Pushing high active reg: %ld\n", highest_active_reg);\
1725
PUSH_FAILURE_INT (highest_active_reg); \
1726
\
1727
DEBUG_PRINT2 (" Pushing pattern %p:\n", pattern_place); \
1728
DEBUG_PRINT_COMPILED_PATTERN (bufp, pattern_place, pend); \
1729
PUSH_FAILURE_POINTER (pattern_place); \
1730
\
1731
DEBUG_PRINT2 (" Pushing string %p: `", string_place); \
1732
DEBUG_PRINT_DOUBLE_STRING (string_place, string1, size1, string2, \
1733
size2); \
1734
DEBUG_PRINT1 ("'\n"); \
1735
PUSH_FAILURE_POINTER (string_place); \
1736
\
1737
DEBUG_PRINT2 (" Pushing failure id: %u\n", failure_id); \
1738
DEBUG_PUSH (failure_id); \
1739
} while (0)
1740
1741
# ifndef DEFINED_ONCE
1742
/* This is the number of items that are pushed and popped on the stack
1743
for each register. */
1744
# define NUM_REG_ITEMS 3
1745
1746
/* Individual items aside from the registers. */
1747
# ifdef DEBUG
1748
# define NUM_NONREG_ITEMS 5 /* Includes failure point id. */
1749
# else
1750
# define NUM_NONREG_ITEMS 4
1751
# endif
1752
1753
/* We push at most this many items on the stack. */
1754
/* We used to use (num_regs - 1), which is the number of registers
1755
this regexp will save; but that was changed to 5
1756
to avoid stack overflow for a regexp with lots of parens. */
1757
# define MAX_FAILURE_ITEMS (5 * NUM_REG_ITEMS + NUM_NONREG_ITEMS)
1758
1759
/* We actually push this many items. */
1760
# define NUM_FAILURE_ITEMS \
1761
(((0 \
1762
? 0 : highest_active_reg - lowest_active_reg + 1) \
1763
* NUM_REG_ITEMS) \
1764
+ NUM_NONREG_ITEMS)
1765
1766
/* How many items can still be added to the stack without overflowing it. */
1767
# define REMAINING_AVAIL_SLOTS ((fail_stack).size - (fail_stack).avail)
1768
# endif /* not DEFINED_ONCE */
1769
1770
1771
/* Pops what PUSH_FAIL_STACK pushes.
1772
1773
We restore into the parameters, all of which should be lvalues:
1774
STR -- the saved data position.
1775
PAT -- the saved pattern position.
1776
LOW_REG, HIGH_REG -- the highest and lowest active registers.
1777
REGSTART, REGEND -- arrays of string positions.
1778
REG_INFO -- array of information about each subexpression.
1779
1780
Also assumes the variables `fail_stack' and (if debugging), `bufp',
1781
`pend', `string1', `size1', `string2', and `size2'. */
1782
# define POP_FAILURE_POINT(str, pat, low_reg, high_reg, regstart, regend, reg_info)\
1783
{ \
1784
DEBUG_STATEMENT (unsigned failure_id;) \
1785
active_reg_t this_reg; \
1786
const UCHAR_T *string_temp; \
1787
\
1788
assert (!FAIL_STACK_EMPTY ()); \
1789
\
1790
/* Remove failure points and point to how many regs pushed. */ \
1791
DEBUG_PRINT1 ("POP_FAILURE_POINT:\n"); \
1792
DEBUG_PRINT2 (" Before pop, next avail: %d\n", fail_stack.avail); \
1793
DEBUG_PRINT2 (" size: %d\n", fail_stack.size); \
1794
\
1795
assert (fail_stack.avail >= NUM_NONREG_ITEMS); \
1796
\
1797
DEBUG_POP (&failure_id); \
1798
DEBUG_PRINT2 (" Popping failure id: %u\n", failure_id); \
1799
\
1800
/* If the saved string location is NULL, it came from an \
1801
on_failure_keep_string_jump opcode, and we want to throw away the \
1802
saved NULL, thus retaining our current position in the string. */ \
1803
string_temp = POP_FAILURE_POINTER (); \
1804
if (string_temp != NULL) \
1805
str = (const CHAR_T *) string_temp; \
1806
\
1807
DEBUG_PRINT2 (" Popping string %p: `", str); \
1808
DEBUG_PRINT_DOUBLE_STRING (str, string1, size1, string2, size2); \
1809
DEBUG_PRINT1 ("'\n"); \
1810
\
1811
pat = (UCHAR_T *) POP_FAILURE_POINTER (); \
1812
DEBUG_PRINT2 (" Popping pattern %p:\n", pat); \
1813
DEBUG_PRINT_COMPILED_PATTERN (bufp, pat, pend); \
1814
\
1815
/* Restore register info. */ \
1816
high_reg = (active_reg_t) POP_FAILURE_INT (); \
1817
DEBUG_PRINT2 (" Popping high active reg: %ld\n", high_reg); \
1818
\
1819
low_reg = (active_reg_t) POP_FAILURE_INT (); \
1820
DEBUG_PRINT2 (" Popping low active reg: %ld\n", low_reg); \
1821
\
1822
if (1) \
1823
for (this_reg = high_reg; this_reg >= low_reg; this_reg--) \
1824
{ \
1825
DEBUG_PRINT2 (" Popping reg: %ld\n", this_reg); \
1826
\
1827
reg_info[this_reg].word = POP_FAILURE_ELT (); \
1828
DEBUG_PRINT2 (" info: %p\n", \
1829
reg_info[this_reg].word.pointer); \
1830
\
1831
regend[this_reg] = (const CHAR_T *) POP_FAILURE_POINTER (); \
1832
DEBUG_PRINT2 (" end: %p\n", regend[this_reg]); \
1833
\
1834
regstart[this_reg] = (const CHAR_T *) POP_FAILURE_POINTER (); \
1835
DEBUG_PRINT2 (" start: %p\n", regstart[this_reg]); \
1836
} \
1837
else \
1838
{ \
1839
for (this_reg = highest_active_reg; this_reg > high_reg; this_reg--) \
1840
{ \
1841
reg_info[this_reg].word.integer = 0; \
1842
regend[this_reg] = 0; \
1843
regstart[this_reg] = 0; \
1844
} \
1845
highest_active_reg = high_reg; \
1846
} \
1847
\
1848
set_regs_matched_done = 0; \
1849
DEBUG_STATEMENT (nfailure_points_popped++); \
1850
} /* POP_FAILURE_POINT */
1851
1852
/* Structure for per-register (a.k.a. per-group) information.
1853
Other register information, such as the
1854
starting and ending positions (which are addresses), and the list of
1855
inner groups (which is a bits list) are maintained in separate
1856
variables.
1857
1858
We are making a (strictly speaking) nonportable assumption here: that
1859
the compiler will pack our bit fields into something that fits into
1860
the type of `word', i.e., is something that fits into one item on the
1861
failure stack. */
1862
1863
1864
/* Declarations and macros for re_match_2. */
1865
1866
typedef union
1867
{
1868
PREFIX(fail_stack_elt_t) word;
1869
struct
1870
{
1871
/* This field is one if this group can match the empty string,
1872
zero if not. If not yet determined, `MATCH_NULL_UNSET_VALUE'. */
1873
# define MATCH_NULL_UNSET_VALUE 3
1874
unsigned match_null_string_p : 2;
1875
unsigned is_active : 1;
1876
unsigned matched_something : 1;
1877
unsigned ever_matched_something : 1;
1878
} bits;
1879
} PREFIX(register_info_type);
1880
1881
# ifndef DEFINED_ONCE
1882
# define REG_MATCH_NULL_STRING_P(R) ((R).bits.match_null_string_p)
1883
# define IS_ACTIVE(R) ((R).bits.is_active)
1884
# define MATCHED_SOMETHING(R) ((R).bits.matched_something)
1885
# define EVER_MATCHED_SOMETHING(R) ((R).bits.ever_matched_something)
1886
1887
1888
/* Call this when have matched a real character; it sets `matched' flags
1889
for the subexpressions which we are currently inside. Also records
1890
that those subexprs have matched. */
1891
# define SET_REGS_MATCHED() \
1892
do \
1893
{ \
1894
if (!set_regs_matched_done) \
1895
{ \
1896
active_reg_t r; \
1897
set_regs_matched_done = 1; \
1898
for (r = lowest_active_reg; r <= highest_active_reg; r++) \
1899
{ \
1900
MATCHED_SOMETHING (reg_info[r]) \
1901
= EVER_MATCHED_SOMETHING (reg_info[r]) \
1902
= 1; \
1903
} \
1904
} \
1905
} \
1906
while (0)
1907
# endif /* not DEFINED_ONCE */
1908
1909
/* Registers are set to a sentinel when they haven't yet matched. */
1910
static CHAR_T PREFIX(reg_unset_dummy);
1911
# define REG_UNSET_VALUE (&PREFIX(reg_unset_dummy))
1912
# define REG_UNSET(e) ((e) == REG_UNSET_VALUE)
1913
1914
/* Subroutine declarations and macros for regex_compile. */
1915
static void PREFIX(store_op1) _RE_ARGS ((re_opcode_t op, UCHAR_T *loc, int arg));
1916
static void PREFIX(store_op2) _RE_ARGS ((re_opcode_t op, UCHAR_T *loc,
1917
int arg1, int arg2));
1918
static void PREFIX(insert_op1) _RE_ARGS ((re_opcode_t op, UCHAR_T *loc,
1919
int arg, UCHAR_T *end));
1920
static void PREFIX(insert_op2) _RE_ARGS ((re_opcode_t op, UCHAR_T *loc,
1921
int arg1, int arg2, UCHAR_T *end));
1922
static boolean PREFIX(at_begline_loc_p) _RE_ARGS ((const CHAR_T *pattern,
1923
const CHAR_T *p,
1924
reg_syntax_t syntax));
1925
static boolean PREFIX(at_endline_loc_p) _RE_ARGS ((const CHAR_T *p,
1926
const CHAR_T *pend,
1927
reg_syntax_t syntax));
1928
# ifdef WCHAR
1929
static reg_errcode_t wcs_compile_range _RE_ARGS ((CHAR_T range_start,
1930
const CHAR_T **p_ptr,
1931
const CHAR_T *pend,
1932
char *translate,
1933
reg_syntax_t syntax,
1934
UCHAR_T *b,
1935
CHAR_T *char_set));
1936
static void insert_space _RE_ARGS ((int num, CHAR_T *loc, CHAR_T *end));
1937
# else /* BYTE */
1938
static reg_errcode_t byte_compile_range _RE_ARGS ((unsigned int range_start,
1939
const char **p_ptr,
1940
const char *pend,
1941
char *translate,
1942
reg_syntax_t syntax,
1943
unsigned char *b));
1944
# endif /* WCHAR */
1945
1946
/* Fetch the next character in the uncompiled pattern---translating it
1947
if necessary. Also cast from a signed character in the constant
1948
string passed to us by the user to an unsigned char that we can use
1949
as an array index (in, e.g., `translate'). */
1950
/* ifdef MBS_SUPPORT, we translate only if character <= 0xff,
1951
because it is impossible to allocate 4GB array for some encodings
1952
which have 4 byte character_set like UCS4. */
1953
# ifndef PATFETCH
1954
# ifdef WCHAR
1955
# define PATFETCH(c) \
1956
do {if (p == pend) return REG_EEND; \
1957
c = (UCHAR_T) *p++; \
1958
if (translate && (c <= 0xff)) c = (UCHAR_T) translate[c]; \
1959
} while (0)
1960
# else /* BYTE */
1961
# define PATFETCH(c) \
1962
do {if (p == pend) return REG_EEND; \
1963
c = (unsigned char) *p++; \
1964
if (translate) c = (unsigned char) translate[c]; \
1965
} while (0)
1966
# endif /* WCHAR */
1967
# endif
1968
1969
/* Fetch the next character in the uncompiled pattern, with no
1970
translation. */
1971
# define PATFETCH_RAW(c) \
1972
do {if (p == pend) return REG_EEND; \
1973
c = (UCHAR_T) *p++; \
1974
} while (0)
1975
1976
/* Go backwards one character in the pattern. */
1977
# define PATUNFETCH p--
1978
1979
1980
/* If `translate' is non-null, return translate[D], else just D. We
1981
cast the subscript to translate because some data is declared as
1982
`char *', to avoid warnings when a string constant is passed. But
1983
when we use a character as a subscript we must make it unsigned. */
1984
/* ifdef MBS_SUPPORT, we translate only if character <= 0xff,
1985
because it is impossible to allocate 4GB array for some encodings
1986
which have 4 byte character_set like UCS4. */
1987
1988
# ifndef TRANSLATE
1989
# ifdef WCHAR
1990
# define TRANSLATE(d) \
1991
((translate && ((UCHAR_T) (d)) <= 0xff) \
1992
? (char) translate[(unsigned char) (d)] : (d))
1993
# else /* BYTE */
1994
# define TRANSLATE(d) \
1995
(translate ? (char) translate[(unsigned char) (d)] : (d))
1996
# endif /* WCHAR */
1997
# endif
1998
1999
2000
/* Macros for outputting the compiled pattern into `buffer'. */
2001
2002
/* If the buffer isn't allocated when it comes in, use this. */
2003
# define INIT_BUF_SIZE (32 * sizeof(UCHAR_T))
2004
2005
/* Make sure we have at least N more bytes of space in buffer. */
2006
# ifdef WCHAR
2007
# define GET_BUFFER_SPACE(n) \
2008
while (((unsigned long)b - (unsigned long)COMPILED_BUFFER_VAR \
2009
+ (n)*sizeof(CHAR_T)) > bufp->allocated) \
2010
EXTEND_BUFFER ()
2011
# else /* BYTE */
2012
# define GET_BUFFER_SPACE(n) \
2013
while ((unsigned long) (b - bufp->buffer + (n)) > bufp->allocated) \
2014
EXTEND_BUFFER ()
2015
# endif /* WCHAR */
2016
2017
/* Make sure we have one more byte of buffer space and then add C to it. */
2018
# define BUF_PUSH(c) \
2019
do { \
2020
GET_BUFFER_SPACE (1); \
2021
*b++ = (UCHAR_T) (c); \
2022
} while (0)
2023
2024
2025
/* Ensure we have two more bytes of buffer space and then append C1 and C2. */
2026
# define BUF_PUSH_2(c1, c2) \
2027
do { \
2028
GET_BUFFER_SPACE (2); \
2029
*b++ = (UCHAR_T) (c1); \
2030
*b++ = (UCHAR_T) (c2); \
2031
} while (0)
2032
2033
2034
/* As with BUF_PUSH_2, except for three bytes. */
2035
# define BUF_PUSH_3(c1, c2, c3) \
2036
do { \
2037
GET_BUFFER_SPACE (3); \
2038
*b++ = (UCHAR_T) (c1); \
2039
*b++ = (UCHAR_T) (c2); \
2040
*b++ = (UCHAR_T) (c3); \
2041
} while (0)
2042
2043
/* Store a jump with opcode OP at LOC to location TO. We store a
2044
relative address offset by the three bytes the jump itself occupies. */
2045
# define STORE_JUMP(op, loc, to) \
2046
PREFIX(store_op1) (op, loc, (int) ((to) - (loc) - (1 + OFFSET_ADDRESS_SIZE)))
2047
2048
/* Likewise, for a two-argument jump. */
2049
# define STORE_JUMP2(op, loc, to, arg) \
2050
PREFIX(store_op2) (op, loc, (int) ((to) - (loc) - (1 + OFFSET_ADDRESS_SIZE)), arg)
2051
2052
/* Like `STORE_JUMP', but for inserting. Assume `b' is the buffer end. */
2053
# define INSERT_JUMP(op, loc, to) \
2054
PREFIX(insert_op1) (op, loc, (int) ((to) - (loc) - (1 + OFFSET_ADDRESS_SIZE)), b)
2055
2056
/* Like `STORE_JUMP2', but for inserting. Assume `b' is the buffer end. */
2057
# define INSERT_JUMP2(op, loc, to, arg) \
2058
PREFIX(insert_op2) (op, loc, (int) ((to) - (loc) - (1 + OFFSET_ADDRESS_SIZE)),\
2059
arg, b)
2060
2061
/* This is not an arbitrary limit: the arguments which represent offsets
2062
into the pattern are two bytes long. So if 2^16 bytes turns out to
2063
be too small, many things would have to change. */
2064
/* Any other compiler which, like MSC, has allocation limit below 2^16
2065
bytes will have to use approach similar to what was done below for
2066
MSC and drop MAX_BUF_SIZE a bit. Otherwise you may end up
2067
reallocating to 0 bytes. Such thing is not going to work too well.
2068
You have been warned!! */
2069
# ifndef DEFINED_ONCE
2070
# if defined _MSC_VER && !defined WIN32
2071
/* Microsoft C 16-bit versions limit malloc to approx 65512 bytes.
2072
The REALLOC define eliminates a flurry of conversion warnings,
2073
but is not required. */
2074
# define MAX_BUF_SIZE 65500L
2075
# define REALLOC(p,s) realloc ((p), (size_t) (s))
2076
# else
2077
# define MAX_BUF_SIZE (1L << 16)
2078
# define REALLOC(p,s) realloc ((p), (s))
2079
# endif
2080
2081
/* Extend the buffer by twice its current size via realloc and
2082
reset the pointers that pointed into the old block to point to the
2083
correct places in the new one. If extending the buffer results in it
2084
being larger than MAX_BUF_SIZE, then flag memory exhausted. */
2085
# if __BOUNDED_POINTERS__
2086
# define SET_HIGH_BOUND(P) (__ptrhigh (P) = __ptrlow (P) + bufp->allocated)
2087
# define MOVE_BUFFER_POINTER(P) \
2088
(__ptrlow (P) += incr, SET_HIGH_BOUND (P), __ptrvalue (P) += incr)
2089
# define ELSE_EXTEND_BUFFER_HIGH_BOUND \
2090
else \
2091
{ \
2092
SET_HIGH_BOUND (b); \
2093
SET_HIGH_BOUND (begalt); \
2094
if (fixup_alt_jump) \
2095
SET_HIGH_BOUND (fixup_alt_jump); \
2096
if (laststart) \
2097
SET_HIGH_BOUND (laststart); \
2098
if (pending_exact) \
2099
SET_HIGH_BOUND (pending_exact); \
2100
}
2101
# else
2102
# define MOVE_BUFFER_POINTER(P) (P) += incr
2103
# define ELSE_EXTEND_BUFFER_HIGH_BOUND
2104
# endif
2105
# endif /* not DEFINED_ONCE */
2106
2107
# ifdef WCHAR
2108
# define EXTEND_BUFFER() \
2109
do { \
2110
UCHAR_T *old_buffer = COMPILED_BUFFER_VAR; \
2111
int wchar_count; \
2112
if (bufp->allocated + sizeof(UCHAR_T) > MAX_BUF_SIZE) \
2113
return REG_ESIZE; \
2114
bufp->allocated <<= 1; \
2115
if (bufp->allocated > MAX_BUF_SIZE) \
2116
bufp->allocated = MAX_BUF_SIZE; \
2117
/* How many characters the new buffer can have? */ \
2118
wchar_count = bufp->allocated / sizeof(UCHAR_T); \
2119
if (wchar_count == 0) wchar_count = 1; \
2120
/* Truncate the buffer to CHAR_T align. */ \
2121
bufp->allocated = wchar_count * sizeof(UCHAR_T); \
2122
RETALLOC (COMPILED_BUFFER_VAR, wchar_count, UCHAR_T); \
2123
bufp->buffer = (char*)COMPILED_BUFFER_VAR; \
2124
if (COMPILED_BUFFER_VAR == NULL) \
2125
return REG_ESPACE; \
2126
/* If the buffer moved, move all the pointers into it. */ \
2127
if (old_buffer != COMPILED_BUFFER_VAR) \
2128
{ \
2129
int incr = COMPILED_BUFFER_VAR - old_buffer; \
2130
MOVE_BUFFER_POINTER (b); \
2131
MOVE_BUFFER_POINTER (begalt); \
2132
if (fixup_alt_jump) \
2133
MOVE_BUFFER_POINTER (fixup_alt_jump); \
2134
if (laststart) \
2135
MOVE_BUFFER_POINTER (laststart); \
2136
if (pending_exact) \
2137
MOVE_BUFFER_POINTER (pending_exact); \
2138
} \
2139
ELSE_EXTEND_BUFFER_HIGH_BOUND \
2140
} while (0)
2141
# else /* BYTE */
2142
# define EXTEND_BUFFER() \
2143
do { \
2144
UCHAR_T *old_buffer = COMPILED_BUFFER_VAR; \
2145
if (bufp->allocated == MAX_BUF_SIZE) \
2146
return REG_ESIZE; \
2147
bufp->allocated <<= 1; \
2148
if (bufp->allocated > MAX_BUF_SIZE) \
2149
bufp->allocated = MAX_BUF_SIZE; \
2150
bufp->buffer = (UCHAR_T *) REALLOC (COMPILED_BUFFER_VAR, \
2151
bufp->allocated); \
2152
if (COMPILED_BUFFER_VAR == NULL) \
2153
return REG_ESPACE; \
2154
/* If the buffer moved, move all the pointers into it. */ \
2155
if (old_buffer != COMPILED_BUFFER_VAR) \
2156
{ \
2157
int incr = COMPILED_BUFFER_VAR - old_buffer; \
2158
MOVE_BUFFER_POINTER (b); \
2159
MOVE_BUFFER_POINTER (begalt); \
2160
if (fixup_alt_jump) \
2161
MOVE_BUFFER_POINTER (fixup_alt_jump); \
2162
if (laststart) \
2163
MOVE_BUFFER_POINTER (laststart); \
2164
if (pending_exact) \
2165
MOVE_BUFFER_POINTER (pending_exact); \
2166
} \
2167
ELSE_EXTEND_BUFFER_HIGH_BOUND \
2168
} while (0)
2169
# endif /* WCHAR */
2170
2171
# ifndef DEFINED_ONCE
2172
/* Since we have one byte reserved for the register number argument to
2173
{start,stop}_memory, the maximum number of groups we can report
2174
things about is what fits in that byte. */
2175
# define MAX_REGNUM 255
2176
2177
/* But patterns can have more than `MAX_REGNUM' registers. We just
2178
ignore the excess. */
2179
typedef unsigned regnum_t;
2180
2181
2182
/* Macros for the compile stack. */
2183
2184
/* Since offsets can go either forwards or backwards, this type needs to
2185
be able to hold values from -(MAX_BUF_SIZE - 1) to MAX_BUF_SIZE - 1. */
2186
/* int may be not enough when sizeof(int) == 2. */
2187
typedef long pattern_offset_t;
2188
2189
typedef struct
2190
{
2191
pattern_offset_t begalt_offset;
2192
pattern_offset_t fixup_alt_jump;
2193
pattern_offset_t inner_group_offset;
2194
pattern_offset_t laststart_offset;
2195
regnum_t regnum;
2196
} compile_stack_elt_t;
2197
2198
2199
typedef struct
2200
{
2201
compile_stack_elt_t *stack;
2202
unsigned size;
2203
unsigned avail; /* Offset of next open position. */
2204
} compile_stack_type;
2205
2206
2207
# define INIT_COMPILE_STACK_SIZE 32
2208
2209
# define COMPILE_STACK_EMPTY (compile_stack.avail == 0)
2210
# define COMPILE_STACK_FULL (compile_stack.avail == compile_stack.size)
2211
2212
/* The next available element. */
2213
# define COMPILE_STACK_TOP (compile_stack.stack[compile_stack.avail])
2214
2215
# endif /* not DEFINED_ONCE */
2216
2217
/* Set the bit for character C in a list. */
2218
# ifndef DEFINED_ONCE
2219
# define SET_LIST_BIT(c) \
2220
(b[((unsigned char) (c)) / BYTEWIDTH] \
2221
|= 1 << (((unsigned char) c) % BYTEWIDTH))
2222
# endif /* DEFINED_ONCE */
2223
2224
/* Get the next unsigned number in the uncompiled pattern. */
2225
# define GET_UNSIGNED_NUMBER(num) \
2226
{ \
2227
while (p != pend) \
2228
{ \
2229
PATFETCH (c); \
2230
if (c < '0' || c > '9') \
2231
break; \
2232
if (num <= RE_DUP_MAX) \
2233
{ \
2234
if (num < 0) \
2235
num = 0; \
2236
num = num * 10 + c - '0'; \
2237
} \
2238
} \
2239
}
2240
2241
# ifndef DEFINED_ONCE
2242
# if defined _LIBC || WIDE_CHAR_SUPPORT
2243
/* The GNU C library provides support for user-defined character classes
2244
and the functions from ISO C amendement 1. */
2245
# ifdef CHARCLASS_NAME_MAX
2246
# define CHAR_CLASS_MAX_LENGTH CHARCLASS_NAME_MAX
2247
# else
2248
/* This shouldn't happen but some implementation might still have this
2249
problem. Use a reasonable default value. */
2250
# define CHAR_CLASS_MAX_LENGTH 256
2251
# endif
2252
2253
# ifdef _LIBC
2254
# define IS_CHAR_CLASS(string) __wctype (string)
2255
# else
2256
# define IS_CHAR_CLASS(string) wctype (string)
2257
# endif
2258
# else
2259
# define CHAR_CLASS_MAX_LENGTH 6 /* Namely, `xdigit'. */
2260
2261
# define IS_CHAR_CLASS(string) \
2262
(STREQ (string, "alpha") || STREQ (string, "upper") \
2263
|| STREQ (string, "lower") || STREQ (string, "digit") \
2264
|| STREQ (string, "alnum") || STREQ (string, "xdigit") \
2265
|| STREQ (string, "space") || STREQ (string, "print") \
2266
|| STREQ (string, "punct") || STREQ (string, "graph") \
2267
|| STREQ (string, "cntrl") || STREQ (string, "blank"))
2268
# endif
2269
# endif /* DEFINED_ONCE */
2270
2271
# ifndef MATCH_MAY_ALLOCATE
2272
2273
/* If we cannot allocate large objects within re_match_2_internal,
2274
we make the fail stack and register vectors global.
2275
The fail stack, we grow to the maximum size when a regexp
2276
is compiled.
2277
The register vectors, we adjust in size each time we
2278
compile a regexp, according to the number of registers it needs. */
2279
2280
static PREFIX(fail_stack_type) fail_stack;
2281
2282
/* Size with which the following vectors are currently allocated.
2283
That is so we can make them bigger as needed,
2284
but never make them smaller. */
2285
# ifdef DEFINED_ONCE
2286
static int regs_allocated_size;
2287
2288
static const char ** regstart, ** regend;
2289
static const char ** old_regstart, ** old_regend;
2290
static const char **best_regstart, **best_regend;
2291
static const char **reg_dummy;
2292
# endif /* DEFINED_ONCE */
2293
2294
static PREFIX(register_info_type) *PREFIX(reg_info);
2295
static PREFIX(register_info_type) *PREFIX(reg_info_dummy);
2296
2297
/* Make the register vectors big enough for NUM_REGS registers,
2298
but don't make them smaller. */
2299
2300
static void
2301
PREFIX(regex_grow_registers) (num_regs)
2302
int num_regs;
2303
{
2304
if (num_regs > regs_allocated_size)
2305
{
2306
RETALLOC_IF (regstart, num_regs, const char *);
2307
RETALLOC_IF (regend, num_regs, const char *);
2308
RETALLOC_IF (old_regstart, num_regs, const char *);
2309
RETALLOC_IF (old_regend, num_regs, const char *);
2310
RETALLOC_IF (best_regstart, num_regs, const char *);
2311
RETALLOC_IF (best_regend, num_regs, const char *);
2312
RETALLOC_IF (PREFIX(reg_info), num_regs, PREFIX(register_info_type));
2313
RETALLOC_IF (reg_dummy, num_regs, const char *);
2314
RETALLOC_IF (PREFIX(reg_info_dummy), num_regs, PREFIX(register_info_type));
2315
2316
regs_allocated_size = num_regs;
2317
}
2318
}
2319
2320
# endif /* not MATCH_MAY_ALLOCATE */
2321
2322
# ifndef DEFINED_ONCE
2323
static boolean group_in_compile_stack _RE_ARGS ((compile_stack_type
2324
compile_stack,
2325
regnum_t regnum));
2326
# endif /* not DEFINED_ONCE */
2327
2328
/* `regex_compile' compiles PATTERN (of length SIZE) according to SYNTAX.
2329
Returns one of error codes defined in `regex.h', or zero for success.
2330
2331
Assumes the `allocated' (and perhaps `buffer') and `translate'
2332
fields are set in BUFP on entry.
2333
2334
If it succeeds, results are put in BUFP (if it returns an error, the
2335
contents of BUFP are undefined):
2336
`buffer' is the compiled pattern;
2337
`syntax' is set to SYNTAX;
2338
`used' is set to the length of the compiled pattern;
2339
`fastmap_accurate' is zero;
2340
`re_nsub' is the number of subexpressions in PATTERN;
2341
`not_bol' and `not_eol' are zero;
2342
2343
The `fastmap' and `newline_anchor' fields are neither
2344
examined nor set. */
2345
2346
/* Return, freeing storage we allocated. */
2347
# ifdef WCHAR
2348
# define FREE_STACK_RETURN(value) \
2349
return (free(pattern), free(mbs_offset), free(is_binary), free (compile_stack.stack), value)
2350
# else
2351
# define FREE_STACK_RETURN(value) \
2352
return (free (compile_stack.stack), value)
2353
# endif /* WCHAR */
2354
2355
static reg_errcode_t
2356
PREFIX(regex_compile) (ARG_PREFIX(pattern), ARG_PREFIX(size), syntax, bufp)
2357
const char *ARG_PREFIX(pattern);
2358
size_t ARG_PREFIX(size);
2359
reg_syntax_t syntax;
2360
struct re_pattern_buffer *bufp;
2361
{
2362
/* We fetch characters from PATTERN here. Even though PATTERN is
2363
`char *' (i.e., signed), we declare these variables as unsigned, so
2364
they can be reliably used as array indices. */
2365
register UCHAR_T c, c1;
2366
2367
#ifdef WCHAR
2368
/* A temporary space to keep wchar_t pattern and compiled pattern. */
2369
CHAR_T *pattern, *COMPILED_BUFFER_VAR;
2370
size_t size;
2371
/* offset buffer for optimization. See convert_mbs_to_wc. */
2372
int *mbs_offset = NULL;
2373
/* It hold whether each wchar_t is binary data or not. */
2374
char *is_binary = NULL;
2375
/* A flag whether exactn is handling binary data or not. */
2376
char is_exactn_bin = FALSE;
2377
#endif /* WCHAR */
2378
2379
/* A random temporary spot in PATTERN. */
2380
const CHAR_T *p1;
2381
2382
/* Points to the end of the buffer, where we should append. */
2383
register UCHAR_T *b;
2384
2385
/* Keeps track of unclosed groups. */
2386
compile_stack_type compile_stack;
2387
2388
/* Points to the current (ending) position in the pattern. */
2389
#ifdef WCHAR
2390
const CHAR_T *p;
2391
const CHAR_T *pend;
2392
#else /* BYTE */
2393
const CHAR_T *p = pattern;
2394
const CHAR_T *pend = pattern + size;
2395
#endif /* WCHAR */
2396
2397
/* How to translate the characters in the pattern. */
2398
RE_TRANSLATE_TYPE translate = bufp->translate;
2399
2400
/* Address of the count-byte of the most recently inserted `exactn'
2401
command. This makes it possible to tell if a new exact-match
2402
character can be added to that command or if the character requires
2403
a new `exactn' command. */
2404
UCHAR_T *pending_exact = 0;
2405
2406
/* Address of start of the most recently finished expression.
2407
This tells, e.g., postfix * where to find the start of its
2408
operand. Reset at the beginning of groups and alternatives. */
2409
UCHAR_T *laststart = 0;
2410
2411
/* Address of beginning of regexp, or inside of last group. */
2412
UCHAR_T *begalt;
2413
2414
/* Address of the place where a forward jump should go to the end of
2415
the containing expression. Each alternative of an `or' -- except the
2416
last -- ends with a forward jump of this sort. */
2417
UCHAR_T *fixup_alt_jump = 0;
2418
2419
/* Counts open-groups as they are encountered. Remembered for the
2420
matching close-group on the compile stack, so the same register
2421
number is put in the stop_memory as the start_memory. */
2422
regnum_t regnum = 0;
2423
2424
#ifdef WCHAR
2425
/* Initialize the wchar_t PATTERN and offset_buffer. */
2426
p = pend = pattern = TALLOC(csize + 1, CHAR_T);
2427
mbs_offset = TALLOC(csize + 1, int);
2428
is_binary = TALLOC(csize + 1, char);
2429
if (pattern == NULL || mbs_offset == NULL || is_binary == NULL)
2430
{
2431
free(pattern);
2432
free(mbs_offset);
2433
free(is_binary);
2434
return REG_ESPACE;
2435
}
2436
pattern[csize] = L'\0'; /* sentinel */
2437
size = convert_mbs_to_wcs(pattern, cpattern, csize, mbs_offset, is_binary);
2438
pend = p + size;
2439
if (size < 0)
2440
{
2441
free(pattern);
2442
free(mbs_offset);
2443
free(is_binary);
2444
return REG_BADPAT;
2445
}
2446
#endif
2447
2448
#ifdef DEBUG
2449
DEBUG_PRINT1 ("\nCompiling pattern: ");
2450
if (debug)
2451
{
2452
unsigned debug_count;
2453
2454
for (debug_count = 0; debug_count < size; debug_count++)
2455
PUT_CHAR (pattern[debug_count]);
2456
putchar ('\n');
2457
}
2458
#endif /* DEBUG */
2459
2460
/* Initialize the compile stack. */
2461
compile_stack.stack = TALLOC (INIT_COMPILE_STACK_SIZE, compile_stack_elt_t);
2462
if (compile_stack.stack == NULL)
2463
{
2464
#ifdef WCHAR
2465
free(pattern);
2466
free(mbs_offset);
2467
free(is_binary);
2468
#endif
2469
return REG_ESPACE;
2470
}
2471
2472
compile_stack.size = INIT_COMPILE_STACK_SIZE;
2473
compile_stack.avail = 0;
2474
2475
/* Initialize the pattern buffer. */
2476
bufp->syntax = syntax;
2477
bufp->fastmap_accurate = 0;
2478
bufp->not_bol = bufp->not_eol = 0;
2479
2480
/* Set `used' to zero, so that if we return an error, the pattern
2481
printer (for debugging) will think there's no pattern. We reset it
2482
at the end. */
2483
bufp->used = 0;
2484
2485
/* Always count groups, whether or not bufp->no_sub is set. */
2486
bufp->re_nsub = 0;
2487
2488
#if !defined emacs && !defined SYNTAX_TABLE
2489
/* Initialize the syntax table. */
2490
init_syntax_once ();
2491
#endif
2492
2493
if (bufp->allocated == 0)
2494
{
2495
if (bufp->buffer)
2496
{ /* If zero allocated, but buffer is non-null, try to realloc
2497
enough space. This loses if buffer's address is bogus, but
2498
that is the user's responsibility. */
2499
#ifdef WCHAR
2500
/* Free bufp->buffer and allocate an array for wchar_t pattern
2501
buffer. */
2502
free(bufp->buffer);
2503
COMPILED_BUFFER_VAR = TALLOC (INIT_BUF_SIZE/sizeof(UCHAR_T),
2504
UCHAR_T);
2505
#else
2506
RETALLOC (COMPILED_BUFFER_VAR, INIT_BUF_SIZE, UCHAR_T);
2507
#endif /* WCHAR */
2508
}
2509
else
2510
{ /* Caller did not allocate a buffer. Do it for them. */
2511
COMPILED_BUFFER_VAR = TALLOC (INIT_BUF_SIZE / sizeof(UCHAR_T),
2512
UCHAR_T);
2513
}
2514
2515
if (!COMPILED_BUFFER_VAR) FREE_STACK_RETURN (REG_ESPACE);
2516
#ifdef WCHAR
2517
bufp->buffer = (char*)COMPILED_BUFFER_VAR;
2518
#endif /* WCHAR */
2519
bufp->allocated = INIT_BUF_SIZE;
2520
}
2521
#ifdef WCHAR
2522
else
2523
COMPILED_BUFFER_VAR = (UCHAR_T*) bufp->buffer;
2524
#endif
2525
2526
begalt = b = COMPILED_BUFFER_VAR;
2527
2528
/* Loop through the uncompiled pattern until we're at the end. */
2529
while (p != pend)
2530
{
2531
PATFETCH (c);
2532
2533
switch (c)
2534
{
2535
case '^':
2536
{
2537
if ( /* If at start of pattern, it's an operator. */
2538
p == pattern + 1
2539
/* If context independent, it's an operator. */
2540
|| syntax & RE_CONTEXT_INDEP_ANCHORS
2541
/* Otherwise, depends on what's come before. */
2542
|| PREFIX(at_begline_loc_p) (pattern, p, syntax))
2543
BUF_PUSH (begline);
2544
else
2545
goto normal_char;
2546
}
2547
break;
2548
2549
2550
case '$':
2551
{
2552
if ( /* If at end of pattern, it's an operator. */
2553
p == pend
2554
/* If context independent, it's an operator. */
2555
|| syntax & RE_CONTEXT_INDEP_ANCHORS
2556
/* Otherwise, depends on what's next. */
2557
|| PREFIX(at_endline_loc_p) (p, pend, syntax))
2558
BUF_PUSH (endline);
2559
else
2560
goto normal_char;
2561
}
2562
break;
2563
2564
2565
case '+':
2566
case '?':
2567
if ((syntax & RE_BK_PLUS_QM)
2568
|| (syntax & RE_LIMITED_OPS))
2569
goto normal_char;
2570
handle_plus:
2571
case '*':
2572
/* If there is no previous pattern... */
2573
if (!laststart)
2574
{
2575
if (syntax & RE_CONTEXT_INVALID_OPS)
2576
FREE_STACK_RETURN (REG_BADRPT);
2577
else if (!(syntax & RE_CONTEXT_INDEP_OPS))
2578
goto normal_char;
2579
}
2580
2581
{
2582
/* Are we optimizing this jump? */
2583
boolean keep_string_p = false;
2584
2585
/* 1 means zero (many) matches is allowed. */
2586
char zero_times_ok = 0, many_times_ok = 0;
2587
2588
/* If there is a sequence of repetition chars, collapse it
2589
down to just one (the right one). We can't combine
2590
interval operators with these because of, e.g., `a{2}*',
2591
which should only match an even number of `a's. */
2592
2593
for (;;)
2594
{
2595
zero_times_ok |= c != '+';
2596
many_times_ok |= c != '?';
2597
2598
if (p == pend)
2599
break;
2600
2601
PATFETCH (c);
2602
2603
if (c == '*'
2604
|| (!(syntax & RE_BK_PLUS_QM) && (c == '+' || c == '?')))
2605
;
2606
2607
else if (syntax & RE_BK_PLUS_QM && c == '\\')
2608
{
2609
if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
2610
2611
PATFETCH (c1);
2612
if (!(c1 == '+' || c1 == '?'))
2613
{
2614
PATUNFETCH;
2615
PATUNFETCH;
2616
break;
2617
}
2618
2619
c = c1;
2620
}
2621
else
2622
{
2623
PATUNFETCH;
2624
break;
2625
}
2626
2627
/* If we get here, we found another repeat character. */
2628
}
2629
2630
/* Star, etc. applied to an empty pattern is equivalent
2631
to an empty pattern. */
2632
if (!laststart)
2633
break;
2634
2635
/* Now we know whether or not zero matches is allowed
2636
and also whether or not two or more matches is allowed. */
2637
if (many_times_ok)
2638
{ /* More than one repetition is allowed, so put in at the
2639
end a backward relative jump from `b' to before the next
2640
jump we're going to put in below (which jumps from
2641
laststart to after this jump).
2642
2643
But if we are at the `*' in the exact sequence `.*\n',
2644
insert an unconditional jump backwards to the .,
2645
instead of the beginning of the loop. This way we only
2646
push a failure point once, instead of every time
2647
through the loop. */
2648
assert (p - 1 > pattern);
2649
2650
/* Allocate the space for the jump. */
2651
GET_BUFFER_SPACE (1 + OFFSET_ADDRESS_SIZE);
2652
2653
/* We know we are not at the first character of the pattern,
2654
because laststart was nonzero. And we've already
2655
incremented `p', by the way, to be the character after
2656
the `*'. Do we have to do something analogous here
2657
for null bytes, because of RE_DOT_NOT_NULL? */
2658
if (TRANSLATE (*(p - 2)) == TRANSLATE ('.')
2659
&& zero_times_ok
2660
&& p < pend && TRANSLATE (*p) == TRANSLATE ('\n')
2661
&& !(syntax & RE_DOT_NEWLINE))
2662
{ /* We have .*\n. */
2663
STORE_JUMP (jump, b, laststart);
2664
keep_string_p = true;
2665
}
2666
else
2667
/* Anything else. */
2668
STORE_JUMP (maybe_pop_jump, b, laststart -
2669
(1 + OFFSET_ADDRESS_SIZE));
2670
2671
/* We've added more stuff to the buffer. */
2672
b += 1 + OFFSET_ADDRESS_SIZE;
2673
}
2674
2675
/* On failure, jump from laststart to b + 3, which will be the
2676
end of the buffer after this jump is inserted. */
2677
/* ifdef WCHAR, 'b + 1 + OFFSET_ADDRESS_SIZE' instead of
2678
'b + 3'. */
2679
GET_BUFFER_SPACE (1 + OFFSET_ADDRESS_SIZE);
2680
INSERT_JUMP (keep_string_p ? on_failure_keep_string_jump
2681
: on_failure_jump,
2682
laststart, b + 1 + OFFSET_ADDRESS_SIZE);
2683
pending_exact = 0;
2684
b += 1 + OFFSET_ADDRESS_SIZE;
2685
2686
if (!zero_times_ok)
2687
{
2688
/* At least one repetition is required, so insert a
2689
`dummy_failure_jump' before the initial
2690
`on_failure_jump' instruction of the loop. This
2691
effects a skip over that instruction the first time
2692
we hit that loop. */
2693
GET_BUFFER_SPACE (1 + OFFSET_ADDRESS_SIZE);
2694
INSERT_JUMP (dummy_failure_jump, laststart, laststart +
2695
2 + 2 * OFFSET_ADDRESS_SIZE);
2696
b += 1 + OFFSET_ADDRESS_SIZE;
2697
}
2698
}
2699
break;
2700
2701
2702
case '.':
2703
laststart = b;
2704
BUF_PUSH (anychar);
2705
break;
2706
2707
2708
case '[':
2709
{
2710
boolean had_char_class = false;
2711
#ifdef WCHAR
2712
CHAR_T range_start = 0xffffffff;
2713
#else
2714
unsigned int range_start = 0xffffffff;
2715
#endif
2716
if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2717
2718
#ifdef WCHAR
2719
/* We assume a charset(_not) structure as a wchar_t array.
2720
charset[0] = (re_opcode_t) charset(_not)
2721
charset[1] = l (= length of char_classes)
2722
charset[2] = m (= length of collating_symbols)
2723
charset[3] = n (= length of equivalence_classes)
2724
charset[4] = o (= length of char_ranges)
2725
charset[5] = p (= length of chars)
2726
2727
charset[6] = char_class (wctype_t)
2728
charset[6+CHAR_CLASS_SIZE] = char_class (wctype_t)
2729
...
2730
charset[l+5] = char_class (wctype_t)
2731
2732
charset[l+6] = collating_symbol (wchar_t)
2733
...
2734
charset[l+m+5] = collating_symbol (wchar_t)
2735
ifdef _LIBC we use the index if
2736
_NL_COLLATE_SYMB_EXTRAMB instead of
2737
wchar_t string.
2738
2739
charset[l+m+6] = equivalence_classes (wchar_t)
2740
...
2741
charset[l+m+n+5] = equivalence_classes (wchar_t)
2742
ifdef _LIBC we use the index in
2743
_NL_COLLATE_WEIGHT instead of
2744
wchar_t string.
2745
2746
charset[l+m+n+6] = range_start
2747
charset[l+m+n+7] = range_end
2748
...
2749
charset[l+m+n+2o+4] = range_start
2750
charset[l+m+n+2o+5] = range_end
2751
ifdef _LIBC we use the value looked up
2752
in _NL_COLLATE_COLLSEQ instead of
2753
wchar_t character.
2754
2755
charset[l+m+n+2o+6] = char
2756
...
2757
charset[l+m+n+2o+p+5] = char
2758
2759
*/
2760
2761
/* We need at least 6 spaces: the opcode, the length of
2762
char_classes, the length of collating_symbols, the length of
2763
equivalence_classes, the length of char_ranges, the length of
2764
chars. */
2765
GET_BUFFER_SPACE (6);
2766
2767
/* Save b as laststart. And We use laststart as the pointer
2768
to the first element of the charset here.
2769
In other words, laststart[i] indicates charset[i]. */
2770
laststart = b;
2771
2772
/* We test `*p == '^' twice, instead of using an if
2773
statement, so we only need one BUF_PUSH. */
2774
BUF_PUSH (*p == '^' ? charset_not : charset);
2775
if (*p == '^')
2776
p++;
2777
2778
/* Push the length of char_classes, the length of
2779
collating_symbols, the length of equivalence_classes, the
2780
length of char_ranges and the length of chars. */
2781
BUF_PUSH_3 (0, 0, 0);
2782
BUF_PUSH_2 (0, 0);
2783
2784
/* Remember the first position in the bracket expression. */
2785
p1 = p;
2786
2787
/* charset_not matches newline according to a syntax bit. */
2788
if ((re_opcode_t) b[-6] == charset_not
2789
&& (syntax & RE_HAT_LISTS_NOT_NEWLINE))
2790
{
2791
BUF_PUSH('\n');
2792
laststart[5]++; /* Update the length of characters */
2793
}
2794
2795
/* Read in characters and ranges, setting map bits. */
2796
for (;;)
2797
{
2798
if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2799
2800
PATFETCH (c);
2801
2802
/* \ might escape characters inside [...] and [^...]. */
2803
if ((syntax & RE_BACKSLASH_ESCAPE_IN_LISTS) && c == '\\')
2804
{
2805
if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
2806
2807
PATFETCH (c1);
2808
BUF_PUSH(c1);
2809
laststart[5]++; /* Update the length of chars */
2810
range_start = c1;
2811
continue;
2812
}
2813
2814
/* Could be the end of the bracket expression. If it's
2815
not (i.e., when the bracket expression is `[]' so
2816
far), the ']' character bit gets set way below. */
2817
if (c == ']' && p != p1 + 1)
2818
break;
2819
2820
/* Look ahead to see if it's a range when the last thing
2821
was a character class. */
2822
if (had_char_class && c == '-' && *p != ']')
2823
FREE_STACK_RETURN (REG_ERANGE);
2824
2825
/* Look ahead to see if it's a range when the last thing
2826
was a character: if this is a hyphen not at the
2827
beginning or the end of a list, then it's the range
2828
operator. */
2829
if (c == '-'
2830
&& !(p - 2 >= pattern && p[-2] == '[')
2831
&& !(p - 3 >= pattern && p[-3] == '[' && p[-2] == '^')
2832
&& *p != ']')
2833
{
2834
reg_errcode_t ret;
2835
/* Allocate the space for range_start and range_end. */
2836
GET_BUFFER_SPACE (2);
2837
/* Update the pointer to indicate end of buffer. */
2838
b += 2;
2839
ret = wcs_compile_range (range_start, &p, pend, translate,
2840
syntax, b, laststart);
2841
if (ret != REG_NOERROR) FREE_STACK_RETURN (ret);
2842
range_start = 0xffffffff;
2843
}
2844
else if (p[0] == '-' && p[1] != ']')
2845
{ /* This handles ranges made up of characters only. */
2846
reg_errcode_t ret;
2847
2848
/* Move past the `-'. */
2849
PATFETCH (c1);
2850
/* Allocate the space for range_start and range_end. */
2851
GET_BUFFER_SPACE (2);
2852
/* Update the pointer to indicate end of buffer. */
2853
b += 2;
2854
ret = wcs_compile_range (c, &p, pend, translate, syntax, b,
2855
laststart);
2856
if (ret != REG_NOERROR) FREE_STACK_RETURN (ret);
2857
range_start = 0xffffffff;
2858
}
2859
2860
/* See if we're at the beginning of a possible character
2861
class. */
2862
else if (syntax & RE_CHAR_CLASSES && c == '[' && *p == ':')
2863
{ /* Leave room for the null. */
2864
char str[CHAR_CLASS_MAX_LENGTH + 1];
2865
2866
PATFETCH (c);
2867
c1 = 0;
2868
2869
/* If pattern is `[[:'. */
2870
if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2871
2872
for (;;)
2873
{
2874
PATFETCH (c);
2875
if ((c == ':' && *p == ']') || p == pend)
2876
break;
2877
if (c1 < CHAR_CLASS_MAX_LENGTH)
2878
str[c1++] = c;
2879
else
2880
/* This is in any case an invalid class name. */
2881
str[0] = '\0';
2882
}
2883
str[c1] = '\0';
2884
2885
/* If isn't a word bracketed by `[:' and `:]':
2886
undo the ending character, the letters, and leave
2887
the leading `:' and `[' (but store them as character). */
2888
if (c == ':' && *p == ']')
2889
{
2890
wctype_t wt;
2891
uintptr_t alignedp;
2892
2893
/* Query the character class as wctype_t. */
2894
wt = IS_CHAR_CLASS (str);
2895
if (wt == 0)
2896
FREE_STACK_RETURN (REG_ECTYPE);
2897
2898
/* Throw away the ] at the end of the character
2899
class. */
2900
PATFETCH (c);
2901
2902
if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2903
2904
/* Allocate the space for character class. */
2905
GET_BUFFER_SPACE(CHAR_CLASS_SIZE);
2906
/* Update the pointer to indicate end of buffer. */
2907
b += CHAR_CLASS_SIZE;
2908
/* Move data which follow character classes
2909
not to violate the data. */
2910
insert_space(CHAR_CLASS_SIZE,
2911
laststart + 6 + laststart[1],
2912
b - 1);
2913
alignedp = ((uintptr_t)(laststart + 6 + laststart[1])
2914
+ __alignof__(wctype_t) - 1)
2915
& ~(uintptr_t)(__alignof__(wctype_t) - 1);
2916
/* Store the character class. */
2917
*((wctype_t*)alignedp) = wt;
2918
/* Update length of char_classes */
2919
laststart[1] += CHAR_CLASS_SIZE;
2920
2921
had_char_class = true;
2922
}
2923
else
2924
{
2925
c1++;
2926
while (c1--)
2927
PATUNFETCH;
2928
BUF_PUSH ('[');
2929
BUF_PUSH (':');
2930
laststart[5] += 2; /* Update the length of characters */
2931
range_start = ':';
2932
had_char_class = false;
2933
}
2934
}
2935
else if (syntax & RE_CHAR_CLASSES && c == '[' && (*p == '='
2936
|| *p == '.'))
2937
{
2938
CHAR_T str[128]; /* Should be large enough. */
2939
CHAR_T delim = *p; /* '=' or '.' */
2940
# ifdef _LIBC
2941
uint32_t nrules =
2942
_NL_CURRENT_WORD (LC_COLLATE, _NL_COLLATE_NRULES);
2943
# endif
2944
PATFETCH (c);
2945
c1 = 0;
2946
2947
/* If pattern is `[[=' or '[[.'. */
2948
if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2949
2950
for (;;)
2951
{
2952
PATFETCH (c);
2953
if ((c == delim && *p == ']') || p == pend)
2954
break;
2955
if (c1 < sizeof (str) - 1)
2956
str[c1++] = c;
2957
else
2958
/* This is in any case an invalid class name. */
2959
str[0] = '\0';
2960
}
2961
str[c1] = '\0';
2962
2963
if (c == delim && *p == ']' && str[0] != '\0')
2964
{
2965
unsigned int i, offset;
2966
/* If we have no collation data we use the default
2967
collation in which each character is in a class
2968
by itself. It also means that ASCII is the
2969
character set and therefore we cannot have character
2970
with more than one byte in the multibyte
2971
representation. */
2972
2973
/* If not defined _LIBC, we push the name and
2974
`\0' for the sake of matching performance. */
2975
int datasize = c1 + 1;
2976
2977
# ifdef _LIBC
2978
int32_t idx = 0;
2979
if (nrules == 0)
2980
# endif
2981
{
2982
if (c1 != 1)
2983
FREE_STACK_RETURN (REG_ECOLLATE);
2984
}
2985
# ifdef _LIBC
2986
else
2987
{
2988
const int32_t *table;
2989
const int32_t *weights;
2990
const int32_t *extra;
2991
const int32_t *indirect;
2992
wint_t *cp;
2993
2994
/* This #include defines a local function! */
2995
# include <locale/weightwc.h>
2996
2997
if(delim == '=')
2998
{
2999
/* We push the index for equivalence class. */
3000
cp = (wint_t*)str;
3001
3002
table = (const int32_t *)
3003
_NL_CURRENT (LC_COLLATE,
3004
_NL_COLLATE_TABLEWC);
3005
weights = (const int32_t *)
3006
_NL_CURRENT (LC_COLLATE,
3007
_NL_COLLATE_WEIGHTWC);
3008
extra = (const int32_t *)
3009
_NL_CURRENT (LC_COLLATE,
3010
_NL_COLLATE_EXTRAWC);
3011
indirect = (const int32_t *)
3012
_NL_CURRENT (LC_COLLATE,
3013
_NL_COLLATE_INDIRECTWC);
3014
3015
idx = findidx ((const wint_t**)&cp);
3016
if (idx == 0 || cp < (wint_t*) str + c1)
3017
/* This is no valid character. */
3018
FREE_STACK_RETURN (REG_ECOLLATE);
3019
3020
str[0] = (wchar_t)idx;
3021
}
3022
else /* delim == '.' */
3023
{
3024
/* We push collation sequence value
3025
for collating symbol. */
3026
int32_t table_size;
3027
const int32_t *symb_table;
3028
const unsigned char *extra;
3029
int32_t idx;
3030
int32_t elem;
3031
int32_t second;
3032
int32_t hash;
3033
char char_str[c1];
3034
3035
/* We have to convert the name to a single-byte
3036
string. This is possible since the names
3037
consist of ASCII characters and the internal
3038
representation is UCS4. */
3039
for (i = 0; i < c1; ++i)
3040
char_str[i] = str[i];
3041
3042
table_size =
3043
_NL_CURRENT_WORD (LC_COLLATE,
3044
_NL_COLLATE_SYMB_HASH_SIZEMB);
3045
symb_table = (const int32_t *)
3046
_NL_CURRENT (LC_COLLATE,
3047
_NL_COLLATE_SYMB_TABLEMB);
3048
extra = (const unsigned char *)
3049
_NL_CURRENT (LC_COLLATE,
3050
_NL_COLLATE_SYMB_EXTRAMB);
3051
3052
/* Locate the character in the hashing table. */
3053
hash = elem_hash (char_str, c1);
3054
3055
idx = 0;
3056
elem = hash % table_size;
3057
second = hash % (table_size - 2);
3058
while (symb_table[2 * elem] != 0)
3059
{
3060
/* First compare the hashing value. */
3061
if (symb_table[2 * elem] == hash
3062
&& c1 == extra[symb_table[2 * elem + 1]]
3063
&& memcmp (char_str,
3064
&extra[symb_table[2 * elem + 1]
3065
+ 1], c1) == 0)
3066
{
3067
/* Yep, this is the entry. */
3068
idx = symb_table[2 * elem + 1];
3069
idx += 1 + extra[idx];
3070
break;
3071
}
3072
3073
/* Next entry. */
3074
elem += second;
3075
}
3076
3077
if (symb_table[2 * elem] != 0)
3078
{
3079
/* Compute the index of the byte sequence
3080
in the table. */
3081
idx += 1 + extra[idx];
3082
/* Adjust for the alignment. */
3083
idx = (idx + 3) & ~3;
3084
3085
str[0] = (wchar_t) idx + 4;
3086
}
3087
else if (symb_table[2 * elem] == 0 && c1 == 1)
3088
{
3089
/* No valid character. Match it as a
3090
single byte character. */
3091
had_char_class = false;
3092
BUF_PUSH(str[0]);
3093
/* Update the length of characters */
3094
laststart[5]++;
3095
range_start = str[0];
3096
3097
/* Throw away the ] at the end of the
3098
collating symbol. */
3099
PATFETCH (c);
3100
/* exit from the switch block. */
3101
continue;
3102
}
3103
else
3104
FREE_STACK_RETURN (REG_ECOLLATE);
3105
}
3106
datasize = 1;
3107
}
3108
# endif
3109
/* Throw away the ] at the end of the equivalence
3110
class (or collating symbol). */
3111
PATFETCH (c);
3112
3113
/* Allocate the space for the equivalence class
3114
(or collating symbol) (and '\0' if needed). */
3115
GET_BUFFER_SPACE(datasize);
3116
/* Update the pointer to indicate end of buffer. */
3117
b += datasize;
3118
3119
if (delim == '=')
3120
{ /* equivalence class */
3121
/* Calculate the offset of char_ranges,
3122
which is next to equivalence_classes. */
3123
offset = laststart[1] + laststart[2]
3124
+ laststart[3] +6;
3125
/* Insert space. */
3126
insert_space(datasize, laststart + offset, b - 1);
3127
3128
/* Write the equivalence_class and \0. */
3129
for (i = 0 ; i < datasize ; i++)
3130
laststart[offset + i] = str[i];
3131
3132
/* Update the length of equivalence_classes. */
3133
laststart[3] += datasize;
3134
had_char_class = true;
3135
}
3136
else /* delim == '.' */
3137
{ /* collating symbol */
3138
/* Calculate the offset of the equivalence_classes,
3139
which is next to collating_symbols. */
3140
offset = laststart[1] + laststart[2] + 6;
3141
/* Insert space and write the collationg_symbol
3142
and \0. */
3143
insert_space(datasize, laststart + offset, b-1);
3144
for (i = 0 ; i < datasize ; i++)
3145
laststart[offset + i] = str[i];
3146
3147
/* In re_match_2_internal if range_start < -1, we
3148
assume -range_start is the offset of the
3149
collating symbol which is specified as
3150
the character of the range start. So we assign
3151
-(laststart[1] + laststart[2] + 6) to
3152
range_start. */
3153
range_start = -(laststart[1] + laststart[2] + 6);
3154
/* Update the length of collating_symbol. */
3155
laststart[2] += datasize;
3156
had_char_class = false;
3157
}
3158
}
3159
else
3160
{
3161
c1++;
3162
while (c1--)
3163
PATUNFETCH;
3164
BUF_PUSH ('[');
3165
BUF_PUSH (delim);
3166
laststart[5] += 2; /* Update the length of characters */
3167
range_start = delim;
3168
had_char_class = false;
3169
}
3170
}
3171
else
3172
{
3173
had_char_class = false;
3174
BUF_PUSH(c);
3175
laststart[5]++; /* Update the length of characters */
3176
range_start = c;
3177
}
3178
}
3179
3180
#else /* BYTE */
3181
/* Ensure that we have enough space to push a charset: the
3182
opcode, the length count, and the bitset; 34 bytes in all. */
3183
GET_BUFFER_SPACE (34);
3184
3185
laststart = b;
3186
3187
/* We test `*p == '^' twice, instead of using an if
3188
statement, so we only need one BUF_PUSH. */
3189
BUF_PUSH (*p == '^' ? charset_not : charset);
3190
if (*p == '^')
3191
p++;
3192
3193
/* Remember the first position in the bracket expression. */
3194
p1 = p;
3195
3196
/* Push the number of bytes in the bitmap. */
3197
BUF_PUSH ((1 << BYTEWIDTH) / BYTEWIDTH);
3198
3199
/* Clear the whole map. */
3200
bzero (b, (1 << BYTEWIDTH) / BYTEWIDTH);
3201
3202
/* charset_not matches newline according to a syntax bit. */
3203
if ((re_opcode_t) b[-2] == charset_not
3204
&& (syntax & RE_HAT_LISTS_NOT_NEWLINE))
3205
SET_LIST_BIT ('\n');
3206
3207
/* Read in characters and ranges, setting map bits. */
3208
for (;;)
3209
{
3210
if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
3211
3212
PATFETCH (c);
3213
3214
/* \ might escape characters inside [...] and [^...]. */
3215
if ((syntax & RE_BACKSLASH_ESCAPE_IN_LISTS) && c == '\\')
3216
{
3217
if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
3218
3219
PATFETCH (c1);
3220
SET_LIST_BIT (c1);
3221
range_start = c1;
3222
continue;
3223
}
3224
3225
/* Could be the end of the bracket expression. If it's
3226
not (i.e., when the bracket expression is `[]' so
3227
far), the ']' character bit gets set way below. */
3228
if (c == ']' && p != p1 + 1)
3229
break;
3230
3231
/* Look ahead to see if it's a range when the last thing
3232
was a character class. */
3233
if (had_char_class && c == '-' && *p != ']')
3234
FREE_STACK_RETURN (REG_ERANGE);
3235
3236
/* Look ahead to see if it's a range when the last thing
3237
was a character: if this is a hyphen not at the
3238
beginning or the end of a list, then it's the range
3239
operator. */
3240
if (c == '-'
3241
&& !(p - 2 >= pattern && p[-2] == '[')
3242
&& !(p - 3 >= pattern && p[-3] == '[' && p[-2] == '^')
3243
&& *p != ']')
3244
{
3245
reg_errcode_t ret
3246
= byte_compile_range (range_start, &p, pend, translate,
3247
syntax, b);
3248
if (ret != REG_NOERROR) FREE_STACK_RETURN (ret);
3249
range_start = 0xffffffff;
3250
}
3251
3252
else if (p[0] == '-' && p[1] != ']')
3253
{ /* This handles ranges made up of characters only. */
3254
reg_errcode_t ret;
3255
3256
/* Move past the `-'. */
3257
PATFETCH (c1);
3258
3259
ret = byte_compile_range (c, &p, pend, translate, syntax, b);
3260
if (ret != REG_NOERROR) FREE_STACK_RETURN (ret);
3261
range_start = 0xffffffff;
3262
}
3263
3264
/* See if we're at the beginning of a possible character
3265
class. */
3266
3267
else if (syntax & RE_CHAR_CLASSES && c == '[' && *p == ':')
3268
{ /* Leave room for the null. */
3269
char str[CHAR_CLASS_MAX_LENGTH + 1];
3270
3271
PATFETCH (c);
3272
c1 = 0;
3273
3274
/* If pattern is `[[:'. */
3275
if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
3276
3277
for (;;)
3278
{
3279
PATFETCH (c);
3280
if ((c == ':' && *p == ']') || p == pend)
3281
break;
3282
if (c1 < CHAR_CLASS_MAX_LENGTH)
3283
str[c1++] = c;
3284
else
3285
/* This is in any case an invalid class name. */
3286
str[0] = '\0';
3287
}
3288
str[c1] = '\0';
3289
3290
/* If isn't a word bracketed by `[:' and `:]':
3291
undo the ending character, the letters, and leave
3292
the leading `:' and `[' (but set bits for them). */
3293
if (c == ':' && *p == ']')
3294
{
3295
# if defined _LIBC || WIDE_CHAR_SUPPORT
3296
boolean is_lower = STREQ (str, "lower");
3297
boolean is_upper = STREQ (str, "upper");
3298
wctype_t wt;
3299
int ch;
3300
3301
wt = IS_CHAR_CLASS (str);
3302
if (wt == 0)
3303
FREE_STACK_RETURN (REG_ECTYPE);
3304
3305
/* Throw away the ] at the end of the character
3306
class. */
3307
PATFETCH (c);
3308
3309
if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
3310
3311
for (ch = 0; ch < 1 << BYTEWIDTH; ++ch)
3312
{
3313
if (iswctype (btowc (ch), wt))
3314
SET_LIST_BIT (ch);
3315
3316
if (translate && (is_upper || is_lower)
3317
&& (ISUPPER (ch) || ISLOWER (ch)))
3318
SET_LIST_BIT (ch);
3319
}
3320
3321
had_char_class = true;
3322
# else
3323
int ch;
3324
boolean is_alnum = STREQ (str, "alnum");
3325
boolean is_alpha = STREQ (str, "alpha");
3326
boolean is_blank = STREQ (str, "blank");
3327
boolean is_cntrl = STREQ (str, "cntrl");
3328
boolean is_digit = STREQ (str, "digit");
3329
boolean is_graph = STREQ (str, "graph");
3330
boolean is_lower = STREQ (str, "lower");
3331
boolean is_print = STREQ (str, "print");
3332
boolean is_punct = STREQ (str, "punct");
3333
boolean is_space = STREQ (str, "space");
3334
boolean is_upper = STREQ (str, "upper");
3335
boolean is_xdigit = STREQ (str, "xdigit");
3336
3337
if (!IS_CHAR_CLASS (str))
3338
FREE_STACK_RETURN (REG_ECTYPE);
3339
3340
/* Throw away the ] at the end of the character
3341
class. */
3342
PATFETCH (c);
3343
3344
if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
3345
3346
for (ch = 0; ch < 1 << BYTEWIDTH; ch++)
3347
{
3348
/* This was split into 3 if's to
3349
avoid an arbitrary limit in some compiler. */
3350
if ( (is_alnum && ISALNUM (ch))
3351
|| (is_alpha && ISALPHA (ch))
3352
|| (is_blank && ISBLANK (ch))
3353
|| (is_cntrl && ISCNTRL (ch)))
3354
SET_LIST_BIT (ch);
3355
if ( (is_digit && ISDIGIT (ch))
3356
|| (is_graph && ISGRAPH (ch))
3357
|| (is_lower && ISLOWER (ch))
3358
|| (is_print && ISPRINT (ch)))
3359
SET_LIST_BIT (ch);
3360
if ( (is_punct && ISPUNCT (ch))
3361
|| (is_space && ISSPACE (ch))
3362
|| (is_upper && ISUPPER (ch))
3363
|| (is_xdigit && ISXDIGIT (ch)))
3364
SET_LIST_BIT (ch);
3365
if ( translate && (is_upper || is_lower)
3366
&& (ISUPPER (ch) || ISLOWER (ch)))
3367
SET_LIST_BIT (ch);
3368
}
3369
had_char_class = true;
3370
# endif /* libc || wctype.h */
3371
}
3372
else
3373
{
3374
c1++;
3375
while (c1--)
3376
PATUNFETCH;
3377
SET_LIST_BIT ('[');
3378
SET_LIST_BIT (':');
3379
range_start = ':';
3380
had_char_class = false;
3381
}
3382
}
3383
else if (syntax & RE_CHAR_CLASSES && c == '[' && *p == '=')
3384
{
3385
unsigned char str[MB_LEN_MAX + 1];
3386
# ifdef _LIBC
3387
uint32_t nrules =
3388
_NL_CURRENT_WORD (LC_COLLATE, _NL_COLLATE_NRULES);
3389
# endif
3390
3391
PATFETCH (c);
3392
c1 = 0;
3393
3394
/* If pattern is `[[='. */
3395
if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
3396
3397
for (;;)
3398
{
3399
PATFETCH (c);
3400
if ((c == '=' && *p == ']') || p == pend)
3401
break;
3402
if (c1 < MB_LEN_MAX)
3403
str[c1++] = c;
3404
else
3405
/* This is in any case an invalid class name. */
3406
str[0] = '\0';
3407
}
3408
str[c1] = '\0';
3409
3410
if (c == '=' && *p == ']' && str[0] != '\0')
3411
{
3412
/* If we have no collation data we use the default
3413
collation in which each character is in a class
3414
by itself. It also means that ASCII is the
3415
character set and therefore we cannot have character
3416
with more than one byte in the multibyte
3417
representation. */
3418
# ifdef _LIBC
3419
if (nrules == 0)
3420
# endif
3421
{
3422
if (c1 != 1)
3423
FREE_STACK_RETURN (REG_ECOLLATE);
3424
3425
/* Throw away the ] at the end of the equivalence
3426
class. */
3427
PATFETCH (c);
3428
3429
/* Set the bit for the character. */
3430
SET_LIST_BIT (str[0]);
3431
}
3432
# ifdef _LIBC
3433
else
3434
{
3435
/* Try to match the byte sequence in `str' against
3436
those known to the collate implementation.
3437
First find out whether the bytes in `str' are
3438
actually from exactly one character. */
3439
const int32_t *table;
3440
const unsigned char *weights;
3441
const unsigned char *extra;
3442
const int32_t *indirect;
3443
int32_t idx;
3444
const unsigned char *cp = str;
3445
int ch;
3446
3447
/* This #include defines a local function! */
3448
# include <locale/weight.h>
3449
3450
table = (const int32_t *)
3451
_NL_CURRENT (LC_COLLATE, _NL_COLLATE_TABLEMB);
3452
weights = (const unsigned char *)
3453
_NL_CURRENT (LC_COLLATE, _NL_COLLATE_WEIGHTMB);
3454
extra = (const unsigned char *)
3455
_NL_CURRENT (LC_COLLATE, _NL_COLLATE_EXTRAMB);
3456
indirect = (const int32_t *)
3457
_NL_CURRENT (LC_COLLATE, _NL_COLLATE_INDIRECTMB);
3458
3459
idx = findidx (&cp);
3460
if (idx == 0 || cp < str + c1)
3461
/* This is no valid character. */
3462
FREE_STACK_RETURN (REG_ECOLLATE);
3463
3464
/* Throw away the ] at the end of the equivalence
3465
class. */
3466
PATFETCH (c);
3467
3468
/* Now we have to go throught the whole table
3469
and find all characters which have the same
3470
first level weight.
3471
3472
XXX Note that this is not entirely correct.
3473
we would have to match multibyte sequences
3474
but this is not possible with the current
3475
implementation. */
3476
for (ch = 1; ch < 256; ++ch)
3477
/* XXX This test would have to be changed if we
3478
would allow matching multibyte sequences. */
3479
if (table[ch] > 0)
3480
{
3481
int32_t idx2 = table[ch];
3482
size_t len = weights[idx2];
3483
3484
/* Test whether the lenghts match. */
3485
if (weights[idx] == len)
3486
{
3487
/* They do. New compare the bytes of
3488
the weight. */
3489
size_t cnt = 0;
3490
3491
while (cnt < len
3492
&& (weights[idx + 1 + cnt]
3493
== weights[idx2 + 1 + cnt]))
3494
++cnt;
3495
3496
if (cnt == len)
3497
/* They match. Mark the character as
3498
acceptable. */
3499
SET_LIST_BIT (ch);
3500
}
3501
}
3502
}
3503
# endif
3504
had_char_class = true;
3505
}
3506
else
3507
{
3508
c1++;
3509
while (c1--)
3510
PATUNFETCH;
3511
SET_LIST_BIT ('[');
3512
SET_LIST_BIT ('=');
3513
range_start = '=';
3514
had_char_class = false;
3515
}
3516
}
3517
else if (syntax & RE_CHAR_CLASSES && c == '[' && *p == '.')
3518
{
3519
unsigned char str[128]; /* Should be large enough. */
3520
# ifdef _LIBC
3521
uint32_t nrules =
3522
_NL_CURRENT_WORD (LC_COLLATE, _NL_COLLATE_NRULES);
3523
# endif
3524
3525
PATFETCH (c);
3526
c1 = 0;
3527
3528
/* If pattern is `[[.'. */
3529
if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
3530
3531
for (;;)
3532
{
3533
PATFETCH (c);
3534
if ((c == '.' && *p == ']') || p == pend)
3535
break;
3536
if (c1 < sizeof (str))
3537
str[c1++] = c;
3538
else
3539
/* This is in any case an invalid class name. */
3540
str[0] = '\0';
3541
}
3542
str[c1] = '\0';
3543
3544
if (c == '.' && *p == ']' && str[0] != '\0')
3545
{
3546
/* If we have no collation data we use the default
3547
collation in which each character is the name
3548
for its own class which contains only the one
3549
character. It also means that ASCII is the
3550
character set and therefore we cannot have character
3551
with more than one byte in the multibyte
3552
representation. */
3553
# ifdef _LIBC
3554
if (nrules == 0)
3555
# endif
3556
{
3557
if (c1 != 1)
3558
FREE_STACK_RETURN (REG_ECOLLATE);
3559
3560
/* Throw away the ] at the end of the equivalence
3561
class. */
3562
PATFETCH (c);
3563
3564
/* Set the bit for the character. */
3565
SET_LIST_BIT (str[0]);
3566
range_start = ((const unsigned char *) str)[0];
3567
}
3568
# ifdef _LIBC
3569
else
3570
{
3571
/* Try to match the byte sequence in `str' against
3572
those known to the collate implementation.
3573
First find out whether the bytes in `str' are
3574
actually from exactly one character. */
3575
int32_t table_size;
3576
const int32_t *symb_table;
3577
const unsigned char *extra;
3578
int32_t idx;
3579
int32_t elem;
3580
int32_t second;
3581
int32_t hash;
3582
3583
table_size =
3584
_NL_CURRENT_WORD (LC_COLLATE,
3585
_NL_COLLATE_SYMB_HASH_SIZEMB);
3586
symb_table = (const int32_t *)
3587
_NL_CURRENT (LC_COLLATE,
3588
_NL_COLLATE_SYMB_TABLEMB);
3589
extra = (const unsigned char *)
3590
_NL_CURRENT (LC_COLLATE,
3591
_NL_COLLATE_SYMB_EXTRAMB);
3592
3593
/* Locate the character in the hashing table. */
3594
hash = elem_hash (str, c1);
3595
3596
idx = 0;
3597
elem = hash % table_size;
3598
second = hash % (table_size - 2);
3599
while (symb_table[2 * elem] != 0)
3600
{
3601
/* First compare the hashing value. */
3602
if (symb_table[2 * elem] == hash
3603
&& c1 == extra[symb_table[2 * elem + 1]]
3604
&& memcmp (str,
3605
&extra[symb_table[2 * elem + 1]
3606
+ 1],
3607
c1) == 0)
3608
{
3609
/* Yep, this is the entry. */
3610
idx = symb_table[2 * elem + 1];
3611
idx += 1 + extra[idx];
3612
break;
3613
}
3614
3615
/* Next entry. */
3616
elem += second;
3617
}
3618
3619
if (symb_table[2 * elem] == 0)
3620
/* This is no valid character. */
3621
FREE_STACK_RETURN (REG_ECOLLATE);
3622
3623
/* Throw away the ] at the end of the equivalence
3624
class. */
3625
PATFETCH (c);
3626
3627
/* Now add the multibyte character(s) we found
3628
to the accept list.
3629
3630
XXX Note that this is not entirely correct.
3631
we would have to match multibyte sequences
3632
but this is not possible with the current
3633
implementation. Also, we have to match
3634
collating symbols, which expand to more than
3635
one file, as a whole and not allow the
3636
individual bytes. */
3637
c1 = extra[idx++];
3638
if (c1 == 1)
3639
range_start = extra[idx];
3640
while (c1-- > 0)
3641
{
3642
SET_LIST_BIT (extra[idx]);
3643
++idx;
3644
}
3645
}
3646
# endif
3647
had_char_class = false;
3648
}
3649
else
3650
{
3651
c1++;
3652
while (c1--)
3653
PATUNFETCH;
3654
SET_LIST_BIT ('[');
3655
SET_LIST_BIT ('.');
3656
range_start = '.';
3657
had_char_class = false;
3658
}
3659
}
3660
else
3661
{
3662
had_char_class = false;
3663
SET_LIST_BIT (c);
3664
range_start = c;
3665
}
3666
}
3667
3668
/* Discard any (non)matching list bytes that are all 0 at the
3669
end of the map. Decrease the map-length byte too. */
3670
while ((int) b[-1] > 0 && b[b[-1] - 1] == 0)
3671
b[-1]--;
3672
b += b[-1];
3673
#endif /* WCHAR */
3674
}
3675
break;
3676
3677
3678
case '(':
3679
if (syntax & RE_NO_BK_PARENS)
3680
goto handle_open;
3681
else
3682
goto normal_char;
3683
3684
3685
case ')':
3686
if (syntax & RE_NO_BK_PARENS)
3687
goto handle_close;
3688
else
3689
goto normal_char;
3690
3691
3692
case '\n':
3693
if (syntax & RE_NEWLINE_ALT)
3694
goto handle_alt;
3695
else
3696
goto normal_char;
3697
3698
3699
case '|':
3700
if (syntax & RE_NO_BK_VBAR)
3701
goto handle_alt;
3702
else
3703
goto normal_char;
3704
3705
3706
case '{':
3707
if (syntax & RE_INTERVALS && syntax & RE_NO_BK_BRACES)
3708
goto handle_interval;
3709
else
3710
goto normal_char;
3711
3712
3713
case '\\':
3714
if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
3715
3716
/* Do not translate the character after the \, so that we can
3717
distinguish, e.g., \B from \b, even if we normally would
3718
translate, e.g., B to b. */
3719
PATFETCH_RAW (c);
3720
3721
switch (c)
3722
{
3723
case '(':
3724
if (syntax & RE_NO_BK_PARENS)
3725
goto normal_backslash;
3726
3727
handle_open:
3728
bufp->re_nsub++;
3729
regnum++;
3730
3731
if (COMPILE_STACK_FULL)
3732
{
3733
RETALLOC (compile_stack.stack, compile_stack.size << 1,
3734
compile_stack_elt_t);
3735
if (compile_stack.stack == NULL) return REG_ESPACE;
3736
3737
compile_stack.size <<= 1;
3738
}
3739
3740
/* These are the values to restore when we hit end of this
3741
group. They are all relative offsets, so that if the
3742
whole pattern moves because of realloc, they will still
3743
be valid. */
3744
COMPILE_STACK_TOP.begalt_offset = begalt - COMPILED_BUFFER_VAR;
3745
COMPILE_STACK_TOP.fixup_alt_jump
3746
= fixup_alt_jump ? fixup_alt_jump - COMPILED_BUFFER_VAR + 1 : 0;
3747
COMPILE_STACK_TOP.laststart_offset = b - COMPILED_BUFFER_VAR;
3748
COMPILE_STACK_TOP.regnum = regnum;
3749
3750
/* We will eventually replace the 0 with the number of
3751
groups inner to this one. But do not push a
3752
start_memory for groups beyond the last one we can
3753
represent in the compiled pattern. */
3754
if (regnum <= MAX_REGNUM)
3755
{
3756
COMPILE_STACK_TOP.inner_group_offset = b
3757
- COMPILED_BUFFER_VAR + 2;
3758
BUF_PUSH_3 (start_memory, regnum, 0);
3759
}
3760
3761
compile_stack.avail++;
3762
3763
fixup_alt_jump = 0;
3764
laststart = 0;
3765
begalt = b;
3766
/* If we've reached MAX_REGNUM groups, then this open
3767
won't actually generate any code, so we'll have to
3768
clear pending_exact explicitly. */
3769
pending_exact = 0;
3770
break;
3771
3772
3773
case ')':
3774
if (syntax & RE_NO_BK_PARENS) goto normal_backslash;
3775
3776
if (COMPILE_STACK_EMPTY)
3777
{
3778
if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD)
3779
goto normal_backslash;
3780
else
3781
FREE_STACK_RETURN (REG_ERPAREN);
3782
}
3783
3784
handle_close:
3785
if (fixup_alt_jump)
3786
{ /* Push a dummy failure point at the end of the
3787
alternative for a possible future
3788
`pop_failure_jump' to pop. See comments at
3789
`push_dummy_failure' in `re_match_2'. */
3790
BUF_PUSH (push_dummy_failure);
3791
3792
/* We allocated space for this jump when we assigned
3793
to `fixup_alt_jump', in the `handle_alt' case below. */
3794
STORE_JUMP (jump_past_alt, fixup_alt_jump, b - 1);
3795
}
3796
3797
/* See similar code for backslashed left paren above. */
3798
if (COMPILE_STACK_EMPTY)
3799
{
3800
if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD)
3801
goto normal_char;
3802
else
3803
FREE_STACK_RETURN (REG_ERPAREN);
3804
}
3805
3806
/* Since we just checked for an empty stack above, this
3807
``can't happen''. */
3808
assert (compile_stack.avail != 0);
3809
{
3810
/* We don't just want to restore into `regnum', because
3811
later groups should continue to be numbered higher,
3812
as in `(ab)c(de)' -- the second group is #2. */
3813
regnum_t this_group_regnum;
3814
3815
compile_stack.avail--;
3816
begalt = COMPILED_BUFFER_VAR + COMPILE_STACK_TOP.begalt_offset;
3817
fixup_alt_jump
3818
= COMPILE_STACK_TOP.fixup_alt_jump
3819
? COMPILED_BUFFER_VAR + COMPILE_STACK_TOP.fixup_alt_jump - 1
3820
: 0;
3821
laststart = COMPILED_BUFFER_VAR + COMPILE_STACK_TOP.laststart_offset;
3822
this_group_regnum = COMPILE_STACK_TOP.regnum;
3823
/* If we've reached MAX_REGNUM groups, then this open
3824
won't actually generate any code, so we'll have to
3825
clear pending_exact explicitly. */
3826
pending_exact = 0;
3827
3828
/* We're at the end of the group, so now we know how many
3829
groups were inside this one. */
3830
if (this_group_regnum <= MAX_REGNUM)
3831
{
3832
UCHAR_T *inner_group_loc
3833
= COMPILED_BUFFER_VAR + COMPILE_STACK_TOP.inner_group_offset;
3834
3835
*inner_group_loc = regnum - this_group_regnum;
3836
BUF_PUSH_3 (stop_memory, this_group_regnum,
3837
regnum - this_group_regnum);
3838
}
3839
}
3840
break;
3841
3842
3843
case '|': /* `\|'. */
3844
if (syntax & RE_LIMITED_OPS || syntax & RE_NO_BK_VBAR)
3845
goto normal_backslash;
3846
handle_alt:
3847
if (syntax & RE_LIMITED_OPS)
3848
goto normal_char;
3849
3850
/* Insert before the previous alternative a jump which
3851
jumps to this alternative if the former fails. */
3852
GET_BUFFER_SPACE (1 + OFFSET_ADDRESS_SIZE);
3853
INSERT_JUMP (on_failure_jump, begalt,
3854
b + 2 + 2 * OFFSET_ADDRESS_SIZE);
3855
pending_exact = 0;
3856
b += 1 + OFFSET_ADDRESS_SIZE;
3857
3858
/* The alternative before this one has a jump after it
3859
which gets executed if it gets matched. Adjust that
3860
jump so it will jump to this alternative's analogous
3861
jump (put in below, which in turn will jump to the next
3862
(if any) alternative's such jump, etc.). The last such
3863
jump jumps to the correct final destination. A picture:
3864
_____ _____
3865
| | | |
3866
| v | v
3867
a | b | c
3868
3869
If we are at `b', then fixup_alt_jump right now points to a
3870
three-byte space after `a'. We'll put in the jump, set
3871
fixup_alt_jump to right after `b', and leave behind three
3872
bytes which we'll fill in when we get to after `c'. */
3873
3874
if (fixup_alt_jump)
3875
STORE_JUMP (jump_past_alt, fixup_alt_jump, b);
3876
3877
/* Mark and leave space for a jump after this alternative,
3878
to be filled in later either by next alternative or
3879
when know we're at the end of a series of alternatives. */
3880
fixup_alt_jump = b;
3881
GET_BUFFER_SPACE (1 + OFFSET_ADDRESS_SIZE);
3882
b += 1 + OFFSET_ADDRESS_SIZE;
3883
3884
laststart = 0;
3885
begalt = b;
3886
break;
3887
3888
3889
case '{':
3890
/* If \{ is a literal. */
3891
if (!(syntax & RE_INTERVALS)
3892
/* If we're at `\{' and it's not the open-interval
3893
operator. */
3894
|| (syntax & RE_NO_BK_BRACES))
3895
goto normal_backslash;
3896
3897
handle_interval:
3898
{
3899
/* If got here, then the syntax allows intervals. */
3900
3901
/* At least (most) this many matches must be made. */
3902
int lower_bound = -1, upper_bound = -1;
3903
3904
/* Place in the uncompiled pattern (i.e., just after
3905
the '{') to go back to if the interval is invalid. */
3906
const CHAR_T *beg_interval = p;
3907
3908
if (p == pend)
3909
goto invalid_interval;
3910
3911
GET_UNSIGNED_NUMBER (lower_bound);
3912
3913
if (c == ',')
3914
{
3915
GET_UNSIGNED_NUMBER (upper_bound);
3916
if (upper_bound < 0)
3917
upper_bound = RE_DUP_MAX;
3918
}
3919
else
3920
/* Interval such as `{1}' => match exactly once. */
3921
upper_bound = lower_bound;
3922
3923
if (! (0 <= lower_bound && lower_bound <= upper_bound))
3924
goto invalid_interval;
3925
3926
if (!(syntax & RE_NO_BK_BRACES))
3927
{
3928
if (c != '\\' || p == pend)
3929
goto invalid_interval;
3930
PATFETCH (c);
3931
}
3932
3933
if (c != '}')
3934
goto invalid_interval;
3935
3936
/* If it's invalid to have no preceding re. */
3937
if (!laststart)
3938
{
3939
if (syntax & RE_CONTEXT_INVALID_OPS
3940
&& !(syntax & RE_INVALID_INTERVAL_ORD))
3941
FREE_STACK_RETURN (REG_BADRPT);
3942
else if (syntax & RE_CONTEXT_INDEP_OPS)
3943
laststart = b;
3944
else
3945
goto unfetch_interval;
3946
}
3947
3948
/* We just parsed a valid interval. */
3949
3950
if (RE_DUP_MAX < upper_bound)
3951
FREE_STACK_RETURN (REG_BADBR);
3952
3953
/* If the upper bound is zero, don't want to succeed at
3954
all; jump from `laststart' to `b + 3', which will be
3955
the end of the buffer after we insert the jump. */
3956
/* ifdef WCHAR, 'b + 1 + OFFSET_ADDRESS_SIZE'
3957
instead of 'b + 3'. */
3958
if (upper_bound == 0)
3959
{
3960
GET_BUFFER_SPACE (1 + OFFSET_ADDRESS_SIZE);
3961
INSERT_JUMP (jump, laststart, b + 1
3962
+ OFFSET_ADDRESS_SIZE);
3963
b += 1 + OFFSET_ADDRESS_SIZE;
3964
}
3965
3966
/* Otherwise, we have a nontrivial interval. When
3967
we're all done, the pattern will look like:
3968
set_number_at <jump count> <upper bound>
3969
set_number_at <succeed_n count> <lower bound>
3970
succeed_n <after jump addr> <succeed_n count>
3971
<body of loop>
3972
jump_n <succeed_n addr> <jump count>
3973
(The upper bound and `jump_n' are omitted if
3974
`upper_bound' is 1, though.) */
3975
else
3976
{ /* If the upper bound is > 1, we need to insert
3977
more at the end of the loop. */
3978
unsigned nbytes = 2 + 4 * OFFSET_ADDRESS_SIZE +
3979
(upper_bound > 1) * (2 + 4 * OFFSET_ADDRESS_SIZE);
3980
3981
GET_BUFFER_SPACE (nbytes);
3982
3983
/* Initialize lower bound of the `succeed_n', even
3984
though it will be set during matching by its
3985
attendant `set_number_at' (inserted next),
3986
because `re_compile_fastmap' needs to know.
3987
Jump to the `jump_n' we might insert below. */
3988
INSERT_JUMP2 (succeed_n, laststart,
3989
b + 1 + 2 * OFFSET_ADDRESS_SIZE
3990
+ (upper_bound > 1) * (1 + 2 * OFFSET_ADDRESS_SIZE)
3991
, lower_bound);
3992
b += 1 + 2 * OFFSET_ADDRESS_SIZE;
3993
3994
/* Code to initialize the lower bound. Insert
3995
before the `succeed_n'. The `5' is the last two
3996
bytes of this `set_number_at', plus 3 bytes of
3997
the following `succeed_n'. */
3998
/* ifdef WCHAR, The '1+2*OFFSET_ADDRESS_SIZE'
3999
is the 'set_number_at', plus '1+OFFSET_ADDRESS_SIZE'
4000
of the following `succeed_n'. */
4001
PREFIX(insert_op2) (set_number_at, laststart, 1
4002
+ 2 * OFFSET_ADDRESS_SIZE, lower_bound, b);
4003
b += 1 + 2 * OFFSET_ADDRESS_SIZE;
4004
4005
if (upper_bound > 1)
4006
{ /* More than one repetition is allowed, so
4007
append a backward jump to the `succeed_n'
4008
that starts this interval.
4009
4010
When we've reached this during matching,
4011
we'll have matched the interval once, so
4012
jump back only `upper_bound - 1' times. */
4013
STORE_JUMP2 (jump_n, b, laststart
4014
+ 2 * OFFSET_ADDRESS_SIZE + 1,
4015
upper_bound - 1);
4016
b += 1 + 2 * OFFSET_ADDRESS_SIZE;
4017
4018
/* The location we want to set is the second
4019
parameter of the `jump_n'; that is `b-2' as
4020
an absolute address. `laststart' will be
4021
the `set_number_at' we're about to insert;
4022
`laststart+3' the number to set, the source
4023
for the relative address. But we are
4024
inserting into the middle of the pattern --
4025
so everything is getting moved up by 5.
4026
Conclusion: (b - 2) - (laststart + 3) + 5,
4027
i.e., b - laststart.
4028
4029
We insert this at the beginning of the loop
4030
so that if we fail during matching, we'll
4031
reinitialize the bounds. */
4032
PREFIX(insert_op2) (set_number_at, laststart,
4033
b - laststart,
4034
upper_bound - 1, b);
4035
b += 1 + 2 * OFFSET_ADDRESS_SIZE;
4036
}
4037
}
4038
pending_exact = 0;
4039
break;
4040
4041
invalid_interval:
4042
if (!(syntax & RE_INVALID_INTERVAL_ORD))
4043
FREE_STACK_RETURN (p == pend ? REG_EBRACE : REG_BADBR);
4044
unfetch_interval:
4045
/* Match the characters as literals. */
4046
p = beg_interval;
4047
c = '{';
4048
if (syntax & RE_NO_BK_BRACES)
4049
goto normal_char;
4050
else
4051
goto normal_backslash;
4052
}
4053
4054
#ifdef emacs
4055
/* There is no way to specify the before_dot and after_dot
4056
operators. rms says this is ok. --karl */
4057
case '=':
4058
BUF_PUSH (at_dot);
4059
break;
4060
4061
case 's':
4062
laststart = b;
4063
PATFETCH (c);
4064
BUF_PUSH_2 (syntaxspec, syntax_spec_code[c]);
4065
break;
4066
4067
case 'S':
4068
laststart = b;
4069
PATFETCH (c);
4070
BUF_PUSH_2 (notsyntaxspec, syntax_spec_code[c]);
4071
break;
4072
#endif /* emacs */
4073
4074
4075
case 'w':
4076
if (syntax & RE_NO_GNU_OPS)
4077
goto normal_char;
4078
laststart = b;
4079
BUF_PUSH (wordchar);
4080
break;
4081
4082
4083
case 'W':
4084
if (syntax & RE_NO_GNU_OPS)
4085
goto normal_char;
4086
laststart = b;
4087
BUF_PUSH (notwordchar);
4088
break;
4089
4090
4091
case '<':
4092
if (syntax & RE_NO_GNU_OPS)
4093
goto normal_char;
4094
BUF_PUSH (wordbeg);
4095
break;
4096
4097
case '>':
4098
if (syntax & RE_NO_GNU_OPS)
4099
goto normal_char;
4100
BUF_PUSH (wordend);
4101
break;
4102
4103
case 'b':
4104
if (syntax & RE_NO_GNU_OPS)
4105
goto normal_char;
4106
BUF_PUSH (wordbound);
4107
break;
4108
4109
case 'B':
4110
if (syntax & RE_NO_GNU_OPS)
4111
goto normal_char;
4112
BUF_PUSH (notwordbound);
4113
break;
4114
4115
case '`':
4116
if (syntax & RE_NO_GNU_OPS)
4117
goto normal_char;
4118
BUF_PUSH (begbuf);
4119
break;
4120
4121
case '\'':
4122
if (syntax & RE_NO_GNU_OPS)
4123
goto normal_char;
4124
BUF_PUSH (endbuf);
4125
break;
4126
4127
case '1': case '2': case '3': case '4': case '5':
4128
case '6': case '7': case '8': case '9':
4129
if (syntax & RE_NO_BK_REFS)
4130
goto normal_char;
4131
4132
c1 = c - '0';
4133
4134
if (c1 > regnum)
4135
FREE_STACK_RETURN (REG_ESUBREG);
4136
4137
/* Can't back reference to a subexpression if inside of it. */
4138
if (group_in_compile_stack (compile_stack, (regnum_t) c1))
4139
goto normal_char;
4140
4141
laststart = b;
4142
BUF_PUSH_2 (duplicate, c1);
4143
break;
4144
4145
4146
case '+':
4147
case '?':
4148
if (syntax & RE_BK_PLUS_QM)
4149
goto handle_plus;
4150
else
4151
goto normal_backslash;
4152
4153
default:
4154
normal_backslash:
4155
/* You might think it would be useful for \ to mean
4156
not to translate; but if we don't translate it
4157
it will never match anything. */
4158
c = TRANSLATE (c);
4159
goto normal_char;
4160
}
4161
break;
4162
4163
4164
default:
4165
/* Expects the character in `c'. */
4166
normal_char:
4167
/* If no exactn currently being built. */
4168
if (!pending_exact
4169
#ifdef WCHAR
4170
/* If last exactn handle binary(or character) and
4171
new exactn handle character(or binary). */
4172
|| is_exactn_bin != is_binary[p - 1 - pattern]
4173
#endif /* WCHAR */
4174
4175
/* If last exactn not at current position. */
4176
|| pending_exact + *pending_exact + 1 != b
4177
4178
/* We have only one byte following the exactn for the count. */
4179
|| *pending_exact == (1 << BYTEWIDTH) - 1
4180
4181
/* If followed by a repetition operator. */
4182
|| *p == '*' || *p == '^'
4183
|| ((syntax & RE_BK_PLUS_QM)
4184
? *p == '\\' && (p[1] == '+' || p[1] == '?')
4185
: (*p == '+' || *p == '?'))
4186
|| ((syntax & RE_INTERVALS)
4187
&& ((syntax & RE_NO_BK_BRACES)
4188
? *p == '{'
4189
: (p[0] == '\\' && p[1] == '{'))))
4190
{
4191
/* Start building a new exactn. */
4192
4193
laststart = b;
4194
4195
#ifdef WCHAR
4196
/* Is this exactn binary data or character? */
4197
is_exactn_bin = is_binary[p - 1 - pattern];
4198
if (is_exactn_bin)
4199
BUF_PUSH_2 (exactn_bin, 0);
4200
else
4201
BUF_PUSH_2 (exactn, 0);
4202
#else
4203
BUF_PUSH_2 (exactn, 0);
4204
#endif /* WCHAR */
4205
pending_exact = b - 1;
4206
}
4207
4208
BUF_PUSH (c);
4209
(*pending_exact)++;
4210
break;
4211
} /* switch (c) */
4212
} /* while p != pend */
4213
4214
4215
/* Through the pattern now. */
4216
4217
if (fixup_alt_jump)
4218
STORE_JUMP (jump_past_alt, fixup_alt_jump, b);
4219
4220
if (!COMPILE_STACK_EMPTY)
4221
FREE_STACK_RETURN (REG_EPAREN);
4222
4223
/* If we don't want backtracking, force success
4224
the first time we reach the end of the compiled pattern. */
4225
if (syntax & RE_NO_POSIX_BACKTRACKING)
4226
BUF_PUSH (succeed);
4227
4228
#ifdef WCHAR
4229
free (pattern);
4230
free (mbs_offset);
4231
free (is_binary);
4232
#endif
4233
free (compile_stack.stack);
4234
4235
/* We have succeeded; set the length of the buffer. */
4236
#ifdef WCHAR
4237
bufp->used = (uintptr_t) b - (uintptr_t) COMPILED_BUFFER_VAR;
4238
#else
4239
bufp->used = b - bufp->buffer;
4240
#endif
4241
4242
#ifdef DEBUG
4243
if (debug)
4244
{
4245
DEBUG_PRINT1 ("\nCompiled pattern: \n");
4246
PREFIX(print_compiled_pattern) (bufp);
4247
}
4248
#endif /* DEBUG */
4249
4250
#ifndef MATCH_MAY_ALLOCATE
4251
/* Initialize the failure stack to the largest possible stack. This
4252
isn't necessary unless we're trying to avoid calling alloca in
4253
the search and match routines. */
4254
{
4255
int num_regs = bufp->re_nsub + 1;
4256
4257
/* Since DOUBLE_FAIL_STACK refuses to double only if the current size
4258
is strictly greater than re_max_failures, the largest possible stack
4259
is 2 * re_max_failures failure points. */
4260
if (fail_stack.size < (2 * re_max_failures * MAX_FAILURE_ITEMS))
4261
{
4262
fail_stack.size = (2 * re_max_failures * MAX_FAILURE_ITEMS);
4263
4264
# ifdef emacs
4265
if (! fail_stack.stack)
4266
fail_stack.stack
4267
= (PREFIX(fail_stack_elt_t) *) xmalloc (fail_stack.size
4268
* sizeof (PREFIX(fail_stack_elt_t)));
4269
else
4270
fail_stack.stack
4271
= (PREFIX(fail_stack_elt_t) *) xrealloc (fail_stack.stack,
4272
(fail_stack.size
4273
* sizeof (PREFIX(fail_stack_elt_t))));
4274
# else /* not emacs */
4275
if (! fail_stack.stack)
4276
fail_stack.stack
4277
= (PREFIX(fail_stack_elt_t) *) malloc (fail_stack.size
4278
* sizeof (PREFIX(fail_stack_elt_t)));
4279
else
4280
fail_stack.stack
4281
= (PREFIX(fail_stack_elt_t) *) realloc (fail_stack.stack,
4282
(fail_stack.size
4283
* sizeof (PREFIX(fail_stack_elt_t))));
4284
# endif /* not emacs */
4285
}
4286
4287
PREFIX(regex_grow_registers) (num_regs);
4288
}
4289
#endif /* not MATCH_MAY_ALLOCATE */
4290
4291
return REG_NOERROR;
4292
} /* regex_compile */
4293
4294
/* Subroutines for `regex_compile'. */
4295
4296
/* Store OP at LOC followed by two-byte integer parameter ARG. */
4297
/* ifdef WCHAR, integer parameter is 1 wchar_t. */
4298
4299
static void
4300
PREFIX(store_op1) (op, loc, arg)
4301
re_opcode_t op;
4302
UCHAR_T *loc;
4303
int arg;
4304
{
4305
*loc = (UCHAR_T) op;
4306
STORE_NUMBER (loc + 1, arg);
4307
}
4308
4309
4310
/* Like `store_op1', but for two two-byte parameters ARG1 and ARG2. */
4311
/* ifdef WCHAR, integer parameter is 1 wchar_t. */
4312
4313
static void
4314
PREFIX(store_op2) (op, loc, arg1, arg2)
4315
re_opcode_t op;
4316
UCHAR_T *loc;
4317
int arg1, arg2;
4318
{
4319
*loc = (UCHAR_T) op;
4320
STORE_NUMBER (loc + 1, arg1);
4321
STORE_NUMBER (loc + 1 + OFFSET_ADDRESS_SIZE, arg2);
4322
}
4323
4324
4325
/* Copy the bytes from LOC to END to open up three bytes of space at LOC
4326
for OP followed by two-byte integer parameter ARG. */
4327
/* ifdef WCHAR, integer parameter is 1 wchar_t. */
4328
4329
static void
4330
PREFIX(insert_op1) (op, loc, arg, end)
4331
re_opcode_t op;
4332
UCHAR_T *loc;
4333
int arg;
4334
UCHAR_T *end;
4335
{
4336
register UCHAR_T *pfrom = end;
4337
register UCHAR_T *pto = end + 1 + OFFSET_ADDRESS_SIZE;
4338
4339
while (pfrom != loc)
4340
*--pto = *--pfrom;
4341
4342
PREFIX(store_op1) (op, loc, arg);
4343
}
4344
4345
4346
/* Like `insert_op1', but for two two-byte parameters ARG1 and ARG2. */
4347
/* ifdef WCHAR, integer parameter is 1 wchar_t. */
4348
4349
static void
4350
PREFIX(insert_op2) (op, loc, arg1, arg2, end)
4351
re_opcode_t op;
4352
UCHAR_T *loc;
4353
int arg1, arg2;
4354
UCHAR_T *end;
4355
{
4356
register UCHAR_T *pfrom = end;
4357
register UCHAR_T *pto = end + 1 + 2 * OFFSET_ADDRESS_SIZE;
4358
4359
while (pfrom != loc)
4360
*--pto = *--pfrom;
4361
4362
PREFIX(store_op2) (op, loc, arg1, arg2);
4363
}
4364
4365
4366
/* P points to just after a ^ in PATTERN. Return true if that ^ comes
4367
after an alternative or a begin-subexpression. We assume there is at
4368
least one character before the ^. */
4369
4370
static boolean
4371
PREFIX(at_begline_loc_p) (pattern, p, syntax)
4372
const CHAR_T *pattern, *p;
4373
reg_syntax_t syntax;
4374
{
4375
const CHAR_T *prev = p - 2;
4376
boolean prev_prev_backslash = prev > pattern && prev[-1] == '\\';
4377
4378
return
4379
/* After a subexpression? */
4380
(*prev == '(' && (syntax & RE_NO_BK_PARENS || prev_prev_backslash))
4381
/* After an alternative? */
4382
|| (*prev == '|' && (syntax & RE_NO_BK_VBAR || prev_prev_backslash));
4383
}
4384
4385
4386
/* The dual of at_begline_loc_p. This one is for $. We assume there is
4387
at least one character after the $, i.e., `P < PEND'. */
4388
4389
static boolean
4390
PREFIX(at_endline_loc_p) (p, pend, syntax)
4391
const CHAR_T *p, *pend;
4392
reg_syntax_t syntax;
4393
{
4394
const CHAR_T *next = p;
4395
boolean next_backslash = *next == '\\';
4396
const CHAR_T *next_next = p + 1 < pend ? p + 1 : 0;
4397
4398
return
4399
/* Before a subexpression? */
4400
(syntax & RE_NO_BK_PARENS ? *next == ')'
4401
: next_backslash && next_next && *next_next == ')')
4402
/* Before an alternative? */
4403
|| (syntax & RE_NO_BK_VBAR ? *next == '|'
4404
: next_backslash && next_next && *next_next == '|');
4405
}
4406
4407
#else /* not INSIDE_RECURSION */
4408
4409
/* Returns true if REGNUM is in one of COMPILE_STACK's elements and
4410
false if it's not. */
4411
4412
static boolean
4413
group_in_compile_stack (compile_stack, regnum)
4414
compile_stack_type compile_stack;
4415
regnum_t regnum;
4416
{
4417
int this_element;
4418
4419
for (this_element = compile_stack.avail - 1;
4420
this_element >= 0;
4421
this_element--)
4422
if (compile_stack.stack[this_element].regnum == regnum)
4423
return true;
4424
4425
return false;
4426
}
4427
#endif /* not INSIDE_RECURSION */
4428
4429
#ifdef INSIDE_RECURSION
4430
4431
#ifdef WCHAR
4432
/* This insert space, which size is "num", into the pattern at "loc".
4433
"end" must point the end of the allocated buffer. */
4434
static void
4435
insert_space (num, loc, end)
4436
int num;
4437
CHAR_T *loc;
4438
CHAR_T *end;
4439
{
4440
register CHAR_T *pto = end;
4441
register CHAR_T *pfrom = end - num;
4442
4443
while (pfrom >= loc)
4444
*pto-- = *pfrom--;
4445
}
4446
#endif /* WCHAR */
4447
4448
#ifdef WCHAR
4449
static reg_errcode_t
4450
wcs_compile_range (range_start_char, p_ptr, pend, translate, syntax, b,
4451
char_set)
4452
CHAR_T range_start_char;
4453
const CHAR_T **p_ptr, *pend;
4454
CHAR_T *char_set, *b;
4455
RE_TRANSLATE_TYPE translate;
4456
reg_syntax_t syntax;
4457
{
4458
const CHAR_T *p = *p_ptr;
4459
CHAR_T range_start, range_end;
4460
reg_errcode_t ret;
4461
# ifdef _LIBC
4462
uint32_t nrules;
4463
uint32_t start_val, end_val;
4464
# endif
4465
if (p == pend)
4466
return REG_ERANGE;
4467
4468
# ifdef _LIBC
4469
nrules = _NL_CURRENT_WORD (LC_COLLATE, _NL_COLLATE_NRULES);
4470
if (nrules != 0)
4471
{
4472
const char *collseq = (const char *) _NL_CURRENT(LC_COLLATE,
4473
_NL_COLLATE_COLLSEQWC);
4474
const unsigned char *extra = (const unsigned char *)
4475
_NL_CURRENT (LC_COLLATE, _NL_COLLATE_SYMB_EXTRAMB);
4476
4477
if (range_start_char < -1)
4478
{
4479
/* range_start is a collating symbol. */
4480
int32_t *wextra;
4481
/* Retreive the index and get collation sequence value. */
4482
wextra = (int32_t*)(extra + char_set[-range_start_char]);
4483
start_val = wextra[1 + *wextra];
4484
}
4485
else
4486
start_val = collseq_table_lookup(collseq, TRANSLATE(range_start_char));
4487
4488
end_val = collseq_table_lookup (collseq, TRANSLATE (p[0]));
4489
4490
/* Report an error if the range is empty and the syntax prohibits
4491
this. */
4492
ret = ((syntax & RE_NO_EMPTY_RANGES)
4493
&& (start_val > end_val))? REG_ERANGE : REG_NOERROR;
4494
4495
/* Insert space to the end of the char_ranges. */
4496
insert_space(2, b - char_set[5] - 2, b - 1);
4497
*(b - char_set[5] - 2) = (wchar_t)start_val;
4498
*(b - char_set[5] - 1) = (wchar_t)end_val;
4499
char_set[4]++; /* ranges_index */
4500
}
4501
else
4502
# endif
4503
{
4504
range_start = (range_start_char >= 0)? TRANSLATE (range_start_char):
4505
range_start_char;
4506
range_end = TRANSLATE (p[0]);
4507
/* Report an error if the range is empty and the syntax prohibits
4508
this. */
4509
ret = ((syntax & RE_NO_EMPTY_RANGES)
4510
&& (range_start > range_end))? REG_ERANGE : REG_NOERROR;
4511
4512
/* Insert space to the end of the char_ranges. */
4513
insert_space(2, b - char_set[5] - 2, b - 1);
4514
*(b - char_set[5] - 2) = range_start;
4515
*(b - char_set[5] - 1) = range_end;
4516
char_set[4]++; /* ranges_index */
4517
}
4518
/* Have to increment the pointer into the pattern string, so the
4519
caller isn't still at the ending character. */
4520
(*p_ptr)++;
4521
4522
return ret;
4523
}
4524
#else /* BYTE */
4525
/* Read the ending character of a range (in a bracket expression) from the
4526
uncompiled pattern *P_PTR (which ends at PEND). We assume the
4527
starting character is in `P[-2]'. (`P[-1]' is the character `-'.)
4528
Then we set the translation of all bits between the starting and
4529
ending characters (inclusive) in the compiled pattern B.
4530
4531
Return an error code.
4532
4533
We use these short variable names so we can use the same macros as
4534
`regex_compile' itself. */
4535
4536
static reg_errcode_t
4537
byte_compile_range (range_start_char, p_ptr, pend, translate, syntax, b)
4538
unsigned int range_start_char;
4539
const char **p_ptr, *pend;
4540
RE_TRANSLATE_TYPE translate;
4541
reg_syntax_t syntax;
4542
unsigned char *b;
4543
{
4544
unsigned this_char;
4545
const char *p = *p_ptr;
4546
reg_errcode_t ret;
4547
# if _LIBC
4548
const unsigned char *collseq;
4549
unsigned int start_colseq;
4550
unsigned int end_colseq;
4551
# else
4552
unsigned end_char;
4553
# endif
4554
4555
if (p == pend)
4556
return REG_ERANGE;
4557
4558
/* Have to increment the pointer into the pattern string, so the
4559
caller isn't still at the ending character. */
4560
(*p_ptr)++;
4561
4562
/* Report an error if the range is empty and the syntax prohibits this. */
4563
ret = syntax & RE_NO_EMPTY_RANGES ? REG_ERANGE : REG_NOERROR;
4564
4565
# if _LIBC
4566
collseq = (const unsigned char *) _NL_CURRENT (LC_COLLATE,
4567
_NL_COLLATE_COLLSEQMB);
4568
4569
start_colseq = collseq[(unsigned char) TRANSLATE (range_start_char)];
4570
end_colseq = collseq[(unsigned char) TRANSLATE (p[0])];
4571
for (this_char = 0; this_char <= (unsigned char) -1; ++this_char)
4572
{
4573
unsigned int this_colseq = collseq[(unsigned char) TRANSLATE (this_char)];
4574
4575
if (start_colseq <= this_colseq && this_colseq <= end_colseq)
4576
{
4577
SET_LIST_BIT (TRANSLATE (this_char));
4578
ret = REG_NOERROR;
4579
}
4580
}
4581
# else
4582
/* Here we see why `this_char' has to be larger than an `unsigned
4583
char' -- we would otherwise go into an infinite loop, since all
4584
characters <= 0xff. */
4585
range_start_char = TRANSLATE (range_start_char);
4586
/* TRANSLATE(p[0]) is casted to char (not unsigned char) in TRANSLATE,
4587
and some compilers cast it to int implicitly, so following for_loop
4588
may fall to (almost) infinite loop.
4589
e.g. If translate[p[0]] = 0xff, end_char may equals to 0xffffffff.
4590
To avoid this, we cast p[0] to unsigned int and truncate it. */
4591
end_char = ((unsigned)TRANSLATE(p[0]) & ((1 << BYTEWIDTH) - 1));
4592
4593
for (this_char = range_start_char; this_char <= end_char; ++this_char)
4594
{
4595
SET_LIST_BIT (TRANSLATE (this_char));
4596
ret = REG_NOERROR;
4597
}
4598
# endif
4599
4600
return ret;
4601
}
4602
#endif /* WCHAR */
4603
4604
/* re_compile_fastmap computes a ``fastmap'' for the compiled pattern in
4605
BUFP. A fastmap records which of the (1 << BYTEWIDTH) possible
4606
characters can start a string that matches the pattern. This fastmap
4607
is used by re_search to skip quickly over impossible starting points.
4608
4609
The caller must supply the address of a (1 << BYTEWIDTH)-byte data
4610
area as BUFP->fastmap.
4611
4612
We set the `fastmap', `fastmap_accurate', and `can_be_null' fields in
4613
the pattern buffer.
4614
4615
Returns 0 if we succeed, -2 if an internal error. */
4616
4617
#ifdef WCHAR
4618
/* local function for re_compile_fastmap.
4619
truncate wchar_t character to char. */
4620
static unsigned char truncate_wchar (CHAR_T c);
4621
4622
static unsigned char
4623
truncate_wchar (c)
4624
CHAR_T c;
4625
{
4626
unsigned char buf[MB_CUR_MAX];
4627
mbstate_t state;
4628
int retval;
4629
memset (&state, '\0', sizeof (state));
4630
retval = wcrtomb (buf, c, &state);
4631
return retval > 0 ? buf[0] : (unsigned char) c;
4632
}
4633
#endif /* WCHAR */
4634
4635
static int
4636
PREFIX(re_compile_fastmap) (bufp)
4637
struct re_pattern_buffer *bufp;
4638
{
4639
int j, k;
4640
#ifdef MATCH_MAY_ALLOCATE
4641
PREFIX(fail_stack_type) fail_stack;
4642
#endif
4643
#ifndef REGEX_MALLOC
4644
char *destination;
4645
#endif
4646
4647
register char *fastmap = bufp->fastmap;
4648
4649
#ifdef WCHAR
4650
/* We need to cast pattern to (wchar_t*), because we casted this compiled
4651
pattern to (char*) in regex_compile. */
4652
UCHAR_T *pattern = (UCHAR_T*)bufp->buffer;
4653
register UCHAR_T *pend = (UCHAR_T*) (bufp->buffer + bufp->used);
4654
#else /* BYTE */
4655
UCHAR_T *pattern = bufp->buffer;
4656
register UCHAR_T *pend = pattern + bufp->used;
4657
#endif /* WCHAR */
4658
UCHAR_T *p = pattern;
4659
4660
#ifdef REL_ALLOC
4661
/* This holds the pointer to the failure stack, when
4662
it is allocated relocatably. */
4663
fail_stack_elt_t *failure_stack_ptr;
4664
#endif
4665
4666
/* Assume that each path through the pattern can be null until
4667
proven otherwise. We set this false at the bottom of switch
4668
statement, to which we get only if a particular path doesn't
4669
match the empty string. */
4670
boolean path_can_be_null = true;
4671
4672
/* We aren't doing a `succeed_n' to begin with. */
4673
boolean succeed_n_p = false;
4674
4675
assert (fastmap != NULL && p != NULL);
4676
4677
INIT_FAIL_STACK ();
4678
bzero (fastmap, 1 << BYTEWIDTH); /* Assume nothing's valid. */
4679
bufp->fastmap_accurate = 1; /* It will be when we're done. */
4680
bufp->can_be_null = 0;
4681
4682
while (1)
4683
{
4684
if (p == pend || *p == succeed)
4685
{
4686
/* We have reached the (effective) end of pattern. */
4687
if (!FAIL_STACK_EMPTY ())
4688
{
4689
bufp->can_be_null |= path_can_be_null;
4690
4691
/* Reset for next path. */
4692
path_can_be_null = true;
4693
4694
p = fail_stack.stack[--fail_stack.avail].pointer;
4695
4696
continue;
4697
}
4698
else
4699
break;
4700
}
4701
4702
/* We should never be about to go beyond the end of the pattern. */
4703
assert (p < pend);
4704
4705
switch (SWITCH_ENUM_CAST ((re_opcode_t) *p++))
4706
{
4707
4708
/* I guess the idea here is to simply not bother with a fastmap
4709
if a backreference is used, since it's too hard to figure out
4710
the fastmap for the corresponding group. Setting
4711
`can_be_null' stops `re_search_2' from using the fastmap, so
4712
that is all we do. */
4713
case duplicate:
4714
bufp->can_be_null = 1;
4715
goto done;
4716
4717
4718
/* Following are the cases which match a character. These end
4719
with `break'. */
4720
4721
#ifdef WCHAR
4722
case exactn:
4723
fastmap[truncate_wchar(p[1])] = 1;
4724
break;
4725
#else /* BYTE */
4726
case exactn:
4727
fastmap[p[1]] = 1;
4728
break;
4729
#endif /* WCHAR */
4730
#ifdef MBS_SUPPORT
4731
case exactn_bin:
4732
fastmap[p[1]] = 1;
4733
break;
4734
#endif
4735
4736
#ifdef WCHAR
4737
/* It is hard to distinguish fastmap from (multi byte) characters
4738
which depends on current locale. */
4739
case charset:
4740
case charset_not:
4741
case wordchar:
4742
case notwordchar:
4743
bufp->can_be_null = 1;
4744
goto done;
4745
#else /* BYTE */
4746
case charset:
4747
for (j = *p++ * BYTEWIDTH - 1; j >= 0; j--)
4748
if (p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH)))
4749
fastmap[j] = 1;
4750
break;
4751
4752
4753
case charset_not:
4754
/* Chars beyond end of map must be allowed. */
4755
for (j = *p * BYTEWIDTH; j < (1 << BYTEWIDTH); j++)
4756
fastmap[j] = 1;
4757
4758
for (j = *p++ * BYTEWIDTH - 1; j >= 0; j--)
4759
if (!(p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH))))
4760
fastmap[j] = 1;
4761
break;
4762
4763
4764
case wordchar:
4765
for (j = 0; j < (1 << BYTEWIDTH); j++)
4766
if (SYNTAX (j) == Sword)
4767
fastmap[j] = 1;
4768
break;
4769
4770
4771
case notwordchar:
4772
for (j = 0; j < (1 << BYTEWIDTH); j++)
4773
if (SYNTAX (j) != Sword)
4774
fastmap[j] = 1;
4775
break;
4776
#endif /* WCHAR */
4777
4778
case anychar:
4779
{
4780
int fastmap_newline = fastmap['\n'];
4781
4782
/* `.' matches anything ... */
4783
for (j = 0; j < (1 << BYTEWIDTH); j++)
4784
fastmap[j] = 1;
4785
4786
/* ... except perhaps newline. */
4787
if (!(bufp->syntax & RE_DOT_NEWLINE))
4788
fastmap['\n'] = fastmap_newline;
4789
4790
/* Return if we have already set `can_be_null'; if we have,
4791
then the fastmap is irrelevant. Something's wrong here. */
4792
else if (bufp->can_be_null)
4793
goto done;
4794
4795
/* Otherwise, have to check alternative paths. */
4796
break;
4797
}
4798
4799
#ifdef emacs
4800
case syntaxspec:
4801
k = *p++;
4802
for (j = 0; j < (1 << BYTEWIDTH); j++)
4803
if (SYNTAX (j) == (enum syntaxcode) k)
4804
fastmap[j] = 1;
4805
break;
4806
4807
4808
case notsyntaxspec:
4809
k = *p++;
4810
for (j = 0; j < (1 << BYTEWIDTH); j++)
4811
if (SYNTAX (j) != (enum syntaxcode) k)
4812
fastmap[j] = 1;
4813
break;
4814
4815
4816
/* All cases after this match the empty string. These end with
4817
`continue'. */
4818
4819
4820
case before_dot:
4821
case at_dot:
4822
case after_dot:
4823
continue;
4824
#endif /* emacs */
4825
4826
4827
case no_op:
4828
case begline:
4829
case endline:
4830
case begbuf:
4831
case endbuf:
4832
case wordbound:
4833
case notwordbound:
4834
case wordbeg:
4835
case wordend:
4836
case push_dummy_failure:
4837
continue;
4838
4839
4840
case jump_n:
4841
case pop_failure_jump:
4842
case maybe_pop_jump:
4843
case jump:
4844
case jump_past_alt:
4845
case dummy_failure_jump:
4846
EXTRACT_NUMBER_AND_INCR (j, p);
4847
p += j;
4848
if (j > 0)
4849
continue;
4850
4851
/* Jump backward implies we just went through the body of a
4852
loop and matched nothing. Opcode jumped to should be
4853
`on_failure_jump' or `succeed_n'. Just treat it like an
4854
ordinary jump. For a * loop, it has pushed its failure
4855
point already; if so, discard that as redundant. */
4856
if ((re_opcode_t) *p != on_failure_jump
4857
&& (re_opcode_t) *p != succeed_n)
4858
continue;
4859
4860
p++;
4861
EXTRACT_NUMBER_AND_INCR (j, p);
4862
p += j;
4863
4864
/* If what's on the stack is where we are now, pop it. */
4865
if (!FAIL_STACK_EMPTY ()
4866
&& fail_stack.stack[fail_stack.avail - 1].pointer == p)
4867
fail_stack.avail--;
4868
4869
continue;
4870
4871
4872
case on_failure_jump:
4873
case on_failure_keep_string_jump:
4874
handle_on_failure_jump:
4875
EXTRACT_NUMBER_AND_INCR (j, p);
4876
4877
/* For some patterns, e.g., `(a?)?', `p+j' here points to the
4878
end of the pattern. We don't want to push such a point,
4879
since when we restore it above, entering the switch will
4880
increment `p' past the end of the pattern. We don't need
4881
to push such a point since we obviously won't find any more
4882
fastmap entries beyond `pend'. Such a pattern can match
4883
the null string, though. */
4884
if (p + j < pend)
4885
{
4886
if (!PUSH_PATTERN_OP (p + j, fail_stack))
4887
{
4888
RESET_FAIL_STACK ();
4889
return -2;
4890
}
4891
}
4892
else
4893
bufp->can_be_null = 1;
4894
4895
if (succeed_n_p)
4896
{
4897
EXTRACT_NUMBER_AND_INCR (k, p); /* Skip the n. */
4898
succeed_n_p = false;
4899
}
4900
4901
continue;
4902
4903
4904
case succeed_n:
4905
/* Get to the number of times to succeed. */
4906
p += OFFSET_ADDRESS_SIZE;
4907
4908
/* Increment p past the n for when k != 0. */
4909
EXTRACT_NUMBER_AND_INCR (k, p);
4910
if (k == 0)
4911
{
4912
p -= 2 * OFFSET_ADDRESS_SIZE;
4913
succeed_n_p = true; /* Spaghetti code alert. */
4914
goto handle_on_failure_jump;
4915
}
4916
continue;
4917
4918
4919
case set_number_at:
4920
p += 2 * OFFSET_ADDRESS_SIZE;
4921
continue;
4922
4923
4924
case start_memory:
4925
case stop_memory:
4926
p += 2;
4927
continue;
4928
4929
4930
default:
4931
abort (); /* We have listed all the cases. */
4932
} /* switch *p++ */
4933
4934
/* Getting here means we have found the possible starting
4935
characters for one path of the pattern -- and that the empty
4936
string does not match. We need not follow this path further.
4937
Instead, look at the next alternative (remembered on the
4938
stack), or quit if no more. The test at the top of the loop
4939
does these things. */
4940
path_can_be_null = false;
4941
p = pend;
4942
} /* while p */
4943
4944
/* Set `can_be_null' for the last path (also the first path, if the
4945
pattern is empty). */
4946
bufp->can_be_null |= path_can_be_null;
4947
4948
done:
4949
RESET_FAIL_STACK ();
4950
return 0;
4951
}
4952
4953
#else /* not INSIDE_RECURSION */
4954
4955
int
4956
re_compile_fastmap (bufp)
4957
struct re_pattern_buffer *bufp;
4958
{
4959
# ifdef MBS_SUPPORT
4960
if (MB_CUR_MAX != 1)
4961
return wcs_re_compile_fastmap(bufp);
4962
else
4963
# endif
4964
return byte_re_compile_fastmap(bufp);
4965
} /* re_compile_fastmap */
4966
#ifdef _LIBC
4967
weak_alias (__re_compile_fastmap, re_compile_fastmap)
4968
#endif
4969
4970
4971
/* Set REGS to hold NUM_REGS registers, storing them in STARTS and
4972
ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use
4973
this memory for recording register information. STARTS and ENDS
4974
must be allocated using the malloc library routine, and must each
4975
be at least NUM_REGS * sizeof (regoff_t) bytes long.
4976
4977
If NUM_REGS == 0, then subsequent matches should allocate their own
4978
register data.
4979
4980
Unless this function is called, the first search or match using
4981
PATTERN_BUFFER will allocate its own register data, without
4982
freeing the old data. */
4983
4984
void
4985
re_set_registers (bufp, regs, num_regs, starts, ends)
4986
struct re_pattern_buffer *bufp;
4987
struct re_registers *regs;
4988
unsigned num_regs;
4989
regoff_t *starts, *ends;
4990
{
4991
if (num_regs)
4992
{
4993
bufp->regs_allocated = REGS_REALLOCATE;
4994
regs->num_regs = num_regs;
4995
regs->start = starts;
4996
regs->end = ends;
4997
}
4998
else
4999
{
5000
bufp->regs_allocated = REGS_UNALLOCATED;
5001
regs->num_regs = 0;
5002
regs->start = regs->end = (regoff_t *) 0;
5003
}
5004
}
5005
#ifdef _LIBC
5006
weak_alias (__re_set_registers, re_set_registers)
5007
#endif
5008
5009
/* Searching routines. */
5010
5011
/* Like re_search_2, below, but only one string is specified, and
5012
doesn't let you say where to stop matching. */
5013
5014
int
5015
re_search (bufp, string, size, startpos, range, regs)
5016
struct re_pattern_buffer *bufp;
5017
const char *string;
5018
int size, startpos, range;
5019
struct re_registers *regs;
5020
{
5021
return re_search_2 (bufp, NULL, 0, string, size, startpos, range,
5022
regs, size);
5023
}
5024
#ifdef _LIBC
5025
weak_alias (__re_search, re_search)
5026
#endif
5027
5028
5029
/* Using the compiled pattern in BUFP->buffer, first tries to match the
5030
virtual concatenation of STRING1 and STRING2, starting first at index
5031
STARTPOS, then at STARTPOS + 1, and so on.
5032
5033
STRING1 and STRING2 have length SIZE1 and SIZE2, respectively.
5034
5035
RANGE is how far to scan while trying to match. RANGE = 0 means try
5036
only at STARTPOS; in general, the last start tried is STARTPOS +
5037
RANGE.
5038
5039
In REGS, return the indices of the virtual concatenation of STRING1
5040
and STRING2 that matched the entire BUFP->buffer and its contained
5041
subexpressions.
5042
5043
Do not consider matching one past the index STOP in the virtual
5044
concatenation of STRING1 and STRING2.
5045
5046
We return either the position in the strings at which the match was
5047
found, -1 if no match, or -2 if error (such as failure
5048
stack overflow). */
5049
5050
int
5051
re_search_2 (bufp, string1, size1, string2, size2, startpos, range, regs, stop)
5052
struct re_pattern_buffer *bufp;
5053
const char *string1, *string2;
5054
int size1, size2;
5055
int startpos;
5056
int range;
5057
struct re_registers *regs;
5058
int stop;
5059
{
5060
# ifdef MBS_SUPPORT
5061
if (MB_CUR_MAX != 1)
5062
return wcs_re_search_2 (bufp, string1, size1, string2, size2, startpos,
5063
range, regs, stop);
5064
else
5065
# endif
5066
return byte_re_search_2 (bufp, string1, size1, string2, size2, startpos,
5067
range, regs, stop);
5068
} /* re_search_2 */
5069
#ifdef _LIBC
5070
weak_alias (__re_search_2, re_search_2)
5071
#endif
5072
5073
#endif /* not INSIDE_RECURSION */
5074
5075
#ifdef INSIDE_RECURSION
5076
5077
#ifdef MATCH_MAY_ALLOCATE
5078
# define FREE_VAR(var) if (var) REGEX_FREE (var); var = NULL
5079
#else
5080
# define FREE_VAR(var) if (var) free (var); var = NULL
5081
#endif
5082
5083
#ifdef WCHAR
5084
# define MAX_ALLOCA_SIZE 2000
5085
5086
# define FREE_WCS_BUFFERS() \
5087
do { \
5088
if (size1 > MAX_ALLOCA_SIZE) \
5089
{ \
5090
free (wcs_string1); \
5091
free (mbs_offset1); \
5092
} \
5093
else \
5094
{ \
5095
FREE_VAR (wcs_string1); \
5096
FREE_VAR (mbs_offset1); \
5097
} \
5098
if (size2 > MAX_ALLOCA_SIZE) \
5099
{ \
5100
free (wcs_string2); \
5101
free (mbs_offset2); \
5102
} \
5103
else \
5104
{ \
5105
FREE_VAR (wcs_string2); \
5106
FREE_VAR (mbs_offset2); \
5107
} \
5108
} while (0)
5109
5110
#endif
5111
5112
5113
static int
5114
PREFIX(re_search_2) (bufp, string1, size1, string2, size2, startpos, range,
5115
regs, stop)
5116
struct re_pattern_buffer *bufp;
5117
const char *string1, *string2;
5118
int size1, size2;
5119
int startpos;
5120
int range;
5121
struct re_registers *regs;
5122
int stop;
5123
{
5124
int val;
5125
register char *fastmap = bufp->fastmap;
5126
register RE_TRANSLATE_TYPE translate = bufp->translate;
5127
int total_size = size1 + size2;
5128
int endpos = startpos + range;
5129
#ifdef WCHAR
5130
/* We need wchar_t* buffers correspond to cstring1, cstring2. */
5131
wchar_t *wcs_string1 = NULL, *wcs_string2 = NULL;
5132
/* We need the size of wchar_t buffers correspond to csize1, csize2. */
5133
int wcs_size1 = 0, wcs_size2 = 0;
5134
/* offset buffer for optimizatoin. See convert_mbs_to_wc. */
5135
int *mbs_offset1 = NULL, *mbs_offset2 = NULL;
5136
/* They hold whether each wchar_t is binary data or not. */
5137
char *is_binary = NULL;
5138
#endif /* WCHAR */
5139
5140
/* Check for out-of-range STARTPOS. */
5141
if (startpos < 0 || startpos > total_size)
5142
return -1;
5143
5144
/* Fix up RANGE if it might eventually take us outside
5145
the virtual concatenation of STRING1 and STRING2.
5146
Make sure we won't move STARTPOS below 0 or above TOTAL_SIZE. */
5147
if (endpos < 0)
5148
range = 0 - startpos;
5149
else if (endpos > total_size)
5150
range = total_size - startpos;
5151
5152
/* If the search isn't to be a backwards one, don't waste time in a
5153
search for a pattern that must be anchored. */
5154
if (bufp->used > 0 && range > 0
5155
&& ((re_opcode_t) bufp->buffer[0] == begbuf
5156
/* `begline' is like `begbuf' if it cannot match at newlines. */
5157
|| ((re_opcode_t) bufp->buffer[0] == begline
5158
&& !bufp->newline_anchor)))
5159
{
5160
if (startpos > 0)
5161
return -1;
5162
else
5163
range = 1;
5164
}
5165
5166
#ifdef emacs
5167
/* In a forward search for something that starts with \=.
5168
don't keep searching past point. */
5169
if (bufp->used > 0 && (re_opcode_t) bufp->buffer[0] == at_dot && range > 0)
5170
{
5171
range = PT - startpos;
5172
if (range <= 0)
5173
return -1;
5174
}
5175
#endif /* emacs */
5176
5177
/* Update the fastmap now if not correct already. */
5178
if (fastmap && !bufp->fastmap_accurate)
5179
if (re_compile_fastmap (bufp) == -2)
5180
return -2;
5181
5182
#ifdef WCHAR
5183
/* Allocate wchar_t array for wcs_string1 and wcs_string2 and
5184
fill them with converted string. */
5185
if (size1 != 0)
5186
{
5187
if (size1 > MAX_ALLOCA_SIZE)
5188
{
5189
wcs_string1 = TALLOC (size1 + 1, CHAR_T);
5190
mbs_offset1 = TALLOC (size1 + 1, int);
5191
is_binary = TALLOC (size1 + 1, char);
5192
}
5193
else
5194
{
5195
wcs_string1 = REGEX_TALLOC (size1 + 1, CHAR_T);
5196
mbs_offset1 = REGEX_TALLOC (size1 + 1, int);
5197
is_binary = REGEX_TALLOC (size1 + 1, char);
5198
}
5199
if (!wcs_string1 || !mbs_offset1 || !is_binary)
5200
{
5201
if (size1 > MAX_ALLOCA_SIZE)
5202
{
5203
free (wcs_string1);
5204
free (mbs_offset1);
5205
free (is_binary);
5206
}
5207
else
5208
{
5209
FREE_VAR (wcs_string1);
5210
FREE_VAR (mbs_offset1);
5211
FREE_VAR (is_binary);
5212
}
5213
return -2;
5214
}
5215
wcs_size1 = convert_mbs_to_wcs(wcs_string1, string1, size1,
5216
mbs_offset1, is_binary);
5217
wcs_string1[wcs_size1] = L'\0'; /* for a sentinel */
5218
if (size1 > MAX_ALLOCA_SIZE)
5219
free (is_binary);
5220
else
5221
FREE_VAR (is_binary);
5222
}
5223
if (size2 != 0)
5224
{
5225
if (size2 > MAX_ALLOCA_SIZE)
5226
{
5227
wcs_string2 = TALLOC (size2 + 1, CHAR_T);
5228
mbs_offset2 = TALLOC (size2 + 1, int);
5229
is_binary = TALLOC (size2 + 1, char);
5230
}
5231
else
5232
{
5233
wcs_string2 = REGEX_TALLOC (size2 + 1, CHAR_T);
5234
mbs_offset2 = REGEX_TALLOC (size2 + 1, int);
5235
is_binary = REGEX_TALLOC (size2 + 1, char);
5236
}
5237
if (!wcs_string2 || !mbs_offset2 || !is_binary)
5238
{
5239
FREE_WCS_BUFFERS ();
5240
if (size2 > MAX_ALLOCA_SIZE)
5241
free (is_binary);
5242
else
5243
FREE_VAR (is_binary);
5244
return -2;
5245
}
5246
wcs_size2 = convert_mbs_to_wcs(wcs_string2, string2, size2,
5247
mbs_offset2, is_binary);
5248
wcs_string2[wcs_size2] = L'\0'; /* for a sentinel */
5249
if (size2 > MAX_ALLOCA_SIZE)
5250
free (is_binary);
5251
else
5252
FREE_VAR (is_binary);
5253
}
5254
#endif /* WCHAR */
5255
5256
5257
/* Loop through the string, looking for a place to start matching. */
5258
for (;;)
5259
{
5260
/* If a fastmap is supplied, skip quickly over characters that
5261
cannot be the start of a match. If the pattern can match the
5262
null string, however, we don't need to skip characters; we want
5263
the first null string. */
5264
if (fastmap && startpos < total_size && !bufp->can_be_null)
5265
{
5266
if (range > 0) /* Searching forwards. */
5267
{
5268
register const char *d;
5269
register int lim = 0;
5270
int irange = range;
5271
5272
if (startpos < size1 && startpos + range >= size1)
5273
lim = range - (size1 - startpos);
5274
5275
d = (startpos >= size1 ? string2 - size1 : string1) + startpos;
5276
5277
/* Written out as an if-else to avoid testing `translate'
5278
inside the loop. */
5279
if (translate)
5280
while (range > lim
5281
&& !fastmap[(unsigned char)
5282
translate[(unsigned char) *d++]])
5283
range--;
5284
else
5285
while (range > lim && !fastmap[(unsigned char) *d++])
5286
range--;
5287
5288
startpos += irange - range;
5289
}
5290
else /* Searching backwards. */
5291
{
5292
register CHAR_T c = (size1 == 0 || startpos >= size1
5293
? string2[startpos - size1]
5294
: string1[startpos]);
5295
5296
if (!fastmap[(unsigned char) TRANSLATE (c)])
5297
goto advance;
5298
}
5299
}
5300
5301
/* If can't match the null string, and that's all we have left, fail. */
5302
if (range >= 0 && startpos == total_size && fastmap
5303
&& !bufp->can_be_null)
5304
{
5305
#ifdef WCHAR
5306
FREE_WCS_BUFFERS ();
5307
#endif
5308
return -1;
5309
}
5310
5311
#ifdef WCHAR
5312
val = wcs_re_match_2_internal (bufp, string1, size1, string2,
5313
size2, startpos, regs, stop,
5314
wcs_string1, wcs_size1,
5315
wcs_string2, wcs_size2,
5316
mbs_offset1, mbs_offset2);
5317
#else /* BYTE */
5318
val = byte_re_match_2_internal (bufp, string1, size1, string2,
5319
size2, startpos, regs, stop);
5320
#endif /* BYTE */
5321
5322
#ifndef REGEX_MALLOC
5323
# ifdef C_ALLOCA
5324
alloca (0);
5325
# endif
5326
#endif
5327
5328
if (val >= 0)
5329
{
5330
#ifdef WCHAR
5331
FREE_WCS_BUFFERS ();
5332
#endif
5333
return startpos;
5334
}
5335
5336
if (val == -2)
5337
{
5338
#ifdef WCHAR
5339
FREE_WCS_BUFFERS ();
5340
#endif
5341
return -2;
5342
}
5343
5344
advance:
5345
if (!range)
5346
break;
5347
else if (range > 0)
5348
{
5349
range--;
5350
startpos++;
5351
}
5352
else
5353
{
5354
range++;
5355
startpos--;
5356
}
5357
}
5358
#ifdef WCHAR
5359
FREE_WCS_BUFFERS ();
5360
#endif
5361
return -1;
5362
}
5363
5364
#ifdef WCHAR
5365
/* This converts PTR, a pointer into one of the search wchar_t strings
5366
`string1' and `string2' into an multibyte string offset from the
5367
beginning of that string. We use mbs_offset to optimize.
5368
See convert_mbs_to_wcs. */
5369
# define POINTER_TO_OFFSET(ptr) \
5370
(FIRST_STRING_P (ptr) \
5371
? ((regoff_t)(mbs_offset1 != NULL? mbs_offset1[(ptr)-string1] : 0)) \
5372
: ((regoff_t)((mbs_offset2 != NULL? mbs_offset2[(ptr)-string2] : 0) \
5373
+ csize1)))
5374
#else /* BYTE */
5375
/* This converts PTR, a pointer into one of the search strings `string1'
5376
and `string2' into an offset from the beginning of that string. */
5377
# define POINTER_TO_OFFSET(ptr) \
5378
(FIRST_STRING_P (ptr) \
5379
? ((regoff_t) ((ptr) - string1)) \
5380
: ((regoff_t) ((ptr) - string2 + size1)))
5381
#endif /* WCHAR */
5382
5383
/* Macros for dealing with the split strings in re_match_2. */
5384
5385
#define MATCHING_IN_FIRST_STRING (dend == end_match_1)
5386
5387
/* Call before fetching a character with *d. This switches over to
5388
string2 if necessary. */
5389
#define PREFETCH() \
5390
while (d == dend) \
5391
{ \
5392
/* End of string2 => fail. */ \
5393
if (dend == end_match_2) \
5394
goto fail; \
5395
/* End of string1 => advance to string2. */ \
5396
d = string2; \
5397
dend = end_match_2; \
5398
}
5399
5400
/* Test if at very beginning or at very end of the virtual concatenation
5401
of `string1' and `string2'. If only one string, it's `string2'. */
5402
#define AT_STRINGS_BEG(d) ((d) == (size1 ? string1 : string2) || !size2)
5403
#define AT_STRINGS_END(d) ((d) == end2)
5404
5405
5406
/* Test if D points to a character which is word-constituent. We have
5407
two special cases to check for: if past the end of string1, look at
5408
the first character in string2; and if before the beginning of
5409
string2, look at the last character in string1. */
5410
#ifdef WCHAR
5411
/* Use internationalized API instead of SYNTAX. */
5412
# define WORDCHAR_P(d) \
5413
(iswalnum ((wint_t)((d) == end1 ? *string2 \
5414
: (d) == string2 - 1 ? *(end1 - 1) : *(d))) != 0 \
5415
|| ((d) == end1 ? *string2 \
5416
: (d) == string2 - 1 ? *(end1 - 1) : *(d)) == L'_')
5417
#else /* BYTE */
5418
# define WORDCHAR_P(d) \
5419
(SYNTAX ((d) == end1 ? *string2 \
5420
: (d) == string2 - 1 ? *(end1 - 1) : *(d)) \
5421
== Sword)
5422
#endif /* WCHAR */
5423
5424
/* Disabled due to a compiler bug -- see comment at case wordbound */
5425
#if 0
5426
/* Test if the character before D and the one at D differ with respect
5427
to being word-constituent. */
5428
#define AT_WORD_BOUNDARY(d) \
5429
(AT_STRINGS_BEG (d) || AT_STRINGS_END (d) \
5430
|| WORDCHAR_P (d - 1) != WORDCHAR_P (d))
5431
#endif
5432
5433
/* Free everything we malloc. */
5434
#ifdef MATCH_MAY_ALLOCATE
5435
# ifdef WCHAR
5436
# define FREE_VARIABLES() \
5437
do { \
5438
REGEX_FREE_STACK (fail_stack.stack); \
5439
FREE_VAR (regstart); \
5440
FREE_VAR (regend); \
5441
FREE_VAR (old_regstart); \
5442
FREE_VAR (old_regend); \
5443
FREE_VAR (best_regstart); \
5444
FREE_VAR (best_regend); \
5445
FREE_VAR (reg_info); \
5446
FREE_VAR (reg_dummy); \
5447
FREE_VAR (reg_info_dummy); \
5448
if (!cant_free_wcs_buf) \
5449
{ \
5450
FREE_VAR (string1); \
5451
FREE_VAR (string2); \
5452
FREE_VAR (mbs_offset1); \
5453
FREE_VAR (mbs_offset2); \
5454
} \
5455
} while (0)
5456
# else /* BYTE */
5457
# define FREE_VARIABLES() \
5458
do { \
5459
REGEX_FREE_STACK (fail_stack.stack); \
5460
FREE_VAR (regstart); \
5461
FREE_VAR (regend); \
5462
FREE_VAR (old_regstart); \
5463
FREE_VAR (old_regend); \
5464
FREE_VAR (best_regstart); \
5465
FREE_VAR (best_regend); \
5466
FREE_VAR (reg_info); \
5467
FREE_VAR (reg_dummy); \
5468
FREE_VAR (reg_info_dummy); \
5469
} while (0)
5470
# endif /* WCHAR */
5471
#else
5472
# ifdef WCHAR
5473
# define FREE_VARIABLES() \
5474
do { \
5475
if (!cant_free_wcs_buf) \
5476
{ \
5477
FREE_VAR (string1); \
5478
FREE_VAR (string2); \
5479
FREE_VAR (mbs_offset1); \
5480
FREE_VAR (mbs_offset2); \
5481
} \
5482
} while (0)
5483
# else /* BYTE */
5484
# define FREE_VARIABLES() ((void)0) /* Do nothing! But inhibit gcc warning. */
5485
# endif /* WCHAR */
5486
#endif /* not MATCH_MAY_ALLOCATE */
5487
5488
/* These values must meet several constraints. They must not be valid
5489
register values; since we have a limit of 255 registers (because
5490
we use only one byte in the pattern for the register number), we can
5491
use numbers larger than 255. They must differ by 1, because of
5492
NUM_FAILURE_ITEMS above. And the value for the lowest register must
5493
be larger than the value for the highest register, so we do not try
5494
to actually save any registers when none are active. */
5495
#define NO_HIGHEST_ACTIVE_REG (1 << BYTEWIDTH)
5496
#define NO_LOWEST_ACTIVE_REG (NO_HIGHEST_ACTIVE_REG + 1)
5497
5498
#else /* not INSIDE_RECURSION */
5499
/* Matching routines. */
5500
5501
#ifndef emacs /* Emacs never uses this. */
5502
/* re_match is like re_match_2 except it takes only a single string. */
5503
5504
int
5505
re_match (bufp, string, size, pos, regs)
5506
struct re_pattern_buffer *bufp;
5507
const char *string;
5508
int size, pos;
5509
struct re_registers *regs;
5510
{
5511
int result;
5512
# ifdef MBS_SUPPORT
5513
if (MB_CUR_MAX != 1)
5514
result = wcs_re_match_2_internal (bufp, NULL, 0, string, size,
5515
pos, regs, size,
5516
NULL, 0, NULL, 0, NULL, NULL);
5517
else
5518
# endif
5519
result = byte_re_match_2_internal (bufp, NULL, 0, string, size,
5520
pos, regs, size);
5521
# ifndef REGEX_MALLOC
5522
# ifdef C_ALLOCA
5523
alloca (0);
5524
# endif
5525
# endif
5526
return result;
5527
}
5528
# ifdef _LIBC
5529
weak_alias (__re_match, re_match)
5530
# endif
5531
#endif /* not emacs */
5532
5533
#endif /* not INSIDE_RECURSION */
5534
5535
#ifdef INSIDE_RECURSION
5536
static boolean PREFIX(group_match_null_string_p) _RE_ARGS ((UCHAR_T **p,
5537
UCHAR_T *end,
5538
PREFIX(register_info_type) *reg_info));
5539
static boolean PREFIX(alt_match_null_string_p) _RE_ARGS ((UCHAR_T *p,
5540
UCHAR_T *end,
5541
PREFIX(register_info_type) *reg_info));
5542
static boolean PREFIX(common_op_match_null_string_p) _RE_ARGS ((UCHAR_T **p,
5543
UCHAR_T *end,
5544
PREFIX(register_info_type) *reg_info));
5545
static int PREFIX(bcmp_translate) _RE_ARGS ((const CHAR_T *s1, const CHAR_T *s2,
5546
int len, char *translate));
5547
#else /* not INSIDE_RECURSION */
5548
5549
/* re_match_2 matches the compiled pattern in BUFP against the
5550
the (virtual) concatenation of STRING1 and STRING2 (of length SIZE1
5551
and SIZE2, respectively). We start matching at POS, and stop
5552
matching at STOP.
5553
5554
If REGS is non-null and the `no_sub' field of BUFP is nonzero, we
5555
store offsets for the substring each group matched in REGS. See the
5556
documentation for exactly how many groups we fill.
5557
5558
We return -1 if no match, -2 if an internal error (such as the
5559
failure stack overflowing). Otherwise, we return the length of the
5560
matched substring. */
5561
5562
int
5563
re_match_2 (bufp, string1, size1, string2, size2, pos, regs, stop)
5564
struct re_pattern_buffer *bufp;
5565
const char *string1, *string2;
5566
int size1, size2;
5567
int pos;
5568
struct re_registers *regs;
5569
int stop;
5570
{
5571
int result;
5572
# ifdef MBS_SUPPORT
5573
if (MB_CUR_MAX != 1)
5574
result = wcs_re_match_2_internal (bufp, string1, size1, string2, size2,
5575
pos, regs, stop,
5576
NULL, 0, NULL, 0, NULL, NULL);
5577
else
5578
# endif
5579
result = byte_re_match_2_internal (bufp, string1, size1, string2, size2,
5580
pos, regs, stop);
5581
5582
#ifndef REGEX_MALLOC
5583
# ifdef C_ALLOCA
5584
alloca (0);
5585
# endif
5586
#endif
5587
return result;
5588
}
5589
#ifdef _LIBC
5590
weak_alias (__re_match_2, re_match_2)
5591
#endif
5592
5593
#endif /* not INSIDE_RECURSION */
5594
5595
#ifdef INSIDE_RECURSION
5596
5597
#ifdef WCHAR
5598
static int count_mbs_length PARAMS ((int *, int));
5599
5600
/* This check the substring (from 0, to length) of the multibyte string,
5601
to which offset_buffer correspond. And count how many wchar_t_characters
5602
the substring occupy. We use offset_buffer to optimization.
5603
See convert_mbs_to_wcs. */
5604
5605
static int
5606
count_mbs_length(offset_buffer, length)
5607
int *offset_buffer;
5608
int length;
5609
{
5610
int upper, lower;
5611
5612
/* Check whether the size is valid. */
5613
if (length < 0)
5614
return -1;
5615
5616
if (offset_buffer == NULL)
5617
return 0;
5618
5619
/* If there are no multibyte character, offset_buffer[i] == i.
5620
Optmize for this case. */
5621
if (offset_buffer[length] == length)
5622
return length;
5623
5624
/* Set up upper with length. (because for all i, offset_buffer[i] >= i) */
5625
upper = length;
5626
lower = 0;
5627
5628
while (true)
5629
{
5630
int middle = (lower + upper) / 2;
5631
if (middle == lower || middle == upper)
5632
break;
5633
if (offset_buffer[middle] > length)
5634
upper = middle;
5635
else if (offset_buffer[middle] < length)
5636
lower = middle;
5637
else
5638
return middle;
5639
}
5640
5641
return -1;
5642
}
5643
#endif /* WCHAR */
5644
5645
/* This is a separate function so that we can force an alloca cleanup
5646
afterwards. */
5647
#ifdef WCHAR
5648
static int
5649
wcs_re_match_2_internal (bufp, cstring1, csize1, cstring2, csize2, pos,
5650
regs, stop, string1, size1, string2, size2,
5651
mbs_offset1, mbs_offset2)
5652
struct re_pattern_buffer *bufp;
5653
const char *cstring1, *cstring2;
5654
int csize1, csize2;
5655
int pos;
5656
struct re_registers *regs;
5657
int stop;
5658
/* string1 == string2 == NULL means string1/2, size1/2 and
5659
mbs_offset1/2 need seting up in this function. */
5660
/* We need wchar_t* buffers correspond to cstring1, cstring2. */
5661
wchar_t *string1, *string2;
5662
/* We need the size of wchar_t buffers correspond to csize1, csize2. */
5663
int size1, size2;
5664
/* offset buffer for optimizatoin. See convert_mbs_to_wc. */
5665
int *mbs_offset1, *mbs_offset2;
5666
#else /* BYTE */
5667
static int
5668
byte_re_match_2_internal (bufp, string1, size1,string2, size2, pos,
5669
regs, stop)
5670
struct re_pattern_buffer *bufp;
5671
const char *string1, *string2;
5672
int size1, size2;
5673
int pos;
5674
struct re_registers *regs;
5675
int stop;
5676
#endif /* BYTE */
5677
{
5678
/* General temporaries. */
5679
int mcnt;
5680
UCHAR_T *p1;
5681
#ifdef WCHAR
5682
/* They hold whether each wchar_t is binary data or not. */
5683
char *is_binary = NULL;
5684
/* If true, we can't free string1/2, mbs_offset1/2. */
5685
int cant_free_wcs_buf = 1;
5686
#endif /* WCHAR */
5687
5688
/* Just past the end of the corresponding string. */
5689
const CHAR_T *end1, *end2;
5690
5691
/* Pointers into string1 and string2, just past the last characters in
5692
each to consider matching. */
5693
const CHAR_T *end_match_1, *end_match_2;
5694
5695
/* Where we are in the data, and the end of the current string. */
5696
const CHAR_T *d, *dend;
5697
5698
/* Where we are in the pattern, and the end of the pattern. */
5699
#ifdef WCHAR
5700
UCHAR_T *pattern, *p;
5701
register UCHAR_T *pend;
5702
#else /* BYTE */
5703
UCHAR_T *p = bufp->buffer;
5704
register UCHAR_T *pend = p + bufp->used;
5705
#endif /* WCHAR */
5706
5707
/* Mark the opcode just after a start_memory, so we can test for an
5708
empty subpattern when we get to the stop_memory. */
5709
UCHAR_T *just_past_start_mem = 0;
5710
5711
/* We use this to map every character in the string. */
5712
RE_TRANSLATE_TYPE translate = bufp->translate;
5713
5714
/* Failure point stack. Each place that can handle a failure further
5715
down the line pushes a failure point on this stack. It consists of
5716
restart, regend, and reg_info for all registers corresponding to
5717
the subexpressions we're currently inside, plus the number of such
5718
registers, and, finally, two char *'s. The first char * is where
5719
to resume scanning the pattern; the second one is where to resume
5720
scanning the strings. If the latter is zero, the failure point is
5721
a ``dummy''; if a failure happens and the failure point is a dummy,
5722
it gets discarded and the next next one is tried. */
5723
#ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
5724
PREFIX(fail_stack_type) fail_stack;
5725
#endif
5726
#ifdef DEBUG
5727
static unsigned failure_id;
5728
unsigned nfailure_points_pushed = 0, nfailure_points_popped = 0;
5729
#endif
5730
5731
#ifdef REL_ALLOC
5732
/* This holds the pointer to the failure stack, when
5733
it is allocated relocatably. */
5734
fail_stack_elt_t *failure_stack_ptr;
5735
#endif
5736
5737
/* We fill all the registers internally, independent of what we
5738
return, for use in backreferences. The number here includes
5739
an element for register zero. */
5740
size_t num_regs = bufp->re_nsub + 1;
5741
5742
/* The currently active registers. */
5743
active_reg_t lowest_active_reg = NO_LOWEST_ACTIVE_REG;
5744
active_reg_t highest_active_reg = NO_HIGHEST_ACTIVE_REG;
5745
5746
/* Information on the contents of registers. These are pointers into
5747
the input strings; they record just what was matched (on this
5748
attempt) by a subexpression part of the pattern, that is, the
5749
regnum-th regstart pointer points to where in the pattern we began
5750
matching and the regnum-th regend points to right after where we
5751
stopped matching the regnum-th subexpression. (The zeroth register
5752
keeps track of what the whole pattern matches.) */
5753
#ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
5754
const CHAR_T **regstart, **regend;
5755
#endif
5756
5757
/* If a group that's operated upon by a repetition operator fails to
5758
match anything, then the register for its start will need to be
5759
restored because it will have been set to wherever in the string we
5760
are when we last see its open-group operator. Similarly for a
5761
register's end. */
5762
#ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
5763
const CHAR_T **old_regstart, **old_regend;
5764
#endif
5765
5766
/* The is_active field of reg_info helps us keep track of which (possibly
5767
nested) subexpressions we are currently in. The matched_something
5768
field of reg_info[reg_num] helps us tell whether or not we have
5769
matched any of the pattern so far this time through the reg_num-th
5770
subexpression. These two fields get reset each time through any
5771
loop their register is in. */
5772
#ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
5773
PREFIX(register_info_type) *reg_info;
5774
#endif
5775
5776
/* The following record the register info as found in the above
5777
variables when we find a match better than any we've seen before.
5778
This happens as we backtrack through the failure points, which in
5779
turn happens only if we have not yet matched the entire string. */
5780
unsigned best_regs_set = false;
5781
#ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
5782
const CHAR_T **best_regstart, **best_regend;
5783
#endif
5784
5785
/* Logically, this is `best_regend[0]'. But we don't want to have to
5786
allocate space for that if we're not allocating space for anything
5787
else (see below). Also, we never need info about register 0 for
5788
any of the other register vectors, and it seems rather a kludge to
5789
treat `best_regend' differently than the rest. So we keep track of
5790
the end of the best match so far in a separate variable. We
5791
initialize this to NULL so that when we backtrack the first time
5792
and need to test it, it's not garbage. */
5793
const CHAR_T *match_end = NULL;
5794
5795
/* This helps SET_REGS_MATCHED avoid doing redundant work. */
5796
int set_regs_matched_done = 0;
5797
5798
/* Used when we pop values we don't care about. */
5799
#ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
5800
const CHAR_T **reg_dummy;
5801
PREFIX(register_info_type) *reg_info_dummy;
5802
#endif
5803
5804
#ifdef DEBUG
5805
/* Counts the total number of registers pushed. */
5806
unsigned num_regs_pushed = 0;
5807
#endif
5808
5809
/* Definitions for state transitions. More efficiently for gcc. */
5810
#ifdef __GNUC__
5811
# if defined HAVE_SUBTRACT_LOCAL_LABELS && defined SHARED
5812
# define NEXT \
5813
do \
5814
{ \
5815
int offset; \
5816
const void *__unbounded ptr; \
5817
offset = (p == pend \
5818
? 0 : jmptable[SWITCH_ENUM_CAST ((re_opcode_t) *p++)]); \
5819
ptr = &&end_of_pattern + offset; \
5820
goto *ptr; \
5821
} \
5822
while (0)
5823
# define REF(x) \
5824
&&label_##x - &&end_of_pattern
5825
# define JUMP_TABLE_TYPE const int
5826
# else
5827
# define NEXT \
5828
do \
5829
{ \
5830
const void *__unbounded ptr; \
5831
ptr = (p == pend ? &&end_of_pattern \
5832
: jmptable[SWITCH_ENUM_CAST ((re_opcode_t) *p++)]); \
5833
goto *ptr; \
5834
} \
5835
while (0)
5836
# define REF(x) \
5837
&&label_##x
5838
# define JUMP_TABLE_TYPE const void *const
5839
# endif
5840
# define CASE(x) label_##x
5841
static JUMP_TABLE_TYPE jmptable[] =
5842
{
5843
REF (no_op),
5844
REF (succeed),
5845
REF (exactn),
5846
# ifdef MBS_SUPPORT
5847
REF (exactn_bin),
5848
# endif
5849
REF (anychar),
5850
REF (charset),
5851
REF (charset_not),
5852
REF (start_memory),
5853
REF (stop_memory),
5854
REF (duplicate),
5855
REF (begline),
5856
REF (endline),
5857
REF (begbuf),
5858
REF (endbuf),
5859
REF (jump),
5860
REF (jump_past_alt),
5861
REF (on_failure_jump),
5862
REF (on_failure_keep_string_jump),
5863
REF (pop_failure_jump),
5864
REF (maybe_pop_jump),
5865
REF (dummy_failure_jump),
5866
REF (push_dummy_failure),
5867
REF (succeed_n),
5868
REF (jump_n),
5869
REF (set_number_at),
5870
REF (wordchar),
5871
REF (notwordchar),
5872
REF (wordbeg),
5873
REF (wordend),
5874
REF (wordbound),
5875
REF (notwordbound)
5876
# ifdef emacs
5877
,REF (before_dot),
5878
REF (at_dot),
5879
REF (after_dot),
5880
REF (syntaxspec),
5881
REF (notsyntaxspec)
5882
# endif
5883
};
5884
#else
5885
# define NEXT \
5886
break
5887
# define CASE(x) \
5888
case x
5889
#endif
5890
5891
DEBUG_PRINT1 ("\n\nEntering re_match_2.\n");
5892
5893
INIT_FAIL_STACK ();
5894
5895
#ifdef MATCH_MAY_ALLOCATE
5896
/* Do not bother to initialize all the register variables if there are
5897
no groups in the pattern, as it takes a fair amount of time. If
5898
there are groups, we include space for register 0 (the whole
5899
pattern), even though we never use it, since it simplifies the
5900
array indexing. We should fix this. */
5901
if (bufp->re_nsub)
5902
{
5903
regstart = REGEX_TALLOC (num_regs, const CHAR_T *);
5904
regend = REGEX_TALLOC (num_regs, const CHAR_T *);
5905
old_regstart = REGEX_TALLOC (num_regs, const CHAR_T *);
5906
old_regend = REGEX_TALLOC (num_regs, const CHAR_T *);
5907
best_regstart = REGEX_TALLOC (num_regs, const CHAR_T *);
5908
best_regend = REGEX_TALLOC (num_regs, const CHAR_T *);
5909
reg_info = REGEX_TALLOC (num_regs, PREFIX(register_info_type));
5910
reg_dummy = REGEX_TALLOC (num_regs, const CHAR_T *);
5911
reg_info_dummy = REGEX_TALLOC (num_regs, PREFIX(register_info_type));
5912
5913
if (!(regstart && regend && old_regstart && old_regend && reg_info
5914
&& best_regstart && best_regend && reg_dummy && reg_info_dummy))
5915
{
5916
FREE_VARIABLES ();
5917
return -2;
5918
}
5919
}
5920
else
5921
{
5922
/* We must initialize all our variables to NULL, so that
5923
`FREE_VARIABLES' doesn't try to free them. */
5924
regstart = regend = old_regstart = old_regend = best_regstart
5925
= best_regend = reg_dummy = NULL;
5926
reg_info = reg_info_dummy = (PREFIX(register_info_type) *) NULL;
5927
}
5928
#endif /* MATCH_MAY_ALLOCATE */
5929
5930
/* The starting position is bogus. */
5931
#ifdef WCHAR
5932
if (pos < 0 || pos > csize1 + csize2)
5933
#else /* BYTE */
5934
if (pos < 0 || pos > size1 + size2)
5935
#endif
5936
{
5937
FREE_VARIABLES ();
5938
return -1;
5939
}
5940
5941
#ifdef WCHAR
5942
/* Allocate wchar_t array for string1 and string2 and
5943
fill them with converted string. */
5944
if (string1 == NULL && string2 == NULL)
5945
{
5946
/* We need seting up buffers here. */
5947
5948
/* We must free wcs buffers in this function. */
5949
cant_free_wcs_buf = 0;
5950
5951
if (csize1 != 0)
5952
{
5953
string1 = REGEX_TALLOC (csize1 + 1, CHAR_T);
5954
mbs_offset1 = REGEX_TALLOC (csize1 + 1, int);
5955
is_binary = REGEX_TALLOC (csize1 + 1, char);
5956
if (!string1 || !mbs_offset1 || !is_binary)
5957
{
5958
FREE_VAR (string1);
5959
FREE_VAR (mbs_offset1);
5960
FREE_VAR (is_binary);
5961
return -2;
5962
}
5963
}
5964
if (csize2 != 0)
5965
{
5966
string2 = REGEX_TALLOC (csize2 + 1, CHAR_T);
5967
mbs_offset2 = REGEX_TALLOC (csize2 + 1, int);
5968
is_binary = REGEX_TALLOC (csize2 + 1, char);
5969
if (!string2 || !mbs_offset2 || !is_binary)
5970
{
5971
FREE_VAR (string1);
5972
FREE_VAR (mbs_offset1);
5973
FREE_VAR (string2);
5974
FREE_VAR (mbs_offset2);
5975
FREE_VAR (is_binary);
5976
return -2;
5977
}
5978
size2 = convert_mbs_to_wcs(string2, cstring2, csize2,
5979
mbs_offset2, is_binary);
5980
string2[size2] = L'\0'; /* for a sentinel */
5981
FREE_VAR (is_binary);
5982
}
5983
}
5984
5985
/* We need to cast pattern to (wchar_t*), because we casted this compiled
5986
pattern to (char*) in regex_compile. */
5987
p = pattern = (CHAR_T*)bufp->buffer;
5988
pend = (CHAR_T*)(bufp->buffer + bufp->used);
5989
5990
#endif /* WCHAR */
5991
5992
/* Initialize subexpression text positions to -1 to mark ones that no
5993
start_memory/stop_memory has been seen for. Also initialize the
5994
register information struct. */
5995
for (mcnt = 1; (unsigned) mcnt < num_regs; mcnt++)
5996
{
5997
regstart[mcnt] = regend[mcnt]
5998
= old_regstart[mcnt] = old_regend[mcnt] = REG_UNSET_VALUE;
5999
6000
REG_MATCH_NULL_STRING_P (reg_info[mcnt]) = MATCH_NULL_UNSET_VALUE;
6001
IS_ACTIVE (reg_info[mcnt]) = 0;
6002
MATCHED_SOMETHING (reg_info[mcnt]) = 0;
6003
EVER_MATCHED_SOMETHING (reg_info[mcnt]) = 0;
6004
}
6005
6006
/* We move `string1' into `string2' if the latter's empty -- but not if
6007
`string1' is null. */
6008
if (size2 == 0 && string1 != NULL)
6009
{
6010
string2 = string1;
6011
size2 = size1;
6012
string1 = 0;
6013
size1 = 0;
6014
#ifdef WCHAR
6015
mbs_offset2 = mbs_offset1;
6016
csize2 = csize1;
6017
mbs_offset1 = NULL;
6018
csize1 = 0;
6019
#endif
6020
}
6021
end1 = string1 + size1;
6022
end2 = string2 + size2;
6023
6024
/* Compute where to stop matching, within the two strings. */
6025
#ifdef WCHAR
6026
if (stop <= csize1)
6027
{
6028
mcnt = count_mbs_length(mbs_offset1, stop);
6029
end_match_1 = string1 + mcnt;
6030
end_match_2 = string2;
6031
}
6032
else
6033
{
6034
if (stop > csize1 + csize2)
6035
stop = csize1 + csize2;
6036
end_match_1 = end1;
6037
mcnt = count_mbs_length(mbs_offset2, stop-csize1);
6038
end_match_2 = string2 + mcnt;
6039
}
6040
if (mcnt < 0)
6041
{ /* count_mbs_length return error. */
6042
FREE_VARIABLES ();
6043
return -1;
6044
}
6045
#else
6046
if (stop <= size1)
6047
{
6048
end_match_1 = string1 + stop;
6049
end_match_2 = string2;
6050
}
6051
else
6052
{
6053
end_match_1 = end1;
6054
end_match_2 = string2 + stop - size1;
6055
}
6056
#endif /* WCHAR */
6057
6058
/* `p' scans through the pattern as `d' scans through the data.
6059
`dend' is the end of the input string that `d' points within. `d'
6060
is advanced into the following input string whenever necessary, but
6061
this happens before fetching; therefore, at the beginning of the
6062
loop, `d' can be pointing at the end of a string, but it cannot
6063
equal `string2'. */
6064
#ifdef WCHAR
6065
if (size1 > 0 && pos <= csize1)
6066
{
6067
mcnt = count_mbs_length(mbs_offset1, pos);
6068
d = string1 + mcnt;
6069
dend = end_match_1;
6070
}
6071
else
6072
{
6073
mcnt = count_mbs_length(mbs_offset2, pos-csize1);
6074
d = string2 + mcnt;
6075
dend = end_match_2;
6076
}
6077
6078
if (mcnt < 0)
6079
{ /* count_mbs_length return error. */
6080
FREE_VARIABLES ();
6081
return -1;
6082
}
6083
#else
6084
if (size1 > 0 && pos <= size1)
6085
{
6086
d = string1 + pos;
6087
dend = end_match_1;
6088
}
6089
else
6090
{
6091
d = string2 + pos - size1;
6092
dend = end_match_2;
6093
}
6094
#endif /* WCHAR */
6095
6096
DEBUG_PRINT1 ("The compiled pattern is:\n");
6097
DEBUG_PRINT_COMPILED_PATTERN (bufp, p, pend);
6098
DEBUG_PRINT1 ("The string to match is: `");
6099
DEBUG_PRINT_DOUBLE_STRING (d, string1, size1, string2, size2);
6100
DEBUG_PRINT1 ("'\n");
6101
6102
/* This loops over pattern commands. It exits by returning from the
6103
function if the match is complete, or it drops through if the match
6104
fails at this starting point in the input data. */
6105
for (;;)
6106
{
6107
#ifdef _LIBC
6108
DEBUG_PRINT2 ("\n%p: ", p);
6109
#else
6110
DEBUG_PRINT2 ("\n0x%x: ", p);
6111
#endif
6112
6113
#ifdef __GNUC__
6114
NEXT;
6115
#else
6116
if (p == pend)
6117
#endif
6118
{
6119
#ifdef __GNUC__
6120
end_of_pattern:
6121
#endif
6122
/* End of pattern means we might have succeeded. */
6123
DEBUG_PRINT1 ("end of pattern ... ");
6124
6125
/* If we haven't matched the entire string, and we want the
6126
longest match, try backtracking. */
6127
if (d != end_match_2)
6128
{
6129
/* 1 if this match ends in the same string (string1 or string2)
6130
as the best previous match. */
6131
boolean same_str_p = (FIRST_STRING_P (match_end)
6132
== MATCHING_IN_FIRST_STRING);
6133
/* 1 if this match is the best seen so far. */
6134
boolean best_match_p;
6135
6136
/* AIX compiler got confused when this was combined
6137
with the previous declaration. */
6138
if (same_str_p)
6139
best_match_p = d > match_end;
6140
else
6141
best_match_p = !MATCHING_IN_FIRST_STRING;
6142
6143
DEBUG_PRINT1 ("backtracking.\n");
6144
6145
if (!FAIL_STACK_EMPTY ())
6146
{ /* More failure points to try. */
6147
6148
/* If exceeds best match so far, save it. */
6149
if (!best_regs_set || best_match_p)
6150
{
6151
best_regs_set = true;
6152
match_end = d;
6153
6154
DEBUG_PRINT1 ("\nSAVING match as best so far.\n");
6155
6156
for (mcnt = 1; (unsigned) mcnt < num_regs; mcnt++)
6157
{
6158
best_regstart[mcnt] = regstart[mcnt];
6159
best_regend[mcnt] = regend[mcnt];
6160
}
6161
}
6162
goto fail;
6163
}
6164
6165
/* If no failure points, don't restore garbage. And if
6166
last match is real best match, don't restore second
6167
best one. */
6168
else if (best_regs_set && !best_match_p)
6169
{
6170
restore_best_regs:
6171
/* Restore best match. It may happen that `dend ==
6172
end_match_1' while the restored d is in string2.
6173
For example, the pattern `x.*y.*z' against the
6174
strings `x-' and `y-z-', if the two strings are
6175
not consecutive in memory. */
6176
DEBUG_PRINT1 ("Restoring best registers.\n");
6177
6178
d = match_end;
6179
dend = ((d >= string1 && d <= end1)
6180
? end_match_1 : end_match_2);
6181
6182
for (mcnt = 1; (unsigned) mcnt < num_regs; mcnt++)
6183
{
6184
regstart[mcnt] = best_regstart[mcnt];
6185
regend[mcnt] = best_regend[mcnt];
6186
}
6187
}
6188
} /* d != end_match_2 */
6189
6190
succeed_label:
6191
DEBUG_PRINT1 ("Accepting match.\n");
6192
/* If caller wants register contents data back, do it. */
6193
if (regs && !bufp->no_sub)
6194
{
6195
/* Have the register data arrays been allocated? */
6196
if (bufp->regs_allocated == REGS_UNALLOCATED)
6197
{ /* No. So allocate them with malloc. We need one
6198
extra element beyond `num_regs' for the `-1' marker
6199
GNU code uses. */
6200
regs->num_regs = MAX (RE_NREGS, num_regs + 1);
6201
regs->start = TALLOC (regs->num_regs, regoff_t);
6202
regs->end = TALLOC (regs->num_regs, regoff_t);
6203
if (regs->start == NULL || regs->end == NULL)
6204
{
6205
FREE_VARIABLES ();
6206
return -2;
6207
}
6208
bufp->regs_allocated = REGS_REALLOCATE;
6209
}
6210
else if (bufp->regs_allocated == REGS_REALLOCATE)
6211
{ /* Yes. If we need more elements than were already
6212
allocated, reallocate them. If we need fewer, just
6213
leave it alone. */
6214
if (regs->num_regs < num_regs + 1)
6215
{
6216
regs->num_regs = num_regs + 1;
6217
RETALLOC (regs->start, regs->num_regs, regoff_t);
6218
RETALLOC (regs->end, regs->num_regs, regoff_t);
6219
if (regs->start == NULL || regs->end == NULL)
6220
{
6221
FREE_VARIABLES ();
6222
return -2;
6223
}
6224
}
6225
}
6226
else
6227
{
6228
/* These braces fend off a "empty body in an else-statement"
6229
warning under GCC when assert expands to nothing. */
6230
assert (bufp->regs_allocated == REGS_FIXED);
6231
}
6232
6233
/* Convert the pointer data in `regstart' and `regend' to
6234
indices. Register zero has to be set differently,
6235
since we haven't kept track of any info for it. */
6236
if (regs->num_regs > 0)
6237
{
6238
regs->start[0] = pos;
6239
#ifdef WCHAR
6240
if (MATCHING_IN_FIRST_STRING)
6241
regs->end[0] = (mbs_offset1 != NULL ?
6242
mbs_offset1[d-string1] : 0);
6243
else
6244
regs->end[0] = csize1 + (mbs_offset2 != NULL
6245
? mbs_offset2[d-string2] : 0);
6246
#else
6247
regs->end[0] = (MATCHING_IN_FIRST_STRING
6248
? ((regoff_t) (d - string1))
6249
: ((regoff_t) (d - string2 + size1)));
6250
#endif /* WCHAR */
6251
}
6252
6253
/* Go through the first `min (num_regs, regs->num_regs)'
6254
registers, since that is all we initialized. */
6255
for (mcnt = 1; (unsigned) mcnt < MIN (num_regs, regs->num_regs);
6256
mcnt++)
6257
{
6258
if (REG_UNSET (regstart[mcnt]) || REG_UNSET (regend[mcnt]))
6259
regs->start[mcnt] = regs->end[mcnt] = -1;
6260
else
6261
{
6262
regs->start[mcnt]
6263
= (regoff_t) POINTER_TO_OFFSET (regstart[mcnt]);
6264
regs->end[mcnt]
6265
= (regoff_t) POINTER_TO_OFFSET (regend[mcnt]);
6266
}
6267
}
6268
6269
/* If the regs structure we return has more elements than
6270
were in the pattern, set the extra elements to -1. If
6271
we (re)allocated the registers, this is the case,
6272
because we always allocate enough to have at least one
6273
-1 at the end. */
6274
for (mcnt = num_regs; (unsigned) mcnt < regs->num_regs; mcnt++)
6275
regs->start[mcnt] = regs->end[mcnt] = -1;
6276
} /* regs && !bufp->no_sub */
6277
6278
DEBUG_PRINT4 ("%u failure points pushed, %u popped (%u remain).\n",
6279
nfailure_points_pushed, nfailure_points_popped,
6280
nfailure_points_pushed - nfailure_points_popped);
6281
DEBUG_PRINT2 ("%u registers pushed.\n", num_regs_pushed);
6282
6283
#ifdef WCHAR
6284
if (MATCHING_IN_FIRST_STRING)
6285
mcnt = mbs_offset1 != NULL ? mbs_offset1[d-string1] : 0;
6286
else
6287
mcnt = (mbs_offset2 != NULL ? mbs_offset2[d-string2] : 0) +
6288
csize1;
6289
mcnt -= pos;
6290
#else
6291
mcnt = d - pos - (MATCHING_IN_FIRST_STRING
6292
? string1 : string2 - size1);
6293
#endif /* WCHAR */
6294
6295
DEBUG_PRINT2 ("Returning %d from re_match_2.\n", mcnt);
6296
6297
FREE_VARIABLES ();
6298
return mcnt;
6299
}
6300
6301
#ifndef __GNUC__
6302
/* Otherwise match next pattern command. */
6303
switch (SWITCH_ENUM_CAST ((re_opcode_t) *p++))
6304
{
6305
#endif
6306
/* Ignore these. Used to ignore the n of succeed_n's which
6307
currently have n == 0. */
6308
CASE (no_op):
6309
DEBUG_PRINT1 ("EXECUTING no_op.\n");
6310
NEXT;
6311
6312
CASE (succeed):
6313
DEBUG_PRINT1 ("EXECUTING succeed.\n");
6314
goto succeed_label;
6315
6316
/* Match the next n pattern characters exactly. The following
6317
byte in the pattern defines n, and the n bytes after that
6318
are the characters to match. */
6319
CASE (exactn):
6320
#ifdef MBS_SUPPORT
6321
CASE (exactn_bin):
6322
#endif
6323
mcnt = *p++;
6324
DEBUG_PRINT2 ("EXECUTING exactn %d.\n", mcnt);
6325
6326
/* This is written out as an if-else so we don't waste time
6327
testing `translate' inside the loop. */
6328
if (translate)
6329
{
6330
do
6331
{
6332
PREFETCH ();
6333
#ifdef WCHAR
6334
if (*d <= 0xff)
6335
{
6336
if ((UCHAR_T) translate[(unsigned char) *d++]
6337
!= (UCHAR_T) *p++)
6338
goto fail;
6339
}
6340
else
6341
{
6342
if (*d++ != (CHAR_T) *p++)
6343
goto fail;
6344
}
6345
#else
6346
if ((UCHAR_T) translate[(unsigned char) *d++]
6347
!= (UCHAR_T) *p++)
6348
goto fail;
6349
#endif /* WCHAR */
6350
}
6351
while (--mcnt);
6352
}
6353
else
6354
{
6355
do
6356
{
6357
PREFETCH ();
6358
if (*d++ != (CHAR_T) *p++) goto fail;
6359
}
6360
while (--mcnt);
6361
}
6362
SET_REGS_MATCHED ();
6363
NEXT;
6364
6365
6366
/* Match any character except possibly a newline or a null. */
6367
CASE (anychar):
6368
DEBUG_PRINT1 ("EXECUTING anychar.\n");
6369
6370
PREFETCH ();
6371
6372
if ((!(bufp->syntax & RE_DOT_NEWLINE) && TRANSLATE (*d) == '\n')
6373
|| (bufp->syntax & RE_DOT_NOT_NULL && TRANSLATE (*d) == '\000'))
6374
goto fail;
6375
6376
SET_REGS_MATCHED ();
6377
DEBUG_PRINT2 (" Matched `%ld'.\n", (long int) *d);
6378
d++;
6379
NEXT;
6380
6381
6382
CASE (charset):
6383
CASE (charset_not):
6384
{
6385
register UCHAR_T c;
6386
#ifdef WCHAR
6387
unsigned int i, char_class_length, coll_symbol_length,
6388
equiv_class_length, ranges_length, chars_length, length;
6389
CHAR_T *workp, *workp2, *charset_top;
6390
#define WORK_BUFFER_SIZE 128
6391
CHAR_T str_buf[WORK_BUFFER_SIZE];
6392
# ifdef _LIBC
6393
uint32_t nrules;
6394
# endif /* _LIBC */
6395
#endif /* WCHAR */
6396
boolean not = (re_opcode_t) *(p - 1) == charset_not;
6397
6398
DEBUG_PRINT2 ("EXECUTING charset%s.\n", not ? "_not" : "");
6399
PREFETCH ();
6400
c = TRANSLATE (*d); /* The character to match. */
6401
#ifdef WCHAR
6402
# ifdef _LIBC
6403
nrules = _NL_CURRENT_WORD (LC_COLLATE, _NL_COLLATE_NRULES);
6404
# endif /* _LIBC */
6405
charset_top = p - 1;
6406
char_class_length = *p++;
6407
coll_symbol_length = *p++;
6408
equiv_class_length = *p++;
6409
ranges_length = *p++;
6410
chars_length = *p++;
6411
/* p points charset[6], so the address of the next instruction
6412
(charset[l+m+n+2o+k+p']) equals p[l+m+n+2*o+p'],
6413
where l=length of char_classes, m=length of collating_symbol,
6414
n=equivalence_class, o=length of char_range,
6415
p'=length of character. */
6416
workp = p;
6417
/* Update p to indicate the next instruction. */
6418
p += char_class_length + coll_symbol_length+ equiv_class_length +
6419
2*ranges_length + chars_length;
6420
6421
/* match with char_class? */
6422
for (i = 0; i < char_class_length ; i += CHAR_CLASS_SIZE)
6423
{
6424
wctype_t wctype;
6425
uintptr_t alignedp = ((uintptr_t)workp
6426
+ __alignof__(wctype_t) - 1)
6427
& ~(uintptr_t)(__alignof__(wctype_t) - 1);
6428
wctype = *((wctype_t*)alignedp);
6429
workp += CHAR_CLASS_SIZE;
6430
if (iswctype((wint_t)c, wctype))
6431
goto char_set_matched;
6432
}
6433
6434
/* match with collating_symbol? */
6435
# ifdef _LIBC
6436
if (nrules != 0)
6437
{
6438
const unsigned char *extra = (const unsigned char *)
6439
_NL_CURRENT (LC_COLLATE, _NL_COLLATE_SYMB_EXTRAMB);
6440
6441
for (workp2 = workp + coll_symbol_length ; workp < workp2 ;
6442
workp++)
6443
{
6444
int32_t *wextra;
6445
wextra = (int32_t*)(extra + *workp++);
6446
for (i = 0; i < *wextra; ++i)
6447
if (TRANSLATE(d[i]) != wextra[1 + i])
6448
break;
6449
6450
if (i == *wextra)
6451
{
6452
/* Update d, however d will be incremented at
6453
char_set_matched:, we decrement d here. */
6454
d += i - 1;
6455
goto char_set_matched;
6456
}
6457
}
6458
}
6459
else /* (nrules == 0) */
6460
# endif
6461
/* If we can't look up collation data, we use wcscoll
6462
instead. */
6463
{
6464
for (workp2 = workp + coll_symbol_length ; workp < workp2 ;)
6465
{
6466
const CHAR_T *backup_d = d, *backup_dend = dend;
6467
length = wcslen (workp);
6468
6469
/* If wcscoll(the collating symbol, whole string) > 0,
6470
any substring of the string never match with the
6471
collating symbol. */
6472
if (wcscoll (workp, d) > 0)
6473
{
6474
workp += length + 1;
6475
continue;
6476
}
6477
6478
/* First, we compare the collating symbol with
6479
the first character of the string.
6480
If it don't match, we add the next character to
6481
the compare buffer in turn. */
6482
for (i = 0 ; i < WORK_BUFFER_SIZE-1 ; i++, d++)
6483
{
6484
int match;
6485
if (d == dend)
6486
{
6487
if (dend == end_match_2)
6488
break;
6489
d = string2;
6490
dend = end_match_2;
6491
}
6492
6493
/* add next character to the compare buffer. */
6494
str_buf[i] = TRANSLATE(*d);
6495
str_buf[i+1] = '\0';
6496
6497
match = wcscoll (workp, str_buf);
6498
if (match == 0)
6499
goto char_set_matched;
6500
6501
if (match < 0)
6502
/* (str_buf > workp) indicate (str_buf + X > workp),
6503
because for all X (str_buf + X > str_buf).
6504
So we don't need continue this loop. */
6505
break;
6506
6507
/* Otherwise(str_buf < workp),
6508
(str_buf+next_character) may equals (workp).
6509
So we continue this loop. */
6510
}
6511
/* not matched */
6512
d = backup_d;
6513
dend = backup_dend;
6514
workp += length + 1;
6515
}
6516
}
6517
/* match with equivalence_class? */
6518
# ifdef _LIBC
6519
if (nrules != 0)
6520
{
6521
const CHAR_T *backup_d = d, *backup_dend = dend;
6522
/* Try to match the equivalence class against
6523
those known to the collate implementation. */
6524
const int32_t *table;
6525
const int32_t *weights;
6526
const int32_t *extra;
6527
const int32_t *indirect;
6528
int32_t idx, idx2;
6529
wint_t *cp;
6530
size_t len;
6531
6532
/* This #include defines a local function! */
6533
# include <locale/weightwc.h>
6534
6535
table = (const int32_t *)
6536
_NL_CURRENT (LC_COLLATE, _NL_COLLATE_TABLEWC);
6537
weights = (const wint_t *)
6538
_NL_CURRENT (LC_COLLATE, _NL_COLLATE_WEIGHTWC);
6539
extra = (const wint_t *)
6540
_NL_CURRENT (LC_COLLATE, _NL_COLLATE_EXTRAWC);
6541
indirect = (const int32_t *)
6542
_NL_CURRENT (LC_COLLATE, _NL_COLLATE_INDIRECTWC);
6543
6544
/* Write 1 collating element to str_buf, and
6545
get its index. */
6546
idx2 = 0;
6547
6548
for (i = 0 ; idx2 == 0 && i < WORK_BUFFER_SIZE - 1; i++)
6549
{
6550
cp = (wint_t*)str_buf;
6551
if (d == dend)
6552
{
6553
if (dend == end_match_2)
6554
break;
6555
d = string2;
6556
dend = end_match_2;
6557
}
6558
str_buf[i] = TRANSLATE(*(d+i));
6559
str_buf[i+1] = '\0'; /* sentinel */
6560
idx2 = findidx ((const wint_t**)&cp);
6561
}
6562
6563
/* Update d, however d will be incremented at
6564
char_set_matched:, we decrement d here. */
6565
d = backup_d + ((wchar_t*)cp - (wchar_t*)str_buf - 1);
6566
if (d >= dend)
6567
{
6568
if (dend == end_match_2)
6569
d = dend;
6570
else
6571
{
6572
d = string2;
6573
dend = end_match_2;
6574
}
6575
}
6576
6577
len = weights[idx2];
6578
6579
for (workp2 = workp + equiv_class_length ; workp < workp2 ;
6580
workp++)
6581
{
6582
idx = (int32_t)*workp;
6583
/* We already checked idx != 0 in regex_compile. */
6584
6585
if (idx2 != 0 && len == weights[idx])
6586
{
6587
int cnt = 0;
6588
while (cnt < len && (weights[idx + 1 + cnt]
6589
== weights[idx2 + 1 + cnt]))
6590
++cnt;
6591
6592
if (cnt == len)
6593
goto char_set_matched;
6594
}
6595
}
6596
/* not matched */
6597
d = backup_d;
6598
dend = backup_dend;
6599
}
6600
else /* (nrules == 0) */
6601
# endif
6602
/* If we can't look up collation data, we use wcscoll
6603
instead. */
6604
{
6605
for (workp2 = workp + equiv_class_length ; workp < workp2 ;)
6606
{
6607
const CHAR_T *backup_d = d, *backup_dend = dend;
6608
length = wcslen (workp);
6609
6610
/* If wcscoll(the collating symbol, whole string) > 0,
6611
any substring of the string never match with the
6612
collating symbol. */
6613
if (wcscoll (workp, d) > 0)
6614
{
6615
workp += length + 1;
6616
break;
6617
}
6618
6619
/* First, we compare the equivalence class with
6620
the first character of the string.
6621
If it don't match, we add the next character to
6622
the compare buffer in turn. */
6623
for (i = 0 ; i < WORK_BUFFER_SIZE - 1 ; i++, d++)
6624
{
6625
int match;
6626
if (d == dend)
6627
{
6628
if (dend == end_match_2)
6629
break;
6630
d = string2;
6631
dend = end_match_2;
6632
}
6633
6634
/* add next character to the compare buffer. */
6635
str_buf[i] = TRANSLATE(*d);
6636
str_buf[i+1] = '\0';
6637
6638
match = wcscoll (workp, str_buf);
6639
6640
if (match == 0)
6641
goto char_set_matched;
6642
6643
if (match < 0)
6644
/* (str_buf > workp) indicate (str_buf + X > workp),
6645
because for all X (str_buf + X > str_buf).
6646
So we don't need continue this loop. */
6647
break;
6648
6649
/* Otherwise(str_buf < workp),
6650
(str_buf+next_character) may equals (workp).
6651
So we continue this loop. */
6652
}
6653
/* not matched */
6654
d = backup_d;
6655
dend = backup_dend;
6656
workp += length + 1;
6657
}
6658
}
6659
6660
/* match with char_range? */
6661
# ifdef _LIBC
6662
if (nrules != 0)
6663
{
6664
uint32_t collseqval;
6665
const char *collseq = (const char *)
6666
_NL_CURRENT(LC_COLLATE, _NL_COLLATE_COLLSEQWC);
6667
6668
collseqval = collseq_table_lookup (collseq, c);
6669
6670
for (; workp < p - chars_length ;)
6671
{
6672
uint32_t start_val, end_val;
6673
6674
/* We already compute the collation sequence value
6675
of the characters (or collating symbols). */
6676
start_val = (uint32_t) *workp++; /* range_start */
6677
end_val = (uint32_t) *workp++; /* range_end */
6678
6679
if (start_val <= collseqval && collseqval <= end_val)
6680
goto char_set_matched;
6681
}
6682
}
6683
else
6684
# endif
6685
{
6686
/* We set range_start_char at str_buf[0], range_end_char
6687
at str_buf[4], and compared char at str_buf[2]. */
6688
str_buf[1] = 0;
6689
str_buf[2] = c;
6690
str_buf[3] = 0;
6691
str_buf[5] = 0;
6692
for (; workp < p - chars_length ;)
6693
{
6694
wchar_t *range_start_char, *range_end_char;
6695
6696
/* match if (range_start_char <= c <= range_end_char). */
6697
6698
/* If range_start(or end) < 0, we assume -range_start(end)
6699
is the offset of the collating symbol which is specified
6700
as the character of the range start(end). */
6701
6702
/* range_start */
6703
if (*workp < 0)
6704
range_start_char = charset_top - (*workp++);
6705
else
6706
{
6707
str_buf[0] = *workp++;
6708
range_start_char = str_buf;
6709
}
6710
6711
/* range_end */
6712
if (*workp < 0)
6713
range_end_char = charset_top - (*workp++);
6714
else
6715
{
6716
str_buf[4] = *workp++;
6717
range_end_char = str_buf + 4;
6718
}
6719
6720
if (wcscoll (range_start_char, str_buf+2) <= 0
6721
&& wcscoll (str_buf+2, range_end_char) <= 0)
6722
goto char_set_matched;
6723
}
6724
}
6725
6726
/* match with char? */
6727
for (; workp < p ; workp++)
6728
if (c == *workp)
6729
goto char_set_matched;
6730
6731
not = !not;
6732
6733
char_set_matched:
6734
if (not) goto fail;
6735
#else
6736
/* Cast to `unsigned' instead of `unsigned char' in case the
6737
bit list is a full 32 bytes long. */
6738
if (c < (unsigned) (*p * BYTEWIDTH)
6739
&& p[1 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH)))
6740
not = !not;
6741
6742
p += 1 + *p;
6743
6744
if (!not) goto fail;
6745
#undef WORK_BUFFER_SIZE
6746
#endif /* WCHAR */
6747
SET_REGS_MATCHED ();
6748
d++;
6749
NEXT;
6750
}
6751
6752
6753
/* The beginning of a group is represented by start_memory.
6754
The arguments are the register number in the next byte, and the
6755
number of groups inner to this one in the next. The text
6756
matched within the group is recorded (in the internal
6757
registers data structure) under the register number. */
6758
CASE (start_memory):
6759
DEBUG_PRINT3 ("EXECUTING start_memory %ld (%ld):\n",
6760
(long int) *p, (long int) p[1]);
6761
6762
/* Find out if this group can match the empty string. */
6763
p1 = p; /* To send to group_match_null_string_p. */
6764
6765
if (REG_MATCH_NULL_STRING_P (reg_info[*p]) == MATCH_NULL_UNSET_VALUE)
6766
REG_MATCH_NULL_STRING_P (reg_info[*p])
6767
= PREFIX(group_match_null_string_p) (&p1, pend, reg_info);
6768
6769
/* Save the position in the string where we were the last time
6770
we were at this open-group operator in case the group is
6771
operated upon by a repetition operator, e.g., with `(a*)*b'
6772
against `ab'; then we want to ignore where we are now in
6773
the string in case this attempt to match fails. */
6774
old_regstart[*p] = REG_MATCH_NULL_STRING_P (reg_info[*p])
6775
? REG_UNSET (regstart[*p]) ? d : regstart[*p]
6776
: regstart[*p];
6777
DEBUG_PRINT2 (" old_regstart: %d\n",
6778
POINTER_TO_OFFSET (old_regstart[*p]));
6779
6780
regstart[*p] = d;
6781
DEBUG_PRINT2 (" regstart: %d\n", POINTER_TO_OFFSET (regstart[*p]));
6782
6783
IS_ACTIVE (reg_info[*p]) = 1;
6784
MATCHED_SOMETHING (reg_info[*p]) = 0;
6785
6786
/* Clear this whenever we change the register activity status. */
6787
set_regs_matched_done = 0;
6788
6789
/* This is the new highest active register. */
6790
highest_active_reg = *p;
6791
6792
/* If nothing was active before, this is the new lowest active
6793
register. */
6794
if (lowest_active_reg == NO_LOWEST_ACTIVE_REG)
6795
lowest_active_reg = *p;
6796
6797
/* Move past the register number and inner group count. */
6798
p += 2;
6799
just_past_start_mem = p;
6800
6801
NEXT;
6802
6803
6804
/* The stop_memory opcode represents the end of a group. Its
6805
arguments are the same as start_memory's: the register
6806
number, and the number of inner groups. */
6807
CASE (stop_memory):
6808
DEBUG_PRINT3 ("EXECUTING stop_memory %ld (%ld):\n",
6809
(long int) *p, (long int) p[1]);
6810
6811
/* We need to save the string position the last time we were at
6812
this close-group operator in case the group is operated
6813
upon by a repetition operator, e.g., with `((a*)*(b*)*)*'
6814
against `aba'; then we want to ignore where we are now in
6815
the string in case this attempt to match fails. */
6816
old_regend[*p] = REG_MATCH_NULL_STRING_P (reg_info[*p])
6817
? REG_UNSET (regend[*p]) ? d : regend[*p]
6818
: regend[*p];
6819
DEBUG_PRINT2 (" old_regend: %d\n",
6820
POINTER_TO_OFFSET (old_regend[*p]));
6821
6822
regend[*p] = d;
6823
DEBUG_PRINT2 (" regend: %d\n", POINTER_TO_OFFSET (regend[*p]));
6824
6825
/* This register isn't active anymore. */
6826
IS_ACTIVE (reg_info[*p]) = 0;
6827
6828
/* Clear this whenever we change the register activity status. */
6829
set_regs_matched_done = 0;
6830
6831
/* If this was the only register active, nothing is active
6832
anymore. */
6833
if (lowest_active_reg == highest_active_reg)
6834
{
6835
lowest_active_reg = NO_LOWEST_ACTIVE_REG;
6836
highest_active_reg = NO_HIGHEST_ACTIVE_REG;
6837
}
6838
else
6839
{ /* We must scan for the new highest active register, since
6840
it isn't necessarily one less than now: consider
6841
(a(b)c(d(e)f)g). When group 3 ends, after the f), the
6842
new highest active register is 1. */
6843
UCHAR_T r = *p - 1;
6844
while (r > 0 && !IS_ACTIVE (reg_info[r]))
6845
r--;
6846
6847
/* If we end up at register zero, that means that we saved
6848
the registers as the result of an `on_failure_jump', not
6849
a `start_memory', and we jumped to past the innermost
6850
`stop_memory'. For example, in ((.)*) we save
6851
registers 1 and 2 as a result of the *, but when we pop
6852
back to the second ), we are at the stop_memory 1.
6853
Thus, nothing is active. */
6854
if (r == 0)
6855
{
6856
lowest_active_reg = NO_LOWEST_ACTIVE_REG;
6857
highest_active_reg = NO_HIGHEST_ACTIVE_REG;
6858
}
6859
else
6860
highest_active_reg = r;
6861
}
6862
6863
/* If just failed to match something this time around with a
6864
group that's operated on by a repetition operator, try to
6865
force exit from the ``loop'', and restore the register
6866
information for this group that we had before trying this
6867
last match. */
6868
if ((!MATCHED_SOMETHING (reg_info[*p])
6869
|| just_past_start_mem == p - 1)
6870
&& (p + 2) < pend)
6871
{
6872
boolean is_a_jump_n = false;
6873
6874
p1 = p + 2;
6875
mcnt = 0;
6876
switch ((re_opcode_t) *p1++)
6877
{
6878
case jump_n:
6879
is_a_jump_n = true;
6880
case pop_failure_jump:
6881
case maybe_pop_jump:
6882
case jump:
6883
case dummy_failure_jump:
6884
EXTRACT_NUMBER_AND_INCR (mcnt, p1);
6885
if (is_a_jump_n)
6886
p1 += OFFSET_ADDRESS_SIZE;
6887
break;
6888
6889
default:
6890
/* do nothing */ ;
6891
}
6892
p1 += mcnt;
6893
6894
/* If the next operation is a jump backwards in the pattern
6895
to an on_failure_jump right before the start_memory
6896
corresponding to this stop_memory, exit from the loop
6897
by forcing a failure after pushing on the stack the
6898
on_failure_jump's jump in the pattern, and d. */
6899
if (mcnt < 0 && (re_opcode_t) *p1 == on_failure_jump
6900
&& (re_opcode_t) p1[1+OFFSET_ADDRESS_SIZE] == start_memory
6901
&& p1[2+OFFSET_ADDRESS_SIZE] == *p)
6902
{
6903
/* If this group ever matched anything, then restore
6904
what its registers were before trying this last
6905
failed match, e.g., with `(a*)*b' against `ab' for
6906
regstart[1], and, e.g., with `((a*)*(b*)*)*'
6907
against `aba' for regend[3].
6908
6909
Also restore the registers for inner groups for,
6910
e.g., `((a*)(b*))*' against `aba' (register 3 would
6911
otherwise get trashed). */
6912
6913
if (EVER_MATCHED_SOMETHING (reg_info[*p]))
6914
{
6915
unsigned r;
6916
6917
EVER_MATCHED_SOMETHING (reg_info[*p]) = 0;
6918
6919
/* Restore this and inner groups' (if any) registers. */
6920
for (r = *p; r < (unsigned) *p + (unsigned) *(p + 1);
6921
r++)
6922
{
6923
regstart[r] = old_regstart[r];
6924
6925
/* xx why this test? */
6926
if (old_regend[r] >= regstart[r])
6927
regend[r] = old_regend[r];
6928
}
6929
}
6930
p1++;
6931
EXTRACT_NUMBER_AND_INCR (mcnt, p1);
6932
PUSH_FAILURE_POINT (p1 + mcnt, d, -2);
6933
6934
goto fail;
6935
}
6936
}
6937
6938
/* Move past the register number and the inner group count. */
6939
p += 2;
6940
NEXT;
6941
6942
6943
/* \<digit> has been turned into a `duplicate' command which is
6944
followed by the numeric value of <digit> as the register number. */
6945
CASE (duplicate):
6946
{
6947
register const CHAR_T *d2, *dend2;
6948
int regno = *p++; /* Get which register to match against. */
6949
DEBUG_PRINT2 ("EXECUTING duplicate %d.\n", regno);
6950
6951
/* Can't back reference a group which we've never matched. */
6952
if (REG_UNSET (regstart[regno]) || REG_UNSET (regend[regno]))
6953
goto fail;
6954
6955
/* Where in input to try to start matching. */
6956
d2 = regstart[regno];
6957
6958
/* Where to stop matching; if both the place to start and
6959
the place to stop matching are in the same string, then
6960
set to the place to stop, otherwise, for now have to use
6961
the end of the first string. */
6962
6963
dend2 = ((FIRST_STRING_P (regstart[regno])
6964
== FIRST_STRING_P (regend[regno]))
6965
? regend[regno] : end_match_1);
6966
for (;;)
6967
{
6968
/* If necessary, advance to next segment in register
6969
contents. */
6970
while (d2 == dend2)
6971
{
6972
if (dend2 == end_match_2) break;
6973
if (dend2 == regend[regno]) break;
6974
6975
/* End of string1 => advance to string2. */
6976
d2 = string2;
6977
dend2 = regend[regno];
6978
}
6979
/* At end of register contents => success */
6980
if (d2 == dend2) break;
6981
6982
/* If necessary, advance to next segment in data. */
6983
PREFETCH ();
6984
6985
/* How many characters left in this segment to match. */
6986
mcnt = dend - d;
6987
6988
/* Want how many consecutive characters we can match in
6989
one shot, so, if necessary, adjust the count. */
6990
if (mcnt > dend2 - d2)
6991
mcnt = dend2 - d2;
6992
6993
/* Compare that many; failure if mismatch, else move
6994
past them. */
6995
if (translate
6996
? PREFIX(bcmp_translate) (d, d2, mcnt, translate)
6997
: memcmp (d, d2, mcnt*sizeof(UCHAR_T)))
6998
goto fail;
6999
d += mcnt, d2 += mcnt;
7000
7001
/* Do this because we've match some characters. */
7002
SET_REGS_MATCHED ();
7003
}
7004
}
7005
NEXT;
7006
7007
7008
/* begline matches the empty string at the beginning of the string
7009
(unless `not_bol' is set in `bufp'), and, if
7010
`newline_anchor' is set, after newlines. */
7011
CASE (begline):
7012
DEBUG_PRINT1 ("EXECUTING begline.\n");
7013
7014
if (AT_STRINGS_BEG (d))
7015
{
7016
if (!bufp->not_bol)
7017
{
7018
NEXT;
7019
}
7020
}
7021
else if (d[-1] == '\n' && bufp->newline_anchor)
7022
{
7023
NEXT;
7024
}
7025
/* In all other cases, we fail. */
7026
goto fail;
7027
7028
7029
/* endline is the dual of begline. */
7030
CASE (endline):
7031
DEBUG_PRINT1 ("EXECUTING endline.\n");
7032
7033
if (AT_STRINGS_END (d))
7034
{
7035
if (!bufp->not_eol)
7036
{
7037
NEXT;
7038
}
7039
}
7040
7041
/* We have to ``prefetch'' the next character. */
7042
else if ((d == end1 ? *string2 : *d) == '\n'
7043
&& bufp->newline_anchor)
7044
{
7045
NEXT;
7046
}
7047
goto fail;
7048
7049
7050
/* Match at the very beginning of the data. */
7051
CASE (begbuf):
7052
DEBUG_PRINT1 ("EXECUTING begbuf.\n");
7053
if (AT_STRINGS_BEG (d))
7054
{
7055
NEXT;
7056
}
7057
goto fail;
7058
7059
7060
/* Match at the very end of the data. */
7061
CASE (endbuf):
7062
DEBUG_PRINT1 ("EXECUTING endbuf.\n");
7063
if (AT_STRINGS_END (d))
7064
{
7065
NEXT;
7066
}
7067
goto fail;
7068
7069
7070
/* on_failure_keep_string_jump is used to optimize `.*\n'. It
7071
pushes NULL as the value for the string on the stack. Then
7072
`pop_failure_point' will keep the current value for the
7073
string, instead of restoring it. To see why, consider
7074
matching `foo\nbar' against `.*\n'. The .* matches the foo;
7075
then the . fails against the \n. But the next thing we want
7076
to do is match the \n against the \n; if we restored the
7077
string value, we would be back at the foo.
7078
7079
Because this is used only in specific cases, we don't need to
7080
check all the things that `on_failure_jump' does, to make
7081
sure the right things get saved on the stack. Hence we don't
7082
share its code. The only reason to push anything on the
7083
stack at all is that otherwise we would have to change
7084
`anychar's code to do something besides goto fail in this
7085
case; that seems worse than this. */
7086
CASE (on_failure_keep_string_jump):
7087
DEBUG_PRINT1 ("EXECUTING on_failure_keep_string_jump");
7088
7089
EXTRACT_NUMBER_AND_INCR (mcnt, p);
7090
#ifdef _LIBC
7091
DEBUG_PRINT3 (" %d (to %p):\n", mcnt, p + mcnt);
7092
#else
7093
DEBUG_PRINT3 (" %d (to 0x%x):\n", mcnt, p + mcnt);
7094
#endif
7095
7096
PUSH_FAILURE_POINT (p + mcnt, NULL, -2);
7097
NEXT;
7098
7099
7100
/* Uses of on_failure_jump:
7101
7102
Each alternative starts with an on_failure_jump that points
7103
to the beginning of the next alternative. Each alternative
7104
except the last ends with a jump that in effect jumps past
7105
the rest of the alternatives. (They really jump to the
7106
ending jump of the following alternative, because tensioning
7107
these jumps is a hassle.)
7108
7109
Repeats start with an on_failure_jump that points past both
7110
the repetition text and either the following jump or
7111
pop_failure_jump back to this on_failure_jump. */
7112
CASE (on_failure_jump):
7113
on_failure:
7114
DEBUG_PRINT1 ("EXECUTING on_failure_jump");
7115
7116
EXTRACT_NUMBER_AND_INCR (mcnt, p);
7117
#ifdef _LIBC
7118
DEBUG_PRINT3 (" %d (to %p)", mcnt, p + mcnt);
7119
#else
7120
DEBUG_PRINT3 (" %d (to 0x%x)", mcnt, p + mcnt);
7121
#endif
7122
7123
/* If this on_failure_jump comes right before a group (i.e.,
7124
the original * applied to a group), save the information
7125
for that group and all inner ones, so that if we fail back
7126
to this point, the group's information will be correct.
7127
For example, in \(a*\)*\1, we need the preceding group,
7128
and in \(zz\(a*\)b*\)\2, we need the inner group. */
7129
7130
/* We can't use `p' to check ahead because we push
7131
a failure point to `p + mcnt' after we do this. */
7132
p1 = p;
7133
7134
/* We need to skip no_op's before we look for the
7135
start_memory in case this on_failure_jump is happening as
7136
the result of a completed succeed_n, as in \(a\)\{1,3\}b\1
7137
against aba. */
7138
while (p1 < pend && (re_opcode_t) *p1 == no_op)
7139
p1++;
7140
7141
if (p1 < pend && (re_opcode_t) *p1 == start_memory)
7142
{
7143
/* We have a new highest active register now. This will
7144
get reset at the start_memory we are about to get to,
7145
but we will have saved all the registers relevant to
7146
this repetition op, as described above. */
7147
highest_active_reg = *(p1 + 1) + *(p1 + 2);
7148
if (lowest_active_reg == NO_LOWEST_ACTIVE_REG)
7149
lowest_active_reg = *(p1 + 1);
7150
}
7151
7152
DEBUG_PRINT1 (":\n");
7153
PUSH_FAILURE_POINT (p + mcnt, d, -2);
7154
NEXT;
7155
7156
7157
/* A smart repeat ends with `maybe_pop_jump'.
7158
We change it to either `pop_failure_jump' or `jump'. */
7159
CASE (maybe_pop_jump):
7160
EXTRACT_NUMBER_AND_INCR (mcnt, p);
7161
DEBUG_PRINT2 ("EXECUTING maybe_pop_jump %d.\n", mcnt);
7162
{
7163
register UCHAR_T *p2 = p;
7164
7165
/* Compare the beginning of the repeat with what in the
7166
pattern follows its end. If we can establish that there
7167
is nothing that they would both match, i.e., that we
7168
would have to backtrack because of (as in, e.g., `a*a')
7169
then we can change to pop_failure_jump, because we'll
7170
never have to backtrack.
7171
7172
This is not true in the case of alternatives: in
7173
`(a|ab)*' we do need to backtrack to the `ab' alternative
7174
(e.g., if the string was `ab'). But instead of trying to
7175
detect that here, the alternative has put on a dummy
7176
failure point which is what we will end up popping. */
7177
7178
/* Skip over open/close-group commands.
7179
If what follows this loop is a ...+ construct,
7180
look at what begins its body, since we will have to
7181
match at least one of that. */
7182
while (1)
7183
{
7184
if (p2 + 2 < pend
7185
&& ((re_opcode_t) *p2 == stop_memory
7186
|| (re_opcode_t) *p2 == start_memory))
7187
p2 += 3;
7188
else if (p2 + 2 + 2 * OFFSET_ADDRESS_SIZE < pend
7189
&& (re_opcode_t) *p2 == dummy_failure_jump)
7190
p2 += 2 + 2 * OFFSET_ADDRESS_SIZE;
7191
else
7192
break;
7193
}
7194
7195
p1 = p + mcnt;
7196
/* p1[0] ... p1[2] are the `on_failure_jump' corresponding
7197
to the `maybe_finalize_jump' of this case. Examine what
7198
follows. */
7199
7200
/* If we're at the end of the pattern, we can change. */
7201
if (p2 == pend)
7202
{
7203
/* Consider what happens when matching ":\(.*\)"
7204
against ":/". I don't really understand this code
7205
yet. */
7206
p[-(1+OFFSET_ADDRESS_SIZE)] = (UCHAR_T)
7207
pop_failure_jump;
7208
DEBUG_PRINT1
7209
(" End of pattern: change to `pop_failure_jump'.\n");
7210
}
7211
7212
else if ((re_opcode_t) *p2 == exactn
7213
#ifdef MBS_SUPPORT
7214
|| (re_opcode_t) *p2 == exactn_bin
7215
#endif
7216
|| (bufp->newline_anchor && (re_opcode_t) *p2 == endline))
7217
{
7218
register UCHAR_T c
7219
= *p2 == (UCHAR_T) endline ? '\n' : p2[2];
7220
7221
if (((re_opcode_t) p1[1+OFFSET_ADDRESS_SIZE] == exactn
7222
#ifdef MBS_SUPPORT
7223
|| (re_opcode_t) p1[1+OFFSET_ADDRESS_SIZE] == exactn_bin
7224
#endif
7225
) && p1[3+OFFSET_ADDRESS_SIZE] != c)
7226
{
7227
p[-(1+OFFSET_ADDRESS_SIZE)] = (UCHAR_T)
7228
pop_failure_jump;
7229
#ifdef WCHAR
7230
DEBUG_PRINT3 (" %C != %C => pop_failure_jump.\n",
7231
(wint_t) c,
7232
(wint_t) p1[3+OFFSET_ADDRESS_SIZE]);
7233
#else
7234
DEBUG_PRINT3 (" %c != %c => pop_failure_jump.\n",
7235
(char) c,
7236
(char) p1[3+OFFSET_ADDRESS_SIZE]);
7237
#endif
7238
}
7239
7240
#ifndef WCHAR
7241
else if ((re_opcode_t) p1[3] == charset
7242
|| (re_opcode_t) p1[3] == charset_not)
7243
{
7244
int not = (re_opcode_t) p1[3] == charset_not;
7245
7246
if (c < (unsigned) (p1[4] * BYTEWIDTH)
7247
&& p1[5 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH)))
7248
not = !not;
7249
7250
/* `not' is equal to 1 if c would match, which means
7251
that we can't change to pop_failure_jump. */
7252
if (!not)
7253
{
7254
p[-3] = (unsigned char) pop_failure_jump;
7255
DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
7256
}
7257
}
7258
#endif /* not WCHAR */
7259
}
7260
#ifndef WCHAR
7261
else if ((re_opcode_t) *p2 == charset)
7262
{
7263
/* We win if the first character of the loop is not part
7264
of the charset. */
7265
if ((re_opcode_t) p1[3] == exactn
7266
&& ! ((int) p2[1] * BYTEWIDTH > (int) p1[5]
7267
&& (p2[2 + p1[5] / BYTEWIDTH]
7268
& (1 << (p1[5] % BYTEWIDTH)))))
7269
{
7270
p[-3] = (unsigned char) pop_failure_jump;
7271
DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
7272
}
7273
7274
else if ((re_opcode_t) p1[3] == charset_not)
7275
{
7276
int idx;
7277
/* We win if the charset_not inside the loop
7278
lists every character listed in the charset after. */
7279
for (idx = 0; idx < (int) p2[1]; idx++)
7280
if (! (p2[2 + idx] == 0
7281
|| (idx < (int) p1[4]
7282
&& ((p2[2 + idx] & ~ p1[5 + idx]) == 0))))
7283
break;
7284
7285
if (idx == p2[1])
7286
{
7287
p[-3] = (unsigned char) pop_failure_jump;
7288
DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
7289
}
7290
}
7291
else if ((re_opcode_t) p1[3] == charset)
7292
{
7293
int idx;
7294
/* We win if the charset inside the loop
7295
has no overlap with the one after the loop. */
7296
for (idx = 0;
7297
idx < (int) p2[1] && idx < (int) p1[4];
7298
idx++)
7299
if ((p2[2 + idx] & p1[5 + idx]) != 0)
7300
break;
7301
7302
if (idx == p2[1] || idx == p1[4])
7303
{
7304
p[-3] = (unsigned char) pop_failure_jump;
7305
DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
7306
}
7307
}
7308
}
7309
#endif /* not WCHAR */
7310
}
7311
p -= OFFSET_ADDRESS_SIZE; /* Point at relative address again. */
7312
if ((re_opcode_t) p[-1] != pop_failure_jump)
7313
{
7314
p[-1] = (UCHAR_T) jump;
7315
DEBUG_PRINT1 (" Match => jump.\n");
7316
goto unconditional_jump;
7317
}
7318
/* Note fall through. */
7319
7320
7321
/* The end of a simple repeat has a pop_failure_jump back to
7322
its matching on_failure_jump, where the latter will push a
7323
failure point. The pop_failure_jump takes off failure
7324
points put on by this pop_failure_jump's matching
7325
on_failure_jump; we got through the pattern to here from the
7326
matching on_failure_jump, so didn't fail. */
7327
CASE (pop_failure_jump):
7328
{
7329
/* We need to pass separate storage for the lowest and
7330
highest registers, even though we don't care about the
7331
actual values. Otherwise, we will restore only one
7332
register from the stack, since lowest will == highest in
7333
`pop_failure_point'. */
7334
active_reg_t dummy_low_reg, dummy_high_reg;
7335
UCHAR_T *pdummy = NULL;
7336
const CHAR_T *sdummy = NULL;
7337
7338
DEBUG_PRINT1 ("EXECUTING pop_failure_jump.\n");
7339
POP_FAILURE_POINT (sdummy, pdummy,
7340
dummy_low_reg, dummy_high_reg,
7341
reg_dummy, reg_dummy, reg_info_dummy);
7342
}
7343
/* Note fall through. */
7344
7345
unconditional_jump:
7346
#ifdef _LIBC
7347
DEBUG_PRINT2 ("\n%p: ", p);
7348
#else
7349
DEBUG_PRINT2 ("\n0x%x: ", p);
7350
#endif
7351
/* Note fall through. */
7352
7353
/* Unconditionally jump (without popping any failure points). */
7354
CASE (jump):
7355
EXTRACT_NUMBER_AND_INCR (mcnt, p); /* Get the amount to jump. */
7356
DEBUG_PRINT2 ("EXECUTING jump %d ", mcnt);
7357
p += mcnt; /* Do the jump. */
7358
#ifdef _LIBC
7359
DEBUG_PRINT2 ("(to %p).\n", p);
7360
#else
7361
DEBUG_PRINT2 ("(to 0x%x).\n", p);
7362
#endif
7363
NEXT;
7364
7365
7366
/* We need this opcode so we can detect where alternatives end
7367
in `group_match_null_string_p' et al. */
7368
CASE (jump_past_alt):
7369
DEBUG_PRINT1 ("EXECUTING jump_past_alt.\n");
7370
goto unconditional_jump;
7371
7372
7373
/* Normally, the on_failure_jump pushes a failure point, which
7374
then gets popped at pop_failure_jump. We will end up at
7375
pop_failure_jump, also, and with a pattern of, say, `a+', we
7376
are skipping over the on_failure_jump, so we have to push
7377
something meaningless for pop_failure_jump to pop. */
7378
CASE (dummy_failure_jump):
7379
DEBUG_PRINT1 ("EXECUTING dummy_failure_jump.\n");
7380
/* It doesn't matter what we push for the string here. What
7381
the code at `fail' tests is the value for the pattern. */
7382
PUSH_FAILURE_POINT (NULL, NULL, -2);
7383
goto unconditional_jump;
7384
7385
7386
/* At the end of an alternative, we need to push a dummy failure
7387
point in case we are followed by a `pop_failure_jump', because
7388
we don't want the failure point for the alternative to be
7389
popped. For example, matching `(a|ab)*' against `aab'
7390
requires that we match the `ab' alternative. */
7391
CASE (push_dummy_failure):
7392
DEBUG_PRINT1 ("EXECUTING push_dummy_failure.\n");
7393
/* See comments just above at `dummy_failure_jump' about the
7394
two zeroes. */
7395
PUSH_FAILURE_POINT (NULL, NULL, -2);
7396
NEXT;
7397
7398
/* Have to succeed matching what follows at least n times.
7399
After that, handle like `on_failure_jump'. */
7400
CASE (succeed_n):
7401
EXTRACT_NUMBER (mcnt, p + OFFSET_ADDRESS_SIZE);
7402
DEBUG_PRINT2 ("EXECUTING succeed_n %d.\n", mcnt);
7403
7404
assert (mcnt >= 0);
7405
/* Originally, this is how many times we HAVE to succeed. */
7406
if (mcnt > 0)
7407
{
7408
mcnt--;
7409
p += OFFSET_ADDRESS_SIZE;
7410
STORE_NUMBER_AND_INCR (p, mcnt);
7411
#ifdef _LIBC
7412
DEBUG_PRINT3 (" Setting %p to %d.\n", p - OFFSET_ADDRESS_SIZE
7413
, mcnt);
7414
#else
7415
DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p - OFFSET_ADDRESS_SIZE
7416
, mcnt);
7417
#endif
7418
}
7419
else if (mcnt == 0)
7420
{
7421
#ifdef _LIBC
7422
DEBUG_PRINT2 (" Setting two bytes from %p to no_op.\n",
7423
p + OFFSET_ADDRESS_SIZE);
7424
#else
7425
DEBUG_PRINT2 (" Setting two bytes from 0x%x to no_op.\n",
7426
p + OFFSET_ADDRESS_SIZE);
7427
#endif /* _LIBC */
7428
7429
#ifdef WCHAR
7430
p[1] = (UCHAR_T) no_op;
7431
#else
7432
p[2] = (UCHAR_T) no_op;
7433
p[3] = (UCHAR_T) no_op;
7434
#endif /* WCHAR */
7435
goto on_failure;
7436
}
7437
NEXT;
7438
7439
CASE (jump_n):
7440
EXTRACT_NUMBER (mcnt, p + OFFSET_ADDRESS_SIZE);
7441
DEBUG_PRINT2 ("EXECUTING jump_n %d.\n", mcnt);
7442
7443
/* Originally, this is how many times we CAN jump. */
7444
if (mcnt)
7445
{
7446
mcnt--;
7447
STORE_NUMBER (p + OFFSET_ADDRESS_SIZE, mcnt);
7448
7449
#ifdef _LIBC
7450
DEBUG_PRINT3 (" Setting %p to %d.\n", p + OFFSET_ADDRESS_SIZE,
7451
mcnt);
7452
#else
7453
DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p + OFFSET_ADDRESS_SIZE,
7454
mcnt);
7455
#endif /* _LIBC */
7456
goto unconditional_jump;
7457
}
7458
/* If don't have to jump any more, skip over the rest of command. */
7459
else
7460
p += 2 * OFFSET_ADDRESS_SIZE;
7461
NEXT;
7462
7463
CASE (set_number_at):
7464
{
7465
DEBUG_PRINT1 ("EXECUTING set_number_at.\n");
7466
7467
EXTRACT_NUMBER_AND_INCR (mcnt, p);
7468
p1 = p + mcnt;
7469
EXTRACT_NUMBER_AND_INCR (mcnt, p);
7470
#ifdef _LIBC
7471
DEBUG_PRINT3 (" Setting %p to %d.\n", p1, mcnt);
7472
#else
7473
DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p1, mcnt);
7474
#endif
7475
STORE_NUMBER (p1, mcnt);
7476
NEXT;
7477
}
7478
7479
#if 0
7480
/* The DEC Alpha C compiler 3.x generates incorrect code for the
7481
test WORDCHAR_P (d - 1) != WORDCHAR_P (d) in the expansion of
7482
AT_WORD_BOUNDARY, so this code is disabled. Expanding the
7483
macro and introducing temporary variables works around the bug. */
7484
7485
CASE (wordbound):
7486
DEBUG_PRINT1 ("EXECUTING wordbound.\n");
7487
if (AT_WORD_BOUNDARY (d))
7488
{
7489
NEXT;
7490
}
7491
goto fail;
7492
7493
CASE (notwordbound):
7494
DEBUG_PRINT1 ("EXECUTING notwordbound.\n");
7495
if (AT_WORD_BOUNDARY (d))
7496
goto fail;
7497
NEXT;
7498
#else
7499
CASE (wordbound):
7500
{
7501
boolean prevchar, thischar;
7502
7503
DEBUG_PRINT1 ("EXECUTING wordbound.\n");
7504
if (AT_STRINGS_BEG (d) || AT_STRINGS_END (d))
7505
{
7506
NEXT;
7507
}
7508
7509
prevchar = WORDCHAR_P (d - 1);
7510
thischar = WORDCHAR_P (d);
7511
if (prevchar != thischar)
7512
{
7513
NEXT;
7514
}
7515
goto fail;
7516
}
7517
7518
CASE (notwordbound):
7519
{
7520
boolean prevchar, thischar;
7521
7522
DEBUG_PRINT1 ("EXECUTING notwordbound.\n");
7523
if (AT_STRINGS_BEG (d) || AT_STRINGS_END (d))
7524
goto fail;
7525
7526
prevchar = WORDCHAR_P (d - 1);
7527
thischar = WORDCHAR_P (d);
7528
if (prevchar != thischar)
7529
goto fail;
7530
NEXT;
7531
}
7532
#endif
7533
7534
CASE (wordbeg):
7535
DEBUG_PRINT1 ("EXECUTING wordbeg.\n");
7536
if (!AT_STRINGS_END (d) && WORDCHAR_P (d)
7537
&& (AT_STRINGS_BEG (d) || !WORDCHAR_P (d - 1)))
7538
{
7539
NEXT;
7540
}
7541
goto fail;
7542
7543
CASE (wordend):
7544
DEBUG_PRINT1 ("EXECUTING wordend.\n");
7545
if (!AT_STRINGS_BEG (d) && WORDCHAR_P (d - 1)
7546
&& (AT_STRINGS_END (d) || !WORDCHAR_P (d)))
7547
{
7548
NEXT;
7549
}
7550
goto fail;
7551
7552
#ifdef emacs
7553
CASE (before_dot):
7554
DEBUG_PRINT1 ("EXECUTING before_dot.\n");
7555
if (PTR_CHAR_POS ((unsigned char *) d) >= point)
7556
goto fail;
7557
NEXT;
7558
7559
CASE (at_dot):
7560
DEBUG_PRINT1 ("EXECUTING at_dot.\n");
7561
if (PTR_CHAR_POS ((unsigned char *) d) != point)
7562
goto fail;
7563
NEXT;
7564
7565
CASE (after_dot):
7566
DEBUG_PRINT1 ("EXECUTING after_dot.\n");
7567
if (PTR_CHAR_POS ((unsigned char *) d) <= point)
7568
goto fail;
7569
NEXT;
7570
7571
CASE (syntaxspec):
7572
DEBUG_PRINT2 ("EXECUTING syntaxspec %d.\n", mcnt);
7573
mcnt = *p++;
7574
goto matchsyntax;
7575
7576
CASE (wordchar):
7577
DEBUG_PRINT1 ("EXECUTING Emacs wordchar.\n");
7578
mcnt = (int) Sword;
7579
matchsyntax:
7580
PREFETCH ();
7581
/* Can't use *d++ here; SYNTAX may be an unsafe macro. */
7582
d++;
7583
if (SYNTAX (d[-1]) != (enum syntaxcode) mcnt)
7584
goto fail;
7585
SET_REGS_MATCHED ();
7586
NEXT;
7587
7588
CASE (notsyntaxspec):
7589
DEBUG_PRINT2 ("EXECUTING notsyntaxspec %d.\n", mcnt);
7590
mcnt = *p++;
7591
goto matchnotsyntax;
7592
7593
CASE (notwordchar):
7594
DEBUG_PRINT1 ("EXECUTING Emacs notwordchar.\n");
7595
mcnt = (int) Sword;
7596
matchnotsyntax:
7597
PREFETCH ();
7598
/* Can't use *d++ here; SYNTAX may be an unsafe macro. */
7599
d++;
7600
if (SYNTAX (d[-1]) == (enum syntaxcode) mcnt)
7601
goto fail;
7602
SET_REGS_MATCHED ();
7603
NEXT;
7604
7605
#else /* not emacs */
7606
CASE (wordchar):
7607
DEBUG_PRINT1 ("EXECUTING non-Emacs wordchar.\n");
7608
PREFETCH ();
7609
if (!WORDCHAR_P (d))
7610
goto fail;
7611
SET_REGS_MATCHED ();
7612
d++;
7613
NEXT;
7614
7615
CASE (notwordchar):
7616
DEBUG_PRINT1 ("EXECUTING non-Emacs notwordchar.\n");
7617
PREFETCH ();
7618
if (WORDCHAR_P (d))
7619
goto fail;
7620
SET_REGS_MATCHED ();
7621
d++;
7622
NEXT;
7623
#endif /* not emacs */
7624
7625
#ifndef __GNUC__
7626
default:
7627
abort ();
7628
}
7629
continue; /* Successfully executed one pattern command; keep going. */
7630
#endif
7631
7632
7633
/* We goto here if a matching operation fails. */
7634
fail:
7635
if (!FAIL_STACK_EMPTY ())
7636
{ /* A restart point is known. Restore to that state. */
7637
DEBUG_PRINT1 ("\nFAIL:\n");
7638
POP_FAILURE_POINT (d, p,
7639
lowest_active_reg, highest_active_reg,
7640
regstart, regend, reg_info);
7641
7642
/* If this failure point is a dummy, try the next one. */
7643
if (!p)
7644
goto fail;
7645
7646
/* If we failed to the end of the pattern, don't examine *p. */
7647
assert (p <= pend);
7648
if (p < pend)
7649
{
7650
boolean is_a_jump_n = false;
7651
7652
/* If failed to a backwards jump that's part of a repetition
7653
loop, need to pop this failure point and use the next one. */
7654
switch ((re_opcode_t) *p)
7655
{
7656
case jump_n:
7657
is_a_jump_n = true;
7658
case maybe_pop_jump:
7659
case pop_failure_jump:
7660
case jump:
7661
p1 = p + 1;
7662
EXTRACT_NUMBER_AND_INCR (mcnt, p1);
7663
p1 += mcnt;
7664
7665
if ((is_a_jump_n && (re_opcode_t) *p1 == succeed_n)
7666
|| (!is_a_jump_n
7667
&& (re_opcode_t) *p1 == on_failure_jump))
7668
goto fail;
7669
break;
7670
default:
7671
/* do nothing */ ;
7672
}
7673
}
7674
7675
if (d >= string1 && d <= end1)
7676
dend = end_match_1;
7677
}
7678
else
7679
break; /* Matching at this starting point really fails. */
7680
} /* for (;;) */
7681
7682
if (best_regs_set)
7683
goto restore_best_regs;
7684
7685
FREE_VARIABLES ();
7686
7687
return -1; /* Failure to match. */
7688
} /* re_match_2 */
7689
7690
/* Subroutine definitions for re_match_2. */
7691
7692
7693
/* We are passed P pointing to a register number after a start_memory.
7694
7695
Return true if the pattern up to the corresponding stop_memory can
7696
match the empty string, and false otherwise.
7697
7698
If we find the matching stop_memory, sets P to point to one past its number.
7699
Otherwise, sets P to an undefined byte less than or equal to END.
7700
7701
We don't handle duplicates properly (yet). */
7702
7703
static boolean
7704
PREFIX(group_match_null_string_p) (p, end, reg_info)
7705
UCHAR_T **p, *end;
7706
PREFIX(register_info_type) *reg_info;
7707
{
7708
int mcnt;
7709
/* Point to after the args to the start_memory. */
7710
UCHAR_T *p1 = *p + 2;
7711
7712
while (p1 < end)
7713
{
7714
/* Skip over opcodes that can match nothing, and return true or
7715
false, as appropriate, when we get to one that can't, or to the
7716
matching stop_memory. */
7717
7718
switch ((re_opcode_t) *p1)
7719
{
7720
/* Could be either a loop or a series of alternatives. */
7721
case on_failure_jump:
7722
p1++;
7723
EXTRACT_NUMBER_AND_INCR (mcnt, p1);
7724
7725
/* If the next operation is not a jump backwards in the
7726
pattern. */
7727
7728
if (mcnt >= 0)
7729
{
7730
/* Go through the on_failure_jumps of the alternatives,
7731
seeing if any of the alternatives cannot match nothing.
7732
The last alternative starts with only a jump,
7733
whereas the rest start with on_failure_jump and end
7734
with a jump, e.g., here is the pattern for `a|b|c':
7735
7736
/on_failure_jump/0/6/exactn/1/a/jump_past_alt/0/6
7737
/on_failure_jump/0/6/exactn/1/b/jump_past_alt/0/3
7738
/exactn/1/c
7739
7740
So, we have to first go through the first (n-1)
7741
alternatives and then deal with the last one separately. */
7742
7743
7744
/* Deal with the first (n-1) alternatives, which start
7745
with an on_failure_jump (see above) that jumps to right
7746
past a jump_past_alt. */
7747
7748
while ((re_opcode_t) p1[mcnt-(1+OFFSET_ADDRESS_SIZE)] ==
7749
jump_past_alt)
7750
{
7751
/* `mcnt' holds how many bytes long the alternative
7752
is, including the ending `jump_past_alt' and
7753
its number. */
7754
7755
if (!PREFIX(alt_match_null_string_p) (p1, p1 + mcnt -
7756
(1 + OFFSET_ADDRESS_SIZE),
7757
reg_info))
7758
return false;
7759
7760
/* Move to right after this alternative, including the
7761
jump_past_alt. */
7762
p1 += mcnt;
7763
7764
/* Break if it's the beginning of an n-th alternative
7765
that doesn't begin with an on_failure_jump. */
7766
if ((re_opcode_t) *p1 != on_failure_jump)
7767
break;
7768
7769
/* Still have to check that it's not an n-th
7770
alternative that starts with an on_failure_jump. */
7771
p1++;
7772
EXTRACT_NUMBER_AND_INCR (mcnt, p1);
7773
if ((re_opcode_t) p1[mcnt-(1+OFFSET_ADDRESS_SIZE)] !=
7774
jump_past_alt)
7775
{
7776
/* Get to the beginning of the n-th alternative. */
7777
p1 -= 1 + OFFSET_ADDRESS_SIZE;
7778
break;
7779
}
7780
}
7781
7782
/* Deal with the last alternative: go back and get number
7783
of the `jump_past_alt' just before it. `mcnt' contains
7784
the length of the alternative. */
7785
EXTRACT_NUMBER (mcnt, p1 - OFFSET_ADDRESS_SIZE);
7786
7787
if (!PREFIX(alt_match_null_string_p) (p1, p1 + mcnt, reg_info))
7788
return false;
7789
7790
p1 += mcnt; /* Get past the n-th alternative. */
7791
} /* if mcnt > 0 */
7792
break;
7793
7794
7795
case stop_memory:
7796
assert (p1[1] == **p);
7797
*p = p1 + 2;
7798
return true;
7799
7800
7801
default:
7802
if (!PREFIX(common_op_match_null_string_p) (&p1, end, reg_info))
7803
return false;
7804
}
7805
} /* while p1 < end */
7806
7807
return false;
7808
} /* group_match_null_string_p */
7809
7810
7811
/* Similar to group_match_null_string_p, but doesn't deal with alternatives:
7812
It expects P to be the first byte of a single alternative and END one
7813
byte past the last. The alternative can contain groups. */
7814
7815
static boolean
7816
PREFIX(alt_match_null_string_p) (p, end, reg_info)
7817
UCHAR_T *p, *end;
7818
PREFIX(register_info_type) *reg_info;
7819
{
7820
int mcnt;
7821
UCHAR_T *p1 = p;
7822
7823
while (p1 < end)
7824
{
7825
/* Skip over opcodes that can match nothing, and break when we get
7826
to one that can't. */
7827
7828
switch ((re_opcode_t) *p1)
7829
{
7830
/* It's a loop. */
7831
case on_failure_jump:
7832
p1++;
7833
EXTRACT_NUMBER_AND_INCR (mcnt, p1);
7834
p1 += mcnt;
7835
break;
7836
7837
default:
7838
if (!PREFIX(common_op_match_null_string_p) (&p1, end, reg_info))
7839
return false;
7840
}
7841
} /* while p1 < end */
7842
7843
return true;
7844
} /* alt_match_null_string_p */
7845
7846
7847
/* Deals with the ops common to group_match_null_string_p and
7848
alt_match_null_string_p.
7849
7850
Sets P to one after the op and its arguments, if any. */
7851
7852
static boolean
7853
PREFIX(common_op_match_null_string_p) (p, end, reg_info)
7854
UCHAR_T **p, *end;
7855
PREFIX(register_info_type) *reg_info;
7856
{
7857
int mcnt;
7858
boolean ret;
7859
int reg_no;
7860
UCHAR_T *p1 = *p;
7861
7862
switch ((re_opcode_t) *p1++)
7863
{
7864
case no_op:
7865
case begline:
7866
case endline:
7867
case begbuf:
7868
case endbuf:
7869
case wordbeg:
7870
case wordend:
7871
case wordbound:
7872
case notwordbound:
7873
#ifdef emacs
7874
case before_dot:
7875
case at_dot:
7876
case after_dot:
7877
#endif
7878
break;
7879
7880
case start_memory:
7881
reg_no = *p1;
7882
assert (reg_no > 0 && reg_no <= MAX_REGNUM);
7883
ret = PREFIX(group_match_null_string_p) (&p1, end, reg_info);
7884
7885
/* Have to set this here in case we're checking a group which
7886
contains a group and a back reference to it. */
7887
7888
if (REG_MATCH_NULL_STRING_P (reg_info[reg_no]) == MATCH_NULL_UNSET_VALUE)
7889
REG_MATCH_NULL_STRING_P (reg_info[reg_no]) = ret;
7890
7891
if (!ret)
7892
return false;
7893
break;
7894
7895
/* If this is an optimized succeed_n for zero times, make the jump. */
7896
case jump:
7897
EXTRACT_NUMBER_AND_INCR (mcnt, p1);
7898
if (mcnt >= 0)
7899
p1 += mcnt;
7900
else
7901
return false;
7902
break;
7903
7904
case succeed_n:
7905
/* Get to the number of times to succeed. */
7906
p1 += OFFSET_ADDRESS_SIZE;
7907
EXTRACT_NUMBER_AND_INCR (mcnt, p1);
7908
7909
if (mcnt == 0)
7910
{
7911
p1 -= 2 * OFFSET_ADDRESS_SIZE;
7912
EXTRACT_NUMBER_AND_INCR (mcnt, p1);
7913
p1 += mcnt;
7914
}
7915
else
7916
return false;
7917
break;
7918
7919
case duplicate:
7920
if (!REG_MATCH_NULL_STRING_P (reg_info[*p1]))
7921
return false;
7922
break;
7923
7924
case set_number_at:
7925
p1 += 2 * OFFSET_ADDRESS_SIZE;
7926
7927
default:
7928
/* All other opcodes mean we cannot match the empty string. */
7929
return false;
7930
}
7931
7932
*p = p1;
7933
return true;
7934
} /* common_op_match_null_string_p */
7935
7936
7937
/* Return zero if TRANSLATE[S1] and TRANSLATE[S2] are identical for LEN
7938
bytes; nonzero otherwise. */
7939
7940
static int
7941
PREFIX(bcmp_translate) (s1, s2, len, translate)
7942
const CHAR_T *s1, *s2;
7943
register int len;
7944
RE_TRANSLATE_TYPE translate;
7945
{
7946
register const UCHAR_T *p1 = (const UCHAR_T *) s1;
7947
register const UCHAR_T *p2 = (const UCHAR_T *) s2;
7948
while (len)
7949
{
7950
#ifdef WCHAR
7951
if (((*p1<=0xff)?translate[*p1++]:*p1++)
7952
!= ((*p2<=0xff)?translate[*p2++]:*p2++))
7953
return 1;
7954
#else /* BYTE */
7955
if (translate[*p1++] != translate[*p2++]) return 1;
7956
#endif /* WCHAR */
7957
len--;
7958
}
7959
return 0;
7960
}
7961
7962
7963
#else /* not INSIDE_RECURSION */
7964
7965
/* Entry points for GNU code. */
7966
7967
/* re_compile_pattern is the GNU regular expression compiler: it
7968
compiles PATTERN (of length SIZE) and puts the result in BUFP.
7969
Returns 0 if the pattern was valid, otherwise an error string.
7970
7971
Assumes the `allocated' (and perhaps `buffer') and `translate' fields
7972
are set in BUFP on entry.
7973
7974
We call regex_compile to do the actual compilation. */
7975
7976
const char *
7977
re_compile_pattern (pattern, length, bufp)
7978
const char *pattern;
7979
size_t length;
7980
struct re_pattern_buffer *bufp;
7981
{
7982
reg_errcode_t ret;
7983
7984
/* GNU code is written to assume at least RE_NREGS registers will be set
7985
(and at least one extra will be -1). */
7986
bufp->regs_allocated = REGS_UNALLOCATED;
7987
7988
/* And GNU code determines whether or not to get register information
7989
by passing null for the REGS argument to re_match, etc., not by
7990
setting no_sub. */
7991
bufp->no_sub = 0;
7992
7993
/* Match anchors at newline. */
7994
bufp->newline_anchor = 1;
7995
7996
# ifdef MBS_SUPPORT
7997
if (MB_CUR_MAX != 1)
7998
ret = wcs_regex_compile (pattern, length, re_syntax_options, bufp);
7999
else
8000
# endif
8001
ret = byte_regex_compile (pattern, length, re_syntax_options, bufp);
8002
8003
if (!ret)
8004
return NULL;
8005
return gettext (re_error_msgid + re_error_msgid_idx[(int) ret]);
8006
}
8007
#ifdef _LIBC
8008
weak_alias (__re_compile_pattern, re_compile_pattern)
8009
#endif
8010
8011
/* Entry points compatible with 4.2 BSD regex library. We don't define
8012
them unless specifically requested. */
8013
8014
#if defined _REGEX_RE_COMP || defined _LIBC
8015
8016
/* BSD has one and only one pattern buffer. */
8017
static struct re_pattern_buffer re_comp_buf;
8018
8019
char *
8020
#ifdef _LIBC
8021
/* Make these definitions weak in libc, so POSIX programs can redefine
8022
these names if they don't use our functions, and still use
8023
regcomp/regexec below without link errors. */
8024
weak_function
8025
#endif
8026
re_comp (s)
8027
const char *s;
8028
{
8029
reg_errcode_t ret;
8030
8031
if (!s)
8032
{
8033
if (!re_comp_buf.buffer)
8034
return gettext ("No previous regular expression");
8035
return 0;
8036
}
8037
8038
if (!re_comp_buf.buffer)
8039
{
8040
re_comp_buf.buffer = (unsigned char *) malloc (200);
8041
if (re_comp_buf.buffer == NULL)
8042
return (char *) gettext (re_error_msgid
8043
+ re_error_msgid_idx[(int) REG_ESPACE]);
8044
re_comp_buf.allocated = 200;
8045
8046
re_comp_buf.fastmap = (char *) malloc (1 << BYTEWIDTH);
8047
if (re_comp_buf.fastmap == NULL)
8048
return (char *) gettext (re_error_msgid
8049
+ re_error_msgid_idx[(int) REG_ESPACE]);
8050
}
8051
8052
/* Since `re_exec' always passes NULL for the `regs' argument, we
8053
don't need to initialize the pattern buffer fields which affect it. */
8054
8055
/* Match anchors at newlines. */
8056
re_comp_buf.newline_anchor = 1;
8057
8058
# ifdef MBS_SUPPORT
8059
if (MB_CUR_MAX != 1)
8060
ret = wcs_regex_compile (s, strlen (s), re_syntax_options, &re_comp_buf);
8061
else
8062
# endif
8063
ret = byte_regex_compile (s, strlen (s), re_syntax_options, &re_comp_buf);
8064
8065
if (!ret)
8066
return NULL;
8067
8068
/* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
8069
return (char *) gettext (re_error_msgid + re_error_msgid_idx[(int) ret]);
8070
}
8071
8072
8073
int
8074
#ifdef _LIBC
8075
weak_function
8076
#endif
8077
re_exec (s)
8078
const char *s;
8079
{
8080
const int len = strlen (s);
8081
return
8082
0 <= re_search (&re_comp_buf, s, len, 0, len, (struct re_registers *) 0);
8083
}
8084
8085
#endif /* _REGEX_RE_COMP */
8086
8087
/* POSIX.2 functions. Don't define these for Emacs. */
8088
8089
#ifndef emacs
8090
8091
/* regcomp takes a regular expression as a string and compiles it.
8092
8093
PREG is a regex_t *. We do not expect any fields to be initialized,
8094
since POSIX says we shouldn't. Thus, we set
8095
8096
`buffer' to the compiled pattern;
8097
`used' to the length of the compiled pattern;
8098
`syntax' to RE_SYNTAX_POSIX_EXTENDED if the
8099
REG_EXTENDED bit in CFLAGS is set; otherwise, to
8100
RE_SYNTAX_POSIX_BASIC;
8101
`newline_anchor' to REG_NEWLINE being set in CFLAGS;
8102
`fastmap' to an allocated space for the fastmap;
8103
`fastmap_accurate' to zero;
8104
`re_nsub' to the number of subexpressions in PATTERN.
8105
8106
PATTERN is the address of the pattern string.
8107
8108
CFLAGS is a series of bits which affect compilation.
8109
8110
If REG_EXTENDED is set, we use POSIX extended syntax; otherwise, we
8111
use POSIX basic syntax.
8112
8113
If REG_NEWLINE is set, then . and [^...] don't match newline.
8114
Also, regexec will try a match beginning after every newline.
8115
8116
If REG_ICASE is set, then we considers upper- and lowercase
8117
versions of letters to be equivalent when matching.
8118
8119
If REG_NOSUB is set, then when PREG is passed to regexec, that
8120
routine will report only success or failure, and nothing about the
8121
registers.
8122
8123
It returns 0 if it succeeds, nonzero if it doesn't. (See regex.h for
8124
the return codes and their meanings.) */
8125
8126
int
8127
regcomp (preg, pattern, cflags)
8128
regex_t *preg;
8129
const char *pattern;
8130
int cflags;
8131
{
8132
reg_errcode_t ret;
8133
reg_syntax_t syntax
8134
= (cflags & REG_EXTENDED) ?
8135
RE_SYNTAX_POSIX_EXTENDED : RE_SYNTAX_POSIX_BASIC;
8136
8137
/* regex_compile will allocate the space for the compiled pattern. */
8138
preg->buffer = 0;
8139
preg->allocated = 0;
8140
preg->used = 0;
8141
8142
/* Try to allocate space for the fastmap. */
8143
preg->fastmap = (char *) malloc (1 << BYTEWIDTH);
8144
8145
if (cflags & REG_ICASE)
8146
{
8147
unsigned i;
8148
8149
preg->translate
8150
= (RE_TRANSLATE_TYPE) malloc (CHAR_SET_SIZE
8151
* sizeof (*(RE_TRANSLATE_TYPE)0));
8152
if (preg->translate == NULL)
8153
return (int) REG_ESPACE;
8154
8155
/* Map uppercase characters to corresponding lowercase ones. */
8156
for (i = 0; i < CHAR_SET_SIZE; i++)
8157
preg->translate[i] = ISUPPER (i) ? TOLOWER (i) : i;
8158
}
8159
else
8160
preg->translate = NULL;
8161
8162
/* If REG_NEWLINE is set, newlines are treated differently. */
8163
if (cflags & REG_NEWLINE)
8164
{ /* REG_NEWLINE implies neither . nor [^...] match newline. */
8165
syntax &= ~RE_DOT_NEWLINE;
8166
syntax |= RE_HAT_LISTS_NOT_NEWLINE;
8167
/* It also changes the matching behavior. */
8168
preg->newline_anchor = 1;
8169
}
8170
else
8171
preg->newline_anchor = 0;
8172
8173
preg->no_sub = !!(cflags & REG_NOSUB);
8174
8175
/* POSIX says a null character in the pattern terminates it, so we
8176
can use strlen here in compiling the pattern. */
8177
# ifdef MBS_SUPPORT
8178
if (MB_CUR_MAX != 1)
8179
ret = wcs_regex_compile (pattern, strlen (pattern), syntax, preg);
8180
else
8181
# endif
8182
ret = byte_regex_compile (pattern, strlen (pattern), syntax, preg);
8183
8184
/* POSIX doesn't distinguish between an unmatched open-group and an
8185
unmatched close-group: both are REG_EPAREN. */
8186
if (ret == REG_ERPAREN) ret = REG_EPAREN;
8187
8188
if (ret == REG_NOERROR && preg->fastmap)
8189
{
8190
/* Compute the fastmap now, since regexec cannot modify the pattern
8191
buffer. */
8192
if (re_compile_fastmap (preg) == -2)
8193
{
8194
/* Some error occurred while computing the fastmap, just forget
8195
about it. */
8196
free (preg->fastmap);
8197
preg->fastmap = NULL;
8198
}
8199
}
8200
8201
return (int) ret;
8202
}
8203
#ifdef _LIBC
8204
weak_alias (__regcomp, regcomp)
8205
#endif
8206
8207
8208
/* regexec searches for a given pattern, specified by PREG, in the
8209
string STRING.
8210
8211
If NMATCH is zero or REG_NOSUB was set in the cflags argument to
8212
`regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at
8213
least NMATCH elements, and we set them to the offsets of the
8214
corresponding matched substrings.
8215
8216
EFLAGS specifies `execution flags' which affect matching: if
8217
REG_NOTBOL is set, then ^ does not match at the beginning of the
8218
string; if REG_NOTEOL is set, then $ does not match at the end.
8219
8220
We return 0 if we find a match and REG_NOMATCH if not. */
8221
8222
int
8223
regexec (preg, string, nmatch, pmatch, eflags)
8224
const regex_t *preg;
8225
const char *string;
8226
size_t nmatch;
8227
regmatch_t pmatch[];
8228
int eflags;
8229
{
8230
int ret;
8231
struct re_registers regs;
8232
regex_t private_preg;
8233
int len = strlen (string);
8234
boolean want_reg_info = !preg->no_sub && nmatch > 0;
8235
8236
private_preg = *preg;
8237
8238
private_preg.not_bol = !!(eflags & REG_NOTBOL);
8239
private_preg.not_eol = !!(eflags & REG_NOTEOL);
8240
8241
/* The user has told us exactly how many registers to return
8242
information about, via `nmatch'. We have to pass that on to the
8243
matching routines. */
8244
private_preg.regs_allocated = REGS_FIXED;
8245
8246
if (want_reg_info)
8247
{
8248
regs.num_regs = nmatch;
8249
regs.start = TALLOC (nmatch * 2, regoff_t);
8250
if (regs.start == NULL)
8251
return (int) REG_NOMATCH;
8252
regs.end = regs.start + nmatch;
8253
}
8254
8255
/* Perform the searching operation. */
8256
ret = re_search (&private_preg, string, len,
8257
/* start: */ 0, /* range: */ len,
8258
want_reg_info ? ®s : (struct re_registers *) 0);
8259
8260
/* Copy the register information to the POSIX structure. */
8261
if (want_reg_info)
8262
{
8263
if (ret >= 0)
8264
{
8265
unsigned r;
8266
8267
for (r = 0; r < nmatch; r++)
8268
{
8269
pmatch[r].rm_so = regs.start[r];
8270
pmatch[r].rm_eo = regs.end[r];
8271
}
8272
}
8273
8274
/* If we needed the temporary register info, free the space now. */
8275
free (regs.start);
8276
}
8277
8278
/* We want zero return to mean success, unlike `re_search'. */
8279
return ret >= 0 ? (int) REG_NOERROR : (int) REG_NOMATCH;
8280
}
8281
#ifdef _LIBC
8282
weak_alias (__regexec, regexec)
8283
#endif
8284
8285
8286
/* Returns a message corresponding to an error code, ERRCODE, returned
8287
from either regcomp or regexec. We don't use PREG here. */
8288
8289
size_t
8290
regerror (errcode, preg, errbuf, errbuf_size)
8291
int errcode;
8292
const regex_t *preg;
8293
char *errbuf;
8294
size_t errbuf_size;
8295
{
8296
const char *msg;
8297
size_t msg_size;
8298
8299
if (errcode < 0
8300
|| errcode >= (int) (sizeof (re_error_msgid_idx)
8301
/ sizeof (re_error_msgid_idx[0])))
8302
/* Only error codes returned by the rest of the code should be passed
8303
to this routine. If we are given anything else, or if other regex
8304
code generates an invalid error code, then the program has a bug.
8305
Dump core so we can fix it. */
8306
abort ();
8307
8308
msg = gettext (re_error_msgid + re_error_msgid_idx[errcode]);
8309
8310
msg_size = strlen (msg) + 1; /* Includes the null. */
8311
8312
if (errbuf_size != 0)
8313
{
8314
if (msg_size > errbuf_size)
8315
{
8316
#if defined HAVE_MEMPCPY || defined _LIBC
8317
*((char *) __mempcpy (errbuf, msg, errbuf_size - 1)) = '\0';
8318
#else
8319
memcpy (errbuf, msg, errbuf_size - 1);
8320
errbuf[errbuf_size - 1] = 0;
8321
#endif
8322
}
8323
else
8324
memcpy (errbuf, msg, msg_size);
8325
}
8326
8327
return msg_size;
8328
}
8329
#ifdef _LIBC
8330
weak_alias (__regerror, regerror)
8331
#endif
8332
8333
8334
/* Free dynamically allocated space used by PREG. */
8335
8336
void
8337
regfree (preg)
8338
regex_t *preg;
8339
{
8340
if (preg->buffer != NULL)
8341
free (preg->buffer);
8342
preg->buffer = NULL;
8343
8344
preg->allocated = 0;
8345
preg->used = 0;
8346
8347
if (preg->fastmap != NULL)
8348
free (preg->fastmap);
8349
preg->fastmap = NULL;
8350
preg->fastmap_accurate = 0;
8351
8352
if (preg->translate != NULL)
8353
free (preg->translate);
8354
preg->translate = NULL;
8355
}
8356
#ifdef _LIBC
8357
weak_alias (__regfree, regfree)
8358
#endif
8359
8360
#endif /* not emacs */
8361
8362
#endif /* not INSIDE_RECURSION */
8363
8364
8365
#undef STORE_NUMBER
8366
#undef STORE_NUMBER_AND_INCR
8367
#undef EXTRACT_NUMBER
8368
#undef EXTRACT_NUMBER_AND_INCR
8369
8370
#undef DEBUG_PRINT_COMPILED_PATTERN
8371
#undef DEBUG_PRINT_DOUBLE_STRING
8372
8373
#undef INIT_FAIL_STACK
8374
#undef RESET_FAIL_STACK
8375
#undef DOUBLE_FAIL_STACK
8376
#undef PUSH_PATTERN_OP
8377
#undef PUSH_FAILURE_POINTER
8378
#undef PUSH_FAILURE_INT
8379
#undef PUSH_FAILURE_ELT
8380
#undef POP_FAILURE_POINTER
8381
#undef POP_FAILURE_INT
8382
#undef POP_FAILURE_ELT
8383
#undef DEBUG_PUSH
8384
#undef DEBUG_POP
8385
#undef PUSH_FAILURE_POINT
8386
#undef POP_FAILURE_POINT
8387
8388
#undef REG_UNSET_VALUE
8389
#undef REG_UNSET
8390
8391
#undef PATFETCH
8392
#undef PATFETCH_RAW
8393
#undef PATUNFETCH
8394
#undef TRANSLATE
8395
8396
#undef INIT_BUF_SIZE
8397
#undef GET_BUFFER_SPACE
8398
#undef BUF_PUSH
8399
#undef BUF_PUSH_2
8400
#undef BUF_PUSH_3
8401
#undef STORE_JUMP
8402
#undef STORE_JUMP2
8403
#undef INSERT_JUMP
8404
#undef INSERT_JUMP2
8405
#undef EXTEND_BUFFER
8406
#undef GET_UNSIGNED_NUMBER
8407
#undef FREE_STACK_RETURN
8408
8409
# undef POINTER_TO_OFFSET
8410
# undef MATCHING_IN_FRST_STRING
8411
# undef PREFETCH
8412
# undef AT_STRINGS_BEG
8413
# undef AT_STRINGS_END
8414
# undef WORDCHAR_P
8415
# undef FREE_VAR
8416
# undef FREE_VARIABLES
8417
# undef NO_HIGHEST_ACTIVE_REG
8418
# undef NO_LOWEST_ACTIVE_REG
8419
8420
# undef CHAR_T
8421
# undef UCHAR_T
8422
# undef COMPILED_BUFFER_VAR
8423
# undef OFFSET_ADDRESS_SIZE
8424
# undef CHAR_CLASS_SIZE
8425
# undef PREFIX
8426
# undef ARG_PREFIX
8427
# undef PUT_CHAR
8428
# undef BYTE
8429
# undef WCHAR
8430
8431
# define DEFINED_ONCE
Loggerhead is a web-based interface for Breezy
Version: 2.0.1