1
/* Extended regular expression matching and search library,
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
This file is part of the GNU C Library.
8
The GNU C Library is free software; you can redistribute it and/or
9
modify it under the terms of the GNU Lesser General Public
10
License as published by the Free Software Foundation; either
11
version 2.1 of the License, or (at your option) any later version.
13
The GNU C Library is distributed in the hope that it will be useful,
14
but WITHOUT ANY WARRANTY; without even the implied warranty of
15
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
16
Lesser General Public License for more details.
18
You should have received a copy of the GNU Lesser General Public
19
License along with the GNU C Library; if not, write to the Free
20
Software Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA
23
/* This file has been modified for usage in libiberty. It includes "xregex.h"
24
instead of <regex.h>. The "xregex.h" header file renames all external
25
routines with an "x" prefix so they do not collide with the native regex
26
routines or with other components regex routines. */
27
/* AIX requires this to be the first thing in the file. */
28
#if defined _AIX && !defined REGEX_MALLOC
40
# if defined __GNUC__ || (defined __STDC__ && __STDC__)
41
# define PARAMS(args) args
43
# define PARAMS(args) ()
45
#endif /* Not PARAMS. */
47
#ifndef INSIDE_RECURSION
49
# if defined STDC_HEADERS && !defined emacs
52
/* We need this for `regex.h', and perhaps for the Emacs include files. */
53
# include <sys/types.h>
56
# define WIDE_CHAR_SUPPORT (HAVE_WCTYPE_H && HAVE_WCHAR_H && HAVE_BTOWC)
58
/* For platform which support the ISO C amendement 1 functionality we
59
support user defined character classes. */
60
# if defined _LIBC || WIDE_CHAR_SUPPORT
61
/* Solaris 2.5 has a bug: <wchar.h> must be included before <wctype.h>. */
67
/* We have to keep the namespace clean. */
68
# define regfree(preg) __regfree (preg)
69
# define regexec(pr, st, nm, pm, ef) __regexec (pr, st, nm, pm, ef)
70
# define regcomp(preg, pattern, cflags) __regcomp (preg, pattern, cflags)
71
# define regerror(errcode, preg, errbuf, errbuf_size) \
72
__regerror(errcode, preg, errbuf, errbuf_size)
73
# define re_set_registers(bu, re, nu, st, en) \
74
__re_set_registers (bu, re, nu, st, en)
75
# define re_match_2(bufp, string1, size1, string2, size2, pos, regs, stop) \
76
__re_match_2 (bufp, string1, size1, string2, size2, pos, regs, stop)
77
# define re_match(bufp, string, size, pos, regs) \
78
__re_match (bufp, string, size, pos, regs)
79
# define re_search(bufp, string, size, startpos, range, regs) \
80
__re_search (bufp, string, size, startpos, range, regs)
81
# define re_compile_pattern(pattern, length, bufp) \
82
__re_compile_pattern (pattern, length, bufp)
83
# define re_set_syntax(syntax) __re_set_syntax (syntax)
84
# define re_search_2(bufp, st1, s1, st2, s2, startpos, range, regs, stop) \
85
__re_search_2 (bufp, st1, s1, st2, s2, startpos, range, regs, stop)
86
# define re_compile_fastmap(bufp) __re_compile_fastmap (bufp)
88
# define btowc __btowc
90
/* We are also using some library internals. */
91
# include <locale/localeinfo.h>
92
# include <locale/elem-hash.h>
93
# include <langinfo.h>
94
# include <locale/coll-lookup.h>
97
/* This is for other GNU distributions with internationalized messages. */
98
# if (HAVE_LIBINTL_H && ENABLE_NLS) || defined _LIBC
102
# define gettext(msgid) __dcgettext ("libc", msgid, LC_MESSAGES)
105
# define gettext(msgid) (msgid)
108
# ifndef gettext_noop
109
/* This define is so xgettext can find the internationalizable
111
# define gettext_noop(String) String
114
/* The `emacs' switch turns on certain matching commands
115
that make sense only in Emacs. */
122
# else /* not emacs */
124
/* If we are not linking with Emacs proper,
125
we can't use the relocating allocator
126
even if config.h says that we can. */
129
# if defined STDC_HEADERS || defined _LIBC
136
/* When used in Emacs's lib-src, we need to get bzero and bcopy somehow.
137
If nothing else has been done, use the method below. */
138
# ifdef INHIBIT_STRING_HEADER
139
# if !(defined HAVE_BZERO && defined HAVE_BCOPY)
140
# if !defined bzero && !defined bcopy
141
# undef INHIBIT_STRING_HEADER
146
/* This is the normal way of making sure we have a bcopy and a bzero.
147
This is used in most programs--a few other programs avoid this
148
by defining INHIBIT_STRING_HEADER. */
149
# ifndef INHIBIT_STRING_HEADER
150
# if defined HAVE_STRING_H || defined STDC_HEADERS || defined _LIBC
154
# define bzero(s, n) (memset (s, '\0', n), (s))
156
# define bzero(s, n) __bzero (s, n)
160
# include <strings.h>
162
# define memcmp(s1, s2, n) bcmp (s1, s2, n)
165
# define memcpy(d, s, n) (bcopy (s, d, n), (d))
170
/* Define the syntax stuff for \<, \>, etc. */
172
/* This must be nonzero for the wordchar and notwordchar pattern
173
commands in re_match_2. */
178
# ifdef SWITCH_ENUM_BUG
179
# define SWITCH_ENUM_CAST(x) ((int)(x))
181
# define SWITCH_ENUM_CAST(x) (x)
184
# endif /* not emacs */
186
# if defined _LIBC || HAVE_LIMITS_H
191
# define MB_LEN_MAX 1
194
/* Get the interface, including the syntax bits. */
195
# include "xregex.h" /* change for libiberty */
197
/* isalpha etc. are used for the character classes. */
200
/* Jim Meyering writes:
202
"... Some ctype macros are valid only for character codes that
203
isascii says are ASCII (SGI's IRIX-4.0.5 is one such system --when
204
using /bin/cc or gcc but without giving an ansi option). So, all
205
ctype uses should be through macros like ISPRINT... If
206
STDC_HEADERS is defined, then autoconf has verified that the ctype
207
macros don't need to be guarded with references to isascii. ...
208
Defining isascii to 1 should let any compiler worth its salt
209
eliminate the && through constant folding."
210
Solaris defines some of these symbols so we must undefine them first. */
213
# if defined STDC_HEADERS || (!defined isascii && !defined HAVE_ISASCII)
214
# define ISASCII(c) 1
216
# define ISASCII(c) isascii(c)
220
# define ISBLANK(c) (ISASCII (c) && isblank (c))
222
# define ISBLANK(c) ((c) == ' ' || (c) == '\t')
225
# define ISGRAPH(c) (ISASCII (c) && isgraph (c))
227
# define ISGRAPH(c) (ISASCII (c) && isprint (c) && !isspace (c))
231
# define ISPRINT(c) (ISASCII (c) && isprint (c))
232
# define ISDIGIT(c) (ISASCII (c) && isdigit (c))
233
# define ISALNUM(c) (ISASCII (c) && isalnum (c))
234
# define ISALPHA(c) (ISASCII (c) && isalpha (c))
235
# define ISCNTRL(c) (ISASCII (c) && iscntrl (c))
236
# define ISLOWER(c) (ISASCII (c) && islower (c))
237
# define ISPUNCT(c) (ISASCII (c) && ispunct (c))
238
# define ISSPACE(c) (ISASCII (c) && isspace (c))
239
# define ISUPPER(c) (ISASCII (c) && isupper (c))
240
# define ISXDIGIT(c) (ISASCII (c) && isxdigit (c))
243
# define TOLOWER(c) _tolower(c)
245
# define TOLOWER(c) tolower(c)
249
# define NULL (void *)0
252
/* We remove any previous definition of `SIGN_EXTEND_CHAR',
253
since ours (we hope) works properly with all combinations of
254
machines, compilers, `char' and `unsigned char' argument types.
255
(Per Bothner suggested the basic approach.) */
256
# undef SIGN_EXTEND_CHAR
258
# define SIGN_EXTEND_CHAR(c) ((signed char) (c))
259
# else /* not __STDC__ */
260
/* As in Harbison and Steele. */
261
# define SIGN_EXTEND_CHAR(c) ((((unsigned char) (c)) ^ 128) - 128)
265
/* How many characters in the character set. */
266
# define CHAR_SET_SIZE 256
270
extern char *re_syntax_table;
272
# else /* not SYNTAX_TABLE */
274
static char re_syntax_table[CHAR_SET_SIZE];
276
static void init_syntax_once PARAMS ((void));
286
bzero (re_syntax_table, sizeof re_syntax_table);
288
for (c = 0; c < CHAR_SET_SIZE; ++c)
290
re_syntax_table[c] = Sword;
292
re_syntax_table['_'] = Sword;
297
# endif /* not SYNTAX_TABLE */
299
# define SYNTAX(c) re_syntax_table[(unsigned char) (c)]
303
/* Integer type for pointers. */
304
# if !defined _LIBC && !defined HAVE_UINTPTR_T
305
typedef unsigned long int uintptr_t;
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.
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. */
320
# define REGEX_ALLOCATE malloc
321
# define REGEX_REALLOCATE(source, osize, nsize) realloc (source, nsize)
322
# define REGEX_FREE free
324
# else /* not REGEX_MALLOC */
326
/* Emacs already defines alloca, sometimes. */
329
/* Make alloca work the best possible way. */
331
# define alloca __builtin_alloca
332
# else /* not __GNUC__ */
335
# endif /* HAVE_ALLOCA_H */
336
# endif /* not __GNUC__ */
338
# endif /* not alloca */
340
# define REGEX_ALLOCATE alloca
342
/* Assumes a `char *destination' variable. */
343
# define REGEX_REALLOCATE(source, osize, nsize) \
344
(destination = (char *) alloca (nsize), \
345
memcpy (destination, source, osize))
347
/* No need to do anything to free, after alloca. */
348
# define REGEX_FREE(arg) ((void)0) /* Do nothing! But inhibit gcc warning. */
350
# endif /* not REGEX_MALLOC */
352
/* Define how to allocate the failure stack. */
354
# if defined REL_ALLOC && defined REGEX_MALLOC
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)
363
# else /* not using relocating allocator */
367
# define REGEX_ALLOCATE_STACK malloc
368
# define REGEX_REALLOCATE_STACK(source, osize, nsize) realloc (source, nsize)
369
# define REGEX_FREE_STACK free
371
# else /* not REGEX_MALLOC */
373
# define REGEX_ALLOCATE_STACK alloca
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)
380
# endif /* not REGEX_MALLOC */
381
# endif /* not using relocating allocator */
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
387
# define FIRST_STRING_P(ptr) \
388
(size1 && string1 <= (ptr) && (ptr) <= string1 + size1)
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)))
397
# define BYTEWIDTH 8 /* In bits. */
399
# define STREQ(s1, s2) ((strcmp (s1, s2) == 0))
403
# define MAX(a, b) ((a) > (b) ? (a) : (b))
404
# define MIN(a, b) ((a) < (b) ? (a) : (b))
406
typedef char boolean;
410
static reg_errcode_t byte_regex_compile _RE_ARGS ((const char *pattern, size_t size,
412
struct re_pattern_buffer *bufp));
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,
418
struct re_registers *regs,
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));
428
static reg_errcode_t wcs_regex_compile _RE_ARGS ((const char *pattern, size_t size,
430
struct re_pattern_buffer *bufp));
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,
437
struct re_registers *regs,
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));
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. */
459
/* Succeed right away--no more backtracking. */
462
/* Followed by one byte giving n, then by n literal bytes. */
466
/* Same as exactn, but contains binary data. */
470
/* Matches any (more or less) character. */
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
485
See regex_compile function. */
488
/* Same parameters as charset, but match any character that is
489
not one of those specified. */
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
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.) */
510
/* Match a duplicate of something remembered. Followed by one
511
byte containing the register number. */
514
/* Fail unless at beginning of line. */
517
/* Fail unless at end of line. */
520
/* Succeeds if at beginning of buffer (if emacs) or at beginning
521
of string to be matched (if not). */
524
/* Analogously, for end of buffer/string. */
527
/* Followed by two byte relative address to which to jump. */
530
/* Same as jump, but marks the end of an alternative. */
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. */
538
/* Like on_failure_jump, but pushes a placeholder instead of the
539
current string position when executed. */
540
on_failure_keep_string_jump,
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. */
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. */
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. */
565
/* Push a dummy failure point and continue. Used at the end of
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. */
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. */
579
/* Set the following two-byte relative address to the
580
subsequent two-byte number. The address *includes* the two
582
/* ifdef MBS_SUPPORT, the size of address is 1. */
585
wordchar, /* Matches any word-constituent character. */
586
notwordchar, /* Matches any char that is not a word-constituent. */
588
wordbeg, /* Succeeds if at word beginning. */
589
wordend, /* Succeeds if at word end. */
591
wordbound, /* Succeeds if at a word boundary. */
592
notwordbound /* Succeeds if not at a word boundary. */
595
,before_dot, /* Succeeds if before point. */
596
at_dot, /* Succeeds if at point. */
597
after_dot, /* Succeeds if after point. */
599
/* Matches any character whose syntax is specified. Followed by
600
a byte which contains a syntax code, e.g., Sword. */
603
/* Matches any character whose syntax is not that specified. */
607
#endif /* not INSIDE_RECURSION */
612
# define UCHAR_T unsigned char
613
# define COMPILED_BUFFER_VAR bufp->buffer
614
# define OFFSET_ADDRESS_SIZE 2
615
# if defined (__STDC__) || defined (ALMOST_STDC) || defined (HAVE_STRINGIZE)
616
# define PREFIX(name) byte_##name
618
# define PREFIX(name) byte_/**/name
620
# define ARG_PREFIX(name) name
621
# define PUT_CHAR(c) putchar (c)
624
# define CHAR_T wchar_t
625
# define UCHAR_T wchar_t
626
# define COMPILED_BUFFER_VAR wc_buffer
627
# define OFFSET_ADDRESS_SIZE 1 /* the size which STORE_NUMBER macro use */
628
# define CHAR_CLASS_SIZE ((__alignof__(wctype_t)+sizeof(wctype_t))/sizeof(CHAR_T)+1)
629
# if defined (__STDC__) || defined (ALMOST_STDC) || defined (HAVE_STRINGIZE)
630
# define PREFIX(name) wcs_##name
631
# define ARG_PREFIX(name) c##name
633
# define PREFIX(name) wcs_/**/name
634
# define ARG_PREFIX(name) c/**/name
636
/* Should we use wide stream?? */
637
# define PUT_CHAR(c) printf ("%C", c);
643
# define INSIDE_RECURSION
645
# undef INSIDE_RECURSION
648
# define INSIDE_RECURSION
650
# undef INSIDE_RECURSION
654
#ifdef INSIDE_RECURSION
655
/* Common operations on the compiled pattern. */
657
/* Store NUMBER in two contiguous bytes starting at DESTINATION. */
658
/* ifdef MBS_SUPPORT, we store NUMBER in 1 element. */
661
# define STORE_NUMBER(destination, number) \
663
*(destination) = (UCHAR_T)(number); \
666
# define STORE_NUMBER(destination, number) \
668
(destination)[0] = (number) & 0377; \
669
(destination)[1] = (number) >> 8; \
673
/* Same as STORE_NUMBER, except increment DESTINATION to
674
the byte after where the number is stored. Therefore, DESTINATION
675
must be an lvalue. */
676
/* ifdef MBS_SUPPORT, we store NUMBER in 1 element. */
678
# define STORE_NUMBER_AND_INCR(destination, number) \
680
STORE_NUMBER (destination, number); \
681
(destination) += OFFSET_ADDRESS_SIZE; \
684
/* Put into DESTINATION a number stored in two contiguous bytes starting
686
/* ifdef MBS_SUPPORT, we store NUMBER in 1 element. */
689
# define EXTRACT_NUMBER(destination, source) \
691
(destination) = *(source); \
694
# define EXTRACT_NUMBER(destination, source) \
696
(destination) = *(source) & 0377; \
697
(destination) += SIGN_EXTEND_CHAR (*((source) + 1)) << 8; \
702
static void PREFIX(extract_number) _RE_ARGS ((int *dest, UCHAR_T *source));
704
PREFIX(extract_number) (dest, source)
711
int temp = SIGN_EXTEND_CHAR (*(source + 1));
712
*dest = *source & 0377;
717
# ifndef EXTRACT_MACROS /* To debug the macros. */
718
# undef EXTRACT_NUMBER
719
# define EXTRACT_NUMBER(dest, src) PREFIX(extract_number) (&dest, src)
720
# endif /* not EXTRACT_MACROS */
724
/* Same as EXTRACT_NUMBER, except increment SOURCE to after the number.
725
SOURCE must be an lvalue. */
727
# define EXTRACT_NUMBER_AND_INCR(destination, source) \
729
EXTRACT_NUMBER (destination, source); \
730
(source) += OFFSET_ADDRESS_SIZE; \
734
static void PREFIX(extract_number_and_incr) _RE_ARGS ((int *destination,
737
PREFIX(extract_number_and_incr) (destination, source)
741
PREFIX(extract_number) (destination, *source);
742
*source += OFFSET_ADDRESS_SIZE;
745
# ifndef EXTRACT_MACROS
746
# undef EXTRACT_NUMBER_AND_INCR
747
# define EXTRACT_NUMBER_AND_INCR(dest, src) \
748
PREFIX(extract_number_and_incr) (&dest, &src)
749
# endif /* not EXTRACT_MACROS */
755
/* If DEBUG is defined, Regex prints many voluminous messages about what
756
it is doing (if the variable `debug' is nonzero). If linked with the
757
main program in `iregex.c', you can enter patterns and strings
758
interactively. And if linked with the main program in `main.c' and
759
the other test files, you can run the already-written tests. */
763
# ifndef DEFINED_ONCE
765
/* We use standard I/O for debugging. */
768
/* It is useful to test things that ``must'' be true when debugging. */
773
# define DEBUG_STATEMENT(e) e
774
# define DEBUG_PRINT1(x) if (debug) printf (x)
775
# define DEBUG_PRINT2(x1, x2) if (debug) printf (x1, x2)
776
# define DEBUG_PRINT3(x1, x2, x3) if (debug) printf (x1, x2, x3)
777
# define DEBUG_PRINT4(x1, x2, x3, x4) if (debug) printf (x1, x2, x3, x4)
778
# endif /* not DEFINED_ONCE */
780
# define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) \
781
if (debug) PREFIX(print_partial_compiled_pattern) (s, e)
782
# define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) \
783
if (debug) PREFIX(print_double_string) (w, s1, sz1, s2, sz2)
786
/* Print the fastmap in human-readable form. */
788
# ifndef DEFINED_ONCE
790
print_fastmap (fastmap)
793
unsigned was_a_range = 0;
796
while (i < (1 << BYTEWIDTH))
802
while (i < (1 << BYTEWIDTH) && fastmap[i])
816
# endif /* not DEFINED_ONCE */
819
/* Print a compiled pattern string in human-readable form, starting at
820
the START pointer into it and ending just before the pointer END. */
823
PREFIX(print_partial_compiled_pattern) (start, end)
838
/* Loop over pattern commands. */
842
printf ("%td:\t", p - start);
844
printf ("%ld:\t", (long int) (p - start));
847
switch ((re_opcode_t) *p++)
855
printf ("/exactn/%d", mcnt);
867
printf ("/exactn_bin/%d", mcnt);
870
printf("/%lx", (long int) *p++);
874
# endif /* MBS_SUPPORT */
878
printf ("/start_memory/%d/%ld", mcnt, (long int) *p++);
883
printf ("/stop_memory/%d/%ld", mcnt, (long int) *p++);
887
printf ("/duplicate/%ld", (long int) *p++);
900
printf ("/charset [%s",
901
(re_opcode_t) *(workp - 1) == charset_not ? "^" : "");
903
length = *workp++; /* the length of char_classes */
904
for (i=0 ; i<length ; i++)
905
printf("[:%lx:]", (long int) *p++);
906
length = *workp++; /* the length of collating_symbol */
907
for (i=0 ; i<length ;)
911
PUT_CHAR((i++,*p++));
915
length = *workp++; /* the length of equivalence_class */
916
for (i=0 ; i<length ;)
920
PUT_CHAR((i++,*p++));
924
length = *workp++; /* the length of char_range */
925
for (i=0 ; i<length ; i++)
927
wchar_t range_start = *p++;
928
wchar_t range_end = *p++;
929
printf("%C-%C", range_start, range_end);
931
length = *workp++; /* the length of char */
932
for (i=0 ; i<length ; i++)
936
register int c, last = -100;
937
register int in_range = 0;
939
printf ("/charset [%s",
940
(re_opcode_t) *(p - 1) == charset_not ? "^" : "");
942
assert (p + *p < pend);
944
for (c = 0; c < 256; c++)
946
&& (p[1 + (c/8)] & (1 << (c % 8))))
948
/* Are we starting a range? */
949
if (last + 1 == c && ! in_range)
954
/* Have we broken a range? */
955
else if (last + 1 != c && in_range)
985
case on_failure_jump:
986
PREFIX(extract_number_and_incr) (&mcnt, &p);
988
printf ("/on_failure_jump to %td", p + mcnt - start);
990
printf ("/on_failure_jump to %ld", (long int) (p + mcnt - start));
994
case on_failure_keep_string_jump:
995
PREFIX(extract_number_and_incr) (&mcnt, &p);
997
printf ("/on_failure_keep_string_jump to %td", p + mcnt - start);
999
printf ("/on_failure_keep_string_jump to %ld",
1000
(long int) (p + mcnt - start));
1004
case dummy_failure_jump:
1005
PREFIX(extract_number_and_incr) (&mcnt, &p);
1007
printf ("/dummy_failure_jump to %td", p + mcnt - start);
1009
printf ("/dummy_failure_jump to %ld", (long int) (p + mcnt - start));
1013
case push_dummy_failure:
1014
printf ("/push_dummy_failure");
1017
case maybe_pop_jump:
1018
PREFIX(extract_number_and_incr) (&mcnt, &p);
1020
printf ("/maybe_pop_jump to %td", p + mcnt - start);
1022
printf ("/maybe_pop_jump to %ld", (long int) (p + mcnt - start));
1026
case pop_failure_jump:
1027
PREFIX(extract_number_and_incr) (&mcnt, &p);
1029
printf ("/pop_failure_jump to %td", p + mcnt - start);
1031
printf ("/pop_failure_jump to %ld", (long int) (p + mcnt - start));
1036
PREFIX(extract_number_and_incr) (&mcnt, &p);
1038
printf ("/jump_past_alt to %td", p + mcnt - start);
1040
printf ("/jump_past_alt to %ld", (long int) (p + mcnt - start));
1045
PREFIX(extract_number_and_incr) (&mcnt, &p);
1047
printf ("/jump to %td", p + mcnt - start);
1049
printf ("/jump to %ld", (long int) (p + mcnt - start));
1054
PREFIX(extract_number_and_incr) (&mcnt, &p);
1056
PREFIX(extract_number_and_incr) (&mcnt2, &p);
1058
printf ("/succeed_n to %td, %d times", p1 - start, mcnt2);
1060
printf ("/succeed_n to %ld, %d times",
1061
(long int) (p1 - start), mcnt2);
1066
PREFIX(extract_number_and_incr) (&mcnt, &p);
1068
PREFIX(extract_number_and_incr) (&mcnt2, &p);
1069
printf ("/jump_n to %d, %d times", p1 - start, mcnt2);
1073
PREFIX(extract_number_and_incr) (&mcnt, &p);
1075
PREFIX(extract_number_and_incr) (&mcnt2, &p);
1077
printf ("/set_number_at location %td to %d", p1 - start, mcnt2);
1079
printf ("/set_number_at location %ld to %d",
1080
(long int) (p1 - start), mcnt2);
1085
printf ("/wordbound");
1089
printf ("/notwordbound");
1093
printf ("/wordbeg");
1097
printf ("/wordend");
1102
printf ("/before_dot");
1110
printf ("/after_dot");
1114
printf ("/syntaxspec");
1116
printf ("/%d", mcnt);
1120
printf ("/notsyntaxspec");
1122
printf ("/%d", mcnt);
1127
printf ("/wordchar");
1131
printf ("/notwordchar");
1143
printf ("?%ld", (long int) *(p-1));
1150
printf ("%td:\tend of pattern.\n", p - start);
1152
printf ("%ld:\tend of pattern.\n", (long int) (p - start));
1158
PREFIX(print_compiled_pattern) (bufp)
1159
struct re_pattern_buffer *bufp;
1161
UCHAR_T *buffer = (UCHAR_T*) bufp->buffer;
1163
PREFIX(print_partial_compiled_pattern) (buffer, buffer
1164
+ bufp->used / sizeof(UCHAR_T));
1165
printf ("%ld bytes used/%ld bytes allocated.\n",
1166
bufp->used, bufp->allocated);
1168
if (bufp->fastmap_accurate && bufp->fastmap)
1170
printf ("fastmap: ");
1171
print_fastmap (bufp->fastmap);
1175
printf ("re_nsub: %Zd\t", bufp->re_nsub);
1177
printf ("re_nsub: %ld\t", (long int) bufp->re_nsub);
1179
printf ("regs_alloc: %d\t", bufp->regs_allocated);
1180
printf ("can_be_null: %d\t", bufp->can_be_null);
1181
printf ("newline_anchor: %d\n", bufp->newline_anchor);
1182
printf ("no_sub: %d\t", bufp->no_sub);
1183
printf ("not_bol: %d\t", bufp->not_bol);
1184
printf ("not_eol: %d\t", bufp->not_eol);
1185
printf ("syntax: %lx\n", bufp->syntax);
1186
/* Perhaps we should print the translate table? */
1191
PREFIX(print_double_string) (where, string1, size1, string2, size2)
1192
const CHAR_T *where;
1193
const CHAR_T *string1;
1194
const CHAR_T *string2;
1206
if (FIRST_STRING_P (where))
1208
for (this_char = where - string1; this_char < size1; this_char++)
1209
PUT_CHAR (string1[this_char]);
1215
for (this_char = where - string2; this_char < size2; this_char++)
1217
PUT_CHAR (string2[this_char]);
1220
fputs ("...", stdout);
1227
# ifndef DEFINED_ONCE
1236
# else /* not DEBUG */
1238
# ifndef DEFINED_ONCE
1242
# define DEBUG_STATEMENT(e)
1243
# define DEBUG_PRINT1(x)
1244
# define DEBUG_PRINT2(x1, x2)
1245
# define DEBUG_PRINT3(x1, x2, x3)
1246
# define DEBUG_PRINT4(x1, x2, x3, x4)
1247
# endif /* not DEFINED_ONCE */
1248
# define DEBUG_PRINT_COMPILED_PATTERN(p, s, e)
1249
# define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2)
1251
# endif /* not DEBUG */
1256
/* This convert a multibyte string to a wide character string.
1257
And write their correspondances to offset_buffer(see below)
1258
and write whether each wchar_t is binary data to is_binary.
1259
This assume invalid multibyte sequences as binary data.
1260
We assume offset_buffer and is_binary is already allocated
1263
static size_t convert_mbs_to_wcs (CHAR_T *dest, const unsigned char* src,
1264
size_t len, int *offset_buffer,
1267
convert_mbs_to_wcs (dest, src, len, offset_buffer, is_binary)
1269
const unsigned char* src;
1270
size_t len; /* the length of multibyte string. */
1272
/* It hold correspondances between src(char string) and
1273
dest(wchar_t string) for optimization.
1275
dest = {'X', 'Y', 'Z'}
1276
(each "xxx", "y" and "zz" represent one multibyte character
1277
corresponding to 'X', 'Y' and 'Z'.)
1278
offset_buffer = {0, 0+3("xxx"), 0+3+1("y"), 0+3+1+2("zz")}
1284
wchar_t *pdest = dest;
1285
const unsigned char *psrc = src;
1286
size_t wc_count = 0;
1290
size_t mb_remain = len;
1291
size_t mb_count = 0;
1293
/* Initialize the conversion state. */
1294
memset (&mbs, 0, sizeof (mbstate_t));
1296
offset_buffer[0] = 0;
1297
for( ; mb_remain > 0 ; ++wc_count, ++pdest, mb_remain -= consumed,
1301
consumed = __mbrtowc (pdest, psrc, mb_remain, &mbs);
1303
consumed = mbrtowc (pdest, psrc, mb_remain, &mbs);
1307
/* failed to convert. maybe src contains binary data.
1308
So we consume 1 byte manualy. */
1312
is_binary[wc_count] = TRUE;
1315
is_binary[wc_count] = FALSE;
1316
/* In sjis encoding, we use yen sign as escape character in
1317
place of reverse solidus. So we convert 0x5c(yen sign in
1318
sjis) to not 0xa5(yen sign in UCS2) but 0x5c(reverse
1319
solidus in UCS2). */
1320
if (consumed == 1 && (int) *psrc == 0x5c && (int) *pdest == 0xa5)
1321
*pdest = (wchar_t) *psrc;
1323
offset_buffer[wc_count + 1] = mb_count += consumed;
1326
/* Fill remain of the buffer with sentinel. */
1327
for (i = wc_count + 1 ; i <= len ; i++)
1328
offset_buffer[i] = mb_count + 1;
1335
#else /* not INSIDE_RECURSION */
1337
/* Set by `re_set_syntax' to the current regexp syntax to recognize. Can
1338
also be assigned to arbitrarily: each pattern buffer stores its own
1339
syntax, so it can be changed between regex compilations. */
1340
/* This has no initializer because initialized variables in Emacs
1341
become read-only after dumping. */
1342
reg_syntax_t re_syntax_options;
1345
/* Specify the precise syntax of regexps for compilation. This provides
1346
for compatibility for various utilities which historically have
1347
different, incompatible syntaxes.
1349
The argument SYNTAX is a bit mask comprised of the various bits
1350
defined in regex.h. We return the old syntax. */
1353
re_set_syntax (syntax)
1354
reg_syntax_t syntax;
1356
reg_syntax_t ret = re_syntax_options;
1358
re_syntax_options = syntax;
1360
if (syntax & RE_DEBUG)
1362
else if (debug) /* was on but now is not */
1368
weak_alias (__re_set_syntax, re_set_syntax)
1371
/* This table gives an error message for each of the error codes listed
1372
in regex.h. Obviously the order here has to be same as there.
1373
POSIX doesn't require that we do anything for REG_NOERROR,
1374
but why not be nice? */
1376
static const char re_error_msgid[] =
1378
# define REG_NOERROR_IDX 0
1379
gettext_noop ("Success") /* REG_NOERROR */
1381
# define REG_NOMATCH_IDX (REG_NOERROR_IDX + sizeof "Success")
1382
gettext_noop ("No match") /* REG_NOMATCH */
1384
# define REG_BADPAT_IDX (REG_NOMATCH_IDX + sizeof "No match")
1385
gettext_noop ("Invalid regular expression") /* REG_BADPAT */
1387
# define REG_ECOLLATE_IDX (REG_BADPAT_IDX + sizeof "Invalid regular expression")
1388
gettext_noop ("Invalid collation character") /* REG_ECOLLATE */
1390
# define REG_ECTYPE_IDX (REG_ECOLLATE_IDX + sizeof "Invalid collation character")
1391
gettext_noop ("Invalid character class name") /* REG_ECTYPE */
1393
# define REG_EESCAPE_IDX (REG_ECTYPE_IDX + sizeof "Invalid character class name")
1394
gettext_noop ("Trailing backslash") /* REG_EESCAPE */
1396
# define REG_ESUBREG_IDX (REG_EESCAPE_IDX + sizeof "Trailing backslash")
1397
gettext_noop ("Invalid back reference") /* REG_ESUBREG */
1399
# define REG_EBRACK_IDX (REG_ESUBREG_IDX + sizeof "Invalid back reference")
1400
gettext_noop ("Unmatched [ or [^") /* REG_EBRACK */
1402
# define REG_EPAREN_IDX (REG_EBRACK_IDX + sizeof "Unmatched [ or [^")
1403
gettext_noop ("Unmatched ( or \\(") /* REG_EPAREN */
1405
# define REG_EBRACE_IDX (REG_EPAREN_IDX + sizeof "Unmatched ( or \\(")
1406
gettext_noop ("Unmatched \\{") /* REG_EBRACE */
1408
# define REG_BADBR_IDX (REG_EBRACE_IDX + sizeof "Unmatched \\{")
1409
gettext_noop ("Invalid content of \\{\\}") /* REG_BADBR */
1411
# define REG_ERANGE_IDX (REG_BADBR_IDX + sizeof "Invalid content of \\{\\}")
1412
gettext_noop ("Invalid range end") /* REG_ERANGE */
1414
# define REG_ESPACE_IDX (REG_ERANGE_IDX + sizeof "Invalid range end")
1415
gettext_noop ("Memory exhausted") /* REG_ESPACE */
1417
# define REG_BADRPT_IDX (REG_ESPACE_IDX + sizeof "Memory exhausted")
1418
gettext_noop ("Invalid preceding regular expression") /* REG_BADRPT */
1420
# define REG_EEND_IDX (REG_BADRPT_IDX + sizeof "Invalid preceding regular expression")
1421
gettext_noop ("Premature end of regular expression") /* REG_EEND */
1423
# define REG_ESIZE_IDX (REG_EEND_IDX + sizeof "Premature end of regular expression")
1424
gettext_noop ("Regular expression too big") /* REG_ESIZE */
1426
# define REG_ERPAREN_IDX (REG_ESIZE_IDX + sizeof "Regular expression too big")
1427
gettext_noop ("Unmatched ) or \\)") /* REG_ERPAREN */
1430
static const size_t re_error_msgid_idx[] =
1451
#endif /* INSIDE_RECURSION */
1453
#ifndef DEFINED_ONCE
1454
/* Avoiding alloca during matching, to placate r_alloc. */
1456
/* Define MATCH_MAY_ALLOCATE unless we need to make sure that the
1457
searching and matching functions should not call alloca. On some
1458
systems, alloca is implemented in terms of malloc, and if we're
1459
using the relocating allocator routines, then malloc could cause a
1460
relocation, which might (if the strings being searched are in the
1461
ralloc heap) shift the data out from underneath the regexp
1464
Here's another reason to avoid allocation: Emacs
1465
processes input from X in a signal handler; processing X input may
1466
call malloc; if input arrives while a matching routine is calling
1467
malloc, then we're scrod. But Emacs can't just block input while
1468
calling matching routines; then we don't notice interrupts when
1469
they come in. So, Emacs blocks input around all regexp calls
1470
except the matching calls, which it leaves unprotected, in the
1471
faith that they will not malloc. */
1473
/* Normally, this is fine. */
1474
# define MATCH_MAY_ALLOCATE
1476
/* When using GNU C, we are not REALLY using the C alloca, no matter
1477
what config.h may say. So don't take precautions for it. */
1482
/* The match routines may not allocate if (1) they would do it with malloc
1483
and (2) it's not safe for them to use malloc.
1484
Note that if REL_ALLOC is defined, matching would not use malloc for the
1485
failure stack, but we would still use it for the register vectors;
1486
so REL_ALLOC should not affect this. */
1487
# if (defined C_ALLOCA || defined REGEX_MALLOC) && defined emacs
1488
# undef MATCH_MAY_ALLOCATE
1490
#endif /* not DEFINED_ONCE */
1492
#ifdef INSIDE_RECURSION
1493
/* Failure stack declarations and macros; both re_compile_fastmap and
1494
re_match_2 use a failure stack. These have to be macros because of
1495
REGEX_ALLOCATE_STACK. */
1498
/* Number of failure points for which to initially allocate space
1499
when matching. If this number is exceeded, we allocate more
1500
space, so it is not a hard limit. */
1501
# ifndef INIT_FAILURE_ALLOC
1502
# define INIT_FAILURE_ALLOC 5
1505
/* Roughly the maximum number of failure points on the stack. Would be
1506
exactly that if always used MAX_FAILURE_ITEMS items each time we failed.
1507
This is a variable only so users of regex can assign to it; we never
1508
change it ourselves. */
1510
# ifdef INT_IS_16BIT
1512
# ifndef DEFINED_ONCE
1513
# if defined MATCH_MAY_ALLOCATE
1514
/* 4400 was enough to cause a crash on Alpha OSF/1,
1515
whose default stack limit is 2mb. */
1516
long int re_max_failures = 4000;
1518
long int re_max_failures = 2000;
1522
union PREFIX(fail_stack_elt)
1528
typedef union PREFIX(fail_stack_elt) PREFIX(fail_stack_elt_t);
1532
PREFIX(fail_stack_elt_t) *stack;
1533
unsigned long int size;
1534
unsigned long int avail; /* Offset of next open position. */
1535
} PREFIX(fail_stack_type);
1537
# else /* not INT_IS_16BIT */
1539
# ifndef DEFINED_ONCE
1540
# if defined MATCH_MAY_ALLOCATE
1541
/* 4400 was enough to cause a crash on Alpha OSF/1,
1542
whose default stack limit is 2mb. */
1543
int re_max_failures = 4000;
1545
int re_max_failures = 2000;
1549
union PREFIX(fail_stack_elt)
1555
typedef union PREFIX(fail_stack_elt) PREFIX(fail_stack_elt_t);
1559
PREFIX(fail_stack_elt_t) *stack;
1561
unsigned avail; /* Offset of next open position. */
1562
} PREFIX(fail_stack_type);
1564
# endif /* INT_IS_16BIT */
1566
# ifndef DEFINED_ONCE
1567
# define FAIL_STACK_EMPTY() (fail_stack.avail == 0)
1568
# define FAIL_STACK_PTR_EMPTY() (fail_stack_ptr->avail == 0)
1569
# define FAIL_STACK_FULL() (fail_stack.avail == fail_stack.size)
1573
/* Define macros to initialize and free the failure stack.
1574
Do `return -2' if the alloc fails. */
1576
# ifdef MATCH_MAY_ALLOCATE
1577
# define INIT_FAIL_STACK() \
1579
fail_stack.stack = (PREFIX(fail_stack_elt_t) *) \
1580
REGEX_ALLOCATE_STACK (INIT_FAILURE_ALLOC * sizeof (PREFIX(fail_stack_elt_t))); \
1582
if (fail_stack.stack == NULL) \
1585
fail_stack.size = INIT_FAILURE_ALLOC; \
1586
fail_stack.avail = 0; \
1589
# define RESET_FAIL_STACK() REGEX_FREE_STACK (fail_stack.stack)
1591
# define INIT_FAIL_STACK() \
1593
fail_stack.avail = 0; \
1596
# define RESET_FAIL_STACK()
1600
/* Double the size of FAIL_STACK, up to approximately `re_max_failures' items.
1602
Return 1 if succeeds, and 0 if either ran out of memory
1603
allocating space for it or it was already too large.
1605
REGEX_REALLOCATE_STACK requires `destination' be declared. */
1607
# define DOUBLE_FAIL_STACK(fail_stack) \
1608
((fail_stack).size > (unsigned) (re_max_failures * MAX_FAILURE_ITEMS) \
1610
: ((fail_stack).stack = (PREFIX(fail_stack_elt_t) *) \
1611
REGEX_REALLOCATE_STACK ((fail_stack).stack, \
1612
(fail_stack).size * sizeof (PREFIX(fail_stack_elt_t)), \
1613
((fail_stack).size << 1) * sizeof (PREFIX(fail_stack_elt_t))),\
1615
(fail_stack).stack == NULL \
1617
: ((fail_stack).size <<= 1, \
1621
/* Push pointer POINTER on FAIL_STACK.
1622
Return 1 if was able to do so and 0 if ran out of memory allocating
1624
# define PUSH_PATTERN_OP(POINTER, FAIL_STACK) \
1625
((FAIL_STACK_FULL () \
1626
&& !DOUBLE_FAIL_STACK (FAIL_STACK)) \
1628
: ((FAIL_STACK).stack[(FAIL_STACK).avail++].pointer = POINTER, \
1631
/* Push a pointer 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_POINTER(item) \
1635
fail_stack.stack[fail_stack.avail++].pointer = (UCHAR_T *) (item)
1637
/* This pushes an integer-valued item onto the failure stack.
1638
Assumes the variable `fail_stack'. Probably should only
1639
be called from within `PUSH_FAILURE_POINT'. */
1640
# define PUSH_FAILURE_INT(item) \
1641
fail_stack.stack[fail_stack.avail++].integer = (item)
1643
/* Push a fail_stack_elt_t value onto the failure stack.
1644
Assumes the variable `fail_stack'. Probably should only
1645
be called from within `PUSH_FAILURE_POINT'. */
1646
# define PUSH_FAILURE_ELT(item) \
1647
fail_stack.stack[fail_stack.avail++] = (item)
1649
/* These three POP... operations complement the three PUSH... operations.
1650
All assume that `fail_stack' is nonempty. */
1651
# define POP_FAILURE_POINTER() fail_stack.stack[--fail_stack.avail].pointer
1652
# define POP_FAILURE_INT() fail_stack.stack[--fail_stack.avail].integer
1653
# define POP_FAILURE_ELT() fail_stack.stack[--fail_stack.avail]
1655
/* Used to omit pushing failure point id's when we're not debugging. */
1657
# define DEBUG_PUSH PUSH_FAILURE_INT
1658
# define DEBUG_POP(item_addr) *(item_addr) = POP_FAILURE_INT ()
1660
# define DEBUG_PUSH(item)
1661
# define DEBUG_POP(item_addr)
1665
/* Push the information about the state we will need
1666
if we ever fail back to it.
1668
Requires variables fail_stack, regstart, regend, reg_info, and
1669
num_regs_pushed be declared. DOUBLE_FAIL_STACK requires `destination'
1672
Does `return FAILURE_CODE' if runs out of memory. */
1674
# define PUSH_FAILURE_POINT(pattern_place, string_place, failure_code) \
1676
char *destination; \
1677
/* Must be int, so when we don't save any registers, the arithmetic \
1678
of 0 + -1 isn't done as unsigned. */ \
1679
/* Can't be int, since there is not a shred of a guarantee that int \
1680
is wide enough to hold a value of something to which pointer can \
1682
active_reg_t this_reg; \
1684
DEBUG_STATEMENT (failure_id++); \
1685
DEBUG_STATEMENT (nfailure_points_pushed++); \
1686
DEBUG_PRINT2 ("\nPUSH_FAILURE_POINT #%u:\n", failure_id); \
1687
DEBUG_PRINT2 (" Before push, next avail: %d\n", (fail_stack).avail);\
1688
DEBUG_PRINT2 (" size: %d\n", (fail_stack).size);\
1690
DEBUG_PRINT2 (" slots needed: %ld\n", NUM_FAILURE_ITEMS); \
1691
DEBUG_PRINT2 (" available: %d\n", REMAINING_AVAIL_SLOTS); \
1693
/* Ensure we have enough space allocated for what we will push. */ \
1694
while (REMAINING_AVAIL_SLOTS < NUM_FAILURE_ITEMS) \
1696
if (!DOUBLE_FAIL_STACK (fail_stack)) \
1697
return failure_code; \
1699
DEBUG_PRINT2 ("\n Doubled stack; size now: %d\n", \
1700
(fail_stack).size); \
1701
DEBUG_PRINT2 (" slots available: %d\n", REMAINING_AVAIL_SLOTS);\
1704
/* Push the info, starting with the registers. */ \
1705
DEBUG_PRINT1 ("\n"); \
1708
for (this_reg = lowest_active_reg; this_reg <= highest_active_reg; \
1711
DEBUG_PRINT2 (" Pushing reg: %lu\n", this_reg); \
1712
DEBUG_STATEMENT (num_regs_pushed++); \
1714
DEBUG_PRINT2 (" start: %p\n", regstart[this_reg]); \
1715
PUSH_FAILURE_POINTER (regstart[this_reg]); \
1717
DEBUG_PRINT2 (" end: %p\n", regend[this_reg]); \
1718
PUSH_FAILURE_POINTER (regend[this_reg]); \
1720
DEBUG_PRINT2 (" info: %p\n ", \
1721
reg_info[this_reg].word.pointer); \
1722
DEBUG_PRINT2 (" match_null=%d", \
1723
REG_MATCH_NULL_STRING_P (reg_info[this_reg])); \
1724
DEBUG_PRINT2 (" active=%d", IS_ACTIVE (reg_info[this_reg])); \
1725
DEBUG_PRINT2 (" matched_something=%d", \
1726
MATCHED_SOMETHING (reg_info[this_reg])); \
1727
DEBUG_PRINT2 (" ever_matched=%d", \
1728
EVER_MATCHED_SOMETHING (reg_info[this_reg])); \
1729
DEBUG_PRINT1 ("\n"); \
1730
PUSH_FAILURE_ELT (reg_info[this_reg].word); \
1733
DEBUG_PRINT2 (" Pushing low active reg: %ld\n", lowest_active_reg);\
1734
PUSH_FAILURE_INT (lowest_active_reg); \
1736
DEBUG_PRINT2 (" Pushing high active reg: %ld\n", highest_active_reg);\
1737
PUSH_FAILURE_INT (highest_active_reg); \
1739
DEBUG_PRINT2 (" Pushing pattern %p:\n", pattern_place); \
1740
DEBUG_PRINT_COMPILED_PATTERN (bufp, pattern_place, pend); \
1741
PUSH_FAILURE_POINTER (pattern_place); \
1743
DEBUG_PRINT2 (" Pushing string %p: `", string_place); \
1744
DEBUG_PRINT_DOUBLE_STRING (string_place, string1, size1, string2, \
1746
DEBUG_PRINT1 ("'\n"); \
1747
PUSH_FAILURE_POINTER (string_place); \
1749
DEBUG_PRINT2 (" Pushing failure id: %u\n", failure_id); \
1750
DEBUG_PUSH (failure_id); \
1753
# ifndef DEFINED_ONCE
1754
/* This is the number of items that are pushed and popped on the stack
1755
for each register. */
1756
# define NUM_REG_ITEMS 3
1758
/* Individual items aside from the registers. */
1760
# define NUM_NONREG_ITEMS 5 /* Includes failure point id. */
1762
# define NUM_NONREG_ITEMS 4
1765
/* We push at most this many items on the stack. */
1766
/* We used to use (num_regs - 1), which is the number of registers
1767
this regexp will save; but that was changed to 5
1768
to avoid stack overflow for a regexp with lots of parens. */
1769
# define MAX_FAILURE_ITEMS (5 * NUM_REG_ITEMS + NUM_NONREG_ITEMS)
1771
/* We actually push this many items. */
1772
# define NUM_FAILURE_ITEMS \
1774
? 0 : highest_active_reg - lowest_active_reg + 1) \
1778
/* How many items can still be added to the stack without overflowing it. */
1779
# define REMAINING_AVAIL_SLOTS ((fail_stack).size - (fail_stack).avail)
1780
# endif /* not DEFINED_ONCE */
1783
/* Pops what PUSH_FAIL_STACK pushes.
1785
We restore into the parameters, all of which should be lvalues:
1786
STR -- the saved data position.
1787
PAT -- the saved pattern position.
1788
LOW_REG, HIGH_REG -- the highest and lowest active registers.
1789
REGSTART, REGEND -- arrays of string positions.
1790
REG_INFO -- array of information about each subexpression.
1792
Also assumes the variables `fail_stack' and (if debugging), `bufp',
1793
`pend', `string1', `size1', `string2', and `size2'. */
1794
# define POP_FAILURE_POINT(str, pat, low_reg, high_reg, regstart, regend, reg_info)\
1796
DEBUG_STATEMENT (unsigned failure_id;) \
1797
active_reg_t this_reg; \
1798
const UCHAR_T *string_temp; \
1800
assert (!FAIL_STACK_EMPTY ()); \
1802
/* Remove failure points and point to how many regs pushed. */ \
1803
DEBUG_PRINT1 ("POP_FAILURE_POINT:\n"); \
1804
DEBUG_PRINT2 (" Before pop, next avail: %d\n", fail_stack.avail); \
1805
DEBUG_PRINT2 (" size: %d\n", fail_stack.size); \
1807
assert (fail_stack.avail >= NUM_NONREG_ITEMS); \
1809
DEBUG_POP (&failure_id); \
1810
DEBUG_PRINT2 (" Popping failure id: %u\n", failure_id); \
1812
/* If the saved string location is NULL, it came from an \
1813
on_failure_keep_string_jump opcode, and we want to throw away the \
1814
saved NULL, thus retaining our current position in the string. */ \
1815
string_temp = POP_FAILURE_POINTER (); \
1816
if (string_temp != NULL) \
1817
str = (const CHAR_T *) string_temp; \
1819
DEBUG_PRINT2 (" Popping string %p: `", str); \
1820
DEBUG_PRINT_DOUBLE_STRING (str, string1, size1, string2, size2); \
1821
DEBUG_PRINT1 ("'\n"); \
1823
pat = (UCHAR_T *) POP_FAILURE_POINTER (); \
1824
DEBUG_PRINT2 (" Popping pattern %p:\n", pat); \
1825
DEBUG_PRINT_COMPILED_PATTERN (bufp, pat, pend); \
1827
/* Restore register info. */ \
1828
high_reg = (active_reg_t) POP_FAILURE_INT (); \
1829
DEBUG_PRINT2 (" Popping high active reg: %ld\n", high_reg); \
1831
low_reg = (active_reg_t) POP_FAILURE_INT (); \
1832
DEBUG_PRINT2 (" Popping low active reg: %ld\n", low_reg); \
1835
for (this_reg = high_reg; this_reg >= low_reg; this_reg--) \
1837
DEBUG_PRINT2 (" Popping reg: %ld\n", this_reg); \
1839
reg_info[this_reg].word = POP_FAILURE_ELT (); \
1840
DEBUG_PRINT2 (" info: %p\n", \
1841
reg_info[this_reg].word.pointer); \
1843
regend[this_reg] = (const CHAR_T *) POP_FAILURE_POINTER (); \
1844
DEBUG_PRINT2 (" end: %p\n", regend[this_reg]); \
1846
regstart[this_reg] = (const CHAR_T *) POP_FAILURE_POINTER (); \
1847
DEBUG_PRINT2 (" start: %p\n", regstart[this_reg]); \
1851
for (this_reg = highest_active_reg; this_reg > high_reg; this_reg--) \
1853
reg_info[this_reg].word.integer = 0; \
1854
regend[this_reg] = 0; \
1855
regstart[this_reg] = 0; \
1857
highest_active_reg = high_reg; \
1860
set_regs_matched_done = 0; \
1861
DEBUG_STATEMENT (nfailure_points_popped++); \
1862
} /* POP_FAILURE_POINT */
1864
/* Structure for per-register (a.k.a. per-group) information.
1865
Other register information, such as the
1866
starting and ending positions (which are addresses), and the list of
1867
inner groups (which is a bits list) are maintained in separate
1870
We are making a (strictly speaking) nonportable assumption here: that
1871
the compiler will pack our bit fields into something that fits into
1872
the type of `word', i.e., is something that fits into one item on the
1876
/* Declarations and macros for re_match_2. */
1880
PREFIX(fail_stack_elt_t) word;
1883
/* This field is one if this group can match the empty string,
1884
zero if not. If not yet determined, `MATCH_NULL_UNSET_VALUE'. */
1885
# define MATCH_NULL_UNSET_VALUE 3
1886
unsigned match_null_string_p : 2;
1887
unsigned is_active : 1;
1888
unsigned matched_something : 1;
1889
unsigned ever_matched_something : 1;
1891
} PREFIX(register_info_type);
1893
# ifndef DEFINED_ONCE
1894
# define REG_MATCH_NULL_STRING_P(R) ((R).bits.match_null_string_p)
1895
# define IS_ACTIVE(R) ((R).bits.is_active)
1896
# define MATCHED_SOMETHING(R) ((R).bits.matched_something)
1897
# define EVER_MATCHED_SOMETHING(R) ((R).bits.ever_matched_something)
1900
/* Call this when have matched a real character; it sets `matched' flags
1901
for the subexpressions which we are currently inside. Also records
1902
that those subexprs have matched. */
1903
# define SET_REGS_MATCHED() \
1906
if (!set_regs_matched_done) \
1909
set_regs_matched_done = 1; \
1910
for (r = lowest_active_reg; r <= highest_active_reg; r++) \
1912
MATCHED_SOMETHING (reg_info[r]) \
1913
= EVER_MATCHED_SOMETHING (reg_info[r]) \
1919
# endif /* not DEFINED_ONCE */
1921
/* Registers are set to a sentinel when they haven't yet matched. */
1922
static CHAR_T PREFIX(reg_unset_dummy);
1923
# define REG_UNSET_VALUE (&PREFIX(reg_unset_dummy))
1924
# define REG_UNSET(e) ((e) == REG_UNSET_VALUE)
1926
/* Subroutine declarations and macros for regex_compile. */
1927
static void PREFIX(store_op1) _RE_ARGS ((re_opcode_t op, UCHAR_T *loc, int arg));
1928
static void PREFIX(store_op2) _RE_ARGS ((re_opcode_t op, UCHAR_T *loc,
1929
int arg1, int arg2));
1930
static void PREFIX(insert_op1) _RE_ARGS ((re_opcode_t op, UCHAR_T *loc,
1931
int arg, UCHAR_T *end));
1932
static void PREFIX(insert_op2) _RE_ARGS ((re_opcode_t op, UCHAR_T *loc,
1933
int arg1, int arg2, UCHAR_T *end));
1934
static boolean PREFIX(at_begline_loc_p) _RE_ARGS ((const CHAR_T *pattern,
1936
reg_syntax_t syntax));
1937
static boolean PREFIX(at_endline_loc_p) _RE_ARGS ((const CHAR_T *p,
1939
reg_syntax_t syntax));
1941
static reg_errcode_t wcs_compile_range _RE_ARGS ((CHAR_T range_start,
1942
const CHAR_T **p_ptr,
1945
reg_syntax_t syntax,
1948
static void insert_space _RE_ARGS ((int num, CHAR_T *loc, CHAR_T *end));
1950
static reg_errcode_t byte_compile_range _RE_ARGS ((unsigned int range_start,
1954
reg_syntax_t syntax,
1958
/* Fetch the next character in the uncompiled pattern---translating it
1959
if necessary. Also cast from a signed character in the constant
1960
string passed to us by the user to an unsigned char that we can use
1961
as an array index (in, e.g., `translate'). */
1962
/* ifdef MBS_SUPPORT, we translate only if character <= 0xff,
1963
because it is impossible to allocate 4GB array for some encodings
1964
which have 4 byte character_set like UCS4. */
1967
# define PATFETCH(c) \
1968
do {if (p == pend) return REG_EEND; \
1969
c = (UCHAR_T) *p++; \
1970
if (translate && (c <= 0xff)) c = (UCHAR_T) translate[c]; \
1973
# define PATFETCH(c) \
1974
do {if (p == pend) return REG_EEND; \
1975
c = (unsigned char) *p++; \
1976
if (translate) c = (unsigned char) translate[c]; \
1981
/* Fetch the next character in the uncompiled pattern, with no
1983
# define PATFETCH_RAW(c) \
1984
do {if (p == pend) return REG_EEND; \
1985
c = (UCHAR_T) *p++; \
1988
/* Go backwards one character in the pattern. */
1989
# define PATUNFETCH p--
1992
/* If `translate' is non-null, return translate[D], else just D. We
1993
cast the subscript to translate because some data is declared as
1994
`char *', to avoid warnings when a string constant is passed. But
1995
when we use a character as a subscript we must make it unsigned. */
1996
/* ifdef MBS_SUPPORT, we translate only if character <= 0xff,
1997
because it is impossible to allocate 4GB array for some encodings
1998
which have 4 byte character_set like UCS4. */
2002
# define TRANSLATE(d) \
2003
((translate && ((UCHAR_T) (d)) <= 0xff) \
2004
? (char) translate[(unsigned char) (d)] : (d))
2006
# define TRANSLATE(d) \
2007
(translate ? (char) translate[(unsigned char) (d)] : (d))
2012
/* Macros for outputting the compiled pattern into `buffer'. */
2014
/* If the buffer isn't allocated when it comes in, use this. */
2015
# define INIT_BUF_SIZE (32 * sizeof(UCHAR_T))
2017
/* Make sure we have at least N more bytes of space in buffer. */
2019
# define GET_BUFFER_SPACE(n) \
2020
while (((unsigned long)b - (unsigned long)COMPILED_BUFFER_VAR \
2021
+ (n)*sizeof(CHAR_T)) > bufp->allocated) \
2024
# define GET_BUFFER_SPACE(n) \
2025
while ((unsigned long) (b - bufp->buffer + (n)) > bufp->allocated) \
2029
/* Make sure we have one more byte of buffer space and then add C to it. */
2030
# define BUF_PUSH(c) \
2032
GET_BUFFER_SPACE (1); \
2033
*b++ = (UCHAR_T) (c); \
2037
/* Ensure we have two more bytes of buffer space and then append C1 and C2. */
2038
# define BUF_PUSH_2(c1, c2) \
2040
GET_BUFFER_SPACE (2); \
2041
*b++ = (UCHAR_T) (c1); \
2042
*b++ = (UCHAR_T) (c2); \
2046
/* As with BUF_PUSH_2, except for three bytes. */
2047
# define BUF_PUSH_3(c1, c2, c3) \
2049
GET_BUFFER_SPACE (3); \
2050
*b++ = (UCHAR_T) (c1); \
2051
*b++ = (UCHAR_T) (c2); \
2052
*b++ = (UCHAR_T) (c3); \
2055
/* Store a jump with opcode OP at LOC to location TO. We store a
2056
relative address offset by the three bytes the jump itself occupies. */
2057
# define STORE_JUMP(op, loc, to) \
2058
PREFIX(store_op1) (op, loc, (int) ((to) - (loc) - (1 + OFFSET_ADDRESS_SIZE)))
2060
/* Likewise, for a two-argument jump. */
2061
# define STORE_JUMP2(op, loc, to, arg) \
2062
PREFIX(store_op2) (op, loc, (int) ((to) - (loc) - (1 + OFFSET_ADDRESS_SIZE)), arg)
2064
/* Like `STORE_JUMP', but for inserting. Assume `b' is the buffer end. */
2065
# define INSERT_JUMP(op, loc, to) \
2066
PREFIX(insert_op1) (op, loc, (int) ((to) - (loc) - (1 + OFFSET_ADDRESS_SIZE)), b)
2068
/* Like `STORE_JUMP2', but for inserting. Assume `b' is the buffer end. */
2069
# define INSERT_JUMP2(op, loc, to, arg) \
2070
PREFIX(insert_op2) (op, loc, (int) ((to) - (loc) - (1 + OFFSET_ADDRESS_SIZE)),\
2073
/* This is not an arbitrary limit: the arguments which represent offsets
2074
into the pattern are two bytes long. So if 2^16 bytes turns out to
2075
be too small, many things would have to change. */
2076
/* Any other compiler which, like MSC, has allocation limit below 2^16
2077
bytes will have to use approach similar to what was done below for
2078
MSC and drop MAX_BUF_SIZE a bit. Otherwise you may end up
2079
reallocating to 0 bytes. Such thing is not going to work too well.
2080
You have been warned!! */
2081
# ifndef DEFINED_ONCE
2082
# if defined _MSC_VER && !defined WIN32
2083
/* Microsoft C 16-bit versions limit malloc to approx 65512 bytes.
2084
The REALLOC define eliminates a flurry of conversion warnings,
2085
but is not required. */
2086
# define MAX_BUF_SIZE 65500L
2087
# define REALLOC(p,s) realloc ((p), (size_t) (s))
2089
# define MAX_BUF_SIZE (1L << 16)
2090
# define REALLOC(p,s) realloc ((p), (s))
2093
/* Extend the buffer by twice its current size via realloc and
2094
reset the pointers that pointed into the old block to point to the
2095
correct places in the new one. If extending the buffer results in it
2096
being larger than MAX_BUF_SIZE, then flag memory exhausted. */
2097
# if __BOUNDED_POINTERS__
2098
# define SET_HIGH_BOUND(P) (__ptrhigh (P) = __ptrlow (P) + bufp->allocated)
2099
# define MOVE_BUFFER_POINTER(P) \
2100
(__ptrlow (P) += incr, SET_HIGH_BOUND (P), __ptrvalue (P) += incr)
2101
# define ELSE_EXTEND_BUFFER_HIGH_BOUND \
2104
SET_HIGH_BOUND (b); \
2105
SET_HIGH_BOUND (begalt); \
2106
if (fixup_alt_jump) \
2107
SET_HIGH_BOUND (fixup_alt_jump); \
2109
SET_HIGH_BOUND (laststart); \
2110
if (pending_exact) \
2111
SET_HIGH_BOUND (pending_exact); \
2114
# define MOVE_BUFFER_POINTER(P) (P) += incr
2115
# define ELSE_EXTEND_BUFFER_HIGH_BOUND
2117
# endif /* not DEFINED_ONCE */
2120
# define EXTEND_BUFFER() \
2122
UCHAR_T *old_buffer = COMPILED_BUFFER_VAR; \
2124
if (bufp->allocated + sizeof(UCHAR_T) > MAX_BUF_SIZE) \
2126
bufp->allocated <<= 1; \
2127
if (bufp->allocated > MAX_BUF_SIZE) \
2128
bufp->allocated = MAX_BUF_SIZE; \
2129
/* How many characters the new buffer can have? */ \
2130
wchar_count = bufp->allocated / sizeof(UCHAR_T); \
2131
if (wchar_count == 0) wchar_count = 1; \
2132
/* Truncate the buffer to CHAR_T align. */ \
2133
bufp->allocated = wchar_count * sizeof(UCHAR_T); \
2134
RETALLOC (COMPILED_BUFFER_VAR, wchar_count, UCHAR_T); \
2135
bufp->buffer = (char*)COMPILED_BUFFER_VAR; \
2136
if (COMPILED_BUFFER_VAR == NULL) \
2137
return REG_ESPACE; \
2138
/* If the buffer moved, move all the pointers into it. */ \
2139
if (old_buffer != COMPILED_BUFFER_VAR) \
2141
int incr = COMPILED_BUFFER_VAR - old_buffer; \
2142
MOVE_BUFFER_POINTER (b); \
2143
MOVE_BUFFER_POINTER (begalt); \
2144
if (fixup_alt_jump) \
2145
MOVE_BUFFER_POINTER (fixup_alt_jump); \
2147
MOVE_BUFFER_POINTER (laststart); \
2148
if (pending_exact) \
2149
MOVE_BUFFER_POINTER (pending_exact); \
2151
ELSE_EXTEND_BUFFER_HIGH_BOUND \
2154
# define EXTEND_BUFFER() \
2156
UCHAR_T *old_buffer = COMPILED_BUFFER_VAR; \
2157
if (bufp->allocated == MAX_BUF_SIZE) \
2159
bufp->allocated <<= 1; \
2160
if (bufp->allocated > MAX_BUF_SIZE) \
2161
bufp->allocated = MAX_BUF_SIZE; \
2162
bufp->buffer = (UCHAR_T *) REALLOC (COMPILED_BUFFER_VAR, \
2164
if (COMPILED_BUFFER_VAR == NULL) \
2165
return REG_ESPACE; \
2166
/* If the buffer moved, move all the pointers into it. */ \
2167
if (old_buffer != COMPILED_BUFFER_VAR) \
2169
int incr = COMPILED_BUFFER_VAR - old_buffer; \
2170
MOVE_BUFFER_POINTER (b); \
2171
MOVE_BUFFER_POINTER (begalt); \
2172
if (fixup_alt_jump) \
2173
MOVE_BUFFER_POINTER (fixup_alt_jump); \
2175
MOVE_BUFFER_POINTER (laststart); \
2176
if (pending_exact) \
2177
MOVE_BUFFER_POINTER (pending_exact); \
2179
ELSE_EXTEND_BUFFER_HIGH_BOUND \
2183
# ifndef DEFINED_ONCE
2184
/* Since we have one byte reserved for the register number argument to
2185
{start,stop}_memory, the maximum number of groups we can report
2186
things about is what fits in that byte. */
2187
# define MAX_REGNUM 255
2189
/* But patterns can have more than `MAX_REGNUM' registers. We just
2190
ignore the excess. */
2191
typedef unsigned regnum_t;
2194
/* Macros for the compile stack. */
2196
/* Since offsets can go either forwards or backwards, this type needs to
2197
be able to hold values from -(MAX_BUF_SIZE - 1) to MAX_BUF_SIZE - 1. */
2198
/* int may be not enough when sizeof(int) == 2. */
2199
typedef long pattern_offset_t;
2203
pattern_offset_t begalt_offset;
2204
pattern_offset_t fixup_alt_jump;
2205
pattern_offset_t inner_group_offset;
2206
pattern_offset_t laststart_offset;
2208
} compile_stack_elt_t;
2213
compile_stack_elt_t *stack;
2215
unsigned avail; /* Offset of next open position. */
2216
} compile_stack_type;
2219
# define INIT_COMPILE_STACK_SIZE 32
2221
# define COMPILE_STACK_EMPTY (compile_stack.avail == 0)
2222
# define COMPILE_STACK_FULL (compile_stack.avail == compile_stack.size)
2224
/* The next available element. */
2225
# define COMPILE_STACK_TOP (compile_stack.stack[compile_stack.avail])
2227
# endif /* not DEFINED_ONCE */
2229
/* Set the bit for character C in a list. */
2230
# ifndef DEFINED_ONCE
2231
# define SET_LIST_BIT(c) \
2232
(b[((unsigned char) (c)) / BYTEWIDTH] \
2233
|= 1 << (((unsigned char) c) % BYTEWIDTH))
2234
# endif /* DEFINED_ONCE */
2236
/* Get the next unsigned number in the uncompiled pattern. */
2237
# define GET_UNSIGNED_NUMBER(num) \
2242
if (c < '0' || c > '9') \
2244
if (num <= RE_DUP_MAX) \
2248
num = num * 10 + c - '0'; \
2253
# ifndef DEFINED_ONCE
2254
# if defined _LIBC || WIDE_CHAR_SUPPORT
2255
/* The GNU C library provides support for user-defined character classes
2256
and the functions from ISO C amendement 1. */
2257
# ifdef CHARCLASS_NAME_MAX
2258
# define CHAR_CLASS_MAX_LENGTH CHARCLASS_NAME_MAX
2260
/* This shouldn't happen but some implementation might still have this
2261
problem. Use a reasonable default value. */
2262
# define CHAR_CLASS_MAX_LENGTH 256
2266
# define IS_CHAR_CLASS(string) __wctype (string)
2268
# define IS_CHAR_CLASS(string) wctype (string)
2271
# define CHAR_CLASS_MAX_LENGTH 6 /* Namely, `xdigit'. */
2273
# define IS_CHAR_CLASS(string) \
2274
(STREQ (string, "alpha") || STREQ (string, "upper") \
2275
|| STREQ (string, "lower") || STREQ (string, "digit") \
2276
|| STREQ (string, "alnum") || STREQ (string, "xdigit") \
2277
|| STREQ (string, "space") || STREQ (string, "print") \
2278
|| STREQ (string, "punct") || STREQ (string, "graph") \
2279
|| STREQ (string, "cntrl") || STREQ (string, "blank"))
2281
# endif /* DEFINED_ONCE */
2283
# ifndef MATCH_MAY_ALLOCATE
2285
/* If we cannot allocate large objects within re_match_2_internal,
2286
we make the fail stack and register vectors global.
2287
The fail stack, we grow to the maximum size when a regexp
2289
The register vectors, we adjust in size each time we
2290
compile a regexp, according to the number of registers it needs. */
2292
static PREFIX(fail_stack_type) fail_stack;
2294
/* Size with which the following vectors are currently allocated.
2295
That is so we can make them bigger as needed,
2296
but never make them smaller. */
2297
# ifdef DEFINED_ONCE
2298
static int regs_allocated_size;
2300
static const char ** regstart, ** regend;
2301
static const char ** old_regstart, ** old_regend;
2302
static const char **best_regstart, **best_regend;
2303
static const char **reg_dummy;
2304
# endif /* DEFINED_ONCE */
2306
static PREFIX(register_info_type) *PREFIX(reg_info);
2307
static PREFIX(register_info_type) *PREFIX(reg_info_dummy);
2309
/* Make the register vectors big enough for NUM_REGS registers,
2310
but don't make them smaller. */
2313
PREFIX(regex_grow_registers) (num_regs)
2316
if (num_regs > regs_allocated_size)
2318
RETALLOC_IF (regstart, num_regs, const char *);
2319
RETALLOC_IF (regend, num_regs, const char *);
2320
RETALLOC_IF (old_regstart, num_regs, const char *);
2321
RETALLOC_IF (old_regend, num_regs, const char *);
2322
RETALLOC_IF (best_regstart, num_regs, const char *);
2323
RETALLOC_IF (best_regend, num_regs, const char *);
2324
RETALLOC_IF (PREFIX(reg_info), num_regs, PREFIX(register_info_type));
2325
RETALLOC_IF (reg_dummy, num_regs, const char *);
2326
RETALLOC_IF (PREFIX(reg_info_dummy), num_regs, PREFIX(register_info_type));
2328
regs_allocated_size = num_regs;
2332
# endif /* not MATCH_MAY_ALLOCATE */
2334
# ifndef DEFINED_ONCE
2335
static boolean group_in_compile_stack _RE_ARGS ((compile_stack_type
2338
# endif /* not DEFINED_ONCE */
2340
/* `regex_compile' compiles PATTERN (of length SIZE) according to SYNTAX.
2341
Returns one of error codes defined in `regex.h', or zero for success.
2343
Assumes the `allocated' (and perhaps `buffer') and `translate'
2344
fields are set in BUFP on entry.
2346
If it succeeds, results are put in BUFP (if it returns an error, the
2347
contents of BUFP are undefined):
2348
`buffer' is the compiled pattern;
2349
`syntax' is set to SYNTAX;
2350
`used' is set to the length of the compiled pattern;
2351
`fastmap_accurate' is zero;
2352
`re_nsub' is the number of subexpressions in PATTERN;
2353
`not_bol' and `not_eol' are zero;
2355
The `fastmap' and `newline_anchor' fields are neither
2356
examined nor set. */
2358
/* Return, freeing storage we allocated. */
2360
# define FREE_STACK_RETURN(value) \
2361
return (free(pattern), free(mbs_offset), free(is_binary), free (compile_stack.stack), value)
2363
# define FREE_STACK_RETURN(value) \
2364
return (free (compile_stack.stack), value)
2367
static reg_errcode_t
2368
PREFIX(regex_compile) (ARG_PREFIX(pattern), ARG_PREFIX(size), syntax, bufp)
2369
const char *ARG_PREFIX(pattern);
2370
size_t ARG_PREFIX(size);
2371
reg_syntax_t syntax;
2372
struct re_pattern_buffer *bufp;
2374
/* We fetch characters from PATTERN here. Even though PATTERN is
2375
`char *' (i.e., signed), we declare these variables as unsigned, so
2376
they can be reliably used as array indices. */
2377
register UCHAR_T c, c1;
2380
/* A temporary space to keep wchar_t pattern and compiled pattern. */
2381
CHAR_T *pattern, *COMPILED_BUFFER_VAR;
2383
/* offset buffer for optimization. See convert_mbs_to_wc. */
2384
int *mbs_offset = NULL;
2385
/* It hold whether each wchar_t is binary data or not. */
2386
char *is_binary = NULL;
2387
/* A flag whether exactn is handling binary data or not. */
2388
char is_exactn_bin = FALSE;
2391
/* A random temporary spot in PATTERN. */
2394
/* Points to the end of the buffer, where we should append. */
2395
register UCHAR_T *b;
2397
/* Keeps track of unclosed groups. */
2398
compile_stack_type compile_stack;
2400
/* Points to the current (ending) position in the pattern. */
2405
const CHAR_T *p = pattern;
2406
const CHAR_T *pend = pattern + size;
2409
/* How to translate the characters in the pattern. */
2410
RE_TRANSLATE_TYPE translate = bufp->translate;
2412
/* Address of the count-byte of the most recently inserted `exactn'
2413
command. This makes it possible to tell if a new exact-match
2414
character can be added to that command or if the character requires
2415
a new `exactn' command. */
2416
UCHAR_T *pending_exact = 0;
2418
/* Address of start of the most recently finished expression.
2419
This tells, e.g., postfix * where to find the start of its
2420
operand. Reset at the beginning of groups and alternatives. */
2421
UCHAR_T *laststart = 0;
2423
/* Address of beginning of regexp, or inside of last group. */
2426
/* Address of the place where a forward jump should go to the end of
2427
the containing expression. Each alternative of an `or' -- except the
2428
last -- ends with a forward jump of this sort. */
2429
UCHAR_T *fixup_alt_jump = 0;
2431
/* Counts open-groups as they are encountered. Remembered for the
2432
matching close-group on the compile stack, so the same register
2433
number is put in the stop_memory as the start_memory. */
2434
regnum_t regnum = 0;
2437
/* Initialize the wchar_t PATTERN and offset_buffer. */
2438
p = pend = pattern = TALLOC(csize + 1, CHAR_T);
2439
mbs_offset = TALLOC(csize + 1, int);
2440
is_binary = TALLOC(csize + 1, char);
2441
if (pattern == NULL || mbs_offset == NULL || is_binary == NULL)
2448
pattern[csize] = L'\0'; /* sentinel */
2449
size = convert_mbs_to_wcs(pattern, cpattern, csize, mbs_offset, is_binary);
2461
DEBUG_PRINT1 ("\nCompiling pattern: ");
2464
unsigned debug_count;
2466
for (debug_count = 0; debug_count < size; debug_count++)
2467
PUT_CHAR (pattern[debug_count]);
2472
/* Initialize the compile stack. */
2473
compile_stack.stack = TALLOC (INIT_COMPILE_STACK_SIZE, compile_stack_elt_t);
2474
if (compile_stack.stack == NULL)
2484
compile_stack.size = INIT_COMPILE_STACK_SIZE;
2485
compile_stack.avail = 0;
2487
/* Initialize the pattern buffer. */
2488
bufp->syntax = syntax;
2489
bufp->fastmap_accurate = 0;
2490
bufp->not_bol = bufp->not_eol = 0;
2492
/* Set `used' to zero, so that if we return an error, the pattern
2493
printer (for debugging) will think there's no pattern. We reset it
2497
/* Always count groups, whether or not bufp->no_sub is set. */
2500
#if !defined emacs && !defined SYNTAX_TABLE
2501
/* Initialize the syntax table. */
2502
init_syntax_once ();
2505
if (bufp->allocated == 0)
2508
{ /* If zero allocated, but buffer is non-null, try to realloc
2509
enough space. This loses if buffer's address is bogus, but
2510
that is the user's responsibility. */
2512
/* Free bufp->buffer and allocate an array for wchar_t pattern
2515
COMPILED_BUFFER_VAR = TALLOC (INIT_BUF_SIZE/sizeof(UCHAR_T),
2518
RETALLOC (COMPILED_BUFFER_VAR, INIT_BUF_SIZE, UCHAR_T);
2522
{ /* Caller did not allocate a buffer. Do it for them. */
2523
COMPILED_BUFFER_VAR = TALLOC (INIT_BUF_SIZE / sizeof(UCHAR_T),
2527
if (!COMPILED_BUFFER_VAR) FREE_STACK_RETURN (REG_ESPACE);
2529
bufp->buffer = (char*)COMPILED_BUFFER_VAR;
2531
bufp->allocated = INIT_BUF_SIZE;
2535
COMPILED_BUFFER_VAR = (UCHAR_T*) bufp->buffer;
2538
begalt = b = COMPILED_BUFFER_VAR;
2540
/* Loop through the uncompiled pattern until we're at the end. */
2549
if ( /* If at start of pattern, it's an operator. */
2551
/* If context independent, it's an operator. */
2552
|| syntax & RE_CONTEXT_INDEP_ANCHORS
2553
/* Otherwise, depends on what's come before. */
2554
|| PREFIX(at_begline_loc_p) (pattern, p, syntax))
2564
if ( /* If at end of pattern, it's an operator. */
2566
/* If context independent, it's an operator. */
2567
|| syntax & RE_CONTEXT_INDEP_ANCHORS
2568
/* Otherwise, depends on what's next. */
2569
|| PREFIX(at_endline_loc_p) (p, pend, syntax))
2579
if ((syntax & RE_BK_PLUS_QM)
2580
|| (syntax & RE_LIMITED_OPS))
2584
/* If there is no previous pattern... */
2587
if (syntax & RE_CONTEXT_INVALID_OPS)
2588
FREE_STACK_RETURN (REG_BADRPT);
2589
else if (!(syntax & RE_CONTEXT_INDEP_OPS))
2594
/* Are we optimizing this jump? */
2595
boolean keep_string_p = false;
2597
/* 1 means zero (many) matches is allowed. */
2598
char zero_times_ok = 0, many_times_ok = 0;
2600
/* If there is a sequence of repetition chars, collapse it
2601
down to just one (the right one). We can't combine
2602
interval operators with these because of, e.g., `a{2}*',
2603
which should only match an even number of `a's. */
2607
zero_times_ok |= c != '+';
2608
many_times_ok |= c != '?';
2616
|| (!(syntax & RE_BK_PLUS_QM) && (c == '+' || c == '?')))
2619
else if (syntax & RE_BK_PLUS_QM && c == '\\')
2621
if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
2624
if (!(c1 == '+' || c1 == '?'))
2639
/* If we get here, we found another repeat character. */
2642
/* Star, etc. applied to an empty pattern is equivalent
2643
to an empty pattern. */
2647
/* Now we know whether or not zero matches is allowed
2648
and also whether or not two or more matches is allowed. */
2650
{ /* More than one repetition is allowed, so put in at the
2651
end a backward relative jump from `b' to before the next
2652
jump we're going to put in below (which jumps from
2653
laststart to after this jump).
2655
But if we are at the `*' in the exact sequence `.*\n',
2656
insert an unconditional jump backwards to the .,
2657
instead of the beginning of the loop. This way we only
2658
push a failure point once, instead of every time
2659
through the loop. */
2660
assert (p - 1 > pattern);
2662
/* Allocate the space for the jump. */
2663
GET_BUFFER_SPACE (1 + OFFSET_ADDRESS_SIZE);
2665
/* We know we are not at the first character of the pattern,
2666
because laststart was nonzero. And we've already
2667
incremented `p', by the way, to be the character after
2668
the `*'. Do we have to do something analogous here
2669
for null bytes, because of RE_DOT_NOT_NULL? */
2670
if (TRANSLATE (*(p - 2)) == TRANSLATE ('.')
2672
&& p < pend && TRANSLATE (*p) == TRANSLATE ('\n')
2673
&& !(syntax & RE_DOT_NEWLINE))
2674
{ /* We have .*\n. */
2675
STORE_JUMP (jump, b, laststart);
2676
keep_string_p = true;
2679
/* Anything else. */
2680
STORE_JUMP (maybe_pop_jump, b, laststart -
2681
(1 + OFFSET_ADDRESS_SIZE));
2683
/* We've added more stuff to the buffer. */
2684
b += 1 + OFFSET_ADDRESS_SIZE;
2687
/* On failure, jump from laststart to b + 3, which will be the
2688
end of the buffer after this jump is inserted. */
2689
/* ifdef WCHAR, 'b + 1 + OFFSET_ADDRESS_SIZE' instead of
2691
GET_BUFFER_SPACE (1 + OFFSET_ADDRESS_SIZE);
2692
INSERT_JUMP (keep_string_p ? on_failure_keep_string_jump
2694
laststart, b + 1 + OFFSET_ADDRESS_SIZE);
2696
b += 1 + OFFSET_ADDRESS_SIZE;
2700
/* At least one repetition is required, so insert a
2701
`dummy_failure_jump' before the initial
2702
`on_failure_jump' instruction of the loop. This
2703
effects a skip over that instruction the first time
2704
we hit that loop. */
2705
GET_BUFFER_SPACE (1 + OFFSET_ADDRESS_SIZE);
2706
INSERT_JUMP (dummy_failure_jump, laststart, laststart +
2707
2 + 2 * OFFSET_ADDRESS_SIZE);
2708
b += 1 + OFFSET_ADDRESS_SIZE;
2722
boolean had_char_class = false;
2724
CHAR_T range_start = 0xffffffff;
2726
unsigned int range_start = 0xffffffff;
2728
if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2731
/* We assume a charset(_not) structure as a wchar_t array.
2732
charset[0] = (re_opcode_t) charset(_not)
2733
charset[1] = l (= length of char_classes)
2734
charset[2] = m (= length of collating_symbols)
2735
charset[3] = n (= length of equivalence_classes)
2736
charset[4] = o (= length of char_ranges)
2737
charset[5] = p (= length of chars)
2739
charset[6] = char_class (wctype_t)
2740
charset[6+CHAR_CLASS_SIZE] = char_class (wctype_t)
2742
charset[l+5] = char_class (wctype_t)
2744
charset[l+6] = collating_symbol (wchar_t)
2746
charset[l+m+5] = collating_symbol (wchar_t)
2747
ifdef _LIBC we use the index if
2748
_NL_COLLATE_SYMB_EXTRAMB instead of
2751
charset[l+m+6] = equivalence_classes (wchar_t)
2753
charset[l+m+n+5] = equivalence_classes (wchar_t)
2754
ifdef _LIBC we use the index in
2755
_NL_COLLATE_WEIGHT instead of
2758
charset[l+m+n+6] = range_start
2759
charset[l+m+n+7] = range_end
2761
charset[l+m+n+2o+4] = range_start
2762
charset[l+m+n+2o+5] = range_end
2763
ifdef _LIBC we use the value looked up
2764
in _NL_COLLATE_COLLSEQ instead of
2767
charset[l+m+n+2o+6] = char
2769
charset[l+m+n+2o+p+5] = char
2773
/* We need at least 6 spaces: the opcode, the length of
2774
char_classes, the length of collating_symbols, the length of
2775
equivalence_classes, the length of char_ranges, the length of
2777
GET_BUFFER_SPACE (6);
2779
/* Save b as laststart. And We use laststart as the pointer
2780
to the first element of the charset here.
2781
In other words, laststart[i] indicates charset[i]. */
2784
/* We test `*p == '^' twice, instead of using an if
2785
statement, so we only need one BUF_PUSH. */
2786
BUF_PUSH (*p == '^' ? charset_not : charset);
2790
/* Push the length of char_classes, the length of
2791
collating_symbols, the length of equivalence_classes, the
2792
length of char_ranges and the length of chars. */
2793
BUF_PUSH_3 (0, 0, 0);
2796
/* Remember the first position in the bracket expression. */
2799
/* charset_not matches newline according to a syntax bit. */
2800
if ((re_opcode_t) b[-6] == charset_not
2801
&& (syntax & RE_HAT_LISTS_NOT_NEWLINE))
2804
laststart[5]++; /* Update the length of characters */
2807
/* Read in characters and ranges, setting map bits. */
2810
if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2814
/* \ might escape characters inside [...] and [^...]. */
2815
if ((syntax & RE_BACKSLASH_ESCAPE_IN_LISTS) && c == '\\')
2817
if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
2821
laststart[5]++; /* Update the length of chars */
2826
/* Could be the end of the bracket expression. If it's
2827
not (i.e., when the bracket expression is `[]' so
2828
far), the ']' character bit gets set way below. */
2829
if (c == ']' && p != p1 + 1)
2832
/* Look ahead to see if it's a range when the last thing
2833
was a character class. */
2834
if (had_char_class && c == '-' && *p != ']')
2835
FREE_STACK_RETURN (REG_ERANGE);
2837
/* Look ahead to see if it's a range when the last thing
2838
was a character: if this is a hyphen not at the
2839
beginning or the end of a list, then it's the range
2842
&& !(p - 2 >= pattern && p[-2] == '[')
2843
&& !(p - 3 >= pattern && p[-3] == '[' && p[-2] == '^')
2847
/* Allocate the space for range_start and range_end. */
2848
GET_BUFFER_SPACE (2);
2849
/* Update the pointer to indicate end of buffer. */
2851
ret = wcs_compile_range (range_start, &p, pend, translate,
2852
syntax, b, laststart);
2853
if (ret != REG_NOERROR) FREE_STACK_RETURN (ret);
2854
range_start = 0xffffffff;
2856
else if (p[0] == '-' && p[1] != ']')
2857
{ /* This handles ranges made up of characters only. */
2860
/* Move past the `-'. */
2862
/* Allocate the space for range_start and range_end. */
2863
GET_BUFFER_SPACE (2);
2864
/* Update the pointer to indicate end of buffer. */
2866
ret = wcs_compile_range (c, &p, pend, translate, syntax, b,
2868
if (ret != REG_NOERROR) FREE_STACK_RETURN (ret);
2869
range_start = 0xffffffff;
2872
/* See if we're at the beginning of a possible character
2874
else if (syntax & RE_CHAR_CLASSES && c == '[' && *p == ':')
2875
{ /* Leave room for the null. */
2876
char str[CHAR_CLASS_MAX_LENGTH + 1];
2881
/* If pattern is `[[:'. */
2882
if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2887
if ((c == ':' && *p == ']') || p == pend)
2889
if (c1 < CHAR_CLASS_MAX_LENGTH)
2892
/* This is in any case an invalid class name. */
2897
/* If isn't a word bracketed by `[:' and `:]':
2898
undo the ending character, the letters, and leave
2899
the leading `:' and `[' (but store them as character). */
2900
if (c == ':' && *p == ']')
2905
/* Query the character class as wctype_t. */
2906
wt = IS_CHAR_CLASS (str);
2908
FREE_STACK_RETURN (REG_ECTYPE);
2910
/* Throw away the ] at the end of the character
2914
if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2916
/* Allocate the space for character class. */
2917
GET_BUFFER_SPACE(CHAR_CLASS_SIZE);
2918
/* Update the pointer to indicate end of buffer. */
2919
b += CHAR_CLASS_SIZE;
2920
/* Move data which follow character classes
2921
not to violate the data. */
2922
insert_space(CHAR_CLASS_SIZE,
2923
laststart + 6 + laststart[1],
2925
alignedp = ((uintptr_t)(laststart + 6 + laststart[1])
2926
+ __alignof__(wctype_t) - 1)
2927
& ~(uintptr_t)(__alignof__(wctype_t) - 1);
2928
/* Store the character class. */
2929
*((wctype_t*)alignedp) = wt;
2930
/* Update length of char_classes */
2931
laststart[1] += CHAR_CLASS_SIZE;
2933
had_char_class = true;
2942
laststart[5] += 2; /* Update the length of characters */
2944
had_char_class = false;
2947
else if (syntax & RE_CHAR_CLASSES && c == '[' && (*p == '='
2950
CHAR_T str[128]; /* Should be large enough. */
2951
CHAR_T delim = *p; /* '=' or '.' */
2954
_NL_CURRENT_WORD (LC_COLLATE, _NL_COLLATE_NRULES);
2959
/* If pattern is `[[=' or '[[.'. */
2960
if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2965
if ((c == delim && *p == ']') || p == pend)
2967
if (c1 < sizeof (str) - 1)
2970
/* This is in any case an invalid class name. */
2975
if (c == delim && *p == ']' && str[0] != '\0')
2977
unsigned int i, offset;
2978
/* If we have no collation data we use the default
2979
collation in which each character is in a class
2980
by itself. It also means that ASCII is the
2981
character set and therefore we cannot have character
2982
with more than one byte in the multibyte
2985
/* If not defined _LIBC, we push the name and
2986
`\0' for the sake of matching performance. */
2987
int datasize = c1 + 1;
2995
FREE_STACK_RETURN (REG_ECOLLATE);
3000
const int32_t *table;
3001
const int32_t *weights;
3002
const int32_t *extra;
3003
const int32_t *indirect;
3006
/* This #include defines a local function! */
3007
# include <locale/weightwc.h>
3011
/* We push the index for equivalence class. */
3014
table = (const int32_t *)
3015
_NL_CURRENT (LC_COLLATE,
3016
_NL_COLLATE_TABLEWC);
3017
weights = (const int32_t *)
3018
_NL_CURRENT (LC_COLLATE,
3019
_NL_COLLATE_WEIGHTWC);
3020
extra = (const int32_t *)
3021
_NL_CURRENT (LC_COLLATE,
3022
_NL_COLLATE_EXTRAWC);
3023
indirect = (const int32_t *)
3024
_NL_CURRENT (LC_COLLATE,
3025
_NL_COLLATE_INDIRECTWC);
3027
idx = findidx ((const wint_t**)&cp);
3028
if (idx == 0 || cp < (wint_t*) str + c1)
3029
/* This is no valid character. */
3030
FREE_STACK_RETURN (REG_ECOLLATE);
3032
str[0] = (wchar_t)idx;
3034
else /* delim == '.' */
3036
/* We push collation sequence value
3037
for collating symbol. */
3039
const int32_t *symb_table;
3040
const unsigned char *extra;
3047
/* We have to convert the name to a single-byte
3048
string. This is possible since the names
3049
consist of ASCII characters and the internal
3050
representation is UCS4. */
3051
for (i = 0; i < c1; ++i)
3052
char_str[i] = str[i];
3055
_NL_CURRENT_WORD (LC_COLLATE,
3056
_NL_COLLATE_SYMB_HASH_SIZEMB);
3057
symb_table = (const int32_t *)
3058
_NL_CURRENT (LC_COLLATE,
3059
_NL_COLLATE_SYMB_TABLEMB);
3060
extra = (const unsigned char *)
3061
_NL_CURRENT (LC_COLLATE,
3062
_NL_COLLATE_SYMB_EXTRAMB);
3064
/* Locate the character in the hashing table. */
3065
hash = elem_hash (char_str, c1);
3068
elem = hash % table_size;
3069
second = hash % (table_size - 2);
3070
while (symb_table[2 * elem] != 0)
3072
/* First compare the hashing value. */
3073
if (symb_table[2 * elem] == hash
3074
&& c1 == extra[symb_table[2 * elem + 1]]
3075
&& memcmp (char_str,
3076
&extra[symb_table[2 * elem + 1]
3079
/* Yep, this is the entry. */
3080
idx = symb_table[2 * elem + 1];
3081
idx += 1 + extra[idx];
3089
if (symb_table[2 * elem] != 0)
3091
/* Compute the index of the byte sequence
3093
idx += 1 + extra[idx];
3094
/* Adjust for the alignment. */
3095
idx = (idx + 3) & ~3;
3097
str[0] = (wchar_t) idx + 4;
3099
else if (symb_table[2 * elem] == 0 && c1 == 1)
3101
/* No valid character. Match it as a
3102
single byte character. */
3103
had_char_class = false;
3105
/* Update the length of characters */
3107
range_start = str[0];
3109
/* Throw away the ] at the end of the
3110
collating symbol. */
3112
/* exit from the switch block. */
3116
FREE_STACK_RETURN (REG_ECOLLATE);
3121
/* Throw away the ] at the end of the equivalence
3122
class (or collating symbol). */
3125
/* Allocate the space for the equivalence class
3126
(or collating symbol) (and '\0' if needed). */
3127
GET_BUFFER_SPACE(datasize);
3128
/* Update the pointer to indicate end of buffer. */
3132
{ /* equivalence class */
3133
/* Calculate the offset of char_ranges,
3134
which is next to equivalence_classes. */
3135
offset = laststart[1] + laststart[2]
3138
insert_space(datasize, laststart + offset, b - 1);
3140
/* Write the equivalence_class and \0. */
3141
for (i = 0 ; i < datasize ; i++)
3142
laststart[offset + i] = str[i];
3144
/* Update the length of equivalence_classes. */
3145
laststart[3] += datasize;
3146
had_char_class = true;
3148
else /* delim == '.' */
3149
{ /* collating symbol */
3150
/* Calculate the offset of the equivalence_classes,
3151
which is next to collating_symbols. */
3152
offset = laststart[1] + laststart[2] + 6;
3153
/* Insert space and write the collationg_symbol
3155
insert_space(datasize, laststart + offset, b-1);
3156
for (i = 0 ; i < datasize ; i++)
3157
laststart[offset + i] = str[i];
3159
/* In re_match_2_internal if range_start < -1, we
3160
assume -range_start is the offset of the
3161
collating symbol which is specified as
3162
the character of the range start. So we assign
3163
-(laststart[1] + laststart[2] + 6) to
3165
range_start = -(laststart[1] + laststart[2] + 6);
3166
/* Update the length of collating_symbol. */
3167
laststart[2] += datasize;
3168
had_char_class = false;
3178
laststart[5] += 2; /* Update the length of characters */
3179
range_start = delim;
3180
had_char_class = false;
3185
had_char_class = false;
3187
laststart[5]++; /* Update the length of characters */
3193
/* Ensure that we have enough space to push a charset: the
3194
opcode, the length count, and the bitset; 34 bytes in all. */
3195
GET_BUFFER_SPACE (34);
3199
/* We test `*p == '^' twice, instead of using an if
3200
statement, so we only need one BUF_PUSH. */
3201
BUF_PUSH (*p == '^' ? charset_not : charset);
3205
/* Remember the first position in the bracket expression. */
3208
/* Push the number of bytes in the bitmap. */
3209
BUF_PUSH ((1 << BYTEWIDTH) / BYTEWIDTH);
3211
/* Clear the whole map. */
3212
bzero (b, (1 << BYTEWIDTH) / BYTEWIDTH);
3214
/* charset_not matches newline according to a syntax bit. */
3215
if ((re_opcode_t) b[-2] == charset_not
3216
&& (syntax & RE_HAT_LISTS_NOT_NEWLINE))
3217
SET_LIST_BIT ('\n');
3219
/* Read in characters and ranges, setting map bits. */
3222
if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
3226
/* \ might escape characters inside [...] and [^...]. */
3227
if ((syntax & RE_BACKSLASH_ESCAPE_IN_LISTS) && c == '\\')
3229
if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
3237
/* Could be the end of the bracket expression. If it's
3238
not (i.e., when the bracket expression is `[]' so
3239
far), the ']' character bit gets set way below. */
3240
if (c == ']' && p != p1 + 1)
3243
/* Look ahead to see if it's a range when the last thing
3244
was a character class. */
3245
if (had_char_class && c == '-' && *p != ']')
3246
FREE_STACK_RETURN (REG_ERANGE);
3248
/* Look ahead to see if it's a range when the last thing
3249
was a character: if this is a hyphen not at the
3250
beginning or the end of a list, then it's the range
3253
&& !(p - 2 >= pattern && p[-2] == '[')
3254
&& !(p - 3 >= pattern && p[-3] == '[' && p[-2] == '^')
3258
= byte_compile_range (range_start, &p, pend, translate,
3260
if (ret != REG_NOERROR) FREE_STACK_RETURN (ret);
3261
range_start = 0xffffffff;
3264
else if (p[0] == '-' && p[1] != ']')
3265
{ /* This handles ranges made up of characters only. */
3268
/* Move past the `-'. */
3271
ret = byte_compile_range (c, &p, pend, translate, syntax, b);
3272
if (ret != REG_NOERROR) FREE_STACK_RETURN (ret);
3273
range_start = 0xffffffff;
3276
/* See if we're at the beginning of a possible character
3279
else if (syntax & RE_CHAR_CLASSES && c == '[' && *p == ':')
3280
{ /* Leave room for the null. */
3281
char str[CHAR_CLASS_MAX_LENGTH + 1];
3286
/* If pattern is `[[:'. */
3287
if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
3292
if ((c == ':' && *p == ']') || p == pend)
3294
if (c1 < CHAR_CLASS_MAX_LENGTH)
3297
/* This is in any case an invalid class name. */
3302
/* If isn't a word bracketed by `[:' and `:]':
3303
undo the ending character, the letters, and leave
3304
the leading `:' and `[' (but set bits for them). */
3305
if (c == ':' && *p == ']')
3307
# if defined _LIBC || WIDE_CHAR_SUPPORT
3308
boolean is_lower = STREQ (str, "lower");
3309
boolean is_upper = STREQ (str, "upper");
3313
wt = IS_CHAR_CLASS (str);
3315
FREE_STACK_RETURN (REG_ECTYPE);
3317
/* Throw away the ] at the end of the character
3321
if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
3323
for (ch = 0; ch < 1 << BYTEWIDTH; ++ch)
3326
if (__iswctype (__btowc (ch), wt))
3329
if (iswctype (btowc (ch), wt))
3333
if (translate && (is_upper || is_lower)
3334
&& (ISUPPER (ch) || ISLOWER (ch)))
3338
had_char_class = true;
3341
boolean is_alnum = STREQ (str, "alnum");
3342
boolean is_alpha = STREQ (str, "alpha");
3343
boolean is_blank = STREQ (str, "blank");
3344
boolean is_cntrl = STREQ (str, "cntrl");
3345
boolean is_digit = STREQ (str, "digit");
3346
boolean is_graph = STREQ (str, "graph");
3347
boolean is_lower = STREQ (str, "lower");
3348
boolean is_print = STREQ (str, "print");
3349
boolean is_punct = STREQ (str, "punct");
3350
boolean is_space = STREQ (str, "space");
3351
boolean is_upper = STREQ (str, "upper");
3352
boolean is_xdigit = STREQ (str, "xdigit");
3354
if (!IS_CHAR_CLASS (str))
3355
FREE_STACK_RETURN (REG_ECTYPE);
3357
/* Throw away the ] at the end of the character
3361
if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
3363
for (ch = 0; ch < 1 << BYTEWIDTH; ch++)
3365
/* This was split into 3 if's to
3366
avoid an arbitrary limit in some compiler. */
3367
if ( (is_alnum && ISALNUM (ch))
3368
|| (is_alpha && ISALPHA (ch))
3369
|| (is_blank && ISBLANK (ch))
3370
|| (is_cntrl && ISCNTRL (ch)))
3372
if ( (is_digit && ISDIGIT (ch))
3373
|| (is_graph && ISGRAPH (ch))
3374
|| (is_lower && ISLOWER (ch))
3375
|| (is_print && ISPRINT (ch)))
3377
if ( (is_punct && ISPUNCT (ch))
3378
|| (is_space && ISSPACE (ch))
3379
|| (is_upper && ISUPPER (ch))
3380
|| (is_xdigit && ISXDIGIT (ch)))
3382
if ( translate && (is_upper || is_lower)
3383
&& (ISUPPER (ch) || ISLOWER (ch)))
3386
had_char_class = true;
3387
# endif /* libc || wctype.h */
3397
had_char_class = false;
3400
else if (syntax & RE_CHAR_CLASSES && c == '[' && *p == '=')
3402
unsigned char str[MB_LEN_MAX + 1];
3405
_NL_CURRENT_WORD (LC_COLLATE, _NL_COLLATE_NRULES);
3411
/* If pattern is `[[='. */
3412
if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
3417
if ((c == '=' && *p == ']') || p == pend)
3419
if (c1 < MB_LEN_MAX)
3422
/* This is in any case an invalid class name. */
3427
if (c == '=' && *p == ']' && str[0] != '\0')
3429
/* If we have no collation data we use the default
3430
collation in which each character is in a class
3431
by itself. It also means that ASCII is the
3432
character set and therefore we cannot have character
3433
with more than one byte in the multibyte
3440
FREE_STACK_RETURN (REG_ECOLLATE);
3442
/* Throw away the ] at the end of the equivalence
3446
/* Set the bit for the character. */
3447
SET_LIST_BIT (str[0]);
3452
/* Try to match the byte sequence in `str' against
3453
those known to the collate implementation.
3454
First find out whether the bytes in `str' are
3455
actually from exactly one character. */
3456
const int32_t *table;
3457
const unsigned char *weights;
3458
const unsigned char *extra;
3459
const int32_t *indirect;
3461
const unsigned char *cp = str;
3464
/* This #include defines a local function! */
3465
# include <locale/weight.h>
3467
table = (const int32_t *)
3468
_NL_CURRENT (LC_COLLATE, _NL_COLLATE_TABLEMB);
3469
weights = (const unsigned char *)
3470
_NL_CURRENT (LC_COLLATE, _NL_COLLATE_WEIGHTMB);
3471
extra = (const unsigned char *)
3472
_NL_CURRENT (LC_COLLATE, _NL_COLLATE_EXTRAMB);
3473
indirect = (const int32_t *)
3474
_NL_CURRENT (LC_COLLATE, _NL_COLLATE_INDIRECTMB);
3476
idx = findidx (&cp);
3477
if (idx == 0 || cp < str + c1)
3478
/* This is no valid character. */
3479
FREE_STACK_RETURN (REG_ECOLLATE);
3481
/* Throw away the ] at the end of the equivalence
3485
/* Now we have to go throught the whole table
3486
and find all characters which have the same
3489
XXX Note that this is not entirely correct.
3490
we would have to match multibyte sequences
3491
but this is not possible with the current
3493
for (ch = 1; ch < 256; ++ch)
3494
/* XXX This test would have to be changed if we
3495
would allow matching multibyte sequences. */
3498
int32_t idx2 = table[ch];
3499
size_t len = weights[idx2];
3501
/* Test whether the lenghts match. */
3502
if (weights[idx] == len)
3504
/* They do. New compare the bytes of
3509
&& (weights[idx + 1 + cnt]
3510
== weights[idx2 + 1 + cnt]))
3514
/* They match. Mark the character as
3521
had_char_class = true;
3531
had_char_class = false;
3534
else if (syntax & RE_CHAR_CLASSES && c == '[' && *p == '.')
3536
unsigned char str[128]; /* Should be large enough. */
3539
_NL_CURRENT_WORD (LC_COLLATE, _NL_COLLATE_NRULES);
3545
/* If pattern is `[[.'. */
3546
if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
3551
if ((c == '.' && *p == ']') || p == pend)
3553
if (c1 < sizeof (str))
3556
/* This is in any case an invalid class name. */
3561
if (c == '.' && *p == ']' && str[0] != '\0')
3563
/* If we have no collation data we use the default
3564
collation in which each character is the name
3565
for its own class which contains only the one
3566
character. It also means that ASCII is the
3567
character set and therefore we cannot have character
3568
with more than one byte in the multibyte
3575
FREE_STACK_RETURN (REG_ECOLLATE);
3577
/* Throw away the ] at the end of the equivalence
3581
/* Set the bit for the character. */
3582
SET_LIST_BIT (str[0]);
3583
range_start = ((const unsigned char *) str)[0];
3588
/* Try to match the byte sequence in `str' against
3589
those known to the collate implementation.
3590
First find out whether the bytes in `str' are
3591
actually from exactly one character. */
3593
const int32_t *symb_table;
3594
const unsigned char *extra;
3601
_NL_CURRENT_WORD (LC_COLLATE,
3602
_NL_COLLATE_SYMB_HASH_SIZEMB);
3603
symb_table = (const int32_t *)
3604
_NL_CURRENT (LC_COLLATE,
3605
_NL_COLLATE_SYMB_TABLEMB);
3606
extra = (const unsigned char *)
3607
_NL_CURRENT (LC_COLLATE,
3608
_NL_COLLATE_SYMB_EXTRAMB);
3610
/* Locate the character in the hashing table. */
3611
hash = elem_hash (str, c1);
3614
elem = hash % table_size;
3615
second = hash % (table_size - 2);
3616
while (symb_table[2 * elem] != 0)
3618
/* First compare the hashing value. */
3619
if (symb_table[2 * elem] == hash
3620
&& c1 == extra[symb_table[2 * elem + 1]]
3622
&extra[symb_table[2 * elem + 1]
3626
/* Yep, this is the entry. */
3627
idx = symb_table[2 * elem + 1];
3628
idx += 1 + extra[idx];
3636
if (symb_table[2 * elem] == 0)
3637
/* This is no valid character. */
3638
FREE_STACK_RETURN (REG_ECOLLATE);
3640
/* Throw away the ] at the end of the equivalence
3644
/* Now add the multibyte character(s) we found
3647
XXX Note that this is not entirely correct.
3648
we would have to match multibyte sequences
3649
but this is not possible with the current
3650
implementation. Also, we have to match
3651
collating symbols, which expand to more than
3652
one file, as a whole and not allow the
3653
individual bytes. */
3656
range_start = extra[idx];
3659
SET_LIST_BIT (extra[idx]);
3664
had_char_class = false;
3674
had_char_class = false;
3679
had_char_class = false;
3685
/* Discard any (non)matching list bytes that are all 0 at the
3686
end of the map. Decrease the map-length byte too. */
3687
while ((int) b[-1] > 0 && b[b[-1] - 1] == 0)
3696
if (syntax & RE_NO_BK_PARENS)
3703
if (syntax & RE_NO_BK_PARENS)
3710
if (syntax & RE_NEWLINE_ALT)
3717
if (syntax & RE_NO_BK_VBAR)
3724
if (syntax & RE_INTERVALS && syntax & RE_NO_BK_BRACES)
3725
goto handle_interval;
3731
if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
3733
/* Do not translate the character after the \, so that we can
3734
distinguish, e.g., \B from \b, even if we normally would
3735
translate, e.g., B to b. */
3741
if (syntax & RE_NO_BK_PARENS)
3742
goto normal_backslash;
3748
if (COMPILE_STACK_FULL)
3750
RETALLOC (compile_stack.stack, compile_stack.size << 1,
3751
compile_stack_elt_t);
3752
if (compile_stack.stack == NULL) return REG_ESPACE;
3754
compile_stack.size <<= 1;
3757
/* These are the values to restore when we hit end of this
3758
group. They are all relative offsets, so that if the
3759
whole pattern moves because of realloc, they will still
3761
COMPILE_STACK_TOP.begalt_offset = begalt - COMPILED_BUFFER_VAR;
3762
COMPILE_STACK_TOP.fixup_alt_jump
3763
= fixup_alt_jump ? fixup_alt_jump - COMPILED_BUFFER_VAR + 1 : 0;
3764
COMPILE_STACK_TOP.laststart_offset = b - COMPILED_BUFFER_VAR;
3765
COMPILE_STACK_TOP.regnum = regnum;
3767
/* We will eventually replace the 0 with the number of
3768
groups inner to this one. But do not push a
3769
start_memory for groups beyond the last one we can
3770
represent in the compiled pattern. */
3771
if (regnum <= MAX_REGNUM)
3773
COMPILE_STACK_TOP.inner_group_offset = b
3774
- COMPILED_BUFFER_VAR + 2;
3775
BUF_PUSH_3 (start_memory, regnum, 0);
3778
compile_stack.avail++;
3783
/* If we've reached MAX_REGNUM groups, then this open
3784
won't actually generate any code, so we'll have to
3785
clear pending_exact explicitly. */
3791
if (syntax & RE_NO_BK_PARENS) goto normal_backslash;
3793
if (COMPILE_STACK_EMPTY)
3795
if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD)
3796
goto normal_backslash;
3798
FREE_STACK_RETURN (REG_ERPAREN);
3803
{ /* Push a dummy failure point at the end of the
3804
alternative for a possible future
3805
`pop_failure_jump' to pop. See comments at
3806
`push_dummy_failure' in `re_match_2'. */
3807
BUF_PUSH (push_dummy_failure);
3809
/* We allocated space for this jump when we assigned
3810
to `fixup_alt_jump', in the `handle_alt' case below. */
3811
STORE_JUMP (jump_past_alt, fixup_alt_jump, b - 1);
3814
/* See similar code for backslashed left paren above. */
3815
if (COMPILE_STACK_EMPTY)
3817
if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD)
3820
FREE_STACK_RETURN (REG_ERPAREN);
3823
/* Since we just checked for an empty stack above, this
3824
``can't happen''. */
3825
assert (compile_stack.avail != 0);
3827
/* We don't just want to restore into `regnum', because
3828
later groups should continue to be numbered higher,
3829
as in `(ab)c(de)' -- the second group is #2. */
3830
regnum_t this_group_regnum;
3832
compile_stack.avail--;
3833
begalt = COMPILED_BUFFER_VAR + COMPILE_STACK_TOP.begalt_offset;
3835
= COMPILE_STACK_TOP.fixup_alt_jump
3836
? COMPILED_BUFFER_VAR + COMPILE_STACK_TOP.fixup_alt_jump - 1
3838
laststart = COMPILED_BUFFER_VAR + COMPILE_STACK_TOP.laststart_offset;
3839
this_group_regnum = COMPILE_STACK_TOP.regnum;
3840
/* If we've reached MAX_REGNUM groups, then this open
3841
won't actually generate any code, so we'll have to
3842
clear pending_exact explicitly. */
3845
/* We're at the end of the group, so now we know how many
3846
groups were inside this one. */
3847
if (this_group_regnum <= MAX_REGNUM)
3849
UCHAR_T *inner_group_loc
3850
= COMPILED_BUFFER_VAR + COMPILE_STACK_TOP.inner_group_offset;
3852
*inner_group_loc = regnum - this_group_regnum;
3853
BUF_PUSH_3 (stop_memory, this_group_regnum,
3854
regnum - this_group_regnum);
3860
case '|': /* `\|'. */
3861
if (syntax & RE_LIMITED_OPS || syntax & RE_NO_BK_VBAR)
3862
goto normal_backslash;
3864
if (syntax & RE_LIMITED_OPS)
3867
/* Insert before the previous alternative a jump which
3868
jumps to this alternative if the former fails. */
3869
GET_BUFFER_SPACE (1 + OFFSET_ADDRESS_SIZE);
3870
INSERT_JUMP (on_failure_jump, begalt,
3871
b + 2 + 2 * OFFSET_ADDRESS_SIZE);
3873
b += 1 + OFFSET_ADDRESS_SIZE;
3875
/* The alternative before this one has a jump after it
3876
which gets executed if it gets matched. Adjust that
3877
jump so it will jump to this alternative's analogous
3878
jump (put in below, which in turn will jump to the next
3879
(if any) alternative's such jump, etc.). The last such
3880
jump jumps to the correct final destination. A picture:
3886
If we are at `b', then fixup_alt_jump right now points to a
3887
three-byte space after `a'. We'll put in the jump, set
3888
fixup_alt_jump to right after `b', and leave behind three
3889
bytes which we'll fill in when we get to after `c'. */
3892
STORE_JUMP (jump_past_alt, fixup_alt_jump, b);
3894
/* Mark and leave space for a jump after this alternative,
3895
to be filled in later either by next alternative or
3896
when know we're at the end of a series of alternatives. */
3898
GET_BUFFER_SPACE (1 + OFFSET_ADDRESS_SIZE);
3899
b += 1 + OFFSET_ADDRESS_SIZE;
3907
/* If \{ is a literal. */
3908
if (!(syntax & RE_INTERVALS)
3909
/* If we're at `\{' and it's not the open-interval
3911
|| (syntax & RE_NO_BK_BRACES))
3912
goto normal_backslash;
3916
/* If got here, then the syntax allows intervals. */
3918
/* At least (most) this many matches must be made. */
3919
int lower_bound = -1, upper_bound = -1;
3921
/* Place in the uncompiled pattern (i.e., just after
3922
the '{') to go back to if the interval is invalid. */
3923
const CHAR_T *beg_interval = p;
3926
goto invalid_interval;
3928
GET_UNSIGNED_NUMBER (lower_bound);
3932
GET_UNSIGNED_NUMBER (upper_bound);
3933
if (upper_bound < 0)
3934
upper_bound = RE_DUP_MAX;
3937
/* Interval such as `{1}' => match exactly once. */
3938
upper_bound = lower_bound;
3940
if (! (0 <= lower_bound && lower_bound <= upper_bound))
3941
goto invalid_interval;
3943
if (!(syntax & RE_NO_BK_BRACES))
3945
if (c != '\\' || p == pend)
3946
goto invalid_interval;
3951
goto invalid_interval;
3953
/* If it's invalid to have no preceding re. */
3956
if (syntax & RE_CONTEXT_INVALID_OPS
3957
&& !(syntax & RE_INVALID_INTERVAL_ORD))
3958
FREE_STACK_RETURN (REG_BADRPT);
3959
else if (syntax & RE_CONTEXT_INDEP_OPS)
3962
goto unfetch_interval;
3965
/* We just parsed a valid interval. */
3967
if (RE_DUP_MAX < upper_bound)
3968
FREE_STACK_RETURN (REG_BADBR);
3970
/* If the upper bound is zero, don't want to succeed at
3971
all; jump from `laststart' to `b + 3', which will be
3972
the end of the buffer after we insert the jump. */
3973
/* ifdef WCHAR, 'b + 1 + OFFSET_ADDRESS_SIZE'
3974
instead of 'b + 3'. */
3975
if (upper_bound == 0)
3977
GET_BUFFER_SPACE (1 + OFFSET_ADDRESS_SIZE);
3978
INSERT_JUMP (jump, laststart, b + 1
3979
+ OFFSET_ADDRESS_SIZE);
3980
b += 1 + OFFSET_ADDRESS_SIZE;
3983
/* Otherwise, we have a nontrivial interval. When
3984
we're all done, the pattern will look like:
3985
set_number_at <jump count> <upper bound>
3986
set_number_at <succeed_n count> <lower bound>
3987
succeed_n <after jump addr> <succeed_n count>
3989
jump_n <succeed_n addr> <jump count>
3990
(The upper bound and `jump_n' are omitted if
3991
`upper_bound' is 1, though.) */
3993
{ /* If the upper bound is > 1, we need to insert
3994
more at the end of the loop. */
3995
unsigned nbytes = 2 + 4 * OFFSET_ADDRESS_SIZE +
3996
(upper_bound > 1) * (2 + 4 * OFFSET_ADDRESS_SIZE);
3998
GET_BUFFER_SPACE (nbytes);
4000
/* Initialize lower bound of the `succeed_n', even
4001
though it will be set during matching by its
4002
attendant `set_number_at' (inserted next),
4003
because `re_compile_fastmap' needs to know.
4004
Jump to the `jump_n' we might insert below. */
4005
INSERT_JUMP2 (succeed_n, laststart,
4006
b + 1 + 2 * OFFSET_ADDRESS_SIZE
4007
+ (upper_bound > 1) * (1 + 2 * OFFSET_ADDRESS_SIZE)
4009
b += 1 + 2 * OFFSET_ADDRESS_SIZE;
4011
/* Code to initialize the lower bound. Insert
4012
before the `succeed_n'. The `5' is the last two
4013
bytes of this `set_number_at', plus 3 bytes of
4014
the following `succeed_n'. */
4015
/* ifdef WCHAR, The '1+2*OFFSET_ADDRESS_SIZE'
4016
is the 'set_number_at', plus '1+OFFSET_ADDRESS_SIZE'
4017
of the following `succeed_n'. */
4018
PREFIX(insert_op2) (set_number_at, laststart, 1
4019
+ 2 * OFFSET_ADDRESS_SIZE, lower_bound, b);
4020
b += 1 + 2 * OFFSET_ADDRESS_SIZE;
4022
if (upper_bound > 1)
4023
{ /* More than one repetition is allowed, so
4024
append a backward jump to the `succeed_n'
4025
that starts this interval.
4027
When we've reached this during matching,
4028
we'll have matched the interval once, so
4029
jump back only `upper_bound - 1' times. */
4030
STORE_JUMP2 (jump_n, b, laststart
4031
+ 2 * OFFSET_ADDRESS_SIZE + 1,
4033
b += 1 + 2 * OFFSET_ADDRESS_SIZE;
4035
/* The location we want to set is the second
4036
parameter of the `jump_n'; that is `b-2' as
4037
an absolute address. `laststart' will be
4038
the `set_number_at' we're about to insert;
4039
`laststart+3' the number to set, the source
4040
for the relative address. But we are
4041
inserting into the middle of the pattern --
4042
so everything is getting moved up by 5.
4043
Conclusion: (b - 2) - (laststart + 3) + 5,
4044
i.e., b - laststart.
4046
We insert this at the beginning of the loop
4047
so that if we fail during matching, we'll
4048
reinitialize the bounds. */
4049
PREFIX(insert_op2) (set_number_at, laststart,
4051
upper_bound - 1, b);
4052
b += 1 + 2 * OFFSET_ADDRESS_SIZE;
4059
if (!(syntax & RE_INVALID_INTERVAL_ORD))
4060
FREE_STACK_RETURN (p == pend ? REG_EBRACE : REG_BADBR);
4062
/* Match the characters as literals. */
4065
if (syntax & RE_NO_BK_BRACES)
4068
goto normal_backslash;
4072
/* There is no way to specify the before_dot and after_dot
4073
operators. rms says this is ok. --karl */
4081
BUF_PUSH_2 (syntaxspec, syntax_spec_code[c]);
4087
BUF_PUSH_2 (notsyntaxspec, syntax_spec_code[c]);
4093
if (syntax & RE_NO_GNU_OPS)
4096
BUF_PUSH (wordchar);
4101
if (syntax & RE_NO_GNU_OPS)
4104
BUF_PUSH (notwordchar);
4109
if (syntax & RE_NO_GNU_OPS)
4115
if (syntax & RE_NO_GNU_OPS)
4121
if (syntax & RE_NO_GNU_OPS)
4123
BUF_PUSH (wordbound);
4127
if (syntax & RE_NO_GNU_OPS)
4129
BUF_PUSH (notwordbound);
4133
if (syntax & RE_NO_GNU_OPS)
4139
if (syntax & RE_NO_GNU_OPS)
4144
case '1': case '2': case '3': case '4': case '5':
4145
case '6': case '7': case '8': case '9':
4146
if (syntax & RE_NO_BK_REFS)
4152
FREE_STACK_RETURN (REG_ESUBREG);
4154
/* Can't back reference to a subexpression if inside of it. */
4155
if (group_in_compile_stack (compile_stack, (regnum_t) c1))
4159
BUF_PUSH_2 (duplicate, c1);
4165
if (syntax & RE_BK_PLUS_QM)
4168
goto normal_backslash;
4172
/* You might think it would be useful for \ to mean
4173
not to translate; but if we don't translate it
4174
it will never match anything. */
4182
/* Expects the character in `c'. */
4184
/* If no exactn currently being built. */
4187
/* If last exactn handle binary(or character) and
4188
new exactn handle character(or binary). */
4189
|| is_exactn_bin != is_binary[p - 1 - pattern]
4192
/* If last exactn not at current position. */
4193
|| pending_exact + *pending_exact + 1 != b
4195
/* We have only one byte following the exactn for the count. */
4196
|| *pending_exact == (1 << BYTEWIDTH) - 1
4198
/* If followed by a repetition operator. */
4199
|| *p == '*' || *p == '^'
4200
|| ((syntax & RE_BK_PLUS_QM)
4201
? *p == '\\' && (p[1] == '+' || p[1] == '?')
4202
: (*p == '+' || *p == '?'))
4203
|| ((syntax & RE_INTERVALS)
4204
&& ((syntax & RE_NO_BK_BRACES)
4206
: (p[0] == '\\' && p[1] == '{'))))
4208
/* Start building a new exactn. */
4213
/* Is this exactn binary data or character? */
4214
is_exactn_bin = is_binary[p - 1 - pattern];
4216
BUF_PUSH_2 (exactn_bin, 0);
4218
BUF_PUSH_2 (exactn, 0);
4220
BUF_PUSH_2 (exactn, 0);
4222
pending_exact = b - 1;
4229
} /* while p != pend */
4232
/* Through the pattern now. */
4235
STORE_JUMP (jump_past_alt, fixup_alt_jump, b);
4237
if (!COMPILE_STACK_EMPTY)
4238
FREE_STACK_RETURN (REG_EPAREN);
4240
/* If we don't want backtracking, force success
4241
the first time we reach the end of the compiled pattern. */
4242
if (syntax & RE_NO_POSIX_BACKTRACKING)
4250
free (compile_stack.stack);
4252
/* We have succeeded; set the length of the buffer. */
4254
bufp->used = (uintptr_t) b - (uintptr_t) COMPILED_BUFFER_VAR;
4256
bufp->used = b - bufp->buffer;
4262
DEBUG_PRINT1 ("\nCompiled pattern: \n");
4263
PREFIX(print_compiled_pattern) (bufp);
4267
#ifndef MATCH_MAY_ALLOCATE
4268
/* Initialize the failure stack to the largest possible stack. This
4269
isn't necessary unless we're trying to avoid calling alloca in
4270
the search and match routines. */
4272
int num_regs = bufp->re_nsub + 1;
4274
/* Since DOUBLE_FAIL_STACK refuses to double only if the current size
4275
is strictly greater than re_max_failures, the largest possible stack
4276
is 2 * re_max_failures failure points. */
4277
if (fail_stack.size < (2 * re_max_failures * MAX_FAILURE_ITEMS))
4279
fail_stack.size = (2 * re_max_failures * MAX_FAILURE_ITEMS);
4282
if (! fail_stack.stack)
4284
= (PREFIX(fail_stack_elt_t) *) xmalloc (fail_stack.size
4285
* sizeof (PREFIX(fail_stack_elt_t)));
4288
= (PREFIX(fail_stack_elt_t) *) xrealloc (fail_stack.stack,
4290
* sizeof (PREFIX(fail_stack_elt_t))));
4291
# else /* not emacs */
4292
if (! fail_stack.stack)
4294
= (PREFIX(fail_stack_elt_t) *) malloc (fail_stack.size
4295
* sizeof (PREFIX(fail_stack_elt_t)));
4298
= (PREFIX(fail_stack_elt_t) *) realloc (fail_stack.stack,
4300
* sizeof (PREFIX(fail_stack_elt_t))));
4301
# endif /* not emacs */
4304
PREFIX(regex_grow_registers) (num_regs);
4306
#endif /* not MATCH_MAY_ALLOCATE */
4309
} /* regex_compile */
4311
/* Subroutines for `regex_compile'. */
4313
/* Store OP at LOC followed by two-byte integer parameter ARG. */
4314
/* ifdef WCHAR, integer parameter is 1 wchar_t. */
4317
PREFIX(store_op1) (op, loc, arg)
4322
*loc = (UCHAR_T) op;
4323
STORE_NUMBER (loc + 1, arg);
4327
/* Like `store_op1', but for two two-byte parameters ARG1 and ARG2. */
4328
/* ifdef WCHAR, integer parameter is 1 wchar_t. */
4331
PREFIX(store_op2) (op, loc, arg1, arg2)
4336
*loc = (UCHAR_T) op;
4337
STORE_NUMBER (loc + 1, arg1);
4338
STORE_NUMBER (loc + 1 + OFFSET_ADDRESS_SIZE, arg2);
4342
/* Copy the bytes from LOC to END to open up three bytes of space at LOC
4343
for OP followed by two-byte integer parameter ARG. */
4344
/* ifdef WCHAR, integer parameter is 1 wchar_t. */
4347
PREFIX(insert_op1) (op, loc, arg, end)
4353
register UCHAR_T *pfrom = end;
4354
register UCHAR_T *pto = end + 1 + OFFSET_ADDRESS_SIZE;
4356
while (pfrom != loc)
4359
PREFIX(store_op1) (op, loc, arg);
4363
/* Like `insert_op1', but for two two-byte parameters ARG1 and ARG2. */
4364
/* ifdef WCHAR, integer parameter is 1 wchar_t. */
4367
PREFIX(insert_op2) (op, loc, arg1, arg2, end)
4373
register UCHAR_T *pfrom = end;
4374
register UCHAR_T *pto = end + 1 + 2 * OFFSET_ADDRESS_SIZE;
4376
while (pfrom != loc)
4379
PREFIX(store_op2) (op, loc, arg1, arg2);
4383
/* P points to just after a ^ in PATTERN. Return true if that ^ comes
4384
after an alternative or a begin-subexpression. We assume there is at
4385
least one character before the ^. */
4388
PREFIX(at_begline_loc_p) (pattern, p, syntax)
4389
const CHAR_T *pattern, *p;
4390
reg_syntax_t syntax;
4392
const CHAR_T *prev = p - 2;
4393
boolean prev_prev_backslash = prev > pattern && prev[-1] == '\\';
4396
/* After a subexpression? */
4397
(*prev == '(' && (syntax & RE_NO_BK_PARENS || prev_prev_backslash))
4398
/* After an alternative? */
4399
|| (*prev == '|' && (syntax & RE_NO_BK_VBAR || prev_prev_backslash));
4403
/* The dual of at_begline_loc_p. This one is for $. We assume there is
4404
at least one character after the $, i.e., `P < PEND'. */
4407
PREFIX(at_endline_loc_p) (p, pend, syntax)
4408
const CHAR_T *p, *pend;
4409
reg_syntax_t syntax;
4411
const CHAR_T *next = p;
4412
boolean next_backslash = *next == '\\';
4413
const CHAR_T *next_next = p + 1 < pend ? p + 1 : 0;
4416
/* Before a subexpression? */
4417
(syntax & RE_NO_BK_PARENS ? *next == ')'
4418
: next_backslash && next_next && *next_next == ')')
4419
/* Before an alternative? */
4420
|| (syntax & RE_NO_BK_VBAR ? *next == '|'
4421
: next_backslash && next_next && *next_next == '|');
4424
#else /* not INSIDE_RECURSION */
4426
/* Returns true if REGNUM is in one of COMPILE_STACK's elements and
4427
false if it's not. */
4430
group_in_compile_stack (compile_stack, regnum)
4431
compile_stack_type compile_stack;
4436
for (this_element = compile_stack.avail - 1;
4439
if (compile_stack.stack[this_element].regnum == regnum)
4444
#endif /* not INSIDE_RECURSION */
4446
#ifdef INSIDE_RECURSION
4449
/* This insert space, which size is "num", into the pattern at "loc".
4450
"end" must point the end of the allocated buffer. */
4452
insert_space (num, loc, end)
4457
register CHAR_T *pto = end;
4458
register CHAR_T *pfrom = end - num;
4460
while (pfrom >= loc)
4466
static reg_errcode_t
4467
wcs_compile_range (range_start_char, p_ptr, pend, translate, syntax, b,
4469
CHAR_T range_start_char;
4470
const CHAR_T **p_ptr, *pend;
4471
CHAR_T *char_set, *b;
4472
RE_TRANSLATE_TYPE translate;
4473
reg_syntax_t syntax;
4475
const CHAR_T *p = *p_ptr;
4476
CHAR_T range_start, range_end;
4480
uint32_t start_val, end_val;
4486
nrules = _NL_CURRENT_WORD (LC_COLLATE, _NL_COLLATE_NRULES);
4489
const char *collseq = (const char *) _NL_CURRENT(LC_COLLATE,
4490
_NL_COLLATE_COLLSEQWC);
4491
const unsigned char *extra = (const unsigned char *)
4492
_NL_CURRENT (LC_COLLATE, _NL_COLLATE_SYMB_EXTRAMB);
4494
if (range_start_char < -1)
4496
/* range_start is a collating symbol. */
4498
/* Retreive the index and get collation sequence value. */
4499
wextra = (int32_t*)(extra + char_set[-range_start_char]);
4500
start_val = wextra[1 + *wextra];
4503
start_val = collseq_table_lookup(collseq, TRANSLATE(range_start_char));
4505
end_val = collseq_table_lookup (collseq, TRANSLATE (p[0]));
4507
/* Report an error if the range is empty and the syntax prohibits
4509
ret = ((syntax & RE_NO_EMPTY_RANGES)
4510
&& (start_val > end_val))? REG_ERANGE : REG_NOERROR;
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) = (wchar_t)start_val;
4515
*(b - char_set[5] - 1) = (wchar_t)end_val;
4516
char_set[4]++; /* ranges_index */
4521
range_start = (range_start_char >= 0)? TRANSLATE (range_start_char):
4523
range_end = TRANSLATE (p[0]);
4524
/* Report an error if the range is empty and the syntax prohibits
4526
ret = ((syntax & RE_NO_EMPTY_RANGES)
4527
&& (range_start > range_end))? REG_ERANGE : REG_NOERROR;
4529
/* Insert space to the end of the char_ranges. */
4530
insert_space(2, b - char_set[5] - 2, b - 1);
4531
*(b - char_set[5] - 2) = range_start;
4532
*(b - char_set[5] - 1) = range_end;
4533
char_set[4]++; /* ranges_index */
4535
/* Have to increment the pointer into the pattern string, so the
4536
caller isn't still at the ending character. */
4542
/* Read the ending character of a range (in a bracket expression) from the
4543
uncompiled pattern *P_PTR (which ends at PEND). We assume the
4544
starting character is in `P[-2]'. (`P[-1]' is the character `-'.)
4545
Then we set the translation of all bits between the starting and
4546
ending characters (inclusive) in the compiled pattern B.
4548
Return an error code.
4550
We use these short variable names so we can use the same macros as
4551
`regex_compile' itself. */
4553
static reg_errcode_t
4554
byte_compile_range (range_start_char, p_ptr, pend, translate, syntax, b)
4555
unsigned int range_start_char;
4556
const char **p_ptr, *pend;
4557
RE_TRANSLATE_TYPE translate;
4558
reg_syntax_t syntax;
4562
const char *p = *p_ptr;
4565
const unsigned char *collseq;
4566
unsigned int start_colseq;
4567
unsigned int end_colseq;
4575
/* Have to increment the pointer into the pattern string, so the
4576
caller isn't still at the ending character. */
4579
/* Report an error if the range is empty and the syntax prohibits this. */
4580
ret = syntax & RE_NO_EMPTY_RANGES ? REG_ERANGE : REG_NOERROR;
4583
collseq = (const unsigned char *) _NL_CURRENT (LC_COLLATE,
4584
_NL_COLLATE_COLLSEQMB);
4586
start_colseq = collseq[(unsigned char) TRANSLATE (range_start_char)];
4587
end_colseq = collseq[(unsigned char) TRANSLATE (p[0])];
4588
for (this_char = 0; this_char <= (unsigned char) -1; ++this_char)
4590
unsigned int this_colseq = collseq[(unsigned char) TRANSLATE (this_char)];
4592
if (start_colseq <= this_colseq && this_colseq <= end_colseq)
4594
SET_LIST_BIT (TRANSLATE (this_char));
4599
/* Here we see why `this_char' has to be larger than an `unsigned
4600
char' -- we would otherwise go into an infinite loop, since all
4601
characters <= 0xff. */
4602
range_start_char = TRANSLATE (range_start_char);
4603
/* TRANSLATE(p[0]) is casted to char (not unsigned char) in TRANSLATE,
4604
and some compilers cast it to int implicitly, so following for_loop
4605
may fall to (almost) infinite loop.
4606
e.g. If translate[p[0]] = 0xff, end_char may equals to 0xffffffff.
4607
To avoid this, we cast p[0] to unsigned int and truncate it. */
4608
end_char = ((unsigned)TRANSLATE(p[0]) & ((1 << BYTEWIDTH) - 1));
4610
for (this_char = range_start_char; this_char <= end_char; ++this_char)
4612
SET_LIST_BIT (TRANSLATE (this_char));
4621
/* re_compile_fastmap computes a ``fastmap'' for the compiled pattern in
4622
BUFP. A fastmap records which of the (1 << BYTEWIDTH) possible
4623
characters can start a string that matches the pattern. This fastmap
4624
is used by re_search to skip quickly over impossible starting points.
4626
The caller must supply the address of a (1 << BYTEWIDTH)-byte data
4627
area as BUFP->fastmap.
4629
We set the `fastmap', `fastmap_accurate', and `can_be_null' fields in
4632
Returns 0 if we succeed, -2 if an internal error. */
4635
/* local function for re_compile_fastmap.
4636
truncate wchar_t character to char. */
4637
static unsigned char truncate_wchar (CHAR_T c);
4639
static unsigned char
4643
unsigned char buf[MB_CUR_MAX];
4646
memset (&state, '\0', sizeof (state));
4648
retval = __wcrtomb (buf, c, &state);
4650
retval = wcrtomb (buf, c, &state);
4652
return retval > 0 ? buf[0] : (unsigned char) c;
4657
PREFIX(re_compile_fastmap) (bufp)
4658
struct re_pattern_buffer *bufp;
4661
#ifdef MATCH_MAY_ALLOCATE
4662
PREFIX(fail_stack_type) fail_stack;
4664
#ifndef REGEX_MALLOC
4668
register char *fastmap = bufp->fastmap;
4671
/* We need to cast pattern to (wchar_t*), because we casted this compiled
4672
pattern to (char*) in regex_compile. */
4673
UCHAR_T *pattern = (UCHAR_T*)bufp->buffer;
4674
register UCHAR_T *pend = (UCHAR_T*) (bufp->buffer + bufp->used);
4676
UCHAR_T *pattern = bufp->buffer;
4677
register UCHAR_T *pend = pattern + bufp->used;
4679
UCHAR_T *p = pattern;
4682
/* This holds the pointer to the failure stack, when
4683
it is allocated relocatably. */
4684
fail_stack_elt_t *failure_stack_ptr;
4687
/* Assume that each path through the pattern can be null until
4688
proven otherwise. We set this false at the bottom of switch
4689
statement, to which we get only if a particular path doesn't
4690
match the empty string. */
4691
boolean path_can_be_null = true;
4693
/* We aren't doing a `succeed_n' to begin with. */
4694
boolean succeed_n_p = false;
4696
assert (fastmap != NULL && p != NULL);
4699
bzero (fastmap, 1 << BYTEWIDTH); /* Assume nothing's valid. */
4700
bufp->fastmap_accurate = 1; /* It will be when we're done. */
4701
bufp->can_be_null = 0;
4705
if (p == pend || *p == succeed)
4707
/* We have reached the (effective) end of pattern. */
4708
if (!FAIL_STACK_EMPTY ())
4710
bufp->can_be_null |= path_can_be_null;
4712
/* Reset for next path. */
4713
path_can_be_null = true;
4715
p = fail_stack.stack[--fail_stack.avail].pointer;
4723
/* We should never be about to go beyond the end of the pattern. */
4726
switch (SWITCH_ENUM_CAST ((re_opcode_t) *p++))
4729
/* I guess the idea here is to simply not bother with a fastmap
4730
if a backreference is used, since it's too hard to figure out
4731
the fastmap for the corresponding group. Setting
4732
`can_be_null' stops `re_search_2' from using the fastmap, so
4733
that is all we do. */
4735
bufp->can_be_null = 1;
4739
/* Following are the cases which match a character. These end
4744
fastmap[truncate_wchar(p[1])] = 1;
4758
/* It is hard to distinguish fastmap from (multi byte) characters
4759
which depends on current locale. */
4764
bufp->can_be_null = 1;
4768
for (j = *p++ * BYTEWIDTH - 1; j >= 0; j--)
4769
if (p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH)))
4775
/* Chars beyond end of map must be allowed. */
4776
for (j = *p * BYTEWIDTH; j < (1 << BYTEWIDTH); j++)
4779
for (j = *p++ * BYTEWIDTH - 1; j >= 0; j--)
4780
if (!(p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH))))
4786
for (j = 0; j < (1 << BYTEWIDTH); j++)
4787
if (SYNTAX (j) == Sword)
4793
for (j = 0; j < (1 << BYTEWIDTH); j++)
4794
if (SYNTAX (j) != Sword)
4801
int fastmap_newline = fastmap['\n'];
4803
/* `.' matches anything ... */
4804
for (j = 0; j < (1 << BYTEWIDTH); j++)
4807
/* ... except perhaps newline. */
4808
if (!(bufp->syntax & RE_DOT_NEWLINE))
4809
fastmap['\n'] = fastmap_newline;
4811
/* Return if we have already set `can_be_null'; if we have,
4812
then the fastmap is irrelevant. Something's wrong here. */
4813
else if (bufp->can_be_null)
4816
/* Otherwise, have to check alternative paths. */
4823
for (j = 0; j < (1 << BYTEWIDTH); j++)
4824
if (SYNTAX (j) == (enum syntaxcode) k)
4831
for (j = 0; j < (1 << BYTEWIDTH); j++)
4832
if (SYNTAX (j) != (enum syntaxcode) k)
4837
/* All cases after this match the empty string. These end with
4857
case push_dummy_failure:
4862
case pop_failure_jump:
4863
case maybe_pop_jump:
4866
case dummy_failure_jump:
4867
EXTRACT_NUMBER_AND_INCR (j, p);
4872
/* Jump backward implies we just went through the body of a
4873
loop and matched nothing. Opcode jumped to should be
4874
`on_failure_jump' or `succeed_n'. Just treat it like an
4875
ordinary jump. For a * loop, it has pushed its failure
4876
point already; if so, discard that as redundant. */
4877
if ((re_opcode_t) *p != on_failure_jump
4878
&& (re_opcode_t) *p != succeed_n)
4882
EXTRACT_NUMBER_AND_INCR (j, p);
4885
/* If what's on the stack is where we are now, pop it. */
4886
if (!FAIL_STACK_EMPTY ()
4887
&& fail_stack.stack[fail_stack.avail - 1].pointer == p)
4893
case on_failure_jump:
4894
case on_failure_keep_string_jump:
4895
handle_on_failure_jump:
4896
EXTRACT_NUMBER_AND_INCR (j, p);
4898
/* For some patterns, e.g., `(a?)?', `p+j' here points to the
4899
end of the pattern. We don't want to push such a point,
4900
since when we restore it above, entering the switch will
4901
increment `p' past the end of the pattern. We don't need
4902
to push such a point since we obviously won't find any more
4903
fastmap entries beyond `pend'. Such a pattern can match
4904
the null string, though. */
4907
if (!PUSH_PATTERN_OP (p + j, fail_stack))
4909
RESET_FAIL_STACK ();
4914
bufp->can_be_null = 1;
4918
EXTRACT_NUMBER_AND_INCR (k, p); /* Skip the n. */
4919
succeed_n_p = false;
4926
/* Get to the number of times to succeed. */
4927
p += OFFSET_ADDRESS_SIZE;
4929
/* Increment p past the n for when k != 0. */
4930
EXTRACT_NUMBER_AND_INCR (k, p);
4933
p -= 2 * OFFSET_ADDRESS_SIZE;
4934
succeed_n_p = true; /* Spaghetti code alert. */
4935
goto handle_on_failure_jump;
4941
p += 2 * OFFSET_ADDRESS_SIZE;
4952
abort (); /* We have listed all the cases. */
4955
/* Getting here means we have found the possible starting
4956
characters for one path of the pattern -- and that the empty
4957
string does not match. We need not follow this path further.
4958
Instead, look at the next alternative (remembered on the
4959
stack), or quit if no more. The test at the top of the loop
4960
does these things. */
4961
path_can_be_null = false;
4965
/* Set `can_be_null' for the last path (also the first path, if the
4966
pattern is empty). */
4967
bufp->can_be_null |= path_can_be_null;
4970
RESET_FAIL_STACK ();
4974
#else /* not INSIDE_RECURSION */
4977
re_compile_fastmap (bufp)
4978
struct re_pattern_buffer *bufp;
4981
if (MB_CUR_MAX != 1)
4982
return wcs_re_compile_fastmap(bufp);
4985
return byte_re_compile_fastmap(bufp);
4986
} /* re_compile_fastmap */
4988
weak_alias (__re_compile_fastmap, re_compile_fastmap)
4992
/* Set REGS to hold NUM_REGS registers, storing them in STARTS and
4993
ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use
4994
this memory for recording register information. STARTS and ENDS
4995
must be allocated using the malloc library routine, and must each
4996
be at least NUM_REGS * sizeof (regoff_t) bytes long.
4998
If NUM_REGS == 0, then subsequent matches should allocate their own
5001
Unless this function is called, the first search or match using
5002
PATTERN_BUFFER will allocate its own register data, without
5003
freeing the old data. */
5006
re_set_registers (bufp, regs, num_regs, starts, ends)
5007
struct re_pattern_buffer *bufp;
5008
struct re_registers *regs;
5010
regoff_t *starts, *ends;
5014
bufp->regs_allocated = REGS_REALLOCATE;
5015
regs->num_regs = num_regs;
5016
regs->start = starts;
5021
bufp->regs_allocated = REGS_UNALLOCATED;
5023
regs->start = regs->end = (regoff_t *) 0;
5027
weak_alias (__re_set_registers, re_set_registers)
5030
/* Searching routines. */
5032
/* Like re_search_2, below, but only one string is specified, and
5033
doesn't let you say where to stop matching. */
5036
re_search (bufp, string, size, startpos, range, regs)
5037
struct re_pattern_buffer *bufp;
5039
int size, startpos, range;
5040
struct re_registers *regs;
5042
return re_search_2 (bufp, NULL, 0, string, size, startpos, range,
5046
weak_alias (__re_search, re_search)
5050
/* Using the compiled pattern in BUFP->buffer, first tries to match the
5051
virtual concatenation of STRING1 and STRING2, starting first at index
5052
STARTPOS, then at STARTPOS + 1, and so on.
5054
STRING1 and STRING2 have length SIZE1 and SIZE2, respectively.
5056
RANGE is how far to scan while trying to match. RANGE = 0 means try
5057
only at STARTPOS; in general, the last start tried is STARTPOS +
5060
In REGS, return the indices of the virtual concatenation of STRING1
5061
and STRING2 that matched the entire BUFP->buffer and its contained
5064
Do not consider matching one past the index STOP in the virtual
5065
concatenation of STRING1 and STRING2.
5067
We return either the position in the strings at which the match was
5068
found, -1 if no match, or -2 if error (such as failure
5072
re_search_2 (bufp, string1, size1, string2, size2, startpos, range, regs, stop)
5073
struct re_pattern_buffer *bufp;
5074
const char *string1, *string2;
5078
struct re_registers *regs;
5082
if (MB_CUR_MAX != 1)
5083
return wcs_re_search_2 (bufp, string1, size1, string2, size2, startpos,
5087
return byte_re_search_2 (bufp, string1, size1, string2, size2, startpos,
5091
weak_alias (__re_search_2, re_search_2)
5094
#endif /* not INSIDE_RECURSION */
5096
#ifdef INSIDE_RECURSION
5098
#ifdef MATCH_MAY_ALLOCATE
5099
# define FREE_VAR(var) if (var) REGEX_FREE (var); var = NULL
5101
# define FREE_VAR(var) if (var) free (var); var = NULL
5105
# define MAX_ALLOCA_SIZE 2000
5107
# define FREE_WCS_BUFFERS() \
5109
if (size1 > MAX_ALLOCA_SIZE) \
5111
free (wcs_string1); \
5112
free (mbs_offset1); \
5116
FREE_VAR (wcs_string1); \
5117
FREE_VAR (mbs_offset1); \
5119
if (size2 > MAX_ALLOCA_SIZE) \
5121
free (wcs_string2); \
5122
free (mbs_offset2); \
5126
FREE_VAR (wcs_string2); \
5127
FREE_VAR (mbs_offset2); \
5135
PREFIX(re_search_2) (bufp, string1, size1, string2, size2, startpos, range,
5137
struct re_pattern_buffer *bufp;
5138
const char *string1, *string2;
5142
struct re_registers *regs;
5146
register char *fastmap = bufp->fastmap;
5147
register RE_TRANSLATE_TYPE translate = bufp->translate;
5148
int total_size = size1 + size2;
5149
int endpos = startpos + range;
5151
/* We need wchar_t* buffers correspond to cstring1, cstring2. */
5152
wchar_t *wcs_string1 = NULL, *wcs_string2 = NULL;
5153
/* We need the size of wchar_t buffers correspond to csize1, csize2. */
5154
int wcs_size1 = 0, wcs_size2 = 0;
5155
/* offset buffer for optimizatoin. See convert_mbs_to_wc. */
5156
int *mbs_offset1 = NULL, *mbs_offset2 = NULL;
5157
/* They hold whether each wchar_t is binary data or not. */
5158
char *is_binary = NULL;
5161
/* Check for out-of-range STARTPOS. */
5162
if (startpos < 0 || startpos > total_size)
5165
/* Fix up RANGE if it might eventually take us outside
5166
the virtual concatenation of STRING1 and STRING2.
5167
Make sure we won't move STARTPOS below 0 or above TOTAL_SIZE. */
5169
range = 0 - startpos;
5170
else if (endpos > total_size)
5171
range = total_size - startpos;
5173
/* If the search isn't to be a backwards one, don't waste time in a
5174
search for a pattern that must be anchored. */
5175
if (bufp->used > 0 && range > 0
5176
&& ((re_opcode_t) bufp->buffer[0] == begbuf
5177
/* `begline' is like `begbuf' if it cannot match at newlines. */
5178
|| ((re_opcode_t) bufp->buffer[0] == begline
5179
&& !bufp->newline_anchor)))
5188
/* In a forward search for something that starts with \=.
5189
don't keep searching past point. */
5190
if (bufp->used > 0 && (re_opcode_t) bufp->buffer[0] == at_dot && range > 0)
5192
range = PT - startpos;
5198
/* Update the fastmap now if not correct already. */
5199
if (fastmap && !bufp->fastmap_accurate)
5200
if (re_compile_fastmap (bufp) == -2)
5204
/* Allocate wchar_t array for wcs_string1 and wcs_string2 and
5205
fill them with converted string. */
5208
if (size1 > MAX_ALLOCA_SIZE)
5210
wcs_string1 = TALLOC (size1 + 1, CHAR_T);
5211
mbs_offset1 = TALLOC (size1 + 1, int);
5212
is_binary = TALLOC (size1 + 1, char);
5216
wcs_string1 = REGEX_TALLOC (size1 + 1, CHAR_T);
5217
mbs_offset1 = REGEX_TALLOC (size1 + 1, int);
5218
is_binary = REGEX_TALLOC (size1 + 1, char);
5220
if (!wcs_string1 || !mbs_offset1 || !is_binary)
5222
if (size1 > MAX_ALLOCA_SIZE)
5230
FREE_VAR (wcs_string1);
5231
FREE_VAR (mbs_offset1);
5232
FREE_VAR (is_binary);
5236
wcs_size1 = convert_mbs_to_wcs(wcs_string1, string1, size1,
5237
mbs_offset1, is_binary);
5238
wcs_string1[wcs_size1] = L'\0'; /* for a sentinel */
5239
if (size1 > MAX_ALLOCA_SIZE)
5242
FREE_VAR (is_binary);
5246
if (size2 > MAX_ALLOCA_SIZE)
5248
wcs_string2 = TALLOC (size2 + 1, CHAR_T);
5249
mbs_offset2 = TALLOC (size2 + 1, int);
5250
is_binary = TALLOC (size2 + 1, char);
5254
wcs_string2 = REGEX_TALLOC (size2 + 1, CHAR_T);
5255
mbs_offset2 = REGEX_TALLOC (size2 + 1, int);
5256
is_binary = REGEX_TALLOC (size2 + 1, char);
5258
if (!wcs_string2 || !mbs_offset2 || !is_binary)
5260
FREE_WCS_BUFFERS ();
5261
if (size2 > MAX_ALLOCA_SIZE)
5264
FREE_VAR (is_binary);
5267
wcs_size2 = convert_mbs_to_wcs(wcs_string2, string2, size2,
5268
mbs_offset2, is_binary);
5269
wcs_string2[wcs_size2] = L'\0'; /* for a sentinel */
5270
if (size2 > MAX_ALLOCA_SIZE)
5273
FREE_VAR (is_binary);
5278
/* Loop through the string, looking for a place to start matching. */
5281
/* If a fastmap is supplied, skip quickly over characters that
5282
cannot be the start of a match. If the pattern can match the
5283
null string, however, we don't need to skip characters; we want
5284
the first null string. */
5285
if (fastmap && startpos < total_size && !bufp->can_be_null)
5287
if (range > 0) /* Searching forwards. */
5289
register const char *d;
5290
register int lim = 0;
5293
if (startpos < size1 && startpos + range >= size1)
5294
lim = range - (size1 - startpos);
5296
d = (startpos >= size1 ? string2 - size1 : string1) + startpos;
5298
/* Written out as an if-else to avoid testing `translate'
5302
&& !fastmap[(unsigned char)
5303
translate[(unsigned char) *d++]])
5306
while (range > lim && !fastmap[(unsigned char) *d++])
5309
startpos += irange - range;
5311
else /* Searching backwards. */
5313
register CHAR_T c = (size1 == 0 || startpos >= size1
5314
? string2[startpos - size1]
5315
: string1[startpos]);
5317
if (!fastmap[(unsigned char) TRANSLATE (c)])
5322
/* If can't match the null string, and that's all we have left, fail. */
5323
if (range >= 0 && startpos == total_size && fastmap
5324
&& !bufp->can_be_null)
5327
FREE_WCS_BUFFERS ();
5333
val = wcs_re_match_2_internal (bufp, string1, size1, string2,
5334
size2, startpos, regs, stop,
5335
wcs_string1, wcs_size1,
5336
wcs_string2, wcs_size2,
5337
mbs_offset1, mbs_offset2);
5339
val = byte_re_match_2_internal (bufp, string1, size1, string2,
5340
size2, startpos, regs, stop);
5343
#ifndef REGEX_MALLOC
5352
FREE_WCS_BUFFERS ();
5360
FREE_WCS_BUFFERS ();
5380
FREE_WCS_BUFFERS ();
5386
/* This converts PTR, a pointer into one of the search wchar_t strings
5387
`string1' and `string2' into an multibyte string offset from the
5388
beginning of that string. We use mbs_offset to optimize.
5389
See convert_mbs_to_wcs. */
5390
# define POINTER_TO_OFFSET(ptr) \
5391
(FIRST_STRING_P (ptr) \
5392
? ((regoff_t)(mbs_offset1 != NULL? mbs_offset1[(ptr)-string1] : 0)) \
5393
: ((regoff_t)((mbs_offset2 != NULL? mbs_offset2[(ptr)-string2] : 0) \
5396
/* This converts PTR, a pointer into one of the search strings `string1'
5397
and `string2' into an offset from the beginning of that string. */
5398
# define POINTER_TO_OFFSET(ptr) \
5399
(FIRST_STRING_P (ptr) \
5400
? ((regoff_t) ((ptr) - string1)) \
5401
: ((regoff_t) ((ptr) - string2 + size1)))
5404
/* Macros for dealing with the split strings in re_match_2. */
5406
#define MATCHING_IN_FIRST_STRING (dend == end_match_1)
5408
/* Call before fetching a character with *d. This switches over to
5409
string2 if necessary. */
5410
#define PREFETCH() \
5413
/* End of string2 => fail. */ \
5414
if (dend == end_match_2) \
5416
/* End of string1 => advance to string2. */ \
5418
dend = end_match_2; \
5421
/* Test if at very beginning or at very end of the virtual concatenation
5422
of `string1' and `string2'. If only one string, it's `string2'. */
5423
#define AT_STRINGS_BEG(d) ((d) == (size1 ? string1 : string2) || !size2)
5424
#define AT_STRINGS_END(d) ((d) == end2)
5427
/* Test if D points to a character which is word-constituent. We have
5428
two special cases to check for: if past the end of string1, look at
5429
the first character in string2; and if before the beginning of
5430
string2, look at the last character in string1. */
5432
/* Use internationalized API instead of SYNTAX. */
5433
# define WORDCHAR_P(d) \
5434
(iswalnum ((wint_t)((d) == end1 ? *string2 \
5435
: (d) == string2 - 1 ? *(end1 - 1) : *(d))) != 0 \
5436
|| ((d) == end1 ? *string2 \
5437
: (d) == string2 - 1 ? *(end1 - 1) : *(d)) == L'_')
5439
# define WORDCHAR_P(d) \
5440
(SYNTAX ((d) == end1 ? *string2 \
5441
: (d) == string2 - 1 ? *(end1 - 1) : *(d)) \
5445
/* Disabled due to a compiler bug -- see comment at case wordbound */
5447
/* Test if the character before D and the one at D differ with respect
5448
to being word-constituent. */
5449
#define AT_WORD_BOUNDARY(d) \
5450
(AT_STRINGS_BEG (d) || AT_STRINGS_END (d) \
5451
|| WORDCHAR_P (d - 1) != WORDCHAR_P (d))
5454
/* Free everything we malloc. */
5455
#ifdef MATCH_MAY_ALLOCATE
5457
# define FREE_VARIABLES() \
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
if (!cant_free_wcs_buf) \
5471
FREE_VAR (string1); \
5472
FREE_VAR (string2); \
5473
FREE_VAR (mbs_offset1); \
5474
FREE_VAR (mbs_offset2); \
5478
# define FREE_VARIABLES() \
5480
REGEX_FREE_STACK (fail_stack.stack); \
5481
FREE_VAR (regstart); \
5482
FREE_VAR (regend); \
5483
FREE_VAR (old_regstart); \
5484
FREE_VAR (old_regend); \
5485
FREE_VAR (best_regstart); \
5486
FREE_VAR (best_regend); \
5487
FREE_VAR (reg_info); \
5488
FREE_VAR (reg_dummy); \
5489
FREE_VAR (reg_info_dummy); \
5494
# define FREE_VARIABLES() \
5496
if (!cant_free_wcs_buf) \
5498
FREE_VAR (string1); \
5499
FREE_VAR (string2); \
5500
FREE_VAR (mbs_offset1); \
5501
FREE_VAR (mbs_offset2); \
5505
# define FREE_VARIABLES() ((void)0) /* Do nothing! But inhibit gcc warning. */
5507
#endif /* not MATCH_MAY_ALLOCATE */
5509
/* These values must meet several constraints. They must not be valid
5510
register values; since we have a limit of 255 registers (because
5511
we use only one byte in the pattern for the register number), we can
5512
use numbers larger than 255. They must differ by 1, because of
5513
NUM_FAILURE_ITEMS above. And the value for the lowest register must
5514
be larger than the value for the highest register, so we do not try
5515
to actually save any registers when none are active. */
5516
#define NO_HIGHEST_ACTIVE_REG (1 << BYTEWIDTH)
5517
#define NO_LOWEST_ACTIVE_REG (NO_HIGHEST_ACTIVE_REG + 1)
5519
#else /* not INSIDE_RECURSION */
5520
/* Matching routines. */
5522
#ifndef emacs /* Emacs never uses this. */
5523
/* re_match is like re_match_2 except it takes only a single string. */
5526
re_match (bufp, string, size, pos, regs)
5527
struct re_pattern_buffer *bufp;
5530
struct re_registers *regs;
5534
if (MB_CUR_MAX != 1)
5535
result = wcs_re_match_2_internal (bufp, NULL, 0, string, size,
5537
NULL, 0, NULL, 0, NULL, NULL);
5540
result = byte_re_match_2_internal (bufp, NULL, 0, string, size,
5542
# ifndef REGEX_MALLOC
5550
weak_alias (__re_match, re_match)
5552
#endif /* not emacs */
5554
#endif /* not INSIDE_RECURSION */
5556
#ifdef INSIDE_RECURSION
5557
static boolean PREFIX(group_match_null_string_p) _RE_ARGS ((UCHAR_T **p,
5559
PREFIX(register_info_type) *reg_info));
5560
static boolean PREFIX(alt_match_null_string_p) _RE_ARGS ((UCHAR_T *p,
5562
PREFIX(register_info_type) *reg_info));
5563
static boolean PREFIX(common_op_match_null_string_p) _RE_ARGS ((UCHAR_T **p,
5565
PREFIX(register_info_type) *reg_info));
5566
static int PREFIX(bcmp_translate) _RE_ARGS ((const CHAR_T *s1, const CHAR_T *s2,
5567
int len, char *translate));
5568
#else /* not INSIDE_RECURSION */
5570
/* re_match_2 matches the compiled pattern in BUFP against the
5571
the (virtual) concatenation of STRING1 and STRING2 (of length SIZE1
5572
and SIZE2, respectively). We start matching at POS, and stop
5575
If REGS is non-null and the `no_sub' field of BUFP is nonzero, we
5576
store offsets for the substring each group matched in REGS. See the
5577
documentation for exactly how many groups we fill.
5579
We return -1 if no match, -2 if an internal error (such as the
5580
failure stack overflowing). Otherwise, we return the length of the
5581
matched substring. */
5584
re_match_2 (bufp, string1, size1, string2, size2, pos, regs, stop)
5585
struct re_pattern_buffer *bufp;
5586
const char *string1, *string2;
5589
struct re_registers *regs;
5594
if (MB_CUR_MAX != 1)
5595
result = wcs_re_match_2_internal (bufp, string1, size1, string2, size2,
5597
NULL, 0, NULL, 0, NULL, NULL);
5600
result = byte_re_match_2_internal (bufp, string1, size1, string2, size2,
5603
#ifndef REGEX_MALLOC
5611
weak_alias (__re_match_2, re_match_2)
5614
#endif /* not INSIDE_RECURSION */
5616
#ifdef INSIDE_RECURSION
5619
static int count_mbs_length PARAMS ((int *, int));
5621
/* This check the substring (from 0, to length) of the multibyte string,
5622
to which offset_buffer correspond. And count how many wchar_t_characters
5623
the substring occupy. We use offset_buffer to optimization.
5624
See convert_mbs_to_wcs. */
5627
count_mbs_length(offset_buffer, length)
5633
/* Check whether the size is valid. */
5637
if (offset_buffer == NULL)
5640
/* If there are no multibyte character, offset_buffer[i] == i.
5641
Optmize for this case. */
5642
if (offset_buffer[length] == length)
5645
/* Set up upper with length. (because for all i, offset_buffer[i] >= i) */
5651
int middle = (lower + upper) / 2;
5652
if (middle == lower || middle == upper)
5654
if (offset_buffer[middle] > length)
5656
else if (offset_buffer[middle] < length)
5666
/* This is a separate function so that we can force an alloca cleanup
5670
wcs_re_match_2_internal (bufp, cstring1, csize1, cstring2, csize2, pos,
5671
regs, stop, string1, size1, string2, size2,
5672
mbs_offset1, mbs_offset2)
5673
struct re_pattern_buffer *bufp;
5674
const char *cstring1, *cstring2;
5677
struct re_registers *regs;
5679
/* string1 == string2 == NULL means string1/2, size1/2 and
5680
mbs_offset1/2 need seting up in this function. */
5681
/* We need wchar_t* buffers correspond to cstring1, cstring2. */
5682
wchar_t *string1, *string2;
5683
/* We need the size of wchar_t buffers correspond to csize1, csize2. */
5685
/* offset buffer for optimizatoin. See convert_mbs_to_wc. */
5686
int *mbs_offset1, *mbs_offset2;
5689
byte_re_match_2_internal (bufp, string1, size1,string2, size2, pos,
5691
struct re_pattern_buffer *bufp;
5692
const char *string1, *string2;
5695
struct re_registers *regs;
5699
/* General temporaries. */
5703
/* They hold whether each wchar_t is binary data or not. */
5704
char *is_binary = NULL;
5705
/* If true, we can't free string1/2, mbs_offset1/2. */
5706
int cant_free_wcs_buf = 1;
5709
/* Just past the end of the corresponding string. */
5710
const CHAR_T *end1, *end2;
5712
/* Pointers into string1 and string2, just past the last characters in
5713
each to consider matching. */
5714
const CHAR_T *end_match_1, *end_match_2;
5716
/* Where we are in the data, and the end of the current string. */
5717
const CHAR_T *d, *dend;
5719
/* Where we are in the pattern, and the end of the pattern. */
5721
UCHAR_T *pattern, *p;
5722
register UCHAR_T *pend;
5724
UCHAR_T *p = bufp->buffer;
5725
register UCHAR_T *pend = p + bufp->used;
5728
/* Mark the opcode just after a start_memory, so we can test for an
5729
empty subpattern when we get to the stop_memory. */
5730
UCHAR_T *just_past_start_mem = 0;
5732
/* We use this to map every character in the string. */
5733
RE_TRANSLATE_TYPE translate = bufp->translate;
5735
/* Failure point stack. Each place that can handle a failure further
5736
down the line pushes a failure point on this stack. It consists of
5737
restart, regend, and reg_info for all registers corresponding to
5738
the subexpressions we're currently inside, plus the number of such
5739
registers, and, finally, two char *'s. The first char * is where
5740
to resume scanning the pattern; the second one is where to resume
5741
scanning the strings. If the latter is zero, the failure point is
5742
a ``dummy''; if a failure happens and the failure point is a dummy,
5743
it gets discarded and the next next one is tried. */
5744
#ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
5745
PREFIX(fail_stack_type) fail_stack;
5748
static unsigned failure_id;
5749
unsigned nfailure_points_pushed = 0, nfailure_points_popped = 0;
5753
/* This holds the pointer to the failure stack, when
5754
it is allocated relocatably. */
5755
fail_stack_elt_t *failure_stack_ptr;
5758
/* We fill all the registers internally, independent of what we
5759
return, for use in backreferences. The number here includes
5760
an element for register zero. */
5761
size_t num_regs = bufp->re_nsub + 1;
5763
/* The currently active registers. */
5764
active_reg_t lowest_active_reg = NO_LOWEST_ACTIVE_REG;
5765
active_reg_t highest_active_reg = NO_HIGHEST_ACTIVE_REG;
5767
/* Information on the contents of registers. These are pointers into
5768
the input strings; they record just what was matched (on this
5769
attempt) by a subexpression part of the pattern, that is, the
5770
regnum-th regstart pointer points to where in the pattern we began
5771
matching and the regnum-th regend points to right after where we
5772
stopped matching the regnum-th subexpression. (The zeroth register
5773
keeps track of what the whole pattern matches.) */
5774
#ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
5775
const CHAR_T **regstart, **regend;
5778
/* If a group that's operated upon by a repetition operator fails to
5779
match anything, then the register for its start will need to be
5780
restored because it will have been set to wherever in the string we
5781
are when we last see its open-group operator. Similarly for a
5783
#ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
5784
const CHAR_T **old_regstart, **old_regend;
5787
/* The is_active field of reg_info helps us keep track of which (possibly
5788
nested) subexpressions we are currently in. The matched_something
5789
field of reg_info[reg_num] helps us tell whether or not we have
5790
matched any of the pattern so far this time through the reg_num-th
5791
subexpression. These two fields get reset each time through any
5792
loop their register is in. */
5793
#ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
5794
PREFIX(register_info_type) *reg_info;
5797
/* The following record the register info as found in the above
5798
variables when we find a match better than any we've seen before.
5799
This happens as we backtrack through the failure points, which in
5800
turn happens only if we have not yet matched the entire string. */
5801
unsigned best_regs_set = false;
5802
#ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
5803
const CHAR_T **best_regstart, **best_regend;
5806
/* Logically, this is `best_regend[0]'. But we don't want to have to
5807
allocate space for that if we're not allocating space for anything
5808
else (see below). Also, we never need info about register 0 for
5809
any of the other register vectors, and it seems rather a kludge to
5810
treat `best_regend' differently than the rest. So we keep track of
5811
the end of the best match so far in a separate variable. We
5812
initialize this to NULL so that when we backtrack the first time
5813
and need to test it, it's not garbage. */
5814
const CHAR_T *match_end = NULL;
5816
/* This helps SET_REGS_MATCHED avoid doing redundant work. */
5817
int set_regs_matched_done = 0;
5819
/* Used when we pop values we don't care about. */
5820
#ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
5821
const CHAR_T **reg_dummy;
5822
PREFIX(register_info_type) *reg_info_dummy;
5826
/* Counts the total number of registers pushed. */
5827
unsigned num_regs_pushed = 0;
5830
DEBUG_PRINT1 ("\n\nEntering re_match_2.\n");
5834
#ifdef MATCH_MAY_ALLOCATE
5835
/* Do not bother to initialize all the register variables if there are
5836
no groups in the pattern, as it takes a fair amount of time. If
5837
there are groups, we include space for register 0 (the whole
5838
pattern), even though we never use it, since it simplifies the
5839
array indexing. We should fix this. */
5842
regstart = REGEX_TALLOC (num_regs, const CHAR_T *);
5843
regend = REGEX_TALLOC (num_regs, const CHAR_T *);
5844
old_regstart = REGEX_TALLOC (num_regs, const CHAR_T *);
5845
old_regend = REGEX_TALLOC (num_regs, const CHAR_T *);
5846
best_regstart = REGEX_TALLOC (num_regs, const CHAR_T *);
5847
best_regend = REGEX_TALLOC (num_regs, const CHAR_T *);
5848
reg_info = REGEX_TALLOC (num_regs, PREFIX(register_info_type));
5849
reg_dummy = REGEX_TALLOC (num_regs, const CHAR_T *);
5850
reg_info_dummy = REGEX_TALLOC (num_regs, PREFIX(register_info_type));
5852
if (!(regstart && regend && old_regstart && old_regend && reg_info
5853
&& best_regstart && best_regend && reg_dummy && reg_info_dummy))
5861
/* We must initialize all our variables to NULL, so that
5862
`FREE_VARIABLES' doesn't try to free them. */
5863
regstart = regend = old_regstart = old_regend = best_regstart
5864
= best_regend = reg_dummy = NULL;
5865
reg_info = reg_info_dummy = (PREFIX(register_info_type) *) NULL;
5867
#endif /* MATCH_MAY_ALLOCATE */
5869
/* The starting position is bogus. */
5871
if (pos < 0 || pos > csize1 + csize2)
5873
if (pos < 0 || pos > size1 + size2)
5881
/* Allocate wchar_t array for string1 and string2 and
5882
fill them with converted string. */
5883
if (string1 == NULL && string2 == NULL)
5885
/* We need seting up buffers here. */
5887
/* We must free wcs buffers in this function. */
5888
cant_free_wcs_buf = 0;
5892
string1 = REGEX_TALLOC (csize1 + 1, CHAR_T);
5893
mbs_offset1 = REGEX_TALLOC (csize1 + 1, int);
5894
is_binary = REGEX_TALLOC (csize1 + 1, char);
5895
if (!string1 || !mbs_offset1 || !is_binary)
5898
FREE_VAR (mbs_offset1);
5899
FREE_VAR (is_binary);
5905
string2 = REGEX_TALLOC (csize2 + 1, CHAR_T);
5906
mbs_offset2 = REGEX_TALLOC (csize2 + 1, int);
5907
is_binary = REGEX_TALLOC (csize2 + 1, char);
5908
if (!string2 || !mbs_offset2 || !is_binary)
5911
FREE_VAR (mbs_offset1);
5913
FREE_VAR (mbs_offset2);
5914
FREE_VAR (is_binary);
5917
size2 = convert_mbs_to_wcs(string2, cstring2, csize2,
5918
mbs_offset2, is_binary);
5919
string2[size2] = L'\0'; /* for a sentinel */
5920
FREE_VAR (is_binary);
5924
/* We need to cast pattern to (wchar_t*), because we casted this compiled
5925
pattern to (char*) in regex_compile. */
5926
p = pattern = (CHAR_T*)bufp->buffer;
5927
pend = (CHAR_T*)(bufp->buffer + bufp->used);
5931
/* Initialize subexpression text positions to -1 to mark ones that no
5932
start_memory/stop_memory has been seen for. Also initialize the
5933
register information struct. */
5934
for (mcnt = 1; (unsigned) mcnt < num_regs; mcnt++)
5936
regstart[mcnt] = regend[mcnt]
5937
= old_regstart[mcnt] = old_regend[mcnt] = REG_UNSET_VALUE;
5939
REG_MATCH_NULL_STRING_P (reg_info[mcnt]) = MATCH_NULL_UNSET_VALUE;
5940
IS_ACTIVE (reg_info[mcnt]) = 0;
5941
MATCHED_SOMETHING (reg_info[mcnt]) = 0;
5942
EVER_MATCHED_SOMETHING (reg_info[mcnt]) = 0;
5945
/* We move `string1' into `string2' if the latter's empty -- but not if
5946
`string1' is null. */
5947
if (size2 == 0 && string1 != NULL)
5954
mbs_offset2 = mbs_offset1;
5960
end1 = string1 + size1;
5961
end2 = string2 + size2;
5963
/* Compute where to stop matching, within the two strings. */
5967
mcnt = count_mbs_length(mbs_offset1, stop);
5968
end_match_1 = string1 + mcnt;
5969
end_match_2 = string2;
5973
if (stop > csize1 + csize2)
5974
stop = csize1 + csize2;
5976
mcnt = count_mbs_length(mbs_offset2, stop-csize1);
5977
end_match_2 = string2 + mcnt;
5980
{ /* count_mbs_length return error. */
5987
end_match_1 = string1 + stop;
5988
end_match_2 = string2;
5993
end_match_2 = string2 + stop - size1;
5997
/* `p' scans through the pattern as `d' scans through the data.
5998
`dend' is the end of the input string that `d' points within. `d'
5999
is advanced into the following input string whenever necessary, but
6000
this happens before fetching; therefore, at the beginning of the
6001
loop, `d' can be pointing at the end of a string, but it cannot
6004
if (size1 > 0 && pos <= csize1)
6006
mcnt = count_mbs_length(mbs_offset1, pos);
6012
mcnt = count_mbs_length(mbs_offset2, pos-csize1);
6018
{ /* count_mbs_length return error. */
6023
if (size1 > 0 && pos <= size1)
6030
d = string2 + pos - size1;
6035
DEBUG_PRINT1 ("The compiled pattern is:\n");
6036
DEBUG_PRINT_COMPILED_PATTERN (bufp, p, pend);
6037
DEBUG_PRINT1 ("The string to match is: `");
6038
DEBUG_PRINT_DOUBLE_STRING (d, string1, size1, string2, size2);
6039
DEBUG_PRINT1 ("'\n");
6041
/* This loops over pattern commands. It exits by returning from the
6042
function if the match is complete, or it drops through if the match
6043
fails at this starting point in the input data. */
6047
DEBUG_PRINT2 ("\n%p: ", p);
6049
DEBUG_PRINT2 ("\n0x%x: ", p);
6053
{ /* End of pattern means we might have succeeded. */
6054
DEBUG_PRINT1 ("end of pattern ... ");
6056
/* If we haven't matched the entire string, and we want the
6057
longest match, try backtracking. */
6058
if (d != end_match_2)
6060
/* 1 if this match ends in the same string (string1 or string2)
6061
as the best previous match. */
6062
boolean same_str_p = (FIRST_STRING_P (match_end)
6063
== MATCHING_IN_FIRST_STRING);
6064
/* 1 if this match is the best seen so far. */
6065
boolean best_match_p;
6067
/* AIX compiler got confused when this was combined
6068
with the previous declaration. */
6070
best_match_p = d > match_end;
6072
best_match_p = !MATCHING_IN_FIRST_STRING;
6074
DEBUG_PRINT1 ("backtracking.\n");
6076
if (!FAIL_STACK_EMPTY ())
6077
{ /* More failure points to try. */
6079
/* If exceeds best match so far, save it. */
6080
if (!best_regs_set || best_match_p)
6082
best_regs_set = true;
6085
DEBUG_PRINT1 ("\nSAVING match as best so far.\n");
6087
for (mcnt = 1; (unsigned) mcnt < num_regs; mcnt++)
6089
best_regstart[mcnt] = regstart[mcnt];
6090
best_regend[mcnt] = regend[mcnt];
6096
/* If no failure points, don't restore garbage. And if
6097
last match is real best match, don't restore second
6099
else if (best_regs_set && !best_match_p)
6102
/* Restore best match. It may happen that `dend ==
6103
end_match_1' while the restored d is in string2.
6104
For example, the pattern `x.*y.*z' against the
6105
strings `x-' and `y-z-', if the two strings are
6106
not consecutive in memory. */
6107
DEBUG_PRINT1 ("Restoring best registers.\n");
6110
dend = ((d >= string1 && d <= end1)
6111
? end_match_1 : end_match_2);
6113
for (mcnt = 1; (unsigned) mcnt < num_regs; mcnt++)
6115
regstart[mcnt] = best_regstart[mcnt];
6116
regend[mcnt] = best_regend[mcnt];
6119
} /* d != end_match_2 */
6122
DEBUG_PRINT1 ("Accepting match.\n");
6123
/* If caller wants register contents data back, do it. */
6124
if (regs && !bufp->no_sub)
6126
/* Have the register data arrays been allocated? */
6127
if (bufp->regs_allocated == REGS_UNALLOCATED)
6128
{ /* No. So allocate them with malloc. We need one
6129
extra element beyond `num_regs' for the `-1' marker
6131
regs->num_regs = MAX (RE_NREGS, num_regs + 1);
6132
regs->start = TALLOC (regs->num_regs, regoff_t);
6133
regs->end = TALLOC (regs->num_regs, regoff_t);
6134
if (regs->start == NULL || regs->end == NULL)
6139
bufp->regs_allocated = REGS_REALLOCATE;
6141
else if (bufp->regs_allocated == REGS_REALLOCATE)
6142
{ /* Yes. If we need more elements than were already
6143
allocated, reallocate them. If we need fewer, just
6145
if (regs->num_regs < num_regs + 1)
6147
regs->num_regs = num_regs + 1;
6148
RETALLOC (regs->start, regs->num_regs, regoff_t);
6149
RETALLOC (regs->end, regs->num_regs, regoff_t);
6150
if (regs->start == NULL || regs->end == NULL)
6159
/* These braces fend off a "empty body in an else-statement"
6160
warning under GCC when assert expands to nothing. */
6161
assert (bufp->regs_allocated == REGS_FIXED);
6164
/* Convert the pointer data in `regstart' and `regend' to
6165
indices. Register zero has to be set differently,
6166
since we haven't kept track of any info for it. */
6167
if (regs->num_regs > 0)
6169
regs->start[0] = pos;
6171
if (MATCHING_IN_FIRST_STRING)
6172
regs->end[0] = mbs_offset1 != NULL ?
6173
mbs_offset1[d-string1] : 0;
6175
regs->end[0] = csize1 + (mbs_offset2 != NULL ?
6176
mbs_offset2[d-string2] : 0);
6178
regs->end[0] = (MATCHING_IN_FIRST_STRING
6179
? ((regoff_t) (d - string1))
6180
: ((regoff_t) (d - string2 + size1)));
6184
/* Go through the first `min (num_regs, regs->num_regs)'
6185
registers, since that is all we initialized. */
6186
for (mcnt = 1; (unsigned) mcnt < MIN (num_regs, regs->num_regs);
6189
if (REG_UNSET (regstart[mcnt]) || REG_UNSET (regend[mcnt]))
6190
regs->start[mcnt] = regs->end[mcnt] = -1;
6194
= (regoff_t) POINTER_TO_OFFSET (regstart[mcnt]);
6196
= (regoff_t) POINTER_TO_OFFSET (regend[mcnt]);
6200
/* If the regs structure we return has more elements than
6201
were in the pattern, set the extra elements to -1. If
6202
we (re)allocated the registers, this is the case,
6203
because we always allocate enough to have at least one
6205
for (mcnt = num_regs; (unsigned) mcnt < regs->num_regs; mcnt++)
6206
regs->start[mcnt] = regs->end[mcnt] = -1;
6207
} /* regs && !bufp->no_sub */
6209
DEBUG_PRINT4 ("%u failure points pushed, %u popped (%u remain).\n",
6210
nfailure_points_pushed, nfailure_points_popped,
6211
nfailure_points_pushed - nfailure_points_popped);
6212
DEBUG_PRINT2 ("%u registers pushed.\n", num_regs_pushed);
6215
if (MATCHING_IN_FIRST_STRING)
6216
mcnt = mbs_offset1 != NULL ? mbs_offset1[d-string1] : 0;
6218
mcnt = (mbs_offset2 != NULL ? mbs_offset2[d-string2] : 0) +
6222
mcnt = d - pos - (MATCHING_IN_FIRST_STRING
6227
DEBUG_PRINT2 ("Returning %d from re_match_2.\n", mcnt);
6233
/* Otherwise match next pattern command. */
6234
switch (SWITCH_ENUM_CAST ((re_opcode_t) *p++))
6236
/* Ignore these. Used to ignore the n of succeed_n's which
6237
currently have n == 0. */
6239
DEBUG_PRINT1 ("EXECUTING no_op.\n");
6243
DEBUG_PRINT1 ("EXECUTING succeed.\n");
6246
/* Match the next n pattern characters exactly. The following
6247
byte in the pattern defines n, and the n bytes after that
6248
are the characters to match. */
6254
DEBUG_PRINT2 ("EXECUTING exactn %d.\n", mcnt);
6256
/* This is written out as an if-else so we don't waste time
6257
testing `translate' inside the loop. */
6266
if ((UCHAR_T) translate[(unsigned char) *d++]
6272
if (*d++ != (CHAR_T) *p++)
6276
if ((UCHAR_T) translate[(unsigned char) *d++]
6288
if (*d++ != (CHAR_T) *p++) goto fail;
6292
SET_REGS_MATCHED ();
6296
/* Match any character except possibly a newline or a null. */
6298
DEBUG_PRINT1 ("EXECUTING anychar.\n");
6302
if ((!(bufp->syntax & RE_DOT_NEWLINE) && TRANSLATE (*d) == '\n')
6303
|| (bufp->syntax & RE_DOT_NOT_NULL && TRANSLATE (*d) == '\000'))
6306
SET_REGS_MATCHED ();
6307
DEBUG_PRINT2 (" Matched `%ld'.\n", (long int) *d);
6317
unsigned int i, char_class_length, coll_symbol_length,
6318
equiv_class_length, ranges_length, chars_length, length;
6319
CHAR_T *workp, *workp2, *charset_top;
6320
#define WORK_BUFFER_SIZE 128
6321
CHAR_T str_buf[WORK_BUFFER_SIZE];
6326
boolean not = (re_opcode_t) *(p - 1) == charset_not;
6328
DEBUG_PRINT2 ("EXECUTING charset%s.\n", not ? "_not" : "");
6330
c = TRANSLATE (*d); /* The character to match. */
6333
nrules = _NL_CURRENT_WORD (LC_COLLATE, _NL_COLLATE_NRULES);
6335
charset_top = p - 1;
6336
char_class_length = *p++;
6337
coll_symbol_length = *p++;
6338
equiv_class_length = *p++;
6339
ranges_length = *p++;
6340
chars_length = *p++;
6341
/* p points charset[6], so the address of the next instruction
6342
(charset[l+m+n+2o+k+p']) equals p[l+m+n+2*o+p'],
6343
where l=length of char_classes, m=length of collating_symbol,
6344
n=equivalence_class, o=length of char_range,
6345
p'=length of character. */
6347
/* Update p to indicate the next instruction. */
6348
p += char_class_length + coll_symbol_length+ equiv_class_length +
6349
2*ranges_length + chars_length;
6351
/* match with char_class? */
6352
for (i = 0; i < char_class_length ; i += CHAR_CLASS_SIZE)
6355
uintptr_t alignedp = ((uintptr_t)workp
6356
+ __alignof__(wctype_t) - 1)
6357
& ~(uintptr_t)(__alignof__(wctype_t) - 1);
6358
wctype = *((wctype_t*)alignedp);
6359
workp += CHAR_CLASS_SIZE;
6361
if (__iswctype((wint_t)c, wctype))
6362
goto char_set_matched;
6364
if (iswctype((wint_t)c, wctype))
6365
goto char_set_matched;
6369
/* match with collating_symbol? */
6373
const unsigned char *extra = (const unsigned char *)
6374
_NL_CURRENT (LC_COLLATE, _NL_COLLATE_SYMB_EXTRAMB);
6376
for (workp2 = workp + coll_symbol_length ; workp < workp2 ;
6380
wextra = (int32_t*)(extra + *workp++);
6381
for (i = 0; i < *wextra; ++i)
6382
if (TRANSLATE(d[i]) != wextra[1 + i])
6387
/* Update d, however d will be incremented at
6388
char_set_matched:, we decrement d here. */
6390
goto char_set_matched;
6394
else /* (nrules == 0) */
6396
/* If we can't look up collation data, we use wcscoll
6399
for (workp2 = workp + coll_symbol_length ; workp < workp2 ;)
6401
const CHAR_T *backup_d = d, *backup_dend = dend;
6403
length = __wcslen (workp);
6405
length = wcslen (workp);
6408
/* If wcscoll(the collating symbol, whole string) > 0,
6409
any substring of the string never match with the
6410
collating symbol. */
6412
if (__wcscoll (workp, d) > 0)
6414
if (wcscoll (workp, d) > 0)
6417
workp += length + 1;
6421
/* First, we compare the collating symbol with
6422
the first character of the string.
6423
If it don't match, we add the next character to
6424
the compare buffer in turn. */
6425
for (i = 0 ; i < WORK_BUFFER_SIZE-1 ; i++, d++)
6430
if (dend == end_match_2)
6436
/* add next character to the compare buffer. */
6437
str_buf[i] = TRANSLATE(*d);
6438
str_buf[i+1] = '\0';
6441
match = __wcscoll (workp, str_buf);
6443
match = wcscoll (workp, str_buf);
6446
goto char_set_matched;
6449
/* (str_buf > workp) indicate (str_buf + X > workp),
6450
because for all X (str_buf + X > str_buf).
6451
So we don't need continue this loop. */
6454
/* Otherwise(str_buf < workp),
6455
(str_buf+next_character) may equals (workp).
6456
So we continue this loop. */
6461
workp += length + 1;
6464
/* match with equivalence_class? */
6468
const CHAR_T *backup_d = d, *backup_dend = dend;
6469
/* Try to match the equivalence class against
6470
those known to the collate implementation. */
6471
const int32_t *table;
6472
const int32_t *weights;
6473
const int32_t *extra;
6474
const int32_t *indirect;
6479
/* This #include defines a local function! */
6480
# include <locale/weightwc.h>
6482
table = (const int32_t *)
6483
_NL_CURRENT (LC_COLLATE, _NL_COLLATE_TABLEWC);
6484
weights = (const wint_t *)
6485
_NL_CURRENT (LC_COLLATE, _NL_COLLATE_WEIGHTWC);
6486
extra = (const wint_t *)
6487
_NL_CURRENT (LC_COLLATE, _NL_COLLATE_EXTRAWC);
6488
indirect = (const int32_t *)
6489
_NL_CURRENT (LC_COLLATE, _NL_COLLATE_INDIRECTWC);
6491
/* Write 1 collating element to str_buf, and
6495
for (i = 0 ; idx2 == 0 && i < WORK_BUFFER_SIZE - 1; i++)
6497
cp = (wint_t*)str_buf;
6500
if (dend == end_match_2)
6505
str_buf[i] = TRANSLATE(*(d+i));
6506
str_buf[i+1] = '\0'; /* sentinel */
6507
idx2 = findidx ((const wint_t**)&cp);
6510
/* Update d, however d will be incremented at
6511
char_set_matched:, we decrement d here. */
6512
d = backup_d + ((wchar_t*)cp - (wchar_t*)str_buf - 1);
6515
if (dend == end_match_2)
6524
len = weights[idx2];
6526
for (workp2 = workp + equiv_class_length ; workp < workp2 ;
6529
idx = (int32_t)*workp;
6530
/* We already checked idx != 0 in regex_compile. */
6532
if (idx2 != 0 && len == weights[idx])
6535
while (cnt < len && (weights[idx + 1 + cnt]
6536
== weights[idx2 + 1 + cnt]))
6540
goto char_set_matched;
6547
else /* (nrules == 0) */
6549
/* If we can't look up collation data, we use wcscoll
6552
for (workp2 = workp + equiv_class_length ; workp < workp2 ;)
6554
const CHAR_T *backup_d = d, *backup_dend = dend;
6556
length = __wcslen (workp);
6558
length = wcslen (workp);
6561
/* If wcscoll(the collating symbol, whole string) > 0,
6562
any substring of the string never match with the
6563
collating symbol. */
6565
if (__wcscoll (workp, d) > 0)
6567
if (wcscoll (workp, d) > 0)
6570
workp += length + 1;
6574
/* First, we compare the equivalence class with
6575
the first character of the string.
6576
If it don't match, we add the next character to
6577
the compare buffer in turn. */
6578
for (i = 0 ; i < WORK_BUFFER_SIZE - 1 ; i++, d++)
6583
if (dend == end_match_2)
6589
/* add next character to the compare buffer. */
6590
str_buf[i] = TRANSLATE(*d);
6591
str_buf[i+1] = '\0';
6594
match = __wcscoll (workp, str_buf);
6596
match = wcscoll (workp, str_buf);
6600
goto char_set_matched;
6603
/* (str_buf > workp) indicate (str_buf + X > workp),
6604
because for all X (str_buf + X > str_buf).
6605
So we don't need continue this loop. */
6608
/* Otherwise(str_buf < workp),
6609
(str_buf+next_character) may equals (workp).
6610
So we continue this loop. */
6615
workp += length + 1;
6619
/* match with char_range? */
6623
uint32_t collseqval;
6624
const char *collseq = (const char *)
6625
_NL_CURRENT(LC_COLLATE, _NL_COLLATE_COLLSEQWC);
6627
collseqval = collseq_table_lookup (collseq, c);
6629
for (; workp < p - chars_length ;)
6631
uint32_t start_val, end_val;
6633
/* We already compute the collation sequence value
6634
of the characters (or collating symbols). */
6635
start_val = (uint32_t) *workp++; /* range_start */
6636
end_val = (uint32_t) *workp++; /* range_end */
6638
if (start_val <= collseqval && collseqval <= end_val)
6639
goto char_set_matched;
6645
/* We set range_start_char at str_buf[0], range_end_char
6646
at str_buf[4], and compared char at str_buf[2]. */
6651
for (; workp < p - chars_length ;)
6653
wchar_t *range_start_char, *range_end_char;
6655
/* match if (range_start_char <= c <= range_end_char). */
6657
/* If range_start(or end) < 0, we assume -range_start(end)
6658
is the offset of the collating symbol which is specified
6659
as the character of the range start(end). */
6663
range_start_char = charset_top - (*workp++);
6666
str_buf[0] = *workp++;
6667
range_start_char = str_buf;
6672
range_end_char = charset_top - (*workp++);
6675
str_buf[4] = *workp++;
6676
range_end_char = str_buf + 4;
6680
if (__wcscoll (range_start_char, str_buf+2) <= 0
6681
&& __wcscoll (str_buf+2, range_end_char) <= 0)
6683
if (wcscoll (range_start_char, str_buf+2) <= 0
6684
&& wcscoll (str_buf+2, range_end_char) <= 0)
6686
goto char_set_matched;
6690
/* match with char? */
6691
for (; workp < p ; workp++)
6693
goto char_set_matched;
6700
/* Cast to `unsigned' instead of `unsigned char' in case the
6701
bit list is a full 32 bytes long. */
6702
if (c < (unsigned) (*p * BYTEWIDTH)
6703
&& p[1 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH)))
6708
if (!not) goto fail;
6709
#undef WORK_BUFFER_SIZE
6711
SET_REGS_MATCHED ();
6717
/* The beginning of a group is represented by start_memory.
6718
The arguments are the register number in the next byte, and the
6719
number of groups inner to this one in the next. The text
6720
matched within the group is recorded (in the internal
6721
registers data structure) under the register number. */
6723
DEBUG_PRINT3 ("EXECUTING start_memory %ld (%ld):\n",
6724
(long int) *p, (long int) p[1]);
6726
/* Find out if this group can match the empty string. */
6727
p1 = p; /* To send to group_match_null_string_p. */
6729
if (REG_MATCH_NULL_STRING_P (reg_info[*p]) == MATCH_NULL_UNSET_VALUE)
6730
REG_MATCH_NULL_STRING_P (reg_info[*p])
6731
= PREFIX(group_match_null_string_p) (&p1, pend, reg_info);
6733
/* Save the position in the string where we were the last time
6734
we were at this open-group operator in case the group is
6735
operated upon by a repetition operator, e.g., with `(a*)*b'
6736
against `ab'; then we want to ignore where we are now in
6737
the string in case this attempt to match fails. */
6738
old_regstart[*p] = REG_MATCH_NULL_STRING_P (reg_info[*p])
6739
? REG_UNSET (regstart[*p]) ? d : regstart[*p]
6741
DEBUG_PRINT2 (" old_regstart: %d\n",
6742
POINTER_TO_OFFSET (old_regstart[*p]));
6745
DEBUG_PRINT2 (" regstart: %d\n", POINTER_TO_OFFSET (regstart[*p]));
6747
IS_ACTIVE (reg_info[*p]) = 1;
6748
MATCHED_SOMETHING (reg_info[*p]) = 0;
6750
/* Clear this whenever we change the register activity status. */
6751
set_regs_matched_done = 0;
6753
/* This is the new highest active register. */
6754
highest_active_reg = *p;
6756
/* If nothing was active before, this is the new lowest active
6758
if (lowest_active_reg == NO_LOWEST_ACTIVE_REG)
6759
lowest_active_reg = *p;
6761
/* Move past the register number and inner group count. */
6763
just_past_start_mem = p;
6768
/* The stop_memory opcode represents the end of a group. Its
6769
arguments are the same as start_memory's: the register
6770
number, and the number of inner groups. */
6772
DEBUG_PRINT3 ("EXECUTING stop_memory %ld (%ld):\n",
6773
(long int) *p, (long int) p[1]);
6775
/* We need to save the string position the last time we were at
6776
this close-group operator in case the group is operated
6777
upon by a repetition operator, e.g., with `((a*)*(b*)*)*'
6778
against `aba'; then we want to ignore where we are now in
6779
the string in case this attempt to match fails. */
6780
old_regend[*p] = REG_MATCH_NULL_STRING_P (reg_info[*p])
6781
? REG_UNSET (regend[*p]) ? d : regend[*p]
6783
DEBUG_PRINT2 (" old_regend: %d\n",
6784
POINTER_TO_OFFSET (old_regend[*p]));
6787
DEBUG_PRINT2 (" regend: %d\n", POINTER_TO_OFFSET (regend[*p]));
6789
/* This register isn't active anymore. */
6790
IS_ACTIVE (reg_info[*p]) = 0;
6792
/* Clear this whenever we change the register activity status. */
6793
set_regs_matched_done = 0;
6795
/* If this was the only register active, nothing is active
6797
if (lowest_active_reg == highest_active_reg)
6799
lowest_active_reg = NO_LOWEST_ACTIVE_REG;
6800
highest_active_reg = NO_HIGHEST_ACTIVE_REG;
6803
{ /* We must scan for the new highest active register, since
6804
it isn't necessarily one less than now: consider
6805
(a(b)c(d(e)f)g). When group 3 ends, after the f), the
6806
new highest active register is 1. */
6808
while (r > 0 && !IS_ACTIVE (reg_info[r]))
6811
/* If we end up at register zero, that means that we saved
6812
the registers as the result of an `on_failure_jump', not
6813
a `start_memory', and we jumped to past the innermost
6814
`stop_memory'. For example, in ((.)*) we save
6815
registers 1 and 2 as a result of the *, but when we pop
6816
back to the second ), we are at the stop_memory 1.
6817
Thus, nothing is active. */
6820
lowest_active_reg = NO_LOWEST_ACTIVE_REG;
6821
highest_active_reg = NO_HIGHEST_ACTIVE_REG;
6824
highest_active_reg = r;
6827
/* If just failed to match something this time around with a
6828
group that's operated on by a repetition operator, try to
6829
force exit from the ``loop'', and restore the register
6830
information for this group that we had before trying this
6832
if ((!MATCHED_SOMETHING (reg_info[*p])
6833
|| just_past_start_mem == p - 1)
6836
boolean is_a_jump_n = false;
6840
switch ((re_opcode_t) *p1++)
6844
case pop_failure_jump:
6845
case maybe_pop_jump:
6847
case dummy_failure_jump:
6848
EXTRACT_NUMBER_AND_INCR (mcnt, p1);
6850
p1 += OFFSET_ADDRESS_SIZE;
6858
/* If the next operation is a jump backwards in the pattern
6859
to an on_failure_jump right before the start_memory
6860
corresponding to this stop_memory, exit from the loop
6861
by forcing a failure after pushing on the stack the
6862
on_failure_jump's jump in the pattern, and d. */
6863
if (mcnt < 0 && (re_opcode_t) *p1 == on_failure_jump
6864
&& (re_opcode_t) p1[1+OFFSET_ADDRESS_SIZE] == start_memory
6865
&& p1[2+OFFSET_ADDRESS_SIZE] == *p)
6867
/* If this group ever matched anything, then restore
6868
what its registers were before trying this last
6869
failed match, e.g., with `(a*)*b' against `ab' for
6870
regstart[1], and, e.g., with `((a*)*(b*)*)*'
6871
against `aba' for regend[3].
6873
Also restore the registers for inner groups for,
6874
e.g., `((a*)(b*))*' against `aba' (register 3 would
6875
otherwise get trashed). */
6877
if (EVER_MATCHED_SOMETHING (reg_info[*p]))
6881
EVER_MATCHED_SOMETHING (reg_info[*p]) = 0;
6883
/* Restore this and inner groups' (if any) registers. */
6884
for (r = *p; r < (unsigned) *p + (unsigned) *(p + 1);
6887
regstart[r] = old_regstart[r];
6889
/* xx why this test? */
6890
if (old_regend[r] >= regstart[r])
6891
regend[r] = old_regend[r];
6895
EXTRACT_NUMBER_AND_INCR (mcnt, p1);
6896
PUSH_FAILURE_POINT (p1 + mcnt, d, -2);
6902
/* Move past the register number and the inner group count. */
6907
/* \<digit> has been turned into a `duplicate' command which is
6908
followed by the numeric value of <digit> as the register number. */
6911
register const CHAR_T *d2, *dend2;
6912
int regno = *p++; /* Get which register to match against. */
6913
DEBUG_PRINT2 ("EXECUTING duplicate %d.\n", regno);
6915
/* Can't back reference a group which we've never matched. */
6916
if (REG_UNSET (regstart[regno]) || REG_UNSET (regend[regno]))
6919
/* Where in input to try to start matching. */
6920
d2 = regstart[regno];
6922
/* Where to stop matching; if both the place to start and
6923
the place to stop matching are in the same string, then
6924
set to the place to stop, otherwise, for now have to use
6925
the end of the first string. */
6927
dend2 = ((FIRST_STRING_P (regstart[regno])
6928
== FIRST_STRING_P (regend[regno]))
6929
? regend[regno] : end_match_1);
6932
/* If necessary, advance to next segment in register
6936
if (dend2 == end_match_2) break;
6937
if (dend2 == regend[regno]) break;
6939
/* End of string1 => advance to string2. */
6941
dend2 = regend[regno];
6943
/* At end of register contents => success */
6944
if (d2 == dend2) break;
6946
/* If necessary, advance to next segment in data. */
6949
/* How many characters left in this segment to match. */
6952
/* Want how many consecutive characters we can match in
6953
one shot, so, if necessary, adjust the count. */
6954
if (mcnt > dend2 - d2)
6957
/* Compare that many; failure if mismatch, else move
6960
? PREFIX(bcmp_translate) (d, d2, mcnt, translate)
6961
: memcmp (d, d2, mcnt*sizeof(UCHAR_T)))
6963
d += mcnt, d2 += mcnt;
6965
/* Do this because we've match some characters. */
6966
SET_REGS_MATCHED ();
6972
/* begline matches the empty string at the beginning of the string
6973
(unless `not_bol' is set in `bufp'), and, if
6974
`newline_anchor' is set, after newlines. */
6976
DEBUG_PRINT1 ("EXECUTING begline.\n");
6978
if (AT_STRINGS_BEG (d))
6980
if (!bufp->not_bol) break;
6982
else if (d[-1] == '\n' && bufp->newline_anchor)
6986
/* In all other cases, we fail. */
6990
/* endline is the dual of begline. */
6992
DEBUG_PRINT1 ("EXECUTING endline.\n");
6994
if (AT_STRINGS_END (d))
6996
if (!bufp->not_eol) break;
6999
/* We have to ``prefetch'' the next character. */
7000
else if ((d == end1 ? *string2 : *d) == '\n'
7001
&& bufp->newline_anchor)
7008
/* Match at the very beginning of the data. */
7010
DEBUG_PRINT1 ("EXECUTING begbuf.\n");
7011
if (AT_STRINGS_BEG (d))
7016
/* Match at the very end of the data. */
7018
DEBUG_PRINT1 ("EXECUTING endbuf.\n");
7019
if (AT_STRINGS_END (d))
7024
/* on_failure_keep_string_jump is used to optimize `.*\n'. It
7025
pushes NULL as the value for the string on the stack. Then
7026
`pop_failure_point' will keep the current value for the
7027
string, instead of restoring it. To see why, consider
7028
matching `foo\nbar' against `.*\n'. The .* matches the foo;
7029
then the . fails against the \n. But the next thing we want
7030
to do is match the \n against the \n; if we restored the
7031
string value, we would be back at the foo.
7033
Because this is used only in specific cases, we don't need to
7034
check all the things that `on_failure_jump' does, to make
7035
sure the right things get saved on the stack. Hence we don't
7036
share its code. The only reason to push anything on the
7037
stack at all is that otherwise we would have to change
7038
`anychar's code to do something besides goto fail in this
7039
case; that seems worse than this. */
7040
case on_failure_keep_string_jump:
7041
DEBUG_PRINT1 ("EXECUTING on_failure_keep_string_jump");
7043
EXTRACT_NUMBER_AND_INCR (mcnt, p);
7045
DEBUG_PRINT3 (" %d (to %p):\n", mcnt, p + mcnt);
7047
DEBUG_PRINT3 (" %d (to 0x%x):\n", mcnt, p + mcnt);
7050
PUSH_FAILURE_POINT (p + mcnt, NULL, -2);
7054
/* Uses of on_failure_jump:
7056
Each alternative starts with an on_failure_jump that points
7057
to the beginning of the next alternative. Each alternative
7058
except the last ends with a jump that in effect jumps past
7059
the rest of the alternatives. (They really jump to the
7060
ending jump of the following alternative, because tensioning
7061
these jumps is a hassle.)
7063
Repeats start with an on_failure_jump that points past both
7064
the repetition text and either the following jump or
7065
pop_failure_jump back to this on_failure_jump. */
7066
case on_failure_jump:
7068
DEBUG_PRINT1 ("EXECUTING on_failure_jump");
7070
EXTRACT_NUMBER_AND_INCR (mcnt, p);
7072
DEBUG_PRINT3 (" %d (to %p)", mcnt, p + mcnt);
7074
DEBUG_PRINT3 (" %d (to 0x%x)", mcnt, p + mcnt);
7077
/* If this on_failure_jump comes right before a group (i.e.,
7078
the original * applied to a group), save the information
7079
for that group and all inner ones, so that if we fail back
7080
to this point, the group's information will be correct.
7081
For example, in \(a*\)*\1, we need the preceding group,
7082
and in \(zz\(a*\)b*\)\2, we need the inner group. */
7084
/* We can't use `p' to check ahead because we push
7085
a failure point to `p + mcnt' after we do this. */
7088
/* We need to skip no_op's before we look for the
7089
start_memory in case this on_failure_jump is happening as
7090
the result of a completed succeed_n, as in \(a\)\{1,3\}b\1
7092
while (p1 < pend && (re_opcode_t) *p1 == no_op)
7095
if (p1 < pend && (re_opcode_t) *p1 == start_memory)
7097
/* We have a new highest active register now. This will
7098
get reset at the start_memory we are about to get to,
7099
but we will have saved all the registers relevant to
7100
this repetition op, as described above. */
7101
highest_active_reg = *(p1 + 1) + *(p1 + 2);
7102
if (lowest_active_reg == NO_LOWEST_ACTIVE_REG)
7103
lowest_active_reg = *(p1 + 1);
7106
DEBUG_PRINT1 (":\n");
7107
PUSH_FAILURE_POINT (p + mcnt, d, -2);
7111
/* A smart repeat ends with `maybe_pop_jump'.
7112
We change it to either `pop_failure_jump' or `jump'. */
7113
case maybe_pop_jump:
7114
EXTRACT_NUMBER_AND_INCR (mcnt, p);
7115
DEBUG_PRINT2 ("EXECUTING maybe_pop_jump %d.\n", mcnt);
7117
register UCHAR_T *p2 = p;
7119
/* Compare the beginning of the repeat with what in the
7120
pattern follows its end. If we can establish that there
7121
is nothing that they would both match, i.e., that we
7122
would have to backtrack because of (as in, e.g., `a*a')
7123
then we can change to pop_failure_jump, because we'll
7124
never have to backtrack.
7126
This is not true in the case of alternatives: in
7127
`(a|ab)*' we do need to backtrack to the `ab' alternative
7128
(e.g., if the string was `ab'). But instead of trying to
7129
detect that here, the alternative has put on a dummy
7130
failure point which is what we will end up popping. */
7132
/* Skip over open/close-group commands.
7133
If what follows this loop is a ...+ construct,
7134
look at what begins its body, since we will have to
7135
match at least one of that. */
7139
&& ((re_opcode_t) *p2 == stop_memory
7140
|| (re_opcode_t) *p2 == start_memory))
7142
else if (p2 + 2 + 2 * OFFSET_ADDRESS_SIZE < pend
7143
&& (re_opcode_t) *p2 == dummy_failure_jump)
7144
p2 += 2 + 2 * OFFSET_ADDRESS_SIZE;
7150
/* p1[0] ... p1[2] are the `on_failure_jump' corresponding
7151
to the `maybe_finalize_jump' of this case. Examine what
7154
/* If we're at the end of the pattern, we can change. */
7157
/* Consider what happens when matching ":\(.*\)"
7158
against ":/". I don't really understand this code
7160
p[-(1+OFFSET_ADDRESS_SIZE)] = (UCHAR_T)
7163
(" End of pattern: change to `pop_failure_jump'.\n");
7166
else if ((re_opcode_t) *p2 == exactn
7168
|| (re_opcode_t) *p2 == exactn_bin
7170
|| (bufp->newline_anchor && (re_opcode_t) *p2 == endline))
7173
= *p2 == (UCHAR_T) endline ? '\n' : p2[2];
7175
if (((re_opcode_t) p1[1+OFFSET_ADDRESS_SIZE] == exactn
7177
|| (re_opcode_t) p1[1+OFFSET_ADDRESS_SIZE] == exactn_bin
7179
) && p1[3+OFFSET_ADDRESS_SIZE] != c)
7181
p[-(1+OFFSET_ADDRESS_SIZE)] = (UCHAR_T)
7184
DEBUG_PRINT3 (" %C != %C => pop_failure_jump.\n",
7186
(wint_t) p1[3+OFFSET_ADDRESS_SIZE]);
7188
DEBUG_PRINT3 (" %c != %c => pop_failure_jump.\n",
7190
(char) p1[3+OFFSET_ADDRESS_SIZE]);
7195
else if ((re_opcode_t) p1[3] == charset
7196
|| (re_opcode_t) p1[3] == charset_not)
7198
int not = (re_opcode_t) p1[3] == charset_not;
7200
if (c < (unsigned) (p1[4] * BYTEWIDTH)
7201
&& p1[5 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH)))
7204
/* `not' is equal to 1 if c would match, which means
7205
that we can't change to pop_failure_jump. */
7208
p[-3] = (unsigned char) pop_failure_jump;
7209
DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
7212
#endif /* not WCHAR */
7215
else if ((re_opcode_t) *p2 == charset)
7217
/* We win if the first character of the loop is not part
7219
if ((re_opcode_t) p1[3] == exactn
7220
&& ! ((int) p2[1] * BYTEWIDTH > (int) p1[5]
7221
&& (p2[2 + p1[5] / BYTEWIDTH]
7222
& (1 << (p1[5] % BYTEWIDTH)))))
7224
p[-3] = (unsigned char) pop_failure_jump;
7225
DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
7228
else if ((re_opcode_t) p1[3] == charset_not)
7231
/* We win if the charset_not inside the loop
7232
lists every character listed in the charset after. */
7233
for (idx = 0; idx < (int) p2[1]; idx++)
7234
if (! (p2[2 + idx] == 0
7235
|| (idx < (int) p1[4]
7236
&& ((p2[2 + idx] & ~ p1[5 + idx]) == 0))))
7241
p[-3] = (unsigned char) pop_failure_jump;
7242
DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
7245
else if ((re_opcode_t) p1[3] == charset)
7248
/* We win if the charset inside the loop
7249
has no overlap with the one after the loop. */
7251
idx < (int) p2[1] && idx < (int) p1[4];
7253
if ((p2[2 + idx] & p1[5 + idx]) != 0)
7256
if (idx == p2[1] || idx == p1[4])
7258
p[-3] = (unsigned char) pop_failure_jump;
7259
DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
7263
#endif /* not WCHAR */
7265
p -= OFFSET_ADDRESS_SIZE; /* Point at relative address again. */
7266
if ((re_opcode_t) p[-1] != pop_failure_jump)
7268
p[-1] = (UCHAR_T) jump;
7269
DEBUG_PRINT1 (" Match => jump.\n");
7270
goto unconditional_jump;
7272
/* Note fall through. */
7275
/* The end of a simple repeat has a pop_failure_jump back to
7276
its matching on_failure_jump, where the latter will push a
7277
failure point. The pop_failure_jump takes off failure
7278
points put on by this pop_failure_jump's matching
7279
on_failure_jump; we got through the pattern to here from the
7280
matching on_failure_jump, so didn't fail. */
7281
case pop_failure_jump:
7283
/* We need to pass separate storage for the lowest and
7284
highest registers, even though we don't care about the
7285
actual values. Otherwise, we will restore only one
7286
register from the stack, since lowest will == highest in
7287
`pop_failure_point'. */
7288
active_reg_t dummy_low_reg, dummy_high_reg;
7289
UCHAR_T *pdummy = NULL;
7290
const CHAR_T *sdummy = NULL;
7292
DEBUG_PRINT1 ("EXECUTING pop_failure_jump.\n");
7293
POP_FAILURE_POINT (sdummy, pdummy,
7294
dummy_low_reg, dummy_high_reg,
7295
reg_dummy, reg_dummy, reg_info_dummy);
7297
/* Note fall through. */
7301
DEBUG_PRINT2 ("\n%p: ", p);
7303
DEBUG_PRINT2 ("\n0x%x: ", p);
7305
/* Note fall through. */
7307
/* Unconditionally jump (without popping any failure points). */
7309
EXTRACT_NUMBER_AND_INCR (mcnt, p); /* Get the amount to jump. */
7310
DEBUG_PRINT2 ("EXECUTING jump %d ", mcnt);
7311
p += mcnt; /* Do the jump. */
7313
DEBUG_PRINT2 ("(to %p).\n", p);
7315
DEBUG_PRINT2 ("(to 0x%x).\n", p);
7320
/* We need this opcode so we can detect where alternatives end
7321
in `group_match_null_string_p' et al. */
7323
DEBUG_PRINT1 ("EXECUTING jump_past_alt.\n");
7324
goto unconditional_jump;
7327
/* Normally, the on_failure_jump pushes a failure point, which
7328
then gets popped at pop_failure_jump. We will end up at
7329
pop_failure_jump, also, and with a pattern of, say, `a+', we
7330
are skipping over the on_failure_jump, so we have to push
7331
something meaningless for pop_failure_jump to pop. */
7332
case dummy_failure_jump:
7333
DEBUG_PRINT1 ("EXECUTING dummy_failure_jump.\n");
7334
/* It doesn't matter what we push for the string here. What
7335
the code at `fail' tests is the value for the pattern. */
7336
PUSH_FAILURE_POINT (NULL, NULL, -2);
7337
goto unconditional_jump;
7340
/* At the end of an alternative, we need to push a dummy failure
7341
point in case we are followed by a `pop_failure_jump', because
7342
we don't want the failure point for the alternative to be
7343
popped. For example, matching `(a|ab)*' against `aab'
7344
requires that we match the `ab' alternative. */
7345
case push_dummy_failure:
7346
DEBUG_PRINT1 ("EXECUTING push_dummy_failure.\n");
7347
/* See comments just above at `dummy_failure_jump' about the
7349
PUSH_FAILURE_POINT (NULL, NULL, -2);
7352
/* Have to succeed matching what follows at least n times.
7353
After that, handle like `on_failure_jump'. */
7355
EXTRACT_NUMBER (mcnt, p + OFFSET_ADDRESS_SIZE);
7356
DEBUG_PRINT2 ("EXECUTING succeed_n %d.\n", mcnt);
7359
/* Originally, this is how many times we HAVE to succeed. */
7363
p += OFFSET_ADDRESS_SIZE;
7364
STORE_NUMBER_AND_INCR (p, mcnt);
7366
DEBUG_PRINT3 (" Setting %p to %d.\n", p - OFFSET_ADDRESS_SIZE
7369
DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p - OFFSET_ADDRESS_SIZE
7376
DEBUG_PRINT2 (" Setting two bytes from %p to no_op.\n",
7377
p + OFFSET_ADDRESS_SIZE);
7379
DEBUG_PRINT2 (" Setting two bytes from 0x%x to no_op.\n",
7380
p + OFFSET_ADDRESS_SIZE);
7384
p[1] = (UCHAR_T) no_op;
7386
p[2] = (UCHAR_T) no_op;
7387
p[3] = (UCHAR_T) no_op;
7394
EXTRACT_NUMBER (mcnt, p + OFFSET_ADDRESS_SIZE);
7395
DEBUG_PRINT2 ("EXECUTING jump_n %d.\n", mcnt);
7397
/* Originally, this is how many times we CAN jump. */
7401
STORE_NUMBER (p + OFFSET_ADDRESS_SIZE, mcnt);
7404
DEBUG_PRINT3 (" Setting %p to %d.\n", p + OFFSET_ADDRESS_SIZE,
7407
DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p + OFFSET_ADDRESS_SIZE,
7410
goto unconditional_jump;
7412
/* If don't have to jump any more, skip over the rest of command. */
7414
p += 2 * OFFSET_ADDRESS_SIZE;
7419
DEBUG_PRINT1 ("EXECUTING set_number_at.\n");
7421
EXTRACT_NUMBER_AND_INCR (mcnt, p);
7423
EXTRACT_NUMBER_AND_INCR (mcnt, p);
7425
DEBUG_PRINT3 (" Setting %p to %d.\n", p1, mcnt);
7427
DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p1, mcnt);
7429
STORE_NUMBER (p1, mcnt);
7434
/* The DEC Alpha C compiler 3.x generates incorrect code for the
7435
test WORDCHAR_P (d - 1) != WORDCHAR_P (d) in the expansion of
7436
AT_WORD_BOUNDARY, so this code is disabled. Expanding the
7437
macro and introducing temporary variables works around the bug. */
7440
DEBUG_PRINT1 ("EXECUTING wordbound.\n");
7441
if (AT_WORD_BOUNDARY (d))
7446
DEBUG_PRINT1 ("EXECUTING notwordbound.\n");
7447
if (AT_WORD_BOUNDARY (d))
7453
boolean prevchar, thischar;
7455
DEBUG_PRINT1 ("EXECUTING wordbound.\n");
7456
if (AT_STRINGS_BEG (d) || AT_STRINGS_END (d))
7459
prevchar = WORDCHAR_P (d - 1);
7460
thischar = WORDCHAR_P (d);
7461
if (prevchar != thischar)
7468
boolean prevchar, thischar;
7470
DEBUG_PRINT1 ("EXECUTING notwordbound.\n");
7471
if (AT_STRINGS_BEG (d) || AT_STRINGS_END (d))
7474
prevchar = WORDCHAR_P (d - 1);
7475
thischar = WORDCHAR_P (d);
7476
if (prevchar != thischar)
7483
DEBUG_PRINT1 ("EXECUTING wordbeg.\n");
7484
if (!AT_STRINGS_END (d) && WORDCHAR_P (d)
7485
&& (AT_STRINGS_BEG (d) || !WORDCHAR_P (d - 1)))
7490
DEBUG_PRINT1 ("EXECUTING wordend.\n");
7491
if (!AT_STRINGS_BEG (d) && WORDCHAR_P (d - 1)
7492
&& (AT_STRINGS_END (d) || !WORDCHAR_P (d)))
7498
DEBUG_PRINT1 ("EXECUTING before_dot.\n");
7499
if (PTR_CHAR_POS ((unsigned char *) d) >= point)
7504
DEBUG_PRINT1 ("EXECUTING at_dot.\n");
7505
if (PTR_CHAR_POS ((unsigned char *) d) != point)
7510
DEBUG_PRINT1 ("EXECUTING after_dot.\n");
7511
if (PTR_CHAR_POS ((unsigned char *) d) <= point)
7516
DEBUG_PRINT2 ("EXECUTING syntaxspec %d.\n", mcnt);
7521
DEBUG_PRINT1 ("EXECUTING Emacs wordchar.\n");
7525
/* Can't use *d++ here; SYNTAX may be an unsafe macro. */
7527
if (SYNTAX (d[-1]) != (enum syntaxcode) mcnt)
7529
SET_REGS_MATCHED ();
7533
DEBUG_PRINT2 ("EXECUTING notsyntaxspec %d.\n", mcnt);
7535
goto matchnotsyntax;
7538
DEBUG_PRINT1 ("EXECUTING Emacs notwordchar.\n");
7542
/* Can't use *d++ here; SYNTAX may be an unsafe macro. */
7544
if (SYNTAX (d[-1]) == (enum syntaxcode) mcnt)
7546
SET_REGS_MATCHED ();
7549
#else /* not emacs */
7551
DEBUG_PRINT1 ("EXECUTING non-Emacs wordchar.\n");
7553
if (!WORDCHAR_P (d))
7555
SET_REGS_MATCHED ();
7560
DEBUG_PRINT1 ("EXECUTING non-Emacs notwordchar.\n");
7564
SET_REGS_MATCHED ();
7567
#endif /* not emacs */
7572
continue; /* Successfully executed one pattern command; keep going. */
7575
/* We goto here if a matching operation fails. */
7577
if (!FAIL_STACK_EMPTY ())
7578
{ /* A restart point is known. Restore to that state. */
7579
DEBUG_PRINT1 ("\nFAIL:\n");
7580
POP_FAILURE_POINT (d, p,
7581
lowest_active_reg, highest_active_reg,
7582
regstart, regend, reg_info);
7584
/* If this failure point is a dummy, try the next one. */
7588
/* If we failed to the end of the pattern, don't examine *p. */
7592
boolean is_a_jump_n = false;
7594
/* If failed to a backwards jump that's part of a repetition
7595
loop, need to pop this failure point and use the next one. */
7596
switch ((re_opcode_t) *p)
7600
case maybe_pop_jump:
7601
case pop_failure_jump:
7604
EXTRACT_NUMBER_AND_INCR (mcnt, p1);
7607
if ((is_a_jump_n && (re_opcode_t) *p1 == succeed_n)
7609
&& (re_opcode_t) *p1 == on_failure_jump))
7617
if (d >= string1 && d <= end1)
7621
break; /* Matching at this starting point really fails. */
7625
goto restore_best_regs;
7629
return -1; /* Failure to match. */
7632
/* Subroutine definitions for re_match_2. */
7635
/* We are passed P pointing to a register number after a start_memory.
7637
Return true if the pattern up to the corresponding stop_memory can
7638
match the empty string, and false otherwise.
7640
If we find the matching stop_memory, sets P to point to one past its number.
7641
Otherwise, sets P to an undefined byte less than or equal to END.
7643
We don't handle duplicates properly (yet). */
7646
PREFIX(group_match_null_string_p) (p, end, reg_info)
7648
PREFIX(register_info_type) *reg_info;
7651
/* Point to after the args to the start_memory. */
7652
UCHAR_T *p1 = *p + 2;
7656
/* Skip over opcodes that can match nothing, and return true or
7657
false, as appropriate, when we get to one that can't, or to the
7658
matching stop_memory. */
7660
switch ((re_opcode_t) *p1)
7662
/* Could be either a loop or a series of alternatives. */
7663
case on_failure_jump:
7665
EXTRACT_NUMBER_AND_INCR (mcnt, p1);
7667
/* If the next operation is not a jump backwards in the
7672
/* Go through the on_failure_jumps of the alternatives,
7673
seeing if any of the alternatives cannot match nothing.
7674
The last alternative starts with only a jump,
7675
whereas the rest start with on_failure_jump and end
7676
with a jump, e.g., here is the pattern for `a|b|c':
7678
/on_failure_jump/0/6/exactn/1/a/jump_past_alt/0/6
7679
/on_failure_jump/0/6/exactn/1/b/jump_past_alt/0/3
7682
So, we have to first go through the first (n-1)
7683
alternatives and then deal with the last one separately. */
7686
/* Deal with the first (n-1) alternatives, which start
7687
with an on_failure_jump (see above) that jumps to right
7688
past a jump_past_alt. */
7690
while ((re_opcode_t) p1[mcnt-(1+OFFSET_ADDRESS_SIZE)] ==
7693
/* `mcnt' holds how many bytes long the alternative
7694
is, including the ending `jump_past_alt' and
7697
if (!PREFIX(alt_match_null_string_p) (p1, p1 + mcnt -
7698
(1 + OFFSET_ADDRESS_SIZE),
7702
/* Move to right after this alternative, including the
7706
/* Break if it's the beginning of an n-th alternative
7707
that doesn't begin with an on_failure_jump. */
7708
if ((re_opcode_t) *p1 != on_failure_jump)
7711
/* Still have to check that it's not an n-th
7712
alternative that starts with an on_failure_jump. */
7714
EXTRACT_NUMBER_AND_INCR (mcnt, p1);
7715
if ((re_opcode_t) p1[mcnt-(1+OFFSET_ADDRESS_SIZE)] !=
7718
/* Get to the beginning of the n-th alternative. */
7719
p1 -= 1 + OFFSET_ADDRESS_SIZE;
7724
/* Deal with the last alternative: go back and get number
7725
of the `jump_past_alt' just before it. `mcnt' contains
7726
the length of the alternative. */
7727
EXTRACT_NUMBER (mcnt, p1 - OFFSET_ADDRESS_SIZE);
7729
if (!PREFIX(alt_match_null_string_p) (p1, p1 + mcnt, reg_info))
7732
p1 += mcnt; /* Get past the n-th alternative. */
7738
assert (p1[1] == **p);
7744
if (!PREFIX(common_op_match_null_string_p) (&p1, end, reg_info))
7747
} /* while p1 < end */
7750
} /* group_match_null_string_p */
7753
/* Similar to group_match_null_string_p, but doesn't deal with alternatives:
7754
It expects P to be the first byte of a single alternative and END one
7755
byte past the last. The alternative can contain groups. */
7758
PREFIX(alt_match_null_string_p) (p, end, reg_info)
7760
PREFIX(register_info_type) *reg_info;
7767
/* Skip over opcodes that can match nothing, and break when we get
7768
to one that can't. */
7770
switch ((re_opcode_t) *p1)
7773
case on_failure_jump:
7775
EXTRACT_NUMBER_AND_INCR (mcnt, p1);
7780
if (!PREFIX(common_op_match_null_string_p) (&p1, end, reg_info))
7783
} /* while p1 < end */
7786
} /* alt_match_null_string_p */
7789
/* Deals with the ops common to group_match_null_string_p and
7790
alt_match_null_string_p.
7792
Sets P to one after the op and its arguments, if any. */
7795
PREFIX(common_op_match_null_string_p) (p, end, reg_info)
7797
PREFIX(register_info_type) *reg_info;
7804
switch ((re_opcode_t) *p1++)
7824
assert (reg_no > 0 && reg_no <= MAX_REGNUM);
7825
ret = PREFIX(group_match_null_string_p) (&p1, end, reg_info);
7827
/* Have to set this here in case we're checking a group which
7828
contains a group and a back reference to it. */
7830
if (REG_MATCH_NULL_STRING_P (reg_info[reg_no]) == MATCH_NULL_UNSET_VALUE)
7831
REG_MATCH_NULL_STRING_P (reg_info[reg_no]) = ret;
7837
/* If this is an optimized succeed_n for zero times, make the jump. */
7839
EXTRACT_NUMBER_AND_INCR (mcnt, p1);
7847
/* Get to the number of times to succeed. */
7848
p1 += OFFSET_ADDRESS_SIZE;
7849
EXTRACT_NUMBER_AND_INCR (mcnt, p1);
7853
p1 -= 2 * OFFSET_ADDRESS_SIZE;
7854
EXTRACT_NUMBER_AND_INCR (mcnt, p1);
7862
if (!REG_MATCH_NULL_STRING_P (reg_info[*p1]))
7867
p1 += 2 * OFFSET_ADDRESS_SIZE;
7870
/* All other opcodes mean we cannot match the empty string. */
7876
} /* common_op_match_null_string_p */
7879
/* Return zero if TRANSLATE[S1] and TRANSLATE[S2] are identical for LEN
7880
bytes; nonzero otherwise. */
7883
PREFIX(bcmp_translate) (s1, s2, len, translate)
7884
const CHAR_T *s1, *s2;
7886
RE_TRANSLATE_TYPE translate;
7888
register const UCHAR_T *p1 = (const UCHAR_T *) s1;
7889
register const UCHAR_T *p2 = (const UCHAR_T *) s2;
7893
if (((*p1<=0xff)?translate[*p1++]:*p1++)
7894
!= ((*p2<=0xff)?translate[*p2++]:*p2++))
7897
if (translate[*p1++] != translate[*p2++]) return 1;
7905
#else /* not INSIDE_RECURSION */
7907
/* Entry points for GNU code. */
7909
/* re_compile_pattern is the GNU regular expression compiler: it
7910
compiles PATTERN (of length SIZE) and puts the result in BUFP.
7911
Returns 0 if the pattern was valid, otherwise an error string.
7913
Assumes the `allocated' (and perhaps `buffer') and `translate' fields
7914
are set in BUFP on entry.
7916
We call regex_compile to do the actual compilation. */
7919
re_compile_pattern (pattern, length, bufp)
7920
const char *pattern;
7922
struct re_pattern_buffer *bufp;
7926
/* GNU code is written to assume at least RE_NREGS registers will be set
7927
(and at least one extra will be -1). */
7928
bufp->regs_allocated = REGS_UNALLOCATED;
7930
/* And GNU code determines whether or not to get register information
7931
by passing null for the REGS argument to re_match, etc., not by
7935
/* Match anchors at newline. */
7936
bufp->newline_anchor = 1;
7939
if (MB_CUR_MAX != 1)
7940
ret = wcs_regex_compile (pattern, length, re_syntax_options, bufp);
7943
ret = byte_regex_compile (pattern, length, re_syntax_options, bufp);
7947
return gettext (re_error_msgid + re_error_msgid_idx[(int) ret]);
7950
weak_alias (__re_compile_pattern, re_compile_pattern)
7953
/* Entry points compatible with 4.2 BSD regex library. We don't define
7954
them unless specifically requested. */
7956
#if defined _REGEX_RE_COMP || defined _LIBC
7958
/* BSD has one and only one pattern buffer. */
7959
static struct re_pattern_buffer re_comp_buf;
7963
/* Make these definitions weak in libc, so POSIX programs can redefine
7964
these names if they don't use our functions, and still use
7965
regcomp/regexec below without link errors. */
7975
if (!re_comp_buf.buffer)
7976
return gettext ("No previous regular expression");
7980
if (!re_comp_buf.buffer)
7982
re_comp_buf.buffer = (unsigned char *) malloc (200);
7983
if (re_comp_buf.buffer == NULL)
7984
return (char *) gettext (re_error_msgid
7985
+ re_error_msgid_idx[(int) REG_ESPACE]);
7986
re_comp_buf.allocated = 200;
7988
re_comp_buf.fastmap = (char *) malloc (1 << BYTEWIDTH);
7989
if (re_comp_buf.fastmap == NULL)
7990
return (char *) gettext (re_error_msgid
7991
+ re_error_msgid_idx[(int) REG_ESPACE]);
7994
/* Since `re_exec' always passes NULL for the `regs' argument, we
7995
don't need to initialize the pattern buffer fields which affect it. */
7997
/* Match anchors at newlines. */
7998
re_comp_buf.newline_anchor = 1;
8001
if (MB_CUR_MAX != 1)
8002
ret = wcs_regex_compile (s, strlen (s), re_syntax_options, &re_comp_buf);
8005
ret = byte_regex_compile (s, strlen (s), re_syntax_options, &re_comp_buf);
8010
/* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
8011
return (char *) gettext (re_error_msgid + re_error_msgid_idx[(int) ret]);
8022
const int len = strlen (s);
8024
0 <= re_search (&re_comp_buf, s, len, 0, len, (struct re_registers *) 0);
8027
#endif /* _REGEX_RE_COMP */
8029
/* POSIX.2 functions. Don't define these for Emacs. */
8033
/* regcomp takes a regular expression as a string and compiles it.
8035
PREG is a regex_t *. We do not expect any fields to be initialized,
8036
since POSIX says we shouldn't. Thus, we set
8038
`buffer' to the compiled pattern;
8039
`used' to the length of the compiled pattern;
8040
`syntax' to RE_SYNTAX_POSIX_EXTENDED if the
8041
REG_EXTENDED bit in CFLAGS is set; otherwise, to
8042
RE_SYNTAX_POSIX_BASIC;
8043
`newline_anchor' to REG_NEWLINE being set in CFLAGS;
8044
`fastmap' to an allocated space for the fastmap;
8045
`fastmap_accurate' to zero;
8046
`re_nsub' to the number of subexpressions in PATTERN.
8048
PATTERN is the address of the pattern string.
8050
CFLAGS is a series of bits which affect compilation.
8052
If REG_EXTENDED is set, we use POSIX extended syntax; otherwise, we
8053
use POSIX basic syntax.
8055
If REG_NEWLINE is set, then . and [^...] don't match newline.
8056
Also, regexec will try a match beginning after every newline.
8058
If REG_ICASE is set, then we considers upper- and lowercase
8059
versions of letters to be equivalent when matching.
8061
If REG_NOSUB is set, then when PREG is passed to regexec, that
8062
routine will report only success or failure, and nothing about the
8065
It returns 0 if it succeeds, nonzero if it doesn't. (See regex.h for
8066
the return codes and their meanings.) */
8069
regcomp (preg, pattern, cflags)
8071
const char *pattern;
8076
= (cflags & REG_EXTENDED) ?
8077
RE_SYNTAX_POSIX_EXTENDED : RE_SYNTAX_POSIX_BASIC;
8079
/* regex_compile will allocate the space for the compiled pattern. */
8081
preg->allocated = 0;
8084
/* Try to allocate space for the fastmap. */
8085
preg->fastmap = (char *) malloc (1 << BYTEWIDTH);
8087
if (cflags & REG_ICASE)
8092
= (RE_TRANSLATE_TYPE) malloc (CHAR_SET_SIZE
8093
* sizeof (*(RE_TRANSLATE_TYPE)0));
8094
if (preg->translate == NULL)
8095
return (int) REG_ESPACE;
8097
/* Map uppercase characters to corresponding lowercase ones. */
8098
for (i = 0; i < CHAR_SET_SIZE; i++)
8099
preg->translate[i] = ISUPPER (i) ? TOLOWER (i) : i;
8102
preg->translate = NULL;
8104
/* If REG_NEWLINE is set, newlines are treated differently. */
8105
if (cflags & REG_NEWLINE)
8106
{ /* REG_NEWLINE implies neither . nor [^...] match newline. */
8107
syntax &= ~RE_DOT_NEWLINE;
8108
syntax |= RE_HAT_LISTS_NOT_NEWLINE;
8109
/* It also changes the matching behavior. */
8110
preg->newline_anchor = 1;
8113
preg->newline_anchor = 0;
8115
preg->no_sub = !!(cflags & REG_NOSUB);
8117
/* POSIX says a null character in the pattern terminates it, so we
8118
can use strlen here in compiling the pattern. */
8120
if (MB_CUR_MAX != 1)
8121
ret = wcs_regex_compile (pattern, strlen (pattern), syntax, preg);
8124
ret = byte_regex_compile (pattern, strlen (pattern), syntax, preg);
8126
/* POSIX doesn't distinguish between an unmatched open-group and an
8127
unmatched close-group: both are REG_EPAREN. */
8128
if (ret == REG_ERPAREN) ret = REG_EPAREN;
8130
if (ret == REG_NOERROR && preg->fastmap)
8132
/* Compute the fastmap now, since regexec cannot modify the pattern
8134
if (re_compile_fastmap (preg) == -2)
8136
/* Some error occurred while computing the fastmap, just forget
8138
free (preg->fastmap);
8139
preg->fastmap = NULL;
8146
weak_alias (__regcomp, regcomp)
8150
/* regexec searches for a given pattern, specified by PREG, in the
8153
If NMATCH is zero or REG_NOSUB was set in the cflags argument to
8154
`regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at
8155
least NMATCH elements, and we set them to the offsets of the
8156
corresponding matched substrings.
8158
EFLAGS specifies `execution flags' which affect matching: if
8159
REG_NOTBOL is set, then ^ does not match at the beginning of the
8160
string; if REG_NOTEOL is set, then $ does not match at the end.
8162
We return 0 if we find a match and REG_NOMATCH if not. */
8165
regexec (preg, string, nmatch, pmatch, eflags)
8166
const regex_t *preg;
8169
regmatch_t pmatch[];
8173
struct re_registers regs;
8174
regex_t private_preg;
8175
int len = strlen (string);
8176
boolean want_reg_info = !preg->no_sub && nmatch > 0;
8178
private_preg = *preg;
8180
private_preg.not_bol = !!(eflags & REG_NOTBOL);
8181
private_preg.not_eol = !!(eflags & REG_NOTEOL);
8183
/* The user has told us exactly how many registers to return
8184
information about, via `nmatch'. We have to pass that on to the
8185
matching routines. */
8186
private_preg.regs_allocated = REGS_FIXED;
8190
regs.num_regs = nmatch;
8191
regs.start = TALLOC (nmatch * 2, regoff_t);
8192
if (regs.start == NULL)
8193
return (int) REG_NOMATCH;
8194
regs.end = regs.start + nmatch;
8197
/* Perform the searching operation. */
8198
ret = re_search (&private_preg, string, len,
8199
/* start: */ 0, /* range: */ len,
8200
want_reg_info ? ®s : (struct re_registers *) 0);
8202
/* Copy the register information to the POSIX structure. */
8209
for (r = 0; r < nmatch; r++)
8211
pmatch[r].rm_so = regs.start[r];
8212
pmatch[r].rm_eo = regs.end[r];
8216
/* If we needed the temporary register info, free the space now. */
8220
/* We want zero return to mean success, unlike `re_search'. */
8221
return ret >= 0 ? (int) REG_NOERROR : (int) REG_NOMATCH;
8224
weak_alias (__regexec, regexec)
8228
/* Returns a message corresponding to an error code, ERRCODE, returned
8229
from either regcomp or regexec. We don't use PREG here. */
8232
regerror (errcode, preg, errbuf, errbuf_size)
8234
const regex_t *preg;
8242
|| errcode >= (int) (sizeof (re_error_msgid_idx)
8243
/ sizeof (re_error_msgid_idx[0])))
8244
/* Only error codes returned by the rest of the code should be passed
8245
to this routine. If we are given anything else, or if other regex
8246
code generates an invalid error code, then the program has a bug.
8247
Dump core so we can fix it. */
8250
msg = gettext (re_error_msgid + re_error_msgid_idx[errcode]);
8252
msg_size = strlen (msg) + 1; /* Includes the null. */
8254
if (errbuf_size != 0)
8256
if (msg_size > errbuf_size)
8258
#if defined HAVE_MEMPCPY || defined _LIBC
8259
*((char *) __mempcpy (errbuf, msg, errbuf_size - 1)) = '\0';
8261
memcpy (errbuf, msg, errbuf_size - 1);
8262
errbuf[errbuf_size - 1] = 0;
8266
memcpy (errbuf, msg, msg_size);
8272
weak_alias (__regerror, regerror)
8276
/* Free dynamically allocated space used by PREG. */
8282
if (preg->buffer != NULL)
8283
free (preg->buffer);
8284
preg->buffer = NULL;
8286
preg->allocated = 0;
8289
if (preg->fastmap != NULL)
8290
free (preg->fastmap);
8291
preg->fastmap = NULL;
8292
preg->fastmap_accurate = 0;
8294
if (preg->translate != NULL)
8295
free (preg->translate);
8296
preg->translate = NULL;
8299
weak_alias (__regfree, regfree)
8302
#endif /* not emacs */
8304
#endif /* not INSIDE_RECURSION */
8308
#undef STORE_NUMBER_AND_INCR
8309
#undef EXTRACT_NUMBER
8310
#undef EXTRACT_NUMBER_AND_INCR
8312
#undef DEBUG_PRINT_COMPILED_PATTERN
8313
#undef DEBUG_PRINT_DOUBLE_STRING
8315
#undef INIT_FAIL_STACK
8316
#undef RESET_FAIL_STACK
8317
#undef DOUBLE_FAIL_STACK
8318
#undef PUSH_PATTERN_OP
8319
#undef PUSH_FAILURE_POINTER
8320
#undef PUSH_FAILURE_INT
8321
#undef PUSH_FAILURE_ELT
8322
#undef POP_FAILURE_POINTER
8323
#undef POP_FAILURE_INT
8324
#undef POP_FAILURE_ELT
8327
#undef PUSH_FAILURE_POINT
8328
#undef POP_FAILURE_POINT
8330
#undef REG_UNSET_VALUE
8338
#undef INIT_BUF_SIZE
8339
#undef GET_BUFFER_SPACE
8347
#undef EXTEND_BUFFER
8348
#undef GET_UNSIGNED_NUMBER
8349
#undef FREE_STACK_RETURN
8351
# undef POINTER_TO_OFFSET
8352
# undef MATCHING_IN_FRST_STRING
8354
# undef AT_STRINGS_BEG
8355
# undef AT_STRINGS_END
8358
# undef FREE_VARIABLES
8359
# undef NO_HIGHEST_ACTIVE_REG
8360
# undef NO_LOWEST_ACTIVE_REG
8364
# undef COMPILED_BUFFER_VAR
8365
# undef OFFSET_ADDRESS_SIZE
8366
# undef CHAR_CLASS_SIZE
8373
# define DEFINED_ONCE