1 /* *INDENT-OFF* */ /* keep in sync with glibc */
2 /* Extended regular expression matching and search library,
4 (Implements POSIX draft P1003.2/D11.2, except for some of the
5 internationalization features.)
6 Copyright (C) 1993, 94, 95, 96, 97, 98 Free Software Foundation, Inc.
8 NOTE: The canonical source of this file is maintained with the
9 GNU C Library. Bugs can be reported to bug-glibc@prep.ai.mit.edu.
11 This program is free software; you can redistribute it and/or modify it
12 under the terms of the GNU General Public License as published by the
13 Free Software Foundation; either version 2, or (at your option) any
16 This program is distributed in the hope that it will be useful,
17 but WITHOUT ANY WARRANTY; without even the implied warranty of
18 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
19 GNU General Public License for more details.
21 You should have received a copy of the GNU General Public License
22 along with this program; if not, write to the Free Software Foundation,
23 Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. */
25 /* AIX requires this to be the first thing in the file. */
26 #if defined _AIX && !defined REGEX_MALLOC
38 # if defined __GNUC__ || (defined __STDC__ && __STDC__)
39 # define PARAMS(args) args
41 # define PARAMS(args) ()
43 #endif /* Not PARAMS. */
45 #if defined STDC_HEADERS && !defined emacs
48 /* We need this for `gnu-regex.h', and perhaps for the Emacs include files. */
49 # include <sys/types.h>
52 /* For platform which support the ISO C amendement 1 functionality we
53 support user defined character classes. */
54 #if defined _LIBC || (defined HAVE_WCTYPE_H && defined HAVE_WCHAR_H)
55 /* Solaris 2.5 has a bug: <wchar.h> must be included before <wctype.h>. */
60 /* This is for other GNU distributions with internationalized messages. */
61 /* CYGNUS LOCAL: ../intl will handle this for us */
65 # define gettext(msgid) (msgid)
69 /* This define is so xgettext can find the internationalizable
71 # define gettext_noop(String) String
74 /* The `emacs' switch turns on certain matching commands
75 that make sense only in Emacs. */
84 /* If we are not linking with Emacs proper,
85 we can't use the relocating allocator
86 even if config.h says that we can. */
89 # if defined STDC_HEADERS || defined _LIBC
96 /* When used in Emacs's lib-src, we need to get bzero and bcopy somehow.
97 If nothing else has been done, use the method below. */
98 # ifdef INHIBIT_STRING_HEADER
99 # if !(defined HAVE_BZERO && defined HAVE_BCOPY)
100 # if !defined bzero && !defined bcopy
101 # undef INHIBIT_STRING_HEADER
106 /* This is the normal way of making sure we have a bcopy and a bzero.
107 This is used in most programs--a few other programs avoid this
108 by defining INHIBIT_STRING_HEADER. */
109 # ifndef INHIBIT_STRING_HEADER
110 # if defined HAVE_STRING_H || defined STDC_HEADERS || defined _LIBC
114 # define bzero(s, n) (memset (s, '\0', n), (s))
116 # define bzero(s, n) __bzero (s, n)
120 # include <strings.h>
122 # define memcmp(s1, s2, n) bcmp (s1, s2, n)
125 # define memcpy(d, s, n) (bcopy (s, d, n), (d))
130 /* Define the syntax stuff for \<, \>, etc. */
132 /* This must be nonzero for the wordchar and notwordchar pattern
133 commands in re_match_2. */
138 # ifdef SWITCH_ENUM_BUG
139 # define SWITCH_ENUM_CAST(x) ((int)(x))
141 # define SWITCH_ENUM_CAST(x) (x)
144 /* How many characters in the character set. */
145 # define CHAR_SET_SIZE 256
147 /* GDB LOCAL: define _REGEX_RE_COMP to get BSD style re_comp and re_exec */
148 #ifndef _REGEX_RE_COMP
149 #define _REGEX_RE_COMP
154 extern char *re_syntax_table
;
156 # else /* not SYNTAX_TABLE */
158 static char re_syntax_table
[CHAR_SET_SIZE
];
169 bzero (re_syntax_table
, sizeof re_syntax_table
);
171 for (c
= 'a'; c
<= 'z'; c
++)
172 re_syntax_table
[c
] = Sword
;
174 for (c
= 'A'; c
<= 'Z'; c
++)
175 re_syntax_table
[c
] = Sword
;
177 for (c
= '0'; c
<= '9'; c
++)
178 re_syntax_table
[c
] = Sword
;
180 re_syntax_table
['_'] = Sword
;
185 # endif /* not SYNTAX_TABLE */
187 # define SYNTAX(c) re_syntax_table[c]
189 #endif /* not emacs */
191 /* Get the interface, including the syntax bits. */
192 /* CYGNUS LOCAL: call it gnu-regex.h, not regex.h, to avoid name conflicts */
193 #include "gnu-regex.h"
195 /* isalpha etc. are used for the character classes. */
198 /* Jim Meyering writes:
200 "... Some ctype macros are valid only for character codes that
201 isascii says are ASCII (SGI's IRIX-4.0.5 is one such system --when
202 using /bin/cc or gcc but without giving an ansi option). So, all
203 ctype uses should be through macros like ISPRINT... If
204 STDC_HEADERS is defined, then autoconf has verified that the ctype
205 macros don't need to be guarded with references to isascii. ...
206 Defining isascii to 1 should let any compiler worth its salt
207 eliminate the && through constant folding."
208 Solaris defines some of these symbols so we must undefine them first. */
211 #if defined STDC_HEADERS || (!defined isascii && !defined HAVE_ISASCII)
212 # define ISASCII(c) 1
214 # define ISASCII(c) isascii(c)
218 # define ISBLANK(c) (ISASCII (c) && isblank (c))
220 # define ISBLANK(c) ((c) == ' ' || (c) == '\t')
223 # define ISGRAPH(c) (ISASCII (c) && isgraph (c))
225 # define ISGRAPH(c) (ISASCII (c) && isprint (c) && !isspace (c))
229 #define ISPRINT(c) (ISASCII (c) && isprint (c))
230 #define ISDIGIT(c) (ISASCII (c) && isdigit (c))
231 #define ISALNUM(c) (ISASCII (c) && isalnum (c))
232 #define ISALPHA(c) (ISASCII (c) && isalpha (c))
233 #define ISCNTRL(c) (ISASCII (c) && iscntrl (c))
234 #define ISLOWER(c) (ISASCII (c) && islower (c))
235 #define ISPUNCT(c) (ISASCII (c) && ispunct (c))
236 #define ISSPACE(c) (ISASCII (c) && isspace (c))
237 #define ISUPPER(c) (ISASCII (c) && isupper (c))
238 #define ISXDIGIT(c) (ISASCII (c) && isxdigit (c))
241 # define NULL (void *)0
244 /* We remove any previous definition of `SIGN_EXTEND_CHAR',
245 since ours (we hope) works properly with all combinations of
246 machines, compilers, `char' and `unsigned char' argument types.
247 (Per Bothner suggested the basic approach.) */
248 #undef SIGN_EXTEND_CHAR
250 # define SIGN_EXTEND_CHAR(c) ((signed char) (c))
251 #else /* not __STDC__ */
252 /* As in Harbison and Steele. */
253 # define SIGN_EXTEND_CHAR(c) ((((unsigned char) (c)) ^ 128) - 128)
256 /* Should we use malloc or alloca? If REGEX_MALLOC is not defined, we
257 use `alloca' instead of `malloc'. This is because using malloc in
258 re_search* or re_match* could cause memory leaks when C-g is used in
259 Emacs; also, malloc is slower and causes storage fragmentation. On
260 the other hand, malloc is more portable, and easier to debug.
262 Because we sometimes use alloca, some routines have to be macros,
263 not functions -- `alloca'-allocated space disappears at the end of the
264 function it is called in. */
268 # define REGEX_ALLOCATE malloc
269 # define REGEX_REALLOCATE(source, osize, nsize) realloc (source, nsize)
270 # define REGEX_FREE free
272 #else /* not REGEX_MALLOC */
274 /* Emacs already defines alloca, sometimes. */
277 /* Make alloca work the best possible way. */
279 # define alloca __builtin_alloca
280 # else /* not __GNUC__ */
283 # endif /* HAVE_ALLOCA_H */
284 # endif /* not __GNUC__ */
286 # endif /* not alloca */
288 # define REGEX_ALLOCATE alloca
290 /* Assumes a `char *destination' variable. */
291 # define REGEX_REALLOCATE(source, osize, nsize) \
292 (destination = (char *) alloca (nsize), \
293 memcpy (destination, source, osize))
295 /* No need to do anything to free, after alloca. */
296 # define REGEX_FREE(arg) ((void)0) /* Do nothing! But inhibit gcc warning. */
298 #endif /* not REGEX_MALLOC */
300 /* Define how to allocate the failure stack. */
302 #if defined REL_ALLOC && defined REGEX_MALLOC
304 # define REGEX_ALLOCATE_STACK(size) \
305 r_alloc (&failure_stack_ptr, (size))
306 # define REGEX_REALLOCATE_STACK(source, osize, nsize) \
307 r_re_alloc (&failure_stack_ptr, (nsize))
308 # define REGEX_FREE_STACK(ptr) \
309 r_alloc_free (&failure_stack_ptr)
311 #else /* not using relocating allocator */
315 # define REGEX_ALLOCATE_STACK malloc
316 # define REGEX_REALLOCATE_STACK(source, osize, nsize) realloc (source, nsize)
317 # define REGEX_FREE_STACK free
319 # else /* not REGEX_MALLOC */
321 # define REGEX_ALLOCATE_STACK alloca
323 # define REGEX_REALLOCATE_STACK(source, osize, nsize) \
324 REGEX_REALLOCATE (source, osize, nsize)
325 /* No need to explicitly free anything. */
326 # define REGEX_FREE_STACK(arg)
328 # endif /* not REGEX_MALLOC */
329 #endif /* not using relocating allocator */
332 /* True if `size1' is non-NULL and PTR is pointing anywhere inside
333 `string1' or just past its end. This works if PTR is NULL, which is
335 #define FIRST_STRING_P(ptr) \
336 (size1 && string1 <= (ptr) && (ptr) <= string1 + size1)
338 /* (Re)Allocate N items of type T using malloc, or fail. */
339 #define TALLOC(n, t) ((t *) malloc ((n) * sizeof (t)))
340 #define RETALLOC(addr, n, t) ((addr) = (t *) realloc (addr, (n) * sizeof (t)))
341 #define RETALLOC_IF(addr, n, t) \
342 if (addr) RETALLOC((addr), (n), t); else (addr) = TALLOC ((n), t)
343 #define REGEX_TALLOC(n, t) ((t *) REGEX_ALLOCATE ((n) * sizeof (t)))
345 #define BYTEWIDTH 8 /* In bits. */
347 #define STREQ(s1, s2) ((strcmp (s1, s2) == 0))
351 #define MAX(a, b) ((a) > (b) ? (a) : (b))
352 #define MIN(a, b) ((a) < (b) ? (a) : (b))
354 typedef char boolean
;
358 static int re_match_2_internal
PARAMS ((struct re_pattern_buffer
*bufp
,
359 const char *string1
, int size1
,
360 const char *string2
, int size2
,
362 struct re_registers
*regs
,
365 /* These are the command codes that appear in compiled regular
366 expressions. Some opcodes are followed by argument bytes. A
367 command code can specify any interpretation whatsoever for its
368 arguments. Zero bytes may appear in the compiled regular expression. */
374 /* Succeed right away--no more backtracking. */
377 /* Followed by one byte giving n, then by n literal bytes. */
380 /* Matches any (more or less) character. */
383 /* Matches any one char belonging to specified set. First
384 following byte is number of bitmap bytes. Then come bytes
385 for a bitmap saying which chars are in. Bits in each byte
386 are ordered low-bit-first. A character is in the set if its
387 bit is 1. A character too large to have a bit in the map is
388 automatically not in the set. */
391 /* Same parameters as charset, but match any character that is
392 not one of those specified. */
395 /* Start remembering the text that is matched, for storing in a
396 register. Followed by one byte with the register number, in
397 the range 0 to one less than the pattern buffer's re_nsub
398 field. Then followed by one byte with the number of groups
399 inner to this one. (This last has to be part of the
400 start_memory only because we need it in the on_failure_jump
404 /* Stop remembering the text that is matched and store it in a
405 memory register. Followed by one byte with the register
406 number, in the range 0 to one less than `re_nsub' in the
407 pattern buffer, and one byte with the number of inner groups,
408 just like `start_memory'. (We need the number of inner
409 groups here because we don't have any easy way of finding the
410 corresponding start_memory when we're at a stop_memory.) */
413 /* Match a duplicate of something remembered. Followed by one
414 byte containing the register number. */
417 /* Fail unless at beginning of line. */
420 /* Fail unless at end of line. */
423 /* Succeeds if at beginning of buffer (if emacs) or at beginning
424 of string to be matched (if not). */
427 /* Analogously, for end of buffer/string. */
430 /* Followed by two byte relative address to which to jump. */
433 /* Same as jump, but marks the end of an alternative. */
436 /* Followed by two-byte relative address of place to resume at
437 in case of failure. */
440 /* Like on_failure_jump, but pushes a placeholder instead of the
441 current string position when executed. */
442 on_failure_keep_string_jump
,
444 /* Throw away latest failure point and then jump to following
445 two-byte relative address. */
448 /* Change to pop_failure_jump if know won't have to backtrack to
449 match; otherwise change to jump. This is used to jump
450 back to the beginning of a repeat. If what follows this jump
451 clearly won't match what the repeat does, such that we can be
452 sure that there is no use backtracking out of repetitions
453 already matched, then we change it to a pop_failure_jump.
454 Followed by two-byte address. */
457 /* Jump to following two-byte address, and push a dummy failure
458 point. This failure point will be thrown away if an attempt
459 is made to use it for a failure. A `+' construct makes this
460 before the first repeat. Also used as an intermediary kind
461 of jump when compiling an alternative. */
464 /* Push a dummy failure point and continue. Used at the end of
468 /* Followed by two-byte relative address and two-byte number n.
469 After matching N times, jump to the address upon failure. */
472 /* Followed by two-byte relative address, and two-byte number n.
473 Jump to the address N times, then fail. */
476 /* Set the following two-byte relative address to the
477 subsequent two-byte number. The address *includes* the two
481 wordchar
, /* Matches any word-constituent character. */
482 notwordchar
, /* Matches any char that is not a word-constituent. */
484 wordbeg
, /* Succeeds if at word beginning. */
485 wordend
, /* Succeeds if at word end. */
487 wordbound
, /* Succeeds if at a word boundary. */
488 notwordbound
/* Succeeds if not at a word boundary. */
491 ,before_dot
, /* Succeeds if before point. */
492 at_dot
, /* Succeeds if at point. */
493 after_dot
, /* Succeeds if after point. */
495 /* Matches any character whose syntax is specified. Followed by
496 a byte which contains a syntax code, e.g., Sword. */
499 /* Matches any character whose syntax is not that specified. */
504 /* Common operations on the compiled pattern. */
506 /* Store NUMBER in two contiguous bytes starting at DESTINATION. */
508 #define STORE_NUMBER(destination, number) \
510 (destination)[0] = (number) & 0377; \
511 (destination)[1] = (number) >> 8; \
514 /* Same as STORE_NUMBER, except increment DESTINATION to
515 the byte after where the number is stored. Therefore, DESTINATION
516 must be an lvalue. */
518 #define STORE_NUMBER_AND_INCR(destination, number) \
520 STORE_NUMBER (destination, number); \
521 (destination) += 2; \
524 /* Put into DESTINATION a number stored in two contiguous bytes starting
527 #define EXTRACT_NUMBER(destination, source) \
529 (destination) = *(source) & 0377; \
530 (destination) += SIGN_EXTEND_CHAR (*((source) + 1)) << 8; \
534 static void extract_number
_RE_ARGS ((int *dest
, unsigned char *source
));
536 extract_number (dest
, source
)
538 unsigned char *source
;
540 int temp
= SIGN_EXTEND_CHAR (*(source
+ 1));
541 *dest
= *source
& 0377;
545 # ifndef EXTRACT_MACROS /* To debug the macros. */
546 # undef EXTRACT_NUMBER
547 # define EXTRACT_NUMBER(dest, src) extract_number (&dest, src)
548 # endif /* not EXTRACT_MACROS */
552 /* Same as EXTRACT_NUMBER, except increment SOURCE to after the number.
553 SOURCE must be an lvalue. */
555 #define EXTRACT_NUMBER_AND_INCR(destination, source) \
557 EXTRACT_NUMBER (destination, source); \
562 static void extract_number_and_incr
_RE_ARGS ((int *destination
,
563 unsigned char **source
));
565 extract_number_and_incr (destination
, source
)
567 unsigned char **source
;
569 extract_number (destination
, *source
);
573 # ifndef EXTRACT_MACROS
574 # undef EXTRACT_NUMBER_AND_INCR
575 # define EXTRACT_NUMBER_AND_INCR(dest, src) \
576 extract_number_and_incr (&dest, &src)
577 # endif /* not EXTRACT_MACROS */
581 /* If DEBUG is defined, Regex prints many voluminous messages about what
582 it is doing (if the variable `debug' is nonzero). If linked with the
583 main program in `iregex.c', you can enter patterns and strings
584 interactively. And if linked with the main program in `main.c' and
585 the other test files, you can run the already-written tests. */
589 /* We use standard I/O for debugging. */
592 /* It is useful to test things that ``must'' be true when debugging. */
595 static int debug
= 0;
597 # define DEBUG_STATEMENT(e) e
598 # define DEBUG_PRINT1(x) if (debug) printf (x)
599 # define DEBUG_PRINT2(x1, x2) if (debug) printf (x1, x2)
600 # define DEBUG_PRINT3(x1, x2, x3) if (debug) printf (x1, x2, x3)
601 # define DEBUG_PRINT4(x1, x2, x3, x4) if (debug) printf (x1, x2, x3, x4)
602 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) \
603 if (debug) print_partial_compiled_pattern (s, e)
604 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) \
605 if (debug) print_double_string (w, s1, sz1, s2, sz2)
608 /* Print the fastmap in human-readable form. */
611 print_fastmap (fastmap
)
614 unsigned was_a_range
= 0;
617 while (i
< (1 << BYTEWIDTH
))
623 while (i
< (1 << BYTEWIDTH
) && fastmap
[i
])
639 /* Print a compiled pattern string in human-readable form, starting at
640 the START pointer into it and ending just before the pointer END. */
643 print_partial_compiled_pattern (start
, end
)
644 unsigned char *start
;
649 unsigned char *p
= start
;
650 unsigned char *pend
= end
;
658 /* Loop over pattern commands. */
661 printf ("%d:\t", p
- start
);
663 switch ((re_opcode_t
) *p
++)
671 printf ("/exactn/%d", mcnt
);
682 printf ("/start_memory/%d/%d", mcnt
, *p
++);
687 printf ("/stop_memory/%d/%d", mcnt
, *p
++);
691 printf ("/duplicate/%d", *p
++);
701 register int c
, last
= -100;
702 register int in_range
= 0;
704 printf ("/charset [%s",
705 (re_opcode_t
) *(p
- 1) == charset_not
? "^" : "");
707 assert (p
+ *p
< pend
);
709 for (c
= 0; c
< 256; c
++)
711 && (p
[1 + (c
/8)] & (1 << (c
% 8))))
713 /* Are we starting a range? */
714 if (last
+ 1 == c
&& ! in_range
)
719 /* Have we broken a range? */
720 else if (last
+ 1 != c
&& in_range
)
749 case on_failure_jump
:
750 extract_number_and_incr (&mcnt
, &p
);
751 printf ("/on_failure_jump to %d", p
+ mcnt
- start
);
754 case on_failure_keep_string_jump
:
755 extract_number_and_incr (&mcnt
, &p
);
756 printf ("/on_failure_keep_string_jump to %d", p
+ mcnt
- start
);
759 case dummy_failure_jump
:
760 extract_number_and_incr (&mcnt
, &p
);
761 printf ("/dummy_failure_jump to %d", p
+ mcnt
- start
);
764 case push_dummy_failure
:
765 printf ("/push_dummy_failure");
769 extract_number_and_incr (&mcnt
, &p
);
770 printf ("/maybe_pop_jump to %d", p
+ mcnt
- start
);
773 case pop_failure_jump
:
774 extract_number_and_incr (&mcnt
, &p
);
775 printf ("/pop_failure_jump to %d", p
+ mcnt
- start
);
779 extract_number_and_incr (&mcnt
, &p
);
780 printf ("/jump_past_alt to %d", p
+ mcnt
- start
);
784 extract_number_and_incr (&mcnt
, &p
);
785 printf ("/jump to %d", p
+ mcnt
- start
);
789 extract_number_and_incr (&mcnt
, &p
);
791 extract_number_and_incr (&mcnt2
, &p
);
792 printf ("/succeed_n to %d, %d times", p1
- start
, mcnt2
);
796 extract_number_and_incr (&mcnt
, &p
);
798 extract_number_and_incr (&mcnt2
, &p
);
799 printf ("/jump_n to %d, %d times", p1
- start
, mcnt2
);
803 extract_number_and_incr (&mcnt
, &p
);
805 extract_number_and_incr (&mcnt2
, &p
);
806 printf ("/set_number_at location %d to %d", p1
- start
, mcnt2
);
810 printf ("/wordbound");
814 printf ("/notwordbound");
826 printf ("/before_dot");
834 printf ("/after_dot");
838 printf ("/syntaxspec");
840 printf ("/%d", mcnt
);
844 printf ("/notsyntaxspec");
846 printf ("/%d", mcnt
);
851 printf ("/wordchar");
855 printf ("/notwordchar");
867 printf ("?%d", *(p
-1));
873 printf ("%d:\tend of pattern.\n", p
- start
);
878 print_compiled_pattern (bufp
)
879 struct re_pattern_buffer
*bufp
;
881 unsigned char *buffer
= bufp
->buffer
;
883 print_partial_compiled_pattern (buffer
, buffer
+ bufp
->used
);
884 printf ("%ld bytes used/%ld bytes allocated.\n",
885 bufp
->used
, bufp
->allocated
);
887 if (bufp
->fastmap_accurate
&& bufp
->fastmap
)
889 printf ("fastmap: ");
890 print_fastmap (bufp
->fastmap
);
893 printf ("re_nsub: %d\t", bufp
->re_nsub
);
894 printf ("regs_alloc: %d\t", bufp
->regs_allocated
);
895 printf ("can_be_null: %d\t", bufp
->can_be_null
);
896 printf ("newline_anchor: %d\n", bufp
->newline_anchor
);
897 printf ("no_sub: %d\t", bufp
->no_sub
);
898 printf ("not_bol: %d\t", bufp
->not_bol
);
899 printf ("not_eol: %d\t", bufp
->not_eol
);
900 printf ("syntax: %lx\n", bufp
->syntax
);
901 /* Perhaps we should print the translate table? */
906 print_double_string (where
, string1
, size1
, string2
, size2
)
919 if (FIRST_STRING_P (where
))
921 for (this_char
= where
- string1
; this_char
< size1
; this_char
++)
922 putchar (string1
[this_char
]);
927 for (this_char
= where
- string2
; this_char
< size2
; this_char
++)
928 putchar (string2
[this_char
]);
939 #else /* not DEBUG */
944 # define DEBUG_STATEMENT(e)
945 # define DEBUG_PRINT1(x)
946 # define DEBUG_PRINT2(x1, x2)
947 # define DEBUG_PRINT3(x1, x2, x3)
948 # define DEBUG_PRINT4(x1, x2, x3, x4)
949 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e)
950 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2)
952 #endif /* not DEBUG */
954 /* Set by `re_set_syntax' to the current regexp syntax to recognize. Can
955 also be assigned to arbitrarily: each pattern buffer stores its own
956 syntax, so it can be changed between regex compilations. */
957 /* This has no initializer because initialized variables in Emacs
958 become read-only after dumping. */
959 reg_syntax_t re_syntax_options
;
962 /* Specify the precise syntax of regexps for compilation. This provides
963 for compatibility for various utilities which historically have
964 different, incompatible syntaxes.
966 The argument SYNTAX is a bit mask comprised of the various bits
967 defined in gnu-regex.h. We return the old syntax. */
970 re_set_syntax (syntax
)
973 reg_syntax_t ret
= re_syntax_options
;
975 re_syntax_options
= syntax
;
977 if (syntax
& RE_DEBUG
)
979 else if (debug
) /* was on but now is not */
985 weak_alias (__re_set_syntax
, re_set_syntax
)
988 /* This table gives an error message for each of the error codes listed
989 in gnu-regex.h. Obviously the order here has to be same as there.
990 POSIX doesn't require that we do anything for REG_NOERROR,
991 but why not be nice? */
993 static const char *re_error_msgid
[] =
995 gettext_noop ("Success"), /* REG_NOERROR */
996 gettext_noop ("No match"), /* REG_NOMATCH */
997 gettext_noop ("Invalid regular expression"), /* REG_BADPAT */
998 gettext_noop ("Invalid collation character"), /* REG_ECOLLATE */
999 gettext_noop ("Invalid character class name"), /* REG_ECTYPE */
1000 gettext_noop ("Trailing backslash"), /* REG_EESCAPE */
1001 gettext_noop ("Invalid back reference"), /* REG_ESUBREG */
1002 gettext_noop ("Unmatched [ or [^"), /* REG_EBRACK */
1003 gettext_noop ("Unmatched ( or \\("), /* REG_EPAREN */
1004 gettext_noop ("Unmatched \\{"), /* REG_EBRACE */
1005 gettext_noop ("Invalid content of \\{\\}"), /* REG_BADBR */
1006 gettext_noop ("Invalid range end"), /* REG_ERANGE */
1007 gettext_noop ("Memory exhausted"), /* REG_ESPACE */
1008 gettext_noop ("Invalid preceding regular expression"), /* REG_BADRPT */
1009 gettext_noop ("Premature end of regular expression"), /* REG_EEND */
1010 gettext_noop ("Regular expression too big"), /* REG_ESIZE */
1011 gettext_noop ("Unmatched ) or \\)"), /* REG_ERPAREN */
1014 /* Avoiding alloca during matching, to placate r_alloc. */
1016 /* Define MATCH_MAY_ALLOCATE unless we need to make sure that the
1017 searching and matching functions should not call alloca. On some
1018 systems, alloca is implemented in terms of malloc, and if we're
1019 using the relocating allocator routines, then malloc could cause a
1020 relocation, which might (if the strings being searched are in the
1021 ralloc heap) shift the data out from underneath the regexp
1024 Here's another reason to avoid allocation: Emacs
1025 processes input from X in a signal handler; processing X input may
1026 call malloc; if input arrives while a matching routine is calling
1027 malloc, then we're scrod. But Emacs can't just block input while
1028 calling matching routines; then we don't notice interrupts when
1029 they come in. So, Emacs blocks input around all regexp calls
1030 except the matching calls, which it leaves unprotected, in the
1031 faith that they will not malloc. */
1033 /* Normally, this is fine. */
1034 #define MATCH_MAY_ALLOCATE
1036 /* When using GNU C, we are not REALLY using the C alloca, no matter
1037 what config.h may say. So don't take precautions for it. */
1042 /* The match routines may not allocate if (1) they would do it with malloc
1043 and (2) it's not safe for them to use malloc.
1044 Note that if REL_ALLOC is defined, matching would not use malloc for the
1045 failure stack, but we would still use it for the register vectors;
1046 so REL_ALLOC should not affect this. */
1047 #if (defined C_ALLOCA || defined REGEX_MALLOC) && defined emacs
1048 # undef MATCH_MAY_ALLOCATE
1052 /* Failure stack declarations and macros; both re_compile_fastmap and
1053 re_match_2 use a failure stack. These have to be macros because of
1054 REGEX_ALLOCATE_STACK. */
1057 /* Number of failure points for which to initially allocate space
1058 when matching. If this number is exceeded, we allocate more
1059 space, so it is not a hard limit. */
1060 #ifndef INIT_FAILURE_ALLOC
1061 # define INIT_FAILURE_ALLOC 5
1064 /* Roughly the maximum number of failure points on the stack. Would be
1065 exactly that if always used MAX_FAILURE_ITEMS items each time we failed.
1066 This is a variable only so users of regex can assign to it; we never
1067 change it ourselves. */
1071 # if defined MATCH_MAY_ALLOCATE
1072 /* 4400 was enough to cause a crash on Alpha OSF/1,
1073 whose default stack limit is 2mb. */
1074 long int re_max_failures
= 4000;
1076 long int re_max_failures
= 2000;
1079 union fail_stack_elt
1081 unsigned char *pointer
;
1085 typedef union fail_stack_elt fail_stack_elt_t
;
1089 fail_stack_elt_t
*stack
;
1090 unsigned long int size
;
1091 unsigned long int avail
; /* Offset of next open position. */
1094 #else /* not INT_IS_16BIT */
1096 # if defined MATCH_MAY_ALLOCATE
1097 /* 4400 was enough to cause a crash on Alpha OSF/1,
1098 whose default stack limit is 2mb. */
1099 int re_max_failures
= 20000;
1101 int re_max_failures
= 2000;
1104 union fail_stack_elt
1106 unsigned char *pointer
;
1110 typedef union fail_stack_elt fail_stack_elt_t
;
1114 fail_stack_elt_t
*stack
;
1116 unsigned avail
; /* Offset of next open position. */
1119 #endif /* INT_IS_16BIT */
1121 #define FAIL_STACK_EMPTY() (fail_stack.avail == 0)
1122 #define FAIL_STACK_PTR_EMPTY() (fail_stack_ptr->avail == 0)
1123 #define FAIL_STACK_FULL() (fail_stack.avail == fail_stack.size)
1126 /* Define macros to initialize and free the failure stack.
1127 Do `return -2' if the alloc fails. */
1129 #ifdef MATCH_MAY_ALLOCATE
1130 # define INIT_FAIL_STACK() \
1132 fail_stack.stack = (fail_stack_elt_t *) \
1133 REGEX_ALLOCATE_STACK (INIT_FAILURE_ALLOC * sizeof (fail_stack_elt_t)); \
1135 if (fail_stack.stack == NULL) \
1138 fail_stack.size = INIT_FAILURE_ALLOC; \
1139 fail_stack.avail = 0; \
1142 # define RESET_FAIL_STACK() REGEX_FREE_STACK (fail_stack.stack)
1144 # define INIT_FAIL_STACK() \
1146 fail_stack.avail = 0; \
1149 # define RESET_FAIL_STACK()
1153 /* Double the size of FAIL_STACK, up to approximately `re_max_failures' items.
1155 Return 1 if succeeds, and 0 if either ran out of memory
1156 allocating space for it or it was already too large.
1158 REGEX_REALLOCATE_STACK requires `destination' be declared. */
1160 #define DOUBLE_FAIL_STACK(fail_stack) \
1161 ((fail_stack).size > (unsigned) (re_max_failures * MAX_FAILURE_ITEMS) \
1163 : ((fail_stack).stack = (fail_stack_elt_t *) \
1164 REGEX_REALLOCATE_STACK ((fail_stack).stack, \
1165 (fail_stack).size * sizeof (fail_stack_elt_t), \
1166 ((fail_stack).size << 1) * sizeof (fail_stack_elt_t)), \
1168 (fail_stack).stack == NULL \
1170 : ((fail_stack).size <<= 1, \
1174 /* Push pointer POINTER on FAIL_STACK.
1175 Return 1 if was able to do so and 0 if ran out of memory allocating
1177 #define PUSH_PATTERN_OP(POINTER, FAIL_STACK) \
1178 ((FAIL_STACK_FULL () \
1179 && !DOUBLE_FAIL_STACK (FAIL_STACK)) \
1181 : ((FAIL_STACK).stack[(FAIL_STACK).avail++].pointer = POINTER, \
1184 /* Push a pointer value onto the failure stack.
1185 Assumes the variable `fail_stack'. Probably should only
1186 be called from within `PUSH_FAILURE_POINT'. */
1187 #define PUSH_FAILURE_POINTER(item) \
1188 fail_stack.stack[fail_stack.avail++].pointer = (unsigned char *) (item)
1190 /* This pushes an integer-valued item onto the failure stack.
1191 Assumes the variable `fail_stack'. Probably should only
1192 be called from within `PUSH_FAILURE_POINT'. */
1193 #define PUSH_FAILURE_INT(item) \
1194 fail_stack.stack[fail_stack.avail++].integer = (item)
1196 /* Push a fail_stack_elt_t value onto the failure stack.
1197 Assumes the variable `fail_stack'. Probably should only
1198 be called from within `PUSH_FAILURE_POINT'. */
1199 #define PUSH_FAILURE_ELT(item) \
1200 fail_stack.stack[fail_stack.avail++] = (item)
1202 /* These three POP... operations complement the three PUSH... operations.
1203 All assume that `fail_stack' is nonempty. */
1204 #define POP_FAILURE_POINTER() fail_stack.stack[--fail_stack.avail].pointer
1205 #define POP_FAILURE_INT() fail_stack.stack[--fail_stack.avail].integer
1206 #define POP_FAILURE_ELT() fail_stack.stack[--fail_stack.avail]
1208 /* Used to omit pushing failure point id's when we're not debugging. */
1210 # define DEBUG_PUSH PUSH_FAILURE_INT
1211 # define DEBUG_POP(item_addr) *(item_addr) = POP_FAILURE_INT ()
1213 # define DEBUG_PUSH(item)
1214 # define DEBUG_POP(item_addr)
1218 /* Push the information about the state we will need
1219 if we ever fail back to it.
1221 Requires variables fail_stack, regstart, regend, reg_info, and
1222 num_regs_pushed be declared. DOUBLE_FAIL_STACK requires `destination'
1225 Does `return FAILURE_CODE' if runs out of memory. */
1227 #define PUSH_FAILURE_POINT(pattern_place, string_place, failure_code) \
1229 char *destination; \
1230 /* Must be int, so when we don't save any registers, the arithmetic \
1231 of 0 + -1 isn't done as unsigned. */ \
1232 /* Can't be int, since there is not a shred of a guarantee that int \
1233 is wide enough to hold a value of something to which pointer can \
1235 active_reg_t this_reg; \
1237 DEBUG_STATEMENT (failure_id++); \
1238 DEBUG_STATEMENT (nfailure_points_pushed++); \
1239 DEBUG_PRINT2 ("\nPUSH_FAILURE_POINT #%u:\n", failure_id); \
1240 DEBUG_PRINT2 (" Before push, next avail: %d\n", (fail_stack).avail);\
1241 DEBUG_PRINT2 (" size: %d\n", (fail_stack).size);\
1243 DEBUG_PRINT2 (" slots needed: %ld\n", NUM_FAILURE_ITEMS); \
1244 DEBUG_PRINT2 (" available: %d\n", REMAINING_AVAIL_SLOTS); \
1246 /* Ensure we have enough space allocated for what we will push. */ \
1247 while (REMAINING_AVAIL_SLOTS < NUM_FAILURE_ITEMS) \
1249 if (!DOUBLE_FAIL_STACK (fail_stack)) \
1250 return failure_code; \
1252 DEBUG_PRINT2 ("\n Doubled stack; size now: %d\n", \
1253 (fail_stack).size); \
1254 DEBUG_PRINT2 (" slots available: %d\n", REMAINING_AVAIL_SLOTS);\
1257 /* Push the info, starting with the registers. */ \
1258 DEBUG_PRINT1 ("\n"); \
1261 for (this_reg = lowest_active_reg; this_reg <= highest_active_reg; \
1264 DEBUG_PRINT2 (" Pushing reg: %lu\n", this_reg); \
1265 DEBUG_STATEMENT (num_regs_pushed++); \
1267 DEBUG_PRINT2 (" start: %p\n", regstart[this_reg]); \
1268 PUSH_FAILURE_POINTER (regstart[this_reg]); \
1270 DEBUG_PRINT2 (" end: %p\n", regend[this_reg]); \
1271 PUSH_FAILURE_POINTER (regend[this_reg]); \
1273 DEBUG_PRINT2 (" info: %p\n ", \
1274 reg_info[this_reg].word.pointer); \
1275 DEBUG_PRINT2 (" match_null=%d", \
1276 REG_MATCH_NULL_STRING_P (reg_info[this_reg])); \
1277 DEBUG_PRINT2 (" active=%d", IS_ACTIVE (reg_info[this_reg])); \
1278 DEBUG_PRINT2 (" matched_something=%d", \
1279 MATCHED_SOMETHING (reg_info[this_reg])); \
1280 DEBUG_PRINT2 (" ever_matched=%d", \
1281 EVER_MATCHED_SOMETHING (reg_info[this_reg])); \
1282 DEBUG_PRINT1 ("\n"); \
1283 PUSH_FAILURE_ELT (reg_info[this_reg].word); \
1286 DEBUG_PRINT2 (" Pushing low active reg: %ld\n", lowest_active_reg);\
1287 PUSH_FAILURE_INT (lowest_active_reg); \
1289 DEBUG_PRINT2 (" Pushing high active reg: %ld\n", highest_active_reg);\
1290 PUSH_FAILURE_INT (highest_active_reg); \
1292 DEBUG_PRINT2 (" Pushing pattern %p:\n", pattern_place); \
1293 DEBUG_PRINT_COMPILED_PATTERN (bufp, pattern_place, pend); \
1294 PUSH_FAILURE_POINTER (pattern_place); \
1296 DEBUG_PRINT2 (" Pushing string %p: `", string_place); \
1297 DEBUG_PRINT_DOUBLE_STRING (string_place, string1, size1, string2, \
1299 DEBUG_PRINT1 ("'\n"); \
1300 PUSH_FAILURE_POINTER (string_place); \
1302 DEBUG_PRINT2 (" Pushing failure id: %u\n", failure_id); \
1303 DEBUG_PUSH (failure_id); \
1306 /* This is the number of items that are pushed and popped on the stack
1307 for each register. */
1308 #define NUM_REG_ITEMS 3
1310 /* Individual items aside from the registers. */
1312 # define NUM_NONREG_ITEMS 5 /* Includes failure point id. */
1314 # define NUM_NONREG_ITEMS 4
1317 /* We push at most this many items on the stack. */
1318 /* We used to use (num_regs - 1), which is the number of registers
1319 this regexp will save; but that was changed to 5
1320 to avoid stack overflow for a regexp with lots of parens. */
1321 #define MAX_FAILURE_ITEMS (5 * NUM_REG_ITEMS + NUM_NONREG_ITEMS)
1323 /* We actually push this many items. */
1324 #define NUM_FAILURE_ITEMS \
1326 ? 0 : highest_active_reg - lowest_active_reg + 1) \
1330 /* How many items can still be added to the stack without overflowing it. */
1331 #define REMAINING_AVAIL_SLOTS ((fail_stack).size - (fail_stack).avail)
1334 /* Pops what PUSH_FAIL_STACK pushes.
1336 We restore into the parameters, all of which should be lvalues:
1337 STR -- the saved data position.
1338 PAT -- the saved pattern position.
1339 LOW_REG, HIGH_REG -- the highest and lowest active registers.
1340 REGSTART, REGEND -- arrays of string positions.
1341 REG_INFO -- array of information about each subexpression.
1343 Also assumes the variables `fail_stack' and (if debugging), `bufp',
1344 `pend', `string1', `size1', `string2', and `size2'. */
1346 #define POP_FAILURE_POINT(str, pat, low_reg, high_reg, regstart, regend, reg_info)\
1348 DEBUG_STATEMENT (unsigned failure_id;) \
1349 active_reg_t this_reg; \
1350 const unsigned char *string_temp; \
1352 assert (!FAIL_STACK_EMPTY ()); \
1354 /* Remove failure points and point to how many regs pushed. */ \
1355 DEBUG_PRINT1 ("POP_FAILURE_POINT:\n"); \
1356 DEBUG_PRINT2 (" Before pop, next avail: %d\n", fail_stack.avail); \
1357 DEBUG_PRINT2 (" size: %d\n", fail_stack.size); \
1359 assert (fail_stack.avail >= NUM_NONREG_ITEMS); \
1361 DEBUG_POP (&failure_id); \
1362 DEBUG_PRINT2 (" Popping failure id: %u\n", failure_id); \
1364 /* If the saved string location is NULL, it came from an \
1365 on_failure_keep_string_jump opcode, and we want to throw away the \
1366 saved NULL, thus retaining our current position in the string. */ \
1367 string_temp = POP_FAILURE_POINTER (); \
1368 if (string_temp != NULL) \
1369 str = (const char *) string_temp; \
1371 DEBUG_PRINT2 (" Popping string %p: `", str); \
1372 DEBUG_PRINT_DOUBLE_STRING (str, string1, size1, string2, size2); \
1373 DEBUG_PRINT1 ("'\n"); \
1375 pat = (unsigned char *) POP_FAILURE_POINTER (); \
1376 DEBUG_PRINT2 (" Popping pattern %p:\n", pat); \
1377 DEBUG_PRINT_COMPILED_PATTERN (bufp, pat, pend); \
1379 /* Restore register info. */ \
1380 high_reg = (active_reg_t) POP_FAILURE_INT (); \
1381 DEBUG_PRINT2 (" Popping high active reg: %ld\n", high_reg); \
1383 low_reg = (active_reg_t) POP_FAILURE_INT (); \
1384 DEBUG_PRINT2 (" Popping low active reg: %ld\n", low_reg); \
1387 for (this_reg = high_reg; this_reg >= low_reg; this_reg--) \
1389 DEBUG_PRINT2 (" Popping reg: %ld\n", this_reg); \
1391 reg_info[this_reg].word = POP_FAILURE_ELT (); \
1392 DEBUG_PRINT2 (" info: %p\n", \
1393 reg_info[this_reg].word.pointer); \
1395 regend[this_reg] = (const char *) POP_FAILURE_POINTER (); \
1396 DEBUG_PRINT2 (" end: %p\n", regend[this_reg]); \
1398 regstart[this_reg] = (const char *) POP_FAILURE_POINTER (); \
1399 DEBUG_PRINT2 (" start: %p\n", regstart[this_reg]); \
1403 for (this_reg = highest_active_reg; this_reg > high_reg; this_reg--) \
1405 reg_info[this_reg].word.integer = 0; \
1406 regend[this_reg] = 0; \
1407 regstart[this_reg] = 0; \
1409 highest_active_reg = high_reg; \
1412 set_regs_matched_done = 0; \
1413 DEBUG_STATEMENT (nfailure_points_popped++); \
1414 } /* POP_FAILURE_POINT */
1418 /* Structure for per-register (a.k.a. per-group) information.
1419 Other register information, such as the
1420 starting and ending positions (which are addresses), and the list of
1421 inner groups (which is a bits list) are maintained in separate
1424 We are making a (strictly speaking) nonportable assumption here: that
1425 the compiler will pack our bit fields into something that fits into
1426 the type of `word', i.e., is something that fits into one item on the
1430 /* Declarations and macros for re_match_2. */
1434 fail_stack_elt_t word
;
1437 /* This field is one if this group can match the empty string,
1438 zero if not. If not yet determined, `MATCH_NULL_UNSET_VALUE'. */
1439 #define MATCH_NULL_UNSET_VALUE 3
1440 unsigned match_null_string_p
: 2;
1441 unsigned is_active
: 1;
1442 unsigned matched_something
: 1;
1443 unsigned ever_matched_something
: 1;
1445 } register_info_type
;
1447 #define REG_MATCH_NULL_STRING_P(R) ((R).bits.match_null_string_p)
1448 #define IS_ACTIVE(R) ((R).bits.is_active)
1449 #define MATCHED_SOMETHING(R) ((R).bits.matched_something)
1450 #define EVER_MATCHED_SOMETHING(R) ((R).bits.ever_matched_something)
1453 /* Call this when have matched a real character; it sets `matched' flags
1454 for the subexpressions which we are currently inside. Also records
1455 that those subexprs have matched. */
1456 #define SET_REGS_MATCHED() \
1459 if (!set_regs_matched_done) \
1462 set_regs_matched_done = 1; \
1463 for (r = lowest_active_reg; r <= highest_active_reg; r++) \
1465 MATCHED_SOMETHING (reg_info[r]) \
1466 = EVER_MATCHED_SOMETHING (reg_info[r]) \
1473 /* Registers are set to a sentinel when they haven't yet matched. */
1474 static char reg_unset_dummy
;
1475 #define REG_UNSET_VALUE (®_unset_dummy)
1476 #define REG_UNSET(e) ((e) == REG_UNSET_VALUE)
1478 /* Subroutine declarations and macros for regex_compile. */
1480 static reg_errcode_t regex_compile
_RE_ARGS ((const char *pattern
, size_t size
,
1481 reg_syntax_t syntax
,
1482 struct re_pattern_buffer
*bufp
));
1483 static void store_op1
_RE_ARGS ((re_opcode_t op
, unsigned char *loc
, int arg
));
1484 static void store_op2
_RE_ARGS ((re_opcode_t op
, unsigned char *loc
,
1485 int arg1
, int arg2
));
1486 static void insert_op1
_RE_ARGS ((re_opcode_t op
, unsigned char *loc
,
1487 int arg
, unsigned char *end
));
1488 static void insert_op2
_RE_ARGS ((re_opcode_t op
, unsigned char *loc
,
1489 int arg1
, int arg2
, unsigned char *end
));
1490 static boolean at_begline_loc_p
_RE_ARGS ((const char *pattern
, const char *p
,
1491 reg_syntax_t syntax
));
1492 static boolean at_endline_loc_p
_RE_ARGS ((const char *p
, const char *pend
,
1493 reg_syntax_t syntax
));
1494 static reg_errcode_t compile_range
_RE_ARGS ((const char **p_ptr
,
1497 reg_syntax_t syntax
,
1500 /* Fetch the next character in the uncompiled pattern---translating it
1501 if necessary. Also cast from a signed character in the constant
1502 string passed to us by the user to an unsigned char that we can use
1503 as an array index (in, e.g., `translate'). */
1505 # define PATFETCH(c) \
1506 do {if (p == pend) return REG_EEND; \
1507 c = (unsigned char) *p++; \
1508 if (translate) c = (unsigned char) translate[c]; \
1512 /* Fetch the next character in the uncompiled pattern, with no
1514 #define PATFETCH_RAW(c) \
1515 do {if (p == pend) return REG_EEND; \
1516 c = (unsigned char) *p++; \
1519 /* Go backwards one character in the pattern. */
1520 #define PATUNFETCH p--
1523 /* If `translate' is non-null, return translate[D], else just D. We
1524 cast the subscript to translate because some data is declared as
1525 `char *', to avoid warnings when a string constant is passed. But
1526 when we use a character as a subscript we must make it unsigned. */
1528 # define TRANSLATE(d) \
1529 (translate ? (char) translate[(unsigned char) (d)] : (d))
1533 /* Macros for outputting the compiled pattern into `buffer'. */
1535 /* If the buffer isn't allocated when it comes in, use this. */
1536 #define INIT_BUF_SIZE 32
1538 /* Make sure we have at least N more bytes of space in buffer. */
1539 #define GET_BUFFER_SPACE(n) \
1540 while ((unsigned long) (b - bufp->buffer + (n)) > bufp->allocated) \
1543 /* Make sure we have one more byte of buffer space and then add C to it. */
1544 #define BUF_PUSH(c) \
1546 GET_BUFFER_SPACE (1); \
1547 *b++ = (unsigned char) (c); \
1551 /* Ensure we have two more bytes of buffer space and then append C1 and C2. */
1552 #define BUF_PUSH_2(c1, c2) \
1554 GET_BUFFER_SPACE (2); \
1555 *b++ = (unsigned char) (c1); \
1556 *b++ = (unsigned char) (c2); \
1560 /* As with BUF_PUSH_2, except for three bytes. */
1561 #define BUF_PUSH_3(c1, c2, c3) \
1563 GET_BUFFER_SPACE (3); \
1564 *b++ = (unsigned char) (c1); \
1565 *b++ = (unsigned char) (c2); \
1566 *b++ = (unsigned char) (c3); \
1570 /* Store a jump with opcode OP at LOC to location TO. We store a
1571 relative address offset by the three bytes the jump itself occupies. */
1572 #define STORE_JUMP(op, loc, to) \
1573 store_op1 (op, loc, (int) ((to) - (loc) - 3))
1575 /* Likewise, for a two-argument jump. */
1576 #define STORE_JUMP2(op, loc, to, arg) \
1577 store_op2 (op, loc, (int) ((to) - (loc) - 3), arg)
1579 /* Like `STORE_JUMP', but for inserting. Assume `b' is the buffer end. */
1580 #define INSERT_JUMP(op, loc, to) \
1581 insert_op1 (op, loc, (int) ((to) - (loc) - 3), b)
1583 /* Like `STORE_JUMP2', but for inserting. Assume `b' is the buffer end. */
1584 #define INSERT_JUMP2(op, loc, to, arg) \
1585 insert_op2 (op, loc, (int) ((to) - (loc) - 3), arg, b)
1588 /* This is not an arbitrary limit: the arguments which represent offsets
1589 into the pattern are two bytes long. So if 2^16 bytes turns out to
1590 be too small, many things would have to change. */
1591 /* Any other compiler which, like MSC, has allocation limit below 2^16
1592 bytes will have to use approach similar to what was done below for
1593 MSC and drop MAX_BUF_SIZE a bit. Otherwise you may end up
1594 reallocating to 0 bytes. Such thing is not going to work too well.
1595 You have been warned!! */
1596 #if defined _MSC_VER && !defined WIN32
1597 /* Microsoft C 16-bit versions limit malloc to approx 65512 bytes.
1598 The REALLOC define eliminates a flurry of conversion warnings,
1599 but is not required. */
1600 # define MAX_BUF_SIZE 65500L
1601 # define REALLOC(p,s) realloc ((p), (size_t) (s))
1603 # define MAX_BUF_SIZE (1L << 16)
1604 # define REALLOC(p,s) realloc ((p), (s))
1607 /* Extend the buffer by twice its current size via realloc and
1608 reset the pointers that pointed into the old block to point to the
1609 correct places in the new one. If extending the buffer results in it
1610 being larger than MAX_BUF_SIZE, then flag memory exhausted. */
1611 #define EXTEND_BUFFER() \
1613 unsigned char *old_buffer = bufp->buffer; \
1614 if (bufp->allocated == MAX_BUF_SIZE) \
1616 bufp->allocated <<= 1; \
1617 if (bufp->allocated > MAX_BUF_SIZE) \
1618 bufp->allocated = MAX_BUF_SIZE; \
1619 bufp->buffer = (unsigned char *) REALLOC (bufp->buffer, bufp->allocated);\
1620 if (bufp->buffer == NULL) \
1621 return REG_ESPACE; \
1622 /* If the buffer moved, move all the pointers into it. */ \
1623 if (old_buffer != bufp->buffer) \
1625 b = (b - old_buffer) + bufp->buffer; \
1626 begalt = (begalt - old_buffer) + bufp->buffer; \
1627 if (fixup_alt_jump) \
1628 fixup_alt_jump = (fixup_alt_jump - old_buffer) + bufp->buffer;\
1630 laststart = (laststart - old_buffer) + bufp->buffer; \
1631 if (pending_exact) \
1632 pending_exact = (pending_exact - old_buffer) + bufp->buffer; \
1637 /* Since we have one byte reserved for the register number argument to
1638 {start,stop}_memory, the maximum number of groups we can report
1639 things about is what fits in that byte. */
1640 #define MAX_REGNUM 255
1642 /* But patterns can have more than `MAX_REGNUM' registers. We just
1643 ignore the excess. */
1644 typedef unsigned regnum_t
;
1647 /* Macros for the compile stack. */
1649 /* Since offsets can go either forwards or backwards, this type needs to
1650 be able to hold values from -(MAX_BUF_SIZE - 1) to MAX_BUF_SIZE - 1. */
1651 /* int may be not enough when sizeof(int) == 2. */
1652 typedef long pattern_offset_t
;
1656 pattern_offset_t begalt_offset
;
1657 pattern_offset_t fixup_alt_jump
;
1658 pattern_offset_t inner_group_offset
;
1659 pattern_offset_t laststart_offset
;
1661 } compile_stack_elt_t
;
1666 compile_stack_elt_t
*stack
;
1668 unsigned avail
; /* Offset of next open position. */
1669 } compile_stack_type
;
1672 #define INIT_COMPILE_STACK_SIZE 32
1674 #define COMPILE_STACK_EMPTY (compile_stack.avail == 0)
1675 #define COMPILE_STACK_FULL (compile_stack.avail == compile_stack.size)
1677 /* The next available element. */
1678 #define COMPILE_STACK_TOP (compile_stack.stack[compile_stack.avail])
1681 /* Set the bit for character C in a list. */
1682 #define SET_LIST_BIT(c) \
1683 (b[((unsigned char) (c)) / BYTEWIDTH] \
1684 |= 1 << (((unsigned char) c) % BYTEWIDTH))
1687 /* Get the next unsigned number in the uncompiled pattern. */
1688 #define GET_UNSIGNED_NUMBER(num) \
1692 while (ISDIGIT (c)) \
1696 num = num * 10 + c - '0'; \
1704 #if defined _LIBC || (defined HAVE_WCTYPE_H && defined HAVE_WCHAR_H)
1705 /* The GNU C library provides support for user-defined character classes
1706 and the functions from ISO C amendement 1. */
1707 # ifdef CHARCLASS_NAME_MAX
1708 # define CHAR_CLASS_MAX_LENGTH CHARCLASS_NAME_MAX
1710 /* This shouldn't happen but some implementation might still have this
1711 problem. Use a reasonable default value. */
1712 # define CHAR_CLASS_MAX_LENGTH 256
1716 # define IS_CHAR_CLASS(string) __wctype (string)
1718 # define IS_CHAR_CLASS(string) wctype (string)
1721 # define CHAR_CLASS_MAX_LENGTH 6 /* Namely, `xdigit'. */
1723 # define IS_CHAR_CLASS(string) \
1724 (STREQ (string, "alpha") || STREQ (string, "upper") \
1725 || STREQ (string, "lower") || STREQ (string, "digit") \
1726 || STREQ (string, "alnum") || STREQ (string, "xdigit") \
1727 || STREQ (string, "space") || STREQ (string, "print") \
1728 || STREQ (string, "punct") || STREQ (string, "graph") \
1729 || STREQ (string, "cntrl") || STREQ (string, "blank"))
1732 #ifndef MATCH_MAY_ALLOCATE
1734 /* If we cannot allocate large objects within re_match_2_internal,
1735 we make the fail stack and register vectors global.
1736 The fail stack, we grow to the maximum size when a regexp
1738 The register vectors, we adjust in size each time we
1739 compile a regexp, according to the number of registers it needs. */
1741 static fail_stack_type fail_stack
;
1743 /* Size with which the following vectors are currently allocated.
1744 That is so we can make them bigger as needed,
1745 but never make them smaller. */
1746 static int regs_allocated_size
;
1748 static const char ** regstart
, ** regend
;
1749 static const char ** old_regstart
, ** old_regend
;
1750 static const char **best_regstart
, **best_regend
;
1751 static register_info_type
*reg_info
;
1752 static const char **reg_dummy
;
1753 static register_info_type
*reg_info_dummy
;
1755 /* Make the register vectors big enough for NUM_REGS registers,
1756 but don't make them smaller. */
1759 regex_grow_registers (num_regs
)
1762 if (num_regs
> regs_allocated_size
)
1764 RETALLOC_IF (regstart
, num_regs
, const char *);
1765 RETALLOC_IF (regend
, num_regs
, const char *);
1766 RETALLOC_IF (old_regstart
, num_regs
, const char *);
1767 RETALLOC_IF (old_regend
, num_regs
, const char *);
1768 RETALLOC_IF (best_regstart
, num_regs
, const char *);
1769 RETALLOC_IF (best_regend
, num_regs
, const char *);
1770 RETALLOC_IF (reg_info
, num_regs
, register_info_type
);
1771 RETALLOC_IF (reg_dummy
, num_regs
, const char *);
1772 RETALLOC_IF (reg_info_dummy
, num_regs
, register_info_type
);
1774 regs_allocated_size
= num_regs
;
1778 #endif /* not MATCH_MAY_ALLOCATE */
1780 static boolean group_in_compile_stack
_RE_ARGS ((compile_stack_type
1784 /* `regex_compile' compiles PATTERN (of length SIZE) according to SYNTAX.
1785 Returns one of error codes defined in `gnu-regex.h', or zero for success.
1787 Assumes the `allocated' (and perhaps `buffer') and `translate'
1788 fields are set in BUFP on entry.
1790 If it succeeds, results are put in BUFP (if it returns an error, the
1791 contents of BUFP are undefined):
1792 `buffer' is the compiled pattern;
1793 `syntax' is set to SYNTAX;
1794 `used' is set to the length of the compiled pattern;
1795 `fastmap_accurate' is zero;
1796 `re_nsub' is the number of subexpressions in PATTERN;
1797 `not_bol' and `not_eol' are zero;
1799 The `fastmap' and `newline_anchor' fields are neither
1800 examined nor set. */
1802 /* Return, freeing storage we allocated. */
1803 #define FREE_STACK_RETURN(value) \
1804 return (free (compile_stack.stack), value)
1806 static reg_errcode_t
1807 regex_compile (pattern
, size
, syntax
, bufp
)
1808 const char *pattern
;
1810 reg_syntax_t syntax
;
1811 struct re_pattern_buffer
*bufp
;
1813 /* We fetch characters from PATTERN here. Even though PATTERN is
1814 `char *' (i.e., signed), we declare these variables as unsigned, so
1815 they can be reliably used as array indices. */
1816 register unsigned char c
, c1
;
1818 /* A random temporary spot in PATTERN. */
1821 /* Points to the end of the buffer, where we should append. */
1822 register unsigned char *b
;
1824 /* Keeps track of unclosed groups. */
1825 compile_stack_type compile_stack
;
1827 /* Points to the current (ending) position in the pattern. */
1828 const char *p
= pattern
;
1829 const char *pend
= pattern
+ size
;
1831 /* How to translate the characters in the pattern. */
1832 RE_TRANSLATE_TYPE translate
= bufp
->translate
;
1834 /* Address of the count-byte of the most recently inserted `exactn'
1835 command. This makes it possible to tell if a new exact-match
1836 character can be added to that command or if the character requires
1837 a new `exactn' command. */
1838 unsigned char *pending_exact
= 0;
1840 /* Address of start of the most recently finished expression.
1841 This tells, e.g., postfix * where to find the start of its
1842 operand. Reset at the beginning of groups and alternatives. */
1843 unsigned char *laststart
= 0;
1845 /* Address of beginning of regexp, or inside of last group. */
1846 unsigned char *begalt
;
1848 /* Place in the uncompiled pattern (i.e., the {) to
1849 which to go back if the interval is invalid. */
1850 const char *beg_interval
;
1852 /* Address of the place where a forward jump should go to the end of
1853 the containing expression. Each alternative of an `or' -- except the
1854 last -- ends with a forward jump of this sort. */
1855 unsigned char *fixup_alt_jump
= 0;
1857 /* Counts open-groups as they are encountered. Remembered for the
1858 matching close-group on the compile stack, so the same register
1859 number is put in the stop_memory as the start_memory. */
1860 regnum_t regnum
= 0;
1863 DEBUG_PRINT1 ("\nCompiling pattern: ");
1866 unsigned debug_count
;
1868 for (debug_count
= 0; debug_count
< size
; debug_count
++)
1869 putchar (pattern
[debug_count
]);
1874 /* Initialize the compile stack. */
1875 compile_stack
.stack
= TALLOC (INIT_COMPILE_STACK_SIZE
, compile_stack_elt_t
);
1876 if (compile_stack
.stack
== NULL
)
1879 compile_stack
.size
= INIT_COMPILE_STACK_SIZE
;
1880 compile_stack
.avail
= 0;
1882 /* Initialize the pattern buffer. */
1883 bufp
->syntax
= syntax
;
1884 bufp
->fastmap_accurate
= 0;
1885 bufp
->not_bol
= bufp
->not_eol
= 0;
1887 /* Set `used' to zero, so that if we return an error, the pattern
1888 printer (for debugging) will think there's no pattern. We reset it
1892 /* Always count groups, whether or not bufp->no_sub is set. */
1895 #if !defined emacs && !defined SYNTAX_TABLE
1896 /* Initialize the syntax table. */
1897 init_syntax_once ();
1900 if (bufp
->allocated
== 0)
1903 { /* If zero allocated, but buffer is non-null, try to realloc
1904 enough space. This loses if buffer's address is bogus, but
1905 that is the user's responsibility. */
1906 RETALLOC (bufp
->buffer
, INIT_BUF_SIZE
, unsigned char);
1909 { /* Caller did not allocate a buffer. Do it for them. */
1910 bufp
->buffer
= TALLOC (INIT_BUF_SIZE
, unsigned char);
1912 if (!bufp
->buffer
) FREE_STACK_RETURN (REG_ESPACE
);
1914 bufp
->allocated
= INIT_BUF_SIZE
;
1917 begalt
= b
= bufp
->buffer
;
1919 /* Loop through the uncompiled pattern until we're at the end. */
1928 if ( /* If at start of pattern, it's an operator. */
1930 /* If context independent, it's an operator. */
1931 || syntax
& RE_CONTEXT_INDEP_ANCHORS
1932 /* Otherwise, depends on what's come before. */
1933 || at_begline_loc_p (pattern
, p
, syntax
))
1943 if ( /* If at end of pattern, it's an operator. */
1945 /* If context independent, it's an operator. */
1946 || syntax
& RE_CONTEXT_INDEP_ANCHORS
1947 /* Otherwise, depends on what's next. */
1948 || at_endline_loc_p (p
, pend
, syntax
))
1958 if ((syntax
& RE_BK_PLUS_QM
)
1959 || (syntax
& RE_LIMITED_OPS
))
1963 /* If there is no previous pattern... */
1966 if (syntax
& RE_CONTEXT_INVALID_OPS
)
1967 FREE_STACK_RETURN (REG_BADRPT
);
1968 else if (!(syntax
& RE_CONTEXT_INDEP_OPS
))
1973 /* Are we optimizing this jump? */
1974 boolean keep_string_p
= false;
1976 /* 1 means zero (many) matches is allowed. */
1977 char zero_times_ok
= 0, many_times_ok
= 0;
1979 /* If there is a sequence of repetition chars, collapse it
1980 down to just one (the right one). We can't combine
1981 interval operators with these because of, e.g., `a{2}*',
1982 which should only match an even number of `a's. */
1986 zero_times_ok
|= c
!= '+';
1987 many_times_ok
|= c
!= '?';
1995 || (!(syntax
& RE_BK_PLUS_QM
) && (c
== '+' || c
== '?')))
1998 else if (syntax
& RE_BK_PLUS_QM
&& c
== '\\')
2000 if (p
== pend
) FREE_STACK_RETURN (REG_EESCAPE
);
2003 if (!(c1
== '+' || c1
== '?'))
2018 /* If we get here, we found another repeat character. */
2021 /* Star, etc. applied to an empty pattern is equivalent
2022 to an empty pattern. */
2026 /* Now we know whether or not zero matches is allowed
2027 and also whether or not two or more matches is allowed. */
2029 { /* More than one repetition is allowed, so put in at the
2030 end a backward relative jump from `b' to before the next
2031 jump we're going to put in below (which jumps from
2032 laststart to after this jump).
2034 But if we are at the `*' in the exact sequence `.*\n',
2035 insert an unconditional jump backwards to the .,
2036 instead of the beginning of the loop. This way we only
2037 push a failure point once, instead of every time
2038 through the loop. */
2039 assert (p
- 1 > pattern
);
2041 /* Allocate the space for the jump. */
2042 GET_BUFFER_SPACE (3);
2044 /* We know we are not at the first character of the pattern,
2045 because laststart was nonzero. And we've already
2046 incremented `p', by the way, to be the character after
2047 the `*'. Do we have to do something analogous here
2048 for null bytes, because of RE_DOT_NOT_NULL? */
2049 if (TRANSLATE (*(p
- 2)) == TRANSLATE ('.')
2051 && p
< pend
&& TRANSLATE (*p
) == TRANSLATE ('\n')
2052 && !(syntax
& RE_DOT_NEWLINE
))
2053 { /* We have .*\n. */
2054 STORE_JUMP (jump
, b
, laststart
);
2055 keep_string_p
= true;
2058 /* Anything else. */
2059 STORE_JUMP (maybe_pop_jump
, b
, laststart
- 3);
2061 /* We've added more stuff to the buffer. */
2065 /* On failure, jump from laststart to b + 3, which will be the
2066 end of the buffer after this jump is inserted. */
2067 GET_BUFFER_SPACE (3);
2068 INSERT_JUMP (keep_string_p
? on_failure_keep_string_jump
2076 /* At least one repetition is required, so insert a
2077 `dummy_failure_jump' before the initial
2078 `on_failure_jump' instruction of the loop. This
2079 effects a skip over that instruction the first time
2080 we hit that loop. */
2081 GET_BUFFER_SPACE (3);
2082 INSERT_JUMP (dummy_failure_jump
, laststart
, laststart
+ 6);
2097 boolean had_char_class
= false;
2099 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2101 /* Ensure that we have enough space to push a charset: the
2102 opcode, the length count, and the bitset; 34 bytes in all. */
2103 GET_BUFFER_SPACE (34);
2107 /* We test `*p == '^' twice, instead of using an if
2108 statement, so we only need one BUF_PUSH. */
2109 BUF_PUSH (*p
== '^' ? charset_not
: charset
);
2113 /* Remember the first position in the bracket expression. */
2116 /* Push the number of bytes in the bitmap. */
2117 BUF_PUSH ((1 << BYTEWIDTH
) / BYTEWIDTH
);
2119 /* Clear the whole map. */
2120 bzero (b
, (1 << BYTEWIDTH
) / BYTEWIDTH
);
2122 /* charset_not matches newline according to a syntax bit. */
2123 if ((re_opcode_t
) b
[-2] == charset_not
2124 && (syntax
& RE_HAT_LISTS_NOT_NEWLINE
))
2125 SET_LIST_BIT ('\n');
2127 /* Read in characters and ranges, setting map bits. */
2130 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2134 /* \ might escape characters inside [...] and [^...]. */
2135 if ((syntax
& RE_BACKSLASH_ESCAPE_IN_LISTS
) && c
== '\\')
2137 if (p
== pend
) FREE_STACK_RETURN (REG_EESCAPE
);
2144 /* Could be the end of the bracket expression. If it's
2145 not (i.e., when the bracket expression is `[]' so
2146 far), the ']' character bit gets set way below. */
2147 if (c
== ']' && p
!= p1
+ 1)
2150 /* Look ahead to see if it's a range when the last thing
2151 was a character class. */
2152 if (had_char_class
&& c
== '-' && *p
!= ']')
2153 FREE_STACK_RETURN (REG_ERANGE
);
2155 /* Look ahead to see if it's a range when the last thing
2156 was a character: if this is a hyphen not at the
2157 beginning or the end of a list, then it's the range
2160 && !(p
- 2 >= pattern
&& p
[-2] == '[')
2161 && !(p
- 3 >= pattern
&& p
[-3] == '[' && p
[-2] == '^')
2165 = compile_range (&p
, pend
, translate
, syntax
, b
);
2166 if (ret
!= REG_NOERROR
) FREE_STACK_RETURN (ret
);
2169 else if (p
[0] == '-' && p
[1] != ']')
2170 { /* This handles ranges made up of characters only. */
2173 /* Move past the `-'. */
2176 ret
= compile_range (&p
, pend
, translate
, syntax
, b
);
2177 if (ret
!= REG_NOERROR
) FREE_STACK_RETURN (ret
);
2180 /* See if we're at the beginning of a possible character
2183 else if (syntax
& RE_CHAR_CLASSES
&& c
== '[' && *p
== ':')
2184 { /* Leave room for the null. */
2185 char str
[CHAR_CLASS_MAX_LENGTH
+ 1];
2190 /* If pattern is `[[:'. */
2191 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2196 if ((c
== ':' && *p
== ']') || p
== pend
2197 || c1
== CHAR_CLASS_MAX_LENGTH
)
2203 /* If isn't a word bracketed by `[:' and `:]':
2204 undo the ending character, the letters, and leave
2205 the leading `:' and `[' (but set bits for them). */
2206 if (c
== ':' && *p
== ']')
2208 /* CYGNUS LOCAL: Skip this code if we don't have btowc(). btowc() is */
2209 /* defined in the 1994 Amendment 1 to ISO C and may not be present on */
2210 /* systems where we have wchar.h and wctype.h. */
2211 #if defined _LIBC || (defined HAVE_WCTYPE_H && defined HAVE_WCHAR_H && defined HAVE_BTOWC)
2212 boolean is_lower
= STREQ (str
, "lower");
2213 boolean is_upper
= STREQ (str
, "upper");
2217 wt
= IS_CHAR_CLASS (str
);
2219 FREE_STACK_RETURN (REG_ECTYPE
);
2221 /* Throw away the ] at the end of the character
2225 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2227 for (ch
= 0; ch
< 1 << BYTEWIDTH
; ++ch
)
2230 if (__iswctype (__btowc (ch
), wt
))
2233 if (iswctype (btowc (ch
), wt
))
2237 if (translate
&& (is_upper
|| is_lower
)
2238 && (ISUPPER (ch
) || ISLOWER (ch
)))
2242 had_char_class
= true;
2245 boolean is_alnum
= STREQ (str
, "alnum");
2246 boolean is_alpha
= STREQ (str
, "alpha");
2247 boolean is_blank
= STREQ (str
, "blank");
2248 boolean is_cntrl
= STREQ (str
, "cntrl");
2249 boolean is_digit
= STREQ (str
, "digit");
2250 boolean is_graph
= STREQ (str
, "graph");
2251 boolean is_lower
= STREQ (str
, "lower");
2252 boolean is_print
= STREQ (str
, "print");
2253 boolean is_punct
= STREQ (str
, "punct");
2254 boolean is_space
= STREQ (str
, "space");
2255 boolean is_upper
= STREQ (str
, "upper");
2256 boolean is_xdigit
= STREQ (str
, "xdigit");
2258 if (!IS_CHAR_CLASS (str
))
2259 FREE_STACK_RETURN (REG_ECTYPE
);
2261 /* Throw away the ] at the end of the character
2265 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2267 for (ch
= 0; ch
< 1 << BYTEWIDTH
; ch
++)
2269 /* This was split into 3 if's to
2270 avoid an arbitrary limit in some compiler. */
2271 if ( (is_alnum
&& ISALNUM (ch
))
2272 || (is_alpha
&& ISALPHA (ch
))
2273 || (is_blank
&& ISBLANK (ch
))
2274 || (is_cntrl
&& ISCNTRL (ch
)))
2276 if ( (is_digit
&& ISDIGIT (ch
))
2277 || (is_graph
&& ISGRAPH (ch
))
2278 || (is_lower
&& ISLOWER (ch
))
2279 || (is_print
&& ISPRINT (ch
)))
2281 if ( (is_punct
&& ISPUNCT (ch
))
2282 || (is_space
&& ISSPACE (ch
))
2283 || (is_upper
&& ISUPPER (ch
))
2284 || (is_xdigit
&& ISXDIGIT (ch
)))
2286 if ( translate
&& (is_upper
|| is_lower
)
2287 && (ISUPPER (ch
) || ISLOWER (ch
)))
2290 had_char_class
= true;
2291 #endif /* libc || wctype.h */
2300 had_char_class
= false;
2305 had_char_class
= false;
2310 /* Discard any (non)matching list bytes that are all 0 at the
2311 end of the map. Decrease the map-length byte too. */
2312 while ((int) b
[-1] > 0 && b
[b
[-1] - 1] == 0)
2320 if (syntax
& RE_NO_BK_PARENS
)
2327 if (syntax
& RE_NO_BK_PARENS
)
2334 if (syntax
& RE_NEWLINE_ALT
)
2341 if (syntax
& RE_NO_BK_VBAR
)
2348 if (syntax
& RE_INTERVALS
&& syntax
& RE_NO_BK_BRACES
)
2349 goto handle_interval
;
2355 if (p
== pend
) FREE_STACK_RETURN (REG_EESCAPE
);
2357 /* Do not translate the character after the \, so that we can
2358 distinguish, e.g., \B from \b, even if we normally would
2359 translate, e.g., B to b. */
2365 if (syntax
& RE_NO_BK_PARENS
)
2366 goto normal_backslash
;
2372 if (COMPILE_STACK_FULL
)
2374 RETALLOC (compile_stack
.stack
, compile_stack
.size
<< 1,
2375 compile_stack_elt_t
);
2376 if (compile_stack
.stack
== NULL
) return REG_ESPACE
;
2378 compile_stack
.size
<<= 1;
2381 /* These are the values to restore when we hit end of this
2382 group. They are all relative offsets, so that if the
2383 whole pattern moves because of realloc, they will still
2385 COMPILE_STACK_TOP
.begalt_offset
= begalt
- bufp
->buffer
;
2386 COMPILE_STACK_TOP
.fixup_alt_jump
2387 = fixup_alt_jump
? fixup_alt_jump
- bufp
->buffer
+ 1 : 0;
2388 COMPILE_STACK_TOP
.laststart_offset
= b
- bufp
->buffer
;
2389 COMPILE_STACK_TOP
.regnum
= regnum
;
2391 /* We will eventually replace the 0 with the number of
2392 groups inner to this one. But do not push a
2393 start_memory for groups beyond the last one we can
2394 represent in the compiled pattern. */
2395 if (regnum
<= MAX_REGNUM
)
2397 COMPILE_STACK_TOP
.inner_group_offset
= b
- bufp
->buffer
+ 2;
2398 BUF_PUSH_3 (start_memory
, regnum
, 0);
2401 compile_stack
.avail
++;
2406 /* If we've reached MAX_REGNUM groups, then this open
2407 won't actually generate any code, so we'll have to
2408 clear pending_exact explicitly. */
2414 if (syntax
& RE_NO_BK_PARENS
) goto normal_backslash
;
2416 if (COMPILE_STACK_EMPTY
)
2418 if (syntax
& RE_UNMATCHED_RIGHT_PAREN_ORD
)
2419 goto normal_backslash
;
2421 FREE_STACK_RETURN (REG_ERPAREN
);
2426 { /* Push a dummy failure point at the end of the
2427 alternative for a possible future
2428 `pop_failure_jump' to pop. See comments at
2429 `push_dummy_failure' in `re_match_2'. */
2430 BUF_PUSH (push_dummy_failure
);
2432 /* We allocated space for this jump when we assigned
2433 to `fixup_alt_jump', in the `handle_alt' case below. */
2434 STORE_JUMP (jump_past_alt
, fixup_alt_jump
, b
- 1);
2437 /* See similar code for backslashed left paren above. */
2438 if (COMPILE_STACK_EMPTY
)
2440 if (syntax
& RE_UNMATCHED_RIGHT_PAREN_ORD
)
2443 FREE_STACK_RETURN (REG_ERPAREN
);
2446 /* Since we just checked for an empty stack above, this
2447 ``can't happen''. */
2448 assert (compile_stack
.avail
!= 0);
2450 /* We don't just want to restore into `regnum', because
2451 later groups should continue to be numbered higher,
2452 as in `(ab)c(de)' -- the second group is #2. */
2453 regnum_t this_group_regnum
;
2455 compile_stack
.avail
--;
2456 begalt
= bufp
->buffer
+ COMPILE_STACK_TOP
.begalt_offset
;
2458 = COMPILE_STACK_TOP
.fixup_alt_jump
2459 ? bufp
->buffer
+ COMPILE_STACK_TOP
.fixup_alt_jump
- 1
2461 laststart
= bufp
->buffer
+ COMPILE_STACK_TOP
.laststart_offset
;
2462 this_group_regnum
= COMPILE_STACK_TOP
.regnum
;
2463 /* If we've reached MAX_REGNUM groups, then this open
2464 won't actually generate any code, so we'll have to
2465 clear pending_exact explicitly. */
2468 /* We're at the end of the group, so now we know how many
2469 groups were inside this one. */
2470 if (this_group_regnum
<= MAX_REGNUM
)
2472 unsigned char *inner_group_loc
2473 = bufp
->buffer
+ COMPILE_STACK_TOP
.inner_group_offset
;
2475 *inner_group_loc
= regnum
- this_group_regnum
;
2476 BUF_PUSH_3 (stop_memory
, this_group_regnum
,
2477 regnum
- this_group_regnum
);
2483 case '|': /* `\|'. */
2484 if (syntax
& RE_LIMITED_OPS
|| syntax
& RE_NO_BK_VBAR
)
2485 goto normal_backslash
;
2487 if (syntax
& RE_LIMITED_OPS
)
2490 /* Insert before the previous alternative a jump which
2491 jumps to this alternative if the former fails. */
2492 GET_BUFFER_SPACE (3);
2493 INSERT_JUMP (on_failure_jump
, begalt
, b
+ 6);
2497 /* The alternative before this one has a jump after it
2498 which gets executed if it gets matched. Adjust that
2499 jump so it will jump to this alternative's analogous
2500 jump (put in below, which in turn will jump to the next
2501 (if any) alternative's such jump, etc.). The last such
2502 jump jumps to the correct final destination. A picture:
2508 If we are at `b', then fixup_alt_jump right now points to a
2509 three-byte space after `a'. We'll put in the jump, set
2510 fixup_alt_jump to right after `b', and leave behind three
2511 bytes which we'll fill in when we get to after `c'. */
2514 STORE_JUMP (jump_past_alt
, fixup_alt_jump
, b
);
2516 /* Mark and leave space for a jump after this alternative,
2517 to be filled in later either by next alternative or
2518 when know we're at the end of a series of alternatives. */
2520 GET_BUFFER_SPACE (3);
2529 /* If \{ is a literal. */
2530 if (!(syntax
& RE_INTERVALS
)
2531 /* If we're at `\{' and it's not the open-interval
2533 || ((syntax
& RE_INTERVALS
) && (syntax
& RE_NO_BK_BRACES
))
2534 || (p
- 2 == pattern
&& p
== pend
))
2535 goto normal_backslash
;
2539 /* If got here, then the syntax allows intervals. */
2541 /* At least (most) this many matches must be made. */
2542 int lower_bound
= -1, upper_bound
= -1;
2544 beg_interval
= p
- 1;
2548 if (syntax
& RE_NO_BK_BRACES
)
2549 goto unfetch_interval
;
2551 FREE_STACK_RETURN (REG_EBRACE
);
2554 GET_UNSIGNED_NUMBER (lower_bound
);
2558 GET_UNSIGNED_NUMBER (upper_bound
);
2559 if (upper_bound
< 0) upper_bound
= RE_DUP_MAX
;
2562 /* Interval such as `{1}' => match exactly once. */
2563 upper_bound
= lower_bound
;
2565 if (lower_bound
< 0 || upper_bound
> RE_DUP_MAX
2566 || lower_bound
> upper_bound
)
2568 if (syntax
& RE_NO_BK_BRACES
)
2569 goto unfetch_interval
;
2571 FREE_STACK_RETURN (REG_BADBR
);
2574 if (!(syntax
& RE_NO_BK_BRACES
))
2576 if (c
!= '\\') FREE_STACK_RETURN (REG_EBRACE
);
2583 if (syntax
& RE_NO_BK_BRACES
)
2584 goto unfetch_interval
;
2586 FREE_STACK_RETURN (REG_BADBR
);
2589 /* We just parsed a valid interval. */
2591 /* If it's invalid to have no preceding re. */
2594 if (syntax
& RE_CONTEXT_INVALID_OPS
)
2595 FREE_STACK_RETURN (REG_BADRPT
);
2596 else if (syntax
& RE_CONTEXT_INDEP_OPS
)
2599 goto unfetch_interval
;
2602 /* If the upper bound is zero, don't want to succeed at
2603 all; jump from `laststart' to `b + 3', which will be
2604 the end of the buffer after we insert the jump. */
2605 if (upper_bound
== 0)
2607 GET_BUFFER_SPACE (3);
2608 INSERT_JUMP (jump
, laststart
, b
+ 3);
2612 /* Otherwise, we have a nontrivial interval. When
2613 we're all done, the pattern will look like:
2614 set_number_at <jump count> <upper bound>
2615 set_number_at <succeed_n count> <lower bound>
2616 succeed_n <after jump addr> <succeed_n count>
2618 jump_n <succeed_n addr> <jump count>
2619 (The upper bound and `jump_n' are omitted if
2620 `upper_bound' is 1, though.) */
2622 { /* If the upper bound is > 1, we need to insert
2623 more at the end of the loop. */
2624 unsigned nbytes
= 10 + (upper_bound
> 1) * 10;
2626 GET_BUFFER_SPACE (nbytes
);
2628 /* Initialize lower bound of the `succeed_n', even
2629 though it will be set during matching by its
2630 attendant `set_number_at' (inserted next),
2631 because `re_compile_fastmap' needs to know.
2632 Jump to the `jump_n' we might insert below. */
2633 INSERT_JUMP2 (succeed_n
, laststart
,
2634 b
+ 5 + (upper_bound
> 1) * 5,
2638 /* Code to initialize the lower bound. Insert
2639 before the `succeed_n'. The `5' is the last two
2640 bytes of this `set_number_at', plus 3 bytes of
2641 the following `succeed_n'. */
2642 insert_op2 (set_number_at
, laststart
, 5, lower_bound
, b
);
2645 if (upper_bound
> 1)
2646 { /* More than one repetition is allowed, so
2647 append a backward jump to the `succeed_n'
2648 that starts this interval.
2650 When we've reached this during matching,
2651 we'll have matched the interval once, so
2652 jump back only `upper_bound - 1' times. */
2653 STORE_JUMP2 (jump_n
, b
, laststart
+ 5,
2657 /* The location we want to set is the second
2658 parameter of the `jump_n'; that is `b-2' as
2659 an absolute address. `laststart' will be
2660 the `set_number_at' we're about to insert;
2661 `laststart+3' the number to set, the source
2662 for the relative address. But we are
2663 inserting into the middle of the pattern --
2664 so everything is getting moved up by 5.
2665 Conclusion: (b - 2) - (laststart + 3) + 5,
2666 i.e., b - laststart.
2668 We insert this at the beginning of the loop
2669 so that if we fail during matching, we'll
2670 reinitialize the bounds. */
2671 insert_op2 (set_number_at
, laststart
, b
- laststart
,
2672 upper_bound
- 1, b
);
2677 beg_interval
= NULL
;
2682 /* If an invalid interval, match the characters as literals. */
2683 assert (beg_interval
);
2685 beg_interval
= NULL
;
2687 /* normal_char and normal_backslash need `c'. */
2690 if (!(syntax
& RE_NO_BK_BRACES
))
2692 if (p
> pattern
&& p
[-1] == '\\')
2693 goto normal_backslash
;
2698 /* There is no way to specify the before_dot and after_dot
2699 operators. rms says this is ok. --karl */
2707 BUF_PUSH_2 (syntaxspec
, syntax_spec_code
[c
]);
2713 BUF_PUSH_2 (notsyntaxspec
, syntax_spec_code
[c
]);
2719 if (syntax
& RE_NO_GNU_OPS
)
2722 BUF_PUSH (wordchar
);
2727 if (syntax
& RE_NO_GNU_OPS
)
2730 BUF_PUSH (notwordchar
);
2735 if (syntax
& RE_NO_GNU_OPS
)
2741 if (syntax
& RE_NO_GNU_OPS
)
2747 if (syntax
& RE_NO_GNU_OPS
)
2749 BUF_PUSH (wordbound
);
2753 if (syntax
& RE_NO_GNU_OPS
)
2755 BUF_PUSH (notwordbound
);
2759 if (syntax
& RE_NO_GNU_OPS
)
2765 if (syntax
& RE_NO_GNU_OPS
)
2770 case '1': case '2': case '3': case '4': case '5':
2771 case '6': case '7': case '8': case '9':
2772 if (syntax
& RE_NO_BK_REFS
)
2778 FREE_STACK_RETURN (REG_ESUBREG
);
2780 /* Can't back reference to a subexpression if inside of it. */
2781 if (group_in_compile_stack (compile_stack
, (regnum_t
) c1
))
2785 BUF_PUSH_2 (duplicate
, c1
);
2791 if (syntax
& RE_BK_PLUS_QM
)
2794 goto normal_backslash
;
2798 /* You might think it would be useful for \ to mean
2799 not to translate; but if we don't translate it
2800 it will never match anything. */
2808 /* Expects the character in `c'. */
2810 /* If no exactn currently being built. */
2813 /* If last exactn not at current position. */
2814 || pending_exact
+ *pending_exact
+ 1 != b
2816 /* We have only one byte following the exactn for the count. */
2817 || *pending_exact
== (1 << BYTEWIDTH
) - 1
2819 /* If followed by a repetition operator. */
2820 || *p
== '*' || *p
== '^'
2821 || ((syntax
& RE_BK_PLUS_QM
)
2822 ? *p
== '\\' && (p
[1] == '+' || p
[1] == '?')
2823 : (*p
== '+' || *p
== '?'))
2824 || ((syntax
& RE_INTERVALS
)
2825 && ((syntax
& RE_NO_BK_BRACES
)
2827 : (p
[0] == '\\' && p
[1] == '{'))))
2829 /* Start building a new exactn. */
2833 BUF_PUSH_2 (exactn
, 0);
2834 pending_exact
= b
- 1;
2841 } /* while p != pend */
2844 /* Through the pattern now. */
2847 STORE_JUMP (jump_past_alt
, fixup_alt_jump
, b
);
2849 if (!COMPILE_STACK_EMPTY
)
2850 FREE_STACK_RETURN (REG_EPAREN
);
2852 /* If we don't want backtracking, force success
2853 the first time we reach the end of the compiled pattern. */
2854 if (syntax
& RE_NO_POSIX_BACKTRACKING
)
2857 free (compile_stack
.stack
);
2859 /* We have succeeded; set the length of the buffer. */
2860 bufp
->used
= b
- bufp
->buffer
;
2865 DEBUG_PRINT1 ("\nCompiled pattern: \n");
2866 print_compiled_pattern (bufp
);
2870 #ifndef MATCH_MAY_ALLOCATE
2871 /* Initialize the failure stack to the largest possible stack. This
2872 isn't necessary unless we're trying to avoid calling alloca in
2873 the search and match routines. */
2875 int num_regs
= bufp
->re_nsub
+ 1;
2877 /* Since DOUBLE_FAIL_STACK refuses to double only if the current size
2878 is strictly greater than re_max_failures, the largest possible stack
2879 is 2 * re_max_failures failure points. */
2880 if (fail_stack
.size
< (2 * re_max_failures
* MAX_FAILURE_ITEMS
))
2882 fail_stack
.size
= (2 * re_max_failures
* MAX_FAILURE_ITEMS
);
2885 if (! fail_stack
.stack
)
2887 = (fail_stack_elt_t
*) xmalloc (fail_stack
.size
2888 * sizeof (fail_stack_elt_t
));
2891 = (fail_stack_elt_t
*) xrealloc (fail_stack
.stack
,
2893 * sizeof (fail_stack_elt_t
)));
2894 # else /* not emacs */
2895 if (! fail_stack
.stack
)
2897 = (fail_stack_elt_t
*) malloc (fail_stack
.size
2898 * sizeof (fail_stack_elt_t
));
2901 = (fail_stack_elt_t
*) realloc (fail_stack
.stack
,
2903 * sizeof (fail_stack_elt_t
)));
2904 # endif /* not emacs */
2907 regex_grow_registers (num_regs
);
2909 #endif /* not MATCH_MAY_ALLOCATE */
2912 } /* regex_compile */
2914 /* Subroutines for `regex_compile'. */
2916 /* Store OP at LOC followed by two-byte integer parameter ARG. */
2919 store_op1 (op
, loc
, arg
)
2924 *loc
= (unsigned char) op
;
2925 STORE_NUMBER (loc
+ 1, arg
);
2929 /* Like `store_op1', but for two two-byte parameters ARG1 and ARG2. */
2932 store_op2 (op
, loc
, arg1
, arg2
)
2937 *loc
= (unsigned char) op
;
2938 STORE_NUMBER (loc
+ 1, arg1
);
2939 STORE_NUMBER (loc
+ 3, arg2
);
2943 /* Copy the bytes from LOC to END to open up three bytes of space at LOC
2944 for OP followed by two-byte integer parameter ARG. */
2947 insert_op1 (op
, loc
, arg
, end
)
2953 register unsigned char *pfrom
= end
;
2954 register unsigned char *pto
= end
+ 3;
2956 while (pfrom
!= loc
)
2959 store_op1 (op
, loc
, arg
);
2963 /* Like `insert_op1', but for two two-byte parameters ARG1 and ARG2. */
2966 insert_op2 (op
, loc
, arg1
, arg2
, end
)
2972 register unsigned char *pfrom
= end
;
2973 register unsigned char *pto
= end
+ 5;
2975 while (pfrom
!= loc
)
2978 store_op2 (op
, loc
, arg1
, arg2
);
2982 /* P points to just after a ^ in PATTERN. Return true if that ^ comes
2983 after an alternative or a begin-subexpression. We assume there is at
2984 least one character before the ^. */
2987 at_begline_loc_p (pattern
, p
, syntax
)
2988 const char *pattern
, *p
;
2989 reg_syntax_t syntax
;
2991 const char *prev
= p
- 2;
2992 boolean prev_prev_backslash
= prev
> pattern
&& prev
[-1] == '\\';
2995 /* After a subexpression? */
2996 (*prev
== '(' && (syntax
& RE_NO_BK_PARENS
|| prev_prev_backslash
))
2997 /* After an alternative? */
2998 || (*prev
== '|' && (syntax
& RE_NO_BK_VBAR
|| prev_prev_backslash
));
3002 /* The dual of at_begline_loc_p. This one is for $. We assume there is
3003 at least one character after the $, i.e., `P < PEND'. */
3006 at_endline_loc_p (p
, pend
, syntax
)
3007 const char *p
, *pend
;
3008 reg_syntax_t syntax
;
3010 const char *next
= p
;
3011 boolean next_backslash
= *next
== '\\';
3012 const char *next_next
= p
+ 1 < pend
? p
+ 1 : 0;
3015 /* Before a subexpression? */
3016 (syntax
& RE_NO_BK_PARENS
? *next
== ')'
3017 : next_backslash
&& next_next
&& *next_next
== ')')
3018 /* Before an alternative? */
3019 || (syntax
& RE_NO_BK_VBAR
? *next
== '|'
3020 : next_backslash
&& next_next
&& *next_next
== '|');
3024 /* Returns true if REGNUM is in one of COMPILE_STACK's elements and
3025 false if it's not. */
3028 group_in_compile_stack (compile_stack
, regnum
)
3029 compile_stack_type compile_stack
;
3034 for (this_element
= compile_stack
.avail
- 1;
3037 if (compile_stack
.stack
[this_element
].regnum
== regnum
)
3044 /* Read the ending character of a range (in a bracket expression) from the
3045 uncompiled pattern *P_PTR (which ends at PEND). We assume the
3046 starting character is in `P[-2]'. (`P[-1]' is the character `-'.)
3047 Then we set the translation of all bits between the starting and
3048 ending characters (inclusive) in the compiled pattern B.
3050 Return an error code.
3052 We use these short variable names so we can use the same macros as
3053 `regex_compile' itself. */
3055 static reg_errcode_t
3056 compile_range (p_ptr
, pend
, translate
, syntax
, b
)
3057 const char **p_ptr
, *pend
;
3058 RE_TRANSLATE_TYPE translate
;
3059 reg_syntax_t syntax
;
3064 const char *p
= *p_ptr
;
3065 unsigned int range_start
, range_end
;
3070 /* Even though the pattern is a signed `char *', we need to fetch
3071 with unsigned char *'s; if the high bit of the pattern character
3072 is set, the range endpoints will be negative if we fetch using a
3075 We also want to fetch the endpoints without translating them; the
3076 appropriate translation is done in the bit-setting loop below. */
3077 /* The SVR4 compiler on the 3B2 had trouble with unsigned const char *. */
3078 range_start
= ((const unsigned char *) p
)[-2];
3079 range_end
= ((const unsigned char *) p
)[0];
3081 /* Have to increment the pointer into the pattern string, so the
3082 caller isn't still at the ending character. */
3085 /* If the start is after the end, the range is empty. */
3086 if (range_start
> range_end
)
3087 return syntax
& RE_NO_EMPTY_RANGES
? REG_ERANGE
: REG_NOERROR
;
3089 /* Here we see why `this_char' has to be larger than an `unsigned
3090 char' -- the range is inclusive, so if `range_end' == 0xff
3091 (assuming 8-bit characters), we would otherwise go into an infinite
3092 loop, since all characters <= 0xff. */
3093 for (this_char
= range_start
; this_char
<= range_end
; this_char
++)
3095 SET_LIST_BIT (TRANSLATE (this_char
));
3101 /* re_compile_fastmap computes a ``fastmap'' for the compiled pattern in
3102 BUFP. A fastmap records which of the (1 << BYTEWIDTH) possible
3103 characters can start a string that matches the pattern. This fastmap
3104 is used by re_search to skip quickly over impossible starting points.
3106 The caller must supply the address of a (1 << BYTEWIDTH)-byte data
3107 area as BUFP->fastmap.
3109 We set the `fastmap', `fastmap_accurate', and `can_be_null' fields in
3112 Returns 0 if we succeed, -2 if an internal error. */
3115 re_compile_fastmap (bufp
)
3116 struct re_pattern_buffer
*bufp
;
3119 #ifdef MATCH_MAY_ALLOCATE
3120 fail_stack_type fail_stack
;
3122 #ifndef REGEX_MALLOC
3126 register char *fastmap
= bufp
->fastmap
;
3127 unsigned char *pattern
= bufp
->buffer
;
3128 unsigned char *p
= pattern
;
3129 register unsigned char *pend
= pattern
+ bufp
->used
;
3132 /* This holds the pointer to the failure stack, when
3133 it is allocated relocatably. */
3134 fail_stack_elt_t
*failure_stack_ptr
;
3137 /* Assume that each path through the pattern can be null until
3138 proven otherwise. We set this false at the bottom of switch
3139 statement, to which we get only if a particular path doesn't
3140 match the empty string. */
3141 boolean path_can_be_null
= true;
3143 /* We aren't doing a `succeed_n' to begin with. */
3144 boolean succeed_n_p
= false;
3146 assert (fastmap
!= NULL
&& p
!= NULL
);
3149 bzero (fastmap
, 1 << BYTEWIDTH
); /* Assume nothing's valid. */
3150 bufp
->fastmap_accurate
= 1; /* It will be when we're done. */
3151 bufp
->can_be_null
= 0;
3155 if (p
== pend
|| *p
== succeed
)
3157 /* We have reached the (effective) end of pattern. */
3158 if (!FAIL_STACK_EMPTY ())
3160 bufp
->can_be_null
|= path_can_be_null
;
3162 /* Reset for next path. */
3163 path_can_be_null
= true;
3165 p
= fail_stack
.stack
[--fail_stack
.avail
].pointer
;
3173 /* We should never be about to go beyond the end of the pattern. */
3176 switch (SWITCH_ENUM_CAST ((re_opcode_t
) *p
++))
3179 /* I guess the idea here is to simply not bother with a fastmap
3180 if a backreference is used, since it's too hard to figure out
3181 the fastmap for the corresponding group. Setting
3182 `can_be_null' stops `re_search_2' from using the fastmap, so
3183 that is all we do. */
3185 bufp
->can_be_null
= 1;
3189 /* Following are the cases which match a character. These end
3198 for (j
= *p
++ * BYTEWIDTH
- 1; j
>= 0; j
--)
3199 if (p
[j
/ BYTEWIDTH
] & (1 << (j
% BYTEWIDTH
)))
3205 /* Chars beyond end of map must be allowed. */
3206 for (j
= *p
* BYTEWIDTH
; j
< (1 << BYTEWIDTH
); j
++)
3209 for (j
= *p
++ * BYTEWIDTH
- 1; j
>= 0; j
--)
3210 if (!(p
[j
/ BYTEWIDTH
] & (1 << (j
% BYTEWIDTH
))))
3216 for (j
= 0; j
< (1 << BYTEWIDTH
); j
++)
3217 if (SYNTAX (j
) == Sword
)
3223 for (j
= 0; j
< (1 << BYTEWIDTH
); j
++)
3224 if (SYNTAX (j
) != Sword
)
3231 int fastmap_newline
= fastmap
['\n'];
3233 /* `.' matches anything ... */
3234 for (j
= 0; j
< (1 << BYTEWIDTH
); j
++)
3237 /* ... except perhaps newline. */
3238 if (!(bufp
->syntax
& RE_DOT_NEWLINE
))
3239 fastmap
['\n'] = fastmap_newline
;
3241 /* Return if we have already set `can_be_null'; if we have,
3242 then the fastmap is irrelevant. Something's wrong here. */
3243 else if (bufp
->can_be_null
)
3246 /* Otherwise, have to check alternative paths. */
3253 for (j
= 0; j
< (1 << BYTEWIDTH
); j
++)
3254 if (SYNTAX (j
) == (enum syntaxcode
) k
)
3261 for (j
= 0; j
< (1 << BYTEWIDTH
); j
++)
3262 if (SYNTAX (j
) != (enum syntaxcode
) k
)
3267 /* All cases after this match the empty string. These end with
3287 case push_dummy_failure
:
3292 case pop_failure_jump
:
3293 case maybe_pop_jump
:
3296 case dummy_failure_jump
:
3297 EXTRACT_NUMBER_AND_INCR (j
, p
);
3302 /* Jump backward implies we just went through the body of a
3303 loop and matched nothing. Opcode jumped to should be
3304 `on_failure_jump' or `succeed_n'. Just treat it like an
3305 ordinary jump. For a * loop, it has pushed its failure
3306 point already; if so, discard that as redundant. */
3307 if ((re_opcode_t
) *p
!= on_failure_jump
3308 && (re_opcode_t
) *p
!= succeed_n
)
3312 EXTRACT_NUMBER_AND_INCR (j
, p
);
3315 /* If what's on the stack is where we are now, pop it. */
3316 if (!FAIL_STACK_EMPTY ()
3317 && fail_stack
.stack
[fail_stack
.avail
- 1].pointer
== p
)
3323 case on_failure_jump
:
3324 case on_failure_keep_string_jump
:
3325 handle_on_failure_jump
:
3326 EXTRACT_NUMBER_AND_INCR (j
, p
);
3328 /* For some patterns, e.g., `(a?)?', `p+j' here points to the
3329 end of the pattern. We don't want to push such a point,
3330 since when we restore it above, entering the switch will
3331 increment `p' past the end of the pattern. We don't need
3332 to push such a point since we obviously won't find any more
3333 fastmap entries beyond `pend'. Such a pattern can match
3334 the null string, though. */
3337 if (!PUSH_PATTERN_OP (p
+ j
, fail_stack
))
3339 RESET_FAIL_STACK ();
3344 bufp
->can_be_null
= 1;
3348 EXTRACT_NUMBER_AND_INCR (k
, p
); /* Skip the n. */
3349 succeed_n_p
= false;
3356 /* Get to the number of times to succeed. */
3359 /* Increment p past the n for when k != 0. */
3360 EXTRACT_NUMBER_AND_INCR (k
, p
);
3364 succeed_n_p
= true; /* Spaghetti code alert. */
3365 goto handle_on_failure_jump
;
3382 abort (); /* We have listed all the cases. */
3385 /* Getting here means we have found the possible starting
3386 characters for one path of the pattern -- and that the empty
3387 string does not match. We need not follow this path further.
3388 Instead, look at the next alternative (remembered on the
3389 stack), or quit if no more. The test at the top of the loop
3390 does these things. */
3391 path_can_be_null
= false;
3395 /* Set `can_be_null' for the last path (also the first path, if the
3396 pattern is empty). */
3397 bufp
->can_be_null
|= path_can_be_null
;
3400 RESET_FAIL_STACK ();
3402 } /* re_compile_fastmap */
3404 weak_alias (__re_compile_fastmap
, re_compile_fastmap
)
3407 /* Set REGS to hold NUM_REGS registers, storing them in STARTS and
3408 ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use
3409 this memory for recording register information. STARTS and ENDS
3410 must be allocated using the malloc library routine, and must each
3411 be at least NUM_REGS * sizeof (regoff_t) bytes long.
3413 If NUM_REGS == 0, then subsequent matches should allocate their own
3416 Unless this function is called, the first search or match using
3417 PATTERN_BUFFER will allocate its own register data, without
3418 freeing the old data. */
3421 re_set_registers (bufp
, regs
, num_regs
, starts
, ends
)
3422 struct re_pattern_buffer
*bufp
;
3423 struct re_registers
*regs
;
3425 regoff_t
*starts
, *ends
;
3429 bufp
->regs_allocated
= REGS_REALLOCATE
;
3430 regs
->num_regs
= num_regs
;
3431 regs
->start
= starts
;
3436 bufp
->regs_allocated
= REGS_UNALLOCATED
;
3438 regs
->start
= regs
->end
= (regoff_t
*) 0;
3442 weak_alias (__re_set_registers
, re_set_registers
)
3445 /* Searching routines. */
3447 /* Like re_search_2, below, but only one string is specified, and
3448 doesn't let you say where to stop matching. */
3451 re_search (bufp
, string
, size
, startpos
, range
, regs
)
3452 struct re_pattern_buffer
*bufp
;
3454 int size
, startpos
, range
;
3455 struct re_registers
*regs
;
3457 return re_search_2 (bufp
, NULL
, 0, string
, size
, startpos
, range
,
3461 weak_alias (__re_search
, re_search
)
3465 /* Using the compiled pattern in BUFP->buffer, first tries to match the
3466 virtual concatenation of STRING1 and STRING2, starting first at index
3467 STARTPOS, then at STARTPOS + 1, and so on.
3469 STRING1 and STRING2 have length SIZE1 and SIZE2, respectively.
3471 RANGE is how far to scan while trying to match. RANGE = 0 means try
3472 only at STARTPOS; in general, the last start tried is STARTPOS +
3475 In REGS, return the indices of the virtual concatenation of STRING1
3476 and STRING2 that matched the entire BUFP->buffer and its contained
3479 Do not consider matching one past the index STOP in the virtual
3480 concatenation of STRING1 and STRING2.
3482 We return either the position in the strings at which the match was
3483 found, -1 if no match, or -2 if error (such as failure
3487 re_search_2 (bufp
, string1
, size1
, string2
, size2
, startpos
, range
, regs
, stop
)
3488 struct re_pattern_buffer
*bufp
;
3489 const char *string1
, *string2
;
3493 struct re_registers
*regs
;
3497 register char *fastmap
= bufp
->fastmap
;
3498 register RE_TRANSLATE_TYPE translate
= bufp
->translate
;
3499 int total_size
= size1
+ size2
;
3500 int endpos
= startpos
+ range
;
3502 /* Check for out-of-range STARTPOS. */
3503 if (startpos
< 0 || startpos
> total_size
)
3506 /* Fix up RANGE if it might eventually take us outside
3507 the virtual concatenation of STRING1 and STRING2.
3508 Make sure we won't move STARTPOS below 0 or above TOTAL_SIZE. */
3510 range
= 0 - startpos
;
3511 else if (endpos
> total_size
)
3512 range
= total_size
- startpos
;
3514 /* If the search isn't to be a backwards one, don't waste time in a
3515 search for a pattern that must be anchored. */
3516 if (bufp
->used
> 0 && (re_opcode_t
) bufp
->buffer
[0] == begbuf
&& range
> 0)
3525 /* In a forward search for something that starts with \=.
3526 don't keep searching past point. */
3527 if (bufp
->used
> 0 && (re_opcode_t
) bufp
->buffer
[0] == at_dot
&& range
> 0)
3529 range
= PT
- startpos
;
3535 /* Update the fastmap now if not correct already. */
3536 if (fastmap
&& !bufp
->fastmap_accurate
)
3537 if (re_compile_fastmap (bufp
) == -2)
3540 /* Loop through the string, looking for a place to start matching. */
3543 /* If a fastmap is supplied, skip quickly over characters that
3544 cannot be the start of a match. If the pattern can match the
3545 null string, however, we don't need to skip characters; we want
3546 the first null string. */
3547 if (fastmap
&& startpos
< total_size
&& !bufp
->can_be_null
)
3549 if (range
> 0) /* Searching forwards. */
3551 register const char *d
;
3552 register int lim
= 0;
3555 if (startpos
< size1
&& startpos
+ range
>= size1
)
3556 lim
= range
- (size1
- startpos
);
3558 d
= (startpos
>= size1
? string2
- size1
: string1
) + startpos
;
3560 /* Written out as an if-else to avoid testing `translate'
3564 && !fastmap
[(unsigned char)
3565 translate
[(unsigned char) *d
++]])
3568 while (range
> lim
&& !fastmap
[(unsigned char) *d
++])
3571 startpos
+= irange
- range
;
3573 else /* Searching backwards. */
3575 register char c
= (size1
== 0 || startpos
>= size1
3576 ? string2
[startpos
- size1
]
3577 : string1
[startpos
]);
3579 if (!fastmap
[(unsigned char) TRANSLATE (c
)])
3584 /* If can't match the null string, and that's all we have left, fail. */
3585 if (range
>= 0 && startpos
== total_size
&& fastmap
3586 && !bufp
->can_be_null
)
3589 val
= re_match_2_internal (bufp
, string1
, size1
, string2
, size2
,
3590 startpos
, regs
, stop
);
3591 #ifndef REGEX_MALLOC
3620 weak_alias (__re_search_2
, re_search_2
)
3623 /* This converts PTR, a pointer into one of the search strings `string1'
3624 and `string2' into an offset from the beginning of that string. */
3625 #define POINTER_TO_OFFSET(ptr) \
3626 (FIRST_STRING_P (ptr) \
3627 ? ((regoff_t) ((ptr) - string1)) \
3628 : ((regoff_t) ((ptr) - string2 + size1)))
3630 /* Macros for dealing with the split strings in re_match_2. */
3632 #define MATCHING_IN_FIRST_STRING (dend == end_match_1)
3634 /* Call before fetching a character with *d. This switches over to
3635 string2 if necessary. */
3636 #define PREFETCH() \
3639 /* End of string2 => fail. */ \
3640 if (dend == end_match_2) \
3642 /* End of string1 => advance to string2. */ \
3644 dend = end_match_2; \
3648 /* Test if at very beginning or at very end of the virtual concatenation
3649 of `string1' and `string2'. If only one string, it's `string2'. */
3650 #define AT_STRINGS_BEG(d) ((d) == (size1 ? string1 : string2) || !size2)
3651 #define AT_STRINGS_END(d) ((d) == end2)
3654 /* Test if D points to a character which is word-constituent. We have
3655 two special cases to check for: if past the end of string1, look at
3656 the first character in string2; and if before the beginning of
3657 string2, look at the last character in string1. */
3658 #define WORDCHAR_P(d) \
3659 (SYNTAX ((d) == end1 ? *string2 \
3660 : (d) == string2 - 1 ? *(end1 - 1) : *(d)) \
3663 /* Disabled due to a compiler bug -- see comment at case wordbound */
3665 /* Test if the character before D and the one at D differ with respect
3666 to being word-constituent. */
3667 #define AT_WORD_BOUNDARY(d) \
3668 (AT_STRINGS_BEG (d) || AT_STRINGS_END (d) \
3669 || WORDCHAR_P (d - 1) != WORDCHAR_P (d))
3672 /* Free everything we malloc. */
3673 #ifdef MATCH_MAY_ALLOCATE
3674 # define FREE_VAR(var) if (var) REGEX_FREE (var); var = NULL
3675 # define FREE_VARIABLES() \
3677 REGEX_FREE_STACK (fail_stack.stack); \
3678 FREE_VAR (regstart); \
3679 FREE_VAR (regend); \
3680 FREE_VAR (old_regstart); \
3681 FREE_VAR (old_regend); \
3682 FREE_VAR (best_regstart); \
3683 FREE_VAR (best_regend); \
3684 FREE_VAR (reg_info); \
3685 FREE_VAR (reg_dummy); \
3686 FREE_VAR (reg_info_dummy); \
3689 # define FREE_VARIABLES() ((void)0) /* Do nothing! But inhibit gcc warning. */
3690 #endif /* not MATCH_MAY_ALLOCATE */
3692 /* These values must meet several constraints. They must not be valid
3693 register values; since we have a limit of 255 registers (because
3694 we use only one byte in the pattern for the register number), we can
3695 use numbers larger than 255. They must differ by 1, because of
3696 NUM_FAILURE_ITEMS above. And the value for the lowest register must
3697 be larger than the value for the highest register, so we do not try
3698 to actually save any registers when none are active. */
3699 #define NO_HIGHEST_ACTIVE_REG (1 << BYTEWIDTH)
3700 #define NO_LOWEST_ACTIVE_REG (NO_HIGHEST_ACTIVE_REG + 1)
3702 /* Matching routines. */
3704 #ifndef emacs /* Emacs never uses this. */
3705 /* re_match is like re_match_2 except it takes only a single string. */
3708 re_match (bufp
, string
, size
, pos
, regs
)
3709 struct re_pattern_buffer
*bufp
;
3712 struct re_registers
*regs
;
3714 int result
= re_match_2_internal (bufp
, NULL
, 0, string
, size
,
3716 # ifndef REGEX_MALLOC
3724 weak_alias (__re_match
, re_match
)
3726 #endif /* not emacs */
3728 static boolean group_match_null_string_p
_RE_ARGS ((unsigned char **p
,
3730 register_info_type
*reg_info
));
3731 static boolean alt_match_null_string_p
_RE_ARGS ((unsigned char *p
,
3733 register_info_type
*reg_info
));
3734 static boolean common_op_match_null_string_p
_RE_ARGS ((unsigned char **p
,
3736 register_info_type
*reg_info
));
3737 static int bcmp_translate
_RE_ARGS ((const char *s1
, const char *s2
,
3738 int len
, char *translate
));
3740 /* re_match_2 matches the compiled pattern in BUFP against the
3741 the (virtual) concatenation of STRING1 and STRING2 (of length SIZE1
3742 and SIZE2, respectively). We start matching at POS, and stop
3745 If REGS is non-null and the `no_sub' field of BUFP is nonzero, we
3746 store offsets for the substring each group matched in REGS. See the
3747 documentation for exactly how many groups we fill.
3749 We return -1 if no match, -2 if an internal error (such as the
3750 failure stack overflowing). Otherwise, we return the length of the
3751 matched substring. */
3754 re_match_2 (bufp
, string1
, size1
, string2
, size2
, pos
, regs
, stop
)
3755 struct re_pattern_buffer
*bufp
;
3756 const char *string1
, *string2
;
3759 struct re_registers
*regs
;
3762 int result
= re_match_2_internal (bufp
, string1
, size1
, string2
, size2
,
3764 #ifndef REGEX_MALLOC
3772 weak_alias (__re_match_2
, re_match_2
)
3775 /* This is a separate function so that we can force an alloca cleanup
3778 re_match_2_internal (bufp
, string1
, size1
, string2
, size2
, pos
, regs
, stop
)
3779 struct re_pattern_buffer
*bufp
;
3780 const char *string1
, *string2
;
3783 struct re_registers
*regs
;
3786 /* General temporaries. */
3790 /* Just past the end of the corresponding string. */
3791 const char *end1
, *end2
;
3793 /* Pointers into string1 and string2, just past the last characters in
3794 each to consider matching. */
3795 const char *end_match_1
, *end_match_2
;
3797 /* Where we are in the data, and the end of the current string. */
3798 const char *d
, *dend
;
3800 /* Where we are in the pattern, and the end of the pattern. */
3801 unsigned char *p
= bufp
->buffer
;
3802 register unsigned char *pend
= p
+ bufp
->used
;
3804 /* Mark the opcode just after a start_memory, so we can test for an
3805 empty subpattern when we get to the stop_memory. */
3806 unsigned char *just_past_start_mem
= 0;
3808 /* We use this to map every character in the string. */
3809 RE_TRANSLATE_TYPE translate
= bufp
->translate
;
3811 /* Failure point stack. Each place that can handle a failure further
3812 down the line pushes a failure point on this stack. It consists of
3813 restart, regend, and reg_info for all registers corresponding to
3814 the subexpressions we're currently inside, plus the number of such
3815 registers, and, finally, two char *'s. The first char * is where
3816 to resume scanning the pattern; the second one is where to resume
3817 scanning the strings. If the latter is zero, the failure point is
3818 a ``dummy''; if a failure happens and the failure point is a dummy,
3819 it gets discarded and the next next one is tried. */
3820 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
3821 fail_stack_type fail_stack
;
3824 static unsigned failure_id
= 0;
3825 unsigned nfailure_points_pushed
= 0, nfailure_points_popped
= 0;
3829 /* This holds the pointer to the failure stack, when
3830 it is allocated relocatably. */
3831 fail_stack_elt_t
*failure_stack_ptr
;
3834 /* We fill all the registers internally, independent of what we
3835 return, for use in backreferences. The number here includes
3836 an element for register zero. */
3837 size_t num_regs
= bufp
->re_nsub
+ 1;
3839 /* The currently active registers. */
3840 active_reg_t lowest_active_reg
= NO_LOWEST_ACTIVE_REG
;
3841 active_reg_t highest_active_reg
= NO_HIGHEST_ACTIVE_REG
;
3843 /* Information on the contents of registers. These are pointers into
3844 the input strings; they record just what was matched (on this
3845 attempt) by a subexpression part of the pattern, that is, the
3846 regnum-th regstart pointer points to where in the pattern we began
3847 matching and the regnum-th regend points to right after where we
3848 stopped matching the regnum-th subexpression. (The zeroth register
3849 keeps track of what the whole pattern matches.) */
3850 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
3851 const char **regstart
, **regend
;
3854 /* If a group that's operated upon by a repetition operator fails to
3855 match anything, then the register for its start will need to be
3856 restored because it will have been set to wherever in the string we
3857 are when we last see its open-group operator. Similarly for a
3859 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
3860 const char **old_regstart
, **old_regend
;
3863 /* The is_active field of reg_info helps us keep track of which (possibly
3864 nested) subexpressions we are currently in. The matched_something
3865 field of reg_info[reg_num] helps us tell whether or not we have
3866 matched any of the pattern so far this time through the reg_num-th
3867 subexpression. These two fields get reset each time through any
3868 loop their register is in. */
3869 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
3870 register_info_type
*reg_info
;
3873 /* The following record the register info as found in the above
3874 variables when we find a match better than any we've seen before.
3875 This happens as we backtrack through the failure points, which in
3876 turn happens only if we have not yet matched the entire string. */
3877 unsigned best_regs_set
= false;
3878 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
3879 const char **best_regstart
, **best_regend
;
3882 /* Logically, this is `best_regend[0]'. But we don't want to have to
3883 allocate space for that if we're not allocating space for anything
3884 else (see below). Also, we never need info about register 0 for
3885 any of the other register vectors, and it seems rather a kludge to
3886 treat `best_regend' differently than the rest. So we keep track of
3887 the end of the best match so far in a separate variable. We
3888 initialize this to NULL so that when we backtrack the first time
3889 and need to test it, it's not garbage. */
3890 const char *match_end
= NULL
;
3892 /* This helps SET_REGS_MATCHED avoid doing redundant work. */
3893 int set_regs_matched_done
= 0;
3895 /* Used when we pop values we don't care about. */
3896 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
3897 const char **reg_dummy
;
3898 register_info_type
*reg_info_dummy
;
3902 /* Counts the total number of registers pushed. */
3903 unsigned num_regs_pushed
= 0;
3906 DEBUG_PRINT1 ("\n\nEntering re_match_2.\n");
3910 #ifdef MATCH_MAY_ALLOCATE
3911 /* Do not bother to initialize all the register variables if there are
3912 no groups in the pattern, as it takes a fair amount of time. If
3913 there are groups, we include space for register 0 (the whole
3914 pattern), even though we never use it, since it simplifies the
3915 array indexing. We should fix this. */
3918 regstart
= REGEX_TALLOC (num_regs
, const char *);
3919 regend
= REGEX_TALLOC (num_regs
, const char *);
3920 old_regstart
= REGEX_TALLOC (num_regs
, const char *);
3921 old_regend
= REGEX_TALLOC (num_regs
, const char *);
3922 best_regstart
= REGEX_TALLOC (num_regs
, const char *);
3923 best_regend
= REGEX_TALLOC (num_regs
, const char *);
3924 reg_info
= REGEX_TALLOC (num_regs
, register_info_type
);
3925 reg_dummy
= REGEX_TALLOC (num_regs
, const char *);
3926 reg_info_dummy
= REGEX_TALLOC (num_regs
, register_info_type
);
3928 if (!(regstart
&& regend
&& old_regstart
&& old_regend
&& reg_info
3929 && best_regstart
&& best_regend
&& reg_dummy
&& reg_info_dummy
))
3937 /* We must initialize all our variables to NULL, so that
3938 `FREE_VARIABLES' doesn't try to free them. */
3939 regstart
= regend
= old_regstart
= old_regend
= best_regstart
3940 = best_regend
= reg_dummy
= NULL
;
3941 reg_info
= reg_info_dummy
= (register_info_type
*) NULL
;
3943 #endif /* MATCH_MAY_ALLOCATE */
3945 /* The starting position is bogus. */
3946 if (pos
< 0 || pos
> size1
+ size2
)
3952 /* Initialize subexpression text positions to -1 to mark ones that no
3953 start_memory/stop_memory has been seen for. Also initialize the
3954 register information struct. */
3955 for (mcnt
= 1; (unsigned) mcnt
< num_regs
; mcnt
++)
3957 regstart
[mcnt
] = regend
[mcnt
]
3958 = old_regstart
[mcnt
] = old_regend
[mcnt
] = REG_UNSET_VALUE
;
3960 REG_MATCH_NULL_STRING_P (reg_info
[mcnt
]) = MATCH_NULL_UNSET_VALUE
;
3961 IS_ACTIVE (reg_info
[mcnt
]) = 0;
3962 MATCHED_SOMETHING (reg_info
[mcnt
]) = 0;
3963 EVER_MATCHED_SOMETHING (reg_info
[mcnt
]) = 0;
3966 /* We move `string1' into `string2' if the latter's empty -- but not if
3967 `string1' is null. */
3968 if (size2
== 0 && string1
!= NULL
)
3975 end1
= string1
+ size1
;
3976 end2
= string2
+ size2
;
3978 /* Compute where to stop matching, within the two strings. */
3981 end_match_1
= string1
+ stop
;
3982 end_match_2
= string2
;
3987 end_match_2
= string2
+ stop
- size1
;
3990 /* `p' scans through the pattern as `d' scans through the data.
3991 `dend' is the end of the input string that `d' points within. `d'
3992 is advanced into the following input string whenever necessary, but
3993 this happens before fetching; therefore, at the beginning of the
3994 loop, `d' can be pointing at the end of a string, but it cannot
3996 if (size1
> 0 && pos
<= size1
)
4003 d
= string2
+ pos
- size1
;
4007 DEBUG_PRINT1 ("The compiled pattern is:\n");
4008 DEBUG_PRINT_COMPILED_PATTERN (bufp
, p
, pend
);
4009 DEBUG_PRINT1 ("The string to match is: `");
4010 DEBUG_PRINT_DOUBLE_STRING (d
, string1
, size1
, string2
, size2
);
4011 DEBUG_PRINT1 ("'\n");
4013 /* This loops over pattern commands. It exits by returning from the
4014 function if the match is complete, or it drops through if the match
4015 fails at this starting point in the input data. */
4019 DEBUG_PRINT2 ("\n%p: ", p
);
4021 DEBUG_PRINT2 ("\n0x%x: ", p
);
4025 { /* End of pattern means we might have succeeded. */
4026 DEBUG_PRINT1 ("end of pattern ... ");
4028 /* If we haven't matched the entire string, and we want the
4029 longest match, try backtracking. */
4030 if (d
!= end_match_2
)
4032 /* 1 if this match ends in the same string (string1 or string2)
4033 as the best previous match. */
4034 boolean same_str_p
= (FIRST_STRING_P (match_end
)
4035 == MATCHING_IN_FIRST_STRING
);
4036 /* 1 if this match is the best seen so far. */
4037 boolean best_match_p
;
4039 /* AIX compiler got confused when this was combined
4040 with the previous declaration. */
4042 best_match_p
= d
> match_end
;
4044 best_match_p
= !MATCHING_IN_FIRST_STRING
;
4046 DEBUG_PRINT1 ("backtracking.\n");
4048 if (!FAIL_STACK_EMPTY ())
4049 { /* More failure points to try. */
4051 /* If exceeds best match so far, save it. */
4052 if (!best_regs_set
|| best_match_p
)
4054 best_regs_set
= true;
4057 DEBUG_PRINT1 ("\nSAVING match as best so far.\n");
4059 for (mcnt
= 1; (unsigned) mcnt
< num_regs
; mcnt
++)
4061 best_regstart
[mcnt
] = regstart
[mcnt
];
4062 best_regend
[mcnt
] = regend
[mcnt
];
4068 /* If no failure points, don't restore garbage. And if
4069 last match is real best match, don't restore second
4071 else if (best_regs_set
&& !best_match_p
)
4074 /* Restore best match. It may happen that `dend ==
4075 end_match_1' while the restored d is in string2.
4076 For example, the pattern `x.*y.*z' against the
4077 strings `x-' and `y-z-', if the two strings are
4078 not consecutive in memory. */
4079 DEBUG_PRINT1 ("Restoring best registers.\n");
4082 dend
= ((d
>= string1
&& d
<= end1
)
4083 ? end_match_1
: end_match_2
);
4085 for (mcnt
= 1; (unsigned) mcnt
< num_regs
; mcnt
++)
4087 regstart
[mcnt
] = best_regstart
[mcnt
];
4088 regend
[mcnt
] = best_regend
[mcnt
];
4091 } /* d != end_match_2 */
4094 DEBUG_PRINT1 ("Accepting match.\n");
4096 /* If caller wants register contents data back, do it. */
4097 if (regs
&& !bufp
->no_sub
)
4099 /* Have the register data arrays been allocated? */
4100 if (bufp
->regs_allocated
== REGS_UNALLOCATED
)
4101 { /* No. So allocate them with malloc. We need one
4102 extra element beyond `num_regs' for the `-1' marker
4104 regs
->num_regs
= MAX (RE_NREGS
, num_regs
+ 1);
4105 regs
->start
= TALLOC (regs
->num_regs
, regoff_t
);
4106 regs
->end
= TALLOC (regs
->num_regs
, regoff_t
);
4107 if (regs
->start
== NULL
|| regs
->end
== NULL
)
4112 bufp
->regs_allocated
= REGS_REALLOCATE
;
4114 else if (bufp
->regs_allocated
== REGS_REALLOCATE
)
4115 { /* Yes. If we need more elements than were already
4116 allocated, reallocate them. If we need fewer, just
4118 if (regs
->num_regs
< num_regs
+ 1)
4120 regs
->num_regs
= num_regs
+ 1;
4121 RETALLOC (regs
->start
, regs
->num_regs
, regoff_t
);
4122 RETALLOC (regs
->end
, regs
->num_regs
, regoff_t
);
4123 if (regs
->start
== NULL
|| regs
->end
== NULL
)
4132 /* These braces fend off a "empty body in an else-statement"
4133 warning under GCC when assert expands to nothing. */
4134 assert (bufp
->regs_allocated
== REGS_FIXED
);
4137 /* Convert the pointer data in `regstart' and `regend' to
4138 indices. Register zero has to be set differently,
4139 since we haven't kept track of any info for it. */
4140 if (regs
->num_regs
> 0)
4142 regs
->start
[0] = pos
;
4143 regs
->end
[0] = (MATCHING_IN_FIRST_STRING
4144 ? ((regoff_t
) (d
- string1
))
4145 : ((regoff_t
) (d
- string2
+ size1
)));
4148 /* Go through the first `min (num_regs, regs->num_regs)'
4149 registers, since that is all we initialized. */
4150 for (mcnt
= 1; (unsigned) mcnt
< MIN (num_regs
, regs
->num_regs
);
4153 if (REG_UNSET (regstart
[mcnt
]) || REG_UNSET (regend
[mcnt
]))
4154 regs
->start
[mcnt
] = regs
->end
[mcnt
] = -1;
4158 = (regoff_t
) POINTER_TO_OFFSET (regstart
[mcnt
]);
4160 = (regoff_t
) POINTER_TO_OFFSET (regend
[mcnt
]);
4164 /* If the regs structure we return has more elements than
4165 were in the pattern, set the extra elements to -1. If
4166 we (re)allocated the registers, this is the case,
4167 because we always allocate enough to have at least one
4169 for (mcnt
= num_regs
; (unsigned) mcnt
< regs
->num_regs
; mcnt
++)
4170 regs
->start
[mcnt
] = regs
->end
[mcnt
] = -1;
4171 } /* regs && !bufp->no_sub */
4173 DEBUG_PRINT4 ("%u failure points pushed, %u popped (%u remain).\n",
4174 nfailure_points_pushed
, nfailure_points_popped
,
4175 nfailure_points_pushed
- nfailure_points_popped
);
4176 DEBUG_PRINT2 ("%u registers pushed.\n", num_regs_pushed
);
4178 mcnt
= d
- pos
- (MATCHING_IN_FIRST_STRING
4182 DEBUG_PRINT2 ("Returning %d from re_match_2.\n", mcnt
);
4188 /* Otherwise match next pattern command. */
4189 switch (SWITCH_ENUM_CAST ((re_opcode_t
) *p
++))
4191 /* Ignore these. Used to ignore the n of succeed_n's which
4192 currently have n == 0. */
4194 DEBUG_PRINT1 ("EXECUTING no_op.\n");
4198 DEBUG_PRINT1 ("EXECUTING succeed.\n");
4201 /* Match the next n pattern characters exactly. The following
4202 byte in the pattern defines n, and the n bytes after that
4203 are the characters to match. */
4206 DEBUG_PRINT2 ("EXECUTING exactn %d.\n", mcnt
);
4208 /* This is written out as an if-else so we don't waste time
4209 testing `translate' inside the loop. */
4215 if ((unsigned char) translate
[(unsigned char) *d
++]
4216 != (unsigned char) *p
++)
4226 if (*d
++ != (char) *p
++) goto fail
;
4230 SET_REGS_MATCHED ();
4234 /* Match any character except possibly a newline or a null. */
4236 DEBUG_PRINT1 ("EXECUTING anychar.\n");
4240 if ((!(bufp
->syntax
& RE_DOT_NEWLINE
) && TRANSLATE (*d
) == '\n')
4241 || (bufp
->syntax
& RE_DOT_NOT_NULL
&& TRANSLATE (*d
) == '\000'))
4244 SET_REGS_MATCHED ();
4245 DEBUG_PRINT2 (" Matched `%d'.\n", *d
);
4253 register unsigned char c
;
4254 boolean
not = (re_opcode_t
) *(p
- 1) == charset_not
;
4256 DEBUG_PRINT2 ("EXECUTING charset%s.\n", not ? "_not" : "");
4259 c
= TRANSLATE (*d
); /* The character to match. */
4261 /* Cast to `unsigned' instead of `unsigned char' in case the
4262 bit list is a full 32 bytes long. */
4263 if (c
< (unsigned) (*p
* BYTEWIDTH
)
4264 && p
[1 + c
/ BYTEWIDTH
] & (1 << (c
% BYTEWIDTH
)))
4269 if (!not) goto fail
;
4271 SET_REGS_MATCHED ();
4277 /* The beginning of a group is represented by start_memory.
4278 The arguments are the register number in the next byte, and the
4279 number of groups inner to this one in the next. The text
4280 matched within the group is recorded (in the internal
4281 registers data structure) under the register number. */
4283 DEBUG_PRINT3 ("EXECUTING start_memory %d (%d):\n", *p
, p
[1]);
4285 /* Find out if this group can match the empty string. */
4286 p1
= p
; /* To send to group_match_null_string_p. */
4288 if (REG_MATCH_NULL_STRING_P (reg_info
[*p
]) == MATCH_NULL_UNSET_VALUE
)
4289 REG_MATCH_NULL_STRING_P (reg_info
[*p
])
4290 = group_match_null_string_p (&p1
, pend
, reg_info
);
4292 /* Save the position in the string where we were the last time
4293 we were at this open-group operator in case the group is
4294 operated upon by a repetition operator, e.g., with `(a*)*b'
4295 against `ab'; then we want to ignore where we are now in
4296 the string in case this attempt to match fails. */
4297 old_regstart
[*p
] = REG_MATCH_NULL_STRING_P (reg_info
[*p
])
4298 ? REG_UNSET (regstart
[*p
]) ? d
: regstart
[*p
]
4300 DEBUG_PRINT2 (" old_regstart: %d\n",
4301 POINTER_TO_OFFSET (old_regstart
[*p
]));
4304 DEBUG_PRINT2 (" regstart: %d\n", POINTER_TO_OFFSET (regstart
[*p
]));
4306 IS_ACTIVE (reg_info
[*p
]) = 1;
4307 MATCHED_SOMETHING (reg_info
[*p
]) = 0;
4309 /* Clear this whenever we change the register activity status. */
4310 set_regs_matched_done
= 0;
4312 /* This is the new highest active register. */
4313 highest_active_reg
= *p
;
4315 /* If nothing was active before, this is the new lowest active
4317 if (lowest_active_reg
== NO_LOWEST_ACTIVE_REG
)
4318 lowest_active_reg
= *p
;
4320 /* Move past the register number and inner group count. */
4322 just_past_start_mem
= p
;
4327 /* The stop_memory opcode represents the end of a group. Its
4328 arguments are the same as start_memory's: the register
4329 number, and the number of inner groups. */
4331 DEBUG_PRINT3 ("EXECUTING stop_memory %d (%d):\n", *p
, p
[1]);
4333 /* We need to save the string position the last time we were at
4334 this close-group operator in case the group is operated
4335 upon by a repetition operator, e.g., with `((a*)*(b*)*)*'
4336 against `aba'; then we want to ignore where we are now in
4337 the string in case this attempt to match fails. */
4338 old_regend
[*p
] = REG_MATCH_NULL_STRING_P (reg_info
[*p
])
4339 ? REG_UNSET (regend
[*p
]) ? d
: regend
[*p
]
4341 DEBUG_PRINT2 (" old_regend: %d\n",
4342 POINTER_TO_OFFSET (old_regend
[*p
]));
4345 DEBUG_PRINT2 (" regend: %d\n", POINTER_TO_OFFSET (regend
[*p
]));
4347 /* This register isn't active anymore. */
4348 IS_ACTIVE (reg_info
[*p
]) = 0;
4350 /* Clear this whenever we change the register activity status. */
4351 set_regs_matched_done
= 0;
4353 /* If this was the only register active, nothing is active
4355 if (lowest_active_reg
== highest_active_reg
)
4357 lowest_active_reg
= NO_LOWEST_ACTIVE_REG
;
4358 highest_active_reg
= NO_HIGHEST_ACTIVE_REG
;
4361 { /* We must scan for the new highest active register, since
4362 it isn't necessarily one less than now: consider
4363 (a(b)c(d(e)f)g). When group 3 ends, after the f), the
4364 new highest active register is 1. */
4365 unsigned char r
= *p
- 1;
4366 while (r
> 0 && !IS_ACTIVE (reg_info
[r
]))
4369 /* If we end up at register zero, that means that we saved
4370 the registers as the result of an `on_failure_jump', not
4371 a `start_memory', and we jumped to past the innermost
4372 `stop_memory'. For example, in ((.)*) we save
4373 registers 1 and 2 as a result of the *, but when we pop
4374 back to the second ), we are at the stop_memory 1.
4375 Thus, nothing is active. */
4378 lowest_active_reg
= NO_LOWEST_ACTIVE_REG
;
4379 highest_active_reg
= NO_HIGHEST_ACTIVE_REG
;
4382 highest_active_reg
= r
;
4385 /* If just failed to match something this time around with a
4386 group that's operated on by a repetition operator, try to
4387 force exit from the ``loop'', and restore the register
4388 information for this group that we had before trying this
4390 if ((!MATCHED_SOMETHING (reg_info
[*p
])
4391 || just_past_start_mem
== p
- 1)
4394 boolean is_a_jump_n
= false;
4398 switch ((re_opcode_t
) *p1
++)
4402 case pop_failure_jump
:
4403 case maybe_pop_jump
:
4405 case dummy_failure_jump
:
4406 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
4416 /* If the next operation is a jump backwards in the pattern
4417 to an on_failure_jump right before the start_memory
4418 corresponding to this stop_memory, exit from the loop
4419 by forcing a failure after pushing on the stack the
4420 on_failure_jump's jump in the pattern, and d. */
4421 if (mcnt
< 0 && (re_opcode_t
) *p1
== on_failure_jump
4422 && (re_opcode_t
) p1
[3] == start_memory
&& p1
[4] == *p
)
4424 /* If this group ever matched anything, then restore
4425 what its registers were before trying this last
4426 failed match, e.g., with `(a*)*b' against `ab' for
4427 regstart[1], and, e.g., with `((a*)*(b*)*)*'
4428 against `aba' for regend[3].
4430 Also restore the registers for inner groups for,
4431 e.g., `((a*)(b*))*' against `aba' (register 3 would
4432 otherwise get trashed). */
4434 if (EVER_MATCHED_SOMETHING (reg_info
[*p
]))
4438 EVER_MATCHED_SOMETHING (reg_info
[*p
]) = 0;
4440 /* Restore this and inner groups' (if any) registers. */
4441 for (r
= *p
; r
< (unsigned) *p
+ (unsigned) *(p
+ 1);
4444 regstart
[r
] = old_regstart
[r
];
4446 /* xx why this test? */
4447 if (old_regend
[r
] >= regstart
[r
])
4448 regend
[r
] = old_regend
[r
];
4452 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
4453 PUSH_FAILURE_POINT (p1
+ mcnt
, d
, -2);
4459 /* Move past the register number and the inner group count. */
4464 /* \<digit> has been turned into a `duplicate' command which is
4465 followed by the numeric value of <digit> as the register number. */
4468 register const char *d2
, *dend2
;
4469 int regno
= *p
++; /* Get which register to match against. */
4470 DEBUG_PRINT2 ("EXECUTING duplicate %d.\n", regno
);
4472 /* Can't back reference a group which we've never matched. */
4473 if (REG_UNSET (regstart
[regno
]) || REG_UNSET (regend
[regno
]))
4476 /* Where in input to try to start matching. */
4477 d2
= regstart
[regno
];
4479 /* Where to stop matching; if both the place to start and
4480 the place to stop matching are in the same string, then
4481 set to the place to stop, otherwise, for now have to use
4482 the end of the first string. */
4484 dend2
= ((FIRST_STRING_P (regstart
[regno
])
4485 == FIRST_STRING_P (regend
[regno
]))
4486 ? regend
[regno
] : end_match_1
);
4489 /* If necessary, advance to next segment in register
4493 if (dend2
== end_match_2
) break;
4494 if (dend2
== regend
[regno
]) break;
4496 /* End of string1 => advance to string2. */
4498 dend2
= regend
[regno
];
4500 /* At end of register contents => success */
4501 if (d2
== dend2
) break;
4503 /* If necessary, advance to next segment in data. */
4506 /* How many characters left in this segment to match. */
4509 /* Want how many consecutive characters we can match in
4510 one shot, so, if necessary, adjust the count. */
4511 if (mcnt
> dend2
- d2
)
4514 /* Compare that many; failure if mismatch, else move
4517 ? bcmp_translate (d
, d2
, mcnt
, translate
)
4518 : memcmp (d
, d2
, mcnt
))
4520 d
+= mcnt
, d2
+= mcnt
;
4522 /* Do this because we've match some characters. */
4523 SET_REGS_MATCHED ();
4529 /* begline matches the empty string at the beginning of the string
4530 (unless `not_bol' is set in `bufp'), and, if
4531 `newline_anchor' is set, after newlines. */
4533 DEBUG_PRINT1 ("EXECUTING begline.\n");
4535 if (AT_STRINGS_BEG (d
))
4537 if (!bufp
->not_bol
) break;
4539 else if (d
[-1] == '\n' && bufp
->newline_anchor
)
4543 /* In all other cases, we fail. */
4547 /* endline is the dual of begline. */
4549 DEBUG_PRINT1 ("EXECUTING endline.\n");
4551 if (AT_STRINGS_END (d
))
4553 if (!bufp
->not_eol
) break;
4556 /* We have to ``prefetch'' the next character. */
4557 else if ((d
== end1
? *string2
: *d
) == '\n'
4558 && bufp
->newline_anchor
)
4565 /* Match at the very beginning of the data. */
4567 DEBUG_PRINT1 ("EXECUTING begbuf.\n");
4568 if (AT_STRINGS_BEG (d
))
4573 /* Match at the very end of the data. */
4575 DEBUG_PRINT1 ("EXECUTING endbuf.\n");
4576 if (AT_STRINGS_END (d
))
4581 /* on_failure_keep_string_jump is used to optimize `.*\n'. It
4582 pushes NULL as the value for the string on the stack. Then
4583 `pop_failure_point' will keep the current value for the
4584 string, instead of restoring it. To see why, consider
4585 matching `foo\nbar' against `.*\n'. The .* matches the foo;
4586 then the . fails against the \n. But the next thing we want
4587 to do is match the \n against the \n; if we restored the
4588 string value, we would be back at the foo.
4590 Because this is used only in specific cases, we don't need to
4591 check all the things that `on_failure_jump' does, to make
4592 sure the right things get saved on the stack. Hence we don't
4593 share its code. The only reason to push anything on the
4594 stack at all is that otherwise we would have to change
4595 `anychar's code to do something besides goto fail in this
4596 case; that seems worse than this. */
4597 case on_failure_keep_string_jump
:
4598 DEBUG_PRINT1 ("EXECUTING on_failure_keep_string_jump");
4600 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
4602 DEBUG_PRINT3 (" %d (to %p):\n", mcnt
, p
+ mcnt
);
4604 DEBUG_PRINT3 (" %d (to 0x%x):\n", mcnt
, p
+ mcnt
);
4607 PUSH_FAILURE_POINT (p
+ mcnt
, NULL
, -2);
4611 /* Uses of on_failure_jump:
4613 Each alternative starts with an on_failure_jump that points
4614 to the beginning of the next alternative. Each alternative
4615 except the last ends with a jump that in effect jumps past
4616 the rest of the alternatives. (They really jump to the
4617 ending jump of the following alternative, because tensioning
4618 these jumps is a hassle.)
4620 Repeats start with an on_failure_jump that points past both
4621 the repetition text and either the following jump or
4622 pop_failure_jump back to this on_failure_jump. */
4623 case on_failure_jump
:
4625 DEBUG_PRINT1 ("EXECUTING on_failure_jump");
4627 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
4629 DEBUG_PRINT3 (" %d (to %p)", mcnt
, p
+ mcnt
);
4631 DEBUG_PRINT3 (" %d (to 0x%x)", mcnt
, p
+ mcnt
);
4634 /* If this on_failure_jump comes right before a group (i.e.,
4635 the original * applied to a group), save the information
4636 for that group and all inner ones, so that if we fail back
4637 to this point, the group's information will be correct.
4638 For example, in \(a*\)*\1, we need the preceding group,
4639 and in \(zz\(a*\)b*\)\2, we need the inner group. */
4641 /* We can't use `p' to check ahead because we push
4642 a failure point to `p + mcnt' after we do this. */
4645 /* We need to skip no_op's before we look for the
4646 start_memory in case this on_failure_jump is happening as
4647 the result of a completed succeed_n, as in \(a\)\{1,3\}b\1
4649 while (p1
< pend
&& (re_opcode_t
) *p1
== no_op
)
4652 if (p1
< pend
&& (re_opcode_t
) *p1
== start_memory
)
4654 /* We have a new highest active register now. This will
4655 get reset at the start_memory we are about to get to,
4656 but we will have saved all the registers relevant to
4657 this repetition op, as described above. */
4658 highest_active_reg
= *(p1
+ 1) + *(p1
+ 2);
4659 if (lowest_active_reg
== NO_LOWEST_ACTIVE_REG
)
4660 lowest_active_reg
= *(p1
+ 1);
4663 DEBUG_PRINT1 (":\n");
4664 PUSH_FAILURE_POINT (p
+ mcnt
, d
, -2);
4668 /* A smart repeat ends with `maybe_pop_jump'.
4669 We change it to either `pop_failure_jump' or `jump'. */
4670 case maybe_pop_jump
:
4671 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
4672 DEBUG_PRINT2 ("EXECUTING maybe_pop_jump %d.\n", mcnt
);
4674 register unsigned char *p2
= p
;
4676 /* Compare the beginning of the repeat with what in the
4677 pattern follows its end. If we can establish that there
4678 is nothing that they would both match, i.e., that we
4679 would have to backtrack because of (as in, e.g., `a*a')
4680 then we can change to pop_failure_jump, because we'll
4681 never have to backtrack.
4683 This is not true in the case of alternatives: in
4684 `(a|ab)*' we do need to backtrack to the `ab' alternative
4685 (e.g., if the string was `ab'). But instead of trying to
4686 detect that here, the alternative has put on a dummy
4687 failure point which is what we will end up popping. */
4689 /* Skip over open/close-group commands.
4690 If what follows this loop is a ...+ construct,
4691 look at what begins its body, since we will have to
4692 match at least one of that. */
4696 && ((re_opcode_t
) *p2
== stop_memory
4697 || (re_opcode_t
) *p2
== start_memory
))
4699 else if (p2
+ 6 < pend
4700 && (re_opcode_t
) *p2
== dummy_failure_jump
)
4707 /* p1[0] ... p1[2] are the `on_failure_jump' corresponding
4708 to the `maybe_finalize_jump' of this case. Examine what
4711 /* If we're at the end of the pattern, we can change. */
4714 /* Consider what happens when matching ":\(.*\)"
4715 against ":/". I don't really understand this code
4717 p
[-3] = (unsigned char) pop_failure_jump
;
4719 (" End of pattern: change to `pop_failure_jump'.\n");
4722 else if ((re_opcode_t
) *p2
== exactn
4723 || (bufp
->newline_anchor
&& (re_opcode_t
) *p2
== endline
))
4725 register unsigned char c
4726 = *p2
== (unsigned char) endline
? '\n' : p2
[2];
4728 if ((re_opcode_t
) p1
[3] == exactn
&& p1
[5] != c
)
4730 p
[-3] = (unsigned char) pop_failure_jump
;
4731 DEBUG_PRINT3 (" %c != %c => pop_failure_jump.\n",
4735 else if ((re_opcode_t
) p1
[3] == charset
4736 || (re_opcode_t
) p1
[3] == charset_not
)
4738 int not = (re_opcode_t
) p1
[3] == charset_not
;
4740 if (c
< (unsigned char) (p1
[4] * BYTEWIDTH
)
4741 && p1
[5 + c
/ BYTEWIDTH
] & (1 << (c
% BYTEWIDTH
)))
4744 /* `not' is equal to 1 if c would match, which means
4745 that we can't change to pop_failure_jump. */
4748 p
[-3] = (unsigned char) pop_failure_jump
;
4749 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
4753 else if ((re_opcode_t
) *p2
== charset
)
4756 register unsigned char c
4757 = *p2
== (unsigned char) endline
? '\n' : p2
[2];
4761 if ((re_opcode_t
) p1
[3] == exactn
4762 && ! ((int) p2
[1] * BYTEWIDTH
> (int) p1
[5]
4763 && (p2
[2 + p1
[5] / BYTEWIDTH
]
4764 & (1 << (p1
[5] % BYTEWIDTH
)))))
4766 if ((re_opcode_t
) p1
[3] == exactn
4767 && ! ((int) p2
[1] * BYTEWIDTH
> (int) p1
[4]
4768 && (p2
[2 + p1
[4] / BYTEWIDTH
]
4769 & (1 << (p1
[4] % BYTEWIDTH
)))))
4772 p
[-3] = (unsigned char) pop_failure_jump
;
4773 DEBUG_PRINT3 (" %c != %c => pop_failure_jump.\n",
4777 else if ((re_opcode_t
) p1
[3] == charset_not
)
4780 /* We win if the charset_not inside the loop
4781 lists every character listed in the charset after. */
4782 for (idx
= 0; idx
< (int) p2
[1]; idx
++)
4783 if (! (p2
[2 + idx
] == 0
4784 || (idx
< (int) p1
[4]
4785 && ((p2
[2 + idx
] & ~ p1
[5 + idx
]) == 0))))
4790 p
[-3] = (unsigned char) pop_failure_jump
;
4791 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
4794 else if ((re_opcode_t
) p1
[3] == charset
)
4797 /* We win if the charset inside the loop
4798 has no overlap with the one after the loop. */
4800 idx
< (int) p2
[1] && idx
< (int) p1
[4];
4802 if ((p2
[2 + idx
] & p1
[5 + idx
]) != 0)
4805 if (idx
== p2
[1] || idx
== p1
[4])
4807 p
[-3] = (unsigned char) pop_failure_jump
;
4808 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
4813 p
-= 2; /* Point at relative address again. */
4814 if ((re_opcode_t
) p
[-1] != pop_failure_jump
)
4816 p
[-1] = (unsigned char) jump
;
4817 DEBUG_PRINT1 (" Match => jump.\n");
4818 goto unconditional_jump
;
4820 /* Note fall through. */
4823 /* The end of a simple repeat has a pop_failure_jump back to
4824 its matching on_failure_jump, where the latter will push a
4825 failure point. The pop_failure_jump takes off failure
4826 points put on by this pop_failure_jump's matching
4827 on_failure_jump; we got through the pattern to here from the
4828 matching on_failure_jump, so didn't fail. */
4829 case pop_failure_jump
:
4831 /* We need to pass separate storage for the lowest and
4832 highest registers, even though we don't care about the
4833 actual values. Otherwise, we will restore only one
4834 register from the stack, since lowest will == highest in
4835 `pop_failure_point'. */
4836 active_reg_t dummy_low_reg
, dummy_high_reg
;
4837 unsigned char *pdummy
;
4840 DEBUG_PRINT1 ("EXECUTING pop_failure_jump.\n");
4841 POP_FAILURE_POINT (sdummy
, pdummy
,
4842 dummy_low_reg
, dummy_high_reg
,
4843 reg_dummy
, reg_dummy
, reg_info_dummy
);
4845 /* Note fall through. */
4849 DEBUG_PRINT2 ("\n%p: ", p
);
4851 DEBUG_PRINT2 ("\n0x%x: ", p
);
4853 /* Note fall through. */
4855 /* Unconditionally jump (without popping any failure points). */
4857 EXTRACT_NUMBER_AND_INCR (mcnt
, p
); /* Get the amount to jump. */
4858 DEBUG_PRINT2 ("EXECUTING jump %d ", mcnt
);
4859 p
+= mcnt
; /* Do the jump. */
4861 DEBUG_PRINT2 ("(to %p).\n", p
);
4863 DEBUG_PRINT2 ("(to 0x%x).\n", p
);
4868 /* We need this opcode so we can detect where alternatives end
4869 in `group_match_null_string_p' et al. */
4871 DEBUG_PRINT1 ("EXECUTING jump_past_alt.\n");
4872 goto unconditional_jump
;
4875 /* Normally, the on_failure_jump pushes a failure point, which
4876 then gets popped at pop_failure_jump. We will end up at
4877 pop_failure_jump, also, and with a pattern of, say, `a+', we
4878 are skipping over the on_failure_jump, so we have to push
4879 something meaningless for pop_failure_jump to pop. */
4880 case dummy_failure_jump
:
4881 DEBUG_PRINT1 ("EXECUTING dummy_failure_jump.\n");
4882 /* It doesn't matter what we push for the string here. What
4883 the code at `fail' tests is the value for the pattern. */
4884 PUSH_FAILURE_POINT (NULL
, NULL
, -2);
4885 goto unconditional_jump
;
4888 /* At the end of an alternative, we need to push a dummy failure
4889 point in case we are followed by a `pop_failure_jump', because
4890 we don't want the failure point for the alternative to be
4891 popped. For example, matching `(a|ab)*' against `aab'
4892 requires that we match the `ab' alternative. */
4893 case push_dummy_failure
:
4894 DEBUG_PRINT1 ("EXECUTING push_dummy_failure.\n");
4895 /* See comments just above at `dummy_failure_jump' about the
4897 PUSH_FAILURE_POINT (NULL
, NULL
, -2);
4900 /* Have to succeed matching what follows at least n times.
4901 After that, handle like `on_failure_jump'. */
4903 EXTRACT_NUMBER (mcnt
, p
+ 2);
4904 DEBUG_PRINT2 ("EXECUTING succeed_n %d.\n", mcnt
);
4907 /* Originally, this is how many times we HAVE to succeed. */
4912 STORE_NUMBER_AND_INCR (p
, mcnt
);
4914 DEBUG_PRINT3 (" Setting %p to %d.\n", p
- 2, mcnt
);
4916 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p
- 2, mcnt
);
4922 DEBUG_PRINT2 (" Setting two bytes from %p to no_op.\n", p
+2);
4924 DEBUG_PRINT2 (" Setting two bytes from 0x%x to no_op.\n", p
+2);
4926 p
[2] = (unsigned char) no_op
;
4927 p
[3] = (unsigned char) no_op
;
4933 EXTRACT_NUMBER (mcnt
, p
+ 2);
4934 DEBUG_PRINT2 ("EXECUTING jump_n %d.\n", mcnt
);
4936 /* Originally, this is how many times we CAN jump. */
4940 STORE_NUMBER (p
+ 2, mcnt
);
4942 DEBUG_PRINT3 (" Setting %p to %d.\n", p
+ 2, mcnt
);
4944 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p
+ 2, mcnt
);
4946 goto unconditional_jump
;
4948 /* If don't have to jump any more, skip over the rest of command. */
4955 DEBUG_PRINT1 ("EXECUTING set_number_at.\n");
4957 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
4959 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
4961 DEBUG_PRINT3 (" Setting %p to %d.\n", p1
, mcnt
);
4963 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p1
, mcnt
);
4965 STORE_NUMBER (p1
, mcnt
);
4970 /* The DEC Alpha C compiler 3.x generates incorrect code for the
4971 test WORDCHAR_P (d - 1) != WORDCHAR_P (d) in the expansion of
4972 AT_WORD_BOUNDARY, so this code is disabled. Expanding the
4973 macro and introducing temporary variables works around the bug. */
4976 DEBUG_PRINT1 ("EXECUTING wordbound.\n");
4977 if (AT_WORD_BOUNDARY (d
))
4982 DEBUG_PRINT1 ("EXECUTING notwordbound.\n");
4983 if (AT_WORD_BOUNDARY (d
))
4989 boolean prevchar
, thischar
;
4991 DEBUG_PRINT1 ("EXECUTING wordbound.\n");
4992 if (AT_STRINGS_BEG (d
) || AT_STRINGS_END (d
))
4995 prevchar
= WORDCHAR_P (d
- 1);
4996 thischar
= WORDCHAR_P (d
);
4997 if (prevchar
!= thischar
)
5004 boolean prevchar
, thischar
;
5006 DEBUG_PRINT1 ("EXECUTING notwordbound.\n");
5007 if (AT_STRINGS_BEG (d
) || AT_STRINGS_END (d
))
5010 prevchar
= WORDCHAR_P (d
- 1);
5011 thischar
= WORDCHAR_P (d
);
5012 if (prevchar
!= thischar
)
5019 DEBUG_PRINT1 ("EXECUTING wordbeg.\n");
5020 if (WORDCHAR_P (d
) && (AT_STRINGS_BEG (d
) || !WORDCHAR_P (d
- 1)))
5025 DEBUG_PRINT1 ("EXECUTING wordend.\n");
5026 if (!AT_STRINGS_BEG (d
) && WORDCHAR_P (d
- 1)
5027 && (!WORDCHAR_P (d
) || AT_STRINGS_END (d
)))
5033 DEBUG_PRINT1 ("EXECUTING before_dot.\n");
5034 if (PTR_CHAR_POS ((unsigned char *) d
) >= point
)
5039 DEBUG_PRINT1 ("EXECUTING at_dot.\n");
5040 if (PTR_CHAR_POS ((unsigned char *) d
) != point
)
5045 DEBUG_PRINT1 ("EXECUTING after_dot.\n");
5046 if (PTR_CHAR_POS ((unsigned char *) d
) <= point
)
5051 DEBUG_PRINT2 ("EXECUTING syntaxspec %d.\n", mcnt
);
5056 DEBUG_PRINT1 ("EXECUTING Emacs wordchar.\n");
5060 /* Can't use *d++ here; SYNTAX may be an unsafe macro. */
5062 if (SYNTAX (d
[-1]) != (enum syntaxcode
) mcnt
)
5064 SET_REGS_MATCHED ();
5068 DEBUG_PRINT2 ("EXECUTING notsyntaxspec %d.\n", mcnt
);
5070 goto matchnotsyntax
;
5073 DEBUG_PRINT1 ("EXECUTING Emacs notwordchar.\n");
5077 /* Can't use *d++ here; SYNTAX may be an unsafe macro. */
5079 if (SYNTAX (d
[-1]) == (enum syntaxcode
) mcnt
)
5081 SET_REGS_MATCHED ();
5084 #else /* not emacs */
5086 DEBUG_PRINT1 ("EXECUTING non-Emacs wordchar.\n");
5088 if (!WORDCHAR_P (d
))
5090 SET_REGS_MATCHED ();
5095 DEBUG_PRINT1 ("EXECUTING non-Emacs notwordchar.\n");
5099 SET_REGS_MATCHED ();
5102 #endif /* not emacs */
5107 continue; /* Successfully executed one pattern command; keep going. */
5110 /* We goto here if a matching operation fails. */
5112 if (!FAIL_STACK_EMPTY ())
5113 { /* A restart point is known. Restore to that state. */
5114 DEBUG_PRINT1 ("\nFAIL:\n");
5115 POP_FAILURE_POINT (d
, p
,
5116 lowest_active_reg
, highest_active_reg
,
5117 regstart
, regend
, reg_info
);
5119 /* If this failure point is a dummy, try the next one. */
5123 /* If we failed to the end of the pattern, don't examine *p. */
5127 boolean is_a_jump_n
= false;
5129 /* If failed to a backwards jump that's part of a repetition
5130 loop, need to pop this failure point and use the next one. */
5131 switch ((re_opcode_t
) *p
)
5135 case maybe_pop_jump
:
5136 case pop_failure_jump
:
5139 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
5142 if ((is_a_jump_n
&& (re_opcode_t
) *p1
== succeed_n
)
5144 && (re_opcode_t
) *p1
== on_failure_jump
))
5152 if (d
>= string1
&& d
<= end1
)
5156 break; /* Matching at this starting point really fails. */
5160 goto restore_best_regs
;
5164 return -1; /* Failure to match. */
5167 /* Subroutine definitions for re_match_2. */
5170 /* We are passed P pointing to a register number after a start_memory.
5172 Return true if the pattern up to the corresponding stop_memory can
5173 match the empty string, and false otherwise.
5175 If we find the matching stop_memory, sets P to point to one past its number.
5176 Otherwise, sets P to an undefined byte less than or equal to END.
5178 We don't handle duplicates properly (yet). */
5181 group_match_null_string_p (p
, end
, reg_info
)
5182 unsigned char **p
, *end
;
5183 register_info_type
*reg_info
;
5186 /* Point to after the args to the start_memory. */
5187 unsigned char *p1
= *p
+ 2;
5191 /* Skip over opcodes that can match nothing, and return true or
5192 false, as appropriate, when we get to one that can't, or to the
5193 matching stop_memory. */
5195 switch ((re_opcode_t
) *p1
)
5197 /* Could be either a loop or a series of alternatives. */
5198 case on_failure_jump
:
5200 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
5202 /* If the next operation is not a jump backwards in the
5207 /* Go through the on_failure_jumps of the alternatives,
5208 seeing if any of the alternatives cannot match nothing.
5209 The last alternative starts with only a jump,
5210 whereas the rest start with on_failure_jump and end
5211 with a jump, e.g., here is the pattern for `a|b|c':
5213 /on_failure_jump/0/6/exactn/1/a/jump_past_alt/0/6
5214 /on_failure_jump/0/6/exactn/1/b/jump_past_alt/0/3
5217 So, we have to first go through the first (n-1)
5218 alternatives and then deal with the last one separately. */
5221 /* Deal with the first (n-1) alternatives, which start
5222 with an on_failure_jump (see above) that jumps to right
5223 past a jump_past_alt. */
5225 while ((re_opcode_t
) p1
[mcnt
-3] == jump_past_alt
)
5227 /* `mcnt' holds how many bytes long the alternative
5228 is, including the ending `jump_past_alt' and
5231 if (!alt_match_null_string_p (p1
, p1
+ mcnt
- 3,
5235 /* Move to right after this alternative, including the
5239 /* Break if it's the beginning of an n-th alternative
5240 that doesn't begin with an on_failure_jump. */
5241 if ((re_opcode_t
) *p1
!= on_failure_jump
)
5244 /* Still have to check that it's not an n-th
5245 alternative that starts with an on_failure_jump. */
5247 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
5248 if ((re_opcode_t
) p1
[mcnt
-3] != jump_past_alt
)
5250 /* Get to the beginning of the n-th alternative. */
5256 /* Deal with the last alternative: go back and get number
5257 of the `jump_past_alt' just before it. `mcnt' contains
5258 the length of the alternative. */
5259 EXTRACT_NUMBER (mcnt
, p1
- 2);
5261 if (!alt_match_null_string_p (p1
, p1
+ mcnt
, reg_info
))
5264 p1
+= mcnt
; /* Get past the n-th alternative. */
5270 assert (p1
[1] == **p
);
5276 if (!common_op_match_null_string_p (&p1
, end
, reg_info
))
5279 } /* while p1 < end */
5282 } /* group_match_null_string_p */
5285 /* Similar to group_match_null_string_p, but doesn't deal with alternatives:
5286 It expects P to be the first byte of a single alternative and END one
5287 byte past the last. The alternative can contain groups. */
5290 alt_match_null_string_p (p
, end
, reg_info
)
5291 unsigned char *p
, *end
;
5292 register_info_type
*reg_info
;
5295 unsigned char *p1
= p
;
5299 /* Skip over opcodes that can match nothing, and break when we get
5300 to one that can't. */
5302 switch ((re_opcode_t
) *p1
)
5305 case on_failure_jump
:
5307 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
5312 if (!common_op_match_null_string_p (&p1
, end
, reg_info
))
5315 } /* while p1 < end */
5318 } /* alt_match_null_string_p */
5321 /* Deals with the ops common to group_match_null_string_p and
5322 alt_match_null_string_p.
5324 Sets P to one after the op and its arguments, if any. */
5327 common_op_match_null_string_p (p
, end
, reg_info
)
5328 unsigned char **p
, *end
;
5329 register_info_type
*reg_info
;
5334 unsigned char *p1
= *p
;
5336 switch ((re_opcode_t
) *p1
++)
5356 assert (reg_no
> 0 && reg_no
<= MAX_REGNUM
);
5357 ret
= group_match_null_string_p (&p1
, end
, reg_info
);
5359 /* Have to set this here in case we're checking a group which
5360 contains a group and a back reference to it. */
5362 if (REG_MATCH_NULL_STRING_P (reg_info
[reg_no
]) == MATCH_NULL_UNSET_VALUE
)
5363 REG_MATCH_NULL_STRING_P (reg_info
[reg_no
]) = ret
;
5369 /* If this is an optimized succeed_n for zero times, make the jump. */
5371 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
5379 /* Get to the number of times to succeed. */
5381 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
5386 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
5394 if (!REG_MATCH_NULL_STRING_P (reg_info
[*p1
]))
5402 /* All other opcodes mean we cannot match the empty string. */
5408 } /* common_op_match_null_string_p */
5411 /* Return zero if TRANSLATE[S1] and TRANSLATE[S2] are identical for LEN
5412 bytes; nonzero otherwise. */
5415 bcmp_translate (s1
, s2
, len
, translate
)
5416 const char *s1
, *s2
;
5418 RE_TRANSLATE_TYPE translate
;
5420 register const unsigned char *p1
= (const unsigned char *) s1
;
5421 register const unsigned char *p2
= (const unsigned char *) s2
;
5424 if (translate
[*p1
++] != translate
[*p2
++]) return 1;
5430 /* Entry points for GNU code. */
5432 /* re_compile_pattern is the GNU regular expression compiler: it
5433 compiles PATTERN (of length SIZE) and puts the result in BUFP.
5434 Returns 0 if the pattern was valid, otherwise an error string.
5436 Assumes the `allocated' (and perhaps `buffer') and `translate' fields
5437 are set in BUFP on entry.
5439 We call regex_compile to do the actual compilation. */
5442 re_compile_pattern (pattern
, length
, bufp
)
5443 const char *pattern
;
5445 struct re_pattern_buffer
*bufp
;
5449 /* GNU code is written to assume at least RE_NREGS registers will be set
5450 (and at least one extra will be -1). */
5451 bufp
->regs_allocated
= REGS_UNALLOCATED
;
5453 /* And GNU code determines whether or not to get register information
5454 by passing null for the REGS argument to re_match, etc., not by
5458 /* Match anchors at newline. */
5459 bufp
->newline_anchor
= 1;
5461 ret
= regex_compile (pattern
, length
, re_syntax_options
, bufp
);
5465 return gettext (re_error_msgid
[(int) ret
]);
5468 weak_alias (__re_compile_pattern
, re_compile_pattern
)
5471 /* Entry points compatible with 4.2 BSD regex library. We don't define
5472 them unless specifically requested. */
5474 #if defined _REGEX_RE_COMP || defined _LIBC
5476 /* BSD has one and only one pattern buffer. */
5477 static struct re_pattern_buffer re_comp_buf
;
5481 /* Make these definitions weak in libc, so POSIX programs can redefine
5482 these names if they don't use our functions, and still use
5483 regcomp/regexec below without link errors. */
5493 if (!re_comp_buf
.buffer
)
5494 return gettext ("No previous regular expression");
5498 if (!re_comp_buf
.buffer
)
5500 re_comp_buf
.buffer
= (unsigned char *) malloc (200);
5501 if (re_comp_buf
.buffer
== NULL
)
5502 return (char *) gettext (re_error_msgid
[(int) REG_ESPACE
]);
5503 re_comp_buf
.allocated
= 200;
5505 re_comp_buf
.fastmap
= (char *) malloc (1 << BYTEWIDTH
);
5506 if (re_comp_buf
.fastmap
== NULL
)
5507 return (char *) gettext (re_error_msgid
[(int) REG_ESPACE
]);
5510 /* Since `re_exec' always passes NULL for the `regs' argument, we
5511 don't need to initialize the pattern buffer fields which affect it. */
5513 /* Match anchors at newlines. */
5514 re_comp_buf
.newline_anchor
= 1;
5516 ret
= regex_compile (s
, strlen (s
), re_syntax_options
, &re_comp_buf
);
5521 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
5522 return (char *) gettext (re_error_msgid
[(int) ret
]);
5533 const int len
= strlen (s
);
5535 0 <= re_search (&re_comp_buf
, s
, len
, 0, len
, (struct re_registers
*) 0);
5538 #endif /* _REGEX_RE_COMP */
5540 /* POSIX.2 functions. Don't define these for Emacs. */
5544 /* regcomp takes a regular expression as a string and compiles it.
5546 PREG is a regex_t *. We do not expect any fields to be initialized,
5547 since POSIX says we shouldn't. Thus, we set
5549 `buffer' to the compiled pattern;
5550 `used' to the length of the compiled pattern;
5551 `syntax' to RE_SYNTAX_POSIX_EXTENDED if the
5552 REG_EXTENDED bit in CFLAGS is set; otherwise, to
5553 RE_SYNTAX_POSIX_BASIC;
5554 `newline_anchor' to REG_NEWLINE being set in CFLAGS;
5555 `fastmap' and `fastmap_accurate' to zero;
5556 `re_nsub' to the number of subexpressions in PATTERN.
5558 PATTERN is the address of the pattern string.
5560 CFLAGS is a series of bits which affect compilation.
5562 If REG_EXTENDED is set, we use POSIX extended syntax; otherwise, we
5563 use POSIX basic syntax.
5565 If REG_NEWLINE is set, then . and [^...] don't match newline.
5566 Also, regexec will try a match beginning after every newline.
5568 If REG_ICASE is set, then we considers upper- and lowercase
5569 versions of letters to be equivalent when matching.
5571 If REG_NOSUB is set, then when PREG is passed to regexec, that
5572 routine will report only success or failure, and nothing about the
5575 It returns 0 if it succeeds, nonzero if it doesn't. (See gnu-regex.h for
5576 the return codes and their meanings.) */
5579 regcomp (preg
, pattern
, cflags
)
5581 const char *pattern
;
5586 = (cflags
& REG_EXTENDED
) ?
5587 RE_SYNTAX_POSIX_EXTENDED
: RE_SYNTAX_POSIX_BASIC
;
5589 /* regex_compile will allocate the space for the compiled pattern. */
5591 preg
->allocated
= 0;
5594 /* Don't bother to use a fastmap when searching. This simplifies the
5595 REG_NEWLINE case: if we used a fastmap, we'd have to put all the
5596 characters after newlines into the fastmap. This way, we just try
5600 if (cflags
& REG_ICASE
)
5605 = (RE_TRANSLATE_TYPE
) malloc (CHAR_SET_SIZE
5606 * sizeof (*(RE_TRANSLATE_TYPE
)0));
5607 if (preg
->translate
== NULL
)
5608 return (int) REG_ESPACE
;
5610 /* Map uppercase characters to corresponding lowercase ones. */
5611 for (i
= 0; i
< CHAR_SET_SIZE
; i
++)
5612 preg
->translate
[i
] = ISUPPER (i
) ? tolower (i
) : i
;
5615 preg
->translate
= NULL
;
5617 /* If REG_NEWLINE is set, newlines are treated differently. */
5618 if (cflags
& REG_NEWLINE
)
5619 { /* REG_NEWLINE implies neither . nor [^...] match newline. */
5620 syntax
&= ~RE_DOT_NEWLINE
;
5621 syntax
|= RE_HAT_LISTS_NOT_NEWLINE
;
5622 /* It also changes the matching behavior. */
5623 preg
->newline_anchor
= 1;
5626 preg
->newline_anchor
= 0;
5628 preg
->no_sub
= !!(cflags
& REG_NOSUB
);
5630 /* POSIX says a null character in the pattern terminates it, so we
5631 can use strlen here in compiling the pattern. */
5632 ret
= regex_compile (pattern
, strlen (pattern
), syntax
, preg
);
5634 /* POSIX doesn't distinguish between an unmatched open-group and an
5635 unmatched close-group: both are REG_EPAREN. */
5636 if (ret
== REG_ERPAREN
) ret
= REG_EPAREN
;
5641 weak_alias (__regcomp
, regcomp
)
5645 /* regexec searches for a given pattern, specified by PREG, in the
5648 If NMATCH is zero or REG_NOSUB was set in the cflags argument to
5649 `regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at
5650 least NMATCH elements, and we set them to the offsets of the
5651 corresponding matched substrings.
5653 EFLAGS specifies `execution flags' which affect matching: if
5654 REG_NOTBOL is set, then ^ does not match at the beginning of the
5655 string; if REG_NOTEOL is set, then $ does not match at the end.
5657 We return 0 if we find a match and REG_NOMATCH if not. */
5660 regexec (preg
, string
, nmatch
, pmatch
, eflags
)
5661 const regex_t
*preg
;
5664 regmatch_t pmatch
[];
5668 struct re_registers regs
;
5669 regex_t private_preg
;
5670 int len
= strlen (string
);
5671 boolean want_reg_info
= !preg
->no_sub
&& nmatch
> 0;
5673 private_preg
= *preg
;
5675 private_preg
.not_bol
= !!(eflags
& REG_NOTBOL
);
5676 private_preg
.not_eol
= !!(eflags
& REG_NOTEOL
);
5678 /* The user has told us exactly how many registers to return
5679 information about, via `nmatch'. We have to pass that on to the
5680 matching routines. */
5681 private_preg
.regs_allocated
= REGS_FIXED
;
5685 regs
.num_regs
= nmatch
;
5686 regs
.start
= TALLOC (nmatch
, regoff_t
);
5687 regs
.end
= TALLOC (nmatch
, regoff_t
);
5688 if (regs
.start
== NULL
|| regs
.end
== NULL
)
5689 return (int) REG_NOMATCH
;
5692 /* Perform the searching operation. */
5693 ret
= re_search (&private_preg
, string
, len
,
5694 /* start: */ 0, /* range: */ len
,
5695 want_reg_info
? ®s
: (struct re_registers
*) 0);
5697 /* Copy the register information to the POSIX structure. */
5704 for (r
= 0; r
< nmatch
; r
++)
5706 pmatch
[r
].rm_so
= regs
.start
[r
];
5707 pmatch
[r
].rm_eo
= regs
.end
[r
];
5711 /* If we needed the temporary register info, free the space now. */
5716 /* We want zero return to mean success, unlike `re_search'. */
5717 return ret
>= 0 ? (int) REG_NOERROR
: (int) REG_NOMATCH
;
5720 weak_alias (__regexec
, regexec
)
5724 /* Returns a message corresponding to an error code, ERRCODE, returned
5725 from either regcomp or regexec. We don't use PREG here. */
5728 __regerror (errcode
, preg
, errbuf
, errbuf_size
)
5730 const regex_t
*preg
;
5738 || errcode
>= (int) (sizeof (re_error_msgid
)
5739 / sizeof (re_error_msgid
[0])))
5740 /* Only error codes returned by the rest of the code should be passed
5741 to this routine. If we are given anything else, or if other regex
5742 code generates an invalid error code, then the program has a bug.
5743 Dump core so we can fix it. */
5746 msg
= gettext (re_error_msgid
[errcode
]);
5748 msg_size
= strlen (msg
) + 1; /* Includes the null. */
5750 if (errbuf_size
!= 0)
5752 if (msg_size
> errbuf_size
)
5754 #if defined HAVE_MEMPCPY || defined _LIBC
5755 *((char *) __mempcpy (errbuf
, msg
, errbuf_size
- 1)) = '\0';
5757 memcpy (errbuf
, msg
, errbuf_size
- 1);
5758 errbuf
[errbuf_size
- 1] = 0;
5762 memcpy (errbuf
, msg
, msg_size
);
5768 weak_alias (__regerror
, regerror
)
5772 /* Free dynamically allocated space used by PREG. */
5778 if (preg
->buffer
!= NULL
)
5779 free (preg
->buffer
);
5780 preg
->buffer
= NULL
;
5782 preg
->allocated
= 0;
5785 if (preg
->fastmap
!= NULL
)
5786 free (preg
->fastmap
);
5787 preg
->fastmap
= NULL
;
5788 preg
->fastmap_accurate
= 0;
5790 if (preg
->translate
!= NULL
)
5791 free (preg
->translate
);
5792 preg
->translate
= NULL
;
5795 weak_alias (__regfree
, regfree
)
5798 #endif /* not emacs */