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,
24 Boston, MA 02111-1307, USA. */
26 /* AIX requires this to be the first thing in the file. */
27 #if defined _AIX && !defined REGEX_MALLOC
39 # if defined __GNUC__ || (defined __STDC__ && __STDC__)
40 # define PARAMS(args) args
42 # define PARAMS(args) ()
44 #endif /* Not PARAMS. */
46 #if defined STDC_HEADERS && !defined emacs
49 /* We need this for `gnu-regex.h', and perhaps for the Emacs include files. */
50 # include <sys/types.h>
53 /* For platform which support the ISO C amendement 1 functionality we
54 support user defined character classes. */
55 #if defined _LIBC || (defined HAVE_WCTYPE_H && defined HAVE_WCHAR_H)
56 /* Solaris 2.5 has a bug: <wchar.h> must be included before <wctype.h>. */
61 /* This is for other GNU distributions with internationalized messages. */
62 /* CYGNUS LOCAL: ../intl will handle this for us */
66 # define gettext(msgid) (msgid)
70 /* This define is so xgettext can find the internationalizable
72 # define gettext_noop(String) String
75 /* The `emacs' switch turns on certain matching commands
76 that make sense only in Emacs. */
85 /* If we are not linking with Emacs proper,
86 we can't use the relocating allocator
87 even if config.h says that we can. */
90 # if defined STDC_HEADERS || defined _LIBC
97 /* When used in Emacs's lib-src, we need to get bzero and bcopy somehow.
98 If nothing else has been done, use the method below. */
99 # ifdef INHIBIT_STRING_HEADER
100 # if !(defined HAVE_BZERO && defined HAVE_BCOPY)
101 # if !defined bzero && !defined bcopy
102 # undef INHIBIT_STRING_HEADER
107 /* This is the normal way of making sure we have a bcopy and a bzero.
108 This is used in most programs--a few other programs avoid this
109 by defining INHIBIT_STRING_HEADER. */
110 # ifndef INHIBIT_STRING_HEADER
111 # if defined HAVE_STRING_H || defined STDC_HEADERS || defined _LIBC
115 # define bzero(s, n) (memset (s, '\0', n), (s))
117 # define bzero(s, n) __bzero (s, n)
121 # include <strings.h>
123 # define memcmp(s1, s2, n) bcmp (s1, s2, n)
126 # define memcpy(d, s, n) (bcopy (s, d, n), (d))
131 /* Define the syntax stuff for \<, \>, etc. */
133 /* This must be nonzero for the wordchar and notwordchar pattern
134 commands in re_match_2. */
139 # ifdef SWITCH_ENUM_BUG
140 # define SWITCH_ENUM_CAST(x) ((int)(x))
142 # define SWITCH_ENUM_CAST(x) (x)
145 /* How many characters in the character set. */
146 # define CHAR_SET_SIZE 256
148 /* GDB LOCAL: define _REGEX_RE_COMP to get BSD style re_comp and re_exec */
149 #ifndef _REGEX_RE_COMP
150 #define _REGEX_RE_COMP
155 extern char *re_syntax_table
;
157 # else /* not SYNTAX_TABLE */
159 static char re_syntax_table
[CHAR_SET_SIZE
];
170 bzero (re_syntax_table
, sizeof re_syntax_table
);
172 for (c
= 'a'; c
<= 'z'; c
++)
173 re_syntax_table
[c
] = Sword
;
175 for (c
= 'A'; c
<= 'Z'; c
++)
176 re_syntax_table
[c
] = Sword
;
178 for (c
= '0'; c
<= '9'; c
++)
179 re_syntax_table
[c
] = Sword
;
181 re_syntax_table
['_'] = Sword
;
186 # endif /* not SYNTAX_TABLE */
188 # define SYNTAX(c) re_syntax_table[c]
190 #endif /* not emacs */
192 /* Get the interface, including the syntax bits. */
193 /* CYGNUS LOCAL: call it gnu-regex.h, not regex.h, to avoid name conflicts */
194 #include "gnu-regex.h"
196 /* isalpha etc. are used for the character classes. */
199 /* Jim Meyering writes:
201 "... Some ctype macros are valid only for character codes that
202 isascii says are ASCII (SGI's IRIX-4.0.5 is one such system --when
203 using /bin/cc or gcc but without giving an ansi option). So, all
204 ctype uses should be through macros like ISPRINT... If
205 STDC_HEADERS is defined, then autoconf has verified that the ctype
206 macros don't need to be guarded with references to isascii. ...
207 Defining isascii to 1 should let any compiler worth its salt
208 eliminate the && through constant folding."
209 Solaris defines some of these symbols so we must undefine them first. */
212 #if defined STDC_HEADERS || (!defined isascii && !defined HAVE_ISASCII)
213 # define ISASCII(c) 1
215 # define ISASCII(c) isascii(c)
219 # define ISBLANK(c) (ISASCII (c) && isblank (c))
221 # define ISBLANK(c) ((c) == ' ' || (c) == '\t')
224 # define ISGRAPH(c) (ISASCII (c) && isgraph (c))
226 # define ISGRAPH(c) (ISASCII (c) && isprint (c) && !isspace (c))
230 #define ISPRINT(c) (ISASCII (c) && isprint (c))
231 #define ISDIGIT(c) (ISASCII (c) && isdigit (c))
232 #define ISALNUM(c) (ISASCII (c) && isalnum (c))
233 #define ISALPHA(c) (ISASCII (c) && isalpha (c))
234 #define ISCNTRL(c) (ISASCII (c) && iscntrl (c))
235 #define ISLOWER(c) (ISASCII (c) && islower (c))
236 #define ISPUNCT(c) (ISASCII (c) && ispunct (c))
237 #define ISSPACE(c) (ISASCII (c) && isspace (c))
238 #define ISUPPER(c) (ISASCII (c) && isupper (c))
239 #define ISXDIGIT(c) (ISASCII (c) && isxdigit (c))
242 # define NULL (void *)0
245 /* We remove any previous definition of `SIGN_EXTEND_CHAR',
246 since ours (we hope) works properly with all combinations of
247 machines, compilers, `char' and `unsigned char' argument types.
248 (Per Bothner suggested the basic approach.) */
249 #undef SIGN_EXTEND_CHAR
251 # define SIGN_EXTEND_CHAR(c) ((signed char) (c))
252 #else /* not __STDC__ */
253 /* As in Harbison and Steele. */
254 # define SIGN_EXTEND_CHAR(c) ((((unsigned char) (c)) ^ 128) - 128)
257 /* Should we use malloc or alloca? If REGEX_MALLOC is not defined, we
258 use `alloca' instead of `malloc'. This is because using malloc in
259 re_search* or re_match* could cause memory leaks when C-g is used in
260 Emacs; also, malloc is slower and causes storage fragmentation. On
261 the other hand, malloc is more portable, and easier to debug.
263 Because we sometimes use alloca, some routines have to be macros,
264 not functions -- `alloca'-allocated space disappears at the end of the
265 function it is called in. */
269 # define REGEX_ALLOCATE malloc
270 # define REGEX_REALLOCATE(source, osize, nsize) realloc (source, nsize)
271 # define REGEX_FREE free
273 #else /* not REGEX_MALLOC */
275 /* Emacs already defines alloca, sometimes. */
278 /* Make alloca work the best possible way. */
280 # define alloca __builtin_alloca
281 # else /* not __GNUC__ */
284 # endif /* HAVE_ALLOCA_H */
285 # endif /* not __GNUC__ */
287 # endif /* not alloca */
289 # define REGEX_ALLOCATE alloca
291 /* Assumes a `char *destination' variable. */
292 # define REGEX_REALLOCATE(source, osize, nsize) \
293 (destination = (char *) alloca (nsize), \
294 memcpy (destination, source, osize))
296 /* No need to do anything to free, after alloca. */
297 # define REGEX_FREE(arg) ((void)0) /* Do nothing! But inhibit gcc warning. */
299 #endif /* not REGEX_MALLOC */
301 /* Define how to allocate the failure stack. */
303 #if defined REL_ALLOC && defined REGEX_MALLOC
305 # define REGEX_ALLOCATE_STACK(size) \
306 r_alloc (&failure_stack_ptr, (size))
307 # define REGEX_REALLOCATE_STACK(source, osize, nsize) \
308 r_re_alloc (&failure_stack_ptr, (nsize))
309 # define REGEX_FREE_STACK(ptr) \
310 r_alloc_free (&failure_stack_ptr)
312 #else /* not using relocating allocator */
316 # define REGEX_ALLOCATE_STACK malloc
317 # define REGEX_REALLOCATE_STACK(source, osize, nsize) realloc (source, nsize)
318 # define REGEX_FREE_STACK free
320 # else /* not REGEX_MALLOC */
322 # define REGEX_ALLOCATE_STACK alloca
324 # define REGEX_REALLOCATE_STACK(source, osize, nsize) \
325 REGEX_REALLOCATE (source, osize, nsize)
326 /* No need to explicitly free anything. */
327 # define REGEX_FREE_STACK(arg)
329 # endif /* not REGEX_MALLOC */
330 #endif /* not using relocating allocator */
333 /* True if `size1' is non-NULL and PTR is pointing anywhere inside
334 `string1' or just past its end. This works if PTR is NULL, which is
336 #define FIRST_STRING_P(ptr) \
337 (size1 && string1 <= (ptr) && (ptr) <= string1 + size1)
339 /* (Re)Allocate N items of type T using malloc, or fail. */
340 #define TALLOC(n, t) ((t *) malloc ((n) * sizeof (t)))
341 #define RETALLOC(addr, n, t) ((addr) = (t *) realloc (addr, (n) * sizeof (t)))
342 #define RETALLOC_IF(addr, n, t) \
343 if (addr) RETALLOC((addr), (n), t); else (addr) = TALLOC ((n), t)
344 #define REGEX_TALLOC(n, t) ((t *) REGEX_ALLOCATE ((n) * sizeof (t)))
346 #define BYTEWIDTH 8 /* In bits. */
348 #define STREQ(s1, s2) ((strcmp (s1, s2) == 0))
352 #define MAX(a, b) ((a) > (b) ? (a) : (b))
353 #define MIN(a, b) ((a) < (b) ? (a) : (b))
355 typedef char boolean
;
359 static int re_match_2_internal
PARAMS ((struct re_pattern_buffer
*bufp
,
360 const char *string1
, int size1
,
361 const char *string2
, int size2
,
363 struct re_registers
*regs
,
366 /* These are the command codes that appear in compiled regular
367 expressions. Some opcodes are followed by argument bytes. A
368 command code can specify any interpretation whatsoever for its
369 arguments. Zero bytes may appear in the compiled regular expression. */
375 /* Succeed right away--no more backtracking. */
378 /* Followed by one byte giving n, then by n literal bytes. */
381 /* Matches any (more or less) character. */
384 /* Matches any one char belonging to specified set. First
385 following byte is number of bitmap bytes. Then come bytes
386 for a bitmap saying which chars are in. Bits in each byte
387 are ordered low-bit-first. A character is in the set if its
388 bit is 1. A character too large to have a bit in the map is
389 automatically not in the set. */
392 /* Same parameters as charset, but match any character that is
393 not one of those specified. */
396 /* Start remembering the text that is matched, for storing in a
397 register. Followed by one byte with the register number, in
398 the range 0 to one less than the pattern buffer's re_nsub
399 field. Then followed by one byte with the number of groups
400 inner to this one. (This last has to be part of the
401 start_memory only because we need it in the on_failure_jump
405 /* Stop remembering the text that is matched and store it in a
406 memory register. Followed by one byte with the register
407 number, in the range 0 to one less than `re_nsub' in the
408 pattern buffer, and one byte with the number of inner groups,
409 just like `start_memory'. (We need the number of inner
410 groups here because we don't have any easy way of finding the
411 corresponding start_memory when we're at a stop_memory.) */
414 /* Match a duplicate of something remembered. Followed by one
415 byte containing the register number. */
418 /* Fail unless at beginning of line. */
421 /* Fail unless at end of line. */
424 /* Succeeds if at beginning of buffer (if emacs) or at beginning
425 of string to be matched (if not). */
428 /* Analogously, for end of buffer/string. */
431 /* Followed by two byte relative address to which to jump. */
434 /* Same as jump, but marks the end of an alternative. */
437 /* Followed by two-byte relative address of place to resume at
438 in case of failure. */
441 /* Like on_failure_jump, but pushes a placeholder instead of the
442 current string position when executed. */
443 on_failure_keep_string_jump
,
445 /* Throw away latest failure point and then jump to following
446 two-byte relative address. */
449 /* Change to pop_failure_jump if know won't have to backtrack to
450 match; otherwise change to jump. This is used to jump
451 back to the beginning of a repeat. If what follows this jump
452 clearly won't match what the repeat does, such that we can be
453 sure that there is no use backtracking out of repetitions
454 already matched, then we change it to a pop_failure_jump.
455 Followed by two-byte address. */
458 /* Jump to following two-byte address, and push a dummy failure
459 point. This failure point will be thrown away if an attempt
460 is made to use it for a failure. A `+' construct makes this
461 before the first repeat. Also used as an intermediary kind
462 of jump when compiling an alternative. */
465 /* Push a dummy failure point and continue. Used at the end of
469 /* Followed by two-byte relative address and two-byte number n.
470 After matching N times, jump to the address upon failure. */
473 /* Followed by two-byte relative address, and two-byte number n.
474 Jump to the address N times, then fail. */
477 /* Set the following two-byte relative address to the
478 subsequent two-byte number. The address *includes* the two
482 wordchar
, /* Matches any word-constituent character. */
483 notwordchar
, /* Matches any char that is not a word-constituent. */
485 wordbeg
, /* Succeeds if at word beginning. */
486 wordend
, /* Succeeds if at word end. */
488 wordbound
, /* Succeeds if at a word boundary. */
489 notwordbound
/* Succeeds if not at a word boundary. */
492 ,before_dot
, /* Succeeds if before point. */
493 at_dot
, /* Succeeds if at point. */
494 after_dot
, /* Succeeds if after point. */
496 /* Matches any character whose syntax is specified. Followed by
497 a byte which contains a syntax code, e.g., Sword. */
500 /* Matches any character whose syntax is not that specified. */
505 /* Common operations on the compiled pattern. */
507 /* Store NUMBER in two contiguous bytes starting at DESTINATION. */
509 #define STORE_NUMBER(destination, number) \
511 (destination)[0] = (number) & 0377; \
512 (destination)[1] = (number) >> 8; \
515 /* Same as STORE_NUMBER, except increment DESTINATION to
516 the byte after where the number is stored. Therefore, DESTINATION
517 must be an lvalue. */
519 #define STORE_NUMBER_AND_INCR(destination, number) \
521 STORE_NUMBER (destination, number); \
522 (destination) += 2; \
525 /* Put into DESTINATION a number stored in two contiguous bytes starting
528 #define EXTRACT_NUMBER(destination, source) \
530 (destination) = *(source) & 0377; \
531 (destination) += SIGN_EXTEND_CHAR (*((source) + 1)) << 8; \
535 static void extract_number
_RE_ARGS ((int *dest
, unsigned char *source
));
537 extract_number (dest
, source
)
539 unsigned char *source
;
541 int temp
= SIGN_EXTEND_CHAR (*(source
+ 1));
542 *dest
= *source
& 0377;
546 # ifndef EXTRACT_MACROS /* To debug the macros. */
547 # undef EXTRACT_NUMBER
548 # define EXTRACT_NUMBER(dest, src) extract_number (&dest, src)
549 # endif /* not EXTRACT_MACROS */
553 /* Same as EXTRACT_NUMBER, except increment SOURCE to after the number.
554 SOURCE must be an lvalue. */
556 #define EXTRACT_NUMBER_AND_INCR(destination, source) \
558 EXTRACT_NUMBER (destination, source); \
563 static void extract_number_and_incr
_RE_ARGS ((int *destination
,
564 unsigned char **source
));
566 extract_number_and_incr (destination
, source
)
568 unsigned char **source
;
570 extract_number (destination
, *source
);
574 # ifndef EXTRACT_MACROS
575 # undef EXTRACT_NUMBER_AND_INCR
576 # define EXTRACT_NUMBER_AND_INCR(dest, src) \
577 extract_number_and_incr (&dest, &src)
578 # endif /* not EXTRACT_MACROS */
582 /* If DEBUG is defined, Regex prints many voluminous messages about what
583 it is doing (if the variable `debug' is nonzero). If linked with the
584 main program in `iregex.c', you can enter patterns and strings
585 interactively. And if linked with the main program in `main.c' and
586 the other test files, you can run the already-written tests. */
590 /* We use standard I/O for debugging. */
593 /* It is useful to test things that ``must'' be true when debugging. */
596 static int debug
= 0;
598 # define DEBUG_STATEMENT(e) e
599 # define DEBUG_PRINT1(x) if (debug) printf (x)
600 # define DEBUG_PRINT2(x1, x2) if (debug) printf (x1, x2)
601 # define DEBUG_PRINT3(x1, x2, x3) if (debug) printf (x1, x2, x3)
602 # define DEBUG_PRINT4(x1, x2, x3, x4) if (debug) printf (x1, x2, x3, x4)
603 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) \
604 if (debug) print_partial_compiled_pattern (s, e)
605 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) \
606 if (debug) print_double_string (w, s1, sz1, s2, sz2)
609 /* Print the fastmap in human-readable form. */
612 print_fastmap (fastmap
)
615 unsigned was_a_range
= 0;
618 while (i
< (1 << BYTEWIDTH
))
624 while (i
< (1 << BYTEWIDTH
) && fastmap
[i
])
640 /* Print a compiled pattern string in human-readable form, starting at
641 the START pointer into it and ending just before the pointer END. */
644 print_partial_compiled_pattern (start
, end
)
645 unsigned char *start
;
650 unsigned char *p
= start
;
651 unsigned char *pend
= end
;
659 /* Loop over pattern commands. */
662 printf ("%d:\t", p
- start
);
664 switch ((re_opcode_t
) *p
++)
672 printf ("/exactn/%d", mcnt
);
683 printf ("/start_memory/%d/%d", mcnt
, *p
++);
688 printf ("/stop_memory/%d/%d", mcnt
, *p
++);
692 printf ("/duplicate/%d", *p
++);
702 register int c
, last
= -100;
703 register int in_range
= 0;
705 printf ("/charset [%s",
706 (re_opcode_t
) *(p
- 1) == charset_not
? "^" : "");
708 assert (p
+ *p
< pend
);
710 for (c
= 0; c
< 256; c
++)
712 && (p
[1 + (c
/8)] & (1 << (c
% 8))))
714 /* Are we starting a range? */
715 if (last
+ 1 == c
&& ! in_range
)
720 /* Have we broken a range? */
721 else if (last
+ 1 != c
&& in_range
)
750 case on_failure_jump
:
751 extract_number_and_incr (&mcnt
, &p
);
752 printf ("/on_failure_jump to %d", p
+ mcnt
- start
);
755 case on_failure_keep_string_jump
:
756 extract_number_and_incr (&mcnt
, &p
);
757 printf ("/on_failure_keep_string_jump to %d", p
+ mcnt
- start
);
760 case dummy_failure_jump
:
761 extract_number_and_incr (&mcnt
, &p
);
762 printf ("/dummy_failure_jump to %d", p
+ mcnt
- start
);
765 case push_dummy_failure
:
766 printf ("/push_dummy_failure");
770 extract_number_and_incr (&mcnt
, &p
);
771 printf ("/maybe_pop_jump to %d", p
+ mcnt
- start
);
774 case pop_failure_jump
:
775 extract_number_and_incr (&mcnt
, &p
);
776 printf ("/pop_failure_jump to %d", p
+ mcnt
- start
);
780 extract_number_and_incr (&mcnt
, &p
);
781 printf ("/jump_past_alt to %d", p
+ mcnt
- start
);
785 extract_number_and_incr (&mcnt
, &p
);
786 printf ("/jump to %d", p
+ mcnt
- start
);
790 extract_number_and_incr (&mcnt
, &p
);
792 extract_number_and_incr (&mcnt2
, &p
);
793 printf ("/succeed_n to %d, %d times", p1
- start
, mcnt2
);
797 extract_number_and_incr (&mcnt
, &p
);
799 extract_number_and_incr (&mcnt2
, &p
);
800 printf ("/jump_n to %d, %d times", p1
- start
, mcnt2
);
804 extract_number_and_incr (&mcnt
, &p
);
806 extract_number_and_incr (&mcnt2
, &p
);
807 printf ("/set_number_at location %d to %d", p1
- start
, mcnt2
);
811 printf ("/wordbound");
815 printf ("/notwordbound");
827 printf ("/before_dot");
835 printf ("/after_dot");
839 printf ("/syntaxspec");
841 printf ("/%d", mcnt
);
845 printf ("/notsyntaxspec");
847 printf ("/%d", mcnt
);
852 printf ("/wordchar");
856 printf ("/notwordchar");
868 printf ("?%d", *(p
-1));
874 printf ("%d:\tend of pattern.\n", p
- start
);
879 print_compiled_pattern (bufp
)
880 struct re_pattern_buffer
*bufp
;
882 unsigned char *buffer
= bufp
->buffer
;
884 print_partial_compiled_pattern (buffer
, buffer
+ bufp
->used
);
885 printf ("%ld bytes used/%ld bytes allocated.\n",
886 bufp
->used
, bufp
->allocated
);
888 if (bufp
->fastmap_accurate
&& bufp
->fastmap
)
890 printf ("fastmap: ");
891 print_fastmap (bufp
->fastmap
);
894 printf ("re_nsub: %d\t", bufp
->re_nsub
);
895 printf ("regs_alloc: %d\t", bufp
->regs_allocated
);
896 printf ("can_be_null: %d\t", bufp
->can_be_null
);
897 printf ("newline_anchor: %d\n", bufp
->newline_anchor
);
898 printf ("no_sub: %d\t", bufp
->no_sub
);
899 printf ("not_bol: %d\t", bufp
->not_bol
);
900 printf ("not_eol: %d\t", bufp
->not_eol
);
901 printf ("syntax: %lx\n", bufp
->syntax
);
902 /* Perhaps we should print the translate table? */
907 print_double_string (where
, string1
, size1
, string2
, size2
)
920 if (FIRST_STRING_P (where
))
922 for (this_char
= where
- string1
; this_char
< size1
; this_char
++)
923 putchar (string1
[this_char
]);
928 for (this_char
= where
- string2
; this_char
< size2
; this_char
++)
929 putchar (string2
[this_char
]);
940 #else /* not DEBUG */
945 # define DEBUG_STATEMENT(e)
946 # define DEBUG_PRINT1(x)
947 # define DEBUG_PRINT2(x1, x2)
948 # define DEBUG_PRINT3(x1, x2, x3)
949 # define DEBUG_PRINT4(x1, x2, x3, x4)
950 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e)
951 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2)
953 #endif /* not DEBUG */
955 /* Set by `re_set_syntax' to the current regexp syntax to recognize. Can
956 also be assigned to arbitrarily: each pattern buffer stores its own
957 syntax, so it can be changed between regex compilations. */
958 /* This has no initializer because initialized variables in Emacs
959 become read-only after dumping. */
960 reg_syntax_t re_syntax_options
;
963 /* Specify the precise syntax of regexps for compilation. This provides
964 for compatibility for various utilities which historically have
965 different, incompatible syntaxes.
967 The argument SYNTAX is a bit mask comprised of the various bits
968 defined in gnu-regex.h. We return the old syntax. */
971 re_set_syntax (syntax
)
974 reg_syntax_t ret
= re_syntax_options
;
976 re_syntax_options
= syntax
;
978 if (syntax
& RE_DEBUG
)
980 else if (debug
) /* was on but now is not */
986 weak_alias (__re_set_syntax
, re_set_syntax
)
989 /* This table gives an error message for each of the error codes listed
990 in gnu-regex.h. Obviously the order here has to be same as there.
991 POSIX doesn't require that we do anything for REG_NOERROR,
992 but why not be nice? */
994 static const char *re_error_msgid
[] =
996 gettext_noop ("Success"), /* REG_NOERROR */
997 gettext_noop ("No match"), /* REG_NOMATCH */
998 gettext_noop ("Invalid regular expression"), /* REG_BADPAT */
999 gettext_noop ("Invalid collation character"), /* REG_ECOLLATE */
1000 gettext_noop ("Invalid character class name"), /* REG_ECTYPE */
1001 gettext_noop ("Trailing backslash"), /* REG_EESCAPE */
1002 gettext_noop ("Invalid back reference"), /* REG_ESUBREG */
1003 gettext_noop ("Unmatched [ or [^"), /* REG_EBRACK */
1004 gettext_noop ("Unmatched ( or \\("), /* REG_EPAREN */
1005 gettext_noop ("Unmatched \\{"), /* REG_EBRACE */
1006 gettext_noop ("Invalid content of \\{\\}"), /* REG_BADBR */
1007 gettext_noop ("Invalid range end"), /* REG_ERANGE */
1008 gettext_noop ("Memory exhausted"), /* REG_ESPACE */
1009 gettext_noop ("Invalid preceding regular expression"), /* REG_BADRPT */
1010 gettext_noop ("Premature end of regular expression"), /* REG_EEND */
1011 gettext_noop ("Regular expression too big"), /* REG_ESIZE */
1012 gettext_noop ("Unmatched ) or \\)"), /* REG_ERPAREN */
1015 /* Avoiding alloca during matching, to placate r_alloc. */
1017 /* Define MATCH_MAY_ALLOCATE unless we need to make sure that the
1018 searching and matching functions should not call alloca. On some
1019 systems, alloca is implemented in terms of malloc, and if we're
1020 using the relocating allocator routines, then malloc could cause a
1021 relocation, which might (if the strings being searched are in the
1022 ralloc heap) shift the data out from underneath the regexp
1025 Here's another reason to avoid allocation: Emacs
1026 processes input from X in a signal handler; processing X input may
1027 call malloc; if input arrives while a matching routine is calling
1028 malloc, then we're scrod. But Emacs can't just block input while
1029 calling matching routines; then we don't notice interrupts when
1030 they come in. So, Emacs blocks input around all regexp calls
1031 except the matching calls, which it leaves unprotected, in the
1032 faith that they will not malloc. */
1034 /* Normally, this is fine. */
1035 #define MATCH_MAY_ALLOCATE
1037 /* When using GNU C, we are not REALLY using the C alloca, no matter
1038 what config.h may say. So don't take precautions for it. */
1043 /* The match routines may not allocate if (1) they would do it with malloc
1044 and (2) it's not safe for them to use malloc.
1045 Note that if REL_ALLOC is defined, matching would not use malloc for the
1046 failure stack, but we would still use it for the register vectors;
1047 so REL_ALLOC should not affect this. */
1048 #if (defined C_ALLOCA || defined REGEX_MALLOC) && defined emacs
1049 # undef MATCH_MAY_ALLOCATE
1053 /* Failure stack declarations and macros; both re_compile_fastmap and
1054 re_match_2 use a failure stack. These have to be macros because of
1055 REGEX_ALLOCATE_STACK. */
1058 /* Number of failure points for which to initially allocate space
1059 when matching. If this number is exceeded, we allocate more
1060 space, so it is not a hard limit. */
1061 #ifndef INIT_FAILURE_ALLOC
1062 # define INIT_FAILURE_ALLOC 5
1065 /* Roughly the maximum number of failure points on the stack. Would be
1066 exactly that if always used MAX_FAILURE_ITEMS items each time we failed.
1067 This is a variable only so users of regex can assign to it; we never
1068 change it ourselves. */
1072 # if defined MATCH_MAY_ALLOCATE
1073 /* 4400 was enough to cause a crash on Alpha OSF/1,
1074 whose default stack limit is 2mb. */
1075 long int re_max_failures
= 4000;
1077 long int re_max_failures
= 2000;
1080 union fail_stack_elt
1082 unsigned char *pointer
;
1086 typedef union fail_stack_elt fail_stack_elt_t
;
1090 fail_stack_elt_t
*stack
;
1091 unsigned long int size
;
1092 unsigned long int avail
; /* Offset of next open position. */
1095 #else /* not INT_IS_16BIT */
1097 # if defined MATCH_MAY_ALLOCATE
1098 /* 4400 was enough to cause a crash on Alpha OSF/1,
1099 whose default stack limit is 2mb. */
1100 int re_max_failures
= 20000;
1102 int re_max_failures
= 2000;
1105 union fail_stack_elt
1107 unsigned char *pointer
;
1111 typedef union fail_stack_elt fail_stack_elt_t
;
1115 fail_stack_elt_t
*stack
;
1117 unsigned avail
; /* Offset of next open position. */
1120 #endif /* INT_IS_16BIT */
1122 #define FAIL_STACK_EMPTY() (fail_stack.avail == 0)
1123 #define FAIL_STACK_PTR_EMPTY() (fail_stack_ptr->avail == 0)
1124 #define FAIL_STACK_FULL() (fail_stack.avail == fail_stack.size)
1127 /* Define macros to initialize and free the failure stack.
1128 Do `return -2' if the alloc fails. */
1130 #ifdef MATCH_MAY_ALLOCATE
1131 # define INIT_FAIL_STACK() \
1133 fail_stack.stack = (fail_stack_elt_t *) \
1134 REGEX_ALLOCATE_STACK (INIT_FAILURE_ALLOC * sizeof (fail_stack_elt_t)); \
1136 if (fail_stack.stack == NULL) \
1139 fail_stack.size = INIT_FAILURE_ALLOC; \
1140 fail_stack.avail = 0; \
1143 # define RESET_FAIL_STACK() REGEX_FREE_STACK (fail_stack.stack)
1145 # define INIT_FAIL_STACK() \
1147 fail_stack.avail = 0; \
1150 # define RESET_FAIL_STACK()
1154 /* Double the size of FAIL_STACK, up to approximately `re_max_failures' items.
1156 Return 1 if succeeds, and 0 if either ran out of memory
1157 allocating space for it or it was already too large.
1159 REGEX_REALLOCATE_STACK requires `destination' be declared. */
1161 #define DOUBLE_FAIL_STACK(fail_stack) \
1162 ((fail_stack).size > (unsigned) (re_max_failures * MAX_FAILURE_ITEMS) \
1164 : ((fail_stack).stack = (fail_stack_elt_t *) \
1165 REGEX_REALLOCATE_STACK ((fail_stack).stack, \
1166 (fail_stack).size * sizeof (fail_stack_elt_t), \
1167 ((fail_stack).size << 1) * sizeof (fail_stack_elt_t)), \
1169 (fail_stack).stack == NULL \
1171 : ((fail_stack).size <<= 1, \
1175 /* Push pointer POINTER on FAIL_STACK.
1176 Return 1 if was able to do so and 0 if ran out of memory allocating
1178 #define PUSH_PATTERN_OP(POINTER, FAIL_STACK) \
1179 ((FAIL_STACK_FULL () \
1180 && !DOUBLE_FAIL_STACK (FAIL_STACK)) \
1182 : ((FAIL_STACK).stack[(FAIL_STACK).avail++].pointer = POINTER, \
1185 /* Push a pointer value onto the failure stack.
1186 Assumes the variable `fail_stack'. Probably should only
1187 be called from within `PUSH_FAILURE_POINT'. */
1188 #define PUSH_FAILURE_POINTER(item) \
1189 fail_stack.stack[fail_stack.avail++].pointer = (unsigned char *) (item)
1191 /* This pushes an integer-valued item onto the failure stack.
1192 Assumes the variable `fail_stack'. Probably should only
1193 be called from within `PUSH_FAILURE_POINT'. */
1194 #define PUSH_FAILURE_INT(item) \
1195 fail_stack.stack[fail_stack.avail++].integer = (item)
1197 /* Push a fail_stack_elt_t value onto the failure stack.
1198 Assumes the variable `fail_stack'. Probably should only
1199 be called from within `PUSH_FAILURE_POINT'. */
1200 #define PUSH_FAILURE_ELT(item) \
1201 fail_stack.stack[fail_stack.avail++] = (item)
1203 /* These three POP... operations complement the three PUSH... operations.
1204 All assume that `fail_stack' is nonempty. */
1205 #define POP_FAILURE_POINTER() fail_stack.stack[--fail_stack.avail].pointer
1206 #define POP_FAILURE_INT() fail_stack.stack[--fail_stack.avail].integer
1207 #define POP_FAILURE_ELT() fail_stack.stack[--fail_stack.avail]
1209 /* Used to omit pushing failure point id's when we're not debugging. */
1211 # define DEBUG_PUSH PUSH_FAILURE_INT
1212 # define DEBUG_POP(item_addr) *(item_addr) = POP_FAILURE_INT ()
1214 # define DEBUG_PUSH(item)
1215 # define DEBUG_POP(item_addr)
1219 /* Push the information about the state we will need
1220 if we ever fail back to it.
1222 Requires variables fail_stack, regstart, regend, reg_info, and
1223 num_regs_pushed be declared. DOUBLE_FAIL_STACK requires `destination'
1226 Does `return FAILURE_CODE' if runs out of memory. */
1228 #define PUSH_FAILURE_POINT(pattern_place, string_place, failure_code) \
1230 char *destination; \
1231 /* Must be int, so when we don't save any registers, the arithmetic \
1232 of 0 + -1 isn't done as unsigned. */ \
1233 /* Can't be int, since there is not a shred of a guarantee that int \
1234 is wide enough to hold a value of something to which pointer can \
1236 active_reg_t this_reg; \
1238 DEBUG_STATEMENT (failure_id++); \
1239 DEBUG_STATEMENT (nfailure_points_pushed++); \
1240 DEBUG_PRINT2 ("\nPUSH_FAILURE_POINT #%u:\n", failure_id); \
1241 DEBUG_PRINT2 (" Before push, next avail: %d\n", (fail_stack).avail);\
1242 DEBUG_PRINT2 (" size: %d\n", (fail_stack).size);\
1244 DEBUG_PRINT2 (" slots needed: %ld\n", NUM_FAILURE_ITEMS); \
1245 DEBUG_PRINT2 (" available: %d\n", REMAINING_AVAIL_SLOTS); \
1247 /* Ensure we have enough space allocated for what we will push. */ \
1248 while (REMAINING_AVAIL_SLOTS < NUM_FAILURE_ITEMS) \
1250 if (!DOUBLE_FAIL_STACK (fail_stack)) \
1251 return failure_code; \
1253 DEBUG_PRINT2 ("\n Doubled stack; size now: %d\n", \
1254 (fail_stack).size); \
1255 DEBUG_PRINT2 (" slots available: %d\n", REMAINING_AVAIL_SLOTS);\
1258 /* Push the info, starting with the registers. */ \
1259 DEBUG_PRINT1 ("\n"); \
1262 for (this_reg = lowest_active_reg; this_reg <= highest_active_reg; \
1265 DEBUG_PRINT2 (" Pushing reg: %lu\n", this_reg); \
1266 DEBUG_STATEMENT (num_regs_pushed++); \
1268 DEBUG_PRINT2 (" start: %p\n", regstart[this_reg]); \
1269 PUSH_FAILURE_POINTER (regstart[this_reg]); \
1271 DEBUG_PRINT2 (" end: %p\n", regend[this_reg]); \
1272 PUSH_FAILURE_POINTER (regend[this_reg]); \
1274 DEBUG_PRINT2 (" info: %p\n ", \
1275 reg_info[this_reg].word.pointer); \
1276 DEBUG_PRINT2 (" match_null=%d", \
1277 REG_MATCH_NULL_STRING_P (reg_info[this_reg])); \
1278 DEBUG_PRINT2 (" active=%d", IS_ACTIVE (reg_info[this_reg])); \
1279 DEBUG_PRINT2 (" matched_something=%d", \
1280 MATCHED_SOMETHING (reg_info[this_reg])); \
1281 DEBUG_PRINT2 (" ever_matched=%d", \
1282 EVER_MATCHED_SOMETHING (reg_info[this_reg])); \
1283 DEBUG_PRINT1 ("\n"); \
1284 PUSH_FAILURE_ELT (reg_info[this_reg].word); \
1287 DEBUG_PRINT2 (" Pushing low active reg: %ld\n", lowest_active_reg);\
1288 PUSH_FAILURE_INT (lowest_active_reg); \
1290 DEBUG_PRINT2 (" Pushing high active reg: %ld\n", highest_active_reg);\
1291 PUSH_FAILURE_INT (highest_active_reg); \
1293 DEBUG_PRINT2 (" Pushing pattern %p:\n", pattern_place); \
1294 DEBUG_PRINT_COMPILED_PATTERN (bufp, pattern_place, pend); \
1295 PUSH_FAILURE_POINTER (pattern_place); \
1297 DEBUG_PRINT2 (" Pushing string %p: `", string_place); \
1298 DEBUG_PRINT_DOUBLE_STRING (string_place, string1, size1, string2, \
1300 DEBUG_PRINT1 ("'\n"); \
1301 PUSH_FAILURE_POINTER (string_place); \
1303 DEBUG_PRINT2 (" Pushing failure id: %u\n", failure_id); \
1304 DEBUG_PUSH (failure_id); \
1307 /* This is the number of items that are pushed and popped on the stack
1308 for each register. */
1309 #define NUM_REG_ITEMS 3
1311 /* Individual items aside from the registers. */
1313 # define NUM_NONREG_ITEMS 5 /* Includes failure point id. */
1315 # define NUM_NONREG_ITEMS 4
1318 /* We push at most this many items on the stack. */
1319 /* We used to use (num_regs - 1), which is the number of registers
1320 this regexp will save; but that was changed to 5
1321 to avoid stack overflow for a regexp with lots of parens. */
1322 #define MAX_FAILURE_ITEMS (5 * NUM_REG_ITEMS + NUM_NONREG_ITEMS)
1324 /* We actually push this many items. */
1325 #define NUM_FAILURE_ITEMS \
1327 ? 0 : highest_active_reg - lowest_active_reg + 1) \
1331 /* How many items can still be added to the stack without overflowing it. */
1332 #define REMAINING_AVAIL_SLOTS ((fail_stack).size - (fail_stack).avail)
1335 /* Pops what PUSH_FAIL_STACK pushes.
1337 We restore into the parameters, all of which should be lvalues:
1338 STR -- the saved data position.
1339 PAT -- the saved pattern position.
1340 LOW_REG, HIGH_REG -- the highest and lowest active registers.
1341 REGSTART, REGEND -- arrays of string positions.
1342 REG_INFO -- array of information about each subexpression.
1344 Also assumes the variables `fail_stack' and (if debugging), `bufp',
1345 `pend', `string1', `size1', `string2', and `size2'. */
1347 #define POP_FAILURE_POINT(str, pat, low_reg, high_reg, regstart, regend, reg_info)\
1349 DEBUG_STATEMENT (unsigned failure_id;) \
1350 active_reg_t this_reg; \
1351 const unsigned char *string_temp; \
1353 assert (!FAIL_STACK_EMPTY ()); \
1355 /* Remove failure points and point to how many regs pushed. */ \
1356 DEBUG_PRINT1 ("POP_FAILURE_POINT:\n"); \
1357 DEBUG_PRINT2 (" Before pop, next avail: %d\n", fail_stack.avail); \
1358 DEBUG_PRINT2 (" size: %d\n", fail_stack.size); \
1360 assert (fail_stack.avail >= NUM_NONREG_ITEMS); \
1362 DEBUG_POP (&failure_id); \
1363 DEBUG_PRINT2 (" Popping failure id: %u\n", failure_id); \
1365 /* If the saved string location is NULL, it came from an \
1366 on_failure_keep_string_jump opcode, and we want to throw away the \
1367 saved NULL, thus retaining our current position in the string. */ \
1368 string_temp = POP_FAILURE_POINTER (); \
1369 if (string_temp != NULL) \
1370 str = (const char *) string_temp; \
1372 DEBUG_PRINT2 (" Popping string %p: `", str); \
1373 DEBUG_PRINT_DOUBLE_STRING (str, string1, size1, string2, size2); \
1374 DEBUG_PRINT1 ("'\n"); \
1376 pat = (unsigned char *) POP_FAILURE_POINTER (); \
1377 DEBUG_PRINT2 (" Popping pattern %p:\n", pat); \
1378 DEBUG_PRINT_COMPILED_PATTERN (bufp, pat, pend); \
1380 /* Restore register info. */ \
1381 high_reg = (active_reg_t) POP_FAILURE_INT (); \
1382 DEBUG_PRINT2 (" Popping high active reg: %ld\n", high_reg); \
1384 low_reg = (active_reg_t) POP_FAILURE_INT (); \
1385 DEBUG_PRINT2 (" Popping low active reg: %ld\n", low_reg); \
1388 for (this_reg = high_reg; this_reg >= low_reg; this_reg--) \
1390 DEBUG_PRINT2 (" Popping reg: %ld\n", this_reg); \
1392 reg_info[this_reg].word = POP_FAILURE_ELT (); \
1393 DEBUG_PRINT2 (" info: %p\n", \
1394 reg_info[this_reg].word.pointer); \
1396 regend[this_reg] = (const char *) POP_FAILURE_POINTER (); \
1397 DEBUG_PRINT2 (" end: %p\n", regend[this_reg]); \
1399 regstart[this_reg] = (const char *) POP_FAILURE_POINTER (); \
1400 DEBUG_PRINT2 (" start: %p\n", regstart[this_reg]); \
1404 for (this_reg = highest_active_reg; this_reg > high_reg; this_reg--) \
1406 reg_info[this_reg].word.integer = 0; \
1407 regend[this_reg] = 0; \
1408 regstart[this_reg] = 0; \
1410 highest_active_reg = high_reg; \
1413 set_regs_matched_done = 0; \
1414 DEBUG_STATEMENT (nfailure_points_popped++); \
1415 } /* POP_FAILURE_POINT */
1419 /* Structure for per-register (a.k.a. per-group) information.
1420 Other register information, such as the
1421 starting and ending positions (which are addresses), and the list of
1422 inner groups (which is a bits list) are maintained in separate
1425 We are making a (strictly speaking) nonportable assumption here: that
1426 the compiler will pack our bit fields into something that fits into
1427 the type of `word', i.e., is something that fits into one item on the
1431 /* Declarations and macros for re_match_2. */
1435 fail_stack_elt_t word
;
1438 /* This field is one if this group can match the empty string,
1439 zero if not. If not yet determined, `MATCH_NULL_UNSET_VALUE'. */
1440 #define MATCH_NULL_UNSET_VALUE 3
1441 unsigned match_null_string_p
: 2;
1442 unsigned is_active
: 1;
1443 unsigned matched_something
: 1;
1444 unsigned ever_matched_something
: 1;
1446 } register_info_type
;
1448 #define REG_MATCH_NULL_STRING_P(R) ((R).bits.match_null_string_p)
1449 #define IS_ACTIVE(R) ((R).bits.is_active)
1450 #define MATCHED_SOMETHING(R) ((R).bits.matched_something)
1451 #define EVER_MATCHED_SOMETHING(R) ((R).bits.ever_matched_something)
1454 /* Call this when have matched a real character; it sets `matched' flags
1455 for the subexpressions which we are currently inside. Also records
1456 that those subexprs have matched. */
1457 #define SET_REGS_MATCHED() \
1460 if (!set_regs_matched_done) \
1463 set_regs_matched_done = 1; \
1464 for (r = lowest_active_reg; r <= highest_active_reg; r++) \
1466 MATCHED_SOMETHING (reg_info[r]) \
1467 = EVER_MATCHED_SOMETHING (reg_info[r]) \
1474 /* Registers are set to a sentinel when they haven't yet matched. */
1475 static char reg_unset_dummy
;
1476 #define REG_UNSET_VALUE (®_unset_dummy)
1477 #define REG_UNSET(e) ((e) == REG_UNSET_VALUE)
1479 /* Subroutine declarations and macros for regex_compile. */
1481 static reg_errcode_t regex_compile
_RE_ARGS ((const char *pattern
, size_t size
,
1482 reg_syntax_t syntax
,
1483 struct re_pattern_buffer
*bufp
));
1484 static void store_op1
_RE_ARGS ((re_opcode_t op
, unsigned char *loc
, int arg
));
1485 static void store_op2
_RE_ARGS ((re_opcode_t op
, unsigned char *loc
,
1486 int arg1
, int arg2
));
1487 static void insert_op1
_RE_ARGS ((re_opcode_t op
, unsigned char *loc
,
1488 int arg
, unsigned char *end
));
1489 static void insert_op2
_RE_ARGS ((re_opcode_t op
, unsigned char *loc
,
1490 int arg1
, int arg2
, unsigned char *end
));
1491 static boolean at_begline_loc_p
_RE_ARGS ((const char *pattern
, const char *p
,
1492 reg_syntax_t syntax
));
1493 static boolean at_endline_loc_p
_RE_ARGS ((const char *p
, const char *pend
,
1494 reg_syntax_t syntax
));
1495 static reg_errcode_t compile_range
_RE_ARGS ((const char **p_ptr
,
1498 reg_syntax_t syntax
,
1501 /* Fetch the next character in the uncompiled pattern---translating it
1502 if necessary. Also cast from a signed character in the constant
1503 string passed to us by the user to an unsigned char that we can use
1504 as an array index (in, e.g., `translate'). */
1506 # define PATFETCH(c) \
1507 do {if (p == pend) return REG_EEND; \
1508 c = (unsigned char) *p++; \
1509 if (translate) c = (unsigned char) translate[c]; \
1513 /* Fetch the next character in the uncompiled pattern, with no
1515 #define PATFETCH_RAW(c) \
1516 do {if (p == pend) return REG_EEND; \
1517 c = (unsigned char) *p++; \
1520 /* Go backwards one character in the pattern. */
1521 #define PATUNFETCH p--
1524 /* If `translate' is non-null, return translate[D], else just D. We
1525 cast the subscript to translate because some data is declared as
1526 `char *', to avoid warnings when a string constant is passed. But
1527 when we use a character as a subscript we must make it unsigned. */
1529 # define TRANSLATE(d) \
1530 (translate ? (char) translate[(unsigned char) (d)] : (d))
1534 /* Macros for outputting the compiled pattern into `buffer'. */
1536 /* If the buffer isn't allocated when it comes in, use this. */
1537 #define INIT_BUF_SIZE 32
1539 /* Make sure we have at least N more bytes of space in buffer. */
1540 #define GET_BUFFER_SPACE(n) \
1541 while ((unsigned long) (b - bufp->buffer + (n)) > bufp->allocated) \
1544 /* Make sure we have one more byte of buffer space and then add C to it. */
1545 #define BUF_PUSH(c) \
1547 GET_BUFFER_SPACE (1); \
1548 *b++ = (unsigned char) (c); \
1552 /* Ensure we have two more bytes of buffer space and then append C1 and C2. */
1553 #define BUF_PUSH_2(c1, c2) \
1555 GET_BUFFER_SPACE (2); \
1556 *b++ = (unsigned char) (c1); \
1557 *b++ = (unsigned char) (c2); \
1561 /* As with BUF_PUSH_2, except for three bytes. */
1562 #define BUF_PUSH_3(c1, c2, c3) \
1564 GET_BUFFER_SPACE (3); \
1565 *b++ = (unsigned char) (c1); \
1566 *b++ = (unsigned char) (c2); \
1567 *b++ = (unsigned char) (c3); \
1571 /* Store a jump with opcode OP at LOC to location TO. We store a
1572 relative address offset by the three bytes the jump itself occupies. */
1573 #define STORE_JUMP(op, loc, to) \
1574 store_op1 (op, loc, (int) ((to) - (loc) - 3))
1576 /* Likewise, for a two-argument jump. */
1577 #define STORE_JUMP2(op, loc, to, arg) \
1578 store_op2 (op, loc, (int) ((to) - (loc) - 3), arg)
1580 /* Like `STORE_JUMP', but for inserting. Assume `b' is the buffer end. */
1581 #define INSERT_JUMP(op, loc, to) \
1582 insert_op1 (op, loc, (int) ((to) - (loc) - 3), b)
1584 /* Like `STORE_JUMP2', but for inserting. Assume `b' is the buffer end. */
1585 #define INSERT_JUMP2(op, loc, to, arg) \
1586 insert_op2 (op, loc, (int) ((to) - (loc) - 3), arg, b)
1589 /* This is not an arbitrary limit: the arguments which represent offsets
1590 into the pattern are two bytes long. So if 2^16 bytes turns out to
1591 be too small, many things would have to change. */
1592 /* Any other compiler which, like MSC, has allocation limit below 2^16
1593 bytes will have to use approach similar to what was done below for
1594 MSC and drop MAX_BUF_SIZE a bit. Otherwise you may end up
1595 reallocating to 0 bytes. Such thing is not going to work too well.
1596 You have been warned!! */
1597 #if defined _MSC_VER && !defined WIN32
1598 /* Microsoft C 16-bit versions limit malloc to approx 65512 bytes.
1599 The REALLOC define eliminates a flurry of conversion warnings,
1600 but is not required. */
1601 # define MAX_BUF_SIZE 65500L
1602 # define REALLOC(p,s) realloc ((p), (size_t) (s))
1604 # define MAX_BUF_SIZE (1L << 16)
1605 # define REALLOC(p,s) realloc ((p), (s))
1608 /* Extend the buffer by twice its current size via realloc and
1609 reset the pointers that pointed into the old block to point to the
1610 correct places in the new one. If extending the buffer results in it
1611 being larger than MAX_BUF_SIZE, then flag memory exhausted. */
1612 #define EXTEND_BUFFER() \
1614 unsigned char *old_buffer = bufp->buffer; \
1615 if (bufp->allocated == MAX_BUF_SIZE) \
1617 bufp->allocated <<= 1; \
1618 if (bufp->allocated > MAX_BUF_SIZE) \
1619 bufp->allocated = MAX_BUF_SIZE; \
1620 bufp->buffer = (unsigned char *) REALLOC (bufp->buffer, bufp->allocated);\
1621 if (bufp->buffer == NULL) \
1622 return REG_ESPACE; \
1623 /* If the buffer moved, move all the pointers into it. */ \
1624 if (old_buffer != bufp->buffer) \
1626 b = (b - old_buffer) + bufp->buffer; \
1627 begalt = (begalt - old_buffer) + bufp->buffer; \
1628 if (fixup_alt_jump) \
1629 fixup_alt_jump = (fixup_alt_jump - old_buffer) + bufp->buffer;\
1631 laststart = (laststart - old_buffer) + bufp->buffer; \
1632 if (pending_exact) \
1633 pending_exact = (pending_exact - old_buffer) + bufp->buffer; \
1638 /* Since we have one byte reserved for the register number argument to
1639 {start,stop}_memory, the maximum number of groups we can report
1640 things about is what fits in that byte. */
1641 #define MAX_REGNUM 255
1643 /* But patterns can have more than `MAX_REGNUM' registers. We just
1644 ignore the excess. */
1645 typedef unsigned regnum_t
;
1648 /* Macros for the compile stack. */
1650 /* Since offsets can go either forwards or backwards, this type needs to
1651 be able to hold values from -(MAX_BUF_SIZE - 1) to MAX_BUF_SIZE - 1. */
1652 /* int may be not enough when sizeof(int) == 2. */
1653 typedef long pattern_offset_t
;
1657 pattern_offset_t begalt_offset
;
1658 pattern_offset_t fixup_alt_jump
;
1659 pattern_offset_t inner_group_offset
;
1660 pattern_offset_t laststart_offset
;
1662 } compile_stack_elt_t
;
1667 compile_stack_elt_t
*stack
;
1669 unsigned avail
; /* Offset of next open position. */
1670 } compile_stack_type
;
1673 #define INIT_COMPILE_STACK_SIZE 32
1675 #define COMPILE_STACK_EMPTY (compile_stack.avail == 0)
1676 #define COMPILE_STACK_FULL (compile_stack.avail == compile_stack.size)
1678 /* The next available element. */
1679 #define COMPILE_STACK_TOP (compile_stack.stack[compile_stack.avail])
1682 /* Set the bit for character C in a list. */
1683 #define SET_LIST_BIT(c) \
1684 (b[((unsigned char) (c)) / BYTEWIDTH] \
1685 |= 1 << (((unsigned char) c) % BYTEWIDTH))
1688 /* Get the next unsigned number in the uncompiled pattern. */
1689 #define GET_UNSIGNED_NUMBER(num) \
1693 while (ISDIGIT (c)) \
1697 num = num * 10 + c - '0'; \
1705 #if defined _LIBC || (defined HAVE_WCTYPE_H && defined HAVE_WCHAR_H)
1706 /* The GNU C library provides support for user-defined character classes
1707 and the functions from ISO C amendement 1. */
1708 # ifdef CHARCLASS_NAME_MAX
1709 # define CHAR_CLASS_MAX_LENGTH CHARCLASS_NAME_MAX
1711 /* This shouldn't happen but some implementation might still have this
1712 problem. Use a reasonable default value. */
1713 # define CHAR_CLASS_MAX_LENGTH 256
1717 # define IS_CHAR_CLASS(string) __wctype (string)
1719 # define IS_CHAR_CLASS(string) wctype (string)
1722 # define CHAR_CLASS_MAX_LENGTH 6 /* Namely, `xdigit'. */
1724 # define IS_CHAR_CLASS(string) \
1725 (STREQ (string, "alpha") || STREQ (string, "upper") \
1726 || STREQ (string, "lower") || STREQ (string, "digit") \
1727 || STREQ (string, "alnum") || STREQ (string, "xdigit") \
1728 || STREQ (string, "space") || STREQ (string, "print") \
1729 || STREQ (string, "punct") || STREQ (string, "graph") \
1730 || STREQ (string, "cntrl") || STREQ (string, "blank"))
1733 #ifndef MATCH_MAY_ALLOCATE
1735 /* If we cannot allocate large objects within re_match_2_internal,
1736 we make the fail stack and register vectors global.
1737 The fail stack, we grow to the maximum size when a regexp
1739 The register vectors, we adjust in size each time we
1740 compile a regexp, according to the number of registers it needs. */
1742 static fail_stack_type fail_stack
;
1744 /* Size with which the following vectors are currently allocated.
1745 That is so we can make them bigger as needed,
1746 but never make them smaller. */
1747 static int regs_allocated_size
;
1749 static const char ** regstart
, ** regend
;
1750 static const char ** old_regstart
, ** old_regend
;
1751 static const char **best_regstart
, **best_regend
;
1752 static register_info_type
*reg_info
;
1753 static const char **reg_dummy
;
1754 static register_info_type
*reg_info_dummy
;
1756 /* Make the register vectors big enough for NUM_REGS registers,
1757 but don't make them smaller. */
1760 regex_grow_registers (num_regs
)
1763 if (num_regs
> regs_allocated_size
)
1765 RETALLOC_IF (regstart
, num_regs
, const char *);
1766 RETALLOC_IF (regend
, num_regs
, const char *);
1767 RETALLOC_IF (old_regstart
, num_regs
, const char *);
1768 RETALLOC_IF (old_regend
, num_regs
, const char *);
1769 RETALLOC_IF (best_regstart
, num_regs
, const char *);
1770 RETALLOC_IF (best_regend
, num_regs
, const char *);
1771 RETALLOC_IF (reg_info
, num_regs
, register_info_type
);
1772 RETALLOC_IF (reg_dummy
, num_regs
, const char *);
1773 RETALLOC_IF (reg_info_dummy
, num_regs
, register_info_type
);
1775 regs_allocated_size
= num_regs
;
1779 #endif /* not MATCH_MAY_ALLOCATE */
1781 static boolean group_in_compile_stack
_RE_ARGS ((compile_stack_type
1785 /* `regex_compile' compiles PATTERN (of length SIZE) according to SYNTAX.
1786 Returns one of error codes defined in `gnu-regex.h', or zero for success.
1788 Assumes the `allocated' (and perhaps `buffer') and `translate'
1789 fields are set in BUFP on entry.
1791 If it succeeds, results are put in BUFP (if it returns an error, the
1792 contents of BUFP are undefined):
1793 `buffer' is the compiled pattern;
1794 `syntax' is set to SYNTAX;
1795 `used' is set to the length of the compiled pattern;
1796 `fastmap_accurate' is zero;
1797 `re_nsub' is the number of subexpressions in PATTERN;
1798 `not_bol' and `not_eol' are zero;
1800 The `fastmap' and `newline_anchor' fields are neither
1801 examined nor set. */
1803 /* Return, freeing storage we allocated. */
1804 #define FREE_STACK_RETURN(value) \
1805 return (free (compile_stack.stack), value)
1807 static reg_errcode_t
1808 regex_compile (pattern
, size
, syntax
, bufp
)
1809 const char *pattern
;
1811 reg_syntax_t syntax
;
1812 struct re_pattern_buffer
*bufp
;
1814 /* We fetch characters from PATTERN here. Even though PATTERN is
1815 `char *' (i.e., signed), we declare these variables as unsigned, so
1816 they can be reliably used as array indices. */
1817 register unsigned char c
, c1
;
1819 /* A random temporary spot in PATTERN. */
1822 /* Points to the end of the buffer, where we should append. */
1823 register unsigned char *b
;
1825 /* Keeps track of unclosed groups. */
1826 compile_stack_type compile_stack
;
1828 /* Points to the current (ending) position in the pattern. */
1829 const char *p
= pattern
;
1830 const char *pend
= pattern
+ size
;
1832 /* How to translate the characters in the pattern. */
1833 RE_TRANSLATE_TYPE translate
= bufp
->translate
;
1835 /* Address of the count-byte of the most recently inserted `exactn'
1836 command. This makes it possible to tell if a new exact-match
1837 character can be added to that command or if the character requires
1838 a new `exactn' command. */
1839 unsigned char *pending_exact
= 0;
1841 /* Address of start of the most recently finished expression.
1842 This tells, e.g., postfix * where to find the start of its
1843 operand. Reset at the beginning of groups and alternatives. */
1844 unsigned char *laststart
= 0;
1846 /* Address of beginning of regexp, or inside of last group. */
1847 unsigned char *begalt
;
1849 /* Place in the uncompiled pattern (i.e., the {) to
1850 which to go back if the interval is invalid. */
1851 const char *beg_interval
;
1853 /* Address of the place where a forward jump should go to the end of
1854 the containing expression. Each alternative of an `or' -- except the
1855 last -- ends with a forward jump of this sort. */
1856 unsigned char *fixup_alt_jump
= 0;
1858 /* Counts open-groups as they are encountered. Remembered for the
1859 matching close-group on the compile stack, so the same register
1860 number is put in the stop_memory as the start_memory. */
1861 regnum_t regnum
= 0;
1864 DEBUG_PRINT1 ("\nCompiling pattern: ");
1867 unsigned debug_count
;
1869 for (debug_count
= 0; debug_count
< size
; debug_count
++)
1870 putchar (pattern
[debug_count
]);
1875 /* Initialize the compile stack. */
1876 compile_stack
.stack
= TALLOC (INIT_COMPILE_STACK_SIZE
, compile_stack_elt_t
);
1877 if (compile_stack
.stack
== NULL
)
1880 compile_stack
.size
= INIT_COMPILE_STACK_SIZE
;
1881 compile_stack
.avail
= 0;
1883 /* Initialize the pattern buffer. */
1884 bufp
->syntax
= syntax
;
1885 bufp
->fastmap_accurate
= 0;
1886 bufp
->not_bol
= bufp
->not_eol
= 0;
1888 /* Set `used' to zero, so that if we return an error, the pattern
1889 printer (for debugging) will think there's no pattern. We reset it
1893 /* Always count groups, whether or not bufp->no_sub is set. */
1896 #if !defined emacs && !defined SYNTAX_TABLE
1897 /* Initialize the syntax table. */
1898 init_syntax_once ();
1901 if (bufp
->allocated
== 0)
1904 { /* If zero allocated, but buffer is non-null, try to realloc
1905 enough space. This loses if buffer's address is bogus, but
1906 that is the user's responsibility. */
1907 RETALLOC (bufp
->buffer
, INIT_BUF_SIZE
, unsigned char);
1910 { /* Caller did not allocate a buffer. Do it for them. */
1911 bufp
->buffer
= TALLOC (INIT_BUF_SIZE
, unsigned char);
1913 if (!bufp
->buffer
) FREE_STACK_RETURN (REG_ESPACE
);
1915 bufp
->allocated
= INIT_BUF_SIZE
;
1918 begalt
= b
= bufp
->buffer
;
1920 /* Loop through the uncompiled pattern until we're at the end. */
1929 if ( /* If at start of pattern, it's an operator. */
1931 /* If context independent, it's an operator. */
1932 || syntax
& RE_CONTEXT_INDEP_ANCHORS
1933 /* Otherwise, depends on what's come before. */
1934 || at_begline_loc_p (pattern
, p
, syntax
))
1944 if ( /* If at end of pattern, it's an operator. */
1946 /* If context independent, it's an operator. */
1947 || syntax
& RE_CONTEXT_INDEP_ANCHORS
1948 /* Otherwise, depends on what's next. */
1949 || at_endline_loc_p (p
, pend
, syntax
))
1959 if ((syntax
& RE_BK_PLUS_QM
)
1960 || (syntax
& RE_LIMITED_OPS
))
1964 /* If there is no previous pattern... */
1967 if (syntax
& RE_CONTEXT_INVALID_OPS
)
1968 FREE_STACK_RETURN (REG_BADRPT
);
1969 else if (!(syntax
& RE_CONTEXT_INDEP_OPS
))
1974 /* Are we optimizing this jump? */
1975 boolean keep_string_p
= false;
1977 /* 1 means zero (many) matches is allowed. */
1978 char zero_times_ok
= 0, many_times_ok
= 0;
1980 /* If there is a sequence of repetition chars, collapse it
1981 down to just one (the right one). We can't combine
1982 interval operators with these because of, e.g., `a{2}*',
1983 which should only match an even number of `a's. */
1987 zero_times_ok
|= c
!= '+';
1988 many_times_ok
|= c
!= '?';
1996 || (!(syntax
& RE_BK_PLUS_QM
) && (c
== '+' || c
== '?')))
1999 else if (syntax
& RE_BK_PLUS_QM
&& c
== '\\')
2001 if (p
== pend
) FREE_STACK_RETURN (REG_EESCAPE
);
2004 if (!(c1
== '+' || c1
== '?'))
2019 /* If we get here, we found another repeat character. */
2022 /* Star, etc. applied to an empty pattern is equivalent
2023 to an empty pattern. */
2027 /* Now we know whether or not zero matches is allowed
2028 and also whether or not two or more matches is allowed. */
2030 { /* More than one repetition is allowed, so put in at the
2031 end a backward relative jump from `b' to before the next
2032 jump we're going to put in below (which jumps from
2033 laststart to after this jump).
2035 But if we are at the `*' in the exact sequence `.*\n',
2036 insert an unconditional jump backwards to the .,
2037 instead of the beginning of the loop. This way we only
2038 push a failure point once, instead of every time
2039 through the loop. */
2040 assert (p
- 1 > pattern
);
2042 /* Allocate the space for the jump. */
2043 GET_BUFFER_SPACE (3);
2045 /* We know we are not at the first character of the pattern,
2046 because laststart was nonzero. And we've already
2047 incremented `p', by the way, to be the character after
2048 the `*'. Do we have to do something analogous here
2049 for null bytes, because of RE_DOT_NOT_NULL? */
2050 if (TRANSLATE (*(p
- 2)) == TRANSLATE ('.')
2052 && p
< pend
&& TRANSLATE (*p
) == TRANSLATE ('\n')
2053 && !(syntax
& RE_DOT_NEWLINE
))
2054 { /* We have .*\n. */
2055 STORE_JUMP (jump
, b
, laststart
);
2056 keep_string_p
= true;
2059 /* Anything else. */
2060 STORE_JUMP (maybe_pop_jump
, b
, laststart
- 3);
2062 /* We've added more stuff to the buffer. */
2066 /* On failure, jump from laststart to b + 3, which will be the
2067 end of the buffer after this jump is inserted. */
2068 GET_BUFFER_SPACE (3);
2069 INSERT_JUMP (keep_string_p
? on_failure_keep_string_jump
2077 /* At least one repetition is required, so insert a
2078 `dummy_failure_jump' before the initial
2079 `on_failure_jump' instruction of the loop. This
2080 effects a skip over that instruction the first time
2081 we hit that loop. */
2082 GET_BUFFER_SPACE (3);
2083 INSERT_JUMP (dummy_failure_jump
, laststart
, laststart
+ 6);
2098 boolean had_char_class
= false;
2100 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2102 /* Ensure that we have enough space to push a charset: the
2103 opcode, the length count, and the bitset; 34 bytes in all. */
2104 GET_BUFFER_SPACE (34);
2108 /* We test `*p == '^' twice, instead of using an if
2109 statement, so we only need one BUF_PUSH. */
2110 BUF_PUSH (*p
== '^' ? charset_not
: charset
);
2114 /* Remember the first position in the bracket expression. */
2117 /* Push the number of bytes in the bitmap. */
2118 BUF_PUSH ((1 << BYTEWIDTH
) / BYTEWIDTH
);
2120 /* Clear the whole map. */
2121 bzero (b
, (1 << BYTEWIDTH
) / BYTEWIDTH
);
2123 /* charset_not matches newline according to a syntax bit. */
2124 if ((re_opcode_t
) b
[-2] == charset_not
2125 && (syntax
& RE_HAT_LISTS_NOT_NEWLINE
))
2126 SET_LIST_BIT ('\n');
2128 /* Read in characters and ranges, setting map bits. */
2131 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2135 /* \ might escape characters inside [...] and [^...]. */
2136 if ((syntax
& RE_BACKSLASH_ESCAPE_IN_LISTS
) && c
== '\\')
2138 if (p
== pend
) FREE_STACK_RETURN (REG_EESCAPE
);
2145 /* Could be the end of the bracket expression. If it's
2146 not (i.e., when the bracket expression is `[]' so
2147 far), the ']' character bit gets set way below. */
2148 if (c
== ']' && p
!= p1
+ 1)
2151 /* Look ahead to see if it's a range when the last thing
2152 was a character class. */
2153 if (had_char_class
&& c
== '-' && *p
!= ']')
2154 FREE_STACK_RETURN (REG_ERANGE
);
2156 /* Look ahead to see if it's a range when the last thing
2157 was a character: if this is a hyphen not at the
2158 beginning or the end of a list, then it's the range
2161 && !(p
- 2 >= pattern
&& p
[-2] == '[')
2162 && !(p
- 3 >= pattern
&& p
[-3] == '[' && p
[-2] == '^')
2166 = compile_range (&p
, pend
, translate
, syntax
, b
);
2167 if (ret
!= REG_NOERROR
) FREE_STACK_RETURN (ret
);
2170 else if (p
[0] == '-' && p
[1] != ']')
2171 { /* This handles ranges made up of characters only. */
2174 /* Move past the `-'. */
2177 ret
= compile_range (&p
, pend
, translate
, syntax
, b
);
2178 if (ret
!= REG_NOERROR
) FREE_STACK_RETURN (ret
);
2181 /* See if we're at the beginning of a possible character
2184 else if (syntax
& RE_CHAR_CLASSES
&& c
== '[' && *p
== ':')
2185 { /* Leave room for the null. */
2186 char str
[CHAR_CLASS_MAX_LENGTH
+ 1];
2191 /* If pattern is `[[:'. */
2192 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2197 if ((c
== ':' && *p
== ']') || p
== pend
2198 || c1
== CHAR_CLASS_MAX_LENGTH
)
2204 /* If isn't a word bracketed by `[:' and `:]':
2205 undo the ending character, the letters, and leave
2206 the leading `:' and `[' (but set bits for them). */
2207 if (c
== ':' && *p
== ']')
2209 /* CYGNUS LOCAL: Skip this code if we don't have btowc(). btowc() is */
2210 /* defined in the 1994 Amendment 1 to ISO C and may not be present on */
2211 /* systems where we have wchar.h and wctype.h. */
2212 #if defined _LIBC || (defined HAVE_WCTYPE_H && defined HAVE_WCHAR_H && defined HAVE_BTOWC)
2213 boolean is_lower
= STREQ (str
, "lower");
2214 boolean is_upper
= STREQ (str
, "upper");
2218 wt
= IS_CHAR_CLASS (str
);
2220 FREE_STACK_RETURN (REG_ECTYPE
);
2222 /* Throw away the ] at the end of the character
2226 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2228 for (ch
= 0; ch
< 1 << BYTEWIDTH
; ++ch
)
2231 if (__iswctype (__btowc (ch
), wt
))
2234 if (iswctype (btowc (ch
), wt
))
2238 if (translate
&& (is_upper
|| is_lower
)
2239 && (ISUPPER (ch
) || ISLOWER (ch
)))
2243 had_char_class
= true;
2246 boolean is_alnum
= STREQ (str
, "alnum");
2247 boolean is_alpha
= STREQ (str
, "alpha");
2248 boolean is_blank
= STREQ (str
, "blank");
2249 boolean is_cntrl
= STREQ (str
, "cntrl");
2250 boolean is_digit
= STREQ (str
, "digit");
2251 boolean is_graph
= STREQ (str
, "graph");
2252 boolean is_lower
= STREQ (str
, "lower");
2253 boolean is_print
= STREQ (str
, "print");
2254 boolean is_punct
= STREQ (str
, "punct");
2255 boolean is_space
= STREQ (str
, "space");
2256 boolean is_upper
= STREQ (str
, "upper");
2257 boolean is_xdigit
= STREQ (str
, "xdigit");
2259 if (!IS_CHAR_CLASS (str
))
2260 FREE_STACK_RETURN (REG_ECTYPE
);
2262 /* Throw away the ] at the end of the character
2266 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2268 for (ch
= 0; ch
< 1 << BYTEWIDTH
; ch
++)
2270 /* This was split into 3 if's to
2271 avoid an arbitrary limit in some compiler. */
2272 if ( (is_alnum
&& ISALNUM (ch
))
2273 || (is_alpha
&& ISALPHA (ch
))
2274 || (is_blank
&& ISBLANK (ch
))
2275 || (is_cntrl
&& ISCNTRL (ch
)))
2277 if ( (is_digit
&& ISDIGIT (ch
))
2278 || (is_graph
&& ISGRAPH (ch
))
2279 || (is_lower
&& ISLOWER (ch
))
2280 || (is_print
&& ISPRINT (ch
)))
2282 if ( (is_punct
&& ISPUNCT (ch
))
2283 || (is_space
&& ISSPACE (ch
))
2284 || (is_upper
&& ISUPPER (ch
))
2285 || (is_xdigit
&& ISXDIGIT (ch
)))
2287 if ( translate
&& (is_upper
|| is_lower
)
2288 && (ISUPPER (ch
) || ISLOWER (ch
)))
2291 had_char_class
= true;
2292 #endif /* libc || wctype.h */
2301 had_char_class
= false;
2306 had_char_class
= false;
2311 /* Discard any (non)matching list bytes that are all 0 at the
2312 end of the map. Decrease the map-length byte too. */
2313 while ((int) b
[-1] > 0 && b
[b
[-1] - 1] == 0)
2321 if (syntax
& RE_NO_BK_PARENS
)
2328 if (syntax
& RE_NO_BK_PARENS
)
2335 if (syntax
& RE_NEWLINE_ALT
)
2342 if (syntax
& RE_NO_BK_VBAR
)
2349 if (syntax
& RE_INTERVALS
&& syntax
& RE_NO_BK_BRACES
)
2350 goto handle_interval
;
2356 if (p
== pend
) FREE_STACK_RETURN (REG_EESCAPE
);
2358 /* Do not translate the character after the \, so that we can
2359 distinguish, e.g., \B from \b, even if we normally would
2360 translate, e.g., B to b. */
2366 if (syntax
& RE_NO_BK_PARENS
)
2367 goto normal_backslash
;
2373 if (COMPILE_STACK_FULL
)
2375 RETALLOC (compile_stack
.stack
, compile_stack
.size
<< 1,
2376 compile_stack_elt_t
);
2377 if (compile_stack
.stack
== NULL
) return REG_ESPACE
;
2379 compile_stack
.size
<<= 1;
2382 /* These are the values to restore when we hit end of this
2383 group. They are all relative offsets, so that if the
2384 whole pattern moves because of realloc, they will still
2386 COMPILE_STACK_TOP
.begalt_offset
= begalt
- bufp
->buffer
;
2387 COMPILE_STACK_TOP
.fixup_alt_jump
2388 = fixup_alt_jump
? fixup_alt_jump
- bufp
->buffer
+ 1 : 0;
2389 COMPILE_STACK_TOP
.laststart_offset
= b
- bufp
->buffer
;
2390 COMPILE_STACK_TOP
.regnum
= regnum
;
2392 /* We will eventually replace the 0 with the number of
2393 groups inner to this one. But do not push a
2394 start_memory for groups beyond the last one we can
2395 represent in the compiled pattern. */
2396 if (regnum
<= MAX_REGNUM
)
2398 COMPILE_STACK_TOP
.inner_group_offset
= b
- bufp
->buffer
+ 2;
2399 BUF_PUSH_3 (start_memory
, regnum
, 0);
2402 compile_stack
.avail
++;
2407 /* If we've reached MAX_REGNUM groups, then this open
2408 won't actually generate any code, so we'll have to
2409 clear pending_exact explicitly. */
2415 if (syntax
& RE_NO_BK_PARENS
) goto normal_backslash
;
2417 if (COMPILE_STACK_EMPTY
)
2419 if (syntax
& RE_UNMATCHED_RIGHT_PAREN_ORD
)
2420 goto normal_backslash
;
2422 FREE_STACK_RETURN (REG_ERPAREN
);
2427 { /* Push a dummy failure point at the end of the
2428 alternative for a possible future
2429 `pop_failure_jump' to pop. See comments at
2430 `push_dummy_failure' in `re_match_2'. */
2431 BUF_PUSH (push_dummy_failure
);
2433 /* We allocated space for this jump when we assigned
2434 to `fixup_alt_jump', in the `handle_alt' case below. */
2435 STORE_JUMP (jump_past_alt
, fixup_alt_jump
, b
- 1);
2438 /* See similar code for backslashed left paren above. */
2439 if (COMPILE_STACK_EMPTY
)
2441 if (syntax
& RE_UNMATCHED_RIGHT_PAREN_ORD
)
2444 FREE_STACK_RETURN (REG_ERPAREN
);
2447 /* Since we just checked for an empty stack above, this
2448 ``can't happen''. */
2449 assert (compile_stack
.avail
!= 0);
2451 /* We don't just want to restore into `regnum', because
2452 later groups should continue to be numbered higher,
2453 as in `(ab)c(de)' -- the second group is #2. */
2454 regnum_t this_group_regnum
;
2456 compile_stack
.avail
--;
2457 begalt
= bufp
->buffer
+ COMPILE_STACK_TOP
.begalt_offset
;
2459 = COMPILE_STACK_TOP
.fixup_alt_jump
2460 ? bufp
->buffer
+ COMPILE_STACK_TOP
.fixup_alt_jump
- 1
2462 laststart
= bufp
->buffer
+ COMPILE_STACK_TOP
.laststart_offset
;
2463 this_group_regnum
= COMPILE_STACK_TOP
.regnum
;
2464 /* If we've reached MAX_REGNUM groups, then this open
2465 won't actually generate any code, so we'll have to
2466 clear pending_exact explicitly. */
2469 /* We're at the end of the group, so now we know how many
2470 groups were inside this one. */
2471 if (this_group_regnum
<= MAX_REGNUM
)
2473 unsigned char *inner_group_loc
2474 = bufp
->buffer
+ COMPILE_STACK_TOP
.inner_group_offset
;
2476 *inner_group_loc
= regnum
- this_group_regnum
;
2477 BUF_PUSH_3 (stop_memory
, this_group_regnum
,
2478 regnum
- this_group_regnum
);
2484 case '|': /* `\|'. */
2485 if (syntax
& RE_LIMITED_OPS
|| syntax
& RE_NO_BK_VBAR
)
2486 goto normal_backslash
;
2488 if (syntax
& RE_LIMITED_OPS
)
2491 /* Insert before the previous alternative a jump which
2492 jumps to this alternative if the former fails. */
2493 GET_BUFFER_SPACE (3);
2494 INSERT_JUMP (on_failure_jump
, begalt
, b
+ 6);
2498 /* The alternative before this one has a jump after it
2499 which gets executed if it gets matched. Adjust that
2500 jump so it will jump to this alternative's analogous
2501 jump (put in below, which in turn will jump to the next
2502 (if any) alternative's such jump, etc.). The last such
2503 jump jumps to the correct final destination. A picture:
2509 If we are at `b', then fixup_alt_jump right now points to a
2510 three-byte space after `a'. We'll put in the jump, set
2511 fixup_alt_jump to right after `b', and leave behind three
2512 bytes which we'll fill in when we get to after `c'. */
2515 STORE_JUMP (jump_past_alt
, fixup_alt_jump
, b
);
2517 /* Mark and leave space for a jump after this alternative,
2518 to be filled in later either by next alternative or
2519 when know we're at the end of a series of alternatives. */
2521 GET_BUFFER_SPACE (3);
2530 /* If \{ is a literal. */
2531 if (!(syntax
& RE_INTERVALS
)
2532 /* If we're at `\{' and it's not the open-interval
2534 || ((syntax
& RE_INTERVALS
) && (syntax
& RE_NO_BK_BRACES
))
2535 || (p
- 2 == pattern
&& p
== pend
))
2536 goto normal_backslash
;
2540 /* If got here, then the syntax allows intervals. */
2542 /* At least (most) this many matches must be made. */
2543 int lower_bound
= -1, upper_bound
= -1;
2545 beg_interval
= p
- 1;
2549 if (syntax
& RE_NO_BK_BRACES
)
2550 goto unfetch_interval
;
2552 FREE_STACK_RETURN (REG_EBRACE
);
2555 GET_UNSIGNED_NUMBER (lower_bound
);
2559 GET_UNSIGNED_NUMBER (upper_bound
);
2560 if (upper_bound
< 0) upper_bound
= RE_DUP_MAX
;
2563 /* Interval such as `{1}' => match exactly once. */
2564 upper_bound
= lower_bound
;
2566 if (lower_bound
< 0 || upper_bound
> RE_DUP_MAX
2567 || lower_bound
> upper_bound
)
2569 if (syntax
& RE_NO_BK_BRACES
)
2570 goto unfetch_interval
;
2572 FREE_STACK_RETURN (REG_BADBR
);
2575 if (!(syntax
& RE_NO_BK_BRACES
))
2577 if (c
!= '\\') FREE_STACK_RETURN (REG_EBRACE
);
2584 if (syntax
& RE_NO_BK_BRACES
)
2585 goto unfetch_interval
;
2587 FREE_STACK_RETURN (REG_BADBR
);
2590 /* We just parsed a valid interval. */
2592 /* If it's invalid to have no preceding re. */
2595 if (syntax
& RE_CONTEXT_INVALID_OPS
)
2596 FREE_STACK_RETURN (REG_BADRPT
);
2597 else if (syntax
& RE_CONTEXT_INDEP_OPS
)
2600 goto unfetch_interval
;
2603 /* If the upper bound is zero, don't want to succeed at
2604 all; jump from `laststart' to `b + 3', which will be
2605 the end of the buffer after we insert the jump. */
2606 if (upper_bound
== 0)
2608 GET_BUFFER_SPACE (3);
2609 INSERT_JUMP (jump
, laststart
, b
+ 3);
2613 /* Otherwise, we have a nontrivial interval. When
2614 we're all done, the pattern will look like:
2615 set_number_at <jump count> <upper bound>
2616 set_number_at <succeed_n count> <lower bound>
2617 succeed_n <after jump addr> <succeed_n count>
2619 jump_n <succeed_n addr> <jump count>
2620 (The upper bound and `jump_n' are omitted if
2621 `upper_bound' is 1, though.) */
2623 { /* If the upper bound is > 1, we need to insert
2624 more at the end of the loop. */
2625 unsigned nbytes
= 10 + (upper_bound
> 1) * 10;
2627 GET_BUFFER_SPACE (nbytes
);
2629 /* Initialize lower bound of the `succeed_n', even
2630 though it will be set during matching by its
2631 attendant `set_number_at' (inserted next),
2632 because `re_compile_fastmap' needs to know.
2633 Jump to the `jump_n' we might insert below. */
2634 INSERT_JUMP2 (succeed_n
, laststart
,
2635 b
+ 5 + (upper_bound
> 1) * 5,
2639 /* Code to initialize the lower bound. Insert
2640 before the `succeed_n'. The `5' is the last two
2641 bytes of this `set_number_at', plus 3 bytes of
2642 the following `succeed_n'. */
2643 insert_op2 (set_number_at
, laststart
, 5, lower_bound
, b
);
2646 if (upper_bound
> 1)
2647 { /* More than one repetition is allowed, so
2648 append a backward jump to the `succeed_n'
2649 that starts this interval.
2651 When we've reached this during matching,
2652 we'll have matched the interval once, so
2653 jump back only `upper_bound - 1' times. */
2654 STORE_JUMP2 (jump_n
, b
, laststart
+ 5,
2658 /* The location we want to set is the second
2659 parameter of the `jump_n'; that is `b-2' as
2660 an absolute address. `laststart' will be
2661 the `set_number_at' we're about to insert;
2662 `laststart+3' the number to set, the source
2663 for the relative address. But we are
2664 inserting into the middle of the pattern --
2665 so everything is getting moved up by 5.
2666 Conclusion: (b - 2) - (laststart + 3) + 5,
2667 i.e., b - laststart.
2669 We insert this at the beginning of the loop
2670 so that if we fail during matching, we'll
2671 reinitialize the bounds. */
2672 insert_op2 (set_number_at
, laststart
, b
- laststart
,
2673 upper_bound
- 1, b
);
2678 beg_interval
= NULL
;
2683 /* If an invalid interval, match the characters as literals. */
2684 assert (beg_interval
);
2686 beg_interval
= NULL
;
2688 /* normal_char and normal_backslash need `c'. */
2691 if (!(syntax
& RE_NO_BK_BRACES
))
2693 if (p
> pattern
&& p
[-1] == '\\')
2694 goto normal_backslash
;
2699 /* There is no way to specify the before_dot and after_dot
2700 operators. rms says this is ok. --karl */
2708 BUF_PUSH_2 (syntaxspec
, syntax_spec_code
[c
]);
2714 BUF_PUSH_2 (notsyntaxspec
, syntax_spec_code
[c
]);
2720 if (syntax
& RE_NO_GNU_OPS
)
2723 BUF_PUSH (wordchar
);
2728 if (syntax
& RE_NO_GNU_OPS
)
2731 BUF_PUSH (notwordchar
);
2736 if (syntax
& RE_NO_GNU_OPS
)
2742 if (syntax
& RE_NO_GNU_OPS
)
2748 if (syntax
& RE_NO_GNU_OPS
)
2750 BUF_PUSH (wordbound
);
2754 if (syntax
& RE_NO_GNU_OPS
)
2756 BUF_PUSH (notwordbound
);
2760 if (syntax
& RE_NO_GNU_OPS
)
2766 if (syntax
& RE_NO_GNU_OPS
)
2771 case '1': case '2': case '3': case '4': case '5':
2772 case '6': case '7': case '8': case '9':
2773 if (syntax
& RE_NO_BK_REFS
)
2779 FREE_STACK_RETURN (REG_ESUBREG
);
2781 /* Can't back reference to a subexpression if inside of it. */
2782 if (group_in_compile_stack (compile_stack
, (regnum_t
) c1
))
2786 BUF_PUSH_2 (duplicate
, c1
);
2792 if (syntax
& RE_BK_PLUS_QM
)
2795 goto normal_backslash
;
2799 /* You might think it would be useful for \ to mean
2800 not to translate; but if we don't translate it
2801 it will never match anything. */
2809 /* Expects the character in `c'. */
2811 /* If no exactn currently being built. */
2814 /* If last exactn not at current position. */
2815 || pending_exact
+ *pending_exact
+ 1 != b
2817 /* We have only one byte following the exactn for the count. */
2818 || *pending_exact
== (1 << BYTEWIDTH
) - 1
2820 /* If followed by a repetition operator. */
2821 || *p
== '*' || *p
== '^'
2822 || ((syntax
& RE_BK_PLUS_QM
)
2823 ? *p
== '\\' && (p
[1] == '+' || p
[1] == '?')
2824 : (*p
== '+' || *p
== '?'))
2825 || ((syntax
& RE_INTERVALS
)
2826 && ((syntax
& RE_NO_BK_BRACES
)
2828 : (p
[0] == '\\' && p
[1] == '{'))))
2830 /* Start building a new exactn. */
2834 BUF_PUSH_2 (exactn
, 0);
2835 pending_exact
= b
- 1;
2842 } /* while p != pend */
2845 /* Through the pattern now. */
2848 STORE_JUMP (jump_past_alt
, fixup_alt_jump
, b
);
2850 if (!COMPILE_STACK_EMPTY
)
2851 FREE_STACK_RETURN (REG_EPAREN
);
2853 /* If we don't want backtracking, force success
2854 the first time we reach the end of the compiled pattern. */
2855 if (syntax
& RE_NO_POSIX_BACKTRACKING
)
2858 free (compile_stack
.stack
);
2860 /* We have succeeded; set the length of the buffer. */
2861 bufp
->used
= b
- bufp
->buffer
;
2866 DEBUG_PRINT1 ("\nCompiled pattern: \n");
2867 print_compiled_pattern (bufp
);
2871 #ifndef MATCH_MAY_ALLOCATE
2872 /* Initialize the failure stack to the largest possible stack. This
2873 isn't necessary unless we're trying to avoid calling alloca in
2874 the search and match routines. */
2876 int num_regs
= bufp
->re_nsub
+ 1;
2878 /* Since DOUBLE_FAIL_STACK refuses to double only if the current size
2879 is strictly greater than re_max_failures, the largest possible stack
2880 is 2 * re_max_failures failure points. */
2881 if (fail_stack
.size
< (2 * re_max_failures
* MAX_FAILURE_ITEMS
))
2883 fail_stack
.size
= (2 * re_max_failures
* MAX_FAILURE_ITEMS
);
2886 if (! fail_stack
.stack
)
2888 = (fail_stack_elt_t
*) xmalloc (fail_stack
.size
2889 * sizeof (fail_stack_elt_t
));
2892 = (fail_stack_elt_t
*) xrealloc (fail_stack
.stack
,
2894 * sizeof (fail_stack_elt_t
)));
2895 # else /* not emacs */
2896 if (! fail_stack
.stack
)
2898 = (fail_stack_elt_t
*) malloc (fail_stack
.size
2899 * sizeof (fail_stack_elt_t
));
2902 = (fail_stack_elt_t
*) realloc (fail_stack
.stack
,
2904 * sizeof (fail_stack_elt_t
)));
2905 # endif /* not emacs */
2908 regex_grow_registers (num_regs
);
2910 #endif /* not MATCH_MAY_ALLOCATE */
2913 } /* regex_compile */
2915 /* Subroutines for `regex_compile'. */
2917 /* Store OP at LOC followed by two-byte integer parameter ARG. */
2920 store_op1 (op
, loc
, arg
)
2925 *loc
= (unsigned char) op
;
2926 STORE_NUMBER (loc
+ 1, arg
);
2930 /* Like `store_op1', but for two two-byte parameters ARG1 and ARG2. */
2933 store_op2 (op
, loc
, arg1
, arg2
)
2938 *loc
= (unsigned char) op
;
2939 STORE_NUMBER (loc
+ 1, arg1
);
2940 STORE_NUMBER (loc
+ 3, arg2
);
2944 /* Copy the bytes from LOC to END to open up three bytes of space at LOC
2945 for OP followed by two-byte integer parameter ARG. */
2948 insert_op1 (op
, loc
, arg
, end
)
2954 register unsigned char *pfrom
= end
;
2955 register unsigned char *pto
= end
+ 3;
2957 while (pfrom
!= loc
)
2960 store_op1 (op
, loc
, arg
);
2964 /* Like `insert_op1', but for two two-byte parameters ARG1 and ARG2. */
2967 insert_op2 (op
, loc
, arg1
, arg2
, end
)
2973 register unsigned char *pfrom
= end
;
2974 register unsigned char *pto
= end
+ 5;
2976 while (pfrom
!= loc
)
2979 store_op2 (op
, loc
, arg1
, arg2
);
2983 /* P points to just after a ^ in PATTERN. Return true if that ^ comes
2984 after an alternative or a begin-subexpression. We assume there is at
2985 least one character before the ^. */
2988 at_begline_loc_p (pattern
, p
, syntax
)
2989 const char *pattern
, *p
;
2990 reg_syntax_t syntax
;
2992 const char *prev
= p
- 2;
2993 boolean prev_prev_backslash
= prev
> pattern
&& prev
[-1] == '\\';
2996 /* After a subexpression? */
2997 (*prev
== '(' && (syntax
& RE_NO_BK_PARENS
|| prev_prev_backslash
))
2998 /* After an alternative? */
2999 || (*prev
== '|' && (syntax
& RE_NO_BK_VBAR
|| prev_prev_backslash
));
3003 /* The dual of at_begline_loc_p. This one is for $. We assume there is
3004 at least one character after the $, i.e., `P < PEND'. */
3007 at_endline_loc_p (p
, pend
, syntax
)
3008 const char *p
, *pend
;
3009 reg_syntax_t syntax
;
3011 const char *next
= p
;
3012 boolean next_backslash
= *next
== '\\';
3013 const char *next_next
= p
+ 1 < pend
? p
+ 1 : 0;
3016 /* Before a subexpression? */
3017 (syntax
& RE_NO_BK_PARENS
? *next
== ')'
3018 : next_backslash
&& next_next
&& *next_next
== ')')
3019 /* Before an alternative? */
3020 || (syntax
& RE_NO_BK_VBAR
? *next
== '|'
3021 : next_backslash
&& next_next
&& *next_next
== '|');
3025 /* Returns true if REGNUM is in one of COMPILE_STACK's elements and
3026 false if it's not. */
3029 group_in_compile_stack (compile_stack
, regnum
)
3030 compile_stack_type compile_stack
;
3035 for (this_element
= compile_stack
.avail
- 1;
3038 if (compile_stack
.stack
[this_element
].regnum
== regnum
)
3045 /* Read the ending character of a range (in a bracket expression) from the
3046 uncompiled pattern *P_PTR (which ends at PEND). We assume the
3047 starting character is in `P[-2]'. (`P[-1]' is the character `-'.)
3048 Then we set the translation of all bits between the starting and
3049 ending characters (inclusive) in the compiled pattern B.
3051 Return an error code.
3053 We use these short variable names so we can use the same macros as
3054 `regex_compile' itself. */
3056 static reg_errcode_t
3057 compile_range (p_ptr
, pend
, translate
, syntax
, b
)
3058 const char **p_ptr
, *pend
;
3059 RE_TRANSLATE_TYPE translate
;
3060 reg_syntax_t syntax
;
3065 const char *p
= *p_ptr
;
3066 unsigned int range_start
, range_end
;
3071 /* Even though the pattern is a signed `char *', we need to fetch
3072 with unsigned char *'s; if the high bit of the pattern character
3073 is set, the range endpoints will be negative if we fetch using a
3076 We also want to fetch the endpoints without translating them; the
3077 appropriate translation is done in the bit-setting loop below. */
3078 /* The SVR4 compiler on the 3B2 had trouble with unsigned const char *. */
3079 range_start
= ((const unsigned char *) p
)[-2];
3080 range_end
= ((const unsigned char *) p
)[0];
3082 /* Have to increment the pointer into the pattern string, so the
3083 caller isn't still at the ending character. */
3086 /* If the start is after the end, the range is empty. */
3087 if (range_start
> range_end
)
3088 return syntax
& RE_NO_EMPTY_RANGES
? REG_ERANGE
: REG_NOERROR
;
3090 /* Here we see why `this_char' has to be larger than an `unsigned
3091 char' -- the range is inclusive, so if `range_end' == 0xff
3092 (assuming 8-bit characters), we would otherwise go into an infinite
3093 loop, since all characters <= 0xff. */
3094 for (this_char
= range_start
; this_char
<= range_end
; this_char
++)
3096 SET_LIST_BIT (TRANSLATE (this_char
));
3102 /* re_compile_fastmap computes a ``fastmap'' for the compiled pattern in
3103 BUFP. A fastmap records which of the (1 << BYTEWIDTH) possible
3104 characters can start a string that matches the pattern. This fastmap
3105 is used by re_search to skip quickly over impossible starting points.
3107 The caller must supply the address of a (1 << BYTEWIDTH)-byte data
3108 area as BUFP->fastmap.
3110 We set the `fastmap', `fastmap_accurate', and `can_be_null' fields in
3113 Returns 0 if we succeed, -2 if an internal error. */
3116 re_compile_fastmap (bufp
)
3117 struct re_pattern_buffer
*bufp
;
3120 #ifdef MATCH_MAY_ALLOCATE
3121 fail_stack_type fail_stack
;
3123 #ifndef REGEX_MALLOC
3127 register char *fastmap
= bufp
->fastmap
;
3128 unsigned char *pattern
= bufp
->buffer
;
3129 unsigned char *p
= pattern
;
3130 register unsigned char *pend
= pattern
+ bufp
->used
;
3133 /* This holds the pointer to the failure stack, when
3134 it is allocated relocatably. */
3135 fail_stack_elt_t
*failure_stack_ptr
;
3138 /* Assume that each path through the pattern can be null until
3139 proven otherwise. We set this false at the bottom of switch
3140 statement, to which we get only if a particular path doesn't
3141 match the empty string. */
3142 boolean path_can_be_null
= true;
3144 /* We aren't doing a `succeed_n' to begin with. */
3145 boolean succeed_n_p
= false;
3147 assert (fastmap
!= NULL
&& p
!= NULL
);
3150 bzero (fastmap
, 1 << BYTEWIDTH
); /* Assume nothing's valid. */
3151 bufp
->fastmap_accurate
= 1; /* It will be when we're done. */
3152 bufp
->can_be_null
= 0;
3156 if (p
== pend
|| *p
== succeed
)
3158 /* We have reached the (effective) end of pattern. */
3159 if (!FAIL_STACK_EMPTY ())
3161 bufp
->can_be_null
|= path_can_be_null
;
3163 /* Reset for next path. */
3164 path_can_be_null
= true;
3166 p
= fail_stack
.stack
[--fail_stack
.avail
].pointer
;
3174 /* We should never be about to go beyond the end of the pattern. */
3177 switch (SWITCH_ENUM_CAST ((re_opcode_t
) *p
++))
3180 /* I guess the idea here is to simply not bother with a fastmap
3181 if a backreference is used, since it's too hard to figure out
3182 the fastmap for the corresponding group. Setting
3183 `can_be_null' stops `re_search_2' from using the fastmap, so
3184 that is all we do. */
3186 bufp
->can_be_null
= 1;
3190 /* Following are the cases which match a character. These end
3199 for (j
= *p
++ * BYTEWIDTH
- 1; j
>= 0; j
--)
3200 if (p
[j
/ BYTEWIDTH
] & (1 << (j
% BYTEWIDTH
)))
3206 /* Chars beyond end of map must be allowed. */
3207 for (j
= *p
* BYTEWIDTH
; j
< (1 << BYTEWIDTH
); j
++)
3210 for (j
= *p
++ * BYTEWIDTH
- 1; j
>= 0; j
--)
3211 if (!(p
[j
/ BYTEWIDTH
] & (1 << (j
% BYTEWIDTH
))))
3217 for (j
= 0; j
< (1 << BYTEWIDTH
); j
++)
3218 if (SYNTAX (j
) == Sword
)
3224 for (j
= 0; j
< (1 << BYTEWIDTH
); j
++)
3225 if (SYNTAX (j
) != Sword
)
3232 int fastmap_newline
= fastmap
['\n'];
3234 /* `.' matches anything ... */
3235 for (j
= 0; j
< (1 << BYTEWIDTH
); j
++)
3238 /* ... except perhaps newline. */
3239 if (!(bufp
->syntax
& RE_DOT_NEWLINE
))
3240 fastmap
['\n'] = fastmap_newline
;
3242 /* Return if we have already set `can_be_null'; if we have,
3243 then the fastmap is irrelevant. Something's wrong here. */
3244 else if (bufp
->can_be_null
)
3247 /* Otherwise, have to check alternative paths. */
3254 for (j
= 0; j
< (1 << BYTEWIDTH
); j
++)
3255 if (SYNTAX (j
) == (enum syntaxcode
) k
)
3262 for (j
= 0; j
< (1 << BYTEWIDTH
); j
++)
3263 if (SYNTAX (j
) != (enum syntaxcode
) k
)
3268 /* All cases after this match the empty string. These end with
3288 case push_dummy_failure
:
3293 case pop_failure_jump
:
3294 case maybe_pop_jump
:
3297 case dummy_failure_jump
:
3298 EXTRACT_NUMBER_AND_INCR (j
, p
);
3303 /* Jump backward implies we just went through the body of a
3304 loop and matched nothing. Opcode jumped to should be
3305 `on_failure_jump' or `succeed_n'. Just treat it like an
3306 ordinary jump. For a * loop, it has pushed its failure
3307 point already; if so, discard that as redundant. */
3308 if ((re_opcode_t
) *p
!= on_failure_jump
3309 && (re_opcode_t
) *p
!= succeed_n
)
3313 EXTRACT_NUMBER_AND_INCR (j
, p
);
3316 /* If what's on the stack is where we are now, pop it. */
3317 if (!FAIL_STACK_EMPTY ()
3318 && fail_stack
.stack
[fail_stack
.avail
- 1].pointer
== p
)
3324 case on_failure_jump
:
3325 case on_failure_keep_string_jump
:
3326 handle_on_failure_jump
:
3327 EXTRACT_NUMBER_AND_INCR (j
, p
);
3329 /* For some patterns, e.g., `(a?)?', `p+j' here points to the
3330 end of the pattern. We don't want to push such a point,
3331 since when we restore it above, entering the switch will
3332 increment `p' past the end of the pattern. We don't need
3333 to push such a point since we obviously won't find any more
3334 fastmap entries beyond `pend'. Such a pattern can match
3335 the null string, though. */
3338 if (!PUSH_PATTERN_OP (p
+ j
, fail_stack
))
3340 RESET_FAIL_STACK ();
3345 bufp
->can_be_null
= 1;
3349 EXTRACT_NUMBER_AND_INCR (k
, p
); /* Skip the n. */
3350 succeed_n_p
= false;
3357 /* Get to the number of times to succeed. */
3360 /* Increment p past the n for when k != 0. */
3361 EXTRACT_NUMBER_AND_INCR (k
, p
);
3365 succeed_n_p
= true; /* Spaghetti code alert. */
3366 goto handle_on_failure_jump
;
3383 abort (); /* We have listed all the cases. */
3386 /* Getting here means we have found the possible starting
3387 characters for one path of the pattern -- and that the empty
3388 string does not match. We need not follow this path further.
3389 Instead, look at the next alternative (remembered on the
3390 stack), or quit if no more. The test at the top of the loop
3391 does these things. */
3392 path_can_be_null
= false;
3396 /* Set `can_be_null' for the last path (also the first path, if the
3397 pattern is empty). */
3398 bufp
->can_be_null
|= path_can_be_null
;
3401 RESET_FAIL_STACK ();
3403 } /* re_compile_fastmap */
3405 weak_alias (__re_compile_fastmap
, re_compile_fastmap
)
3408 /* Set REGS to hold NUM_REGS registers, storing them in STARTS and
3409 ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use
3410 this memory for recording register information. STARTS and ENDS
3411 must be allocated using the malloc library routine, and must each
3412 be at least NUM_REGS * sizeof (regoff_t) bytes long.
3414 If NUM_REGS == 0, then subsequent matches should allocate their own
3417 Unless this function is called, the first search or match using
3418 PATTERN_BUFFER will allocate its own register data, without
3419 freeing the old data. */
3422 re_set_registers (bufp
, regs
, num_regs
, starts
, ends
)
3423 struct re_pattern_buffer
*bufp
;
3424 struct re_registers
*regs
;
3426 regoff_t
*starts
, *ends
;
3430 bufp
->regs_allocated
= REGS_REALLOCATE
;
3431 regs
->num_regs
= num_regs
;
3432 regs
->start
= starts
;
3437 bufp
->regs_allocated
= REGS_UNALLOCATED
;
3439 regs
->start
= regs
->end
= (regoff_t
*) 0;
3443 weak_alias (__re_set_registers
, re_set_registers
)
3446 /* Searching routines. */
3448 /* Like re_search_2, below, but only one string is specified, and
3449 doesn't let you say where to stop matching. */
3452 re_search (bufp
, string
, size
, startpos
, range
, regs
)
3453 struct re_pattern_buffer
*bufp
;
3455 int size
, startpos
, range
;
3456 struct re_registers
*regs
;
3458 return re_search_2 (bufp
, NULL
, 0, string
, size
, startpos
, range
,
3462 weak_alias (__re_search
, re_search
)
3466 /* Using the compiled pattern in BUFP->buffer, first tries to match the
3467 virtual concatenation of STRING1 and STRING2, starting first at index
3468 STARTPOS, then at STARTPOS + 1, and so on.
3470 STRING1 and STRING2 have length SIZE1 and SIZE2, respectively.
3472 RANGE is how far to scan while trying to match. RANGE = 0 means try
3473 only at STARTPOS; in general, the last start tried is STARTPOS +
3476 In REGS, return the indices of the virtual concatenation of STRING1
3477 and STRING2 that matched the entire BUFP->buffer and its contained
3480 Do not consider matching one past the index STOP in the virtual
3481 concatenation of STRING1 and STRING2.
3483 We return either the position in the strings at which the match was
3484 found, -1 if no match, or -2 if error (such as failure
3488 re_search_2 (bufp
, string1
, size1
, string2
, size2
, startpos
, range
, regs
, stop
)
3489 struct re_pattern_buffer
*bufp
;
3490 const char *string1
, *string2
;
3494 struct re_registers
*regs
;
3498 register char *fastmap
= bufp
->fastmap
;
3499 register RE_TRANSLATE_TYPE translate
= bufp
->translate
;
3500 int total_size
= size1
+ size2
;
3501 int endpos
= startpos
+ range
;
3503 /* Check for out-of-range STARTPOS. */
3504 if (startpos
< 0 || startpos
> total_size
)
3507 /* Fix up RANGE if it might eventually take us outside
3508 the virtual concatenation of STRING1 and STRING2.
3509 Make sure we won't move STARTPOS below 0 or above TOTAL_SIZE. */
3511 range
= 0 - startpos
;
3512 else if (endpos
> total_size
)
3513 range
= total_size
- startpos
;
3515 /* If the search isn't to be a backwards one, don't waste time in a
3516 search for a pattern that must be anchored. */
3517 if (bufp
->used
> 0 && (re_opcode_t
) bufp
->buffer
[0] == begbuf
&& range
> 0)
3526 /* In a forward search for something that starts with \=.
3527 don't keep searching past point. */
3528 if (bufp
->used
> 0 && (re_opcode_t
) bufp
->buffer
[0] == at_dot
&& range
> 0)
3530 range
= PT
- startpos
;
3536 /* Update the fastmap now if not correct already. */
3537 if (fastmap
&& !bufp
->fastmap_accurate
)
3538 if (re_compile_fastmap (bufp
) == -2)
3541 /* Loop through the string, looking for a place to start matching. */
3544 /* If a fastmap is supplied, skip quickly over characters that
3545 cannot be the start of a match. If the pattern can match the
3546 null string, however, we don't need to skip characters; we want
3547 the first null string. */
3548 if (fastmap
&& startpos
< total_size
&& !bufp
->can_be_null
)
3550 if (range
> 0) /* Searching forwards. */
3552 register const char *d
;
3553 register int lim
= 0;
3556 if (startpos
< size1
&& startpos
+ range
>= size1
)
3557 lim
= range
- (size1
- startpos
);
3559 d
= (startpos
>= size1
? string2
- size1
: string1
) + startpos
;
3561 /* Written out as an if-else to avoid testing `translate'
3565 && !fastmap
[(unsigned char)
3566 translate
[(unsigned char) *d
++]])
3569 while (range
> lim
&& !fastmap
[(unsigned char) *d
++])
3572 startpos
+= irange
- range
;
3574 else /* Searching backwards. */
3576 register char c
= (size1
== 0 || startpos
>= size1
3577 ? string2
[startpos
- size1
]
3578 : string1
[startpos
]);
3580 if (!fastmap
[(unsigned char) TRANSLATE (c
)])
3585 /* If can't match the null string, and that's all we have left, fail. */
3586 if (range
>= 0 && startpos
== total_size
&& fastmap
3587 && !bufp
->can_be_null
)
3590 val
= re_match_2_internal (bufp
, string1
, size1
, string2
, size2
,
3591 startpos
, regs
, stop
);
3592 #ifndef REGEX_MALLOC
3621 weak_alias (__re_search_2
, re_search_2
)
3624 /* This converts PTR, a pointer into one of the search strings `string1'
3625 and `string2' into an offset from the beginning of that string. */
3626 #define POINTER_TO_OFFSET(ptr) \
3627 (FIRST_STRING_P (ptr) \
3628 ? ((regoff_t) ((ptr) - string1)) \
3629 : ((regoff_t) ((ptr) - string2 + size1)))
3631 /* Macros for dealing with the split strings in re_match_2. */
3633 #define MATCHING_IN_FIRST_STRING (dend == end_match_1)
3635 /* Call before fetching a character with *d. This switches over to
3636 string2 if necessary. */
3637 #define PREFETCH() \
3640 /* End of string2 => fail. */ \
3641 if (dend == end_match_2) \
3643 /* End of string1 => advance to string2. */ \
3645 dend = end_match_2; \
3649 /* Test if at very beginning or at very end of the virtual concatenation
3650 of `string1' and `string2'. If only one string, it's `string2'. */
3651 #define AT_STRINGS_BEG(d) ((d) == (size1 ? string1 : string2) || !size2)
3652 #define AT_STRINGS_END(d) ((d) == end2)
3655 /* Test if D points to a character which is word-constituent. We have
3656 two special cases to check for: if past the end of string1, look at
3657 the first character in string2; and if before the beginning of
3658 string2, look at the last character in string1. */
3659 #define WORDCHAR_P(d) \
3660 (SYNTAX ((d) == end1 ? *string2 \
3661 : (d) == string2 - 1 ? *(end1 - 1) : *(d)) \
3664 /* Disabled due to a compiler bug -- see comment at case wordbound */
3666 /* Test if the character before D and the one at D differ with respect
3667 to being word-constituent. */
3668 #define AT_WORD_BOUNDARY(d) \
3669 (AT_STRINGS_BEG (d) || AT_STRINGS_END (d) \
3670 || WORDCHAR_P (d - 1) != WORDCHAR_P (d))
3673 /* Free everything we malloc. */
3674 #ifdef MATCH_MAY_ALLOCATE
3675 # define FREE_VAR(var) if (var) REGEX_FREE (var); var = NULL
3676 # define FREE_VARIABLES() \
3678 REGEX_FREE_STACK (fail_stack.stack); \
3679 FREE_VAR (regstart); \
3680 FREE_VAR (regend); \
3681 FREE_VAR (old_regstart); \
3682 FREE_VAR (old_regend); \
3683 FREE_VAR (best_regstart); \
3684 FREE_VAR (best_regend); \
3685 FREE_VAR (reg_info); \
3686 FREE_VAR (reg_dummy); \
3687 FREE_VAR (reg_info_dummy); \
3690 # define FREE_VARIABLES() ((void)0) /* Do nothing! But inhibit gcc warning. */
3691 #endif /* not MATCH_MAY_ALLOCATE */
3693 /* These values must meet several constraints. They must not be valid
3694 register values; since we have a limit of 255 registers (because
3695 we use only one byte in the pattern for the register number), we can
3696 use numbers larger than 255. They must differ by 1, because of
3697 NUM_FAILURE_ITEMS above. And the value for the lowest register must
3698 be larger than the value for the highest register, so we do not try
3699 to actually save any registers when none are active. */
3700 #define NO_HIGHEST_ACTIVE_REG (1 << BYTEWIDTH)
3701 #define NO_LOWEST_ACTIVE_REG (NO_HIGHEST_ACTIVE_REG + 1)
3703 /* Matching routines. */
3705 #ifndef emacs /* Emacs never uses this. */
3706 /* re_match is like re_match_2 except it takes only a single string. */
3709 re_match (bufp
, string
, size
, pos
, regs
)
3710 struct re_pattern_buffer
*bufp
;
3713 struct re_registers
*regs
;
3715 int result
= re_match_2_internal (bufp
, NULL
, 0, string
, size
,
3717 # ifndef REGEX_MALLOC
3725 weak_alias (__re_match
, re_match
)
3727 #endif /* not emacs */
3729 static boolean group_match_null_string_p
_RE_ARGS ((unsigned char **p
,
3731 register_info_type
*reg_info
));
3732 static boolean alt_match_null_string_p
_RE_ARGS ((unsigned char *p
,
3734 register_info_type
*reg_info
));
3735 static boolean common_op_match_null_string_p
_RE_ARGS ((unsigned char **p
,
3737 register_info_type
*reg_info
));
3738 static int bcmp_translate
_RE_ARGS ((const char *s1
, const char *s2
,
3739 int len
, char *translate
));
3741 /* re_match_2 matches the compiled pattern in BUFP against the
3742 the (virtual) concatenation of STRING1 and STRING2 (of length SIZE1
3743 and SIZE2, respectively). We start matching at POS, and stop
3746 If REGS is non-null and the `no_sub' field of BUFP is nonzero, we
3747 store offsets for the substring each group matched in REGS. See the
3748 documentation for exactly how many groups we fill.
3750 We return -1 if no match, -2 if an internal error (such as the
3751 failure stack overflowing). Otherwise, we return the length of the
3752 matched substring. */
3755 re_match_2 (bufp
, string1
, size1
, string2
, size2
, pos
, regs
, stop
)
3756 struct re_pattern_buffer
*bufp
;
3757 const char *string1
, *string2
;
3760 struct re_registers
*regs
;
3763 int result
= re_match_2_internal (bufp
, string1
, size1
, string2
, size2
,
3765 #ifndef REGEX_MALLOC
3773 weak_alias (__re_match_2
, re_match_2
)
3776 /* This is a separate function so that we can force an alloca cleanup
3779 re_match_2_internal (bufp
, string1
, size1
, string2
, size2
, pos
, regs
, stop
)
3780 struct re_pattern_buffer
*bufp
;
3781 const char *string1
, *string2
;
3784 struct re_registers
*regs
;
3787 /* General temporaries. */
3791 /* Just past the end of the corresponding string. */
3792 const char *end1
, *end2
;
3794 /* Pointers into string1 and string2, just past the last characters in
3795 each to consider matching. */
3796 const char *end_match_1
, *end_match_2
;
3798 /* Where we are in the data, and the end of the current string. */
3799 const char *d
, *dend
;
3801 /* Where we are in the pattern, and the end of the pattern. */
3802 unsigned char *p
= bufp
->buffer
;
3803 register unsigned char *pend
= p
+ bufp
->used
;
3805 /* Mark the opcode just after a start_memory, so we can test for an
3806 empty subpattern when we get to the stop_memory. */
3807 unsigned char *just_past_start_mem
= 0;
3809 /* We use this to map every character in the string. */
3810 RE_TRANSLATE_TYPE translate
= bufp
->translate
;
3812 /* Failure point stack. Each place that can handle a failure further
3813 down the line pushes a failure point on this stack. It consists of
3814 restart, regend, and reg_info for all registers corresponding to
3815 the subexpressions we're currently inside, plus the number of such
3816 registers, and, finally, two char *'s. The first char * is where
3817 to resume scanning the pattern; the second one is where to resume
3818 scanning the strings. If the latter is zero, the failure point is
3819 a ``dummy''; if a failure happens and the failure point is a dummy,
3820 it gets discarded and the next next one is tried. */
3821 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
3822 fail_stack_type fail_stack
;
3825 static unsigned failure_id
= 0;
3826 unsigned nfailure_points_pushed
= 0, nfailure_points_popped
= 0;
3830 /* This holds the pointer to the failure stack, when
3831 it is allocated relocatably. */
3832 fail_stack_elt_t
*failure_stack_ptr
;
3835 /* We fill all the registers internally, independent of what we
3836 return, for use in backreferences. The number here includes
3837 an element for register zero. */
3838 size_t num_regs
= bufp
->re_nsub
+ 1;
3840 /* The currently active registers. */
3841 active_reg_t lowest_active_reg
= NO_LOWEST_ACTIVE_REG
;
3842 active_reg_t highest_active_reg
= NO_HIGHEST_ACTIVE_REG
;
3844 /* Information on the contents of registers. These are pointers into
3845 the input strings; they record just what was matched (on this
3846 attempt) by a subexpression part of the pattern, that is, the
3847 regnum-th regstart pointer points to where in the pattern we began
3848 matching and the regnum-th regend points to right after where we
3849 stopped matching the regnum-th subexpression. (The zeroth register
3850 keeps track of what the whole pattern matches.) */
3851 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
3852 const char **regstart
, **regend
;
3855 /* If a group that's operated upon by a repetition operator fails to
3856 match anything, then the register for its start will need to be
3857 restored because it will have been set to wherever in the string we
3858 are when we last see its open-group operator. Similarly for a
3860 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
3861 const char **old_regstart
, **old_regend
;
3864 /* The is_active field of reg_info helps us keep track of which (possibly
3865 nested) subexpressions we are currently in. The matched_something
3866 field of reg_info[reg_num] helps us tell whether or not we have
3867 matched any of the pattern so far this time through the reg_num-th
3868 subexpression. These two fields get reset each time through any
3869 loop their register is in. */
3870 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
3871 register_info_type
*reg_info
;
3874 /* The following record the register info as found in the above
3875 variables when we find a match better than any we've seen before.
3876 This happens as we backtrack through the failure points, which in
3877 turn happens only if we have not yet matched the entire string. */
3878 unsigned best_regs_set
= false;
3879 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
3880 const char **best_regstart
, **best_regend
;
3883 /* Logically, this is `best_regend[0]'. But we don't want to have to
3884 allocate space for that if we're not allocating space for anything
3885 else (see below). Also, we never need info about register 0 for
3886 any of the other register vectors, and it seems rather a kludge to
3887 treat `best_regend' differently than the rest. So we keep track of
3888 the end of the best match so far in a separate variable. We
3889 initialize this to NULL so that when we backtrack the first time
3890 and need to test it, it's not garbage. */
3891 const char *match_end
= NULL
;
3893 /* This helps SET_REGS_MATCHED avoid doing redundant work. */
3894 int set_regs_matched_done
= 0;
3896 /* Used when we pop values we don't care about. */
3897 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
3898 const char **reg_dummy
;
3899 register_info_type
*reg_info_dummy
;
3903 /* Counts the total number of registers pushed. */
3904 unsigned num_regs_pushed
= 0;
3907 DEBUG_PRINT1 ("\n\nEntering re_match_2.\n");
3911 #ifdef MATCH_MAY_ALLOCATE
3912 /* Do not bother to initialize all the register variables if there are
3913 no groups in the pattern, as it takes a fair amount of time. If
3914 there are groups, we include space for register 0 (the whole
3915 pattern), even though we never use it, since it simplifies the
3916 array indexing. We should fix this. */
3919 regstart
= REGEX_TALLOC (num_regs
, const char *);
3920 regend
= REGEX_TALLOC (num_regs
, const char *);
3921 old_regstart
= REGEX_TALLOC (num_regs
, const char *);
3922 old_regend
= REGEX_TALLOC (num_regs
, const char *);
3923 best_regstart
= REGEX_TALLOC (num_regs
, const char *);
3924 best_regend
= REGEX_TALLOC (num_regs
, const char *);
3925 reg_info
= REGEX_TALLOC (num_regs
, register_info_type
);
3926 reg_dummy
= REGEX_TALLOC (num_regs
, const char *);
3927 reg_info_dummy
= REGEX_TALLOC (num_regs
, register_info_type
);
3929 if (!(regstart
&& regend
&& old_regstart
&& old_regend
&& reg_info
3930 && best_regstart
&& best_regend
&& reg_dummy
&& reg_info_dummy
))
3938 /* We must initialize all our variables to NULL, so that
3939 `FREE_VARIABLES' doesn't try to free them. */
3940 regstart
= regend
= old_regstart
= old_regend
= best_regstart
3941 = best_regend
= reg_dummy
= NULL
;
3942 reg_info
= reg_info_dummy
= (register_info_type
*) NULL
;
3944 #endif /* MATCH_MAY_ALLOCATE */
3946 /* The starting position is bogus. */
3947 if (pos
< 0 || pos
> size1
+ size2
)
3953 /* Initialize subexpression text positions to -1 to mark ones that no
3954 start_memory/stop_memory has been seen for. Also initialize the
3955 register information struct. */
3956 for (mcnt
= 1; (unsigned) mcnt
< num_regs
; mcnt
++)
3958 regstart
[mcnt
] = regend
[mcnt
]
3959 = old_regstart
[mcnt
] = old_regend
[mcnt
] = REG_UNSET_VALUE
;
3961 REG_MATCH_NULL_STRING_P (reg_info
[mcnt
]) = MATCH_NULL_UNSET_VALUE
;
3962 IS_ACTIVE (reg_info
[mcnt
]) = 0;
3963 MATCHED_SOMETHING (reg_info
[mcnt
]) = 0;
3964 EVER_MATCHED_SOMETHING (reg_info
[mcnt
]) = 0;
3967 /* We move `string1' into `string2' if the latter's empty -- but not if
3968 `string1' is null. */
3969 if (size2
== 0 && string1
!= NULL
)
3976 end1
= string1
+ size1
;
3977 end2
= string2
+ size2
;
3979 /* Compute where to stop matching, within the two strings. */
3982 end_match_1
= string1
+ stop
;
3983 end_match_2
= string2
;
3988 end_match_2
= string2
+ stop
- size1
;
3991 /* `p' scans through the pattern as `d' scans through the data.
3992 `dend' is the end of the input string that `d' points within. `d'
3993 is advanced into the following input string whenever necessary, but
3994 this happens before fetching; therefore, at the beginning of the
3995 loop, `d' can be pointing at the end of a string, but it cannot
3997 if (size1
> 0 && pos
<= size1
)
4004 d
= string2
+ pos
- size1
;
4008 DEBUG_PRINT1 ("The compiled pattern is:\n");
4009 DEBUG_PRINT_COMPILED_PATTERN (bufp
, p
, pend
);
4010 DEBUG_PRINT1 ("The string to match is: `");
4011 DEBUG_PRINT_DOUBLE_STRING (d
, string1
, size1
, string2
, size2
);
4012 DEBUG_PRINT1 ("'\n");
4014 /* This loops over pattern commands. It exits by returning from the
4015 function if the match is complete, or it drops through if the match
4016 fails at this starting point in the input data. */
4020 DEBUG_PRINT2 ("\n%p: ", p
);
4022 DEBUG_PRINT2 ("\n0x%x: ", p
);
4026 { /* End of pattern means we might have succeeded. */
4027 DEBUG_PRINT1 ("end of pattern ... ");
4029 /* If we haven't matched the entire string, and we want the
4030 longest match, try backtracking. */
4031 if (d
!= end_match_2
)
4033 /* 1 if this match ends in the same string (string1 or string2)
4034 as the best previous match. */
4035 boolean same_str_p
= (FIRST_STRING_P (match_end
)
4036 == MATCHING_IN_FIRST_STRING
);
4037 /* 1 if this match is the best seen so far. */
4038 boolean best_match_p
;
4040 /* AIX compiler got confused when this was combined
4041 with the previous declaration. */
4043 best_match_p
= d
> match_end
;
4045 best_match_p
= !MATCHING_IN_FIRST_STRING
;
4047 DEBUG_PRINT1 ("backtracking.\n");
4049 if (!FAIL_STACK_EMPTY ())
4050 { /* More failure points to try. */
4052 /* If exceeds best match so far, save it. */
4053 if (!best_regs_set
|| best_match_p
)
4055 best_regs_set
= true;
4058 DEBUG_PRINT1 ("\nSAVING match as best so far.\n");
4060 for (mcnt
= 1; (unsigned) mcnt
< num_regs
; mcnt
++)
4062 best_regstart
[mcnt
] = regstart
[mcnt
];
4063 best_regend
[mcnt
] = regend
[mcnt
];
4069 /* If no failure points, don't restore garbage. And if
4070 last match is real best match, don't restore second
4072 else if (best_regs_set
&& !best_match_p
)
4075 /* Restore best match. It may happen that `dend ==
4076 end_match_1' while the restored d is in string2.
4077 For example, the pattern `x.*y.*z' against the
4078 strings `x-' and `y-z-', if the two strings are
4079 not consecutive in memory. */
4080 DEBUG_PRINT1 ("Restoring best registers.\n");
4083 dend
= ((d
>= string1
&& d
<= end1
)
4084 ? end_match_1
: end_match_2
);
4086 for (mcnt
= 1; (unsigned) mcnt
< num_regs
; mcnt
++)
4088 regstart
[mcnt
] = best_regstart
[mcnt
];
4089 regend
[mcnt
] = best_regend
[mcnt
];
4092 } /* d != end_match_2 */
4095 DEBUG_PRINT1 ("Accepting match.\n");
4097 /* If caller wants register contents data back, do it. */
4098 if (regs
&& !bufp
->no_sub
)
4100 /* Have the register data arrays been allocated? */
4101 if (bufp
->regs_allocated
== REGS_UNALLOCATED
)
4102 { /* No. So allocate them with malloc. We need one
4103 extra element beyond `num_regs' for the `-1' marker
4105 regs
->num_regs
= MAX (RE_NREGS
, num_regs
+ 1);
4106 regs
->start
= TALLOC (regs
->num_regs
, regoff_t
);
4107 regs
->end
= TALLOC (regs
->num_regs
, regoff_t
);
4108 if (regs
->start
== NULL
|| regs
->end
== NULL
)
4113 bufp
->regs_allocated
= REGS_REALLOCATE
;
4115 else if (bufp
->regs_allocated
== REGS_REALLOCATE
)
4116 { /* Yes. If we need more elements than were already
4117 allocated, reallocate them. If we need fewer, just
4119 if (regs
->num_regs
< num_regs
+ 1)
4121 regs
->num_regs
= num_regs
+ 1;
4122 RETALLOC (regs
->start
, regs
->num_regs
, regoff_t
);
4123 RETALLOC (regs
->end
, regs
->num_regs
, regoff_t
);
4124 if (regs
->start
== NULL
|| regs
->end
== NULL
)
4133 /* These braces fend off a "empty body in an else-statement"
4134 warning under GCC when assert expands to nothing. */
4135 assert (bufp
->regs_allocated
== REGS_FIXED
);
4138 /* Convert the pointer data in `regstart' and `regend' to
4139 indices. Register zero has to be set differently,
4140 since we haven't kept track of any info for it. */
4141 if (regs
->num_regs
> 0)
4143 regs
->start
[0] = pos
;
4144 regs
->end
[0] = (MATCHING_IN_FIRST_STRING
4145 ? ((regoff_t
) (d
- string1
))
4146 : ((regoff_t
) (d
- string2
+ size1
)));
4149 /* Go through the first `min (num_regs, regs->num_regs)'
4150 registers, since that is all we initialized. */
4151 for (mcnt
= 1; (unsigned) mcnt
< MIN (num_regs
, regs
->num_regs
);
4154 if (REG_UNSET (regstart
[mcnt
]) || REG_UNSET (regend
[mcnt
]))
4155 regs
->start
[mcnt
] = regs
->end
[mcnt
] = -1;
4159 = (regoff_t
) POINTER_TO_OFFSET (regstart
[mcnt
]);
4161 = (regoff_t
) POINTER_TO_OFFSET (regend
[mcnt
]);
4165 /* If the regs structure we return has more elements than
4166 were in the pattern, set the extra elements to -1. If
4167 we (re)allocated the registers, this is the case,
4168 because we always allocate enough to have at least one
4170 for (mcnt
= num_regs
; (unsigned) mcnt
< regs
->num_regs
; mcnt
++)
4171 regs
->start
[mcnt
] = regs
->end
[mcnt
] = -1;
4172 } /* regs && !bufp->no_sub */
4174 DEBUG_PRINT4 ("%u failure points pushed, %u popped (%u remain).\n",
4175 nfailure_points_pushed
, nfailure_points_popped
,
4176 nfailure_points_pushed
- nfailure_points_popped
);
4177 DEBUG_PRINT2 ("%u registers pushed.\n", num_regs_pushed
);
4179 mcnt
= d
- pos
- (MATCHING_IN_FIRST_STRING
4183 DEBUG_PRINT2 ("Returning %d from re_match_2.\n", mcnt
);
4189 /* Otherwise match next pattern command. */
4190 switch (SWITCH_ENUM_CAST ((re_opcode_t
) *p
++))
4192 /* Ignore these. Used to ignore the n of succeed_n's which
4193 currently have n == 0. */
4195 DEBUG_PRINT1 ("EXECUTING no_op.\n");
4199 DEBUG_PRINT1 ("EXECUTING succeed.\n");
4202 /* Match the next n pattern characters exactly. The following
4203 byte in the pattern defines n, and the n bytes after that
4204 are the characters to match. */
4207 DEBUG_PRINT2 ("EXECUTING exactn %d.\n", mcnt
);
4209 /* This is written out as an if-else so we don't waste time
4210 testing `translate' inside the loop. */
4216 if ((unsigned char) translate
[(unsigned char) *d
++]
4217 != (unsigned char) *p
++)
4227 if (*d
++ != (char) *p
++) goto fail
;
4231 SET_REGS_MATCHED ();
4235 /* Match any character except possibly a newline or a null. */
4237 DEBUG_PRINT1 ("EXECUTING anychar.\n");
4241 if ((!(bufp
->syntax
& RE_DOT_NEWLINE
) && TRANSLATE (*d
) == '\n')
4242 || (bufp
->syntax
& RE_DOT_NOT_NULL
&& TRANSLATE (*d
) == '\000'))
4245 SET_REGS_MATCHED ();
4246 DEBUG_PRINT2 (" Matched `%d'.\n", *d
);
4254 register unsigned char c
;
4255 boolean
not = (re_opcode_t
) *(p
- 1) == charset_not
;
4257 DEBUG_PRINT2 ("EXECUTING charset%s.\n", not ? "_not" : "");
4260 c
= TRANSLATE (*d
); /* The character to match. */
4262 /* Cast to `unsigned' instead of `unsigned char' in case the
4263 bit list is a full 32 bytes long. */
4264 if (c
< (unsigned) (*p
* BYTEWIDTH
)
4265 && p
[1 + c
/ BYTEWIDTH
] & (1 << (c
% BYTEWIDTH
)))
4270 if (!not) goto fail
;
4272 SET_REGS_MATCHED ();
4278 /* The beginning of a group is represented by start_memory.
4279 The arguments are the register number in the next byte, and the
4280 number of groups inner to this one in the next. The text
4281 matched within the group is recorded (in the internal
4282 registers data structure) under the register number. */
4284 DEBUG_PRINT3 ("EXECUTING start_memory %d (%d):\n", *p
, p
[1]);
4286 /* Find out if this group can match the empty string. */
4287 p1
= p
; /* To send to group_match_null_string_p. */
4289 if (REG_MATCH_NULL_STRING_P (reg_info
[*p
]) == MATCH_NULL_UNSET_VALUE
)
4290 REG_MATCH_NULL_STRING_P (reg_info
[*p
])
4291 = group_match_null_string_p (&p1
, pend
, reg_info
);
4293 /* Save the position in the string where we were the last time
4294 we were at this open-group operator in case the group is
4295 operated upon by a repetition operator, e.g., with `(a*)*b'
4296 against `ab'; then we want to ignore where we are now in
4297 the string in case this attempt to match fails. */
4298 old_regstart
[*p
] = REG_MATCH_NULL_STRING_P (reg_info
[*p
])
4299 ? REG_UNSET (regstart
[*p
]) ? d
: regstart
[*p
]
4301 DEBUG_PRINT2 (" old_regstart: %d\n",
4302 POINTER_TO_OFFSET (old_regstart
[*p
]));
4305 DEBUG_PRINT2 (" regstart: %d\n", POINTER_TO_OFFSET (regstart
[*p
]));
4307 IS_ACTIVE (reg_info
[*p
]) = 1;
4308 MATCHED_SOMETHING (reg_info
[*p
]) = 0;
4310 /* Clear this whenever we change the register activity status. */
4311 set_regs_matched_done
= 0;
4313 /* This is the new highest active register. */
4314 highest_active_reg
= *p
;
4316 /* If nothing was active before, this is the new lowest active
4318 if (lowest_active_reg
== NO_LOWEST_ACTIVE_REG
)
4319 lowest_active_reg
= *p
;
4321 /* Move past the register number and inner group count. */
4323 just_past_start_mem
= p
;
4328 /* The stop_memory opcode represents the end of a group. Its
4329 arguments are the same as start_memory's: the register
4330 number, and the number of inner groups. */
4332 DEBUG_PRINT3 ("EXECUTING stop_memory %d (%d):\n", *p
, p
[1]);
4334 /* We need to save the string position the last time we were at
4335 this close-group operator in case the group is operated
4336 upon by a repetition operator, e.g., with `((a*)*(b*)*)*'
4337 against `aba'; then we want to ignore where we are now in
4338 the string in case this attempt to match fails. */
4339 old_regend
[*p
] = REG_MATCH_NULL_STRING_P (reg_info
[*p
])
4340 ? REG_UNSET (regend
[*p
]) ? d
: regend
[*p
]
4342 DEBUG_PRINT2 (" old_regend: %d\n",
4343 POINTER_TO_OFFSET (old_regend
[*p
]));
4346 DEBUG_PRINT2 (" regend: %d\n", POINTER_TO_OFFSET (regend
[*p
]));
4348 /* This register isn't active anymore. */
4349 IS_ACTIVE (reg_info
[*p
]) = 0;
4351 /* Clear this whenever we change the register activity status. */
4352 set_regs_matched_done
= 0;
4354 /* If this was the only register active, nothing is active
4356 if (lowest_active_reg
== highest_active_reg
)
4358 lowest_active_reg
= NO_LOWEST_ACTIVE_REG
;
4359 highest_active_reg
= NO_HIGHEST_ACTIVE_REG
;
4362 { /* We must scan for the new highest active register, since
4363 it isn't necessarily one less than now: consider
4364 (a(b)c(d(e)f)g). When group 3 ends, after the f), the
4365 new highest active register is 1. */
4366 unsigned char r
= *p
- 1;
4367 while (r
> 0 && !IS_ACTIVE (reg_info
[r
]))
4370 /* If we end up at register zero, that means that we saved
4371 the registers as the result of an `on_failure_jump', not
4372 a `start_memory', and we jumped to past the innermost
4373 `stop_memory'. For example, in ((.)*) we save
4374 registers 1 and 2 as a result of the *, but when we pop
4375 back to the second ), we are at the stop_memory 1.
4376 Thus, nothing is active. */
4379 lowest_active_reg
= NO_LOWEST_ACTIVE_REG
;
4380 highest_active_reg
= NO_HIGHEST_ACTIVE_REG
;
4383 highest_active_reg
= r
;
4386 /* If just failed to match something this time around with a
4387 group that's operated on by a repetition operator, try to
4388 force exit from the ``loop'', and restore the register
4389 information for this group that we had before trying this
4391 if ((!MATCHED_SOMETHING (reg_info
[*p
])
4392 || just_past_start_mem
== p
- 1)
4395 boolean is_a_jump_n
= false;
4399 switch ((re_opcode_t
) *p1
++)
4403 case pop_failure_jump
:
4404 case maybe_pop_jump
:
4406 case dummy_failure_jump
:
4407 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
4417 /* If the next operation is a jump backwards in the pattern
4418 to an on_failure_jump right before the start_memory
4419 corresponding to this stop_memory, exit from the loop
4420 by forcing a failure after pushing on the stack the
4421 on_failure_jump's jump in the pattern, and d. */
4422 if (mcnt
< 0 && (re_opcode_t
) *p1
== on_failure_jump
4423 && (re_opcode_t
) p1
[3] == start_memory
&& p1
[4] == *p
)
4425 /* If this group ever matched anything, then restore
4426 what its registers were before trying this last
4427 failed match, e.g., with `(a*)*b' against `ab' for
4428 regstart[1], and, e.g., with `((a*)*(b*)*)*'
4429 against `aba' for regend[3].
4431 Also restore the registers for inner groups for,
4432 e.g., `((a*)(b*))*' against `aba' (register 3 would
4433 otherwise get trashed). */
4435 if (EVER_MATCHED_SOMETHING (reg_info
[*p
]))
4439 EVER_MATCHED_SOMETHING (reg_info
[*p
]) = 0;
4441 /* Restore this and inner groups' (if any) registers. */
4442 for (r
= *p
; r
< (unsigned) *p
+ (unsigned) *(p
+ 1);
4445 regstart
[r
] = old_regstart
[r
];
4447 /* xx why this test? */
4448 if (old_regend
[r
] >= regstart
[r
])
4449 regend
[r
] = old_regend
[r
];
4453 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
4454 PUSH_FAILURE_POINT (p1
+ mcnt
, d
, -2);
4460 /* Move past the register number and the inner group count. */
4465 /* \<digit> has been turned into a `duplicate' command which is
4466 followed by the numeric value of <digit> as the register number. */
4469 register const char *d2
, *dend2
;
4470 int regno
= *p
++; /* Get which register to match against. */
4471 DEBUG_PRINT2 ("EXECUTING duplicate %d.\n", regno
);
4473 /* Can't back reference a group which we've never matched. */
4474 if (REG_UNSET (regstart
[regno
]) || REG_UNSET (regend
[regno
]))
4477 /* Where in input to try to start matching. */
4478 d2
= regstart
[regno
];
4480 /* Where to stop matching; if both the place to start and
4481 the place to stop matching are in the same string, then
4482 set to the place to stop, otherwise, for now have to use
4483 the end of the first string. */
4485 dend2
= ((FIRST_STRING_P (regstart
[regno
])
4486 == FIRST_STRING_P (regend
[regno
]))
4487 ? regend
[regno
] : end_match_1
);
4490 /* If necessary, advance to next segment in register
4494 if (dend2
== end_match_2
) break;
4495 if (dend2
== regend
[regno
]) break;
4497 /* End of string1 => advance to string2. */
4499 dend2
= regend
[regno
];
4501 /* At end of register contents => success */
4502 if (d2
== dend2
) break;
4504 /* If necessary, advance to next segment in data. */
4507 /* How many characters left in this segment to match. */
4510 /* Want how many consecutive characters we can match in
4511 one shot, so, if necessary, adjust the count. */
4512 if (mcnt
> dend2
- d2
)
4515 /* Compare that many; failure if mismatch, else move
4518 ? bcmp_translate (d
, d2
, mcnt
, translate
)
4519 : memcmp (d
, d2
, mcnt
))
4521 d
+= mcnt
, d2
+= mcnt
;
4523 /* Do this because we've match some characters. */
4524 SET_REGS_MATCHED ();
4530 /* begline matches the empty string at the beginning of the string
4531 (unless `not_bol' is set in `bufp'), and, if
4532 `newline_anchor' is set, after newlines. */
4534 DEBUG_PRINT1 ("EXECUTING begline.\n");
4536 if (AT_STRINGS_BEG (d
))
4538 if (!bufp
->not_bol
) break;
4540 else if (d
[-1] == '\n' && bufp
->newline_anchor
)
4544 /* In all other cases, we fail. */
4548 /* endline is the dual of begline. */
4550 DEBUG_PRINT1 ("EXECUTING endline.\n");
4552 if (AT_STRINGS_END (d
))
4554 if (!bufp
->not_eol
) break;
4557 /* We have to ``prefetch'' the next character. */
4558 else if ((d
== end1
? *string2
: *d
) == '\n'
4559 && bufp
->newline_anchor
)
4566 /* Match at the very beginning of the data. */
4568 DEBUG_PRINT1 ("EXECUTING begbuf.\n");
4569 if (AT_STRINGS_BEG (d
))
4574 /* Match at the very end of the data. */
4576 DEBUG_PRINT1 ("EXECUTING endbuf.\n");
4577 if (AT_STRINGS_END (d
))
4582 /* on_failure_keep_string_jump is used to optimize `.*\n'. It
4583 pushes NULL as the value for the string on the stack. Then
4584 `pop_failure_point' will keep the current value for the
4585 string, instead of restoring it. To see why, consider
4586 matching `foo\nbar' against `.*\n'. The .* matches the foo;
4587 then the . fails against the \n. But the next thing we want
4588 to do is match the \n against the \n; if we restored the
4589 string value, we would be back at the foo.
4591 Because this is used only in specific cases, we don't need to
4592 check all the things that `on_failure_jump' does, to make
4593 sure the right things get saved on the stack. Hence we don't
4594 share its code. The only reason to push anything on the
4595 stack at all is that otherwise we would have to change
4596 `anychar's code to do something besides goto fail in this
4597 case; that seems worse than this. */
4598 case on_failure_keep_string_jump
:
4599 DEBUG_PRINT1 ("EXECUTING on_failure_keep_string_jump");
4601 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
4603 DEBUG_PRINT3 (" %d (to %p):\n", mcnt
, p
+ mcnt
);
4605 DEBUG_PRINT3 (" %d (to 0x%x):\n", mcnt
, p
+ mcnt
);
4608 PUSH_FAILURE_POINT (p
+ mcnt
, NULL
, -2);
4612 /* Uses of on_failure_jump:
4614 Each alternative starts with an on_failure_jump that points
4615 to the beginning of the next alternative. Each alternative
4616 except the last ends with a jump that in effect jumps past
4617 the rest of the alternatives. (They really jump to the
4618 ending jump of the following alternative, because tensioning
4619 these jumps is a hassle.)
4621 Repeats start with an on_failure_jump that points past both
4622 the repetition text and either the following jump or
4623 pop_failure_jump back to this on_failure_jump. */
4624 case on_failure_jump
:
4626 DEBUG_PRINT1 ("EXECUTING on_failure_jump");
4628 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
4630 DEBUG_PRINT3 (" %d (to %p)", mcnt
, p
+ mcnt
);
4632 DEBUG_PRINT3 (" %d (to 0x%x)", mcnt
, p
+ mcnt
);
4635 /* If this on_failure_jump comes right before a group (i.e.,
4636 the original * applied to a group), save the information
4637 for that group and all inner ones, so that if we fail back
4638 to this point, the group's information will be correct.
4639 For example, in \(a*\)*\1, we need the preceding group,
4640 and in \(zz\(a*\)b*\)\2, we need the inner group. */
4642 /* We can't use `p' to check ahead because we push
4643 a failure point to `p + mcnt' after we do this. */
4646 /* We need to skip no_op's before we look for the
4647 start_memory in case this on_failure_jump is happening as
4648 the result of a completed succeed_n, as in \(a\)\{1,3\}b\1
4650 while (p1
< pend
&& (re_opcode_t
) *p1
== no_op
)
4653 if (p1
< pend
&& (re_opcode_t
) *p1
== start_memory
)
4655 /* We have a new highest active register now. This will
4656 get reset at the start_memory we are about to get to,
4657 but we will have saved all the registers relevant to
4658 this repetition op, as described above. */
4659 highest_active_reg
= *(p1
+ 1) + *(p1
+ 2);
4660 if (lowest_active_reg
== NO_LOWEST_ACTIVE_REG
)
4661 lowest_active_reg
= *(p1
+ 1);
4664 DEBUG_PRINT1 (":\n");
4665 PUSH_FAILURE_POINT (p
+ mcnt
, d
, -2);
4669 /* A smart repeat ends with `maybe_pop_jump'.
4670 We change it to either `pop_failure_jump' or `jump'. */
4671 case maybe_pop_jump
:
4672 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
4673 DEBUG_PRINT2 ("EXECUTING maybe_pop_jump %d.\n", mcnt
);
4675 register unsigned char *p2
= p
;
4677 /* Compare the beginning of the repeat with what in the
4678 pattern follows its end. If we can establish that there
4679 is nothing that they would both match, i.e., that we
4680 would have to backtrack because of (as in, e.g., `a*a')
4681 then we can change to pop_failure_jump, because we'll
4682 never have to backtrack.
4684 This is not true in the case of alternatives: in
4685 `(a|ab)*' we do need to backtrack to the `ab' alternative
4686 (e.g., if the string was `ab'). But instead of trying to
4687 detect that here, the alternative has put on a dummy
4688 failure point which is what we will end up popping. */
4690 /* Skip over open/close-group commands.
4691 If what follows this loop is a ...+ construct,
4692 look at what begins its body, since we will have to
4693 match at least one of that. */
4697 && ((re_opcode_t
) *p2
== stop_memory
4698 || (re_opcode_t
) *p2
== start_memory
))
4700 else if (p2
+ 6 < pend
4701 && (re_opcode_t
) *p2
== dummy_failure_jump
)
4708 /* p1[0] ... p1[2] are the `on_failure_jump' corresponding
4709 to the `maybe_finalize_jump' of this case. Examine what
4712 /* If we're at the end of the pattern, we can change. */
4715 /* Consider what happens when matching ":\(.*\)"
4716 against ":/". I don't really understand this code
4718 p
[-3] = (unsigned char) pop_failure_jump
;
4720 (" End of pattern: change to `pop_failure_jump'.\n");
4723 else if ((re_opcode_t
) *p2
== exactn
4724 || (bufp
->newline_anchor
&& (re_opcode_t
) *p2
== endline
))
4726 register unsigned char c
4727 = *p2
== (unsigned char) endline
? '\n' : p2
[2];
4729 if ((re_opcode_t
) p1
[3] == exactn
&& p1
[5] != c
)
4731 p
[-3] = (unsigned char) pop_failure_jump
;
4732 DEBUG_PRINT3 (" %c != %c => pop_failure_jump.\n",
4736 else if ((re_opcode_t
) p1
[3] == charset
4737 || (re_opcode_t
) p1
[3] == charset_not
)
4739 int not = (re_opcode_t
) p1
[3] == charset_not
;
4741 if (c
< (unsigned char) (p1
[4] * BYTEWIDTH
)
4742 && p1
[5 + c
/ BYTEWIDTH
] & (1 << (c
% BYTEWIDTH
)))
4745 /* `not' is equal to 1 if c would match, which means
4746 that we can't change to pop_failure_jump. */
4749 p
[-3] = (unsigned char) pop_failure_jump
;
4750 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
4754 else if ((re_opcode_t
) *p2
== charset
)
4757 register unsigned char c
4758 = *p2
== (unsigned char) endline
? '\n' : p2
[2];
4762 if ((re_opcode_t
) p1
[3] == exactn
4763 && ! ((int) p2
[1] * BYTEWIDTH
> (int) p1
[5]
4764 && (p2
[2 + p1
[5] / BYTEWIDTH
]
4765 & (1 << (p1
[5] % BYTEWIDTH
)))))
4767 if ((re_opcode_t
) p1
[3] == exactn
4768 && ! ((int) p2
[1] * BYTEWIDTH
> (int) p1
[4]
4769 && (p2
[2 + p1
[4] / BYTEWIDTH
]
4770 & (1 << (p1
[4] % BYTEWIDTH
)))))
4773 p
[-3] = (unsigned char) pop_failure_jump
;
4774 DEBUG_PRINT3 (" %c != %c => pop_failure_jump.\n",
4778 else if ((re_opcode_t
) p1
[3] == charset_not
)
4781 /* We win if the charset_not inside the loop
4782 lists every character listed in the charset after. */
4783 for (idx
= 0; idx
< (int) p2
[1]; idx
++)
4784 if (! (p2
[2 + idx
] == 0
4785 || (idx
< (int) p1
[4]
4786 && ((p2
[2 + idx
] & ~ p1
[5 + idx
]) == 0))))
4791 p
[-3] = (unsigned char) pop_failure_jump
;
4792 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
4795 else if ((re_opcode_t
) p1
[3] == charset
)
4798 /* We win if the charset inside the loop
4799 has no overlap with the one after the loop. */
4801 idx
< (int) p2
[1] && idx
< (int) p1
[4];
4803 if ((p2
[2 + idx
] & p1
[5 + idx
]) != 0)
4806 if (idx
== p2
[1] || idx
== p1
[4])
4808 p
[-3] = (unsigned char) pop_failure_jump
;
4809 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
4814 p
-= 2; /* Point at relative address again. */
4815 if ((re_opcode_t
) p
[-1] != pop_failure_jump
)
4817 p
[-1] = (unsigned char) jump
;
4818 DEBUG_PRINT1 (" Match => jump.\n");
4819 goto unconditional_jump
;
4821 /* Note fall through. */
4824 /* The end of a simple repeat has a pop_failure_jump back to
4825 its matching on_failure_jump, where the latter will push a
4826 failure point. The pop_failure_jump takes off failure
4827 points put on by this pop_failure_jump's matching
4828 on_failure_jump; we got through the pattern to here from the
4829 matching on_failure_jump, so didn't fail. */
4830 case pop_failure_jump
:
4832 /* We need to pass separate storage for the lowest and
4833 highest registers, even though we don't care about the
4834 actual values. Otherwise, we will restore only one
4835 register from the stack, since lowest will == highest in
4836 `pop_failure_point'. */
4837 active_reg_t dummy_low_reg
, dummy_high_reg
;
4838 unsigned char *pdummy
;
4841 DEBUG_PRINT1 ("EXECUTING pop_failure_jump.\n");
4842 POP_FAILURE_POINT (sdummy
, pdummy
,
4843 dummy_low_reg
, dummy_high_reg
,
4844 reg_dummy
, reg_dummy
, reg_info_dummy
);
4846 /* Note fall through. */
4850 DEBUG_PRINT2 ("\n%p: ", p
);
4852 DEBUG_PRINT2 ("\n0x%x: ", p
);
4854 /* Note fall through. */
4856 /* Unconditionally jump (without popping any failure points). */
4858 EXTRACT_NUMBER_AND_INCR (mcnt
, p
); /* Get the amount to jump. */
4859 DEBUG_PRINT2 ("EXECUTING jump %d ", mcnt
);
4860 p
+= mcnt
; /* Do the jump. */
4862 DEBUG_PRINT2 ("(to %p).\n", p
);
4864 DEBUG_PRINT2 ("(to 0x%x).\n", p
);
4869 /* We need this opcode so we can detect where alternatives end
4870 in `group_match_null_string_p' et al. */
4872 DEBUG_PRINT1 ("EXECUTING jump_past_alt.\n");
4873 goto unconditional_jump
;
4876 /* Normally, the on_failure_jump pushes a failure point, which
4877 then gets popped at pop_failure_jump. We will end up at
4878 pop_failure_jump, also, and with a pattern of, say, `a+', we
4879 are skipping over the on_failure_jump, so we have to push
4880 something meaningless for pop_failure_jump to pop. */
4881 case dummy_failure_jump
:
4882 DEBUG_PRINT1 ("EXECUTING dummy_failure_jump.\n");
4883 /* It doesn't matter what we push for the string here. What
4884 the code at `fail' tests is the value for the pattern. */
4885 PUSH_FAILURE_POINT (NULL
, NULL
, -2);
4886 goto unconditional_jump
;
4889 /* At the end of an alternative, we need to push a dummy failure
4890 point in case we are followed by a `pop_failure_jump', because
4891 we don't want the failure point for the alternative to be
4892 popped. For example, matching `(a|ab)*' against `aab'
4893 requires that we match the `ab' alternative. */
4894 case push_dummy_failure
:
4895 DEBUG_PRINT1 ("EXECUTING push_dummy_failure.\n");
4896 /* See comments just above at `dummy_failure_jump' about the
4898 PUSH_FAILURE_POINT (NULL
, NULL
, -2);
4901 /* Have to succeed matching what follows at least n times.
4902 After that, handle like `on_failure_jump'. */
4904 EXTRACT_NUMBER (mcnt
, p
+ 2);
4905 DEBUG_PRINT2 ("EXECUTING succeed_n %d.\n", mcnt
);
4908 /* Originally, this is how many times we HAVE to succeed. */
4913 STORE_NUMBER_AND_INCR (p
, mcnt
);
4915 DEBUG_PRINT3 (" Setting %p to %d.\n", p
- 2, mcnt
);
4917 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p
- 2, mcnt
);
4923 DEBUG_PRINT2 (" Setting two bytes from %p to no_op.\n", p
+2);
4925 DEBUG_PRINT2 (" Setting two bytes from 0x%x to no_op.\n", p
+2);
4927 p
[2] = (unsigned char) no_op
;
4928 p
[3] = (unsigned char) no_op
;
4934 EXTRACT_NUMBER (mcnt
, p
+ 2);
4935 DEBUG_PRINT2 ("EXECUTING jump_n %d.\n", mcnt
);
4937 /* Originally, this is how many times we CAN jump. */
4941 STORE_NUMBER (p
+ 2, mcnt
);
4943 DEBUG_PRINT3 (" Setting %p to %d.\n", p
+ 2, mcnt
);
4945 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p
+ 2, mcnt
);
4947 goto unconditional_jump
;
4949 /* If don't have to jump any more, skip over the rest of command. */
4956 DEBUG_PRINT1 ("EXECUTING set_number_at.\n");
4958 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
4960 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
4962 DEBUG_PRINT3 (" Setting %p to %d.\n", p1
, mcnt
);
4964 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p1
, mcnt
);
4966 STORE_NUMBER (p1
, mcnt
);
4971 /* The DEC Alpha C compiler 3.x generates incorrect code for the
4972 test WORDCHAR_P (d - 1) != WORDCHAR_P (d) in the expansion of
4973 AT_WORD_BOUNDARY, so this code is disabled. Expanding the
4974 macro and introducing temporary variables works around the bug. */
4977 DEBUG_PRINT1 ("EXECUTING wordbound.\n");
4978 if (AT_WORD_BOUNDARY (d
))
4983 DEBUG_PRINT1 ("EXECUTING notwordbound.\n");
4984 if (AT_WORD_BOUNDARY (d
))
4990 boolean prevchar
, thischar
;
4992 DEBUG_PRINT1 ("EXECUTING wordbound.\n");
4993 if (AT_STRINGS_BEG (d
) || AT_STRINGS_END (d
))
4996 prevchar
= WORDCHAR_P (d
- 1);
4997 thischar
= WORDCHAR_P (d
);
4998 if (prevchar
!= thischar
)
5005 boolean prevchar
, thischar
;
5007 DEBUG_PRINT1 ("EXECUTING notwordbound.\n");
5008 if (AT_STRINGS_BEG (d
) || AT_STRINGS_END (d
))
5011 prevchar
= WORDCHAR_P (d
- 1);
5012 thischar
= WORDCHAR_P (d
);
5013 if (prevchar
!= thischar
)
5020 DEBUG_PRINT1 ("EXECUTING wordbeg.\n");
5021 if (WORDCHAR_P (d
) && (AT_STRINGS_BEG (d
) || !WORDCHAR_P (d
- 1)))
5026 DEBUG_PRINT1 ("EXECUTING wordend.\n");
5027 if (!AT_STRINGS_BEG (d
) && WORDCHAR_P (d
- 1)
5028 && (!WORDCHAR_P (d
) || AT_STRINGS_END (d
)))
5034 DEBUG_PRINT1 ("EXECUTING before_dot.\n");
5035 if (PTR_CHAR_POS ((unsigned char *) d
) >= point
)
5040 DEBUG_PRINT1 ("EXECUTING at_dot.\n");
5041 if (PTR_CHAR_POS ((unsigned char *) d
) != point
)
5046 DEBUG_PRINT1 ("EXECUTING after_dot.\n");
5047 if (PTR_CHAR_POS ((unsigned char *) d
) <= point
)
5052 DEBUG_PRINT2 ("EXECUTING syntaxspec %d.\n", mcnt
);
5057 DEBUG_PRINT1 ("EXECUTING Emacs wordchar.\n");
5061 /* Can't use *d++ here; SYNTAX may be an unsafe macro. */
5063 if (SYNTAX (d
[-1]) != (enum syntaxcode
) mcnt
)
5065 SET_REGS_MATCHED ();
5069 DEBUG_PRINT2 ("EXECUTING notsyntaxspec %d.\n", mcnt
);
5071 goto matchnotsyntax
;
5074 DEBUG_PRINT1 ("EXECUTING Emacs notwordchar.\n");
5078 /* Can't use *d++ here; SYNTAX may be an unsafe macro. */
5080 if (SYNTAX (d
[-1]) == (enum syntaxcode
) mcnt
)
5082 SET_REGS_MATCHED ();
5085 #else /* not emacs */
5087 DEBUG_PRINT1 ("EXECUTING non-Emacs wordchar.\n");
5089 if (!WORDCHAR_P (d
))
5091 SET_REGS_MATCHED ();
5096 DEBUG_PRINT1 ("EXECUTING non-Emacs notwordchar.\n");
5100 SET_REGS_MATCHED ();
5103 #endif /* not emacs */
5108 continue; /* Successfully executed one pattern command; keep going. */
5111 /* We goto here if a matching operation fails. */
5113 if (!FAIL_STACK_EMPTY ())
5114 { /* A restart point is known. Restore to that state. */
5115 DEBUG_PRINT1 ("\nFAIL:\n");
5116 POP_FAILURE_POINT (d
, p
,
5117 lowest_active_reg
, highest_active_reg
,
5118 regstart
, regend
, reg_info
);
5120 /* If this failure point is a dummy, try the next one. */
5124 /* If we failed to the end of the pattern, don't examine *p. */
5128 boolean is_a_jump_n
= false;
5130 /* If failed to a backwards jump that's part of a repetition
5131 loop, need to pop this failure point and use the next one. */
5132 switch ((re_opcode_t
) *p
)
5136 case maybe_pop_jump
:
5137 case pop_failure_jump
:
5140 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
5143 if ((is_a_jump_n
&& (re_opcode_t
) *p1
== succeed_n
)
5145 && (re_opcode_t
) *p1
== on_failure_jump
))
5153 if (d
>= string1
&& d
<= end1
)
5157 break; /* Matching at this starting point really fails. */
5161 goto restore_best_regs
;
5165 return -1; /* Failure to match. */
5168 /* Subroutine definitions for re_match_2. */
5171 /* We are passed P pointing to a register number after a start_memory.
5173 Return true if the pattern up to the corresponding stop_memory can
5174 match the empty string, and false otherwise.
5176 If we find the matching stop_memory, sets P to point to one past its number.
5177 Otherwise, sets P to an undefined byte less than or equal to END.
5179 We don't handle duplicates properly (yet). */
5182 group_match_null_string_p (p
, end
, reg_info
)
5183 unsigned char **p
, *end
;
5184 register_info_type
*reg_info
;
5187 /* Point to after the args to the start_memory. */
5188 unsigned char *p1
= *p
+ 2;
5192 /* Skip over opcodes that can match nothing, and return true or
5193 false, as appropriate, when we get to one that can't, or to the
5194 matching stop_memory. */
5196 switch ((re_opcode_t
) *p1
)
5198 /* Could be either a loop or a series of alternatives. */
5199 case on_failure_jump
:
5201 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
5203 /* If the next operation is not a jump backwards in the
5208 /* Go through the on_failure_jumps of the alternatives,
5209 seeing if any of the alternatives cannot match nothing.
5210 The last alternative starts with only a jump,
5211 whereas the rest start with on_failure_jump and end
5212 with a jump, e.g., here is the pattern for `a|b|c':
5214 /on_failure_jump/0/6/exactn/1/a/jump_past_alt/0/6
5215 /on_failure_jump/0/6/exactn/1/b/jump_past_alt/0/3
5218 So, we have to first go through the first (n-1)
5219 alternatives and then deal with the last one separately. */
5222 /* Deal with the first (n-1) alternatives, which start
5223 with an on_failure_jump (see above) that jumps to right
5224 past a jump_past_alt. */
5226 while ((re_opcode_t
) p1
[mcnt
-3] == jump_past_alt
)
5228 /* `mcnt' holds how many bytes long the alternative
5229 is, including the ending `jump_past_alt' and
5232 if (!alt_match_null_string_p (p1
, p1
+ mcnt
- 3,
5236 /* Move to right after this alternative, including the
5240 /* Break if it's the beginning of an n-th alternative
5241 that doesn't begin with an on_failure_jump. */
5242 if ((re_opcode_t
) *p1
!= on_failure_jump
)
5245 /* Still have to check that it's not an n-th
5246 alternative that starts with an on_failure_jump. */
5248 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
5249 if ((re_opcode_t
) p1
[mcnt
-3] != jump_past_alt
)
5251 /* Get to the beginning of the n-th alternative. */
5257 /* Deal with the last alternative: go back and get number
5258 of the `jump_past_alt' just before it. `mcnt' contains
5259 the length of the alternative. */
5260 EXTRACT_NUMBER (mcnt
, p1
- 2);
5262 if (!alt_match_null_string_p (p1
, p1
+ mcnt
, reg_info
))
5265 p1
+= mcnt
; /* Get past the n-th alternative. */
5271 assert (p1
[1] == **p
);
5277 if (!common_op_match_null_string_p (&p1
, end
, reg_info
))
5280 } /* while p1 < end */
5283 } /* group_match_null_string_p */
5286 /* Similar to group_match_null_string_p, but doesn't deal with alternatives:
5287 It expects P to be the first byte of a single alternative and END one
5288 byte past the last. The alternative can contain groups. */
5291 alt_match_null_string_p (p
, end
, reg_info
)
5292 unsigned char *p
, *end
;
5293 register_info_type
*reg_info
;
5296 unsigned char *p1
= p
;
5300 /* Skip over opcodes that can match nothing, and break when we get
5301 to one that can't. */
5303 switch ((re_opcode_t
) *p1
)
5306 case on_failure_jump
:
5308 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
5313 if (!common_op_match_null_string_p (&p1
, end
, reg_info
))
5316 } /* while p1 < end */
5319 } /* alt_match_null_string_p */
5322 /* Deals with the ops common to group_match_null_string_p and
5323 alt_match_null_string_p.
5325 Sets P to one after the op and its arguments, if any. */
5328 common_op_match_null_string_p (p
, end
, reg_info
)
5329 unsigned char **p
, *end
;
5330 register_info_type
*reg_info
;
5335 unsigned char *p1
= *p
;
5337 switch ((re_opcode_t
) *p1
++)
5357 assert (reg_no
> 0 && reg_no
<= MAX_REGNUM
);
5358 ret
= group_match_null_string_p (&p1
, end
, reg_info
);
5360 /* Have to set this here in case we're checking a group which
5361 contains a group and a back reference to it. */
5363 if (REG_MATCH_NULL_STRING_P (reg_info
[reg_no
]) == MATCH_NULL_UNSET_VALUE
)
5364 REG_MATCH_NULL_STRING_P (reg_info
[reg_no
]) = ret
;
5370 /* If this is an optimized succeed_n for zero times, make the jump. */
5372 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
5380 /* Get to the number of times to succeed. */
5382 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
5387 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
5395 if (!REG_MATCH_NULL_STRING_P (reg_info
[*p1
]))
5403 /* All other opcodes mean we cannot match the empty string. */
5409 } /* common_op_match_null_string_p */
5412 /* Return zero if TRANSLATE[S1] and TRANSLATE[S2] are identical for LEN
5413 bytes; nonzero otherwise. */
5416 bcmp_translate (s1
, s2
, len
, translate
)
5417 const char *s1
, *s2
;
5419 RE_TRANSLATE_TYPE translate
;
5421 register const unsigned char *p1
= (const unsigned char *) s1
;
5422 register const unsigned char *p2
= (const unsigned char *) s2
;
5425 if (translate
[*p1
++] != translate
[*p2
++]) return 1;
5431 /* Entry points for GNU code. */
5433 /* re_compile_pattern is the GNU regular expression compiler: it
5434 compiles PATTERN (of length SIZE) and puts the result in BUFP.
5435 Returns 0 if the pattern was valid, otherwise an error string.
5437 Assumes the `allocated' (and perhaps `buffer') and `translate' fields
5438 are set in BUFP on entry.
5440 We call regex_compile to do the actual compilation. */
5443 re_compile_pattern (pattern
, length
, bufp
)
5444 const char *pattern
;
5446 struct re_pattern_buffer
*bufp
;
5450 /* GNU code is written to assume at least RE_NREGS registers will be set
5451 (and at least one extra will be -1). */
5452 bufp
->regs_allocated
= REGS_UNALLOCATED
;
5454 /* And GNU code determines whether or not to get register information
5455 by passing null for the REGS argument to re_match, etc., not by
5459 /* Match anchors at newline. */
5460 bufp
->newline_anchor
= 1;
5462 ret
= regex_compile (pattern
, length
, re_syntax_options
, bufp
);
5466 return gettext (re_error_msgid
[(int) ret
]);
5469 weak_alias (__re_compile_pattern
, re_compile_pattern
)
5472 /* Entry points compatible with 4.2 BSD regex library. We don't define
5473 them unless specifically requested. */
5475 #if defined _REGEX_RE_COMP || defined _LIBC
5477 /* BSD has one and only one pattern buffer. */
5478 static struct re_pattern_buffer re_comp_buf
;
5482 /* Make these definitions weak in libc, so POSIX programs can redefine
5483 these names if they don't use our functions, and still use
5484 regcomp/regexec below without link errors. */
5494 if (!re_comp_buf
.buffer
)
5495 return gettext ("No previous regular expression");
5499 if (!re_comp_buf
.buffer
)
5501 re_comp_buf
.buffer
= (unsigned char *) malloc (200);
5502 if (re_comp_buf
.buffer
== NULL
)
5503 return (char *) gettext (re_error_msgid
[(int) REG_ESPACE
]);
5504 re_comp_buf
.allocated
= 200;
5506 re_comp_buf
.fastmap
= (char *) malloc (1 << BYTEWIDTH
);
5507 if (re_comp_buf
.fastmap
== NULL
)
5508 return (char *) gettext (re_error_msgid
[(int) REG_ESPACE
]);
5511 /* Since `re_exec' always passes NULL for the `regs' argument, we
5512 don't need to initialize the pattern buffer fields which affect it. */
5514 /* Match anchors at newlines. */
5515 re_comp_buf
.newline_anchor
= 1;
5517 ret
= regex_compile (s
, strlen (s
), re_syntax_options
, &re_comp_buf
);
5522 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
5523 return (char *) gettext (re_error_msgid
[(int) ret
]);
5534 const int len
= strlen (s
);
5536 0 <= re_search (&re_comp_buf
, s
, len
, 0, len
, (struct re_registers
*) 0);
5539 #endif /* _REGEX_RE_COMP */
5541 /* POSIX.2 functions. Don't define these for Emacs. */
5545 /* regcomp takes a regular expression as a string and compiles it.
5547 PREG is a regex_t *. We do not expect any fields to be initialized,
5548 since POSIX says we shouldn't. Thus, we set
5550 `buffer' to the compiled pattern;
5551 `used' to the length of the compiled pattern;
5552 `syntax' to RE_SYNTAX_POSIX_EXTENDED if the
5553 REG_EXTENDED bit in CFLAGS is set; otherwise, to
5554 RE_SYNTAX_POSIX_BASIC;
5555 `newline_anchor' to REG_NEWLINE being set in CFLAGS;
5556 `fastmap' and `fastmap_accurate' to zero;
5557 `re_nsub' to the number of subexpressions in PATTERN.
5559 PATTERN is the address of the pattern string.
5561 CFLAGS is a series of bits which affect compilation.
5563 If REG_EXTENDED is set, we use POSIX extended syntax; otherwise, we
5564 use POSIX basic syntax.
5566 If REG_NEWLINE is set, then . and [^...] don't match newline.
5567 Also, regexec will try a match beginning after every newline.
5569 If REG_ICASE is set, then we considers upper- and lowercase
5570 versions of letters to be equivalent when matching.
5572 If REG_NOSUB is set, then when PREG is passed to regexec, that
5573 routine will report only success or failure, and nothing about the
5576 It returns 0 if it succeeds, nonzero if it doesn't. (See gnu-regex.h for
5577 the return codes and their meanings.) */
5580 regcomp (preg
, pattern
, cflags
)
5582 const char *pattern
;
5587 = (cflags
& REG_EXTENDED
) ?
5588 RE_SYNTAX_POSIX_EXTENDED
: RE_SYNTAX_POSIX_BASIC
;
5590 /* regex_compile will allocate the space for the compiled pattern. */
5592 preg
->allocated
= 0;
5595 /* Don't bother to use a fastmap when searching. This simplifies the
5596 REG_NEWLINE case: if we used a fastmap, we'd have to put all the
5597 characters after newlines into the fastmap. This way, we just try
5601 if (cflags
& REG_ICASE
)
5606 = (RE_TRANSLATE_TYPE
) malloc (CHAR_SET_SIZE
5607 * sizeof (*(RE_TRANSLATE_TYPE
)0));
5608 if (preg
->translate
== NULL
)
5609 return (int) REG_ESPACE
;
5611 /* Map uppercase characters to corresponding lowercase ones. */
5612 for (i
= 0; i
< CHAR_SET_SIZE
; i
++)
5613 preg
->translate
[i
] = ISUPPER (i
) ? tolower (i
) : i
;
5616 preg
->translate
= NULL
;
5618 /* If REG_NEWLINE is set, newlines are treated differently. */
5619 if (cflags
& REG_NEWLINE
)
5620 { /* REG_NEWLINE implies neither . nor [^...] match newline. */
5621 syntax
&= ~RE_DOT_NEWLINE
;
5622 syntax
|= RE_HAT_LISTS_NOT_NEWLINE
;
5623 /* It also changes the matching behavior. */
5624 preg
->newline_anchor
= 1;
5627 preg
->newline_anchor
= 0;
5629 preg
->no_sub
= !!(cflags
& REG_NOSUB
);
5631 /* POSIX says a null character in the pattern terminates it, so we
5632 can use strlen here in compiling the pattern. */
5633 ret
= regex_compile (pattern
, strlen (pattern
), syntax
, preg
);
5635 /* POSIX doesn't distinguish between an unmatched open-group and an
5636 unmatched close-group: both are REG_EPAREN. */
5637 if (ret
== REG_ERPAREN
) ret
= REG_EPAREN
;
5642 weak_alias (__regcomp
, regcomp
)
5646 /* regexec searches for a given pattern, specified by PREG, in the
5649 If NMATCH is zero or REG_NOSUB was set in the cflags argument to
5650 `regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at
5651 least NMATCH elements, and we set them to the offsets of the
5652 corresponding matched substrings.
5654 EFLAGS specifies `execution flags' which affect matching: if
5655 REG_NOTBOL is set, then ^ does not match at the beginning of the
5656 string; if REG_NOTEOL is set, then $ does not match at the end.
5658 We return 0 if we find a match and REG_NOMATCH if not. */
5661 regexec (preg
, string
, nmatch
, pmatch
, eflags
)
5662 const regex_t
*preg
;
5665 regmatch_t pmatch
[];
5669 struct re_registers regs
;
5670 regex_t private_preg
;
5671 int len
= strlen (string
);
5672 boolean want_reg_info
= !preg
->no_sub
&& nmatch
> 0;
5674 private_preg
= *preg
;
5676 private_preg
.not_bol
= !!(eflags
& REG_NOTBOL
);
5677 private_preg
.not_eol
= !!(eflags
& REG_NOTEOL
);
5679 /* The user has told us exactly how many registers to return
5680 information about, via `nmatch'. We have to pass that on to the
5681 matching routines. */
5682 private_preg
.regs_allocated
= REGS_FIXED
;
5686 regs
.num_regs
= nmatch
;
5687 regs
.start
= TALLOC (nmatch
, regoff_t
);
5688 regs
.end
= TALLOC (nmatch
, regoff_t
);
5689 if (regs
.start
== NULL
|| regs
.end
== NULL
)
5690 return (int) REG_NOMATCH
;
5693 /* Perform the searching operation. */
5694 ret
= re_search (&private_preg
, string
, len
,
5695 /* start: */ 0, /* range: */ len
,
5696 want_reg_info
? ®s
: (struct re_registers
*) 0);
5698 /* Copy the register information to the POSIX structure. */
5705 for (r
= 0; r
< nmatch
; r
++)
5707 pmatch
[r
].rm_so
= regs
.start
[r
];
5708 pmatch
[r
].rm_eo
= regs
.end
[r
];
5712 /* If we needed the temporary register info, free the space now. */
5717 /* We want zero return to mean success, unlike `re_search'. */
5718 return ret
>= 0 ? (int) REG_NOERROR
: (int) REG_NOMATCH
;
5721 weak_alias (__regexec
, regexec
)
5725 /* Returns a message corresponding to an error code, ERRCODE, returned
5726 from either regcomp or regexec. We don't use PREG here. */
5729 __regerror (errcode
, preg
, errbuf
, errbuf_size
)
5731 const regex_t
*preg
;
5739 || errcode
>= (int) (sizeof (re_error_msgid
)
5740 / sizeof (re_error_msgid
[0])))
5741 /* Only error codes returned by the rest of the code should be passed
5742 to this routine. If we are given anything else, or if other regex
5743 code generates an invalid error code, then the program has a bug.
5744 Dump core so we can fix it. */
5747 msg
= gettext (re_error_msgid
[errcode
]);
5749 msg_size
= strlen (msg
) + 1; /* Includes the null. */
5751 if (errbuf_size
!= 0)
5753 if (msg_size
> errbuf_size
)
5755 #if defined HAVE_MEMPCPY || defined _LIBC
5756 *((char *) __mempcpy (errbuf
, msg
, errbuf_size
- 1)) = '\0';
5758 memcpy (errbuf
, msg
, errbuf_size
- 1);
5759 errbuf
[errbuf_size
- 1] = 0;
5763 memcpy (errbuf
, msg
, msg_size
);
5769 weak_alias (__regerror
, regerror
)
5773 /* Free dynamically allocated space used by PREG. */
5779 if (preg
->buffer
!= NULL
)
5780 free (preg
->buffer
);
5781 preg
->buffer
= NULL
;
5783 preg
->allocated
= 0;
5786 if (preg
->fastmap
!= NULL
)
5787 free (preg
->fastmap
);
5788 preg
->fastmap
= NULL
;
5789 preg
->fastmap_accurate
= 0;
5791 if (preg
->translate
!= NULL
)
5792 free (preg
->translate
);
5793 preg
->translate
= NULL
;
5796 weak_alias (__regfree
, regfree
)
5799 #endif /* not emacs */