1 /* Extended regular expression matching and search library,
3 (Implements POSIX draft P1003.2/D11.2, except for some of the
4 internationalization features.)
5 Copyright (C) 1993, 94, 95, 96, 97, 98 Free Software Foundation, Inc.
7 NOTE: The canonical source of this file is maintained with the
8 GNU C Library. Bugs can be reported to bug-glibc@prep.ai.mit.edu.
10 This program is free software; you can redistribute it and/or modify it
11 under the terms of the GNU General Public License as published by the
12 Free Software Foundation; either version 2, or (at your option) any
15 This program is distributed in the hope that it will be useful,
16 but WITHOUT ANY WARRANTY; without even the implied warranty of
17 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
18 GNU General Public License for more details.
20 You should have received a copy of the GNU General Public License
21 along with this program; if not, write to the Free Software Foundation,
22 Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. */
24 /* AIX requires this to be the first thing in the file. */
25 #if defined _AIX && !defined REGEX_MALLOC
37 # if defined __GNUC__ || (defined __STDC__ && __STDC__)
38 # define PARAMS(args) args
40 # define PARAMS(args) ()
42 #endif /* Not PARAMS. */
44 #if defined STDC_HEADERS && !defined emacs
47 /* We need this for `gnu-regex.h', and perhaps for the Emacs include files. */
48 # include <sys/types.h>
51 /* For platform which support the ISO C amendement 1 functionality we
52 support user defined character classes. */
53 #if defined _LIBC || (defined HAVE_WCTYPE_H && defined HAVE_WCHAR_H)
54 /* Solaris 2.5 has a bug: <wchar.h> must be included before <wctype.h>. */
59 /* This is for other GNU distributions with internationalized messages. */
60 #if HAVE_LIBINTL_H || defined _LIBC
63 # define gettext(msgid) (msgid)
67 /* This define is so xgettext can find the internationalizable
69 # define gettext_noop(String) String
72 /* The `emacs' switch turns on certain matching commands
73 that make sense only in Emacs. */
82 /* If we are not linking with Emacs proper,
83 we can't use the relocating allocator
84 even if config.h says that we can. */
87 # if defined STDC_HEADERS || defined _LIBC
94 /* When used in Emacs's lib-src, we need to get bzero and bcopy somehow.
95 If nothing else has been done, use the method below. */
96 # ifdef INHIBIT_STRING_HEADER
97 # if !(defined HAVE_BZERO && defined HAVE_BCOPY)
98 # if !defined bzero && !defined bcopy
99 # undef INHIBIT_STRING_HEADER
104 /* This is the normal way of making sure we have a bcopy and a bzero.
105 This is used in most programs--a few other programs avoid this
106 by defining INHIBIT_STRING_HEADER. */
107 # ifndef INHIBIT_STRING_HEADER
108 # if defined HAVE_STRING_H || defined STDC_HEADERS || defined _LIBC
112 # define bzero(s, n) (memset (s, '\0', n), (s))
114 # define bzero(s, n) __bzero (s, n)
118 # include <strings.h>
120 # define memcmp(s1, s2, n) bcmp (s1, s2, n)
123 # define memcpy(d, s, n) (bcopy (s, d, n), (d))
128 /* Define the syntax stuff for \<, \>, etc. */
130 /* This must be nonzero for the wordchar and notwordchar pattern
131 commands in re_match_2. */
136 # ifdef SWITCH_ENUM_BUG
137 # define SWITCH_ENUM_CAST(x) ((int)(x))
139 # define SWITCH_ENUM_CAST(x) (x)
142 /* How many characters in the character set. */
143 # define CHAR_SET_SIZE 256
147 extern char *re_syntax_table
;
149 # else /* not SYNTAX_TABLE */
151 static char re_syntax_table
[CHAR_SET_SIZE
];
162 bzero (re_syntax_table
, sizeof re_syntax_table
);
164 for (c
= 'a'; c
<= 'z'; c
++)
165 re_syntax_table
[c
] = Sword
;
167 for (c
= 'A'; c
<= 'Z'; c
++)
168 re_syntax_table
[c
] = Sword
;
170 for (c
= '0'; c
<= '9'; c
++)
171 re_syntax_table
[c
] = Sword
;
173 re_syntax_table
['_'] = Sword
;
178 # endif /* not SYNTAX_TABLE */
180 # define SYNTAX(c) re_syntax_table[c]
182 #endif /* not emacs */
184 /* Get the interface, including the syntax bits. */
185 /* CYGNUS LOCAL: call it gnu-regex.h, not regex.h, to avoid name conflicts */
186 #include "gnu-regex.h"
188 /* isalpha etc. are used for the character classes. */
191 /* Jim Meyering writes:
193 "... Some ctype macros are valid only for character codes that
194 isascii says are ASCII (SGI's IRIX-4.0.5 is one such system --when
195 using /bin/cc or gcc but without giving an ansi option). So, all
196 ctype uses should be through macros like ISPRINT... If
197 STDC_HEADERS is defined, then autoconf has verified that the ctype
198 macros don't need to be guarded with references to isascii. ...
199 Defining isascii to 1 should let any compiler worth its salt
200 eliminate the && through constant folding."
201 Solaris defines some of these symbols so we must undefine them first. */
204 #if defined STDC_HEADERS || (!defined isascii && !defined HAVE_ISASCII)
205 # define ISASCII(c) 1
207 # define ISASCII(c) isascii(c)
211 # define ISBLANK(c) (ISASCII (c) && isblank (c))
213 # define ISBLANK(c) ((c) == ' ' || (c) == '\t')
216 # define ISGRAPH(c) (ISASCII (c) && isgraph (c))
218 # define ISGRAPH(c) (ISASCII (c) && isprint (c) && !isspace (c))
222 #define ISPRINT(c) (ISASCII (c) && isprint (c))
223 #define ISDIGIT(c) (ISASCII (c) && isdigit (c))
224 #define ISALNUM(c) (ISASCII (c) && isalnum (c))
225 #define ISALPHA(c) (ISASCII (c) && isalpha (c))
226 #define ISCNTRL(c) (ISASCII (c) && iscntrl (c))
227 #define ISLOWER(c) (ISASCII (c) && islower (c))
228 #define ISPUNCT(c) (ISASCII (c) && ispunct (c))
229 #define ISSPACE(c) (ISASCII (c) && isspace (c))
230 #define ISUPPER(c) (ISASCII (c) && isupper (c))
231 #define ISXDIGIT(c) (ISASCII (c) && isxdigit (c))
234 # define NULL (void *)0
237 /* We remove any previous definition of `SIGN_EXTEND_CHAR',
238 since ours (we hope) works properly with all combinations of
239 machines, compilers, `char' and `unsigned char' argument types.
240 (Per Bothner suggested the basic approach.) */
241 #undef SIGN_EXTEND_CHAR
243 # define SIGN_EXTEND_CHAR(c) ((signed char) (c))
244 #else /* not __STDC__ */
245 /* As in Harbison and Steele. */
246 # define SIGN_EXTEND_CHAR(c) ((((unsigned char) (c)) ^ 128) - 128)
249 /* Should we use malloc or alloca? If REGEX_MALLOC is not defined, we
250 use `alloca' instead of `malloc'. This is because using malloc in
251 re_search* or re_match* could cause memory leaks when C-g is used in
252 Emacs; also, malloc is slower and causes storage fragmentation. On
253 the other hand, malloc is more portable, and easier to debug.
255 Because we sometimes use alloca, some routines have to be macros,
256 not functions -- `alloca'-allocated space disappears at the end of the
257 function it is called in. */
261 # define REGEX_ALLOCATE malloc
262 # define REGEX_REALLOCATE(source, osize, nsize) realloc (source, nsize)
263 # define REGEX_FREE free
265 #else /* not REGEX_MALLOC */
267 /* Emacs already defines alloca, sometimes. */
270 /* Make alloca work the best possible way. */
272 # define alloca __builtin_alloca
273 # else /* not __GNUC__ */
276 # endif /* HAVE_ALLOCA_H */
277 # endif /* not __GNUC__ */
279 # endif /* not alloca */
281 # define REGEX_ALLOCATE alloca
283 /* Assumes a `char *destination' variable. */
284 # define REGEX_REALLOCATE(source, osize, nsize) \
285 (destination = (char *) alloca (nsize), \
286 memcpy (destination, source, osize))
288 /* No need to do anything to free, after alloca. */
289 # define REGEX_FREE(arg) ((void)0) /* Do nothing! But inhibit gcc warning. */
291 #endif /* not REGEX_MALLOC */
293 /* Define how to allocate the failure stack. */
295 #if defined REL_ALLOC && defined REGEX_MALLOC
297 # define REGEX_ALLOCATE_STACK(size) \
298 r_alloc (&failure_stack_ptr, (size))
299 # define REGEX_REALLOCATE_STACK(source, osize, nsize) \
300 r_re_alloc (&failure_stack_ptr, (nsize))
301 # define REGEX_FREE_STACK(ptr) \
302 r_alloc_free (&failure_stack_ptr)
304 #else /* not using relocating allocator */
308 # define REGEX_ALLOCATE_STACK malloc
309 # define REGEX_REALLOCATE_STACK(source, osize, nsize) realloc (source, nsize)
310 # define REGEX_FREE_STACK free
312 # else /* not REGEX_MALLOC */
314 # define REGEX_ALLOCATE_STACK alloca
316 # define REGEX_REALLOCATE_STACK(source, osize, nsize) \
317 REGEX_REALLOCATE (source, osize, nsize)
318 /* No need to explicitly free anything. */
319 # define REGEX_FREE_STACK(arg)
321 # endif /* not REGEX_MALLOC */
322 #endif /* not using relocating allocator */
325 /* True if `size1' is non-NULL and PTR is pointing anywhere inside
326 `string1' or just past its end. This works if PTR is NULL, which is
328 #define FIRST_STRING_P(ptr) \
329 (size1 && string1 <= (ptr) && (ptr) <= string1 + size1)
331 /* (Re)Allocate N items of type T using malloc, or fail. */
332 #define TALLOC(n, t) ((t *) malloc ((n) * sizeof (t)))
333 #define RETALLOC(addr, n, t) ((addr) = (t *) realloc (addr, (n) * sizeof (t)))
334 #define RETALLOC_IF(addr, n, t) \
335 if (addr) RETALLOC((addr), (n), t); else (addr) = TALLOC ((n), t)
336 #define REGEX_TALLOC(n, t) ((t *) REGEX_ALLOCATE ((n) * sizeof (t)))
338 #define BYTEWIDTH 8 /* In bits. */
340 #define STREQ(s1, s2) ((strcmp (s1, s2) == 0))
344 #define MAX(a, b) ((a) > (b) ? (a) : (b))
345 #define MIN(a, b) ((a) < (b) ? (a) : (b))
347 typedef char boolean
;
351 static int re_match_2_internal
PARAMS ((struct re_pattern_buffer
*bufp
,
352 const char *string1
, int size1
,
353 const char *string2
, int size2
,
355 struct re_registers
*regs
,
358 /* These are the command codes that appear in compiled regular
359 expressions. Some opcodes are followed by argument bytes. A
360 command code can specify any interpretation whatsoever for its
361 arguments. Zero bytes may appear in the compiled regular expression. */
367 /* Succeed right away--no more backtracking. */
370 /* Followed by one byte giving n, then by n literal bytes. */
373 /* Matches any (more or less) character. */
376 /* Matches any one char belonging to specified set. First
377 following byte is number of bitmap bytes. Then come bytes
378 for a bitmap saying which chars are in. Bits in each byte
379 are ordered low-bit-first. A character is in the set if its
380 bit is 1. A character too large to have a bit in the map is
381 automatically not in the set. */
384 /* Same parameters as charset, but match any character that is
385 not one of those specified. */
388 /* Start remembering the text that is matched, for storing in a
389 register. Followed by one byte with the register number, in
390 the range 0 to one less than the pattern buffer's re_nsub
391 field. Then followed by one byte with the number of groups
392 inner to this one. (This last has to be part of the
393 start_memory only because we need it in the on_failure_jump
397 /* Stop remembering the text that is matched and store it in a
398 memory register. Followed by one byte with the register
399 number, in the range 0 to one less than `re_nsub' in the
400 pattern buffer, and one byte with the number of inner groups,
401 just like `start_memory'. (We need the number of inner
402 groups here because we don't have any easy way of finding the
403 corresponding start_memory when we're at a stop_memory.) */
406 /* Match a duplicate of something remembered. Followed by one
407 byte containing the register number. */
410 /* Fail unless at beginning of line. */
413 /* Fail unless at end of line. */
416 /* Succeeds if at beginning of buffer (if emacs) or at beginning
417 of string to be matched (if not). */
420 /* Analogously, for end of buffer/string. */
423 /* Followed by two byte relative address to which to jump. */
426 /* Same as jump, but marks the end of an alternative. */
429 /* Followed by two-byte relative address of place to resume at
430 in case of failure. */
433 /* Like on_failure_jump, but pushes a placeholder instead of the
434 current string position when executed. */
435 on_failure_keep_string_jump
,
437 /* Throw away latest failure point and then jump to following
438 two-byte relative address. */
441 /* Change to pop_failure_jump if know won't have to backtrack to
442 match; otherwise change to jump. This is used to jump
443 back to the beginning of a repeat. If what follows this jump
444 clearly won't match what the repeat does, such that we can be
445 sure that there is no use backtracking out of repetitions
446 already matched, then we change it to a pop_failure_jump.
447 Followed by two-byte address. */
450 /* Jump to following two-byte address, and push a dummy failure
451 point. This failure point will be thrown away if an attempt
452 is made to use it for a failure. A `+' construct makes this
453 before the first repeat. Also used as an intermediary kind
454 of jump when compiling an alternative. */
457 /* Push a dummy failure point and continue. Used at the end of
461 /* Followed by two-byte relative address and two-byte number n.
462 After matching N times, jump to the address upon failure. */
465 /* Followed by two-byte relative address, and two-byte number n.
466 Jump to the address N times, then fail. */
469 /* Set the following two-byte relative address to the
470 subsequent two-byte number. The address *includes* the two
474 wordchar
, /* Matches any word-constituent character. */
475 notwordchar
, /* Matches any char that is not a word-constituent. */
477 wordbeg
, /* Succeeds if at word beginning. */
478 wordend
, /* Succeeds if at word end. */
480 wordbound
, /* Succeeds if at a word boundary. */
481 notwordbound
/* Succeeds if not at a word boundary. */
484 ,before_dot
, /* Succeeds if before point. */
485 at_dot
, /* Succeeds if at point. */
486 after_dot
, /* Succeeds if after point. */
488 /* Matches any character whose syntax is specified. Followed by
489 a byte which contains a syntax code, e.g., Sword. */
492 /* Matches any character whose syntax is not that specified. */
497 /* Common operations on the compiled pattern. */
499 /* Store NUMBER in two contiguous bytes starting at DESTINATION. */
501 #define STORE_NUMBER(destination, number) \
503 (destination)[0] = (number) & 0377; \
504 (destination)[1] = (number) >> 8; \
507 /* Same as STORE_NUMBER, except increment DESTINATION to
508 the byte after where the number is stored. Therefore, DESTINATION
509 must be an lvalue. */
511 #define STORE_NUMBER_AND_INCR(destination, number) \
513 STORE_NUMBER (destination, number); \
514 (destination) += 2; \
517 /* Put into DESTINATION a number stored in two contiguous bytes starting
520 #define EXTRACT_NUMBER(destination, source) \
522 (destination) = *(source) & 0377; \
523 (destination) += SIGN_EXTEND_CHAR (*((source) + 1)) << 8; \
527 static void extract_number
_RE_ARGS ((int *dest
, unsigned char *source
));
529 extract_number (dest
, source
)
531 unsigned char *source
;
533 int temp
= SIGN_EXTEND_CHAR (*(source
+ 1));
534 *dest
= *source
& 0377;
538 # ifndef EXTRACT_MACROS /* To debug the macros. */
539 # undef EXTRACT_NUMBER
540 # define EXTRACT_NUMBER(dest, src) extract_number (&dest, src)
541 # endif /* not EXTRACT_MACROS */
545 /* Same as EXTRACT_NUMBER, except increment SOURCE to after the number.
546 SOURCE must be an lvalue. */
548 #define EXTRACT_NUMBER_AND_INCR(destination, source) \
550 EXTRACT_NUMBER (destination, source); \
555 static void extract_number_and_incr
_RE_ARGS ((int *destination
,
556 unsigned char **source
));
558 extract_number_and_incr (destination
, source
)
560 unsigned char **source
;
562 extract_number (destination
, *source
);
566 # ifndef EXTRACT_MACROS
567 # undef EXTRACT_NUMBER_AND_INCR
568 # define EXTRACT_NUMBER_AND_INCR(dest, src) \
569 extract_number_and_incr (&dest, &src)
570 # endif /* not EXTRACT_MACROS */
574 /* If DEBUG is defined, Regex prints many voluminous messages about what
575 it is doing (if the variable `debug' is nonzero). If linked with the
576 main program in `iregex.c', you can enter patterns and strings
577 interactively. And if linked with the main program in `main.c' and
578 the other test files, you can run the already-written tests. */
582 /* We use standard I/O for debugging. */
585 /* It is useful to test things that ``must'' be true when debugging. */
588 static int debug
= 0;
590 # define DEBUG_STATEMENT(e) e
591 # define DEBUG_PRINT1(x) if (debug) printf (x)
592 # define DEBUG_PRINT2(x1, x2) if (debug) printf (x1, x2)
593 # define DEBUG_PRINT3(x1, x2, x3) if (debug) printf (x1, x2, x3)
594 # define DEBUG_PRINT4(x1, x2, x3, x4) if (debug) printf (x1, x2, x3, x4)
595 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) \
596 if (debug) print_partial_compiled_pattern (s, e)
597 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) \
598 if (debug) print_double_string (w, s1, sz1, s2, sz2)
601 /* Print the fastmap in human-readable form. */
604 print_fastmap (fastmap
)
607 unsigned was_a_range
= 0;
610 while (i
< (1 << BYTEWIDTH
))
616 while (i
< (1 << BYTEWIDTH
) && fastmap
[i
])
632 /* Print a compiled pattern string in human-readable form, starting at
633 the START pointer into it and ending just before the pointer END. */
636 print_partial_compiled_pattern (start
, end
)
637 unsigned char *start
;
642 unsigned char *p
= start
;
643 unsigned char *pend
= end
;
651 /* Loop over pattern commands. */
654 printf ("%d:\t", p
- start
);
656 switch ((re_opcode_t
) *p
++)
664 printf ("/exactn/%d", mcnt
);
675 printf ("/start_memory/%d/%d", mcnt
, *p
++);
680 printf ("/stop_memory/%d/%d", mcnt
, *p
++);
684 printf ("/duplicate/%d", *p
++);
694 register int c
, last
= -100;
695 register int in_range
= 0;
697 printf ("/charset [%s",
698 (re_opcode_t
) *(p
- 1) == charset_not
? "^" : "");
700 assert (p
+ *p
< pend
);
702 for (c
= 0; c
< 256; c
++)
704 && (p
[1 + (c
/8)] & (1 << (c
% 8))))
706 /* Are we starting a range? */
707 if (last
+ 1 == c
&& ! in_range
)
712 /* Have we broken a range? */
713 else if (last
+ 1 != c
&& in_range
)
742 case on_failure_jump
:
743 extract_number_and_incr (&mcnt
, &p
);
744 printf ("/on_failure_jump to %d", p
+ mcnt
- start
);
747 case on_failure_keep_string_jump
:
748 extract_number_and_incr (&mcnt
, &p
);
749 printf ("/on_failure_keep_string_jump to %d", p
+ mcnt
- start
);
752 case dummy_failure_jump
:
753 extract_number_and_incr (&mcnt
, &p
);
754 printf ("/dummy_failure_jump to %d", p
+ mcnt
- start
);
757 case push_dummy_failure
:
758 printf ("/push_dummy_failure");
762 extract_number_and_incr (&mcnt
, &p
);
763 printf ("/maybe_pop_jump to %d", p
+ mcnt
- start
);
766 case pop_failure_jump
:
767 extract_number_and_incr (&mcnt
, &p
);
768 printf ("/pop_failure_jump to %d", p
+ mcnt
- start
);
772 extract_number_and_incr (&mcnt
, &p
);
773 printf ("/jump_past_alt to %d", p
+ mcnt
- start
);
777 extract_number_and_incr (&mcnt
, &p
);
778 printf ("/jump to %d", p
+ mcnt
- start
);
782 extract_number_and_incr (&mcnt
, &p
);
784 extract_number_and_incr (&mcnt2
, &p
);
785 printf ("/succeed_n to %d, %d times", p1
- start
, mcnt2
);
789 extract_number_and_incr (&mcnt
, &p
);
791 extract_number_and_incr (&mcnt2
, &p
);
792 printf ("/jump_n to %d, %d times", p1
- start
, mcnt2
);
796 extract_number_and_incr (&mcnt
, &p
);
798 extract_number_and_incr (&mcnt2
, &p
);
799 printf ("/set_number_at location %d to %d", p1
- start
, mcnt2
);
803 printf ("/wordbound");
807 printf ("/notwordbound");
819 printf ("/before_dot");
827 printf ("/after_dot");
831 printf ("/syntaxspec");
833 printf ("/%d", mcnt
);
837 printf ("/notsyntaxspec");
839 printf ("/%d", mcnt
);
844 printf ("/wordchar");
848 printf ("/notwordchar");
860 printf ("?%d", *(p
-1));
866 printf ("%d:\tend of pattern.\n", p
- start
);
871 print_compiled_pattern (bufp
)
872 struct re_pattern_buffer
*bufp
;
874 unsigned char *buffer
= bufp
->buffer
;
876 print_partial_compiled_pattern (buffer
, buffer
+ bufp
->used
);
877 printf ("%ld bytes used/%ld bytes allocated.\n",
878 bufp
->used
, bufp
->allocated
);
880 if (bufp
->fastmap_accurate
&& bufp
->fastmap
)
882 printf ("fastmap: ");
883 print_fastmap (bufp
->fastmap
);
886 printf ("re_nsub: %d\t", bufp
->re_nsub
);
887 printf ("regs_alloc: %d\t", bufp
->regs_allocated
);
888 printf ("can_be_null: %d\t", bufp
->can_be_null
);
889 printf ("newline_anchor: %d\n", bufp
->newline_anchor
);
890 printf ("no_sub: %d\t", bufp
->no_sub
);
891 printf ("not_bol: %d\t", bufp
->not_bol
);
892 printf ("not_eol: %d\t", bufp
->not_eol
);
893 printf ("syntax: %lx\n", bufp
->syntax
);
894 /* Perhaps we should print the translate table? */
899 print_double_string (where
, string1
, size1
, string2
, size2
)
912 if (FIRST_STRING_P (where
))
914 for (this_char
= where
- string1
; this_char
< size1
; this_char
++)
915 putchar (string1
[this_char
]);
920 for (this_char
= where
- string2
; this_char
< size2
; this_char
++)
921 putchar (string2
[this_char
]);
932 #else /* not DEBUG */
937 # define DEBUG_STATEMENT(e)
938 # define DEBUG_PRINT1(x)
939 # define DEBUG_PRINT2(x1, x2)
940 # define DEBUG_PRINT3(x1, x2, x3)
941 # define DEBUG_PRINT4(x1, x2, x3, x4)
942 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e)
943 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2)
945 #endif /* not DEBUG */
947 /* Set by `re_set_syntax' to the current regexp syntax to recognize. Can
948 also be assigned to arbitrarily: each pattern buffer stores its own
949 syntax, so it can be changed between regex compilations. */
950 /* This has no initializer because initialized variables in Emacs
951 become read-only after dumping. */
952 reg_syntax_t re_syntax_options
;
955 /* Specify the precise syntax of regexps for compilation. This provides
956 for compatibility for various utilities which historically have
957 different, incompatible syntaxes.
959 The argument SYNTAX is a bit mask comprised of the various bits
960 defined in gnu-regex.h. We return the old syntax. */
963 re_set_syntax (syntax
)
966 reg_syntax_t ret
= re_syntax_options
;
968 re_syntax_options
= syntax
;
970 if (syntax
& RE_DEBUG
)
972 else if (debug
) /* was on but now is not */
978 weak_alias (__re_set_syntax
, re_set_syntax
)
981 /* This table gives an error message for each of the error codes listed
982 in gnu-regex.h. Obviously the order here has to be same as there.
983 POSIX doesn't require that we do anything for REG_NOERROR,
984 but why not be nice? */
986 static const char *re_error_msgid
[] =
988 gettext_noop ("Success"), /* REG_NOERROR */
989 gettext_noop ("No match"), /* REG_NOMATCH */
990 gettext_noop ("Invalid regular expression"), /* REG_BADPAT */
991 gettext_noop ("Invalid collation character"), /* REG_ECOLLATE */
992 gettext_noop ("Invalid character class name"), /* REG_ECTYPE */
993 gettext_noop ("Trailing backslash"), /* REG_EESCAPE */
994 gettext_noop ("Invalid back reference"), /* REG_ESUBREG */
995 gettext_noop ("Unmatched [ or [^"), /* REG_EBRACK */
996 gettext_noop ("Unmatched ( or \\("), /* REG_EPAREN */
997 gettext_noop ("Unmatched \\{"), /* REG_EBRACE */
998 gettext_noop ("Invalid content of \\{\\}"), /* REG_BADBR */
999 gettext_noop ("Invalid range end"), /* REG_ERANGE */
1000 gettext_noop ("Memory exhausted"), /* REG_ESPACE */
1001 gettext_noop ("Invalid preceding regular expression"), /* REG_BADRPT */
1002 gettext_noop ("Premature end of regular expression"), /* REG_EEND */
1003 gettext_noop ("Regular expression too big"), /* REG_ESIZE */
1004 gettext_noop ("Unmatched ) or \\)"), /* REG_ERPAREN */
1007 /* Avoiding alloca during matching, to placate r_alloc. */
1009 /* Define MATCH_MAY_ALLOCATE unless we need to make sure that the
1010 searching and matching functions should not call alloca. On some
1011 systems, alloca is implemented in terms of malloc, and if we're
1012 using the relocating allocator routines, then malloc could cause a
1013 relocation, which might (if the strings being searched are in the
1014 ralloc heap) shift the data out from underneath the regexp
1017 Here's another reason to avoid allocation: Emacs
1018 processes input from X in a signal handler; processing X input may
1019 call malloc; if input arrives while a matching routine is calling
1020 malloc, then we're scrod. But Emacs can't just block input while
1021 calling matching routines; then we don't notice interrupts when
1022 they come in. So, Emacs blocks input around all regexp calls
1023 except the matching calls, which it leaves unprotected, in the
1024 faith that they will not malloc. */
1026 /* Normally, this is fine. */
1027 #define MATCH_MAY_ALLOCATE
1029 /* When using GNU C, we are not REALLY using the C alloca, no matter
1030 what config.h may say. So don't take precautions for it. */
1035 /* The match routines may not allocate if (1) they would do it with malloc
1036 and (2) it's not safe for them to use malloc.
1037 Note that if REL_ALLOC is defined, matching would not use malloc for the
1038 failure stack, but we would still use it for the register vectors;
1039 so REL_ALLOC should not affect this. */
1040 #if (defined C_ALLOCA || defined REGEX_MALLOC) && defined emacs
1041 # undef MATCH_MAY_ALLOCATE
1045 /* Failure stack declarations and macros; both re_compile_fastmap and
1046 re_match_2 use a failure stack. These have to be macros because of
1047 REGEX_ALLOCATE_STACK. */
1050 /* Number of failure points for which to initially allocate space
1051 when matching. If this number is exceeded, we allocate more
1052 space, so it is not a hard limit. */
1053 #ifndef INIT_FAILURE_ALLOC
1054 # define INIT_FAILURE_ALLOC 5
1057 /* Roughly the maximum number of failure points on the stack. Would be
1058 exactly that if always used MAX_FAILURE_ITEMS items each time we failed.
1059 This is a variable only so users of regex can assign to it; we never
1060 change it ourselves. */
1064 # if defined MATCH_MAY_ALLOCATE
1065 /* 4400 was enough to cause a crash on Alpha OSF/1,
1066 whose default stack limit is 2mb. */
1067 long int re_max_failures
= 4000;
1069 long int re_max_failures
= 2000;
1072 union fail_stack_elt
1074 unsigned char *pointer
;
1078 typedef union fail_stack_elt fail_stack_elt_t
;
1082 fail_stack_elt_t
*stack
;
1083 unsigned long int size
;
1084 unsigned long int avail
; /* Offset of next open position. */
1087 #else /* not INT_IS_16BIT */
1089 # if defined MATCH_MAY_ALLOCATE
1090 /* 4400 was enough to cause a crash on Alpha OSF/1,
1091 whose default stack limit is 2mb. */
1092 int re_max_failures
= 20000;
1094 int re_max_failures
= 2000;
1097 union fail_stack_elt
1099 unsigned char *pointer
;
1103 typedef union fail_stack_elt fail_stack_elt_t
;
1107 fail_stack_elt_t
*stack
;
1109 unsigned avail
; /* Offset of next open position. */
1112 #endif /* INT_IS_16BIT */
1114 #define FAIL_STACK_EMPTY() (fail_stack.avail == 0)
1115 #define FAIL_STACK_PTR_EMPTY() (fail_stack_ptr->avail == 0)
1116 #define FAIL_STACK_FULL() (fail_stack.avail == fail_stack.size)
1119 /* Define macros to initialize and free the failure stack.
1120 Do `return -2' if the alloc fails. */
1122 #ifdef MATCH_MAY_ALLOCATE
1123 # define INIT_FAIL_STACK() \
1125 fail_stack.stack = (fail_stack_elt_t *) \
1126 REGEX_ALLOCATE_STACK (INIT_FAILURE_ALLOC * sizeof (fail_stack_elt_t)); \
1128 if (fail_stack.stack == NULL) \
1131 fail_stack.size = INIT_FAILURE_ALLOC; \
1132 fail_stack.avail = 0; \
1135 # define RESET_FAIL_STACK() REGEX_FREE_STACK (fail_stack.stack)
1137 # define INIT_FAIL_STACK() \
1139 fail_stack.avail = 0; \
1142 # define RESET_FAIL_STACK()
1146 /* Double the size of FAIL_STACK, up to approximately `re_max_failures' items.
1148 Return 1 if succeeds, and 0 if either ran out of memory
1149 allocating space for it or it was already too large.
1151 REGEX_REALLOCATE_STACK requires `destination' be declared. */
1153 #define DOUBLE_FAIL_STACK(fail_stack) \
1154 ((fail_stack).size > (unsigned) (re_max_failures * MAX_FAILURE_ITEMS) \
1156 : ((fail_stack).stack = (fail_stack_elt_t *) \
1157 REGEX_REALLOCATE_STACK ((fail_stack).stack, \
1158 (fail_stack).size * sizeof (fail_stack_elt_t), \
1159 ((fail_stack).size << 1) * sizeof (fail_stack_elt_t)), \
1161 (fail_stack).stack == NULL \
1163 : ((fail_stack).size <<= 1, \
1167 /* Push pointer POINTER on FAIL_STACK.
1168 Return 1 if was able to do so and 0 if ran out of memory allocating
1170 #define PUSH_PATTERN_OP(POINTER, FAIL_STACK) \
1171 ((FAIL_STACK_FULL () \
1172 && !DOUBLE_FAIL_STACK (FAIL_STACK)) \
1174 : ((FAIL_STACK).stack[(FAIL_STACK).avail++].pointer = POINTER, \
1177 /* Push a pointer value onto the failure stack.
1178 Assumes the variable `fail_stack'. Probably should only
1179 be called from within `PUSH_FAILURE_POINT'. */
1180 #define PUSH_FAILURE_POINTER(item) \
1181 fail_stack.stack[fail_stack.avail++].pointer = (unsigned char *) (item)
1183 /* This pushes an integer-valued item onto the failure stack.
1184 Assumes the variable `fail_stack'. Probably should only
1185 be called from within `PUSH_FAILURE_POINT'. */
1186 #define PUSH_FAILURE_INT(item) \
1187 fail_stack.stack[fail_stack.avail++].integer = (item)
1189 /* Push a fail_stack_elt_t value onto the failure stack.
1190 Assumes the variable `fail_stack'. Probably should only
1191 be called from within `PUSH_FAILURE_POINT'. */
1192 #define PUSH_FAILURE_ELT(item) \
1193 fail_stack.stack[fail_stack.avail++] = (item)
1195 /* These three POP... operations complement the three PUSH... operations.
1196 All assume that `fail_stack' is nonempty. */
1197 #define POP_FAILURE_POINTER() fail_stack.stack[--fail_stack.avail].pointer
1198 #define POP_FAILURE_INT() fail_stack.stack[--fail_stack.avail].integer
1199 #define POP_FAILURE_ELT() fail_stack.stack[--fail_stack.avail]
1201 /* Used to omit pushing failure point id's when we're not debugging. */
1203 # define DEBUG_PUSH PUSH_FAILURE_INT
1204 # define DEBUG_POP(item_addr) *(item_addr) = POP_FAILURE_INT ()
1206 # define DEBUG_PUSH(item)
1207 # define DEBUG_POP(item_addr)
1211 /* Push the information about the state we will need
1212 if we ever fail back to it.
1214 Requires variables fail_stack, regstart, regend, reg_info, and
1215 num_regs_pushed be declared. DOUBLE_FAIL_STACK requires `destination'
1218 Does `return FAILURE_CODE' if runs out of memory. */
1220 #define PUSH_FAILURE_POINT(pattern_place, string_place, failure_code) \
1222 char *destination; \
1223 /* Must be int, so when we don't save any registers, the arithmetic \
1224 of 0 + -1 isn't done as unsigned. */ \
1225 /* Can't be int, since there is not a shred of a guarantee that int \
1226 is wide enough to hold a value of something to which pointer can \
1228 active_reg_t this_reg; \
1230 DEBUG_STATEMENT (failure_id++); \
1231 DEBUG_STATEMENT (nfailure_points_pushed++); \
1232 DEBUG_PRINT2 ("\nPUSH_FAILURE_POINT #%u:\n", failure_id); \
1233 DEBUG_PRINT2 (" Before push, next avail: %d\n", (fail_stack).avail);\
1234 DEBUG_PRINT2 (" size: %d\n", (fail_stack).size);\
1236 DEBUG_PRINT2 (" slots needed: %ld\n", NUM_FAILURE_ITEMS); \
1237 DEBUG_PRINT2 (" available: %d\n", REMAINING_AVAIL_SLOTS); \
1239 /* Ensure we have enough space allocated for what we will push. */ \
1240 while (REMAINING_AVAIL_SLOTS < NUM_FAILURE_ITEMS) \
1242 if (!DOUBLE_FAIL_STACK (fail_stack)) \
1243 return failure_code; \
1245 DEBUG_PRINT2 ("\n Doubled stack; size now: %d\n", \
1246 (fail_stack).size); \
1247 DEBUG_PRINT2 (" slots available: %d\n", REMAINING_AVAIL_SLOTS);\
1250 /* Push the info, starting with the registers. */ \
1251 DEBUG_PRINT1 ("\n"); \
1254 for (this_reg = lowest_active_reg; this_reg <= highest_active_reg; \
1257 DEBUG_PRINT2 (" Pushing reg: %lu\n", this_reg); \
1258 DEBUG_STATEMENT (num_regs_pushed++); \
1260 DEBUG_PRINT2 (" start: %p\n", regstart[this_reg]); \
1261 PUSH_FAILURE_POINTER (regstart[this_reg]); \
1263 DEBUG_PRINT2 (" end: %p\n", regend[this_reg]); \
1264 PUSH_FAILURE_POINTER (regend[this_reg]); \
1266 DEBUG_PRINT2 (" info: %p\n ", \
1267 reg_info[this_reg].word.pointer); \
1268 DEBUG_PRINT2 (" match_null=%d", \
1269 REG_MATCH_NULL_STRING_P (reg_info[this_reg])); \
1270 DEBUG_PRINT2 (" active=%d", IS_ACTIVE (reg_info[this_reg])); \
1271 DEBUG_PRINT2 (" matched_something=%d", \
1272 MATCHED_SOMETHING (reg_info[this_reg])); \
1273 DEBUG_PRINT2 (" ever_matched=%d", \
1274 EVER_MATCHED_SOMETHING (reg_info[this_reg])); \
1275 DEBUG_PRINT1 ("\n"); \
1276 PUSH_FAILURE_ELT (reg_info[this_reg].word); \
1279 DEBUG_PRINT2 (" Pushing low active reg: %ld\n", lowest_active_reg);\
1280 PUSH_FAILURE_INT (lowest_active_reg); \
1282 DEBUG_PRINT2 (" Pushing high active reg: %ld\n", highest_active_reg);\
1283 PUSH_FAILURE_INT (highest_active_reg); \
1285 DEBUG_PRINT2 (" Pushing pattern %p:\n", pattern_place); \
1286 DEBUG_PRINT_COMPILED_PATTERN (bufp, pattern_place, pend); \
1287 PUSH_FAILURE_POINTER (pattern_place); \
1289 DEBUG_PRINT2 (" Pushing string %p: `", string_place); \
1290 DEBUG_PRINT_DOUBLE_STRING (string_place, string1, size1, string2, \
1292 DEBUG_PRINT1 ("'\n"); \
1293 PUSH_FAILURE_POINTER (string_place); \
1295 DEBUG_PRINT2 (" Pushing failure id: %u\n", failure_id); \
1296 DEBUG_PUSH (failure_id); \
1299 /* This is the number of items that are pushed and popped on the stack
1300 for each register. */
1301 #define NUM_REG_ITEMS 3
1303 /* Individual items aside from the registers. */
1305 # define NUM_NONREG_ITEMS 5 /* Includes failure point id. */
1307 # define NUM_NONREG_ITEMS 4
1310 /* We push at most this many items on the stack. */
1311 /* We used to use (num_regs - 1), which is the number of registers
1312 this regexp will save; but that was changed to 5
1313 to avoid stack overflow for a regexp with lots of parens. */
1314 #define MAX_FAILURE_ITEMS (5 * NUM_REG_ITEMS + NUM_NONREG_ITEMS)
1316 /* We actually push this many items. */
1317 #define NUM_FAILURE_ITEMS \
1319 ? 0 : highest_active_reg - lowest_active_reg + 1) \
1323 /* How many items can still be added to the stack without overflowing it. */
1324 #define REMAINING_AVAIL_SLOTS ((fail_stack).size - (fail_stack).avail)
1327 /* Pops what PUSH_FAIL_STACK pushes.
1329 We restore into the parameters, all of which should be lvalues:
1330 STR -- the saved data position.
1331 PAT -- the saved pattern position.
1332 LOW_REG, HIGH_REG -- the highest and lowest active registers.
1333 REGSTART, REGEND -- arrays of string positions.
1334 REG_INFO -- array of information about each subexpression.
1336 Also assumes the variables `fail_stack' and (if debugging), `bufp',
1337 `pend', `string1', `size1', `string2', and `size2'. */
1339 #define POP_FAILURE_POINT(str, pat, low_reg, high_reg, regstart, regend, reg_info)\
1341 DEBUG_STATEMENT (unsigned failure_id;) \
1342 active_reg_t this_reg; \
1343 const unsigned char *string_temp; \
1345 assert (!FAIL_STACK_EMPTY ()); \
1347 /* Remove failure points and point to how many regs pushed. */ \
1348 DEBUG_PRINT1 ("POP_FAILURE_POINT:\n"); \
1349 DEBUG_PRINT2 (" Before pop, next avail: %d\n", fail_stack.avail); \
1350 DEBUG_PRINT2 (" size: %d\n", fail_stack.size); \
1352 assert (fail_stack.avail >= NUM_NONREG_ITEMS); \
1354 DEBUG_POP (&failure_id); \
1355 DEBUG_PRINT2 (" Popping failure id: %u\n", failure_id); \
1357 /* If the saved string location is NULL, it came from an \
1358 on_failure_keep_string_jump opcode, and we want to throw away the \
1359 saved NULL, thus retaining our current position in the string. */ \
1360 string_temp = POP_FAILURE_POINTER (); \
1361 if (string_temp != NULL) \
1362 str = (const char *) string_temp; \
1364 DEBUG_PRINT2 (" Popping string %p: `", str); \
1365 DEBUG_PRINT_DOUBLE_STRING (str, string1, size1, string2, size2); \
1366 DEBUG_PRINT1 ("'\n"); \
1368 pat = (unsigned char *) POP_FAILURE_POINTER (); \
1369 DEBUG_PRINT2 (" Popping pattern %p:\n", pat); \
1370 DEBUG_PRINT_COMPILED_PATTERN (bufp, pat, pend); \
1372 /* Restore register info. */ \
1373 high_reg = (active_reg_t) POP_FAILURE_INT (); \
1374 DEBUG_PRINT2 (" Popping high active reg: %ld\n", high_reg); \
1376 low_reg = (active_reg_t) POP_FAILURE_INT (); \
1377 DEBUG_PRINT2 (" Popping low active reg: %ld\n", low_reg); \
1380 for (this_reg = high_reg; this_reg >= low_reg; this_reg--) \
1382 DEBUG_PRINT2 (" Popping reg: %ld\n", this_reg); \
1384 reg_info[this_reg].word = POP_FAILURE_ELT (); \
1385 DEBUG_PRINT2 (" info: %p\n", \
1386 reg_info[this_reg].word.pointer); \
1388 regend[this_reg] = (const char *) POP_FAILURE_POINTER (); \
1389 DEBUG_PRINT2 (" end: %p\n", regend[this_reg]); \
1391 regstart[this_reg] = (const char *) POP_FAILURE_POINTER (); \
1392 DEBUG_PRINT2 (" start: %p\n", regstart[this_reg]); \
1396 for (this_reg = highest_active_reg; this_reg > high_reg; this_reg--) \
1398 reg_info[this_reg].word.integer = 0; \
1399 regend[this_reg] = 0; \
1400 regstart[this_reg] = 0; \
1402 highest_active_reg = high_reg; \
1405 set_regs_matched_done = 0; \
1406 DEBUG_STATEMENT (nfailure_points_popped++); \
1407 } /* POP_FAILURE_POINT */
1411 /* Structure for per-register (a.k.a. per-group) information.
1412 Other register information, such as the
1413 starting and ending positions (which are addresses), and the list of
1414 inner groups (which is a bits list) are maintained in separate
1417 We are making a (strictly speaking) nonportable assumption here: that
1418 the compiler will pack our bit fields into something that fits into
1419 the type of `word', i.e., is something that fits into one item on the
1423 /* Declarations and macros for re_match_2. */
1427 fail_stack_elt_t word
;
1430 /* This field is one if this group can match the empty string,
1431 zero if not. If not yet determined, `MATCH_NULL_UNSET_VALUE'. */
1432 #define MATCH_NULL_UNSET_VALUE 3
1433 unsigned match_null_string_p
: 2;
1434 unsigned is_active
: 1;
1435 unsigned matched_something
: 1;
1436 unsigned ever_matched_something
: 1;
1438 } register_info_type
;
1440 #define REG_MATCH_NULL_STRING_P(R) ((R).bits.match_null_string_p)
1441 #define IS_ACTIVE(R) ((R).bits.is_active)
1442 #define MATCHED_SOMETHING(R) ((R).bits.matched_something)
1443 #define EVER_MATCHED_SOMETHING(R) ((R).bits.ever_matched_something)
1446 /* Call this when have matched a real character; it sets `matched' flags
1447 for the subexpressions which we are currently inside. Also records
1448 that those subexprs have matched. */
1449 #define SET_REGS_MATCHED() \
1452 if (!set_regs_matched_done) \
1455 set_regs_matched_done = 1; \
1456 for (r = lowest_active_reg; r <= highest_active_reg; r++) \
1458 MATCHED_SOMETHING (reg_info[r]) \
1459 = EVER_MATCHED_SOMETHING (reg_info[r]) \
1466 /* Registers are set to a sentinel when they haven't yet matched. */
1467 static char reg_unset_dummy
;
1468 #define REG_UNSET_VALUE (®_unset_dummy)
1469 #define REG_UNSET(e) ((e) == REG_UNSET_VALUE)
1471 /* Subroutine declarations and macros for regex_compile. */
1473 static reg_errcode_t regex_compile
_RE_ARGS ((const char *pattern
, size_t size
,
1474 reg_syntax_t syntax
,
1475 struct re_pattern_buffer
*bufp
));
1476 static void store_op1
_RE_ARGS ((re_opcode_t op
, unsigned char *loc
, int arg
));
1477 static void store_op2
_RE_ARGS ((re_opcode_t op
, unsigned char *loc
,
1478 int arg1
, int arg2
));
1479 static void insert_op1
_RE_ARGS ((re_opcode_t op
, unsigned char *loc
,
1480 int arg
, unsigned char *end
));
1481 static void insert_op2
_RE_ARGS ((re_opcode_t op
, unsigned char *loc
,
1482 int arg1
, int arg2
, unsigned char *end
));
1483 static boolean at_begline_loc_p
_RE_ARGS ((const char *pattern
, const char *p
,
1484 reg_syntax_t syntax
));
1485 static boolean at_endline_loc_p
_RE_ARGS ((const char *p
, const char *pend
,
1486 reg_syntax_t syntax
));
1487 static reg_errcode_t compile_range
_RE_ARGS ((const char **p_ptr
,
1490 reg_syntax_t syntax
,
1493 /* Fetch the next character in the uncompiled pattern---translating it
1494 if necessary. Also cast from a signed character in the constant
1495 string passed to us by the user to an unsigned char that we can use
1496 as an array index (in, e.g., `translate'). */
1498 # define PATFETCH(c) \
1499 do {if (p == pend) return REG_EEND; \
1500 c = (unsigned char) *p++; \
1501 if (translate) c = (unsigned char) translate[c]; \
1505 /* Fetch the next character in the uncompiled pattern, with no
1507 #define PATFETCH_RAW(c) \
1508 do {if (p == pend) return REG_EEND; \
1509 c = (unsigned char) *p++; \
1512 /* Go backwards one character in the pattern. */
1513 #define PATUNFETCH p--
1516 /* If `translate' is non-null, return translate[D], else just D. We
1517 cast the subscript to translate because some data is declared as
1518 `char *', to avoid warnings when a string constant is passed. But
1519 when we use a character as a subscript we must make it unsigned. */
1521 # define TRANSLATE(d) \
1522 (translate ? (char) translate[(unsigned char) (d)] : (d))
1526 /* Macros for outputting the compiled pattern into `buffer'. */
1528 /* If the buffer isn't allocated when it comes in, use this. */
1529 #define INIT_BUF_SIZE 32
1531 /* Make sure we have at least N more bytes of space in buffer. */
1532 #define GET_BUFFER_SPACE(n) \
1533 while ((unsigned long) (b - bufp->buffer + (n)) > bufp->allocated) \
1536 /* Make sure we have one more byte of buffer space and then add C to it. */
1537 #define BUF_PUSH(c) \
1539 GET_BUFFER_SPACE (1); \
1540 *b++ = (unsigned char) (c); \
1544 /* Ensure we have two more bytes of buffer space and then append C1 and C2. */
1545 #define BUF_PUSH_2(c1, c2) \
1547 GET_BUFFER_SPACE (2); \
1548 *b++ = (unsigned char) (c1); \
1549 *b++ = (unsigned char) (c2); \
1553 /* As with BUF_PUSH_2, except for three bytes. */
1554 #define BUF_PUSH_3(c1, c2, c3) \
1556 GET_BUFFER_SPACE (3); \
1557 *b++ = (unsigned char) (c1); \
1558 *b++ = (unsigned char) (c2); \
1559 *b++ = (unsigned char) (c3); \
1563 /* Store a jump with opcode OP at LOC to location TO. We store a
1564 relative address offset by the three bytes the jump itself occupies. */
1565 #define STORE_JUMP(op, loc, to) \
1566 store_op1 (op, loc, (int) ((to) - (loc) - 3))
1568 /* Likewise, for a two-argument jump. */
1569 #define STORE_JUMP2(op, loc, to, arg) \
1570 store_op2 (op, loc, (int) ((to) - (loc) - 3), arg)
1572 /* Like `STORE_JUMP', but for inserting. Assume `b' is the buffer end. */
1573 #define INSERT_JUMP(op, loc, to) \
1574 insert_op1 (op, loc, (int) ((to) - (loc) - 3), b)
1576 /* Like `STORE_JUMP2', but for inserting. Assume `b' is the buffer end. */
1577 #define INSERT_JUMP2(op, loc, to, arg) \
1578 insert_op2 (op, loc, (int) ((to) - (loc) - 3), arg, b)
1581 /* This is not an arbitrary limit: the arguments which represent offsets
1582 into the pattern are two bytes long. So if 2^16 bytes turns out to
1583 be too small, many things would have to change. */
1584 /* Any other compiler which, like MSC, has allocation limit below 2^16
1585 bytes will have to use approach similar to what was done below for
1586 MSC and drop MAX_BUF_SIZE a bit. Otherwise you may end up
1587 reallocating to 0 bytes. Such thing is not going to work too well.
1588 You have been warned!! */
1589 #if defined _MSC_VER && !defined WIN32
1590 /* Microsoft C 16-bit versions limit malloc to approx 65512 bytes.
1591 The REALLOC define eliminates a flurry of conversion warnings,
1592 but is not required. */
1593 # define MAX_BUF_SIZE 65500L
1594 # define REALLOC(p,s) realloc ((p), (size_t) (s))
1596 # define MAX_BUF_SIZE (1L << 16)
1597 # define REALLOC(p,s) realloc ((p), (s))
1600 /* Extend the buffer by twice its current size via realloc and
1601 reset the pointers that pointed into the old block to point to the
1602 correct places in the new one. If extending the buffer results in it
1603 being larger than MAX_BUF_SIZE, then flag memory exhausted. */
1604 #define EXTEND_BUFFER() \
1606 unsigned char *old_buffer = bufp->buffer; \
1607 if (bufp->allocated == MAX_BUF_SIZE) \
1609 bufp->allocated <<= 1; \
1610 if (bufp->allocated > MAX_BUF_SIZE) \
1611 bufp->allocated = MAX_BUF_SIZE; \
1612 bufp->buffer = (unsigned char *) REALLOC (bufp->buffer, bufp->allocated);\
1613 if (bufp->buffer == NULL) \
1614 return REG_ESPACE; \
1615 /* If the buffer moved, move all the pointers into it. */ \
1616 if (old_buffer != bufp->buffer) \
1618 b = (b - old_buffer) + bufp->buffer; \
1619 begalt = (begalt - old_buffer) + bufp->buffer; \
1620 if (fixup_alt_jump) \
1621 fixup_alt_jump = (fixup_alt_jump - old_buffer) + bufp->buffer;\
1623 laststart = (laststart - old_buffer) + bufp->buffer; \
1624 if (pending_exact) \
1625 pending_exact = (pending_exact - old_buffer) + bufp->buffer; \
1630 /* Since we have one byte reserved for the register number argument to
1631 {start,stop}_memory, the maximum number of groups we can report
1632 things about is what fits in that byte. */
1633 #define MAX_REGNUM 255
1635 /* But patterns can have more than `MAX_REGNUM' registers. We just
1636 ignore the excess. */
1637 typedef unsigned regnum_t
;
1640 /* Macros for the compile stack. */
1642 /* Since offsets can go either forwards or backwards, this type needs to
1643 be able to hold values from -(MAX_BUF_SIZE - 1) to MAX_BUF_SIZE - 1. */
1644 /* int may be not enough when sizeof(int) == 2. */
1645 typedef long pattern_offset_t
;
1649 pattern_offset_t begalt_offset
;
1650 pattern_offset_t fixup_alt_jump
;
1651 pattern_offset_t inner_group_offset
;
1652 pattern_offset_t laststart_offset
;
1654 } compile_stack_elt_t
;
1659 compile_stack_elt_t
*stack
;
1661 unsigned avail
; /* Offset of next open position. */
1662 } compile_stack_type
;
1665 #define INIT_COMPILE_STACK_SIZE 32
1667 #define COMPILE_STACK_EMPTY (compile_stack.avail == 0)
1668 #define COMPILE_STACK_FULL (compile_stack.avail == compile_stack.size)
1670 /* The next available element. */
1671 #define COMPILE_STACK_TOP (compile_stack.stack[compile_stack.avail])
1674 /* Set the bit for character C in a list. */
1675 #define SET_LIST_BIT(c) \
1676 (b[((unsigned char) (c)) / BYTEWIDTH] \
1677 |= 1 << (((unsigned char) c) % BYTEWIDTH))
1680 /* Get the next unsigned number in the uncompiled pattern. */
1681 #define GET_UNSIGNED_NUMBER(num) \
1685 while (ISDIGIT (c)) \
1689 num = num * 10 + c - '0'; \
1697 #if defined _LIBC || (defined HAVE_WCTYPE_H && defined HAVE_WCHAR_H)
1698 /* The GNU C library provides support for user-defined character classes
1699 and the functions from ISO C amendement 1. */
1700 # ifdef CHARCLASS_NAME_MAX
1701 # define CHAR_CLASS_MAX_LENGTH CHARCLASS_NAME_MAX
1703 /* This shouldn't happen but some implementation might still have this
1704 problem. Use a reasonable default value. */
1705 # define CHAR_CLASS_MAX_LENGTH 256
1709 # define IS_CHAR_CLASS(string) __wctype (string)
1711 # define IS_CHAR_CLASS(string) wctype (string)
1714 # define CHAR_CLASS_MAX_LENGTH 6 /* Namely, `xdigit'. */
1716 # define IS_CHAR_CLASS(string) \
1717 (STREQ (string, "alpha") || STREQ (string, "upper") \
1718 || STREQ (string, "lower") || STREQ (string, "digit") \
1719 || STREQ (string, "alnum") || STREQ (string, "xdigit") \
1720 || STREQ (string, "space") || STREQ (string, "print") \
1721 || STREQ (string, "punct") || STREQ (string, "graph") \
1722 || STREQ (string, "cntrl") || STREQ (string, "blank"))
1725 #ifndef MATCH_MAY_ALLOCATE
1727 /* If we cannot allocate large objects within re_match_2_internal,
1728 we make the fail stack and register vectors global.
1729 The fail stack, we grow to the maximum size when a regexp
1731 The register vectors, we adjust in size each time we
1732 compile a regexp, according to the number of registers it needs. */
1734 static fail_stack_type fail_stack
;
1736 /* Size with which the following vectors are currently allocated.
1737 That is so we can make them bigger as needed,
1738 but never make them smaller. */
1739 static int regs_allocated_size
;
1741 static const char ** regstart
, ** regend
;
1742 static const char ** old_regstart
, ** old_regend
;
1743 static const char **best_regstart
, **best_regend
;
1744 static register_info_type
*reg_info
;
1745 static const char **reg_dummy
;
1746 static register_info_type
*reg_info_dummy
;
1748 /* Make the register vectors big enough for NUM_REGS registers,
1749 but don't make them smaller. */
1752 regex_grow_registers (num_regs
)
1755 if (num_regs
> regs_allocated_size
)
1757 RETALLOC_IF (regstart
, num_regs
, const char *);
1758 RETALLOC_IF (regend
, num_regs
, const char *);
1759 RETALLOC_IF (old_regstart
, num_regs
, const char *);
1760 RETALLOC_IF (old_regend
, num_regs
, const char *);
1761 RETALLOC_IF (best_regstart
, num_regs
, const char *);
1762 RETALLOC_IF (best_regend
, num_regs
, const char *);
1763 RETALLOC_IF (reg_info
, num_regs
, register_info_type
);
1764 RETALLOC_IF (reg_dummy
, num_regs
, const char *);
1765 RETALLOC_IF (reg_info_dummy
, num_regs
, register_info_type
);
1767 regs_allocated_size
= num_regs
;
1771 #endif /* not MATCH_MAY_ALLOCATE */
1773 static boolean group_in_compile_stack
_RE_ARGS ((compile_stack_type
1777 /* `regex_compile' compiles PATTERN (of length SIZE) according to SYNTAX.
1778 Returns one of error codes defined in `gnu-regex.h', or zero for success.
1780 Assumes the `allocated' (and perhaps `buffer') and `translate'
1781 fields are set in BUFP on entry.
1783 If it succeeds, results are put in BUFP (if it returns an error, the
1784 contents of BUFP are undefined):
1785 `buffer' is the compiled pattern;
1786 `syntax' is set to SYNTAX;
1787 `used' is set to the length of the compiled pattern;
1788 `fastmap_accurate' is zero;
1789 `re_nsub' is the number of subexpressions in PATTERN;
1790 `not_bol' and `not_eol' are zero;
1792 The `fastmap' and `newline_anchor' fields are neither
1793 examined nor set. */
1795 /* Return, freeing storage we allocated. */
1796 #define FREE_STACK_RETURN(value) \
1797 return (free (compile_stack.stack), value)
1799 static reg_errcode_t
1800 regex_compile (pattern
, size
, syntax
, bufp
)
1801 const char *pattern
;
1803 reg_syntax_t syntax
;
1804 struct re_pattern_buffer
*bufp
;
1806 /* We fetch characters from PATTERN here. Even though PATTERN is
1807 `char *' (i.e., signed), we declare these variables as unsigned, so
1808 they can be reliably used as array indices. */
1809 register unsigned char c
, c1
;
1811 /* A random temporary spot in PATTERN. */
1814 /* Points to the end of the buffer, where we should append. */
1815 register unsigned char *b
;
1817 /* Keeps track of unclosed groups. */
1818 compile_stack_type compile_stack
;
1820 /* Points to the current (ending) position in the pattern. */
1821 const char *p
= pattern
;
1822 const char *pend
= pattern
+ size
;
1824 /* How to translate the characters in the pattern. */
1825 RE_TRANSLATE_TYPE translate
= bufp
->translate
;
1827 /* Address of the count-byte of the most recently inserted `exactn'
1828 command. This makes it possible to tell if a new exact-match
1829 character can be added to that command or if the character requires
1830 a new `exactn' command. */
1831 unsigned char *pending_exact
= 0;
1833 /* Address of start of the most recently finished expression.
1834 This tells, e.g., postfix * where to find the start of its
1835 operand. Reset at the beginning of groups and alternatives. */
1836 unsigned char *laststart
= 0;
1838 /* Address of beginning of regexp, or inside of last group. */
1839 unsigned char *begalt
;
1841 /* Place in the uncompiled pattern (i.e., the {) to
1842 which to go back if the interval is invalid. */
1843 const char *beg_interval
;
1845 /* Address of the place where a forward jump should go to the end of
1846 the containing expression. Each alternative of an `or' -- except the
1847 last -- ends with a forward jump of this sort. */
1848 unsigned char *fixup_alt_jump
= 0;
1850 /* Counts open-groups as they are encountered. Remembered for the
1851 matching close-group on the compile stack, so the same register
1852 number is put in the stop_memory as the start_memory. */
1853 regnum_t regnum
= 0;
1856 DEBUG_PRINT1 ("\nCompiling pattern: ");
1859 unsigned debug_count
;
1861 for (debug_count
= 0; debug_count
< size
; debug_count
++)
1862 putchar (pattern
[debug_count
]);
1867 /* Initialize the compile stack. */
1868 compile_stack
.stack
= TALLOC (INIT_COMPILE_STACK_SIZE
, compile_stack_elt_t
);
1869 if (compile_stack
.stack
== NULL
)
1872 compile_stack
.size
= INIT_COMPILE_STACK_SIZE
;
1873 compile_stack
.avail
= 0;
1875 /* Initialize the pattern buffer. */
1876 bufp
->syntax
= syntax
;
1877 bufp
->fastmap_accurate
= 0;
1878 bufp
->not_bol
= bufp
->not_eol
= 0;
1880 /* Set `used' to zero, so that if we return an error, the pattern
1881 printer (for debugging) will think there's no pattern. We reset it
1885 /* Always count groups, whether or not bufp->no_sub is set. */
1888 #if !defined emacs && !defined SYNTAX_TABLE
1889 /* Initialize the syntax table. */
1890 init_syntax_once ();
1893 if (bufp
->allocated
== 0)
1896 { /* If zero allocated, but buffer is non-null, try to realloc
1897 enough space. This loses if buffer's address is bogus, but
1898 that is the user's responsibility. */
1899 RETALLOC (bufp
->buffer
, INIT_BUF_SIZE
, unsigned char);
1902 { /* Caller did not allocate a buffer. Do it for them. */
1903 bufp
->buffer
= TALLOC (INIT_BUF_SIZE
, unsigned char);
1905 if (!bufp
->buffer
) FREE_STACK_RETURN (REG_ESPACE
);
1907 bufp
->allocated
= INIT_BUF_SIZE
;
1910 begalt
= b
= bufp
->buffer
;
1912 /* Loop through the uncompiled pattern until we're at the end. */
1921 if ( /* If at start of pattern, it's an operator. */
1923 /* If context independent, it's an operator. */
1924 || syntax
& RE_CONTEXT_INDEP_ANCHORS
1925 /* Otherwise, depends on what's come before. */
1926 || at_begline_loc_p (pattern
, p
, syntax
))
1936 if ( /* If at end of pattern, it's an operator. */
1938 /* If context independent, it's an operator. */
1939 || syntax
& RE_CONTEXT_INDEP_ANCHORS
1940 /* Otherwise, depends on what's next. */
1941 || at_endline_loc_p (p
, pend
, syntax
))
1951 if ((syntax
& RE_BK_PLUS_QM
)
1952 || (syntax
& RE_LIMITED_OPS
))
1956 /* If there is no previous pattern... */
1959 if (syntax
& RE_CONTEXT_INVALID_OPS
)
1960 FREE_STACK_RETURN (REG_BADRPT
);
1961 else if (!(syntax
& RE_CONTEXT_INDEP_OPS
))
1966 /* Are we optimizing this jump? */
1967 boolean keep_string_p
= false;
1969 /* 1 means zero (many) matches is allowed. */
1970 char zero_times_ok
= 0, many_times_ok
= 0;
1972 /* If there is a sequence of repetition chars, collapse it
1973 down to just one (the right one). We can't combine
1974 interval operators with these because of, e.g., `a{2}*',
1975 which should only match an even number of `a's. */
1979 zero_times_ok
|= c
!= '+';
1980 many_times_ok
|= c
!= '?';
1988 || (!(syntax
& RE_BK_PLUS_QM
) && (c
== '+' || c
== '?')))
1991 else if (syntax
& RE_BK_PLUS_QM
&& c
== '\\')
1993 if (p
== pend
) FREE_STACK_RETURN (REG_EESCAPE
);
1996 if (!(c1
== '+' || c1
== '?'))
2011 /* If we get here, we found another repeat character. */
2014 /* Star, etc. applied to an empty pattern is equivalent
2015 to an empty pattern. */
2019 /* Now we know whether or not zero matches is allowed
2020 and also whether or not two or more matches is allowed. */
2022 { /* More than one repetition is allowed, so put in at the
2023 end a backward relative jump from `b' to before the next
2024 jump we're going to put in below (which jumps from
2025 laststart to after this jump).
2027 But if we are at the `*' in the exact sequence `.*\n',
2028 insert an unconditional jump backwards to the .,
2029 instead of the beginning of the loop. This way we only
2030 push a failure point once, instead of every time
2031 through the loop. */
2032 assert (p
- 1 > pattern
);
2034 /* Allocate the space for the jump. */
2035 GET_BUFFER_SPACE (3);
2037 /* We know we are not at the first character of the pattern,
2038 because laststart was nonzero. And we've already
2039 incremented `p', by the way, to be the character after
2040 the `*'. Do we have to do something analogous here
2041 for null bytes, because of RE_DOT_NOT_NULL? */
2042 if (TRANSLATE (*(p
- 2)) == TRANSLATE ('.')
2044 && p
< pend
&& TRANSLATE (*p
) == TRANSLATE ('\n')
2045 && !(syntax
& RE_DOT_NEWLINE
))
2046 { /* We have .*\n. */
2047 STORE_JUMP (jump
, b
, laststart
);
2048 keep_string_p
= true;
2051 /* Anything else. */
2052 STORE_JUMP (maybe_pop_jump
, b
, laststart
- 3);
2054 /* We've added more stuff to the buffer. */
2058 /* On failure, jump from laststart to b + 3, which will be the
2059 end of the buffer after this jump is inserted. */
2060 GET_BUFFER_SPACE (3);
2061 INSERT_JUMP (keep_string_p
? on_failure_keep_string_jump
2069 /* At least one repetition is required, so insert a
2070 `dummy_failure_jump' before the initial
2071 `on_failure_jump' instruction of the loop. This
2072 effects a skip over that instruction the first time
2073 we hit that loop. */
2074 GET_BUFFER_SPACE (3);
2075 INSERT_JUMP (dummy_failure_jump
, laststart
, laststart
+ 6);
2090 boolean had_char_class
= false;
2092 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2094 /* Ensure that we have enough space to push a charset: the
2095 opcode, the length count, and the bitset; 34 bytes in all. */
2096 GET_BUFFER_SPACE (34);
2100 /* We test `*p == '^' twice, instead of using an if
2101 statement, so we only need one BUF_PUSH. */
2102 BUF_PUSH (*p
== '^' ? charset_not
: charset
);
2106 /* Remember the first position in the bracket expression. */
2109 /* Push the number of bytes in the bitmap. */
2110 BUF_PUSH ((1 << BYTEWIDTH
) / BYTEWIDTH
);
2112 /* Clear the whole map. */
2113 bzero (b
, (1 << BYTEWIDTH
) / BYTEWIDTH
);
2115 /* charset_not matches newline according to a syntax bit. */
2116 if ((re_opcode_t
) b
[-2] == charset_not
2117 && (syntax
& RE_HAT_LISTS_NOT_NEWLINE
))
2118 SET_LIST_BIT ('\n');
2120 /* Read in characters and ranges, setting map bits. */
2123 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2127 /* \ might escape characters inside [...] and [^...]. */
2128 if ((syntax
& RE_BACKSLASH_ESCAPE_IN_LISTS
) && c
== '\\')
2130 if (p
== pend
) FREE_STACK_RETURN (REG_EESCAPE
);
2137 /* Could be the end of the bracket expression. If it's
2138 not (i.e., when the bracket expression is `[]' so
2139 far), the ']' character bit gets set way below. */
2140 if (c
== ']' && p
!= p1
+ 1)
2143 /* Look ahead to see if it's a range when the last thing
2144 was a character class. */
2145 if (had_char_class
&& c
== '-' && *p
!= ']')
2146 FREE_STACK_RETURN (REG_ERANGE
);
2148 /* Look ahead to see if it's a range when the last thing
2149 was a character: if this is a hyphen not at the
2150 beginning or the end of a list, then it's the range
2153 && !(p
- 2 >= pattern
&& p
[-2] == '[')
2154 && !(p
- 3 >= pattern
&& p
[-3] == '[' && p
[-2] == '^')
2158 = compile_range (&p
, pend
, translate
, syntax
, b
);
2159 if (ret
!= REG_NOERROR
) FREE_STACK_RETURN (ret
);
2162 else if (p
[0] == '-' && p
[1] != ']')
2163 { /* This handles ranges made up of characters only. */
2166 /* Move past the `-'. */
2169 ret
= compile_range (&p
, pend
, translate
, syntax
, b
);
2170 if (ret
!= REG_NOERROR
) FREE_STACK_RETURN (ret
);
2173 /* See if we're at the beginning of a possible character
2176 else if (syntax
& RE_CHAR_CLASSES
&& c
== '[' && *p
== ':')
2177 { /* Leave room for the null. */
2178 char str
[CHAR_CLASS_MAX_LENGTH
+ 1];
2183 /* If pattern is `[[:'. */
2184 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2189 if ((c
== ':' && *p
== ']') || p
== pend
2190 || c1
== CHAR_CLASS_MAX_LENGTH
)
2196 /* If isn't a word bracketed by `[:' and `:]':
2197 undo the ending character, the letters, and leave
2198 the leading `:' and `[' (but set bits for them). */
2199 if (c
== ':' && *p
== ']')
2201 #if defined _LIBC || (defined HAVE_WCTYPE_H && defined HAVE_WCHAR_H)
2202 boolean is_lower
= STREQ (str
, "lower");
2203 boolean is_upper
= STREQ (str
, "upper");
2207 wt
= IS_CHAR_CLASS (str
);
2209 FREE_STACK_RETURN (REG_ECTYPE
);
2211 /* Throw away the ] at the end of the character
2215 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2217 for (ch
= 0; ch
< 1 << BYTEWIDTH
; ++ch
)
2220 if (__iswctype (__btowc (ch
), wt
))
2223 if (iswctype (btowc (ch
), wt
))
2227 if (translate
&& (is_upper
|| is_lower
)
2228 && (ISUPPER (ch
) || ISLOWER (ch
)))
2232 had_char_class
= true;
2235 boolean is_alnum
= STREQ (str
, "alnum");
2236 boolean is_alpha
= STREQ (str
, "alpha");
2237 boolean is_blank
= STREQ (str
, "blank");
2238 boolean is_cntrl
= STREQ (str
, "cntrl");
2239 boolean is_digit
= STREQ (str
, "digit");
2240 boolean is_graph
= STREQ (str
, "graph");
2241 boolean is_lower
= STREQ (str
, "lower");
2242 boolean is_print
= STREQ (str
, "print");
2243 boolean is_punct
= STREQ (str
, "punct");
2244 boolean is_space
= STREQ (str
, "space");
2245 boolean is_upper
= STREQ (str
, "upper");
2246 boolean is_xdigit
= STREQ (str
, "xdigit");
2248 if (!IS_CHAR_CLASS (str
))
2249 FREE_STACK_RETURN (REG_ECTYPE
);
2251 /* Throw away the ] at the end of the character
2255 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2257 for (ch
= 0; ch
< 1 << BYTEWIDTH
; ch
++)
2259 /* This was split into 3 if's to
2260 avoid an arbitrary limit in some compiler. */
2261 if ( (is_alnum
&& ISALNUM (ch
))
2262 || (is_alpha
&& ISALPHA (ch
))
2263 || (is_blank
&& ISBLANK (ch
))
2264 || (is_cntrl
&& ISCNTRL (ch
)))
2266 if ( (is_digit
&& ISDIGIT (ch
))
2267 || (is_graph
&& ISGRAPH (ch
))
2268 || (is_lower
&& ISLOWER (ch
))
2269 || (is_print
&& ISPRINT (ch
)))
2271 if ( (is_punct
&& ISPUNCT (ch
))
2272 || (is_space
&& ISSPACE (ch
))
2273 || (is_upper
&& ISUPPER (ch
))
2274 || (is_xdigit
&& ISXDIGIT (ch
)))
2276 if ( translate
&& (is_upper
|| is_lower
)
2277 && (ISUPPER (ch
) || ISLOWER (ch
)))
2280 had_char_class
= true;
2281 #endif /* libc || wctype.h */
2290 had_char_class
= false;
2295 had_char_class
= false;
2300 /* Discard any (non)matching list bytes that are all 0 at the
2301 end of the map. Decrease the map-length byte too. */
2302 while ((int) b
[-1] > 0 && b
[b
[-1] - 1] == 0)
2310 if (syntax
& RE_NO_BK_PARENS
)
2317 if (syntax
& RE_NO_BK_PARENS
)
2324 if (syntax
& RE_NEWLINE_ALT
)
2331 if (syntax
& RE_NO_BK_VBAR
)
2338 if (syntax
& RE_INTERVALS
&& syntax
& RE_NO_BK_BRACES
)
2339 goto handle_interval
;
2345 if (p
== pend
) FREE_STACK_RETURN (REG_EESCAPE
);
2347 /* Do not translate the character after the \, so that we can
2348 distinguish, e.g., \B from \b, even if we normally would
2349 translate, e.g., B to b. */
2355 if (syntax
& RE_NO_BK_PARENS
)
2356 goto normal_backslash
;
2362 if (COMPILE_STACK_FULL
)
2364 RETALLOC (compile_stack
.stack
, compile_stack
.size
<< 1,
2365 compile_stack_elt_t
);
2366 if (compile_stack
.stack
== NULL
) return REG_ESPACE
;
2368 compile_stack
.size
<<= 1;
2371 /* These are the values to restore when we hit end of this
2372 group. They are all relative offsets, so that if the
2373 whole pattern moves because of realloc, they will still
2375 COMPILE_STACK_TOP
.begalt_offset
= begalt
- bufp
->buffer
;
2376 COMPILE_STACK_TOP
.fixup_alt_jump
2377 = fixup_alt_jump
? fixup_alt_jump
- bufp
->buffer
+ 1 : 0;
2378 COMPILE_STACK_TOP
.laststart_offset
= b
- bufp
->buffer
;
2379 COMPILE_STACK_TOP
.regnum
= regnum
;
2381 /* We will eventually replace the 0 with the number of
2382 groups inner to this one. But do not push a
2383 start_memory for groups beyond the last one we can
2384 represent in the compiled pattern. */
2385 if (regnum
<= MAX_REGNUM
)
2387 COMPILE_STACK_TOP
.inner_group_offset
= b
- bufp
->buffer
+ 2;
2388 BUF_PUSH_3 (start_memory
, regnum
, 0);
2391 compile_stack
.avail
++;
2396 /* If we've reached MAX_REGNUM groups, then this open
2397 won't actually generate any code, so we'll have to
2398 clear pending_exact explicitly. */
2404 if (syntax
& RE_NO_BK_PARENS
) goto normal_backslash
;
2406 if (COMPILE_STACK_EMPTY
)
2408 if (syntax
& RE_UNMATCHED_RIGHT_PAREN_ORD
)
2409 goto normal_backslash
;
2411 FREE_STACK_RETURN (REG_ERPAREN
);
2416 { /* Push a dummy failure point at the end of the
2417 alternative for a possible future
2418 `pop_failure_jump' to pop. See comments at
2419 `push_dummy_failure' in `re_match_2'. */
2420 BUF_PUSH (push_dummy_failure
);
2422 /* We allocated space for this jump when we assigned
2423 to `fixup_alt_jump', in the `handle_alt' case below. */
2424 STORE_JUMP (jump_past_alt
, fixup_alt_jump
, b
- 1);
2427 /* See similar code for backslashed left paren above. */
2428 if (COMPILE_STACK_EMPTY
)
2430 if (syntax
& RE_UNMATCHED_RIGHT_PAREN_ORD
)
2433 FREE_STACK_RETURN (REG_ERPAREN
);
2436 /* Since we just checked for an empty stack above, this
2437 ``can't happen''. */
2438 assert (compile_stack
.avail
!= 0);
2440 /* We don't just want to restore into `regnum', because
2441 later groups should continue to be numbered higher,
2442 as in `(ab)c(de)' -- the second group is #2. */
2443 regnum_t this_group_regnum
;
2445 compile_stack
.avail
--;
2446 begalt
= bufp
->buffer
+ COMPILE_STACK_TOP
.begalt_offset
;
2448 = COMPILE_STACK_TOP
.fixup_alt_jump
2449 ? bufp
->buffer
+ COMPILE_STACK_TOP
.fixup_alt_jump
- 1
2451 laststart
= bufp
->buffer
+ COMPILE_STACK_TOP
.laststart_offset
;
2452 this_group_regnum
= COMPILE_STACK_TOP
.regnum
;
2453 /* If we've reached MAX_REGNUM groups, then this open
2454 won't actually generate any code, so we'll have to
2455 clear pending_exact explicitly. */
2458 /* We're at the end of the group, so now we know how many
2459 groups were inside this one. */
2460 if (this_group_regnum
<= MAX_REGNUM
)
2462 unsigned char *inner_group_loc
2463 = bufp
->buffer
+ COMPILE_STACK_TOP
.inner_group_offset
;
2465 *inner_group_loc
= regnum
- this_group_regnum
;
2466 BUF_PUSH_3 (stop_memory
, this_group_regnum
,
2467 regnum
- this_group_regnum
);
2473 case '|': /* `\|'. */
2474 if (syntax
& RE_LIMITED_OPS
|| syntax
& RE_NO_BK_VBAR
)
2475 goto normal_backslash
;
2477 if (syntax
& RE_LIMITED_OPS
)
2480 /* Insert before the previous alternative a jump which
2481 jumps to this alternative if the former fails. */
2482 GET_BUFFER_SPACE (3);
2483 INSERT_JUMP (on_failure_jump
, begalt
, b
+ 6);
2487 /* The alternative before this one has a jump after it
2488 which gets executed if it gets matched. Adjust that
2489 jump so it will jump to this alternative's analogous
2490 jump (put in below, which in turn will jump to the next
2491 (if any) alternative's such jump, etc.). The last such
2492 jump jumps to the correct final destination. A picture:
2498 If we are at `b', then fixup_alt_jump right now points to a
2499 three-byte space after `a'. We'll put in the jump, set
2500 fixup_alt_jump to right after `b', and leave behind three
2501 bytes which we'll fill in when we get to after `c'. */
2504 STORE_JUMP (jump_past_alt
, fixup_alt_jump
, b
);
2506 /* Mark and leave space for a jump after this alternative,
2507 to be filled in later either by next alternative or
2508 when know we're at the end of a series of alternatives. */
2510 GET_BUFFER_SPACE (3);
2519 /* If \{ is a literal. */
2520 if (!(syntax
& RE_INTERVALS
)
2521 /* If we're at `\{' and it's not the open-interval
2523 || ((syntax
& RE_INTERVALS
) && (syntax
& RE_NO_BK_BRACES
))
2524 || (p
- 2 == pattern
&& p
== pend
))
2525 goto normal_backslash
;
2529 /* If got here, then the syntax allows intervals. */
2531 /* At least (most) this many matches must be made. */
2532 int lower_bound
= -1, upper_bound
= -1;
2534 beg_interval
= p
- 1;
2538 if (syntax
& RE_NO_BK_BRACES
)
2539 goto unfetch_interval
;
2541 FREE_STACK_RETURN (REG_EBRACE
);
2544 GET_UNSIGNED_NUMBER (lower_bound
);
2548 GET_UNSIGNED_NUMBER (upper_bound
);
2549 if (upper_bound
< 0) upper_bound
= RE_DUP_MAX
;
2552 /* Interval such as `{1}' => match exactly once. */
2553 upper_bound
= lower_bound
;
2555 if (lower_bound
< 0 || upper_bound
> RE_DUP_MAX
2556 || lower_bound
> upper_bound
)
2558 if (syntax
& RE_NO_BK_BRACES
)
2559 goto unfetch_interval
;
2561 FREE_STACK_RETURN (REG_BADBR
);
2564 if (!(syntax
& RE_NO_BK_BRACES
))
2566 if (c
!= '\\') FREE_STACK_RETURN (REG_EBRACE
);
2573 if (syntax
& RE_NO_BK_BRACES
)
2574 goto unfetch_interval
;
2576 FREE_STACK_RETURN (REG_BADBR
);
2579 /* We just parsed a valid interval. */
2581 /* If it's invalid to have no preceding re. */
2584 if (syntax
& RE_CONTEXT_INVALID_OPS
)
2585 FREE_STACK_RETURN (REG_BADRPT
);
2586 else if (syntax
& RE_CONTEXT_INDEP_OPS
)
2589 goto unfetch_interval
;
2592 /* If the upper bound is zero, don't want to succeed at
2593 all; jump from `laststart' to `b + 3', which will be
2594 the end of the buffer after we insert the jump. */
2595 if (upper_bound
== 0)
2597 GET_BUFFER_SPACE (3);
2598 INSERT_JUMP (jump
, laststart
, b
+ 3);
2602 /* Otherwise, we have a nontrivial interval. When
2603 we're all done, the pattern will look like:
2604 set_number_at <jump count> <upper bound>
2605 set_number_at <succeed_n count> <lower bound>
2606 succeed_n <after jump addr> <succeed_n count>
2608 jump_n <succeed_n addr> <jump count>
2609 (The upper bound and `jump_n' are omitted if
2610 `upper_bound' is 1, though.) */
2612 { /* If the upper bound is > 1, we need to insert
2613 more at the end of the loop. */
2614 unsigned nbytes
= 10 + (upper_bound
> 1) * 10;
2616 GET_BUFFER_SPACE (nbytes
);
2618 /* Initialize lower bound of the `succeed_n', even
2619 though it will be set during matching by its
2620 attendant `set_number_at' (inserted next),
2621 because `re_compile_fastmap' needs to know.
2622 Jump to the `jump_n' we might insert below. */
2623 INSERT_JUMP2 (succeed_n
, laststart
,
2624 b
+ 5 + (upper_bound
> 1) * 5,
2628 /* Code to initialize the lower bound. Insert
2629 before the `succeed_n'. The `5' is the last two
2630 bytes of this `set_number_at', plus 3 bytes of
2631 the following `succeed_n'. */
2632 insert_op2 (set_number_at
, laststart
, 5, lower_bound
, b
);
2635 if (upper_bound
> 1)
2636 { /* More than one repetition is allowed, so
2637 append a backward jump to the `succeed_n'
2638 that starts this interval.
2640 When we've reached this during matching,
2641 we'll have matched the interval once, so
2642 jump back only `upper_bound - 1' times. */
2643 STORE_JUMP2 (jump_n
, b
, laststart
+ 5,
2647 /* The location we want to set is the second
2648 parameter of the `jump_n'; that is `b-2' as
2649 an absolute address. `laststart' will be
2650 the `set_number_at' we're about to insert;
2651 `laststart+3' the number to set, the source
2652 for the relative address. But we are
2653 inserting into the middle of the pattern --
2654 so everything is getting moved up by 5.
2655 Conclusion: (b - 2) - (laststart + 3) + 5,
2656 i.e., b - laststart.
2658 We insert this at the beginning of the loop
2659 so that if we fail during matching, we'll
2660 reinitialize the bounds. */
2661 insert_op2 (set_number_at
, laststart
, b
- laststart
,
2662 upper_bound
- 1, b
);
2667 beg_interval
= NULL
;
2672 /* If an invalid interval, match the characters as literals. */
2673 assert (beg_interval
);
2675 beg_interval
= NULL
;
2677 /* normal_char and normal_backslash need `c'. */
2680 if (!(syntax
& RE_NO_BK_BRACES
))
2682 if (p
> pattern
&& p
[-1] == '\\')
2683 goto normal_backslash
;
2688 /* There is no way to specify the before_dot and after_dot
2689 operators. rms says this is ok. --karl */
2697 BUF_PUSH_2 (syntaxspec
, syntax_spec_code
[c
]);
2703 BUF_PUSH_2 (notsyntaxspec
, syntax_spec_code
[c
]);
2709 if (syntax
& RE_NO_GNU_OPS
)
2712 BUF_PUSH (wordchar
);
2717 if (syntax
& RE_NO_GNU_OPS
)
2720 BUF_PUSH (notwordchar
);
2725 if (syntax
& RE_NO_GNU_OPS
)
2731 if (syntax
& RE_NO_GNU_OPS
)
2737 if (syntax
& RE_NO_GNU_OPS
)
2739 BUF_PUSH (wordbound
);
2743 if (syntax
& RE_NO_GNU_OPS
)
2745 BUF_PUSH (notwordbound
);
2749 if (syntax
& RE_NO_GNU_OPS
)
2755 if (syntax
& RE_NO_GNU_OPS
)
2760 case '1': case '2': case '3': case '4': case '5':
2761 case '6': case '7': case '8': case '9':
2762 if (syntax
& RE_NO_BK_REFS
)
2768 FREE_STACK_RETURN (REG_ESUBREG
);
2770 /* Can't back reference to a subexpression if inside of it. */
2771 if (group_in_compile_stack (compile_stack
, (regnum_t
) c1
))
2775 BUF_PUSH_2 (duplicate
, c1
);
2781 if (syntax
& RE_BK_PLUS_QM
)
2784 goto normal_backslash
;
2788 /* You might think it would be useful for \ to mean
2789 not to translate; but if we don't translate it
2790 it will never match anything. */
2798 /* Expects the character in `c'. */
2800 /* If no exactn currently being built. */
2803 /* If last exactn not at current position. */
2804 || pending_exact
+ *pending_exact
+ 1 != b
2806 /* We have only one byte following the exactn for the count. */
2807 || *pending_exact
== (1 << BYTEWIDTH
) - 1
2809 /* If followed by a repetition operator. */
2810 || *p
== '*' || *p
== '^'
2811 || ((syntax
& RE_BK_PLUS_QM
)
2812 ? *p
== '\\' && (p
[1] == '+' || p
[1] == '?')
2813 : (*p
== '+' || *p
== '?'))
2814 || ((syntax
& RE_INTERVALS
)
2815 && ((syntax
& RE_NO_BK_BRACES
)
2817 : (p
[0] == '\\' && p
[1] == '{'))))
2819 /* Start building a new exactn. */
2823 BUF_PUSH_2 (exactn
, 0);
2824 pending_exact
= b
- 1;
2831 } /* while p != pend */
2834 /* Through the pattern now. */
2837 STORE_JUMP (jump_past_alt
, fixup_alt_jump
, b
);
2839 if (!COMPILE_STACK_EMPTY
)
2840 FREE_STACK_RETURN (REG_EPAREN
);
2842 /* If we don't want backtracking, force success
2843 the first time we reach the end of the compiled pattern. */
2844 if (syntax
& RE_NO_POSIX_BACKTRACKING
)
2847 free (compile_stack
.stack
);
2849 /* We have succeeded; set the length of the buffer. */
2850 bufp
->used
= b
- bufp
->buffer
;
2855 DEBUG_PRINT1 ("\nCompiled pattern: \n");
2856 print_compiled_pattern (bufp
);
2860 #ifndef MATCH_MAY_ALLOCATE
2861 /* Initialize the failure stack to the largest possible stack. This
2862 isn't necessary unless we're trying to avoid calling alloca in
2863 the search and match routines. */
2865 int num_regs
= bufp
->re_nsub
+ 1;
2867 /* Since DOUBLE_FAIL_STACK refuses to double only if the current size
2868 is strictly greater than re_max_failures, the largest possible stack
2869 is 2 * re_max_failures failure points. */
2870 if (fail_stack
.size
< (2 * re_max_failures
* MAX_FAILURE_ITEMS
))
2872 fail_stack
.size
= (2 * re_max_failures
* MAX_FAILURE_ITEMS
);
2875 if (! fail_stack
.stack
)
2877 = (fail_stack_elt_t
*) xmalloc (fail_stack
.size
2878 * sizeof (fail_stack_elt_t
));
2881 = (fail_stack_elt_t
*) xrealloc (fail_stack
.stack
,
2883 * sizeof (fail_stack_elt_t
)));
2884 # else /* not emacs */
2885 if (! fail_stack
.stack
)
2887 = (fail_stack_elt_t
*) malloc (fail_stack
.size
2888 * sizeof (fail_stack_elt_t
));
2891 = (fail_stack_elt_t
*) realloc (fail_stack
.stack
,
2893 * sizeof (fail_stack_elt_t
)));
2894 # endif /* not emacs */
2897 regex_grow_registers (num_regs
);
2899 #endif /* not MATCH_MAY_ALLOCATE */
2902 } /* regex_compile */
2904 /* Subroutines for `regex_compile'. */
2906 /* Store OP at LOC followed by two-byte integer parameter ARG. */
2909 store_op1 (op
, loc
, arg
)
2914 *loc
= (unsigned char) op
;
2915 STORE_NUMBER (loc
+ 1, arg
);
2919 /* Like `store_op1', but for two two-byte parameters ARG1 and ARG2. */
2922 store_op2 (op
, loc
, arg1
, arg2
)
2927 *loc
= (unsigned char) op
;
2928 STORE_NUMBER (loc
+ 1, arg1
);
2929 STORE_NUMBER (loc
+ 3, arg2
);
2933 /* Copy the bytes from LOC to END to open up three bytes of space at LOC
2934 for OP followed by two-byte integer parameter ARG. */
2937 insert_op1 (op
, loc
, arg
, end
)
2943 register unsigned char *pfrom
= end
;
2944 register unsigned char *pto
= end
+ 3;
2946 while (pfrom
!= loc
)
2949 store_op1 (op
, loc
, arg
);
2953 /* Like `insert_op1', but for two two-byte parameters ARG1 and ARG2. */
2956 insert_op2 (op
, loc
, arg1
, arg2
, end
)
2962 register unsigned char *pfrom
= end
;
2963 register unsigned char *pto
= end
+ 5;
2965 while (pfrom
!= loc
)
2968 store_op2 (op
, loc
, arg1
, arg2
);
2972 /* P points to just after a ^ in PATTERN. Return true if that ^ comes
2973 after an alternative or a begin-subexpression. We assume there is at
2974 least one character before the ^. */
2977 at_begline_loc_p (pattern
, p
, syntax
)
2978 const char *pattern
, *p
;
2979 reg_syntax_t syntax
;
2981 const char *prev
= p
- 2;
2982 boolean prev_prev_backslash
= prev
> pattern
&& prev
[-1] == '\\';
2985 /* After a subexpression? */
2986 (*prev
== '(' && (syntax
& RE_NO_BK_PARENS
|| prev_prev_backslash
))
2987 /* After an alternative? */
2988 || (*prev
== '|' && (syntax
& RE_NO_BK_VBAR
|| prev_prev_backslash
));
2992 /* The dual of at_begline_loc_p. This one is for $. We assume there is
2993 at least one character after the $, i.e., `P < PEND'. */
2996 at_endline_loc_p (p
, pend
, syntax
)
2997 const char *p
, *pend
;
2998 reg_syntax_t syntax
;
3000 const char *next
= p
;
3001 boolean next_backslash
= *next
== '\\';
3002 const char *next_next
= p
+ 1 < pend
? p
+ 1 : 0;
3005 /* Before a subexpression? */
3006 (syntax
& RE_NO_BK_PARENS
? *next
== ')'
3007 : next_backslash
&& next_next
&& *next_next
== ')')
3008 /* Before an alternative? */
3009 || (syntax
& RE_NO_BK_VBAR
? *next
== '|'
3010 : next_backslash
&& next_next
&& *next_next
== '|');
3014 /* Returns true if REGNUM is in one of COMPILE_STACK's elements and
3015 false if it's not. */
3018 group_in_compile_stack (compile_stack
, regnum
)
3019 compile_stack_type compile_stack
;
3024 for (this_element
= compile_stack
.avail
- 1;
3027 if (compile_stack
.stack
[this_element
].regnum
== regnum
)
3034 /* Read the ending character of a range (in a bracket expression) from the
3035 uncompiled pattern *P_PTR (which ends at PEND). We assume the
3036 starting character is in `P[-2]'. (`P[-1]' is the character `-'.)
3037 Then we set the translation of all bits between the starting and
3038 ending characters (inclusive) in the compiled pattern B.
3040 Return an error code.
3042 We use these short variable names so we can use the same macros as
3043 `regex_compile' itself. */
3045 static reg_errcode_t
3046 compile_range (p_ptr
, pend
, translate
, syntax
, b
)
3047 const char **p_ptr
, *pend
;
3048 RE_TRANSLATE_TYPE translate
;
3049 reg_syntax_t syntax
;
3054 const char *p
= *p_ptr
;
3055 unsigned int range_start
, range_end
;
3060 /* Even though the pattern is a signed `char *', we need to fetch
3061 with unsigned char *'s; if the high bit of the pattern character
3062 is set, the range endpoints will be negative if we fetch using a
3065 We also want to fetch the endpoints without translating them; the
3066 appropriate translation is done in the bit-setting loop below. */
3067 /* The SVR4 compiler on the 3B2 had trouble with unsigned const char *. */
3068 range_start
= ((const unsigned char *) p
)[-2];
3069 range_end
= ((const unsigned char *) p
)[0];
3071 /* Have to increment the pointer into the pattern string, so the
3072 caller isn't still at the ending character. */
3075 /* If the start is after the end, the range is empty. */
3076 if (range_start
> range_end
)
3077 return syntax
& RE_NO_EMPTY_RANGES
? REG_ERANGE
: REG_NOERROR
;
3079 /* Here we see why `this_char' has to be larger than an `unsigned
3080 char' -- the range is inclusive, so if `range_end' == 0xff
3081 (assuming 8-bit characters), we would otherwise go into an infinite
3082 loop, since all characters <= 0xff. */
3083 for (this_char
= range_start
; this_char
<= range_end
; this_char
++)
3085 SET_LIST_BIT (TRANSLATE (this_char
));
3091 /* re_compile_fastmap computes a ``fastmap'' for the compiled pattern in
3092 BUFP. A fastmap records which of the (1 << BYTEWIDTH) possible
3093 characters can start a string that matches the pattern. This fastmap
3094 is used by re_search to skip quickly over impossible starting points.
3096 The caller must supply the address of a (1 << BYTEWIDTH)-byte data
3097 area as BUFP->fastmap.
3099 We set the `fastmap', `fastmap_accurate', and `can_be_null' fields in
3102 Returns 0 if we succeed, -2 if an internal error. */
3105 re_compile_fastmap (bufp
)
3106 struct re_pattern_buffer
*bufp
;
3109 #ifdef MATCH_MAY_ALLOCATE
3110 fail_stack_type fail_stack
;
3112 #ifndef REGEX_MALLOC
3116 register char *fastmap
= bufp
->fastmap
;
3117 unsigned char *pattern
= bufp
->buffer
;
3118 unsigned char *p
= pattern
;
3119 register unsigned char *pend
= pattern
+ bufp
->used
;
3122 /* This holds the pointer to the failure stack, when
3123 it is allocated relocatably. */
3124 fail_stack_elt_t
*failure_stack_ptr
;
3127 /* Assume that each path through the pattern can be null until
3128 proven otherwise. We set this false at the bottom of switch
3129 statement, to which we get only if a particular path doesn't
3130 match the empty string. */
3131 boolean path_can_be_null
= true;
3133 /* We aren't doing a `succeed_n' to begin with. */
3134 boolean succeed_n_p
= false;
3136 assert (fastmap
!= NULL
&& p
!= NULL
);
3139 bzero (fastmap
, 1 << BYTEWIDTH
); /* Assume nothing's valid. */
3140 bufp
->fastmap_accurate
= 1; /* It will be when we're done. */
3141 bufp
->can_be_null
= 0;
3145 if (p
== pend
|| *p
== succeed
)
3147 /* We have reached the (effective) end of pattern. */
3148 if (!FAIL_STACK_EMPTY ())
3150 bufp
->can_be_null
|= path_can_be_null
;
3152 /* Reset for next path. */
3153 path_can_be_null
= true;
3155 p
= fail_stack
.stack
[--fail_stack
.avail
].pointer
;
3163 /* We should never be about to go beyond the end of the pattern. */
3166 switch (SWITCH_ENUM_CAST ((re_opcode_t
) *p
++))
3169 /* I guess the idea here is to simply not bother with a fastmap
3170 if a backreference is used, since it's too hard to figure out
3171 the fastmap for the corresponding group. Setting
3172 `can_be_null' stops `re_search_2' from using the fastmap, so
3173 that is all we do. */
3175 bufp
->can_be_null
= 1;
3179 /* Following are the cases which match a character. These end
3188 for (j
= *p
++ * BYTEWIDTH
- 1; j
>= 0; j
--)
3189 if (p
[j
/ BYTEWIDTH
] & (1 << (j
% BYTEWIDTH
)))
3195 /* Chars beyond end of map must be allowed. */
3196 for (j
= *p
* BYTEWIDTH
; j
< (1 << BYTEWIDTH
); j
++)
3199 for (j
= *p
++ * BYTEWIDTH
- 1; j
>= 0; j
--)
3200 if (!(p
[j
/ BYTEWIDTH
] & (1 << (j
% BYTEWIDTH
))))
3206 for (j
= 0; j
< (1 << BYTEWIDTH
); j
++)
3207 if (SYNTAX (j
) == Sword
)
3213 for (j
= 0; j
< (1 << BYTEWIDTH
); j
++)
3214 if (SYNTAX (j
) != Sword
)
3221 int fastmap_newline
= fastmap
['\n'];
3223 /* `.' matches anything ... */
3224 for (j
= 0; j
< (1 << BYTEWIDTH
); j
++)
3227 /* ... except perhaps newline. */
3228 if (!(bufp
->syntax
& RE_DOT_NEWLINE
))
3229 fastmap
['\n'] = fastmap_newline
;
3231 /* Return if we have already set `can_be_null'; if we have,
3232 then the fastmap is irrelevant. Something's wrong here. */
3233 else if (bufp
->can_be_null
)
3236 /* Otherwise, have to check alternative paths. */
3243 for (j
= 0; j
< (1 << BYTEWIDTH
); j
++)
3244 if (SYNTAX (j
) == (enum syntaxcode
) k
)
3251 for (j
= 0; j
< (1 << BYTEWIDTH
); j
++)
3252 if (SYNTAX (j
) != (enum syntaxcode
) k
)
3257 /* All cases after this match the empty string. These end with
3277 case push_dummy_failure
:
3282 case pop_failure_jump
:
3283 case maybe_pop_jump
:
3286 case dummy_failure_jump
:
3287 EXTRACT_NUMBER_AND_INCR (j
, p
);
3292 /* Jump backward implies we just went through the body of a
3293 loop and matched nothing. Opcode jumped to should be
3294 `on_failure_jump' or `succeed_n'. Just treat it like an
3295 ordinary jump. For a * loop, it has pushed its failure
3296 point already; if so, discard that as redundant. */
3297 if ((re_opcode_t
) *p
!= on_failure_jump
3298 && (re_opcode_t
) *p
!= succeed_n
)
3302 EXTRACT_NUMBER_AND_INCR (j
, p
);
3305 /* If what's on the stack is where we are now, pop it. */
3306 if (!FAIL_STACK_EMPTY ()
3307 && fail_stack
.stack
[fail_stack
.avail
- 1].pointer
== p
)
3313 case on_failure_jump
:
3314 case on_failure_keep_string_jump
:
3315 handle_on_failure_jump
:
3316 EXTRACT_NUMBER_AND_INCR (j
, p
);
3318 /* For some patterns, e.g., `(a?)?', `p+j' here points to the
3319 end of the pattern. We don't want to push such a point,
3320 since when we restore it above, entering the switch will
3321 increment `p' past the end of the pattern. We don't need
3322 to push such a point since we obviously won't find any more
3323 fastmap entries beyond `pend'. Such a pattern can match
3324 the null string, though. */
3327 if (!PUSH_PATTERN_OP (p
+ j
, fail_stack
))
3329 RESET_FAIL_STACK ();
3334 bufp
->can_be_null
= 1;
3338 EXTRACT_NUMBER_AND_INCR (k
, p
); /* Skip the n. */
3339 succeed_n_p
= false;
3346 /* Get to the number of times to succeed. */
3349 /* Increment p past the n for when k != 0. */
3350 EXTRACT_NUMBER_AND_INCR (k
, p
);
3354 succeed_n_p
= true; /* Spaghetti code alert. */
3355 goto handle_on_failure_jump
;
3372 abort (); /* We have listed all the cases. */
3375 /* Getting here means we have found the possible starting
3376 characters for one path of the pattern -- and that the empty
3377 string does not match. We need not follow this path further.
3378 Instead, look at the next alternative (remembered on the
3379 stack), or quit if no more. The test at the top of the loop
3380 does these things. */
3381 path_can_be_null
= false;
3385 /* Set `can_be_null' for the last path (also the first path, if the
3386 pattern is empty). */
3387 bufp
->can_be_null
|= path_can_be_null
;
3390 RESET_FAIL_STACK ();
3392 } /* re_compile_fastmap */
3394 weak_alias (__re_compile_fastmap
, re_compile_fastmap
)
3397 /* Set REGS to hold NUM_REGS registers, storing them in STARTS and
3398 ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use
3399 this memory for recording register information. STARTS and ENDS
3400 must be allocated using the malloc library routine, and must each
3401 be at least NUM_REGS * sizeof (regoff_t) bytes long.
3403 If NUM_REGS == 0, then subsequent matches should allocate their own
3406 Unless this function is called, the first search or match using
3407 PATTERN_BUFFER will allocate its own register data, without
3408 freeing the old data. */
3411 re_set_registers (bufp
, regs
, num_regs
, starts
, ends
)
3412 struct re_pattern_buffer
*bufp
;
3413 struct re_registers
*regs
;
3415 regoff_t
*starts
, *ends
;
3419 bufp
->regs_allocated
= REGS_REALLOCATE
;
3420 regs
->num_regs
= num_regs
;
3421 regs
->start
= starts
;
3426 bufp
->regs_allocated
= REGS_UNALLOCATED
;
3428 regs
->start
= regs
->end
= (regoff_t
*) 0;
3432 weak_alias (__re_set_registers
, re_set_registers
)
3435 /* Searching routines. */
3437 /* Like re_search_2, below, but only one string is specified, and
3438 doesn't let you say where to stop matching. */
3441 re_search (bufp
, string
, size
, startpos
, range
, regs
)
3442 struct re_pattern_buffer
*bufp
;
3444 int size
, startpos
, range
;
3445 struct re_registers
*regs
;
3447 return re_search_2 (bufp
, NULL
, 0, string
, size
, startpos
, range
,
3451 weak_alias (__re_search
, re_search
)
3455 /* Using the compiled pattern in BUFP->buffer, first tries to match the
3456 virtual concatenation of STRING1 and STRING2, starting first at index
3457 STARTPOS, then at STARTPOS + 1, and so on.
3459 STRING1 and STRING2 have length SIZE1 and SIZE2, respectively.
3461 RANGE is how far to scan while trying to match. RANGE = 0 means try
3462 only at STARTPOS; in general, the last start tried is STARTPOS +
3465 In REGS, return the indices of the virtual concatenation of STRING1
3466 and STRING2 that matched the entire BUFP->buffer and its contained
3469 Do not consider matching one past the index STOP in the virtual
3470 concatenation of STRING1 and STRING2.
3472 We return either the position in the strings at which the match was
3473 found, -1 if no match, or -2 if error (such as failure
3477 re_search_2 (bufp
, string1
, size1
, string2
, size2
, startpos
, range
, regs
, stop
)
3478 struct re_pattern_buffer
*bufp
;
3479 const char *string1
, *string2
;
3483 struct re_registers
*regs
;
3487 register char *fastmap
= bufp
->fastmap
;
3488 register RE_TRANSLATE_TYPE translate
= bufp
->translate
;
3489 int total_size
= size1
+ size2
;
3490 int endpos
= startpos
+ range
;
3492 /* Check for out-of-range STARTPOS. */
3493 if (startpos
< 0 || startpos
> total_size
)
3496 /* Fix up RANGE if it might eventually take us outside
3497 the virtual concatenation of STRING1 and STRING2.
3498 Make sure we won't move STARTPOS below 0 or above TOTAL_SIZE. */
3500 range
= 0 - startpos
;
3501 else if (endpos
> total_size
)
3502 range
= total_size
- startpos
;
3504 /* If the search isn't to be a backwards one, don't waste time in a
3505 search for a pattern that must be anchored. */
3506 if (bufp
->used
> 0 && (re_opcode_t
) bufp
->buffer
[0] == begbuf
&& range
> 0)
3515 /* In a forward search for something that starts with \=.
3516 don't keep searching past point. */
3517 if (bufp
->used
> 0 && (re_opcode_t
) bufp
->buffer
[0] == at_dot
&& range
> 0)
3519 range
= PT
- startpos
;
3525 /* Update the fastmap now if not correct already. */
3526 if (fastmap
&& !bufp
->fastmap_accurate
)
3527 if (re_compile_fastmap (bufp
) == -2)
3530 /* Loop through the string, looking for a place to start matching. */
3533 /* If a fastmap is supplied, skip quickly over characters that
3534 cannot be the start of a match. If the pattern can match the
3535 null string, however, we don't need to skip characters; we want
3536 the first null string. */
3537 if (fastmap
&& startpos
< total_size
&& !bufp
->can_be_null
)
3539 if (range
> 0) /* Searching forwards. */
3541 register const char *d
;
3542 register int lim
= 0;
3545 if (startpos
< size1
&& startpos
+ range
>= size1
)
3546 lim
= range
- (size1
- startpos
);
3548 d
= (startpos
>= size1
? string2
- size1
: string1
) + startpos
;
3550 /* Written out as an if-else to avoid testing `translate'
3554 && !fastmap
[(unsigned char)
3555 translate
[(unsigned char) *d
++]])
3558 while (range
> lim
&& !fastmap
[(unsigned char) *d
++])
3561 startpos
+= irange
- range
;
3563 else /* Searching backwards. */
3565 register char c
= (size1
== 0 || startpos
>= size1
3566 ? string2
[startpos
- size1
]
3567 : string1
[startpos
]);
3569 if (!fastmap
[(unsigned char) TRANSLATE (c
)])
3574 /* If can't match the null string, and that's all we have left, fail. */
3575 if (range
>= 0 && startpos
== total_size
&& fastmap
3576 && !bufp
->can_be_null
)
3579 val
= re_match_2_internal (bufp
, string1
, size1
, string2
, size2
,
3580 startpos
, regs
, stop
);
3581 #ifndef REGEX_MALLOC
3610 weak_alias (__re_search_2
, re_search_2
)
3613 /* This converts PTR, a pointer into one of the search strings `string1'
3614 and `string2' into an offset from the beginning of that string. */
3615 #define POINTER_TO_OFFSET(ptr) \
3616 (FIRST_STRING_P (ptr) \
3617 ? ((regoff_t) ((ptr) - string1)) \
3618 : ((regoff_t) ((ptr) - string2 + size1)))
3620 /* Macros for dealing with the split strings in re_match_2. */
3622 #define MATCHING_IN_FIRST_STRING (dend == end_match_1)
3624 /* Call before fetching a character with *d. This switches over to
3625 string2 if necessary. */
3626 #define PREFETCH() \
3629 /* End of string2 => fail. */ \
3630 if (dend == end_match_2) \
3632 /* End of string1 => advance to string2. */ \
3634 dend = end_match_2; \
3638 /* Test if at very beginning or at very end of the virtual concatenation
3639 of `string1' and `string2'. If only one string, it's `string2'. */
3640 #define AT_STRINGS_BEG(d) ((d) == (size1 ? string1 : string2) || !size2)
3641 #define AT_STRINGS_END(d) ((d) == end2)
3644 /* Test if D points to a character which is word-constituent. We have
3645 two special cases to check for: if past the end of string1, look at
3646 the first character in string2; and if before the beginning of
3647 string2, look at the last character in string1. */
3648 #define WORDCHAR_P(d) \
3649 (SYNTAX ((d) == end1 ? *string2 \
3650 : (d) == string2 - 1 ? *(end1 - 1) : *(d)) \
3653 /* Disabled due to a compiler bug -- see comment at case wordbound */
3655 /* Test if the character before D and the one at D differ with respect
3656 to being word-constituent. */
3657 #define AT_WORD_BOUNDARY(d) \
3658 (AT_STRINGS_BEG (d) || AT_STRINGS_END (d) \
3659 || WORDCHAR_P (d - 1) != WORDCHAR_P (d))
3662 /* Free everything we malloc. */
3663 #ifdef MATCH_MAY_ALLOCATE
3664 # define FREE_VAR(var) if (var) REGEX_FREE (var); var = NULL
3665 # define FREE_VARIABLES() \
3667 REGEX_FREE_STACK (fail_stack.stack); \
3668 FREE_VAR (regstart); \
3669 FREE_VAR (regend); \
3670 FREE_VAR (old_regstart); \
3671 FREE_VAR (old_regend); \
3672 FREE_VAR (best_regstart); \
3673 FREE_VAR (best_regend); \
3674 FREE_VAR (reg_info); \
3675 FREE_VAR (reg_dummy); \
3676 FREE_VAR (reg_info_dummy); \
3679 # define FREE_VARIABLES() ((void)0) /* Do nothing! But inhibit gcc warning. */
3680 #endif /* not MATCH_MAY_ALLOCATE */
3682 /* These values must meet several constraints. They must not be valid
3683 register values; since we have a limit of 255 registers (because
3684 we use only one byte in the pattern for the register number), we can
3685 use numbers larger than 255. They must differ by 1, because of
3686 NUM_FAILURE_ITEMS above. And the value for the lowest register must
3687 be larger than the value for the highest register, so we do not try
3688 to actually save any registers when none are active. */
3689 #define NO_HIGHEST_ACTIVE_REG (1 << BYTEWIDTH)
3690 #define NO_LOWEST_ACTIVE_REG (NO_HIGHEST_ACTIVE_REG + 1)
3692 /* Matching routines. */
3694 #ifndef emacs /* Emacs never uses this. */
3695 /* re_match is like re_match_2 except it takes only a single string. */
3698 re_match (bufp
, string
, size
, pos
, regs
)
3699 struct re_pattern_buffer
*bufp
;
3702 struct re_registers
*regs
;
3704 int result
= re_match_2_internal (bufp
, NULL
, 0, string
, size
,
3706 # ifndef REGEX_MALLOC
3714 weak_alias (__re_match
, re_match
)
3716 #endif /* not emacs */
3718 static boolean group_match_null_string_p
_RE_ARGS ((unsigned char **p
,
3720 register_info_type
*reg_info
));
3721 static boolean alt_match_null_string_p
_RE_ARGS ((unsigned char *p
,
3723 register_info_type
*reg_info
));
3724 static boolean common_op_match_null_string_p
_RE_ARGS ((unsigned char **p
,
3726 register_info_type
*reg_info
));
3727 static int bcmp_translate
_RE_ARGS ((const char *s1
, const char *s2
,
3728 int len
, char *translate
));
3730 /* re_match_2 matches the compiled pattern in BUFP against the
3731 the (virtual) concatenation of STRING1 and STRING2 (of length SIZE1
3732 and SIZE2, respectively). We start matching at POS, and stop
3735 If REGS is non-null and the `no_sub' field of BUFP is nonzero, we
3736 store offsets for the substring each group matched in REGS. See the
3737 documentation for exactly how many groups we fill.
3739 We return -1 if no match, -2 if an internal error (such as the
3740 failure stack overflowing). Otherwise, we return the length of the
3741 matched substring. */
3744 re_match_2 (bufp
, string1
, size1
, string2
, size2
, pos
, regs
, stop
)
3745 struct re_pattern_buffer
*bufp
;
3746 const char *string1
, *string2
;
3749 struct re_registers
*regs
;
3752 int result
= re_match_2_internal (bufp
, string1
, size1
, string2
, size2
,
3754 #ifndef REGEX_MALLOC
3762 weak_alias (__re_match_2
, re_match_2
)
3765 /* This is a separate function so that we can force an alloca cleanup
3768 re_match_2_internal (bufp
, string1
, size1
, string2
, size2
, pos
, regs
, stop
)
3769 struct re_pattern_buffer
*bufp
;
3770 const char *string1
, *string2
;
3773 struct re_registers
*regs
;
3776 /* General temporaries. */
3780 /* Just past the end of the corresponding string. */
3781 const char *end1
, *end2
;
3783 /* Pointers into string1 and string2, just past the last characters in
3784 each to consider matching. */
3785 const char *end_match_1
, *end_match_2
;
3787 /* Where we are in the data, and the end of the current string. */
3788 const char *d
, *dend
;
3790 /* Where we are in the pattern, and the end of the pattern. */
3791 unsigned char *p
= bufp
->buffer
;
3792 register unsigned char *pend
= p
+ bufp
->used
;
3794 /* Mark the opcode just after a start_memory, so we can test for an
3795 empty subpattern when we get to the stop_memory. */
3796 unsigned char *just_past_start_mem
= 0;
3798 /* We use this to map every character in the string. */
3799 RE_TRANSLATE_TYPE translate
= bufp
->translate
;
3801 /* Failure point stack. Each place that can handle a failure further
3802 down the line pushes a failure point on this stack. It consists of
3803 restart, regend, and reg_info for all registers corresponding to
3804 the subexpressions we're currently inside, plus the number of such
3805 registers, and, finally, two char *'s. The first char * is where
3806 to resume scanning the pattern; the second one is where to resume
3807 scanning the strings. If the latter is zero, the failure point is
3808 a ``dummy''; if a failure happens and the failure point is a dummy,
3809 it gets discarded and the next next one is tried. */
3810 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
3811 fail_stack_type fail_stack
;
3814 static unsigned failure_id
= 0;
3815 unsigned nfailure_points_pushed
= 0, nfailure_points_popped
= 0;
3819 /* This holds the pointer to the failure stack, when
3820 it is allocated relocatably. */
3821 fail_stack_elt_t
*failure_stack_ptr
;
3824 /* We fill all the registers internally, independent of what we
3825 return, for use in backreferences. The number here includes
3826 an element for register zero. */
3827 size_t num_regs
= bufp
->re_nsub
+ 1;
3829 /* The currently active registers. */
3830 active_reg_t lowest_active_reg
= NO_LOWEST_ACTIVE_REG
;
3831 active_reg_t highest_active_reg
= NO_HIGHEST_ACTIVE_REG
;
3833 /* Information on the contents of registers. These are pointers into
3834 the input strings; they record just what was matched (on this
3835 attempt) by a subexpression part of the pattern, that is, the
3836 regnum-th regstart pointer points to where in the pattern we began
3837 matching and the regnum-th regend points to right after where we
3838 stopped matching the regnum-th subexpression. (The zeroth register
3839 keeps track of what the whole pattern matches.) */
3840 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
3841 const char **regstart
, **regend
;
3844 /* If a group that's operated upon by a repetition operator fails to
3845 match anything, then the register for its start will need to be
3846 restored because it will have been set to wherever in the string we
3847 are when we last see its open-group operator. Similarly for a
3849 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
3850 const char **old_regstart
, **old_regend
;
3853 /* The is_active field of reg_info helps us keep track of which (possibly
3854 nested) subexpressions we are currently in. The matched_something
3855 field of reg_info[reg_num] helps us tell whether or not we have
3856 matched any of the pattern so far this time through the reg_num-th
3857 subexpression. These two fields get reset each time through any
3858 loop their register is in. */
3859 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
3860 register_info_type
*reg_info
;
3863 /* The following record the register info as found in the above
3864 variables when we find a match better than any we've seen before.
3865 This happens as we backtrack through the failure points, which in
3866 turn happens only if we have not yet matched the entire string. */
3867 unsigned best_regs_set
= false;
3868 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
3869 const char **best_regstart
, **best_regend
;
3872 /* Logically, this is `best_regend[0]'. But we don't want to have to
3873 allocate space for that if we're not allocating space for anything
3874 else (see below). Also, we never need info about register 0 for
3875 any of the other register vectors, and it seems rather a kludge to
3876 treat `best_regend' differently than the rest. So we keep track of
3877 the end of the best match so far in a separate variable. We
3878 initialize this to NULL so that when we backtrack the first time
3879 and need to test it, it's not garbage. */
3880 const char *match_end
= NULL
;
3882 /* This helps SET_REGS_MATCHED avoid doing redundant work. */
3883 int set_regs_matched_done
= 0;
3885 /* Used when we pop values we don't care about. */
3886 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
3887 const char **reg_dummy
;
3888 register_info_type
*reg_info_dummy
;
3892 /* Counts the total number of registers pushed. */
3893 unsigned num_regs_pushed
= 0;
3896 DEBUG_PRINT1 ("\n\nEntering re_match_2.\n");
3900 #ifdef MATCH_MAY_ALLOCATE
3901 /* Do not bother to initialize all the register variables if there are
3902 no groups in the pattern, as it takes a fair amount of time. If
3903 there are groups, we include space for register 0 (the whole
3904 pattern), even though we never use it, since it simplifies the
3905 array indexing. We should fix this. */
3908 regstart
= REGEX_TALLOC (num_regs
, const char *);
3909 regend
= REGEX_TALLOC (num_regs
, const char *);
3910 old_regstart
= REGEX_TALLOC (num_regs
, const char *);
3911 old_regend
= REGEX_TALLOC (num_regs
, const char *);
3912 best_regstart
= REGEX_TALLOC (num_regs
, const char *);
3913 best_regend
= REGEX_TALLOC (num_regs
, const char *);
3914 reg_info
= REGEX_TALLOC (num_regs
, register_info_type
);
3915 reg_dummy
= REGEX_TALLOC (num_regs
, const char *);
3916 reg_info_dummy
= REGEX_TALLOC (num_regs
, register_info_type
);
3918 if (!(regstart
&& regend
&& old_regstart
&& old_regend
&& reg_info
3919 && best_regstart
&& best_regend
&& reg_dummy
&& reg_info_dummy
))
3927 /* We must initialize all our variables to NULL, so that
3928 `FREE_VARIABLES' doesn't try to free them. */
3929 regstart
= regend
= old_regstart
= old_regend
= best_regstart
3930 = best_regend
= reg_dummy
= NULL
;
3931 reg_info
= reg_info_dummy
= (register_info_type
*) NULL
;
3933 #endif /* MATCH_MAY_ALLOCATE */
3935 /* The starting position is bogus. */
3936 if (pos
< 0 || pos
> size1
+ size2
)
3942 /* Initialize subexpression text positions to -1 to mark ones that no
3943 start_memory/stop_memory has been seen for. Also initialize the
3944 register information struct. */
3945 for (mcnt
= 1; (unsigned) mcnt
< num_regs
; mcnt
++)
3947 regstart
[mcnt
] = regend
[mcnt
]
3948 = old_regstart
[mcnt
] = old_regend
[mcnt
] = REG_UNSET_VALUE
;
3950 REG_MATCH_NULL_STRING_P (reg_info
[mcnt
]) = MATCH_NULL_UNSET_VALUE
;
3951 IS_ACTIVE (reg_info
[mcnt
]) = 0;
3952 MATCHED_SOMETHING (reg_info
[mcnt
]) = 0;
3953 EVER_MATCHED_SOMETHING (reg_info
[mcnt
]) = 0;
3956 /* We move `string1' into `string2' if the latter's empty -- but not if
3957 `string1' is null. */
3958 if (size2
== 0 && string1
!= NULL
)
3965 end1
= string1
+ size1
;
3966 end2
= string2
+ size2
;
3968 /* Compute where to stop matching, within the two strings. */
3971 end_match_1
= string1
+ stop
;
3972 end_match_2
= string2
;
3977 end_match_2
= string2
+ stop
- size1
;
3980 /* `p' scans through the pattern as `d' scans through the data.
3981 `dend' is the end of the input string that `d' points within. `d'
3982 is advanced into the following input string whenever necessary, but
3983 this happens before fetching; therefore, at the beginning of the
3984 loop, `d' can be pointing at the end of a string, but it cannot
3986 if (size1
> 0 && pos
<= size1
)
3993 d
= string2
+ pos
- size1
;
3997 DEBUG_PRINT1 ("The compiled pattern is:\n");
3998 DEBUG_PRINT_COMPILED_PATTERN (bufp
, p
, pend
);
3999 DEBUG_PRINT1 ("The string to match is: `");
4000 DEBUG_PRINT_DOUBLE_STRING (d
, string1
, size1
, string2
, size2
);
4001 DEBUG_PRINT1 ("'\n");
4003 /* This loops over pattern commands. It exits by returning from the
4004 function if the match is complete, or it drops through if the match
4005 fails at this starting point in the input data. */
4009 DEBUG_PRINT2 ("\n%p: ", p
);
4011 DEBUG_PRINT2 ("\n0x%x: ", p
);
4015 { /* End of pattern means we might have succeeded. */
4016 DEBUG_PRINT1 ("end of pattern ... ");
4018 /* If we haven't matched the entire string, and we want the
4019 longest match, try backtracking. */
4020 if (d
!= end_match_2
)
4022 /* 1 if this match ends in the same string (string1 or string2)
4023 as the best previous match. */
4024 boolean same_str_p
= (FIRST_STRING_P (match_end
)
4025 == MATCHING_IN_FIRST_STRING
);
4026 /* 1 if this match is the best seen so far. */
4027 boolean best_match_p
;
4029 /* AIX compiler got confused when this was combined
4030 with the previous declaration. */
4032 best_match_p
= d
> match_end
;
4034 best_match_p
= !MATCHING_IN_FIRST_STRING
;
4036 DEBUG_PRINT1 ("backtracking.\n");
4038 if (!FAIL_STACK_EMPTY ())
4039 { /* More failure points to try. */
4041 /* If exceeds best match so far, save it. */
4042 if (!best_regs_set
|| best_match_p
)
4044 best_regs_set
= true;
4047 DEBUG_PRINT1 ("\nSAVING match as best so far.\n");
4049 for (mcnt
= 1; (unsigned) mcnt
< num_regs
; mcnt
++)
4051 best_regstart
[mcnt
] = regstart
[mcnt
];
4052 best_regend
[mcnt
] = regend
[mcnt
];
4058 /* If no failure points, don't restore garbage. And if
4059 last match is real best match, don't restore second
4061 else if (best_regs_set
&& !best_match_p
)
4064 /* Restore best match. It may happen that `dend ==
4065 end_match_1' while the restored d is in string2.
4066 For example, the pattern `x.*y.*z' against the
4067 strings `x-' and `y-z-', if the two strings are
4068 not consecutive in memory. */
4069 DEBUG_PRINT1 ("Restoring best registers.\n");
4072 dend
= ((d
>= string1
&& d
<= end1
)
4073 ? end_match_1
: end_match_2
);
4075 for (mcnt
= 1; (unsigned) mcnt
< num_regs
; mcnt
++)
4077 regstart
[mcnt
] = best_regstart
[mcnt
];
4078 regend
[mcnt
] = best_regend
[mcnt
];
4081 } /* d != end_match_2 */
4084 DEBUG_PRINT1 ("Accepting match.\n");
4086 /* If caller wants register contents data back, do it. */
4087 if (regs
&& !bufp
->no_sub
)
4089 /* Have the register data arrays been allocated? */
4090 if (bufp
->regs_allocated
== REGS_UNALLOCATED
)
4091 { /* No. So allocate them with malloc. We need one
4092 extra element beyond `num_regs' for the `-1' marker
4094 regs
->num_regs
= MAX (RE_NREGS
, num_regs
+ 1);
4095 regs
->start
= TALLOC (regs
->num_regs
, regoff_t
);
4096 regs
->end
= TALLOC (regs
->num_regs
, regoff_t
);
4097 if (regs
->start
== NULL
|| regs
->end
== NULL
)
4102 bufp
->regs_allocated
= REGS_REALLOCATE
;
4104 else if (bufp
->regs_allocated
== REGS_REALLOCATE
)
4105 { /* Yes. If we need more elements than were already
4106 allocated, reallocate them. If we need fewer, just
4108 if (regs
->num_regs
< num_regs
+ 1)
4110 regs
->num_regs
= num_regs
+ 1;
4111 RETALLOC (regs
->start
, regs
->num_regs
, regoff_t
);
4112 RETALLOC (regs
->end
, regs
->num_regs
, regoff_t
);
4113 if (regs
->start
== NULL
|| regs
->end
== NULL
)
4122 /* These braces fend off a "empty body in an else-statement"
4123 warning under GCC when assert expands to nothing. */
4124 assert (bufp
->regs_allocated
== REGS_FIXED
);
4127 /* Convert the pointer data in `regstart' and `regend' to
4128 indices. Register zero has to be set differently,
4129 since we haven't kept track of any info for it. */
4130 if (regs
->num_regs
> 0)
4132 regs
->start
[0] = pos
;
4133 regs
->end
[0] = (MATCHING_IN_FIRST_STRING
4134 ? ((regoff_t
) (d
- string1
))
4135 : ((regoff_t
) (d
- string2
+ size1
)));
4138 /* Go through the first `min (num_regs, regs->num_regs)'
4139 registers, since that is all we initialized. */
4140 for (mcnt
= 1; (unsigned) mcnt
< MIN (num_regs
, regs
->num_regs
);
4143 if (REG_UNSET (regstart
[mcnt
]) || REG_UNSET (regend
[mcnt
]))
4144 regs
->start
[mcnt
] = regs
->end
[mcnt
] = -1;
4148 = (regoff_t
) POINTER_TO_OFFSET (regstart
[mcnt
]);
4150 = (regoff_t
) POINTER_TO_OFFSET (regend
[mcnt
]);
4154 /* If the regs structure we return has more elements than
4155 were in the pattern, set the extra elements to -1. If
4156 we (re)allocated the registers, this is the case,
4157 because we always allocate enough to have at least one
4159 for (mcnt
= num_regs
; (unsigned) mcnt
< regs
->num_regs
; mcnt
++)
4160 regs
->start
[mcnt
] = regs
->end
[mcnt
] = -1;
4161 } /* regs && !bufp->no_sub */
4163 DEBUG_PRINT4 ("%u failure points pushed, %u popped (%u remain).\n",
4164 nfailure_points_pushed
, nfailure_points_popped
,
4165 nfailure_points_pushed
- nfailure_points_popped
);
4166 DEBUG_PRINT2 ("%u registers pushed.\n", num_regs_pushed
);
4168 mcnt
= d
- pos
- (MATCHING_IN_FIRST_STRING
4172 DEBUG_PRINT2 ("Returning %d from re_match_2.\n", mcnt
);
4178 /* Otherwise match next pattern command. */
4179 switch (SWITCH_ENUM_CAST ((re_opcode_t
) *p
++))
4181 /* Ignore these. Used to ignore the n of succeed_n's which
4182 currently have n == 0. */
4184 DEBUG_PRINT1 ("EXECUTING no_op.\n");
4188 DEBUG_PRINT1 ("EXECUTING succeed.\n");
4191 /* Match the next n pattern characters exactly. The following
4192 byte in the pattern defines n, and the n bytes after that
4193 are the characters to match. */
4196 DEBUG_PRINT2 ("EXECUTING exactn %d.\n", mcnt
);
4198 /* This is written out as an if-else so we don't waste time
4199 testing `translate' inside the loop. */
4205 if ((unsigned char) translate
[(unsigned char) *d
++]
4206 != (unsigned char) *p
++)
4216 if (*d
++ != (char) *p
++) goto fail
;
4220 SET_REGS_MATCHED ();
4224 /* Match any character except possibly a newline or a null. */
4226 DEBUG_PRINT1 ("EXECUTING anychar.\n");
4230 if ((!(bufp
->syntax
& RE_DOT_NEWLINE
) && TRANSLATE (*d
) == '\n')
4231 || (bufp
->syntax
& RE_DOT_NOT_NULL
&& TRANSLATE (*d
) == '\000'))
4234 SET_REGS_MATCHED ();
4235 DEBUG_PRINT2 (" Matched `%d'.\n", *d
);
4243 register unsigned char c
;
4244 boolean
not = (re_opcode_t
) *(p
- 1) == charset_not
;
4246 DEBUG_PRINT2 ("EXECUTING charset%s.\n", not ? "_not" : "");
4249 c
= TRANSLATE (*d
); /* The character to match. */
4251 /* Cast to `unsigned' instead of `unsigned char' in case the
4252 bit list is a full 32 bytes long. */
4253 if (c
< (unsigned) (*p
* BYTEWIDTH
)
4254 && p
[1 + c
/ BYTEWIDTH
] & (1 << (c
% BYTEWIDTH
)))
4259 if (!not) goto fail
;
4261 SET_REGS_MATCHED ();
4267 /* The beginning of a group is represented by start_memory.
4268 The arguments are the register number in the next byte, and the
4269 number of groups inner to this one in the next. The text
4270 matched within the group is recorded (in the internal
4271 registers data structure) under the register number. */
4273 DEBUG_PRINT3 ("EXECUTING start_memory %d (%d):\n", *p
, p
[1]);
4275 /* Find out if this group can match the empty string. */
4276 p1
= p
; /* To send to group_match_null_string_p. */
4278 if (REG_MATCH_NULL_STRING_P (reg_info
[*p
]) == MATCH_NULL_UNSET_VALUE
)
4279 REG_MATCH_NULL_STRING_P (reg_info
[*p
])
4280 = group_match_null_string_p (&p1
, pend
, reg_info
);
4282 /* Save the position in the string where we were the last time
4283 we were at this open-group operator in case the group is
4284 operated upon by a repetition operator, e.g., with `(a*)*b'
4285 against `ab'; then we want to ignore where we are now in
4286 the string in case this attempt to match fails. */
4287 old_regstart
[*p
] = REG_MATCH_NULL_STRING_P (reg_info
[*p
])
4288 ? REG_UNSET (regstart
[*p
]) ? d
: regstart
[*p
]
4290 DEBUG_PRINT2 (" old_regstart: %d\n",
4291 POINTER_TO_OFFSET (old_regstart
[*p
]));
4294 DEBUG_PRINT2 (" regstart: %d\n", POINTER_TO_OFFSET (regstart
[*p
]));
4296 IS_ACTIVE (reg_info
[*p
]) = 1;
4297 MATCHED_SOMETHING (reg_info
[*p
]) = 0;
4299 /* Clear this whenever we change the register activity status. */
4300 set_regs_matched_done
= 0;
4302 /* This is the new highest active register. */
4303 highest_active_reg
= *p
;
4305 /* If nothing was active before, this is the new lowest active
4307 if (lowest_active_reg
== NO_LOWEST_ACTIVE_REG
)
4308 lowest_active_reg
= *p
;
4310 /* Move past the register number and inner group count. */
4312 just_past_start_mem
= p
;
4317 /* The stop_memory opcode represents the end of a group. Its
4318 arguments are the same as start_memory's: the register
4319 number, and the number of inner groups. */
4321 DEBUG_PRINT3 ("EXECUTING stop_memory %d (%d):\n", *p
, p
[1]);
4323 /* We need to save the string position the last time we were at
4324 this close-group operator in case the group is operated
4325 upon by a repetition operator, e.g., with `((a*)*(b*)*)*'
4326 against `aba'; then we want to ignore where we are now in
4327 the string in case this attempt to match fails. */
4328 old_regend
[*p
] = REG_MATCH_NULL_STRING_P (reg_info
[*p
])
4329 ? REG_UNSET (regend
[*p
]) ? d
: regend
[*p
]
4331 DEBUG_PRINT2 (" old_regend: %d\n",
4332 POINTER_TO_OFFSET (old_regend
[*p
]));
4335 DEBUG_PRINT2 (" regend: %d\n", POINTER_TO_OFFSET (regend
[*p
]));
4337 /* This register isn't active anymore. */
4338 IS_ACTIVE (reg_info
[*p
]) = 0;
4340 /* Clear this whenever we change the register activity status. */
4341 set_regs_matched_done
= 0;
4343 /* If this was the only register active, nothing is active
4345 if (lowest_active_reg
== highest_active_reg
)
4347 lowest_active_reg
= NO_LOWEST_ACTIVE_REG
;
4348 highest_active_reg
= NO_HIGHEST_ACTIVE_REG
;
4351 { /* We must scan for the new highest active register, since
4352 it isn't necessarily one less than now: consider
4353 (a(b)c(d(e)f)g). When group 3 ends, after the f), the
4354 new highest active register is 1. */
4355 unsigned char r
= *p
- 1;
4356 while (r
> 0 && !IS_ACTIVE (reg_info
[r
]))
4359 /* If we end up at register zero, that means that we saved
4360 the registers as the result of an `on_failure_jump', not
4361 a `start_memory', and we jumped to past the innermost
4362 `stop_memory'. For example, in ((.)*) we save
4363 registers 1 and 2 as a result of the *, but when we pop
4364 back to the second ), we are at the stop_memory 1.
4365 Thus, nothing is active. */
4368 lowest_active_reg
= NO_LOWEST_ACTIVE_REG
;
4369 highest_active_reg
= NO_HIGHEST_ACTIVE_REG
;
4372 highest_active_reg
= r
;
4375 /* If just failed to match something this time around with a
4376 group that's operated on by a repetition operator, try to
4377 force exit from the ``loop'', and restore the register
4378 information for this group that we had before trying this
4380 if ((!MATCHED_SOMETHING (reg_info
[*p
])
4381 || just_past_start_mem
== p
- 1)
4384 boolean is_a_jump_n
= false;
4388 switch ((re_opcode_t
) *p1
++)
4392 case pop_failure_jump
:
4393 case maybe_pop_jump
:
4395 case dummy_failure_jump
:
4396 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
4406 /* If the next operation is a jump backwards in the pattern
4407 to an on_failure_jump right before the start_memory
4408 corresponding to this stop_memory, exit from the loop
4409 by forcing a failure after pushing on the stack the
4410 on_failure_jump's jump in the pattern, and d. */
4411 if (mcnt
< 0 && (re_opcode_t
) *p1
== on_failure_jump
4412 && (re_opcode_t
) p1
[3] == start_memory
&& p1
[4] == *p
)
4414 /* If this group ever matched anything, then restore
4415 what its registers were before trying this last
4416 failed match, e.g., with `(a*)*b' against `ab' for
4417 regstart[1], and, e.g., with `((a*)*(b*)*)*'
4418 against `aba' for regend[3].
4420 Also restore the registers for inner groups for,
4421 e.g., `((a*)(b*))*' against `aba' (register 3 would
4422 otherwise get trashed). */
4424 if (EVER_MATCHED_SOMETHING (reg_info
[*p
]))
4428 EVER_MATCHED_SOMETHING (reg_info
[*p
]) = 0;
4430 /* Restore this and inner groups' (if any) registers. */
4431 for (r
= *p
; r
< (unsigned) *p
+ (unsigned) *(p
+ 1);
4434 regstart
[r
] = old_regstart
[r
];
4436 /* xx why this test? */
4437 if (old_regend
[r
] >= regstart
[r
])
4438 regend
[r
] = old_regend
[r
];
4442 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
4443 PUSH_FAILURE_POINT (p1
+ mcnt
, d
, -2);
4449 /* Move past the register number and the inner group count. */
4454 /* \<digit> has been turned into a `duplicate' command which is
4455 followed by the numeric value of <digit> as the register number. */
4458 register const char *d2
, *dend2
;
4459 int regno
= *p
++; /* Get which register to match against. */
4460 DEBUG_PRINT2 ("EXECUTING duplicate %d.\n", regno
);
4462 /* Can't back reference a group which we've never matched. */
4463 if (REG_UNSET (regstart
[regno
]) || REG_UNSET (regend
[regno
]))
4466 /* Where in input to try to start matching. */
4467 d2
= regstart
[regno
];
4469 /* Where to stop matching; if both the place to start and
4470 the place to stop matching are in the same string, then
4471 set to the place to stop, otherwise, for now have to use
4472 the end of the first string. */
4474 dend2
= ((FIRST_STRING_P (regstart
[regno
])
4475 == FIRST_STRING_P (regend
[regno
]))
4476 ? regend
[regno
] : end_match_1
);
4479 /* If necessary, advance to next segment in register
4483 if (dend2
== end_match_2
) break;
4484 if (dend2
== regend
[regno
]) break;
4486 /* End of string1 => advance to string2. */
4488 dend2
= regend
[regno
];
4490 /* At end of register contents => success */
4491 if (d2
== dend2
) break;
4493 /* If necessary, advance to next segment in data. */
4496 /* How many characters left in this segment to match. */
4499 /* Want how many consecutive characters we can match in
4500 one shot, so, if necessary, adjust the count. */
4501 if (mcnt
> dend2
- d2
)
4504 /* Compare that many; failure if mismatch, else move
4507 ? bcmp_translate (d
, d2
, mcnt
, translate
)
4508 : memcmp (d
, d2
, mcnt
))
4510 d
+= mcnt
, d2
+= mcnt
;
4512 /* Do this because we've match some characters. */
4513 SET_REGS_MATCHED ();
4519 /* begline matches the empty string at the beginning of the string
4520 (unless `not_bol' is set in `bufp'), and, if
4521 `newline_anchor' is set, after newlines. */
4523 DEBUG_PRINT1 ("EXECUTING begline.\n");
4525 if (AT_STRINGS_BEG (d
))
4527 if (!bufp
->not_bol
) break;
4529 else if (d
[-1] == '\n' && bufp
->newline_anchor
)
4533 /* In all other cases, we fail. */
4537 /* endline is the dual of begline. */
4539 DEBUG_PRINT1 ("EXECUTING endline.\n");
4541 if (AT_STRINGS_END (d
))
4543 if (!bufp
->not_eol
) break;
4546 /* We have to ``prefetch'' the next character. */
4547 else if ((d
== end1
? *string2
: *d
) == '\n'
4548 && bufp
->newline_anchor
)
4555 /* Match at the very beginning of the data. */
4557 DEBUG_PRINT1 ("EXECUTING begbuf.\n");
4558 if (AT_STRINGS_BEG (d
))
4563 /* Match at the very end of the data. */
4565 DEBUG_PRINT1 ("EXECUTING endbuf.\n");
4566 if (AT_STRINGS_END (d
))
4571 /* on_failure_keep_string_jump is used to optimize `.*\n'. It
4572 pushes NULL as the value for the string on the stack. Then
4573 `pop_failure_point' will keep the current value for the
4574 string, instead of restoring it. To see why, consider
4575 matching `foo\nbar' against `.*\n'. The .* matches the foo;
4576 then the . fails against the \n. But the next thing we want
4577 to do is match the \n against the \n; if we restored the
4578 string value, we would be back at the foo.
4580 Because this is used only in specific cases, we don't need to
4581 check all the things that `on_failure_jump' does, to make
4582 sure the right things get saved on the stack. Hence we don't
4583 share its code. The only reason to push anything on the
4584 stack at all is that otherwise we would have to change
4585 `anychar's code to do something besides goto fail in this
4586 case; that seems worse than this. */
4587 case on_failure_keep_string_jump
:
4588 DEBUG_PRINT1 ("EXECUTING on_failure_keep_string_jump");
4590 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
4592 DEBUG_PRINT3 (" %d (to %p):\n", mcnt
, p
+ mcnt
);
4594 DEBUG_PRINT3 (" %d (to 0x%x):\n", mcnt
, p
+ mcnt
);
4597 PUSH_FAILURE_POINT (p
+ mcnt
, NULL
, -2);
4601 /* Uses of on_failure_jump:
4603 Each alternative starts with an on_failure_jump that points
4604 to the beginning of the next alternative. Each alternative
4605 except the last ends with a jump that in effect jumps past
4606 the rest of the alternatives. (They really jump to the
4607 ending jump of the following alternative, because tensioning
4608 these jumps is a hassle.)
4610 Repeats start with an on_failure_jump that points past both
4611 the repetition text and either the following jump or
4612 pop_failure_jump back to this on_failure_jump. */
4613 case on_failure_jump
:
4615 DEBUG_PRINT1 ("EXECUTING on_failure_jump");
4617 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
4619 DEBUG_PRINT3 (" %d (to %p)", mcnt
, p
+ mcnt
);
4621 DEBUG_PRINT3 (" %d (to 0x%x)", mcnt
, p
+ mcnt
);
4624 /* If this on_failure_jump comes right before a group (i.e.,
4625 the original * applied to a group), save the information
4626 for that group and all inner ones, so that if we fail back
4627 to this point, the group's information will be correct.
4628 For example, in \(a*\)*\1, we need the preceding group,
4629 and in \(zz\(a*\)b*\)\2, we need the inner group. */
4631 /* We can't use `p' to check ahead because we push
4632 a failure point to `p + mcnt' after we do this. */
4635 /* We need to skip no_op's before we look for the
4636 start_memory in case this on_failure_jump is happening as
4637 the result of a completed succeed_n, as in \(a\)\{1,3\}b\1
4639 while (p1
< pend
&& (re_opcode_t
) *p1
== no_op
)
4642 if (p1
< pend
&& (re_opcode_t
) *p1
== start_memory
)
4644 /* We have a new highest active register now. This will
4645 get reset at the start_memory we are about to get to,
4646 but we will have saved all the registers relevant to
4647 this repetition op, as described above. */
4648 highest_active_reg
= *(p1
+ 1) + *(p1
+ 2);
4649 if (lowest_active_reg
== NO_LOWEST_ACTIVE_REG
)
4650 lowest_active_reg
= *(p1
+ 1);
4653 DEBUG_PRINT1 (":\n");
4654 PUSH_FAILURE_POINT (p
+ mcnt
, d
, -2);
4658 /* A smart repeat ends with `maybe_pop_jump'.
4659 We change it to either `pop_failure_jump' or `jump'. */
4660 case maybe_pop_jump
:
4661 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
4662 DEBUG_PRINT2 ("EXECUTING maybe_pop_jump %d.\n", mcnt
);
4664 register unsigned char *p2
= p
;
4666 /* Compare the beginning of the repeat with what in the
4667 pattern follows its end. If we can establish that there
4668 is nothing that they would both match, i.e., that we
4669 would have to backtrack because of (as in, e.g., `a*a')
4670 then we can change to pop_failure_jump, because we'll
4671 never have to backtrack.
4673 This is not true in the case of alternatives: in
4674 `(a|ab)*' we do need to backtrack to the `ab' alternative
4675 (e.g., if the string was `ab'). But instead of trying to
4676 detect that here, the alternative has put on a dummy
4677 failure point which is what we will end up popping. */
4679 /* Skip over open/close-group commands.
4680 If what follows this loop is a ...+ construct,
4681 look at what begins its body, since we will have to
4682 match at least one of that. */
4686 && ((re_opcode_t
) *p2
== stop_memory
4687 || (re_opcode_t
) *p2
== start_memory
))
4689 else if (p2
+ 6 < pend
4690 && (re_opcode_t
) *p2
== dummy_failure_jump
)
4697 /* p1[0] ... p1[2] are the `on_failure_jump' corresponding
4698 to the `maybe_finalize_jump' of this case. Examine what
4701 /* If we're at the end of the pattern, we can change. */
4704 /* Consider what happens when matching ":\(.*\)"
4705 against ":/". I don't really understand this code
4707 p
[-3] = (unsigned char) pop_failure_jump
;
4709 (" End of pattern: change to `pop_failure_jump'.\n");
4712 else if ((re_opcode_t
) *p2
== exactn
4713 || (bufp
->newline_anchor
&& (re_opcode_t
) *p2
== endline
))
4715 register unsigned char c
4716 = *p2
== (unsigned char) endline
? '\n' : p2
[2];
4718 if ((re_opcode_t
) p1
[3] == exactn
&& p1
[5] != c
)
4720 p
[-3] = (unsigned char) pop_failure_jump
;
4721 DEBUG_PRINT3 (" %c != %c => pop_failure_jump.\n",
4725 else if ((re_opcode_t
) p1
[3] == charset
4726 || (re_opcode_t
) p1
[3] == charset_not
)
4728 int not = (re_opcode_t
) p1
[3] == charset_not
;
4730 if (c
< (unsigned char) (p1
[4] * BYTEWIDTH
)
4731 && p1
[5 + c
/ BYTEWIDTH
] & (1 << (c
% BYTEWIDTH
)))
4734 /* `not' is equal to 1 if c would match, which means
4735 that we can't change to pop_failure_jump. */
4738 p
[-3] = (unsigned char) pop_failure_jump
;
4739 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
4743 else if ((re_opcode_t
) *p2
== charset
)
4746 register unsigned char c
4747 = *p2
== (unsigned char) endline
? '\n' : p2
[2];
4751 if ((re_opcode_t
) p1
[3] == exactn
4752 && ! ((int) p2
[1] * BYTEWIDTH
> (int) p1
[5]
4753 && (p2
[2 + p1
[5] / BYTEWIDTH
]
4754 & (1 << (p1
[5] % BYTEWIDTH
)))))
4756 if ((re_opcode_t
) p1
[3] == exactn
4757 && ! ((int) p2
[1] * BYTEWIDTH
> (int) p1
[4]
4758 && (p2
[2 + p1
[4] / BYTEWIDTH
]
4759 & (1 << (p1
[4] % BYTEWIDTH
)))))
4762 p
[-3] = (unsigned char) pop_failure_jump
;
4763 DEBUG_PRINT3 (" %c != %c => pop_failure_jump.\n",
4767 else if ((re_opcode_t
) p1
[3] == charset_not
)
4770 /* We win if the charset_not inside the loop
4771 lists every character listed in the charset after. */
4772 for (idx
= 0; idx
< (int) p2
[1]; idx
++)
4773 if (! (p2
[2 + idx
] == 0
4774 || (idx
< (int) p1
[4]
4775 && ((p2
[2 + idx
] & ~ p1
[5 + idx
]) == 0))))
4780 p
[-3] = (unsigned char) pop_failure_jump
;
4781 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
4784 else if ((re_opcode_t
) p1
[3] == charset
)
4787 /* We win if the charset inside the loop
4788 has no overlap with the one after the loop. */
4790 idx
< (int) p2
[1] && idx
< (int) p1
[4];
4792 if ((p2
[2 + idx
] & p1
[5 + idx
]) != 0)
4795 if (idx
== p2
[1] || idx
== p1
[4])
4797 p
[-3] = (unsigned char) pop_failure_jump
;
4798 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
4803 p
-= 2; /* Point at relative address again. */
4804 if ((re_opcode_t
) p
[-1] != pop_failure_jump
)
4806 p
[-1] = (unsigned char) jump
;
4807 DEBUG_PRINT1 (" Match => jump.\n");
4808 goto unconditional_jump
;
4810 /* Note fall through. */
4813 /* The end of a simple repeat has a pop_failure_jump back to
4814 its matching on_failure_jump, where the latter will push a
4815 failure point. The pop_failure_jump takes off failure
4816 points put on by this pop_failure_jump's matching
4817 on_failure_jump; we got through the pattern to here from the
4818 matching on_failure_jump, so didn't fail. */
4819 case pop_failure_jump
:
4821 /* We need to pass separate storage for the lowest and
4822 highest registers, even though we don't care about the
4823 actual values. Otherwise, we will restore only one
4824 register from the stack, since lowest will == highest in
4825 `pop_failure_point'. */
4826 active_reg_t dummy_low_reg
, dummy_high_reg
;
4827 unsigned char *pdummy
;
4830 DEBUG_PRINT1 ("EXECUTING pop_failure_jump.\n");
4831 POP_FAILURE_POINT (sdummy
, pdummy
,
4832 dummy_low_reg
, dummy_high_reg
,
4833 reg_dummy
, reg_dummy
, reg_info_dummy
);
4835 /* Note fall through. */
4839 DEBUG_PRINT2 ("\n%p: ", p
);
4841 DEBUG_PRINT2 ("\n0x%x: ", p
);
4843 /* Note fall through. */
4845 /* Unconditionally jump (without popping any failure points). */
4847 EXTRACT_NUMBER_AND_INCR (mcnt
, p
); /* Get the amount to jump. */
4848 DEBUG_PRINT2 ("EXECUTING jump %d ", mcnt
);
4849 p
+= mcnt
; /* Do the jump. */
4851 DEBUG_PRINT2 ("(to %p).\n", p
);
4853 DEBUG_PRINT2 ("(to 0x%x).\n", p
);
4858 /* We need this opcode so we can detect where alternatives end
4859 in `group_match_null_string_p' et al. */
4861 DEBUG_PRINT1 ("EXECUTING jump_past_alt.\n");
4862 goto unconditional_jump
;
4865 /* Normally, the on_failure_jump pushes a failure point, which
4866 then gets popped at pop_failure_jump. We will end up at
4867 pop_failure_jump, also, and with a pattern of, say, `a+', we
4868 are skipping over the on_failure_jump, so we have to push
4869 something meaningless for pop_failure_jump to pop. */
4870 case dummy_failure_jump
:
4871 DEBUG_PRINT1 ("EXECUTING dummy_failure_jump.\n");
4872 /* It doesn't matter what we push for the string here. What
4873 the code at `fail' tests is the value for the pattern. */
4874 PUSH_FAILURE_POINT (NULL
, NULL
, -2);
4875 goto unconditional_jump
;
4878 /* At the end of an alternative, we need to push a dummy failure
4879 point in case we are followed by a `pop_failure_jump', because
4880 we don't want the failure point for the alternative to be
4881 popped. For example, matching `(a|ab)*' against `aab'
4882 requires that we match the `ab' alternative. */
4883 case push_dummy_failure
:
4884 DEBUG_PRINT1 ("EXECUTING push_dummy_failure.\n");
4885 /* See comments just above at `dummy_failure_jump' about the
4887 PUSH_FAILURE_POINT (NULL
, NULL
, -2);
4890 /* Have to succeed matching what follows at least n times.
4891 After that, handle like `on_failure_jump'. */
4893 EXTRACT_NUMBER (mcnt
, p
+ 2);
4894 DEBUG_PRINT2 ("EXECUTING succeed_n %d.\n", mcnt
);
4897 /* Originally, this is how many times we HAVE to succeed. */
4902 STORE_NUMBER_AND_INCR (p
, mcnt
);
4904 DEBUG_PRINT3 (" Setting %p to %d.\n", p
- 2, mcnt
);
4906 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p
- 2, mcnt
);
4912 DEBUG_PRINT2 (" Setting two bytes from %p to no_op.\n", p
+2);
4914 DEBUG_PRINT2 (" Setting two bytes from 0x%x to no_op.\n", p
+2);
4916 p
[2] = (unsigned char) no_op
;
4917 p
[3] = (unsigned char) no_op
;
4923 EXTRACT_NUMBER (mcnt
, p
+ 2);
4924 DEBUG_PRINT2 ("EXECUTING jump_n %d.\n", mcnt
);
4926 /* Originally, this is how many times we CAN jump. */
4930 STORE_NUMBER (p
+ 2, mcnt
);
4932 DEBUG_PRINT3 (" Setting %p to %d.\n", p
+ 2, mcnt
);
4934 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p
+ 2, mcnt
);
4936 goto unconditional_jump
;
4938 /* If don't have to jump any more, skip over the rest of command. */
4945 DEBUG_PRINT1 ("EXECUTING set_number_at.\n");
4947 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
4949 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
4951 DEBUG_PRINT3 (" Setting %p to %d.\n", p1
, mcnt
);
4953 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p1
, mcnt
);
4955 STORE_NUMBER (p1
, mcnt
);
4960 /* The DEC Alpha C compiler 3.x generates incorrect code for the
4961 test WORDCHAR_P (d - 1) != WORDCHAR_P (d) in the expansion of
4962 AT_WORD_BOUNDARY, so this code is disabled. Expanding the
4963 macro and introducing temporary variables works around the bug. */
4966 DEBUG_PRINT1 ("EXECUTING wordbound.\n");
4967 if (AT_WORD_BOUNDARY (d
))
4972 DEBUG_PRINT1 ("EXECUTING notwordbound.\n");
4973 if (AT_WORD_BOUNDARY (d
))
4979 boolean prevchar
, thischar
;
4981 DEBUG_PRINT1 ("EXECUTING wordbound.\n");
4982 if (AT_STRINGS_BEG (d
) || AT_STRINGS_END (d
))
4985 prevchar
= WORDCHAR_P (d
- 1);
4986 thischar
= WORDCHAR_P (d
);
4987 if (prevchar
!= thischar
)
4994 boolean prevchar
, thischar
;
4996 DEBUG_PRINT1 ("EXECUTING notwordbound.\n");
4997 if (AT_STRINGS_BEG (d
) || AT_STRINGS_END (d
))
5000 prevchar
= WORDCHAR_P (d
- 1);
5001 thischar
= WORDCHAR_P (d
);
5002 if (prevchar
!= thischar
)
5009 DEBUG_PRINT1 ("EXECUTING wordbeg.\n");
5010 if (WORDCHAR_P (d
) && (AT_STRINGS_BEG (d
) || !WORDCHAR_P (d
- 1)))
5015 DEBUG_PRINT1 ("EXECUTING wordend.\n");
5016 if (!AT_STRINGS_BEG (d
) && WORDCHAR_P (d
- 1)
5017 && (!WORDCHAR_P (d
) || AT_STRINGS_END (d
)))
5023 DEBUG_PRINT1 ("EXECUTING before_dot.\n");
5024 if (PTR_CHAR_POS ((unsigned char *) d
) >= point
)
5029 DEBUG_PRINT1 ("EXECUTING at_dot.\n");
5030 if (PTR_CHAR_POS ((unsigned char *) d
) != point
)
5035 DEBUG_PRINT1 ("EXECUTING after_dot.\n");
5036 if (PTR_CHAR_POS ((unsigned char *) d
) <= point
)
5041 DEBUG_PRINT2 ("EXECUTING syntaxspec %d.\n", mcnt
);
5046 DEBUG_PRINT1 ("EXECUTING Emacs wordchar.\n");
5050 /* Can't use *d++ here; SYNTAX may be an unsafe macro. */
5052 if (SYNTAX (d
[-1]) != (enum syntaxcode
) mcnt
)
5054 SET_REGS_MATCHED ();
5058 DEBUG_PRINT2 ("EXECUTING notsyntaxspec %d.\n", mcnt
);
5060 goto matchnotsyntax
;
5063 DEBUG_PRINT1 ("EXECUTING Emacs notwordchar.\n");
5067 /* Can't use *d++ here; SYNTAX may be an unsafe macro. */
5069 if (SYNTAX (d
[-1]) == (enum syntaxcode
) mcnt
)
5071 SET_REGS_MATCHED ();
5074 #else /* not emacs */
5076 DEBUG_PRINT1 ("EXECUTING non-Emacs wordchar.\n");
5078 if (!WORDCHAR_P (d
))
5080 SET_REGS_MATCHED ();
5085 DEBUG_PRINT1 ("EXECUTING non-Emacs notwordchar.\n");
5089 SET_REGS_MATCHED ();
5092 #endif /* not emacs */
5097 continue; /* Successfully executed one pattern command; keep going. */
5100 /* We goto here if a matching operation fails. */
5102 if (!FAIL_STACK_EMPTY ())
5103 { /* A restart point is known. Restore to that state. */
5104 DEBUG_PRINT1 ("\nFAIL:\n");
5105 POP_FAILURE_POINT (d
, p
,
5106 lowest_active_reg
, highest_active_reg
,
5107 regstart
, regend
, reg_info
);
5109 /* If this failure point is a dummy, try the next one. */
5113 /* If we failed to the end of the pattern, don't examine *p. */
5117 boolean is_a_jump_n
= false;
5119 /* If failed to a backwards jump that's part of a repetition
5120 loop, need to pop this failure point and use the next one. */
5121 switch ((re_opcode_t
) *p
)
5125 case maybe_pop_jump
:
5126 case pop_failure_jump
:
5129 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
5132 if ((is_a_jump_n
&& (re_opcode_t
) *p1
== succeed_n
)
5134 && (re_opcode_t
) *p1
== on_failure_jump
))
5142 if (d
>= string1
&& d
<= end1
)
5146 break; /* Matching at this starting point really fails. */
5150 goto restore_best_regs
;
5154 return -1; /* Failure to match. */
5157 /* Subroutine definitions for re_match_2. */
5160 /* We are passed P pointing to a register number after a start_memory.
5162 Return true if the pattern up to the corresponding stop_memory can
5163 match the empty string, and false otherwise.
5165 If we find the matching stop_memory, sets P to point to one past its number.
5166 Otherwise, sets P to an undefined byte less than or equal to END.
5168 We don't handle duplicates properly (yet). */
5171 group_match_null_string_p (p
, end
, reg_info
)
5172 unsigned char **p
, *end
;
5173 register_info_type
*reg_info
;
5176 /* Point to after the args to the start_memory. */
5177 unsigned char *p1
= *p
+ 2;
5181 /* Skip over opcodes that can match nothing, and return true or
5182 false, as appropriate, when we get to one that can't, or to the
5183 matching stop_memory. */
5185 switch ((re_opcode_t
) *p1
)
5187 /* Could be either a loop or a series of alternatives. */
5188 case on_failure_jump
:
5190 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
5192 /* If the next operation is not a jump backwards in the
5197 /* Go through the on_failure_jumps of the alternatives,
5198 seeing if any of the alternatives cannot match nothing.
5199 The last alternative starts with only a jump,
5200 whereas the rest start with on_failure_jump and end
5201 with a jump, e.g., here is the pattern for `a|b|c':
5203 /on_failure_jump/0/6/exactn/1/a/jump_past_alt/0/6
5204 /on_failure_jump/0/6/exactn/1/b/jump_past_alt/0/3
5207 So, we have to first go through the first (n-1)
5208 alternatives and then deal with the last one separately. */
5211 /* Deal with the first (n-1) alternatives, which start
5212 with an on_failure_jump (see above) that jumps to right
5213 past a jump_past_alt. */
5215 while ((re_opcode_t
) p1
[mcnt
-3] == jump_past_alt
)
5217 /* `mcnt' holds how many bytes long the alternative
5218 is, including the ending `jump_past_alt' and
5221 if (!alt_match_null_string_p (p1
, p1
+ mcnt
- 3,
5225 /* Move to right after this alternative, including the
5229 /* Break if it's the beginning of an n-th alternative
5230 that doesn't begin with an on_failure_jump. */
5231 if ((re_opcode_t
) *p1
!= on_failure_jump
)
5234 /* Still have to check that it's not an n-th
5235 alternative that starts with an on_failure_jump. */
5237 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
5238 if ((re_opcode_t
) p1
[mcnt
-3] != jump_past_alt
)
5240 /* Get to the beginning of the n-th alternative. */
5246 /* Deal with the last alternative: go back and get number
5247 of the `jump_past_alt' just before it. `mcnt' contains
5248 the length of the alternative. */
5249 EXTRACT_NUMBER (mcnt
, p1
- 2);
5251 if (!alt_match_null_string_p (p1
, p1
+ mcnt
, reg_info
))
5254 p1
+= mcnt
; /* Get past the n-th alternative. */
5260 assert (p1
[1] == **p
);
5266 if (!common_op_match_null_string_p (&p1
, end
, reg_info
))
5269 } /* while p1 < end */
5272 } /* group_match_null_string_p */
5275 /* Similar to group_match_null_string_p, but doesn't deal with alternatives:
5276 It expects P to be the first byte of a single alternative and END one
5277 byte past the last. The alternative can contain groups. */
5280 alt_match_null_string_p (p
, end
, reg_info
)
5281 unsigned char *p
, *end
;
5282 register_info_type
*reg_info
;
5285 unsigned char *p1
= p
;
5289 /* Skip over opcodes that can match nothing, and break when we get
5290 to one that can't. */
5292 switch ((re_opcode_t
) *p1
)
5295 case on_failure_jump
:
5297 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
5302 if (!common_op_match_null_string_p (&p1
, end
, reg_info
))
5305 } /* while p1 < end */
5308 } /* alt_match_null_string_p */
5311 /* Deals with the ops common to group_match_null_string_p and
5312 alt_match_null_string_p.
5314 Sets P to one after the op and its arguments, if any. */
5317 common_op_match_null_string_p (p
, end
, reg_info
)
5318 unsigned char **p
, *end
;
5319 register_info_type
*reg_info
;
5324 unsigned char *p1
= *p
;
5326 switch ((re_opcode_t
) *p1
++)
5346 assert (reg_no
> 0 && reg_no
<= MAX_REGNUM
);
5347 ret
= group_match_null_string_p (&p1
, end
, reg_info
);
5349 /* Have to set this here in case we're checking a group which
5350 contains a group and a back reference to it. */
5352 if (REG_MATCH_NULL_STRING_P (reg_info
[reg_no
]) == MATCH_NULL_UNSET_VALUE
)
5353 REG_MATCH_NULL_STRING_P (reg_info
[reg_no
]) = ret
;
5359 /* If this is an optimized succeed_n for zero times, make the jump. */
5361 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
5369 /* Get to the number of times to succeed. */
5371 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
5376 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
5384 if (!REG_MATCH_NULL_STRING_P (reg_info
[*p1
]))
5392 /* All other opcodes mean we cannot match the empty string. */
5398 } /* common_op_match_null_string_p */
5401 /* Return zero if TRANSLATE[S1] and TRANSLATE[S2] are identical for LEN
5402 bytes; nonzero otherwise. */
5405 bcmp_translate (s1
, s2
, len
, translate
)
5406 const char *s1
, *s2
;
5408 RE_TRANSLATE_TYPE translate
;
5410 register const unsigned char *p1
= (const unsigned char *) s1
;
5411 register const unsigned char *p2
= (const unsigned char *) s2
;
5414 if (translate
[*p1
++] != translate
[*p2
++]) return 1;
5420 /* Entry points for GNU code. */
5422 /* re_compile_pattern is the GNU regular expression compiler: it
5423 compiles PATTERN (of length SIZE) and puts the result in BUFP.
5424 Returns 0 if the pattern was valid, otherwise an error string.
5426 Assumes the `allocated' (and perhaps `buffer') and `translate' fields
5427 are set in BUFP on entry.
5429 We call regex_compile to do the actual compilation. */
5432 re_compile_pattern (pattern
, length
, bufp
)
5433 const char *pattern
;
5435 struct re_pattern_buffer
*bufp
;
5439 /* GNU code is written to assume at least RE_NREGS registers will be set
5440 (and at least one extra will be -1). */
5441 bufp
->regs_allocated
= REGS_UNALLOCATED
;
5443 /* And GNU code determines whether or not to get register information
5444 by passing null for the REGS argument to re_match, etc., not by
5448 /* Match anchors at newline. */
5449 bufp
->newline_anchor
= 1;
5451 ret
= regex_compile (pattern
, length
, re_syntax_options
, bufp
);
5455 return gettext (re_error_msgid
[(int) ret
]);
5458 weak_alias (__re_compile_pattern
, re_compile_pattern
)
5461 /* Entry points compatible with 4.2 BSD regex library. We don't define
5462 them unless specifically requested. */
5464 #if defined _REGEX_RE_COMP || defined _LIBC
5466 /* BSD has one and only one pattern buffer. */
5467 static struct re_pattern_buffer re_comp_buf
;
5471 /* Make these definitions weak in libc, so POSIX programs can redefine
5472 these names if they don't use our functions, and still use
5473 regcomp/regexec below without link errors. */
5483 if (!re_comp_buf
.buffer
)
5484 return gettext ("No previous regular expression");
5488 if (!re_comp_buf
.buffer
)
5490 re_comp_buf
.buffer
= (unsigned char *) malloc (200);
5491 if (re_comp_buf
.buffer
== NULL
)
5492 return gettext (re_error_msgid
[(int) REG_ESPACE
]);
5493 re_comp_buf
.allocated
= 200;
5495 re_comp_buf
.fastmap
= (char *) malloc (1 << BYTEWIDTH
);
5496 if (re_comp_buf
.fastmap
== NULL
)
5497 return gettext (re_error_msgid
[(int) REG_ESPACE
]);
5500 /* Since `re_exec' always passes NULL for the `regs' argument, we
5501 don't need to initialize the pattern buffer fields which affect it. */
5503 /* Match anchors at newlines. */
5504 re_comp_buf
.newline_anchor
= 1;
5506 ret
= regex_compile (s
, strlen (s
), re_syntax_options
, &re_comp_buf
);
5511 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
5512 return (char *) gettext (re_error_msgid
[(int) ret
]);
5523 const int len
= strlen (s
);
5525 0 <= re_search (&re_comp_buf
, s
, len
, 0, len
, (struct re_registers
*) 0);
5528 #endif /* _REGEX_RE_COMP */
5530 /* POSIX.2 functions. Don't define these for Emacs. */
5534 /* regcomp takes a regular expression as a string and compiles it.
5536 PREG is a regex_t *. We do not expect any fields to be initialized,
5537 since POSIX says we shouldn't. Thus, we set
5539 `buffer' to the compiled pattern;
5540 `used' to the length of the compiled pattern;
5541 `syntax' to RE_SYNTAX_POSIX_EXTENDED if the
5542 REG_EXTENDED bit in CFLAGS is set; otherwise, to
5543 RE_SYNTAX_POSIX_BASIC;
5544 `newline_anchor' to REG_NEWLINE being set in CFLAGS;
5545 `fastmap' and `fastmap_accurate' to zero;
5546 `re_nsub' to the number of subexpressions in PATTERN.
5548 PATTERN is the address of the pattern string.
5550 CFLAGS is a series of bits which affect compilation.
5552 If REG_EXTENDED is set, we use POSIX extended syntax; otherwise, we
5553 use POSIX basic syntax.
5555 If REG_NEWLINE is set, then . and [^...] don't match newline.
5556 Also, regexec will try a match beginning after every newline.
5558 If REG_ICASE is set, then we considers upper- and lowercase
5559 versions of letters to be equivalent when matching.
5561 If REG_NOSUB is set, then when PREG is passed to regexec, that
5562 routine will report only success or failure, and nothing about the
5565 It returns 0 if it succeeds, nonzero if it doesn't. (See gnu-regex.h for
5566 the return codes and their meanings.) */
5569 regcomp (preg
, pattern
, cflags
)
5571 const char *pattern
;
5576 = (cflags
& REG_EXTENDED
) ?
5577 RE_SYNTAX_POSIX_EXTENDED
: RE_SYNTAX_POSIX_BASIC
;
5579 /* regex_compile will allocate the space for the compiled pattern. */
5581 preg
->allocated
= 0;
5584 /* Don't bother to use a fastmap when searching. This simplifies the
5585 REG_NEWLINE case: if we used a fastmap, we'd have to put all the
5586 characters after newlines into the fastmap. This way, we just try
5590 if (cflags
& REG_ICASE
)
5595 = (RE_TRANSLATE_TYPE
) malloc (CHAR_SET_SIZE
5596 * sizeof (*(RE_TRANSLATE_TYPE
)0));
5597 if (preg
->translate
== NULL
)
5598 return (int) REG_ESPACE
;
5600 /* Map uppercase characters to corresponding lowercase ones. */
5601 for (i
= 0; i
< CHAR_SET_SIZE
; i
++)
5602 preg
->translate
[i
] = ISUPPER (i
) ? tolower (i
) : i
;
5605 preg
->translate
= NULL
;
5607 /* If REG_NEWLINE is set, newlines are treated differently. */
5608 if (cflags
& REG_NEWLINE
)
5609 { /* REG_NEWLINE implies neither . nor [^...] match newline. */
5610 syntax
&= ~RE_DOT_NEWLINE
;
5611 syntax
|= RE_HAT_LISTS_NOT_NEWLINE
;
5612 /* It also changes the matching behavior. */
5613 preg
->newline_anchor
= 1;
5616 preg
->newline_anchor
= 0;
5618 preg
->no_sub
= !!(cflags
& REG_NOSUB
);
5620 /* POSIX says a null character in the pattern terminates it, so we
5621 can use strlen here in compiling the pattern. */
5622 ret
= regex_compile (pattern
, strlen (pattern
), syntax
, preg
);
5624 /* POSIX doesn't distinguish between an unmatched open-group and an
5625 unmatched close-group: both are REG_EPAREN. */
5626 if (ret
== REG_ERPAREN
) ret
= REG_EPAREN
;
5631 weak_alias (__regcomp
, regcomp
)
5635 /* regexec searches for a given pattern, specified by PREG, in the
5638 If NMATCH is zero or REG_NOSUB was set in the cflags argument to
5639 `regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at
5640 least NMATCH elements, and we set them to the offsets of the
5641 corresponding matched substrings.
5643 EFLAGS specifies `execution flags' which affect matching: if
5644 REG_NOTBOL is set, then ^ does not match at the beginning of the
5645 string; if REG_NOTEOL is set, then $ does not match at the end.
5647 We return 0 if we find a match and REG_NOMATCH if not. */
5650 regexec (preg
, string
, nmatch
, pmatch
, eflags
)
5651 const regex_t
*preg
;
5654 regmatch_t pmatch
[];
5658 struct re_registers regs
;
5659 regex_t private_preg
;
5660 int len
= strlen (string
);
5661 boolean want_reg_info
= !preg
->no_sub
&& nmatch
> 0;
5663 private_preg
= *preg
;
5665 private_preg
.not_bol
= !!(eflags
& REG_NOTBOL
);
5666 private_preg
.not_eol
= !!(eflags
& REG_NOTEOL
);
5668 /* The user has told us exactly how many registers to return
5669 information about, via `nmatch'. We have to pass that on to the
5670 matching routines. */
5671 private_preg
.regs_allocated
= REGS_FIXED
;
5675 regs
.num_regs
= nmatch
;
5676 regs
.start
= TALLOC (nmatch
, regoff_t
);
5677 regs
.end
= TALLOC (nmatch
, regoff_t
);
5678 if (regs
.start
== NULL
|| regs
.end
== NULL
)
5679 return (int) REG_NOMATCH
;
5682 /* Perform the searching operation. */
5683 ret
= re_search (&private_preg
, string
, len
,
5684 /* start: */ 0, /* range: */ len
,
5685 want_reg_info
? ®s
: (struct re_registers
*) 0);
5687 /* Copy the register information to the POSIX structure. */
5694 for (r
= 0; r
< nmatch
; r
++)
5696 pmatch
[r
].rm_so
= regs
.start
[r
];
5697 pmatch
[r
].rm_eo
= regs
.end
[r
];
5701 /* If we needed the temporary register info, free the space now. */
5706 /* We want zero return to mean success, unlike `re_search'. */
5707 return ret
>= 0 ? (int) REG_NOERROR
: (int) REG_NOMATCH
;
5710 weak_alias (__regexec
, regexec
)
5714 /* Returns a message corresponding to an error code, ERRCODE, returned
5715 from either regcomp or regexec. We don't use PREG here. */
5718 __regerror (errcode
, preg
, errbuf
, errbuf_size
)
5720 const regex_t
*preg
;
5728 || errcode
>= (int) (sizeof (re_error_msgid
)
5729 / sizeof (re_error_msgid
[0])))
5730 /* Only error codes returned by the rest of the code should be passed
5731 to this routine. If we are given anything else, or if other regex
5732 code generates an invalid error code, then the program has a bug.
5733 Dump core so we can fix it. */
5736 msg
= gettext (re_error_msgid
[errcode
]);
5738 msg_size
= strlen (msg
) + 1; /* Includes the null. */
5740 if (errbuf_size
!= 0)
5742 if (msg_size
> errbuf_size
)
5744 #if defined HAVE_MEMPCPY || defined _LIBC
5745 *((char *) __mempcpy (errbuf
, msg
, errbuf_size
- 1)) = '\0';
5747 memcpy (errbuf
, msg
, errbuf_size
- 1);
5748 errbuf
[errbuf_size
- 1] = 0;
5752 memcpy (errbuf
, msg
, msg_size
);
5758 weak_alias (__regerror
, regerror
)
5762 /* Free dynamically allocated space used by PREG. */
5768 if (preg
->buffer
!= NULL
)
5769 free (preg
->buffer
);
5770 preg
->buffer
= NULL
;
5772 preg
->allocated
= 0;
5775 if (preg
->fastmap
!= NULL
)
5776 free (preg
->fastmap
);
5777 preg
->fastmap
= NULL
;
5778 preg
->fastmap_accurate
= 0;
5780 if (preg
->translate
!= NULL
)
5781 free (preg
->translate
);
5782 preg
->translate
= NULL
;
5785 weak_alias (__regfree
, regfree
)
5788 #endif /* not emacs */