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, 1994, 1995, 1996, 1997, 1998
6 Free Software Foundation, Inc.
8 NOTE: The canonical source of this file is maintained with the
9 GNU C Library. Bugs can be reported to bug-glibc@prep.ai.mit.edu.
11 This program is free software; you can redistribute it and/or modify it
12 under the terms of the GNU General Public License as published by the
13 Free Software Foundation; either version 2, or (at your option) any
16 This program is distributed in the hope that it will be useful,
17 but WITHOUT ANY WARRANTY; without even the implied warranty of
18 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
19 GNU General Public License for more details.
21 You should have received a copy of the GNU General Public License
22 along with this program; if not, write to the Free Software Foundation,
23 Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. */
29 /* GCC LOCAL: we don't need NLS here. */
31 /* GCC LOCAL: to handle defining alloca. */
32 #include "libiberty.h"
34 /* Do not use a C alloca, we will leak memory and crash. */
39 /* AIX requires this to be the first thing in the file. */
40 #if defined _AIX && !defined REGEX_MALLOC
45 # if defined __GNUC__ || (defined __STDC__ && __STDC__)
46 # define PARAMS(args) args
48 # define PARAMS(args) ()
50 #endif /* Not PARAMS. */
52 #if defined STDC_HEADERS && !defined emacs
55 /* We need this for `gnu-regex.h', and perhaps for the Emacs include files. */
56 # include <sys/types.h>
59 /* For platform which support the ISO C amendement 1 functionality we
60 support user defined character classes. */
61 #if defined _LIBC || (defined HAVE_WCTYPE_H && defined HAVE_WCHAR_H)
62 /* Solaris 2.5 has a bug: <wchar.h> must be included before <wctype.h>. */
67 /* This is for other GNU distributions with internationalized messages. */
68 /* GCC LOCAL: ../intl will handle this for us */
72 # define gettext(msgid) (msgid)
76 /* This define is so xgettext can find the internationalizable
78 # define gettext_noop(String) String
81 /* If we are not linking with Emacs proper,
82 we can't use the relocating allocator
83 even if config.h says that we can. */
86 # if defined STDC_HEADERS || defined _LIBC
93 /* When used in Emacs's lib-src, we need to get bzero and bcopy somehow.
94 If nothing else has been done, use the method below. */
95 # ifdef INHIBIT_STRING_HEADER
96 # if !(defined HAVE_BZERO && defined HAVE_BCOPY)
97 # if !defined bzero && !defined bcopy
98 # undef INHIBIT_STRING_HEADER
103 /* This is the normal way of making sure we have a bcopy and a bzero.
104 This is used in most programs--a few other programs avoid this
105 by defining INHIBIT_STRING_HEADER. */
106 # ifndef INHIBIT_STRING_HEADER
107 # if defined HAVE_STRING_H || defined STDC_HEADERS || defined _LIBC
111 # define bzero(s, n) (memset (s, '\0', n), (s))
113 # define bzero(s, n) __bzero (s, n)
117 # include <strings.h>
119 # define memcmp(s1, s2, n) bcmp (s1, s2, n)
122 # define memcpy(d, s, n) (bcopy (s, d, n), (d))
127 /* Define the syntax stuff for \<, \>, etc. */
129 /* This must be nonzero for the wordchar and notwordchar pattern
130 commands in re_match_2. */
135 # ifdef SWITCH_ENUM_BUG
136 # define SWITCH_ENUM_CAST(x) ((int)(x))
138 # define SWITCH_ENUM_CAST(x) (x)
141 /* How many characters in the character set. */
142 # define CHAR_SET_SIZE 256
146 extern char *re_syntax_table
;
148 # else /* not SYNTAX_TABLE */
150 static char re_syntax_table
[CHAR_SET_SIZE
];
152 static void init_syntax_once
PARAMS ((void));
163 bzero (re_syntax_table
, sizeof re_syntax_table
);
165 for (c
= 'a'; c
<= 'z'; c
++)
166 re_syntax_table
[c
] = Sword
;
168 for (c
= 'A'; c
<= 'Z'; c
++)
169 re_syntax_table
[c
] = Sword
;
171 for (c
= '0'; c
<= '9'; c
++)
172 re_syntax_table
[c
] = Sword
;
174 re_syntax_table
['_'] = Sword
;
179 # endif /* not SYNTAX_TABLE */
181 # define SYNTAX(c) re_syntax_table[c]
183 /* Get the interface, including the syntax bits. */
184 /* GCC LOCAL: call it gnu-regex.h, not regex.h, to avoid name conflicts */
185 #include "gnu-regex.h"
187 /* ISALPHA etc. are used for the character classes. */
188 /* GCC LOCAL: use libiberty's safe-ctype.h, don't bother defining
189 wrapper macros ourselves. */
190 #include <safe-ctype.h>
193 # define NULL (void *)0
196 /* We remove any previous definition of `SIGN_EXTEND_CHAR',
197 since ours (we hope) works properly with all combinations of
198 machines, compilers, `char' and `unsigned char' argument types.
199 (Per Bothner suggested the basic approach.) */
200 #undef SIGN_EXTEND_CHAR
202 # define SIGN_EXTEND_CHAR(c) ((signed char) (c))
203 #else /* not __STDC__ */
204 /* As in Harbison and Steele. */
205 # define SIGN_EXTEND_CHAR(c) ((((unsigned char) (c)) ^ 128) - 128)
208 /* Should we use malloc or alloca? If REGEX_MALLOC is not defined, we
209 use `alloca' instead of `malloc'. This is because using malloc in
210 re_search* or re_match* could cause memory leaks when C-g is used in
211 Emacs; also, malloc is slower and causes storage fragmentation. On
212 the other hand, malloc is more portable, and easier to debug.
214 Because we sometimes use alloca, some routines have to be macros,
215 not functions -- `alloca'-allocated space disappears at the end of the
216 function it is called in. */
220 # define REGEX_ALLOCATE malloc
221 # define REGEX_REALLOCATE(source, osize, nsize) realloc (source, nsize)
222 # define REGEX_FREE free
224 #else /* not REGEX_MALLOC */
226 /* Emacs already defines alloca, sometimes. */
229 /* Make alloca work the best possible way. */
231 # define alloca __builtin_alloca
232 # else /* not __GNUC__ */
235 # endif /* HAVE_ALLOCA_H */
236 # endif /* not __GNUC__ */
238 # endif /* not alloca */
240 # define REGEX_ALLOCATE alloca
242 /* Assumes a `char *destination' variable. */
243 # define REGEX_REALLOCATE(source, osize, nsize) \
244 (destination = (char *) alloca (nsize), \
245 memcpy (destination, source, osize))
247 /* No need to do anything to free, after alloca. */
248 # define REGEX_FREE(arg) ((void)0) /* Do nothing! But inhibit gcc warning. */
250 #endif /* not REGEX_MALLOC */
252 /* Define how to allocate the failure stack. */
254 #if defined REL_ALLOC && defined REGEX_MALLOC
256 # define REGEX_ALLOCATE_STACK(size) \
257 r_alloc (&failure_stack_ptr, (size))
258 # define REGEX_REALLOCATE_STACK(source, osize, nsize) \
259 r_re_alloc (&failure_stack_ptr, (nsize))
260 # define REGEX_FREE_STACK(ptr) \
261 r_alloc_free (&failure_stack_ptr)
263 #else /* not using relocating allocator */
267 # define REGEX_ALLOCATE_STACK malloc
268 # define REGEX_REALLOCATE_STACK(source, osize, nsize) realloc (source, nsize)
269 # define REGEX_FREE_STACK free
271 # else /* not REGEX_MALLOC */
273 # define REGEX_ALLOCATE_STACK alloca
275 # define REGEX_REALLOCATE_STACK(source, osize, nsize) \
276 REGEX_REALLOCATE (source, osize, nsize)
277 /* No need to explicitly free anything. */
278 # define REGEX_FREE_STACK(arg)
280 # endif /* not REGEX_MALLOC */
281 #endif /* not using relocating allocator */
284 /* True if `size1' is non-NULL and PTR is pointing anywhere inside
285 `string1' or just past its end. This works if PTR is NULL, which is
287 #define FIRST_STRING_P(ptr) \
288 (size1 && string1 <= (ptr) && (ptr) <= string1 + size1)
290 /* (Re)Allocate N items of type T using malloc, or fail. */
291 #define TALLOC(n, t) ((t *) malloc ((n) * sizeof (t)))
292 #define RETALLOC(addr, n, t) ((addr) = (t *) realloc (addr, (n) * sizeof (t)))
293 #define RETALLOC_IF(addr, n, t) \
294 if (addr) RETALLOC((addr), (n), t); else (addr) = TALLOC ((n), t)
295 #define REGEX_TALLOC(n, t) ((t *) REGEX_ALLOCATE ((n) * sizeof (t)))
297 #define BYTEWIDTH 8 /* In bits. */
299 #define STREQ(s1, s2) ((strcmp (s1, s2) == 0))
303 #define MAX(a, b) ((a) > (b) ? (a) : (b))
304 #define MIN(a, b) ((a) < (b) ? (a) : (b))
306 typedef char boolean
;
310 static int re_match_2_internal
PARAMS ((struct re_pattern_buffer
*bufp
,
311 const char *string1
, int size1
,
312 const char *string2
, int size2
,
314 struct re_registers
*regs
,
317 /* These are the command codes that appear in compiled regular
318 expressions. Some opcodes are followed by argument bytes. A
319 command code can specify any interpretation whatsoever for its
320 arguments. Zero bytes may appear in the compiled regular expression. */
326 /* Succeed right away--no more backtracking. */
329 /* Followed by one byte giving n, then by n literal bytes. */
332 /* Matches any (more or less) character. */
335 /* Matches any one char belonging to specified set. First
336 following byte is number of bitmap bytes. Then come bytes
337 for a bitmap saying which chars are in. Bits in each byte
338 are ordered low-bit-first. A character is in the set if its
339 bit is 1. A character too large to have a bit in the map is
340 automatically not in the set. */
343 /* Same parameters as charset, but match any character that is
344 not one of those specified. */
347 /* Start remembering the text that is matched, for storing in a
348 register. Followed by one byte with the register number, in
349 the range 0 to one less than the pattern buffer's re_nsub
350 field. Then followed by one byte with the number of groups
351 inner to this one. (This last has to be part of the
352 start_memory only because we need it in the on_failure_jump
356 /* Stop remembering the text that is matched and store it in a
357 memory register. Followed by one byte with the register
358 number, in the range 0 to one less than `re_nsub' in the
359 pattern buffer, and one byte with the number of inner groups,
360 just like `start_memory'. (We need the number of inner
361 groups here because we don't have any easy way of finding the
362 corresponding start_memory when we're at a stop_memory.) */
365 /* Match a duplicate of something remembered. Followed by one
366 byte containing the register number. */
369 /* Fail unless at beginning of line. */
372 /* Fail unless at end of line. */
375 /* Succeeds if at beginning of buffer (if emacs) or at beginning
376 of string to be matched (if not). */
379 /* Analogously, for end of buffer/string. */
382 /* Followed by two byte relative address to which to jump. */
385 /* Same as jump, but marks the end of an alternative. */
388 /* Followed by two-byte relative address of place to resume at
389 in case of failure. */
392 /* Like on_failure_jump, but pushes a placeholder instead of the
393 current string position when executed. */
394 on_failure_keep_string_jump
,
396 /* Throw away latest failure point and then jump to following
397 two-byte relative address. */
400 /* Change to pop_failure_jump if know won't have to backtrack to
401 match; otherwise change to jump. This is used to jump
402 back to the beginning of a repeat. If what follows this jump
403 clearly won't match what the repeat does, such that we can be
404 sure that there is no use backtracking out of repetitions
405 already matched, then we change it to a pop_failure_jump.
406 Followed by two-byte address. */
409 /* Jump to following two-byte address, and push a dummy failure
410 point. This failure point will be thrown away if an attempt
411 is made to use it for a failure. A `+' construct makes this
412 before the first repeat. Also used as an intermediary kind
413 of jump when compiling an alternative. */
416 /* Push a dummy failure point and continue. Used at the end of
420 /* Followed by two-byte relative address and two-byte number n.
421 After matching N times, jump to the address upon failure. */
424 /* Followed by two-byte relative address, and two-byte number n.
425 Jump to the address N times, then fail. */
428 /* Set the following two-byte relative address to the
429 subsequent two-byte number. The address *includes* the two
433 wordchar
, /* Matches any word-constituent character. */
434 notwordchar
, /* Matches any char that is not a word-constituent. */
436 wordbeg
, /* Succeeds if at word beginning. */
437 wordend
, /* Succeeds if at word end. */
439 wordbound
, /* Succeeds if at a word boundary. */
440 notwordbound
/* Succeeds if not at a word boundary. */
443 ,before_dot
, /* Succeeds if before point. */
444 at_dot
, /* Succeeds if at point. */
445 after_dot
, /* Succeeds if after point. */
447 /* Matches any character whose syntax is specified. Followed by
448 a byte which contains a syntax code, e.g., Sword. */
451 /* Matches any character whose syntax is not that specified. */
456 /* Common operations on the compiled pattern. */
458 /* Store NUMBER in two contiguous bytes starting at DESTINATION. */
460 #define STORE_NUMBER(destination, number) \
462 (destination)[0] = (number) & 0377; \
463 (destination)[1] = (number) >> 8; \
466 /* Same as STORE_NUMBER, except increment DESTINATION to
467 the byte after where the number is stored. Therefore, DESTINATION
468 must be an lvalue. */
470 #define STORE_NUMBER_AND_INCR(destination, number) \
472 STORE_NUMBER (destination, number); \
473 (destination) += 2; \
476 /* Put into DESTINATION a number stored in two contiguous bytes starting
479 #define EXTRACT_NUMBER(destination, source) \
481 (destination) = *(source) & 0377; \
482 (destination) += SIGN_EXTEND_CHAR (*((source) + 1)) << 8; \
486 static void extract_number
_RE_ARGS ((int *dest
, unsigned char *source
));
488 extract_number (dest
, source
)
490 unsigned char *source
;
492 int temp
= SIGN_EXTEND_CHAR (*(source
+ 1));
493 *dest
= *source
& 0377;
497 # ifndef EXTRACT_MACROS /* To debug the macros. */
498 # undef EXTRACT_NUMBER
499 # define EXTRACT_NUMBER(dest, src) extract_number (&dest, src)
500 # endif /* not EXTRACT_MACROS */
504 /* Same as EXTRACT_NUMBER, except increment SOURCE to after the number.
505 SOURCE must be an lvalue. */
507 #define EXTRACT_NUMBER_AND_INCR(destination, source) \
509 EXTRACT_NUMBER (destination, source); \
514 static void extract_number_and_incr
_RE_ARGS ((int *destination
,
515 unsigned char **source
));
517 extract_number_and_incr (destination
, source
)
519 unsigned char **source
;
521 extract_number (destination
, *source
);
525 # ifndef EXTRACT_MACROS
526 # undef EXTRACT_NUMBER_AND_INCR
527 # define EXTRACT_NUMBER_AND_INCR(dest, src) \
528 extract_number_and_incr (&dest, &src)
529 # endif /* not EXTRACT_MACROS */
533 /* If DEBUG is defined, Regex prints many voluminous messages about what
534 it is doing (if the variable `debug' is nonzero). If linked with the
535 main program in `iregex.c', you can enter patterns and strings
536 interactively. And if linked with the main program in `main.c' and
537 the other test files, you can run the already-written tests. */
541 /* We use standard I/O for debugging. */
544 /* It is useful to test things that ``must'' be true when debugging. */
547 static int debug
= 0;
549 # define DEBUG_STATEMENT(e) e
550 # define DEBUG_PRINT1(x) if (debug) printf (x)
551 # define DEBUG_PRINT2(x1, x2) if (debug) printf (x1, x2)
552 # define DEBUG_PRINT3(x1, x2, x3) if (debug) printf (x1, x2, x3)
553 # define DEBUG_PRINT4(x1, x2, x3, x4) if (debug) printf (x1, x2, x3, x4)
554 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) \
555 if (debug) print_partial_compiled_pattern (s, e)
556 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) \
557 if (debug) print_double_string (w, s1, sz1, s2, sz2)
560 /* Print the fastmap in human-readable form. */
563 print_fastmap (fastmap
)
566 unsigned was_a_range
= 0;
569 while (i
< (1 << BYTEWIDTH
))
575 while (i
< (1 << BYTEWIDTH
) && fastmap
[i
])
591 /* Print a compiled pattern string in human-readable form, starting at
592 the START pointer into it and ending just before the pointer END. */
595 print_partial_compiled_pattern (start
, end
)
596 unsigned char *start
;
601 unsigned char *p
= start
;
602 unsigned char *pend
= end
;
610 /* Loop over pattern commands. */
613 printf ("%d:\t", p
- start
);
615 switch ((re_opcode_t
) *p
++)
623 printf ("/exactn/%d", mcnt
);
634 printf ("/start_memory/%d/%d", mcnt
, *p
++);
639 printf ("/stop_memory/%d/%d", mcnt
, *p
++);
643 printf ("/duplicate/%d", *p
++);
653 register int c
, last
= -100;
654 register int in_range
= 0;
656 printf ("/charset [%s",
657 (re_opcode_t
) *(p
- 1) == charset_not
? "^" : "");
659 assert (p
+ *p
< pend
);
661 for (c
= 0; c
< 256; c
++)
663 && (p
[1 + (c
/8)] & (1 << (c
% 8))))
665 /* Are we starting a range? */
666 if (last
+ 1 == c
&& ! in_range
)
671 /* Have we broken a range? */
672 else if (last
+ 1 != c
&& in_range
)
701 case on_failure_jump
:
702 extract_number_and_incr (&mcnt
, &p
);
703 printf ("/on_failure_jump to %d", p
+ mcnt
- start
);
706 case on_failure_keep_string_jump
:
707 extract_number_and_incr (&mcnt
, &p
);
708 printf ("/on_failure_keep_string_jump to %d", p
+ mcnt
- start
);
711 case dummy_failure_jump
:
712 extract_number_and_incr (&mcnt
, &p
);
713 printf ("/dummy_failure_jump to %d", p
+ mcnt
- start
);
716 case push_dummy_failure
:
717 printf ("/push_dummy_failure");
721 extract_number_and_incr (&mcnt
, &p
);
722 printf ("/maybe_pop_jump to %d", p
+ mcnt
- start
);
725 case pop_failure_jump
:
726 extract_number_and_incr (&mcnt
, &p
);
727 printf ("/pop_failure_jump to %d", p
+ mcnt
- start
);
731 extract_number_and_incr (&mcnt
, &p
);
732 printf ("/jump_past_alt to %d", p
+ mcnt
- start
);
736 extract_number_and_incr (&mcnt
, &p
);
737 printf ("/jump to %d", p
+ mcnt
- start
);
741 extract_number_and_incr (&mcnt
, &p
);
743 extract_number_and_incr (&mcnt2
, &p
);
744 printf ("/succeed_n to %d, %d times", p1
- start
, mcnt2
);
748 extract_number_and_incr (&mcnt
, &p
);
750 extract_number_and_incr (&mcnt2
, &p
);
751 printf ("/jump_n to %d, %d times", p1
- start
, mcnt2
);
755 extract_number_and_incr (&mcnt
, &p
);
757 extract_number_and_incr (&mcnt2
, &p
);
758 printf ("/set_number_at location %d to %d", p1
- start
, mcnt2
);
762 printf ("/wordbound");
766 printf ("/notwordbound");
778 printf ("/before_dot");
786 printf ("/after_dot");
790 printf ("/syntaxspec");
792 printf ("/%d", mcnt
);
796 printf ("/notsyntaxspec");
798 printf ("/%d", mcnt
);
803 printf ("/wordchar");
807 printf ("/notwordchar");
819 printf ("?%d", *(p
-1));
825 printf ("%d:\tend of pattern.\n", p
- start
);
830 print_compiled_pattern (bufp
)
831 struct re_pattern_buffer
*bufp
;
833 unsigned char *buffer
= bufp
->buffer
;
835 print_partial_compiled_pattern (buffer
, buffer
+ bufp
->used
);
836 printf ("%ld bytes used/%ld bytes allocated.\n",
837 bufp
->used
, bufp
->allocated
);
839 if (bufp
->fastmap_accurate
&& bufp
->fastmap
)
841 printf ("fastmap: ");
842 print_fastmap (bufp
->fastmap
);
845 printf ("re_nsub: %d\t", bufp
->re_nsub
);
846 printf ("regs_alloc: %d\t", bufp
->regs_allocated
);
847 printf ("can_be_null: %d\t", bufp
->can_be_null
);
848 printf ("newline_anchor: %d\n", bufp
->newline_anchor
);
849 printf ("no_sub: %d\t", bufp
->no_sub
);
850 printf ("not_bol: %d\t", bufp
->not_bol
);
851 printf ("not_eol: %d\t", bufp
->not_eol
);
852 printf ("syntax: %lx\n", bufp
->syntax
);
853 /* Perhaps we should print the translate table? */
858 print_double_string (where
, string1
, size1
, string2
, size2
)
871 if (FIRST_STRING_P (where
))
873 for (this_char
= where
- string1
; this_char
< size1
; this_char
++)
874 putchar (string1
[this_char
]);
879 for (this_char
= where
- string2
; this_char
< size2
; this_char
++)
880 putchar (string2
[this_char
]);
891 #else /* not DEBUG */
896 # define DEBUG_STATEMENT(e)
897 # define DEBUG_PRINT1(x)
898 # define DEBUG_PRINT2(x1, x2)
899 # define DEBUG_PRINT3(x1, x2, x3)
900 # define DEBUG_PRINT4(x1, x2, x3, x4)
901 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e)
902 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2)
904 #endif /* not DEBUG */
906 /* Set by `re_set_syntax' to the current regexp syntax to recognize. Can
907 also be assigned to arbitrarily: each pattern buffer stores its own
908 syntax, so it can be changed between regex compilations. */
909 /* This has no initializer because initialized variables in Emacs
910 become read-only after dumping. */
911 reg_syntax_t re_syntax_options
;
914 /* Specify the precise syntax of regexps for compilation. This provides
915 for compatibility for various utilities which historically have
916 different, incompatible syntaxes.
918 The argument SYNTAX is a bit mask comprised of the various bits
919 defined in gnu-regex.h. We return the old syntax. */
922 re_set_syntax (syntax
)
925 reg_syntax_t ret
= re_syntax_options
;
927 re_syntax_options
= syntax
;
929 if (syntax
& RE_DEBUG
)
931 else if (debug
) /* was on but now is not */
937 weak_alias (__re_set_syntax
, re_set_syntax
)
940 /* This table gives an error message for each of the error codes listed
941 in gnu-regex.h. Obviously the order here has to be same as there.
942 POSIX doesn't require that we do anything for REG_NOERROR,
943 but why not be nice? */
945 static const char *const re_error_msgid
[] =
947 gettext_noop ("Success"), /* REG_NOERROR */
948 gettext_noop ("No match"), /* REG_NOMATCH */
949 gettext_noop ("Invalid regular expression"), /* REG_BADPAT */
950 gettext_noop ("Invalid collation character"), /* REG_ECOLLATE */
951 gettext_noop ("Invalid character class name"), /* REG_ECTYPE */
952 gettext_noop ("Trailing backslash"), /* REG_EESCAPE */
953 gettext_noop ("Invalid back reference"), /* REG_ESUBREG */
954 gettext_noop ("Unmatched [ or [^"), /* REG_EBRACK */
955 gettext_noop ("Unmatched ( or \\("), /* REG_EPAREN */
956 gettext_noop ("Unmatched \\{"), /* REG_EBRACE */
957 gettext_noop ("Invalid content of \\{\\}"), /* REG_BADBR */
958 gettext_noop ("Invalid range end"), /* REG_ERANGE */
959 gettext_noop ("Memory exhausted"), /* REG_ESPACE */
960 gettext_noop ("Invalid preceding regular expression"), /* REG_BADRPT */
961 gettext_noop ("Premature end of regular expression"), /* REG_EEND */
962 gettext_noop ("Regular expression too big"), /* REG_ESIZE */
963 gettext_noop ("Unmatched ) or \\)"), /* REG_ERPAREN */
966 /* Avoiding alloca during matching, to placate r_alloc. */
968 /* Define MATCH_MAY_ALLOCATE unless we need to make sure that the
969 searching and matching functions should not call alloca. On some
970 systems, alloca is implemented in terms of malloc, and if we're
971 using the relocating allocator routines, then malloc could cause a
972 relocation, which might (if the strings being searched are in the
973 ralloc heap) shift the data out from underneath the regexp
976 Here's another reason to avoid allocation: Emacs
977 processes input from X in a signal handler; processing X input may
978 call malloc; if input arrives while a matching routine is calling
979 malloc, then we're scrod. But Emacs can't just block input while
980 calling matching routines; then we don't notice interrupts when
981 they come in. So, Emacs blocks input around all regexp calls
982 except the matching calls, which it leaves unprotected, in the
983 faith that they will not malloc. */
985 /* Normally, this is fine. */
986 #define MATCH_MAY_ALLOCATE
988 /* When using GNU C, we are not REALLY using the C alloca, no matter
989 what config.h may say. So don't take precautions for it. */
994 /* The match routines may not allocate if (1) they would do it with malloc
995 and (2) it's not safe for them to use malloc.
996 Note that if REL_ALLOC is defined, matching would not use malloc for the
997 failure stack, but we would still use it for the register vectors;
998 so REL_ALLOC should not affect this. */
999 #if (defined C_ALLOCA || defined REGEX_MALLOC) && defined emacs
1000 # undef MATCH_MAY_ALLOCATE
1004 /* Failure stack declarations and macros; both re_compile_fastmap and
1005 re_match_2 use a failure stack. These have to be macros because of
1006 REGEX_ALLOCATE_STACK. */
1009 /* Number of failure points for which to initially allocate space
1010 when matching. If this number is exceeded, we allocate more
1011 space, so it is not a hard limit. */
1012 #ifndef INIT_FAILURE_ALLOC
1013 # define INIT_FAILURE_ALLOC 5
1016 /* Roughly the maximum number of failure points on the stack. Would be
1017 exactly that if always used MAX_FAILURE_ITEMS items each time we failed.
1018 This is a variable only so users of regex can assign to it; we never
1019 change it ourselves. */
1023 # if defined MATCH_MAY_ALLOCATE
1024 /* 4400 was enough to cause a crash on Alpha OSF/1,
1025 whose default stack limit is 2mb. */
1026 long int re_max_failures
= 4000;
1028 long int re_max_failures
= 2000;
1031 union fail_stack_elt
1033 unsigned char *pointer
;
1037 typedef union fail_stack_elt fail_stack_elt_t
;
1041 fail_stack_elt_t
*stack
;
1042 unsigned long int size
;
1043 unsigned long int avail
; /* Offset of next open position. */
1046 #else /* not INT_IS_16BIT */
1048 # if defined MATCH_MAY_ALLOCATE
1049 /* 4400 was enough to cause a crash on Alpha OSF/1,
1050 whose default stack limit is 2mb. */
1051 int re_max_failures
= 20000;
1053 int re_max_failures
= 2000;
1056 union fail_stack_elt
1058 unsigned char *pointer
;
1062 typedef union fail_stack_elt fail_stack_elt_t
;
1066 fail_stack_elt_t
*stack
;
1068 unsigned avail
; /* Offset of next open position. */
1071 #endif /* INT_IS_16BIT */
1073 #define FAIL_STACK_EMPTY() (fail_stack.avail == 0)
1074 #define FAIL_STACK_PTR_EMPTY() (fail_stack_ptr->avail == 0)
1075 #define FAIL_STACK_FULL() (fail_stack.avail == fail_stack.size)
1078 /* Define macros to initialize and free the failure stack.
1079 Do `return -2' if the alloc fails. */
1081 #ifdef MATCH_MAY_ALLOCATE
1082 # define INIT_FAIL_STACK() \
1084 fail_stack.stack = (fail_stack_elt_t *) \
1085 REGEX_ALLOCATE_STACK (INIT_FAILURE_ALLOC * sizeof (fail_stack_elt_t)); \
1087 if (fail_stack.stack == NULL) \
1090 fail_stack.size = INIT_FAILURE_ALLOC; \
1091 fail_stack.avail = 0; \
1094 # define RESET_FAIL_STACK() REGEX_FREE_STACK (fail_stack.stack)
1096 # define INIT_FAIL_STACK() \
1098 fail_stack.avail = 0; \
1101 # define RESET_FAIL_STACK()
1105 /* Double the size of FAIL_STACK, up to approximately `re_max_failures' items.
1107 Return 1 if succeeds, and 0 if either ran out of memory
1108 allocating space for it or it was already too large.
1110 REGEX_REALLOCATE_STACK requires `destination' be declared. */
1112 #define DOUBLE_FAIL_STACK(fail_stack) \
1113 ((fail_stack).size > (unsigned) (re_max_failures * MAX_FAILURE_ITEMS) \
1115 : ((fail_stack).stack = (fail_stack_elt_t *) \
1116 REGEX_REALLOCATE_STACK ((fail_stack).stack, \
1117 (fail_stack).size * sizeof (fail_stack_elt_t), \
1118 ((fail_stack).size << 1) * sizeof (fail_stack_elt_t)), \
1120 (fail_stack).stack == NULL \
1122 : ((fail_stack).size <<= 1, \
1126 /* Push pointer POINTER on FAIL_STACK.
1127 Return 1 if was able to do so and 0 if ran out of memory allocating
1129 #define PUSH_PATTERN_OP(POINTER, FAIL_STACK) \
1130 ((FAIL_STACK_FULL () \
1131 && !DOUBLE_FAIL_STACK (FAIL_STACK)) \
1133 : ((FAIL_STACK).stack[(FAIL_STACK).avail++].pointer = POINTER, \
1136 /* Push a pointer value onto the failure stack.
1137 Assumes the variable `fail_stack'. Probably should only
1138 be called from within `PUSH_FAILURE_POINT'. */
1139 #define PUSH_FAILURE_POINTER(item) \
1140 fail_stack.stack[fail_stack.avail++].pointer = (unsigned char *) (item)
1142 /* This pushes an integer-valued item onto the failure stack.
1143 Assumes the variable `fail_stack'. Probably should only
1144 be called from within `PUSH_FAILURE_POINT'. */
1145 #define PUSH_FAILURE_INT(item) \
1146 fail_stack.stack[fail_stack.avail++].integer = (item)
1148 /* Push a fail_stack_elt_t value onto the failure stack.
1149 Assumes the variable `fail_stack'. Probably should only
1150 be called from within `PUSH_FAILURE_POINT'. */
1151 #define PUSH_FAILURE_ELT(item) \
1152 fail_stack.stack[fail_stack.avail++] = (item)
1154 /* These three POP... operations complement the three PUSH... operations.
1155 All assume that `fail_stack' is nonempty. */
1156 #define POP_FAILURE_POINTER() fail_stack.stack[--fail_stack.avail].pointer
1157 #define POP_FAILURE_INT() fail_stack.stack[--fail_stack.avail].integer
1158 #define POP_FAILURE_ELT() fail_stack.stack[--fail_stack.avail]
1160 /* Used to omit pushing failure point id's when we're not debugging. */
1162 # define DEBUG_PUSH PUSH_FAILURE_INT
1163 # define DEBUG_POP(item_addr) *(item_addr) = POP_FAILURE_INT ()
1165 # define DEBUG_PUSH(item)
1166 # define DEBUG_POP(item_addr)
1170 /* Push the information about the state we will need
1171 if we ever fail back to it.
1173 Requires variables fail_stack, regstart, regend, reg_info, and
1174 num_regs_pushed be declared. DOUBLE_FAIL_STACK requires `destination'
1177 Does `return FAILURE_CODE' if runs out of memory. */
1179 #define PUSH_FAILURE_POINT(pattern_place, string_place, failure_code) \
1181 char *destination; \
1182 /* Must be int, so when we don't save any registers, the arithmetic \
1183 of 0 + -1 isn't done as unsigned. */ \
1184 /* Can't be int, since there is not a shred of a guarantee that int \
1185 is wide enough to hold a value of something to which pointer can \
1187 active_reg_t this_reg; \
1189 DEBUG_STATEMENT (failure_id++); \
1190 DEBUG_STATEMENT (nfailure_points_pushed++); \
1191 DEBUG_PRINT2 ("\nPUSH_FAILURE_POINT #%u:\n", failure_id); \
1192 DEBUG_PRINT2 (" Before push, next avail: %d\n", (fail_stack).avail);\
1193 DEBUG_PRINT2 (" size: %d\n", (fail_stack).size);\
1195 DEBUG_PRINT2 (" slots needed: %ld\n", NUM_FAILURE_ITEMS); \
1196 DEBUG_PRINT2 (" available: %d\n", REMAINING_AVAIL_SLOTS); \
1198 /* Ensure we have enough space allocated for what we will push. */ \
1199 while (REMAINING_AVAIL_SLOTS < NUM_FAILURE_ITEMS) \
1201 if (!DOUBLE_FAIL_STACK (fail_stack)) \
1202 return failure_code; \
1204 DEBUG_PRINT2 ("\n Doubled stack; size now: %d\n", \
1205 (fail_stack).size); \
1206 DEBUG_PRINT2 (" slots available: %d\n", REMAINING_AVAIL_SLOTS);\
1209 /* Push the info, starting with the registers. */ \
1210 DEBUG_PRINT1 ("\n"); \
1213 for (this_reg = lowest_active_reg; this_reg <= highest_active_reg; \
1216 DEBUG_PRINT2 (" Pushing reg: %lu\n", this_reg); \
1217 DEBUG_STATEMENT (num_regs_pushed++); \
1219 DEBUG_PRINT2 (" start: %p\n", regstart[this_reg]); \
1220 PUSH_FAILURE_POINTER (regstart[this_reg]); \
1222 DEBUG_PRINT2 (" end: %p\n", regend[this_reg]); \
1223 PUSH_FAILURE_POINTER (regend[this_reg]); \
1225 DEBUG_PRINT2 (" info: %p\n ", \
1226 reg_info[this_reg].word.pointer); \
1227 DEBUG_PRINT2 (" match_null=%d", \
1228 REG_MATCH_NULL_STRING_P (reg_info[this_reg])); \
1229 DEBUG_PRINT2 (" active=%d", IS_ACTIVE (reg_info[this_reg])); \
1230 DEBUG_PRINT2 (" matched_something=%d", \
1231 MATCHED_SOMETHING (reg_info[this_reg])); \
1232 DEBUG_PRINT2 (" ever_matched=%d", \
1233 EVER_MATCHED_SOMETHING (reg_info[this_reg])); \
1234 DEBUG_PRINT1 ("\n"); \
1235 PUSH_FAILURE_ELT (reg_info[this_reg].word); \
1238 DEBUG_PRINT2 (" Pushing low active reg: %ld\n", lowest_active_reg);\
1239 PUSH_FAILURE_INT (lowest_active_reg); \
1241 DEBUG_PRINT2 (" Pushing high active reg: %ld\n", highest_active_reg);\
1242 PUSH_FAILURE_INT (highest_active_reg); \
1244 DEBUG_PRINT2 (" Pushing pattern %p:\n", pattern_place); \
1245 DEBUG_PRINT_COMPILED_PATTERN (bufp, pattern_place, pend); \
1246 PUSH_FAILURE_POINTER (pattern_place); \
1248 DEBUG_PRINT2 (" Pushing string %p: `", string_place); \
1249 DEBUG_PRINT_DOUBLE_STRING (string_place, string1, size1, string2, \
1251 DEBUG_PRINT1 ("'\n"); \
1252 PUSH_FAILURE_POINTER (string_place); \
1254 DEBUG_PRINT2 (" Pushing failure id: %u\n", failure_id); \
1255 DEBUG_PUSH (failure_id); \
1258 /* This is the number of items that are pushed and popped on the stack
1259 for each register. */
1260 #define NUM_REG_ITEMS 3
1262 /* Individual items aside from the registers. */
1264 # define NUM_NONREG_ITEMS 5 /* Includes failure point id. */
1266 # define NUM_NONREG_ITEMS 4
1269 /* We push at most this many items on the stack. */
1270 /* We used to use (num_regs - 1), which is the number of registers
1271 this regexp will save; but that was changed to 5
1272 to avoid stack overflow for a regexp with lots of parens. */
1273 #define MAX_FAILURE_ITEMS (5 * NUM_REG_ITEMS + NUM_NONREG_ITEMS)
1275 /* We actually push this many items. */
1276 #define NUM_FAILURE_ITEMS \
1278 ? 0 : highest_active_reg - lowest_active_reg + 1) \
1282 /* How many items can still be added to the stack without overflowing it. */
1283 #define REMAINING_AVAIL_SLOTS ((fail_stack).size - (fail_stack).avail)
1286 /* Pops what PUSH_FAIL_STACK pushes.
1288 We restore into the parameters, all of which should be lvalues:
1289 STR -- the saved data position.
1290 PAT -- the saved pattern position.
1291 LOW_REG, HIGH_REG -- the highest and lowest active registers.
1292 REGSTART, REGEND -- arrays of string positions.
1293 REG_INFO -- array of information about each subexpression.
1295 Also assumes the variables `fail_stack' and (if debugging), `bufp',
1296 `pend', `string1', `size1', `string2', and `size2'. */
1298 #define POP_FAILURE_POINT(str, pat, low_reg, high_reg, regstart, regend, reg_info)\
1300 DEBUG_STATEMENT (unsigned failure_id;) \
1301 active_reg_t this_reg; \
1302 const unsigned char *string_temp; \
1304 assert (!FAIL_STACK_EMPTY ()); \
1306 /* Remove failure points and point to how many regs pushed. */ \
1307 DEBUG_PRINT1 ("POP_FAILURE_POINT:\n"); \
1308 DEBUG_PRINT2 (" Before pop, next avail: %d\n", fail_stack.avail); \
1309 DEBUG_PRINT2 (" size: %d\n", fail_stack.size); \
1311 assert (fail_stack.avail >= NUM_NONREG_ITEMS); \
1313 DEBUG_POP (&failure_id); \
1314 DEBUG_PRINT2 (" Popping failure id: %u\n", failure_id); \
1316 /* If the saved string location is NULL, it came from an \
1317 on_failure_keep_string_jump opcode, and we want to throw away the \
1318 saved NULL, thus retaining our current position in the string. */ \
1319 string_temp = POP_FAILURE_POINTER (); \
1320 if (string_temp != NULL) \
1321 str = (const char *) string_temp; \
1323 DEBUG_PRINT2 (" Popping string %p: `", str); \
1324 DEBUG_PRINT_DOUBLE_STRING (str, string1, size1, string2, size2); \
1325 DEBUG_PRINT1 ("'\n"); \
1327 pat = (unsigned char *) POP_FAILURE_POINTER (); \
1328 DEBUG_PRINT2 (" Popping pattern %p:\n", pat); \
1329 DEBUG_PRINT_COMPILED_PATTERN (bufp, pat, pend); \
1331 /* Restore register info. */ \
1332 high_reg = (active_reg_t) POP_FAILURE_INT (); \
1333 DEBUG_PRINT2 (" Popping high active reg: %ld\n", high_reg); \
1335 low_reg = (active_reg_t) POP_FAILURE_INT (); \
1336 DEBUG_PRINT2 (" Popping low active reg: %ld\n", low_reg); \
1339 for (this_reg = high_reg; this_reg >= low_reg; this_reg--) \
1341 DEBUG_PRINT2 (" Popping reg: %ld\n", this_reg); \
1343 reg_info[this_reg].word = POP_FAILURE_ELT (); \
1344 DEBUG_PRINT2 (" info: %p\n", \
1345 reg_info[this_reg].word.pointer); \
1347 regend[this_reg] = (const char *) POP_FAILURE_POINTER (); \
1348 DEBUG_PRINT2 (" end: %p\n", regend[this_reg]); \
1350 regstart[this_reg] = (const char *) POP_FAILURE_POINTER (); \
1351 DEBUG_PRINT2 (" start: %p\n", regstart[this_reg]); \
1355 for (this_reg = highest_active_reg; this_reg > high_reg; this_reg--) \
1357 reg_info[this_reg].word.integer = 0; \
1358 regend[this_reg] = 0; \
1359 regstart[this_reg] = 0; \
1361 highest_active_reg = high_reg; \
1364 set_regs_matched_done = 0; \
1365 DEBUG_STATEMENT (nfailure_points_popped++); \
1366 } /* POP_FAILURE_POINT */
1370 /* Structure for per-register (a.k.a. per-group) information.
1371 Other register information, such as the
1372 starting and ending positions (which are addresses), and the list of
1373 inner groups (which is a bits list) are maintained in separate
1376 We are making a (strictly speaking) nonportable assumption here: that
1377 the compiler will pack our bit fields into something that fits into
1378 the type of `word', i.e., is something that fits into one item on the
1382 /* Declarations and macros for re_match_2. */
1386 fail_stack_elt_t word
;
1389 /* This field is one if this group can match the empty string,
1390 zero if not. If not yet determined, `MATCH_NULL_UNSET_VALUE'. */
1391 #define MATCH_NULL_UNSET_VALUE 3
1392 unsigned match_null_string_p
: 2;
1393 unsigned is_active
: 1;
1394 unsigned matched_something
: 1;
1395 unsigned ever_matched_something
: 1;
1397 } register_info_type
;
1399 #define REG_MATCH_NULL_STRING_P(R) ((R).bits.match_null_string_p)
1400 #define IS_ACTIVE(R) ((R).bits.is_active)
1401 #define MATCHED_SOMETHING(R) ((R).bits.matched_something)
1402 #define EVER_MATCHED_SOMETHING(R) ((R).bits.ever_matched_something)
1405 /* Call this when have matched a real character; it sets `matched' flags
1406 for the subexpressions which we are currently inside. Also records
1407 that those subexprs have matched. */
1408 #define SET_REGS_MATCHED() \
1411 if (!set_regs_matched_done) \
1414 set_regs_matched_done = 1; \
1415 for (r = lowest_active_reg; r <= highest_active_reg; r++) \
1417 MATCHED_SOMETHING (reg_info[r]) \
1418 = EVER_MATCHED_SOMETHING (reg_info[r]) \
1425 /* Registers are set to a sentinel when they haven't yet matched. */
1426 static char reg_unset_dummy
;
1427 #define REG_UNSET_VALUE (®_unset_dummy)
1428 #define REG_UNSET(e) ((e) == REG_UNSET_VALUE)
1430 /* Subroutine declarations and macros for regex_compile. */
1432 static reg_errcode_t regex_compile
_RE_ARGS ((const char *pattern
, size_t size
,
1433 reg_syntax_t syntax
,
1434 struct re_pattern_buffer
*bufp
));
1435 static void store_op1
_RE_ARGS ((re_opcode_t op
, unsigned char *loc
, int arg
));
1436 static void store_op2
_RE_ARGS ((re_opcode_t op
, unsigned char *loc
,
1437 int arg1
, int arg2
));
1438 static void insert_op1
_RE_ARGS ((re_opcode_t op
, unsigned char *loc
,
1439 int arg
, unsigned char *end
));
1440 static void insert_op2
_RE_ARGS ((re_opcode_t op
, unsigned char *loc
,
1441 int arg1
, int arg2
, unsigned char *end
));
1442 static boolean at_begline_loc_p
_RE_ARGS ((const char *pattern
, const char *p
,
1443 reg_syntax_t syntax
));
1444 static boolean at_endline_loc_p
_RE_ARGS ((const char *p
, const char *pend
,
1445 reg_syntax_t syntax
));
1446 static reg_errcode_t compile_range
_RE_ARGS ((const char **p_ptr
,
1449 reg_syntax_t syntax
,
1452 /* Fetch the next character in the uncompiled pattern---translating it
1453 if necessary. Also cast from a signed character in the constant
1454 string passed to us by the user to an unsigned char that we can use
1455 as an array index (in, e.g., `translate'). */
1457 # define PATFETCH(c) \
1458 do {if (p == pend) return REG_EEND; \
1459 c = (unsigned char) *p++; \
1460 if (translate) c = (unsigned char) translate[c]; \
1464 /* Fetch the next character in the uncompiled pattern, with no
1466 #define PATFETCH_RAW(c) \
1467 do {if (p == pend) return REG_EEND; \
1468 c = (unsigned char) *p++; \
1471 /* Go backwards one character in the pattern. */
1472 #define PATUNFETCH p--
1475 /* If `translate' is non-null, return translate[D], else just D. We
1476 cast the subscript to translate because some data is declared as
1477 `char *', to avoid warnings when a string constant is passed. But
1478 when we use a character as a subscript we must make it unsigned. */
1480 # define TRANSLATE(d) \
1481 (translate ? (char) translate[(unsigned char) (d)] : (d))
1485 /* Macros for outputting the compiled pattern into `buffer'. */
1487 /* If the buffer isn't allocated when it comes in, use this. */
1488 #define INIT_BUF_SIZE 32
1490 /* Make sure we have at least N more bytes of space in buffer. */
1491 #define GET_BUFFER_SPACE(n) \
1492 while ((unsigned long) (b - bufp->buffer + (n)) > bufp->allocated) \
1495 /* Make sure we have one more byte of buffer space and then add C to it. */
1496 #define BUF_PUSH(c) \
1498 GET_BUFFER_SPACE (1); \
1499 *b++ = (unsigned char) (c); \
1503 /* Ensure we have two more bytes of buffer space and then append C1 and C2. */
1504 #define BUF_PUSH_2(c1, c2) \
1506 GET_BUFFER_SPACE (2); \
1507 *b++ = (unsigned char) (c1); \
1508 *b++ = (unsigned char) (c2); \
1512 /* As with BUF_PUSH_2, except for three bytes. */
1513 #define BUF_PUSH_3(c1, c2, c3) \
1515 GET_BUFFER_SPACE (3); \
1516 *b++ = (unsigned char) (c1); \
1517 *b++ = (unsigned char) (c2); \
1518 *b++ = (unsigned char) (c3); \
1522 /* Store a jump with opcode OP at LOC to location TO. We store a
1523 relative address offset by the three bytes the jump itself occupies. */
1524 #define STORE_JUMP(op, loc, to) \
1525 store_op1 (op, loc, (int) ((to) - (loc) - 3))
1527 /* Likewise, for a two-argument jump. */
1528 #define STORE_JUMP2(op, loc, to, arg) \
1529 store_op2 (op, loc, (int) ((to) - (loc) - 3), arg)
1531 /* Like `STORE_JUMP', but for inserting. Assume `b' is the buffer end. */
1532 #define INSERT_JUMP(op, loc, to) \
1533 insert_op1 (op, loc, (int) ((to) - (loc) - 3), b)
1535 /* Like `STORE_JUMP2', but for inserting. Assume `b' is the buffer end. */
1536 #define INSERT_JUMP2(op, loc, to, arg) \
1537 insert_op2 (op, loc, (int) ((to) - (loc) - 3), arg, b)
1540 /* This is not an arbitrary limit: the arguments which represent offsets
1541 into the pattern are two bytes long. So if 2^16 bytes turns out to
1542 be too small, many things would have to change. */
1543 /* Any other compiler which, like MSC, has allocation limit below 2^16
1544 bytes will have to use approach similar to what was done below for
1545 MSC and drop MAX_BUF_SIZE a bit. Otherwise you may end up
1546 reallocating to 0 bytes. Such thing is not going to work too well.
1547 You have been warned!! */
1548 #if defined _MSC_VER && !defined WIN32
1549 /* Microsoft C 16-bit versions limit malloc to approx 65512 bytes.
1550 The REALLOC define eliminates a flurry of conversion warnings,
1551 but is not required. */
1552 # define MAX_BUF_SIZE 65500L
1553 # define REALLOC(p,s) realloc ((p), (size_t) (s))
1555 # define MAX_BUF_SIZE (1L << 16)
1556 # define REALLOC(p,s) realloc ((p), (s))
1559 /* Extend the buffer by twice its current size via realloc and
1560 reset the pointers that pointed into the old block to point to the
1561 correct places in the new one. If extending the buffer results in it
1562 being larger than MAX_BUF_SIZE, then flag memory exhausted. */
1563 #define EXTEND_BUFFER() \
1565 unsigned char *old_buffer = bufp->buffer; \
1566 if (bufp->allocated == MAX_BUF_SIZE) \
1568 bufp->allocated <<= 1; \
1569 if (bufp->allocated > MAX_BUF_SIZE) \
1570 bufp->allocated = MAX_BUF_SIZE; \
1571 bufp->buffer = (unsigned char *) REALLOC (bufp->buffer, bufp->allocated);\
1572 if (bufp->buffer == NULL) \
1573 return REG_ESPACE; \
1574 /* If the buffer moved, move all the pointers into it. */ \
1575 if (old_buffer != bufp->buffer) \
1577 b = (b - old_buffer) + bufp->buffer; \
1578 begalt = (begalt - old_buffer) + bufp->buffer; \
1579 if (fixup_alt_jump) \
1580 fixup_alt_jump = (fixup_alt_jump - old_buffer) + bufp->buffer;\
1582 laststart = (laststart - old_buffer) + bufp->buffer; \
1583 if (pending_exact) \
1584 pending_exact = (pending_exact - old_buffer) + bufp->buffer; \
1589 /* Since we have one byte reserved for the register number argument to
1590 {start,stop}_memory, the maximum number of groups we can report
1591 things about is what fits in that byte. */
1592 #define MAX_REGNUM 255
1594 /* But patterns can have more than `MAX_REGNUM' registers. We just
1595 ignore the excess. */
1596 typedef unsigned regnum_t
;
1599 /* Macros for the compile stack. */
1601 /* Since offsets can go either forwards or backwards, this type needs to
1602 be able to hold values from -(MAX_BUF_SIZE - 1) to MAX_BUF_SIZE - 1. */
1603 /* int may be not enough when sizeof(int) == 2. */
1604 typedef long pattern_offset_t
;
1608 pattern_offset_t begalt_offset
;
1609 pattern_offset_t fixup_alt_jump
;
1610 pattern_offset_t inner_group_offset
;
1611 pattern_offset_t laststart_offset
;
1613 } compile_stack_elt_t
;
1618 compile_stack_elt_t
*stack
;
1620 unsigned avail
; /* Offset of next open position. */
1621 } compile_stack_type
;
1624 #define INIT_COMPILE_STACK_SIZE 32
1626 #define COMPILE_STACK_EMPTY (compile_stack.avail == 0)
1627 #define COMPILE_STACK_FULL (compile_stack.avail == compile_stack.size)
1629 /* The next available element. */
1630 #define COMPILE_STACK_TOP (compile_stack.stack[compile_stack.avail])
1633 /* Set the bit for character C in a list. */
1634 #define SET_LIST_BIT(c) \
1635 (b[((unsigned char) (c)) / BYTEWIDTH] \
1636 |= 1 << (((unsigned char) c) % BYTEWIDTH))
1639 /* Get the next unsigned number in the uncompiled pattern. */
1640 #define GET_UNSIGNED_NUMBER(num) \
1644 while (ISDIGIT (c)) \
1648 num = num * 10 + c - '0'; \
1656 #if defined _LIBC || (defined HAVE_WCTYPE_H && defined HAVE_WCHAR_H)
1657 /* The GNU C library provides support for user-defined character classes
1658 and the functions from ISO C amendement 1. */
1659 # ifdef CHARCLASS_NAME_MAX
1660 # define CHAR_CLASS_MAX_LENGTH CHARCLASS_NAME_MAX
1662 /* This shouldn't happen but some implementation might still have this
1663 problem. Use a reasonable default value. */
1664 # define CHAR_CLASS_MAX_LENGTH 256
1668 # define IS_CHAR_CLASS(string) __wctype (string)
1670 # define IS_CHAR_CLASS(string) wctype (string)
1673 # define CHAR_CLASS_MAX_LENGTH 6 /* Namely, `xdigit'. */
1675 # define IS_CHAR_CLASS(string) \
1676 (STREQ (string, "alpha") || STREQ (string, "upper") \
1677 || STREQ (string, "lower") || STREQ (string, "digit") \
1678 || STREQ (string, "alnum") || STREQ (string, "xdigit") \
1679 || STREQ (string, "space") || STREQ (string, "print") \
1680 || STREQ (string, "punct") || STREQ (string, "graph") \
1681 || STREQ (string, "cntrl") || STREQ (string, "blank"))
1684 #ifndef MATCH_MAY_ALLOCATE
1686 /* If we cannot allocate large objects within re_match_2_internal,
1687 we make the fail stack and register vectors global.
1688 The fail stack, we grow to the maximum size when a regexp
1690 The register vectors, we adjust in size each time we
1691 compile a regexp, according to the number of registers it needs. */
1693 static fail_stack_type fail_stack
;
1695 /* Size with which the following vectors are currently allocated.
1696 That is so we can make them bigger as needed,
1697 but never make them smaller. */
1698 static int regs_allocated_size
;
1700 static const char ** regstart
, ** regend
;
1701 static const char ** old_regstart
, ** old_regend
;
1702 static const char **best_regstart
, **best_regend
;
1703 static register_info_type
*reg_info
;
1704 static const char **reg_dummy
;
1705 static register_info_type
*reg_info_dummy
;
1707 /* Make the register vectors big enough for NUM_REGS registers,
1708 but don't make them smaller. */
1711 regex_grow_registers (num_regs
)
1714 if (num_regs
> regs_allocated_size
)
1716 RETALLOC_IF (regstart
, num_regs
, const char *);
1717 RETALLOC_IF (regend
, num_regs
, const char *);
1718 RETALLOC_IF (old_regstart
, num_regs
, const char *);
1719 RETALLOC_IF (old_regend
, num_regs
, const char *);
1720 RETALLOC_IF (best_regstart
, num_regs
, const char *);
1721 RETALLOC_IF (best_regend
, num_regs
, const char *);
1722 RETALLOC_IF (reg_info
, num_regs
, register_info_type
);
1723 RETALLOC_IF (reg_dummy
, num_regs
, const char *);
1724 RETALLOC_IF (reg_info_dummy
, num_regs
, register_info_type
);
1726 regs_allocated_size
= num_regs
;
1730 #endif /* not MATCH_MAY_ALLOCATE */
1732 static boolean group_in_compile_stack
_RE_ARGS ((compile_stack_type
1736 /* `regex_compile' compiles PATTERN (of length SIZE) according to SYNTAX.
1737 Returns one of error codes defined in `gnu-regex.h', or zero for success.
1739 Assumes the `allocated' (and perhaps `buffer') and `translate'
1740 fields are set in BUFP on entry.
1742 If it succeeds, results are put in BUFP (if it returns an error, the
1743 contents of BUFP are undefined):
1744 `buffer' is the compiled pattern;
1745 `syntax' is set to SYNTAX;
1746 `used' is set to the length of the compiled pattern;
1747 `fastmap_accurate' is zero;
1748 `re_nsub' is the number of subexpressions in PATTERN;
1749 `not_bol' and `not_eol' are zero;
1751 The `fastmap' and `newline_anchor' fields are neither
1752 examined nor set. */
1754 /* Return, freeing storage we allocated. */
1755 #define FREE_STACK_RETURN(value) \
1756 return (free (compile_stack.stack), value)
1758 static reg_errcode_t
1759 regex_compile (pattern
, size
, syntax
, bufp
)
1760 const char *pattern
;
1762 reg_syntax_t syntax
;
1763 struct re_pattern_buffer
*bufp
;
1765 /* We fetch characters from PATTERN here. Even though PATTERN is
1766 `char *' (i.e., signed), we declare these variables as unsigned, so
1767 they can be reliably used as array indices. */
1768 register unsigned char c
, c1
;
1770 /* A random temporary spot in PATTERN. */
1773 /* Points to the end of the buffer, where we should append. */
1774 register unsigned char *b
;
1776 /* Keeps track of unclosed groups. */
1777 compile_stack_type compile_stack
;
1779 /* Points to the current (ending) position in the pattern. */
1780 const char *p
= pattern
;
1781 const char *pend
= pattern
+ size
;
1783 /* How to translate the characters in the pattern. */
1784 RE_TRANSLATE_TYPE translate
= bufp
->translate
;
1786 /* Address of the count-byte of the most recently inserted `exactn'
1787 command. This makes it possible to tell if a new exact-match
1788 character can be added to that command or if the character requires
1789 a new `exactn' command. */
1790 unsigned char *pending_exact
= 0;
1792 /* Address of start of the most recently finished expression.
1793 This tells, e.g., postfix * where to find the start of its
1794 operand. Reset at the beginning of groups and alternatives. */
1795 unsigned char *laststart
= 0;
1797 /* Address of beginning of regexp, or inside of last group. */
1798 unsigned char *begalt
;
1800 /* Place in the uncompiled pattern (i.e., the {) to
1801 which to go back if the interval is invalid. */
1802 const char *beg_interval
;
1804 /* Address of the place where a forward jump should go to the end of
1805 the containing expression. Each alternative of an `or' -- except the
1806 last -- ends with a forward jump of this sort. */
1807 unsigned char *fixup_alt_jump
= 0;
1809 /* Counts open-groups as they are encountered. Remembered for the
1810 matching close-group on the compile stack, so the same register
1811 number is put in the stop_memory as the start_memory. */
1812 regnum_t regnum
= 0;
1815 DEBUG_PRINT1 ("\nCompiling pattern: ");
1818 unsigned debug_count
;
1820 for (debug_count
= 0; debug_count
< size
; debug_count
++)
1821 putchar (pattern
[debug_count
]);
1826 /* Initialize the compile stack. */
1827 compile_stack
.stack
= TALLOC (INIT_COMPILE_STACK_SIZE
, compile_stack_elt_t
);
1828 if (compile_stack
.stack
== NULL
)
1831 compile_stack
.size
= INIT_COMPILE_STACK_SIZE
;
1832 compile_stack
.avail
= 0;
1834 /* Initialize the pattern buffer. */
1835 bufp
->syntax
= syntax
;
1836 bufp
->fastmap_accurate
= 0;
1837 bufp
->not_bol
= bufp
->not_eol
= 0;
1839 /* Set `used' to zero, so that if we return an error, the pattern
1840 printer (for debugging) will think there's no pattern. We reset it
1844 /* Always count groups, whether or not bufp->no_sub is set. */
1847 #if !defined emacs && !defined SYNTAX_TABLE
1848 /* Initialize the syntax table. */
1849 init_syntax_once ();
1852 if (bufp
->allocated
== 0)
1855 { /* If zero allocated, but buffer is non-null, try to realloc
1856 enough space. This loses if buffer's address is bogus, but
1857 that is the user's responsibility. */
1858 RETALLOC (bufp
->buffer
, INIT_BUF_SIZE
, unsigned char);
1861 { /* Caller did not allocate a buffer. Do it for them. */
1862 bufp
->buffer
= TALLOC (INIT_BUF_SIZE
, unsigned char);
1864 if (!bufp
->buffer
) FREE_STACK_RETURN (REG_ESPACE
);
1866 bufp
->allocated
= INIT_BUF_SIZE
;
1869 begalt
= b
= bufp
->buffer
;
1871 /* Loop through the uncompiled pattern until we're at the end. */
1880 if ( /* If at start of pattern, it's an operator. */
1882 /* If context independent, it's an operator. */
1883 || syntax
& RE_CONTEXT_INDEP_ANCHORS
1884 /* Otherwise, depends on what's come before. */
1885 || at_begline_loc_p (pattern
, p
, syntax
))
1895 if ( /* If at end of pattern, it's an operator. */
1897 /* If context independent, it's an operator. */
1898 || syntax
& RE_CONTEXT_INDEP_ANCHORS
1899 /* Otherwise, depends on what's next. */
1900 || at_endline_loc_p (p
, pend
, syntax
))
1910 if ((syntax
& RE_BK_PLUS_QM
)
1911 || (syntax
& RE_LIMITED_OPS
))
1915 /* If there is no previous pattern... */
1918 if (syntax
& RE_CONTEXT_INVALID_OPS
)
1919 FREE_STACK_RETURN (REG_BADRPT
);
1920 else if (!(syntax
& RE_CONTEXT_INDEP_OPS
))
1925 /* Are we optimizing this jump? */
1926 boolean keep_string_p
= false;
1928 /* 1 means zero (many) matches is allowed. */
1929 char zero_times_ok
= 0, many_times_ok
= 0;
1931 /* If there is a sequence of repetition chars, collapse it
1932 down to just one (the right one). We can't combine
1933 interval operators with these because of, e.g., `a{2}*',
1934 which should only match an even number of `a's. */
1938 zero_times_ok
|= c
!= '+';
1939 many_times_ok
|= c
!= '?';
1947 || (!(syntax
& RE_BK_PLUS_QM
) && (c
== '+' || c
== '?')))
1950 else if (syntax
& RE_BK_PLUS_QM
&& c
== '\\')
1952 if (p
== pend
) FREE_STACK_RETURN (REG_EESCAPE
);
1955 if (!(c1
== '+' || c1
== '?'))
1970 /* If we get here, we found another repeat character. */
1973 /* Star, etc. applied to an empty pattern is equivalent
1974 to an empty pattern. */
1978 /* Now we know whether or not zero matches is allowed
1979 and also whether or not two or more matches is allowed. */
1981 { /* More than one repetition is allowed, so put in at the
1982 end a backward relative jump from `b' to before the next
1983 jump we're going to put in below (which jumps from
1984 laststart to after this jump).
1986 But if we are at the `*' in the exact sequence `.*\n',
1987 insert an unconditional jump backwards to the .,
1988 instead of the beginning of the loop. This way we only
1989 push a failure point once, instead of every time
1990 through the loop. */
1991 assert (p
- 1 > pattern
);
1993 /* Allocate the space for the jump. */
1994 GET_BUFFER_SPACE (3);
1996 /* We know we are not at the first character of the pattern,
1997 because laststart was nonzero. And we've already
1998 incremented `p', by the way, to be the character after
1999 the `*'. Do we have to do something analogous here
2000 for null bytes, because of RE_DOT_NOT_NULL? */
2001 if (TRANSLATE (*(p
- 2)) == TRANSLATE ('.')
2003 && p
< pend
&& TRANSLATE (*p
) == TRANSLATE ('\n')
2004 && !(syntax
& RE_DOT_NEWLINE
))
2005 { /* We have .*\n. */
2006 STORE_JUMP (jump
, b
, laststart
);
2007 keep_string_p
= true;
2010 /* Anything else. */
2011 STORE_JUMP (maybe_pop_jump
, b
, laststart
- 3);
2013 /* We've added more stuff to the buffer. */
2017 /* On failure, jump from laststart to b + 3, which will be the
2018 end of the buffer after this jump is inserted. */
2019 GET_BUFFER_SPACE (3);
2020 INSERT_JUMP (keep_string_p
? on_failure_keep_string_jump
2028 /* At least one repetition is required, so insert a
2029 `dummy_failure_jump' before the initial
2030 `on_failure_jump' instruction of the loop. This
2031 effects a skip over that instruction the first time
2032 we hit that loop. */
2033 GET_BUFFER_SPACE (3);
2034 INSERT_JUMP (dummy_failure_jump
, laststart
, laststart
+ 6);
2049 boolean had_char_class
= false;
2051 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2053 /* Ensure that we have enough space to push a charset: the
2054 opcode, the length count, and the bitset; 34 bytes in all. */
2055 GET_BUFFER_SPACE (34);
2059 /* We test `*p == '^' twice, instead of using an if
2060 statement, so we only need one BUF_PUSH. */
2061 BUF_PUSH (*p
== '^' ? charset_not
: charset
);
2065 /* Remember the first position in the bracket expression. */
2068 /* Push the number of bytes in the bitmap. */
2069 BUF_PUSH ((1 << BYTEWIDTH
) / BYTEWIDTH
);
2071 /* Clear the whole map. */
2072 bzero (b
, (1 << BYTEWIDTH
) / BYTEWIDTH
);
2074 /* charset_not matches newline according to a syntax bit. */
2075 if ((re_opcode_t
) b
[-2] == charset_not
2076 && (syntax
& RE_HAT_LISTS_NOT_NEWLINE
))
2077 SET_LIST_BIT ('\n');
2079 /* Read in characters and ranges, setting map bits. */
2082 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2086 /* \ might escape characters inside [...] and [^...]. */
2087 if ((syntax
& RE_BACKSLASH_ESCAPE_IN_LISTS
) && c
== '\\')
2089 if (p
== pend
) FREE_STACK_RETURN (REG_EESCAPE
);
2096 /* Could be the end of the bracket expression. If it's
2097 not (i.e., when the bracket expression is `[]' so
2098 far), the ']' character bit gets set way below. */
2099 if (c
== ']' && p
!= p1
+ 1)
2102 /* Look ahead to see if it's a range when the last thing
2103 was a character class. */
2104 if (had_char_class
&& c
== '-' && *p
!= ']')
2105 FREE_STACK_RETURN (REG_ERANGE
);
2107 /* Look ahead to see if it's a range when the last thing
2108 was a character: if this is a hyphen not at the
2109 beginning or the end of a list, then it's the range
2112 && !(p
- 2 >= pattern
&& p
[-2] == '[')
2113 && !(p
- 3 >= pattern
&& p
[-3] == '[' && p
[-2] == '^')
2117 = compile_range (&p
, pend
, translate
, syntax
, b
);
2118 if (ret
!= REG_NOERROR
) FREE_STACK_RETURN (ret
);
2121 else if (p
[0] == '-' && p
[1] != ']')
2122 { /* This handles ranges made up of characters only. */
2125 /* Move past the `-'. */
2128 ret
= compile_range (&p
, pend
, translate
, syntax
, b
);
2129 if (ret
!= REG_NOERROR
) FREE_STACK_RETURN (ret
);
2132 /* See if we're at the beginning of a possible character
2135 else if (syntax
& RE_CHAR_CLASSES
&& c
== '[' && *p
== ':')
2136 { /* Leave room for the null. */
2137 char str
[CHAR_CLASS_MAX_LENGTH
+ 1];
2142 /* If pattern is `[[:'. */
2143 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2148 if ((c
== ':' && *p
== ']') || p
== pend
2149 || c1
== CHAR_CLASS_MAX_LENGTH
)
2155 /* If isn't a word bracketed by `[:' and `:]':
2156 undo the ending character, the letters, and leave
2157 the leading `:' and `[' (but set bits for them). */
2158 if (c
== ':' && *p
== ']')
2160 /* GCC LOCAL: Skip this code if we don't have btowc(). btowc() is */
2161 /* defined in the 1994 Amendment 1 to ISO C and may not be present on */
2162 /* systems where we have wchar.h and wctype.h. */
2163 #if defined _LIBC || (defined HAVE_WCTYPE_H && defined HAVE_WCHAR_H && defined HAVE_BTOWC)
2164 boolean is_lower
= STREQ (str
, "lower");
2165 boolean is_upper
= STREQ (str
, "upper");
2169 wt
= IS_CHAR_CLASS (str
);
2171 FREE_STACK_RETURN (REG_ECTYPE
);
2173 /* Throw away the ] at the end of the character
2177 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2179 for (ch
= 0; ch
< 1 << BYTEWIDTH
; ++ch
)
2182 if (__iswctype (__btowc (ch
), wt
))
2185 if (iswctype (btowc (ch
), wt
))
2189 if (translate
&& (is_upper
|| is_lower
)
2190 && (ISUPPER (ch
) || ISLOWER (ch
)))
2194 had_char_class
= true;
2197 boolean is_alnum
= STREQ (str
, "alnum");
2198 boolean is_alpha
= STREQ (str
, "alpha");
2199 boolean is_blank
= STREQ (str
, "blank");
2200 boolean is_cntrl
= STREQ (str
, "cntrl");
2201 boolean is_digit
= STREQ (str
, "digit");
2202 boolean is_graph
= STREQ (str
, "graph");
2203 boolean is_lower
= STREQ (str
, "lower");
2204 boolean is_print
= STREQ (str
, "print");
2205 boolean is_punct
= STREQ (str
, "punct");
2206 boolean is_space
= STREQ (str
, "space");
2207 boolean is_upper
= STREQ (str
, "upper");
2208 boolean is_xdigit
= STREQ (str
, "xdigit");
2210 if (!IS_CHAR_CLASS (str
))
2211 FREE_STACK_RETURN (REG_ECTYPE
);
2213 /* Throw away the ] at the end of the character
2217 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2219 for (ch
= 0; ch
< 1 << BYTEWIDTH
; ch
++)
2221 /* This was split into 3 if's to
2222 avoid an arbitrary limit in some compiler. */
2223 if ( (is_alnum
&& ISALNUM (ch
))
2224 || (is_alpha
&& ISALPHA (ch
))
2225 || (is_blank
&& ISBLANK (ch
))
2226 || (is_cntrl
&& ISCNTRL (ch
)))
2228 if ( (is_digit
&& ISDIGIT (ch
))
2229 || (is_graph
&& ISGRAPH (ch
))
2230 || (is_lower
&& ISLOWER (ch
))
2231 || (is_print
&& ISPRINT (ch
)))
2233 if ( (is_punct
&& ISPUNCT (ch
))
2234 || (is_space
&& ISSPACE (ch
))
2235 || (is_upper
&& ISUPPER (ch
))
2236 || (is_xdigit
&& ISXDIGIT (ch
)))
2238 if ( translate
&& (is_upper
|| is_lower
)
2239 && (ISUPPER (ch
) || ISLOWER (ch
)))
2242 had_char_class
= true;
2243 #endif /* libc || wctype.h */
2252 had_char_class
= false;
2257 had_char_class
= false;
2262 /* Discard any (non)matching list bytes that are all 0 at the
2263 end of the map. Decrease the map-length byte too. */
2264 while ((int) b
[-1] > 0 && b
[b
[-1] - 1] == 0)
2272 if (syntax
& RE_NO_BK_PARENS
)
2279 if (syntax
& RE_NO_BK_PARENS
)
2286 if (syntax
& RE_NEWLINE_ALT
)
2293 if (syntax
& RE_NO_BK_VBAR
)
2300 if (syntax
& RE_INTERVALS
&& syntax
& RE_NO_BK_BRACES
)
2301 goto handle_interval
;
2307 if (p
== pend
) FREE_STACK_RETURN (REG_EESCAPE
);
2309 /* Do not translate the character after the \, so that we can
2310 distinguish, e.g., \B from \b, even if we normally would
2311 translate, e.g., B to b. */
2317 if (syntax
& RE_NO_BK_PARENS
)
2318 goto normal_backslash
;
2324 if (COMPILE_STACK_FULL
)
2326 RETALLOC (compile_stack
.stack
, compile_stack
.size
<< 1,
2327 compile_stack_elt_t
);
2328 if (compile_stack
.stack
== NULL
) return REG_ESPACE
;
2330 compile_stack
.size
<<= 1;
2333 /* These are the values to restore when we hit end of this
2334 group. They are all relative offsets, so that if the
2335 whole pattern moves because of realloc, they will still
2337 COMPILE_STACK_TOP
.begalt_offset
= begalt
- bufp
->buffer
;
2338 COMPILE_STACK_TOP
.fixup_alt_jump
2339 = fixup_alt_jump
? fixup_alt_jump
- bufp
->buffer
+ 1 : 0;
2340 COMPILE_STACK_TOP
.laststart_offset
= b
- bufp
->buffer
;
2341 COMPILE_STACK_TOP
.regnum
= regnum
;
2343 /* We will eventually replace the 0 with the number of
2344 groups inner to this one. But do not push a
2345 start_memory for groups beyond the last one we can
2346 represent in the compiled pattern. */
2347 if (regnum
<= MAX_REGNUM
)
2349 COMPILE_STACK_TOP
.inner_group_offset
= b
- bufp
->buffer
+ 2;
2350 BUF_PUSH_3 (start_memory
, regnum
, 0);
2353 compile_stack
.avail
++;
2358 /* If we've reached MAX_REGNUM groups, then this open
2359 won't actually generate any code, so we'll have to
2360 clear pending_exact explicitly. */
2366 if (syntax
& RE_NO_BK_PARENS
) goto normal_backslash
;
2368 if (COMPILE_STACK_EMPTY
)
2370 if (syntax
& RE_UNMATCHED_RIGHT_PAREN_ORD
)
2371 goto normal_backslash
;
2373 FREE_STACK_RETURN (REG_ERPAREN
);
2378 { /* Push a dummy failure point at the end of the
2379 alternative for a possible future
2380 `pop_failure_jump' to pop. See comments at
2381 `push_dummy_failure' in `re_match_2'. */
2382 BUF_PUSH (push_dummy_failure
);
2384 /* We allocated space for this jump when we assigned
2385 to `fixup_alt_jump', in the `handle_alt' case below. */
2386 STORE_JUMP (jump_past_alt
, fixup_alt_jump
, b
- 1);
2389 /* See similar code for backslashed left paren above. */
2390 if (COMPILE_STACK_EMPTY
)
2392 if (syntax
& RE_UNMATCHED_RIGHT_PAREN_ORD
)
2395 FREE_STACK_RETURN (REG_ERPAREN
);
2398 /* Since we just checked for an empty stack above, this
2399 ``can't happen''. */
2400 assert (compile_stack
.avail
!= 0);
2402 /* We don't just want to restore into `regnum', because
2403 later groups should continue to be numbered higher,
2404 as in `(ab)c(de)' -- the second group is #2. */
2405 regnum_t this_group_regnum
;
2407 compile_stack
.avail
--;
2408 begalt
= bufp
->buffer
+ COMPILE_STACK_TOP
.begalt_offset
;
2410 = COMPILE_STACK_TOP
.fixup_alt_jump
2411 ? bufp
->buffer
+ COMPILE_STACK_TOP
.fixup_alt_jump
- 1
2413 laststart
= bufp
->buffer
+ COMPILE_STACK_TOP
.laststart_offset
;
2414 this_group_regnum
= COMPILE_STACK_TOP
.regnum
;
2415 /* If we've reached MAX_REGNUM groups, then this open
2416 won't actually generate any code, so we'll have to
2417 clear pending_exact explicitly. */
2420 /* We're at the end of the group, so now we know how many
2421 groups were inside this one. */
2422 if (this_group_regnum
<= MAX_REGNUM
)
2424 unsigned char *inner_group_loc
2425 = bufp
->buffer
+ COMPILE_STACK_TOP
.inner_group_offset
;
2427 *inner_group_loc
= regnum
- this_group_regnum
;
2428 BUF_PUSH_3 (stop_memory
, this_group_regnum
,
2429 regnum
- this_group_regnum
);
2435 case '|': /* `\|'. */
2436 if (syntax
& RE_LIMITED_OPS
|| syntax
& RE_NO_BK_VBAR
)
2437 goto normal_backslash
;
2439 if (syntax
& RE_LIMITED_OPS
)
2442 /* Insert before the previous alternative a jump which
2443 jumps to this alternative if the former fails. */
2444 GET_BUFFER_SPACE (3);
2445 INSERT_JUMP (on_failure_jump
, begalt
, b
+ 6);
2449 /* The alternative before this one has a jump after it
2450 which gets executed if it gets matched. Adjust that
2451 jump so it will jump to this alternative's analogous
2452 jump (put in below, which in turn will jump to the next
2453 (if any) alternative's such jump, etc.). The last such
2454 jump jumps to the correct final destination. A picture:
2460 If we are at `b', then fixup_alt_jump right now points to a
2461 three-byte space after `a'. We'll put in the jump, set
2462 fixup_alt_jump to right after `b', and leave behind three
2463 bytes which we'll fill in when we get to after `c'. */
2466 STORE_JUMP (jump_past_alt
, fixup_alt_jump
, b
);
2468 /* Mark and leave space for a jump after this alternative,
2469 to be filled in later either by next alternative or
2470 when know we're at the end of a series of alternatives. */
2472 GET_BUFFER_SPACE (3);
2481 /* If \{ is a literal. */
2482 if (!(syntax
& RE_INTERVALS
)
2483 /* If we're at `\{' and it's not the open-interval
2485 || ((syntax
& RE_INTERVALS
) && (syntax
& RE_NO_BK_BRACES
))
2486 || (p
- 2 == pattern
&& p
== pend
))
2487 goto normal_backslash
;
2491 /* If got here, then the syntax allows intervals. */
2493 /* At least (most) this many matches must be made. */
2494 int lower_bound
= -1, upper_bound
= -1;
2496 beg_interval
= p
- 1;
2500 if (syntax
& RE_NO_BK_BRACES
)
2501 goto unfetch_interval
;
2503 FREE_STACK_RETURN (REG_EBRACE
);
2506 GET_UNSIGNED_NUMBER (lower_bound
);
2510 GET_UNSIGNED_NUMBER (upper_bound
);
2511 if (upper_bound
< 0) upper_bound
= RE_DUP_MAX
;
2514 /* Interval such as `{1}' => match exactly once. */
2515 upper_bound
= lower_bound
;
2517 if (lower_bound
< 0 || upper_bound
> RE_DUP_MAX
2518 || lower_bound
> upper_bound
)
2520 if (syntax
& RE_NO_BK_BRACES
)
2521 goto unfetch_interval
;
2523 FREE_STACK_RETURN (REG_BADBR
);
2526 if (!(syntax
& RE_NO_BK_BRACES
))
2528 if (c
!= '\\') FREE_STACK_RETURN (REG_EBRACE
);
2535 if (syntax
& RE_NO_BK_BRACES
)
2536 goto unfetch_interval
;
2538 FREE_STACK_RETURN (REG_BADBR
);
2541 /* We just parsed a valid interval. */
2543 /* If it's invalid to have no preceding re. */
2546 if (syntax
& RE_CONTEXT_INVALID_OPS
)
2547 FREE_STACK_RETURN (REG_BADRPT
);
2548 else if (syntax
& RE_CONTEXT_INDEP_OPS
)
2551 goto unfetch_interval
;
2554 /* If the upper bound is zero, don't want to succeed at
2555 all; jump from `laststart' to `b + 3', which will be
2556 the end of the buffer after we insert the jump. */
2557 if (upper_bound
== 0)
2559 GET_BUFFER_SPACE (3);
2560 INSERT_JUMP (jump
, laststart
, b
+ 3);
2564 /* Otherwise, we have a nontrivial interval. When
2565 we're all done, the pattern will look like:
2566 set_number_at <jump count> <upper bound>
2567 set_number_at <succeed_n count> <lower bound>
2568 succeed_n <after jump addr> <succeed_n count>
2570 jump_n <succeed_n addr> <jump count>
2571 (The upper bound and `jump_n' are omitted if
2572 `upper_bound' is 1, though.) */
2574 { /* If the upper bound is > 1, we need to insert
2575 more at the end of the loop. */
2576 unsigned nbytes
= 10 + (upper_bound
> 1) * 10;
2578 GET_BUFFER_SPACE (nbytes
);
2580 /* Initialize lower bound of the `succeed_n', even
2581 though it will be set during matching by its
2582 attendant `set_number_at' (inserted next),
2583 because `re_compile_fastmap' needs to know.
2584 Jump to the `jump_n' we might insert below. */
2585 INSERT_JUMP2 (succeed_n
, laststart
,
2586 b
+ 5 + (upper_bound
> 1) * 5,
2590 /* Code to initialize the lower bound. Insert
2591 before the `succeed_n'. The `5' is the last two
2592 bytes of this `set_number_at', plus 3 bytes of
2593 the following `succeed_n'. */
2594 insert_op2 (set_number_at
, laststart
, 5, lower_bound
, b
);
2597 if (upper_bound
> 1)
2598 { /* More than one repetition is allowed, so
2599 append a backward jump to the `succeed_n'
2600 that starts this interval.
2602 When we've reached this during matching,
2603 we'll have matched the interval once, so
2604 jump back only `upper_bound - 1' times. */
2605 STORE_JUMP2 (jump_n
, b
, laststart
+ 5,
2609 /* The location we want to set is the second
2610 parameter of the `jump_n'; that is `b-2' as
2611 an absolute address. `laststart' will be
2612 the `set_number_at' we're about to insert;
2613 `laststart+3' the number to set, the source
2614 for the relative address. But we are
2615 inserting into the middle of the pattern --
2616 so everything is getting moved up by 5.
2617 Conclusion: (b - 2) - (laststart + 3) + 5,
2618 i.e., b - laststart.
2620 We insert this at the beginning of the loop
2621 so that if we fail during matching, we'll
2622 reinitialize the bounds. */
2623 insert_op2 (set_number_at
, laststart
, b
- laststart
,
2624 upper_bound
- 1, b
);
2629 beg_interval
= NULL
;
2634 /* If an invalid interval, match the characters as literals. */
2635 assert (beg_interval
);
2637 beg_interval
= NULL
;
2639 /* normal_char and normal_backslash need `c'. */
2642 if (!(syntax
& RE_NO_BK_BRACES
))
2644 if (p
> pattern
&& p
[-1] == '\\')
2645 goto normal_backslash
;
2650 /* There is no way to specify the before_dot and after_dot
2651 operators. rms says this is ok. --karl */
2659 BUF_PUSH_2 (syntaxspec
, syntax_spec_code
[c
]);
2665 BUF_PUSH_2 (notsyntaxspec
, syntax_spec_code
[c
]);
2671 if (syntax
& RE_NO_GNU_OPS
)
2674 BUF_PUSH (wordchar
);
2679 if (syntax
& RE_NO_GNU_OPS
)
2682 BUF_PUSH (notwordchar
);
2687 if (syntax
& RE_NO_GNU_OPS
)
2693 if (syntax
& RE_NO_GNU_OPS
)
2699 if (syntax
& RE_NO_GNU_OPS
)
2701 BUF_PUSH (wordbound
);
2705 if (syntax
& RE_NO_GNU_OPS
)
2707 BUF_PUSH (notwordbound
);
2711 if (syntax
& RE_NO_GNU_OPS
)
2717 if (syntax
& RE_NO_GNU_OPS
)
2722 case '1': case '2': case '3': case '4': case '5':
2723 case '6': case '7': case '8': case '9':
2724 if (syntax
& RE_NO_BK_REFS
)
2730 FREE_STACK_RETURN (REG_ESUBREG
);
2732 /* Can't back reference to a subexpression if inside of it. */
2733 if (group_in_compile_stack (compile_stack
, (regnum_t
) c1
))
2737 BUF_PUSH_2 (duplicate
, c1
);
2743 if (syntax
& RE_BK_PLUS_QM
)
2746 goto normal_backslash
;
2750 /* You might think it would be useful for \ to mean
2751 not to translate; but if we don't translate it
2752 it will never match anything. */
2760 /* Expects the character in `c'. */
2762 /* If no exactn currently being built. */
2765 /* If last exactn not at current position. */
2766 || pending_exact
+ *pending_exact
+ 1 != b
2768 /* We have only one byte following the exactn for the count. */
2769 || *pending_exact
== (1 << BYTEWIDTH
) - 1
2771 /* If followed by a repetition operator. */
2772 || *p
== '*' || *p
== '^'
2773 || ((syntax
& RE_BK_PLUS_QM
)
2774 ? *p
== '\\' && (p
[1] == '+' || p
[1] == '?')
2775 : (*p
== '+' || *p
== '?'))
2776 || ((syntax
& RE_INTERVALS
)
2777 && ((syntax
& RE_NO_BK_BRACES
)
2779 : (p
[0] == '\\' && p
[1] == '{'))))
2781 /* Start building a new exactn. */
2785 BUF_PUSH_2 (exactn
, 0);
2786 pending_exact
= b
- 1;
2793 } /* while p != pend */
2796 /* Through the pattern now. */
2799 STORE_JUMP (jump_past_alt
, fixup_alt_jump
, b
);
2801 if (!COMPILE_STACK_EMPTY
)
2802 FREE_STACK_RETURN (REG_EPAREN
);
2804 /* If we don't want backtracking, force success
2805 the first time we reach the end of the compiled pattern. */
2806 if (syntax
& RE_NO_POSIX_BACKTRACKING
)
2809 free (compile_stack
.stack
);
2811 /* We have succeeded; set the length of the buffer. */
2812 bufp
->used
= b
- bufp
->buffer
;
2817 DEBUG_PRINT1 ("\nCompiled pattern: \n");
2818 print_compiled_pattern (bufp
);
2822 #ifndef MATCH_MAY_ALLOCATE
2823 /* Initialize the failure stack to the largest possible stack. This
2824 isn't necessary unless we're trying to avoid calling alloca in
2825 the search and match routines. */
2827 int num_regs
= bufp
->re_nsub
+ 1;
2829 /* Since DOUBLE_FAIL_STACK refuses to double only if the current size
2830 is strictly greater than re_max_failures, the largest possible stack
2831 is 2 * re_max_failures failure points. */
2832 if (fail_stack
.size
< (2 * re_max_failures
* MAX_FAILURE_ITEMS
))
2834 fail_stack
.size
= (2 * re_max_failures
* MAX_FAILURE_ITEMS
);
2837 if (! fail_stack
.stack
)
2839 = (fail_stack_elt_t
*) xmalloc (fail_stack
.size
2840 * sizeof (fail_stack_elt_t
));
2843 = (fail_stack_elt_t
*) xrealloc (fail_stack
.stack
,
2845 * sizeof (fail_stack_elt_t
)));
2846 # else /* not emacs */
2847 if (! fail_stack
.stack
)
2849 = (fail_stack_elt_t
*) malloc (fail_stack
.size
2850 * sizeof (fail_stack_elt_t
));
2853 = (fail_stack_elt_t
*) realloc (fail_stack
.stack
,
2855 * sizeof (fail_stack_elt_t
)));
2856 # endif /* not emacs */
2859 regex_grow_registers (num_regs
);
2861 #endif /* not MATCH_MAY_ALLOCATE */
2864 } /* regex_compile */
2866 /* Subroutines for `regex_compile'. */
2868 /* Store OP at LOC followed by two-byte integer parameter ARG. */
2871 store_op1 (op
, loc
, arg
)
2876 *loc
= (unsigned char) op
;
2877 STORE_NUMBER (loc
+ 1, arg
);
2881 /* Like `store_op1', but for two two-byte parameters ARG1 and ARG2. */
2884 store_op2 (op
, loc
, arg1
, arg2
)
2889 *loc
= (unsigned char) op
;
2890 STORE_NUMBER (loc
+ 1, arg1
);
2891 STORE_NUMBER (loc
+ 3, arg2
);
2895 /* Copy the bytes from LOC to END to open up three bytes of space at LOC
2896 for OP followed by two-byte integer parameter ARG. */
2899 insert_op1 (op
, loc
, arg
, end
)
2905 register unsigned char *pfrom
= end
;
2906 register unsigned char *pto
= end
+ 3;
2908 while (pfrom
!= loc
)
2911 store_op1 (op
, loc
, arg
);
2915 /* Like `insert_op1', but for two two-byte parameters ARG1 and ARG2. */
2918 insert_op2 (op
, loc
, arg1
, arg2
, end
)
2924 register unsigned char *pfrom
= end
;
2925 register unsigned char *pto
= end
+ 5;
2927 while (pfrom
!= loc
)
2930 store_op2 (op
, loc
, arg1
, arg2
);
2934 /* P points to just after a ^ in PATTERN. Return true if that ^ comes
2935 after an alternative or a begin-subexpression. We assume there is at
2936 least one character before the ^. */
2939 at_begline_loc_p (pattern
, p
, syntax
)
2940 const char *pattern
, *p
;
2941 reg_syntax_t syntax
;
2943 const char *prev
= p
- 2;
2944 boolean prev_prev_backslash
= prev
> pattern
&& prev
[-1] == '\\';
2947 /* After a subexpression? */
2948 (*prev
== '(' && (syntax
& RE_NO_BK_PARENS
|| prev_prev_backslash
))
2949 /* After an alternative? */
2950 || (*prev
== '|' && (syntax
& RE_NO_BK_VBAR
|| prev_prev_backslash
));
2954 /* The dual of at_begline_loc_p. This one is for $. We assume there is
2955 at least one character after the $, i.e., `P < PEND'. */
2958 at_endline_loc_p (p
, pend
, syntax
)
2959 const char *p
, *pend
;
2960 reg_syntax_t syntax
;
2962 const char *next
= p
;
2963 boolean next_backslash
= *next
== '\\';
2964 const char *next_next
= p
+ 1 < pend
? p
+ 1 : 0;
2967 /* Before a subexpression? */
2968 (syntax
& RE_NO_BK_PARENS
? *next
== ')'
2969 : next_backslash
&& next_next
&& *next_next
== ')')
2970 /* Before an alternative? */
2971 || (syntax
& RE_NO_BK_VBAR
? *next
== '|'
2972 : next_backslash
&& next_next
&& *next_next
== '|');
2976 /* Returns true if REGNUM is in one of COMPILE_STACK's elements and
2977 false if it's not. */
2980 group_in_compile_stack (compile_stack
, regnum
)
2981 compile_stack_type compile_stack
;
2986 for (this_element
= compile_stack
.avail
- 1;
2989 if (compile_stack
.stack
[this_element
].regnum
== regnum
)
2996 /* Read the ending character of a range (in a bracket expression) from the
2997 uncompiled pattern *P_PTR (which ends at PEND). We assume the
2998 starting character is in `P[-2]'. (`P[-1]' is the character `-'.)
2999 Then we set the translation of all bits between the starting and
3000 ending characters (inclusive) in the compiled pattern B.
3002 Return an error code.
3004 We use these short variable names so we can use the same macros as
3005 `regex_compile' itself. */
3007 static reg_errcode_t
3008 compile_range (p_ptr
, pend
, translate
, syntax
, b
)
3009 const char **p_ptr
, *pend
;
3010 RE_TRANSLATE_TYPE translate
;
3011 reg_syntax_t syntax
;
3016 const char *p
= *p_ptr
;
3017 unsigned int range_start
, range_end
;
3022 /* Even though the pattern is a signed `char *', we need to fetch
3023 with unsigned char *'s; if the high bit of the pattern character
3024 is set, the range endpoints will be negative if we fetch using a
3027 We also want to fetch the endpoints without translating them; the
3028 appropriate translation is done in the bit-setting loop below. */
3029 /* The SVR4 compiler on the 3B2 had trouble with unsigned const char *. */
3030 range_start
= ((const unsigned char *) p
)[-2];
3031 range_end
= ((const unsigned char *) p
)[0];
3033 /* Have to increment the pointer into the pattern string, so the
3034 caller isn't still at the ending character. */
3037 /* If the start is after the end, the range is empty. */
3038 if (range_start
> range_end
)
3039 return syntax
& RE_NO_EMPTY_RANGES
? REG_ERANGE
: REG_NOERROR
;
3041 /* Here we see why `this_char' has to be larger than an `unsigned
3042 char' -- the range is inclusive, so if `range_end' == 0xff
3043 (assuming 8-bit characters), we would otherwise go into an infinite
3044 loop, since all characters <= 0xff. */
3045 for (this_char
= range_start
; this_char
<= range_end
; this_char
++)
3047 SET_LIST_BIT (TRANSLATE (this_char
));
3053 /* re_compile_fastmap computes a ``fastmap'' for the compiled pattern in
3054 BUFP. A fastmap records which of the (1 << BYTEWIDTH) possible
3055 characters can start a string that matches the pattern. This fastmap
3056 is used by re_search to skip quickly over impossible starting points.
3058 The caller must supply the address of a (1 << BYTEWIDTH)-byte data
3059 area as BUFP->fastmap.
3061 We set the `fastmap', `fastmap_accurate', and `can_be_null' fields in
3064 Returns 0 if we succeed, -2 if an internal error. */
3067 re_compile_fastmap (bufp
)
3068 struct re_pattern_buffer
*bufp
;
3071 #ifdef MATCH_MAY_ALLOCATE
3072 fail_stack_type fail_stack
;
3074 #ifndef REGEX_MALLOC
3078 register char *fastmap
= bufp
->fastmap
;
3079 unsigned char *pattern
= bufp
->buffer
;
3080 unsigned char *p
= pattern
;
3081 register unsigned char *pend
= pattern
+ bufp
->used
;
3084 /* This holds the pointer to the failure stack, when
3085 it is allocated relocatably. */
3086 fail_stack_elt_t
*failure_stack_ptr
;
3089 /* Assume that each path through the pattern can be null until
3090 proven otherwise. We set this false at the bottom of switch
3091 statement, to which we get only if a particular path doesn't
3092 match the empty string. */
3093 boolean path_can_be_null
= true;
3095 /* We aren't doing a `succeed_n' to begin with. */
3096 boolean succeed_n_p
= false;
3098 assert (fastmap
!= NULL
&& p
!= NULL
);
3101 bzero (fastmap
, 1 << BYTEWIDTH
); /* Assume nothing's valid. */
3102 bufp
->fastmap_accurate
= 1; /* It will be when we're done. */
3103 bufp
->can_be_null
= 0;
3107 if (p
== pend
|| *p
== succeed
)
3109 /* We have reached the (effective) end of pattern. */
3110 if (!FAIL_STACK_EMPTY ())
3112 bufp
->can_be_null
|= path_can_be_null
;
3114 /* Reset for next path. */
3115 path_can_be_null
= true;
3117 p
= fail_stack
.stack
[--fail_stack
.avail
].pointer
;
3125 /* We should never be about to go beyond the end of the pattern. */
3128 switch (SWITCH_ENUM_CAST ((re_opcode_t
) *p
++))
3131 /* I guess the idea here is to simply not bother with a fastmap
3132 if a backreference is used, since it's too hard to figure out
3133 the fastmap for the corresponding group. Setting
3134 `can_be_null' stops `re_search_2' from using the fastmap, so
3135 that is all we do. */
3137 bufp
->can_be_null
= 1;
3141 /* Following are the cases which match a character. These end
3150 for (j
= *p
++ * BYTEWIDTH
- 1; j
>= 0; j
--)
3151 if (p
[j
/ BYTEWIDTH
] & (1 << (j
% BYTEWIDTH
)))
3157 /* Chars beyond end of map must be allowed. */
3158 for (j
= *p
* BYTEWIDTH
; j
< (1 << BYTEWIDTH
); j
++)
3161 for (j
= *p
++ * BYTEWIDTH
- 1; j
>= 0; j
--)
3162 if (!(p
[j
/ BYTEWIDTH
] & (1 << (j
% BYTEWIDTH
))))
3168 for (j
= 0; j
< (1 << BYTEWIDTH
); j
++)
3169 if (SYNTAX (j
) == Sword
)
3175 for (j
= 0; j
< (1 << BYTEWIDTH
); j
++)
3176 if (SYNTAX (j
) != Sword
)
3183 int fastmap_newline
= fastmap
['\n'];
3185 /* `.' matches anything ... */
3186 for (j
= 0; j
< (1 << BYTEWIDTH
); j
++)
3189 /* ... except perhaps newline. */
3190 if (!(bufp
->syntax
& RE_DOT_NEWLINE
))
3191 fastmap
['\n'] = fastmap_newline
;
3193 /* Return if we have already set `can_be_null'; if we have,
3194 then the fastmap is irrelevant. Something's wrong here. */
3195 else if (bufp
->can_be_null
)
3198 /* Otherwise, have to check alternative paths. */
3205 for (j
= 0; j
< (1 << BYTEWIDTH
); j
++)
3206 if (SYNTAX (j
) == (enum syntaxcode
) k
)
3213 for (j
= 0; j
< (1 << BYTEWIDTH
); j
++)
3214 if (SYNTAX (j
) != (enum syntaxcode
) k
)
3219 /* All cases after this match the empty string. These end with
3239 case push_dummy_failure
:
3244 case pop_failure_jump
:
3245 case maybe_pop_jump
:
3248 case dummy_failure_jump
:
3249 EXTRACT_NUMBER_AND_INCR (j
, p
);
3254 /* Jump backward implies we just went through the body of a
3255 loop and matched nothing. Opcode jumped to should be
3256 `on_failure_jump' or `succeed_n'. Just treat it like an
3257 ordinary jump. For a * loop, it has pushed its failure
3258 point already; if so, discard that as redundant. */
3259 if ((re_opcode_t
) *p
!= on_failure_jump
3260 && (re_opcode_t
) *p
!= succeed_n
)
3264 EXTRACT_NUMBER_AND_INCR (j
, p
);
3267 /* If what's on the stack is where we are now, pop it. */
3268 if (!FAIL_STACK_EMPTY ()
3269 && fail_stack
.stack
[fail_stack
.avail
- 1].pointer
== p
)
3275 case on_failure_jump
:
3276 case on_failure_keep_string_jump
:
3277 handle_on_failure_jump
:
3278 EXTRACT_NUMBER_AND_INCR (j
, p
);
3280 /* For some patterns, e.g., `(a?)?', `p+j' here points to the
3281 end of the pattern. We don't want to push such a point,
3282 since when we restore it above, entering the switch will
3283 increment `p' past the end of the pattern. We don't need
3284 to push such a point since we obviously won't find any more
3285 fastmap entries beyond `pend'. Such a pattern can match
3286 the null string, though. */
3289 if (!PUSH_PATTERN_OP (p
+ j
, fail_stack
))
3291 RESET_FAIL_STACK ();
3296 bufp
->can_be_null
= 1;
3300 EXTRACT_NUMBER_AND_INCR (k
, p
); /* Skip the n. */
3301 succeed_n_p
= false;
3308 /* Get to the number of times to succeed. */
3311 /* Increment p past the n for when k != 0. */
3312 EXTRACT_NUMBER_AND_INCR (k
, p
);
3316 succeed_n_p
= true; /* Spaghetti code alert. */
3317 goto handle_on_failure_jump
;
3334 abort (); /* We have listed all the cases. */
3337 /* Getting here means we have found the possible starting
3338 characters for one path of the pattern -- and that the empty
3339 string does not match. We need not follow this path further.
3340 Instead, look at the next alternative (remembered on the
3341 stack), or quit if no more. The test at the top of the loop
3342 does these things. */
3343 path_can_be_null
= false;
3347 /* Set `can_be_null' for the last path (also the first path, if the
3348 pattern is empty). */
3349 bufp
->can_be_null
|= path_can_be_null
;
3352 RESET_FAIL_STACK ();
3354 } /* re_compile_fastmap */
3356 weak_alias (__re_compile_fastmap
, re_compile_fastmap
)
3359 /* Set REGS to hold NUM_REGS registers, storing them in STARTS and
3360 ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use
3361 this memory for recording register information. STARTS and ENDS
3362 must be allocated using the malloc library routine, and must each
3363 be at least NUM_REGS * sizeof (regoff_t) bytes long.
3365 If NUM_REGS == 0, then subsequent matches should allocate their own
3368 Unless this function is called, the first search or match using
3369 PATTERN_BUFFER will allocate its own register data, without
3370 freeing the old data. */
3373 re_set_registers (bufp
, regs
, num_regs
, starts
, ends
)
3374 struct re_pattern_buffer
*bufp
;
3375 struct re_registers
*regs
;
3377 regoff_t
*starts
, *ends
;
3381 bufp
->regs_allocated
= REGS_REALLOCATE
;
3382 regs
->num_regs
= num_regs
;
3383 regs
->start
= starts
;
3388 bufp
->regs_allocated
= REGS_UNALLOCATED
;
3390 regs
->start
= regs
->end
= (regoff_t
*) 0;
3394 weak_alias (__re_set_registers
, re_set_registers
)
3397 /* Searching routines. */
3399 /* Like re_search_2, below, but only one string is specified, and
3400 doesn't let you say where to stop matching. */
3403 re_search (bufp
, string
, size
, startpos
, range
, regs
)
3404 struct re_pattern_buffer
*bufp
;
3406 int size
, startpos
, range
;
3407 struct re_registers
*regs
;
3409 return re_search_2 (bufp
, NULL
, 0, string
, size
, startpos
, range
,
3413 weak_alias (__re_search
, re_search
)
3417 /* Using the compiled pattern in BUFP->buffer, first tries to match the
3418 virtual concatenation of STRING1 and STRING2, starting first at index
3419 STARTPOS, then at STARTPOS + 1, and so on.
3421 STRING1 and STRING2 have length SIZE1 and SIZE2, respectively.
3423 RANGE is how far to scan while trying to match. RANGE = 0 means try
3424 only at STARTPOS; in general, the last start tried is STARTPOS +
3427 In REGS, return the indices of the virtual concatenation of STRING1
3428 and STRING2 that matched the entire BUFP->buffer and its contained
3431 Do not consider matching one past the index STOP in the virtual
3432 concatenation of STRING1 and STRING2.
3434 We return either the position in the strings at which the match was
3435 found, -1 if no match, or -2 if error (such as failure
3439 re_search_2 (bufp
, string1
, size1
, string2
, size2
, startpos
, range
, regs
, stop
)
3440 struct re_pattern_buffer
*bufp
;
3441 const char *string1
, *string2
;
3445 struct re_registers
*regs
;
3449 register char *fastmap
= bufp
->fastmap
;
3450 register RE_TRANSLATE_TYPE translate
= bufp
->translate
;
3451 int total_size
= size1
+ size2
;
3452 int endpos
= startpos
+ range
;
3454 /* Check for out-of-range STARTPOS. */
3455 if (startpos
< 0 || startpos
> total_size
)
3458 /* Fix up RANGE if it might eventually take us outside
3459 the virtual concatenation of STRING1 and STRING2.
3460 Make sure we won't move STARTPOS below 0 or above TOTAL_SIZE. */
3462 range
= 0 - startpos
;
3463 else if (endpos
> total_size
)
3464 range
= total_size
- startpos
;
3466 /* If the search isn't to be a backwards one, don't waste time in a
3467 search for a pattern that must be anchored. */
3468 if (bufp
->used
> 0 && (re_opcode_t
) bufp
->buffer
[0] == begbuf
&& range
> 0)
3477 /* In a forward search for something that starts with \=.
3478 don't keep searching past point. */
3479 if (bufp
->used
> 0 && (re_opcode_t
) bufp
->buffer
[0] == at_dot
&& range
> 0)
3481 range
= PT
- startpos
;
3487 /* Update the fastmap now if not correct already. */
3488 if (fastmap
&& !bufp
->fastmap_accurate
)
3489 if (re_compile_fastmap (bufp
) == -2)
3492 /* Loop through the string, looking for a place to start matching. */
3495 /* If a fastmap is supplied, skip quickly over characters that
3496 cannot be the start of a match. If the pattern can match the
3497 null string, however, we don't need to skip characters; we want
3498 the first null string. */
3499 if (fastmap
&& startpos
< total_size
&& !bufp
->can_be_null
)
3501 if (range
> 0) /* Searching forwards. */
3503 register const char *d
;
3504 register int lim
= 0;
3507 if (startpos
< size1
&& startpos
+ range
>= size1
)
3508 lim
= range
- (size1
- startpos
);
3510 d
= (startpos
>= size1
? string2
- size1
: string1
) + startpos
;
3512 /* Written out as an if-else to avoid testing `translate'
3516 && !fastmap
[(unsigned char)
3517 translate
[(unsigned char) *d
++]])
3520 while (range
> lim
&& !fastmap
[(unsigned char) *d
++])
3523 startpos
+= irange
- range
;
3525 else /* Searching backwards. */
3527 register char c
= (size1
== 0 || startpos
>= size1
3528 ? string2
[startpos
- size1
]
3529 : string1
[startpos
]);
3531 if (!fastmap
[(unsigned char) TRANSLATE (c
)])
3536 /* If can't match the null string, and that's all we have left, fail. */
3537 if (range
>= 0 && startpos
== total_size
&& fastmap
3538 && !bufp
->can_be_null
)
3541 val
= re_match_2_internal (bufp
, string1
, size1
, string2
, size2
,
3542 startpos
, regs
, stop
);
3543 #ifndef REGEX_MALLOC
3572 weak_alias (__re_search_2
, re_search_2
)
3575 /* This converts PTR, a pointer into one of the search strings `string1'
3576 and `string2' into an offset from the beginning of that string. */
3577 #define POINTER_TO_OFFSET(ptr) \
3578 (FIRST_STRING_P (ptr) \
3579 ? ((regoff_t) ((ptr) - string1)) \
3580 : ((regoff_t) ((ptr) - string2 + size1)))
3582 /* Macros for dealing with the split strings in re_match_2. */
3584 #define MATCHING_IN_FIRST_STRING (dend == end_match_1)
3586 /* Call before fetching a character with *d. This switches over to
3587 string2 if necessary. */
3588 #define PREFETCH() \
3591 /* End of string2 => fail. */ \
3592 if (dend == end_match_2) \
3594 /* End of string1 => advance to string2. */ \
3596 dend = end_match_2; \
3600 /* Test if at very beginning or at very end of the virtual concatenation
3601 of `string1' and `string2'. If only one string, it's `string2'. */
3602 #define AT_STRINGS_BEG(d) ((d) == (size1 ? string1 : string2) || !size2)
3603 #define AT_STRINGS_END(d) ((d) == end2)
3606 /* Test if D points to a character which is word-constituent. We have
3607 two special cases to check for: if past the end of string1, look at
3608 the first character in string2; and if before the beginning of
3609 string2, look at the last character in string1. */
3610 #define WORDCHAR_P(d) \
3611 (SYNTAX ((d) == end1 ? *string2 \
3612 : (d) == string2 - 1 ? *(end1 - 1) : *(d)) \
3615 /* Disabled due to a compiler bug -- see comment at case wordbound */
3617 /* Test if the character before D and the one at D differ with respect
3618 to being word-constituent. */
3619 #define AT_WORD_BOUNDARY(d) \
3620 (AT_STRINGS_BEG (d) || AT_STRINGS_END (d) \
3621 || WORDCHAR_P (d - 1) != WORDCHAR_P (d))
3624 /* Free everything we malloc. */
3625 #ifdef MATCH_MAY_ALLOCATE
3626 # define FREE_VAR(var) if (var) REGEX_FREE (var); var = NULL
3627 # define FREE_VARIABLES() \
3629 REGEX_FREE_STACK (fail_stack.stack); \
3630 FREE_VAR (regstart); \
3631 FREE_VAR (regend); \
3632 FREE_VAR (old_regstart); \
3633 FREE_VAR (old_regend); \
3634 FREE_VAR (best_regstart); \
3635 FREE_VAR (best_regend); \
3636 FREE_VAR (reg_info); \
3637 FREE_VAR (reg_dummy); \
3638 FREE_VAR (reg_info_dummy); \
3641 # define FREE_VARIABLES() ((void)0) /* Do nothing! But inhibit gcc warning. */
3642 #endif /* not MATCH_MAY_ALLOCATE */
3644 /* These values must meet several constraints. They must not be valid
3645 register values; since we have a limit of 255 registers (because
3646 we use only one byte in the pattern for the register number), we can
3647 use numbers larger than 255. They must differ by 1, because of
3648 NUM_FAILURE_ITEMS above. And the value for the lowest register must
3649 be larger than the value for the highest register, so we do not try
3650 to actually save any registers when none are active. */
3651 #define NO_HIGHEST_ACTIVE_REG (1 << BYTEWIDTH)
3652 #define NO_LOWEST_ACTIVE_REG (NO_HIGHEST_ACTIVE_REG + 1)
3654 /* Matching routines. */
3656 #ifndef emacs /* Emacs never uses this. */
3657 /* re_match is like re_match_2 except it takes only a single string. */
3660 re_match (bufp
, string
, size
, pos
, regs
)
3661 struct re_pattern_buffer
*bufp
;
3664 struct re_registers
*regs
;
3666 int result
= re_match_2_internal (bufp
, NULL
, 0, string
, size
,
3668 # ifndef REGEX_MALLOC
3676 weak_alias (__re_match
, re_match
)
3678 #endif /* not emacs */
3680 static boolean group_match_null_string_p
_RE_ARGS ((unsigned char **p
,
3682 register_info_type
*reg_info
));
3683 static boolean alt_match_null_string_p
_RE_ARGS ((unsigned char *p
,
3685 register_info_type
*reg_info
));
3686 static boolean common_op_match_null_string_p
_RE_ARGS ((unsigned char **p
,
3688 register_info_type
*reg_info
));
3689 static int bcmp_translate
_RE_ARGS ((const char *s1
, const char *s2
,
3690 int len
, char *translate
));
3692 /* re_match_2 matches the compiled pattern in BUFP against the
3693 the (virtual) concatenation of STRING1 and STRING2 (of length SIZE1
3694 and SIZE2, respectively). We start matching at POS, and stop
3697 If REGS is non-null and the `no_sub' field of BUFP is nonzero, we
3698 store offsets for the substring each group matched in REGS. See the
3699 documentation for exactly how many groups we fill.
3701 We return -1 if no match, -2 if an internal error (such as the
3702 failure stack overflowing). Otherwise, we return the length of the
3703 matched substring. */
3706 re_match_2 (bufp
, string1
, size1
, string2
, size2
, pos
, regs
, stop
)
3707 struct re_pattern_buffer
*bufp
;
3708 const char *string1
, *string2
;
3711 struct re_registers
*regs
;
3714 int result
= re_match_2_internal (bufp
, string1
, size1
, string2
, size2
,
3716 #ifndef REGEX_MALLOC
3724 weak_alias (__re_match_2
, re_match_2
)
3727 /* This is a separate function so that we can force an alloca cleanup
3730 re_match_2_internal (bufp
, string1
, size1
, string2
, size2
, pos
, regs
, stop
)
3731 struct re_pattern_buffer
*bufp
;
3732 const char *string1
, *string2
;
3735 struct re_registers
*regs
;
3738 /* General temporaries. */
3742 /* Just past the end of the corresponding string. */
3743 const char *end1
, *end2
;
3745 /* Pointers into string1 and string2, just past the last characters in
3746 each to consider matching. */
3747 const char *end_match_1
, *end_match_2
;
3749 /* Where we are in the data, and the end of the current string. */
3750 const char *d
, *dend
;
3752 /* Where we are in the pattern, and the end of the pattern. */
3753 unsigned char *p
= bufp
->buffer
;
3754 register unsigned char *pend
= p
+ bufp
->used
;
3756 /* Mark the opcode just after a start_memory, so we can test for an
3757 empty subpattern when we get to the stop_memory. */
3758 unsigned char *just_past_start_mem
= 0;
3760 /* We use this to map every character in the string. */
3761 RE_TRANSLATE_TYPE translate
= bufp
->translate
;
3763 /* Failure point stack. Each place that can handle a failure further
3764 down the line pushes a failure point on this stack. It consists of
3765 restart, regend, and reg_info for all registers corresponding to
3766 the subexpressions we're currently inside, plus the number of such
3767 registers, and, finally, two char *'s. The first char * is where
3768 to resume scanning the pattern; the second one is where to resume
3769 scanning the strings. If the latter is zero, the failure point is
3770 a ``dummy''; if a failure happens and the failure point is a dummy,
3771 it gets discarded and the next next one is tried. */
3772 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
3773 fail_stack_type fail_stack
;
3776 static unsigned failure_id
= 0;
3777 unsigned nfailure_points_pushed
= 0, nfailure_points_popped
= 0;
3781 /* This holds the pointer to the failure stack, when
3782 it is allocated relocatably. */
3783 fail_stack_elt_t
*failure_stack_ptr
;
3786 /* We fill all the registers internally, independent of what we
3787 return, for use in backreferences. The number here includes
3788 an element for register zero. */
3789 size_t num_regs
= bufp
->re_nsub
+ 1;
3791 /* The currently active registers. */
3792 active_reg_t lowest_active_reg
= NO_LOWEST_ACTIVE_REG
;
3793 active_reg_t highest_active_reg
= NO_HIGHEST_ACTIVE_REG
;
3795 /* Information on the contents of registers. These are pointers into
3796 the input strings; they record just what was matched (on this
3797 attempt) by a subexpression part of the pattern, that is, the
3798 regnum-th regstart pointer points to where in the pattern we began
3799 matching and the regnum-th regend points to right after where we
3800 stopped matching the regnum-th subexpression. (The zeroth register
3801 keeps track of what the whole pattern matches.) */
3802 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
3803 const char **regstart
, **regend
;
3806 /* If a group that's operated upon by a repetition operator fails to
3807 match anything, then the register for its start will need to be
3808 restored because it will have been set to wherever in the string we
3809 are when we last see its open-group operator. Similarly for a
3811 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
3812 const char **old_regstart
, **old_regend
;
3815 /* The is_active field of reg_info helps us keep track of which (possibly
3816 nested) subexpressions we are currently in. The matched_something
3817 field of reg_info[reg_num] helps us tell whether or not we have
3818 matched any of the pattern so far this time through the reg_num-th
3819 subexpression. These two fields get reset each time through any
3820 loop their register is in. */
3821 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
3822 register_info_type
*reg_info
;
3825 /* The following record the register info as found in the above
3826 variables when we find a match better than any we've seen before.
3827 This happens as we backtrack through the failure points, which in
3828 turn happens only if we have not yet matched the entire string. */
3829 unsigned best_regs_set
= false;
3830 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
3831 const char **best_regstart
, **best_regend
;
3834 /* Logically, this is `best_regend[0]'. But we don't want to have to
3835 allocate space for that if we're not allocating space for anything
3836 else (see below). Also, we never need info about register 0 for
3837 any of the other register vectors, and it seems rather a kludge to
3838 treat `best_regend' differently than the rest. So we keep track of
3839 the end of the best match so far in a separate variable. We
3840 initialize this to NULL so that when we backtrack the first time
3841 and need to test it, it's not garbage. */
3842 const char *match_end
= NULL
;
3844 /* This helps SET_REGS_MATCHED avoid doing redundant work. */
3845 int set_regs_matched_done
= 0;
3847 /* Used when we pop values we don't care about. */
3848 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
3849 const char **reg_dummy
;
3850 register_info_type
*reg_info_dummy
;
3854 /* Counts the total number of registers pushed. */
3855 unsigned num_regs_pushed
= 0;
3858 DEBUG_PRINT1 ("\n\nEntering re_match_2.\n");
3862 #ifdef MATCH_MAY_ALLOCATE
3863 /* Do not bother to initialize all the register variables if there are
3864 no groups in the pattern, as it takes a fair amount of time. If
3865 there are groups, we include space for register 0 (the whole
3866 pattern), even though we never use it, since it simplifies the
3867 array indexing. We should fix this. */
3870 regstart
= REGEX_TALLOC (num_regs
, const char *);
3871 regend
= REGEX_TALLOC (num_regs
, const char *);
3872 old_regstart
= REGEX_TALLOC (num_regs
, const char *);
3873 old_regend
= REGEX_TALLOC (num_regs
, const char *);
3874 best_regstart
= REGEX_TALLOC (num_regs
, const char *);
3875 best_regend
= REGEX_TALLOC (num_regs
, const char *);
3876 reg_info
= REGEX_TALLOC (num_regs
, register_info_type
);
3877 reg_dummy
= REGEX_TALLOC (num_regs
, const char *);
3878 reg_info_dummy
= REGEX_TALLOC (num_regs
, register_info_type
);
3880 if (!(regstart
&& regend
&& old_regstart
&& old_regend
&& reg_info
3881 && best_regstart
&& best_regend
&& reg_dummy
&& reg_info_dummy
))
3889 /* We must initialize all our variables to NULL, so that
3890 `FREE_VARIABLES' doesn't try to free them. */
3891 regstart
= regend
= old_regstart
= old_regend
= best_regstart
3892 = best_regend
= reg_dummy
= NULL
;
3893 reg_info
= reg_info_dummy
= (register_info_type
*) NULL
;
3895 #endif /* MATCH_MAY_ALLOCATE */
3897 /* The starting position is bogus. */
3898 if (pos
< 0 || pos
> size1
+ size2
)
3904 /* Initialize subexpression text positions to -1 to mark ones that no
3905 start_memory/stop_memory has been seen for. Also initialize the
3906 register information struct. */
3907 for (mcnt
= 1; (unsigned) mcnt
< num_regs
; mcnt
++)
3909 regstart
[mcnt
] = regend
[mcnt
]
3910 = old_regstart
[mcnt
] = old_regend
[mcnt
] = REG_UNSET_VALUE
;
3912 REG_MATCH_NULL_STRING_P (reg_info
[mcnt
]) = MATCH_NULL_UNSET_VALUE
;
3913 IS_ACTIVE (reg_info
[mcnt
]) = 0;
3914 MATCHED_SOMETHING (reg_info
[mcnt
]) = 0;
3915 EVER_MATCHED_SOMETHING (reg_info
[mcnt
]) = 0;
3918 /* We move `string1' into `string2' if the latter's empty -- but not if
3919 `string1' is null. */
3920 if (size2
== 0 && string1
!= NULL
)
3927 end1
= string1
+ size1
;
3928 end2
= string2
+ size2
;
3930 /* Compute where to stop matching, within the two strings. */
3933 end_match_1
= string1
+ stop
;
3934 end_match_2
= string2
;
3939 end_match_2
= string2
+ stop
- size1
;
3942 /* `p' scans through the pattern as `d' scans through the data.
3943 `dend' is the end of the input string that `d' points within. `d'
3944 is advanced into the following input string whenever necessary, but
3945 this happens before fetching; therefore, at the beginning of the
3946 loop, `d' can be pointing at the end of a string, but it cannot
3948 if (size1
> 0 && pos
<= size1
)
3955 d
= string2
+ pos
- size1
;
3959 DEBUG_PRINT1 ("The compiled pattern is:\n");
3960 DEBUG_PRINT_COMPILED_PATTERN (bufp
, p
, pend
);
3961 DEBUG_PRINT1 ("The string to match is: `");
3962 DEBUG_PRINT_DOUBLE_STRING (d
, string1
, size1
, string2
, size2
);
3963 DEBUG_PRINT1 ("'\n");
3965 /* This loops over pattern commands. It exits by returning from the
3966 function if the match is complete, or it drops through if the match
3967 fails at this starting point in the input data. */
3971 DEBUG_PRINT2 ("\n%p: ", p
);
3973 DEBUG_PRINT2 ("\n0x%x: ", p
);
3977 { /* End of pattern means we might have succeeded. */
3978 DEBUG_PRINT1 ("end of pattern ... ");
3980 /* If we haven't matched the entire string, and we want the
3981 longest match, try backtracking. */
3982 if (d
!= end_match_2
)
3984 /* 1 if this match ends in the same string (string1 or string2)
3985 as the best previous match. */
3986 boolean same_str_p
= (FIRST_STRING_P (match_end
)
3987 == MATCHING_IN_FIRST_STRING
);
3988 /* 1 if this match is the best seen so far. */
3989 boolean best_match_p
;
3991 /* AIX compiler got confused when this was combined
3992 with the previous declaration. */
3994 best_match_p
= d
> match_end
;
3996 best_match_p
= !MATCHING_IN_FIRST_STRING
;
3998 DEBUG_PRINT1 ("backtracking.\n");
4000 if (!FAIL_STACK_EMPTY ())
4001 { /* More failure points to try. */
4003 /* If exceeds best match so far, save it. */
4004 if (!best_regs_set
|| best_match_p
)
4006 best_regs_set
= true;
4009 DEBUG_PRINT1 ("\nSAVING match as best so far.\n");
4011 for (mcnt
= 1; (unsigned) mcnt
< num_regs
; mcnt
++)
4013 best_regstart
[mcnt
] = regstart
[mcnt
];
4014 best_regend
[mcnt
] = regend
[mcnt
];
4020 /* If no failure points, don't restore garbage. And if
4021 last match is real best match, don't restore second
4023 else if (best_regs_set
&& !best_match_p
)
4026 /* Restore best match. It may happen that `dend ==
4027 end_match_1' while the restored d is in string2.
4028 For example, the pattern `x.*y.*z' against the
4029 strings `x-' and `y-z-', if the two strings are
4030 not consecutive in memory. */
4031 DEBUG_PRINT1 ("Restoring best registers.\n");
4034 dend
= ((d
>= string1
&& d
<= end1
)
4035 ? end_match_1
: end_match_2
);
4037 for (mcnt
= 1; (unsigned) mcnt
< num_regs
; mcnt
++)
4039 regstart
[mcnt
] = best_regstart
[mcnt
];
4040 regend
[mcnt
] = best_regend
[mcnt
];
4043 } /* d != end_match_2 */
4046 DEBUG_PRINT1 ("Accepting match.\n");
4048 /* If caller wants register contents data back, do it. */
4049 if (regs
&& !bufp
->no_sub
)
4051 /* Have the register data arrays been allocated? */
4052 if (bufp
->regs_allocated
== REGS_UNALLOCATED
)
4053 { /* No. So allocate them with malloc. We need one
4054 extra element beyond `num_regs' for the `-1' marker
4056 regs
->num_regs
= MAX (RE_NREGS
, num_regs
+ 1);
4057 regs
->start
= TALLOC (regs
->num_regs
, regoff_t
);
4058 regs
->end
= TALLOC (regs
->num_regs
, regoff_t
);
4059 if (regs
->start
== NULL
|| regs
->end
== NULL
)
4064 bufp
->regs_allocated
= REGS_REALLOCATE
;
4066 else if (bufp
->regs_allocated
== REGS_REALLOCATE
)
4067 { /* Yes. If we need more elements than were already
4068 allocated, reallocate them. If we need fewer, just
4070 if (regs
->num_regs
< num_regs
+ 1)
4072 regs
->num_regs
= num_regs
+ 1;
4073 RETALLOC (regs
->start
, regs
->num_regs
, regoff_t
);
4074 RETALLOC (regs
->end
, regs
->num_regs
, regoff_t
);
4075 if (regs
->start
== NULL
|| regs
->end
== NULL
)
4084 /* These braces fend off a "empty body in an else-statement"
4085 warning under GCC when assert expands to nothing. */
4086 assert (bufp
->regs_allocated
== REGS_FIXED
);
4089 /* Convert the pointer data in `regstart' and `regend' to
4090 indices. Register zero has to be set differently,
4091 since we haven't kept track of any info for it. */
4092 if (regs
->num_regs
> 0)
4094 regs
->start
[0] = pos
;
4095 regs
->end
[0] = (MATCHING_IN_FIRST_STRING
4096 ? ((regoff_t
) (d
- string1
))
4097 : ((regoff_t
) (d
- string2
+ size1
)));
4100 /* Go through the first `min (num_regs, regs->num_regs)'
4101 registers, since that is all we initialized. */
4102 for (mcnt
= 1; (unsigned) mcnt
< MIN (num_regs
, regs
->num_regs
);
4105 if (REG_UNSET (regstart
[mcnt
]) || REG_UNSET (regend
[mcnt
]))
4106 regs
->start
[mcnt
] = regs
->end
[mcnt
] = -1;
4110 = (regoff_t
) POINTER_TO_OFFSET (regstart
[mcnt
]);
4112 = (regoff_t
) POINTER_TO_OFFSET (regend
[mcnt
]);
4116 /* If the regs structure we return has more elements than
4117 were in the pattern, set the extra elements to -1. If
4118 we (re)allocated the registers, this is the case,
4119 because we always allocate enough to have at least one
4121 for (mcnt
= num_regs
; (unsigned) mcnt
< regs
->num_regs
; mcnt
++)
4122 regs
->start
[mcnt
] = regs
->end
[mcnt
] = -1;
4123 } /* regs && !bufp->no_sub */
4125 DEBUG_PRINT4 ("%u failure points pushed, %u popped (%u remain).\n",
4126 nfailure_points_pushed
, nfailure_points_popped
,
4127 nfailure_points_pushed
- nfailure_points_popped
);
4128 DEBUG_PRINT2 ("%u registers pushed.\n", num_regs_pushed
);
4130 mcnt
= d
- pos
- (MATCHING_IN_FIRST_STRING
4134 DEBUG_PRINT2 ("Returning %d from re_match_2.\n", mcnt
);
4140 /* Otherwise match next pattern command. */
4141 switch (SWITCH_ENUM_CAST ((re_opcode_t
) *p
++))
4143 /* Ignore these. Used to ignore the n of succeed_n's which
4144 currently have n == 0. */
4146 DEBUG_PRINT1 ("EXECUTING no_op.\n");
4150 DEBUG_PRINT1 ("EXECUTING succeed.\n");
4153 /* Match the next n pattern characters exactly. The following
4154 byte in the pattern defines n, and the n bytes after that
4155 are the characters to match. */
4158 DEBUG_PRINT2 ("EXECUTING exactn %d.\n", mcnt
);
4160 /* This is written out as an if-else so we don't waste time
4161 testing `translate' inside the loop. */
4167 if ((unsigned char) translate
[(unsigned char) *d
++]
4168 != (unsigned char) *p
++)
4178 if (*d
++ != (char) *p
++) goto fail
;
4182 SET_REGS_MATCHED ();
4186 /* Match any character except possibly a newline or a null. */
4188 DEBUG_PRINT1 ("EXECUTING anychar.\n");
4192 if ((!(bufp
->syntax
& RE_DOT_NEWLINE
) && TRANSLATE (*d
) == '\n')
4193 || (bufp
->syntax
& RE_DOT_NOT_NULL
&& TRANSLATE (*d
) == '\000'))
4196 SET_REGS_MATCHED ();
4197 DEBUG_PRINT2 (" Matched `%d'.\n", *d
);
4205 register unsigned char c
;
4206 boolean
not = (re_opcode_t
) *(p
- 1) == charset_not
;
4208 DEBUG_PRINT2 ("EXECUTING charset%s.\n", not ? "_not" : "");
4211 c
= TRANSLATE (*d
); /* The character to match. */
4213 /* Cast to `unsigned' instead of `unsigned char' in case the
4214 bit list is a full 32 bytes long. */
4215 if (c
< (unsigned) (*p
* BYTEWIDTH
)
4216 && p
[1 + c
/ BYTEWIDTH
] & (1 << (c
% BYTEWIDTH
)))
4221 if (!not) goto fail
;
4223 SET_REGS_MATCHED ();
4229 /* The beginning of a group is represented by start_memory.
4230 The arguments are the register number in the next byte, and the
4231 number of groups inner to this one in the next. The text
4232 matched within the group is recorded (in the internal
4233 registers data structure) under the register number. */
4235 DEBUG_PRINT3 ("EXECUTING start_memory %d (%d):\n", *p
, p
[1]);
4237 /* Find out if this group can match the empty string. */
4238 p1
= p
; /* To send to group_match_null_string_p. */
4240 if (REG_MATCH_NULL_STRING_P (reg_info
[*p
]) == MATCH_NULL_UNSET_VALUE
)
4241 REG_MATCH_NULL_STRING_P (reg_info
[*p
])
4242 = group_match_null_string_p (&p1
, pend
, reg_info
);
4244 /* Save the position in the string where we were the last time
4245 we were at this open-group operator in case the group is
4246 operated upon by a repetition operator, e.g., with `(a*)*b'
4247 against `ab'; then we want to ignore where we are now in
4248 the string in case this attempt to match fails. */
4249 old_regstart
[*p
] = REG_MATCH_NULL_STRING_P (reg_info
[*p
])
4250 ? REG_UNSET (regstart
[*p
]) ? d
: regstart
[*p
]
4252 DEBUG_PRINT2 (" old_regstart: %d\n",
4253 POINTER_TO_OFFSET (old_regstart
[*p
]));
4256 DEBUG_PRINT2 (" regstart: %d\n", POINTER_TO_OFFSET (regstart
[*p
]));
4258 IS_ACTIVE (reg_info
[*p
]) = 1;
4259 MATCHED_SOMETHING (reg_info
[*p
]) = 0;
4261 /* Clear this whenever we change the register activity status. */
4262 set_regs_matched_done
= 0;
4264 /* This is the new highest active register. */
4265 highest_active_reg
= *p
;
4267 /* If nothing was active before, this is the new lowest active
4269 if (lowest_active_reg
== NO_LOWEST_ACTIVE_REG
)
4270 lowest_active_reg
= *p
;
4272 /* Move past the register number and inner group count. */
4274 just_past_start_mem
= p
;
4279 /* The stop_memory opcode represents the end of a group. Its
4280 arguments are the same as start_memory's: the register
4281 number, and the number of inner groups. */
4283 DEBUG_PRINT3 ("EXECUTING stop_memory %d (%d):\n", *p
, p
[1]);
4285 /* We need to save the string position the last time we were at
4286 this close-group operator in case the group is operated
4287 upon by a repetition operator, e.g., with `((a*)*(b*)*)*'
4288 against `aba'; then we want to ignore where we are now in
4289 the string in case this attempt to match fails. */
4290 old_regend
[*p
] = REG_MATCH_NULL_STRING_P (reg_info
[*p
])
4291 ? REG_UNSET (regend
[*p
]) ? d
: regend
[*p
]
4293 DEBUG_PRINT2 (" old_regend: %d\n",
4294 POINTER_TO_OFFSET (old_regend
[*p
]));
4297 DEBUG_PRINT2 (" regend: %d\n", POINTER_TO_OFFSET (regend
[*p
]));
4299 /* This register isn't active anymore. */
4300 IS_ACTIVE (reg_info
[*p
]) = 0;
4302 /* Clear this whenever we change the register activity status. */
4303 set_regs_matched_done
= 0;
4305 /* If this was the only register active, nothing is active
4307 if (lowest_active_reg
== highest_active_reg
)
4309 lowest_active_reg
= NO_LOWEST_ACTIVE_REG
;
4310 highest_active_reg
= NO_HIGHEST_ACTIVE_REG
;
4313 { /* We must scan for the new highest active register, since
4314 it isn't necessarily one less than now: consider
4315 (a(b)c(d(e)f)g). When group 3 ends, after the f), the
4316 new highest active register is 1. */
4317 unsigned char r
= *p
- 1;
4318 while (r
> 0 && !IS_ACTIVE (reg_info
[r
]))
4321 /* If we end up at register zero, that means that we saved
4322 the registers as the result of an `on_failure_jump', not
4323 a `start_memory', and we jumped to past the innermost
4324 `stop_memory'. For example, in ((.)*) we save
4325 registers 1 and 2 as a result of the *, but when we pop
4326 back to the second ), we are at the stop_memory 1.
4327 Thus, nothing is active. */
4330 lowest_active_reg
= NO_LOWEST_ACTIVE_REG
;
4331 highest_active_reg
= NO_HIGHEST_ACTIVE_REG
;
4334 highest_active_reg
= r
;
4337 /* If just failed to match something this time around with a
4338 group that's operated on by a repetition operator, try to
4339 force exit from the ``loop'', and restore the register
4340 information for this group that we had before trying this
4342 if ((!MATCHED_SOMETHING (reg_info
[*p
])
4343 || just_past_start_mem
== p
- 1)
4346 boolean is_a_jump_n
= false;
4350 switch ((re_opcode_t
) *p1
++)
4354 case pop_failure_jump
:
4355 case maybe_pop_jump
:
4357 case dummy_failure_jump
:
4358 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
4368 /* If the next operation is a jump backwards in the pattern
4369 to an on_failure_jump right before the start_memory
4370 corresponding to this stop_memory, exit from the loop
4371 by forcing a failure after pushing on the stack the
4372 on_failure_jump's jump in the pattern, and d. */
4373 if (mcnt
< 0 && (re_opcode_t
) *p1
== on_failure_jump
4374 && (re_opcode_t
) p1
[3] == start_memory
&& p1
[4] == *p
)
4376 /* If this group ever matched anything, then restore
4377 what its registers were before trying this last
4378 failed match, e.g., with `(a*)*b' against `ab' for
4379 regstart[1], and, e.g., with `((a*)*(b*)*)*'
4380 against `aba' for regend[3].
4382 Also restore the registers for inner groups for,
4383 e.g., `((a*)(b*))*' against `aba' (register 3 would
4384 otherwise get trashed). */
4386 if (EVER_MATCHED_SOMETHING (reg_info
[*p
]))
4390 EVER_MATCHED_SOMETHING (reg_info
[*p
]) = 0;
4392 /* Restore this and inner groups' (if any) registers. */
4393 for (r
= *p
; r
< (unsigned) *p
+ (unsigned) *(p
+ 1);
4396 regstart
[r
] = old_regstart
[r
];
4398 /* xx why this test? */
4399 if (old_regend
[r
] >= regstart
[r
])
4400 regend
[r
] = old_regend
[r
];
4404 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
4405 PUSH_FAILURE_POINT (p1
+ mcnt
, d
, -2);
4411 /* Move past the register number and the inner group count. */
4416 /* \<digit> has been turned into a `duplicate' command which is
4417 followed by the numeric value of <digit> as the register number. */
4420 register const char *d2
, *dend2
;
4421 int regno
= *p
++; /* Get which register to match against. */
4422 DEBUG_PRINT2 ("EXECUTING duplicate %d.\n", regno
);
4424 /* Can't back reference a group which we've never matched. */
4425 if (REG_UNSET (regstart
[regno
]) || REG_UNSET (regend
[regno
]))
4428 /* Where in input to try to start matching. */
4429 d2
= regstart
[regno
];
4431 /* Where to stop matching; if both the place to start and
4432 the place to stop matching are in the same string, then
4433 set to the place to stop, otherwise, for now have to use
4434 the end of the first string. */
4436 dend2
= ((FIRST_STRING_P (regstart
[regno
])
4437 == FIRST_STRING_P (regend
[regno
]))
4438 ? regend
[regno
] : end_match_1
);
4441 /* If necessary, advance to next segment in register
4445 if (dend2
== end_match_2
) break;
4446 if (dend2
== regend
[regno
]) break;
4448 /* End of string1 => advance to string2. */
4450 dend2
= regend
[regno
];
4452 /* At end of register contents => success */
4453 if (d2
== dend2
) break;
4455 /* If necessary, advance to next segment in data. */
4458 /* How many characters left in this segment to match. */
4461 /* Want how many consecutive characters we can match in
4462 one shot, so, if necessary, adjust the count. */
4463 if (mcnt
> dend2
- d2
)
4466 /* Compare that many; failure if mismatch, else move
4469 ? bcmp_translate (d
, d2
, mcnt
, translate
)
4470 : memcmp (d
, d2
, mcnt
))
4472 d
+= mcnt
, d2
+= mcnt
;
4474 /* Do this because we've match some characters. */
4475 SET_REGS_MATCHED ();
4481 /* begline matches the empty string at the beginning of the string
4482 (unless `not_bol' is set in `bufp'), and, if
4483 `newline_anchor' is set, after newlines. */
4485 DEBUG_PRINT1 ("EXECUTING begline.\n");
4487 if (AT_STRINGS_BEG (d
))
4489 if (!bufp
->not_bol
) break;
4491 else if (d
[-1] == '\n' && bufp
->newline_anchor
)
4495 /* In all other cases, we fail. */
4499 /* endline is the dual of begline. */
4501 DEBUG_PRINT1 ("EXECUTING endline.\n");
4503 if (AT_STRINGS_END (d
))
4505 if (!bufp
->not_eol
) break;
4508 /* We have to ``prefetch'' the next character. */
4509 else if ((d
== end1
? *string2
: *d
) == '\n'
4510 && bufp
->newline_anchor
)
4517 /* Match at the very beginning of the data. */
4519 DEBUG_PRINT1 ("EXECUTING begbuf.\n");
4520 if (AT_STRINGS_BEG (d
))
4525 /* Match at the very end of the data. */
4527 DEBUG_PRINT1 ("EXECUTING endbuf.\n");
4528 if (AT_STRINGS_END (d
))
4533 /* on_failure_keep_string_jump is used to optimize `.*\n'. It
4534 pushes NULL as the value for the string on the stack. Then
4535 `pop_failure_point' will keep the current value for the
4536 string, instead of restoring it. To see why, consider
4537 matching `foo\nbar' against `.*\n'. The .* matches the foo;
4538 then the . fails against the \n. But the next thing we want
4539 to do is match the \n against the \n; if we restored the
4540 string value, we would be back at the foo.
4542 Because this is used only in specific cases, we don't need to
4543 check all the things that `on_failure_jump' does, to make
4544 sure the right things get saved on the stack. Hence we don't
4545 share its code. The only reason to push anything on the
4546 stack at all is that otherwise we would have to change
4547 `anychar's code to do something besides goto fail in this
4548 case; that seems worse than this. */
4549 case on_failure_keep_string_jump
:
4550 DEBUG_PRINT1 ("EXECUTING on_failure_keep_string_jump");
4552 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
4554 DEBUG_PRINT3 (" %d (to %p):\n", mcnt
, p
+ mcnt
);
4556 DEBUG_PRINT3 (" %d (to 0x%x):\n", mcnt
, p
+ mcnt
);
4559 PUSH_FAILURE_POINT (p
+ mcnt
, NULL
, -2);
4563 /* Uses of on_failure_jump:
4565 Each alternative starts with an on_failure_jump that points
4566 to the beginning of the next alternative. Each alternative
4567 except the last ends with a jump that in effect jumps past
4568 the rest of the alternatives. (They really jump to the
4569 ending jump of the following alternative, because tensioning
4570 these jumps is a hassle.)
4572 Repeats start with an on_failure_jump that points past both
4573 the repetition text and either the following jump or
4574 pop_failure_jump back to this on_failure_jump. */
4575 case on_failure_jump
:
4577 DEBUG_PRINT1 ("EXECUTING on_failure_jump");
4579 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
4581 DEBUG_PRINT3 (" %d (to %p)", mcnt
, p
+ mcnt
);
4583 DEBUG_PRINT3 (" %d (to 0x%x)", mcnt
, p
+ mcnt
);
4586 /* If this on_failure_jump comes right before a group (i.e.,
4587 the original * applied to a group), save the information
4588 for that group and all inner ones, so that if we fail back
4589 to this point, the group's information will be correct.
4590 For example, in \(a*\)*\1, we need the preceding group,
4591 and in \(zz\(a*\)b*\)\2, we need the inner group. */
4593 /* We can't use `p' to check ahead because we push
4594 a failure point to `p + mcnt' after we do this. */
4597 /* We need to skip no_op's before we look for the
4598 start_memory in case this on_failure_jump is happening as
4599 the result of a completed succeed_n, as in \(a\)\{1,3\}b\1
4601 while (p1
< pend
&& (re_opcode_t
) *p1
== no_op
)
4604 if (p1
< pend
&& (re_opcode_t
) *p1
== start_memory
)
4606 /* We have a new highest active register now. This will
4607 get reset at the start_memory we are about to get to,
4608 but we will have saved all the registers relevant to
4609 this repetition op, as described above. */
4610 highest_active_reg
= *(p1
+ 1) + *(p1
+ 2);
4611 if (lowest_active_reg
== NO_LOWEST_ACTIVE_REG
)
4612 lowest_active_reg
= *(p1
+ 1);
4615 DEBUG_PRINT1 (":\n");
4616 PUSH_FAILURE_POINT (p
+ mcnt
, d
, -2);
4620 /* A smart repeat ends with `maybe_pop_jump'.
4621 We change it to either `pop_failure_jump' or `jump'. */
4622 case maybe_pop_jump
:
4623 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
4624 DEBUG_PRINT2 ("EXECUTING maybe_pop_jump %d.\n", mcnt
);
4626 register unsigned char *p2
= p
;
4628 /* Compare the beginning of the repeat with what in the
4629 pattern follows its end. If we can establish that there
4630 is nothing that they would both match, i.e., that we
4631 would have to backtrack because of (as in, e.g., `a*a')
4632 then we can change to pop_failure_jump, because we'll
4633 never have to backtrack.
4635 This is not true in the case of alternatives: in
4636 `(a|ab)*' we do need to backtrack to the `ab' alternative
4637 (e.g., if the string was `ab'). But instead of trying to
4638 detect that here, the alternative has put on a dummy
4639 failure point which is what we will end up popping. */
4641 /* Skip over open/close-group commands.
4642 If what follows this loop is a ...+ construct,
4643 look at what begins its body, since we will have to
4644 match at least one of that. */
4648 && ((re_opcode_t
) *p2
== stop_memory
4649 || (re_opcode_t
) *p2
== start_memory
))
4651 else if (p2
+ 6 < pend
4652 && (re_opcode_t
) *p2
== dummy_failure_jump
)
4659 /* p1[0] ... p1[2] are the `on_failure_jump' corresponding
4660 to the `maybe_finalize_jump' of this case. Examine what
4663 /* If we're at the end of the pattern, we can change. */
4666 /* Consider what happens when matching ":\(.*\)"
4667 against ":/". I don't really understand this code
4669 p
[-3] = (unsigned char) pop_failure_jump
;
4671 (" End of pattern: change to `pop_failure_jump'.\n");
4674 else if ((re_opcode_t
) *p2
== exactn
4675 || (bufp
->newline_anchor
&& (re_opcode_t
) *p2
== endline
))
4677 register unsigned char c
4678 = *p2
== (unsigned char) endline
? '\n' : p2
[2];
4680 if ((re_opcode_t
) p1
[3] == exactn
&& p1
[5] != c
)
4682 p
[-3] = (unsigned char) pop_failure_jump
;
4683 DEBUG_PRINT3 (" %c != %c => pop_failure_jump.\n",
4687 else if ((re_opcode_t
) p1
[3] == charset
4688 || (re_opcode_t
) p1
[3] == charset_not
)
4690 int not = (re_opcode_t
) p1
[3] == charset_not
;
4692 if (c
< (unsigned char) (p1
[4] * BYTEWIDTH
)
4693 && p1
[5 + c
/ BYTEWIDTH
] & (1 << (c
% BYTEWIDTH
)))
4696 /* `not' is equal to 1 if c would match, which means
4697 that we can't change to pop_failure_jump. */
4700 p
[-3] = (unsigned char) pop_failure_jump
;
4701 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
4705 else if ((re_opcode_t
) *p2
== charset
)
4708 register unsigned char c
4709 = *p2
== (unsigned char) endline
? '\n' : p2
[2];
4713 if ((re_opcode_t
) p1
[3] == exactn
4714 && ! ((int) p2
[1] * BYTEWIDTH
> (int) p1
[5]
4715 && (p2
[2 + p1
[5] / BYTEWIDTH
]
4716 & (1 << (p1
[5] % BYTEWIDTH
)))))
4718 if ((re_opcode_t
) p1
[3] == exactn
4719 && ! ((int) p2
[1] * BYTEWIDTH
> (int) p1
[4]
4720 && (p2
[2 + p1
[4] / BYTEWIDTH
]
4721 & (1 << (p1
[4] % BYTEWIDTH
)))))
4724 p
[-3] = (unsigned char) pop_failure_jump
;
4725 DEBUG_PRINT3 (" %c != %c => pop_failure_jump.\n",
4729 else if ((re_opcode_t
) p1
[3] == charset_not
)
4732 /* We win if the charset_not inside the loop
4733 lists every character listed in the charset after. */
4734 for (idx
= 0; idx
< (int) p2
[1]; idx
++)
4735 if (! (p2
[2 + idx
] == 0
4736 || (idx
< (int) p1
[4]
4737 && ((p2
[2 + idx
] & ~ p1
[5 + idx
]) == 0))))
4742 p
[-3] = (unsigned char) pop_failure_jump
;
4743 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
4746 else if ((re_opcode_t
) p1
[3] == charset
)
4749 /* We win if the charset inside the loop
4750 has no overlap with the one after the loop. */
4752 idx
< (int) p2
[1] && idx
< (int) p1
[4];
4754 if ((p2
[2 + idx
] & p1
[5 + idx
]) != 0)
4757 if (idx
== p2
[1] || idx
== p1
[4])
4759 p
[-3] = (unsigned char) pop_failure_jump
;
4760 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
4765 p
-= 2; /* Point at relative address again. */
4766 if ((re_opcode_t
) p
[-1] != pop_failure_jump
)
4768 p
[-1] = (unsigned char) jump
;
4769 DEBUG_PRINT1 (" Match => jump.\n");
4770 goto unconditional_jump
;
4772 /* Note fall through. */
4775 /* The end of a simple repeat has a pop_failure_jump back to
4776 its matching on_failure_jump, where the latter will push a
4777 failure point. The pop_failure_jump takes off failure
4778 points put on by this pop_failure_jump's matching
4779 on_failure_jump; we got through the pattern to here from the
4780 matching on_failure_jump, so didn't fail. */
4781 case pop_failure_jump
:
4783 /* We need to pass separate storage for the lowest and
4784 highest registers, even though we don't care about the
4785 actual values. Otherwise, we will restore only one
4786 register from the stack, since lowest will == highest in
4787 `pop_failure_point'. */
4788 active_reg_t dummy_low_reg
, dummy_high_reg
;
4789 unsigned char *pdummy
;
4792 DEBUG_PRINT1 ("EXECUTING pop_failure_jump.\n");
4793 POP_FAILURE_POINT (sdummy
, pdummy
,
4794 dummy_low_reg
, dummy_high_reg
,
4795 reg_dummy
, reg_dummy
, reg_info_dummy
);
4797 /* Note fall through. */
4801 DEBUG_PRINT2 ("\n%p: ", p
);
4803 DEBUG_PRINT2 ("\n0x%x: ", p
);
4805 /* Note fall through. */
4807 /* Unconditionally jump (without popping any failure points). */
4809 EXTRACT_NUMBER_AND_INCR (mcnt
, p
); /* Get the amount to jump. */
4810 DEBUG_PRINT2 ("EXECUTING jump %d ", mcnt
);
4811 p
+= mcnt
; /* Do the jump. */
4813 DEBUG_PRINT2 ("(to %p).\n", p
);
4815 DEBUG_PRINT2 ("(to 0x%x).\n", p
);
4820 /* We need this opcode so we can detect where alternatives end
4821 in `group_match_null_string_p' et al. */
4823 DEBUG_PRINT1 ("EXECUTING jump_past_alt.\n");
4824 goto unconditional_jump
;
4827 /* Normally, the on_failure_jump pushes a failure point, which
4828 then gets popped at pop_failure_jump. We will end up at
4829 pop_failure_jump, also, and with a pattern of, say, `a+', we
4830 are skipping over the on_failure_jump, so we have to push
4831 something meaningless for pop_failure_jump to pop. */
4832 case dummy_failure_jump
:
4833 DEBUG_PRINT1 ("EXECUTING dummy_failure_jump.\n");
4834 /* It doesn't matter what we push for the string here. What
4835 the code at `fail' tests is the value for the pattern. */
4836 PUSH_FAILURE_POINT (NULL
, NULL
, -2);
4837 goto unconditional_jump
;
4840 /* At the end of an alternative, we need to push a dummy failure
4841 point in case we are followed by a `pop_failure_jump', because
4842 we don't want the failure point for the alternative to be
4843 popped. For example, matching `(a|ab)*' against `aab'
4844 requires that we match the `ab' alternative. */
4845 case push_dummy_failure
:
4846 DEBUG_PRINT1 ("EXECUTING push_dummy_failure.\n");
4847 /* See comments just above at `dummy_failure_jump' about the
4849 PUSH_FAILURE_POINT (NULL
, NULL
, -2);
4852 /* Have to succeed matching what follows at least n times.
4853 After that, handle like `on_failure_jump'. */
4855 EXTRACT_NUMBER (mcnt
, p
+ 2);
4856 DEBUG_PRINT2 ("EXECUTING succeed_n %d.\n", mcnt
);
4859 /* Originally, this is how many times we HAVE to succeed. */
4864 STORE_NUMBER_AND_INCR (p
, mcnt
);
4866 DEBUG_PRINT3 (" Setting %p to %d.\n", p
- 2, mcnt
);
4868 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p
- 2, mcnt
);
4874 DEBUG_PRINT2 (" Setting two bytes from %p to no_op.\n", p
+2);
4876 DEBUG_PRINT2 (" Setting two bytes from 0x%x to no_op.\n", p
+2);
4878 p
[2] = (unsigned char) no_op
;
4879 p
[3] = (unsigned char) no_op
;
4885 EXTRACT_NUMBER (mcnt
, p
+ 2);
4886 DEBUG_PRINT2 ("EXECUTING jump_n %d.\n", mcnt
);
4888 /* Originally, this is how many times we CAN jump. */
4892 STORE_NUMBER (p
+ 2, mcnt
);
4894 DEBUG_PRINT3 (" Setting %p to %d.\n", p
+ 2, mcnt
);
4896 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p
+ 2, mcnt
);
4898 goto unconditional_jump
;
4900 /* If don't have to jump any more, skip over the rest of command. */
4907 DEBUG_PRINT1 ("EXECUTING set_number_at.\n");
4909 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
4911 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
4913 DEBUG_PRINT3 (" Setting %p to %d.\n", p1
, mcnt
);
4915 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p1
, mcnt
);
4917 STORE_NUMBER (p1
, mcnt
);
4922 /* The DEC Alpha C compiler 3.x generates incorrect code for the
4923 test WORDCHAR_P (d - 1) != WORDCHAR_P (d) in the expansion of
4924 AT_WORD_BOUNDARY, so this code is disabled. Expanding the
4925 macro and introducing temporary variables works around the bug. */
4928 DEBUG_PRINT1 ("EXECUTING wordbound.\n");
4929 if (AT_WORD_BOUNDARY (d
))
4934 DEBUG_PRINT1 ("EXECUTING notwordbound.\n");
4935 if (AT_WORD_BOUNDARY (d
))
4941 boolean prevchar
, thischar
;
4943 DEBUG_PRINT1 ("EXECUTING wordbound.\n");
4944 if (AT_STRINGS_BEG (d
) || AT_STRINGS_END (d
))
4947 prevchar
= WORDCHAR_P (d
- 1);
4948 thischar
= WORDCHAR_P (d
);
4949 if (prevchar
!= thischar
)
4956 boolean prevchar
, thischar
;
4958 DEBUG_PRINT1 ("EXECUTING notwordbound.\n");
4959 if (AT_STRINGS_BEG (d
) || AT_STRINGS_END (d
))
4962 prevchar
= WORDCHAR_P (d
- 1);
4963 thischar
= WORDCHAR_P (d
);
4964 if (prevchar
!= thischar
)
4971 DEBUG_PRINT1 ("EXECUTING wordbeg.\n");
4972 if (WORDCHAR_P (d
) && (AT_STRINGS_BEG (d
) || !WORDCHAR_P (d
- 1)))
4977 DEBUG_PRINT1 ("EXECUTING wordend.\n");
4978 if (!AT_STRINGS_BEG (d
) && WORDCHAR_P (d
- 1)
4979 && (!WORDCHAR_P (d
) || AT_STRINGS_END (d
)))
4985 DEBUG_PRINT1 ("EXECUTING before_dot.\n");
4986 if (PTR_CHAR_POS ((unsigned char *) d
) >= point
)
4991 DEBUG_PRINT1 ("EXECUTING at_dot.\n");
4992 if (PTR_CHAR_POS ((unsigned char *) d
) != point
)
4997 DEBUG_PRINT1 ("EXECUTING after_dot.\n");
4998 if (PTR_CHAR_POS ((unsigned char *) d
) <= point
)
5003 DEBUG_PRINT2 ("EXECUTING syntaxspec %d.\n", mcnt
);
5008 DEBUG_PRINT1 ("EXECUTING Emacs wordchar.\n");
5012 /* Can't use *d++ here; SYNTAX may be an unsafe macro. */
5014 if (SYNTAX (d
[-1]) != (enum syntaxcode
) mcnt
)
5016 SET_REGS_MATCHED ();
5020 DEBUG_PRINT2 ("EXECUTING notsyntaxspec %d.\n", mcnt
);
5022 goto matchnotsyntax
;
5025 DEBUG_PRINT1 ("EXECUTING Emacs notwordchar.\n");
5029 /* Can't use *d++ here; SYNTAX may be an unsafe macro. */
5031 if (SYNTAX (d
[-1]) == (enum syntaxcode
) mcnt
)
5033 SET_REGS_MATCHED ();
5036 #else /* not emacs */
5038 DEBUG_PRINT1 ("EXECUTING non-Emacs wordchar.\n");
5040 if (!WORDCHAR_P (d
))
5042 SET_REGS_MATCHED ();
5047 DEBUG_PRINT1 ("EXECUTING non-Emacs notwordchar.\n");
5051 SET_REGS_MATCHED ();
5054 #endif /* not emacs */
5059 continue; /* Successfully executed one pattern command; keep going. */
5062 /* We goto here if a matching operation fails. */
5064 if (!FAIL_STACK_EMPTY ())
5065 { /* A restart point is known. Restore to that state. */
5066 DEBUG_PRINT1 ("\nFAIL:\n");
5067 POP_FAILURE_POINT (d
, p
,
5068 lowest_active_reg
, highest_active_reg
,
5069 regstart
, regend
, reg_info
);
5071 /* If this failure point is a dummy, try the next one. */
5075 /* If we failed to the end of the pattern, don't examine *p. */
5079 boolean is_a_jump_n
= false;
5081 /* If failed to a backwards jump that's part of a repetition
5082 loop, need to pop this failure point and use the next one. */
5083 switch ((re_opcode_t
) *p
)
5087 case maybe_pop_jump
:
5088 case pop_failure_jump
:
5091 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
5094 if ((is_a_jump_n
&& (re_opcode_t
) *p1
== succeed_n
)
5096 && (re_opcode_t
) *p1
== on_failure_jump
))
5104 if (d
>= string1
&& d
<= end1
)
5108 break; /* Matching at this starting point really fails. */
5112 goto restore_best_regs
;
5116 return -1; /* Failure to match. */
5119 /* Subroutine definitions for re_match_2. */
5122 /* We are passed P pointing to a register number after a start_memory.
5124 Return true if the pattern up to the corresponding stop_memory can
5125 match the empty string, and false otherwise.
5127 If we find the matching stop_memory, sets P to point to one past its number.
5128 Otherwise, sets P to an undefined byte less than or equal to END.
5130 We don't handle duplicates properly (yet). */
5133 group_match_null_string_p (p
, end
, reg_info
)
5134 unsigned char **p
, *end
;
5135 register_info_type
*reg_info
;
5138 /* Point to after the args to the start_memory. */
5139 unsigned char *p1
= *p
+ 2;
5143 /* Skip over opcodes that can match nothing, and return true or
5144 false, as appropriate, when we get to one that can't, or to the
5145 matching stop_memory. */
5147 switch ((re_opcode_t
) *p1
)
5149 /* Could be either a loop or a series of alternatives. */
5150 case on_failure_jump
:
5152 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
5154 /* If the next operation is not a jump backwards in the
5159 /* Go through the on_failure_jumps of the alternatives,
5160 seeing if any of the alternatives cannot match nothing.
5161 The last alternative starts with only a jump,
5162 whereas the rest start with on_failure_jump and end
5163 with a jump, e.g., here is the pattern for `a|b|c':
5165 /on_failure_jump/0/6/exactn/1/a/jump_past_alt/0/6
5166 /on_failure_jump/0/6/exactn/1/b/jump_past_alt/0/3
5169 So, we have to first go through the first (n-1)
5170 alternatives and then deal with the last one separately. */
5173 /* Deal with the first (n-1) alternatives, which start
5174 with an on_failure_jump (see above) that jumps to right
5175 past a jump_past_alt. */
5177 while ((re_opcode_t
) p1
[mcnt
-3] == jump_past_alt
)
5179 /* `mcnt' holds how many bytes long the alternative
5180 is, including the ending `jump_past_alt' and
5183 if (!alt_match_null_string_p (p1
, p1
+ mcnt
- 3,
5187 /* Move to right after this alternative, including the
5191 /* Break if it's the beginning of an n-th alternative
5192 that doesn't begin with an on_failure_jump. */
5193 if ((re_opcode_t
) *p1
!= on_failure_jump
)
5196 /* Still have to check that it's not an n-th
5197 alternative that starts with an on_failure_jump. */
5199 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
5200 if ((re_opcode_t
) p1
[mcnt
-3] != jump_past_alt
)
5202 /* Get to the beginning of the n-th alternative. */
5208 /* Deal with the last alternative: go back and get number
5209 of the `jump_past_alt' just before it. `mcnt' contains
5210 the length of the alternative. */
5211 EXTRACT_NUMBER (mcnt
, p1
- 2);
5213 if (!alt_match_null_string_p (p1
, p1
+ mcnt
, reg_info
))
5216 p1
+= mcnt
; /* Get past the n-th alternative. */
5222 assert (p1
[1] == **p
);
5228 if (!common_op_match_null_string_p (&p1
, end
, reg_info
))
5231 } /* while p1 < end */
5234 } /* group_match_null_string_p */
5237 /* Similar to group_match_null_string_p, but doesn't deal with alternatives:
5238 It expects P to be the first byte of a single alternative and END one
5239 byte past the last. The alternative can contain groups. */
5242 alt_match_null_string_p (p
, end
, reg_info
)
5243 unsigned char *p
, *end
;
5244 register_info_type
*reg_info
;
5247 unsigned char *p1
= p
;
5251 /* Skip over opcodes that can match nothing, and break when we get
5252 to one that can't. */
5254 switch ((re_opcode_t
) *p1
)
5257 case on_failure_jump
:
5259 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
5264 if (!common_op_match_null_string_p (&p1
, end
, reg_info
))
5267 } /* while p1 < end */
5270 } /* alt_match_null_string_p */
5273 /* Deals with the ops common to group_match_null_string_p and
5274 alt_match_null_string_p.
5276 Sets P to one after the op and its arguments, if any. */
5279 common_op_match_null_string_p (p
, end
, reg_info
)
5280 unsigned char **p
, *end
;
5281 register_info_type
*reg_info
;
5286 unsigned char *p1
= *p
;
5288 switch ((re_opcode_t
) *p1
++)
5308 assert (reg_no
> 0 && reg_no
<= MAX_REGNUM
);
5309 ret
= group_match_null_string_p (&p1
, end
, reg_info
);
5311 /* Have to set this here in case we're checking a group which
5312 contains a group and a back reference to it. */
5314 if (REG_MATCH_NULL_STRING_P (reg_info
[reg_no
]) == MATCH_NULL_UNSET_VALUE
)
5315 REG_MATCH_NULL_STRING_P (reg_info
[reg_no
]) = ret
;
5321 /* If this is an optimized succeed_n for zero times, make the jump. */
5323 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
5331 /* Get to the number of times to succeed. */
5333 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
5338 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
5346 if (!REG_MATCH_NULL_STRING_P (reg_info
[*p1
]))
5354 /* All other opcodes mean we cannot match the empty string. */
5360 } /* common_op_match_null_string_p */
5363 /* Return zero if TRANSLATE[S1] and TRANSLATE[S2] are identical for LEN
5364 bytes; nonzero otherwise. */
5367 bcmp_translate (s1
, s2
, len
, translate
)
5368 const char *s1
, *s2
;
5370 RE_TRANSLATE_TYPE translate
;
5372 register const unsigned char *p1
= (const unsigned char *) s1
;
5373 register const unsigned char *p2
= (const unsigned char *) s2
;
5376 if (translate
[*p1
++] != translate
[*p2
++]) return 1;
5382 /* Entry points for GNU code. */
5384 /* re_compile_pattern is the GNU regular expression compiler: it
5385 compiles PATTERN (of length SIZE) and puts the result in BUFP.
5386 Returns 0 if the pattern was valid, otherwise an error string.
5388 Assumes the `allocated' (and perhaps `buffer') and `translate' fields
5389 are set in BUFP on entry.
5391 We call regex_compile to do the actual compilation. */
5394 re_compile_pattern (pattern
, length
, bufp
)
5395 const char *pattern
;
5397 struct re_pattern_buffer
*bufp
;
5401 /* GNU code is written to assume at least RE_NREGS registers will be set
5402 (and at least one extra will be -1). */
5403 bufp
->regs_allocated
= REGS_UNALLOCATED
;
5405 /* And GNU code determines whether or not to get register information
5406 by passing null for the REGS argument to re_match, etc., not by
5410 /* Match anchors at newline. */
5411 bufp
->newline_anchor
= 1;
5413 ret
= regex_compile (pattern
, length
, re_syntax_options
, bufp
);
5417 return gettext (re_error_msgid
[(int) ret
]);
5420 weak_alias (__re_compile_pattern
, re_compile_pattern
)
5423 /* Entry points compatible with 4.2 BSD regex library. We don't define
5424 them unless specifically requested. */
5426 #if defined _REGEX_RE_COMP || defined _LIBC
5428 /* BSD has one and only one pattern buffer. */
5429 static struct re_pattern_buffer re_comp_buf
;
5433 /* Make these definitions weak in libc, so POSIX programs can redefine
5434 these names if they don't use our functions, and still use
5435 regcomp/regexec below without link errors. */
5445 if (!re_comp_buf
.buffer
)
5446 return gettext ("No previous regular expression");
5450 if (!re_comp_buf
.buffer
)
5452 re_comp_buf
.buffer
= (unsigned char *) malloc (200);
5453 if (re_comp_buf
.buffer
== NULL
)
5454 return (char *) gettext (re_error_msgid
[(int) REG_ESPACE
]);
5455 re_comp_buf
.allocated
= 200;
5457 re_comp_buf
.fastmap
= (char *) malloc (1 << BYTEWIDTH
);
5458 if (re_comp_buf
.fastmap
== NULL
)
5459 return (char *) gettext (re_error_msgid
[(int) REG_ESPACE
]);
5462 /* Since `re_exec' always passes NULL for the `regs' argument, we
5463 don't need to initialize the pattern buffer fields which affect it. */
5465 /* Match anchors at newlines. */
5466 re_comp_buf
.newline_anchor
= 1;
5468 ret
= regex_compile (s
, strlen (s
), re_syntax_options
, &re_comp_buf
);
5473 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
5474 return (char *) gettext (re_error_msgid
[(int) ret
]);
5485 const int len
= strlen (s
);
5487 0 <= re_search (&re_comp_buf
, s
, len
, 0, len
, (struct re_registers
*) 0);
5490 #endif /* _REGEX_RE_COMP */
5492 /* POSIX.2 functions. Don't define these for Emacs. */
5496 /* regcomp takes a regular expression as a string and compiles it.
5498 PREG is a regex_t *. We do not expect any fields to be initialized,
5499 since POSIX says we shouldn't. Thus, we set
5501 `buffer' to the compiled pattern;
5502 `used' to the length of the compiled pattern;
5503 `syntax' to RE_SYNTAX_POSIX_EXTENDED if the
5504 REG_EXTENDED bit in CFLAGS is set; otherwise, to
5505 RE_SYNTAX_POSIX_BASIC;
5506 `newline_anchor' to REG_NEWLINE being set in CFLAGS;
5507 `fastmap' to an allocated space for the fastmap;
5508 `fastmap_accurate' to 1;
5509 `re_nsub' to the number of subexpressions in PATTERN.
5511 PATTERN is the address of the pattern string.
5513 CFLAGS is a series of bits which affect compilation.
5515 If REG_EXTENDED is set, we use POSIX extended syntax; otherwise, we
5516 use POSIX basic syntax.
5518 If REG_NEWLINE is set, then . and [^...] don't match newline.
5519 Also, regexec will try a match beginning after every newline.
5521 If REG_ICASE is set, then we considers upper- and lowercase
5522 versions of letters to be equivalent when matching.
5524 If REG_NOSUB is set, then when PREG is passed to regexec, that
5525 routine will report only success or failure, and nothing about the
5528 It returns 0 if it succeeds, nonzero if it doesn't. (See gnu-regex.h for
5529 the return codes and their meanings.) */
5532 regcomp (preg
, pattern
, cflags
)
5534 const char *pattern
;
5539 = (cflags
& REG_EXTENDED
) ?
5540 RE_SYNTAX_POSIX_EXTENDED
: RE_SYNTAX_POSIX_BASIC
;
5542 /* regex_compile will allocate the space for the compiled pattern. */
5544 preg
->allocated
= 0;
5547 /* Try to allocate space for the fastmap. */
5548 preg
->fastmap
= (char *) malloc (1 << BYTEWIDTH
);
5550 if (cflags
& REG_ICASE
)
5555 = (RE_TRANSLATE_TYPE
) malloc (CHAR_SET_SIZE
5556 * sizeof (*(RE_TRANSLATE_TYPE
)0));
5557 if (preg
->translate
== NULL
)
5558 return (int) REG_ESPACE
;
5560 /* Map uppercase characters to corresponding lowercase ones. */
5561 for (i
= 0; i
< CHAR_SET_SIZE
; i
++)
5562 preg
->translate
[i
] = TOLOWER (i
);
5565 preg
->translate
= NULL
;
5567 /* If REG_NEWLINE is set, newlines are treated differently. */
5568 if (cflags
& REG_NEWLINE
)
5569 { /* REG_NEWLINE implies neither . nor [^...] match newline. */
5570 syntax
&= ~RE_DOT_NEWLINE
;
5571 syntax
|= RE_HAT_LISTS_NOT_NEWLINE
;
5572 /* It also changes the matching behavior. */
5573 preg
->newline_anchor
= 1;
5576 preg
->newline_anchor
= 0;
5578 preg
->no_sub
= !!(cflags
& REG_NOSUB
);
5580 /* POSIX says a null character in the pattern terminates it, so we
5581 can use strlen here in compiling the pattern. */
5582 ret
= regex_compile (pattern
, strlen (pattern
), syntax
, preg
);
5584 /* POSIX doesn't distinguish between an unmatched open-group and an
5585 unmatched close-group: both are REG_EPAREN. */
5586 if (ret
== REG_ERPAREN
) ret
= REG_EPAREN
;
5588 if (ret
== REG_NOERROR
&& preg
->fastmap
)
5590 /* Compute the fastmap now, since regexec cannot modify the pattern
5592 if (re_compile_fastmap (preg
) == -2)
5594 /* Some error occured while computing the fastmap, just forget
5596 free (preg
->fastmap
);
5597 preg
->fastmap
= NULL
;
5604 weak_alias (__regcomp
, regcomp
)
5608 /* regexec searches for a given pattern, specified by PREG, in the
5611 If NMATCH is zero or REG_NOSUB was set in the cflags argument to
5612 `regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at
5613 least NMATCH elements, and we set them to the offsets of the
5614 corresponding matched substrings.
5616 EFLAGS specifies `execution flags' which affect matching: if
5617 REG_NOTBOL is set, then ^ does not match at the beginning of the
5618 string; if REG_NOTEOL is set, then $ does not match at the end.
5620 We return 0 if we find a match and REG_NOMATCH if not. */
5623 regexec (preg
, string
, nmatch
, pmatch
, eflags
)
5624 const regex_t
*preg
;
5627 regmatch_t pmatch
[];
5631 struct re_registers regs
;
5632 regex_t private_preg
;
5633 int len
= strlen (string
);
5634 boolean want_reg_info
= !preg
->no_sub
&& nmatch
> 0;
5636 private_preg
= *preg
;
5638 private_preg
.not_bol
= !!(eflags
& REG_NOTBOL
);
5639 private_preg
.not_eol
= !!(eflags
& REG_NOTEOL
);
5641 /* The user has told us exactly how many registers to return
5642 information about, via `nmatch'. We have to pass that on to the
5643 matching routines. */
5644 private_preg
.regs_allocated
= REGS_FIXED
;
5648 regs
.num_regs
= nmatch
;
5649 regs
.start
= TALLOC (nmatch
, regoff_t
);
5650 regs
.end
= TALLOC (nmatch
, regoff_t
);
5651 if (regs
.start
== NULL
|| regs
.end
== NULL
)
5652 return (int) REG_NOMATCH
;
5655 /* Perform the searching operation. */
5656 ret
= re_search (&private_preg
, string
, len
,
5657 /* start: */ 0, /* range: */ len
,
5658 want_reg_info
? ®s
: (struct re_registers
*) 0);
5660 /* Copy the register information to the POSIX structure. */
5667 for (r
= 0; r
< nmatch
; r
++)
5669 pmatch
[r
].rm_so
= regs
.start
[r
];
5670 pmatch
[r
].rm_eo
= regs
.end
[r
];
5674 /* If we needed the temporary register info, free the space now. */
5679 /* We want zero return to mean success, unlike `re_search'. */
5680 return ret
>= 0 ? (int) REG_NOERROR
: (int) REG_NOMATCH
;
5683 weak_alias (__regexec
, regexec
)
5687 /* Returns a message corresponding to an error code, ERRCODE, returned
5688 from either regcomp or regexec. We don't use PREG here. */
5691 regerror (errcode
, preg
, errbuf
, errbuf_size
)
5693 const regex_t
*preg
;
5702 || errcode
>= (int) (sizeof (re_error_msgid
)
5703 / sizeof (re_error_msgid
[0])))
5704 /* Only error codes returned by the rest of the code should be passed
5705 to this routine. If we are given anything else, or if other regex
5706 code generates an invalid error code, then the program has a bug.
5707 Dump core so we can fix it. */
5710 msg
= gettext (re_error_msgid
[errcode
]);
5712 msg_size
= strlen (msg
) + 1; /* Includes the null. */
5714 if (errbuf_size
!= 0)
5716 if (msg_size
> errbuf_size
)
5718 #if defined HAVE_MEMPCPY || defined _LIBC
5719 *((char *) mempcpy (errbuf
, msg
, errbuf_size
- 1)) = '\0';
5721 memcpy (errbuf
, msg
, errbuf_size
- 1);
5722 errbuf
[errbuf_size
- 1] = 0;
5726 memcpy (errbuf
, msg
, msg_size
);
5732 weak_alias (__regerror
, regerror
)
5736 /* Free dynamically allocated space used by PREG. */
5742 if (preg
->buffer
!= NULL
)
5743 free (preg
->buffer
);
5744 preg
->buffer
= NULL
;
5746 preg
->allocated
= 0;
5749 if (preg
->fastmap
!= NULL
)
5750 free (preg
->fastmap
);
5751 preg
->fastmap
= NULL
;
5752 preg
->fastmap_accurate
= 0;
5754 if (preg
->translate
!= NULL
)
5755 free (preg
->translate
);
5756 preg
->translate
= NULL
;
5759 weak_alias (__regfree
, regfree
)
5762 #endif /* not emacs */