1 /* Extended regular expression matching and search library,
3 (Implements POSIX draft P10003.2/D11.2, except for
4 internationalization features.)
6 Copyright (C) 1993, 1994, 1995 Free Software Foundation, Inc.
8 This file is part of the GNU C Library. Its master source is NOT part of
9 the C library, however. The master source lives in /gd/gnu/lib.
11 The GNU C Library is free software; you can redistribute it and/or
12 modify it under the terms of the GNU Library General Public License as
13 published by the Free Software Foundation; either version 2 of the
14 License, or (at your option) any later version.
16 The GNU C Library 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 GNU
19 Library General Public License for more details.
21 You should have received a copy of the GNU Library General Public
22 License along with the GNU C Library; see the file COPYING.LIB. If
23 not, write to the Free Software Foundation, Inc., 675 Mass Ave,
24 Cambridge, MA 02139, USA. */
26 /* AIX requires this to be the first thing in the file. */
27 #if defined (_AIX) && !defined (REGEX_MALLOC)
37 /* We need this for `regex.h', and perhaps for the Emacs include files. */
38 #include <sys/types.h>
40 /* This is for other GNU distributions with internationalized messages. */
41 #if HAVE_LIBINTL_H || defined (_LIBC)
44 # define gettext(msgid) (msgid)
47 /* The `emacs' switch turns on certain matching commands
48 that make sense only in Emacs. */
57 /* If we are not linking with Emacs proper,
58 we can't use the relocating allocator
59 even if config.h says that we can. */
62 #if defined (STDC_HEADERS) || defined (_LIBC)
69 /* When used in Emacs's lib-src, we need to get bzero and bcopy somehow.
70 If nothing else has been done, use the method below. */
71 #ifdef INHIBIT_STRING_HEADER
72 #if !(defined (HAVE_BZERO) && defined (HAVE_BCOPY))
73 #if !defined (bzero) && !defined (bcopy)
74 #undef INHIBIT_STRING_HEADER
79 /* This is the normal way of making sure we have a bcopy and a bzero.
80 This is used in most programs--a few other programs avoid this
81 by defining INHIBIT_STRING_HEADER. */
82 #ifndef INHIBIT_STRING_HEADER
83 #if defined (HAVE_STRING_H) || defined (STDC_HEADERS) || defined (_LIBC)
86 #define bcmp(s1, s2, n) memcmp ((s1), (s2), (n))
89 #define bcopy(s, d, n) memcpy ((d), (s), (n))
92 #define bzero(s, n) memset ((s), 0, (n))
99 /* Define the syntax stuff for \<, \>, etc. */
101 /* This must be nonzero for the wordchar and notwordchar pattern
102 commands in re_match_2. */
107 #ifdef SWITCH_ENUM_BUG
108 #define SWITCH_ENUM_CAST(x) ((int)(x))
110 #define SWITCH_ENUM_CAST(x) (x)
115 extern char *re_syntax_table
;
117 #else /* not SYNTAX_TABLE */
119 /* How many characters in the character set. */
120 #define CHAR_SET_SIZE 256
122 static char re_syntax_table
[CHAR_SET_SIZE
];
133 bzero (re_syntax_table
, sizeof re_syntax_table
);
135 for (c
= 'a'; c
<= 'z'; c
++)
136 re_syntax_table
[c
] = Sword
;
138 for (c
= 'A'; c
<= 'Z'; c
++)
139 re_syntax_table
[c
] = Sword
;
141 for (c
= '0'; c
<= '9'; c
++)
142 re_syntax_table
[c
] = Sword
;
144 re_syntax_table
['_'] = Sword
;
149 #endif /* not SYNTAX_TABLE */
151 #define SYNTAX(c) re_syntax_table[c]
153 #endif /* not emacs */
155 /* Get the interface, including the syntax bits. */
158 /* isalpha etc. are used for the character classes. */
161 /* Jim Meyering writes:
163 "... Some ctype macros are valid only for character codes that
164 isascii says are ASCII (SGI's IRIX-4.0.5 is one such system --when
165 using /bin/cc or gcc but without giving an ansi option). So, all
166 ctype uses should be through macros like ISPRINT... If
167 STDC_HEADERS is defined, then autoconf has verified that the ctype
168 macros don't need to be guarded with references to isascii. ...
169 Defining isascii to 1 should let any compiler worth its salt
170 eliminate the && through constant folding." */
172 #if defined (STDC_HEADERS) || (!defined (isascii) && !defined (HAVE_ISASCII))
175 #define ISASCII(c) isascii(c)
179 #define ISBLANK(c) (ISASCII (c) && isblank (c))
181 #define ISBLANK(c) ((c) == ' ' || (c) == '\t')
184 #define ISGRAPH(c) (ISASCII (c) && isgraph (c))
186 #define ISGRAPH(c) (ISASCII (c) && isprint (c) && !isspace (c))
189 #define ISPRINT(c) (ISASCII (c) && isprint (c))
190 #define ISDIGIT(c) (ISASCII (c) && isdigit (c))
191 #define ISALNUM(c) (ISASCII (c) && isalnum (c))
192 #define ISALPHA(c) (ISASCII (c) && isalpha (c))
193 #define ISCNTRL(c) (ISASCII (c) && iscntrl (c))
194 #define ISLOWER(c) (ISASCII (c) && islower (c))
195 #define ISPUNCT(c) (ISASCII (c) && ispunct (c))
196 #define ISSPACE(c) (ISASCII (c) && isspace (c))
197 #define ISUPPER(c) (ISASCII (c) && isupper (c))
198 #define ISXDIGIT(c) (ISASCII (c) && isxdigit (c))
201 #define NULL (void *)0
204 /* We remove any previous definition of `SIGN_EXTEND_CHAR',
205 since ours (we hope) works properly with all combinations of
206 machines, compilers, `char' and `unsigned char' argument types.
207 (Per Bothner suggested the basic approach.) */
208 #undef SIGN_EXTEND_CHAR
210 #define SIGN_EXTEND_CHAR(c) ((signed char) (c))
211 #else /* not __STDC__ */
212 /* As in Harbison and Steele. */
213 #define SIGN_EXTEND_CHAR(c) ((((unsigned char) (c)) ^ 128) - 128)
216 /* Should we use malloc or alloca? If REGEX_MALLOC is not defined, we
217 use `alloca' instead of `malloc'. This is because using malloc in
218 re_search* or re_match* could cause memory leaks when C-g is used in
219 Emacs; also, malloc is slower and causes storage fragmentation. On
220 the other hand, malloc is more portable, and easier to debug.
222 Because we sometimes use alloca, some routines have to be macros,
223 not functions -- `alloca'-allocated space disappears at the end of the
224 function it is called in. */
228 #define REGEX_ALLOCATE malloc
229 #define REGEX_REALLOCATE(source, osize, nsize) realloc (source, nsize)
230 #define REGEX_FREE free
232 #else /* not REGEX_MALLOC */
234 /* Emacs already defines alloca, sometimes. */
237 /* Make alloca work the best possible way. */
239 #define alloca __builtin_alloca
240 #else /* not __GNUC__ */
243 #else /* not __GNUC__ or HAVE_ALLOCA_H */
244 #if 0 /* It is a bad idea to declare alloca. We always cast the result. */
245 #ifndef _AIX /* Already did AIX, up at the top. */
247 #endif /* not _AIX */
249 #endif /* not HAVE_ALLOCA_H */
250 #endif /* not __GNUC__ */
252 #endif /* not alloca */
254 #define REGEX_ALLOCATE alloca
256 /* Assumes a `char *destination' variable. */
257 #define REGEX_REALLOCATE(source, osize, nsize) \
258 (destination = (char *) alloca (nsize), \
259 bcopy (source, destination, osize), \
262 /* No need to do anything to free, after alloca. */
263 #define REGEX_FREE(arg) ((void)0) /* Do nothing! But inhibit gcc warning. */
265 #endif /* not REGEX_MALLOC */
267 /* Define how to allocate the failure stack. */
269 #if defined (REL_ALLOC) && defined (REGEX_MALLOC)
271 #define REGEX_ALLOCATE_STACK(size) \
272 r_alloc (&failure_stack_ptr, (size))
273 #define REGEX_REALLOCATE_STACK(source, osize, nsize) \
274 r_re_alloc (&failure_stack_ptr, (nsize))
275 #define REGEX_FREE_STACK(ptr) \
276 r_alloc_free (&failure_stack_ptr)
278 #else /* not using relocating allocator */
282 #define REGEX_ALLOCATE_STACK malloc
283 #define REGEX_REALLOCATE_STACK(source, osize, nsize) realloc (source, nsize)
284 #define REGEX_FREE_STACK free
286 #else /* not REGEX_MALLOC */
288 #define REGEX_ALLOCATE_STACK alloca
290 #define REGEX_REALLOCATE_STACK(source, osize, nsize) \
291 REGEX_REALLOCATE (source, osize, nsize)
292 /* No need to explicitly free anything. */
293 #define REGEX_FREE_STACK(arg)
295 #endif /* not REGEX_MALLOC */
296 #endif /* not using relocating allocator */
299 /* True if `size1' is non-NULL and PTR is pointing anywhere inside
300 `string1' or just past its end. This works if PTR is NULL, which is
302 #define FIRST_STRING_P(ptr) \
303 (size1 && string1 <= (ptr) && (ptr) <= string1 + size1)
305 /* (Re)Allocate N items of type T using malloc, or fail. */
306 #define TALLOC(n, t) ((t *) malloc ((n) * sizeof (t)))
307 #define RETALLOC(addr, n, t) ((addr) = (t *) realloc (addr, (n) * sizeof (t)))
308 #define RETALLOC_IF(addr, n, t) \
309 if (addr) RETALLOC((addr), (n), t); else (addr) = TALLOC ((n), t)
310 #define REGEX_TALLOC(n, t) ((t *) REGEX_ALLOCATE ((n) * sizeof (t)))
312 #define BYTEWIDTH 8 /* In bits. */
314 #define STREQ(s1, s2) ((strcmp (s1, s2) == 0))
318 #define MAX(a, b) ((a) > (b) ? (a) : (b))
319 #define MIN(a, b) ((a) < (b) ? (a) : (b))
321 typedef char boolean
;
325 static int re_match_2_internal ();
327 /* These are the command codes that appear in compiled regular
328 expressions. Some opcodes are followed by argument bytes. A
329 command code can specify any interpretation whatsoever for its
330 arguments. Zero bytes may appear in the compiled regular expression. */
336 /* Succeed right away--no more backtracking. */
339 /* Followed by one byte giving n, then by n literal bytes. */
342 /* Matches any (more or less) character. */
345 /* Matches any one char belonging to specified set. First
346 following byte is number of bitmap bytes. Then come bytes
347 for a bitmap saying which chars are in. Bits in each byte
348 are ordered low-bit-first. A character is in the set if its
349 bit is 1. A character too large to have a bit in the map is
350 automatically not in the set. */
353 /* Same parameters as charset, but match any character that is
354 not one of those specified. */
357 /* Start remembering the text that is matched, for storing in a
358 register. Followed by one byte with the register number, in
359 the range 0 to one less than the pattern buffer's re_nsub
360 field. Then followed by one byte with the number of groups
361 inner to this one. (This last has to be part of the
362 start_memory only because we need it in the on_failure_jump
366 /* Stop remembering the text that is matched and store it in a
367 memory register. Followed by one byte with the register
368 number, in the range 0 to one less than `re_nsub' in the
369 pattern buffer, and one byte with the number of inner groups,
370 just like `start_memory'. (We need the number of inner
371 groups here because we don't have any easy way of finding the
372 corresponding start_memory when we're at a stop_memory.) */
375 /* Match a duplicate of something remembered. Followed by one
376 byte containing the register number. */
379 /* Fail unless at beginning of line. */
382 /* Fail unless at end of line. */
385 /* Succeeds if at beginning of buffer (if emacs) or at beginning
386 of string to be matched (if not). */
389 /* Analogously, for end of buffer/string. */
392 /* Followed by two byte relative address to which to jump. */
395 /* Same as jump, but marks the end of an alternative. */
398 /* Followed by two-byte relative address of place to resume at
399 in case of failure. */
402 /* Like on_failure_jump, but pushes a placeholder instead of the
403 current string position when executed. */
404 on_failure_keep_string_jump
,
406 /* Throw away latest failure point and then jump to following
407 two-byte relative address. */
410 /* Change to pop_failure_jump if know won't have to backtrack to
411 match; otherwise change to jump. This is used to jump
412 back to the beginning of a repeat. If what follows this jump
413 clearly won't match what the repeat does, such that we can be
414 sure that there is no use backtracking out of repetitions
415 already matched, then we change it to a pop_failure_jump.
416 Followed by two-byte address. */
419 /* Jump to following two-byte address, and push a dummy failure
420 point. This failure point will be thrown away if an attempt
421 is made to use it for a failure. A `+' construct makes this
422 before the first repeat. Also used as an intermediary kind
423 of jump when compiling an alternative. */
426 /* Push a dummy failure point and continue. Used at the end of
430 /* Followed by two-byte relative address and two-byte number n.
431 After matching N times, jump to the address upon failure. */
434 /* Followed by two-byte relative address, and two-byte number n.
435 Jump to the address N times, then fail. */
438 /* Set the following two-byte relative address to the
439 subsequent two-byte number. The address *includes* the two
443 wordchar
, /* Matches any word-constituent character. */
444 notwordchar
, /* Matches any char that is not a word-constituent. */
446 wordbeg
, /* Succeeds if at word beginning. */
447 wordend
, /* Succeeds if at word end. */
449 wordbound
, /* Succeeds if at a word boundary. */
450 notwordbound
/* Succeeds if not at a word boundary. */
453 ,before_dot
, /* Succeeds if before point. */
454 at_dot
, /* Succeeds if at point. */
455 after_dot
, /* Succeeds if after point. */
457 /* Matches any character whose syntax is specified. Followed by
458 a byte which contains a syntax code, e.g., Sword. */
461 /* Matches any character whose syntax is not that specified. */
466 /* Common operations on the compiled pattern. */
468 /* Store NUMBER in two contiguous bytes starting at DESTINATION. */
470 #define STORE_NUMBER(destination, number) \
472 (destination)[0] = (number) & 0377; \
473 (destination)[1] = (number) >> 8; \
476 /* Same as STORE_NUMBER, except increment DESTINATION to
477 the byte after where the number is stored. Therefore, DESTINATION
478 must be an lvalue. */
480 #define STORE_NUMBER_AND_INCR(destination, number) \
482 STORE_NUMBER (destination, number); \
483 (destination) += 2; \
486 /* Put into DESTINATION a number stored in two contiguous bytes starting
489 #define EXTRACT_NUMBER(destination, source) \
491 (destination) = *(source) & 0377; \
492 (destination) += SIGN_EXTEND_CHAR (*((source) + 1)) << 8; \
497 extract_number (dest
, source
)
499 unsigned char *source
;
501 int temp
= SIGN_EXTEND_CHAR (*(source
+ 1));
502 *dest
= *source
& 0377;
506 #ifndef EXTRACT_MACROS /* To debug the macros. */
507 #undef EXTRACT_NUMBER
508 #define EXTRACT_NUMBER(dest, src) extract_number (&dest, src)
509 #endif /* not EXTRACT_MACROS */
513 /* Same as EXTRACT_NUMBER, except increment SOURCE to after the number.
514 SOURCE must be an lvalue. */
516 #define EXTRACT_NUMBER_AND_INCR(destination, source) \
518 EXTRACT_NUMBER (destination, source); \
524 extract_number_and_incr (destination
, source
)
526 unsigned char **source
;
528 extract_number (destination
, *source
);
532 #ifndef EXTRACT_MACROS
533 #undef EXTRACT_NUMBER_AND_INCR
534 #define EXTRACT_NUMBER_AND_INCR(dest, src) \
535 extract_number_and_incr (&dest, &src)
536 #endif /* not EXTRACT_MACROS */
540 /* If DEBUG is defined, Regex prints many voluminous messages about what
541 it is doing (if the variable `debug' is nonzero). If linked with the
542 main program in `iregex.c', you can enter patterns and strings
543 interactively. And if linked with the main program in `main.c' and
544 the other test files, you can run the already-written tests. */
548 /* We use standard I/O for debugging. */
551 /* It is useful to test things that ``must'' be true when debugging. */
554 static int debug
= 0;
556 #define DEBUG_STATEMENT(e) e
557 #define DEBUG_PRINT1(x) if (debug) printf (x)
558 #define DEBUG_PRINT2(x1, x2) if (debug) printf (x1, x2)
559 #define DEBUG_PRINT3(x1, x2, x3) if (debug) printf (x1, x2, x3)
560 #define DEBUG_PRINT4(x1, x2, x3, x4) if (debug) printf (x1, x2, x3, x4)
561 #define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) \
562 if (debug) print_partial_compiled_pattern (s, e)
563 #define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) \
564 if (debug) print_double_string (w, s1, sz1, s2, sz2)
567 /* Print the fastmap in human-readable form. */
570 print_fastmap (fastmap
)
573 unsigned was_a_range
= 0;
576 while (i
< (1 << BYTEWIDTH
))
582 while (i
< (1 << BYTEWIDTH
) && fastmap
[i
])
598 /* Print a compiled pattern string in human-readable form, starting at
599 the START pointer into it and ending just before the pointer END. */
602 print_partial_compiled_pattern (start
, end
)
603 unsigned char *start
;
607 unsigned char *p
= start
;
608 unsigned char *pend
= end
;
616 /* Loop over pattern commands. */
619 printf ("%d:\t", p
- start
);
621 switch ((re_opcode_t
) *p
++)
629 printf ("/exactn/%d", mcnt
);
640 printf ("/start_memory/%d/%d", mcnt
, *p
++);
645 printf ("/stop_memory/%d/%d", mcnt
, *p
++);
649 printf ("/duplicate/%d", *p
++);
659 register int c
, last
= -100;
660 register int in_range
= 0;
662 printf ("/charset [%s",
663 (re_opcode_t
) *(p
- 1) == charset_not
? "^" : "");
665 assert (p
+ *p
< pend
);
667 for (c
= 0; c
< 256; c
++)
669 && (p
[1 + (c
/8)] & (1 << (c
% 8))))
671 /* Are we starting a range? */
672 if (last
+ 1 == c
&& ! in_range
)
677 /* Have we broken a range? */
678 else if (last
+ 1 != c
&& in_range
)
707 case on_failure_jump
:
708 extract_number_and_incr (&mcnt
, &p
);
709 printf ("/on_failure_jump to %d", p
+ mcnt
- start
);
712 case on_failure_keep_string_jump
:
713 extract_number_and_incr (&mcnt
, &p
);
714 printf ("/on_failure_keep_string_jump to %d", p
+ mcnt
- start
);
717 case dummy_failure_jump
:
718 extract_number_and_incr (&mcnt
, &p
);
719 printf ("/dummy_failure_jump to %d", p
+ mcnt
- start
);
722 case push_dummy_failure
:
723 printf ("/push_dummy_failure");
727 extract_number_and_incr (&mcnt
, &p
);
728 printf ("/maybe_pop_jump to %d", p
+ mcnt
- start
);
731 case pop_failure_jump
:
732 extract_number_and_incr (&mcnt
, &p
);
733 printf ("/pop_failure_jump to %d", p
+ mcnt
- start
);
737 extract_number_and_incr (&mcnt
, &p
);
738 printf ("/jump_past_alt to %d", p
+ mcnt
- start
);
742 extract_number_and_incr (&mcnt
, &p
);
743 printf ("/jump to %d", p
+ mcnt
- start
);
747 extract_number_and_incr (&mcnt
, &p
);
748 extract_number_and_incr (&mcnt2
, &p
);
749 printf ("/succeed_n to %d, %d times", p
+ mcnt
- start
, mcnt2
);
753 extract_number_and_incr (&mcnt
, &p
);
754 extract_number_and_incr (&mcnt2
, &p
);
755 printf ("/jump_n to %d, %d times", p
+ mcnt
- start
, mcnt2
);
759 extract_number_and_incr (&mcnt
, &p
);
760 extract_number_and_incr (&mcnt2
, &p
);
761 printf ("/set_number_at location %d to %d", p
+ mcnt
- start
, mcnt2
);
765 printf ("/wordbound");
769 printf ("/notwordbound");
781 printf ("/before_dot");
789 printf ("/after_dot");
793 printf ("/syntaxspec");
795 printf ("/%d", mcnt
);
799 printf ("/notsyntaxspec");
801 printf ("/%d", mcnt
);
806 printf ("/wordchar");
810 printf ("/notwordchar");
822 printf ("?%d", *(p
-1));
828 printf ("%d:\tend of pattern.\n", p
- start
);
833 print_compiled_pattern (bufp
)
834 struct re_pattern_buffer
*bufp
;
836 unsigned char *buffer
= bufp
->buffer
;
838 print_partial_compiled_pattern (buffer
, buffer
+ bufp
->used
);
839 printf ("%d bytes used/%d bytes allocated.\n", bufp
->used
, bufp
->allocated
);
841 if (bufp
->fastmap_accurate
&& bufp
->fastmap
)
843 printf ("fastmap: ");
844 print_fastmap (bufp
->fastmap
);
847 printf ("re_nsub: %d\t", bufp
->re_nsub
);
848 printf ("regs_alloc: %d\t", bufp
->regs_allocated
);
849 printf ("can_be_null: %d\t", bufp
->can_be_null
);
850 printf ("newline_anchor: %d\n", bufp
->newline_anchor
);
851 printf ("no_sub: %d\t", bufp
->no_sub
);
852 printf ("not_bol: %d\t", bufp
->not_bol
);
853 printf ("not_eol: %d\t", bufp
->not_eol
);
854 printf ("syntax: %d\n", bufp
->syntax
);
855 /* Perhaps we should print the translate table? */
860 print_double_string (where
, string1
, size1
, string2
, size2
)
873 if (FIRST_STRING_P (where
))
875 for (this_char
= where
- string1
; this_char
< size1
; this_char
++)
876 putchar (string1
[this_char
]);
881 for (this_char
= where
- string2
; this_char
< size2
; this_char
++)
882 putchar (string2
[this_char
]);
886 #else /* not DEBUG */
891 #define DEBUG_STATEMENT(e)
892 #define DEBUG_PRINT1(x)
893 #define DEBUG_PRINT2(x1, x2)
894 #define DEBUG_PRINT3(x1, x2, x3)
895 #define DEBUG_PRINT4(x1, x2, x3, x4)
896 #define DEBUG_PRINT_COMPILED_PATTERN(p, s, e)
897 #define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2)
899 #endif /* not DEBUG */
901 /* Set by `re_set_syntax' to the current regexp syntax to recognize. Can
902 also be assigned to arbitrarily: each pattern buffer stores its own
903 syntax, so it can be changed between regex compilations. */
904 /* This has no initializer because initialized variables in Emacs
905 become read-only after dumping. */
906 reg_syntax_t re_syntax_options
;
909 /* Specify the precise syntax of regexps for compilation. This provides
910 for compatibility for various utilities which historically have
911 different, incompatible syntaxes.
913 The argument SYNTAX is a bit mask comprised of the various bits
914 defined in regex.h. We return the old syntax. */
917 re_set_syntax (syntax
)
920 reg_syntax_t ret
= re_syntax_options
;
922 re_syntax_options
= syntax
;
926 /* This table gives an error message for each of the error codes listed
927 in regex.h. Obviously the order here has to be same as there.
928 POSIX doesn't require that we do anything for REG_NOERROR,
929 but why not be nice? */
931 static const char *re_error_msgid
[] =
932 { "Success", /* REG_NOERROR */
933 "No match", /* REG_NOMATCH */
934 "Invalid regular expression", /* REG_BADPAT */
935 "Invalid collation character", /* REG_ECOLLATE */
936 "Invalid character class name", /* REG_ECTYPE */
937 "Trailing backslash", /* REG_EESCAPE */
938 "Invalid back reference", /* REG_ESUBREG */
939 "Unmatched [ or [^", /* REG_EBRACK */
940 "Unmatched ( or \\(", /* REG_EPAREN */
941 "Unmatched \\{", /* REG_EBRACE */
942 "Invalid content of \\{\\}", /* REG_BADBR */
943 "Invalid range end", /* REG_ERANGE */
944 "Memory exhausted", /* REG_ESPACE */
945 "Invalid preceding regular expression", /* REG_BADRPT */
946 "Premature end of regular expression", /* REG_EEND */
947 "Regular expression too big", /* REG_ESIZE */
948 "Unmatched ) or \\)", /* REG_ERPAREN */
951 /* Avoiding alloca during matching, to placate r_alloc. */
953 /* Define MATCH_MAY_ALLOCATE unless we need to make sure that the
954 searching and matching functions should not call alloca. On some
955 systems, alloca is implemented in terms of malloc, and if we're
956 using the relocating allocator routines, then malloc could cause a
957 relocation, which might (if the strings being searched are in the
958 ralloc heap) shift the data out from underneath the regexp
961 Here's another reason to avoid allocation: Emacs
962 processes input from X in a signal handler; processing X input may
963 call malloc; if input arrives while a matching routine is calling
964 malloc, then we're scrod. But Emacs can't just block input while
965 calling matching routines; then we don't notice interrupts when
966 they come in. So, Emacs blocks input around all regexp calls
967 except the matching calls, which it leaves unprotected, in the
968 faith that they will not malloc. */
970 /* Normally, this is fine. */
971 #define MATCH_MAY_ALLOCATE
973 /* When using GNU C, we are not REALLY using the C alloca, no matter
974 what config.h may say. So don't take precautions for it. */
979 /* The match routines may not allocate if (1) they would do it with malloc
980 and (2) it's not safe for them to use malloc.
981 Note that if REL_ALLOC is defined, matching would not use malloc for the
982 failure stack, but we would still use it for the register vectors;
983 so REL_ALLOC should not affect this. */
984 #if (defined (C_ALLOCA) || defined (REGEX_MALLOC)) && defined (emacs)
985 #undef MATCH_MAY_ALLOCATE
989 /* Failure stack declarations and macros; both re_compile_fastmap and
990 re_match_2 use a failure stack. These have to be macros because of
991 REGEX_ALLOCATE_STACK. */
994 /* Number of failure points for which to initially allocate space
995 when matching. If this number is exceeded, we allocate more
996 space, so it is not a hard limit. */
997 #ifndef INIT_FAILURE_ALLOC
998 #define INIT_FAILURE_ALLOC 5
1001 /* Roughly the maximum number of failure points on the stack. Would be
1002 exactly that if always used MAX_FAILURE_SPACE each time we failed.
1003 This is a variable only so users of regex can assign to it; we never
1004 change it ourselves. */
1005 #if defined (MATCH_MAY_ALLOCATE)
1006 int re_max_failures
= 200000;
1008 int re_max_failures
= 2000;
1011 union fail_stack_elt
1013 unsigned char *pointer
;
1017 typedef union fail_stack_elt fail_stack_elt_t
;
1021 fail_stack_elt_t
*stack
;
1023 unsigned avail
; /* Offset of next open position. */
1026 #define FAIL_STACK_EMPTY() (fail_stack.avail == 0)
1027 #define FAIL_STACK_PTR_EMPTY() (fail_stack_ptr->avail == 0)
1028 #define FAIL_STACK_FULL() (fail_stack.avail == fail_stack.size)
1031 /* Define macros to initialize and free the failure stack.
1032 Do `return -2' if the alloc fails. */
1034 #ifdef MATCH_MAY_ALLOCATE
1035 #define INIT_FAIL_STACK() \
1037 fail_stack.stack = (fail_stack_elt_t *) \
1038 REGEX_ALLOCATE_STACK (INIT_FAILURE_ALLOC * sizeof (fail_stack_elt_t)); \
1040 if (fail_stack.stack == NULL) \
1043 fail_stack.size = INIT_FAILURE_ALLOC; \
1044 fail_stack.avail = 0; \
1047 #define RESET_FAIL_STACK() REGEX_FREE_STACK (fail_stack.stack)
1049 #define INIT_FAIL_STACK() \
1051 fail_stack.avail = 0; \
1054 #define RESET_FAIL_STACK()
1058 /* Double the size of FAIL_STACK, up to approximately `re_max_failures' items.
1060 Return 1 if succeeds, and 0 if either ran out of memory
1061 allocating space for it or it was already too large.
1063 REGEX_REALLOCATE_STACK requires `destination' be declared. */
1065 #define DOUBLE_FAIL_STACK(fail_stack) \
1066 ((fail_stack).size > re_max_failures * MAX_FAILURE_ITEMS \
1068 : ((fail_stack).stack = (fail_stack_elt_t *) \
1069 REGEX_REALLOCATE_STACK ((fail_stack).stack, \
1070 (fail_stack).size * sizeof (fail_stack_elt_t), \
1071 ((fail_stack).size << 1) * sizeof (fail_stack_elt_t)), \
1073 (fail_stack).stack == NULL \
1075 : ((fail_stack).size <<= 1, \
1079 /* Push pointer POINTER on FAIL_STACK.
1080 Return 1 if was able to do so and 0 if ran out of memory allocating
1082 #define PUSH_PATTERN_OP(POINTER, FAIL_STACK) \
1083 ((FAIL_STACK_FULL () \
1084 && !DOUBLE_FAIL_STACK (FAIL_STACK)) \
1086 : ((FAIL_STACK).stack[(FAIL_STACK).avail++].pointer = POINTER, \
1089 /* Push a pointer value onto the failure stack.
1090 Assumes the variable `fail_stack'. Probably should only
1091 be called from within `PUSH_FAILURE_POINT'. */
1092 #define PUSH_FAILURE_POINTER(item) \
1093 fail_stack.stack[fail_stack.avail++].pointer = (unsigned char *) (item)
1095 /* This pushes an integer-valued item onto the failure stack.
1096 Assumes the variable `fail_stack'. Probably should only
1097 be called from within `PUSH_FAILURE_POINT'. */
1098 #define PUSH_FAILURE_INT(item) \
1099 fail_stack.stack[fail_stack.avail++].integer = (item)
1101 /* Push a fail_stack_elt_t value onto the failure stack.
1102 Assumes the variable `fail_stack'. Probably should only
1103 be called from within `PUSH_FAILURE_POINT'. */
1104 #define PUSH_FAILURE_ELT(item) \
1105 fail_stack.stack[fail_stack.avail++] = (item)
1107 /* These three POP... operations complement the three PUSH... operations.
1108 All assume that `fail_stack' is nonempty. */
1109 #define POP_FAILURE_POINTER() fail_stack.stack[--fail_stack.avail].pointer
1110 #define POP_FAILURE_INT() fail_stack.stack[--fail_stack.avail].integer
1111 #define POP_FAILURE_ELT() fail_stack.stack[--fail_stack.avail]
1113 /* Used to omit pushing failure point id's when we're not debugging. */
1115 #define DEBUG_PUSH PUSH_FAILURE_INT
1116 #define DEBUG_POP(item_addr) *(item_addr) = POP_FAILURE_INT ()
1118 #define DEBUG_PUSH(item)
1119 #define DEBUG_POP(item_addr)
1123 /* Push the information about the state we will need
1124 if we ever fail back to it.
1126 Requires variables fail_stack, regstart, regend, reg_info, and
1127 num_regs be declared. DOUBLE_FAIL_STACK requires `destination' be
1130 Does `return FAILURE_CODE' if runs out of memory. */
1132 #define PUSH_FAILURE_POINT(pattern_place, string_place, failure_code) \
1134 char *destination; \
1135 /* Must be int, so when we don't save any registers, the arithmetic \
1136 of 0 + -1 isn't done as unsigned. */ \
1139 DEBUG_STATEMENT (failure_id++); \
1140 DEBUG_STATEMENT (nfailure_points_pushed++); \
1141 DEBUG_PRINT2 ("\nPUSH_FAILURE_POINT #%u:\n", failure_id); \
1142 DEBUG_PRINT2 (" Before push, next avail: %d\n", (fail_stack).avail);\
1143 DEBUG_PRINT2 (" size: %d\n", (fail_stack).size);\
1145 DEBUG_PRINT2 (" slots needed: %d\n", NUM_FAILURE_ITEMS); \
1146 DEBUG_PRINT2 (" available: %d\n", REMAINING_AVAIL_SLOTS); \
1148 /* Ensure we have enough space allocated for what we will push. */ \
1149 while (REMAINING_AVAIL_SLOTS < NUM_FAILURE_ITEMS) \
1151 if (!DOUBLE_FAIL_STACK (fail_stack)) \
1152 return failure_code; \
1154 DEBUG_PRINT2 ("\n Doubled stack; size now: %d\n", \
1155 (fail_stack).size); \
1156 DEBUG_PRINT2 (" slots available: %d\n", REMAINING_AVAIL_SLOTS);\
1159 /* Push the info, starting with the registers. */ \
1160 DEBUG_PRINT1 ("\n"); \
1162 if (!(RE_NO_POSIX_BACKTRACKING & bufp->syntax)) \
1163 for (this_reg = lowest_active_reg; this_reg <= highest_active_reg; \
1166 DEBUG_PRINT2 (" Pushing reg: %d\n", this_reg); \
1167 DEBUG_STATEMENT (num_regs_pushed++); \
1169 DEBUG_PRINT2 (" start: 0x%x\n", regstart[this_reg]); \
1170 PUSH_FAILURE_POINTER (regstart[this_reg]); \
1172 DEBUG_PRINT2 (" end: 0x%x\n", regend[this_reg]); \
1173 PUSH_FAILURE_POINTER (regend[this_reg]); \
1175 DEBUG_PRINT2 (" info: 0x%x\n ", reg_info[this_reg]); \
1176 DEBUG_PRINT2 (" match_null=%d", \
1177 REG_MATCH_NULL_STRING_P (reg_info[this_reg])); \
1178 DEBUG_PRINT2 (" active=%d", IS_ACTIVE (reg_info[this_reg])); \
1179 DEBUG_PRINT2 (" matched_something=%d", \
1180 MATCHED_SOMETHING (reg_info[this_reg])); \
1181 DEBUG_PRINT2 (" ever_matched=%d", \
1182 EVER_MATCHED_SOMETHING (reg_info[this_reg])); \
1183 DEBUG_PRINT1 ("\n"); \
1184 PUSH_FAILURE_ELT (reg_info[this_reg].word); \
1187 DEBUG_PRINT2 (" Pushing low active reg: %d\n", lowest_active_reg);\
1188 PUSH_FAILURE_INT (lowest_active_reg); \
1190 DEBUG_PRINT2 (" Pushing high active reg: %d\n", highest_active_reg);\
1191 PUSH_FAILURE_INT (highest_active_reg); \
1193 DEBUG_PRINT2 (" Pushing pattern 0x%x: ", pattern_place); \
1194 DEBUG_PRINT_COMPILED_PATTERN (bufp, pattern_place, pend); \
1195 PUSH_FAILURE_POINTER (pattern_place); \
1197 DEBUG_PRINT2 (" Pushing string 0x%x: `", string_place); \
1198 DEBUG_PRINT_DOUBLE_STRING (string_place, string1, size1, string2, \
1200 DEBUG_PRINT1 ("'\n"); \
1201 PUSH_FAILURE_POINTER (string_place); \
1203 DEBUG_PRINT2 (" Pushing failure id: %u\n", failure_id); \
1204 DEBUG_PUSH (failure_id); \
1207 /* This is the number of items that are pushed and popped on the stack
1208 for each register. */
1209 #define NUM_REG_ITEMS 3
1211 /* Individual items aside from the registers. */
1213 #define NUM_NONREG_ITEMS 5 /* Includes failure point id. */
1215 #define NUM_NONREG_ITEMS 4
1218 /* We push at most this many items on the stack. */
1219 #define MAX_FAILURE_ITEMS ((num_regs - 1) * NUM_REG_ITEMS + NUM_NONREG_ITEMS)
1221 /* We actually push this many items. */
1222 #define NUM_FAILURE_ITEMS \
1223 (((RE_NO_POSIX_BACKTRACKING & bufp->syntax \
1224 ? 0 : highest_active_reg - lowest_active_reg + 1) \
1228 /* How many items can still be added to the stack without overflowing it. */
1229 #define REMAINING_AVAIL_SLOTS ((fail_stack).size - (fail_stack).avail)
1232 /* Pops what PUSH_FAIL_STACK pushes.
1234 We restore into the parameters, all of which should be lvalues:
1235 STR -- the saved data position.
1236 PAT -- the saved pattern position.
1237 LOW_REG, HIGH_REG -- the highest and lowest active registers.
1238 REGSTART, REGEND -- arrays of string positions.
1239 REG_INFO -- array of information about each subexpression.
1241 Also assumes the variables `fail_stack' and (if debugging), `bufp',
1242 `pend', `string1', `size1', `string2', and `size2'. */
1244 #define POP_FAILURE_POINT(str, pat, low_reg, high_reg, regstart, regend, reg_info)\
1246 DEBUG_STATEMENT (fail_stack_elt_t failure_id;) \
1248 const unsigned char *string_temp; \
1250 assert (!FAIL_STACK_EMPTY ()); \
1252 /* Remove failure points and point to how many regs pushed. */ \
1253 DEBUG_PRINT1 ("POP_FAILURE_POINT:\n"); \
1254 DEBUG_PRINT2 (" Before pop, next avail: %d\n", fail_stack.avail); \
1255 DEBUG_PRINT2 (" size: %d\n", fail_stack.size); \
1257 assert (fail_stack.avail >= NUM_NONREG_ITEMS); \
1259 DEBUG_POP (&failure_id); \
1260 DEBUG_PRINT2 (" Popping failure id: %u\n", failure_id); \
1262 /* If the saved string location is NULL, it came from an \
1263 on_failure_keep_string_jump opcode, and we want to throw away the \
1264 saved NULL, thus retaining our current position in the string. */ \
1265 string_temp = POP_FAILURE_POINTER (); \
1266 if (string_temp != NULL) \
1267 str = (const char *) string_temp; \
1269 DEBUG_PRINT2 (" Popping string 0x%x: `", str); \
1270 DEBUG_PRINT_DOUBLE_STRING (str, string1, size1, string2, size2); \
1271 DEBUG_PRINT1 ("'\n"); \
1273 pat = (unsigned char *) POP_FAILURE_POINTER (); \
1274 DEBUG_PRINT2 (" Popping pattern 0x%x: ", pat); \
1275 DEBUG_PRINT_COMPILED_PATTERN (bufp, pat, pend); \
1277 /* Restore register info. */ \
1278 high_reg = (unsigned) POP_FAILURE_INT (); \
1279 DEBUG_PRINT2 (" Popping high active reg: %d\n", high_reg); \
1281 low_reg = (unsigned) POP_FAILURE_INT (); \
1282 DEBUG_PRINT2 (" Popping low active reg: %d\n", low_reg); \
1284 if (!(RE_NO_POSIX_BACKTRACKING & bufp->syntax)) \
1285 for (this_reg = high_reg; this_reg >= low_reg; this_reg--) \
1287 DEBUG_PRINT2 (" Popping reg: %d\n", this_reg); \
1289 reg_info[this_reg].word = POP_FAILURE_ELT (); \
1290 DEBUG_PRINT2 (" info: 0x%x\n", reg_info[this_reg]); \
1292 regend[this_reg] = (const char *) POP_FAILURE_POINTER (); \
1293 DEBUG_PRINT2 (" end: 0x%x\n", regend[this_reg]); \
1295 regstart[this_reg] = (const char *) POP_FAILURE_POINTER (); \
1296 DEBUG_PRINT2 (" start: 0x%x\n", regstart[this_reg]); \
1300 for (this_reg = highest_active_reg; this_reg > high_reg; this_reg--) \
1302 reg_info[this_reg].word = 0; \
1303 regend[this_reg] = 0; \
1304 regstart[this_reg] = 0; \
1306 highest_active_reg = high_reg; \
1309 set_regs_matched_done = 0; \
1310 DEBUG_STATEMENT (nfailure_points_popped++); \
1311 } /* POP_FAILURE_POINT */
1315 /* Structure for per-register (a.k.a. per-group) information.
1316 Other register information, such as the
1317 starting and ending positions (which are addresses), and the list of
1318 inner groups (which is a bits list) are maintained in separate
1321 We are making a (strictly speaking) nonportable assumption here: that
1322 the compiler will pack our bit fields into something that fits into
1323 the type of `word', i.e., is something that fits into one item on the
1328 fail_stack_elt_t word
;
1331 /* This field is one if this group can match the empty string,
1332 zero if not. If not yet determined, `MATCH_NULL_UNSET_VALUE'. */
1333 #define MATCH_NULL_UNSET_VALUE 3
1334 unsigned match_null_string_p
: 2;
1335 unsigned is_active
: 1;
1336 unsigned matched_something
: 1;
1337 unsigned ever_matched_something
: 1;
1339 } register_info_type
;
1341 #define REG_MATCH_NULL_STRING_P(R) ((R).bits.match_null_string_p)
1342 #define IS_ACTIVE(R) ((R).bits.is_active)
1343 #define MATCHED_SOMETHING(R) ((R).bits.matched_something)
1344 #define EVER_MATCHED_SOMETHING(R) ((R).bits.ever_matched_something)
1347 /* Call this when have matched a real character; it sets `matched' flags
1348 for the subexpressions which we are currently inside. Also records
1349 that those subexprs have matched. */
1350 #define SET_REGS_MATCHED() \
1353 if (!set_regs_matched_done) \
1356 set_regs_matched_done = 1; \
1357 for (r = lowest_active_reg; r <= highest_active_reg; r++) \
1359 MATCHED_SOMETHING (reg_info[r]) \
1360 = EVER_MATCHED_SOMETHING (reg_info[r]) \
1367 /* Registers are set to a sentinel when they haven't yet matched. */
1368 static char reg_unset_dummy
;
1369 #define REG_UNSET_VALUE (®_unset_dummy)
1370 #define REG_UNSET(e) ((e) == REG_UNSET_VALUE)
1372 /* Subroutine declarations and macros for regex_compile. */
1374 static void store_op1 (), store_op2 ();
1375 static void insert_op1 (), insert_op2 ();
1376 static boolean
at_begline_loc_p (), at_endline_loc_p ();
1377 static boolean
group_in_compile_stack ();
1378 static reg_errcode_t
compile_range ();
1380 /* Fetch the next character in the uncompiled pattern---translating it
1381 if necessary. Also cast from a signed character in the constant
1382 string passed to us by the user to an unsigned char that we can use
1383 as an array index (in, e.g., `translate'). */
1385 #define PATFETCH(c) \
1386 do {if (p == pend) return REG_EEND; \
1387 c = (unsigned char) *p++; \
1388 if (translate) c = (unsigned char) translate[c]; \
1392 /* Fetch the next character in the uncompiled pattern, with no
1394 #define PATFETCH_RAW(c) \
1395 do {if (p == pend) return REG_EEND; \
1396 c = (unsigned char) *p++; \
1399 /* Go backwards one character in the pattern. */
1400 #define PATUNFETCH p--
1403 /* If `translate' is non-null, return translate[D], else just D. We
1404 cast the subscript to translate because some data is declared as
1405 `char *', to avoid warnings when a string constant is passed. But
1406 when we use a character as a subscript we must make it unsigned. */
1408 #define TRANSLATE(d) \
1409 (translate ? (char) translate[(unsigned char) (d)] : (d))
1413 /* Macros for outputting the compiled pattern into `buffer'. */
1415 /* If the buffer isn't allocated when it comes in, use this. */
1416 #define INIT_BUF_SIZE 32
1418 /* Make sure we have at least N more bytes of space in buffer. */
1419 #define GET_BUFFER_SPACE(n) \
1420 while (b - bufp->buffer + (n) > bufp->allocated) \
1423 /* Make sure we have one more byte of buffer space and then add C to it. */
1424 #define BUF_PUSH(c) \
1426 GET_BUFFER_SPACE (1); \
1427 *b++ = (unsigned char) (c); \
1431 /* Ensure we have two more bytes of buffer space and then append C1 and C2. */
1432 #define BUF_PUSH_2(c1, c2) \
1434 GET_BUFFER_SPACE (2); \
1435 *b++ = (unsigned char) (c1); \
1436 *b++ = (unsigned char) (c2); \
1440 /* As with BUF_PUSH_2, except for three bytes. */
1441 #define BUF_PUSH_3(c1, c2, c3) \
1443 GET_BUFFER_SPACE (3); \
1444 *b++ = (unsigned char) (c1); \
1445 *b++ = (unsigned char) (c2); \
1446 *b++ = (unsigned char) (c3); \
1450 /* Store a jump with opcode OP at LOC to location TO. We store a
1451 relative address offset by the three bytes the jump itself occupies. */
1452 #define STORE_JUMP(op, loc, to) \
1453 store_op1 (op, loc, (to) - (loc) - 3)
1455 /* Likewise, for a two-argument jump. */
1456 #define STORE_JUMP2(op, loc, to, arg) \
1457 store_op2 (op, loc, (to) - (loc) - 3, arg)
1459 /* Like `STORE_JUMP', but for inserting. Assume `b' is the buffer end. */
1460 #define INSERT_JUMP(op, loc, to) \
1461 insert_op1 (op, loc, (to) - (loc) - 3, b)
1463 /* Like `STORE_JUMP2', but for inserting. Assume `b' is the buffer end. */
1464 #define INSERT_JUMP2(op, loc, to, arg) \
1465 insert_op2 (op, loc, (to) - (loc) - 3, arg, b)
1468 /* This is not an arbitrary limit: the arguments which represent offsets
1469 into the pattern are two bytes long. So if 2^16 bytes turns out to
1470 be too small, many things would have to change. */
1471 #define MAX_BUF_SIZE (1L << 16)
1474 /* Extend the buffer by twice its current size via realloc and
1475 reset the pointers that pointed into the old block to point to the
1476 correct places in the new one. If extending the buffer results in it
1477 being larger than MAX_BUF_SIZE, then flag memory exhausted. */
1478 #define EXTEND_BUFFER() \
1480 unsigned char *old_buffer = bufp->buffer; \
1481 if (bufp->allocated == MAX_BUF_SIZE) \
1483 bufp->allocated <<= 1; \
1484 if (bufp->allocated > MAX_BUF_SIZE) \
1485 bufp->allocated = MAX_BUF_SIZE; \
1486 bufp->buffer = (unsigned char *) realloc (bufp->buffer, bufp->allocated);\
1487 if (bufp->buffer == NULL) \
1488 return REG_ESPACE; \
1489 /* If the buffer moved, move all the pointers into it. */ \
1490 if (old_buffer != bufp->buffer) \
1492 b = (b - old_buffer) + bufp->buffer; \
1493 begalt = (begalt - old_buffer) + bufp->buffer; \
1494 if (fixup_alt_jump) \
1495 fixup_alt_jump = (fixup_alt_jump - old_buffer) + bufp->buffer;\
1497 laststart = (laststart - old_buffer) + bufp->buffer; \
1498 if (pending_exact) \
1499 pending_exact = (pending_exact - old_buffer) + bufp->buffer; \
1504 /* Since we have one byte reserved for the register number argument to
1505 {start,stop}_memory, the maximum number of groups we can report
1506 things about is what fits in that byte. */
1507 #define MAX_REGNUM 255
1509 /* But patterns can have more than `MAX_REGNUM' registers. We just
1510 ignore the excess. */
1511 typedef unsigned regnum_t
;
1514 /* Macros for the compile stack. */
1516 /* Since offsets can go either forwards or backwards, this type needs to
1517 be able to hold values from -(MAX_BUF_SIZE - 1) to MAX_BUF_SIZE - 1. */
1518 typedef int pattern_offset_t
;
1522 pattern_offset_t begalt_offset
;
1523 pattern_offset_t fixup_alt_jump
;
1524 pattern_offset_t inner_group_offset
;
1525 pattern_offset_t laststart_offset
;
1527 } compile_stack_elt_t
;
1532 compile_stack_elt_t
*stack
;
1534 unsigned avail
; /* Offset of next open position. */
1535 } compile_stack_type
;
1538 #define INIT_COMPILE_STACK_SIZE 32
1540 #define COMPILE_STACK_EMPTY (compile_stack.avail == 0)
1541 #define COMPILE_STACK_FULL (compile_stack.avail == compile_stack.size)
1543 /* The next available element. */
1544 #define COMPILE_STACK_TOP (compile_stack.stack[compile_stack.avail])
1547 /* Set the bit for character C in a list. */
1548 #define SET_LIST_BIT(c) \
1549 (b[((unsigned char) (c)) / BYTEWIDTH] \
1550 |= 1 << (((unsigned char) c) % BYTEWIDTH))
1553 /* Get the next unsigned number in the uncompiled pattern. */
1554 #define GET_UNSIGNED_NUMBER(num) \
1558 while (ISDIGIT (c)) \
1562 num = num * 10 + c - '0'; \
1570 #define CHAR_CLASS_MAX_LENGTH 6 /* Namely, `xdigit'. */
1572 #define IS_CHAR_CLASS(string) \
1573 (STREQ (string, "alpha") || STREQ (string, "upper") \
1574 || STREQ (string, "lower") || STREQ (string, "digit") \
1575 || STREQ (string, "alnum") || STREQ (string, "xdigit") \
1576 || STREQ (string, "space") || STREQ (string, "print") \
1577 || STREQ (string, "punct") || STREQ (string, "graph") \
1578 || STREQ (string, "cntrl") || STREQ (string, "blank"))
1580 #ifndef MATCH_MAY_ALLOCATE
1582 /* If we cannot allocate large objects within re_match_2_internal,
1583 we make the fail stack and register vectors global.
1584 The fail stack, we grow to the maximum size when a regexp
1586 The register vectors, we adjust in size each time we
1587 compile a regexp, according to the number of registers it needs. */
1589 static fail_stack_type fail_stack
;
1591 /* Size with which the following vectors are currently allocated.
1592 That is so we can make them bigger as needed,
1593 but never make them smaller. */
1594 static int regs_allocated_size
;
1596 static const char ** regstart
, ** regend
;
1597 static const char ** old_regstart
, ** old_regend
;
1598 static const char **best_regstart
, **best_regend
;
1599 static register_info_type
*reg_info
;
1600 static const char **reg_dummy
;
1601 static register_info_type
*reg_info_dummy
;
1603 /* Make the register vectors big enough for NUM_REGS registers,
1604 but don't make them smaller. */
1607 regex_grow_registers (num_regs
)
1610 if (num_regs
> regs_allocated_size
)
1612 RETALLOC_IF (regstart
, num_regs
, const char *);
1613 RETALLOC_IF (regend
, num_regs
, const char *);
1614 RETALLOC_IF (old_regstart
, num_regs
, const char *);
1615 RETALLOC_IF (old_regend
, num_regs
, const char *);
1616 RETALLOC_IF (best_regstart
, num_regs
, const char *);
1617 RETALLOC_IF (best_regend
, num_regs
, const char *);
1618 RETALLOC_IF (reg_info
, num_regs
, register_info_type
);
1619 RETALLOC_IF (reg_dummy
, num_regs
, const char *);
1620 RETALLOC_IF (reg_info_dummy
, num_regs
, register_info_type
);
1622 regs_allocated_size
= num_regs
;
1626 #endif /* not MATCH_MAY_ALLOCATE */
1628 /* `regex_compile' compiles PATTERN (of length SIZE) according to SYNTAX.
1629 Returns one of error codes defined in `regex.h', or zero for success.
1631 Assumes the `allocated' (and perhaps `buffer') and `translate'
1632 fields are set in BUFP on entry.
1634 If it succeeds, results are put in BUFP (if it returns an error, the
1635 contents of BUFP are undefined):
1636 `buffer' is the compiled pattern;
1637 `syntax' is set to SYNTAX;
1638 `used' is set to the length of the compiled pattern;
1639 `fastmap_accurate' is zero;
1640 `re_nsub' is the number of subexpressions in PATTERN;
1641 `not_bol' and `not_eol' are zero;
1643 The `fastmap' and `newline_anchor' fields are neither
1644 examined nor set. */
1646 /* Return, freeing storage we allocated. */
1647 #define FREE_STACK_RETURN(value) \
1648 return (free (compile_stack.stack), value)
1650 static reg_errcode_t
1651 regex_compile (pattern
, size
, syntax
, bufp
)
1652 const char *pattern
;
1654 reg_syntax_t syntax
;
1655 struct re_pattern_buffer
*bufp
;
1657 /* We fetch characters from PATTERN here. Even though PATTERN is
1658 `char *' (i.e., signed), we declare these variables as unsigned, so
1659 they can be reliably used as array indices. */
1660 register unsigned char c
, c1
;
1662 /* A random temporary spot in PATTERN. */
1665 /* Points to the end of the buffer, where we should append. */
1666 register unsigned char *b
;
1668 /* Keeps track of unclosed groups. */
1669 compile_stack_type compile_stack
;
1671 /* Points to the current (ending) position in the pattern. */
1672 const char *p
= pattern
;
1673 const char *pend
= pattern
+ size
;
1675 /* How to translate the characters in the pattern. */
1676 RE_TRANSLATE_TYPE translate
= bufp
->translate
;
1678 /* Address of the count-byte of the most recently inserted `exactn'
1679 command. This makes it possible to tell if a new exact-match
1680 character can be added to that command or if the character requires
1681 a new `exactn' command. */
1682 unsigned char *pending_exact
= 0;
1684 /* Address of start of the most recently finished expression.
1685 This tells, e.g., postfix * where to find the start of its
1686 operand. Reset at the beginning of groups and alternatives. */
1687 unsigned char *laststart
= 0;
1689 /* Address of beginning of regexp, or inside of last group. */
1690 unsigned char *begalt
;
1692 /* Place in the uncompiled pattern (i.e., the {) to
1693 which to go back if the interval is invalid. */
1694 const char *beg_interval
;
1696 /* Address of the place where a forward jump should go to the end of
1697 the containing expression. Each alternative of an `or' -- except the
1698 last -- ends with a forward jump of this sort. */
1699 unsigned char *fixup_alt_jump
= 0;
1701 /* Counts open-groups as they are encountered. Remembered for the
1702 matching close-group on the compile stack, so the same register
1703 number is put in the stop_memory as the start_memory. */
1704 regnum_t regnum
= 0;
1707 DEBUG_PRINT1 ("\nCompiling pattern: ");
1710 unsigned debug_count
;
1712 for (debug_count
= 0; debug_count
< size
; debug_count
++)
1713 putchar (pattern
[debug_count
]);
1718 /* Initialize the compile stack. */
1719 compile_stack
.stack
= TALLOC (INIT_COMPILE_STACK_SIZE
, compile_stack_elt_t
);
1720 if (compile_stack
.stack
== NULL
)
1723 compile_stack
.size
= INIT_COMPILE_STACK_SIZE
;
1724 compile_stack
.avail
= 0;
1726 /* Initialize the pattern buffer. */
1727 bufp
->syntax
= syntax
;
1728 bufp
->fastmap_accurate
= 0;
1729 bufp
->not_bol
= bufp
->not_eol
= 0;
1731 /* Set `used' to zero, so that if we return an error, the pattern
1732 printer (for debugging) will think there's no pattern. We reset it
1736 /* Always count groups, whether or not bufp->no_sub is set. */
1739 #if !defined (emacs) && !defined (SYNTAX_TABLE)
1740 /* Initialize the syntax table. */
1741 init_syntax_once ();
1744 if (bufp
->allocated
== 0)
1747 { /* If zero allocated, but buffer is non-null, try to realloc
1748 enough space. This loses if buffer's address is bogus, but
1749 that is the user's responsibility. */
1750 RETALLOC (bufp
->buffer
, INIT_BUF_SIZE
, unsigned char);
1753 { /* Caller did not allocate a buffer. Do it for them. */
1754 bufp
->buffer
= TALLOC (INIT_BUF_SIZE
, unsigned char);
1756 if (!bufp
->buffer
) FREE_STACK_RETURN (REG_ESPACE
);
1758 bufp
->allocated
= INIT_BUF_SIZE
;
1761 begalt
= b
= bufp
->buffer
;
1763 /* Loop through the uncompiled pattern until we're at the end. */
1772 if ( /* If at start of pattern, it's an operator. */
1774 /* If context independent, it's an operator. */
1775 || syntax
& RE_CONTEXT_INDEP_ANCHORS
1776 /* Otherwise, depends on what's come before. */
1777 || at_begline_loc_p (pattern
, p
, syntax
))
1787 if ( /* If at end of pattern, it's an operator. */
1789 /* If context independent, it's an operator. */
1790 || syntax
& RE_CONTEXT_INDEP_ANCHORS
1791 /* Otherwise, depends on what's next. */
1792 || at_endline_loc_p (p
, pend
, syntax
))
1802 if ((syntax
& RE_BK_PLUS_QM
)
1803 || (syntax
& RE_LIMITED_OPS
))
1807 /* If there is no previous pattern... */
1810 if (syntax
& RE_CONTEXT_INVALID_OPS
)
1811 FREE_STACK_RETURN (REG_BADRPT
);
1812 else if (!(syntax
& RE_CONTEXT_INDEP_OPS
))
1817 /* Are we optimizing this jump? */
1818 boolean keep_string_p
= false;
1820 /* 1 means zero (many) matches is allowed. */
1821 char zero_times_ok
= 0, many_times_ok
= 0;
1823 /* If there is a sequence of repetition chars, collapse it
1824 down to just one (the right one). We can't combine
1825 interval operators with these because of, e.g., `a{2}*',
1826 which should only match an even number of `a's. */
1830 zero_times_ok
|= c
!= '+';
1831 many_times_ok
|= c
!= '?';
1839 || (!(syntax
& RE_BK_PLUS_QM
) && (c
== '+' || c
== '?')))
1842 else if (syntax
& RE_BK_PLUS_QM
&& c
== '\\')
1844 if (p
== pend
) FREE_STACK_RETURN (REG_EESCAPE
);
1847 if (!(c1
== '+' || c1
== '?'))
1862 /* If we get here, we found another repeat character. */
1865 /* Star, etc. applied to an empty pattern is equivalent
1866 to an empty pattern. */
1870 /* Now we know whether or not zero matches is allowed
1871 and also whether or not two or more matches is allowed. */
1873 { /* More than one repetition is allowed, so put in at the
1874 end a backward relative jump from `b' to before the next
1875 jump we're going to put in below (which jumps from
1876 laststart to after this jump).
1878 But if we are at the `*' in the exact sequence `.*\n',
1879 insert an unconditional jump backwards to the .,
1880 instead of the beginning of the loop. This way we only
1881 push a failure point once, instead of every time
1882 through the loop. */
1883 assert (p
- 1 > pattern
);
1885 /* Allocate the space for the jump. */
1886 GET_BUFFER_SPACE (3);
1888 /* We know we are not at the first character of the pattern,
1889 because laststart was nonzero. And we've already
1890 incremented `p', by the way, to be the character after
1891 the `*'. Do we have to do something analogous here
1892 for null bytes, because of RE_DOT_NOT_NULL? */
1893 if (TRANSLATE (*(p
- 2)) == TRANSLATE ('.')
1895 && p
< pend
&& TRANSLATE (*p
) == TRANSLATE ('\n')
1896 && !(syntax
& RE_DOT_NEWLINE
))
1897 { /* We have .*\n. */
1898 STORE_JUMP (jump
, b
, laststart
);
1899 keep_string_p
= true;
1902 /* Anything else. */
1903 STORE_JUMP (maybe_pop_jump
, b
, laststart
- 3);
1905 /* We've added more stuff to the buffer. */
1909 /* On failure, jump from laststart to b + 3, which will be the
1910 end of the buffer after this jump is inserted. */
1911 GET_BUFFER_SPACE (3);
1912 INSERT_JUMP (keep_string_p
? on_failure_keep_string_jump
1920 /* At least one repetition is required, so insert a
1921 `dummy_failure_jump' before the initial
1922 `on_failure_jump' instruction of the loop. This
1923 effects a skip over that instruction the first time
1924 we hit that loop. */
1925 GET_BUFFER_SPACE (3);
1926 INSERT_JUMP (dummy_failure_jump
, laststart
, laststart
+ 6);
1941 boolean had_char_class
= false;
1943 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
1945 /* Ensure that we have enough space to push a charset: the
1946 opcode, the length count, and the bitset; 34 bytes in all. */
1947 GET_BUFFER_SPACE (34);
1951 /* We test `*p == '^' twice, instead of using an if
1952 statement, so we only need one BUF_PUSH. */
1953 BUF_PUSH (*p
== '^' ? charset_not
: charset
);
1957 /* Remember the first position in the bracket expression. */
1960 /* Push the number of bytes in the bitmap. */
1961 BUF_PUSH ((1 << BYTEWIDTH
) / BYTEWIDTH
);
1963 /* Clear the whole map. */
1964 bzero (b
, (1 << BYTEWIDTH
) / BYTEWIDTH
);
1966 /* charset_not matches newline according to a syntax bit. */
1967 if ((re_opcode_t
) b
[-2] == charset_not
1968 && (syntax
& RE_HAT_LISTS_NOT_NEWLINE
))
1969 SET_LIST_BIT ('\n');
1971 /* Read in characters and ranges, setting map bits. */
1974 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
1978 /* \ might escape characters inside [...] and [^...]. */
1979 if ((syntax
& RE_BACKSLASH_ESCAPE_IN_LISTS
) && c
== '\\')
1981 if (p
== pend
) FREE_STACK_RETURN (REG_EESCAPE
);
1988 /* Could be the end of the bracket expression. If it's
1989 not (i.e., when the bracket expression is `[]' so
1990 far), the ']' character bit gets set way below. */
1991 if (c
== ']' && p
!= p1
+ 1)
1994 /* Look ahead to see if it's a range when the last thing
1995 was a character class. */
1996 if (had_char_class
&& c
== '-' && *p
!= ']')
1997 FREE_STACK_RETURN (REG_ERANGE
);
1999 /* Look ahead to see if it's a range when the last thing
2000 was a character: if this is a hyphen not at the
2001 beginning or the end of a list, then it's the range
2004 && !(p
- 2 >= pattern
&& p
[-2] == '[')
2005 && !(p
- 3 >= pattern
&& p
[-3] == '[' && p
[-2] == '^')
2009 = compile_range (&p
, pend
, translate
, syntax
, b
);
2010 if (ret
!= REG_NOERROR
) FREE_STACK_RETURN (ret
);
2013 else if (p
[0] == '-' && p
[1] != ']')
2014 { /* This handles ranges made up of characters only. */
2017 /* Move past the `-'. */
2020 ret
= compile_range (&p
, pend
, translate
, syntax
, b
);
2021 if (ret
!= REG_NOERROR
) FREE_STACK_RETURN (ret
);
2024 /* See if we're at the beginning of a possible character
2027 else if (syntax
& RE_CHAR_CLASSES
&& c
== '[' && *p
== ':')
2028 { /* Leave room for the null. */
2029 char str
[CHAR_CLASS_MAX_LENGTH
+ 1];
2034 /* If pattern is `[[:'. */
2035 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2040 if (c
== ':' || c
== ']' || p
== pend
2041 || c1
== CHAR_CLASS_MAX_LENGTH
)
2047 /* If isn't a word bracketed by `[:' and:`]':
2048 undo the ending character, the letters, and leave
2049 the leading `:' and `[' (but set bits for them). */
2050 if (c
== ':' && *p
== ']')
2053 boolean is_alnum
= STREQ (str
, "alnum");
2054 boolean is_alpha
= STREQ (str
, "alpha");
2055 boolean is_blank
= STREQ (str
, "blank");
2056 boolean is_cntrl
= STREQ (str
, "cntrl");
2057 boolean is_digit
= STREQ (str
, "digit");
2058 boolean is_graph
= STREQ (str
, "graph");
2059 boolean is_lower
= STREQ (str
, "lower");
2060 boolean is_print
= STREQ (str
, "print");
2061 boolean is_punct
= STREQ (str
, "punct");
2062 boolean is_space
= STREQ (str
, "space");
2063 boolean is_upper
= STREQ (str
, "upper");
2064 boolean is_xdigit
= STREQ (str
, "xdigit");
2066 if (!IS_CHAR_CLASS (str
))
2067 FREE_STACK_RETURN (REG_ECTYPE
);
2069 /* Throw away the ] at the end of the character
2073 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2075 for (ch
= 0; ch
< 1 << BYTEWIDTH
; ch
++)
2077 /* This was split into 3 if's to
2078 avoid an arbitrary limit in some compiler. */
2079 if ( (is_alnum
&& ISALNUM (ch
))
2080 || (is_alpha
&& ISALPHA (ch
))
2081 || (is_blank
&& ISBLANK (ch
))
2082 || (is_cntrl
&& ISCNTRL (ch
)))
2084 if ( (is_digit
&& ISDIGIT (ch
))
2085 || (is_graph
&& ISGRAPH (ch
))
2086 || (is_lower
&& ISLOWER (ch
))
2087 || (is_print
&& ISPRINT (ch
)))
2089 if ( (is_punct
&& ISPUNCT (ch
))
2090 || (is_space
&& ISSPACE (ch
))
2091 || (is_upper
&& ISUPPER (ch
))
2092 || (is_xdigit
&& ISXDIGIT (ch
)))
2095 had_char_class
= true;
2104 had_char_class
= false;
2109 had_char_class
= false;
2114 /* Discard any (non)matching list bytes that are all 0 at the
2115 end of the map. Decrease the map-length byte too. */
2116 while ((int) b
[-1] > 0 && b
[b
[-1] - 1] == 0)
2124 if (syntax
& RE_NO_BK_PARENS
)
2131 if (syntax
& RE_NO_BK_PARENS
)
2138 if (syntax
& RE_NEWLINE_ALT
)
2145 if (syntax
& RE_NO_BK_VBAR
)
2152 if (syntax
& RE_INTERVALS
&& syntax
& RE_NO_BK_BRACES
)
2153 goto handle_interval
;
2159 if (p
== pend
) FREE_STACK_RETURN (REG_EESCAPE
);
2161 /* Do not translate the character after the \, so that we can
2162 distinguish, e.g., \B from \b, even if we normally would
2163 translate, e.g., B to b. */
2169 if (syntax
& RE_NO_BK_PARENS
)
2170 goto normal_backslash
;
2176 if (COMPILE_STACK_FULL
)
2178 RETALLOC (compile_stack
.stack
, compile_stack
.size
<< 1,
2179 compile_stack_elt_t
);
2180 if (compile_stack
.stack
== NULL
) return REG_ESPACE
;
2182 compile_stack
.size
<<= 1;
2185 /* These are the values to restore when we hit end of this
2186 group. They are all relative offsets, so that if the
2187 whole pattern moves because of realloc, they will still
2189 COMPILE_STACK_TOP
.begalt_offset
= begalt
- bufp
->buffer
;
2190 COMPILE_STACK_TOP
.fixup_alt_jump
2191 = fixup_alt_jump
? fixup_alt_jump
- bufp
->buffer
+ 1 : 0;
2192 COMPILE_STACK_TOP
.laststart_offset
= b
- bufp
->buffer
;
2193 COMPILE_STACK_TOP
.regnum
= regnum
;
2195 /* We will eventually replace the 0 with the number of
2196 groups inner to this one. But do not push a
2197 start_memory for groups beyond the last one we can
2198 represent in the compiled pattern. */
2199 if (regnum
<= MAX_REGNUM
)
2201 COMPILE_STACK_TOP
.inner_group_offset
= b
- bufp
->buffer
+ 2;
2202 BUF_PUSH_3 (start_memory
, regnum
, 0);
2205 compile_stack
.avail
++;
2210 /* If we've reached MAX_REGNUM groups, then this open
2211 won't actually generate any code, so we'll have to
2212 clear pending_exact explicitly. */
2218 if (syntax
& RE_NO_BK_PARENS
) goto normal_backslash
;
2220 if (COMPILE_STACK_EMPTY
)
2221 if (syntax
& RE_UNMATCHED_RIGHT_PAREN_ORD
)
2222 goto normal_backslash
;
2224 FREE_STACK_RETURN (REG_ERPAREN
);
2228 { /* Push a dummy failure point at the end of the
2229 alternative for a possible future
2230 `pop_failure_jump' to pop. See comments at
2231 `push_dummy_failure' in `re_match_2'. */
2232 BUF_PUSH (push_dummy_failure
);
2234 /* We allocated space for this jump when we assigned
2235 to `fixup_alt_jump', in the `handle_alt' case below. */
2236 STORE_JUMP (jump_past_alt
, fixup_alt_jump
, b
- 1);
2239 /* See similar code for backslashed left paren above. */
2240 if (COMPILE_STACK_EMPTY
)
2241 if (syntax
& RE_UNMATCHED_RIGHT_PAREN_ORD
)
2244 FREE_STACK_RETURN (REG_ERPAREN
);
2246 /* Since we just checked for an empty stack above, this
2247 ``can't happen''. */
2248 assert (compile_stack
.avail
!= 0);
2250 /* We don't just want to restore into `regnum', because
2251 later groups should continue to be numbered higher,
2252 as in `(ab)c(de)' -- the second group is #2. */
2253 regnum_t this_group_regnum
;
2255 compile_stack
.avail
--;
2256 begalt
= bufp
->buffer
+ COMPILE_STACK_TOP
.begalt_offset
;
2258 = COMPILE_STACK_TOP
.fixup_alt_jump
2259 ? bufp
->buffer
+ COMPILE_STACK_TOP
.fixup_alt_jump
- 1
2261 laststart
= bufp
->buffer
+ COMPILE_STACK_TOP
.laststart_offset
;
2262 this_group_regnum
= COMPILE_STACK_TOP
.regnum
;
2263 /* If we've reached MAX_REGNUM groups, then this open
2264 won't actually generate any code, so we'll have to
2265 clear pending_exact explicitly. */
2268 /* We're at the end of the group, so now we know how many
2269 groups were inside this one. */
2270 if (this_group_regnum
<= MAX_REGNUM
)
2272 unsigned char *inner_group_loc
2273 = bufp
->buffer
+ COMPILE_STACK_TOP
.inner_group_offset
;
2275 *inner_group_loc
= regnum
- this_group_regnum
;
2276 BUF_PUSH_3 (stop_memory
, this_group_regnum
,
2277 regnum
- this_group_regnum
);
2283 case '|': /* `\|'. */
2284 if (syntax
& RE_LIMITED_OPS
|| syntax
& RE_NO_BK_VBAR
)
2285 goto normal_backslash
;
2287 if (syntax
& RE_LIMITED_OPS
)
2290 /* Insert before the previous alternative a jump which
2291 jumps to this alternative if the former fails. */
2292 GET_BUFFER_SPACE (3);
2293 INSERT_JUMP (on_failure_jump
, begalt
, b
+ 6);
2297 /* The alternative before this one has a jump after it
2298 which gets executed if it gets matched. Adjust that
2299 jump so it will jump to this alternative's analogous
2300 jump (put in below, which in turn will jump to the next
2301 (if any) alternative's such jump, etc.). The last such
2302 jump jumps to the correct final destination. A picture:
2308 If we are at `b', then fixup_alt_jump right now points to a
2309 three-byte space after `a'. We'll put in the jump, set
2310 fixup_alt_jump to right after `b', and leave behind three
2311 bytes which we'll fill in when we get to after `c'. */
2314 STORE_JUMP (jump_past_alt
, fixup_alt_jump
, b
);
2316 /* Mark and leave space for a jump after this alternative,
2317 to be filled in later either by next alternative or
2318 when know we're at the end of a series of alternatives. */
2320 GET_BUFFER_SPACE (3);
2329 /* If \{ is a literal. */
2330 if (!(syntax
& RE_INTERVALS
)
2331 /* If we're at `\{' and it's not the open-interval
2333 || ((syntax
& RE_INTERVALS
) && (syntax
& RE_NO_BK_BRACES
))
2334 || (p
- 2 == pattern
&& p
== pend
))
2335 goto normal_backslash
;
2339 /* If got here, then the syntax allows intervals. */
2341 /* At least (most) this many matches must be made. */
2342 int lower_bound
= -1, upper_bound
= -1;
2344 beg_interval
= p
- 1;
2348 if (syntax
& RE_NO_BK_BRACES
)
2349 goto unfetch_interval
;
2351 FREE_STACK_RETURN (REG_EBRACE
);
2354 GET_UNSIGNED_NUMBER (lower_bound
);
2358 GET_UNSIGNED_NUMBER (upper_bound
);
2359 if (upper_bound
< 0) upper_bound
= RE_DUP_MAX
;
2362 /* Interval such as `{1}' => match exactly once. */
2363 upper_bound
= lower_bound
;
2365 if (lower_bound
< 0 || upper_bound
> RE_DUP_MAX
2366 || lower_bound
> upper_bound
)
2368 if (syntax
& RE_NO_BK_BRACES
)
2369 goto unfetch_interval
;
2371 FREE_STACK_RETURN (REG_BADBR
);
2374 if (!(syntax
& RE_NO_BK_BRACES
))
2376 if (c
!= '\\') FREE_STACK_RETURN (REG_EBRACE
);
2383 if (syntax
& RE_NO_BK_BRACES
)
2384 goto unfetch_interval
;
2386 FREE_STACK_RETURN (REG_BADBR
);
2389 /* We just parsed a valid interval. */
2391 /* If it's invalid to have no preceding re. */
2394 if (syntax
& RE_CONTEXT_INVALID_OPS
)
2395 FREE_STACK_RETURN (REG_BADRPT
);
2396 else if (syntax
& RE_CONTEXT_INDEP_OPS
)
2399 goto unfetch_interval
;
2402 /* If the upper bound is zero, don't want to succeed at
2403 all; jump from `laststart' to `b + 3', which will be
2404 the end of the buffer after we insert the jump. */
2405 if (upper_bound
== 0)
2407 GET_BUFFER_SPACE (3);
2408 INSERT_JUMP (jump
, laststart
, b
+ 3);
2412 /* Otherwise, we have a nontrivial interval. When
2413 we're all done, the pattern will look like:
2414 set_number_at <jump count> <upper bound>
2415 set_number_at <succeed_n count> <lower bound>
2416 succeed_n <after jump addr> <succeed_n count>
2418 jump_n <succeed_n addr> <jump count>
2419 (The upper bound and `jump_n' are omitted if
2420 `upper_bound' is 1, though.) */
2422 { /* If the upper bound is > 1, we need to insert
2423 more at the end of the loop. */
2424 unsigned nbytes
= 10 + (upper_bound
> 1) * 10;
2426 GET_BUFFER_SPACE (nbytes
);
2428 /* Initialize lower bound of the `succeed_n', even
2429 though it will be set during matching by its
2430 attendant `set_number_at' (inserted next),
2431 because `re_compile_fastmap' needs to know.
2432 Jump to the `jump_n' we might insert below. */
2433 INSERT_JUMP2 (succeed_n
, laststart
,
2434 b
+ 5 + (upper_bound
> 1) * 5,
2438 /* Code to initialize the lower bound. Insert
2439 before the `succeed_n'. The `5' is the last two
2440 bytes of this `set_number_at', plus 3 bytes of
2441 the following `succeed_n'. */
2442 insert_op2 (set_number_at
, laststart
, 5, lower_bound
, b
);
2445 if (upper_bound
> 1)
2446 { /* More than one repetition is allowed, so
2447 append a backward jump to the `succeed_n'
2448 that starts this interval.
2450 When we've reached this during matching,
2451 we'll have matched the interval once, so
2452 jump back only `upper_bound - 1' times. */
2453 STORE_JUMP2 (jump_n
, b
, laststart
+ 5,
2457 /* The location we want to set is the second
2458 parameter of the `jump_n'; that is `b-2' as
2459 an absolute address. `laststart' will be
2460 the `set_number_at' we're about to insert;
2461 `laststart+3' the number to set, the source
2462 for the relative address. But we are
2463 inserting into the middle of the pattern --
2464 so everything is getting moved up by 5.
2465 Conclusion: (b - 2) - (laststart + 3) + 5,
2466 i.e., b - laststart.
2468 We insert this at the beginning of the loop
2469 so that if we fail during matching, we'll
2470 reinitialize the bounds. */
2471 insert_op2 (set_number_at
, laststart
, b
- laststart
,
2472 upper_bound
- 1, b
);
2477 beg_interval
= NULL
;
2482 /* If an invalid interval, match the characters as literals. */
2483 assert (beg_interval
);
2485 beg_interval
= NULL
;
2487 /* normal_char and normal_backslash need `c'. */
2490 if (!(syntax
& RE_NO_BK_BRACES
))
2492 if (p
> pattern
&& p
[-1] == '\\')
2493 goto normal_backslash
;
2498 /* There is no way to specify the before_dot and after_dot
2499 operators. rms says this is ok. --karl */
2507 BUF_PUSH_2 (syntaxspec
, syntax_spec_code
[c
]);
2513 BUF_PUSH_2 (notsyntaxspec
, syntax_spec_code
[c
]);
2520 BUF_PUSH (wordchar
);
2526 BUF_PUSH (notwordchar
);
2539 BUF_PUSH (wordbound
);
2543 BUF_PUSH (notwordbound
);
2554 case '1': case '2': case '3': case '4': case '5':
2555 case '6': case '7': case '8': case '9':
2556 if (syntax
& RE_NO_BK_REFS
)
2562 FREE_STACK_RETURN (REG_ESUBREG
);
2564 /* Can't back reference to a subexpression if inside of it. */
2565 if (group_in_compile_stack (compile_stack
, c1
))
2569 BUF_PUSH_2 (duplicate
, c1
);
2575 if (syntax
& RE_BK_PLUS_QM
)
2578 goto normal_backslash
;
2582 /* You might think it would be useful for \ to mean
2583 not to translate; but if we don't translate it
2584 it will never match anything. */
2592 /* Expects the character in `c'. */
2594 /* If no exactn currently being built. */
2597 /* If last exactn not at current position. */
2598 || pending_exact
+ *pending_exact
+ 1 != b
2600 /* We have only one byte following the exactn for the count. */
2601 || *pending_exact
== (1 << BYTEWIDTH
) - 1
2603 /* If followed by a repetition operator. */
2604 || *p
== '*' || *p
== '^'
2605 || ((syntax
& RE_BK_PLUS_QM
)
2606 ? *p
== '\\' && (p
[1] == '+' || p
[1] == '?')
2607 : (*p
== '+' || *p
== '?'))
2608 || ((syntax
& RE_INTERVALS
)
2609 && ((syntax
& RE_NO_BK_BRACES
)
2611 : (p
[0] == '\\' && p
[1] == '{'))))
2613 /* Start building a new exactn. */
2617 BUF_PUSH_2 (exactn
, 0);
2618 pending_exact
= b
- 1;
2625 } /* while p != pend */
2628 /* Through the pattern now. */
2631 STORE_JUMP (jump_past_alt
, fixup_alt_jump
, b
);
2633 if (!COMPILE_STACK_EMPTY
)
2634 FREE_STACK_RETURN (REG_EPAREN
);
2636 /* If we don't want backtracking, force success
2637 the first time we reach the end of the compiled pattern. */
2638 if (syntax
& RE_NO_POSIX_BACKTRACKING
)
2641 free (compile_stack
.stack
);
2643 /* We have succeeded; set the length of the buffer. */
2644 bufp
->used
= b
- bufp
->buffer
;
2649 DEBUG_PRINT1 ("\nCompiled pattern: \n");
2650 print_compiled_pattern (bufp
);
2654 #ifndef MATCH_MAY_ALLOCATE
2655 /* Initialize the failure stack to the largest possible stack. This
2656 isn't necessary unless we're trying to avoid calling alloca in
2657 the search and match routines. */
2659 int num_regs
= bufp
->re_nsub
+ 1;
2661 /* Since DOUBLE_FAIL_STACK refuses to double only if the current size
2662 is strictly greater than re_max_failures, the largest possible stack
2663 is 2 * re_max_failures failure points. */
2664 if (fail_stack
.size
< (2 * re_max_failures
* MAX_FAILURE_ITEMS
))
2666 fail_stack
.size
= (2 * re_max_failures
* MAX_FAILURE_ITEMS
);
2669 if (! fail_stack
.stack
)
2671 = (fail_stack_elt_t
*) xmalloc (fail_stack
.size
2672 * sizeof (fail_stack_elt_t
));
2675 = (fail_stack_elt_t
*) xrealloc (fail_stack
.stack
,
2677 * sizeof (fail_stack_elt_t
)));
2678 #else /* not emacs */
2679 if (! fail_stack
.stack
)
2681 = (fail_stack_elt_t
*) malloc (fail_stack
.size
2682 * sizeof (fail_stack_elt_t
));
2685 = (fail_stack_elt_t
*) realloc (fail_stack
.stack
,
2687 * sizeof (fail_stack_elt_t
)));
2688 #endif /* not emacs */
2691 regex_grow_registers (num_regs
);
2693 #endif /* not MATCH_MAY_ALLOCATE */
2696 } /* regex_compile */
2698 /* Subroutines for `regex_compile'. */
2700 /* Store OP at LOC followed by two-byte integer parameter ARG. */
2703 store_op1 (op
, loc
, arg
)
2708 *loc
= (unsigned char) op
;
2709 STORE_NUMBER (loc
+ 1, arg
);
2713 /* Like `store_op1', but for two two-byte parameters ARG1 and ARG2. */
2716 store_op2 (op
, loc
, arg1
, arg2
)
2721 *loc
= (unsigned char) op
;
2722 STORE_NUMBER (loc
+ 1, arg1
);
2723 STORE_NUMBER (loc
+ 3, arg2
);
2727 /* Copy the bytes from LOC to END to open up three bytes of space at LOC
2728 for OP followed by two-byte integer parameter ARG. */
2731 insert_op1 (op
, loc
, arg
, end
)
2737 register unsigned char *pfrom
= end
;
2738 register unsigned char *pto
= end
+ 3;
2740 while (pfrom
!= loc
)
2743 store_op1 (op
, loc
, arg
);
2747 /* Like `insert_op1', but for two two-byte parameters ARG1 and ARG2. */
2750 insert_op2 (op
, loc
, arg1
, arg2
, end
)
2756 register unsigned char *pfrom
= end
;
2757 register unsigned char *pto
= end
+ 5;
2759 while (pfrom
!= loc
)
2762 store_op2 (op
, loc
, arg1
, arg2
);
2766 /* P points to just after a ^ in PATTERN. Return true if that ^ comes
2767 after an alternative or a begin-subexpression. We assume there is at
2768 least one character before the ^. */
2771 at_begline_loc_p (pattern
, p
, syntax
)
2772 const char *pattern
, *p
;
2773 reg_syntax_t syntax
;
2775 const char *prev
= p
- 2;
2776 boolean prev_prev_backslash
= prev
> pattern
&& prev
[-1] == '\\';
2779 /* After a subexpression? */
2780 (*prev
== '(' && (syntax
& RE_NO_BK_PARENS
|| prev_prev_backslash
))
2781 /* After an alternative? */
2782 || (*prev
== '|' && (syntax
& RE_NO_BK_VBAR
|| prev_prev_backslash
));
2786 /* The dual of at_begline_loc_p. This one is for $. We assume there is
2787 at least one character after the $, i.e., `P < PEND'. */
2790 at_endline_loc_p (p
, pend
, syntax
)
2791 const char *p
, *pend
;
2794 const char *next
= p
;
2795 boolean next_backslash
= *next
== '\\';
2796 const char *next_next
= p
+ 1 < pend
? p
+ 1 : 0;
2799 /* Before a subexpression? */
2800 (syntax
& RE_NO_BK_PARENS
? *next
== ')'
2801 : next_backslash
&& next_next
&& *next_next
== ')')
2802 /* Before an alternative? */
2803 || (syntax
& RE_NO_BK_VBAR
? *next
== '|'
2804 : next_backslash
&& next_next
&& *next_next
== '|');
2808 /* Returns true if REGNUM is in one of COMPILE_STACK's elements and
2809 false if it's not. */
2812 group_in_compile_stack (compile_stack
, regnum
)
2813 compile_stack_type compile_stack
;
2818 for (this_element
= compile_stack
.avail
- 1;
2821 if (compile_stack
.stack
[this_element
].regnum
== regnum
)
2828 /* Read the ending character of a range (in a bracket expression) from the
2829 uncompiled pattern *P_PTR (which ends at PEND). We assume the
2830 starting character is in `P[-2]'. (`P[-1]' is the character `-'.)
2831 Then we set the translation of all bits between the starting and
2832 ending characters (inclusive) in the compiled pattern B.
2834 Return an error code.
2836 We use these short variable names so we can use the same macros as
2837 `regex_compile' itself. */
2839 static reg_errcode_t
2840 compile_range (p_ptr
, pend
, translate
, syntax
, b
)
2841 const char **p_ptr
, *pend
;
2842 RE_TRANSLATE_TYPE translate
;
2843 reg_syntax_t syntax
;
2848 const char *p
= *p_ptr
;
2849 int range_start
, range_end
;
2854 /* Even though the pattern is a signed `char *', we need to fetch
2855 with unsigned char *'s; if the high bit of the pattern character
2856 is set, the range endpoints will be negative if we fetch using a
2859 We also want to fetch the endpoints without translating them; the
2860 appropriate translation is done in the bit-setting loop below. */
2861 /* The SVR4 compiler on the 3B2 had trouble with unsigned const char *. */
2862 range_start
= ((const unsigned char *) p
)[-2];
2863 range_end
= ((const unsigned char *) p
)[0];
2865 /* Have to increment the pointer into the pattern string, so the
2866 caller isn't still at the ending character. */
2869 /* If the start is after the end, the range is empty. */
2870 if (range_start
> range_end
)
2871 return syntax
& RE_NO_EMPTY_RANGES
? REG_ERANGE
: REG_NOERROR
;
2873 /* Here we see why `this_char' has to be larger than an `unsigned
2874 char' -- the range is inclusive, so if `range_end' == 0xff
2875 (assuming 8-bit characters), we would otherwise go into an infinite
2876 loop, since all characters <= 0xff. */
2877 for (this_char
= range_start
; this_char
<= range_end
; this_char
++)
2879 SET_LIST_BIT (TRANSLATE (this_char
));
2885 /* re_compile_fastmap computes a ``fastmap'' for the compiled pattern in
2886 BUFP. A fastmap records which of the (1 << BYTEWIDTH) possible
2887 characters can start a string that matches the pattern. This fastmap
2888 is used by re_search to skip quickly over impossible starting points.
2890 The caller must supply the address of a (1 << BYTEWIDTH)-byte data
2891 area as BUFP->fastmap.
2893 We set the `fastmap', `fastmap_accurate', and `can_be_null' fields in
2896 Returns 0 if we succeed, -2 if an internal error. */
2899 re_compile_fastmap (bufp
)
2900 struct re_pattern_buffer
*bufp
;
2903 #ifdef MATCH_MAY_ALLOCATE
2904 fail_stack_type fail_stack
;
2906 #ifndef REGEX_MALLOC
2909 /* We don't push any register information onto the failure stack. */
2910 unsigned num_regs
= 0;
2912 register char *fastmap
= bufp
->fastmap
;
2913 unsigned char *pattern
= bufp
->buffer
;
2914 unsigned long size
= bufp
->used
;
2915 unsigned char *p
= pattern
;
2916 register unsigned char *pend
= pattern
+ size
;
2918 /* This holds the pointer to the failure stack, when
2919 it is allocated relocatably. */
2920 fail_stack_elt_t
*failure_stack_ptr
;
2922 /* Assume that each path through the pattern can be null until
2923 proven otherwise. We set this false at the bottom of switch
2924 statement, to which we get only if a particular path doesn't
2925 match the empty string. */
2926 boolean path_can_be_null
= true;
2928 /* We aren't doing a `succeed_n' to begin with. */
2929 boolean succeed_n_p
= false;
2931 assert (fastmap
!= NULL
&& p
!= NULL
);
2934 bzero (fastmap
, 1 << BYTEWIDTH
); /* Assume nothing's valid. */
2935 bufp
->fastmap_accurate
= 1; /* It will be when we're done. */
2936 bufp
->can_be_null
= 0;
2940 if (p
== pend
|| *p
== succeed
)
2942 /* We have reached the (effective) end of pattern. */
2943 if (!FAIL_STACK_EMPTY ())
2945 bufp
->can_be_null
|= path_can_be_null
;
2947 /* Reset for next path. */
2948 path_can_be_null
= true;
2950 p
= fail_stack
.stack
[--fail_stack
.avail
].pointer
;
2958 /* We should never be about to go beyond the end of the pattern. */
2961 switch (SWITCH_ENUM_CAST ((re_opcode_t
) *p
++))
2964 /* I guess the idea here is to simply not bother with a fastmap
2965 if a backreference is used, since it's too hard to figure out
2966 the fastmap for the corresponding group. Setting
2967 `can_be_null' stops `re_search_2' from using the fastmap, so
2968 that is all we do. */
2970 bufp
->can_be_null
= 1;
2974 /* Following are the cases which match a character. These end
2983 for (j
= *p
++ * BYTEWIDTH
- 1; j
>= 0; j
--)
2984 if (p
[j
/ BYTEWIDTH
] & (1 << (j
% BYTEWIDTH
)))
2990 /* Chars beyond end of map must be allowed. */
2991 for (j
= *p
* BYTEWIDTH
; j
< (1 << BYTEWIDTH
); j
++)
2994 for (j
= *p
++ * BYTEWIDTH
- 1; j
>= 0; j
--)
2995 if (!(p
[j
/ BYTEWIDTH
] & (1 << (j
% BYTEWIDTH
))))
3001 for (j
= 0; j
< (1 << BYTEWIDTH
); j
++)
3002 if (SYNTAX (j
) == Sword
)
3008 for (j
= 0; j
< (1 << BYTEWIDTH
); j
++)
3009 if (SYNTAX (j
) != Sword
)
3016 int fastmap_newline
= fastmap
['\n'];
3018 /* `.' matches anything ... */
3019 for (j
= 0; j
< (1 << BYTEWIDTH
); j
++)
3022 /* ... except perhaps newline. */
3023 if (!(bufp
->syntax
& RE_DOT_NEWLINE
))
3024 fastmap
['\n'] = fastmap_newline
;
3026 /* Return if we have already set `can_be_null'; if we have,
3027 then the fastmap is irrelevant. Something's wrong here. */
3028 else if (bufp
->can_be_null
)
3031 /* Otherwise, have to check alternative paths. */
3038 for (j
= 0; j
< (1 << BYTEWIDTH
); j
++)
3039 if (SYNTAX (j
) == (enum syntaxcode
) k
)
3046 for (j
= 0; j
< (1 << BYTEWIDTH
); j
++)
3047 if (SYNTAX (j
) != (enum syntaxcode
) k
)
3052 /* All cases after this match the empty string. These end with
3072 case push_dummy_failure
:
3077 case pop_failure_jump
:
3078 case maybe_pop_jump
:
3081 case dummy_failure_jump
:
3082 EXTRACT_NUMBER_AND_INCR (j
, p
);
3087 /* Jump backward implies we just went through the body of a
3088 loop and matched nothing. Opcode jumped to should be
3089 `on_failure_jump' or `succeed_n'. Just treat it like an
3090 ordinary jump. For a * loop, it has pushed its failure
3091 point already; if so, discard that as redundant. */
3092 if ((re_opcode_t
) *p
!= on_failure_jump
3093 && (re_opcode_t
) *p
!= succeed_n
)
3097 EXTRACT_NUMBER_AND_INCR (j
, p
);
3100 /* If what's on the stack is where we are now, pop it. */
3101 if (!FAIL_STACK_EMPTY ()
3102 && fail_stack
.stack
[fail_stack
.avail
- 1].pointer
== p
)
3108 case on_failure_jump
:
3109 case on_failure_keep_string_jump
:
3110 handle_on_failure_jump
:
3111 EXTRACT_NUMBER_AND_INCR (j
, p
);
3113 /* For some patterns, e.g., `(a?)?', `p+j' here points to the
3114 end of the pattern. We don't want to push such a point,
3115 since when we restore it above, entering the switch will
3116 increment `p' past the end of the pattern. We don't need
3117 to push such a point since we obviously won't find any more
3118 fastmap entries beyond `pend'. Such a pattern can match
3119 the null string, though. */
3122 if (!PUSH_PATTERN_OP (p
+ j
, fail_stack
))
3124 RESET_FAIL_STACK ();
3129 bufp
->can_be_null
= 1;
3133 EXTRACT_NUMBER_AND_INCR (k
, p
); /* Skip the n. */
3134 succeed_n_p
= false;
3141 /* Get to the number of times to succeed. */
3144 /* Increment p past the n for when k != 0. */
3145 EXTRACT_NUMBER_AND_INCR (k
, p
);
3149 succeed_n_p
= true; /* Spaghetti code alert. */
3150 goto handle_on_failure_jump
;
3167 abort (); /* We have listed all the cases. */
3170 /* Getting here means we have found the possible starting
3171 characters for one path of the pattern -- and that the empty
3172 string does not match. We need not follow this path further.
3173 Instead, look at the next alternative (remembered on the
3174 stack), or quit if no more. The test at the top of the loop
3175 does these things. */
3176 path_can_be_null
= false;
3180 /* Set `can_be_null' for the last path (also the first path, if the
3181 pattern is empty). */
3182 bufp
->can_be_null
|= path_can_be_null
;
3185 RESET_FAIL_STACK ();
3187 } /* re_compile_fastmap */
3189 /* Set REGS to hold NUM_REGS registers, storing them in STARTS and
3190 ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use
3191 this memory for recording register information. STARTS and ENDS
3192 must be allocated using the malloc library routine, and must each
3193 be at least NUM_REGS * sizeof (regoff_t) bytes long.
3195 If NUM_REGS == 0, then subsequent matches should allocate their own
3198 Unless this function is called, the first search or match using
3199 PATTERN_BUFFER will allocate its own register data, without
3200 freeing the old data. */
3203 re_set_registers (bufp
, regs
, num_regs
, starts
, ends
)
3204 struct re_pattern_buffer
*bufp
;
3205 struct re_registers
*regs
;
3207 regoff_t
*starts
, *ends
;
3211 bufp
->regs_allocated
= REGS_REALLOCATE
;
3212 regs
->num_regs
= num_regs
;
3213 regs
->start
= starts
;
3218 bufp
->regs_allocated
= REGS_UNALLOCATED
;
3220 regs
->start
= regs
->end
= (regoff_t
*) 0;
3224 /* Searching routines. */
3226 /* Like re_search_2, below, but only one string is specified, and
3227 doesn't let you say where to stop matching. */
3230 re_search (bufp
, string
, size
, startpos
, range
, regs
)
3231 struct re_pattern_buffer
*bufp
;
3233 int size
, startpos
, range
;
3234 struct re_registers
*regs
;
3236 return re_search_2 (bufp
, NULL
, 0, string
, size
, startpos
, range
,
3241 /* Using the compiled pattern in BUFP->buffer, first tries to match the
3242 virtual concatenation of STRING1 and STRING2, starting first at index
3243 STARTPOS, then at STARTPOS + 1, and so on.
3245 STRING1 and STRING2 have length SIZE1 and SIZE2, respectively.
3247 RANGE is how far to scan while trying to match. RANGE = 0 means try
3248 only at STARTPOS; in general, the last start tried is STARTPOS +
3251 In REGS, return the indices of the virtual concatenation of STRING1
3252 and STRING2 that matched the entire BUFP->buffer and its contained
3255 Do not consider matching one past the index STOP in the virtual
3256 concatenation of STRING1 and STRING2.
3258 We return either the position in the strings at which the match was
3259 found, -1 if no match, or -2 if error (such as failure
3263 re_search_2 (bufp
, string1
, size1
, string2
, size2
, startpos
, range
, regs
, stop
)
3264 struct re_pattern_buffer
*bufp
;
3265 const char *string1
, *string2
;
3269 struct re_registers
*regs
;
3273 register char *fastmap
= bufp
->fastmap
;
3274 register RE_TRANSLATE_TYPE translate
= bufp
->translate
;
3275 int total_size
= size1
+ size2
;
3276 int endpos
= startpos
+ range
;
3278 /* Check for out-of-range STARTPOS. */
3279 if (startpos
< 0 || startpos
> total_size
)
3282 /* Fix up RANGE if it might eventually take us outside
3283 the virtual concatenation of STRING1 and STRING2.
3284 Make sure we won't move STARTPOS below 0 or above TOTAL_SIZE. */
3286 range
= 0 - startpos
;
3287 else if (endpos
> total_size
)
3288 range
= total_size
- startpos
;
3290 /* If the search isn't to be a backwards one, don't waste time in a
3291 search for a pattern that must be anchored. */
3292 if (bufp
->used
> 0 && (re_opcode_t
) bufp
->buffer
[0] == begbuf
&& range
> 0)
3301 /* In a forward search for something that starts with \=.
3302 don't keep searching past point. */
3303 if (bufp
->used
> 0 && (re_opcode_t
) bufp
->buffer
[0] == at_dot
&& range
> 0)
3305 range
= PT
- startpos
;
3311 /* Update the fastmap now if not correct already. */
3312 if (fastmap
&& !bufp
->fastmap_accurate
)
3313 if (re_compile_fastmap (bufp
) == -2)
3316 /* Loop through the string, looking for a place to start matching. */
3319 /* If a fastmap is supplied, skip quickly over characters that
3320 cannot be the start of a match. If the pattern can match the
3321 null string, however, we don't need to skip characters; we want
3322 the first null string. */
3323 if (fastmap
&& startpos
< total_size
&& !bufp
->can_be_null
)
3325 if (range
> 0) /* Searching forwards. */
3327 register const char *d
;
3328 register int lim
= 0;
3331 if (startpos
< size1
&& startpos
+ range
>= size1
)
3332 lim
= range
- (size1
- startpos
);
3334 d
= (startpos
>= size1
? string2
- size1
: string1
) + startpos
;
3336 /* Written out as an if-else to avoid testing `translate'
3340 && !fastmap
[(unsigned char)
3341 translate
[(unsigned char) *d
++]])
3344 while (range
> lim
&& !fastmap
[(unsigned char) *d
++])
3347 startpos
+= irange
- range
;
3349 else /* Searching backwards. */
3351 register char c
= (size1
== 0 || startpos
>= size1
3352 ? string2
[startpos
- size1
]
3353 : string1
[startpos
]);
3355 if (!fastmap
[(unsigned char) TRANSLATE (c
)])
3360 /* If can't match the null string, and that's all we have left, fail. */
3361 if (range
>= 0 && startpos
== total_size
&& fastmap
3362 && !bufp
->can_be_null
)
3365 val
= re_match_2_internal (bufp
, string1
, size1
, string2
, size2
,
3366 startpos
, regs
, stop
);
3367 #ifndef REGEX_MALLOC
3396 /* Declarations and macros for re_match_2. */
3398 static int bcmp_translate ();
3399 static boolean
alt_match_null_string_p (),
3400 common_op_match_null_string_p (),
3401 group_match_null_string_p ();
3403 /* This converts PTR, a pointer into one of the search strings `string1'
3404 and `string2' into an offset from the beginning of that string. */
3405 #define POINTER_TO_OFFSET(ptr) \
3406 (FIRST_STRING_P (ptr) \
3407 ? ((regoff_t) ((ptr) - string1)) \
3408 : ((regoff_t) ((ptr) - string2 + size1)))
3410 /* Macros for dealing with the split strings in re_match_2. */
3412 #define MATCHING_IN_FIRST_STRING (dend == end_match_1)
3414 /* Call before fetching a character with *d. This switches over to
3415 string2 if necessary. */
3416 #define PREFETCH() \
3419 /* End of string2 => fail. */ \
3420 if (dend == end_match_2) \
3422 /* End of string1 => advance to string2. */ \
3424 dend = end_match_2; \
3428 /* Test if at very beginning or at very end of the virtual concatenation
3429 of `string1' and `string2'. If only one string, it's `string2'. */
3430 #define AT_STRINGS_BEG(d) ((d) == (size1 ? string1 : string2) || !size2)
3431 #define AT_STRINGS_END(d) ((d) == end2)
3434 /* Test if D points to a character which is word-constituent. We have
3435 two special cases to check for: if past the end of string1, look at
3436 the first character in string2; and if before the beginning of
3437 string2, look at the last character in string1. */
3438 #define WORDCHAR_P(d) \
3439 (SYNTAX ((d) == end1 ? *string2 \
3440 : (d) == string2 - 1 ? *(end1 - 1) : *(d)) \
3443 /* Test if the character before D and the one at D differ with respect
3444 to being word-constituent. */
3445 #define AT_WORD_BOUNDARY(d) \
3446 (AT_STRINGS_BEG (d) || AT_STRINGS_END (d) \
3447 || WORDCHAR_P (d - 1) != WORDCHAR_P (d))
3450 /* Free everything we malloc. */
3451 #ifdef MATCH_MAY_ALLOCATE
3452 #define FREE_VAR(var) if (var) REGEX_FREE (var); var = NULL
3453 #define FREE_VARIABLES() \
3455 REGEX_FREE_STACK (fail_stack.stack); \
3456 FREE_VAR (regstart); \
3457 FREE_VAR (regend); \
3458 FREE_VAR (old_regstart); \
3459 FREE_VAR (old_regend); \
3460 FREE_VAR (best_regstart); \
3461 FREE_VAR (best_regend); \
3462 FREE_VAR (reg_info); \
3463 FREE_VAR (reg_dummy); \
3464 FREE_VAR (reg_info_dummy); \
3467 #define FREE_VARIABLES() ((void)0) /* Do nothing! But inhibit gcc warning. */
3468 #endif /* not MATCH_MAY_ALLOCATE */
3470 /* These values must meet several constraints. They must not be valid
3471 register values; since we have a limit of 255 registers (because
3472 we use only one byte in the pattern for the register number), we can
3473 use numbers larger than 255. They must differ by 1, because of
3474 NUM_FAILURE_ITEMS above. And the value for the lowest register must
3475 be larger than the value for the highest register, so we do not try
3476 to actually save any registers when none are active. */
3477 #define NO_HIGHEST_ACTIVE_REG (1 << BYTEWIDTH)
3478 #define NO_LOWEST_ACTIVE_REG (NO_HIGHEST_ACTIVE_REG + 1)
3480 /* Matching routines. */
3482 #ifndef emacs /* Emacs never uses this. */
3483 /* re_match is like re_match_2 except it takes only a single string. */
3486 re_match (bufp
, string
, size
, pos
, regs
)
3487 struct re_pattern_buffer
*bufp
;
3490 struct re_registers
*regs
;
3492 int result
= re_match_2_internal (bufp
, NULL
, 0, string
, size
,
3497 #endif /* not emacs */
3500 /* re_match_2 matches the compiled pattern in BUFP against the
3501 the (virtual) concatenation of STRING1 and STRING2 (of length SIZE1
3502 and SIZE2, respectively). We start matching at POS, and stop
3505 If REGS is non-null and the `no_sub' field of BUFP is nonzero, we
3506 store offsets for the substring each group matched in REGS. See the
3507 documentation for exactly how many groups we fill.
3509 We return -1 if no match, -2 if an internal error (such as the
3510 failure stack overflowing). Otherwise, we return the length of the
3511 matched substring. */
3514 re_match_2 (bufp
, string1
, size1
, string2
, size2
, pos
, regs
, stop
)
3515 struct re_pattern_buffer
*bufp
;
3516 const char *string1
, *string2
;
3519 struct re_registers
*regs
;
3522 int result
= re_match_2_internal (bufp
, string1
, size1
, string2
, size2
,
3528 /* This is a separate function so that we can force an alloca cleanup
3531 re_match_2_internal (bufp
, string1
, size1
, string2
, size2
, pos
, regs
, stop
)
3532 struct re_pattern_buffer
*bufp
;
3533 const char *string1
, *string2
;
3536 struct re_registers
*regs
;
3539 /* General temporaries. */
3543 /* Just past the end of the corresponding string. */
3544 const char *end1
, *end2
;
3546 /* Pointers into string1 and string2, just past the last characters in
3547 each to consider matching. */
3548 const char *end_match_1
, *end_match_2
;
3550 /* Where we are in the data, and the end of the current string. */
3551 const char *d
, *dend
;
3553 /* Where we are in the pattern, and the end of the pattern. */
3554 unsigned char *p
= bufp
->buffer
;
3555 register unsigned char *pend
= p
+ bufp
->used
;
3557 /* Mark the opcode just after a start_memory, so we can test for an
3558 empty subpattern when we get to the stop_memory. */
3559 unsigned char *just_past_start_mem
= 0;
3561 /* We use this to map every character in the string. */
3562 RE_TRANSLATE_TYPE translate
= bufp
->translate
;
3564 /* Failure point stack. Each place that can handle a failure further
3565 down the line pushes a failure point on this stack. It consists of
3566 restart, regend, and reg_info for all registers corresponding to
3567 the subexpressions we're currently inside, plus the number of such
3568 registers, and, finally, two char *'s. The first char * is where
3569 to resume scanning the pattern; the second one is where to resume
3570 scanning the strings. If the latter is zero, the failure point is
3571 a ``dummy''; if a failure happens and the failure point is a dummy,
3572 it gets discarded and the next next one is tried. */
3573 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
3574 fail_stack_type fail_stack
;
3577 static unsigned failure_id
= 0;
3578 unsigned nfailure_points_pushed
= 0, nfailure_points_popped
= 0;
3581 /* This holds the pointer to the failure stack, when
3582 it is allocated relocatably. */
3583 fail_stack_elt_t
*failure_stack_ptr
;
3585 /* We fill all the registers internally, independent of what we
3586 return, for use in backreferences. The number here includes
3587 an element for register zero. */
3588 unsigned num_regs
= bufp
->re_nsub
+ 1;
3590 /* The currently active registers. */
3591 unsigned lowest_active_reg
= NO_LOWEST_ACTIVE_REG
;
3592 unsigned highest_active_reg
= NO_HIGHEST_ACTIVE_REG
;
3594 /* Information on the contents of registers. These are pointers into
3595 the input strings; they record just what was matched (on this
3596 attempt) by a subexpression part of the pattern, that is, the
3597 regnum-th regstart pointer points to where in the pattern we began
3598 matching and the regnum-th regend points to right after where we
3599 stopped matching the regnum-th subexpression. (The zeroth register
3600 keeps track of what the whole pattern matches.) */
3601 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
3602 const char **regstart
, **regend
;
3605 /* If a group that's operated upon by a repetition operator fails to
3606 match anything, then the register for its start will need to be
3607 restored because it will have been set to wherever in the string we
3608 are when we last see its open-group operator. Similarly for a
3610 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
3611 const char **old_regstart
, **old_regend
;
3614 /* The is_active field of reg_info helps us keep track of which (possibly
3615 nested) subexpressions we are currently in. The matched_something
3616 field of reg_info[reg_num] helps us tell whether or not we have
3617 matched any of the pattern so far this time through the reg_num-th
3618 subexpression. These two fields get reset each time through any
3619 loop their register is in. */
3620 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
3621 register_info_type
*reg_info
;
3624 /* The following record the register info as found in the above
3625 variables when we find a match better than any we've seen before.
3626 This happens as we backtrack through the failure points, which in
3627 turn happens only if we have not yet matched the entire string. */
3628 unsigned best_regs_set
= false;
3629 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
3630 const char **best_regstart
, **best_regend
;
3633 /* Logically, this is `best_regend[0]'. But we don't want to have to
3634 allocate space for that if we're not allocating space for anything
3635 else (see below). Also, we never need info about register 0 for
3636 any of the other register vectors, and it seems rather a kludge to
3637 treat `best_regend' differently than the rest. So we keep track of
3638 the end of the best match so far in a separate variable. We
3639 initialize this to NULL so that when we backtrack the first time
3640 and need to test it, it's not garbage. */
3641 const char *match_end
= NULL
;
3643 /* This helps SET_REGS_MATCHED avoid doing redundant work. */
3644 int set_regs_matched_done
= 0;
3646 /* Used when we pop values we don't care about. */
3647 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
3648 const char **reg_dummy
;
3649 register_info_type
*reg_info_dummy
;
3653 /* Counts the total number of registers pushed. */
3654 unsigned num_regs_pushed
= 0;
3657 DEBUG_PRINT1 ("\n\nEntering re_match_2.\n");
3661 #ifdef MATCH_MAY_ALLOCATE
3662 /* Do not bother to initialize all the register variables if there are
3663 no groups in the pattern, as it takes a fair amount of time. If
3664 there are groups, we include space for register 0 (the whole
3665 pattern), even though we never use it, since it simplifies the
3666 array indexing. We should fix this. */
3669 regstart
= REGEX_TALLOC (num_regs
, const char *);
3670 regend
= REGEX_TALLOC (num_regs
, const char *);
3671 old_regstart
= REGEX_TALLOC (num_regs
, const char *);
3672 old_regend
= REGEX_TALLOC (num_regs
, const char *);
3673 best_regstart
= REGEX_TALLOC (num_regs
, const char *);
3674 best_regend
= REGEX_TALLOC (num_regs
, const char *);
3675 reg_info
= REGEX_TALLOC (num_regs
, register_info_type
);
3676 reg_dummy
= REGEX_TALLOC (num_regs
, const char *);
3677 reg_info_dummy
= REGEX_TALLOC (num_regs
, register_info_type
);
3679 if (!(regstart
&& regend
&& old_regstart
&& old_regend
&& reg_info
3680 && best_regstart
&& best_regend
&& reg_dummy
&& reg_info_dummy
))
3688 /* We must initialize all our variables to NULL, so that
3689 `FREE_VARIABLES' doesn't try to free them. */
3690 regstart
= regend
= old_regstart
= old_regend
= best_regstart
3691 = best_regend
= reg_dummy
= NULL
;
3692 reg_info
= reg_info_dummy
= (register_info_type
*) NULL
;
3694 #endif /* MATCH_MAY_ALLOCATE */
3696 /* The starting position is bogus. */
3697 if (pos
< 0 || pos
> size1
+ size2
)
3703 /* Initialize subexpression text positions to -1 to mark ones that no
3704 start_memory/stop_memory has been seen for. Also initialize the
3705 register information struct. */
3706 for (mcnt
= 1; mcnt
< num_regs
; mcnt
++)
3708 regstart
[mcnt
] = regend
[mcnt
]
3709 = old_regstart
[mcnt
] = old_regend
[mcnt
] = REG_UNSET_VALUE
;
3711 REG_MATCH_NULL_STRING_P (reg_info
[mcnt
]) = MATCH_NULL_UNSET_VALUE
;
3712 IS_ACTIVE (reg_info
[mcnt
]) = 0;
3713 MATCHED_SOMETHING (reg_info
[mcnt
]) = 0;
3714 EVER_MATCHED_SOMETHING (reg_info
[mcnt
]) = 0;
3717 /* We move `string1' into `string2' if the latter's empty -- but not if
3718 `string1' is null. */
3719 if (size2
== 0 && string1
!= NULL
)
3726 end1
= string1
+ size1
;
3727 end2
= string2
+ size2
;
3729 /* Compute where to stop matching, within the two strings. */
3732 end_match_1
= string1
+ stop
;
3733 end_match_2
= string2
;
3738 end_match_2
= string2
+ stop
- size1
;
3741 /* `p' scans through the pattern as `d' scans through the data.
3742 `dend' is the end of the input string that `d' points within. `d'
3743 is advanced into the following input string whenever necessary, but
3744 this happens before fetching; therefore, at the beginning of the
3745 loop, `d' can be pointing at the end of a string, but it cannot
3747 if (size1
> 0 && pos
<= size1
)
3754 d
= string2
+ pos
- size1
;
3758 DEBUG_PRINT1 ("The compiled pattern is: ");
3759 DEBUG_PRINT_COMPILED_PATTERN (bufp
, p
, pend
);
3760 DEBUG_PRINT1 ("The string to match is: `");
3761 DEBUG_PRINT_DOUBLE_STRING (d
, string1
, size1
, string2
, size2
);
3762 DEBUG_PRINT1 ("'\n");
3764 /* This loops over pattern commands. It exits by returning from the
3765 function if the match is complete, or it drops through if the match
3766 fails at this starting point in the input data. */
3769 DEBUG_PRINT2 ("\n0x%x: ", p
);
3772 { /* End of pattern means we might have succeeded. */
3773 DEBUG_PRINT1 ("end of pattern ... ");
3775 /* If we haven't matched the entire string, and we want the
3776 longest match, try backtracking. */
3777 if (d
!= end_match_2
)
3779 /* 1 if this match ends in the same string (string1 or string2)
3780 as the best previous match. */
3781 boolean same_str_p
= (FIRST_STRING_P (match_end
)
3782 == MATCHING_IN_FIRST_STRING
);
3783 /* 1 if this match is the best seen so far. */
3784 boolean best_match_p
;
3786 /* AIX compiler got confused when this was combined
3787 with the previous declaration. */
3789 best_match_p
= d
> match_end
;
3791 best_match_p
= !MATCHING_IN_FIRST_STRING
;
3793 DEBUG_PRINT1 ("backtracking.\n");
3795 if (!FAIL_STACK_EMPTY ())
3796 { /* More failure points to try. */
3798 /* If exceeds best match so far, save it. */
3799 if (!best_regs_set
|| best_match_p
)
3801 best_regs_set
= true;
3804 DEBUG_PRINT1 ("\nSAVING match as best so far.\n");
3806 for (mcnt
= 1; mcnt
< num_regs
; mcnt
++)
3808 best_regstart
[mcnt
] = regstart
[mcnt
];
3809 best_regend
[mcnt
] = regend
[mcnt
];
3815 /* If no failure points, don't restore garbage. And if
3816 last match is real best match, don't restore second
3818 else if (best_regs_set
&& !best_match_p
)
3821 /* Restore best match. It may happen that `dend ==
3822 end_match_1' while the restored d is in string2.
3823 For example, the pattern `x.*y.*z' against the
3824 strings `x-' and `y-z-', if the two strings are
3825 not consecutive in memory. */
3826 DEBUG_PRINT1 ("Restoring best registers.\n");
3829 dend
= ((d
>= string1
&& d
<= end1
)
3830 ? end_match_1
: end_match_2
);
3832 for (mcnt
= 1; mcnt
< num_regs
; mcnt
++)
3834 regstart
[mcnt
] = best_regstart
[mcnt
];
3835 regend
[mcnt
] = best_regend
[mcnt
];
3838 } /* d != end_match_2 */
3841 DEBUG_PRINT1 ("Accepting match.\n");
3843 /* If caller wants register contents data back, do it. */
3844 if (regs
&& !bufp
->no_sub
)
3846 /* Have the register data arrays been allocated? */
3847 if (bufp
->regs_allocated
== REGS_UNALLOCATED
)
3848 { /* No. So allocate them with malloc. We need one
3849 extra element beyond `num_regs' for the `-1' marker
3851 regs
->num_regs
= MAX (RE_NREGS
, num_regs
+ 1);
3852 regs
->start
= TALLOC (regs
->num_regs
, regoff_t
);
3853 regs
->end
= TALLOC (regs
->num_regs
, regoff_t
);
3854 if (regs
->start
== NULL
|| regs
->end
== NULL
)
3859 bufp
->regs_allocated
= REGS_REALLOCATE
;
3861 else if (bufp
->regs_allocated
== REGS_REALLOCATE
)
3862 { /* Yes. If we need more elements than were already
3863 allocated, reallocate them. If we need fewer, just
3865 if (regs
->num_regs
< num_regs
+ 1)
3867 regs
->num_regs
= num_regs
+ 1;
3868 RETALLOC (regs
->start
, regs
->num_regs
, regoff_t
);
3869 RETALLOC (regs
->end
, regs
->num_regs
, regoff_t
);
3870 if (regs
->start
== NULL
|| regs
->end
== NULL
)
3879 /* These braces fend off a "empty body in an else-statement"
3880 warning under GCC when assert expands to nothing. */
3881 assert (bufp
->regs_allocated
== REGS_FIXED
);
3884 /* Convert the pointer data in `regstart' and `regend' to
3885 indices. Register zero has to be set differently,
3886 since we haven't kept track of any info for it. */
3887 if (regs
->num_regs
> 0)
3889 regs
->start
[0] = pos
;
3890 regs
->end
[0] = (MATCHING_IN_FIRST_STRING
3891 ? ((regoff_t
) (d
- string1
))
3892 : ((regoff_t
) (d
- string2
+ size1
)));
3895 /* Go through the first `min (num_regs, regs->num_regs)'
3896 registers, since that is all we initialized. */
3897 for (mcnt
= 1; mcnt
< MIN (num_regs
, regs
->num_regs
); mcnt
++)
3899 if (REG_UNSET (regstart
[mcnt
]) || REG_UNSET (regend
[mcnt
]))
3900 regs
->start
[mcnt
] = regs
->end
[mcnt
] = -1;
3904 = (regoff_t
) POINTER_TO_OFFSET (regstart
[mcnt
]);
3906 = (regoff_t
) POINTER_TO_OFFSET (regend
[mcnt
]);
3910 /* If the regs structure we return has more elements than
3911 were in the pattern, set the extra elements to -1. If
3912 we (re)allocated the registers, this is the case,
3913 because we always allocate enough to have at least one
3915 for (mcnt
= num_regs
; mcnt
< regs
->num_regs
; mcnt
++)
3916 regs
->start
[mcnt
] = regs
->end
[mcnt
] = -1;
3917 } /* regs && !bufp->no_sub */
3919 DEBUG_PRINT4 ("%u failure points pushed, %u popped (%u remain).\n",
3920 nfailure_points_pushed
, nfailure_points_popped
,
3921 nfailure_points_pushed
- nfailure_points_popped
);
3922 DEBUG_PRINT2 ("%u registers pushed.\n", num_regs_pushed
);
3924 mcnt
= d
- pos
- (MATCHING_IN_FIRST_STRING
3928 DEBUG_PRINT2 ("Returning %d from re_match_2.\n", mcnt
);
3934 /* Otherwise match next pattern command. */
3935 switch (SWITCH_ENUM_CAST ((re_opcode_t
) *p
++))
3937 /* Ignore these. Used to ignore the n of succeed_n's which
3938 currently have n == 0. */
3940 DEBUG_PRINT1 ("EXECUTING no_op.\n");
3944 DEBUG_PRINT1 ("EXECUTING succeed.\n");
3947 /* Match the next n pattern characters exactly. The following
3948 byte in the pattern defines n, and the n bytes after that
3949 are the characters to match. */
3952 DEBUG_PRINT2 ("EXECUTING exactn %d.\n", mcnt
);
3954 /* This is written out as an if-else so we don't waste time
3955 testing `translate' inside the loop. */
3961 if ((unsigned char) translate
[(unsigned char) *d
++]
3962 != (unsigned char) *p
++)
3972 if (*d
++ != (char) *p
++) goto fail
;
3976 SET_REGS_MATCHED ();
3980 /* Match any character except possibly a newline or a null. */
3982 DEBUG_PRINT1 ("EXECUTING anychar.\n");
3986 if ((!(bufp
->syntax
& RE_DOT_NEWLINE
) && TRANSLATE (*d
) == '\n')
3987 || (bufp
->syntax
& RE_DOT_NOT_NULL
&& TRANSLATE (*d
) == '\000'))
3990 SET_REGS_MATCHED ();
3991 DEBUG_PRINT2 (" Matched `%d'.\n", *d
);
3999 register unsigned char c
;
4000 boolean
not = (re_opcode_t
) *(p
- 1) == charset_not
;
4002 DEBUG_PRINT2 ("EXECUTING charset%s.\n", not ? "_not" : "");
4005 c
= TRANSLATE (*d
); /* The character to match. */
4007 /* Cast to `unsigned' instead of `unsigned char' in case the
4008 bit list is a full 32 bytes long. */
4009 if (c
< (unsigned) (*p
* BYTEWIDTH
)
4010 && p
[1 + c
/ BYTEWIDTH
] & (1 << (c
% BYTEWIDTH
)))
4015 if (!not) goto fail
;
4017 SET_REGS_MATCHED ();
4023 /* The beginning of a group is represented by start_memory.
4024 The arguments are the register number in the next byte, and the
4025 number of groups inner to this one in the next. The text
4026 matched within the group is recorded (in the internal
4027 registers data structure) under the register number. */
4029 DEBUG_PRINT3 ("EXECUTING start_memory %d (%d):\n", *p
, p
[1]);
4031 /* Find out if this group can match the empty string. */
4032 p1
= p
; /* To send to group_match_null_string_p. */
4034 if (REG_MATCH_NULL_STRING_P (reg_info
[*p
]) == MATCH_NULL_UNSET_VALUE
)
4035 REG_MATCH_NULL_STRING_P (reg_info
[*p
])
4036 = group_match_null_string_p (&p1
, pend
, reg_info
);
4038 /* Save the position in the string where we were the last time
4039 we were at this open-group operator in case the group is
4040 operated upon by a repetition operator, e.g., with `(a*)*b'
4041 against `ab'; then we want to ignore where we are now in
4042 the string in case this attempt to match fails. */
4043 old_regstart
[*p
] = REG_MATCH_NULL_STRING_P (reg_info
[*p
])
4044 ? REG_UNSET (regstart
[*p
]) ? d
: regstart
[*p
]
4046 DEBUG_PRINT2 (" old_regstart: %d\n",
4047 POINTER_TO_OFFSET (old_regstart
[*p
]));
4050 DEBUG_PRINT2 (" regstart: %d\n", POINTER_TO_OFFSET (regstart
[*p
]));
4052 IS_ACTIVE (reg_info
[*p
]) = 1;
4053 MATCHED_SOMETHING (reg_info
[*p
]) = 0;
4055 /* Clear this whenever we change the register activity status. */
4056 set_regs_matched_done
= 0;
4058 /* This is the new highest active register. */
4059 highest_active_reg
= *p
;
4061 /* If nothing was active before, this is the new lowest active
4063 if (lowest_active_reg
== NO_LOWEST_ACTIVE_REG
)
4064 lowest_active_reg
= *p
;
4066 /* Move past the register number and inner group count. */
4068 just_past_start_mem
= p
;
4073 /* The stop_memory opcode represents the end of a group. Its
4074 arguments are the same as start_memory's: the register
4075 number, and the number of inner groups. */
4077 DEBUG_PRINT3 ("EXECUTING stop_memory %d (%d):\n", *p
, p
[1]);
4079 /* We need to save the string position the last time we were at
4080 this close-group operator in case the group is operated
4081 upon by a repetition operator, e.g., with `((a*)*(b*)*)*'
4082 against `aba'; then we want to ignore where we are now in
4083 the string in case this attempt to match fails. */
4084 old_regend
[*p
] = REG_MATCH_NULL_STRING_P (reg_info
[*p
])
4085 ? REG_UNSET (regend
[*p
]) ? d
: regend
[*p
]
4087 DEBUG_PRINT2 (" old_regend: %d\n",
4088 POINTER_TO_OFFSET (old_regend
[*p
]));
4091 DEBUG_PRINT2 (" regend: %d\n", POINTER_TO_OFFSET (regend
[*p
]));
4093 /* This register isn't active anymore. */
4094 IS_ACTIVE (reg_info
[*p
]) = 0;
4096 /* Clear this whenever we change the register activity status. */
4097 set_regs_matched_done
= 0;
4099 /* If this was the only register active, nothing is active
4101 if (lowest_active_reg
== highest_active_reg
)
4103 lowest_active_reg
= NO_LOWEST_ACTIVE_REG
;
4104 highest_active_reg
= NO_HIGHEST_ACTIVE_REG
;
4107 { /* We must scan for the new highest active register, since
4108 it isn't necessarily one less than now: consider
4109 (a(b)c(d(e)f)g). When group 3 ends, after the f), the
4110 new highest active register is 1. */
4111 unsigned char r
= *p
- 1;
4112 while (r
> 0 && !IS_ACTIVE (reg_info
[r
]))
4115 /* If we end up at register zero, that means that we saved
4116 the registers as the result of an `on_failure_jump', not
4117 a `start_memory', and we jumped to past the innermost
4118 `stop_memory'. For example, in ((.)*) we save
4119 registers 1 and 2 as a result of the *, but when we pop
4120 back to the second ), we are at the stop_memory 1.
4121 Thus, nothing is active. */
4124 lowest_active_reg
= NO_LOWEST_ACTIVE_REG
;
4125 highest_active_reg
= NO_HIGHEST_ACTIVE_REG
;
4128 highest_active_reg
= r
;
4131 /* If just failed to match something this time around with a
4132 group that's operated on by a repetition operator, try to
4133 force exit from the ``loop'', and restore the register
4134 information for this group that we had before trying this
4136 if ((!MATCHED_SOMETHING (reg_info
[*p
])
4137 || just_past_start_mem
== p
- 1)
4140 boolean is_a_jump_n
= false;
4144 switch ((re_opcode_t
) *p1
++)
4148 case pop_failure_jump
:
4149 case maybe_pop_jump
:
4151 case dummy_failure_jump
:
4152 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
4162 /* If the next operation is a jump backwards in the pattern
4163 to an on_failure_jump right before the start_memory
4164 corresponding to this stop_memory, exit from the loop
4165 by forcing a failure after pushing on the stack the
4166 on_failure_jump's jump in the pattern, and d. */
4167 if (mcnt
< 0 && (re_opcode_t
) *p1
== on_failure_jump
4168 && (re_opcode_t
) p1
[3] == start_memory
&& p1
[4] == *p
)
4170 /* If this group ever matched anything, then restore
4171 what its registers were before trying this last
4172 failed match, e.g., with `(a*)*b' against `ab' for
4173 regstart[1], and, e.g., with `((a*)*(b*)*)*'
4174 against `aba' for regend[3].
4176 Also restore the registers for inner groups for,
4177 e.g., `((a*)(b*))*' against `aba' (register 3 would
4178 otherwise get trashed). */
4180 if (EVER_MATCHED_SOMETHING (reg_info
[*p
]))
4184 EVER_MATCHED_SOMETHING (reg_info
[*p
]) = 0;
4186 /* Restore this and inner groups' (if any) registers. */
4187 for (r
= *p
; r
< *p
+ *(p
+ 1); r
++)
4189 regstart
[r
] = old_regstart
[r
];
4191 /* xx why this test? */
4192 if (old_regend
[r
] >= regstart
[r
])
4193 regend
[r
] = old_regend
[r
];
4197 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
4198 PUSH_FAILURE_POINT (p1
+ mcnt
, d
, -2);
4204 /* Move past the register number and the inner group count. */
4209 /* \<digit> has been turned into a `duplicate' command which is
4210 followed by the numeric value of <digit> as the register number. */
4213 register const char *d2
, *dend2
;
4214 int regno
= *p
++; /* Get which register to match against. */
4215 DEBUG_PRINT2 ("EXECUTING duplicate %d.\n", regno
);
4217 /* Can't back reference a group which we've never matched. */
4218 if (REG_UNSET (regstart
[regno
]) || REG_UNSET (regend
[regno
]))
4221 /* Where in input to try to start matching. */
4222 d2
= regstart
[regno
];
4224 /* Where to stop matching; if both the place to start and
4225 the place to stop matching are in the same string, then
4226 set to the place to stop, otherwise, for now have to use
4227 the end of the first string. */
4229 dend2
= ((FIRST_STRING_P (regstart
[regno
])
4230 == FIRST_STRING_P (regend
[regno
]))
4231 ? regend
[regno
] : end_match_1
);
4234 /* If necessary, advance to next segment in register
4238 if (dend2
== end_match_2
) break;
4239 if (dend2
== regend
[regno
]) break;
4241 /* End of string1 => advance to string2. */
4243 dend2
= regend
[regno
];
4245 /* At end of register contents => success */
4246 if (d2
== dend2
) break;
4248 /* If necessary, advance to next segment in data. */
4251 /* How many characters left in this segment to match. */
4254 /* Want how many consecutive characters we can match in
4255 one shot, so, if necessary, adjust the count. */
4256 if (mcnt
> dend2
- d2
)
4259 /* Compare that many; failure if mismatch, else move
4262 ? bcmp_translate (d
, d2
, mcnt
, translate
)
4263 : bcmp (d
, d2
, mcnt
))
4265 d
+= mcnt
, d2
+= mcnt
;
4267 /* Do this because we've match some characters. */
4268 SET_REGS_MATCHED ();
4274 /* begline matches the empty string at the beginning of the string
4275 (unless `not_bol' is set in `bufp'), and, if
4276 `newline_anchor' is set, after newlines. */
4278 DEBUG_PRINT1 ("EXECUTING begline.\n");
4280 if (AT_STRINGS_BEG (d
))
4282 if (!bufp
->not_bol
) break;
4284 else if (d
[-1] == '\n' && bufp
->newline_anchor
)
4288 /* In all other cases, we fail. */
4292 /* endline is the dual of begline. */
4294 DEBUG_PRINT1 ("EXECUTING endline.\n");
4296 if (AT_STRINGS_END (d
))
4298 if (!bufp
->not_eol
) break;
4301 /* We have to ``prefetch'' the next character. */
4302 else if ((d
== end1
? *string2
: *d
) == '\n'
4303 && bufp
->newline_anchor
)
4310 /* Match at the very beginning of the data. */
4312 DEBUG_PRINT1 ("EXECUTING begbuf.\n");
4313 if (AT_STRINGS_BEG (d
))
4318 /* Match at the very end of the data. */
4320 DEBUG_PRINT1 ("EXECUTING endbuf.\n");
4321 if (AT_STRINGS_END (d
))
4326 /* on_failure_keep_string_jump is used to optimize `.*\n'. It
4327 pushes NULL as the value for the string on the stack. Then
4328 `pop_failure_point' will keep the current value for the
4329 string, instead of restoring it. To see why, consider
4330 matching `foo\nbar' against `.*\n'. The .* matches the foo;
4331 then the . fails against the \n. But the next thing we want
4332 to do is match the \n against the \n; if we restored the
4333 string value, we would be back at the foo.
4335 Because this is used only in specific cases, we don't need to
4336 check all the things that `on_failure_jump' does, to make
4337 sure the right things get saved on the stack. Hence we don't
4338 share its code. The only reason to push anything on the
4339 stack at all is that otherwise we would have to change
4340 `anychar's code to do something besides goto fail in this
4341 case; that seems worse than this. */
4342 case on_failure_keep_string_jump
:
4343 DEBUG_PRINT1 ("EXECUTING on_failure_keep_string_jump");
4345 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
4346 DEBUG_PRINT3 (" %d (to 0x%x):\n", mcnt
, p
+ mcnt
);
4348 PUSH_FAILURE_POINT (p
+ mcnt
, NULL
, -2);
4352 /* Uses of on_failure_jump:
4354 Each alternative starts with an on_failure_jump that points
4355 to the beginning of the next alternative. Each alternative
4356 except the last ends with a jump that in effect jumps past
4357 the rest of the alternatives. (They really jump to the
4358 ending jump of the following alternative, because tensioning
4359 these jumps is a hassle.)
4361 Repeats start with an on_failure_jump that points past both
4362 the repetition text and either the following jump or
4363 pop_failure_jump back to this on_failure_jump. */
4364 case on_failure_jump
:
4366 DEBUG_PRINT1 ("EXECUTING on_failure_jump");
4368 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
4369 DEBUG_PRINT3 (" %d (to 0x%x)", mcnt
, p
+ mcnt
);
4371 /* If this on_failure_jump comes right before a group (i.e.,
4372 the original * applied to a group), save the information
4373 for that group and all inner ones, so that if we fail back
4374 to this point, the group's information will be correct.
4375 For example, in \(a*\)*\1, we need the preceding group,
4376 and in \(zz\(a*\)b*\)\2, we need the inner group. */
4378 /* We can't use `p' to check ahead because we push
4379 a failure point to `p + mcnt' after we do this. */
4382 /* We need to skip no_op's before we look for the
4383 start_memory in case this on_failure_jump is happening as
4384 the result of a completed succeed_n, as in \(a\)\{1,3\}b\1
4386 while (p1
< pend
&& (re_opcode_t
) *p1
== no_op
)
4389 if (p1
< pend
&& (re_opcode_t
) *p1
== start_memory
)
4391 /* We have a new highest active register now. This will
4392 get reset at the start_memory we are about to get to,
4393 but we will have saved all the registers relevant to
4394 this repetition op, as described above. */
4395 highest_active_reg
= *(p1
+ 1) + *(p1
+ 2);
4396 if (lowest_active_reg
== NO_LOWEST_ACTIVE_REG
)
4397 lowest_active_reg
= *(p1
+ 1);
4400 DEBUG_PRINT1 (":\n");
4401 PUSH_FAILURE_POINT (p
+ mcnt
, d
, -2);
4405 /* A smart repeat ends with `maybe_pop_jump'.
4406 We change it to either `pop_failure_jump' or `jump'. */
4407 case maybe_pop_jump
:
4408 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
4409 DEBUG_PRINT2 ("EXECUTING maybe_pop_jump %d.\n", mcnt
);
4411 register unsigned char *p2
= p
;
4413 /* Compare the beginning of the repeat with what in the
4414 pattern follows its end. If we can establish that there
4415 is nothing that they would both match, i.e., that we
4416 would have to backtrack because of (as in, e.g., `a*a')
4417 then we can change to pop_failure_jump, because we'll
4418 never have to backtrack.
4420 This is not true in the case of alternatives: in
4421 `(a|ab)*' we do need to backtrack to the `ab' alternative
4422 (e.g., if the string was `ab'). But instead of trying to
4423 detect that here, the alternative has put on a dummy
4424 failure point which is what we will end up popping. */
4426 /* Skip over open/close-group commands.
4427 If what follows this loop is a ...+ construct,
4428 look at what begins its body, since we will have to
4429 match at least one of that. */
4433 && ((re_opcode_t
) *p2
== stop_memory
4434 || (re_opcode_t
) *p2
== start_memory
))
4436 else if (p2
+ 6 < pend
4437 && (re_opcode_t
) *p2
== dummy_failure_jump
)
4444 /* p1[0] ... p1[2] are the `on_failure_jump' corresponding
4445 to the `maybe_finalize_jump' of this case. Examine what
4448 /* If we're at the end of the pattern, we can change. */
4451 /* Consider what happens when matching ":\(.*\)"
4452 against ":/". I don't really understand this code
4454 p
[-3] = (unsigned char) pop_failure_jump
;
4456 (" End of pattern: change to `pop_failure_jump'.\n");
4459 else if ((re_opcode_t
) *p2
== exactn
4460 || (bufp
->newline_anchor
&& (re_opcode_t
) *p2
== endline
))
4462 register unsigned char c
4463 = *p2
== (unsigned char) endline
? '\n' : p2
[2];
4465 if ((re_opcode_t
) p1
[3] == exactn
&& p1
[5] != c
)
4467 p
[-3] = (unsigned char) pop_failure_jump
;
4468 DEBUG_PRINT3 (" %c != %c => pop_failure_jump.\n",
4472 else if ((re_opcode_t
) p1
[3] == charset
4473 || (re_opcode_t
) p1
[3] == charset_not
)
4475 int not = (re_opcode_t
) p1
[3] == charset_not
;
4477 if (c
< (unsigned char) (p1
[4] * BYTEWIDTH
)
4478 && p1
[5 + c
/ BYTEWIDTH
] & (1 << (c
% BYTEWIDTH
)))
4481 /* `not' is equal to 1 if c would match, which means
4482 that we can't change to pop_failure_jump. */
4485 p
[-3] = (unsigned char) pop_failure_jump
;
4486 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
4490 else if ((re_opcode_t
) *p2
== charset
)
4493 register unsigned char c
4494 = *p2
== (unsigned char) endline
? '\n' : p2
[2];
4497 if ((re_opcode_t
) p1
[3] == exactn
4498 && ! ((int) p2
[1] * BYTEWIDTH
> (int) p1
[4]
4499 && (p2
[1 + p1
[4] / BYTEWIDTH
]
4500 & (1 << (p1
[4] % BYTEWIDTH
)))))
4502 p
[-3] = (unsigned char) pop_failure_jump
;
4503 DEBUG_PRINT3 (" %c != %c => pop_failure_jump.\n",
4507 else if ((re_opcode_t
) p1
[3] == charset_not
)
4510 /* We win if the charset_not inside the loop
4511 lists every character listed in the charset after. */
4512 for (idx
= 0; idx
< (int) p2
[1]; idx
++)
4513 if (! (p2
[2 + idx
] == 0
4514 || (idx
< (int) p1
[4]
4515 && ((p2
[2 + idx
] & ~ p1
[5 + idx
]) == 0))))
4520 p
[-3] = (unsigned char) pop_failure_jump
;
4521 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
4524 else if ((re_opcode_t
) p1
[3] == charset
)
4527 /* We win if the charset inside the loop
4528 has no overlap with the one after the loop. */
4530 idx
< (int) p2
[1] && idx
< (int) p1
[4];
4532 if ((p2
[2 + idx
] & p1
[5 + idx
]) != 0)
4535 if (idx
== p2
[1] || idx
== p1
[4])
4537 p
[-3] = (unsigned char) pop_failure_jump
;
4538 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
4543 p
-= 2; /* Point at relative address again. */
4544 if ((re_opcode_t
) p
[-1] != pop_failure_jump
)
4546 p
[-1] = (unsigned char) jump
;
4547 DEBUG_PRINT1 (" Match => jump.\n");
4548 goto unconditional_jump
;
4550 /* Note fall through. */
4553 /* The end of a simple repeat has a pop_failure_jump back to
4554 its matching on_failure_jump, where the latter will push a
4555 failure point. The pop_failure_jump takes off failure
4556 points put on by this pop_failure_jump's matching
4557 on_failure_jump; we got through the pattern to here from the
4558 matching on_failure_jump, so didn't fail. */
4559 case pop_failure_jump
:
4561 /* We need to pass separate storage for the lowest and
4562 highest registers, even though we don't care about the
4563 actual values. Otherwise, we will restore only one
4564 register from the stack, since lowest will == highest in
4565 `pop_failure_point'. */
4566 unsigned dummy_low_reg
, dummy_high_reg
;
4567 unsigned char *pdummy
;
4570 DEBUG_PRINT1 ("EXECUTING pop_failure_jump.\n");
4571 POP_FAILURE_POINT (sdummy
, pdummy
,
4572 dummy_low_reg
, dummy_high_reg
,
4573 reg_dummy
, reg_dummy
, reg_info_dummy
);
4575 /* Note fall through. */
4578 /* Unconditionally jump (without popping any failure points). */
4581 EXTRACT_NUMBER_AND_INCR (mcnt
, p
); /* Get the amount to jump. */
4582 DEBUG_PRINT2 ("EXECUTING jump %d ", mcnt
);
4583 p
+= mcnt
; /* Do the jump. */
4584 DEBUG_PRINT2 ("(to 0x%x).\n", p
);
4588 /* We need this opcode so we can detect where alternatives end
4589 in `group_match_null_string_p' et al. */
4591 DEBUG_PRINT1 ("EXECUTING jump_past_alt.\n");
4592 goto unconditional_jump
;
4595 /* Normally, the on_failure_jump pushes a failure point, which
4596 then gets popped at pop_failure_jump. We will end up at
4597 pop_failure_jump, also, and with a pattern of, say, `a+', we
4598 are skipping over the on_failure_jump, so we have to push
4599 something meaningless for pop_failure_jump to pop. */
4600 case dummy_failure_jump
:
4601 DEBUG_PRINT1 ("EXECUTING dummy_failure_jump.\n");
4602 /* It doesn't matter what we push for the string here. What
4603 the code at `fail' tests is the value for the pattern. */
4604 PUSH_FAILURE_POINT (0, 0, -2);
4605 goto unconditional_jump
;
4608 /* At the end of an alternative, we need to push a dummy failure
4609 point in case we are followed by a `pop_failure_jump', because
4610 we don't want the failure point for the alternative to be
4611 popped. For example, matching `(a|ab)*' against `aab'
4612 requires that we match the `ab' alternative. */
4613 case push_dummy_failure
:
4614 DEBUG_PRINT1 ("EXECUTING push_dummy_failure.\n");
4615 /* See comments just above at `dummy_failure_jump' about the
4617 PUSH_FAILURE_POINT (0, 0, -2);
4620 /* Have to succeed matching what follows at least n times.
4621 After that, handle like `on_failure_jump'. */
4623 EXTRACT_NUMBER (mcnt
, p
+ 2);
4624 DEBUG_PRINT2 ("EXECUTING succeed_n %d.\n", mcnt
);
4627 /* Originally, this is how many times we HAVE to succeed. */
4632 STORE_NUMBER_AND_INCR (p
, mcnt
);
4633 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p
, mcnt
);
4637 DEBUG_PRINT2 (" Setting two bytes from 0x%x to no_op.\n", p
+2);
4638 p
[2] = (unsigned char) no_op
;
4639 p
[3] = (unsigned char) no_op
;
4645 EXTRACT_NUMBER (mcnt
, p
+ 2);
4646 DEBUG_PRINT2 ("EXECUTING jump_n %d.\n", mcnt
);
4648 /* Originally, this is how many times we CAN jump. */
4652 STORE_NUMBER (p
+ 2, mcnt
);
4653 goto unconditional_jump
;
4655 /* If don't have to jump any more, skip over the rest of command. */
4662 DEBUG_PRINT1 ("EXECUTING set_number_at.\n");
4664 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
4666 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
4667 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p1
, mcnt
);
4668 STORE_NUMBER (p1
, mcnt
);
4673 DEBUG_PRINT1 ("EXECUTING wordbound.\n");
4674 if (AT_WORD_BOUNDARY (d
))
4679 DEBUG_PRINT1 ("EXECUTING notwordbound.\n");
4680 if (AT_WORD_BOUNDARY (d
))
4685 DEBUG_PRINT1 ("EXECUTING wordbeg.\n");
4686 if (WORDCHAR_P (d
) && (AT_STRINGS_BEG (d
) || !WORDCHAR_P (d
- 1)))
4691 DEBUG_PRINT1 ("EXECUTING wordend.\n");
4692 if (!AT_STRINGS_BEG (d
) && WORDCHAR_P (d
- 1)
4693 && (!WORDCHAR_P (d
) || AT_STRINGS_END (d
)))
4699 DEBUG_PRINT1 ("EXECUTING before_dot.\n");
4700 if (PTR_CHAR_POS ((unsigned char *) d
) >= point
)
4705 DEBUG_PRINT1 ("EXECUTING at_dot.\n");
4706 if (PTR_CHAR_POS ((unsigned char *) d
) != point
)
4711 DEBUG_PRINT1 ("EXECUTING after_dot.\n");
4712 if (PTR_CHAR_POS ((unsigned char *) d
) <= point
)
4717 DEBUG_PRINT2 ("EXECUTING syntaxspec %d.\n", mcnt
);
4722 DEBUG_PRINT1 ("EXECUTING Emacs wordchar.\n");
4726 /* Can't use *d++ here; SYNTAX may be an unsafe macro. */
4728 if (SYNTAX (d
[-1]) != (enum syntaxcode
) mcnt
)
4730 SET_REGS_MATCHED ();
4734 DEBUG_PRINT2 ("EXECUTING notsyntaxspec %d.\n", mcnt
);
4736 goto matchnotsyntax
;
4739 DEBUG_PRINT1 ("EXECUTING Emacs notwordchar.\n");
4743 /* Can't use *d++ here; SYNTAX may be an unsafe macro. */
4745 if (SYNTAX (d
[-1]) == (enum syntaxcode
) mcnt
)
4747 SET_REGS_MATCHED ();
4750 #else /* not emacs */
4752 DEBUG_PRINT1 ("EXECUTING non-Emacs wordchar.\n");
4754 if (!WORDCHAR_P (d
))
4756 SET_REGS_MATCHED ();
4761 DEBUG_PRINT1 ("EXECUTING non-Emacs notwordchar.\n");
4765 SET_REGS_MATCHED ();
4768 #endif /* not emacs */
4773 continue; /* Successfully executed one pattern command; keep going. */
4776 /* We goto here if a matching operation fails. */
4778 if (!FAIL_STACK_EMPTY ())
4779 { /* A restart point is known. Restore to that state. */
4780 DEBUG_PRINT1 ("\nFAIL:\n");
4781 POP_FAILURE_POINT (d
, p
,
4782 lowest_active_reg
, highest_active_reg
,
4783 regstart
, regend
, reg_info
);
4785 /* If this failure point is a dummy, try the next one. */
4789 /* If we failed to the end of the pattern, don't examine *p. */
4793 boolean is_a_jump_n
= false;
4795 /* If failed to a backwards jump that's part of a repetition
4796 loop, need to pop this failure point and use the next one. */
4797 switch ((re_opcode_t
) *p
)
4801 case maybe_pop_jump
:
4802 case pop_failure_jump
:
4805 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
4808 if ((is_a_jump_n
&& (re_opcode_t
) *p1
== succeed_n
)
4810 && (re_opcode_t
) *p1
== on_failure_jump
))
4818 if (d
>= string1
&& d
<= end1
)
4822 break; /* Matching at this starting point really fails. */
4826 goto restore_best_regs
;
4830 return -1; /* Failure to match. */
4833 /* Subroutine definitions for re_match_2. */
4836 /* We are passed P pointing to a register number after a start_memory.
4838 Return true if the pattern up to the corresponding stop_memory can
4839 match the empty string, and false otherwise.
4841 If we find the matching stop_memory, sets P to point to one past its number.
4842 Otherwise, sets P to an undefined byte less than or equal to END.
4844 We don't handle duplicates properly (yet). */
4847 group_match_null_string_p (p
, end
, reg_info
)
4848 unsigned char **p
, *end
;
4849 register_info_type
*reg_info
;
4852 /* Point to after the args to the start_memory. */
4853 unsigned char *p1
= *p
+ 2;
4857 /* Skip over opcodes that can match nothing, and return true or
4858 false, as appropriate, when we get to one that can't, or to the
4859 matching stop_memory. */
4861 switch ((re_opcode_t
) *p1
)
4863 /* Could be either a loop or a series of alternatives. */
4864 case on_failure_jump
:
4866 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
4868 /* If the next operation is not a jump backwards in the
4873 /* Go through the on_failure_jumps of the alternatives,
4874 seeing if any of the alternatives cannot match nothing.
4875 The last alternative starts with only a jump,
4876 whereas the rest start with on_failure_jump and end
4877 with a jump, e.g., here is the pattern for `a|b|c':
4879 /on_failure_jump/0/6/exactn/1/a/jump_past_alt/0/6
4880 /on_failure_jump/0/6/exactn/1/b/jump_past_alt/0/3
4883 So, we have to first go through the first (n-1)
4884 alternatives and then deal with the last one separately. */
4887 /* Deal with the first (n-1) alternatives, which start
4888 with an on_failure_jump (see above) that jumps to right
4889 past a jump_past_alt. */
4891 while ((re_opcode_t
) p1
[mcnt
-3] == jump_past_alt
)
4893 /* `mcnt' holds how many bytes long the alternative
4894 is, including the ending `jump_past_alt' and
4897 if (!alt_match_null_string_p (p1
, p1
+ mcnt
- 3,
4901 /* Move to right after this alternative, including the
4905 /* Break if it's the beginning of an n-th alternative
4906 that doesn't begin with an on_failure_jump. */
4907 if ((re_opcode_t
) *p1
!= on_failure_jump
)
4910 /* Still have to check that it's not an n-th
4911 alternative that starts with an on_failure_jump. */
4913 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
4914 if ((re_opcode_t
) p1
[mcnt
-3] != jump_past_alt
)
4916 /* Get to the beginning of the n-th alternative. */
4922 /* Deal with the last alternative: go back and get number
4923 of the `jump_past_alt' just before it. `mcnt' contains
4924 the length of the alternative. */
4925 EXTRACT_NUMBER (mcnt
, p1
- 2);
4927 if (!alt_match_null_string_p (p1
, p1
+ mcnt
, reg_info
))
4930 p1
+= mcnt
; /* Get past the n-th alternative. */
4936 assert (p1
[1] == **p
);
4942 if (!common_op_match_null_string_p (&p1
, end
, reg_info
))
4945 } /* while p1 < end */
4948 } /* group_match_null_string_p */
4951 /* Similar to group_match_null_string_p, but doesn't deal with alternatives:
4952 It expects P to be the first byte of a single alternative and END one
4953 byte past the last. The alternative can contain groups. */
4956 alt_match_null_string_p (p
, end
, reg_info
)
4957 unsigned char *p
, *end
;
4958 register_info_type
*reg_info
;
4961 unsigned char *p1
= p
;
4965 /* Skip over opcodes that can match nothing, and break when we get
4966 to one that can't. */
4968 switch ((re_opcode_t
) *p1
)
4971 case on_failure_jump
:
4973 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
4978 if (!common_op_match_null_string_p (&p1
, end
, reg_info
))
4981 } /* while p1 < end */
4984 } /* alt_match_null_string_p */
4987 /* Deals with the ops common to group_match_null_string_p and
4988 alt_match_null_string_p.
4990 Sets P to one after the op and its arguments, if any. */
4993 common_op_match_null_string_p (p
, end
, reg_info
)
4994 unsigned char **p
, *end
;
4995 register_info_type
*reg_info
;
5000 unsigned char *p1
= *p
;
5002 switch ((re_opcode_t
) *p1
++)
5022 assert (reg_no
> 0 && reg_no
<= MAX_REGNUM
);
5023 ret
= group_match_null_string_p (&p1
, end
, reg_info
);
5025 /* Have to set this here in case we're checking a group which
5026 contains a group and a back reference to it. */
5028 if (REG_MATCH_NULL_STRING_P (reg_info
[reg_no
]) == MATCH_NULL_UNSET_VALUE
)
5029 REG_MATCH_NULL_STRING_P (reg_info
[reg_no
]) = ret
;
5035 /* If this is an optimized succeed_n for zero times, make the jump. */
5037 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
5045 /* Get to the number of times to succeed. */
5047 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
5052 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
5060 if (!REG_MATCH_NULL_STRING_P (reg_info
[*p1
]))
5068 /* All other opcodes mean we cannot match the empty string. */
5074 } /* common_op_match_null_string_p */
5077 /* Return zero if TRANSLATE[S1] and TRANSLATE[S2] are identical for LEN
5078 bytes; nonzero otherwise. */
5081 bcmp_translate (s1
, s2
, len
, translate
)
5082 unsigned char *s1
, *s2
;
5084 RE_TRANSLATE_TYPE translate
;
5086 register unsigned char *p1
= s1
, *p2
= s2
;
5089 if (translate
[*p1
++] != translate
[*p2
++]) return 1;
5095 /* Entry points for GNU code. */
5097 /* re_compile_pattern is the GNU regular expression compiler: it
5098 compiles PATTERN (of length SIZE) and puts the result in BUFP.
5099 Returns 0 if the pattern was valid, otherwise an error string.
5101 Assumes the `allocated' (and perhaps `buffer') and `translate' fields
5102 are set in BUFP on entry.
5104 We call regex_compile to do the actual compilation. */
5107 re_compile_pattern (pattern
, length
, bufp
)
5108 const char *pattern
;
5110 struct re_pattern_buffer
*bufp
;
5114 /* GNU code is written to assume at least RE_NREGS registers will be set
5115 (and at least one extra will be -1). */
5116 bufp
->regs_allocated
= REGS_UNALLOCATED
;
5118 /* And GNU code determines whether or not to get register information
5119 by passing null for the REGS argument to re_match, etc., not by
5123 /* Match anchors at newline. */
5124 bufp
->newline_anchor
= 1;
5126 ret
= regex_compile (pattern
, length
, re_syntax_options
, bufp
);
5130 return gettext (re_error_msgid
[(int) ret
]);
5133 /* Entry points compatible with 4.2 BSD regex library. We don't define
5134 them unless specifically requested. */
5136 #ifdef _REGEX_RE_COMP
5138 /* BSD has one and only one pattern buffer. */
5139 static struct re_pattern_buffer re_comp_buf
;
5149 if (!re_comp_buf
.buffer
)
5150 return gettext ("No previous regular expression");
5154 if (!re_comp_buf
.buffer
)
5156 re_comp_buf
.buffer
= (unsigned char *) malloc (200);
5157 if (re_comp_buf
.buffer
== NULL
)
5158 return gettext (re_error_msgid
[(int) REG_ESPACE
]);
5159 re_comp_buf
.allocated
= 200;
5161 re_comp_buf
.fastmap
= (char *) malloc (1 << BYTEWIDTH
);
5162 if (re_comp_buf
.fastmap
== NULL
)
5163 return gettext (re_error_msgid
[(int) REG_ESPACE
]);
5166 /* Since `re_exec' always passes NULL for the `regs' argument, we
5167 don't need to initialize the pattern buffer fields which affect it. */
5169 /* Match anchors at newlines. */
5170 re_comp_buf
.newline_anchor
= 1;
5172 ret
= regex_compile (s
, strlen (s
), re_syntax_options
, &re_comp_buf
);
5177 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
5178 return (char *) gettext (re_error_msgid
[(int) ret
]);
5186 const int len
= strlen (s
);
5188 0 <= re_search (&re_comp_buf
, s
, len
, 0, len
, (struct re_registers
*) 0);
5190 #endif /* _REGEX_RE_COMP */
5192 /* POSIX.2 functions. Don't define these for Emacs. */
5196 /* regcomp takes a regular expression as a string and compiles it.
5198 PREG is a regex_t *. We do not expect any fields to be initialized,
5199 since POSIX says we shouldn't. Thus, we set
5201 `buffer' to the compiled pattern;
5202 `used' to the length of the compiled pattern;
5203 `syntax' to RE_SYNTAX_POSIX_EXTENDED if the
5204 REG_EXTENDED bit in CFLAGS is set; otherwise, to
5205 RE_SYNTAX_POSIX_BASIC;
5206 `newline_anchor' to REG_NEWLINE being set in CFLAGS;
5207 `fastmap' and `fastmap_accurate' to zero;
5208 `re_nsub' to the number of subexpressions in PATTERN.
5210 PATTERN is the address of the pattern string.
5212 CFLAGS is a series of bits which affect compilation.
5214 If REG_EXTENDED is set, we use POSIX extended syntax; otherwise, we
5215 use POSIX basic syntax.
5217 If REG_NEWLINE is set, then . and [^...] don't match newline.
5218 Also, regexec will try a match beginning after every newline.
5220 If REG_ICASE is set, then we considers upper- and lowercase
5221 versions of letters to be equivalent when matching.
5223 If REG_NOSUB is set, then when PREG is passed to regexec, that
5224 routine will report only success or failure, and nothing about the
5227 It returns 0 if it succeeds, nonzero if it doesn't. (See regex.h for
5228 the return codes and their meanings.) */
5231 regcomp (preg
, pattern
, cflags
)
5233 const char *pattern
;
5238 = (cflags
& REG_EXTENDED
) ?
5239 RE_SYNTAX_POSIX_EXTENDED
: RE_SYNTAX_POSIX_BASIC
;
5241 /* regex_compile will allocate the space for the compiled pattern. */
5243 preg
->allocated
= 0;
5246 /* Don't bother to use a fastmap when searching. This simplifies the
5247 REG_NEWLINE case: if we used a fastmap, we'd have to put all the
5248 characters after newlines into the fastmap. This way, we just try
5252 if (cflags
& REG_ICASE
)
5257 = (RE_TRANSLATE_TYPE
) malloc (CHAR_SET_SIZE
5258 * sizeof (*(RE_TRANSLATE_TYPE
)0));
5259 if (preg
->translate
== NULL
)
5260 return (int) REG_ESPACE
;
5262 /* Map uppercase characters to corresponding lowercase ones. */
5263 for (i
= 0; i
< CHAR_SET_SIZE
; i
++)
5264 preg
->translate
[i
] = ISUPPER (i
) ? tolower (i
) : i
;
5267 preg
->translate
= NULL
;
5269 /* If REG_NEWLINE is set, newlines are treated differently. */
5270 if (cflags
& REG_NEWLINE
)
5271 { /* REG_NEWLINE implies neither . nor [^...] match newline. */
5272 syntax
&= ~RE_DOT_NEWLINE
;
5273 syntax
|= RE_HAT_LISTS_NOT_NEWLINE
;
5274 /* It also changes the matching behavior. */
5275 preg
->newline_anchor
= 1;
5278 preg
->newline_anchor
= 0;
5280 preg
->no_sub
= !!(cflags
& REG_NOSUB
);
5282 /* POSIX says a null character in the pattern terminates it, so we
5283 can use strlen here in compiling the pattern. */
5284 ret
= regex_compile (pattern
, strlen (pattern
), syntax
, preg
);
5286 /* POSIX doesn't distinguish between an unmatched open-group and an
5287 unmatched close-group: both are REG_EPAREN. */
5288 if (ret
== REG_ERPAREN
) ret
= REG_EPAREN
;
5294 /* regexec searches for a given pattern, specified by PREG, in the
5297 If NMATCH is zero or REG_NOSUB was set in the cflags argument to
5298 `regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at
5299 least NMATCH elements, and we set them to the offsets of the
5300 corresponding matched substrings.
5302 EFLAGS specifies `execution flags' which affect matching: if
5303 REG_NOTBOL is set, then ^ does not match at the beginning of the
5304 string; if REG_NOTEOL is set, then $ does not match at the end.
5306 We return 0 if we find a match and REG_NOMATCH if not. */
5309 regexec (preg
, string
, nmatch
, pmatch
, eflags
)
5310 const regex_t
*preg
;
5313 regmatch_t pmatch
[];
5317 struct re_registers regs
;
5318 regex_t private_preg
;
5319 int len
= strlen (string
);
5320 boolean want_reg_info
= !preg
->no_sub
&& nmatch
> 0;
5322 private_preg
= *preg
;
5324 private_preg
.not_bol
= !!(eflags
& REG_NOTBOL
);
5325 private_preg
.not_eol
= !!(eflags
& REG_NOTEOL
);
5327 /* The user has told us exactly how many registers to return
5328 information about, via `nmatch'. We have to pass that on to the
5329 matching routines. */
5330 private_preg
.regs_allocated
= REGS_FIXED
;
5334 regs
.num_regs
= nmatch
;
5335 regs
.start
= TALLOC (nmatch
, regoff_t
);
5336 regs
.end
= TALLOC (nmatch
, regoff_t
);
5337 if (regs
.start
== NULL
|| regs
.end
== NULL
)
5338 return (int) REG_NOMATCH
;
5341 /* Perform the searching operation. */
5342 ret
= re_search (&private_preg
, string
, len
,
5343 /* start: */ 0, /* range: */ len
,
5344 want_reg_info
? ®s
: (struct re_registers
*) 0);
5346 /* Copy the register information to the POSIX structure. */
5353 for (r
= 0; r
< nmatch
; r
++)
5355 pmatch
[r
].rm_so
= regs
.start
[r
];
5356 pmatch
[r
].rm_eo
= regs
.end
[r
];
5360 /* If we needed the temporary register info, free the space now. */
5365 /* We want zero return to mean success, unlike `re_search'. */
5366 return ret
>= 0 ? (int) REG_NOERROR
: (int) REG_NOMATCH
;
5370 /* Returns a message corresponding to an error code, ERRCODE, returned
5371 from either regcomp or regexec. We don't use PREG here. */
5374 regerror (errcode
, preg
, errbuf
, errbuf_size
)
5376 const regex_t
*preg
;
5384 || errcode
>= (sizeof (re_error_msgid
) / sizeof (re_error_msgid
[0])))
5385 /* Only error codes returned by the rest of the code should be passed
5386 to this routine. If we are given anything else, or if other regex
5387 code generates an invalid error code, then the program has a bug.
5388 Dump core so we can fix it. */
5391 msg
= gettext (re_error_msgid
[errcode
]);
5393 msg_size
= strlen (msg
) + 1; /* Includes the null. */
5395 if (errbuf_size
!= 0)
5397 if (msg_size
> errbuf_size
)
5399 strncpy (errbuf
, msg
, errbuf_size
- 1);
5400 errbuf
[errbuf_size
- 1] = 0;
5403 strcpy (errbuf
, msg
);
5410 /* Free dynamically allocated space used by PREG. */
5416 if (preg
->buffer
!= NULL
)
5417 free (preg
->buffer
);
5418 preg
->buffer
= NULL
;
5420 preg
->allocated
= 0;
5423 if (preg
->fastmap
!= NULL
)
5424 free (preg
->fastmap
);
5425 preg
->fastmap
= NULL
;
5426 preg
->fastmap_accurate
= 0;
5428 if (preg
->translate
!= NULL
)
5429 free (preg
->translate
);
5430 preg
->translate
= NULL
;
5433 #endif /* not emacs */
5437 make-backup-files: t
5439 trim-versions-without-asking: nil