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, 1996 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)
38 /* We need this for `regex.h', and perhaps for the Emacs include files. */
39 #include <sys/types.h>
41 /* This is for other GNU distributions with internationalized messages. */
42 #if HAVE_LIBINTL_H || defined (_LIBC)
45 # define gettext(msgid) (msgid)
49 /* This define is so xgettext can find the internationalizable
51 #define gettext_noop(String) String
54 /* The `emacs' switch turns on certain matching commands
55 that make sense only in Emacs. */
64 /* If we are not linking with Emacs proper,
65 we can't use the relocating allocator
66 even if config.h says that we can. */
69 #if defined (STDC_HEADERS) || defined (_LIBC)
76 /* When used in Emacs's lib-src, we need to get bzero and bcopy somehow.
77 If nothing else has been done, use the method below. */
78 #ifdef INHIBIT_STRING_HEADER
79 #if !(defined (HAVE_BZERO) && defined (HAVE_BCOPY))
80 #if !defined (bzero) && !defined (bcopy)
81 #undef INHIBIT_STRING_HEADER
86 /* This is the normal way of making sure we have a bcopy and a bzero.
87 This is used in most programs--a few other programs avoid this
88 by defining INHIBIT_STRING_HEADER. */
89 #ifndef INHIBIT_STRING_HEADER
90 #if defined (HAVE_STRING_H) || defined (STDC_HEADERS) || defined (_LIBC)
93 #define bcmp(s1, s2, n) memcmp ((s1), (s2), (n))
96 #define bcopy(s, d, n) memcpy ((d), (s), (n))
99 #define bzero(s, n) memset ((s), 0, (n))
106 /* Define the syntax stuff for \<, \>, etc. */
108 /* This must be nonzero for the wordchar and notwordchar pattern
109 commands in re_match_2. */
114 #ifdef SWITCH_ENUM_BUG
115 #define SWITCH_ENUM_CAST(x) ((int)(x))
117 #define SWITCH_ENUM_CAST(x) (x)
122 extern char *re_syntax_table
;
124 #else /* not SYNTAX_TABLE */
126 /* How many characters in the character set. */
127 #define CHAR_SET_SIZE 256
129 static char re_syntax_table
[CHAR_SET_SIZE
];
140 bzero (re_syntax_table
, sizeof re_syntax_table
);
142 for (c
= 'a'; c
<= 'z'; c
++)
143 re_syntax_table
[c
] = Sword
;
145 for (c
= 'A'; c
<= 'Z'; c
++)
146 re_syntax_table
[c
] = Sword
;
148 for (c
= '0'; c
<= '9'; c
++)
149 re_syntax_table
[c
] = Sword
;
151 re_syntax_table
['_'] = Sword
;
156 #endif /* not SYNTAX_TABLE */
158 #define SYNTAX(c) re_syntax_table[c]
160 #endif /* not emacs */
162 /* Get the interface, including the syntax bits. */
165 /* isalpha etc. are used for the character classes. */
168 /* Jim Meyering writes:
170 "... Some ctype macros are valid only for character codes that
171 isascii says are ASCII (SGI's IRIX-4.0.5 is one such system --when
172 using /bin/cc or gcc but without giving an ansi option). So, all
173 ctype uses should be through macros like ISPRINT... If
174 STDC_HEADERS is defined, then autoconf has verified that the ctype
175 macros don't need to be guarded with references to isascii. ...
176 Defining isascii to 1 should let any compiler worth its salt
177 eliminate the && through constant folding." */
179 #if defined (STDC_HEADERS) || (!defined (isascii) && !defined (HAVE_ISASCII))
182 #define ISASCII(c) isascii(c)
186 #define ISBLANK(c) (ISASCII (c) && isblank (c))
188 #define ISBLANK(c) ((c) == ' ' || (c) == '\t')
191 #define ISGRAPH(c) (ISASCII (c) && isgraph (c))
193 #define ISGRAPH(c) (ISASCII (c) && isprint (c) && !isspace (c))
196 #define ISPRINT(c) (ISASCII (c) && isprint (c))
197 #define ISDIGIT(c) (ISASCII (c) && isdigit (c))
198 #define ISALNUM(c) (ISASCII (c) && isalnum (c))
199 #define ISALPHA(c) (ISASCII (c) && isalpha (c))
200 #define ISCNTRL(c) (ISASCII (c) && iscntrl (c))
201 #define ISLOWER(c) (ISASCII (c) && islower (c))
202 #define ISPUNCT(c) (ISASCII (c) && ispunct (c))
203 #define ISSPACE(c) (ISASCII (c) && isspace (c))
204 #define ISUPPER(c) (ISASCII (c) && isupper (c))
205 #define ISXDIGIT(c) (ISASCII (c) && isxdigit (c))
208 #define NULL (void *)0
211 /* We remove any previous definition of `SIGN_EXTEND_CHAR',
212 since ours (we hope) works properly with all combinations of
213 machines, compilers, `char' and `unsigned char' argument types.
214 (Per Bothner suggested the basic approach.) */
215 #undef SIGN_EXTEND_CHAR
217 #define SIGN_EXTEND_CHAR(c) ((signed char) (c))
218 #else /* not __STDC__ */
219 /* As in Harbison and Steele. */
220 #define SIGN_EXTEND_CHAR(c) ((((unsigned char) (c)) ^ 128) - 128)
223 /* Should we use malloc or alloca? If REGEX_MALLOC is not defined, we
224 use `alloca' instead of `malloc'. This is because using malloc in
225 re_search* or re_match* could cause memory leaks when C-g is used in
226 Emacs; also, malloc is slower and causes storage fragmentation. On
227 the other hand, malloc is more portable, and easier to debug.
229 Because we sometimes use alloca, some routines have to be macros,
230 not functions -- `alloca'-allocated space disappears at the end of the
231 function it is called in. */
235 #define REGEX_ALLOCATE malloc
236 #define REGEX_REALLOCATE(source, osize, nsize) realloc (source, nsize)
237 #define REGEX_FREE free
239 #else /* not REGEX_MALLOC */
241 /* Emacs already defines alloca, sometimes. */
244 /* Make alloca work the best possible way. */
246 #define alloca __builtin_alloca
247 #else /* not __GNUC__ */
250 #else /* not __GNUC__ or HAVE_ALLOCA_H */
251 #if 0 /* It is a bad idea to declare alloca. We always cast the result. */
252 #ifndef _AIX /* Already did AIX, up at the top. */
254 #endif /* not _AIX */
256 #endif /* not HAVE_ALLOCA_H */
257 #endif /* not __GNUC__ */
259 #endif /* not alloca */
261 #define REGEX_ALLOCATE alloca
263 /* Assumes a `char *destination' variable. */
264 #define REGEX_REALLOCATE(source, osize, nsize) \
265 (destination = (char *) alloca (nsize), \
266 bcopy (source, destination, osize), \
269 /* No need to do anything to free, after alloca. */
270 #define REGEX_FREE(arg) ((void)0) /* Do nothing! But inhibit gcc warning. */
272 #endif /* not REGEX_MALLOC */
274 /* Define how to allocate the failure stack. */
276 #if defined (REL_ALLOC) && defined (REGEX_MALLOC)
278 #define REGEX_ALLOCATE_STACK(size) \
279 r_alloc (&failure_stack_ptr, (size))
280 #define REGEX_REALLOCATE_STACK(source, osize, nsize) \
281 r_re_alloc (&failure_stack_ptr, (nsize))
282 #define REGEX_FREE_STACK(ptr) \
283 r_alloc_free (&failure_stack_ptr)
285 #else /* not using relocating allocator */
289 #define REGEX_ALLOCATE_STACK malloc
290 #define REGEX_REALLOCATE_STACK(source, osize, nsize) realloc (source, nsize)
291 #define REGEX_FREE_STACK free
293 #else /* not REGEX_MALLOC */
295 #define REGEX_ALLOCATE_STACK alloca
297 #define REGEX_REALLOCATE_STACK(source, osize, nsize) \
298 REGEX_REALLOCATE (source, osize, nsize)
299 /* No need to explicitly free anything. */
300 #define REGEX_FREE_STACK(arg)
302 #endif /* not REGEX_MALLOC */
303 #endif /* not using relocating allocator */
306 /* True if `size1' is non-NULL and PTR is pointing anywhere inside
307 `string1' or just past its end. This works if PTR is NULL, which is
309 #define FIRST_STRING_P(ptr) \
310 (size1 && string1 <= (ptr) && (ptr) <= string1 + size1)
312 /* (Re)Allocate N items of type T using malloc, or fail. */
313 #define TALLOC(n, t) ((t *) malloc ((n) * sizeof (t)))
314 #define RETALLOC(addr, n, t) ((addr) = (t *) realloc (addr, (n) * sizeof (t)))
315 #define RETALLOC_IF(addr, n, t) \
316 if (addr) RETALLOC((addr), (n), t); else (addr) = TALLOC ((n), t)
317 #define REGEX_TALLOC(n, t) ((t *) REGEX_ALLOCATE ((n) * sizeof (t)))
319 #define BYTEWIDTH 8 /* In bits. */
321 #define STREQ(s1, s2) ((strcmp (s1, s2) == 0))
325 #define MAX(a, b) ((a) > (b) ? (a) : (b))
326 #define MIN(a, b) ((a) < (b) ? (a) : (b))
328 typedef char boolean
;
332 static int re_match_2_internal ();
334 /* These are the command codes that appear in compiled regular
335 expressions. Some opcodes are followed by argument bytes. A
336 command code can specify any interpretation whatsoever for its
337 arguments. Zero bytes may appear in the compiled regular expression. */
343 /* Succeed right away--no more backtracking. */
346 /* Followed by one byte giving n, then by n literal bytes. */
349 /* Matches any (more or less) character. */
352 /* Matches any one char belonging to specified set. First
353 following byte is number of bitmap bytes. Then come bytes
354 for a bitmap saying which chars are in. Bits in each byte
355 are ordered low-bit-first. A character is in the set if its
356 bit is 1. A character too large to have a bit in the map is
357 automatically not in the set. */
360 /* Same parameters as charset, but match any character that is
361 not one of those specified. */
364 /* Start remembering the text that is matched, for storing in a
365 register. Followed by one byte with the register number, in
366 the range 0 to one less than the pattern buffer's re_nsub
367 field. Then followed by one byte with the number of groups
368 inner to this one. (This last has to be part of the
369 start_memory only because we need it in the on_failure_jump
373 /* Stop remembering the text that is matched and store it in a
374 memory register. Followed by one byte with the register
375 number, in the range 0 to one less than `re_nsub' in the
376 pattern buffer, and one byte with the number of inner groups,
377 just like `start_memory'. (We need the number of inner
378 groups here because we don't have any easy way of finding the
379 corresponding start_memory when we're at a stop_memory.) */
382 /* Match a duplicate of something remembered. Followed by one
383 byte containing the register number. */
386 /* Fail unless at beginning of line. */
389 /* Fail unless at end of line. */
392 /* Succeeds if at beginning of buffer (if emacs) or at beginning
393 of string to be matched (if not). */
396 /* Analogously, for end of buffer/string. */
399 /* Followed by two byte relative address to which to jump. */
402 /* Same as jump, but marks the end of an alternative. */
405 /* Followed by two-byte relative address of place to resume at
406 in case of failure. */
409 /* Like on_failure_jump, but pushes a placeholder instead of the
410 current string position when executed. */
411 on_failure_keep_string_jump
,
413 /* Throw away latest failure point and then jump to following
414 two-byte relative address. */
417 /* Change to pop_failure_jump if know won't have to backtrack to
418 match; otherwise change to jump. This is used to jump
419 back to the beginning of a repeat. If what follows this jump
420 clearly won't match what the repeat does, such that we can be
421 sure that there is no use backtracking out of repetitions
422 already matched, then we change it to a pop_failure_jump.
423 Followed by two-byte address. */
426 /* Jump to following two-byte address, and push a dummy failure
427 point. This failure point will be thrown away if an attempt
428 is made to use it for a failure. A `+' construct makes this
429 before the first repeat. Also used as an intermediary kind
430 of jump when compiling an alternative. */
433 /* Push a dummy failure point and continue. Used at the end of
437 /* Followed by two-byte relative address and two-byte number n.
438 After matching N times, jump to the address upon failure. */
441 /* Followed by two-byte relative address, and two-byte number n.
442 Jump to the address N times, then fail. */
445 /* Set the following two-byte relative address to the
446 subsequent two-byte number. The address *includes* the two
450 wordchar
, /* Matches any word-constituent character. */
451 notwordchar
, /* Matches any char that is not a word-constituent. */
453 wordbeg
, /* Succeeds if at word beginning. */
454 wordend
, /* Succeeds if at word end. */
456 wordbound
, /* Succeeds if at a word boundary. */
457 notwordbound
/* Succeeds if not at a word boundary. */
460 ,before_dot
, /* Succeeds if before point. */
461 at_dot
, /* Succeeds if at point. */
462 after_dot
, /* Succeeds if after point. */
464 /* Matches any character whose syntax is specified. Followed by
465 a byte which contains a syntax code, e.g., Sword. */
468 /* Matches any character whose syntax is not that specified. */
473 /* Common operations on the compiled pattern. */
475 /* Store NUMBER in two contiguous bytes starting at DESTINATION. */
477 #define STORE_NUMBER(destination, number) \
479 (destination)[0] = (number) & 0377; \
480 (destination)[1] = (number) >> 8; \
483 /* Same as STORE_NUMBER, except increment DESTINATION to
484 the byte after where the number is stored. Therefore, DESTINATION
485 must be an lvalue. */
487 #define STORE_NUMBER_AND_INCR(destination, number) \
489 STORE_NUMBER (destination, number); \
490 (destination) += 2; \
493 /* Put into DESTINATION a number stored in two contiguous bytes starting
496 #define EXTRACT_NUMBER(destination, source) \
498 (destination) = *(source) & 0377; \
499 (destination) += SIGN_EXTEND_CHAR (*((source) + 1)) << 8; \
504 extract_number (dest
, source
)
506 unsigned char *source
;
508 int temp
= SIGN_EXTEND_CHAR (*(source
+ 1));
509 *dest
= *source
& 0377;
513 #ifndef EXTRACT_MACROS /* To debug the macros. */
514 #undef EXTRACT_NUMBER
515 #define EXTRACT_NUMBER(dest, src) extract_number (&dest, src)
516 #endif /* not EXTRACT_MACROS */
520 /* Same as EXTRACT_NUMBER, except increment SOURCE to after the number.
521 SOURCE must be an lvalue. */
523 #define EXTRACT_NUMBER_AND_INCR(destination, source) \
525 EXTRACT_NUMBER (destination, source); \
531 extract_number_and_incr (destination
, source
)
533 unsigned char **source
;
535 extract_number (destination
, *source
);
539 #ifndef EXTRACT_MACROS
540 #undef EXTRACT_NUMBER_AND_INCR
541 #define EXTRACT_NUMBER_AND_INCR(dest, src) \
542 extract_number_and_incr (&dest, &src)
543 #endif /* not EXTRACT_MACROS */
547 /* If DEBUG is defined, Regex prints many voluminous messages about what
548 it is doing (if the variable `debug' is nonzero). If linked with the
549 main program in `iregex.c', you can enter patterns and strings
550 interactively. And if linked with the main program in `main.c' and
551 the other test files, you can run the already-written tests. */
555 /* We use standard I/O for debugging. */
558 /* It is useful to test things that ``must'' be true when debugging. */
561 static int debug
= 0;
563 #define DEBUG_STATEMENT(e) e
564 #define DEBUG_PRINT1(x) if (debug) printf (x)
565 #define DEBUG_PRINT2(x1, x2) if (debug) printf (x1, x2)
566 #define DEBUG_PRINT3(x1, x2, x3) if (debug) printf (x1, x2, x3)
567 #define DEBUG_PRINT4(x1, x2, x3, x4) if (debug) printf (x1, x2, x3, x4)
568 #define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) \
569 if (debug) print_partial_compiled_pattern (s, e)
570 #define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) \
571 if (debug) print_double_string (w, s1, sz1, s2, sz2)
574 /* Print the fastmap in human-readable form. */
577 print_fastmap (fastmap
)
580 unsigned was_a_range
= 0;
583 while (i
< (1 << BYTEWIDTH
))
589 while (i
< (1 << BYTEWIDTH
) && fastmap
[i
])
605 /* Print a compiled pattern string in human-readable form, starting at
606 the START pointer into it and ending just before the pointer END. */
609 print_partial_compiled_pattern (start
, end
)
610 unsigned char *start
;
614 unsigned char *p
= start
;
615 unsigned char *pend
= end
;
623 /* Loop over pattern commands. */
626 printf ("%d:\t", p
- start
);
628 switch ((re_opcode_t
) *p
++)
636 printf ("/exactn/%d", mcnt
);
647 printf ("/start_memory/%d/%d", mcnt
, *p
++);
652 printf ("/stop_memory/%d/%d", mcnt
, *p
++);
656 printf ("/duplicate/%d", *p
++);
666 register int c
, last
= -100;
667 register int in_range
= 0;
669 printf ("/charset [%s",
670 (re_opcode_t
) *(p
- 1) == charset_not
? "^" : "");
672 assert (p
+ *p
< pend
);
674 for (c
= 0; c
< 256; c
++)
676 && (p
[1 + (c
/8)] & (1 << (c
% 8))))
678 /* Are we starting a range? */
679 if (last
+ 1 == c
&& ! in_range
)
684 /* Have we broken a range? */
685 else if (last
+ 1 != c
&& in_range
)
714 case on_failure_jump
:
715 extract_number_and_incr (&mcnt
, &p
);
716 printf ("/on_failure_jump to %d", p
+ mcnt
- start
);
719 case on_failure_keep_string_jump
:
720 extract_number_and_incr (&mcnt
, &p
);
721 printf ("/on_failure_keep_string_jump to %d", p
+ mcnt
- start
);
724 case dummy_failure_jump
:
725 extract_number_and_incr (&mcnt
, &p
);
726 printf ("/dummy_failure_jump to %d", p
+ mcnt
- start
);
729 case push_dummy_failure
:
730 printf ("/push_dummy_failure");
734 extract_number_and_incr (&mcnt
, &p
);
735 printf ("/maybe_pop_jump to %d", p
+ mcnt
- start
);
738 case pop_failure_jump
:
739 extract_number_and_incr (&mcnt
, &p
);
740 printf ("/pop_failure_jump to %d", p
+ mcnt
- start
);
744 extract_number_and_incr (&mcnt
, &p
);
745 printf ("/jump_past_alt to %d", p
+ mcnt
- start
);
749 extract_number_and_incr (&mcnt
, &p
);
750 printf ("/jump to %d", p
+ mcnt
- start
);
754 extract_number_and_incr (&mcnt
, &p
);
755 extract_number_and_incr (&mcnt2
, &p
);
756 printf ("/succeed_n to %d, %d times", p
+ mcnt
- start
, mcnt2
);
760 extract_number_and_incr (&mcnt
, &p
);
761 extract_number_and_incr (&mcnt2
, &p
);
762 printf ("/jump_n to %d, %d times", p
+ mcnt
- start
, mcnt2
);
766 extract_number_and_incr (&mcnt
, &p
);
767 extract_number_and_incr (&mcnt2
, &p
);
768 printf ("/set_number_at location %d to %d", p
+ mcnt
- start
, mcnt2
);
772 printf ("/wordbound");
776 printf ("/notwordbound");
788 printf ("/before_dot");
796 printf ("/after_dot");
800 printf ("/syntaxspec");
802 printf ("/%d", mcnt
);
806 printf ("/notsyntaxspec");
808 printf ("/%d", mcnt
);
813 printf ("/wordchar");
817 printf ("/notwordchar");
829 printf ("?%d", *(p
-1));
835 printf ("%d:\tend of pattern.\n", p
- start
);
840 print_compiled_pattern (bufp
)
841 struct re_pattern_buffer
*bufp
;
843 unsigned char *buffer
= bufp
->buffer
;
845 print_partial_compiled_pattern (buffer
, buffer
+ bufp
->used
);
846 printf ("%d bytes used/%d bytes allocated.\n", bufp
->used
, bufp
->allocated
);
848 if (bufp
->fastmap_accurate
&& bufp
->fastmap
)
850 printf ("fastmap: ");
851 print_fastmap (bufp
->fastmap
);
854 printf ("re_nsub: %d\t", bufp
->re_nsub
);
855 printf ("regs_alloc: %d\t", bufp
->regs_allocated
);
856 printf ("can_be_null: %d\t", bufp
->can_be_null
);
857 printf ("newline_anchor: %d\n", bufp
->newline_anchor
);
858 printf ("no_sub: %d\t", bufp
->no_sub
);
859 printf ("not_bol: %d\t", bufp
->not_bol
);
860 printf ("not_eol: %d\t", bufp
->not_eol
);
861 printf ("syntax: %d\n", bufp
->syntax
);
862 /* Perhaps we should print the translate table? */
867 print_double_string (where
, string1
, size1
, string2
, size2
)
880 if (FIRST_STRING_P (where
))
882 for (this_char
= where
- string1
; this_char
< size1
; this_char
++)
883 putchar (string1
[this_char
]);
888 for (this_char
= where
- string2
; this_char
< size2
; this_char
++)
889 putchar (string2
[this_char
]);
893 #else /* not DEBUG */
898 #define DEBUG_STATEMENT(e)
899 #define DEBUG_PRINT1(x)
900 #define DEBUG_PRINT2(x1, x2)
901 #define DEBUG_PRINT3(x1, x2, x3)
902 #define DEBUG_PRINT4(x1, x2, x3, x4)
903 #define DEBUG_PRINT_COMPILED_PATTERN(p, s, e)
904 #define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2)
906 #endif /* not DEBUG */
908 /* Set by `re_set_syntax' to the current regexp syntax to recognize. Can
909 also be assigned to arbitrarily: each pattern buffer stores its own
910 syntax, so it can be changed between regex compilations. */
911 /* This has no initializer because initialized variables in Emacs
912 become read-only after dumping. */
913 reg_syntax_t re_syntax_options
;
916 /* Specify the precise syntax of regexps for compilation. This provides
917 for compatibility for various utilities which historically have
918 different, incompatible syntaxes.
920 The argument SYNTAX is a bit mask comprised of the various bits
921 defined in regex.h. We return the old syntax. */
924 re_set_syntax (syntax
)
927 reg_syntax_t ret
= re_syntax_options
;
929 re_syntax_options
= syntax
;
933 /* This table gives an error message for each of the error codes listed
934 in regex.h. Obviously the order here has to be same as there.
935 POSIX doesn't require that we do anything for REG_NOERROR,
936 but why not be nice? */
938 static const char *re_error_msgid
[] =
940 gettext_noop ("Success"), /* REG_NOERROR */
941 gettext_noop ("No match"), /* REG_NOMATCH */
942 gettext_noop ("Invalid regular expression"), /* REG_BADPAT */
943 gettext_noop ("Invalid collation character"), /* REG_ECOLLATE */
944 gettext_noop ("Invalid character class name"), /* REG_ECTYPE */
945 gettext_noop ("Trailing backslash"), /* REG_EESCAPE */
946 gettext_noop ("Invalid back reference"), /* REG_ESUBREG */
947 gettext_noop ("Unmatched [ or [^"), /* REG_EBRACK */
948 gettext_noop ("Unmatched ( or \\("), /* REG_EPAREN */
949 gettext_noop ("Unmatched \\{"), /* REG_EBRACE */
950 gettext_noop ("Invalid content of \\{\\}"), /* REG_BADBR */
951 gettext_noop ("Invalid range end"), /* REG_ERANGE */
952 gettext_noop ("Memory exhausted"), /* REG_ESPACE */
953 gettext_noop ("Invalid preceding regular expression"), /* REG_BADRPT */
954 gettext_noop ("Premature end of regular expression"), /* REG_EEND */
955 gettext_noop ("Regular expression too big"), /* REG_ESIZE */
956 gettext_noop ("Unmatched ) or \\)"), /* REG_ERPAREN */
959 /* Avoiding alloca during matching, to placate r_alloc. */
961 /* Define MATCH_MAY_ALLOCATE unless we need to make sure that the
962 searching and matching functions should not call alloca. On some
963 systems, alloca is implemented in terms of malloc, and if we're
964 using the relocating allocator routines, then malloc could cause a
965 relocation, which might (if the strings being searched are in the
966 ralloc heap) shift the data out from underneath the regexp
969 Here's another reason to avoid allocation: Emacs
970 processes input from X in a signal handler; processing X input may
971 call malloc; if input arrives while a matching routine is calling
972 malloc, then we're scrod. But Emacs can't just block input while
973 calling matching routines; then we don't notice interrupts when
974 they come in. So, Emacs blocks input around all regexp calls
975 except the matching calls, which it leaves unprotected, in the
976 faith that they will not malloc. */
978 /* Normally, this is fine. */
979 #define MATCH_MAY_ALLOCATE
981 /* When using GNU C, we are not REALLY using the C alloca, no matter
982 what config.h may say. So don't take precautions for it. */
987 /* The match routines may not allocate if (1) they would do it with malloc
988 and (2) it's not safe for them to use malloc.
989 Note that if REL_ALLOC is defined, matching would not use malloc for the
990 failure stack, but we would still use it for the register vectors;
991 so REL_ALLOC should not affect this. */
992 #if (defined (C_ALLOCA) || defined (REGEX_MALLOC)) && defined (emacs)
993 #undef MATCH_MAY_ALLOCATE
997 /* Failure stack declarations and macros; both re_compile_fastmap and
998 re_match_2 use a failure stack. These have to be macros because of
999 REGEX_ALLOCATE_STACK. */
1002 /* Number of failure points for which to initially allocate space
1003 when matching. If this number is exceeded, we allocate more
1004 space, so it is not a hard limit. */
1005 #ifndef INIT_FAILURE_ALLOC
1006 #define INIT_FAILURE_ALLOC 5
1009 /* Roughly the maximum number of failure points on the stack. Would be
1010 exactly that if always used MAX_FAILURE_ITEMS items each time we failed.
1011 This is a variable only so users of regex can assign to it; we never
1012 change it ourselves. */
1013 #if defined (MATCH_MAY_ALLOCATE)
1014 int re_max_failures
= 20000;
1016 int re_max_failures
= 2000;
1019 union fail_stack_elt
1021 unsigned char *pointer
;
1025 typedef union fail_stack_elt fail_stack_elt_t
;
1029 fail_stack_elt_t
*stack
;
1031 unsigned avail
; /* Offset of next open position. */
1034 #define FAIL_STACK_EMPTY() (fail_stack.avail == 0)
1035 #define FAIL_STACK_PTR_EMPTY() (fail_stack_ptr->avail == 0)
1036 #define FAIL_STACK_FULL() (fail_stack.avail == fail_stack.size)
1039 /* Define macros to initialize and free the failure stack.
1040 Do `return -2' if the alloc fails. */
1042 #ifdef MATCH_MAY_ALLOCATE
1043 #define INIT_FAIL_STACK() \
1045 fail_stack.stack = (fail_stack_elt_t *) \
1046 REGEX_ALLOCATE_STACK (INIT_FAILURE_ALLOC * sizeof (fail_stack_elt_t)); \
1048 if (fail_stack.stack == NULL) \
1051 fail_stack.size = INIT_FAILURE_ALLOC; \
1052 fail_stack.avail = 0; \
1055 #define RESET_FAIL_STACK() REGEX_FREE_STACK (fail_stack.stack)
1057 #define INIT_FAIL_STACK() \
1059 fail_stack.avail = 0; \
1062 #define RESET_FAIL_STACK()
1066 /* Double the size of FAIL_STACK, up to approximately `re_max_failures' items.
1068 Return 1 if succeeds, and 0 if either ran out of memory
1069 allocating space for it or it was already too large.
1071 REGEX_REALLOCATE_STACK requires `destination' be declared. */
1073 #define DOUBLE_FAIL_STACK(fail_stack) \
1074 ((fail_stack).size > re_max_failures * MAX_FAILURE_ITEMS \
1076 : ((fail_stack).stack = (fail_stack_elt_t *) \
1077 REGEX_REALLOCATE_STACK ((fail_stack).stack, \
1078 (fail_stack).size * sizeof (fail_stack_elt_t), \
1079 ((fail_stack).size << 1) * sizeof (fail_stack_elt_t)), \
1081 (fail_stack).stack == NULL \
1083 : ((fail_stack).size <<= 1, \
1087 /* Push pointer POINTER on FAIL_STACK.
1088 Return 1 if was able to do so and 0 if ran out of memory allocating
1090 #define PUSH_PATTERN_OP(POINTER, FAIL_STACK) \
1091 ((FAIL_STACK_FULL () \
1092 && !DOUBLE_FAIL_STACK (FAIL_STACK)) \
1094 : ((FAIL_STACK).stack[(FAIL_STACK).avail++].pointer = POINTER, \
1097 /* Push a pointer value onto the failure stack.
1098 Assumes the variable `fail_stack'. Probably should only
1099 be called from within `PUSH_FAILURE_POINT'. */
1100 #define PUSH_FAILURE_POINTER(item) \
1101 fail_stack.stack[fail_stack.avail++].pointer = (unsigned char *) (item)
1103 /* This pushes an integer-valued item onto the failure stack.
1104 Assumes the variable `fail_stack'. Probably should only
1105 be called from within `PUSH_FAILURE_POINT'. */
1106 #define PUSH_FAILURE_INT(item) \
1107 fail_stack.stack[fail_stack.avail++].integer = (item)
1109 /* Push a fail_stack_elt_t value onto the failure stack.
1110 Assumes the variable `fail_stack'. Probably should only
1111 be called from within `PUSH_FAILURE_POINT'. */
1112 #define PUSH_FAILURE_ELT(item) \
1113 fail_stack.stack[fail_stack.avail++] = (item)
1115 /* These three POP... operations complement the three PUSH... operations.
1116 All assume that `fail_stack' is nonempty. */
1117 #define POP_FAILURE_POINTER() fail_stack.stack[--fail_stack.avail].pointer
1118 #define POP_FAILURE_INT() fail_stack.stack[--fail_stack.avail].integer
1119 #define POP_FAILURE_ELT() fail_stack.stack[--fail_stack.avail]
1121 /* Used to omit pushing failure point id's when we're not debugging. */
1123 #define DEBUG_PUSH PUSH_FAILURE_INT
1124 #define DEBUG_POP(item_addr) *(item_addr) = POP_FAILURE_INT ()
1126 #define DEBUG_PUSH(item)
1127 #define DEBUG_POP(item_addr)
1131 /* Push the information about the state we will need
1132 if we ever fail back to it.
1134 Requires variables fail_stack, regstart, regend, reg_info, and
1135 num_regs be declared. DOUBLE_FAIL_STACK requires `destination' be
1138 Does `return FAILURE_CODE' if runs out of memory. */
1140 #define PUSH_FAILURE_POINT(pattern_place, string_place, failure_code) \
1142 char *destination; \
1143 /* Must be int, so when we don't save any registers, the arithmetic \
1144 of 0 + -1 isn't done as unsigned. */ \
1147 DEBUG_STATEMENT (failure_id++); \
1148 DEBUG_STATEMENT (nfailure_points_pushed++); \
1149 DEBUG_PRINT2 ("\nPUSH_FAILURE_POINT #%u:\n", failure_id); \
1150 DEBUG_PRINT2 (" Before push, next avail: %d\n", (fail_stack).avail);\
1151 DEBUG_PRINT2 (" size: %d\n", (fail_stack).size);\
1153 DEBUG_PRINT2 (" slots needed: %d\n", NUM_FAILURE_ITEMS); \
1154 DEBUG_PRINT2 (" available: %d\n", REMAINING_AVAIL_SLOTS); \
1156 /* Ensure we have enough space allocated for what we will push. */ \
1157 while (REMAINING_AVAIL_SLOTS < NUM_FAILURE_ITEMS) \
1159 if (!DOUBLE_FAIL_STACK (fail_stack)) \
1160 return failure_code; \
1162 DEBUG_PRINT2 ("\n Doubled stack; size now: %d\n", \
1163 (fail_stack).size); \
1164 DEBUG_PRINT2 (" slots available: %d\n", REMAINING_AVAIL_SLOTS);\
1167 /* Push the info, starting with the registers. */ \
1168 DEBUG_PRINT1 ("\n"); \
1171 for (this_reg = lowest_active_reg; this_reg <= highest_active_reg; \
1174 DEBUG_PRINT2 (" Pushing reg: %d\n", this_reg); \
1175 DEBUG_STATEMENT (num_regs_pushed++); \
1177 DEBUG_PRINT2 (" start: 0x%x\n", regstart[this_reg]); \
1178 PUSH_FAILURE_POINTER (regstart[this_reg]); \
1180 DEBUG_PRINT2 (" end: 0x%x\n", regend[this_reg]); \
1181 PUSH_FAILURE_POINTER (regend[this_reg]); \
1183 DEBUG_PRINT2 (" info: 0x%x\n ", reg_info[this_reg]); \
1184 DEBUG_PRINT2 (" match_null=%d", \
1185 REG_MATCH_NULL_STRING_P (reg_info[this_reg])); \
1186 DEBUG_PRINT2 (" active=%d", IS_ACTIVE (reg_info[this_reg])); \
1187 DEBUG_PRINT2 (" matched_something=%d", \
1188 MATCHED_SOMETHING (reg_info[this_reg])); \
1189 DEBUG_PRINT2 (" ever_matched=%d", \
1190 EVER_MATCHED_SOMETHING (reg_info[this_reg])); \
1191 DEBUG_PRINT1 ("\n"); \
1192 PUSH_FAILURE_ELT (reg_info[this_reg].word); \
1195 DEBUG_PRINT2 (" Pushing low active reg: %d\n", lowest_active_reg);\
1196 PUSH_FAILURE_INT (lowest_active_reg); \
1198 DEBUG_PRINT2 (" Pushing high active reg: %d\n", highest_active_reg);\
1199 PUSH_FAILURE_INT (highest_active_reg); \
1201 DEBUG_PRINT2 (" Pushing pattern 0x%x: ", pattern_place); \
1202 DEBUG_PRINT_COMPILED_PATTERN (bufp, pattern_place, pend); \
1203 PUSH_FAILURE_POINTER (pattern_place); \
1205 DEBUG_PRINT2 (" Pushing string 0x%x: `", string_place); \
1206 DEBUG_PRINT_DOUBLE_STRING (string_place, string1, size1, string2, \
1208 DEBUG_PRINT1 ("'\n"); \
1209 PUSH_FAILURE_POINTER (string_place); \
1211 DEBUG_PRINT2 (" Pushing failure id: %u\n", failure_id); \
1212 DEBUG_PUSH (failure_id); \
1215 /* This is the number of items that are pushed and popped on the stack
1216 for each register. */
1217 #define NUM_REG_ITEMS 3
1219 /* Individual items aside from the registers. */
1221 #define NUM_NONREG_ITEMS 5 /* Includes failure point id. */
1223 #define NUM_NONREG_ITEMS 4
1226 /* We push at most this many items on the stack. */
1227 /* We used to use (num_regs - 1), which is the number of registers
1228 this regexp will save; but that was changed to 5
1229 to avoid stack overflow for a regexp with lots of parens. */
1230 #define MAX_FAILURE_ITEMS (5 * NUM_REG_ITEMS + NUM_NONREG_ITEMS)
1232 /* We actually push this many items. */
1233 #define NUM_FAILURE_ITEMS \
1235 ? 0 : highest_active_reg - lowest_active_reg + 1) \
1239 /* How many items can still be added to the stack without overflowing it. */
1240 #define REMAINING_AVAIL_SLOTS ((fail_stack).size - (fail_stack).avail)
1243 /* Pops what PUSH_FAIL_STACK pushes.
1245 We restore into the parameters, all of which should be lvalues:
1246 STR -- the saved data position.
1247 PAT -- the saved pattern position.
1248 LOW_REG, HIGH_REG -- the highest and lowest active registers.
1249 REGSTART, REGEND -- arrays of string positions.
1250 REG_INFO -- array of information about each subexpression.
1252 Also assumes the variables `fail_stack' and (if debugging), `bufp',
1253 `pend', `string1', `size1', `string2', and `size2'. */
1255 #define POP_FAILURE_POINT(str, pat, low_reg, high_reg, regstart, regend, reg_info)\
1257 DEBUG_STATEMENT (fail_stack_elt_t failure_id;) \
1259 const unsigned char *string_temp; \
1261 assert (!FAIL_STACK_EMPTY ()); \
1263 /* Remove failure points and point to how many regs pushed. */ \
1264 DEBUG_PRINT1 ("POP_FAILURE_POINT:\n"); \
1265 DEBUG_PRINT2 (" Before pop, next avail: %d\n", fail_stack.avail); \
1266 DEBUG_PRINT2 (" size: %d\n", fail_stack.size); \
1268 assert (fail_stack.avail >= NUM_NONREG_ITEMS); \
1270 DEBUG_POP (&failure_id); \
1271 DEBUG_PRINT2 (" Popping failure id: %u\n", failure_id); \
1273 /* If the saved string location is NULL, it came from an \
1274 on_failure_keep_string_jump opcode, and we want to throw away the \
1275 saved NULL, thus retaining our current position in the string. */ \
1276 string_temp = POP_FAILURE_POINTER (); \
1277 if (string_temp != NULL) \
1278 str = (const char *) string_temp; \
1280 DEBUG_PRINT2 (" Popping string 0x%x: `", str); \
1281 DEBUG_PRINT_DOUBLE_STRING (str, string1, size1, string2, size2); \
1282 DEBUG_PRINT1 ("'\n"); \
1284 pat = (unsigned char *) POP_FAILURE_POINTER (); \
1285 DEBUG_PRINT2 (" Popping pattern 0x%x: ", pat); \
1286 DEBUG_PRINT_COMPILED_PATTERN (bufp, pat, pend); \
1288 /* Restore register info. */ \
1289 high_reg = (unsigned) POP_FAILURE_INT (); \
1290 DEBUG_PRINT2 (" Popping high active reg: %d\n", high_reg); \
1292 low_reg = (unsigned) POP_FAILURE_INT (); \
1293 DEBUG_PRINT2 (" Popping low active reg: %d\n", low_reg); \
1296 for (this_reg = high_reg; this_reg >= low_reg; this_reg--) \
1298 DEBUG_PRINT2 (" Popping reg: %d\n", this_reg); \
1300 reg_info[this_reg].word = POP_FAILURE_ELT (); \
1301 DEBUG_PRINT2 (" info: 0x%x\n", reg_info[this_reg]); \
1303 regend[this_reg] = (const char *) POP_FAILURE_POINTER (); \
1304 DEBUG_PRINT2 (" end: 0x%x\n", regend[this_reg]); \
1306 regstart[this_reg] = (const char *) POP_FAILURE_POINTER (); \
1307 DEBUG_PRINT2 (" start: 0x%x\n", regstart[this_reg]); \
1311 for (this_reg = highest_active_reg; this_reg > high_reg; this_reg--) \
1313 reg_info[this_reg].word.integer = 0; \
1314 regend[this_reg] = 0; \
1315 regstart[this_reg] = 0; \
1317 highest_active_reg = high_reg; \
1320 set_regs_matched_done = 0; \
1321 DEBUG_STATEMENT (nfailure_points_popped++); \
1322 } /* POP_FAILURE_POINT */
1326 /* Structure for per-register (a.k.a. per-group) information.
1327 Other register information, such as the
1328 starting and ending positions (which are addresses), and the list of
1329 inner groups (which is a bits list) are maintained in separate
1332 We are making a (strictly speaking) nonportable assumption here: that
1333 the compiler will pack our bit fields into something that fits into
1334 the type of `word', i.e., is something that fits into one item on the
1339 fail_stack_elt_t word
;
1342 /* This field is one if this group can match the empty string,
1343 zero if not. If not yet determined, `MATCH_NULL_UNSET_VALUE'. */
1344 #define MATCH_NULL_UNSET_VALUE 3
1345 unsigned match_null_string_p
: 2;
1346 unsigned is_active
: 1;
1347 unsigned matched_something
: 1;
1348 unsigned ever_matched_something
: 1;
1350 } register_info_type
;
1352 #define REG_MATCH_NULL_STRING_P(R) ((R).bits.match_null_string_p)
1353 #define IS_ACTIVE(R) ((R).bits.is_active)
1354 #define MATCHED_SOMETHING(R) ((R).bits.matched_something)
1355 #define EVER_MATCHED_SOMETHING(R) ((R).bits.ever_matched_something)
1358 /* Call this when have matched a real character; it sets `matched' flags
1359 for the subexpressions which we are currently inside. Also records
1360 that those subexprs have matched. */
1361 #define SET_REGS_MATCHED() \
1364 if (!set_regs_matched_done) \
1367 set_regs_matched_done = 1; \
1368 for (r = lowest_active_reg; r <= highest_active_reg; r++) \
1370 MATCHED_SOMETHING (reg_info[r]) \
1371 = EVER_MATCHED_SOMETHING (reg_info[r]) \
1378 /* Registers are set to a sentinel when they haven't yet matched. */
1379 static char reg_unset_dummy
;
1380 #define REG_UNSET_VALUE (®_unset_dummy)
1381 #define REG_UNSET(e) ((e) == REG_UNSET_VALUE)
1383 /* Subroutine declarations and macros for regex_compile. */
1385 static void store_op1 (), store_op2 ();
1386 static void insert_op1 (), insert_op2 ();
1387 static boolean
at_begline_loc_p (), at_endline_loc_p ();
1388 static boolean
group_in_compile_stack ();
1389 static reg_errcode_t
compile_range ();
1391 /* Fetch the next character in the uncompiled pattern---translating it
1392 if necessary. Also cast from a signed character in the constant
1393 string passed to us by the user to an unsigned char that we can use
1394 as an array index (in, e.g., `translate'). */
1396 #define PATFETCH(c) \
1397 do {if (p == pend) return REG_EEND; \
1398 c = (unsigned char) *p++; \
1399 if (translate) c = (unsigned char) translate[c]; \
1403 /* Fetch the next character in the uncompiled pattern, with no
1405 #define PATFETCH_RAW(c) \
1406 do {if (p == pend) return REG_EEND; \
1407 c = (unsigned char) *p++; \
1410 /* Go backwards one character in the pattern. */
1411 #define PATUNFETCH p--
1414 /* If `translate' is non-null, return translate[D], else just D. We
1415 cast the subscript to translate because some data is declared as
1416 `char *', to avoid warnings when a string constant is passed. But
1417 when we use a character as a subscript we must make it unsigned. */
1419 #define TRANSLATE(d) \
1420 (translate ? (char) translate[(unsigned char) (d)] : (d))
1424 /* Macros for outputting the compiled pattern into `buffer'. */
1426 /* If the buffer isn't allocated when it comes in, use this. */
1427 #define INIT_BUF_SIZE 32
1429 /* Make sure we have at least N more bytes of space in buffer. */
1430 #define GET_BUFFER_SPACE(n) \
1431 while (b - bufp->buffer + (n) > bufp->allocated) \
1434 /* Make sure we have one more byte of buffer space and then add C to it. */
1435 #define BUF_PUSH(c) \
1437 GET_BUFFER_SPACE (1); \
1438 *b++ = (unsigned char) (c); \
1442 /* Ensure we have two more bytes of buffer space and then append C1 and C2. */
1443 #define BUF_PUSH_2(c1, c2) \
1445 GET_BUFFER_SPACE (2); \
1446 *b++ = (unsigned char) (c1); \
1447 *b++ = (unsigned char) (c2); \
1451 /* As with BUF_PUSH_2, except for three bytes. */
1452 #define BUF_PUSH_3(c1, c2, c3) \
1454 GET_BUFFER_SPACE (3); \
1455 *b++ = (unsigned char) (c1); \
1456 *b++ = (unsigned char) (c2); \
1457 *b++ = (unsigned char) (c3); \
1461 /* Store a jump with opcode OP at LOC to location TO. We store a
1462 relative address offset by the three bytes the jump itself occupies. */
1463 #define STORE_JUMP(op, loc, to) \
1464 store_op1 (op, loc, (to) - (loc) - 3)
1466 /* Likewise, for a two-argument jump. */
1467 #define STORE_JUMP2(op, loc, to, arg) \
1468 store_op2 (op, loc, (to) - (loc) - 3, arg)
1470 /* Like `STORE_JUMP', but for inserting. Assume `b' is the buffer end. */
1471 #define INSERT_JUMP(op, loc, to) \
1472 insert_op1 (op, loc, (to) - (loc) - 3, b)
1474 /* Like `STORE_JUMP2', but for inserting. Assume `b' is the buffer end. */
1475 #define INSERT_JUMP2(op, loc, to, arg) \
1476 insert_op2 (op, loc, (to) - (loc) - 3, arg, b)
1479 /* This is not an arbitrary limit: the arguments which represent offsets
1480 into the pattern are two bytes long. So if 2^16 bytes turns out to
1481 be too small, many things would have to change. */
1482 #define MAX_BUF_SIZE (1L << 16)
1485 /* Extend the buffer by twice its current size via realloc and
1486 reset the pointers that pointed into the old block to point to the
1487 correct places in the new one. If extending the buffer results in it
1488 being larger than MAX_BUF_SIZE, then flag memory exhausted. */
1489 #define EXTEND_BUFFER() \
1491 unsigned char *old_buffer = bufp->buffer; \
1492 if (bufp->allocated == MAX_BUF_SIZE) \
1494 bufp->allocated <<= 1; \
1495 if (bufp->allocated > MAX_BUF_SIZE) \
1496 bufp->allocated = MAX_BUF_SIZE; \
1497 bufp->buffer = (unsigned char *) realloc (bufp->buffer, bufp->allocated);\
1498 if (bufp->buffer == NULL) \
1499 return REG_ESPACE; \
1500 /* If the buffer moved, move all the pointers into it. */ \
1501 if (old_buffer != bufp->buffer) \
1503 b = (b - old_buffer) + bufp->buffer; \
1504 begalt = (begalt - old_buffer) + bufp->buffer; \
1505 if (fixup_alt_jump) \
1506 fixup_alt_jump = (fixup_alt_jump - old_buffer) + bufp->buffer;\
1508 laststart = (laststart - old_buffer) + bufp->buffer; \
1509 if (pending_exact) \
1510 pending_exact = (pending_exact - old_buffer) + bufp->buffer; \
1515 /* Since we have one byte reserved for the register number argument to
1516 {start,stop}_memory, the maximum number of groups we can report
1517 things about is what fits in that byte. */
1518 #define MAX_REGNUM 255
1520 /* But patterns can have more than `MAX_REGNUM' registers. We just
1521 ignore the excess. */
1522 typedef unsigned regnum_t
;
1525 /* Macros for the compile stack. */
1527 /* Since offsets can go either forwards or backwards, this type needs to
1528 be able to hold values from -(MAX_BUF_SIZE - 1) to MAX_BUF_SIZE - 1. */
1529 typedef int pattern_offset_t
;
1533 pattern_offset_t begalt_offset
;
1534 pattern_offset_t fixup_alt_jump
;
1535 pattern_offset_t inner_group_offset
;
1536 pattern_offset_t laststart_offset
;
1538 } compile_stack_elt_t
;
1543 compile_stack_elt_t
*stack
;
1545 unsigned avail
; /* Offset of next open position. */
1546 } compile_stack_type
;
1549 #define INIT_COMPILE_STACK_SIZE 32
1551 #define COMPILE_STACK_EMPTY (compile_stack.avail == 0)
1552 #define COMPILE_STACK_FULL (compile_stack.avail == compile_stack.size)
1554 /* The next available element. */
1555 #define COMPILE_STACK_TOP (compile_stack.stack[compile_stack.avail])
1558 /* Set the bit for character C in a list. */
1559 #define SET_LIST_BIT(c) \
1560 (b[((unsigned char) (c)) / BYTEWIDTH] \
1561 |= 1 << (((unsigned char) c) % BYTEWIDTH))
1564 /* Get the next unsigned number in the uncompiled pattern. */
1565 #define GET_UNSIGNED_NUMBER(num) \
1569 while (ISDIGIT (c)) \
1573 num = num * 10 + c - '0'; \
1581 #define CHAR_CLASS_MAX_LENGTH 6 /* Namely, `xdigit'. */
1583 #define IS_CHAR_CLASS(string) \
1584 (STREQ (string, "alpha") || STREQ (string, "upper") \
1585 || STREQ (string, "lower") || STREQ (string, "digit") \
1586 || STREQ (string, "alnum") || STREQ (string, "xdigit") \
1587 || STREQ (string, "space") || STREQ (string, "print") \
1588 || STREQ (string, "punct") || STREQ (string, "graph") \
1589 || STREQ (string, "cntrl") || STREQ (string, "blank"))
1591 #ifndef MATCH_MAY_ALLOCATE
1593 /* If we cannot allocate large objects within re_match_2_internal,
1594 we make the fail stack and register vectors global.
1595 The fail stack, we grow to the maximum size when a regexp
1597 The register vectors, we adjust in size each time we
1598 compile a regexp, according to the number of registers it needs. */
1600 static fail_stack_type fail_stack
;
1602 /* Size with which the following vectors are currently allocated.
1603 That is so we can make them bigger as needed,
1604 but never make them smaller. */
1605 static int regs_allocated_size
;
1607 static const char ** regstart
, ** regend
;
1608 static const char ** old_regstart
, ** old_regend
;
1609 static const char **best_regstart
, **best_regend
;
1610 static register_info_type
*reg_info
;
1611 static const char **reg_dummy
;
1612 static register_info_type
*reg_info_dummy
;
1614 /* Make the register vectors big enough for NUM_REGS registers,
1615 but don't make them smaller. */
1618 regex_grow_registers (num_regs
)
1621 if (num_regs
> regs_allocated_size
)
1623 RETALLOC_IF (regstart
, num_regs
, const char *);
1624 RETALLOC_IF (regend
, num_regs
, const char *);
1625 RETALLOC_IF (old_regstart
, num_regs
, const char *);
1626 RETALLOC_IF (old_regend
, num_regs
, const char *);
1627 RETALLOC_IF (best_regstart
, num_regs
, const char *);
1628 RETALLOC_IF (best_regend
, num_regs
, const char *);
1629 RETALLOC_IF (reg_info
, num_regs
, register_info_type
);
1630 RETALLOC_IF (reg_dummy
, num_regs
, const char *);
1631 RETALLOC_IF (reg_info_dummy
, num_regs
, register_info_type
);
1633 regs_allocated_size
= num_regs
;
1637 #endif /* not MATCH_MAY_ALLOCATE */
1639 /* `regex_compile' compiles PATTERN (of length SIZE) according to SYNTAX.
1640 Returns one of error codes defined in `regex.h', or zero for success.
1642 Assumes the `allocated' (and perhaps `buffer') and `translate'
1643 fields are set in BUFP on entry.
1645 If it succeeds, results are put in BUFP (if it returns an error, the
1646 contents of BUFP are undefined):
1647 `buffer' is the compiled pattern;
1648 `syntax' is set to SYNTAX;
1649 `used' is set to the length of the compiled pattern;
1650 `fastmap_accurate' is zero;
1651 `re_nsub' is the number of subexpressions in PATTERN;
1652 `not_bol' and `not_eol' are zero;
1654 The `fastmap' and `newline_anchor' fields are neither
1655 examined nor set. */
1657 /* Return, freeing storage we allocated. */
1658 #define FREE_STACK_RETURN(value) \
1659 return (free (compile_stack.stack), value)
1661 static reg_errcode_t
1662 regex_compile (pattern
, size
, syntax
, bufp
)
1663 const char *pattern
;
1665 reg_syntax_t syntax
;
1666 struct re_pattern_buffer
*bufp
;
1668 /* We fetch characters from PATTERN here. Even though PATTERN is
1669 `char *' (i.e., signed), we declare these variables as unsigned, so
1670 they can be reliably used as array indices. */
1671 register unsigned char c
, c1
;
1673 /* A random temporary spot in PATTERN. */
1676 /* Points to the end of the buffer, where we should append. */
1677 register unsigned char *b
;
1679 /* Keeps track of unclosed groups. */
1680 compile_stack_type compile_stack
;
1682 /* Points to the current (ending) position in the pattern. */
1683 const char *p
= pattern
;
1684 const char *pend
= pattern
+ size
;
1686 /* How to translate the characters in the pattern. */
1687 RE_TRANSLATE_TYPE translate
= bufp
->translate
;
1689 /* Address of the count-byte of the most recently inserted `exactn'
1690 command. This makes it possible to tell if a new exact-match
1691 character can be added to that command or if the character requires
1692 a new `exactn' command. */
1693 unsigned char *pending_exact
= 0;
1695 /* Address of start of the most recently finished expression.
1696 This tells, e.g., postfix * where to find the start of its
1697 operand. Reset at the beginning of groups and alternatives. */
1698 unsigned char *laststart
= 0;
1700 /* Address of beginning of regexp, or inside of last group. */
1701 unsigned char *begalt
;
1703 /* Place in the uncompiled pattern (i.e., the {) to
1704 which to go back if the interval is invalid. */
1705 const char *beg_interval
;
1707 /* Address of the place where a forward jump should go to the end of
1708 the containing expression. Each alternative of an `or' -- except the
1709 last -- ends with a forward jump of this sort. */
1710 unsigned char *fixup_alt_jump
= 0;
1712 /* Counts open-groups as they are encountered. Remembered for the
1713 matching close-group on the compile stack, so the same register
1714 number is put in the stop_memory as the start_memory. */
1715 regnum_t regnum
= 0;
1718 DEBUG_PRINT1 ("\nCompiling pattern: ");
1721 unsigned debug_count
;
1723 for (debug_count
= 0; debug_count
< size
; debug_count
++)
1724 putchar (pattern
[debug_count
]);
1729 /* Initialize the compile stack. */
1730 compile_stack
.stack
= TALLOC (INIT_COMPILE_STACK_SIZE
, compile_stack_elt_t
);
1731 if (compile_stack
.stack
== NULL
)
1734 compile_stack
.size
= INIT_COMPILE_STACK_SIZE
;
1735 compile_stack
.avail
= 0;
1737 /* Initialize the pattern buffer. */
1738 bufp
->syntax
= syntax
;
1739 bufp
->fastmap_accurate
= 0;
1740 bufp
->not_bol
= bufp
->not_eol
= 0;
1742 /* Set `used' to zero, so that if we return an error, the pattern
1743 printer (for debugging) will think there's no pattern. We reset it
1747 /* Always count groups, whether or not bufp->no_sub is set. */
1750 #if !defined (emacs) && !defined (SYNTAX_TABLE)
1751 /* Initialize the syntax table. */
1752 init_syntax_once ();
1755 if (bufp
->allocated
== 0)
1758 { /* If zero allocated, but buffer is non-null, try to realloc
1759 enough space. This loses if buffer's address is bogus, but
1760 that is the user's responsibility. */
1761 RETALLOC (bufp
->buffer
, INIT_BUF_SIZE
, unsigned char);
1764 { /* Caller did not allocate a buffer. Do it for them. */
1765 bufp
->buffer
= TALLOC (INIT_BUF_SIZE
, unsigned char);
1767 if (!bufp
->buffer
) FREE_STACK_RETURN (REG_ESPACE
);
1769 bufp
->allocated
= INIT_BUF_SIZE
;
1772 begalt
= b
= bufp
->buffer
;
1774 /* Loop through the uncompiled pattern until we're at the end. */
1783 if ( /* If at start of pattern, it's an operator. */
1785 /* If context independent, it's an operator. */
1786 || syntax
& RE_CONTEXT_INDEP_ANCHORS
1787 /* Otherwise, depends on what's come before. */
1788 || at_begline_loc_p (pattern
, p
, syntax
))
1798 if ( /* If at end of pattern, it's an operator. */
1800 /* If context independent, it's an operator. */
1801 || syntax
& RE_CONTEXT_INDEP_ANCHORS
1802 /* Otherwise, depends on what's next. */
1803 || at_endline_loc_p (p
, pend
, syntax
))
1813 if ((syntax
& RE_BK_PLUS_QM
)
1814 || (syntax
& RE_LIMITED_OPS
))
1818 /* If there is no previous pattern... */
1821 if (syntax
& RE_CONTEXT_INVALID_OPS
)
1822 FREE_STACK_RETURN (REG_BADRPT
);
1823 else if (!(syntax
& RE_CONTEXT_INDEP_OPS
))
1828 /* Are we optimizing this jump? */
1829 boolean keep_string_p
= false;
1831 /* 1 means zero (many) matches is allowed. */
1832 char zero_times_ok
= 0, many_times_ok
= 0;
1834 /* If there is a sequence of repetition chars, collapse it
1835 down to just one (the right one). We can't combine
1836 interval operators with these because of, e.g., `a{2}*',
1837 which should only match an even number of `a's. */
1841 zero_times_ok
|= c
!= '+';
1842 many_times_ok
|= c
!= '?';
1850 || (!(syntax
& RE_BK_PLUS_QM
) && (c
== '+' || c
== '?')))
1853 else if (syntax
& RE_BK_PLUS_QM
&& c
== '\\')
1855 if (p
== pend
) FREE_STACK_RETURN (REG_EESCAPE
);
1858 if (!(c1
== '+' || c1
== '?'))
1873 /* If we get here, we found another repeat character. */
1876 /* Star, etc. applied to an empty pattern is equivalent
1877 to an empty pattern. */
1881 /* Now we know whether or not zero matches is allowed
1882 and also whether or not two or more matches is allowed. */
1884 { /* More than one repetition is allowed, so put in at the
1885 end a backward relative jump from `b' to before the next
1886 jump we're going to put in below (which jumps from
1887 laststart to after this jump).
1889 But if we are at the `*' in the exact sequence `.*\n',
1890 insert an unconditional jump backwards to the .,
1891 instead of the beginning of the loop. This way we only
1892 push a failure point once, instead of every time
1893 through the loop. */
1894 assert (p
- 1 > pattern
);
1896 /* Allocate the space for the jump. */
1897 GET_BUFFER_SPACE (3);
1899 /* We know we are not at the first character of the pattern,
1900 because laststart was nonzero. And we've already
1901 incremented `p', by the way, to be the character after
1902 the `*'. Do we have to do something analogous here
1903 for null bytes, because of RE_DOT_NOT_NULL? */
1904 if (TRANSLATE (*(p
- 2)) == TRANSLATE ('.')
1906 && p
< pend
&& TRANSLATE (*p
) == TRANSLATE ('\n')
1907 && !(syntax
& RE_DOT_NEWLINE
))
1908 { /* We have .*\n. */
1909 STORE_JUMP (jump
, b
, laststart
);
1910 keep_string_p
= true;
1913 /* Anything else. */
1914 STORE_JUMP (maybe_pop_jump
, b
, laststart
- 3);
1916 /* We've added more stuff to the buffer. */
1920 /* On failure, jump from laststart to b + 3, which will be the
1921 end of the buffer after this jump is inserted. */
1922 GET_BUFFER_SPACE (3);
1923 INSERT_JUMP (keep_string_p
? on_failure_keep_string_jump
1931 /* At least one repetition is required, so insert a
1932 `dummy_failure_jump' before the initial
1933 `on_failure_jump' instruction of the loop. This
1934 effects a skip over that instruction the first time
1935 we hit that loop. */
1936 GET_BUFFER_SPACE (3);
1937 INSERT_JUMP (dummy_failure_jump
, laststart
, laststart
+ 6);
1952 boolean had_char_class
= false;
1954 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
1956 /* Ensure that we have enough space to push a charset: the
1957 opcode, the length count, and the bitset; 34 bytes in all. */
1958 GET_BUFFER_SPACE (34);
1962 /* We test `*p == '^' twice, instead of using an if
1963 statement, so we only need one BUF_PUSH. */
1964 BUF_PUSH (*p
== '^' ? charset_not
: charset
);
1968 /* Remember the first position in the bracket expression. */
1971 /* Push the number of bytes in the bitmap. */
1972 BUF_PUSH ((1 << BYTEWIDTH
) / BYTEWIDTH
);
1974 /* Clear the whole map. */
1975 bzero (b
, (1 << BYTEWIDTH
) / BYTEWIDTH
);
1977 /* charset_not matches newline according to a syntax bit. */
1978 if ((re_opcode_t
) b
[-2] == charset_not
1979 && (syntax
& RE_HAT_LISTS_NOT_NEWLINE
))
1980 SET_LIST_BIT ('\n');
1982 /* Read in characters and ranges, setting map bits. */
1985 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
1989 /* \ might escape characters inside [...] and [^...]. */
1990 if ((syntax
& RE_BACKSLASH_ESCAPE_IN_LISTS
) && c
== '\\')
1992 if (p
== pend
) FREE_STACK_RETURN (REG_EESCAPE
);
1999 /* Could be the end of the bracket expression. If it's
2000 not (i.e., when the bracket expression is `[]' so
2001 far), the ']' character bit gets set way below. */
2002 if (c
== ']' && p
!= p1
+ 1)
2005 /* Look ahead to see if it's a range when the last thing
2006 was a character class. */
2007 if (had_char_class
&& c
== '-' && *p
!= ']')
2008 FREE_STACK_RETURN (REG_ERANGE
);
2010 /* Look ahead to see if it's a range when the last thing
2011 was a character: if this is a hyphen not at the
2012 beginning or the end of a list, then it's the range
2015 && !(p
- 2 >= pattern
&& p
[-2] == '[')
2016 && !(p
- 3 >= pattern
&& p
[-3] == '[' && p
[-2] == '^')
2020 = compile_range (&p
, pend
, translate
, syntax
, b
);
2021 if (ret
!= REG_NOERROR
) FREE_STACK_RETURN (ret
);
2024 else if (p
[0] == '-' && p
[1] != ']')
2025 { /* This handles ranges made up of characters only. */
2028 /* Move past the `-'. */
2031 ret
= compile_range (&p
, pend
, translate
, syntax
, b
);
2032 if (ret
!= REG_NOERROR
) FREE_STACK_RETURN (ret
);
2035 /* See if we're at the beginning of a possible character
2038 else if (syntax
& RE_CHAR_CLASSES
&& c
== '[' && *p
== ':')
2039 { /* Leave room for the null. */
2040 char str
[CHAR_CLASS_MAX_LENGTH
+ 1];
2045 /* If pattern is `[[:'. */
2046 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2051 if (c
== ':' || c
== ']' || p
== pend
2052 || c1
== CHAR_CLASS_MAX_LENGTH
)
2058 /* If isn't a word bracketed by `[:' and:`]':
2059 undo the ending character, the letters, and leave
2060 the leading `:' and `[' (but set bits for them). */
2061 if (c
== ':' && *p
== ']')
2064 boolean is_alnum
= STREQ (str
, "alnum");
2065 boolean is_alpha
= STREQ (str
, "alpha");
2066 boolean is_blank
= STREQ (str
, "blank");
2067 boolean is_cntrl
= STREQ (str
, "cntrl");
2068 boolean is_digit
= STREQ (str
, "digit");
2069 boolean is_graph
= STREQ (str
, "graph");
2070 boolean is_lower
= STREQ (str
, "lower");
2071 boolean is_print
= STREQ (str
, "print");
2072 boolean is_punct
= STREQ (str
, "punct");
2073 boolean is_space
= STREQ (str
, "space");
2074 boolean is_upper
= STREQ (str
, "upper");
2075 boolean is_xdigit
= STREQ (str
, "xdigit");
2077 if (!IS_CHAR_CLASS (str
))
2078 FREE_STACK_RETURN (REG_ECTYPE
);
2080 /* Throw away the ] at the end of the character
2084 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2086 for (ch
= 0; ch
< 1 << BYTEWIDTH
; ch
++)
2088 /* This was split into 3 if's to
2089 avoid an arbitrary limit in some compiler. */
2090 if ( (is_alnum
&& ISALNUM (ch
))
2091 || (is_alpha
&& ISALPHA (ch
))
2092 || (is_blank
&& ISBLANK (ch
))
2093 || (is_cntrl
&& ISCNTRL (ch
)))
2095 if ( (is_digit
&& ISDIGIT (ch
))
2096 || (is_graph
&& ISGRAPH (ch
))
2097 || (is_lower
&& ISLOWER (ch
))
2098 || (is_print
&& ISPRINT (ch
)))
2100 if ( (is_punct
&& ISPUNCT (ch
))
2101 || (is_space
&& ISSPACE (ch
))
2102 || (is_upper
&& ISUPPER (ch
))
2103 || (is_xdigit
&& ISXDIGIT (ch
)))
2106 had_char_class
= true;
2115 had_char_class
= false;
2120 had_char_class
= false;
2125 /* Discard any (non)matching list bytes that are all 0 at the
2126 end of the map. Decrease the map-length byte too. */
2127 while ((int) b
[-1] > 0 && b
[b
[-1] - 1] == 0)
2135 if (syntax
& RE_NO_BK_PARENS
)
2142 if (syntax
& RE_NO_BK_PARENS
)
2149 if (syntax
& RE_NEWLINE_ALT
)
2156 if (syntax
& RE_NO_BK_VBAR
)
2163 if (syntax
& RE_INTERVALS
&& syntax
& RE_NO_BK_BRACES
)
2164 goto handle_interval
;
2170 if (p
== pend
) FREE_STACK_RETURN (REG_EESCAPE
);
2172 /* Do not translate the character after the \, so that we can
2173 distinguish, e.g., \B from \b, even if we normally would
2174 translate, e.g., B to b. */
2180 if (syntax
& RE_NO_BK_PARENS
)
2181 goto normal_backslash
;
2187 if (COMPILE_STACK_FULL
)
2189 RETALLOC (compile_stack
.stack
, compile_stack
.size
<< 1,
2190 compile_stack_elt_t
);
2191 if (compile_stack
.stack
== NULL
) return REG_ESPACE
;
2193 compile_stack
.size
<<= 1;
2196 /* These are the values to restore when we hit end of this
2197 group. They are all relative offsets, so that if the
2198 whole pattern moves because of realloc, they will still
2200 COMPILE_STACK_TOP
.begalt_offset
= begalt
- bufp
->buffer
;
2201 COMPILE_STACK_TOP
.fixup_alt_jump
2202 = fixup_alt_jump
? fixup_alt_jump
- bufp
->buffer
+ 1 : 0;
2203 COMPILE_STACK_TOP
.laststart_offset
= b
- bufp
->buffer
;
2204 COMPILE_STACK_TOP
.regnum
= regnum
;
2206 /* We will eventually replace the 0 with the number of
2207 groups inner to this one. But do not push a
2208 start_memory for groups beyond the last one we can
2209 represent in the compiled pattern. */
2210 if (regnum
<= MAX_REGNUM
)
2212 COMPILE_STACK_TOP
.inner_group_offset
= b
- bufp
->buffer
+ 2;
2213 BUF_PUSH_3 (start_memory
, regnum
, 0);
2216 compile_stack
.avail
++;
2221 /* If we've reached MAX_REGNUM groups, then this open
2222 won't actually generate any code, so we'll have to
2223 clear pending_exact explicitly. */
2229 if (syntax
& RE_NO_BK_PARENS
) goto normal_backslash
;
2231 if (COMPILE_STACK_EMPTY
)
2232 if (syntax
& RE_UNMATCHED_RIGHT_PAREN_ORD
)
2233 goto normal_backslash
;
2235 FREE_STACK_RETURN (REG_ERPAREN
);
2239 { /* Push a dummy failure point at the end of the
2240 alternative for a possible future
2241 `pop_failure_jump' to pop. See comments at
2242 `push_dummy_failure' in `re_match_2'. */
2243 BUF_PUSH (push_dummy_failure
);
2245 /* We allocated space for this jump when we assigned
2246 to `fixup_alt_jump', in the `handle_alt' case below. */
2247 STORE_JUMP (jump_past_alt
, fixup_alt_jump
, b
- 1);
2250 /* See similar code for backslashed left paren above. */
2251 if (COMPILE_STACK_EMPTY
)
2252 if (syntax
& RE_UNMATCHED_RIGHT_PAREN_ORD
)
2255 FREE_STACK_RETURN (REG_ERPAREN
);
2257 /* Since we just checked for an empty stack above, this
2258 ``can't happen''. */
2259 assert (compile_stack
.avail
!= 0);
2261 /* We don't just want to restore into `regnum', because
2262 later groups should continue to be numbered higher,
2263 as in `(ab)c(de)' -- the second group is #2. */
2264 regnum_t this_group_regnum
;
2266 compile_stack
.avail
--;
2267 begalt
= bufp
->buffer
+ COMPILE_STACK_TOP
.begalt_offset
;
2269 = COMPILE_STACK_TOP
.fixup_alt_jump
2270 ? bufp
->buffer
+ COMPILE_STACK_TOP
.fixup_alt_jump
- 1
2272 laststart
= bufp
->buffer
+ COMPILE_STACK_TOP
.laststart_offset
;
2273 this_group_regnum
= COMPILE_STACK_TOP
.regnum
;
2274 /* If we've reached MAX_REGNUM groups, then this open
2275 won't actually generate any code, so we'll have to
2276 clear pending_exact explicitly. */
2279 /* We're at the end of the group, so now we know how many
2280 groups were inside this one. */
2281 if (this_group_regnum
<= MAX_REGNUM
)
2283 unsigned char *inner_group_loc
2284 = bufp
->buffer
+ COMPILE_STACK_TOP
.inner_group_offset
;
2286 *inner_group_loc
= regnum
- this_group_regnum
;
2287 BUF_PUSH_3 (stop_memory
, this_group_regnum
,
2288 regnum
- this_group_regnum
);
2294 case '|': /* `\|'. */
2295 if (syntax
& RE_LIMITED_OPS
|| syntax
& RE_NO_BK_VBAR
)
2296 goto normal_backslash
;
2298 if (syntax
& RE_LIMITED_OPS
)
2301 /* Insert before the previous alternative a jump which
2302 jumps to this alternative if the former fails. */
2303 GET_BUFFER_SPACE (3);
2304 INSERT_JUMP (on_failure_jump
, begalt
, b
+ 6);
2308 /* The alternative before this one has a jump after it
2309 which gets executed if it gets matched. Adjust that
2310 jump so it will jump to this alternative's analogous
2311 jump (put in below, which in turn will jump to the next
2312 (if any) alternative's such jump, etc.). The last such
2313 jump jumps to the correct final destination. A picture:
2319 If we are at `b', then fixup_alt_jump right now points to a
2320 three-byte space after `a'. We'll put in the jump, set
2321 fixup_alt_jump to right after `b', and leave behind three
2322 bytes which we'll fill in when we get to after `c'. */
2325 STORE_JUMP (jump_past_alt
, fixup_alt_jump
, b
);
2327 /* Mark and leave space for a jump after this alternative,
2328 to be filled in later either by next alternative or
2329 when know we're at the end of a series of alternatives. */
2331 GET_BUFFER_SPACE (3);
2340 /* If \{ is a literal. */
2341 if (!(syntax
& RE_INTERVALS
)
2342 /* If we're at `\{' and it's not the open-interval
2344 || ((syntax
& RE_INTERVALS
) && (syntax
& RE_NO_BK_BRACES
))
2345 || (p
- 2 == pattern
&& p
== pend
))
2346 goto normal_backslash
;
2350 /* If got here, then the syntax allows intervals. */
2352 /* At least (most) this many matches must be made. */
2353 int lower_bound
= -1, upper_bound
= -1;
2355 beg_interval
= p
- 1;
2359 if (syntax
& RE_NO_BK_BRACES
)
2360 goto unfetch_interval
;
2362 FREE_STACK_RETURN (REG_EBRACE
);
2365 GET_UNSIGNED_NUMBER (lower_bound
);
2369 GET_UNSIGNED_NUMBER (upper_bound
);
2370 if (upper_bound
< 0) upper_bound
= RE_DUP_MAX
;
2373 /* Interval such as `{1}' => match exactly once. */
2374 upper_bound
= lower_bound
;
2376 if (lower_bound
< 0 || upper_bound
> RE_DUP_MAX
2377 || lower_bound
> upper_bound
)
2379 if (syntax
& RE_NO_BK_BRACES
)
2380 goto unfetch_interval
;
2382 FREE_STACK_RETURN (REG_BADBR
);
2385 if (!(syntax
& RE_NO_BK_BRACES
))
2387 if (c
!= '\\') FREE_STACK_RETURN (REG_EBRACE
);
2394 if (syntax
& RE_NO_BK_BRACES
)
2395 goto unfetch_interval
;
2397 FREE_STACK_RETURN (REG_BADBR
);
2400 /* We just parsed a valid interval. */
2402 /* If it's invalid to have no preceding re. */
2405 if (syntax
& RE_CONTEXT_INVALID_OPS
)
2406 FREE_STACK_RETURN (REG_BADRPT
);
2407 else if (syntax
& RE_CONTEXT_INDEP_OPS
)
2410 goto unfetch_interval
;
2413 /* If the upper bound is zero, don't want to succeed at
2414 all; jump from `laststart' to `b + 3', which will be
2415 the end of the buffer after we insert the jump. */
2416 if (upper_bound
== 0)
2418 GET_BUFFER_SPACE (3);
2419 INSERT_JUMP (jump
, laststart
, b
+ 3);
2423 /* Otherwise, we have a nontrivial interval. When
2424 we're all done, the pattern will look like:
2425 set_number_at <jump count> <upper bound>
2426 set_number_at <succeed_n count> <lower bound>
2427 succeed_n <after jump addr> <succeed_n count>
2429 jump_n <succeed_n addr> <jump count>
2430 (The upper bound and `jump_n' are omitted if
2431 `upper_bound' is 1, though.) */
2433 { /* If the upper bound is > 1, we need to insert
2434 more at the end of the loop. */
2435 unsigned nbytes
= 10 + (upper_bound
> 1) * 10;
2437 GET_BUFFER_SPACE (nbytes
);
2439 /* Initialize lower bound of the `succeed_n', even
2440 though it will be set during matching by its
2441 attendant `set_number_at' (inserted next),
2442 because `re_compile_fastmap' needs to know.
2443 Jump to the `jump_n' we might insert below. */
2444 INSERT_JUMP2 (succeed_n
, laststart
,
2445 b
+ 5 + (upper_bound
> 1) * 5,
2449 /* Code to initialize the lower bound. Insert
2450 before the `succeed_n'. The `5' is the last two
2451 bytes of this `set_number_at', plus 3 bytes of
2452 the following `succeed_n'. */
2453 insert_op2 (set_number_at
, laststart
, 5, lower_bound
, b
);
2456 if (upper_bound
> 1)
2457 { /* More than one repetition is allowed, so
2458 append a backward jump to the `succeed_n'
2459 that starts this interval.
2461 When we've reached this during matching,
2462 we'll have matched the interval once, so
2463 jump back only `upper_bound - 1' times. */
2464 STORE_JUMP2 (jump_n
, b
, laststart
+ 5,
2468 /* The location we want to set is the second
2469 parameter of the `jump_n'; that is `b-2' as
2470 an absolute address. `laststart' will be
2471 the `set_number_at' we're about to insert;
2472 `laststart+3' the number to set, the source
2473 for the relative address. But we are
2474 inserting into the middle of the pattern --
2475 so everything is getting moved up by 5.
2476 Conclusion: (b - 2) - (laststart + 3) + 5,
2477 i.e., b - laststart.
2479 We insert this at the beginning of the loop
2480 so that if we fail during matching, we'll
2481 reinitialize the bounds. */
2482 insert_op2 (set_number_at
, laststart
, b
- laststart
,
2483 upper_bound
- 1, b
);
2488 beg_interval
= NULL
;
2493 /* If an invalid interval, match the characters as literals. */
2494 assert (beg_interval
);
2496 beg_interval
= NULL
;
2498 /* normal_char and normal_backslash need `c'. */
2501 if (!(syntax
& RE_NO_BK_BRACES
))
2503 if (p
> pattern
&& p
[-1] == '\\')
2504 goto normal_backslash
;
2509 /* There is no way to specify the before_dot and after_dot
2510 operators. rms says this is ok. --karl */
2518 BUF_PUSH_2 (syntaxspec
, syntax_spec_code
[c
]);
2524 BUF_PUSH_2 (notsyntaxspec
, syntax_spec_code
[c
]);
2531 BUF_PUSH (wordchar
);
2537 BUF_PUSH (notwordchar
);
2550 BUF_PUSH (wordbound
);
2554 BUF_PUSH (notwordbound
);
2565 case '1': case '2': case '3': case '4': case '5':
2566 case '6': case '7': case '8': case '9':
2567 if (syntax
& RE_NO_BK_REFS
)
2573 FREE_STACK_RETURN (REG_ESUBREG
);
2575 /* Can't back reference to a subexpression if inside of it. */
2576 if (group_in_compile_stack (compile_stack
, c1
))
2580 BUF_PUSH_2 (duplicate
, c1
);
2586 if (syntax
& RE_BK_PLUS_QM
)
2589 goto normal_backslash
;
2593 /* You might think it would be useful for \ to mean
2594 not to translate; but if we don't translate it
2595 it will never match anything. */
2603 /* Expects the character in `c'. */
2605 /* If no exactn currently being built. */
2608 /* If last exactn not at current position. */
2609 || pending_exact
+ *pending_exact
+ 1 != b
2611 /* We have only one byte following the exactn for the count. */
2612 || *pending_exact
== (1 << BYTEWIDTH
) - 1
2614 /* If followed by a repetition operator. */
2615 || *p
== '*' || *p
== '^'
2616 || ((syntax
& RE_BK_PLUS_QM
)
2617 ? *p
== '\\' && (p
[1] == '+' || p
[1] == '?')
2618 : (*p
== '+' || *p
== '?'))
2619 || ((syntax
& RE_INTERVALS
)
2620 && ((syntax
& RE_NO_BK_BRACES
)
2622 : (p
[0] == '\\' && p
[1] == '{'))))
2624 /* Start building a new exactn. */
2628 BUF_PUSH_2 (exactn
, 0);
2629 pending_exact
= b
- 1;
2636 } /* while p != pend */
2639 /* Through the pattern now. */
2642 STORE_JUMP (jump_past_alt
, fixup_alt_jump
, b
);
2644 if (!COMPILE_STACK_EMPTY
)
2645 FREE_STACK_RETURN (REG_EPAREN
);
2647 /* If we don't want backtracking, force success
2648 the first time we reach the end of the compiled pattern. */
2649 if (syntax
& RE_NO_POSIX_BACKTRACKING
)
2652 free (compile_stack
.stack
);
2654 /* We have succeeded; set the length of the buffer. */
2655 bufp
->used
= b
- bufp
->buffer
;
2660 DEBUG_PRINT1 ("\nCompiled pattern: \n");
2661 print_compiled_pattern (bufp
);
2665 #ifndef MATCH_MAY_ALLOCATE
2666 /* Initialize the failure stack to the largest possible stack. This
2667 isn't necessary unless we're trying to avoid calling alloca in
2668 the search and match routines. */
2670 int num_regs
= bufp
->re_nsub
+ 1;
2672 /* Since DOUBLE_FAIL_STACK refuses to double only if the current size
2673 is strictly greater than re_max_failures, the largest possible stack
2674 is 2 * re_max_failures failure points. */
2675 if (fail_stack
.size
< (2 * re_max_failures
* MAX_FAILURE_ITEMS
))
2677 fail_stack
.size
= (2 * re_max_failures
* MAX_FAILURE_ITEMS
);
2680 if (! fail_stack
.stack
)
2682 = (fail_stack_elt_t
*) xmalloc (fail_stack
.size
2683 * sizeof (fail_stack_elt_t
));
2686 = (fail_stack_elt_t
*) xrealloc (fail_stack
.stack
,
2688 * sizeof (fail_stack_elt_t
)));
2689 #else /* not emacs */
2690 if (! fail_stack
.stack
)
2692 = (fail_stack_elt_t
*) malloc (fail_stack
.size
2693 * sizeof (fail_stack_elt_t
));
2696 = (fail_stack_elt_t
*) realloc (fail_stack
.stack
,
2698 * sizeof (fail_stack_elt_t
)));
2699 #endif /* not emacs */
2702 regex_grow_registers (num_regs
);
2704 #endif /* not MATCH_MAY_ALLOCATE */
2707 } /* regex_compile */
2709 /* Subroutines for `regex_compile'. */
2711 /* Store OP at LOC followed by two-byte integer parameter ARG. */
2714 store_op1 (op
, loc
, arg
)
2719 *loc
= (unsigned char) op
;
2720 STORE_NUMBER (loc
+ 1, arg
);
2724 /* Like `store_op1', but for two two-byte parameters ARG1 and ARG2. */
2727 store_op2 (op
, loc
, arg1
, arg2
)
2732 *loc
= (unsigned char) op
;
2733 STORE_NUMBER (loc
+ 1, arg1
);
2734 STORE_NUMBER (loc
+ 3, arg2
);
2738 /* Copy the bytes from LOC to END to open up three bytes of space at LOC
2739 for OP followed by two-byte integer parameter ARG. */
2742 insert_op1 (op
, loc
, arg
, end
)
2748 register unsigned char *pfrom
= end
;
2749 register unsigned char *pto
= end
+ 3;
2751 while (pfrom
!= loc
)
2754 store_op1 (op
, loc
, arg
);
2758 /* Like `insert_op1', but for two two-byte parameters ARG1 and ARG2. */
2761 insert_op2 (op
, loc
, arg1
, arg2
, end
)
2767 register unsigned char *pfrom
= end
;
2768 register unsigned char *pto
= end
+ 5;
2770 while (pfrom
!= loc
)
2773 store_op2 (op
, loc
, arg1
, arg2
);
2777 /* P points to just after a ^ in PATTERN. Return true if that ^ comes
2778 after an alternative or a begin-subexpression. We assume there is at
2779 least one character before the ^. */
2782 at_begline_loc_p (pattern
, p
, syntax
)
2783 const char *pattern
, *p
;
2784 reg_syntax_t syntax
;
2786 const char *prev
= p
- 2;
2787 boolean prev_prev_backslash
= prev
> pattern
&& prev
[-1] == '\\';
2790 /* After a subexpression? */
2791 (*prev
== '(' && (syntax
& RE_NO_BK_PARENS
|| prev_prev_backslash
))
2792 /* After an alternative? */
2793 || (*prev
== '|' && (syntax
& RE_NO_BK_VBAR
|| prev_prev_backslash
));
2797 /* The dual of at_begline_loc_p. This one is for $. We assume there is
2798 at least one character after the $, i.e., `P < PEND'. */
2801 at_endline_loc_p (p
, pend
, syntax
)
2802 const char *p
, *pend
;
2805 const char *next
= p
;
2806 boolean next_backslash
= *next
== '\\';
2807 const char *next_next
= p
+ 1 < pend
? p
+ 1 : 0;
2810 /* Before a subexpression? */
2811 (syntax
& RE_NO_BK_PARENS
? *next
== ')'
2812 : next_backslash
&& next_next
&& *next_next
== ')')
2813 /* Before an alternative? */
2814 || (syntax
& RE_NO_BK_VBAR
? *next
== '|'
2815 : next_backslash
&& next_next
&& *next_next
== '|');
2819 /* Returns true if REGNUM is in one of COMPILE_STACK's elements and
2820 false if it's not. */
2823 group_in_compile_stack (compile_stack
, regnum
)
2824 compile_stack_type compile_stack
;
2829 for (this_element
= compile_stack
.avail
- 1;
2832 if (compile_stack
.stack
[this_element
].regnum
== regnum
)
2839 /* Read the ending character of a range (in a bracket expression) from the
2840 uncompiled pattern *P_PTR (which ends at PEND). We assume the
2841 starting character is in `P[-2]'. (`P[-1]' is the character `-'.)
2842 Then we set the translation of all bits between the starting and
2843 ending characters (inclusive) in the compiled pattern B.
2845 Return an error code.
2847 We use these short variable names so we can use the same macros as
2848 `regex_compile' itself. */
2850 static reg_errcode_t
2851 compile_range (p_ptr
, pend
, translate
, syntax
, b
)
2852 const char **p_ptr
, *pend
;
2853 RE_TRANSLATE_TYPE translate
;
2854 reg_syntax_t syntax
;
2859 const char *p
= *p_ptr
;
2860 int range_start
, range_end
;
2865 /* Even though the pattern is a signed `char *', we need to fetch
2866 with unsigned char *'s; if the high bit of the pattern character
2867 is set, the range endpoints will be negative if we fetch using a
2870 We also want to fetch the endpoints without translating them; the
2871 appropriate translation is done in the bit-setting loop below. */
2872 /* The SVR4 compiler on the 3B2 had trouble with unsigned const char *. */
2873 range_start
= ((const unsigned char *) p
)[-2];
2874 range_end
= ((const unsigned char *) p
)[0];
2876 /* Have to increment the pointer into the pattern string, so the
2877 caller isn't still at the ending character. */
2880 /* If the start is after the end, the range is empty. */
2881 if (range_start
> range_end
)
2882 return syntax
& RE_NO_EMPTY_RANGES
? REG_ERANGE
: REG_NOERROR
;
2884 /* Here we see why `this_char' has to be larger than an `unsigned
2885 char' -- the range is inclusive, so if `range_end' == 0xff
2886 (assuming 8-bit characters), we would otherwise go into an infinite
2887 loop, since all characters <= 0xff. */
2888 for (this_char
= range_start
; this_char
<= range_end
; this_char
++)
2890 SET_LIST_BIT (TRANSLATE (this_char
));
2896 /* re_compile_fastmap computes a ``fastmap'' for the compiled pattern in
2897 BUFP. A fastmap records which of the (1 << BYTEWIDTH) possible
2898 characters can start a string that matches the pattern. This fastmap
2899 is used by re_search to skip quickly over impossible starting points.
2901 The caller must supply the address of a (1 << BYTEWIDTH)-byte data
2902 area as BUFP->fastmap.
2904 We set the `fastmap', `fastmap_accurate', and `can_be_null' fields in
2907 Returns 0 if we succeed, -2 if an internal error. */
2910 re_compile_fastmap (bufp
)
2911 struct re_pattern_buffer
*bufp
;
2914 #ifdef MATCH_MAY_ALLOCATE
2915 fail_stack_type fail_stack
;
2917 #ifndef REGEX_MALLOC
2920 /* We don't push any register information onto the failure stack. */
2921 unsigned num_regs
= 0;
2923 register char *fastmap
= bufp
->fastmap
;
2924 unsigned char *pattern
= bufp
->buffer
;
2925 unsigned long size
= bufp
->used
;
2926 unsigned char *p
= pattern
;
2927 register unsigned char *pend
= pattern
+ size
;
2929 /* This holds the pointer to the failure stack, when
2930 it is allocated relocatably. */
2931 fail_stack_elt_t
*failure_stack_ptr
;
2933 /* Assume that each path through the pattern can be null until
2934 proven otherwise. We set this false at the bottom of switch
2935 statement, to which we get only if a particular path doesn't
2936 match the empty string. */
2937 boolean path_can_be_null
= true;
2939 /* We aren't doing a `succeed_n' to begin with. */
2940 boolean succeed_n_p
= false;
2942 assert (fastmap
!= NULL
&& p
!= NULL
);
2945 bzero (fastmap
, 1 << BYTEWIDTH
); /* Assume nothing's valid. */
2946 bufp
->fastmap_accurate
= 1; /* It will be when we're done. */
2947 bufp
->can_be_null
= 0;
2951 if (p
== pend
|| *p
== succeed
)
2953 /* We have reached the (effective) end of pattern. */
2954 if (!FAIL_STACK_EMPTY ())
2956 bufp
->can_be_null
|= path_can_be_null
;
2958 /* Reset for next path. */
2959 path_can_be_null
= true;
2961 p
= fail_stack
.stack
[--fail_stack
.avail
].pointer
;
2969 /* We should never be about to go beyond the end of the pattern. */
2972 switch (SWITCH_ENUM_CAST ((re_opcode_t
) *p
++))
2975 /* I guess the idea here is to simply not bother with a fastmap
2976 if a backreference is used, since it's too hard to figure out
2977 the fastmap for the corresponding group. Setting
2978 `can_be_null' stops `re_search_2' from using the fastmap, so
2979 that is all we do. */
2981 bufp
->can_be_null
= 1;
2985 /* Following are the cases which match a character. These end
2994 for (j
= *p
++ * BYTEWIDTH
- 1; j
>= 0; j
--)
2995 if (p
[j
/ BYTEWIDTH
] & (1 << (j
% BYTEWIDTH
)))
3001 /* Chars beyond end of map must be allowed. */
3002 for (j
= *p
* BYTEWIDTH
; j
< (1 << BYTEWIDTH
); j
++)
3005 for (j
= *p
++ * BYTEWIDTH
- 1; j
>= 0; j
--)
3006 if (!(p
[j
/ BYTEWIDTH
] & (1 << (j
% BYTEWIDTH
))))
3012 for (j
= 0; j
< (1 << BYTEWIDTH
); j
++)
3013 if (SYNTAX (j
) == Sword
)
3019 for (j
= 0; j
< (1 << BYTEWIDTH
); j
++)
3020 if (SYNTAX (j
) != Sword
)
3027 int fastmap_newline
= fastmap
['\n'];
3029 /* `.' matches anything ... */
3030 for (j
= 0; j
< (1 << BYTEWIDTH
); j
++)
3033 /* ... except perhaps newline. */
3034 if (!(bufp
->syntax
& RE_DOT_NEWLINE
))
3035 fastmap
['\n'] = fastmap_newline
;
3037 /* Return if we have already set `can_be_null'; if we have,
3038 then the fastmap is irrelevant. Something's wrong here. */
3039 else if (bufp
->can_be_null
)
3042 /* Otherwise, have to check alternative paths. */
3049 for (j
= 0; j
< (1 << BYTEWIDTH
); j
++)
3050 if (SYNTAX (j
) == (enum syntaxcode
) k
)
3057 for (j
= 0; j
< (1 << BYTEWIDTH
); j
++)
3058 if (SYNTAX (j
) != (enum syntaxcode
) k
)
3063 /* All cases after this match the empty string. These end with
3083 case push_dummy_failure
:
3088 case pop_failure_jump
:
3089 case maybe_pop_jump
:
3092 case dummy_failure_jump
:
3093 EXTRACT_NUMBER_AND_INCR (j
, p
);
3098 /* Jump backward implies we just went through the body of a
3099 loop and matched nothing. Opcode jumped to should be
3100 `on_failure_jump' or `succeed_n'. Just treat it like an
3101 ordinary jump. For a * loop, it has pushed its failure
3102 point already; if so, discard that as redundant. */
3103 if ((re_opcode_t
) *p
!= on_failure_jump
3104 && (re_opcode_t
) *p
!= succeed_n
)
3108 EXTRACT_NUMBER_AND_INCR (j
, p
);
3111 /* If what's on the stack is where we are now, pop it. */
3112 if (!FAIL_STACK_EMPTY ()
3113 && fail_stack
.stack
[fail_stack
.avail
- 1].pointer
== p
)
3119 case on_failure_jump
:
3120 case on_failure_keep_string_jump
:
3121 handle_on_failure_jump
:
3122 EXTRACT_NUMBER_AND_INCR (j
, p
);
3124 /* For some patterns, e.g., `(a?)?', `p+j' here points to the
3125 end of the pattern. We don't want to push such a point,
3126 since when we restore it above, entering the switch will
3127 increment `p' past the end of the pattern. We don't need
3128 to push such a point since we obviously won't find any more
3129 fastmap entries beyond `pend'. Such a pattern can match
3130 the null string, though. */
3133 if (!PUSH_PATTERN_OP (p
+ j
, fail_stack
))
3135 RESET_FAIL_STACK ();
3140 bufp
->can_be_null
= 1;
3144 EXTRACT_NUMBER_AND_INCR (k
, p
); /* Skip the n. */
3145 succeed_n_p
= false;
3152 /* Get to the number of times to succeed. */
3155 /* Increment p past the n for when k != 0. */
3156 EXTRACT_NUMBER_AND_INCR (k
, p
);
3160 succeed_n_p
= true; /* Spaghetti code alert. */
3161 goto handle_on_failure_jump
;
3178 abort (); /* We have listed all the cases. */
3181 /* Getting here means we have found the possible starting
3182 characters for one path of the pattern -- and that the empty
3183 string does not match. We need not follow this path further.
3184 Instead, look at the next alternative (remembered on the
3185 stack), or quit if no more. The test at the top of the loop
3186 does these things. */
3187 path_can_be_null
= false;
3191 /* Set `can_be_null' for the last path (also the first path, if the
3192 pattern is empty). */
3193 bufp
->can_be_null
|= path_can_be_null
;
3196 RESET_FAIL_STACK ();
3198 } /* re_compile_fastmap */
3200 /* Set REGS to hold NUM_REGS registers, storing them in STARTS and
3201 ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use
3202 this memory for recording register information. STARTS and ENDS
3203 must be allocated using the malloc library routine, and must each
3204 be at least NUM_REGS * sizeof (regoff_t) bytes long.
3206 If NUM_REGS == 0, then subsequent matches should allocate their own
3209 Unless this function is called, the first search or match using
3210 PATTERN_BUFFER will allocate its own register data, without
3211 freeing the old data. */
3214 re_set_registers (bufp
, regs
, num_regs
, starts
, ends
)
3215 struct re_pattern_buffer
*bufp
;
3216 struct re_registers
*regs
;
3218 regoff_t
*starts
, *ends
;
3222 bufp
->regs_allocated
= REGS_REALLOCATE
;
3223 regs
->num_regs
= num_regs
;
3224 regs
->start
= starts
;
3229 bufp
->regs_allocated
= REGS_UNALLOCATED
;
3231 regs
->start
= regs
->end
= (regoff_t
*) 0;
3235 /* Searching routines. */
3237 /* Like re_search_2, below, but only one string is specified, and
3238 doesn't let you say where to stop matching. */
3241 re_search (bufp
, string
, size
, startpos
, range
, regs
)
3242 struct re_pattern_buffer
*bufp
;
3244 int size
, startpos
, range
;
3245 struct re_registers
*regs
;
3247 return re_search_2 (bufp
, NULL
, 0, string
, size
, startpos
, range
,
3252 /* Using the compiled pattern in BUFP->buffer, first tries to match the
3253 virtual concatenation of STRING1 and STRING2, starting first at index
3254 STARTPOS, then at STARTPOS + 1, and so on.
3256 STRING1 and STRING2 have length SIZE1 and SIZE2, respectively.
3258 RANGE is how far to scan while trying to match. RANGE = 0 means try
3259 only at STARTPOS; in general, the last start tried is STARTPOS +
3262 In REGS, return the indices of the virtual concatenation of STRING1
3263 and STRING2 that matched the entire BUFP->buffer and its contained
3266 Do not consider matching one past the index STOP in the virtual
3267 concatenation of STRING1 and STRING2.
3269 We return either the position in the strings at which the match was
3270 found, -1 if no match, or -2 if error (such as failure
3274 re_search_2 (bufp
, string1
, size1
, string2
, size2
, startpos
, range
, regs
, stop
)
3275 struct re_pattern_buffer
*bufp
;
3276 const char *string1
, *string2
;
3280 struct re_registers
*regs
;
3284 register char *fastmap
= bufp
->fastmap
;
3285 register RE_TRANSLATE_TYPE translate
= bufp
->translate
;
3286 int total_size
= size1
+ size2
;
3287 int endpos
= startpos
+ range
;
3289 /* Check for out-of-range STARTPOS. */
3290 if (startpos
< 0 || startpos
> total_size
)
3293 /* Fix up RANGE if it might eventually take us outside
3294 the virtual concatenation of STRING1 and STRING2.
3295 Make sure we won't move STARTPOS below 0 or above TOTAL_SIZE. */
3297 range
= 0 - startpos
;
3298 else if (endpos
> total_size
)
3299 range
= total_size
- startpos
;
3301 /* If the search isn't to be a backwards one, don't waste time in a
3302 search for a pattern that must be anchored. */
3303 if (bufp
->used
> 0 && (re_opcode_t
) bufp
->buffer
[0] == begbuf
&& range
> 0)
3312 /* In a forward search for something that starts with \=.
3313 don't keep searching past point. */
3314 if (bufp
->used
> 0 && (re_opcode_t
) bufp
->buffer
[0] == at_dot
&& range
> 0)
3316 range
= PT
- startpos
;
3322 /* Update the fastmap now if not correct already. */
3323 if (fastmap
&& !bufp
->fastmap_accurate
)
3324 if (re_compile_fastmap (bufp
) == -2)
3327 /* Loop through the string, looking for a place to start matching. */
3330 /* If a fastmap is supplied, skip quickly over characters that
3331 cannot be the start of a match. If the pattern can match the
3332 null string, however, we don't need to skip characters; we want
3333 the first null string. */
3334 if (fastmap
&& startpos
< total_size
&& !bufp
->can_be_null
)
3336 if (range
> 0) /* Searching forwards. */
3338 register const char *d
;
3339 register int lim
= 0;
3342 if (startpos
< size1
&& startpos
+ range
>= size1
)
3343 lim
= range
- (size1
- startpos
);
3345 d
= (startpos
>= size1
? string2
- size1
: string1
) + startpos
;
3347 /* Written out as an if-else to avoid testing `translate'
3351 && !fastmap
[(unsigned char)
3352 translate
[(unsigned char) *d
++]])
3355 while (range
> lim
&& !fastmap
[(unsigned char) *d
++])
3358 startpos
+= irange
- range
;
3360 else /* Searching backwards. */
3362 register char c
= (size1
== 0 || startpos
>= size1
3363 ? string2
[startpos
- size1
]
3364 : string1
[startpos
]);
3366 if (!fastmap
[(unsigned char) TRANSLATE (c
)])
3371 /* If can't match the null string, and that's all we have left, fail. */
3372 if (range
>= 0 && startpos
== total_size
&& fastmap
3373 && !bufp
->can_be_null
)
3376 val
= re_match_2_internal (bufp
, string1
, size1
, string2
, size2
,
3377 startpos
, regs
, stop
);
3378 #ifndef REGEX_MALLOC
3407 /* Declarations and macros for re_match_2. */
3409 static int bcmp_translate ();
3410 static boolean
alt_match_null_string_p (),
3411 common_op_match_null_string_p (),
3412 group_match_null_string_p ();
3414 /* This converts PTR, a pointer into one of the search strings `string1'
3415 and `string2' into an offset from the beginning of that string. */
3416 #define POINTER_TO_OFFSET(ptr) \
3417 (FIRST_STRING_P (ptr) \
3418 ? ((regoff_t) ((ptr) - string1)) \
3419 : ((regoff_t) ((ptr) - string2 + size1)))
3421 /* Macros for dealing with the split strings in re_match_2. */
3423 #define MATCHING_IN_FIRST_STRING (dend == end_match_1)
3425 /* Call before fetching a character with *d. This switches over to
3426 string2 if necessary. */
3427 #define PREFETCH() \
3430 /* End of string2 => fail. */ \
3431 if (dend == end_match_2) \
3433 /* End of string1 => advance to string2. */ \
3435 dend = end_match_2; \
3439 /* Test if at very beginning or at very end of the virtual concatenation
3440 of `string1' and `string2'. If only one string, it's `string2'. */
3441 #define AT_STRINGS_BEG(d) ((d) == (size1 ? string1 : string2) || !size2)
3442 #define AT_STRINGS_END(d) ((d) == end2)
3445 /* Test if D points to a character which is word-constituent. We have
3446 two special cases to check for: if past the end of string1, look at
3447 the first character in string2; and if before the beginning of
3448 string2, look at the last character in string1. */
3449 #define WORDCHAR_P(d) \
3450 (SYNTAX ((d) == end1 ? *string2 \
3451 : (d) == string2 - 1 ? *(end1 - 1) : *(d)) \
3454 /* Disabled due to a compiler bug -- see comment at case wordbound */
3456 /* Test if the character before D and the one at D differ with respect
3457 to being word-constituent. */
3458 #define AT_WORD_BOUNDARY(d) \
3459 (AT_STRINGS_BEG (d) || AT_STRINGS_END (d) \
3460 || WORDCHAR_P (d - 1) != WORDCHAR_P (d))
3463 /* Free everything we malloc. */
3464 #ifdef MATCH_MAY_ALLOCATE
3465 #define FREE_VAR(var) if (var) REGEX_FREE (var); var = NULL
3466 #define FREE_VARIABLES() \
3468 REGEX_FREE_STACK (fail_stack.stack); \
3469 FREE_VAR (regstart); \
3470 FREE_VAR (regend); \
3471 FREE_VAR (old_regstart); \
3472 FREE_VAR (old_regend); \
3473 FREE_VAR (best_regstart); \
3474 FREE_VAR (best_regend); \
3475 FREE_VAR (reg_info); \
3476 FREE_VAR (reg_dummy); \
3477 FREE_VAR (reg_info_dummy); \
3480 #define FREE_VARIABLES() ((void)0) /* Do nothing! But inhibit gcc warning. */
3481 #endif /* not MATCH_MAY_ALLOCATE */
3483 /* These values must meet several constraints. They must not be valid
3484 register values; since we have a limit of 255 registers (because
3485 we use only one byte in the pattern for the register number), we can
3486 use numbers larger than 255. They must differ by 1, because of
3487 NUM_FAILURE_ITEMS above. And the value for the lowest register must
3488 be larger than the value for the highest register, so we do not try
3489 to actually save any registers when none are active. */
3490 #define NO_HIGHEST_ACTIVE_REG (1 << BYTEWIDTH)
3491 #define NO_LOWEST_ACTIVE_REG (NO_HIGHEST_ACTIVE_REG + 1)
3493 /* Matching routines. */
3495 #ifndef emacs /* Emacs never uses this. */
3496 /* re_match is like re_match_2 except it takes only a single string. */
3499 re_match (bufp
, string
, size
, pos
, regs
)
3500 struct re_pattern_buffer
*bufp
;
3503 struct re_registers
*regs
;
3505 int result
= re_match_2_internal (bufp
, NULL
, 0, string
, size
,
3510 #endif /* not emacs */
3513 /* re_match_2 matches the compiled pattern in BUFP against the
3514 the (virtual) concatenation of STRING1 and STRING2 (of length SIZE1
3515 and SIZE2, respectively). We start matching at POS, and stop
3518 If REGS is non-null and the `no_sub' field of BUFP is nonzero, we
3519 store offsets for the substring each group matched in REGS. See the
3520 documentation for exactly how many groups we fill.
3522 We return -1 if no match, -2 if an internal error (such as the
3523 failure stack overflowing). Otherwise, we return the length of the
3524 matched substring. */
3527 re_match_2 (bufp
, string1
, size1
, string2
, size2
, pos
, regs
, stop
)
3528 struct re_pattern_buffer
*bufp
;
3529 const char *string1
, *string2
;
3532 struct re_registers
*regs
;
3535 int result
= re_match_2_internal (bufp
, string1
, size1
, string2
, size2
,
3541 /* This is a separate function so that we can force an alloca cleanup
3544 re_match_2_internal (bufp
, string1
, size1
, string2
, size2
, pos
, regs
, stop
)
3545 struct re_pattern_buffer
*bufp
;
3546 const char *string1
, *string2
;
3549 struct re_registers
*regs
;
3552 /* General temporaries. */
3556 /* Just past the end of the corresponding string. */
3557 const char *end1
, *end2
;
3559 /* Pointers into string1 and string2, just past the last characters in
3560 each to consider matching. */
3561 const char *end_match_1
, *end_match_2
;
3563 /* Where we are in the data, and the end of the current string. */
3564 const char *d
, *dend
;
3566 /* Where we are in the pattern, and the end of the pattern. */
3567 unsigned char *p
= bufp
->buffer
;
3568 register unsigned char *pend
= p
+ bufp
->used
;
3570 /* Mark the opcode just after a start_memory, so we can test for an
3571 empty subpattern when we get to the stop_memory. */
3572 unsigned char *just_past_start_mem
= 0;
3574 /* We use this to map every character in the string. */
3575 RE_TRANSLATE_TYPE translate
= bufp
->translate
;
3577 /* Failure point stack. Each place that can handle a failure further
3578 down the line pushes a failure point on this stack. It consists of
3579 restart, regend, and reg_info for all registers corresponding to
3580 the subexpressions we're currently inside, plus the number of such
3581 registers, and, finally, two char *'s. The first char * is where
3582 to resume scanning the pattern; the second one is where to resume
3583 scanning the strings. If the latter is zero, the failure point is
3584 a ``dummy''; if a failure happens and the failure point is a dummy,
3585 it gets discarded and the next next one is tried. */
3586 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
3587 fail_stack_type fail_stack
;
3590 static unsigned failure_id
= 0;
3591 unsigned nfailure_points_pushed
= 0, nfailure_points_popped
= 0;
3594 /* This holds the pointer to the failure stack, when
3595 it is allocated relocatably. */
3596 fail_stack_elt_t
*failure_stack_ptr
;
3598 /* We fill all the registers internally, independent of what we
3599 return, for use in backreferences. The number here includes
3600 an element for register zero. */
3601 unsigned num_regs
= bufp
->re_nsub
+ 1;
3603 /* The currently active registers. */
3604 unsigned lowest_active_reg
= NO_LOWEST_ACTIVE_REG
;
3605 unsigned highest_active_reg
= NO_HIGHEST_ACTIVE_REG
;
3607 /* Information on the contents of registers. These are pointers into
3608 the input strings; they record just what was matched (on this
3609 attempt) by a subexpression part of the pattern, that is, the
3610 regnum-th regstart pointer points to where in the pattern we began
3611 matching and the regnum-th regend points to right after where we
3612 stopped matching the regnum-th subexpression. (The zeroth register
3613 keeps track of what the whole pattern matches.) */
3614 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
3615 const char **regstart
, **regend
;
3618 /* If a group that's operated upon by a repetition operator fails to
3619 match anything, then the register for its start will need to be
3620 restored because it will have been set to wherever in the string we
3621 are when we last see its open-group operator. Similarly for a
3623 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
3624 const char **old_regstart
, **old_regend
;
3627 /* The is_active field of reg_info helps us keep track of which (possibly
3628 nested) subexpressions we are currently in. The matched_something
3629 field of reg_info[reg_num] helps us tell whether or not we have
3630 matched any of the pattern so far this time through the reg_num-th
3631 subexpression. These two fields get reset each time through any
3632 loop their register is in. */
3633 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
3634 register_info_type
*reg_info
;
3637 /* The following record the register info as found in the above
3638 variables when we find a match better than any we've seen before.
3639 This happens as we backtrack through the failure points, which in
3640 turn happens only if we have not yet matched the entire string. */
3641 unsigned best_regs_set
= false;
3642 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
3643 const char **best_regstart
, **best_regend
;
3646 /* Logically, this is `best_regend[0]'. But we don't want to have to
3647 allocate space for that if we're not allocating space for anything
3648 else (see below). Also, we never need info about register 0 for
3649 any of the other register vectors, and it seems rather a kludge to
3650 treat `best_regend' differently than the rest. So we keep track of
3651 the end of the best match so far in a separate variable. We
3652 initialize this to NULL so that when we backtrack the first time
3653 and need to test it, it's not garbage. */
3654 const char *match_end
= NULL
;
3656 /* This helps SET_REGS_MATCHED avoid doing redundant work. */
3657 int set_regs_matched_done
= 0;
3659 /* Used when we pop values we don't care about. */
3660 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
3661 const char **reg_dummy
;
3662 register_info_type
*reg_info_dummy
;
3666 /* Counts the total number of registers pushed. */
3667 unsigned num_regs_pushed
= 0;
3670 DEBUG_PRINT1 ("\n\nEntering re_match_2.\n");
3674 #ifdef MATCH_MAY_ALLOCATE
3675 /* Do not bother to initialize all the register variables if there are
3676 no groups in the pattern, as it takes a fair amount of time. If
3677 there are groups, we include space for register 0 (the whole
3678 pattern), even though we never use it, since it simplifies the
3679 array indexing. We should fix this. */
3682 regstart
= REGEX_TALLOC (num_regs
, const char *);
3683 regend
= REGEX_TALLOC (num_regs
, const char *);
3684 old_regstart
= REGEX_TALLOC (num_regs
, const char *);
3685 old_regend
= REGEX_TALLOC (num_regs
, const char *);
3686 best_regstart
= REGEX_TALLOC (num_regs
, const char *);
3687 best_regend
= REGEX_TALLOC (num_regs
, const char *);
3688 reg_info
= REGEX_TALLOC (num_regs
, register_info_type
);
3689 reg_dummy
= REGEX_TALLOC (num_regs
, const char *);
3690 reg_info_dummy
= REGEX_TALLOC (num_regs
, register_info_type
);
3692 if (!(regstart
&& regend
&& old_regstart
&& old_regend
&& reg_info
3693 && best_regstart
&& best_regend
&& reg_dummy
&& reg_info_dummy
))
3701 /* We must initialize all our variables to NULL, so that
3702 `FREE_VARIABLES' doesn't try to free them. */
3703 regstart
= regend
= old_regstart
= old_regend
= best_regstart
3704 = best_regend
= reg_dummy
= NULL
;
3705 reg_info
= reg_info_dummy
= (register_info_type
*) NULL
;
3707 #endif /* MATCH_MAY_ALLOCATE */
3709 /* The starting position is bogus. */
3710 if (pos
< 0 || pos
> size1
+ size2
)
3716 /* Initialize subexpression text positions to -1 to mark ones that no
3717 start_memory/stop_memory has been seen for. Also initialize the
3718 register information struct. */
3719 for (mcnt
= 1; mcnt
< num_regs
; mcnt
++)
3721 regstart
[mcnt
] = regend
[mcnt
]
3722 = old_regstart
[mcnt
] = old_regend
[mcnt
] = REG_UNSET_VALUE
;
3724 REG_MATCH_NULL_STRING_P (reg_info
[mcnt
]) = MATCH_NULL_UNSET_VALUE
;
3725 IS_ACTIVE (reg_info
[mcnt
]) = 0;
3726 MATCHED_SOMETHING (reg_info
[mcnt
]) = 0;
3727 EVER_MATCHED_SOMETHING (reg_info
[mcnt
]) = 0;
3730 /* We move `string1' into `string2' if the latter's empty -- but not if
3731 `string1' is null. */
3732 if (size2
== 0 && string1
!= NULL
)
3739 end1
= string1
+ size1
;
3740 end2
= string2
+ size2
;
3742 /* Compute where to stop matching, within the two strings. */
3745 end_match_1
= string1
+ stop
;
3746 end_match_2
= string2
;
3751 end_match_2
= string2
+ stop
- size1
;
3754 /* `p' scans through the pattern as `d' scans through the data.
3755 `dend' is the end of the input string that `d' points within. `d'
3756 is advanced into the following input string whenever necessary, but
3757 this happens before fetching; therefore, at the beginning of the
3758 loop, `d' can be pointing at the end of a string, but it cannot
3760 if (size1
> 0 && pos
<= size1
)
3767 d
= string2
+ pos
- size1
;
3771 DEBUG_PRINT1 ("The compiled pattern is: ");
3772 DEBUG_PRINT_COMPILED_PATTERN (bufp
, p
, pend
);
3773 DEBUG_PRINT1 ("The string to match is: `");
3774 DEBUG_PRINT_DOUBLE_STRING (d
, string1
, size1
, string2
, size2
);
3775 DEBUG_PRINT1 ("'\n");
3777 /* This loops over pattern commands. It exits by returning from the
3778 function if the match is complete, or it drops through if the match
3779 fails at this starting point in the input data. */
3782 DEBUG_PRINT2 ("\n0x%x: ", p
);
3785 { /* End of pattern means we might have succeeded. */
3786 DEBUG_PRINT1 ("end of pattern ... ");
3788 /* If we haven't matched the entire string, and we want the
3789 longest match, try backtracking. */
3790 if (d
!= end_match_2
)
3792 /* 1 if this match ends in the same string (string1 or string2)
3793 as the best previous match. */
3794 boolean same_str_p
= (FIRST_STRING_P (match_end
)
3795 == MATCHING_IN_FIRST_STRING
);
3796 /* 1 if this match is the best seen so far. */
3797 boolean best_match_p
;
3799 /* AIX compiler got confused when this was combined
3800 with the previous declaration. */
3802 best_match_p
= d
> match_end
;
3804 best_match_p
= !MATCHING_IN_FIRST_STRING
;
3806 DEBUG_PRINT1 ("backtracking.\n");
3808 if (!FAIL_STACK_EMPTY ())
3809 { /* More failure points to try. */
3811 /* If exceeds best match so far, save it. */
3812 if (!best_regs_set
|| best_match_p
)
3814 best_regs_set
= true;
3817 DEBUG_PRINT1 ("\nSAVING match as best so far.\n");
3819 for (mcnt
= 1; mcnt
< num_regs
; mcnt
++)
3821 best_regstart
[mcnt
] = regstart
[mcnt
];
3822 best_regend
[mcnt
] = regend
[mcnt
];
3828 /* If no failure points, don't restore garbage. And if
3829 last match is real best match, don't restore second
3831 else if (best_regs_set
&& !best_match_p
)
3834 /* Restore best match. It may happen that `dend ==
3835 end_match_1' while the restored d is in string2.
3836 For example, the pattern `x.*y.*z' against the
3837 strings `x-' and `y-z-', if the two strings are
3838 not consecutive in memory. */
3839 DEBUG_PRINT1 ("Restoring best registers.\n");
3842 dend
= ((d
>= string1
&& d
<= end1
)
3843 ? end_match_1
: end_match_2
);
3845 for (mcnt
= 1; mcnt
< num_regs
; mcnt
++)
3847 regstart
[mcnt
] = best_regstart
[mcnt
];
3848 regend
[mcnt
] = best_regend
[mcnt
];
3851 } /* d != end_match_2 */
3854 DEBUG_PRINT1 ("Accepting match.\n");
3856 /* If caller wants register contents data back, do it. */
3857 if (regs
&& !bufp
->no_sub
)
3859 /* Have the register data arrays been allocated? */
3860 if (bufp
->regs_allocated
== REGS_UNALLOCATED
)
3861 { /* No. So allocate them with malloc. We need one
3862 extra element beyond `num_regs' for the `-1' marker
3864 regs
->num_regs
= MAX (RE_NREGS
, num_regs
+ 1);
3865 regs
->start
= TALLOC (regs
->num_regs
, regoff_t
);
3866 regs
->end
= TALLOC (regs
->num_regs
, regoff_t
);
3867 if (regs
->start
== NULL
|| regs
->end
== NULL
)
3872 bufp
->regs_allocated
= REGS_REALLOCATE
;
3874 else if (bufp
->regs_allocated
== REGS_REALLOCATE
)
3875 { /* Yes. If we need more elements than were already
3876 allocated, reallocate them. If we need fewer, just
3878 if (regs
->num_regs
< num_regs
+ 1)
3880 regs
->num_regs
= num_regs
+ 1;
3881 RETALLOC (regs
->start
, regs
->num_regs
, regoff_t
);
3882 RETALLOC (regs
->end
, regs
->num_regs
, regoff_t
);
3883 if (regs
->start
== NULL
|| regs
->end
== NULL
)
3892 /* These braces fend off a "empty body in an else-statement"
3893 warning under GCC when assert expands to nothing. */
3894 assert (bufp
->regs_allocated
== REGS_FIXED
);
3897 /* Convert the pointer data in `regstart' and `regend' to
3898 indices. Register zero has to be set differently,
3899 since we haven't kept track of any info for it. */
3900 if (regs
->num_regs
> 0)
3902 regs
->start
[0] = pos
;
3903 regs
->end
[0] = (MATCHING_IN_FIRST_STRING
3904 ? ((regoff_t
) (d
- string1
))
3905 : ((regoff_t
) (d
- string2
+ size1
)));
3908 /* Go through the first `min (num_regs, regs->num_regs)'
3909 registers, since that is all we initialized. */
3910 for (mcnt
= 1; mcnt
< MIN (num_regs
, regs
->num_regs
); mcnt
++)
3912 if (REG_UNSET (regstart
[mcnt
]) || REG_UNSET (regend
[mcnt
]))
3913 regs
->start
[mcnt
] = regs
->end
[mcnt
] = -1;
3917 = (regoff_t
) POINTER_TO_OFFSET (regstart
[mcnt
]);
3919 = (regoff_t
) POINTER_TO_OFFSET (regend
[mcnt
]);
3923 /* If the regs structure we return has more elements than
3924 were in the pattern, set the extra elements to -1. If
3925 we (re)allocated the registers, this is the case,
3926 because we always allocate enough to have at least one
3928 for (mcnt
= num_regs
; mcnt
< regs
->num_regs
; mcnt
++)
3929 regs
->start
[mcnt
] = regs
->end
[mcnt
] = -1;
3930 } /* regs && !bufp->no_sub */
3932 DEBUG_PRINT4 ("%u failure points pushed, %u popped (%u remain).\n",
3933 nfailure_points_pushed
, nfailure_points_popped
,
3934 nfailure_points_pushed
- nfailure_points_popped
);
3935 DEBUG_PRINT2 ("%u registers pushed.\n", num_regs_pushed
);
3937 mcnt
= d
- pos
- (MATCHING_IN_FIRST_STRING
3941 DEBUG_PRINT2 ("Returning %d from re_match_2.\n", mcnt
);
3947 /* Otherwise match next pattern command. */
3948 switch (SWITCH_ENUM_CAST ((re_opcode_t
) *p
++))
3950 /* Ignore these. Used to ignore the n of succeed_n's which
3951 currently have n == 0. */
3953 DEBUG_PRINT1 ("EXECUTING no_op.\n");
3957 DEBUG_PRINT1 ("EXECUTING succeed.\n");
3960 /* Match the next n pattern characters exactly. The following
3961 byte in the pattern defines n, and the n bytes after that
3962 are the characters to match. */
3965 DEBUG_PRINT2 ("EXECUTING exactn %d.\n", mcnt
);
3967 /* This is written out as an if-else so we don't waste time
3968 testing `translate' inside the loop. */
3974 if ((unsigned char) translate
[(unsigned char) *d
++]
3975 != (unsigned char) *p
++)
3985 if (*d
++ != (char) *p
++) goto fail
;
3989 SET_REGS_MATCHED ();
3993 /* Match any character except possibly a newline or a null. */
3995 DEBUG_PRINT1 ("EXECUTING anychar.\n");
3999 if ((!(bufp
->syntax
& RE_DOT_NEWLINE
) && TRANSLATE (*d
) == '\n')
4000 || (bufp
->syntax
& RE_DOT_NOT_NULL
&& TRANSLATE (*d
) == '\000'))
4003 SET_REGS_MATCHED ();
4004 DEBUG_PRINT2 (" Matched `%d'.\n", *d
);
4012 register unsigned char c
;
4013 boolean
not = (re_opcode_t
) *(p
- 1) == charset_not
;
4015 DEBUG_PRINT2 ("EXECUTING charset%s.\n", not ? "_not" : "");
4018 c
= TRANSLATE (*d
); /* The character to match. */
4020 /* Cast to `unsigned' instead of `unsigned char' in case the
4021 bit list is a full 32 bytes long. */
4022 if (c
< (unsigned) (*p
* BYTEWIDTH
)
4023 && p
[1 + c
/ BYTEWIDTH
] & (1 << (c
% BYTEWIDTH
)))
4028 if (!not) goto fail
;
4030 SET_REGS_MATCHED ();
4036 /* The beginning of a group is represented by start_memory.
4037 The arguments are the register number in the next byte, and the
4038 number of groups inner to this one in the next. The text
4039 matched within the group is recorded (in the internal
4040 registers data structure) under the register number. */
4042 DEBUG_PRINT3 ("EXECUTING start_memory %d (%d):\n", *p
, p
[1]);
4044 /* Find out if this group can match the empty string. */
4045 p1
= p
; /* To send to group_match_null_string_p. */
4047 if (REG_MATCH_NULL_STRING_P (reg_info
[*p
]) == MATCH_NULL_UNSET_VALUE
)
4048 REG_MATCH_NULL_STRING_P (reg_info
[*p
])
4049 = group_match_null_string_p (&p1
, pend
, reg_info
);
4051 /* Save the position in the string where we were the last time
4052 we were at this open-group operator in case the group is
4053 operated upon by a repetition operator, e.g., with `(a*)*b'
4054 against `ab'; then we want to ignore where we are now in
4055 the string in case this attempt to match fails. */
4056 old_regstart
[*p
] = REG_MATCH_NULL_STRING_P (reg_info
[*p
])
4057 ? REG_UNSET (regstart
[*p
]) ? d
: regstart
[*p
]
4059 DEBUG_PRINT2 (" old_regstart: %d\n",
4060 POINTER_TO_OFFSET (old_regstart
[*p
]));
4063 DEBUG_PRINT2 (" regstart: %d\n", POINTER_TO_OFFSET (regstart
[*p
]));
4065 IS_ACTIVE (reg_info
[*p
]) = 1;
4066 MATCHED_SOMETHING (reg_info
[*p
]) = 0;
4068 /* Clear this whenever we change the register activity status. */
4069 set_regs_matched_done
= 0;
4071 /* This is the new highest active register. */
4072 highest_active_reg
= *p
;
4074 /* If nothing was active before, this is the new lowest active
4076 if (lowest_active_reg
== NO_LOWEST_ACTIVE_REG
)
4077 lowest_active_reg
= *p
;
4079 /* Move past the register number and inner group count. */
4081 just_past_start_mem
= p
;
4086 /* The stop_memory opcode represents the end of a group. Its
4087 arguments are the same as start_memory's: the register
4088 number, and the number of inner groups. */
4090 DEBUG_PRINT3 ("EXECUTING stop_memory %d (%d):\n", *p
, p
[1]);
4092 /* We need to save the string position the last time we were at
4093 this close-group operator in case the group is operated
4094 upon by a repetition operator, e.g., with `((a*)*(b*)*)*'
4095 against `aba'; then we want to ignore where we are now in
4096 the string in case this attempt to match fails. */
4097 old_regend
[*p
] = REG_MATCH_NULL_STRING_P (reg_info
[*p
])
4098 ? REG_UNSET (regend
[*p
]) ? d
: regend
[*p
]
4100 DEBUG_PRINT2 (" old_regend: %d\n",
4101 POINTER_TO_OFFSET (old_regend
[*p
]));
4104 DEBUG_PRINT2 (" regend: %d\n", POINTER_TO_OFFSET (regend
[*p
]));
4106 /* This register isn't active anymore. */
4107 IS_ACTIVE (reg_info
[*p
]) = 0;
4109 /* Clear this whenever we change the register activity status. */
4110 set_regs_matched_done
= 0;
4112 /* If this was the only register active, nothing is active
4114 if (lowest_active_reg
== highest_active_reg
)
4116 lowest_active_reg
= NO_LOWEST_ACTIVE_REG
;
4117 highest_active_reg
= NO_HIGHEST_ACTIVE_REG
;
4120 { /* We must scan for the new highest active register, since
4121 it isn't necessarily one less than now: consider
4122 (a(b)c(d(e)f)g). When group 3 ends, after the f), the
4123 new highest active register is 1. */
4124 unsigned char r
= *p
- 1;
4125 while (r
> 0 && !IS_ACTIVE (reg_info
[r
]))
4128 /* If we end up at register zero, that means that we saved
4129 the registers as the result of an `on_failure_jump', not
4130 a `start_memory', and we jumped to past the innermost
4131 `stop_memory'. For example, in ((.)*) we save
4132 registers 1 and 2 as a result of the *, but when we pop
4133 back to the second ), we are at the stop_memory 1.
4134 Thus, nothing is active. */
4137 lowest_active_reg
= NO_LOWEST_ACTIVE_REG
;
4138 highest_active_reg
= NO_HIGHEST_ACTIVE_REG
;
4141 highest_active_reg
= r
;
4144 /* If just failed to match something this time around with a
4145 group that's operated on by a repetition operator, try to
4146 force exit from the ``loop'', and restore the register
4147 information for this group that we had before trying this
4149 if ((!MATCHED_SOMETHING (reg_info
[*p
])
4150 || just_past_start_mem
== p
- 1)
4153 boolean is_a_jump_n
= false;
4157 switch ((re_opcode_t
) *p1
++)
4161 case pop_failure_jump
:
4162 case maybe_pop_jump
:
4164 case dummy_failure_jump
:
4165 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
4175 /* If the next operation is a jump backwards in the pattern
4176 to an on_failure_jump right before the start_memory
4177 corresponding to this stop_memory, exit from the loop
4178 by forcing a failure after pushing on the stack the
4179 on_failure_jump's jump in the pattern, and d. */
4180 if (mcnt
< 0 && (re_opcode_t
) *p1
== on_failure_jump
4181 && (re_opcode_t
) p1
[3] == start_memory
&& p1
[4] == *p
)
4183 /* If this group ever matched anything, then restore
4184 what its registers were before trying this last
4185 failed match, e.g., with `(a*)*b' against `ab' for
4186 regstart[1], and, e.g., with `((a*)*(b*)*)*'
4187 against `aba' for regend[3].
4189 Also restore the registers for inner groups for,
4190 e.g., `((a*)(b*))*' against `aba' (register 3 would
4191 otherwise get trashed). */
4193 if (EVER_MATCHED_SOMETHING (reg_info
[*p
]))
4197 EVER_MATCHED_SOMETHING (reg_info
[*p
]) = 0;
4199 /* Restore this and inner groups' (if any) registers. */
4200 for (r
= *p
; r
< *p
+ *(p
+ 1); r
++)
4202 regstart
[r
] = old_regstart
[r
];
4204 /* xx why this test? */
4205 if (old_regend
[r
] >= regstart
[r
])
4206 regend
[r
] = old_regend
[r
];
4210 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
4211 PUSH_FAILURE_POINT (p1
+ mcnt
, d
, -2);
4217 /* Move past the register number and the inner group count. */
4222 /* \<digit> has been turned into a `duplicate' command which is
4223 followed by the numeric value of <digit> as the register number. */
4226 register const char *d2
, *dend2
;
4227 int regno
= *p
++; /* Get which register to match against. */
4228 DEBUG_PRINT2 ("EXECUTING duplicate %d.\n", regno
);
4230 /* Can't back reference a group which we've never matched. */
4231 if (REG_UNSET (regstart
[regno
]) || REG_UNSET (regend
[regno
]))
4234 /* Where in input to try to start matching. */
4235 d2
= regstart
[regno
];
4237 /* Where to stop matching; if both the place to start and
4238 the place to stop matching are in the same string, then
4239 set to the place to stop, otherwise, for now have to use
4240 the end of the first string. */
4242 dend2
= ((FIRST_STRING_P (regstart
[regno
])
4243 == FIRST_STRING_P (regend
[regno
]))
4244 ? regend
[regno
] : end_match_1
);
4247 /* If necessary, advance to next segment in register
4251 if (dend2
== end_match_2
) break;
4252 if (dend2
== regend
[regno
]) break;
4254 /* End of string1 => advance to string2. */
4256 dend2
= regend
[regno
];
4258 /* At end of register contents => success */
4259 if (d2
== dend2
) break;
4261 /* If necessary, advance to next segment in data. */
4264 /* How many characters left in this segment to match. */
4267 /* Want how many consecutive characters we can match in
4268 one shot, so, if necessary, adjust the count. */
4269 if (mcnt
> dend2
- d2
)
4272 /* Compare that many; failure if mismatch, else move
4275 ? bcmp_translate (d
, d2
, mcnt
, translate
)
4276 : bcmp (d
, d2
, mcnt
))
4278 d
+= mcnt
, d2
+= mcnt
;
4280 /* Do this because we've match some characters. */
4281 SET_REGS_MATCHED ();
4287 /* begline matches the empty string at the beginning of the string
4288 (unless `not_bol' is set in `bufp'), and, if
4289 `newline_anchor' is set, after newlines. */
4291 DEBUG_PRINT1 ("EXECUTING begline.\n");
4293 if (AT_STRINGS_BEG (d
))
4295 if (!bufp
->not_bol
) break;
4297 else if (d
[-1] == '\n' && bufp
->newline_anchor
)
4301 /* In all other cases, we fail. */
4305 /* endline is the dual of begline. */
4307 DEBUG_PRINT1 ("EXECUTING endline.\n");
4309 if (AT_STRINGS_END (d
))
4311 if (!bufp
->not_eol
) break;
4314 /* We have to ``prefetch'' the next character. */
4315 else if ((d
== end1
? *string2
: *d
) == '\n'
4316 && bufp
->newline_anchor
)
4323 /* Match at the very beginning of the data. */
4325 DEBUG_PRINT1 ("EXECUTING begbuf.\n");
4326 if (AT_STRINGS_BEG (d
))
4331 /* Match at the very end of the data. */
4333 DEBUG_PRINT1 ("EXECUTING endbuf.\n");
4334 if (AT_STRINGS_END (d
))
4339 /* on_failure_keep_string_jump is used to optimize `.*\n'. It
4340 pushes NULL as the value for the string on the stack. Then
4341 `pop_failure_point' will keep the current value for the
4342 string, instead of restoring it. To see why, consider
4343 matching `foo\nbar' against `.*\n'. The .* matches the foo;
4344 then the . fails against the \n. But the next thing we want
4345 to do is match the \n against the \n; if we restored the
4346 string value, we would be back at the foo.
4348 Because this is used only in specific cases, we don't need to
4349 check all the things that `on_failure_jump' does, to make
4350 sure the right things get saved on the stack. Hence we don't
4351 share its code. The only reason to push anything on the
4352 stack at all is that otherwise we would have to change
4353 `anychar's code to do something besides goto fail in this
4354 case; that seems worse than this. */
4355 case on_failure_keep_string_jump
:
4356 DEBUG_PRINT1 ("EXECUTING on_failure_keep_string_jump");
4358 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
4359 DEBUG_PRINT3 (" %d (to 0x%x):\n", mcnt
, p
+ mcnt
);
4361 PUSH_FAILURE_POINT (p
+ mcnt
, NULL
, -2);
4365 /* Uses of on_failure_jump:
4367 Each alternative starts with an on_failure_jump that points
4368 to the beginning of the next alternative. Each alternative
4369 except the last ends with a jump that in effect jumps past
4370 the rest of the alternatives. (They really jump to the
4371 ending jump of the following alternative, because tensioning
4372 these jumps is a hassle.)
4374 Repeats start with an on_failure_jump that points past both
4375 the repetition text and either the following jump or
4376 pop_failure_jump back to this on_failure_jump. */
4377 case on_failure_jump
:
4379 DEBUG_PRINT1 ("EXECUTING on_failure_jump");
4381 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
4382 DEBUG_PRINT3 (" %d (to 0x%x)", mcnt
, p
+ mcnt
);
4384 /* If this on_failure_jump comes right before a group (i.e.,
4385 the original * applied to a group), save the information
4386 for that group and all inner ones, so that if we fail back
4387 to this point, the group's information will be correct.
4388 For example, in \(a*\)*\1, we need the preceding group,
4389 and in \(zz\(a*\)b*\)\2, we need the inner group. */
4391 /* We can't use `p' to check ahead because we push
4392 a failure point to `p + mcnt' after we do this. */
4395 /* We need to skip no_op's before we look for the
4396 start_memory in case this on_failure_jump is happening as
4397 the result of a completed succeed_n, as in \(a\)\{1,3\}b\1
4399 while (p1
< pend
&& (re_opcode_t
) *p1
== no_op
)
4402 if (p1
< pend
&& (re_opcode_t
) *p1
== start_memory
)
4404 /* We have a new highest active register now. This will
4405 get reset at the start_memory we are about to get to,
4406 but we will have saved all the registers relevant to
4407 this repetition op, as described above. */
4408 highest_active_reg
= *(p1
+ 1) + *(p1
+ 2);
4409 if (lowest_active_reg
== NO_LOWEST_ACTIVE_REG
)
4410 lowest_active_reg
= *(p1
+ 1);
4413 DEBUG_PRINT1 (":\n");
4414 PUSH_FAILURE_POINT (p
+ mcnt
, d
, -2);
4418 /* A smart repeat ends with `maybe_pop_jump'.
4419 We change it to either `pop_failure_jump' or `jump'. */
4420 case maybe_pop_jump
:
4421 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
4422 DEBUG_PRINT2 ("EXECUTING maybe_pop_jump %d.\n", mcnt
);
4424 register unsigned char *p2
= p
;
4426 /* Compare the beginning of the repeat with what in the
4427 pattern follows its end. If we can establish that there
4428 is nothing that they would both match, i.e., that we
4429 would have to backtrack because of (as in, e.g., `a*a')
4430 then we can change to pop_failure_jump, because we'll
4431 never have to backtrack.
4433 This is not true in the case of alternatives: in
4434 `(a|ab)*' we do need to backtrack to the `ab' alternative
4435 (e.g., if the string was `ab'). But instead of trying to
4436 detect that here, the alternative has put on a dummy
4437 failure point which is what we will end up popping. */
4439 /* Skip over open/close-group commands.
4440 If what follows this loop is a ...+ construct,
4441 look at what begins its body, since we will have to
4442 match at least one of that. */
4446 && ((re_opcode_t
) *p2
== stop_memory
4447 || (re_opcode_t
) *p2
== start_memory
))
4449 else if (p2
+ 6 < pend
4450 && (re_opcode_t
) *p2
== dummy_failure_jump
)
4457 /* p1[0] ... p1[2] are the `on_failure_jump' corresponding
4458 to the `maybe_finalize_jump' of this case. Examine what
4461 /* If we're at the end of the pattern, we can change. */
4464 /* Consider what happens when matching ":\(.*\)"
4465 against ":/". I don't really understand this code
4467 p
[-3] = (unsigned char) pop_failure_jump
;
4469 (" End of pattern: change to `pop_failure_jump'.\n");
4472 else if ((re_opcode_t
) *p2
== exactn
4473 || (bufp
->newline_anchor
&& (re_opcode_t
) *p2
== endline
))
4475 register unsigned char c
4476 = *p2
== (unsigned char) endline
? '\n' : p2
[2];
4478 if ((re_opcode_t
) p1
[3] == exactn
&& p1
[5] != c
)
4480 p
[-3] = (unsigned char) pop_failure_jump
;
4481 DEBUG_PRINT3 (" %c != %c => pop_failure_jump.\n",
4485 else if ((re_opcode_t
) p1
[3] == charset
4486 || (re_opcode_t
) p1
[3] == charset_not
)
4488 int not = (re_opcode_t
) p1
[3] == charset_not
;
4490 if (c
< (unsigned char) (p1
[4] * BYTEWIDTH
)
4491 && p1
[5 + c
/ BYTEWIDTH
] & (1 << (c
% BYTEWIDTH
)))
4494 /* `not' is equal to 1 if c would match, which means
4495 that we can't change to pop_failure_jump. */
4498 p
[-3] = (unsigned char) pop_failure_jump
;
4499 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
4503 else if ((re_opcode_t
) *p2
== charset
)
4506 register unsigned char c
4507 = *p2
== (unsigned char) endline
? '\n' : p2
[2];
4510 if ((re_opcode_t
) p1
[3] == exactn
4511 && ! ((int) p2
[1] * BYTEWIDTH
> (int) p1
[5]
4512 && (p2
[2 + p1
[5] / BYTEWIDTH
]
4513 & (1 << (p1
[5] % BYTEWIDTH
)))))
4515 p
[-3] = (unsigned char) pop_failure_jump
;
4516 DEBUG_PRINT3 (" %c != %c => pop_failure_jump.\n",
4520 else if ((re_opcode_t
) p1
[3] == charset_not
)
4523 /* We win if the charset_not inside the loop
4524 lists every character listed in the charset after. */
4525 for (idx
= 0; idx
< (int) p2
[1]; idx
++)
4526 if (! (p2
[2 + idx
] == 0
4527 || (idx
< (int) p1
[4]
4528 && ((p2
[2 + idx
] & ~ p1
[5 + idx
]) == 0))))
4533 p
[-3] = (unsigned char) pop_failure_jump
;
4534 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
4537 else if ((re_opcode_t
) p1
[3] == charset
)
4540 /* We win if the charset inside the loop
4541 has no overlap with the one after the loop. */
4543 idx
< (int) p2
[1] && idx
< (int) p1
[4];
4545 if ((p2
[2 + idx
] & p1
[5 + idx
]) != 0)
4548 if (idx
== p2
[1] || idx
== p1
[4])
4550 p
[-3] = (unsigned char) pop_failure_jump
;
4551 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
4556 p
-= 2; /* Point at relative address again. */
4557 if ((re_opcode_t
) p
[-1] != pop_failure_jump
)
4559 p
[-1] = (unsigned char) jump
;
4560 DEBUG_PRINT1 (" Match => jump.\n");
4561 goto unconditional_jump
;
4563 /* Note fall through. */
4566 /* The end of a simple repeat has a pop_failure_jump back to
4567 its matching on_failure_jump, where the latter will push a
4568 failure point. The pop_failure_jump takes off failure
4569 points put on by this pop_failure_jump's matching
4570 on_failure_jump; we got through the pattern to here from the
4571 matching on_failure_jump, so didn't fail. */
4572 case pop_failure_jump
:
4574 /* We need to pass separate storage for the lowest and
4575 highest registers, even though we don't care about the
4576 actual values. Otherwise, we will restore only one
4577 register from the stack, since lowest will == highest in
4578 `pop_failure_point'. */
4579 unsigned dummy_low_reg
, dummy_high_reg
;
4580 unsigned char *pdummy
;
4583 DEBUG_PRINT1 ("EXECUTING pop_failure_jump.\n");
4584 POP_FAILURE_POINT (sdummy
, pdummy
,
4585 dummy_low_reg
, dummy_high_reg
,
4586 reg_dummy
, reg_dummy
, reg_info_dummy
);
4588 /* Note fall through. */
4591 /* Unconditionally jump (without popping any failure points). */
4594 EXTRACT_NUMBER_AND_INCR (mcnt
, p
); /* Get the amount to jump. */
4595 DEBUG_PRINT2 ("EXECUTING jump %d ", mcnt
);
4596 p
+= mcnt
; /* Do the jump. */
4597 DEBUG_PRINT2 ("(to 0x%x).\n", p
);
4601 /* We need this opcode so we can detect where alternatives end
4602 in `group_match_null_string_p' et al. */
4604 DEBUG_PRINT1 ("EXECUTING jump_past_alt.\n");
4605 goto unconditional_jump
;
4608 /* Normally, the on_failure_jump pushes a failure point, which
4609 then gets popped at pop_failure_jump. We will end up at
4610 pop_failure_jump, also, and with a pattern of, say, `a+', we
4611 are skipping over the on_failure_jump, so we have to push
4612 something meaningless for pop_failure_jump to pop. */
4613 case dummy_failure_jump
:
4614 DEBUG_PRINT1 ("EXECUTING dummy_failure_jump.\n");
4615 /* It doesn't matter what we push for the string here. What
4616 the code at `fail' tests is the value for the pattern. */
4617 PUSH_FAILURE_POINT (0, 0, -2);
4618 goto unconditional_jump
;
4621 /* At the end of an alternative, we need to push a dummy failure
4622 point in case we are followed by a `pop_failure_jump', because
4623 we don't want the failure point for the alternative to be
4624 popped. For example, matching `(a|ab)*' against `aab'
4625 requires that we match the `ab' alternative. */
4626 case push_dummy_failure
:
4627 DEBUG_PRINT1 ("EXECUTING push_dummy_failure.\n");
4628 /* See comments just above at `dummy_failure_jump' about the
4630 PUSH_FAILURE_POINT (0, 0, -2);
4633 /* Have to succeed matching what follows at least n times.
4634 After that, handle like `on_failure_jump'. */
4636 EXTRACT_NUMBER (mcnt
, p
+ 2);
4637 DEBUG_PRINT2 ("EXECUTING succeed_n %d.\n", mcnt
);
4640 /* Originally, this is how many times we HAVE to succeed. */
4645 STORE_NUMBER_AND_INCR (p
, mcnt
);
4646 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p
, mcnt
);
4650 DEBUG_PRINT2 (" Setting two bytes from 0x%x to no_op.\n", p
+2);
4651 p
[2] = (unsigned char) no_op
;
4652 p
[3] = (unsigned char) no_op
;
4658 EXTRACT_NUMBER (mcnt
, p
+ 2);
4659 DEBUG_PRINT2 ("EXECUTING jump_n %d.\n", mcnt
);
4661 /* Originally, this is how many times we CAN jump. */
4665 STORE_NUMBER (p
+ 2, mcnt
);
4666 goto unconditional_jump
;
4668 /* If don't have to jump any more, skip over the rest of command. */
4675 DEBUG_PRINT1 ("EXECUTING set_number_at.\n");
4677 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
4679 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
4680 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p1
, mcnt
);
4681 STORE_NUMBER (p1
, mcnt
);
4686 /* The DEC Alpha C compiler 3.x generates incorrect code for the
4687 test WORDCHAR_P (d - 1) != WORDCHAR_P (d) in the expansion of
4688 AT_WORD_BOUNDARY, so this code is disabled. Expanding the
4689 macro and introducing temporary variables works around the bug. */
4692 DEBUG_PRINT1 ("EXECUTING wordbound.\n");
4693 if (AT_WORD_BOUNDARY (d
))
4698 DEBUG_PRINT1 ("EXECUTING notwordbound.\n");
4699 if (AT_WORD_BOUNDARY (d
))
4705 boolean prevchar
, thischar
;
4707 DEBUG_PRINT1 ("EXECUTING wordbound.\n");
4708 if (AT_STRINGS_BEG (d
) || AT_STRINGS_END (d
))
4711 prevchar
= WORDCHAR_P (d
- 1);
4712 thischar
= WORDCHAR_P (d
);
4713 if (prevchar
!= thischar
)
4720 boolean prevchar
, thischar
;
4722 DEBUG_PRINT1 ("EXECUTING notwordbound.\n");
4723 if (AT_STRINGS_BEG (d
) || AT_STRINGS_END (d
))
4726 prevchar
= WORDCHAR_P (d
- 1);
4727 thischar
= WORDCHAR_P (d
);
4728 if (prevchar
!= thischar
)
4735 DEBUG_PRINT1 ("EXECUTING wordbeg.\n");
4736 if (WORDCHAR_P (d
) && (AT_STRINGS_BEG (d
) || !WORDCHAR_P (d
- 1)))
4741 DEBUG_PRINT1 ("EXECUTING wordend.\n");
4742 if (!AT_STRINGS_BEG (d
) && WORDCHAR_P (d
- 1)
4743 && (!WORDCHAR_P (d
) || AT_STRINGS_END (d
)))
4749 DEBUG_PRINT1 ("EXECUTING before_dot.\n");
4750 if (PTR_CHAR_POS ((unsigned char *) d
) >= point
)
4755 DEBUG_PRINT1 ("EXECUTING at_dot.\n");
4756 if (PTR_CHAR_POS ((unsigned char *) d
) != point
)
4761 DEBUG_PRINT1 ("EXECUTING after_dot.\n");
4762 if (PTR_CHAR_POS ((unsigned char *) d
) <= point
)
4767 DEBUG_PRINT2 ("EXECUTING syntaxspec %d.\n", mcnt
);
4772 DEBUG_PRINT1 ("EXECUTING Emacs wordchar.\n");
4776 /* Can't use *d++ here; SYNTAX may be an unsafe macro. */
4778 if (SYNTAX (d
[-1]) != (enum syntaxcode
) mcnt
)
4780 SET_REGS_MATCHED ();
4784 DEBUG_PRINT2 ("EXECUTING notsyntaxspec %d.\n", mcnt
);
4786 goto matchnotsyntax
;
4789 DEBUG_PRINT1 ("EXECUTING Emacs notwordchar.\n");
4793 /* Can't use *d++ here; SYNTAX may be an unsafe macro. */
4795 if (SYNTAX (d
[-1]) == (enum syntaxcode
) mcnt
)
4797 SET_REGS_MATCHED ();
4800 #else /* not emacs */
4802 DEBUG_PRINT1 ("EXECUTING non-Emacs wordchar.\n");
4804 if (!WORDCHAR_P (d
))
4806 SET_REGS_MATCHED ();
4811 DEBUG_PRINT1 ("EXECUTING non-Emacs notwordchar.\n");
4815 SET_REGS_MATCHED ();
4818 #endif /* not emacs */
4823 continue; /* Successfully executed one pattern command; keep going. */
4826 /* We goto here if a matching operation fails. */
4828 if (!FAIL_STACK_EMPTY ())
4829 { /* A restart point is known. Restore to that state. */
4830 DEBUG_PRINT1 ("\nFAIL:\n");
4831 POP_FAILURE_POINT (d
, p
,
4832 lowest_active_reg
, highest_active_reg
,
4833 regstart
, regend
, reg_info
);
4835 /* If this failure point is a dummy, try the next one. */
4839 /* If we failed to the end of the pattern, don't examine *p. */
4843 boolean is_a_jump_n
= false;
4845 /* If failed to a backwards jump that's part of a repetition
4846 loop, need to pop this failure point and use the next one. */
4847 switch ((re_opcode_t
) *p
)
4851 case maybe_pop_jump
:
4852 case pop_failure_jump
:
4855 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
4858 if ((is_a_jump_n
&& (re_opcode_t
) *p1
== succeed_n
)
4860 && (re_opcode_t
) *p1
== on_failure_jump
))
4868 if (d
>= string1
&& d
<= end1
)
4872 break; /* Matching at this starting point really fails. */
4876 goto restore_best_regs
;
4880 return -1; /* Failure to match. */
4883 /* Subroutine definitions for re_match_2. */
4886 /* We are passed P pointing to a register number after a start_memory.
4888 Return true if the pattern up to the corresponding stop_memory can
4889 match the empty string, and false otherwise.
4891 If we find the matching stop_memory, sets P to point to one past its number.
4892 Otherwise, sets P to an undefined byte less than or equal to END.
4894 We don't handle duplicates properly (yet). */
4897 group_match_null_string_p (p
, end
, reg_info
)
4898 unsigned char **p
, *end
;
4899 register_info_type
*reg_info
;
4902 /* Point to after the args to the start_memory. */
4903 unsigned char *p1
= *p
+ 2;
4907 /* Skip over opcodes that can match nothing, and return true or
4908 false, as appropriate, when we get to one that can't, or to the
4909 matching stop_memory. */
4911 switch ((re_opcode_t
) *p1
)
4913 /* Could be either a loop or a series of alternatives. */
4914 case on_failure_jump
:
4916 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
4918 /* If the next operation is not a jump backwards in the
4923 /* Go through the on_failure_jumps of the alternatives,
4924 seeing if any of the alternatives cannot match nothing.
4925 The last alternative starts with only a jump,
4926 whereas the rest start with on_failure_jump and end
4927 with a jump, e.g., here is the pattern for `a|b|c':
4929 /on_failure_jump/0/6/exactn/1/a/jump_past_alt/0/6
4930 /on_failure_jump/0/6/exactn/1/b/jump_past_alt/0/3
4933 So, we have to first go through the first (n-1)
4934 alternatives and then deal with the last one separately. */
4937 /* Deal with the first (n-1) alternatives, which start
4938 with an on_failure_jump (see above) that jumps to right
4939 past a jump_past_alt. */
4941 while ((re_opcode_t
) p1
[mcnt
-3] == jump_past_alt
)
4943 /* `mcnt' holds how many bytes long the alternative
4944 is, including the ending `jump_past_alt' and
4947 if (!alt_match_null_string_p (p1
, p1
+ mcnt
- 3,
4951 /* Move to right after this alternative, including the
4955 /* Break if it's the beginning of an n-th alternative
4956 that doesn't begin with an on_failure_jump. */
4957 if ((re_opcode_t
) *p1
!= on_failure_jump
)
4960 /* Still have to check that it's not an n-th
4961 alternative that starts with an on_failure_jump. */
4963 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
4964 if ((re_opcode_t
) p1
[mcnt
-3] != jump_past_alt
)
4966 /* Get to the beginning of the n-th alternative. */
4972 /* Deal with the last alternative: go back and get number
4973 of the `jump_past_alt' just before it. `mcnt' contains
4974 the length of the alternative. */
4975 EXTRACT_NUMBER (mcnt
, p1
- 2);
4977 if (!alt_match_null_string_p (p1
, p1
+ mcnt
, reg_info
))
4980 p1
+= mcnt
; /* Get past the n-th alternative. */
4986 assert (p1
[1] == **p
);
4992 if (!common_op_match_null_string_p (&p1
, end
, reg_info
))
4995 } /* while p1 < end */
4998 } /* group_match_null_string_p */
5001 /* Similar to group_match_null_string_p, but doesn't deal with alternatives:
5002 It expects P to be the first byte of a single alternative and END one
5003 byte past the last. The alternative can contain groups. */
5006 alt_match_null_string_p (p
, end
, reg_info
)
5007 unsigned char *p
, *end
;
5008 register_info_type
*reg_info
;
5011 unsigned char *p1
= p
;
5015 /* Skip over opcodes that can match nothing, and break when we get
5016 to one that can't. */
5018 switch ((re_opcode_t
) *p1
)
5021 case on_failure_jump
:
5023 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
5028 if (!common_op_match_null_string_p (&p1
, end
, reg_info
))
5031 } /* while p1 < end */
5034 } /* alt_match_null_string_p */
5037 /* Deals with the ops common to group_match_null_string_p and
5038 alt_match_null_string_p.
5040 Sets P to one after the op and its arguments, if any. */
5043 common_op_match_null_string_p (p
, end
, reg_info
)
5044 unsigned char **p
, *end
;
5045 register_info_type
*reg_info
;
5050 unsigned char *p1
= *p
;
5052 switch ((re_opcode_t
) *p1
++)
5072 assert (reg_no
> 0 && reg_no
<= MAX_REGNUM
);
5073 ret
= group_match_null_string_p (&p1
, end
, reg_info
);
5075 /* Have to set this here in case we're checking a group which
5076 contains a group and a back reference to it. */
5078 if (REG_MATCH_NULL_STRING_P (reg_info
[reg_no
]) == MATCH_NULL_UNSET_VALUE
)
5079 REG_MATCH_NULL_STRING_P (reg_info
[reg_no
]) = ret
;
5085 /* If this is an optimized succeed_n for zero times, make the jump. */
5087 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
5095 /* Get to the number of times to succeed. */
5097 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
5102 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
5110 if (!REG_MATCH_NULL_STRING_P (reg_info
[*p1
]))
5118 /* All other opcodes mean we cannot match the empty string. */
5124 } /* common_op_match_null_string_p */
5127 /* Return zero if TRANSLATE[S1] and TRANSLATE[S2] are identical for LEN
5128 bytes; nonzero otherwise. */
5131 bcmp_translate (s1
, s2
, len
, translate
)
5132 unsigned char *s1
, *s2
;
5134 RE_TRANSLATE_TYPE translate
;
5136 register unsigned char *p1
= s1
, *p2
= s2
;
5139 if (translate
[*p1
++] != translate
[*p2
++]) return 1;
5145 /* Entry points for GNU code. */
5147 /* re_compile_pattern is the GNU regular expression compiler: it
5148 compiles PATTERN (of length SIZE) and puts the result in BUFP.
5149 Returns 0 if the pattern was valid, otherwise an error string.
5151 Assumes the `allocated' (and perhaps `buffer') and `translate' fields
5152 are set in BUFP on entry.
5154 We call regex_compile to do the actual compilation. */
5157 re_compile_pattern (pattern
, length
, bufp
)
5158 const char *pattern
;
5160 struct re_pattern_buffer
*bufp
;
5164 /* GNU code is written to assume at least RE_NREGS registers will be set
5165 (and at least one extra will be -1). */
5166 bufp
->regs_allocated
= REGS_UNALLOCATED
;
5168 /* And GNU code determines whether or not to get register information
5169 by passing null for the REGS argument to re_match, etc., not by
5173 /* Match anchors at newline. */
5174 bufp
->newline_anchor
= 1;
5176 ret
= regex_compile (pattern
, length
, re_syntax_options
, bufp
);
5180 return gettext (re_error_msgid
[(int) ret
]);
5183 /* Entry points compatible with 4.2 BSD regex library. We don't define
5184 them unless specifically requested. */
5186 #if defined (_REGEX_RE_COMP) || defined (_LIBC)
5188 /* BSD has one and only one pattern buffer. */
5189 static struct re_pattern_buffer re_comp_buf
;
5199 if (!re_comp_buf
.buffer
)
5200 return gettext ("No previous regular expression");
5204 if (!re_comp_buf
.buffer
)
5206 re_comp_buf
.buffer
= (unsigned char *) malloc (200);
5207 if (re_comp_buf
.buffer
== NULL
)
5208 return gettext (re_error_msgid
[(int) REG_ESPACE
]);
5209 re_comp_buf
.allocated
= 200;
5211 re_comp_buf
.fastmap
= (char *) malloc (1 << BYTEWIDTH
);
5212 if (re_comp_buf
.fastmap
== NULL
)
5213 return gettext (re_error_msgid
[(int) REG_ESPACE
]);
5216 /* Since `re_exec' always passes NULL for the `regs' argument, we
5217 don't need to initialize the pattern buffer fields which affect it. */
5219 /* Match anchors at newlines. */
5220 re_comp_buf
.newline_anchor
= 1;
5222 ret
= regex_compile (s
, strlen (s
), re_syntax_options
, &re_comp_buf
);
5227 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
5228 return (char *) gettext (re_error_msgid
[(int) ret
]);
5236 const int len
= strlen (s
);
5238 0 <= re_search (&re_comp_buf
, s
, len
, 0, len
, (struct re_registers
*) 0);
5242 /* Make these definitions weak in libc, so POSIX programs can redefine
5243 these names if they don't use our functions, and still use
5244 regcomp/regexec below without link errors. */
5245 weak_symbol (re_comp
)
5246 weak_symbol (re_exec
)
5249 #endif /* _REGEX_RE_COMP */
5251 /* POSIX.2 functions. Don't define these for Emacs. */
5255 /* regcomp takes a regular expression as a string and compiles it.
5257 PREG is a regex_t *. We do not expect any fields to be initialized,
5258 since POSIX says we shouldn't. Thus, we set
5260 `buffer' to the compiled pattern;
5261 `used' to the length of the compiled pattern;
5262 `syntax' to RE_SYNTAX_POSIX_EXTENDED if the
5263 REG_EXTENDED bit in CFLAGS is set; otherwise, to
5264 RE_SYNTAX_POSIX_BASIC;
5265 `newline_anchor' to REG_NEWLINE being set in CFLAGS;
5266 `fastmap' and `fastmap_accurate' to zero;
5267 `re_nsub' to the number of subexpressions in PATTERN.
5269 PATTERN is the address of the pattern string.
5271 CFLAGS is a series of bits which affect compilation.
5273 If REG_EXTENDED is set, we use POSIX extended syntax; otherwise, we
5274 use POSIX basic syntax.
5276 If REG_NEWLINE is set, then . and [^...] don't match newline.
5277 Also, regexec will try a match beginning after every newline.
5279 If REG_ICASE is set, then we considers upper- and lowercase
5280 versions of letters to be equivalent when matching.
5282 If REG_NOSUB is set, then when PREG is passed to regexec, that
5283 routine will report only success or failure, and nothing about the
5286 It returns 0 if it succeeds, nonzero if it doesn't. (See regex.h for
5287 the return codes and their meanings.) */
5290 regcomp (preg
, pattern
, cflags
)
5292 const char *pattern
;
5297 = (cflags
& REG_EXTENDED
) ?
5298 RE_SYNTAX_POSIX_EXTENDED
: RE_SYNTAX_POSIX_BASIC
;
5300 /* regex_compile will allocate the space for the compiled pattern. */
5302 preg
->allocated
= 0;
5305 /* Don't bother to use a fastmap when searching. This simplifies the
5306 REG_NEWLINE case: if we used a fastmap, we'd have to put all the
5307 characters after newlines into the fastmap. This way, we just try
5311 if (cflags
& REG_ICASE
)
5316 = (RE_TRANSLATE_TYPE
) malloc (CHAR_SET_SIZE
5317 * sizeof (*(RE_TRANSLATE_TYPE
)0));
5318 if (preg
->translate
== NULL
)
5319 return (int) REG_ESPACE
;
5321 /* Map uppercase characters to corresponding lowercase ones. */
5322 for (i
= 0; i
< CHAR_SET_SIZE
; i
++)
5323 preg
->translate
[i
] = ISUPPER (i
) ? tolower (i
) : i
;
5326 preg
->translate
= NULL
;
5328 /* If REG_NEWLINE is set, newlines are treated differently. */
5329 if (cflags
& REG_NEWLINE
)
5330 { /* REG_NEWLINE implies neither . nor [^...] match newline. */
5331 syntax
&= ~RE_DOT_NEWLINE
;
5332 syntax
|= RE_HAT_LISTS_NOT_NEWLINE
;
5333 /* It also changes the matching behavior. */
5334 preg
->newline_anchor
= 1;
5337 preg
->newline_anchor
= 0;
5339 preg
->no_sub
= !!(cflags
& REG_NOSUB
);
5341 /* POSIX says a null character in the pattern terminates it, so we
5342 can use strlen here in compiling the pattern. */
5343 ret
= regex_compile (pattern
, strlen (pattern
), syntax
, preg
);
5345 /* POSIX doesn't distinguish between an unmatched open-group and an
5346 unmatched close-group: both are REG_EPAREN. */
5347 if (ret
== REG_ERPAREN
) ret
= REG_EPAREN
;
5353 /* regexec searches for a given pattern, specified by PREG, in the
5356 If NMATCH is zero or REG_NOSUB was set in the cflags argument to
5357 `regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at
5358 least NMATCH elements, and we set them to the offsets of the
5359 corresponding matched substrings.
5361 EFLAGS specifies `execution flags' which affect matching: if
5362 REG_NOTBOL is set, then ^ does not match at the beginning of the
5363 string; if REG_NOTEOL is set, then $ does not match at the end.
5365 We return 0 if we find a match and REG_NOMATCH if not. */
5368 regexec (preg
, string
, nmatch
, pmatch
, eflags
)
5369 const regex_t
*preg
;
5372 regmatch_t pmatch
[];
5376 struct re_registers regs
;
5377 regex_t private_preg
;
5378 int len
= strlen (string
);
5379 boolean want_reg_info
= !preg
->no_sub
&& nmatch
> 0;
5381 private_preg
= *preg
;
5383 private_preg
.not_bol
= !!(eflags
& REG_NOTBOL
);
5384 private_preg
.not_eol
= !!(eflags
& REG_NOTEOL
);
5386 /* The user has told us exactly how many registers to return
5387 information about, via `nmatch'. We have to pass that on to the
5388 matching routines. */
5389 private_preg
.regs_allocated
= REGS_FIXED
;
5393 regs
.num_regs
= nmatch
;
5394 regs
.start
= TALLOC (nmatch
, regoff_t
);
5395 regs
.end
= TALLOC (nmatch
, regoff_t
);
5396 if (regs
.start
== NULL
|| regs
.end
== NULL
)
5397 return (int) REG_NOMATCH
;
5400 /* Perform the searching operation. */
5401 ret
= re_search (&private_preg
, string
, len
,
5402 /* start: */ 0, /* range: */ len
,
5403 want_reg_info
? ®s
: (struct re_registers
*) 0);
5405 /* Copy the register information to the POSIX structure. */
5412 for (r
= 0; r
< nmatch
; r
++)
5414 pmatch
[r
].rm_so
= regs
.start
[r
];
5415 pmatch
[r
].rm_eo
= regs
.end
[r
];
5419 /* If we needed the temporary register info, free the space now. */
5424 /* We want zero return to mean success, unlike `re_search'. */
5425 return ret
>= 0 ? (int) REG_NOERROR
: (int) REG_NOMATCH
;
5429 /* Returns a message corresponding to an error code, ERRCODE, returned
5430 from either regcomp or regexec. We don't use PREG here. */
5433 regerror (errcode
, preg
, errbuf
, errbuf_size
)
5435 const regex_t
*preg
;
5443 || errcode
>= (sizeof (re_error_msgid
) / sizeof (re_error_msgid
[0])))
5444 /* Only error codes returned by the rest of the code should be passed
5445 to this routine. If we are given anything else, or if other regex
5446 code generates an invalid error code, then the program has a bug.
5447 Dump core so we can fix it. */
5450 msg
= gettext (re_error_msgid
[errcode
]);
5452 msg_size
= strlen (msg
) + 1; /* Includes the null. */
5454 if (errbuf_size
!= 0)
5456 if (msg_size
> errbuf_size
)
5458 strncpy (errbuf
, msg
, errbuf_size
- 1);
5459 errbuf
[errbuf_size
- 1] = 0;
5462 strcpy (errbuf
, msg
);
5469 /* Free dynamically allocated space used by PREG. */
5475 if (preg
->buffer
!= NULL
)
5476 free (preg
->buffer
);
5477 preg
->buffer
= NULL
;
5479 preg
->allocated
= 0;
5482 if (preg
->fastmap
!= NULL
)
5483 free (preg
->fastmap
);
5484 preg
->fastmap
= NULL
;
5485 preg
->fastmap_accurate
= 0;
5487 if (preg
->translate
!= NULL
)
5488 free (preg
->translate
);
5489 preg
->translate
= NULL
;
5492 #endif /* not emacs */