1 /* Extended regular expression matching and search library,
3 (Implements POSIX draft P1003.2/D11.2, except for some of the
4 internationalization features.)
5 Copyright (C) 1993-1999, 2000 Free Software Foundation, Inc.
7 The GNU C Library is free software; you can redistribute it and/or
8 modify it under the terms of the GNU Library General Public License as
9 published by the Free Software Foundation; either version 2 of the
10 License, or (at your option) any later version.
12 The GNU C Library is distributed in the hope that it will be useful,
13 but WITHOUT ANY WARRANTY; without even the implied warranty of
14 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 Library General Public License for more details.
17 You should have received a copy of the GNU Library General Public
18 License along with the GNU C Library; see the file COPYING.LIB. If not,
19 write to the Free Software Foundation, Inc., 59 Temple Place - Suite 330,
20 Boston, MA 02111-1307, USA. */
22 /* AIX requires this to be the first thing in the file. */
23 #if defined _AIX && !defined REGEX_MALLOC
35 # if defined __GNUC__ || (defined __STDC__ && __STDC__)
36 # define PARAMS(args) args
38 # define PARAMS(args) ()
40 #endif /* Not PARAMS. */
42 #if defined STDC_HEADERS && !defined emacs
45 /* We need this for `regex.h', and perhaps for the Emacs include files. */
46 # include <sys/types.h>
49 #define WIDE_CHAR_SUPPORT (HAVE_WCTYPE_H && HAVE_WCHAR_H && HAVE_BTOWC)
51 /* For platform which support the ISO C amendement 1 functionality we
52 support user defined character classes. */
53 #if defined _LIBC || WIDE_CHAR_SUPPORT
54 /* Solaris 2.5 has a bug: <wchar.h> must be included before <wctype.h>. */
60 /* We have to keep the namespace clean. */
61 # define regfree(preg) __regfree (preg)
62 # define regexec(pr, st, nm, pm, ef) __regexec (pr, st, nm, pm, ef)
63 # define regcomp(preg, pattern, cflags) __regcomp (preg, pattern, cflags)
64 # define regerror(errcode, preg, errbuf, errbuf_size) \
65 __regerror(errcode, preg, errbuf, errbuf_size)
66 # define re_set_registers(bu, re, nu, st, en) \
67 __re_set_registers (bu, re, nu, st, en)
68 # define re_match_2(bufp, string1, size1, string2, size2, pos, regs, stop) \
69 __re_match_2 (bufp, string1, size1, string2, size2, pos, regs, stop)
70 # define re_match(bufp, string, size, pos, regs) \
71 __re_match (bufp, string, size, pos, regs)
72 # define re_search(bufp, string, size, startpos, range, regs) \
73 __re_search (bufp, string, size, startpos, range, regs)
74 # define re_compile_pattern(pattern, length, bufp) \
75 __re_compile_pattern (pattern, length, bufp)
76 # define re_set_syntax(syntax) __re_set_syntax (syntax)
77 # define re_search_2(bufp, st1, s1, st2, s2, startpos, range, regs, stop) \
78 __re_search_2 (bufp, st1, s1, st2, s2, startpos, range, regs, stop)
79 # define re_compile_fastmap(bufp) __re_compile_fastmap (bufp)
81 # define btowc __btowc
83 /* We are also using some library internals. */
84 # include <locale/localeinfo.h>
85 # include <locale/elem-hash.h>
86 # include <langinfo.h>
89 /* This is for other GNU distributions with internationalized messages. */
90 #if HAVE_LIBINTL_H || defined _LIBC
93 # define gettext(msgid) (msgid)
97 /* This define is so xgettext can find the internationalizable
99 # define gettext_noop(String) String
102 /* The `emacs' switch turns on certain matching commands
103 that make sense only in Emacs. */
110 #else /* not emacs */
112 /* If we are not linking with Emacs proper,
113 we can't use the relocating allocator
114 even if config.h says that we can. */
117 # if defined STDC_HEADERS || defined _LIBC
124 /* When used in Emacs's lib-src, we need to get bzero and bcopy somehow.
125 If nothing else has been done, use the method below. */
126 # ifdef INHIBIT_STRING_HEADER
127 # if !(defined HAVE_BZERO && defined HAVE_BCOPY)
128 # if !defined bzero && !defined bcopy
129 # undef INHIBIT_STRING_HEADER
134 /* This is the normal way of making sure we have a bcopy and a bzero.
135 This is used in most programs--a few other programs avoid this
136 by defining INHIBIT_STRING_HEADER. */
137 # ifndef INHIBIT_STRING_HEADER
138 # if defined HAVE_STRING_H || defined STDC_HEADERS || defined _LIBC
142 # define bzero(s, n) (memset (s, '\0', n), (s))
144 # define bzero(s, n) __bzero (s, n)
148 # include <strings.h>
150 # define memcmp(s1, s2, n) bcmp (s1, s2, n)
153 # define memcpy(d, s, n) (bcopy (s, d, n), (d))
158 /* Define the syntax stuff for \<, \>, etc. */
160 /* This must be nonzero for the wordchar and notwordchar pattern
161 commands in re_match_2. */
166 # ifdef SWITCH_ENUM_BUG
167 # define SWITCH_ENUM_CAST(x) ((int)(x))
169 # define SWITCH_ENUM_CAST(x) (x)
172 #endif /* not emacs */
174 #if defined _LIBC || HAVE_LIMITS_H
179 # define MB_LEN_MAX 1
182 /* Get the interface, including the syntax bits. */
185 /* isalpha etc. are used for the character classes. */
188 /* Jim Meyering writes:
190 "... Some ctype macros are valid only for character codes that
191 isascii says are ASCII (SGI's IRIX-4.0.5 is one such system --when
192 using /bin/cc or gcc but without giving an ansi option). So, all
193 ctype uses should be through macros like ISPRINT... If
194 STDC_HEADERS is defined, then autoconf has verified that the ctype
195 macros don't need to be guarded with references to isascii. ...
196 Defining isascii to 1 should let any compiler worth its salt
197 eliminate the && through constant folding."
198 Solaris defines some of these symbols so we must undefine them first. */
201 #if defined STDC_HEADERS || (!defined isascii && !defined HAVE_ISASCII)
202 # define ISASCII(c) 1
204 # define ISASCII(c) isascii(c)
208 # define ISBLANK(c) (ISASCII (c) && isblank (c))
210 # define ISBLANK(c) ((c) == ' ' || (c) == '\t')
213 # define ISGRAPH(c) (ISASCII (c) && isgraph (c))
215 # define ISGRAPH(c) (ISASCII (c) && isprint (c) && !isspace (c))
219 #define ISPRINT(c) (ISASCII (c) && isprint (c))
220 #define ISDIGIT(c) (ISASCII (c) && isdigit (c))
221 #define ISALNUM(c) (ISASCII (c) && isalnum (c))
222 #define ISALPHA(c) (ISASCII (c) && isalpha (c))
223 #define ISCNTRL(c) (ISASCII (c) && iscntrl (c))
224 #define ISLOWER(c) (ISASCII (c) && islower (c))
225 #define ISPUNCT(c) (ISASCII (c) && ispunct (c))
226 #define ISSPACE(c) (ISASCII (c) && isspace (c))
227 #define ISUPPER(c) (ISASCII (c) && isupper (c))
228 #define ISXDIGIT(c) (ISASCII (c) && isxdigit (c))
231 # define TOLOWER(c) _tolower(c)
233 # define TOLOWER(c) tolower(c)
237 # define NULL (void *)0
240 /* We remove any previous definition of `SIGN_EXTEND_CHAR',
241 since ours (we hope) works properly with all combinations of
242 machines, compilers, `char' and `unsigned char' argument types.
243 (Per Bothner suggested the basic approach.) */
244 #undef SIGN_EXTEND_CHAR
246 # define SIGN_EXTEND_CHAR(c) ((signed char) (c))
247 #else /* not __STDC__ */
248 /* As in Harbison and Steele. */
249 # define SIGN_EXTEND_CHAR(c) ((((unsigned char) (c)) ^ 128) - 128)
253 /* How many characters in the character set. */
254 # define CHAR_SET_SIZE 256
258 extern char *re_syntax_table
;
260 # else /* not SYNTAX_TABLE */
262 static char re_syntax_table
[CHAR_SET_SIZE
];
272 bzero (re_syntax_table
, sizeof re_syntax_table
);
274 for (c
= 0; c
< CHAR_SET_SIZE
; ++c
)
276 re_syntax_table
[c
] = Sword
;
278 re_syntax_table
['_'] = Sword
;
283 # endif /* not SYNTAX_TABLE */
285 # define SYNTAX(c) re_syntax_table[(unsigned char) (c)]
289 /* Should we use malloc or alloca? If REGEX_MALLOC is not defined, we
290 use `alloca' instead of `malloc'. This is because using malloc in
291 re_search* or re_match* could cause memory leaks when C-g is used in
292 Emacs; also, malloc is slower and causes storage fragmentation. On
293 the other hand, malloc is more portable, and easier to debug.
295 Because we sometimes use alloca, some routines have to be macros,
296 not functions -- `alloca'-allocated space disappears at the end of the
297 function it is called in. */
301 # define REGEX_ALLOCATE malloc
302 # define REGEX_REALLOCATE(source, osize, nsize) realloc (source, nsize)
303 # define REGEX_FREE free
305 #else /* not REGEX_MALLOC */
307 /* Emacs already defines alloca, sometimes. */
310 /* Make alloca work the best possible way. */
312 # define alloca __builtin_alloca
313 # else /* not __GNUC__ */
316 # endif /* HAVE_ALLOCA_H */
317 # endif /* not __GNUC__ */
319 # endif /* not alloca */
321 # define REGEX_ALLOCATE alloca
323 /* Assumes a `char *destination' variable. */
324 # define REGEX_REALLOCATE(source, osize, nsize) \
325 (destination = (char *) alloca (nsize), \
326 memcpy (destination, source, osize))
328 /* No need to do anything to free, after alloca. */
329 # define REGEX_FREE(arg) ((void)0) /* Do nothing! But inhibit gcc warning. */
331 #endif /* not REGEX_MALLOC */
333 /* Define how to allocate the failure stack. */
335 #if defined REL_ALLOC && defined REGEX_MALLOC
337 # define REGEX_ALLOCATE_STACK(size) \
338 r_alloc (&failure_stack_ptr, (size))
339 # define REGEX_REALLOCATE_STACK(source, osize, nsize) \
340 r_re_alloc (&failure_stack_ptr, (nsize))
341 # define REGEX_FREE_STACK(ptr) \
342 r_alloc_free (&failure_stack_ptr)
344 #else /* not using relocating allocator */
348 # define REGEX_ALLOCATE_STACK malloc
349 # define REGEX_REALLOCATE_STACK(source, osize, nsize) realloc (source, nsize)
350 # define REGEX_FREE_STACK free
352 # else /* not REGEX_MALLOC */
354 # define REGEX_ALLOCATE_STACK alloca
356 # define REGEX_REALLOCATE_STACK(source, osize, nsize) \
357 REGEX_REALLOCATE (source, osize, nsize)
358 /* No need to explicitly free anything. */
359 # define REGEX_FREE_STACK(arg)
361 # endif /* not REGEX_MALLOC */
362 #endif /* not using relocating allocator */
365 /* True if `size1' is non-NULL and PTR is pointing anywhere inside
366 `string1' or just past its end. This works if PTR is NULL, which is
368 #define FIRST_STRING_P(ptr) \
369 (size1 && string1 <= (ptr) && (ptr) <= string1 + size1)
371 /* (Re)Allocate N items of type T using malloc, or fail. */
372 #define TALLOC(n, t) ((t *) malloc ((n) * sizeof (t)))
373 #define RETALLOC(addr, n, t) ((addr) = (t *) realloc (addr, (n) * sizeof (t)))
374 #define RETALLOC_IF(addr, n, t) \
375 if (addr) RETALLOC((addr), (n), t); else (addr) = TALLOC ((n), t)
376 #define REGEX_TALLOC(n, t) ((t *) REGEX_ALLOCATE ((n) * sizeof (t)))
378 #define BYTEWIDTH 8 /* In bits. */
380 #define STREQ(s1, s2) ((strcmp (s1, s2) == 0))
384 #define MAX(a, b) ((a) > (b) ? (a) : (b))
385 #define MIN(a, b) ((a) < (b) ? (a) : (b))
387 typedef char boolean
;
391 static int re_match_2_internal
PARAMS ((struct re_pattern_buffer
*bufp
,
392 const char *string1
, int size1
,
393 const char *string2
, int size2
,
395 struct re_registers
*regs
,
398 /* These are the command codes that appear in compiled regular
399 expressions. Some opcodes are followed by argument bytes. A
400 command code can specify any interpretation whatsoever for its
401 arguments. Zero bytes may appear in the compiled regular expression. */
407 /* Succeed right away--no more backtracking. */
410 /* Followed by one byte giving n, then by n literal bytes. */
413 /* Matches any (more or less) character. */
416 /* Matches any one char belonging to specified set. First
417 following byte is number of bitmap bytes. Then come bytes
418 for a bitmap saying which chars are in. Bits in each byte
419 are ordered low-bit-first. A character is in the set if its
420 bit is 1. A character too large to have a bit in the map is
421 automatically not in the set. */
424 /* Same parameters as charset, but match any character that is
425 not one of those specified. */
428 /* Start remembering the text that is matched, for storing in a
429 register. Followed by one byte with the register number, in
430 the range 0 to one less than the pattern buffer's re_nsub
431 field. Then followed by one byte with the number of groups
432 inner to this one. (This last has to be part of the
433 start_memory only because we need it in the on_failure_jump
437 /* Stop remembering the text that is matched and store it in a
438 memory register. Followed by one byte with the register
439 number, in the range 0 to one less than `re_nsub' in the
440 pattern buffer, and one byte with the number of inner groups,
441 just like `start_memory'. (We need the number of inner
442 groups here because we don't have any easy way of finding the
443 corresponding start_memory when we're at a stop_memory.) */
446 /* Match a duplicate of something remembered. Followed by one
447 byte containing the register number. */
450 /* Fail unless at beginning of line. */
453 /* Fail unless at end of line. */
456 /* Succeeds if at beginning of buffer (if emacs) or at beginning
457 of string to be matched (if not). */
460 /* Analogously, for end of buffer/string. */
463 /* Followed by two byte relative address to which to jump. */
466 /* Same as jump, but marks the end of an alternative. */
469 /* Followed by two-byte relative address of place to resume at
470 in case of failure. */
473 /* Like on_failure_jump, but pushes a placeholder instead of the
474 current string position when executed. */
475 on_failure_keep_string_jump
,
477 /* Throw away latest failure point and then jump to following
478 two-byte relative address. */
481 /* Change to pop_failure_jump if know won't have to backtrack to
482 match; otherwise change to jump. This is used to jump
483 back to the beginning of a repeat. If what follows this jump
484 clearly won't match what the repeat does, such that we can be
485 sure that there is no use backtracking out of repetitions
486 already matched, then we change it to a pop_failure_jump.
487 Followed by two-byte address. */
490 /* Jump to following two-byte address, and push a dummy failure
491 point. This failure point will be thrown away if an attempt
492 is made to use it for a failure. A `+' construct makes this
493 before the first repeat. Also used as an intermediary kind
494 of jump when compiling an alternative. */
497 /* Push a dummy failure point and continue. Used at the end of
501 /* Followed by two-byte relative address and two-byte number n.
502 After matching N times, jump to the address upon failure. */
505 /* Followed by two-byte relative address, and two-byte number n.
506 Jump to the address N times, then fail. */
509 /* Set the following two-byte relative address to the
510 subsequent two-byte number. The address *includes* the two
514 wordchar
, /* Matches any word-constituent character. */
515 notwordchar
, /* Matches any char that is not a word-constituent. */
517 wordbeg
, /* Succeeds if at word beginning. */
518 wordend
, /* Succeeds if at word end. */
520 wordbound
, /* Succeeds if at a word boundary. */
521 notwordbound
/* Succeeds if not at a word boundary. */
524 ,before_dot
, /* Succeeds if before point. */
525 at_dot
, /* Succeeds if at point. */
526 after_dot
, /* Succeeds if after point. */
528 /* Matches any character whose syntax is specified. Followed by
529 a byte which contains a syntax code, e.g., Sword. */
532 /* Matches any character whose syntax is not that specified. */
537 /* Common operations on the compiled pattern. */
539 /* Store NUMBER in two contiguous bytes starting at DESTINATION. */
541 #define STORE_NUMBER(destination, number) \
543 (destination)[0] = (number) & 0377; \
544 (destination)[1] = (number) >> 8; \
547 /* Same as STORE_NUMBER, except increment DESTINATION to
548 the byte after where the number is stored. Therefore, DESTINATION
549 must be an lvalue. */
551 #define STORE_NUMBER_AND_INCR(destination, number) \
553 STORE_NUMBER (destination, number); \
554 (destination) += 2; \
557 /* Put into DESTINATION a number stored in two contiguous bytes starting
560 #define EXTRACT_NUMBER(destination, source) \
562 (destination) = *(source) & 0377; \
563 (destination) += SIGN_EXTEND_CHAR (*((source) + 1)) << 8; \
567 static void extract_number
_RE_ARGS ((int *dest
, unsigned char *source
));
569 extract_number (dest
, source
)
571 unsigned char *source
;
573 int temp
= SIGN_EXTEND_CHAR (*(source
+ 1));
574 *dest
= *source
& 0377;
578 # ifndef EXTRACT_MACROS /* To debug the macros. */
579 # undef EXTRACT_NUMBER
580 # define EXTRACT_NUMBER(dest, src) extract_number (&dest, src)
581 # endif /* not EXTRACT_MACROS */
585 /* Same as EXTRACT_NUMBER, except increment SOURCE to after the number.
586 SOURCE must be an lvalue. */
588 #define EXTRACT_NUMBER_AND_INCR(destination, source) \
590 EXTRACT_NUMBER (destination, source); \
595 static void extract_number_and_incr
_RE_ARGS ((int *destination
,
596 unsigned char **source
));
598 extract_number_and_incr (destination
, source
)
600 unsigned char **source
;
602 extract_number (destination
, *source
);
606 # ifndef EXTRACT_MACROS
607 # undef EXTRACT_NUMBER_AND_INCR
608 # define EXTRACT_NUMBER_AND_INCR(dest, src) \
609 extract_number_and_incr (&dest, &src)
610 # endif /* not EXTRACT_MACROS */
614 /* If DEBUG is defined, Regex prints many voluminous messages about what
615 it is doing (if the variable `debug' is nonzero). If linked with the
616 main program in `iregex.c', you can enter patterns and strings
617 interactively. And if linked with the main program in `main.c' and
618 the other test files, you can run the already-written tests. */
622 /* We use standard I/O for debugging. */
625 /* It is useful to test things that ``must'' be true when debugging. */
630 # define DEBUG_STATEMENT(e) e
631 # define DEBUG_PRINT1(x) if (debug) printf (x)
632 # define DEBUG_PRINT2(x1, x2) if (debug) printf (x1, x2)
633 # define DEBUG_PRINT3(x1, x2, x3) if (debug) printf (x1, x2, x3)
634 # define DEBUG_PRINT4(x1, x2, x3, x4) if (debug) printf (x1, x2, x3, x4)
635 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) \
636 if (debug) print_partial_compiled_pattern (s, e)
637 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) \
638 if (debug) print_double_string (w, s1, sz1, s2, sz2)
641 /* Print the fastmap in human-readable form. */
644 print_fastmap (fastmap
)
647 unsigned was_a_range
= 0;
650 while (i
< (1 << BYTEWIDTH
))
656 while (i
< (1 << BYTEWIDTH
) && fastmap
[i
])
672 /* Print a compiled pattern string in human-readable form, starting at
673 the START pointer into it and ending just before the pointer END. */
676 print_partial_compiled_pattern (start
, end
)
677 unsigned char *start
;
682 unsigned char *p
= start
;
683 unsigned char *pend
= end
;
691 /* Loop over pattern commands. */
695 printf ("%t:\t", p
- start
);
697 printf ("%ld:\t", (long int) (p
- start
));
700 switch ((re_opcode_t
) *p
++)
708 printf ("/exactn/%d", mcnt
);
719 printf ("/start_memory/%d/%d", mcnt
, *p
++);
724 printf ("/stop_memory/%d/%d", mcnt
, *p
++);
728 printf ("/duplicate/%d", *p
++);
738 register int c
, last
= -100;
739 register int in_range
= 0;
741 printf ("/charset [%s",
742 (re_opcode_t
) *(p
- 1) == charset_not
? "^" : "");
744 assert (p
+ *p
< pend
);
746 for (c
= 0; c
< 256; c
++)
748 && (p
[1 + (c
/8)] & (1 << (c
% 8))))
750 /* Are we starting a range? */
751 if (last
+ 1 == c
&& ! in_range
)
756 /* Have we broken a range? */
757 else if (last
+ 1 != c
&& in_range
)
786 case on_failure_jump
:
787 extract_number_and_incr (&mcnt
, &p
);
789 printf ("/on_failure_jump to %t", p
+ mcnt
- start
);
791 printf ("/on_failure_jump to %ld", (long int) (p
+ mcnt
- start
));
795 case on_failure_keep_string_jump
:
796 extract_number_and_incr (&mcnt
, &p
);
798 printf ("/on_failure_keep_string_jump to %t", p
+ mcnt
- start
);
800 printf ("/on_failure_keep_string_jump to %ld",
801 (long int) (p
+ mcnt
- start
));
805 case dummy_failure_jump
:
806 extract_number_and_incr (&mcnt
, &p
);
808 printf ("/dummy_failure_jump to %t", p
+ mcnt
- start
);
810 printf ("/dummy_failure_jump to %ld", (long int) (p
+ mcnt
- start
));
814 case push_dummy_failure
:
815 printf ("/push_dummy_failure");
819 extract_number_and_incr (&mcnt
, &p
);
821 printf ("/maybe_pop_jump to %t", p
+ mcnt
- start
);
823 printf ("/maybe_pop_jump to %ld", (long int) (p
+ mcnt
- start
));
827 case pop_failure_jump
:
828 extract_number_and_incr (&mcnt
, &p
);
830 printf ("/pop_failure_jump to %t", p
+ mcnt
- start
);
832 printf ("/pop_failure_jump to %ld", (long int) (p
+ mcnt
- start
));
837 extract_number_and_incr (&mcnt
, &p
);
839 printf ("/jump_past_alt to %t", p
+ mcnt
- start
);
841 printf ("/jump_past_alt to %ld", (long int) (p
+ mcnt
- start
));
846 extract_number_and_incr (&mcnt
, &p
);
848 printf ("/jump to %t", p
+ mcnt
- start
);
850 printf ("/jump to %ld", (long int) (p
+ mcnt
- start
));
855 extract_number_and_incr (&mcnt
, &p
);
857 extract_number_and_incr (&mcnt2
, &p
);
859 printf ("/succeed_n to %t, %d times", p1
- start
, mcnt2
);
861 printf ("/succeed_n to %ld, %d times",
862 (long int) (p1
- start
), mcnt2
);
867 extract_number_and_incr (&mcnt
, &p
);
869 extract_number_and_incr (&mcnt2
, &p
);
870 printf ("/jump_n to %d, %d times", p1
- start
, mcnt2
);
874 extract_number_and_incr (&mcnt
, &p
);
876 extract_number_and_incr (&mcnt2
, &p
);
878 printf ("/set_number_at location %t to %d", p1
- start
, mcnt2
);
880 printf ("/set_number_at location %ld to %d",
881 (long int) (p1
- start
), mcnt2
);
886 printf ("/wordbound");
890 printf ("/notwordbound");
902 printf ("/before_dot");
910 printf ("/after_dot");
914 printf ("/syntaxspec");
916 printf ("/%d", mcnt
);
920 printf ("/notsyntaxspec");
922 printf ("/%d", mcnt
);
927 printf ("/wordchar");
931 printf ("/notwordchar");
943 printf ("?%d", *(p
-1));
950 printf ("%t:\tend of pattern.\n", p
- start
);
952 printf ("%ld:\tend of pattern.\n", (long int) (p
- start
));
958 print_compiled_pattern (bufp
)
959 struct re_pattern_buffer
*bufp
;
961 unsigned char *buffer
= bufp
->buffer
;
963 print_partial_compiled_pattern (buffer
, buffer
+ bufp
->used
);
964 printf ("%ld bytes used/%ld bytes allocated.\n",
965 bufp
->used
, bufp
->allocated
);
967 if (bufp
->fastmap_accurate
&& bufp
->fastmap
)
969 printf ("fastmap: ");
970 print_fastmap (bufp
->fastmap
);
974 printf ("re_nsub: %Zd\t", bufp
->re_nsub
);
976 printf ("re_nsub: %ld\t", (long int) bufp
->re_nsub
);
978 printf ("regs_alloc: %d\t", bufp
->regs_allocated
);
979 printf ("can_be_null: %d\t", bufp
->can_be_null
);
980 printf ("newline_anchor: %d\n", bufp
->newline_anchor
);
981 printf ("no_sub: %d\t", bufp
->no_sub
);
982 printf ("not_bol: %d\t", bufp
->not_bol
);
983 printf ("not_eol: %d\t", bufp
->not_eol
);
984 printf ("syntax: %lx\n", bufp
->syntax
);
985 /* Perhaps we should print the translate table? */
990 print_double_string (where
, string1
, size1
, string2
, size2
)
1003 if (FIRST_STRING_P (where
))
1005 for (this_char
= where
- string1
; this_char
< size1
; this_char
++)
1006 putchar (string1
[this_char
]);
1011 for (this_char
= where
- string2
; this_char
< size2
; this_char
++)
1012 putchar (string2
[this_char
]);
1023 #else /* not DEBUG */
1028 # define DEBUG_STATEMENT(e)
1029 # define DEBUG_PRINT1(x)
1030 # define DEBUG_PRINT2(x1, x2)
1031 # define DEBUG_PRINT3(x1, x2, x3)
1032 # define DEBUG_PRINT4(x1, x2, x3, x4)
1033 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e)
1034 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2)
1036 #endif /* not DEBUG */
1038 /* Set by `re_set_syntax' to the current regexp syntax to recognize. Can
1039 also be assigned to arbitrarily: each pattern buffer stores its own
1040 syntax, so it can be changed between regex compilations. */
1041 /* This has no initializer because initialized variables in Emacs
1042 become read-only after dumping. */
1043 reg_syntax_t re_syntax_options
;
1046 /* Specify the precise syntax of regexps for compilation. This provides
1047 for compatibility for various utilities which historically have
1048 different, incompatible syntaxes.
1050 The argument SYNTAX is a bit mask comprised of the various bits
1051 defined in regex.h. We return the old syntax. */
1054 re_set_syntax (syntax
)
1055 reg_syntax_t syntax
;
1057 reg_syntax_t ret
= re_syntax_options
;
1059 re_syntax_options
= syntax
;
1061 if (syntax
& RE_DEBUG
)
1063 else if (debug
) /* was on but now is not */
1069 weak_alias (__re_set_syntax
, re_set_syntax
)
1072 /* This table gives an error message for each of the error codes listed
1073 in regex.h. Obviously the order here has to be same as there.
1074 POSIX doesn't require that we do anything for REG_NOERROR,
1075 but why not be nice? */
1077 static const char re_error_msgid
[] =
1079 #define REG_NOERROR_IDX 0
1080 gettext_noop ("Success") /* REG_NOERROR */
1082 #define REG_NOMATCH_IDX (REG_NOERROR_IDX + sizeof "Success")
1083 gettext_noop ("No match") /* REG_NOMATCH */
1085 #define REG_BADPAT_IDX (REG_NOMATCH_IDX + sizeof "No match")
1086 gettext_noop ("Invalid regular expression") /* REG_BADPAT */
1088 #define REG_ECOLLATE_IDX (REG_BADPAT_IDX + sizeof "Invalid regular expression")
1089 gettext_noop ("Invalid collation character") /* REG_ECOLLATE */
1091 #define REG_ECTYPE_IDX (REG_ECOLLATE_IDX + sizeof "Invalid collation character")
1092 gettext_noop ("Invalid character class name") /* REG_ECTYPE */
1094 #define REG_EESCAPE_IDX (REG_ECTYPE_IDX + sizeof "Invalid character class name")
1095 gettext_noop ("Trailing backslash") /* REG_EESCAPE */
1097 #define REG_ESUBREG_IDX (REG_EESCAPE_IDX + sizeof "Trailing backslash")
1098 gettext_noop ("Invalid back reference") /* REG_ESUBREG */
1100 #define REG_EBRACK_IDX (REG_ESUBREG_IDX + sizeof "Invalid back reference")
1101 gettext_noop ("Unmatched [ or [^") /* REG_EBRACK */
1103 #define REG_EPAREN_IDX (REG_EBRACK_IDX + sizeof "Unmatched [ or [^")
1104 gettext_noop ("Unmatched ( or \\(") /* REG_EPAREN */
1106 #define REG_EBRACE_IDX (REG_EPAREN_IDX + sizeof "Unmatched ( or \\(")
1107 gettext_noop ("Unmatched \\{") /* REG_EBRACE */
1109 #define REG_BADBR_IDX (REG_EBRACE_IDX + sizeof "Unmatched \\{")
1110 gettext_noop ("Invalid content of \\{\\}") /* REG_BADBR */
1112 #define REG_ERANGE_IDX (REG_BADBR_IDX + sizeof "Invalid content of \\{\\}")
1113 gettext_noop ("Invalid range end") /* REG_ERANGE */
1115 #define REG_ESPACE_IDX (REG_ERANGE_IDX + sizeof "Invalid range end")
1116 gettext_noop ("Memory exhausted") /* REG_ESPACE */
1118 #define REG_BADRPT_IDX (REG_ESPACE_IDX + sizeof "Memory exhausted")
1119 gettext_noop ("Invalid preceding regular expression") /* REG_BADRPT */
1121 #define REG_EEND_IDX (REG_BADRPT_IDX + sizeof "Invalid preceding regular expression")
1122 gettext_noop ("Premature end of regular expression") /* REG_EEND */
1124 #define REG_ESIZE_IDX (REG_EEND_IDX + sizeof "Premature end of regular expression")
1125 gettext_noop ("Regular expression too big") /* REG_ESIZE */
1127 #define REG_ERPAREN_IDX (REG_ESIZE_IDX + sizeof "Regular expression too big")
1128 gettext_noop ("Unmatched ) or \\)") /* REG_ERPAREN */
1131 static const size_t re_error_msgid_idx
[] =
1152 /* Avoiding alloca during matching, to placate r_alloc. */
1154 /* Define MATCH_MAY_ALLOCATE unless we need to make sure that the
1155 searching and matching functions should not call alloca. On some
1156 systems, alloca is implemented in terms of malloc, and if we're
1157 using the relocating allocator routines, then malloc could cause a
1158 relocation, which might (if the strings being searched are in the
1159 ralloc heap) shift the data out from underneath the regexp
1162 Here's another reason to avoid allocation: Emacs
1163 processes input from X in a signal handler; processing X input may
1164 call malloc; if input arrives while a matching routine is calling
1165 malloc, then we're scrod. But Emacs can't just block input while
1166 calling matching routines; then we don't notice interrupts when
1167 they come in. So, Emacs blocks input around all regexp calls
1168 except the matching calls, which it leaves unprotected, in the
1169 faith that they will not malloc. */
1171 /* Normally, this is fine. */
1172 #define MATCH_MAY_ALLOCATE
1174 /* When using GNU C, we are not REALLY using the C alloca, no matter
1175 what config.h may say. So don't take precautions for it. */
1180 /* The match routines may not allocate if (1) they would do it with malloc
1181 and (2) it's not safe for them to use malloc.
1182 Note that if REL_ALLOC is defined, matching would not use malloc for the
1183 failure stack, but we would still use it for the register vectors;
1184 so REL_ALLOC should not affect this. */
1185 #if (defined C_ALLOCA || defined REGEX_MALLOC) && defined emacs
1186 # undef MATCH_MAY_ALLOCATE
1190 /* Failure stack declarations and macros; both re_compile_fastmap and
1191 re_match_2 use a failure stack. These have to be macros because of
1192 REGEX_ALLOCATE_STACK. */
1195 /* Number of failure points for which to initially allocate space
1196 when matching. If this number is exceeded, we allocate more
1197 space, so it is not a hard limit. */
1198 #ifndef INIT_FAILURE_ALLOC
1199 # define INIT_FAILURE_ALLOC 5
1202 /* Roughly the maximum number of failure points on the stack. Would be
1203 exactly that if always used MAX_FAILURE_ITEMS items each time we failed.
1204 This is a variable only so users of regex can assign to it; we never
1205 change it ourselves. */
1209 # if defined MATCH_MAY_ALLOCATE
1210 /* 4400 was enough to cause a crash on Alpha OSF/1,
1211 whose default stack limit is 2mb. */
1212 long int re_max_failures
= 4000;
1214 long int re_max_failures
= 2000;
1217 union fail_stack_elt
1219 unsigned char *pointer
;
1223 typedef union fail_stack_elt fail_stack_elt_t
;
1227 fail_stack_elt_t
*stack
;
1228 unsigned long int size
;
1229 unsigned long int avail
; /* Offset of next open position. */
1232 #else /* not INT_IS_16BIT */
1234 # if defined MATCH_MAY_ALLOCATE
1235 /* 4400 was enough to cause a crash on Alpha OSF/1,
1236 whose default stack limit is 2mb. */
1237 int re_max_failures
= 20000;
1239 int re_max_failures
= 2000;
1242 union fail_stack_elt
1244 unsigned char *pointer
;
1248 typedef union fail_stack_elt fail_stack_elt_t
;
1252 fail_stack_elt_t
*stack
;
1254 unsigned avail
; /* Offset of next open position. */
1257 #endif /* INT_IS_16BIT */
1259 #define FAIL_STACK_EMPTY() (fail_stack.avail == 0)
1260 #define FAIL_STACK_PTR_EMPTY() (fail_stack_ptr->avail == 0)
1261 #define FAIL_STACK_FULL() (fail_stack.avail == fail_stack.size)
1264 /* Define macros to initialize and free the failure stack.
1265 Do `return -2' if the alloc fails. */
1267 #ifdef MATCH_MAY_ALLOCATE
1268 # define INIT_FAIL_STACK() \
1270 fail_stack.stack = (fail_stack_elt_t *) \
1271 REGEX_ALLOCATE_STACK (INIT_FAILURE_ALLOC * sizeof (fail_stack_elt_t)); \
1273 if (fail_stack.stack == NULL) \
1276 fail_stack.size = INIT_FAILURE_ALLOC; \
1277 fail_stack.avail = 0; \
1280 # define RESET_FAIL_STACK() REGEX_FREE_STACK (fail_stack.stack)
1282 # define INIT_FAIL_STACK() \
1284 fail_stack.avail = 0; \
1287 # define RESET_FAIL_STACK()
1291 /* Double the size of FAIL_STACK, up to approximately `re_max_failures' items.
1293 Return 1 if succeeds, and 0 if either ran out of memory
1294 allocating space for it or it was already too large.
1296 REGEX_REALLOCATE_STACK requires `destination' be declared. */
1298 #define DOUBLE_FAIL_STACK(fail_stack) \
1299 ((fail_stack).size > (unsigned) (re_max_failures * MAX_FAILURE_ITEMS) \
1301 : ((fail_stack).stack = (fail_stack_elt_t *) \
1302 REGEX_REALLOCATE_STACK ((fail_stack).stack, \
1303 (fail_stack).size * sizeof (fail_stack_elt_t), \
1304 ((fail_stack).size << 1) * sizeof (fail_stack_elt_t)), \
1306 (fail_stack).stack == NULL \
1308 : ((fail_stack).size <<= 1, \
1312 /* Push pointer POINTER on FAIL_STACK.
1313 Return 1 if was able to do so and 0 if ran out of memory allocating
1315 #define PUSH_PATTERN_OP(POINTER, FAIL_STACK) \
1316 ((FAIL_STACK_FULL () \
1317 && !DOUBLE_FAIL_STACK (FAIL_STACK)) \
1319 : ((FAIL_STACK).stack[(FAIL_STACK).avail++].pointer = POINTER, \
1322 /* Push a pointer value onto the failure stack.
1323 Assumes the variable `fail_stack'. Probably should only
1324 be called from within `PUSH_FAILURE_POINT'. */
1325 #define PUSH_FAILURE_POINTER(item) \
1326 fail_stack.stack[fail_stack.avail++].pointer = (unsigned char *) (item)
1328 /* This pushes an integer-valued item onto the failure stack.
1329 Assumes the variable `fail_stack'. Probably should only
1330 be called from within `PUSH_FAILURE_POINT'. */
1331 #define PUSH_FAILURE_INT(item) \
1332 fail_stack.stack[fail_stack.avail++].integer = (item)
1334 /* Push a fail_stack_elt_t value onto the failure stack.
1335 Assumes the variable `fail_stack'. Probably should only
1336 be called from within `PUSH_FAILURE_POINT'. */
1337 #define PUSH_FAILURE_ELT(item) \
1338 fail_stack.stack[fail_stack.avail++] = (item)
1340 /* These three POP... operations complement the three PUSH... operations.
1341 All assume that `fail_stack' is nonempty. */
1342 #define POP_FAILURE_POINTER() fail_stack.stack[--fail_stack.avail].pointer
1343 #define POP_FAILURE_INT() fail_stack.stack[--fail_stack.avail].integer
1344 #define POP_FAILURE_ELT() fail_stack.stack[--fail_stack.avail]
1346 /* Used to omit pushing failure point id's when we're not debugging. */
1348 # define DEBUG_PUSH PUSH_FAILURE_INT
1349 # define DEBUG_POP(item_addr) *(item_addr) = POP_FAILURE_INT ()
1351 # define DEBUG_PUSH(item)
1352 # define DEBUG_POP(item_addr)
1356 /* Push the information about the state we will need
1357 if we ever fail back to it.
1359 Requires variables fail_stack, regstart, regend, reg_info, and
1360 num_regs_pushed be declared. DOUBLE_FAIL_STACK requires `destination'
1363 Does `return FAILURE_CODE' if runs out of memory. */
1365 #define PUSH_FAILURE_POINT(pattern_place, string_place, failure_code) \
1367 char *destination; \
1368 /* Must be int, so when we don't save any registers, the arithmetic \
1369 of 0 + -1 isn't done as unsigned. */ \
1370 /* Can't be int, since there is not a shred of a guarantee that int \
1371 is wide enough to hold a value of something to which pointer can \
1373 active_reg_t this_reg; \
1375 DEBUG_STATEMENT (failure_id++); \
1376 DEBUG_STATEMENT (nfailure_points_pushed++); \
1377 DEBUG_PRINT2 ("\nPUSH_FAILURE_POINT #%u:\n", failure_id); \
1378 DEBUG_PRINT2 (" Before push, next avail: %d\n", (fail_stack).avail);\
1379 DEBUG_PRINT2 (" size: %d\n", (fail_stack).size);\
1381 DEBUG_PRINT2 (" slots needed: %ld\n", NUM_FAILURE_ITEMS); \
1382 DEBUG_PRINT2 (" available: %d\n", REMAINING_AVAIL_SLOTS); \
1384 /* Ensure we have enough space allocated for what we will push. */ \
1385 while (REMAINING_AVAIL_SLOTS < NUM_FAILURE_ITEMS) \
1387 if (!DOUBLE_FAIL_STACK (fail_stack)) \
1388 return failure_code; \
1390 DEBUG_PRINT2 ("\n Doubled stack; size now: %d\n", \
1391 (fail_stack).size); \
1392 DEBUG_PRINT2 (" slots available: %d\n", REMAINING_AVAIL_SLOTS);\
1395 /* Push the info, starting with the registers. */ \
1396 DEBUG_PRINT1 ("\n"); \
1399 for (this_reg = lowest_active_reg; this_reg <= highest_active_reg; \
1402 DEBUG_PRINT2 (" Pushing reg: %lu\n", this_reg); \
1403 DEBUG_STATEMENT (num_regs_pushed++); \
1405 DEBUG_PRINT2 (" start: %p\n", regstart[this_reg]); \
1406 PUSH_FAILURE_POINTER (regstart[this_reg]); \
1408 DEBUG_PRINT2 (" end: %p\n", regend[this_reg]); \
1409 PUSH_FAILURE_POINTER (regend[this_reg]); \
1411 DEBUG_PRINT2 (" info: %p\n ", \
1412 reg_info[this_reg].word.pointer); \
1413 DEBUG_PRINT2 (" match_null=%d", \
1414 REG_MATCH_NULL_STRING_P (reg_info[this_reg])); \
1415 DEBUG_PRINT2 (" active=%d", IS_ACTIVE (reg_info[this_reg])); \
1416 DEBUG_PRINT2 (" matched_something=%d", \
1417 MATCHED_SOMETHING (reg_info[this_reg])); \
1418 DEBUG_PRINT2 (" ever_matched=%d", \
1419 EVER_MATCHED_SOMETHING (reg_info[this_reg])); \
1420 DEBUG_PRINT1 ("\n"); \
1421 PUSH_FAILURE_ELT (reg_info[this_reg].word); \
1424 DEBUG_PRINT2 (" Pushing low active reg: %ld\n", lowest_active_reg);\
1425 PUSH_FAILURE_INT (lowest_active_reg); \
1427 DEBUG_PRINT2 (" Pushing high active reg: %ld\n", highest_active_reg);\
1428 PUSH_FAILURE_INT (highest_active_reg); \
1430 DEBUG_PRINT2 (" Pushing pattern %p:\n", pattern_place); \
1431 DEBUG_PRINT_COMPILED_PATTERN (bufp, pattern_place, pend); \
1432 PUSH_FAILURE_POINTER (pattern_place); \
1434 DEBUG_PRINT2 (" Pushing string %p: `", string_place); \
1435 DEBUG_PRINT_DOUBLE_STRING (string_place, string1, size1, string2, \
1437 DEBUG_PRINT1 ("'\n"); \
1438 PUSH_FAILURE_POINTER (string_place); \
1440 DEBUG_PRINT2 (" Pushing failure id: %u\n", failure_id); \
1441 DEBUG_PUSH (failure_id); \
1444 /* This is the number of items that are pushed and popped on the stack
1445 for each register. */
1446 #define NUM_REG_ITEMS 3
1448 /* Individual items aside from the registers. */
1450 # define NUM_NONREG_ITEMS 5 /* Includes failure point id. */
1452 # define NUM_NONREG_ITEMS 4
1455 /* We push at most this many items on the stack. */
1456 /* We used to use (num_regs - 1), which is the number of registers
1457 this regexp will save; but that was changed to 5
1458 to avoid stack overflow for a regexp with lots of parens. */
1459 #define MAX_FAILURE_ITEMS (5 * NUM_REG_ITEMS + NUM_NONREG_ITEMS)
1461 /* We actually push this many items. */
1462 #define NUM_FAILURE_ITEMS \
1464 ? 0 : highest_active_reg - lowest_active_reg + 1) \
1468 /* How many items can still be added to the stack without overflowing it. */
1469 #define REMAINING_AVAIL_SLOTS ((fail_stack).size - (fail_stack).avail)
1472 /* Pops what PUSH_FAIL_STACK pushes.
1474 We restore into the parameters, all of which should be lvalues:
1475 STR -- the saved data position.
1476 PAT -- the saved pattern position.
1477 LOW_REG, HIGH_REG -- the highest and lowest active registers.
1478 REGSTART, REGEND -- arrays of string positions.
1479 REG_INFO -- array of information about each subexpression.
1481 Also assumes the variables `fail_stack' and (if debugging), `bufp',
1482 `pend', `string1', `size1', `string2', and `size2'. */
1484 #define POP_FAILURE_POINT(str, pat, low_reg, high_reg, regstart, regend, reg_info)\
1486 DEBUG_STATEMENT (unsigned failure_id;) \
1487 active_reg_t this_reg; \
1488 const unsigned char *string_temp; \
1490 assert (!FAIL_STACK_EMPTY ()); \
1492 /* Remove failure points and point to how many regs pushed. */ \
1493 DEBUG_PRINT1 ("POP_FAILURE_POINT:\n"); \
1494 DEBUG_PRINT2 (" Before pop, next avail: %d\n", fail_stack.avail); \
1495 DEBUG_PRINT2 (" size: %d\n", fail_stack.size); \
1497 assert (fail_stack.avail >= NUM_NONREG_ITEMS); \
1499 DEBUG_POP (&failure_id); \
1500 DEBUG_PRINT2 (" Popping failure id: %u\n", failure_id); \
1502 /* If the saved string location is NULL, it came from an \
1503 on_failure_keep_string_jump opcode, and we want to throw away the \
1504 saved NULL, thus retaining our current position in the string. */ \
1505 string_temp = POP_FAILURE_POINTER (); \
1506 if (string_temp != NULL) \
1507 str = (const char *) string_temp; \
1509 DEBUG_PRINT2 (" Popping string %p: `", str); \
1510 DEBUG_PRINT_DOUBLE_STRING (str, string1, size1, string2, size2); \
1511 DEBUG_PRINT1 ("'\n"); \
1513 pat = (unsigned char *) POP_FAILURE_POINTER (); \
1514 DEBUG_PRINT2 (" Popping pattern %p:\n", pat); \
1515 DEBUG_PRINT_COMPILED_PATTERN (bufp, pat, pend); \
1517 /* Restore register info. */ \
1518 high_reg = (active_reg_t) POP_FAILURE_INT (); \
1519 DEBUG_PRINT2 (" Popping high active reg: %ld\n", high_reg); \
1521 low_reg = (active_reg_t) POP_FAILURE_INT (); \
1522 DEBUG_PRINT2 (" Popping low active reg: %ld\n", low_reg); \
1525 for (this_reg = high_reg; this_reg >= low_reg; this_reg--) \
1527 DEBUG_PRINT2 (" Popping reg: %ld\n", this_reg); \
1529 reg_info[this_reg].word = POP_FAILURE_ELT (); \
1530 DEBUG_PRINT2 (" info: %p\n", \
1531 reg_info[this_reg].word.pointer); \
1533 regend[this_reg] = (const char *) POP_FAILURE_POINTER (); \
1534 DEBUG_PRINT2 (" end: %p\n", regend[this_reg]); \
1536 regstart[this_reg] = (const char *) POP_FAILURE_POINTER (); \
1537 DEBUG_PRINT2 (" start: %p\n", regstart[this_reg]); \
1541 for (this_reg = highest_active_reg; this_reg > high_reg; this_reg--) \
1543 reg_info[this_reg].word.integer = 0; \
1544 regend[this_reg] = 0; \
1545 regstart[this_reg] = 0; \
1547 highest_active_reg = high_reg; \
1550 set_regs_matched_done = 0; \
1551 DEBUG_STATEMENT (nfailure_points_popped++); \
1552 } /* POP_FAILURE_POINT */
1556 /* Structure for per-register (a.k.a. per-group) information.
1557 Other register information, such as the
1558 starting and ending positions (which are addresses), and the list of
1559 inner groups (which is a bits list) are maintained in separate
1562 We are making a (strictly speaking) nonportable assumption here: that
1563 the compiler will pack our bit fields into something that fits into
1564 the type of `word', i.e., is something that fits into one item on the
1568 /* Declarations and macros for re_match_2. */
1572 fail_stack_elt_t word
;
1575 /* This field is one if this group can match the empty string,
1576 zero if not. If not yet determined, `MATCH_NULL_UNSET_VALUE'. */
1577 #define MATCH_NULL_UNSET_VALUE 3
1578 unsigned match_null_string_p
: 2;
1579 unsigned is_active
: 1;
1580 unsigned matched_something
: 1;
1581 unsigned ever_matched_something
: 1;
1583 } register_info_type
;
1585 #define REG_MATCH_NULL_STRING_P(R) ((R).bits.match_null_string_p)
1586 #define IS_ACTIVE(R) ((R).bits.is_active)
1587 #define MATCHED_SOMETHING(R) ((R).bits.matched_something)
1588 #define EVER_MATCHED_SOMETHING(R) ((R).bits.ever_matched_something)
1591 /* Call this when have matched a real character; it sets `matched' flags
1592 for the subexpressions which we are currently inside. Also records
1593 that those subexprs have matched. */
1594 #define SET_REGS_MATCHED() \
1597 if (!set_regs_matched_done) \
1600 set_regs_matched_done = 1; \
1601 for (r = lowest_active_reg; r <= highest_active_reg; r++) \
1603 MATCHED_SOMETHING (reg_info[r]) \
1604 = EVER_MATCHED_SOMETHING (reg_info[r]) \
1611 /* Registers are set to a sentinel when they haven't yet matched. */
1612 static char reg_unset_dummy
;
1613 #define REG_UNSET_VALUE (®_unset_dummy)
1614 #define REG_UNSET(e) ((e) == REG_UNSET_VALUE)
1616 /* Subroutine declarations and macros for regex_compile. */
1618 static reg_errcode_t regex_compile
_RE_ARGS ((const char *pattern
, size_t size
,
1619 reg_syntax_t syntax
,
1620 struct re_pattern_buffer
*bufp
));
1621 static void store_op1
_RE_ARGS ((re_opcode_t op
, unsigned char *loc
, int arg
));
1622 static void store_op2
_RE_ARGS ((re_opcode_t op
, unsigned char *loc
,
1623 int arg1
, int arg2
));
1624 static void insert_op1
_RE_ARGS ((re_opcode_t op
, unsigned char *loc
,
1625 int arg
, unsigned char *end
));
1626 static void insert_op2
_RE_ARGS ((re_opcode_t op
, unsigned char *loc
,
1627 int arg1
, int arg2
, unsigned char *end
));
1628 static boolean at_begline_loc_p
_RE_ARGS ((const char *pattern
, const char *p
,
1629 reg_syntax_t syntax
));
1630 static boolean at_endline_loc_p
_RE_ARGS ((const char *p
, const char *pend
,
1631 reg_syntax_t syntax
));
1632 static reg_errcode_t compile_range
_RE_ARGS ((unsigned int range_start
,
1636 reg_syntax_t syntax
,
1639 /* Fetch the next character in the uncompiled pattern---translating it
1640 if necessary. Also cast from a signed character in the constant
1641 string passed to us by the user to an unsigned char that we can use
1642 as an array index (in, e.g., `translate'). */
1644 # define PATFETCH(c) \
1645 do {if (p == pend) return REG_EEND; \
1646 c = (unsigned char) *p++; \
1647 if (translate) c = (unsigned char) translate[c]; \
1651 /* Fetch the next character in the uncompiled pattern, with no
1653 #define PATFETCH_RAW(c) \
1654 do {if (p == pend) return REG_EEND; \
1655 c = (unsigned char) *p++; \
1658 /* Go backwards one character in the pattern. */
1659 #define PATUNFETCH p--
1662 /* If `translate' is non-null, return translate[D], else just D. We
1663 cast the subscript to translate because some data is declared as
1664 `char *', to avoid warnings when a string constant is passed. But
1665 when we use a character as a subscript we must make it unsigned. */
1667 # define TRANSLATE(d) \
1668 (translate ? (char) translate[(unsigned char) (d)] : (d))
1672 /* Macros for outputting the compiled pattern into `buffer'. */
1674 /* If the buffer isn't allocated when it comes in, use this. */
1675 #define INIT_BUF_SIZE 32
1677 /* Make sure we have at least N more bytes of space in buffer. */
1678 #define GET_BUFFER_SPACE(n) \
1679 while ((unsigned long) (b - bufp->buffer + (n)) > bufp->allocated) \
1682 /* Make sure we have one more byte of buffer space and then add C to it. */
1683 #define BUF_PUSH(c) \
1685 GET_BUFFER_SPACE (1); \
1686 *b++ = (unsigned char) (c); \
1690 /* Ensure we have two more bytes of buffer space and then append C1 and C2. */
1691 #define BUF_PUSH_2(c1, c2) \
1693 GET_BUFFER_SPACE (2); \
1694 *b++ = (unsigned char) (c1); \
1695 *b++ = (unsigned char) (c2); \
1699 /* As with BUF_PUSH_2, except for three bytes. */
1700 #define BUF_PUSH_3(c1, c2, c3) \
1702 GET_BUFFER_SPACE (3); \
1703 *b++ = (unsigned char) (c1); \
1704 *b++ = (unsigned char) (c2); \
1705 *b++ = (unsigned char) (c3); \
1709 /* Store a jump with opcode OP at LOC to location TO. We store a
1710 relative address offset by the three bytes the jump itself occupies. */
1711 #define STORE_JUMP(op, loc, to) \
1712 store_op1 (op, loc, (int) ((to) - (loc) - 3))
1714 /* Likewise, for a two-argument jump. */
1715 #define STORE_JUMP2(op, loc, to, arg) \
1716 store_op2 (op, loc, (int) ((to) - (loc) - 3), arg)
1718 /* Like `STORE_JUMP', but for inserting. Assume `b' is the buffer end. */
1719 #define INSERT_JUMP(op, loc, to) \
1720 insert_op1 (op, loc, (int) ((to) - (loc) - 3), b)
1722 /* Like `STORE_JUMP2', but for inserting. Assume `b' is the buffer end. */
1723 #define INSERT_JUMP2(op, loc, to, arg) \
1724 insert_op2 (op, loc, (int) ((to) - (loc) - 3), arg, b)
1727 /* This is not an arbitrary limit: the arguments which represent offsets
1728 into the pattern are two bytes long. So if 2^16 bytes turns out to
1729 be too small, many things would have to change. */
1730 /* Any other compiler which, like MSC, has allocation limit below 2^16
1731 bytes will have to use approach similar to what was done below for
1732 MSC and drop MAX_BUF_SIZE a bit. Otherwise you may end up
1733 reallocating to 0 bytes. Such thing is not going to work too well.
1734 You have been warned!! */
1735 #if defined _MSC_VER && !defined WIN32
1736 /* Microsoft C 16-bit versions limit malloc to approx 65512 bytes.
1737 The REALLOC define eliminates a flurry of conversion warnings,
1738 but is not required. */
1739 # define MAX_BUF_SIZE 65500L
1740 # define REALLOC(p,s) realloc ((p), (size_t) (s))
1742 # define MAX_BUF_SIZE (1L << 16)
1743 # define REALLOC(p,s) realloc ((p), (s))
1746 /* Extend the buffer by twice its current size via realloc and
1747 reset the pointers that pointed into the old block to point to the
1748 correct places in the new one. If extending the buffer results in it
1749 being larger than MAX_BUF_SIZE, then flag memory exhausted. */
1750 #define EXTEND_BUFFER() \
1752 unsigned char *old_buffer = bufp->buffer; \
1753 if (bufp->allocated == MAX_BUF_SIZE) \
1755 bufp->allocated <<= 1; \
1756 if (bufp->allocated > MAX_BUF_SIZE) \
1757 bufp->allocated = MAX_BUF_SIZE; \
1758 bufp->buffer = (unsigned char *) REALLOC (bufp->buffer, bufp->allocated);\
1759 if (bufp->buffer == NULL) \
1760 return REG_ESPACE; \
1761 /* If the buffer moved, move all the pointers into it. */ \
1762 if (old_buffer != bufp->buffer) \
1764 b = (b - old_buffer) + bufp->buffer; \
1765 begalt = (begalt - old_buffer) + bufp->buffer; \
1766 if (fixup_alt_jump) \
1767 fixup_alt_jump = (fixup_alt_jump - old_buffer) + bufp->buffer;\
1769 laststart = (laststart - old_buffer) + bufp->buffer; \
1770 if (pending_exact) \
1771 pending_exact = (pending_exact - old_buffer) + bufp->buffer; \
1776 /* Since we have one byte reserved for the register number argument to
1777 {start,stop}_memory, the maximum number of groups we can report
1778 things about is what fits in that byte. */
1779 #define MAX_REGNUM 255
1781 /* But patterns can have more than `MAX_REGNUM' registers. We just
1782 ignore the excess. */
1783 typedef unsigned regnum_t
;
1786 /* Macros for the compile stack. */
1788 /* Since offsets can go either forwards or backwards, this type needs to
1789 be able to hold values from -(MAX_BUF_SIZE - 1) to MAX_BUF_SIZE - 1. */
1790 /* int may be not enough when sizeof(int) == 2. */
1791 typedef long pattern_offset_t
;
1795 pattern_offset_t begalt_offset
;
1796 pattern_offset_t fixup_alt_jump
;
1797 pattern_offset_t inner_group_offset
;
1798 pattern_offset_t laststart_offset
;
1800 } compile_stack_elt_t
;
1805 compile_stack_elt_t
*stack
;
1807 unsigned avail
; /* Offset of next open position. */
1808 } compile_stack_type
;
1811 #define INIT_COMPILE_STACK_SIZE 32
1813 #define COMPILE_STACK_EMPTY (compile_stack.avail == 0)
1814 #define COMPILE_STACK_FULL (compile_stack.avail == compile_stack.size)
1816 /* The next available element. */
1817 #define COMPILE_STACK_TOP (compile_stack.stack[compile_stack.avail])
1820 /* Set the bit for character C in a list. */
1821 #define SET_LIST_BIT(c) \
1822 (b[((unsigned char) (c)) / BYTEWIDTH] \
1823 |= 1 << (((unsigned char) c) % BYTEWIDTH))
1826 /* Get the next unsigned number in the uncompiled pattern. */
1827 #define GET_UNSIGNED_NUMBER(num) \
1831 while ('0' <= c && c <= '9') \
1835 num = num * 10 + c - '0'; \
1843 #if defined _LIBC || WIDE_CHAR_SUPPORT
1844 /* The GNU C library provides support for user-defined character classes
1845 and the functions from ISO C amendement 1. */
1846 # ifdef CHARCLASS_NAME_MAX
1847 # define CHAR_CLASS_MAX_LENGTH CHARCLASS_NAME_MAX
1849 /* This shouldn't happen but some implementation might still have this
1850 problem. Use a reasonable default value. */
1851 # define CHAR_CLASS_MAX_LENGTH 256
1855 # define IS_CHAR_CLASS(string) __wctype (string)
1857 # define IS_CHAR_CLASS(string) wctype (string)
1860 # define CHAR_CLASS_MAX_LENGTH 6 /* Namely, `xdigit'. */
1862 # define IS_CHAR_CLASS(string) \
1863 (STREQ (string, "alpha") || STREQ (string, "upper") \
1864 || STREQ (string, "lower") || STREQ (string, "digit") \
1865 || STREQ (string, "alnum") || STREQ (string, "xdigit") \
1866 || STREQ (string, "space") || STREQ (string, "print") \
1867 || STREQ (string, "punct") || STREQ (string, "graph") \
1868 || STREQ (string, "cntrl") || STREQ (string, "blank"))
1871 #ifndef MATCH_MAY_ALLOCATE
1873 /* If we cannot allocate large objects within re_match_2_internal,
1874 we make the fail stack and register vectors global.
1875 The fail stack, we grow to the maximum size when a regexp
1877 The register vectors, we adjust in size each time we
1878 compile a regexp, according to the number of registers it needs. */
1880 static fail_stack_type fail_stack
;
1882 /* Size with which the following vectors are currently allocated.
1883 That is so we can make them bigger as needed,
1884 but never make them smaller. */
1885 static int regs_allocated_size
;
1887 static const char ** regstart
, ** regend
;
1888 static const char ** old_regstart
, ** old_regend
;
1889 static const char **best_regstart
, **best_regend
;
1890 static register_info_type
*reg_info
;
1891 static const char **reg_dummy
;
1892 static register_info_type
*reg_info_dummy
;
1894 /* Make the register vectors big enough for NUM_REGS registers,
1895 but don't make them smaller. */
1898 regex_grow_registers (num_regs
)
1901 if (num_regs
> regs_allocated_size
)
1903 RETALLOC_IF (regstart
, num_regs
, const char *);
1904 RETALLOC_IF (regend
, num_regs
, const char *);
1905 RETALLOC_IF (old_regstart
, num_regs
, const char *);
1906 RETALLOC_IF (old_regend
, num_regs
, const char *);
1907 RETALLOC_IF (best_regstart
, num_regs
, const char *);
1908 RETALLOC_IF (best_regend
, num_regs
, const char *);
1909 RETALLOC_IF (reg_info
, num_regs
, register_info_type
);
1910 RETALLOC_IF (reg_dummy
, num_regs
, const char *);
1911 RETALLOC_IF (reg_info_dummy
, num_regs
, register_info_type
);
1913 regs_allocated_size
= num_regs
;
1917 #endif /* not MATCH_MAY_ALLOCATE */
1919 static boolean group_in_compile_stack
_RE_ARGS ((compile_stack_type
1923 /* `regex_compile' compiles PATTERN (of length SIZE) according to SYNTAX.
1924 Returns one of error codes defined in `regex.h', or zero for success.
1926 Assumes the `allocated' (and perhaps `buffer') and `translate'
1927 fields are set in BUFP on entry.
1929 If it succeeds, results are put in BUFP (if it returns an error, the
1930 contents of BUFP are undefined):
1931 `buffer' is the compiled pattern;
1932 `syntax' is set to SYNTAX;
1933 `used' is set to the length of the compiled pattern;
1934 `fastmap_accurate' is zero;
1935 `re_nsub' is the number of subexpressions in PATTERN;
1936 `not_bol' and `not_eol' are zero;
1938 The `fastmap' and `newline_anchor' fields are neither
1939 examined nor set. */
1941 /* Return, freeing storage we allocated. */
1942 #define FREE_STACK_RETURN(value) \
1943 return (free (compile_stack.stack), value)
1945 static reg_errcode_t
1946 regex_compile (pattern
, size
, syntax
, bufp
)
1947 const char *pattern
;
1949 reg_syntax_t syntax
;
1950 struct re_pattern_buffer
*bufp
;
1952 /* We fetch characters from PATTERN here. Even though PATTERN is
1953 `char *' (i.e., signed), we declare these variables as unsigned, so
1954 they can be reliably used as array indices. */
1955 register unsigned char c
, c1
;
1957 /* A random temporary spot in PATTERN. */
1960 /* Points to the end of the buffer, where we should append. */
1961 register unsigned char *b
;
1963 /* Keeps track of unclosed groups. */
1964 compile_stack_type compile_stack
;
1966 /* Points to the current (ending) position in the pattern. */
1967 const char *p
= pattern
;
1968 const char *pend
= pattern
+ size
;
1970 /* How to translate the characters in the pattern. */
1971 RE_TRANSLATE_TYPE translate
= bufp
->translate
;
1973 /* Address of the count-byte of the most recently inserted `exactn'
1974 command. This makes it possible to tell if a new exact-match
1975 character can be added to that command or if the character requires
1976 a new `exactn' command. */
1977 unsigned char *pending_exact
= 0;
1979 /* Address of start of the most recently finished expression.
1980 This tells, e.g., postfix * where to find the start of its
1981 operand. Reset at the beginning of groups and alternatives. */
1982 unsigned char *laststart
= 0;
1984 /* Address of beginning of regexp, or inside of last group. */
1985 unsigned char *begalt
;
1987 /* Place in the uncompiled pattern (i.e., the {) to
1988 which to go back if the interval is invalid. */
1989 const char *beg_interval
;
1991 /* Address of the place where a forward jump should go to the end of
1992 the containing expression. Each alternative of an `or' -- except the
1993 last -- ends with a forward jump of this sort. */
1994 unsigned char *fixup_alt_jump
= 0;
1996 /* Counts open-groups as they are encountered. Remembered for the
1997 matching close-group on the compile stack, so the same register
1998 number is put in the stop_memory as the start_memory. */
1999 regnum_t regnum
= 0;
2002 DEBUG_PRINT1 ("\nCompiling pattern: ");
2005 unsigned debug_count
;
2007 for (debug_count
= 0; debug_count
< size
; debug_count
++)
2008 putchar (pattern
[debug_count
]);
2013 /* Initialize the compile stack. */
2014 compile_stack
.stack
= TALLOC (INIT_COMPILE_STACK_SIZE
, compile_stack_elt_t
);
2015 if (compile_stack
.stack
== NULL
)
2018 compile_stack
.size
= INIT_COMPILE_STACK_SIZE
;
2019 compile_stack
.avail
= 0;
2021 /* Initialize the pattern buffer. */
2022 bufp
->syntax
= syntax
;
2023 bufp
->fastmap_accurate
= 0;
2024 bufp
->not_bol
= bufp
->not_eol
= 0;
2026 /* Set `used' to zero, so that if we return an error, the pattern
2027 printer (for debugging) will think there's no pattern. We reset it
2031 /* Always count groups, whether or not bufp->no_sub is set. */
2034 #if !defined emacs && !defined SYNTAX_TABLE
2035 /* Initialize the syntax table. */
2036 init_syntax_once ();
2039 if (bufp
->allocated
== 0)
2042 { /* If zero allocated, but buffer is non-null, try to realloc
2043 enough space. This loses if buffer's address is bogus, but
2044 that is the user's responsibility. */
2045 RETALLOC (bufp
->buffer
, INIT_BUF_SIZE
, unsigned char);
2048 { /* Caller did not allocate a buffer. Do it for them. */
2049 bufp
->buffer
= TALLOC (INIT_BUF_SIZE
, unsigned char);
2051 if (!bufp
->buffer
) FREE_STACK_RETURN (REG_ESPACE
);
2053 bufp
->allocated
= INIT_BUF_SIZE
;
2056 begalt
= b
= bufp
->buffer
;
2058 /* Loop through the uncompiled pattern until we're at the end. */
2067 if ( /* If at start of pattern, it's an operator. */
2069 /* If context independent, it's an operator. */
2070 || syntax
& RE_CONTEXT_INDEP_ANCHORS
2071 /* Otherwise, depends on what's come before. */
2072 || at_begline_loc_p (pattern
, p
, syntax
))
2082 if ( /* If at end of pattern, it's an operator. */
2084 /* If context independent, it's an operator. */
2085 || syntax
& RE_CONTEXT_INDEP_ANCHORS
2086 /* Otherwise, depends on what's next. */
2087 || at_endline_loc_p (p
, pend
, syntax
))
2097 if ((syntax
& RE_BK_PLUS_QM
)
2098 || (syntax
& RE_LIMITED_OPS
))
2102 /* If there is no previous pattern... */
2105 if (syntax
& RE_CONTEXT_INVALID_OPS
)
2106 FREE_STACK_RETURN (REG_BADRPT
);
2107 else if (!(syntax
& RE_CONTEXT_INDEP_OPS
))
2112 /* Are we optimizing this jump? */
2113 boolean keep_string_p
= false;
2115 /* 1 means zero (many) matches is allowed. */
2116 char zero_times_ok
= 0, many_times_ok
= 0;
2118 /* If there is a sequence of repetition chars, collapse it
2119 down to just one (the right one). We can't combine
2120 interval operators with these because of, e.g., `a{2}*',
2121 which should only match an even number of `a's. */
2125 zero_times_ok
|= c
!= '+';
2126 many_times_ok
|= c
!= '?';
2134 || (!(syntax
& RE_BK_PLUS_QM
) && (c
== '+' || c
== '?')))
2137 else if (syntax
& RE_BK_PLUS_QM
&& c
== '\\')
2139 if (p
== pend
) FREE_STACK_RETURN (REG_EESCAPE
);
2142 if (!(c1
== '+' || c1
== '?'))
2157 /* If we get here, we found another repeat character. */
2160 /* Star, etc. applied to an empty pattern is equivalent
2161 to an empty pattern. */
2165 /* Now we know whether or not zero matches is allowed
2166 and also whether or not two or more matches is allowed. */
2168 { /* More than one repetition is allowed, so put in at the
2169 end a backward relative jump from `b' to before the next
2170 jump we're going to put in below (which jumps from
2171 laststart to after this jump).
2173 But if we are at the `*' in the exact sequence `.*\n',
2174 insert an unconditional jump backwards to the .,
2175 instead of the beginning of the loop. This way we only
2176 push a failure point once, instead of every time
2177 through the loop. */
2178 assert (p
- 1 > pattern
);
2180 /* Allocate the space for the jump. */
2181 GET_BUFFER_SPACE (3);
2183 /* We know we are not at the first character of the pattern,
2184 because laststart was nonzero. And we've already
2185 incremented `p', by the way, to be the character after
2186 the `*'. Do we have to do something analogous here
2187 for null bytes, because of RE_DOT_NOT_NULL? */
2188 if (TRANSLATE (*(p
- 2)) == TRANSLATE ('.')
2190 && p
< pend
&& TRANSLATE (*p
) == TRANSLATE ('\n')
2191 && !(syntax
& RE_DOT_NEWLINE
))
2192 { /* We have .*\n. */
2193 STORE_JUMP (jump
, b
, laststart
);
2194 keep_string_p
= true;
2197 /* Anything else. */
2198 STORE_JUMP (maybe_pop_jump
, b
, laststart
- 3);
2200 /* We've added more stuff to the buffer. */
2204 /* On failure, jump from laststart to b + 3, which will be the
2205 end of the buffer after this jump is inserted. */
2206 GET_BUFFER_SPACE (3);
2207 INSERT_JUMP (keep_string_p
? on_failure_keep_string_jump
2215 /* At least one repetition is required, so insert a
2216 `dummy_failure_jump' before the initial
2217 `on_failure_jump' instruction of the loop. This
2218 effects a skip over that instruction the first time
2219 we hit that loop. */
2220 GET_BUFFER_SPACE (3);
2221 INSERT_JUMP (dummy_failure_jump
, laststart
, laststart
+ 6);
2236 boolean had_char_class
= false;
2237 unsigned int range_start
= 0xffffffff;
2239 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2241 /* Ensure that we have enough space to push a charset: the
2242 opcode, the length count, and the bitset; 34 bytes in all. */
2243 GET_BUFFER_SPACE (34);
2247 /* We test `*p == '^' twice, instead of using an if
2248 statement, so we only need one BUF_PUSH. */
2249 BUF_PUSH (*p
== '^' ? charset_not
: charset
);
2253 /* Remember the first position in the bracket expression. */
2256 /* Push the number of bytes in the bitmap. */
2257 BUF_PUSH ((1 << BYTEWIDTH
) / BYTEWIDTH
);
2259 /* Clear the whole map. */
2260 bzero (b
, (1 << BYTEWIDTH
) / BYTEWIDTH
);
2262 /* charset_not matches newline according to a syntax bit. */
2263 if ((re_opcode_t
) b
[-2] == charset_not
2264 && (syntax
& RE_HAT_LISTS_NOT_NEWLINE
))
2265 SET_LIST_BIT ('\n');
2267 /* Read in characters and ranges, setting map bits. */
2270 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2274 /* \ might escape characters inside [...] and [^...]. */
2275 if ((syntax
& RE_BACKSLASH_ESCAPE_IN_LISTS
) && c
== '\\')
2277 if (p
== pend
) FREE_STACK_RETURN (REG_EESCAPE
);
2285 /* Could be the end of the bracket expression. If it's
2286 not (i.e., when the bracket expression is `[]' so
2287 far), the ']' character bit gets set way below. */
2288 if (c
== ']' && p
!= p1
+ 1)
2291 /* Look ahead to see if it's a range when the last thing
2292 was a character class. */
2293 if (had_char_class
&& c
== '-' && *p
!= ']')
2294 FREE_STACK_RETURN (REG_ERANGE
);
2296 /* Look ahead to see if it's a range when the last thing
2297 was a character: if this is a hyphen not at the
2298 beginning or the end of a list, then it's the range
2301 && !(p
- 2 >= pattern
&& p
[-2] == '[')
2302 && !(p
- 3 >= pattern
&& p
[-3] == '[' && p
[-2] == '^')
2306 = compile_range (range_start
, &p
, pend
, translate
,
2308 if (ret
!= REG_NOERROR
) FREE_STACK_RETURN (ret
);
2309 range_start
= 0xffffffff;
2312 else if (p
[0] == '-' && p
[1] != ']')
2313 { /* This handles ranges made up of characters only. */
2316 /* Move past the `-'. */
2319 ret
= compile_range (c
, &p
, pend
, translate
, syntax
, b
);
2320 if (ret
!= REG_NOERROR
) FREE_STACK_RETURN (ret
);
2321 range_start
= 0xffffffff;
2324 /* See if we're at the beginning of a possible character
2327 else if (syntax
& RE_CHAR_CLASSES
&& c
== '[' && *p
== ':')
2328 { /* Leave room for the null. */
2329 char str
[CHAR_CLASS_MAX_LENGTH
+ 1];
2334 /* If pattern is `[[:'. */
2335 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2340 if ((c
== ':' && *p
== ']') || p
== pend
)
2342 if (c1
< CHAR_CLASS_MAX_LENGTH
)
2345 /* This is in any case an invalid class name. */
2350 /* If isn't a word bracketed by `[:' and `:]':
2351 undo the ending character, the letters, and leave
2352 the leading `:' and `[' (but set bits for them). */
2353 if (c
== ':' && *p
== ']')
2355 #if defined _LIBC || WIDE_CHAR_SUPPORT
2356 boolean is_lower
= STREQ (str
, "lower");
2357 boolean is_upper
= STREQ (str
, "upper");
2361 wt
= IS_CHAR_CLASS (str
);
2363 FREE_STACK_RETURN (REG_ECTYPE
);
2365 /* Throw away the ] at the end of the character
2369 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2371 for (ch
= 0; ch
< 1 << BYTEWIDTH
; ++ch
)
2374 if (__iswctype (__btowc (ch
), wt
))
2377 if (iswctype (btowc (ch
), wt
))
2381 if (translate
&& (is_upper
|| is_lower
)
2382 && (ISUPPER (ch
) || ISLOWER (ch
)))
2386 had_char_class
= true;
2389 boolean is_alnum
= STREQ (str
, "alnum");
2390 boolean is_alpha
= STREQ (str
, "alpha");
2391 boolean is_blank
= STREQ (str
, "blank");
2392 boolean is_cntrl
= STREQ (str
, "cntrl");
2393 boolean is_digit
= STREQ (str
, "digit");
2394 boolean is_graph
= STREQ (str
, "graph");
2395 boolean is_lower
= STREQ (str
, "lower");
2396 boolean is_print
= STREQ (str
, "print");
2397 boolean is_punct
= STREQ (str
, "punct");
2398 boolean is_space
= STREQ (str
, "space");
2399 boolean is_upper
= STREQ (str
, "upper");
2400 boolean is_xdigit
= STREQ (str
, "xdigit");
2402 if (!IS_CHAR_CLASS (str
))
2403 FREE_STACK_RETURN (REG_ECTYPE
);
2405 /* Throw away the ] at the end of the character
2409 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2411 for (ch
= 0; ch
< 1 << BYTEWIDTH
; ch
++)
2413 /* This was split into 3 if's to
2414 avoid an arbitrary limit in some compiler. */
2415 if ( (is_alnum
&& ISALNUM (ch
))
2416 || (is_alpha
&& ISALPHA (ch
))
2417 || (is_blank
&& ISBLANK (ch
))
2418 || (is_cntrl
&& ISCNTRL (ch
)))
2420 if ( (is_digit
&& ISDIGIT (ch
))
2421 || (is_graph
&& ISGRAPH (ch
))
2422 || (is_lower
&& ISLOWER (ch
))
2423 || (is_print
&& ISPRINT (ch
)))
2425 if ( (is_punct
&& ISPUNCT (ch
))
2426 || (is_space
&& ISSPACE (ch
))
2427 || (is_upper
&& ISUPPER (ch
))
2428 || (is_xdigit
&& ISXDIGIT (ch
)))
2430 if ( translate
&& (is_upper
|| is_lower
)
2431 && (ISUPPER (ch
) || ISLOWER (ch
)))
2434 had_char_class
= true;
2435 #endif /* libc || wctype.h */
2445 had_char_class
= false;
2448 else if (syntax
& RE_CHAR_CLASSES
&& c
== '[' && *p
== '=')
2450 unsigned char str
[MB_LEN_MAX
+ 1];
2453 _NL_CURRENT_WORD (LC_COLLATE
, _NL_COLLATE_NRULES
);
2459 /* If pattern is `[[='. */
2460 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2465 if ((c
== '=' && *p
== ']') || p
== pend
)
2467 if (c1
< MB_LEN_MAX
)
2470 /* This is in any case an invalid class name. */
2475 if (c
== '=' && *p
== ']' && str
[0] != '\0')
2477 /* If we have no collation data we use the default
2478 collation in which each character is in a class
2479 by itself. It also means that ASCII is the
2480 character set and therefore we cannot have character
2481 with more than one byte in the multibyte
2488 FREE_STACK_RETURN (REG_ECOLLATE
);
2490 /* Throw away the ] at the end of the equivalence
2494 /* Set the bit for the character. */
2495 SET_LIST_BIT (str
[0]);
2500 /* Try to match the byte sequence in `str' against
2501 those known to the collate implementation.
2502 First find out whether the bytes in `str' are
2503 actually from exactly one character. */
2504 const int32_t *table
;
2505 const unsigned char *weights
;
2506 const unsigned char *extra
;
2507 const int32_t *indirect
;
2509 const unsigned char *cp
= str
;
2512 /* This #include defines a local function! */
2513 # include <locale/weight.h>
2515 table
= (const int32_t *)
2516 _NL_CURRENT (LC_COLLATE
, _NL_COLLATE_TABLEMB
);
2517 weights
= (const unsigned char *)
2518 _NL_CURRENT (LC_COLLATE
, _NL_COLLATE_WEIGHTMB
);
2519 extra
= (const unsigned char *)
2520 _NL_CURRENT (LC_COLLATE
, _NL_COLLATE_EXTRAMB
);
2521 indirect
= (const int32_t *)
2522 _NL_CURRENT (LC_COLLATE
, _NL_COLLATE_INDIRECTMB
);
2524 idx
= findidx (&cp
);
2525 if (idx
== 0 || cp
< str
+ c1
)
2526 /* This is no valid character. */
2527 FREE_STACK_RETURN (REG_ECOLLATE
);
2529 /* Throw away the ] at the end of the equivalence
2533 /* Now we have to go throught the whole table
2534 and find all characters which have the same
2537 XXX Note that this is not entirely correct.
2538 we would have to match multibyte sequences
2539 but this is not possible with the current
2541 for (ch
= 1; ch
< 256; ++ch
)
2542 /* XXX This test would have to be changed if we
2543 would allow matching multibyte sequences. */
2546 int32_t idx2
= table
[ch
];
2547 size_t len
= weights
[idx2
];
2549 /* Test whether the lenghts match. */
2550 if (weights
[idx
] == len
)
2552 /* They do. New compare the bytes of
2557 && (weights
[idx
+ 1 + cnt
]
2558 == weights
[idx2
+ 1 + cnt
]))
2562 /* They match. Mark the character as
2569 had_char_class
= true;
2579 had_char_class
= false;
2582 else if (syntax
& RE_CHAR_CLASSES
&& c
== '[' && *p
== '.')
2584 unsigned char str
[128]; /* Should be large enough. */
2587 _NL_CURRENT_WORD (LC_COLLATE
, _NL_COLLATE_NRULES
);
2593 /* If pattern is `[[='. */
2594 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2599 if ((c
== '.' && *p
== ']') || p
== pend
)
2601 if (c1
< sizeof (str
))
2604 /* This is in any case an invalid class name. */
2609 if (c
== '.' && *p
== ']' && str
[0] != '\0')
2611 /* If we have no collation data we use the default
2612 collation in which each character is the name
2613 for its own class which contains only the one
2614 character. It also means that ASCII is the
2615 character set and therefore we cannot have character
2616 with more than one byte in the multibyte
2623 FREE_STACK_RETURN (REG_ECOLLATE
);
2625 /* Throw away the ] at the end of the equivalence
2629 /* Set the bit for the character. */
2630 SET_LIST_BIT (str
[0]);
2631 range_start
= ((const unsigned char *) str
)[0];
2636 /* Try to match the byte sequence in `str' against
2637 those known to the collate implementation.
2638 First find out whether the bytes in `str' are
2639 actually from exactly one character. */
2641 const int32_t *symb_table
;
2642 const unsigned char *extra
;
2649 _NL_CURRENT_WORD (LC_COLLATE
,
2650 _NL_COLLATE_SYMB_HASH_SIZEMB
);
2651 symb_table
= (const int32_t *)
2652 _NL_CURRENT (LC_COLLATE
,
2653 _NL_COLLATE_SYMB_TABLEMB
);
2654 extra
= (const unsigned char *)
2655 _NL_CURRENT (LC_COLLATE
,
2656 _NL_COLLATE_SYMB_EXTRAMB
);
2658 /* Locate the character in the hashing table. */
2659 hash
= elem_hash (str
, c1
);
2662 elem
= hash
% table_size
;
2663 second
= hash
% (table_size
- 2);
2664 while (symb_table
[2 * elem
] != 0)
2666 /* First compare the hashing value. */
2667 if (symb_table
[2 * elem
] == hash
2668 && c1
== extra
[symb_table
[2 * elem
+ 1]]
2670 &extra
[symb_table
[2 * elem
+ 1]
2674 /* Yep, this is the entry. */
2675 idx
= symb_table
[2 * elem
+ 1];
2676 idx
+= 1 + extra
[idx
];
2684 if (symb_table
[2 * elem
] == 0)
2685 /* This is no valid character. */
2686 FREE_STACK_RETURN (REG_ECOLLATE
);
2688 /* Throw away the ] at the end of the equivalence
2692 /* Now add the multibyte character(s) we found
2693 to the acceptabed list.
2695 XXX Note that this is not entirely correct.
2696 we would have to match multibyte sequences
2697 but this is not possible with the current
2698 implementation. Also, we have to match
2699 collating symbols, which expand to more than
2700 one file, as a whole and not allow the
2701 individual bytes. */
2704 range_start
= extra
[idx
];
2706 SET_LIST_BIT (extra
[idx
++]);
2709 had_char_class
= false;
2719 had_char_class
= false;
2724 had_char_class
= false;
2730 /* Discard any (non)matching list bytes that are all 0 at the
2731 end of the map. Decrease the map-length byte too. */
2732 while ((int) b
[-1] > 0 && b
[b
[-1] - 1] == 0)
2740 if (syntax
& RE_NO_BK_PARENS
)
2747 if (syntax
& RE_NO_BK_PARENS
)
2754 if (syntax
& RE_NEWLINE_ALT
)
2761 if (syntax
& RE_NO_BK_VBAR
)
2768 if (syntax
& RE_INTERVALS
&& syntax
& RE_NO_BK_BRACES
)
2769 goto handle_interval
;
2775 if (p
== pend
) FREE_STACK_RETURN (REG_EESCAPE
);
2777 /* Do not translate the character after the \, so that we can
2778 distinguish, e.g., \B from \b, even if we normally would
2779 translate, e.g., B to b. */
2785 if (syntax
& RE_NO_BK_PARENS
)
2786 goto normal_backslash
;
2792 if (COMPILE_STACK_FULL
)
2794 RETALLOC (compile_stack
.stack
, compile_stack
.size
<< 1,
2795 compile_stack_elt_t
);
2796 if (compile_stack
.stack
== NULL
) return REG_ESPACE
;
2798 compile_stack
.size
<<= 1;
2801 /* These are the values to restore when we hit end of this
2802 group. They are all relative offsets, so that if the
2803 whole pattern moves because of realloc, they will still
2805 COMPILE_STACK_TOP
.begalt_offset
= begalt
- bufp
->buffer
;
2806 COMPILE_STACK_TOP
.fixup_alt_jump
2807 = fixup_alt_jump
? fixup_alt_jump
- bufp
->buffer
+ 1 : 0;
2808 COMPILE_STACK_TOP
.laststart_offset
= b
- bufp
->buffer
;
2809 COMPILE_STACK_TOP
.regnum
= regnum
;
2811 /* We will eventually replace the 0 with the number of
2812 groups inner to this one. But do not push a
2813 start_memory for groups beyond the last one we can
2814 represent in the compiled pattern. */
2815 if (regnum
<= MAX_REGNUM
)
2817 COMPILE_STACK_TOP
.inner_group_offset
= b
- bufp
->buffer
+ 2;
2818 BUF_PUSH_3 (start_memory
, regnum
, 0);
2821 compile_stack
.avail
++;
2826 /* If we've reached MAX_REGNUM groups, then this open
2827 won't actually generate any code, so we'll have to
2828 clear pending_exact explicitly. */
2834 if (syntax
& RE_NO_BK_PARENS
) goto normal_backslash
;
2836 if (COMPILE_STACK_EMPTY
)
2838 if (syntax
& RE_UNMATCHED_RIGHT_PAREN_ORD
)
2839 goto normal_backslash
;
2841 FREE_STACK_RETURN (REG_ERPAREN
);
2846 { /* Push a dummy failure point at the end of the
2847 alternative for a possible future
2848 `pop_failure_jump' to pop. See comments at
2849 `push_dummy_failure' in `re_match_2'. */
2850 BUF_PUSH (push_dummy_failure
);
2852 /* We allocated space for this jump when we assigned
2853 to `fixup_alt_jump', in the `handle_alt' case below. */
2854 STORE_JUMP (jump_past_alt
, fixup_alt_jump
, b
- 1);
2857 /* See similar code for backslashed left paren above. */
2858 if (COMPILE_STACK_EMPTY
)
2860 if (syntax
& RE_UNMATCHED_RIGHT_PAREN_ORD
)
2863 FREE_STACK_RETURN (REG_ERPAREN
);
2866 /* Since we just checked for an empty stack above, this
2867 ``can't happen''. */
2868 assert (compile_stack
.avail
!= 0);
2870 /* We don't just want to restore into `regnum', because
2871 later groups should continue to be numbered higher,
2872 as in `(ab)c(de)' -- the second group is #2. */
2873 regnum_t this_group_regnum
;
2875 compile_stack
.avail
--;
2876 begalt
= bufp
->buffer
+ COMPILE_STACK_TOP
.begalt_offset
;
2878 = COMPILE_STACK_TOP
.fixup_alt_jump
2879 ? bufp
->buffer
+ COMPILE_STACK_TOP
.fixup_alt_jump
- 1
2881 laststart
= bufp
->buffer
+ COMPILE_STACK_TOP
.laststart_offset
;
2882 this_group_regnum
= COMPILE_STACK_TOP
.regnum
;
2883 /* If we've reached MAX_REGNUM groups, then this open
2884 won't actually generate any code, so we'll have to
2885 clear pending_exact explicitly. */
2888 /* We're at the end of the group, so now we know how many
2889 groups were inside this one. */
2890 if (this_group_regnum
<= MAX_REGNUM
)
2892 unsigned char *inner_group_loc
2893 = bufp
->buffer
+ COMPILE_STACK_TOP
.inner_group_offset
;
2895 *inner_group_loc
= regnum
- this_group_regnum
;
2896 BUF_PUSH_3 (stop_memory
, this_group_regnum
,
2897 regnum
- this_group_regnum
);
2903 case '|': /* `\|'. */
2904 if (syntax
& RE_LIMITED_OPS
|| syntax
& RE_NO_BK_VBAR
)
2905 goto normal_backslash
;
2907 if (syntax
& RE_LIMITED_OPS
)
2910 /* Insert before the previous alternative a jump which
2911 jumps to this alternative if the former fails. */
2912 GET_BUFFER_SPACE (3);
2913 INSERT_JUMP (on_failure_jump
, begalt
, b
+ 6);
2917 /* The alternative before this one has a jump after it
2918 which gets executed if it gets matched. Adjust that
2919 jump so it will jump to this alternative's analogous
2920 jump (put in below, which in turn will jump to the next
2921 (if any) alternative's such jump, etc.). The last such
2922 jump jumps to the correct final destination. A picture:
2928 If we are at `b', then fixup_alt_jump right now points to a
2929 three-byte space after `a'. We'll put in the jump, set
2930 fixup_alt_jump to right after `b', and leave behind three
2931 bytes which we'll fill in when we get to after `c'. */
2934 STORE_JUMP (jump_past_alt
, fixup_alt_jump
, b
);
2936 /* Mark and leave space for a jump after this alternative,
2937 to be filled in later either by next alternative or
2938 when know we're at the end of a series of alternatives. */
2940 GET_BUFFER_SPACE (3);
2949 /* If \{ is a literal. */
2950 if (!(syntax
& RE_INTERVALS
)
2951 /* If we're at `\{' and it's not the open-interval
2953 || (syntax
& RE_NO_BK_BRACES
))
2954 goto normal_backslash
;
2958 /* If got here, then the syntax allows intervals. */
2960 /* At least (most) this many matches must be made. */
2961 int lower_bound
= -1, upper_bound
= -1;
2963 beg_interval
= p
- 1;
2967 if (!(syntax
& RE_INTERVALS
) && (syntax
& RE_NO_BK_BRACES
))
2968 goto unfetch_interval
;
2970 FREE_STACK_RETURN (REG_EBRACE
);
2973 GET_UNSIGNED_NUMBER (lower_bound
);
2977 GET_UNSIGNED_NUMBER (upper_bound
);
2978 if ((!(syntax
& RE_NO_BK_BRACES
) && c
!= '\\')
2979 || ((syntax
& RE_NO_BK_BRACES
) && c
!= '}'))
2980 FREE_STACK_RETURN (REG_BADBR
);
2982 if (upper_bound
< 0)
2983 upper_bound
= RE_DUP_MAX
;
2986 /* Interval such as `{1}' => match exactly once. */
2987 upper_bound
= lower_bound
;
2989 if (lower_bound
< 0 || upper_bound
> RE_DUP_MAX
2990 || lower_bound
> upper_bound
)
2992 if (!(syntax
& RE_INTERVALS
) && (syntax
& RE_NO_BK_BRACES
))
2993 goto unfetch_interval
;
2995 FREE_STACK_RETURN (REG_BADBR
);
2998 if (!(syntax
& RE_NO_BK_BRACES
))
3000 if (c
!= '\\') FREE_STACK_RETURN (REG_EBRACE
);
3007 if (!(syntax
& RE_INTERVALS
) && (syntax
& RE_NO_BK_BRACES
))
3008 goto unfetch_interval
;
3010 FREE_STACK_RETURN (REG_BADBR
);
3013 /* We just parsed a valid interval. */
3015 /* If it's invalid to have no preceding re. */
3018 if (syntax
& RE_CONTEXT_INVALID_OPS
)
3019 FREE_STACK_RETURN (REG_BADRPT
);
3020 else if (syntax
& RE_CONTEXT_INDEP_OPS
)
3023 goto unfetch_interval
;
3026 /* If the upper bound is zero, don't want to succeed at
3027 all; jump from `laststart' to `b + 3', which will be
3028 the end of the buffer after we insert the jump. */
3029 if (upper_bound
== 0)
3031 GET_BUFFER_SPACE (3);
3032 INSERT_JUMP (jump
, laststart
, b
+ 3);
3036 /* Otherwise, we have a nontrivial interval. When
3037 we're all done, the pattern will look like:
3038 set_number_at <jump count> <upper bound>
3039 set_number_at <succeed_n count> <lower bound>
3040 succeed_n <after jump addr> <succeed_n count>
3042 jump_n <succeed_n addr> <jump count>
3043 (The upper bound and `jump_n' are omitted if
3044 `upper_bound' is 1, though.) */
3046 { /* If the upper bound is > 1, we need to insert
3047 more at the end of the loop. */
3048 unsigned nbytes
= 10 + (upper_bound
> 1) * 10;
3050 GET_BUFFER_SPACE (nbytes
);
3052 /* Initialize lower bound of the `succeed_n', even
3053 though it will be set during matching by its
3054 attendant `set_number_at' (inserted next),
3055 because `re_compile_fastmap' needs to know.
3056 Jump to the `jump_n' we might insert below. */
3057 INSERT_JUMP2 (succeed_n
, laststart
,
3058 b
+ 5 + (upper_bound
> 1) * 5,
3062 /* Code to initialize the lower bound. Insert
3063 before the `succeed_n'. The `5' is the last two
3064 bytes of this `set_number_at', plus 3 bytes of
3065 the following `succeed_n'. */
3066 insert_op2 (set_number_at
, laststart
, 5, lower_bound
, b
);
3069 if (upper_bound
> 1)
3070 { /* More than one repetition is allowed, so
3071 append a backward jump to the `succeed_n'
3072 that starts this interval.
3074 When we've reached this during matching,
3075 we'll have matched the interval once, so
3076 jump back only `upper_bound - 1' times. */
3077 STORE_JUMP2 (jump_n
, b
, laststart
+ 5,
3081 /* The location we want to set is the second
3082 parameter of the `jump_n'; that is `b-2' as
3083 an absolute address. `laststart' will be
3084 the `set_number_at' we're about to insert;
3085 `laststart+3' the number to set, the source
3086 for the relative address. But we are
3087 inserting into the middle of the pattern --
3088 so everything is getting moved up by 5.
3089 Conclusion: (b - 2) - (laststart + 3) + 5,
3090 i.e., b - laststart.
3092 We insert this at the beginning of the loop
3093 so that if we fail during matching, we'll
3094 reinitialize the bounds. */
3095 insert_op2 (set_number_at
, laststart
, b
- laststart
,
3096 upper_bound
- 1, b
);
3101 beg_interval
= NULL
;
3106 /* If an invalid interval, match the characters as literals. */
3107 assert (beg_interval
);
3109 beg_interval
= NULL
;
3111 /* normal_char and normal_backslash need `c'. */
3114 if (!(syntax
& RE_NO_BK_BRACES
))
3116 if (p
> pattern
&& p
[-1] == '\\')
3117 goto normal_backslash
;
3122 /* There is no way to specify the before_dot and after_dot
3123 operators. rms says this is ok. --karl */
3131 BUF_PUSH_2 (syntaxspec
, syntax_spec_code
[c
]);
3137 BUF_PUSH_2 (notsyntaxspec
, syntax_spec_code
[c
]);
3143 if (syntax
& RE_NO_GNU_OPS
)
3146 BUF_PUSH (wordchar
);
3151 if (syntax
& RE_NO_GNU_OPS
)
3154 BUF_PUSH (notwordchar
);
3159 if (syntax
& RE_NO_GNU_OPS
)
3165 if (syntax
& RE_NO_GNU_OPS
)
3171 if (syntax
& RE_NO_GNU_OPS
)
3173 BUF_PUSH (wordbound
);
3177 if (syntax
& RE_NO_GNU_OPS
)
3179 BUF_PUSH (notwordbound
);
3183 if (syntax
& RE_NO_GNU_OPS
)
3189 if (syntax
& RE_NO_GNU_OPS
)
3194 case '1': case '2': case '3': case '4': case '5':
3195 case '6': case '7': case '8': case '9':
3196 if (syntax
& RE_NO_BK_REFS
)
3202 FREE_STACK_RETURN (REG_ESUBREG
);
3204 /* Can't back reference to a subexpression if inside of it. */
3205 if (group_in_compile_stack (compile_stack
, (regnum_t
) c1
))
3209 BUF_PUSH_2 (duplicate
, c1
);
3215 if (syntax
& RE_BK_PLUS_QM
)
3218 goto normal_backslash
;
3222 /* You might think it would be useful for \ to mean
3223 not to translate; but if we don't translate it
3224 it will never match anything. */
3232 /* Expects the character in `c'. */
3234 /* If no exactn currently being built. */
3237 /* If last exactn not at current position. */
3238 || pending_exact
+ *pending_exact
+ 1 != b
3240 /* We have only one byte following the exactn for the count. */
3241 || *pending_exact
== (1 << BYTEWIDTH
) - 1
3243 /* If followed by a repetition operator. */
3244 || *p
== '*' || *p
== '^'
3245 || ((syntax
& RE_BK_PLUS_QM
)
3246 ? *p
== '\\' && (p
[1] == '+' || p
[1] == '?')
3247 : (*p
== '+' || *p
== '?'))
3248 || ((syntax
& RE_INTERVALS
)
3249 && ((syntax
& RE_NO_BK_BRACES
)
3251 : (p
[0] == '\\' && p
[1] == '{'))))
3253 /* Start building a new exactn. */
3257 BUF_PUSH_2 (exactn
, 0);
3258 pending_exact
= b
- 1;
3265 } /* while p != pend */
3268 /* Through the pattern now. */
3271 STORE_JUMP (jump_past_alt
, fixup_alt_jump
, b
);
3273 if (!COMPILE_STACK_EMPTY
)
3274 FREE_STACK_RETURN (REG_EPAREN
);
3276 /* If we don't want backtracking, force success
3277 the first time we reach the end of the compiled pattern. */
3278 if (syntax
& RE_NO_POSIX_BACKTRACKING
)
3281 free (compile_stack
.stack
);
3283 /* We have succeeded; set the length of the buffer. */
3284 bufp
->used
= b
- bufp
->buffer
;
3289 DEBUG_PRINT1 ("\nCompiled pattern: \n");
3290 print_compiled_pattern (bufp
);
3294 #ifndef MATCH_MAY_ALLOCATE
3295 /* Initialize the failure stack to the largest possible stack. This
3296 isn't necessary unless we're trying to avoid calling alloca in
3297 the search and match routines. */
3299 int num_regs
= bufp
->re_nsub
+ 1;
3301 /* Since DOUBLE_FAIL_STACK refuses to double only if the current size
3302 is strictly greater than re_max_failures, the largest possible stack
3303 is 2 * re_max_failures failure points. */
3304 if (fail_stack
.size
< (2 * re_max_failures
* MAX_FAILURE_ITEMS
))
3306 fail_stack
.size
= (2 * re_max_failures
* MAX_FAILURE_ITEMS
);
3309 if (! fail_stack
.stack
)
3311 = (fail_stack_elt_t
*) xmalloc (fail_stack
.size
3312 * sizeof (fail_stack_elt_t
));
3315 = (fail_stack_elt_t
*) xrealloc (fail_stack
.stack
,
3317 * sizeof (fail_stack_elt_t
)));
3318 # else /* not emacs */
3319 if (! fail_stack
.stack
)
3321 = (fail_stack_elt_t
*) malloc (fail_stack
.size
3322 * sizeof (fail_stack_elt_t
));
3325 = (fail_stack_elt_t
*) realloc (fail_stack
.stack
,
3327 * sizeof (fail_stack_elt_t
)));
3328 # endif /* not emacs */
3331 regex_grow_registers (num_regs
);
3333 #endif /* not MATCH_MAY_ALLOCATE */
3336 } /* regex_compile */
3338 /* Subroutines for `regex_compile'. */
3340 /* Store OP at LOC followed by two-byte integer parameter ARG. */
3343 store_op1 (op
, loc
, arg
)
3348 *loc
= (unsigned char) op
;
3349 STORE_NUMBER (loc
+ 1, arg
);
3353 /* Like `store_op1', but for two two-byte parameters ARG1 and ARG2. */
3356 store_op2 (op
, loc
, arg1
, arg2
)
3361 *loc
= (unsigned char) op
;
3362 STORE_NUMBER (loc
+ 1, arg1
);
3363 STORE_NUMBER (loc
+ 3, arg2
);
3367 /* Copy the bytes from LOC to END to open up three bytes of space at LOC
3368 for OP followed by two-byte integer parameter ARG. */
3371 insert_op1 (op
, loc
, arg
, end
)
3377 register unsigned char *pfrom
= end
;
3378 register unsigned char *pto
= end
+ 3;
3380 while (pfrom
!= loc
)
3383 store_op1 (op
, loc
, arg
);
3387 /* Like `insert_op1', but for two two-byte parameters ARG1 and ARG2. */
3390 insert_op2 (op
, loc
, arg1
, arg2
, end
)
3396 register unsigned char *pfrom
= end
;
3397 register unsigned char *pto
= end
+ 5;
3399 while (pfrom
!= loc
)
3402 store_op2 (op
, loc
, arg1
, arg2
);
3406 /* P points to just after a ^ in PATTERN. Return true if that ^ comes
3407 after an alternative or a begin-subexpression. We assume there is at
3408 least one character before the ^. */
3411 at_begline_loc_p (pattern
, p
, syntax
)
3412 const char *pattern
, *p
;
3413 reg_syntax_t syntax
;
3415 const char *prev
= p
- 2;
3416 boolean prev_prev_backslash
= prev
> pattern
&& prev
[-1] == '\\';
3419 /* After a subexpression? */
3420 (*prev
== '(' && (syntax
& RE_NO_BK_PARENS
|| prev_prev_backslash
))
3421 /* After an alternative? */
3422 || (*prev
== '|' && (syntax
& RE_NO_BK_VBAR
|| prev_prev_backslash
));
3426 /* The dual of at_begline_loc_p. This one is for $. We assume there is
3427 at least one character after the $, i.e., `P < PEND'. */
3430 at_endline_loc_p (p
, pend
, syntax
)
3431 const char *p
, *pend
;
3432 reg_syntax_t syntax
;
3434 const char *next
= p
;
3435 boolean next_backslash
= *next
== '\\';
3436 const char *next_next
= p
+ 1 < pend
? p
+ 1 : 0;
3439 /* Before a subexpression? */
3440 (syntax
& RE_NO_BK_PARENS
? *next
== ')'
3441 : next_backslash
&& next_next
&& *next_next
== ')')
3442 /* Before an alternative? */
3443 || (syntax
& RE_NO_BK_VBAR
? *next
== '|'
3444 : next_backslash
&& next_next
&& *next_next
== '|');
3448 /* Returns true if REGNUM is in one of COMPILE_STACK's elements and
3449 false if it's not. */
3452 group_in_compile_stack (compile_stack
, regnum
)
3453 compile_stack_type compile_stack
;
3458 for (this_element
= compile_stack
.avail
- 1;
3461 if (compile_stack
.stack
[this_element
].regnum
== regnum
)
3468 /* Read the ending character of a range (in a bracket expression) from the
3469 uncompiled pattern *P_PTR (which ends at PEND). We assume the
3470 starting character is in `P[-2]'. (`P[-1]' is the character `-'.)
3471 Then we set the translation of all bits between the starting and
3472 ending characters (inclusive) in the compiled pattern B.
3474 Return an error code.
3476 We use these short variable names so we can use the same macros as
3477 `regex_compile' itself. */
3479 static reg_errcode_t
3480 compile_range (range_start_char
, p_ptr
, pend
, translate
, syntax
, b
)
3481 unsigned int range_start_char
;
3482 const char **p_ptr
, *pend
;
3483 RE_TRANSLATE_TYPE translate
;
3484 reg_syntax_t syntax
;
3489 const char *p
= *p_ptr
;
3491 char range_start
[2];
3498 /* Fetch the endpoints without translating them; the
3499 appropriate translation is done in the bit-setting loop below. */
3500 range_start
[0] = range_start_char
;
3501 range_start
[1] = '\0';
3502 range_end
[0] = p
[0];
3503 range_end
[1] = '\0';
3505 /* Have to increment the pointer into the pattern string, so the
3506 caller isn't still at the ending character. */
3509 /* Report an error if the range is empty and the syntax prohibits this. */
3510 ret
= syntax
& RE_NO_EMPTY_RANGES
? REG_ERANGE
: REG_NOERROR
;
3512 /* Here we see why `this_char' has to be larger than an `unsigned
3513 char' -- we would otherwise go into an infinite loop, since all
3514 characters <= 0xff. */
3516 for (this_char
= 0; this_char
<= (unsigned char) -1; ++this_char
)
3519 if (strcoll (range_start
, ch
) <= 0 && strcoll (ch
, range_end
) <= 0)
3521 SET_LIST_BIT (TRANSLATE (this_char
));
3529 /* re_compile_fastmap computes a ``fastmap'' for the compiled pattern in
3530 BUFP. A fastmap records which of the (1 << BYTEWIDTH) possible
3531 characters can start a string that matches the pattern. This fastmap
3532 is used by re_search to skip quickly over impossible starting points.
3534 The caller must supply the address of a (1 << BYTEWIDTH)-byte data
3535 area as BUFP->fastmap.
3537 We set the `fastmap', `fastmap_accurate', and `can_be_null' fields in
3540 Returns 0 if we succeed, -2 if an internal error. */
3543 re_compile_fastmap (bufp
)
3544 struct re_pattern_buffer
*bufp
;
3547 #ifdef MATCH_MAY_ALLOCATE
3548 fail_stack_type fail_stack
;
3550 #ifndef REGEX_MALLOC
3554 register char *fastmap
= bufp
->fastmap
;
3555 unsigned char *pattern
= bufp
->buffer
;
3556 unsigned char *p
= pattern
;
3557 register unsigned char *pend
= pattern
+ bufp
->used
;
3560 /* This holds the pointer to the failure stack, when
3561 it is allocated relocatably. */
3562 fail_stack_elt_t
*failure_stack_ptr
;
3565 /* Assume that each path through the pattern can be null until
3566 proven otherwise. We set this false at the bottom of switch
3567 statement, to which we get only if a particular path doesn't
3568 match the empty string. */
3569 boolean path_can_be_null
= true;
3571 /* We aren't doing a `succeed_n' to begin with. */
3572 boolean succeed_n_p
= false;
3574 assert (fastmap
!= NULL
&& p
!= NULL
);
3577 bzero (fastmap
, 1 << BYTEWIDTH
); /* Assume nothing's valid. */
3578 bufp
->fastmap_accurate
= 1; /* It will be when we're done. */
3579 bufp
->can_be_null
= 0;
3583 if (p
== pend
|| *p
== succeed
)
3585 /* We have reached the (effective) end of pattern. */
3586 if (!FAIL_STACK_EMPTY ())
3588 bufp
->can_be_null
|= path_can_be_null
;
3590 /* Reset for next path. */
3591 path_can_be_null
= true;
3593 p
= fail_stack
.stack
[--fail_stack
.avail
].pointer
;
3601 /* We should never be about to go beyond the end of the pattern. */
3604 switch (SWITCH_ENUM_CAST ((re_opcode_t
) *p
++))
3607 /* I guess the idea here is to simply not bother with a fastmap
3608 if a backreference is used, since it's too hard to figure out
3609 the fastmap for the corresponding group. Setting
3610 `can_be_null' stops `re_search_2' from using the fastmap, so
3611 that is all we do. */
3613 bufp
->can_be_null
= 1;
3617 /* Following are the cases which match a character. These end
3626 for (j
= *p
++ * BYTEWIDTH
- 1; j
>= 0; j
--)
3627 if (p
[j
/ BYTEWIDTH
] & (1 << (j
% BYTEWIDTH
)))
3633 /* Chars beyond end of map must be allowed. */
3634 for (j
= *p
* BYTEWIDTH
; j
< (1 << BYTEWIDTH
); j
++)
3637 for (j
= *p
++ * BYTEWIDTH
- 1; j
>= 0; j
--)
3638 if (!(p
[j
/ BYTEWIDTH
] & (1 << (j
% BYTEWIDTH
))))
3644 for (j
= 0; j
< (1 << BYTEWIDTH
); j
++)
3645 if (SYNTAX (j
) == Sword
)
3651 for (j
= 0; j
< (1 << BYTEWIDTH
); j
++)
3652 if (SYNTAX (j
) != Sword
)
3659 int fastmap_newline
= fastmap
['\n'];
3661 /* `.' matches anything ... */
3662 for (j
= 0; j
< (1 << BYTEWIDTH
); j
++)
3665 /* ... except perhaps newline. */
3666 if (!(bufp
->syntax
& RE_DOT_NEWLINE
))
3667 fastmap
['\n'] = fastmap_newline
;
3669 /* Return if we have already set `can_be_null'; if we have,
3670 then the fastmap is irrelevant. Something's wrong here. */
3671 else if (bufp
->can_be_null
)
3674 /* Otherwise, have to check alternative paths. */
3681 for (j
= 0; j
< (1 << BYTEWIDTH
); j
++)
3682 if (SYNTAX (j
) == (enum syntaxcode
) k
)
3689 for (j
= 0; j
< (1 << BYTEWIDTH
); j
++)
3690 if (SYNTAX (j
) != (enum syntaxcode
) k
)
3695 /* All cases after this match the empty string. These end with
3715 case push_dummy_failure
:
3720 case pop_failure_jump
:
3721 case maybe_pop_jump
:
3724 case dummy_failure_jump
:
3725 EXTRACT_NUMBER_AND_INCR (j
, p
);
3730 /* Jump backward implies we just went through the body of a
3731 loop and matched nothing. Opcode jumped to should be
3732 `on_failure_jump' or `succeed_n'. Just treat it like an
3733 ordinary jump. For a * loop, it has pushed its failure
3734 point already; if so, discard that as redundant. */
3735 if ((re_opcode_t
) *p
!= on_failure_jump
3736 && (re_opcode_t
) *p
!= succeed_n
)
3740 EXTRACT_NUMBER_AND_INCR (j
, p
);
3743 /* If what's on the stack is where we are now, pop it. */
3744 if (!FAIL_STACK_EMPTY ()
3745 && fail_stack
.stack
[fail_stack
.avail
- 1].pointer
== p
)
3751 case on_failure_jump
:
3752 case on_failure_keep_string_jump
:
3753 handle_on_failure_jump
:
3754 EXTRACT_NUMBER_AND_INCR (j
, p
);
3756 /* For some patterns, e.g., `(a?)?', `p+j' here points to the
3757 end of the pattern. We don't want to push such a point,
3758 since when we restore it above, entering the switch will
3759 increment `p' past the end of the pattern. We don't need
3760 to push such a point since we obviously won't find any more
3761 fastmap entries beyond `pend'. Such a pattern can match
3762 the null string, though. */
3765 if (!PUSH_PATTERN_OP (p
+ j
, fail_stack
))
3767 RESET_FAIL_STACK ();
3772 bufp
->can_be_null
= 1;
3776 EXTRACT_NUMBER_AND_INCR (k
, p
); /* Skip the n. */
3777 succeed_n_p
= false;
3784 /* Get to the number of times to succeed. */
3787 /* Increment p past the n for when k != 0. */
3788 EXTRACT_NUMBER_AND_INCR (k
, p
);
3792 succeed_n_p
= true; /* Spaghetti code alert. */
3793 goto handle_on_failure_jump
;
3810 abort (); /* We have listed all the cases. */
3813 /* Getting here means we have found the possible starting
3814 characters for one path of the pattern -- and that the empty
3815 string does not match. We need not follow this path further.
3816 Instead, look at the next alternative (remembered on the
3817 stack), or quit if no more. The test at the top of the loop
3818 does these things. */
3819 path_can_be_null
= false;
3823 /* Set `can_be_null' for the last path (also the first path, if the
3824 pattern is empty). */
3825 bufp
->can_be_null
|= path_can_be_null
;
3828 RESET_FAIL_STACK ();
3830 } /* re_compile_fastmap */
3832 weak_alias (__re_compile_fastmap
, re_compile_fastmap
)
3835 /* Set REGS to hold NUM_REGS registers, storing them in STARTS and
3836 ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use
3837 this memory for recording register information. STARTS and ENDS
3838 must be allocated using the malloc library routine, and must each
3839 be at least NUM_REGS * sizeof (regoff_t) bytes long.
3841 If NUM_REGS == 0, then subsequent matches should allocate their own
3844 Unless this function is called, the first search or match using
3845 PATTERN_BUFFER will allocate its own register data, without
3846 freeing the old data. */
3849 re_set_registers (bufp
, regs
, num_regs
, starts
, ends
)
3850 struct re_pattern_buffer
*bufp
;
3851 struct re_registers
*regs
;
3853 regoff_t
*starts
, *ends
;
3857 bufp
->regs_allocated
= REGS_REALLOCATE
;
3858 regs
->num_regs
= num_regs
;
3859 regs
->start
= starts
;
3864 bufp
->regs_allocated
= REGS_UNALLOCATED
;
3866 regs
->start
= regs
->end
= (regoff_t
*) 0;
3870 weak_alias (__re_set_registers
, re_set_registers
)
3873 /* Searching routines. */
3875 /* Like re_search_2, below, but only one string is specified, and
3876 doesn't let you say where to stop matching. */
3879 re_search (bufp
, string
, size
, startpos
, range
, regs
)
3880 struct re_pattern_buffer
*bufp
;
3882 int size
, startpos
, range
;
3883 struct re_registers
*regs
;
3885 return re_search_2 (bufp
, NULL
, 0, string
, size
, startpos
, range
,
3889 weak_alias (__re_search
, re_search
)
3893 /* Using the compiled pattern in BUFP->buffer, first tries to match the
3894 virtual concatenation of STRING1 and STRING2, starting first at index
3895 STARTPOS, then at STARTPOS + 1, and so on.
3897 STRING1 and STRING2 have length SIZE1 and SIZE2, respectively.
3899 RANGE is how far to scan while trying to match. RANGE = 0 means try
3900 only at STARTPOS; in general, the last start tried is STARTPOS +
3903 In REGS, return the indices of the virtual concatenation of STRING1
3904 and STRING2 that matched the entire BUFP->buffer and its contained
3907 Do not consider matching one past the index STOP in the virtual
3908 concatenation of STRING1 and STRING2.
3910 We return either the position in the strings at which the match was
3911 found, -1 if no match, or -2 if error (such as failure
3915 re_search_2 (bufp
, string1
, size1
, string2
, size2
, startpos
, range
, regs
, stop
)
3916 struct re_pattern_buffer
*bufp
;
3917 const char *string1
, *string2
;
3921 struct re_registers
*regs
;
3925 register char *fastmap
= bufp
->fastmap
;
3926 register RE_TRANSLATE_TYPE translate
= bufp
->translate
;
3927 int total_size
= size1
+ size2
;
3928 int endpos
= startpos
+ range
;
3930 /* Check for out-of-range STARTPOS. */
3931 if (startpos
< 0 || startpos
> total_size
)
3934 /* Fix up RANGE if it might eventually take us outside
3935 the virtual concatenation of STRING1 and STRING2.
3936 Make sure we won't move STARTPOS below 0 or above TOTAL_SIZE. */
3938 range
= 0 - startpos
;
3939 else if (endpos
> total_size
)
3940 range
= total_size
- startpos
;
3942 /* If the search isn't to be a backwards one, don't waste time in a
3943 search for a pattern that must be anchored. */
3944 if (bufp
->used
> 0 && range
> 0
3945 && ((re_opcode_t
) bufp
->buffer
[0] == begbuf
3946 /* `begline' is like `begbuf' if it cannot match at newlines. */
3947 || ((re_opcode_t
) bufp
->buffer
[0] == begline
3948 && !bufp
->newline_anchor
)))
3957 /* In a forward search for something that starts with \=.
3958 don't keep searching past point. */
3959 if (bufp
->used
> 0 && (re_opcode_t
) bufp
->buffer
[0] == at_dot
&& range
> 0)
3961 range
= PT
- startpos
;
3967 /* Update the fastmap now if not correct already. */
3968 if (fastmap
&& !bufp
->fastmap_accurate
)
3969 if (re_compile_fastmap (bufp
) == -2)
3972 /* Loop through the string, looking for a place to start matching. */
3975 /* If a fastmap is supplied, skip quickly over characters that
3976 cannot be the start of a match. If the pattern can match the
3977 null string, however, we don't need to skip characters; we want
3978 the first null string. */
3979 if (fastmap
&& startpos
< total_size
&& !bufp
->can_be_null
)
3981 if (range
> 0) /* Searching forwards. */
3983 register const char *d
;
3984 register int lim
= 0;
3987 if (startpos
< size1
&& startpos
+ range
>= size1
)
3988 lim
= range
- (size1
- startpos
);
3990 d
= (startpos
>= size1
? string2
- size1
: string1
) + startpos
;
3992 /* Written out as an if-else to avoid testing `translate'
3996 && !fastmap
[(unsigned char)
3997 translate
[(unsigned char) *d
++]])
4000 while (range
> lim
&& !fastmap
[(unsigned char) *d
++])
4003 startpos
+= irange
- range
;
4005 else /* Searching backwards. */
4007 register char c
= (size1
== 0 || startpos
>= size1
4008 ? string2
[startpos
- size1
]
4009 : string1
[startpos
]);
4011 if (!fastmap
[(unsigned char) TRANSLATE (c
)])
4016 /* If can't match the null string, and that's all we have left, fail. */
4017 if (range
>= 0 && startpos
== total_size
&& fastmap
4018 && !bufp
->can_be_null
)
4021 val
= re_match_2_internal (bufp
, string1
, size1
, string2
, size2
,
4022 startpos
, regs
, stop
);
4023 #ifndef REGEX_MALLOC
4052 weak_alias (__re_search_2
, re_search_2
)
4055 /* This converts PTR, a pointer into one of the search strings `string1'
4056 and `string2' into an offset from the beginning of that string. */
4057 #define POINTER_TO_OFFSET(ptr) \
4058 (FIRST_STRING_P (ptr) \
4059 ? ((regoff_t) ((ptr) - string1)) \
4060 : ((regoff_t) ((ptr) - string2 + size1)))
4062 /* Macros for dealing with the split strings in re_match_2. */
4064 #define MATCHING_IN_FIRST_STRING (dend == end_match_1)
4066 /* Call before fetching a character with *d. This switches over to
4067 string2 if necessary. */
4068 #define PREFETCH() \
4071 /* End of string2 => fail. */ \
4072 if (dend == end_match_2) \
4074 /* End of string1 => advance to string2. */ \
4076 dend = end_match_2; \
4080 /* Test if at very beginning or at very end of the virtual concatenation
4081 of `string1' and `string2'. If only one string, it's `string2'. */
4082 #define AT_STRINGS_BEG(d) ((d) == (size1 ? string1 : string2) || !size2)
4083 #define AT_STRINGS_END(d) ((d) == end2)
4086 /* Test if D points to a character which is word-constituent. We have
4087 two special cases to check for: if past the end of string1, look at
4088 the first character in string2; and if before the beginning of
4089 string2, look at the last character in string1. */
4090 #define WORDCHAR_P(d) \
4091 (SYNTAX ((d) == end1 ? *string2 \
4092 : (d) == string2 - 1 ? *(end1 - 1) : *(d)) \
4095 /* Disabled due to a compiler bug -- see comment at case wordbound */
4097 /* Test if the character before D and the one at D differ with respect
4098 to being word-constituent. */
4099 #define AT_WORD_BOUNDARY(d) \
4100 (AT_STRINGS_BEG (d) || AT_STRINGS_END (d) \
4101 || WORDCHAR_P (d - 1) != WORDCHAR_P (d))
4104 /* Free everything we malloc. */
4105 #ifdef MATCH_MAY_ALLOCATE
4106 # define FREE_VAR(var) if (var) REGEX_FREE (var); var = NULL
4107 # define FREE_VARIABLES() \
4109 REGEX_FREE_STACK (fail_stack.stack); \
4110 FREE_VAR (regstart); \
4111 FREE_VAR (regend); \
4112 FREE_VAR (old_regstart); \
4113 FREE_VAR (old_regend); \
4114 FREE_VAR (best_regstart); \
4115 FREE_VAR (best_regend); \
4116 FREE_VAR (reg_info); \
4117 FREE_VAR (reg_dummy); \
4118 FREE_VAR (reg_info_dummy); \
4121 # define FREE_VARIABLES() ((void)0) /* Do nothing! But inhibit gcc warning. */
4122 #endif /* not MATCH_MAY_ALLOCATE */
4124 /* These values must meet several constraints. They must not be valid
4125 register values; since we have a limit of 255 registers (because
4126 we use only one byte in the pattern for the register number), we can
4127 use numbers larger than 255. They must differ by 1, because of
4128 NUM_FAILURE_ITEMS above. And the value for the lowest register must
4129 be larger than the value for the highest register, so we do not try
4130 to actually save any registers when none are active. */
4131 #define NO_HIGHEST_ACTIVE_REG (1 << BYTEWIDTH)
4132 #define NO_LOWEST_ACTIVE_REG (NO_HIGHEST_ACTIVE_REG + 1)
4134 /* Matching routines. */
4136 #ifndef emacs /* Emacs never uses this. */
4137 /* re_match is like re_match_2 except it takes only a single string. */
4140 re_match (bufp
, string
, size
, pos
, regs
)
4141 struct re_pattern_buffer
*bufp
;
4144 struct re_registers
*regs
;
4146 int result
= re_match_2_internal (bufp
, NULL
, 0, string
, size
,
4148 # ifndef REGEX_MALLOC
4156 weak_alias (__re_match
, re_match
)
4158 #endif /* not emacs */
4160 static boolean group_match_null_string_p
_RE_ARGS ((unsigned char **p
,
4162 register_info_type
*reg_info
));
4163 static boolean alt_match_null_string_p
_RE_ARGS ((unsigned char *p
,
4165 register_info_type
*reg_info
));
4166 static boolean common_op_match_null_string_p
_RE_ARGS ((unsigned char **p
,
4168 register_info_type
*reg_info
));
4169 static int bcmp_translate
_RE_ARGS ((const char *s1
, const char *s2
,
4170 int len
, char *translate
));
4172 /* re_match_2 matches the compiled pattern in BUFP against the
4173 the (virtual) concatenation of STRING1 and STRING2 (of length SIZE1
4174 and SIZE2, respectively). We start matching at POS, and stop
4177 If REGS is non-null and the `no_sub' field of BUFP is nonzero, we
4178 store offsets for the substring each group matched in REGS. See the
4179 documentation for exactly how many groups we fill.
4181 We return -1 if no match, -2 if an internal error (such as the
4182 failure stack overflowing). Otherwise, we return the length of the
4183 matched substring. */
4186 re_match_2 (bufp
, string1
, size1
, string2
, size2
, pos
, regs
, stop
)
4187 struct re_pattern_buffer
*bufp
;
4188 const char *string1
, *string2
;
4191 struct re_registers
*regs
;
4194 int result
= re_match_2_internal (bufp
, string1
, size1
, string2
, size2
,
4196 #ifndef REGEX_MALLOC
4204 weak_alias (__re_match_2
, re_match_2
)
4207 /* This is a separate function so that we can force an alloca cleanup
4210 re_match_2_internal (bufp
, string1
, size1
, string2
, size2
, pos
, regs
, stop
)
4211 struct re_pattern_buffer
*bufp
;
4212 const char *string1
, *string2
;
4215 struct re_registers
*regs
;
4218 /* General temporaries. */
4222 /* Just past the end of the corresponding string. */
4223 const char *end1
, *end2
;
4225 /* Pointers into string1 and string2, just past the last characters in
4226 each to consider matching. */
4227 const char *end_match_1
, *end_match_2
;
4229 /* Where we are in the data, and the end of the current string. */
4230 const char *d
, *dend
;
4232 /* Where we are in the pattern, and the end of the pattern. */
4233 unsigned char *p
= bufp
->buffer
;
4234 register unsigned char *pend
= p
+ bufp
->used
;
4236 /* Mark the opcode just after a start_memory, so we can test for an
4237 empty subpattern when we get to the stop_memory. */
4238 unsigned char *just_past_start_mem
= 0;
4240 /* We use this to map every character in the string. */
4241 RE_TRANSLATE_TYPE translate
= bufp
->translate
;
4243 /* Failure point stack. Each place that can handle a failure further
4244 down the line pushes a failure point on this stack. It consists of
4245 restart, regend, and reg_info for all registers corresponding to
4246 the subexpressions we're currently inside, plus the number of such
4247 registers, and, finally, two char *'s. The first char * is where
4248 to resume scanning the pattern; the second one is where to resume
4249 scanning the strings. If the latter is zero, the failure point is
4250 a ``dummy''; if a failure happens and the failure point is a dummy,
4251 it gets discarded and the next next one is tried. */
4252 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
4253 fail_stack_type fail_stack
;
4256 static unsigned failure_id
;
4257 unsigned nfailure_points_pushed
= 0, nfailure_points_popped
= 0;
4261 /* This holds the pointer to the failure stack, when
4262 it is allocated relocatably. */
4263 fail_stack_elt_t
*failure_stack_ptr
;
4266 /* We fill all the registers internally, independent of what we
4267 return, for use in backreferences. The number here includes
4268 an element for register zero. */
4269 size_t num_regs
= bufp
->re_nsub
+ 1;
4271 /* The currently active registers. */
4272 active_reg_t lowest_active_reg
= NO_LOWEST_ACTIVE_REG
;
4273 active_reg_t highest_active_reg
= NO_HIGHEST_ACTIVE_REG
;
4275 /* Information on the contents of registers. These are pointers into
4276 the input strings; they record just what was matched (on this
4277 attempt) by a subexpression part of the pattern, that is, the
4278 regnum-th regstart pointer points to where in the pattern we began
4279 matching and the regnum-th regend points to right after where we
4280 stopped matching the regnum-th subexpression. (The zeroth register
4281 keeps track of what the whole pattern matches.) */
4282 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
4283 const char **regstart
, **regend
;
4286 /* If a group that's operated upon by a repetition operator fails to
4287 match anything, then the register for its start will need to be
4288 restored because it will have been set to wherever in the string we
4289 are when we last see its open-group operator. Similarly for a
4291 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
4292 const char **old_regstart
, **old_regend
;
4295 /* The is_active field of reg_info helps us keep track of which (possibly
4296 nested) subexpressions we are currently in. The matched_something
4297 field of reg_info[reg_num] helps us tell whether or not we have
4298 matched any of the pattern so far this time through the reg_num-th
4299 subexpression. These two fields get reset each time through any
4300 loop their register is in. */
4301 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
4302 register_info_type
*reg_info
;
4305 /* The following record the register info as found in the above
4306 variables when we find a match better than any we've seen before.
4307 This happens as we backtrack through the failure points, which in
4308 turn happens only if we have not yet matched the entire string. */
4309 unsigned best_regs_set
= false;
4310 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
4311 const char **best_regstart
, **best_regend
;
4314 /* Logically, this is `best_regend[0]'. But we don't want to have to
4315 allocate space for that if we're not allocating space for anything
4316 else (see below). Also, we never need info about register 0 for
4317 any of the other register vectors, and it seems rather a kludge to
4318 treat `best_regend' differently than the rest. So we keep track of
4319 the end of the best match so far in a separate variable. We
4320 initialize this to NULL so that when we backtrack the first time
4321 and need to test it, it's not garbage. */
4322 const char *match_end
= NULL
;
4324 /* This helps SET_REGS_MATCHED avoid doing redundant work. */
4325 int set_regs_matched_done
= 0;
4327 /* Used when we pop values we don't care about. */
4328 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
4329 const char **reg_dummy
;
4330 register_info_type
*reg_info_dummy
;
4334 /* Counts the total number of registers pushed. */
4335 unsigned num_regs_pushed
= 0;
4338 DEBUG_PRINT1 ("\n\nEntering re_match_2.\n");
4342 #ifdef MATCH_MAY_ALLOCATE
4343 /* Do not bother to initialize all the register variables if there are
4344 no groups in the pattern, as it takes a fair amount of time. If
4345 there are groups, we include space for register 0 (the whole
4346 pattern), even though we never use it, since it simplifies the
4347 array indexing. We should fix this. */
4350 regstart
= REGEX_TALLOC (num_regs
, const char *);
4351 regend
= REGEX_TALLOC (num_regs
, const char *);
4352 old_regstart
= REGEX_TALLOC (num_regs
, const char *);
4353 old_regend
= REGEX_TALLOC (num_regs
, const char *);
4354 best_regstart
= REGEX_TALLOC (num_regs
, const char *);
4355 best_regend
= REGEX_TALLOC (num_regs
, const char *);
4356 reg_info
= REGEX_TALLOC (num_regs
, register_info_type
);
4357 reg_dummy
= REGEX_TALLOC (num_regs
, const char *);
4358 reg_info_dummy
= REGEX_TALLOC (num_regs
, register_info_type
);
4360 if (!(regstart
&& regend
&& old_regstart
&& old_regend
&& reg_info
4361 && best_regstart
&& best_regend
&& reg_dummy
&& reg_info_dummy
))
4369 /* We must initialize all our variables to NULL, so that
4370 `FREE_VARIABLES' doesn't try to free them. */
4371 regstart
= regend
= old_regstart
= old_regend
= best_regstart
4372 = best_regend
= reg_dummy
= NULL
;
4373 reg_info
= reg_info_dummy
= (register_info_type
*) NULL
;
4375 #endif /* MATCH_MAY_ALLOCATE */
4377 /* The starting position is bogus. */
4378 if (pos
< 0 || pos
> size1
+ size2
)
4384 /* Initialize subexpression text positions to -1 to mark ones that no
4385 start_memory/stop_memory has been seen for. Also initialize the
4386 register information struct. */
4387 for (mcnt
= 1; (unsigned) mcnt
< num_regs
; mcnt
++)
4389 regstart
[mcnt
] = regend
[mcnt
]
4390 = old_regstart
[mcnt
] = old_regend
[mcnt
] = REG_UNSET_VALUE
;
4392 REG_MATCH_NULL_STRING_P (reg_info
[mcnt
]) = MATCH_NULL_UNSET_VALUE
;
4393 IS_ACTIVE (reg_info
[mcnt
]) = 0;
4394 MATCHED_SOMETHING (reg_info
[mcnt
]) = 0;
4395 EVER_MATCHED_SOMETHING (reg_info
[mcnt
]) = 0;
4398 /* We move `string1' into `string2' if the latter's empty -- but not if
4399 `string1' is null. */
4400 if (size2
== 0 && string1
!= NULL
)
4407 end1
= string1
+ size1
;
4408 end2
= string2
+ size2
;
4410 /* Compute where to stop matching, within the two strings. */
4413 end_match_1
= string1
+ stop
;
4414 end_match_2
= string2
;
4419 end_match_2
= string2
+ stop
- size1
;
4422 /* `p' scans through the pattern as `d' scans through the data.
4423 `dend' is the end of the input string that `d' points within. `d'
4424 is advanced into the following input string whenever necessary, but
4425 this happens before fetching; therefore, at the beginning of the
4426 loop, `d' can be pointing at the end of a string, but it cannot
4428 if (size1
> 0 && pos
<= size1
)
4435 d
= string2
+ pos
- size1
;
4439 DEBUG_PRINT1 ("The compiled pattern is:\n");
4440 DEBUG_PRINT_COMPILED_PATTERN (bufp
, p
, pend
);
4441 DEBUG_PRINT1 ("The string to match is: `");
4442 DEBUG_PRINT_DOUBLE_STRING (d
, string1
, size1
, string2
, size2
);
4443 DEBUG_PRINT1 ("'\n");
4445 /* This loops over pattern commands. It exits by returning from the
4446 function if the match is complete, or it drops through if the match
4447 fails at this starting point in the input data. */
4451 DEBUG_PRINT2 ("\n%p: ", p
);
4453 DEBUG_PRINT2 ("\n0x%x: ", p
);
4457 { /* End of pattern means we might have succeeded. */
4458 DEBUG_PRINT1 ("end of pattern ... ");
4460 /* If we haven't matched the entire string, and we want the
4461 longest match, try backtracking. */
4462 if (d
!= end_match_2
)
4464 /* 1 if this match ends in the same string (string1 or string2)
4465 as the best previous match. */
4466 boolean same_str_p
= (FIRST_STRING_P (match_end
)
4467 == MATCHING_IN_FIRST_STRING
);
4468 /* 1 if this match is the best seen so far. */
4469 boolean best_match_p
;
4471 /* AIX compiler got confused when this was combined
4472 with the previous declaration. */
4474 best_match_p
= d
> match_end
;
4476 best_match_p
= !MATCHING_IN_FIRST_STRING
;
4478 DEBUG_PRINT1 ("backtracking.\n");
4480 if (!FAIL_STACK_EMPTY ())
4481 { /* More failure points to try. */
4483 /* If exceeds best match so far, save it. */
4484 if (!best_regs_set
|| best_match_p
)
4486 best_regs_set
= true;
4489 DEBUG_PRINT1 ("\nSAVING match as best so far.\n");
4491 for (mcnt
= 1; (unsigned) mcnt
< num_regs
; mcnt
++)
4493 best_regstart
[mcnt
] = regstart
[mcnt
];
4494 best_regend
[mcnt
] = regend
[mcnt
];
4500 /* If no failure points, don't restore garbage. And if
4501 last match is real best match, don't restore second
4503 else if (best_regs_set
&& !best_match_p
)
4506 /* Restore best match. It may happen that `dend ==
4507 end_match_1' while the restored d is in string2.
4508 For example, the pattern `x.*y.*z' against the
4509 strings `x-' and `y-z-', if the two strings are
4510 not consecutive in memory. */
4511 DEBUG_PRINT1 ("Restoring best registers.\n");
4514 dend
= ((d
>= string1
&& d
<= end1
)
4515 ? end_match_1
: end_match_2
);
4517 for (mcnt
= 1; (unsigned) mcnt
< num_regs
; mcnt
++)
4519 regstart
[mcnt
] = best_regstart
[mcnt
];
4520 regend
[mcnt
] = best_regend
[mcnt
];
4523 } /* d != end_match_2 */
4526 DEBUG_PRINT1 ("Accepting match.\n");
4528 /* If caller wants register contents data back, do it. */
4529 if (regs
&& !bufp
->no_sub
)
4531 /* Have the register data arrays been allocated? */
4532 if (bufp
->regs_allocated
== REGS_UNALLOCATED
)
4533 { /* No. So allocate them with malloc. We need one
4534 extra element beyond `num_regs' for the `-1' marker
4536 regs
->num_regs
= MAX (RE_NREGS
, num_regs
+ 1);
4537 regs
->start
= TALLOC (regs
->num_regs
, regoff_t
);
4538 regs
->end
= TALLOC (regs
->num_regs
, regoff_t
);
4539 if (regs
->start
== NULL
|| regs
->end
== NULL
)
4544 bufp
->regs_allocated
= REGS_REALLOCATE
;
4546 else if (bufp
->regs_allocated
== REGS_REALLOCATE
)
4547 { /* Yes. If we need more elements than were already
4548 allocated, reallocate them. If we need fewer, just
4550 if (regs
->num_regs
< num_regs
+ 1)
4552 regs
->num_regs
= num_regs
+ 1;
4553 RETALLOC (regs
->start
, regs
->num_regs
, regoff_t
);
4554 RETALLOC (regs
->end
, regs
->num_regs
, regoff_t
);
4555 if (regs
->start
== NULL
|| regs
->end
== NULL
)
4564 /* These braces fend off a "empty body in an else-statement"
4565 warning under GCC when assert expands to nothing. */
4566 assert (bufp
->regs_allocated
== REGS_FIXED
);
4569 /* Convert the pointer data in `regstart' and `regend' to
4570 indices. Register zero has to be set differently,
4571 since we haven't kept track of any info for it. */
4572 if (regs
->num_regs
> 0)
4574 regs
->start
[0] = pos
;
4575 regs
->end
[0] = (MATCHING_IN_FIRST_STRING
4576 ? ((regoff_t
) (d
- string1
))
4577 : ((regoff_t
) (d
- string2
+ size1
)));
4580 /* Go through the first `min (num_regs, regs->num_regs)'
4581 registers, since that is all we initialized. */
4582 for (mcnt
= 1; (unsigned) mcnt
< MIN (num_regs
, regs
->num_regs
);
4585 if (REG_UNSET (regstart
[mcnt
]) || REG_UNSET (regend
[mcnt
]))
4586 regs
->start
[mcnt
] = regs
->end
[mcnt
] = -1;
4590 = (regoff_t
) POINTER_TO_OFFSET (regstart
[mcnt
]);
4592 = (regoff_t
) POINTER_TO_OFFSET (regend
[mcnt
]);
4596 /* If the regs structure we return has more elements than
4597 were in the pattern, set the extra elements to -1. If
4598 we (re)allocated the registers, this is the case,
4599 because we always allocate enough to have at least one
4601 for (mcnt
= num_regs
; (unsigned) mcnt
< regs
->num_regs
; mcnt
++)
4602 regs
->start
[mcnt
] = regs
->end
[mcnt
] = -1;
4603 } /* regs && !bufp->no_sub */
4605 DEBUG_PRINT4 ("%u failure points pushed, %u popped (%u remain).\n",
4606 nfailure_points_pushed
, nfailure_points_popped
,
4607 nfailure_points_pushed
- nfailure_points_popped
);
4608 DEBUG_PRINT2 ("%u registers pushed.\n", num_regs_pushed
);
4610 mcnt
= d
- pos
- (MATCHING_IN_FIRST_STRING
4614 DEBUG_PRINT2 ("Returning %d from re_match_2.\n", mcnt
);
4620 /* Otherwise match next pattern command. */
4621 switch (SWITCH_ENUM_CAST ((re_opcode_t
) *p
++))
4623 /* Ignore these. Used to ignore the n of succeed_n's which
4624 currently have n == 0. */
4626 DEBUG_PRINT1 ("EXECUTING no_op.\n");
4630 DEBUG_PRINT1 ("EXECUTING succeed.\n");
4633 /* Match the next n pattern characters exactly. The following
4634 byte in the pattern defines n, and the n bytes after that
4635 are the characters to match. */
4638 DEBUG_PRINT2 ("EXECUTING exactn %d.\n", mcnt
);
4640 /* This is written out as an if-else so we don't waste time
4641 testing `translate' inside the loop. */
4647 if ((unsigned char) translate
[(unsigned char) *d
++]
4648 != (unsigned char) *p
++)
4658 if (*d
++ != (char) *p
++) goto fail
;
4662 SET_REGS_MATCHED ();
4666 /* Match any character except possibly a newline or a null. */
4668 DEBUG_PRINT1 ("EXECUTING anychar.\n");
4672 if ((!(bufp
->syntax
& RE_DOT_NEWLINE
) && TRANSLATE (*d
) == '\n')
4673 || (bufp
->syntax
& RE_DOT_NOT_NULL
&& TRANSLATE (*d
) == '\000'))
4676 SET_REGS_MATCHED ();
4677 DEBUG_PRINT2 (" Matched `%d'.\n", *d
);
4685 register unsigned char c
;
4686 boolean
not = (re_opcode_t
) *(p
- 1) == charset_not
;
4688 DEBUG_PRINT2 ("EXECUTING charset%s.\n", not ? "_not" : "");
4691 c
= TRANSLATE (*d
); /* The character to match. */
4693 /* Cast to `unsigned' instead of `unsigned char' in case the
4694 bit list is a full 32 bytes long. */
4695 if (c
< (unsigned) (*p
* BYTEWIDTH
)
4696 && p
[1 + c
/ BYTEWIDTH
] & (1 << (c
% BYTEWIDTH
)))
4701 if (!not) goto fail
;
4703 SET_REGS_MATCHED ();
4709 /* The beginning of a group is represented by start_memory.
4710 The arguments are the register number in the next byte, and the
4711 number of groups inner to this one in the next. The text
4712 matched within the group is recorded (in the internal
4713 registers data structure) under the register number. */
4715 DEBUG_PRINT3 ("EXECUTING start_memory %d (%d):\n", *p
, p
[1]);
4717 /* Find out if this group can match the empty string. */
4718 p1
= p
; /* To send to group_match_null_string_p. */
4720 if (REG_MATCH_NULL_STRING_P (reg_info
[*p
]) == MATCH_NULL_UNSET_VALUE
)
4721 REG_MATCH_NULL_STRING_P (reg_info
[*p
])
4722 = group_match_null_string_p (&p1
, pend
, reg_info
);
4724 /* Save the position in the string where we were the last time
4725 we were at this open-group operator in case the group is
4726 operated upon by a repetition operator, e.g., with `(a*)*b'
4727 against `ab'; then we want to ignore where we are now in
4728 the string in case this attempt to match fails. */
4729 old_regstart
[*p
] = REG_MATCH_NULL_STRING_P (reg_info
[*p
])
4730 ? REG_UNSET (regstart
[*p
]) ? d
: regstart
[*p
]
4732 DEBUG_PRINT2 (" old_regstart: %d\n",
4733 POINTER_TO_OFFSET (old_regstart
[*p
]));
4736 DEBUG_PRINT2 (" regstart: %d\n", POINTER_TO_OFFSET (regstart
[*p
]));
4738 IS_ACTIVE (reg_info
[*p
]) = 1;
4739 MATCHED_SOMETHING (reg_info
[*p
]) = 0;
4741 /* Clear this whenever we change the register activity status. */
4742 set_regs_matched_done
= 0;
4744 /* This is the new highest active register. */
4745 highest_active_reg
= *p
;
4747 /* If nothing was active before, this is the new lowest active
4749 if (lowest_active_reg
== NO_LOWEST_ACTIVE_REG
)
4750 lowest_active_reg
= *p
;
4752 /* Move past the register number and inner group count. */
4754 just_past_start_mem
= p
;
4759 /* The stop_memory opcode represents the end of a group. Its
4760 arguments are the same as start_memory's: the register
4761 number, and the number of inner groups. */
4763 DEBUG_PRINT3 ("EXECUTING stop_memory %d (%d):\n", *p
, p
[1]);
4765 /* We need to save the string position the last time we were at
4766 this close-group operator in case the group is operated
4767 upon by a repetition operator, e.g., with `((a*)*(b*)*)*'
4768 against `aba'; then we want to ignore where we are now in
4769 the string in case this attempt to match fails. */
4770 old_regend
[*p
] = REG_MATCH_NULL_STRING_P (reg_info
[*p
])
4771 ? REG_UNSET (regend
[*p
]) ? d
: regend
[*p
]
4773 DEBUG_PRINT2 (" old_regend: %d\n",
4774 POINTER_TO_OFFSET (old_regend
[*p
]));
4777 DEBUG_PRINT2 (" regend: %d\n", POINTER_TO_OFFSET (regend
[*p
]));
4779 /* This register isn't active anymore. */
4780 IS_ACTIVE (reg_info
[*p
]) = 0;
4782 /* Clear this whenever we change the register activity status. */
4783 set_regs_matched_done
= 0;
4785 /* If this was the only register active, nothing is active
4787 if (lowest_active_reg
== highest_active_reg
)
4789 lowest_active_reg
= NO_LOWEST_ACTIVE_REG
;
4790 highest_active_reg
= NO_HIGHEST_ACTIVE_REG
;
4793 { /* We must scan for the new highest active register, since
4794 it isn't necessarily one less than now: consider
4795 (a(b)c(d(e)f)g). When group 3 ends, after the f), the
4796 new highest active register is 1. */
4797 unsigned char r
= *p
- 1;
4798 while (r
> 0 && !IS_ACTIVE (reg_info
[r
]))
4801 /* If we end up at register zero, that means that we saved
4802 the registers as the result of an `on_failure_jump', not
4803 a `start_memory', and we jumped to past the innermost
4804 `stop_memory'. For example, in ((.)*) we save
4805 registers 1 and 2 as a result of the *, but when we pop
4806 back to the second ), we are at the stop_memory 1.
4807 Thus, nothing is active. */
4810 lowest_active_reg
= NO_LOWEST_ACTIVE_REG
;
4811 highest_active_reg
= NO_HIGHEST_ACTIVE_REG
;
4814 highest_active_reg
= r
;
4817 /* If just failed to match something this time around with a
4818 group that's operated on by a repetition operator, try to
4819 force exit from the ``loop'', and restore the register
4820 information for this group that we had before trying this
4822 if ((!MATCHED_SOMETHING (reg_info
[*p
])
4823 || just_past_start_mem
== p
- 1)
4826 boolean is_a_jump_n
= false;
4830 switch ((re_opcode_t
) *p1
++)
4834 case pop_failure_jump
:
4835 case maybe_pop_jump
:
4837 case dummy_failure_jump
:
4838 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
4848 /* If the next operation is a jump backwards in the pattern
4849 to an on_failure_jump right before the start_memory
4850 corresponding to this stop_memory, exit from the loop
4851 by forcing a failure after pushing on the stack the
4852 on_failure_jump's jump in the pattern, and d. */
4853 if (mcnt
< 0 && (re_opcode_t
) *p1
== on_failure_jump
4854 && (re_opcode_t
) p1
[3] == start_memory
&& p1
[4] == *p
)
4856 /* If this group ever matched anything, then restore
4857 what its registers were before trying this last
4858 failed match, e.g., with `(a*)*b' against `ab' for
4859 regstart[1], and, e.g., with `((a*)*(b*)*)*'
4860 against `aba' for regend[3].
4862 Also restore the registers for inner groups for,
4863 e.g., `((a*)(b*))*' against `aba' (register 3 would
4864 otherwise get trashed). */
4866 if (EVER_MATCHED_SOMETHING (reg_info
[*p
]))
4870 EVER_MATCHED_SOMETHING (reg_info
[*p
]) = 0;
4872 /* Restore this and inner groups' (if any) registers. */
4873 for (r
= *p
; r
< (unsigned) *p
+ (unsigned) *(p
+ 1);
4876 regstart
[r
] = old_regstart
[r
];
4878 /* xx why this test? */
4879 if (old_regend
[r
] >= regstart
[r
])
4880 regend
[r
] = old_regend
[r
];
4884 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
4885 PUSH_FAILURE_POINT (p1
+ mcnt
, d
, -2);
4891 /* Move past the register number and the inner group count. */
4896 /* \<digit> has been turned into a `duplicate' command which is
4897 followed by the numeric value of <digit> as the register number. */
4900 register const char *d2
, *dend2
;
4901 int regno
= *p
++; /* Get which register to match against. */
4902 DEBUG_PRINT2 ("EXECUTING duplicate %d.\n", regno
);
4904 /* Can't back reference a group which we've never matched. */
4905 if (REG_UNSET (regstart
[regno
]) || REG_UNSET (regend
[regno
]))
4908 /* Where in input to try to start matching. */
4909 d2
= regstart
[regno
];
4911 /* Where to stop matching; if both the place to start and
4912 the place to stop matching are in the same string, then
4913 set to the place to stop, otherwise, for now have to use
4914 the end of the first string. */
4916 dend2
= ((FIRST_STRING_P (regstart
[regno
])
4917 == FIRST_STRING_P (regend
[regno
]))
4918 ? regend
[regno
] : end_match_1
);
4921 /* If necessary, advance to next segment in register
4925 if (dend2
== end_match_2
) break;
4926 if (dend2
== regend
[regno
]) break;
4928 /* End of string1 => advance to string2. */
4930 dend2
= regend
[regno
];
4932 /* At end of register contents => success */
4933 if (d2
== dend2
) break;
4935 /* If necessary, advance to next segment in data. */
4938 /* How many characters left in this segment to match. */
4941 /* Want how many consecutive characters we can match in
4942 one shot, so, if necessary, adjust the count. */
4943 if (mcnt
> dend2
- d2
)
4946 /* Compare that many; failure if mismatch, else move
4949 ? bcmp_translate (d
, d2
, mcnt
, translate
)
4950 : memcmp (d
, d2
, mcnt
))
4952 d
+= mcnt
, d2
+= mcnt
;
4954 /* Do this because we've match some characters. */
4955 SET_REGS_MATCHED ();
4961 /* begline matches the empty string at the beginning of the string
4962 (unless `not_bol' is set in `bufp'), and, if
4963 `newline_anchor' is set, after newlines. */
4965 DEBUG_PRINT1 ("EXECUTING begline.\n");
4967 if (AT_STRINGS_BEG (d
))
4969 if (!bufp
->not_bol
) break;
4971 else if (d
[-1] == '\n' && bufp
->newline_anchor
)
4975 /* In all other cases, we fail. */
4979 /* endline is the dual of begline. */
4981 DEBUG_PRINT1 ("EXECUTING endline.\n");
4983 if (AT_STRINGS_END (d
))
4985 if (!bufp
->not_eol
) break;
4988 /* We have to ``prefetch'' the next character. */
4989 else if ((d
== end1
? *string2
: *d
) == '\n'
4990 && bufp
->newline_anchor
)
4997 /* Match at the very beginning of the data. */
4999 DEBUG_PRINT1 ("EXECUTING begbuf.\n");
5000 if (AT_STRINGS_BEG (d
))
5005 /* Match at the very end of the data. */
5007 DEBUG_PRINT1 ("EXECUTING endbuf.\n");
5008 if (AT_STRINGS_END (d
))
5013 /* on_failure_keep_string_jump is used to optimize `.*\n'. It
5014 pushes NULL as the value for the string on the stack. Then
5015 `pop_failure_point' will keep the current value for the
5016 string, instead of restoring it. To see why, consider
5017 matching `foo\nbar' against `.*\n'. The .* matches the foo;
5018 then the . fails against the \n. But the next thing we want
5019 to do is match the \n against the \n; if we restored the
5020 string value, we would be back at the foo.
5022 Because this is used only in specific cases, we don't need to
5023 check all the things that `on_failure_jump' does, to make
5024 sure the right things get saved on the stack. Hence we don't
5025 share its code. The only reason to push anything on the
5026 stack at all is that otherwise we would have to change
5027 `anychar's code to do something besides goto fail in this
5028 case; that seems worse than this. */
5029 case on_failure_keep_string_jump
:
5030 DEBUG_PRINT1 ("EXECUTING on_failure_keep_string_jump");
5032 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5034 DEBUG_PRINT3 (" %d (to %p):\n", mcnt
, p
+ mcnt
);
5036 DEBUG_PRINT3 (" %d (to 0x%x):\n", mcnt
, p
+ mcnt
);
5039 PUSH_FAILURE_POINT (p
+ mcnt
, NULL
, -2);
5043 /* Uses of on_failure_jump:
5045 Each alternative starts with an on_failure_jump that points
5046 to the beginning of the next alternative. Each alternative
5047 except the last ends with a jump that in effect jumps past
5048 the rest of the alternatives. (They really jump to the
5049 ending jump of the following alternative, because tensioning
5050 these jumps is a hassle.)
5052 Repeats start with an on_failure_jump that points past both
5053 the repetition text and either the following jump or
5054 pop_failure_jump back to this on_failure_jump. */
5055 case on_failure_jump
:
5057 DEBUG_PRINT1 ("EXECUTING on_failure_jump");
5059 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5061 DEBUG_PRINT3 (" %d (to %p)", mcnt
, p
+ mcnt
);
5063 DEBUG_PRINT3 (" %d (to 0x%x)", mcnt
, p
+ mcnt
);
5066 /* If this on_failure_jump comes right before a group (i.e.,
5067 the original * applied to a group), save the information
5068 for that group and all inner ones, so that if we fail back
5069 to this point, the group's information will be correct.
5070 For example, in \(a*\)*\1, we need the preceding group,
5071 and in \(zz\(a*\)b*\)\2, we need the inner group. */
5073 /* We can't use `p' to check ahead because we push
5074 a failure point to `p + mcnt' after we do this. */
5077 /* We need to skip no_op's before we look for the
5078 start_memory in case this on_failure_jump is happening as
5079 the result of a completed succeed_n, as in \(a\)\{1,3\}b\1
5081 while (p1
< pend
&& (re_opcode_t
) *p1
== no_op
)
5084 if (p1
< pend
&& (re_opcode_t
) *p1
== start_memory
)
5086 /* We have a new highest active register now. This will
5087 get reset at the start_memory we are about to get to,
5088 but we will have saved all the registers relevant to
5089 this repetition op, as described above. */
5090 highest_active_reg
= *(p1
+ 1) + *(p1
+ 2);
5091 if (lowest_active_reg
== NO_LOWEST_ACTIVE_REG
)
5092 lowest_active_reg
= *(p1
+ 1);
5095 DEBUG_PRINT1 (":\n");
5096 PUSH_FAILURE_POINT (p
+ mcnt
, d
, -2);
5100 /* A smart repeat ends with `maybe_pop_jump'.
5101 We change it to either `pop_failure_jump' or `jump'. */
5102 case maybe_pop_jump
:
5103 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5104 DEBUG_PRINT2 ("EXECUTING maybe_pop_jump %d.\n", mcnt
);
5106 register unsigned char *p2
= p
;
5108 /* Compare the beginning of the repeat with what in the
5109 pattern follows its end. If we can establish that there
5110 is nothing that they would both match, i.e., that we
5111 would have to backtrack because of (as in, e.g., `a*a')
5112 then we can change to pop_failure_jump, because we'll
5113 never have to backtrack.
5115 This is not true in the case of alternatives: in
5116 `(a|ab)*' we do need to backtrack to the `ab' alternative
5117 (e.g., if the string was `ab'). But instead of trying to
5118 detect that here, the alternative has put on a dummy
5119 failure point which is what we will end up popping. */
5121 /* Skip over open/close-group commands.
5122 If what follows this loop is a ...+ construct,
5123 look at what begins its body, since we will have to
5124 match at least one of that. */
5128 && ((re_opcode_t
) *p2
== stop_memory
5129 || (re_opcode_t
) *p2
== start_memory
))
5131 else if (p2
+ 6 < pend
5132 && (re_opcode_t
) *p2
== dummy_failure_jump
)
5139 /* p1[0] ... p1[2] are the `on_failure_jump' corresponding
5140 to the `maybe_finalize_jump' of this case. Examine what
5143 /* If we're at the end of the pattern, we can change. */
5146 /* Consider what happens when matching ":\(.*\)"
5147 against ":/". I don't really understand this code
5149 p
[-3] = (unsigned char) pop_failure_jump
;
5151 (" End of pattern: change to `pop_failure_jump'.\n");
5154 else if ((re_opcode_t
) *p2
== exactn
5155 || (bufp
->newline_anchor
&& (re_opcode_t
) *p2
== endline
))
5157 register unsigned char c
5158 = *p2
== (unsigned char) endline
? '\n' : p2
[2];
5160 if ((re_opcode_t
) p1
[3] == exactn
&& p1
[5] != c
)
5162 p
[-3] = (unsigned char) pop_failure_jump
;
5163 DEBUG_PRINT3 (" %c != %c => pop_failure_jump.\n",
5167 else if ((re_opcode_t
) p1
[3] == charset
5168 || (re_opcode_t
) p1
[3] == charset_not
)
5170 int not = (re_opcode_t
) p1
[3] == charset_not
;
5172 if (c
< (unsigned char) (p1
[4] * BYTEWIDTH
)
5173 && p1
[5 + c
/ BYTEWIDTH
] & (1 << (c
% BYTEWIDTH
)))
5176 /* `not' is equal to 1 if c would match, which means
5177 that we can't change to pop_failure_jump. */
5180 p
[-3] = (unsigned char) pop_failure_jump
;
5181 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
5185 else if ((re_opcode_t
) *p2
== charset
)
5187 /* We win if the first character of the loop is not part
5189 if ((re_opcode_t
) p1
[3] == exactn
5190 && ! ((int) p2
[1] * BYTEWIDTH
> (int) p1
[5]
5191 && (p2
[2 + p1
[5] / BYTEWIDTH
]
5192 & (1 << (p1
[5] % BYTEWIDTH
)))))
5194 p
[-3] = (unsigned char) pop_failure_jump
;
5195 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
5198 else if ((re_opcode_t
) p1
[3] == charset_not
)
5201 /* We win if the charset_not inside the loop
5202 lists every character listed in the charset after. */
5203 for (idx
= 0; idx
< (int) p2
[1]; idx
++)
5204 if (! (p2
[2 + idx
] == 0
5205 || (idx
< (int) p1
[4]
5206 && ((p2
[2 + idx
] & ~ p1
[5 + idx
]) == 0))))
5211 p
[-3] = (unsigned char) pop_failure_jump
;
5212 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
5215 else if ((re_opcode_t
) p1
[3] == charset
)
5218 /* We win if the charset inside the loop
5219 has no overlap with the one after the loop. */
5221 idx
< (int) p2
[1] && idx
< (int) p1
[4];
5223 if ((p2
[2 + idx
] & p1
[5 + idx
]) != 0)
5226 if (idx
== p2
[1] || idx
== p1
[4])
5228 p
[-3] = (unsigned char) pop_failure_jump
;
5229 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
5234 p
-= 2; /* Point at relative address again. */
5235 if ((re_opcode_t
) p
[-1] != pop_failure_jump
)
5237 p
[-1] = (unsigned char) jump
;
5238 DEBUG_PRINT1 (" Match => jump.\n");
5239 goto unconditional_jump
;
5241 /* Note fall through. */
5244 /* The end of a simple repeat has a pop_failure_jump back to
5245 its matching on_failure_jump, where the latter will push a
5246 failure point. The pop_failure_jump takes off failure
5247 points put on by this pop_failure_jump's matching
5248 on_failure_jump; we got through the pattern to here from the
5249 matching on_failure_jump, so didn't fail. */
5250 case pop_failure_jump
:
5252 /* We need to pass separate storage for the lowest and
5253 highest registers, even though we don't care about the
5254 actual values. Otherwise, we will restore only one
5255 register from the stack, since lowest will == highest in
5256 `pop_failure_point'. */
5257 active_reg_t dummy_low_reg
, dummy_high_reg
;
5258 unsigned char *pdummy
;
5261 DEBUG_PRINT1 ("EXECUTING pop_failure_jump.\n");
5262 POP_FAILURE_POINT (sdummy
, pdummy
,
5263 dummy_low_reg
, dummy_high_reg
,
5264 reg_dummy
, reg_dummy
, reg_info_dummy
);
5266 /* Note fall through. */
5270 DEBUG_PRINT2 ("\n%p: ", p
);
5272 DEBUG_PRINT2 ("\n0x%x: ", p
);
5274 /* Note fall through. */
5276 /* Unconditionally jump (without popping any failure points). */
5278 EXTRACT_NUMBER_AND_INCR (mcnt
, p
); /* Get the amount to jump. */
5279 DEBUG_PRINT2 ("EXECUTING jump %d ", mcnt
);
5280 p
+= mcnt
; /* Do the jump. */
5282 DEBUG_PRINT2 ("(to %p).\n", p
);
5284 DEBUG_PRINT2 ("(to 0x%x).\n", p
);
5289 /* We need this opcode so we can detect where alternatives end
5290 in `group_match_null_string_p' et al. */
5292 DEBUG_PRINT1 ("EXECUTING jump_past_alt.\n");
5293 goto unconditional_jump
;
5296 /* Normally, the on_failure_jump pushes a failure point, which
5297 then gets popped at pop_failure_jump. We will end up at
5298 pop_failure_jump, also, and with a pattern of, say, `a+', we
5299 are skipping over the on_failure_jump, so we have to push
5300 something meaningless for pop_failure_jump to pop. */
5301 case dummy_failure_jump
:
5302 DEBUG_PRINT1 ("EXECUTING dummy_failure_jump.\n");
5303 /* It doesn't matter what we push for the string here. What
5304 the code at `fail' tests is the value for the pattern. */
5305 PUSH_FAILURE_POINT (NULL
, NULL
, -2);
5306 goto unconditional_jump
;
5309 /* At the end of an alternative, we need to push a dummy failure
5310 point in case we are followed by a `pop_failure_jump', because
5311 we don't want the failure point for the alternative to be
5312 popped. For example, matching `(a|ab)*' against `aab'
5313 requires that we match the `ab' alternative. */
5314 case push_dummy_failure
:
5315 DEBUG_PRINT1 ("EXECUTING push_dummy_failure.\n");
5316 /* See comments just above at `dummy_failure_jump' about the
5318 PUSH_FAILURE_POINT (NULL
, NULL
, -2);
5321 /* Have to succeed matching what follows at least n times.
5322 After that, handle like `on_failure_jump'. */
5324 EXTRACT_NUMBER (mcnt
, p
+ 2);
5325 DEBUG_PRINT2 ("EXECUTING succeed_n %d.\n", mcnt
);
5328 /* Originally, this is how many times we HAVE to succeed. */
5333 STORE_NUMBER_AND_INCR (p
, mcnt
);
5335 DEBUG_PRINT3 (" Setting %p to %d.\n", p
- 2, mcnt
);
5337 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p
- 2, mcnt
);
5343 DEBUG_PRINT2 (" Setting two bytes from %p to no_op.\n", p
+2);
5345 DEBUG_PRINT2 (" Setting two bytes from 0x%x to no_op.\n", p
+2);
5347 p
[2] = (unsigned char) no_op
;
5348 p
[3] = (unsigned char) no_op
;
5354 EXTRACT_NUMBER (mcnt
, p
+ 2);
5355 DEBUG_PRINT2 ("EXECUTING jump_n %d.\n", mcnt
);
5357 /* Originally, this is how many times we CAN jump. */
5361 STORE_NUMBER (p
+ 2, mcnt
);
5363 DEBUG_PRINT3 (" Setting %p to %d.\n", p
+ 2, mcnt
);
5365 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p
+ 2, mcnt
);
5367 goto unconditional_jump
;
5369 /* If don't have to jump any more, skip over the rest of command. */
5376 DEBUG_PRINT1 ("EXECUTING set_number_at.\n");
5378 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5380 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5382 DEBUG_PRINT3 (" Setting %p to %d.\n", p1
, mcnt
);
5384 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p1
, mcnt
);
5386 STORE_NUMBER (p1
, mcnt
);
5391 /* The DEC Alpha C compiler 3.x generates incorrect code for the
5392 test WORDCHAR_P (d - 1) != WORDCHAR_P (d) in the expansion of
5393 AT_WORD_BOUNDARY, so this code is disabled. Expanding the
5394 macro and introducing temporary variables works around the bug. */
5397 DEBUG_PRINT1 ("EXECUTING wordbound.\n");
5398 if (AT_WORD_BOUNDARY (d
))
5403 DEBUG_PRINT1 ("EXECUTING notwordbound.\n");
5404 if (AT_WORD_BOUNDARY (d
))
5410 boolean prevchar
, thischar
;
5412 DEBUG_PRINT1 ("EXECUTING wordbound.\n");
5413 if (AT_STRINGS_BEG (d
) || AT_STRINGS_END (d
))
5416 prevchar
= WORDCHAR_P (d
- 1);
5417 thischar
= WORDCHAR_P (d
);
5418 if (prevchar
!= thischar
)
5425 boolean prevchar
, thischar
;
5427 DEBUG_PRINT1 ("EXECUTING notwordbound.\n");
5428 if (AT_STRINGS_BEG (d
) || AT_STRINGS_END (d
))
5431 prevchar
= WORDCHAR_P (d
- 1);
5432 thischar
= WORDCHAR_P (d
);
5433 if (prevchar
!= thischar
)
5440 DEBUG_PRINT1 ("EXECUTING wordbeg.\n");
5441 if (WORDCHAR_P (d
) && (AT_STRINGS_BEG (d
) || !WORDCHAR_P (d
- 1)))
5446 DEBUG_PRINT1 ("EXECUTING wordend.\n");
5447 if (!AT_STRINGS_BEG (d
) && WORDCHAR_P (d
- 1)
5448 && (!WORDCHAR_P (d
) || AT_STRINGS_END (d
)))
5454 DEBUG_PRINT1 ("EXECUTING before_dot.\n");
5455 if (PTR_CHAR_POS ((unsigned char *) d
) >= point
)
5460 DEBUG_PRINT1 ("EXECUTING at_dot.\n");
5461 if (PTR_CHAR_POS ((unsigned char *) d
) != point
)
5466 DEBUG_PRINT1 ("EXECUTING after_dot.\n");
5467 if (PTR_CHAR_POS ((unsigned char *) d
) <= point
)
5472 DEBUG_PRINT2 ("EXECUTING syntaxspec %d.\n", mcnt
);
5477 DEBUG_PRINT1 ("EXECUTING Emacs wordchar.\n");
5481 /* Can't use *d++ here; SYNTAX may be an unsafe macro. */
5483 if (SYNTAX (d
[-1]) != (enum syntaxcode
) mcnt
)
5485 SET_REGS_MATCHED ();
5489 DEBUG_PRINT2 ("EXECUTING notsyntaxspec %d.\n", mcnt
);
5491 goto matchnotsyntax
;
5494 DEBUG_PRINT1 ("EXECUTING Emacs notwordchar.\n");
5498 /* Can't use *d++ here; SYNTAX may be an unsafe macro. */
5500 if (SYNTAX (d
[-1]) == (enum syntaxcode
) mcnt
)
5502 SET_REGS_MATCHED ();
5505 #else /* not emacs */
5507 DEBUG_PRINT1 ("EXECUTING non-Emacs wordchar.\n");
5509 if (!WORDCHAR_P (d
))
5511 SET_REGS_MATCHED ();
5516 DEBUG_PRINT1 ("EXECUTING non-Emacs notwordchar.\n");
5520 SET_REGS_MATCHED ();
5523 #endif /* not emacs */
5528 continue; /* Successfully executed one pattern command; keep going. */
5531 /* We goto here if a matching operation fails. */
5533 if (!FAIL_STACK_EMPTY ())
5534 { /* A restart point is known. Restore to that state. */
5535 DEBUG_PRINT1 ("\nFAIL:\n");
5536 POP_FAILURE_POINT (d
, p
,
5537 lowest_active_reg
, highest_active_reg
,
5538 regstart
, regend
, reg_info
);
5540 /* If this failure point is a dummy, try the next one. */
5544 /* If we failed to the end of the pattern, don't examine *p. */
5548 boolean is_a_jump_n
= false;
5550 /* If failed to a backwards jump that's part of a repetition
5551 loop, need to pop this failure point and use the next one. */
5552 switch ((re_opcode_t
) *p
)
5556 case maybe_pop_jump
:
5557 case pop_failure_jump
:
5560 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
5563 if ((is_a_jump_n
&& (re_opcode_t
) *p1
== succeed_n
)
5565 && (re_opcode_t
) *p1
== on_failure_jump
))
5573 if (d
>= string1
&& d
<= end1
)
5577 break; /* Matching at this starting point really fails. */
5581 goto restore_best_regs
;
5585 return -1; /* Failure to match. */
5588 /* Subroutine definitions for re_match_2. */
5591 /* We are passed P pointing to a register number after a start_memory.
5593 Return true if the pattern up to the corresponding stop_memory can
5594 match the empty string, and false otherwise.
5596 If we find the matching stop_memory, sets P to point to one past its number.
5597 Otherwise, sets P to an undefined byte less than or equal to END.
5599 We don't handle duplicates properly (yet). */
5602 group_match_null_string_p (p
, end
, reg_info
)
5603 unsigned char **p
, *end
;
5604 register_info_type
*reg_info
;
5607 /* Point to after the args to the start_memory. */
5608 unsigned char *p1
= *p
+ 2;
5612 /* Skip over opcodes that can match nothing, and return true or
5613 false, as appropriate, when we get to one that can't, or to the
5614 matching stop_memory. */
5616 switch ((re_opcode_t
) *p1
)
5618 /* Could be either a loop or a series of alternatives. */
5619 case on_failure_jump
:
5621 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
5623 /* If the next operation is not a jump backwards in the
5628 /* Go through the on_failure_jumps of the alternatives,
5629 seeing if any of the alternatives cannot match nothing.
5630 The last alternative starts with only a jump,
5631 whereas the rest start with on_failure_jump and end
5632 with a jump, e.g., here is the pattern for `a|b|c':
5634 /on_failure_jump/0/6/exactn/1/a/jump_past_alt/0/6
5635 /on_failure_jump/0/6/exactn/1/b/jump_past_alt/0/3
5638 So, we have to first go through the first (n-1)
5639 alternatives and then deal with the last one separately. */
5642 /* Deal with the first (n-1) alternatives, which start
5643 with an on_failure_jump (see above) that jumps to right
5644 past a jump_past_alt. */
5646 while ((re_opcode_t
) p1
[mcnt
-3] == jump_past_alt
)
5648 /* `mcnt' holds how many bytes long the alternative
5649 is, including the ending `jump_past_alt' and
5652 if (!alt_match_null_string_p (p1
, p1
+ mcnt
- 3,
5656 /* Move to right after this alternative, including the
5660 /* Break if it's the beginning of an n-th alternative
5661 that doesn't begin with an on_failure_jump. */
5662 if ((re_opcode_t
) *p1
!= on_failure_jump
)
5665 /* Still have to check that it's not an n-th
5666 alternative that starts with an on_failure_jump. */
5668 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
5669 if ((re_opcode_t
) p1
[mcnt
-3] != jump_past_alt
)
5671 /* Get to the beginning of the n-th alternative. */
5677 /* Deal with the last alternative: go back and get number
5678 of the `jump_past_alt' just before it. `mcnt' contains
5679 the length of the alternative. */
5680 EXTRACT_NUMBER (mcnt
, p1
- 2);
5682 if (!alt_match_null_string_p (p1
, p1
+ mcnt
, reg_info
))
5685 p1
+= mcnt
; /* Get past the n-th alternative. */
5691 assert (p1
[1] == **p
);
5697 if (!common_op_match_null_string_p (&p1
, end
, reg_info
))
5700 } /* while p1 < end */
5703 } /* group_match_null_string_p */
5706 /* Similar to group_match_null_string_p, but doesn't deal with alternatives:
5707 It expects P to be the first byte of a single alternative and END one
5708 byte past the last. The alternative can contain groups. */
5711 alt_match_null_string_p (p
, end
, reg_info
)
5712 unsigned char *p
, *end
;
5713 register_info_type
*reg_info
;
5716 unsigned char *p1
= p
;
5720 /* Skip over opcodes that can match nothing, and break when we get
5721 to one that can't. */
5723 switch ((re_opcode_t
) *p1
)
5726 case on_failure_jump
:
5728 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
5733 if (!common_op_match_null_string_p (&p1
, end
, reg_info
))
5736 } /* while p1 < end */
5739 } /* alt_match_null_string_p */
5742 /* Deals with the ops common to group_match_null_string_p and
5743 alt_match_null_string_p.
5745 Sets P to one after the op and its arguments, if any. */
5748 common_op_match_null_string_p (p
, end
, reg_info
)
5749 unsigned char **p
, *end
;
5750 register_info_type
*reg_info
;
5755 unsigned char *p1
= *p
;
5757 switch ((re_opcode_t
) *p1
++)
5777 assert (reg_no
> 0 && reg_no
<= MAX_REGNUM
);
5778 ret
= group_match_null_string_p (&p1
, end
, reg_info
);
5780 /* Have to set this here in case we're checking a group which
5781 contains a group and a back reference to it. */
5783 if (REG_MATCH_NULL_STRING_P (reg_info
[reg_no
]) == MATCH_NULL_UNSET_VALUE
)
5784 REG_MATCH_NULL_STRING_P (reg_info
[reg_no
]) = ret
;
5790 /* If this is an optimized succeed_n for zero times, make the jump. */
5792 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
5800 /* Get to the number of times to succeed. */
5802 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
5807 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
5815 if (!REG_MATCH_NULL_STRING_P (reg_info
[*p1
]))
5823 /* All other opcodes mean we cannot match the empty string. */
5829 } /* common_op_match_null_string_p */
5832 /* Return zero if TRANSLATE[S1] and TRANSLATE[S2] are identical for LEN
5833 bytes; nonzero otherwise. */
5836 bcmp_translate (s1
, s2
, len
, translate
)
5837 const char *s1
, *s2
;
5839 RE_TRANSLATE_TYPE translate
;
5841 register const unsigned char *p1
= (const unsigned char *) s1
;
5842 register const unsigned char *p2
= (const unsigned char *) s2
;
5845 if (translate
[*p1
++] != translate
[*p2
++]) return 1;
5851 /* Entry points for GNU code. */
5853 /* re_compile_pattern is the GNU regular expression compiler: it
5854 compiles PATTERN (of length SIZE) and puts the result in BUFP.
5855 Returns 0 if the pattern was valid, otherwise an error string.
5857 Assumes the `allocated' (and perhaps `buffer') and `translate' fields
5858 are set in BUFP on entry.
5860 We call regex_compile to do the actual compilation. */
5863 re_compile_pattern (pattern
, length
, bufp
)
5864 const char *pattern
;
5866 struct re_pattern_buffer
*bufp
;
5870 /* GNU code is written to assume at least RE_NREGS registers will be set
5871 (and at least one extra will be -1). */
5872 bufp
->regs_allocated
= REGS_UNALLOCATED
;
5874 /* And GNU code determines whether or not to get register information
5875 by passing null for the REGS argument to re_match, etc., not by
5879 /* Match anchors at newline. */
5880 bufp
->newline_anchor
= 1;
5882 ret
= regex_compile (pattern
, length
, re_syntax_options
, bufp
);
5886 return gettext (re_error_msgid
+ re_error_msgid_idx
[(int) ret
]);
5889 weak_alias (__re_compile_pattern
, re_compile_pattern
)
5892 /* Entry points compatible with 4.2 BSD regex library. We don't define
5893 them unless specifically requested. */
5895 #if defined _REGEX_RE_COMP || defined _LIBC
5897 /* BSD has one and only one pattern buffer. */
5898 static struct re_pattern_buffer re_comp_buf
;
5902 /* Make these definitions weak in libc, so POSIX programs can redefine
5903 these names if they don't use our functions, and still use
5904 regcomp/regexec below without link errors. */
5914 if (!re_comp_buf
.buffer
)
5915 return gettext ("No previous regular expression");
5919 if (!re_comp_buf
.buffer
)
5921 re_comp_buf
.buffer
= (unsigned char *) malloc (200);
5922 if (re_comp_buf
.buffer
== NULL
)
5923 return (char *) gettext (re_error_msgid
5924 + re_error_msgid_idx
[(int) REG_ESPACE
]);
5925 re_comp_buf
.allocated
= 200;
5927 re_comp_buf
.fastmap
= (char *) malloc (1 << BYTEWIDTH
);
5928 if (re_comp_buf
.fastmap
== NULL
)
5929 return (char *) gettext (re_error_msgid
5930 + re_error_msgid_idx
[(int) REG_ESPACE
]);
5933 /* Since `re_exec' always passes NULL for the `regs' argument, we
5934 don't need to initialize the pattern buffer fields which affect it. */
5936 /* Match anchors at newlines. */
5937 re_comp_buf
.newline_anchor
= 1;
5939 ret
= regex_compile (s
, strlen (s
), re_syntax_options
, &re_comp_buf
);
5944 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
5945 return (char *) gettext (re_error_msgid
+ re_error_msgid_idx
[(int) ret
]);
5956 const int len
= strlen (s
);
5958 0 <= re_search (&re_comp_buf
, s
, len
, 0, len
, (struct re_registers
*) 0);
5961 #endif /* _REGEX_RE_COMP */
5963 /* POSIX.2 functions. Don't define these for Emacs. */
5967 /* regcomp takes a regular expression as a string and compiles it.
5969 PREG is a regex_t *. We do not expect any fields to be initialized,
5970 since POSIX says we shouldn't. Thus, we set
5972 `buffer' to the compiled pattern;
5973 `used' to the length of the compiled pattern;
5974 `syntax' to RE_SYNTAX_POSIX_EXTENDED if the
5975 REG_EXTENDED bit in CFLAGS is set; otherwise, to
5976 RE_SYNTAX_POSIX_BASIC;
5977 `newline_anchor' to REG_NEWLINE being set in CFLAGS;
5978 `fastmap' to an allocated space for the fastmap;
5979 `fastmap_accurate' to zero;
5980 `re_nsub' to the number of subexpressions in PATTERN.
5982 PATTERN is the address of the pattern string.
5984 CFLAGS is a series of bits which affect compilation.
5986 If REG_EXTENDED is set, we use POSIX extended syntax; otherwise, we
5987 use POSIX basic syntax.
5989 If REG_NEWLINE is set, then . and [^...] don't match newline.
5990 Also, regexec will try a match beginning after every newline.
5992 If REG_ICASE is set, then we considers upper- and lowercase
5993 versions of letters to be equivalent when matching.
5995 If REG_NOSUB is set, then when PREG is passed to regexec, that
5996 routine will report only success or failure, and nothing about the
5999 It returns 0 if it succeeds, nonzero if it doesn't. (See regex.h for
6000 the return codes and their meanings.) */
6003 regcomp (preg
, pattern
, cflags
)
6005 const char *pattern
;
6010 = (cflags
& REG_EXTENDED
) ?
6011 RE_SYNTAX_POSIX_EXTENDED
: RE_SYNTAX_POSIX_BASIC
;
6013 /* regex_compile will allocate the space for the compiled pattern. */
6015 preg
->allocated
= 0;
6018 /* Try to allocate space for the fastmap. */
6019 preg
->fastmap
= (char *) malloc (1 << BYTEWIDTH
);
6021 if (cflags
& REG_ICASE
)
6026 = (RE_TRANSLATE_TYPE
) malloc (CHAR_SET_SIZE
6027 * sizeof (*(RE_TRANSLATE_TYPE
)0));
6028 if (preg
->translate
== NULL
)
6029 return (int) REG_ESPACE
;
6031 /* Map uppercase characters to corresponding lowercase ones. */
6032 for (i
= 0; i
< CHAR_SET_SIZE
; i
++)
6033 preg
->translate
[i
] = ISUPPER (i
) ? TOLOWER (i
) : i
;
6036 preg
->translate
= NULL
;
6038 /* If REG_NEWLINE is set, newlines are treated differently. */
6039 if (cflags
& REG_NEWLINE
)
6040 { /* REG_NEWLINE implies neither . nor [^...] match newline. */
6041 syntax
&= ~RE_DOT_NEWLINE
;
6042 syntax
|= RE_HAT_LISTS_NOT_NEWLINE
;
6043 /* It also changes the matching behavior. */
6044 preg
->newline_anchor
= 1;
6047 preg
->newline_anchor
= 0;
6049 preg
->no_sub
= !!(cflags
& REG_NOSUB
);
6051 /* POSIX says a null character in the pattern terminates it, so we
6052 can use strlen here in compiling the pattern. */
6053 ret
= regex_compile (pattern
, strlen (pattern
), syntax
, preg
);
6055 /* POSIX doesn't distinguish between an unmatched open-group and an
6056 unmatched close-group: both are REG_EPAREN. */
6057 if (ret
== REG_ERPAREN
) ret
= REG_EPAREN
;
6059 if (ret
== REG_NOERROR
&& preg
->fastmap
)
6061 /* Compute the fastmap now, since regexec cannot modify the pattern
6063 if (re_compile_fastmap (preg
) == -2)
6065 /* Some error occurred while computing the fastmap, just forget
6067 free (preg
->fastmap
);
6068 preg
->fastmap
= NULL
;
6075 weak_alias (__regcomp
, regcomp
)
6079 /* regexec searches for a given pattern, specified by PREG, in the
6082 If NMATCH is zero or REG_NOSUB was set in the cflags argument to
6083 `regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at
6084 least NMATCH elements, and we set them to the offsets of the
6085 corresponding matched substrings.
6087 EFLAGS specifies `execution flags' which affect matching: if
6088 REG_NOTBOL is set, then ^ does not match at the beginning of the
6089 string; if REG_NOTEOL is set, then $ does not match at the end.
6091 We return 0 if we find a match and REG_NOMATCH if not. */
6094 regexec (preg
, string
, nmatch
, pmatch
, eflags
)
6095 const regex_t
*preg
;
6098 regmatch_t pmatch
[];
6102 struct re_registers regs
;
6103 regex_t private_preg
;
6104 int len
= strlen (string
);
6105 boolean want_reg_info
= !preg
->no_sub
&& nmatch
> 0;
6107 private_preg
= *preg
;
6109 private_preg
.not_bol
= !!(eflags
& REG_NOTBOL
);
6110 private_preg
.not_eol
= !!(eflags
& REG_NOTEOL
);
6112 /* The user has told us exactly how many registers to return
6113 information about, via `nmatch'. We have to pass that on to the
6114 matching routines. */
6115 private_preg
.regs_allocated
= REGS_FIXED
;
6119 regs
.num_regs
= nmatch
;
6120 regs
.start
= TALLOC (nmatch
* 2, regoff_t
);
6121 if (regs
.start
== NULL
)
6122 return (int) REG_NOMATCH
;
6123 regs
.end
= regs
.start
+ nmatch
;
6126 /* Perform the searching operation. */
6127 ret
= re_search (&private_preg
, string
, len
,
6128 /* start: */ 0, /* range: */ len
,
6129 want_reg_info
? ®s
: (struct re_registers
*) 0);
6131 /* Copy the register information to the POSIX structure. */
6138 for (r
= 0; r
< nmatch
; r
++)
6140 pmatch
[r
].rm_so
= regs
.start
[r
];
6141 pmatch
[r
].rm_eo
= regs
.end
[r
];
6145 /* If we needed the temporary register info, free the space now. */
6149 /* We want zero return to mean success, unlike `re_search'. */
6150 return ret
>= 0 ? (int) REG_NOERROR
: (int) REG_NOMATCH
;
6153 weak_alias (__regexec
, regexec
)
6157 /* Returns a message corresponding to an error code, ERRCODE, returned
6158 from either regcomp or regexec. We don't use PREG here. */
6161 regerror (errcode
, preg
, errbuf
, errbuf_size
)
6163 const regex_t
*preg
;
6171 || errcode
>= (int) (sizeof (re_error_msgid_idx
)
6172 / sizeof (re_error_msgid_idx
[0])))
6173 /* Only error codes returned by the rest of the code should be passed
6174 to this routine. If we are given anything else, or if other regex
6175 code generates an invalid error code, then the program has a bug.
6176 Dump core so we can fix it. */
6179 msg
= gettext (re_error_msgid
+ re_error_msgid_idx
[errcode
]);
6181 msg_size
= strlen (msg
) + 1; /* Includes the null. */
6183 if (errbuf_size
!= 0)
6185 if (msg_size
> errbuf_size
)
6187 #if defined HAVE_MEMPCPY || defined _LIBC
6188 *((char *) __mempcpy (errbuf
, msg
, errbuf_size
- 1)) = '\0';
6190 memcpy (errbuf
, msg
, errbuf_size
- 1);
6191 errbuf
[errbuf_size
- 1] = 0;
6195 memcpy (errbuf
, msg
, msg_size
);
6201 weak_alias (__regerror
, regerror
)
6205 /* Free dynamically allocated space used by PREG. */
6211 if (preg
->buffer
!= NULL
)
6212 free (preg
->buffer
);
6213 preg
->buffer
= NULL
;
6215 preg
->allocated
= 0;
6218 if (preg
->fastmap
!= NULL
)
6219 free (preg
->fastmap
);
6220 preg
->fastmap
= NULL
;
6221 preg
->fastmap_accurate
= 0;
6223 if (preg
->translate
!= NULL
)
6224 free (preg
->translate
);
6225 preg
->translate
= NULL
;
6228 weak_alias (__regfree
, regfree
)
6231 #endif /* not emacs */