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. */
694 printf ("%d:\t", p
- start
);
696 switch ((re_opcode_t
) *p
++)
704 printf ("/exactn/%d", mcnt
);
715 printf ("/start_memory/%d/%d", mcnt
, *p
++);
720 printf ("/stop_memory/%d/%d", mcnt
, *p
++);
724 printf ("/duplicate/%d", *p
++);
734 register int c
, last
= -100;
735 register int in_range
= 0;
737 printf ("/charset [%s",
738 (re_opcode_t
) *(p
- 1) == charset_not
? "^" : "");
740 assert (p
+ *p
< pend
);
742 for (c
= 0; c
< 256; c
++)
744 && (p
[1 + (c
/8)] & (1 << (c
% 8))))
746 /* Are we starting a range? */
747 if (last
+ 1 == c
&& ! in_range
)
752 /* Have we broken a range? */
753 else if (last
+ 1 != c
&& in_range
)
782 case on_failure_jump
:
783 extract_number_and_incr (&mcnt
, &p
);
784 printf ("/on_failure_jump to %d", p
+ mcnt
- start
);
787 case on_failure_keep_string_jump
:
788 extract_number_and_incr (&mcnt
, &p
);
789 printf ("/on_failure_keep_string_jump to %d", p
+ mcnt
- start
);
792 case dummy_failure_jump
:
793 extract_number_and_incr (&mcnt
, &p
);
794 printf ("/dummy_failure_jump to %d", p
+ mcnt
- start
);
797 case push_dummy_failure
:
798 printf ("/push_dummy_failure");
802 extract_number_and_incr (&mcnt
, &p
);
803 printf ("/maybe_pop_jump to %d", p
+ mcnt
- start
);
806 case pop_failure_jump
:
807 extract_number_and_incr (&mcnt
, &p
);
808 printf ("/pop_failure_jump to %d", p
+ mcnt
- start
);
812 extract_number_and_incr (&mcnt
, &p
);
813 printf ("/jump_past_alt to %d", p
+ mcnt
- start
);
817 extract_number_and_incr (&mcnt
, &p
);
818 printf ("/jump to %d", p
+ mcnt
- start
);
822 extract_number_and_incr (&mcnt
, &p
);
824 extract_number_and_incr (&mcnt2
, &p
);
825 printf ("/succeed_n to %d, %d times", p1
- start
, mcnt2
);
829 extract_number_and_incr (&mcnt
, &p
);
831 extract_number_and_incr (&mcnt2
, &p
);
832 printf ("/jump_n to %d, %d times", p1
- start
, mcnt2
);
836 extract_number_and_incr (&mcnt
, &p
);
838 extract_number_and_incr (&mcnt2
, &p
);
839 printf ("/set_number_at location %d to %d", p1
- start
, mcnt2
);
843 printf ("/wordbound");
847 printf ("/notwordbound");
859 printf ("/before_dot");
867 printf ("/after_dot");
871 printf ("/syntaxspec");
873 printf ("/%d", mcnt
);
877 printf ("/notsyntaxspec");
879 printf ("/%d", mcnt
);
884 printf ("/wordchar");
888 printf ("/notwordchar");
900 printf ("?%d", *(p
-1));
906 printf ("%d:\tend of pattern.\n", p
- start
);
911 print_compiled_pattern (bufp
)
912 struct re_pattern_buffer
*bufp
;
914 unsigned char *buffer
= bufp
->buffer
;
916 print_partial_compiled_pattern (buffer
, buffer
+ bufp
->used
);
917 printf ("%ld bytes used/%ld bytes allocated.\n",
918 bufp
->used
, bufp
->allocated
);
920 if (bufp
->fastmap_accurate
&& bufp
->fastmap
)
922 printf ("fastmap: ");
923 print_fastmap (bufp
->fastmap
);
926 printf ("re_nsub: %d\t", bufp
->re_nsub
);
927 printf ("regs_alloc: %d\t", bufp
->regs_allocated
);
928 printf ("can_be_null: %d\t", bufp
->can_be_null
);
929 printf ("newline_anchor: %d\n", bufp
->newline_anchor
);
930 printf ("no_sub: %d\t", bufp
->no_sub
);
931 printf ("not_bol: %d\t", bufp
->not_bol
);
932 printf ("not_eol: %d\t", bufp
->not_eol
);
933 printf ("syntax: %lx\n", bufp
->syntax
);
934 /* Perhaps we should print the translate table? */
939 print_double_string (where
, string1
, size1
, string2
, size2
)
952 if (FIRST_STRING_P (where
))
954 for (this_char
= where
- string1
; this_char
< size1
; this_char
++)
955 putchar (string1
[this_char
]);
960 for (this_char
= where
- string2
; this_char
< size2
; this_char
++)
961 putchar (string2
[this_char
]);
972 #else /* not DEBUG */
977 # define DEBUG_STATEMENT(e)
978 # define DEBUG_PRINT1(x)
979 # define DEBUG_PRINT2(x1, x2)
980 # define DEBUG_PRINT3(x1, x2, x3)
981 # define DEBUG_PRINT4(x1, x2, x3, x4)
982 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e)
983 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2)
985 #endif /* not DEBUG */
987 /* Set by `re_set_syntax' to the current regexp syntax to recognize. Can
988 also be assigned to arbitrarily: each pattern buffer stores its own
989 syntax, so it can be changed between regex compilations. */
990 /* This has no initializer because initialized variables in Emacs
991 become read-only after dumping. */
992 reg_syntax_t re_syntax_options
;
995 /* Specify the precise syntax of regexps for compilation. This provides
996 for compatibility for various utilities which historically have
997 different, incompatible syntaxes.
999 The argument SYNTAX is a bit mask comprised of the various bits
1000 defined in regex.h. We return the old syntax. */
1003 re_set_syntax (syntax
)
1004 reg_syntax_t syntax
;
1006 reg_syntax_t ret
= re_syntax_options
;
1008 re_syntax_options
= syntax
;
1010 if (syntax
& RE_DEBUG
)
1012 else if (debug
) /* was on but now is not */
1018 weak_alias (__re_set_syntax
, re_set_syntax
)
1021 /* This table gives an error message for each of the error codes listed
1022 in regex.h. Obviously the order here has to be same as there.
1023 POSIX doesn't require that we do anything for REG_NOERROR,
1024 but why not be nice? */
1026 static const char re_error_msgid
[] =
1028 #define REG_NOERROR_IDX 0
1029 gettext_noop ("Success") /* REG_NOERROR */
1031 #define REG_NOMATCH_IDX (REG_NOERROR_IDX + sizeof "Success")
1032 gettext_noop ("No match") /* REG_NOMATCH */
1034 #define REG_BADPAT_IDX (REG_NOMATCH_IDX + sizeof "No match")
1035 gettext_noop ("Invalid regular expression") /* REG_BADPAT */
1037 #define REG_ECOLLATE_IDX (REG_BADPAT_IDX + sizeof "Invalid regular expression")
1038 gettext_noop ("Invalid collation character") /* REG_ECOLLATE */
1040 #define REG_ECTYPE_IDX (REG_ECOLLATE_IDX + sizeof "Invalid collation character")
1041 gettext_noop ("Invalid character class name") /* REG_ECTYPE */
1043 #define REG_EESCAPE_IDX (REG_ECTYPE_IDX + sizeof "Invalid character class name")
1044 gettext_noop ("Trailing backslash") /* REG_EESCAPE */
1046 #define REG_ESUBREG_IDX (REG_EESCAPE_IDX + sizeof "Trailing backslash")
1047 gettext_noop ("Invalid back reference") /* REG_ESUBREG */
1049 #define REG_EBRACK_IDX (REG_ESUBREG_IDX + sizeof "Invalid back reference")
1050 gettext_noop ("Unmatched [ or [^") /* REG_EBRACK */
1052 #define REG_EPAREN_IDX (REG_EBRACK_IDX + sizeof "Unmatched [ or [^")
1053 gettext_noop ("Unmatched ( or \\(") /* REG_EPAREN */
1055 #define REG_EBRACE_IDX (REG_EPAREN_IDX + sizeof "Unmatched ( or \\(")
1056 gettext_noop ("Unmatched \\{") /* REG_EBRACE */
1058 #define REG_BADBR_IDX (REG_EBRACE_IDX + sizeof "Unmatched \\{")
1059 gettext_noop ("Invalid content of \\{\\}") /* REG_BADBR */
1061 #define REG_ERANGE_IDX (REG_BADBR_IDX + sizeof "Invalid content of \\{\\}")
1062 gettext_noop ("Invalid range end") /* REG_ERANGE */
1064 #define REG_ESPACE_IDX (REG_ERANGE_IDX + sizeof "Invalid range end")
1065 gettext_noop ("Memory exhausted") /* REG_ESPACE */
1067 #define REG_BADRPT_IDX (REG_ESPACE_IDX + sizeof "Memory exhausted")
1068 gettext_noop ("Invalid preceding regular expression") /* REG_BADRPT */
1070 #define REG_EEND_IDX (REG_BADRPT_IDX + sizeof "Invalid preceding regular expression")
1071 gettext_noop ("Premature end of regular expression") /* REG_EEND */
1073 #define REG_ESIZE_IDX (REG_EEND_IDX + sizeof "Premature end of regular expression")
1074 gettext_noop ("Regular expression too big") /* REG_ESIZE */
1076 #define REG_ERPAREN_IDX (REG_ESIZE_IDX + sizeof "Regular expression too big")
1077 gettext_noop ("Unmatched ) or \\)") /* REG_ERPAREN */
1080 static const size_t re_error_msgid_idx
[] =
1101 /* Avoiding alloca during matching, to placate r_alloc. */
1103 /* Define MATCH_MAY_ALLOCATE unless we need to make sure that the
1104 searching and matching functions should not call alloca. On some
1105 systems, alloca is implemented in terms of malloc, and if we're
1106 using the relocating allocator routines, then malloc could cause a
1107 relocation, which might (if the strings being searched are in the
1108 ralloc heap) shift the data out from underneath the regexp
1111 Here's another reason to avoid allocation: Emacs
1112 processes input from X in a signal handler; processing X input may
1113 call malloc; if input arrives while a matching routine is calling
1114 malloc, then we're scrod. But Emacs can't just block input while
1115 calling matching routines; then we don't notice interrupts when
1116 they come in. So, Emacs blocks input around all regexp calls
1117 except the matching calls, which it leaves unprotected, in the
1118 faith that they will not malloc. */
1120 /* Normally, this is fine. */
1121 #define MATCH_MAY_ALLOCATE
1123 /* When using GNU C, we are not REALLY using the C alloca, no matter
1124 what config.h may say. So don't take precautions for it. */
1129 /* The match routines may not allocate if (1) they would do it with malloc
1130 and (2) it's not safe for them to use malloc.
1131 Note that if REL_ALLOC is defined, matching would not use malloc for the
1132 failure stack, but we would still use it for the register vectors;
1133 so REL_ALLOC should not affect this. */
1134 #if (defined C_ALLOCA || defined REGEX_MALLOC) && defined emacs
1135 # undef MATCH_MAY_ALLOCATE
1139 /* Failure stack declarations and macros; both re_compile_fastmap and
1140 re_match_2 use a failure stack. These have to be macros because of
1141 REGEX_ALLOCATE_STACK. */
1144 /* Number of failure points for which to initially allocate space
1145 when matching. If this number is exceeded, we allocate more
1146 space, so it is not a hard limit. */
1147 #ifndef INIT_FAILURE_ALLOC
1148 # define INIT_FAILURE_ALLOC 5
1151 /* Roughly the maximum number of failure points on the stack. Would be
1152 exactly that if always used MAX_FAILURE_ITEMS items each time we failed.
1153 This is a variable only so users of regex can assign to it; we never
1154 change it ourselves. */
1158 # if defined MATCH_MAY_ALLOCATE
1159 /* 4400 was enough to cause a crash on Alpha OSF/1,
1160 whose default stack limit is 2mb. */
1161 long int re_max_failures
= 4000;
1163 long int re_max_failures
= 2000;
1166 union fail_stack_elt
1168 unsigned char *pointer
;
1172 typedef union fail_stack_elt fail_stack_elt_t
;
1176 fail_stack_elt_t
*stack
;
1177 unsigned long int size
;
1178 unsigned long int avail
; /* Offset of next open position. */
1181 #else /* not INT_IS_16BIT */
1183 # if defined MATCH_MAY_ALLOCATE
1184 /* 4400 was enough to cause a crash on Alpha OSF/1,
1185 whose default stack limit is 2mb. */
1186 int re_max_failures
= 20000;
1188 int re_max_failures
= 2000;
1191 union fail_stack_elt
1193 unsigned char *pointer
;
1197 typedef union fail_stack_elt fail_stack_elt_t
;
1201 fail_stack_elt_t
*stack
;
1203 unsigned avail
; /* Offset of next open position. */
1206 #endif /* INT_IS_16BIT */
1208 #define FAIL_STACK_EMPTY() (fail_stack.avail == 0)
1209 #define FAIL_STACK_PTR_EMPTY() (fail_stack_ptr->avail == 0)
1210 #define FAIL_STACK_FULL() (fail_stack.avail == fail_stack.size)
1213 /* Define macros to initialize and free the failure stack.
1214 Do `return -2' if the alloc fails. */
1216 #ifdef MATCH_MAY_ALLOCATE
1217 # define INIT_FAIL_STACK() \
1219 fail_stack.stack = (fail_stack_elt_t *) \
1220 REGEX_ALLOCATE_STACK (INIT_FAILURE_ALLOC * sizeof (fail_stack_elt_t)); \
1222 if (fail_stack.stack == NULL) \
1225 fail_stack.size = INIT_FAILURE_ALLOC; \
1226 fail_stack.avail = 0; \
1229 # define RESET_FAIL_STACK() REGEX_FREE_STACK (fail_stack.stack)
1231 # define INIT_FAIL_STACK() \
1233 fail_stack.avail = 0; \
1236 # define RESET_FAIL_STACK()
1240 /* Double the size of FAIL_STACK, up to approximately `re_max_failures' items.
1242 Return 1 if succeeds, and 0 if either ran out of memory
1243 allocating space for it or it was already too large.
1245 REGEX_REALLOCATE_STACK requires `destination' be declared. */
1247 #define DOUBLE_FAIL_STACK(fail_stack) \
1248 ((fail_stack).size > (unsigned) (re_max_failures * MAX_FAILURE_ITEMS) \
1250 : ((fail_stack).stack = (fail_stack_elt_t *) \
1251 REGEX_REALLOCATE_STACK ((fail_stack).stack, \
1252 (fail_stack).size * sizeof (fail_stack_elt_t), \
1253 ((fail_stack).size << 1) * sizeof (fail_stack_elt_t)), \
1255 (fail_stack).stack == NULL \
1257 : ((fail_stack).size <<= 1, \
1261 /* Push pointer POINTER on FAIL_STACK.
1262 Return 1 if was able to do so and 0 if ran out of memory allocating
1264 #define PUSH_PATTERN_OP(POINTER, FAIL_STACK) \
1265 ((FAIL_STACK_FULL () \
1266 && !DOUBLE_FAIL_STACK (FAIL_STACK)) \
1268 : ((FAIL_STACK).stack[(FAIL_STACK).avail++].pointer = POINTER, \
1271 /* Push a pointer value onto the failure stack.
1272 Assumes the variable `fail_stack'. Probably should only
1273 be called from within `PUSH_FAILURE_POINT'. */
1274 #define PUSH_FAILURE_POINTER(item) \
1275 fail_stack.stack[fail_stack.avail++].pointer = (unsigned char *) (item)
1277 /* This pushes an integer-valued item onto the failure stack.
1278 Assumes the variable `fail_stack'. Probably should only
1279 be called from within `PUSH_FAILURE_POINT'. */
1280 #define PUSH_FAILURE_INT(item) \
1281 fail_stack.stack[fail_stack.avail++].integer = (item)
1283 /* Push a fail_stack_elt_t value onto the failure stack.
1284 Assumes the variable `fail_stack'. Probably should only
1285 be called from within `PUSH_FAILURE_POINT'. */
1286 #define PUSH_FAILURE_ELT(item) \
1287 fail_stack.stack[fail_stack.avail++] = (item)
1289 /* These three POP... operations complement the three PUSH... operations.
1290 All assume that `fail_stack' is nonempty. */
1291 #define POP_FAILURE_POINTER() fail_stack.stack[--fail_stack.avail].pointer
1292 #define POP_FAILURE_INT() fail_stack.stack[--fail_stack.avail].integer
1293 #define POP_FAILURE_ELT() fail_stack.stack[--fail_stack.avail]
1295 /* Used to omit pushing failure point id's when we're not debugging. */
1297 # define DEBUG_PUSH PUSH_FAILURE_INT
1298 # define DEBUG_POP(item_addr) *(item_addr) = POP_FAILURE_INT ()
1300 # define DEBUG_PUSH(item)
1301 # define DEBUG_POP(item_addr)
1305 /* Push the information about the state we will need
1306 if we ever fail back to it.
1308 Requires variables fail_stack, regstart, regend, reg_info, and
1309 num_regs_pushed be declared. DOUBLE_FAIL_STACK requires `destination'
1312 Does `return FAILURE_CODE' if runs out of memory. */
1314 #define PUSH_FAILURE_POINT(pattern_place, string_place, failure_code) \
1316 char *destination; \
1317 /* Must be int, so when we don't save any registers, the arithmetic \
1318 of 0 + -1 isn't done as unsigned. */ \
1319 /* Can't be int, since there is not a shred of a guarantee that int \
1320 is wide enough to hold a value of something to which pointer can \
1322 active_reg_t this_reg; \
1324 DEBUG_STATEMENT (failure_id++); \
1325 DEBUG_STATEMENT (nfailure_points_pushed++); \
1326 DEBUG_PRINT2 ("\nPUSH_FAILURE_POINT #%u:\n", failure_id); \
1327 DEBUG_PRINT2 (" Before push, next avail: %d\n", (fail_stack).avail);\
1328 DEBUG_PRINT2 (" size: %d\n", (fail_stack).size);\
1330 DEBUG_PRINT2 (" slots needed: %ld\n", NUM_FAILURE_ITEMS); \
1331 DEBUG_PRINT2 (" available: %d\n", REMAINING_AVAIL_SLOTS); \
1333 /* Ensure we have enough space allocated for what we will push. */ \
1334 while (REMAINING_AVAIL_SLOTS < NUM_FAILURE_ITEMS) \
1336 if (!DOUBLE_FAIL_STACK (fail_stack)) \
1337 return failure_code; \
1339 DEBUG_PRINT2 ("\n Doubled stack; size now: %d\n", \
1340 (fail_stack).size); \
1341 DEBUG_PRINT2 (" slots available: %d\n", REMAINING_AVAIL_SLOTS);\
1344 /* Push the info, starting with the registers. */ \
1345 DEBUG_PRINT1 ("\n"); \
1348 for (this_reg = lowest_active_reg; this_reg <= highest_active_reg; \
1351 DEBUG_PRINT2 (" Pushing reg: %lu\n", this_reg); \
1352 DEBUG_STATEMENT (num_regs_pushed++); \
1354 DEBUG_PRINT2 (" start: %p\n", regstart[this_reg]); \
1355 PUSH_FAILURE_POINTER (regstart[this_reg]); \
1357 DEBUG_PRINT2 (" end: %p\n", regend[this_reg]); \
1358 PUSH_FAILURE_POINTER (regend[this_reg]); \
1360 DEBUG_PRINT2 (" info: %p\n ", \
1361 reg_info[this_reg].word.pointer); \
1362 DEBUG_PRINT2 (" match_null=%d", \
1363 REG_MATCH_NULL_STRING_P (reg_info[this_reg])); \
1364 DEBUG_PRINT2 (" active=%d", IS_ACTIVE (reg_info[this_reg])); \
1365 DEBUG_PRINT2 (" matched_something=%d", \
1366 MATCHED_SOMETHING (reg_info[this_reg])); \
1367 DEBUG_PRINT2 (" ever_matched=%d", \
1368 EVER_MATCHED_SOMETHING (reg_info[this_reg])); \
1369 DEBUG_PRINT1 ("\n"); \
1370 PUSH_FAILURE_ELT (reg_info[this_reg].word); \
1373 DEBUG_PRINT2 (" Pushing low active reg: %ld\n", lowest_active_reg);\
1374 PUSH_FAILURE_INT (lowest_active_reg); \
1376 DEBUG_PRINT2 (" Pushing high active reg: %ld\n", highest_active_reg);\
1377 PUSH_FAILURE_INT (highest_active_reg); \
1379 DEBUG_PRINT2 (" Pushing pattern %p:\n", pattern_place); \
1380 DEBUG_PRINT_COMPILED_PATTERN (bufp, pattern_place, pend); \
1381 PUSH_FAILURE_POINTER (pattern_place); \
1383 DEBUG_PRINT2 (" Pushing string %p: `", string_place); \
1384 DEBUG_PRINT_DOUBLE_STRING (string_place, string1, size1, string2, \
1386 DEBUG_PRINT1 ("'\n"); \
1387 PUSH_FAILURE_POINTER (string_place); \
1389 DEBUG_PRINT2 (" Pushing failure id: %u\n", failure_id); \
1390 DEBUG_PUSH (failure_id); \
1393 /* This is the number of items that are pushed and popped on the stack
1394 for each register. */
1395 #define NUM_REG_ITEMS 3
1397 /* Individual items aside from the registers. */
1399 # define NUM_NONREG_ITEMS 5 /* Includes failure point id. */
1401 # define NUM_NONREG_ITEMS 4
1404 /* We push at most this many items on the stack. */
1405 /* We used to use (num_regs - 1), which is the number of registers
1406 this regexp will save; but that was changed to 5
1407 to avoid stack overflow for a regexp with lots of parens. */
1408 #define MAX_FAILURE_ITEMS (5 * NUM_REG_ITEMS + NUM_NONREG_ITEMS)
1410 /* We actually push this many items. */
1411 #define NUM_FAILURE_ITEMS \
1413 ? 0 : highest_active_reg - lowest_active_reg + 1) \
1417 /* How many items can still be added to the stack without overflowing it. */
1418 #define REMAINING_AVAIL_SLOTS ((fail_stack).size - (fail_stack).avail)
1421 /* Pops what PUSH_FAIL_STACK pushes.
1423 We restore into the parameters, all of which should be lvalues:
1424 STR -- the saved data position.
1425 PAT -- the saved pattern position.
1426 LOW_REG, HIGH_REG -- the highest and lowest active registers.
1427 REGSTART, REGEND -- arrays of string positions.
1428 REG_INFO -- array of information about each subexpression.
1430 Also assumes the variables `fail_stack' and (if debugging), `bufp',
1431 `pend', `string1', `size1', `string2', and `size2'. */
1433 #define POP_FAILURE_POINT(str, pat, low_reg, high_reg, regstart, regend, reg_info)\
1435 DEBUG_STATEMENT (unsigned failure_id;) \
1436 active_reg_t this_reg; \
1437 const unsigned char *string_temp; \
1439 assert (!FAIL_STACK_EMPTY ()); \
1441 /* Remove failure points and point to how many regs pushed. */ \
1442 DEBUG_PRINT1 ("POP_FAILURE_POINT:\n"); \
1443 DEBUG_PRINT2 (" Before pop, next avail: %d\n", fail_stack.avail); \
1444 DEBUG_PRINT2 (" size: %d\n", fail_stack.size); \
1446 assert (fail_stack.avail >= NUM_NONREG_ITEMS); \
1448 DEBUG_POP (&failure_id); \
1449 DEBUG_PRINT2 (" Popping failure id: %u\n", failure_id); \
1451 /* If the saved string location is NULL, it came from an \
1452 on_failure_keep_string_jump opcode, and we want to throw away the \
1453 saved NULL, thus retaining our current position in the string. */ \
1454 string_temp = POP_FAILURE_POINTER (); \
1455 if (string_temp != NULL) \
1456 str = (const char *) string_temp; \
1458 DEBUG_PRINT2 (" Popping string %p: `", str); \
1459 DEBUG_PRINT_DOUBLE_STRING (str, string1, size1, string2, size2); \
1460 DEBUG_PRINT1 ("'\n"); \
1462 pat = (unsigned char *) POP_FAILURE_POINTER (); \
1463 DEBUG_PRINT2 (" Popping pattern %p:\n", pat); \
1464 DEBUG_PRINT_COMPILED_PATTERN (bufp, pat, pend); \
1466 /* Restore register info. */ \
1467 high_reg = (active_reg_t) POP_FAILURE_INT (); \
1468 DEBUG_PRINT2 (" Popping high active reg: %ld\n", high_reg); \
1470 low_reg = (active_reg_t) POP_FAILURE_INT (); \
1471 DEBUG_PRINT2 (" Popping low active reg: %ld\n", low_reg); \
1474 for (this_reg = high_reg; this_reg >= low_reg; this_reg--) \
1476 DEBUG_PRINT2 (" Popping reg: %ld\n", this_reg); \
1478 reg_info[this_reg].word = POP_FAILURE_ELT (); \
1479 DEBUG_PRINT2 (" info: %p\n", \
1480 reg_info[this_reg].word.pointer); \
1482 regend[this_reg] = (const char *) POP_FAILURE_POINTER (); \
1483 DEBUG_PRINT2 (" end: %p\n", regend[this_reg]); \
1485 regstart[this_reg] = (const char *) POP_FAILURE_POINTER (); \
1486 DEBUG_PRINT2 (" start: %p\n", regstart[this_reg]); \
1490 for (this_reg = highest_active_reg; this_reg > high_reg; this_reg--) \
1492 reg_info[this_reg].word.integer = 0; \
1493 regend[this_reg] = 0; \
1494 regstart[this_reg] = 0; \
1496 highest_active_reg = high_reg; \
1499 set_regs_matched_done = 0; \
1500 DEBUG_STATEMENT (nfailure_points_popped++); \
1501 } /* POP_FAILURE_POINT */
1505 /* Structure for per-register (a.k.a. per-group) information.
1506 Other register information, such as the
1507 starting and ending positions (which are addresses), and the list of
1508 inner groups (which is a bits list) are maintained in separate
1511 We are making a (strictly speaking) nonportable assumption here: that
1512 the compiler will pack our bit fields into something that fits into
1513 the type of `word', i.e., is something that fits into one item on the
1517 /* Declarations and macros for re_match_2. */
1521 fail_stack_elt_t word
;
1524 /* This field is one if this group can match the empty string,
1525 zero if not. If not yet determined, `MATCH_NULL_UNSET_VALUE'. */
1526 #define MATCH_NULL_UNSET_VALUE 3
1527 unsigned match_null_string_p
: 2;
1528 unsigned is_active
: 1;
1529 unsigned matched_something
: 1;
1530 unsigned ever_matched_something
: 1;
1532 } register_info_type
;
1534 #define REG_MATCH_NULL_STRING_P(R) ((R).bits.match_null_string_p)
1535 #define IS_ACTIVE(R) ((R).bits.is_active)
1536 #define MATCHED_SOMETHING(R) ((R).bits.matched_something)
1537 #define EVER_MATCHED_SOMETHING(R) ((R).bits.ever_matched_something)
1540 /* Call this when have matched a real character; it sets `matched' flags
1541 for the subexpressions which we are currently inside. Also records
1542 that those subexprs have matched. */
1543 #define SET_REGS_MATCHED() \
1546 if (!set_regs_matched_done) \
1549 set_regs_matched_done = 1; \
1550 for (r = lowest_active_reg; r <= highest_active_reg; r++) \
1552 MATCHED_SOMETHING (reg_info[r]) \
1553 = EVER_MATCHED_SOMETHING (reg_info[r]) \
1560 /* Registers are set to a sentinel when they haven't yet matched. */
1561 static char reg_unset_dummy
;
1562 #define REG_UNSET_VALUE (®_unset_dummy)
1563 #define REG_UNSET(e) ((e) == REG_UNSET_VALUE)
1565 /* Subroutine declarations and macros for regex_compile. */
1567 static reg_errcode_t regex_compile
_RE_ARGS ((const char *pattern
, size_t size
,
1568 reg_syntax_t syntax
,
1569 struct re_pattern_buffer
*bufp
));
1570 static void store_op1
_RE_ARGS ((re_opcode_t op
, unsigned char *loc
, int arg
));
1571 static void store_op2
_RE_ARGS ((re_opcode_t op
, unsigned char *loc
,
1572 int arg1
, int arg2
));
1573 static void insert_op1
_RE_ARGS ((re_opcode_t op
, unsigned char *loc
,
1574 int arg
, unsigned char *end
));
1575 static void insert_op2
_RE_ARGS ((re_opcode_t op
, unsigned char *loc
,
1576 int arg1
, int arg2
, unsigned char *end
));
1577 static boolean at_begline_loc_p
_RE_ARGS ((const char *pattern
, const char *p
,
1578 reg_syntax_t syntax
));
1579 static boolean at_endline_loc_p
_RE_ARGS ((const char *p
, const char *pend
,
1580 reg_syntax_t syntax
));
1581 static reg_errcode_t compile_range
_RE_ARGS ((unsigned int range_start
,
1585 reg_syntax_t syntax
,
1588 /* Fetch the next character in the uncompiled pattern---translating it
1589 if necessary. Also cast from a signed character in the constant
1590 string passed to us by the user to an unsigned char that we can use
1591 as an array index (in, e.g., `translate'). */
1593 # define PATFETCH(c) \
1594 do {if (p == pend) return REG_EEND; \
1595 c = (unsigned char) *p++; \
1596 if (translate) c = (unsigned char) translate[c]; \
1600 /* Fetch the next character in the uncompiled pattern, with no
1602 #define PATFETCH_RAW(c) \
1603 do {if (p == pend) return REG_EEND; \
1604 c = (unsigned char) *p++; \
1607 /* Go backwards one character in the pattern. */
1608 #define PATUNFETCH p--
1611 /* If `translate' is non-null, return translate[D], else just D. We
1612 cast the subscript to translate because some data is declared as
1613 `char *', to avoid warnings when a string constant is passed. But
1614 when we use a character as a subscript we must make it unsigned. */
1616 # define TRANSLATE(d) \
1617 (translate ? (char) translate[(unsigned char) (d)] : (d))
1621 /* Macros for outputting the compiled pattern into `buffer'. */
1623 /* If the buffer isn't allocated when it comes in, use this. */
1624 #define INIT_BUF_SIZE 32
1626 /* Make sure we have at least N more bytes of space in buffer. */
1627 #define GET_BUFFER_SPACE(n) \
1628 while ((unsigned long) (b - bufp->buffer + (n)) > bufp->allocated) \
1631 /* Make sure we have one more byte of buffer space and then add C to it. */
1632 #define BUF_PUSH(c) \
1634 GET_BUFFER_SPACE (1); \
1635 *b++ = (unsigned char) (c); \
1639 /* Ensure we have two more bytes of buffer space and then append C1 and C2. */
1640 #define BUF_PUSH_2(c1, c2) \
1642 GET_BUFFER_SPACE (2); \
1643 *b++ = (unsigned char) (c1); \
1644 *b++ = (unsigned char) (c2); \
1648 /* As with BUF_PUSH_2, except for three bytes. */
1649 #define BUF_PUSH_3(c1, c2, c3) \
1651 GET_BUFFER_SPACE (3); \
1652 *b++ = (unsigned char) (c1); \
1653 *b++ = (unsigned char) (c2); \
1654 *b++ = (unsigned char) (c3); \
1658 /* Store a jump with opcode OP at LOC to location TO. We store a
1659 relative address offset by the three bytes the jump itself occupies. */
1660 #define STORE_JUMP(op, loc, to) \
1661 store_op1 (op, loc, (int) ((to) - (loc) - 3))
1663 /* Likewise, for a two-argument jump. */
1664 #define STORE_JUMP2(op, loc, to, arg) \
1665 store_op2 (op, loc, (int) ((to) - (loc) - 3), arg)
1667 /* Like `STORE_JUMP', but for inserting. Assume `b' is the buffer end. */
1668 #define INSERT_JUMP(op, loc, to) \
1669 insert_op1 (op, loc, (int) ((to) - (loc) - 3), b)
1671 /* Like `STORE_JUMP2', but for inserting. Assume `b' is the buffer end. */
1672 #define INSERT_JUMP2(op, loc, to, arg) \
1673 insert_op2 (op, loc, (int) ((to) - (loc) - 3), arg, b)
1676 /* This is not an arbitrary limit: the arguments which represent offsets
1677 into the pattern are two bytes long. So if 2^16 bytes turns out to
1678 be too small, many things would have to change. */
1679 /* Any other compiler which, like MSC, has allocation limit below 2^16
1680 bytes will have to use approach similar to what was done below for
1681 MSC and drop MAX_BUF_SIZE a bit. Otherwise you may end up
1682 reallocating to 0 bytes. Such thing is not going to work too well.
1683 You have been warned!! */
1684 #if defined _MSC_VER && !defined WIN32
1685 /* Microsoft C 16-bit versions limit malloc to approx 65512 bytes.
1686 The REALLOC define eliminates a flurry of conversion warnings,
1687 but is not required. */
1688 # define MAX_BUF_SIZE 65500L
1689 # define REALLOC(p,s) realloc ((p), (size_t) (s))
1691 # define MAX_BUF_SIZE (1L << 16)
1692 # define REALLOC(p,s) realloc ((p), (s))
1695 /* Extend the buffer by twice its current size via realloc and
1696 reset the pointers that pointed into the old block to point to the
1697 correct places in the new one. If extending the buffer results in it
1698 being larger than MAX_BUF_SIZE, then flag memory exhausted. */
1699 #define EXTEND_BUFFER() \
1701 unsigned char *old_buffer = bufp->buffer; \
1702 if (bufp->allocated == MAX_BUF_SIZE) \
1704 bufp->allocated <<= 1; \
1705 if (bufp->allocated > MAX_BUF_SIZE) \
1706 bufp->allocated = MAX_BUF_SIZE; \
1707 bufp->buffer = (unsigned char *) REALLOC (bufp->buffer, bufp->allocated);\
1708 if (bufp->buffer == NULL) \
1709 return REG_ESPACE; \
1710 /* If the buffer moved, move all the pointers into it. */ \
1711 if (old_buffer != bufp->buffer) \
1713 b = (b - old_buffer) + bufp->buffer; \
1714 begalt = (begalt - old_buffer) + bufp->buffer; \
1715 if (fixup_alt_jump) \
1716 fixup_alt_jump = (fixup_alt_jump - old_buffer) + bufp->buffer;\
1718 laststart = (laststart - old_buffer) + bufp->buffer; \
1719 if (pending_exact) \
1720 pending_exact = (pending_exact - old_buffer) + bufp->buffer; \
1725 /* Since we have one byte reserved for the register number argument to
1726 {start,stop}_memory, the maximum number of groups we can report
1727 things about is what fits in that byte. */
1728 #define MAX_REGNUM 255
1730 /* But patterns can have more than `MAX_REGNUM' registers. We just
1731 ignore the excess. */
1732 typedef unsigned regnum_t
;
1735 /* Macros for the compile stack. */
1737 /* Since offsets can go either forwards or backwards, this type needs to
1738 be able to hold values from -(MAX_BUF_SIZE - 1) to MAX_BUF_SIZE - 1. */
1739 /* int may be not enough when sizeof(int) == 2. */
1740 typedef long pattern_offset_t
;
1744 pattern_offset_t begalt_offset
;
1745 pattern_offset_t fixup_alt_jump
;
1746 pattern_offset_t inner_group_offset
;
1747 pattern_offset_t laststart_offset
;
1749 } compile_stack_elt_t
;
1754 compile_stack_elt_t
*stack
;
1756 unsigned avail
; /* Offset of next open position. */
1757 } compile_stack_type
;
1760 #define INIT_COMPILE_STACK_SIZE 32
1762 #define COMPILE_STACK_EMPTY (compile_stack.avail == 0)
1763 #define COMPILE_STACK_FULL (compile_stack.avail == compile_stack.size)
1765 /* The next available element. */
1766 #define COMPILE_STACK_TOP (compile_stack.stack[compile_stack.avail])
1769 /* Set the bit for character C in a list. */
1770 #define SET_LIST_BIT(c) \
1771 (b[((unsigned char) (c)) / BYTEWIDTH] \
1772 |= 1 << (((unsigned char) c) % BYTEWIDTH))
1775 /* Get the next unsigned number in the uncompiled pattern. */
1776 #define GET_UNSIGNED_NUMBER(num) \
1780 while ('0' <= c && c <= '9') \
1784 num = num * 10 + c - '0'; \
1792 #if defined _LIBC || WIDE_CHAR_SUPPORT
1793 /* The GNU C library provides support for user-defined character classes
1794 and the functions from ISO C amendement 1. */
1795 # ifdef CHARCLASS_NAME_MAX
1796 # define CHAR_CLASS_MAX_LENGTH CHARCLASS_NAME_MAX
1798 /* This shouldn't happen but some implementation might still have this
1799 problem. Use a reasonable default value. */
1800 # define CHAR_CLASS_MAX_LENGTH 256
1804 # define IS_CHAR_CLASS(string) __wctype (string)
1806 # define IS_CHAR_CLASS(string) wctype (string)
1809 # define CHAR_CLASS_MAX_LENGTH 6 /* Namely, `xdigit'. */
1811 # define IS_CHAR_CLASS(string) \
1812 (STREQ (string, "alpha") || STREQ (string, "upper") \
1813 || STREQ (string, "lower") || STREQ (string, "digit") \
1814 || STREQ (string, "alnum") || STREQ (string, "xdigit") \
1815 || STREQ (string, "space") || STREQ (string, "print") \
1816 || STREQ (string, "punct") || STREQ (string, "graph") \
1817 || STREQ (string, "cntrl") || STREQ (string, "blank"))
1820 #ifndef MATCH_MAY_ALLOCATE
1822 /* If we cannot allocate large objects within re_match_2_internal,
1823 we make the fail stack and register vectors global.
1824 The fail stack, we grow to the maximum size when a regexp
1826 The register vectors, we adjust in size each time we
1827 compile a regexp, according to the number of registers it needs. */
1829 static fail_stack_type fail_stack
;
1831 /* Size with which the following vectors are currently allocated.
1832 That is so we can make them bigger as needed,
1833 but never make them smaller. */
1834 static int regs_allocated_size
;
1836 static const char ** regstart
, ** regend
;
1837 static const char ** old_regstart
, ** old_regend
;
1838 static const char **best_regstart
, **best_regend
;
1839 static register_info_type
*reg_info
;
1840 static const char **reg_dummy
;
1841 static register_info_type
*reg_info_dummy
;
1843 /* Make the register vectors big enough for NUM_REGS registers,
1844 but don't make them smaller. */
1847 regex_grow_registers (num_regs
)
1850 if (num_regs
> regs_allocated_size
)
1852 RETALLOC_IF (regstart
, num_regs
, const char *);
1853 RETALLOC_IF (regend
, num_regs
, const char *);
1854 RETALLOC_IF (old_regstart
, num_regs
, const char *);
1855 RETALLOC_IF (old_regend
, num_regs
, const char *);
1856 RETALLOC_IF (best_regstart
, num_regs
, const char *);
1857 RETALLOC_IF (best_regend
, num_regs
, const char *);
1858 RETALLOC_IF (reg_info
, num_regs
, register_info_type
);
1859 RETALLOC_IF (reg_dummy
, num_regs
, const char *);
1860 RETALLOC_IF (reg_info_dummy
, num_regs
, register_info_type
);
1862 regs_allocated_size
= num_regs
;
1866 #endif /* not MATCH_MAY_ALLOCATE */
1868 static boolean group_in_compile_stack
_RE_ARGS ((compile_stack_type
1872 /* `regex_compile' compiles PATTERN (of length SIZE) according to SYNTAX.
1873 Returns one of error codes defined in `regex.h', or zero for success.
1875 Assumes the `allocated' (and perhaps `buffer') and `translate'
1876 fields are set in BUFP on entry.
1878 If it succeeds, results are put in BUFP (if it returns an error, the
1879 contents of BUFP are undefined):
1880 `buffer' is the compiled pattern;
1881 `syntax' is set to SYNTAX;
1882 `used' is set to the length of the compiled pattern;
1883 `fastmap_accurate' is zero;
1884 `re_nsub' is the number of subexpressions in PATTERN;
1885 `not_bol' and `not_eol' are zero;
1887 The `fastmap' and `newline_anchor' fields are neither
1888 examined nor set. */
1890 /* Return, freeing storage we allocated. */
1891 #define FREE_STACK_RETURN(value) \
1892 return (free (compile_stack.stack), value)
1894 static reg_errcode_t
1895 regex_compile (pattern
, size
, syntax
, bufp
)
1896 const char *pattern
;
1898 reg_syntax_t syntax
;
1899 struct re_pattern_buffer
*bufp
;
1901 /* We fetch characters from PATTERN here. Even though PATTERN is
1902 `char *' (i.e., signed), we declare these variables as unsigned, so
1903 they can be reliably used as array indices. */
1904 register unsigned char c
, c1
;
1906 /* A random temporary spot in PATTERN. */
1909 /* Points to the end of the buffer, where we should append. */
1910 register unsigned char *b
;
1912 /* Keeps track of unclosed groups. */
1913 compile_stack_type compile_stack
;
1915 /* Points to the current (ending) position in the pattern. */
1916 const char *p
= pattern
;
1917 const char *pend
= pattern
+ size
;
1919 /* How to translate the characters in the pattern. */
1920 RE_TRANSLATE_TYPE translate
= bufp
->translate
;
1922 /* Address of the count-byte of the most recently inserted `exactn'
1923 command. This makes it possible to tell if a new exact-match
1924 character can be added to that command or if the character requires
1925 a new `exactn' command. */
1926 unsigned char *pending_exact
= 0;
1928 /* Address of start of the most recently finished expression.
1929 This tells, e.g., postfix * where to find the start of its
1930 operand. Reset at the beginning of groups and alternatives. */
1931 unsigned char *laststart
= 0;
1933 /* Address of beginning of regexp, or inside of last group. */
1934 unsigned char *begalt
;
1936 /* Place in the uncompiled pattern (i.e., the {) to
1937 which to go back if the interval is invalid. */
1938 const char *beg_interval
;
1940 /* Address of the place where a forward jump should go to the end of
1941 the containing expression. Each alternative of an `or' -- except the
1942 last -- ends with a forward jump of this sort. */
1943 unsigned char *fixup_alt_jump
= 0;
1945 /* Counts open-groups as they are encountered. Remembered for the
1946 matching close-group on the compile stack, so the same register
1947 number is put in the stop_memory as the start_memory. */
1948 regnum_t regnum
= 0;
1951 DEBUG_PRINT1 ("\nCompiling pattern: ");
1954 unsigned debug_count
;
1956 for (debug_count
= 0; debug_count
< size
; debug_count
++)
1957 putchar (pattern
[debug_count
]);
1962 /* Initialize the compile stack. */
1963 compile_stack
.stack
= TALLOC (INIT_COMPILE_STACK_SIZE
, compile_stack_elt_t
);
1964 if (compile_stack
.stack
== NULL
)
1967 compile_stack
.size
= INIT_COMPILE_STACK_SIZE
;
1968 compile_stack
.avail
= 0;
1970 /* Initialize the pattern buffer. */
1971 bufp
->syntax
= syntax
;
1972 bufp
->fastmap_accurate
= 0;
1973 bufp
->not_bol
= bufp
->not_eol
= 0;
1975 /* Set `used' to zero, so that if we return an error, the pattern
1976 printer (for debugging) will think there's no pattern. We reset it
1980 /* Always count groups, whether or not bufp->no_sub is set. */
1983 #if !defined emacs && !defined SYNTAX_TABLE
1984 /* Initialize the syntax table. */
1985 init_syntax_once ();
1988 if (bufp
->allocated
== 0)
1991 { /* If zero allocated, but buffer is non-null, try to realloc
1992 enough space. This loses if buffer's address is bogus, but
1993 that is the user's responsibility. */
1994 RETALLOC (bufp
->buffer
, INIT_BUF_SIZE
, unsigned char);
1997 { /* Caller did not allocate a buffer. Do it for them. */
1998 bufp
->buffer
= TALLOC (INIT_BUF_SIZE
, unsigned char);
2000 if (!bufp
->buffer
) FREE_STACK_RETURN (REG_ESPACE
);
2002 bufp
->allocated
= INIT_BUF_SIZE
;
2005 begalt
= b
= bufp
->buffer
;
2007 /* Loop through the uncompiled pattern until we're at the end. */
2016 if ( /* If at start of pattern, it's an operator. */
2018 /* If context independent, it's an operator. */
2019 || syntax
& RE_CONTEXT_INDEP_ANCHORS
2020 /* Otherwise, depends on what's come before. */
2021 || at_begline_loc_p (pattern
, p
, syntax
))
2031 if ( /* If at end of pattern, it's an operator. */
2033 /* If context independent, it's an operator. */
2034 || syntax
& RE_CONTEXT_INDEP_ANCHORS
2035 /* Otherwise, depends on what's next. */
2036 || at_endline_loc_p (p
, pend
, syntax
))
2046 if ((syntax
& RE_BK_PLUS_QM
)
2047 || (syntax
& RE_LIMITED_OPS
))
2051 /* If there is no previous pattern... */
2054 if (syntax
& RE_CONTEXT_INVALID_OPS
)
2055 FREE_STACK_RETURN (REG_BADRPT
);
2056 else if (!(syntax
& RE_CONTEXT_INDEP_OPS
))
2061 /* Are we optimizing this jump? */
2062 boolean keep_string_p
= false;
2064 /* 1 means zero (many) matches is allowed. */
2065 char zero_times_ok
= 0, many_times_ok
= 0;
2067 /* If there is a sequence of repetition chars, collapse it
2068 down to just one (the right one). We can't combine
2069 interval operators with these because of, e.g., `a{2}*',
2070 which should only match an even number of `a's. */
2074 zero_times_ok
|= c
!= '+';
2075 many_times_ok
|= c
!= '?';
2083 || (!(syntax
& RE_BK_PLUS_QM
) && (c
== '+' || c
== '?')))
2086 else if (syntax
& RE_BK_PLUS_QM
&& c
== '\\')
2088 if (p
== pend
) FREE_STACK_RETURN (REG_EESCAPE
);
2091 if (!(c1
== '+' || c1
== '?'))
2106 /* If we get here, we found another repeat character. */
2109 /* Star, etc. applied to an empty pattern is equivalent
2110 to an empty pattern. */
2114 /* Now we know whether or not zero matches is allowed
2115 and also whether or not two or more matches is allowed. */
2117 { /* More than one repetition is allowed, so put in at the
2118 end a backward relative jump from `b' to before the next
2119 jump we're going to put in below (which jumps from
2120 laststart to after this jump).
2122 But if we are at the `*' in the exact sequence `.*\n',
2123 insert an unconditional jump backwards to the .,
2124 instead of the beginning of the loop. This way we only
2125 push a failure point once, instead of every time
2126 through the loop. */
2127 assert (p
- 1 > pattern
);
2129 /* Allocate the space for the jump. */
2130 GET_BUFFER_SPACE (3);
2132 /* We know we are not at the first character of the pattern,
2133 because laststart was nonzero. And we've already
2134 incremented `p', by the way, to be the character after
2135 the `*'. Do we have to do something analogous here
2136 for null bytes, because of RE_DOT_NOT_NULL? */
2137 if (TRANSLATE (*(p
- 2)) == TRANSLATE ('.')
2139 && p
< pend
&& TRANSLATE (*p
) == TRANSLATE ('\n')
2140 && !(syntax
& RE_DOT_NEWLINE
))
2141 { /* We have .*\n. */
2142 STORE_JUMP (jump
, b
, laststart
);
2143 keep_string_p
= true;
2146 /* Anything else. */
2147 STORE_JUMP (maybe_pop_jump
, b
, laststart
- 3);
2149 /* We've added more stuff to the buffer. */
2153 /* On failure, jump from laststart to b + 3, which will be the
2154 end of the buffer after this jump is inserted. */
2155 GET_BUFFER_SPACE (3);
2156 INSERT_JUMP (keep_string_p
? on_failure_keep_string_jump
2164 /* At least one repetition is required, so insert a
2165 `dummy_failure_jump' before the initial
2166 `on_failure_jump' instruction of the loop. This
2167 effects a skip over that instruction the first time
2168 we hit that loop. */
2169 GET_BUFFER_SPACE (3);
2170 INSERT_JUMP (dummy_failure_jump
, laststart
, laststart
+ 6);
2185 boolean had_char_class
= false;
2186 unsigned int range_start
= 0xffffffff;
2188 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2190 /* Ensure that we have enough space to push a charset: the
2191 opcode, the length count, and the bitset; 34 bytes in all. */
2192 GET_BUFFER_SPACE (34);
2196 /* We test `*p == '^' twice, instead of using an if
2197 statement, so we only need one BUF_PUSH. */
2198 BUF_PUSH (*p
== '^' ? charset_not
: charset
);
2202 /* Remember the first position in the bracket expression. */
2205 /* Push the number of bytes in the bitmap. */
2206 BUF_PUSH ((1 << BYTEWIDTH
) / BYTEWIDTH
);
2208 /* Clear the whole map. */
2209 bzero (b
, (1 << BYTEWIDTH
) / BYTEWIDTH
);
2211 /* charset_not matches newline according to a syntax bit. */
2212 if ((re_opcode_t
) b
[-2] == charset_not
2213 && (syntax
& RE_HAT_LISTS_NOT_NEWLINE
))
2214 SET_LIST_BIT ('\n');
2216 /* Read in characters and ranges, setting map bits. */
2219 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2223 /* \ might escape characters inside [...] and [^...]. */
2224 if ((syntax
& RE_BACKSLASH_ESCAPE_IN_LISTS
) && c
== '\\')
2226 if (p
== pend
) FREE_STACK_RETURN (REG_EESCAPE
);
2234 /* Could be the end of the bracket expression. If it's
2235 not (i.e., when the bracket expression is `[]' so
2236 far), the ']' character bit gets set way below. */
2237 if (c
== ']' && p
!= p1
+ 1)
2240 /* Look ahead to see if it's a range when the last thing
2241 was a character class. */
2242 if (had_char_class
&& c
== '-' && *p
!= ']')
2243 FREE_STACK_RETURN (REG_ERANGE
);
2245 /* Look ahead to see if it's a range when the last thing
2246 was a character: if this is a hyphen not at the
2247 beginning or the end of a list, then it's the range
2250 && !(p
- 2 >= pattern
&& p
[-2] == '[')
2251 && !(p
- 3 >= pattern
&& p
[-3] == '[' && p
[-2] == '^')
2255 = compile_range (range_start
, &p
, pend
, translate
,
2257 if (ret
!= REG_NOERROR
) FREE_STACK_RETURN (ret
);
2258 range_start
= 0xffffffff;
2261 else if (p
[0] == '-' && p
[1] != ']')
2262 { /* This handles ranges made up of characters only. */
2265 /* Move past the `-'. */
2268 ret
= compile_range (c
, &p
, pend
, translate
, syntax
, b
);
2269 if (ret
!= REG_NOERROR
) FREE_STACK_RETURN (ret
);
2270 range_start
= 0xffffffff;
2273 /* See if we're at the beginning of a possible character
2276 else if (syntax
& RE_CHAR_CLASSES
&& c
== '[' && *p
== ':')
2277 { /* Leave room for the null. */
2278 char str
[CHAR_CLASS_MAX_LENGTH
+ 1];
2283 /* If pattern is `[[:'. */
2284 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2289 if ((c
== ':' && *p
== ']') || p
== pend
)
2291 if (c1
< CHAR_CLASS_MAX_LENGTH
)
2294 /* This is in any case an invalid class name. */
2299 /* If isn't a word bracketed by `[:' and `:]':
2300 undo the ending character, the letters, and leave
2301 the leading `:' and `[' (but set bits for them). */
2302 if (c
== ':' && *p
== ']')
2304 #if defined _LIBC || WIDE_CHAR_SUPPORT
2305 boolean is_lower
= STREQ (str
, "lower");
2306 boolean is_upper
= STREQ (str
, "upper");
2310 wt
= IS_CHAR_CLASS (str
);
2312 FREE_STACK_RETURN (REG_ECTYPE
);
2314 /* Throw away the ] at the end of the character
2318 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2320 for (ch
= 0; ch
< 1 << BYTEWIDTH
; ++ch
)
2323 if (__iswctype (__btowc (ch
), wt
))
2326 if (iswctype (btowc (ch
), wt
))
2330 if (translate
&& (is_upper
|| is_lower
)
2331 && (ISUPPER (ch
) || ISLOWER (ch
)))
2335 had_char_class
= true;
2338 boolean is_alnum
= STREQ (str
, "alnum");
2339 boolean is_alpha
= STREQ (str
, "alpha");
2340 boolean is_blank
= STREQ (str
, "blank");
2341 boolean is_cntrl
= STREQ (str
, "cntrl");
2342 boolean is_digit
= STREQ (str
, "digit");
2343 boolean is_graph
= STREQ (str
, "graph");
2344 boolean is_lower
= STREQ (str
, "lower");
2345 boolean is_print
= STREQ (str
, "print");
2346 boolean is_punct
= STREQ (str
, "punct");
2347 boolean is_space
= STREQ (str
, "space");
2348 boolean is_upper
= STREQ (str
, "upper");
2349 boolean is_xdigit
= STREQ (str
, "xdigit");
2351 if (!IS_CHAR_CLASS (str
))
2352 FREE_STACK_RETURN (REG_ECTYPE
);
2354 /* Throw away the ] at the end of the character
2358 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2360 for (ch
= 0; ch
< 1 << BYTEWIDTH
; ch
++)
2362 /* This was split into 3 if's to
2363 avoid an arbitrary limit in some compiler. */
2364 if ( (is_alnum
&& ISALNUM (ch
))
2365 || (is_alpha
&& ISALPHA (ch
))
2366 || (is_blank
&& ISBLANK (ch
))
2367 || (is_cntrl
&& ISCNTRL (ch
)))
2369 if ( (is_digit
&& ISDIGIT (ch
))
2370 || (is_graph
&& ISGRAPH (ch
))
2371 || (is_lower
&& ISLOWER (ch
))
2372 || (is_print
&& ISPRINT (ch
)))
2374 if ( (is_punct
&& ISPUNCT (ch
))
2375 || (is_space
&& ISSPACE (ch
))
2376 || (is_upper
&& ISUPPER (ch
))
2377 || (is_xdigit
&& ISXDIGIT (ch
)))
2379 if ( translate
&& (is_upper
|| is_lower
)
2380 && (ISUPPER (ch
) || ISLOWER (ch
)))
2383 had_char_class
= true;
2384 #endif /* libc || wctype.h */
2394 had_char_class
= false;
2397 else if (syntax
& RE_CHAR_CLASSES
&& c
== '[' && *p
== '=')
2399 unsigned char str
[MB_LEN_MAX
+ 1];
2402 _NL_CURRENT_WORD (LC_COLLATE
, _NL_COLLATE_NRULES
);
2408 /* If pattern is `[[='. */
2409 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2414 if ((c
== '=' && *p
== ']') || p
== pend
)
2416 if (c1
< MB_LEN_MAX
)
2419 /* This is in any case an invalid class name. */
2424 if (c
== '=' && *p
== ']' && str
[0] != '\0')
2426 /* If we have no collation data we use the default
2427 collation in which each character is in a class
2428 by itself. It also means that ASCII is the
2429 character set and therefore we cannot have character
2430 with more than one byte in the multibyte
2437 FREE_STACK_RETURN (REG_ECOLLATE
);
2439 /* Throw away the ] at the end of the equivalence
2443 /* Set the bit for the character. */
2444 SET_LIST_BIT (str
[0]);
2449 /* Try to match the byte sequence in `str' against
2450 those known to the collate implementation.
2451 First find out whether the bytes in `str' are
2452 actually from exactly one character. */
2453 const int32_t *table
;
2454 const unsigned char *weights
;
2455 const unsigned char *extra
;
2456 const int32_t *indirect
;
2458 const unsigned char *cp
= str
;
2461 /* This #include defines a local function! */
2462 # include <locale/weight.h>
2464 table
= (const int32_t *)
2465 _NL_CURRENT (LC_COLLATE
, _NL_COLLATE_TABLEMB
);
2466 weights
= (const unsigned char *)
2467 _NL_CURRENT (LC_COLLATE
, _NL_COLLATE_WEIGHTMB
);
2468 extra
= (const unsigned char *)
2469 _NL_CURRENT (LC_COLLATE
, _NL_COLLATE_EXTRAMB
);
2470 indirect
= (const int32_t *)
2471 _NL_CURRENT (LC_COLLATE
, _NL_COLLATE_INDIRECTMB
);
2473 idx
= findidx (&cp
);
2474 if (idx
== 0 || cp
< str
+ c1
)
2475 /* This is no valid character. */
2476 FREE_STACK_RETURN (REG_ECOLLATE
);
2478 /* Throw away the ] at the end of the equivalence
2482 /* Now we have to go throught the whole table
2483 and find all characters which have the same
2486 XXX Note that this is not entirely correct.
2487 we would have to match multibyte sequences
2488 but this is not possible with the current
2490 for (ch
= 1; ch
< 256; ++ch
)
2491 /* XXX This test would have to be changed if we
2492 would allow matching multibyte sequences. */
2495 int32_t idx2
= table
[ch
];
2496 size_t len
= weights
[idx2
];
2498 /* Test whether the lenghts match. */
2499 if (weights
[idx
] == len
)
2501 /* They do. New compare the bytes of
2506 && (weights
[idx
+ 1 + cnt
]
2507 == weights
[idx2
+ 1 + cnt
]))
2511 /* They match. Mark the character as
2518 had_char_class
= true;
2528 had_char_class
= false;
2531 else if (syntax
& RE_CHAR_CLASSES
&& c
== '[' && *p
== '.')
2533 unsigned char str
[128]; /* Should be large enough. */
2536 _NL_CURRENT_WORD (LC_COLLATE
, _NL_COLLATE_NRULES
);
2542 /* If pattern is `[[='. */
2543 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2548 if ((c
== '.' && *p
== ']') || p
== pend
)
2550 if (c1
< sizeof (str
))
2553 /* This is in any case an invalid class name. */
2558 if (c
== '.' && *p
== ']' && str
[0] != '\0')
2560 /* If we have no collation data we use the default
2561 collation in which each character is the name
2562 for its own class which contains only the one
2563 character. It also means that ASCII is the
2564 character set and therefore we cannot have character
2565 with more than one byte in the multibyte
2572 FREE_STACK_RETURN (REG_ECOLLATE
);
2574 /* Throw away the ] at the end of the equivalence
2578 /* Set the bit for the character. */
2579 SET_LIST_BIT (str
[0]);
2580 range_start
= ((const unsigned char *) str
)[0];
2585 /* Try to match the byte sequence in `str' against
2586 those known to the collate implementation.
2587 First find out whether the bytes in `str' are
2588 actually from exactly one character. */
2590 const int32_t *symb_table
;
2591 const unsigned char *extra
;
2598 _NL_CURRENT_WORD (LC_COLLATE
,
2599 _NL_COLLATE_SYMB_HASH_SIZEMB
);
2600 symb_table
= (const int32_t *)
2601 _NL_CURRENT (LC_COLLATE
,
2602 _NL_COLLATE_SYMB_TABLEMB
);
2603 extra
= (const unsigned char *)
2604 _NL_CURRENT (LC_COLLATE
,
2605 _NL_COLLATE_SYMB_EXTRAMB
);
2607 /* Locate the character in the hashing table. */
2608 hash
= elem_hash (str
, c1
);
2611 elem
= hash
% table_size
;
2612 second
= hash
% (table_size
- 2);
2613 while (symb_table
[2 * elem
] != 0)
2615 /* First compare the hashing value. */
2616 if (symb_table
[2 * elem
] == hash
2617 && c1
== extra
[symb_table
[2 * elem
+ 1]]
2619 &extra
[symb_table
[2 * elem
+ 1]
2623 /* Yep, this is the entry. */
2624 idx
= symb_table
[2 * elem
+ 1];
2625 idx
+= 1 + extra
[idx
];
2633 if (symb_table
[2 * elem
] == 0)
2634 /* This is no valid character. */
2635 FREE_STACK_RETURN (REG_ECOLLATE
);
2637 /* Throw away the ] at the end of the equivalence
2641 /* Now add the multibyte character(s) we found
2642 to the acceptabed list.
2644 XXX Note that this is not entirely correct.
2645 we would have to match multibyte sequences
2646 but this is not possible with the current
2647 implementation. Also, we have to match
2648 collating symbols, which expand to more than
2649 one file, as a whole and not allow the
2650 individual bytes. */
2653 range_start
= extra
[idx
];
2655 SET_LIST_BIT (extra
[idx
++]);
2658 had_char_class
= false;
2668 had_char_class
= false;
2673 had_char_class
= false;
2679 /* Discard any (non)matching list bytes that are all 0 at the
2680 end of the map. Decrease the map-length byte too. */
2681 while ((int) b
[-1] > 0 && b
[b
[-1] - 1] == 0)
2689 if (syntax
& RE_NO_BK_PARENS
)
2696 if (syntax
& RE_NO_BK_PARENS
)
2703 if (syntax
& RE_NEWLINE_ALT
)
2710 if (syntax
& RE_NO_BK_VBAR
)
2717 if (syntax
& RE_INTERVALS
&& syntax
& RE_NO_BK_BRACES
)
2718 goto handle_interval
;
2724 if (p
== pend
) FREE_STACK_RETURN (REG_EESCAPE
);
2726 /* Do not translate the character after the \, so that we can
2727 distinguish, e.g., \B from \b, even if we normally would
2728 translate, e.g., B to b. */
2734 if (syntax
& RE_NO_BK_PARENS
)
2735 goto normal_backslash
;
2741 if (COMPILE_STACK_FULL
)
2743 RETALLOC (compile_stack
.stack
, compile_stack
.size
<< 1,
2744 compile_stack_elt_t
);
2745 if (compile_stack
.stack
== NULL
) return REG_ESPACE
;
2747 compile_stack
.size
<<= 1;
2750 /* These are the values to restore when we hit end of this
2751 group. They are all relative offsets, so that if the
2752 whole pattern moves because of realloc, they will still
2754 COMPILE_STACK_TOP
.begalt_offset
= begalt
- bufp
->buffer
;
2755 COMPILE_STACK_TOP
.fixup_alt_jump
2756 = fixup_alt_jump
? fixup_alt_jump
- bufp
->buffer
+ 1 : 0;
2757 COMPILE_STACK_TOP
.laststart_offset
= b
- bufp
->buffer
;
2758 COMPILE_STACK_TOP
.regnum
= regnum
;
2760 /* We will eventually replace the 0 with the number of
2761 groups inner to this one. But do not push a
2762 start_memory for groups beyond the last one we can
2763 represent in the compiled pattern. */
2764 if (regnum
<= MAX_REGNUM
)
2766 COMPILE_STACK_TOP
.inner_group_offset
= b
- bufp
->buffer
+ 2;
2767 BUF_PUSH_3 (start_memory
, regnum
, 0);
2770 compile_stack
.avail
++;
2775 /* If we've reached MAX_REGNUM groups, then this open
2776 won't actually generate any code, so we'll have to
2777 clear pending_exact explicitly. */
2783 if (syntax
& RE_NO_BK_PARENS
) goto normal_backslash
;
2785 if (COMPILE_STACK_EMPTY
)
2787 if (syntax
& RE_UNMATCHED_RIGHT_PAREN_ORD
)
2788 goto normal_backslash
;
2790 FREE_STACK_RETURN (REG_ERPAREN
);
2795 { /* Push a dummy failure point at the end of the
2796 alternative for a possible future
2797 `pop_failure_jump' to pop. See comments at
2798 `push_dummy_failure' in `re_match_2'. */
2799 BUF_PUSH (push_dummy_failure
);
2801 /* We allocated space for this jump when we assigned
2802 to `fixup_alt_jump', in the `handle_alt' case below. */
2803 STORE_JUMP (jump_past_alt
, fixup_alt_jump
, b
- 1);
2806 /* See similar code for backslashed left paren above. */
2807 if (COMPILE_STACK_EMPTY
)
2809 if (syntax
& RE_UNMATCHED_RIGHT_PAREN_ORD
)
2812 FREE_STACK_RETURN (REG_ERPAREN
);
2815 /* Since we just checked for an empty stack above, this
2816 ``can't happen''. */
2817 assert (compile_stack
.avail
!= 0);
2819 /* We don't just want to restore into `regnum', because
2820 later groups should continue to be numbered higher,
2821 as in `(ab)c(de)' -- the second group is #2. */
2822 regnum_t this_group_regnum
;
2824 compile_stack
.avail
--;
2825 begalt
= bufp
->buffer
+ COMPILE_STACK_TOP
.begalt_offset
;
2827 = COMPILE_STACK_TOP
.fixup_alt_jump
2828 ? bufp
->buffer
+ COMPILE_STACK_TOP
.fixup_alt_jump
- 1
2830 laststart
= bufp
->buffer
+ COMPILE_STACK_TOP
.laststart_offset
;
2831 this_group_regnum
= COMPILE_STACK_TOP
.regnum
;
2832 /* If we've reached MAX_REGNUM groups, then this open
2833 won't actually generate any code, so we'll have to
2834 clear pending_exact explicitly. */
2837 /* We're at the end of the group, so now we know how many
2838 groups were inside this one. */
2839 if (this_group_regnum
<= MAX_REGNUM
)
2841 unsigned char *inner_group_loc
2842 = bufp
->buffer
+ COMPILE_STACK_TOP
.inner_group_offset
;
2844 *inner_group_loc
= regnum
- this_group_regnum
;
2845 BUF_PUSH_3 (stop_memory
, this_group_regnum
,
2846 regnum
- this_group_regnum
);
2852 case '|': /* `\|'. */
2853 if (syntax
& RE_LIMITED_OPS
|| syntax
& RE_NO_BK_VBAR
)
2854 goto normal_backslash
;
2856 if (syntax
& RE_LIMITED_OPS
)
2859 /* Insert before the previous alternative a jump which
2860 jumps to this alternative if the former fails. */
2861 GET_BUFFER_SPACE (3);
2862 INSERT_JUMP (on_failure_jump
, begalt
, b
+ 6);
2866 /* The alternative before this one has a jump after it
2867 which gets executed if it gets matched. Adjust that
2868 jump so it will jump to this alternative's analogous
2869 jump (put in below, which in turn will jump to the next
2870 (if any) alternative's such jump, etc.). The last such
2871 jump jumps to the correct final destination. A picture:
2877 If we are at `b', then fixup_alt_jump right now points to a
2878 three-byte space after `a'. We'll put in the jump, set
2879 fixup_alt_jump to right after `b', and leave behind three
2880 bytes which we'll fill in when we get to after `c'. */
2883 STORE_JUMP (jump_past_alt
, fixup_alt_jump
, b
);
2885 /* Mark and leave space for a jump after this alternative,
2886 to be filled in later either by next alternative or
2887 when know we're at the end of a series of alternatives. */
2889 GET_BUFFER_SPACE (3);
2898 /* If \{ is a literal. */
2899 if (!(syntax
& RE_INTERVALS
)
2900 /* If we're at `\{' and it's not the open-interval
2902 || (syntax
& RE_NO_BK_BRACES
)
2903 || (p
- 2 == pattern
&& p
== pend
))
2904 goto normal_backslash
;
2908 /* If got here, then the syntax allows intervals. */
2910 /* At least (most) this many matches must be made. */
2911 int lower_bound
= -1, upper_bound
= -1;
2913 beg_interval
= p
- 1;
2917 if (syntax
& RE_NO_BK_BRACES
)
2918 goto unfetch_interval
;
2920 FREE_STACK_RETURN (REG_EBRACE
);
2923 GET_UNSIGNED_NUMBER (lower_bound
);
2927 GET_UNSIGNED_NUMBER (upper_bound
);
2928 if (upper_bound
< 0)
2929 upper_bound
= RE_DUP_MAX
;
2932 /* Interval such as `{1}' => match exactly once. */
2933 upper_bound
= lower_bound
;
2935 if (lower_bound
< 0 || upper_bound
> RE_DUP_MAX
2936 || lower_bound
> upper_bound
)
2938 if (syntax
& RE_NO_BK_BRACES
)
2939 goto unfetch_interval
;
2941 FREE_STACK_RETURN (REG_BADBR
);
2944 if (!(syntax
& RE_NO_BK_BRACES
))
2946 if (c
!= '\\') FREE_STACK_RETURN (REG_EBRACE
);
2953 if (syntax
& RE_NO_BK_BRACES
)
2954 goto unfetch_interval
;
2956 FREE_STACK_RETURN (REG_BADBR
);
2959 /* We just parsed a valid interval. */
2961 /* If it's invalid to have no preceding re. */
2964 if (syntax
& RE_CONTEXT_INVALID_OPS
)
2965 FREE_STACK_RETURN (REG_BADRPT
);
2966 else if (syntax
& RE_CONTEXT_INDEP_OPS
)
2969 goto unfetch_interval
;
2972 /* If the upper bound is zero, don't want to succeed at
2973 all; jump from `laststart' to `b + 3', which will be
2974 the end of the buffer after we insert the jump. */
2975 if (upper_bound
== 0)
2977 GET_BUFFER_SPACE (3);
2978 INSERT_JUMP (jump
, laststart
, b
+ 3);
2982 /* Otherwise, we have a nontrivial interval. When
2983 we're all done, the pattern will look like:
2984 set_number_at <jump count> <upper bound>
2985 set_number_at <succeed_n count> <lower bound>
2986 succeed_n <after jump addr> <succeed_n count>
2988 jump_n <succeed_n addr> <jump count>
2989 (The upper bound and `jump_n' are omitted if
2990 `upper_bound' is 1, though.) */
2992 { /* If the upper bound is > 1, we need to insert
2993 more at the end of the loop. */
2994 unsigned nbytes
= 10 + (upper_bound
> 1) * 10;
2996 GET_BUFFER_SPACE (nbytes
);
2998 /* Initialize lower bound of the `succeed_n', even
2999 though it will be set during matching by its
3000 attendant `set_number_at' (inserted next),
3001 because `re_compile_fastmap' needs to know.
3002 Jump to the `jump_n' we might insert below. */
3003 INSERT_JUMP2 (succeed_n
, laststart
,
3004 b
+ 5 + (upper_bound
> 1) * 5,
3008 /* Code to initialize the lower bound. Insert
3009 before the `succeed_n'. The `5' is the last two
3010 bytes of this `set_number_at', plus 3 bytes of
3011 the following `succeed_n'. */
3012 insert_op2 (set_number_at
, laststart
, 5, lower_bound
, b
);
3015 if (upper_bound
> 1)
3016 { /* More than one repetition is allowed, so
3017 append a backward jump to the `succeed_n'
3018 that starts this interval.
3020 When we've reached this during matching,
3021 we'll have matched the interval once, so
3022 jump back only `upper_bound - 1' times. */
3023 STORE_JUMP2 (jump_n
, b
, laststart
+ 5,
3027 /* The location we want to set is the second
3028 parameter of the `jump_n'; that is `b-2' as
3029 an absolute address. `laststart' will be
3030 the `set_number_at' we're about to insert;
3031 `laststart+3' the number to set, the source
3032 for the relative address. But we are
3033 inserting into the middle of the pattern --
3034 so everything is getting moved up by 5.
3035 Conclusion: (b - 2) - (laststart + 3) + 5,
3036 i.e., b - laststart.
3038 We insert this at the beginning of the loop
3039 so that if we fail during matching, we'll
3040 reinitialize the bounds. */
3041 insert_op2 (set_number_at
, laststart
, b
- laststart
,
3042 upper_bound
- 1, b
);
3047 beg_interval
= NULL
;
3052 /* If an invalid interval, match the characters as literals. */
3053 assert (beg_interval
);
3055 beg_interval
= NULL
;
3057 /* normal_char and normal_backslash need `c'. */
3060 if (!(syntax
& RE_NO_BK_BRACES
))
3062 if (p
> pattern
&& p
[-1] == '\\')
3063 goto normal_backslash
;
3068 /* There is no way to specify the before_dot and after_dot
3069 operators. rms says this is ok. --karl */
3077 BUF_PUSH_2 (syntaxspec
, syntax_spec_code
[c
]);
3083 BUF_PUSH_2 (notsyntaxspec
, syntax_spec_code
[c
]);
3089 if (syntax
& RE_NO_GNU_OPS
)
3092 BUF_PUSH (wordchar
);
3097 if (syntax
& RE_NO_GNU_OPS
)
3100 BUF_PUSH (notwordchar
);
3105 if (syntax
& RE_NO_GNU_OPS
)
3111 if (syntax
& RE_NO_GNU_OPS
)
3117 if (syntax
& RE_NO_GNU_OPS
)
3119 BUF_PUSH (wordbound
);
3123 if (syntax
& RE_NO_GNU_OPS
)
3125 BUF_PUSH (notwordbound
);
3129 if (syntax
& RE_NO_GNU_OPS
)
3135 if (syntax
& RE_NO_GNU_OPS
)
3140 case '1': case '2': case '3': case '4': case '5':
3141 case '6': case '7': case '8': case '9':
3142 if (syntax
& RE_NO_BK_REFS
)
3148 FREE_STACK_RETURN (REG_ESUBREG
);
3150 /* Can't back reference to a subexpression if inside of it. */
3151 if (group_in_compile_stack (compile_stack
, (regnum_t
) c1
))
3155 BUF_PUSH_2 (duplicate
, c1
);
3161 if (syntax
& RE_BK_PLUS_QM
)
3164 goto normal_backslash
;
3168 /* You might think it would be useful for \ to mean
3169 not to translate; but if we don't translate it
3170 it will never match anything. */
3178 /* Expects the character in `c'. */
3180 /* If no exactn currently being built. */
3183 /* If last exactn not at current position. */
3184 || pending_exact
+ *pending_exact
+ 1 != b
3186 /* We have only one byte following the exactn for the count. */
3187 || *pending_exact
== (1 << BYTEWIDTH
) - 1
3189 /* If followed by a repetition operator. */
3190 || *p
== '*' || *p
== '^'
3191 || ((syntax
& RE_BK_PLUS_QM
)
3192 ? *p
== '\\' && (p
[1] == '+' || p
[1] == '?')
3193 : (*p
== '+' || *p
== '?'))
3194 || ((syntax
& RE_INTERVALS
)
3195 && ((syntax
& RE_NO_BK_BRACES
)
3197 : (p
[0] == '\\' && p
[1] == '{'))))
3199 /* Start building a new exactn. */
3203 BUF_PUSH_2 (exactn
, 0);
3204 pending_exact
= b
- 1;
3211 } /* while p != pend */
3214 /* Through the pattern now. */
3217 STORE_JUMP (jump_past_alt
, fixup_alt_jump
, b
);
3219 if (!COMPILE_STACK_EMPTY
)
3220 FREE_STACK_RETURN (REG_EPAREN
);
3222 /* If we don't want backtracking, force success
3223 the first time we reach the end of the compiled pattern. */
3224 if (syntax
& RE_NO_POSIX_BACKTRACKING
)
3227 free (compile_stack
.stack
);
3229 /* We have succeeded; set the length of the buffer. */
3230 bufp
->used
= b
- bufp
->buffer
;
3235 DEBUG_PRINT1 ("\nCompiled pattern: \n");
3236 print_compiled_pattern (bufp
);
3240 #ifndef MATCH_MAY_ALLOCATE
3241 /* Initialize the failure stack to the largest possible stack. This
3242 isn't necessary unless we're trying to avoid calling alloca in
3243 the search and match routines. */
3245 int num_regs
= bufp
->re_nsub
+ 1;
3247 /* Since DOUBLE_FAIL_STACK refuses to double only if the current size
3248 is strictly greater than re_max_failures, the largest possible stack
3249 is 2 * re_max_failures failure points. */
3250 if (fail_stack
.size
< (2 * re_max_failures
* MAX_FAILURE_ITEMS
))
3252 fail_stack
.size
= (2 * re_max_failures
* MAX_FAILURE_ITEMS
);
3255 if (! fail_stack
.stack
)
3257 = (fail_stack_elt_t
*) xmalloc (fail_stack
.size
3258 * sizeof (fail_stack_elt_t
));
3261 = (fail_stack_elt_t
*) xrealloc (fail_stack
.stack
,
3263 * sizeof (fail_stack_elt_t
)));
3264 # else /* not emacs */
3265 if (! fail_stack
.stack
)
3267 = (fail_stack_elt_t
*) malloc (fail_stack
.size
3268 * sizeof (fail_stack_elt_t
));
3271 = (fail_stack_elt_t
*) realloc (fail_stack
.stack
,
3273 * sizeof (fail_stack_elt_t
)));
3274 # endif /* not emacs */
3277 regex_grow_registers (num_regs
);
3279 #endif /* not MATCH_MAY_ALLOCATE */
3282 } /* regex_compile */
3284 /* Subroutines for `regex_compile'. */
3286 /* Store OP at LOC followed by two-byte integer parameter ARG. */
3289 store_op1 (op
, loc
, arg
)
3294 *loc
= (unsigned char) op
;
3295 STORE_NUMBER (loc
+ 1, arg
);
3299 /* Like `store_op1', but for two two-byte parameters ARG1 and ARG2. */
3302 store_op2 (op
, loc
, arg1
, arg2
)
3307 *loc
= (unsigned char) op
;
3308 STORE_NUMBER (loc
+ 1, arg1
);
3309 STORE_NUMBER (loc
+ 3, arg2
);
3313 /* Copy the bytes from LOC to END to open up three bytes of space at LOC
3314 for OP followed by two-byte integer parameter ARG. */
3317 insert_op1 (op
, loc
, arg
, end
)
3323 register unsigned char *pfrom
= end
;
3324 register unsigned char *pto
= end
+ 3;
3326 while (pfrom
!= loc
)
3329 store_op1 (op
, loc
, arg
);
3333 /* Like `insert_op1', but for two two-byte parameters ARG1 and ARG2. */
3336 insert_op2 (op
, loc
, arg1
, arg2
, end
)
3342 register unsigned char *pfrom
= end
;
3343 register unsigned char *pto
= end
+ 5;
3345 while (pfrom
!= loc
)
3348 store_op2 (op
, loc
, arg1
, arg2
);
3352 /* P points to just after a ^ in PATTERN. Return true if that ^ comes
3353 after an alternative or a begin-subexpression. We assume there is at
3354 least one character before the ^. */
3357 at_begline_loc_p (pattern
, p
, syntax
)
3358 const char *pattern
, *p
;
3359 reg_syntax_t syntax
;
3361 const char *prev
= p
- 2;
3362 boolean prev_prev_backslash
= prev
> pattern
&& prev
[-1] == '\\';
3365 /* After a subexpression? */
3366 (*prev
== '(' && (syntax
& RE_NO_BK_PARENS
|| prev_prev_backslash
))
3367 /* After an alternative? */
3368 || (*prev
== '|' && (syntax
& RE_NO_BK_VBAR
|| prev_prev_backslash
));
3372 /* The dual of at_begline_loc_p. This one is for $. We assume there is
3373 at least one character after the $, i.e., `P < PEND'. */
3376 at_endline_loc_p (p
, pend
, syntax
)
3377 const char *p
, *pend
;
3378 reg_syntax_t syntax
;
3380 const char *next
= p
;
3381 boolean next_backslash
= *next
== '\\';
3382 const char *next_next
= p
+ 1 < pend
? p
+ 1 : 0;
3385 /* Before a subexpression? */
3386 (syntax
& RE_NO_BK_PARENS
? *next
== ')'
3387 : next_backslash
&& next_next
&& *next_next
== ')')
3388 /* Before an alternative? */
3389 || (syntax
& RE_NO_BK_VBAR
? *next
== '|'
3390 : next_backslash
&& next_next
&& *next_next
== '|');
3394 /* Returns true if REGNUM is in one of COMPILE_STACK's elements and
3395 false if it's not. */
3398 group_in_compile_stack (compile_stack
, regnum
)
3399 compile_stack_type compile_stack
;
3404 for (this_element
= compile_stack
.avail
- 1;
3407 if (compile_stack
.stack
[this_element
].regnum
== regnum
)
3414 /* Read the ending character of a range (in a bracket expression) from the
3415 uncompiled pattern *P_PTR (which ends at PEND). We assume the
3416 starting character is in `P[-2]'. (`P[-1]' is the character `-'.)
3417 Then we set the translation of all bits between the starting and
3418 ending characters (inclusive) in the compiled pattern B.
3420 Return an error code.
3422 We use these short variable names so we can use the same macros as
3423 `regex_compile' itself. */
3425 static reg_errcode_t
3426 compile_range (range_start_char
, p_ptr
, pend
, translate
, syntax
, b
)
3427 unsigned int range_start_char
;
3428 const char **p_ptr
, *pend
;
3429 RE_TRANSLATE_TYPE translate
;
3430 reg_syntax_t syntax
;
3435 const char *p
= *p_ptr
;
3437 char range_start
[2];
3444 /* Fetch the endpoints without translating them; the
3445 appropriate translation is done in the bit-setting loop below. */
3446 range_start
[0] = range_start_char
;
3447 range_start
[1] = '\0';
3448 range_end
[0] = p
[0];
3449 range_end
[1] = '\0';
3451 /* Have to increment the pointer into the pattern string, so the
3452 caller isn't still at the ending character. */
3455 /* Report an error if the range is empty and the syntax prohibits this. */
3456 ret
= syntax
& RE_NO_EMPTY_RANGES
? REG_ERANGE
: REG_NOERROR
;
3458 /* Here we see why `this_char' has to be larger than an `unsigned
3459 char' -- we would otherwise go into an infinite loop, since all
3460 characters <= 0xff. */
3462 for (this_char
= 0; this_char
<= (unsigned char) -1; ++this_char
)
3465 if (strcoll (range_start
, ch
) <= 0 && strcoll (ch
, range_end
) <= 0)
3467 SET_LIST_BIT (TRANSLATE (this_char
));
3475 /* re_compile_fastmap computes a ``fastmap'' for the compiled pattern in
3476 BUFP. A fastmap records which of the (1 << BYTEWIDTH) possible
3477 characters can start a string that matches the pattern. This fastmap
3478 is used by re_search to skip quickly over impossible starting points.
3480 The caller must supply the address of a (1 << BYTEWIDTH)-byte data
3481 area as BUFP->fastmap.
3483 We set the `fastmap', `fastmap_accurate', and `can_be_null' fields in
3486 Returns 0 if we succeed, -2 if an internal error. */
3489 re_compile_fastmap (bufp
)
3490 struct re_pattern_buffer
*bufp
;
3493 #ifdef MATCH_MAY_ALLOCATE
3494 fail_stack_type fail_stack
;
3496 #ifndef REGEX_MALLOC
3500 register char *fastmap
= bufp
->fastmap
;
3501 unsigned char *pattern
= bufp
->buffer
;
3502 unsigned char *p
= pattern
;
3503 register unsigned char *pend
= pattern
+ bufp
->used
;
3506 /* This holds the pointer to the failure stack, when
3507 it is allocated relocatably. */
3508 fail_stack_elt_t
*failure_stack_ptr
;
3511 /* Assume that each path through the pattern can be null until
3512 proven otherwise. We set this false at the bottom of switch
3513 statement, to which we get only if a particular path doesn't
3514 match the empty string. */
3515 boolean path_can_be_null
= true;
3517 /* We aren't doing a `succeed_n' to begin with. */
3518 boolean succeed_n_p
= false;
3520 assert (fastmap
!= NULL
&& p
!= NULL
);
3523 bzero (fastmap
, 1 << BYTEWIDTH
); /* Assume nothing's valid. */
3524 bufp
->fastmap_accurate
= 1; /* It will be when we're done. */
3525 bufp
->can_be_null
= 0;
3529 if (p
== pend
|| *p
== succeed
)
3531 /* We have reached the (effective) end of pattern. */
3532 if (!FAIL_STACK_EMPTY ())
3534 bufp
->can_be_null
|= path_can_be_null
;
3536 /* Reset for next path. */
3537 path_can_be_null
= true;
3539 p
= fail_stack
.stack
[--fail_stack
.avail
].pointer
;
3547 /* We should never be about to go beyond the end of the pattern. */
3550 switch (SWITCH_ENUM_CAST ((re_opcode_t
) *p
++))
3553 /* I guess the idea here is to simply not bother with a fastmap
3554 if a backreference is used, since it's too hard to figure out
3555 the fastmap for the corresponding group. Setting
3556 `can_be_null' stops `re_search_2' from using the fastmap, so
3557 that is all we do. */
3559 bufp
->can_be_null
= 1;
3563 /* Following are the cases which match a character. These end
3572 for (j
= *p
++ * BYTEWIDTH
- 1; j
>= 0; j
--)
3573 if (p
[j
/ BYTEWIDTH
] & (1 << (j
% BYTEWIDTH
)))
3579 /* Chars beyond end of map must be allowed. */
3580 for (j
= *p
* BYTEWIDTH
; j
< (1 << BYTEWIDTH
); j
++)
3583 for (j
= *p
++ * BYTEWIDTH
- 1; j
>= 0; j
--)
3584 if (!(p
[j
/ BYTEWIDTH
] & (1 << (j
% BYTEWIDTH
))))
3590 for (j
= 0; j
< (1 << BYTEWIDTH
); j
++)
3591 if (SYNTAX (j
) == Sword
)
3597 for (j
= 0; j
< (1 << BYTEWIDTH
); j
++)
3598 if (SYNTAX (j
) != Sword
)
3605 int fastmap_newline
= fastmap
['\n'];
3607 /* `.' matches anything ... */
3608 for (j
= 0; j
< (1 << BYTEWIDTH
); j
++)
3611 /* ... except perhaps newline. */
3612 if (!(bufp
->syntax
& RE_DOT_NEWLINE
))
3613 fastmap
['\n'] = fastmap_newline
;
3615 /* Return if we have already set `can_be_null'; if we have,
3616 then the fastmap is irrelevant. Something's wrong here. */
3617 else if (bufp
->can_be_null
)
3620 /* Otherwise, have to check alternative paths. */
3627 for (j
= 0; j
< (1 << BYTEWIDTH
); j
++)
3628 if (SYNTAX (j
) == (enum syntaxcode
) k
)
3635 for (j
= 0; j
< (1 << BYTEWIDTH
); j
++)
3636 if (SYNTAX (j
) != (enum syntaxcode
) k
)
3641 /* All cases after this match the empty string. These end with
3661 case push_dummy_failure
:
3666 case pop_failure_jump
:
3667 case maybe_pop_jump
:
3670 case dummy_failure_jump
:
3671 EXTRACT_NUMBER_AND_INCR (j
, p
);
3676 /* Jump backward implies we just went through the body of a
3677 loop and matched nothing. Opcode jumped to should be
3678 `on_failure_jump' or `succeed_n'. Just treat it like an
3679 ordinary jump. For a * loop, it has pushed its failure
3680 point already; if so, discard that as redundant. */
3681 if ((re_opcode_t
) *p
!= on_failure_jump
3682 && (re_opcode_t
) *p
!= succeed_n
)
3686 EXTRACT_NUMBER_AND_INCR (j
, p
);
3689 /* If what's on the stack is where we are now, pop it. */
3690 if (!FAIL_STACK_EMPTY ()
3691 && fail_stack
.stack
[fail_stack
.avail
- 1].pointer
== p
)
3697 case on_failure_jump
:
3698 case on_failure_keep_string_jump
:
3699 handle_on_failure_jump
:
3700 EXTRACT_NUMBER_AND_INCR (j
, p
);
3702 /* For some patterns, e.g., `(a?)?', `p+j' here points to the
3703 end of the pattern. We don't want to push such a point,
3704 since when we restore it above, entering the switch will
3705 increment `p' past the end of the pattern. We don't need
3706 to push such a point since we obviously won't find any more
3707 fastmap entries beyond `pend'. Such a pattern can match
3708 the null string, though. */
3711 if (!PUSH_PATTERN_OP (p
+ j
, fail_stack
))
3713 RESET_FAIL_STACK ();
3718 bufp
->can_be_null
= 1;
3722 EXTRACT_NUMBER_AND_INCR (k
, p
); /* Skip the n. */
3723 succeed_n_p
= false;
3730 /* Get to the number of times to succeed. */
3733 /* Increment p past the n for when k != 0. */
3734 EXTRACT_NUMBER_AND_INCR (k
, p
);
3738 succeed_n_p
= true; /* Spaghetti code alert. */
3739 goto handle_on_failure_jump
;
3756 abort (); /* We have listed all the cases. */
3759 /* Getting here means we have found the possible starting
3760 characters for one path of the pattern -- and that the empty
3761 string does not match. We need not follow this path further.
3762 Instead, look at the next alternative (remembered on the
3763 stack), or quit if no more. The test at the top of the loop
3764 does these things. */
3765 path_can_be_null
= false;
3769 /* Set `can_be_null' for the last path (also the first path, if the
3770 pattern is empty). */
3771 bufp
->can_be_null
|= path_can_be_null
;
3774 RESET_FAIL_STACK ();
3776 } /* re_compile_fastmap */
3778 weak_alias (__re_compile_fastmap
, re_compile_fastmap
)
3781 /* Set REGS to hold NUM_REGS registers, storing them in STARTS and
3782 ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use
3783 this memory for recording register information. STARTS and ENDS
3784 must be allocated using the malloc library routine, and must each
3785 be at least NUM_REGS * sizeof (regoff_t) bytes long.
3787 If NUM_REGS == 0, then subsequent matches should allocate their own
3790 Unless this function is called, the first search or match using
3791 PATTERN_BUFFER will allocate its own register data, without
3792 freeing the old data. */
3795 re_set_registers (bufp
, regs
, num_regs
, starts
, ends
)
3796 struct re_pattern_buffer
*bufp
;
3797 struct re_registers
*regs
;
3799 regoff_t
*starts
, *ends
;
3803 bufp
->regs_allocated
= REGS_REALLOCATE
;
3804 regs
->num_regs
= num_regs
;
3805 regs
->start
= starts
;
3810 bufp
->regs_allocated
= REGS_UNALLOCATED
;
3812 regs
->start
= regs
->end
= (regoff_t
*) 0;
3816 weak_alias (__re_set_registers
, re_set_registers
)
3819 /* Searching routines. */
3821 /* Like re_search_2, below, but only one string is specified, and
3822 doesn't let you say where to stop matching. */
3825 re_search (bufp
, string
, size
, startpos
, range
, regs
)
3826 struct re_pattern_buffer
*bufp
;
3828 int size
, startpos
, range
;
3829 struct re_registers
*regs
;
3831 return re_search_2 (bufp
, NULL
, 0, string
, size
, startpos
, range
,
3835 weak_alias (__re_search
, re_search
)
3839 /* Using the compiled pattern in BUFP->buffer, first tries to match the
3840 virtual concatenation of STRING1 and STRING2, starting first at index
3841 STARTPOS, then at STARTPOS + 1, and so on.
3843 STRING1 and STRING2 have length SIZE1 and SIZE2, respectively.
3845 RANGE is how far to scan while trying to match. RANGE = 0 means try
3846 only at STARTPOS; in general, the last start tried is STARTPOS +
3849 In REGS, return the indices of the virtual concatenation of STRING1
3850 and STRING2 that matched the entire BUFP->buffer and its contained
3853 Do not consider matching one past the index STOP in the virtual
3854 concatenation of STRING1 and STRING2.
3856 We return either the position in the strings at which the match was
3857 found, -1 if no match, or -2 if error (such as failure
3861 re_search_2 (bufp
, string1
, size1
, string2
, size2
, startpos
, range
, regs
, stop
)
3862 struct re_pattern_buffer
*bufp
;
3863 const char *string1
, *string2
;
3867 struct re_registers
*regs
;
3871 register char *fastmap
= bufp
->fastmap
;
3872 register RE_TRANSLATE_TYPE translate
= bufp
->translate
;
3873 int total_size
= size1
+ size2
;
3874 int endpos
= startpos
+ range
;
3876 /* Check for out-of-range STARTPOS. */
3877 if (startpos
< 0 || startpos
> total_size
)
3880 /* Fix up RANGE if it might eventually take us outside
3881 the virtual concatenation of STRING1 and STRING2.
3882 Make sure we won't move STARTPOS below 0 or above TOTAL_SIZE. */
3884 range
= 0 - startpos
;
3885 else if (endpos
> total_size
)
3886 range
= total_size
- startpos
;
3888 /* If the search isn't to be a backwards one, don't waste time in a
3889 search for a pattern that must be anchored. */
3890 if (bufp
->used
> 0 && range
> 0
3891 && ((re_opcode_t
) bufp
->buffer
[0] == begbuf
3892 /* `begline' is like `begbuf' if it cannot match at newlines. */
3893 || ((re_opcode_t
) bufp
->buffer
[0] == begline
3894 && !bufp
->newline_anchor
)))
3903 /* In a forward search for something that starts with \=.
3904 don't keep searching past point. */
3905 if (bufp
->used
> 0 && (re_opcode_t
) bufp
->buffer
[0] == at_dot
&& range
> 0)
3907 range
= PT
- startpos
;
3913 /* Update the fastmap now if not correct already. */
3914 if (fastmap
&& !bufp
->fastmap_accurate
)
3915 if (re_compile_fastmap (bufp
) == -2)
3918 /* Loop through the string, looking for a place to start matching. */
3921 /* If a fastmap is supplied, skip quickly over characters that
3922 cannot be the start of a match. If the pattern can match the
3923 null string, however, we don't need to skip characters; we want
3924 the first null string. */
3925 if (fastmap
&& startpos
< total_size
&& !bufp
->can_be_null
)
3927 if (range
> 0) /* Searching forwards. */
3929 register const char *d
;
3930 register int lim
= 0;
3933 if (startpos
< size1
&& startpos
+ range
>= size1
)
3934 lim
= range
- (size1
- startpos
);
3936 d
= (startpos
>= size1
? string2
- size1
: string1
) + startpos
;
3938 /* Written out as an if-else to avoid testing `translate'
3942 && !fastmap
[(unsigned char)
3943 translate
[(unsigned char) *d
++]])
3946 while (range
> lim
&& !fastmap
[(unsigned char) *d
++])
3949 startpos
+= irange
- range
;
3951 else /* Searching backwards. */
3953 register char c
= (size1
== 0 || startpos
>= size1
3954 ? string2
[startpos
- size1
]
3955 : string1
[startpos
]);
3957 if (!fastmap
[(unsigned char) TRANSLATE (c
)])
3962 /* If can't match the null string, and that's all we have left, fail. */
3963 if (range
>= 0 && startpos
== total_size
&& fastmap
3964 && !bufp
->can_be_null
)
3967 val
= re_match_2_internal (bufp
, string1
, size1
, string2
, size2
,
3968 startpos
, regs
, stop
);
3969 #ifndef REGEX_MALLOC
3998 weak_alias (__re_search_2
, re_search_2
)
4001 /* This converts PTR, a pointer into one of the search strings `string1'
4002 and `string2' into an offset from the beginning of that string. */
4003 #define POINTER_TO_OFFSET(ptr) \
4004 (FIRST_STRING_P (ptr) \
4005 ? ((regoff_t) ((ptr) - string1)) \
4006 : ((regoff_t) ((ptr) - string2 + size1)))
4008 /* Macros for dealing with the split strings in re_match_2. */
4010 #define MATCHING_IN_FIRST_STRING (dend == end_match_1)
4012 /* Call before fetching a character with *d. This switches over to
4013 string2 if necessary. */
4014 #define PREFETCH() \
4017 /* End of string2 => fail. */ \
4018 if (dend == end_match_2) \
4020 /* End of string1 => advance to string2. */ \
4022 dend = end_match_2; \
4026 /* Test if at very beginning or at very end of the virtual concatenation
4027 of `string1' and `string2'. If only one string, it's `string2'. */
4028 #define AT_STRINGS_BEG(d) ((d) == (size1 ? string1 : string2) || !size2)
4029 #define AT_STRINGS_END(d) ((d) == end2)
4032 /* Test if D points to a character which is word-constituent. We have
4033 two special cases to check for: if past the end of string1, look at
4034 the first character in string2; and if before the beginning of
4035 string2, look at the last character in string1. */
4036 #define WORDCHAR_P(d) \
4037 (SYNTAX ((d) == end1 ? *string2 \
4038 : (d) == string2 - 1 ? *(end1 - 1) : *(d)) \
4041 /* Disabled due to a compiler bug -- see comment at case wordbound */
4043 /* Test if the character before D and the one at D differ with respect
4044 to being word-constituent. */
4045 #define AT_WORD_BOUNDARY(d) \
4046 (AT_STRINGS_BEG (d) || AT_STRINGS_END (d) \
4047 || WORDCHAR_P (d - 1) != WORDCHAR_P (d))
4050 /* Free everything we malloc. */
4051 #ifdef MATCH_MAY_ALLOCATE
4052 # define FREE_VAR(var) if (var) REGEX_FREE (var); var = NULL
4053 # define FREE_VARIABLES() \
4055 REGEX_FREE_STACK (fail_stack.stack); \
4056 FREE_VAR (regstart); \
4057 FREE_VAR (regend); \
4058 FREE_VAR (old_regstart); \
4059 FREE_VAR (old_regend); \
4060 FREE_VAR (best_regstart); \
4061 FREE_VAR (best_regend); \
4062 FREE_VAR (reg_info); \
4063 FREE_VAR (reg_dummy); \
4064 FREE_VAR (reg_info_dummy); \
4067 # define FREE_VARIABLES() ((void)0) /* Do nothing! But inhibit gcc warning. */
4068 #endif /* not MATCH_MAY_ALLOCATE */
4070 /* These values must meet several constraints. They must not be valid
4071 register values; since we have a limit of 255 registers (because
4072 we use only one byte in the pattern for the register number), we can
4073 use numbers larger than 255. They must differ by 1, because of
4074 NUM_FAILURE_ITEMS above. And the value for the lowest register must
4075 be larger than the value for the highest register, so we do not try
4076 to actually save any registers when none are active. */
4077 #define NO_HIGHEST_ACTIVE_REG (1 << BYTEWIDTH)
4078 #define NO_LOWEST_ACTIVE_REG (NO_HIGHEST_ACTIVE_REG + 1)
4080 /* Matching routines. */
4082 #ifndef emacs /* Emacs never uses this. */
4083 /* re_match is like re_match_2 except it takes only a single string. */
4086 re_match (bufp
, string
, size
, pos
, regs
)
4087 struct re_pattern_buffer
*bufp
;
4090 struct re_registers
*regs
;
4092 int result
= re_match_2_internal (bufp
, NULL
, 0, string
, size
,
4094 # ifndef REGEX_MALLOC
4102 weak_alias (__re_match
, re_match
)
4104 #endif /* not emacs */
4106 static boolean group_match_null_string_p
_RE_ARGS ((unsigned char **p
,
4108 register_info_type
*reg_info
));
4109 static boolean alt_match_null_string_p
_RE_ARGS ((unsigned char *p
,
4111 register_info_type
*reg_info
));
4112 static boolean common_op_match_null_string_p
_RE_ARGS ((unsigned char **p
,
4114 register_info_type
*reg_info
));
4115 static int bcmp_translate
_RE_ARGS ((const char *s1
, const char *s2
,
4116 int len
, char *translate
));
4118 /* re_match_2 matches the compiled pattern in BUFP against the
4119 the (virtual) concatenation of STRING1 and STRING2 (of length SIZE1
4120 and SIZE2, respectively). We start matching at POS, and stop
4123 If REGS is non-null and the `no_sub' field of BUFP is nonzero, we
4124 store offsets for the substring each group matched in REGS. See the
4125 documentation for exactly how many groups we fill.
4127 We return -1 if no match, -2 if an internal error (such as the
4128 failure stack overflowing). Otherwise, we return the length of the
4129 matched substring. */
4132 re_match_2 (bufp
, string1
, size1
, string2
, size2
, pos
, regs
, stop
)
4133 struct re_pattern_buffer
*bufp
;
4134 const char *string1
, *string2
;
4137 struct re_registers
*regs
;
4140 int result
= re_match_2_internal (bufp
, string1
, size1
, string2
, size2
,
4142 #ifndef REGEX_MALLOC
4150 weak_alias (__re_match_2
, re_match_2
)
4153 /* This is a separate function so that we can force an alloca cleanup
4156 re_match_2_internal (bufp
, string1
, size1
, string2
, size2
, pos
, regs
, stop
)
4157 struct re_pattern_buffer
*bufp
;
4158 const char *string1
, *string2
;
4161 struct re_registers
*regs
;
4164 /* General temporaries. */
4168 /* Just past the end of the corresponding string. */
4169 const char *end1
, *end2
;
4171 /* Pointers into string1 and string2, just past the last characters in
4172 each to consider matching. */
4173 const char *end_match_1
, *end_match_2
;
4175 /* Where we are in the data, and the end of the current string. */
4176 const char *d
, *dend
;
4178 /* Where we are in the pattern, and the end of the pattern. */
4179 unsigned char *p
= bufp
->buffer
;
4180 register unsigned char *pend
= p
+ bufp
->used
;
4182 /* Mark the opcode just after a start_memory, so we can test for an
4183 empty subpattern when we get to the stop_memory. */
4184 unsigned char *just_past_start_mem
= 0;
4186 /* We use this to map every character in the string. */
4187 RE_TRANSLATE_TYPE translate
= bufp
->translate
;
4189 /* Failure point stack. Each place that can handle a failure further
4190 down the line pushes a failure point on this stack. It consists of
4191 restart, regend, and reg_info for all registers corresponding to
4192 the subexpressions we're currently inside, plus the number of such
4193 registers, and, finally, two char *'s. The first char * is where
4194 to resume scanning the pattern; the second one is where to resume
4195 scanning the strings. If the latter is zero, the failure point is
4196 a ``dummy''; if a failure happens and the failure point is a dummy,
4197 it gets discarded and the next next one is tried. */
4198 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
4199 fail_stack_type fail_stack
;
4202 static unsigned failure_id
;
4203 unsigned nfailure_points_pushed
= 0, nfailure_points_popped
= 0;
4207 /* This holds the pointer to the failure stack, when
4208 it is allocated relocatably. */
4209 fail_stack_elt_t
*failure_stack_ptr
;
4212 /* We fill all the registers internally, independent of what we
4213 return, for use in backreferences. The number here includes
4214 an element for register zero. */
4215 size_t num_regs
= bufp
->re_nsub
+ 1;
4217 /* The currently active registers. */
4218 active_reg_t lowest_active_reg
= NO_LOWEST_ACTIVE_REG
;
4219 active_reg_t highest_active_reg
= NO_HIGHEST_ACTIVE_REG
;
4221 /* Information on the contents of registers. These are pointers into
4222 the input strings; they record just what was matched (on this
4223 attempt) by a subexpression part of the pattern, that is, the
4224 regnum-th regstart pointer points to where in the pattern we began
4225 matching and the regnum-th regend points to right after where we
4226 stopped matching the regnum-th subexpression. (The zeroth register
4227 keeps track of what the whole pattern matches.) */
4228 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
4229 const char **regstart
, **regend
;
4232 /* If a group that's operated upon by a repetition operator fails to
4233 match anything, then the register for its start will need to be
4234 restored because it will have been set to wherever in the string we
4235 are when we last see its open-group operator. Similarly for a
4237 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
4238 const char **old_regstart
, **old_regend
;
4241 /* The is_active field of reg_info helps us keep track of which (possibly
4242 nested) subexpressions we are currently in. The matched_something
4243 field of reg_info[reg_num] helps us tell whether or not we have
4244 matched any of the pattern so far this time through the reg_num-th
4245 subexpression. These two fields get reset each time through any
4246 loop their register is in. */
4247 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
4248 register_info_type
*reg_info
;
4251 /* The following record the register info as found in the above
4252 variables when we find a match better than any we've seen before.
4253 This happens as we backtrack through the failure points, which in
4254 turn happens only if we have not yet matched the entire string. */
4255 unsigned best_regs_set
= false;
4256 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
4257 const char **best_regstart
, **best_regend
;
4260 /* Logically, this is `best_regend[0]'. But we don't want to have to
4261 allocate space for that if we're not allocating space for anything
4262 else (see below). Also, we never need info about register 0 for
4263 any of the other register vectors, and it seems rather a kludge to
4264 treat `best_regend' differently than the rest. So we keep track of
4265 the end of the best match so far in a separate variable. We
4266 initialize this to NULL so that when we backtrack the first time
4267 and need to test it, it's not garbage. */
4268 const char *match_end
= NULL
;
4270 /* This helps SET_REGS_MATCHED avoid doing redundant work. */
4271 int set_regs_matched_done
= 0;
4273 /* Used when we pop values we don't care about. */
4274 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
4275 const char **reg_dummy
;
4276 register_info_type
*reg_info_dummy
;
4280 /* Counts the total number of registers pushed. */
4281 unsigned num_regs_pushed
= 0;
4284 DEBUG_PRINT1 ("\n\nEntering re_match_2.\n");
4288 #ifdef MATCH_MAY_ALLOCATE
4289 /* Do not bother to initialize all the register variables if there are
4290 no groups in the pattern, as it takes a fair amount of time. If
4291 there are groups, we include space for register 0 (the whole
4292 pattern), even though we never use it, since it simplifies the
4293 array indexing. We should fix this. */
4296 regstart
= REGEX_TALLOC (num_regs
, const char *);
4297 regend
= REGEX_TALLOC (num_regs
, const char *);
4298 old_regstart
= REGEX_TALLOC (num_regs
, const char *);
4299 old_regend
= REGEX_TALLOC (num_regs
, const char *);
4300 best_regstart
= REGEX_TALLOC (num_regs
, const char *);
4301 best_regend
= REGEX_TALLOC (num_regs
, const char *);
4302 reg_info
= REGEX_TALLOC (num_regs
, register_info_type
);
4303 reg_dummy
= REGEX_TALLOC (num_regs
, const char *);
4304 reg_info_dummy
= REGEX_TALLOC (num_regs
, register_info_type
);
4306 if (!(regstart
&& regend
&& old_regstart
&& old_regend
&& reg_info
4307 && best_regstart
&& best_regend
&& reg_dummy
&& reg_info_dummy
))
4315 /* We must initialize all our variables to NULL, so that
4316 `FREE_VARIABLES' doesn't try to free them. */
4317 regstart
= regend
= old_regstart
= old_regend
= best_regstart
4318 = best_regend
= reg_dummy
= NULL
;
4319 reg_info
= reg_info_dummy
= (register_info_type
*) NULL
;
4321 #endif /* MATCH_MAY_ALLOCATE */
4323 /* The starting position is bogus. */
4324 if (pos
< 0 || pos
> size1
+ size2
)
4330 /* Initialize subexpression text positions to -1 to mark ones that no
4331 start_memory/stop_memory has been seen for. Also initialize the
4332 register information struct. */
4333 for (mcnt
= 1; (unsigned) mcnt
< num_regs
; mcnt
++)
4335 regstart
[mcnt
] = regend
[mcnt
]
4336 = old_regstart
[mcnt
] = old_regend
[mcnt
] = REG_UNSET_VALUE
;
4338 REG_MATCH_NULL_STRING_P (reg_info
[mcnt
]) = MATCH_NULL_UNSET_VALUE
;
4339 IS_ACTIVE (reg_info
[mcnt
]) = 0;
4340 MATCHED_SOMETHING (reg_info
[mcnt
]) = 0;
4341 EVER_MATCHED_SOMETHING (reg_info
[mcnt
]) = 0;
4344 /* We move `string1' into `string2' if the latter's empty -- but not if
4345 `string1' is null. */
4346 if (size2
== 0 && string1
!= NULL
)
4353 end1
= string1
+ size1
;
4354 end2
= string2
+ size2
;
4356 /* Compute where to stop matching, within the two strings. */
4359 end_match_1
= string1
+ stop
;
4360 end_match_2
= string2
;
4365 end_match_2
= string2
+ stop
- size1
;
4368 /* `p' scans through the pattern as `d' scans through the data.
4369 `dend' is the end of the input string that `d' points within. `d'
4370 is advanced into the following input string whenever necessary, but
4371 this happens before fetching; therefore, at the beginning of the
4372 loop, `d' can be pointing at the end of a string, but it cannot
4374 if (size1
> 0 && pos
<= size1
)
4381 d
= string2
+ pos
- size1
;
4385 DEBUG_PRINT1 ("The compiled pattern is:\n");
4386 DEBUG_PRINT_COMPILED_PATTERN (bufp
, p
, pend
);
4387 DEBUG_PRINT1 ("The string to match is: `");
4388 DEBUG_PRINT_DOUBLE_STRING (d
, string1
, size1
, string2
, size2
);
4389 DEBUG_PRINT1 ("'\n");
4391 /* This loops over pattern commands. It exits by returning from the
4392 function if the match is complete, or it drops through if the match
4393 fails at this starting point in the input data. */
4397 DEBUG_PRINT2 ("\n%p: ", p
);
4399 DEBUG_PRINT2 ("\n0x%x: ", p
);
4403 { /* End of pattern means we might have succeeded. */
4404 DEBUG_PRINT1 ("end of pattern ... ");
4406 /* If we haven't matched the entire string, and we want the
4407 longest match, try backtracking. */
4408 if (d
!= end_match_2
)
4410 /* 1 if this match ends in the same string (string1 or string2)
4411 as the best previous match. */
4412 boolean same_str_p
= (FIRST_STRING_P (match_end
)
4413 == MATCHING_IN_FIRST_STRING
);
4414 /* 1 if this match is the best seen so far. */
4415 boolean best_match_p
;
4417 /* AIX compiler got confused when this was combined
4418 with the previous declaration. */
4420 best_match_p
= d
> match_end
;
4422 best_match_p
= !MATCHING_IN_FIRST_STRING
;
4424 DEBUG_PRINT1 ("backtracking.\n");
4426 if (!FAIL_STACK_EMPTY ())
4427 { /* More failure points to try. */
4429 /* If exceeds best match so far, save it. */
4430 if (!best_regs_set
|| best_match_p
)
4432 best_regs_set
= true;
4435 DEBUG_PRINT1 ("\nSAVING match as best so far.\n");
4437 for (mcnt
= 1; (unsigned) mcnt
< num_regs
; mcnt
++)
4439 best_regstart
[mcnt
] = regstart
[mcnt
];
4440 best_regend
[mcnt
] = regend
[mcnt
];
4446 /* If no failure points, don't restore garbage. And if
4447 last match is real best match, don't restore second
4449 else if (best_regs_set
&& !best_match_p
)
4452 /* Restore best match. It may happen that `dend ==
4453 end_match_1' while the restored d is in string2.
4454 For example, the pattern `x.*y.*z' against the
4455 strings `x-' and `y-z-', if the two strings are
4456 not consecutive in memory. */
4457 DEBUG_PRINT1 ("Restoring best registers.\n");
4460 dend
= ((d
>= string1
&& d
<= end1
)
4461 ? end_match_1
: end_match_2
);
4463 for (mcnt
= 1; (unsigned) mcnt
< num_regs
; mcnt
++)
4465 regstart
[mcnt
] = best_regstart
[mcnt
];
4466 regend
[mcnt
] = best_regend
[mcnt
];
4469 } /* d != end_match_2 */
4472 DEBUG_PRINT1 ("Accepting match.\n");
4474 /* If caller wants register contents data back, do it. */
4475 if (regs
&& !bufp
->no_sub
)
4477 /* Have the register data arrays been allocated? */
4478 if (bufp
->regs_allocated
== REGS_UNALLOCATED
)
4479 { /* No. So allocate them with malloc. We need one
4480 extra element beyond `num_regs' for the `-1' marker
4482 regs
->num_regs
= MAX (RE_NREGS
, num_regs
+ 1);
4483 regs
->start
= TALLOC (regs
->num_regs
, regoff_t
);
4484 regs
->end
= TALLOC (regs
->num_regs
, regoff_t
);
4485 if (regs
->start
== NULL
|| regs
->end
== NULL
)
4490 bufp
->regs_allocated
= REGS_REALLOCATE
;
4492 else if (bufp
->regs_allocated
== REGS_REALLOCATE
)
4493 { /* Yes. If we need more elements than were already
4494 allocated, reallocate them. If we need fewer, just
4496 if (regs
->num_regs
< num_regs
+ 1)
4498 regs
->num_regs
= num_regs
+ 1;
4499 RETALLOC (regs
->start
, regs
->num_regs
, regoff_t
);
4500 RETALLOC (regs
->end
, regs
->num_regs
, regoff_t
);
4501 if (regs
->start
== NULL
|| regs
->end
== NULL
)
4510 /* These braces fend off a "empty body in an else-statement"
4511 warning under GCC when assert expands to nothing. */
4512 assert (bufp
->regs_allocated
== REGS_FIXED
);
4515 /* Convert the pointer data in `regstart' and `regend' to
4516 indices. Register zero has to be set differently,
4517 since we haven't kept track of any info for it. */
4518 if (regs
->num_regs
> 0)
4520 regs
->start
[0] = pos
;
4521 regs
->end
[0] = (MATCHING_IN_FIRST_STRING
4522 ? ((regoff_t
) (d
- string1
))
4523 : ((regoff_t
) (d
- string2
+ size1
)));
4526 /* Go through the first `min (num_regs, regs->num_regs)'
4527 registers, since that is all we initialized. */
4528 for (mcnt
= 1; (unsigned) mcnt
< MIN (num_regs
, regs
->num_regs
);
4531 if (REG_UNSET (regstart
[mcnt
]) || REG_UNSET (regend
[mcnt
]))
4532 regs
->start
[mcnt
] = regs
->end
[mcnt
] = -1;
4536 = (regoff_t
) POINTER_TO_OFFSET (regstart
[mcnt
]);
4538 = (regoff_t
) POINTER_TO_OFFSET (regend
[mcnt
]);
4542 /* If the regs structure we return has more elements than
4543 were in the pattern, set the extra elements to -1. If
4544 we (re)allocated the registers, this is the case,
4545 because we always allocate enough to have at least one
4547 for (mcnt
= num_regs
; (unsigned) mcnt
< regs
->num_regs
; mcnt
++)
4548 regs
->start
[mcnt
] = regs
->end
[mcnt
] = -1;
4549 } /* regs && !bufp->no_sub */
4551 DEBUG_PRINT4 ("%u failure points pushed, %u popped (%u remain).\n",
4552 nfailure_points_pushed
, nfailure_points_popped
,
4553 nfailure_points_pushed
- nfailure_points_popped
);
4554 DEBUG_PRINT2 ("%u registers pushed.\n", num_regs_pushed
);
4556 mcnt
= d
- pos
- (MATCHING_IN_FIRST_STRING
4560 DEBUG_PRINT2 ("Returning %d from re_match_2.\n", mcnt
);
4566 /* Otherwise match next pattern command. */
4567 switch (SWITCH_ENUM_CAST ((re_opcode_t
) *p
++))
4569 /* Ignore these. Used to ignore the n of succeed_n's which
4570 currently have n == 0. */
4572 DEBUG_PRINT1 ("EXECUTING no_op.\n");
4576 DEBUG_PRINT1 ("EXECUTING succeed.\n");
4579 /* Match the next n pattern characters exactly. The following
4580 byte in the pattern defines n, and the n bytes after that
4581 are the characters to match. */
4584 DEBUG_PRINT2 ("EXECUTING exactn %d.\n", mcnt
);
4586 /* This is written out as an if-else so we don't waste time
4587 testing `translate' inside the loop. */
4593 if ((unsigned char) translate
[(unsigned char) *d
++]
4594 != (unsigned char) *p
++)
4604 if (*d
++ != (char) *p
++) goto fail
;
4608 SET_REGS_MATCHED ();
4612 /* Match any character except possibly a newline or a null. */
4614 DEBUG_PRINT1 ("EXECUTING anychar.\n");
4618 if ((!(bufp
->syntax
& RE_DOT_NEWLINE
) && TRANSLATE (*d
) == '\n')
4619 || (bufp
->syntax
& RE_DOT_NOT_NULL
&& TRANSLATE (*d
) == '\000'))
4622 SET_REGS_MATCHED ();
4623 DEBUG_PRINT2 (" Matched `%d'.\n", *d
);
4631 register unsigned char c
;
4632 boolean
not = (re_opcode_t
) *(p
- 1) == charset_not
;
4634 DEBUG_PRINT2 ("EXECUTING charset%s.\n", not ? "_not" : "");
4637 c
= TRANSLATE (*d
); /* The character to match. */
4639 /* Cast to `unsigned' instead of `unsigned char' in case the
4640 bit list is a full 32 bytes long. */
4641 if (c
< (unsigned) (*p
* BYTEWIDTH
)
4642 && p
[1 + c
/ BYTEWIDTH
] & (1 << (c
% BYTEWIDTH
)))
4647 if (!not) goto fail
;
4649 SET_REGS_MATCHED ();
4655 /* The beginning of a group is represented by start_memory.
4656 The arguments are the register number in the next byte, and the
4657 number of groups inner to this one in the next. The text
4658 matched within the group is recorded (in the internal
4659 registers data structure) under the register number. */
4661 DEBUG_PRINT3 ("EXECUTING start_memory %d (%d):\n", *p
, p
[1]);
4663 /* Find out if this group can match the empty string. */
4664 p1
= p
; /* To send to group_match_null_string_p. */
4666 if (REG_MATCH_NULL_STRING_P (reg_info
[*p
]) == MATCH_NULL_UNSET_VALUE
)
4667 REG_MATCH_NULL_STRING_P (reg_info
[*p
])
4668 = group_match_null_string_p (&p1
, pend
, reg_info
);
4670 /* Save the position in the string where we were the last time
4671 we were at this open-group operator in case the group is
4672 operated upon by a repetition operator, e.g., with `(a*)*b'
4673 against `ab'; then we want to ignore where we are now in
4674 the string in case this attempt to match fails. */
4675 old_regstart
[*p
] = REG_MATCH_NULL_STRING_P (reg_info
[*p
])
4676 ? REG_UNSET (regstart
[*p
]) ? d
: regstart
[*p
]
4678 DEBUG_PRINT2 (" old_regstart: %d\n",
4679 POINTER_TO_OFFSET (old_regstart
[*p
]));
4682 DEBUG_PRINT2 (" regstart: %d\n", POINTER_TO_OFFSET (regstart
[*p
]));
4684 IS_ACTIVE (reg_info
[*p
]) = 1;
4685 MATCHED_SOMETHING (reg_info
[*p
]) = 0;
4687 /* Clear this whenever we change the register activity status. */
4688 set_regs_matched_done
= 0;
4690 /* This is the new highest active register. */
4691 highest_active_reg
= *p
;
4693 /* If nothing was active before, this is the new lowest active
4695 if (lowest_active_reg
== NO_LOWEST_ACTIVE_REG
)
4696 lowest_active_reg
= *p
;
4698 /* Move past the register number and inner group count. */
4700 just_past_start_mem
= p
;
4705 /* The stop_memory opcode represents the end of a group. Its
4706 arguments are the same as start_memory's: the register
4707 number, and the number of inner groups. */
4709 DEBUG_PRINT3 ("EXECUTING stop_memory %d (%d):\n", *p
, p
[1]);
4711 /* We need to save the string position the last time we were at
4712 this close-group operator in case the group is operated
4713 upon by a repetition operator, e.g., with `((a*)*(b*)*)*'
4714 against `aba'; then we want to ignore where we are now in
4715 the string in case this attempt to match fails. */
4716 old_regend
[*p
] = REG_MATCH_NULL_STRING_P (reg_info
[*p
])
4717 ? REG_UNSET (regend
[*p
]) ? d
: regend
[*p
]
4719 DEBUG_PRINT2 (" old_regend: %d\n",
4720 POINTER_TO_OFFSET (old_regend
[*p
]));
4723 DEBUG_PRINT2 (" regend: %d\n", POINTER_TO_OFFSET (regend
[*p
]));
4725 /* This register isn't active anymore. */
4726 IS_ACTIVE (reg_info
[*p
]) = 0;
4728 /* Clear this whenever we change the register activity status. */
4729 set_regs_matched_done
= 0;
4731 /* If this was the only register active, nothing is active
4733 if (lowest_active_reg
== highest_active_reg
)
4735 lowest_active_reg
= NO_LOWEST_ACTIVE_REG
;
4736 highest_active_reg
= NO_HIGHEST_ACTIVE_REG
;
4739 { /* We must scan for the new highest active register, since
4740 it isn't necessarily one less than now: consider
4741 (a(b)c(d(e)f)g). When group 3 ends, after the f), the
4742 new highest active register is 1. */
4743 unsigned char r
= *p
- 1;
4744 while (r
> 0 && !IS_ACTIVE (reg_info
[r
]))
4747 /* If we end up at register zero, that means that we saved
4748 the registers as the result of an `on_failure_jump', not
4749 a `start_memory', and we jumped to past the innermost
4750 `stop_memory'. For example, in ((.)*) we save
4751 registers 1 and 2 as a result of the *, but when we pop
4752 back to the second ), we are at the stop_memory 1.
4753 Thus, nothing is active. */
4756 lowest_active_reg
= NO_LOWEST_ACTIVE_REG
;
4757 highest_active_reg
= NO_HIGHEST_ACTIVE_REG
;
4760 highest_active_reg
= r
;
4763 /* If just failed to match something this time around with a
4764 group that's operated on by a repetition operator, try to
4765 force exit from the ``loop'', and restore the register
4766 information for this group that we had before trying this
4768 if ((!MATCHED_SOMETHING (reg_info
[*p
])
4769 || just_past_start_mem
== p
- 1)
4772 boolean is_a_jump_n
= false;
4776 switch ((re_opcode_t
) *p1
++)
4780 case pop_failure_jump
:
4781 case maybe_pop_jump
:
4783 case dummy_failure_jump
:
4784 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
4794 /* If the next operation is a jump backwards in the pattern
4795 to an on_failure_jump right before the start_memory
4796 corresponding to this stop_memory, exit from the loop
4797 by forcing a failure after pushing on the stack the
4798 on_failure_jump's jump in the pattern, and d. */
4799 if (mcnt
< 0 && (re_opcode_t
) *p1
== on_failure_jump
4800 && (re_opcode_t
) p1
[3] == start_memory
&& p1
[4] == *p
)
4802 /* If this group ever matched anything, then restore
4803 what its registers were before trying this last
4804 failed match, e.g., with `(a*)*b' against `ab' for
4805 regstart[1], and, e.g., with `((a*)*(b*)*)*'
4806 against `aba' for regend[3].
4808 Also restore the registers for inner groups for,
4809 e.g., `((a*)(b*))*' against `aba' (register 3 would
4810 otherwise get trashed). */
4812 if (EVER_MATCHED_SOMETHING (reg_info
[*p
]))
4816 EVER_MATCHED_SOMETHING (reg_info
[*p
]) = 0;
4818 /* Restore this and inner groups' (if any) registers. */
4819 for (r
= *p
; r
< (unsigned) *p
+ (unsigned) *(p
+ 1);
4822 regstart
[r
] = old_regstart
[r
];
4824 /* xx why this test? */
4825 if (old_regend
[r
] >= regstart
[r
])
4826 regend
[r
] = old_regend
[r
];
4830 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
4831 PUSH_FAILURE_POINT (p1
+ mcnt
, d
, -2);
4837 /* Move past the register number and the inner group count. */
4842 /* \<digit> has been turned into a `duplicate' command which is
4843 followed by the numeric value of <digit> as the register number. */
4846 register const char *d2
, *dend2
;
4847 int regno
= *p
++; /* Get which register to match against. */
4848 DEBUG_PRINT2 ("EXECUTING duplicate %d.\n", regno
);
4850 /* Can't back reference a group which we've never matched. */
4851 if (REG_UNSET (regstart
[regno
]) || REG_UNSET (regend
[regno
]))
4854 /* Where in input to try to start matching. */
4855 d2
= regstart
[regno
];
4857 /* Where to stop matching; if both the place to start and
4858 the place to stop matching are in the same string, then
4859 set to the place to stop, otherwise, for now have to use
4860 the end of the first string. */
4862 dend2
= ((FIRST_STRING_P (regstart
[regno
])
4863 == FIRST_STRING_P (regend
[regno
]))
4864 ? regend
[regno
] : end_match_1
);
4867 /* If necessary, advance to next segment in register
4871 if (dend2
== end_match_2
) break;
4872 if (dend2
== regend
[regno
]) break;
4874 /* End of string1 => advance to string2. */
4876 dend2
= regend
[regno
];
4878 /* At end of register contents => success */
4879 if (d2
== dend2
) break;
4881 /* If necessary, advance to next segment in data. */
4884 /* How many characters left in this segment to match. */
4887 /* Want how many consecutive characters we can match in
4888 one shot, so, if necessary, adjust the count. */
4889 if (mcnt
> dend2
- d2
)
4892 /* Compare that many; failure if mismatch, else move
4895 ? bcmp_translate (d
, d2
, mcnt
, translate
)
4896 : memcmp (d
, d2
, mcnt
))
4898 d
+= mcnt
, d2
+= mcnt
;
4900 /* Do this because we've match some characters. */
4901 SET_REGS_MATCHED ();
4907 /* begline matches the empty string at the beginning of the string
4908 (unless `not_bol' is set in `bufp'), and, if
4909 `newline_anchor' is set, after newlines. */
4911 DEBUG_PRINT1 ("EXECUTING begline.\n");
4913 if (AT_STRINGS_BEG (d
))
4915 if (!bufp
->not_bol
) break;
4917 else if (d
[-1] == '\n' && bufp
->newline_anchor
)
4921 /* In all other cases, we fail. */
4925 /* endline is the dual of begline. */
4927 DEBUG_PRINT1 ("EXECUTING endline.\n");
4929 if (AT_STRINGS_END (d
))
4931 if (!bufp
->not_eol
) break;
4934 /* We have to ``prefetch'' the next character. */
4935 else if ((d
== end1
? *string2
: *d
) == '\n'
4936 && bufp
->newline_anchor
)
4943 /* Match at the very beginning of the data. */
4945 DEBUG_PRINT1 ("EXECUTING begbuf.\n");
4946 if (AT_STRINGS_BEG (d
))
4951 /* Match at the very end of the data. */
4953 DEBUG_PRINT1 ("EXECUTING endbuf.\n");
4954 if (AT_STRINGS_END (d
))
4959 /* on_failure_keep_string_jump is used to optimize `.*\n'. It
4960 pushes NULL as the value for the string on the stack. Then
4961 `pop_failure_point' will keep the current value for the
4962 string, instead of restoring it. To see why, consider
4963 matching `foo\nbar' against `.*\n'. The .* matches the foo;
4964 then the . fails against the \n. But the next thing we want
4965 to do is match the \n against the \n; if we restored the
4966 string value, we would be back at the foo.
4968 Because this is used only in specific cases, we don't need to
4969 check all the things that `on_failure_jump' does, to make
4970 sure the right things get saved on the stack. Hence we don't
4971 share its code. The only reason to push anything on the
4972 stack at all is that otherwise we would have to change
4973 `anychar's code to do something besides goto fail in this
4974 case; that seems worse than this. */
4975 case on_failure_keep_string_jump
:
4976 DEBUG_PRINT1 ("EXECUTING on_failure_keep_string_jump");
4978 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
4980 DEBUG_PRINT3 (" %d (to %p):\n", mcnt
, p
+ mcnt
);
4982 DEBUG_PRINT3 (" %d (to 0x%x):\n", mcnt
, p
+ mcnt
);
4985 PUSH_FAILURE_POINT (p
+ mcnt
, NULL
, -2);
4989 /* Uses of on_failure_jump:
4991 Each alternative starts with an on_failure_jump that points
4992 to the beginning of the next alternative. Each alternative
4993 except the last ends with a jump that in effect jumps past
4994 the rest of the alternatives. (They really jump to the
4995 ending jump of the following alternative, because tensioning
4996 these jumps is a hassle.)
4998 Repeats start with an on_failure_jump that points past both
4999 the repetition text and either the following jump or
5000 pop_failure_jump back to this on_failure_jump. */
5001 case on_failure_jump
:
5003 DEBUG_PRINT1 ("EXECUTING on_failure_jump");
5005 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5007 DEBUG_PRINT3 (" %d (to %p)", mcnt
, p
+ mcnt
);
5009 DEBUG_PRINT3 (" %d (to 0x%x)", mcnt
, p
+ mcnt
);
5012 /* If this on_failure_jump comes right before a group (i.e.,
5013 the original * applied to a group), save the information
5014 for that group and all inner ones, so that if we fail back
5015 to this point, the group's information will be correct.
5016 For example, in \(a*\)*\1, we need the preceding group,
5017 and in \(zz\(a*\)b*\)\2, we need the inner group. */
5019 /* We can't use `p' to check ahead because we push
5020 a failure point to `p + mcnt' after we do this. */
5023 /* We need to skip no_op's before we look for the
5024 start_memory in case this on_failure_jump is happening as
5025 the result of a completed succeed_n, as in \(a\)\{1,3\}b\1
5027 while (p1
< pend
&& (re_opcode_t
) *p1
== no_op
)
5030 if (p1
< pend
&& (re_opcode_t
) *p1
== start_memory
)
5032 /* We have a new highest active register now. This will
5033 get reset at the start_memory we are about to get to,
5034 but we will have saved all the registers relevant to
5035 this repetition op, as described above. */
5036 highest_active_reg
= *(p1
+ 1) + *(p1
+ 2);
5037 if (lowest_active_reg
== NO_LOWEST_ACTIVE_REG
)
5038 lowest_active_reg
= *(p1
+ 1);
5041 DEBUG_PRINT1 (":\n");
5042 PUSH_FAILURE_POINT (p
+ mcnt
, d
, -2);
5046 /* A smart repeat ends with `maybe_pop_jump'.
5047 We change it to either `pop_failure_jump' or `jump'. */
5048 case maybe_pop_jump
:
5049 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5050 DEBUG_PRINT2 ("EXECUTING maybe_pop_jump %d.\n", mcnt
);
5052 register unsigned char *p2
= p
;
5054 /* Compare the beginning of the repeat with what in the
5055 pattern follows its end. If we can establish that there
5056 is nothing that they would both match, i.e., that we
5057 would have to backtrack because of (as in, e.g., `a*a')
5058 then we can change to pop_failure_jump, because we'll
5059 never have to backtrack.
5061 This is not true in the case of alternatives: in
5062 `(a|ab)*' we do need to backtrack to the `ab' alternative
5063 (e.g., if the string was `ab'). But instead of trying to
5064 detect that here, the alternative has put on a dummy
5065 failure point which is what we will end up popping. */
5067 /* Skip over open/close-group commands.
5068 If what follows this loop is a ...+ construct,
5069 look at what begins its body, since we will have to
5070 match at least one of that. */
5074 && ((re_opcode_t
) *p2
== stop_memory
5075 || (re_opcode_t
) *p2
== start_memory
))
5077 else if (p2
+ 6 < pend
5078 && (re_opcode_t
) *p2
== dummy_failure_jump
)
5085 /* p1[0] ... p1[2] are the `on_failure_jump' corresponding
5086 to the `maybe_finalize_jump' of this case. Examine what
5089 /* If we're at the end of the pattern, we can change. */
5092 /* Consider what happens when matching ":\(.*\)"
5093 against ":/". I don't really understand this code
5095 p
[-3] = (unsigned char) pop_failure_jump
;
5097 (" End of pattern: change to `pop_failure_jump'.\n");
5100 else if ((re_opcode_t
) *p2
== exactn
5101 || (bufp
->newline_anchor
&& (re_opcode_t
) *p2
== endline
))
5103 register unsigned char c
5104 = *p2
== (unsigned char) endline
? '\n' : p2
[2];
5106 if ((re_opcode_t
) p1
[3] == exactn
&& p1
[5] != c
)
5108 p
[-3] = (unsigned char) pop_failure_jump
;
5109 DEBUG_PRINT3 (" %c != %c => pop_failure_jump.\n",
5113 else if ((re_opcode_t
) p1
[3] == charset
5114 || (re_opcode_t
) p1
[3] == charset_not
)
5116 int not = (re_opcode_t
) p1
[3] == charset_not
;
5118 if (c
< (unsigned char) (p1
[4] * BYTEWIDTH
)
5119 && p1
[5 + c
/ BYTEWIDTH
] & (1 << (c
% BYTEWIDTH
)))
5122 /* `not' is equal to 1 if c would match, which means
5123 that we can't change to pop_failure_jump. */
5126 p
[-3] = (unsigned char) pop_failure_jump
;
5127 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
5131 else if ((re_opcode_t
) *p2
== charset
)
5133 /* We win if the first character of the loop is not part
5135 if ((re_opcode_t
) p1
[3] == exactn
5136 && ! ((int) p2
[1] * BYTEWIDTH
> (int) p1
[5]
5137 && (p2
[2 + p1
[5] / BYTEWIDTH
]
5138 & (1 << (p1
[5] % BYTEWIDTH
)))))
5140 p
[-3] = (unsigned char) pop_failure_jump
;
5141 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
5144 else if ((re_opcode_t
) p1
[3] == charset_not
)
5147 /* We win if the charset_not inside the loop
5148 lists every character listed in the charset after. */
5149 for (idx
= 0; idx
< (int) p2
[1]; idx
++)
5150 if (! (p2
[2 + idx
] == 0
5151 || (idx
< (int) p1
[4]
5152 && ((p2
[2 + idx
] & ~ p1
[5 + idx
]) == 0))))
5157 p
[-3] = (unsigned char) pop_failure_jump
;
5158 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
5161 else if ((re_opcode_t
) p1
[3] == charset
)
5164 /* We win if the charset inside the loop
5165 has no overlap with the one after the loop. */
5167 idx
< (int) p2
[1] && idx
< (int) p1
[4];
5169 if ((p2
[2 + idx
] & p1
[5 + idx
]) != 0)
5172 if (idx
== p2
[1] || idx
== p1
[4])
5174 p
[-3] = (unsigned char) pop_failure_jump
;
5175 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
5180 p
-= 2; /* Point at relative address again. */
5181 if ((re_opcode_t
) p
[-1] != pop_failure_jump
)
5183 p
[-1] = (unsigned char) jump
;
5184 DEBUG_PRINT1 (" Match => jump.\n");
5185 goto unconditional_jump
;
5187 /* Note fall through. */
5190 /* The end of a simple repeat has a pop_failure_jump back to
5191 its matching on_failure_jump, where the latter will push a
5192 failure point. The pop_failure_jump takes off failure
5193 points put on by this pop_failure_jump's matching
5194 on_failure_jump; we got through the pattern to here from the
5195 matching on_failure_jump, so didn't fail. */
5196 case pop_failure_jump
:
5198 /* We need to pass separate storage for the lowest and
5199 highest registers, even though we don't care about the
5200 actual values. Otherwise, we will restore only one
5201 register from the stack, since lowest will == highest in
5202 `pop_failure_point'. */
5203 active_reg_t dummy_low_reg
, dummy_high_reg
;
5204 unsigned char *pdummy
;
5207 DEBUG_PRINT1 ("EXECUTING pop_failure_jump.\n");
5208 POP_FAILURE_POINT (sdummy
, pdummy
,
5209 dummy_low_reg
, dummy_high_reg
,
5210 reg_dummy
, reg_dummy
, reg_info_dummy
);
5212 /* Note fall through. */
5216 DEBUG_PRINT2 ("\n%p: ", p
);
5218 DEBUG_PRINT2 ("\n0x%x: ", p
);
5220 /* Note fall through. */
5222 /* Unconditionally jump (without popping any failure points). */
5224 EXTRACT_NUMBER_AND_INCR (mcnt
, p
); /* Get the amount to jump. */
5225 DEBUG_PRINT2 ("EXECUTING jump %d ", mcnt
);
5226 p
+= mcnt
; /* Do the jump. */
5228 DEBUG_PRINT2 ("(to %p).\n", p
);
5230 DEBUG_PRINT2 ("(to 0x%x).\n", p
);
5235 /* We need this opcode so we can detect where alternatives end
5236 in `group_match_null_string_p' et al. */
5238 DEBUG_PRINT1 ("EXECUTING jump_past_alt.\n");
5239 goto unconditional_jump
;
5242 /* Normally, the on_failure_jump pushes a failure point, which
5243 then gets popped at pop_failure_jump. We will end up at
5244 pop_failure_jump, also, and with a pattern of, say, `a+', we
5245 are skipping over the on_failure_jump, so we have to push
5246 something meaningless for pop_failure_jump to pop. */
5247 case dummy_failure_jump
:
5248 DEBUG_PRINT1 ("EXECUTING dummy_failure_jump.\n");
5249 /* It doesn't matter what we push for the string here. What
5250 the code at `fail' tests is the value for the pattern. */
5251 PUSH_FAILURE_POINT (NULL
, NULL
, -2);
5252 goto unconditional_jump
;
5255 /* At the end of an alternative, we need to push a dummy failure
5256 point in case we are followed by a `pop_failure_jump', because
5257 we don't want the failure point for the alternative to be
5258 popped. For example, matching `(a|ab)*' against `aab'
5259 requires that we match the `ab' alternative. */
5260 case push_dummy_failure
:
5261 DEBUG_PRINT1 ("EXECUTING push_dummy_failure.\n");
5262 /* See comments just above at `dummy_failure_jump' about the
5264 PUSH_FAILURE_POINT (NULL
, NULL
, -2);
5267 /* Have to succeed matching what follows at least n times.
5268 After that, handle like `on_failure_jump'. */
5270 EXTRACT_NUMBER (mcnt
, p
+ 2);
5271 DEBUG_PRINT2 ("EXECUTING succeed_n %d.\n", mcnt
);
5274 /* Originally, this is how many times we HAVE to succeed. */
5279 STORE_NUMBER_AND_INCR (p
, mcnt
);
5281 DEBUG_PRINT3 (" Setting %p to %d.\n", p
- 2, mcnt
);
5283 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p
- 2, mcnt
);
5289 DEBUG_PRINT2 (" Setting two bytes from %p to no_op.\n", p
+2);
5291 DEBUG_PRINT2 (" Setting two bytes from 0x%x to no_op.\n", p
+2);
5293 p
[2] = (unsigned char) no_op
;
5294 p
[3] = (unsigned char) no_op
;
5300 EXTRACT_NUMBER (mcnt
, p
+ 2);
5301 DEBUG_PRINT2 ("EXECUTING jump_n %d.\n", mcnt
);
5303 /* Originally, this is how many times we CAN jump. */
5307 STORE_NUMBER (p
+ 2, mcnt
);
5309 DEBUG_PRINT3 (" Setting %p to %d.\n", p
+ 2, mcnt
);
5311 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p
+ 2, mcnt
);
5313 goto unconditional_jump
;
5315 /* If don't have to jump any more, skip over the rest of command. */
5322 DEBUG_PRINT1 ("EXECUTING set_number_at.\n");
5324 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5326 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5328 DEBUG_PRINT3 (" Setting %p to %d.\n", p1
, mcnt
);
5330 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p1
, mcnt
);
5332 STORE_NUMBER (p1
, mcnt
);
5337 /* The DEC Alpha C compiler 3.x generates incorrect code for the
5338 test WORDCHAR_P (d - 1) != WORDCHAR_P (d) in the expansion of
5339 AT_WORD_BOUNDARY, so this code is disabled. Expanding the
5340 macro and introducing temporary variables works around the bug. */
5343 DEBUG_PRINT1 ("EXECUTING wordbound.\n");
5344 if (AT_WORD_BOUNDARY (d
))
5349 DEBUG_PRINT1 ("EXECUTING notwordbound.\n");
5350 if (AT_WORD_BOUNDARY (d
))
5356 boolean prevchar
, thischar
;
5358 DEBUG_PRINT1 ("EXECUTING wordbound.\n");
5359 if (AT_STRINGS_BEG (d
) || AT_STRINGS_END (d
))
5362 prevchar
= WORDCHAR_P (d
- 1);
5363 thischar
= WORDCHAR_P (d
);
5364 if (prevchar
!= thischar
)
5371 boolean prevchar
, thischar
;
5373 DEBUG_PRINT1 ("EXECUTING notwordbound.\n");
5374 if (AT_STRINGS_BEG (d
) || AT_STRINGS_END (d
))
5377 prevchar
= WORDCHAR_P (d
- 1);
5378 thischar
= WORDCHAR_P (d
);
5379 if (prevchar
!= thischar
)
5386 DEBUG_PRINT1 ("EXECUTING wordbeg.\n");
5387 if (WORDCHAR_P (d
) && (AT_STRINGS_BEG (d
) || !WORDCHAR_P (d
- 1)))
5392 DEBUG_PRINT1 ("EXECUTING wordend.\n");
5393 if (!AT_STRINGS_BEG (d
) && WORDCHAR_P (d
- 1)
5394 && (!WORDCHAR_P (d
) || AT_STRINGS_END (d
)))
5400 DEBUG_PRINT1 ("EXECUTING before_dot.\n");
5401 if (PTR_CHAR_POS ((unsigned char *) d
) >= point
)
5406 DEBUG_PRINT1 ("EXECUTING at_dot.\n");
5407 if (PTR_CHAR_POS ((unsigned char *) d
) != point
)
5412 DEBUG_PRINT1 ("EXECUTING after_dot.\n");
5413 if (PTR_CHAR_POS ((unsigned char *) d
) <= point
)
5418 DEBUG_PRINT2 ("EXECUTING syntaxspec %d.\n", mcnt
);
5423 DEBUG_PRINT1 ("EXECUTING Emacs wordchar.\n");
5427 /* Can't use *d++ here; SYNTAX may be an unsafe macro. */
5429 if (SYNTAX (d
[-1]) != (enum syntaxcode
) mcnt
)
5431 SET_REGS_MATCHED ();
5435 DEBUG_PRINT2 ("EXECUTING notsyntaxspec %d.\n", mcnt
);
5437 goto matchnotsyntax
;
5440 DEBUG_PRINT1 ("EXECUTING Emacs notwordchar.\n");
5444 /* Can't use *d++ here; SYNTAX may be an unsafe macro. */
5446 if (SYNTAX (d
[-1]) == (enum syntaxcode
) mcnt
)
5448 SET_REGS_MATCHED ();
5451 #else /* not emacs */
5453 DEBUG_PRINT1 ("EXECUTING non-Emacs wordchar.\n");
5455 if (!WORDCHAR_P (d
))
5457 SET_REGS_MATCHED ();
5462 DEBUG_PRINT1 ("EXECUTING non-Emacs notwordchar.\n");
5466 SET_REGS_MATCHED ();
5469 #endif /* not emacs */
5474 continue; /* Successfully executed one pattern command; keep going. */
5477 /* We goto here if a matching operation fails. */
5479 if (!FAIL_STACK_EMPTY ())
5480 { /* A restart point is known. Restore to that state. */
5481 DEBUG_PRINT1 ("\nFAIL:\n");
5482 POP_FAILURE_POINT (d
, p
,
5483 lowest_active_reg
, highest_active_reg
,
5484 regstart
, regend
, reg_info
);
5486 /* If this failure point is a dummy, try the next one. */
5490 /* If we failed to the end of the pattern, don't examine *p. */
5494 boolean is_a_jump_n
= false;
5496 /* If failed to a backwards jump that's part of a repetition
5497 loop, need to pop this failure point and use the next one. */
5498 switch ((re_opcode_t
) *p
)
5502 case maybe_pop_jump
:
5503 case pop_failure_jump
:
5506 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
5509 if ((is_a_jump_n
&& (re_opcode_t
) *p1
== succeed_n
)
5511 && (re_opcode_t
) *p1
== on_failure_jump
))
5519 if (d
>= string1
&& d
<= end1
)
5523 break; /* Matching at this starting point really fails. */
5527 goto restore_best_regs
;
5531 return -1; /* Failure to match. */
5534 /* Subroutine definitions for re_match_2. */
5537 /* We are passed P pointing to a register number after a start_memory.
5539 Return true if the pattern up to the corresponding stop_memory can
5540 match the empty string, and false otherwise.
5542 If we find the matching stop_memory, sets P to point to one past its number.
5543 Otherwise, sets P to an undefined byte less than or equal to END.
5545 We don't handle duplicates properly (yet). */
5548 group_match_null_string_p (p
, end
, reg_info
)
5549 unsigned char **p
, *end
;
5550 register_info_type
*reg_info
;
5553 /* Point to after the args to the start_memory. */
5554 unsigned char *p1
= *p
+ 2;
5558 /* Skip over opcodes that can match nothing, and return true or
5559 false, as appropriate, when we get to one that can't, or to the
5560 matching stop_memory. */
5562 switch ((re_opcode_t
) *p1
)
5564 /* Could be either a loop or a series of alternatives. */
5565 case on_failure_jump
:
5567 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
5569 /* If the next operation is not a jump backwards in the
5574 /* Go through the on_failure_jumps of the alternatives,
5575 seeing if any of the alternatives cannot match nothing.
5576 The last alternative starts with only a jump,
5577 whereas the rest start with on_failure_jump and end
5578 with a jump, e.g., here is the pattern for `a|b|c':
5580 /on_failure_jump/0/6/exactn/1/a/jump_past_alt/0/6
5581 /on_failure_jump/0/6/exactn/1/b/jump_past_alt/0/3
5584 So, we have to first go through the first (n-1)
5585 alternatives and then deal with the last one separately. */
5588 /* Deal with the first (n-1) alternatives, which start
5589 with an on_failure_jump (see above) that jumps to right
5590 past a jump_past_alt. */
5592 while ((re_opcode_t
) p1
[mcnt
-3] == jump_past_alt
)
5594 /* `mcnt' holds how many bytes long the alternative
5595 is, including the ending `jump_past_alt' and
5598 if (!alt_match_null_string_p (p1
, p1
+ mcnt
- 3,
5602 /* Move to right after this alternative, including the
5606 /* Break if it's the beginning of an n-th alternative
5607 that doesn't begin with an on_failure_jump. */
5608 if ((re_opcode_t
) *p1
!= on_failure_jump
)
5611 /* Still have to check that it's not an n-th
5612 alternative that starts with an on_failure_jump. */
5614 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
5615 if ((re_opcode_t
) p1
[mcnt
-3] != jump_past_alt
)
5617 /* Get to the beginning of the n-th alternative. */
5623 /* Deal with the last alternative: go back and get number
5624 of the `jump_past_alt' just before it. `mcnt' contains
5625 the length of the alternative. */
5626 EXTRACT_NUMBER (mcnt
, p1
- 2);
5628 if (!alt_match_null_string_p (p1
, p1
+ mcnt
, reg_info
))
5631 p1
+= mcnt
; /* Get past the n-th alternative. */
5637 assert (p1
[1] == **p
);
5643 if (!common_op_match_null_string_p (&p1
, end
, reg_info
))
5646 } /* while p1 < end */
5649 } /* group_match_null_string_p */
5652 /* Similar to group_match_null_string_p, but doesn't deal with alternatives:
5653 It expects P to be the first byte of a single alternative and END one
5654 byte past the last. The alternative can contain groups. */
5657 alt_match_null_string_p (p
, end
, reg_info
)
5658 unsigned char *p
, *end
;
5659 register_info_type
*reg_info
;
5662 unsigned char *p1
= p
;
5666 /* Skip over opcodes that can match nothing, and break when we get
5667 to one that can't. */
5669 switch ((re_opcode_t
) *p1
)
5672 case on_failure_jump
:
5674 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
5679 if (!common_op_match_null_string_p (&p1
, end
, reg_info
))
5682 } /* while p1 < end */
5685 } /* alt_match_null_string_p */
5688 /* Deals with the ops common to group_match_null_string_p and
5689 alt_match_null_string_p.
5691 Sets P to one after the op and its arguments, if any. */
5694 common_op_match_null_string_p (p
, end
, reg_info
)
5695 unsigned char **p
, *end
;
5696 register_info_type
*reg_info
;
5701 unsigned char *p1
= *p
;
5703 switch ((re_opcode_t
) *p1
++)
5723 assert (reg_no
> 0 && reg_no
<= MAX_REGNUM
);
5724 ret
= group_match_null_string_p (&p1
, end
, reg_info
);
5726 /* Have to set this here in case we're checking a group which
5727 contains a group and a back reference to it. */
5729 if (REG_MATCH_NULL_STRING_P (reg_info
[reg_no
]) == MATCH_NULL_UNSET_VALUE
)
5730 REG_MATCH_NULL_STRING_P (reg_info
[reg_no
]) = ret
;
5736 /* If this is an optimized succeed_n for zero times, make the jump. */
5738 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
5746 /* Get to the number of times to succeed. */
5748 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
5753 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
5761 if (!REG_MATCH_NULL_STRING_P (reg_info
[*p1
]))
5769 /* All other opcodes mean we cannot match the empty string. */
5775 } /* common_op_match_null_string_p */
5778 /* Return zero if TRANSLATE[S1] and TRANSLATE[S2] are identical for LEN
5779 bytes; nonzero otherwise. */
5782 bcmp_translate (s1
, s2
, len
, translate
)
5783 const char *s1
, *s2
;
5785 RE_TRANSLATE_TYPE translate
;
5787 register const unsigned char *p1
= (const unsigned char *) s1
;
5788 register const unsigned char *p2
= (const unsigned char *) s2
;
5791 if (translate
[*p1
++] != translate
[*p2
++]) return 1;
5797 /* Entry points for GNU code. */
5799 /* re_compile_pattern is the GNU regular expression compiler: it
5800 compiles PATTERN (of length SIZE) and puts the result in BUFP.
5801 Returns 0 if the pattern was valid, otherwise an error string.
5803 Assumes the `allocated' (and perhaps `buffer') and `translate' fields
5804 are set in BUFP on entry.
5806 We call regex_compile to do the actual compilation. */
5809 re_compile_pattern (pattern
, length
, bufp
)
5810 const char *pattern
;
5812 struct re_pattern_buffer
*bufp
;
5816 /* GNU code is written to assume at least RE_NREGS registers will be set
5817 (and at least one extra will be -1). */
5818 bufp
->regs_allocated
= REGS_UNALLOCATED
;
5820 /* And GNU code determines whether or not to get register information
5821 by passing null for the REGS argument to re_match, etc., not by
5825 /* Match anchors at newline. */
5826 bufp
->newline_anchor
= 1;
5828 ret
= regex_compile (pattern
, length
, re_syntax_options
, bufp
);
5832 return gettext (re_error_msgid
+ re_error_msgid_idx
[(int) ret
]);
5835 weak_alias (__re_compile_pattern
, re_compile_pattern
)
5838 /* Entry points compatible with 4.2 BSD regex library. We don't define
5839 them unless specifically requested. */
5841 #if defined _REGEX_RE_COMP || defined _LIBC
5843 /* BSD has one and only one pattern buffer. */
5844 static struct re_pattern_buffer re_comp_buf
;
5848 /* Make these definitions weak in libc, so POSIX programs can redefine
5849 these names if they don't use our functions, and still use
5850 regcomp/regexec below without link errors. */
5860 if (!re_comp_buf
.buffer
)
5861 return gettext ("No previous regular expression");
5865 if (!re_comp_buf
.buffer
)
5867 re_comp_buf
.buffer
= (unsigned char *) malloc (200);
5868 if (re_comp_buf
.buffer
== NULL
)
5869 return (char *) gettext (re_error_msgid
5870 + re_error_msgid_idx
[(int) REG_ESPACE
]);
5871 re_comp_buf
.allocated
= 200;
5873 re_comp_buf
.fastmap
= (char *) malloc (1 << BYTEWIDTH
);
5874 if (re_comp_buf
.fastmap
== NULL
)
5875 return (char *) gettext (re_error_msgid
5876 + re_error_msgid_idx
[(int) REG_ESPACE
]);
5879 /* Since `re_exec' always passes NULL for the `regs' argument, we
5880 don't need to initialize the pattern buffer fields which affect it. */
5882 /* Match anchors at newlines. */
5883 re_comp_buf
.newline_anchor
= 1;
5885 ret
= regex_compile (s
, strlen (s
), re_syntax_options
, &re_comp_buf
);
5890 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
5891 return (char *) gettext (re_error_msgid
+ re_error_msgid_idx
[(int) ret
]);
5902 const int len
= strlen (s
);
5904 0 <= re_search (&re_comp_buf
, s
, len
, 0, len
, (struct re_registers
*) 0);
5907 #endif /* _REGEX_RE_COMP */
5909 /* POSIX.2 functions. Don't define these for Emacs. */
5913 /* regcomp takes a regular expression as a string and compiles it.
5915 PREG is a regex_t *. We do not expect any fields to be initialized,
5916 since POSIX says we shouldn't. Thus, we set
5918 `buffer' to the compiled pattern;
5919 `used' to the length of the compiled pattern;
5920 `syntax' to RE_SYNTAX_POSIX_EXTENDED if the
5921 REG_EXTENDED bit in CFLAGS is set; otherwise, to
5922 RE_SYNTAX_POSIX_BASIC;
5923 `newline_anchor' to REG_NEWLINE being set in CFLAGS;
5924 `fastmap' to an allocated space for the fastmap;
5925 `fastmap_accurate' to zero;
5926 `re_nsub' to the number of subexpressions in PATTERN.
5928 PATTERN is the address of the pattern string.
5930 CFLAGS is a series of bits which affect compilation.
5932 If REG_EXTENDED is set, we use POSIX extended syntax; otherwise, we
5933 use POSIX basic syntax.
5935 If REG_NEWLINE is set, then . and [^...] don't match newline.
5936 Also, regexec will try a match beginning after every newline.
5938 If REG_ICASE is set, then we considers upper- and lowercase
5939 versions of letters to be equivalent when matching.
5941 If REG_NOSUB is set, then when PREG is passed to regexec, that
5942 routine will report only success or failure, and nothing about the
5945 It returns 0 if it succeeds, nonzero if it doesn't. (See regex.h for
5946 the return codes and their meanings.) */
5949 regcomp (preg
, pattern
, cflags
)
5951 const char *pattern
;
5956 = (cflags
& REG_EXTENDED
) ?
5957 RE_SYNTAX_POSIX_EXTENDED
: RE_SYNTAX_POSIX_BASIC
;
5959 /* regex_compile will allocate the space for the compiled pattern. */
5961 preg
->allocated
= 0;
5964 /* Try to allocate space for the fastmap. */
5965 preg
->fastmap
= (char *) malloc (1 << BYTEWIDTH
);
5967 if (cflags
& REG_ICASE
)
5972 = (RE_TRANSLATE_TYPE
) malloc (CHAR_SET_SIZE
5973 * sizeof (*(RE_TRANSLATE_TYPE
)0));
5974 if (preg
->translate
== NULL
)
5975 return (int) REG_ESPACE
;
5977 /* Map uppercase characters to corresponding lowercase ones. */
5978 for (i
= 0; i
< CHAR_SET_SIZE
; i
++)
5979 preg
->translate
[i
] = ISUPPER (i
) ? TOLOWER (i
) : i
;
5982 preg
->translate
= NULL
;
5984 /* If REG_NEWLINE is set, newlines are treated differently. */
5985 if (cflags
& REG_NEWLINE
)
5986 { /* REG_NEWLINE implies neither . nor [^...] match newline. */
5987 syntax
&= ~RE_DOT_NEWLINE
;
5988 syntax
|= RE_HAT_LISTS_NOT_NEWLINE
;
5989 /* It also changes the matching behavior. */
5990 preg
->newline_anchor
= 1;
5993 preg
->newline_anchor
= 0;
5995 preg
->no_sub
= !!(cflags
& REG_NOSUB
);
5997 /* POSIX says a null character in the pattern terminates it, so we
5998 can use strlen here in compiling the pattern. */
5999 ret
= regex_compile (pattern
, strlen (pattern
), syntax
, preg
);
6001 /* POSIX doesn't distinguish between an unmatched open-group and an
6002 unmatched close-group: both are REG_EPAREN. */
6003 if (ret
== REG_ERPAREN
) ret
= REG_EPAREN
;
6005 if (ret
== REG_NOERROR
&& preg
->fastmap
)
6007 /* Compute the fastmap now, since regexec cannot modify the pattern
6009 if (re_compile_fastmap (preg
) == -2)
6011 /* Some error occurred while computing the fastmap, just forget
6013 free (preg
->fastmap
);
6014 preg
->fastmap
= NULL
;
6021 weak_alias (__regcomp
, regcomp
)
6025 /* regexec searches for a given pattern, specified by PREG, in the
6028 If NMATCH is zero or REG_NOSUB was set in the cflags argument to
6029 `regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at
6030 least NMATCH elements, and we set them to the offsets of the
6031 corresponding matched substrings.
6033 EFLAGS specifies `execution flags' which affect matching: if
6034 REG_NOTBOL is set, then ^ does not match at the beginning of the
6035 string; if REG_NOTEOL is set, then $ does not match at the end.
6037 We return 0 if we find a match and REG_NOMATCH if not. */
6040 regexec (preg
, string
, nmatch
, pmatch
, eflags
)
6041 const regex_t
*preg
;
6044 regmatch_t pmatch
[];
6048 struct re_registers regs
;
6049 regex_t private_preg
;
6050 int len
= strlen (string
);
6051 boolean want_reg_info
= !preg
->no_sub
&& nmatch
> 0;
6053 private_preg
= *preg
;
6055 private_preg
.not_bol
= !!(eflags
& REG_NOTBOL
);
6056 private_preg
.not_eol
= !!(eflags
& REG_NOTEOL
);
6058 /* The user has told us exactly how many registers to return
6059 information about, via `nmatch'. We have to pass that on to the
6060 matching routines. */
6061 private_preg
.regs_allocated
= REGS_FIXED
;
6065 regs
.num_regs
= nmatch
;
6066 regs
.start
= TALLOC (nmatch
* 2, regoff_t
);
6067 if (regs
.start
== NULL
)
6068 return (int) REG_NOMATCH
;
6069 regs
.end
= regs
.start
+ nmatch
;
6072 /* Perform the searching operation. */
6073 ret
= re_search (&private_preg
, string
, len
,
6074 /* start: */ 0, /* range: */ len
,
6075 want_reg_info
? ®s
: (struct re_registers
*) 0);
6077 /* Copy the register information to the POSIX structure. */
6084 for (r
= 0; r
< nmatch
; r
++)
6086 pmatch
[r
].rm_so
= regs
.start
[r
];
6087 pmatch
[r
].rm_eo
= regs
.end
[r
];
6091 /* If we needed the temporary register info, free the space now. */
6095 /* We want zero return to mean success, unlike `re_search'. */
6096 return ret
>= 0 ? (int) REG_NOERROR
: (int) REG_NOMATCH
;
6099 weak_alias (__regexec
, regexec
)
6103 /* Returns a message corresponding to an error code, ERRCODE, returned
6104 from either regcomp or regexec. We don't use PREG here. */
6107 regerror (errcode
, preg
, errbuf
, errbuf_size
)
6109 const regex_t
*preg
;
6117 || errcode
>= (int) (sizeof (re_error_msgid_idx
)
6118 / sizeof (re_error_msgid_idx
[0])))
6119 /* Only error codes returned by the rest of the code should be passed
6120 to this routine. If we are given anything else, or if other regex
6121 code generates an invalid error code, then the program has a bug.
6122 Dump core so we can fix it. */
6125 msg
= gettext (re_error_msgid
+ re_error_msgid_idx
[errcode
]);
6127 msg_size
= strlen (msg
) + 1; /* Includes the null. */
6129 if (errbuf_size
!= 0)
6131 if (msg_size
> errbuf_size
)
6133 #if defined HAVE_MEMPCPY || defined _LIBC
6134 *((char *) __mempcpy (errbuf
, msg
, errbuf_size
- 1)) = '\0';
6136 memcpy (errbuf
, msg
, errbuf_size
- 1);
6137 errbuf
[errbuf_size
- 1] = 0;
6141 memcpy (errbuf
, msg
, msg_size
);
6147 weak_alias (__regerror
, regerror
)
6151 /* Free dynamically allocated space used by PREG. */
6157 if (preg
->buffer
!= NULL
)
6158 free (preg
->buffer
);
6159 preg
->buffer
= NULL
;
6161 preg
->allocated
= 0;
6164 if (preg
->fastmap
!= NULL
)
6165 free (preg
->fastmap
);
6166 preg
->fastmap
= NULL
;
6167 preg
->fastmap_accurate
= 0;
6169 if (preg
->translate
!= NULL
)
6170 free (preg
->translate
);
6171 preg
->translate
= NULL
;
6174 weak_alias (__regfree
, regfree
)
6177 #endif /* not emacs */