1 /* Extended regular expression matching and search library, version
2 0.12. (Implements POSIX draft P10003.2/D11.2, except for
3 internationalization features.)
5 Copyright (C) 1993, 1994, 1995, 1996, 1997 Free Software Foundation, Inc.
7 This program is free software; you can redistribute it and/or modify
8 it under the terms of the GNU General Public License as published by
9 the Free Software Foundation; either version 2, or (at your option)
12 This program 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
15 GNU General Public License for more details.
17 You should have received a copy of the GNU General Public License
18 along with this program; if not, write to the Free Software
19 Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307,
22 /* AIX requires this to be the first thing in the file. */
23 #if defined (_AIX) && !defined (REGEX_MALLOC)
30 /* Converts the pointer to the char to BEG-based offset from the start. */
31 #define PTR_TO_OFFSET(d) \
32 POS_AS_IN_BUFFER (MATCHING_IN_FIRST_STRING \
33 ? (d) - string1 : (d) - (string2 - size1))
34 #define POS_AS_IN_BUFFER(p) ((p) + 1)
40 /* We need this for `regex.h', and perhaps for the Emacs include files. */
41 #include <sys/types.h>
43 /* This is for other GNU distributions with internationalized messages. */
44 #if HAVE_LIBINTL_H || defined (_LIBC)
47 # define gettext(msgid) (msgid)
51 /* This define is so xgettext can find the internationalizable
53 #define gettext_noop(String) String
56 /* The `emacs' switch turns on certain matching commands
57 that make sense only in Emacs. */
63 /* Make syntax table lookup grant data in gl_state. */
64 #define SYNTAX_ENTRY_VIA_PROPERTY
70 #define malloc xmalloc
75 /* If we are not linking with Emacs proper,
76 we can't use the relocating allocator
77 even if config.h says that we can. */
80 #if defined (STDC_HEADERS) || defined (_LIBC)
87 /* When used in Emacs's lib-src, we need to get bzero and bcopy somehow.
88 If nothing else has been done, use the method below. */
89 #ifdef INHIBIT_STRING_HEADER
90 #if !(defined (HAVE_BZERO) && defined (HAVE_BCOPY))
91 #if !defined (bzero) && !defined (bcopy)
92 #undef INHIBIT_STRING_HEADER
97 /* This is the normal way of making sure we have a bcopy and a bzero.
98 This is used in most programs--a few other programs avoid this
99 by defining INHIBIT_STRING_HEADER. */
100 #ifndef INHIBIT_STRING_HEADER
101 #if defined (HAVE_STRING_H) || defined (STDC_HEADERS) || defined (_LIBC)
104 #define bcmp(s1, s2, n) memcmp ((s1), (s2), (n))
107 #define bcopy(s, d, n) memcpy ((d), (s), (n))
110 #define bzero(s, n) memset ((s), 0, (n))
117 /* Define the syntax stuff for \<, \>, etc. */
119 /* This must be nonzero for the wordchar and notwordchar pattern
120 commands in re_match_2. */
125 #ifdef SWITCH_ENUM_BUG
126 #define SWITCH_ENUM_CAST(x) ((int)(x))
128 #define SWITCH_ENUM_CAST(x) (x)
133 extern char *re_syntax_table
;
135 #else /* not SYNTAX_TABLE */
137 /* How many characters in the character set. */
138 #define CHAR_SET_SIZE 256
140 static char re_syntax_table
[CHAR_SET_SIZE
];
151 bzero (re_syntax_table
, sizeof re_syntax_table
);
153 for (c
= 'a'; c
<= 'z'; c
++)
154 re_syntax_table
[c
] = Sword
;
156 for (c
= 'A'; c
<= 'Z'; c
++)
157 re_syntax_table
[c
] = Sword
;
159 for (c
= '0'; c
<= '9'; c
++)
160 re_syntax_table
[c
] = Sword
;
162 re_syntax_table
['_'] = Sword
;
167 #endif /* not SYNTAX_TABLE */
169 #define SYNTAX(c) re_syntax_table[c]
171 /* Dummy macro for non emacs environments. */
172 #define BASE_LEADING_CODE_P(c) (0)
173 #define WORD_BOUNDARY_P(c1, c2) (0)
174 #define CHAR_HEAD_P(p) (1)
175 #define SINGLE_BYTE_CHAR_P(c) (1)
176 #define SAME_CHARSET_P(c1, c2) (1)
177 #define MULTIBYTE_FORM_LENGTH(p, s) (1)
178 #define STRING_CHAR(p, s) (*(p))
179 #define STRING_CHAR_AND_LENGTH(p, s, actual_len) ((actual_len) = 1, *(p))
180 #define GET_CHAR_AFTER_2(c, p, str1, end1, str2, end2) \
181 (c = ((p) == (end1) ? *(str2) : *(p)))
182 #define GET_CHAR_BEFORE_2(c, p, str1, end1, str2, end2) \
183 (c = ((p) == (str2) ? *((end1) - 1) : *((p) - 1)))
184 #endif /* not emacs */
186 /* Get the interface, including the syntax bits. */
189 /* isalpha etc. are used for the character classes. */
192 /* Jim Meyering writes:
194 "... Some ctype macros are valid only for character codes that
195 isascii says are ASCII (SGI's IRIX-4.0.5 is one such system --when
196 using /bin/cc or gcc but without giving an ansi option). So, all
197 ctype uses should be through macros like ISPRINT... If
198 STDC_HEADERS is defined, then autoconf has verified that the ctype
199 macros don't need to be guarded with references to isascii. ...
200 Defining isascii to 1 should let any compiler worth its salt
201 eliminate the && through constant folding." */
203 #if defined (STDC_HEADERS) || (!defined (isascii) && !defined (HAVE_ISASCII))
206 #define ISASCII(c) isascii(c)
210 #define ISBLANK(c) (ISASCII (c) && isblank (c))
212 #define ISBLANK(c) ((c) == ' ' || (c) == '\t')
215 #define ISGRAPH(c) (ISASCII (c) && isgraph (c))
217 #define ISGRAPH(c) (ISASCII (c) && isprint (c) && !isspace (c))
220 #define ISPRINT(c) (ISASCII (c) && isprint (c))
221 #define ISDIGIT(c) (ISASCII (c) && isdigit (c))
222 #define ISALNUM(c) (ISASCII (c) && isalnum (c))
223 #define ISALPHA(c) (ISASCII (c) && isalpha (c))
224 #define ISCNTRL(c) (ISASCII (c) && iscntrl (c))
225 #define ISLOWER(c) (ISASCII (c) && islower (c))
226 #define ISPUNCT(c) (ISASCII (c) && ispunct (c))
227 #define ISSPACE(c) (ISASCII (c) && isspace (c))
228 #define ISUPPER(c) (ISASCII (c) && isupper (c))
229 #define ISXDIGIT(c) (ISASCII (c) && isxdigit (c))
232 #define NULL (void *)0
235 /* We remove any previous definition of `SIGN_EXTEND_CHAR',
236 since ours (we hope) works properly with all combinations of
237 machines, compilers, `char' and `unsigned char' argument types.
238 (Per Bothner suggested the basic approach.) */
239 #undef SIGN_EXTEND_CHAR
241 #define SIGN_EXTEND_CHAR(c) ((signed char) (c))
242 #else /* not __STDC__ */
243 /* As in Harbison and Steele. */
244 #define SIGN_EXTEND_CHAR(c) ((((unsigned char) (c)) ^ 128) - 128)
247 /* Should we use malloc or alloca? If REGEX_MALLOC is not defined, we
248 use `alloca' instead of `malloc'. This is because using malloc in
249 re_search* or re_match* could cause memory leaks when C-g is used in
250 Emacs; also, malloc is slower and causes storage fragmentation. On
251 the other hand, malloc is more portable, and easier to debug.
253 Because we sometimes use alloca, some routines have to be macros,
254 not functions -- `alloca'-allocated space disappears at the end of the
255 function it is called in. */
259 #define REGEX_ALLOCATE malloc
260 #define REGEX_REALLOCATE(source, osize, nsize) realloc (source, nsize)
261 #define REGEX_FREE free
263 #else /* not REGEX_MALLOC */
265 /* Emacs already defines alloca, sometimes. */
268 /* Make alloca work the best possible way. */
270 #define alloca __builtin_alloca
271 #else /* not __GNUC__ */
274 #else /* not __GNUC__ or HAVE_ALLOCA_H */
275 #if 0 /* It is a bad idea to declare alloca. We always cast the result. */
276 #ifndef _AIX /* Already did AIX, up at the top. */
278 #endif /* not _AIX */
280 #endif /* not HAVE_ALLOCA_H */
281 #endif /* not __GNUC__ */
283 #endif /* not alloca */
285 #define REGEX_ALLOCATE alloca
287 /* Assumes a `char *destination' variable. */
288 #define REGEX_REALLOCATE(source, osize, nsize) \
289 (destination = (char *) alloca (nsize), \
290 bcopy (source, destination, osize), \
293 /* No need to do anything to free, after alloca. */
294 #define REGEX_FREE(arg) ((void)0) /* Do nothing! But inhibit gcc warning. */
296 #endif /* not REGEX_MALLOC */
298 /* Define how to allocate the failure stack. */
300 #if defined (REL_ALLOC) && defined (REGEX_MALLOC)
302 #define REGEX_ALLOCATE_STACK(size) \
303 r_alloc (&failure_stack_ptr, (size))
304 #define REGEX_REALLOCATE_STACK(source, osize, nsize) \
305 r_re_alloc (&failure_stack_ptr, (nsize))
306 #define REGEX_FREE_STACK(ptr) \
307 r_alloc_free (&failure_stack_ptr)
309 #else /* not using relocating allocator */
313 #define REGEX_ALLOCATE_STACK malloc
314 #define REGEX_REALLOCATE_STACK(source, osize, nsize) realloc (source, nsize)
315 #define REGEX_FREE_STACK free
317 #else /* not REGEX_MALLOC */
319 #define REGEX_ALLOCATE_STACK alloca
321 #define REGEX_REALLOCATE_STACK(source, osize, nsize) \
322 REGEX_REALLOCATE (source, osize, nsize)
323 /* No need to explicitly free anything. */
324 #define REGEX_FREE_STACK(arg)
326 #endif /* not REGEX_MALLOC */
327 #endif /* not using relocating allocator */
330 /* True if `size1' is non-NULL and PTR is pointing anywhere inside
331 `string1' or just past its end. This works if PTR is NULL, which is
333 #define FIRST_STRING_P(ptr) \
334 (size1 && string1 <= (ptr) && (ptr) <= string1 + size1)
336 /* (Re)Allocate N items of type T using malloc, or fail. */
337 #define TALLOC(n, t) ((t *) malloc ((n) * sizeof (t)))
338 #define RETALLOC(addr, n, t) ((addr) = (t *) realloc (addr, (n) * sizeof (t)))
339 #define RETALLOC_IF(addr, n, t) \
340 if (addr) RETALLOC((addr), (n), t); else (addr) = TALLOC ((n), t)
341 #define REGEX_TALLOC(n, t) ((t *) REGEX_ALLOCATE ((n) * sizeof (t)))
343 #define BYTEWIDTH 8 /* In bits. */
345 #define STREQ(s1, s2) ((strcmp (s1, s2) == 0))
349 #define MAX(a, b) ((a) > (b) ? (a) : (b))
350 #define MIN(a, b) ((a) < (b) ? (a) : (b))
352 typedef char boolean
;
356 static int re_match_2_internal ();
358 /* These are the command codes that appear in compiled regular
359 expressions. Some opcodes are followed by argument bytes. A
360 command code can specify any interpretation whatsoever for its
361 arguments. Zero bytes may appear in the compiled regular expression. */
367 /* Succeed right away--no more backtracking. */
370 /* Followed by one byte giving n, then by n literal bytes. */
373 /* Matches any (more or less) character. */
376 /* Matches any one char belonging to specified set. First
377 following byte is number of bitmap bytes. Then come bytes
378 for a bitmap saying which chars are in. Bits in each byte
379 are ordered low-bit-first. A character is in the set if its
380 bit is 1. A character too large to have a bit in the map is
381 automatically not in the set. */
384 /* Same parameters as charset, but match any character that is
385 not one of those specified. */
388 /* Start remembering the text that is matched, for storing in a
389 register. Followed by one byte with the register number, in
390 the range 0 to one less than the pattern buffer's re_nsub
391 field. Then followed by one byte with the number of groups
392 inner to this one. (This last has to be part of the
393 start_memory only because we need it in the on_failure_jump
397 /* Stop remembering the text that is matched and store it in a
398 memory register. Followed by one byte with the register
399 number, in the range 0 to one less than `re_nsub' in the
400 pattern buffer, and one byte with the number of inner groups,
401 just like `start_memory'. (We need the number of inner
402 groups here because we don't have any easy way of finding the
403 corresponding start_memory when we're at a stop_memory.) */
406 /* Match a duplicate of something remembered. Followed by one
407 byte containing the register number. */
410 /* Fail unless at beginning of line. */
413 /* Fail unless at end of line. */
416 /* Succeeds if at beginning of buffer (if emacs) or at beginning
417 of string to be matched (if not). */
420 /* Analogously, for end of buffer/string. */
423 /* Followed by two byte relative address to which to jump. */
426 /* Same as jump, but marks the end of an alternative. */
429 /* Followed by two-byte relative address of place to resume at
430 in case of failure. */
433 /* Like on_failure_jump, but pushes a placeholder instead of the
434 current string position when executed. */
435 on_failure_keep_string_jump
,
437 /* Throw away latest failure point and then jump to following
438 two-byte relative address. */
441 /* Change to pop_failure_jump if know won't have to backtrack to
442 match; otherwise change to jump. This is used to jump
443 back to the beginning of a repeat. If what follows this jump
444 clearly won't match what the repeat does, such that we can be
445 sure that there is no use backtracking out of repetitions
446 already matched, then we change it to a pop_failure_jump.
447 Followed by two-byte address. */
450 /* Jump to following two-byte address, and push a dummy failure
451 point. This failure point will be thrown away if an attempt
452 is made to use it for a failure. A `+' construct makes this
453 before the first repeat. Also used as an intermediary kind
454 of jump when compiling an alternative. */
457 /* Push a dummy failure point and continue. Used at the end of
461 /* Followed by two-byte relative address and two-byte number n.
462 After matching N times, jump to the address upon failure. */
465 /* Followed by two-byte relative address, and two-byte number n.
466 Jump to the address N times, then fail. */
469 /* Set the following two-byte relative address to the
470 subsequent two-byte number. The address *includes* the two
474 wordchar
, /* Matches any word-constituent character. */
475 notwordchar
, /* Matches any char that is not a word-constituent. */
477 wordbeg
, /* Succeeds if at word beginning. */
478 wordend
, /* Succeeds if at word end. */
480 wordbound
, /* Succeeds if at a word boundary. */
481 notwordbound
/* Succeeds if not at a word boundary. */
484 ,before_dot
, /* Succeeds if before point. */
485 at_dot
, /* Succeeds if at point. */
486 after_dot
, /* Succeeds if after point. */
488 /* Matches any character whose syntax is specified. Followed by
489 a byte which contains a syntax code, e.g., Sword. */
492 /* Matches any character whose syntax is not that specified. */
495 /* Matches any character whose category-set contains the specified
496 category. The operator is followed by a byte which contains a
497 category code (mnemonic ASCII character). */
500 /* Matches any character whose category-set does not contain the
501 specified category. The operator is followed by a byte which
502 contains the category code (mnemonic ASCII character). */
507 /* Common operations on the compiled pattern. */
509 /* Store NUMBER in two contiguous bytes starting at DESTINATION. */
511 #define STORE_NUMBER(destination, number) \
513 (destination)[0] = (number) & 0377; \
514 (destination)[1] = (number) >> 8; \
517 /* Same as STORE_NUMBER, except increment DESTINATION to
518 the byte after where the number is stored. Therefore, DESTINATION
519 must be an lvalue. */
521 #define STORE_NUMBER_AND_INCR(destination, number) \
523 STORE_NUMBER (destination, number); \
524 (destination) += 2; \
527 /* Put into DESTINATION a number stored in two contiguous bytes starting
530 #define EXTRACT_NUMBER(destination, source) \
532 (destination) = *(source) & 0377; \
533 (destination) += SIGN_EXTEND_CHAR (*((source) + 1)) << 8; \
538 extract_number (dest
, source
)
540 unsigned char *source
;
542 int temp
= SIGN_EXTEND_CHAR (*(source
+ 1));
543 *dest
= *source
& 0377;
547 #ifndef EXTRACT_MACROS /* To debug the macros. */
548 #undef EXTRACT_NUMBER
549 #define EXTRACT_NUMBER(dest, src) extract_number (&dest, src)
550 #endif /* not EXTRACT_MACROS */
554 /* Same as EXTRACT_NUMBER, except increment SOURCE to after the number.
555 SOURCE must be an lvalue. */
557 #define EXTRACT_NUMBER_AND_INCR(destination, source) \
559 EXTRACT_NUMBER (destination, source); \
565 extract_number_and_incr (destination
, source
)
567 unsigned char **source
;
569 extract_number (destination
, *source
);
573 #ifndef EXTRACT_MACROS
574 #undef EXTRACT_NUMBER_AND_INCR
575 #define EXTRACT_NUMBER_AND_INCR(dest, src) \
576 extract_number_and_incr (&dest, &src)
577 #endif /* not EXTRACT_MACROS */
581 /* Store a multibyte character in three contiguous bytes starting
582 DESTINATION, and increment DESTINATION to the byte after where the
583 character is stored. Therefore, DESTINATION must be an lvalue. */
585 #define STORE_CHARACTER_AND_INCR(destination, character) \
587 (destination)[0] = (character) & 0377; \
588 (destination)[1] = ((character) >> 8) & 0377; \
589 (destination)[2] = (character) >> 16; \
590 (destination) += 3; \
593 /* Put into DESTINATION a character stored in three contiguous bytes
594 starting at SOURCE. */
596 #define EXTRACT_CHARACTER(destination, source) \
598 (destination) = ((source)[0] \
599 | ((source)[1] << 8) \
600 | ((source)[2] << 16)); \
604 /* Macros for charset. */
606 /* Size of bitmap of charset P in bytes. P is a start of charset,
607 i.e. *P is (re_opcode_t) charset or (re_opcode_t) charset_not. */
608 #define CHARSET_BITMAP_SIZE(p) ((p)[1] & 0x7F)
610 /* Nonzero if charset P has range table. */
611 #define CHARSET_RANGE_TABLE_EXISTS_P(p) ((p)[1] & 0x80)
613 /* Return the address of range table of charset P. But not the start
614 of table itself, but the before where the number of ranges is
615 stored. `2 +' means to skip re_opcode_t and size of bitmap. */
616 #define CHARSET_RANGE_TABLE(p) (&(p)[2 + CHARSET_BITMAP_SIZE (p)])
618 /* Test if C is listed in the bitmap of charset P. */
619 #define CHARSET_LOOKUP_BITMAP(p, c) \
620 ((c) < CHARSET_BITMAP_SIZE (p) * BYTEWIDTH \
621 && (p)[2 + (c) / BYTEWIDTH] & (1 << ((c) % BYTEWIDTH)))
623 /* Return the address of end of RANGE_TABLE. COUNT is number of
624 ranges (which is a pair of (start, end)) in the RANGE_TABLE. `* 2'
625 is start of range and end of range. `* 3' is size of each start
627 #define CHARSET_RANGE_TABLE_END(range_table, count) \
628 ((range_table) + (count) * 2 * 3)
630 /* Test if C is in RANGE_TABLE. A flag NOT is negated if C is in.
631 COUNT is number of ranges in RANGE_TABLE. */
632 #define CHARSET_LOOKUP_RANGE_TABLE_RAW(not, c, range_table, count) \
635 int range_start, range_end; \
637 unsigned char *range_table_end \
638 = CHARSET_RANGE_TABLE_END ((range_table), (count)); \
640 for (p = (range_table); p < range_table_end; p += 2 * 3) \
642 EXTRACT_CHARACTER (range_start, p); \
643 EXTRACT_CHARACTER (range_end, p + 3); \
645 if (range_start <= (c) && (c) <= range_end) \
654 /* Test if C is in range table of CHARSET. The flag NOT is negated if
655 C is listed in it. */
656 #define CHARSET_LOOKUP_RANGE_TABLE(not, c, charset) \
659 /* Number of ranges in range table. */ \
661 unsigned char *range_table = CHARSET_RANGE_TABLE (charset); \
663 EXTRACT_NUMBER_AND_INCR (count, range_table); \
664 CHARSET_LOOKUP_RANGE_TABLE_RAW ((not), (c), range_table, count); \
668 /* If DEBUG is defined, Regex prints many voluminous messages about what
669 it is doing (if the variable `debug' is nonzero). If linked with the
670 main program in `iregex.c', you can enter patterns and strings
671 interactively. And if linked with the main program in `main.c' and
672 the other test files, you can run the already-written tests. */
676 /* We use standard I/O for debugging. */
679 /* It is useful to test things that ``must'' be true when debugging. */
682 static int debug
= 0;
684 #define DEBUG_STATEMENT(e) e
685 #define DEBUG_PRINT1(x) if (debug) printf (x)
686 #define DEBUG_PRINT2(x1, x2) if (debug) printf (x1, x2)
687 #define DEBUG_PRINT3(x1, x2, x3) if (debug) printf (x1, x2, x3)
688 #define DEBUG_PRINT4(x1, x2, x3, x4) if (debug) printf (x1, x2, x3, x4)
689 #define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) \
690 if (debug) print_partial_compiled_pattern (s, e)
691 #define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) \
692 if (debug) print_double_string (w, s1, sz1, s2, sz2)
695 /* Print the fastmap in human-readable form. */
698 print_fastmap (fastmap
)
701 unsigned was_a_range
= 0;
704 while (i
< (1 << BYTEWIDTH
))
710 while (i
< (1 << BYTEWIDTH
) && fastmap
[i
])
726 /* Print a compiled pattern string in human-readable form, starting at
727 the START pointer into it and ending just before the pointer END. */
730 print_partial_compiled_pattern (start
, end
)
731 unsigned char *start
;
735 unsigned char *p
= start
;
736 unsigned char *pend
= end
;
744 /* Loop over pattern commands. */
747 printf ("%d:\t", p
- start
);
749 switch ((re_opcode_t
) *p
++)
757 printf ("/exactn/%d", mcnt
);
768 printf ("/start_memory/%d/%d", mcnt
, *p
++);
773 printf ("/stop_memory/%d/%d", mcnt
, *p
++);
777 printf ("/duplicate/%d", *p
++);
787 register int c
, last
= -100;
788 register int in_range
= 0;
790 printf ("/charset [%s",
791 (re_opcode_t
) *(p
- 1) == charset_not
? "^" : "");
793 assert (p
+ *p
< pend
);
795 for (c
= 0; c
< 256; c
++)
797 && (p
[1 + (c
/8)] & (1 << (c
% 8))))
799 /* Are we starting a range? */
800 if (last
+ 1 == c
&& ! in_range
)
805 /* Have we broken a range? */
806 else if (last
+ 1 != c
&& in_range
)
835 case on_failure_jump
:
836 extract_number_and_incr (&mcnt
, &p
);
837 printf ("/on_failure_jump to %d", p
+ mcnt
- start
);
840 case on_failure_keep_string_jump
:
841 extract_number_and_incr (&mcnt
, &p
);
842 printf ("/on_failure_keep_string_jump to %d", p
+ mcnt
- start
);
845 case dummy_failure_jump
:
846 extract_number_and_incr (&mcnt
, &p
);
847 printf ("/dummy_failure_jump to %d", p
+ mcnt
- start
);
850 case push_dummy_failure
:
851 printf ("/push_dummy_failure");
855 extract_number_and_incr (&mcnt
, &p
);
856 printf ("/maybe_pop_jump to %d", p
+ mcnt
- start
);
859 case pop_failure_jump
:
860 extract_number_and_incr (&mcnt
, &p
);
861 printf ("/pop_failure_jump to %d", p
+ mcnt
- start
);
865 extract_number_and_incr (&mcnt
, &p
);
866 printf ("/jump_past_alt to %d", p
+ mcnt
- start
);
870 extract_number_and_incr (&mcnt
, &p
);
871 printf ("/jump to %d", p
+ mcnt
- start
);
875 extract_number_and_incr (&mcnt
, &p
);
876 extract_number_and_incr (&mcnt2
, &p
);
877 printf ("/succeed_n to %d, %d times", p
+ mcnt
- start
, mcnt2
);
881 extract_number_and_incr (&mcnt
, &p
);
882 extract_number_and_incr (&mcnt2
, &p
);
883 printf ("/jump_n to %d, %d times", p
+ mcnt
- start
, mcnt2
);
887 extract_number_and_incr (&mcnt
, &p
);
888 extract_number_and_incr (&mcnt2
, &p
);
889 printf ("/set_number_at location %d to %d", p
+ mcnt
- start
, mcnt2
);
893 printf ("/wordbound");
897 printf ("/notwordbound");
909 printf ("/before_dot");
917 printf ("/after_dot");
921 printf ("/syntaxspec");
923 printf ("/%d", mcnt
);
927 printf ("/notsyntaxspec");
929 printf ("/%d", mcnt
);
934 printf ("/wordchar");
938 printf ("/notwordchar");
950 printf ("?%d", *(p
-1));
956 printf ("%d:\tend of pattern.\n", p
- start
);
961 print_compiled_pattern (bufp
)
962 struct re_pattern_buffer
*bufp
;
964 unsigned char *buffer
= bufp
->buffer
;
966 print_partial_compiled_pattern (buffer
, buffer
+ bufp
->used
);
967 printf ("%d bytes used/%d bytes allocated.\n", bufp
->used
, bufp
->allocated
);
969 if (bufp
->fastmap_accurate
&& bufp
->fastmap
)
971 printf ("fastmap: ");
972 print_fastmap (bufp
->fastmap
);
975 printf ("re_nsub: %d\t", bufp
->re_nsub
);
976 printf ("regs_alloc: %d\t", bufp
->regs_allocated
);
977 printf ("can_be_null: %d\t", bufp
->can_be_null
);
978 printf ("newline_anchor: %d\n", bufp
->newline_anchor
);
979 printf ("no_sub: %d\t", bufp
->no_sub
);
980 printf ("not_bol: %d\t", bufp
->not_bol
);
981 printf ("not_eol: %d\t", bufp
->not_eol
);
982 printf ("syntax: %d\n", bufp
->syntax
);
983 /* Perhaps we should print the translate table? */
988 print_double_string (where
, string1
, size1
, string2
, size2
)
1001 if (FIRST_STRING_P (where
))
1003 for (this_char
= where
- string1
; this_char
< size1
; this_char
++)
1004 putchar (string1
[this_char
]);
1009 for (this_char
= where
- string2
; this_char
< size2
; this_char
++)
1010 putchar (string2
[this_char
]);
1014 #else /* not DEBUG */
1019 #define DEBUG_STATEMENT(e)
1020 #define DEBUG_PRINT1(x)
1021 #define DEBUG_PRINT2(x1, x2)
1022 #define DEBUG_PRINT3(x1, x2, x3)
1023 #define DEBUG_PRINT4(x1, x2, x3, x4)
1024 #define DEBUG_PRINT_COMPILED_PATTERN(p, s, e)
1025 #define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2)
1027 #endif /* not DEBUG */
1029 /* Set by `re_set_syntax' to the current regexp syntax to recognize. Can
1030 also be assigned to arbitrarily: each pattern buffer stores its own
1031 syntax, so it can be changed between regex compilations. */
1032 /* This has no initializer because initialized variables in Emacs
1033 become read-only after dumping. */
1034 reg_syntax_t re_syntax_options
;
1037 /* Specify the precise syntax of regexps for compilation. This provides
1038 for compatibility for various utilities which historically have
1039 different, incompatible syntaxes.
1041 The argument SYNTAX is a bit mask comprised of the various bits
1042 defined in regex.h. We return the old syntax. */
1045 re_set_syntax (syntax
)
1046 reg_syntax_t syntax
;
1048 reg_syntax_t ret
= re_syntax_options
;
1050 re_syntax_options
= syntax
;
1054 /* This table gives an error message for each of the error codes listed
1055 in regex.h. Obviously the order here has to be same as there.
1056 POSIX doesn't require that we do anything for REG_NOERROR,
1057 but why not be nice? */
1059 static const char *re_error_msgid
[] =
1061 gettext_noop ("Success"), /* REG_NOERROR */
1062 gettext_noop ("No match"), /* REG_NOMATCH */
1063 gettext_noop ("Invalid regular expression"), /* REG_BADPAT */
1064 gettext_noop ("Invalid collation character"), /* REG_ECOLLATE */
1065 gettext_noop ("Invalid character class name"), /* REG_ECTYPE */
1066 gettext_noop ("Trailing backslash"), /* REG_EESCAPE */
1067 gettext_noop ("Invalid back reference"), /* REG_ESUBREG */
1068 gettext_noop ("Unmatched [ or [^"), /* REG_EBRACK */
1069 gettext_noop ("Unmatched ( or \\("), /* REG_EPAREN */
1070 gettext_noop ("Unmatched \\{"), /* REG_EBRACE */
1071 gettext_noop ("Invalid content of \\{\\}"), /* REG_BADBR */
1072 gettext_noop ("Invalid range end"), /* REG_ERANGE */
1073 gettext_noop ("Memory exhausted"), /* REG_ESPACE */
1074 gettext_noop ("Invalid preceding regular expression"), /* REG_BADRPT */
1075 gettext_noop ("Premature end of regular expression"), /* REG_EEND */
1076 gettext_noop ("Regular expression too big"), /* REG_ESIZE */
1077 gettext_noop ("Unmatched ) or \\)"), /* REG_ERPAREN */
1080 /* Avoiding alloca during matching, to placate r_alloc. */
1082 /* Define MATCH_MAY_ALLOCATE unless we need to make sure that the
1083 searching and matching functions should not call alloca. On some
1084 systems, alloca is implemented in terms of malloc, and if we're
1085 using the relocating allocator routines, then malloc could cause a
1086 relocation, which might (if the strings being searched are in the
1087 ralloc heap) shift the data out from underneath the regexp
1090 Here's another reason to avoid allocation: Emacs
1091 processes input from X in a signal handler; processing X input may
1092 call malloc; if input arrives while a matching routine is calling
1093 malloc, then we're scrod. But Emacs can't just block input while
1094 calling matching routines; then we don't notice interrupts when
1095 they come in. So, Emacs blocks input around all regexp calls
1096 except the matching calls, which it leaves unprotected, in the
1097 faith that they will not malloc. */
1099 /* Normally, this is fine. */
1100 #define MATCH_MAY_ALLOCATE
1102 /* When using GNU C, we are not REALLY using the C alloca, no matter
1103 what config.h may say. So don't take precautions for it. */
1108 /* The match routines may not allocate if (1) they would do it with malloc
1109 and (2) it's not safe for them to use malloc.
1110 Note that if REL_ALLOC is defined, matching would not use malloc for the
1111 failure stack, but we would still use it for the register vectors;
1112 so REL_ALLOC should not affect this. */
1113 #if (defined (C_ALLOCA) || defined (REGEX_MALLOC)) && defined (emacs)
1114 #undef MATCH_MAY_ALLOCATE
1118 /* Failure stack declarations and macros; both re_compile_fastmap and
1119 re_match_2 use a failure stack. These have to be macros because of
1120 REGEX_ALLOCATE_STACK. */
1123 /* Approximate number of failure points for which to initially allocate space
1124 when matching. If this number is exceeded, we allocate more
1125 space, so it is not a hard limit. */
1126 #ifndef INIT_FAILURE_ALLOC
1127 #define INIT_FAILURE_ALLOC 20
1130 /* Roughly the maximum number of failure points on the stack. Would be
1131 exactly that if always used TYPICAL_FAILURE_SIZE items each time we failed.
1132 This is a variable only so users of regex can assign to it; we never
1133 change it ourselves. */
1134 #if defined (MATCH_MAY_ALLOCATE)
1135 /* Note that 4400 is enough to cause a crash on Alpha OSF/1,
1136 whose default stack limit is 2mb. In order for a larger
1137 value to work reliably, you have to try to make it accord
1138 with the process stack limit. */
1139 int re_max_failures
= 40000;
1141 int re_max_failures
= 4000;
1144 union fail_stack_elt
1146 unsigned char *pointer
;
1150 typedef union fail_stack_elt fail_stack_elt_t
;
1154 fail_stack_elt_t
*stack
;
1156 unsigned avail
; /* Offset of next open position. */
1159 #define FAIL_STACK_EMPTY() (fail_stack.avail == 0)
1160 #define FAIL_STACK_PTR_EMPTY() (fail_stack_ptr->avail == 0)
1161 #define FAIL_STACK_FULL() (fail_stack.avail == fail_stack.size)
1164 /* Define macros to initialize and free the failure stack.
1165 Do `return -2' if the alloc fails. */
1167 #ifdef MATCH_MAY_ALLOCATE
1168 #define INIT_FAIL_STACK() \
1170 fail_stack.stack = (fail_stack_elt_t *) \
1171 REGEX_ALLOCATE_STACK (INIT_FAILURE_ALLOC * TYPICAL_FAILURE_SIZE \
1172 * sizeof (fail_stack_elt_t)); \
1174 if (fail_stack.stack == NULL) \
1177 fail_stack.size = INIT_FAILURE_ALLOC; \
1178 fail_stack.avail = 0; \
1181 #define RESET_FAIL_STACK() REGEX_FREE_STACK (fail_stack.stack)
1183 #define INIT_FAIL_STACK() \
1185 fail_stack.avail = 0; \
1188 #define RESET_FAIL_STACK()
1192 /* Double the size of FAIL_STACK, up to a limit
1193 which allows approximately `re_max_failures' items.
1195 Return 1 if succeeds, and 0 if either ran out of memory
1196 allocating space for it or it was already too large.
1198 REGEX_REALLOCATE_STACK requires `destination' be declared. */
1200 /* Factor to increase the failure stack size by
1201 when we increase it.
1202 This used to be 2, but 2 was too wasteful
1203 because the old discarded stacks added up to as much space
1204 were as ultimate, maximum-size stack. */
1205 #define FAIL_STACK_GROWTH_FACTOR 4
1207 #define GROW_FAIL_STACK(fail_stack) \
1208 ((fail_stack).size >= re_max_failures * TYPICAL_FAILURE_SIZE \
1210 : ((fail_stack).stack \
1211 = (fail_stack_elt_t *) \
1212 REGEX_REALLOCATE_STACK ((fail_stack).stack, \
1213 (fail_stack).size * sizeof (fail_stack_elt_t), \
1214 MIN (re_max_failures * TYPICAL_FAILURE_SIZE, \
1215 ((fail_stack).size * sizeof (fail_stack_elt_t) \
1216 * FAIL_STACK_GROWTH_FACTOR))), \
1218 (fail_stack).stack == NULL \
1220 : (MIN (re_max_failures * TYPICAL_FAILURE_SIZE, \
1221 ((fail_stack).size * sizeof (fail_stack_elt_t) \
1222 * FAIL_STACK_GROWTH_FACTOR)), \
1226 /* Push pointer POINTER on FAIL_STACK.
1227 Return 1 if was able to do so and 0 if ran out of memory allocating
1229 #define PUSH_PATTERN_OP(POINTER, FAIL_STACK) \
1230 ((FAIL_STACK_FULL () \
1231 && !GROW_FAIL_STACK (FAIL_STACK)) \
1233 : ((FAIL_STACK).stack[(FAIL_STACK).avail++].pointer = POINTER, \
1236 /* Push a pointer value onto the failure stack.
1237 Assumes the variable `fail_stack'. Probably should only
1238 be called from within `PUSH_FAILURE_POINT'. */
1239 #define PUSH_FAILURE_POINTER(item) \
1240 fail_stack.stack[fail_stack.avail++].pointer = (unsigned char *) (item)
1242 /* This pushes an integer-valued item onto the failure stack.
1243 Assumes the variable `fail_stack'. Probably should only
1244 be called from within `PUSH_FAILURE_POINT'. */
1245 #define PUSH_FAILURE_INT(item) \
1246 fail_stack.stack[fail_stack.avail++].integer = (item)
1248 /* Push a fail_stack_elt_t value onto the failure stack.
1249 Assumes the variable `fail_stack'. Probably should only
1250 be called from within `PUSH_FAILURE_POINT'. */
1251 #define PUSH_FAILURE_ELT(item) \
1252 fail_stack.stack[fail_stack.avail++] = (item)
1254 /* These three POP... operations complement the three PUSH... operations.
1255 All assume that `fail_stack' is nonempty. */
1256 #define POP_FAILURE_POINTER() fail_stack.stack[--fail_stack.avail].pointer
1257 #define POP_FAILURE_INT() fail_stack.stack[--fail_stack.avail].integer
1258 #define POP_FAILURE_ELT() fail_stack.stack[--fail_stack.avail]
1260 /* Used to omit pushing failure point id's when we're not debugging. */
1262 #define DEBUG_PUSH PUSH_FAILURE_INT
1263 #define DEBUG_POP(item_addr) *(item_addr) = POP_FAILURE_INT ()
1265 #define DEBUG_PUSH(item)
1266 #define DEBUG_POP(item_addr)
1270 /* Push the information about the state we will need
1271 if we ever fail back to it.
1273 Requires variables fail_stack, regstart, regend, reg_info, and
1274 num_regs be declared. GROW_FAIL_STACK requires `destination' be
1277 Does `return FAILURE_CODE' if runs out of memory. */
1279 #define PUSH_FAILURE_POINT(pattern_place, string_place, failure_code) \
1281 char *destination; \
1282 /* Must be int, so when we don't save any registers, the arithmetic \
1283 of 0 + -1 isn't done as unsigned. */ \
1286 DEBUG_STATEMENT (failure_id++); \
1287 DEBUG_STATEMENT (nfailure_points_pushed++); \
1288 DEBUG_PRINT2 ("\nPUSH_FAILURE_POINT #%u:\n", failure_id); \
1289 DEBUG_PRINT2 (" Before push, next avail: %d\n", (fail_stack).avail);\
1290 DEBUG_PRINT2 (" size: %d\n", (fail_stack).size);\
1292 DEBUG_PRINT2 (" slots needed: %d\n", NUM_FAILURE_ITEMS); \
1293 DEBUG_PRINT2 (" available: %d\n", REMAINING_AVAIL_SLOTS); \
1295 /* Ensure we have enough space allocated for what we will push. */ \
1296 while (REMAINING_AVAIL_SLOTS < NUM_FAILURE_ITEMS) \
1298 if (!GROW_FAIL_STACK (fail_stack)) \
1299 return failure_code; \
1301 DEBUG_PRINT2 ("\n Doubled stack; size now: %d\n", \
1302 (fail_stack).size); \
1303 DEBUG_PRINT2 (" slots available: %d\n", REMAINING_AVAIL_SLOTS);\
1306 /* Push the info, starting with the registers. */ \
1307 DEBUG_PRINT1 ("\n"); \
1310 for (this_reg = lowest_active_reg; this_reg <= highest_active_reg; \
1313 DEBUG_PRINT2 (" Pushing reg: %d\n", this_reg); \
1314 DEBUG_STATEMENT (num_regs_pushed++); \
1316 DEBUG_PRINT2 (" start: 0x%x\n", regstart[this_reg]); \
1317 PUSH_FAILURE_POINTER (regstart[this_reg]); \
1319 DEBUG_PRINT2 (" end: 0x%x\n", regend[this_reg]); \
1320 PUSH_FAILURE_POINTER (regend[this_reg]); \
1322 DEBUG_PRINT2 (" info: 0x%x\n ", reg_info[this_reg]); \
1323 DEBUG_PRINT2 (" match_null=%d", \
1324 REG_MATCH_NULL_STRING_P (reg_info[this_reg])); \
1325 DEBUG_PRINT2 (" active=%d", IS_ACTIVE (reg_info[this_reg])); \
1326 DEBUG_PRINT2 (" matched_something=%d", \
1327 MATCHED_SOMETHING (reg_info[this_reg])); \
1328 DEBUG_PRINT2 (" ever_matched=%d", \
1329 EVER_MATCHED_SOMETHING (reg_info[this_reg])); \
1330 DEBUG_PRINT1 ("\n"); \
1331 PUSH_FAILURE_ELT (reg_info[this_reg].word); \
1334 DEBUG_PRINT2 (" Pushing low active reg: %d\n", lowest_active_reg);\
1335 PUSH_FAILURE_INT (lowest_active_reg); \
1337 DEBUG_PRINT2 (" Pushing high active reg: %d\n", highest_active_reg);\
1338 PUSH_FAILURE_INT (highest_active_reg); \
1340 DEBUG_PRINT2 (" Pushing pattern 0x%x: ", pattern_place); \
1341 DEBUG_PRINT_COMPILED_PATTERN (bufp, pattern_place, pend); \
1342 PUSH_FAILURE_POINTER (pattern_place); \
1344 DEBUG_PRINT2 (" Pushing string 0x%x: `", string_place); \
1345 DEBUG_PRINT_DOUBLE_STRING (string_place, string1, size1, string2, \
1347 DEBUG_PRINT1 ("'\n"); \
1348 PUSH_FAILURE_POINTER (string_place); \
1350 DEBUG_PRINT2 (" Pushing failure id: %u\n", failure_id); \
1351 DEBUG_PUSH (failure_id); \
1354 /* This is the number of items that are pushed and popped on the stack
1355 for each register. */
1356 #define NUM_REG_ITEMS 3
1358 /* Individual items aside from the registers. */
1360 #define NUM_NONREG_ITEMS 5 /* Includes failure point id. */
1362 #define NUM_NONREG_ITEMS 4
1365 /* Estimate the size of data pushed by a typical failure stack entry.
1366 An estimate is all we need, because all we use this for
1367 is to choose a limit for how big to make the failure stack. */
1369 #define TYPICAL_FAILURE_SIZE 20
1371 /* This is how many items we actually use for a failure point.
1372 It depends on the regexp. */
1373 #define NUM_FAILURE_ITEMS \
1375 ? 0 : highest_active_reg - lowest_active_reg + 1) \
1379 /* How many items can still be added to the stack without overflowing it. */
1380 #define REMAINING_AVAIL_SLOTS ((fail_stack).size - (fail_stack).avail)
1383 /* Pops what PUSH_FAIL_STACK pushes.
1385 We restore into the parameters, all of which should be lvalues:
1386 STR -- the saved data position.
1387 PAT -- the saved pattern position.
1388 LOW_REG, HIGH_REG -- the highest and lowest active registers.
1389 REGSTART, REGEND -- arrays of string positions.
1390 REG_INFO -- array of information about each subexpression.
1392 Also assumes the variables `fail_stack' and (if debugging), `bufp',
1393 `pend', `string1', `size1', `string2', and `size2'. */
1395 #define POP_FAILURE_POINT(str, pat, low_reg, high_reg, regstart, regend, reg_info)\
1397 DEBUG_STATEMENT (fail_stack_elt_t failure_id;) \
1399 const unsigned char *string_temp; \
1401 assert (!FAIL_STACK_EMPTY ()); \
1403 /* Remove failure points and point to how many regs pushed. */ \
1404 DEBUG_PRINT1 ("POP_FAILURE_POINT:\n"); \
1405 DEBUG_PRINT2 (" Before pop, next avail: %d\n", fail_stack.avail); \
1406 DEBUG_PRINT2 (" size: %d\n", fail_stack.size); \
1408 assert (fail_stack.avail >= NUM_NONREG_ITEMS); \
1410 DEBUG_POP (&failure_id); \
1411 DEBUG_PRINT2 (" Popping failure id: %u\n", failure_id); \
1413 /* If the saved string location is NULL, it came from an \
1414 on_failure_keep_string_jump opcode, and we want to throw away the \
1415 saved NULL, thus retaining our current position in the string. */ \
1416 string_temp = POP_FAILURE_POINTER (); \
1417 if (string_temp != NULL) \
1418 str = (const char *) string_temp; \
1420 DEBUG_PRINT2 (" Popping string 0x%x: `", str); \
1421 DEBUG_PRINT_DOUBLE_STRING (str, string1, size1, string2, size2); \
1422 DEBUG_PRINT1 ("'\n"); \
1424 pat = (unsigned char *) POP_FAILURE_POINTER (); \
1425 DEBUG_PRINT2 (" Popping pattern 0x%x: ", pat); \
1426 DEBUG_PRINT_COMPILED_PATTERN (bufp, pat, pend); \
1428 /* Restore register info. */ \
1429 high_reg = (unsigned) POP_FAILURE_INT (); \
1430 DEBUG_PRINT2 (" Popping high active reg: %d\n", high_reg); \
1432 low_reg = (unsigned) POP_FAILURE_INT (); \
1433 DEBUG_PRINT2 (" Popping low active reg: %d\n", low_reg); \
1436 for (this_reg = high_reg; this_reg >= low_reg; this_reg--) \
1438 DEBUG_PRINT2 (" Popping reg: %d\n", this_reg); \
1440 reg_info[this_reg].word = POP_FAILURE_ELT (); \
1441 DEBUG_PRINT2 (" info: 0x%x\n", reg_info[this_reg]); \
1443 regend[this_reg] = (const char *) POP_FAILURE_POINTER (); \
1444 DEBUG_PRINT2 (" end: 0x%x\n", regend[this_reg]); \
1446 regstart[this_reg] = (const char *) POP_FAILURE_POINTER (); \
1447 DEBUG_PRINT2 (" start: 0x%x\n", regstart[this_reg]); \
1451 for (this_reg = highest_active_reg; this_reg > high_reg; this_reg--) \
1453 reg_info[this_reg].word.integer = 0; \
1454 regend[this_reg] = 0; \
1455 regstart[this_reg] = 0; \
1457 highest_active_reg = high_reg; \
1460 set_regs_matched_done = 0; \
1461 DEBUG_STATEMENT (nfailure_points_popped++); \
1462 } /* POP_FAILURE_POINT */
1466 /* Structure for per-register (a.k.a. per-group) information.
1467 Other register information, such as the
1468 starting and ending positions (which are addresses), and the list of
1469 inner groups (which is a bits list) are maintained in separate
1472 We are making a (strictly speaking) nonportable assumption here: that
1473 the compiler will pack our bit fields into something that fits into
1474 the type of `word', i.e., is something that fits into one item on the
1479 fail_stack_elt_t word
;
1482 /* This field is one if this group can match the empty string,
1483 zero if not. If not yet determined, `MATCH_NULL_UNSET_VALUE'. */
1484 #define MATCH_NULL_UNSET_VALUE 3
1485 unsigned match_null_string_p
: 2;
1486 unsigned is_active
: 1;
1487 unsigned matched_something
: 1;
1488 unsigned ever_matched_something
: 1;
1490 } register_info_type
;
1492 #define REG_MATCH_NULL_STRING_P(R) ((R).bits.match_null_string_p)
1493 #define IS_ACTIVE(R) ((R).bits.is_active)
1494 #define MATCHED_SOMETHING(R) ((R).bits.matched_something)
1495 #define EVER_MATCHED_SOMETHING(R) ((R).bits.ever_matched_something)
1498 /* Call this when have matched a real character; it sets `matched' flags
1499 for the subexpressions which we are currently inside. Also records
1500 that those subexprs have matched. */
1501 #define SET_REGS_MATCHED() \
1504 if (!set_regs_matched_done) \
1507 set_regs_matched_done = 1; \
1508 for (r = lowest_active_reg; r <= highest_active_reg; r++) \
1510 MATCHED_SOMETHING (reg_info[r]) \
1511 = EVER_MATCHED_SOMETHING (reg_info[r]) \
1518 /* Registers are set to a sentinel when they haven't yet matched. */
1519 static char reg_unset_dummy
;
1520 #define REG_UNSET_VALUE (®_unset_dummy)
1521 #define REG_UNSET(e) ((e) == REG_UNSET_VALUE)
1523 /* Subroutine declarations and macros for regex_compile. */
1525 static void store_op1 (), store_op2 ();
1526 static void insert_op1 (), insert_op2 ();
1527 static boolean
at_begline_loc_p (), at_endline_loc_p ();
1528 static boolean
group_in_compile_stack ();
1529 static reg_errcode_t
compile_range ();
1531 /* Fetch the next character in the uncompiled pattern---translating it
1532 if necessary. Also cast from a signed character in the constant
1533 string passed to us by the user to an unsigned char that we can use
1534 as an array index (in, e.g., `translate'). */
1536 #define PATFETCH(c) \
1537 do {if (p == pend) return REG_EEND; \
1538 c = (unsigned char) *p++; \
1539 if (translate) c = (unsigned char) translate[c]; \
1543 /* Fetch the next character in the uncompiled pattern, with no
1545 #define PATFETCH_RAW(c) \
1546 do {if (p == pend) return REG_EEND; \
1547 c = (unsigned char) *p++; \
1550 /* Go backwards one character in the pattern. */
1551 #define PATUNFETCH p--
1554 /* If `translate' is non-null, return translate[D], else just D. We
1555 cast the subscript to translate because some data is declared as
1556 `char *', to avoid warnings when a string constant is passed. But
1557 when we use a character as a subscript we must make it unsigned. */
1559 #define TRANSLATE(d) \
1560 (translate ? (unsigned char) RE_TRANSLATE (translate, (unsigned char) (d)) : (d))
1564 /* Macros for outputting the compiled pattern into `buffer'. */
1566 /* If the buffer isn't allocated when it comes in, use this. */
1567 #define INIT_BUF_SIZE 32
1569 /* Make sure we have at least N more bytes of space in buffer. */
1570 #define GET_BUFFER_SPACE(n) \
1571 while (b - bufp->buffer + (n) > bufp->allocated) \
1574 /* Make sure we have one more byte of buffer space and then add C to it. */
1575 #define BUF_PUSH(c) \
1577 GET_BUFFER_SPACE (1); \
1578 *b++ = (unsigned char) (c); \
1582 /* Ensure we have two more bytes of buffer space and then append C1 and C2. */
1583 #define BUF_PUSH_2(c1, c2) \
1585 GET_BUFFER_SPACE (2); \
1586 *b++ = (unsigned char) (c1); \
1587 *b++ = (unsigned char) (c2); \
1591 /* As with BUF_PUSH_2, except for three bytes. */
1592 #define BUF_PUSH_3(c1, c2, c3) \
1594 GET_BUFFER_SPACE (3); \
1595 *b++ = (unsigned char) (c1); \
1596 *b++ = (unsigned char) (c2); \
1597 *b++ = (unsigned char) (c3); \
1601 /* Store a jump with opcode OP at LOC to location TO. We store a
1602 relative address offset by the three bytes the jump itself occupies. */
1603 #define STORE_JUMP(op, loc, to) \
1604 store_op1 (op, loc, (to) - (loc) - 3)
1606 /* Likewise, for a two-argument jump. */
1607 #define STORE_JUMP2(op, loc, to, arg) \
1608 store_op2 (op, loc, (to) - (loc) - 3, arg)
1610 /* Like `STORE_JUMP', but for inserting. Assume `b' is the buffer end. */
1611 #define INSERT_JUMP(op, loc, to) \
1612 insert_op1 (op, loc, (to) - (loc) - 3, b)
1614 /* Like `STORE_JUMP2', but for inserting. Assume `b' is the buffer end. */
1615 #define INSERT_JUMP2(op, loc, to, arg) \
1616 insert_op2 (op, loc, (to) - (loc) - 3, arg, b)
1619 /* This is not an arbitrary limit: the arguments which represent offsets
1620 into the pattern are two bytes long. So if 2^16 bytes turns out to
1621 be too small, many things would have to change. */
1622 #define MAX_BUF_SIZE (1L << 16)
1625 /* Extend the buffer by twice its current size via realloc and
1626 reset the pointers that pointed into the old block to point to the
1627 correct places in the new one. If extending the buffer results in it
1628 being larger than MAX_BUF_SIZE, then flag memory exhausted. */
1629 #define EXTEND_BUFFER() \
1631 unsigned char *old_buffer = bufp->buffer; \
1632 if (bufp->allocated == MAX_BUF_SIZE) \
1634 bufp->allocated <<= 1; \
1635 if (bufp->allocated > MAX_BUF_SIZE) \
1636 bufp->allocated = MAX_BUF_SIZE; \
1637 bufp->buffer = (unsigned char *) realloc (bufp->buffer, bufp->allocated);\
1638 if (bufp->buffer == NULL) \
1639 return REG_ESPACE; \
1640 /* If the buffer moved, move all the pointers into it. */ \
1641 if (old_buffer != bufp->buffer) \
1643 b = (b - old_buffer) + bufp->buffer; \
1644 begalt = (begalt - old_buffer) + bufp->buffer; \
1645 if (fixup_alt_jump) \
1646 fixup_alt_jump = (fixup_alt_jump - old_buffer) + bufp->buffer;\
1648 laststart = (laststart - old_buffer) + bufp->buffer; \
1649 if (pending_exact) \
1650 pending_exact = (pending_exact - old_buffer) + bufp->buffer; \
1655 /* Since we have one byte reserved for the register number argument to
1656 {start,stop}_memory, the maximum number of groups we can report
1657 things about is what fits in that byte. */
1658 #define MAX_REGNUM 255
1660 /* But patterns can have more than `MAX_REGNUM' registers. We just
1661 ignore the excess. */
1662 typedef unsigned regnum_t
;
1665 /* Macros for the compile stack. */
1667 /* Since offsets can go either forwards or backwards, this type needs to
1668 be able to hold values from -(MAX_BUF_SIZE - 1) to MAX_BUF_SIZE - 1. */
1669 typedef int pattern_offset_t
;
1673 pattern_offset_t begalt_offset
;
1674 pattern_offset_t fixup_alt_jump
;
1675 pattern_offset_t inner_group_offset
;
1676 pattern_offset_t laststart_offset
;
1678 } compile_stack_elt_t
;
1683 compile_stack_elt_t
*stack
;
1685 unsigned avail
; /* Offset of next open position. */
1686 } compile_stack_type
;
1689 #define INIT_COMPILE_STACK_SIZE 32
1691 #define COMPILE_STACK_EMPTY (compile_stack.avail == 0)
1692 #define COMPILE_STACK_FULL (compile_stack.avail == compile_stack.size)
1694 /* The next available element. */
1695 #define COMPILE_STACK_TOP (compile_stack.stack[compile_stack.avail])
1698 /* Structure to manage work area for range table. */
1699 struct range_table_work_area
1701 int *table
; /* actual work area. */
1702 int allocated
; /* allocated size for work area in bytes. */
1703 int used
; /* actually used size in words. */
1706 /* Make sure that WORK_AREA can hold more N multibyte characters. */
1707 #define EXTEND_RANGE_TABLE_WORK_AREA(work_area, n) \
1709 if (((work_area).used + (n)) * sizeof (int) > (work_area).allocated) \
1711 (work_area).allocated += 16 * sizeof (int); \
1712 if ((work_area).table) \
1714 = (int *) realloc ((work_area).table, (work_area).allocated); \
1717 = (int *) malloc ((work_area).allocated); \
1718 if ((work_area).table == 0) \
1719 FREE_STACK_RETURN (REG_ESPACE); \
1723 /* Set a range (RANGE_START, RANGE_END) to WORK_AREA. */
1724 #define SET_RANGE_TABLE_WORK_AREA(work_area, range_start, range_end) \
1726 EXTEND_RANGE_TABLE_WORK_AREA ((work_area), 2); \
1727 (work_area).table[(work_area).used++] = (range_start); \
1728 (work_area).table[(work_area).used++] = (range_end); \
1731 /* Free allocated memory for WORK_AREA. */
1732 #define FREE_RANGE_TABLE_WORK_AREA(work_area) \
1734 if ((work_area).table) \
1735 free ((work_area).table); \
1738 #define CLEAR_RANGE_TABLE_WORK_USED(work_area) ((work_area).used = 0)
1739 #define RANGE_TABLE_WORK_USED(work_area) ((work_area).used)
1740 #define RANGE_TABLE_WORK_ELT(work_area, i) ((work_area).table[i])
1743 /* Set the bit for character C in a list. */
1744 #define SET_LIST_BIT(c) \
1745 (b[((unsigned char) (c)) / BYTEWIDTH] \
1746 |= 1 << (((unsigned char) c) % BYTEWIDTH))
1749 /* Get the next unsigned number in the uncompiled pattern. */
1750 #define GET_UNSIGNED_NUMBER(num) \
1754 while (ISDIGIT (c)) \
1758 num = num * 10 + c - '0'; \
1766 #define CHAR_CLASS_MAX_LENGTH 6 /* Namely, `xdigit'. */
1768 #define IS_CHAR_CLASS(string) \
1769 (STREQ (string, "alpha") || STREQ (string, "upper") \
1770 || STREQ (string, "lower") || STREQ (string, "digit") \
1771 || STREQ (string, "alnum") || STREQ (string, "xdigit") \
1772 || STREQ (string, "space") || STREQ (string, "print") \
1773 || STREQ (string, "punct") || STREQ (string, "graph") \
1774 || STREQ (string, "cntrl") || STREQ (string, "blank"))
1776 #ifndef MATCH_MAY_ALLOCATE
1778 /* If we cannot allocate large objects within re_match_2_internal,
1779 we make the fail stack and register vectors global.
1780 The fail stack, we grow to the maximum size when a regexp
1782 The register vectors, we adjust in size each time we
1783 compile a regexp, according to the number of registers it needs. */
1785 static fail_stack_type fail_stack
;
1787 /* Size with which the following vectors are currently allocated.
1788 That is so we can make them bigger as needed,
1789 but never make them smaller. */
1790 static int regs_allocated_size
;
1792 static const char ** regstart
, ** regend
;
1793 static const char ** old_regstart
, ** old_regend
;
1794 static const char **best_regstart
, **best_regend
;
1795 static register_info_type
*reg_info
;
1796 static const char **reg_dummy
;
1797 static register_info_type
*reg_info_dummy
;
1799 /* Make the register vectors big enough for NUM_REGS registers,
1800 but don't make them smaller. */
1803 regex_grow_registers (num_regs
)
1806 if (num_regs
> regs_allocated_size
)
1808 RETALLOC_IF (regstart
, num_regs
, const char *);
1809 RETALLOC_IF (regend
, num_regs
, const char *);
1810 RETALLOC_IF (old_regstart
, num_regs
, const char *);
1811 RETALLOC_IF (old_regend
, num_regs
, const char *);
1812 RETALLOC_IF (best_regstart
, num_regs
, const char *);
1813 RETALLOC_IF (best_regend
, num_regs
, const char *);
1814 RETALLOC_IF (reg_info
, num_regs
, register_info_type
);
1815 RETALLOC_IF (reg_dummy
, num_regs
, const char *);
1816 RETALLOC_IF (reg_info_dummy
, num_regs
, register_info_type
);
1818 regs_allocated_size
= num_regs
;
1822 #endif /* not MATCH_MAY_ALLOCATE */
1824 /* `regex_compile' compiles PATTERN (of length SIZE) according to SYNTAX.
1825 Returns one of error codes defined in `regex.h', or zero for success.
1827 Assumes the `allocated' (and perhaps `buffer') and `translate'
1828 fields are set in BUFP on entry.
1830 If it succeeds, results are put in BUFP (if it returns an error, the
1831 contents of BUFP are undefined):
1832 `buffer' is the compiled pattern;
1833 `syntax' is set to SYNTAX;
1834 `used' is set to the length of the compiled pattern;
1835 `fastmap_accurate' is zero;
1836 `re_nsub' is the number of subexpressions in PATTERN;
1837 `not_bol' and `not_eol' are zero;
1839 The `fastmap' and `newline_anchor' fields are neither
1840 examined nor set. */
1842 /* Return, freeing storage we allocated. */
1843 #define FREE_STACK_RETURN(value) \
1845 FREE_RANGE_TABLE_WORK_AREA (range_table_work); \
1846 free (compile_stack.stack); \
1850 static reg_errcode_t
1851 regex_compile (pattern
, size
, syntax
, bufp
)
1852 const char *pattern
;
1854 reg_syntax_t syntax
;
1855 struct re_pattern_buffer
*bufp
;
1857 /* We fetch characters from PATTERN here. Even though PATTERN is
1858 `char *' (i.e., signed), we declare these variables as unsigned, so
1859 they can be reliably used as array indices. */
1860 register unsigned int c
, c1
;
1862 /* A random temporary spot in PATTERN. */
1865 /* Points to the end of the buffer, where we should append. */
1866 register unsigned char *b
;
1868 /* Keeps track of unclosed groups. */
1869 compile_stack_type compile_stack
;
1871 /* Points to the current (ending) position in the pattern. */
1872 const char *p
= pattern
;
1873 const char *pend
= pattern
+ size
;
1875 /* How to translate the characters in the pattern. */
1876 RE_TRANSLATE_TYPE translate
= bufp
->translate
;
1878 /* Address of the count-byte of the most recently inserted `exactn'
1879 command. This makes it possible to tell if a new exact-match
1880 character can be added to that command or if the character requires
1881 a new `exactn' command. */
1882 unsigned char *pending_exact
= 0;
1884 /* Address of start of the most recently finished expression.
1885 This tells, e.g., postfix * where to find the start of its
1886 operand. Reset at the beginning of groups and alternatives. */
1887 unsigned char *laststart
= 0;
1889 /* Address of beginning of regexp, or inside of last group. */
1890 unsigned char *begalt
;
1892 /* Place in the uncompiled pattern (i.e., the {) to
1893 which to go back if the interval is invalid. */
1894 const char *beg_interval
;
1896 /* Address of the place where a forward jump should go to the end of
1897 the containing expression. Each alternative of an `or' -- except the
1898 last -- ends with a forward jump of this sort. */
1899 unsigned char *fixup_alt_jump
= 0;
1901 /* Counts open-groups as they are encountered. Remembered for the
1902 matching close-group on the compile stack, so the same register
1903 number is put in the stop_memory as the start_memory. */
1904 regnum_t regnum
= 0;
1906 /* Work area for range table of charset. */
1907 struct range_table_work_area range_table_work
;
1910 DEBUG_PRINT1 ("\nCompiling pattern: ");
1913 unsigned debug_count
;
1915 for (debug_count
= 0; debug_count
< size
; debug_count
++)
1916 putchar (pattern
[debug_count
]);
1921 /* Initialize the compile stack. */
1922 compile_stack
.stack
= TALLOC (INIT_COMPILE_STACK_SIZE
, compile_stack_elt_t
);
1923 if (compile_stack
.stack
== NULL
)
1926 compile_stack
.size
= INIT_COMPILE_STACK_SIZE
;
1927 compile_stack
.avail
= 0;
1929 range_table_work
.table
= 0;
1930 range_table_work
.allocated
= 0;
1932 /* Initialize the pattern buffer. */
1933 bufp
->syntax
= syntax
;
1934 bufp
->fastmap_accurate
= 0;
1935 bufp
->not_bol
= bufp
->not_eol
= 0;
1937 /* Set `used' to zero, so that if we return an error, the pattern
1938 printer (for debugging) will think there's no pattern. We reset it
1942 /* Always count groups, whether or not bufp->no_sub is set. */
1946 /* bufp->multibyte is set before regex_compile is called, so don't alter
1948 #else /* not emacs */
1949 /* Nothing is recognized as a multibyte character. */
1950 bufp
->multibyte
= 0;
1953 #if !defined (emacs) && !defined (SYNTAX_TABLE)
1954 /* Initialize the syntax table. */
1955 init_syntax_once ();
1958 if (bufp
->allocated
== 0)
1961 { /* If zero allocated, but buffer is non-null, try to realloc
1962 enough space. This loses if buffer's address is bogus, but
1963 that is the user's responsibility. */
1964 RETALLOC (bufp
->buffer
, INIT_BUF_SIZE
, unsigned char);
1967 { /* Caller did not allocate a buffer. Do it for them. */
1968 bufp
->buffer
= TALLOC (INIT_BUF_SIZE
, unsigned char);
1970 if (!bufp
->buffer
) FREE_STACK_RETURN (REG_ESPACE
);
1972 bufp
->allocated
= INIT_BUF_SIZE
;
1975 begalt
= b
= bufp
->buffer
;
1977 /* Loop through the uncompiled pattern until we're at the end. */
1986 if ( /* If at start of pattern, it's an operator. */
1988 /* If context independent, it's an operator. */
1989 || syntax
& RE_CONTEXT_INDEP_ANCHORS
1990 /* Otherwise, depends on what's come before. */
1991 || at_begline_loc_p (pattern
, p
, syntax
))
2001 if ( /* If at end of pattern, it's an operator. */
2003 /* If context independent, it's an operator. */
2004 || syntax
& RE_CONTEXT_INDEP_ANCHORS
2005 /* Otherwise, depends on what's next. */
2006 || at_endline_loc_p (p
, pend
, syntax
))
2016 if ((syntax
& RE_BK_PLUS_QM
)
2017 || (syntax
& RE_LIMITED_OPS
))
2021 /* If there is no previous pattern... */
2024 if (syntax
& RE_CONTEXT_INVALID_OPS
)
2025 FREE_STACK_RETURN (REG_BADRPT
);
2026 else if (!(syntax
& RE_CONTEXT_INDEP_OPS
))
2031 /* Are we optimizing this jump? */
2032 boolean keep_string_p
= false;
2034 /* 1 means zero (many) matches is allowed. */
2035 char zero_times_ok
= 0, many_times_ok
= 0;
2037 /* If there is a sequence of repetition chars, collapse it
2038 down to just one (the right one). We can't combine
2039 interval operators with these because of, e.g., `a{2}*',
2040 which should only match an even number of `a's. */
2044 zero_times_ok
|= c
!= '+';
2045 many_times_ok
|= c
!= '?';
2053 || (!(syntax
& RE_BK_PLUS_QM
) && (c
== '+' || c
== '?')))
2056 else if (syntax
& RE_BK_PLUS_QM
&& c
== '\\')
2058 if (p
== pend
) FREE_STACK_RETURN (REG_EESCAPE
);
2061 if (!(c1
== '+' || c1
== '?'))
2076 /* If we get here, we found another repeat character. */
2079 /* Star, etc. applied to an empty pattern is equivalent
2080 to an empty pattern. */
2084 /* Now we know whether or not zero matches is allowed
2085 and also whether or not two or more matches is allowed. */
2087 { /* More than one repetition is allowed, so put in at the
2088 end a backward relative jump from `b' to before the next
2089 jump we're going to put in below (which jumps from
2090 laststart to after this jump).
2092 But if we are at the `*' in the exact sequence `.*\n',
2093 insert an unconditional jump backwards to the .,
2094 instead of the beginning of the loop. This way we only
2095 push a failure point once, instead of every time
2096 through the loop. */
2097 assert (p
- 1 > pattern
);
2099 /* Allocate the space for the jump. */
2100 GET_BUFFER_SPACE (3);
2102 /* We know we are not at the first character of the pattern,
2103 because laststart was nonzero. And we've already
2104 incremented `p', by the way, to be the character after
2105 the `*'. Do we have to do something analogous here
2106 for null bytes, because of RE_DOT_NOT_NULL? */
2107 if (TRANSLATE (*(p
- 2)) == TRANSLATE ('.')
2109 && p
< pend
&& TRANSLATE (*p
) == TRANSLATE ('\n')
2110 && !(syntax
& RE_DOT_NEWLINE
))
2111 { /* We have .*\n. */
2112 STORE_JUMP (jump
, b
, laststart
);
2113 keep_string_p
= true;
2116 /* Anything else. */
2117 STORE_JUMP (maybe_pop_jump
, b
, laststart
- 3);
2119 /* We've added more stuff to the buffer. */
2123 /* On failure, jump from laststart to b + 3, which will be the
2124 end of the buffer after this jump is inserted. */
2125 GET_BUFFER_SPACE (3);
2126 INSERT_JUMP (keep_string_p
? on_failure_keep_string_jump
2134 /* At least one repetition is required, so insert a
2135 `dummy_failure_jump' before the initial
2136 `on_failure_jump' instruction of the loop. This
2137 effects a skip over that instruction the first time
2138 we hit that loop. */
2139 GET_BUFFER_SPACE (3);
2140 INSERT_JUMP (dummy_failure_jump
, laststart
, laststart
+ 6);
2155 CLEAR_RANGE_TABLE_WORK_USED (range_table_work
);
2157 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2159 /* Ensure that we have enough space to push a charset: the
2160 opcode, the length count, and the bitset; 34 bytes in all. */
2161 GET_BUFFER_SPACE (34);
2165 /* We test `*p == '^' twice, instead of using an if
2166 statement, so we only need one BUF_PUSH. */
2167 BUF_PUSH (*p
== '^' ? charset_not
: charset
);
2171 /* Remember the first position in the bracket expression. */
2174 /* Push the number of bytes in the bitmap. */
2175 BUF_PUSH ((1 << BYTEWIDTH
) / BYTEWIDTH
);
2177 /* Clear the whole map. */
2178 bzero (b
, (1 << BYTEWIDTH
) / BYTEWIDTH
);
2180 /* charset_not matches newline according to a syntax bit. */
2181 if ((re_opcode_t
) b
[-2] == charset_not
2182 && (syntax
& RE_HAT_LISTS_NOT_NEWLINE
))
2183 SET_LIST_BIT ('\n');
2185 /* Read in characters and ranges, setting map bits. */
2189 boolean escaped_char
= false;
2191 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2195 /* \ might escape characters inside [...] and [^...]. */
2196 if ((syntax
& RE_BACKSLASH_ESCAPE_IN_LISTS
) && c
== '\\')
2198 if (p
== pend
) FREE_STACK_RETURN (REG_EESCAPE
);
2201 escaped_char
= true;
2205 /* Could be the end of the bracket expression. If it's
2206 not (i.e., when the bracket expression is `[]' so
2207 far), the ']' character bit gets set way below. */
2208 if (c
== ']' && p
!= p1
+ 1)
2212 /* If C indicates start of multibyte char, get the
2213 actual character code in C, and set the pattern
2214 pointer P to the next character boundary. */
2215 if (bufp
->multibyte
&& BASE_LEADING_CODE_P (c
))
2218 c
= STRING_CHAR_AND_LENGTH (p
, pend
- p
, len
);
2221 /* What should we do for the character which is
2222 greater than 0x7F, but not BASE_LEADING_CODE_P?
2225 /* See if we're at the beginning of a possible character
2228 else if (!escaped_char
&&
2229 syntax
& RE_CHAR_CLASSES
&& c
== '[' && *p
== ':')
2231 /* Leave room for the null. */
2232 char str
[CHAR_CLASS_MAX_LENGTH
+ 1];
2237 /* If pattern is `[[:'. */
2238 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2243 if (c
== ':' || c
== ']' || p
== pend
2244 || c1
== CHAR_CLASS_MAX_LENGTH
)
2250 /* If isn't a word bracketed by `[:' and `:]':
2251 undo the ending character, the letters, and
2252 leave the leading `:' and `[' (but set bits for
2254 if (c
== ':' && *p
== ']')
2257 boolean is_alnum
= STREQ (str
, "alnum");
2258 boolean is_alpha
= STREQ (str
, "alpha");
2259 boolean is_blank
= STREQ (str
, "blank");
2260 boolean is_cntrl
= STREQ (str
, "cntrl");
2261 boolean is_digit
= STREQ (str
, "digit");
2262 boolean is_graph
= STREQ (str
, "graph");
2263 boolean is_lower
= STREQ (str
, "lower");
2264 boolean is_print
= STREQ (str
, "print");
2265 boolean is_punct
= STREQ (str
, "punct");
2266 boolean is_space
= STREQ (str
, "space");
2267 boolean is_upper
= STREQ (str
, "upper");
2268 boolean is_xdigit
= STREQ (str
, "xdigit");
2270 if (!IS_CHAR_CLASS (str
))
2271 FREE_STACK_RETURN (REG_ECTYPE
);
2273 /* Throw away the ] at the end of the character
2277 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2279 for (ch
= 0; ch
< 1 << BYTEWIDTH
; ch
++)
2281 int translated
= TRANSLATE (ch
);
2282 /* This was split into 3 if's to
2283 avoid an arbitrary limit in some compiler. */
2284 if ( (is_alnum
&& ISALNUM (ch
))
2285 || (is_alpha
&& ISALPHA (ch
))
2286 || (is_blank
&& ISBLANK (ch
))
2287 || (is_cntrl
&& ISCNTRL (ch
)))
2288 SET_LIST_BIT (translated
);
2289 if ( (is_digit
&& ISDIGIT (ch
))
2290 || (is_graph
&& ISGRAPH (ch
))
2291 || (is_lower
&& ISLOWER (ch
))
2292 || (is_print
&& ISPRINT (ch
)))
2293 SET_LIST_BIT (translated
);
2294 if ( (is_punct
&& ISPUNCT (ch
))
2295 || (is_space
&& ISSPACE (ch
))
2296 || (is_upper
&& ISUPPER (ch
))
2297 || (is_xdigit
&& ISXDIGIT (ch
)))
2298 SET_LIST_BIT (translated
);
2301 /* Repeat the loop. */
2311 /* Because the `:' may starts the range, we
2312 can't simply set bit and repeat the loop.
2313 Instead, just set it to C and handle below. */
2318 if (p
< pend
&& p
[0] == '-' && p
[1] != ']')
2321 /* Discard the `-'. */
2324 /* Fetch the character which ends the range. */
2326 if (bufp
->multibyte
&& BASE_LEADING_CODE_P (c1
))
2329 c1
= STRING_CHAR_AND_LENGTH (p
, pend
- p
, len
);
2333 if (!SAME_CHARSET_P (c
, c1
))
2334 FREE_STACK_RETURN (REG_ERANGE
);
2337 /* Range from C to C. */
2340 /* Set the range ... */
2341 if (SINGLE_BYTE_CHAR_P (c
))
2342 /* ... into bitmap. */
2345 int range_start
= c
, range_end
= c1
;
2347 /* If the start is after the end, the range is empty. */
2348 if (range_start
> range_end
)
2350 if (syntax
& RE_NO_EMPTY_RANGES
)
2351 FREE_STACK_RETURN (REG_ERANGE
);
2352 /* Else, repeat the loop. */
2356 for (this_char
= range_start
; this_char
<= range_end
;
2358 SET_LIST_BIT (TRANSLATE (this_char
));
2362 /* ... into range table. */
2363 SET_RANGE_TABLE_WORK_AREA (range_table_work
, c
, c1
);
2366 /* Discard any (non)matching list bytes that are all 0 at the
2367 end of the map. Decrease the map-length byte too. */
2368 while ((int) b
[-1] > 0 && b
[b
[-1] - 1] == 0)
2372 /* Build real range table from work area. */
2373 if (RANGE_TABLE_WORK_USED (range_table_work
))
2376 int used
= RANGE_TABLE_WORK_USED (range_table_work
);
2378 /* Allocate space for COUNT + RANGE_TABLE. Needs two
2379 bytes for COUNT and three bytes for each character. */
2380 GET_BUFFER_SPACE (2 + used
* 3);
2382 /* Indicate the existence of range table. */
2383 laststart
[1] |= 0x80;
2385 STORE_NUMBER_AND_INCR (b
, used
/ 2);
2386 for (i
= 0; i
< used
; i
++)
2387 STORE_CHARACTER_AND_INCR
2388 (b
, RANGE_TABLE_WORK_ELT (range_table_work
, i
));
2395 if (syntax
& RE_NO_BK_PARENS
)
2402 if (syntax
& RE_NO_BK_PARENS
)
2409 if (syntax
& RE_NEWLINE_ALT
)
2416 if (syntax
& RE_NO_BK_VBAR
)
2423 if (syntax
& RE_INTERVALS
&& syntax
& RE_NO_BK_BRACES
)
2424 goto handle_interval
;
2430 if (p
== pend
) FREE_STACK_RETURN (REG_EESCAPE
);
2432 /* Do not translate the character after the \, so that we can
2433 distinguish, e.g., \B from \b, even if we normally would
2434 translate, e.g., B to b. */
2440 if (syntax
& RE_NO_BK_PARENS
)
2441 goto normal_backslash
;
2447 if (COMPILE_STACK_FULL
)
2449 RETALLOC (compile_stack
.stack
, compile_stack
.size
<< 1,
2450 compile_stack_elt_t
);
2451 if (compile_stack
.stack
== NULL
) return REG_ESPACE
;
2453 compile_stack
.size
<<= 1;
2456 /* These are the values to restore when we hit end of this
2457 group. They are all relative offsets, so that if the
2458 whole pattern moves because of realloc, they will still
2460 COMPILE_STACK_TOP
.begalt_offset
= begalt
- bufp
->buffer
;
2461 COMPILE_STACK_TOP
.fixup_alt_jump
2462 = fixup_alt_jump
? fixup_alt_jump
- bufp
->buffer
+ 1 : 0;
2463 COMPILE_STACK_TOP
.laststart_offset
= b
- bufp
->buffer
;
2464 COMPILE_STACK_TOP
.regnum
= regnum
;
2466 /* We will eventually replace the 0 with the number of
2467 groups inner to this one. But do not push a
2468 start_memory for groups beyond the last one we can
2469 represent in the compiled pattern. */
2470 if (regnum
<= MAX_REGNUM
)
2472 COMPILE_STACK_TOP
.inner_group_offset
= b
- bufp
->buffer
+ 2;
2473 BUF_PUSH_3 (start_memory
, regnum
, 0);
2476 compile_stack
.avail
++;
2481 /* If we've reached MAX_REGNUM groups, then this open
2482 won't actually generate any code, so we'll have to
2483 clear pending_exact explicitly. */
2489 if (syntax
& RE_NO_BK_PARENS
) goto normal_backslash
;
2491 if (COMPILE_STACK_EMPTY
)
2492 if (syntax
& RE_UNMATCHED_RIGHT_PAREN_ORD
)
2493 goto normal_backslash
;
2495 FREE_STACK_RETURN (REG_ERPAREN
);
2499 { /* Push a dummy failure point at the end of the
2500 alternative for a possible future
2501 `pop_failure_jump' to pop. See comments at
2502 `push_dummy_failure' in `re_match_2'. */
2503 BUF_PUSH (push_dummy_failure
);
2505 /* We allocated space for this jump when we assigned
2506 to `fixup_alt_jump', in the `handle_alt' case below. */
2507 STORE_JUMP (jump_past_alt
, fixup_alt_jump
, b
- 1);
2510 /* See similar code for backslashed left paren above. */
2511 if (COMPILE_STACK_EMPTY
)
2512 if (syntax
& RE_UNMATCHED_RIGHT_PAREN_ORD
)
2515 FREE_STACK_RETURN (REG_ERPAREN
);
2517 /* Since we just checked for an empty stack above, this
2518 ``can't happen''. */
2519 assert (compile_stack
.avail
!= 0);
2521 /* We don't just want to restore into `regnum', because
2522 later groups should continue to be numbered higher,
2523 as in `(ab)c(de)' -- the second group is #2. */
2524 regnum_t this_group_regnum
;
2526 compile_stack
.avail
--;
2527 begalt
= bufp
->buffer
+ COMPILE_STACK_TOP
.begalt_offset
;
2529 = COMPILE_STACK_TOP
.fixup_alt_jump
2530 ? bufp
->buffer
+ COMPILE_STACK_TOP
.fixup_alt_jump
- 1
2532 laststart
= bufp
->buffer
+ COMPILE_STACK_TOP
.laststart_offset
;
2533 this_group_regnum
= COMPILE_STACK_TOP
.regnum
;
2534 /* If we've reached MAX_REGNUM groups, then this open
2535 won't actually generate any code, so we'll have to
2536 clear pending_exact explicitly. */
2539 /* We're at the end of the group, so now we know how many
2540 groups were inside this one. */
2541 if (this_group_regnum
<= MAX_REGNUM
)
2543 unsigned char *inner_group_loc
2544 = bufp
->buffer
+ COMPILE_STACK_TOP
.inner_group_offset
;
2546 *inner_group_loc
= regnum
- this_group_regnum
;
2547 BUF_PUSH_3 (stop_memory
, this_group_regnum
,
2548 regnum
- this_group_regnum
);
2554 case '|': /* `\|'. */
2555 if (syntax
& RE_LIMITED_OPS
|| syntax
& RE_NO_BK_VBAR
)
2556 goto normal_backslash
;
2558 if (syntax
& RE_LIMITED_OPS
)
2561 /* Insert before the previous alternative a jump which
2562 jumps to this alternative if the former fails. */
2563 GET_BUFFER_SPACE (3);
2564 INSERT_JUMP (on_failure_jump
, begalt
, b
+ 6);
2568 /* The alternative before this one has a jump after it
2569 which gets executed if it gets matched. Adjust that
2570 jump so it will jump to this alternative's analogous
2571 jump (put in below, which in turn will jump to the next
2572 (if any) alternative's such jump, etc.). The last such
2573 jump jumps to the correct final destination. A picture:
2579 If we are at `b', then fixup_alt_jump right now points to a
2580 three-byte space after `a'. We'll put in the jump, set
2581 fixup_alt_jump to right after `b', and leave behind three
2582 bytes which we'll fill in when we get to after `c'. */
2585 STORE_JUMP (jump_past_alt
, fixup_alt_jump
, b
);
2587 /* Mark and leave space for a jump after this alternative,
2588 to be filled in later either by next alternative or
2589 when know we're at the end of a series of alternatives. */
2591 GET_BUFFER_SPACE (3);
2600 /* If \{ is a literal. */
2601 if (!(syntax
& RE_INTERVALS
)
2602 /* If we're at `\{' and it's not the open-interval
2604 || ((syntax
& RE_INTERVALS
) && (syntax
& RE_NO_BK_BRACES
))
2605 || (p
- 2 == pattern
&& p
== pend
))
2606 goto normal_backslash
;
2610 /* If got here, then the syntax allows intervals. */
2612 /* At least (most) this many matches must be made. */
2613 int lower_bound
= -1, upper_bound
= -1;
2615 beg_interval
= p
- 1;
2619 if (syntax
& RE_NO_BK_BRACES
)
2620 goto unfetch_interval
;
2622 FREE_STACK_RETURN (REG_EBRACE
);
2625 GET_UNSIGNED_NUMBER (lower_bound
);
2629 GET_UNSIGNED_NUMBER (upper_bound
);
2630 if (upper_bound
< 0) upper_bound
= RE_DUP_MAX
;
2633 /* Interval such as `{1}' => match exactly once. */
2634 upper_bound
= lower_bound
;
2636 if (lower_bound
< 0 || upper_bound
> RE_DUP_MAX
2637 || lower_bound
> upper_bound
)
2639 if (syntax
& RE_NO_BK_BRACES
)
2640 goto unfetch_interval
;
2642 FREE_STACK_RETURN (REG_BADBR
);
2645 if (!(syntax
& RE_NO_BK_BRACES
))
2647 if (c
!= '\\') FREE_STACK_RETURN (REG_EBRACE
);
2654 if (syntax
& RE_NO_BK_BRACES
)
2655 goto unfetch_interval
;
2657 FREE_STACK_RETURN (REG_BADBR
);
2660 /* We just parsed a valid interval. */
2662 /* If it's invalid to have no preceding re. */
2665 if (syntax
& RE_CONTEXT_INVALID_OPS
)
2666 FREE_STACK_RETURN (REG_BADRPT
);
2667 else if (syntax
& RE_CONTEXT_INDEP_OPS
)
2670 goto unfetch_interval
;
2673 /* If the upper bound is zero, don't want to succeed at
2674 all; jump from `laststart' to `b + 3', which will be
2675 the end of the buffer after we insert the jump. */
2676 if (upper_bound
== 0)
2678 GET_BUFFER_SPACE (3);
2679 INSERT_JUMP (jump
, laststart
, b
+ 3);
2683 /* Otherwise, we have a nontrivial interval. When
2684 we're all done, the pattern will look like:
2685 set_number_at <jump count> <upper bound>
2686 set_number_at <succeed_n count> <lower bound>
2687 succeed_n <after jump addr> <succeed_n count>
2689 jump_n <succeed_n addr> <jump count>
2690 (The upper bound and `jump_n' are omitted if
2691 `upper_bound' is 1, though.) */
2693 { /* If the upper bound is > 1, we need to insert
2694 more at the end of the loop. */
2695 unsigned nbytes
= 10 + (upper_bound
> 1) * 10;
2697 GET_BUFFER_SPACE (nbytes
);
2699 /* Initialize lower bound of the `succeed_n', even
2700 though it will be set during matching by its
2701 attendant `set_number_at' (inserted next),
2702 because `re_compile_fastmap' needs to know.
2703 Jump to the `jump_n' we might insert below. */
2704 INSERT_JUMP2 (succeed_n
, laststart
,
2705 b
+ 5 + (upper_bound
> 1) * 5,
2709 /* Code to initialize the lower bound. Insert
2710 before the `succeed_n'. The `5' is the last two
2711 bytes of this `set_number_at', plus 3 bytes of
2712 the following `succeed_n'. */
2713 insert_op2 (set_number_at
, laststart
, 5, lower_bound
, b
);
2716 if (upper_bound
> 1)
2717 { /* More than one repetition is allowed, so
2718 append a backward jump to the `succeed_n'
2719 that starts this interval.
2721 When we've reached this during matching,
2722 we'll have matched the interval once, so
2723 jump back only `upper_bound - 1' times. */
2724 STORE_JUMP2 (jump_n
, b
, laststart
+ 5,
2728 /* The location we want to set is the second
2729 parameter of the `jump_n'; that is `b-2' as
2730 an absolute address. `laststart' will be
2731 the `set_number_at' we're about to insert;
2732 `laststart+3' the number to set, the source
2733 for the relative address. But we are
2734 inserting into the middle of the pattern --
2735 so everything is getting moved up by 5.
2736 Conclusion: (b - 2) - (laststart + 3) + 5,
2737 i.e., b - laststart.
2739 We insert this at the beginning of the loop
2740 so that if we fail during matching, we'll
2741 reinitialize the bounds. */
2742 insert_op2 (set_number_at
, laststart
, b
- laststart
,
2743 upper_bound
- 1, b
);
2748 beg_interval
= NULL
;
2753 /* If an invalid interval, match the characters as literals. */
2754 assert (beg_interval
);
2756 beg_interval
= NULL
;
2758 /* normal_char and normal_backslash need `c'. */
2761 if (!(syntax
& RE_NO_BK_BRACES
))
2763 if (p
> pattern
&& p
[-1] == '\\')
2764 goto normal_backslash
;
2769 /* There is no way to specify the before_dot and after_dot
2770 operators. rms says this is ok. --karl */
2778 BUF_PUSH_2 (syntaxspec
, syntax_spec_code
[c
]);
2784 BUF_PUSH_2 (notsyntaxspec
, syntax_spec_code
[c
]);
2790 BUF_PUSH_2 (categoryspec
, c
);
2796 BUF_PUSH_2 (notcategoryspec
, c
);
2803 BUF_PUSH (wordchar
);
2809 BUF_PUSH (notwordchar
);
2822 BUF_PUSH (wordbound
);
2826 BUF_PUSH (notwordbound
);
2837 case '1': case '2': case '3': case '4': case '5':
2838 case '6': case '7': case '8': case '9':
2839 if (syntax
& RE_NO_BK_REFS
)
2845 FREE_STACK_RETURN (REG_ESUBREG
);
2847 /* Can't back reference to a subexpression if inside of it. */
2848 if (group_in_compile_stack (compile_stack
, c1
))
2852 BUF_PUSH_2 (duplicate
, c1
);
2858 if (syntax
& RE_BK_PLUS_QM
)
2861 goto normal_backslash
;
2865 /* You might think it would be useful for \ to mean
2866 not to translate; but if we don't translate it
2867 it will never match anything. */
2875 /* Expects the character in `c'. */
2877 p1
= p
- 1; /* P1 points the head of C. */
2879 if (bufp
->multibyte
)
2880 /* Set P to the next character boundary. */
2881 p
+= MULTIBYTE_FORM_LENGTH (p1
, pend
- p1
) - 1;
2883 /* If no exactn currently being built. */
2886 /* If last exactn not at current position. */
2887 || pending_exact
+ *pending_exact
+ 1 != b
2889 /* We have only one byte following the exactn for the count. */
2890 || *pending_exact
>= (1 << BYTEWIDTH
) - (p
- p1
)
2892 /* If followed by a repetition operator. */
2893 || *p
== '*' || *p
== '^'
2894 || ((syntax
& RE_BK_PLUS_QM
)
2895 ? *p
== '\\' && (p
[1] == '+' || p
[1] == '?')
2896 : (*p
== '+' || *p
== '?'))
2897 || ((syntax
& RE_INTERVALS
)
2898 && ((syntax
& RE_NO_BK_BRACES
)
2900 : (p
[0] == '\\' && p
[1] == '{'))))
2902 /* Start building a new exactn. */
2906 BUF_PUSH_2 (exactn
, 0);
2907 pending_exact
= b
- 1;
2910 /* Here, C may translated, therefore C may not equal to *P1. */
2918 /* Rest of multibyte form should be copied literally. */
2919 c
= *(unsigned char *)p1
;
2923 } /* while p != pend */
2926 /* Through the pattern now. */
2929 STORE_JUMP (jump_past_alt
, fixup_alt_jump
, b
);
2931 if (!COMPILE_STACK_EMPTY
)
2932 FREE_STACK_RETURN (REG_EPAREN
);
2934 /* If we don't want backtracking, force success
2935 the first time we reach the end of the compiled pattern. */
2936 if (syntax
& RE_NO_POSIX_BACKTRACKING
)
2939 free (compile_stack
.stack
);
2941 /* We have succeeded; set the length of the buffer. */
2942 bufp
->used
= b
- bufp
->buffer
;
2947 DEBUG_PRINT1 ("\nCompiled pattern: \n");
2948 print_compiled_pattern (bufp
);
2952 #ifndef MATCH_MAY_ALLOCATE
2953 /* Initialize the failure stack to the largest possible stack. This
2954 isn't necessary unless we're trying to avoid calling alloca in
2955 the search and match routines. */
2957 int num_regs
= bufp
->re_nsub
+ 1;
2959 if (fail_stack
.size
< re_max_failures
* TYPICAL_FAILURE_SIZE
)
2961 fail_stack
.size
= re_max_failures
* TYPICAL_FAILURE_SIZE
);
2964 if (! fail_stack
.stack
)
2966 = (fail_stack_elt_t
*) xmalloc (fail_stack
.size
2967 * sizeof (fail_stack_elt_t
));
2970 = (fail_stack_elt_t
*) xrealloc (fail_stack
.stack
,
2972 * sizeof (fail_stack_elt_t
)));
2973 #else /* not emacs */
2974 if (! fail_stack
.stack
)
2976 = (fail_stack_elt_t
*) malloc (fail_stack
.size
2977 * sizeof (fail_stack_elt_t
));
2980 = (fail_stack_elt_t
*) realloc (fail_stack
.stack
,
2982 * sizeof (fail_stack_elt_t
)));
2983 #endif /* not emacs */
2986 regex_grow_registers (num_regs
);
2988 #endif /* not MATCH_MAY_ALLOCATE */
2991 } /* regex_compile */
2993 /* Subroutines for `regex_compile'. */
2995 /* Store OP at LOC followed by two-byte integer parameter ARG. */
2998 store_op1 (op
, loc
, arg
)
3003 *loc
= (unsigned char) op
;
3004 STORE_NUMBER (loc
+ 1, arg
);
3008 /* Like `store_op1', but for two two-byte parameters ARG1 and ARG2. */
3011 store_op2 (op
, loc
, arg1
, arg2
)
3016 *loc
= (unsigned char) op
;
3017 STORE_NUMBER (loc
+ 1, arg1
);
3018 STORE_NUMBER (loc
+ 3, arg2
);
3022 /* Copy the bytes from LOC to END to open up three bytes of space at LOC
3023 for OP followed by two-byte integer parameter ARG. */
3026 insert_op1 (op
, loc
, arg
, end
)
3032 register unsigned char *pfrom
= end
;
3033 register unsigned char *pto
= end
+ 3;
3035 while (pfrom
!= loc
)
3038 store_op1 (op
, loc
, arg
);
3042 /* Like `insert_op1', but for two two-byte parameters ARG1 and ARG2. */
3045 insert_op2 (op
, loc
, arg1
, arg2
, end
)
3051 register unsigned char *pfrom
= end
;
3052 register unsigned char *pto
= end
+ 5;
3054 while (pfrom
!= loc
)
3057 store_op2 (op
, loc
, arg1
, arg2
);
3061 /* P points to just after a ^ in PATTERN. Return true if that ^ comes
3062 after an alternative or a begin-subexpression. We assume there is at
3063 least one character before the ^. */
3066 at_begline_loc_p (pattern
, p
, syntax
)
3067 const char *pattern
, *p
;
3068 reg_syntax_t syntax
;
3070 const char *prev
= p
- 2;
3071 boolean prev_prev_backslash
= prev
> pattern
&& prev
[-1] == '\\';
3074 /* After a subexpression? */
3075 (*prev
== '(' && (syntax
& RE_NO_BK_PARENS
|| prev_prev_backslash
))
3076 /* After an alternative? */
3077 || (*prev
== '|' && (syntax
& RE_NO_BK_VBAR
|| prev_prev_backslash
));
3081 /* The dual of at_begline_loc_p. This one is for $. We assume there is
3082 at least one character after the $, i.e., `P < PEND'. */
3085 at_endline_loc_p (p
, pend
, syntax
)
3086 const char *p
, *pend
;
3089 const char *next
= p
;
3090 boolean next_backslash
= *next
== '\\';
3091 const char *next_next
= p
+ 1 < pend
? p
+ 1 : 0;
3094 /* Before a subexpression? */
3095 (syntax
& RE_NO_BK_PARENS
? *next
== ')'
3096 : next_backslash
&& next_next
&& *next_next
== ')')
3097 /* Before an alternative? */
3098 || (syntax
& RE_NO_BK_VBAR
? *next
== '|'
3099 : next_backslash
&& next_next
&& *next_next
== '|');
3103 /* Returns true if REGNUM is in one of COMPILE_STACK's elements and
3104 false if it's not. */
3107 group_in_compile_stack (compile_stack
, regnum
)
3108 compile_stack_type compile_stack
;
3113 for (this_element
= compile_stack
.avail
- 1;
3116 if (compile_stack
.stack
[this_element
].regnum
== regnum
)
3123 /* Read the ending character of a range (in a bracket expression) from the
3124 uncompiled pattern *P_PTR (which ends at PEND). We assume the
3125 starting character is in `P[-2]'. (`P[-1]' is the character `-'.)
3126 Then we set the translation of all bits between the starting and
3127 ending characters (inclusive) in the compiled pattern B.
3129 Return an error code.
3131 We use these short variable names so we can use the same macros as
3132 `regex_compile' itself. */
3134 static reg_errcode_t
3135 compile_range (p_ptr
, pend
, translate
, syntax
, b
)
3136 const char **p_ptr
, *pend
;
3137 RE_TRANSLATE_TYPE translate
;
3138 reg_syntax_t syntax
;
3143 const char *p
= *p_ptr
;
3144 int range_start
, range_end
;
3149 /* Even though the pattern is a signed `char *', we need to fetch
3150 with unsigned char *'s; if the high bit of the pattern character
3151 is set, the range endpoints will be negative if we fetch using a
3154 We also want to fetch the endpoints without translating them; the
3155 appropriate translation is done in the bit-setting loop below. */
3156 /* The SVR4 compiler on the 3B2 had trouble with unsigned const char *. */
3157 range_start
= ((const unsigned char *) p
)[-2];
3158 range_end
= ((const unsigned char *) p
)[0];
3160 /* Have to increment the pointer into the pattern string, so the
3161 caller isn't still at the ending character. */
3164 /* If the start is after the end, the range is empty. */
3165 if (range_start
> range_end
)
3166 return syntax
& RE_NO_EMPTY_RANGES
? REG_ERANGE
: REG_NOERROR
;
3168 /* Here we see why `this_char' has to be larger than an `unsigned
3169 char' -- the range is inclusive, so if `range_end' == 0xff
3170 (assuming 8-bit characters), we would otherwise go into an infinite
3171 loop, since all characters <= 0xff. */
3172 for (this_char
= range_start
; this_char
<= range_end
; this_char
++)
3174 SET_LIST_BIT (TRANSLATE (this_char
));
3180 /* re_compile_fastmap computes a ``fastmap'' for the compiled pattern in
3181 BUFP. A fastmap records which of the (1 << BYTEWIDTH) possible
3182 characters can start a string that matches the pattern. This fastmap
3183 is used by re_search to skip quickly over impossible starting points.
3185 The caller must supply the address of a (1 << BYTEWIDTH)-byte data
3186 area as BUFP->fastmap.
3188 We set the `fastmap', `fastmap_accurate', and `can_be_null' fields in
3191 Returns 0 if we succeed, -2 if an internal error. */
3194 re_compile_fastmap (bufp
)
3195 struct re_pattern_buffer
*bufp
;
3198 #ifdef MATCH_MAY_ALLOCATE
3199 fail_stack_type fail_stack
;
3201 #ifndef REGEX_MALLOC
3204 /* We don't push any register information onto the failure stack. */
3205 unsigned num_regs
= 0;
3207 register char *fastmap
= bufp
->fastmap
;
3208 unsigned char *pattern
= bufp
->buffer
;
3209 unsigned long size
= bufp
->used
;
3210 unsigned char *p
= pattern
;
3211 register unsigned char *pend
= pattern
+ size
;
3213 /* This holds the pointer to the failure stack, when
3214 it is allocated relocatably. */
3215 fail_stack_elt_t
*failure_stack_ptr
;
3217 /* Assume that each path through the pattern can be null until
3218 proven otherwise. We set this false at the bottom of switch
3219 statement, to which we get only if a particular path doesn't
3220 match the empty string. */
3221 boolean path_can_be_null
= true;
3223 /* We aren't doing a `succeed_n' to begin with. */
3224 boolean succeed_n_p
= false;
3226 /* If all elements for base leading-codes in fastmap is set, this
3227 flag is set true. */
3228 boolean match_any_multibyte_characters
= false;
3230 /* Maximum code of simple (single byte) character. */
3231 int simple_char_max
;
3233 assert (fastmap
!= NULL
&& p
!= NULL
);
3236 bzero (fastmap
, 1 << BYTEWIDTH
); /* Assume nothing's valid. */
3237 bufp
->fastmap_accurate
= 1; /* It will be when we're done. */
3238 bufp
->can_be_null
= 0;
3242 if (p
== pend
|| *p
== succeed
)
3244 /* We have reached the (effective) end of pattern. */
3245 if (!FAIL_STACK_EMPTY ())
3247 bufp
->can_be_null
|= path_can_be_null
;
3249 /* Reset for next path. */
3250 path_can_be_null
= true;
3252 p
= fail_stack
.stack
[--fail_stack
.avail
].pointer
;
3260 /* We should never be about to go beyond the end of the pattern. */
3263 switch (SWITCH_ENUM_CAST ((re_opcode_t
) *p
++))
3266 /* I guess the idea here is to simply not bother with a fastmap
3267 if a backreference is used, since it's too hard to figure out
3268 the fastmap for the corresponding group. Setting
3269 `can_be_null' stops `re_search_2' from using the fastmap, so
3270 that is all we do. */
3272 bufp
->can_be_null
= 1;
3276 /* Following are the cases which match a character. These end
3286 for (j
= *p
++ * BYTEWIDTH
- 1; j
>= 0; j
--)
3287 if (p
[j
/ BYTEWIDTH
] & (1 << (j
% BYTEWIDTH
)))
3293 /* Chars beyond end of map must be allowed. */
3294 for (j
= *p
* BYTEWIDTH
; j
< (1 << BYTEWIDTH
); j
++)
3297 for (j
= *p
++ * BYTEWIDTH
- 1; j
>= 0; j
--)
3298 if (!(p
[j
/ BYTEWIDTH
] & (1 << (j
% BYTEWIDTH
))))
3304 for (j
= 0; j
< (1 << BYTEWIDTH
); j
++)
3305 if (SYNTAX (j
) == Sword
)
3311 for (j
= 0; j
< (1 << BYTEWIDTH
); j
++)
3312 if (SYNTAX (j
) != Sword
)
3317 for (j
= CHARSET_BITMAP_SIZE (&p
[-1]) * BYTEWIDTH
- 1, p
++;
3319 if (p
[j
/ BYTEWIDTH
] & (1 << (j
% BYTEWIDTH
)))
3322 if (CHARSET_RANGE_TABLE_EXISTS_P (&p
[-2])
3323 && match_any_multibyte_characters
== false)
3325 /* Set fastmap[I] 1 where I is a base leading code of each
3326 multibyte character in the range table. */
3329 /* Make P points the range table. */
3330 p
+= CHARSET_BITMAP_SIZE (&p
[-2]);
3332 /* Extract the number of ranges in range table into
3334 EXTRACT_NUMBER_AND_INCR (count
, p
);
3335 for (; count
> 0; count
--, p
+= 2 * 3) /* XXX */
3337 /* Extract the start of each range. */
3338 EXTRACT_CHARACTER (c
, p
);
3339 j
= CHAR_CHARSET (c
);
3340 fastmap
[CHARSET_LEADING_CODE_BASE (j
)] = 1;
3347 /* Chars beyond end of map must be allowed. End of map is
3348 `127' if bufp->multibyte is nonzero. */
3349 simple_char_max
= bufp
->multibyte
? 0x80 : (1 << BYTEWIDTH
);
3350 for (j
= CHARSET_BITMAP_SIZE (&p
[-1]) * BYTEWIDTH
;
3351 j
< simple_char_max
; j
++)
3354 for (j
= CHARSET_BITMAP_SIZE (&p
[-1]) * BYTEWIDTH
- 1, p
++;
3356 if (!(p
[j
/ BYTEWIDTH
] & (1 << (j
% BYTEWIDTH
))))
3359 if (bufp
->multibyte
)
3360 /* Any character set can possibly contain a character
3361 which doesn't match the specified set of characters. */
3363 set_fastmap_for_multibyte_characters
:
3364 if (match_any_multibyte_characters
== false)
3366 for (j
= 0x80; j
< 0xA0; j
++) /* XXX */
3367 if (BASE_LEADING_CODE_P (j
))
3369 match_any_multibyte_characters
= true;
3376 simple_char_max
= bufp
->multibyte
? 0x80 : (1 << BYTEWIDTH
);
3377 for (j
= 0; j
< simple_char_max
; j
++)
3378 if (SYNTAX (j
) == Sword
)
3381 if (bufp
->multibyte
)
3382 /* Any character set can possibly contain a character
3383 whose syntax is `Sword'. */
3384 goto set_fastmap_for_multibyte_characters
;
3389 simple_char_max
= bufp
->multibyte
? 0x80 : (1 << BYTEWIDTH
);
3390 for (j
= 0; j
< simple_char_max
; j
++)
3391 if (SYNTAX (j
) != Sword
)
3394 if (bufp
->multibyte
)
3395 /* Any character set can possibly contain a character
3396 whose syntax is not `Sword'. */
3397 goto set_fastmap_for_multibyte_characters
;
3403 int fastmap_newline
= fastmap
['\n'];
3405 /* `.' matches anything (but if bufp->multibyte is
3406 nonzero, matches `\000' .. `\127' and possible multibyte
3408 if (bufp
->multibyte
)
3410 simple_char_max
= 0x80;
3412 for (j
= 0x80; j
< 0xA0; j
++)
3413 if (BASE_LEADING_CODE_P (j
))
3415 match_any_multibyte_characters
= true;
3418 simple_char_max
= (1 << BYTEWIDTH
);
3420 for (j
= 0; j
< simple_char_max
; j
++)
3423 /* ... except perhaps newline. */
3424 if (!(bufp
->syntax
& RE_DOT_NEWLINE
))
3425 fastmap
['\n'] = fastmap_newline
;
3427 /* Return if we have already set `can_be_null'; if we have,
3428 then the fastmap is irrelevant. Something's wrong here. */
3429 else if (bufp
->can_be_null
)
3432 /* Otherwise, have to check alternative paths. */
3443 /* This match depends on text properties. These end with
3444 aborting optimizations. */
3445 bufp
->can_be_null
= 1;
3449 simple_char_max
= bufp
->multibyte
? 0x80 : (1 << BYTEWIDTH
);
3450 for (j
= 0; j
< simple_char_max
; j
++)
3451 if (SYNTAX (j
) == (enum syntaxcode
) k
)
3454 if (bufp
->multibyte
)
3455 /* Any character set can possibly contain a character
3456 whose syntax is K. */
3457 goto set_fastmap_for_multibyte_characters
;
3462 simple_char_max
= bufp
->multibyte
? 0x80 : (1 << BYTEWIDTH
);
3463 for (j
= 0; j
< simple_char_max
; j
++)
3464 if (SYNTAX (j
) != (enum syntaxcode
) k
)
3467 if (bufp
->multibyte
)
3468 /* Any character set can possibly contain a character
3469 whose syntax is not K. */
3470 goto set_fastmap_for_multibyte_characters
;
3477 simple_char_max
= bufp
->multibyte
? 0x80 : (1 << BYTEWIDTH
);
3478 for (j
= 0; j
< simple_char_max
; j
++)
3479 if (CHAR_HAS_CATEGORY (j
, k
))
3482 if (bufp
->multibyte
)
3483 /* Any character set can possibly contain a character
3484 whose category is K. */
3485 goto set_fastmap_for_multibyte_characters
;
3489 case notcategoryspec
:
3491 simple_char_max
= bufp
->multibyte
? 0x80 : (1 << BYTEWIDTH
);
3492 for (j
= 0; j
< simple_char_max
; j
++)
3493 if (!CHAR_HAS_CATEGORY (j
, k
))
3496 if (bufp
->multibyte
)
3497 /* Any character set can possibly contain a character
3498 whose category is not K. */
3499 goto set_fastmap_for_multibyte_characters
;
3502 /* All cases after this match the empty string. These end with
3524 case push_dummy_failure
:
3529 case pop_failure_jump
:
3530 case maybe_pop_jump
:
3533 case dummy_failure_jump
:
3534 EXTRACT_NUMBER_AND_INCR (j
, p
);
3539 /* Jump backward implies we just went through the body of a
3540 loop and matched nothing. Opcode jumped to should be
3541 `on_failure_jump' or `succeed_n'. Just treat it like an
3542 ordinary jump. For a * loop, it has pushed its failure
3543 point already; if so, discard that as redundant. */
3544 if ((re_opcode_t
) *p
!= on_failure_jump
3545 && (re_opcode_t
) *p
!= succeed_n
)
3549 EXTRACT_NUMBER_AND_INCR (j
, p
);
3552 /* If what's on the stack is where we are now, pop it. */
3553 if (!FAIL_STACK_EMPTY ()
3554 && fail_stack
.stack
[fail_stack
.avail
- 1].pointer
== p
)
3560 case on_failure_jump
:
3561 case on_failure_keep_string_jump
:
3562 handle_on_failure_jump
:
3563 EXTRACT_NUMBER_AND_INCR (j
, p
);
3565 /* For some patterns, e.g., `(a?)?', `p+j' here points to the
3566 end of the pattern. We don't want to push such a point,
3567 since when we restore it above, entering the switch will
3568 increment `p' past the end of the pattern. We don't need
3569 to push such a point since we obviously won't find any more
3570 fastmap entries beyond `pend'. Such a pattern can match
3571 the null string, though. */
3574 if (!PUSH_PATTERN_OP (p
+ j
, fail_stack
))
3576 RESET_FAIL_STACK ();
3581 bufp
->can_be_null
= 1;
3585 EXTRACT_NUMBER_AND_INCR (k
, p
); /* Skip the n. */
3586 succeed_n_p
= false;
3593 /* Get to the number of times to succeed. */
3596 /* Increment p past the n for when k != 0. */
3597 EXTRACT_NUMBER_AND_INCR (k
, p
);
3601 succeed_n_p
= true; /* Spaghetti code alert. */
3602 goto handle_on_failure_jump
;
3619 abort (); /* We have listed all the cases. */
3622 /* Getting here means we have found the possible starting
3623 characters for one path of the pattern -- and that the empty
3624 string does not match. We need not follow this path further.
3625 Instead, look at the next alternative (remembered on the
3626 stack), or quit if no more. The test at the top of the loop
3627 does these things. */
3628 path_can_be_null
= false;
3632 /* Set `can_be_null' for the last path (also the first path, if the
3633 pattern is empty). */
3634 bufp
->can_be_null
|= path_can_be_null
;
3637 RESET_FAIL_STACK ();
3639 } /* re_compile_fastmap */
3641 /* Set REGS to hold NUM_REGS registers, storing them in STARTS and
3642 ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use
3643 this memory for recording register information. STARTS and ENDS
3644 must be allocated using the malloc library routine, and must each
3645 be at least NUM_REGS * sizeof (regoff_t) bytes long.
3647 If NUM_REGS == 0, then subsequent matches should allocate their own
3650 Unless this function is called, the first search or match using
3651 PATTERN_BUFFER will allocate its own register data, without
3652 freeing the old data. */
3655 re_set_registers (bufp
, regs
, num_regs
, starts
, ends
)
3656 struct re_pattern_buffer
*bufp
;
3657 struct re_registers
*regs
;
3659 regoff_t
*starts
, *ends
;
3663 bufp
->regs_allocated
= REGS_REALLOCATE
;
3664 regs
->num_regs
= num_regs
;
3665 regs
->start
= starts
;
3670 bufp
->regs_allocated
= REGS_UNALLOCATED
;
3672 regs
->start
= regs
->end
= (regoff_t
*) 0;
3676 /* Searching routines. */
3678 /* Like re_search_2, below, but only one string is specified, and
3679 doesn't let you say where to stop matching. */
3682 re_search (bufp
, string
, size
, startpos
, range
, regs
)
3683 struct re_pattern_buffer
*bufp
;
3685 int size
, startpos
, range
;
3686 struct re_registers
*regs
;
3688 return re_search_2 (bufp
, NULL
, 0, string
, size
, startpos
, range
,
3692 /* End address of virtual concatenation of string. */
3693 #define STOP_ADDR_VSTRING(P) \
3694 (((P) >= size1 ? string2 + size2 : string1 + size1))
3696 /* Address of POS in the concatenation of virtual string. */
3697 #define POS_ADDR_VSTRING(POS) \
3698 (((POS) >= size1 ? string2 - size1 : string1) + (POS))
3700 /* Using the compiled pattern in BUFP->buffer, first tries to match the
3701 virtual concatenation of STRING1 and STRING2, starting first at index
3702 STARTPOS, then at STARTPOS + 1, and so on.
3704 STRING1 and STRING2 have length SIZE1 and SIZE2, respectively.
3706 RANGE is how far to scan while trying to match. RANGE = 0 means try
3707 only at STARTPOS; in general, the last start tried is STARTPOS +
3710 In REGS, return the indices of the virtual concatenation of STRING1
3711 and STRING2 that matched the entire BUFP->buffer and its contained
3714 Do not consider matching one past the index STOP in the virtual
3715 concatenation of STRING1 and STRING2.
3717 We return either the position in the strings at which the match was
3718 found, -1 if no match, or -2 if error (such as failure
3722 re_search_2 (bufp
, string1
, size1
, string2
, size2
, startpos
, range
, regs
, stop
)
3723 struct re_pattern_buffer
*bufp
;
3724 const char *string1
, *string2
;
3728 struct re_registers
*regs
;
3732 register char *fastmap
= bufp
->fastmap
;
3733 register RE_TRANSLATE_TYPE translate
= bufp
->translate
;
3734 int total_size
= size1
+ size2
;
3735 int endpos
= startpos
+ range
;
3736 int anchored_start
= 0;
3738 /* Nonzero if we have to concern multibyte character. */
3739 int multibyte
= bufp
->multibyte
;
3741 /* Check for out-of-range STARTPOS. */
3742 if (startpos
< 0 || startpos
> total_size
)
3745 /* Fix up RANGE if it might eventually take us outside
3746 the virtual concatenation of STRING1 and STRING2.
3747 Make sure we won't move STARTPOS below 0 or above TOTAL_SIZE. */
3749 range
= 0 - startpos
;
3750 else if (endpos
> total_size
)
3751 range
= total_size
- startpos
;
3753 /* If the search isn't to be a backwards one, don't waste time in a
3754 search for a pattern that must be anchored. */
3755 if (bufp
->used
> 0 && (re_opcode_t
) bufp
->buffer
[0] == begbuf
&& range
> 0)
3764 /* In a forward search for something that starts with \=.
3765 don't keep searching past point. */
3766 if (bufp
->used
> 0 && (re_opcode_t
) bufp
->buffer
[0] == at_dot
&& range
> 0)
3768 range
= PT
- startpos
;
3774 /* Update the fastmap now if not correct already. */
3775 if (fastmap
&& !bufp
->fastmap_accurate
)
3776 if (re_compile_fastmap (bufp
) == -2)
3779 /* See whether the pattern is anchored. */
3780 if (bufp
->buffer
[0] == begline
)
3784 SETUP_SYNTAX_TABLE_FOR_OBJECT (re_match_object
,
3785 POS_AS_IN_BUFFER (startpos
> 0
3786 ? startpos
- 1 : startpos
),
3790 /* Loop through the string, looking for a place to start matching. */
3793 /* If the pattern is anchored,
3794 skip quickly past places we cannot match.
3795 We don't bother to treat startpos == 0 specially
3796 because that case doesn't repeat. */
3797 if (anchored_start
&& startpos
> 0)
3799 if (! (bufp
->newline_anchor
3800 && ((startpos
<= size1
? string1
[startpos
- 1]
3801 : string2
[startpos
- size1
- 1])
3806 /* If a fastmap is supplied, skip quickly over characters that
3807 cannot be the start of a match. If the pattern can match the
3808 null string, however, we don't need to skip characters; we want
3809 the first null string. */
3810 if (fastmap
&& startpos
< total_size
&& !bufp
->can_be_null
)
3812 if (range
> 0) /* Searching forwards. */
3814 register const char *d
;
3815 register int lim
= 0;
3818 if (startpos
< size1
&& startpos
+ range
>= size1
)
3819 lim
= range
- (size1
- startpos
);
3821 d
= POS_ADDR_VSTRING (startpos
);
3823 /* Written out as an if-else to avoid testing `translate'
3827 && !fastmap
[(unsigned char)
3828 RE_TRANSLATE (translate
, (unsigned char) *d
++)])
3831 while (range
> lim
&& !fastmap
[(unsigned char) *d
++])
3834 startpos
+= irange
- range
;
3836 else /* Searching backwards. */
3838 register char c
= (size1
== 0 || startpos
>= size1
3839 ? string2
[startpos
- size1
]
3840 : string1
[startpos
]);
3842 if (!fastmap
[(unsigned char) TRANSLATE (c
)])
3847 /* If can't match the null string, and that's all we have left, fail. */
3848 if (range
>= 0 && startpos
== total_size
&& fastmap
3849 && !bufp
->can_be_null
)
3852 val
= re_match_2_internal (bufp
, string1
, size1
, string2
, size2
,
3853 startpos
, regs
, stop
);
3854 #ifndef REGEX_MALLOC
3871 /* Update STARTPOS to the next character boundary. */
3874 const unsigned char *p
3875 = (const unsigned char *) POS_ADDR_VSTRING (startpos
);
3876 const unsigned char *pend
3877 = (const unsigned char *) STOP_ADDR_VSTRING (startpos
);
3878 int len
= MULTIBYTE_FORM_LENGTH (p
, pend
- p
);
3896 /* Update STARTPOS to the previous character boundary. */
3899 const unsigned char *p
3900 = (const unsigned char *) POS_ADDR_VSTRING (startpos
);
3903 /* Find the head of multibyte form. */
3904 while (!CHAR_HEAD_P (p
))
3909 if (MULTIBYTE_FORM_LENGTH (p
, len
+ 1) != (len
+ 1))
3926 /* Declarations and macros for re_match_2. */
3928 static int bcmp_translate ();
3929 static boolean
alt_match_null_string_p (),
3930 common_op_match_null_string_p (),
3931 group_match_null_string_p ();
3933 /* This converts PTR, a pointer into one of the search strings `string1'
3934 and `string2' into an offset from the beginning of that string. */
3935 #define POINTER_TO_OFFSET(ptr) \
3936 (FIRST_STRING_P (ptr) \
3937 ? ((regoff_t) ((ptr) - string1)) \
3938 : ((regoff_t) ((ptr) - string2 + size1)))
3940 /* Macros for dealing with the split strings in re_match_2. */
3942 #define MATCHING_IN_FIRST_STRING (dend == end_match_1)
3944 /* Call before fetching a character with *d. This switches over to
3945 string2 if necessary. */
3946 #define PREFETCH() \
3949 /* End of string2 => fail. */ \
3950 if (dend == end_match_2) \
3952 /* End of string1 => advance to string2. */ \
3954 dend = end_match_2; \
3958 /* Test if at very beginning or at very end of the virtual concatenation
3959 of `string1' and `string2'. If only one string, it's `string2'. */
3960 #define AT_STRINGS_BEG(d) ((d) == (size1 ? string1 : string2) || !size2)
3961 #define AT_STRINGS_END(d) ((d) == end2)
3964 /* Test if D points to a character which is word-constituent. We have
3965 two special cases to check for: if past the end of string1, look at
3966 the first character in string2; and if before the beginning of
3967 string2, look at the last character in string1. */
3968 #define WORDCHAR_P(d) \
3969 (SYNTAX ((d) == end1 ? *string2 \
3970 : (d) == string2 - 1 ? *(end1 - 1) : *(d)) \
3973 /* Disabled due to a compiler bug -- see comment at case wordbound */
3975 /* The comment at case wordbound is following one, but we don't use
3976 AT_WORD_BOUNDARY anymore to support multibyte form.
3978 The DEC Alpha C compiler 3.x generates incorrect code for the
3979 test WORDCHAR_P (d - 1) != WORDCHAR_P (d) in the expansion of
3980 AT_WORD_BOUNDARY, so this code is disabled. Expanding the
3981 macro and introducing temporary variables works around the bug. */
3984 /* Test if the character before D and the one at D differ with respect
3985 to being word-constituent. */
3986 #define AT_WORD_BOUNDARY(d) \
3987 (AT_STRINGS_BEG (d) || AT_STRINGS_END (d) \
3988 || WORDCHAR_P (d - 1) != WORDCHAR_P (d))
3991 /* Free everything we malloc. */
3992 #ifdef MATCH_MAY_ALLOCATE
3993 #define FREE_VAR(var) if (var) { REGEX_FREE (var); var = NULL; } else
3994 #define FREE_VARIABLES() \
3996 REGEX_FREE_STACK (fail_stack.stack); \
3997 FREE_VAR (regstart); \
3998 FREE_VAR (regend); \
3999 FREE_VAR (old_regstart); \
4000 FREE_VAR (old_regend); \
4001 FREE_VAR (best_regstart); \
4002 FREE_VAR (best_regend); \
4003 FREE_VAR (reg_info); \
4004 FREE_VAR (reg_dummy); \
4005 FREE_VAR (reg_info_dummy); \
4008 #define FREE_VARIABLES() ((void)0) /* Do nothing! But inhibit gcc warning. */
4009 #endif /* not MATCH_MAY_ALLOCATE */
4011 /* These values must meet several constraints. They must not be valid
4012 register values; since we have a limit of 255 registers (because
4013 we use only one byte in the pattern for the register number), we can
4014 use numbers larger than 255. They must differ by 1, because of
4015 NUM_FAILURE_ITEMS above. And the value for the lowest register must
4016 be larger than the value for the highest register, so we do not try
4017 to actually save any registers when none are active. */
4018 #define NO_HIGHEST_ACTIVE_REG (1 << BYTEWIDTH)
4019 #define NO_LOWEST_ACTIVE_REG (NO_HIGHEST_ACTIVE_REG + 1)
4021 /* Matching routines. */
4023 #ifndef emacs /* Emacs never uses this. */
4024 /* re_match is like re_match_2 except it takes only a single string. */
4027 re_match (bufp
, string
, size
, pos
, regs
)
4028 struct re_pattern_buffer
*bufp
;
4031 struct re_registers
*regs
;
4033 int result
= re_match_2_internal (bufp
, NULL
, 0, string
, size
,
4038 #endif /* not emacs */
4041 /* In Emacs, this is the string or buffer in which we
4042 are matching. It is used for looking up syntax properties. */
4043 Lisp_Object re_match_object
;
4046 /* re_match_2 matches the compiled pattern in BUFP against the
4047 the (virtual) concatenation of STRING1 and STRING2 (of length SIZE1
4048 and SIZE2, respectively). We start matching at POS, and stop
4051 If REGS is non-null and the `no_sub' field of BUFP is nonzero, we
4052 store offsets for the substring each group matched in REGS. See the
4053 documentation for exactly how many groups we fill.
4055 We return -1 if no match, -2 if an internal error (such as the
4056 failure stack overflowing). Otherwise, we return the length of the
4057 matched substring. */
4060 re_match_2 (bufp
, string1
, size1
, string2
, size2
, pos
, regs
, stop
)
4061 struct re_pattern_buffer
*bufp
;
4062 const char *string1
, *string2
;
4065 struct re_registers
*regs
;
4071 SETUP_SYNTAX_TABLE_FOR_OBJECT (re_match_object
,
4072 POS_AS_IN_BUFFER (pos
> 0 ? pos
- 1 : pos
),
4076 result
= re_match_2_internal (bufp
, string1
, size1
, string2
, size2
,
4082 /* This is a separate function so that we can force an alloca cleanup
4085 re_match_2_internal (bufp
, string1
, size1
, string2
, size2
, pos
, regs
, stop
)
4086 struct re_pattern_buffer
*bufp
;
4087 const char *string1
, *string2
;
4090 struct re_registers
*regs
;
4093 /* General temporaries. */
4097 /* Just past the end of the corresponding string. */
4098 const char *end1
, *end2
;
4100 /* Pointers into string1 and string2, just past the last characters in
4101 each to consider matching. */
4102 const char *end_match_1
, *end_match_2
;
4104 /* Where we are in the data, and the end of the current string. */
4105 const char *d
, *dend
;
4107 /* Where we are in the pattern, and the end of the pattern. */
4108 unsigned char *p
= bufp
->buffer
;
4109 register unsigned char *pend
= p
+ bufp
->used
;
4111 /* Mark the opcode just after a start_memory, so we can test for an
4112 empty subpattern when we get to the stop_memory. */
4113 unsigned char *just_past_start_mem
= 0;
4115 /* We use this to map every character in the string. */
4116 RE_TRANSLATE_TYPE translate
= bufp
->translate
;
4118 /* Nonzero if we have to concern multibyte character. */
4119 int multibyte
= bufp
->multibyte
;
4121 /* Failure point stack. Each place that can handle a failure further
4122 down the line pushes a failure point on this stack. It consists of
4123 restart, regend, and reg_info for all registers corresponding to
4124 the subexpressions we're currently inside, plus the number of such
4125 registers, and, finally, two char *'s. The first char * is where
4126 to resume scanning the pattern; the second one is where to resume
4127 scanning the strings. If the latter is zero, the failure point is
4128 a ``dummy''; if a failure happens and the failure point is a dummy,
4129 it gets discarded and the next next one is tried. */
4130 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
4131 fail_stack_type fail_stack
;
4134 static unsigned failure_id
= 0;
4135 unsigned nfailure_points_pushed
= 0, nfailure_points_popped
= 0;
4138 /* This holds the pointer to the failure stack, when
4139 it is allocated relocatably. */
4140 fail_stack_elt_t
*failure_stack_ptr
;
4142 /* We fill all the registers internally, independent of what we
4143 return, for use in backreferences. The number here includes
4144 an element for register zero. */
4145 unsigned num_regs
= bufp
->re_nsub
+ 1;
4147 /* The currently active registers. */
4148 unsigned lowest_active_reg
= NO_LOWEST_ACTIVE_REG
;
4149 unsigned highest_active_reg
= NO_HIGHEST_ACTIVE_REG
;
4151 /* Information on the contents of registers. These are pointers into
4152 the input strings; they record just what was matched (on this
4153 attempt) by a subexpression part of the pattern, that is, the
4154 regnum-th regstart pointer points to where in the pattern we began
4155 matching and the regnum-th regend points to right after where we
4156 stopped matching the regnum-th subexpression. (The zeroth register
4157 keeps track of what the whole pattern matches.) */
4158 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
4159 const char **regstart
, **regend
;
4162 /* If a group that's operated upon by a repetition operator fails to
4163 match anything, then the register for its start will need to be
4164 restored because it will have been set to wherever in the string we
4165 are when we last see its open-group operator. Similarly for a
4167 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
4168 const char **old_regstart
, **old_regend
;
4171 /* The is_active field of reg_info helps us keep track of which (possibly
4172 nested) subexpressions we are currently in. The matched_something
4173 field of reg_info[reg_num] helps us tell whether or not we have
4174 matched any of the pattern so far this time through the reg_num-th
4175 subexpression. These two fields get reset each time through any
4176 loop their register is in. */
4177 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
4178 register_info_type
*reg_info
;
4181 /* The following record the register info as found in the above
4182 variables when we find a match better than any we've seen before.
4183 This happens as we backtrack through the failure points, which in
4184 turn happens only if we have not yet matched the entire string. */
4185 unsigned best_regs_set
= false;
4186 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
4187 const char **best_regstart
, **best_regend
;
4190 /* Logically, this is `best_regend[0]'. But we don't want to have to
4191 allocate space for that if we're not allocating space for anything
4192 else (see below). Also, we never need info about register 0 for
4193 any of the other register vectors, and it seems rather a kludge to
4194 treat `best_regend' differently than the rest. So we keep track of
4195 the end of the best match so far in a separate variable. We
4196 initialize this to NULL so that when we backtrack the first time
4197 and need to test it, it's not garbage. */
4198 const char *match_end
= NULL
;
4200 /* This helps SET_REGS_MATCHED avoid doing redundant work. */
4201 int set_regs_matched_done
= 0;
4203 /* Used when we pop values we don't care about. */
4204 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
4205 const char **reg_dummy
;
4206 register_info_type
*reg_info_dummy
;
4210 /* Counts the total number of registers pushed. */
4211 unsigned num_regs_pushed
= 0;
4214 DEBUG_PRINT1 ("\n\nEntering re_match_2.\n");
4218 #ifdef MATCH_MAY_ALLOCATE
4219 /* Do not bother to initialize all the register variables if there are
4220 no groups in the pattern, as it takes a fair amount of time. If
4221 there are groups, we include space for register 0 (the whole
4222 pattern), even though we never use it, since it simplifies the
4223 array indexing. We should fix this. */
4226 regstart
= REGEX_TALLOC (num_regs
, const char *);
4227 regend
= REGEX_TALLOC (num_regs
, const char *);
4228 old_regstart
= REGEX_TALLOC (num_regs
, const char *);
4229 old_regend
= REGEX_TALLOC (num_regs
, const char *);
4230 best_regstart
= REGEX_TALLOC (num_regs
, const char *);
4231 best_regend
= REGEX_TALLOC (num_regs
, const char *);
4232 reg_info
= REGEX_TALLOC (num_regs
, register_info_type
);
4233 reg_dummy
= REGEX_TALLOC (num_regs
, const char *);
4234 reg_info_dummy
= REGEX_TALLOC (num_regs
, register_info_type
);
4236 if (!(regstart
&& regend
&& old_regstart
&& old_regend
&& reg_info
4237 && best_regstart
&& best_regend
&& reg_dummy
&& reg_info_dummy
))
4245 /* We must initialize all our variables to NULL, so that
4246 `FREE_VARIABLES' doesn't try to free them. */
4247 regstart
= regend
= old_regstart
= old_regend
= best_regstart
4248 = best_regend
= reg_dummy
= NULL
;
4249 reg_info
= reg_info_dummy
= (register_info_type
*) NULL
;
4251 #endif /* MATCH_MAY_ALLOCATE */
4253 /* The starting position is bogus. */
4254 if (pos
< 0 || pos
> size1
+ size2
)
4260 /* Initialize subexpression text positions to -1 to mark ones that no
4261 start_memory/stop_memory has been seen for. Also initialize the
4262 register information struct. */
4263 for (mcnt
= 1; mcnt
< num_regs
; mcnt
++)
4265 regstart
[mcnt
] = regend
[mcnt
]
4266 = old_regstart
[mcnt
] = old_regend
[mcnt
] = REG_UNSET_VALUE
;
4268 REG_MATCH_NULL_STRING_P (reg_info
[mcnt
]) = MATCH_NULL_UNSET_VALUE
;
4269 IS_ACTIVE (reg_info
[mcnt
]) = 0;
4270 MATCHED_SOMETHING (reg_info
[mcnt
]) = 0;
4271 EVER_MATCHED_SOMETHING (reg_info
[mcnt
]) = 0;
4274 /* We move `string1' into `string2' if the latter's empty -- but not if
4275 `string1' is null. */
4276 if (size2
== 0 && string1
!= NULL
)
4283 end1
= string1
+ size1
;
4284 end2
= string2
+ size2
;
4286 /* Compute where to stop matching, within the two strings. */
4289 end_match_1
= string1
+ stop
;
4290 end_match_2
= string2
;
4295 end_match_2
= string2
+ stop
- size1
;
4298 /* `p' scans through the pattern as `d' scans through the data.
4299 `dend' is the end of the input string that `d' points within. `d'
4300 is advanced into the following input string whenever necessary, but
4301 this happens before fetching; therefore, at the beginning of the
4302 loop, `d' can be pointing at the end of a string, but it cannot
4304 if (size1
> 0 && pos
<= size1
)
4311 d
= string2
+ pos
- size1
;
4315 DEBUG_PRINT1 ("The compiled pattern is: ");
4316 DEBUG_PRINT_COMPILED_PATTERN (bufp
, p
, pend
);
4317 DEBUG_PRINT1 ("The string to match is: `");
4318 DEBUG_PRINT_DOUBLE_STRING (d
, string1
, size1
, string2
, size2
);
4319 DEBUG_PRINT1 ("'\n");
4321 /* This loops over pattern commands. It exits by returning from the
4322 function if the match is complete, or it drops through if the match
4323 fails at this starting point in the input data. */
4326 DEBUG_PRINT2 ("\n0x%x: ", p
);
4329 { /* End of pattern means we might have succeeded. */
4330 DEBUG_PRINT1 ("end of pattern ... ");
4332 /* If we haven't matched the entire string, and we want the
4333 longest match, try backtracking. */
4334 if (d
!= end_match_2
)
4336 /* 1 if this match ends in the same string (string1 or string2)
4337 as the best previous match. */
4338 boolean same_str_p
= (FIRST_STRING_P (match_end
)
4339 == MATCHING_IN_FIRST_STRING
);
4340 /* 1 if this match is the best seen so far. */
4341 boolean best_match_p
;
4343 /* AIX compiler got confused when this was combined
4344 with the previous declaration. */
4346 best_match_p
= d
> match_end
;
4348 best_match_p
= !MATCHING_IN_FIRST_STRING
;
4350 DEBUG_PRINT1 ("backtracking.\n");
4352 if (!FAIL_STACK_EMPTY ())
4353 { /* More failure points to try. */
4355 /* If exceeds best match so far, save it. */
4356 if (!best_regs_set
|| best_match_p
)
4358 best_regs_set
= true;
4361 DEBUG_PRINT1 ("\nSAVING match as best so far.\n");
4363 for (mcnt
= 1; mcnt
< num_regs
; mcnt
++)
4365 best_regstart
[mcnt
] = regstart
[mcnt
];
4366 best_regend
[mcnt
] = regend
[mcnt
];
4372 /* If no failure points, don't restore garbage. And if
4373 last match is real best match, don't restore second
4375 else if (best_regs_set
&& !best_match_p
)
4378 /* Restore best match. It may happen that `dend ==
4379 end_match_1' while the restored d is in string2.
4380 For example, the pattern `x.*y.*z' against the
4381 strings `x-' and `y-z-', if the two strings are
4382 not consecutive in memory. */
4383 DEBUG_PRINT1 ("Restoring best registers.\n");
4386 dend
= ((d
>= string1
&& d
<= end1
)
4387 ? end_match_1
: end_match_2
);
4389 for (mcnt
= 1; mcnt
< num_regs
; mcnt
++)
4391 regstart
[mcnt
] = best_regstart
[mcnt
];
4392 regend
[mcnt
] = best_regend
[mcnt
];
4395 } /* d != end_match_2 */
4398 DEBUG_PRINT1 ("Accepting match.\n");
4400 /* If caller wants register contents data back, do it. */
4401 if (regs
&& !bufp
->no_sub
)
4403 /* Have the register data arrays been allocated? */
4404 if (bufp
->regs_allocated
== REGS_UNALLOCATED
)
4405 { /* No. So allocate them with malloc. We need one
4406 extra element beyond `num_regs' for the `-1' marker
4408 regs
->num_regs
= MAX (RE_NREGS
, num_regs
+ 1);
4409 regs
->start
= TALLOC (regs
->num_regs
, regoff_t
);
4410 regs
->end
= TALLOC (regs
->num_regs
, regoff_t
);
4411 if (regs
->start
== NULL
|| regs
->end
== NULL
)
4416 bufp
->regs_allocated
= REGS_REALLOCATE
;
4418 else if (bufp
->regs_allocated
== REGS_REALLOCATE
)
4419 { /* Yes. If we need more elements than were already
4420 allocated, reallocate them. If we need fewer, just
4422 if (regs
->num_regs
< num_regs
+ 1)
4424 regs
->num_regs
= num_regs
+ 1;
4425 RETALLOC (regs
->start
, regs
->num_regs
, regoff_t
);
4426 RETALLOC (regs
->end
, regs
->num_regs
, regoff_t
);
4427 if (regs
->start
== NULL
|| regs
->end
== NULL
)
4436 /* These braces fend off a "empty body in an else-statement"
4437 warning under GCC when assert expands to nothing. */
4438 assert (bufp
->regs_allocated
== REGS_FIXED
);
4441 /* Convert the pointer data in `regstart' and `regend' to
4442 indices. Register zero has to be set differently,
4443 since we haven't kept track of any info for it. */
4444 if (regs
->num_regs
> 0)
4446 regs
->start
[0] = pos
;
4447 regs
->end
[0] = (MATCHING_IN_FIRST_STRING
4448 ? ((regoff_t
) (d
- string1
))
4449 : ((regoff_t
) (d
- string2
+ size1
)));
4452 /* Go through the first `min (num_regs, regs->num_regs)'
4453 registers, since that is all we initialized. */
4454 for (mcnt
= 1; mcnt
< MIN (num_regs
, regs
->num_regs
); mcnt
++)
4456 if (REG_UNSET (regstart
[mcnt
]) || REG_UNSET (regend
[mcnt
]))
4457 regs
->start
[mcnt
] = regs
->end
[mcnt
] = -1;
4461 = (regoff_t
) POINTER_TO_OFFSET (regstart
[mcnt
]);
4463 = (regoff_t
) POINTER_TO_OFFSET (regend
[mcnt
]);
4467 /* If the regs structure we return has more elements than
4468 were in the pattern, set the extra elements to -1. If
4469 we (re)allocated the registers, this is the case,
4470 because we always allocate enough to have at least one
4472 for (mcnt
= num_regs
; mcnt
< regs
->num_regs
; mcnt
++)
4473 regs
->start
[mcnt
] = regs
->end
[mcnt
] = -1;
4474 } /* regs && !bufp->no_sub */
4476 DEBUG_PRINT4 ("%u failure points pushed, %u popped (%u remain).\n",
4477 nfailure_points_pushed
, nfailure_points_popped
,
4478 nfailure_points_pushed
- nfailure_points_popped
);
4479 DEBUG_PRINT2 ("%u registers pushed.\n", num_regs_pushed
);
4481 mcnt
= d
- pos
- (MATCHING_IN_FIRST_STRING
4485 DEBUG_PRINT2 ("Returning %d from re_match_2.\n", mcnt
);
4491 /* Otherwise match next pattern command. */
4492 switch (SWITCH_ENUM_CAST ((re_opcode_t
) *p
++))
4494 /* Ignore these. Used to ignore the n of succeed_n's which
4495 currently have n == 0. */
4497 DEBUG_PRINT1 ("EXECUTING no_op.\n");
4501 DEBUG_PRINT1 ("EXECUTING succeed.\n");
4504 /* Match the next n pattern characters exactly. The following
4505 byte in the pattern defines n, and the n bytes after that
4506 are the characters to match. */
4509 DEBUG_PRINT2 ("EXECUTING exactn %d.\n", mcnt
);
4511 /* This is written out as an if-else so we don't waste time
4512 testing `translate' inside the loop. */
4518 if ((unsigned char) RE_TRANSLATE (translate
, (unsigned char) *d
++)
4519 != (unsigned char) *p
++)
4529 if (*d
++ != (char) *p
++) goto fail
;
4533 SET_REGS_MATCHED ();
4537 /* Match any character except possibly a newline or a null. */
4539 DEBUG_PRINT1 ("EXECUTING anychar.\n");
4543 if ((!(bufp
->syntax
& RE_DOT_NEWLINE
) && TRANSLATE (*d
) == '\n')
4544 || (bufp
->syntax
& RE_DOT_NOT_NULL
&& TRANSLATE (*d
) == '\000'))
4547 SET_REGS_MATCHED ();
4548 DEBUG_PRINT2 (" Matched `%d'.\n", *d
);
4549 d
+= multibyte
? MULTIBYTE_FORM_LENGTH (d
, dend
- d
) : 1;
4556 register unsigned int c
;
4557 boolean
not = (re_opcode_t
) *(p
- 1) == charset_not
;
4560 /* Start of actual range_table, or end of bitmap if there is no
4562 unsigned char *range_table
;
4564 /* Nonzero if there is range table. */
4565 int range_table_exists
;
4567 /* Number of ranges of range table. Not in bytes. */
4570 DEBUG_PRINT2 ("EXECUTING charset%s.\n", not ? "_not" : "");
4573 c
= (unsigned char) *d
;
4575 range_table
= CHARSET_RANGE_TABLE (&p
[-1]); /* Past the bitmap. */
4576 range_table_exists
= CHARSET_RANGE_TABLE_EXISTS_P (&p
[-1]);
4577 if (range_table_exists
)
4578 EXTRACT_NUMBER_AND_INCR (count
, range_table
);
4582 if (multibyte
&& BASE_LEADING_CODE_P (c
))
4583 c
= STRING_CHAR_AND_LENGTH (d
, dend
- d
, len
);
4585 if (SINGLE_BYTE_CHAR_P (c
))
4586 { /* Lookup bitmap. */
4587 c
= TRANSLATE (c
); /* The character to match. */
4590 /* Cast to `unsigned' instead of `unsigned char' in
4591 case the bit list is a full 32 bytes long. */
4592 if (c
< (unsigned) (CHARSET_BITMAP_SIZE (&p
[-1]) * BYTEWIDTH
)
4593 && p
[1 + c
/ BYTEWIDTH
] & (1 << (c
% BYTEWIDTH
)))
4596 else if (range_table_exists
)
4597 CHARSET_LOOKUP_RANGE_TABLE_RAW (not, c
, range_table
, count
);
4599 p
= CHARSET_RANGE_TABLE_END (range_table
, count
);
4601 if (!not) goto fail
;
4603 SET_REGS_MATCHED ();
4609 /* The beginning of a group is represented by start_memory.
4610 The arguments are the register number in the next byte, and the
4611 number of groups inner to this one in the next. The text
4612 matched within the group is recorded (in the internal
4613 registers data structure) under the register number. */
4615 DEBUG_PRINT3 ("EXECUTING start_memory %d (%d):\n", *p
, p
[1]);
4617 /* Find out if this group can match the empty string. */
4618 p1
= p
; /* To send to group_match_null_string_p. */
4620 if (REG_MATCH_NULL_STRING_P (reg_info
[*p
]) == MATCH_NULL_UNSET_VALUE
)
4621 REG_MATCH_NULL_STRING_P (reg_info
[*p
])
4622 = group_match_null_string_p (&p1
, pend
, reg_info
);
4624 /* Save the position in the string where we were the last time
4625 we were at this open-group operator in case the group is
4626 operated upon by a repetition operator, e.g., with `(a*)*b'
4627 against `ab'; then we want to ignore where we are now in
4628 the string in case this attempt to match fails. */
4629 old_regstart
[*p
] = REG_MATCH_NULL_STRING_P (reg_info
[*p
])
4630 ? REG_UNSET (regstart
[*p
]) ? d
: regstart
[*p
]
4632 DEBUG_PRINT2 (" old_regstart: %d\n",
4633 POINTER_TO_OFFSET (old_regstart
[*p
]));
4636 DEBUG_PRINT2 (" regstart: %d\n", POINTER_TO_OFFSET (regstart
[*p
]));
4638 IS_ACTIVE (reg_info
[*p
]) = 1;
4639 MATCHED_SOMETHING (reg_info
[*p
]) = 0;
4641 /* Clear this whenever we change the register activity status. */
4642 set_regs_matched_done
= 0;
4644 /* This is the new highest active register. */
4645 highest_active_reg
= *p
;
4647 /* If nothing was active before, this is the new lowest active
4649 if (lowest_active_reg
== NO_LOWEST_ACTIVE_REG
)
4650 lowest_active_reg
= *p
;
4652 /* Move past the register number and inner group count. */
4654 just_past_start_mem
= p
;
4659 /* The stop_memory opcode represents the end of a group. Its
4660 arguments are the same as start_memory's: the register
4661 number, and the number of inner groups. */
4663 DEBUG_PRINT3 ("EXECUTING stop_memory %d (%d):\n", *p
, p
[1]);
4665 /* We need to save the string position the last time we were at
4666 this close-group operator in case the group is operated
4667 upon by a repetition operator, e.g., with `((a*)*(b*)*)*'
4668 against `aba'; then we want to ignore where we are now in
4669 the string in case this attempt to match fails. */
4670 old_regend
[*p
] = REG_MATCH_NULL_STRING_P (reg_info
[*p
])
4671 ? REG_UNSET (regend
[*p
]) ? d
: regend
[*p
]
4673 DEBUG_PRINT2 (" old_regend: %d\n",
4674 POINTER_TO_OFFSET (old_regend
[*p
]));
4677 DEBUG_PRINT2 (" regend: %d\n", POINTER_TO_OFFSET (regend
[*p
]));
4679 /* This register isn't active anymore. */
4680 IS_ACTIVE (reg_info
[*p
]) = 0;
4682 /* Clear this whenever we change the register activity status. */
4683 set_regs_matched_done
= 0;
4685 /* If this was the only register active, nothing is active
4687 if (lowest_active_reg
== highest_active_reg
)
4689 lowest_active_reg
= NO_LOWEST_ACTIVE_REG
;
4690 highest_active_reg
= NO_HIGHEST_ACTIVE_REG
;
4693 { /* We must scan for the new highest active register, since
4694 it isn't necessarily one less than now: consider
4695 (a(b)c(d(e)f)g). When group 3 ends, after the f), the
4696 new highest active register is 1. */
4697 unsigned char r
= *p
- 1;
4698 while (r
> 0 && !IS_ACTIVE (reg_info
[r
]))
4701 /* If we end up at register zero, that means that we saved
4702 the registers as the result of an `on_failure_jump', not
4703 a `start_memory', and we jumped to past the innermost
4704 `stop_memory'. For example, in ((.)*) we save
4705 registers 1 and 2 as a result of the *, but when we pop
4706 back to the second ), we are at the stop_memory 1.
4707 Thus, nothing is active. */
4710 lowest_active_reg
= NO_LOWEST_ACTIVE_REG
;
4711 highest_active_reg
= NO_HIGHEST_ACTIVE_REG
;
4714 highest_active_reg
= r
;
4717 /* If just failed to match something this time around with a
4718 group that's operated on by a repetition operator, try to
4719 force exit from the ``loop'', and restore the register
4720 information for this group that we had before trying this
4722 if ((!MATCHED_SOMETHING (reg_info
[*p
])
4723 || just_past_start_mem
== p
- 1)
4726 boolean is_a_jump_n
= false;
4730 switch ((re_opcode_t
) *p1
++)
4734 case pop_failure_jump
:
4735 case maybe_pop_jump
:
4737 case dummy_failure_jump
:
4738 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
4748 /* If the next operation is a jump backwards in the pattern
4749 to an on_failure_jump right before the start_memory
4750 corresponding to this stop_memory, exit from the loop
4751 by forcing a failure after pushing on the stack the
4752 on_failure_jump's jump in the pattern, and d. */
4753 if (mcnt
< 0 && (re_opcode_t
) *p1
== on_failure_jump
4754 && (re_opcode_t
) p1
[3] == start_memory
&& p1
[4] == *p
)
4756 /* If this group ever matched anything, then restore
4757 what its registers were before trying this last
4758 failed match, e.g., with `(a*)*b' against `ab' for
4759 regstart[1], and, e.g., with `((a*)*(b*)*)*'
4760 against `aba' for regend[3].
4762 Also restore the registers for inner groups for,
4763 e.g., `((a*)(b*))*' against `aba' (register 3 would
4764 otherwise get trashed). */
4766 if (EVER_MATCHED_SOMETHING (reg_info
[*p
]))
4770 EVER_MATCHED_SOMETHING (reg_info
[*p
]) = 0;
4772 /* Restore this and inner groups' (if any) registers. */
4773 for (r
= *p
; r
< *p
+ *(p
+ 1); r
++)
4775 regstart
[r
] = old_regstart
[r
];
4777 /* xx why this test? */
4778 if (old_regend
[r
] >= regstart
[r
])
4779 regend
[r
] = old_regend
[r
];
4783 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
4784 PUSH_FAILURE_POINT (p1
+ mcnt
, d
, -2);
4790 /* Move past the register number and the inner group count. */
4795 /* \<digit> has been turned into a `duplicate' command which is
4796 followed by the numeric value of <digit> as the register number. */
4799 register const char *d2
, *dend2
;
4800 int regno
= *p
++; /* Get which register to match against. */
4801 DEBUG_PRINT2 ("EXECUTING duplicate %d.\n", regno
);
4803 /* Can't back reference a group which we've never matched. */
4804 if (REG_UNSET (regstart
[regno
]) || REG_UNSET (regend
[regno
]))
4807 /* Where in input to try to start matching. */
4808 d2
= regstart
[regno
];
4810 /* Where to stop matching; if both the place to start and
4811 the place to stop matching are in the same string, then
4812 set to the place to stop, otherwise, for now have to use
4813 the end of the first string. */
4815 dend2
= ((FIRST_STRING_P (regstart
[regno
])
4816 == FIRST_STRING_P (regend
[regno
]))
4817 ? regend
[regno
] : end_match_1
);
4820 /* If necessary, advance to next segment in register
4824 if (dend2
== end_match_2
) break;
4825 if (dend2
== regend
[regno
]) break;
4827 /* End of string1 => advance to string2. */
4829 dend2
= regend
[regno
];
4831 /* At end of register contents => success */
4832 if (d2
== dend2
) break;
4834 /* If necessary, advance to next segment in data. */
4837 /* How many characters left in this segment to match. */
4840 /* Want how many consecutive characters we can match in
4841 one shot, so, if necessary, adjust the count. */
4842 if (mcnt
> dend2
- d2
)
4845 /* Compare that many; failure if mismatch, else move
4848 ? bcmp_translate (d
, d2
, mcnt
, translate
)
4849 : bcmp (d
, d2
, mcnt
))
4851 d
+= mcnt
, d2
+= mcnt
;
4853 /* Do this because we've match some characters. */
4854 SET_REGS_MATCHED ();
4860 /* begline matches the empty string at the beginning of the string
4861 (unless `not_bol' is set in `bufp'), and, if
4862 `newline_anchor' is set, after newlines. */
4864 DEBUG_PRINT1 ("EXECUTING begline.\n");
4866 if (AT_STRINGS_BEG (d
))
4868 if (!bufp
->not_bol
) break;
4870 else if (d
[-1] == '\n' && bufp
->newline_anchor
)
4874 /* In all other cases, we fail. */
4878 /* endline is the dual of begline. */
4880 DEBUG_PRINT1 ("EXECUTING endline.\n");
4882 if (AT_STRINGS_END (d
))
4884 if (!bufp
->not_eol
) break;
4887 /* We have to ``prefetch'' the next character. */
4888 else if ((d
== end1
? *string2
: *d
) == '\n'
4889 && bufp
->newline_anchor
)
4896 /* Match at the very beginning of the data. */
4898 DEBUG_PRINT1 ("EXECUTING begbuf.\n");
4899 if (AT_STRINGS_BEG (d
))
4904 /* Match at the very end of the data. */
4906 DEBUG_PRINT1 ("EXECUTING endbuf.\n");
4907 if (AT_STRINGS_END (d
))
4912 /* on_failure_keep_string_jump is used to optimize `.*\n'. It
4913 pushes NULL as the value for the string on the stack. Then
4914 `pop_failure_point' will keep the current value for the
4915 string, instead of restoring it. To see why, consider
4916 matching `foo\nbar' against `.*\n'. The .* matches the foo;
4917 then the . fails against the \n. But the next thing we want
4918 to do is match the \n against the \n; if we restored the
4919 string value, we would be back at the foo.
4921 Because this is used only in specific cases, we don't need to
4922 check all the things that `on_failure_jump' does, to make
4923 sure the right things get saved on the stack. Hence we don't
4924 share its code. The only reason to push anything on the
4925 stack at all is that otherwise we would have to change
4926 `anychar's code to do something besides goto fail in this
4927 case; that seems worse than this. */
4928 case on_failure_keep_string_jump
:
4929 DEBUG_PRINT1 ("EXECUTING on_failure_keep_string_jump");
4931 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
4932 DEBUG_PRINT3 (" %d (to 0x%x):\n", mcnt
, p
+ mcnt
);
4934 PUSH_FAILURE_POINT (p
+ mcnt
, NULL
, -2);
4938 /* Uses of on_failure_jump:
4940 Each alternative starts with an on_failure_jump that points
4941 to the beginning of the next alternative. Each alternative
4942 except the last ends with a jump that in effect jumps past
4943 the rest of the alternatives. (They really jump to the
4944 ending jump of the following alternative, because tensioning
4945 these jumps is a hassle.)
4947 Repeats start with an on_failure_jump that points past both
4948 the repetition text and either the following jump or
4949 pop_failure_jump back to this on_failure_jump. */
4950 case on_failure_jump
:
4952 DEBUG_PRINT1 ("EXECUTING on_failure_jump");
4954 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
4955 DEBUG_PRINT3 (" %d (to 0x%x)", mcnt
, p
+ mcnt
);
4957 /* If this on_failure_jump comes right before a group (i.e.,
4958 the original * applied to a group), save the information
4959 for that group and all inner ones, so that if we fail back
4960 to this point, the group's information will be correct.
4961 For example, in \(a*\)*\1, we need the preceding group,
4962 and in \(zz\(a*\)b*\)\2, we need the inner group. */
4964 /* We can't use `p' to check ahead because we push
4965 a failure point to `p + mcnt' after we do this. */
4968 /* We need to skip no_op's before we look for the
4969 start_memory in case this on_failure_jump is happening as
4970 the result of a completed succeed_n, as in \(a\)\{1,3\}b\1
4972 while (p1
< pend
&& (re_opcode_t
) *p1
== no_op
)
4975 if (p1
< pend
&& (re_opcode_t
) *p1
== start_memory
)
4977 /* We have a new highest active register now. This will
4978 get reset at the start_memory we are about to get to,
4979 but we will have saved all the registers relevant to
4980 this repetition op, as described above. */
4981 highest_active_reg
= *(p1
+ 1) + *(p1
+ 2);
4982 if (lowest_active_reg
== NO_LOWEST_ACTIVE_REG
)
4983 lowest_active_reg
= *(p1
+ 1);
4986 DEBUG_PRINT1 (":\n");
4987 PUSH_FAILURE_POINT (p
+ mcnt
, d
, -2);
4991 /* A smart repeat ends with `maybe_pop_jump'.
4992 We change it to either `pop_failure_jump' or `jump'. */
4993 case maybe_pop_jump
:
4994 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
4995 DEBUG_PRINT2 ("EXECUTING maybe_pop_jump %d.\n", mcnt
);
4997 register unsigned char *p2
= p
;
4999 /* Compare the beginning of the repeat with what in the
5000 pattern follows its end. If we can establish that there
5001 is nothing that they would both match, i.e., that we
5002 would have to backtrack because of (as in, e.g., `a*a')
5003 then we can change to pop_failure_jump, because we'll
5004 never have to backtrack.
5006 This is not true in the case of alternatives: in
5007 `(a|ab)*' we do need to backtrack to the `ab' alternative
5008 (e.g., if the string was `ab'). But instead of trying to
5009 detect that here, the alternative has put on a dummy
5010 failure point which is what we will end up popping. */
5012 /* Skip over open/close-group commands.
5013 If what follows this loop is a ...+ construct,
5014 look at what begins its body, since we will have to
5015 match at least one of that. */
5019 && ((re_opcode_t
) *p2
== stop_memory
5020 || (re_opcode_t
) *p2
== start_memory
))
5022 else if (p2
+ 6 < pend
5023 && (re_opcode_t
) *p2
== dummy_failure_jump
)
5030 /* p1[0] ... p1[2] are the `on_failure_jump' corresponding
5031 to the `maybe_finalize_jump' of this case. Examine what
5034 /* If we're at the end of the pattern, we can change. */
5037 /* Consider what happens when matching ":\(.*\)"
5038 against ":/". I don't really understand this code
5040 p
[-3] = (unsigned char) pop_failure_jump
;
5042 (" End of pattern: change to `pop_failure_jump'.\n");
5045 else if ((re_opcode_t
) *p2
== exactn
5046 || (bufp
->newline_anchor
&& (re_opcode_t
) *p2
== endline
))
5048 register unsigned int c
5049 = *p2
== (unsigned char) endline
? '\n' : p2
[2];
5051 if ((re_opcode_t
) p1
[3] == exactn
)
5053 if (!(multibyte
/* && (c != '\n') */
5054 && BASE_LEADING_CODE_P (c
))
5056 : (STRING_CHAR (&p2
[2], pend
- &p2
[2])
5057 != STRING_CHAR (&p1
[5], pend
- &p1
[5])))
5059 p
[-3] = (unsigned char) pop_failure_jump
;
5060 DEBUG_PRINT3 (" %c != %c => pop_failure_jump.\n",
5065 else if ((re_opcode_t
) p1
[3] == charset
5066 || (re_opcode_t
) p1
[3] == charset_not
)
5068 int not = (re_opcode_t
) p1
[3] == charset_not
;
5070 if (multibyte
/* && (c != '\n') */
5071 && BASE_LEADING_CODE_P (c
))
5072 c
= STRING_CHAR (&p2
[2], pend
- &p2
[2]);
5074 /* Test if C is listed in charset (or charset_not)
5076 if (SINGLE_BYTE_CHAR_P (c
))
5078 if (c
< CHARSET_BITMAP_SIZE (&p1
[3]) * BYTEWIDTH
5079 && p1
[5 + c
/ BYTEWIDTH
] & (1 << (c
% BYTEWIDTH
)))
5082 else if (CHARSET_RANGE_TABLE_EXISTS_P (&p1
[3]))
5083 CHARSET_LOOKUP_RANGE_TABLE (not, c
, &p1
[3]);
5085 /* `not' is equal to 1 if c would match, which means
5086 that we can't change to pop_failure_jump. */
5089 p
[-3] = (unsigned char) pop_failure_jump
;
5090 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
5094 else if ((re_opcode_t
) *p2
== charset
)
5096 if ((re_opcode_t
) p1
[3] == exactn
)
5098 register unsigned int c
= p1
[5];
5101 if (multibyte
&& BASE_LEADING_CODE_P (c
))
5102 c
= STRING_CHAR (&p1
[5], pend
- &p1
[5]);
5104 /* Test if C is listed in charset at `p2'. */
5105 if (SINGLE_BYTE_CHAR_P (c
))
5107 if (c
< CHARSET_BITMAP_SIZE (p2
) * BYTEWIDTH
5108 && (p2
[2 + c
/ BYTEWIDTH
]
5109 & (1 << (c
% BYTEWIDTH
))))
5112 else if (CHARSET_RANGE_TABLE_EXISTS_P (p2
))
5113 CHARSET_LOOKUP_RANGE_TABLE (not, c
, p2
);
5117 p
[-3] = (unsigned char) pop_failure_jump
;
5118 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
5122 /* It is hard to list up all the character in charset
5123 P2 if it includes multibyte character. Give up in
5125 else if (!multibyte
|| !CHARSET_RANGE_TABLE_EXISTS_P (p2
))
5127 /* Now, we are sure that P2 has no range table.
5128 So, for the size of bitmap in P2, `p2[1]' is
5129 enough. But P1 may have range table, so the
5130 size of bitmap table of P1 is extracted by
5131 using macro `CHARSET_BITMAP_SIZE'.
5133 Since we know that all the character listed in
5134 P2 is ASCII, it is enough to test only bitmap
5137 if ((re_opcode_t
) p1
[3] == charset_not
)
5140 /* We win if the charset_not inside the loop lists
5141 every character listed in the charset after. */
5142 for (idx
= 0; idx
< (int) p2
[1]; idx
++)
5143 if (! (p2
[2 + idx
] == 0
5144 || (idx
< CHARSET_BITMAP_SIZE (&p1
[3])
5145 && ((p2
[2 + idx
] & ~ p1
[5 + idx
]) == 0))))
5150 p
[-3] = (unsigned char) pop_failure_jump
;
5151 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
5154 else if ((re_opcode_t
) p1
[3] == charset
)
5157 /* We win if the charset inside the loop
5158 has no overlap with the one after the loop. */
5161 && idx
< CHARSET_BITMAP_SIZE (&p1
[3]));
5163 if ((p2
[2 + idx
] & p1
[5 + idx
]) != 0)
5167 || idx
== CHARSET_BITMAP_SIZE (&p1
[3]))
5169 p
[-3] = (unsigned char) pop_failure_jump
;
5170 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
5176 p
-= 2; /* Point at relative address again. */
5177 if ((re_opcode_t
) p
[-1] != pop_failure_jump
)
5179 p
[-1] = (unsigned char) jump
;
5180 DEBUG_PRINT1 (" Match => jump.\n");
5181 goto unconditional_jump
;
5183 /* Note fall through. */
5186 /* The end of a simple repeat has a pop_failure_jump back to
5187 its matching on_failure_jump, where the latter will push a
5188 failure point. The pop_failure_jump takes off failure
5189 points put on by this pop_failure_jump's matching
5190 on_failure_jump; we got through the pattern to here from the
5191 matching on_failure_jump, so didn't fail. */
5192 case pop_failure_jump
:
5194 /* We need to pass separate storage for the lowest and
5195 highest registers, even though we don't care about the
5196 actual values. Otherwise, we will restore only one
5197 register from the stack, since lowest will == highest in
5198 `pop_failure_point'. */
5199 unsigned dummy_low_reg
, dummy_high_reg
;
5200 unsigned char *pdummy
;
5203 DEBUG_PRINT1 ("EXECUTING pop_failure_jump.\n");
5204 POP_FAILURE_POINT (sdummy
, pdummy
,
5205 dummy_low_reg
, dummy_high_reg
,
5206 reg_dummy
, reg_dummy
, reg_info_dummy
);
5208 /* Note fall through. */
5211 /* Unconditionally jump (without popping any failure points). */
5214 EXTRACT_NUMBER_AND_INCR (mcnt
, p
); /* Get the amount to jump. */
5215 DEBUG_PRINT2 ("EXECUTING jump %d ", mcnt
);
5216 p
+= mcnt
; /* Do the jump. */
5217 DEBUG_PRINT2 ("(to 0x%x).\n", p
);
5221 /* We need this opcode so we can detect where alternatives end
5222 in `group_match_null_string_p' et al. */
5224 DEBUG_PRINT1 ("EXECUTING jump_past_alt.\n");
5225 goto unconditional_jump
;
5228 /* Normally, the on_failure_jump pushes a failure point, which
5229 then gets popped at pop_failure_jump. We will end up at
5230 pop_failure_jump, also, and with a pattern of, say, `a+', we
5231 are skipping over the on_failure_jump, so we have to push
5232 something meaningless for pop_failure_jump to pop. */
5233 case dummy_failure_jump
:
5234 DEBUG_PRINT1 ("EXECUTING dummy_failure_jump.\n");
5235 /* It doesn't matter what we push for the string here. What
5236 the code at `fail' tests is the value for the pattern. */
5237 PUSH_FAILURE_POINT (0, 0, -2);
5238 goto unconditional_jump
;
5241 /* At the end of an alternative, we need to push a dummy failure
5242 point in case we are followed by a `pop_failure_jump', because
5243 we don't want the failure point for the alternative to be
5244 popped. For example, matching `(a|ab)*' against `aab'
5245 requires that we match the `ab' alternative. */
5246 case push_dummy_failure
:
5247 DEBUG_PRINT1 ("EXECUTING push_dummy_failure.\n");
5248 /* See comments just above at `dummy_failure_jump' about the
5250 PUSH_FAILURE_POINT (0, 0, -2);
5253 /* Have to succeed matching what follows at least n times.
5254 After that, handle like `on_failure_jump'. */
5256 EXTRACT_NUMBER (mcnt
, p
+ 2);
5257 DEBUG_PRINT2 ("EXECUTING succeed_n %d.\n", mcnt
);
5260 /* Originally, this is how many times we HAVE to succeed. */
5265 STORE_NUMBER_AND_INCR (p
, mcnt
);
5266 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p
, mcnt
);
5270 DEBUG_PRINT2 (" Setting two bytes from 0x%x to no_op.\n", p
+2);
5271 p
[2] = (unsigned char) no_op
;
5272 p
[3] = (unsigned char) no_op
;
5278 EXTRACT_NUMBER (mcnt
, p
+ 2);
5279 DEBUG_PRINT2 ("EXECUTING jump_n %d.\n", mcnt
);
5281 /* Originally, this is how many times we CAN jump. */
5285 STORE_NUMBER (p
+ 2, mcnt
);
5286 goto unconditional_jump
;
5288 /* If don't have to jump any more, skip over the rest of command. */
5295 DEBUG_PRINT1 ("EXECUTING set_number_at.\n");
5297 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5299 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5300 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p1
, mcnt
);
5301 STORE_NUMBER (p1
, mcnt
);
5306 DEBUG_PRINT1 ("EXECUTING wordbound.\n");
5308 /* We SUCCEED in one of the following cases: */
5310 /* Case 1: D is at the beginning or the end of string. */
5311 if (AT_STRINGS_BEG (d
) || AT_STRINGS_END (d
))
5315 /* C1 is the character before D, S1 is the syntax of C1, C2
5316 is the character at D, and S2 is the syntax of C2. */
5318 int pos1
= PTR_TO_OFFSET (d
- 1);
5320 GET_CHAR_BEFORE_2 (c1
, d
, string1
, end1
, string2
, end2
);
5321 GET_CHAR_AFTER_2 (c2
, d
, string1
, end1
, string2
, end2
);
5323 UPDATE_SYNTAX_TABLE (pos1
? pos1
: 1);
5327 UPDATE_SYNTAX_TABLE_FORWARD (pos1
+ 1);
5331 if (/* Case 2: Only one of S1 and S2 is Sword. */
5332 ((s1
== Sword
) != (s2
== Sword
))
5333 /* Case 3: Both of S1 and S2 are Sword, and macro
5334 WORD_BOUNDARY_P (C1, C2) returns nonzero. */
5335 || ((s1
== Sword
) && WORD_BOUNDARY_P (c1
, c2
)))
5341 DEBUG_PRINT1 ("EXECUTING notwordbound.\n");
5343 /* We FAIL in one of the following cases: */
5345 /* Case 1: D is at the beginning or the end of string. */
5346 if (AT_STRINGS_BEG (d
) || AT_STRINGS_END (d
))
5350 /* C1 is the character before D, S1 is the syntax of C1, C2
5351 is the character at D, and S2 is the syntax of C2. */
5353 int pos1
= PTR_TO_OFFSET (d
- 1);
5355 GET_CHAR_BEFORE_2 (c1
, d
, string1
, end1
, string2
, end2
);
5356 GET_CHAR_AFTER_2 (c2
, d
, string1
, end1
, string2
, end2
);
5358 UPDATE_SYNTAX_TABLE (pos1
);
5362 UPDATE_SYNTAX_TABLE_FORWARD (pos1
+ 1);
5366 if (/* Case 2: Only one of S1 and S2 is Sword. */
5367 ((s1
== Sword
) != (s2
== Sword
))
5368 /* Case 3: Both of S1 and S2 are Sword, and macro
5369 WORD_BOUNDARY_P (C1, C2) returns nonzero. */
5370 || ((s1
== Sword
) && WORD_BOUNDARY_P (c1
, c2
)))
5376 DEBUG_PRINT1 ("EXECUTING wordbeg.\n");
5378 /* We FAIL in one of the following cases: */
5380 /* Case 1: D is at the end of string. */
5381 if (AT_STRINGS_END (d
))
5385 /* C1 is the character before D, S1 is the syntax of C1, C2
5386 is the character at D, and S2 is the syntax of C2. */
5388 int pos1
= PTR_TO_OFFSET (d
);
5390 GET_CHAR_AFTER_2 (c2
, d
, string1
, end1
, string2
, end2
);
5392 UPDATE_SYNTAX_TABLE (pos1
);
5396 /* Case 2: S2 is not Sword. */
5400 /* Case 3: D is not at the beginning of string ... */
5401 if (!AT_STRINGS_BEG (d
))
5403 GET_CHAR_BEFORE_2 (c1
, d
, string1
, end1
, string2
, end2
);
5405 UPDATE_SYNTAX_TABLE_BACKWARD (pos1
- 1);
5409 /* ... and S1 is Sword, and WORD_BOUNDARY_P (C1, C2)
5411 if ((s1
== Sword
) && !WORD_BOUNDARY_P (c1
, c2
))
5418 DEBUG_PRINT1 ("EXECUTING wordend.\n");
5420 /* We FAIL in one of the following cases: */
5422 /* Case 1: D is at the beginning of string. */
5423 if (AT_STRINGS_BEG (d
))
5427 /* C1 is the character before D, S1 is the syntax of C1, C2
5428 is the character at D, and S2 is the syntax of C2. */
5431 GET_CHAR_BEFORE_2 (c1
, d
, string1
, end1
, string2
, end2
);
5434 /* Case 2: S1 is not Sword. */
5438 /* Case 3: D is not at the end of string ... */
5439 if (!AT_STRINGS_END (d
))
5441 GET_CHAR_AFTER_2 (c2
, d
, string1
, end1
, string2
, end2
);
5444 /* ... and S2 is Sword, and WORD_BOUNDARY_P (C1, C2)
5446 if ((s2
== Sword
) && !WORD_BOUNDARY_P (c1
, c2
))
5454 DEBUG_PRINT1 ("EXECUTING before_dot.\n");
5455 if (PTR_CHAR_POS ((unsigned char *) d
) >= PT
)
5460 DEBUG_PRINT1 ("EXECUTING at_dot.\n");
5461 if (PTR_CHAR_POS ((unsigned char *) d
) != PT
)
5466 DEBUG_PRINT1 ("EXECUTING after_dot.\n");
5467 if (PTR_CHAR_POS ((unsigned char *) d
) <= PT
)
5472 DEBUG_PRINT2 ("EXECUTING syntaxspec %d.\n", mcnt
);
5477 DEBUG_PRINT1 ("EXECUTING Emacs wordchar.\n");
5483 int pos1
= PTR_TO_OFFSET (d
);
5484 UPDATE_SYNTAX_TABLE (pos1
);
5491 /* we must concern about multibyte form, ... */
5492 c
= STRING_CHAR_AND_LENGTH (d
, dend
- d
, len
);
5494 /* everything should be handled as ASCII, even though it
5495 looks like multibyte form. */
5498 if (SYNTAX (c
) != (enum syntaxcode
) mcnt
)
5502 SET_REGS_MATCHED ();
5506 DEBUG_PRINT2 ("EXECUTING notsyntaxspec %d.\n", mcnt
);
5508 goto matchnotsyntax
;
5511 DEBUG_PRINT1 ("EXECUTING Emacs notwordchar.\n");
5517 int pos1
= PTR_TO_OFFSET (d
);
5518 UPDATE_SYNTAX_TABLE (pos1
);
5525 c
= STRING_CHAR_AND_LENGTH (d
, dend
- d
, len
);
5529 if (SYNTAX (c
) == (enum syntaxcode
) mcnt
)
5533 SET_REGS_MATCHED ();
5537 DEBUG_PRINT2 ("EXECUTING categoryspec %d.\n", *p
);
5544 c
= STRING_CHAR_AND_LENGTH (d
, dend
- d
, len
);
5548 if (!CHAR_HAS_CATEGORY (c
, mcnt
))
5552 SET_REGS_MATCHED ();
5555 case notcategoryspec
:
5556 DEBUG_PRINT2 ("EXECUTING notcategoryspec %d.\n", *p
);
5563 c
= STRING_CHAR_AND_LENGTH (d
, dend
- d
, len
);
5567 if (CHAR_HAS_CATEGORY (c
, mcnt
))
5571 SET_REGS_MATCHED ();
5574 #else /* not emacs */
5576 DEBUG_PRINT1 ("EXECUTING non-Emacs wordchar.\n");
5578 if (!WORDCHAR_P (d
))
5580 SET_REGS_MATCHED ();
5585 DEBUG_PRINT1 ("EXECUTING non-Emacs notwordchar.\n");
5589 SET_REGS_MATCHED ();
5592 #endif /* not emacs */
5597 continue; /* Successfully executed one pattern command; keep going. */
5600 /* We goto here if a matching operation fails. */
5602 if (!FAIL_STACK_EMPTY ())
5603 { /* A restart point is known. Restore to that state. */
5604 DEBUG_PRINT1 ("\nFAIL:\n");
5605 POP_FAILURE_POINT (d
, p
,
5606 lowest_active_reg
, highest_active_reg
,
5607 regstart
, regend
, reg_info
);
5609 /* If this failure point is a dummy, try the next one. */
5613 /* If we failed to the end of the pattern, don't examine *p. */
5617 boolean is_a_jump_n
= false;
5619 /* If failed to a backwards jump that's part of a repetition
5620 loop, need to pop this failure point and use the next one. */
5621 switch ((re_opcode_t
) *p
)
5625 case maybe_pop_jump
:
5626 case pop_failure_jump
:
5629 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
5632 if ((is_a_jump_n
&& (re_opcode_t
) *p1
== succeed_n
)
5634 && (re_opcode_t
) *p1
== on_failure_jump
))
5642 if (d
>= string1
&& d
<= end1
)
5646 break; /* Matching at this starting point really fails. */
5650 goto restore_best_regs
;
5654 return -1; /* Failure to match. */
5657 /* Subroutine definitions for re_match_2. */
5660 /* We are passed P pointing to a register number after a start_memory.
5662 Return true if the pattern up to the corresponding stop_memory can
5663 match the empty string, and false otherwise.
5665 If we find the matching stop_memory, sets P to point to one past its number.
5666 Otherwise, sets P to an undefined byte less than or equal to END.
5668 We don't handle duplicates properly (yet). */
5671 group_match_null_string_p (p
, end
, reg_info
)
5672 unsigned char **p
, *end
;
5673 register_info_type
*reg_info
;
5676 /* Point to after the args to the start_memory. */
5677 unsigned char *p1
= *p
+ 2;
5681 /* Skip over opcodes that can match nothing, and return true or
5682 false, as appropriate, when we get to one that can't, or to the
5683 matching stop_memory. */
5685 switch ((re_opcode_t
) *p1
)
5687 /* Could be either a loop or a series of alternatives. */
5688 case on_failure_jump
:
5690 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
5692 /* If the next operation is not a jump backwards in the
5697 /* Go through the on_failure_jumps of the alternatives,
5698 seeing if any of the alternatives cannot match nothing.
5699 The last alternative starts with only a jump,
5700 whereas the rest start with on_failure_jump and end
5701 with a jump, e.g., here is the pattern for `a|b|c':
5703 /on_failure_jump/0/6/exactn/1/a/jump_past_alt/0/6
5704 /on_failure_jump/0/6/exactn/1/b/jump_past_alt/0/3
5707 So, we have to first go through the first (n-1)
5708 alternatives and then deal with the last one separately. */
5711 /* Deal with the first (n-1) alternatives, which start
5712 with an on_failure_jump (see above) that jumps to right
5713 past a jump_past_alt. */
5715 while ((re_opcode_t
) p1
[mcnt
-3] == jump_past_alt
)
5717 /* `mcnt' holds how many bytes long the alternative
5718 is, including the ending `jump_past_alt' and
5721 if (!alt_match_null_string_p (p1
, p1
+ mcnt
- 3,
5725 /* Move to right after this alternative, including the
5729 /* Break if it's the beginning of an n-th alternative
5730 that doesn't begin with an on_failure_jump. */
5731 if ((re_opcode_t
) *p1
!= on_failure_jump
)
5734 /* Still have to check that it's not an n-th
5735 alternative that starts with an on_failure_jump. */
5737 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
5738 if ((re_opcode_t
) p1
[mcnt
-3] != jump_past_alt
)
5740 /* Get to the beginning of the n-th alternative. */
5746 /* Deal with the last alternative: go back and get number
5747 of the `jump_past_alt' just before it. `mcnt' contains
5748 the length of the alternative. */
5749 EXTRACT_NUMBER (mcnt
, p1
- 2);
5751 if (!alt_match_null_string_p (p1
, p1
+ mcnt
, reg_info
))
5754 p1
+= mcnt
; /* Get past the n-th alternative. */
5760 assert (p1
[1] == **p
);
5766 if (!common_op_match_null_string_p (&p1
, end
, reg_info
))
5769 } /* while p1 < end */
5772 } /* group_match_null_string_p */
5775 /* Similar to group_match_null_string_p, but doesn't deal with alternatives:
5776 It expects P to be the first byte of a single alternative and END one
5777 byte past the last. The alternative can contain groups. */
5780 alt_match_null_string_p (p
, end
, reg_info
)
5781 unsigned char *p
, *end
;
5782 register_info_type
*reg_info
;
5785 unsigned char *p1
= p
;
5789 /* Skip over opcodes that can match nothing, and break when we get
5790 to one that can't. */
5792 switch ((re_opcode_t
) *p1
)
5795 case on_failure_jump
:
5797 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
5802 if (!common_op_match_null_string_p (&p1
, end
, reg_info
))
5805 } /* while p1 < end */
5808 } /* alt_match_null_string_p */
5811 /* Deals with the ops common to group_match_null_string_p and
5812 alt_match_null_string_p.
5814 Sets P to one after the op and its arguments, if any. */
5817 common_op_match_null_string_p (p
, end
, reg_info
)
5818 unsigned char **p
, *end
;
5819 register_info_type
*reg_info
;
5824 unsigned char *p1
= *p
;
5826 switch ((re_opcode_t
) *p1
++)
5846 assert (reg_no
> 0 && reg_no
<= MAX_REGNUM
);
5847 ret
= group_match_null_string_p (&p1
, end
, reg_info
);
5849 /* Have to set this here in case we're checking a group which
5850 contains a group and a back reference to it. */
5852 if (REG_MATCH_NULL_STRING_P (reg_info
[reg_no
]) == MATCH_NULL_UNSET_VALUE
)
5853 REG_MATCH_NULL_STRING_P (reg_info
[reg_no
]) = ret
;
5859 /* If this is an optimized succeed_n for zero times, make the jump. */
5861 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
5869 /* Get to the number of times to succeed. */
5871 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
5876 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
5884 if (!REG_MATCH_NULL_STRING_P (reg_info
[*p1
]))
5892 /* All other opcodes mean we cannot match the empty string. */
5898 } /* common_op_match_null_string_p */
5901 /* Return zero if TRANSLATE[S1] and TRANSLATE[S2] are identical for LEN
5902 bytes; nonzero otherwise. */
5905 bcmp_translate (s1
, s2
, len
, translate
)
5906 unsigned char *s1
, *s2
;
5908 RE_TRANSLATE_TYPE translate
;
5910 register unsigned char *p1
= s1
, *p2
= s2
;
5913 if (RE_TRANSLATE (translate
, *p1
++) != RE_TRANSLATE (translate
, *p2
++))
5920 /* Entry points for GNU code. */
5922 /* re_compile_pattern is the GNU regular expression compiler: it
5923 compiles PATTERN (of length SIZE) and puts the result in BUFP.
5924 Returns 0 if the pattern was valid, otherwise an error string.
5926 Assumes the `allocated' (and perhaps `buffer') and `translate' fields
5927 are set in BUFP on entry.
5929 We call regex_compile to do the actual compilation. */
5932 re_compile_pattern (pattern
, length
, bufp
)
5933 const char *pattern
;
5935 struct re_pattern_buffer
*bufp
;
5939 /* GNU code is written to assume at least RE_NREGS registers will be set
5940 (and at least one extra will be -1). */
5941 bufp
->regs_allocated
= REGS_UNALLOCATED
;
5943 /* And GNU code determines whether or not to get register information
5944 by passing null for the REGS argument to re_match, etc., not by
5948 /* Match anchors at newline. */
5949 bufp
->newline_anchor
= 1;
5951 ret
= regex_compile (pattern
, length
, re_syntax_options
, bufp
);
5955 return gettext (re_error_msgid
[(int) ret
]);
5958 /* Entry points compatible with 4.2 BSD regex library. We don't define
5959 them unless specifically requested. */
5961 #if defined (_REGEX_RE_COMP) || defined (_LIBC)
5963 /* BSD has one and only one pattern buffer. */
5964 static struct re_pattern_buffer re_comp_buf
;
5968 /* Make these definitions weak in libc, so POSIX programs can redefine
5969 these names if they don't use our functions, and still use
5970 regcomp/regexec below without link errors. */
5980 if (!re_comp_buf
.buffer
)
5981 return gettext ("No previous regular expression");
5985 if (!re_comp_buf
.buffer
)
5987 re_comp_buf
.buffer
= (unsigned char *) malloc (200);
5988 if (re_comp_buf
.buffer
== NULL
)
5989 return gettext (re_error_msgid
[(int) REG_ESPACE
]);
5990 re_comp_buf
.allocated
= 200;
5992 re_comp_buf
.fastmap
= (char *) malloc (1 << BYTEWIDTH
);
5993 if (re_comp_buf
.fastmap
== NULL
)
5994 return gettext (re_error_msgid
[(int) REG_ESPACE
]);
5997 /* Since `re_exec' always passes NULL for the `regs' argument, we
5998 don't need to initialize the pattern buffer fields which affect it. */
6000 /* Match anchors at newlines. */
6001 re_comp_buf
.newline_anchor
= 1;
6003 ret
= regex_compile (s
, strlen (s
), re_syntax_options
, &re_comp_buf
);
6008 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6009 return (char *) gettext (re_error_msgid
[(int) ret
]);
6020 const int len
= strlen (s
);
6022 0 <= re_search (&re_comp_buf
, s
, len
, 0, len
, (struct re_registers
*) 0);
6024 #endif /* _REGEX_RE_COMP */
6026 /* POSIX.2 functions. Don't define these for Emacs. */
6030 /* regcomp takes a regular expression as a string and compiles it.
6032 PREG is a regex_t *. We do not expect any fields to be initialized,
6033 since POSIX says we shouldn't. Thus, we set
6035 `buffer' to the compiled pattern;
6036 `used' to the length of the compiled pattern;
6037 `syntax' to RE_SYNTAX_POSIX_EXTENDED if the
6038 REG_EXTENDED bit in CFLAGS is set; otherwise, to
6039 RE_SYNTAX_POSIX_BASIC;
6040 `newline_anchor' to REG_NEWLINE being set in CFLAGS;
6041 `fastmap' and `fastmap_accurate' to zero;
6042 `re_nsub' to the number of subexpressions in PATTERN.
6044 PATTERN is the address of the pattern string.
6046 CFLAGS is a series of bits which affect compilation.
6048 If REG_EXTENDED is set, we use POSIX extended syntax; otherwise, we
6049 use POSIX basic syntax.
6051 If REG_NEWLINE is set, then . and [^...] don't match newline.
6052 Also, regexec will try a match beginning after every newline.
6054 If REG_ICASE is set, then we considers upper- and lowercase
6055 versions of letters to be equivalent when matching.
6057 If REG_NOSUB is set, then when PREG is passed to regexec, that
6058 routine will report only success or failure, and nothing about the
6061 It returns 0 if it succeeds, nonzero if it doesn't. (See regex.h for
6062 the return codes and their meanings.) */
6065 regcomp (preg
, pattern
, cflags
)
6067 const char *pattern
;
6072 = (cflags
& REG_EXTENDED
) ?
6073 RE_SYNTAX_POSIX_EXTENDED
: RE_SYNTAX_POSIX_BASIC
;
6075 /* regex_compile will allocate the space for the compiled pattern. */
6077 preg
->allocated
= 0;
6080 /* Don't bother to use a fastmap when searching. This simplifies the
6081 REG_NEWLINE case: if we used a fastmap, we'd have to put all the
6082 characters after newlines into the fastmap. This way, we just try
6086 if (cflags
& REG_ICASE
)
6091 = (RE_TRANSLATE_TYPE
) malloc (CHAR_SET_SIZE
6092 * sizeof (*(RE_TRANSLATE_TYPE
)0));
6093 if (preg
->translate
== NULL
)
6094 return (int) REG_ESPACE
;
6096 /* Map uppercase characters to corresponding lowercase ones. */
6097 for (i
= 0; i
< CHAR_SET_SIZE
; i
++)
6098 preg
->translate
[i
] = ISUPPER (i
) ? tolower (i
) : i
;
6101 preg
->translate
= NULL
;
6103 /* If REG_NEWLINE is set, newlines are treated differently. */
6104 if (cflags
& REG_NEWLINE
)
6105 { /* REG_NEWLINE implies neither . nor [^...] match newline. */
6106 syntax
&= ~RE_DOT_NEWLINE
;
6107 syntax
|= RE_HAT_LISTS_NOT_NEWLINE
;
6108 /* It also changes the matching behavior. */
6109 preg
->newline_anchor
= 1;
6112 preg
->newline_anchor
= 0;
6114 preg
->no_sub
= !!(cflags
& REG_NOSUB
);
6116 /* POSIX says a null character in the pattern terminates it, so we
6117 can use strlen here in compiling the pattern. */
6118 ret
= regex_compile (pattern
, strlen (pattern
), syntax
, preg
);
6120 /* POSIX doesn't distinguish between an unmatched open-group and an
6121 unmatched close-group: both are REG_EPAREN. */
6122 if (ret
== REG_ERPAREN
) ret
= REG_EPAREN
;
6128 /* regexec searches for a given pattern, specified by PREG, in the
6131 If NMATCH is zero or REG_NOSUB was set in the cflags argument to
6132 `regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at
6133 least NMATCH elements, and we set them to the offsets of the
6134 corresponding matched substrings.
6136 EFLAGS specifies `execution flags' which affect matching: if
6137 REG_NOTBOL is set, then ^ does not match at the beginning of the
6138 string; if REG_NOTEOL is set, then $ does not match at the end.
6140 We return 0 if we find a match and REG_NOMATCH if not. */
6143 regexec (preg
, string
, nmatch
, pmatch
, eflags
)
6144 const regex_t
*preg
;
6147 regmatch_t pmatch
[];
6151 struct re_registers regs
;
6152 regex_t private_preg
;
6153 int len
= strlen (string
);
6154 boolean want_reg_info
= !preg
->no_sub
&& nmatch
> 0;
6156 private_preg
= *preg
;
6158 private_preg
.not_bol
= !!(eflags
& REG_NOTBOL
);
6159 private_preg
.not_eol
= !!(eflags
& REG_NOTEOL
);
6161 /* The user has told us exactly how many registers to return
6162 information about, via `nmatch'. We have to pass that on to the
6163 matching routines. */
6164 private_preg
.regs_allocated
= REGS_FIXED
;
6168 regs
.num_regs
= nmatch
;
6169 regs
.start
= TALLOC (nmatch
, regoff_t
);
6170 regs
.end
= TALLOC (nmatch
, regoff_t
);
6171 if (regs
.start
== NULL
|| regs
.end
== NULL
)
6172 return (int) REG_NOMATCH
;
6175 /* Perform the searching operation. */
6176 ret
= re_search (&private_preg
, string
, len
,
6177 /* start: */ 0, /* range: */ len
,
6178 want_reg_info
? ®s
: (struct re_registers
*) 0);
6180 /* Copy the register information to the POSIX structure. */
6187 for (r
= 0; r
< nmatch
; r
++)
6189 pmatch
[r
].rm_so
= regs
.start
[r
];
6190 pmatch
[r
].rm_eo
= regs
.end
[r
];
6194 /* If we needed the temporary register info, free the space now. */
6199 /* We want zero return to mean success, unlike `re_search'. */
6200 return ret
>= 0 ? (int) REG_NOERROR
: (int) REG_NOMATCH
;
6204 /* Returns a message corresponding to an error code, ERRCODE, returned
6205 from either regcomp or regexec. We don't use PREG here. */
6208 regerror (errcode
, preg
, errbuf
, errbuf_size
)
6210 const regex_t
*preg
;
6218 || errcode
>= (sizeof (re_error_msgid
) / sizeof (re_error_msgid
[0])))
6219 /* Only error codes returned by the rest of the code should be passed
6220 to this routine. If we are given anything else, or if other regex
6221 code generates an invalid error code, then the program has a bug.
6222 Dump core so we can fix it. */
6225 msg
= gettext (re_error_msgid
[errcode
]);
6227 msg_size
= strlen (msg
) + 1; /* Includes the null. */
6229 if (errbuf_size
!= 0)
6231 if (msg_size
> errbuf_size
)
6233 strncpy (errbuf
, msg
, errbuf_size
- 1);
6234 errbuf
[errbuf_size
- 1] = 0;
6237 strcpy (errbuf
, msg
);
6244 /* Free dynamically allocated space used by PREG. */
6250 if (preg
->buffer
!= NULL
)
6251 free (preg
->buffer
);
6252 preg
->buffer
= NULL
;
6254 preg
->allocated
= 0;
6257 if (preg
->fastmap
!= NULL
)
6258 free (preg
->fastmap
);
6259 preg
->fastmap
= NULL
;
6260 preg
->fastmap_accurate
= 0;
6262 if (preg
->translate
!= NULL
)
6263 free (preg
->translate
);
6264 preg
->translate
= NULL
;
6267 #endif /* not emacs */