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, 1998 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)
31 /* Converts the pointer to the char to BEG-based offset from the start. */
32 #define PTR_TO_OFFSET(d) \
33 POS_AS_IN_BUFFER (MATCHING_IN_FIRST_STRING \
34 ? (d) - string1 : (d) - (string2 - size1))
35 #define POS_AS_IN_BUFFER(p) ((p) + (NILP (re_match_object) || BUFFERP (re_match_object)))
37 #define PTR_TO_OFFSET(d) 0
44 /* We need this for `regex.h', and perhaps for the Emacs include files. */
45 #include <sys/types.h>
47 /* This is for other GNU distributions with internationalized messages. */
48 #if HAVE_LIBINTL_H || defined (_LIBC)
51 # define gettext(msgid) (msgid)
55 /* This define is so xgettext can find the internationalizable
57 #define gettext_noop(String) String
60 /* The `emacs' switch turns on certain matching commands
61 that make sense only in Emacs. */
67 /* Make syntax table lookup grant data in gl_state. */
68 #define SYNTAX_ENTRY_VIA_PROPERTY
74 #define malloc xmalloc
75 #define realloc xrealloc
80 /* If we are not linking with Emacs proper,
81 we can't use the relocating allocator
82 even if config.h says that we can. */
85 #if defined (STDC_HEADERS) || defined (_LIBC)
92 /* When used in Emacs's lib-src, we need to get bzero and bcopy somehow.
93 If nothing else has been done, use the method below. */
94 #ifdef INHIBIT_STRING_HEADER
95 #if !(defined (HAVE_BZERO) && defined (HAVE_BCOPY))
96 #if !defined (bzero) && !defined (bcopy)
97 #undef INHIBIT_STRING_HEADER
102 /* This is the normal way of making sure we have a bcopy and a bzero.
103 This is used in most programs--a few other programs avoid this
104 by defining INHIBIT_STRING_HEADER. */
105 #ifndef INHIBIT_STRING_HEADER
106 #if defined (HAVE_STRING_H) || defined (STDC_HEADERS) || defined (_LIBC)
109 #define bcmp(s1, s2, n) memcmp ((s1), (s2), (n))
112 #define bcopy(s, d, n) memcpy ((d), (s), (n))
115 #define bzero(s, n) memset ((s), 0, (n))
122 /* Define the syntax stuff for \<, \>, etc. */
124 /* This must be nonzero for the wordchar and notwordchar pattern
125 commands in re_match_2. */
130 #ifdef SWITCH_ENUM_BUG
131 #define SWITCH_ENUM_CAST(x) ((int)(x))
133 #define SWITCH_ENUM_CAST(x) (x)
138 extern char *re_syntax_table
;
140 #else /* not SYNTAX_TABLE */
142 /* How many characters in the character set. */
143 #define CHAR_SET_SIZE 256
145 static char re_syntax_table
[CHAR_SET_SIZE
];
156 bzero (re_syntax_table
, sizeof re_syntax_table
);
158 for (c
= 'a'; c
<= 'z'; c
++)
159 re_syntax_table
[c
] = Sword
;
161 for (c
= 'A'; c
<= 'Z'; c
++)
162 re_syntax_table
[c
] = Sword
;
164 for (c
= '0'; c
<= '9'; c
++)
165 re_syntax_table
[c
] = Sword
;
167 re_syntax_table
['_'] = Sword
;
172 #endif /* not SYNTAX_TABLE */
174 #define SYNTAX(c) re_syntax_table[c]
176 /* Dummy macros for non-Emacs environments. */
177 #define BASE_LEADING_CODE_P(c) (0)
178 #define WORD_BOUNDARY_P(c1, c2) (0)
179 #define CHAR_HEAD_P(p) (1)
180 #define SINGLE_BYTE_CHAR_P(c) (1)
181 #define SAME_CHARSET_P(c1, c2) (1)
182 #define MULTIBYTE_FORM_LENGTH(p, s) (1)
183 #define STRING_CHAR(p, s) (*(p))
184 #define STRING_CHAR_AND_LENGTH(p, s, actual_len) ((actual_len) = 1, *(p))
185 #define GET_CHAR_AFTER_2(c, p, str1, end1, str2, end2) \
186 (c = ((p) == (end1) ? *(str2) : *(p)))
187 #define GET_CHAR_BEFORE_2(c, p, str1, end1, str2, end2) \
188 (c = ((p) == (str2) ? *((end1) - 1) : *((p) - 1)))
189 #endif /* not emacs */
191 /* Get the interface, including the syntax bits. */
194 /* isalpha etc. are used for the character classes. */
197 /* Jim Meyering writes:
199 "... Some ctype macros are valid only for character codes that
200 isascii says are ASCII (SGI's IRIX-4.0.5 is one such system --when
201 using /bin/cc or gcc but without giving an ansi option). So, all
202 ctype uses should be through macros like ISPRINT... If
203 STDC_HEADERS is defined, then autoconf has verified that the ctype
204 macros don't need to be guarded with references to isascii. ...
205 Defining isascii to 1 should let any compiler worth its salt
206 eliminate the && through constant folding." */
208 #if defined (STDC_HEADERS) || (!defined (isascii) && !defined (HAVE_ISASCII))
211 #define ISASCII(c) isascii(c)
215 #define ISBLANK(c) (ISASCII (c) && isblank (c))
217 #define ISBLANK(c) ((c) == ' ' || (c) == '\t')
220 #define ISGRAPH(c) (ISASCII (c) && isgraph (c))
222 #define ISGRAPH(c) (ISASCII (c) && isprint (c) && !isspace (c))
225 #define ISPRINT(c) (ISASCII (c) && isprint (c))
226 #define ISDIGIT(c) (ISASCII (c) && isdigit (c))
227 #define ISALNUM(c) (ISASCII (c) && isalnum (c))
228 #define ISALPHA(c) (ISASCII (c) && isalpha (c))
229 #define ISCNTRL(c) (ISASCII (c) && iscntrl (c))
230 #define ISLOWER(c) (ISASCII (c) && islower (c))
231 #define ISPUNCT(c) (ISASCII (c) && ispunct (c))
232 #define ISSPACE(c) (ISASCII (c) && isspace (c))
233 #define ISUPPER(c) (ISASCII (c) && isupper (c))
234 #define ISXDIGIT(c) (ISASCII (c) && isxdigit (c))
237 #define NULL (void *)0
240 /* We remove any previous definition of `SIGN_EXTEND_CHAR',
241 since ours (we hope) works properly with all combinations of
242 machines, compilers, `char' and `unsigned char' argument types.
243 (Per Bothner suggested the basic approach.) */
244 #undef SIGN_EXTEND_CHAR
246 #define SIGN_EXTEND_CHAR(c) ((signed char) (c))
247 #else /* not __STDC__ */
248 /* As in Harbison and Steele. */
249 #define SIGN_EXTEND_CHAR(c) ((((unsigned char) (c)) ^ 128) - 128)
252 /* Should we use malloc or alloca? If REGEX_MALLOC is not defined, we
253 use `alloca' instead of `malloc'. This is because using malloc in
254 re_search* or re_match* could cause memory leaks when C-g is used in
255 Emacs; also, malloc is slower and causes storage fragmentation. On
256 the other hand, malloc is more portable, and easier to debug.
258 Because we sometimes use alloca, some routines have to be macros,
259 not functions -- `alloca'-allocated space disappears at the end of the
260 function it is called in. */
264 #define REGEX_ALLOCATE malloc
265 #define REGEX_REALLOCATE(source, osize, nsize) realloc (source, nsize)
266 #define REGEX_FREE free
268 #else /* not REGEX_MALLOC */
270 /* Emacs already defines alloca, sometimes. */
273 /* Make alloca work the best possible way. */
275 #define alloca __builtin_alloca
276 #else /* not __GNUC__ */
279 #else /* not __GNUC__ or HAVE_ALLOCA_H */
280 #if 0 /* It is a bad idea to declare alloca. We always cast the result. */
281 #ifndef _AIX /* Already did AIX, up at the top. */
283 #endif /* not _AIX */
285 #endif /* not HAVE_ALLOCA_H */
286 #endif /* not __GNUC__ */
288 #endif /* not alloca */
290 #define REGEX_ALLOCATE alloca
292 /* Assumes a `char *destination' variable. */
293 #define REGEX_REALLOCATE(source, osize, nsize) \
294 (destination = (char *) alloca (nsize), \
295 bcopy (source, destination, osize), \
298 /* No need to do anything to free, after alloca. */
299 #define REGEX_FREE(arg) ((void)0) /* Do nothing! But inhibit gcc warning. */
301 #endif /* not REGEX_MALLOC */
303 /* Define how to allocate the failure stack. */
305 #if defined (REL_ALLOC) && defined (REGEX_MALLOC)
307 #define REGEX_ALLOCATE_STACK(size) \
308 r_alloc (&failure_stack_ptr, (size))
309 #define REGEX_REALLOCATE_STACK(source, osize, nsize) \
310 r_re_alloc (&failure_stack_ptr, (nsize))
311 #define REGEX_FREE_STACK(ptr) \
312 r_alloc_free (&failure_stack_ptr)
314 #else /* not using relocating allocator */
318 #define REGEX_ALLOCATE_STACK malloc
319 #define REGEX_REALLOCATE_STACK(source, osize, nsize) realloc (source, nsize)
320 #define REGEX_FREE_STACK free
322 #else /* not REGEX_MALLOC */
324 #define REGEX_ALLOCATE_STACK alloca
326 #define REGEX_REALLOCATE_STACK(source, osize, nsize) \
327 REGEX_REALLOCATE (source, osize, nsize)
328 /* No need to explicitly free anything. */
329 #define REGEX_FREE_STACK(arg)
331 #endif /* not REGEX_MALLOC */
332 #endif /* not using relocating allocator */
335 /* True if `size1' is non-NULL and PTR is pointing anywhere inside
336 `string1' or just past its end. This works if PTR is NULL, which is
338 #define FIRST_STRING_P(ptr) \
339 (size1 && string1 <= (ptr) && (ptr) <= string1 + size1)
341 /* (Re)Allocate N items of type T using malloc, or fail. */
342 #define TALLOC(n, t) ((t *) malloc ((n) * sizeof (t)))
343 #define RETALLOC(addr, n, t) ((addr) = (t *) realloc (addr, (n) * sizeof (t)))
344 #define RETALLOC_IF(addr, n, t) \
345 if (addr) RETALLOC((addr), (n), t); else (addr) = TALLOC ((n), t)
346 #define REGEX_TALLOC(n, t) ((t *) REGEX_ALLOCATE ((n) * sizeof (t)))
348 #define BYTEWIDTH 8 /* In bits. */
350 #define STREQ(s1, s2) ((strcmp (s1, s2) == 0))
354 #define MAX(a, b) ((a) > (b) ? (a) : (b))
355 #define MIN(a, b) ((a) < (b) ? (a) : (b))
357 typedef char boolean
;
361 static int re_match_2_internal ();
363 /* These are the command codes that appear in compiled regular
364 expressions. Some opcodes are followed by argument bytes. A
365 command code can specify any interpretation whatsoever for its
366 arguments. Zero bytes may appear in the compiled regular expression. */
372 /* Succeed right away--no more backtracking. */
375 /* Followed by one byte giving n, then by n literal bytes. */
378 /* Matches any (more or less) character. */
381 /* Matches any one char belonging to specified set. First
382 following byte is number of bitmap bytes. Then come bytes
383 for a bitmap saying which chars are in. Bits in each byte
384 are ordered low-bit-first. A character is in the set if its
385 bit is 1. A character too large to have a bit in the map is
386 automatically not in the set. */
389 /* Same parameters as charset, but match any character that is
390 not one of those specified. */
393 /* Start remembering the text that is matched, for storing in a
394 register. Followed by one byte with the register number, in
395 the range 0 to one less than the pattern buffer's re_nsub
396 field. Then followed by one byte with the number of groups
397 inner to this one. (This last has to be part of the
398 start_memory only because we need it in the on_failure_jump
402 /* Stop remembering the text that is matched and store it in a
403 memory register. Followed by one byte with the register
404 number, in the range 0 to one less than `re_nsub' in the
405 pattern buffer, and one byte with the number of inner groups,
406 just like `start_memory'. (We need the number of inner
407 groups here because we don't have any easy way of finding the
408 corresponding start_memory when we're at a stop_memory.) */
411 /* Match a duplicate of something remembered. Followed by one
412 byte containing the register number. */
415 /* Fail unless at beginning of line. */
418 /* Fail unless at end of line. */
421 /* Succeeds if at beginning of buffer (if emacs) or at beginning
422 of string to be matched (if not). */
425 /* Analogously, for end of buffer/string. */
428 /* Followed by two byte relative address to which to jump. */
431 /* Same as jump, but marks the end of an alternative. */
434 /* Followed by two-byte relative address of place to resume at
435 in case of failure. */
438 /* Like on_failure_jump, but pushes a placeholder instead of the
439 current string position when executed. */
440 on_failure_keep_string_jump
,
442 /* Throw away latest failure point and then jump to following
443 two-byte relative address. */
446 /* Change to pop_failure_jump if know won't have to backtrack to
447 match; otherwise change to jump. This is used to jump
448 back to the beginning of a repeat. If what follows this jump
449 clearly won't match what the repeat does, such that we can be
450 sure that there is no use backtracking out of repetitions
451 already matched, then we change it to a pop_failure_jump.
452 Followed by two-byte address. */
455 /* Jump to following two-byte address, and push a dummy failure
456 point. This failure point will be thrown away if an attempt
457 is made to use it for a failure. A `+' construct makes this
458 before the first repeat. Also used as an intermediary kind
459 of jump when compiling an alternative. */
462 /* Push a dummy failure point and continue. Used at the end of
466 /* Followed by two-byte relative address and two-byte number n.
467 After matching N times, jump to the address upon failure. */
470 /* Followed by two-byte relative address, and two-byte number n.
471 Jump to the address N times, then fail. */
474 /* Set the following two-byte relative address to the
475 subsequent two-byte number. The address *includes* the two
479 wordchar
, /* Matches any word-constituent character. */
480 notwordchar
, /* Matches any char that is not a word-constituent. */
482 wordbeg
, /* Succeeds if at word beginning. */
483 wordend
, /* Succeeds if at word end. */
485 wordbound
, /* Succeeds if at a word boundary. */
486 notwordbound
/* Succeeds if not at a word boundary. */
489 ,before_dot
, /* Succeeds if before point. */
490 at_dot
, /* Succeeds if at point. */
491 after_dot
, /* Succeeds if after point. */
493 /* Matches any character whose syntax is specified. Followed by
494 a byte which contains a syntax code, e.g., Sword. */
497 /* Matches any character whose syntax is not that specified. */
500 /* Matches any character whose category-set contains the specified
501 category. The operator is followed by a byte which contains a
502 category code (mnemonic ASCII character). */
505 /* Matches any character whose category-set does not contain the
506 specified category. The operator is followed by a byte which
507 contains the category code (mnemonic ASCII character). */
512 /* Common operations on the compiled pattern. */
514 /* Store NUMBER in two contiguous bytes starting at DESTINATION. */
516 #define STORE_NUMBER(destination, number) \
518 (destination)[0] = (number) & 0377; \
519 (destination)[1] = (number) >> 8; \
522 /* Same as STORE_NUMBER, except increment DESTINATION to
523 the byte after where the number is stored. Therefore, DESTINATION
524 must be an lvalue. */
526 #define STORE_NUMBER_AND_INCR(destination, number) \
528 STORE_NUMBER (destination, number); \
529 (destination) += 2; \
532 /* Put into DESTINATION a number stored in two contiguous bytes starting
535 #define EXTRACT_NUMBER(destination, source) \
537 (destination) = *(source) & 0377; \
538 (destination) += SIGN_EXTEND_CHAR (*((source) + 1)) << 8; \
543 extract_number (dest
, source
)
545 unsigned char *source
;
547 int temp
= SIGN_EXTEND_CHAR (*(source
+ 1));
548 *dest
= *source
& 0377;
552 #ifndef EXTRACT_MACROS /* To debug the macros. */
553 #undef EXTRACT_NUMBER
554 #define EXTRACT_NUMBER(dest, src) extract_number (&dest, src)
555 #endif /* not EXTRACT_MACROS */
559 /* Same as EXTRACT_NUMBER, except increment SOURCE to after the number.
560 SOURCE must be an lvalue. */
562 #define EXTRACT_NUMBER_AND_INCR(destination, source) \
564 EXTRACT_NUMBER (destination, source); \
570 extract_number_and_incr (destination
, source
)
572 unsigned char **source
;
574 extract_number (destination
, *source
);
578 #ifndef EXTRACT_MACROS
579 #undef EXTRACT_NUMBER_AND_INCR
580 #define EXTRACT_NUMBER_AND_INCR(dest, src) \
581 extract_number_and_incr (&dest, &src)
582 #endif /* not EXTRACT_MACROS */
586 /* Store a multibyte character in three contiguous bytes starting
587 DESTINATION, and increment DESTINATION to the byte after where the
588 character is stored. Therefore, DESTINATION must be an lvalue. */
590 #define STORE_CHARACTER_AND_INCR(destination, character) \
592 (destination)[0] = (character) & 0377; \
593 (destination)[1] = ((character) >> 8) & 0377; \
594 (destination)[2] = (character) >> 16; \
595 (destination) += 3; \
598 /* Put into DESTINATION a character stored in three contiguous bytes
599 starting at SOURCE. */
601 #define EXTRACT_CHARACTER(destination, source) \
603 (destination) = ((source)[0] \
604 | ((source)[1] << 8) \
605 | ((source)[2] << 16)); \
609 /* Macros for charset. */
611 /* Size of bitmap of charset P in bytes. P is a start of charset,
612 i.e. *P is (re_opcode_t) charset or (re_opcode_t) charset_not. */
613 #define CHARSET_BITMAP_SIZE(p) ((p)[1] & 0x7F)
615 /* Nonzero if charset P has range table. */
616 #define CHARSET_RANGE_TABLE_EXISTS_P(p) ((p)[1] & 0x80)
618 /* Return the address of range table of charset P. But not the start
619 of table itself, but the before where the number of ranges is
620 stored. `2 +' means to skip re_opcode_t and size of bitmap. */
621 #define CHARSET_RANGE_TABLE(p) (&(p)[2 + CHARSET_BITMAP_SIZE (p)])
623 /* Test if C is listed in the bitmap of charset P. */
624 #define CHARSET_LOOKUP_BITMAP(p, c) \
625 ((c) < CHARSET_BITMAP_SIZE (p) * BYTEWIDTH \
626 && (p)[2 + (c) / BYTEWIDTH] & (1 << ((c) % BYTEWIDTH)))
628 /* Return the address of end of RANGE_TABLE. COUNT is number of
629 ranges (which is a pair of (start, end)) in the RANGE_TABLE. `* 2'
630 is start of range and end of range. `* 3' is size of each start
632 #define CHARSET_RANGE_TABLE_END(range_table, count) \
633 ((range_table) + (count) * 2 * 3)
635 /* Test if C is in RANGE_TABLE. A flag NOT is negated if C is in.
636 COUNT is number of ranges in RANGE_TABLE. */
637 #define CHARSET_LOOKUP_RANGE_TABLE_RAW(not, c, range_table, count) \
640 int range_start, range_end; \
642 unsigned char *range_table_end \
643 = CHARSET_RANGE_TABLE_END ((range_table), (count)); \
645 for (p = (range_table); p < range_table_end; p += 2 * 3) \
647 EXTRACT_CHARACTER (range_start, p); \
648 EXTRACT_CHARACTER (range_end, p + 3); \
650 if (range_start <= (c) && (c) <= range_end) \
659 /* Test if C is in range table of CHARSET. The flag NOT is negated if
660 C is listed in it. */
661 #define CHARSET_LOOKUP_RANGE_TABLE(not, c, charset) \
664 /* Number of ranges in range table. */ \
666 unsigned char *range_table = CHARSET_RANGE_TABLE (charset); \
668 EXTRACT_NUMBER_AND_INCR (count, range_table); \
669 CHARSET_LOOKUP_RANGE_TABLE_RAW ((not), (c), range_table, count); \
673 /* If DEBUG is defined, Regex prints many voluminous messages about what
674 it is doing (if the variable `debug' is nonzero). If linked with the
675 main program in `iregex.c', you can enter patterns and strings
676 interactively. And if linked with the main program in `main.c' and
677 the other test files, you can run the already-written tests. */
681 /* We use standard I/O for debugging. */
684 /* It is useful to test things that ``must'' be true when debugging. */
687 static int debug
= 0;
689 #define DEBUG_STATEMENT(e) e
690 #define DEBUG_PRINT1(x) if (debug) printf (x)
691 #define DEBUG_PRINT2(x1, x2) if (debug) printf (x1, x2)
692 #define DEBUG_PRINT3(x1, x2, x3) if (debug) printf (x1, x2, x3)
693 #define DEBUG_PRINT4(x1, x2, x3, x4) if (debug) printf (x1, x2, x3, x4)
694 #define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) \
695 if (debug) print_partial_compiled_pattern (s, e)
696 #define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) \
697 if (debug) print_double_string (w, s1, sz1, s2, sz2)
700 /* Print the fastmap in human-readable form. */
703 print_fastmap (fastmap
)
706 unsigned was_a_range
= 0;
709 while (i
< (1 << BYTEWIDTH
))
715 while (i
< (1 << BYTEWIDTH
) && fastmap
[i
])
731 /* Print a compiled pattern string in human-readable form, starting at
732 the START pointer into it and ending just before the pointer END. */
735 print_partial_compiled_pattern (start
, end
)
736 unsigned char *start
;
740 unsigned char *p
= start
;
741 unsigned char *pend
= end
;
749 /* Loop over pattern commands. */
752 printf ("%d:\t", p
- start
);
754 switch ((re_opcode_t
) *p
++)
762 printf ("/exactn/%d", mcnt
);
773 printf ("/start_memory/%d/%d", mcnt
, *p
++);
778 printf ("/stop_memory/%d/%d", mcnt
, *p
++);
782 printf ("/duplicate/%d", *p
++);
792 register int c
, last
= -100;
793 register int in_range
= 0;
795 printf ("/charset [%s",
796 (re_opcode_t
) *(p
- 1) == charset_not
? "^" : "");
798 assert (p
+ *p
< pend
);
800 for (c
= 0; c
< 256; c
++)
802 && (p
[1 + (c
/8)] & (1 << (c
% 8))))
804 /* Are we starting a range? */
805 if (last
+ 1 == c
&& ! in_range
)
810 /* Have we broken a range? */
811 else if (last
+ 1 != c
&& in_range
)
840 case on_failure_jump
:
841 extract_number_and_incr (&mcnt
, &p
);
842 printf ("/on_failure_jump to %d", p
+ mcnt
- start
);
845 case on_failure_keep_string_jump
:
846 extract_number_and_incr (&mcnt
, &p
);
847 printf ("/on_failure_keep_string_jump to %d", p
+ mcnt
- start
);
850 case dummy_failure_jump
:
851 extract_number_and_incr (&mcnt
, &p
);
852 printf ("/dummy_failure_jump to %d", p
+ mcnt
- start
);
855 case push_dummy_failure
:
856 printf ("/push_dummy_failure");
860 extract_number_and_incr (&mcnt
, &p
);
861 printf ("/maybe_pop_jump to %d", p
+ mcnt
- start
);
864 case pop_failure_jump
:
865 extract_number_and_incr (&mcnt
, &p
);
866 printf ("/pop_failure_jump to %d", p
+ mcnt
- start
);
870 extract_number_and_incr (&mcnt
, &p
);
871 printf ("/jump_past_alt to %d", p
+ mcnt
- start
);
875 extract_number_and_incr (&mcnt
, &p
);
876 printf ("/jump to %d", p
+ mcnt
- start
);
880 extract_number_and_incr (&mcnt
, &p
);
881 extract_number_and_incr (&mcnt2
, &p
);
882 printf ("/succeed_n to %d, %d times", p
+ mcnt
- start
, mcnt2
);
886 extract_number_and_incr (&mcnt
, &p
);
887 extract_number_and_incr (&mcnt2
, &p
);
888 printf ("/jump_n to %d, %d times", p
+ mcnt
- start
, mcnt2
);
892 extract_number_and_incr (&mcnt
, &p
);
893 extract_number_and_incr (&mcnt2
, &p
);
894 printf ("/set_number_at location %d to %d", p
+ mcnt
- start
, mcnt2
);
898 printf ("/wordbound");
902 printf ("/notwordbound");
914 printf ("/before_dot");
922 printf ("/after_dot");
926 printf ("/syntaxspec");
928 printf ("/%d", mcnt
);
932 printf ("/notsyntaxspec");
934 printf ("/%d", mcnt
);
939 printf ("/wordchar");
943 printf ("/notwordchar");
955 printf ("?%d", *(p
-1));
961 printf ("%d:\tend of pattern.\n", p
- start
);
966 print_compiled_pattern (bufp
)
967 struct re_pattern_buffer
*bufp
;
969 unsigned char *buffer
= bufp
->buffer
;
971 print_partial_compiled_pattern (buffer
, buffer
+ bufp
->used
);
972 printf ("%d bytes used/%d bytes allocated.\n", bufp
->used
, bufp
->allocated
);
974 if (bufp
->fastmap_accurate
&& bufp
->fastmap
)
976 printf ("fastmap: ");
977 print_fastmap (bufp
->fastmap
);
980 printf ("re_nsub: %d\t", bufp
->re_nsub
);
981 printf ("regs_alloc: %d\t", bufp
->regs_allocated
);
982 printf ("can_be_null: %d\t", bufp
->can_be_null
);
983 printf ("newline_anchor: %d\n", bufp
->newline_anchor
);
984 printf ("no_sub: %d\t", bufp
->no_sub
);
985 printf ("not_bol: %d\t", bufp
->not_bol
);
986 printf ("not_eol: %d\t", bufp
->not_eol
);
987 printf ("syntax: %d\n", bufp
->syntax
);
988 /* Perhaps we should print the translate table? */
993 print_double_string (where
, string1
, size1
, string2
, size2
)
1006 if (FIRST_STRING_P (where
))
1008 for (this_char
= where
- string1
; this_char
< size1
; this_char
++)
1009 putchar (string1
[this_char
]);
1014 for (this_char
= where
- string2
; this_char
< size2
; this_char
++)
1015 putchar (string2
[this_char
]);
1019 #else /* not DEBUG */
1024 #define DEBUG_STATEMENT(e)
1025 #define DEBUG_PRINT1(x)
1026 #define DEBUG_PRINT2(x1, x2)
1027 #define DEBUG_PRINT3(x1, x2, x3)
1028 #define DEBUG_PRINT4(x1, x2, x3, x4)
1029 #define DEBUG_PRINT_COMPILED_PATTERN(p, s, e)
1030 #define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2)
1032 #endif /* not DEBUG */
1034 /* Set by `re_set_syntax' to the current regexp syntax to recognize. Can
1035 also be assigned to arbitrarily: each pattern buffer stores its own
1036 syntax, so it can be changed between regex compilations. */
1037 /* This has no initializer because initialized variables in Emacs
1038 become read-only after dumping. */
1039 reg_syntax_t re_syntax_options
;
1042 /* Specify the precise syntax of regexps for compilation. This provides
1043 for compatibility for various utilities which historically have
1044 different, incompatible syntaxes.
1046 The argument SYNTAX is a bit mask comprised of the various bits
1047 defined in regex.h. We return the old syntax. */
1050 re_set_syntax (syntax
)
1051 reg_syntax_t syntax
;
1053 reg_syntax_t ret
= re_syntax_options
;
1055 re_syntax_options
= syntax
;
1059 /* This table gives an error message for each of the error codes listed
1060 in regex.h. Obviously the order here has to be same as there.
1061 POSIX doesn't require that we do anything for REG_NOERROR,
1062 but why not be nice? */
1064 static const char *re_error_msgid
[] =
1066 gettext_noop ("Success"), /* REG_NOERROR */
1067 gettext_noop ("No match"), /* REG_NOMATCH */
1068 gettext_noop ("Invalid regular expression"), /* REG_BADPAT */
1069 gettext_noop ("Invalid collation character"), /* REG_ECOLLATE */
1070 gettext_noop ("Invalid character class name"), /* REG_ECTYPE */
1071 gettext_noop ("Trailing backslash"), /* REG_EESCAPE */
1072 gettext_noop ("Invalid back reference"), /* REG_ESUBREG */
1073 gettext_noop ("Unmatched [ or [^"), /* REG_EBRACK */
1074 gettext_noop ("Unmatched ( or \\("), /* REG_EPAREN */
1075 gettext_noop ("Unmatched \\{"), /* REG_EBRACE */
1076 gettext_noop ("Invalid content of \\{\\}"), /* REG_BADBR */
1077 gettext_noop ("Invalid range end"), /* REG_ERANGE */
1078 gettext_noop ("Memory exhausted"), /* REG_ESPACE */
1079 gettext_noop ("Invalid preceding regular expression"), /* REG_BADRPT */
1080 gettext_noop ("Premature end of regular expression"), /* REG_EEND */
1081 gettext_noop ("Regular expression too big"), /* REG_ESIZE */
1082 gettext_noop ("Unmatched ) or \\)"), /* REG_ERPAREN */
1085 /* Avoiding alloca during matching, to placate r_alloc. */
1087 /* Define MATCH_MAY_ALLOCATE unless we need to make sure that the
1088 searching and matching functions should not call alloca. On some
1089 systems, alloca is implemented in terms of malloc, and if we're
1090 using the relocating allocator routines, then malloc could cause a
1091 relocation, which might (if the strings being searched are in the
1092 ralloc heap) shift the data out from underneath the regexp
1095 Here's another reason to avoid allocation: Emacs
1096 processes input from X in a signal handler; processing X input may
1097 call malloc; if input arrives while a matching routine is calling
1098 malloc, then we're scrod. But Emacs can't just block input while
1099 calling matching routines; then we don't notice interrupts when
1100 they come in. So, Emacs blocks input around all regexp calls
1101 except the matching calls, which it leaves unprotected, in the
1102 faith that they will not malloc. */
1104 /* Normally, this is fine. */
1105 #define MATCH_MAY_ALLOCATE
1107 /* When using GNU C, we are not REALLY using the C alloca, no matter
1108 what config.h may say. So don't take precautions for it. */
1113 /* The match routines may not allocate if (1) they would do it with malloc
1114 and (2) it's not safe for them to use malloc.
1115 Note that if REL_ALLOC is defined, matching would not use malloc for the
1116 failure stack, but we would still use it for the register vectors;
1117 so REL_ALLOC should not affect this. */
1118 #if (defined (C_ALLOCA) || defined (REGEX_MALLOC)) && defined (emacs)
1119 #undef MATCH_MAY_ALLOCATE
1123 /* Failure stack declarations and macros; both re_compile_fastmap and
1124 re_match_2 use a failure stack. These have to be macros because of
1125 REGEX_ALLOCATE_STACK. */
1128 /* Approximate number of failure points for which to initially allocate space
1129 when matching. If this number is exceeded, we allocate more
1130 space, so it is not a hard limit. */
1131 #ifndef INIT_FAILURE_ALLOC
1132 #define INIT_FAILURE_ALLOC 20
1135 /* Roughly the maximum number of failure points on the stack. Would be
1136 exactly that if always used TYPICAL_FAILURE_SIZE items each time we failed.
1137 This is a variable only so users of regex can assign to it; we never
1138 change it ourselves. */
1139 #if defined (MATCH_MAY_ALLOCATE)
1140 /* Note that 4400 is enough to cause a crash on Alpha OSF/1,
1141 whose default stack limit is 2mb. In order for a larger
1142 value to work reliably, you have to try to make it accord
1143 with the process stack limit. */
1144 int re_max_failures
= 40000;
1146 int re_max_failures
= 4000;
1149 union fail_stack_elt
1151 unsigned char *pointer
;
1155 typedef union fail_stack_elt fail_stack_elt_t
;
1159 fail_stack_elt_t
*stack
;
1161 unsigned avail
; /* Offset of next open position. */
1164 #define FAIL_STACK_EMPTY() (fail_stack.avail == 0)
1165 #define FAIL_STACK_PTR_EMPTY() (fail_stack_ptr->avail == 0)
1166 #define FAIL_STACK_FULL() (fail_stack.avail == fail_stack.size)
1169 /* Define macros to initialize and free the failure stack.
1170 Do `return -2' if the alloc fails. */
1172 #ifdef MATCH_MAY_ALLOCATE
1173 #define INIT_FAIL_STACK() \
1175 fail_stack.stack = (fail_stack_elt_t *) \
1176 REGEX_ALLOCATE_STACK (INIT_FAILURE_ALLOC * TYPICAL_FAILURE_SIZE \
1177 * sizeof (fail_stack_elt_t)); \
1179 if (fail_stack.stack == NULL) \
1182 fail_stack.size = INIT_FAILURE_ALLOC; \
1183 fail_stack.avail = 0; \
1186 #define RESET_FAIL_STACK() REGEX_FREE_STACK (fail_stack.stack)
1188 #define INIT_FAIL_STACK() \
1190 fail_stack.avail = 0; \
1193 #define RESET_FAIL_STACK()
1197 /* Double the size of FAIL_STACK, up to a limit
1198 which allows approximately `re_max_failures' items.
1200 Return 1 if succeeds, and 0 if either ran out of memory
1201 allocating space for it or it was already too large.
1203 REGEX_REALLOCATE_STACK requires `destination' be declared. */
1205 /* Factor to increase the failure stack size by
1206 when we increase it.
1207 This used to be 2, but 2 was too wasteful
1208 because the old discarded stacks added up to as much space
1209 were as ultimate, maximum-size stack. */
1210 #define FAIL_STACK_GROWTH_FACTOR 4
1212 #define GROW_FAIL_STACK(fail_stack) \
1213 (((fail_stack).size * sizeof (fail_stack_elt_t) \
1214 >= re_max_failures * TYPICAL_FAILURE_SIZE) \
1216 : ((fail_stack).stack \
1217 = (fail_stack_elt_t *) \
1218 REGEX_REALLOCATE_STACK ((fail_stack).stack, \
1219 (fail_stack).size * sizeof (fail_stack_elt_t), \
1220 MIN (re_max_failures * TYPICAL_FAILURE_SIZE, \
1221 ((fail_stack).size * sizeof (fail_stack_elt_t) \
1222 * FAIL_STACK_GROWTH_FACTOR))), \
1224 (fail_stack).stack == NULL \
1226 : ((fail_stack).size \
1227 = (MIN (re_max_failures * TYPICAL_FAILURE_SIZE, \
1228 ((fail_stack).size * sizeof (fail_stack_elt_t) \
1229 * FAIL_STACK_GROWTH_FACTOR)) \
1230 / sizeof (fail_stack_elt_t)), \
1234 /* Push pointer POINTER on FAIL_STACK.
1235 Return 1 if was able to do so and 0 if ran out of memory allocating
1237 #define PUSH_PATTERN_OP(POINTER, FAIL_STACK) \
1238 ((FAIL_STACK_FULL () \
1239 && !GROW_FAIL_STACK (FAIL_STACK)) \
1241 : ((FAIL_STACK).stack[(FAIL_STACK).avail++].pointer = POINTER, \
1244 /* Push a pointer value onto the failure stack.
1245 Assumes the variable `fail_stack'. Probably should only
1246 be called from within `PUSH_FAILURE_POINT'. */
1247 #define PUSH_FAILURE_POINTER(item) \
1248 fail_stack.stack[fail_stack.avail++].pointer = (unsigned char *) (item)
1250 /* This pushes an integer-valued item onto the failure stack.
1251 Assumes the variable `fail_stack'. Probably should only
1252 be called from within `PUSH_FAILURE_POINT'. */
1253 #define PUSH_FAILURE_INT(item) \
1254 fail_stack.stack[fail_stack.avail++].integer = (item)
1256 /* Push a fail_stack_elt_t value onto the failure stack.
1257 Assumes the variable `fail_stack'. Probably should only
1258 be called from within `PUSH_FAILURE_POINT'. */
1259 #define PUSH_FAILURE_ELT(item) \
1260 fail_stack.stack[fail_stack.avail++] = (item)
1262 /* These three POP... operations complement the three PUSH... operations.
1263 All assume that `fail_stack' is nonempty. */
1264 #define POP_FAILURE_POINTER() fail_stack.stack[--fail_stack.avail].pointer
1265 #define POP_FAILURE_INT() fail_stack.stack[--fail_stack.avail].integer
1266 #define POP_FAILURE_ELT() fail_stack.stack[--fail_stack.avail]
1268 /* Used to omit pushing failure point id's when we're not debugging. */
1270 #define DEBUG_PUSH PUSH_FAILURE_INT
1271 #define DEBUG_POP(item_addr) *(item_addr) = POP_FAILURE_INT ()
1273 #define DEBUG_PUSH(item)
1274 #define DEBUG_POP(item_addr)
1278 /* Push the information about the state we will need
1279 if we ever fail back to it.
1281 Requires variables fail_stack, regstart, regend, reg_info, and
1282 num_regs be declared. GROW_FAIL_STACK requires `destination' be
1285 Does `return FAILURE_CODE' if runs out of memory. */
1287 #define PUSH_FAILURE_POINT(pattern_place, string_place, failure_code) \
1289 char *destination; \
1290 /* Must be int, so when we don't save any registers, the arithmetic \
1291 of 0 + -1 isn't done as unsigned. */ \
1294 DEBUG_STATEMENT (failure_id++); \
1295 DEBUG_STATEMENT (nfailure_points_pushed++); \
1296 DEBUG_PRINT2 ("\nPUSH_FAILURE_POINT #%u:\n", failure_id); \
1297 DEBUG_PRINT2 (" Before push, next avail: %d\n", (fail_stack).avail);\
1298 DEBUG_PRINT2 (" size: %d\n", (fail_stack).size);\
1300 DEBUG_PRINT2 (" slots needed: %d\n", NUM_FAILURE_ITEMS); \
1301 DEBUG_PRINT2 (" available: %d\n", REMAINING_AVAIL_SLOTS); \
1303 /* Ensure we have enough space allocated for what we will push. */ \
1304 while (REMAINING_AVAIL_SLOTS < NUM_FAILURE_ITEMS) \
1306 if (!GROW_FAIL_STACK (fail_stack)) \
1307 return failure_code; \
1309 DEBUG_PRINT2 ("\n Doubled stack; size now: %d\n", \
1310 (fail_stack).size); \
1311 DEBUG_PRINT2 (" slots available: %d\n", REMAINING_AVAIL_SLOTS);\
1314 /* Push the info, starting with the registers. */ \
1315 DEBUG_PRINT1 ("\n"); \
1318 for (this_reg = lowest_active_reg; this_reg <= highest_active_reg; \
1321 DEBUG_PRINT2 (" Pushing reg: %d\n", this_reg); \
1322 DEBUG_STATEMENT (num_regs_pushed++); \
1324 DEBUG_PRINT2 (" start: 0x%x\n", regstart[this_reg]); \
1325 PUSH_FAILURE_POINTER (regstart[this_reg]); \
1327 DEBUG_PRINT2 (" end: 0x%x\n", regend[this_reg]); \
1328 PUSH_FAILURE_POINTER (regend[this_reg]); \
1330 DEBUG_PRINT2 (" info: 0x%x\n ", reg_info[this_reg]); \
1331 DEBUG_PRINT2 (" match_null=%d", \
1332 REG_MATCH_NULL_STRING_P (reg_info[this_reg])); \
1333 DEBUG_PRINT2 (" active=%d", IS_ACTIVE (reg_info[this_reg])); \
1334 DEBUG_PRINT2 (" matched_something=%d", \
1335 MATCHED_SOMETHING (reg_info[this_reg])); \
1336 DEBUG_PRINT2 (" ever_matched=%d", \
1337 EVER_MATCHED_SOMETHING (reg_info[this_reg])); \
1338 DEBUG_PRINT1 ("\n"); \
1339 PUSH_FAILURE_ELT (reg_info[this_reg].word); \
1342 DEBUG_PRINT2 (" Pushing low active reg: %d\n", lowest_active_reg);\
1343 PUSH_FAILURE_INT (lowest_active_reg); \
1345 DEBUG_PRINT2 (" Pushing high active reg: %d\n", highest_active_reg);\
1346 PUSH_FAILURE_INT (highest_active_reg); \
1348 DEBUG_PRINT2 (" Pushing pattern 0x%x: ", pattern_place); \
1349 DEBUG_PRINT_COMPILED_PATTERN (bufp, pattern_place, pend); \
1350 PUSH_FAILURE_POINTER (pattern_place); \
1352 DEBUG_PRINT2 (" Pushing string 0x%x: `", string_place); \
1353 DEBUG_PRINT_DOUBLE_STRING (string_place, string1, size1, string2, \
1355 DEBUG_PRINT1 ("'\n"); \
1356 PUSH_FAILURE_POINTER (string_place); \
1358 DEBUG_PRINT2 (" Pushing failure id: %u\n", failure_id); \
1359 DEBUG_PUSH (failure_id); \
1362 /* This is the number of items that are pushed and popped on the stack
1363 for each register. */
1364 #define NUM_REG_ITEMS 3
1366 /* Individual items aside from the registers. */
1368 #define NUM_NONREG_ITEMS 5 /* Includes failure point id. */
1370 #define NUM_NONREG_ITEMS 4
1373 /* Estimate the size of data pushed by a typical failure stack entry.
1374 An estimate is all we need, because all we use this for
1375 is to choose a limit for how big to make the failure stack. */
1377 #define TYPICAL_FAILURE_SIZE 20
1379 /* This is how many items we actually use for a failure point.
1380 It depends on the regexp. */
1381 #define NUM_FAILURE_ITEMS \
1383 ? 0 : highest_active_reg - lowest_active_reg + 1) \
1387 /* How many items can still be added to the stack without overflowing it. */
1388 #define REMAINING_AVAIL_SLOTS ((fail_stack).size - (fail_stack).avail)
1391 /* Pops what PUSH_FAIL_STACK pushes.
1393 We restore into the parameters, all of which should be lvalues:
1394 STR -- the saved data position.
1395 PAT -- the saved pattern position.
1396 LOW_REG, HIGH_REG -- the highest and lowest active registers.
1397 REGSTART, REGEND -- arrays of string positions.
1398 REG_INFO -- array of information about each subexpression.
1400 Also assumes the variables `fail_stack' and (if debugging), `bufp',
1401 `pend', `string1', `size1', `string2', and `size2'. */
1403 #define POP_FAILURE_POINT(str, pat, low_reg, high_reg, regstart, regend, reg_info)\
1405 DEBUG_STATEMENT (fail_stack_elt_t failure_id;) \
1407 const unsigned char *string_temp; \
1409 assert (!FAIL_STACK_EMPTY ()); \
1411 /* Remove failure points and point to how many regs pushed. */ \
1412 DEBUG_PRINT1 ("POP_FAILURE_POINT:\n"); \
1413 DEBUG_PRINT2 (" Before pop, next avail: %d\n", fail_stack.avail); \
1414 DEBUG_PRINT2 (" size: %d\n", fail_stack.size); \
1416 assert (fail_stack.avail >= NUM_NONREG_ITEMS); \
1418 DEBUG_POP (&failure_id); \
1419 DEBUG_PRINT2 (" Popping failure id: %u\n", failure_id); \
1421 /* If the saved string location is NULL, it came from an \
1422 on_failure_keep_string_jump opcode, and we want to throw away the \
1423 saved NULL, thus retaining our current position in the string. */ \
1424 string_temp = POP_FAILURE_POINTER (); \
1425 if (string_temp != NULL) \
1426 str = (const char *) string_temp; \
1428 DEBUG_PRINT2 (" Popping string 0x%x: `", str); \
1429 DEBUG_PRINT_DOUBLE_STRING (str, string1, size1, string2, size2); \
1430 DEBUG_PRINT1 ("'\n"); \
1432 pat = (unsigned char *) POP_FAILURE_POINTER (); \
1433 DEBUG_PRINT2 (" Popping pattern 0x%x: ", pat); \
1434 DEBUG_PRINT_COMPILED_PATTERN (bufp, pat, pend); \
1436 /* Restore register info. */ \
1437 high_reg = (unsigned) POP_FAILURE_INT (); \
1438 DEBUG_PRINT2 (" Popping high active reg: %d\n", high_reg); \
1440 low_reg = (unsigned) POP_FAILURE_INT (); \
1441 DEBUG_PRINT2 (" Popping low active reg: %d\n", low_reg); \
1444 for (this_reg = high_reg; this_reg >= low_reg; this_reg--) \
1446 DEBUG_PRINT2 (" Popping reg: %d\n", this_reg); \
1448 reg_info[this_reg].word = POP_FAILURE_ELT (); \
1449 DEBUG_PRINT2 (" info: 0x%x\n", reg_info[this_reg]); \
1451 regend[this_reg] = (const char *) POP_FAILURE_POINTER (); \
1452 DEBUG_PRINT2 (" end: 0x%x\n", regend[this_reg]); \
1454 regstart[this_reg] = (const char *) POP_FAILURE_POINTER (); \
1455 DEBUG_PRINT2 (" start: 0x%x\n", regstart[this_reg]); \
1459 for (this_reg = highest_active_reg; this_reg > high_reg; this_reg--) \
1461 reg_info[this_reg].word.integer = 0; \
1462 regend[this_reg] = 0; \
1463 regstart[this_reg] = 0; \
1465 highest_active_reg = high_reg; \
1468 set_regs_matched_done = 0; \
1469 DEBUG_STATEMENT (nfailure_points_popped++); \
1470 } /* POP_FAILURE_POINT */
1474 /* Structure for per-register (a.k.a. per-group) information.
1475 Other register information, such as the
1476 starting and ending positions (which are addresses), and the list of
1477 inner groups (which is a bits list) are maintained in separate
1480 We are making a (strictly speaking) nonportable assumption here: that
1481 the compiler will pack our bit fields into something that fits into
1482 the type of `word', i.e., is something that fits into one item on the
1487 fail_stack_elt_t word
;
1490 /* This field is one if this group can match the empty string,
1491 zero if not. If not yet determined, `MATCH_NULL_UNSET_VALUE'. */
1492 #define MATCH_NULL_UNSET_VALUE 3
1493 unsigned match_null_string_p
: 2;
1494 unsigned is_active
: 1;
1495 unsigned matched_something
: 1;
1496 unsigned ever_matched_something
: 1;
1498 } register_info_type
;
1500 #define REG_MATCH_NULL_STRING_P(R) ((R).bits.match_null_string_p)
1501 #define IS_ACTIVE(R) ((R).bits.is_active)
1502 #define MATCHED_SOMETHING(R) ((R).bits.matched_something)
1503 #define EVER_MATCHED_SOMETHING(R) ((R).bits.ever_matched_something)
1506 /* Call this when have matched a real character; it sets `matched' flags
1507 for the subexpressions which we are currently inside. Also records
1508 that those subexprs have matched. */
1509 #define SET_REGS_MATCHED() \
1512 if (!set_regs_matched_done) \
1515 set_regs_matched_done = 1; \
1516 for (r = lowest_active_reg; r <= highest_active_reg; r++) \
1518 MATCHED_SOMETHING (reg_info[r]) \
1519 = EVER_MATCHED_SOMETHING (reg_info[r]) \
1526 /* Registers are set to a sentinel when they haven't yet matched. */
1527 static char reg_unset_dummy
;
1528 #define REG_UNSET_VALUE (®_unset_dummy)
1529 #define REG_UNSET(e) ((e) == REG_UNSET_VALUE)
1531 /* Subroutine declarations and macros for regex_compile. */
1533 static void store_op1 (), store_op2 ();
1534 static void insert_op1 (), insert_op2 ();
1535 static boolean
at_begline_loc_p (), at_endline_loc_p ();
1536 static boolean
group_in_compile_stack ();
1537 static reg_errcode_t
compile_range ();
1539 /* Fetch the next character in the uncompiled pattern---translating it
1540 if necessary. Also cast from a signed character in the constant
1541 string passed to us by the user to an unsigned char that we can use
1542 as an array index (in, e.g., `translate'). */
1544 #define PATFETCH(c) \
1545 do {if (p == pend) return REG_EEND; \
1546 c = (unsigned char) *p++; \
1547 if (RE_TRANSLATE_P (translate)) c = RE_TRANSLATE (translate, c); \
1551 /* Fetch the next character in the uncompiled pattern, with no
1553 #define PATFETCH_RAW(c) \
1554 do {if (p == pend) return REG_EEND; \
1555 c = (unsigned char) *p++; \
1558 /* Go backwards one character in the pattern. */
1559 #define PATUNFETCH p--
1562 /* If `translate' is non-null, return translate[D], else just D. We
1563 cast the subscript to translate because some data is declared as
1564 `char *', to avoid warnings when a string constant is passed. But
1565 when we use a character as a subscript we must make it unsigned. */
1567 #define TRANSLATE(d) \
1568 (RE_TRANSLATE_P (translate) \
1569 ? (unsigned) RE_TRANSLATE (translate, (unsigned) (d)) : (d))
1573 /* Macros for outputting the compiled pattern into `buffer'. */
1575 /* If the buffer isn't allocated when it comes in, use this. */
1576 #define INIT_BUF_SIZE 32
1578 /* Make sure we have at least N more bytes of space in buffer. */
1579 #define GET_BUFFER_SPACE(n) \
1580 while (b - bufp->buffer + (n) > bufp->allocated) \
1583 /* Make sure we have one more byte of buffer space and then add C to it. */
1584 #define BUF_PUSH(c) \
1586 GET_BUFFER_SPACE (1); \
1587 *b++ = (unsigned char) (c); \
1591 /* Ensure we have two more bytes of buffer space and then append C1 and C2. */
1592 #define BUF_PUSH_2(c1, c2) \
1594 GET_BUFFER_SPACE (2); \
1595 *b++ = (unsigned char) (c1); \
1596 *b++ = (unsigned char) (c2); \
1600 /* As with BUF_PUSH_2, except for three bytes. */
1601 #define BUF_PUSH_3(c1, c2, c3) \
1603 GET_BUFFER_SPACE (3); \
1604 *b++ = (unsigned char) (c1); \
1605 *b++ = (unsigned char) (c2); \
1606 *b++ = (unsigned char) (c3); \
1610 /* Store a jump with opcode OP at LOC to location TO. We store a
1611 relative address offset by the three bytes the jump itself occupies. */
1612 #define STORE_JUMP(op, loc, to) \
1613 store_op1 (op, loc, (to) - (loc) - 3)
1615 /* Likewise, for a two-argument jump. */
1616 #define STORE_JUMP2(op, loc, to, arg) \
1617 store_op2 (op, loc, (to) - (loc) - 3, arg)
1619 /* Like `STORE_JUMP', but for inserting. Assume `b' is the buffer end. */
1620 #define INSERT_JUMP(op, loc, to) \
1621 insert_op1 (op, loc, (to) - (loc) - 3, b)
1623 /* Like `STORE_JUMP2', but for inserting. Assume `b' is the buffer end. */
1624 #define INSERT_JUMP2(op, loc, to, arg) \
1625 insert_op2 (op, loc, (to) - (loc) - 3, arg, b)
1628 /* This is not an arbitrary limit: the arguments which represent offsets
1629 into the pattern are two bytes long. So if 2^16 bytes turns out to
1630 be too small, many things would have to change. */
1631 #define MAX_BUF_SIZE (1L << 16)
1634 /* Extend the buffer by twice its current size via realloc and
1635 reset the pointers that pointed into the old block to point to the
1636 correct places in the new one. If extending the buffer results in it
1637 being larger than MAX_BUF_SIZE, then flag memory exhausted. */
1638 #define EXTEND_BUFFER() \
1640 unsigned char *old_buffer = bufp->buffer; \
1641 if (bufp->allocated == MAX_BUF_SIZE) \
1643 bufp->allocated <<= 1; \
1644 if (bufp->allocated > MAX_BUF_SIZE) \
1645 bufp->allocated = MAX_BUF_SIZE; \
1646 bufp->buffer = (unsigned char *) realloc (bufp->buffer, bufp->allocated);\
1647 if (bufp->buffer == NULL) \
1648 return REG_ESPACE; \
1649 /* If the buffer moved, move all the pointers into it. */ \
1650 if (old_buffer != bufp->buffer) \
1652 b = (b - old_buffer) + bufp->buffer; \
1653 begalt = (begalt - old_buffer) + bufp->buffer; \
1654 if (fixup_alt_jump) \
1655 fixup_alt_jump = (fixup_alt_jump - old_buffer) + bufp->buffer;\
1657 laststart = (laststart - old_buffer) + bufp->buffer; \
1658 if (pending_exact) \
1659 pending_exact = (pending_exact - old_buffer) + bufp->buffer; \
1664 /* Since we have one byte reserved for the register number argument to
1665 {start,stop}_memory, the maximum number of groups we can report
1666 things about is what fits in that byte. */
1667 #define MAX_REGNUM 255
1669 /* But patterns can have more than `MAX_REGNUM' registers. We just
1670 ignore the excess. */
1671 typedef unsigned regnum_t
;
1674 /* Macros for the compile stack. */
1676 /* Since offsets can go either forwards or backwards, this type needs to
1677 be able to hold values from -(MAX_BUF_SIZE - 1) to MAX_BUF_SIZE - 1. */
1678 typedef int pattern_offset_t
;
1682 pattern_offset_t begalt_offset
;
1683 pattern_offset_t fixup_alt_jump
;
1684 pattern_offset_t inner_group_offset
;
1685 pattern_offset_t laststart_offset
;
1687 } compile_stack_elt_t
;
1692 compile_stack_elt_t
*stack
;
1694 unsigned avail
; /* Offset of next open position. */
1695 } compile_stack_type
;
1698 #define INIT_COMPILE_STACK_SIZE 32
1700 #define COMPILE_STACK_EMPTY (compile_stack.avail == 0)
1701 #define COMPILE_STACK_FULL (compile_stack.avail == compile_stack.size)
1703 /* The next available element. */
1704 #define COMPILE_STACK_TOP (compile_stack.stack[compile_stack.avail])
1707 /* Structure to manage work area for range table. */
1708 struct range_table_work_area
1710 int *table
; /* actual work area. */
1711 int allocated
; /* allocated size for work area in bytes. */
1712 int used
; /* actually used size in words. */
1715 /* Make sure that WORK_AREA can hold more N multibyte characters. */
1716 #define EXTEND_RANGE_TABLE_WORK_AREA(work_area, n) \
1718 if (((work_area).used + (n)) * sizeof (int) > (work_area).allocated) \
1720 (work_area).allocated += 16 * sizeof (int); \
1721 if ((work_area).table) \
1723 = (int *) realloc ((work_area).table, (work_area).allocated); \
1726 = (int *) malloc ((work_area).allocated); \
1727 if ((work_area).table == 0) \
1728 FREE_STACK_RETURN (REG_ESPACE); \
1732 /* Set a range (RANGE_START, RANGE_END) to WORK_AREA. */
1733 #define SET_RANGE_TABLE_WORK_AREA(work_area, range_start, range_end) \
1735 EXTEND_RANGE_TABLE_WORK_AREA ((work_area), 2); \
1736 (work_area).table[(work_area).used++] = (range_start); \
1737 (work_area).table[(work_area).used++] = (range_end); \
1740 /* Free allocated memory for WORK_AREA. */
1741 #define FREE_RANGE_TABLE_WORK_AREA(work_area) \
1743 if ((work_area).table) \
1744 free ((work_area).table); \
1747 #define CLEAR_RANGE_TABLE_WORK_USED(work_area) ((work_area).used = 0)
1748 #define RANGE_TABLE_WORK_USED(work_area) ((work_area).used)
1749 #define RANGE_TABLE_WORK_ELT(work_area, i) ((work_area).table[i])
1752 /* Set the bit for character C in a list. */
1753 #define SET_LIST_BIT(c) \
1754 (b[((unsigned char) (c)) / BYTEWIDTH] \
1755 |= 1 << (((unsigned char) c) % BYTEWIDTH))
1758 /* Get the next unsigned number in the uncompiled pattern. */
1759 #define GET_UNSIGNED_NUMBER(num) \
1763 while (ISDIGIT (c)) \
1767 num = num * 10 + c - '0'; \
1775 #define CHAR_CLASS_MAX_LENGTH 6 /* Namely, `xdigit'. */
1777 #define IS_CHAR_CLASS(string) \
1778 (STREQ (string, "alpha") || STREQ (string, "upper") \
1779 || STREQ (string, "lower") || STREQ (string, "digit") \
1780 || STREQ (string, "alnum") || STREQ (string, "xdigit") \
1781 || STREQ (string, "space") || STREQ (string, "print") \
1782 || STREQ (string, "punct") || STREQ (string, "graph") \
1783 || STREQ (string, "cntrl") || STREQ (string, "blank"))
1785 #ifndef MATCH_MAY_ALLOCATE
1787 /* If we cannot allocate large objects within re_match_2_internal,
1788 we make the fail stack and register vectors global.
1789 The fail stack, we grow to the maximum size when a regexp
1791 The register vectors, we adjust in size each time we
1792 compile a regexp, according to the number of registers it needs. */
1794 static fail_stack_type fail_stack
;
1796 /* Size with which the following vectors are currently allocated.
1797 That is so we can make them bigger as needed,
1798 but never make them smaller. */
1799 static int regs_allocated_size
;
1801 static const char ** regstart
, ** regend
;
1802 static const char ** old_regstart
, ** old_regend
;
1803 static const char **best_regstart
, **best_regend
;
1804 static register_info_type
*reg_info
;
1805 static const char **reg_dummy
;
1806 static register_info_type
*reg_info_dummy
;
1808 /* Make the register vectors big enough for NUM_REGS registers,
1809 but don't make them smaller. */
1812 regex_grow_registers (num_regs
)
1815 if (num_regs
> regs_allocated_size
)
1817 RETALLOC_IF (regstart
, num_regs
, const char *);
1818 RETALLOC_IF (regend
, num_regs
, const char *);
1819 RETALLOC_IF (old_regstart
, num_regs
, const char *);
1820 RETALLOC_IF (old_regend
, num_regs
, const char *);
1821 RETALLOC_IF (best_regstart
, num_regs
, const char *);
1822 RETALLOC_IF (best_regend
, num_regs
, const char *);
1823 RETALLOC_IF (reg_info
, num_regs
, register_info_type
);
1824 RETALLOC_IF (reg_dummy
, num_regs
, const char *);
1825 RETALLOC_IF (reg_info_dummy
, num_regs
, register_info_type
);
1827 regs_allocated_size
= num_regs
;
1831 #endif /* not MATCH_MAY_ALLOCATE */
1833 /* `regex_compile' compiles PATTERN (of length SIZE) according to SYNTAX.
1834 Returns one of error codes defined in `regex.h', or zero for success.
1836 Assumes the `allocated' (and perhaps `buffer') and `translate'
1837 fields are set in BUFP on entry.
1839 If it succeeds, results are put in BUFP (if it returns an error, the
1840 contents of BUFP are undefined):
1841 `buffer' is the compiled pattern;
1842 `syntax' is set to SYNTAX;
1843 `used' is set to the length of the compiled pattern;
1844 `fastmap_accurate' is zero;
1845 `re_nsub' is the number of subexpressions in PATTERN;
1846 `not_bol' and `not_eol' are zero;
1848 The `fastmap' and `newline_anchor' fields are neither
1849 examined nor set. */
1851 /* Return, freeing storage we allocated. */
1852 #define FREE_STACK_RETURN(value) \
1854 FREE_RANGE_TABLE_WORK_AREA (range_table_work); \
1855 free (compile_stack.stack); \
1859 static reg_errcode_t
1860 regex_compile (pattern
, size
, syntax
, bufp
)
1861 const char *pattern
;
1863 reg_syntax_t syntax
;
1864 struct re_pattern_buffer
*bufp
;
1866 /* We fetch characters from PATTERN here. Even though PATTERN is
1867 `char *' (i.e., signed), we declare these variables as unsigned, so
1868 they can be reliably used as array indices. */
1869 register unsigned int c
, c1
;
1871 /* A random temporary spot in PATTERN. */
1874 /* Points to the end of the buffer, where we should append. */
1875 register unsigned char *b
;
1877 /* Keeps track of unclosed groups. */
1878 compile_stack_type compile_stack
;
1880 /* Points to the current (ending) position in the pattern. */
1882 /* `const' makes AIX compiler fail. */
1885 const char *p
= pattern
;
1887 const char *pend
= pattern
+ size
;
1889 /* How to translate the characters in the pattern. */
1890 RE_TRANSLATE_TYPE translate
= bufp
->translate
;
1892 /* Address of the count-byte of the most recently inserted `exactn'
1893 command. This makes it possible to tell if a new exact-match
1894 character can be added to that command or if the character requires
1895 a new `exactn' command. */
1896 unsigned char *pending_exact
= 0;
1898 /* Address of start of the most recently finished expression.
1899 This tells, e.g., postfix * where to find the start of its
1900 operand. Reset at the beginning of groups and alternatives. */
1901 unsigned char *laststart
= 0;
1903 /* Address of beginning of regexp, or inside of last group. */
1904 unsigned char *begalt
;
1906 /* Place in the uncompiled pattern (i.e., the {) to
1907 which to go back if the interval is invalid. */
1908 const char *beg_interval
;
1910 /* Address of the place where a forward jump should go to the end of
1911 the containing expression. Each alternative of an `or' -- except the
1912 last -- ends with a forward jump of this sort. */
1913 unsigned char *fixup_alt_jump
= 0;
1915 /* Counts open-groups as they are encountered. Remembered for the
1916 matching close-group on the compile stack, so the same register
1917 number is put in the stop_memory as the start_memory. */
1918 regnum_t regnum
= 0;
1920 /* Work area for range table of charset. */
1921 struct range_table_work_area range_table_work
;
1924 DEBUG_PRINT1 ("\nCompiling pattern: ");
1927 unsigned debug_count
;
1929 for (debug_count
= 0; debug_count
< size
; debug_count
++)
1930 putchar (pattern
[debug_count
]);
1935 /* Initialize the compile stack. */
1936 compile_stack
.stack
= TALLOC (INIT_COMPILE_STACK_SIZE
, compile_stack_elt_t
);
1937 if (compile_stack
.stack
== NULL
)
1940 compile_stack
.size
= INIT_COMPILE_STACK_SIZE
;
1941 compile_stack
.avail
= 0;
1943 range_table_work
.table
= 0;
1944 range_table_work
.allocated
= 0;
1946 /* Initialize the pattern buffer. */
1947 bufp
->syntax
= syntax
;
1948 bufp
->fastmap_accurate
= 0;
1949 bufp
->not_bol
= bufp
->not_eol
= 0;
1951 /* Set `used' to zero, so that if we return an error, the pattern
1952 printer (for debugging) will think there's no pattern. We reset it
1956 /* Always count groups, whether or not bufp->no_sub is set. */
1960 /* bufp->multibyte is set before regex_compile is called, so don't alter
1962 #else /* not emacs */
1963 /* Nothing is recognized as a multibyte character. */
1964 bufp
->multibyte
= 0;
1967 #if !defined (emacs) && !defined (SYNTAX_TABLE)
1968 /* Initialize the syntax table. */
1969 init_syntax_once ();
1972 if (bufp
->allocated
== 0)
1975 { /* If zero allocated, but buffer is non-null, try to realloc
1976 enough space. This loses if buffer's address is bogus, but
1977 that is the user's responsibility. */
1978 RETALLOC (bufp
->buffer
, INIT_BUF_SIZE
, unsigned char);
1981 { /* Caller did not allocate a buffer. Do it for them. */
1982 bufp
->buffer
= TALLOC (INIT_BUF_SIZE
, unsigned char);
1984 if (!bufp
->buffer
) FREE_STACK_RETURN (REG_ESPACE
);
1986 bufp
->allocated
= INIT_BUF_SIZE
;
1989 begalt
= b
= bufp
->buffer
;
1991 /* Loop through the uncompiled pattern until we're at the end. */
2000 if ( /* If at start of pattern, it's an operator. */
2002 /* If context independent, it's an operator. */
2003 || syntax
& RE_CONTEXT_INDEP_ANCHORS
2004 /* Otherwise, depends on what's come before. */
2005 || at_begline_loc_p (pattern
, p
, syntax
))
2015 if ( /* If at end of pattern, it's an operator. */
2017 /* If context independent, it's an operator. */
2018 || syntax
& RE_CONTEXT_INDEP_ANCHORS
2019 /* Otherwise, depends on what's next. */
2020 || at_endline_loc_p (p
, pend
, syntax
))
2030 if ((syntax
& RE_BK_PLUS_QM
)
2031 || (syntax
& RE_LIMITED_OPS
))
2035 /* If there is no previous pattern... */
2038 if (syntax
& RE_CONTEXT_INVALID_OPS
)
2039 FREE_STACK_RETURN (REG_BADRPT
);
2040 else if (!(syntax
& RE_CONTEXT_INDEP_OPS
))
2045 /* Are we optimizing this jump? */
2046 boolean keep_string_p
= false;
2048 /* 1 means zero (many) matches is allowed. */
2049 char zero_times_ok
= 0, many_times_ok
= 0;
2051 /* If there is a sequence of repetition chars, collapse it
2052 down to just one (the right one). We can't combine
2053 interval operators with these because of, e.g., `a{2}*',
2054 which should only match an even number of `a's. */
2058 zero_times_ok
|= c
!= '+';
2059 many_times_ok
|= c
!= '?';
2067 || (!(syntax
& RE_BK_PLUS_QM
) && (c
== '+' || c
== '?')))
2070 else if (syntax
& RE_BK_PLUS_QM
&& c
== '\\')
2072 if (p
== pend
) FREE_STACK_RETURN (REG_EESCAPE
);
2075 if (!(c1
== '+' || c1
== '?'))
2090 /* If we get here, we found another repeat character. */
2093 /* Star, etc. applied to an empty pattern is equivalent
2094 to an empty pattern. */
2098 /* Now we know whether or not zero matches is allowed
2099 and also whether or not two or more matches is allowed. */
2101 { /* More than one repetition is allowed, so put in at the
2102 end a backward relative jump from `b' to before the next
2103 jump we're going to put in below (which jumps from
2104 laststart to after this jump).
2106 But if we are at the `*' in the exact sequence `.*\n',
2107 insert an unconditional jump backwards to the .,
2108 instead of the beginning of the loop. This way we only
2109 push a failure point once, instead of every time
2110 through the loop. */
2111 assert (p
- 1 > pattern
);
2113 /* Allocate the space for the jump. */
2114 GET_BUFFER_SPACE (3);
2116 /* We know we are not at the first character of the pattern,
2117 because laststart was nonzero. And we've already
2118 incremented `p', by the way, to be the character after
2119 the `*'. Do we have to do something analogous here
2120 for null bytes, because of RE_DOT_NOT_NULL? */
2121 if (TRANSLATE ((unsigned char)*(p
- 2)) == TRANSLATE ('.')
2124 && TRANSLATE ((unsigned char)*p
) == TRANSLATE ('\n')
2125 && !(syntax
& RE_DOT_NEWLINE
))
2126 { /* We have .*\n. */
2127 STORE_JUMP (jump
, b
, laststart
);
2128 keep_string_p
= true;
2131 /* Anything else. */
2132 STORE_JUMP (maybe_pop_jump
, b
, laststart
- 3);
2134 /* We've added more stuff to the buffer. */
2138 /* On failure, jump from laststart to b + 3, which will be the
2139 end of the buffer after this jump is inserted. */
2140 GET_BUFFER_SPACE (3);
2141 INSERT_JUMP (keep_string_p
? on_failure_keep_string_jump
2149 /* At least one repetition is required, so insert a
2150 `dummy_failure_jump' before the initial
2151 `on_failure_jump' instruction of the loop. This
2152 effects a skip over that instruction the first time
2153 we hit that loop. */
2154 GET_BUFFER_SPACE (3);
2155 INSERT_JUMP (dummy_failure_jump
, laststart
, laststart
+ 6);
2170 CLEAR_RANGE_TABLE_WORK_USED (range_table_work
);
2172 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2174 /* Ensure that we have enough space to push a charset: the
2175 opcode, the length count, and the bitset; 34 bytes in all. */
2176 GET_BUFFER_SPACE (34);
2180 /* We test `*p == '^' twice, instead of using an if
2181 statement, so we only need one BUF_PUSH. */
2182 BUF_PUSH (*p
== '^' ? charset_not
: charset
);
2186 /* Remember the first position in the bracket expression. */
2189 /* Push the number of bytes in the bitmap. */
2190 BUF_PUSH ((1 << BYTEWIDTH
) / BYTEWIDTH
);
2192 /* Clear the whole map. */
2193 bzero (b
, (1 << BYTEWIDTH
) / BYTEWIDTH
);
2195 /* charset_not matches newline according to a syntax bit. */
2196 if ((re_opcode_t
) b
[-2] == charset_not
2197 && (syntax
& RE_HAT_LISTS_NOT_NEWLINE
))
2198 SET_LIST_BIT ('\n');
2200 /* Read in characters and ranges, setting map bits. */
2204 boolean escaped_char
= false;
2206 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2210 /* \ might escape characters inside [...] and [^...]. */
2211 if ((syntax
& RE_BACKSLASH_ESCAPE_IN_LISTS
) && c
== '\\')
2213 if (p
== pend
) FREE_STACK_RETURN (REG_EESCAPE
);
2216 escaped_char
= true;
2220 /* Could be the end of the bracket expression. If it's
2221 not (i.e., when the bracket expression is `[]' so
2222 far), the ']' character bit gets set way below. */
2223 if (c
== ']' && p
!= p1
+ 1)
2227 /* If C indicates start of multibyte char, get the
2228 actual character code in C, and set the pattern
2229 pointer P to the next character boundary. */
2230 if (bufp
->multibyte
&& BASE_LEADING_CODE_P (c
))
2233 c
= STRING_CHAR_AND_LENGTH (p
, pend
- p
, len
);
2236 /* What should we do for the character which is
2237 greater than 0x7F, but not BASE_LEADING_CODE_P?
2240 /* See if we're at the beginning of a possible character
2243 else if (!escaped_char
&&
2244 syntax
& RE_CHAR_CLASSES
&& c
== '[' && *p
== ':')
2246 /* Leave room for the null. */
2247 char str
[CHAR_CLASS_MAX_LENGTH
+ 1];
2252 /* If pattern is `[[:'. */
2253 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2258 if (c
== ':' || c
== ']' || p
== pend
2259 || c1
== CHAR_CLASS_MAX_LENGTH
)
2265 /* If isn't a word bracketed by `[:' and `:]':
2266 undo the ending character, the letters, and
2267 leave the leading `:' and `[' (but set bits for
2269 if (c
== ':' && *p
== ']')
2272 boolean is_alnum
= STREQ (str
, "alnum");
2273 boolean is_alpha
= STREQ (str
, "alpha");
2274 boolean is_blank
= STREQ (str
, "blank");
2275 boolean is_cntrl
= STREQ (str
, "cntrl");
2276 boolean is_digit
= STREQ (str
, "digit");
2277 boolean is_graph
= STREQ (str
, "graph");
2278 boolean is_lower
= STREQ (str
, "lower");
2279 boolean is_print
= STREQ (str
, "print");
2280 boolean is_punct
= STREQ (str
, "punct");
2281 boolean is_space
= STREQ (str
, "space");
2282 boolean is_upper
= STREQ (str
, "upper");
2283 boolean is_xdigit
= STREQ (str
, "xdigit");
2285 if (!IS_CHAR_CLASS (str
))
2286 FREE_STACK_RETURN (REG_ECTYPE
);
2288 /* Throw away the ] at the end of the character
2292 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2294 for (ch
= 0; ch
< 1 << BYTEWIDTH
; ch
++)
2296 int translated
= TRANSLATE (ch
);
2297 /* This was split into 3 if's to
2298 avoid an arbitrary limit in some compiler. */
2299 if ( (is_alnum
&& ISALNUM (ch
))
2300 || (is_alpha
&& ISALPHA (ch
))
2301 || (is_blank
&& ISBLANK (ch
))
2302 || (is_cntrl
&& ISCNTRL (ch
)))
2303 SET_LIST_BIT (translated
);
2304 if ( (is_digit
&& ISDIGIT (ch
))
2305 || (is_graph
&& ISGRAPH (ch
))
2306 || (is_lower
&& ISLOWER (ch
))
2307 || (is_print
&& ISPRINT (ch
)))
2308 SET_LIST_BIT (translated
);
2309 if ( (is_punct
&& ISPUNCT (ch
))
2310 || (is_space
&& ISSPACE (ch
))
2311 || (is_upper
&& ISUPPER (ch
))
2312 || (is_xdigit
&& ISXDIGIT (ch
)))
2313 SET_LIST_BIT (translated
);
2316 /* Repeat the loop. */
2326 /* Because the `:' may starts the range, we
2327 can't simply set bit and repeat the loop.
2328 Instead, just set it to C and handle below. */
2333 if (p
< pend
&& p
[0] == '-' && p
[1] != ']')
2336 /* Discard the `-'. */
2339 /* Fetch the character which ends the range. */
2341 if (bufp
->multibyte
&& BASE_LEADING_CODE_P (c1
))
2344 c1
= STRING_CHAR_AND_LENGTH (p
, pend
- p
, len
);
2348 if (SINGLE_BYTE_CHAR_P (c
)
2349 && ! SINGLE_BYTE_CHAR_P (c1
))
2351 /* Handle a range such as \177-\377 in multibyte mode.
2352 Split that into two ranges,,
2353 the low one ending at 0237, and the high one
2354 starting at ...040. */
2355 int c1_base
= (c1
& ~0177) | 040;
2356 SET_RANGE_TABLE_WORK_AREA (range_table_work
, c
, c1
);
2359 else if (!SAME_CHARSET_P (c
, c1
))
2360 FREE_STACK_RETURN (REG_ERANGE
);
2363 /* Range from C to C. */
2366 /* Set the range ... */
2367 if (SINGLE_BYTE_CHAR_P (c
))
2368 /* ... into bitmap. */
2371 int range_start
= c
, range_end
= c1
;
2373 /* If the start is after the end, the range is empty. */
2374 if (range_start
> range_end
)
2376 if (syntax
& RE_NO_EMPTY_RANGES
)
2377 FREE_STACK_RETURN (REG_ERANGE
);
2378 /* Else, repeat the loop. */
2382 for (this_char
= range_start
; this_char
<= range_end
;
2384 SET_LIST_BIT (TRANSLATE (this_char
));
2388 /* ... into range table. */
2389 SET_RANGE_TABLE_WORK_AREA (range_table_work
, c
, c1
);
2392 /* Discard any (non)matching list bytes that are all 0 at the
2393 end of the map. Decrease the map-length byte too. */
2394 while ((int) b
[-1] > 0 && b
[b
[-1] - 1] == 0)
2398 /* Build real range table from work area. */
2399 if (RANGE_TABLE_WORK_USED (range_table_work
))
2402 int used
= RANGE_TABLE_WORK_USED (range_table_work
);
2404 /* Allocate space for COUNT + RANGE_TABLE. Needs two
2405 bytes for COUNT and three bytes for each character. */
2406 GET_BUFFER_SPACE (2 + used
* 3);
2408 /* Indicate the existence of range table. */
2409 laststart
[1] |= 0x80;
2411 STORE_NUMBER_AND_INCR (b
, used
/ 2);
2412 for (i
= 0; i
< used
; i
++)
2413 STORE_CHARACTER_AND_INCR
2414 (b
, RANGE_TABLE_WORK_ELT (range_table_work
, i
));
2421 if (syntax
& RE_NO_BK_PARENS
)
2428 if (syntax
& RE_NO_BK_PARENS
)
2435 if (syntax
& RE_NEWLINE_ALT
)
2442 if (syntax
& RE_NO_BK_VBAR
)
2449 if (syntax
& RE_INTERVALS
&& syntax
& RE_NO_BK_BRACES
)
2450 goto handle_interval
;
2456 if (p
== pend
) FREE_STACK_RETURN (REG_EESCAPE
);
2458 /* Do not translate the character after the \, so that we can
2459 distinguish, e.g., \B from \b, even if we normally would
2460 translate, e.g., B to b. */
2466 if (syntax
& RE_NO_BK_PARENS
)
2467 goto normal_backslash
;
2473 if (COMPILE_STACK_FULL
)
2475 RETALLOC (compile_stack
.stack
, compile_stack
.size
<< 1,
2476 compile_stack_elt_t
);
2477 if (compile_stack
.stack
== NULL
) return REG_ESPACE
;
2479 compile_stack
.size
<<= 1;
2482 /* These are the values to restore when we hit end of this
2483 group. They are all relative offsets, so that if the
2484 whole pattern moves because of realloc, they will still
2486 COMPILE_STACK_TOP
.begalt_offset
= begalt
- bufp
->buffer
;
2487 COMPILE_STACK_TOP
.fixup_alt_jump
2488 = fixup_alt_jump
? fixup_alt_jump
- bufp
->buffer
+ 1 : 0;
2489 COMPILE_STACK_TOP
.laststart_offset
= b
- bufp
->buffer
;
2490 COMPILE_STACK_TOP
.regnum
= regnum
;
2492 /* We will eventually replace the 0 with the number of
2493 groups inner to this one. But do not push a
2494 start_memory for groups beyond the last one we can
2495 represent in the compiled pattern. */
2496 if (regnum
<= MAX_REGNUM
)
2498 COMPILE_STACK_TOP
.inner_group_offset
= b
- bufp
->buffer
+ 2;
2499 BUF_PUSH_3 (start_memory
, regnum
, 0);
2502 compile_stack
.avail
++;
2507 /* If we've reached MAX_REGNUM groups, then this open
2508 won't actually generate any code, so we'll have to
2509 clear pending_exact explicitly. */
2515 if (syntax
& RE_NO_BK_PARENS
) goto normal_backslash
;
2517 if (COMPILE_STACK_EMPTY
)
2518 if (syntax
& RE_UNMATCHED_RIGHT_PAREN_ORD
)
2519 goto normal_backslash
;
2521 FREE_STACK_RETURN (REG_ERPAREN
);
2525 { /* Push a dummy failure point at the end of the
2526 alternative for a possible future
2527 `pop_failure_jump' to pop. See comments at
2528 `push_dummy_failure' in `re_match_2'. */
2529 BUF_PUSH (push_dummy_failure
);
2531 /* We allocated space for this jump when we assigned
2532 to `fixup_alt_jump', in the `handle_alt' case below. */
2533 STORE_JUMP (jump_past_alt
, fixup_alt_jump
, b
- 1);
2536 /* See similar code for backslashed left paren above. */
2537 if (COMPILE_STACK_EMPTY
)
2538 if (syntax
& RE_UNMATCHED_RIGHT_PAREN_ORD
)
2541 FREE_STACK_RETURN (REG_ERPAREN
);
2543 /* Since we just checked for an empty stack above, this
2544 ``can't happen''. */
2545 assert (compile_stack
.avail
!= 0);
2547 /* We don't just want to restore into `regnum', because
2548 later groups should continue to be numbered higher,
2549 as in `(ab)c(de)' -- the second group is #2. */
2550 regnum_t this_group_regnum
;
2552 compile_stack
.avail
--;
2553 begalt
= bufp
->buffer
+ COMPILE_STACK_TOP
.begalt_offset
;
2555 = COMPILE_STACK_TOP
.fixup_alt_jump
2556 ? bufp
->buffer
+ COMPILE_STACK_TOP
.fixup_alt_jump
- 1
2558 laststart
= bufp
->buffer
+ COMPILE_STACK_TOP
.laststart_offset
;
2559 this_group_regnum
= COMPILE_STACK_TOP
.regnum
;
2560 /* If we've reached MAX_REGNUM groups, then this open
2561 won't actually generate any code, so we'll have to
2562 clear pending_exact explicitly. */
2565 /* We're at the end of the group, so now we know how many
2566 groups were inside this one. */
2567 if (this_group_regnum
<= MAX_REGNUM
)
2569 unsigned char *inner_group_loc
2570 = bufp
->buffer
+ COMPILE_STACK_TOP
.inner_group_offset
;
2572 *inner_group_loc
= regnum
- this_group_regnum
;
2573 BUF_PUSH_3 (stop_memory
, this_group_regnum
,
2574 regnum
- this_group_regnum
);
2580 case '|': /* `\|'. */
2581 if (syntax
& RE_LIMITED_OPS
|| syntax
& RE_NO_BK_VBAR
)
2582 goto normal_backslash
;
2584 if (syntax
& RE_LIMITED_OPS
)
2587 /* Insert before the previous alternative a jump which
2588 jumps to this alternative if the former fails. */
2589 GET_BUFFER_SPACE (3);
2590 INSERT_JUMP (on_failure_jump
, begalt
, b
+ 6);
2594 /* The alternative before this one has a jump after it
2595 which gets executed if it gets matched. Adjust that
2596 jump so it will jump to this alternative's analogous
2597 jump (put in below, which in turn will jump to the next
2598 (if any) alternative's such jump, etc.). The last such
2599 jump jumps to the correct final destination. A picture:
2605 If we are at `b', then fixup_alt_jump right now points to a
2606 three-byte space after `a'. We'll put in the jump, set
2607 fixup_alt_jump to right after `b', and leave behind three
2608 bytes which we'll fill in when we get to after `c'. */
2611 STORE_JUMP (jump_past_alt
, fixup_alt_jump
, b
);
2613 /* Mark and leave space for a jump after this alternative,
2614 to be filled in later either by next alternative or
2615 when know we're at the end of a series of alternatives. */
2617 GET_BUFFER_SPACE (3);
2626 /* If \{ is a literal. */
2627 if (!(syntax
& RE_INTERVALS
)
2628 /* If we're at `\{' and it's not the open-interval
2630 || ((syntax
& RE_INTERVALS
) && (syntax
& RE_NO_BK_BRACES
))
2631 || (p
- 2 == pattern
&& p
== pend
))
2632 goto normal_backslash
;
2636 /* If got here, then the syntax allows intervals. */
2638 /* At least (most) this many matches must be made. */
2639 int lower_bound
= -1, upper_bound
= -1;
2641 beg_interval
= p
- 1;
2645 if (syntax
& RE_NO_BK_BRACES
)
2646 goto unfetch_interval
;
2648 FREE_STACK_RETURN (REG_EBRACE
);
2651 GET_UNSIGNED_NUMBER (lower_bound
);
2655 GET_UNSIGNED_NUMBER (upper_bound
);
2656 if (upper_bound
< 0) upper_bound
= RE_DUP_MAX
;
2659 /* Interval such as `{1}' => match exactly once. */
2660 upper_bound
= lower_bound
;
2662 if (lower_bound
< 0 || upper_bound
> RE_DUP_MAX
2663 || lower_bound
> upper_bound
)
2665 if (syntax
& RE_NO_BK_BRACES
)
2666 goto unfetch_interval
;
2668 FREE_STACK_RETURN (REG_BADBR
);
2671 if (!(syntax
& RE_NO_BK_BRACES
))
2673 if (c
!= '\\') FREE_STACK_RETURN (REG_EBRACE
);
2680 if (syntax
& RE_NO_BK_BRACES
)
2681 goto unfetch_interval
;
2683 FREE_STACK_RETURN (REG_BADBR
);
2686 /* We just parsed a valid interval. */
2688 /* If it's invalid to have no preceding re. */
2691 if (syntax
& RE_CONTEXT_INVALID_OPS
)
2692 FREE_STACK_RETURN (REG_BADRPT
);
2693 else if (syntax
& RE_CONTEXT_INDEP_OPS
)
2696 goto unfetch_interval
;
2699 /* If the upper bound is zero, don't want to succeed at
2700 all; jump from `laststart' to `b + 3', which will be
2701 the end of the buffer after we insert the jump. */
2702 if (upper_bound
== 0)
2704 GET_BUFFER_SPACE (3);
2705 INSERT_JUMP (jump
, laststart
, b
+ 3);
2709 /* Otherwise, we have a nontrivial interval. When
2710 we're all done, the pattern will look like:
2711 set_number_at <jump count> <upper bound>
2712 set_number_at <succeed_n count> <lower bound>
2713 succeed_n <after jump addr> <succeed_n count>
2715 jump_n <succeed_n addr> <jump count>
2716 (The upper bound and `jump_n' are omitted if
2717 `upper_bound' is 1, though.) */
2719 { /* If the upper bound is > 1, we need to insert
2720 more at the end of the loop. */
2721 unsigned nbytes
= 10 + (upper_bound
> 1) * 10;
2723 GET_BUFFER_SPACE (nbytes
);
2725 /* Initialize lower bound of the `succeed_n', even
2726 though it will be set during matching by its
2727 attendant `set_number_at' (inserted next),
2728 because `re_compile_fastmap' needs to know.
2729 Jump to the `jump_n' we might insert below. */
2730 INSERT_JUMP2 (succeed_n
, laststart
,
2731 b
+ 5 + (upper_bound
> 1) * 5,
2735 /* Code to initialize the lower bound. Insert
2736 before the `succeed_n'. The `5' is the last two
2737 bytes of this `set_number_at', plus 3 bytes of
2738 the following `succeed_n'. */
2739 insert_op2 (set_number_at
, laststart
, 5, lower_bound
, b
);
2742 if (upper_bound
> 1)
2743 { /* More than one repetition is allowed, so
2744 append a backward jump to the `succeed_n'
2745 that starts this interval.
2747 When we've reached this during matching,
2748 we'll have matched the interval once, so
2749 jump back only `upper_bound - 1' times. */
2750 STORE_JUMP2 (jump_n
, b
, laststart
+ 5,
2754 /* The location we want to set is the second
2755 parameter of the `jump_n'; that is `b-2' as
2756 an absolute address. `laststart' will be
2757 the `set_number_at' we're about to insert;
2758 `laststart+3' the number to set, the source
2759 for the relative address. But we are
2760 inserting into the middle of the pattern --
2761 so everything is getting moved up by 5.
2762 Conclusion: (b - 2) - (laststart + 3) + 5,
2763 i.e., b - laststart.
2765 We insert this at the beginning of the loop
2766 so that if we fail during matching, we'll
2767 reinitialize the bounds. */
2768 insert_op2 (set_number_at
, laststart
, b
- laststart
,
2769 upper_bound
- 1, b
);
2774 beg_interval
= NULL
;
2779 /* If an invalid interval, match the characters as literals. */
2780 assert (beg_interval
);
2782 beg_interval
= NULL
;
2784 /* normal_char and normal_backslash need `c'. */
2787 if (!(syntax
& RE_NO_BK_BRACES
))
2789 if (p
> pattern
&& p
[-1] == '\\')
2790 goto normal_backslash
;
2795 /* There is no way to specify the before_dot and after_dot
2796 operators. rms says this is ok. --karl */
2804 BUF_PUSH_2 (syntaxspec
, syntax_spec_code
[c
]);
2810 BUF_PUSH_2 (notsyntaxspec
, syntax_spec_code
[c
]);
2816 BUF_PUSH_2 (categoryspec
, c
);
2822 BUF_PUSH_2 (notcategoryspec
, c
);
2829 BUF_PUSH (wordchar
);
2835 BUF_PUSH (notwordchar
);
2848 BUF_PUSH (wordbound
);
2852 BUF_PUSH (notwordbound
);
2863 case '1': case '2': case '3': case '4': case '5':
2864 case '6': case '7': case '8': case '9':
2865 if (syntax
& RE_NO_BK_REFS
)
2871 FREE_STACK_RETURN (REG_ESUBREG
);
2873 /* Can't back reference to a subexpression if inside of it. */
2874 if (group_in_compile_stack (compile_stack
, c1
))
2878 BUF_PUSH_2 (duplicate
, c1
);
2884 if (syntax
& RE_BK_PLUS_QM
)
2887 goto normal_backslash
;
2891 /* You might think it would be useful for \ to mean
2892 not to translate; but if we don't translate it
2893 it will never match anything. */
2901 /* Expects the character in `c'. */
2903 p1
= p
- 1; /* P1 points the head of C. */
2905 if (bufp
->multibyte
)
2907 c
= STRING_CHAR (p1
, pend
- p1
);
2909 /* Set P to the next character boundary. */
2910 p
+= MULTIBYTE_FORM_LENGTH (p1
, pend
- p1
) - 1;
2913 /* If no exactn currently being built. */
2916 /* If last exactn not at current position. */
2917 || pending_exact
+ *pending_exact
+ 1 != b
2919 /* We have only one byte following the exactn for the count. */
2920 || *pending_exact
>= (1 << BYTEWIDTH
) - (p
- p1
)
2922 /* If followed by a repetition operator. */
2923 || (p
!= pend
&& (*p
== '*' || *p
== '^'))
2924 || ((syntax
& RE_BK_PLUS_QM
)
2925 ? p
+ 1 < pend
&& *p
== '\\' && (p
[1] == '+' || p
[1] == '?')
2926 : p
!= pend
&& (*p
== '+' || *p
== '?'))
2927 || ((syntax
& RE_INTERVALS
)
2928 && ((syntax
& RE_NO_BK_BRACES
)
2929 ? p
!= pend
&& *p
== '{'
2930 : p
+ 1 < pend
&& p
[0] == '\\' && p
[1] == '{')))
2932 /* Start building a new exactn. */
2936 BUF_PUSH_2 (exactn
, 0);
2937 pending_exact
= b
- 1;
2941 if (! SINGLE_BYTE_CHAR_P (c
))
2943 unsigned char work
[4], *str
;
2944 int i
= CHAR_STRING (c
, work
, str
);
2946 for (j
= 0; j
< i
; j
++)
2960 } /* while p != pend */
2963 /* Through the pattern now. */
2966 STORE_JUMP (jump_past_alt
, fixup_alt_jump
, b
);
2968 if (!COMPILE_STACK_EMPTY
)
2969 FREE_STACK_RETURN (REG_EPAREN
);
2971 /* If we don't want backtracking, force success
2972 the first time we reach the end of the compiled pattern. */
2973 if (syntax
& RE_NO_POSIX_BACKTRACKING
)
2976 free (compile_stack
.stack
);
2978 /* We have succeeded; set the length of the buffer. */
2979 bufp
->used
= b
- bufp
->buffer
;
2984 DEBUG_PRINT1 ("\nCompiled pattern: \n");
2985 print_compiled_pattern (bufp
);
2989 #ifndef MATCH_MAY_ALLOCATE
2990 /* Initialize the failure stack to the largest possible stack. This
2991 isn't necessary unless we're trying to avoid calling alloca in
2992 the search and match routines. */
2994 int num_regs
= bufp
->re_nsub
+ 1;
2996 if (fail_stack
.size
< re_max_failures
* TYPICAL_FAILURE_SIZE
)
2998 fail_stack
.size
= re_max_failures
* TYPICAL_FAILURE_SIZE
;
3001 if (! fail_stack
.stack
)
3003 = (fail_stack_elt_t
*) xmalloc (fail_stack
.size
3004 * sizeof (fail_stack_elt_t
));
3007 = (fail_stack_elt_t
*) xrealloc (fail_stack
.stack
,
3009 * sizeof (fail_stack_elt_t
)));
3010 #else /* not emacs */
3011 if (! fail_stack
.stack
)
3013 = (fail_stack_elt_t
*) malloc (fail_stack
.size
3014 * sizeof (fail_stack_elt_t
));
3017 = (fail_stack_elt_t
*) realloc (fail_stack
.stack
,
3019 * sizeof (fail_stack_elt_t
)));
3020 #endif /* not emacs */
3023 regex_grow_registers (num_regs
);
3025 #endif /* not MATCH_MAY_ALLOCATE */
3028 } /* regex_compile */
3030 /* Subroutines for `regex_compile'. */
3032 /* Store OP at LOC followed by two-byte integer parameter ARG. */
3035 store_op1 (op
, loc
, arg
)
3040 *loc
= (unsigned char) op
;
3041 STORE_NUMBER (loc
+ 1, arg
);
3045 /* Like `store_op1', but for two two-byte parameters ARG1 and ARG2. */
3048 store_op2 (op
, loc
, arg1
, arg2
)
3053 *loc
= (unsigned char) op
;
3054 STORE_NUMBER (loc
+ 1, arg1
);
3055 STORE_NUMBER (loc
+ 3, arg2
);
3059 /* Copy the bytes from LOC to END to open up three bytes of space at LOC
3060 for OP followed by two-byte integer parameter ARG. */
3063 insert_op1 (op
, loc
, arg
, end
)
3069 register unsigned char *pfrom
= end
;
3070 register unsigned char *pto
= end
+ 3;
3072 while (pfrom
!= loc
)
3075 store_op1 (op
, loc
, arg
);
3079 /* Like `insert_op1', but for two two-byte parameters ARG1 and ARG2. */
3082 insert_op2 (op
, loc
, arg1
, arg2
, end
)
3088 register unsigned char *pfrom
= end
;
3089 register unsigned char *pto
= end
+ 5;
3091 while (pfrom
!= loc
)
3094 store_op2 (op
, loc
, arg1
, arg2
);
3098 /* P points to just after a ^ in PATTERN. Return true if that ^ comes
3099 after an alternative or a begin-subexpression. We assume there is at
3100 least one character before the ^. */
3103 at_begline_loc_p (pattern
, p
, syntax
)
3104 const char *pattern
, *p
;
3105 reg_syntax_t syntax
;
3107 const char *prev
= p
- 2;
3108 boolean prev_prev_backslash
= prev
> pattern
&& prev
[-1] == '\\';
3111 /* After a subexpression? */
3112 (*prev
== '(' && (syntax
& RE_NO_BK_PARENS
|| prev_prev_backslash
))
3113 /* After an alternative? */
3114 || (*prev
== '|' && (syntax
& RE_NO_BK_VBAR
|| prev_prev_backslash
));
3118 /* The dual of at_begline_loc_p. This one is for $. We assume there is
3119 at least one character after the $, i.e., `P < PEND'. */
3122 at_endline_loc_p (p
, pend
, syntax
)
3123 const char *p
, *pend
;
3126 const char *next
= p
;
3127 boolean next_backslash
= *next
== '\\';
3128 const char *next_next
= p
+ 1 < pend
? p
+ 1 : 0;
3131 /* Before a subexpression? */
3132 (syntax
& RE_NO_BK_PARENS
? *next
== ')'
3133 : next_backslash
&& next_next
&& *next_next
== ')')
3134 /* Before an alternative? */
3135 || (syntax
& RE_NO_BK_VBAR
? *next
== '|'
3136 : next_backslash
&& next_next
&& *next_next
== '|');
3140 /* Returns true if REGNUM is in one of COMPILE_STACK's elements and
3141 false if it's not. */
3144 group_in_compile_stack (compile_stack
, regnum
)
3145 compile_stack_type compile_stack
;
3150 for (this_element
= compile_stack
.avail
- 1;
3153 if (compile_stack
.stack
[this_element
].regnum
== regnum
)
3159 /* re_compile_fastmap computes a ``fastmap'' for the compiled pattern in
3160 BUFP. A fastmap records which of the (1 << BYTEWIDTH) possible
3161 characters can start a string that matches the pattern. This fastmap
3162 is used by re_search to skip quickly over impossible starting points.
3164 The caller must supply the address of a (1 << BYTEWIDTH)-byte data
3165 area as BUFP->fastmap.
3167 We set the `fastmap', `fastmap_accurate', and `can_be_null' fields in
3170 Returns 0 if we succeed, -2 if an internal error. */
3173 re_compile_fastmap (bufp
)
3174 struct re_pattern_buffer
*bufp
;
3177 #ifdef MATCH_MAY_ALLOCATE
3178 fail_stack_type fail_stack
;
3180 #ifndef REGEX_MALLOC
3183 /* We don't push any register information onto the failure stack. */
3184 unsigned num_regs
= 0;
3186 register char *fastmap
= bufp
->fastmap
;
3187 unsigned char *pattern
= bufp
->buffer
;
3188 unsigned long size
= bufp
->used
;
3189 unsigned char *p
= pattern
;
3190 register unsigned char *pend
= pattern
+ size
;
3192 /* This holds the pointer to the failure stack, when
3193 it is allocated relocatably. */
3194 fail_stack_elt_t
*failure_stack_ptr
;
3196 /* Assume that each path through the pattern can be null until
3197 proven otherwise. We set this false at the bottom of switch
3198 statement, to which we get only if a particular path doesn't
3199 match the empty string. */
3200 boolean path_can_be_null
= true;
3202 /* We aren't doing a `succeed_n' to begin with. */
3203 boolean succeed_n_p
= false;
3205 /* If all elements for base leading-codes in fastmap is set, this
3206 flag is set true. */
3207 boolean match_any_multibyte_characters
= false;
3209 /* Maximum code of simple (single byte) character. */
3210 int simple_char_max
;
3212 assert (fastmap
!= NULL
&& p
!= NULL
);
3215 bzero (fastmap
, 1 << BYTEWIDTH
); /* Assume nothing's valid. */
3216 bufp
->fastmap_accurate
= 1; /* It will be when we're done. */
3217 bufp
->can_be_null
= 0;
3221 if (p
== pend
|| *p
== succeed
)
3223 /* We have reached the (effective) end of pattern. */
3224 if (!FAIL_STACK_EMPTY ())
3226 bufp
->can_be_null
|= path_can_be_null
;
3228 /* Reset for next path. */
3229 path_can_be_null
= true;
3231 p
= fail_stack
.stack
[--fail_stack
.avail
].pointer
;
3239 /* We should never be about to go beyond the end of the pattern. */
3242 switch (SWITCH_ENUM_CAST ((re_opcode_t
) *p
++))
3245 /* I guess the idea here is to simply not bother with a fastmap
3246 if a backreference is used, since it's too hard to figure out
3247 the fastmap for the corresponding group. Setting
3248 `can_be_null' stops `re_search_2' from using the fastmap, so
3249 that is all we do. */
3251 bufp
->can_be_null
= 1;
3255 /* Following are the cases which match a character. These end
3265 for (j
= *p
++ * BYTEWIDTH
- 1; j
>= 0; j
--)
3266 if (p
[j
/ BYTEWIDTH
] & (1 << (j
% BYTEWIDTH
)))
3272 /* Chars beyond end of map must be allowed. */
3273 for (j
= *p
* BYTEWIDTH
; j
< (1 << BYTEWIDTH
); j
++)
3276 for (j
= *p
++ * BYTEWIDTH
- 1; j
>= 0; j
--)
3277 if (!(p
[j
/ BYTEWIDTH
] & (1 << (j
% BYTEWIDTH
))))
3283 for (j
= 0; j
< (1 << BYTEWIDTH
); j
++)
3284 if (SYNTAX (j
) == Sword
)
3290 for (j
= 0; j
< (1 << BYTEWIDTH
); j
++)
3291 if (SYNTAX (j
) != Sword
)
3296 for (j
= CHARSET_BITMAP_SIZE (&p
[-1]) * BYTEWIDTH
- 1, p
++;
3298 if (p
[j
/ BYTEWIDTH
] & (1 << (j
% BYTEWIDTH
)))
3301 if (CHARSET_RANGE_TABLE_EXISTS_P (&p
[-2])
3302 && match_any_multibyte_characters
== false)
3304 /* Set fastmap[I] 1 where I is a base leading code of each
3305 multibyte character in the range table. */
3308 /* Make P points the range table. */
3309 p
+= CHARSET_BITMAP_SIZE (&p
[-2]);
3311 /* Extract the number of ranges in range table into
3313 EXTRACT_NUMBER_AND_INCR (count
, p
);
3314 for (; count
> 0; count
--, p
+= 2 * 3) /* XXX */
3316 /* Extract the start of each range. */
3317 EXTRACT_CHARACTER (c
, p
);
3318 j
= CHAR_CHARSET (c
);
3319 fastmap
[CHARSET_LEADING_CODE_BASE (j
)] = 1;
3326 /* Chars beyond end of map must be allowed. End of map is
3327 `127' if bufp->multibyte is nonzero. */
3328 simple_char_max
= bufp
->multibyte
? 0x80 : (1 << BYTEWIDTH
);
3329 for (j
= CHARSET_BITMAP_SIZE (&p
[-1]) * BYTEWIDTH
;
3330 j
< simple_char_max
; j
++)
3333 for (j
= CHARSET_BITMAP_SIZE (&p
[-1]) * BYTEWIDTH
- 1, p
++;
3335 if (!(p
[j
/ BYTEWIDTH
] & (1 << (j
% BYTEWIDTH
))))
3338 if (bufp
->multibyte
)
3339 /* Any character set can possibly contain a character
3340 which doesn't match the specified set of characters. */
3342 set_fastmap_for_multibyte_characters
:
3343 if (match_any_multibyte_characters
== false)
3345 for (j
= 0x80; j
< 0xA0; j
++) /* XXX */
3346 if (BASE_LEADING_CODE_P (j
))
3348 match_any_multibyte_characters
= true;
3355 simple_char_max
= bufp
->multibyte
? 0x80 : (1 << BYTEWIDTH
);
3356 for (j
= 0; j
< simple_char_max
; j
++)
3357 if (SYNTAX (j
) == Sword
)
3360 if (bufp
->multibyte
)
3361 /* Any character set can possibly contain a character
3362 whose syntax is `Sword'. */
3363 goto set_fastmap_for_multibyte_characters
;
3368 simple_char_max
= bufp
->multibyte
? 0x80 : (1 << BYTEWIDTH
);
3369 for (j
= 0; j
< simple_char_max
; j
++)
3370 if (SYNTAX (j
) != Sword
)
3373 if (bufp
->multibyte
)
3374 /* Any character set can possibly contain a character
3375 whose syntax is not `Sword'. */
3376 goto set_fastmap_for_multibyte_characters
;
3382 int fastmap_newline
= fastmap
['\n'];
3384 /* `.' matches anything, except perhaps newline.
3385 Even in a multibyte buffer, it should match any
3386 conceivable byte value for the fastmap. */
3387 if (bufp
->multibyte
)
3388 match_any_multibyte_characters
= true;
3390 simple_char_max
= (1 << BYTEWIDTH
);
3391 for (j
= 0; j
< simple_char_max
; j
++)
3394 /* ... except perhaps newline. */
3395 if (!(bufp
->syntax
& RE_DOT_NEWLINE
))
3396 fastmap
['\n'] = fastmap_newline
;
3398 /* Return if we have already set `can_be_null'; if we have,
3399 then the fastmap is irrelevant. Something's wrong here. */
3400 else if (bufp
->can_be_null
)
3403 /* Otherwise, have to check alternative paths. */
3414 /* This match depends on text properties. These end with
3415 aborting optimizations. */
3416 bufp
->can_be_null
= 1;
3420 simple_char_max
= bufp
->multibyte
? 0x80 : (1 << BYTEWIDTH
);
3421 for (j
= 0; j
< simple_char_max
; j
++)
3422 if (SYNTAX (j
) == (enum syntaxcode
) k
)
3425 if (bufp
->multibyte
)
3426 /* Any character set can possibly contain a character
3427 whose syntax is K. */
3428 goto set_fastmap_for_multibyte_characters
;
3433 simple_char_max
= bufp
->multibyte
? 0x80 : (1 << BYTEWIDTH
);
3434 for (j
= 0; j
< simple_char_max
; j
++)
3435 if (SYNTAX (j
) != (enum syntaxcode
) k
)
3438 if (bufp
->multibyte
)
3439 /* Any character set can possibly contain a character
3440 whose syntax is not K. */
3441 goto set_fastmap_for_multibyte_characters
;
3448 simple_char_max
= bufp
->multibyte
? 0x80 : (1 << BYTEWIDTH
);
3449 for (j
= 0; j
< simple_char_max
; j
++)
3450 if (CHAR_HAS_CATEGORY (j
, k
))
3453 if (bufp
->multibyte
)
3454 /* Any character set can possibly contain a character
3455 whose category is K. */
3456 goto set_fastmap_for_multibyte_characters
;
3460 case notcategoryspec
:
3462 simple_char_max
= bufp
->multibyte
? 0x80 : (1 << BYTEWIDTH
);
3463 for (j
= 0; j
< simple_char_max
; j
++)
3464 if (!CHAR_HAS_CATEGORY (j
, k
))
3467 if (bufp
->multibyte
)
3468 /* Any character set can possibly contain a character
3469 whose category is not K. */
3470 goto set_fastmap_for_multibyte_characters
;
3473 /* All cases after this match the empty string. These end with
3495 case push_dummy_failure
:
3500 case pop_failure_jump
:
3501 case maybe_pop_jump
:
3504 case dummy_failure_jump
:
3505 EXTRACT_NUMBER_AND_INCR (j
, p
);
3510 /* Jump backward implies we just went through the body of a
3511 loop and matched nothing. Opcode jumped to should be
3512 `on_failure_jump' or `succeed_n'. Just treat it like an
3513 ordinary jump. For a * loop, it has pushed its failure
3514 point already; if so, discard that as redundant. */
3515 if ((re_opcode_t
) *p
!= on_failure_jump
3516 && (re_opcode_t
) *p
!= succeed_n
)
3520 EXTRACT_NUMBER_AND_INCR (j
, p
);
3523 /* If what's on the stack is where we are now, pop it. */
3524 if (!FAIL_STACK_EMPTY ()
3525 && fail_stack
.stack
[fail_stack
.avail
- 1].pointer
== p
)
3531 case on_failure_jump
:
3532 case on_failure_keep_string_jump
:
3533 handle_on_failure_jump
:
3534 EXTRACT_NUMBER_AND_INCR (j
, p
);
3536 /* For some patterns, e.g., `(a?)?', `p+j' here points to the
3537 end of the pattern. We don't want to push such a point,
3538 since when we restore it above, entering the switch will
3539 increment `p' past the end of the pattern. We don't need
3540 to push such a point since we obviously won't find any more
3541 fastmap entries beyond `pend'. Such a pattern can match
3542 the null string, though. */
3545 if (!PUSH_PATTERN_OP (p
+ j
, fail_stack
))
3547 RESET_FAIL_STACK ();
3552 bufp
->can_be_null
= 1;
3556 EXTRACT_NUMBER_AND_INCR (k
, p
); /* Skip the n. */
3557 succeed_n_p
= false;
3564 /* Get to the number of times to succeed. */
3567 /* Increment p past the n for when k != 0. */
3568 EXTRACT_NUMBER_AND_INCR (k
, p
);
3572 succeed_n_p
= true; /* Spaghetti code alert. */
3573 goto handle_on_failure_jump
;
3590 abort (); /* We have listed all the cases. */
3593 /* Getting here means we have found the possible starting
3594 characters for one path of the pattern -- and that the empty
3595 string does not match. We need not follow this path further.
3596 Instead, look at the next alternative (remembered on the
3597 stack), or quit if no more. The test at the top of the loop
3598 does these things. */
3599 path_can_be_null
= false;
3603 /* Set `can_be_null' for the last path (also the first path, if the
3604 pattern is empty). */
3605 bufp
->can_be_null
|= path_can_be_null
;
3608 RESET_FAIL_STACK ();
3610 } /* re_compile_fastmap */
3612 /* Set REGS to hold NUM_REGS registers, storing them in STARTS and
3613 ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use
3614 this memory for recording register information. STARTS and ENDS
3615 must be allocated using the malloc library routine, and must each
3616 be at least NUM_REGS * sizeof (regoff_t) bytes long.
3618 If NUM_REGS == 0, then subsequent matches should allocate their own
3621 Unless this function is called, the first search or match using
3622 PATTERN_BUFFER will allocate its own register data, without
3623 freeing the old data. */
3626 re_set_registers (bufp
, regs
, num_regs
, starts
, ends
)
3627 struct re_pattern_buffer
*bufp
;
3628 struct re_registers
*regs
;
3630 regoff_t
*starts
, *ends
;
3634 bufp
->regs_allocated
= REGS_REALLOCATE
;
3635 regs
->num_regs
= num_regs
;
3636 regs
->start
= starts
;
3641 bufp
->regs_allocated
= REGS_UNALLOCATED
;
3643 regs
->start
= regs
->end
= (regoff_t
*) 0;
3647 /* Searching routines. */
3649 /* Like re_search_2, below, but only one string is specified, and
3650 doesn't let you say where to stop matching. */
3653 re_search (bufp
, string
, size
, startpos
, range
, regs
)
3654 struct re_pattern_buffer
*bufp
;
3656 int size
, startpos
, range
;
3657 struct re_registers
*regs
;
3659 return re_search_2 (bufp
, NULL
, 0, string
, size
, startpos
, range
,
3663 /* End address of virtual concatenation of string. */
3664 #define STOP_ADDR_VSTRING(P) \
3665 (((P) >= size1 ? string2 + size2 : string1 + size1))
3667 /* Address of POS in the concatenation of virtual string. */
3668 #define POS_ADDR_VSTRING(POS) \
3669 (((POS) >= size1 ? string2 - size1 : string1) + (POS))
3671 /* Using the compiled pattern in BUFP->buffer, first tries to match the
3672 virtual concatenation of STRING1 and STRING2, starting first at index
3673 STARTPOS, then at STARTPOS + 1, and so on.
3675 STRING1 and STRING2 have length SIZE1 and SIZE2, respectively.
3677 RANGE is how far to scan while trying to match. RANGE = 0 means try
3678 only at STARTPOS; in general, the last start tried is STARTPOS +
3681 In REGS, return the indices of the virtual concatenation of STRING1
3682 and STRING2 that matched the entire BUFP->buffer and its contained
3685 Do not consider matching one past the index STOP in the virtual
3686 concatenation of STRING1 and STRING2.
3688 We return either the position in the strings at which the match was
3689 found, -1 if no match, or -2 if error (such as failure
3693 re_search_2 (bufp
, string1
, size1
, string2
, size2
, startpos
, range
, regs
, stop
)
3694 struct re_pattern_buffer
*bufp
;
3695 const char *string1
, *string2
;
3699 struct re_registers
*regs
;
3703 register char *fastmap
= bufp
->fastmap
;
3704 register RE_TRANSLATE_TYPE translate
= bufp
->translate
;
3705 int total_size
= size1
+ size2
;
3706 int endpos
= startpos
+ range
;
3707 int anchored_start
= 0;
3709 /* Nonzero if we have to concern multibyte character. */
3710 int multibyte
= bufp
->multibyte
;
3712 /* Check for out-of-range STARTPOS. */
3713 if (startpos
< 0 || startpos
> total_size
)
3716 /* Fix up RANGE if it might eventually take us outside
3717 the virtual concatenation of STRING1 and STRING2.
3718 Make sure we won't move STARTPOS below 0 or above TOTAL_SIZE. */
3720 range
= 0 - startpos
;
3721 else if (endpos
> total_size
)
3722 range
= total_size
- startpos
;
3724 /* If the search isn't to be a backwards one, don't waste time in a
3725 search for a pattern anchored at beginning of buffer. */
3726 if (bufp
->used
> 0 && (re_opcode_t
) bufp
->buffer
[0] == begbuf
&& range
> 0)
3735 /* In a forward search for something that starts with \=.
3736 don't keep searching past point. */
3737 if (bufp
->used
> 0 && (re_opcode_t
) bufp
->buffer
[0] == at_dot
&& range
> 0)
3739 range
= PT_BYTE
- BEGV_BYTE
- startpos
;
3745 /* Update the fastmap now if not correct already. */
3746 if (fastmap
&& !bufp
->fastmap_accurate
)
3747 if (re_compile_fastmap (bufp
) == -2)
3750 /* See whether the pattern is anchored. */
3751 if (bufp
->buffer
[0] == begline
)
3755 gl_state
.object
= re_match_object
;
3757 int adjpos
= NILP (re_match_object
) || BUFFERP (re_match_object
);
3758 int charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (startpos
+ adjpos
);
3760 SETUP_SYNTAX_TABLE_FOR_OBJECT (re_match_object
, charpos
, 1);
3764 /* Loop through the string, looking for a place to start matching. */
3767 /* If the pattern is anchored,
3768 skip quickly past places we cannot match.
3769 We don't bother to treat startpos == 0 specially
3770 because that case doesn't repeat. */
3771 if (anchored_start
&& startpos
> 0)
3773 if (! (bufp
->newline_anchor
3774 && ((startpos
<= size1
? string1
[startpos
- 1]
3775 : string2
[startpos
- size1
- 1])
3780 /* If a fastmap is supplied, skip quickly over characters that
3781 cannot be the start of a match. If the pattern can match the
3782 null string, however, we don't need to skip characters; we want
3783 the first null string. */
3784 if (fastmap
&& startpos
< total_size
&& !bufp
->can_be_null
)
3786 register const char *d
;
3787 register unsigned int buf_ch
;
3789 d
= POS_ADDR_VSTRING (startpos
);
3791 if (range
> 0) /* Searching forwards. */
3793 register int lim
= 0;
3796 if (startpos
< size1
&& startpos
+ range
>= size1
)
3797 lim
= range
- (size1
- startpos
);
3799 /* Written out as an if-else to avoid testing `translate'
3801 if (RE_TRANSLATE_P (translate
))
3808 buf_ch
= STRING_CHAR_AND_LENGTH (d
, range
- lim
,
3811 buf_ch
= RE_TRANSLATE (translate
, buf_ch
);
3816 range
-= buf_charlen
;
3821 && !fastmap
[(unsigned char)
3822 RE_TRANSLATE (translate
, (unsigned char) *d
)])
3829 while (range
> lim
&& !fastmap
[(unsigned char) *d
])
3835 startpos
+= irange
- range
;
3837 else /* Searching backwards. */
3839 int room
= (size1
== 0 || startpos
>= size1
3840 ? size2
+ size1
- startpos
3841 : size1
- startpos
);
3843 buf_ch
= STRING_CHAR (d
, room
);
3844 if (RE_TRANSLATE_P (translate
))
3845 buf_ch
= RE_TRANSLATE (translate
, buf_ch
);
3847 if (! (buf_ch
>= 0400
3848 || fastmap
[buf_ch
]))
3853 /* If can't match the null string, and that's all we have left, fail. */
3854 if (range
>= 0 && startpos
== total_size
&& fastmap
3855 && !bufp
->can_be_null
)
3858 val
= re_match_2_internal (bufp
, string1
, size1
, string2
, size2
,
3859 startpos
, regs
, stop
);
3860 #ifndef REGEX_MALLOC
3877 /* Update STARTPOS to the next character boundary. */
3880 const unsigned char *p
3881 = (const unsigned char *) POS_ADDR_VSTRING (startpos
);
3882 const unsigned char *pend
3883 = (const unsigned char *) STOP_ADDR_VSTRING (startpos
);
3884 int len
= MULTIBYTE_FORM_LENGTH (p
, pend
- p
);
3902 /* Update STARTPOS to the previous character boundary. */
3905 const unsigned char *p
3906 = (const unsigned char *) POS_ADDR_VSTRING (startpos
);
3909 /* Find the head of multibyte form. */
3910 while (!CHAR_HEAD_P (*p
))
3915 if (MULTIBYTE_FORM_LENGTH (p
, len
+ 1) != (len
+ 1))
3932 /* Declarations and macros for re_match_2. */
3934 static int bcmp_translate ();
3935 static boolean
alt_match_null_string_p (),
3936 common_op_match_null_string_p (),
3937 group_match_null_string_p ();
3939 /* This converts PTR, a pointer into one of the search strings `string1'
3940 and `string2' into an offset from the beginning of that string. */
3941 #define POINTER_TO_OFFSET(ptr) \
3942 (FIRST_STRING_P (ptr) \
3943 ? ((regoff_t) ((ptr) - string1)) \
3944 : ((regoff_t) ((ptr) - string2 + size1)))
3946 /* Macros for dealing with the split strings in re_match_2. */
3948 #define MATCHING_IN_FIRST_STRING (dend == end_match_1)
3950 /* Call before fetching a character with *d. This switches over to
3951 string2 if necessary. */
3952 #define PREFETCH() \
3955 /* End of string2 => fail. */ \
3956 if (dend == end_match_2) \
3958 /* End of string1 => advance to string2. */ \
3960 dend = end_match_2; \
3964 /* Test if at very beginning or at very end of the virtual concatenation
3965 of `string1' and `string2'. If only one string, it's `string2'. */
3966 #define AT_STRINGS_BEG(d) ((d) == (size1 ? string1 : string2) || !size2)
3967 #define AT_STRINGS_END(d) ((d) == end2)
3970 /* Test if D points to a character which is word-constituent. We have
3971 two special cases to check for: if past the end of string1, look at
3972 the first character in string2; and if before the beginning of
3973 string2, look at the last character in string1. */
3974 #define WORDCHAR_P(d) \
3975 (SYNTAX ((d) == end1 ? *string2 \
3976 : (d) == string2 - 1 ? *(end1 - 1) : *(d)) \
3979 /* Disabled due to a compiler bug -- see comment at case wordbound */
3981 /* The comment at case wordbound is following one, but we don't use
3982 AT_WORD_BOUNDARY anymore to support multibyte form.
3984 The DEC Alpha C compiler 3.x generates incorrect code for the
3985 test WORDCHAR_P (d - 1) != WORDCHAR_P (d) in the expansion of
3986 AT_WORD_BOUNDARY, so this code is disabled. Expanding the
3987 macro and introducing temporary variables works around the bug. */
3990 /* Test if the character before D and the one at D differ with respect
3991 to being word-constituent. */
3992 #define AT_WORD_BOUNDARY(d) \
3993 (AT_STRINGS_BEG (d) || AT_STRINGS_END (d) \
3994 || WORDCHAR_P (d - 1) != WORDCHAR_P (d))
3997 /* Free everything we malloc. */
3998 #ifdef MATCH_MAY_ALLOCATE
3999 #define FREE_VAR(var) if (var) { REGEX_FREE (var); var = NULL; } else
4000 #define FREE_VARIABLES() \
4002 REGEX_FREE_STACK (fail_stack.stack); \
4003 FREE_VAR (regstart); \
4004 FREE_VAR (regend); \
4005 FREE_VAR (old_regstart); \
4006 FREE_VAR (old_regend); \
4007 FREE_VAR (best_regstart); \
4008 FREE_VAR (best_regend); \
4009 FREE_VAR (reg_info); \
4010 FREE_VAR (reg_dummy); \
4011 FREE_VAR (reg_info_dummy); \
4014 #define FREE_VARIABLES() ((void)0) /* Do nothing! But inhibit gcc warning. */
4015 #endif /* not MATCH_MAY_ALLOCATE */
4017 /* These values must meet several constraints. They must not be valid
4018 register values; since we have a limit of 255 registers (because
4019 we use only one byte in the pattern for the register number), we can
4020 use numbers larger than 255. They must differ by 1, because of
4021 NUM_FAILURE_ITEMS above. And the value for the lowest register must
4022 be larger than the value for the highest register, so we do not try
4023 to actually save any registers when none are active. */
4024 #define NO_HIGHEST_ACTIVE_REG (1 << BYTEWIDTH)
4025 #define NO_LOWEST_ACTIVE_REG (NO_HIGHEST_ACTIVE_REG + 1)
4027 /* Matching routines. */
4029 #ifndef emacs /* Emacs never uses this. */
4030 /* re_match is like re_match_2 except it takes only a single string. */
4033 re_match (bufp
, string
, size
, pos
, regs
)
4034 struct re_pattern_buffer
*bufp
;
4037 struct re_registers
*regs
;
4039 int result
= re_match_2_internal (bufp
, NULL
, 0, string
, size
,
4044 #endif /* not emacs */
4047 /* In Emacs, this is the string or buffer in which we
4048 are matching. It is used for looking up syntax properties. */
4049 Lisp_Object re_match_object
;
4052 /* re_match_2 matches the compiled pattern in BUFP against the
4053 the (virtual) concatenation of STRING1 and STRING2 (of length SIZE1
4054 and SIZE2, respectively). We start matching at POS, and stop
4057 If REGS is non-null and the `no_sub' field of BUFP is nonzero, we
4058 store offsets for the substring each group matched in REGS. See the
4059 documentation for exactly how many groups we fill.
4061 We return -1 if no match, -2 if an internal error (such as the
4062 failure stack overflowing). Otherwise, we return the length of the
4063 matched substring. */
4066 re_match_2 (bufp
, string1
, size1
, string2
, size2
, pos
, regs
, stop
)
4067 struct re_pattern_buffer
*bufp
;
4068 const char *string1
, *string2
;
4071 struct re_registers
*regs
;
4078 int adjpos
= NILP (re_match_object
) || BUFFERP (re_match_object
);
4079 gl_state
.object
= re_match_object
;
4080 charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (pos
+ adjpos
);
4081 SETUP_SYNTAX_TABLE_FOR_OBJECT (re_match_object
, charpos
, 1);
4084 result
= re_match_2_internal (bufp
, string1
, size1
, string2
, size2
,
4090 /* This is a separate function so that we can force an alloca cleanup
4093 re_match_2_internal (bufp
, string1
, size1
, string2
, size2
, pos
, regs
, stop
)
4094 struct re_pattern_buffer
*bufp
;
4095 const char *string1
, *string2
;
4098 struct re_registers
*regs
;
4101 /* General temporaries. */
4105 /* Just past the end of the corresponding string. */
4106 const char *end1
, *end2
;
4108 /* Pointers into string1 and string2, just past the last characters in
4109 each to consider matching. */
4110 const char *end_match_1
, *end_match_2
;
4112 /* Where we are in the data, and the end of the current string. */
4113 const char *d
, *dend
;
4115 /* Where we are in the pattern, and the end of the pattern. */
4116 unsigned char *p
= bufp
->buffer
;
4117 register unsigned char *pend
= p
+ bufp
->used
;
4119 /* Mark the opcode just after a start_memory, so we can test for an
4120 empty subpattern when we get to the stop_memory. */
4121 unsigned char *just_past_start_mem
= 0;
4123 /* We use this to map every character in the string. */
4124 RE_TRANSLATE_TYPE translate
= bufp
->translate
;
4126 /* Nonzero if we have to concern multibyte character. */
4127 int multibyte
= bufp
->multibyte
;
4129 /* Failure point stack. Each place that can handle a failure further
4130 down the line pushes a failure point on this stack. It consists of
4131 restart, regend, and reg_info for all registers corresponding to
4132 the subexpressions we're currently inside, plus the number of such
4133 registers, and, finally, two char *'s. The first char * is where
4134 to resume scanning the pattern; the second one is where to resume
4135 scanning the strings. If the latter is zero, the failure point is
4136 a ``dummy''; if a failure happens and the failure point is a dummy,
4137 it gets discarded and the next next one is tried. */
4138 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
4139 fail_stack_type fail_stack
;
4142 static unsigned failure_id
= 0;
4143 unsigned nfailure_points_pushed
= 0, nfailure_points_popped
= 0;
4146 /* This holds the pointer to the failure stack, when
4147 it is allocated relocatably. */
4148 fail_stack_elt_t
*failure_stack_ptr
;
4150 /* We fill all the registers internally, independent of what we
4151 return, for use in backreferences. The number here includes
4152 an element for register zero. */
4153 unsigned num_regs
= bufp
->re_nsub
+ 1;
4155 /* The currently active registers. */
4156 unsigned lowest_active_reg
= NO_LOWEST_ACTIVE_REG
;
4157 unsigned highest_active_reg
= NO_HIGHEST_ACTIVE_REG
;
4159 /* Information on the contents of registers. These are pointers into
4160 the input strings; they record just what was matched (on this
4161 attempt) by a subexpression part of the pattern, that is, the
4162 regnum-th regstart pointer points to where in the pattern we began
4163 matching and the regnum-th regend points to right after where we
4164 stopped matching the regnum-th subexpression. (The zeroth register
4165 keeps track of what the whole pattern matches.) */
4166 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
4167 const char **regstart
, **regend
;
4170 /* If a group that's operated upon by a repetition operator fails to
4171 match anything, then the register for its start will need to be
4172 restored because it will have been set to wherever in the string we
4173 are when we last see its open-group operator. Similarly for a
4175 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
4176 const char **old_regstart
, **old_regend
;
4179 /* The is_active field of reg_info helps us keep track of which (possibly
4180 nested) subexpressions we are currently in. The matched_something
4181 field of reg_info[reg_num] helps us tell whether or not we have
4182 matched any of the pattern so far this time through the reg_num-th
4183 subexpression. These two fields get reset each time through any
4184 loop their register is in. */
4185 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
4186 register_info_type
*reg_info
;
4189 /* The following record the register info as found in the above
4190 variables when we find a match better than any we've seen before.
4191 This happens as we backtrack through the failure points, which in
4192 turn happens only if we have not yet matched the entire string. */
4193 unsigned best_regs_set
= false;
4194 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
4195 const char **best_regstart
, **best_regend
;
4198 /* Logically, this is `best_regend[0]'. But we don't want to have to
4199 allocate space for that if we're not allocating space for anything
4200 else (see below). Also, we never need info about register 0 for
4201 any of the other register vectors, and it seems rather a kludge to
4202 treat `best_regend' differently than the rest. So we keep track of
4203 the end of the best match so far in a separate variable. We
4204 initialize this to NULL so that when we backtrack the first time
4205 and need to test it, it's not garbage. */
4206 const char *match_end
= NULL
;
4208 /* This helps SET_REGS_MATCHED avoid doing redundant work. */
4209 int set_regs_matched_done
= 0;
4211 /* Used when we pop values we don't care about. */
4212 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
4213 const char **reg_dummy
;
4214 register_info_type
*reg_info_dummy
;
4218 /* Counts the total number of registers pushed. */
4219 unsigned num_regs_pushed
= 0;
4222 DEBUG_PRINT1 ("\n\nEntering re_match_2.\n");
4226 #ifdef MATCH_MAY_ALLOCATE
4227 /* Do not bother to initialize all the register variables if there are
4228 no groups in the pattern, as it takes a fair amount of time. If
4229 there are groups, we include space for register 0 (the whole
4230 pattern), even though we never use it, since it simplifies the
4231 array indexing. We should fix this. */
4234 regstart
= REGEX_TALLOC (num_regs
, const char *);
4235 regend
= REGEX_TALLOC (num_regs
, const char *);
4236 old_regstart
= REGEX_TALLOC (num_regs
, const char *);
4237 old_regend
= REGEX_TALLOC (num_regs
, const char *);
4238 best_regstart
= REGEX_TALLOC (num_regs
, const char *);
4239 best_regend
= REGEX_TALLOC (num_regs
, const char *);
4240 reg_info
= REGEX_TALLOC (num_regs
, register_info_type
);
4241 reg_dummy
= REGEX_TALLOC (num_regs
, const char *);
4242 reg_info_dummy
= REGEX_TALLOC (num_regs
, register_info_type
);
4244 if (!(regstart
&& regend
&& old_regstart
&& old_regend
&& reg_info
4245 && best_regstart
&& best_regend
&& reg_dummy
&& reg_info_dummy
))
4253 /* We must initialize all our variables to NULL, so that
4254 `FREE_VARIABLES' doesn't try to free them. */
4255 regstart
= regend
= old_regstart
= old_regend
= best_regstart
4256 = best_regend
= reg_dummy
= NULL
;
4257 reg_info
= reg_info_dummy
= (register_info_type
*) NULL
;
4259 #endif /* MATCH_MAY_ALLOCATE */
4261 /* The starting position is bogus. */
4262 if (pos
< 0 || pos
> size1
+ size2
)
4268 /* Initialize subexpression text positions to -1 to mark ones that no
4269 start_memory/stop_memory has been seen for. Also initialize the
4270 register information struct. */
4271 for (mcnt
= 1; mcnt
< num_regs
; mcnt
++)
4273 regstart
[mcnt
] = regend
[mcnt
]
4274 = old_regstart
[mcnt
] = old_regend
[mcnt
] = REG_UNSET_VALUE
;
4276 REG_MATCH_NULL_STRING_P (reg_info
[mcnt
]) = MATCH_NULL_UNSET_VALUE
;
4277 IS_ACTIVE (reg_info
[mcnt
]) = 0;
4278 MATCHED_SOMETHING (reg_info
[mcnt
]) = 0;
4279 EVER_MATCHED_SOMETHING (reg_info
[mcnt
]) = 0;
4282 /* We move `string1' into `string2' if the latter's empty -- but not if
4283 `string1' is null. */
4284 if (size2
== 0 && string1
!= NULL
)
4291 end1
= string1
+ size1
;
4292 end2
= string2
+ size2
;
4294 /* Compute where to stop matching, within the two strings. */
4297 end_match_1
= string1
+ stop
;
4298 end_match_2
= string2
;
4303 end_match_2
= string2
+ stop
- size1
;
4306 /* `p' scans through the pattern as `d' scans through the data.
4307 `dend' is the end of the input string that `d' points within. `d'
4308 is advanced into the following input string whenever necessary, but
4309 this happens before fetching; therefore, at the beginning of the
4310 loop, `d' can be pointing at the end of a string, but it cannot
4312 if (size1
> 0 && pos
<= size1
)
4319 d
= string2
+ pos
- size1
;
4323 DEBUG_PRINT1 ("The compiled pattern is: ");
4324 DEBUG_PRINT_COMPILED_PATTERN (bufp
, p
, pend
);
4325 DEBUG_PRINT1 ("The string to match is: `");
4326 DEBUG_PRINT_DOUBLE_STRING (d
, string1
, size1
, string2
, size2
);
4327 DEBUG_PRINT1 ("'\n");
4329 /* This loops over pattern commands. It exits by returning from the
4330 function if the match is complete, or it drops through if the match
4331 fails at this starting point in the input data. */
4334 DEBUG_PRINT2 ("\n0x%x: ", p
);
4337 { /* End of pattern means we might have succeeded. */
4338 DEBUG_PRINT1 ("end of pattern ... ");
4340 /* If we haven't matched the entire string, and we want the
4341 longest match, try backtracking. */
4342 if (d
!= end_match_2
)
4344 /* 1 if this match ends in the same string (string1 or string2)
4345 as the best previous match. */
4346 boolean same_str_p
= (FIRST_STRING_P (match_end
)
4347 == MATCHING_IN_FIRST_STRING
);
4348 /* 1 if this match is the best seen so far. */
4349 boolean best_match_p
;
4351 /* AIX compiler got confused when this was combined
4352 with the previous declaration. */
4354 best_match_p
= d
> match_end
;
4356 best_match_p
= !MATCHING_IN_FIRST_STRING
;
4358 DEBUG_PRINT1 ("backtracking.\n");
4360 if (!FAIL_STACK_EMPTY ())
4361 { /* More failure points to try. */
4363 /* If exceeds best match so far, save it. */
4364 if (!best_regs_set
|| best_match_p
)
4366 best_regs_set
= true;
4369 DEBUG_PRINT1 ("\nSAVING match as best so far.\n");
4371 for (mcnt
= 1; mcnt
< num_regs
; mcnt
++)
4373 best_regstart
[mcnt
] = regstart
[mcnt
];
4374 best_regend
[mcnt
] = regend
[mcnt
];
4380 /* If no failure points, don't restore garbage. And if
4381 last match is real best match, don't restore second
4383 else if (best_regs_set
&& !best_match_p
)
4386 /* Restore best match. It may happen that `dend ==
4387 end_match_1' while the restored d is in string2.
4388 For example, the pattern `x.*y.*z' against the
4389 strings `x-' and `y-z-', if the two strings are
4390 not consecutive in memory. */
4391 DEBUG_PRINT1 ("Restoring best registers.\n");
4394 dend
= ((d
>= string1
&& d
<= end1
)
4395 ? end_match_1
: end_match_2
);
4397 for (mcnt
= 1; mcnt
< num_regs
; mcnt
++)
4399 regstart
[mcnt
] = best_regstart
[mcnt
];
4400 regend
[mcnt
] = best_regend
[mcnt
];
4403 } /* d != end_match_2 */
4406 DEBUG_PRINT1 ("Accepting match.\n");
4408 /* If caller wants register contents data back, do it. */
4409 if (regs
&& !bufp
->no_sub
)
4411 /* Have the register data arrays been allocated? */
4412 if (bufp
->regs_allocated
== REGS_UNALLOCATED
)
4413 { /* No. So allocate them with malloc. We need one
4414 extra element beyond `num_regs' for the `-1' marker
4416 regs
->num_regs
= MAX (RE_NREGS
, num_regs
+ 1);
4417 regs
->start
= TALLOC (regs
->num_regs
, regoff_t
);
4418 regs
->end
= TALLOC (regs
->num_regs
, regoff_t
);
4419 if (regs
->start
== NULL
|| regs
->end
== NULL
)
4424 bufp
->regs_allocated
= REGS_REALLOCATE
;
4426 else if (bufp
->regs_allocated
== REGS_REALLOCATE
)
4427 { /* Yes. If we need more elements than were already
4428 allocated, reallocate them. If we need fewer, just
4430 if (regs
->num_regs
< num_regs
+ 1)
4432 regs
->num_regs
= num_regs
+ 1;
4433 RETALLOC (regs
->start
, regs
->num_regs
, regoff_t
);
4434 RETALLOC (regs
->end
, regs
->num_regs
, regoff_t
);
4435 if (regs
->start
== NULL
|| regs
->end
== NULL
)
4444 /* These braces fend off a "empty body in an else-statement"
4445 warning under GCC when assert expands to nothing. */
4446 assert (bufp
->regs_allocated
== REGS_FIXED
);
4449 /* Convert the pointer data in `regstart' and `regend' to
4450 indices. Register zero has to be set differently,
4451 since we haven't kept track of any info for it. */
4452 if (regs
->num_regs
> 0)
4454 regs
->start
[0] = pos
;
4455 regs
->end
[0] = (MATCHING_IN_FIRST_STRING
4456 ? ((regoff_t
) (d
- string1
))
4457 : ((regoff_t
) (d
- string2
+ size1
)));
4460 /* Go through the first `min (num_regs, regs->num_regs)'
4461 registers, since that is all we initialized. */
4462 for (mcnt
= 1; mcnt
< MIN (num_regs
, regs
->num_regs
); mcnt
++)
4464 if (REG_UNSET (regstart
[mcnt
]) || REG_UNSET (regend
[mcnt
]))
4465 regs
->start
[mcnt
] = regs
->end
[mcnt
] = -1;
4469 = (regoff_t
) POINTER_TO_OFFSET (regstart
[mcnt
]);
4471 = (regoff_t
) POINTER_TO_OFFSET (regend
[mcnt
]);
4475 /* If the regs structure we return has more elements than
4476 were in the pattern, set the extra elements to -1. If
4477 we (re)allocated the registers, this is the case,
4478 because we always allocate enough to have at least one
4480 for (mcnt
= num_regs
; mcnt
< regs
->num_regs
; mcnt
++)
4481 regs
->start
[mcnt
] = regs
->end
[mcnt
] = -1;
4482 } /* regs && !bufp->no_sub */
4484 DEBUG_PRINT4 ("%u failure points pushed, %u popped (%u remain).\n",
4485 nfailure_points_pushed
, nfailure_points_popped
,
4486 nfailure_points_pushed
- nfailure_points_popped
);
4487 DEBUG_PRINT2 ("%u registers pushed.\n", num_regs_pushed
);
4489 mcnt
= d
- pos
- (MATCHING_IN_FIRST_STRING
4493 DEBUG_PRINT2 ("Returning %d from re_match_2.\n", mcnt
);
4499 /* Otherwise match next pattern command. */
4500 switch (SWITCH_ENUM_CAST ((re_opcode_t
) *p
++))
4502 /* Ignore these. Used to ignore the n of succeed_n's which
4503 currently have n == 0. */
4505 DEBUG_PRINT1 ("EXECUTING no_op.\n");
4509 DEBUG_PRINT1 ("EXECUTING succeed.\n");
4512 /* Match the next n pattern characters exactly. The following
4513 byte in the pattern defines n, and the n bytes after that
4514 are the characters to match. */
4517 DEBUG_PRINT2 ("EXECUTING exactn %d.\n", mcnt
);
4519 /* This is written out as an if-else so we don't waste time
4520 testing `translate' inside the loop. */
4521 if (RE_TRANSLATE_P (translate
))
4527 int pat_charlen
, buf_charlen
;
4528 unsigned int pat_ch
, buf_ch
;
4531 pat_ch
= STRING_CHAR_AND_LENGTH (p
, pend
- p
, pat_charlen
);
4532 buf_ch
= STRING_CHAR_AND_LENGTH (d
, dend
- d
, buf_charlen
);
4534 if (RE_TRANSLATE (translate
, buf_ch
)
4540 mcnt
-= pat_charlen
;
4544 #endif /* not emacs */
4548 if ((unsigned char) RE_TRANSLATE (translate
, (unsigned char) *d
)
4549 != (unsigned char) *p
++)
4560 if (*d
++ != (char) *p
++) goto fail
;
4564 SET_REGS_MATCHED ();
4568 /* Match any character except possibly a newline or a null. */
4572 unsigned int buf_ch
;
4574 DEBUG_PRINT1 ("EXECUTING anychar.\n");
4580 buf_ch
= STRING_CHAR_AND_LENGTH (d
, dend
- d
, buf_charlen
);
4582 #endif /* not emacs */
4584 buf_ch
= (unsigned char) *d
;
4588 buf_ch
= TRANSLATE (buf_ch
);
4590 if ((!(bufp
->syntax
& RE_DOT_NEWLINE
)
4592 || ((bufp
->syntax
& RE_DOT_NOT_NULL
)
4593 && buf_ch
== '\000'))
4596 SET_REGS_MATCHED ();
4597 DEBUG_PRINT2 (" Matched `%d'.\n", *d
);
4606 register unsigned int c
;
4607 boolean
not = (re_opcode_t
) *(p
- 1) == charset_not
;
4610 /* Start of actual range_table, or end of bitmap if there is no
4612 unsigned char *range_table
;
4614 /* Nonzero if there is range table. */
4615 int range_table_exists
;
4617 /* Number of ranges of range table. Not in bytes. */
4620 DEBUG_PRINT2 ("EXECUTING charset%s.\n", not ? "_not" : "");
4623 c
= (unsigned char) *d
;
4625 range_table
= CHARSET_RANGE_TABLE (&p
[-1]); /* Past the bitmap. */
4626 range_table_exists
= CHARSET_RANGE_TABLE_EXISTS_P (&p
[-1]);
4627 if (range_table_exists
)
4628 EXTRACT_NUMBER_AND_INCR (count
, range_table
);
4632 if (multibyte
&& BASE_LEADING_CODE_P (c
))
4633 c
= STRING_CHAR_AND_LENGTH (d
, dend
- d
, len
);
4635 if (SINGLE_BYTE_CHAR_P (c
))
4636 { /* Lookup bitmap. */
4637 c
= TRANSLATE (c
); /* The character to match. */
4640 /* Cast to `unsigned' instead of `unsigned char' in
4641 case the bit list is a full 32 bytes long. */
4642 if (c
< (unsigned) (CHARSET_BITMAP_SIZE (&p
[-1]) * BYTEWIDTH
)
4643 && p
[1 + c
/ BYTEWIDTH
] & (1 << (c
% BYTEWIDTH
)))
4646 else if (range_table_exists
)
4647 CHARSET_LOOKUP_RANGE_TABLE_RAW (not, c
, range_table
, count
);
4649 p
= CHARSET_RANGE_TABLE_END (range_table
, count
);
4651 if (!not) goto fail
;
4653 SET_REGS_MATCHED ();
4659 /* The beginning of a group is represented by start_memory.
4660 The arguments are the register number in the next byte, and the
4661 number of groups inner to this one in the next. The text
4662 matched within the group is recorded (in the internal
4663 registers data structure) under the register number. */
4665 DEBUG_PRINT3 ("EXECUTING start_memory %d (%d):\n", *p
, p
[1]);
4667 /* Find out if this group can match the empty string. */
4668 p1
= p
; /* To send to group_match_null_string_p. */
4670 if (REG_MATCH_NULL_STRING_P (reg_info
[*p
]) == MATCH_NULL_UNSET_VALUE
)
4671 REG_MATCH_NULL_STRING_P (reg_info
[*p
])
4672 = group_match_null_string_p (&p1
, pend
, reg_info
);
4674 /* Save the position in the string where we were the last time
4675 we were at this open-group operator in case the group is
4676 operated upon by a repetition operator, e.g., with `(a*)*b'
4677 against `ab'; then we want to ignore where we are now in
4678 the string in case this attempt to match fails. */
4679 old_regstart
[*p
] = REG_MATCH_NULL_STRING_P (reg_info
[*p
])
4680 ? REG_UNSET (regstart
[*p
]) ? d
: regstart
[*p
]
4682 DEBUG_PRINT2 (" old_regstart: %d\n",
4683 POINTER_TO_OFFSET (old_regstart
[*p
]));
4686 DEBUG_PRINT2 (" regstart: %d\n", POINTER_TO_OFFSET (regstart
[*p
]));
4688 IS_ACTIVE (reg_info
[*p
]) = 1;
4689 MATCHED_SOMETHING (reg_info
[*p
]) = 0;
4691 /* Clear this whenever we change the register activity status. */
4692 set_regs_matched_done
= 0;
4694 /* This is the new highest active register. */
4695 highest_active_reg
= *p
;
4697 /* If nothing was active before, this is the new lowest active
4699 if (lowest_active_reg
== NO_LOWEST_ACTIVE_REG
)
4700 lowest_active_reg
= *p
;
4702 /* Move past the register number and inner group count. */
4704 just_past_start_mem
= p
;
4709 /* The stop_memory opcode represents the end of a group. Its
4710 arguments are the same as start_memory's: the register
4711 number, and the number of inner groups. */
4713 DEBUG_PRINT3 ("EXECUTING stop_memory %d (%d):\n", *p
, p
[1]);
4715 /* We need to save the string position the last time we were at
4716 this close-group operator in case the group is operated
4717 upon by a repetition operator, e.g., with `((a*)*(b*)*)*'
4718 against `aba'; then we want to ignore where we are now in
4719 the string in case this attempt to match fails. */
4720 old_regend
[*p
] = REG_MATCH_NULL_STRING_P (reg_info
[*p
])
4721 ? REG_UNSET (regend
[*p
]) ? d
: regend
[*p
]
4723 DEBUG_PRINT2 (" old_regend: %d\n",
4724 POINTER_TO_OFFSET (old_regend
[*p
]));
4727 DEBUG_PRINT2 (" regend: %d\n", POINTER_TO_OFFSET (regend
[*p
]));
4729 /* This register isn't active anymore. */
4730 IS_ACTIVE (reg_info
[*p
]) = 0;
4732 /* Clear this whenever we change the register activity status. */
4733 set_regs_matched_done
= 0;
4735 /* If this was the only register active, nothing is active
4737 if (lowest_active_reg
== highest_active_reg
)
4739 lowest_active_reg
= NO_LOWEST_ACTIVE_REG
;
4740 highest_active_reg
= NO_HIGHEST_ACTIVE_REG
;
4743 { /* We must scan for the new highest active register, since
4744 it isn't necessarily one less than now: consider
4745 (a(b)c(d(e)f)g). When group 3 ends, after the f), the
4746 new highest active register is 1. */
4747 unsigned char r
= *p
- 1;
4748 while (r
> 0 && !IS_ACTIVE (reg_info
[r
]))
4751 /* If we end up at register zero, that means that we saved
4752 the registers as the result of an `on_failure_jump', not
4753 a `start_memory', and we jumped to past the innermost
4754 `stop_memory'. For example, in ((.)*) we save
4755 registers 1 and 2 as a result of the *, but when we pop
4756 back to the second ), we are at the stop_memory 1.
4757 Thus, nothing is active. */
4760 lowest_active_reg
= NO_LOWEST_ACTIVE_REG
;
4761 highest_active_reg
= NO_HIGHEST_ACTIVE_REG
;
4764 highest_active_reg
= r
;
4767 /* If just failed to match something this time around with a
4768 group that's operated on by a repetition operator, try to
4769 force exit from the ``loop'', and restore the register
4770 information for this group that we had before trying this
4772 if ((!MATCHED_SOMETHING (reg_info
[*p
])
4773 || just_past_start_mem
== p
- 1)
4776 boolean is_a_jump_n
= false;
4780 switch ((re_opcode_t
) *p1
++)
4784 case pop_failure_jump
:
4785 case maybe_pop_jump
:
4787 case dummy_failure_jump
:
4788 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
4798 /* If the next operation is a jump backwards in the pattern
4799 to an on_failure_jump right before the start_memory
4800 corresponding to this stop_memory, exit from the loop
4801 by forcing a failure after pushing on the stack the
4802 on_failure_jump's jump in the pattern, and d. */
4803 if (mcnt
< 0 && (re_opcode_t
) *p1
== on_failure_jump
4804 && (re_opcode_t
) p1
[3] == start_memory
&& p1
[4] == *p
)
4806 /* If this group ever matched anything, then restore
4807 what its registers were before trying this last
4808 failed match, e.g., with `(a*)*b' against `ab' for
4809 regstart[1], and, e.g., with `((a*)*(b*)*)*'
4810 against `aba' for regend[3].
4812 Also restore the registers for inner groups for,
4813 e.g., `((a*)(b*))*' against `aba' (register 3 would
4814 otherwise get trashed). */
4816 if (EVER_MATCHED_SOMETHING (reg_info
[*p
]))
4820 EVER_MATCHED_SOMETHING (reg_info
[*p
]) = 0;
4822 /* Restore this and inner groups' (if any) registers. */
4823 for (r
= *p
; r
< *p
+ *(p
+ 1); r
++)
4825 regstart
[r
] = old_regstart
[r
];
4827 /* xx why this test? */
4828 if (old_regend
[r
] >= regstart
[r
])
4829 regend
[r
] = old_regend
[r
];
4833 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
4834 PUSH_FAILURE_POINT (p1
+ mcnt
, d
, -2);
4840 /* Move past the register number and the inner group count. */
4845 /* \<digit> has been turned into a `duplicate' command which is
4846 followed by the numeric value of <digit> as the register number. */
4849 register const char *d2
, *dend2
;
4850 int regno
= *p
++; /* Get which register to match against. */
4851 DEBUG_PRINT2 ("EXECUTING duplicate %d.\n", regno
);
4853 /* Can't back reference a group which we've never matched. */
4854 if (REG_UNSET (regstart
[regno
]) || REG_UNSET (regend
[regno
]))
4857 /* Where in input to try to start matching. */
4858 d2
= regstart
[regno
];
4860 /* Where to stop matching; if both the place to start and
4861 the place to stop matching are in the same string, then
4862 set to the place to stop, otherwise, for now have to use
4863 the end of the first string. */
4865 dend2
= ((FIRST_STRING_P (regstart
[regno
])
4866 == FIRST_STRING_P (regend
[regno
]))
4867 ? regend
[regno
] : end_match_1
);
4870 /* If necessary, advance to next segment in register
4874 if (dend2
== end_match_2
) break;
4875 if (dend2
== regend
[regno
]) break;
4877 /* End of string1 => advance to string2. */
4879 dend2
= regend
[regno
];
4881 /* At end of register contents => success */
4882 if (d2
== dend2
) break;
4884 /* If necessary, advance to next segment in data. */
4887 /* How many characters left in this segment to match. */
4890 /* Want how many consecutive characters we can match in
4891 one shot, so, if necessary, adjust the count. */
4892 if (mcnt
> dend2
- d2
)
4895 /* Compare that many; failure if mismatch, else move
4897 if (RE_TRANSLATE_P (translate
)
4898 ? bcmp_translate (d
, d2
, mcnt
, translate
)
4899 : bcmp (d
, d2
, mcnt
))
4901 d
+= mcnt
, d2
+= mcnt
;
4903 /* Do this because we've match some characters. */
4904 SET_REGS_MATCHED ();
4910 /* begline matches the empty string at the beginning of the string
4911 (unless `not_bol' is set in `bufp'), and, if
4912 `newline_anchor' is set, after newlines. */
4914 DEBUG_PRINT1 ("EXECUTING begline.\n");
4916 if (AT_STRINGS_BEG (d
))
4918 if (!bufp
->not_bol
) break;
4920 else if (d
[-1] == '\n' && bufp
->newline_anchor
)
4924 /* In all other cases, we fail. */
4928 /* endline is the dual of begline. */
4930 DEBUG_PRINT1 ("EXECUTING endline.\n");
4932 if (AT_STRINGS_END (d
))
4934 if (!bufp
->not_eol
) break;
4937 /* We have to ``prefetch'' the next character. */
4938 else if ((d
== end1
? *string2
: *d
) == '\n'
4939 && bufp
->newline_anchor
)
4946 /* Match at the very beginning of the data. */
4948 DEBUG_PRINT1 ("EXECUTING begbuf.\n");
4949 if (AT_STRINGS_BEG (d
))
4954 /* Match at the very end of the data. */
4956 DEBUG_PRINT1 ("EXECUTING endbuf.\n");
4957 if (AT_STRINGS_END (d
))
4962 /* on_failure_keep_string_jump is used to optimize `.*\n'. It
4963 pushes NULL as the value for the string on the stack. Then
4964 `pop_failure_point' will keep the current value for the
4965 string, instead of restoring it. To see why, consider
4966 matching `foo\nbar' against `.*\n'. The .* matches the foo;
4967 then the . fails against the \n. But the next thing we want
4968 to do is match the \n against the \n; if we restored the
4969 string value, we would be back at the foo.
4971 Because this is used only in specific cases, we don't need to
4972 check all the things that `on_failure_jump' does, to make
4973 sure the right things get saved on the stack. Hence we don't
4974 share its code. The only reason to push anything on the
4975 stack at all is that otherwise we would have to change
4976 `anychar's code to do something besides goto fail in this
4977 case; that seems worse than this. */
4978 case on_failure_keep_string_jump
:
4979 DEBUG_PRINT1 ("EXECUTING on_failure_keep_string_jump");
4981 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
4982 DEBUG_PRINT3 (" %d (to 0x%x):\n", mcnt
, p
+ mcnt
);
4984 PUSH_FAILURE_POINT (p
+ mcnt
, NULL
, -2);
4988 /* Uses of on_failure_jump:
4990 Each alternative starts with an on_failure_jump that points
4991 to the beginning of the next alternative. Each alternative
4992 except the last ends with a jump that in effect jumps past
4993 the rest of the alternatives. (They really jump to the
4994 ending jump of the following alternative, because tensioning
4995 these jumps is a hassle.)
4997 Repeats start with an on_failure_jump that points past both
4998 the repetition text and either the following jump or
4999 pop_failure_jump back to this on_failure_jump. */
5000 case on_failure_jump
:
5002 DEBUG_PRINT1 ("EXECUTING on_failure_jump");
5004 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5005 DEBUG_PRINT3 (" %d (to 0x%x)", mcnt
, p
+ mcnt
);
5007 /* If this on_failure_jump comes right before a group (i.e.,
5008 the original * applied to a group), save the information
5009 for that group and all inner ones, so that if we fail back
5010 to this point, the group's information will be correct.
5011 For example, in \(a*\)*\1, we need the preceding group,
5012 and in \(zz\(a*\)b*\)\2, we need the inner group. */
5014 /* We can't use `p' to check ahead because we push
5015 a failure point to `p + mcnt' after we do this. */
5018 /* We need to skip no_op's before we look for the
5019 start_memory in case this on_failure_jump is happening as
5020 the result of a completed succeed_n, as in \(a\)\{1,3\}b\1
5022 while (p1
< pend
&& (re_opcode_t
) *p1
== no_op
)
5025 if (p1
< pend
&& (re_opcode_t
) *p1
== start_memory
)
5027 /* We have a new highest active register now. This will
5028 get reset at the start_memory we are about to get to,
5029 but we will have saved all the registers relevant to
5030 this repetition op, as described above. */
5031 highest_active_reg
= *(p1
+ 1) + *(p1
+ 2);
5032 if (lowest_active_reg
== NO_LOWEST_ACTIVE_REG
)
5033 lowest_active_reg
= *(p1
+ 1);
5036 DEBUG_PRINT1 (":\n");
5037 PUSH_FAILURE_POINT (p
+ mcnt
, d
, -2);
5041 /* A smart repeat ends with `maybe_pop_jump'.
5042 We change it to either `pop_failure_jump' or `jump'. */
5043 case maybe_pop_jump
:
5044 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5045 DEBUG_PRINT2 ("EXECUTING maybe_pop_jump %d.\n", mcnt
);
5047 register unsigned char *p2
= p
;
5049 /* Compare the beginning of the repeat with what in the
5050 pattern follows its end. If we can establish that there
5051 is nothing that they would both match, i.e., that we
5052 would have to backtrack because of (as in, e.g., `a*a')
5053 then we can change to pop_failure_jump, because we'll
5054 never have to backtrack.
5056 This is not true in the case of alternatives: in
5057 `(a|ab)*' we do need to backtrack to the `ab' alternative
5058 (e.g., if the string was `ab'). But instead of trying to
5059 detect that here, the alternative has put on a dummy
5060 failure point which is what we will end up popping. */
5062 /* Skip over open/close-group commands.
5063 If what follows this loop is a ...+ construct,
5064 look at what begins its body, since we will have to
5065 match at least one of that. */
5069 && ((re_opcode_t
) *p2
== stop_memory
5070 || (re_opcode_t
) *p2
== start_memory
))
5072 else if (p2
+ 6 < pend
5073 && (re_opcode_t
) *p2
== dummy_failure_jump
)
5080 /* p1[0] ... p1[2] are the `on_failure_jump' corresponding
5081 to the `maybe_finalize_jump' of this case. Examine what
5084 /* If we're at the end of the pattern, we can change. */
5087 /* Consider what happens when matching ":\(.*\)"
5088 against ":/". I don't really understand this code
5090 p
[-3] = (unsigned char) pop_failure_jump
;
5092 (" End of pattern: change to `pop_failure_jump'.\n");
5095 else if ((re_opcode_t
) *p2
== exactn
5096 || (bufp
->newline_anchor
&& (re_opcode_t
) *p2
== endline
))
5098 register unsigned int c
5099 = *p2
== (unsigned char) endline
? '\n' : p2
[2];
5101 if ((re_opcode_t
) p1
[3] == exactn
)
5103 if (!(multibyte
/* && (c != '\n') */
5104 && BASE_LEADING_CODE_P (c
))
5106 : (STRING_CHAR (&p2
[2], pend
- &p2
[2])
5107 != STRING_CHAR (&p1
[5], pend
- &p1
[5])))
5109 p
[-3] = (unsigned char) pop_failure_jump
;
5110 DEBUG_PRINT3 (" %c != %c => pop_failure_jump.\n",
5115 else if ((re_opcode_t
) p1
[3] == charset
5116 || (re_opcode_t
) p1
[3] == charset_not
)
5118 int not = (re_opcode_t
) p1
[3] == charset_not
;
5120 if (multibyte
/* && (c != '\n') */
5121 && BASE_LEADING_CODE_P (c
))
5122 c
= STRING_CHAR (&p2
[2], pend
- &p2
[2]);
5124 /* Test if C is listed in charset (or charset_not)
5126 if (SINGLE_BYTE_CHAR_P (c
))
5128 if (c
< CHARSET_BITMAP_SIZE (&p1
[3]) * BYTEWIDTH
5129 && p1
[5 + c
/ BYTEWIDTH
] & (1 << (c
% BYTEWIDTH
)))
5132 else if (CHARSET_RANGE_TABLE_EXISTS_P (&p1
[3]))
5133 CHARSET_LOOKUP_RANGE_TABLE (not, c
, &p1
[3]);
5135 /* `not' is equal to 1 if c would match, which means
5136 that we can't change to pop_failure_jump. */
5139 p
[-3] = (unsigned char) pop_failure_jump
;
5140 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
5144 else if ((re_opcode_t
) *p2
== charset
)
5146 if ((re_opcode_t
) p1
[3] == exactn
)
5148 register unsigned int c
= p1
[5];
5151 if (multibyte
&& BASE_LEADING_CODE_P (c
))
5152 c
= STRING_CHAR (&p1
[5], pend
- &p1
[5]);
5154 /* Test if C is listed in charset at `p2'. */
5155 if (SINGLE_BYTE_CHAR_P (c
))
5157 if (c
< CHARSET_BITMAP_SIZE (p2
) * BYTEWIDTH
5158 && (p2
[2 + c
/ BYTEWIDTH
]
5159 & (1 << (c
% BYTEWIDTH
))))
5162 else if (CHARSET_RANGE_TABLE_EXISTS_P (p2
))
5163 CHARSET_LOOKUP_RANGE_TABLE (not, c
, p2
);
5167 p
[-3] = (unsigned char) pop_failure_jump
;
5168 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
5172 /* It is hard to list up all the character in charset
5173 P2 if it includes multibyte character. Give up in
5175 else if (!multibyte
|| !CHARSET_RANGE_TABLE_EXISTS_P (p2
))
5177 /* Now, we are sure that P2 has no range table.
5178 So, for the size of bitmap in P2, `p2[1]' is
5179 enough. But P1 may have range table, so the
5180 size of bitmap table of P1 is extracted by
5181 using macro `CHARSET_BITMAP_SIZE'.
5183 Since we know that all the character listed in
5184 P2 is ASCII, it is enough to test only bitmap
5187 if ((re_opcode_t
) p1
[3] == charset_not
)
5190 /* We win if the charset_not inside the loop lists
5191 every character listed in the charset after. */
5192 for (idx
= 0; idx
< (int) p2
[1]; idx
++)
5193 if (! (p2
[2 + idx
] == 0
5194 || (idx
< CHARSET_BITMAP_SIZE (&p1
[3])
5195 && ((p2
[2 + idx
] & ~ p1
[5 + idx
]) == 0))))
5200 p
[-3] = (unsigned char) pop_failure_jump
;
5201 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
5204 else if ((re_opcode_t
) p1
[3] == charset
)
5207 /* We win if the charset inside the loop
5208 has no overlap with the one after the loop. */
5211 && idx
< CHARSET_BITMAP_SIZE (&p1
[3]));
5213 if ((p2
[2 + idx
] & p1
[5 + idx
]) != 0)
5217 || idx
== CHARSET_BITMAP_SIZE (&p1
[3]))
5219 p
[-3] = (unsigned char) pop_failure_jump
;
5220 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
5226 p
-= 2; /* Point at relative address again. */
5227 if ((re_opcode_t
) p
[-1] != pop_failure_jump
)
5229 p
[-1] = (unsigned char) jump
;
5230 DEBUG_PRINT1 (" Match => jump.\n");
5231 goto unconditional_jump
;
5233 /* Note fall through. */
5236 /* The end of a simple repeat has a pop_failure_jump back to
5237 its matching on_failure_jump, where the latter will push a
5238 failure point. The pop_failure_jump takes off failure
5239 points put on by this pop_failure_jump's matching
5240 on_failure_jump; we got through the pattern to here from the
5241 matching on_failure_jump, so didn't fail. */
5242 case pop_failure_jump
:
5244 /* We need to pass separate storage for the lowest and
5245 highest registers, even though we don't care about the
5246 actual values. Otherwise, we will restore only one
5247 register from the stack, since lowest will == highest in
5248 `pop_failure_point'. */
5249 unsigned dummy_low_reg
, dummy_high_reg
;
5250 unsigned char *pdummy
;
5253 DEBUG_PRINT1 ("EXECUTING pop_failure_jump.\n");
5254 POP_FAILURE_POINT (sdummy
, pdummy
,
5255 dummy_low_reg
, dummy_high_reg
,
5256 reg_dummy
, reg_dummy
, reg_info_dummy
);
5258 /* Note fall through. */
5261 /* Unconditionally jump (without popping any failure points). */
5264 EXTRACT_NUMBER_AND_INCR (mcnt
, p
); /* Get the amount to jump. */
5265 DEBUG_PRINT2 ("EXECUTING jump %d ", mcnt
);
5266 p
+= mcnt
; /* Do the jump. */
5267 DEBUG_PRINT2 ("(to 0x%x).\n", p
);
5271 /* We need this opcode so we can detect where alternatives end
5272 in `group_match_null_string_p' et al. */
5274 DEBUG_PRINT1 ("EXECUTING jump_past_alt.\n");
5275 goto unconditional_jump
;
5278 /* Normally, the on_failure_jump pushes a failure point, which
5279 then gets popped at pop_failure_jump. We will end up at
5280 pop_failure_jump, also, and with a pattern of, say, `a+', we
5281 are skipping over the on_failure_jump, so we have to push
5282 something meaningless for pop_failure_jump to pop. */
5283 case dummy_failure_jump
:
5284 DEBUG_PRINT1 ("EXECUTING dummy_failure_jump.\n");
5285 /* It doesn't matter what we push for the string here. What
5286 the code at `fail' tests is the value for the pattern. */
5287 PUSH_FAILURE_POINT (0, 0, -2);
5288 goto unconditional_jump
;
5291 /* At the end of an alternative, we need to push a dummy failure
5292 point in case we are followed by a `pop_failure_jump', because
5293 we don't want the failure point for the alternative to be
5294 popped. For example, matching `(a|ab)*' against `aab'
5295 requires that we match the `ab' alternative. */
5296 case push_dummy_failure
:
5297 DEBUG_PRINT1 ("EXECUTING push_dummy_failure.\n");
5298 /* See comments just above at `dummy_failure_jump' about the
5300 PUSH_FAILURE_POINT (0, 0, -2);
5303 /* Have to succeed matching what follows at least n times.
5304 After that, handle like `on_failure_jump'. */
5306 EXTRACT_NUMBER (mcnt
, p
+ 2);
5307 DEBUG_PRINT2 ("EXECUTING succeed_n %d.\n", mcnt
);
5310 /* Originally, this is how many times we HAVE to succeed. */
5315 STORE_NUMBER_AND_INCR (p
, mcnt
);
5316 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p
, mcnt
);
5320 DEBUG_PRINT2 (" Setting two bytes from 0x%x to no_op.\n", p
+2);
5321 p
[2] = (unsigned char) no_op
;
5322 p
[3] = (unsigned char) no_op
;
5328 EXTRACT_NUMBER (mcnt
, p
+ 2);
5329 DEBUG_PRINT2 ("EXECUTING jump_n %d.\n", mcnt
);
5331 /* Originally, this is how many times we CAN jump. */
5335 STORE_NUMBER (p
+ 2, mcnt
);
5336 goto unconditional_jump
;
5338 /* If don't have to jump any more, skip over the rest of command. */
5345 DEBUG_PRINT1 ("EXECUTING set_number_at.\n");
5347 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5349 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
5350 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p1
, mcnt
);
5351 STORE_NUMBER (p1
, mcnt
);
5356 DEBUG_PRINT1 ("EXECUTING wordbound.\n");
5358 /* We SUCCEED in one of the following cases: */
5360 /* Case 1: D is at the beginning or the end of string. */
5361 if (AT_STRINGS_BEG (d
) || AT_STRINGS_END (d
))
5365 /* C1 is the character before D, S1 is the syntax of C1, C2
5366 is the character at D, and S2 is the syntax of C2. */
5368 int pos1
= PTR_TO_OFFSET (d
- 1);
5371 GET_CHAR_BEFORE_2 (c1
, d
, string1
, end1
, string2
, end2
);
5372 GET_CHAR_AFTER_2 (c2
, d
, string1
, end1
, string2
, end2
);
5374 charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (pos1
);
5375 UPDATE_SYNTAX_TABLE (charpos
);
5379 UPDATE_SYNTAX_TABLE_FORWARD (charpos
+ 1);
5383 if (/* Case 2: Only one of S1 and S2 is Sword. */
5384 ((s1
== Sword
) != (s2
== Sword
))
5385 /* Case 3: Both of S1 and S2 are Sword, and macro
5386 WORD_BOUNDARY_P (C1, C2) returns nonzero. */
5387 || ((s1
== Sword
) && WORD_BOUNDARY_P (c1
, c2
)))
5393 DEBUG_PRINT1 ("EXECUTING notwordbound.\n");
5395 /* We FAIL in one of the following cases: */
5397 /* Case 1: D is at the beginning or the end of string. */
5398 if (AT_STRINGS_BEG (d
) || AT_STRINGS_END (d
))
5402 /* C1 is the character before D, S1 is the syntax of C1, C2
5403 is the character at D, and S2 is the syntax of C2. */
5405 int pos1
= PTR_TO_OFFSET (d
- 1);
5408 GET_CHAR_BEFORE_2 (c1
, d
, string1
, end1
, string2
, end2
);
5409 GET_CHAR_AFTER_2 (c2
, d
, string1
, end1
, string2
, end2
);
5411 charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (pos1
);
5412 UPDATE_SYNTAX_TABLE (charpos
);
5416 UPDATE_SYNTAX_TABLE_FORWARD (charpos
+ 1);
5420 if (/* Case 2: Only one of S1 and S2 is Sword. */
5421 ((s1
== Sword
) != (s2
== Sword
))
5422 /* Case 3: Both of S1 and S2 are Sword, and macro
5423 WORD_BOUNDARY_P (C1, C2) returns nonzero. */
5424 || ((s1
== Sword
) && WORD_BOUNDARY_P (c1
, c2
)))
5430 DEBUG_PRINT1 ("EXECUTING wordbeg.\n");
5432 /* We FAIL in one of the following cases: */
5434 /* Case 1: D is at the end of string. */
5435 if (AT_STRINGS_END (d
))
5439 /* C1 is the character before D, S1 is the syntax of C1, C2
5440 is the character at D, and S2 is the syntax of C2. */
5442 int pos1
= PTR_TO_OFFSET (d
);
5445 GET_CHAR_AFTER_2 (c2
, d
, string1
, end1
, string2
, end2
);
5447 charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (pos1
);
5448 UPDATE_SYNTAX_TABLE (charpos
);
5452 /* Case 2: S2 is not Sword. */
5456 /* Case 3: D is not at the beginning of string ... */
5457 if (!AT_STRINGS_BEG (d
))
5459 GET_CHAR_BEFORE_2 (c1
, d
, string1
, end1
, string2
, end2
);
5461 UPDATE_SYNTAX_TABLE_BACKWARD (charpos
- 1);
5465 /* ... and S1 is Sword, and WORD_BOUNDARY_P (C1, C2)
5467 if ((s1
== Sword
) && !WORD_BOUNDARY_P (c1
, c2
))
5474 DEBUG_PRINT1 ("EXECUTING wordend.\n");
5476 /* We FAIL in one of the following cases: */
5478 /* Case 1: D is at the beginning of string. */
5479 if (AT_STRINGS_BEG (d
))
5483 /* C1 is the character before D, S1 is the syntax of C1, C2
5484 is the character at D, and S2 is the syntax of C2. */
5486 int pos1
= PTR_TO_OFFSET (d
);
5489 GET_CHAR_BEFORE_2 (c1
, d
, string1
, end1
, string2
, end2
);
5491 charpos
= SYNTAX_TABLE_BYTE_TO_CHAR (pos1
- 1);
5492 UPDATE_SYNTAX_TABLE (charpos
);
5496 /* Case 2: S1 is not Sword. */
5500 /* Case 3: D is not at the end of string ... */
5501 if (!AT_STRINGS_END (d
))
5503 GET_CHAR_AFTER_2 (c2
, d
, string1
, end1
, string2
, end2
);
5505 UPDATE_SYNTAX_TABLE_FORWARD (charpos
);
5509 /* ... and S2 is Sword, and WORD_BOUNDARY_P (C1, C2)
5511 if ((s2
== Sword
) && !WORD_BOUNDARY_P (c1
, c2
))
5519 DEBUG_PRINT1 ("EXECUTING before_dot.\n");
5520 if (PTR_BYTE_POS ((unsigned char *) d
) >= PT_BYTE
)
5525 DEBUG_PRINT1 ("EXECUTING at_dot.\n");
5526 if (PTR_BYTE_POS ((unsigned char *) d
) != PT_BYTE
)
5531 DEBUG_PRINT1 ("EXECUTING after_dot.\n");
5532 if (PTR_BYTE_POS ((unsigned char *) d
) <= PT_BYTE
)
5537 DEBUG_PRINT2 ("EXECUTING syntaxspec %d.\n", mcnt
);
5542 DEBUG_PRINT1 ("EXECUTING Emacs wordchar.\n");
5548 int pos1
= SYNTAX_TABLE_BYTE_TO_CHAR (PTR_TO_OFFSET (d
));
5549 UPDATE_SYNTAX_TABLE (pos1
);
5556 /* we must concern about multibyte form, ... */
5557 c
= STRING_CHAR_AND_LENGTH (d
, dend
- d
, len
);
5559 /* everything should be handled as ASCII, even though it
5560 looks like multibyte form. */
5563 if (SYNTAX (c
) != (enum syntaxcode
) mcnt
)
5567 SET_REGS_MATCHED ();
5571 DEBUG_PRINT2 ("EXECUTING notsyntaxspec %d.\n", mcnt
);
5573 goto matchnotsyntax
;
5576 DEBUG_PRINT1 ("EXECUTING Emacs notwordchar.\n");
5582 int pos1
= SYNTAX_TABLE_BYTE_TO_CHAR (PTR_TO_OFFSET (d
));
5583 UPDATE_SYNTAX_TABLE (pos1
);
5590 c
= STRING_CHAR_AND_LENGTH (d
, dend
- d
, len
);
5594 if (SYNTAX (c
) == (enum syntaxcode
) mcnt
)
5598 SET_REGS_MATCHED ();
5602 DEBUG_PRINT2 ("EXECUTING categoryspec %d.\n", *p
);
5609 c
= STRING_CHAR_AND_LENGTH (d
, dend
- d
, len
);
5613 if (!CHAR_HAS_CATEGORY (c
, mcnt
))
5617 SET_REGS_MATCHED ();
5620 case notcategoryspec
:
5621 DEBUG_PRINT2 ("EXECUTING notcategoryspec %d.\n", *p
);
5628 c
= STRING_CHAR_AND_LENGTH (d
, dend
- d
, len
);
5632 if (CHAR_HAS_CATEGORY (c
, mcnt
))
5636 SET_REGS_MATCHED ();
5639 #else /* not emacs */
5641 DEBUG_PRINT1 ("EXECUTING non-Emacs wordchar.\n");
5643 if (!WORDCHAR_P (d
))
5645 SET_REGS_MATCHED ();
5650 DEBUG_PRINT1 ("EXECUTING non-Emacs notwordchar.\n");
5654 SET_REGS_MATCHED ();
5657 #endif /* not emacs */
5662 continue; /* Successfully executed one pattern command; keep going. */
5665 /* We goto here if a matching operation fails. */
5667 if (!FAIL_STACK_EMPTY ())
5668 { /* A restart point is known. Restore to that state. */
5669 DEBUG_PRINT1 ("\nFAIL:\n");
5670 POP_FAILURE_POINT (d
, p
,
5671 lowest_active_reg
, highest_active_reg
,
5672 regstart
, regend
, reg_info
);
5674 /* If this failure point is a dummy, try the next one. */
5678 /* If we failed to the end of the pattern, don't examine *p. */
5682 boolean is_a_jump_n
= false;
5684 /* If failed to a backwards jump that's part of a repetition
5685 loop, need to pop this failure point and use the next one. */
5686 switch ((re_opcode_t
) *p
)
5690 case maybe_pop_jump
:
5691 case pop_failure_jump
:
5694 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
5697 if ((is_a_jump_n
&& (re_opcode_t
) *p1
== succeed_n
)
5699 && (re_opcode_t
) *p1
== on_failure_jump
))
5707 if (d
>= string1
&& d
<= end1
)
5711 break; /* Matching at this starting point really fails. */
5715 goto restore_best_regs
;
5719 return -1; /* Failure to match. */
5722 /* Subroutine definitions for re_match_2. */
5725 /* We are passed P pointing to a register number after a start_memory.
5727 Return true if the pattern up to the corresponding stop_memory can
5728 match the empty string, and false otherwise.
5730 If we find the matching stop_memory, sets P to point to one past its number.
5731 Otherwise, sets P to an undefined byte less than or equal to END.
5733 We don't handle duplicates properly (yet). */
5736 group_match_null_string_p (p
, end
, reg_info
)
5737 unsigned char **p
, *end
;
5738 register_info_type
*reg_info
;
5741 /* Point to after the args to the start_memory. */
5742 unsigned char *p1
= *p
+ 2;
5746 /* Skip over opcodes that can match nothing, and return true or
5747 false, as appropriate, when we get to one that can't, or to the
5748 matching stop_memory. */
5750 switch ((re_opcode_t
) *p1
)
5752 /* Could be either a loop or a series of alternatives. */
5753 case on_failure_jump
:
5755 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
5757 /* If the next operation is not a jump backwards in the
5762 /* Go through the on_failure_jumps of the alternatives,
5763 seeing if any of the alternatives cannot match nothing.
5764 The last alternative starts with only a jump,
5765 whereas the rest start with on_failure_jump and end
5766 with a jump, e.g., here is the pattern for `a|b|c':
5768 /on_failure_jump/0/6/exactn/1/a/jump_past_alt/0/6
5769 /on_failure_jump/0/6/exactn/1/b/jump_past_alt/0/3
5772 So, we have to first go through the first (n-1)
5773 alternatives and then deal with the last one separately. */
5776 /* Deal with the first (n-1) alternatives, which start
5777 with an on_failure_jump (see above) that jumps to right
5778 past a jump_past_alt. */
5780 while ((re_opcode_t
) p1
[mcnt
-3] == jump_past_alt
)
5782 /* `mcnt' holds how many bytes long the alternative
5783 is, including the ending `jump_past_alt' and
5786 if (!alt_match_null_string_p (p1
, p1
+ mcnt
- 3,
5790 /* Move to right after this alternative, including the
5794 /* Break if it's the beginning of an n-th alternative
5795 that doesn't begin with an on_failure_jump. */
5796 if ((re_opcode_t
) *p1
!= on_failure_jump
)
5799 /* Still have to check that it's not an n-th
5800 alternative that starts with an on_failure_jump. */
5802 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
5803 if ((re_opcode_t
) p1
[mcnt
-3] != jump_past_alt
)
5805 /* Get to the beginning of the n-th alternative. */
5811 /* Deal with the last alternative: go back and get number
5812 of the `jump_past_alt' just before it. `mcnt' contains
5813 the length of the alternative. */
5814 EXTRACT_NUMBER (mcnt
, p1
- 2);
5816 if (!alt_match_null_string_p (p1
, p1
+ mcnt
, reg_info
))
5819 p1
+= mcnt
; /* Get past the n-th alternative. */
5825 assert (p1
[1] == **p
);
5831 if (!common_op_match_null_string_p (&p1
, end
, reg_info
))
5834 } /* while p1 < end */
5837 } /* group_match_null_string_p */
5840 /* Similar to group_match_null_string_p, but doesn't deal with alternatives:
5841 It expects P to be the first byte of a single alternative and END one
5842 byte past the last. The alternative can contain groups. */
5845 alt_match_null_string_p (p
, end
, reg_info
)
5846 unsigned char *p
, *end
;
5847 register_info_type
*reg_info
;
5850 unsigned char *p1
= p
;
5854 /* Skip over opcodes that can match nothing, and break when we get
5855 to one that can't. */
5857 switch ((re_opcode_t
) *p1
)
5860 case on_failure_jump
:
5862 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
5867 if (!common_op_match_null_string_p (&p1
, end
, reg_info
))
5870 } /* while p1 < end */
5873 } /* alt_match_null_string_p */
5876 /* Deals with the ops common to group_match_null_string_p and
5877 alt_match_null_string_p.
5879 Sets P to one after the op and its arguments, if any. */
5882 common_op_match_null_string_p (p
, end
, reg_info
)
5883 unsigned char **p
, *end
;
5884 register_info_type
*reg_info
;
5889 unsigned char *p1
= *p
;
5891 switch ((re_opcode_t
) *p1
++)
5911 assert (reg_no
> 0 && reg_no
<= MAX_REGNUM
);
5912 ret
= group_match_null_string_p (&p1
, end
, reg_info
);
5914 /* Have to set this here in case we're checking a group which
5915 contains a group and a back reference to it. */
5917 if (REG_MATCH_NULL_STRING_P (reg_info
[reg_no
]) == MATCH_NULL_UNSET_VALUE
)
5918 REG_MATCH_NULL_STRING_P (reg_info
[reg_no
]) = ret
;
5924 /* If this is an optimized succeed_n for zero times, make the jump. */
5926 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
5934 /* Get to the number of times to succeed. */
5936 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
5941 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
5949 if (!REG_MATCH_NULL_STRING_P (reg_info
[*p1
]))
5957 /* All other opcodes mean we cannot match the empty string. */
5963 } /* common_op_match_null_string_p */
5966 /* Return zero if TRANSLATE[S1] and TRANSLATE[S2] are identical for LEN
5967 bytes; nonzero otherwise. */
5970 bcmp_translate (s1
, s2
, len
, translate
)
5971 unsigned char *s1
, *s2
;
5973 RE_TRANSLATE_TYPE translate
;
5975 register unsigned char *p1
= s1
, *p2
= s2
;
5976 unsigned char *p1_end
= s1
+ len
;
5977 unsigned char *p2_end
= s2
+ len
;
5979 while (p1
!= p1_end
&& p2
!= p2_end
)
5981 int p1_charlen
, p2_charlen
;
5984 p1_ch
= STRING_CHAR_AND_LENGTH (p1
, p1_end
- p1
, p1_charlen
);
5985 p2_ch
= STRING_CHAR_AND_LENGTH (p2
, p2_end
- p2
, p2_charlen
);
5987 if (RE_TRANSLATE (translate
, p1_ch
)
5988 != RE_TRANSLATE (translate
, p2_ch
))
5991 p1
+= p1_charlen
, p2
+= p2_charlen
;
5994 if (p1
!= p1_end
|| p2
!= p2_end
)
6000 /* Entry points for GNU code. */
6002 /* re_compile_pattern is the GNU regular expression compiler: it
6003 compiles PATTERN (of length SIZE) and puts the result in BUFP.
6004 Returns 0 if the pattern was valid, otherwise an error string.
6006 Assumes the `allocated' (and perhaps `buffer') and `translate' fields
6007 are set in BUFP on entry.
6009 We call regex_compile to do the actual compilation. */
6012 re_compile_pattern (pattern
, length
, bufp
)
6013 const char *pattern
;
6015 struct re_pattern_buffer
*bufp
;
6019 /* GNU code is written to assume at least RE_NREGS registers will be set
6020 (and at least one extra will be -1). */
6021 bufp
->regs_allocated
= REGS_UNALLOCATED
;
6023 /* And GNU code determines whether or not to get register information
6024 by passing null for the REGS argument to re_match, etc., not by
6028 /* Match anchors at newline. */
6029 bufp
->newline_anchor
= 1;
6031 ret
= regex_compile (pattern
, length
, re_syntax_options
, bufp
);
6035 return gettext (re_error_msgid
[(int) ret
]);
6038 /* Entry points compatible with 4.2 BSD regex library. We don't define
6039 them unless specifically requested. */
6041 #if defined (_REGEX_RE_COMP) || defined (_LIBC)
6043 /* BSD has one and only one pattern buffer. */
6044 static struct re_pattern_buffer re_comp_buf
;
6048 /* Make these definitions weak in libc, so POSIX programs can redefine
6049 these names if they don't use our functions, and still use
6050 regcomp/regexec below without link errors. */
6060 if (!re_comp_buf
.buffer
)
6061 return gettext ("No previous regular expression");
6065 if (!re_comp_buf
.buffer
)
6067 re_comp_buf
.buffer
= (unsigned char *) malloc (200);
6068 if (re_comp_buf
.buffer
== NULL
)
6069 return gettext (re_error_msgid
[(int) REG_ESPACE
]);
6070 re_comp_buf
.allocated
= 200;
6072 re_comp_buf
.fastmap
= (char *) malloc (1 << BYTEWIDTH
);
6073 if (re_comp_buf
.fastmap
== NULL
)
6074 return gettext (re_error_msgid
[(int) REG_ESPACE
]);
6077 /* Since `re_exec' always passes NULL for the `regs' argument, we
6078 don't need to initialize the pattern buffer fields which affect it. */
6080 /* Match anchors at newlines. */
6081 re_comp_buf
.newline_anchor
= 1;
6083 ret
= regex_compile (s
, strlen (s
), re_syntax_options
, &re_comp_buf
);
6088 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6089 return (char *) gettext (re_error_msgid
[(int) ret
]);
6100 const int len
= strlen (s
);
6102 0 <= re_search (&re_comp_buf
, s
, len
, 0, len
, (struct re_registers
*) 0);
6104 #endif /* _REGEX_RE_COMP */
6106 /* POSIX.2 functions. Don't define these for Emacs. */
6110 /* regcomp takes a regular expression as a string and compiles it.
6112 PREG is a regex_t *. We do not expect any fields to be initialized,
6113 since POSIX says we shouldn't. Thus, we set
6115 `buffer' to the compiled pattern;
6116 `used' to the length of the compiled pattern;
6117 `syntax' to RE_SYNTAX_POSIX_EXTENDED if the
6118 REG_EXTENDED bit in CFLAGS is set; otherwise, to
6119 RE_SYNTAX_POSIX_BASIC;
6120 `newline_anchor' to REG_NEWLINE being set in CFLAGS;
6121 `fastmap' and `fastmap_accurate' to zero;
6122 `re_nsub' to the number of subexpressions in PATTERN.
6124 PATTERN is the address of the pattern string.
6126 CFLAGS is a series of bits which affect compilation.
6128 If REG_EXTENDED is set, we use POSIX extended syntax; otherwise, we
6129 use POSIX basic syntax.
6131 If REG_NEWLINE is set, then . and [^...] don't match newline.
6132 Also, regexec will try a match beginning after every newline.
6134 If REG_ICASE is set, then we considers upper- and lowercase
6135 versions of letters to be equivalent when matching.
6137 If REG_NOSUB is set, then when PREG is passed to regexec, that
6138 routine will report only success or failure, and nothing about the
6141 It returns 0 if it succeeds, nonzero if it doesn't. (See regex.h for
6142 the return codes and their meanings.) */
6145 regcomp (preg
, pattern
, cflags
)
6147 const char *pattern
;
6152 = (cflags
& REG_EXTENDED
) ?
6153 RE_SYNTAX_POSIX_EXTENDED
: RE_SYNTAX_POSIX_BASIC
;
6155 /* regex_compile will allocate the space for the compiled pattern. */
6157 preg
->allocated
= 0;
6160 /* Don't bother to use a fastmap when searching. This simplifies the
6161 REG_NEWLINE case: if we used a fastmap, we'd have to put all the
6162 characters after newlines into the fastmap. This way, we just try
6166 if (cflags
& REG_ICASE
)
6171 = (RE_TRANSLATE_TYPE
) malloc (CHAR_SET_SIZE
6172 * sizeof (*(RE_TRANSLATE_TYPE
)0));
6173 if (preg
->translate
== NULL
)
6174 return (int) REG_ESPACE
;
6176 /* Map uppercase characters to corresponding lowercase ones. */
6177 for (i
= 0; i
< CHAR_SET_SIZE
; i
++)
6178 preg
->translate
[i
] = ISUPPER (i
) ? tolower (i
) : i
;
6181 preg
->translate
= NULL
;
6183 /* If REG_NEWLINE is set, newlines are treated differently. */
6184 if (cflags
& REG_NEWLINE
)
6185 { /* REG_NEWLINE implies neither . nor [^...] match newline. */
6186 syntax
&= ~RE_DOT_NEWLINE
;
6187 syntax
|= RE_HAT_LISTS_NOT_NEWLINE
;
6188 /* It also changes the matching behavior. */
6189 preg
->newline_anchor
= 1;
6192 preg
->newline_anchor
= 0;
6194 preg
->no_sub
= !!(cflags
& REG_NOSUB
);
6196 /* POSIX says a null character in the pattern terminates it, so we
6197 can use strlen here in compiling the pattern. */
6198 ret
= regex_compile (pattern
, strlen (pattern
), syntax
, preg
);
6200 /* POSIX doesn't distinguish between an unmatched open-group and an
6201 unmatched close-group: both are REG_EPAREN. */
6202 if (ret
== REG_ERPAREN
) ret
= REG_EPAREN
;
6208 /* regexec searches for a given pattern, specified by PREG, in the
6211 If NMATCH is zero or REG_NOSUB was set in the cflags argument to
6212 `regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at
6213 least NMATCH elements, and we set them to the offsets of the
6214 corresponding matched substrings.
6216 EFLAGS specifies `execution flags' which affect matching: if
6217 REG_NOTBOL is set, then ^ does not match at the beginning of the
6218 string; if REG_NOTEOL is set, then $ does not match at the end.
6220 We return 0 if we find a match and REG_NOMATCH if not. */
6223 regexec (preg
, string
, nmatch
, pmatch
, eflags
)
6224 const regex_t
*preg
;
6227 regmatch_t pmatch
[];
6231 struct re_registers regs
;
6232 regex_t private_preg
;
6233 int len
= strlen (string
);
6234 boolean want_reg_info
= !preg
->no_sub
&& nmatch
> 0;
6236 private_preg
= *preg
;
6238 private_preg
.not_bol
= !!(eflags
& REG_NOTBOL
);
6239 private_preg
.not_eol
= !!(eflags
& REG_NOTEOL
);
6241 /* The user has told us exactly how many registers to return
6242 information about, via `nmatch'. We have to pass that on to the
6243 matching routines. */
6244 private_preg
.regs_allocated
= REGS_FIXED
;
6248 regs
.num_regs
= nmatch
;
6249 regs
.start
= TALLOC (nmatch
, regoff_t
);
6250 regs
.end
= TALLOC (nmatch
, regoff_t
);
6251 if (regs
.start
== NULL
|| regs
.end
== NULL
)
6252 return (int) REG_NOMATCH
;
6255 /* Perform the searching operation. */
6256 ret
= re_search (&private_preg
, string
, len
,
6257 /* start: */ 0, /* range: */ len
,
6258 want_reg_info
? ®s
: (struct re_registers
*) 0);
6260 /* Copy the register information to the POSIX structure. */
6267 for (r
= 0; r
< nmatch
; r
++)
6269 pmatch
[r
].rm_so
= regs
.start
[r
];
6270 pmatch
[r
].rm_eo
= regs
.end
[r
];
6274 /* If we needed the temporary register info, free the space now. */
6279 /* We want zero return to mean success, unlike `re_search'. */
6280 return ret
>= 0 ? (int) REG_NOERROR
: (int) REG_NOMATCH
;
6284 /* Returns a message corresponding to an error code, ERRCODE, returned
6285 from either regcomp or regexec. We don't use PREG here. */
6288 regerror (errcode
, preg
, errbuf
, errbuf_size
)
6290 const regex_t
*preg
;
6298 || errcode
>= (sizeof (re_error_msgid
) / sizeof (re_error_msgid
[0])))
6299 /* Only error codes returned by the rest of the code should be passed
6300 to this routine. If we are given anything else, or if other regex
6301 code generates an invalid error code, then the program has a bug.
6302 Dump core so we can fix it. */
6305 msg
= gettext (re_error_msgid
[errcode
]);
6307 msg_size
= strlen (msg
) + 1; /* Includes the null. */
6309 if (errbuf_size
!= 0)
6311 if (msg_size
> errbuf_size
)
6313 strncpy (errbuf
, msg
, errbuf_size
- 1);
6314 errbuf
[errbuf_size
- 1] = 0;
6317 strcpy (errbuf
, msg
);
6324 /* Free dynamically allocated space used by PREG. */
6330 if (preg
->buffer
!= NULL
)
6331 free (preg
->buffer
);
6332 preg
->buffer
= NULL
;
6334 preg
->allocated
= 0;
6337 if (preg
->fastmap
!= NULL
)
6338 free (preg
->fastmap
);
6339 preg
->fastmap
= NULL
;
6340 preg
->fastmap_accurate
= 0;
6342 if (preg
->translate
!= NULL
)
6343 free (preg
->translate
);
6344 preg
->translate
= NULL
;
6347 #endif /* not emacs */