1 /* Extended regular expression matching and search library,
3 (Implements POSIX draft P10003.2/D11.2, except for
4 internationalization features.)
6 Copyright (C) 1993, 1994 Free Software Foundation, Inc.
8 This program is free software; you can redistribute it and/or modify
9 it under the terms of the GNU General Public License as published by
10 the Free Software Foundation; either version 2, or (at your option)
13 This program is distributed in the hope that it will be useful,
14 but WITHOUT ANY WARRANTY; without even the implied warranty of
15 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
16 GNU General Public License for more details.
18 You should have received a copy of the GNU General Public License
19 along with this program; if not, write to the Free Software
20 Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. */
22 /* AIX requires this to be the first thing in the file. */
23 #if defined (_AIX) && !defined (REGEX_MALLOC)
33 /* We need this for `regex.h', and perhaps for the Emacs include files. */
34 #include <sys/types.h>
36 /* This is for other GNU distributions with internationalized messages.
37 The GNU C Library itself does not yet support such messages. */
41 # define gettext(msgid) (msgid)
44 /* The `emacs' switch turns on certain matching commands
45 that make sense only in Emacs. */
62 /* We used to test for `BSTRING' here, but only GCC and Emacs define
63 `BSTRING', as far as I know, and neither of them use this code. */
64 #ifndef INHIBIT_STRING_HEADER
65 #if HAVE_STRING_H || STDC_HEADERS
68 #define bcmp(s1, s2, n) memcmp ((s1), (s2), (n))
71 #define bcopy(s, d, n) memcpy ((d), (s), (n))
74 #define bzero(s, n) memset ((s), 0, (n))
81 /* Define the syntax stuff for \<, \>, etc. */
83 /* This must be nonzero for the wordchar and notwordchar pattern
84 commands in re_match_2. */
89 #ifdef SWITCH_ENUM_BUG
90 #define SWITCH_ENUM_CAST(x) ((int)(x))
92 #define SWITCH_ENUM_CAST(x) (x)
97 extern char *re_syntax_table
;
99 #else /* not SYNTAX_TABLE */
101 /* How many characters in the character set. */
102 #define CHAR_SET_SIZE 256
104 static char re_syntax_table
[CHAR_SET_SIZE
];
115 bzero (re_syntax_table
, sizeof re_syntax_table
);
117 for (c
= 'a'; c
<= 'z'; c
++)
118 re_syntax_table
[c
] = Sword
;
120 for (c
= 'A'; c
<= 'Z'; c
++)
121 re_syntax_table
[c
] = Sword
;
123 for (c
= '0'; c
<= '9'; c
++)
124 re_syntax_table
[c
] = Sword
;
126 re_syntax_table
['_'] = Sword
;
131 #endif /* not SYNTAX_TABLE */
133 #define SYNTAX(c) re_syntax_table[c]
135 #endif /* not emacs */
137 /* Get the interface, including the syntax bits. */
140 /* isalpha etc. are used for the character classes. */
143 /* Jim Meyering writes:
145 "... Some ctype macros are valid only for character codes that
146 isascii says are ASCII (SGI's IRIX-4.0.5 is one such system --when
147 using /bin/cc or gcc but without giving an ansi option). So, all
148 ctype uses should be through macros like ISPRINT... If
149 STDC_HEADERS is defined, then autoconf has verified that the ctype
150 macros don't need to be guarded with references to isascii. ...
151 Defining isascii to 1 should let any compiler worth its salt
152 eliminate the && through constant folding." */
154 #if defined (STDC_HEADERS) || (!defined (isascii) && !defined (HAVE_ISASCII))
157 #define ISASCII(c) isascii(c)
161 #define ISBLANK(c) (ISASCII (c) && isblank (c))
163 #define ISBLANK(c) ((c) == ' ' || (c) == '\t')
166 #define ISGRAPH(c) (ISASCII (c) && isgraph (c))
168 #define ISGRAPH(c) (ISASCII (c) && isprint (c) && !isspace (c))
171 #define ISPRINT(c) (ISASCII (c) && isprint (c))
172 #define ISDIGIT(c) (ISASCII (c) && isdigit (c))
173 #define ISALNUM(c) (ISASCII (c) && isalnum (c))
174 #define ISALPHA(c) (ISASCII (c) && isalpha (c))
175 #define ISCNTRL(c) (ISASCII (c) && iscntrl (c))
176 #define ISLOWER(c) (ISASCII (c) && islower (c))
177 #define ISPUNCT(c) (ISASCII (c) && ispunct (c))
178 #define ISSPACE(c) (ISASCII (c) && isspace (c))
179 #define ISUPPER(c) (ISASCII (c) && isupper (c))
180 #define ISXDIGIT(c) (ISASCII (c) && isxdigit (c))
186 /* We remove any previous definition of `SIGN_EXTEND_CHAR',
187 since ours (we hope) works properly with all combinations of
188 machines, compilers, `char' and `unsigned char' argument types.
189 (Per Bothner suggested the basic approach.) */
190 #undef SIGN_EXTEND_CHAR
192 #define SIGN_EXTEND_CHAR(c) ((signed char) (c))
193 #else /* not __STDC__ */
194 /* As in Harbison and Steele. */
195 #define SIGN_EXTEND_CHAR(c) ((((unsigned char) (c)) ^ 128) - 128)
198 /* Should we use malloc or alloca? If REGEX_MALLOC is not defined, we
199 use `alloca' instead of `malloc'. This is because using malloc in
200 re_search* or re_match* could cause memory leaks when C-g is used in
201 Emacs; also, malloc is slower and causes storage fragmentation. On
202 the other hand, malloc is more portable, and easier to debug.
204 Because we sometimes use alloca, some routines have to be macros,
205 not functions -- `alloca'-allocated space disappears at the end of the
206 function it is called in. */
210 #define REGEX_ALLOCATE malloc
211 #define REGEX_REALLOCATE(source, osize, nsize) realloc (source, nsize)
213 #else /* not REGEX_MALLOC */
215 /* Emacs already defines alloca, sometimes. */
218 /* Make alloca work the best possible way. */
220 #define alloca __builtin_alloca
221 #else /* not __GNUC__ */
224 #else /* not __GNUC__ or HAVE_ALLOCA_H */
225 #ifndef _AIX /* Already did AIX, up at the top. */
227 #endif /* not _AIX */
228 #endif /* not HAVE_ALLOCA_H */
229 #endif /* not __GNUC__ */
231 #endif /* not alloca */
233 #define REGEX_ALLOCATE alloca
235 /* Assumes a `char *destination' variable. */
236 #define REGEX_REALLOCATE(source, osize, nsize) \
237 (destination = (char *) alloca (nsize), \
238 bcopy (source, destination, osize), \
241 #endif /* not REGEX_MALLOC */
244 /* True if `size1' is non-NULL and PTR is pointing anywhere inside
245 `string1' or just past its end. This works if PTR is NULL, which is
247 #define FIRST_STRING_P(ptr) \
248 (size1 && string1 <= (ptr) && (ptr) <= string1 + size1)
250 /* (Re)Allocate N items of type T using malloc, or fail. */
251 #define TALLOC(n, t) ((t *) malloc ((n) * sizeof (t)))
252 #define RETALLOC(addr, n, t) ((addr) = (t *) realloc (addr, (n) * sizeof (t)))
253 #define RETALLOC_IF(addr, n, t) \
254 if (addr) RETALLOC((addr), (n), t); else (addr) = TALLOC ((n), t)
255 #define REGEX_TALLOC(n, t) ((t *) REGEX_ALLOCATE ((n) * sizeof (t)))
257 #define BYTEWIDTH 8 /* In bits. */
259 #define STREQ(s1, s2) ((strcmp (s1, s2) == 0))
263 #define MAX(a, b) ((a) > (b) ? (a) : (b))
264 #define MIN(a, b) ((a) < (b) ? (a) : (b))
266 typedef char boolean
;
270 static int re_match_2_internal ();
272 /* These are the command codes that appear in compiled regular
273 expressions. Some opcodes are followed by argument bytes. A
274 command code can specify any interpretation whatsoever for its
275 arguments. Zero bytes may appear in the compiled regular expression. */
281 /* Succeed right away--no more backtracking. */
284 /* Followed by one byte giving n, then by n literal bytes. */
287 /* Matches any (more or less) character. */
290 /* Matches any one char belonging to specified set. First
291 following byte is number of bitmap bytes. Then come bytes
292 for a bitmap saying which chars are in. Bits in each byte
293 are ordered low-bit-first. A character is in the set if its
294 bit is 1. A character too large to have a bit in the map is
295 automatically not in the set. */
298 /* Same parameters as charset, but match any character that is
299 not one of those specified. */
302 /* Start remembering the text that is matched, for storing in a
303 register. Followed by one byte with the register number, in
304 the range 0 to one less than the pattern buffer's re_nsub
305 field. Then followed by one byte with the number of groups
306 inner to this one. (This last has to be part of the
307 start_memory only because we need it in the on_failure_jump
311 /* Stop remembering the text that is matched and store it in a
312 memory register. Followed by one byte with the register
313 number, in the range 0 to one less than `re_nsub' in the
314 pattern buffer, and one byte with the number of inner groups,
315 just like `start_memory'. (We need the number of inner
316 groups here because we don't have any easy way of finding the
317 corresponding start_memory when we're at a stop_memory.) */
320 /* Match a duplicate of something remembered. Followed by one
321 byte containing the register number. */
324 /* Fail unless at beginning of line. */
327 /* Fail unless at end of line. */
330 /* Succeeds if at beginning of buffer (if emacs) or at beginning
331 of string to be matched (if not). */
334 /* Analogously, for end of buffer/string. */
337 /* Followed by two byte relative address to which to jump. */
340 /* Same as jump, but marks the end of an alternative. */
343 /* Followed by two-byte relative address of place to resume at
344 in case of failure. */
347 /* Like on_failure_jump, but pushes a placeholder instead of the
348 current string position when executed. */
349 on_failure_keep_string_jump
,
351 /* Throw away latest failure point and then jump to following
352 two-byte relative address. */
355 /* Change to pop_failure_jump if know won't have to backtrack to
356 match; otherwise change to jump. This is used to jump
357 back to the beginning of a repeat. If what follows this jump
358 clearly won't match what the repeat does, such that we can be
359 sure that there is no use backtracking out of repetitions
360 already matched, then we change it to a pop_failure_jump.
361 Followed by two-byte address. */
364 /* Jump to following two-byte address, and push a dummy failure
365 point. This failure point will be thrown away if an attempt
366 is made to use it for a failure. A `+' construct makes this
367 before the first repeat. Also used as an intermediary kind
368 of jump when compiling an alternative. */
371 /* Push a dummy failure point and continue. Used at the end of
375 /* Followed by two-byte relative address and two-byte number n.
376 After matching N times, jump to the address upon failure. */
379 /* Followed by two-byte relative address, and two-byte number n.
380 Jump to the address N times, then fail. */
383 /* Set the following two-byte relative address to the
384 subsequent two-byte number. The address *includes* the two
388 wordchar
, /* Matches any word-constituent character. */
389 notwordchar
, /* Matches any char that is not a word-constituent. */
391 wordbeg
, /* Succeeds if at word beginning. */
392 wordend
, /* Succeeds if at word end. */
394 wordbound
, /* Succeeds if at a word boundary. */
395 notwordbound
/* Succeeds if not at a word boundary. */
398 ,before_dot
, /* Succeeds if before point. */
399 at_dot
, /* Succeeds if at point. */
400 after_dot
, /* Succeeds if after point. */
402 /* Matches any character whose syntax is specified. Followed by
403 a byte which contains a syntax code, e.g., Sword. */
406 /* Matches any character whose syntax is not that specified. */
411 /* Common operations on the compiled pattern. */
413 /* Store NUMBER in two contiguous bytes starting at DESTINATION. */
415 #define STORE_NUMBER(destination, number) \
417 (destination)[0] = (number) & 0377; \
418 (destination)[1] = (number) >> 8; \
421 /* Same as STORE_NUMBER, except increment DESTINATION to
422 the byte after where the number is stored. Therefore, DESTINATION
423 must be an lvalue. */
425 #define STORE_NUMBER_AND_INCR(destination, number) \
427 STORE_NUMBER (destination, number); \
428 (destination) += 2; \
431 /* Put into DESTINATION a number stored in two contiguous bytes starting
434 #define EXTRACT_NUMBER(destination, source) \
436 (destination) = *(source) & 0377; \
437 (destination) += SIGN_EXTEND_CHAR (*((source) + 1)) << 8; \
442 extract_number (dest
, source
)
444 unsigned char *source
;
446 int temp
= SIGN_EXTEND_CHAR (*(source
+ 1));
447 *dest
= *source
& 0377;
451 #ifndef EXTRACT_MACROS /* To debug the macros. */
452 #undef EXTRACT_NUMBER
453 #define EXTRACT_NUMBER(dest, src) extract_number (&dest, src)
454 #endif /* not EXTRACT_MACROS */
458 /* Same as EXTRACT_NUMBER, except increment SOURCE to after the number.
459 SOURCE must be an lvalue. */
461 #define EXTRACT_NUMBER_AND_INCR(destination, source) \
463 EXTRACT_NUMBER (destination, source); \
469 extract_number_and_incr (destination
, source
)
471 unsigned char **source
;
473 extract_number (destination
, *source
);
477 #ifndef EXTRACT_MACROS
478 #undef EXTRACT_NUMBER_AND_INCR
479 #define EXTRACT_NUMBER_AND_INCR(dest, src) \
480 extract_number_and_incr (&dest, &src)
481 #endif /* not EXTRACT_MACROS */
485 /* If DEBUG is defined, Regex prints many voluminous messages about what
486 it is doing (if the variable `debug' is nonzero). If linked with the
487 main program in `iregex.c', you can enter patterns and strings
488 interactively. And if linked with the main program in `main.c' and
489 the other test files, you can run the already-written tests. */
493 /* We use standard I/O for debugging. */
496 /* It is useful to test things that ``must'' be true when debugging. */
499 static int debug
= 0;
501 #define DEBUG_STATEMENT(e) e
502 #define DEBUG_PRINT1(x) if (debug) printf (x)
503 #define DEBUG_PRINT2(x1, x2) if (debug) printf (x1, x2)
504 #define DEBUG_PRINT3(x1, x2, x3) if (debug) printf (x1, x2, x3)
505 #define DEBUG_PRINT4(x1, x2, x3, x4) if (debug) printf (x1, x2, x3, x4)
506 #define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) \
507 if (debug) print_partial_compiled_pattern (s, e)
508 #define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) \
509 if (debug) print_double_string (w, s1, sz1, s2, sz2)
512 /* Print the fastmap in human-readable form. */
515 print_fastmap (fastmap
)
518 unsigned was_a_range
= 0;
521 while (i
< (1 << BYTEWIDTH
))
527 while (i
< (1 << BYTEWIDTH
) && fastmap
[i
])
543 /* Print a compiled pattern string in human-readable form, starting at
544 the START pointer into it and ending just before the pointer END. */
547 print_partial_compiled_pattern (start
, end
)
548 unsigned char *start
;
552 unsigned char *p
= start
;
553 unsigned char *pend
= end
;
561 /* Loop over pattern commands. */
564 printf ("%d:\t", p
- start
);
566 switch ((re_opcode_t
) *p
++)
574 printf ("/exactn/%d", mcnt
);
585 printf ("/start_memory/%d/%d", mcnt
, *p
++);
590 printf ("/stop_memory/%d/%d", mcnt
, *p
++);
594 printf ("/duplicate/%d", *p
++);
604 register int c
, last
= -100;
605 register int in_range
= 0;
607 printf ("/charset [%s",
608 (re_opcode_t
) *(p
- 1) == charset_not
? "^" : "");
610 assert (p
+ *p
< pend
);
612 for (c
= 0; c
< 256; c
++)
614 && (p
[1 + (c
/8)] & (1 << (c
% 8))))
616 /* Are we starting a range? */
617 if (last
+ 1 == c
&& ! in_range
)
622 /* Have we broken a range? */
623 else if (last
+ 1 != c
&& in_range
)
652 case on_failure_jump
:
653 extract_number_and_incr (&mcnt
, &p
);
654 printf ("/on_failure_jump to %d", p
+ mcnt
- start
);
657 case on_failure_keep_string_jump
:
658 extract_number_and_incr (&mcnt
, &p
);
659 printf ("/on_failure_keep_string_jump to %d", p
+ mcnt
- start
);
662 case dummy_failure_jump
:
663 extract_number_and_incr (&mcnt
, &p
);
664 printf ("/dummy_failure_jump to %d", p
+ mcnt
- start
);
667 case push_dummy_failure
:
668 printf ("/push_dummy_failure");
672 extract_number_and_incr (&mcnt
, &p
);
673 printf ("/maybe_pop_jump to %d", p
+ mcnt
- start
);
676 case pop_failure_jump
:
677 extract_number_and_incr (&mcnt
, &p
);
678 printf ("/pop_failure_jump to %d", p
+ mcnt
- start
);
682 extract_number_and_incr (&mcnt
, &p
);
683 printf ("/jump_past_alt to %d", p
+ mcnt
- start
);
687 extract_number_and_incr (&mcnt
, &p
);
688 printf ("/jump to %d", p
+ mcnt
- start
);
692 extract_number_and_incr (&mcnt
, &p
);
693 extract_number_and_incr (&mcnt2
, &p
);
694 printf ("/succeed_n to %d, %d times", p
+ mcnt
- start
, mcnt2
);
698 extract_number_and_incr (&mcnt
, &p
);
699 extract_number_and_incr (&mcnt2
, &p
);
700 printf ("/jump_n to %d, %d times", p
+ mcnt
- start
, mcnt2
);
704 extract_number_and_incr (&mcnt
, &p
);
705 extract_number_and_incr (&mcnt2
, &p
);
706 printf ("/set_number_at location %d to %d", p
+ mcnt
- start
, mcnt2
);
710 printf ("/wordbound");
714 printf ("/notwordbound");
726 printf ("/before_dot");
734 printf ("/after_dot");
738 printf ("/syntaxspec");
740 printf ("/%d", mcnt
);
744 printf ("/notsyntaxspec");
746 printf ("/%d", mcnt
);
751 printf ("/wordchar");
755 printf ("/notwordchar");
767 printf ("?%d", *(p
-1));
773 printf ("%d:\tend of pattern.\n", p
- start
);
778 print_compiled_pattern (bufp
)
779 struct re_pattern_buffer
*bufp
;
781 unsigned char *buffer
= bufp
->buffer
;
783 print_partial_compiled_pattern (buffer
, buffer
+ bufp
->used
);
784 printf ("%d bytes used/%d bytes allocated.\n", bufp
->used
, bufp
->allocated
);
786 if (bufp
->fastmap_accurate
&& bufp
->fastmap
)
788 printf ("fastmap: ");
789 print_fastmap (bufp
->fastmap
);
792 printf ("re_nsub: %d\t", bufp
->re_nsub
);
793 printf ("regs_alloc: %d\t", bufp
->regs_allocated
);
794 printf ("can_be_null: %d\t", bufp
->can_be_null
);
795 printf ("newline_anchor: %d\n", bufp
->newline_anchor
);
796 printf ("no_sub: %d\t", bufp
->no_sub
);
797 printf ("not_bol: %d\t", bufp
->not_bol
);
798 printf ("not_eol: %d\t", bufp
->not_eol
);
799 printf ("syntax: %d\n", bufp
->syntax
);
800 /* Perhaps we should print the translate table? */
805 print_double_string (where
, string1
, size1
, string2
, size2
)
818 if (FIRST_STRING_P (where
))
820 for (this_char
= where
- string1
; this_char
< size1
; this_char
++)
821 putchar (string1
[this_char
]);
826 for (this_char
= where
- string2
; this_char
< size2
; this_char
++)
827 putchar (string2
[this_char
]);
831 #else /* not DEBUG */
836 #define DEBUG_STATEMENT(e)
837 #define DEBUG_PRINT1(x)
838 #define DEBUG_PRINT2(x1, x2)
839 #define DEBUG_PRINT3(x1, x2, x3)
840 #define DEBUG_PRINT4(x1, x2, x3, x4)
841 #define DEBUG_PRINT_COMPILED_PATTERN(p, s, e)
842 #define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2)
844 #endif /* not DEBUG */
846 /* Set by `re_set_syntax' to the current regexp syntax to recognize. Can
847 also be assigned to arbitrarily: each pattern buffer stores its own
848 syntax, so it can be changed between regex compilations. */
849 /* This has no initializer because initialized variables in Emacs
850 become read-only after dumping. */
851 reg_syntax_t re_syntax_options
;
854 /* Specify the precise syntax of regexps for compilation. This provides
855 for compatibility for various utilities which historically have
856 different, incompatible syntaxes.
858 The argument SYNTAX is a bit mask comprised of the various bits
859 defined in regex.h. We return the old syntax. */
862 re_set_syntax (syntax
)
865 reg_syntax_t ret
= re_syntax_options
;
867 re_syntax_options
= syntax
;
871 /* This table gives an error message for each of the error codes listed
872 in regex.h. Obviously the order here has to be same as there.
873 POSIX doesn't require that we do anything for REG_NOERROR,
874 but why not be nice? */
876 static const char *re_error_msgid
[] =
877 { "Success", /* REG_NOERROR */
878 "No match", /* REG_NOMATCH */
879 "Invalid regular expression", /* REG_BADPAT */
880 "Invalid collation character", /* REG_ECOLLATE */
881 "Invalid character class name", /* REG_ECTYPE */
882 "Trailing backslash", /* REG_EESCAPE */
883 "Invalid back reference", /* REG_ESUBREG */
884 "Unmatched [ or [^", /* REG_EBRACK */
885 "Unmatched ( or \\(", /* REG_EPAREN */
886 "Unmatched \\{", /* REG_EBRACE */
887 "Invalid content of \\{\\}", /* REG_BADBR */
888 "Invalid range end", /* REG_ERANGE */
889 "Memory exhausted", /* REG_ESPACE */
890 "Invalid preceding regular expression", /* REG_BADRPT */
891 "Premature end of regular expression", /* REG_EEND */
892 "Regular expression too big", /* REG_ESIZE */
893 "Unmatched ) or \\)", /* REG_ERPAREN */
896 /* Avoiding alloca during matching, to placate r_alloc. */
898 /* Define MATCH_MAY_ALLOCATE unless we need to make sure that the
899 searching and matching functions should not call alloca. On some
900 systems, alloca is implemented in terms of malloc, and if we're
901 using the relocating allocator routines, then malloc could cause a
902 relocation, which might (if the strings being searched are in the
903 ralloc heap) shift the data out from underneath the regexp
906 Here's another reason to avoid allocation: Emacs
907 processes input from X in a signal handler; processing X input may
908 call malloc; if input arrives while a matching routine is calling
909 malloc, then we're scrod. But Emacs can't just block input while
910 calling matching routines; then we don't notice interrupts when
911 they come in. So, Emacs blocks input around all regexp calls
912 except the matching calls, which it leaves unprotected, in the
913 faith that they will not malloc. */
915 /* Normally, this is fine. */
916 #define MATCH_MAY_ALLOCATE
918 /* The match routines may not allocate if (1) they would do it with malloc
919 and (2) it's not safe for them to use malloc. */
920 #if (defined (C_ALLOCA) || defined (REGEX_MALLOC)) && (defined (emacs) || defined (REL_ALLOC))
921 #undef MATCH_MAY_ALLOCATE
925 /* Failure stack declarations and macros; both re_compile_fastmap and
926 re_match_2 use a failure stack. These have to be macros because of
930 /* Number of failure points for which to initially allocate space
931 when matching. If this number is exceeded, we allocate more
932 space, so it is not a hard limit. */
933 #ifndef INIT_FAILURE_ALLOC
934 #define INIT_FAILURE_ALLOC 5
937 /* Roughly the maximum number of failure points on the stack. Would be
938 exactly that if always used MAX_FAILURE_SPACE each time we failed.
939 This is a variable only so users of regex can assign to it; we never
940 change it ourselves. */
941 int re_max_failures
= 2000;
943 typedef unsigned char *fail_stack_elt_t
;
947 fail_stack_elt_t
*stack
;
949 unsigned avail
; /* Offset of next open position. */
952 #define FAIL_STACK_EMPTY() (fail_stack.avail == 0)
953 #define FAIL_STACK_PTR_EMPTY() (fail_stack_ptr->avail == 0)
954 #define FAIL_STACK_FULL() (fail_stack.avail == fail_stack.size)
955 #define FAIL_STACK_TOP() (fail_stack.stack[fail_stack.avail])
958 /* Initialize `fail_stack'. Do `return -2' if the alloc fails. */
960 #ifdef MATCH_MAY_ALLOCATE
961 #define INIT_FAIL_STACK() \
963 fail_stack.stack = (fail_stack_elt_t *) \
964 REGEX_ALLOCATE (INIT_FAILURE_ALLOC * sizeof (fail_stack_elt_t)); \
966 if (fail_stack.stack == NULL) \
969 fail_stack.size = INIT_FAILURE_ALLOC; \
970 fail_stack.avail = 0; \
973 #define INIT_FAIL_STACK() \
975 fail_stack.avail = 0; \
980 /* Double the size of FAIL_STACK, up to approximately `re_max_failures' items.
982 Return 1 if succeeds, and 0 if either ran out of memory
983 allocating space for it or it was already too large.
985 REGEX_REALLOCATE requires `destination' be declared. */
987 #define DOUBLE_FAIL_STACK(fail_stack) \
988 ((fail_stack).size > re_max_failures * MAX_FAILURE_ITEMS \
990 : ((fail_stack).stack = (fail_stack_elt_t *) \
991 REGEX_REALLOCATE ((fail_stack).stack, \
992 (fail_stack).size * sizeof (fail_stack_elt_t), \
993 ((fail_stack).size << 1) * sizeof (fail_stack_elt_t)), \
995 (fail_stack).stack == NULL \
997 : ((fail_stack).size <<= 1, \
1001 /* Push PATTERN_OP on FAIL_STACK.
1003 Return 1 if was able to do so and 0 if ran out of memory allocating
1005 #define PUSH_PATTERN_OP(pattern_op, fail_stack) \
1006 ((FAIL_STACK_FULL () \
1007 && !DOUBLE_FAIL_STACK (fail_stack)) \
1009 : ((fail_stack).stack[(fail_stack).avail++] = pattern_op, \
1012 /* Push a pointer value onto the failure stack.
1013 Assumes the variable `fail_stack'. Probably should only
1014 be called from within `PUSH_FAILURE_POINT'. */
1015 #define PUSH_FAILURE_POINTER(item) \
1016 fail_stack.stack[fail_stack.avail++] = (fail_stack_elt_t) (item)
1018 /* This pushes an integer-valued item onto the failure stack.
1019 Assumes the variable `fail_stack'. Probably should only
1020 be called from within `PUSH_FAILURE_POINT'. */
1021 #define PUSH_FAILURE_INT(item) \
1022 fail_stack.stack[fail_stack.avail++] = (fail_stack_elt_t) (EMACS_INT) (item)
1024 /* The complement operation. Assumes `fail_stack' is nonempty. */
1025 #define POP_FAILURE_POINTER() fail_stack.stack[--fail_stack.avail]
1027 /* The complement operation. Assumes `fail_stack' is nonempty. */
1028 #define POP_FAILURE_INT() (EMACS_INT) fail_stack.stack[--fail_stack.avail]
1030 /* Used to omit pushing failure point id's when we're not debugging. */
1032 #define DEBUG_PUSH PUSH_FAILURE_INT
1033 #define DEBUG_POP(item_addr) *(item_addr) = POP_FAILURE_INT ()
1035 #define DEBUG_PUSH(item)
1036 #define DEBUG_POP(item_addr)
1040 /* Push the information about the state we will need
1041 if we ever fail back to it.
1043 Requires variables fail_stack, regstart, regend, reg_info, and
1044 num_regs be declared. DOUBLE_FAIL_STACK requires `destination' be
1047 Does `return FAILURE_CODE' if runs out of memory. */
1049 #define PUSH_FAILURE_POINT(pattern_place, string_place, failure_code) \
1051 char *destination; \
1052 /* Must be int, so when we don't save any registers, the arithmetic \
1053 of 0 + -1 isn't done as unsigned. */ \
1056 DEBUG_STATEMENT (failure_id++); \
1057 DEBUG_STATEMENT (nfailure_points_pushed++); \
1058 DEBUG_PRINT2 ("\nPUSH_FAILURE_POINT #%u:\n", failure_id); \
1059 DEBUG_PRINT2 (" Before push, next avail: %d\n", (fail_stack).avail);\
1060 DEBUG_PRINT2 (" size: %d\n", (fail_stack).size);\
1062 DEBUG_PRINT2 (" slots needed: %d\n", NUM_FAILURE_ITEMS); \
1063 DEBUG_PRINT2 (" available: %d\n", REMAINING_AVAIL_SLOTS); \
1065 /* Ensure we have enough space allocated for what we will push. */ \
1066 while (REMAINING_AVAIL_SLOTS < NUM_FAILURE_ITEMS) \
1068 if (!DOUBLE_FAIL_STACK (fail_stack)) \
1069 return failure_code; \
1071 DEBUG_PRINT2 ("\n Doubled stack; size now: %d\n", \
1072 (fail_stack).size); \
1073 DEBUG_PRINT2 (" slots available: %d\n", REMAINING_AVAIL_SLOTS);\
1076 /* Push the info, starting with the registers. */ \
1077 DEBUG_PRINT1 ("\n"); \
1079 for (this_reg = lowest_active_reg; this_reg <= highest_active_reg; \
1082 DEBUG_PRINT2 (" Pushing reg: %d\n", this_reg); \
1083 DEBUG_STATEMENT (num_regs_pushed++); \
1085 DEBUG_PRINT2 (" start: 0x%x\n", regstart[this_reg]); \
1086 PUSH_FAILURE_POINTER (regstart[this_reg]); \
1088 DEBUG_PRINT2 (" end: 0x%x\n", regend[this_reg]); \
1089 PUSH_FAILURE_POINTER (regend[this_reg]); \
1091 DEBUG_PRINT2 (" info: 0x%x\n ", reg_info[this_reg]); \
1092 DEBUG_PRINT2 (" match_null=%d", \
1093 REG_MATCH_NULL_STRING_P (reg_info[this_reg])); \
1094 DEBUG_PRINT2 (" active=%d", IS_ACTIVE (reg_info[this_reg])); \
1095 DEBUG_PRINT2 (" matched_something=%d", \
1096 MATCHED_SOMETHING (reg_info[this_reg])); \
1097 DEBUG_PRINT2 (" ever_matched=%d", \
1098 EVER_MATCHED_SOMETHING (reg_info[this_reg])); \
1099 DEBUG_PRINT1 ("\n"); \
1100 PUSH_FAILURE_POINTER (reg_info[this_reg].word); \
1103 DEBUG_PRINT2 (" Pushing low active reg: %d\n", lowest_active_reg);\
1104 PUSH_FAILURE_INT (lowest_active_reg); \
1106 DEBUG_PRINT2 (" Pushing high active reg: %d\n", highest_active_reg);\
1107 PUSH_FAILURE_INT (highest_active_reg); \
1109 DEBUG_PRINT2 (" Pushing pattern 0x%x: ", pattern_place); \
1110 DEBUG_PRINT_COMPILED_PATTERN (bufp, pattern_place, pend); \
1111 PUSH_FAILURE_POINTER (pattern_place); \
1113 DEBUG_PRINT2 (" Pushing string 0x%x: `", string_place); \
1114 DEBUG_PRINT_DOUBLE_STRING (string_place, string1, size1, string2, \
1116 DEBUG_PRINT1 ("'\n"); \
1117 PUSH_FAILURE_POINTER (string_place); \
1119 DEBUG_PRINT2 (" Pushing failure id: %u\n", failure_id); \
1120 DEBUG_PUSH (failure_id); \
1123 /* This is the number of items that are pushed and popped on the stack
1124 for each register. */
1125 #define NUM_REG_ITEMS 3
1127 /* Individual items aside from the registers. */
1129 #define NUM_NONREG_ITEMS 5 /* Includes failure point id. */
1131 #define NUM_NONREG_ITEMS 4
1134 /* We push at most this many items on the stack. */
1135 #define MAX_FAILURE_ITEMS ((num_regs - 1) * NUM_REG_ITEMS + NUM_NONREG_ITEMS)
1137 /* We actually push this many items. */
1138 #define NUM_FAILURE_ITEMS \
1139 ((highest_active_reg - lowest_active_reg + 1) * NUM_REG_ITEMS \
1142 /* How many items can still be added to the stack without overflowing it. */
1143 #define REMAINING_AVAIL_SLOTS ((fail_stack).size - (fail_stack).avail)
1146 /* Pops what PUSH_FAIL_STACK pushes.
1148 We restore into the parameters, all of which should be lvalues:
1149 STR -- the saved data position.
1150 PAT -- the saved pattern position.
1151 LOW_REG, HIGH_REG -- the highest and lowest active registers.
1152 REGSTART, REGEND -- arrays of string positions.
1153 REG_INFO -- array of information about each subexpression.
1155 Also assumes the variables `fail_stack' and (if debugging), `bufp',
1156 `pend', `string1', `size1', `string2', and `size2'. */
1158 #define POP_FAILURE_POINT(str, pat, low_reg, high_reg, regstart, regend, reg_info)\
1160 DEBUG_STATEMENT (fail_stack_elt_t failure_id;) \
1162 const unsigned char *string_temp; \
1164 assert (!FAIL_STACK_EMPTY ()); \
1166 /* Remove failure points and point to how many regs pushed. */ \
1167 DEBUG_PRINT1 ("POP_FAILURE_POINT:\n"); \
1168 DEBUG_PRINT2 (" Before pop, next avail: %d\n", fail_stack.avail); \
1169 DEBUG_PRINT2 (" size: %d\n", fail_stack.size); \
1171 assert (fail_stack.avail >= NUM_NONREG_ITEMS); \
1173 DEBUG_POP (&failure_id); \
1174 DEBUG_PRINT2 (" Popping failure id: %u\n", failure_id); \
1176 /* If the saved string location is NULL, it came from an \
1177 on_failure_keep_string_jump opcode, and we want to throw away the \
1178 saved NULL, thus retaining our current position in the string. */ \
1179 string_temp = POP_FAILURE_POINTER (); \
1180 if (string_temp != NULL) \
1181 str = (const char *) string_temp; \
1183 DEBUG_PRINT2 (" Popping string 0x%x: `", str); \
1184 DEBUG_PRINT_DOUBLE_STRING (str, string1, size1, string2, size2); \
1185 DEBUG_PRINT1 ("'\n"); \
1187 pat = (unsigned char *) POP_FAILURE_POINTER (); \
1188 DEBUG_PRINT2 (" Popping pattern 0x%x: ", pat); \
1189 DEBUG_PRINT_COMPILED_PATTERN (bufp, pat, pend); \
1191 /* Restore register info. */ \
1192 high_reg = (unsigned) POP_FAILURE_INT (); \
1193 DEBUG_PRINT2 (" Popping high active reg: %d\n", high_reg); \
1195 low_reg = (unsigned) POP_FAILURE_INT (); \
1196 DEBUG_PRINT2 (" Popping low active reg: %d\n", low_reg); \
1198 for (this_reg = high_reg; this_reg >= low_reg; this_reg--) \
1200 DEBUG_PRINT2 (" Popping reg: %d\n", this_reg); \
1202 reg_info[this_reg].word = POP_FAILURE_POINTER (); \
1203 DEBUG_PRINT2 (" info: 0x%x\n", reg_info[this_reg]); \
1205 regend[this_reg] = (const char *) POP_FAILURE_POINTER (); \
1206 DEBUG_PRINT2 (" end: 0x%x\n", regend[this_reg]); \
1208 regstart[this_reg] = (const char *) POP_FAILURE_POINTER (); \
1209 DEBUG_PRINT2 (" start: 0x%x\n", regstart[this_reg]); \
1212 set_regs_matched_done = 0; \
1213 DEBUG_STATEMENT (nfailure_points_popped++); \
1214 } /* POP_FAILURE_POINT */
1218 /* Structure for per-register (a.k.a. per-group) information.
1219 This must not be longer than one word, because we push this value
1220 onto the failure stack. Other register information, such as the
1221 starting and ending positions (which are addresses), and the list of
1222 inner groups (which is a bits list) are maintained in separate
1225 We are making a (strictly speaking) nonportable assumption here: that
1226 the compiler will pack our bit fields into something that fits into
1227 the type of `word', i.e., is something that fits into one item on the
1231 fail_stack_elt_t word
;
1234 /* This field is one if this group can match the empty string,
1235 zero if not. If not yet determined, `MATCH_NULL_UNSET_VALUE'. */
1236 #define MATCH_NULL_UNSET_VALUE 3
1237 unsigned match_null_string_p
: 2;
1238 unsigned is_active
: 1;
1239 unsigned matched_something
: 1;
1240 unsigned ever_matched_something
: 1;
1242 } register_info_type
;
1244 #define REG_MATCH_NULL_STRING_P(R) ((R).bits.match_null_string_p)
1245 #define IS_ACTIVE(R) ((R).bits.is_active)
1246 #define MATCHED_SOMETHING(R) ((R).bits.matched_something)
1247 #define EVER_MATCHED_SOMETHING(R) ((R).bits.ever_matched_something)
1250 /* Call this when have matched a real character; it sets `matched' flags
1251 for the subexpressions which we are currently inside. Also records
1252 that those subexprs have matched. */
1253 #define SET_REGS_MATCHED() \
1256 if (!set_regs_matched_done) \
1259 set_regs_matched_done = 1; \
1260 for (r = lowest_active_reg; r <= highest_active_reg; r++) \
1262 MATCHED_SOMETHING (reg_info[r]) \
1263 = EVER_MATCHED_SOMETHING (reg_info[r]) \
1270 /* Registers are set to a sentinel when they haven't yet matched. */
1271 static char reg_unset_dummy
;
1272 #define REG_UNSET_VALUE (®_unset_dummy)
1273 #define REG_UNSET(e) ((e) == REG_UNSET_VALUE)
1277 /* How do we implement a missing MATCH_MAY_ALLOCATE?
1278 We make the fail stack a global thing, and then grow it to
1279 re_max_failures when we compile. */
1280 #ifndef MATCH_MAY_ALLOCATE
1281 static fail_stack_type fail_stack
;
1283 static const char ** regstart
, ** regend
;
1284 static const char ** old_regstart
, ** old_regend
;
1285 static const char **best_regstart
, **best_regend
;
1286 static register_info_type
*reg_info
;
1287 static const char **reg_dummy
;
1288 static register_info_type
*reg_info_dummy
;
1292 /* Subroutine declarations and macros for regex_compile. */
1294 static void store_op1 (), store_op2 ();
1295 static void insert_op1 (), insert_op2 ();
1296 static boolean
at_begline_loc_p (), at_endline_loc_p ();
1297 static boolean
group_in_compile_stack ();
1298 static reg_errcode_t
compile_range ();
1300 /* Fetch the next character in the uncompiled pattern---translating it
1301 if necessary. Also cast from a signed character in the constant
1302 string passed to us by the user to an unsigned char that we can use
1303 as an array index (in, e.g., `translate'). */
1304 #define PATFETCH(c) \
1305 do {if (p == pend) return REG_EEND; \
1306 c = (unsigned char) *p++; \
1307 if (translate) c = translate[c]; \
1310 /* Fetch the next character in the uncompiled pattern, with no
1312 #define PATFETCH_RAW(c) \
1313 do {if (p == pend) return REG_EEND; \
1314 c = (unsigned char) *p++; \
1317 /* Go backwards one character in the pattern. */
1318 #define PATUNFETCH p--
1321 /* If `translate' is non-null, return translate[D], else just D. We
1322 cast the subscript to translate because some data is declared as
1323 `char *', to avoid warnings when a string constant is passed. But
1324 when we use a character as a subscript we must make it unsigned. */
1325 #define TRANSLATE(d) (translate ? translate[(unsigned char) (d)] : (d))
1328 /* Macros for outputting the compiled pattern into `buffer'. */
1330 /* If the buffer isn't allocated when it comes in, use this. */
1331 #define INIT_BUF_SIZE 32
1333 /* Make sure we have at least N more bytes of space in buffer. */
1334 #define GET_BUFFER_SPACE(n) \
1335 while (b - bufp->buffer + (n) > bufp->allocated) \
1338 /* Make sure we have one more byte of buffer space and then add C to it. */
1339 #define BUF_PUSH(c) \
1341 GET_BUFFER_SPACE (1); \
1342 *b++ = (unsigned char) (c); \
1346 /* Ensure we have two more bytes of buffer space and then append C1 and C2. */
1347 #define BUF_PUSH_2(c1, c2) \
1349 GET_BUFFER_SPACE (2); \
1350 *b++ = (unsigned char) (c1); \
1351 *b++ = (unsigned char) (c2); \
1355 /* As with BUF_PUSH_2, except for three bytes. */
1356 #define BUF_PUSH_3(c1, c2, c3) \
1358 GET_BUFFER_SPACE (3); \
1359 *b++ = (unsigned char) (c1); \
1360 *b++ = (unsigned char) (c2); \
1361 *b++ = (unsigned char) (c3); \
1365 /* Store a jump with opcode OP at LOC to location TO. We store a
1366 relative address offset by the three bytes the jump itself occupies. */
1367 #define STORE_JUMP(op, loc, to) \
1368 store_op1 (op, loc, (to) - (loc) - 3)
1370 /* Likewise, for a two-argument jump. */
1371 #define STORE_JUMP2(op, loc, to, arg) \
1372 store_op2 (op, loc, (to) - (loc) - 3, arg)
1374 /* Like `STORE_JUMP', but for inserting. Assume `b' is the buffer end. */
1375 #define INSERT_JUMP(op, loc, to) \
1376 insert_op1 (op, loc, (to) - (loc) - 3, b)
1378 /* Like `STORE_JUMP2', but for inserting. Assume `b' is the buffer end. */
1379 #define INSERT_JUMP2(op, loc, to, arg) \
1380 insert_op2 (op, loc, (to) - (loc) - 3, arg, b)
1383 /* This is not an arbitrary limit: the arguments which represent offsets
1384 into the pattern are two bytes long. So if 2^16 bytes turns out to
1385 be too small, many things would have to change. */
1386 #define MAX_BUF_SIZE (1L << 16)
1389 /* Extend the buffer by twice its current size via realloc and
1390 reset the pointers that pointed into the old block to point to the
1391 correct places in the new one. If extending the buffer results in it
1392 being larger than MAX_BUF_SIZE, then flag memory exhausted. */
1393 #define EXTEND_BUFFER() \
1395 unsigned char *old_buffer = bufp->buffer; \
1396 if (bufp->allocated == MAX_BUF_SIZE) \
1398 bufp->allocated <<= 1; \
1399 if (bufp->allocated > MAX_BUF_SIZE) \
1400 bufp->allocated = MAX_BUF_SIZE; \
1401 bufp->buffer = (unsigned char *) realloc (bufp->buffer, bufp->allocated);\
1402 if (bufp->buffer == NULL) \
1403 return REG_ESPACE; \
1404 /* If the buffer moved, move all the pointers into it. */ \
1405 if (old_buffer != bufp->buffer) \
1407 b = (b - old_buffer) + bufp->buffer; \
1408 begalt = (begalt - old_buffer) + bufp->buffer; \
1409 if (fixup_alt_jump) \
1410 fixup_alt_jump = (fixup_alt_jump - old_buffer) + bufp->buffer;\
1412 laststart = (laststart - old_buffer) + bufp->buffer; \
1413 if (pending_exact) \
1414 pending_exact = (pending_exact - old_buffer) + bufp->buffer; \
1419 /* Since we have one byte reserved for the register number argument to
1420 {start,stop}_memory, the maximum number of groups we can report
1421 things about is what fits in that byte. */
1422 #define MAX_REGNUM 255
1424 /* But patterns can have more than `MAX_REGNUM' registers. We just
1425 ignore the excess. */
1426 typedef unsigned regnum_t
;
1429 /* Macros for the compile stack. */
1431 /* Since offsets can go either forwards or backwards, this type needs to
1432 be able to hold values from -(MAX_BUF_SIZE - 1) to MAX_BUF_SIZE - 1. */
1433 typedef int pattern_offset_t
;
1437 pattern_offset_t begalt_offset
;
1438 pattern_offset_t fixup_alt_jump
;
1439 pattern_offset_t inner_group_offset
;
1440 pattern_offset_t laststart_offset
;
1442 } compile_stack_elt_t
;
1447 compile_stack_elt_t
*stack
;
1449 unsigned avail
; /* Offset of next open position. */
1450 } compile_stack_type
;
1453 #define INIT_COMPILE_STACK_SIZE 32
1455 #define COMPILE_STACK_EMPTY (compile_stack.avail == 0)
1456 #define COMPILE_STACK_FULL (compile_stack.avail == compile_stack.size)
1458 /* The next available element. */
1459 #define COMPILE_STACK_TOP (compile_stack.stack[compile_stack.avail])
1462 /* Set the bit for character C in a list. */
1463 #define SET_LIST_BIT(c) \
1464 (b[((unsigned char) (c)) / BYTEWIDTH] \
1465 |= 1 << (((unsigned char) c) % BYTEWIDTH))
1468 /* Get the next unsigned number in the uncompiled pattern. */
1469 #define GET_UNSIGNED_NUMBER(num) \
1473 while (ISDIGIT (c)) \
1477 num = num * 10 + c - '0'; \
1485 #define CHAR_CLASS_MAX_LENGTH 6 /* Namely, `xdigit'. */
1487 #define IS_CHAR_CLASS(string) \
1488 (STREQ (string, "alpha") || STREQ (string, "upper") \
1489 || STREQ (string, "lower") || STREQ (string, "digit") \
1490 || STREQ (string, "alnum") || STREQ (string, "xdigit") \
1491 || STREQ (string, "space") || STREQ (string, "print") \
1492 || STREQ (string, "punct") || STREQ (string, "graph") \
1493 || STREQ (string, "cntrl") || STREQ (string, "blank"))
1495 /* `regex_compile' compiles PATTERN (of length SIZE) according to SYNTAX.
1496 Returns one of error codes defined in `regex.h', or zero for success.
1498 Assumes the `allocated' (and perhaps `buffer') and `translate'
1499 fields are set in BUFP on entry.
1501 If it succeeds, results are put in BUFP (if it returns an error, the
1502 contents of BUFP are undefined):
1503 `buffer' is the compiled pattern;
1504 `syntax' is set to SYNTAX;
1505 `used' is set to the length of the compiled pattern;
1506 `fastmap_accurate' is zero;
1507 `re_nsub' is the number of subexpressions in PATTERN;
1508 `not_bol' and `not_eol' are zero;
1510 The `fastmap' and `newline_anchor' fields are neither
1511 examined nor set. */
1513 /* Return, freeing storage we allocated. */
1514 #define FREE_STACK_RETURN(value) \
1515 return (free (compile_stack.stack), value)
1517 static reg_errcode_t
1518 regex_compile (pattern
, size
, syntax
, bufp
)
1519 const char *pattern
;
1521 reg_syntax_t syntax
;
1522 struct re_pattern_buffer
*bufp
;
1524 /* We fetch characters from PATTERN here. Even though PATTERN is
1525 `char *' (i.e., signed), we declare these variables as unsigned, so
1526 they can be reliably used as array indices. */
1527 register unsigned char c
, c1
;
1529 /* A random temporary spot in PATTERN. */
1532 /* Points to the end of the buffer, where we should append. */
1533 register unsigned char *b
;
1535 /* Keeps track of unclosed groups. */
1536 compile_stack_type compile_stack
;
1538 /* Points to the current (ending) position in the pattern. */
1539 const char *p
= pattern
;
1540 const char *pend
= pattern
+ size
;
1542 /* How to translate the characters in the pattern. */
1543 char *translate
= bufp
->translate
;
1545 /* Address of the count-byte of the most recently inserted `exactn'
1546 command. This makes it possible to tell if a new exact-match
1547 character can be added to that command or if the character requires
1548 a new `exactn' command. */
1549 unsigned char *pending_exact
= 0;
1551 /* Address of start of the most recently finished expression.
1552 This tells, e.g., postfix * where to find the start of its
1553 operand. Reset at the beginning of groups and alternatives. */
1554 unsigned char *laststart
= 0;
1556 /* Address of beginning of regexp, or inside of last group. */
1557 unsigned char *begalt
;
1559 /* Place in the uncompiled pattern (i.e., the {) to
1560 which to go back if the interval is invalid. */
1561 const char *beg_interval
;
1563 /* Address of the place where a forward jump should go to the end of
1564 the containing expression. Each alternative of an `or' -- except the
1565 last -- ends with a forward jump of this sort. */
1566 unsigned char *fixup_alt_jump
= 0;
1568 /* Counts open-groups as they are encountered. Remembered for the
1569 matching close-group on the compile stack, so the same register
1570 number is put in the stop_memory as the start_memory. */
1571 regnum_t regnum
= 0;
1574 DEBUG_PRINT1 ("\nCompiling pattern: ");
1577 unsigned debug_count
;
1579 for (debug_count
= 0; debug_count
< size
; debug_count
++)
1580 putchar (pattern
[debug_count
]);
1585 /* Initialize the compile stack. */
1586 compile_stack
.stack
= TALLOC (INIT_COMPILE_STACK_SIZE
, compile_stack_elt_t
);
1587 if (compile_stack
.stack
== NULL
)
1590 compile_stack
.size
= INIT_COMPILE_STACK_SIZE
;
1591 compile_stack
.avail
= 0;
1593 /* Initialize the pattern buffer. */
1594 bufp
->syntax
= syntax
;
1595 bufp
->fastmap_accurate
= 0;
1596 bufp
->not_bol
= bufp
->not_eol
= 0;
1598 /* Set `used' to zero, so that if we return an error, the pattern
1599 printer (for debugging) will think there's no pattern. We reset it
1603 /* Always count groups, whether or not bufp->no_sub is set. */
1606 #if !defined (emacs) && !defined (SYNTAX_TABLE)
1607 /* Initialize the syntax table. */
1608 init_syntax_once ();
1611 if (bufp
->allocated
== 0)
1614 { /* If zero allocated, but buffer is non-null, try to realloc
1615 enough space. This loses if buffer's address is bogus, but
1616 that is the user's responsibility. */
1617 RETALLOC (bufp
->buffer
, INIT_BUF_SIZE
, unsigned char);
1620 { /* Caller did not allocate a buffer. Do it for them. */
1621 bufp
->buffer
= TALLOC (INIT_BUF_SIZE
, unsigned char);
1623 if (!bufp
->buffer
) FREE_STACK_RETURN (REG_ESPACE
);
1625 bufp
->allocated
= INIT_BUF_SIZE
;
1628 begalt
= b
= bufp
->buffer
;
1630 /* Loop through the uncompiled pattern until we're at the end. */
1639 if ( /* If at start of pattern, it's an operator. */
1641 /* If context independent, it's an operator. */
1642 || syntax
& RE_CONTEXT_INDEP_ANCHORS
1643 /* Otherwise, depends on what's come before. */
1644 || at_begline_loc_p (pattern
, p
, syntax
))
1654 if ( /* If at end of pattern, it's an operator. */
1656 /* If context independent, it's an operator. */
1657 || syntax
& RE_CONTEXT_INDEP_ANCHORS
1658 /* Otherwise, depends on what's next. */
1659 || at_endline_loc_p (p
, pend
, syntax
))
1669 if ((syntax
& RE_BK_PLUS_QM
)
1670 || (syntax
& RE_LIMITED_OPS
))
1674 /* If there is no previous pattern... */
1677 if (syntax
& RE_CONTEXT_INVALID_OPS
)
1678 FREE_STACK_RETURN (REG_BADRPT
);
1679 else if (!(syntax
& RE_CONTEXT_INDEP_OPS
))
1684 /* Are we optimizing this jump? */
1685 boolean keep_string_p
= false;
1687 /* 1 means zero (many) matches is allowed. */
1688 char zero_times_ok
= 0, many_times_ok
= 0;
1690 /* If there is a sequence of repetition chars, collapse it
1691 down to just one (the right one). We can't combine
1692 interval operators with these because of, e.g., `a{2}*',
1693 which should only match an even number of `a's. */
1697 zero_times_ok
|= c
!= '+';
1698 many_times_ok
|= c
!= '?';
1706 || (!(syntax
& RE_BK_PLUS_QM
) && (c
== '+' || c
== '?')))
1709 else if (syntax
& RE_BK_PLUS_QM
&& c
== '\\')
1711 if (p
== pend
) FREE_STACK_RETURN (REG_EESCAPE
);
1714 if (!(c1
== '+' || c1
== '?'))
1729 /* If we get here, we found another repeat character. */
1732 /* Star, etc. applied to an empty pattern is equivalent
1733 to an empty pattern. */
1737 /* Now we know whether or not zero matches is allowed
1738 and also whether or not two or more matches is allowed. */
1740 { /* More than one repetition is allowed, so put in at the
1741 end a backward relative jump from `b' to before the next
1742 jump we're going to put in below (which jumps from
1743 laststart to after this jump).
1745 But if we are at the `*' in the exact sequence `.*\n',
1746 insert an unconditional jump backwards to the .,
1747 instead of the beginning of the loop. This way we only
1748 push a failure point once, instead of every time
1749 through the loop. */
1750 assert (p
- 1 > pattern
);
1752 /* Allocate the space for the jump. */
1753 GET_BUFFER_SPACE (3);
1755 /* We know we are not at the first character of the pattern,
1756 because laststart was nonzero. And we've already
1757 incremented `p', by the way, to be the character after
1758 the `*'. Do we have to do something analogous here
1759 for null bytes, because of RE_DOT_NOT_NULL? */
1760 if (TRANSLATE (*(p
- 2)) == TRANSLATE ('.')
1762 && p
< pend
&& TRANSLATE (*p
) == TRANSLATE ('\n')
1763 && !(syntax
& RE_DOT_NEWLINE
))
1764 { /* We have .*\n. */
1765 STORE_JUMP (jump
, b
, laststart
);
1766 keep_string_p
= true;
1769 /* Anything else. */
1770 STORE_JUMP (maybe_pop_jump
, b
, laststart
- 3);
1772 /* We've added more stuff to the buffer. */
1776 /* On failure, jump from laststart to b + 3, which will be the
1777 end of the buffer after this jump is inserted. */
1778 GET_BUFFER_SPACE (3);
1779 INSERT_JUMP (keep_string_p
? on_failure_keep_string_jump
1787 /* At least one repetition is required, so insert a
1788 `dummy_failure_jump' before the initial
1789 `on_failure_jump' instruction of the loop. This
1790 effects a skip over that instruction the first time
1791 we hit that loop. */
1792 GET_BUFFER_SPACE (3);
1793 INSERT_JUMP (dummy_failure_jump
, laststart
, laststart
+ 6);
1808 boolean had_char_class
= false;
1810 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
1812 /* Ensure that we have enough space to push a charset: the
1813 opcode, the length count, and the bitset; 34 bytes in all. */
1814 GET_BUFFER_SPACE (34);
1818 /* We test `*p == '^' twice, instead of using an if
1819 statement, so we only need one BUF_PUSH. */
1820 BUF_PUSH (*p
== '^' ? charset_not
: charset
);
1824 /* Remember the first position in the bracket expression. */
1827 /* Push the number of bytes in the bitmap. */
1828 BUF_PUSH ((1 << BYTEWIDTH
) / BYTEWIDTH
);
1830 /* Clear the whole map. */
1831 bzero (b
, (1 << BYTEWIDTH
) / BYTEWIDTH
);
1833 /* charset_not matches newline according to a syntax bit. */
1834 if ((re_opcode_t
) b
[-2] == charset_not
1835 && (syntax
& RE_HAT_LISTS_NOT_NEWLINE
))
1836 SET_LIST_BIT ('\n');
1838 /* Read in characters and ranges, setting map bits. */
1841 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
1845 /* \ might escape characters inside [...] and [^...]. */
1846 if ((syntax
& RE_BACKSLASH_ESCAPE_IN_LISTS
) && c
== '\\')
1848 if (p
== pend
) FREE_STACK_RETURN (REG_EESCAPE
);
1855 /* Could be the end of the bracket expression. If it's
1856 not (i.e., when the bracket expression is `[]' so
1857 far), the ']' character bit gets set way below. */
1858 if (c
== ']' && p
!= p1
+ 1)
1861 /* Look ahead to see if it's a range when the last thing
1862 was a character class. */
1863 if (had_char_class
&& c
== '-' && *p
!= ']')
1864 FREE_STACK_RETURN (REG_ERANGE
);
1866 /* Look ahead to see if it's a range when the last thing
1867 was a character: if this is a hyphen not at the
1868 beginning or the end of a list, then it's the range
1871 && !(p
- 2 >= pattern
&& p
[-2] == '[')
1872 && !(p
- 3 >= pattern
&& p
[-3] == '[' && p
[-2] == '^')
1876 = compile_range (&p
, pend
, translate
, syntax
, b
);
1877 if (ret
!= REG_NOERROR
) FREE_STACK_RETURN (ret
);
1880 else if (p
[0] == '-' && p
[1] != ']')
1881 { /* This handles ranges made up of characters only. */
1884 /* Move past the `-'. */
1887 ret
= compile_range (&p
, pend
, translate
, syntax
, b
);
1888 if (ret
!= REG_NOERROR
) FREE_STACK_RETURN (ret
);
1891 /* See if we're at the beginning of a possible character
1894 else if (syntax
& RE_CHAR_CLASSES
&& c
== '[' && *p
== ':')
1895 { /* Leave room for the null. */
1896 char str
[CHAR_CLASS_MAX_LENGTH
+ 1];
1901 /* If pattern is `[[:'. */
1902 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
1907 if (c
== ':' || c
== ']' || p
== pend
1908 || c1
== CHAR_CLASS_MAX_LENGTH
)
1914 /* If isn't a word bracketed by `[:' and:`]':
1915 undo the ending character, the letters, and leave
1916 the leading `:' and `[' (but set bits for them). */
1917 if (c
== ':' && *p
== ']')
1920 boolean is_alnum
= STREQ (str
, "alnum");
1921 boolean is_alpha
= STREQ (str
, "alpha");
1922 boolean is_blank
= STREQ (str
, "blank");
1923 boolean is_cntrl
= STREQ (str
, "cntrl");
1924 boolean is_digit
= STREQ (str
, "digit");
1925 boolean is_graph
= STREQ (str
, "graph");
1926 boolean is_lower
= STREQ (str
, "lower");
1927 boolean is_print
= STREQ (str
, "print");
1928 boolean is_punct
= STREQ (str
, "punct");
1929 boolean is_space
= STREQ (str
, "space");
1930 boolean is_upper
= STREQ (str
, "upper");
1931 boolean is_xdigit
= STREQ (str
, "xdigit");
1933 if (!IS_CHAR_CLASS (str
))
1934 FREE_STACK_RETURN (REG_ECTYPE
);
1936 /* Throw away the ] at the end of the character
1940 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
1942 for (ch
= 0; ch
< 1 << BYTEWIDTH
; ch
++)
1944 /* This was split into 3 if's to
1945 avoid an arbitrary limit in some compiler. */
1946 if ( (is_alnum
&& ISALNUM (ch
))
1947 || (is_alpha
&& ISALPHA (ch
))
1948 || (is_blank
&& ISBLANK (ch
))
1949 || (is_cntrl
&& ISCNTRL (ch
)))
1951 if ( (is_digit
&& ISDIGIT (ch
))
1952 || (is_graph
&& ISGRAPH (ch
))
1953 || (is_lower
&& ISLOWER (ch
))
1954 || (is_print
&& ISPRINT (ch
)))
1956 if ( (is_punct
&& ISPUNCT (ch
))
1957 || (is_space
&& ISSPACE (ch
))
1958 || (is_upper
&& ISUPPER (ch
))
1959 || (is_xdigit
&& ISXDIGIT (ch
)))
1962 had_char_class
= true;
1971 had_char_class
= false;
1976 had_char_class
= false;
1981 /* Discard any (non)matching list bytes that are all 0 at the
1982 end of the map. Decrease the map-length byte too. */
1983 while ((int) b
[-1] > 0 && b
[b
[-1] - 1] == 0)
1991 if (syntax
& RE_NO_BK_PARENS
)
1998 if (syntax
& RE_NO_BK_PARENS
)
2005 if (syntax
& RE_NEWLINE_ALT
)
2012 if (syntax
& RE_NO_BK_VBAR
)
2019 if (syntax
& RE_INTERVALS
&& syntax
& RE_NO_BK_BRACES
)
2020 goto handle_interval
;
2026 if (p
== pend
) FREE_STACK_RETURN (REG_EESCAPE
);
2028 /* Do not translate the character after the \, so that we can
2029 distinguish, e.g., \B from \b, even if we normally would
2030 translate, e.g., B to b. */
2036 if (syntax
& RE_NO_BK_PARENS
)
2037 goto normal_backslash
;
2043 if (COMPILE_STACK_FULL
)
2045 RETALLOC (compile_stack
.stack
, compile_stack
.size
<< 1,
2046 compile_stack_elt_t
);
2047 if (compile_stack
.stack
== NULL
) return REG_ESPACE
;
2049 compile_stack
.size
<<= 1;
2052 /* These are the values to restore when we hit end of this
2053 group. They are all relative offsets, so that if the
2054 whole pattern moves because of realloc, they will still
2056 COMPILE_STACK_TOP
.begalt_offset
= begalt
- bufp
->buffer
;
2057 COMPILE_STACK_TOP
.fixup_alt_jump
2058 = fixup_alt_jump
? fixup_alt_jump
- bufp
->buffer
+ 1 : 0;
2059 COMPILE_STACK_TOP
.laststart_offset
= b
- bufp
->buffer
;
2060 COMPILE_STACK_TOP
.regnum
= regnum
;
2062 /* We will eventually replace the 0 with the number of
2063 groups inner to this one. But do not push a
2064 start_memory for groups beyond the last one we can
2065 represent in the compiled pattern. */
2066 if (regnum
<= MAX_REGNUM
)
2068 COMPILE_STACK_TOP
.inner_group_offset
= b
- bufp
->buffer
+ 2;
2069 BUF_PUSH_3 (start_memory
, regnum
, 0);
2072 compile_stack
.avail
++;
2077 /* If we've reached MAX_REGNUM groups, then this open
2078 won't actually generate any code, so we'll have to
2079 clear pending_exact explicitly. */
2085 if (syntax
& RE_NO_BK_PARENS
) goto normal_backslash
;
2087 if (COMPILE_STACK_EMPTY
)
2088 if (syntax
& RE_UNMATCHED_RIGHT_PAREN_ORD
)
2089 goto normal_backslash
;
2091 FREE_STACK_RETURN (REG_ERPAREN
);
2095 { /* Push a dummy failure point at the end of the
2096 alternative for a possible future
2097 `pop_failure_jump' to pop. See comments at
2098 `push_dummy_failure' in `re_match_2'. */
2099 BUF_PUSH (push_dummy_failure
);
2101 /* We allocated space for this jump when we assigned
2102 to `fixup_alt_jump', in the `handle_alt' case below. */
2103 STORE_JUMP (jump_past_alt
, fixup_alt_jump
, b
- 1);
2106 /* See similar code for backslashed left paren above. */
2107 if (COMPILE_STACK_EMPTY
)
2108 if (syntax
& RE_UNMATCHED_RIGHT_PAREN_ORD
)
2111 FREE_STACK_RETURN (REG_ERPAREN
);
2113 /* Since we just checked for an empty stack above, this
2114 ``can't happen''. */
2115 assert (compile_stack
.avail
!= 0);
2117 /* We don't just want to restore into `regnum', because
2118 later groups should continue to be numbered higher,
2119 as in `(ab)c(de)' -- the second group is #2. */
2120 regnum_t this_group_regnum
;
2122 compile_stack
.avail
--;
2123 begalt
= bufp
->buffer
+ COMPILE_STACK_TOP
.begalt_offset
;
2125 = COMPILE_STACK_TOP
.fixup_alt_jump
2126 ? bufp
->buffer
+ COMPILE_STACK_TOP
.fixup_alt_jump
- 1
2128 laststart
= bufp
->buffer
+ COMPILE_STACK_TOP
.laststart_offset
;
2129 this_group_regnum
= COMPILE_STACK_TOP
.regnum
;
2130 /* If we've reached MAX_REGNUM groups, then this open
2131 won't actually generate any code, so we'll have to
2132 clear pending_exact explicitly. */
2135 /* We're at the end of the group, so now we know how many
2136 groups were inside this one. */
2137 if (this_group_regnum
<= MAX_REGNUM
)
2139 unsigned char *inner_group_loc
2140 = bufp
->buffer
+ COMPILE_STACK_TOP
.inner_group_offset
;
2142 *inner_group_loc
= regnum
- this_group_regnum
;
2143 BUF_PUSH_3 (stop_memory
, this_group_regnum
,
2144 regnum
- this_group_regnum
);
2150 case '|': /* `\|'. */
2151 if (syntax
& RE_LIMITED_OPS
|| syntax
& RE_NO_BK_VBAR
)
2152 goto normal_backslash
;
2154 if (syntax
& RE_LIMITED_OPS
)
2157 /* Insert before the previous alternative a jump which
2158 jumps to this alternative if the former fails. */
2159 GET_BUFFER_SPACE (3);
2160 INSERT_JUMP (on_failure_jump
, begalt
, b
+ 6);
2164 /* The alternative before this one has a jump after it
2165 which gets executed if it gets matched. Adjust that
2166 jump so it will jump to this alternative's analogous
2167 jump (put in below, which in turn will jump to the next
2168 (if any) alternative's such jump, etc.). The last such
2169 jump jumps to the correct final destination. A picture:
2175 If we are at `b', then fixup_alt_jump right now points to a
2176 three-byte space after `a'. We'll put in the jump, set
2177 fixup_alt_jump to right after `b', and leave behind three
2178 bytes which we'll fill in when we get to after `c'. */
2181 STORE_JUMP (jump_past_alt
, fixup_alt_jump
, b
);
2183 /* Mark and leave space for a jump after this alternative,
2184 to be filled in later either by next alternative or
2185 when know we're at the end of a series of alternatives. */
2187 GET_BUFFER_SPACE (3);
2196 /* If \{ is a literal. */
2197 if (!(syntax
& RE_INTERVALS
)
2198 /* If we're at `\{' and it's not the open-interval
2200 || ((syntax
& RE_INTERVALS
) && (syntax
& RE_NO_BK_BRACES
))
2201 || (p
- 2 == pattern
&& p
== pend
))
2202 goto normal_backslash
;
2206 /* If got here, then the syntax allows intervals. */
2208 /* At least (most) this many matches must be made. */
2209 int lower_bound
= -1, upper_bound
= -1;
2211 beg_interval
= p
- 1;
2215 if (syntax
& RE_NO_BK_BRACES
)
2216 goto unfetch_interval
;
2218 FREE_STACK_RETURN (REG_EBRACE
);
2221 GET_UNSIGNED_NUMBER (lower_bound
);
2225 GET_UNSIGNED_NUMBER (upper_bound
);
2226 if (upper_bound
< 0) upper_bound
= RE_DUP_MAX
;
2229 /* Interval such as `{1}' => match exactly once. */
2230 upper_bound
= lower_bound
;
2232 if (lower_bound
< 0 || upper_bound
> RE_DUP_MAX
2233 || lower_bound
> upper_bound
)
2235 if (syntax
& RE_NO_BK_BRACES
)
2236 goto unfetch_interval
;
2238 FREE_STACK_RETURN (REG_BADBR
);
2241 if (!(syntax
& RE_NO_BK_BRACES
))
2243 if (c
!= '\\') FREE_STACK_RETURN (REG_EBRACE
);
2250 if (syntax
& RE_NO_BK_BRACES
)
2251 goto unfetch_interval
;
2253 FREE_STACK_RETURN (REG_BADBR
);
2256 /* We just parsed a valid interval. */
2258 /* If it's invalid to have no preceding re. */
2261 if (syntax
& RE_CONTEXT_INVALID_OPS
)
2262 FREE_STACK_RETURN (REG_BADRPT
);
2263 else if (syntax
& RE_CONTEXT_INDEP_OPS
)
2266 goto unfetch_interval
;
2269 /* If the upper bound is zero, don't want to succeed at
2270 all; jump from `laststart' to `b + 3', which will be
2271 the end of the buffer after we insert the jump. */
2272 if (upper_bound
== 0)
2274 GET_BUFFER_SPACE (3);
2275 INSERT_JUMP (jump
, laststart
, b
+ 3);
2279 /* Otherwise, we have a nontrivial interval. When
2280 we're all done, the pattern will look like:
2281 set_number_at <jump count> <upper bound>
2282 set_number_at <succeed_n count> <lower bound>
2283 succeed_n <after jump addr> <succeed_n count>
2285 jump_n <succeed_n addr> <jump count>
2286 (The upper bound and `jump_n' are omitted if
2287 `upper_bound' is 1, though.) */
2289 { /* If the upper bound is > 1, we need to insert
2290 more at the end of the loop. */
2291 unsigned nbytes
= 10 + (upper_bound
> 1) * 10;
2293 GET_BUFFER_SPACE (nbytes
);
2295 /* Initialize lower bound of the `succeed_n', even
2296 though it will be set during matching by its
2297 attendant `set_number_at' (inserted next),
2298 because `re_compile_fastmap' needs to know.
2299 Jump to the `jump_n' we might insert below. */
2300 INSERT_JUMP2 (succeed_n
, laststart
,
2301 b
+ 5 + (upper_bound
> 1) * 5,
2305 /* Code to initialize the lower bound. Insert
2306 before the `succeed_n'. The `5' is the last two
2307 bytes of this `set_number_at', plus 3 bytes of
2308 the following `succeed_n'. */
2309 insert_op2 (set_number_at
, laststart
, 5, lower_bound
, b
);
2312 if (upper_bound
> 1)
2313 { /* More than one repetition is allowed, so
2314 append a backward jump to the `succeed_n'
2315 that starts this interval.
2317 When we've reached this during matching,
2318 we'll have matched the interval once, so
2319 jump back only `upper_bound - 1' times. */
2320 STORE_JUMP2 (jump_n
, b
, laststart
+ 5,
2324 /* The location we want to set is the second
2325 parameter of the `jump_n'; that is `b-2' as
2326 an absolute address. `laststart' will be
2327 the `set_number_at' we're about to insert;
2328 `laststart+3' the number to set, the source
2329 for the relative address. But we are
2330 inserting into the middle of the pattern --
2331 so everything is getting moved up by 5.
2332 Conclusion: (b - 2) - (laststart + 3) + 5,
2333 i.e., b - laststart.
2335 We insert this at the beginning of the loop
2336 so that if we fail during matching, we'll
2337 reinitialize the bounds. */
2338 insert_op2 (set_number_at
, laststart
, b
- laststart
,
2339 upper_bound
- 1, b
);
2344 beg_interval
= NULL
;
2349 /* If an invalid interval, match the characters as literals. */
2350 assert (beg_interval
);
2352 beg_interval
= NULL
;
2354 /* normal_char and normal_backslash need `c'. */
2357 if (!(syntax
& RE_NO_BK_BRACES
))
2359 if (p
> pattern
&& p
[-1] == '\\')
2360 goto normal_backslash
;
2365 /* There is no way to specify the before_dot and after_dot
2366 operators. rms says this is ok. --karl */
2374 BUF_PUSH_2 (syntaxspec
, syntax_spec_code
[c
]);
2380 BUF_PUSH_2 (notsyntaxspec
, syntax_spec_code
[c
]);
2387 BUF_PUSH (wordchar
);
2393 BUF_PUSH (notwordchar
);
2406 BUF_PUSH (wordbound
);
2410 BUF_PUSH (notwordbound
);
2421 case '1': case '2': case '3': case '4': case '5':
2422 case '6': case '7': case '8': case '9':
2423 if (syntax
& RE_NO_BK_REFS
)
2429 FREE_STACK_RETURN (REG_ESUBREG
);
2431 /* Can't back reference to a subexpression if inside of it. */
2432 if (group_in_compile_stack (compile_stack
, c1
))
2436 BUF_PUSH_2 (duplicate
, c1
);
2442 if (syntax
& RE_BK_PLUS_QM
)
2445 goto normal_backslash
;
2449 /* You might think it would be useful for \ to mean
2450 not to translate; but if we don't translate it
2451 it will never match anything. */
2459 /* Expects the character in `c'. */
2461 /* If no exactn currently being built. */
2464 /* If last exactn not at current position. */
2465 || pending_exact
+ *pending_exact
+ 1 != b
2467 /* We have only one byte following the exactn for the count. */
2468 || *pending_exact
== (1 << BYTEWIDTH
) - 1
2470 /* If followed by a repetition operator. */
2471 || *p
== '*' || *p
== '^'
2472 || ((syntax
& RE_BK_PLUS_QM
)
2473 ? *p
== '\\' && (p
[1] == '+' || p
[1] == '?')
2474 : (*p
== '+' || *p
== '?'))
2475 || ((syntax
& RE_INTERVALS
)
2476 && ((syntax
& RE_NO_BK_BRACES
)
2478 : (p
[0] == '\\' && p
[1] == '{'))))
2480 /* Start building a new exactn. */
2484 BUF_PUSH_2 (exactn
, 0);
2485 pending_exact
= b
- 1;
2492 } /* while p != pend */
2495 /* Through the pattern now. */
2498 STORE_JUMP (jump_past_alt
, fixup_alt_jump
, b
);
2500 if (!COMPILE_STACK_EMPTY
)
2501 FREE_STACK_RETURN (REG_EPAREN
);
2503 /* If we don't want backtracking, force success
2504 the first time we reach the end of the compiled pattern. */
2505 if (syntax
& RE_NO_POSIX_BACKTRACKING
)
2508 free (compile_stack
.stack
);
2510 /* We have succeeded; set the length of the buffer. */
2511 bufp
->used
= b
- bufp
->buffer
;
2516 DEBUG_PRINT1 ("\nCompiled pattern: \n");
2517 print_compiled_pattern (bufp
);
2521 #ifndef MATCH_MAY_ALLOCATE
2522 /* Initialize the failure stack to the largest possible stack. This
2523 isn't necessary unless we're trying to avoid calling alloca in
2524 the search and match routines. */
2526 int num_regs
= bufp
->re_nsub
+ 1;
2528 /* Since DOUBLE_FAIL_STACK refuses to double only if the current size
2529 is strictly greater than re_max_failures, the largest possible stack
2530 is 2 * re_max_failures failure points. */
2531 if (fail_stack
.size
< (2 * re_max_failures
* MAX_FAILURE_ITEMS
))
2533 fail_stack
.size
= (2 * re_max_failures
* MAX_FAILURE_ITEMS
);
2536 if (! fail_stack
.stack
)
2538 = (fail_stack_elt_t
*) xmalloc (fail_stack
.size
2539 * sizeof (fail_stack_elt_t
));
2542 = (fail_stack_elt_t
*) xrealloc (fail_stack
.stack
,
2544 * sizeof (fail_stack_elt_t
)));
2545 #else /* not emacs */
2546 if (! fail_stack
.stack
)
2548 = (fail_stack_elt_t
*) malloc (fail_stack
.size
2549 * sizeof (fail_stack_elt_t
));
2552 = (fail_stack_elt_t
*) realloc (fail_stack
.stack
,
2554 * sizeof (fail_stack_elt_t
)));
2555 #endif /* not emacs */
2558 /* Initialize some other variables the matcher uses. */
2559 RETALLOC_IF (regstart
, num_regs
, const char *);
2560 RETALLOC_IF (regend
, num_regs
, const char *);
2561 RETALLOC_IF (old_regstart
, num_regs
, const char *);
2562 RETALLOC_IF (old_regend
, num_regs
, const char *);
2563 RETALLOC_IF (best_regstart
, num_regs
, const char *);
2564 RETALLOC_IF (best_regend
, num_regs
, const char *);
2565 RETALLOC_IF (reg_info
, num_regs
, register_info_type
);
2566 RETALLOC_IF (reg_dummy
, num_regs
, const char *);
2567 RETALLOC_IF (reg_info_dummy
, num_regs
, register_info_type
);
2572 } /* regex_compile */
2574 /* Subroutines for `regex_compile'. */
2576 /* Store OP at LOC followed by two-byte integer parameter ARG. */
2579 store_op1 (op
, loc
, arg
)
2584 *loc
= (unsigned char) op
;
2585 STORE_NUMBER (loc
+ 1, arg
);
2589 /* Like `store_op1', but for two two-byte parameters ARG1 and ARG2. */
2592 store_op2 (op
, loc
, arg1
, arg2
)
2597 *loc
= (unsigned char) op
;
2598 STORE_NUMBER (loc
+ 1, arg1
);
2599 STORE_NUMBER (loc
+ 3, arg2
);
2603 /* Copy the bytes from LOC to END to open up three bytes of space at LOC
2604 for OP followed by two-byte integer parameter ARG. */
2607 insert_op1 (op
, loc
, arg
, end
)
2613 register unsigned char *pfrom
= end
;
2614 register unsigned char *pto
= end
+ 3;
2616 while (pfrom
!= loc
)
2619 store_op1 (op
, loc
, arg
);
2623 /* Like `insert_op1', but for two two-byte parameters ARG1 and ARG2. */
2626 insert_op2 (op
, loc
, arg1
, arg2
, end
)
2632 register unsigned char *pfrom
= end
;
2633 register unsigned char *pto
= end
+ 5;
2635 while (pfrom
!= loc
)
2638 store_op2 (op
, loc
, arg1
, arg2
);
2642 /* P points to just after a ^ in PATTERN. Return true if that ^ comes
2643 after an alternative or a begin-subexpression. We assume there is at
2644 least one character before the ^. */
2647 at_begline_loc_p (pattern
, p
, syntax
)
2648 const char *pattern
, *p
;
2649 reg_syntax_t syntax
;
2651 const char *prev
= p
- 2;
2652 boolean prev_prev_backslash
= prev
> pattern
&& prev
[-1] == '\\';
2655 /* After a subexpression? */
2656 (*prev
== '(' && (syntax
& RE_NO_BK_PARENS
|| prev_prev_backslash
))
2657 /* After an alternative? */
2658 || (*prev
== '|' && (syntax
& RE_NO_BK_VBAR
|| prev_prev_backslash
));
2662 /* The dual of at_begline_loc_p. This one is for $. We assume there is
2663 at least one character after the $, i.e., `P < PEND'. */
2666 at_endline_loc_p (p
, pend
, syntax
)
2667 const char *p
, *pend
;
2670 const char *next
= p
;
2671 boolean next_backslash
= *next
== '\\';
2672 const char *next_next
= p
+ 1 < pend
? p
+ 1 : NULL
;
2675 /* Before a subexpression? */
2676 (syntax
& RE_NO_BK_PARENS
? *next
== ')'
2677 : next_backslash
&& next_next
&& *next_next
== ')')
2678 /* Before an alternative? */
2679 || (syntax
& RE_NO_BK_VBAR
? *next
== '|'
2680 : next_backslash
&& next_next
&& *next_next
== '|');
2684 /* Returns true if REGNUM is in one of COMPILE_STACK's elements and
2685 false if it's not. */
2688 group_in_compile_stack (compile_stack
, regnum
)
2689 compile_stack_type compile_stack
;
2694 for (this_element
= compile_stack
.avail
- 1;
2697 if (compile_stack
.stack
[this_element
].regnum
== regnum
)
2704 /* Read the ending character of a range (in a bracket expression) from the
2705 uncompiled pattern *P_PTR (which ends at PEND). We assume the
2706 starting character is in `P[-2]'. (`P[-1]' is the character `-'.)
2707 Then we set the translation of all bits between the starting and
2708 ending characters (inclusive) in the compiled pattern B.
2710 Return an error code.
2712 We use these short variable names so we can use the same macros as
2713 `regex_compile' itself. */
2715 static reg_errcode_t
2716 compile_range (p_ptr
, pend
, translate
, syntax
, b
)
2717 const char **p_ptr
, *pend
;
2719 reg_syntax_t syntax
;
2724 const char *p
= *p_ptr
;
2725 int range_start
, range_end
;
2730 /* Even though the pattern is a signed `char *', we need to fetch
2731 with unsigned char *'s; if the high bit of the pattern character
2732 is set, the range endpoints will be negative if we fetch using a
2735 We also want to fetch the endpoints without translating them; the
2736 appropriate translation is done in the bit-setting loop below. */
2737 /* The SVR4 compiler on the 3B2 had trouble with unsigned const char *. */
2738 range_start
= ((const unsigned char *) p
)[-2];
2739 range_end
= ((const unsigned char *) p
)[0];
2741 /* Have to increment the pointer into the pattern string, so the
2742 caller isn't still at the ending character. */
2745 /* If the start is after the end, the range is empty. */
2746 if (range_start
> range_end
)
2747 return syntax
& RE_NO_EMPTY_RANGES
? REG_ERANGE
: REG_NOERROR
;
2749 /* Here we see why `this_char' has to be larger than an `unsigned
2750 char' -- the range is inclusive, so if `range_end' == 0xff
2751 (assuming 8-bit characters), we would otherwise go into an infinite
2752 loop, since all characters <= 0xff. */
2753 for (this_char
= range_start
; this_char
<= range_end
; this_char
++)
2755 SET_LIST_BIT (TRANSLATE (this_char
));
2761 /* re_compile_fastmap computes a ``fastmap'' for the compiled pattern in
2762 BUFP. A fastmap records which of the (1 << BYTEWIDTH) possible
2763 characters can start a string that matches the pattern. This fastmap
2764 is used by re_search to skip quickly over impossible starting points.
2766 The caller must supply the address of a (1 << BYTEWIDTH)-byte data
2767 area as BUFP->fastmap.
2769 We set the `fastmap', `fastmap_accurate', and `can_be_null' fields in
2772 Returns 0 if we succeed, -2 if an internal error. */
2775 re_compile_fastmap (bufp
)
2776 struct re_pattern_buffer
*bufp
;
2779 #ifdef MATCH_MAY_ALLOCATE
2780 fail_stack_type fail_stack
;
2782 #ifndef REGEX_MALLOC
2785 /* We don't push any register information onto the failure stack. */
2786 unsigned num_regs
= 0;
2788 register char *fastmap
= bufp
->fastmap
;
2789 unsigned char *pattern
= bufp
->buffer
;
2790 unsigned long size
= bufp
->used
;
2791 unsigned char *p
= pattern
;
2792 register unsigned char *pend
= pattern
+ size
;
2794 /* Assume that each path through the pattern can be null until
2795 proven otherwise. We set this false at the bottom of switch
2796 statement, to which we get only if a particular path doesn't
2797 match the empty string. */
2798 boolean path_can_be_null
= true;
2800 /* We aren't doing a `succeed_n' to begin with. */
2801 boolean succeed_n_p
= false;
2803 assert (fastmap
!= NULL
&& p
!= NULL
);
2806 bzero (fastmap
, 1 << BYTEWIDTH
); /* Assume nothing's valid. */
2807 bufp
->fastmap_accurate
= 1; /* It will be when we're done. */
2808 bufp
->can_be_null
= 0;
2812 if (p
== pend
|| *p
== succeed
)
2814 /* We have reached the (effective) end of pattern. */
2815 if (!FAIL_STACK_EMPTY ())
2817 bufp
->can_be_null
|= path_can_be_null
;
2819 /* Reset for next path. */
2820 path_can_be_null
= true;
2822 p
= fail_stack
.stack
[--fail_stack
.avail
];
2830 /* We should never be about to go beyond the end of the pattern. */
2833 switch (SWITCH_ENUM_CAST ((re_opcode_t
) *p
++))
2836 /* I guess the idea here is to simply not bother with a fastmap
2837 if a backreference is used, since it's too hard to figure out
2838 the fastmap for the corresponding group. Setting
2839 `can_be_null' stops `re_search_2' from using the fastmap, so
2840 that is all we do. */
2842 bufp
->can_be_null
= 1;
2846 /* Following are the cases which match a character. These end
2855 for (j
= *p
++ * BYTEWIDTH
- 1; j
>= 0; j
--)
2856 if (p
[j
/ BYTEWIDTH
] & (1 << (j
% BYTEWIDTH
)))
2862 /* Chars beyond end of map must be allowed. */
2863 for (j
= *p
* BYTEWIDTH
; j
< (1 << BYTEWIDTH
); j
++)
2866 for (j
= *p
++ * BYTEWIDTH
- 1; j
>= 0; j
--)
2867 if (!(p
[j
/ BYTEWIDTH
] & (1 << (j
% BYTEWIDTH
))))
2873 for (j
= 0; j
< (1 << BYTEWIDTH
); j
++)
2874 if (SYNTAX (j
) == Sword
)
2880 for (j
= 0; j
< (1 << BYTEWIDTH
); j
++)
2881 if (SYNTAX (j
) != Sword
)
2888 int fastmap_newline
= fastmap
['\n'];
2890 /* `.' matches anything ... */
2891 for (j
= 0; j
< (1 << BYTEWIDTH
); j
++)
2894 /* ... except perhaps newline. */
2895 if (!(bufp
->syntax
& RE_DOT_NEWLINE
))
2896 fastmap
['\n'] = fastmap_newline
;
2898 /* Return if we have already set `can_be_null'; if we have,
2899 then the fastmap is irrelevant. Something's wrong here. */
2900 else if (bufp
->can_be_null
)
2903 /* Otherwise, have to check alternative paths. */
2910 for (j
= 0; j
< (1 << BYTEWIDTH
); j
++)
2911 if (SYNTAX (j
) == (enum syntaxcode
) k
)
2918 for (j
= 0; j
< (1 << BYTEWIDTH
); j
++)
2919 if (SYNTAX (j
) != (enum syntaxcode
) k
)
2924 /* All cases after this match the empty string. These end with
2932 #endif /* not emacs */
2944 case push_dummy_failure
:
2949 case pop_failure_jump
:
2950 case maybe_pop_jump
:
2953 case dummy_failure_jump
:
2954 EXTRACT_NUMBER_AND_INCR (j
, p
);
2959 /* Jump backward implies we just went through the body of a
2960 loop and matched nothing. Opcode jumped to should be
2961 `on_failure_jump' or `succeed_n'. Just treat it like an
2962 ordinary jump. For a * loop, it has pushed its failure
2963 point already; if so, discard that as redundant. */
2964 if ((re_opcode_t
) *p
!= on_failure_jump
2965 && (re_opcode_t
) *p
!= succeed_n
)
2969 EXTRACT_NUMBER_AND_INCR (j
, p
);
2972 /* If what's on the stack is where we are now, pop it. */
2973 if (!FAIL_STACK_EMPTY ()
2974 && fail_stack
.stack
[fail_stack
.avail
- 1] == p
)
2980 case on_failure_jump
:
2981 case on_failure_keep_string_jump
:
2982 handle_on_failure_jump
:
2983 EXTRACT_NUMBER_AND_INCR (j
, p
);
2985 /* For some patterns, e.g., `(a?)?', `p+j' here points to the
2986 end of the pattern. We don't want to push such a point,
2987 since when we restore it above, entering the switch will
2988 increment `p' past the end of the pattern. We don't need
2989 to push such a point since we obviously won't find any more
2990 fastmap entries beyond `pend'. Such a pattern can match
2991 the null string, though. */
2994 if (!PUSH_PATTERN_OP (p
+ j
, fail_stack
))
2998 bufp
->can_be_null
= 1;
3002 EXTRACT_NUMBER_AND_INCR (k
, p
); /* Skip the n. */
3003 succeed_n_p
= false;
3010 /* Get to the number of times to succeed. */
3013 /* Increment p past the n for when k != 0. */
3014 EXTRACT_NUMBER_AND_INCR (k
, p
);
3018 succeed_n_p
= true; /* Spaghetti code alert. */
3019 goto handle_on_failure_jump
;
3036 abort (); /* We have listed all the cases. */
3039 /* Getting here means we have found the possible starting
3040 characters for one path of the pattern -- and that the empty
3041 string does not match. We need not follow this path further.
3042 Instead, look at the next alternative (remembered on the
3043 stack), or quit if no more. The test at the top of the loop
3044 does these things. */
3045 path_can_be_null
= false;
3049 /* Set `can_be_null' for the last path (also the first path, if the
3050 pattern is empty). */
3051 bufp
->can_be_null
|= path_can_be_null
;
3053 } /* re_compile_fastmap */
3055 /* Set REGS to hold NUM_REGS registers, storing them in STARTS and
3056 ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use
3057 this memory for recording register information. STARTS and ENDS
3058 must be allocated using the malloc library routine, and must each
3059 be at least NUM_REGS * sizeof (regoff_t) bytes long.
3061 If NUM_REGS == 0, then subsequent matches should allocate their own
3064 Unless this function is called, the first search or match using
3065 PATTERN_BUFFER will allocate its own register data, without
3066 freeing the old data. */
3069 re_set_registers (bufp
, regs
, num_regs
, starts
, ends
)
3070 struct re_pattern_buffer
*bufp
;
3071 struct re_registers
*regs
;
3073 regoff_t
*starts
, *ends
;
3077 bufp
->regs_allocated
= REGS_REALLOCATE
;
3078 regs
->num_regs
= num_regs
;
3079 regs
->start
= starts
;
3084 bufp
->regs_allocated
= REGS_UNALLOCATED
;
3086 regs
->start
= regs
->end
= (regoff_t
*) 0;
3090 /* Searching routines. */
3092 /* Like re_search_2, below, but only one string is specified, and
3093 doesn't let you say where to stop matching. */
3096 re_search (bufp
, string
, size
, startpos
, range
, regs
)
3097 struct re_pattern_buffer
*bufp
;
3099 int size
, startpos
, range
;
3100 struct re_registers
*regs
;
3102 return re_search_2 (bufp
, NULL
, 0, string
, size
, startpos
, range
,
3107 /* Using the compiled pattern in BUFP->buffer, first tries to match the
3108 virtual concatenation of STRING1 and STRING2, starting first at index
3109 STARTPOS, then at STARTPOS + 1, and so on.
3111 STRING1 and STRING2 have length SIZE1 and SIZE2, respectively.
3113 RANGE is how far to scan while trying to match. RANGE = 0 means try
3114 only at STARTPOS; in general, the last start tried is STARTPOS +
3117 In REGS, return the indices of the virtual concatenation of STRING1
3118 and STRING2 that matched the entire BUFP->buffer and its contained
3121 Do not consider matching one past the index STOP in the virtual
3122 concatenation of STRING1 and STRING2.
3124 We return either the position in the strings at which the match was
3125 found, -1 if no match, or -2 if error (such as failure
3129 re_search_2 (bufp
, string1
, size1
, string2
, size2
, startpos
, range
, regs
, stop
)
3130 struct re_pattern_buffer
*bufp
;
3131 const char *string1
, *string2
;
3135 struct re_registers
*regs
;
3139 register char *fastmap
= bufp
->fastmap
;
3140 register char *translate
= bufp
->translate
;
3141 int total_size
= size1
+ size2
;
3142 int endpos
= startpos
+ range
;
3144 /* Check for out-of-range STARTPOS. */
3145 if (startpos
< 0 || startpos
> total_size
)
3148 /* Fix up RANGE if it might eventually take us outside
3149 the virtual concatenation of STRING1 and STRING2. */
3151 range
= -1 - startpos
;
3152 else if (endpos
> total_size
)
3153 range
= total_size
- startpos
;
3155 /* If the search isn't to be a backwards one, don't waste time in a
3156 search for a pattern that must be anchored. */
3157 if (bufp
->used
> 0 && (re_opcode_t
) bufp
->buffer
[0] == begbuf
&& range
> 0)
3165 /* Update the fastmap now if not correct already. */
3166 if (fastmap
&& !bufp
->fastmap_accurate
)
3167 if (re_compile_fastmap (bufp
) == -2)
3170 /* Loop through the string, looking for a place to start matching. */
3173 /* If a fastmap is supplied, skip quickly over characters that
3174 cannot be the start of a match. If the pattern can match the
3175 null string, however, we don't need to skip characters; we want
3176 the first null string. */
3177 if (fastmap
&& startpos
< total_size
&& !bufp
->can_be_null
)
3179 if (range
> 0) /* Searching forwards. */
3181 register const char *d
;
3182 register int lim
= 0;
3185 if (startpos
< size1
&& startpos
+ range
>= size1
)
3186 lim
= range
- (size1
- startpos
);
3188 d
= (startpos
>= size1
? string2
- size1
: string1
) + startpos
;
3190 /* Written out as an if-else to avoid testing `translate'
3194 && !fastmap
[(unsigned char)
3195 translate
[(unsigned char) *d
++]])
3198 while (range
> lim
&& !fastmap
[(unsigned char) *d
++])
3201 startpos
+= irange
- range
;
3203 else /* Searching backwards. */
3205 register char c
= (size1
== 0 || startpos
>= size1
3206 ? string2
[startpos
- size1
]
3207 : string1
[startpos
]);
3209 if (!fastmap
[(unsigned char) TRANSLATE (c
)])
3214 /* If can't match the null string, and that's all we have left, fail. */
3215 if (range
>= 0 && startpos
== total_size
&& fastmap
3216 && !bufp
->can_be_null
)
3219 val
= re_match_2_internal (bufp
, string1
, size1
, string2
, size2
,
3220 startpos
, regs
, stop
);
3221 #ifndef REGEX_MALLOC
3250 /* Declarations and macros for re_match_2. */
3252 static int bcmp_translate ();
3253 static boolean
alt_match_null_string_p (),
3254 common_op_match_null_string_p (),
3255 group_match_null_string_p ();
3257 /* This converts PTR, a pointer into one of the search strings `string1'
3258 and `string2' into an offset from the beginning of that string. */
3259 #define POINTER_TO_OFFSET(ptr) \
3260 (FIRST_STRING_P (ptr) \
3261 ? ((regoff_t) ((ptr) - string1)) \
3262 : ((regoff_t) ((ptr) - string2 + size1)))
3264 /* Macros for dealing with the split strings in re_match_2. */
3266 #define MATCHING_IN_FIRST_STRING (dend == end_match_1)
3268 /* Call before fetching a character with *d. This switches over to
3269 string2 if necessary. */
3270 #define PREFETCH() \
3273 /* End of string2 => fail. */ \
3274 if (dend == end_match_2) \
3276 /* End of string1 => advance to string2. */ \
3278 dend = end_match_2; \
3282 /* Test if at very beginning or at very end of the virtual concatenation
3283 of `string1' and `string2'. If only one string, it's `string2'. */
3284 #define AT_STRINGS_BEG(d) ((d) == (size1 ? string1 : string2) || !size2)
3285 #define AT_STRINGS_END(d) ((d) == end2)
3288 /* Test if D points to a character which is word-constituent. We have
3289 two special cases to check for: if past the end of string1, look at
3290 the first character in string2; and if before the beginning of
3291 string2, look at the last character in string1. */
3292 #define WORDCHAR_P(d) \
3293 (SYNTAX ((d) == end1 ? *string2 \
3294 : (d) == string2 - 1 ? *(end1 - 1) : *(d)) \
3297 /* Test if the character before D and the one at D differ with respect
3298 to being word-constituent. */
3299 #define AT_WORD_BOUNDARY(d) \
3300 (AT_STRINGS_BEG (d) || AT_STRINGS_END (d) \
3301 || WORDCHAR_P (d - 1) != WORDCHAR_P (d))
3304 /* Free everything we malloc. */
3305 #ifdef MATCH_MAY_ALLOCATE
3307 #define FREE_VAR(var) if (var) free (var); var = NULL
3308 #define FREE_VARIABLES() \
3310 FREE_VAR (fail_stack.stack); \
3311 FREE_VAR (regstart); \
3312 FREE_VAR (regend); \
3313 FREE_VAR (old_regstart); \
3314 FREE_VAR (old_regend); \
3315 FREE_VAR (best_regstart); \
3316 FREE_VAR (best_regend); \
3317 FREE_VAR (reg_info); \
3318 FREE_VAR (reg_dummy); \
3319 FREE_VAR (reg_info_dummy); \
3321 #else /* not REGEX_MALLOC */
3322 /* This used to do alloca (0), but now we do that in the caller. */
3323 #define FREE_VARIABLES() /* Nothing */
3324 #endif /* not REGEX_MALLOC */
3326 #define FREE_VARIABLES() /* Do nothing! */
3327 #endif /* not MATCH_MAY_ALLOCATE */
3329 /* These values must meet several constraints. They must not be valid
3330 register values; since we have a limit of 255 registers (because
3331 we use only one byte in the pattern for the register number), we can
3332 use numbers larger than 255. They must differ by 1, because of
3333 NUM_FAILURE_ITEMS above. And the value for the lowest register must
3334 be larger than the value for the highest register, so we do not try
3335 to actually save any registers when none are active. */
3336 #define NO_HIGHEST_ACTIVE_REG (1 << BYTEWIDTH)
3337 #define NO_LOWEST_ACTIVE_REG (NO_HIGHEST_ACTIVE_REG + 1)
3339 /* Matching routines. */
3341 #ifndef emacs /* Emacs never uses this. */
3342 /* re_match is like re_match_2 except it takes only a single string. */
3345 re_match (bufp
, string
, size
, pos
, regs
)
3346 struct re_pattern_buffer
*bufp
;
3349 struct re_registers
*regs
;
3351 int result
= re_match_2_internal (bufp
, NULL
, 0, string
, size
,
3356 #endif /* not emacs */
3359 /* re_match_2 matches the compiled pattern in BUFP against the
3360 the (virtual) concatenation of STRING1 and STRING2 (of length SIZE1
3361 and SIZE2, respectively). We start matching at POS, and stop
3364 If REGS is non-null and the `no_sub' field of BUFP is nonzero, we
3365 store offsets for the substring each group matched in REGS. See the
3366 documentation for exactly how many groups we fill.
3368 We return -1 if no match, -2 if an internal error (such as the
3369 failure stack overflowing). Otherwise, we return the length of the
3370 matched substring. */
3373 re_match_2 (bufp
, string1
, size1
, string2
, size2
, pos
, regs
, stop
)
3374 struct re_pattern_buffer
*bufp
;
3375 const char *string1
, *string2
;
3378 struct re_registers
*regs
;
3381 int result
= re_match_2_internal (bufp
, string1
, size1
, string2
, size2
,
3387 /* This is a separate function so that we can force an alloca cleanup
3390 re_match_2_internal (bufp
, string1
, size1
, string2
, size2
, pos
, regs
, stop
)
3391 struct re_pattern_buffer
*bufp
;
3392 const char *string1
, *string2
;
3395 struct re_registers
*regs
;
3398 /* General temporaries. */
3402 /* Just past the end of the corresponding string. */
3403 const char *end1
, *end2
;
3405 /* Pointers into string1 and string2, just past the last characters in
3406 each to consider matching. */
3407 const char *end_match_1
, *end_match_2
;
3409 /* Where we are in the data, and the end of the current string. */
3410 const char *d
, *dend
;
3412 /* Where we are in the pattern, and the end of the pattern. */
3413 unsigned char *p
= bufp
->buffer
;
3414 register unsigned char *pend
= p
+ bufp
->used
;
3416 /* Mark the opcode just after a start_memory, so we can test for an
3417 empty subpattern when we get to the stop_memory. */
3418 unsigned char *just_past_start_mem
= 0;
3420 /* We use this to map every character in the string. */
3421 char *translate
= bufp
->translate
;
3423 /* Failure point stack. Each place that can handle a failure further
3424 down the line pushes a failure point on this stack. It consists of
3425 restart, regend, and reg_info for all registers corresponding to
3426 the subexpressions we're currently inside, plus the number of such
3427 registers, and, finally, two char *'s. The first char * is where
3428 to resume scanning the pattern; the second one is where to resume
3429 scanning the strings. If the latter is zero, the failure point is
3430 a ``dummy''; if a failure happens and the failure point is a dummy,
3431 it gets discarded and the next next one is tried. */
3432 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
3433 fail_stack_type fail_stack
;
3436 static unsigned failure_id
= 0;
3437 unsigned nfailure_points_pushed
= 0, nfailure_points_popped
= 0;
3440 /* We fill all the registers internally, independent of what we
3441 return, for use in backreferences. The number here includes
3442 an element for register zero. */
3443 unsigned num_regs
= bufp
->re_nsub
+ 1;
3445 /* The currently active registers. */
3446 unsigned lowest_active_reg
= NO_LOWEST_ACTIVE_REG
;
3447 unsigned highest_active_reg
= NO_HIGHEST_ACTIVE_REG
;
3449 /* Information on the contents of registers. These are pointers into
3450 the input strings; they record just what was matched (on this
3451 attempt) by a subexpression part of the pattern, that is, the
3452 regnum-th regstart pointer points to where in the pattern we began
3453 matching and the regnum-th regend points to right after where we
3454 stopped matching the regnum-th subexpression. (The zeroth register
3455 keeps track of what the whole pattern matches.) */
3456 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
3457 const char **regstart
, **regend
;
3460 /* If a group that's operated upon by a repetition operator fails to
3461 match anything, then the register for its start will need to be
3462 restored because it will have been set to wherever in the string we
3463 are when we last see its open-group operator. Similarly for a
3465 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
3466 const char **old_regstart
, **old_regend
;
3469 /* The is_active field of reg_info helps us keep track of which (possibly
3470 nested) subexpressions we are currently in. The matched_something
3471 field of reg_info[reg_num] helps us tell whether or not we have
3472 matched any of the pattern so far this time through the reg_num-th
3473 subexpression. These two fields get reset each time through any
3474 loop their register is in. */
3475 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
3476 register_info_type
*reg_info
;
3479 /* The following record the register info as found in the above
3480 variables when we find a match better than any we've seen before.
3481 This happens as we backtrack through the failure points, which in
3482 turn happens only if we have not yet matched the entire string. */
3483 unsigned best_regs_set
= false;
3484 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
3485 const char **best_regstart
, **best_regend
;
3488 /* Logically, this is `best_regend[0]'. But we don't want to have to
3489 allocate space for that if we're not allocating space for anything
3490 else (see below). Also, we never need info about register 0 for
3491 any of the other register vectors, and it seems rather a kludge to
3492 treat `best_regend' differently than the rest. So we keep track of
3493 the end of the best match so far in a separate variable. We
3494 initialize this to NULL so that when we backtrack the first time
3495 and need to test it, it's not garbage. */
3496 const char *match_end
= NULL
;
3498 /* This helps SET_REGS_MATCHED avoid doing redundant work. */
3499 int set_regs_matched_done
= 0;
3501 /* Used when we pop values we don't care about. */
3502 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
3503 const char **reg_dummy
;
3504 register_info_type
*reg_info_dummy
;
3508 /* Counts the total number of registers pushed. */
3509 unsigned num_regs_pushed
= 0;
3512 DEBUG_PRINT1 ("\n\nEntering re_match_2.\n");
3516 #ifdef MATCH_MAY_ALLOCATE
3517 /* Do not bother to initialize all the register variables if there are
3518 no groups in the pattern, as it takes a fair amount of time. If
3519 there are groups, we include space for register 0 (the whole
3520 pattern), even though we never use it, since it simplifies the
3521 array indexing. We should fix this. */
3524 regstart
= REGEX_TALLOC (num_regs
, const char *);
3525 regend
= REGEX_TALLOC (num_regs
, const char *);
3526 old_regstart
= REGEX_TALLOC (num_regs
, const char *);
3527 old_regend
= REGEX_TALLOC (num_regs
, const char *);
3528 best_regstart
= REGEX_TALLOC (num_regs
, const char *);
3529 best_regend
= REGEX_TALLOC (num_regs
, const char *);
3530 reg_info
= REGEX_TALLOC (num_regs
, register_info_type
);
3531 reg_dummy
= REGEX_TALLOC (num_regs
, const char *);
3532 reg_info_dummy
= REGEX_TALLOC (num_regs
, register_info_type
);
3534 if (!(regstart
&& regend
&& old_regstart
&& old_regend
&& reg_info
3535 && best_regstart
&& best_regend
&& reg_dummy
&& reg_info_dummy
))
3541 #if defined (REGEX_MALLOC)
3544 /* We must initialize all our variables to NULL, so that
3545 `FREE_VARIABLES' doesn't try to free them. */
3546 regstart
= regend
= old_regstart
= old_regend
= best_regstart
3547 = best_regend
= reg_dummy
= NULL
;
3548 reg_info
= reg_info_dummy
= (register_info_type
*) NULL
;
3550 #endif /* REGEX_MALLOC */
3551 #endif /* MATCH_MAY_ALLOCATE */
3553 /* The starting position is bogus. */
3554 if (pos
< 0 || pos
> size1
+ size2
)
3560 /* Initialize subexpression text positions to -1 to mark ones that no
3561 start_memory/stop_memory has been seen for. Also initialize the
3562 register information struct. */
3563 for (mcnt
= 1; mcnt
< num_regs
; mcnt
++)
3565 regstart
[mcnt
] = regend
[mcnt
]
3566 = old_regstart
[mcnt
] = old_regend
[mcnt
] = REG_UNSET_VALUE
;
3568 REG_MATCH_NULL_STRING_P (reg_info
[mcnt
]) = MATCH_NULL_UNSET_VALUE
;
3569 IS_ACTIVE (reg_info
[mcnt
]) = 0;
3570 MATCHED_SOMETHING (reg_info
[mcnt
]) = 0;
3571 EVER_MATCHED_SOMETHING (reg_info
[mcnt
]) = 0;
3574 /* We move `string1' into `string2' if the latter's empty -- but not if
3575 `string1' is null. */
3576 if (size2
== 0 && string1
!= NULL
)
3583 end1
= string1
+ size1
;
3584 end2
= string2
+ size2
;
3586 /* Compute where to stop matching, within the two strings. */
3589 end_match_1
= string1
+ stop
;
3590 end_match_2
= string2
;
3595 end_match_2
= string2
+ stop
- size1
;
3598 /* `p' scans through the pattern as `d' scans through the data.
3599 `dend' is the end of the input string that `d' points within. `d'
3600 is advanced into the following input string whenever necessary, but
3601 this happens before fetching; therefore, at the beginning of the
3602 loop, `d' can be pointing at the end of a string, but it cannot
3604 if (size1
> 0 && pos
<= size1
)
3611 d
= string2
+ pos
- size1
;
3615 DEBUG_PRINT1 ("The compiled pattern is: ");
3616 DEBUG_PRINT_COMPILED_PATTERN (bufp
, p
, pend
);
3617 DEBUG_PRINT1 ("The string to match is: `");
3618 DEBUG_PRINT_DOUBLE_STRING (d
, string1
, size1
, string2
, size2
);
3619 DEBUG_PRINT1 ("'\n");
3621 /* This loops over pattern commands. It exits by returning from the
3622 function if the match is complete, or it drops through if the match
3623 fails at this starting point in the input data. */
3626 DEBUG_PRINT2 ("\n0x%x: ", p
);
3629 { /* End of pattern means we might have succeeded. */
3630 DEBUG_PRINT1 ("end of pattern ... ");
3632 /* If we haven't matched the entire string, and we want the
3633 longest match, try backtracking. */
3634 if (d
!= end_match_2
)
3636 /* 1 if this match ends in the same string (string1 or string2)
3637 as the best previous match. */
3638 boolean same_str_p
= (FIRST_STRING_P (match_end
)
3639 == MATCHING_IN_FIRST_STRING
);
3640 /* 1 if this match is the best seen so far. */
3641 boolean best_match_p
;
3643 /* AIX compiler got confused when this was combined
3644 with the previous declaration. */
3646 best_match_p
= d
> match_end
;
3648 best_match_p
= !MATCHING_IN_FIRST_STRING
;
3650 DEBUG_PRINT1 ("backtracking.\n");
3652 if (!FAIL_STACK_EMPTY ())
3653 { /* More failure points to try. */
3655 /* If exceeds best match so far, save it. */
3656 if (!best_regs_set
|| best_match_p
)
3658 best_regs_set
= true;
3661 DEBUG_PRINT1 ("\nSAVING match as best so far.\n");
3663 for (mcnt
= 1; mcnt
< num_regs
; mcnt
++)
3665 best_regstart
[mcnt
] = regstart
[mcnt
];
3666 best_regend
[mcnt
] = regend
[mcnt
];
3672 /* If no failure points, don't restore garbage. And if
3673 last match is real best match, don't restore second
3675 else if (best_regs_set
&& !best_match_p
)
3678 /* Restore best match. It may happen that `dend ==
3679 end_match_1' while the restored d is in string2.
3680 For example, the pattern `x.*y.*z' against the
3681 strings `x-' and `y-z-', if the two strings are
3682 not consecutive in memory. */
3683 DEBUG_PRINT1 ("Restoring best registers.\n");
3686 dend
= ((d
>= string1
&& d
<= end1
)
3687 ? end_match_1
: end_match_2
);
3689 for (mcnt
= 1; mcnt
< num_regs
; mcnt
++)
3691 regstart
[mcnt
] = best_regstart
[mcnt
];
3692 regend
[mcnt
] = best_regend
[mcnt
];
3695 } /* d != end_match_2 */
3698 DEBUG_PRINT1 ("Accepting match.\n");
3700 /* If caller wants register contents data back, do it. */
3701 if (regs
&& !bufp
->no_sub
)
3703 /* Have the register data arrays been allocated? */
3704 if (bufp
->regs_allocated
== REGS_UNALLOCATED
)
3705 { /* No. So allocate them with malloc. We need one
3706 extra element beyond `num_regs' for the `-1' marker
3708 regs
->num_regs
= MAX (RE_NREGS
, num_regs
+ 1);
3709 regs
->start
= TALLOC (regs
->num_regs
, regoff_t
);
3710 regs
->end
= TALLOC (regs
->num_regs
, regoff_t
);
3711 if (regs
->start
== NULL
|| regs
->end
== NULL
)
3713 bufp
->regs_allocated
= REGS_REALLOCATE
;
3715 else if (bufp
->regs_allocated
== REGS_REALLOCATE
)
3716 { /* Yes. If we need more elements than were already
3717 allocated, reallocate them. If we need fewer, just
3719 if (regs
->num_regs
< num_regs
+ 1)
3721 regs
->num_regs
= num_regs
+ 1;
3722 RETALLOC (regs
->start
, regs
->num_regs
, regoff_t
);
3723 RETALLOC (regs
->end
, regs
->num_regs
, regoff_t
);
3724 if (regs
->start
== NULL
|| regs
->end
== NULL
)
3730 /* These braces fend off a "empty body in an else-statement"
3731 warning under GCC when assert expands to nothing. */
3732 assert (bufp
->regs_allocated
== REGS_FIXED
);
3735 /* Convert the pointer data in `regstart' and `regend' to
3736 indices. Register zero has to be set differently,
3737 since we haven't kept track of any info for it. */
3738 if (regs
->num_regs
> 0)
3740 regs
->start
[0] = pos
;
3741 regs
->end
[0] = (MATCHING_IN_FIRST_STRING
3742 ? ((regoff_t
) (d
- string1
))
3743 : ((regoff_t
) (d
- string2
+ size1
)));
3746 /* Go through the first `min (num_regs, regs->num_regs)'
3747 registers, since that is all we initialized. */
3748 for (mcnt
= 1; mcnt
< MIN (num_regs
, regs
->num_regs
); mcnt
++)
3750 if (REG_UNSET (regstart
[mcnt
]) || REG_UNSET (regend
[mcnt
]))
3751 regs
->start
[mcnt
] = regs
->end
[mcnt
] = -1;
3755 = (regoff_t
) POINTER_TO_OFFSET (regstart
[mcnt
]);
3757 = (regoff_t
) POINTER_TO_OFFSET (regend
[mcnt
]);
3761 /* If the regs structure we return has more elements than
3762 were in the pattern, set the extra elements to -1. If
3763 we (re)allocated the registers, this is the case,
3764 because we always allocate enough to have at least one
3766 for (mcnt
= num_regs
; mcnt
< regs
->num_regs
; mcnt
++)
3767 regs
->start
[mcnt
] = regs
->end
[mcnt
] = -1;
3768 } /* regs && !bufp->no_sub */
3771 DEBUG_PRINT4 ("%u failure points pushed, %u popped (%u remain).\n",
3772 nfailure_points_pushed
, nfailure_points_popped
,
3773 nfailure_points_pushed
- nfailure_points_popped
);
3774 DEBUG_PRINT2 ("%u registers pushed.\n", num_regs_pushed
);
3776 mcnt
= d
- pos
- (MATCHING_IN_FIRST_STRING
3780 DEBUG_PRINT2 ("Returning %d from re_match_2.\n", mcnt
);
3785 /* Otherwise match next pattern command. */
3786 switch (SWITCH_ENUM_CAST ((re_opcode_t
) *p
++))
3788 /* Ignore these. Used to ignore the n of succeed_n's which
3789 currently have n == 0. */
3791 DEBUG_PRINT1 ("EXECUTING no_op.\n");
3795 DEBUG_PRINT1 ("EXECUTING succeed.\n");
3798 /* Match the next n pattern characters exactly. The following
3799 byte in the pattern defines n, and the n bytes after that
3800 are the characters to match. */
3803 DEBUG_PRINT2 ("EXECUTING exactn %d.\n", mcnt
);
3805 /* This is written out as an if-else so we don't waste time
3806 testing `translate' inside the loop. */
3812 if (translate
[(unsigned char) *d
++] != (char) *p
++)
3822 if (*d
++ != (char) *p
++) goto fail
;
3826 SET_REGS_MATCHED ();
3830 /* Match any character except possibly a newline or a null. */
3832 DEBUG_PRINT1 ("EXECUTING anychar.\n");
3836 if ((!(bufp
->syntax
& RE_DOT_NEWLINE
) && TRANSLATE (*d
) == '\n')
3837 || (bufp
->syntax
& RE_DOT_NOT_NULL
&& TRANSLATE (*d
) == '\000'))
3840 SET_REGS_MATCHED ();
3841 DEBUG_PRINT2 (" Matched `%d'.\n", *d
);
3849 register unsigned char c
;
3850 boolean
not = (re_opcode_t
) *(p
- 1) == charset_not
;
3852 DEBUG_PRINT2 ("EXECUTING charset%s.\n", not ? "_not" : "");
3855 c
= TRANSLATE (*d
); /* The character to match. */
3857 /* Cast to `unsigned' instead of `unsigned char' in case the
3858 bit list is a full 32 bytes long. */
3859 if (c
< (unsigned) (*p
* BYTEWIDTH
)
3860 && p
[1 + c
/ BYTEWIDTH
] & (1 << (c
% BYTEWIDTH
)))
3865 if (!not) goto fail
;
3867 SET_REGS_MATCHED ();
3873 /* The beginning of a group is represented by start_memory.
3874 The arguments are the register number in the next byte, and the
3875 number of groups inner to this one in the next. The text
3876 matched within the group is recorded (in the internal
3877 registers data structure) under the register number. */
3879 DEBUG_PRINT3 ("EXECUTING start_memory %d (%d):\n", *p
, p
[1]);
3881 /* Find out if this group can match the empty string. */
3882 p1
= p
; /* To send to group_match_null_string_p. */
3884 if (REG_MATCH_NULL_STRING_P (reg_info
[*p
]) == MATCH_NULL_UNSET_VALUE
)
3885 REG_MATCH_NULL_STRING_P (reg_info
[*p
])
3886 = group_match_null_string_p (&p1
, pend
, reg_info
);
3888 /* Save the position in the string where we were the last time
3889 we were at this open-group operator in case the group is
3890 operated upon by a repetition operator, e.g., with `(a*)*b'
3891 against `ab'; then we want to ignore where we are now in
3892 the string in case this attempt to match fails. */
3893 old_regstart
[*p
] = REG_MATCH_NULL_STRING_P (reg_info
[*p
])
3894 ? REG_UNSET (regstart
[*p
]) ? d
: regstart
[*p
]
3896 DEBUG_PRINT2 (" old_regstart: %d\n",
3897 POINTER_TO_OFFSET (old_regstart
[*p
]));
3900 DEBUG_PRINT2 (" regstart: %d\n", POINTER_TO_OFFSET (regstart
[*p
]));
3902 IS_ACTIVE (reg_info
[*p
]) = 1;
3903 MATCHED_SOMETHING (reg_info
[*p
]) = 0;
3905 /* Clear this whenever we change the register activity status. */
3906 set_regs_matched_done
= 0;
3908 /* This is the new highest active register. */
3909 highest_active_reg
= *p
;
3911 /* If nothing was active before, this is the new lowest active
3913 if (lowest_active_reg
== NO_LOWEST_ACTIVE_REG
)
3914 lowest_active_reg
= *p
;
3916 /* Move past the register number and inner group count. */
3918 just_past_start_mem
= p
;
3923 /* The stop_memory opcode represents the end of a group. Its
3924 arguments are the same as start_memory's: the register
3925 number, and the number of inner groups. */
3927 DEBUG_PRINT3 ("EXECUTING stop_memory %d (%d):\n", *p
, p
[1]);
3929 /* We need to save the string position the last time we were at
3930 this close-group operator in case the group is operated
3931 upon by a repetition operator, e.g., with `((a*)*(b*)*)*'
3932 against `aba'; then we want to ignore where we are now in
3933 the string in case this attempt to match fails. */
3934 old_regend
[*p
] = REG_MATCH_NULL_STRING_P (reg_info
[*p
])
3935 ? REG_UNSET (regend
[*p
]) ? d
: regend
[*p
]
3937 DEBUG_PRINT2 (" old_regend: %d\n",
3938 POINTER_TO_OFFSET (old_regend
[*p
]));
3941 DEBUG_PRINT2 (" regend: %d\n", POINTER_TO_OFFSET (regend
[*p
]));
3943 /* This register isn't active anymore. */
3944 IS_ACTIVE (reg_info
[*p
]) = 0;
3946 /* Clear this whenever we change the register activity status. */
3947 set_regs_matched_done
= 0;
3949 /* If this was the only register active, nothing is active
3951 if (lowest_active_reg
== highest_active_reg
)
3953 lowest_active_reg
= NO_LOWEST_ACTIVE_REG
;
3954 highest_active_reg
= NO_HIGHEST_ACTIVE_REG
;
3957 { /* We must scan for the new highest active register, since
3958 it isn't necessarily one less than now: consider
3959 (a(b)c(d(e)f)g). When group 3 ends, after the f), the
3960 new highest active register is 1. */
3961 unsigned char r
= *p
- 1;
3962 while (r
> 0 && !IS_ACTIVE (reg_info
[r
]))
3965 /* If we end up at register zero, that means that we saved
3966 the registers as the result of an `on_failure_jump', not
3967 a `start_memory', and we jumped to past the innermost
3968 `stop_memory'. For example, in ((.)*) we save
3969 registers 1 and 2 as a result of the *, but when we pop
3970 back to the second ), we are at the stop_memory 1.
3971 Thus, nothing is active. */
3974 lowest_active_reg
= NO_LOWEST_ACTIVE_REG
;
3975 highest_active_reg
= NO_HIGHEST_ACTIVE_REG
;
3978 highest_active_reg
= r
;
3981 /* If just failed to match something this time around with a
3982 group that's operated on by a repetition operator, try to
3983 force exit from the ``loop'', and restore the register
3984 information for this group that we had before trying this
3986 if ((!MATCHED_SOMETHING (reg_info
[*p
])
3987 || just_past_start_mem
== p
- 1)
3990 boolean is_a_jump_n
= false;
3994 switch ((re_opcode_t
) *p1
++)
3998 case pop_failure_jump
:
3999 case maybe_pop_jump
:
4001 case dummy_failure_jump
:
4002 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
4012 /* If the next operation is a jump backwards in the pattern
4013 to an on_failure_jump right before the start_memory
4014 corresponding to this stop_memory, exit from the loop
4015 by forcing a failure after pushing on the stack the
4016 on_failure_jump's jump in the pattern, and d. */
4017 if (mcnt
< 0 && (re_opcode_t
) *p1
== on_failure_jump
4018 && (re_opcode_t
) p1
[3] == start_memory
&& p1
[4] == *p
)
4020 /* If this group ever matched anything, then restore
4021 what its registers were before trying this last
4022 failed match, e.g., with `(a*)*b' against `ab' for
4023 regstart[1], and, e.g., with `((a*)*(b*)*)*'
4024 against `aba' for regend[3].
4026 Also restore the registers for inner groups for,
4027 e.g., `((a*)(b*))*' against `aba' (register 3 would
4028 otherwise get trashed). */
4030 if (EVER_MATCHED_SOMETHING (reg_info
[*p
]))
4034 EVER_MATCHED_SOMETHING (reg_info
[*p
]) = 0;
4036 /* Restore this and inner groups' (if any) registers. */
4037 for (r
= *p
; r
< *p
+ *(p
+ 1); r
++)
4039 regstart
[r
] = old_regstart
[r
];
4041 /* xx why this test? */
4042 if (old_regend
[r
] >= regstart
[r
])
4043 regend
[r
] = old_regend
[r
];
4047 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
4048 PUSH_FAILURE_POINT (p1
+ mcnt
, d
, -2);
4054 /* Move past the register number and the inner group count. */
4059 /* \<digit> has been turned into a `duplicate' command which is
4060 followed by the numeric value of <digit> as the register number. */
4063 register const char *d2
, *dend2
;
4064 int regno
= *p
++; /* Get which register to match against. */
4065 DEBUG_PRINT2 ("EXECUTING duplicate %d.\n", regno
);
4067 /* Can't back reference a group which we've never matched. */
4068 if (REG_UNSET (regstart
[regno
]) || REG_UNSET (regend
[regno
]))
4071 /* Where in input to try to start matching. */
4072 d2
= regstart
[regno
];
4074 /* Where to stop matching; if both the place to start and
4075 the place to stop matching are in the same string, then
4076 set to the place to stop, otherwise, for now have to use
4077 the end of the first string. */
4079 dend2
= ((FIRST_STRING_P (regstart
[regno
])
4080 == FIRST_STRING_P (regend
[regno
]))
4081 ? regend
[regno
] : end_match_1
);
4084 /* If necessary, advance to next segment in register
4088 if (dend2
== end_match_2
) break;
4089 if (dend2
== regend
[regno
]) break;
4091 /* End of string1 => advance to string2. */
4093 dend2
= regend
[regno
];
4095 /* At end of register contents => success */
4096 if (d2
== dend2
) break;
4098 /* If necessary, advance to next segment in data. */
4101 /* How many characters left in this segment to match. */
4104 /* Want how many consecutive characters we can match in
4105 one shot, so, if necessary, adjust the count. */
4106 if (mcnt
> dend2
- d2
)
4109 /* Compare that many; failure if mismatch, else move
4112 ? bcmp_translate (d
, d2
, mcnt
, translate
)
4113 : bcmp (d
, d2
, mcnt
))
4115 d
+= mcnt
, d2
+= mcnt
;
4117 /* Do this because we've match some characters. */
4118 SET_REGS_MATCHED ();
4124 /* begline matches the empty string at the beginning of the string
4125 (unless `not_bol' is set in `bufp'), and, if
4126 `newline_anchor' is set, after newlines. */
4128 DEBUG_PRINT1 ("EXECUTING begline.\n");
4130 if (AT_STRINGS_BEG (d
))
4132 if (!bufp
->not_bol
) break;
4134 else if (d
[-1] == '\n' && bufp
->newline_anchor
)
4138 /* In all other cases, we fail. */
4142 /* endline is the dual of begline. */
4144 DEBUG_PRINT1 ("EXECUTING endline.\n");
4146 if (AT_STRINGS_END (d
))
4148 if (!bufp
->not_eol
) break;
4151 /* We have to ``prefetch'' the next character. */
4152 else if ((d
== end1
? *string2
: *d
) == '\n'
4153 && bufp
->newline_anchor
)
4160 /* Match at the very beginning of the data. */
4162 DEBUG_PRINT1 ("EXECUTING begbuf.\n");
4163 if (AT_STRINGS_BEG (d
))
4168 /* Match at the very end of the data. */
4170 DEBUG_PRINT1 ("EXECUTING endbuf.\n");
4171 if (AT_STRINGS_END (d
))
4176 /* on_failure_keep_string_jump is used to optimize `.*\n'. It
4177 pushes NULL as the value for the string on the stack. Then
4178 `pop_failure_point' will keep the current value for the
4179 string, instead of restoring it. To see why, consider
4180 matching `foo\nbar' against `.*\n'. The .* matches the foo;
4181 then the . fails against the \n. But the next thing we want
4182 to do is match the \n against the \n; if we restored the
4183 string value, we would be back at the foo.
4185 Because this is used only in specific cases, we don't need to
4186 check all the things that `on_failure_jump' does, to make
4187 sure the right things get saved on the stack. Hence we don't
4188 share its code. The only reason to push anything on the
4189 stack at all is that otherwise we would have to change
4190 `anychar's code to do something besides goto fail in this
4191 case; that seems worse than this. */
4192 case on_failure_keep_string_jump
:
4193 DEBUG_PRINT1 ("EXECUTING on_failure_keep_string_jump");
4195 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
4196 DEBUG_PRINT3 (" %d (to 0x%x):\n", mcnt
, p
+ mcnt
);
4198 PUSH_FAILURE_POINT (p
+ mcnt
, NULL
, -2);
4202 /* Uses of on_failure_jump:
4204 Each alternative starts with an on_failure_jump that points
4205 to the beginning of the next alternative. Each alternative
4206 except the last ends with a jump that in effect jumps past
4207 the rest of the alternatives. (They really jump to the
4208 ending jump of the following alternative, because tensioning
4209 these jumps is a hassle.)
4211 Repeats start with an on_failure_jump that points past both
4212 the repetition text and either the following jump or
4213 pop_failure_jump back to this on_failure_jump. */
4214 case on_failure_jump
:
4216 DEBUG_PRINT1 ("EXECUTING on_failure_jump");
4218 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
4219 DEBUG_PRINT3 (" %d (to 0x%x)", mcnt
, p
+ mcnt
);
4221 /* If this on_failure_jump comes right before a group (i.e.,
4222 the original * applied to a group), save the information
4223 for that group and all inner ones, so that if we fail back
4224 to this point, the group's information will be correct.
4225 For example, in \(a*\)*\1, we need the preceding group,
4226 and in \(\(a*\)b*\)\2, we need the inner group. */
4228 /* We can't use `p' to check ahead because we push
4229 a failure point to `p + mcnt' after we do this. */
4232 /* We need to skip no_op's before we look for the
4233 start_memory in case this on_failure_jump is happening as
4234 the result of a completed succeed_n, as in \(a\)\{1,3\}b\1
4236 while (p1
< pend
&& (re_opcode_t
) *p1
== no_op
)
4239 if (p1
< pend
&& (re_opcode_t
) *p1
== start_memory
)
4241 /* We have a new highest active register now. This will
4242 get reset at the start_memory we are about to get to,
4243 but we will have saved all the registers relevant to
4244 this repetition op, as described above. */
4245 highest_active_reg
= *(p1
+ 1) + *(p1
+ 2);
4246 if (lowest_active_reg
== NO_LOWEST_ACTIVE_REG
)
4247 lowest_active_reg
= *(p1
+ 1);
4250 DEBUG_PRINT1 (":\n");
4251 PUSH_FAILURE_POINT (p
+ mcnt
, d
, -2);
4255 /* A smart repeat ends with `maybe_pop_jump'.
4256 We change it to either `pop_failure_jump' or `jump'. */
4257 case maybe_pop_jump
:
4258 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
4259 DEBUG_PRINT2 ("EXECUTING maybe_pop_jump %d.\n", mcnt
);
4261 register unsigned char *p2
= p
;
4263 /* Compare the beginning of the repeat with what in the
4264 pattern follows its end. If we can establish that there
4265 is nothing that they would both match, i.e., that we
4266 would have to backtrack because of (as in, e.g., `a*a')
4267 then we can change to pop_failure_jump, because we'll
4268 never have to backtrack.
4270 This is not true in the case of alternatives: in
4271 `(a|ab)*' we do need to backtrack to the `ab' alternative
4272 (e.g., if the string was `ab'). But instead of trying to
4273 detect that here, the alternative has put on a dummy
4274 failure point which is what we will end up popping. */
4276 /* Skip over open/close-group commands.
4277 If what follows this loop is a ...+ construct,
4278 look at what begins its body, since we will have to
4279 match at least one of that. */
4283 && ((re_opcode_t
) *p2
== stop_memory
4284 || (re_opcode_t
) *p2
== start_memory
))
4286 else if (p2
+ 6 < pend
4287 && (re_opcode_t
) *p2
== dummy_failure_jump
)
4294 /* p1[0] ... p1[2] are the `on_failure_jump' corresponding
4295 to the `maybe_finalize_jump' of this case. Examine what
4298 /* If we're at the end of the pattern, we can change. */
4301 /* Consider what happens when matching ":\(.*\)"
4302 against ":/". I don't really understand this code
4304 p
[-3] = (unsigned char) pop_failure_jump
;
4306 (" End of pattern: change to `pop_failure_jump'.\n");
4309 else if ((re_opcode_t
) *p2
== exactn
4310 || (bufp
->newline_anchor
&& (re_opcode_t
) *p2
== endline
))
4312 register unsigned char c
4313 = *p2
== (unsigned char) endline
? '\n' : p2
[2];
4315 if ((re_opcode_t
) p1
[3] == exactn
&& p1
[5] != c
)
4317 p
[-3] = (unsigned char) pop_failure_jump
;
4318 DEBUG_PRINT3 (" %c != %c => pop_failure_jump.\n",
4322 else if ((re_opcode_t
) p1
[3] == charset
4323 || (re_opcode_t
) p1
[3] == charset_not
)
4325 int not = (re_opcode_t
) p1
[3] == charset_not
;
4327 if (c
< (unsigned char) (p1
[4] * BYTEWIDTH
)
4328 && p1
[5 + c
/ BYTEWIDTH
] & (1 << (c
% BYTEWIDTH
)))
4331 /* `not' is equal to 1 if c would match, which means
4332 that we can't change to pop_failure_jump. */
4335 p
[-3] = (unsigned char) pop_failure_jump
;
4336 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
4340 else if ((re_opcode_t
) *p2
== charset
)
4343 register unsigned char c
4344 = *p2
== (unsigned char) endline
? '\n' : p2
[2];
4347 if ((re_opcode_t
) p1
[3] == exactn
4348 && ! ((int) p2
[1] * BYTEWIDTH
> (int) p1
[4]
4349 && (p2
[1 + p1
[4] / BYTEWIDTH
]
4350 & (1 << (p1
[4] % BYTEWIDTH
)))))
4352 p
[-3] = (unsigned char) pop_failure_jump
;
4353 DEBUG_PRINT3 (" %c != %c => pop_failure_jump.\n",
4357 else if ((re_opcode_t
) p1
[3] == charset_not
)
4360 /* We win if the charset_not inside the loop
4361 lists every character listed in the charset after. */
4362 for (idx
= 0; idx
< (int) p2
[1]; idx
++)
4363 if (! (p2
[2 + idx
] == 0
4364 || (idx
< (int) p1
[4]
4365 && ((p2
[2 + idx
] & ~ p1
[5 + idx
]) == 0))))
4370 p
[-3] = (unsigned char) pop_failure_jump
;
4371 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
4374 else if ((re_opcode_t
) p1
[3] == charset
)
4377 /* We win if the charset inside the loop
4378 has no overlap with the one after the loop. */
4380 idx
< (int) p2
[1] && idx
< (int) p1
[4];
4382 if ((p2
[2 + idx
] & p1
[5 + idx
]) != 0)
4385 if (idx
== p2
[1] || idx
== p1
[4])
4387 p
[-3] = (unsigned char) pop_failure_jump
;
4388 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
4393 p
-= 2; /* Point at relative address again. */
4394 if ((re_opcode_t
) p
[-1] != pop_failure_jump
)
4396 p
[-1] = (unsigned char) jump
;
4397 DEBUG_PRINT1 (" Match => jump.\n");
4398 goto unconditional_jump
;
4400 /* Note fall through. */
4403 /* The end of a simple repeat has a pop_failure_jump back to
4404 its matching on_failure_jump, where the latter will push a
4405 failure point. The pop_failure_jump takes off failure
4406 points put on by this pop_failure_jump's matching
4407 on_failure_jump; we got through the pattern to here from the
4408 matching on_failure_jump, so didn't fail. */
4409 case pop_failure_jump
:
4411 /* We need to pass separate storage for the lowest and
4412 highest registers, even though we don't care about the
4413 actual values. Otherwise, we will restore only one
4414 register from the stack, since lowest will == highest in
4415 `pop_failure_point'. */
4416 unsigned dummy_low_reg
, dummy_high_reg
;
4417 unsigned char *pdummy
;
4420 DEBUG_PRINT1 ("EXECUTING pop_failure_jump.\n");
4421 POP_FAILURE_POINT (sdummy
, pdummy
,
4422 dummy_low_reg
, dummy_high_reg
,
4423 reg_dummy
, reg_dummy
, reg_info_dummy
);
4425 /* Note fall through. */
4428 /* Unconditionally jump (without popping any failure points). */
4431 EXTRACT_NUMBER_AND_INCR (mcnt
, p
); /* Get the amount to jump. */
4432 DEBUG_PRINT2 ("EXECUTING jump %d ", mcnt
);
4433 p
+= mcnt
; /* Do the jump. */
4434 DEBUG_PRINT2 ("(to 0x%x).\n", p
);
4438 /* We need this opcode so we can detect where alternatives end
4439 in `group_match_null_string_p' et al. */
4441 DEBUG_PRINT1 ("EXECUTING jump_past_alt.\n");
4442 goto unconditional_jump
;
4445 /* Normally, the on_failure_jump pushes a failure point, which
4446 then gets popped at pop_failure_jump. We will end up at
4447 pop_failure_jump, also, and with a pattern of, say, `a+', we
4448 are skipping over the on_failure_jump, so we have to push
4449 something meaningless for pop_failure_jump to pop. */
4450 case dummy_failure_jump
:
4451 DEBUG_PRINT1 ("EXECUTING dummy_failure_jump.\n");
4452 /* It doesn't matter what we push for the string here. What
4453 the code at `fail' tests is the value for the pattern. */
4454 PUSH_FAILURE_POINT (0, 0, -2);
4455 goto unconditional_jump
;
4458 /* At the end of an alternative, we need to push a dummy failure
4459 point in case we are followed by a `pop_failure_jump', because
4460 we don't want the failure point for the alternative to be
4461 popped. For example, matching `(a|ab)*' against `aab'
4462 requires that we match the `ab' alternative. */
4463 case push_dummy_failure
:
4464 DEBUG_PRINT1 ("EXECUTING push_dummy_failure.\n");
4465 /* See comments just above at `dummy_failure_jump' about the
4467 PUSH_FAILURE_POINT (0, 0, -2);
4470 /* Have to succeed matching what follows at least n times.
4471 After that, handle like `on_failure_jump'. */
4473 EXTRACT_NUMBER (mcnt
, p
+ 2);
4474 DEBUG_PRINT2 ("EXECUTING succeed_n %d.\n", mcnt
);
4477 /* Originally, this is how many times we HAVE to succeed. */
4482 STORE_NUMBER_AND_INCR (p
, mcnt
);
4483 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p
, mcnt
);
4487 DEBUG_PRINT2 (" Setting two bytes from 0x%x to no_op.\n", p
+2);
4488 p
[2] = (unsigned char) no_op
;
4489 p
[3] = (unsigned char) no_op
;
4495 EXTRACT_NUMBER (mcnt
, p
+ 2);
4496 DEBUG_PRINT2 ("EXECUTING jump_n %d.\n", mcnt
);
4498 /* Originally, this is how many times we CAN jump. */
4502 STORE_NUMBER (p
+ 2, mcnt
);
4503 goto unconditional_jump
;
4505 /* If don't have to jump any more, skip over the rest of command. */
4512 DEBUG_PRINT1 ("EXECUTING set_number_at.\n");
4514 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
4516 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
4517 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p1
, mcnt
);
4518 STORE_NUMBER (p1
, mcnt
);
4523 DEBUG_PRINT1 ("EXECUTING wordbound.\n");
4524 if (AT_WORD_BOUNDARY (d
))
4529 DEBUG_PRINT1 ("EXECUTING notwordbound.\n");
4530 if (AT_WORD_BOUNDARY (d
))
4535 DEBUG_PRINT1 ("EXECUTING wordbeg.\n");
4536 if (WORDCHAR_P (d
) && (AT_STRINGS_BEG (d
) || !WORDCHAR_P (d
- 1)))
4541 DEBUG_PRINT1 ("EXECUTING wordend.\n");
4542 if (!AT_STRINGS_BEG (d
) && WORDCHAR_P (d
- 1)
4543 && (!WORDCHAR_P (d
) || AT_STRINGS_END (d
)))
4549 DEBUG_PRINT1 ("EXECUTING before_dot.\n");
4550 if (PTR_CHAR_POS ((unsigned char *) d
) >= point
)
4555 DEBUG_PRINT1 ("EXECUTING at_dot.\n");
4556 if (PTR_CHAR_POS ((unsigned char *) d
) != point
)
4561 DEBUG_PRINT1 ("EXECUTING after_dot.\n");
4562 if (PTR_CHAR_POS ((unsigned char *) d
) <= point
)
4565 #if 0 /* not emacs19 */
4567 DEBUG_PRINT1 ("EXECUTING at_dot.\n");
4568 if (PTR_CHAR_POS ((unsigned char *) d
) + 1 != point
)
4571 #endif /* not emacs19 */
4574 DEBUG_PRINT2 ("EXECUTING syntaxspec %d.\n", mcnt
);
4579 DEBUG_PRINT1 ("EXECUTING Emacs wordchar.\n");
4583 /* Can't use *d++ here; SYNTAX may be an unsafe macro. */
4585 if (SYNTAX (d
[-1]) != (enum syntaxcode
) mcnt
)
4587 SET_REGS_MATCHED ();
4591 DEBUG_PRINT2 ("EXECUTING notsyntaxspec %d.\n", mcnt
);
4593 goto matchnotsyntax
;
4596 DEBUG_PRINT1 ("EXECUTING Emacs notwordchar.\n");
4600 /* Can't use *d++ here; SYNTAX may be an unsafe macro. */
4602 if (SYNTAX (d
[-1]) == (enum syntaxcode
) mcnt
)
4604 SET_REGS_MATCHED ();
4607 #else /* not emacs */
4609 DEBUG_PRINT1 ("EXECUTING non-Emacs wordchar.\n");
4611 if (!WORDCHAR_P (d
))
4613 SET_REGS_MATCHED ();
4618 DEBUG_PRINT1 ("EXECUTING non-Emacs notwordchar.\n");
4622 SET_REGS_MATCHED ();
4625 #endif /* not emacs */
4630 continue; /* Successfully executed one pattern command; keep going. */
4633 /* We goto here if a matching operation fails. */
4635 if (!FAIL_STACK_EMPTY ())
4636 { /* A restart point is known. Restore to that state. */
4637 DEBUG_PRINT1 ("\nFAIL:\n");
4638 POP_FAILURE_POINT (d
, p
,
4639 lowest_active_reg
, highest_active_reg
,
4640 regstart
, regend
, reg_info
);
4642 /* If this failure point is a dummy, try the next one. */
4646 /* If we failed to the end of the pattern, don't examine *p. */
4650 boolean is_a_jump_n
= false;
4652 /* If failed to a backwards jump that's part of a repetition
4653 loop, need to pop this failure point and use the next one. */
4654 switch ((re_opcode_t
) *p
)
4658 case maybe_pop_jump
:
4659 case pop_failure_jump
:
4662 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
4665 if ((is_a_jump_n
&& (re_opcode_t
) *p1
== succeed_n
)
4667 && (re_opcode_t
) *p1
== on_failure_jump
))
4675 if (d
>= string1
&& d
<= end1
)
4679 break; /* Matching at this starting point really fails. */
4683 goto restore_best_regs
;
4687 return -1; /* Failure to match. */
4690 /* Subroutine definitions for re_match_2. */
4693 /* We are passed P pointing to a register number after a start_memory.
4695 Return true if the pattern up to the corresponding stop_memory can
4696 match the empty string, and false otherwise.
4698 If we find the matching stop_memory, sets P to point to one past its number.
4699 Otherwise, sets P to an undefined byte less than or equal to END.
4701 We don't handle duplicates properly (yet). */
4704 group_match_null_string_p (p
, end
, reg_info
)
4705 unsigned char **p
, *end
;
4706 register_info_type
*reg_info
;
4709 /* Point to after the args to the start_memory. */
4710 unsigned char *p1
= *p
+ 2;
4714 /* Skip over opcodes that can match nothing, and return true or
4715 false, as appropriate, when we get to one that can't, or to the
4716 matching stop_memory. */
4718 switch ((re_opcode_t
) *p1
)
4720 /* Could be either a loop or a series of alternatives. */
4721 case on_failure_jump
:
4723 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
4725 /* If the next operation is not a jump backwards in the
4730 /* Go through the on_failure_jumps of the alternatives,
4731 seeing if any of the alternatives cannot match nothing.
4732 The last alternative starts with only a jump,
4733 whereas the rest start with on_failure_jump and end
4734 with a jump, e.g., here is the pattern for `a|b|c':
4736 /on_failure_jump/0/6/exactn/1/a/jump_past_alt/0/6
4737 /on_failure_jump/0/6/exactn/1/b/jump_past_alt/0/3
4740 So, we have to first go through the first (n-1)
4741 alternatives and then deal with the last one separately. */
4744 /* Deal with the first (n-1) alternatives, which start
4745 with an on_failure_jump (see above) that jumps to right
4746 past a jump_past_alt. */
4748 while ((re_opcode_t
) p1
[mcnt
-3] == jump_past_alt
)
4750 /* `mcnt' holds how many bytes long the alternative
4751 is, including the ending `jump_past_alt' and
4754 if (!alt_match_null_string_p (p1
, p1
+ mcnt
- 3,
4758 /* Move to right after this alternative, including the
4762 /* Break if it's the beginning of an n-th alternative
4763 that doesn't begin with an on_failure_jump. */
4764 if ((re_opcode_t
) *p1
!= on_failure_jump
)
4767 /* Still have to check that it's not an n-th
4768 alternative that starts with an on_failure_jump. */
4770 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
4771 if ((re_opcode_t
) p1
[mcnt
-3] != jump_past_alt
)
4773 /* Get to the beginning of the n-th alternative. */
4779 /* Deal with the last alternative: go back and get number
4780 of the `jump_past_alt' just before it. `mcnt' contains
4781 the length of the alternative. */
4782 EXTRACT_NUMBER (mcnt
, p1
- 2);
4784 if (!alt_match_null_string_p (p1
, p1
+ mcnt
, reg_info
))
4787 p1
+= mcnt
; /* Get past the n-th alternative. */
4793 assert (p1
[1] == **p
);
4799 if (!common_op_match_null_string_p (&p1
, end
, reg_info
))
4802 } /* while p1 < end */
4805 } /* group_match_null_string_p */
4808 /* Similar to group_match_null_string_p, but doesn't deal with alternatives:
4809 It expects P to be the first byte of a single alternative and END one
4810 byte past the last. The alternative can contain groups. */
4813 alt_match_null_string_p (p
, end
, reg_info
)
4814 unsigned char *p
, *end
;
4815 register_info_type
*reg_info
;
4818 unsigned char *p1
= p
;
4822 /* Skip over opcodes that can match nothing, and break when we get
4823 to one that can't. */
4825 switch ((re_opcode_t
) *p1
)
4828 case on_failure_jump
:
4830 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
4835 if (!common_op_match_null_string_p (&p1
, end
, reg_info
))
4838 } /* while p1 < end */
4841 } /* alt_match_null_string_p */
4844 /* Deals with the ops common to group_match_null_string_p and
4845 alt_match_null_string_p.
4847 Sets P to one after the op and its arguments, if any. */
4850 common_op_match_null_string_p (p
, end
, reg_info
)
4851 unsigned char **p
, *end
;
4852 register_info_type
*reg_info
;
4857 unsigned char *p1
= *p
;
4859 switch ((re_opcode_t
) *p1
++)
4879 assert (reg_no
> 0 && reg_no
<= MAX_REGNUM
);
4880 ret
= group_match_null_string_p (&p1
, end
, reg_info
);
4882 /* Have to set this here in case we're checking a group which
4883 contains a group and a back reference to it. */
4885 if (REG_MATCH_NULL_STRING_P (reg_info
[reg_no
]) == MATCH_NULL_UNSET_VALUE
)
4886 REG_MATCH_NULL_STRING_P (reg_info
[reg_no
]) = ret
;
4892 /* If this is an optimized succeed_n for zero times, make the jump. */
4894 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
4902 /* Get to the number of times to succeed. */
4904 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
4909 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
4917 if (!REG_MATCH_NULL_STRING_P (reg_info
[*p1
]))
4925 /* All other opcodes mean we cannot match the empty string. */
4931 } /* common_op_match_null_string_p */
4934 /* Return zero if TRANSLATE[S1] and TRANSLATE[S2] are identical for LEN
4935 bytes; nonzero otherwise. */
4938 bcmp_translate (s1
, s2
, len
, translate
)
4939 unsigned char *s1
, *s2
;
4943 register unsigned char *p1
= s1
, *p2
= s2
;
4946 if (translate
[*p1
++] != translate
[*p2
++]) return 1;
4952 /* Entry points for GNU code. */
4954 /* re_compile_pattern is the GNU regular expression compiler: it
4955 compiles PATTERN (of length SIZE) and puts the result in BUFP.
4956 Returns 0 if the pattern was valid, otherwise an error string.
4958 Assumes the `allocated' (and perhaps `buffer') and `translate' fields
4959 are set in BUFP on entry.
4961 We call regex_compile to do the actual compilation. */
4964 re_compile_pattern (pattern
, length
, bufp
)
4965 const char *pattern
;
4967 struct re_pattern_buffer
*bufp
;
4971 /* GNU code is written to assume at least RE_NREGS registers will be set
4972 (and at least one extra will be -1). */
4973 bufp
->regs_allocated
= REGS_UNALLOCATED
;
4975 /* And GNU code determines whether or not to get register information
4976 by passing null for the REGS argument to re_match, etc., not by
4980 /* Match anchors at newline. */
4981 bufp
->newline_anchor
= 1;
4983 ret
= regex_compile (pattern
, length
, re_syntax_options
, bufp
);
4987 return gettext (re_error_msgid
[(int) ret
]);
4990 /* Entry points compatible with 4.2 BSD regex library. We don't define
4991 them unless specifically requested. */
4993 #ifdef _REGEX_RE_COMP
4995 /* BSD has one and only one pattern buffer. */
4996 static struct re_pattern_buffer re_comp_buf
;
5006 if (!re_comp_buf
.buffer
)
5007 return gettext ("No previous regular expression");
5011 if (!re_comp_buf
.buffer
)
5013 re_comp_buf
.buffer
= (unsigned char *) malloc (200);
5014 if (re_comp_buf
.buffer
== NULL
)
5015 return gettext (re_error_msgid
[(int) REG_ESPACE
]);
5016 re_comp_buf
.allocated
= 200;
5018 re_comp_buf
.fastmap
= (char *) malloc (1 << BYTEWIDTH
);
5019 if (re_comp_buf
.fastmap
== NULL
)
5020 return gettext (re_error_msgid
[(int) REG_ESPACE
]);
5023 /* Since `re_exec' always passes NULL for the `regs' argument, we
5024 don't need to initialize the pattern buffer fields which affect it. */
5026 /* Match anchors at newlines. */
5027 re_comp_buf
.newline_anchor
= 1;
5029 ret
= regex_compile (s
, strlen (s
), re_syntax_options
, &re_comp_buf
);
5034 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
5035 return (char *) gettext (re_error_msgid
[(int) ret
]);
5043 const int len
= strlen (s
);
5045 0 <= re_search (&re_comp_buf
, s
, len
, 0, len
, (struct re_registers
*) 0);
5047 #endif /* _REGEX_RE_COMP */
5049 /* POSIX.2 functions. Don't define these for Emacs. */
5053 /* regcomp takes a regular expression as a string and compiles it.
5055 PREG is a regex_t *. We do not expect any fields to be initialized,
5056 since POSIX says we shouldn't. Thus, we set
5058 `buffer' to the compiled pattern;
5059 `used' to the length of the compiled pattern;
5060 `syntax' to RE_SYNTAX_POSIX_EXTENDED if the
5061 REG_EXTENDED bit in CFLAGS is set; otherwise, to
5062 RE_SYNTAX_POSIX_BASIC;
5063 `newline_anchor' to REG_NEWLINE being set in CFLAGS;
5064 `fastmap' and `fastmap_accurate' to zero;
5065 `re_nsub' to the number of subexpressions in PATTERN.
5067 PATTERN is the address of the pattern string.
5069 CFLAGS is a series of bits which affect compilation.
5071 If REG_EXTENDED is set, we use POSIX extended syntax; otherwise, we
5072 use POSIX basic syntax.
5074 If REG_NEWLINE is set, then . and [^...] don't match newline.
5075 Also, regexec will try a match beginning after every newline.
5077 If REG_ICASE is set, then we considers upper- and lowercase
5078 versions of letters to be equivalent when matching.
5080 If REG_NOSUB is set, then when PREG is passed to regexec, that
5081 routine will report only success or failure, and nothing about the
5084 It returns 0 if it succeeds, nonzero if it doesn't. (See regex.h for
5085 the return codes and their meanings.) */
5088 regcomp (preg
, pattern
, cflags
)
5090 const char *pattern
;
5095 = (cflags
& REG_EXTENDED
) ?
5096 RE_SYNTAX_POSIX_EXTENDED
: RE_SYNTAX_POSIX_BASIC
;
5098 /* regex_compile will allocate the space for the compiled pattern. */
5100 preg
->allocated
= 0;
5103 /* Don't bother to use a fastmap when searching. This simplifies the
5104 REG_NEWLINE case: if we used a fastmap, we'd have to put all the
5105 characters after newlines into the fastmap. This way, we just try
5109 if (cflags
& REG_ICASE
)
5113 preg
->translate
= (char *) malloc (CHAR_SET_SIZE
);
5114 if (preg
->translate
== NULL
)
5115 return (int) REG_ESPACE
;
5117 /* Map uppercase characters to corresponding lowercase ones. */
5118 for (i
= 0; i
< CHAR_SET_SIZE
; i
++)
5119 preg
->translate
[i
] = ISUPPER (i
) ? tolower (i
) : i
;
5122 preg
->translate
= NULL
;
5124 /* If REG_NEWLINE is set, newlines are treated differently. */
5125 if (cflags
& REG_NEWLINE
)
5126 { /* REG_NEWLINE implies neither . nor [^...] match newline. */
5127 syntax
&= ~RE_DOT_NEWLINE
;
5128 syntax
|= RE_HAT_LISTS_NOT_NEWLINE
;
5129 /* It also changes the matching behavior. */
5130 preg
->newline_anchor
= 1;
5133 preg
->newline_anchor
= 0;
5135 preg
->no_sub
= !!(cflags
& REG_NOSUB
);
5137 /* POSIX says a null character in the pattern terminates it, so we
5138 can use strlen here in compiling the pattern. */
5139 ret
= regex_compile (pattern
, strlen (pattern
), syntax
, preg
);
5141 /* POSIX doesn't distinguish between an unmatched open-group and an
5142 unmatched close-group: both are REG_EPAREN. */
5143 if (ret
== REG_ERPAREN
) ret
= REG_EPAREN
;
5149 /* regexec searches for a given pattern, specified by PREG, in the
5152 If NMATCH is zero or REG_NOSUB was set in the cflags argument to
5153 `regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at
5154 least NMATCH elements, and we set them to the offsets of the
5155 corresponding matched substrings.
5157 EFLAGS specifies `execution flags' which affect matching: if
5158 REG_NOTBOL is set, then ^ does not match at the beginning of the
5159 string; if REG_NOTEOL is set, then $ does not match at the end.
5161 We return 0 if we find a match and REG_NOMATCH if not. */
5164 regexec (preg
, string
, nmatch
, pmatch
, eflags
)
5165 const regex_t
*preg
;
5168 regmatch_t pmatch
[];
5172 struct re_registers regs
;
5173 regex_t private_preg
;
5174 int len
= strlen (string
);
5175 boolean want_reg_info
= !preg
->no_sub
&& nmatch
> 0;
5177 private_preg
= *preg
;
5179 private_preg
.not_bol
= !!(eflags
& REG_NOTBOL
);
5180 private_preg
.not_eol
= !!(eflags
& REG_NOTEOL
);
5182 /* The user has told us exactly how many registers to return
5183 information about, via `nmatch'. We have to pass that on to the
5184 matching routines. */
5185 private_preg
.regs_allocated
= REGS_FIXED
;
5189 regs
.num_regs
= nmatch
;
5190 regs
.start
= TALLOC (nmatch
, regoff_t
);
5191 regs
.end
= TALLOC (nmatch
, regoff_t
);
5192 if (regs
.start
== NULL
|| regs
.end
== NULL
)
5193 return (int) REG_NOMATCH
;
5196 /* Perform the searching operation. */
5197 ret
= re_search (&private_preg
, string
, len
,
5198 /* start: */ 0, /* range: */ len
,
5199 want_reg_info
? ®s
: (struct re_registers
*) 0);
5201 /* Copy the register information to the POSIX structure. */
5208 for (r
= 0; r
< nmatch
; r
++)
5210 pmatch
[r
].rm_so
= regs
.start
[r
];
5211 pmatch
[r
].rm_eo
= regs
.end
[r
];
5215 /* If we needed the temporary register info, free the space now. */
5220 /* We want zero return to mean success, unlike `re_search'. */
5221 return ret
>= 0 ? (int) REG_NOERROR
: (int) REG_NOMATCH
;
5225 /* Returns a message corresponding to an error code, ERRCODE, returned
5226 from either regcomp or regexec. We don't use PREG here. */
5229 regerror (errcode
, preg
, errbuf
, errbuf_size
)
5231 const regex_t
*preg
;
5239 || errcode
>= (sizeof (re_error_msgid
) / sizeof (re_error_msgid
[0])))
5240 /* Only error codes returned by the rest of the code should be passed
5241 to this routine. If we are given anything else, or if other regex
5242 code generates an invalid error code, then the program has a bug.
5243 Dump core so we can fix it. */
5246 msg
= gettext (re_error_msgid
[errcode
]);
5248 msg_size
= strlen (msg
) + 1; /* Includes the null. */
5250 if (errbuf_size
!= 0)
5252 if (msg_size
> errbuf_size
)
5254 strncpy (errbuf
, msg
, errbuf_size
- 1);
5255 errbuf
[errbuf_size
- 1] = 0;
5258 strcpy (errbuf
, msg
);
5265 /* Free dynamically allocated space used by PREG. */
5271 if (preg
->buffer
!= NULL
)
5272 free (preg
->buffer
);
5273 preg
->buffer
= NULL
;
5275 preg
->allocated
= 0;
5278 if (preg
->fastmap
!= NULL
)
5279 free (preg
->fastmap
);
5280 preg
->fastmap
= NULL
;
5281 preg
->fastmap_accurate
= 0;
5283 if (preg
->translate
!= NULL
)
5284 free (preg
->translate
);
5285 preg
->translate
= NULL
;
5288 #endif /* not emacs */
5292 make-backup-files: t
5294 trim-versions-without-asking: nil