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 /* The `emacs' switch turns on certain matching commands
37 that make sense only in Emacs. */
44 /* Emacs uses `NULL' as a predicate. */
57 /* We used to test for `BSTRING' here, but only GCC and Emacs define
58 `BSTRING', as far as I know, and neither of them use this code. */
59 #ifndef INHIBIT_STRING_HEADER
60 #if HAVE_STRING_H || STDC_HEADERS
63 #define bcmp(s1, s2, n) memcmp ((s1), (s2), (n))
66 #define bcopy(s, d, n) memcpy ((d), (s), (n))
69 #define bzero(s, n) memset ((s), 0, (n))
76 /* Define the syntax stuff for \<, \>, etc. */
78 /* This must be nonzero for the wordchar and notwordchar pattern
79 commands in re_match_2. */
86 extern char *re_syntax_table
;
88 #else /* not SYNTAX_TABLE */
90 /* How many characters in the character set. */
91 #define CHAR_SET_SIZE 256
93 static char re_syntax_table
[CHAR_SET_SIZE
];
104 bzero (re_syntax_table
, sizeof re_syntax_table
);
106 for (c
= 'a'; c
<= 'z'; c
++)
107 re_syntax_table
[c
] = Sword
;
109 for (c
= 'A'; c
<= 'Z'; c
++)
110 re_syntax_table
[c
] = Sword
;
112 for (c
= '0'; c
<= '9'; c
++)
113 re_syntax_table
[c
] = Sword
;
115 re_syntax_table
['_'] = Sword
;
120 #endif /* not SYNTAX_TABLE */
122 #define SYNTAX(c) re_syntax_table[c]
124 #endif /* not emacs */
126 /* Get the interface, including the syntax bits. */
129 /* isalpha etc. are used for the character classes. */
132 /* Jim Meyering writes:
134 "... Some ctype macros are valid only for character codes that
135 isascii says are ASCII (SGI's IRIX-4.0.5 is one such system --when
136 using /bin/cc or gcc but without giving an ansi option). So, all
137 ctype uses should be through macros like ISPRINT... If
138 STDC_HEADERS is defined, then autoconf has verified that the ctype
139 macros don't need to be guarded with references to isascii. ...
140 Defining isascii to 1 should let any compiler worth its salt
141 eliminate the && through constant folding." */
143 #if defined (STDC_HEADERS) || (!defined (isascii) && !defined (HAVE_ISASCII))
146 #define ISASCII(c) isascii(c)
150 #define ISBLANK(c) (ISASCII (c) && isblank (c))
152 #define ISBLANK(c) ((c) == ' ' || (c) == '\t')
155 #define ISGRAPH(c) (ISASCII (c) && isgraph (c))
157 #define ISGRAPH(c) (ISASCII (c) && isprint (c) && !isspace (c))
160 #define ISPRINT(c) (ISASCII (c) && isprint (c))
161 #define ISDIGIT(c) (ISASCII (c) && isdigit (c))
162 #define ISALNUM(c) (ISASCII (c) && isalnum (c))
163 #define ISALPHA(c) (ISASCII (c) && isalpha (c))
164 #define ISCNTRL(c) (ISASCII (c) && iscntrl (c))
165 #define ISLOWER(c) (ISASCII (c) && islower (c))
166 #define ISPUNCT(c) (ISASCII (c) && ispunct (c))
167 #define ISSPACE(c) (ISASCII (c) && isspace (c))
168 #define ISUPPER(c) (ISASCII (c) && isupper (c))
169 #define ISXDIGIT(c) (ISASCII (c) && isxdigit (c))
175 /* We remove any previous definition of `SIGN_EXTEND_CHAR',
176 since ours (we hope) works properly with all combinations of
177 machines, compilers, `char' and `unsigned char' argument types.
178 (Per Bothner suggested the basic approach.) */
179 #undef SIGN_EXTEND_CHAR
181 #define SIGN_EXTEND_CHAR(c) ((signed char) (c))
182 #else /* not __STDC__ */
183 /* As in Harbison and Steele. */
184 #define SIGN_EXTEND_CHAR(c) ((((unsigned char) (c)) ^ 128) - 128)
187 /* Should we use malloc or alloca? If REGEX_MALLOC is not defined, we
188 use `alloca' instead of `malloc'. This is because using malloc in
189 re_search* or re_match* could cause memory leaks when C-g is used in
190 Emacs; also, malloc is slower and causes storage fragmentation. On
191 the other hand, malloc is more portable, and easier to debug.
193 Because we sometimes use alloca, some routines have to be macros,
194 not functions -- `alloca'-allocated space disappears at the end of the
195 function it is called in. */
199 #define REGEX_ALLOCATE malloc
200 #define REGEX_REALLOCATE(source, osize, nsize) realloc (source, nsize)
202 #else /* not REGEX_MALLOC */
204 /* Emacs already defines alloca, sometimes. */
207 /* Make alloca work the best possible way. */
209 #define alloca __builtin_alloca
210 #else /* not __GNUC__ */
213 #else /* not __GNUC__ or HAVE_ALLOCA_H */
214 #ifndef _AIX /* Already did AIX, up at the top. */
216 #endif /* not _AIX */
217 #endif /* not HAVE_ALLOCA_H */
218 #endif /* not __GNUC__ */
220 #endif /* not alloca */
222 #define REGEX_ALLOCATE alloca
224 /* Assumes a `char *destination' variable. */
225 #define REGEX_REALLOCATE(source, osize, nsize) \
226 (destination = (char *) alloca (nsize), \
227 bcopy (source, destination, osize), \
230 #endif /* not REGEX_MALLOC */
233 /* True if `size1' is non-NULL and PTR is pointing anywhere inside
234 `string1' or just past its end. This works if PTR is NULL, which is
236 #define FIRST_STRING_P(ptr) \
237 (size1 && string1 <= (ptr) && (ptr) <= string1 + size1)
239 /* (Re)Allocate N items of type T using malloc, or fail. */
240 #define TALLOC(n, t) ((t *) malloc ((n) * sizeof (t)))
241 #define RETALLOC(addr, n, t) ((addr) = (t *) realloc (addr, (n) * sizeof (t)))
242 #define RETALLOC_IF(addr, n, t) \
243 if (addr) RETALLOC((addr), (n), t); else (addr) = TALLOC ((n), t)
244 #define REGEX_TALLOC(n, t) ((t *) REGEX_ALLOCATE ((n) * sizeof (t)))
246 #define BYTEWIDTH 8 /* In bits. */
248 #define STREQ(s1, s2) ((strcmp (s1, s2) == 0))
252 #define MAX(a, b) ((a) > (b) ? (a) : (b))
253 #define MIN(a, b) ((a) < (b) ? (a) : (b))
255 typedef char boolean
;
259 static int re_match_2_internal ();
261 /* These are the command codes that appear in compiled regular
262 expressions. Some opcodes are followed by argument bytes. A
263 command code can specify any interpretation whatsoever for its
264 arguments. Zero bytes may appear in the compiled regular expression. */
270 /* Followed by one byte giving n, then by n literal bytes. */
273 /* Matches any (more or less) character. */
276 /* Matches any one char belonging to specified set. First
277 following byte is number of bitmap bytes. Then come bytes
278 for a bitmap saying which chars are in. Bits in each byte
279 are ordered low-bit-first. A character is in the set if its
280 bit is 1. A character too large to have a bit in the map is
281 automatically not in the set. */
284 /* Same parameters as charset, but match any character that is
285 not one of those specified. */
288 /* Start remembering the text that is matched, for storing in a
289 register. Followed by one byte with the register number, in
290 the range 0 to one less than the pattern buffer's re_nsub
291 field. Then followed by one byte with the number of groups
292 inner to this one. (This last has to be part of the
293 start_memory only because we need it in the on_failure_jump
297 /* Stop remembering the text that is matched and store it in a
298 memory register. Followed by one byte with the register
299 number, in the range 0 to one less than `re_nsub' in the
300 pattern buffer, and one byte with the number of inner groups,
301 just like `start_memory'. (We need the number of inner
302 groups here because we don't have any easy way of finding the
303 corresponding start_memory when we're at a stop_memory.) */
306 /* Match a duplicate of something remembered. Followed by one
307 byte containing the register number. */
310 /* Fail unless at beginning of line. */
313 /* Fail unless at end of line. */
316 /* Succeeds if at beginning of buffer (if emacs) or at beginning
317 of string to be matched (if not). */
320 /* Analogously, for end of buffer/string. */
323 /* Followed by two byte relative address to which to jump. */
326 /* Same as jump, but marks the end of an alternative. */
329 /* Followed by two-byte relative address of place to resume at
330 in case of failure. */
333 /* Like on_failure_jump, but pushes a placeholder instead of the
334 current string position when executed. */
335 on_failure_keep_string_jump
,
337 /* Throw away latest failure point and then jump to following
338 two-byte relative address. */
341 /* Change to pop_failure_jump if know won't have to backtrack to
342 match; otherwise change to jump. This is used to jump
343 back to the beginning of a repeat. If what follows this jump
344 clearly won't match what the repeat does, such that we can be
345 sure that there is no use backtracking out of repetitions
346 already matched, then we change it to a pop_failure_jump.
347 Followed by two-byte address. */
350 /* Jump to following two-byte address, and push a dummy failure
351 point. This failure point will be thrown away if an attempt
352 is made to use it for a failure. A `+' construct makes this
353 before the first repeat. Also used as an intermediary kind
354 of jump when compiling an alternative. */
357 /* Push a dummy failure point and continue. Used at the end of
361 /* Followed by two-byte relative address and two-byte number n.
362 After matching N times, jump to the address upon failure. */
365 /* Followed by two-byte relative address, and two-byte number n.
366 Jump to the address N times, then fail. */
369 /* Set the following two-byte relative address to the
370 subsequent two-byte number. The address *includes* the two
374 wordchar
, /* Matches any word-constituent character. */
375 notwordchar
, /* Matches any char that is not a word-constituent. */
377 wordbeg
, /* Succeeds if at word beginning. */
378 wordend
, /* Succeeds if at word end. */
380 wordbound
, /* Succeeds if at a word boundary. */
381 notwordbound
/* Succeeds if not at a word boundary. */
384 ,before_dot
, /* Succeeds if before point. */
385 at_dot
, /* Succeeds if at point. */
386 after_dot
, /* Succeeds if after point. */
388 /* Matches any character whose syntax is specified. Followed by
389 a byte which contains a syntax code, e.g., Sword. */
392 /* Matches any character whose syntax is not that specified. */
397 /* Common operations on the compiled pattern. */
399 /* Store NUMBER in two contiguous bytes starting at DESTINATION. */
401 #define STORE_NUMBER(destination, number) \
403 (destination)[0] = (number) & 0377; \
404 (destination)[1] = (number) >> 8; \
407 /* Same as STORE_NUMBER, except increment DESTINATION to
408 the byte after where the number is stored. Therefore, DESTINATION
409 must be an lvalue. */
411 #define STORE_NUMBER_AND_INCR(destination, number) \
413 STORE_NUMBER (destination, number); \
414 (destination) += 2; \
417 /* Put into DESTINATION a number stored in two contiguous bytes starting
420 #define EXTRACT_NUMBER(destination, source) \
422 (destination) = *(source) & 0377; \
423 (destination) += SIGN_EXTEND_CHAR (*((source) + 1)) << 8; \
428 extract_number (dest
, source
)
430 unsigned char *source
;
432 int temp
= SIGN_EXTEND_CHAR (*(source
+ 1));
433 *dest
= *source
& 0377;
437 #ifndef EXTRACT_MACROS /* To debug the macros. */
438 #undef EXTRACT_NUMBER
439 #define EXTRACT_NUMBER(dest, src) extract_number (&dest, src)
440 #endif /* not EXTRACT_MACROS */
444 /* Same as EXTRACT_NUMBER, except increment SOURCE to after the number.
445 SOURCE must be an lvalue. */
447 #define EXTRACT_NUMBER_AND_INCR(destination, source) \
449 EXTRACT_NUMBER (destination, source); \
455 extract_number_and_incr (destination
, source
)
457 unsigned char **source
;
459 extract_number (destination
, *source
);
463 #ifndef EXTRACT_MACROS
464 #undef EXTRACT_NUMBER_AND_INCR
465 #define EXTRACT_NUMBER_AND_INCR(dest, src) \
466 extract_number_and_incr (&dest, &src)
467 #endif /* not EXTRACT_MACROS */
471 /* If DEBUG is defined, Regex prints many voluminous messages about what
472 it is doing (if the variable `debug' is nonzero). If linked with the
473 main program in `iregex.c', you can enter patterns and strings
474 interactively. And if linked with the main program in `main.c' and
475 the other test files, you can run the already-written tests. */
479 /* We use standard I/O for debugging. */
482 /* It is useful to test things that ``must'' be true when debugging. */
485 static int debug
= 0;
487 #define DEBUG_STATEMENT(e) e
488 #define DEBUG_PRINT1(x) if (debug) printf (x)
489 #define DEBUG_PRINT2(x1, x2) if (debug) printf (x1, x2)
490 #define DEBUG_PRINT3(x1, x2, x3) if (debug) printf (x1, x2, x3)
491 #define DEBUG_PRINT4(x1, x2, x3, x4) if (debug) printf (x1, x2, x3, x4)
492 #define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) \
493 if (debug) print_partial_compiled_pattern (s, e)
494 #define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) \
495 if (debug) print_double_string (w, s1, sz1, s2, sz2)
498 extern void printchar ();
500 /* Print the fastmap in human-readable form. */
503 print_fastmap (fastmap
)
506 unsigned was_a_range
= 0;
509 while (i
< (1 << BYTEWIDTH
))
515 while (i
< (1 << BYTEWIDTH
) && fastmap
[i
])
531 /* Print a compiled pattern string in human-readable form, starting at
532 the START pointer into it and ending just before the pointer END. */
535 print_partial_compiled_pattern (start
, end
)
536 unsigned char *start
;
540 unsigned char *p
= start
;
541 unsigned char *pend
= end
;
549 /* Loop over pattern commands. */
552 printf ("%d:\t", p
- start
);
554 switch ((re_opcode_t
) *p
++)
562 printf ("/exactn/%d", mcnt
);
573 printf ("/start_memory/%d/%d", mcnt
, *p
++);
578 printf ("/stop_memory/%d/%d", mcnt
, *p
++);
582 printf ("/duplicate/%d", *p
++);
592 register int c
, last
= -100;
593 register int in_range
= 0;
595 printf ("/charset [%s",
596 (re_opcode_t
) *(p
- 1) == charset_not
? "^" : "");
598 assert (p
+ *p
< pend
);
600 for (c
= 0; c
< 256; c
++)
602 && (p
[1 + (c
/8)] & (1 << (c
% 8))))
604 /* Are we starting a range? */
605 if (last
+ 1 == c
&& ! in_range
)
610 /* Have we broken a range? */
611 else if (last
+ 1 != c
&& in_range
)
640 case on_failure_jump
:
641 extract_number_and_incr (&mcnt
, &p
);
642 printf ("/on_failure_jump to %d", p
+ mcnt
- start
);
645 case on_failure_keep_string_jump
:
646 extract_number_and_incr (&mcnt
, &p
);
647 printf ("/on_failure_keep_string_jump to %d", p
+ mcnt
- start
);
650 case dummy_failure_jump
:
651 extract_number_and_incr (&mcnt
, &p
);
652 printf ("/dummy_failure_jump to %d", p
+ mcnt
- start
);
655 case push_dummy_failure
:
656 printf ("/push_dummy_failure");
660 extract_number_and_incr (&mcnt
, &p
);
661 printf ("/maybe_pop_jump to %d", p
+ mcnt
- start
);
664 case pop_failure_jump
:
665 extract_number_and_incr (&mcnt
, &p
);
666 printf ("/pop_failure_jump to %d", p
+ mcnt
- start
);
670 extract_number_and_incr (&mcnt
, &p
);
671 printf ("/jump_past_alt to %d", p
+ mcnt
- start
);
675 extract_number_and_incr (&mcnt
, &p
);
676 printf ("/jump to %d", p
+ mcnt
- start
);
680 extract_number_and_incr (&mcnt
, &p
);
681 extract_number_and_incr (&mcnt2
, &p
);
682 printf ("/succeed_n to %d, %d times", p
+ mcnt
- start
, mcnt2
);
686 extract_number_and_incr (&mcnt
, &p
);
687 extract_number_and_incr (&mcnt2
, &p
);
688 printf ("/jump_n to %d, %d times", p
+ mcnt
- start
, mcnt2
);
692 extract_number_and_incr (&mcnt
, &p
);
693 extract_number_and_incr (&mcnt2
, &p
);
694 printf ("/set_number_at location %d to %d", p
+ mcnt
- start
, mcnt2
);
698 printf ("/wordbound");
702 printf ("/notwordbound");
714 printf ("/before_dot");
722 printf ("/after_dot");
726 printf ("/syntaxspec");
728 printf ("/%d", mcnt
);
732 printf ("/notsyntaxspec");
734 printf ("/%d", mcnt
);
739 printf ("/wordchar");
743 printf ("/notwordchar");
755 printf ("?%d", *(p
-1));
761 printf ("%d:\tend of pattern.\n", p
- start
);
766 print_compiled_pattern (bufp
)
767 struct re_pattern_buffer
*bufp
;
769 unsigned char *buffer
= bufp
->buffer
;
771 print_partial_compiled_pattern (buffer
, buffer
+ bufp
->used
);
772 printf ("%d bytes used/%d bytes allocated.\n", bufp
->used
, bufp
->allocated
);
774 if (bufp
->fastmap_accurate
&& bufp
->fastmap
)
776 printf ("fastmap: ");
777 print_fastmap (bufp
->fastmap
);
780 printf ("re_nsub: %d\t", bufp
->re_nsub
);
781 printf ("regs_alloc: %d\t", bufp
->regs_allocated
);
782 printf ("can_be_null: %d\t", bufp
->can_be_null
);
783 printf ("newline_anchor: %d\n", bufp
->newline_anchor
);
784 printf ("no_sub: %d\t", bufp
->no_sub
);
785 printf ("not_bol: %d\t", bufp
->not_bol
);
786 printf ("not_eol: %d\t", bufp
->not_eol
);
787 printf ("syntax: %d\n", bufp
->syntax
);
788 /* Perhaps we should print the translate table? */
793 print_double_string (where
, string1
, size1
, string2
, size2
)
806 if (FIRST_STRING_P (where
))
808 for (this_char
= where
- string1
; this_char
< size1
; this_char
++)
809 printchar (string1
[this_char
]);
814 for (this_char
= where
- string2
; this_char
< size2
; this_char
++)
815 printchar (string2
[this_char
]);
819 #else /* not DEBUG */
824 #define DEBUG_STATEMENT(e)
825 #define DEBUG_PRINT1(x)
826 #define DEBUG_PRINT2(x1, x2)
827 #define DEBUG_PRINT3(x1, x2, x3)
828 #define DEBUG_PRINT4(x1, x2, x3, x4)
829 #define DEBUG_PRINT_COMPILED_PATTERN(p, s, e)
830 #define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2)
832 #endif /* not DEBUG */
834 /* Set by `re_set_syntax' to the current regexp syntax to recognize. Can
835 also be assigned to arbitrarily: each pattern buffer stores its own
836 syntax, so it can be changed between regex compilations. */
837 reg_syntax_t re_syntax_options
= RE_SYNTAX_EMACS
;
840 /* Specify the precise syntax of regexps for compilation. This provides
841 for compatibility for various utilities which historically have
842 different, incompatible syntaxes.
844 The argument SYNTAX is a bit mask comprised of the various bits
845 defined in regex.h. We return the old syntax. */
848 re_set_syntax (syntax
)
851 reg_syntax_t ret
= re_syntax_options
;
853 re_syntax_options
= syntax
;
857 /* This table gives an error message for each of the error codes listed
858 in regex.h. Obviously the order here has to be same as there. */
860 static const char *re_error_msg
[] =
861 { NULL
, /* REG_NOERROR */
862 "No match", /* REG_NOMATCH */
863 "Invalid regular expression", /* REG_BADPAT */
864 "Invalid collation character", /* REG_ECOLLATE */
865 "Invalid character class name", /* REG_ECTYPE */
866 "Trailing backslash", /* REG_EESCAPE */
867 "Invalid back reference", /* REG_ESUBREG */
868 "Unmatched [ or [^", /* REG_EBRACK */
869 "Unmatched ( or \\(", /* REG_EPAREN */
870 "Unmatched \\{", /* REG_EBRACE */
871 "Invalid content of \\{\\}", /* REG_BADBR */
872 "Invalid range end", /* REG_ERANGE */
873 "Memory exhausted", /* REG_ESPACE */
874 "Invalid preceding regular expression", /* REG_BADRPT */
875 "Premature end of regular expression", /* REG_EEND */
876 "Regular expression too big", /* REG_ESIZE */
877 "Unmatched ) or \\)", /* REG_ERPAREN */
880 /* Avoiding alloca during matching, to placate r_alloc. */
882 /* Define MATCH_MAY_ALLOCATE unless we need to make sure that the
883 searching and matching functions should not call alloca. On some
884 systems, alloca is implemented in terms of malloc, and if we're
885 using the relocating allocator routines, then malloc could cause a
886 relocation, which might (if the strings being searched are in the
887 ralloc heap) shift the data out from underneath the regexp
890 Here's another reason to avoid allocation: Emacs
891 processes input from X in a signal handler; processing X input may
892 call malloc; if input arrives while a matching routine is calling
893 malloc, then we're scrod. But Emacs can't just block input while
894 calling matching routines; then we don't notice interrupts when
895 they come in. So, Emacs blocks input around all regexp calls
896 except the matching calls, which it leaves unprotected, in the
897 faith that they will not malloc. */
899 /* Normally, this is fine. */
900 #define MATCH_MAY_ALLOCATE
902 /* The match routines may not allocate if (1) they would do it with malloc
903 and (2) it's not safe for them to use malloc. */
904 #if (defined (C_ALLOCA) || defined (REGEX_MALLOC)) && (defined (emacs) || defined (REL_ALLOC))
905 #undef MATCH_MAY_ALLOCATE
909 /* Failure stack declarations and macros; both re_compile_fastmap and
910 re_match_2 use a failure stack. These have to be macros because of
914 /* Number of failure points for which to initially allocate space
915 when matching. If this number is exceeded, we allocate more
916 space, so it is not a hard limit. */
917 #ifndef INIT_FAILURE_ALLOC
918 #define INIT_FAILURE_ALLOC 5
921 /* Roughly the maximum number of failure points on the stack. Would be
922 exactly that if always used MAX_FAILURE_SPACE each time we failed.
923 This is a variable only so users of regex can assign to it; we never
924 change it ourselves. */
925 int re_max_failures
= 2000;
927 typedef unsigned char *fail_stack_elt_t
;
931 fail_stack_elt_t
*stack
;
933 unsigned avail
; /* Offset of next open position. */
936 #define FAIL_STACK_EMPTY() (fail_stack.avail == 0)
937 #define FAIL_STACK_PTR_EMPTY() (fail_stack_ptr->avail == 0)
938 #define FAIL_STACK_FULL() (fail_stack.avail == fail_stack.size)
939 #define FAIL_STACK_TOP() (fail_stack.stack[fail_stack.avail])
942 /* Initialize `fail_stack'. Do `return -2' if the alloc fails. */
944 #ifdef MATCH_MAY_ALLOCATE
945 #define INIT_FAIL_STACK() \
947 fail_stack.stack = (fail_stack_elt_t *) \
948 REGEX_ALLOCATE (INIT_FAILURE_ALLOC * sizeof (fail_stack_elt_t)); \
950 if (fail_stack.stack == NULL) \
953 fail_stack.size = INIT_FAILURE_ALLOC; \
954 fail_stack.avail = 0; \
957 #define INIT_FAIL_STACK() \
959 fail_stack.avail = 0; \
964 /* Double the size of FAIL_STACK, up to approximately `re_max_failures' items.
966 Return 1 if succeeds, and 0 if either ran out of memory
967 allocating space for it or it was already too large.
969 REGEX_REALLOCATE requires `destination' be declared. */
971 #define DOUBLE_FAIL_STACK(fail_stack) \
972 ((fail_stack).size > re_max_failures * MAX_FAILURE_ITEMS \
974 : ((fail_stack).stack = (fail_stack_elt_t *) \
975 REGEX_REALLOCATE ((fail_stack).stack, \
976 (fail_stack).size * sizeof (fail_stack_elt_t), \
977 ((fail_stack).size << 1) * sizeof (fail_stack_elt_t)), \
979 (fail_stack).stack == NULL \
981 : ((fail_stack).size <<= 1, \
985 /* Push PATTERN_OP on FAIL_STACK.
987 Return 1 if was able to do so and 0 if ran out of memory allocating
989 #define PUSH_PATTERN_OP(pattern_op, fail_stack) \
990 ((FAIL_STACK_FULL () \
991 && !DOUBLE_FAIL_STACK (fail_stack)) \
993 : ((fail_stack).stack[(fail_stack).avail++] = pattern_op, \
996 /* This pushes an item onto the failure stack. Must be a four-byte
997 value. Assumes the variable `fail_stack'. Probably should only
998 be called from within `PUSH_FAILURE_POINT'. */
999 #define PUSH_FAILURE_ITEM(item) \
1000 fail_stack.stack[fail_stack.avail++] = (fail_stack_elt_t) item
1002 /* The complement operation. Assumes `fail_stack' is nonempty. */
1003 #define POP_FAILURE_ITEM() fail_stack.stack[--fail_stack.avail]
1005 /* Used to omit pushing failure point id's when we're not debugging. */
1007 #define DEBUG_PUSH PUSH_FAILURE_ITEM
1008 #define DEBUG_POP(item_addr) *(item_addr) = POP_FAILURE_ITEM ()
1010 #define DEBUG_PUSH(item)
1011 #define DEBUG_POP(item_addr)
1015 /* Push the information about the state we will need
1016 if we ever fail back to it.
1018 Requires variables fail_stack, regstart, regend, reg_info, and
1019 num_regs be declared. DOUBLE_FAIL_STACK requires `destination' be
1022 Does `return FAILURE_CODE' if runs out of memory. */
1024 #define PUSH_FAILURE_POINT(pattern_place, string_place, failure_code) \
1026 char *destination; \
1027 /* Must be int, so when we don't save any registers, the arithmetic \
1028 of 0 + -1 isn't done as unsigned. */ \
1031 DEBUG_STATEMENT (failure_id++); \
1032 DEBUG_STATEMENT (nfailure_points_pushed++); \
1033 DEBUG_PRINT2 ("\nPUSH_FAILURE_POINT #%u:\n", failure_id); \
1034 DEBUG_PRINT2 (" Before push, next avail: %d\n", (fail_stack).avail);\
1035 DEBUG_PRINT2 (" size: %d\n", (fail_stack).size);\
1037 DEBUG_PRINT2 (" slots needed: %d\n", NUM_FAILURE_ITEMS); \
1038 DEBUG_PRINT2 (" available: %d\n", REMAINING_AVAIL_SLOTS); \
1040 /* Ensure we have enough space allocated for what we will push. */ \
1041 while (REMAINING_AVAIL_SLOTS < NUM_FAILURE_ITEMS) \
1043 if (!DOUBLE_FAIL_STACK (fail_stack)) \
1044 return failure_code; \
1046 DEBUG_PRINT2 ("\n Doubled stack; size now: %d\n", \
1047 (fail_stack).size); \
1048 DEBUG_PRINT2 (" slots available: %d\n", REMAINING_AVAIL_SLOTS);\
1051 /* Push the info, starting with the registers. */ \
1052 DEBUG_PRINT1 ("\n"); \
1054 for (this_reg = lowest_active_reg; this_reg <= highest_active_reg; \
1057 DEBUG_PRINT2 (" Pushing reg: %d\n", this_reg); \
1058 DEBUG_STATEMENT (num_regs_pushed++); \
1060 DEBUG_PRINT2 (" start: 0x%x\n", regstart[this_reg]); \
1061 PUSH_FAILURE_ITEM (regstart[this_reg]); \
1063 DEBUG_PRINT2 (" end: 0x%x\n", regend[this_reg]); \
1064 PUSH_FAILURE_ITEM (regend[this_reg]); \
1066 DEBUG_PRINT2 (" info: 0x%x\n ", reg_info[this_reg]); \
1067 DEBUG_PRINT2 (" match_null=%d", \
1068 REG_MATCH_NULL_STRING_P (reg_info[this_reg])); \
1069 DEBUG_PRINT2 (" active=%d", IS_ACTIVE (reg_info[this_reg])); \
1070 DEBUG_PRINT2 (" matched_something=%d", \
1071 MATCHED_SOMETHING (reg_info[this_reg])); \
1072 DEBUG_PRINT2 (" ever_matched=%d", \
1073 EVER_MATCHED_SOMETHING (reg_info[this_reg])); \
1074 DEBUG_PRINT1 ("\n"); \
1075 PUSH_FAILURE_ITEM (reg_info[this_reg].word); \
1078 DEBUG_PRINT2 (" Pushing low active reg: %d\n", lowest_active_reg);\
1079 PUSH_FAILURE_ITEM (lowest_active_reg); \
1081 DEBUG_PRINT2 (" Pushing high active reg: %d\n", highest_active_reg);\
1082 PUSH_FAILURE_ITEM (highest_active_reg); \
1084 DEBUG_PRINT2 (" Pushing pattern 0x%x: ", pattern_place); \
1085 DEBUG_PRINT_COMPILED_PATTERN (bufp, pattern_place, pend); \
1086 PUSH_FAILURE_ITEM (pattern_place); \
1088 DEBUG_PRINT2 (" Pushing string 0x%x: `", string_place); \
1089 DEBUG_PRINT_DOUBLE_STRING (string_place, string1, size1, string2, \
1091 DEBUG_PRINT1 ("'\n"); \
1092 PUSH_FAILURE_ITEM (string_place); \
1094 DEBUG_PRINT2 (" Pushing failure id: %u\n", failure_id); \
1095 DEBUG_PUSH (failure_id); \
1098 /* This is the number of items that are pushed and popped on the stack
1099 for each register. */
1100 #define NUM_REG_ITEMS 3
1102 /* Individual items aside from the registers. */
1104 #define NUM_NONREG_ITEMS 5 /* Includes failure point id. */
1106 #define NUM_NONREG_ITEMS 4
1109 /* We push at most this many items on the stack. */
1110 #define MAX_FAILURE_ITEMS ((num_regs - 1) * NUM_REG_ITEMS + NUM_NONREG_ITEMS)
1112 /* We actually push this many items. */
1113 #define NUM_FAILURE_ITEMS \
1114 ((highest_active_reg - lowest_active_reg + 1) * NUM_REG_ITEMS \
1117 /* How many items can still be added to the stack without overflowing it. */
1118 #define REMAINING_AVAIL_SLOTS ((fail_stack).size - (fail_stack).avail)
1121 /* Pops what PUSH_FAIL_STACK pushes.
1123 We restore into the parameters, all of which should be lvalues:
1124 STR -- the saved data position.
1125 PAT -- the saved pattern position.
1126 LOW_REG, HIGH_REG -- the highest and lowest active registers.
1127 REGSTART, REGEND -- arrays of string positions.
1128 REG_INFO -- array of information about each subexpression.
1130 Also assumes the variables `fail_stack' and (if debugging), `bufp',
1131 `pend', `string1', `size1', `string2', and `size2'. */
1133 #define POP_FAILURE_POINT(str, pat, low_reg, high_reg, regstart, regend, reg_info)\
1135 DEBUG_STATEMENT (fail_stack_elt_t failure_id;) \
1137 const unsigned char *string_temp; \
1139 assert (!FAIL_STACK_EMPTY ()); \
1141 /* Remove failure points and point to how many regs pushed. */ \
1142 DEBUG_PRINT1 ("POP_FAILURE_POINT:\n"); \
1143 DEBUG_PRINT2 (" Before pop, next avail: %d\n", fail_stack.avail); \
1144 DEBUG_PRINT2 (" size: %d\n", fail_stack.size); \
1146 assert (fail_stack.avail >= NUM_NONREG_ITEMS); \
1148 DEBUG_POP (&failure_id); \
1149 DEBUG_PRINT2 (" Popping failure id: %u\n", failure_id); \
1151 /* If the saved string location is NULL, it came from an \
1152 on_failure_keep_string_jump opcode, and we want to throw away the \
1153 saved NULL, thus retaining our current position in the string. */ \
1154 string_temp = POP_FAILURE_ITEM (); \
1155 if (string_temp != NULL) \
1156 str = (const char *) string_temp; \
1158 DEBUG_PRINT2 (" Popping string 0x%x: `", str); \
1159 DEBUG_PRINT_DOUBLE_STRING (str, string1, size1, string2, size2); \
1160 DEBUG_PRINT1 ("'\n"); \
1162 pat = (unsigned char *) POP_FAILURE_ITEM (); \
1163 DEBUG_PRINT2 (" Popping pattern 0x%x: ", pat); \
1164 DEBUG_PRINT_COMPILED_PATTERN (bufp, pat, pend); \
1166 /* Restore register info. */ \
1167 high_reg = (unsigned) POP_FAILURE_ITEM (); \
1168 DEBUG_PRINT2 (" Popping high active reg: %d\n", high_reg); \
1170 low_reg = (unsigned) POP_FAILURE_ITEM (); \
1171 DEBUG_PRINT2 (" Popping low active reg: %d\n", low_reg); \
1173 for (this_reg = high_reg; this_reg >= low_reg; this_reg--) \
1175 DEBUG_PRINT2 (" Popping reg: %d\n", this_reg); \
1177 reg_info[this_reg].word = POP_FAILURE_ITEM (); \
1178 DEBUG_PRINT2 (" info: 0x%x\n", reg_info[this_reg]); \
1180 regend[this_reg] = (const char *) POP_FAILURE_ITEM (); \
1181 DEBUG_PRINT2 (" end: 0x%x\n", regend[this_reg]); \
1183 regstart[this_reg] = (const char *) POP_FAILURE_ITEM (); \
1184 DEBUG_PRINT2 (" start: 0x%x\n", regstart[this_reg]); \
1187 DEBUG_STATEMENT (nfailure_points_popped++); \
1188 } /* POP_FAILURE_POINT */
1192 /* Structure for per-register (a.k.a. per-group) information.
1193 This must not be longer than one word, because we push this value
1194 onto the failure stack. Other register information, such as the
1195 starting and ending positions (which are addresses), and the list of
1196 inner groups (which is a bits list) are maintained in separate
1199 We are making a (strictly speaking) nonportable assumption here: that
1200 the compiler will pack our bit fields into something that fits into
1201 the type of `word', i.e., is something that fits into one item on the
1205 fail_stack_elt_t word
;
1208 /* This field is one if this group can match the empty string,
1209 zero if not. If not yet determined, `MATCH_NULL_UNSET_VALUE'. */
1210 #define MATCH_NULL_UNSET_VALUE 3
1211 unsigned match_null_string_p
: 2;
1212 unsigned is_active
: 1;
1213 unsigned matched_something
: 1;
1214 unsigned ever_matched_something
: 1;
1216 } register_info_type
;
1218 #define REG_MATCH_NULL_STRING_P(R) ((R).bits.match_null_string_p)
1219 #define IS_ACTIVE(R) ((R).bits.is_active)
1220 #define MATCHED_SOMETHING(R) ((R).bits.matched_something)
1221 #define EVER_MATCHED_SOMETHING(R) ((R).bits.ever_matched_something)
1224 /* Call this when have matched a real character; it sets `matched' flags
1225 for the subexpressions which we are currently inside. Also records
1226 that those subexprs have matched. */
1227 #define SET_REGS_MATCHED() \
1231 for (r = lowest_active_reg; r <= highest_active_reg; r++) \
1233 MATCHED_SOMETHING (reg_info[r]) \
1234 = EVER_MATCHED_SOMETHING (reg_info[r]) \
1241 /* Registers are set to a sentinel when they haven't yet matched. */
1242 #define REG_UNSET_VALUE ((char *) -1)
1243 #define REG_UNSET(e) ((e) == REG_UNSET_VALUE)
1247 /* How do we implement a missing MATCH_MAY_ALLOCATE?
1248 We make the fail stack a global thing, and then grow it to
1249 re_max_failures when we compile. */
1250 #ifndef MATCH_MAY_ALLOCATE
1251 static fail_stack_type fail_stack
;
1253 static const char ** regstart
, ** regend
;
1254 static const char ** old_regstart
, ** old_regend
;
1255 static const char **best_regstart
, **best_regend
;
1256 static register_info_type
*reg_info
;
1257 static const char **reg_dummy
;
1258 static register_info_type
*reg_info_dummy
;
1262 /* Subroutine declarations and macros for regex_compile. */
1264 static void store_op1 (), store_op2 ();
1265 static void insert_op1 (), insert_op2 ();
1266 static boolean
at_begline_loc_p (), at_endline_loc_p ();
1267 static boolean
group_in_compile_stack ();
1268 static reg_errcode_t
compile_range ();
1270 /* Fetch the next character in the uncompiled pattern---translating it
1271 if necessary. Also cast from a signed character in the constant
1272 string passed to us by the user to an unsigned char that we can use
1273 as an array index (in, e.g., `translate'). */
1274 #define PATFETCH(c) \
1275 do {if (p == pend) return REG_EEND; \
1276 c = (unsigned char) *p++; \
1277 if (translate) c = translate[c]; \
1280 /* Fetch the next character in the uncompiled pattern, with no
1282 #define PATFETCH_RAW(c) \
1283 do {if (p == pend) return REG_EEND; \
1284 c = (unsigned char) *p++; \
1287 /* Go backwards one character in the pattern. */
1288 #define PATUNFETCH p--
1291 /* If `translate' is non-null, return translate[D], else just D. We
1292 cast the subscript to translate because some data is declared as
1293 `char *', to avoid warnings when a string constant is passed. But
1294 when we use a character as a subscript we must make it unsigned. */
1295 #define TRANSLATE(d) (translate ? translate[(unsigned char) (d)] : (d))
1298 /* Macros for outputting the compiled pattern into `buffer'. */
1300 /* If the buffer isn't allocated when it comes in, use this. */
1301 #define INIT_BUF_SIZE 32
1303 /* Make sure we have at least N more bytes of space in buffer. */
1304 #define GET_BUFFER_SPACE(n) \
1305 while (b - bufp->buffer + (n) > bufp->allocated) \
1308 /* Make sure we have one more byte of buffer space and then add C to it. */
1309 #define BUF_PUSH(c) \
1311 GET_BUFFER_SPACE (1); \
1312 *b++ = (unsigned char) (c); \
1316 /* Ensure we have two more bytes of buffer space and then append C1 and C2. */
1317 #define BUF_PUSH_2(c1, c2) \
1319 GET_BUFFER_SPACE (2); \
1320 *b++ = (unsigned char) (c1); \
1321 *b++ = (unsigned char) (c2); \
1325 /* As with BUF_PUSH_2, except for three bytes. */
1326 #define BUF_PUSH_3(c1, c2, c3) \
1328 GET_BUFFER_SPACE (3); \
1329 *b++ = (unsigned char) (c1); \
1330 *b++ = (unsigned char) (c2); \
1331 *b++ = (unsigned char) (c3); \
1335 /* Store a jump with opcode OP at LOC to location TO. We store a
1336 relative address offset by the three bytes the jump itself occupies. */
1337 #define STORE_JUMP(op, loc, to) \
1338 store_op1 (op, loc, (to) - (loc) - 3)
1340 /* Likewise, for a two-argument jump. */
1341 #define STORE_JUMP2(op, loc, to, arg) \
1342 store_op2 (op, loc, (to) - (loc) - 3, arg)
1344 /* Like `STORE_JUMP', but for inserting. Assume `b' is the buffer end. */
1345 #define INSERT_JUMP(op, loc, to) \
1346 insert_op1 (op, loc, (to) - (loc) - 3, b)
1348 /* Like `STORE_JUMP2', but for inserting. Assume `b' is the buffer end. */
1349 #define INSERT_JUMP2(op, loc, to, arg) \
1350 insert_op2 (op, loc, (to) - (loc) - 3, arg, b)
1353 /* This is not an arbitrary limit: the arguments which represent offsets
1354 into the pattern are two bytes long. So if 2^16 bytes turns out to
1355 be too small, many things would have to change. */
1356 #define MAX_BUF_SIZE (1L << 16)
1359 /* Extend the buffer by twice its current size via realloc and
1360 reset the pointers that pointed into the old block to point to the
1361 correct places in the new one. If extending the buffer results in it
1362 being larger than MAX_BUF_SIZE, then flag memory exhausted. */
1363 #define EXTEND_BUFFER() \
1365 unsigned char *old_buffer = bufp->buffer; \
1366 if (bufp->allocated == MAX_BUF_SIZE) \
1368 bufp->allocated <<= 1; \
1369 if (bufp->allocated > MAX_BUF_SIZE) \
1370 bufp->allocated = MAX_BUF_SIZE; \
1371 bufp->buffer = (unsigned char *) realloc (bufp->buffer, bufp->allocated);\
1372 if (bufp->buffer == NULL) \
1373 return REG_ESPACE; \
1374 /* If the buffer moved, move all the pointers into it. */ \
1375 if (old_buffer != bufp->buffer) \
1377 b = (b - old_buffer) + bufp->buffer; \
1378 begalt = (begalt - old_buffer) + bufp->buffer; \
1379 if (fixup_alt_jump) \
1380 fixup_alt_jump = (fixup_alt_jump - old_buffer) + bufp->buffer;\
1382 laststart = (laststart - old_buffer) + bufp->buffer; \
1383 if (pending_exact) \
1384 pending_exact = (pending_exact - old_buffer) + bufp->buffer; \
1389 /* Since we have one byte reserved for the register number argument to
1390 {start,stop}_memory, the maximum number of groups we can report
1391 things about is what fits in that byte. */
1392 #define MAX_REGNUM 255
1394 /* But patterns can have more than `MAX_REGNUM' registers. We just
1395 ignore the excess. */
1396 typedef unsigned regnum_t
;
1399 /* Macros for the compile stack. */
1401 /* Since offsets can go either forwards or backwards, this type needs to
1402 be able to hold values from -(MAX_BUF_SIZE - 1) to MAX_BUF_SIZE - 1. */
1403 typedef int pattern_offset_t
;
1407 pattern_offset_t begalt_offset
;
1408 pattern_offset_t fixup_alt_jump
;
1409 pattern_offset_t inner_group_offset
;
1410 pattern_offset_t laststart_offset
;
1412 } compile_stack_elt_t
;
1417 compile_stack_elt_t
*stack
;
1419 unsigned avail
; /* Offset of next open position. */
1420 } compile_stack_type
;
1423 #define INIT_COMPILE_STACK_SIZE 32
1425 #define COMPILE_STACK_EMPTY (compile_stack.avail == 0)
1426 #define COMPILE_STACK_FULL (compile_stack.avail == compile_stack.size)
1428 /* The next available element. */
1429 #define COMPILE_STACK_TOP (compile_stack.stack[compile_stack.avail])
1432 /* Set the bit for character C in a list. */
1433 #define SET_LIST_BIT(c) \
1434 (b[((unsigned char) (c)) / BYTEWIDTH] \
1435 |= 1 << (((unsigned char) c) % BYTEWIDTH))
1438 /* Get the next unsigned number in the uncompiled pattern. */
1439 #define GET_UNSIGNED_NUMBER(num) \
1443 while (ISDIGIT (c)) \
1447 num = num * 10 + c - '0'; \
1455 #define CHAR_CLASS_MAX_LENGTH 6 /* Namely, `xdigit'. */
1457 #define IS_CHAR_CLASS(string) \
1458 (STREQ (string, "alpha") || STREQ (string, "upper") \
1459 || STREQ (string, "lower") || STREQ (string, "digit") \
1460 || STREQ (string, "alnum") || STREQ (string, "xdigit") \
1461 || STREQ (string, "space") || STREQ (string, "print") \
1462 || STREQ (string, "punct") || STREQ (string, "graph") \
1463 || STREQ (string, "cntrl") || STREQ (string, "blank"))
1465 /* `regex_compile' compiles PATTERN (of length SIZE) according to SYNTAX.
1466 Returns one of error codes defined in `regex.h', or zero for success.
1468 Assumes the `allocated' (and perhaps `buffer') and `translate'
1469 fields are set in BUFP on entry.
1471 If it succeeds, results are put in BUFP (if it returns an error, the
1472 contents of BUFP are undefined):
1473 `buffer' is the compiled pattern;
1474 `syntax' is set to SYNTAX;
1475 `used' is set to the length of the compiled pattern;
1476 `fastmap_accurate' is zero;
1477 `re_nsub' is the number of subexpressions in PATTERN;
1478 `not_bol' and `not_eol' are zero;
1480 The `fastmap' and `newline_anchor' fields are neither
1481 examined nor set. */
1483 /* Return, freeing storage we allocated. */
1484 #define FREE_STACK_RETURN(value) \
1485 return (free (compile_stack.stack), value)
1487 static reg_errcode_t
1488 regex_compile (pattern
, size
, syntax
, bufp
)
1489 const char *pattern
;
1491 reg_syntax_t syntax
;
1492 struct re_pattern_buffer
*bufp
;
1494 /* We fetch characters from PATTERN here. Even though PATTERN is
1495 `char *' (i.e., signed), we declare these variables as unsigned, so
1496 they can be reliably used as array indices. */
1497 register unsigned char c
, c1
;
1499 /* A random temporary spot in PATTERN. */
1502 /* Points to the end of the buffer, where we should append. */
1503 register unsigned char *b
;
1505 /* Keeps track of unclosed groups. */
1506 compile_stack_type compile_stack
;
1508 /* Points to the current (ending) position in the pattern. */
1509 const char *p
= pattern
;
1510 const char *pend
= pattern
+ size
;
1512 /* How to translate the characters in the pattern. */
1513 char *translate
= bufp
->translate
;
1515 /* Address of the count-byte of the most recently inserted `exactn'
1516 command. This makes it possible to tell if a new exact-match
1517 character can be added to that command or if the character requires
1518 a new `exactn' command. */
1519 unsigned char *pending_exact
= 0;
1521 /* Address of start of the most recently finished expression.
1522 This tells, e.g., postfix * where to find the start of its
1523 operand. Reset at the beginning of groups and alternatives. */
1524 unsigned char *laststart
= 0;
1526 /* Address of beginning of regexp, or inside of last group. */
1527 unsigned char *begalt
;
1529 /* Place in the uncompiled pattern (i.e., the {) to
1530 which to go back if the interval is invalid. */
1531 const char *beg_interval
;
1533 /* Address of the place where a forward jump should go to the end of
1534 the containing expression. Each alternative of an `or' -- except the
1535 last -- ends with a forward jump of this sort. */
1536 unsigned char *fixup_alt_jump
= 0;
1538 /* Counts open-groups as they are encountered. Remembered for the
1539 matching close-group on the compile stack, so the same register
1540 number is put in the stop_memory as the start_memory. */
1541 regnum_t regnum
= 0;
1544 DEBUG_PRINT1 ("\nCompiling pattern: ");
1547 unsigned debug_count
;
1549 for (debug_count
= 0; debug_count
< size
; debug_count
++)
1550 printchar (pattern
[debug_count
]);
1555 /* Initialize the compile stack. */
1556 compile_stack
.stack
= TALLOC (INIT_COMPILE_STACK_SIZE
, compile_stack_elt_t
);
1557 if (compile_stack
.stack
== NULL
)
1560 compile_stack
.size
= INIT_COMPILE_STACK_SIZE
;
1561 compile_stack
.avail
= 0;
1563 /* Initialize the pattern buffer. */
1564 bufp
->syntax
= syntax
;
1565 bufp
->fastmap_accurate
= 0;
1566 bufp
->not_bol
= bufp
->not_eol
= 0;
1568 /* Set `used' to zero, so that if we return an error, the pattern
1569 printer (for debugging) will think there's no pattern. We reset it
1573 /* Always count groups, whether or not bufp->no_sub is set. */
1576 #if !defined (emacs) && !defined (SYNTAX_TABLE)
1577 /* Initialize the syntax table. */
1578 init_syntax_once ();
1581 if (bufp
->allocated
== 0)
1584 { /* If zero allocated, but buffer is non-null, try to realloc
1585 enough space. This loses if buffer's address is bogus, but
1586 that is the user's responsibility. */
1587 RETALLOC (bufp
->buffer
, INIT_BUF_SIZE
, unsigned char);
1590 { /* Caller did not allocate a buffer. Do it for them. */
1591 bufp
->buffer
= TALLOC (INIT_BUF_SIZE
, unsigned char);
1593 if (!bufp
->buffer
) FREE_STACK_RETURN (REG_ESPACE
);
1595 bufp
->allocated
= INIT_BUF_SIZE
;
1598 begalt
= b
= bufp
->buffer
;
1600 /* Loop through the uncompiled pattern until we're at the end. */
1609 if ( /* If at start of pattern, it's an operator. */
1611 /* If context independent, it's an operator. */
1612 || syntax
& RE_CONTEXT_INDEP_ANCHORS
1613 /* Otherwise, depends on what's come before. */
1614 || at_begline_loc_p (pattern
, p
, syntax
))
1624 if ( /* If at end of pattern, it's an operator. */
1626 /* If context independent, it's an operator. */
1627 || syntax
& RE_CONTEXT_INDEP_ANCHORS
1628 /* Otherwise, depends on what's next. */
1629 || at_endline_loc_p (p
, pend
, syntax
))
1639 if ((syntax
& RE_BK_PLUS_QM
)
1640 || (syntax
& RE_LIMITED_OPS
))
1644 /* If there is no previous pattern... */
1647 if (syntax
& RE_CONTEXT_INVALID_OPS
)
1648 FREE_STACK_RETURN (REG_BADRPT
);
1649 else if (!(syntax
& RE_CONTEXT_INDEP_OPS
))
1654 /* Are we optimizing this jump? */
1655 boolean keep_string_p
= false;
1657 /* 1 means zero (many) matches is allowed. */
1658 char zero_times_ok
= 0, many_times_ok
= 0;
1660 /* If there is a sequence of repetition chars, collapse it
1661 down to just one (the right one). We can't combine
1662 interval operators with these because of, e.g., `a{2}*',
1663 which should only match an even number of `a's. */
1667 zero_times_ok
|= c
!= '+';
1668 many_times_ok
|= c
!= '?';
1676 || (!(syntax
& RE_BK_PLUS_QM
) && (c
== '+' || c
== '?')))
1679 else if (syntax
& RE_BK_PLUS_QM
&& c
== '\\')
1681 if (p
== pend
) FREE_STACK_RETURN (REG_EESCAPE
);
1684 if (!(c1
== '+' || c1
== '?'))
1699 /* If we get here, we found another repeat character. */
1702 /* Star, etc. applied to an empty pattern is equivalent
1703 to an empty pattern. */
1707 /* Now we know whether or not zero matches is allowed
1708 and also whether or not two or more matches is allowed. */
1710 { /* More than one repetition is allowed, so put in at the
1711 end a backward relative jump from `b' to before the next
1712 jump we're going to put in below (which jumps from
1713 laststart to after this jump).
1715 But if we are at the `*' in the exact sequence `.*\n',
1716 insert an unconditional jump backwards to the .,
1717 instead of the beginning of the loop. This way we only
1718 push a failure point once, instead of every time
1719 through the loop. */
1720 assert (p
- 1 > pattern
);
1722 /* Allocate the space for the jump. */
1723 GET_BUFFER_SPACE (3);
1725 /* We know we are not at the first character of the pattern,
1726 because laststart was nonzero. And we've already
1727 incremented `p', by the way, to be the character after
1728 the `*'. Do we have to do something analogous here
1729 for null bytes, because of RE_DOT_NOT_NULL? */
1730 if (TRANSLATE (*(p
- 2)) == TRANSLATE ('.')
1732 && p
< pend
&& TRANSLATE (*p
) == TRANSLATE ('\n')
1733 && !(syntax
& RE_DOT_NEWLINE
))
1734 { /* We have .*\n. */
1735 STORE_JUMP (jump
, b
, laststart
);
1736 keep_string_p
= true;
1739 /* Anything else. */
1740 STORE_JUMP (maybe_pop_jump
, b
, laststart
- 3);
1742 /* We've added more stuff to the buffer. */
1746 /* On failure, jump from laststart to b + 3, which will be the
1747 end of the buffer after this jump is inserted. */
1748 GET_BUFFER_SPACE (3);
1749 INSERT_JUMP (keep_string_p
? on_failure_keep_string_jump
1757 /* At least one repetition is required, so insert a
1758 `dummy_failure_jump' before the initial
1759 `on_failure_jump' instruction of the loop. This
1760 effects a skip over that instruction the first time
1761 we hit that loop. */
1762 GET_BUFFER_SPACE (3);
1763 INSERT_JUMP (dummy_failure_jump
, laststart
, laststart
+ 6);
1778 boolean had_char_class
= false;
1780 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
1782 /* Ensure that we have enough space to push a charset: the
1783 opcode, the length count, and the bitset; 34 bytes in all. */
1784 GET_BUFFER_SPACE (34);
1788 /* We test `*p == '^' twice, instead of using an if
1789 statement, so we only need one BUF_PUSH. */
1790 BUF_PUSH (*p
== '^' ? charset_not
: charset
);
1794 /* Remember the first position in the bracket expression. */
1797 /* Push the number of bytes in the bitmap. */
1798 BUF_PUSH ((1 << BYTEWIDTH
) / BYTEWIDTH
);
1800 /* Clear the whole map. */
1801 bzero (b
, (1 << BYTEWIDTH
) / BYTEWIDTH
);
1803 /* charset_not matches newline according to a syntax bit. */
1804 if ((re_opcode_t
) b
[-2] == charset_not
1805 && (syntax
& RE_HAT_LISTS_NOT_NEWLINE
))
1806 SET_LIST_BIT ('\n');
1808 /* Read in characters and ranges, setting map bits. */
1811 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
1815 /* \ might escape characters inside [...] and [^...]. */
1816 if ((syntax
& RE_BACKSLASH_ESCAPE_IN_LISTS
) && c
== '\\')
1818 if (p
== pend
) FREE_STACK_RETURN (REG_EESCAPE
);
1825 /* Could be the end of the bracket expression. If it's
1826 not (i.e., when the bracket expression is `[]' so
1827 far), the ']' character bit gets set way below. */
1828 if (c
== ']' && p
!= p1
+ 1)
1831 /* Look ahead to see if it's a range when the last thing
1832 was a character class. */
1833 if (had_char_class
&& c
== '-' && *p
!= ']')
1834 FREE_STACK_RETURN (REG_ERANGE
);
1836 /* Look ahead to see if it's a range when the last thing
1837 was a character: if this is a hyphen not at the
1838 beginning or the end of a list, then it's the range
1841 && !(p
- 2 >= pattern
&& p
[-2] == '[')
1842 && !(p
- 3 >= pattern
&& p
[-3] == '[' && p
[-2] == '^')
1846 = compile_range (&p
, pend
, translate
, syntax
, b
);
1847 if (ret
!= REG_NOERROR
) FREE_STACK_RETURN (ret
);
1850 else if (p
[0] == '-' && p
[1] != ']')
1851 { /* This handles ranges made up of characters only. */
1854 /* Move past the `-'. */
1857 ret
= compile_range (&p
, pend
, translate
, syntax
, b
);
1858 if (ret
!= REG_NOERROR
) FREE_STACK_RETURN (ret
);
1861 /* See if we're at the beginning of a possible character
1864 else if (syntax
& RE_CHAR_CLASSES
&& c
== '[' && *p
== ':')
1865 { /* Leave room for the null. */
1866 char str
[CHAR_CLASS_MAX_LENGTH
+ 1];
1871 /* If pattern is `[[:'. */
1872 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
1877 if (c
== ':' || c
== ']' || p
== pend
1878 || c1
== CHAR_CLASS_MAX_LENGTH
)
1884 /* If isn't a word bracketed by `[:' and:`]':
1885 undo the ending character, the letters, and leave
1886 the leading `:' and `[' (but set bits for them). */
1887 if (c
== ':' && *p
== ']')
1890 boolean is_alnum
= STREQ (str
, "alnum");
1891 boolean is_alpha
= STREQ (str
, "alpha");
1892 boolean is_blank
= STREQ (str
, "blank");
1893 boolean is_cntrl
= STREQ (str
, "cntrl");
1894 boolean is_digit
= STREQ (str
, "digit");
1895 boolean is_graph
= STREQ (str
, "graph");
1896 boolean is_lower
= STREQ (str
, "lower");
1897 boolean is_print
= STREQ (str
, "print");
1898 boolean is_punct
= STREQ (str
, "punct");
1899 boolean is_space
= STREQ (str
, "space");
1900 boolean is_upper
= STREQ (str
, "upper");
1901 boolean is_xdigit
= STREQ (str
, "xdigit");
1903 if (!IS_CHAR_CLASS (str
))
1904 FREE_STACK_RETURN (REG_ECTYPE
);
1906 /* Throw away the ] at the end of the character
1910 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
1912 for (ch
= 0; ch
< 1 << BYTEWIDTH
; ch
++)
1914 /* This was split into 3 if's to
1915 avoid an arbitrary limit in some compiler. */
1916 if ( (is_alnum
&& ISALNUM (ch
))
1917 || (is_alpha
&& ISALPHA (ch
))
1918 || (is_blank
&& ISBLANK (ch
))
1919 || (is_cntrl
&& ISCNTRL (ch
)))
1921 if ( (is_digit
&& ISDIGIT (ch
))
1922 || (is_graph
&& ISGRAPH (ch
))
1923 || (is_lower
&& ISLOWER (ch
))
1924 || (is_print
&& ISPRINT (ch
)))
1926 if ( (is_punct
&& ISPUNCT (ch
))
1927 || (is_space
&& ISSPACE (ch
))
1928 || (is_upper
&& ISUPPER (ch
))
1929 || (is_xdigit
&& ISXDIGIT (ch
)))
1932 had_char_class
= true;
1941 had_char_class
= false;
1946 had_char_class
= false;
1951 /* Discard any (non)matching list bytes that are all 0 at the
1952 end of the map. Decrease the map-length byte too. */
1953 while ((int) b
[-1] > 0 && b
[b
[-1] - 1] == 0)
1961 if (syntax
& RE_NO_BK_PARENS
)
1968 if (syntax
& RE_NO_BK_PARENS
)
1975 if (syntax
& RE_NEWLINE_ALT
)
1982 if (syntax
& RE_NO_BK_VBAR
)
1989 if (syntax
& RE_INTERVALS
&& syntax
& RE_NO_BK_BRACES
)
1990 goto handle_interval
;
1996 if (p
== pend
) FREE_STACK_RETURN (REG_EESCAPE
);
1998 /* Do not translate the character after the \, so that we can
1999 distinguish, e.g., \B from \b, even if we normally would
2000 translate, e.g., B to b. */
2006 if (syntax
& RE_NO_BK_PARENS
)
2007 goto normal_backslash
;
2013 if (COMPILE_STACK_FULL
)
2015 RETALLOC (compile_stack
.stack
, compile_stack
.size
<< 1,
2016 compile_stack_elt_t
);
2017 if (compile_stack
.stack
== NULL
) return REG_ESPACE
;
2019 compile_stack
.size
<<= 1;
2022 /* These are the values to restore when we hit end of this
2023 group. They are all relative offsets, so that if the
2024 whole pattern moves because of realloc, they will still
2026 COMPILE_STACK_TOP
.begalt_offset
= begalt
- bufp
->buffer
;
2027 COMPILE_STACK_TOP
.fixup_alt_jump
2028 = fixup_alt_jump
? fixup_alt_jump
- bufp
->buffer
+ 1 : 0;
2029 COMPILE_STACK_TOP
.laststart_offset
= b
- bufp
->buffer
;
2030 COMPILE_STACK_TOP
.regnum
= regnum
;
2032 /* We will eventually replace the 0 with the number of
2033 groups inner to this one. But do not push a
2034 start_memory for groups beyond the last one we can
2035 represent in the compiled pattern. */
2036 if (regnum
<= MAX_REGNUM
)
2038 COMPILE_STACK_TOP
.inner_group_offset
= b
- bufp
->buffer
+ 2;
2039 BUF_PUSH_3 (start_memory
, regnum
, 0);
2042 compile_stack
.avail
++;
2047 /* If we've reached MAX_REGNUM groups, then this open
2048 won't actually generate any code, so we'll have to
2049 clear pending_exact explicitly. */
2055 if (syntax
& RE_NO_BK_PARENS
) goto normal_backslash
;
2057 if (COMPILE_STACK_EMPTY
)
2058 if (syntax
& RE_UNMATCHED_RIGHT_PAREN_ORD
)
2059 goto normal_backslash
;
2061 FREE_STACK_RETURN (REG_ERPAREN
);
2065 { /* Push a dummy failure point at the end of the
2066 alternative for a possible future
2067 `pop_failure_jump' to pop. See comments at
2068 `push_dummy_failure' in `re_match_2'. */
2069 BUF_PUSH (push_dummy_failure
);
2071 /* We allocated space for this jump when we assigned
2072 to `fixup_alt_jump', in the `handle_alt' case below. */
2073 STORE_JUMP (jump_past_alt
, fixup_alt_jump
, b
- 1);
2076 /* See similar code for backslashed left paren above. */
2077 if (COMPILE_STACK_EMPTY
)
2078 if (syntax
& RE_UNMATCHED_RIGHT_PAREN_ORD
)
2081 FREE_STACK_RETURN (REG_ERPAREN
);
2083 /* Since we just checked for an empty stack above, this
2084 ``can't happen''. */
2085 assert (compile_stack
.avail
!= 0);
2087 /* We don't just want to restore into `regnum', because
2088 later groups should continue to be numbered higher,
2089 as in `(ab)c(de)' -- the second group is #2. */
2090 regnum_t this_group_regnum
;
2092 compile_stack
.avail
--;
2093 begalt
= bufp
->buffer
+ COMPILE_STACK_TOP
.begalt_offset
;
2095 = COMPILE_STACK_TOP
.fixup_alt_jump
2096 ? bufp
->buffer
+ COMPILE_STACK_TOP
.fixup_alt_jump
- 1
2098 laststart
= bufp
->buffer
+ COMPILE_STACK_TOP
.laststart_offset
;
2099 this_group_regnum
= COMPILE_STACK_TOP
.regnum
;
2100 /* If we've reached MAX_REGNUM groups, then this open
2101 won't actually generate any code, so we'll have to
2102 clear pending_exact explicitly. */
2105 /* We're at the end of the group, so now we know how many
2106 groups were inside this one. */
2107 if (this_group_regnum
<= MAX_REGNUM
)
2109 unsigned char *inner_group_loc
2110 = bufp
->buffer
+ COMPILE_STACK_TOP
.inner_group_offset
;
2112 *inner_group_loc
= regnum
- this_group_regnum
;
2113 BUF_PUSH_3 (stop_memory
, this_group_regnum
,
2114 regnum
- this_group_regnum
);
2120 case '|': /* `\|'. */
2121 if (syntax
& RE_LIMITED_OPS
|| syntax
& RE_NO_BK_VBAR
)
2122 goto normal_backslash
;
2124 if (syntax
& RE_LIMITED_OPS
)
2127 /* Insert before the previous alternative a jump which
2128 jumps to this alternative if the former fails. */
2129 GET_BUFFER_SPACE (3);
2130 INSERT_JUMP (on_failure_jump
, begalt
, b
+ 6);
2134 /* The alternative before this one has a jump after it
2135 which gets executed if it gets matched. Adjust that
2136 jump so it will jump to this alternative's analogous
2137 jump (put in below, which in turn will jump to the next
2138 (if any) alternative's such jump, etc.). The last such
2139 jump jumps to the correct final destination. A picture:
2145 If we are at `b', then fixup_alt_jump right now points to a
2146 three-byte space after `a'. We'll put in the jump, set
2147 fixup_alt_jump to right after `b', and leave behind three
2148 bytes which we'll fill in when we get to after `c'. */
2151 STORE_JUMP (jump_past_alt
, fixup_alt_jump
, b
);
2153 /* Mark and leave space for a jump after this alternative,
2154 to be filled in later either by next alternative or
2155 when know we're at the end of a series of alternatives. */
2157 GET_BUFFER_SPACE (3);
2166 /* If \{ is a literal. */
2167 if (!(syntax
& RE_INTERVALS
)
2168 /* If we're at `\{' and it's not the open-interval
2170 || ((syntax
& RE_INTERVALS
) && (syntax
& RE_NO_BK_BRACES
))
2171 || (p
- 2 == pattern
&& p
== pend
))
2172 goto normal_backslash
;
2176 /* If got here, then the syntax allows intervals. */
2178 /* At least (most) this many matches must be made. */
2179 int lower_bound
= -1, upper_bound
= -1;
2181 beg_interval
= p
- 1;
2185 if (syntax
& RE_NO_BK_BRACES
)
2186 goto unfetch_interval
;
2188 FREE_STACK_RETURN (REG_EBRACE
);
2191 GET_UNSIGNED_NUMBER (lower_bound
);
2195 GET_UNSIGNED_NUMBER (upper_bound
);
2196 if (upper_bound
< 0) upper_bound
= RE_DUP_MAX
;
2199 /* Interval such as `{1}' => match exactly once. */
2200 upper_bound
= lower_bound
;
2202 if (lower_bound
< 0 || upper_bound
> RE_DUP_MAX
2203 || lower_bound
> upper_bound
)
2205 if (syntax
& RE_NO_BK_BRACES
)
2206 goto unfetch_interval
;
2208 FREE_STACK_RETURN (REG_BADBR
);
2211 if (!(syntax
& RE_NO_BK_BRACES
))
2213 if (c
!= '\\') FREE_STACK_RETURN (REG_EBRACE
);
2220 if (syntax
& RE_NO_BK_BRACES
)
2221 goto unfetch_interval
;
2223 FREE_STACK_RETURN (REG_BADBR
);
2226 /* We just parsed a valid interval. */
2228 /* If it's invalid to have no preceding re. */
2231 if (syntax
& RE_CONTEXT_INVALID_OPS
)
2232 FREE_STACK_RETURN (REG_BADRPT
);
2233 else if (syntax
& RE_CONTEXT_INDEP_OPS
)
2236 goto unfetch_interval
;
2239 /* If the upper bound is zero, don't want to succeed at
2240 all; jump from `laststart' to `b + 3', which will be
2241 the end of the buffer after we insert the jump. */
2242 if (upper_bound
== 0)
2244 GET_BUFFER_SPACE (3);
2245 INSERT_JUMP (jump
, laststart
, b
+ 3);
2249 /* Otherwise, we have a nontrivial interval. When
2250 we're all done, the pattern will look like:
2251 set_number_at <jump count> <upper bound>
2252 set_number_at <succeed_n count> <lower bound>
2253 succeed_n <after jump addr> <succeed_n count>
2255 jump_n <succeed_n addr> <jump count>
2256 (The upper bound and `jump_n' are omitted if
2257 `upper_bound' is 1, though.) */
2259 { /* If the upper bound is > 1, we need to insert
2260 more at the end of the loop. */
2261 unsigned nbytes
= 10 + (upper_bound
> 1) * 10;
2263 GET_BUFFER_SPACE (nbytes
);
2265 /* Initialize lower bound of the `succeed_n', even
2266 though it will be set during matching by its
2267 attendant `set_number_at' (inserted next),
2268 because `re_compile_fastmap' needs to know.
2269 Jump to the `jump_n' we might insert below. */
2270 INSERT_JUMP2 (succeed_n
, laststart
,
2271 b
+ 5 + (upper_bound
> 1) * 5,
2275 /* Code to initialize the lower bound. Insert
2276 before the `succeed_n'. The `5' is the last two
2277 bytes of this `set_number_at', plus 3 bytes of
2278 the following `succeed_n'. */
2279 insert_op2 (set_number_at
, laststart
, 5, lower_bound
, b
);
2282 if (upper_bound
> 1)
2283 { /* More than one repetition is allowed, so
2284 append a backward jump to the `succeed_n'
2285 that starts this interval.
2287 When we've reached this during matching,
2288 we'll have matched the interval once, so
2289 jump back only `upper_bound - 1' times. */
2290 STORE_JUMP2 (jump_n
, b
, laststart
+ 5,
2294 /* The location we want to set is the second
2295 parameter of the `jump_n'; that is `b-2' as
2296 an absolute address. `laststart' will be
2297 the `set_number_at' we're about to insert;
2298 `laststart+3' the number to set, the source
2299 for the relative address. But we are
2300 inserting into the middle of the pattern --
2301 so everything is getting moved up by 5.
2302 Conclusion: (b - 2) - (laststart + 3) + 5,
2303 i.e., b - laststart.
2305 We insert this at the beginning of the loop
2306 so that if we fail during matching, we'll
2307 reinitialize the bounds. */
2308 insert_op2 (set_number_at
, laststart
, b
- laststart
,
2309 upper_bound
- 1, b
);
2314 beg_interval
= NULL
;
2319 /* If an invalid interval, match the characters as literals. */
2320 assert (beg_interval
);
2322 beg_interval
= NULL
;
2324 /* normal_char and normal_backslash need `c'. */
2327 if (!(syntax
& RE_NO_BK_BRACES
))
2329 if (p
> pattern
&& p
[-1] == '\\')
2330 goto normal_backslash
;
2335 /* There is no way to specify the before_dot and after_dot
2336 operators. rms says this is ok. --karl */
2344 BUF_PUSH_2 (syntaxspec
, syntax_spec_code
[c
]);
2350 BUF_PUSH_2 (notsyntaxspec
, syntax_spec_code
[c
]);
2357 BUF_PUSH (wordchar
);
2363 BUF_PUSH (notwordchar
);
2376 BUF_PUSH (wordbound
);
2380 BUF_PUSH (notwordbound
);
2391 case '1': case '2': case '3': case '4': case '5':
2392 case '6': case '7': case '8': case '9':
2393 if (syntax
& RE_NO_BK_REFS
)
2399 FREE_STACK_RETURN (REG_ESUBREG
);
2401 /* Can't back reference to a subexpression if inside of it. */
2402 if (group_in_compile_stack (compile_stack
, c1
))
2406 BUF_PUSH_2 (duplicate
, c1
);
2412 if (syntax
& RE_BK_PLUS_QM
)
2415 goto normal_backslash
;
2419 /* You might think it would be useful for \ to mean
2420 not to translate; but if we don't translate it
2421 it will never match anything. */
2429 /* Expects the character in `c'. */
2431 /* If no exactn currently being built. */
2434 /* If last exactn not at current position. */
2435 || pending_exact
+ *pending_exact
+ 1 != b
2437 /* We have only one byte following the exactn for the count. */
2438 || *pending_exact
== (1 << BYTEWIDTH
) - 1
2440 /* If followed by a repetition operator. */
2441 || *p
== '*' || *p
== '^'
2442 || ((syntax
& RE_BK_PLUS_QM
)
2443 ? *p
== '\\' && (p
[1] == '+' || p
[1] == '?')
2444 : (*p
== '+' || *p
== '?'))
2445 || ((syntax
& RE_INTERVALS
)
2446 && ((syntax
& RE_NO_BK_BRACES
)
2448 : (p
[0] == '\\' && p
[1] == '{'))))
2450 /* Start building a new exactn. */
2454 BUF_PUSH_2 (exactn
, 0);
2455 pending_exact
= b
- 1;
2462 } /* while p != pend */
2465 /* Through the pattern now. */
2468 STORE_JUMP (jump_past_alt
, fixup_alt_jump
, b
);
2470 if (!COMPILE_STACK_EMPTY
)
2471 FREE_STACK_RETURN (REG_EPAREN
);
2473 free (compile_stack
.stack
);
2475 /* We have succeeded; set the length of the buffer. */
2476 bufp
->used
= b
- bufp
->buffer
;
2481 DEBUG_PRINT1 ("\nCompiled pattern: \n");
2482 print_compiled_pattern (bufp
);
2486 #ifndef MATCH_MAY_ALLOCATE
2487 /* Initialize the failure stack to the largest possible stack. This
2488 isn't necessary unless we're trying to avoid calling alloca in
2489 the search and match routines. */
2491 int num_regs
= bufp
->re_nsub
+ 1;
2493 /* Since DOUBLE_FAIL_STACK refuses to double only if the current size
2494 is strictly greater than re_max_failures, the largest possible stack
2495 is 2 * re_max_failures failure points. */
2496 if (fail_stack
.size
< (2 * re_max_failures
* MAX_FAILURE_ITEMS
))
2498 fail_stack
.size
= (2 * re_max_failures
* MAX_FAILURE_ITEMS
);
2501 if (! fail_stack
.stack
)
2503 = (fail_stack_elt_t
*) xmalloc (fail_stack
.size
2504 * sizeof (fail_stack_elt_t
));
2507 = (fail_stack_elt_t
*) xrealloc (fail_stack
.stack
,
2509 * sizeof (fail_stack_elt_t
)));
2510 #else /* not emacs */
2511 if (! fail_stack
.stack
)
2513 = (fail_stack_elt_t
*) malloc (fail_stack
.size
2514 * sizeof (fail_stack_elt_t
));
2517 = (fail_stack_elt_t
*) realloc (fail_stack
.stack
,
2519 * sizeof (fail_stack_elt_t
)));
2520 #endif /* not emacs */
2523 /* Initialize some other variables the matcher uses. */
2524 RETALLOC_IF (regstart
, num_regs
, const char *);
2525 RETALLOC_IF (regend
, num_regs
, const char *);
2526 RETALLOC_IF (old_regstart
, num_regs
, const char *);
2527 RETALLOC_IF (old_regend
, num_regs
, const char *);
2528 RETALLOC_IF (best_regstart
, num_regs
, const char *);
2529 RETALLOC_IF (best_regend
, num_regs
, const char *);
2530 RETALLOC_IF (reg_info
, num_regs
, register_info_type
);
2531 RETALLOC_IF (reg_dummy
, num_regs
, const char *);
2532 RETALLOC_IF (reg_info_dummy
, num_regs
, register_info_type
);
2537 } /* regex_compile */
2539 /* Subroutines for `regex_compile'. */
2541 /* Store OP at LOC followed by two-byte integer parameter ARG. */
2544 store_op1 (op
, loc
, arg
)
2549 *loc
= (unsigned char) op
;
2550 STORE_NUMBER (loc
+ 1, arg
);
2554 /* Like `store_op1', but for two two-byte parameters ARG1 and ARG2. */
2557 store_op2 (op
, loc
, arg1
, arg2
)
2562 *loc
= (unsigned char) op
;
2563 STORE_NUMBER (loc
+ 1, arg1
);
2564 STORE_NUMBER (loc
+ 3, arg2
);
2568 /* Copy the bytes from LOC to END to open up three bytes of space at LOC
2569 for OP followed by two-byte integer parameter ARG. */
2572 insert_op1 (op
, loc
, arg
, end
)
2578 register unsigned char *pfrom
= end
;
2579 register unsigned char *pto
= end
+ 3;
2581 while (pfrom
!= loc
)
2584 store_op1 (op
, loc
, arg
);
2588 /* Like `insert_op1', but for two two-byte parameters ARG1 and ARG2. */
2591 insert_op2 (op
, loc
, arg1
, arg2
, end
)
2597 register unsigned char *pfrom
= end
;
2598 register unsigned char *pto
= end
+ 5;
2600 while (pfrom
!= loc
)
2603 store_op2 (op
, loc
, arg1
, arg2
);
2607 /* P points to just after a ^ in PATTERN. Return true if that ^ comes
2608 after an alternative or a begin-subexpression. We assume there is at
2609 least one character before the ^. */
2612 at_begline_loc_p (pattern
, p
, syntax
)
2613 const char *pattern
, *p
;
2614 reg_syntax_t syntax
;
2616 const char *prev
= p
- 2;
2617 boolean prev_prev_backslash
= prev
> pattern
&& prev
[-1] == '\\';
2620 /* After a subexpression? */
2621 (*prev
== '(' && (syntax
& RE_NO_BK_PARENS
|| prev_prev_backslash
))
2622 /* After an alternative? */
2623 || (*prev
== '|' && (syntax
& RE_NO_BK_VBAR
|| prev_prev_backslash
));
2627 /* The dual of at_begline_loc_p. This one is for $. We assume there is
2628 at least one character after the $, i.e., `P < PEND'. */
2631 at_endline_loc_p (p
, pend
, syntax
)
2632 const char *p
, *pend
;
2635 const char *next
= p
;
2636 boolean next_backslash
= *next
== '\\';
2637 const char *next_next
= p
+ 1 < pend
? p
+ 1 : NULL
;
2640 /* Before a subexpression? */
2641 (syntax
& RE_NO_BK_PARENS
? *next
== ')'
2642 : next_backslash
&& next_next
&& *next_next
== ')')
2643 /* Before an alternative? */
2644 || (syntax
& RE_NO_BK_VBAR
? *next
== '|'
2645 : next_backslash
&& next_next
&& *next_next
== '|');
2649 /* Returns true if REGNUM is in one of COMPILE_STACK's elements and
2650 false if it's not. */
2653 group_in_compile_stack (compile_stack
, regnum
)
2654 compile_stack_type compile_stack
;
2659 for (this_element
= compile_stack
.avail
- 1;
2662 if (compile_stack
.stack
[this_element
].regnum
== regnum
)
2669 /* Read the ending character of a range (in a bracket expression) from the
2670 uncompiled pattern *P_PTR (which ends at PEND). We assume the
2671 starting character is in `P[-2]'. (`P[-1]' is the character `-'.)
2672 Then we set the translation of all bits between the starting and
2673 ending characters (inclusive) in the compiled pattern B.
2675 Return an error code.
2677 We use these short variable names so we can use the same macros as
2678 `regex_compile' itself. */
2680 static reg_errcode_t
2681 compile_range (p_ptr
, pend
, translate
, syntax
, b
)
2682 const char **p_ptr
, *pend
;
2684 reg_syntax_t syntax
;
2689 const char *p
= *p_ptr
;
2690 int range_start
, range_end
;
2695 /* Even though the pattern is a signed `char *', we need to fetch
2696 with unsigned char *'s; if the high bit of the pattern character
2697 is set, the range endpoints will be negative if we fetch using a
2700 We also want to fetch the endpoints without translating them; the
2701 appropriate translation is done in the bit-setting loop below. */
2702 /* The SVR4 compiler on the 3B2 had trouble with unsigned const char *. */
2703 range_start
= ((const unsigned char *) p
)[-2];
2704 range_end
= ((const unsigned char *) p
)[0];
2706 /* Have to increment the pointer into the pattern string, so the
2707 caller isn't still at the ending character. */
2710 /* If the start is after the end, the range is empty. */
2711 if (range_start
> range_end
)
2712 return syntax
& RE_NO_EMPTY_RANGES
? REG_ERANGE
: REG_NOERROR
;
2714 /* Here we see why `this_char' has to be larger than an `unsigned
2715 char' -- the range is inclusive, so if `range_end' == 0xff
2716 (assuming 8-bit characters), we would otherwise go into an infinite
2717 loop, since all characters <= 0xff. */
2718 for (this_char
= range_start
; this_char
<= range_end
; this_char
++)
2720 SET_LIST_BIT (TRANSLATE (this_char
));
2726 /* re_compile_fastmap computes a ``fastmap'' for the compiled pattern in
2727 BUFP. A fastmap records which of the (1 << BYTEWIDTH) possible
2728 characters can start a string that matches the pattern. This fastmap
2729 is used by re_search to skip quickly over impossible starting points.
2731 The caller must supply the address of a (1 << BYTEWIDTH)-byte data
2732 area as BUFP->fastmap.
2734 We set the `fastmap', `fastmap_accurate', and `can_be_null' fields in
2737 Returns 0 if we succeed, -2 if an internal error. */
2740 re_compile_fastmap (bufp
)
2741 struct re_pattern_buffer
*bufp
;
2744 #ifdef MATCH_MAY_ALLOCATE
2745 fail_stack_type fail_stack
;
2747 #ifndef REGEX_MALLOC
2750 /* We don't push any register information onto the failure stack. */
2751 unsigned num_regs
= 0;
2753 register char *fastmap
= bufp
->fastmap
;
2754 unsigned char *pattern
= bufp
->buffer
;
2755 unsigned long size
= bufp
->used
;
2756 unsigned char *p
= pattern
;
2757 register unsigned char *pend
= pattern
+ size
;
2759 /* Assume that each path through the pattern can be null until
2760 proven otherwise. We set this false at the bottom of switch
2761 statement, to which we get only if a particular path doesn't
2762 match the empty string. */
2763 boolean path_can_be_null
= true;
2765 /* We aren't doing a `succeed_n' to begin with. */
2766 boolean succeed_n_p
= false;
2768 assert (fastmap
!= NULL
&& p
!= NULL
);
2771 bzero (fastmap
, 1 << BYTEWIDTH
); /* Assume nothing's valid. */
2772 bufp
->fastmap_accurate
= 1; /* It will be when we're done. */
2773 bufp
->can_be_null
= 0;
2775 while (p
!= pend
|| !FAIL_STACK_EMPTY ())
2779 bufp
->can_be_null
|= path_can_be_null
;
2781 /* Reset for next path. */
2782 path_can_be_null
= true;
2784 p
= fail_stack
.stack
[--fail_stack
.avail
];
2787 /* We should never be about to go beyond the end of the pattern. */
2790 #ifdef SWITCH_ENUM_BUG
2791 switch ((int) ((re_opcode_t
) *p
++))
2793 switch ((re_opcode_t
) *p
++)
2797 /* I guess the idea here is to simply not bother with a fastmap
2798 if a backreference is used, since it's too hard to figure out
2799 the fastmap for the corresponding group. Setting
2800 `can_be_null' stops `re_search_2' from using the fastmap, so
2801 that is all we do. */
2803 bufp
->can_be_null
= 1;
2807 /* Following are the cases which match a character. These end
2816 for (j
= *p
++ * BYTEWIDTH
- 1; j
>= 0; j
--)
2817 if (p
[j
/ BYTEWIDTH
] & (1 << (j
% BYTEWIDTH
)))
2823 /* Chars beyond end of map must be allowed. */
2824 for (j
= *p
* BYTEWIDTH
; j
< (1 << BYTEWIDTH
); j
++)
2827 for (j
= *p
++ * BYTEWIDTH
- 1; j
>= 0; j
--)
2828 if (!(p
[j
/ BYTEWIDTH
] & (1 << (j
% BYTEWIDTH
))))
2834 for (j
= 0; j
< (1 << BYTEWIDTH
); j
++)
2835 if (SYNTAX (j
) == Sword
)
2841 for (j
= 0; j
< (1 << BYTEWIDTH
); j
++)
2842 if (SYNTAX (j
) != Sword
)
2849 int fastmap_newline
= fastmap
['\n'];
2851 /* `.' matches anything ... */
2852 for (j
= 0; j
< (1 << BYTEWIDTH
); j
++)
2855 /* ... except perhaps newline. */
2856 if (!(bufp
->syntax
& RE_DOT_NEWLINE
))
2857 fastmap
['\n'] = fastmap_newline
;
2859 /* Return if we have already set `can_be_null'; if we have,
2860 then the fastmap is irrelevant. Something's wrong here. */
2861 else if (bufp
->can_be_null
)
2864 /* Otherwise, have to check alternative paths. */
2871 for (j
= 0; j
< (1 << BYTEWIDTH
); j
++)
2872 if (SYNTAX (j
) == (enum syntaxcode
) k
)
2879 for (j
= 0; j
< (1 << BYTEWIDTH
); j
++)
2880 if (SYNTAX (j
) != (enum syntaxcode
) k
)
2885 /* All cases after this match the empty string. These end with
2893 #endif /* not emacs */
2905 case push_dummy_failure
:
2910 case pop_failure_jump
:
2911 case maybe_pop_jump
:
2914 case dummy_failure_jump
:
2915 EXTRACT_NUMBER_AND_INCR (j
, p
);
2920 /* Jump backward implies we just went through the body of a
2921 loop and matched nothing. Opcode jumped to should be
2922 `on_failure_jump' or `succeed_n'. Just treat it like an
2923 ordinary jump. For a * loop, it has pushed its failure
2924 point already; if so, discard that as redundant. */
2925 if ((re_opcode_t
) *p
!= on_failure_jump
2926 && (re_opcode_t
) *p
!= succeed_n
)
2930 EXTRACT_NUMBER_AND_INCR (j
, p
);
2933 /* If what's on the stack is where we are now, pop it. */
2934 if (!FAIL_STACK_EMPTY ()
2935 && fail_stack
.stack
[fail_stack
.avail
- 1] == p
)
2941 case on_failure_jump
:
2942 case on_failure_keep_string_jump
:
2943 handle_on_failure_jump
:
2944 EXTRACT_NUMBER_AND_INCR (j
, p
);
2946 /* For some patterns, e.g., `(a?)?', `p+j' here points to the
2947 end of the pattern. We don't want to push such a point,
2948 since when we restore it above, entering the switch will
2949 increment `p' past the end of the pattern. We don't need
2950 to push such a point since we obviously won't find any more
2951 fastmap entries beyond `pend'. Such a pattern can match
2952 the null string, though. */
2955 if (!PUSH_PATTERN_OP (p
+ j
, fail_stack
))
2959 bufp
->can_be_null
= 1;
2963 EXTRACT_NUMBER_AND_INCR (k
, p
); /* Skip the n. */
2964 succeed_n_p
= false;
2971 /* Get to the number of times to succeed. */
2974 /* Increment p past the n for when k != 0. */
2975 EXTRACT_NUMBER_AND_INCR (k
, p
);
2979 succeed_n_p
= true; /* Spaghetti code alert. */
2980 goto handle_on_failure_jump
;
2997 abort (); /* We have listed all the cases. */
3000 /* Getting here means we have found the possible starting
3001 characters for one path of the pattern -- and that the empty
3002 string does not match. We need not follow this path further.
3003 Instead, look at the next alternative (remembered on the
3004 stack), or quit if no more. The test at the top of the loop
3005 does these things. */
3006 path_can_be_null
= false;
3010 /* Set `can_be_null' for the last path (also the first path, if the
3011 pattern is empty). */
3012 bufp
->can_be_null
|= path_can_be_null
;
3014 } /* re_compile_fastmap */
3016 /* Set REGS to hold NUM_REGS registers, storing them in STARTS and
3017 ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use
3018 this memory for recording register information. STARTS and ENDS
3019 must be allocated using the malloc library routine, and must each
3020 be at least NUM_REGS * sizeof (regoff_t) bytes long.
3022 If NUM_REGS == 0, then subsequent matches should allocate their own
3025 Unless this function is called, the first search or match using
3026 PATTERN_BUFFER will allocate its own register data, without
3027 freeing the old data. */
3030 re_set_registers (bufp
, regs
, num_regs
, starts
, ends
)
3031 struct re_pattern_buffer
*bufp
;
3032 struct re_registers
*regs
;
3034 regoff_t
*starts
, *ends
;
3038 bufp
->regs_allocated
= REGS_REALLOCATE
;
3039 regs
->num_regs
= num_regs
;
3040 regs
->start
= starts
;
3045 bufp
->regs_allocated
= REGS_UNALLOCATED
;
3047 regs
->start
= regs
->end
= (regoff_t
*) 0;
3051 /* Searching routines. */
3053 /* Like re_search_2, below, but only one string is specified, and
3054 doesn't let you say where to stop matching. */
3057 re_search (bufp
, string
, size
, startpos
, range
, regs
)
3058 struct re_pattern_buffer
*bufp
;
3060 int size
, startpos
, range
;
3061 struct re_registers
*regs
;
3063 return re_search_2 (bufp
, NULL
, 0, string
, size
, startpos
, range
,
3068 /* Using the compiled pattern in BUFP->buffer, first tries to match the
3069 virtual concatenation of STRING1 and STRING2, starting first at index
3070 STARTPOS, then at STARTPOS + 1, and so on.
3072 STRING1 and STRING2 have length SIZE1 and SIZE2, respectively.
3074 RANGE is how far to scan while trying to match. RANGE = 0 means try
3075 only at STARTPOS; in general, the last start tried is STARTPOS +
3078 In REGS, return the indices of the virtual concatenation of STRING1
3079 and STRING2 that matched the entire BUFP->buffer and its contained
3082 Do not consider matching one past the index STOP in the virtual
3083 concatenation of STRING1 and STRING2.
3085 We return either the position in the strings at which the match was
3086 found, -1 if no match, or -2 if error (such as failure
3090 re_search_2 (bufp
, string1
, size1
, string2
, size2
, startpos
, range
, regs
, stop
)
3091 struct re_pattern_buffer
*bufp
;
3092 const char *string1
, *string2
;
3096 struct re_registers
*regs
;
3100 register char *fastmap
= bufp
->fastmap
;
3101 register char *translate
= bufp
->translate
;
3102 int total_size
= size1
+ size2
;
3103 int endpos
= startpos
+ range
;
3105 /* Check for out-of-range STARTPOS. */
3106 if (startpos
< 0 || startpos
> total_size
)
3109 /* Fix up RANGE if it might eventually take us outside
3110 the virtual concatenation of STRING1 and STRING2. */
3112 range
= -1 - startpos
;
3113 else if (endpos
> total_size
)
3114 range
= total_size
- startpos
;
3116 /* If the search isn't to be a backwards one, don't waste time in a
3117 search for a pattern that must be anchored. */
3118 if (bufp
->used
> 0 && (re_opcode_t
) bufp
->buffer
[0] == begbuf
&& range
> 0)
3126 /* Update the fastmap now if not correct already. */
3127 if (fastmap
&& !bufp
->fastmap_accurate
)
3128 if (re_compile_fastmap (bufp
) == -2)
3131 /* Loop through the string, looking for a place to start matching. */
3134 /* If a fastmap is supplied, skip quickly over characters that
3135 cannot be the start of a match. If the pattern can match the
3136 null string, however, we don't need to skip characters; we want
3137 the first null string. */
3138 if (fastmap
&& startpos
< total_size
&& !bufp
->can_be_null
)
3140 if (range
> 0) /* Searching forwards. */
3142 register const char *d
;
3143 register int lim
= 0;
3146 if (startpos
< size1
&& startpos
+ range
>= size1
)
3147 lim
= range
- (size1
- startpos
);
3149 d
= (startpos
>= size1
? string2
- size1
: string1
) + startpos
;
3151 /* Written out as an if-else to avoid testing `translate'
3155 && !fastmap
[(unsigned char)
3156 translate
[(unsigned char) *d
++]])
3159 while (range
> lim
&& !fastmap
[(unsigned char) *d
++])
3162 startpos
+= irange
- range
;
3164 else /* Searching backwards. */
3166 register char c
= (size1
== 0 || startpos
>= size1
3167 ? string2
[startpos
- size1
]
3168 : string1
[startpos
]);
3170 if (!fastmap
[(unsigned char) TRANSLATE (c
)])
3175 /* If can't match the null string, and that's all we have left, fail. */
3176 if (range
>= 0 && startpos
== total_size
&& fastmap
3177 && !bufp
->can_be_null
)
3180 val
= re_match_2_internal (bufp
, string1
, size1
, string2
, size2
,
3181 startpos
, regs
, stop
);
3182 #ifndef REGEX_MALLOC
3211 /* Declarations and macros for re_match_2. */
3213 static int bcmp_translate ();
3214 static boolean
alt_match_null_string_p (),
3215 common_op_match_null_string_p (),
3216 group_match_null_string_p ();
3218 /* This converts PTR, a pointer into one of the search strings `string1'
3219 and `string2' into an offset from the beginning of that string. */
3220 #define POINTER_TO_OFFSET(ptr) \
3221 (FIRST_STRING_P (ptr) \
3222 ? ((regoff_t) ((ptr) - string1)) \
3223 : ((regoff_t) ((ptr) - string2 + size1)))
3225 /* Macros for dealing with the split strings in re_match_2. */
3227 #define MATCHING_IN_FIRST_STRING (dend == end_match_1)
3229 /* Call before fetching a character with *d. This switches over to
3230 string2 if necessary. */
3231 #define PREFETCH() \
3234 /* End of string2 => fail. */ \
3235 if (dend == end_match_2) \
3237 /* End of string1 => advance to string2. */ \
3239 dend = end_match_2; \
3243 /* Test if at very beginning or at very end of the virtual concatenation
3244 of `string1' and `string2'. If only one string, it's `string2'. */
3245 #define AT_STRINGS_BEG(d) ((d) == (size1 ? string1 : string2) || !size2)
3246 #define AT_STRINGS_END(d) ((d) == end2)
3249 /* Test if D points to a character which is word-constituent. We have
3250 two special cases to check for: if past the end of string1, look at
3251 the first character in string2; and if before the beginning of
3252 string2, look at the last character in string1. */
3253 #define WORDCHAR_P(d) \
3254 (SYNTAX ((d) == end1 ? *string2 \
3255 : (d) == string2 - 1 ? *(end1 - 1) : *(d)) \
3258 /* Test if the character before D and the one at D differ with respect
3259 to being word-constituent. */
3260 #define AT_WORD_BOUNDARY(d) \
3261 (AT_STRINGS_BEG (d) || AT_STRINGS_END (d) \
3262 || WORDCHAR_P (d - 1) != WORDCHAR_P (d))
3265 /* Free everything we malloc. */
3266 #ifdef MATCH_MAY_ALLOCATE
3268 #define FREE_VAR(var) if (var) free (var); var = NULL
3269 #define FREE_VARIABLES() \
3271 FREE_VAR (fail_stack.stack); \
3272 FREE_VAR (regstart); \
3273 FREE_VAR (regend); \
3274 FREE_VAR (old_regstart); \
3275 FREE_VAR (old_regend); \
3276 FREE_VAR (best_regstart); \
3277 FREE_VAR (best_regend); \
3278 FREE_VAR (reg_info); \
3279 FREE_VAR (reg_dummy); \
3280 FREE_VAR (reg_info_dummy); \
3282 #else /* not REGEX_MALLOC */
3283 /* This used to do alloca (0), but now we do that in the caller. */
3284 #define FREE_VARIABLES() /* Nothing */
3285 #endif /* not REGEX_MALLOC */
3287 #define FREE_VARIABLES() /* Do nothing! */
3288 #endif /* not MATCH_MAY_ALLOCATE */
3290 /* These values must meet several constraints. They must not be valid
3291 register values; since we have a limit of 255 registers (because
3292 we use only one byte in the pattern for the register number), we can
3293 use numbers larger than 255. They must differ by 1, because of
3294 NUM_FAILURE_ITEMS above. And the value for the lowest register must
3295 be larger than the value for the highest register, so we do not try
3296 to actually save any registers when none are active. */
3297 #define NO_HIGHEST_ACTIVE_REG (1 << BYTEWIDTH)
3298 #define NO_LOWEST_ACTIVE_REG (NO_HIGHEST_ACTIVE_REG + 1)
3300 /* Matching routines. */
3302 #ifndef emacs /* Emacs never uses this. */
3303 /* re_match is like re_match_2 except it takes only a single string. */
3306 re_match (bufp
, string
, size
, pos
, regs
)
3307 struct re_pattern_buffer
*bufp
;
3310 struct re_registers
*regs
;
3312 int result
= re_match_2_internal (bufp
, NULL
, 0, string
, size
,
3317 #endif /* not emacs */
3320 /* re_match_2 matches the compiled pattern in BUFP against the
3321 the (virtual) concatenation of STRING1 and STRING2 (of length SIZE1
3322 and SIZE2, respectively). We start matching at POS, and stop
3325 If REGS is non-null and the `no_sub' field of BUFP is nonzero, we
3326 store offsets for the substring each group matched in REGS. See the
3327 documentation for exactly how many groups we fill.
3329 We return -1 if no match, -2 if an internal error (such as the
3330 failure stack overflowing). Otherwise, we return the length of the
3331 matched substring. */
3334 re_match_2 (bufp
, string1
, size1
, string2
, size2
, pos
, regs
, stop
)
3335 struct re_pattern_buffer
*bufp
;
3336 const char *string1
, *string2
;
3339 struct re_registers
*regs
;
3342 int result
= re_match_2_internal (bufp
, string1
, size1
, string2
, size2
,
3348 /* This is a separate function so that we can force an alloca cleanup
3351 re_match_2_internal (bufp
, string1
, size1
, string2
, size2
, pos
, regs
, stop
)
3352 struct re_pattern_buffer
*bufp
;
3353 const char *string1
, *string2
;
3356 struct re_registers
*regs
;
3359 /* General temporaries. */
3363 /* Just past the end of the corresponding string. */
3364 const char *end1
, *end2
;
3366 /* Pointers into string1 and string2, just past the last characters in
3367 each to consider matching. */
3368 const char *end_match_1
, *end_match_2
;
3370 /* Where we are in the data, and the end of the current string. */
3371 const char *d
, *dend
;
3373 /* Where we are in the pattern, and the end of the pattern. */
3374 unsigned char *p
= bufp
->buffer
;
3375 register unsigned char *pend
= p
+ bufp
->used
;
3377 /* Mark the opcode just after a start_memory, so we can test for an
3378 empty subpattern when we get to the stop_memory. */
3379 unsigned char *just_past_start_mem
= 0;
3381 /* We use this to map every character in the string. */
3382 char *translate
= bufp
->translate
;
3384 /* Failure point stack. Each place that can handle a failure further
3385 down the line pushes a failure point on this stack. It consists of
3386 restart, regend, and reg_info for all registers corresponding to
3387 the subexpressions we're currently inside, plus the number of such
3388 registers, and, finally, two char *'s. The first char * is where
3389 to resume scanning the pattern; the second one is where to resume
3390 scanning the strings. If the latter is zero, the failure point is
3391 a ``dummy''; if a failure happens and the failure point is a dummy,
3392 it gets discarded and the next next one is tried. */
3393 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
3394 fail_stack_type fail_stack
;
3397 static unsigned failure_id
= 0;
3398 unsigned nfailure_points_pushed
= 0, nfailure_points_popped
= 0;
3401 /* We fill all the registers internally, independent of what we
3402 return, for use in backreferences. The number here includes
3403 an element for register zero. */
3404 unsigned num_regs
= bufp
->re_nsub
+ 1;
3406 /* The currently active registers. */
3407 unsigned lowest_active_reg
= NO_LOWEST_ACTIVE_REG
;
3408 unsigned highest_active_reg
= NO_HIGHEST_ACTIVE_REG
;
3410 /* Information on the contents of registers. These are pointers into
3411 the input strings; they record just what was matched (on this
3412 attempt) by a subexpression part of the pattern, that is, the
3413 regnum-th regstart pointer points to where in the pattern we began
3414 matching and the regnum-th regend points to right after where we
3415 stopped matching the regnum-th subexpression. (The zeroth register
3416 keeps track of what the whole pattern matches.) */
3417 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
3418 const char **regstart
, **regend
;
3421 /* If a group that's operated upon by a repetition operator fails to
3422 match anything, then the register for its start will need to be
3423 restored because it will have been set to wherever in the string we
3424 are when we last see its open-group operator. Similarly for a
3426 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
3427 const char **old_regstart
, **old_regend
;
3430 /* The is_active field of reg_info helps us keep track of which (possibly
3431 nested) subexpressions we are currently in. The matched_something
3432 field of reg_info[reg_num] helps us tell whether or not we have
3433 matched any of the pattern so far this time through the reg_num-th
3434 subexpression. These two fields get reset each time through any
3435 loop their register is in. */
3436 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
3437 register_info_type
*reg_info
;
3440 /* The following record the register info as found in the above
3441 variables when we find a match better than any we've seen before.
3442 This happens as we backtrack through the failure points, which in
3443 turn happens only if we have not yet matched the entire string. */
3444 unsigned best_regs_set
= false;
3445 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
3446 const char **best_regstart
, **best_regend
;
3449 /* Logically, this is `best_regend[0]'. But we don't want to have to
3450 allocate space for that if we're not allocating space for anything
3451 else (see below). Also, we never need info about register 0 for
3452 any of the other register vectors, and it seems rather a kludge to
3453 treat `best_regend' differently than the rest. So we keep track of
3454 the end of the best match so far in a separate variable. We
3455 initialize this to NULL so that when we backtrack the first time
3456 and need to test it, it's not garbage. */
3457 const char *match_end
= NULL
;
3459 /* Used when we pop values we don't care about. */
3460 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
3461 const char **reg_dummy
;
3462 register_info_type
*reg_info_dummy
;
3466 /* Counts the total number of registers pushed. */
3467 unsigned num_regs_pushed
= 0;
3470 DEBUG_PRINT1 ("\n\nEntering re_match_2.\n");
3474 #ifdef MATCH_MAY_ALLOCATE
3475 /* Do not bother to initialize all the register variables if there are
3476 no groups in the pattern, as it takes a fair amount of time. If
3477 there are groups, we include space for register 0 (the whole
3478 pattern), even though we never use it, since it simplifies the
3479 array indexing. We should fix this. */
3482 regstart
= REGEX_TALLOC (num_regs
, const char *);
3483 regend
= REGEX_TALLOC (num_regs
, const char *);
3484 old_regstart
= REGEX_TALLOC (num_regs
, const char *);
3485 old_regend
= REGEX_TALLOC (num_regs
, const char *);
3486 best_regstart
= REGEX_TALLOC (num_regs
, const char *);
3487 best_regend
= REGEX_TALLOC (num_regs
, const char *);
3488 reg_info
= REGEX_TALLOC (num_regs
, register_info_type
);
3489 reg_dummy
= REGEX_TALLOC (num_regs
, const char *);
3490 reg_info_dummy
= REGEX_TALLOC (num_regs
, register_info_type
);
3492 if (!(regstart
&& regend
&& old_regstart
&& old_regend
&& reg_info
3493 && best_regstart
&& best_regend
&& reg_dummy
&& reg_info_dummy
))
3499 #if defined (REGEX_MALLOC)
3502 /* We must initialize all our variables to NULL, so that
3503 `FREE_VARIABLES' doesn't try to free them. */
3504 regstart
= regend
= old_regstart
= old_regend
= best_regstart
3505 = best_regend
= reg_dummy
= NULL
;
3506 reg_info
= reg_info_dummy
= (register_info_type
*) NULL
;
3508 #endif /* REGEX_MALLOC */
3509 #endif /* MATCH_MAY_ALLOCATE */
3511 /* The starting position is bogus. */
3512 if (pos
< 0 || pos
> size1
+ size2
)
3518 /* Initialize subexpression text positions to -1 to mark ones that no
3519 start_memory/stop_memory has been seen for. Also initialize the
3520 register information struct. */
3521 for (mcnt
= 1; mcnt
< num_regs
; mcnt
++)
3523 regstart
[mcnt
] = regend
[mcnt
]
3524 = old_regstart
[mcnt
] = old_regend
[mcnt
] = REG_UNSET_VALUE
;
3526 REG_MATCH_NULL_STRING_P (reg_info
[mcnt
]) = MATCH_NULL_UNSET_VALUE
;
3527 IS_ACTIVE (reg_info
[mcnt
]) = 0;
3528 MATCHED_SOMETHING (reg_info
[mcnt
]) = 0;
3529 EVER_MATCHED_SOMETHING (reg_info
[mcnt
]) = 0;
3532 /* We move `string1' into `string2' if the latter's empty -- but not if
3533 `string1' is null. */
3534 if (size2
== 0 && string1
!= NULL
)
3541 end1
= string1
+ size1
;
3542 end2
= string2
+ size2
;
3544 /* Compute where to stop matching, within the two strings. */
3547 end_match_1
= string1
+ stop
;
3548 end_match_2
= string2
;
3553 end_match_2
= string2
+ stop
- size1
;
3556 /* `p' scans through the pattern as `d' scans through the data.
3557 `dend' is the end of the input string that `d' points within. `d'
3558 is advanced into the following input string whenever necessary, but
3559 this happens before fetching; therefore, at the beginning of the
3560 loop, `d' can be pointing at the end of a string, but it cannot
3562 if (size1
> 0 && pos
<= size1
)
3569 d
= string2
+ pos
- size1
;
3573 DEBUG_PRINT1 ("The compiled pattern is: ");
3574 DEBUG_PRINT_COMPILED_PATTERN (bufp
, p
, pend
);
3575 DEBUG_PRINT1 ("The string to match is: `");
3576 DEBUG_PRINT_DOUBLE_STRING (d
, string1
, size1
, string2
, size2
);
3577 DEBUG_PRINT1 ("'\n");
3579 /* This loops over pattern commands. It exits by returning from the
3580 function if the match is complete, or it drops through if the match
3581 fails at this starting point in the input data. */
3584 DEBUG_PRINT2 ("\n0x%x: ", p
);
3587 { /* End of pattern means we might have succeeded. */
3588 DEBUG_PRINT1 ("end of pattern ... ");
3590 /* If we haven't matched the entire string, and we want the
3591 longest match, try backtracking. */
3592 if (d
!= end_match_2
)
3594 /* 1 if this match ends in the same string (string1 or string2)
3595 as the best previous match. */
3596 boolean same_str_p
= (FIRST_STRING_P (match_end
)
3597 == MATCHING_IN_FIRST_STRING
);
3598 /* 1 if this match is the best seen so far. */
3599 boolean best_match_p
;
3601 /* AIX compiler got confused when this was combined
3602 with the previous declaration. */
3604 best_match_p
= d
> match_end
;
3606 best_match_p
= !MATCHING_IN_FIRST_STRING
;
3608 DEBUG_PRINT1 ("backtracking.\n");
3610 if (!FAIL_STACK_EMPTY ())
3611 { /* More failure points to try. */
3613 /* If exceeds best match so far, save it. */
3614 if (!best_regs_set
|| best_match_p
)
3616 best_regs_set
= true;
3619 DEBUG_PRINT1 ("\nSAVING match as best so far.\n");
3621 for (mcnt
= 1; mcnt
< num_regs
; mcnt
++)
3623 best_regstart
[mcnt
] = regstart
[mcnt
];
3624 best_regend
[mcnt
] = regend
[mcnt
];
3630 /* If no failure points, don't restore garbage. And if
3631 last match is real best match, don't restore second
3633 else if (best_regs_set
&& !best_match_p
)
3636 /* Restore best match. It may happen that `dend ==
3637 end_match_1' while the restored d is in string2.
3638 For example, the pattern `x.*y.*z' against the
3639 strings `x-' and `y-z-', if the two strings are
3640 not consecutive in memory. */
3641 DEBUG_PRINT1 ("Restoring best registers.\n");
3644 dend
= ((d
>= string1
&& d
<= end1
)
3645 ? end_match_1
: end_match_2
);
3647 for (mcnt
= 1; mcnt
< num_regs
; mcnt
++)
3649 regstart
[mcnt
] = best_regstart
[mcnt
];
3650 regend
[mcnt
] = best_regend
[mcnt
];
3653 } /* d != end_match_2 */
3655 DEBUG_PRINT1 ("Accepting match.\n");
3657 /* If caller wants register contents data back, do it. */
3658 if (regs
&& !bufp
->no_sub
)
3660 /* Have the register data arrays been allocated? */
3661 if (bufp
->regs_allocated
== REGS_UNALLOCATED
)
3662 { /* No. So allocate them with malloc. We need one
3663 extra element beyond `num_regs' for the `-1' marker
3665 regs
->num_regs
= MAX (RE_NREGS
, num_regs
+ 1);
3666 regs
->start
= TALLOC (regs
->num_regs
, regoff_t
);
3667 regs
->end
= TALLOC (regs
->num_regs
, regoff_t
);
3668 if (regs
->start
== NULL
|| regs
->end
== NULL
)
3670 bufp
->regs_allocated
= REGS_REALLOCATE
;
3672 else if (bufp
->regs_allocated
== REGS_REALLOCATE
)
3673 { /* Yes. If we need more elements than were already
3674 allocated, reallocate them. If we need fewer, just
3676 if (regs
->num_regs
< num_regs
+ 1)
3678 regs
->num_regs
= num_regs
+ 1;
3679 RETALLOC (regs
->start
, regs
->num_regs
, regoff_t
);
3680 RETALLOC (regs
->end
, regs
->num_regs
, regoff_t
);
3681 if (regs
->start
== NULL
|| regs
->end
== NULL
)
3687 /* These braces fend off a "empty body in an else-statement"
3688 warning under GCC when assert expands to nothing. */
3689 assert (bufp
->regs_allocated
== REGS_FIXED
);
3692 /* Convert the pointer data in `regstart' and `regend' to
3693 indices. Register zero has to be set differently,
3694 since we haven't kept track of any info for it. */
3695 if (regs
->num_regs
> 0)
3697 regs
->start
[0] = pos
;
3698 regs
->end
[0] = (MATCHING_IN_FIRST_STRING
3699 ? ((regoff_t
) (d
- string1
))
3700 : ((regoff_t
) (d
- string2
+ size1
)));
3703 /* Go through the first `min (num_regs, regs->num_regs)'
3704 registers, since that is all we initialized. */
3705 for (mcnt
= 1; mcnt
< MIN (num_regs
, regs
->num_regs
); mcnt
++)
3707 if (REG_UNSET (regstart
[mcnt
]) || REG_UNSET (regend
[mcnt
]))
3708 regs
->start
[mcnt
] = regs
->end
[mcnt
] = -1;
3712 = (regoff_t
) POINTER_TO_OFFSET (regstart
[mcnt
]);
3714 = (regoff_t
) POINTER_TO_OFFSET (regend
[mcnt
]);
3718 /* If the regs structure we return has more elements than
3719 were in the pattern, set the extra elements to -1. If
3720 we (re)allocated the registers, this is the case,
3721 because we always allocate enough to have at least one
3723 for (mcnt
= num_regs
; mcnt
< regs
->num_regs
; mcnt
++)
3724 regs
->start
[mcnt
] = regs
->end
[mcnt
] = -1;
3725 } /* regs && !bufp->no_sub */
3728 DEBUG_PRINT4 ("%u failure points pushed, %u popped (%u remain).\n",
3729 nfailure_points_pushed
, nfailure_points_popped
,
3730 nfailure_points_pushed
- nfailure_points_popped
);
3731 DEBUG_PRINT2 ("%u registers pushed.\n", num_regs_pushed
);
3733 mcnt
= d
- pos
- (MATCHING_IN_FIRST_STRING
3737 DEBUG_PRINT2 ("Returning %d from re_match_2.\n", mcnt
);
3742 /* Otherwise match next pattern command. */
3743 #ifdef SWITCH_ENUM_BUG
3744 switch ((int) ((re_opcode_t
) *p
++))
3746 switch ((re_opcode_t
) *p
++)
3749 /* Ignore these. Used to ignore the n of succeed_n's which
3750 currently have n == 0. */
3752 DEBUG_PRINT1 ("EXECUTING no_op.\n");
3756 /* Match the next n pattern characters exactly. The following
3757 byte in the pattern defines n, and the n bytes after that
3758 are the characters to match. */
3761 DEBUG_PRINT2 ("EXECUTING exactn %d.\n", mcnt
);
3763 /* This is written out as an if-else so we don't waste time
3764 testing `translate' inside the loop. */
3770 if (translate
[(unsigned char) *d
++] != (char) *p
++)
3780 if (*d
++ != (char) *p
++) goto fail
;
3784 SET_REGS_MATCHED ();
3788 /* Match any character except possibly a newline or a null. */
3790 DEBUG_PRINT1 ("EXECUTING anychar.\n");
3794 if ((!(bufp
->syntax
& RE_DOT_NEWLINE
) && TRANSLATE (*d
) == '\n')
3795 || (bufp
->syntax
& RE_DOT_NOT_NULL
&& TRANSLATE (*d
) == '\000'))
3798 SET_REGS_MATCHED ();
3799 DEBUG_PRINT2 (" Matched `%d'.\n", *d
);
3807 register unsigned char c
;
3808 boolean
not = (re_opcode_t
) *(p
- 1) == charset_not
;
3810 DEBUG_PRINT2 ("EXECUTING charset%s.\n", not ? "_not" : "");
3813 c
= TRANSLATE (*d
); /* The character to match. */
3815 /* Cast to `unsigned' instead of `unsigned char' in case the
3816 bit list is a full 32 bytes long. */
3817 if (c
< (unsigned) (*p
* BYTEWIDTH
)
3818 && p
[1 + c
/ BYTEWIDTH
] & (1 << (c
% BYTEWIDTH
)))
3823 if (!not) goto fail
;
3825 SET_REGS_MATCHED ();
3831 /* The beginning of a group is represented by start_memory.
3832 The arguments are the register number in the next byte, and the
3833 number of groups inner to this one in the next. The text
3834 matched within the group is recorded (in the internal
3835 registers data structure) under the register number. */
3837 DEBUG_PRINT3 ("EXECUTING start_memory %d (%d):\n", *p
, p
[1]);
3839 /* Find out if this group can match the empty string. */
3840 p1
= p
; /* To send to group_match_null_string_p. */
3842 if (REG_MATCH_NULL_STRING_P (reg_info
[*p
]) == MATCH_NULL_UNSET_VALUE
)
3843 REG_MATCH_NULL_STRING_P (reg_info
[*p
])
3844 = group_match_null_string_p (&p1
, pend
, reg_info
);
3846 /* Save the position in the string where we were the last time
3847 we were at this open-group operator in case the group is
3848 operated upon by a repetition operator, e.g., with `(a*)*b'
3849 against `ab'; then we want to ignore where we are now in
3850 the string in case this attempt to match fails. */
3851 old_regstart
[*p
] = REG_MATCH_NULL_STRING_P (reg_info
[*p
])
3852 ? REG_UNSET (regstart
[*p
]) ? d
: regstart
[*p
]
3854 DEBUG_PRINT2 (" old_regstart: %d\n",
3855 POINTER_TO_OFFSET (old_regstart
[*p
]));
3858 DEBUG_PRINT2 (" regstart: %d\n", POINTER_TO_OFFSET (regstart
[*p
]));
3860 IS_ACTIVE (reg_info
[*p
]) = 1;
3861 MATCHED_SOMETHING (reg_info
[*p
]) = 0;
3863 /* This is the new highest active register. */
3864 highest_active_reg
= *p
;
3866 /* If nothing was active before, this is the new lowest active
3868 if (lowest_active_reg
== NO_LOWEST_ACTIVE_REG
)
3869 lowest_active_reg
= *p
;
3871 /* Move past the register number and inner group count. */
3873 just_past_start_mem
= p
;
3877 /* The stop_memory opcode represents the end of a group. Its
3878 arguments are the same as start_memory's: the register
3879 number, and the number of inner groups. */
3881 DEBUG_PRINT3 ("EXECUTING stop_memory %d (%d):\n", *p
, p
[1]);
3883 /* We need to save the string position the last time we were at
3884 this close-group operator in case the group is operated
3885 upon by a repetition operator, e.g., with `((a*)*(b*)*)*'
3886 against `aba'; then we want to ignore where we are now in
3887 the string in case this attempt to match fails. */
3888 old_regend
[*p
] = REG_MATCH_NULL_STRING_P (reg_info
[*p
])
3889 ? REG_UNSET (regend
[*p
]) ? d
: regend
[*p
]
3891 DEBUG_PRINT2 (" old_regend: %d\n",
3892 POINTER_TO_OFFSET (old_regend
[*p
]));
3895 DEBUG_PRINT2 (" regend: %d\n", POINTER_TO_OFFSET (regend
[*p
]));
3897 /* This register isn't active anymore. */
3898 IS_ACTIVE (reg_info
[*p
]) = 0;
3900 /* If this was the only register active, nothing is active
3902 if (lowest_active_reg
== highest_active_reg
)
3904 lowest_active_reg
= NO_LOWEST_ACTIVE_REG
;
3905 highest_active_reg
= NO_HIGHEST_ACTIVE_REG
;
3908 { /* We must scan for the new highest active register, since
3909 it isn't necessarily one less than now: consider
3910 (a(b)c(d(e)f)g). When group 3 ends, after the f), the
3911 new highest active register is 1. */
3912 unsigned char r
= *p
- 1;
3913 while (r
> 0 && !IS_ACTIVE (reg_info
[r
]))
3916 /* If we end up at register zero, that means that we saved
3917 the registers as the result of an `on_failure_jump', not
3918 a `start_memory', and we jumped to past the innermost
3919 `stop_memory'. For example, in ((.)*) we save
3920 registers 1 and 2 as a result of the *, but when we pop
3921 back to the second ), we are at the stop_memory 1.
3922 Thus, nothing is active. */
3925 lowest_active_reg
= NO_LOWEST_ACTIVE_REG
;
3926 highest_active_reg
= NO_HIGHEST_ACTIVE_REG
;
3929 highest_active_reg
= r
;
3932 /* If just failed to match something this time around with a
3933 group that's operated on by a repetition operator, try to
3934 force exit from the ``loop'', and restore the register
3935 information for this group that we had before trying this
3937 if ((!MATCHED_SOMETHING (reg_info
[*p
])
3938 || just_past_start_mem
== p
- 1)
3941 boolean is_a_jump_n
= false;
3945 switch ((re_opcode_t
) *p1
++)
3949 case pop_failure_jump
:
3950 case maybe_pop_jump
:
3952 case dummy_failure_jump
:
3953 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
3963 /* If the next operation is a jump backwards in the pattern
3964 to an on_failure_jump right before the start_memory
3965 corresponding to this stop_memory, exit from the loop
3966 by forcing a failure after pushing on the stack the
3967 on_failure_jump's jump in the pattern, and d. */
3968 if (mcnt
< 0 && (re_opcode_t
) *p1
== on_failure_jump
3969 && (re_opcode_t
) p1
[3] == start_memory
&& p1
[4] == *p
)
3971 /* If this group ever matched anything, then restore
3972 what its registers were before trying this last
3973 failed match, e.g., with `(a*)*b' against `ab' for
3974 regstart[1], and, e.g., with `((a*)*(b*)*)*'
3975 against `aba' for regend[3].
3977 Also restore the registers for inner groups for,
3978 e.g., `((a*)(b*))*' against `aba' (register 3 would
3979 otherwise get trashed). */
3981 if (EVER_MATCHED_SOMETHING (reg_info
[*p
]))
3985 EVER_MATCHED_SOMETHING (reg_info
[*p
]) = 0;
3987 /* Restore this and inner groups' (if any) registers. */
3988 for (r
= *p
; r
< *p
+ *(p
+ 1); r
++)
3990 regstart
[r
] = old_regstart
[r
];
3992 /* xx why this test? */
3993 if ((int) old_regend
[r
] >= (int) regstart
[r
])
3994 regend
[r
] = old_regend
[r
];
3998 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
3999 PUSH_FAILURE_POINT (p1
+ mcnt
, d
, -2);
4005 /* Move past the register number and the inner group count. */
4010 /* \<digit> has been turned into a `duplicate' command which is
4011 followed by the numeric value of <digit> as the register number. */
4014 register const char *d2
, *dend2
;
4015 int regno
= *p
++; /* Get which register to match against. */
4016 DEBUG_PRINT2 ("EXECUTING duplicate %d.\n", regno
);
4018 /* Can't back reference a group which we've never matched. */
4019 if (REG_UNSET (regstart
[regno
]) || REG_UNSET (regend
[regno
]))
4022 /* Where in input to try to start matching. */
4023 d2
= regstart
[regno
];
4025 /* Where to stop matching; if both the place to start and
4026 the place to stop matching are in the same string, then
4027 set to the place to stop, otherwise, for now have to use
4028 the end of the first string. */
4030 dend2
= ((FIRST_STRING_P (regstart
[regno
])
4031 == FIRST_STRING_P (regend
[regno
]))
4032 ? regend
[regno
] : end_match_1
);
4035 /* If necessary, advance to next segment in register
4039 if (dend2
== end_match_2
) break;
4040 if (dend2
== regend
[regno
]) break;
4042 /* End of string1 => advance to string2. */
4044 dend2
= regend
[regno
];
4046 /* At end of register contents => success */
4047 if (d2
== dend2
) break;
4049 /* If necessary, advance to next segment in data. */
4052 /* How many characters left in this segment to match. */
4055 /* Want how many consecutive characters we can match in
4056 one shot, so, if necessary, adjust the count. */
4057 if (mcnt
> dend2
- d2
)
4060 /* Compare that many; failure if mismatch, else move
4063 ? bcmp_translate (d
, d2
, mcnt
, translate
)
4064 : bcmp (d
, d2
, mcnt
))
4066 d
+= mcnt
, d2
+= mcnt
;
4072 /* begline matches the empty string at the beginning of the string
4073 (unless `not_bol' is set in `bufp'), and, if
4074 `newline_anchor' is set, after newlines. */
4076 DEBUG_PRINT1 ("EXECUTING begline.\n");
4078 if (AT_STRINGS_BEG (d
))
4080 if (!bufp
->not_bol
) break;
4082 else if (d
[-1] == '\n' && bufp
->newline_anchor
)
4086 /* In all other cases, we fail. */
4090 /* endline is the dual of begline. */
4092 DEBUG_PRINT1 ("EXECUTING endline.\n");
4094 if (AT_STRINGS_END (d
))
4096 if (!bufp
->not_eol
) break;
4099 /* We have to ``prefetch'' the next character. */
4100 else if ((d
== end1
? *string2
: *d
) == '\n'
4101 && bufp
->newline_anchor
)
4108 /* Match at the very beginning of the data. */
4110 DEBUG_PRINT1 ("EXECUTING begbuf.\n");
4111 if (AT_STRINGS_BEG (d
))
4116 /* Match at the very end of the data. */
4118 DEBUG_PRINT1 ("EXECUTING endbuf.\n");
4119 if (AT_STRINGS_END (d
))
4124 /* on_failure_keep_string_jump is used to optimize `.*\n'. It
4125 pushes NULL as the value for the string on the stack. Then
4126 `pop_failure_point' will keep the current value for the
4127 string, instead of restoring it. To see why, consider
4128 matching `foo\nbar' against `.*\n'. The .* matches the foo;
4129 then the . fails against the \n. But the next thing we want
4130 to do is match the \n against the \n; if we restored the
4131 string value, we would be back at the foo.
4133 Because this is used only in specific cases, we don't need to
4134 check all the things that `on_failure_jump' does, to make
4135 sure the right things get saved on the stack. Hence we don't
4136 share its code. The only reason to push anything on the
4137 stack at all is that otherwise we would have to change
4138 `anychar's code to do something besides goto fail in this
4139 case; that seems worse than this. */
4140 case on_failure_keep_string_jump
:
4141 DEBUG_PRINT1 ("EXECUTING on_failure_keep_string_jump");
4143 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
4144 DEBUG_PRINT3 (" %d (to 0x%x):\n", mcnt
, p
+ mcnt
);
4146 PUSH_FAILURE_POINT (p
+ mcnt
, NULL
, -2);
4150 /* Uses of on_failure_jump:
4152 Each alternative starts with an on_failure_jump that points
4153 to the beginning of the next alternative. Each alternative
4154 except the last ends with a jump that in effect jumps past
4155 the rest of the alternatives. (They really jump to the
4156 ending jump of the following alternative, because tensioning
4157 these jumps is a hassle.)
4159 Repeats start with an on_failure_jump that points past both
4160 the repetition text and either the following jump or
4161 pop_failure_jump back to this on_failure_jump. */
4162 case on_failure_jump
:
4164 DEBUG_PRINT1 ("EXECUTING on_failure_jump");
4166 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
4167 DEBUG_PRINT3 (" %d (to 0x%x)", mcnt
, p
+ mcnt
);
4169 /* If this on_failure_jump comes right before a group (i.e.,
4170 the original * applied to a group), save the information
4171 for that group and all inner ones, so that if we fail back
4172 to this point, the group's information will be correct.
4173 For example, in \(a*\)*\1, we need the preceding group,
4174 and in \(\(a*\)b*\)\2, we need the inner group. */
4176 /* We can't use `p' to check ahead because we push
4177 a failure point to `p + mcnt' after we do this. */
4180 /* We need to skip no_op's before we look for the
4181 start_memory in case this on_failure_jump is happening as
4182 the result of a completed succeed_n, as in \(a\)\{1,3\}b\1
4184 while (p1
< pend
&& (re_opcode_t
) *p1
== no_op
)
4187 if (p1
< pend
&& (re_opcode_t
) *p1
== start_memory
)
4189 /* We have a new highest active register now. This will
4190 get reset at the start_memory we are about to get to,
4191 but we will have saved all the registers relevant to
4192 this repetition op, as described above. */
4193 highest_active_reg
= *(p1
+ 1) + *(p1
+ 2);
4194 if (lowest_active_reg
== NO_LOWEST_ACTIVE_REG
)
4195 lowest_active_reg
= *(p1
+ 1);
4198 DEBUG_PRINT1 (":\n");
4199 PUSH_FAILURE_POINT (p
+ mcnt
, d
, -2);
4203 /* A smart repeat ends with `maybe_pop_jump'.
4204 We change it to either `pop_failure_jump' or `jump'. */
4205 case maybe_pop_jump
:
4206 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
4207 DEBUG_PRINT2 ("EXECUTING maybe_pop_jump %d.\n", mcnt
);
4209 register unsigned char *p2
= p
;
4211 /* Compare the beginning of the repeat with what in the
4212 pattern follows its end. If we can establish that there
4213 is nothing that they would both match, i.e., that we
4214 would have to backtrack because of (as in, e.g., `a*a')
4215 then we can change to pop_failure_jump, because we'll
4216 never have to backtrack.
4218 This is not true in the case of alternatives: in
4219 `(a|ab)*' we do need to backtrack to the `ab' alternative
4220 (e.g., if the string was `ab'). But instead of trying to
4221 detect that here, the alternative has put on a dummy
4222 failure point which is what we will end up popping. */
4224 /* Skip over open/close-group commands.
4225 If what follows this loop is a ...+ construct,
4226 look at what begins its body, since we will have to
4227 match at least one of that. */
4231 && ((re_opcode_t
) *p2
== stop_memory
4232 || (re_opcode_t
) *p2
== start_memory
))
4234 else if (p2
+ 6 < pend
4235 && (re_opcode_t
) *p2
== dummy_failure_jump
)
4242 /* p1[0] ... p1[2] are the `on_failure_jump' corresponding
4243 to the `maybe_finalize_jump' of this case. Examine what
4246 /* If we're at the end of the pattern, we can change. */
4249 /* Consider what happens when matching ":\(.*\)"
4250 against ":/". I don't really understand this code
4252 p
[-3] = (unsigned char) pop_failure_jump
;
4254 (" End of pattern: change to `pop_failure_jump'.\n");
4257 else if ((re_opcode_t
) *p2
== exactn
4258 || (bufp
->newline_anchor
&& (re_opcode_t
) *p2
== endline
))
4260 register unsigned char c
4261 = *p2
== (unsigned char) endline
? '\n' : p2
[2];
4263 if ((re_opcode_t
) p1
[3] == exactn
&& p1
[5] != c
)
4265 p
[-3] = (unsigned char) pop_failure_jump
;
4266 DEBUG_PRINT3 (" %c != %c => pop_failure_jump.\n",
4270 else if ((re_opcode_t
) p1
[3] == charset
4271 || (re_opcode_t
) p1
[3] == charset_not
)
4273 int not = (re_opcode_t
) p1
[3] == charset_not
;
4275 if (c
< (unsigned char) (p1
[4] * BYTEWIDTH
)
4276 && p1
[5 + c
/ BYTEWIDTH
] & (1 << (c
% BYTEWIDTH
)))
4279 /* `not' is equal to 1 if c would match, which means
4280 that we can't change to pop_failure_jump. */
4283 p
[-3] = (unsigned char) pop_failure_jump
;
4284 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
4288 else if ((re_opcode_t
) *p2
== charset
)
4291 register unsigned char c
4292 = *p2
== (unsigned char) endline
? '\n' : p2
[2];
4295 if ((re_opcode_t
) p1
[3] == exactn
4296 && ! ((int) p2
[1] * BYTEWIDTH
> (int) p1
[4]
4297 && (p2
[1 + p1
[4] / BYTEWIDTH
]
4298 & (1 << (p1
[4] % BYTEWIDTH
)))))
4300 p
[-3] = (unsigned char) pop_failure_jump
;
4301 DEBUG_PRINT3 (" %c != %c => pop_failure_jump.\n",
4305 else if ((re_opcode_t
) p1
[3] == charset_not
)
4308 /* We win if the charset_not inside the loop
4309 lists every character listed in the charset after. */
4310 for (idx
= 0; idx
< (int) p2
[1]; idx
++)
4311 if (! (p2
[2 + idx
] == 0
4312 || (idx
< (int) p1
[4]
4313 && ((p2
[2 + idx
] & ~ p1
[5 + idx
]) == 0))))
4318 p
[-3] = (unsigned char) pop_failure_jump
;
4319 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
4322 else if ((re_opcode_t
) p1
[3] == charset
)
4325 /* We win if the charset inside the loop
4326 has no overlap with the one after the loop. */
4328 idx
< (int) p2
[1] && idx
< (int) p1
[4];
4330 if ((p2
[2 + idx
] & p1
[5 + idx
]) != 0)
4333 if (idx
== p2
[1] || idx
== p1
[4])
4335 p
[-3] = (unsigned char) pop_failure_jump
;
4336 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
4341 p
-= 2; /* Point at relative address again. */
4342 if ((re_opcode_t
) p
[-1] != pop_failure_jump
)
4344 p
[-1] = (unsigned char) jump
;
4345 DEBUG_PRINT1 (" Match => jump.\n");
4346 goto unconditional_jump
;
4348 /* Note fall through. */
4351 /* The end of a simple repeat has a pop_failure_jump back to
4352 its matching on_failure_jump, where the latter will push a
4353 failure point. The pop_failure_jump takes off failure
4354 points put on by this pop_failure_jump's matching
4355 on_failure_jump; we got through the pattern to here from the
4356 matching on_failure_jump, so didn't fail. */
4357 case pop_failure_jump
:
4359 /* We need to pass separate storage for the lowest and
4360 highest registers, even though we don't care about the
4361 actual values. Otherwise, we will restore only one
4362 register from the stack, since lowest will == highest in
4363 `pop_failure_point'. */
4364 unsigned dummy_low_reg
, dummy_high_reg
;
4365 unsigned char *pdummy
;
4368 DEBUG_PRINT1 ("EXECUTING pop_failure_jump.\n");
4369 POP_FAILURE_POINT (sdummy
, pdummy
,
4370 dummy_low_reg
, dummy_high_reg
,
4371 reg_dummy
, reg_dummy
, reg_info_dummy
);
4373 /* Note fall through. */
4376 /* Unconditionally jump (without popping any failure points). */
4379 EXTRACT_NUMBER_AND_INCR (mcnt
, p
); /* Get the amount to jump. */
4380 DEBUG_PRINT2 ("EXECUTING jump %d ", mcnt
);
4381 p
+= mcnt
; /* Do the jump. */
4382 DEBUG_PRINT2 ("(to 0x%x).\n", p
);
4386 /* We need this opcode so we can detect where alternatives end
4387 in `group_match_null_string_p' et al. */
4389 DEBUG_PRINT1 ("EXECUTING jump_past_alt.\n");
4390 goto unconditional_jump
;
4393 /* Normally, the on_failure_jump pushes a failure point, which
4394 then gets popped at pop_failure_jump. We will end up at
4395 pop_failure_jump, also, and with a pattern of, say, `a+', we
4396 are skipping over the on_failure_jump, so we have to push
4397 something meaningless for pop_failure_jump to pop. */
4398 case dummy_failure_jump
:
4399 DEBUG_PRINT1 ("EXECUTING dummy_failure_jump.\n");
4400 /* It doesn't matter what we push for the string here. What
4401 the code at `fail' tests is the value for the pattern. */
4402 PUSH_FAILURE_POINT (0, 0, -2);
4403 goto unconditional_jump
;
4406 /* At the end of an alternative, we need to push a dummy failure
4407 point in case we are followed by a `pop_failure_jump', because
4408 we don't want the failure point for the alternative to be
4409 popped. For example, matching `(a|ab)*' against `aab'
4410 requires that we match the `ab' alternative. */
4411 case push_dummy_failure
:
4412 DEBUG_PRINT1 ("EXECUTING push_dummy_failure.\n");
4413 /* See comments just above at `dummy_failure_jump' about the
4415 PUSH_FAILURE_POINT (0, 0, -2);
4418 /* Have to succeed matching what follows at least n times.
4419 After that, handle like `on_failure_jump'. */
4421 EXTRACT_NUMBER (mcnt
, p
+ 2);
4422 DEBUG_PRINT2 ("EXECUTING succeed_n %d.\n", mcnt
);
4425 /* Originally, this is how many times we HAVE to succeed. */
4430 STORE_NUMBER_AND_INCR (p
, mcnt
);
4431 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p
, mcnt
);
4435 DEBUG_PRINT2 (" Setting two bytes from 0x%x to no_op.\n", p
+2);
4436 p
[2] = (unsigned char) no_op
;
4437 p
[3] = (unsigned char) no_op
;
4443 EXTRACT_NUMBER (mcnt
, p
+ 2);
4444 DEBUG_PRINT2 ("EXECUTING jump_n %d.\n", mcnt
);
4446 /* Originally, this is how many times we CAN jump. */
4450 STORE_NUMBER (p
+ 2, mcnt
);
4451 goto unconditional_jump
;
4453 /* If don't have to jump any more, skip over the rest of command. */
4460 DEBUG_PRINT1 ("EXECUTING set_number_at.\n");
4462 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
4464 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
4465 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p1
, mcnt
);
4466 STORE_NUMBER (p1
, mcnt
);
4471 DEBUG_PRINT1 ("EXECUTING wordbound.\n");
4472 if (AT_WORD_BOUNDARY (d
))
4477 DEBUG_PRINT1 ("EXECUTING notwordbound.\n");
4478 if (AT_WORD_BOUNDARY (d
))
4483 DEBUG_PRINT1 ("EXECUTING wordbeg.\n");
4484 if (WORDCHAR_P (d
) && (AT_STRINGS_BEG (d
) || !WORDCHAR_P (d
- 1)))
4489 DEBUG_PRINT1 ("EXECUTING wordend.\n");
4490 if (!AT_STRINGS_BEG (d
) && WORDCHAR_P (d
- 1)
4491 && (!WORDCHAR_P (d
) || AT_STRINGS_END (d
)))
4497 DEBUG_PRINT1 ("EXECUTING before_dot.\n");
4498 if (PTR_CHAR_POS ((unsigned char *) d
) >= point
)
4503 DEBUG_PRINT1 ("EXECUTING at_dot.\n");
4504 if (PTR_CHAR_POS ((unsigned char *) d
) != point
)
4509 DEBUG_PRINT1 ("EXECUTING after_dot.\n");
4510 if (PTR_CHAR_POS ((unsigned char *) d
) <= point
)
4513 #if 0 /* not emacs19 */
4515 DEBUG_PRINT1 ("EXECUTING at_dot.\n");
4516 if (PTR_CHAR_POS ((unsigned char *) d
) + 1 != point
)
4519 #endif /* not emacs19 */
4522 DEBUG_PRINT2 ("EXECUTING syntaxspec %d.\n", mcnt
);
4527 DEBUG_PRINT1 ("EXECUTING Emacs wordchar.\n");
4531 /* Can't use *d++ here; SYNTAX may be an unsafe macro. */
4533 if (SYNTAX (d
[-1]) != (enum syntaxcode
) mcnt
)
4535 SET_REGS_MATCHED ();
4539 DEBUG_PRINT2 ("EXECUTING notsyntaxspec %d.\n", mcnt
);
4541 goto matchnotsyntax
;
4544 DEBUG_PRINT1 ("EXECUTING Emacs notwordchar.\n");
4548 /* Can't use *d++ here; SYNTAX may be an unsafe macro. */
4550 if (SYNTAX (d
[-1]) == (enum syntaxcode
) mcnt
)
4552 SET_REGS_MATCHED ();
4555 #else /* not emacs */
4557 DEBUG_PRINT1 ("EXECUTING non-Emacs wordchar.\n");
4559 if (!WORDCHAR_P (d
))
4561 SET_REGS_MATCHED ();
4566 DEBUG_PRINT1 ("EXECUTING non-Emacs notwordchar.\n");
4570 SET_REGS_MATCHED ();
4573 #endif /* not emacs */
4578 continue; /* Successfully executed one pattern command; keep going. */
4581 /* We goto here if a matching operation fails. */
4583 if (!FAIL_STACK_EMPTY ())
4584 { /* A restart point is known. Restore to that state. */
4585 DEBUG_PRINT1 ("\nFAIL:\n");
4586 POP_FAILURE_POINT (d
, p
,
4587 lowest_active_reg
, highest_active_reg
,
4588 regstart
, regend
, reg_info
);
4590 /* If this failure point is a dummy, try the next one. */
4594 /* If we failed to the end of the pattern, don't examine *p. */
4598 boolean is_a_jump_n
= false;
4600 /* If failed to a backwards jump that's part of a repetition
4601 loop, need to pop this failure point and use the next one. */
4602 switch ((re_opcode_t
) *p
)
4606 case maybe_pop_jump
:
4607 case pop_failure_jump
:
4610 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
4613 if ((is_a_jump_n
&& (re_opcode_t
) *p1
== succeed_n
)
4615 && (re_opcode_t
) *p1
== on_failure_jump
))
4623 if (d
>= string1
&& d
<= end1
)
4627 break; /* Matching at this starting point really fails. */
4631 goto restore_best_regs
;
4635 return -1; /* Failure to match. */
4638 /* Subroutine definitions for re_match_2. */
4641 /* We are passed P pointing to a register number after a start_memory.
4643 Return true if the pattern up to the corresponding stop_memory can
4644 match the empty string, and false otherwise.
4646 If we find the matching stop_memory, sets P to point to one past its number.
4647 Otherwise, sets P to an undefined byte less than or equal to END.
4649 We don't handle duplicates properly (yet). */
4652 group_match_null_string_p (p
, end
, reg_info
)
4653 unsigned char **p
, *end
;
4654 register_info_type
*reg_info
;
4657 /* Point to after the args to the start_memory. */
4658 unsigned char *p1
= *p
+ 2;
4662 /* Skip over opcodes that can match nothing, and return true or
4663 false, as appropriate, when we get to one that can't, or to the
4664 matching stop_memory. */
4666 switch ((re_opcode_t
) *p1
)
4668 /* Could be either a loop or a series of alternatives. */
4669 case on_failure_jump
:
4671 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
4673 /* If the next operation is not a jump backwards in the
4678 /* Go through the on_failure_jumps of the alternatives,
4679 seeing if any of the alternatives cannot match nothing.
4680 The last alternative starts with only a jump,
4681 whereas the rest start with on_failure_jump and end
4682 with a jump, e.g., here is the pattern for `a|b|c':
4684 /on_failure_jump/0/6/exactn/1/a/jump_past_alt/0/6
4685 /on_failure_jump/0/6/exactn/1/b/jump_past_alt/0/3
4688 So, we have to first go through the first (n-1)
4689 alternatives and then deal with the last one separately. */
4692 /* Deal with the first (n-1) alternatives, which start
4693 with an on_failure_jump (see above) that jumps to right
4694 past a jump_past_alt. */
4696 while ((re_opcode_t
) p1
[mcnt
-3] == jump_past_alt
)
4698 /* `mcnt' holds how many bytes long the alternative
4699 is, including the ending `jump_past_alt' and
4702 if (!alt_match_null_string_p (p1
, p1
+ mcnt
- 3,
4706 /* Move to right after this alternative, including the
4710 /* Break if it's the beginning of an n-th alternative
4711 that doesn't begin with an on_failure_jump. */
4712 if ((re_opcode_t
) *p1
!= on_failure_jump
)
4715 /* Still have to check that it's not an n-th
4716 alternative that starts with an on_failure_jump. */
4718 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
4719 if ((re_opcode_t
) p1
[mcnt
-3] != jump_past_alt
)
4721 /* Get to the beginning of the n-th alternative. */
4727 /* Deal with the last alternative: go back and get number
4728 of the `jump_past_alt' just before it. `mcnt' contains
4729 the length of the alternative. */
4730 EXTRACT_NUMBER (mcnt
, p1
- 2);
4732 if (!alt_match_null_string_p (p1
, p1
+ mcnt
, reg_info
))
4735 p1
+= mcnt
; /* Get past the n-th alternative. */
4741 assert (p1
[1] == **p
);
4747 if (!common_op_match_null_string_p (&p1
, end
, reg_info
))
4750 } /* while p1 < end */
4753 } /* group_match_null_string_p */
4756 /* Similar to group_match_null_string_p, but doesn't deal with alternatives:
4757 It expects P to be the first byte of a single alternative and END one
4758 byte past the last. The alternative can contain groups. */
4761 alt_match_null_string_p (p
, end
, reg_info
)
4762 unsigned char *p
, *end
;
4763 register_info_type
*reg_info
;
4766 unsigned char *p1
= p
;
4770 /* Skip over opcodes that can match nothing, and break when we get
4771 to one that can't. */
4773 switch ((re_opcode_t
) *p1
)
4776 case on_failure_jump
:
4778 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
4783 if (!common_op_match_null_string_p (&p1
, end
, reg_info
))
4786 } /* while p1 < end */
4789 } /* alt_match_null_string_p */
4792 /* Deals with the ops common to group_match_null_string_p and
4793 alt_match_null_string_p.
4795 Sets P to one after the op and its arguments, if any. */
4798 common_op_match_null_string_p (p
, end
, reg_info
)
4799 unsigned char **p
, *end
;
4800 register_info_type
*reg_info
;
4805 unsigned char *p1
= *p
;
4807 switch ((re_opcode_t
) *p1
++)
4827 assert (reg_no
> 0 && reg_no
<= MAX_REGNUM
);
4828 ret
= group_match_null_string_p (&p1
, end
, reg_info
);
4830 /* Have to set this here in case we're checking a group which
4831 contains a group and a back reference to it. */
4833 if (REG_MATCH_NULL_STRING_P (reg_info
[reg_no
]) == MATCH_NULL_UNSET_VALUE
)
4834 REG_MATCH_NULL_STRING_P (reg_info
[reg_no
]) = ret
;
4840 /* If this is an optimized succeed_n for zero times, make the jump. */
4842 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
4850 /* Get to the number of times to succeed. */
4852 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
4857 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
4865 if (!REG_MATCH_NULL_STRING_P (reg_info
[*p1
]))
4873 /* All other opcodes mean we cannot match the empty string. */
4879 } /* common_op_match_null_string_p */
4882 /* Return zero if TRANSLATE[S1] and TRANSLATE[S2] are identical for LEN
4883 bytes; nonzero otherwise. */
4886 bcmp_translate (s1
, s2
, len
, translate
)
4887 unsigned char *s1
, *s2
;
4891 register unsigned char *p1
= s1
, *p2
= s2
;
4894 if (translate
[*p1
++] != translate
[*p2
++]) return 1;
4900 /* Entry points for GNU code. */
4902 /* re_compile_pattern is the GNU regular expression compiler: it
4903 compiles PATTERN (of length SIZE) and puts the result in BUFP.
4904 Returns 0 if the pattern was valid, otherwise an error string.
4906 Assumes the `allocated' (and perhaps `buffer') and `translate' fields
4907 are set in BUFP on entry.
4909 We call regex_compile to do the actual compilation. */
4912 re_compile_pattern (pattern
, length
, bufp
)
4913 const char *pattern
;
4915 struct re_pattern_buffer
*bufp
;
4919 /* GNU code is written to assume at least RE_NREGS registers will be set
4920 (and at least one extra will be -1). */
4921 bufp
->regs_allocated
= REGS_UNALLOCATED
;
4923 /* And GNU code determines whether or not to get register information
4924 by passing null for the REGS argument to re_match, etc., not by
4928 /* Match anchors at newline. */
4929 bufp
->newline_anchor
= 1;
4931 ret
= regex_compile (pattern
, length
, re_syntax_options
, bufp
);
4933 return re_error_msg
[(int) ret
];
4936 /* Entry points compatible with 4.2 BSD regex library. We don't define
4937 them unless specifically requested. */
4939 #ifdef _REGEX_RE_COMP
4941 /* BSD has one and only one pattern buffer. */
4942 static struct re_pattern_buffer re_comp_buf
;
4952 if (!re_comp_buf
.buffer
)
4953 return "No previous regular expression";
4957 if (!re_comp_buf
.buffer
)
4959 re_comp_buf
.buffer
= (unsigned char *) malloc (200);
4960 if (re_comp_buf
.buffer
== NULL
)
4961 return "Memory exhausted";
4962 re_comp_buf
.allocated
= 200;
4964 re_comp_buf
.fastmap
= (char *) malloc (1 << BYTEWIDTH
);
4965 if (re_comp_buf
.fastmap
== NULL
)
4966 return "Memory exhausted";
4969 /* Since `re_exec' always passes NULL for the `regs' argument, we
4970 don't need to initialize the pattern buffer fields which affect it. */
4972 /* Match anchors at newlines. */
4973 re_comp_buf
.newline_anchor
= 1;
4975 ret
= regex_compile (s
, strlen (s
), re_syntax_options
, &re_comp_buf
);
4977 /* Yes, we're discarding `const' here. */
4978 return (char *) re_error_msg
[(int) ret
];
4986 const int len
= strlen (s
);
4988 0 <= re_search (&re_comp_buf
, s
, len
, 0, len
, (struct re_registers
*) 0);
4990 #endif /* _REGEX_RE_COMP */
4992 /* POSIX.2 functions. Don't define these for Emacs. */
4996 /* regcomp takes a regular expression as a string and compiles it.
4998 PREG is a regex_t *. We do not expect any fields to be initialized,
4999 since POSIX says we shouldn't. Thus, we set
5001 `buffer' to the compiled pattern;
5002 `used' to the length of the compiled pattern;
5003 `syntax' to RE_SYNTAX_POSIX_EXTENDED if the
5004 REG_EXTENDED bit in CFLAGS is set; otherwise, to
5005 RE_SYNTAX_POSIX_BASIC;
5006 `newline_anchor' to REG_NEWLINE being set in CFLAGS;
5007 `fastmap' and `fastmap_accurate' to zero;
5008 `re_nsub' to the number of subexpressions in PATTERN.
5010 PATTERN is the address of the pattern string.
5012 CFLAGS is a series of bits which affect compilation.
5014 If REG_EXTENDED is set, we use POSIX extended syntax; otherwise, we
5015 use POSIX basic syntax.
5017 If REG_NEWLINE is set, then . and [^...] don't match newline.
5018 Also, regexec will try a match beginning after every newline.
5020 If REG_ICASE is set, then we considers upper- and lowercase
5021 versions of letters to be equivalent when matching.
5023 If REG_NOSUB is set, then when PREG is passed to regexec, that
5024 routine will report only success or failure, and nothing about the
5027 It returns 0 if it succeeds, nonzero if it doesn't. (See regex.h for
5028 the return codes and their meanings.) */
5031 regcomp (preg
, pattern
, cflags
)
5033 const char *pattern
;
5038 = (cflags
& REG_EXTENDED
) ?
5039 RE_SYNTAX_POSIX_EXTENDED
: RE_SYNTAX_POSIX_BASIC
;
5041 /* regex_compile will allocate the space for the compiled pattern. */
5043 preg
->allocated
= 0;
5046 /* Don't bother to use a fastmap when searching. This simplifies the
5047 REG_NEWLINE case: if we used a fastmap, we'd have to put all the
5048 characters after newlines into the fastmap. This way, we just try
5052 if (cflags
& REG_ICASE
)
5056 preg
->translate
= (char *) malloc (CHAR_SET_SIZE
);
5057 if (preg
->translate
== NULL
)
5058 return (int) REG_ESPACE
;
5060 /* Map uppercase characters to corresponding lowercase ones. */
5061 for (i
= 0; i
< CHAR_SET_SIZE
; i
++)
5062 preg
->translate
[i
] = ISUPPER (i
) ? tolower (i
) : i
;
5065 preg
->translate
= NULL
;
5067 /* If REG_NEWLINE is set, newlines are treated differently. */
5068 if (cflags
& REG_NEWLINE
)
5069 { /* REG_NEWLINE implies neither . nor [^...] match newline. */
5070 syntax
&= ~RE_DOT_NEWLINE
;
5071 syntax
|= RE_HAT_LISTS_NOT_NEWLINE
;
5072 /* It also changes the matching behavior. */
5073 preg
->newline_anchor
= 1;
5076 preg
->newline_anchor
= 0;
5078 preg
->no_sub
= !!(cflags
& REG_NOSUB
);
5080 /* POSIX says a null character in the pattern terminates it, so we
5081 can use strlen here in compiling the pattern. */
5082 ret
= regex_compile (pattern
, strlen (pattern
), syntax
, preg
);
5084 /* POSIX doesn't distinguish between an unmatched open-group and an
5085 unmatched close-group: both are REG_EPAREN. */
5086 if (ret
== REG_ERPAREN
) ret
= REG_EPAREN
;
5092 /* regexec searches for a given pattern, specified by PREG, in the
5095 If NMATCH is zero or REG_NOSUB was set in the cflags argument to
5096 `regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at
5097 least NMATCH elements, and we set them to the offsets of the
5098 corresponding matched substrings.
5100 EFLAGS specifies `execution flags' which affect matching: if
5101 REG_NOTBOL is set, then ^ does not match at the beginning of the
5102 string; if REG_NOTEOL is set, then $ does not match at the end.
5104 We return 0 if we find a match and REG_NOMATCH if not. */
5107 regexec (preg
, string
, nmatch
, pmatch
, eflags
)
5108 const regex_t
*preg
;
5111 regmatch_t pmatch
[];
5115 struct re_registers regs
;
5116 regex_t private_preg
;
5117 int len
= strlen (string
);
5118 boolean want_reg_info
= !preg
->no_sub
&& nmatch
> 0;
5120 private_preg
= *preg
;
5122 private_preg
.not_bol
= !!(eflags
& REG_NOTBOL
);
5123 private_preg
.not_eol
= !!(eflags
& REG_NOTEOL
);
5125 /* The user has told us exactly how many registers to return
5126 information about, via `nmatch'. We have to pass that on to the
5127 matching routines. */
5128 private_preg
.regs_allocated
= REGS_FIXED
;
5132 regs
.num_regs
= nmatch
;
5133 regs
.start
= TALLOC (nmatch
, regoff_t
);
5134 regs
.end
= TALLOC (nmatch
, regoff_t
);
5135 if (regs
.start
== NULL
|| regs
.end
== NULL
)
5136 return (int) REG_NOMATCH
;
5139 /* Perform the searching operation. */
5140 ret
= re_search (&private_preg
, string
, len
,
5141 /* start: */ 0, /* range: */ len
,
5142 want_reg_info
? ®s
: (struct re_registers
*) 0);
5144 /* Copy the register information to the POSIX structure. */
5151 for (r
= 0; r
< nmatch
; r
++)
5153 pmatch
[r
].rm_so
= regs
.start
[r
];
5154 pmatch
[r
].rm_eo
= regs
.end
[r
];
5158 /* If we needed the temporary register info, free the space now. */
5163 /* We want zero return to mean success, unlike `re_search'. */
5164 return ret
>= 0 ? (int) REG_NOERROR
: (int) REG_NOMATCH
;
5168 /* Returns a message corresponding to an error code, ERRCODE, returned
5169 from either regcomp or regexec. We don't use PREG here. */
5172 regerror (errcode
, preg
, errbuf
, errbuf_size
)
5174 const regex_t
*preg
;
5182 || errcode
>= (sizeof (re_error_msg
) / sizeof (re_error_msg
[0])))
5183 /* Only error codes returned by the rest of the code should be passed
5184 to this routine. If we are given anything else, or if other regex
5185 code generates an invalid error code, then the program has a bug.
5186 Dump core so we can fix it. */
5189 msg
= re_error_msg
[errcode
];
5191 /* POSIX doesn't require that we do anything in this case, but why
5196 msg_size
= strlen (msg
) + 1; /* Includes the null. */
5198 if (errbuf_size
!= 0)
5200 if (msg_size
> errbuf_size
)
5202 strncpy (errbuf
, msg
, errbuf_size
- 1);
5203 errbuf
[errbuf_size
- 1] = 0;
5206 strcpy (errbuf
, msg
);
5213 /* Free dynamically allocated space used by PREG. */
5219 if (preg
->buffer
!= NULL
)
5220 free (preg
->buffer
);
5221 preg
->buffer
= NULL
;
5223 preg
->allocated
= 0;
5226 if (preg
->fastmap
!= NULL
)
5227 free (preg
->fastmap
);
5228 preg
->fastmap
= NULL
;
5229 preg
->fastmap_accurate
= 0;
5231 if (preg
->translate
!= NULL
)
5232 free (preg
->translate
);
5233 preg
->translate
= NULL
;
5236 #endif /* not emacs */
5240 make-backup-files: t
5242 trim-versions-without-asking: nil