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. */
54 /* We used to test for `BSTRING' here, but only GCC and Emacs define
55 `BSTRING', as far as I know, and neither of them use this code. */
56 #ifndef INHIBIT_STRING_HEADER
57 #if HAVE_STRING_H || STDC_HEADERS
60 #define bcmp(s1, s2, n) memcmp ((s1), (s2), (n))
63 #define bcopy(s, d, n) memcpy ((d), (s), (n))
66 #define bzero(s, n) memset ((s), 0, (n))
73 /* Define the syntax stuff for \<, \>, etc. */
75 /* This must be nonzero for the wordchar and notwordchar pattern
76 commands in re_match_2. */
83 extern char *re_syntax_table
;
85 #else /* not SYNTAX_TABLE */
87 /* How many characters in the character set. */
88 #define CHAR_SET_SIZE 256
90 static char re_syntax_table
[CHAR_SET_SIZE
];
101 bzero (re_syntax_table
, sizeof re_syntax_table
);
103 for (c
= 'a'; c
<= 'z'; c
++)
104 re_syntax_table
[c
] = Sword
;
106 for (c
= 'A'; c
<= 'Z'; c
++)
107 re_syntax_table
[c
] = Sword
;
109 for (c
= '0'; c
<= '9'; c
++)
110 re_syntax_table
[c
] = Sword
;
112 re_syntax_table
['_'] = Sword
;
117 #endif /* not SYNTAX_TABLE */
119 #define SYNTAX(c) re_syntax_table[c]
121 #endif /* not emacs */
123 /* Get the interface, including the syntax bits. */
126 /* isalpha etc. are used for the character classes. */
129 /* Jim Meyering writes:
131 "... Some ctype macros are valid only for character codes that
132 isascii says are ASCII (SGI's IRIX-4.0.5 is one such system --when
133 using /bin/cc or gcc but without giving an ansi option). So, all
134 ctype uses should be through macros like ISPRINT... If
135 STDC_HEADERS is defined, then autoconf has verified that the ctype
136 macros don't need to be guarded with references to isascii. ...
137 Defining isascii to 1 should let any compiler worth its salt
138 eliminate the && through constant folding." */
140 #if defined (STDC_HEADERS) || (!defined (isascii) && !defined (HAVE_ISASCII))
143 #define ISASCII(c) isascii(c)
147 #define ISBLANK(c) (ISASCII (c) && isblank (c))
149 #define ISBLANK(c) ((c) == ' ' || (c) == '\t')
152 #define ISGRAPH(c) (ISASCII (c) && isgraph (c))
154 #define ISGRAPH(c) (ISASCII (c) && isprint (c) && !isspace (c))
157 #define ISPRINT(c) (ISASCII (c) && isprint (c))
158 #define ISDIGIT(c) (ISASCII (c) && isdigit (c))
159 #define ISALNUM(c) (ISASCII (c) && isalnum (c))
160 #define ISALPHA(c) (ISASCII (c) && isalpha (c))
161 #define ISCNTRL(c) (ISASCII (c) && iscntrl (c))
162 #define ISLOWER(c) (ISASCII (c) && islower (c))
163 #define ISPUNCT(c) (ISASCII (c) && ispunct (c))
164 #define ISSPACE(c) (ISASCII (c) && isspace (c))
165 #define ISUPPER(c) (ISASCII (c) && isupper (c))
166 #define ISXDIGIT(c) (ISASCII (c) && isxdigit (c))
172 /* We remove any previous definition of `SIGN_EXTEND_CHAR',
173 since ours (we hope) works properly with all combinations of
174 machines, compilers, `char' and `unsigned char' argument types.
175 (Per Bothner suggested the basic approach.) */
176 #undef SIGN_EXTEND_CHAR
178 #define SIGN_EXTEND_CHAR(c) ((signed char) (c))
179 #else /* not __STDC__ */
180 /* As in Harbison and Steele. */
181 #define SIGN_EXTEND_CHAR(c) ((((unsigned char) (c)) ^ 128) - 128)
184 /* Should we use malloc or alloca? If REGEX_MALLOC is not defined, we
185 use `alloca' instead of `malloc'. This is because using malloc in
186 re_search* or re_match* could cause memory leaks when C-g is used in
187 Emacs; also, malloc is slower and causes storage fragmentation. On
188 the other hand, malloc is more portable, and easier to debug.
190 Because we sometimes use alloca, some routines have to be macros,
191 not functions -- `alloca'-allocated space disappears at the end of the
192 function it is called in. */
196 #define REGEX_ALLOCATE malloc
197 #define REGEX_REALLOCATE(source, osize, nsize) realloc (source, nsize)
199 #else /* not REGEX_MALLOC */
201 /* Emacs already defines alloca, sometimes. */
204 /* Make alloca work the best possible way. */
206 #define alloca __builtin_alloca
207 #else /* not __GNUC__ */
210 #else /* not __GNUC__ or HAVE_ALLOCA_H */
211 #ifndef _AIX /* Already did AIX, up at the top. */
213 #endif /* not _AIX */
214 #endif /* not HAVE_ALLOCA_H */
215 #endif /* not __GNUC__ */
217 #endif /* not alloca */
219 #define REGEX_ALLOCATE alloca
221 /* Assumes a `char *destination' variable. */
222 #define REGEX_REALLOCATE(source, osize, nsize) \
223 (destination = (char *) alloca (nsize), \
224 bcopy (source, destination, osize), \
227 #endif /* not REGEX_MALLOC */
230 /* True if `size1' is non-NULL and PTR is pointing anywhere inside
231 `string1' or just past its end. This works if PTR is NULL, which is
233 #define FIRST_STRING_P(ptr) \
234 (size1 && string1 <= (ptr) && (ptr) <= string1 + size1)
236 /* (Re)Allocate N items of type T using malloc, or fail. */
237 #define TALLOC(n, t) ((t *) malloc ((n) * sizeof (t)))
238 #define RETALLOC(addr, n, t) ((addr) = (t *) realloc (addr, (n) * sizeof (t)))
239 #define RETALLOC_IF(addr, n, t) \
240 if (addr) RETALLOC((addr), (n), t); else (addr) = TALLOC ((n), t)
241 #define REGEX_TALLOC(n, t) ((t *) REGEX_ALLOCATE ((n) * sizeof (t)))
243 #define BYTEWIDTH 8 /* In bits. */
245 #define STREQ(s1, s2) ((strcmp (s1, s2) == 0))
249 #define MAX(a, b) ((a) > (b) ? (a) : (b))
250 #define MIN(a, b) ((a) < (b) ? (a) : (b))
252 typedef char boolean
;
256 static int re_match_2_internal ();
258 /* These are the command codes that appear in compiled regular
259 expressions. Some opcodes are followed by argument bytes. A
260 command code can specify any interpretation whatsoever for its
261 arguments. Zero bytes may appear in the compiled regular expression. */
267 /* Followed by one byte giving n, then by n literal bytes. */
270 /* Matches any (more or less) character. */
273 /* Matches any one char belonging to specified set. First
274 following byte is number of bitmap bytes. Then come bytes
275 for a bitmap saying which chars are in. Bits in each byte
276 are ordered low-bit-first. A character is in the set if its
277 bit is 1. A character too large to have a bit in the map is
278 automatically not in the set. */
281 /* Same parameters as charset, but match any character that is
282 not one of those specified. */
285 /* Start remembering the text that is matched, for storing in a
286 register. Followed by one byte with the register number, in
287 the range 0 to one less than the pattern buffer's re_nsub
288 field. Then followed by one byte with the number of groups
289 inner to this one. (This last has to be part of the
290 start_memory only because we need it in the on_failure_jump
294 /* Stop remembering the text that is matched and store it in a
295 memory register. Followed by one byte with the register
296 number, in the range 0 to one less than `re_nsub' in the
297 pattern buffer, and one byte with the number of inner groups,
298 just like `start_memory'. (We need the number of inner
299 groups here because we don't have any easy way of finding the
300 corresponding start_memory when we're at a stop_memory.) */
303 /* Match a duplicate of something remembered. Followed by one
304 byte containing the register number. */
307 /* Fail unless at beginning of line. */
310 /* Fail unless at end of line. */
313 /* Succeeds if at beginning of buffer (if emacs) or at beginning
314 of string to be matched (if not). */
317 /* Analogously, for end of buffer/string. */
320 /* Followed by two byte relative address to which to jump. */
323 /* Same as jump, but marks the end of an alternative. */
326 /* Followed by two-byte relative address of place to resume at
327 in case of failure. */
330 /* Like on_failure_jump, but pushes a placeholder instead of the
331 current string position when executed. */
332 on_failure_keep_string_jump
,
334 /* Throw away latest failure point and then jump to following
335 two-byte relative address. */
338 /* Change to pop_failure_jump if know won't have to backtrack to
339 match; otherwise change to jump. This is used to jump
340 back to the beginning of a repeat. If what follows this jump
341 clearly won't match what the repeat does, such that we can be
342 sure that there is no use backtracking out of repetitions
343 already matched, then we change it to a pop_failure_jump.
344 Followed by two-byte address. */
347 /* Jump to following two-byte address, and push a dummy failure
348 point. This failure point will be thrown away if an attempt
349 is made to use it for a failure. A `+' construct makes this
350 before the first repeat. Also used as an intermediary kind
351 of jump when compiling an alternative. */
354 /* Push a dummy failure point and continue. Used at the end of
358 /* Followed by two-byte relative address and two-byte number n.
359 After matching N times, jump to the address upon failure. */
362 /* Followed by two-byte relative address, and two-byte number n.
363 Jump to the address N times, then fail. */
366 /* Set the following two-byte relative address to the
367 subsequent two-byte number. The address *includes* the two
371 wordchar
, /* Matches any word-constituent character. */
372 notwordchar
, /* Matches any char that is not a word-constituent. */
374 wordbeg
, /* Succeeds if at word beginning. */
375 wordend
, /* Succeeds if at word end. */
377 wordbound
, /* Succeeds if at a word boundary. */
378 notwordbound
/* Succeeds if not at a word boundary. */
381 ,before_dot
, /* Succeeds if before point. */
382 at_dot
, /* Succeeds if at point. */
383 after_dot
, /* Succeeds if after point. */
385 /* Matches any character whose syntax is specified. Followed by
386 a byte which contains a syntax code, e.g., Sword. */
389 /* Matches any character whose syntax is not that specified. */
394 /* Common operations on the compiled pattern. */
396 /* Store NUMBER in two contiguous bytes starting at DESTINATION. */
398 #define STORE_NUMBER(destination, number) \
400 (destination)[0] = (number) & 0377; \
401 (destination)[1] = (number) >> 8; \
404 /* Same as STORE_NUMBER, except increment DESTINATION to
405 the byte after where the number is stored. Therefore, DESTINATION
406 must be an lvalue. */
408 #define STORE_NUMBER_AND_INCR(destination, number) \
410 STORE_NUMBER (destination, number); \
411 (destination) += 2; \
414 /* Put into DESTINATION a number stored in two contiguous bytes starting
417 #define EXTRACT_NUMBER(destination, source) \
419 (destination) = *(source) & 0377; \
420 (destination) += SIGN_EXTEND_CHAR (*((source) + 1)) << 8; \
425 extract_number (dest
, source
)
427 unsigned char *source
;
429 int temp
= SIGN_EXTEND_CHAR (*(source
+ 1));
430 *dest
= *source
& 0377;
434 #ifndef EXTRACT_MACROS /* To debug the macros. */
435 #undef EXTRACT_NUMBER
436 #define EXTRACT_NUMBER(dest, src) extract_number (&dest, src)
437 #endif /* not EXTRACT_MACROS */
441 /* Same as EXTRACT_NUMBER, except increment SOURCE to after the number.
442 SOURCE must be an lvalue. */
444 #define EXTRACT_NUMBER_AND_INCR(destination, source) \
446 EXTRACT_NUMBER (destination, source); \
452 extract_number_and_incr (destination
, source
)
454 unsigned char **source
;
456 extract_number (destination
, *source
);
460 #ifndef EXTRACT_MACROS
461 #undef EXTRACT_NUMBER_AND_INCR
462 #define EXTRACT_NUMBER_AND_INCR(dest, src) \
463 extract_number_and_incr (&dest, &src)
464 #endif /* not EXTRACT_MACROS */
468 /* If DEBUG is defined, Regex prints many voluminous messages about what
469 it is doing (if the variable `debug' is nonzero). If linked with the
470 main program in `iregex.c', you can enter patterns and strings
471 interactively. And if linked with the main program in `main.c' and
472 the other test files, you can run the already-written tests. */
476 /* We use standard I/O for debugging. */
479 /* It is useful to test things that ``must'' be true when debugging. */
482 static int debug
= 0;
484 #define DEBUG_STATEMENT(e) e
485 #define DEBUG_PRINT1(x) if (debug) printf (x)
486 #define DEBUG_PRINT2(x1, x2) if (debug) printf (x1, x2)
487 #define DEBUG_PRINT3(x1, x2, x3) if (debug) printf (x1, x2, x3)
488 #define DEBUG_PRINT4(x1, x2, x3, x4) if (debug) printf (x1, x2, x3, x4)
489 #define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) \
490 if (debug) print_partial_compiled_pattern (s, e)
491 #define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) \
492 if (debug) print_double_string (w, s1, sz1, s2, sz2)
495 extern void printchar ();
497 /* Print the fastmap in human-readable form. */
500 print_fastmap (fastmap
)
503 unsigned was_a_range
= 0;
506 while (i
< (1 << BYTEWIDTH
))
512 while (i
< (1 << BYTEWIDTH
) && fastmap
[i
])
528 /* Print a compiled pattern string in human-readable form, starting at
529 the START pointer into it and ending just before the pointer END. */
532 print_partial_compiled_pattern (start
, end
)
533 unsigned char *start
;
537 unsigned char *p
= start
;
538 unsigned char *pend
= end
;
546 /* Loop over pattern commands. */
549 printf ("%d:\t", p
- start
);
551 switch ((re_opcode_t
) *p
++)
559 printf ("/exactn/%d", mcnt
);
570 printf ("/start_memory/%d/%d", mcnt
, *p
++);
575 printf ("/stop_memory/%d/%d", mcnt
, *p
++);
579 printf ("/duplicate/%d", *p
++);
589 register int c
, last
= -100;
590 register int in_range
= 0;
592 printf ("/charset [%s",
593 (re_opcode_t
) *(p
- 1) == charset_not
? "^" : "");
595 assert (p
+ *p
< pend
);
597 for (c
= 0; c
< 256; c
++)
599 && (p
[1 + (c
/8)] & (1 << (c
% 8))))
601 /* Are we starting a range? */
602 if (last
+ 1 == c
&& ! in_range
)
607 /* Have we broken a range? */
608 else if (last
+ 1 != c
&& in_range
)
637 case on_failure_jump
:
638 extract_number_and_incr (&mcnt
, &p
);
639 printf ("/on_failure_jump to %d", p
+ mcnt
- start
);
642 case on_failure_keep_string_jump
:
643 extract_number_and_incr (&mcnt
, &p
);
644 printf ("/on_failure_keep_string_jump to %d", p
+ mcnt
- start
);
647 case dummy_failure_jump
:
648 extract_number_and_incr (&mcnt
, &p
);
649 printf ("/dummy_failure_jump to %d", p
+ mcnt
- start
);
652 case push_dummy_failure
:
653 printf ("/push_dummy_failure");
657 extract_number_and_incr (&mcnt
, &p
);
658 printf ("/maybe_pop_jump to %d", p
+ mcnt
- start
);
661 case pop_failure_jump
:
662 extract_number_and_incr (&mcnt
, &p
);
663 printf ("/pop_failure_jump to %d", p
+ mcnt
- start
);
667 extract_number_and_incr (&mcnt
, &p
);
668 printf ("/jump_past_alt to %d", p
+ mcnt
- start
);
672 extract_number_and_incr (&mcnt
, &p
);
673 printf ("/jump to %d", p
+ mcnt
- start
);
677 extract_number_and_incr (&mcnt
, &p
);
678 extract_number_and_incr (&mcnt2
, &p
);
679 printf ("/succeed_n to %d, %d times", p
+ mcnt
- start
, mcnt2
);
683 extract_number_and_incr (&mcnt
, &p
);
684 extract_number_and_incr (&mcnt2
, &p
);
685 printf ("/jump_n to %d, %d times", p
+ mcnt
- start
, mcnt2
);
689 extract_number_and_incr (&mcnt
, &p
);
690 extract_number_and_incr (&mcnt2
, &p
);
691 printf ("/set_number_at location %d to %d", p
+ mcnt
- start
, mcnt2
);
695 printf ("/wordbound");
699 printf ("/notwordbound");
711 printf ("/before_dot");
719 printf ("/after_dot");
723 printf ("/syntaxspec");
725 printf ("/%d", mcnt
);
729 printf ("/notsyntaxspec");
731 printf ("/%d", mcnt
);
736 printf ("/wordchar");
740 printf ("/notwordchar");
752 printf ("?%d", *(p
-1));
758 printf ("%d:\tend of pattern.\n", p
- start
);
763 print_compiled_pattern (bufp
)
764 struct re_pattern_buffer
*bufp
;
766 unsigned char *buffer
= bufp
->buffer
;
768 print_partial_compiled_pattern (buffer
, buffer
+ bufp
->used
);
769 printf ("%d bytes used/%d bytes allocated.\n", bufp
->used
, bufp
->allocated
);
771 if (bufp
->fastmap_accurate
&& bufp
->fastmap
)
773 printf ("fastmap: ");
774 print_fastmap (bufp
->fastmap
);
777 printf ("re_nsub: %d\t", bufp
->re_nsub
);
778 printf ("regs_alloc: %d\t", bufp
->regs_allocated
);
779 printf ("can_be_null: %d\t", bufp
->can_be_null
);
780 printf ("newline_anchor: %d\n", bufp
->newline_anchor
);
781 printf ("no_sub: %d\t", bufp
->no_sub
);
782 printf ("not_bol: %d\t", bufp
->not_bol
);
783 printf ("not_eol: %d\t", bufp
->not_eol
);
784 printf ("syntax: %d\n", bufp
->syntax
);
785 /* Perhaps we should print the translate table? */
790 print_double_string (where
, string1
, size1
, string2
, size2
)
803 if (FIRST_STRING_P (where
))
805 for (this_char
= where
- string1
; this_char
< size1
; this_char
++)
806 printchar (string1
[this_char
]);
811 for (this_char
= where
- string2
; this_char
< size2
; this_char
++)
812 printchar (string2
[this_char
]);
816 #else /* not DEBUG */
821 #define DEBUG_STATEMENT(e)
822 #define DEBUG_PRINT1(x)
823 #define DEBUG_PRINT2(x1, x2)
824 #define DEBUG_PRINT3(x1, x2, x3)
825 #define DEBUG_PRINT4(x1, x2, x3, x4)
826 #define DEBUG_PRINT_COMPILED_PATTERN(p, s, e)
827 #define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2)
829 #endif /* not DEBUG */
831 /* Set by `re_set_syntax' to the current regexp syntax to recognize. Can
832 also be assigned to arbitrarily: each pattern buffer stores its own
833 syntax, so it can be changed between regex compilations. */
834 reg_syntax_t re_syntax_options
= RE_SYNTAX_EMACS
;
837 /* Specify the precise syntax of regexps for compilation. This provides
838 for compatibility for various utilities which historically have
839 different, incompatible syntaxes.
841 The argument SYNTAX is a bit mask comprised of the various bits
842 defined in regex.h. We return the old syntax. */
845 re_set_syntax (syntax
)
848 reg_syntax_t ret
= re_syntax_options
;
850 re_syntax_options
= syntax
;
854 /* This table gives an error message for each of the error codes listed
855 in regex.h. Obviously the order here has to be same as there. */
857 static const char *re_error_msg
[] =
858 { NULL
, /* REG_NOERROR */
859 "No match", /* REG_NOMATCH */
860 "Invalid regular expression", /* REG_BADPAT */
861 "Invalid collation character", /* REG_ECOLLATE */
862 "Invalid character class name", /* REG_ECTYPE */
863 "Trailing backslash", /* REG_EESCAPE */
864 "Invalid back reference", /* REG_ESUBREG */
865 "Unmatched [ or [^", /* REG_EBRACK */
866 "Unmatched ( or \\(", /* REG_EPAREN */
867 "Unmatched \\{", /* REG_EBRACE */
868 "Invalid content of \\{\\}", /* REG_BADBR */
869 "Invalid range end", /* REG_ERANGE */
870 "Memory exhausted", /* REG_ESPACE */
871 "Invalid preceding regular expression", /* REG_BADRPT */
872 "Premature end of regular expression", /* REG_EEND */
873 "Regular expression too big", /* REG_ESIZE */
874 "Unmatched ) or \\)", /* REG_ERPAREN */
877 /* Avoiding alloca during matching, to placate r_alloc. */
879 /* Define MATCH_MAY_ALLOCATE unless we need to make sure that the
880 searching and matching functions should not call alloca. On some
881 systems, alloca is implemented in terms of malloc, and if we're
882 using the relocating allocator routines, then malloc could cause a
883 relocation, which might (if the strings being searched are in the
884 ralloc heap) shift the data out from underneath the regexp
887 Here's another reason to avoid allocation: Emacs
888 processes input from X in a signal handler; processing X input may
889 call malloc; if input arrives while a matching routine is calling
890 malloc, then we're scrod. But Emacs can't just block input while
891 calling matching routines; then we don't notice interrupts when
892 they come in. So, Emacs blocks input around all regexp calls
893 except the matching calls, which it leaves unprotected, in the
894 faith that they will not malloc. */
896 /* Normally, this is fine. */
897 #define MATCH_MAY_ALLOCATE
899 /* The match routines may not allocate if (1) they would do it with malloc
900 and (2) it's not safe for them to use malloc. */
901 #if (defined (C_ALLOCA) || defined (REGEX_MALLOC)) && (defined (emacs) || defined (REL_ALLOC))
902 #undef MATCH_MAY_ALLOCATE
906 /* Failure stack declarations and macros; both re_compile_fastmap and
907 re_match_2 use a failure stack. These have to be macros because of
911 /* Number of failure points for which to initially allocate space
912 when matching. If this number is exceeded, we allocate more
913 space, so it is not a hard limit. */
914 #ifndef INIT_FAILURE_ALLOC
915 #define INIT_FAILURE_ALLOC 5
918 /* Roughly the maximum number of failure points on the stack. Would be
919 exactly that if always used MAX_FAILURE_SPACE each time we failed.
920 This is a variable only so users of regex can assign to it; we never
921 change it ourselves. */
922 int re_max_failures
= 2000;
924 typedef unsigned char *fail_stack_elt_t
;
928 fail_stack_elt_t
*stack
;
930 unsigned avail
; /* Offset of next open position. */
933 #define FAIL_STACK_EMPTY() (fail_stack.avail == 0)
934 #define FAIL_STACK_PTR_EMPTY() (fail_stack_ptr->avail == 0)
935 #define FAIL_STACK_FULL() (fail_stack.avail == fail_stack.size)
936 #define FAIL_STACK_TOP() (fail_stack.stack[fail_stack.avail])
939 /* Initialize `fail_stack'. Do `return -2' if the alloc fails. */
941 #ifdef MATCH_MAY_ALLOCATE
942 #define INIT_FAIL_STACK() \
944 fail_stack.stack = (fail_stack_elt_t *) \
945 REGEX_ALLOCATE (INIT_FAILURE_ALLOC * sizeof (fail_stack_elt_t)); \
947 if (fail_stack.stack == NULL) \
950 fail_stack.size = INIT_FAILURE_ALLOC; \
951 fail_stack.avail = 0; \
954 #define INIT_FAIL_STACK() \
956 fail_stack.avail = 0; \
961 /* Double the size of FAIL_STACK, up to approximately `re_max_failures' items.
963 Return 1 if succeeds, and 0 if either ran out of memory
964 allocating space for it or it was already too large.
966 REGEX_REALLOCATE requires `destination' be declared. */
968 #define DOUBLE_FAIL_STACK(fail_stack) \
969 ((fail_stack).size > re_max_failures * MAX_FAILURE_ITEMS \
971 : ((fail_stack).stack = (fail_stack_elt_t *) \
972 REGEX_REALLOCATE ((fail_stack).stack, \
973 (fail_stack).size * sizeof (fail_stack_elt_t), \
974 ((fail_stack).size << 1) * sizeof (fail_stack_elt_t)), \
976 (fail_stack).stack == NULL \
978 : ((fail_stack).size <<= 1, \
982 /* Push PATTERN_OP on FAIL_STACK.
984 Return 1 if was able to do so and 0 if ran out of memory allocating
986 #define PUSH_PATTERN_OP(pattern_op, fail_stack) \
987 ((FAIL_STACK_FULL () \
988 && !DOUBLE_FAIL_STACK (fail_stack)) \
990 : ((fail_stack).stack[(fail_stack).avail++] = pattern_op, \
993 /* This pushes an item onto the failure stack. Must be a four-byte
994 value. Assumes the variable `fail_stack'. Probably should only
995 be called from within `PUSH_FAILURE_POINT'. */
996 #define PUSH_FAILURE_ITEM(item) \
997 fail_stack.stack[fail_stack.avail++] = (fail_stack_elt_t) item
999 /* The complement operation. Assumes `fail_stack' is nonempty. */
1000 #define POP_FAILURE_ITEM() fail_stack.stack[--fail_stack.avail]
1002 /* Used to omit pushing failure point id's when we're not debugging. */
1004 #define DEBUG_PUSH PUSH_FAILURE_ITEM
1005 #define DEBUG_POP(item_addr) *(item_addr) = POP_FAILURE_ITEM ()
1007 #define DEBUG_PUSH(item)
1008 #define DEBUG_POP(item_addr)
1012 /* Push the information about the state we will need
1013 if we ever fail back to it.
1015 Requires variables fail_stack, regstart, regend, reg_info, and
1016 num_regs be declared. DOUBLE_FAIL_STACK requires `destination' be
1019 Does `return FAILURE_CODE' if runs out of memory. */
1021 #define PUSH_FAILURE_POINT(pattern_place, string_place, failure_code) \
1023 char *destination; \
1024 /* Must be int, so when we don't save any registers, the arithmetic \
1025 of 0 + -1 isn't done as unsigned. */ \
1028 DEBUG_STATEMENT (failure_id++); \
1029 DEBUG_STATEMENT (nfailure_points_pushed++); \
1030 DEBUG_PRINT2 ("\nPUSH_FAILURE_POINT #%u:\n", failure_id); \
1031 DEBUG_PRINT2 (" Before push, next avail: %d\n", (fail_stack).avail);\
1032 DEBUG_PRINT2 (" size: %d\n", (fail_stack).size);\
1034 DEBUG_PRINT2 (" slots needed: %d\n", NUM_FAILURE_ITEMS); \
1035 DEBUG_PRINT2 (" available: %d\n", REMAINING_AVAIL_SLOTS); \
1037 /* Ensure we have enough space allocated for what we will push. */ \
1038 while (REMAINING_AVAIL_SLOTS < NUM_FAILURE_ITEMS) \
1040 if (!DOUBLE_FAIL_STACK (fail_stack)) \
1041 return failure_code; \
1043 DEBUG_PRINT2 ("\n Doubled stack; size now: %d\n", \
1044 (fail_stack).size); \
1045 DEBUG_PRINT2 (" slots available: %d\n", REMAINING_AVAIL_SLOTS);\
1048 /* Push the info, starting with the registers. */ \
1049 DEBUG_PRINT1 ("\n"); \
1051 for (this_reg = lowest_active_reg; this_reg <= highest_active_reg; \
1054 DEBUG_PRINT2 (" Pushing reg: %d\n", this_reg); \
1055 DEBUG_STATEMENT (num_regs_pushed++); \
1057 DEBUG_PRINT2 (" start: 0x%x\n", regstart[this_reg]); \
1058 PUSH_FAILURE_ITEM (regstart[this_reg]); \
1060 DEBUG_PRINT2 (" end: 0x%x\n", regend[this_reg]); \
1061 PUSH_FAILURE_ITEM (regend[this_reg]); \
1063 DEBUG_PRINT2 (" info: 0x%x\n ", reg_info[this_reg]); \
1064 DEBUG_PRINT2 (" match_null=%d", \
1065 REG_MATCH_NULL_STRING_P (reg_info[this_reg])); \
1066 DEBUG_PRINT2 (" active=%d", IS_ACTIVE (reg_info[this_reg])); \
1067 DEBUG_PRINT2 (" matched_something=%d", \
1068 MATCHED_SOMETHING (reg_info[this_reg])); \
1069 DEBUG_PRINT2 (" ever_matched=%d", \
1070 EVER_MATCHED_SOMETHING (reg_info[this_reg])); \
1071 DEBUG_PRINT1 ("\n"); \
1072 PUSH_FAILURE_ITEM (reg_info[this_reg].word); \
1075 DEBUG_PRINT2 (" Pushing low active reg: %d\n", lowest_active_reg);\
1076 PUSH_FAILURE_ITEM (lowest_active_reg); \
1078 DEBUG_PRINT2 (" Pushing high active reg: %d\n", highest_active_reg);\
1079 PUSH_FAILURE_ITEM (highest_active_reg); \
1081 DEBUG_PRINT2 (" Pushing pattern 0x%x: ", pattern_place); \
1082 DEBUG_PRINT_COMPILED_PATTERN (bufp, pattern_place, pend); \
1083 PUSH_FAILURE_ITEM (pattern_place); \
1085 DEBUG_PRINT2 (" Pushing string 0x%x: `", string_place); \
1086 DEBUG_PRINT_DOUBLE_STRING (string_place, string1, size1, string2, \
1088 DEBUG_PRINT1 ("'\n"); \
1089 PUSH_FAILURE_ITEM (string_place); \
1091 DEBUG_PRINT2 (" Pushing failure id: %u\n", failure_id); \
1092 DEBUG_PUSH (failure_id); \
1095 /* This is the number of items that are pushed and popped on the stack
1096 for each register. */
1097 #define NUM_REG_ITEMS 3
1099 /* Individual items aside from the registers. */
1101 #define NUM_NONREG_ITEMS 5 /* Includes failure point id. */
1103 #define NUM_NONREG_ITEMS 4
1106 /* We push at most this many items on the stack. */
1107 #define MAX_FAILURE_ITEMS ((num_regs - 1) * NUM_REG_ITEMS + NUM_NONREG_ITEMS)
1109 /* We actually push this many items. */
1110 #define NUM_FAILURE_ITEMS \
1111 ((highest_active_reg - lowest_active_reg + 1) * NUM_REG_ITEMS \
1114 /* How many items can still be added to the stack without overflowing it. */
1115 #define REMAINING_AVAIL_SLOTS ((fail_stack).size - (fail_stack).avail)
1118 /* Pops what PUSH_FAIL_STACK pushes.
1120 We restore into the parameters, all of which should be lvalues:
1121 STR -- the saved data position.
1122 PAT -- the saved pattern position.
1123 LOW_REG, HIGH_REG -- the highest and lowest active registers.
1124 REGSTART, REGEND -- arrays of string positions.
1125 REG_INFO -- array of information about each subexpression.
1127 Also assumes the variables `fail_stack' and (if debugging), `bufp',
1128 `pend', `string1', `size1', `string2', and `size2'. */
1130 #define POP_FAILURE_POINT(str, pat, low_reg, high_reg, regstart, regend, reg_info)\
1132 DEBUG_STATEMENT (fail_stack_elt_t failure_id;) \
1134 const unsigned char *string_temp; \
1136 assert (!FAIL_STACK_EMPTY ()); \
1138 /* Remove failure points and point to how many regs pushed. */ \
1139 DEBUG_PRINT1 ("POP_FAILURE_POINT:\n"); \
1140 DEBUG_PRINT2 (" Before pop, next avail: %d\n", fail_stack.avail); \
1141 DEBUG_PRINT2 (" size: %d\n", fail_stack.size); \
1143 assert (fail_stack.avail >= NUM_NONREG_ITEMS); \
1145 DEBUG_POP (&failure_id); \
1146 DEBUG_PRINT2 (" Popping failure id: %u\n", failure_id); \
1148 /* If the saved string location is NULL, it came from an \
1149 on_failure_keep_string_jump opcode, and we want to throw away the \
1150 saved NULL, thus retaining our current position in the string. */ \
1151 string_temp = POP_FAILURE_ITEM (); \
1152 if (string_temp != NULL) \
1153 str = (const char *) string_temp; \
1155 DEBUG_PRINT2 (" Popping string 0x%x: `", str); \
1156 DEBUG_PRINT_DOUBLE_STRING (str, string1, size1, string2, size2); \
1157 DEBUG_PRINT1 ("'\n"); \
1159 pat = (unsigned char *) POP_FAILURE_ITEM (); \
1160 DEBUG_PRINT2 (" Popping pattern 0x%x: ", pat); \
1161 DEBUG_PRINT_COMPILED_PATTERN (bufp, pat, pend); \
1163 /* Restore register info. */ \
1164 high_reg = (unsigned) POP_FAILURE_ITEM (); \
1165 DEBUG_PRINT2 (" Popping high active reg: %d\n", high_reg); \
1167 low_reg = (unsigned) POP_FAILURE_ITEM (); \
1168 DEBUG_PRINT2 (" Popping low active reg: %d\n", low_reg); \
1170 for (this_reg = high_reg; this_reg >= low_reg; this_reg--) \
1172 DEBUG_PRINT2 (" Popping reg: %d\n", this_reg); \
1174 reg_info[this_reg].word = POP_FAILURE_ITEM (); \
1175 DEBUG_PRINT2 (" info: 0x%x\n", reg_info[this_reg]); \
1177 regend[this_reg] = (const char *) POP_FAILURE_ITEM (); \
1178 DEBUG_PRINT2 (" end: 0x%x\n", regend[this_reg]); \
1180 regstart[this_reg] = (const char *) POP_FAILURE_ITEM (); \
1181 DEBUG_PRINT2 (" start: 0x%x\n", regstart[this_reg]); \
1184 DEBUG_STATEMENT (nfailure_points_popped++); \
1185 } /* POP_FAILURE_POINT */
1189 /* Structure for per-register (a.k.a. per-group) information.
1190 This must not be longer than one word, because we push this value
1191 onto the failure stack. Other register information, such as the
1192 starting and ending positions (which are addresses), and the list of
1193 inner groups (which is a bits list) are maintained in separate
1196 We are making a (strictly speaking) nonportable assumption here: that
1197 the compiler will pack our bit fields into something that fits into
1198 the type of `word', i.e., is something that fits into one item on the
1202 fail_stack_elt_t word
;
1205 /* This field is one if this group can match the empty string,
1206 zero if not. If not yet determined, `MATCH_NULL_UNSET_VALUE'. */
1207 #define MATCH_NULL_UNSET_VALUE 3
1208 unsigned match_null_string_p
: 2;
1209 unsigned is_active
: 1;
1210 unsigned matched_something
: 1;
1211 unsigned ever_matched_something
: 1;
1213 } register_info_type
;
1215 #define REG_MATCH_NULL_STRING_P(R) ((R).bits.match_null_string_p)
1216 #define IS_ACTIVE(R) ((R).bits.is_active)
1217 #define MATCHED_SOMETHING(R) ((R).bits.matched_something)
1218 #define EVER_MATCHED_SOMETHING(R) ((R).bits.ever_matched_something)
1221 /* Call this when have matched a real character; it sets `matched' flags
1222 for the subexpressions which we are currently inside. Also records
1223 that those subexprs have matched. */
1224 #define SET_REGS_MATCHED() \
1228 for (r = lowest_active_reg; r <= highest_active_reg; r++) \
1230 MATCHED_SOMETHING (reg_info[r]) \
1231 = EVER_MATCHED_SOMETHING (reg_info[r]) \
1238 /* Registers are set to a sentinel when they haven't yet matched. */
1239 #define REG_UNSET_VALUE ((char *) -1)
1240 #define REG_UNSET(e) ((e) == REG_UNSET_VALUE)
1244 /* How do we implement a missing MATCH_MAY_ALLOCATE?
1245 We make the fail stack a global thing, and then grow it to
1246 re_max_failures when we compile. */
1247 #ifndef MATCH_MAY_ALLOCATE
1248 static fail_stack_type fail_stack
;
1250 static const char ** regstart
, ** regend
;
1251 static const char ** old_regstart
, ** old_regend
;
1252 static const char **best_regstart
, **best_regend
;
1253 static register_info_type
*reg_info
;
1254 static const char **reg_dummy
;
1255 static register_info_type
*reg_info_dummy
;
1259 /* Subroutine declarations and macros for regex_compile. */
1261 static void store_op1 (), store_op2 ();
1262 static void insert_op1 (), insert_op2 ();
1263 static boolean
at_begline_loc_p (), at_endline_loc_p ();
1264 static boolean
group_in_compile_stack ();
1265 static reg_errcode_t
compile_range ();
1267 /* Fetch the next character in the uncompiled pattern---translating it
1268 if necessary. Also cast from a signed character in the constant
1269 string passed to us by the user to an unsigned char that we can use
1270 as an array index (in, e.g., `translate'). */
1271 #define PATFETCH(c) \
1272 do {if (p == pend) return REG_EEND; \
1273 c = (unsigned char) *p++; \
1274 if (translate) c = translate[c]; \
1277 /* Fetch the next character in the uncompiled pattern, with no
1279 #define PATFETCH_RAW(c) \
1280 do {if (p == pend) return REG_EEND; \
1281 c = (unsigned char) *p++; \
1284 /* Go backwards one character in the pattern. */
1285 #define PATUNFETCH p--
1288 /* If `translate' is non-null, return translate[D], else just D. We
1289 cast the subscript to translate because some data is declared as
1290 `char *', to avoid warnings when a string constant is passed. But
1291 when we use a character as a subscript we must make it unsigned. */
1292 #define TRANSLATE(d) (translate ? translate[(unsigned char) (d)] : (d))
1295 /* Macros for outputting the compiled pattern into `buffer'. */
1297 /* If the buffer isn't allocated when it comes in, use this. */
1298 #define INIT_BUF_SIZE 32
1300 /* Make sure we have at least N more bytes of space in buffer. */
1301 #define GET_BUFFER_SPACE(n) \
1302 while (b - bufp->buffer + (n) > bufp->allocated) \
1305 /* Make sure we have one more byte of buffer space and then add C to it. */
1306 #define BUF_PUSH(c) \
1308 GET_BUFFER_SPACE (1); \
1309 *b++ = (unsigned char) (c); \
1313 /* Ensure we have two more bytes of buffer space and then append C1 and C2. */
1314 #define BUF_PUSH_2(c1, c2) \
1316 GET_BUFFER_SPACE (2); \
1317 *b++ = (unsigned char) (c1); \
1318 *b++ = (unsigned char) (c2); \
1322 /* As with BUF_PUSH_2, except for three bytes. */
1323 #define BUF_PUSH_3(c1, c2, c3) \
1325 GET_BUFFER_SPACE (3); \
1326 *b++ = (unsigned char) (c1); \
1327 *b++ = (unsigned char) (c2); \
1328 *b++ = (unsigned char) (c3); \
1332 /* Store a jump with opcode OP at LOC to location TO. We store a
1333 relative address offset by the three bytes the jump itself occupies. */
1334 #define STORE_JUMP(op, loc, to) \
1335 store_op1 (op, loc, (to) - (loc) - 3)
1337 /* Likewise, for a two-argument jump. */
1338 #define STORE_JUMP2(op, loc, to, arg) \
1339 store_op2 (op, loc, (to) - (loc) - 3, arg)
1341 /* Like `STORE_JUMP', but for inserting. Assume `b' is the buffer end. */
1342 #define INSERT_JUMP(op, loc, to) \
1343 insert_op1 (op, loc, (to) - (loc) - 3, b)
1345 /* Like `STORE_JUMP2', but for inserting. Assume `b' is the buffer end. */
1346 #define INSERT_JUMP2(op, loc, to, arg) \
1347 insert_op2 (op, loc, (to) - (loc) - 3, arg, b)
1350 /* This is not an arbitrary limit: the arguments which represent offsets
1351 into the pattern are two bytes long. So if 2^16 bytes turns out to
1352 be too small, many things would have to change. */
1353 #define MAX_BUF_SIZE (1L << 16)
1356 /* Extend the buffer by twice its current size via realloc and
1357 reset the pointers that pointed into the old block to point to the
1358 correct places in the new one. If extending the buffer results in it
1359 being larger than MAX_BUF_SIZE, then flag memory exhausted. */
1360 #define EXTEND_BUFFER() \
1362 unsigned char *old_buffer = bufp->buffer; \
1363 if (bufp->allocated == MAX_BUF_SIZE) \
1365 bufp->allocated <<= 1; \
1366 if (bufp->allocated > MAX_BUF_SIZE) \
1367 bufp->allocated = MAX_BUF_SIZE; \
1368 bufp->buffer = (unsigned char *) realloc (bufp->buffer, bufp->allocated);\
1369 if (bufp->buffer == NULL) \
1370 return REG_ESPACE; \
1371 /* If the buffer moved, move all the pointers into it. */ \
1372 if (old_buffer != bufp->buffer) \
1374 b = (b - old_buffer) + bufp->buffer; \
1375 begalt = (begalt - old_buffer) + bufp->buffer; \
1376 if (fixup_alt_jump) \
1377 fixup_alt_jump = (fixup_alt_jump - old_buffer) + bufp->buffer;\
1379 laststart = (laststart - old_buffer) + bufp->buffer; \
1380 if (pending_exact) \
1381 pending_exact = (pending_exact - old_buffer) + bufp->buffer; \
1386 /* Since we have one byte reserved for the register number argument to
1387 {start,stop}_memory, the maximum number of groups we can report
1388 things about is what fits in that byte. */
1389 #define MAX_REGNUM 255
1391 /* But patterns can have more than `MAX_REGNUM' registers. We just
1392 ignore the excess. */
1393 typedef unsigned regnum_t
;
1396 /* Macros for the compile stack. */
1398 /* Since offsets can go either forwards or backwards, this type needs to
1399 be able to hold values from -(MAX_BUF_SIZE - 1) to MAX_BUF_SIZE - 1. */
1400 typedef int pattern_offset_t
;
1404 pattern_offset_t begalt_offset
;
1405 pattern_offset_t fixup_alt_jump
;
1406 pattern_offset_t inner_group_offset
;
1407 pattern_offset_t laststart_offset
;
1409 } compile_stack_elt_t
;
1414 compile_stack_elt_t
*stack
;
1416 unsigned avail
; /* Offset of next open position. */
1417 } compile_stack_type
;
1420 #define INIT_COMPILE_STACK_SIZE 32
1422 #define COMPILE_STACK_EMPTY (compile_stack.avail == 0)
1423 #define COMPILE_STACK_FULL (compile_stack.avail == compile_stack.size)
1425 /* The next available element. */
1426 #define COMPILE_STACK_TOP (compile_stack.stack[compile_stack.avail])
1429 /* Set the bit for character C in a list. */
1430 #define SET_LIST_BIT(c) \
1431 (b[((unsigned char) (c)) / BYTEWIDTH] \
1432 |= 1 << (((unsigned char) c) % BYTEWIDTH))
1435 /* Get the next unsigned number in the uncompiled pattern. */
1436 #define GET_UNSIGNED_NUMBER(num) \
1440 while (ISDIGIT (c)) \
1444 num = num * 10 + c - '0'; \
1452 #define CHAR_CLASS_MAX_LENGTH 6 /* Namely, `xdigit'. */
1454 #define IS_CHAR_CLASS(string) \
1455 (STREQ (string, "alpha") || STREQ (string, "upper") \
1456 || STREQ (string, "lower") || STREQ (string, "digit") \
1457 || STREQ (string, "alnum") || STREQ (string, "xdigit") \
1458 || STREQ (string, "space") || STREQ (string, "print") \
1459 || STREQ (string, "punct") || STREQ (string, "graph") \
1460 || STREQ (string, "cntrl") || STREQ (string, "blank"))
1462 /* `regex_compile' compiles PATTERN (of length SIZE) according to SYNTAX.
1463 Returns one of error codes defined in `regex.h', or zero for success.
1465 Assumes the `allocated' (and perhaps `buffer') and `translate'
1466 fields are set in BUFP on entry.
1468 If it succeeds, results are put in BUFP (if it returns an error, the
1469 contents of BUFP are undefined):
1470 `buffer' is the compiled pattern;
1471 `syntax' is set to SYNTAX;
1472 `used' is set to the length of the compiled pattern;
1473 `fastmap_accurate' is zero;
1474 `re_nsub' is the number of subexpressions in PATTERN;
1475 `not_bol' and `not_eol' are zero;
1477 The `fastmap' and `newline_anchor' fields are neither
1478 examined nor set. */
1480 /* Return, freeing storage we allocated. */
1481 #define FREE_STACK_RETURN(value) \
1482 return (free (compile_stack.stack), value)
1484 static reg_errcode_t
1485 regex_compile (pattern
, size
, syntax
, bufp
)
1486 const char *pattern
;
1488 reg_syntax_t syntax
;
1489 struct re_pattern_buffer
*bufp
;
1491 /* We fetch characters from PATTERN here. Even though PATTERN is
1492 `char *' (i.e., signed), we declare these variables as unsigned, so
1493 they can be reliably used as array indices. */
1494 register unsigned char c
, c1
;
1496 /* A random temporary spot in PATTERN. */
1499 /* Points to the end of the buffer, where we should append. */
1500 register unsigned char *b
;
1502 /* Keeps track of unclosed groups. */
1503 compile_stack_type compile_stack
;
1505 /* Points to the current (ending) position in the pattern. */
1506 const char *p
= pattern
;
1507 const char *pend
= pattern
+ size
;
1509 /* How to translate the characters in the pattern. */
1510 char *translate
= bufp
->translate
;
1512 /* Address of the count-byte of the most recently inserted `exactn'
1513 command. This makes it possible to tell if a new exact-match
1514 character can be added to that command or if the character requires
1515 a new `exactn' command. */
1516 unsigned char *pending_exact
= 0;
1518 /* Address of start of the most recently finished expression.
1519 This tells, e.g., postfix * where to find the start of its
1520 operand. Reset at the beginning of groups and alternatives. */
1521 unsigned char *laststart
= 0;
1523 /* Address of beginning of regexp, or inside of last group. */
1524 unsigned char *begalt
;
1526 /* Place in the uncompiled pattern (i.e., the {) to
1527 which to go back if the interval is invalid. */
1528 const char *beg_interval
;
1530 /* Address of the place where a forward jump should go to the end of
1531 the containing expression. Each alternative of an `or' -- except the
1532 last -- ends with a forward jump of this sort. */
1533 unsigned char *fixup_alt_jump
= 0;
1535 /* Counts open-groups as they are encountered. Remembered for the
1536 matching close-group on the compile stack, so the same register
1537 number is put in the stop_memory as the start_memory. */
1538 regnum_t regnum
= 0;
1541 DEBUG_PRINT1 ("\nCompiling pattern: ");
1544 unsigned debug_count
;
1546 for (debug_count
= 0; debug_count
< size
; debug_count
++)
1547 printchar (pattern
[debug_count
]);
1552 /* Initialize the compile stack. */
1553 compile_stack
.stack
= TALLOC (INIT_COMPILE_STACK_SIZE
, compile_stack_elt_t
);
1554 if (compile_stack
.stack
== NULL
)
1557 compile_stack
.size
= INIT_COMPILE_STACK_SIZE
;
1558 compile_stack
.avail
= 0;
1560 /* Initialize the pattern buffer. */
1561 bufp
->syntax
= syntax
;
1562 bufp
->fastmap_accurate
= 0;
1563 bufp
->not_bol
= bufp
->not_eol
= 0;
1565 /* Set `used' to zero, so that if we return an error, the pattern
1566 printer (for debugging) will think there's no pattern. We reset it
1570 /* Always count groups, whether or not bufp->no_sub is set. */
1573 #if !defined (emacs) && !defined (SYNTAX_TABLE)
1574 /* Initialize the syntax table. */
1575 init_syntax_once ();
1578 if (bufp
->allocated
== 0)
1581 { /* If zero allocated, but buffer is non-null, try to realloc
1582 enough space. This loses if buffer's address is bogus, but
1583 that is the user's responsibility. */
1584 RETALLOC (bufp
->buffer
, INIT_BUF_SIZE
, unsigned char);
1587 { /* Caller did not allocate a buffer. Do it for them. */
1588 bufp
->buffer
= TALLOC (INIT_BUF_SIZE
, unsigned char);
1590 if (!bufp
->buffer
) FREE_STACK_RETURN (REG_ESPACE
);
1592 bufp
->allocated
= INIT_BUF_SIZE
;
1595 begalt
= b
= bufp
->buffer
;
1597 /* Loop through the uncompiled pattern until we're at the end. */
1606 if ( /* If at start of pattern, it's an operator. */
1608 /* If context independent, it's an operator. */
1609 || syntax
& RE_CONTEXT_INDEP_ANCHORS
1610 /* Otherwise, depends on what's come before. */
1611 || at_begline_loc_p (pattern
, p
, syntax
))
1621 if ( /* If at end of pattern, it's an operator. */
1623 /* If context independent, it's an operator. */
1624 || syntax
& RE_CONTEXT_INDEP_ANCHORS
1625 /* Otherwise, depends on what's next. */
1626 || at_endline_loc_p (p
, pend
, syntax
))
1636 if ((syntax
& RE_BK_PLUS_QM
)
1637 || (syntax
& RE_LIMITED_OPS
))
1641 /* If there is no previous pattern... */
1644 if (syntax
& RE_CONTEXT_INVALID_OPS
)
1645 FREE_STACK_RETURN (REG_BADRPT
);
1646 else if (!(syntax
& RE_CONTEXT_INDEP_OPS
))
1651 /* Are we optimizing this jump? */
1652 boolean keep_string_p
= false;
1654 /* 1 means zero (many) matches is allowed. */
1655 char zero_times_ok
= 0, many_times_ok
= 0;
1657 /* If there is a sequence of repetition chars, collapse it
1658 down to just one (the right one). We can't combine
1659 interval operators with these because of, e.g., `a{2}*',
1660 which should only match an even number of `a's. */
1664 zero_times_ok
|= c
!= '+';
1665 many_times_ok
|= c
!= '?';
1673 || (!(syntax
& RE_BK_PLUS_QM
) && (c
== '+' || c
== '?')))
1676 else if (syntax
& RE_BK_PLUS_QM
&& c
== '\\')
1678 if (p
== pend
) FREE_STACK_RETURN (REG_EESCAPE
);
1681 if (!(c1
== '+' || c1
== '?'))
1696 /* If we get here, we found another repeat character. */
1699 /* Star, etc. applied to an empty pattern is equivalent
1700 to an empty pattern. */
1704 /* Now we know whether or not zero matches is allowed
1705 and also whether or not two or more matches is allowed. */
1707 { /* More than one repetition is allowed, so put in at the
1708 end a backward relative jump from `b' to before the next
1709 jump we're going to put in below (which jumps from
1710 laststart to after this jump).
1712 But if we are at the `*' in the exact sequence `.*\n',
1713 insert an unconditional jump backwards to the .,
1714 instead of the beginning of the loop. This way we only
1715 push a failure point once, instead of every time
1716 through the loop. */
1717 assert (p
- 1 > pattern
);
1719 /* Allocate the space for the jump. */
1720 GET_BUFFER_SPACE (3);
1722 /* We know we are not at the first character of the pattern,
1723 because laststart was nonzero. And we've already
1724 incremented `p', by the way, to be the character after
1725 the `*'. Do we have to do something analogous here
1726 for null bytes, because of RE_DOT_NOT_NULL? */
1727 if (TRANSLATE (*(p
- 2)) == TRANSLATE ('.')
1729 && p
< pend
&& TRANSLATE (*p
) == TRANSLATE ('\n')
1730 && !(syntax
& RE_DOT_NEWLINE
))
1731 { /* We have .*\n. */
1732 STORE_JUMP (jump
, b
, laststart
);
1733 keep_string_p
= true;
1736 /* Anything else. */
1737 STORE_JUMP (maybe_pop_jump
, b
, laststart
- 3);
1739 /* We've added more stuff to the buffer. */
1743 /* On failure, jump from laststart to b + 3, which will be the
1744 end of the buffer after this jump is inserted. */
1745 GET_BUFFER_SPACE (3);
1746 INSERT_JUMP (keep_string_p
? on_failure_keep_string_jump
1754 /* At least one repetition is required, so insert a
1755 `dummy_failure_jump' before the initial
1756 `on_failure_jump' instruction of the loop. This
1757 effects a skip over that instruction the first time
1758 we hit that loop. */
1759 GET_BUFFER_SPACE (3);
1760 INSERT_JUMP (dummy_failure_jump
, laststart
, laststart
+ 6);
1775 boolean had_char_class
= false;
1777 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
1779 /* Ensure that we have enough space to push a charset: the
1780 opcode, the length count, and the bitset; 34 bytes in all. */
1781 GET_BUFFER_SPACE (34);
1785 /* We test `*p == '^' twice, instead of using an if
1786 statement, so we only need one BUF_PUSH. */
1787 BUF_PUSH (*p
== '^' ? charset_not
: charset
);
1791 /* Remember the first position in the bracket expression. */
1794 /* Push the number of bytes in the bitmap. */
1795 BUF_PUSH ((1 << BYTEWIDTH
) / BYTEWIDTH
);
1797 /* Clear the whole map. */
1798 bzero (b
, (1 << BYTEWIDTH
) / BYTEWIDTH
);
1800 /* charset_not matches newline according to a syntax bit. */
1801 if ((re_opcode_t
) b
[-2] == charset_not
1802 && (syntax
& RE_HAT_LISTS_NOT_NEWLINE
))
1803 SET_LIST_BIT ('\n');
1805 /* Read in characters and ranges, setting map bits. */
1808 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
1812 /* \ might escape characters inside [...] and [^...]. */
1813 if ((syntax
& RE_BACKSLASH_ESCAPE_IN_LISTS
) && c
== '\\')
1815 if (p
== pend
) FREE_STACK_RETURN (REG_EESCAPE
);
1822 /* Could be the end of the bracket expression. If it's
1823 not (i.e., when the bracket expression is `[]' so
1824 far), the ']' character bit gets set way below. */
1825 if (c
== ']' && p
!= p1
+ 1)
1828 /* Look ahead to see if it's a range when the last thing
1829 was a character class. */
1830 if (had_char_class
&& c
== '-' && *p
!= ']')
1831 FREE_STACK_RETURN (REG_ERANGE
);
1833 /* Look ahead to see if it's a range when the last thing
1834 was a character: if this is a hyphen not at the
1835 beginning or the end of a list, then it's the range
1838 && !(p
- 2 >= pattern
&& p
[-2] == '[')
1839 && !(p
- 3 >= pattern
&& p
[-3] == '[' && p
[-2] == '^')
1843 = compile_range (&p
, pend
, translate
, syntax
, b
);
1844 if (ret
!= REG_NOERROR
) FREE_STACK_RETURN (ret
);
1847 else if (p
[0] == '-' && p
[1] != ']')
1848 { /* This handles ranges made up of characters only. */
1851 /* Move past the `-'. */
1854 ret
= compile_range (&p
, pend
, translate
, syntax
, b
);
1855 if (ret
!= REG_NOERROR
) FREE_STACK_RETURN (ret
);
1858 /* See if we're at the beginning of a possible character
1861 else if (syntax
& RE_CHAR_CLASSES
&& c
== '[' && *p
== ':')
1862 { /* Leave room for the null. */
1863 char str
[CHAR_CLASS_MAX_LENGTH
+ 1];
1868 /* If pattern is `[[:'. */
1869 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
1874 if (c
== ':' || c
== ']' || p
== pend
1875 || c1
== CHAR_CLASS_MAX_LENGTH
)
1881 /* If isn't a word bracketed by `[:' and:`]':
1882 undo the ending character, the letters, and leave
1883 the leading `:' and `[' (but set bits for them). */
1884 if (c
== ':' && *p
== ']')
1887 boolean is_alnum
= STREQ (str
, "alnum");
1888 boolean is_alpha
= STREQ (str
, "alpha");
1889 boolean is_blank
= STREQ (str
, "blank");
1890 boolean is_cntrl
= STREQ (str
, "cntrl");
1891 boolean is_digit
= STREQ (str
, "digit");
1892 boolean is_graph
= STREQ (str
, "graph");
1893 boolean is_lower
= STREQ (str
, "lower");
1894 boolean is_print
= STREQ (str
, "print");
1895 boolean is_punct
= STREQ (str
, "punct");
1896 boolean is_space
= STREQ (str
, "space");
1897 boolean is_upper
= STREQ (str
, "upper");
1898 boolean is_xdigit
= STREQ (str
, "xdigit");
1900 if (!IS_CHAR_CLASS (str
))
1901 FREE_STACK_RETURN (REG_ECTYPE
);
1903 /* Throw away the ] at the end of the character
1907 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
1909 for (ch
= 0; ch
< 1 << BYTEWIDTH
; ch
++)
1911 /* This was split into 3 if's to
1912 avoid an arbitrary limit in some compiler. */
1913 if ( (is_alnum
&& ISALNUM (ch
))
1914 || (is_alpha
&& ISALPHA (ch
))
1915 || (is_blank
&& ISBLANK (ch
))
1916 || (is_cntrl
&& ISCNTRL (ch
)))
1918 if ( (is_digit
&& ISDIGIT (ch
))
1919 || (is_graph
&& ISGRAPH (ch
))
1920 || (is_lower
&& ISLOWER (ch
))
1921 || (is_print
&& ISPRINT (ch
)))
1923 if ( (is_punct
&& ISPUNCT (ch
))
1924 || (is_space
&& ISSPACE (ch
))
1925 || (is_upper
&& ISUPPER (ch
))
1926 || (is_xdigit
&& ISXDIGIT (ch
)))
1929 had_char_class
= true;
1938 had_char_class
= false;
1943 had_char_class
= false;
1948 /* Discard any (non)matching list bytes that are all 0 at the
1949 end of the map. Decrease the map-length byte too. */
1950 while ((int) b
[-1] > 0 && b
[b
[-1] - 1] == 0)
1958 if (syntax
& RE_NO_BK_PARENS
)
1965 if (syntax
& RE_NO_BK_PARENS
)
1972 if (syntax
& RE_NEWLINE_ALT
)
1979 if (syntax
& RE_NO_BK_VBAR
)
1986 if (syntax
& RE_INTERVALS
&& syntax
& RE_NO_BK_BRACES
)
1987 goto handle_interval
;
1993 if (p
== pend
) FREE_STACK_RETURN (REG_EESCAPE
);
1995 /* Do not translate the character after the \, so that we can
1996 distinguish, e.g., \B from \b, even if we normally would
1997 translate, e.g., B to b. */
2003 if (syntax
& RE_NO_BK_PARENS
)
2004 goto normal_backslash
;
2010 if (COMPILE_STACK_FULL
)
2012 RETALLOC (compile_stack
.stack
, compile_stack
.size
<< 1,
2013 compile_stack_elt_t
);
2014 if (compile_stack
.stack
== NULL
) return REG_ESPACE
;
2016 compile_stack
.size
<<= 1;
2019 /* These are the values to restore when we hit end of this
2020 group. They are all relative offsets, so that if the
2021 whole pattern moves because of realloc, they will still
2023 COMPILE_STACK_TOP
.begalt_offset
= begalt
- bufp
->buffer
;
2024 COMPILE_STACK_TOP
.fixup_alt_jump
2025 = fixup_alt_jump
? fixup_alt_jump
- bufp
->buffer
+ 1 : 0;
2026 COMPILE_STACK_TOP
.laststart_offset
= b
- bufp
->buffer
;
2027 COMPILE_STACK_TOP
.regnum
= regnum
;
2029 /* We will eventually replace the 0 with the number of
2030 groups inner to this one. But do not push a
2031 start_memory for groups beyond the last one we can
2032 represent in the compiled pattern. */
2033 if (regnum
<= MAX_REGNUM
)
2035 COMPILE_STACK_TOP
.inner_group_offset
= b
- bufp
->buffer
+ 2;
2036 BUF_PUSH_3 (start_memory
, regnum
, 0);
2039 compile_stack
.avail
++;
2044 /* If we've reached MAX_REGNUM groups, then this open
2045 won't actually generate any code, so we'll have to
2046 clear pending_exact explicitly. */
2052 if (syntax
& RE_NO_BK_PARENS
) goto normal_backslash
;
2054 if (COMPILE_STACK_EMPTY
)
2055 if (syntax
& RE_UNMATCHED_RIGHT_PAREN_ORD
)
2056 goto normal_backslash
;
2058 FREE_STACK_RETURN (REG_ERPAREN
);
2062 { /* Push a dummy failure point at the end of the
2063 alternative for a possible future
2064 `pop_failure_jump' to pop. See comments at
2065 `push_dummy_failure' in `re_match_2'. */
2066 BUF_PUSH (push_dummy_failure
);
2068 /* We allocated space for this jump when we assigned
2069 to `fixup_alt_jump', in the `handle_alt' case below. */
2070 STORE_JUMP (jump_past_alt
, fixup_alt_jump
, b
- 1);
2073 /* See similar code for backslashed left paren above. */
2074 if (COMPILE_STACK_EMPTY
)
2075 if (syntax
& RE_UNMATCHED_RIGHT_PAREN_ORD
)
2078 FREE_STACK_RETURN (REG_ERPAREN
);
2080 /* Since we just checked for an empty stack above, this
2081 ``can't happen''. */
2082 assert (compile_stack
.avail
!= 0);
2084 /* We don't just want to restore into `regnum', because
2085 later groups should continue to be numbered higher,
2086 as in `(ab)c(de)' -- the second group is #2. */
2087 regnum_t this_group_regnum
;
2089 compile_stack
.avail
--;
2090 begalt
= bufp
->buffer
+ COMPILE_STACK_TOP
.begalt_offset
;
2092 = COMPILE_STACK_TOP
.fixup_alt_jump
2093 ? bufp
->buffer
+ COMPILE_STACK_TOP
.fixup_alt_jump
- 1
2095 laststart
= bufp
->buffer
+ COMPILE_STACK_TOP
.laststart_offset
;
2096 this_group_regnum
= COMPILE_STACK_TOP
.regnum
;
2097 /* If we've reached MAX_REGNUM groups, then this open
2098 won't actually generate any code, so we'll have to
2099 clear pending_exact explicitly. */
2102 /* We're at the end of the group, so now we know how many
2103 groups were inside this one. */
2104 if (this_group_regnum
<= MAX_REGNUM
)
2106 unsigned char *inner_group_loc
2107 = bufp
->buffer
+ COMPILE_STACK_TOP
.inner_group_offset
;
2109 *inner_group_loc
= regnum
- this_group_regnum
;
2110 BUF_PUSH_3 (stop_memory
, this_group_regnum
,
2111 regnum
- this_group_regnum
);
2117 case '|': /* `\|'. */
2118 if (syntax
& RE_LIMITED_OPS
|| syntax
& RE_NO_BK_VBAR
)
2119 goto normal_backslash
;
2121 if (syntax
& RE_LIMITED_OPS
)
2124 /* Insert before the previous alternative a jump which
2125 jumps to this alternative if the former fails. */
2126 GET_BUFFER_SPACE (3);
2127 INSERT_JUMP (on_failure_jump
, begalt
, b
+ 6);
2131 /* The alternative before this one has a jump after it
2132 which gets executed if it gets matched. Adjust that
2133 jump so it will jump to this alternative's analogous
2134 jump (put in below, which in turn will jump to the next
2135 (if any) alternative's such jump, etc.). The last such
2136 jump jumps to the correct final destination. A picture:
2142 If we are at `b', then fixup_alt_jump right now points to a
2143 three-byte space after `a'. We'll put in the jump, set
2144 fixup_alt_jump to right after `b', and leave behind three
2145 bytes which we'll fill in when we get to after `c'. */
2148 STORE_JUMP (jump_past_alt
, fixup_alt_jump
, b
);
2150 /* Mark and leave space for a jump after this alternative,
2151 to be filled in later either by next alternative or
2152 when know we're at the end of a series of alternatives. */
2154 GET_BUFFER_SPACE (3);
2163 /* If \{ is a literal. */
2164 if (!(syntax
& RE_INTERVALS
)
2165 /* If we're at `\{' and it's not the open-interval
2167 || ((syntax
& RE_INTERVALS
) && (syntax
& RE_NO_BK_BRACES
))
2168 || (p
- 2 == pattern
&& p
== pend
))
2169 goto normal_backslash
;
2173 /* If got here, then the syntax allows intervals. */
2175 /* At least (most) this many matches must be made. */
2176 int lower_bound
= -1, upper_bound
= -1;
2178 beg_interval
= p
- 1;
2182 if (syntax
& RE_NO_BK_BRACES
)
2183 goto unfetch_interval
;
2185 FREE_STACK_RETURN (REG_EBRACE
);
2188 GET_UNSIGNED_NUMBER (lower_bound
);
2192 GET_UNSIGNED_NUMBER (upper_bound
);
2193 if (upper_bound
< 0) upper_bound
= RE_DUP_MAX
;
2196 /* Interval such as `{1}' => match exactly once. */
2197 upper_bound
= lower_bound
;
2199 if (lower_bound
< 0 || upper_bound
> RE_DUP_MAX
2200 || lower_bound
> upper_bound
)
2202 if (syntax
& RE_NO_BK_BRACES
)
2203 goto unfetch_interval
;
2205 FREE_STACK_RETURN (REG_BADBR
);
2208 if (!(syntax
& RE_NO_BK_BRACES
))
2210 if (c
!= '\\') FREE_STACK_RETURN (REG_EBRACE
);
2217 if (syntax
& RE_NO_BK_BRACES
)
2218 goto unfetch_interval
;
2220 FREE_STACK_RETURN (REG_BADBR
);
2223 /* We just parsed a valid interval. */
2225 /* If it's invalid to have no preceding re. */
2228 if (syntax
& RE_CONTEXT_INVALID_OPS
)
2229 FREE_STACK_RETURN (REG_BADRPT
);
2230 else if (syntax
& RE_CONTEXT_INDEP_OPS
)
2233 goto unfetch_interval
;
2236 /* If the upper bound is zero, don't want to succeed at
2237 all; jump from `laststart' to `b + 3', which will be
2238 the end of the buffer after we insert the jump. */
2239 if (upper_bound
== 0)
2241 GET_BUFFER_SPACE (3);
2242 INSERT_JUMP (jump
, laststart
, b
+ 3);
2246 /* Otherwise, we have a nontrivial interval. When
2247 we're all done, the pattern will look like:
2248 set_number_at <jump count> <upper bound>
2249 set_number_at <succeed_n count> <lower bound>
2250 succeed_n <after jump addr> <succeed_n count>
2252 jump_n <succeed_n addr> <jump count>
2253 (The upper bound and `jump_n' are omitted if
2254 `upper_bound' is 1, though.) */
2256 { /* If the upper bound is > 1, we need to insert
2257 more at the end of the loop. */
2258 unsigned nbytes
= 10 + (upper_bound
> 1) * 10;
2260 GET_BUFFER_SPACE (nbytes
);
2262 /* Initialize lower bound of the `succeed_n', even
2263 though it will be set during matching by its
2264 attendant `set_number_at' (inserted next),
2265 because `re_compile_fastmap' needs to know.
2266 Jump to the `jump_n' we might insert below. */
2267 INSERT_JUMP2 (succeed_n
, laststart
,
2268 b
+ 5 + (upper_bound
> 1) * 5,
2272 /* Code to initialize the lower bound. Insert
2273 before the `succeed_n'. The `5' is the last two
2274 bytes of this `set_number_at', plus 3 bytes of
2275 the following `succeed_n'. */
2276 insert_op2 (set_number_at
, laststart
, 5, lower_bound
, b
);
2279 if (upper_bound
> 1)
2280 { /* More than one repetition is allowed, so
2281 append a backward jump to the `succeed_n'
2282 that starts this interval.
2284 When we've reached this during matching,
2285 we'll have matched the interval once, so
2286 jump back only `upper_bound - 1' times. */
2287 STORE_JUMP2 (jump_n
, b
, laststart
+ 5,
2291 /* The location we want to set is the second
2292 parameter of the `jump_n'; that is `b-2' as
2293 an absolute address. `laststart' will be
2294 the `set_number_at' we're about to insert;
2295 `laststart+3' the number to set, the source
2296 for the relative address. But we are
2297 inserting into the middle of the pattern --
2298 so everything is getting moved up by 5.
2299 Conclusion: (b - 2) - (laststart + 3) + 5,
2300 i.e., b - laststart.
2302 We insert this at the beginning of the loop
2303 so that if we fail during matching, we'll
2304 reinitialize the bounds. */
2305 insert_op2 (set_number_at
, laststart
, b
- laststart
,
2306 upper_bound
- 1, b
);
2311 beg_interval
= NULL
;
2316 /* If an invalid interval, match the characters as literals. */
2317 assert (beg_interval
);
2319 beg_interval
= NULL
;
2321 /* normal_char and normal_backslash need `c'. */
2324 if (!(syntax
& RE_NO_BK_BRACES
))
2326 if (p
> pattern
&& p
[-1] == '\\')
2327 goto normal_backslash
;
2332 /* There is no way to specify the before_dot and after_dot
2333 operators. rms says this is ok. --karl */
2341 BUF_PUSH_2 (syntaxspec
, syntax_spec_code
[c
]);
2347 BUF_PUSH_2 (notsyntaxspec
, syntax_spec_code
[c
]);
2354 BUF_PUSH (wordchar
);
2360 BUF_PUSH (notwordchar
);
2373 BUF_PUSH (wordbound
);
2377 BUF_PUSH (notwordbound
);
2388 case '1': case '2': case '3': case '4': case '5':
2389 case '6': case '7': case '8': case '9':
2390 if (syntax
& RE_NO_BK_REFS
)
2396 FREE_STACK_RETURN (REG_ESUBREG
);
2398 /* Can't back reference to a subexpression if inside of it. */
2399 if (group_in_compile_stack (compile_stack
, c1
))
2403 BUF_PUSH_2 (duplicate
, c1
);
2409 if (syntax
& RE_BK_PLUS_QM
)
2412 goto normal_backslash
;
2416 /* You might think it would be useful for \ to mean
2417 not to translate; but if we don't translate it
2418 it will never match anything. */
2426 /* Expects the character in `c'. */
2428 /* If no exactn currently being built. */
2431 /* If last exactn not at current position. */
2432 || pending_exact
+ *pending_exact
+ 1 != b
2434 /* We have only one byte following the exactn for the count. */
2435 || *pending_exact
== (1 << BYTEWIDTH
) - 1
2437 /* If followed by a repetition operator. */
2438 || *p
== '*' || *p
== '^'
2439 || ((syntax
& RE_BK_PLUS_QM
)
2440 ? *p
== '\\' && (p
[1] == '+' || p
[1] == '?')
2441 : (*p
== '+' || *p
== '?'))
2442 || ((syntax
& RE_INTERVALS
)
2443 && ((syntax
& RE_NO_BK_BRACES
)
2445 : (p
[0] == '\\' && p
[1] == '{'))))
2447 /* Start building a new exactn. */
2451 BUF_PUSH_2 (exactn
, 0);
2452 pending_exact
= b
- 1;
2459 } /* while p != pend */
2462 /* Through the pattern now. */
2465 STORE_JUMP (jump_past_alt
, fixup_alt_jump
, b
);
2467 if (!COMPILE_STACK_EMPTY
)
2468 FREE_STACK_RETURN (REG_EPAREN
);
2470 free (compile_stack
.stack
);
2472 /* We have succeeded; set the length of the buffer. */
2473 bufp
->used
= b
- bufp
->buffer
;
2478 DEBUG_PRINT1 ("\nCompiled pattern: \n");
2479 print_compiled_pattern (bufp
);
2483 #ifndef MATCH_MAY_ALLOCATE
2484 /* Initialize the failure stack to the largest possible stack. This
2485 isn't necessary unless we're trying to avoid calling alloca in
2486 the search and match routines. */
2488 int num_regs
= bufp
->re_nsub
+ 1;
2490 /* Since DOUBLE_FAIL_STACK refuses to double only if the current size
2491 is strictly greater than re_max_failures, the largest possible stack
2492 is 2 * re_max_failures failure points. */
2493 if (fail_stack
.size
< (2 * re_max_failures
* MAX_FAILURE_ITEMS
))
2495 fail_stack
.size
= (2 * re_max_failures
* MAX_FAILURE_ITEMS
);
2498 if (! fail_stack
.stack
)
2500 = (fail_stack_elt_t
*) xmalloc (fail_stack
.size
2501 * sizeof (fail_stack_elt_t
));
2504 = (fail_stack_elt_t
*) xrealloc (fail_stack
.stack
,
2506 * sizeof (fail_stack_elt_t
)));
2507 #else /* not emacs */
2508 if (! fail_stack
.stack
)
2510 = (fail_stack_elt_t
*) malloc (fail_stack
.size
2511 * sizeof (fail_stack_elt_t
));
2514 = (fail_stack_elt_t
*) realloc (fail_stack
.stack
,
2516 * sizeof (fail_stack_elt_t
)));
2517 #endif /* not emacs */
2520 /* Initialize some other variables the matcher uses. */
2521 RETALLOC_IF (regstart
, num_regs
, const char *);
2522 RETALLOC_IF (regend
, num_regs
, const char *);
2523 RETALLOC_IF (old_regstart
, num_regs
, const char *);
2524 RETALLOC_IF (old_regend
, num_regs
, const char *);
2525 RETALLOC_IF (best_regstart
, num_regs
, const char *);
2526 RETALLOC_IF (best_regend
, num_regs
, const char *);
2527 RETALLOC_IF (reg_info
, num_regs
, register_info_type
);
2528 RETALLOC_IF (reg_dummy
, num_regs
, const char *);
2529 RETALLOC_IF (reg_info_dummy
, num_regs
, register_info_type
);
2534 } /* regex_compile */
2536 /* Subroutines for `regex_compile'. */
2538 /* Store OP at LOC followed by two-byte integer parameter ARG. */
2541 store_op1 (op
, loc
, arg
)
2546 *loc
= (unsigned char) op
;
2547 STORE_NUMBER (loc
+ 1, arg
);
2551 /* Like `store_op1', but for two two-byte parameters ARG1 and ARG2. */
2554 store_op2 (op
, loc
, arg1
, arg2
)
2559 *loc
= (unsigned char) op
;
2560 STORE_NUMBER (loc
+ 1, arg1
);
2561 STORE_NUMBER (loc
+ 3, arg2
);
2565 /* Copy the bytes from LOC to END to open up three bytes of space at LOC
2566 for OP followed by two-byte integer parameter ARG. */
2569 insert_op1 (op
, loc
, arg
, end
)
2575 register unsigned char *pfrom
= end
;
2576 register unsigned char *pto
= end
+ 3;
2578 while (pfrom
!= loc
)
2581 store_op1 (op
, loc
, arg
);
2585 /* Like `insert_op1', but for two two-byte parameters ARG1 and ARG2. */
2588 insert_op2 (op
, loc
, arg1
, arg2
, end
)
2594 register unsigned char *pfrom
= end
;
2595 register unsigned char *pto
= end
+ 5;
2597 while (pfrom
!= loc
)
2600 store_op2 (op
, loc
, arg1
, arg2
);
2604 /* P points to just after a ^ in PATTERN. Return true if that ^ comes
2605 after an alternative or a begin-subexpression. We assume there is at
2606 least one character before the ^. */
2609 at_begline_loc_p (pattern
, p
, syntax
)
2610 const char *pattern
, *p
;
2611 reg_syntax_t syntax
;
2613 const char *prev
= p
- 2;
2614 boolean prev_prev_backslash
= prev
> pattern
&& prev
[-1] == '\\';
2617 /* After a subexpression? */
2618 (*prev
== '(' && (syntax
& RE_NO_BK_PARENS
|| prev_prev_backslash
))
2619 /* After an alternative? */
2620 || (*prev
== '|' && (syntax
& RE_NO_BK_VBAR
|| prev_prev_backslash
));
2624 /* The dual of at_begline_loc_p. This one is for $. We assume there is
2625 at least one character after the $, i.e., `P < PEND'. */
2628 at_endline_loc_p (p
, pend
, syntax
)
2629 const char *p
, *pend
;
2632 const char *next
= p
;
2633 boolean next_backslash
= *next
== '\\';
2634 const char *next_next
= p
+ 1 < pend
? p
+ 1 : NULL
;
2637 /* Before a subexpression? */
2638 (syntax
& RE_NO_BK_PARENS
? *next
== ')'
2639 : next_backslash
&& next_next
&& *next_next
== ')')
2640 /* Before an alternative? */
2641 || (syntax
& RE_NO_BK_VBAR
? *next
== '|'
2642 : next_backslash
&& next_next
&& *next_next
== '|');
2646 /* Returns true if REGNUM is in one of COMPILE_STACK's elements and
2647 false if it's not. */
2650 group_in_compile_stack (compile_stack
, regnum
)
2651 compile_stack_type compile_stack
;
2656 for (this_element
= compile_stack
.avail
- 1;
2659 if (compile_stack
.stack
[this_element
].regnum
== regnum
)
2666 /* Read the ending character of a range (in a bracket expression) from the
2667 uncompiled pattern *P_PTR (which ends at PEND). We assume the
2668 starting character is in `P[-2]'. (`P[-1]' is the character `-'.)
2669 Then we set the translation of all bits between the starting and
2670 ending characters (inclusive) in the compiled pattern B.
2672 Return an error code.
2674 We use these short variable names so we can use the same macros as
2675 `regex_compile' itself. */
2677 static reg_errcode_t
2678 compile_range (p_ptr
, pend
, translate
, syntax
, b
)
2679 const char **p_ptr
, *pend
;
2681 reg_syntax_t syntax
;
2686 const char *p
= *p_ptr
;
2687 int range_start
, range_end
;
2692 /* Even though the pattern is a signed `char *', we need to fetch
2693 with unsigned char *'s; if the high bit of the pattern character
2694 is set, the range endpoints will be negative if we fetch using a
2697 We also want to fetch the endpoints without translating them; the
2698 appropriate translation is done in the bit-setting loop below. */
2699 /* The SVR4 compiler on the 3B2 had trouble with unsigned const char *. */
2700 range_start
= ((const unsigned char *) p
)[-2];
2701 range_end
= ((const unsigned char *) p
)[0];
2703 /* Have to increment the pointer into the pattern string, so the
2704 caller isn't still at the ending character. */
2707 /* If the start is after the end, the range is empty. */
2708 if (range_start
> range_end
)
2709 return syntax
& RE_NO_EMPTY_RANGES
? REG_ERANGE
: REG_NOERROR
;
2711 /* Here we see why `this_char' has to be larger than an `unsigned
2712 char' -- the range is inclusive, so if `range_end' == 0xff
2713 (assuming 8-bit characters), we would otherwise go into an infinite
2714 loop, since all characters <= 0xff. */
2715 for (this_char
= range_start
; this_char
<= range_end
; this_char
++)
2717 SET_LIST_BIT (TRANSLATE (this_char
));
2723 /* re_compile_fastmap computes a ``fastmap'' for the compiled pattern in
2724 BUFP. A fastmap records which of the (1 << BYTEWIDTH) possible
2725 characters can start a string that matches the pattern. This fastmap
2726 is used by re_search to skip quickly over impossible starting points.
2728 The caller must supply the address of a (1 << BYTEWIDTH)-byte data
2729 area as BUFP->fastmap.
2731 We set the `fastmap', `fastmap_accurate', and `can_be_null' fields in
2734 Returns 0 if we succeed, -2 if an internal error. */
2737 re_compile_fastmap (bufp
)
2738 struct re_pattern_buffer
*bufp
;
2741 #ifdef MATCH_MAY_ALLOCATE
2742 fail_stack_type fail_stack
;
2744 #ifndef REGEX_MALLOC
2747 /* We don't push any register information onto the failure stack. */
2748 unsigned num_regs
= 0;
2750 register char *fastmap
= bufp
->fastmap
;
2751 unsigned char *pattern
= bufp
->buffer
;
2752 unsigned long size
= bufp
->used
;
2753 unsigned char *p
= pattern
;
2754 register unsigned char *pend
= pattern
+ size
;
2756 /* Assume that each path through the pattern can be null until
2757 proven otherwise. We set this false at the bottom of switch
2758 statement, to which we get only if a particular path doesn't
2759 match the empty string. */
2760 boolean path_can_be_null
= true;
2762 /* We aren't doing a `succeed_n' to begin with. */
2763 boolean succeed_n_p
= false;
2765 assert (fastmap
!= NULL
&& p
!= NULL
);
2768 bzero (fastmap
, 1 << BYTEWIDTH
); /* Assume nothing's valid. */
2769 bufp
->fastmap_accurate
= 1; /* It will be when we're done. */
2770 bufp
->can_be_null
= 0;
2772 while (p
!= pend
|| !FAIL_STACK_EMPTY ())
2776 bufp
->can_be_null
|= path_can_be_null
;
2778 /* Reset for next path. */
2779 path_can_be_null
= true;
2781 p
= fail_stack
.stack
[--fail_stack
.avail
];
2784 /* We should never be about to go beyond the end of the pattern. */
2787 #ifdef SWITCH_ENUM_BUG
2788 switch ((int) ((re_opcode_t
) *p
++))
2790 switch ((re_opcode_t
) *p
++)
2794 /* I guess the idea here is to simply not bother with a fastmap
2795 if a backreference is used, since it's too hard to figure out
2796 the fastmap for the corresponding group. Setting
2797 `can_be_null' stops `re_search_2' from using the fastmap, so
2798 that is all we do. */
2800 bufp
->can_be_null
= 1;
2804 /* Following are the cases which match a character. These end
2813 for (j
= *p
++ * BYTEWIDTH
- 1; j
>= 0; j
--)
2814 if (p
[j
/ BYTEWIDTH
] & (1 << (j
% BYTEWIDTH
)))
2820 /* Chars beyond end of map must be allowed. */
2821 for (j
= *p
* BYTEWIDTH
; j
< (1 << BYTEWIDTH
); j
++)
2824 for (j
= *p
++ * BYTEWIDTH
- 1; j
>= 0; j
--)
2825 if (!(p
[j
/ BYTEWIDTH
] & (1 << (j
% BYTEWIDTH
))))
2831 for (j
= 0; j
< (1 << BYTEWIDTH
); j
++)
2832 if (SYNTAX (j
) == Sword
)
2838 for (j
= 0; j
< (1 << BYTEWIDTH
); j
++)
2839 if (SYNTAX (j
) != Sword
)
2846 int fastmap_newline
= fastmap
['\n'];
2848 /* `.' matches anything ... */
2849 for (j
= 0; j
< (1 << BYTEWIDTH
); j
++)
2852 /* ... except perhaps newline. */
2853 if (!(bufp
->syntax
& RE_DOT_NEWLINE
))
2854 fastmap
['\n'] = fastmap_newline
;
2856 /* Return if we have already set `can_be_null'; if we have,
2857 then the fastmap is irrelevant. Something's wrong here. */
2858 else if (bufp
->can_be_null
)
2861 /* Otherwise, have to check alternative paths. */
2868 for (j
= 0; j
< (1 << BYTEWIDTH
); j
++)
2869 if (SYNTAX (j
) == (enum syntaxcode
) k
)
2876 for (j
= 0; j
< (1 << BYTEWIDTH
); j
++)
2877 if (SYNTAX (j
) != (enum syntaxcode
) k
)
2882 /* All cases after this match the empty string. These end with
2890 #endif /* not emacs */
2902 case push_dummy_failure
:
2907 case pop_failure_jump
:
2908 case maybe_pop_jump
:
2911 case dummy_failure_jump
:
2912 EXTRACT_NUMBER_AND_INCR (j
, p
);
2917 /* Jump backward implies we just went through the body of a
2918 loop and matched nothing. Opcode jumped to should be
2919 `on_failure_jump' or `succeed_n'. Just treat it like an
2920 ordinary jump. For a * loop, it has pushed its failure
2921 point already; if so, discard that as redundant. */
2922 if ((re_opcode_t
) *p
!= on_failure_jump
2923 && (re_opcode_t
) *p
!= succeed_n
)
2927 EXTRACT_NUMBER_AND_INCR (j
, p
);
2930 /* If what's on the stack is where we are now, pop it. */
2931 if (!FAIL_STACK_EMPTY ()
2932 && fail_stack
.stack
[fail_stack
.avail
- 1] == p
)
2938 case on_failure_jump
:
2939 case on_failure_keep_string_jump
:
2940 handle_on_failure_jump
:
2941 EXTRACT_NUMBER_AND_INCR (j
, p
);
2943 /* For some patterns, e.g., `(a?)?', `p+j' here points to the
2944 end of the pattern. We don't want to push such a point,
2945 since when we restore it above, entering the switch will
2946 increment `p' past the end of the pattern. We don't need
2947 to push such a point since we obviously won't find any more
2948 fastmap entries beyond `pend'. Such a pattern can match
2949 the null string, though. */
2952 if (!PUSH_PATTERN_OP (p
+ j
, fail_stack
))
2956 bufp
->can_be_null
= 1;
2960 EXTRACT_NUMBER_AND_INCR (k
, p
); /* Skip the n. */
2961 succeed_n_p
= false;
2968 /* Get to the number of times to succeed. */
2971 /* Increment p past the n for when k != 0. */
2972 EXTRACT_NUMBER_AND_INCR (k
, p
);
2976 succeed_n_p
= true; /* Spaghetti code alert. */
2977 goto handle_on_failure_jump
;
2994 abort (); /* We have listed all the cases. */
2997 /* Getting here means we have found the possible starting
2998 characters for one path of the pattern -- and that the empty
2999 string does not match. We need not follow this path further.
3000 Instead, look at the next alternative (remembered on the
3001 stack), or quit if no more. The test at the top of the loop
3002 does these things. */
3003 path_can_be_null
= false;
3007 /* Set `can_be_null' for the last path (also the first path, if the
3008 pattern is empty). */
3009 bufp
->can_be_null
|= path_can_be_null
;
3011 } /* re_compile_fastmap */
3013 /* Set REGS to hold NUM_REGS registers, storing them in STARTS and
3014 ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use
3015 this memory for recording register information. STARTS and ENDS
3016 must be allocated using the malloc library routine, and must each
3017 be at least NUM_REGS * sizeof (regoff_t) bytes long.
3019 If NUM_REGS == 0, then subsequent matches should allocate their own
3022 Unless this function is called, the first search or match using
3023 PATTERN_BUFFER will allocate its own register data, without
3024 freeing the old data. */
3027 re_set_registers (bufp
, regs
, num_regs
, starts
, ends
)
3028 struct re_pattern_buffer
*bufp
;
3029 struct re_registers
*regs
;
3031 regoff_t
*starts
, *ends
;
3035 bufp
->regs_allocated
= REGS_REALLOCATE
;
3036 regs
->num_regs
= num_regs
;
3037 regs
->start
= starts
;
3042 bufp
->regs_allocated
= REGS_UNALLOCATED
;
3044 regs
->start
= regs
->end
= (regoff_t
*) 0;
3048 /* Searching routines. */
3050 /* Like re_search_2, below, but only one string is specified, and
3051 doesn't let you say where to stop matching. */
3054 re_search (bufp
, string
, size
, startpos
, range
, regs
)
3055 struct re_pattern_buffer
*bufp
;
3057 int size
, startpos
, range
;
3058 struct re_registers
*regs
;
3060 return re_search_2 (bufp
, NULL
, 0, string
, size
, startpos
, range
,
3065 /* Using the compiled pattern in BUFP->buffer, first tries to match the
3066 virtual concatenation of STRING1 and STRING2, starting first at index
3067 STARTPOS, then at STARTPOS + 1, and so on.
3069 STRING1 and STRING2 have length SIZE1 and SIZE2, respectively.
3071 RANGE is how far to scan while trying to match. RANGE = 0 means try
3072 only at STARTPOS; in general, the last start tried is STARTPOS +
3075 In REGS, return the indices of the virtual concatenation of STRING1
3076 and STRING2 that matched the entire BUFP->buffer and its contained
3079 Do not consider matching one past the index STOP in the virtual
3080 concatenation of STRING1 and STRING2.
3082 We return either the position in the strings at which the match was
3083 found, -1 if no match, or -2 if error (such as failure
3087 re_search_2 (bufp
, string1
, size1
, string2
, size2
, startpos
, range
, regs
, stop
)
3088 struct re_pattern_buffer
*bufp
;
3089 const char *string1
, *string2
;
3093 struct re_registers
*regs
;
3097 register char *fastmap
= bufp
->fastmap
;
3098 register char *translate
= bufp
->translate
;
3099 int total_size
= size1
+ size2
;
3100 int endpos
= startpos
+ range
;
3102 /* Check for out-of-range STARTPOS. */
3103 if (startpos
< 0 || startpos
> total_size
)
3106 /* Fix up RANGE if it might eventually take us outside
3107 the virtual concatenation of STRING1 and STRING2. */
3109 range
= -1 - startpos
;
3110 else if (endpos
> total_size
)
3111 range
= total_size
- startpos
;
3113 /* If the search isn't to be a backwards one, don't waste time in a
3114 search for a pattern that must be anchored. */
3115 if (bufp
->used
> 0 && (re_opcode_t
) bufp
->buffer
[0] == begbuf
&& range
> 0)
3123 /* Update the fastmap now if not correct already. */
3124 if (fastmap
&& !bufp
->fastmap_accurate
)
3125 if (re_compile_fastmap (bufp
) == -2)
3128 /* Loop through the string, looking for a place to start matching. */
3131 /* If a fastmap is supplied, skip quickly over characters that
3132 cannot be the start of a match. If the pattern can match the
3133 null string, however, we don't need to skip characters; we want
3134 the first null string. */
3135 if (fastmap
&& startpos
< total_size
&& !bufp
->can_be_null
)
3137 if (range
> 0) /* Searching forwards. */
3139 register const char *d
;
3140 register int lim
= 0;
3143 if (startpos
< size1
&& startpos
+ range
>= size1
)
3144 lim
= range
- (size1
- startpos
);
3146 d
= (startpos
>= size1
? string2
- size1
: string1
) + startpos
;
3148 /* Written out as an if-else to avoid testing `translate'
3152 && !fastmap
[(unsigned char)
3153 translate
[(unsigned char) *d
++]])
3156 while (range
> lim
&& !fastmap
[(unsigned char) *d
++])
3159 startpos
+= irange
- range
;
3161 else /* Searching backwards. */
3163 register char c
= (size1
== 0 || startpos
>= size1
3164 ? string2
[startpos
- size1
]
3165 : string1
[startpos
]);
3167 if (!fastmap
[(unsigned char) TRANSLATE (c
)])
3172 /* If can't match the null string, and that's all we have left, fail. */
3173 if (range
>= 0 && startpos
== total_size
&& fastmap
3174 && !bufp
->can_be_null
)
3177 val
= re_match_2_internal (bufp
, string1
, size1
, string2
, size2
,
3178 startpos
, regs
, stop
);
3179 #ifndef REGEX_MALLOC
3208 /* Declarations and macros for re_match_2. */
3210 static int bcmp_translate ();
3211 static boolean
alt_match_null_string_p (),
3212 common_op_match_null_string_p (),
3213 group_match_null_string_p ();
3215 /* This converts PTR, a pointer into one of the search strings `string1'
3216 and `string2' into an offset from the beginning of that string. */
3217 #define POINTER_TO_OFFSET(ptr) \
3218 (FIRST_STRING_P (ptr) \
3219 ? ((regoff_t) ((ptr) - string1)) \
3220 : ((regoff_t) ((ptr) - string2 + size1)))
3222 /* Macros for dealing with the split strings in re_match_2. */
3224 #define MATCHING_IN_FIRST_STRING (dend == end_match_1)
3226 /* Call before fetching a character with *d. This switches over to
3227 string2 if necessary. */
3228 #define PREFETCH() \
3231 /* End of string2 => fail. */ \
3232 if (dend == end_match_2) \
3234 /* End of string1 => advance to string2. */ \
3236 dend = end_match_2; \
3240 /* Test if at very beginning or at very end of the virtual concatenation
3241 of `string1' and `string2'. If only one string, it's `string2'. */
3242 #define AT_STRINGS_BEG(d) ((d) == (size1 ? string1 : string2) || !size2)
3243 #define AT_STRINGS_END(d) ((d) == end2)
3246 /* Test if D points to a character which is word-constituent. We have
3247 two special cases to check for: if past the end of string1, look at
3248 the first character in string2; and if before the beginning of
3249 string2, look at the last character in string1. */
3250 #define WORDCHAR_P(d) \
3251 (SYNTAX ((d) == end1 ? *string2 \
3252 : (d) == string2 - 1 ? *(end1 - 1) : *(d)) \
3255 /* Test if the character before D and the one at D differ with respect
3256 to being word-constituent. */
3257 #define AT_WORD_BOUNDARY(d) \
3258 (AT_STRINGS_BEG (d) || AT_STRINGS_END (d) \
3259 || WORDCHAR_P (d - 1) != WORDCHAR_P (d))
3262 /* Free everything we malloc. */
3263 #ifdef MATCH_MAY_ALLOCATE
3265 #define FREE_VAR(var) if (var) free (var); var = NULL
3266 #define FREE_VARIABLES() \
3268 FREE_VAR (fail_stack.stack); \
3269 FREE_VAR (regstart); \
3270 FREE_VAR (regend); \
3271 FREE_VAR (old_regstart); \
3272 FREE_VAR (old_regend); \
3273 FREE_VAR (best_regstart); \
3274 FREE_VAR (best_regend); \
3275 FREE_VAR (reg_info); \
3276 FREE_VAR (reg_dummy); \
3277 FREE_VAR (reg_info_dummy); \
3279 #else /* not REGEX_MALLOC */
3280 /* This used to do alloca (0), but now we do that in the caller. */
3281 #define FREE_VARIABLES() /* Nothing */
3282 #endif /* not REGEX_MALLOC */
3284 #define FREE_VARIABLES() /* Do nothing! */
3285 #endif /* not MATCH_MAY_ALLOCATE */
3287 /* These values must meet several constraints. They must not be valid
3288 register values; since we have a limit of 255 registers (because
3289 we use only one byte in the pattern for the register number), we can
3290 use numbers larger than 255. They must differ by 1, because of
3291 NUM_FAILURE_ITEMS above. And the value for the lowest register must
3292 be larger than the value for the highest register, so we do not try
3293 to actually save any registers when none are active. */
3294 #define NO_HIGHEST_ACTIVE_REG (1 << BYTEWIDTH)
3295 #define NO_LOWEST_ACTIVE_REG (NO_HIGHEST_ACTIVE_REG + 1)
3297 /* Matching routines. */
3299 #ifndef emacs /* Emacs never uses this. */
3300 /* re_match is like re_match_2 except it takes only a single string. */
3303 re_match (bufp
, string
, size
, pos
, regs
)
3304 struct re_pattern_buffer
*bufp
;
3307 struct re_registers
*regs
;
3309 int result
= re_match_2_internal (bufp
, NULL
, 0, string
, size
,
3314 #endif /* not emacs */
3317 /* re_match_2 matches the compiled pattern in BUFP against the
3318 the (virtual) concatenation of STRING1 and STRING2 (of length SIZE1
3319 and SIZE2, respectively). We start matching at POS, and stop
3322 If REGS is non-null and the `no_sub' field of BUFP is nonzero, we
3323 store offsets for the substring each group matched in REGS. See the
3324 documentation for exactly how many groups we fill.
3326 We return -1 if no match, -2 if an internal error (such as the
3327 failure stack overflowing). Otherwise, we return the length of the
3328 matched substring. */
3331 re_match_2 (bufp
, string1
, size1
, string2
, size2
, pos
, regs
, stop
)
3332 struct re_pattern_buffer
*bufp
;
3333 const char *string1
, *string2
;
3336 struct re_registers
*regs
;
3339 int result
= re_match_2_internal (bufp
, string1
, size1
, string2
, size2
,
3345 /* This is a separate function so that we can force an alloca cleanup
3348 re_match_2_internal (bufp
, string1
, size1
, string2
, size2
, pos
, regs
, stop
)
3349 struct re_pattern_buffer
*bufp
;
3350 const char *string1
, *string2
;
3353 struct re_registers
*regs
;
3356 /* General temporaries. */
3360 /* Just past the end of the corresponding string. */
3361 const char *end1
, *end2
;
3363 /* Pointers into string1 and string2, just past the last characters in
3364 each to consider matching. */
3365 const char *end_match_1
, *end_match_2
;
3367 /* Where we are in the data, and the end of the current string. */
3368 const char *d
, *dend
;
3370 /* Where we are in the pattern, and the end of the pattern. */
3371 unsigned char *p
= bufp
->buffer
;
3372 register unsigned char *pend
= p
+ bufp
->used
;
3374 /* Mark the opcode just after a start_memory, so we can test for an
3375 empty subpattern when we get to the stop_memory. */
3376 unsigned char *just_past_start_mem
= 0;
3378 /* We use this to map every character in the string. */
3379 char *translate
= bufp
->translate
;
3381 /* Failure point stack. Each place that can handle a failure further
3382 down the line pushes a failure point on this stack. It consists of
3383 restart, regend, and reg_info for all registers corresponding to
3384 the subexpressions we're currently inside, plus the number of such
3385 registers, and, finally, two char *'s. The first char * is where
3386 to resume scanning the pattern; the second one is where to resume
3387 scanning the strings. If the latter is zero, the failure point is
3388 a ``dummy''; if a failure happens and the failure point is a dummy,
3389 it gets discarded and the next next one is tried. */
3390 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
3391 fail_stack_type fail_stack
;
3394 static unsigned failure_id
= 0;
3395 unsigned nfailure_points_pushed
= 0, nfailure_points_popped
= 0;
3398 /* We fill all the registers internally, independent of what we
3399 return, for use in backreferences. The number here includes
3400 an element for register zero. */
3401 unsigned num_regs
= bufp
->re_nsub
+ 1;
3403 /* The currently active registers. */
3404 unsigned lowest_active_reg
= NO_LOWEST_ACTIVE_REG
;
3405 unsigned highest_active_reg
= NO_HIGHEST_ACTIVE_REG
;
3407 /* Information on the contents of registers. These are pointers into
3408 the input strings; they record just what was matched (on this
3409 attempt) by a subexpression part of the pattern, that is, the
3410 regnum-th regstart pointer points to where in the pattern we began
3411 matching and the regnum-th regend points to right after where we
3412 stopped matching the regnum-th subexpression. (The zeroth register
3413 keeps track of what the whole pattern matches.) */
3414 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
3415 const char **regstart
, **regend
;
3418 /* If a group that's operated upon by a repetition operator fails to
3419 match anything, then the register for its start will need to be
3420 restored because it will have been set to wherever in the string we
3421 are when we last see its open-group operator. Similarly for a
3423 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
3424 const char **old_regstart
, **old_regend
;
3427 /* The is_active field of reg_info helps us keep track of which (possibly
3428 nested) subexpressions we are currently in. The matched_something
3429 field of reg_info[reg_num] helps us tell whether or not we have
3430 matched any of the pattern so far this time through the reg_num-th
3431 subexpression. These two fields get reset each time through any
3432 loop their register is in. */
3433 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
3434 register_info_type
*reg_info
;
3437 /* The following record the register info as found in the above
3438 variables when we find a match better than any we've seen before.
3439 This happens as we backtrack through the failure points, which in
3440 turn happens only if we have not yet matched the entire string. */
3441 unsigned best_regs_set
= false;
3442 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
3443 const char **best_regstart
, **best_regend
;
3446 /* Logically, this is `best_regend[0]'. But we don't want to have to
3447 allocate space for that if we're not allocating space for anything
3448 else (see below). Also, we never need info about register 0 for
3449 any of the other register vectors, and it seems rather a kludge to
3450 treat `best_regend' differently than the rest. So we keep track of
3451 the end of the best match so far in a separate variable. We
3452 initialize this to NULL so that when we backtrack the first time
3453 and need to test it, it's not garbage. */
3454 const char *match_end
= NULL
;
3456 /* Used when we pop values we don't care about. */
3457 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
3458 const char **reg_dummy
;
3459 register_info_type
*reg_info_dummy
;
3463 /* Counts the total number of registers pushed. */
3464 unsigned num_regs_pushed
= 0;
3467 DEBUG_PRINT1 ("\n\nEntering re_match_2.\n");
3471 #ifdef MATCH_MAY_ALLOCATE
3472 /* Do not bother to initialize all the register variables if there are
3473 no groups in the pattern, as it takes a fair amount of time. If
3474 there are groups, we include space for register 0 (the whole
3475 pattern), even though we never use it, since it simplifies the
3476 array indexing. We should fix this. */
3479 regstart
= REGEX_TALLOC (num_regs
, const char *);
3480 regend
= REGEX_TALLOC (num_regs
, const char *);
3481 old_regstart
= REGEX_TALLOC (num_regs
, const char *);
3482 old_regend
= REGEX_TALLOC (num_regs
, const char *);
3483 best_regstart
= REGEX_TALLOC (num_regs
, const char *);
3484 best_regend
= REGEX_TALLOC (num_regs
, const char *);
3485 reg_info
= REGEX_TALLOC (num_regs
, register_info_type
);
3486 reg_dummy
= REGEX_TALLOC (num_regs
, const char *);
3487 reg_info_dummy
= REGEX_TALLOC (num_regs
, register_info_type
);
3489 if (!(regstart
&& regend
&& old_regstart
&& old_regend
&& reg_info
3490 && best_regstart
&& best_regend
&& reg_dummy
&& reg_info_dummy
))
3496 #if defined (REGEX_MALLOC)
3499 /* We must initialize all our variables to NULL, so that
3500 `FREE_VARIABLES' doesn't try to free them. */
3501 regstart
= regend
= old_regstart
= old_regend
= best_regstart
3502 = best_regend
= reg_dummy
= NULL
;
3503 reg_info
= reg_info_dummy
= (register_info_type
*) NULL
;
3505 #endif /* REGEX_MALLOC */
3506 #endif /* MATCH_MAY_ALLOCATE */
3508 /* The starting position is bogus. */
3509 if (pos
< 0 || pos
> size1
+ size2
)
3515 /* Initialize subexpression text positions to -1 to mark ones that no
3516 start_memory/stop_memory has been seen for. Also initialize the
3517 register information struct. */
3518 for (mcnt
= 1; mcnt
< num_regs
; mcnt
++)
3520 regstart
[mcnt
] = regend
[mcnt
]
3521 = old_regstart
[mcnt
] = old_regend
[mcnt
] = REG_UNSET_VALUE
;
3523 REG_MATCH_NULL_STRING_P (reg_info
[mcnt
]) = MATCH_NULL_UNSET_VALUE
;
3524 IS_ACTIVE (reg_info
[mcnt
]) = 0;
3525 MATCHED_SOMETHING (reg_info
[mcnt
]) = 0;
3526 EVER_MATCHED_SOMETHING (reg_info
[mcnt
]) = 0;
3529 /* We move `string1' into `string2' if the latter's empty -- but not if
3530 `string1' is null. */
3531 if (size2
== 0 && string1
!= NULL
)
3538 end1
= string1
+ size1
;
3539 end2
= string2
+ size2
;
3541 /* Compute where to stop matching, within the two strings. */
3544 end_match_1
= string1
+ stop
;
3545 end_match_2
= string2
;
3550 end_match_2
= string2
+ stop
- size1
;
3553 /* `p' scans through the pattern as `d' scans through the data.
3554 `dend' is the end of the input string that `d' points within. `d'
3555 is advanced into the following input string whenever necessary, but
3556 this happens before fetching; therefore, at the beginning of the
3557 loop, `d' can be pointing at the end of a string, but it cannot
3559 if (size1
> 0 && pos
<= size1
)
3566 d
= string2
+ pos
- size1
;
3570 DEBUG_PRINT1 ("The compiled pattern is: ");
3571 DEBUG_PRINT_COMPILED_PATTERN (bufp
, p
, pend
);
3572 DEBUG_PRINT1 ("The string to match is: `");
3573 DEBUG_PRINT_DOUBLE_STRING (d
, string1
, size1
, string2
, size2
);
3574 DEBUG_PRINT1 ("'\n");
3576 /* This loops over pattern commands. It exits by returning from the
3577 function if the match is complete, or it drops through if the match
3578 fails at this starting point in the input data. */
3581 DEBUG_PRINT2 ("\n0x%x: ", p
);
3584 { /* End of pattern means we might have succeeded. */
3585 DEBUG_PRINT1 ("end of pattern ... ");
3587 /* If we haven't matched the entire string, and we want the
3588 longest match, try backtracking. */
3589 if (d
!= end_match_2
)
3591 /* 1 if this match ends in the same string (string1 or string2)
3592 as the best previous match. */
3593 boolean same_str_p
= (FIRST_STRING_P (match_end
)
3594 == MATCHING_IN_FIRST_STRING
);
3595 /* 1 if this match is the best seen so far. */
3596 boolean best_match_p
;
3598 /* AIX compiler got confused when this was combined
3599 with the previous declaration. */
3601 best_match_p
= d
> match_end
;
3603 best_match_p
= !MATCHING_IN_FIRST_STRING
;
3605 DEBUG_PRINT1 ("backtracking.\n");
3607 if (!FAIL_STACK_EMPTY ())
3608 { /* More failure points to try. */
3610 /* If exceeds best match so far, save it. */
3611 if (!best_regs_set
|| best_match_p
)
3613 best_regs_set
= true;
3616 DEBUG_PRINT1 ("\nSAVING match as best so far.\n");
3618 for (mcnt
= 1; mcnt
< num_regs
; mcnt
++)
3620 best_regstart
[mcnt
] = regstart
[mcnt
];
3621 best_regend
[mcnt
] = regend
[mcnt
];
3627 /* If no failure points, don't restore garbage. And if
3628 last match is real best match, don't restore second
3630 else if (best_regs_set
&& !best_match_p
)
3633 /* Restore best match. It may happen that `dend ==
3634 end_match_1' while the restored d is in string2.
3635 For example, the pattern `x.*y.*z' against the
3636 strings `x-' and `y-z-', if the two strings are
3637 not consecutive in memory. */
3638 DEBUG_PRINT1 ("Restoring best registers.\n");
3641 dend
= ((d
>= string1
&& d
<= end1
)
3642 ? end_match_1
: end_match_2
);
3644 for (mcnt
= 1; mcnt
< num_regs
; mcnt
++)
3646 regstart
[mcnt
] = best_regstart
[mcnt
];
3647 regend
[mcnt
] = best_regend
[mcnt
];
3650 } /* d != end_match_2 */
3652 DEBUG_PRINT1 ("Accepting match.\n");
3654 /* If caller wants register contents data back, do it. */
3655 if (regs
&& !bufp
->no_sub
)
3657 /* Have the register data arrays been allocated? */
3658 if (bufp
->regs_allocated
== REGS_UNALLOCATED
)
3659 { /* No. So allocate them with malloc. We need one
3660 extra element beyond `num_regs' for the `-1' marker
3662 regs
->num_regs
= MAX (RE_NREGS
, num_regs
+ 1);
3663 regs
->start
= TALLOC (regs
->num_regs
, regoff_t
);
3664 regs
->end
= TALLOC (regs
->num_regs
, regoff_t
);
3665 if (regs
->start
== NULL
|| regs
->end
== NULL
)
3667 bufp
->regs_allocated
= REGS_REALLOCATE
;
3669 else if (bufp
->regs_allocated
== REGS_REALLOCATE
)
3670 { /* Yes. If we need more elements than were already
3671 allocated, reallocate them. If we need fewer, just
3673 if (regs
->num_regs
< num_regs
+ 1)
3675 regs
->num_regs
= num_regs
+ 1;
3676 RETALLOC (regs
->start
, regs
->num_regs
, regoff_t
);
3677 RETALLOC (regs
->end
, regs
->num_regs
, regoff_t
);
3678 if (regs
->start
== NULL
|| regs
->end
== NULL
)
3684 /* These braces fend off a "empty body in an else-statement"
3685 warning under GCC when assert expands to nothing. */
3686 assert (bufp
->regs_allocated
== REGS_FIXED
);
3689 /* Convert the pointer data in `regstart' and `regend' to
3690 indices. Register zero has to be set differently,
3691 since we haven't kept track of any info for it. */
3692 if (regs
->num_regs
> 0)
3694 regs
->start
[0] = pos
;
3695 regs
->end
[0] = (MATCHING_IN_FIRST_STRING
3696 ? ((regoff_t
) (d
- string1
))
3697 : ((regoff_t
) (d
- string2
+ size1
)));
3700 /* Go through the first `min (num_regs, regs->num_regs)'
3701 registers, since that is all we initialized. */
3702 for (mcnt
= 1; mcnt
< MIN (num_regs
, regs
->num_regs
); mcnt
++)
3704 if (REG_UNSET (regstart
[mcnt
]) || REG_UNSET (regend
[mcnt
]))
3705 regs
->start
[mcnt
] = regs
->end
[mcnt
] = -1;
3709 = (regoff_t
) POINTER_TO_OFFSET (regstart
[mcnt
]);
3711 = (regoff_t
) POINTER_TO_OFFSET (regend
[mcnt
]);
3715 /* If the regs structure we return has more elements than
3716 were in the pattern, set the extra elements to -1. If
3717 we (re)allocated the registers, this is the case,
3718 because we always allocate enough to have at least one
3720 for (mcnt
= num_regs
; mcnt
< regs
->num_regs
; mcnt
++)
3721 regs
->start
[mcnt
] = regs
->end
[mcnt
] = -1;
3722 } /* regs && !bufp->no_sub */
3725 DEBUG_PRINT4 ("%u failure points pushed, %u popped (%u remain).\n",
3726 nfailure_points_pushed
, nfailure_points_popped
,
3727 nfailure_points_pushed
- nfailure_points_popped
);
3728 DEBUG_PRINT2 ("%u registers pushed.\n", num_regs_pushed
);
3730 mcnt
= d
- pos
- (MATCHING_IN_FIRST_STRING
3734 DEBUG_PRINT2 ("Returning %d from re_match_2.\n", mcnt
);
3739 /* Otherwise match next pattern command. */
3740 #ifdef SWITCH_ENUM_BUG
3741 switch ((int) ((re_opcode_t
) *p
++))
3743 switch ((re_opcode_t
) *p
++)
3746 /* Ignore these. Used to ignore the n of succeed_n's which
3747 currently have n == 0. */
3749 DEBUG_PRINT1 ("EXECUTING no_op.\n");
3753 /* Match the next n pattern characters exactly. The following
3754 byte in the pattern defines n, and the n bytes after that
3755 are the characters to match. */
3758 DEBUG_PRINT2 ("EXECUTING exactn %d.\n", mcnt
);
3760 /* This is written out as an if-else so we don't waste time
3761 testing `translate' inside the loop. */
3767 if (translate
[(unsigned char) *d
++] != (char) *p
++)
3777 if (*d
++ != (char) *p
++) goto fail
;
3781 SET_REGS_MATCHED ();
3785 /* Match any character except possibly a newline or a null. */
3787 DEBUG_PRINT1 ("EXECUTING anychar.\n");
3791 if ((!(bufp
->syntax
& RE_DOT_NEWLINE
) && TRANSLATE (*d
) == '\n')
3792 || (bufp
->syntax
& RE_DOT_NOT_NULL
&& TRANSLATE (*d
) == '\000'))
3795 SET_REGS_MATCHED ();
3796 DEBUG_PRINT2 (" Matched `%d'.\n", *d
);
3804 register unsigned char c
;
3805 boolean
not = (re_opcode_t
) *(p
- 1) == charset_not
;
3807 DEBUG_PRINT2 ("EXECUTING charset%s.\n", not ? "_not" : "");
3810 c
= TRANSLATE (*d
); /* The character to match. */
3812 /* Cast to `unsigned' instead of `unsigned char' in case the
3813 bit list is a full 32 bytes long. */
3814 if (c
< (unsigned) (*p
* BYTEWIDTH
)
3815 && p
[1 + c
/ BYTEWIDTH
] & (1 << (c
% BYTEWIDTH
)))
3820 if (!not) goto fail
;
3822 SET_REGS_MATCHED ();
3828 /* The beginning of a group is represented by start_memory.
3829 The arguments are the register number in the next byte, and the
3830 number of groups inner to this one in the next. The text
3831 matched within the group is recorded (in the internal
3832 registers data structure) under the register number. */
3834 DEBUG_PRINT3 ("EXECUTING start_memory %d (%d):\n", *p
, p
[1]);
3836 /* Find out if this group can match the empty string. */
3837 p1
= p
; /* To send to group_match_null_string_p. */
3839 if (REG_MATCH_NULL_STRING_P (reg_info
[*p
]) == MATCH_NULL_UNSET_VALUE
)
3840 REG_MATCH_NULL_STRING_P (reg_info
[*p
])
3841 = group_match_null_string_p (&p1
, pend
, reg_info
);
3843 /* Save the position in the string where we were the last time
3844 we were at this open-group operator in case the group is
3845 operated upon by a repetition operator, e.g., with `(a*)*b'
3846 against `ab'; then we want to ignore where we are now in
3847 the string in case this attempt to match fails. */
3848 old_regstart
[*p
] = REG_MATCH_NULL_STRING_P (reg_info
[*p
])
3849 ? REG_UNSET (regstart
[*p
]) ? d
: regstart
[*p
]
3851 DEBUG_PRINT2 (" old_regstart: %d\n",
3852 POINTER_TO_OFFSET (old_regstart
[*p
]));
3855 DEBUG_PRINT2 (" regstart: %d\n", POINTER_TO_OFFSET (regstart
[*p
]));
3857 IS_ACTIVE (reg_info
[*p
]) = 1;
3858 MATCHED_SOMETHING (reg_info
[*p
]) = 0;
3860 /* This is the new highest active register. */
3861 highest_active_reg
= *p
;
3863 /* If nothing was active before, this is the new lowest active
3865 if (lowest_active_reg
== NO_LOWEST_ACTIVE_REG
)
3866 lowest_active_reg
= *p
;
3868 /* Move past the register number and inner group count. */
3870 just_past_start_mem
= p
;
3874 /* The stop_memory opcode represents the end of a group. Its
3875 arguments are the same as start_memory's: the register
3876 number, and the number of inner groups. */
3878 DEBUG_PRINT3 ("EXECUTING stop_memory %d (%d):\n", *p
, p
[1]);
3880 /* We need to save the string position the last time we were at
3881 this close-group operator in case the group is operated
3882 upon by a repetition operator, e.g., with `((a*)*(b*)*)*'
3883 against `aba'; then we want to ignore where we are now in
3884 the string in case this attempt to match fails. */
3885 old_regend
[*p
] = REG_MATCH_NULL_STRING_P (reg_info
[*p
])
3886 ? REG_UNSET (regend
[*p
]) ? d
: regend
[*p
]
3888 DEBUG_PRINT2 (" old_regend: %d\n",
3889 POINTER_TO_OFFSET (old_regend
[*p
]));
3892 DEBUG_PRINT2 (" regend: %d\n", POINTER_TO_OFFSET (regend
[*p
]));
3894 /* This register isn't active anymore. */
3895 IS_ACTIVE (reg_info
[*p
]) = 0;
3897 /* If this was the only register active, nothing is active
3899 if (lowest_active_reg
== highest_active_reg
)
3901 lowest_active_reg
= NO_LOWEST_ACTIVE_REG
;
3902 highest_active_reg
= NO_HIGHEST_ACTIVE_REG
;
3905 { /* We must scan for the new highest active register, since
3906 it isn't necessarily one less than now: consider
3907 (a(b)c(d(e)f)g). When group 3 ends, after the f), the
3908 new highest active register is 1. */
3909 unsigned char r
= *p
- 1;
3910 while (r
> 0 && !IS_ACTIVE (reg_info
[r
]))
3913 /* If we end up at register zero, that means that we saved
3914 the registers as the result of an `on_failure_jump', not
3915 a `start_memory', and we jumped to past the innermost
3916 `stop_memory'. For example, in ((.)*) we save
3917 registers 1 and 2 as a result of the *, but when we pop
3918 back to the second ), we are at the stop_memory 1.
3919 Thus, nothing is active. */
3922 lowest_active_reg
= NO_LOWEST_ACTIVE_REG
;
3923 highest_active_reg
= NO_HIGHEST_ACTIVE_REG
;
3926 highest_active_reg
= r
;
3929 /* If just failed to match something this time around with a
3930 group that's operated on by a repetition operator, try to
3931 force exit from the ``loop'', and restore the register
3932 information for this group that we had before trying this
3934 if ((!MATCHED_SOMETHING (reg_info
[*p
])
3935 || just_past_start_mem
== p
- 1)
3938 boolean is_a_jump_n
= false;
3942 switch ((re_opcode_t
) *p1
++)
3946 case pop_failure_jump
:
3947 case maybe_pop_jump
:
3949 case dummy_failure_jump
:
3950 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
3960 /* If the next operation is a jump backwards in the pattern
3961 to an on_failure_jump right before the start_memory
3962 corresponding to this stop_memory, exit from the loop
3963 by forcing a failure after pushing on the stack the
3964 on_failure_jump's jump in the pattern, and d. */
3965 if (mcnt
< 0 && (re_opcode_t
) *p1
== on_failure_jump
3966 && (re_opcode_t
) p1
[3] == start_memory
&& p1
[4] == *p
)
3968 /* If this group ever matched anything, then restore
3969 what its registers were before trying this last
3970 failed match, e.g., with `(a*)*b' against `ab' for
3971 regstart[1], and, e.g., with `((a*)*(b*)*)*'
3972 against `aba' for regend[3].
3974 Also restore the registers for inner groups for,
3975 e.g., `((a*)(b*))*' against `aba' (register 3 would
3976 otherwise get trashed). */
3978 if (EVER_MATCHED_SOMETHING (reg_info
[*p
]))
3982 EVER_MATCHED_SOMETHING (reg_info
[*p
]) = 0;
3984 /* Restore this and inner groups' (if any) registers. */
3985 for (r
= *p
; r
< *p
+ *(p
+ 1); r
++)
3987 regstart
[r
] = old_regstart
[r
];
3989 /* xx why this test? */
3990 if ((int) old_regend
[r
] >= (int) regstart
[r
])
3991 regend
[r
] = old_regend
[r
];
3995 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
3996 PUSH_FAILURE_POINT (p1
+ mcnt
, d
, -2);
4002 /* Move past the register number and the inner group count. */
4007 /* \<digit> has been turned into a `duplicate' command which is
4008 followed by the numeric value of <digit> as the register number. */
4011 register const char *d2
, *dend2
;
4012 int regno
= *p
++; /* Get which register to match against. */
4013 DEBUG_PRINT2 ("EXECUTING duplicate %d.\n", regno
);
4015 /* Can't back reference a group which we've never matched. */
4016 if (REG_UNSET (regstart
[regno
]) || REG_UNSET (regend
[regno
]))
4019 /* Where in input to try to start matching. */
4020 d2
= regstart
[regno
];
4022 /* Where to stop matching; if both the place to start and
4023 the place to stop matching are in the same string, then
4024 set to the place to stop, otherwise, for now have to use
4025 the end of the first string. */
4027 dend2
= ((FIRST_STRING_P (regstart
[regno
])
4028 == FIRST_STRING_P (regend
[regno
]))
4029 ? regend
[regno
] : end_match_1
);
4032 /* If necessary, advance to next segment in register
4036 if (dend2
== end_match_2
) break;
4037 if (dend2
== regend
[regno
]) break;
4039 /* End of string1 => advance to string2. */
4041 dend2
= regend
[regno
];
4043 /* At end of register contents => success */
4044 if (d2
== dend2
) break;
4046 /* If necessary, advance to next segment in data. */
4049 /* How many characters left in this segment to match. */
4052 /* Want how many consecutive characters we can match in
4053 one shot, so, if necessary, adjust the count. */
4054 if (mcnt
> dend2
- d2
)
4057 /* Compare that many; failure if mismatch, else move
4060 ? bcmp_translate (d
, d2
, mcnt
, translate
)
4061 : bcmp (d
, d2
, mcnt
))
4063 d
+= mcnt
, d2
+= mcnt
;
4069 /* begline matches the empty string at the beginning of the string
4070 (unless `not_bol' is set in `bufp'), and, if
4071 `newline_anchor' is set, after newlines. */
4073 DEBUG_PRINT1 ("EXECUTING begline.\n");
4075 if (AT_STRINGS_BEG (d
))
4077 if (!bufp
->not_bol
) break;
4079 else if (d
[-1] == '\n' && bufp
->newline_anchor
)
4083 /* In all other cases, we fail. */
4087 /* endline is the dual of begline. */
4089 DEBUG_PRINT1 ("EXECUTING endline.\n");
4091 if (AT_STRINGS_END (d
))
4093 if (!bufp
->not_eol
) break;
4096 /* We have to ``prefetch'' the next character. */
4097 else if ((d
== end1
? *string2
: *d
) == '\n'
4098 && bufp
->newline_anchor
)
4105 /* Match at the very beginning of the data. */
4107 DEBUG_PRINT1 ("EXECUTING begbuf.\n");
4108 if (AT_STRINGS_BEG (d
))
4113 /* Match at the very end of the data. */
4115 DEBUG_PRINT1 ("EXECUTING endbuf.\n");
4116 if (AT_STRINGS_END (d
))
4121 /* on_failure_keep_string_jump is used to optimize `.*\n'. It
4122 pushes NULL as the value for the string on the stack. Then
4123 `pop_failure_point' will keep the current value for the
4124 string, instead of restoring it. To see why, consider
4125 matching `foo\nbar' against `.*\n'. The .* matches the foo;
4126 then the . fails against the \n. But the next thing we want
4127 to do is match the \n against the \n; if we restored the
4128 string value, we would be back at the foo.
4130 Because this is used only in specific cases, we don't need to
4131 check all the things that `on_failure_jump' does, to make
4132 sure the right things get saved on the stack. Hence we don't
4133 share its code. The only reason to push anything on the
4134 stack at all is that otherwise we would have to change
4135 `anychar's code to do something besides goto fail in this
4136 case; that seems worse than this. */
4137 case on_failure_keep_string_jump
:
4138 DEBUG_PRINT1 ("EXECUTING on_failure_keep_string_jump");
4140 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
4141 DEBUG_PRINT3 (" %d (to 0x%x):\n", mcnt
, p
+ mcnt
);
4143 PUSH_FAILURE_POINT (p
+ mcnt
, NULL
, -2);
4147 /* Uses of on_failure_jump:
4149 Each alternative starts with an on_failure_jump that points
4150 to the beginning of the next alternative. Each alternative
4151 except the last ends with a jump that in effect jumps past
4152 the rest of the alternatives. (They really jump to the
4153 ending jump of the following alternative, because tensioning
4154 these jumps is a hassle.)
4156 Repeats start with an on_failure_jump that points past both
4157 the repetition text and either the following jump or
4158 pop_failure_jump back to this on_failure_jump. */
4159 case on_failure_jump
:
4161 DEBUG_PRINT1 ("EXECUTING on_failure_jump");
4163 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
4164 DEBUG_PRINT3 (" %d (to 0x%x)", mcnt
, p
+ mcnt
);
4166 /* If this on_failure_jump comes right before a group (i.e.,
4167 the original * applied to a group), save the information
4168 for that group and all inner ones, so that if we fail back
4169 to this point, the group's information will be correct.
4170 For example, in \(a*\)*\1, we need the preceding group,
4171 and in \(\(a*\)b*\)\2, we need the inner group. */
4173 /* We can't use `p' to check ahead because we push
4174 a failure point to `p + mcnt' after we do this. */
4177 /* We need to skip no_op's before we look for the
4178 start_memory in case this on_failure_jump is happening as
4179 the result of a completed succeed_n, as in \(a\)\{1,3\}b\1
4181 while (p1
< pend
&& (re_opcode_t
) *p1
== no_op
)
4184 if (p1
< pend
&& (re_opcode_t
) *p1
== start_memory
)
4186 /* We have a new highest active register now. This will
4187 get reset at the start_memory we are about to get to,
4188 but we will have saved all the registers relevant to
4189 this repetition op, as described above. */
4190 highest_active_reg
= *(p1
+ 1) + *(p1
+ 2);
4191 if (lowest_active_reg
== NO_LOWEST_ACTIVE_REG
)
4192 lowest_active_reg
= *(p1
+ 1);
4195 DEBUG_PRINT1 (":\n");
4196 PUSH_FAILURE_POINT (p
+ mcnt
, d
, -2);
4200 /* A smart repeat ends with `maybe_pop_jump'.
4201 We change it to either `pop_failure_jump' or `jump'. */
4202 case maybe_pop_jump
:
4203 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
4204 DEBUG_PRINT2 ("EXECUTING maybe_pop_jump %d.\n", mcnt
);
4206 register unsigned char *p2
= p
;
4208 /* Compare the beginning of the repeat with what in the
4209 pattern follows its end. If we can establish that there
4210 is nothing that they would both match, i.e., that we
4211 would have to backtrack because of (as in, e.g., `a*a')
4212 then we can change to pop_failure_jump, because we'll
4213 never have to backtrack.
4215 This is not true in the case of alternatives: in
4216 `(a|ab)*' we do need to backtrack to the `ab' alternative
4217 (e.g., if the string was `ab'). But instead of trying to
4218 detect that here, the alternative has put on a dummy
4219 failure point which is what we will end up popping. */
4221 /* Skip over open/close-group commands.
4222 If what follows this loop is a ...+ construct,
4223 look at what begins its body, since we will have to
4224 match at least one of that. */
4228 && ((re_opcode_t
) *p2
== stop_memory
4229 || (re_opcode_t
) *p2
== start_memory
))
4231 else if (p2
+ 6 < pend
4232 && (re_opcode_t
) *p2
== dummy_failure_jump
)
4239 /* p1[0] ... p1[2] are the `on_failure_jump' corresponding
4240 to the `maybe_finalize_jump' of this case. Examine what
4243 /* If we're at the end of the pattern, we can change. */
4246 /* Consider what happens when matching ":\(.*\)"
4247 against ":/". I don't really understand this code
4249 p
[-3] = (unsigned char) pop_failure_jump
;
4251 (" End of pattern: change to `pop_failure_jump'.\n");
4254 else if ((re_opcode_t
) *p2
== exactn
4255 || (bufp
->newline_anchor
&& (re_opcode_t
) *p2
== endline
))
4257 register unsigned char c
4258 = *p2
== (unsigned char) endline
? '\n' : p2
[2];
4260 if ((re_opcode_t
) p1
[3] == exactn
&& p1
[5] != c
)
4262 p
[-3] = (unsigned char) pop_failure_jump
;
4263 DEBUG_PRINT3 (" %c != %c => pop_failure_jump.\n",
4267 else if ((re_opcode_t
) p1
[3] == charset
4268 || (re_opcode_t
) p1
[3] == charset_not
)
4270 int not = (re_opcode_t
) p1
[3] == charset_not
;
4272 if (c
< (unsigned char) (p1
[4] * BYTEWIDTH
)
4273 && p1
[5 + c
/ BYTEWIDTH
] & (1 << (c
% BYTEWIDTH
)))
4276 /* `not' is equal to 1 if c would match, which means
4277 that we can't change to pop_failure_jump. */
4280 p
[-3] = (unsigned char) pop_failure_jump
;
4281 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
4285 else if ((re_opcode_t
) *p2
== charset
)
4288 register unsigned char c
4289 = *p2
== (unsigned char) endline
? '\n' : p2
[2];
4292 if ((re_opcode_t
) p1
[3] == exactn
4293 && ! ((int) p2
[1] * BYTEWIDTH
> (int) p1
[4]
4294 && (p2
[1 + p1
[4] / BYTEWIDTH
]
4295 & (1 << (p1
[4] % BYTEWIDTH
)))))
4297 p
[-3] = (unsigned char) pop_failure_jump
;
4298 DEBUG_PRINT3 (" %c != %c => pop_failure_jump.\n",
4302 else if ((re_opcode_t
) p1
[3] == charset_not
)
4305 /* We win if the charset_not inside the loop
4306 lists every character listed in the charset after. */
4307 for (idx
= 0; idx
< (int) p2
[1]; idx
++)
4308 if (! (p2
[2 + idx
] == 0
4309 || (idx
< (int) p1
[4]
4310 && ((p2
[2 + idx
] & ~ p1
[5 + idx
]) == 0))))
4315 p
[-3] = (unsigned char) pop_failure_jump
;
4316 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
4319 else if ((re_opcode_t
) p1
[3] == charset
)
4322 /* We win if the charset inside the loop
4323 has no overlap with the one after the loop. */
4325 idx
< (int) p2
[1] && idx
< (int) p1
[4];
4327 if ((p2
[2 + idx
] & p1
[5 + idx
]) != 0)
4330 if (idx
== p2
[1] || idx
== p1
[4])
4332 p
[-3] = (unsigned char) pop_failure_jump
;
4333 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
4338 p
-= 2; /* Point at relative address again. */
4339 if ((re_opcode_t
) p
[-1] != pop_failure_jump
)
4341 p
[-1] = (unsigned char) jump
;
4342 DEBUG_PRINT1 (" Match => jump.\n");
4343 goto unconditional_jump
;
4345 /* Note fall through. */
4348 /* The end of a simple repeat has a pop_failure_jump back to
4349 its matching on_failure_jump, where the latter will push a
4350 failure point. The pop_failure_jump takes off failure
4351 points put on by this pop_failure_jump's matching
4352 on_failure_jump; we got through the pattern to here from the
4353 matching on_failure_jump, so didn't fail. */
4354 case pop_failure_jump
:
4356 /* We need to pass separate storage for the lowest and
4357 highest registers, even though we don't care about the
4358 actual values. Otherwise, we will restore only one
4359 register from the stack, since lowest will == highest in
4360 `pop_failure_point'. */
4361 unsigned dummy_low_reg
, dummy_high_reg
;
4362 unsigned char *pdummy
;
4365 DEBUG_PRINT1 ("EXECUTING pop_failure_jump.\n");
4366 POP_FAILURE_POINT (sdummy
, pdummy
,
4367 dummy_low_reg
, dummy_high_reg
,
4368 reg_dummy
, reg_dummy
, reg_info_dummy
);
4370 /* Note fall through. */
4373 /* Unconditionally jump (without popping any failure points). */
4376 EXTRACT_NUMBER_AND_INCR (mcnt
, p
); /* Get the amount to jump. */
4377 DEBUG_PRINT2 ("EXECUTING jump %d ", mcnt
);
4378 p
+= mcnt
; /* Do the jump. */
4379 DEBUG_PRINT2 ("(to 0x%x).\n", p
);
4383 /* We need this opcode so we can detect where alternatives end
4384 in `group_match_null_string_p' et al. */
4386 DEBUG_PRINT1 ("EXECUTING jump_past_alt.\n");
4387 goto unconditional_jump
;
4390 /* Normally, the on_failure_jump pushes a failure point, which
4391 then gets popped at pop_failure_jump. We will end up at
4392 pop_failure_jump, also, and with a pattern of, say, `a+', we
4393 are skipping over the on_failure_jump, so we have to push
4394 something meaningless for pop_failure_jump to pop. */
4395 case dummy_failure_jump
:
4396 DEBUG_PRINT1 ("EXECUTING dummy_failure_jump.\n");
4397 /* It doesn't matter what we push for the string here. What
4398 the code at `fail' tests is the value for the pattern. */
4399 PUSH_FAILURE_POINT (0, 0, -2);
4400 goto unconditional_jump
;
4403 /* At the end of an alternative, we need to push a dummy failure
4404 point in case we are followed by a `pop_failure_jump', because
4405 we don't want the failure point for the alternative to be
4406 popped. For example, matching `(a|ab)*' against `aab'
4407 requires that we match the `ab' alternative. */
4408 case push_dummy_failure
:
4409 DEBUG_PRINT1 ("EXECUTING push_dummy_failure.\n");
4410 /* See comments just above at `dummy_failure_jump' about the
4412 PUSH_FAILURE_POINT (0, 0, -2);
4415 /* Have to succeed matching what follows at least n times.
4416 After that, handle like `on_failure_jump'. */
4418 EXTRACT_NUMBER (mcnt
, p
+ 2);
4419 DEBUG_PRINT2 ("EXECUTING succeed_n %d.\n", mcnt
);
4422 /* Originally, this is how many times we HAVE to succeed. */
4427 STORE_NUMBER_AND_INCR (p
, mcnt
);
4428 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p
, mcnt
);
4432 DEBUG_PRINT2 (" Setting two bytes from 0x%x to no_op.\n", p
+2);
4433 p
[2] = (unsigned char) no_op
;
4434 p
[3] = (unsigned char) no_op
;
4440 EXTRACT_NUMBER (mcnt
, p
+ 2);
4441 DEBUG_PRINT2 ("EXECUTING jump_n %d.\n", mcnt
);
4443 /* Originally, this is how many times we CAN jump. */
4447 STORE_NUMBER (p
+ 2, mcnt
);
4448 goto unconditional_jump
;
4450 /* If don't have to jump any more, skip over the rest of command. */
4457 DEBUG_PRINT1 ("EXECUTING set_number_at.\n");
4459 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
4461 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
4462 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p1
, mcnt
);
4463 STORE_NUMBER (p1
, mcnt
);
4468 DEBUG_PRINT1 ("EXECUTING wordbound.\n");
4469 if (AT_WORD_BOUNDARY (d
))
4474 DEBUG_PRINT1 ("EXECUTING notwordbound.\n");
4475 if (AT_WORD_BOUNDARY (d
))
4480 DEBUG_PRINT1 ("EXECUTING wordbeg.\n");
4481 if (WORDCHAR_P (d
) && (AT_STRINGS_BEG (d
) || !WORDCHAR_P (d
- 1)))
4486 DEBUG_PRINT1 ("EXECUTING wordend.\n");
4487 if (!AT_STRINGS_BEG (d
) && WORDCHAR_P (d
- 1)
4488 && (!WORDCHAR_P (d
) || AT_STRINGS_END (d
)))
4494 DEBUG_PRINT1 ("EXECUTING before_dot.\n");
4495 if (PTR_CHAR_POS ((unsigned char *) d
) >= point
)
4500 DEBUG_PRINT1 ("EXECUTING at_dot.\n");
4501 if (PTR_CHAR_POS ((unsigned char *) d
) != point
)
4506 DEBUG_PRINT1 ("EXECUTING after_dot.\n");
4507 if (PTR_CHAR_POS ((unsigned char *) d
) <= point
)
4510 #if 0 /* not emacs19 */
4512 DEBUG_PRINT1 ("EXECUTING at_dot.\n");
4513 if (PTR_CHAR_POS ((unsigned char *) d
) + 1 != point
)
4516 #endif /* not emacs19 */
4519 DEBUG_PRINT2 ("EXECUTING syntaxspec %d.\n", mcnt
);
4524 DEBUG_PRINT1 ("EXECUTING Emacs wordchar.\n");
4528 /* Can't use *d++ here; SYNTAX may be an unsafe macro. */
4530 if (SYNTAX (d
[-1]) != (enum syntaxcode
) mcnt
)
4532 SET_REGS_MATCHED ();
4536 DEBUG_PRINT2 ("EXECUTING notsyntaxspec %d.\n", mcnt
);
4538 goto matchnotsyntax
;
4541 DEBUG_PRINT1 ("EXECUTING Emacs notwordchar.\n");
4545 /* Can't use *d++ here; SYNTAX may be an unsafe macro. */
4547 if (SYNTAX (d
[-1]) == (enum syntaxcode
) mcnt
)
4549 SET_REGS_MATCHED ();
4552 #else /* not emacs */
4554 DEBUG_PRINT1 ("EXECUTING non-Emacs wordchar.\n");
4556 if (!WORDCHAR_P (d
))
4558 SET_REGS_MATCHED ();
4563 DEBUG_PRINT1 ("EXECUTING non-Emacs notwordchar.\n");
4567 SET_REGS_MATCHED ();
4570 #endif /* not emacs */
4575 continue; /* Successfully executed one pattern command; keep going. */
4578 /* We goto here if a matching operation fails. */
4580 if (!FAIL_STACK_EMPTY ())
4581 { /* A restart point is known. Restore to that state. */
4582 DEBUG_PRINT1 ("\nFAIL:\n");
4583 POP_FAILURE_POINT (d
, p
,
4584 lowest_active_reg
, highest_active_reg
,
4585 regstart
, regend
, reg_info
);
4587 /* If this failure point is a dummy, try the next one. */
4591 /* If we failed to the end of the pattern, don't examine *p. */
4595 boolean is_a_jump_n
= false;
4597 /* If failed to a backwards jump that's part of a repetition
4598 loop, need to pop this failure point and use the next one. */
4599 switch ((re_opcode_t
) *p
)
4603 case maybe_pop_jump
:
4604 case pop_failure_jump
:
4607 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
4610 if ((is_a_jump_n
&& (re_opcode_t
) *p1
== succeed_n
)
4612 && (re_opcode_t
) *p1
== on_failure_jump
))
4620 if (d
>= string1
&& d
<= end1
)
4624 break; /* Matching at this starting point really fails. */
4628 goto restore_best_regs
;
4632 return -1; /* Failure to match. */
4635 /* Subroutine definitions for re_match_2. */
4638 /* We are passed P pointing to a register number after a start_memory.
4640 Return true if the pattern up to the corresponding stop_memory can
4641 match the empty string, and false otherwise.
4643 If we find the matching stop_memory, sets P to point to one past its number.
4644 Otherwise, sets P to an undefined byte less than or equal to END.
4646 We don't handle duplicates properly (yet). */
4649 group_match_null_string_p (p
, end
, reg_info
)
4650 unsigned char **p
, *end
;
4651 register_info_type
*reg_info
;
4654 /* Point to after the args to the start_memory. */
4655 unsigned char *p1
= *p
+ 2;
4659 /* Skip over opcodes that can match nothing, and return true or
4660 false, as appropriate, when we get to one that can't, or to the
4661 matching stop_memory. */
4663 switch ((re_opcode_t
) *p1
)
4665 /* Could be either a loop or a series of alternatives. */
4666 case on_failure_jump
:
4668 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
4670 /* If the next operation is not a jump backwards in the
4675 /* Go through the on_failure_jumps of the alternatives,
4676 seeing if any of the alternatives cannot match nothing.
4677 The last alternative starts with only a jump,
4678 whereas the rest start with on_failure_jump and end
4679 with a jump, e.g., here is the pattern for `a|b|c':
4681 /on_failure_jump/0/6/exactn/1/a/jump_past_alt/0/6
4682 /on_failure_jump/0/6/exactn/1/b/jump_past_alt/0/3
4685 So, we have to first go through the first (n-1)
4686 alternatives and then deal with the last one separately. */
4689 /* Deal with the first (n-1) alternatives, which start
4690 with an on_failure_jump (see above) that jumps to right
4691 past a jump_past_alt. */
4693 while ((re_opcode_t
) p1
[mcnt
-3] == jump_past_alt
)
4695 /* `mcnt' holds how many bytes long the alternative
4696 is, including the ending `jump_past_alt' and
4699 if (!alt_match_null_string_p (p1
, p1
+ mcnt
- 3,
4703 /* Move to right after this alternative, including the
4707 /* Break if it's the beginning of an n-th alternative
4708 that doesn't begin with an on_failure_jump. */
4709 if ((re_opcode_t
) *p1
!= on_failure_jump
)
4712 /* Still have to check that it's not an n-th
4713 alternative that starts with an on_failure_jump. */
4715 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
4716 if ((re_opcode_t
) p1
[mcnt
-3] != jump_past_alt
)
4718 /* Get to the beginning of the n-th alternative. */
4724 /* Deal with the last alternative: go back and get number
4725 of the `jump_past_alt' just before it. `mcnt' contains
4726 the length of the alternative. */
4727 EXTRACT_NUMBER (mcnt
, p1
- 2);
4729 if (!alt_match_null_string_p (p1
, p1
+ mcnt
, reg_info
))
4732 p1
+= mcnt
; /* Get past the n-th alternative. */
4738 assert (p1
[1] == **p
);
4744 if (!common_op_match_null_string_p (&p1
, end
, reg_info
))
4747 } /* while p1 < end */
4750 } /* group_match_null_string_p */
4753 /* Similar to group_match_null_string_p, but doesn't deal with alternatives:
4754 It expects P to be the first byte of a single alternative and END one
4755 byte past the last. The alternative can contain groups. */
4758 alt_match_null_string_p (p
, end
, reg_info
)
4759 unsigned char *p
, *end
;
4760 register_info_type
*reg_info
;
4763 unsigned char *p1
= p
;
4767 /* Skip over opcodes that can match nothing, and break when we get
4768 to one that can't. */
4770 switch ((re_opcode_t
) *p1
)
4773 case on_failure_jump
:
4775 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
4780 if (!common_op_match_null_string_p (&p1
, end
, reg_info
))
4783 } /* while p1 < end */
4786 } /* alt_match_null_string_p */
4789 /* Deals with the ops common to group_match_null_string_p and
4790 alt_match_null_string_p.
4792 Sets P to one after the op and its arguments, if any. */
4795 common_op_match_null_string_p (p
, end
, reg_info
)
4796 unsigned char **p
, *end
;
4797 register_info_type
*reg_info
;
4802 unsigned char *p1
= *p
;
4804 switch ((re_opcode_t
) *p1
++)
4824 assert (reg_no
> 0 && reg_no
<= MAX_REGNUM
);
4825 ret
= group_match_null_string_p (&p1
, end
, reg_info
);
4827 /* Have to set this here in case we're checking a group which
4828 contains a group and a back reference to it. */
4830 if (REG_MATCH_NULL_STRING_P (reg_info
[reg_no
]) == MATCH_NULL_UNSET_VALUE
)
4831 REG_MATCH_NULL_STRING_P (reg_info
[reg_no
]) = ret
;
4837 /* If this is an optimized succeed_n for zero times, make the jump. */
4839 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
4847 /* Get to the number of times to succeed. */
4849 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
4854 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
4862 if (!REG_MATCH_NULL_STRING_P (reg_info
[*p1
]))
4870 /* All other opcodes mean we cannot match the empty string. */
4876 } /* common_op_match_null_string_p */
4879 /* Return zero if TRANSLATE[S1] and TRANSLATE[S2] are identical for LEN
4880 bytes; nonzero otherwise. */
4883 bcmp_translate (s1
, s2
, len
, translate
)
4884 unsigned char *s1
, *s2
;
4888 register unsigned char *p1
= s1
, *p2
= s2
;
4891 if (translate
[*p1
++] != translate
[*p2
++]) return 1;
4897 /* Entry points for GNU code. */
4899 /* re_compile_pattern is the GNU regular expression compiler: it
4900 compiles PATTERN (of length SIZE) and puts the result in BUFP.
4901 Returns 0 if the pattern was valid, otherwise an error string.
4903 Assumes the `allocated' (and perhaps `buffer') and `translate' fields
4904 are set in BUFP on entry.
4906 We call regex_compile to do the actual compilation. */
4909 re_compile_pattern (pattern
, length
, bufp
)
4910 const char *pattern
;
4912 struct re_pattern_buffer
*bufp
;
4916 /* GNU code is written to assume at least RE_NREGS registers will be set
4917 (and at least one extra will be -1). */
4918 bufp
->regs_allocated
= REGS_UNALLOCATED
;
4920 /* And GNU code determines whether or not to get register information
4921 by passing null for the REGS argument to re_match, etc., not by
4925 /* Match anchors at newline. */
4926 bufp
->newline_anchor
= 1;
4928 ret
= regex_compile (pattern
, length
, re_syntax_options
, bufp
);
4930 return re_error_msg
[(int) ret
];
4933 /* Entry points compatible with 4.2 BSD regex library. We don't define
4934 them unless specifically requested. */
4936 #ifdef _REGEX_RE_COMP
4938 /* BSD has one and only one pattern buffer. */
4939 static struct re_pattern_buffer re_comp_buf
;
4949 if (!re_comp_buf
.buffer
)
4950 return "No previous regular expression";
4954 if (!re_comp_buf
.buffer
)
4956 re_comp_buf
.buffer
= (unsigned char *) malloc (200);
4957 if (re_comp_buf
.buffer
== NULL
)
4958 return "Memory exhausted";
4959 re_comp_buf
.allocated
= 200;
4961 re_comp_buf
.fastmap
= (char *) malloc (1 << BYTEWIDTH
);
4962 if (re_comp_buf
.fastmap
== NULL
)
4963 return "Memory exhausted";
4966 /* Since `re_exec' always passes NULL for the `regs' argument, we
4967 don't need to initialize the pattern buffer fields which affect it. */
4969 /* Match anchors at newlines. */
4970 re_comp_buf
.newline_anchor
= 1;
4972 ret
= regex_compile (s
, strlen (s
), re_syntax_options
, &re_comp_buf
);
4974 /* Yes, we're discarding `const' here. */
4975 return (char *) re_error_msg
[(int) ret
];
4983 const int len
= strlen (s
);
4985 0 <= re_search (&re_comp_buf
, s
, len
, 0, len
, (struct re_registers
*) 0);
4987 #endif /* _REGEX_RE_COMP */
4989 /* POSIX.2 functions. Don't define these for Emacs. */
4993 /* regcomp takes a regular expression as a string and compiles it.
4995 PREG is a regex_t *. We do not expect any fields to be initialized,
4996 since POSIX says we shouldn't. Thus, we set
4998 `buffer' to the compiled pattern;
4999 `used' to the length of the compiled pattern;
5000 `syntax' to RE_SYNTAX_POSIX_EXTENDED if the
5001 REG_EXTENDED bit in CFLAGS is set; otherwise, to
5002 RE_SYNTAX_POSIX_BASIC;
5003 `newline_anchor' to REG_NEWLINE being set in CFLAGS;
5004 `fastmap' and `fastmap_accurate' to zero;
5005 `re_nsub' to the number of subexpressions in PATTERN.
5007 PATTERN is the address of the pattern string.
5009 CFLAGS is a series of bits which affect compilation.
5011 If REG_EXTENDED is set, we use POSIX extended syntax; otherwise, we
5012 use POSIX basic syntax.
5014 If REG_NEWLINE is set, then . and [^...] don't match newline.
5015 Also, regexec will try a match beginning after every newline.
5017 If REG_ICASE is set, then we considers upper- and lowercase
5018 versions of letters to be equivalent when matching.
5020 If REG_NOSUB is set, then when PREG is passed to regexec, that
5021 routine will report only success or failure, and nothing about the
5024 It returns 0 if it succeeds, nonzero if it doesn't. (See regex.h for
5025 the return codes and their meanings.) */
5028 regcomp (preg
, pattern
, cflags
)
5030 const char *pattern
;
5035 = (cflags
& REG_EXTENDED
) ?
5036 RE_SYNTAX_POSIX_EXTENDED
: RE_SYNTAX_POSIX_BASIC
;
5038 /* regex_compile will allocate the space for the compiled pattern. */
5040 preg
->allocated
= 0;
5043 /* Don't bother to use a fastmap when searching. This simplifies the
5044 REG_NEWLINE case: if we used a fastmap, we'd have to put all the
5045 characters after newlines into the fastmap. This way, we just try
5049 if (cflags
& REG_ICASE
)
5053 preg
->translate
= (char *) malloc (CHAR_SET_SIZE
);
5054 if (preg
->translate
== NULL
)
5055 return (int) REG_ESPACE
;
5057 /* Map uppercase characters to corresponding lowercase ones. */
5058 for (i
= 0; i
< CHAR_SET_SIZE
; i
++)
5059 preg
->translate
[i
] = ISUPPER (i
) ? tolower (i
) : i
;
5062 preg
->translate
= NULL
;
5064 /* If REG_NEWLINE is set, newlines are treated differently. */
5065 if (cflags
& REG_NEWLINE
)
5066 { /* REG_NEWLINE implies neither . nor [^...] match newline. */
5067 syntax
&= ~RE_DOT_NEWLINE
;
5068 syntax
|= RE_HAT_LISTS_NOT_NEWLINE
;
5069 /* It also changes the matching behavior. */
5070 preg
->newline_anchor
= 1;
5073 preg
->newline_anchor
= 0;
5075 preg
->no_sub
= !!(cflags
& REG_NOSUB
);
5077 /* POSIX says a null character in the pattern terminates it, so we
5078 can use strlen here in compiling the pattern. */
5079 ret
= regex_compile (pattern
, strlen (pattern
), syntax
, preg
);
5081 /* POSIX doesn't distinguish between an unmatched open-group and an
5082 unmatched close-group: both are REG_EPAREN. */
5083 if (ret
== REG_ERPAREN
) ret
= REG_EPAREN
;
5089 /* regexec searches for a given pattern, specified by PREG, in the
5092 If NMATCH is zero or REG_NOSUB was set in the cflags argument to
5093 `regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at
5094 least NMATCH elements, and we set them to the offsets of the
5095 corresponding matched substrings.
5097 EFLAGS specifies `execution flags' which affect matching: if
5098 REG_NOTBOL is set, then ^ does not match at the beginning of the
5099 string; if REG_NOTEOL is set, then $ does not match at the end.
5101 We return 0 if we find a match and REG_NOMATCH if not. */
5104 regexec (preg
, string
, nmatch
, pmatch
, eflags
)
5105 const regex_t
*preg
;
5108 regmatch_t pmatch
[];
5112 struct re_registers regs
;
5113 regex_t private_preg
;
5114 int len
= strlen (string
);
5115 boolean want_reg_info
= !preg
->no_sub
&& nmatch
> 0;
5117 private_preg
= *preg
;
5119 private_preg
.not_bol
= !!(eflags
& REG_NOTBOL
);
5120 private_preg
.not_eol
= !!(eflags
& REG_NOTEOL
);
5122 /* The user has told us exactly how many registers to return
5123 information about, via `nmatch'. We have to pass that on to the
5124 matching routines. */
5125 private_preg
.regs_allocated
= REGS_FIXED
;
5129 regs
.num_regs
= nmatch
;
5130 regs
.start
= TALLOC (nmatch
, regoff_t
);
5131 regs
.end
= TALLOC (nmatch
, regoff_t
);
5132 if (regs
.start
== NULL
|| regs
.end
== NULL
)
5133 return (int) REG_NOMATCH
;
5136 /* Perform the searching operation. */
5137 ret
= re_search (&private_preg
, string
, len
,
5138 /* start: */ 0, /* range: */ len
,
5139 want_reg_info
? ®s
: (struct re_registers
*) 0);
5141 /* Copy the register information to the POSIX structure. */
5148 for (r
= 0; r
< nmatch
; r
++)
5150 pmatch
[r
].rm_so
= regs
.start
[r
];
5151 pmatch
[r
].rm_eo
= regs
.end
[r
];
5155 /* If we needed the temporary register info, free the space now. */
5160 /* We want zero return to mean success, unlike `re_search'. */
5161 return ret
>= 0 ? (int) REG_NOERROR
: (int) REG_NOMATCH
;
5165 /* Returns a message corresponding to an error code, ERRCODE, returned
5166 from either regcomp or regexec. We don't use PREG here. */
5169 regerror (errcode
, preg
, errbuf
, errbuf_size
)
5171 const regex_t
*preg
;
5179 || errcode
>= (sizeof (re_error_msg
) / sizeof (re_error_msg
[0])))
5180 /* Only error codes returned by the rest of the code should be passed
5181 to this routine. If we are given anything else, or if other regex
5182 code generates an invalid error code, then the program has a bug.
5183 Dump core so we can fix it. */
5186 msg
= re_error_msg
[errcode
];
5188 /* POSIX doesn't require that we do anything in this case, but why
5193 msg_size
= strlen (msg
) + 1; /* Includes the null. */
5195 if (errbuf_size
!= 0)
5197 if (msg_size
> errbuf_size
)
5199 strncpy (errbuf
, msg
, errbuf_size
- 1);
5200 errbuf
[errbuf_size
- 1] = 0;
5203 strcpy (errbuf
, msg
);
5210 /* Free dynamically allocated space used by PREG. */
5216 if (preg
->buffer
!= NULL
)
5217 free (preg
->buffer
);
5218 preg
->buffer
= NULL
;
5220 preg
->allocated
= 0;
5223 if (preg
->fastmap
!= NULL
)
5224 free (preg
->fastmap
);
5225 preg
->fastmap
= NULL
;
5226 preg
->fastmap_accurate
= 0;
5228 if (preg
->translate
!= NULL
)
5229 free (preg
->translate
);
5230 preg
->translate
= NULL
;
5233 #endif /* not emacs */
5237 make-backup-files: t
5239 trim-versions-without-asking: nil