1 /* Extended regular expression matching and search library,
3 (Implements POSIX draft P10003.2/D11.2, except for
4 internationalization features.)
6 Copyright (C) 1993, 1994 Free Software Foundation, Inc.
8 This program is free software; you can redistribute it and/or modify
9 it under the terms of the GNU General Public License as published by
10 the Free Software Foundation; either version 2, or (at your option)
13 This program is distributed in the hope that it will be useful,
14 but WITHOUT ANY WARRANTY; without even the implied warranty of
15 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
16 GNU General Public License for more details.
18 You should have received a copy of the GNU General Public License
19 along with this program; if not, write to the Free Software
20 Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. */
22 /* AIX requires this to be the first thing in the file. */
23 #if defined (_AIX) && !defined (REGEX_MALLOC)
33 /* We need this for `regex.h', and perhaps for the Emacs include files. */
34 #include <sys/types.h>
36 /* This is for other GNU distributions with internationalized messages.
37 The GNU C Library itself does not yet support such messages. */
41 # define gettext(msgid) (msgid)
44 /* The `emacs' switch turns on certain matching commands
45 that make sense only in Emacs. */
62 /* We used to test for `BSTRING' here, but only GCC and Emacs define
63 `BSTRING', as far as I know, and neither of them use this code. */
64 #ifndef INHIBIT_STRING_HEADER
65 #if HAVE_STRING_H || STDC_HEADERS
68 #define bcmp(s1, s2, n) memcmp ((s1), (s2), (n))
71 #define bcopy(s, d, n) memcpy ((d), (s), (n))
74 #define bzero(s, n) memset ((s), 0, (n))
81 /* Define the syntax stuff for \<, \>, etc. */
83 /* This must be nonzero for the wordchar and notwordchar pattern
84 commands in re_match_2. */
89 #ifdef SWITCH_ENUM_BUG
90 #define SWITCH_ENUM_CAST(x) ((int)(x))
92 #define SWITCH_ENUM_CAST(x) (x)
97 extern char *re_syntax_table;
99 #else /* not SYNTAX_TABLE */
101 /* How many characters in the character set. */
102 #define CHAR_SET_SIZE 256
104 static char re_syntax_table[CHAR_SET_SIZE];
115 bzero (re_syntax_table, sizeof re_syntax_table);
117 for (c = 'a'; c <= 'z'; c++)
118 re_syntax_table[c] = Sword;
120 for (c = 'A'; c <= 'Z'; c++)
121 re_syntax_table[c] = Sword;
123 for (c = '0'; c <= '9'; c++)
124 re_syntax_table[c] = Sword;
126 re_syntax_table['_'] = Sword;
131 #endif /* not SYNTAX_TABLE */
133 #define SYNTAX(c) re_syntax_table[c]
135 #endif /* not emacs */
137 /* Get the interface, including the syntax bits. */
140 /* isalpha etc. are used for the character classes. */
143 /* Jim Meyering writes:
145 "... Some ctype macros are valid only for character codes that
146 isascii says are ASCII (SGI's IRIX-4.0.5 is one such system --when
147 using /bin/cc or gcc but without giving an ansi option). So, all
148 ctype uses should be through macros like ISPRINT... If
149 STDC_HEADERS is defined, then autoconf has verified that the ctype
150 macros don't need to be guarded with references to isascii. ...
151 Defining isascii to 1 should let any compiler worth its salt
152 eliminate the && through constant folding." */
154 #if defined (STDC_HEADERS) || (!defined (isascii) && !defined (HAVE_ISASCII))
157 #define ISASCII(c) isascii(c)
161 #define ISBLANK(c) (ISASCII (c) && isblank (c))
163 #define ISBLANK(c) ((c) == ' ' || (c) == '\t')
166 #define ISGRAPH(c) (ISASCII (c) && isgraph (c))
168 #define ISGRAPH(c) (ISASCII (c) && isprint (c) && !isspace (c))
171 #define ISPRINT(c) (ISASCII (c) && isprint (c))
172 #define ISDIGIT(c) (ISASCII (c) && isdigit (c))
173 #define ISALNUM(c) (ISASCII (c) && isalnum (c))
174 #define ISALPHA(c) (ISASCII (c) && isalpha (c))
175 #define ISCNTRL(c) (ISASCII (c) && iscntrl (c))
176 #define ISLOWER(c) (ISASCII (c) && islower (c))
177 #define ISPUNCT(c) (ISASCII (c) && ispunct (c))
178 #define ISSPACE(c) (ISASCII (c) && isspace (c))
179 #define ISUPPER(c) (ISASCII (c) && isupper (c))
180 #define ISXDIGIT(c) (ISASCII (c) && isxdigit (c))
186 /* We remove any previous definition of `SIGN_EXTEND_CHAR',
187 since ours (we hope) works properly with all combinations of
188 machines, compilers, `char' and `unsigned char' argument types.
189 (Per Bothner suggested the basic approach.) */
190 #undef SIGN_EXTEND_CHAR
192 #define SIGN_EXTEND_CHAR(c) ((signed char) (c))
193 #else /* not __STDC__ */
194 /* As in Harbison and Steele. */
195 #define SIGN_EXTEND_CHAR(c) ((((unsigned char) (c)) ^ 128) - 128)
198 /* Should we use malloc or alloca? If REGEX_MALLOC is not defined, we
199 use `alloca' instead of `malloc'. This is because using malloc in
200 re_search* or re_match* could cause memory leaks when C-g is used in
201 Emacs; also, malloc is slower and causes storage fragmentation. On
202 the other hand, malloc is more portable, and easier to debug.
204 Because we sometimes use alloca, some routines have to be macros,
205 not functions -- `alloca'-allocated space disappears at the end of the
206 function it is called in. */
210 #define REGEX_ALLOCATE malloc
211 #define REGEX_REALLOCATE(source, osize, nsize) realloc (source, nsize)
213 #else /* not REGEX_MALLOC */
215 /* Emacs already defines alloca, sometimes. */
218 /* Make alloca work the best possible way. */
220 #define alloca __builtin_alloca
221 #else /* not __GNUC__ */
224 #else /* not __GNUC__ or HAVE_ALLOCA_H */
225 #ifndef _AIX /* Already did AIX, up at the top. */
227 #endif /* not _AIX */
228 #endif /* not HAVE_ALLOCA_H */
229 #endif /* not __GNUC__ */
231 #endif /* not alloca */
233 #define REGEX_ALLOCATE alloca
235 /* Assumes a `char *destination' variable. */
236 #define REGEX_REALLOCATE(source, osize, nsize) \
237 (destination = (char *) alloca (nsize), \
238 bcopy (source, destination, osize), \
241 #endif /* not REGEX_MALLOC */
244 /* True if `size1' is non-NULL and PTR is pointing anywhere inside
245 `string1' or just past its end. This works if PTR is NULL, which is
247 #define FIRST_STRING_P(ptr) \
248 (size1 && string1 <= (ptr) && (ptr) <= string1 + size1)
250 /* (Re)Allocate N items of type T using malloc, or fail. */
251 #define TALLOC(n, t) ((t *) malloc ((n) * sizeof (t)))
252 #define RETALLOC(addr, n, t) ((addr) = (t *) realloc (addr, (n) * sizeof (t)))
253 #define RETALLOC_IF(addr, n, t) \
254 if (addr) RETALLOC((addr), (n), t); else (addr) = TALLOC ((n), t)
255 #define REGEX_TALLOC(n, t) ((t *) REGEX_ALLOCATE ((n) * sizeof (t)))
257 #define BYTEWIDTH 8 /* In bits. */
259 #define STREQ(s1, s2) ((strcmp (s1, s2) == 0))
263 #define MAX(a, b) ((a) > (b) ? (a) : (b))
264 #define MIN(a, b) ((a) < (b) ? (a) : (b))
266 typedef char boolean;
270 static int re_match_2_internal ();
272 /* These are the command codes that appear in compiled regular
273 expressions. Some opcodes are followed by argument bytes. A
274 command code can specify any interpretation whatsoever for its
275 arguments. Zero bytes may appear in the compiled regular expression. */
281 /* Succeed right away--no more backtracking. */
284 /* Followed by one byte giving n, then by n literal bytes. */
287 /* Matches any (more or less) character. */
290 /* Matches any one char belonging to specified set. First
291 following byte is number of bitmap bytes. Then come bytes
292 for a bitmap saying which chars are in. Bits in each byte
293 are ordered low-bit-first. A character is in the set if its
294 bit is 1. A character too large to have a bit in the map is
295 automatically not in the set. */
298 /* Same parameters as charset, but match any character that is
299 not one of those specified. */
302 /* Start remembering the text that is matched, for storing in a
303 register. Followed by one byte with the register number, in
304 the range 0 to one less than the pattern buffer's re_nsub
305 field. Then followed by one byte with the number of groups
306 inner to this one. (This last has to be part of the
307 start_memory only because we need it in the on_failure_jump
311 /* Stop remembering the text that is matched and store it in a
312 memory register. Followed by one byte with the register
313 number, in the range 0 to one less than `re_nsub' in the
314 pattern buffer, and one byte with the number of inner groups,
315 just like `start_memory'. (We need the number of inner
316 groups here because we don't have any easy way of finding the
317 corresponding start_memory when we're at a stop_memory.) */
320 /* Match a duplicate of something remembered. Followed by one
321 byte containing the register number. */
324 /* Fail unless at beginning of line. */
327 /* Fail unless at end of line. */
330 /* Succeeds if at beginning of buffer (if emacs) or at beginning
331 of string to be matched (if not). */
334 /* Analogously, for end of buffer/string. */
337 /* Followed by two byte relative address to which to jump. */
340 /* Same as jump, but marks the end of an alternative. */
343 /* Followed by two-byte relative address of place to resume at
344 in case of failure. */
347 /* Like on_failure_jump, but pushes a placeholder instead of the
348 current string position when executed. */
349 on_failure_keep_string_jump,
351 /* Throw away latest failure point and then jump to following
352 two-byte relative address. */
355 /* Change to pop_failure_jump if know won't have to backtrack to
356 match; otherwise change to jump. This is used to jump
357 back to the beginning of a repeat. If what follows this jump
358 clearly won't match what the repeat does, such that we can be
359 sure that there is no use backtracking out of repetitions
360 already matched, then we change it to a pop_failure_jump.
361 Followed by two-byte address. */
364 /* Jump to following two-byte address, and push a dummy failure
365 point. This failure point will be thrown away if an attempt
366 is made to use it for a failure. A `+' construct makes this
367 before the first repeat. Also used as an intermediary kind
368 of jump when compiling an alternative. */
371 /* Push a dummy failure point and continue. Used at the end of
375 /* Followed by two-byte relative address and two-byte number n.
376 After matching N times, jump to the address upon failure. */
379 /* Followed by two-byte relative address, and two-byte number n.
380 Jump to the address N times, then fail. */
383 /* Set the following two-byte relative address to the
384 subsequent two-byte number. The address *includes* the two
388 wordchar, /* Matches any word-constituent character. */
389 notwordchar, /* Matches any char that is not a word-constituent. */
391 wordbeg, /* Succeeds if at word beginning. */
392 wordend, /* Succeeds if at word end. */
394 wordbound, /* Succeeds if at a word boundary. */
395 notwordbound /* Succeeds if not at a word boundary. */
398 ,before_dot, /* Succeeds if before point. */
399 at_dot, /* Succeeds if at point. */
400 after_dot, /* Succeeds if after point. */
402 /* Matches any character whose syntax is specified. Followed by
403 a byte which contains a syntax code, e.g., Sword. */
406 /* Matches any character whose syntax is not that specified. */
411 /* Common operations on the compiled pattern. */
413 /* Store NUMBER in two contiguous bytes starting at DESTINATION. */
415 #define STORE_NUMBER(destination, number) \
417 (destination)[0] = (number) & 0377; \
418 (destination)[1] = (number) >> 8; \
421 /* Same as STORE_NUMBER, except increment DESTINATION to
422 the byte after where the number is stored. Therefore, DESTINATION
423 must be an lvalue. */
425 #define STORE_NUMBER_AND_INCR(destination, number) \
427 STORE_NUMBER (destination, number); \
428 (destination) += 2; \
431 /* Put into DESTINATION a number stored in two contiguous bytes starting
434 #define EXTRACT_NUMBER(destination, source) \
436 (destination) = *(source) & 0377; \
437 (destination) += SIGN_EXTEND_CHAR (*((source) + 1)) << 8; \
442 extract_number (dest, source)
444 unsigned char *source;
446 int temp = SIGN_EXTEND_CHAR (*(source + 1));
447 *dest = *source & 0377;
451 #ifndef EXTRACT_MACROS /* To debug the macros. */
452 #undef EXTRACT_NUMBER
453 #define EXTRACT_NUMBER(dest, src) extract_number (&dest, src)
454 #endif /* not EXTRACT_MACROS */
458 /* Same as EXTRACT_NUMBER, except increment SOURCE to after the number.
459 SOURCE must be an lvalue. */
461 #define EXTRACT_NUMBER_AND_INCR(destination, source) \
463 EXTRACT_NUMBER (destination, source); \
469 extract_number_and_incr (destination, source)
471 unsigned char **source;
473 extract_number (destination, *source);
477 #ifndef EXTRACT_MACROS
478 #undef EXTRACT_NUMBER_AND_INCR
479 #define EXTRACT_NUMBER_AND_INCR(dest, src) \
480 extract_number_and_incr (&dest, &src)
481 #endif /* not EXTRACT_MACROS */
485 /* If DEBUG is defined, Regex prints many voluminous messages about what
486 it is doing (if the variable `debug' is nonzero). If linked with the
487 main program in `iregex.c', you can enter patterns and strings
488 interactively. And if linked with the main program in `main.c' and
489 the other test files, you can run the already-written tests. */
493 /* We use standard I/O for debugging. */
496 /* It is useful to test things that ``must'' be true when debugging. */
499 static int debug = 0;
501 #define DEBUG_STATEMENT(e) e
502 #define DEBUG_PRINT1(x) if (debug) printf (x)
503 #define DEBUG_PRINT2(x1, x2) if (debug) printf (x1, x2)
504 #define DEBUG_PRINT3(x1, x2, x3) if (debug) printf (x1, x2, x3)
505 #define DEBUG_PRINT4(x1, x2, x3, x4) if (debug) printf (x1, x2, x3, x4)
506 #define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) \
507 if (debug) print_partial_compiled_pattern (s, e)
508 #define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) \
509 if (debug) print_double_string (w, s1, sz1, s2, sz2)
512 /* Print the fastmap in human-readable form. */
515 print_fastmap (fastmap)
518 unsigned was_a_range = 0;
521 while (i < (1 << BYTEWIDTH))
527 while (i < (1 << BYTEWIDTH) && fastmap[i])
543 /* Print a compiled pattern string in human-readable form, starting at
544 the START pointer into it and ending just before the pointer END. */
547 print_partial_compiled_pattern (start, end)
548 unsigned char *start;
552 unsigned char *p = start;
553 unsigned char *pend = end;
561 /* Loop over pattern commands. */
564 printf ("%d:\t", p - start);
566 switch ((re_opcode_t) *p++)
574 printf ("/exactn/%d", mcnt);
585 printf ("/start_memory/%d/%d", mcnt, *p++);
590 printf ("/stop_memory/%d/%d", mcnt, *p++);
594 printf ("/duplicate/%d", *p++);
604 register int c, last = -100;
605 register int in_range = 0;
607 printf ("/charset [%s",
608 (re_opcode_t) *(p - 1) == charset_not ? "^" : "");
610 assert (p + *p < pend);
612 for (c = 0; c < 256; c++)
614 && (p[1 + (c/8)] & (1 << (c % 8))))
616 /* Are we starting a range? */
617 if (last + 1 == c && ! in_range)
622 /* Have we broken a range? */
623 else if (last + 1 != c && in_range)
652 case on_failure_jump:
653 extract_number_and_incr (&mcnt, &p);
654 printf ("/on_failure_jump to %d", p + mcnt - start);
657 case on_failure_keep_string_jump:
658 extract_number_and_incr (&mcnt, &p);
659 printf ("/on_failure_keep_string_jump to %d", p + mcnt - start);
662 case dummy_failure_jump:
663 extract_number_and_incr (&mcnt, &p);
664 printf ("/dummy_failure_jump to %d", p + mcnt - start);
667 case push_dummy_failure:
668 printf ("/push_dummy_failure");
672 extract_number_and_incr (&mcnt, &p);
673 printf ("/maybe_pop_jump to %d", p + mcnt - start);
676 case pop_failure_jump:
677 extract_number_and_incr (&mcnt, &p);
678 printf ("/pop_failure_jump to %d", p + mcnt - start);
682 extract_number_and_incr (&mcnt, &p);
683 printf ("/jump_past_alt to %d", p + mcnt - start);
687 extract_number_and_incr (&mcnt, &p);
688 printf ("/jump to %d", p + mcnt - start);
692 extract_number_and_incr (&mcnt, &p);
693 extract_number_and_incr (&mcnt2, &p);
694 printf ("/succeed_n to %d, %d times", p + mcnt - start, mcnt2);
698 extract_number_and_incr (&mcnt, &p);
699 extract_number_and_incr (&mcnt2, &p);
700 printf ("/jump_n to %d, %d times", p + mcnt - start, mcnt2);
704 extract_number_and_incr (&mcnt, &p);
705 extract_number_and_incr (&mcnt2, &p);
706 printf ("/set_number_at location %d to %d", p + mcnt - start, mcnt2);
710 printf ("/wordbound");
714 printf ("/notwordbound");
726 printf ("/before_dot");
734 printf ("/after_dot");
738 printf ("/syntaxspec");
740 printf ("/%d", mcnt);
744 printf ("/notsyntaxspec");
746 printf ("/%d", mcnt);
751 printf ("/wordchar");
755 printf ("/notwordchar");
767 printf ("?%d", *(p-1));
773 printf ("%d:\tend of pattern.\n", p - start);
778 print_compiled_pattern (bufp)
779 struct re_pattern_buffer *bufp;
781 unsigned char *buffer = bufp->buffer;
783 print_partial_compiled_pattern (buffer, buffer + bufp->used);
784 printf ("%d bytes used/%d bytes allocated.\n", bufp->used, bufp->allocated);
786 if (bufp->fastmap_accurate && bufp->fastmap)
788 printf ("fastmap: ");
789 print_fastmap (bufp->fastmap);
792 printf ("re_nsub: %d\t", bufp->re_nsub);
793 printf ("regs_alloc: %d\t", bufp->regs_allocated);
794 printf ("can_be_null: %d\t", bufp->can_be_null);
795 printf ("newline_anchor: %d\n", bufp->newline_anchor);
796 printf ("no_sub: %d\t", bufp->no_sub);
797 printf ("not_bol: %d\t", bufp->not_bol);
798 printf ("not_eol: %d\t", bufp->not_eol);
799 printf ("syntax: %d\n", bufp->syntax);
800 /* Perhaps we should print the translate table? */
805 print_double_string (where, string1, size1, string2, size2)
818 if (FIRST_STRING_P (where))
820 for (this_char = where - string1; this_char < size1; this_char++)
821 putchar (string1[this_char]);
826 for (this_char = where - string2; this_char < size2; this_char++)
827 putchar (string2[this_char]);
831 #else /* not DEBUG */
836 #define DEBUG_STATEMENT(e)
837 #define DEBUG_PRINT1(x)
838 #define DEBUG_PRINT2(x1, x2)
839 #define DEBUG_PRINT3(x1, x2, x3)
840 #define DEBUG_PRINT4(x1, x2, x3, x4)
841 #define DEBUG_PRINT_COMPILED_PATTERN(p, s, e)
842 #define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2)
844 #endif /* not DEBUG */
846 /* Set by `re_set_syntax' to the current regexp syntax to recognize. Can
847 also be assigned to arbitrarily: each pattern buffer stores its own
848 syntax, so it can be changed between regex compilations. */
849 /* This has no initializer because initialized variables in Emacs
850 become read-only after dumping. */
851 reg_syntax_t re_syntax_options;
854 /* Specify the precise syntax of regexps for compilation. This provides
855 for compatibility for various utilities which historically have
856 different, incompatible syntaxes.
858 The argument SYNTAX is a bit mask comprised of the various bits
859 defined in regex.h. We return the old syntax. */
862 re_set_syntax (syntax)
865 reg_syntax_t ret = re_syntax_options;
867 re_syntax_options = syntax;
871 /* This table gives an error message for each of the error codes listed
872 in regex.h. Obviously the order here has to be same as there.
873 POSIX doesn't require that we do anything for REG_NOERROR,
874 but why not be nice? */
876 static const char *re_error_msgid[] =
877 { "Success", /* REG_NOERROR */
878 "No match", /* REG_NOMATCH */
879 "Invalid regular expression", /* REG_BADPAT */
880 "Invalid collation character", /* REG_ECOLLATE */
881 "Invalid character class name", /* REG_ECTYPE */
882 "Trailing backslash", /* REG_EESCAPE */
883 "Invalid back reference", /* REG_ESUBREG */
884 "Unmatched [ or [^", /* REG_EBRACK */
885 "Unmatched ( or \\(", /* REG_EPAREN */
886 "Unmatched \\{", /* REG_EBRACE */
887 "Invalid content of \\{\\}", /* REG_BADBR */
888 "Invalid range end", /* REG_ERANGE */
889 "Memory exhausted", /* REG_ESPACE */
890 "Invalid preceding regular expression", /* REG_BADRPT */
891 "Premature end of regular expression", /* REG_EEND */
892 "Regular expression too big", /* REG_ESIZE */
893 "Unmatched ) or \\)", /* REG_ERPAREN */
896 /* Avoiding alloca during matching, to placate r_alloc. */
898 /* Define MATCH_MAY_ALLOCATE unless we need to make sure that the
899 searching and matching functions should not call alloca. On some
900 systems, alloca is implemented in terms of malloc, and if we're
901 using the relocating allocator routines, then malloc could cause a
902 relocation, which might (if the strings being searched are in the
903 ralloc heap) shift the data out from underneath the regexp
906 Here's another reason to avoid allocation: Emacs
907 processes input from X in a signal handler; processing X input may
908 call malloc; if input arrives while a matching routine is calling
909 malloc, then we're scrod. But Emacs can't just block input while
910 calling matching routines; then we don't notice interrupts when
911 they come in. So, Emacs blocks input around all regexp calls
912 except the matching calls, which it leaves unprotected, in the
913 faith that they will not malloc. */
915 /* Normally, this is fine. */
916 #define MATCH_MAY_ALLOCATE
918 /* The match routines may not allocate if (1) they would do it with malloc
919 and (2) it's not safe for them to use malloc. */
920 #if (defined (C_ALLOCA) || defined (REGEX_MALLOC)) && (defined (emacs) || defined (REL_ALLOC))
921 #undef MATCH_MAY_ALLOCATE
925 /* Failure stack declarations and macros; both re_compile_fastmap and
926 re_match_2 use a failure stack. These have to be macros because of
930 /* Number of failure points for which to initially allocate space
931 when matching. If this number is exceeded, we allocate more
932 space, so it is not a hard limit. */
933 #ifndef INIT_FAILURE_ALLOC
934 #define INIT_FAILURE_ALLOC 5
937 /* Roughly the maximum number of failure points on the stack. Would be
938 exactly that if always used MAX_FAILURE_SPACE each time we failed.
939 This is a variable only so users of regex can assign to it; we never
940 change it ourselves. */
941 int re_max_failures = 2000;
943 typedef unsigned char *fail_stack_elt_t;
947 fail_stack_elt_t *stack;
949 unsigned avail; /* Offset of next open position. */
952 #define FAIL_STACK_EMPTY() (fail_stack.avail == 0)
953 #define FAIL_STACK_PTR_EMPTY() (fail_stack_ptr->avail == 0)
954 #define FAIL_STACK_FULL() (fail_stack.avail == fail_stack.size)
955 #define FAIL_STACK_TOP() (fail_stack.stack[fail_stack.avail])
958 /* Initialize `fail_stack'. Do `return -2' if the alloc fails. */
960 #ifdef MATCH_MAY_ALLOCATE
961 #define INIT_FAIL_STACK() \
963 fail_stack.stack = (fail_stack_elt_t *) \
964 REGEX_ALLOCATE (INIT_FAILURE_ALLOC * sizeof (fail_stack_elt_t)); \
966 if (fail_stack.stack == NULL) \
969 fail_stack.size = INIT_FAILURE_ALLOC; \
970 fail_stack.avail = 0; \
973 #define INIT_FAIL_STACK() \
975 fail_stack.avail = 0; \
980 /* Double the size of FAIL_STACK, up to approximately `re_max_failures' items.
982 Return 1 if succeeds, and 0 if either ran out of memory
983 allocating space for it or it was already too large.
985 REGEX_REALLOCATE requires `destination' be declared. */
987 #define DOUBLE_FAIL_STACK(fail_stack) \
988 ((fail_stack).size > re_max_failures * MAX_FAILURE_ITEMS \
990 : ((fail_stack).stack = (fail_stack_elt_t *) \
991 REGEX_REALLOCATE ((fail_stack).stack, \
992 (fail_stack).size * sizeof (fail_stack_elt_t), \
993 ((fail_stack).size << 1) * sizeof (fail_stack_elt_t)), \
995 (fail_stack).stack == NULL \
997 : ((fail_stack).size <<= 1, \
1001 /* Push PATTERN_OP on FAIL_STACK.
1003 Return 1 if was able to do so and 0 if ran out of memory allocating
1005 #define PUSH_PATTERN_OP(pattern_op, fail_stack) \
1006 ((FAIL_STACK_FULL () \
1007 && !DOUBLE_FAIL_STACK (fail_stack)) \
1009 : ((fail_stack).stack[(fail_stack).avail++] = pattern_op, \
1012 /* This pushes an item onto the failure stack. Must be a four-byte
1013 value. Assumes the variable `fail_stack'. Probably should only
1014 be called from within `PUSH_FAILURE_POINT'. */
1015 #define PUSH_FAILURE_ITEM(item) \
1016 fail_stack.stack[fail_stack.avail++] = (fail_stack_elt_t) item
1018 /* The complement operation. Assumes `fail_stack' is nonempty. */
1019 #define POP_FAILURE_ITEM() fail_stack.stack[--fail_stack.avail]
1021 /* Used to omit pushing failure point id's when we're not debugging. */
1023 #define DEBUG_PUSH PUSH_FAILURE_ITEM
1024 #define DEBUG_POP(item_addr) *(item_addr) = POP_FAILURE_ITEM ()
1026 #define DEBUG_PUSH(item)
1027 #define DEBUG_POP(item_addr)
1031 /* Push the information about the state we will need
1032 if we ever fail back to it.
1034 Requires variables fail_stack, regstart, regend, reg_info, and
1035 num_regs be declared. DOUBLE_FAIL_STACK requires `destination' be
1038 Does `return FAILURE_CODE' if runs out of memory. */
1040 #define PUSH_FAILURE_POINT(pattern_place, string_place, failure_code) \
1042 char *destination; \
1043 /* Must be int, so when we don't save any registers, the arithmetic \
1044 of 0 + -1 isn't done as unsigned. */ \
1047 DEBUG_STATEMENT (failure_id++); \
1048 DEBUG_STATEMENT (nfailure_points_pushed++); \
1049 DEBUG_PRINT2 ("\nPUSH_FAILURE_POINT #%u:\n", failure_id); \
1050 DEBUG_PRINT2 (" Before push, next avail: %d\n", (fail_stack).avail);\
1051 DEBUG_PRINT2 (" size: %d\n", (fail_stack).size);\
1053 DEBUG_PRINT2 (" slots needed: %d\n", NUM_FAILURE_ITEMS); \
1054 DEBUG_PRINT2 (" available: %d\n", REMAINING_AVAIL_SLOTS); \
1056 /* Ensure we have enough space allocated for what we will push. */ \
1057 while (REMAINING_AVAIL_SLOTS < NUM_FAILURE_ITEMS) \
1059 if (!DOUBLE_FAIL_STACK (fail_stack)) \
1060 return failure_code; \
1062 DEBUG_PRINT2 ("\n Doubled stack; size now: %d\n", \
1063 (fail_stack).size); \
1064 DEBUG_PRINT2 (" slots available: %d\n", REMAINING_AVAIL_SLOTS);\
1067 /* Push the info, starting with the registers. */ \
1068 DEBUG_PRINT1 ("\n"); \
1070 for (this_reg = lowest_active_reg; this_reg <= highest_active_reg; \
1073 DEBUG_PRINT2 (" Pushing reg: %d\n", this_reg); \
1074 DEBUG_STATEMENT (num_regs_pushed++); \
1076 DEBUG_PRINT2 (" start: 0x%x\n", regstart[this_reg]); \
1077 PUSH_FAILURE_ITEM (regstart[this_reg]); \
1079 DEBUG_PRINT2 (" end: 0x%x\n", regend[this_reg]); \
1080 PUSH_FAILURE_ITEM (regend[this_reg]); \
1082 DEBUG_PRINT2 (" info: 0x%x\n ", reg_info[this_reg]); \
1083 DEBUG_PRINT2 (" match_null=%d", \
1084 REG_MATCH_NULL_STRING_P (reg_info[this_reg])); \
1085 DEBUG_PRINT2 (" active=%d", IS_ACTIVE (reg_info[this_reg])); \
1086 DEBUG_PRINT2 (" matched_something=%d", \
1087 MATCHED_SOMETHING (reg_info[this_reg])); \
1088 DEBUG_PRINT2 (" ever_matched=%d", \
1089 EVER_MATCHED_SOMETHING (reg_info[this_reg])); \
1090 DEBUG_PRINT1 ("\n"); \
1091 PUSH_FAILURE_ITEM (reg_info[this_reg].word); \
1094 DEBUG_PRINT2 (" Pushing low active reg: %d\n", lowest_active_reg);\
1095 PUSH_FAILURE_ITEM (lowest_active_reg); \
1097 DEBUG_PRINT2 (" Pushing high active reg: %d\n", highest_active_reg);\
1098 PUSH_FAILURE_ITEM (highest_active_reg); \
1100 DEBUG_PRINT2 (" Pushing pattern 0x%x: ", pattern_place); \
1101 DEBUG_PRINT_COMPILED_PATTERN (bufp, pattern_place, pend); \
1102 PUSH_FAILURE_ITEM (pattern_place); \
1104 DEBUG_PRINT2 (" Pushing string 0x%x: `", string_place); \
1105 DEBUG_PRINT_DOUBLE_STRING (string_place, string1, size1, string2, \
1107 DEBUG_PRINT1 ("'\n"); \
1108 PUSH_FAILURE_ITEM (string_place); \
1110 DEBUG_PRINT2 (" Pushing failure id: %u\n", failure_id); \
1111 DEBUG_PUSH (failure_id); \
1114 /* This is the number of items that are pushed and popped on the stack
1115 for each register. */
1116 #define NUM_REG_ITEMS 3
1118 /* Individual items aside from the registers. */
1120 #define NUM_NONREG_ITEMS 5 /* Includes failure point id. */
1122 #define NUM_NONREG_ITEMS 4
1125 /* We push at most this many items on the stack. */
1126 #define MAX_FAILURE_ITEMS ((num_regs - 1) * NUM_REG_ITEMS + NUM_NONREG_ITEMS)
1128 /* We actually push this many items. */
1129 #define NUM_FAILURE_ITEMS \
1130 ((highest_active_reg - lowest_active_reg + 1) * NUM_REG_ITEMS \
1133 /* How many items can still be added to the stack without overflowing it. */
1134 #define REMAINING_AVAIL_SLOTS ((fail_stack).size - (fail_stack).avail)
1137 /* Pops what PUSH_FAIL_STACK pushes.
1139 We restore into the parameters, all of which should be lvalues:
1140 STR -- the saved data position.
1141 PAT -- the saved pattern position.
1142 LOW_REG, HIGH_REG -- the highest and lowest active registers.
1143 REGSTART, REGEND -- arrays of string positions.
1144 REG_INFO -- array of information about each subexpression.
1146 Also assumes the variables `fail_stack' and (if debugging), `bufp',
1147 `pend', `string1', `size1', `string2', and `size2'. */
1149 #define POP_FAILURE_POINT(str, pat, low_reg, high_reg, regstart, regend, reg_info)\
1151 DEBUG_STATEMENT (fail_stack_elt_t failure_id;) \
1153 const unsigned char *string_temp; \
1155 assert (!FAIL_STACK_EMPTY ()); \
1157 /* Remove failure points and point to how many regs pushed. */ \
1158 DEBUG_PRINT1 ("POP_FAILURE_POINT:\n"); \
1159 DEBUG_PRINT2 (" Before pop, next avail: %d\n", fail_stack.avail); \
1160 DEBUG_PRINT2 (" size: %d\n", fail_stack.size); \
1162 assert (fail_stack.avail >= NUM_NONREG_ITEMS); \
1164 DEBUG_POP (&failure_id); \
1165 DEBUG_PRINT2 (" Popping failure id: %u\n", failure_id); \
1167 /* If the saved string location is NULL, it came from an \
1168 on_failure_keep_string_jump opcode, and we want to throw away the \
1169 saved NULL, thus retaining our current position in the string. */ \
1170 string_temp = POP_FAILURE_ITEM (); \
1171 if (string_temp != NULL) \
1172 str = (const char *) string_temp; \
1174 DEBUG_PRINT2 (" Popping string 0x%x: `", str); \
1175 DEBUG_PRINT_DOUBLE_STRING (str, string1, size1, string2, size2); \
1176 DEBUG_PRINT1 ("'\n"); \
1178 pat = (unsigned char *) POP_FAILURE_ITEM (); \
1179 DEBUG_PRINT2 (" Popping pattern 0x%x: ", pat); \
1180 DEBUG_PRINT_COMPILED_PATTERN (bufp, pat, pend); \
1182 /* Restore register info. */ \
1183 high_reg = (unsigned) POP_FAILURE_ITEM (); \
1184 DEBUG_PRINT2 (" Popping high active reg: %d\n", high_reg); \
1186 low_reg = (unsigned) POP_FAILURE_ITEM (); \
1187 DEBUG_PRINT2 (" Popping low active reg: %d\n", low_reg); \
1189 for (this_reg = high_reg; this_reg >= low_reg; this_reg--) \
1191 DEBUG_PRINT2 (" Popping reg: %d\n", this_reg); \
1193 reg_info[this_reg].word = POP_FAILURE_ITEM (); \
1194 DEBUG_PRINT2 (" info: 0x%x\n", reg_info[this_reg]); \
1196 regend[this_reg] = (const char *) POP_FAILURE_ITEM (); \
1197 DEBUG_PRINT2 (" end: 0x%x\n", regend[this_reg]); \
1199 regstart[this_reg] = (const char *) POP_FAILURE_ITEM (); \
1200 DEBUG_PRINT2 (" start: 0x%x\n", regstart[this_reg]); \
1203 set_regs_matched_done = 0; \
1204 DEBUG_STATEMENT (nfailure_points_popped++); \
1205 } /* POP_FAILURE_POINT */
1209 /* Structure for per-register (a.k.a. per-group) information.
1210 This must not be longer than one word, because we push this value
1211 onto the failure stack. Other register information, such as the
1212 starting and ending positions (which are addresses), and the list of
1213 inner groups (which is a bits list) are maintained in separate
1216 We are making a (strictly speaking) nonportable assumption here: that
1217 the compiler will pack our bit fields into something that fits into
1218 the type of `word', i.e., is something that fits into one item on the
1222 fail_stack_elt_t word;
1225 /* This field is one if this group can match the empty string,
1226 zero if not. If not yet determined, `MATCH_NULL_UNSET_VALUE'. */
1227 #define MATCH_NULL_UNSET_VALUE 3
1228 unsigned match_null_string_p : 2;
1229 unsigned is_active : 1;
1230 unsigned matched_something : 1;
1231 unsigned ever_matched_something : 1;
1233 } register_info_type;
1235 #define REG_MATCH_NULL_STRING_P(R) ((R).bits.match_null_string_p)
1236 #define IS_ACTIVE(R) ((R).bits.is_active)
1237 #define MATCHED_SOMETHING(R) ((R).bits.matched_something)
1238 #define EVER_MATCHED_SOMETHING(R) ((R).bits.ever_matched_something)
1241 /* Call this when have matched a real character; it sets `matched' flags
1242 for the subexpressions which we are currently inside. Also records
1243 that those subexprs have matched. */
1244 #define SET_REGS_MATCHED() \
1247 if (!set_regs_matched_done) \
1250 set_regs_matched_done = 1; \
1251 for (r = lowest_active_reg; r <= highest_active_reg; r++) \
1253 MATCHED_SOMETHING (reg_info[r]) \
1254 = EVER_MATCHED_SOMETHING (reg_info[r]) \
1261 /* Registers are set to a sentinel when they haven't yet matched. */
1262 static char reg_unset_dummy;
1263 #define REG_UNSET_VALUE (®_unset_dummy)
1264 #define REG_UNSET(e) ((e) == REG_UNSET_VALUE)
1268 /* How do we implement a missing MATCH_MAY_ALLOCATE?
1269 We make the fail stack a global thing, and then grow it to
1270 re_max_failures when we compile. */
1271 #ifndef MATCH_MAY_ALLOCATE
1272 static fail_stack_type fail_stack;
1274 static const char ** regstart, ** regend;
1275 static const char ** old_regstart, ** old_regend;
1276 static const char **best_regstart, **best_regend;
1277 static register_info_type *reg_info;
1278 static const char **reg_dummy;
1279 static register_info_type *reg_info_dummy;
1283 /* Subroutine declarations and macros for regex_compile. */
1285 static void store_op1 (), store_op2 ();
1286 static void insert_op1 (), insert_op2 ();
1287 static boolean at_begline_loc_p (), at_endline_loc_p ();
1288 static boolean group_in_compile_stack ();
1289 static reg_errcode_t compile_range ();
1291 /* Fetch the next character in the uncompiled pattern---translating it
1292 if necessary. Also cast from a signed character in the constant
1293 string passed to us by the user to an unsigned char that we can use
1294 as an array index (in, e.g., `translate'). */
1295 #define PATFETCH(c) \
1296 do {if (p == pend) return REG_EEND; \
1297 c = (unsigned char) *p++; \
1298 if (translate) c = translate[c]; \
1301 /* Fetch the next character in the uncompiled pattern, with no
1303 #define PATFETCH_RAW(c) \
1304 do {if (p == pend) return REG_EEND; \
1305 c = (unsigned char) *p++; \
1308 /* Go backwards one character in the pattern. */
1309 #define PATUNFETCH p--
1312 /* If `translate' is non-null, return translate[D], else just D. We
1313 cast the subscript to translate because some data is declared as
1314 `char *', to avoid warnings when a string constant is passed. But
1315 when we use a character as a subscript we must make it unsigned. */
1316 #define TRANSLATE(d) (translate ? translate[(unsigned char) (d)] : (d))
1319 /* Macros for outputting the compiled pattern into `buffer'. */
1321 /* If the buffer isn't allocated when it comes in, use this. */
1322 #define INIT_BUF_SIZE 32
1324 /* Make sure we have at least N more bytes of space in buffer. */
1325 #define GET_BUFFER_SPACE(n) \
1326 while (b - bufp->buffer + (n) > bufp->allocated) \
1329 /* Make sure we have one more byte of buffer space and then add C to it. */
1330 #define BUF_PUSH(c) \
1332 GET_BUFFER_SPACE (1); \
1333 *b++ = (unsigned char) (c); \
1337 /* Ensure we have two more bytes of buffer space and then append C1 and C2. */
1338 #define BUF_PUSH_2(c1, c2) \
1340 GET_BUFFER_SPACE (2); \
1341 *b++ = (unsigned char) (c1); \
1342 *b++ = (unsigned char) (c2); \
1346 /* As with BUF_PUSH_2, except for three bytes. */
1347 #define BUF_PUSH_3(c1, c2, c3) \
1349 GET_BUFFER_SPACE (3); \
1350 *b++ = (unsigned char) (c1); \
1351 *b++ = (unsigned char) (c2); \
1352 *b++ = (unsigned char) (c3); \
1356 /* Store a jump with opcode OP at LOC to location TO. We store a
1357 relative address offset by the three bytes the jump itself occupies. */
1358 #define STORE_JUMP(op, loc, to) \
1359 store_op1 (op, loc, (to) - (loc) - 3)
1361 /* Likewise, for a two-argument jump. */
1362 #define STORE_JUMP2(op, loc, to, arg) \
1363 store_op2 (op, loc, (to) - (loc) - 3, arg)
1365 /* Like `STORE_JUMP', but for inserting. Assume `b' is the buffer end. */
1366 #define INSERT_JUMP(op, loc, to) \
1367 insert_op1 (op, loc, (to) - (loc) - 3, b)
1369 /* Like `STORE_JUMP2', but for inserting. Assume `b' is the buffer end. */
1370 #define INSERT_JUMP2(op, loc, to, arg) \
1371 insert_op2 (op, loc, (to) - (loc) - 3, arg, b)
1374 /* This is not an arbitrary limit: the arguments which represent offsets
1375 into the pattern are two bytes long. So if 2^16 bytes turns out to
1376 be too small, many things would have to change. */
1377 #define MAX_BUF_SIZE (1L << 16)
1380 /* Extend the buffer by twice its current size via realloc and
1381 reset the pointers that pointed into the old block to point to the
1382 correct places in the new one. If extending the buffer results in it
1383 being larger than MAX_BUF_SIZE, then flag memory exhausted. */
1384 #define EXTEND_BUFFER() \
1386 unsigned char *old_buffer = bufp->buffer; \
1387 if (bufp->allocated == MAX_BUF_SIZE) \
1389 bufp->allocated <<= 1; \
1390 if (bufp->allocated > MAX_BUF_SIZE) \
1391 bufp->allocated = MAX_BUF_SIZE; \
1392 bufp->buffer = (unsigned char *) realloc (bufp->buffer, bufp->allocated);\
1393 if (bufp->buffer == NULL) \
1394 return REG_ESPACE; \
1395 /* If the buffer moved, move all the pointers into it. */ \
1396 if (old_buffer != bufp->buffer) \
1398 b = (b - old_buffer) + bufp->buffer; \
1399 begalt = (begalt - old_buffer) + bufp->buffer; \
1400 if (fixup_alt_jump) \
1401 fixup_alt_jump = (fixup_alt_jump - old_buffer) + bufp->buffer;\
1403 laststart = (laststart - old_buffer) + bufp->buffer; \
1404 if (pending_exact) \
1405 pending_exact = (pending_exact - old_buffer) + bufp->buffer; \
1410 /* Since we have one byte reserved for the register number argument to
1411 {start,stop}_memory, the maximum number of groups we can report
1412 things about is what fits in that byte. */
1413 #define MAX_REGNUM 255
1415 /* But patterns can have more than `MAX_REGNUM' registers. We just
1416 ignore the excess. */
1417 typedef unsigned regnum_t;
1420 /* Macros for the compile stack. */
1422 /* Since offsets can go either forwards or backwards, this type needs to
1423 be able to hold values from -(MAX_BUF_SIZE - 1) to MAX_BUF_SIZE - 1. */
1424 typedef int pattern_offset_t;
1428 pattern_offset_t begalt_offset;
1429 pattern_offset_t fixup_alt_jump;
1430 pattern_offset_t inner_group_offset;
1431 pattern_offset_t laststart_offset;
1433 } compile_stack_elt_t;
1438 compile_stack_elt_t *stack;
1440 unsigned avail; /* Offset of next open position. */
1441 } compile_stack_type;
1444 #define INIT_COMPILE_STACK_SIZE 32
1446 #define COMPILE_STACK_EMPTY (compile_stack.avail == 0)
1447 #define COMPILE_STACK_FULL (compile_stack.avail == compile_stack.size)
1449 /* The next available element. */
1450 #define COMPILE_STACK_TOP (compile_stack.stack[compile_stack.avail])
1453 /* Set the bit for character C in a list. */
1454 #define SET_LIST_BIT(c) \
1455 (b[((unsigned char) (c)) / BYTEWIDTH] \
1456 |= 1 << (((unsigned char) c) % BYTEWIDTH))
1459 /* Get the next unsigned number in the uncompiled pattern. */
1460 #define GET_UNSIGNED_NUMBER(num) \
1464 while (ISDIGIT (c)) \
1468 num = num * 10 + c - '0'; \
1476 #define CHAR_CLASS_MAX_LENGTH 6 /* Namely, `xdigit'. */
1478 #define IS_CHAR_CLASS(string) \
1479 (STREQ (string, "alpha") || STREQ (string, "upper") \
1480 || STREQ (string, "lower") || STREQ (string, "digit") \
1481 || STREQ (string, "alnum") || STREQ (string, "xdigit") \
1482 || STREQ (string, "space") || STREQ (string, "print") \
1483 || STREQ (string, "punct") || STREQ (string, "graph") \
1484 || STREQ (string, "cntrl") || STREQ (string, "blank"))
1486 /* `regex_compile' compiles PATTERN (of length SIZE) according to SYNTAX.
1487 Returns one of error codes defined in `regex.h', or zero for success.
1489 Assumes the `allocated' (and perhaps `buffer') and `translate'
1490 fields are set in BUFP on entry.
1492 If it succeeds, results are put in BUFP (if it returns an error, the
1493 contents of BUFP are undefined):
1494 `buffer' is the compiled pattern;
1495 `syntax' is set to SYNTAX;
1496 `used' is set to the length of the compiled pattern;
1497 `fastmap_accurate' is zero;
1498 `re_nsub' is the number of subexpressions in PATTERN;
1499 `not_bol' and `not_eol' are zero;
1501 The `fastmap' and `newline_anchor' fields are neither
1502 examined nor set. */
1504 /* Return, freeing storage we allocated. */
1505 #define FREE_STACK_RETURN(value) \
1506 return (free (compile_stack.stack), value)
1508 static reg_errcode_t
1509 regex_compile (pattern, size, syntax, bufp)
1510 const char *pattern;
1512 reg_syntax_t syntax;
1513 struct re_pattern_buffer *bufp;
1515 /* We fetch characters from PATTERN here. Even though PATTERN is
1516 `char *' (i.e., signed), we declare these variables as unsigned, so
1517 they can be reliably used as array indices. */
1518 register unsigned char c, c1;
1520 /* A random temporary spot in PATTERN. */
1523 /* Points to the end of the buffer, where we should append. */
1524 register unsigned char *b;
1526 /* Keeps track of unclosed groups. */
1527 compile_stack_type compile_stack;
1529 /* Points to the current (ending) position in the pattern. */
1530 const char *p = pattern;
1531 const char *pend = pattern + size;
1533 /* How to translate the characters in the pattern. */
1534 char *translate = bufp->translate;
1536 /* Address of the count-byte of the most recently inserted `exactn'
1537 command. This makes it possible to tell if a new exact-match
1538 character can be added to that command or if the character requires
1539 a new `exactn' command. */
1540 unsigned char *pending_exact = 0;
1542 /* Address of start of the most recently finished expression.
1543 This tells, e.g., postfix * where to find the start of its
1544 operand. Reset at the beginning of groups and alternatives. */
1545 unsigned char *laststart = 0;
1547 /* Address of beginning of regexp, or inside of last group. */
1548 unsigned char *begalt;
1550 /* Place in the uncompiled pattern (i.e., the {) to
1551 which to go back if the interval is invalid. */
1552 const char *beg_interval;
1554 /* Address of the place where a forward jump should go to the end of
1555 the containing expression. Each alternative of an `or' -- except the
1556 last -- ends with a forward jump of this sort. */
1557 unsigned char *fixup_alt_jump = 0;
1559 /* Counts open-groups as they are encountered. Remembered for the
1560 matching close-group on the compile stack, so the same register
1561 number is put in the stop_memory as the start_memory. */
1562 regnum_t regnum = 0;
1565 DEBUG_PRINT1 ("\nCompiling pattern: ");
1568 unsigned debug_count;
1570 for (debug_count = 0; debug_count < size; debug_count++)
1571 putchar (pattern[debug_count]);
1576 /* Initialize the compile stack. */
1577 compile_stack.stack = TALLOC (INIT_COMPILE_STACK_SIZE, compile_stack_elt_t);
1578 if (compile_stack.stack == NULL)
1581 compile_stack.size = INIT_COMPILE_STACK_SIZE;
1582 compile_stack.avail = 0;
1584 /* Initialize the pattern buffer. */
1585 bufp->syntax = syntax;
1586 bufp->fastmap_accurate = 0;
1587 bufp->not_bol = bufp->not_eol = 0;
1589 /* Set `used' to zero, so that if we return an error, the pattern
1590 printer (for debugging) will think there's no pattern. We reset it
1594 /* Always count groups, whether or not bufp->no_sub is set. */
1597 #if !defined (emacs) && !defined (SYNTAX_TABLE)
1598 /* Initialize the syntax table. */
1599 init_syntax_once ();
1602 if (bufp->allocated == 0)
1605 { /* If zero allocated, but buffer is non-null, try to realloc
1606 enough space. This loses if buffer's address is bogus, but
1607 that is the user's responsibility. */
1608 RETALLOC (bufp->buffer, INIT_BUF_SIZE, unsigned char);
1611 { /* Caller did not allocate a buffer. Do it for them. */
1612 bufp->buffer = TALLOC (INIT_BUF_SIZE, unsigned char);
1614 if (!bufp->buffer) FREE_STACK_RETURN (REG_ESPACE);
1616 bufp->allocated = INIT_BUF_SIZE;
1619 begalt = b = bufp->buffer;
1621 /* Loop through the uncompiled pattern until we're at the end. */
1630 if ( /* If at start of pattern, it's an operator. */
1632 /* If context independent, it's an operator. */
1633 || syntax & RE_CONTEXT_INDEP_ANCHORS
1634 /* Otherwise, depends on what's come before. */
1635 || at_begline_loc_p (pattern, p, syntax))
1645 if ( /* If at end of pattern, it's an operator. */
1647 /* If context independent, it's an operator. */
1648 || syntax & RE_CONTEXT_INDEP_ANCHORS
1649 /* Otherwise, depends on what's next. */
1650 || at_endline_loc_p (p, pend, syntax))
1660 if ((syntax & RE_BK_PLUS_QM)
1661 || (syntax & RE_LIMITED_OPS))
1665 /* If there is no previous pattern... */
1668 if (syntax & RE_CONTEXT_INVALID_OPS)
1669 FREE_STACK_RETURN (REG_BADRPT);
1670 else if (!(syntax & RE_CONTEXT_INDEP_OPS))
1675 /* Are we optimizing this jump? */
1676 boolean keep_string_p = false;
1678 /* 1 means zero (many) matches is allowed. */
1679 char zero_times_ok = 0, many_times_ok = 0;
1681 /* If there is a sequence of repetition chars, collapse it
1682 down to just one (the right one). We can't combine
1683 interval operators with these because of, e.g., `a{2}*',
1684 which should only match an even number of `a's. */
1688 zero_times_ok |= c != '+';
1689 many_times_ok |= c != '?';
1697 || (!(syntax & RE_BK_PLUS_QM) && (c == '+' || c == '?')))
1700 else if (syntax & RE_BK_PLUS_QM && c == '\\')
1702 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
1705 if (!(c1 == '+' || c1 == '?'))
1720 /* If we get here, we found another repeat character. */
1723 /* Star, etc. applied to an empty pattern is equivalent
1724 to an empty pattern. */
1728 /* Now we know whether or not zero matches is allowed
1729 and also whether or not two or more matches is allowed. */
1731 { /* More than one repetition is allowed, so put in at the
1732 end a backward relative jump from `b' to before the next
1733 jump we're going to put in below (which jumps from
1734 laststart to after this jump).
1736 But if we are at the `*' in the exact sequence `.*\n',
1737 insert an unconditional jump backwards to the .,
1738 instead of the beginning of the loop. This way we only
1739 push a failure point once, instead of every time
1740 through the loop. */
1741 assert (p - 1 > pattern);
1743 /* Allocate the space for the jump. */
1744 GET_BUFFER_SPACE (3);
1746 /* We know we are not at the first character of the pattern,
1747 because laststart was nonzero. And we've already
1748 incremented `p', by the way, to be the character after
1749 the `*'. Do we have to do something analogous here
1750 for null bytes, because of RE_DOT_NOT_NULL? */
1751 if (TRANSLATE (*(p - 2)) == TRANSLATE ('.')
1753 && p < pend && TRANSLATE (*p) == TRANSLATE ('\n')
1754 && !(syntax & RE_DOT_NEWLINE))
1755 { /* We have .*\n. */
1756 STORE_JUMP (jump, b, laststart);
1757 keep_string_p = true;
1760 /* Anything else. */
1761 STORE_JUMP (maybe_pop_jump, b, laststart - 3);
1763 /* We've added more stuff to the buffer. */
1767 /* On failure, jump from laststart to b + 3, which will be the
1768 end of the buffer after this jump is inserted. */
1769 GET_BUFFER_SPACE (3);
1770 INSERT_JUMP (keep_string_p ? on_failure_keep_string_jump
1778 /* At least one repetition is required, so insert a
1779 `dummy_failure_jump' before the initial
1780 `on_failure_jump' instruction of the loop. This
1781 effects a skip over that instruction the first time
1782 we hit that loop. */
1783 GET_BUFFER_SPACE (3);
1784 INSERT_JUMP (dummy_failure_jump, laststart, laststart + 6);
1799 boolean had_char_class = false;
1801 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
1803 /* Ensure that we have enough space to push a charset: the
1804 opcode, the length count, and the bitset; 34 bytes in all. */
1805 GET_BUFFER_SPACE (34);
1809 /* We test `*p == '^' twice, instead of using an if
1810 statement, so we only need one BUF_PUSH. */
1811 BUF_PUSH (*p == '^' ? charset_not : charset);
1815 /* Remember the first position in the bracket expression. */
1818 /* Push the number of bytes in the bitmap. */
1819 BUF_PUSH ((1 << BYTEWIDTH) / BYTEWIDTH);
1821 /* Clear the whole map. */
1822 bzero (b, (1 << BYTEWIDTH) / BYTEWIDTH);
1824 /* charset_not matches newline according to a syntax bit. */
1825 if ((re_opcode_t) b[-2] == charset_not
1826 && (syntax & RE_HAT_LISTS_NOT_NEWLINE))
1827 SET_LIST_BIT ('\n');
1829 /* Read in characters and ranges, setting map bits. */
1832 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
1836 /* \ might escape characters inside [...] and [^...]. */
1837 if ((syntax & RE_BACKSLASH_ESCAPE_IN_LISTS) && c == '\\')
1839 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
1846 /* Could be the end of the bracket expression. If it's
1847 not (i.e., when the bracket expression is `[]' so
1848 far), the ']' character bit gets set way below. */
1849 if (c == ']' && p != p1 + 1)
1852 /* Look ahead to see if it's a range when the last thing
1853 was a character class. */
1854 if (had_char_class && c == '-' && *p != ']')
1855 FREE_STACK_RETURN (REG_ERANGE);
1857 /* Look ahead to see if it's a range when the last thing
1858 was a character: if this is a hyphen not at the
1859 beginning or the end of a list, then it's the range
1862 && !(p - 2 >= pattern && p[-2] == '[')
1863 && !(p - 3 >= pattern && p[-3] == '[' && p[-2] == '^')
1867 = compile_range (&p, pend, translate, syntax, b);
1868 if (ret != REG_NOERROR) FREE_STACK_RETURN (ret);
1871 else if (p[0] == '-' && p[1] != ']')
1872 { /* This handles ranges made up of characters only. */
1875 /* Move past the `-'. */
1878 ret = compile_range (&p, pend, translate, syntax, b);
1879 if (ret != REG_NOERROR) FREE_STACK_RETURN (ret);
1882 /* See if we're at the beginning of a possible character
1885 else if (syntax & RE_CHAR_CLASSES && c == '[' && *p == ':')
1886 { /* Leave room for the null. */
1887 char str[CHAR_CLASS_MAX_LENGTH + 1];
1892 /* If pattern is `[[:'. */
1893 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
1898 if (c == ':' || c == ']' || p == pend
1899 || c1 == CHAR_CLASS_MAX_LENGTH)
1905 /* If isn't a word bracketed by `[:' and:`]':
1906 undo the ending character, the letters, and leave
1907 the leading `:' and `[' (but set bits for them). */
1908 if (c == ':' && *p == ']')
1911 boolean is_alnum = STREQ (str, "alnum");
1912 boolean is_alpha = STREQ (str, "alpha");
1913 boolean is_blank = STREQ (str, "blank");
1914 boolean is_cntrl = STREQ (str, "cntrl");
1915 boolean is_digit = STREQ (str, "digit");
1916 boolean is_graph = STREQ (str, "graph");
1917 boolean is_lower = STREQ (str, "lower");
1918 boolean is_print = STREQ (str, "print");
1919 boolean is_punct = STREQ (str, "punct");
1920 boolean is_space = STREQ (str, "space");
1921 boolean is_upper = STREQ (str, "upper");
1922 boolean is_xdigit = STREQ (str, "xdigit");
1924 if (!IS_CHAR_CLASS (str))
1925 FREE_STACK_RETURN (REG_ECTYPE);
1927 /* Throw away the ] at the end of the character
1931 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
1933 for (ch = 0; ch < 1 << BYTEWIDTH; ch++)
1935 /* This was split into 3 if's to
1936 avoid an arbitrary limit in some compiler. */
1937 if ( (is_alnum && ISALNUM (ch))
1938 || (is_alpha && ISALPHA (ch))
1939 || (is_blank && ISBLANK (ch))
1940 || (is_cntrl && ISCNTRL (ch)))
1942 if ( (is_digit && ISDIGIT (ch))
1943 || (is_graph && ISGRAPH (ch))
1944 || (is_lower && ISLOWER (ch))
1945 || (is_print && ISPRINT (ch)))
1947 if ( (is_punct && ISPUNCT (ch))
1948 || (is_space && ISSPACE (ch))
1949 || (is_upper && ISUPPER (ch))
1950 || (is_xdigit && ISXDIGIT (ch)))
1953 had_char_class = true;
1962 had_char_class = false;
1967 had_char_class = false;
1972 /* Discard any (non)matching list bytes that are all 0 at the
1973 end of the map. Decrease the map-length byte too. */
1974 while ((int) b[-1] > 0 && b[b[-1] - 1] == 0)
1982 if (syntax & RE_NO_BK_PARENS)
1989 if (syntax & RE_NO_BK_PARENS)
1996 if (syntax & RE_NEWLINE_ALT)
2003 if (syntax & RE_NO_BK_VBAR)
2010 if (syntax & RE_INTERVALS && syntax & RE_NO_BK_BRACES)
2011 goto handle_interval;
2017 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
2019 /* Do not translate the character after the \, so that we can
2020 distinguish, e.g., \B from \b, even if we normally would
2021 translate, e.g., B to b. */
2027 if (syntax & RE_NO_BK_PARENS)
2028 goto normal_backslash;
2034 if (COMPILE_STACK_FULL)
2036 RETALLOC (compile_stack.stack, compile_stack.size << 1,
2037 compile_stack_elt_t);
2038 if (compile_stack.stack == NULL) return REG_ESPACE;
2040 compile_stack.size <<= 1;
2043 /* These are the values to restore when we hit end of this
2044 group. They are all relative offsets, so that if the
2045 whole pattern moves because of realloc, they will still
2047 COMPILE_STACK_TOP.begalt_offset = begalt - bufp->buffer;
2048 COMPILE_STACK_TOP.fixup_alt_jump
2049 = fixup_alt_jump ? fixup_alt_jump - bufp->buffer + 1 : 0;
2050 COMPILE_STACK_TOP.laststart_offset = b - bufp->buffer;
2051 COMPILE_STACK_TOP.regnum = regnum;
2053 /* We will eventually replace the 0 with the number of
2054 groups inner to this one. But do not push a
2055 start_memory for groups beyond the last one we can
2056 represent in the compiled pattern. */
2057 if (regnum <= MAX_REGNUM)
2059 COMPILE_STACK_TOP.inner_group_offset = b - bufp->buffer + 2;
2060 BUF_PUSH_3 (start_memory, regnum, 0);
2063 compile_stack.avail++;
2068 /* If we've reached MAX_REGNUM groups, then this open
2069 won't actually generate any code, so we'll have to
2070 clear pending_exact explicitly. */
2076 if (syntax & RE_NO_BK_PARENS) goto normal_backslash;
2078 if (COMPILE_STACK_EMPTY)
2079 if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD)
2080 goto normal_backslash;
2082 FREE_STACK_RETURN (REG_ERPAREN);
2086 { /* Push a dummy failure point at the end of the
2087 alternative for a possible future
2088 `pop_failure_jump' to pop. See comments at
2089 `push_dummy_failure' in `re_match_2'. */
2090 BUF_PUSH (push_dummy_failure);
2092 /* We allocated space for this jump when we assigned
2093 to `fixup_alt_jump', in the `handle_alt' case below. */
2094 STORE_JUMP (jump_past_alt, fixup_alt_jump, b - 1);
2097 /* See similar code for backslashed left paren above. */
2098 if (COMPILE_STACK_EMPTY)
2099 if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD)
2102 FREE_STACK_RETURN (REG_ERPAREN);
2104 /* Since we just checked for an empty stack above, this
2105 ``can't happen''. */
2106 assert (compile_stack.avail != 0);
2108 /* We don't just want to restore into `regnum', because
2109 later groups should continue to be numbered higher,
2110 as in `(ab)c(de)' -- the second group is #2. */
2111 regnum_t this_group_regnum;
2113 compile_stack.avail--;
2114 begalt = bufp->buffer + COMPILE_STACK_TOP.begalt_offset;
2116 = COMPILE_STACK_TOP.fixup_alt_jump
2117 ? bufp->buffer + COMPILE_STACK_TOP.fixup_alt_jump - 1
2119 laststart = bufp->buffer + COMPILE_STACK_TOP.laststart_offset;
2120 this_group_regnum = COMPILE_STACK_TOP.regnum;
2121 /* If we've reached MAX_REGNUM groups, then this open
2122 won't actually generate any code, so we'll have to
2123 clear pending_exact explicitly. */
2126 /* We're at the end of the group, so now we know how many
2127 groups were inside this one. */
2128 if (this_group_regnum <= MAX_REGNUM)
2130 unsigned char *inner_group_loc
2131 = bufp->buffer + COMPILE_STACK_TOP.inner_group_offset;
2133 *inner_group_loc = regnum - this_group_regnum;
2134 BUF_PUSH_3 (stop_memory, this_group_regnum,
2135 regnum - this_group_regnum);
2141 case '|': /* `\|'. */
2142 if (syntax & RE_LIMITED_OPS || syntax & RE_NO_BK_VBAR)
2143 goto normal_backslash;
2145 if (syntax & RE_LIMITED_OPS)
2148 /* Insert before the previous alternative a jump which
2149 jumps to this alternative if the former fails. */
2150 GET_BUFFER_SPACE (3);
2151 INSERT_JUMP (on_failure_jump, begalt, b + 6);
2155 /* The alternative before this one has a jump after it
2156 which gets executed if it gets matched. Adjust that
2157 jump so it will jump to this alternative's analogous
2158 jump (put in below, which in turn will jump to the next
2159 (if any) alternative's such jump, etc.). The last such
2160 jump jumps to the correct final destination. A picture:
2166 If we are at `b', then fixup_alt_jump right now points to a
2167 three-byte space after `a'. We'll put in the jump, set
2168 fixup_alt_jump to right after `b', and leave behind three
2169 bytes which we'll fill in when we get to after `c'. */
2172 STORE_JUMP (jump_past_alt, fixup_alt_jump, b);
2174 /* Mark and leave space for a jump after this alternative,
2175 to be filled in later either by next alternative or
2176 when know we're at the end of a series of alternatives. */
2178 GET_BUFFER_SPACE (3);
2187 /* If \{ is a literal. */
2188 if (!(syntax & RE_INTERVALS)
2189 /* If we're at `\{' and it's not the open-interval
2191 || ((syntax & RE_INTERVALS) && (syntax & RE_NO_BK_BRACES))
2192 || (p - 2 == pattern && p == pend))
2193 goto normal_backslash;
2197 /* If got here, then the syntax allows intervals. */
2199 /* At least (most) this many matches must be made. */
2200 int lower_bound = -1, upper_bound = -1;
2202 beg_interval = p - 1;
2206 if (syntax & RE_NO_BK_BRACES)
2207 goto unfetch_interval;
2209 FREE_STACK_RETURN (REG_EBRACE);
2212 GET_UNSIGNED_NUMBER (lower_bound);
2216 GET_UNSIGNED_NUMBER (upper_bound);
2217 if (upper_bound < 0) upper_bound = RE_DUP_MAX;
2220 /* Interval such as `{1}' => match exactly once. */
2221 upper_bound = lower_bound;
2223 if (lower_bound < 0 || upper_bound > RE_DUP_MAX
2224 || lower_bound > upper_bound)
2226 if (syntax & RE_NO_BK_BRACES)
2227 goto unfetch_interval;
2229 FREE_STACK_RETURN (REG_BADBR);
2232 if (!(syntax & RE_NO_BK_BRACES))
2234 if (c != '\\') FREE_STACK_RETURN (REG_EBRACE);
2241 if (syntax & RE_NO_BK_BRACES)
2242 goto unfetch_interval;
2244 FREE_STACK_RETURN (REG_BADBR);
2247 /* We just parsed a valid interval. */
2249 /* If it's invalid to have no preceding re. */
2252 if (syntax & RE_CONTEXT_INVALID_OPS)
2253 FREE_STACK_RETURN (REG_BADRPT);
2254 else if (syntax & RE_CONTEXT_INDEP_OPS)
2257 goto unfetch_interval;
2260 /* If the upper bound is zero, don't want to succeed at
2261 all; jump from `laststart' to `b + 3', which will be
2262 the end of the buffer after we insert the jump. */
2263 if (upper_bound == 0)
2265 GET_BUFFER_SPACE (3);
2266 INSERT_JUMP (jump, laststart, b + 3);
2270 /* Otherwise, we have a nontrivial interval. When
2271 we're all done, the pattern will look like:
2272 set_number_at <jump count> <upper bound>
2273 set_number_at <succeed_n count> <lower bound>
2274 succeed_n <after jump addr> <succeed_n count>
2276 jump_n <succeed_n addr> <jump count>
2277 (The upper bound and `jump_n' are omitted if
2278 `upper_bound' is 1, though.) */
2280 { /* If the upper bound is > 1, we need to insert
2281 more at the end of the loop. */
2282 unsigned nbytes = 10 + (upper_bound > 1) * 10;
2284 GET_BUFFER_SPACE (nbytes);
2286 /* Initialize lower bound of the `succeed_n', even
2287 though it will be set during matching by its
2288 attendant `set_number_at' (inserted next),
2289 because `re_compile_fastmap' needs to know.
2290 Jump to the `jump_n' we might insert below. */
2291 INSERT_JUMP2 (succeed_n, laststart,
2292 b + 5 + (upper_bound > 1) * 5,
2296 /* Code to initialize the lower bound. Insert
2297 before the `succeed_n'. The `5' is the last two
2298 bytes of this `set_number_at', plus 3 bytes of
2299 the following `succeed_n'. */
2300 insert_op2 (set_number_at, laststart, 5, lower_bound, b);
2303 if (upper_bound > 1)
2304 { /* More than one repetition is allowed, so
2305 append a backward jump to the `succeed_n'
2306 that starts this interval.
2308 When we've reached this during matching,
2309 we'll have matched the interval once, so
2310 jump back only `upper_bound - 1' times. */
2311 STORE_JUMP2 (jump_n, b, laststart + 5,
2315 /* The location we want to set is the second
2316 parameter of the `jump_n'; that is `b-2' as
2317 an absolute address. `laststart' will be
2318 the `set_number_at' we're about to insert;
2319 `laststart+3' the number to set, the source
2320 for the relative address. But we are
2321 inserting into the middle of the pattern --
2322 so everything is getting moved up by 5.
2323 Conclusion: (b - 2) - (laststart + 3) + 5,
2324 i.e., b - laststart.
2326 We insert this at the beginning of the loop
2327 so that if we fail during matching, we'll
2328 reinitialize the bounds. */
2329 insert_op2 (set_number_at, laststart, b - laststart,
2330 upper_bound - 1, b);
2335 beg_interval = NULL;
2340 /* If an invalid interval, match the characters as literals. */
2341 assert (beg_interval);
2343 beg_interval = NULL;
2345 /* normal_char and normal_backslash need `c'. */
2348 if (!(syntax & RE_NO_BK_BRACES))
2350 if (p > pattern && p[-1] == '\\')
2351 goto normal_backslash;
2356 /* There is no way to specify the before_dot and after_dot
2357 operators. rms says this is ok. --karl */
2365 BUF_PUSH_2 (syntaxspec, syntax_spec_code[c]);
2371 BUF_PUSH_2 (notsyntaxspec, syntax_spec_code[c]);
2378 BUF_PUSH (wordchar);
2384 BUF_PUSH (notwordchar);
2397 BUF_PUSH (wordbound);
2401 BUF_PUSH (notwordbound);
2412 case '1': case '2': case '3': case '4': case '5':
2413 case '6': case '7': case '8': case '9':
2414 if (syntax & RE_NO_BK_REFS)
2420 FREE_STACK_RETURN (REG_ESUBREG);
2422 /* Can't back reference to a subexpression if inside of it. */
2423 if (group_in_compile_stack (compile_stack, c1))
2427 BUF_PUSH_2 (duplicate, c1);
2433 if (syntax & RE_BK_PLUS_QM)
2436 goto normal_backslash;
2440 /* You might think it would be useful for \ to mean
2441 not to translate; but if we don't translate it
2442 it will never match anything. */
2450 /* Expects the character in `c'. */
2452 /* If no exactn currently being built. */
2455 /* If last exactn not at current position. */
2456 || pending_exact + *pending_exact + 1 != b
2458 /* We have only one byte following the exactn for the count. */
2459 || *pending_exact == (1 << BYTEWIDTH) - 1
2461 /* If followed by a repetition operator. */
2462 || *p == '*' || *p == '^'
2463 || ((syntax & RE_BK_PLUS_QM)
2464 ? *p == '\\' && (p[1] == '+' || p[1] == '?')
2465 : (*p == '+' || *p == '?'))
2466 || ((syntax & RE_INTERVALS)
2467 && ((syntax & RE_NO_BK_BRACES)
2469 : (p[0] == '\\' && p[1] == '{'))))
2471 /* Start building a new exactn. */
2475 BUF_PUSH_2 (exactn, 0);
2476 pending_exact = b - 1;
2483 } /* while p != pend */
2486 /* Through the pattern now. */
2489 STORE_JUMP (jump_past_alt, fixup_alt_jump, b);
2491 if (!COMPILE_STACK_EMPTY)
2492 FREE_STACK_RETURN (REG_EPAREN);
2494 /* If we don't want backtracking, force success
2495 the first time we reach the end of the compiled pattern. */
2496 if (syntax & RE_NO_POSIX_BACKTRACKING)
2499 free (compile_stack.stack);
2501 /* We have succeeded; set the length of the buffer. */
2502 bufp->used = b - bufp->buffer;
2507 DEBUG_PRINT1 ("\nCompiled pattern: \n");
2508 print_compiled_pattern (bufp);
2512 #ifndef MATCH_MAY_ALLOCATE
2513 /* Initialize the failure stack to the largest possible stack. This
2514 isn't necessary unless we're trying to avoid calling alloca in
2515 the search and match routines. */
2517 int num_regs = bufp->re_nsub + 1;
2519 /* Since DOUBLE_FAIL_STACK refuses to double only if the current size
2520 is strictly greater than re_max_failures, the largest possible stack
2521 is 2 * re_max_failures failure points. */
2522 if (fail_stack.size < (2 * re_max_failures * MAX_FAILURE_ITEMS))
2524 fail_stack.size = (2 * re_max_failures * MAX_FAILURE_ITEMS);
2527 if (! fail_stack.stack)
2529 = (fail_stack_elt_t *) xmalloc (fail_stack.size
2530 * sizeof (fail_stack_elt_t));
2533 = (fail_stack_elt_t *) xrealloc (fail_stack.stack,
2535 * sizeof (fail_stack_elt_t)));
2536 #else /* not emacs */
2537 if (! fail_stack.stack)
2539 = (fail_stack_elt_t *) malloc (fail_stack.size
2540 * sizeof (fail_stack_elt_t));
2543 = (fail_stack_elt_t *) realloc (fail_stack.stack,
2545 * sizeof (fail_stack_elt_t)));
2546 #endif /* not emacs */
2549 /* Initialize some other variables the matcher uses. */
2550 RETALLOC_IF (regstart, num_regs, const char *);
2551 RETALLOC_IF (regend, num_regs, const char *);
2552 RETALLOC_IF (old_regstart, num_regs, const char *);
2553 RETALLOC_IF (old_regend, num_regs, const char *);
2554 RETALLOC_IF (best_regstart, num_regs, const char *);
2555 RETALLOC_IF (best_regend, num_regs, const char *);
2556 RETALLOC_IF (reg_info, num_regs, register_info_type);
2557 RETALLOC_IF (reg_dummy, num_regs, const char *);
2558 RETALLOC_IF (reg_info_dummy, num_regs, register_info_type);
2563 } /* regex_compile */
2565 /* Subroutines for `regex_compile'. */
2567 /* Store OP at LOC followed by two-byte integer parameter ARG. */
2570 store_op1 (op, loc, arg)
2575 *loc = (unsigned char) op;
2576 STORE_NUMBER (loc + 1, arg);
2580 /* Like `store_op1', but for two two-byte parameters ARG1 and ARG2. */
2583 store_op2 (op, loc, arg1, arg2)
2588 *loc = (unsigned char) op;
2589 STORE_NUMBER (loc + 1, arg1);
2590 STORE_NUMBER (loc + 3, arg2);
2594 /* Copy the bytes from LOC to END to open up three bytes of space at LOC
2595 for OP followed by two-byte integer parameter ARG. */
2598 insert_op1 (op, loc, arg, end)
2604 register unsigned char *pfrom = end;
2605 register unsigned char *pto = end + 3;
2607 while (pfrom != loc)
2610 store_op1 (op, loc, arg);
2614 /* Like `insert_op1', but for two two-byte parameters ARG1 and ARG2. */
2617 insert_op2 (op, loc, arg1, arg2, end)
2623 register unsigned char *pfrom = end;
2624 register unsigned char *pto = end + 5;
2626 while (pfrom != loc)
2629 store_op2 (op, loc, arg1, arg2);
2633 /* P points to just after a ^ in PATTERN. Return true if that ^ comes
2634 after an alternative or a begin-subexpression. We assume there is at
2635 least one character before the ^. */
2638 at_begline_loc_p (pattern, p, syntax)
2639 const char *pattern, *p;
2640 reg_syntax_t syntax;
2642 const char *prev = p - 2;
2643 boolean prev_prev_backslash = prev > pattern && prev[-1] == '\\';
2646 /* After a subexpression? */
2647 (*prev == '(' && (syntax & RE_NO_BK_PARENS || prev_prev_backslash))
2648 /* After an alternative? */
2649 || (*prev == '|' && (syntax & RE_NO_BK_VBAR || prev_prev_backslash));
2653 /* The dual of at_begline_loc_p. This one is for $. We assume there is
2654 at least one character after the $, i.e., `P < PEND'. */
2657 at_endline_loc_p (p, pend, syntax)
2658 const char *p, *pend;
2661 const char *next = p;
2662 boolean next_backslash = *next == '\\';
2663 const char *next_next = p + 1 < pend ? p + 1 : NULL;
2666 /* Before a subexpression? */
2667 (syntax & RE_NO_BK_PARENS ? *next == ')'
2668 : next_backslash && next_next && *next_next == ')')
2669 /* Before an alternative? */
2670 || (syntax & RE_NO_BK_VBAR ? *next == '|'
2671 : next_backslash && next_next && *next_next == '|');
2675 /* Returns true if REGNUM is in one of COMPILE_STACK's elements and
2676 false if it's not. */
2679 group_in_compile_stack (compile_stack, regnum)
2680 compile_stack_type compile_stack;
2685 for (this_element = compile_stack.avail - 1;
2688 if (compile_stack.stack[this_element].regnum == regnum)
2695 /* Read the ending character of a range (in a bracket expression) from the
2696 uncompiled pattern *P_PTR (which ends at PEND). We assume the
2697 starting character is in `P[-2]'. (`P[-1]' is the character `-'.)
2698 Then we set the translation of all bits between the starting and
2699 ending characters (inclusive) in the compiled pattern B.
2701 Return an error code.
2703 We use these short variable names so we can use the same macros as
2704 `regex_compile' itself. */
2706 static reg_errcode_t
2707 compile_range (p_ptr, pend, translate, syntax, b)
2708 const char **p_ptr, *pend;
2710 reg_syntax_t syntax;
2715 const char *p = *p_ptr;
2716 int range_start, range_end;
2721 /* Even though the pattern is a signed `char *', we need to fetch
2722 with unsigned char *'s; if the high bit of the pattern character
2723 is set, the range endpoints will be negative if we fetch using a
2726 We also want to fetch the endpoints without translating them; the
2727 appropriate translation is done in the bit-setting loop below. */
2728 /* The SVR4 compiler on the 3B2 had trouble with unsigned const char *. */
2729 range_start = ((const unsigned char *) p)[-2];
2730 range_end = ((const unsigned char *) p)[0];
2732 /* Have to increment the pointer into the pattern string, so the
2733 caller isn't still at the ending character. */
2736 /* If the start is after the end, the range is empty. */
2737 if (range_start > range_end)
2738 return syntax & RE_NO_EMPTY_RANGES ? REG_ERANGE : REG_NOERROR;
2740 /* Here we see why `this_char' has to be larger than an `unsigned
2741 char' -- the range is inclusive, so if `range_end' == 0xff
2742 (assuming 8-bit characters), we would otherwise go into an infinite
2743 loop, since all characters <= 0xff. */
2744 for (this_char = range_start; this_char <= range_end; this_char++)
2746 SET_LIST_BIT (TRANSLATE (this_char));
2752 /* re_compile_fastmap computes a ``fastmap'' for the compiled pattern in
2753 BUFP. A fastmap records which of the (1 << BYTEWIDTH) possible
2754 characters can start a string that matches the pattern. This fastmap
2755 is used by re_search to skip quickly over impossible starting points.
2757 The caller must supply the address of a (1 << BYTEWIDTH)-byte data
2758 area as BUFP->fastmap.
2760 We set the `fastmap', `fastmap_accurate', and `can_be_null' fields in
2763 Returns 0 if we succeed, -2 if an internal error. */
2766 re_compile_fastmap (bufp)
2767 struct re_pattern_buffer *bufp;
2770 #ifdef MATCH_MAY_ALLOCATE
2771 fail_stack_type fail_stack;
2773 #ifndef REGEX_MALLOC
2776 /* We don't push any register information onto the failure stack. */
2777 unsigned num_regs = 0;
2779 register char *fastmap = bufp->fastmap;
2780 unsigned char *pattern = bufp->buffer;
2781 unsigned long size = bufp->used;
2782 unsigned char *p = pattern;
2783 register unsigned char *pend = pattern + size;
2785 /* Assume that each path through the pattern can be null until
2786 proven otherwise. We set this false at the bottom of switch
2787 statement, to which we get only if a particular path doesn't
2788 match the empty string. */
2789 boolean path_can_be_null = true;
2791 /* We aren't doing a `succeed_n' to begin with. */
2792 boolean succeed_n_p = false;
2794 assert (fastmap != NULL && p != NULL);
2797 bzero (fastmap, 1 << BYTEWIDTH); /* Assume nothing's valid. */
2798 bufp->fastmap_accurate = 1; /* It will be when we're done. */
2799 bufp->can_be_null = 0;
2803 if (p == pend || *p == succeed)
2805 /* We have reached the (effective) end of pattern. */
2806 if (!FAIL_STACK_EMPTY ())
2808 bufp->can_be_null |= path_can_be_null;
2810 /* Reset for next path. */
2811 path_can_be_null = true;
2813 p = fail_stack.stack[--fail_stack.avail];
2821 /* We should never be about to go beyond the end of the pattern. */
2824 switch (SWITCH_ENUM_CAST ((re_opcode_t) *p++))
2827 /* I guess the idea here is to simply not bother with a fastmap
2828 if a backreference is used, since it's too hard to figure out
2829 the fastmap for the corresponding group. Setting
2830 `can_be_null' stops `re_search_2' from using the fastmap, so
2831 that is all we do. */
2833 bufp->can_be_null = 1;
2837 /* Following are the cases which match a character. These end
2846 for (j = *p++ * BYTEWIDTH - 1; j >= 0; j--)
2847 if (p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH)))
2853 /* Chars beyond end of map must be allowed. */
2854 for (j = *p * BYTEWIDTH; j < (1 << BYTEWIDTH); j++)
2857 for (j = *p++ * BYTEWIDTH - 1; j >= 0; j--)
2858 if (!(p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH))))
2864 for (j = 0; j < (1 << BYTEWIDTH); j++)
2865 if (SYNTAX (j) == Sword)
2871 for (j = 0; j < (1 << BYTEWIDTH); j++)
2872 if (SYNTAX (j) != Sword)
2879 int fastmap_newline = fastmap['\n'];
2881 /* `.' matches anything ... */
2882 for (j = 0; j < (1 << BYTEWIDTH); j++)
2885 /* ... except perhaps newline. */
2886 if (!(bufp->syntax & RE_DOT_NEWLINE))
2887 fastmap['\n'] = fastmap_newline;
2889 /* Return if we have already set `can_be_null'; if we have,
2890 then the fastmap is irrelevant. Something's wrong here. */
2891 else if (bufp->can_be_null)
2894 /* Otherwise, have to check alternative paths. */
2901 for (j = 0; j < (1 << BYTEWIDTH); j++)
2902 if (SYNTAX (j) == (enum syntaxcode) k)
2909 for (j = 0; j < (1 << BYTEWIDTH); j++)
2910 if (SYNTAX (j) != (enum syntaxcode) k)
2915 /* All cases after this match the empty string. These end with
2923 #endif /* not emacs */
2935 case push_dummy_failure:
2940 case pop_failure_jump:
2941 case maybe_pop_jump:
2944 case dummy_failure_jump:
2945 EXTRACT_NUMBER_AND_INCR (j, p);
2950 /* Jump backward implies we just went through the body of a
2951 loop and matched nothing. Opcode jumped to should be
2952 `on_failure_jump' or `succeed_n'. Just treat it like an
2953 ordinary jump. For a * loop, it has pushed its failure
2954 point already; if so, discard that as redundant. */
2955 if ((re_opcode_t) *p != on_failure_jump
2956 && (re_opcode_t) *p != succeed_n)
2960 EXTRACT_NUMBER_AND_INCR (j, p);
2963 /* If what's on the stack is where we are now, pop it. */
2964 if (!FAIL_STACK_EMPTY ()
2965 && fail_stack.stack[fail_stack.avail - 1] == p)
2971 case on_failure_jump:
2972 case on_failure_keep_string_jump:
2973 handle_on_failure_jump:
2974 EXTRACT_NUMBER_AND_INCR (j, p);
2976 /* For some patterns, e.g., `(a?)?', `p+j' here points to the
2977 end of the pattern. We don't want to push such a point,
2978 since when we restore it above, entering the switch will
2979 increment `p' past the end of the pattern. We don't need
2980 to push such a point since we obviously won't find any more
2981 fastmap entries beyond `pend'. Such a pattern can match
2982 the null string, though. */
2985 if (!PUSH_PATTERN_OP (p + j, fail_stack))
2989 bufp->can_be_null = 1;
2993 EXTRACT_NUMBER_AND_INCR (k, p); /* Skip the n. */
2994 succeed_n_p = false;
3001 /* Get to the number of times to succeed. */
3004 /* Increment p past the n for when k != 0. */
3005 EXTRACT_NUMBER_AND_INCR (k, p);
3009 succeed_n_p = true; /* Spaghetti code alert. */
3010 goto handle_on_failure_jump;
3027 abort (); /* We have listed all the cases. */
3030 /* Getting here means we have found the possible starting
3031 characters for one path of the pattern -- and that the empty
3032 string does not match. We need not follow this path further.
3033 Instead, look at the next alternative (remembered on the
3034 stack), or quit if no more. The test at the top of the loop
3035 does these things. */
3036 path_can_be_null = false;
3040 /* Set `can_be_null' for the last path (also the first path, if the
3041 pattern is empty). */
3042 bufp->can_be_null |= path_can_be_null;
3044 } /* re_compile_fastmap */
3046 /* Set REGS to hold NUM_REGS registers, storing them in STARTS and
3047 ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use
3048 this memory for recording register information. STARTS and ENDS
3049 must be allocated using the malloc library routine, and must each
3050 be at least NUM_REGS * sizeof (regoff_t) bytes long.
3052 If NUM_REGS == 0, then subsequent matches should allocate their own
3055 Unless this function is called, the first search or match using
3056 PATTERN_BUFFER will allocate its own register data, without
3057 freeing the old data. */
3060 re_set_registers (bufp, regs, num_regs, starts, ends)
3061 struct re_pattern_buffer *bufp;
3062 struct re_registers *regs;
3064 regoff_t *starts, *ends;
3068 bufp->regs_allocated = REGS_REALLOCATE;
3069 regs->num_regs = num_regs;
3070 regs->start = starts;
3075 bufp->regs_allocated = REGS_UNALLOCATED;
3077 regs->start = regs->end = (regoff_t *) 0;
3081 /* Searching routines. */
3083 /* Like re_search_2, below, but only one string is specified, and
3084 doesn't let you say where to stop matching. */
3087 re_search (bufp, string, size, startpos, range, regs)
3088 struct re_pattern_buffer *bufp;
3090 int size, startpos, range;
3091 struct re_registers *regs;
3093 return re_search_2 (bufp, NULL, 0, string, size, startpos, range,
3098 /* Using the compiled pattern in BUFP->buffer, first tries to match the
3099 virtual concatenation of STRING1 and STRING2, starting first at index
3100 STARTPOS, then at STARTPOS + 1, and so on.
3102 STRING1 and STRING2 have length SIZE1 and SIZE2, respectively.
3104 RANGE is how far to scan while trying to match. RANGE = 0 means try
3105 only at STARTPOS; in general, the last start tried is STARTPOS +
3108 In REGS, return the indices of the virtual concatenation of STRING1
3109 and STRING2 that matched the entire BUFP->buffer and its contained
3112 Do not consider matching one past the index STOP in the virtual
3113 concatenation of STRING1 and STRING2.
3115 We return either the position in the strings at which the match was
3116 found, -1 if no match, or -2 if error (such as failure
3120 re_search_2 (bufp, string1, size1, string2, size2, startpos, range, regs, stop)
3121 struct re_pattern_buffer *bufp;
3122 const char *string1, *string2;
3126 struct re_registers *regs;
3130 register char *fastmap = bufp->fastmap;
3131 register char *translate = bufp->translate;
3132 int total_size = size1 + size2;
3133 int endpos = startpos + range;
3135 /* Check for out-of-range STARTPOS. */
3136 if (startpos < 0 || startpos > total_size)
3139 /* Fix up RANGE if it might eventually take us outside
3140 the virtual concatenation of STRING1 and STRING2. */
3142 range = -1 - startpos;
3143 else if (endpos > total_size)
3144 range = total_size - startpos;
3146 /* If the search isn't to be a backwards one, don't waste time in a
3147 search for a pattern that must be anchored. */
3148 if (bufp->used > 0 && (re_opcode_t) bufp->buffer[0] == begbuf && range > 0)
3156 /* Update the fastmap now if not correct already. */
3157 if (fastmap && !bufp->fastmap_accurate)
3158 if (re_compile_fastmap (bufp) == -2)
3161 /* Loop through the string, looking for a place to start matching. */
3164 /* If a fastmap is supplied, skip quickly over characters that
3165 cannot be the start of a match. If the pattern can match the
3166 null string, however, we don't need to skip characters; we want
3167 the first null string. */
3168 if (fastmap && startpos < total_size && !bufp->can_be_null)
3170 if (range > 0) /* Searching forwards. */
3172 register const char *d;
3173 register int lim = 0;
3176 if (startpos < size1 && startpos + range >= size1)
3177 lim = range - (size1 - startpos);
3179 d = (startpos >= size1 ? string2 - size1 : string1) + startpos;
3181 /* Written out as an if-else to avoid testing `translate'
3185 && !fastmap[(unsigned char)
3186 translate[(unsigned char) *d++]])
3189 while (range > lim && !fastmap[(unsigned char) *d++])
3192 startpos += irange - range;
3194 else /* Searching backwards. */
3196 register char c = (size1 == 0 || startpos >= size1
3197 ? string2[startpos - size1]
3198 : string1[startpos]);
3200 if (!fastmap[(unsigned char) TRANSLATE (c)])
3205 /* If can't match the null string, and that's all we have left, fail. */
3206 if (range >= 0 && startpos == total_size && fastmap
3207 && !bufp->can_be_null)
3210 val = re_match_2_internal (bufp, string1, size1, string2, size2,
3211 startpos, regs, stop);
3212 #ifndef REGEX_MALLOC
3241 /* Declarations and macros for re_match_2. */
3243 static int bcmp_translate ();
3244 static boolean alt_match_null_string_p (),
3245 common_op_match_null_string_p (),
3246 group_match_null_string_p ();
3248 /* This converts PTR, a pointer into one of the search strings `string1'
3249 and `string2' into an offset from the beginning of that string. */
3250 #define POINTER_TO_OFFSET(ptr) \
3251 (FIRST_STRING_P (ptr) \
3252 ? ((regoff_t) ((ptr) - string1)) \
3253 : ((regoff_t) ((ptr) - string2 + size1)))
3255 /* Macros for dealing with the split strings in re_match_2. */
3257 #define MATCHING_IN_FIRST_STRING (dend == end_match_1)
3259 /* Call before fetching a character with *d. This switches over to
3260 string2 if necessary. */
3261 #define PREFETCH() \
3264 /* End of string2 => fail. */ \
3265 if (dend == end_match_2) \
3267 /* End of string1 => advance to string2. */ \
3269 dend = end_match_2; \
3273 /* Test if at very beginning or at very end of the virtual concatenation
3274 of `string1' and `string2'. If only one string, it's `string2'. */
3275 #define AT_STRINGS_BEG(d) ((d) == (size1 ? string1 : string2) || !size2)
3276 #define AT_STRINGS_END(d) ((d) == end2)
3279 /* Test if D points to a character which is word-constituent. We have
3280 two special cases to check for: if past the end of string1, look at
3281 the first character in string2; and if before the beginning of
3282 string2, look at the last character in string1. */
3283 #define WORDCHAR_P(d) \
3284 (SYNTAX ((d) == end1 ? *string2 \
3285 : (d) == string2 - 1 ? *(end1 - 1) : *(d)) \
3288 /* Test if the character before D and the one at D differ with respect
3289 to being word-constituent. */
3290 #define AT_WORD_BOUNDARY(d) \
3291 (AT_STRINGS_BEG (d) || AT_STRINGS_END (d) \
3292 || WORDCHAR_P (d - 1) != WORDCHAR_P (d))
3295 /* Free everything we malloc. */
3296 #ifdef MATCH_MAY_ALLOCATE
3298 #define FREE_VAR(var) if (var) free (var); var = NULL
3299 #define FREE_VARIABLES() \
3301 FREE_VAR (fail_stack.stack); \
3302 FREE_VAR (regstart); \
3303 FREE_VAR (regend); \
3304 FREE_VAR (old_regstart); \
3305 FREE_VAR (old_regend); \
3306 FREE_VAR (best_regstart); \
3307 FREE_VAR (best_regend); \
3308 FREE_VAR (reg_info); \
3309 FREE_VAR (reg_dummy); \
3310 FREE_VAR (reg_info_dummy); \
3312 #else /* not REGEX_MALLOC */
3313 /* This used to do alloca (0), but now we do that in the caller. */
3314 #define FREE_VARIABLES() /* Nothing */
3315 #endif /* not REGEX_MALLOC */
3317 #define FREE_VARIABLES() /* Do nothing! */
3318 #endif /* not MATCH_MAY_ALLOCATE */
3320 /* These values must meet several constraints. They must not be valid
3321 register values; since we have a limit of 255 registers (because
3322 we use only one byte in the pattern for the register number), we can
3323 use numbers larger than 255. They must differ by 1, because of
3324 NUM_FAILURE_ITEMS above. And the value for the lowest register must
3325 be larger than the value for the highest register, so we do not try
3326 to actually save any registers when none are active. */
3327 #define NO_HIGHEST_ACTIVE_REG (1 << BYTEWIDTH)
3328 #define NO_LOWEST_ACTIVE_REG (NO_HIGHEST_ACTIVE_REG + 1)
3330 /* Matching routines. */
3332 #ifndef emacs /* Emacs never uses this. */
3333 /* re_match is like re_match_2 except it takes only a single string. */
3336 re_match (bufp, string, size, pos, regs)
3337 struct re_pattern_buffer *bufp;
3340 struct re_registers *regs;
3342 int result = re_match_2_internal (bufp, NULL, 0, string, size,
3347 #endif /* not emacs */
3350 /* re_match_2 matches the compiled pattern in BUFP against the
3351 the (virtual) concatenation of STRING1 and STRING2 (of length SIZE1
3352 and SIZE2, respectively). We start matching at POS, and stop
3355 If REGS is non-null and the `no_sub' field of BUFP is nonzero, we
3356 store offsets for the substring each group matched in REGS. See the
3357 documentation for exactly how many groups we fill.
3359 We return -1 if no match, -2 if an internal error (such as the
3360 failure stack overflowing). Otherwise, we return the length of the
3361 matched substring. */
3364 re_match_2 (bufp, string1, size1, string2, size2, pos, regs, stop)
3365 struct re_pattern_buffer *bufp;
3366 const char *string1, *string2;
3369 struct re_registers *regs;
3372 int result = re_match_2_internal (bufp, string1, size1, string2, size2,
3378 /* This is a separate function so that we can force an alloca cleanup
3381 re_match_2_internal (bufp, string1, size1, string2, size2, pos, regs, stop)
3382 struct re_pattern_buffer *bufp;
3383 const char *string1, *string2;
3386 struct re_registers *regs;
3389 /* General temporaries. */
3393 /* Just past the end of the corresponding string. */
3394 const char *end1, *end2;
3396 /* Pointers into string1 and string2, just past the last characters in
3397 each to consider matching. */
3398 const char *end_match_1, *end_match_2;
3400 /* Where we are in the data, and the end of the current string. */
3401 const char *d, *dend;
3403 /* Where we are in the pattern, and the end of the pattern. */
3404 unsigned char *p = bufp->buffer;
3405 register unsigned char *pend = p + bufp->used;
3407 /* Mark the opcode just after a start_memory, so we can test for an
3408 empty subpattern when we get to the stop_memory. */
3409 unsigned char *just_past_start_mem = 0;
3411 /* We use this to map every character in the string. */
3412 char *translate = bufp->translate;
3414 /* Failure point stack. Each place that can handle a failure further
3415 down the line pushes a failure point on this stack. It consists of
3416 restart, regend, and reg_info for all registers corresponding to
3417 the subexpressions we're currently inside, plus the number of such
3418 registers, and, finally, two char *'s. The first char * is where
3419 to resume scanning the pattern; the second one is where to resume
3420 scanning the strings. If the latter is zero, the failure point is
3421 a ``dummy''; if a failure happens and the failure point is a dummy,
3422 it gets discarded and the next next one is tried. */
3423 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
3424 fail_stack_type fail_stack;
3427 static unsigned failure_id = 0;
3428 unsigned nfailure_points_pushed = 0, nfailure_points_popped = 0;
3431 /* We fill all the registers internally, independent of what we
3432 return, for use in backreferences. The number here includes
3433 an element for register zero. */
3434 unsigned num_regs = bufp->re_nsub + 1;
3436 /* The currently active registers. */
3437 unsigned lowest_active_reg = NO_LOWEST_ACTIVE_REG;
3438 unsigned highest_active_reg = NO_HIGHEST_ACTIVE_REG;
3440 /* Information on the contents of registers. These are pointers into
3441 the input strings; they record just what was matched (on this
3442 attempt) by a subexpression part of the pattern, that is, the
3443 regnum-th regstart pointer points to where in the pattern we began
3444 matching and the regnum-th regend points to right after where we
3445 stopped matching the regnum-th subexpression. (The zeroth register
3446 keeps track of what the whole pattern matches.) */
3447 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
3448 const char **regstart, **regend;
3451 /* If a group that's operated upon by a repetition operator fails to
3452 match anything, then the register for its start will need to be
3453 restored because it will have been set to wherever in the string we
3454 are when we last see its open-group operator. Similarly for a
3456 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
3457 const char **old_regstart, **old_regend;
3460 /* The is_active field of reg_info helps us keep track of which (possibly
3461 nested) subexpressions we are currently in. The matched_something
3462 field of reg_info[reg_num] helps us tell whether or not we have
3463 matched any of the pattern so far this time through the reg_num-th
3464 subexpression. These two fields get reset each time through any
3465 loop their register is in. */
3466 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
3467 register_info_type *reg_info;
3470 /* The following record the register info as found in the above
3471 variables when we find a match better than any we've seen before.
3472 This happens as we backtrack through the failure points, which in
3473 turn happens only if we have not yet matched the entire string. */
3474 unsigned best_regs_set = false;
3475 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
3476 const char **best_regstart, **best_regend;
3479 /* Logically, this is `best_regend[0]'. But we don't want to have to
3480 allocate space for that if we're not allocating space for anything
3481 else (see below). Also, we never need info about register 0 for
3482 any of the other register vectors, and it seems rather a kludge to
3483 treat `best_regend' differently than the rest. So we keep track of
3484 the end of the best match so far in a separate variable. We
3485 initialize this to NULL so that when we backtrack the first time
3486 and need to test it, it's not garbage. */
3487 const char *match_end = NULL;
3489 /* This helps SET_REGS_MATCHED avoid doing redundant work. */
3490 int set_regs_matched_done = 0;
3492 /* Used when we pop values we don't care about. */
3493 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
3494 const char **reg_dummy;
3495 register_info_type *reg_info_dummy;
3499 /* Counts the total number of registers pushed. */
3500 unsigned num_regs_pushed = 0;
3503 DEBUG_PRINT1 ("\n\nEntering re_match_2.\n");
3507 #ifdef MATCH_MAY_ALLOCATE
3508 /* Do not bother to initialize all the register variables if there are
3509 no groups in the pattern, as it takes a fair amount of time. If
3510 there are groups, we include space for register 0 (the whole
3511 pattern), even though we never use it, since it simplifies the
3512 array indexing. We should fix this. */
3515 regstart = REGEX_TALLOC (num_regs, const char *);
3516 regend = REGEX_TALLOC (num_regs, const char *);
3517 old_regstart = REGEX_TALLOC (num_regs, const char *);
3518 old_regend = REGEX_TALLOC (num_regs, const char *);
3519 best_regstart = REGEX_TALLOC (num_regs, const char *);
3520 best_regend = REGEX_TALLOC (num_regs, const char *);
3521 reg_info = REGEX_TALLOC (num_regs, register_info_type);
3522 reg_dummy = REGEX_TALLOC (num_regs, const char *);
3523 reg_info_dummy = REGEX_TALLOC (num_regs, register_info_type);
3525 if (!(regstart && regend && old_regstart && old_regend && reg_info
3526 && best_regstart && best_regend && reg_dummy && reg_info_dummy))
3532 #if defined (REGEX_MALLOC)
3535 /* We must initialize all our variables to NULL, so that
3536 `FREE_VARIABLES' doesn't try to free them. */
3537 regstart = regend = old_regstart = old_regend = best_regstart
3538 = best_regend = reg_dummy = NULL;
3539 reg_info = reg_info_dummy = (register_info_type *) NULL;
3541 #endif /* REGEX_MALLOC */
3542 #endif /* MATCH_MAY_ALLOCATE */
3544 /* The starting position is bogus. */
3545 if (pos < 0 || pos > size1 + size2)
3551 /* Initialize subexpression text positions to -1 to mark ones that no
3552 start_memory/stop_memory has been seen for. Also initialize the
3553 register information struct. */
3554 for (mcnt = 1; mcnt < num_regs; mcnt++)
3556 regstart[mcnt] = regend[mcnt]
3557 = old_regstart[mcnt] = old_regend[mcnt] = REG_UNSET_VALUE;
3559 REG_MATCH_NULL_STRING_P (reg_info[mcnt]) = MATCH_NULL_UNSET_VALUE;
3560 IS_ACTIVE (reg_info[mcnt]) = 0;
3561 MATCHED_SOMETHING (reg_info[mcnt]) = 0;
3562 EVER_MATCHED_SOMETHING (reg_info[mcnt]) = 0;
3565 /* We move `string1' into `string2' if the latter's empty -- but not if
3566 `string1' is null. */
3567 if (size2 == 0 && string1 != NULL)
3574 end1 = string1 + size1;
3575 end2 = string2 + size2;
3577 /* Compute where to stop matching, within the two strings. */
3580 end_match_1 = string1 + stop;
3581 end_match_2 = string2;
3586 end_match_2 = string2 + stop - size1;
3589 /* `p' scans through the pattern as `d' scans through the data.
3590 `dend' is the end of the input string that `d' points within. `d'
3591 is advanced into the following input string whenever necessary, but
3592 this happens before fetching; therefore, at the beginning of the
3593 loop, `d' can be pointing at the end of a string, but it cannot
3595 if (size1 > 0 && pos <= size1)
3602 d = string2 + pos - size1;
3606 DEBUG_PRINT1 ("The compiled pattern is: ");
3607 DEBUG_PRINT_COMPILED_PATTERN (bufp, p, pend);
3608 DEBUG_PRINT1 ("The string to match is: `");
3609 DEBUG_PRINT_DOUBLE_STRING (d, string1, size1, string2, size2);
3610 DEBUG_PRINT1 ("'\n");
3612 /* This loops over pattern commands. It exits by returning from the
3613 function if the match is complete, or it drops through if the match
3614 fails at this starting point in the input data. */
3617 DEBUG_PRINT2 ("\n0x%x: ", p);
3620 { /* End of pattern means we might have succeeded. */
3621 DEBUG_PRINT1 ("end of pattern ... ");
3623 /* If we haven't matched the entire string, and we want the
3624 longest match, try backtracking. */
3625 if (d != end_match_2)
3627 /* 1 if this match ends in the same string (string1 or string2)
3628 as the best previous match. */
3629 boolean same_str_p = (FIRST_STRING_P (match_end)
3630 == MATCHING_IN_FIRST_STRING);
3631 /* 1 if this match is the best seen so far. */
3632 boolean best_match_p;
3634 /* AIX compiler got confused when this was combined
3635 with the previous declaration. */
3637 best_match_p = d > match_end;
3639 best_match_p = !MATCHING_IN_FIRST_STRING;
3641 DEBUG_PRINT1 ("backtracking.\n");
3643 if (!FAIL_STACK_EMPTY ())
3644 { /* More failure points to try. */
3646 /* If exceeds best match so far, save it. */
3647 if (!best_regs_set || best_match_p)
3649 best_regs_set = true;
3652 DEBUG_PRINT1 ("\nSAVING match as best so far.\n");
3654 for (mcnt = 1; mcnt < num_regs; mcnt++)
3656 best_regstart[mcnt] = regstart[mcnt];
3657 best_regend[mcnt] = regend[mcnt];
3663 /* If no failure points, don't restore garbage. And if
3664 last match is real best match, don't restore second
3666 else if (best_regs_set && !best_match_p)
3669 /* Restore best match. It may happen that `dend ==
3670 end_match_1' while the restored d is in string2.
3671 For example, the pattern `x.*y.*z' against the
3672 strings `x-' and `y-z-', if the two strings are
3673 not consecutive in memory. */
3674 DEBUG_PRINT1 ("Restoring best registers.\n");
3677 dend = ((d >= string1 && d <= end1)
3678 ? end_match_1 : end_match_2);
3680 for (mcnt = 1; mcnt < num_regs; mcnt++)
3682 regstart[mcnt] = best_regstart[mcnt];
3683 regend[mcnt] = best_regend[mcnt];
3686 } /* d != end_match_2 */
3689 DEBUG_PRINT1 ("Accepting match.\n");
3691 /* If caller wants register contents data back, do it. */
3692 if (regs && !bufp->no_sub)
3694 /* Have the register data arrays been allocated? */
3695 if (bufp->regs_allocated == REGS_UNALLOCATED)
3696 { /* No. So allocate them with malloc. We need one
3697 extra element beyond `num_regs' for the `-1' marker
3699 regs->num_regs = MAX (RE_NREGS, num_regs + 1);
3700 regs->start = TALLOC (regs->num_regs, regoff_t);
3701 regs->end = TALLOC (regs->num_regs, regoff_t);
3702 if (regs->start == NULL || regs->end == NULL)
3704 bufp->regs_allocated = REGS_REALLOCATE;
3706 else if (bufp->regs_allocated == REGS_REALLOCATE)
3707 { /* Yes. If we need more elements than were already
3708 allocated, reallocate them. If we need fewer, just
3710 if (regs->num_regs < num_regs + 1)
3712 regs->num_regs = num_regs + 1;
3713 RETALLOC (regs->start, regs->num_regs, regoff_t);
3714 RETALLOC (regs->end, regs->num_regs, regoff_t);
3715 if (regs->start == NULL || regs->end == NULL)
3721 /* These braces fend off a "empty body in an else-statement"
3722 warning under GCC when assert expands to nothing. */
3723 assert (bufp->regs_allocated == REGS_FIXED);
3726 /* Convert the pointer data in `regstart' and `regend' to
3727 indices. Register zero has to be set differently,
3728 since we haven't kept track of any info for it. */
3729 if (regs->num_regs > 0)
3731 regs->start[0] = pos;
3732 regs->end[0] = (MATCHING_IN_FIRST_STRING
3733 ? ((regoff_t) (d - string1))
3734 : ((regoff_t) (d - string2 + size1)));
3737 /* Go through the first `min (num_regs, regs->num_regs)'
3738 registers, since that is all we initialized. */
3739 for (mcnt = 1; mcnt < MIN (num_regs, regs->num_regs); mcnt++)
3741 if (REG_UNSET (regstart[mcnt]) || REG_UNSET (regend[mcnt]))
3742 regs->start[mcnt] = regs->end[mcnt] = -1;
3746 = (regoff_t) POINTER_TO_OFFSET (regstart[mcnt]);
3748 = (regoff_t) POINTER_TO_OFFSET (regend[mcnt]);
3752 /* If the regs structure we return has more elements than
3753 were in the pattern, set the extra elements to -1. If
3754 we (re)allocated the registers, this is the case,
3755 because we always allocate enough to have at least one
3757 for (mcnt = num_regs; mcnt < regs->num_regs; mcnt++)
3758 regs->start[mcnt] = regs->end[mcnt] = -1;
3759 } /* regs && !bufp->no_sub */
3762 DEBUG_PRINT4 ("%u failure points pushed, %u popped (%u remain).\n",
3763 nfailure_points_pushed, nfailure_points_popped,
3764 nfailure_points_pushed - nfailure_points_popped);
3765 DEBUG_PRINT2 ("%u registers pushed.\n", num_regs_pushed);
3767 mcnt = d - pos - (MATCHING_IN_FIRST_STRING
3771 DEBUG_PRINT2 ("Returning %d from re_match_2.\n", mcnt);
3776 /* Otherwise match next pattern command. */
3777 switch (SWITCH_ENUM_CAST ((re_opcode_t) *p++))
3779 /* Ignore these. Used to ignore the n of succeed_n's which
3780 currently have n == 0. */
3782 DEBUG_PRINT1 ("EXECUTING no_op.\n");
3786 DEBUG_PRINT1 ("EXECUTING succeed.\n");
3789 /* Match the next n pattern characters exactly. The following
3790 byte in the pattern defines n, and the n bytes after that
3791 are the characters to match. */
3794 DEBUG_PRINT2 ("EXECUTING exactn %d.\n", mcnt);
3796 /* This is written out as an if-else so we don't waste time
3797 testing `translate' inside the loop. */
3803 if (translate[(unsigned char) *d++] != (char) *p++)
3813 if (*d++ != (char) *p++) goto fail;
3817 SET_REGS_MATCHED ();
3821 /* Match any character except possibly a newline or a null. */
3823 DEBUG_PRINT1 ("EXECUTING anychar.\n");
3827 if ((!(bufp->syntax & RE_DOT_NEWLINE) && TRANSLATE (*d) == '\n')
3828 || (bufp->syntax & RE_DOT_NOT_NULL && TRANSLATE (*d) == '\000'))
3831 SET_REGS_MATCHED ();
3832 DEBUG_PRINT2 (" Matched `%d'.\n", *d);
3840 register unsigned char c;
3841 boolean not = (re_opcode_t) *(p - 1) == charset_not;
3843 DEBUG_PRINT2 ("EXECUTING charset%s.\n", not ? "_not" : "");
3846 c = TRANSLATE (*d); /* The character to match. */
3848 /* Cast to `unsigned' instead of `unsigned char' in case the
3849 bit list is a full 32 bytes long. */
3850 if (c < (unsigned) (*p * BYTEWIDTH)
3851 && p[1 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH)))
3856 if (!not) goto fail;
3858 SET_REGS_MATCHED ();
3864 /* The beginning of a group is represented by start_memory.
3865 The arguments are the register number in the next byte, and the
3866 number of groups inner to this one in the next. The text
3867 matched within the group is recorded (in the internal
3868 registers data structure) under the register number. */
3870 DEBUG_PRINT3 ("EXECUTING start_memory %d (%d):\n", *p, p[1]);
3872 /* Find out if this group can match the empty string. */
3873 p1 = p; /* To send to group_match_null_string_p. */
3875 if (REG_MATCH_NULL_STRING_P (reg_info[*p]) == MATCH_NULL_UNSET_VALUE)
3876 REG_MATCH_NULL_STRING_P (reg_info[*p])
3877 = group_match_null_string_p (&p1, pend, reg_info);
3879 /* Save the position in the string where we were the last time
3880 we were at this open-group operator in case the group is
3881 operated upon by a repetition operator, e.g., with `(a*)*b'
3882 against `ab'; then we want to ignore where we are now in
3883 the string in case this attempt to match fails. */
3884 old_regstart[*p] = REG_MATCH_NULL_STRING_P (reg_info[*p])
3885 ? REG_UNSET (regstart[*p]) ? d : regstart[*p]
3887 DEBUG_PRINT2 (" old_regstart: %d\n",
3888 POINTER_TO_OFFSET (old_regstart[*p]));
3891 DEBUG_PRINT2 (" regstart: %d\n", POINTER_TO_OFFSET (regstart[*p]));
3893 IS_ACTIVE (reg_info[*p]) = 1;
3894 MATCHED_SOMETHING (reg_info[*p]) = 0;
3896 /* Clear this whenever we change the register activity status. */
3897 set_regs_matched_done = 0;
3899 /* This is the new highest active register. */
3900 highest_active_reg = *p;
3902 /* If nothing was active before, this is the new lowest active
3904 if (lowest_active_reg == NO_LOWEST_ACTIVE_REG)
3905 lowest_active_reg = *p;
3907 /* Move past the register number and inner group count. */
3909 just_past_start_mem = p;
3914 /* The stop_memory opcode represents the end of a group. Its
3915 arguments are the same as start_memory's: the register
3916 number, and the number of inner groups. */
3918 DEBUG_PRINT3 ("EXECUTING stop_memory %d (%d):\n", *p, p[1]);
3920 /* We need to save the string position the last time we were at
3921 this close-group operator in case the group is operated
3922 upon by a repetition operator, e.g., with `((a*)*(b*)*)*'
3923 against `aba'; then we want to ignore where we are now in
3924 the string in case this attempt to match fails. */
3925 old_regend[*p] = REG_MATCH_NULL_STRING_P (reg_info[*p])
3926 ? REG_UNSET (regend[*p]) ? d : regend[*p]
3928 DEBUG_PRINT2 (" old_regend: %d\n",
3929 POINTER_TO_OFFSET (old_regend[*p]));
3932 DEBUG_PRINT2 (" regend: %d\n", POINTER_TO_OFFSET (regend[*p]));
3934 /* This register isn't active anymore. */
3935 IS_ACTIVE (reg_info[*p]) = 0;
3937 /* Clear this whenever we change the register activity status. */
3938 set_regs_matched_done = 0;
3940 /* If this was the only register active, nothing is active
3942 if (lowest_active_reg == highest_active_reg)
3944 lowest_active_reg = NO_LOWEST_ACTIVE_REG;
3945 highest_active_reg = NO_HIGHEST_ACTIVE_REG;
3948 { /* We must scan for the new highest active register, since
3949 it isn't necessarily one less than now: consider
3950 (a(b)c(d(e)f)g). When group 3 ends, after the f), the
3951 new highest active register is 1. */
3952 unsigned char r = *p - 1;
3953 while (r > 0 && !IS_ACTIVE (reg_info[r]))
3956 /* If we end up at register zero, that means that we saved
3957 the registers as the result of an `on_failure_jump', not
3958 a `start_memory', and we jumped to past the innermost
3959 `stop_memory'. For example, in ((.)*) we save
3960 registers 1 and 2 as a result of the *, but when we pop
3961 back to the second ), we are at the stop_memory 1.
3962 Thus, nothing is active. */
3965 lowest_active_reg = NO_LOWEST_ACTIVE_REG;
3966 highest_active_reg = NO_HIGHEST_ACTIVE_REG;
3969 highest_active_reg = r;
3972 /* If just failed to match something this time around with a
3973 group that's operated on by a repetition operator, try to
3974 force exit from the ``loop'', and restore the register
3975 information for this group that we had before trying this
3977 if ((!MATCHED_SOMETHING (reg_info[*p])
3978 || just_past_start_mem == p - 1)
3981 boolean is_a_jump_n = false;
3985 switch ((re_opcode_t) *p1++)
3989 case pop_failure_jump:
3990 case maybe_pop_jump:
3992 case dummy_failure_jump:
3993 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4003 /* If the next operation is a jump backwards in the pattern
4004 to an on_failure_jump right before the start_memory
4005 corresponding to this stop_memory, exit from the loop
4006 by forcing a failure after pushing on the stack the
4007 on_failure_jump's jump in the pattern, and d. */
4008 if (mcnt < 0 && (re_opcode_t) *p1 == on_failure_jump
4009 && (re_opcode_t) p1[3] == start_memory && p1[4] == *p)
4011 /* If this group ever matched anything, then restore
4012 what its registers were before trying this last
4013 failed match, e.g., with `(a*)*b' against `ab' for
4014 regstart[1], and, e.g., with `((a*)*(b*)*)*'
4015 against `aba' for regend[3].
4017 Also restore the registers for inner groups for,
4018 e.g., `((a*)(b*))*' against `aba' (register 3 would
4019 otherwise get trashed). */
4021 if (EVER_MATCHED_SOMETHING (reg_info[*p]))
4025 EVER_MATCHED_SOMETHING (reg_info[*p]) = 0;
4027 /* Restore this and inner groups' (if any) registers. */
4028 for (r = *p; r < *p + *(p + 1); r++)
4030 regstart[r] = old_regstart[r];
4032 /* xx why this test? */
4033 if ((int) old_regend[r] >= (int) regstart[r])
4034 regend[r] = old_regend[r];
4038 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4039 PUSH_FAILURE_POINT (p1 + mcnt, d, -2);
4045 /* Move past the register number and the inner group count. */
4050 /* \<digit> has been turned into a `duplicate' command which is
4051 followed by the numeric value of <digit> as the register number. */
4054 register const char *d2, *dend2;
4055 int regno = *p++; /* Get which register to match against. */
4056 DEBUG_PRINT2 ("EXECUTING duplicate %d.\n", regno);
4058 /* Can't back reference a group which we've never matched. */
4059 if (REG_UNSET (regstart[regno]) || REG_UNSET (regend[regno]))
4062 /* Where in input to try to start matching. */
4063 d2 = regstart[regno];
4065 /* Where to stop matching; if both the place to start and
4066 the place to stop matching are in the same string, then
4067 set to the place to stop, otherwise, for now have to use
4068 the end of the first string. */
4070 dend2 = ((FIRST_STRING_P (regstart[regno])
4071 == FIRST_STRING_P (regend[regno]))
4072 ? regend[regno] : end_match_1);
4075 /* If necessary, advance to next segment in register
4079 if (dend2 == end_match_2) break;
4080 if (dend2 == regend[regno]) break;
4082 /* End of string1 => advance to string2. */
4084 dend2 = regend[regno];
4086 /* At end of register contents => success */
4087 if (d2 == dend2) break;
4089 /* If necessary, advance to next segment in data. */
4092 /* How many characters left in this segment to match. */
4095 /* Want how many consecutive characters we can match in
4096 one shot, so, if necessary, adjust the count. */
4097 if (mcnt > dend2 - d2)
4100 /* Compare that many; failure if mismatch, else move
4103 ? bcmp_translate (d, d2, mcnt, translate)
4104 : bcmp (d, d2, mcnt))
4106 d += mcnt, d2 += mcnt;
4108 /* Do this because we've match some characters. */
4109 SET_REGS_MATCHED ();
4115 /* begline matches the empty string at the beginning of the string
4116 (unless `not_bol' is set in `bufp'), and, if
4117 `newline_anchor' is set, after newlines. */
4119 DEBUG_PRINT1 ("EXECUTING begline.\n");
4121 if (AT_STRINGS_BEG (d))
4123 if (!bufp->not_bol) break;
4125 else if (d[-1] == '\n' && bufp->newline_anchor)
4129 /* In all other cases, we fail. */
4133 /* endline is the dual of begline. */
4135 DEBUG_PRINT1 ("EXECUTING endline.\n");
4137 if (AT_STRINGS_END (d))
4139 if (!bufp->not_eol) break;
4142 /* We have to ``prefetch'' the next character. */
4143 else if ((d == end1 ? *string2 : *d) == '\n'
4144 && bufp->newline_anchor)
4151 /* Match at the very beginning of the data. */
4153 DEBUG_PRINT1 ("EXECUTING begbuf.\n");
4154 if (AT_STRINGS_BEG (d))
4159 /* Match at the very end of the data. */
4161 DEBUG_PRINT1 ("EXECUTING endbuf.\n");
4162 if (AT_STRINGS_END (d))
4167 /* on_failure_keep_string_jump is used to optimize `.*\n'. It
4168 pushes NULL as the value for the string on the stack. Then
4169 `pop_failure_point' will keep the current value for the
4170 string, instead of restoring it. To see why, consider
4171 matching `foo\nbar' against `.*\n'. The .* matches the foo;
4172 then the . fails against the \n. But the next thing we want
4173 to do is match the \n against the \n; if we restored the
4174 string value, we would be back at the foo.
4176 Because this is used only in specific cases, we don't need to
4177 check all the things that `on_failure_jump' does, to make
4178 sure the right things get saved on the stack. Hence we don't
4179 share its code. The only reason to push anything on the
4180 stack at all is that otherwise we would have to change
4181 `anychar's code to do something besides goto fail in this
4182 case; that seems worse than this. */
4183 case on_failure_keep_string_jump:
4184 DEBUG_PRINT1 ("EXECUTING on_failure_keep_string_jump");
4186 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4187 DEBUG_PRINT3 (" %d (to 0x%x):\n", mcnt, p + mcnt);
4189 PUSH_FAILURE_POINT (p + mcnt, NULL, -2);
4193 /* Uses of on_failure_jump:
4195 Each alternative starts with an on_failure_jump that points
4196 to the beginning of the next alternative. Each alternative
4197 except the last ends with a jump that in effect jumps past
4198 the rest of the alternatives. (They really jump to the
4199 ending jump of the following alternative, because tensioning
4200 these jumps is a hassle.)
4202 Repeats start with an on_failure_jump that points past both
4203 the repetition text and either the following jump or
4204 pop_failure_jump back to this on_failure_jump. */
4205 case on_failure_jump:
4207 DEBUG_PRINT1 ("EXECUTING on_failure_jump");
4209 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4210 DEBUG_PRINT3 (" %d (to 0x%x)", mcnt, p + mcnt);
4212 /* If this on_failure_jump comes right before a group (i.e.,
4213 the original * applied to a group), save the information
4214 for that group and all inner ones, so that if we fail back
4215 to this point, the group's information will be correct.
4216 For example, in \(a*\)*\1, we need the preceding group,
4217 and in \(\(a*\)b*\)\2, we need the inner group. */
4219 /* We can't use `p' to check ahead because we push
4220 a failure point to `p + mcnt' after we do this. */
4223 /* We need to skip no_op's before we look for the
4224 start_memory in case this on_failure_jump is happening as
4225 the result of a completed succeed_n, as in \(a\)\{1,3\}b\1
4227 while (p1 < pend && (re_opcode_t) *p1 == no_op)
4230 if (p1 < pend && (re_opcode_t) *p1 == start_memory)
4232 /* We have a new highest active register now. This will
4233 get reset at the start_memory we are about to get to,
4234 but we will have saved all the registers relevant to
4235 this repetition op, as described above. */
4236 highest_active_reg = *(p1 + 1) + *(p1 + 2);
4237 if (lowest_active_reg == NO_LOWEST_ACTIVE_REG)
4238 lowest_active_reg = *(p1 + 1);
4241 DEBUG_PRINT1 (":\n");
4242 PUSH_FAILURE_POINT (p + mcnt, d, -2);
4246 /* A smart repeat ends with `maybe_pop_jump'.
4247 We change it to either `pop_failure_jump' or `jump'. */
4248 case maybe_pop_jump:
4249 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4250 DEBUG_PRINT2 ("EXECUTING maybe_pop_jump %d.\n", mcnt);
4252 register unsigned char *p2 = p;
4254 /* Compare the beginning of the repeat with what in the
4255 pattern follows its end. If we can establish that there
4256 is nothing that they would both match, i.e., that we
4257 would have to backtrack because of (as in, e.g., `a*a')
4258 then we can change to pop_failure_jump, because we'll
4259 never have to backtrack.
4261 This is not true in the case of alternatives: in
4262 `(a|ab)*' we do need to backtrack to the `ab' alternative
4263 (e.g., if the string was `ab'). But instead of trying to
4264 detect that here, the alternative has put on a dummy
4265 failure point which is what we will end up popping. */
4267 /* Skip over open/close-group commands.
4268 If what follows this loop is a ...+ construct,
4269 look at what begins its body, since we will have to
4270 match at least one of that. */
4274 && ((re_opcode_t) *p2 == stop_memory
4275 || (re_opcode_t) *p2 == start_memory))
4277 else if (p2 + 6 < pend
4278 && (re_opcode_t) *p2 == dummy_failure_jump)
4285 /* p1[0] ... p1[2] are the `on_failure_jump' corresponding
4286 to the `maybe_finalize_jump' of this case. Examine what
4289 /* If we're at the end of the pattern, we can change. */
4292 /* Consider what happens when matching ":\(.*\)"
4293 against ":/". I don't really understand this code
4295 p[-3] = (unsigned char) pop_failure_jump;
4297 (" End of pattern: change to `pop_failure_jump'.\n");
4300 else if ((re_opcode_t) *p2 == exactn
4301 || (bufp->newline_anchor && (re_opcode_t) *p2 == endline))
4303 register unsigned char c
4304 = *p2 == (unsigned char) endline ? '\n' : p2[2];
4306 if ((re_opcode_t) p1[3] == exactn && p1[5] != c)
4308 p[-3] = (unsigned char) pop_failure_jump;
4309 DEBUG_PRINT3 (" %c != %c => pop_failure_jump.\n",
4313 else if ((re_opcode_t) p1[3] == charset
4314 || (re_opcode_t) p1[3] == charset_not)
4316 int not = (re_opcode_t) p1[3] == charset_not;
4318 if (c < (unsigned char) (p1[4] * BYTEWIDTH)
4319 && p1[5 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH)))
4322 /* `not' is equal to 1 if c would match, which means
4323 that we can't change to pop_failure_jump. */
4326 p[-3] = (unsigned char) pop_failure_jump;
4327 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
4331 else if ((re_opcode_t) *p2 == charset)
4334 register unsigned char c
4335 = *p2 == (unsigned char) endline ? '\n' : p2[2];
4338 if ((re_opcode_t) p1[3] == exactn
4339 && ! ((int) p2[1] * BYTEWIDTH > (int) p1[4]
4340 && (p2[1 + p1[4] / BYTEWIDTH]
4341 & (1 << (p1[4] % BYTEWIDTH)))))
4343 p[-3] = (unsigned char) pop_failure_jump;
4344 DEBUG_PRINT3 (" %c != %c => pop_failure_jump.\n",
4348 else if ((re_opcode_t) p1[3] == charset_not)
4351 /* We win if the charset_not inside the loop
4352 lists every character listed in the charset after. */
4353 for (idx = 0; idx < (int) p2[1]; idx++)
4354 if (! (p2[2 + idx] == 0
4355 || (idx < (int) p1[4]
4356 && ((p2[2 + idx] & ~ p1[5 + idx]) == 0))))
4361 p[-3] = (unsigned char) pop_failure_jump;
4362 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
4365 else if ((re_opcode_t) p1[3] == charset)
4368 /* We win if the charset inside the loop
4369 has no overlap with the one after the loop. */
4371 idx < (int) p2[1] && idx < (int) p1[4];
4373 if ((p2[2 + idx] & p1[5 + idx]) != 0)
4376 if (idx == p2[1] || idx == p1[4])
4378 p[-3] = (unsigned char) pop_failure_jump;
4379 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
4384 p -= 2; /* Point at relative address again. */
4385 if ((re_opcode_t) p[-1] != pop_failure_jump)
4387 p[-1] = (unsigned char) jump;
4388 DEBUG_PRINT1 (" Match => jump.\n");
4389 goto unconditional_jump;
4391 /* Note fall through. */
4394 /* The end of a simple repeat has a pop_failure_jump back to
4395 its matching on_failure_jump, where the latter will push a
4396 failure point. The pop_failure_jump takes off failure
4397 points put on by this pop_failure_jump's matching
4398 on_failure_jump; we got through the pattern to here from the
4399 matching on_failure_jump, so didn't fail. */
4400 case pop_failure_jump:
4402 /* We need to pass separate storage for the lowest and
4403 highest registers, even though we don't care about the
4404 actual values. Otherwise, we will restore only one
4405 register from the stack, since lowest will == highest in
4406 `pop_failure_point'. */
4407 unsigned dummy_low_reg, dummy_high_reg;
4408 unsigned char *pdummy;
4411 DEBUG_PRINT1 ("EXECUTING pop_failure_jump.\n");
4412 POP_FAILURE_POINT (sdummy, pdummy,
4413 dummy_low_reg, dummy_high_reg,
4414 reg_dummy, reg_dummy, reg_info_dummy);
4416 /* Note fall through. */
4419 /* Unconditionally jump (without popping any failure points). */
4422 EXTRACT_NUMBER_AND_INCR (mcnt, p); /* Get the amount to jump. */
4423 DEBUG_PRINT2 ("EXECUTING jump %d ", mcnt);
4424 p += mcnt; /* Do the jump. */
4425 DEBUG_PRINT2 ("(to 0x%x).\n", p);
4429 /* We need this opcode so we can detect where alternatives end
4430 in `group_match_null_string_p' et al. */
4432 DEBUG_PRINT1 ("EXECUTING jump_past_alt.\n");
4433 goto unconditional_jump;
4436 /* Normally, the on_failure_jump pushes a failure point, which
4437 then gets popped at pop_failure_jump. We will end up at
4438 pop_failure_jump, also, and with a pattern of, say, `a+', we
4439 are skipping over the on_failure_jump, so we have to push
4440 something meaningless for pop_failure_jump to pop. */
4441 case dummy_failure_jump:
4442 DEBUG_PRINT1 ("EXECUTING dummy_failure_jump.\n");
4443 /* It doesn't matter what we push for the string here. What
4444 the code at `fail' tests is the value for the pattern. */
4445 PUSH_FAILURE_POINT (0, 0, -2);
4446 goto unconditional_jump;
4449 /* At the end of an alternative, we need to push a dummy failure
4450 point in case we are followed by a `pop_failure_jump', because
4451 we don't want the failure point for the alternative to be
4452 popped. For example, matching `(a|ab)*' against `aab'
4453 requires that we match the `ab' alternative. */
4454 case push_dummy_failure:
4455 DEBUG_PRINT1 ("EXECUTING push_dummy_failure.\n");
4456 /* See comments just above at `dummy_failure_jump' about the
4458 PUSH_FAILURE_POINT (0, 0, -2);
4461 /* Have to succeed matching what follows at least n times.
4462 After that, handle like `on_failure_jump'. */
4464 EXTRACT_NUMBER (mcnt, p + 2);
4465 DEBUG_PRINT2 ("EXECUTING succeed_n %d.\n", mcnt);
4468 /* Originally, this is how many times we HAVE to succeed. */
4473 STORE_NUMBER_AND_INCR (p, mcnt);
4474 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p, mcnt);
4478 DEBUG_PRINT2 (" Setting two bytes from 0x%x to no_op.\n", p+2);
4479 p[2] = (unsigned char) no_op;
4480 p[3] = (unsigned char) no_op;
4486 EXTRACT_NUMBER (mcnt, p + 2);
4487 DEBUG_PRINT2 ("EXECUTING jump_n %d.\n", mcnt);
4489 /* Originally, this is how many times we CAN jump. */
4493 STORE_NUMBER (p + 2, mcnt);
4494 goto unconditional_jump;
4496 /* If don't have to jump any more, skip over the rest of command. */
4503 DEBUG_PRINT1 ("EXECUTING set_number_at.\n");
4505 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4507 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4508 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p1, mcnt);
4509 STORE_NUMBER (p1, mcnt);
4514 DEBUG_PRINT1 ("EXECUTING wordbound.\n");
4515 if (AT_WORD_BOUNDARY (d))
4520 DEBUG_PRINT1 ("EXECUTING notwordbound.\n");
4521 if (AT_WORD_BOUNDARY (d))
4526 DEBUG_PRINT1 ("EXECUTING wordbeg.\n");
4527 if (WORDCHAR_P (d) && (AT_STRINGS_BEG (d) || !WORDCHAR_P (d - 1)))
4532 DEBUG_PRINT1 ("EXECUTING wordend.\n");
4533 if (!AT_STRINGS_BEG (d) && WORDCHAR_P (d - 1)
4534 && (!WORDCHAR_P (d) || AT_STRINGS_END (d)))
4540 DEBUG_PRINT1 ("EXECUTING before_dot.\n");
4541 if (PTR_CHAR_POS ((unsigned char *) d) >= point)
4546 DEBUG_PRINT1 ("EXECUTING at_dot.\n");
4547 if (PTR_CHAR_POS ((unsigned char *) d) != point)
4552 DEBUG_PRINT1 ("EXECUTING after_dot.\n");
4553 if (PTR_CHAR_POS ((unsigned char *) d) <= point)
4556 #if 0 /* not emacs19 */
4558 DEBUG_PRINT1 ("EXECUTING at_dot.\n");
4559 if (PTR_CHAR_POS ((unsigned char *) d) + 1 != point)
4562 #endif /* not emacs19 */
4565 DEBUG_PRINT2 ("EXECUTING syntaxspec %d.\n", mcnt);
4570 DEBUG_PRINT1 ("EXECUTING Emacs wordchar.\n");
4574 /* Can't use *d++ here; SYNTAX may be an unsafe macro. */
4576 if (SYNTAX (d[-1]) != (enum syntaxcode) mcnt)
4578 SET_REGS_MATCHED ();
4582 DEBUG_PRINT2 ("EXECUTING notsyntaxspec %d.\n", mcnt);
4584 goto matchnotsyntax;
4587 DEBUG_PRINT1 ("EXECUTING Emacs notwordchar.\n");
4591 /* Can't use *d++ here; SYNTAX may be an unsafe macro. */
4593 if (SYNTAX (d[-1]) == (enum syntaxcode) mcnt)
4595 SET_REGS_MATCHED ();
4598 #else /* not emacs */
4600 DEBUG_PRINT1 ("EXECUTING non-Emacs wordchar.\n");
4602 if (!WORDCHAR_P (d))
4604 SET_REGS_MATCHED ();
4609 DEBUG_PRINT1 ("EXECUTING non-Emacs notwordchar.\n");
4613 SET_REGS_MATCHED ();
4616 #endif /* not emacs */
4621 continue; /* Successfully executed one pattern command; keep going. */
4624 /* We goto here if a matching operation fails. */
4626 if (!FAIL_STACK_EMPTY ())
4627 { /* A restart point is known. Restore to that state. */
4628 DEBUG_PRINT1 ("\nFAIL:\n");
4629 POP_FAILURE_POINT (d, p,
4630 lowest_active_reg, highest_active_reg,
4631 regstart, regend, reg_info);
4633 /* If this failure point is a dummy, try the next one. */
4637 /* If we failed to the end of the pattern, don't examine *p. */
4641 boolean is_a_jump_n = false;
4643 /* If failed to a backwards jump that's part of a repetition
4644 loop, need to pop this failure point and use the next one. */
4645 switch ((re_opcode_t) *p)
4649 case maybe_pop_jump:
4650 case pop_failure_jump:
4653 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4656 if ((is_a_jump_n && (re_opcode_t) *p1 == succeed_n)
4658 && (re_opcode_t) *p1 == on_failure_jump))
4666 if (d >= string1 && d <= end1)
4670 break; /* Matching at this starting point really fails. */
4674 goto restore_best_regs;
4678 return -1; /* Failure to match. */
4681 /* Subroutine definitions for re_match_2. */
4684 /* We are passed P pointing to a register number after a start_memory.
4686 Return true if the pattern up to the corresponding stop_memory can
4687 match the empty string, and false otherwise.
4689 If we find the matching stop_memory, sets P to point to one past its number.
4690 Otherwise, sets P to an undefined byte less than or equal to END.
4692 We don't handle duplicates properly (yet). */
4695 group_match_null_string_p (p, end, reg_info)
4696 unsigned char **p, *end;
4697 register_info_type *reg_info;
4700 /* Point to after the args to the start_memory. */
4701 unsigned char *p1 = *p + 2;
4705 /* Skip over opcodes that can match nothing, and return true or
4706 false, as appropriate, when we get to one that can't, or to the
4707 matching stop_memory. */
4709 switch ((re_opcode_t) *p1)
4711 /* Could be either a loop or a series of alternatives. */
4712 case on_failure_jump:
4714 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4716 /* If the next operation is not a jump backwards in the
4721 /* Go through the on_failure_jumps of the alternatives,
4722 seeing if any of the alternatives cannot match nothing.
4723 The last alternative starts with only a jump,
4724 whereas the rest start with on_failure_jump and end
4725 with a jump, e.g., here is the pattern for `a|b|c':
4727 /on_failure_jump/0/6/exactn/1/a/jump_past_alt/0/6
4728 /on_failure_jump/0/6/exactn/1/b/jump_past_alt/0/3
4731 So, we have to first go through the first (n-1)
4732 alternatives and then deal with the last one separately. */
4735 /* Deal with the first (n-1) alternatives, which start
4736 with an on_failure_jump (see above) that jumps to right
4737 past a jump_past_alt. */
4739 while ((re_opcode_t) p1[mcnt-3] == jump_past_alt)
4741 /* `mcnt' holds how many bytes long the alternative
4742 is, including the ending `jump_past_alt' and
4745 if (!alt_match_null_string_p (p1, p1 + mcnt - 3,
4749 /* Move to right after this alternative, including the
4753 /* Break if it's the beginning of an n-th alternative
4754 that doesn't begin with an on_failure_jump. */
4755 if ((re_opcode_t) *p1 != on_failure_jump)
4758 /* Still have to check that it's not an n-th
4759 alternative that starts with an on_failure_jump. */
4761 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4762 if ((re_opcode_t) p1[mcnt-3] != jump_past_alt)
4764 /* Get to the beginning of the n-th alternative. */
4770 /* Deal with the last alternative: go back and get number
4771 of the `jump_past_alt' just before it. `mcnt' contains
4772 the length of the alternative. */
4773 EXTRACT_NUMBER (mcnt, p1 - 2);
4775 if (!alt_match_null_string_p (p1, p1 + mcnt, reg_info))
4778 p1 += mcnt; /* Get past the n-th alternative. */
4784 assert (p1[1] == **p);
4790 if (!common_op_match_null_string_p (&p1, end, reg_info))
4793 } /* while p1 < end */
4796 } /* group_match_null_string_p */
4799 /* Similar to group_match_null_string_p, but doesn't deal with alternatives:
4800 It expects P to be the first byte of a single alternative and END one
4801 byte past the last. The alternative can contain groups. */
4804 alt_match_null_string_p (p, end, reg_info)
4805 unsigned char *p, *end;
4806 register_info_type *reg_info;
4809 unsigned char *p1 = p;
4813 /* Skip over opcodes that can match nothing, and break when we get
4814 to one that can't. */
4816 switch ((re_opcode_t) *p1)
4819 case on_failure_jump:
4821 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4826 if (!common_op_match_null_string_p (&p1, end, reg_info))
4829 } /* while p1 < end */
4832 } /* alt_match_null_string_p */
4835 /* Deals with the ops common to group_match_null_string_p and
4836 alt_match_null_string_p.
4838 Sets P to one after the op and its arguments, if any. */
4841 common_op_match_null_string_p (p, end, reg_info)
4842 unsigned char **p, *end;
4843 register_info_type *reg_info;
4848 unsigned char *p1 = *p;
4850 switch ((re_opcode_t) *p1++)
4870 assert (reg_no > 0 && reg_no <= MAX_REGNUM);
4871 ret = group_match_null_string_p (&p1, end, reg_info);
4873 /* Have to set this here in case we're checking a group which
4874 contains a group and a back reference to it. */
4876 if (REG_MATCH_NULL_STRING_P (reg_info[reg_no]) == MATCH_NULL_UNSET_VALUE)
4877 REG_MATCH_NULL_STRING_P (reg_info[reg_no]) = ret;
4883 /* If this is an optimized succeed_n for zero times, make the jump. */
4885 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4893 /* Get to the number of times to succeed. */
4895 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4900 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4908 if (!REG_MATCH_NULL_STRING_P (reg_info[*p1]))
4916 /* All other opcodes mean we cannot match the empty string. */
4922 } /* common_op_match_null_string_p */
4925 /* Return zero if TRANSLATE[S1] and TRANSLATE[S2] are identical for LEN
4926 bytes; nonzero otherwise. */
4929 bcmp_translate (s1, s2, len, translate)
4930 unsigned char *s1, *s2;
4934 register unsigned char *p1 = s1, *p2 = s2;
4937 if (translate[*p1++] != translate[*p2++]) return 1;
4943 /* Entry points for GNU code. */
4945 /* re_compile_pattern is the GNU regular expression compiler: it
4946 compiles PATTERN (of length SIZE) and puts the result in BUFP.
4947 Returns 0 if the pattern was valid, otherwise an error string.
4949 Assumes the `allocated' (and perhaps `buffer') and `translate' fields
4950 are set in BUFP on entry.
4952 We call regex_compile to do the actual compilation. */
4955 re_compile_pattern (pattern, length, bufp)
4956 const char *pattern;
4958 struct re_pattern_buffer *bufp;
4962 /* GNU code is written to assume at least RE_NREGS registers will be set
4963 (and at least one extra will be -1). */
4964 bufp->regs_allocated = REGS_UNALLOCATED;
4966 /* And GNU code determines whether or not to get register information
4967 by passing null for the REGS argument to re_match, etc., not by
4971 /* Match anchors at newline. */
4972 bufp->newline_anchor = 1;
4974 ret = regex_compile (pattern, length, re_syntax_options, bufp);
4978 return gettext (re_error_msgid[(int) ret]);
4981 /* Entry points compatible with 4.2 BSD regex library. We don't define
4982 them unless specifically requested. */
4984 #ifdef _REGEX_RE_COMP
4986 /* BSD has one and only one pattern buffer. */
4987 static struct re_pattern_buffer re_comp_buf;
4997 if (!re_comp_buf.buffer)
4998 return gettext ("No previous regular expression");
5002 if (!re_comp_buf.buffer)
5004 re_comp_buf.buffer = (unsigned char *) malloc (200);
5005 if (re_comp_buf.buffer == NULL)
5006 return gettext (re_error_msgid[(int) REG_ESPACE]);
5007 re_comp_buf.allocated = 200;
5009 re_comp_buf.fastmap = (char *) malloc (1 << BYTEWIDTH);
5010 if (re_comp_buf.fastmap == NULL)
5011 return gettext (re_error_msgid[(int) REG_ESPACE]);
5014 /* Since `re_exec' always passes NULL for the `regs' argument, we
5015 don't need to initialize the pattern buffer fields which affect it. */
5017 /* Match anchors at newlines. */
5018 re_comp_buf.newline_anchor = 1;
5020 ret = regex_compile (s, strlen (s), re_syntax_options, &re_comp_buf);
5025 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
5026 return (char *) gettext (re_error_msgid[(int) ret]);
5034 const int len = strlen (s);
5036 0 <= re_search (&re_comp_buf, s, len, 0, len, (struct re_registers *) 0);
5038 #endif /* _REGEX_RE_COMP */
5040 /* POSIX.2 functions. Don't define these for Emacs. */
5044 /* regcomp takes a regular expression as a string and compiles it.
5046 PREG is a regex_t *. We do not expect any fields to be initialized,
5047 since POSIX says we shouldn't. Thus, we set
5049 `buffer' to the compiled pattern;
5050 `used' to the length of the compiled pattern;
5051 `syntax' to RE_SYNTAX_POSIX_EXTENDED if the
5052 REG_EXTENDED bit in CFLAGS is set; otherwise, to
5053 RE_SYNTAX_POSIX_BASIC;
5054 `newline_anchor' to REG_NEWLINE being set in CFLAGS;
5055 `fastmap' and `fastmap_accurate' to zero;
5056 `re_nsub' to the number of subexpressions in PATTERN.
5058 PATTERN is the address of the pattern string.
5060 CFLAGS is a series of bits which affect compilation.
5062 If REG_EXTENDED is set, we use POSIX extended syntax; otherwise, we
5063 use POSIX basic syntax.
5065 If REG_NEWLINE is set, then . and [^...] don't match newline.
5066 Also, regexec will try a match beginning after every newline.
5068 If REG_ICASE is set, then we considers upper- and lowercase
5069 versions of letters to be equivalent when matching.
5071 If REG_NOSUB is set, then when PREG is passed to regexec, that
5072 routine will report only success or failure, and nothing about the
5075 It returns 0 if it succeeds, nonzero if it doesn't. (See regex.h for
5076 the return codes and their meanings.) */
5079 regcomp (preg, pattern, cflags)
5081 const char *pattern;
5086 = (cflags & REG_EXTENDED) ?
5087 RE_SYNTAX_POSIX_EXTENDED : RE_SYNTAX_POSIX_BASIC;
5089 /* regex_compile will allocate the space for the compiled pattern. */
5091 preg->allocated = 0;
5094 /* Don't bother to use a fastmap when searching. This simplifies the
5095 REG_NEWLINE case: if we used a fastmap, we'd have to put all the
5096 characters after newlines into the fastmap. This way, we just try
5100 if (cflags & REG_ICASE)
5104 preg->translate = (char *) malloc (CHAR_SET_SIZE);
5105 if (preg->translate == NULL)
5106 return (int) REG_ESPACE;
5108 /* Map uppercase characters to corresponding lowercase ones. */
5109 for (i = 0; i < CHAR_SET_SIZE; i++)
5110 preg->translate[i] = ISUPPER (i) ? tolower (i) : i;
5113 preg->translate = NULL;
5115 /* If REG_NEWLINE is set, newlines are treated differently. */
5116 if (cflags & REG_NEWLINE)
5117 { /* REG_NEWLINE implies neither . nor [^...] match newline. */
5118 syntax &= ~RE_DOT_NEWLINE;
5119 syntax |= RE_HAT_LISTS_NOT_NEWLINE;
5120 /* It also changes the matching behavior. */
5121 preg->newline_anchor = 1;
5124 preg->newline_anchor = 0;
5126 preg->no_sub = !!(cflags & REG_NOSUB);
5128 /* POSIX says a null character in the pattern terminates it, so we
5129 can use strlen here in compiling the pattern. */
5130 ret = regex_compile (pattern, strlen (pattern), syntax, preg);
5132 /* POSIX doesn't distinguish between an unmatched open-group and an
5133 unmatched close-group: both are REG_EPAREN. */
5134 if (ret == REG_ERPAREN) ret = REG_EPAREN;
5140 /* regexec searches for a given pattern, specified by PREG, in the
5143 If NMATCH is zero or REG_NOSUB was set in the cflags argument to
5144 `regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at
5145 least NMATCH elements, and we set them to the offsets of the
5146 corresponding matched substrings.
5148 EFLAGS specifies `execution flags' which affect matching: if
5149 REG_NOTBOL is set, then ^ does not match at the beginning of the
5150 string; if REG_NOTEOL is set, then $ does not match at the end.
5152 We return 0 if we find a match and REG_NOMATCH if not. */
5155 regexec (preg, string, nmatch, pmatch, eflags)
5156 const regex_t *preg;
5159 regmatch_t pmatch[];
5163 struct re_registers regs;
5164 regex_t private_preg;
5165 int len = strlen (string);
5166 boolean want_reg_info = !preg->no_sub && nmatch > 0;
5168 private_preg = *preg;
5170 private_preg.not_bol = !!(eflags & REG_NOTBOL);
5171 private_preg.not_eol = !!(eflags & REG_NOTEOL);
5173 /* The user has told us exactly how many registers to return
5174 information about, via `nmatch'. We have to pass that on to the
5175 matching routines. */
5176 private_preg.regs_allocated = REGS_FIXED;
5180 regs.num_regs = nmatch;
5181 regs.start = TALLOC (nmatch, regoff_t);
5182 regs.end = TALLOC (nmatch, regoff_t);
5183 if (regs.start == NULL || regs.end == NULL)
5184 return (int) REG_NOMATCH;
5187 /* Perform the searching operation. */
5188 ret = re_search (&private_preg, string, len,
5189 /* start: */ 0, /* range: */ len,
5190 want_reg_info ? ®s : (struct re_registers *) 0);
5192 /* Copy the register information to the POSIX structure. */
5199 for (r = 0; r < nmatch; r++)
5201 pmatch[r].rm_so = regs.start[r];
5202 pmatch[r].rm_eo = regs.end[r];
5206 /* If we needed the temporary register info, free the space now. */
5211 /* We want zero return to mean success, unlike `re_search'. */
5212 return ret >= 0 ? (int) REG_NOERROR : (int) REG_NOMATCH;
5216 /* Returns a message corresponding to an error code, ERRCODE, returned
5217 from either regcomp or regexec. We don't use PREG here. */
5220 regerror (errcode, preg, errbuf, errbuf_size)
5222 const regex_t *preg;
5230 || errcode >= (sizeof (re_error_msgid) / sizeof (re_error_msgid[0])))
5231 /* Only error codes returned by the rest of the code should be passed
5232 to this routine. If we are given anything else, or if other regex
5233 code generates an invalid error code, then the program has a bug.
5234 Dump core so we can fix it. */
5237 msg = gettext (re_error_msgid[errcode]);
5239 msg_size = strlen (msg) + 1; /* Includes the null. */
5241 if (errbuf_size != 0)
5243 if (msg_size > errbuf_size)
5245 strncpy (errbuf, msg, errbuf_size - 1);
5246 errbuf[errbuf_size - 1] = 0;
5249 strcpy (errbuf, msg);
5256 /* Free dynamically allocated space used by PREG. */
5262 if (preg->buffer != NULL)
5263 free (preg->buffer);
5264 preg->buffer = NULL;
5266 preg->allocated = 0;
5269 if (preg->fastmap != NULL)
5270 free (preg->fastmap);
5271 preg->fastmap = NULL;
5272 preg->fastmap_accurate = 0;
5274 if (preg->translate != NULL)
5275 free (preg->translate);
5276 preg->translate = NULL;
5279 #endif /* not emacs */
5283 make-backup-files: t
5285 trim-versions-without-asking: nil