1 /* Extended regular expression matching and search library,
3 (Implements POSIX draft P10003.2/D11.2, except for
4 internationalization features.)
6 Copyright (C) 1993, 1994 Free Software Foundation, Inc.
8 This program is free software; you can redistribute it and/or modify
9 it under the terms of the GNU General Public License as published by
10 the Free Software Foundation; either version 2, or (at your option)
13 This program is distributed in the hope that it will be useful,
14 but WITHOUT ANY WARRANTY; without even the implied warranty of
15 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
16 GNU General Public License for more details.
18 You should have received a copy of the GNU General Public License
19 along with this program; if not, write to the Free Software
20 Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. */
22 /* AIX requires this to be the first thing in the file. */
23 #if defined (_AIX) && !defined (REGEX_MALLOC)
33 /* We need this for `regex.h', and perhaps for the Emacs include files. */
34 #include <sys/types.h>
36 /* The `emacs' switch turns on certain matching commands
37 that make sense only in Emacs. */
44 /* Emacs uses `NULL' as a predicate. */
56 /* This is for other GNU distributions with internationalized messages.
57 The GNU C Library itself does not yet support such messages. */
61 # define gettext(msgid) (msgid)
65 /* We used to test for `BSTRING' here, but only GCC and Emacs define
66 `BSTRING', as far as I know, and neither of them use this code. */
67 #ifndef INHIBIT_STRING_HEADER
68 #if HAVE_STRING_H || STDC_HEADERS
71 #define bcmp(s1, s2, n) memcmp ((s1), (s2), (n))
74 #define bcopy(s, d, n) memcpy ((d), (s), (n))
77 #define bzero(s, n) memset ((s), 0, (n))
84 /* Define the syntax stuff for \<, \>, etc. */
86 /* This must be nonzero for the wordchar and notwordchar pattern
87 commands in re_match_2. */
94 extern char *re_syntax_table;
96 #else /* not SYNTAX_TABLE */
98 /* How many characters in the character set. */
99 #define CHAR_SET_SIZE 256
101 static char re_syntax_table[CHAR_SET_SIZE];
112 bzero (re_syntax_table, sizeof re_syntax_table);
114 for (c = 'a'; c <= 'z'; c++)
115 re_syntax_table[c] = Sword;
117 for (c = 'A'; c <= 'Z'; c++)
118 re_syntax_table[c] = Sword;
120 for (c = '0'; c <= '9'; c++)
121 re_syntax_table[c] = Sword;
123 re_syntax_table['_'] = Sword;
128 #endif /* not SYNTAX_TABLE */
130 #define SYNTAX(c) re_syntax_table[c]
132 #endif /* not emacs */
134 /* Get the interface, including the syntax bits. */
137 /* isalpha etc. are used for the character classes. */
140 /* Jim Meyering writes:
142 "... Some ctype macros are valid only for character codes that
143 isascii says are ASCII (SGI's IRIX-4.0.5 is one such system --when
144 using /bin/cc or gcc but without giving an ansi option). So, all
145 ctype uses should be through macros like ISPRINT... If
146 STDC_HEADERS is defined, then autoconf has verified that the ctype
147 macros don't need to be guarded with references to isascii. ...
148 Defining isascii to 1 should let any compiler worth its salt
149 eliminate the && through constant folding." */
151 #if defined (STDC_HEADERS) || (!defined (isascii) && !defined (HAVE_ISASCII))
154 #define ISASCII(c) isascii(c)
158 #define ISBLANK(c) (ISASCII (c) && isblank (c))
160 #define ISBLANK(c) ((c) == ' ' || (c) == '\t')
163 #define ISGRAPH(c) (ISASCII (c) && isgraph (c))
165 #define ISGRAPH(c) (ISASCII (c) && isprint (c) && !isspace (c))
168 #define ISPRINT(c) (ISASCII (c) && isprint (c))
169 #define ISDIGIT(c) (ISASCII (c) && isdigit (c))
170 #define ISALNUM(c) (ISASCII (c) && isalnum (c))
171 #define ISALPHA(c) (ISASCII (c) && isalpha (c))
172 #define ISCNTRL(c) (ISASCII (c) && iscntrl (c))
173 #define ISLOWER(c) (ISASCII (c) && islower (c))
174 #define ISPUNCT(c) (ISASCII (c) && ispunct (c))
175 #define ISSPACE(c) (ISASCII (c) && isspace (c))
176 #define ISUPPER(c) (ISASCII (c) && isupper (c))
177 #define ISXDIGIT(c) (ISASCII (c) && isxdigit (c))
183 /* We remove any previous definition of `SIGN_EXTEND_CHAR',
184 since ours (we hope) works properly with all combinations of
185 machines, compilers, `char' and `unsigned char' argument types.
186 (Per Bothner suggested the basic approach.) */
187 #undef SIGN_EXTEND_CHAR
189 #define SIGN_EXTEND_CHAR(c) ((signed char) (c))
190 #else /* not __STDC__ */
191 /* As in Harbison and Steele. */
192 #define SIGN_EXTEND_CHAR(c) ((((unsigned char) (c)) ^ 128) - 128)
195 /* Should we use malloc or alloca? If REGEX_MALLOC is not defined, we
196 use `alloca' instead of `malloc'. This is because using malloc in
197 re_search* or re_match* could cause memory leaks when C-g is used in
198 Emacs; also, malloc is slower and causes storage fragmentation. On
199 the other hand, malloc is more portable, and easier to debug.
201 Because we sometimes use alloca, some routines have to be macros,
202 not functions -- `alloca'-allocated space disappears at the end of the
203 function it is called in. */
207 #define REGEX_ALLOCATE malloc
208 #define REGEX_REALLOCATE(source, osize, nsize) realloc (source, nsize)
210 #else /* not REGEX_MALLOC */
212 /* Emacs already defines alloca, sometimes. */
215 /* Make alloca work the best possible way. */
217 #define alloca __builtin_alloca
218 #else /* not __GNUC__ */
221 #else /* not __GNUC__ or HAVE_ALLOCA_H */
222 #ifndef _AIX /* Already did AIX, up at the top. */
224 #endif /* not _AIX */
225 #endif /* not HAVE_ALLOCA_H */
226 #endif /* not __GNUC__ */
228 #endif /* not alloca */
230 #define REGEX_ALLOCATE alloca
232 /* Assumes a `char *destination' variable. */
233 #define REGEX_REALLOCATE(source, osize, nsize) \
234 (destination = (char *) alloca (nsize), \
235 bcopy (source, destination, osize), \
238 #endif /* not REGEX_MALLOC */
241 /* True if `size1' is non-NULL and PTR is pointing anywhere inside
242 `string1' or just past its end. This works if PTR is NULL, which is
244 #define FIRST_STRING_P(ptr) \
245 (size1 && string1 <= (ptr) && (ptr) <= string1 + size1)
247 /* (Re)Allocate N items of type T using malloc, or fail. */
248 #define TALLOC(n, t) ((t *) malloc ((n) * sizeof (t)))
249 #define RETALLOC(addr, n, t) ((addr) = (t *) realloc (addr, (n) * sizeof (t)))
250 #define RETALLOC_IF(addr, n, t) \
251 if (addr) RETALLOC((addr), (n), t); else (addr) = TALLOC ((n), t)
252 #define REGEX_TALLOC(n, t) ((t *) REGEX_ALLOCATE ((n) * sizeof (t)))
254 #define BYTEWIDTH 8 /* In bits. */
256 #define STREQ(s1, s2) ((strcmp (s1, s2) == 0))
260 #define MAX(a, b) ((a) > (b) ? (a) : (b))
261 #define MIN(a, b) ((a) < (b) ? (a) : (b))
263 typedef char boolean;
267 static int re_match_2_internal ();
269 /* These are the command codes that appear in compiled regular
270 expressions. Some opcodes are followed by argument bytes. A
271 command code can specify any interpretation whatsoever for its
272 arguments. Zero bytes may appear in the compiled regular expression. */
278 /* Succeed right away--no more backtracking. */
281 /* Followed by one byte giving n, then by n literal bytes. */
284 /* Matches any (more or less) character. */
287 /* Matches any one char belonging to specified set. First
288 following byte is number of bitmap bytes. Then come bytes
289 for a bitmap saying which chars are in. Bits in each byte
290 are ordered low-bit-first. A character is in the set if its
291 bit is 1. A character too large to have a bit in the map is
292 automatically not in the set. */
295 /* Same parameters as charset, but match any character that is
296 not one of those specified. */
299 /* Start remembering the text that is matched, for storing in a
300 register. Followed by one byte with the register number, in
301 the range 0 to one less than the pattern buffer's re_nsub
302 field. Then followed by one byte with the number of groups
303 inner to this one. (This last has to be part of the
304 start_memory only because we need it in the on_failure_jump
308 /* Stop remembering the text that is matched and store it in a
309 memory register. Followed by one byte with the register
310 number, in the range 0 to one less than `re_nsub' in the
311 pattern buffer, and one byte with the number of inner groups,
312 just like `start_memory'. (We need the number of inner
313 groups here because we don't have any easy way of finding the
314 corresponding start_memory when we're at a stop_memory.) */
317 /* Match a duplicate of something remembered. Followed by one
318 byte containing the register number. */
321 /* Fail unless at beginning of line. */
324 /* Fail unless at end of line. */
327 /* Succeeds if at beginning of buffer (if emacs) or at beginning
328 of string to be matched (if not). */
331 /* Analogously, for end of buffer/string. */
334 /* Followed by two byte relative address to which to jump. */
337 /* Same as jump, but marks the end of an alternative. */
340 /* Followed by two-byte relative address of place to resume at
341 in case of failure. */
344 /* Like on_failure_jump, but pushes a placeholder instead of the
345 current string position when executed. */
346 on_failure_keep_string_jump,
348 /* Throw away latest failure point and then jump to following
349 two-byte relative address. */
352 /* Change to pop_failure_jump if know won't have to backtrack to
353 match; otherwise change to jump. This is used to jump
354 back to the beginning of a repeat. If what follows this jump
355 clearly won't match what the repeat does, such that we can be
356 sure that there is no use backtracking out of repetitions
357 already matched, then we change it to a pop_failure_jump.
358 Followed by two-byte address. */
361 /* Jump to following two-byte address, and push a dummy failure
362 point. This failure point will be thrown away if an attempt
363 is made to use it for a failure. A `+' construct makes this
364 before the first repeat. Also used as an intermediary kind
365 of jump when compiling an alternative. */
368 /* Push a dummy failure point and continue. Used at the end of
372 /* Followed by two-byte relative address and two-byte number n.
373 After matching N times, jump to the address upon failure. */
376 /* Followed by two-byte relative address, and two-byte number n.
377 Jump to the address N times, then fail. */
380 /* Set the following two-byte relative address to the
381 subsequent two-byte number. The address *includes* the two
385 wordchar, /* Matches any word-constituent character. */
386 notwordchar, /* Matches any char that is not a word-constituent. */
388 wordbeg, /* Succeeds if at word beginning. */
389 wordend, /* Succeeds if at word end. */
391 wordbound, /* Succeeds if at a word boundary. */
392 notwordbound /* Succeeds if not at a word boundary. */
395 ,before_dot, /* Succeeds if before point. */
396 at_dot, /* Succeeds if at point. */
397 after_dot, /* Succeeds if after point. */
399 /* Matches any character whose syntax is specified. Followed by
400 a byte which contains a syntax code, e.g., Sword. */
403 /* Matches any character whose syntax is not that specified. */
408 /* Common operations on the compiled pattern. */
410 /* Store NUMBER in two contiguous bytes starting at DESTINATION. */
412 #define STORE_NUMBER(destination, number) \
414 (destination)[0] = (number) & 0377; \
415 (destination)[1] = (number) >> 8; \
418 /* Same as STORE_NUMBER, except increment DESTINATION to
419 the byte after where the number is stored. Therefore, DESTINATION
420 must be an lvalue. */
422 #define STORE_NUMBER_AND_INCR(destination, number) \
424 STORE_NUMBER (destination, number); \
425 (destination) += 2; \
428 /* Put into DESTINATION a number stored in two contiguous bytes starting
431 #define EXTRACT_NUMBER(destination, source) \
433 (destination) = *(source) & 0377; \
434 (destination) += SIGN_EXTEND_CHAR (*((source) + 1)) << 8; \
439 extract_number (dest, source)
441 unsigned char *source;
443 int temp = SIGN_EXTEND_CHAR (*(source + 1));
444 *dest = *source & 0377;
448 #ifndef EXTRACT_MACROS /* To debug the macros. */
449 #undef EXTRACT_NUMBER
450 #define EXTRACT_NUMBER(dest, src) extract_number (&dest, src)
451 #endif /* not EXTRACT_MACROS */
455 /* Same as EXTRACT_NUMBER, except increment SOURCE to after the number.
456 SOURCE must be an lvalue. */
458 #define EXTRACT_NUMBER_AND_INCR(destination, source) \
460 EXTRACT_NUMBER (destination, source); \
466 extract_number_and_incr (destination, source)
468 unsigned char **source;
470 extract_number (destination, *source);
474 #ifndef EXTRACT_MACROS
475 #undef EXTRACT_NUMBER_AND_INCR
476 #define EXTRACT_NUMBER_AND_INCR(dest, src) \
477 extract_number_and_incr (&dest, &src)
478 #endif /* not EXTRACT_MACROS */
482 /* If DEBUG is defined, Regex prints many voluminous messages about what
483 it is doing (if the variable `debug' is nonzero). If linked with the
484 main program in `iregex.c', you can enter patterns and strings
485 interactively. And if linked with the main program in `main.c' and
486 the other test files, you can run the already-written tests. */
490 /* We use standard I/O for debugging. */
493 /* It is useful to test things that ``must'' be true when debugging. */
496 static int debug = 0;
498 #define DEBUG_STATEMENT(e) e
499 #define DEBUG_PRINT1(x) if (debug) printf (x)
500 #define DEBUG_PRINT2(x1, x2) if (debug) printf (x1, x2)
501 #define DEBUG_PRINT3(x1, x2, x3) if (debug) printf (x1, x2, x3)
502 #define DEBUG_PRINT4(x1, x2, x3, x4) if (debug) printf (x1, x2, x3, x4)
503 #define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) \
504 if (debug) print_partial_compiled_pattern (s, e)
505 #define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) \
506 if (debug) print_double_string (w, s1, sz1, s2, sz2)
509 extern void printchar ();
511 /* Print the fastmap in human-readable form. */
514 print_fastmap (fastmap)
517 unsigned was_a_range = 0;
520 while (i < (1 << BYTEWIDTH))
526 while (i < (1 << BYTEWIDTH) && fastmap[i])
542 /* Print a compiled pattern string in human-readable form, starting at
543 the START pointer into it and ending just before the pointer END. */
546 print_partial_compiled_pattern (start, end)
547 unsigned char *start;
551 unsigned char *p = start;
552 unsigned char *pend = end;
560 /* Loop over pattern commands. */
563 printf ("%d:\t", p - start);
565 switch ((re_opcode_t) *p++)
573 printf ("/exactn/%d", mcnt);
584 printf ("/start_memory/%d/%d", mcnt, *p++);
589 printf ("/stop_memory/%d/%d", mcnt, *p++);
593 printf ("/duplicate/%d", *p++);
603 register int c, last = -100;
604 register int in_range = 0;
606 printf ("/charset [%s",
607 (re_opcode_t) *(p - 1) == charset_not ? "^" : "");
609 assert (p + *p < pend);
611 for (c = 0; c < 256; c++)
613 && (p[1 + (c/8)] & (1 << (c % 8))))
615 /* Are we starting a range? */
616 if (last + 1 == c && ! in_range)
621 /* Have we broken a range? */
622 else if (last + 1 != c && in_range)
651 case on_failure_jump:
652 extract_number_and_incr (&mcnt, &p);
653 printf ("/on_failure_jump to %d", p + mcnt - start);
656 case on_failure_keep_string_jump:
657 extract_number_and_incr (&mcnt, &p);
658 printf ("/on_failure_keep_string_jump to %d", p + mcnt - start);
661 case dummy_failure_jump:
662 extract_number_and_incr (&mcnt, &p);
663 printf ("/dummy_failure_jump to %d", p + mcnt - start);
666 case push_dummy_failure:
667 printf ("/push_dummy_failure");
671 extract_number_and_incr (&mcnt, &p);
672 printf ("/maybe_pop_jump to %d", p + mcnt - start);
675 case pop_failure_jump:
676 extract_number_and_incr (&mcnt, &p);
677 printf ("/pop_failure_jump to %d", p + mcnt - start);
681 extract_number_and_incr (&mcnt, &p);
682 printf ("/jump_past_alt to %d", p + mcnt - start);
686 extract_number_and_incr (&mcnt, &p);
687 printf ("/jump to %d", p + mcnt - start);
691 extract_number_and_incr (&mcnt, &p);
692 extract_number_and_incr (&mcnt2, &p);
693 printf ("/succeed_n to %d, %d times", p + mcnt - start, mcnt2);
697 extract_number_and_incr (&mcnt, &p);
698 extract_number_and_incr (&mcnt2, &p);
699 printf ("/jump_n to %d, %d times", p + mcnt - start, mcnt2);
703 extract_number_and_incr (&mcnt, &p);
704 extract_number_and_incr (&mcnt2, &p);
705 printf ("/set_number_at location %d to %d", p + mcnt - start, mcnt2);
709 printf ("/wordbound");
713 printf ("/notwordbound");
725 printf ("/before_dot");
733 printf ("/after_dot");
737 printf ("/syntaxspec");
739 printf ("/%d", mcnt);
743 printf ("/notsyntaxspec");
745 printf ("/%d", mcnt);
750 printf ("/wordchar");
754 printf ("/notwordchar");
766 printf ("?%d", *(p-1));
772 printf ("%d:\tend of pattern.\n", p - start);
777 print_compiled_pattern (bufp)
778 struct re_pattern_buffer *bufp;
780 unsigned char *buffer = bufp->buffer;
782 print_partial_compiled_pattern (buffer, buffer + bufp->used);
783 printf ("%d bytes used/%d bytes allocated.\n", bufp->used, bufp->allocated);
785 if (bufp->fastmap_accurate && bufp->fastmap)
787 printf ("fastmap: ");
788 print_fastmap (bufp->fastmap);
791 printf ("re_nsub: %d\t", bufp->re_nsub);
792 printf ("regs_alloc: %d\t", bufp->regs_allocated);
793 printf ("can_be_null: %d\t", bufp->can_be_null);
794 printf ("newline_anchor: %d\n", bufp->newline_anchor);
795 printf ("no_sub: %d\t", bufp->no_sub);
796 printf ("not_bol: %d\t", bufp->not_bol);
797 printf ("not_eol: %d\t", bufp->not_eol);
798 printf ("syntax: %d\n", bufp->syntax);
799 /* Perhaps we should print the translate table? */
804 print_double_string (where, string1, size1, string2, size2)
817 if (FIRST_STRING_P (where))
819 for (this_char = where - string1; this_char < size1; this_char++)
820 printchar (string1[this_char]);
825 for (this_char = where - string2; this_char < size2; this_char++)
826 printchar (string2[this_char]);
830 #else /* not DEBUG */
835 #define DEBUG_STATEMENT(e)
836 #define DEBUG_PRINT1(x)
837 #define DEBUG_PRINT2(x1, x2)
838 #define DEBUG_PRINT3(x1, x2, x3)
839 #define DEBUG_PRINT4(x1, x2, x3, x4)
840 #define DEBUG_PRINT_COMPILED_PATTERN(p, s, e)
841 #define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2)
843 #endif /* not DEBUG */
845 /* Set by `re_set_syntax' to the current regexp syntax to recognize. Can
846 also be assigned to arbitrarily: each pattern buffer stores its own
847 syntax, so it can be changed between regex compilations. */
848 /* This has no initializer because initialized variables in Emacs
849 become read-only after dumping. */
850 reg_syntax_t re_syntax_options;
853 /* Specify the precise syntax of regexps for compilation. This provides
854 for compatibility for various utilities which historically have
855 different, incompatible syntaxes.
857 The argument SYNTAX is a bit mask comprised of the various bits
858 defined in regex.h. We return the old syntax. */
861 re_set_syntax (syntax)
864 reg_syntax_t ret = re_syntax_options;
866 re_syntax_options = syntax;
870 /* This table gives an error message for each of the error codes listed
871 in regex.h. Obviously the order here has to be same as there.
872 POSIX doesn't require that we do anything for REG_NOERROR,
873 but why not be nice? */
875 static const char *re_error_msgid[] =
876 { "Success", /* REG_NOERROR */
877 "No match", /* REG_NOMATCH */
878 "Invalid regular expression", /* REG_BADPAT */
879 "Invalid collation character", /* REG_ECOLLATE */
880 "Invalid character class name", /* REG_ECTYPE */
881 "Trailing backslash", /* REG_EESCAPE */
882 "Invalid back reference", /* REG_ESUBREG */
883 "Unmatched [ or [^", /* REG_EBRACK */
884 "Unmatched ( or \\(", /* REG_EPAREN */
885 "Unmatched \\{", /* REG_EBRACE */
886 "Invalid content of \\{\\}", /* REG_BADBR */
887 "Invalid range end", /* REG_ERANGE */
888 "Memory exhausted", /* REG_ESPACE */
889 "Invalid preceding regular expression", /* REG_BADRPT */
890 "Premature end of regular expression", /* REG_EEND */
891 "Regular expression too big", /* REG_ESIZE */
892 "Unmatched ) or \\)", /* REG_ERPAREN */
895 /* Avoiding alloca during matching, to placate r_alloc. */
897 /* Define MATCH_MAY_ALLOCATE unless we need to make sure that the
898 searching and matching functions should not call alloca. On some
899 systems, alloca is implemented in terms of malloc, and if we're
900 using the relocating allocator routines, then malloc could cause a
901 relocation, which might (if the strings being searched are in the
902 ralloc heap) shift the data out from underneath the regexp
905 Here's another reason to avoid allocation: Emacs
906 processes input from X in a signal handler; processing X input may
907 call malloc; if input arrives while a matching routine is calling
908 malloc, then we're scrod. But Emacs can't just block input while
909 calling matching routines; then we don't notice interrupts when
910 they come in. So, Emacs blocks input around all regexp calls
911 except the matching calls, which it leaves unprotected, in the
912 faith that they will not malloc. */
914 /* Normally, this is fine. */
915 #define MATCH_MAY_ALLOCATE
917 /* The match routines may not allocate if (1) they would do it with malloc
918 and (2) it's not safe for them to use malloc. */
919 #if (defined (C_ALLOCA) || defined (REGEX_MALLOC)) && (defined (emacs) || defined (REL_ALLOC))
920 #undef MATCH_MAY_ALLOCATE
924 /* Failure stack declarations and macros; both re_compile_fastmap and
925 re_match_2 use a failure stack. These have to be macros because of
929 /* Number of failure points for which to initially allocate space
930 when matching. If this number is exceeded, we allocate more
931 space, so it is not a hard limit. */
932 #ifndef INIT_FAILURE_ALLOC
933 #define INIT_FAILURE_ALLOC 5
936 /* Roughly the maximum number of failure points on the stack. Would be
937 exactly that if always used MAX_FAILURE_SPACE each time we failed.
938 This is a variable only so users of regex can assign to it; we never
939 change it ourselves. */
940 int re_max_failures = 2000;
942 typedef unsigned char *fail_stack_elt_t;
946 fail_stack_elt_t *stack;
948 unsigned avail; /* Offset of next open position. */
951 #define FAIL_STACK_EMPTY() (fail_stack.avail == 0)
952 #define FAIL_STACK_PTR_EMPTY() (fail_stack_ptr->avail == 0)
953 #define FAIL_STACK_FULL() (fail_stack.avail == fail_stack.size)
954 #define FAIL_STACK_TOP() (fail_stack.stack[fail_stack.avail])
957 /* Initialize `fail_stack'. Do `return -2' if the alloc fails. */
959 #ifdef MATCH_MAY_ALLOCATE
960 #define INIT_FAIL_STACK() \
962 fail_stack.stack = (fail_stack_elt_t *) \
963 REGEX_ALLOCATE (INIT_FAILURE_ALLOC * sizeof (fail_stack_elt_t)); \
965 if (fail_stack.stack == NULL) \
968 fail_stack.size = INIT_FAILURE_ALLOC; \
969 fail_stack.avail = 0; \
972 #define INIT_FAIL_STACK() \
974 fail_stack.avail = 0; \
979 /* Double the size of FAIL_STACK, up to approximately `re_max_failures' items.
981 Return 1 if succeeds, and 0 if either ran out of memory
982 allocating space for it or it was already too large.
984 REGEX_REALLOCATE requires `destination' be declared. */
986 #define DOUBLE_FAIL_STACK(fail_stack) \
987 ((fail_stack).size > re_max_failures * MAX_FAILURE_ITEMS \
989 : ((fail_stack).stack = (fail_stack_elt_t *) \
990 REGEX_REALLOCATE ((fail_stack).stack, \
991 (fail_stack).size * sizeof (fail_stack_elt_t), \
992 ((fail_stack).size << 1) * sizeof (fail_stack_elt_t)), \
994 (fail_stack).stack == NULL \
996 : ((fail_stack).size <<= 1, \
1000 /* Push PATTERN_OP on FAIL_STACK.
1002 Return 1 if was able to do so and 0 if ran out of memory allocating
1004 #define PUSH_PATTERN_OP(pattern_op, fail_stack) \
1005 ((FAIL_STACK_FULL () \
1006 && !DOUBLE_FAIL_STACK (fail_stack)) \
1008 : ((fail_stack).stack[(fail_stack).avail++] = pattern_op, \
1011 /* This pushes an item onto the failure stack. Must be a four-byte
1012 value. Assumes the variable `fail_stack'. Probably should only
1013 be called from within `PUSH_FAILURE_POINT'. */
1014 #define PUSH_FAILURE_ITEM(item) \
1015 fail_stack.stack[fail_stack.avail++] = (fail_stack_elt_t) item
1017 /* The complement operation. Assumes `fail_stack' is nonempty. */
1018 #define POP_FAILURE_ITEM() fail_stack.stack[--fail_stack.avail]
1020 /* Used to omit pushing failure point id's when we're not debugging. */
1022 #define DEBUG_PUSH PUSH_FAILURE_ITEM
1023 #define DEBUG_POP(item_addr) *(item_addr) = POP_FAILURE_ITEM ()
1025 #define DEBUG_PUSH(item)
1026 #define DEBUG_POP(item_addr)
1030 /* Push the information about the state we will need
1031 if we ever fail back to it.
1033 Requires variables fail_stack, regstart, regend, reg_info, and
1034 num_regs be declared. DOUBLE_FAIL_STACK requires `destination' be
1037 Does `return FAILURE_CODE' if runs out of memory. */
1039 #define PUSH_FAILURE_POINT(pattern_place, string_place, failure_code) \
1041 char *destination; \
1042 /* Must be int, so when we don't save any registers, the arithmetic \
1043 of 0 + -1 isn't done as unsigned. */ \
1046 DEBUG_STATEMENT (failure_id++); \
1047 DEBUG_STATEMENT (nfailure_points_pushed++); \
1048 DEBUG_PRINT2 ("\nPUSH_FAILURE_POINT #%u:\n", failure_id); \
1049 DEBUG_PRINT2 (" Before push, next avail: %d\n", (fail_stack).avail);\
1050 DEBUG_PRINT2 (" size: %d\n", (fail_stack).size);\
1052 DEBUG_PRINT2 (" slots needed: %d\n", NUM_FAILURE_ITEMS); \
1053 DEBUG_PRINT2 (" available: %d\n", REMAINING_AVAIL_SLOTS); \
1055 /* Ensure we have enough space allocated for what we will push. */ \
1056 while (REMAINING_AVAIL_SLOTS < NUM_FAILURE_ITEMS) \
1058 if (!DOUBLE_FAIL_STACK (fail_stack)) \
1059 return failure_code; \
1061 DEBUG_PRINT2 ("\n Doubled stack; size now: %d\n", \
1062 (fail_stack).size); \
1063 DEBUG_PRINT2 (" slots available: %d\n", REMAINING_AVAIL_SLOTS);\
1066 /* Push the info, starting with the registers. */ \
1067 DEBUG_PRINT1 ("\n"); \
1069 for (this_reg = lowest_active_reg; this_reg <= highest_active_reg; \
1072 DEBUG_PRINT2 (" Pushing reg: %d\n", this_reg); \
1073 DEBUG_STATEMENT (num_regs_pushed++); \
1075 DEBUG_PRINT2 (" start: 0x%x\n", regstart[this_reg]); \
1076 PUSH_FAILURE_ITEM (regstart[this_reg]); \
1078 DEBUG_PRINT2 (" end: 0x%x\n", regend[this_reg]); \
1079 PUSH_FAILURE_ITEM (regend[this_reg]); \
1081 DEBUG_PRINT2 (" info: 0x%x\n ", reg_info[this_reg]); \
1082 DEBUG_PRINT2 (" match_null=%d", \
1083 REG_MATCH_NULL_STRING_P (reg_info[this_reg])); \
1084 DEBUG_PRINT2 (" active=%d", IS_ACTIVE (reg_info[this_reg])); \
1085 DEBUG_PRINT2 (" matched_something=%d", \
1086 MATCHED_SOMETHING (reg_info[this_reg])); \
1087 DEBUG_PRINT2 (" ever_matched=%d", \
1088 EVER_MATCHED_SOMETHING (reg_info[this_reg])); \
1089 DEBUG_PRINT1 ("\n"); \
1090 PUSH_FAILURE_ITEM (reg_info[this_reg].word); \
1093 DEBUG_PRINT2 (" Pushing low active reg: %d\n", lowest_active_reg);\
1094 PUSH_FAILURE_ITEM (lowest_active_reg); \
1096 DEBUG_PRINT2 (" Pushing high active reg: %d\n", highest_active_reg);\
1097 PUSH_FAILURE_ITEM (highest_active_reg); \
1099 DEBUG_PRINT2 (" Pushing pattern 0x%x: ", pattern_place); \
1100 DEBUG_PRINT_COMPILED_PATTERN (bufp, pattern_place, pend); \
1101 PUSH_FAILURE_ITEM (pattern_place); \
1103 DEBUG_PRINT2 (" Pushing string 0x%x: `", string_place); \
1104 DEBUG_PRINT_DOUBLE_STRING (string_place, string1, size1, string2, \
1106 DEBUG_PRINT1 ("'\n"); \
1107 PUSH_FAILURE_ITEM (string_place); \
1109 DEBUG_PRINT2 (" Pushing failure id: %u\n", failure_id); \
1110 DEBUG_PUSH (failure_id); \
1113 /* This is the number of items that are pushed and popped on the stack
1114 for each register. */
1115 #define NUM_REG_ITEMS 3
1117 /* Individual items aside from the registers. */
1119 #define NUM_NONREG_ITEMS 5 /* Includes failure point id. */
1121 #define NUM_NONREG_ITEMS 4
1124 /* We push at most this many items on the stack. */
1125 #define MAX_FAILURE_ITEMS ((num_regs - 1) * NUM_REG_ITEMS + NUM_NONREG_ITEMS)
1127 /* We actually push this many items. */
1128 #define NUM_FAILURE_ITEMS \
1129 ((highest_active_reg - lowest_active_reg + 1) * NUM_REG_ITEMS \
1132 /* How many items can still be added to the stack without overflowing it. */
1133 #define REMAINING_AVAIL_SLOTS ((fail_stack).size - (fail_stack).avail)
1136 /* Pops what PUSH_FAIL_STACK pushes.
1138 We restore into the parameters, all of which should be lvalues:
1139 STR -- the saved data position.
1140 PAT -- the saved pattern position.
1141 LOW_REG, HIGH_REG -- the highest and lowest active registers.
1142 REGSTART, REGEND -- arrays of string positions.
1143 REG_INFO -- array of information about each subexpression.
1145 Also assumes the variables `fail_stack' and (if debugging), `bufp',
1146 `pend', `string1', `size1', `string2', and `size2'. */
1148 #define POP_FAILURE_POINT(str, pat, low_reg, high_reg, regstart, regend, reg_info)\
1150 DEBUG_STATEMENT (fail_stack_elt_t failure_id;) \
1152 const unsigned char *string_temp; \
1154 assert (!FAIL_STACK_EMPTY ()); \
1156 /* Remove failure points and point to how many regs pushed. */ \
1157 DEBUG_PRINT1 ("POP_FAILURE_POINT:\n"); \
1158 DEBUG_PRINT2 (" Before pop, next avail: %d\n", fail_stack.avail); \
1159 DEBUG_PRINT2 (" size: %d\n", fail_stack.size); \
1161 assert (fail_stack.avail >= NUM_NONREG_ITEMS); \
1163 DEBUG_POP (&failure_id); \
1164 DEBUG_PRINT2 (" Popping failure id: %u\n", failure_id); \
1166 /* If the saved string location is NULL, it came from an \
1167 on_failure_keep_string_jump opcode, and we want to throw away the \
1168 saved NULL, thus retaining our current position in the string. */ \
1169 string_temp = POP_FAILURE_ITEM (); \
1170 if (string_temp != NULL) \
1171 str = (const char *) string_temp; \
1173 DEBUG_PRINT2 (" Popping string 0x%x: `", str); \
1174 DEBUG_PRINT_DOUBLE_STRING (str, string1, size1, string2, size2); \
1175 DEBUG_PRINT1 ("'\n"); \
1177 pat = (unsigned char *) POP_FAILURE_ITEM (); \
1178 DEBUG_PRINT2 (" Popping pattern 0x%x: ", pat); \
1179 DEBUG_PRINT_COMPILED_PATTERN (bufp, pat, pend); \
1181 /* Restore register info. */ \
1182 high_reg = (unsigned) POP_FAILURE_ITEM (); \
1183 DEBUG_PRINT2 (" Popping high active reg: %d\n", high_reg); \
1185 low_reg = (unsigned) POP_FAILURE_ITEM (); \
1186 DEBUG_PRINT2 (" Popping low active reg: %d\n", low_reg); \
1188 for (this_reg = high_reg; this_reg >= low_reg; this_reg--) \
1190 DEBUG_PRINT2 (" Popping reg: %d\n", this_reg); \
1192 reg_info[this_reg].word = POP_FAILURE_ITEM (); \
1193 DEBUG_PRINT2 (" info: 0x%x\n", reg_info[this_reg]); \
1195 regend[this_reg] = (const char *) POP_FAILURE_ITEM (); \
1196 DEBUG_PRINT2 (" end: 0x%x\n", regend[this_reg]); \
1198 regstart[this_reg] = (const char *) POP_FAILURE_ITEM (); \
1199 DEBUG_PRINT2 (" start: 0x%x\n", regstart[this_reg]); \
1202 DEBUG_STATEMENT (nfailure_points_popped++); \
1203 } /* POP_FAILURE_POINT */
1207 /* Structure for per-register (a.k.a. per-group) information.
1208 This must not be longer than one word, because we push this value
1209 onto the failure stack. Other register information, such as the
1210 starting and ending positions (which are addresses), and the list of
1211 inner groups (which is a bits list) are maintained in separate
1214 We are making a (strictly speaking) nonportable assumption here: that
1215 the compiler will pack our bit fields into something that fits into
1216 the type of `word', i.e., is something that fits into one item on the
1220 fail_stack_elt_t word;
1223 /* This field is one if this group can match the empty string,
1224 zero if not. If not yet determined, `MATCH_NULL_UNSET_VALUE'. */
1225 #define MATCH_NULL_UNSET_VALUE 3
1226 unsigned match_null_string_p : 2;
1227 unsigned is_active : 1;
1228 unsigned matched_something : 1;
1229 unsigned ever_matched_something : 1;
1231 } register_info_type;
1233 #define REG_MATCH_NULL_STRING_P(R) ((R).bits.match_null_string_p)
1234 #define IS_ACTIVE(R) ((R).bits.is_active)
1235 #define MATCHED_SOMETHING(R) ((R).bits.matched_something)
1236 #define EVER_MATCHED_SOMETHING(R) ((R).bits.ever_matched_something)
1239 /* Call this when have matched a real character; it sets `matched' flags
1240 for the subexpressions which we are currently inside. Also records
1241 that those subexprs have matched. */
1242 #define SET_REGS_MATCHED() \
1246 for (r = lowest_active_reg; r <= highest_active_reg; r++) \
1248 MATCHED_SOMETHING (reg_info[r]) \
1249 = EVER_MATCHED_SOMETHING (reg_info[r]) \
1256 /* Registers are set to a sentinel when they haven't yet matched. */
1257 #define REG_UNSET_VALUE ((char *) -1)
1258 #define REG_UNSET(e) ((e) == REG_UNSET_VALUE)
1262 /* How do we implement a missing MATCH_MAY_ALLOCATE?
1263 We make the fail stack a global thing, and then grow it to
1264 re_max_failures when we compile. */
1265 #ifndef MATCH_MAY_ALLOCATE
1266 static fail_stack_type fail_stack;
1268 static const char ** regstart, ** regend;
1269 static const char ** old_regstart, ** old_regend;
1270 static const char **best_regstart, **best_regend;
1271 static register_info_type *reg_info;
1272 static const char **reg_dummy;
1273 static register_info_type *reg_info_dummy;
1277 /* Subroutine declarations and macros for regex_compile. */
1279 static void store_op1 (), store_op2 ();
1280 static void insert_op1 (), insert_op2 ();
1281 static boolean at_begline_loc_p (), at_endline_loc_p ();
1282 static boolean group_in_compile_stack ();
1283 static reg_errcode_t compile_range ();
1285 /* Fetch the next character in the uncompiled pattern---translating it
1286 if necessary. Also cast from a signed character in the constant
1287 string passed to us by the user to an unsigned char that we can use
1288 as an array index (in, e.g., `translate'). */
1289 #define PATFETCH(c) \
1290 do {if (p == pend) return REG_EEND; \
1291 c = (unsigned char) *p++; \
1292 if (translate) c = translate[c]; \
1295 /* Fetch the next character in the uncompiled pattern, with no
1297 #define PATFETCH_RAW(c) \
1298 do {if (p == pend) return REG_EEND; \
1299 c = (unsigned char) *p++; \
1302 /* Go backwards one character in the pattern. */
1303 #define PATUNFETCH p--
1306 /* If `translate' is non-null, return translate[D], else just D. We
1307 cast the subscript to translate because some data is declared as
1308 `char *', to avoid warnings when a string constant is passed. But
1309 when we use a character as a subscript we must make it unsigned. */
1310 #define TRANSLATE(d) (translate ? translate[(unsigned char) (d)] : (d))
1313 /* Macros for outputting the compiled pattern into `buffer'. */
1315 /* If the buffer isn't allocated when it comes in, use this. */
1316 #define INIT_BUF_SIZE 32
1318 /* Make sure we have at least N more bytes of space in buffer. */
1319 #define GET_BUFFER_SPACE(n) \
1320 while (b - bufp->buffer + (n) > bufp->allocated) \
1323 /* Make sure we have one more byte of buffer space and then add C to it. */
1324 #define BUF_PUSH(c) \
1326 GET_BUFFER_SPACE (1); \
1327 *b++ = (unsigned char) (c); \
1331 /* Ensure we have two more bytes of buffer space and then append C1 and C2. */
1332 #define BUF_PUSH_2(c1, c2) \
1334 GET_BUFFER_SPACE (2); \
1335 *b++ = (unsigned char) (c1); \
1336 *b++ = (unsigned char) (c2); \
1340 /* As with BUF_PUSH_2, except for three bytes. */
1341 #define BUF_PUSH_3(c1, c2, c3) \
1343 GET_BUFFER_SPACE (3); \
1344 *b++ = (unsigned char) (c1); \
1345 *b++ = (unsigned char) (c2); \
1346 *b++ = (unsigned char) (c3); \
1350 /* Store a jump with opcode OP at LOC to location TO. We store a
1351 relative address offset by the three bytes the jump itself occupies. */
1352 #define STORE_JUMP(op, loc, to) \
1353 store_op1 (op, loc, (to) - (loc) - 3)
1355 /* Likewise, for a two-argument jump. */
1356 #define STORE_JUMP2(op, loc, to, arg) \
1357 store_op2 (op, loc, (to) - (loc) - 3, arg)
1359 /* Like `STORE_JUMP', but for inserting. Assume `b' is the buffer end. */
1360 #define INSERT_JUMP(op, loc, to) \
1361 insert_op1 (op, loc, (to) - (loc) - 3, b)
1363 /* Like `STORE_JUMP2', but for inserting. Assume `b' is the buffer end. */
1364 #define INSERT_JUMP2(op, loc, to, arg) \
1365 insert_op2 (op, loc, (to) - (loc) - 3, arg, b)
1368 /* This is not an arbitrary limit: the arguments which represent offsets
1369 into the pattern are two bytes long. So if 2^16 bytes turns out to
1370 be too small, many things would have to change. */
1371 #define MAX_BUF_SIZE (1L << 16)
1374 /* Extend the buffer by twice its current size via realloc and
1375 reset the pointers that pointed into the old block to point to the
1376 correct places in the new one. If extending the buffer results in it
1377 being larger than MAX_BUF_SIZE, then flag memory exhausted. */
1378 #define EXTEND_BUFFER() \
1380 unsigned char *old_buffer = bufp->buffer; \
1381 if (bufp->allocated == MAX_BUF_SIZE) \
1383 bufp->allocated <<= 1; \
1384 if (bufp->allocated > MAX_BUF_SIZE) \
1385 bufp->allocated = MAX_BUF_SIZE; \
1386 bufp->buffer = (unsigned char *) realloc (bufp->buffer, bufp->allocated);\
1387 if (bufp->buffer == NULL) \
1388 return REG_ESPACE; \
1389 /* If the buffer moved, move all the pointers into it. */ \
1390 if (old_buffer != bufp->buffer) \
1392 b = (b - old_buffer) + bufp->buffer; \
1393 begalt = (begalt - old_buffer) + bufp->buffer; \
1394 if (fixup_alt_jump) \
1395 fixup_alt_jump = (fixup_alt_jump - old_buffer) + bufp->buffer;\
1397 laststart = (laststart - old_buffer) + bufp->buffer; \
1398 if (pending_exact) \
1399 pending_exact = (pending_exact - old_buffer) + bufp->buffer; \
1404 /* Since we have one byte reserved for the register number argument to
1405 {start,stop}_memory, the maximum number of groups we can report
1406 things about is what fits in that byte. */
1407 #define MAX_REGNUM 255
1409 /* But patterns can have more than `MAX_REGNUM' registers. We just
1410 ignore the excess. */
1411 typedef unsigned regnum_t;
1414 /* Macros for the compile stack. */
1416 /* Since offsets can go either forwards or backwards, this type needs to
1417 be able to hold values from -(MAX_BUF_SIZE - 1) to MAX_BUF_SIZE - 1. */
1418 typedef int pattern_offset_t;
1422 pattern_offset_t begalt_offset;
1423 pattern_offset_t fixup_alt_jump;
1424 pattern_offset_t inner_group_offset;
1425 pattern_offset_t laststart_offset;
1427 } compile_stack_elt_t;
1432 compile_stack_elt_t *stack;
1434 unsigned avail; /* Offset of next open position. */
1435 } compile_stack_type;
1438 #define INIT_COMPILE_STACK_SIZE 32
1440 #define COMPILE_STACK_EMPTY (compile_stack.avail == 0)
1441 #define COMPILE_STACK_FULL (compile_stack.avail == compile_stack.size)
1443 /* The next available element. */
1444 #define COMPILE_STACK_TOP (compile_stack.stack[compile_stack.avail])
1447 /* Set the bit for character C in a list. */
1448 #define SET_LIST_BIT(c) \
1449 (b[((unsigned char) (c)) / BYTEWIDTH] \
1450 |= 1 << (((unsigned char) c) % BYTEWIDTH))
1453 /* Get the next unsigned number in the uncompiled pattern. */
1454 #define GET_UNSIGNED_NUMBER(num) \
1458 while (ISDIGIT (c)) \
1462 num = num * 10 + c - '0'; \
1470 #define CHAR_CLASS_MAX_LENGTH 6 /* Namely, `xdigit'. */
1472 #define IS_CHAR_CLASS(string) \
1473 (STREQ (string, "alpha") || STREQ (string, "upper") \
1474 || STREQ (string, "lower") || STREQ (string, "digit") \
1475 || STREQ (string, "alnum") || STREQ (string, "xdigit") \
1476 || STREQ (string, "space") || STREQ (string, "print") \
1477 || STREQ (string, "punct") || STREQ (string, "graph") \
1478 || STREQ (string, "cntrl") || STREQ (string, "blank"))
1480 /* `regex_compile' compiles PATTERN (of length SIZE) according to SYNTAX.
1481 Returns one of error codes defined in `regex.h', or zero for success.
1483 Assumes the `allocated' (and perhaps `buffer') and `translate'
1484 fields are set in BUFP on entry.
1486 If it succeeds, results are put in BUFP (if it returns an error, the
1487 contents of BUFP are undefined):
1488 `buffer' is the compiled pattern;
1489 `syntax' is set to SYNTAX;
1490 `used' is set to the length of the compiled pattern;
1491 `fastmap_accurate' is zero;
1492 `re_nsub' is the number of subexpressions in PATTERN;
1493 `not_bol' and `not_eol' are zero;
1495 The `fastmap' and `newline_anchor' fields are neither
1496 examined nor set. */
1498 /* Return, freeing storage we allocated. */
1499 #define FREE_STACK_RETURN(value) \
1500 return (free (compile_stack.stack), value)
1502 static reg_errcode_t
1503 regex_compile (pattern, size, syntax, bufp)
1504 const char *pattern;
1506 reg_syntax_t syntax;
1507 struct re_pattern_buffer *bufp;
1509 /* We fetch characters from PATTERN here. Even though PATTERN is
1510 `char *' (i.e., signed), we declare these variables as unsigned, so
1511 they can be reliably used as array indices. */
1512 register unsigned char c, c1;
1514 /* A random temporary spot in PATTERN. */
1517 /* Points to the end of the buffer, where we should append. */
1518 register unsigned char *b;
1520 /* Keeps track of unclosed groups. */
1521 compile_stack_type compile_stack;
1523 /* Points to the current (ending) position in the pattern. */
1524 const char *p = pattern;
1525 const char *pend = pattern + size;
1527 /* How to translate the characters in the pattern. */
1528 char *translate = bufp->translate;
1530 /* Address of the count-byte of the most recently inserted `exactn'
1531 command. This makes it possible to tell if a new exact-match
1532 character can be added to that command or if the character requires
1533 a new `exactn' command. */
1534 unsigned char *pending_exact = 0;
1536 /* Address of start of the most recently finished expression.
1537 This tells, e.g., postfix * where to find the start of its
1538 operand. Reset at the beginning of groups and alternatives. */
1539 unsigned char *laststart = 0;
1541 /* Address of beginning of regexp, or inside of last group. */
1542 unsigned char *begalt;
1544 /* Place in the uncompiled pattern (i.e., the {) to
1545 which to go back if the interval is invalid. */
1546 const char *beg_interval;
1548 /* Address of the place where a forward jump should go to the end of
1549 the containing expression. Each alternative of an `or' -- except the
1550 last -- ends with a forward jump of this sort. */
1551 unsigned char *fixup_alt_jump = 0;
1553 /* Counts open-groups as they are encountered. Remembered for the
1554 matching close-group on the compile stack, so the same register
1555 number is put in the stop_memory as the start_memory. */
1556 regnum_t regnum = 0;
1559 DEBUG_PRINT1 ("\nCompiling pattern: ");
1562 unsigned debug_count;
1564 for (debug_count = 0; debug_count < size; debug_count++)
1565 printchar (pattern[debug_count]);
1570 /* Initialize the compile stack. */
1571 compile_stack.stack = TALLOC (INIT_COMPILE_STACK_SIZE, compile_stack_elt_t);
1572 if (compile_stack.stack == NULL)
1575 compile_stack.size = INIT_COMPILE_STACK_SIZE;
1576 compile_stack.avail = 0;
1578 /* Initialize the pattern buffer. */
1579 bufp->syntax = syntax;
1580 bufp->fastmap_accurate = 0;
1581 bufp->not_bol = bufp->not_eol = 0;
1583 /* Set `used' to zero, so that if we return an error, the pattern
1584 printer (for debugging) will think there's no pattern. We reset it
1588 /* Always count groups, whether or not bufp->no_sub is set. */
1591 #if !defined (emacs) && !defined (SYNTAX_TABLE)
1592 /* Initialize the syntax table. */
1593 init_syntax_once ();
1596 if (bufp->allocated == 0)
1599 { /* If zero allocated, but buffer is non-null, try to realloc
1600 enough space. This loses if buffer's address is bogus, but
1601 that is the user's responsibility. */
1602 RETALLOC (bufp->buffer, INIT_BUF_SIZE, unsigned char);
1605 { /* Caller did not allocate a buffer. Do it for them. */
1606 bufp->buffer = TALLOC (INIT_BUF_SIZE, unsigned char);
1608 if (!bufp->buffer) FREE_STACK_RETURN (REG_ESPACE);
1610 bufp->allocated = INIT_BUF_SIZE;
1613 begalt = b = bufp->buffer;
1615 /* Loop through the uncompiled pattern until we're at the end. */
1624 if ( /* If at start of pattern, it's an operator. */
1626 /* If context independent, it's an operator. */
1627 || syntax & RE_CONTEXT_INDEP_ANCHORS
1628 /* Otherwise, depends on what's come before. */
1629 || at_begline_loc_p (pattern, p, syntax))
1639 if ( /* If at end of pattern, it's an operator. */
1641 /* If context independent, it's an operator. */
1642 || syntax & RE_CONTEXT_INDEP_ANCHORS
1643 /* Otherwise, depends on what's next. */
1644 || at_endline_loc_p (p, pend, syntax))
1654 if ((syntax & RE_BK_PLUS_QM)
1655 || (syntax & RE_LIMITED_OPS))
1659 /* If there is no previous pattern... */
1662 if (syntax & RE_CONTEXT_INVALID_OPS)
1663 FREE_STACK_RETURN (REG_BADRPT);
1664 else if (!(syntax & RE_CONTEXT_INDEP_OPS))
1669 /* Are we optimizing this jump? */
1670 boolean keep_string_p = false;
1672 /* 1 means zero (many) matches is allowed. */
1673 char zero_times_ok = 0, many_times_ok = 0;
1675 /* If there is a sequence of repetition chars, collapse it
1676 down to just one (the right one). We can't combine
1677 interval operators with these because of, e.g., `a{2}*',
1678 which should only match an even number of `a's. */
1682 zero_times_ok |= c != '+';
1683 many_times_ok |= c != '?';
1691 || (!(syntax & RE_BK_PLUS_QM) && (c == '+' || c == '?')))
1694 else if (syntax & RE_BK_PLUS_QM && c == '\\')
1696 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
1699 if (!(c1 == '+' || c1 == '?'))
1714 /* If we get here, we found another repeat character. */
1717 /* Star, etc. applied to an empty pattern is equivalent
1718 to an empty pattern. */
1722 /* Now we know whether or not zero matches is allowed
1723 and also whether or not two or more matches is allowed. */
1725 { /* More than one repetition is allowed, so put in at the
1726 end a backward relative jump from `b' to before the next
1727 jump we're going to put in below (which jumps from
1728 laststart to after this jump).
1730 But if we are at the `*' in the exact sequence `.*\n',
1731 insert an unconditional jump backwards to the .,
1732 instead of the beginning of the loop. This way we only
1733 push a failure point once, instead of every time
1734 through the loop. */
1735 assert (p - 1 > pattern);
1737 /* Allocate the space for the jump. */
1738 GET_BUFFER_SPACE (3);
1740 /* We know we are not at the first character of the pattern,
1741 because laststart was nonzero. And we've already
1742 incremented `p', by the way, to be the character after
1743 the `*'. Do we have to do something analogous here
1744 for null bytes, because of RE_DOT_NOT_NULL? */
1745 if (TRANSLATE (*(p - 2)) == TRANSLATE ('.')
1747 && p < pend && TRANSLATE (*p) == TRANSLATE ('\n')
1748 && !(syntax & RE_DOT_NEWLINE))
1749 { /* We have .*\n. */
1750 STORE_JUMP (jump, b, laststart);
1751 keep_string_p = true;
1754 /* Anything else. */
1755 STORE_JUMP (maybe_pop_jump, b, laststart - 3);
1757 /* We've added more stuff to the buffer. */
1761 /* On failure, jump from laststart to b + 3, which will be the
1762 end of the buffer after this jump is inserted. */
1763 GET_BUFFER_SPACE (3);
1764 INSERT_JUMP (keep_string_p ? on_failure_keep_string_jump
1772 /* At least one repetition is required, so insert a
1773 `dummy_failure_jump' before the initial
1774 `on_failure_jump' instruction of the loop. This
1775 effects a skip over that instruction the first time
1776 we hit that loop. */
1777 GET_BUFFER_SPACE (3);
1778 INSERT_JUMP (dummy_failure_jump, laststart, laststart + 6);
1793 boolean had_char_class = false;
1795 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
1797 /* Ensure that we have enough space to push a charset: the
1798 opcode, the length count, and the bitset; 34 bytes in all. */
1799 GET_BUFFER_SPACE (34);
1803 /* We test `*p == '^' twice, instead of using an if
1804 statement, so we only need one BUF_PUSH. */
1805 BUF_PUSH (*p == '^' ? charset_not : charset);
1809 /* Remember the first position in the bracket expression. */
1812 /* Push the number of bytes in the bitmap. */
1813 BUF_PUSH ((1 << BYTEWIDTH) / BYTEWIDTH);
1815 /* Clear the whole map. */
1816 bzero (b, (1 << BYTEWIDTH) / BYTEWIDTH);
1818 /* charset_not matches newline according to a syntax bit. */
1819 if ((re_opcode_t) b[-2] == charset_not
1820 && (syntax & RE_HAT_LISTS_NOT_NEWLINE))
1821 SET_LIST_BIT ('\n');
1823 /* Read in characters and ranges, setting map bits. */
1826 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
1830 /* \ might escape characters inside [...] and [^...]. */
1831 if ((syntax & RE_BACKSLASH_ESCAPE_IN_LISTS) && c == '\\')
1833 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
1840 /* Could be the end of the bracket expression. If it's
1841 not (i.e., when the bracket expression is `[]' so
1842 far), the ']' character bit gets set way below. */
1843 if (c == ']' && p != p1 + 1)
1846 /* Look ahead to see if it's a range when the last thing
1847 was a character class. */
1848 if (had_char_class && c == '-' && *p != ']')
1849 FREE_STACK_RETURN (REG_ERANGE);
1851 /* Look ahead to see if it's a range when the last thing
1852 was a character: if this is a hyphen not at the
1853 beginning or the end of a list, then it's the range
1856 && !(p - 2 >= pattern && p[-2] == '[')
1857 && !(p - 3 >= pattern && p[-3] == '[' && p[-2] == '^')
1861 = compile_range (&p, pend, translate, syntax, b);
1862 if (ret != REG_NOERROR) FREE_STACK_RETURN (ret);
1865 else if (p[0] == '-' && p[1] != ']')
1866 { /* This handles ranges made up of characters only. */
1869 /* Move past the `-'. */
1872 ret = compile_range (&p, pend, translate, syntax, b);
1873 if (ret != REG_NOERROR) FREE_STACK_RETURN (ret);
1876 /* See if we're at the beginning of a possible character
1879 else if (syntax & RE_CHAR_CLASSES && c == '[' && *p == ':')
1880 { /* Leave room for the null. */
1881 char str[CHAR_CLASS_MAX_LENGTH + 1];
1886 /* If pattern is `[[:'. */
1887 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
1892 if (c == ':' || c == ']' || p == pend
1893 || c1 == CHAR_CLASS_MAX_LENGTH)
1899 /* If isn't a word bracketed by `[:' and:`]':
1900 undo the ending character, the letters, and leave
1901 the leading `:' and `[' (but set bits for them). */
1902 if (c == ':' && *p == ']')
1905 boolean is_alnum = STREQ (str, "alnum");
1906 boolean is_alpha = STREQ (str, "alpha");
1907 boolean is_blank = STREQ (str, "blank");
1908 boolean is_cntrl = STREQ (str, "cntrl");
1909 boolean is_digit = STREQ (str, "digit");
1910 boolean is_graph = STREQ (str, "graph");
1911 boolean is_lower = STREQ (str, "lower");
1912 boolean is_print = STREQ (str, "print");
1913 boolean is_punct = STREQ (str, "punct");
1914 boolean is_space = STREQ (str, "space");
1915 boolean is_upper = STREQ (str, "upper");
1916 boolean is_xdigit = STREQ (str, "xdigit");
1918 if (!IS_CHAR_CLASS (str))
1919 FREE_STACK_RETURN (REG_ECTYPE);
1921 /* Throw away the ] at the end of the character
1925 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
1927 for (ch = 0; ch < 1 << BYTEWIDTH; ch++)
1929 /* This was split into 3 if's to
1930 avoid an arbitrary limit in some compiler. */
1931 if ( (is_alnum && ISALNUM (ch))
1932 || (is_alpha && ISALPHA (ch))
1933 || (is_blank && ISBLANK (ch))
1934 || (is_cntrl && ISCNTRL (ch)))
1936 if ( (is_digit && ISDIGIT (ch))
1937 || (is_graph && ISGRAPH (ch))
1938 || (is_lower && ISLOWER (ch))
1939 || (is_print && ISPRINT (ch)))
1941 if ( (is_punct && ISPUNCT (ch))
1942 || (is_space && ISSPACE (ch))
1943 || (is_upper && ISUPPER (ch))
1944 || (is_xdigit && ISXDIGIT (ch)))
1947 had_char_class = true;
1956 had_char_class = false;
1961 had_char_class = false;
1966 /* Discard any (non)matching list bytes that are all 0 at the
1967 end of the map. Decrease the map-length byte too. */
1968 while ((int) b[-1] > 0 && b[b[-1] - 1] == 0)
1976 if (syntax & RE_NO_BK_PARENS)
1983 if (syntax & RE_NO_BK_PARENS)
1990 if (syntax & RE_NEWLINE_ALT)
1997 if (syntax & RE_NO_BK_VBAR)
2004 if (syntax & RE_INTERVALS && syntax & RE_NO_BK_BRACES)
2005 goto handle_interval;
2011 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
2013 /* Do not translate the character after the \, so that we can
2014 distinguish, e.g., \B from \b, even if we normally would
2015 translate, e.g., B to b. */
2021 if (syntax & RE_NO_BK_PARENS)
2022 goto normal_backslash;
2028 if (COMPILE_STACK_FULL)
2030 RETALLOC (compile_stack.stack, compile_stack.size << 1,
2031 compile_stack_elt_t);
2032 if (compile_stack.stack == NULL) return REG_ESPACE;
2034 compile_stack.size <<= 1;
2037 /* These are the values to restore when we hit end of this
2038 group. They are all relative offsets, so that if the
2039 whole pattern moves because of realloc, they will still
2041 COMPILE_STACK_TOP.begalt_offset = begalt - bufp->buffer;
2042 COMPILE_STACK_TOP.fixup_alt_jump
2043 = fixup_alt_jump ? fixup_alt_jump - bufp->buffer + 1 : 0;
2044 COMPILE_STACK_TOP.laststart_offset = b - bufp->buffer;
2045 COMPILE_STACK_TOP.regnum = regnum;
2047 /* We will eventually replace the 0 with the number of
2048 groups inner to this one. But do not push a
2049 start_memory for groups beyond the last one we can
2050 represent in the compiled pattern. */
2051 if (regnum <= MAX_REGNUM)
2053 COMPILE_STACK_TOP.inner_group_offset = b - bufp->buffer + 2;
2054 BUF_PUSH_3 (start_memory, regnum, 0);
2057 compile_stack.avail++;
2062 /* If we've reached MAX_REGNUM groups, then this open
2063 won't actually generate any code, so we'll have to
2064 clear pending_exact explicitly. */
2070 if (syntax & RE_NO_BK_PARENS) goto normal_backslash;
2072 if (COMPILE_STACK_EMPTY)
2073 if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD)
2074 goto normal_backslash;
2076 FREE_STACK_RETURN (REG_ERPAREN);
2080 { /* Push a dummy failure point at the end of the
2081 alternative for a possible future
2082 `pop_failure_jump' to pop. See comments at
2083 `push_dummy_failure' in `re_match_2'. */
2084 BUF_PUSH (push_dummy_failure);
2086 /* We allocated space for this jump when we assigned
2087 to `fixup_alt_jump', in the `handle_alt' case below. */
2088 STORE_JUMP (jump_past_alt, fixup_alt_jump, b - 1);
2091 /* See similar code for backslashed left paren above. */
2092 if (COMPILE_STACK_EMPTY)
2093 if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD)
2096 FREE_STACK_RETURN (REG_ERPAREN);
2098 /* Since we just checked for an empty stack above, this
2099 ``can't happen''. */
2100 assert (compile_stack.avail != 0);
2102 /* We don't just want to restore into `regnum', because
2103 later groups should continue to be numbered higher,
2104 as in `(ab)c(de)' -- the second group is #2. */
2105 regnum_t this_group_regnum;
2107 compile_stack.avail--;
2108 begalt = bufp->buffer + COMPILE_STACK_TOP.begalt_offset;
2110 = COMPILE_STACK_TOP.fixup_alt_jump
2111 ? bufp->buffer + COMPILE_STACK_TOP.fixup_alt_jump - 1
2113 laststart = bufp->buffer + COMPILE_STACK_TOP.laststart_offset;
2114 this_group_regnum = COMPILE_STACK_TOP.regnum;
2115 /* If we've reached MAX_REGNUM groups, then this open
2116 won't actually generate any code, so we'll have to
2117 clear pending_exact explicitly. */
2120 /* We're at the end of the group, so now we know how many
2121 groups were inside this one. */
2122 if (this_group_regnum <= MAX_REGNUM)
2124 unsigned char *inner_group_loc
2125 = bufp->buffer + COMPILE_STACK_TOP.inner_group_offset;
2127 *inner_group_loc = regnum - this_group_regnum;
2128 BUF_PUSH_3 (stop_memory, this_group_regnum,
2129 regnum - this_group_regnum);
2135 case '|': /* `\|'. */
2136 if (syntax & RE_LIMITED_OPS || syntax & RE_NO_BK_VBAR)
2137 goto normal_backslash;
2139 if (syntax & RE_LIMITED_OPS)
2142 /* Insert before the previous alternative a jump which
2143 jumps to this alternative if the former fails. */
2144 GET_BUFFER_SPACE (3);
2145 INSERT_JUMP (on_failure_jump, begalt, b + 6);
2149 /* The alternative before this one has a jump after it
2150 which gets executed if it gets matched. Adjust that
2151 jump so it will jump to this alternative's analogous
2152 jump (put in below, which in turn will jump to the next
2153 (if any) alternative's such jump, etc.). The last such
2154 jump jumps to the correct final destination. A picture:
2160 If we are at `b', then fixup_alt_jump right now points to a
2161 three-byte space after `a'. We'll put in the jump, set
2162 fixup_alt_jump to right after `b', and leave behind three
2163 bytes which we'll fill in when we get to after `c'. */
2166 STORE_JUMP (jump_past_alt, fixup_alt_jump, b);
2168 /* Mark and leave space for a jump after this alternative,
2169 to be filled in later either by next alternative or
2170 when know we're at the end of a series of alternatives. */
2172 GET_BUFFER_SPACE (3);
2181 /* If \{ is a literal. */
2182 if (!(syntax & RE_INTERVALS)
2183 /* If we're at `\{' and it's not the open-interval
2185 || ((syntax & RE_INTERVALS) && (syntax & RE_NO_BK_BRACES))
2186 || (p - 2 == pattern && p == pend))
2187 goto normal_backslash;
2191 /* If got here, then the syntax allows intervals. */
2193 /* At least (most) this many matches must be made. */
2194 int lower_bound = -1, upper_bound = -1;
2196 beg_interval = p - 1;
2200 if (syntax & RE_NO_BK_BRACES)
2201 goto unfetch_interval;
2203 FREE_STACK_RETURN (REG_EBRACE);
2206 GET_UNSIGNED_NUMBER (lower_bound);
2210 GET_UNSIGNED_NUMBER (upper_bound);
2211 if (upper_bound < 0) upper_bound = RE_DUP_MAX;
2214 /* Interval such as `{1}' => match exactly once. */
2215 upper_bound = lower_bound;
2217 if (lower_bound < 0 || upper_bound > RE_DUP_MAX
2218 || lower_bound > upper_bound)
2220 if (syntax & RE_NO_BK_BRACES)
2221 goto unfetch_interval;
2223 FREE_STACK_RETURN (REG_BADBR);
2226 if (!(syntax & RE_NO_BK_BRACES))
2228 if (c != '\\') FREE_STACK_RETURN (REG_EBRACE);
2235 if (syntax & RE_NO_BK_BRACES)
2236 goto unfetch_interval;
2238 FREE_STACK_RETURN (REG_BADBR);
2241 /* We just parsed a valid interval. */
2243 /* If it's invalid to have no preceding re. */
2246 if (syntax & RE_CONTEXT_INVALID_OPS)
2247 FREE_STACK_RETURN (REG_BADRPT);
2248 else if (syntax & RE_CONTEXT_INDEP_OPS)
2251 goto unfetch_interval;
2254 /* If the upper bound is zero, don't want to succeed at
2255 all; jump from `laststart' to `b + 3', which will be
2256 the end of the buffer after we insert the jump. */
2257 if (upper_bound == 0)
2259 GET_BUFFER_SPACE (3);
2260 INSERT_JUMP (jump, laststart, b + 3);
2264 /* Otherwise, we have a nontrivial interval. When
2265 we're all done, the pattern will look like:
2266 set_number_at <jump count> <upper bound>
2267 set_number_at <succeed_n count> <lower bound>
2268 succeed_n <after jump addr> <succeed_n count>
2270 jump_n <succeed_n addr> <jump count>
2271 (The upper bound and `jump_n' are omitted if
2272 `upper_bound' is 1, though.) */
2274 { /* If the upper bound is > 1, we need to insert
2275 more at the end of the loop. */
2276 unsigned nbytes = 10 + (upper_bound > 1) * 10;
2278 GET_BUFFER_SPACE (nbytes);
2280 /* Initialize lower bound of the `succeed_n', even
2281 though it will be set during matching by its
2282 attendant `set_number_at' (inserted next),
2283 because `re_compile_fastmap' needs to know.
2284 Jump to the `jump_n' we might insert below. */
2285 INSERT_JUMP2 (succeed_n, laststart,
2286 b + 5 + (upper_bound > 1) * 5,
2290 /* Code to initialize the lower bound. Insert
2291 before the `succeed_n'. The `5' is the last two
2292 bytes of this `set_number_at', plus 3 bytes of
2293 the following `succeed_n'. */
2294 insert_op2 (set_number_at, laststart, 5, lower_bound, b);
2297 if (upper_bound > 1)
2298 { /* More than one repetition is allowed, so
2299 append a backward jump to the `succeed_n'
2300 that starts this interval.
2302 When we've reached this during matching,
2303 we'll have matched the interval once, so
2304 jump back only `upper_bound - 1' times. */
2305 STORE_JUMP2 (jump_n, b, laststart + 5,
2309 /* The location we want to set is the second
2310 parameter of the `jump_n'; that is `b-2' as
2311 an absolute address. `laststart' will be
2312 the `set_number_at' we're about to insert;
2313 `laststart+3' the number to set, the source
2314 for the relative address. But we are
2315 inserting into the middle of the pattern --
2316 so everything is getting moved up by 5.
2317 Conclusion: (b - 2) - (laststart + 3) + 5,
2318 i.e., b - laststart.
2320 We insert this at the beginning of the loop
2321 so that if we fail during matching, we'll
2322 reinitialize the bounds. */
2323 insert_op2 (set_number_at, laststart, b - laststart,
2324 upper_bound - 1, b);
2329 beg_interval = NULL;
2334 /* If an invalid interval, match the characters as literals. */
2335 assert (beg_interval);
2337 beg_interval = NULL;
2339 /* normal_char and normal_backslash need `c'. */
2342 if (!(syntax & RE_NO_BK_BRACES))
2344 if (p > pattern && p[-1] == '\\')
2345 goto normal_backslash;
2350 /* There is no way to specify the before_dot and after_dot
2351 operators. rms says this is ok. --karl */
2359 BUF_PUSH_2 (syntaxspec, syntax_spec_code[c]);
2365 BUF_PUSH_2 (notsyntaxspec, syntax_spec_code[c]);
2372 BUF_PUSH (wordchar);
2378 BUF_PUSH (notwordchar);
2391 BUF_PUSH (wordbound);
2395 BUF_PUSH (notwordbound);
2406 case '1': case '2': case '3': case '4': case '5':
2407 case '6': case '7': case '8': case '9':
2408 if (syntax & RE_NO_BK_REFS)
2414 FREE_STACK_RETURN (REG_ESUBREG);
2416 /* Can't back reference to a subexpression if inside of it. */
2417 if (group_in_compile_stack (compile_stack, c1))
2421 BUF_PUSH_2 (duplicate, c1);
2427 if (syntax & RE_BK_PLUS_QM)
2430 goto normal_backslash;
2434 /* You might think it would be useful for \ to mean
2435 not to translate; but if we don't translate it
2436 it will never match anything. */
2444 /* Expects the character in `c'. */
2446 /* If no exactn currently being built. */
2449 /* If last exactn not at current position. */
2450 || pending_exact + *pending_exact + 1 != b
2452 /* We have only one byte following the exactn for the count. */
2453 || *pending_exact == (1 << BYTEWIDTH) - 1
2455 /* If followed by a repetition operator. */
2456 || *p == '*' || *p == '^'
2457 || ((syntax & RE_BK_PLUS_QM)
2458 ? *p == '\\' && (p[1] == '+' || p[1] == '?')
2459 : (*p == '+' || *p == '?'))
2460 || ((syntax & RE_INTERVALS)
2461 && ((syntax & RE_NO_BK_BRACES)
2463 : (p[0] == '\\' && p[1] == '{'))))
2465 /* Start building a new exactn. */
2469 BUF_PUSH_2 (exactn, 0);
2470 pending_exact = b - 1;
2477 } /* while p != pend */
2480 /* Through the pattern now. */
2483 STORE_JUMP (jump_past_alt, fixup_alt_jump, b);
2485 if (!COMPILE_STACK_EMPTY)
2486 FREE_STACK_RETURN (REG_EPAREN);
2488 /* If we don't want backtracking, force success
2489 the first time we reach the end of the compiled pattern. */
2490 if (syntax & RE_NO_POSIX_BACKTRACKING)
2493 free (compile_stack.stack);
2495 /* We have succeeded; set the length of the buffer. */
2496 bufp->used = b - bufp->buffer;
2501 DEBUG_PRINT1 ("\nCompiled pattern: \n");
2502 print_compiled_pattern (bufp);
2506 #ifndef MATCH_MAY_ALLOCATE
2507 /* Initialize the failure stack to the largest possible stack. This
2508 isn't necessary unless we're trying to avoid calling alloca in
2509 the search and match routines. */
2511 int num_regs = bufp->re_nsub + 1;
2513 /* Since DOUBLE_FAIL_STACK refuses to double only if the current size
2514 is strictly greater than re_max_failures, the largest possible stack
2515 is 2 * re_max_failures failure points. */
2516 if (fail_stack.size < (2 * re_max_failures * MAX_FAILURE_ITEMS))
2518 fail_stack.size = (2 * re_max_failures * MAX_FAILURE_ITEMS);
2521 if (! fail_stack.stack)
2523 = (fail_stack_elt_t *) xmalloc (fail_stack.size
2524 * sizeof (fail_stack_elt_t));
2527 = (fail_stack_elt_t *) xrealloc (fail_stack.stack,
2529 * sizeof (fail_stack_elt_t)));
2530 #else /* not emacs */
2531 if (! fail_stack.stack)
2533 = (fail_stack_elt_t *) malloc (fail_stack.size
2534 * sizeof (fail_stack_elt_t));
2537 = (fail_stack_elt_t *) realloc (fail_stack.stack,
2539 * sizeof (fail_stack_elt_t)));
2540 #endif /* not emacs */
2543 /* Initialize some other variables the matcher uses. */
2544 RETALLOC_IF (regstart, num_regs, const char *);
2545 RETALLOC_IF (regend, num_regs, const char *);
2546 RETALLOC_IF (old_regstart, num_regs, const char *);
2547 RETALLOC_IF (old_regend, num_regs, const char *);
2548 RETALLOC_IF (best_regstart, num_regs, const char *);
2549 RETALLOC_IF (best_regend, num_regs, const char *);
2550 RETALLOC_IF (reg_info, num_regs, register_info_type);
2551 RETALLOC_IF (reg_dummy, num_regs, const char *);
2552 RETALLOC_IF (reg_info_dummy, num_regs, register_info_type);
2557 } /* regex_compile */
2559 /* Subroutines for `regex_compile'. */
2561 /* Store OP at LOC followed by two-byte integer parameter ARG. */
2564 store_op1 (op, loc, arg)
2569 *loc = (unsigned char) op;
2570 STORE_NUMBER (loc + 1, arg);
2574 /* Like `store_op1', but for two two-byte parameters ARG1 and ARG2. */
2577 store_op2 (op, loc, arg1, arg2)
2582 *loc = (unsigned char) op;
2583 STORE_NUMBER (loc + 1, arg1);
2584 STORE_NUMBER (loc + 3, arg2);
2588 /* Copy the bytes from LOC to END to open up three bytes of space at LOC
2589 for OP followed by two-byte integer parameter ARG. */
2592 insert_op1 (op, loc, arg, end)
2598 register unsigned char *pfrom = end;
2599 register unsigned char *pto = end + 3;
2601 while (pfrom != loc)
2604 store_op1 (op, loc, arg);
2608 /* Like `insert_op1', but for two two-byte parameters ARG1 and ARG2. */
2611 insert_op2 (op, loc, arg1, arg2, end)
2617 register unsigned char *pfrom = end;
2618 register unsigned char *pto = end + 5;
2620 while (pfrom != loc)
2623 store_op2 (op, loc, arg1, arg2);
2627 /* P points to just after a ^ in PATTERN. Return true if that ^ comes
2628 after an alternative or a begin-subexpression. We assume there is at
2629 least one character before the ^. */
2632 at_begline_loc_p (pattern, p, syntax)
2633 const char *pattern, *p;
2634 reg_syntax_t syntax;
2636 const char *prev = p - 2;
2637 boolean prev_prev_backslash = prev > pattern && prev[-1] == '\\';
2640 /* After a subexpression? */
2641 (*prev == '(' && (syntax & RE_NO_BK_PARENS || prev_prev_backslash))
2642 /* After an alternative? */
2643 || (*prev == '|' && (syntax & RE_NO_BK_VBAR || prev_prev_backslash));
2647 /* The dual of at_begline_loc_p. This one is for $. We assume there is
2648 at least one character after the $, i.e., `P < PEND'. */
2651 at_endline_loc_p (p, pend, syntax)
2652 const char *p, *pend;
2655 const char *next = p;
2656 boolean next_backslash = *next == '\\';
2657 const char *next_next = p + 1 < pend ? p + 1 : NULL;
2660 /* Before a subexpression? */
2661 (syntax & RE_NO_BK_PARENS ? *next == ')'
2662 : next_backslash && next_next && *next_next == ')')
2663 /* Before an alternative? */
2664 || (syntax & RE_NO_BK_VBAR ? *next == '|'
2665 : next_backslash && next_next && *next_next == '|');
2669 /* Returns true if REGNUM is in one of COMPILE_STACK's elements and
2670 false if it's not. */
2673 group_in_compile_stack (compile_stack, regnum)
2674 compile_stack_type compile_stack;
2679 for (this_element = compile_stack.avail - 1;
2682 if (compile_stack.stack[this_element].regnum == regnum)
2689 /* Read the ending character of a range (in a bracket expression) from the
2690 uncompiled pattern *P_PTR (which ends at PEND). We assume the
2691 starting character is in `P[-2]'. (`P[-1]' is the character `-'.)
2692 Then we set the translation of all bits between the starting and
2693 ending characters (inclusive) in the compiled pattern B.
2695 Return an error code.
2697 We use these short variable names so we can use the same macros as
2698 `regex_compile' itself. */
2700 static reg_errcode_t
2701 compile_range (p_ptr, pend, translate, syntax, b)
2702 const char **p_ptr, *pend;
2704 reg_syntax_t syntax;
2709 const char *p = *p_ptr;
2710 int range_start, range_end;
2715 /* Even though the pattern is a signed `char *', we need to fetch
2716 with unsigned char *'s; if the high bit of the pattern character
2717 is set, the range endpoints will be negative if we fetch using a
2720 We also want to fetch the endpoints without translating them; the
2721 appropriate translation is done in the bit-setting loop below. */
2722 /* The SVR4 compiler on the 3B2 had trouble with unsigned const char *. */
2723 range_start = ((const unsigned char *) p)[-2];
2724 range_end = ((const unsigned char *) p)[0];
2726 /* Have to increment the pointer into the pattern string, so the
2727 caller isn't still at the ending character. */
2730 /* If the start is after the end, the range is empty. */
2731 if (range_start > range_end)
2732 return syntax & RE_NO_EMPTY_RANGES ? REG_ERANGE : REG_NOERROR;
2734 /* Here we see why `this_char' has to be larger than an `unsigned
2735 char' -- the range is inclusive, so if `range_end' == 0xff
2736 (assuming 8-bit characters), we would otherwise go into an infinite
2737 loop, since all characters <= 0xff. */
2738 for (this_char = range_start; this_char <= range_end; this_char++)
2740 SET_LIST_BIT (TRANSLATE (this_char));
2746 /* re_compile_fastmap computes a ``fastmap'' for the compiled pattern in
2747 BUFP. A fastmap records which of the (1 << BYTEWIDTH) possible
2748 characters can start a string that matches the pattern. This fastmap
2749 is used by re_search to skip quickly over impossible starting points.
2751 The caller must supply the address of a (1 << BYTEWIDTH)-byte data
2752 area as BUFP->fastmap.
2754 We set the `fastmap', `fastmap_accurate', and `can_be_null' fields in
2757 Returns 0 if we succeed, -2 if an internal error. */
2760 re_compile_fastmap (bufp)
2761 struct re_pattern_buffer *bufp;
2764 #ifdef MATCH_MAY_ALLOCATE
2765 fail_stack_type fail_stack;
2767 #ifndef REGEX_MALLOC
2770 /* We don't push any register information onto the failure stack. */
2771 unsigned num_regs = 0;
2773 register char *fastmap = bufp->fastmap;
2774 unsigned char *pattern = bufp->buffer;
2775 unsigned long size = bufp->used;
2776 unsigned char *p = pattern;
2777 register unsigned char *pend = pattern + size;
2779 /* Assume that each path through the pattern can be null until
2780 proven otherwise. We set this false at the bottom of switch
2781 statement, to which we get only if a particular path doesn't
2782 match the empty string. */
2783 boolean path_can_be_null = true;
2785 /* We aren't doing a `succeed_n' to begin with. */
2786 boolean succeed_n_p = false;
2788 assert (fastmap != NULL && p != NULL);
2791 bzero (fastmap, 1 << BYTEWIDTH); /* Assume nothing's valid. */
2792 bufp->fastmap_accurate = 1; /* It will be when we're done. */
2793 bufp->can_be_null = 0;
2797 if (p == pend || *p == succeed)
2799 /* We have reached the (effective) end of pattern. */
2800 if (!FAIL_STACK_EMPTY ())
2802 bufp->can_be_null |= path_can_be_null;
2804 /* Reset for next path. */
2805 path_can_be_null = true;
2807 p = fail_stack.stack[--fail_stack.avail];
2815 /* We should never be about to go beyond the end of the pattern. */
2818 #ifdef SWITCH_ENUM_BUG
2819 switch ((int) ((re_opcode_t) *p++))
2821 switch ((re_opcode_t) *p++)
2825 /* I guess the idea here is to simply not bother with a fastmap
2826 if a backreference is used, since it's too hard to figure out
2827 the fastmap for the corresponding group. Setting
2828 `can_be_null' stops `re_search_2' from using the fastmap, so
2829 that is all we do. */
2831 bufp->can_be_null = 1;
2835 /* Following are the cases which match a character. These end
2844 for (j = *p++ * BYTEWIDTH - 1; j >= 0; j--)
2845 if (p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH)))
2851 /* Chars beyond end of map must be allowed. */
2852 for (j = *p * BYTEWIDTH; j < (1 << BYTEWIDTH); j++)
2855 for (j = *p++ * BYTEWIDTH - 1; j >= 0; j--)
2856 if (!(p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH))))
2862 for (j = 0; j < (1 << BYTEWIDTH); j++)
2863 if (SYNTAX (j) == Sword)
2869 for (j = 0; j < (1 << BYTEWIDTH); j++)
2870 if (SYNTAX (j) != Sword)
2877 int fastmap_newline = fastmap['\n'];
2879 /* `.' matches anything ... */
2880 for (j = 0; j < (1 << BYTEWIDTH); j++)
2883 /* ... except perhaps newline. */
2884 if (!(bufp->syntax & RE_DOT_NEWLINE))
2885 fastmap['\n'] = fastmap_newline;
2887 /* Return if we have already set `can_be_null'; if we have,
2888 then the fastmap is irrelevant. Something's wrong here. */
2889 else if (bufp->can_be_null)
2892 /* Otherwise, have to check alternative paths. */
2899 for (j = 0; j < (1 << BYTEWIDTH); j++)
2900 if (SYNTAX (j) == (enum syntaxcode) k)
2907 for (j = 0; j < (1 << BYTEWIDTH); j++)
2908 if (SYNTAX (j) != (enum syntaxcode) k)
2913 /* All cases after this match the empty string. These end with
2921 #endif /* not emacs */
2933 case push_dummy_failure:
2938 case pop_failure_jump:
2939 case maybe_pop_jump:
2942 case dummy_failure_jump:
2943 EXTRACT_NUMBER_AND_INCR (j, p);
2948 /* Jump backward implies we just went through the body of a
2949 loop and matched nothing. Opcode jumped to should be
2950 `on_failure_jump' or `succeed_n'. Just treat it like an
2951 ordinary jump. For a * loop, it has pushed its failure
2952 point already; if so, discard that as redundant. */
2953 if ((re_opcode_t) *p != on_failure_jump
2954 && (re_opcode_t) *p != succeed_n)
2958 EXTRACT_NUMBER_AND_INCR (j, p);
2961 /* If what's on the stack is where we are now, pop it. */
2962 if (!FAIL_STACK_EMPTY ()
2963 && fail_stack.stack[fail_stack.avail - 1] == p)
2969 case on_failure_jump:
2970 case on_failure_keep_string_jump:
2971 handle_on_failure_jump:
2972 EXTRACT_NUMBER_AND_INCR (j, p);
2974 /* For some patterns, e.g., `(a?)?', `p+j' here points to the
2975 end of the pattern. We don't want to push such a point,
2976 since when we restore it above, entering the switch will
2977 increment `p' past the end of the pattern. We don't need
2978 to push such a point since we obviously won't find any more
2979 fastmap entries beyond `pend'. Such a pattern can match
2980 the null string, though. */
2983 if (!PUSH_PATTERN_OP (p + j, fail_stack))
2987 bufp->can_be_null = 1;
2991 EXTRACT_NUMBER_AND_INCR (k, p); /* Skip the n. */
2992 succeed_n_p = false;
2999 /* Get to the number of times to succeed. */
3002 /* Increment p past the n for when k != 0. */
3003 EXTRACT_NUMBER_AND_INCR (k, p);
3007 succeed_n_p = true; /* Spaghetti code alert. */
3008 goto handle_on_failure_jump;
3025 abort (); /* We have listed all the cases. */
3028 /* Getting here means we have found the possible starting
3029 characters for one path of the pattern -- and that the empty
3030 string does not match. We need not follow this path further.
3031 Instead, look at the next alternative (remembered on the
3032 stack), or quit if no more. The test at the top of the loop
3033 does these things. */
3034 path_can_be_null = false;
3038 /* Set `can_be_null' for the last path (also the first path, if the
3039 pattern is empty). */
3040 bufp->can_be_null |= path_can_be_null;
3042 } /* re_compile_fastmap */
3044 /* Set REGS to hold NUM_REGS registers, storing them in STARTS and
3045 ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use
3046 this memory for recording register information. STARTS and ENDS
3047 must be allocated using the malloc library routine, and must each
3048 be at least NUM_REGS * sizeof (regoff_t) bytes long.
3050 If NUM_REGS == 0, then subsequent matches should allocate their own
3053 Unless this function is called, the first search or match using
3054 PATTERN_BUFFER will allocate its own register data, without
3055 freeing the old data. */
3058 re_set_registers (bufp, regs, num_regs, starts, ends)
3059 struct re_pattern_buffer *bufp;
3060 struct re_registers *regs;
3062 regoff_t *starts, *ends;
3066 bufp->regs_allocated = REGS_REALLOCATE;
3067 regs->num_regs = num_regs;
3068 regs->start = starts;
3073 bufp->regs_allocated = REGS_UNALLOCATED;
3075 regs->start = regs->end = (regoff_t *) 0;
3079 /* Searching routines. */
3081 /* Like re_search_2, below, but only one string is specified, and
3082 doesn't let you say where to stop matching. */
3085 re_search (bufp, string, size, startpos, range, regs)
3086 struct re_pattern_buffer *bufp;
3088 int size, startpos, range;
3089 struct re_registers *regs;
3091 return re_search_2 (bufp, NULL, 0, string, size, startpos, range,
3096 /* Using the compiled pattern in BUFP->buffer, first tries to match the
3097 virtual concatenation of STRING1 and STRING2, starting first at index
3098 STARTPOS, then at STARTPOS + 1, and so on.
3100 STRING1 and STRING2 have length SIZE1 and SIZE2, respectively.
3102 RANGE is how far to scan while trying to match. RANGE = 0 means try
3103 only at STARTPOS; in general, the last start tried is STARTPOS +
3106 In REGS, return the indices of the virtual concatenation of STRING1
3107 and STRING2 that matched the entire BUFP->buffer and its contained
3110 Do not consider matching one past the index STOP in the virtual
3111 concatenation of STRING1 and STRING2.
3113 We return either the position in the strings at which the match was
3114 found, -1 if no match, or -2 if error (such as failure
3118 re_search_2 (bufp, string1, size1, string2, size2, startpos, range, regs, stop)
3119 struct re_pattern_buffer *bufp;
3120 const char *string1, *string2;
3124 struct re_registers *regs;
3128 register char *fastmap = bufp->fastmap;
3129 register char *translate = bufp->translate;
3130 int total_size = size1 + size2;
3131 int endpos = startpos + range;
3133 /* Check for out-of-range STARTPOS. */
3134 if (startpos < 0 || startpos > total_size)
3137 /* Fix up RANGE if it might eventually take us outside
3138 the virtual concatenation of STRING1 and STRING2. */
3140 range = -1 - startpos;
3141 else if (endpos > total_size)
3142 range = total_size - startpos;
3144 /* If the search isn't to be a backwards one, don't waste time in a
3145 search for a pattern that must be anchored. */
3146 if (bufp->used > 0 && (re_opcode_t) bufp->buffer[0] == begbuf && range > 0)
3154 /* Update the fastmap now if not correct already. */
3155 if (fastmap && !bufp->fastmap_accurate)
3156 if (re_compile_fastmap (bufp) == -2)
3159 /* Loop through the string, looking for a place to start matching. */
3162 /* If a fastmap is supplied, skip quickly over characters that
3163 cannot be the start of a match. If the pattern can match the
3164 null string, however, we don't need to skip characters; we want
3165 the first null string. */
3166 if (fastmap && startpos < total_size && !bufp->can_be_null)
3168 if (range > 0) /* Searching forwards. */
3170 register const char *d;
3171 register int lim = 0;
3174 if (startpos < size1 && startpos + range >= size1)
3175 lim = range - (size1 - startpos);
3177 d = (startpos >= size1 ? string2 - size1 : string1) + startpos;
3179 /* Written out as an if-else to avoid testing `translate'
3183 && !fastmap[(unsigned char)
3184 translate[(unsigned char) *d++]])
3187 while (range > lim && !fastmap[(unsigned char) *d++])
3190 startpos += irange - range;
3192 else /* Searching backwards. */
3194 register char c = (size1 == 0 || startpos >= size1
3195 ? string2[startpos - size1]
3196 : string1[startpos]);
3198 if (!fastmap[(unsigned char) TRANSLATE (c)])
3203 /* If can't match the null string, and that's all we have left, fail. */
3204 if (range >= 0 && startpos == total_size && fastmap
3205 && !bufp->can_be_null)
3208 val = re_match_2_internal (bufp, string1, size1, string2, size2,
3209 startpos, regs, stop);
3210 #ifndef REGEX_MALLOC
3239 /* Declarations and macros for re_match_2. */
3241 static int bcmp_translate ();
3242 static boolean alt_match_null_string_p (),
3243 common_op_match_null_string_p (),
3244 group_match_null_string_p ();
3246 /* This converts PTR, a pointer into one of the search strings `string1'
3247 and `string2' into an offset from the beginning of that string. */
3248 #define POINTER_TO_OFFSET(ptr) \
3249 (FIRST_STRING_P (ptr) \
3250 ? ((regoff_t) ((ptr) - string1)) \
3251 : ((regoff_t) ((ptr) - string2 + size1)))
3253 /* Macros for dealing with the split strings in re_match_2. */
3255 #define MATCHING_IN_FIRST_STRING (dend == end_match_1)
3257 /* Call before fetching a character with *d. This switches over to
3258 string2 if necessary. */
3259 #define PREFETCH() \
3262 /* End of string2 => fail. */ \
3263 if (dend == end_match_2) \
3265 /* End of string1 => advance to string2. */ \
3267 dend = end_match_2; \
3271 /* Test if at very beginning or at very end of the virtual concatenation
3272 of `string1' and `string2'. If only one string, it's `string2'. */
3273 #define AT_STRINGS_BEG(d) ((d) == (size1 ? string1 : string2) || !size2)
3274 #define AT_STRINGS_END(d) ((d) == end2)
3277 /* Test if D points to a character which is word-constituent. We have
3278 two special cases to check for: if past the end of string1, look at
3279 the first character in string2; and if before the beginning of
3280 string2, look at the last character in string1. */
3281 #define WORDCHAR_P(d) \
3282 (SYNTAX ((d) == end1 ? *string2 \
3283 : (d) == string2 - 1 ? *(end1 - 1) : *(d)) \
3286 /* Test if the character before D and the one at D differ with respect
3287 to being word-constituent. */
3288 #define AT_WORD_BOUNDARY(d) \
3289 (AT_STRINGS_BEG (d) || AT_STRINGS_END (d) \
3290 || WORDCHAR_P (d - 1) != WORDCHAR_P (d))
3293 /* Free everything we malloc. */
3294 #ifdef MATCH_MAY_ALLOCATE
3296 #define FREE_VAR(var) if (var) free (var); var = NULL
3297 #define FREE_VARIABLES() \
3299 FREE_VAR (fail_stack.stack); \
3300 FREE_VAR (regstart); \
3301 FREE_VAR (regend); \
3302 FREE_VAR (old_regstart); \
3303 FREE_VAR (old_regend); \
3304 FREE_VAR (best_regstart); \
3305 FREE_VAR (best_regend); \
3306 FREE_VAR (reg_info); \
3307 FREE_VAR (reg_dummy); \
3308 FREE_VAR (reg_info_dummy); \
3310 #else /* not REGEX_MALLOC */
3311 /* This used to do alloca (0), but now we do that in the caller. */
3312 #define FREE_VARIABLES() /* Nothing */
3313 #endif /* not REGEX_MALLOC */
3315 #define FREE_VARIABLES() /* Do nothing! */
3316 #endif /* not MATCH_MAY_ALLOCATE */
3318 /* These values must meet several constraints. They must not be valid
3319 register values; since we have a limit of 255 registers (because
3320 we use only one byte in the pattern for the register number), we can
3321 use numbers larger than 255. They must differ by 1, because of
3322 NUM_FAILURE_ITEMS above. And the value for the lowest register must
3323 be larger than the value for the highest register, so we do not try
3324 to actually save any registers when none are active. */
3325 #define NO_HIGHEST_ACTIVE_REG (1 << BYTEWIDTH)
3326 #define NO_LOWEST_ACTIVE_REG (NO_HIGHEST_ACTIVE_REG + 1)
3328 /* Matching routines. */
3330 #ifndef emacs /* Emacs never uses this. */
3331 /* re_match is like re_match_2 except it takes only a single string. */
3334 re_match (bufp, string, size, pos, regs)
3335 struct re_pattern_buffer *bufp;
3338 struct re_registers *regs;
3340 int result = re_match_2_internal (bufp, NULL, 0, string, size,
3345 #endif /* not emacs */
3348 /* re_match_2 matches the compiled pattern in BUFP against the
3349 the (virtual) concatenation of STRING1 and STRING2 (of length SIZE1
3350 and SIZE2, respectively). We start matching at POS, and stop
3353 If REGS is non-null and the `no_sub' field of BUFP is nonzero, we
3354 store offsets for the substring each group matched in REGS. See the
3355 documentation for exactly how many groups we fill.
3357 We return -1 if no match, -2 if an internal error (such as the
3358 failure stack overflowing). Otherwise, we return the length of the
3359 matched substring. */
3362 re_match_2 (bufp, string1, size1, string2, size2, pos, regs, stop)
3363 struct re_pattern_buffer *bufp;
3364 const char *string1, *string2;
3367 struct re_registers *regs;
3370 int result = re_match_2_internal (bufp, string1, size1, string2, size2,
3376 /* This is a separate function so that we can force an alloca cleanup
3379 re_match_2_internal (bufp, string1, size1, string2, size2, pos, regs, stop)
3380 struct re_pattern_buffer *bufp;
3381 const char *string1, *string2;
3384 struct re_registers *regs;
3387 /* General temporaries. */
3391 /* Just past the end of the corresponding string. */
3392 const char *end1, *end2;
3394 /* Pointers into string1 and string2, just past the last characters in
3395 each to consider matching. */
3396 const char *end_match_1, *end_match_2;
3398 /* Where we are in the data, and the end of the current string. */
3399 const char *d, *dend;
3401 /* Where we are in the pattern, and the end of the pattern. */
3402 unsigned char *p = bufp->buffer;
3403 register unsigned char *pend = p + bufp->used;
3405 /* Mark the opcode just after a start_memory, so we can test for an
3406 empty subpattern when we get to the stop_memory. */
3407 unsigned char *just_past_start_mem = 0;
3409 /* We use this to map every character in the string. */
3410 char *translate = bufp->translate;
3412 /* Failure point stack. Each place that can handle a failure further
3413 down the line pushes a failure point on this stack. It consists of
3414 restart, regend, and reg_info for all registers corresponding to
3415 the subexpressions we're currently inside, plus the number of such
3416 registers, and, finally, two char *'s. The first char * is where
3417 to resume scanning the pattern; the second one is where to resume
3418 scanning the strings. If the latter is zero, the failure point is
3419 a ``dummy''; if a failure happens and the failure point is a dummy,
3420 it gets discarded and the next next one is tried. */
3421 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
3422 fail_stack_type fail_stack;
3425 static unsigned failure_id = 0;
3426 unsigned nfailure_points_pushed = 0, nfailure_points_popped = 0;
3429 /* We fill all the registers internally, independent of what we
3430 return, for use in backreferences. The number here includes
3431 an element for register zero. */
3432 unsigned num_regs = bufp->re_nsub + 1;
3434 /* The currently active registers. */
3435 unsigned lowest_active_reg = NO_LOWEST_ACTIVE_REG;
3436 unsigned highest_active_reg = NO_HIGHEST_ACTIVE_REG;
3438 /* Information on the contents of registers. These are pointers into
3439 the input strings; they record just what was matched (on this
3440 attempt) by a subexpression part of the pattern, that is, the
3441 regnum-th regstart pointer points to where in the pattern we began
3442 matching and the regnum-th regend points to right after where we
3443 stopped matching the regnum-th subexpression. (The zeroth register
3444 keeps track of what the whole pattern matches.) */
3445 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
3446 const char **regstart, **regend;
3449 /* If a group that's operated upon by a repetition operator fails to
3450 match anything, then the register for its start will need to be
3451 restored because it will have been set to wherever in the string we
3452 are when we last see its open-group operator. Similarly for a
3454 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
3455 const char **old_regstart, **old_regend;
3458 /* The is_active field of reg_info helps us keep track of which (possibly
3459 nested) subexpressions we are currently in. The matched_something
3460 field of reg_info[reg_num] helps us tell whether or not we have
3461 matched any of the pattern so far this time through the reg_num-th
3462 subexpression. These two fields get reset each time through any
3463 loop their register is in. */
3464 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
3465 register_info_type *reg_info;
3468 /* The following record the register info as found in the above
3469 variables when we find a match better than any we've seen before.
3470 This happens as we backtrack through the failure points, which in
3471 turn happens only if we have not yet matched the entire string. */
3472 unsigned best_regs_set = false;
3473 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
3474 const char **best_regstart, **best_regend;
3477 /* Logically, this is `best_regend[0]'. But we don't want to have to
3478 allocate space for that if we're not allocating space for anything
3479 else (see below). Also, we never need info about register 0 for
3480 any of the other register vectors, and it seems rather a kludge to
3481 treat `best_regend' differently than the rest. So we keep track of
3482 the end of the best match so far in a separate variable. We
3483 initialize this to NULL so that when we backtrack the first time
3484 and need to test it, it's not garbage. */
3485 const char *match_end = NULL;
3487 /* Used when we pop values we don't care about. */
3488 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
3489 const char **reg_dummy;
3490 register_info_type *reg_info_dummy;
3494 /* Counts the total number of registers pushed. */
3495 unsigned num_regs_pushed = 0;
3498 DEBUG_PRINT1 ("\n\nEntering re_match_2.\n");
3502 #ifdef MATCH_MAY_ALLOCATE
3503 /* Do not bother to initialize all the register variables if there are
3504 no groups in the pattern, as it takes a fair amount of time. If
3505 there are groups, we include space for register 0 (the whole
3506 pattern), even though we never use it, since it simplifies the
3507 array indexing. We should fix this. */
3510 regstart = REGEX_TALLOC (num_regs, const char *);
3511 regend = REGEX_TALLOC (num_regs, const char *);
3512 old_regstart = REGEX_TALLOC (num_regs, const char *);
3513 old_regend = REGEX_TALLOC (num_regs, const char *);
3514 best_regstart = REGEX_TALLOC (num_regs, const char *);
3515 best_regend = REGEX_TALLOC (num_regs, const char *);
3516 reg_info = REGEX_TALLOC (num_regs, register_info_type);
3517 reg_dummy = REGEX_TALLOC (num_regs, const char *);
3518 reg_info_dummy = REGEX_TALLOC (num_regs, register_info_type);
3520 if (!(regstart && regend && old_regstart && old_regend && reg_info
3521 && best_regstart && best_regend && reg_dummy && reg_info_dummy))
3527 #if defined (REGEX_MALLOC)
3530 /* We must initialize all our variables to NULL, so that
3531 `FREE_VARIABLES' doesn't try to free them. */
3532 regstart = regend = old_regstart = old_regend = best_regstart
3533 = best_regend = reg_dummy = NULL;
3534 reg_info = reg_info_dummy = (register_info_type *) NULL;
3536 #endif /* REGEX_MALLOC */
3537 #endif /* MATCH_MAY_ALLOCATE */
3539 /* The starting position is bogus. */
3540 if (pos < 0 || pos > size1 + size2)
3546 /* Initialize subexpression text positions to -1 to mark ones that no
3547 start_memory/stop_memory has been seen for. Also initialize the
3548 register information struct. */
3549 for (mcnt = 1; mcnt < num_regs; mcnt++)
3551 regstart[mcnt] = regend[mcnt]
3552 = old_regstart[mcnt] = old_regend[mcnt] = REG_UNSET_VALUE;
3554 REG_MATCH_NULL_STRING_P (reg_info[mcnt]) = MATCH_NULL_UNSET_VALUE;
3555 IS_ACTIVE (reg_info[mcnt]) = 0;
3556 MATCHED_SOMETHING (reg_info[mcnt]) = 0;
3557 EVER_MATCHED_SOMETHING (reg_info[mcnt]) = 0;
3560 /* We move `string1' into `string2' if the latter's empty -- but not if
3561 `string1' is null. */
3562 if (size2 == 0 && string1 != NULL)
3569 end1 = string1 + size1;
3570 end2 = string2 + size2;
3572 /* Compute where to stop matching, within the two strings. */
3575 end_match_1 = string1 + stop;
3576 end_match_2 = string2;
3581 end_match_2 = string2 + stop - size1;
3584 /* `p' scans through the pattern as `d' scans through the data.
3585 `dend' is the end of the input string that `d' points within. `d'
3586 is advanced into the following input string whenever necessary, but
3587 this happens before fetching; therefore, at the beginning of the
3588 loop, `d' can be pointing at the end of a string, but it cannot
3590 if (size1 > 0 && pos <= size1)
3597 d = string2 + pos - size1;
3601 DEBUG_PRINT1 ("The compiled pattern is: ");
3602 DEBUG_PRINT_COMPILED_PATTERN (bufp, p, pend);
3603 DEBUG_PRINT1 ("The string to match is: `");
3604 DEBUG_PRINT_DOUBLE_STRING (d, string1, size1, string2, size2);
3605 DEBUG_PRINT1 ("'\n");
3607 /* This loops over pattern commands. It exits by returning from the
3608 function if the match is complete, or it drops through if the match
3609 fails at this starting point in the input data. */
3612 DEBUG_PRINT2 ("\n0x%x: ", p);
3615 { /* End of pattern means we might have succeeded. */
3616 DEBUG_PRINT1 ("end of pattern ... ");
3618 /* If we haven't matched the entire string, and we want the
3619 longest match, try backtracking. */
3620 if (d != end_match_2)
3622 /* 1 if this match ends in the same string (string1 or string2)
3623 as the best previous match. */
3624 boolean same_str_p = (FIRST_STRING_P (match_end)
3625 == MATCHING_IN_FIRST_STRING);
3626 /* 1 if this match is the best seen so far. */
3627 boolean best_match_p;
3629 /* AIX compiler got confused when this was combined
3630 with the previous declaration. */
3632 best_match_p = d > match_end;
3634 best_match_p = !MATCHING_IN_FIRST_STRING;
3636 DEBUG_PRINT1 ("backtracking.\n");
3638 if (!FAIL_STACK_EMPTY ())
3639 { /* More failure points to try. */
3641 /* If exceeds best match so far, save it. */
3642 if (!best_regs_set || best_match_p)
3644 best_regs_set = true;
3647 DEBUG_PRINT1 ("\nSAVING match as best so far.\n");
3649 for (mcnt = 1; mcnt < num_regs; mcnt++)
3651 best_regstart[mcnt] = regstart[mcnt];
3652 best_regend[mcnt] = regend[mcnt];
3658 /* If no failure points, don't restore garbage. And if
3659 last match is real best match, don't restore second
3661 else if (best_regs_set && !best_match_p)
3664 /* Restore best match. It may happen that `dend ==
3665 end_match_1' while the restored d is in string2.
3666 For example, the pattern `x.*y.*z' against the
3667 strings `x-' and `y-z-', if the two strings are
3668 not consecutive in memory. */
3669 DEBUG_PRINT1 ("Restoring best registers.\n");
3672 dend = ((d >= string1 && d <= end1)
3673 ? end_match_1 : end_match_2);
3675 for (mcnt = 1; mcnt < num_regs; mcnt++)
3677 regstart[mcnt] = best_regstart[mcnt];
3678 regend[mcnt] = best_regend[mcnt];
3681 } /* d != end_match_2 */
3684 DEBUG_PRINT1 ("Accepting match.\n");
3686 /* If caller wants register contents data back, do it. */
3687 if (regs && !bufp->no_sub)
3689 /* Have the register data arrays been allocated? */
3690 if (bufp->regs_allocated == REGS_UNALLOCATED)
3691 { /* No. So allocate them with malloc. We need one
3692 extra element beyond `num_regs' for the `-1' marker
3694 regs->num_regs = MAX (RE_NREGS, num_regs + 1);
3695 regs->start = TALLOC (regs->num_regs, regoff_t);
3696 regs->end = TALLOC (regs->num_regs, regoff_t);
3697 if (regs->start == NULL || regs->end == NULL)
3699 bufp->regs_allocated = REGS_REALLOCATE;
3701 else if (bufp->regs_allocated == REGS_REALLOCATE)
3702 { /* Yes. If we need more elements than were already
3703 allocated, reallocate them. If we need fewer, just
3705 if (regs->num_regs < num_regs + 1)
3707 regs->num_regs = num_regs + 1;
3708 RETALLOC (regs->start, regs->num_regs, regoff_t);
3709 RETALLOC (regs->end, regs->num_regs, regoff_t);
3710 if (regs->start == NULL || regs->end == NULL)
3716 /* These braces fend off a "empty body in an else-statement"
3717 warning under GCC when assert expands to nothing. */
3718 assert (bufp->regs_allocated == REGS_FIXED);
3721 /* Convert the pointer data in `regstart' and `regend' to
3722 indices. Register zero has to be set differently,
3723 since we haven't kept track of any info for it. */
3724 if (regs->num_regs > 0)
3726 regs->start[0] = pos;
3727 regs->end[0] = (MATCHING_IN_FIRST_STRING
3728 ? ((regoff_t) (d - string1))
3729 : ((regoff_t) (d - string2 + size1)));
3732 /* Go through the first `min (num_regs, regs->num_regs)'
3733 registers, since that is all we initialized. */
3734 for (mcnt = 1; mcnt < MIN (num_regs, regs->num_regs); mcnt++)
3736 if (REG_UNSET (regstart[mcnt]) || REG_UNSET (regend[mcnt]))
3737 regs->start[mcnt] = regs->end[mcnt] = -1;
3741 = (regoff_t) POINTER_TO_OFFSET (regstart[mcnt]);
3743 = (regoff_t) POINTER_TO_OFFSET (regend[mcnt]);
3747 /* If the regs structure we return has more elements than
3748 were in the pattern, set the extra elements to -1. If
3749 we (re)allocated the registers, this is the case,
3750 because we always allocate enough to have at least one
3752 for (mcnt = num_regs; mcnt < regs->num_regs; mcnt++)
3753 regs->start[mcnt] = regs->end[mcnt] = -1;
3754 } /* regs && !bufp->no_sub */
3757 DEBUG_PRINT4 ("%u failure points pushed, %u popped (%u remain).\n",
3758 nfailure_points_pushed, nfailure_points_popped,
3759 nfailure_points_pushed - nfailure_points_popped);
3760 DEBUG_PRINT2 ("%u registers pushed.\n", num_regs_pushed);
3762 mcnt = d - pos - (MATCHING_IN_FIRST_STRING
3766 DEBUG_PRINT2 ("Returning %d from re_match_2.\n", mcnt);
3771 /* Otherwise match next pattern command. */
3772 #ifdef SWITCH_ENUM_BUG
3773 switch ((int) ((re_opcode_t) *p++))
3775 switch ((re_opcode_t) *p++)
3778 /* Ignore these. Used to ignore the n of succeed_n's which
3779 currently have n == 0. */
3781 DEBUG_PRINT1 ("EXECUTING no_op.\n");
3785 DEBUG_PRINT1 ("EXECUTING succeed.\n");
3788 /* Match the next n pattern characters exactly. The following
3789 byte in the pattern defines n, and the n bytes after that
3790 are the characters to match. */
3793 DEBUG_PRINT2 ("EXECUTING exactn %d.\n", mcnt);
3795 /* This is written out as an if-else so we don't waste time
3796 testing `translate' inside the loop. */
3802 if (translate[(unsigned char) *d++] != (char) *p++)
3812 if (*d++ != (char) *p++) goto fail;
3816 SET_REGS_MATCHED ();
3820 /* Match any character except possibly a newline or a null. */
3822 DEBUG_PRINT1 ("EXECUTING anychar.\n");
3826 if ((!(bufp->syntax & RE_DOT_NEWLINE) && TRANSLATE (*d) == '\n')
3827 || (bufp->syntax & RE_DOT_NOT_NULL && TRANSLATE (*d) == '\000'))
3830 SET_REGS_MATCHED ();
3831 DEBUG_PRINT2 (" Matched `%d'.\n", *d);
3839 register unsigned char c;
3840 boolean not = (re_opcode_t) *(p - 1) == charset_not;
3842 DEBUG_PRINT2 ("EXECUTING charset%s.\n", not ? "_not" : "");
3845 c = TRANSLATE (*d); /* The character to match. */
3847 /* Cast to `unsigned' instead of `unsigned char' in case the
3848 bit list is a full 32 bytes long. */
3849 if (c < (unsigned) (*p * BYTEWIDTH)
3850 && p[1 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH)))
3855 if (!not) goto fail;
3857 SET_REGS_MATCHED ();
3863 /* The beginning of a group is represented by start_memory.
3864 The arguments are the register number in the next byte, and the
3865 number of groups inner to this one in the next. The text
3866 matched within the group is recorded (in the internal
3867 registers data structure) under the register number. */
3869 DEBUG_PRINT3 ("EXECUTING start_memory %d (%d):\n", *p, p[1]);
3871 /* Find out if this group can match the empty string. */
3872 p1 = p; /* To send to group_match_null_string_p. */
3874 if (REG_MATCH_NULL_STRING_P (reg_info[*p]) == MATCH_NULL_UNSET_VALUE)
3875 REG_MATCH_NULL_STRING_P (reg_info[*p])
3876 = group_match_null_string_p (&p1, pend, reg_info);
3878 /* Save the position in the string where we were the last time
3879 we were at this open-group operator in case the group is
3880 operated upon by a repetition operator, e.g., with `(a*)*b'
3881 against `ab'; then we want to ignore where we are now in
3882 the string in case this attempt to match fails. */
3883 old_regstart[*p] = REG_MATCH_NULL_STRING_P (reg_info[*p])
3884 ? REG_UNSET (regstart[*p]) ? d : regstart[*p]
3886 DEBUG_PRINT2 (" old_regstart: %d\n",
3887 POINTER_TO_OFFSET (old_regstart[*p]));
3890 DEBUG_PRINT2 (" regstart: %d\n", POINTER_TO_OFFSET (regstart[*p]));
3892 IS_ACTIVE (reg_info[*p]) = 1;
3893 MATCHED_SOMETHING (reg_info[*p]) = 0;
3895 /* This is the new highest active register. */
3896 highest_active_reg = *p;
3898 /* If nothing was active before, this is the new lowest active
3900 if (lowest_active_reg == NO_LOWEST_ACTIVE_REG)
3901 lowest_active_reg = *p;
3903 /* Move past the register number and inner group count. */
3905 just_past_start_mem = p;
3909 /* The stop_memory opcode represents the end of a group. Its
3910 arguments are the same as start_memory's: the register
3911 number, and the number of inner groups. */
3913 DEBUG_PRINT3 ("EXECUTING stop_memory %d (%d):\n", *p, p[1]);
3915 /* We need to save the string position the last time we were at
3916 this close-group operator in case the group is operated
3917 upon by a repetition operator, e.g., with `((a*)*(b*)*)*'
3918 against `aba'; then we want to ignore where we are now in
3919 the string in case this attempt to match fails. */
3920 old_regend[*p] = REG_MATCH_NULL_STRING_P (reg_info[*p])
3921 ? REG_UNSET (regend[*p]) ? d : regend[*p]
3923 DEBUG_PRINT2 (" old_regend: %d\n",
3924 POINTER_TO_OFFSET (old_regend[*p]));
3927 DEBUG_PRINT2 (" regend: %d\n", POINTER_TO_OFFSET (regend[*p]));
3929 /* This register isn't active anymore. */
3930 IS_ACTIVE (reg_info[*p]) = 0;
3932 /* If this was the only register active, nothing is active
3934 if (lowest_active_reg == highest_active_reg)
3936 lowest_active_reg = NO_LOWEST_ACTIVE_REG;
3937 highest_active_reg = NO_HIGHEST_ACTIVE_REG;
3940 { /* We must scan for the new highest active register, since
3941 it isn't necessarily one less than now: consider
3942 (a(b)c(d(e)f)g). When group 3 ends, after the f), the
3943 new highest active register is 1. */
3944 unsigned char r = *p - 1;
3945 while (r > 0 && !IS_ACTIVE (reg_info[r]))
3948 /* If we end up at register zero, that means that we saved
3949 the registers as the result of an `on_failure_jump', not
3950 a `start_memory', and we jumped to past the innermost
3951 `stop_memory'. For example, in ((.)*) we save
3952 registers 1 and 2 as a result of the *, but when we pop
3953 back to the second ), we are at the stop_memory 1.
3954 Thus, nothing is active. */
3957 lowest_active_reg = NO_LOWEST_ACTIVE_REG;
3958 highest_active_reg = NO_HIGHEST_ACTIVE_REG;
3961 highest_active_reg = r;
3964 /* If just failed to match something this time around with a
3965 group that's operated on by a repetition operator, try to
3966 force exit from the ``loop'', and restore the register
3967 information for this group that we had before trying this
3969 if ((!MATCHED_SOMETHING (reg_info[*p])
3970 || just_past_start_mem == p - 1)
3973 boolean is_a_jump_n = false;
3977 switch ((re_opcode_t) *p1++)
3981 case pop_failure_jump:
3982 case maybe_pop_jump:
3984 case dummy_failure_jump:
3985 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
3995 /* If the next operation is a jump backwards in the pattern
3996 to an on_failure_jump right before the start_memory
3997 corresponding to this stop_memory, exit from the loop
3998 by forcing a failure after pushing on the stack the
3999 on_failure_jump's jump in the pattern, and d. */
4000 if (mcnt < 0 && (re_opcode_t) *p1 == on_failure_jump
4001 && (re_opcode_t) p1[3] == start_memory && p1[4] == *p)
4003 /* If this group ever matched anything, then restore
4004 what its registers were before trying this last
4005 failed match, e.g., with `(a*)*b' against `ab' for
4006 regstart[1], and, e.g., with `((a*)*(b*)*)*'
4007 against `aba' for regend[3].
4009 Also restore the registers for inner groups for,
4010 e.g., `((a*)(b*))*' against `aba' (register 3 would
4011 otherwise get trashed). */
4013 if (EVER_MATCHED_SOMETHING (reg_info[*p]))
4017 EVER_MATCHED_SOMETHING (reg_info[*p]) = 0;
4019 /* Restore this and inner groups' (if any) registers. */
4020 for (r = *p; r < *p + *(p + 1); r++)
4022 regstart[r] = old_regstart[r];
4024 /* xx why this test? */
4025 if ((int) old_regend[r] >= (int) regstart[r])
4026 regend[r] = old_regend[r];
4030 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4031 PUSH_FAILURE_POINT (p1 + mcnt, d, -2);
4037 /* Move past the register number and the inner group count. */
4042 /* \<digit> has been turned into a `duplicate' command which is
4043 followed by the numeric value of <digit> as the register number. */
4046 register const char *d2, *dend2;
4047 int regno = *p++; /* Get which register to match against. */
4048 DEBUG_PRINT2 ("EXECUTING duplicate %d.\n", regno);
4050 /* Can't back reference a group which we've never matched. */
4051 if (REG_UNSET (regstart[regno]) || REG_UNSET (regend[regno]))
4054 /* Where in input to try to start matching. */
4055 d2 = regstart[regno];
4057 /* Where to stop matching; if both the place to start and
4058 the place to stop matching are in the same string, then
4059 set to the place to stop, otherwise, for now have to use
4060 the end of the first string. */
4062 dend2 = ((FIRST_STRING_P (regstart[regno])
4063 == FIRST_STRING_P (regend[regno]))
4064 ? regend[regno] : end_match_1);
4067 /* If necessary, advance to next segment in register
4071 if (dend2 == end_match_2) break;
4072 if (dend2 == regend[regno]) break;
4074 /* End of string1 => advance to string2. */
4076 dend2 = regend[regno];
4078 /* At end of register contents => success */
4079 if (d2 == dend2) break;
4081 /* If necessary, advance to next segment in data. */
4084 /* How many characters left in this segment to match. */
4087 /* Want how many consecutive characters we can match in
4088 one shot, so, if necessary, adjust the count. */
4089 if (mcnt > dend2 - d2)
4092 /* Compare that many; failure if mismatch, else move
4095 ? bcmp_translate (d, d2, mcnt, translate)
4096 : bcmp (d, d2, mcnt))
4098 d += mcnt, d2 += mcnt;
4104 /* begline matches the empty string at the beginning of the string
4105 (unless `not_bol' is set in `bufp'), and, if
4106 `newline_anchor' is set, after newlines. */
4108 DEBUG_PRINT1 ("EXECUTING begline.\n");
4110 if (AT_STRINGS_BEG (d))
4112 if (!bufp->not_bol) break;
4114 else if (d[-1] == '\n' && bufp->newline_anchor)
4118 /* In all other cases, we fail. */
4122 /* endline is the dual of begline. */
4124 DEBUG_PRINT1 ("EXECUTING endline.\n");
4126 if (AT_STRINGS_END (d))
4128 if (!bufp->not_eol) break;
4131 /* We have to ``prefetch'' the next character. */
4132 else if ((d == end1 ? *string2 : *d) == '\n'
4133 && bufp->newline_anchor)
4140 /* Match at the very beginning of the data. */
4142 DEBUG_PRINT1 ("EXECUTING begbuf.\n");
4143 if (AT_STRINGS_BEG (d))
4148 /* Match at the very end of the data. */
4150 DEBUG_PRINT1 ("EXECUTING endbuf.\n");
4151 if (AT_STRINGS_END (d))
4156 /* on_failure_keep_string_jump is used to optimize `.*\n'. It
4157 pushes NULL as the value for the string on the stack. Then
4158 `pop_failure_point' will keep the current value for the
4159 string, instead of restoring it. To see why, consider
4160 matching `foo\nbar' against `.*\n'. The .* matches the foo;
4161 then the . fails against the \n. But the next thing we want
4162 to do is match the \n against the \n; if we restored the
4163 string value, we would be back at the foo.
4165 Because this is used only in specific cases, we don't need to
4166 check all the things that `on_failure_jump' does, to make
4167 sure the right things get saved on the stack. Hence we don't
4168 share its code. The only reason to push anything on the
4169 stack at all is that otherwise we would have to change
4170 `anychar's code to do something besides goto fail in this
4171 case; that seems worse than this. */
4172 case on_failure_keep_string_jump:
4173 DEBUG_PRINT1 ("EXECUTING on_failure_keep_string_jump");
4175 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4176 DEBUG_PRINT3 (" %d (to 0x%x):\n", mcnt, p + mcnt);
4178 PUSH_FAILURE_POINT (p + mcnt, NULL, -2);
4182 /* Uses of on_failure_jump:
4184 Each alternative starts with an on_failure_jump that points
4185 to the beginning of the next alternative. Each alternative
4186 except the last ends with a jump that in effect jumps past
4187 the rest of the alternatives. (They really jump to the
4188 ending jump of the following alternative, because tensioning
4189 these jumps is a hassle.)
4191 Repeats start with an on_failure_jump that points past both
4192 the repetition text and either the following jump or
4193 pop_failure_jump back to this on_failure_jump. */
4194 case on_failure_jump:
4196 DEBUG_PRINT1 ("EXECUTING on_failure_jump");
4198 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4199 DEBUG_PRINT3 (" %d (to 0x%x)", mcnt, p + mcnt);
4201 /* If this on_failure_jump comes right before a group (i.e.,
4202 the original * applied to a group), save the information
4203 for that group and all inner ones, so that if we fail back
4204 to this point, the group's information will be correct.
4205 For example, in \(a*\)*\1, we need the preceding group,
4206 and in \(\(a*\)b*\)\2, we need the inner group. */
4208 /* We can't use `p' to check ahead because we push
4209 a failure point to `p + mcnt' after we do this. */
4212 /* We need to skip no_op's before we look for the
4213 start_memory in case this on_failure_jump is happening as
4214 the result of a completed succeed_n, as in \(a\)\{1,3\}b\1
4216 while (p1 < pend && (re_opcode_t) *p1 == no_op)
4219 if (p1 < pend && (re_opcode_t) *p1 == start_memory)
4221 /* We have a new highest active register now. This will
4222 get reset at the start_memory we are about to get to,
4223 but we will have saved all the registers relevant to
4224 this repetition op, as described above. */
4225 highest_active_reg = *(p1 + 1) + *(p1 + 2);
4226 if (lowest_active_reg == NO_LOWEST_ACTIVE_REG)
4227 lowest_active_reg = *(p1 + 1);
4230 DEBUG_PRINT1 (":\n");
4231 PUSH_FAILURE_POINT (p + mcnt, d, -2);
4235 /* A smart repeat ends with `maybe_pop_jump'.
4236 We change it to either `pop_failure_jump' or `jump'. */
4237 case maybe_pop_jump:
4238 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4239 DEBUG_PRINT2 ("EXECUTING maybe_pop_jump %d.\n", mcnt);
4241 register unsigned char *p2 = p;
4243 /* Compare the beginning of the repeat with what in the
4244 pattern follows its end. If we can establish that there
4245 is nothing that they would both match, i.e., that we
4246 would have to backtrack because of (as in, e.g., `a*a')
4247 then we can change to pop_failure_jump, because we'll
4248 never have to backtrack.
4250 This is not true in the case of alternatives: in
4251 `(a|ab)*' we do need to backtrack to the `ab' alternative
4252 (e.g., if the string was `ab'). But instead of trying to
4253 detect that here, the alternative has put on a dummy
4254 failure point which is what we will end up popping. */
4256 /* Skip over open/close-group commands.
4257 If what follows this loop is a ...+ construct,
4258 look at what begins its body, since we will have to
4259 match at least one of that. */
4263 && ((re_opcode_t) *p2 == stop_memory
4264 || (re_opcode_t) *p2 == start_memory))
4266 else if (p2 + 6 < pend
4267 && (re_opcode_t) *p2 == dummy_failure_jump)
4274 /* p1[0] ... p1[2] are the `on_failure_jump' corresponding
4275 to the `maybe_finalize_jump' of this case. Examine what
4278 /* If we're at the end of the pattern, we can change. */
4281 /* Consider what happens when matching ":\(.*\)"
4282 against ":/". I don't really understand this code
4284 p[-3] = (unsigned char) pop_failure_jump;
4286 (" End of pattern: change to `pop_failure_jump'.\n");
4289 else if ((re_opcode_t) *p2 == exactn
4290 || (bufp->newline_anchor && (re_opcode_t) *p2 == endline))
4292 register unsigned char c
4293 = *p2 == (unsigned char) endline ? '\n' : p2[2];
4295 if ((re_opcode_t) p1[3] == exactn && p1[5] != c)
4297 p[-3] = (unsigned char) pop_failure_jump;
4298 DEBUG_PRINT3 (" %c != %c => pop_failure_jump.\n",
4302 else if ((re_opcode_t) p1[3] == charset
4303 || (re_opcode_t) p1[3] == charset_not)
4305 int not = (re_opcode_t) p1[3] == charset_not;
4307 if (c < (unsigned char) (p1[4] * BYTEWIDTH)
4308 && p1[5 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH)))
4311 /* `not' is equal to 1 if c would match, which means
4312 that we can't change to pop_failure_jump. */
4315 p[-3] = (unsigned char) pop_failure_jump;
4316 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
4320 else if ((re_opcode_t) *p2 == charset)
4323 register unsigned char c
4324 = *p2 == (unsigned char) endline ? '\n' : p2[2];
4327 if ((re_opcode_t) p1[3] == exactn
4328 && ! ((int) p2[1] * BYTEWIDTH > (int) p1[4]
4329 && (p2[1 + p1[4] / BYTEWIDTH]
4330 & (1 << (p1[4] % BYTEWIDTH)))))
4332 p[-3] = (unsigned char) pop_failure_jump;
4333 DEBUG_PRINT3 (" %c != %c => pop_failure_jump.\n",
4337 else if ((re_opcode_t) p1[3] == charset_not)
4340 /* We win if the charset_not inside the loop
4341 lists every character listed in the charset after. */
4342 for (idx = 0; idx < (int) p2[1]; idx++)
4343 if (! (p2[2 + idx] == 0
4344 || (idx < (int) p1[4]
4345 && ((p2[2 + idx] & ~ p1[5 + idx]) == 0))))
4350 p[-3] = (unsigned char) pop_failure_jump;
4351 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
4354 else if ((re_opcode_t) p1[3] == charset)
4357 /* We win if the charset inside the loop
4358 has no overlap with the one after the loop. */
4360 idx < (int) p2[1] && idx < (int) p1[4];
4362 if ((p2[2 + idx] & p1[5 + idx]) != 0)
4365 if (idx == p2[1] || idx == p1[4])
4367 p[-3] = (unsigned char) pop_failure_jump;
4368 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
4373 p -= 2; /* Point at relative address again. */
4374 if ((re_opcode_t) p[-1] != pop_failure_jump)
4376 p[-1] = (unsigned char) jump;
4377 DEBUG_PRINT1 (" Match => jump.\n");
4378 goto unconditional_jump;
4380 /* Note fall through. */
4383 /* The end of a simple repeat has a pop_failure_jump back to
4384 its matching on_failure_jump, where the latter will push a
4385 failure point. The pop_failure_jump takes off failure
4386 points put on by this pop_failure_jump's matching
4387 on_failure_jump; we got through the pattern to here from the
4388 matching on_failure_jump, so didn't fail. */
4389 case pop_failure_jump:
4391 /* We need to pass separate storage for the lowest and
4392 highest registers, even though we don't care about the
4393 actual values. Otherwise, we will restore only one
4394 register from the stack, since lowest will == highest in
4395 `pop_failure_point'. */
4396 unsigned dummy_low_reg, dummy_high_reg;
4397 unsigned char *pdummy;
4400 DEBUG_PRINT1 ("EXECUTING pop_failure_jump.\n");
4401 POP_FAILURE_POINT (sdummy, pdummy,
4402 dummy_low_reg, dummy_high_reg,
4403 reg_dummy, reg_dummy, reg_info_dummy);
4405 /* Note fall through. */
4408 /* Unconditionally jump (without popping any failure points). */
4411 EXTRACT_NUMBER_AND_INCR (mcnt, p); /* Get the amount to jump. */
4412 DEBUG_PRINT2 ("EXECUTING jump %d ", mcnt);
4413 p += mcnt; /* Do the jump. */
4414 DEBUG_PRINT2 ("(to 0x%x).\n", p);
4418 /* We need this opcode so we can detect where alternatives end
4419 in `group_match_null_string_p' et al. */
4421 DEBUG_PRINT1 ("EXECUTING jump_past_alt.\n");
4422 goto unconditional_jump;
4425 /* Normally, the on_failure_jump pushes a failure point, which
4426 then gets popped at pop_failure_jump. We will end up at
4427 pop_failure_jump, also, and with a pattern of, say, `a+', we
4428 are skipping over the on_failure_jump, so we have to push
4429 something meaningless for pop_failure_jump to pop. */
4430 case dummy_failure_jump:
4431 DEBUG_PRINT1 ("EXECUTING dummy_failure_jump.\n");
4432 /* It doesn't matter what we push for the string here. What
4433 the code at `fail' tests is the value for the pattern. */
4434 PUSH_FAILURE_POINT (0, 0, -2);
4435 goto unconditional_jump;
4438 /* At the end of an alternative, we need to push a dummy failure
4439 point in case we are followed by a `pop_failure_jump', because
4440 we don't want the failure point for the alternative to be
4441 popped. For example, matching `(a|ab)*' against `aab'
4442 requires that we match the `ab' alternative. */
4443 case push_dummy_failure:
4444 DEBUG_PRINT1 ("EXECUTING push_dummy_failure.\n");
4445 /* See comments just above at `dummy_failure_jump' about the
4447 PUSH_FAILURE_POINT (0, 0, -2);
4450 /* Have to succeed matching what follows at least n times.
4451 After that, handle like `on_failure_jump'. */
4453 EXTRACT_NUMBER (mcnt, p + 2);
4454 DEBUG_PRINT2 ("EXECUTING succeed_n %d.\n", mcnt);
4457 /* Originally, this is how many times we HAVE to succeed. */
4462 STORE_NUMBER_AND_INCR (p, mcnt);
4463 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p, mcnt);
4467 DEBUG_PRINT2 (" Setting two bytes from 0x%x to no_op.\n", p+2);
4468 p[2] = (unsigned char) no_op;
4469 p[3] = (unsigned char) no_op;
4475 EXTRACT_NUMBER (mcnt, p + 2);
4476 DEBUG_PRINT2 ("EXECUTING jump_n %d.\n", mcnt);
4478 /* Originally, this is how many times we CAN jump. */
4482 STORE_NUMBER (p + 2, mcnt);
4483 goto unconditional_jump;
4485 /* If don't have to jump any more, skip over the rest of command. */
4492 DEBUG_PRINT1 ("EXECUTING set_number_at.\n");
4494 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4496 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4497 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p1, mcnt);
4498 STORE_NUMBER (p1, mcnt);
4503 DEBUG_PRINT1 ("EXECUTING wordbound.\n");
4504 if (AT_WORD_BOUNDARY (d))
4509 DEBUG_PRINT1 ("EXECUTING notwordbound.\n");
4510 if (AT_WORD_BOUNDARY (d))
4515 DEBUG_PRINT1 ("EXECUTING wordbeg.\n");
4516 if (WORDCHAR_P (d) && (AT_STRINGS_BEG (d) || !WORDCHAR_P (d - 1)))
4521 DEBUG_PRINT1 ("EXECUTING wordend.\n");
4522 if (!AT_STRINGS_BEG (d) && WORDCHAR_P (d - 1)
4523 && (!WORDCHAR_P (d) || AT_STRINGS_END (d)))
4529 DEBUG_PRINT1 ("EXECUTING before_dot.\n");
4530 if (PTR_CHAR_POS ((unsigned char *) d) >= point)
4535 DEBUG_PRINT1 ("EXECUTING at_dot.\n");
4536 if (PTR_CHAR_POS ((unsigned char *) d) != point)
4541 DEBUG_PRINT1 ("EXECUTING after_dot.\n");
4542 if (PTR_CHAR_POS ((unsigned char *) d) <= point)
4545 #if 0 /* not emacs19 */
4547 DEBUG_PRINT1 ("EXECUTING at_dot.\n");
4548 if (PTR_CHAR_POS ((unsigned char *) d) + 1 != point)
4551 #endif /* not emacs19 */
4554 DEBUG_PRINT2 ("EXECUTING syntaxspec %d.\n", mcnt);
4559 DEBUG_PRINT1 ("EXECUTING Emacs wordchar.\n");
4563 /* Can't use *d++ here; SYNTAX may be an unsafe macro. */
4565 if (SYNTAX (d[-1]) != (enum syntaxcode) mcnt)
4567 SET_REGS_MATCHED ();
4571 DEBUG_PRINT2 ("EXECUTING notsyntaxspec %d.\n", mcnt);
4573 goto matchnotsyntax;
4576 DEBUG_PRINT1 ("EXECUTING Emacs notwordchar.\n");
4580 /* Can't use *d++ here; SYNTAX may be an unsafe macro. */
4582 if (SYNTAX (d[-1]) == (enum syntaxcode) mcnt)
4584 SET_REGS_MATCHED ();
4587 #else /* not emacs */
4589 DEBUG_PRINT1 ("EXECUTING non-Emacs wordchar.\n");
4591 if (!WORDCHAR_P (d))
4593 SET_REGS_MATCHED ();
4598 DEBUG_PRINT1 ("EXECUTING non-Emacs notwordchar.\n");
4602 SET_REGS_MATCHED ();
4605 #endif /* not emacs */
4610 continue; /* Successfully executed one pattern command; keep going. */
4613 /* We goto here if a matching operation fails. */
4615 if (!FAIL_STACK_EMPTY ())
4616 { /* A restart point is known. Restore to that state. */
4617 DEBUG_PRINT1 ("\nFAIL:\n");
4618 POP_FAILURE_POINT (d, p,
4619 lowest_active_reg, highest_active_reg,
4620 regstart, regend, reg_info);
4622 /* If this failure point is a dummy, try the next one. */
4626 /* If we failed to the end of the pattern, don't examine *p. */
4630 boolean is_a_jump_n = false;
4632 /* If failed to a backwards jump that's part of a repetition
4633 loop, need to pop this failure point and use the next one. */
4634 switch ((re_opcode_t) *p)
4638 case maybe_pop_jump:
4639 case pop_failure_jump:
4642 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4645 if ((is_a_jump_n && (re_opcode_t) *p1 == succeed_n)
4647 && (re_opcode_t) *p1 == on_failure_jump))
4655 if (d >= string1 && d <= end1)
4659 break; /* Matching at this starting point really fails. */
4663 goto restore_best_regs;
4667 return -1; /* Failure to match. */
4670 /* Subroutine definitions for re_match_2. */
4673 /* We are passed P pointing to a register number after a start_memory.
4675 Return true if the pattern up to the corresponding stop_memory can
4676 match the empty string, and false otherwise.
4678 If we find the matching stop_memory, sets P to point to one past its number.
4679 Otherwise, sets P to an undefined byte less than or equal to END.
4681 We don't handle duplicates properly (yet). */
4684 group_match_null_string_p (p, end, reg_info)
4685 unsigned char **p, *end;
4686 register_info_type *reg_info;
4689 /* Point to after the args to the start_memory. */
4690 unsigned char *p1 = *p + 2;
4694 /* Skip over opcodes that can match nothing, and return true or
4695 false, as appropriate, when we get to one that can't, or to the
4696 matching stop_memory. */
4698 switch ((re_opcode_t) *p1)
4700 /* Could be either a loop or a series of alternatives. */
4701 case on_failure_jump:
4703 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4705 /* If the next operation is not a jump backwards in the
4710 /* Go through the on_failure_jumps of the alternatives,
4711 seeing if any of the alternatives cannot match nothing.
4712 The last alternative starts with only a jump,
4713 whereas the rest start with on_failure_jump and end
4714 with a jump, e.g., here is the pattern for `a|b|c':
4716 /on_failure_jump/0/6/exactn/1/a/jump_past_alt/0/6
4717 /on_failure_jump/0/6/exactn/1/b/jump_past_alt/0/3
4720 So, we have to first go through the first (n-1)
4721 alternatives and then deal with the last one separately. */
4724 /* Deal with the first (n-1) alternatives, which start
4725 with an on_failure_jump (see above) that jumps to right
4726 past a jump_past_alt. */
4728 while ((re_opcode_t) p1[mcnt-3] == jump_past_alt)
4730 /* `mcnt' holds how many bytes long the alternative
4731 is, including the ending `jump_past_alt' and
4734 if (!alt_match_null_string_p (p1, p1 + mcnt - 3,
4738 /* Move to right after this alternative, including the
4742 /* Break if it's the beginning of an n-th alternative
4743 that doesn't begin with an on_failure_jump. */
4744 if ((re_opcode_t) *p1 != on_failure_jump)
4747 /* Still have to check that it's not an n-th
4748 alternative that starts with an on_failure_jump. */
4750 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4751 if ((re_opcode_t) p1[mcnt-3] != jump_past_alt)
4753 /* Get to the beginning of the n-th alternative. */
4759 /* Deal with the last alternative: go back and get number
4760 of the `jump_past_alt' just before it. `mcnt' contains
4761 the length of the alternative. */
4762 EXTRACT_NUMBER (mcnt, p1 - 2);
4764 if (!alt_match_null_string_p (p1, p1 + mcnt, reg_info))
4767 p1 += mcnt; /* Get past the n-th alternative. */
4773 assert (p1[1] == **p);
4779 if (!common_op_match_null_string_p (&p1, end, reg_info))
4782 } /* while p1 < end */
4785 } /* group_match_null_string_p */
4788 /* Similar to group_match_null_string_p, but doesn't deal with alternatives:
4789 It expects P to be the first byte of a single alternative and END one
4790 byte past the last. The alternative can contain groups. */
4793 alt_match_null_string_p (p, end, reg_info)
4794 unsigned char *p, *end;
4795 register_info_type *reg_info;
4798 unsigned char *p1 = p;
4802 /* Skip over opcodes that can match nothing, and break when we get
4803 to one that can't. */
4805 switch ((re_opcode_t) *p1)
4808 case on_failure_jump:
4810 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4815 if (!common_op_match_null_string_p (&p1, end, reg_info))
4818 } /* while p1 < end */
4821 } /* alt_match_null_string_p */
4824 /* Deals with the ops common to group_match_null_string_p and
4825 alt_match_null_string_p.
4827 Sets P to one after the op and its arguments, if any. */
4830 common_op_match_null_string_p (p, end, reg_info)
4831 unsigned char **p, *end;
4832 register_info_type *reg_info;
4837 unsigned char *p1 = *p;
4839 switch ((re_opcode_t) *p1++)
4859 assert (reg_no > 0 && reg_no <= MAX_REGNUM);
4860 ret = group_match_null_string_p (&p1, end, reg_info);
4862 /* Have to set this here in case we're checking a group which
4863 contains a group and a back reference to it. */
4865 if (REG_MATCH_NULL_STRING_P (reg_info[reg_no]) == MATCH_NULL_UNSET_VALUE)
4866 REG_MATCH_NULL_STRING_P (reg_info[reg_no]) = ret;
4872 /* If this is an optimized succeed_n for zero times, make the jump. */
4874 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4882 /* Get to the number of times to succeed. */
4884 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4889 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4897 if (!REG_MATCH_NULL_STRING_P (reg_info[*p1]))
4905 /* All other opcodes mean we cannot match the empty string. */
4911 } /* common_op_match_null_string_p */
4914 /* Return zero if TRANSLATE[S1] and TRANSLATE[S2] are identical for LEN
4915 bytes; nonzero otherwise. */
4918 bcmp_translate (s1, s2, len, translate)
4919 unsigned char *s1, *s2;
4923 register unsigned char *p1 = s1, *p2 = s2;
4926 if (translate[*p1++] != translate[*p2++]) return 1;
4932 /* Entry points for GNU code. */
4934 /* re_compile_pattern is the GNU regular expression compiler: it
4935 compiles PATTERN (of length SIZE) and puts the result in BUFP.
4936 Returns 0 if the pattern was valid, otherwise an error string.
4938 Assumes the `allocated' (and perhaps `buffer') and `translate' fields
4939 are set in BUFP on entry.
4941 We call regex_compile to do the actual compilation. */
4944 re_compile_pattern (pattern, length, bufp)
4945 const char *pattern;
4947 struct re_pattern_buffer *bufp;
4951 /* GNU code is written to assume at least RE_NREGS registers will be set
4952 (and at least one extra will be -1). */
4953 bufp->regs_allocated = REGS_UNALLOCATED;
4955 /* And GNU code determines whether or not to get register information
4956 by passing null for the REGS argument to re_match, etc., not by
4960 /* Match anchors at newline. */
4961 bufp->newline_anchor = 1;
4963 ret = regex_compile (pattern, length, re_syntax_options, bufp);
4967 return gettext (re_error_msgid[(int) ret]);
4970 /* Entry points compatible with 4.2 BSD regex library. We don't define
4971 them unless specifically requested. */
4973 #ifdef _REGEX_RE_COMP
4975 /* BSD has one and only one pattern buffer. */
4976 static struct re_pattern_buffer re_comp_buf;
4986 if (!re_comp_buf.buffer)
4987 return gettext ("No previous regular expression");
4991 if (!re_comp_buf.buffer)
4993 re_comp_buf.buffer = (unsigned char *) malloc (200);
4994 if (re_comp_buf.buffer == NULL)
4995 return gettext (re_error_msgid[(int) REG_ESPACE]);
4996 re_comp_buf.allocated = 200;
4998 re_comp_buf.fastmap = (char *) malloc (1 << BYTEWIDTH);
4999 if (re_comp_buf.fastmap == NULL)
5000 return gettext (re_error_msgid[(int) REG_ESPACE]);
5003 /* Since `re_exec' always passes NULL for the `regs' argument, we
5004 don't need to initialize the pattern buffer fields which affect it. */
5006 /* Match anchors at newlines. */
5007 re_comp_buf.newline_anchor = 1;
5009 ret = regex_compile (s, strlen (s), re_syntax_options, &re_comp_buf);
5014 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
5015 return (char *) gettext (re_error_msgid[(int) ret]);
5023 const int len = strlen (s);
5025 0 <= re_search (&re_comp_buf, s, len, 0, len, (struct re_registers *) 0);
5027 #endif /* _REGEX_RE_COMP */
5029 /* POSIX.2 functions. Don't define these for Emacs. */
5033 /* regcomp takes a regular expression as a string and compiles it.
5035 PREG is a regex_t *. We do not expect any fields to be initialized,
5036 since POSIX says we shouldn't. Thus, we set
5038 `buffer' to the compiled pattern;
5039 `used' to the length of the compiled pattern;
5040 `syntax' to RE_SYNTAX_POSIX_EXTENDED if the
5041 REG_EXTENDED bit in CFLAGS is set; otherwise, to
5042 RE_SYNTAX_POSIX_BASIC;
5043 `newline_anchor' to REG_NEWLINE being set in CFLAGS;
5044 `fastmap' and `fastmap_accurate' to zero;
5045 `re_nsub' to the number of subexpressions in PATTERN.
5047 PATTERN is the address of the pattern string.
5049 CFLAGS is a series of bits which affect compilation.
5051 If REG_EXTENDED is set, we use POSIX extended syntax; otherwise, we
5052 use POSIX basic syntax.
5054 If REG_NEWLINE is set, then . and [^...] don't match newline.
5055 Also, regexec will try a match beginning after every newline.
5057 If REG_ICASE is set, then we considers upper- and lowercase
5058 versions of letters to be equivalent when matching.
5060 If REG_NOSUB is set, then when PREG is passed to regexec, that
5061 routine will report only success or failure, and nothing about the
5064 It returns 0 if it succeeds, nonzero if it doesn't. (See regex.h for
5065 the return codes and their meanings.) */
5068 regcomp (preg, pattern, cflags)
5070 const char *pattern;
5075 = (cflags & REG_EXTENDED) ?
5076 RE_SYNTAX_POSIX_EXTENDED : RE_SYNTAX_POSIX_BASIC;
5078 /* regex_compile will allocate the space for the compiled pattern. */
5080 preg->allocated = 0;
5083 /* Don't bother to use a fastmap when searching. This simplifies the
5084 REG_NEWLINE case: if we used a fastmap, we'd have to put all the
5085 characters after newlines into the fastmap. This way, we just try
5089 if (cflags & REG_ICASE)
5093 preg->translate = (char *) malloc (CHAR_SET_SIZE);
5094 if (preg->translate == NULL)
5095 return (int) REG_ESPACE;
5097 /* Map uppercase characters to corresponding lowercase ones. */
5098 for (i = 0; i < CHAR_SET_SIZE; i++)
5099 preg->translate[i] = ISUPPER (i) ? tolower (i) : i;
5102 preg->translate = NULL;
5104 /* If REG_NEWLINE is set, newlines are treated differently. */
5105 if (cflags & REG_NEWLINE)
5106 { /* REG_NEWLINE implies neither . nor [^...] match newline. */
5107 syntax &= ~RE_DOT_NEWLINE;
5108 syntax |= RE_HAT_LISTS_NOT_NEWLINE;
5109 /* It also changes the matching behavior. */
5110 preg->newline_anchor = 1;
5113 preg->newline_anchor = 0;
5115 preg->no_sub = !!(cflags & REG_NOSUB);
5117 /* POSIX says a null character in the pattern terminates it, so we
5118 can use strlen here in compiling the pattern. */
5119 ret = regex_compile (pattern, strlen (pattern), syntax, preg);
5121 /* POSIX doesn't distinguish between an unmatched open-group and an
5122 unmatched close-group: both are REG_EPAREN. */
5123 if (ret == REG_ERPAREN) ret = REG_EPAREN;
5129 /* regexec searches for a given pattern, specified by PREG, in the
5132 If NMATCH is zero or REG_NOSUB was set in the cflags argument to
5133 `regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at
5134 least NMATCH elements, and we set them to the offsets of the
5135 corresponding matched substrings.
5137 EFLAGS specifies `execution flags' which affect matching: if
5138 REG_NOTBOL is set, then ^ does not match at the beginning of the
5139 string; if REG_NOTEOL is set, then $ does not match at the end.
5141 We return 0 if we find a match and REG_NOMATCH if not. */
5144 regexec (preg, string, nmatch, pmatch, eflags)
5145 const regex_t *preg;
5148 regmatch_t pmatch[];
5152 struct re_registers regs;
5153 regex_t private_preg;
5154 int len = strlen (string);
5155 boolean want_reg_info = !preg->no_sub && nmatch > 0;
5157 private_preg = *preg;
5159 private_preg.not_bol = !!(eflags & REG_NOTBOL);
5160 private_preg.not_eol = !!(eflags & REG_NOTEOL);
5162 /* The user has told us exactly how many registers to return
5163 information about, via `nmatch'. We have to pass that on to the
5164 matching routines. */
5165 private_preg.regs_allocated = REGS_FIXED;
5169 regs.num_regs = nmatch;
5170 regs.start = TALLOC (nmatch, regoff_t);
5171 regs.end = TALLOC (nmatch, regoff_t);
5172 if (regs.start == NULL || regs.end == NULL)
5173 return (int) REG_NOMATCH;
5176 /* Perform the searching operation. */
5177 ret = re_search (&private_preg, string, len,
5178 /* start: */ 0, /* range: */ len,
5179 want_reg_info ? ®s : (struct re_registers *) 0);
5181 /* Copy the register information to the POSIX structure. */
5188 for (r = 0; r < nmatch; r++)
5190 pmatch[r].rm_so = regs.start[r];
5191 pmatch[r].rm_eo = regs.end[r];
5195 /* If we needed the temporary register info, free the space now. */
5200 /* We want zero return to mean success, unlike `re_search'. */
5201 return ret >= 0 ? (int) REG_NOERROR : (int) REG_NOMATCH;
5205 /* Returns a message corresponding to an error code, ERRCODE, returned
5206 from either regcomp or regexec. We don't use PREG here. */
5209 regerror (errcode, preg, errbuf, errbuf_size)
5211 const regex_t *preg;
5219 || errcode >= (sizeof (re_error_msgid) / sizeof (re_error_msgid[0])))
5220 /* Only error codes returned by the rest of the code should be passed
5221 to this routine. If we are given anything else, or if other regex
5222 code generates an invalid error code, then the program has a bug.
5223 Dump core so we can fix it. */
5226 msg = gettext (re_error_msgid[errcode]);
5228 msg_size = strlen (msg) + 1; /* Includes the null. */
5230 if (errbuf_size != 0)
5232 if (msg_size > errbuf_size)
5234 strncpy (errbuf, msg, errbuf_size - 1);
5235 errbuf[errbuf_size - 1] = 0;
5238 strcpy (errbuf, msg);
5245 /* Free dynamically allocated space used by PREG. */
5251 if (preg->buffer != NULL)
5252 free (preg->buffer);
5253 preg->buffer = NULL;
5255 preg->allocated = 0;
5258 if (preg->fastmap != NULL)
5259 free (preg->fastmap);
5260 preg->fastmap = NULL;
5261 preg->fastmap_accurate = 0;
5263 if (preg->translate != NULL)
5264 free (preg->translate);
5265 preg->translate = NULL;
5268 #endif /* not emacs */
5272 make-backup-files: t
5274 trim-versions-without-asking: nil