1 /* Extended regular expression matching and search library,
3 (Implements POSIX draft P10003.2/D11.2, except for
4 internationalization features.)
6 Copyright (C) 1993, 1994 Free Software Foundation, Inc.
8 This program is free software; you can redistribute it and/or modify
9 it under the terms of the GNU General Public License as published by
10 the Free Software Foundation; either version 2, or (at your option)
13 This program is distributed in the hope that it will be useful,
14 but WITHOUT ANY WARRANTY; without even the implied warranty of
15 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
16 GNU General Public License for more details.
18 You should have received a copy of the GNU General Public License
19 along with this program; if not, write to the Free Software
20 Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. */
22 /* AIX requires this to be the first thing in the file. */
23 #if defined (_AIX) && !defined (REGEX_MALLOC)
33 /* We need this for `regex.h', and perhaps for the Emacs include files. */
34 #include <sys/types.h>
36 /* This is for other GNU distributions with internationalized messages.
37 The GNU C Library itself does not yet support such messages. */
41 # define gettext(msgid) (msgid)
44 /* The `emacs' switch turns on certain matching commands
45 that make sense only in Emacs. */
62 /* We used to test for `BSTRING' here, but only GCC and Emacs define
63 `BSTRING', as far as I know, and neither of them use this code. */
64 #ifndef INHIBIT_STRING_HEADER
65 #if HAVE_STRING_H || STDC_HEADERS
68 #define bcmp(s1, s2, n) memcmp ((s1), (s2), (n))
71 #define bcopy(s, d, n) memcpy ((d), (s), (n))
74 #define bzero(s, n) memset ((s), 0, (n))
81 /* Define the syntax stuff for \<, \>, etc. */
83 /* This must be nonzero for the wordchar and notwordchar pattern
84 commands in re_match_2. */
91 extern char *re_syntax_table;
93 #else /* not SYNTAX_TABLE */
95 /* How many characters in the character set. */
96 #define CHAR_SET_SIZE 256
98 static char re_syntax_table[CHAR_SET_SIZE];
109 bzero (re_syntax_table, sizeof re_syntax_table);
111 for (c = 'a'; c <= 'z'; c++)
112 re_syntax_table[c] = Sword;
114 for (c = 'A'; c <= 'Z'; c++)
115 re_syntax_table[c] = Sword;
117 for (c = '0'; c <= '9'; c++)
118 re_syntax_table[c] = Sword;
120 re_syntax_table['_'] = Sword;
125 #endif /* not SYNTAX_TABLE */
127 #define SYNTAX(c) re_syntax_table[c]
129 #endif /* not emacs */
131 /* Get the interface, including the syntax bits. */
134 /* isalpha etc. are used for the character classes. */
137 /* Jim Meyering writes:
139 "... Some ctype macros are valid only for character codes that
140 isascii says are ASCII (SGI's IRIX-4.0.5 is one such system --when
141 using /bin/cc or gcc but without giving an ansi option). So, all
142 ctype uses should be through macros like ISPRINT... If
143 STDC_HEADERS is defined, then autoconf has verified that the ctype
144 macros don't need to be guarded with references to isascii. ...
145 Defining isascii to 1 should let any compiler worth its salt
146 eliminate the && through constant folding." */
148 #if defined (STDC_HEADERS) || (!defined (isascii) && !defined (HAVE_ISASCII))
151 #define ISASCII(c) isascii(c)
155 #define ISBLANK(c) (ISASCII (c) && isblank (c))
157 #define ISBLANK(c) ((c) == ' ' || (c) == '\t')
160 #define ISGRAPH(c) (ISASCII (c) && isgraph (c))
162 #define ISGRAPH(c) (ISASCII (c) && isprint (c) && !isspace (c))
165 #define ISPRINT(c) (ISASCII (c) && isprint (c))
166 #define ISDIGIT(c) (ISASCII (c) && isdigit (c))
167 #define ISALNUM(c) (ISASCII (c) && isalnum (c))
168 #define ISALPHA(c) (ISASCII (c) && isalpha (c))
169 #define ISCNTRL(c) (ISASCII (c) && iscntrl (c))
170 #define ISLOWER(c) (ISASCII (c) && islower (c))
171 #define ISPUNCT(c) (ISASCII (c) && ispunct (c))
172 #define ISSPACE(c) (ISASCII (c) && isspace (c))
173 #define ISUPPER(c) (ISASCII (c) && isupper (c))
174 #define ISXDIGIT(c) (ISASCII (c) && isxdigit (c))
180 /* We remove any previous definition of `SIGN_EXTEND_CHAR',
181 since ours (we hope) works properly with all combinations of
182 machines, compilers, `char' and `unsigned char' argument types.
183 (Per Bothner suggested the basic approach.) */
184 #undef SIGN_EXTEND_CHAR
186 #define SIGN_EXTEND_CHAR(c) ((signed char) (c))
187 #else /* not __STDC__ */
188 /* As in Harbison and Steele. */
189 #define SIGN_EXTEND_CHAR(c) ((((unsigned char) (c)) ^ 128) - 128)
192 /* Should we use malloc or alloca? If REGEX_MALLOC is not defined, we
193 use `alloca' instead of `malloc'. This is because using malloc in
194 re_search* or re_match* could cause memory leaks when C-g is used in
195 Emacs; also, malloc is slower and causes storage fragmentation. On
196 the other hand, malloc is more portable, and easier to debug.
198 Because we sometimes use alloca, some routines have to be macros,
199 not functions -- `alloca'-allocated space disappears at the end of the
200 function it is called in. */
204 #define REGEX_ALLOCATE malloc
205 #define REGEX_REALLOCATE(source, osize, nsize) realloc (source, nsize)
207 #else /* not REGEX_MALLOC */
209 /* Emacs already defines alloca, sometimes. */
212 /* Make alloca work the best possible way. */
214 #define alloca __builtin_alloca
215 #else /* not __GNUC__ */
218 #else /* not __GNUC__ or HAVE_ALLOCA_H */
219 #ifndef _AIX /* Already did AIX, up at the top. */
221 #endif /* not _AIX */
222 #endif /* not HAVE_ALLOCA_H */
223 #endif /* not __GNUC__ */
225 #endif /* not alloca */
227 #define REGEX_ALLOCATE alloca
229 /* Assumes a `char *destination' variable. */
230 #define REGEX_REALLOCATE(source, osize, nsize) \
231 (destination = (char *) alloca (nsize), \
232 bcopy (source, destination, osize), \
235 #endif /* not REGEX_MALLOC */
238 /* True if `size1' is non-NULL and PTR is pointing anywhere inside
239 `string1' or just past its end. This works if PTR is NULL, which is
241 #define FIRST_STRING_P(ptr) \
242 (size1 && string1 <= (ptr) && (ptr) <= string1 + size1)
244 /* (Re)Allocate N items of type T using malloc, or fail. */
245 #define TALLOC(n, t) ((t *) malloc ((n) * sizeof (t)))
246 #define RETALLOC(addr, n, t) ((addr) = (t *) realloc (addr, (n) * sizeof (t)))
247 #define RETALLOC_IF(addr, n, t) \
248 if (addr) RETALLOC((addr), (n), t); else (addr) = TALLOC ((n), t)
249 #define REGEX_TALLOC(n, t) ((t *) REGEX_ALLOCATE ((n) * sizeof (t)))
251 #define BYTEWIDTH 8 /* In bits. */
253 #define STREQ(s1, s2) ((strcmp (s1, s2) == 0))
257 #define MAX(a, b) ((a) > (b) ? (a) : (b))
258 #define MIN(a, b) ((a) < (b) ? (a) : (b))
260 typedef char boolean;
264 static int re_match_2_internal ();
266 /* These are the command codes that appear in compiled regular
267 expressions. Some opcodes are followed by argument bytes. A
268 command code can specify any interpretation whatsoever for its
269 arguments. Zero bytes may appear in the compiled regular expression. */
275 /* Succeed right away--no more backtracking. */
278 /* Followed by one byte giving n, then by n literal bytes. */
281 /* Matches any (more or less) character. */
284 /* Matches any one char belonging to specified set. First
285 following byte is number of bitmap bytes. Then come bytes
286 for a bitmap saying which chars are in. Bits in each byte
287 are ordered low-bit-first. A character is in the set if its
288 bit is 1. A character too large to have a bit in the map is
289 automatically not in the set. */
292 /* Same parameters as charset, but match any character that is
293 not one of those specified. */
296 /* Start remembering the text that is matched, for storing in a
297 register. Followed by one byte with the register number, in
298 the range 0 to one less than the pattern buffer's re_nsub
299 field. Then followed by one byte with the number of groups
300 inner to this one. (This last has to be part of the
301 start_memory only because we need it in the on_failure_jump
305 /* Stop remembering the text that is matched and store it in a
306 memory register. Followed by one byte with the register
307 number, in the range 0 to one less than `re_nsub' in the
308 pattern buffer, and one byte with the number of inner groups,
309 just like `start_memory'. (We need the number of inner
310 groups here because we don't have any easy way of finding the
311 corresponding start_memory when we're at a stop_memory.) */
314 /* Match a duplicate of something remembered. Followed by one
315 byte containing the register number. */
318 /* Fail unless at beginning of line. */
321 /* Fail unless at end of line. */
324 /* Succeeds if at beginning of buffer (if emacs) or at beginning
325 of string to be matched (if not). */
328 /* Analogously, for end of buffer/string. */
331 /* Followed by two byte relative address to which to jump. */
334 /* Same as jump, but marks the end of an alternative. */
337 /* Followed by two-byte relative address of place to resume at
338 in case of failure. */
341 /* Like on_failure_jump, but pushes a placeholder instead of the
342 current string position when executed. */
343 on_failure_keep_string_jump,
345 /* Throw away latest failure point and then jump to following
346 two-byte relative address. */
349 /* Change to pop_failure_jump if know won't have to backtrack to
350 match; otherwise change to jump. This is used to jump
351 back to the beginning of a repeat. If what follows this jump
352 clearly won't match what the repeat does, such that we can be
353 sure that there is no use backtracking out of repetitions
354 already matched, then we change it to a pop_failure_jump.
355 Followed by two-byte address. */
358 /* Jump to following two-byte address, and push a dummy failure
359 point. This failure point will be thrown away if an attempt
360 is made to use it for a failure. A `+' construct makes this
361 before the first repeat. Also used as an intermediary kind
362 of jump when compiling an alternative. */
365 /* Push a dummy failure point and continue. Used at the end of
369 /* Followed by two-byte relative address and two-byte number n.
370 After matching N times, jump to the address upon failure. */
373 /* Followed by two-byte relative address, and two-byte number n.
374 Jump to the address N times, then fail. */
377 /* Set the following two-byte relative address to the
378 subsequent two-byte number. The address *includes* the two
382 wordchar, /* Matches any word-constituent character. */
383 notwordchar, /* Matches any char that is not a word-constituent. */
385 wordbeg, /* Succeeds if at word beginning. */
386 wordend, /* Succeeds if at word end. */
388 wordbound, /* Succeeds if at a word boundary. */
389 notwordbound /* Succeeds if not at a word boundary. */
392 ,before_dot, /* Succeeds if before point. */
393 at_dot, /* Succeeds if at point. */
394 after_dot, /* Succeeds if after point. */
396 /* Matches any character whose syntax is specified. Followed by
397 a byte which contains a syntax code, e.g., Sword. */
400 /* Matches any character whose syntax is not that specified. */
405 /* Common operations on the compiled pattern. */
407 /* Store NUMBER in two contiguous bytes starting at DESTINATION. */
409 #define STORE_NUMBER(destination, number) \
411 (destination)[0] = (number) & 0377; \
412 (destination)[1] = (number) >> 8; \
415 /* Same as STORE_NUMBER, except increment DESTINATION to
416 the byte after where the number is stored. Therefore, DESTINATION
417 must be an lvalue. */
419 #define STORE_NUMBER_AND_INCR(destination, number) \
421 STORE_NUMBER (destination, number); \
422 (destination) += 2; \
425 /* Put into DESTINATION a number stored in two contiguous bytes starting
428 #define EXTRACT_NUMBER(destination, source) \
430 (destination) = *(source) & 0377; \
431 (destination) += SIGN_EXTEND_CHAR (*((source) + 1)) << 8; \
436 extract_number (dest, source)
438 unsigned char *source;
440 int temp = SIGN_EXTEND_CHAR (*(source + 1));
441 *dest = *source & 0377;
445 #ifndef EXTRACT_MACROS /* To debug the macros. */
446 #undef EXTRACT_NUMBER
447 #define EXTRACT_NUMBER(dest, src) extract_number (&dest, src)
448 #endif /* not EXTRACT_MACROS */
452 /* Same as EXTRACT_NUMBER, except increment SOURCE to after the number.
453 SOURCE must be an lvalue. */
455 #define EXTRACT_NUMBER_AND_INCR(destination, source) \
457 EXTRACT_NUMBER (destination, source); \
463 extract_number_and_incr (destination, source)
465 unsigned char **source;
467 extract_number (destination, *source);
471 #ifndef EXTRACT_MACROS
472 #undef EXTRACT_NUMBER_AND_INCR
473 #define EXTRACT_NUMBER_AND_INCR(dest, src) \
474 extract_number_and_incr (&dest, &src)
475 #endif /* not EXTRACT_MACROS */
479 /* If DEBUG is defined, Regex prints many voluminous messages about what
480 it is doing (if the variable `debug' is nonzero). If linked with the
481 main program in `iregex.c', you can enter patterns and strings
482 interactively. And if linked with the main program in `main.c' and
483 the other test files, you can run the already-written tests. */
487 /* We use standard I/O for debugging. */
490 /* It is useful to test things that ``must'' be true when debugging. */
493 static int debug = 0;
495 #define DEBUG_STATEMENT(e) e
496 #define DEBUG_PRINT1(x) if (debug) printf (x)
497 #define DEBUG_PRINT2(x1, x2) if (debug) printf (x1, x2)
498 #define DEBUG_PRINT3(x1, x2, x3) if (debug) printf (x1, x2, x3)
499 #define DEBUG_PRINT4(x1, x2, x3, x4) if (debug) printf (x1, x2, x3, x4)
500 #define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) \
501 if (debug) print_partial_compiled_pattern (s, e)
502 #define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) \
503 if (debug) print_double_string (w, s1, sz1, s2, sz2)
506 extern void printchar ();
508 /* Print the fastmap in human-readable form. */
511 print_fastmap (fastmap)
514 unsigned was_a_range = 0;
517 while (i < (1 << BYTEWIDTH))
523 while (i < (1 << BYTEWIDTH) && fastmap[i])
539 /* Print a compiled pattern string in human-readable form, starting at
540 the START pointer into it and ending just before the pointer END. */
543 print_partial_compiled_pattern (start, end)
544 unsigned char *start;
548 unsigned char *p = start;
549 unsigned char *pend = end;
557 /* Loop over pattern commands. */
560 printf ("%d:\t", p - start);
562 switch ((re_opcode_t) *p++)
570 printf ("/exactn/%d", mcnt);
581 printf ("/start_memory/%d/%d", mcnt, *p++);
586 printf ("/stop_memory/%d/%d", mcnt, *p++);
590 printf ("/duplicate/%d", *p++);
600 register int c, last = -100;
601 register int in_range = 0;
603 printf ("/charset [%s",
604 (re_opcode_t) *(p - 1) == charset_not ? "^" : "");
606 assert (p + *p < pend);
608 for (c = 0; c < 256; c++)
610 && (p[1 + (c/8)] & (1 << (c % 8))))
612 /* Are we starting a range? */
613 if (last + 1 == c && ! in_range)
618 /* Have we broken a range? */
619 else if (last + 1 != c && in_range)
648 case on_failure_jump:
649 extract_number_and_incr (&mcnt, &p);
650 printf ("/on_failure_jump to %d", p + mcnt - start);
653 case on_failure_keep_string_jump:
654 extract_number_and_incr (&mcnt, &p);
655 printf ("/on_failure_keep_string_jump to %d", p + mcnt - start);
658 case dummy_failure_jump:
659 extract_number_and_incr (&mcnt, &p);
660 printf ("/dummy_failure_jump to %d", p + mcnt - start);
663 case push_dummy_failure:
664 printf ("/push_dummy_failure");
668 extract_number_and_incr (&mcnt, &p);
669 printf ("/maybe_pop_jump to %d", p + mcnt - start);
672 case pop_failure_jump:
673 extract_number_and_incr (&mcnt, &p);
674 printf ("/pop_failure_jump to %d", p + mcnt - start);
678 extract_number_and_incr (&mcnt, &p);
679 printf ("/jump_past_alt to %d", p + mcnt - start);
683 extract_number_and_incr (&mcnt, &p);
684 printf ("/jump to %d", p + mcnt - start);
688 extract_number_and_incr (&mcnt, &p);
689 extract_number_and_incr (&mcnt2, &p);
690 printf ("/succeed_n to %d, %d times", p + mcnt - start, mcnt2);
694 extract_number_and_incr (&mcnt, &p);
695 extract_number_and_incr (&mcnt2, &p);
696 printf ("/jump_n to %d, %d times", p + mcnt - start, mcnt2);
700 extract_number_and_incr (&mcnt, &p);
701 extract_number_and_incr (&mcnt2, &p);
702 printf ("/set_number_at location %d to %d", p + mcnt - start, mcnt2);
706 printf ("/wordbound");
710 printf ("/notwordbound");
722 printf ("/before_dot");
730 printf ("/after_dot");
734 printf ("/syntaxspec");
736 printf ("/%d", mcnt);
740 printf ("/notsyntaxspec");
742 printf ("/%d", mcnt);
747 printf ("/wordchar");
751 printf ("/notwordchar");
763 printf ("?%d", *(p-1));
769 printf ("%d:\tend of pattern.\n", p - start);
774 print_compiled_pattern (bufp)
775 struct re_pattern_buffer *bufp;
777 unsigned char *buffer = bufp->buffer;
779 print_partial_compiled_pattern (buffer, buffer + bufp->used);
780 printf ("%d bytes used/%d bytes allocated.\n", bufp->used, bufp->allocated);
782 if (bufp->fastmap_accurate && bufp->fastmap)
784 printf ("fastmap: ");
785 print_fastmap (bufp->fastmap);
788 printf ("re_nsub: %d\t", bufp->re_nsub);
789 printf ("regs_alloc: %d\t", bufp->regs_allocated);
790 printf ("can_be_null: %d\t", bufp->can_be_null);
791 printf ("newline_anchor: %d\n", bufp->newline_anchor);
792 printf ("no_sub: %d\t", bufp->no_sub);
793 printf ("not_bol: %d\t", bufp->not_bol);
794 printf ("not_eol: %d\t", bufp->not_eol);
795 printf ("syntax: %d\n", bufp->syntax);
796 /* Perhaps we should print the translate table? */
801 print_double_string (where, string1, size1, string2, size2)
814 if (FIRST_STRING_P (where))
816 for (this_char = where - string1; this_char < size1; this_char++)
817 printchar (string1[this_char]);
822 for (this_char = where - string2; this_char < size2; this_char++)
823 printchar (string2[this_char]);
827 #else /* not DEBUG */
832 #define DEBUG_STATEMENT(e)
833 #define DEBUG_PRINT1(x)
834 #define DEBUG_PRINT2(x1, x2)
835 #define DEBUG_PRINT3(x1, x2, x3)
836 #define DEBUG_PRINT4(x1, x2, x3, x4)
837 #define DEBUG_PRINT_COMPILED_PATTERN(p, s, e)
838 #define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2)
840 #endif /* not DEBUG */
842 /* Set by `re_set_syntax' to the current regexp syntax to recognize. Can
843 also be assigned to arbitrarily: each pattern buffer stores its own
844 syntax, so it can be changed between regex compilations. */
845 /* This has no initializer because initialized variables in Emacs
846 become read-only after dumping. */
847 reg_syntax_t re_syntax_options;
850 /* Specify the precise syntax of regexps for compilation. This provides
851 for compatibility for various utilities which historically have
852 different, incompatible syntaxes.
854 The argument SYNTAX is a bit mask comprised of the various bits
855 defined in regex.h. We return the old syntax. */
858 re_set_syntax (syntax)
861 reg_syntax_t ret = re_syntax_options;
863 re_syntax_options = syntax;
867 /* This table gives an error message for each of the error codes listed
868 in regex.h. Obviously the order here has to be same as there.
869 POSIX doesn't require that we do anything for REG_NOERROR,
870 but why not be nice? */
872 static const char *re_error_msgid[] =
873 { "Success", /* REG_NOERROR */
874 "No match", /* REG_NOMATCH */
875 "Invalid regular expression", /* REG_BADPAT */
876 "Invalid collation character", /* REG_ECOLLATE */
877 "Invalid character class name", /* REG_ECTYPE */
878 "Trailing backslash", /* REG_EESCAPE */
879 "Invalid back reference", /* REG_ESUBREG */
880 "Unmatched [ or [^", /* REG_EBRACK */
881 "Unmatched ( or \\(", /* REG_EPAREN */
882 "Unmatched \\{", /* REG_EBRACE */
883 "Invalid content of \\{\\}", /* REG_BADBR */
884 "Invalid range end", /* REG_ERANGE */
885 "Memory exhausted", /* REG_ESPACE */
886 "Invalid preceding regular expression", /* REG_BADRPT */
887 "Premature end of regular expression", /* REG_EEND */
888 "Regular expression too big", /* REG_ESIZE */
889 "Unmatched ) or \\)", /* REG_ERPAREN */
892 /* Avoiding alloca during matching, to placate r_alloc. */
894 /* Define MATCH_MAY_ALLOCATE unless we need to make sure that the
895 searching and matching functions should not call alloca. On some
896 systems, alloca is implemented in terms of malloc, and if we're
897 using the relocating allocator routines, then malloc could cause a
898 relocation, which might (if the strings being searched are in the
899 ralloc heap) shift the data out from underneath the regexp
902 Here's another reason to avoid allocation: Emacs
903 processes input from X in a signal handler; processing X input may
904 call malloc; if input arrives while a matching routine is calling
905 malloc, then we're scrod. But Emacs can't just block input while
906 calling matching routines; then we don't notice interrupts when
907 they come in. So, Emacs blocks input around all regexp calls
908 except the matching calls, which it leaves unprotected, in the
909 faith that they will not malloc. */
911 /* Normally, this is fine. */
912 #define MATCH_MAY_ALLOCATE
914 /* The match routines may not allocate if (1) they would do it with malloc
915 and (2) it's not safe for them to use malloc. */
916 #if (defined (C_ALLOCA) || defined (REGEX_MALLOC)) && (defined (emacs) || defined (REL_ALLOC))
917 #undef MATCH_MAY_ALLOCATE
921 /* Failure stack declarations and macros; both re_compile_fastmap and
922 re_match_2 use a failure stack. These have to be macros because of
926 /* Number of failure points for which to initially allocate space
927 when matching. If this number is exceeded, we allocate more
928 space, so it is not a hard limit. */
929 #ifndef INIT_FAILURE_ALLOC
930 #define INIT_FAILURE_ALLOC 5
933 /* Roughly the maximum number of failure points on the stack. Would be
934 exactly that if always used MAX_FAILURE_SPACE each time we failed.
935 This is a variable only so users of regex can assign to it; we never
936 change it ourselves. */
937 int re_max_failures = 2000;
939 typedef unsigned char *fail_stack_elt_t;
943 fail_stack_elt_t *stack;
945 unsigned avail; /* Offset of next open position. */
948 #define FAIL_STACK_EMPTY() (fail_stack.avail == 0)
949 #define FAIL_STACK_PTR_EMPTY() (fail_stack_ptr->avail == 0)
950 #define FAIL_STACK_FULL() (fail_stack.avail == fail_stack.size)
951 #define FAIL_STACK_TOP() (fail_stack.stack[fail_stack.avail])
954 /* Initialize `fail_stack'. Do `return -2' if the alloc fails. */
956 #ifdef MATCH_MAY_ALLOCATE
957 #define INIT_FAIL_STACK() \
959 fail_stack.stack = (fail_stack_elt_t *) \
960 REGEX_ALLOCATE (INIT_FAILURE_ALLOC * sizeof (fail_stack_elt_t)); \
962 if (fail_stack.stack == NULL) \
965 fail_stack.size = INIT_FAILURE_ALLOC; \
966 fail_stack.avail = 0; \
969 #define INIT_FAIL_STACK() \
971 fail_stack.avail = 0; \
976 /* Double the size of FAIL_STACK, up to approximately `re_max_failures' items.
978 Return 1 if succeeds, and 0 if either ran out of memory
979 allocating space for it or it was already too large.
981 REGEX_REALLOCATE requires `destination' be declared. */
983 #define DOUBLE_FAIL_STACK(fail_stack) \
984 ((fail_stack).size > re_max_failures * MAX_FAILURE_ITEMS \
986 : ((fail_stack).stack = (fail_stack_elt_t *) \
987 REGEX_REALLOCATE ((fail_stack).stack, \
988 (fail_stack).size * sizeof (fail_stack_elt_t), \
989 ((fail_stack).size << 1) * sizeof (fail_stack_elt_t)), \
991 (fail_stack).stack == NULL \
993 : ((fail_stack).size <<= 1, \
997 /* Push PATTERN_OP on FAIL_STACK.
999 Return 1 if was able to do so and 0 if ran out of memory allocating
1001 #define PUSH_PATTERN_OP(pattern_op, fail_stack) \
1002 ((FAIL_STACK_FULL () \
1003 && !DOUBLE_FAIL_STACK (fail_stack)) \
1005 : ((fail_stack).stack[(fail_stack).avail++] = pattern_op, \
1008 /* This pushes an item onto the failure stack. Must be a four-byte
1009 value. Assumes the variable `fail_stack'. Probably should only
1010 be called from within `PUSH_FAILURE_POINT'. */
1011 #define PUSH_FAILURE_ITEM(item) \
1012 fail_stack.stack[fail_stack.avail++] = (fail_stack_elt_t) item
1014 /* The complement operation. Assumes `fail_stack' is nonempty. */
1015 #define POP_FAILURE_ITEM() fail_stack.stack[--fail_stack.avail]
1017 /* Used to omit pushing failure point id's when we're not debugging. */
1019 #define DEBUG_PUSH PUSH_FAILURE_ITEM
1020 #define DEBUG_POP(item_addr) *(item_addr) = POP_FAILURE_ITEM ()
1022 #define DEBUG_PUSH(item)
1023 #define DEBUG_POP(item_addr)
1027 /* Push the information about the state we will need
1028 if we ever fail back to it.
1030 Requires variables fail_stack, regstart, regend, reg_info, and
1031 num_regs be declared. DOUBLE_FAIL_STACK requires `destination' be
1034 Does `return FAILURE_CODE' if runs out of memory. */
1036 #define PUSH_FAILURE_POINT(pattern_place, string_place, failure_code) \
1038 char *destination; \
1039 /* Must be int, so when we don't save any registers, the arithmetic \
1040 of 0 + -1 isn't done as unsigned. */ \
1043 DEBUG_STATEMENT (failure_id++); \
1044 DEBUG_STATEMENT (nfailure_points_pushed++); \
1045 DEBUG_PRINT2 ("\nPUSH_FAILURE_POINT #%u:\n", failure_id); \
1046 DEBUG_PRINT2 (" Before push, next avail: %d\n", (fail_stack).avail);\
1047 DEBUG_PRINT2 (" size: %d\n", (fail_stack).size);\
1049 DEBUG_PRINT2 (" slots needed: %d\n", NUM_FAILURE_ITEMS); \
1050 DEBUG_PRINT2 (" available: %d\n", REMAINING_AVAIL_SLOTS); \
1052 /* Ensure we have enough space allocated for what we will push. */ \
1053 while (REMAINING_AVAIL_SLOTS < NUM_FAILURE_ITEMS) \
1055 if (!DOUBLE_FAIL_STACK (fail_stack)) \
1056 return failure_code; \
1058 DEBUG_PRINT2 ("\n Doubled stack; size now: %d\n", \
1059 (fail_stack).size); \
1060 DEBUG_PRINT2 (" slots available: %d\n", REMAINING_AVAIL_SLOTS);\
1063 /* Push the info, starting with the registers. */ \
1064 DEBUG_PRINT1 ("\n"); \
1066 for (this_reg = lowest_active_reg; this_reg <= highest_active_reg; \
1069 DEBUG_PRINT2 (" Pushing reg: %d\n", this_reg); \
1070 DEBUG_STATEMENT (num_regs_pushed++); \
1072 DEBUG_PRINT2 (" start: 0x%x\n", regstart[this_reg]); \
1073 PUSH_FAILURE_ITEM (regstart[this_reg]); \
1075 DEBUG_PRINT2 (" end: 0x%x\n", regend[this_reg]); \
1076 PUSH_FAILURE_ITEM (regend[this_reg]); \
1078 DEBUG_PRINT2 (" info: 0x%x\n ", reg_info[this_reg]); \
1079 DEBUG_PRINT2 (" match_null=%d", \
1080 REG_MATCH_NULL_STRING_P (reg_info[this_reg])); \
1081 DEBUG_PRINT2 (" active=%d", IS_ACTIVE (reg_info[this_reg])); \
1082 DEBUG_PRINT2 (" matched_something=%d", \
1083 MATCHED_SOMETHING (reg_info[this_reg])); \
1084 DEBUG_PRINT2 (" ever_matched=%d", \
1085 EVER_MATCHED_SOMETHING (reg_info[this_reg])); \
1086 DEBUG_PRINT1 ("\n"); \
1087 PUSH_FAILURE_ITEM (reg_info[this_reg].word); \
1090 DEBUG_PRINT2 (" Pushing low active reg: %d\n", lowest_active_reg);\
1091 PUSH_FAILURE_ITEM (lowest_active_reg); \
1093 DEBUG_PRINT2 (" Pushing high active reg: %d\n", highest_active_reg);\
1094 PUSH_FAILURE_ITEM (highest_active_reg); \
1096 DEBUG_PRINT2 (" Pushing pattern 0x%x: ", pattern_place); \
1097 DEBUG_PRINT_COMPILED_PATTERN (bufp, pattern_place, pend); \
1098 PUSH_FAILURE_ITEM (pattern_place); \
1100 DEBUG_PRINT2 (" Pushing string 0x%x: `", string_place); \
1101 DEBUG_PRINT_DOUBLE_STRING (string_place, string1, size1, string2, \
1103 DEBUG_PRINT1 ("'\n"); \
1104 PUSH_FAILURE_ITEM (string_place); \
1106 DEBUG_PRINT2 (" Pushing failure id: %u\n", failure_id); \
1107 DEBUG_PUSH (failure_id); \
1110 /* This is the number of items that are pushed and popped on the stack
1111 for each register. */
1112 #define NUM_REG_ITEMS 3
1114 /* Individual items aside from the registers. */
1116 #define NUM_NONREG_ITEMS 5 /* Includes failure point id. */
1118 #define NUM_NONREG_ITEMS 4
1121 /* We push at most this many items on the stack. */
1122 #define MAX_FAILURE_ITEMS ((num_regs - 1) * NUM_REG_ITEMS + NUM_NONREG_ITEMS)
1124 /* We actually push this many items. */
1125 #define NUM_FAILURE_ITEMS \
1126 ((highest_active_reg - lowest_active_reg + 1) * NUM_REG_ITEMS \
1129 /* How many items can still be added to the stack without overflowing it. */
1130 #define REMAINING_AVAIL_SLOTS ((fail_stack).size - (fail_stack).avail)
1133 /* Pops what PUSH_FAIL_STACK pushes.
1135 We restore into the parameters, all of which should be lvalues:
1136 STR -- the saved data position.
1137 PAT -- the saved pattern position.
1138 LOW_REG, HIGH_REG -- the highest and lowest active registers.
1139 REGSTART, REGEND -- arrays of string positions.
1140 REG_INFO -- array of information about each subexpression.
1142 Also assumes the variables `fail_stack' and (if debugging), `bufp',
1143 `pend', `string1', `size1', `string2', and `size2'. */
1145 #define POP_FAILURE_POINT(str, pat, low_reg, high_reg, regstart, regend, reg_info)\
1147 DEBUG_STATEMENT (fail_stack_elt_t failure_id;) \
1149 const unsigned char *string_temp; \
1151 assert (!FAIL_STACK_EMPTY ()); \
1153 /* Remove failure points and point to how many regs pushed. */ \
1154 DEBUG_PRINT1 ("POP_FAILURE_POINT:\n"); \
1155 DEBUG_PRINT2 (" Before pop, next avail: %d\n", fail_stack.avail); \
1156 DEBUG_PRINT2 (" size: %d\n", fail_stack.size); \
1158 assert (fail_stack.avail >= NUM_NONREG_ITEMS); \
1160 DEBUG_POP (&failure_id); \
1161 DEBUG_PRINT2 (" Popping failure id: %u\n", failure_id); \
1163 /* If the saved string location is NULL, it came from an \
1164 on_failure_keep_string_jump opcode, and we want to throw away the \
1165 saved NULL, thus retaining our current position in the string. */ \
1166 string_temp = POP_FAILURE_ITEM (); \
1167 if (string_temp != NULL) \
1168 str = (const char *) string_temp; \
1170 DEBUG_PRINT2 (" Popping string 0x%x: `", str); \
1171 DEBUG_PRINT_DOUBLE_STRING (str, string1, size1, string2, size2); \
1172 DEBUG_PRINT1 ("'\n"); \
1174 pat = (unsigned char *) POP_FAILURE_ITEM (); \
1175 DEBUG_PRINT2 (" Popping pattern 0x%x: ", pat); \
1176 DEBUG_PRINT_COMPILED_PATTERN (bufp, pat, pend); \
1178 /* Restore register info. */ \
1179 high_reg = (unsigned) POP_FAILURE_ITEM (); \
1180 DEBUG_PRINT2 (" Popping high active reg: %d\n", high_reg); \
1182 low_reg = (unsigned) POP_FAILURE_ITEM (); \
1183 DEBUG_PRINT2 (" Popping low active reg: %d\n", low_reg); \
1185 for (this_reg = high_reg; this_reg >= low_reg; this_reg--) \
1187 DEBUG_PRINT2 (" Popping reg: %d\n", this_reg); \
1189 reg_info[this_reg].word = POP_FAILURE_ITEM (); \
1190 DEBUG_PRINT2 (" info: 0x%x\n", reg_info[this_reg]); \
1192 regend[this_reg] = (const char *) POP_FAILURE_ITEM (); \
1193 DEBUG_PRINT2 (" end: 0x%x\n", regend[this_reg]); \
1195 regstart[this_reg] = (const char *) POP_FAILURE_ITEM (); \
1196 DEBUG_PRINT2 (" start: 0x%x\n", regstart[this_reg]); \
1199 DEBUG_STATEMENT (nfailure_points_popped++); \
1200 } /* POP_FAILURE_POINT */
1204 /* Structure for per-register (a.k.a. per-group) information.
1205 This must not be longer than one word, because we push this value
1206 onto the failure stack. Other register information, such as the
1207 starting and ending positions (which are addresses), and the list of
1208 inner groups (which is a bits list) are maintained in separate
1211 We are making a (strictly speaking) nonportable assumption here: that
1212 the compiler will pack our bit fields into something that fits into
1213 the type of `word', i.e., is something that fits into one item on the
1217 fail_stack_elt_t word;
1220 /* This field is one if this group can match the empty string,
1221 zero if not. If not yet determined, `MATCH_NULL_UNSET_VALUE'. */
1222 #define MATCH_NULL_UNSET_VALUE 3
1223 unsigned match_null_string_p : 2;
1224 unsigned is_active : 1;
1225 unsigned matched_something : 1;
1226 unsigned ever_matched_something : 1;
1228 } register_info_type;
1230 #define REG_MATCH_NULL_STRING_P(R) ((R).bits.match_null_string_p)
1231 #define IS_ACTIVE(R) ((R).bits.is_active)
1232 #define MATCHED_SOMETHING(R) ((R).bits.matched_something)
1233 #define EVER_MATCHED_SOMETHING(R) ((R).bits.ever_matched_something)
1236 /* Call this when have matched a real character; it sets `matched' flags
1237 for the subexpressions which we are currently inside. Also records
1238 that those subexprs have matched. */
1239 #define SET_REGS_MATCHED() \
1243 for (r = lowest_active_reg; r <= highest_active_reg; r++) \
1245 MATCHED_SOMETHING (reg_info[r]) \
1246 = EVER_MATCHED_SOMETHING (reg_info[r]) \
1253 /* Registers are set to a sentinel when they haven't yet matched. */
1254 static char reg_unset_dummy;
1255 #define REG_UNSET_VALUE (®_unset_dummy)
1256 #define REG_UNSET(e) ((e) == REG_UNSET_VALUE)
1260 /* How do we implement a missing MATCH_MAY_ALLOCATE?
1261 We make the fail stack a global thing, and then grow it to
1262 re_max_failures when we compile. */
1263 #ifndef MATCH_MAY_ALLOCATE
1264 static fail_stack_type fail_stack;
1266 static const char ** regstart, ** regend;
1267 static const char ** old_regstart, ** old_regend;
1268 static const char **best_regstart, **best_regend;
1269 static register_info_type *reg_info;
1270 static const char **reg_dummy;
1271 static register_info_type *reg_info_dummy;
1275 /* Subroutine declarations and macros for regex_compile. */
1277 static void store_op1 (), store_op2 ();
1278 static void insert_op1 (), insert_op2 ();
1279 static boolean at_begline_loc_p (), at_endline_loc_p ();
1280 static boolean group_in_compile_stack ();
1281 static reg_errcode_t compile_range ();
1283 /* Fetch the next character in the uncompiled pattern---translating it
1284 if necessary. Also cast from a signed character in the constant
1285 string passed to us by the user to an unsigned char that we can use
1286 as an array index (in, e.g., `translate'). */
1287 #define PATFETCH(c) \
1288 do {if (p == pend) return REG_EEND; \
1289 c = (unsigned char) *p++; \
1290 if (translate) c = translate[c]; \
1293 /* Fetch the next character in the uncompiled pattern, with no
1295 #define PATFETCH_RAW(c) \
1296 do {if (p == pend) return REG_EEND; \
1297 c = (unsigned char) *p++; \
1300 /* Go backwards one character in the pattern. */
1301 #define PATUNFETCH p--
1304 /* If `translate' is non-null, return translate[D], else just D. We
1305 cast the subscript to translate because some data is declared as
1306 `char *', to avoid warnings when a string constant is passed. But
1307 when we use a character as a subscript we must make it unsigned. */
1308 #define TRANSLATE(d) (translate ? translate[(unsigned char) (d)] : (d))
1311 /* Macros for outputting the compiled pattern into `buffer'. */
1313 /* If the buffer isn't allocated when it comes in, use this. */
1314 #define INIT_BUF_SIZE 32
1316 /* Make sure we have at least N more bytes of space in buffer. */
1317 #define GET_BUFFER_SPACE(n) \
1318 while (b - bufp->buffer + (n) > bufp->allocated) \
1321 /* Make sure we have one more byte of buffer space and then add C to it. */
1322 #define BUF_PUSH(c) \
1324 GET_BUFFER_SPACE (1); \
1325 *b++ = (unsigned char) (c); \
1329 /* Ensure we have two more bytes of buffer space and then append C1 and C2. */
1330 #define BUF_PUSH_2(c1, c2) \
1332 GET_BUFFER_SPACE (2); \
1333 *b++ = (unsigned char) (c1); \
1334 *b++ = (unsigned char) (c2); \
1338 /* As with BUF_PUSH_2, except for three bytes. */
1339 #define BUF_PUSH_3(c1, c2, c3) \
1341 GET_BUFFER_SPACE (3); \
1342 *b++ = (unsigned char) (c1); \
1343 *b++ = (unsigned char) (c2); \
1344 *b++ = (unsigned char) (c3); \
1348 /* Store a jump with opcode OP at LOC to location TO. We store a
1349 relative address offset by the three bytes the jump itself occupies. */
1350 #define STORE_JUMP(op, loc, to) \
1351 store_op1 (op, loc, (to) - (loc) - 3)
1353 /* Likewise, for a two-argument jump. */
1354 #define STORE_JUMP2(op, loc, to, arg) \
1355 store_op2 (op, loc, (to) - (loc) - 3, arg)
1357 /* Like `STORE_JUMP', but for inserting. Assume `b' is the buffer end. */
1358 #define INSERT_JUMP(op, loc, to) \
1359 insert_op1 (op, loc, (to) - (loc) - 3, b)
1361 /* Like `STORE_JUMP2', but for inserting. Assume `b' is the buffer end. */
1362 #define INSERT_JUMP2(op, loc, to, arg) \
1363 insert_op2 (op, loc, (to) - (loc) - 3, arg, b)
1366 /* This is not an arbitrary limit: the arguments which represent offsets
1367 into the pattern are two bytes long. So if 2^16 bytes turns out to
1368 be too small, many things would have to change. */
1369 #define MAX_BUF_SIZE (1L << 16)
1372 /* Extend the buffer by twice its current size via realloc and
1373 reset the pointers that pointed into the old block to point to the
1374 correct places in the new one. If extending the buffer results in it
1375 being larger than MAX_BUF_SIZE, then flag memory exhausted. */
1376 #define EXTEND_BUFFER() \
1378 unsigned char *old_buffer = bufp->buffer; \
1379 if (bufp->allocated == MAX_BUF_SIZE) \
1381 bufp->allocated <<= 1; \
1382 if (bufp->allocated > MAX_BUF_SIZE) \
1383 bufp->allocated = MAX_BUF_SIZE; \
1384 bufp->buffer = (unsigned char *) realloc (bufp->buffer, bufp->allocated);\
1385 if (bufp->buffer == NULL) \
1386 return REG_ESPACE; \
1387 /* If the buffer moved, move all the pointers into it. */ \
1388 if (old_buffer != bufp->buffer) \
1390 b = (b - old_buffer) + bufp->buffer; \
1391 begalt = (begalt - old_buffer) + bufp->buffer; \
1392 if (fixup_alt_jump) \
1393 fixup_alt_jump = (fixup_alt_jump - old_buffer) + bufp->buffer;\
1395 laststart = (laststart - old_buffer) + bufp->buffer; \
1396 if (pending_exact) \
1397 pending_exact = (pending_exact - old_buffer) + bufp->buffer; \
1402 /* Since we have one byte reserved for the register number argument to
1403 {start,stop}_memory, the maximum number of groups we can report
1404 things about is what fits in that byte. */
1405 #define MAX_REGNUM 255
1407 /* But patterns can have more than `MAX_REGNUM' registers. We just
1408 ignore the excess. */
1409 typedef unsigned regnum_t;
1412 /* Macros for the compile stack. */
1414 /* Since offsets can go either forwards or backwards, this type needs to
1415 be able to hold values from -(MAX_BUF_SIZE - 1) to MAX_BUF_SIZE - 1. */
1416 typedef int pattern_offset_t;
1420 pattern_offset_t begalt_offset;
1421 pattern_offset_t fixup_alt_jump;
1422 pattern_offset_t inner_group_offset;
1423 pattern_offset_t laststart_offset;
1425 } compile_stack_elt_t;
1430 compile_stack_elt_t *stack;
1432 unsigned avail; /* Offset of next open position. */
1433 } compile_stack_type;
1436 #define INIT_COMPILE_STACK_SIZE 32
1438 #define COMPILE_STACK_EMPTY (compile_stack.avail == 0)
1439 #define COMPILE_STACK_FULL (compile_stack.avail == compile_stack.size)
1441 /* The next available element. */
1442 #define COMPILE_STACK_TOP (compile_stack.stack[compile_stack.avail])
1445 /* Set the bit for character C in a list. */
1446 #define SET_LIST_BIT(c) \
1447 (b[((unsigned char) (c)) / BYTEWIDTH] \
1448 |= 1 << (((unsigned char) c) % BYTEWIDTH))
1451 /* Get the next unsigned number in the uncompiled pattern. */
1452 #define GET_UNSIGNED_NUMBER(num) \
1456 while (ISDIGIT (c)) \
1460 num = num * 10 + c - '0'; \
1468 #define CHAR_CLASS_MAX_LENGTH 6 /* Namely, `xdigit'. */
1470 #define IS_CHAR_CLASS(string) \
1471 (STREQ (string, "alpha") || STREQ (string, "upper") \
1472 || STREQ (string, "lower") || STREQ (string, "digit") \
1473 || STREQ (string, "alnum") || STREQ (string, "xdigit") \
1474 || STREQ (string, "space") || STREQ (string, "print") \
1475 || STREQ (string, "punct") || STREQ (string, "graph") \
1476 || STREQ (string, "cntrl") || STREQ (string, "blank"))
1478 /* `regex_compile' compiles PATTERN (of length SIZE) according to SYNTAX.
1479 Returns one of error codes defined in `regex.h', or zero for success.
1481 Assumes the `allocated' (and perhaps `buffer') and `translate'
1482 fields are set in BUFP on entry.
1484 If it succeeds, results are put in BUFP (if it returns an error, the
1485 contents of BUFP are undefined):
1486 `buffer' is the compiled pattern;
1487 `syntax' is set to SYNTAX;
1488 `used' is set to the length of the compiled pattern;
1489 `fastmap_accurate' is zero;
1490 `re_nsub' is the number of subexpressions in PATTERN;
1491 `not_bol' and `not_eol' are zero;
1493 The `fastmap' and `newline_anchor' fields are neither
1494 examined nor set. */
1496 /* Return, freeing storage we allocated. */
1497 #define FREE_STACK_RETURN(value) \
1498 return (free (compile_stack.stack), value)
1500 static reg_errcode_t
1501 regex_compile (pattern, size, syntax, bufp)
1502 const char *pattern;
1504 reg_syntax_t syntax;
1505 struct re_pattern_buffer *bufp;
1507 /* We fetch characters from PATTERN here. Even though PATTERN is
1508 `char *' (i.e., signed), we declare these variables as unsigned, so
1509 they can be reliably used as array indices. */
1510 register unsigned char c, c1;
1512 /* A random temporary spot in PATTERN. */
1515 /* Points to the end of the buffer, where we should append. */
1516 register unsigned char *b;
1518 /* Keeps track of unclosed groups. */
1519 compile_stack_type compile_stack;
1521 /* Points to the current (ending) position in the pattern. */
1522 const char *p = pattern;
1523 const char *pend = pattern + size;
1525 /* How to translate the characters in the pattern. */
1526 char *translate = bufp->translate;
1528 /* Address of the count-byte of the most recently inserted `exactn'
1529 command. This makes it possible to tell if a new exact-match
1530 character can be added to that command or if the character requires
1531 a new `exactn' command. */
1532 unsigned char *pending_exact = 0;
1534 /* Address of start of the most recently finished expression.
1535 This tells, e.g., postfix * where to find the start of its
1536 operand. Reset at the beginning of groups and alternatives. */
1537 unsigned char *laststart = 0;
1539 /* Address of beginning of regexp, or inside of last group. */
1540 unsigned char *begalt;
1542 /* Place in the uncompiled pattern (i.e., the {) to
1543 which to go back if the interval is invalid. */
1544 const char *beg_interval;
1546 /* Address of the place where a forward jump should go to the end of
1547 the containing expression. Each alternative of an `or' -- except the
1548 last -- ends with a forward jump of this sort. */
1549 unsigned char *fixup_alt_jump = 0;
1551 /* Counts open-groups as they are encountered. Remembered for the
1552 matching close-group on the compile stack, so the same register
1553 number is put in the stop_memory as the start_memory. */
1554 regnum_t regnum = 0;
1557 DEBUG_PRINT1 ("\nCompiling pattern: ");
1560 unsigned debug_count;
1562 for (debug_count = 0; debug_count < size; debug_count++)
1563 printchar (pattern[debug_count]);
1568 /* Initialize the compile stack. */
1569 compile_stack.stack = TALLOC (INIT_COMPILE_STACK_SIZE, compile_stack_elt_t);
1570 if (compile_stack.stack == NULL)
1573 compile_stack.size = INIT_COMPILE_STACK_SIZE;
1574 compile_stack.avail = 0;
1576 /* Initialize the pattern buffer. */
1577 bufp->syntax = syntax;
1578 bufp->fastmap_accurate = 0;
1579 bufp->not_bol = bufp->not_eol = 0;
1581 /* Set `used' to zero, so that if we return an error, the pattern
1582 printer (for debugging) will think there's no pattern. We reset it
1586 /* Always count groups, whether or not bufp->no_sub is set. */
1589 #if !defined (emacs) && !defined (SYNTAX_TABLE)
1590 /* Initialize the syntax table. */
1591 init_syntax_once ();
1594 if (bufp->allocated == 0)
1597 { /* If zero allocated, but buffer is non-null, try to realloc
1598 enough space. This loses if buffer's address is bogus, but
1599 that is the user's responsibility. */
1600 RETALLOC (bufp->buffer, INIT_BUF_SIZE, unsigned char);
1603 { /* Caller did not allocate a buffer. Do it for them. */
1604 bufp->buffer = TALLOC (INIT_BUF_SIZE, unsigned char);
1606 if (!bufp->buffer) FREE_STACK_RETURN (REG_ESPACE);
1608 bufp->allocated = INIT_BUF_SIZE;
1611 begalt = b = bufp->buffer;
1613 /* Loop through the uncompiled pattern until we're at the end. */
1622 if ( /* If at start of pattern, it's an operator. */
1624 /* If context independent, it's an operator. */
1625 || syntax & RE_CONTEXT_INDEP_ANCHORS
1626 /* Otherwise, depends on what's come before. */
1627 || at_begline_loc_p (pattern, p, syntax))
1637 if ( /* If at end of pattern, it's an operator. */
1639 /* If context independent, it's an operator. */
1640 || syntax & RE_CONTEXT_INDEP_ANCHORS
1641 /* Otherwise, depends on what's next. */
1642 || at_endline_loc_p (p, pend, syntax))
1652 if ((syntax & RE_BK_PLUS_QM)
1653 || (syntax & RE_LIMITED_OPS))
1657 /* If there is no previous pattern... */
1660 if (syntax & RE_CONTEXT_INVALID_OPS)
1661 FREE_STACK_RETURN (REG_BADRPT);
1662 else if (!(syntax & RE_CONTEXT_INDEP_OPS))
1667 /* Are we optimizing this jump? */
1668 boolean keep_string_p = false;
1670 /* 1 means zero (many) matches is allowed. */
1671 char zero_times_ok = 0, many_times_ok = 0;
1673 /* If there is a sequence of repetition chars, collapse it
1674 down to just one (the right one). We can't combine
1675 interval operators with these because of, e.g., `a{2}*',
1676 which should only match an even number of `a's. */
1680 zero_times_ok |= c != '+';
1681 many_times_ok |= c != '?';
1689 || (!(syntax & RE_BK_PLUS_QM) && (c == '+' || c == '?')))
1692 else if (syntax & RE_BK_PLUS_QM && c == '\\')
1694 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
1697 if (!(c1 == '+' || c1 == '?'))
1712 /* If we get here, we found another repeat character. */
1715 /* Star, etc. applied to an empty pattern is equivalent
1716 to an empty pattern. */
1720 /* Now we know whether or not zero matches is allowed
1721 and also whether or not two or more matches is allowed. */
1723 { /* More than one repetition is allowed, so put in at the
1724 end a backward relative jump from `b' to before the next
1725 jump we're going to put in below (which jumps from
1726 laststart to after this jump).
1728 But if we are at the `*' in the exact sequence `.*\n',
1729 insert an unconditional jump backwards to the .,
1730 instead of the beginning of the loop. This way we only
1731 push a failure point once, instead of every time
1732 through the loop. */
1733 assert (p - 1 > pattern);
1735 /* Allocate the space for the jump. */
1736 GET_BUFFER_SPACE (3);
1738 /* We know we are not at the first character of the pattern,
1739 because laststart was nonzero. And we've already
1740 incremented `p', by the way, to be the character after
1741 the `*'. Do we have to do something analogous here
1742 for null bytes, because of RE_DOT_NOT_NULL? */
1743 if (TRANSLATE (*(p - 2)) == TRANSLATE ('.')
1745 && p < pend && TRANSLATE (*p) == TRANSLATE ('\n')
1746 && !(syntax & RE_DOT_NEWLINE))
1747 { /* We have .*\n. */
1748 STORE_JUMP (jump, b, laststart);
1749 keep_string_p = true;
1752 /* Anything else. */
1753 STORE_JUMP (maybe_pop_jump, b, laststart - 3);
1755 /* We've added more stuff to the buffer. */
1759 /* On failure, jump from laststart to b + 3, which will be the
1760 end of the buffer after this jump is inserted. */
1761 GET_BUFFER_SPACE (3);
1762 INSERT_JUMP (keep_string_p ? on_failure_keep_string_jump
1770 /* At least one repetition is required, so insert a
1771 `dummy_failure_jump' before the initial
1772 `on_failure_jump' instruction of the loop. This
1773 effects a skip over that instruction the first time
1774 we hit that loop. */
1775 GET_BUFFER_SPACE (3);
1776 INSERT_JUMP (dummy_failure_jump, laststart, laststart + 6);
1791 boolean had_char_class = false;
1793 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
1795 /* Ensure that we have enough space to push a charset: the
1796 opcode, the length count, and the bitset; 34 bytes in all. */
1797 GET_BUFFER_SPACE (34);
1801 /* We test `*p == '^' twice, instead of using an if
1802 statement, so we only need one BUF_PUSH. */
1803 BUF_PUSH (*p == '^' ? charset_not : charset);
1807 /* Remember the first position in the bracket expression. */
1810 /* Push the number of bytes in the bitmap. */
1811 BUF_PUSH ((1 << BYTEWIDTH) / BYTEWIDTH);
1813 /* Clear the whole map. */
1814 bzero (b, (1 << BYTEWIDTH) / BYTEWIDTH);
1816 /* charset_not matches newline according to a syntax bit. */
1817 if ((re_opcode_t) b[-2] == charset_not
1818 && (syntax & RE_HAT_LISTS_NOT_NEWLINE))
1819 SET_LIST_BIT ('\n');
1821 /* Read in characters and ranges, setting map bits. */
1824 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
1828 /* \ might escape characters inside [...] and [^...]. */
1829 if ((syntax & RE_BACKSLASH_ESCAPE_IN_LISTS) && c == '\\')
1831 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
1838 /* Could be the end of the bracket expression. If it's
1839 not (i.e., when the bracket expression is `[]' so
1840 far), the ']' character bit gets set way below. */
1841 if (c == ']' && p != p1 + 1)
1844 /* Look ahead to see if it's a range when the last thing
1845 was a character class. */
1846 if (had_char_class && c == '-' && *p != ']')
1847 FREE_STACK_RETURN (REG_ERANGE);
1849 /* Look ahead to see if it's a range when the last thing
1850 was a character: if this is a hyphen not at the
1851 beginning or the end of a list, then it's the range
1854 && !(p - 2 >= pattern && p[-2] == '[')
1855 && !(p - 3 >= pattern && p[-3] == '[' && p[-2] == '^')
1859 = compile_range (&p, pend, translate, syntax, b);
1860 if (ret != REG_NOERROR) FREE_STACK_RETURN (ret);
1863 else if (p[0] == '-' && p[1] != ']')
1864 { /* This handles ranges made up of characters only. */
1867 /* Move past the `-'. */
1870 ret = compile_range (&p, pend, translate, syntax, b);
1871 if (ret != REG_NOERROR) FREE_STACK_RETURN (ret);
1874 /* See if we're at the beginning of a possible character
1877 else if (syntax & RE_CHAR_CLASSES && c == '[' && *p == ':')
1878 { /* Leave room for the null. */
1879 char str[CHAR_CLASS_MAX_LENGTH + 1];
1884 /* If pattern is `[[:'. */
1885 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
1890 if (c == ':' || c == ']' || p == pend
1891 || c1 == CHAR_CLASS_MAX_LENGTH)
1897 /* If isn't a word bracketed by `[:' and:`]':
1898 undo the ending character, the letters, and leave
1899 the leading `:' and `[' (but set bits for them). */
1900 if (c == ':' && *p == ']')
1903 boolean is_alnum = STREQ (str, "alnum");
1904 boolean is_alpha = STREQ (str, "alpha");
1905 boolean is_blank = STREQ (str, "blank");
1906 boolean is_cntrl = STREQ (str, "cntrl");
1907 boolean is_digit = STREQ (str, "digit");
1908 boolean is_graph = STREQ (str, "graph");
1909 boolean is_lower = STREQ (str, "lower");
1910 boolean is_print = STREQ (str, "print");
1911 boolean is_punct = STREQ (str, "punct");
1912 boolean is_space = STREQ (str, "space");
1913 boolean is_upper = STREQ (str, "upper");
1914 boolean is_xdigit = STREQ (str, "xdigit");
1916 if (!IS_CHAR_CLASS (str))
1917 FREE_STACK_RETURN (REG_ECTYPE);
1919 /* Throw away the ] at the end of the character
1923 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
1925 for (ch = 0; ch < 1 << BYTEWIDTH; ch++)
1927 /* This was split into 3 if's to
1928 avoid an arbitrary limit in some compiler. */
1929 if ( (is_alnum && ISALNUM (ch))
1930 || (is_alpha && ISALPHA (ch))
1931 || (is_blank && ISBLANK (ch))
1932 || (is_cntrl && ISCNTRL (ch)))
1934 if ( (is_digit && ISDIGIT (ch))
1935 || (is_graph && ISGRAPH (ch))
1936 || (is_lower && ISLOWER (ch))
1937 || (is_print && ISPRINT (ch)))
1939 if ( (is_punct && ISPUNCT (ch))
1940 || (is_space && ISSPACE (ch))
1941 || (is_upper && ISUPPER (ch))
1942 || (is_xdigit && ISXDIGIT (ch)))
1945 had_char_class = true;
1954 had_char_class = false;
1959 had_char_class = false;
1964 /* Discard any (non)matching list bytes that are all 0 at the
1965 end of the map. Decrease the map-length byte too. */
1966 while ((int) b[-1] > 0 && b[b[-1] - 1] == 0)
1974 if (syntax & RE_NO_BK_PARENS)
1981 if (syntax & RE_NO_BK_PARENS)
1988 if (syntax & RE_NEWLINE_ALT)
1995 if (syntax & RE_NO_BK_VBAR)
2002 if (syntax & RE_INTERVALS && syntax & RE_NO_BK_BRACES)
2003 goto handle_interval;
2009 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
2011 /* Do not translate the character after the \, so that we can
2012 distinguish, e.g., \B from \b, even if we normally would
2013 translate, e.g., B to b. */
2019 if (syntax & RE_NO_BK_PARENS)
2020 goto normal_backslash;
2026 if (COMPILE_STACK_FULL)
2028 RETALLOC (compile_stack.stack, compile_stack.size << 1,
2029 compile_stack_elt_t);
2030 if (compile_stack.stack == NULL) return REG_ESPACE;
2032 compile_stack.size <<= 1;
2035 /* These are the values to restore when we hit end of this
2036 group. They are all relative offsets, so that if the
2037 whole pattern moves because of realloc, they will still
2039 COMPILE_STACK_TOP.begalt_offset = begalt - bufp->buffer;
2040 COMPILE_STACK_TOP.fixup_alt_jump
2041 = fixup_alt_jump ? fixup_alt_jump - bufp->buffer + 1 : 0;
2042 COMPILE_STACK_TOP.laststart_offset = b - bufp->buffer;
2043 COMPILE_STACK_TOP.regnum = regnum;
2045 /* We will eventually replace the 0 with the number of
2046 groups inner to this one. But do not push a
2047 start_memory for groups beyond the last one we can
2048 represent in the compiled pattern. */
2049 if (regnum <= MAX_REGNUM)
2051 COMPILE_STACK_TOP.inner_group_offset = b - bufp->buffer + 2;
2052 BUF_PUSH_3 (start_memory, regnum, 0);
2055 compile_stack.avail++;
2060 /* If we've reached MAX_REGNUM groups, then this open
2061 won't actually generate any code, so we'll have to
2062 clear pending_exact explicitly. */
2068 if (syntax & RE_NO_BK_PARENS) goto normal_backslash;
2070 if (COMPILE_STACK_EMPTY)
2071 if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD)
2072 goto normal_backslash;
2074 FREE_STACK_RETURN (REG_ERPAREN);
2078 { /* Push a dummy failure point at the end of the
2079 alternative for a possible future
2080 `pop_failure_jump' to pop. See comments at
2081 `push_dummy_failure' in `re_match_2'. */
2082 BUF_PUSH (push_dummy_failure);
2084 /* We allocated space for this jump when we assigned
2085 to `fixup_alt_jump', in the `handle_alt' case below. */
2086 STORE_JUMP (jump_past_alt, fixup_alt_jump, b - 1);
2089 /* See similar code for backslashed left paren above. */
2090 if (COMPILE_STACK_EMPTY)
2091 if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD)
2094 FREE_STACK_RETURN (REG_ERPAREN);
2096 /* Since we just checked for an empty stack above, this
2097 ``can't happen''. */
2098 assert (compile_stack.avail != 0);
2100 /* We don't just want to restore into `regnum', because
2101 later groups should continue to be numbered higher,
2102 as in `(ab)c(de)' -- the second group is #2. */
2103 regnum_t this_group_regnum;
2105 compile_stack.avail--;
2106 begalt = bufp->buffer + COMPILE_STACK_TOP.begalt_offset;
2108 = COMPILE_STACK_TOP.fixup_alt_jump
2109 ? bufp->buffer + COMPILE_STACK_TOP.fixup_alt_jump - 1
2111 laststart = bufp->buffer + COMPILE_STACK_TOP.laststart_offset;
2112 this_group_regnum = COMPILE_STACK_TOP.regnum;
2113 /* If we've reached MAX_REGNUM groups, then this open
2114 won't actually generate any code, so we'll have to
2115 clear pending_exact explicitly. */
2118 /* We're at the end of the group, so now we know how many
2119 groups were inside this one. */
2120 if (this_group_regnum <= MAX_REGNUM)
2122 unsigned char *inner_group_loc
2123 = bufp->buffer + COMPILE_STACK_TOP.inner_group_offset;
2125 *inner_group_loc = regnum - this_group_regnum;
2126 BUF_PUSH_3 (stop_memory, this_group_regnum,
2127 regnum - this_group_regnum);
2133 case '|': /* `\|'. */
2134 if (syntax & RE_LIMITED_OPS || syntax & RE_NO_BK_VBAR)
2135 goto normal_backslash;
2137 if (syntax & RE_LIMITED_OPS)
2140 /* Insert before the previous alternative a jump which
2141 jumps to this alternative if the former fails. */
2142 GET_BUFFER_SPACE (3);
2143 INSERT_JUMP (on_failure_jump, begalt, b + 6);
2147 /* The alternative before this one has a jump after it
2148 which gets executed if it gets matched. Adjust that
2149 jump so it will jump to this alternative's analogous
2150 jump (put in below, which in turn will jump to the next
2151 (if any) alternative's such jump, etc.). The last such
2152 jump jumps to the correct final destination. A picture:
2158 If we are at `b', then fixup_alt_jump right now points to a
2159 three-byte space after `a'. We'll put in the jump, set
2160 fixup_alt_jump to right after `b', and leave behind three
2161 bytes which we'll fill in when we get to after `c'. */
2164 STORE_JUMP (jump_past_alt, fixup_alt_jump, b);
2166 /* Mark and leave space for a jump after this alternative,
2167 to be filled in later either by next alternative or
2168 when know we're at the end of a series of alternatives. */
2170 GET_BUFFER_SPACE (3);
2179 /* If \{ is a literal. */
2180 if (!(syntax & RE_INTERVALS)
2181 /* If we're at `\{' and it's not the open-interval
2183 || ((syntax & RE_INTERVALS) && (syntax & RE_NO_BK_BRACES))
2184 || (p - 2 == pattern && p == pend))
2185 goto normal_backslash;
2189 /* If got here, then the syntax allows intervals. */
2191 /* At least (most) this many matches must be made. */
2192 int lower_bound = -1, upper_bound = -1;
2194 beg_interval = p - 1;
2198 if (syntax & RE_NO_BK_BRACES)
2199 goto unfetch_interval;
2201 FREE_STACK_RETURN (REG_EBRACE);
2204 GET_UNSIGNED_NUMBER (lower_bound);
2208 GET_UNSIGNED_NUMBER (upper_bound);
2209 if (upper_bound < 0) upper_bound = RE_DUP_MAX;
2212 /* Interval such as `{1}' => match exactly once. */
2213 upper_bound = lower_bound;
2215 if (lower_bound < 0 || upper_bound > RE_DUP_MAX
2216 || lower_bound > upper_bound)
2218 if (syntax & RE_NO_BK_BRACES)
2219 goto unfetch_interval;
2221 FREE_STACK_RETURN (REG_BADBR);
2224 if (!(syntax & RE_NO_BK_BRACES))
2226 if (c != '\\') FREE_STACK_RETURN (REG_EBRACE);
2233 if (syntax & RE_NO_BK_BRACES)
2234 goto unfetch_interval;
2236 FREE_STACK_RETURN (REG_BADBR);
2239 /* We just parsed a valid interval. */
2241 /* If it's invalid to have no preceding re. */
2244 if (syntax & RE_CONTEXT_INVALID_OPS)
2245 FREE_STACK_RETURN (REG_BADRPT);
2246 else if (syntax & RE_CONTEXT_INDEP_OPS)
2249 goto unfetch_interval;
2252 /* If the upper bound is zero, don't want to succeed at
2253 all; jump from `laststart' to `b + 3', which will be
2254 the end of the buffer after we insert the jump. */
2255 if (upper_bound == 0)
2257 GET_BUFFER_SPACE (3);
2258 INSERT_JUMP (jump, laststart, b + 3);
2262 /* Otherwise, we have a nontrivial interval. When
2263 we're all done, the pattern will look like:
2264 set_number_at <jump count> <upper bound>
2265 set_number_at <succeed_n count> <lower bound>
2266 succeed_n <after jump addr> <succeed_n count>
2268 jump_n <succeed_n addr> <jump count>
2269 (The upper bound and `jump_n' are omitted if
2270 `upper_bound' is 1, though.) */
2272 { /* If the upper bound is > 1, we need to insert
2273 more at the end of the loop. */
2274 unsigned nbytes = 10 + (upper_bound > 1) * 10;
2276 GET_BUFFER_SPACE (nbytes);
2278 /* Initialize lower bound of the `succeed_n', even
2279 though it will be set during matching by its
2280 attendant `set_number_at' (inserted next),
2281 because `re_compile_fastmap' needs to know.
2282 Jump to the `jump_n' we might insert below. */
2283 INSERT_JUMP2 (succeed_n, laststart,
2284 b + 5 + (upper_bound > 1) * 5,
2288 /* Code to initialize the lower bound. Insert
2289 before the `succeed_n'. The `5' is the last two
2290 bytes of this `set_number_at', plus 3 bytes of
2291 the following `succeed_n'. */
2292 insert_op2 (set_number_at, laststart, 5, lower_bound, b);
2295 if (upper_bound > 1)
2296 { /* More than one repetition is allowed, so
2297 append a backward jump to the `succeed_n'
2298 that starts this interval.
2300 When we've reached this during matching,
2301 we'll have matched the interval once, so
2302 jump back only `upper_bound - 1' times. */
2303 STORE_JUMP2 (jump_n, b, laststart + 5,
2307 /* The location we want to set is the second
2308 parameter of the `jump_n'; that is `b-2' as
2309 an absolute address. `laststart' will be
2310 the `set_number_at' we're about to insert;
2311 `laststart+3' the number to set, the source
2312 for the relative address. But we are
2313 inserting into the middle of the pattern --
2314 so everything is getting moved up by 5.
2315 Conclusion: (b - 2) - (laststart + 3) + 5,
2316 i.e., b - laststart.
2318 We insert this at the beginning of the loop
2319 so that if we fail during matching, we'll
2320 reinitialize the bounds. */
2321 insert_op2 (set_number_at, laststart, b - laststart,
2322 upper_bound - 1, b);
2327 beg_interval = NULL;
2332 /* If an invalid interval, match the characters as literals. */
2333 assert (beg_interval);
2335 beg_interval = NULL;
2337 /* normal_char and normal_backslash need `c'. */
2340 if (!(syntax & RE_NO_BK_BRACES))
2342 if (p > pattern && p[-1] == '\\')
2343 goto normal_backslash;
2348 /* There is no way to specify the before_dot and after_dot
2349 operators. rms says this is ok. --karl */
2357 BUF_PUSH_2 (syntaxspec, syntax_spec_code[c]);
2363 BUF_PUSH_2 (notsyntaxspec, syntax_spec_code[c]);
2370 BUF_PUSH (wordchar);
2376 BUF_PUSH (notwordchar);
2389 BUF_PUSH (wordbound);
2393 BUF_PUSH (notwordbound);
2404 case '1': case '2': case '3': case '4': case '5':
2405 case '6': case '7': case '8': case '9':
2406 if (syntax & RE_NO_BK_REFS)
2412 FREE_STACK_RETURN (REG_ESUBREG);
2414 /* Can't back reference to a subexpression if inside of it. */
2415 if (group_in_compile_stack (compile_stack, c1))
2419 BUF_PUSH_2 (duplicate, c1);
2425 if (syntax & RE_BK_PLUS_QM)
2428 goto normal_backslash;
2432 /* You might think it would be useful for \ to mean
2433 not to translate; but if we don't translate it
2434 it will never match anything. */
2442 /* Expects the character in `c'. */
2444 /* If no exactn currently being built. */
2447 /* If last exactn not at current position. */
2448 || pending_exact + *pending_exact + 1 != b
2450 /* We have only one byte following the exactn for the count. */
2451 || *pending_exact == (1 << BYTEWIDTH) - 1
2453 /* If followed by a repetition operator. */
2454 || *p == '*' || *p == '^'
2455 || ((syntax & RE_BK_PLUS_QM)
2456 ? *p == '\\' && (p[1] == '+' || p[1] == '?')
2457 : (*p == '+' || *p == '?'))
2458 || ((syntax & RE_INTERVALS)
2459 && ((syntax & RE_NO_BK_BRACES)
2461 : (p[0] == '\\' && p[1] == '{'))))
2463 /* Start building a new exactn. */
2467 BUF_PUSH_2 (exactn, 0);
2468 pending_exact = b - 1;
2475 } /* while p != pend */
2478 /* Through the pattern now. */
2481 STORE_JUMP (jump_past_alt, fixup_alt_jump, b);
2483 if (!COMPILE_STACK_EMPTY)
2484 FREE_STACK_RETURN (REG_EPAREN);
2486 /* If we don't want backtracking, force success
2487 the first time we reach the end of the compiled pattern. */
2488 if (syntax & RE_NO_POSIX_BACKTRACKING)
2491 free (compile_stack.stack);
2493 /* We have succeeded; set the length of the buffer. */
2494 bufp->used = b - bufp->buffer;
2499 DEBUG_PRINT1 ("\nCompiled pattern: \n");
2500 print_compiled_pattern (bufp);
2504 #ifndef MATCH_MAY_ALLOCATE
2505 /* Initialize the failure stack to the largest possible stack. This
2506 isn't necessary unless we're trying to avoid calling alloca in
2507 the search and match routines. */
2509 int num_regs = bufp->re_nsub + 1;
2511 /* Since DOUBLE_FAIL_STACK refuses to double only if the current size
2512 is strictly greater than re_max_failures, the largest possible stack
2513 is 2 * re_max_failures failure points. */
2514 if (fail_stack.size < (2 * re_max_failures * MAX_FAILURE_ITEMS))
2516 fail_stack.size = (2 * re_max_failures * MAX_FAILURE_ITEMS);
2519 if (! fail_stack.stack)
2521 = (fail_stack_elt_t *) xmalloc (fail_stack.size
2522 * sizeof (fail_stack_elt_t));
2525 = (fail_stack_elt_t *) xrealloc (fail_stack.stack,
2527 * sizeof (fail_stack_elt_t)));
2528 #else /* not emacs */
2529 if (! fail_stack.stack)
2531 = (fail_stack_elt_t *) malloc (fail_stack.size
2532 * sizeof (fail_stack_elt_t));
2535 = (fail_stack_elt_t *) realloc (fail_stack.stack,
2537 * sizeof (fail_stack_elt_t)));
2538 #endif /* not emacs */
2541 /* Initialize some other variables the matcher uses. */
2542 RETALLOC_IF (regstart, num_regs, const char *);
2543 RETALLOC_IF (regend, num_regs, const char *);
2544 RETALLOC_IF (old_regstart, num_regs, const char *);
2545 RETALLOC_IF (old_regend, num_regs, const char *);
2546 RETALLOC_IF (best_regstart, num_regs, const char *);
2547 RETALLOC_IF (best_regend, num_regs, const char *);
2548 RETALLOC_IF (reg_info, num_regs, register_info_type);
2549 RETALLOC_IF (reg_dummy, num_regs, const char *);
2550 RETALLOC_IF (reg_info_dummy, num_regs, register_info_type);
2555 } /* regex_compile */
2557 /* Subroutines for `regex_compile'. */
2559 /* Store OP at LOC followed by two-byte integer parameter ARG. */
2562 store_op1 (op, loc, arg)
2567 *loc = (unsigned char) op;
2568 STORE_NUMBER (loc + 1, arg);
2572 /* Like `store_op1', but for two two-byte parameters ARG1 and ARG2. */
2575 store_op2 (op, loc, arg1, arg2)
2580 *loc = (unsigned char) op;
2581 STORE_NUMBER (loc + 1, arg1);
2582 STORE_NUMBER (loc + 3, arg2);
2586 /* Copy the bytes from LOC to END to open up three bytes of space at LOC
2587 for OP followed by two-byte integer parameter ARG. */
2590 insert_op1 (op, loc, arg, end)
2596 register unsigned char *pfrom = end;
2597 register unsigned char *pto = end + 3;
2599 while (pfrom != loc)
2602 store_op1 (op, loc, arg);
2606 /* Like `insert_op1', but for two two-byte parameters ARG1 and ARG2. */
2609 insert_op2 (op, loc, arg1, arg2, end)
2615 register unsigned char *pfrom = end;
2616 register unsigned char *pto = end + 5;
2618 while (pfrom != loc)
2621 store_op2 (op, loc, arg1, arg2);
2625 /* P points to just after a ^ in PATTERN. Return true if that ^ comes
2626 after an alternative or a begin-subexpression. We assume there is at
2627 least one character before the ^. */
2630 at_begline_loc_p (pattern, p, syntax)
2631 const char *pattern, *p;
2632 reg_syntax_t syntax;
2634 const char *prev = p - 2;
2635 boolean prev_prev_backslash = prev > pattern && prev[-1] == '\\';
2638 /* After a subexpression? */
2639 (*prev == '(' && (syntax & RE_NO_BK_PARENS || prev_prev_backslash))
2640 /* After an alternative? */
2641 || (*prev == '|' && (syntax & RE_NO_BK_VBAR || prev_prev_backslash));
2645 /* The dual of at_begline_loc_p. This one is for $. We assume there is
2646 at least one character after the $, i.e., `P < PEND'. */
2649 at_endline_loc_p (p, pend, syntax)
2650 const char *p, *pend;
2653 const char *next = p;
2654 boolean next_backslash = *next == '\\';
2655 const char *next_next = p + 1 < pend ? p + 1 : NULL;
2658 /* Before a subexpression? */
2659 (syntax & RE_NO_BK_PARENS ? *next == ')'
2660 : next_backslash && next_next && *next_next == ')')
2661 /* Before an alternative? */
2662 || (syntax & RE_NO_BK_VBAR ? *next == '|'
2663 : next_backslash && next_next && *next_next == '|');
2667 /* Returns true if REGNUM is in one of COMPILE_STACK's elements and
2668 false if it's not. */
2671 group_in_compile_stack (compile_stack, regnum)
2672 compile_stack_type compile_stack;
2677 for (this_element = compile_stack.avail - 1;
2680 if (compile_stack.stack[this_element].regnum == regnum)
2687 /* Read the ending character of a range (in a bracket expression) from the
2688 uncompiled pattern *P_PTR (which ends at PEND). We assume the
2689 starting character is in `P[-2]'. (`P[-1]' is the character `-'.)
2690 Then we set the translation of all bits between the starting and
2691 ending characters (inclusive) in the compiled pattern B.
2693 Return an error code.
2695 We use these short variable names so we can use the same macros as
2696 `regex_compile' itself. */
2698 static reg_errcode_t
2699 compile_range (p_ptr, pend, translate, syntax, b)
2700 const char **p_ptr, *pend;
2702 reg_syntax_t syntax;
2707 const char *p = *p_ptr;
2708 int range_start, range_end;
2713 /* Even though the pattern is a signed `char *', we need to fetch
2714 with unsigned char *'s; if the high bit of the pattern character
2715 is set, the range endpoints will be negative if we fetch using a
2718 We also want to fetch the endpoints without translating them; the
2719 appropriate translation is done in the bit-setting loop below. */
2720 /* The SVR4 compiler on the 3B2 had trouble with unsigned const char *. */
2721 range_start = ((const unsigned char *) p)[-2];
2722 range_end = ((const unsigned char *) p)[0];
2724 /* Have to increment the pointer into the pattern string, so the
2725 caller isn't still at the ending character. */
2728 /* If the start is after the end, the range is empty. */
2729 if (range_start > range_end)
2730 return syntax & RE_NO_EMPTY_RANGES ? REG_ERANGE : REG_NOERROR;
2732 /* Here we see why `this_char' has to be larger than an `unsigned
2733 char' -- the range is inclusive, so if `range_end' == 0xff
2734 (assuming 8-bit characters), we would otherwise go into an infinite
2735 loop, since all characters <= 0xff. */
2736 for (this_char = range_start; this_char <= range_end; this_char++)
2738 SET_LIST_BIT (TRANSLATE (this_char));
2744 /* re_compile_fastmap computes a ``fastmap'' for the compiled pattern in
2745 BUFP. A fastmap records which of the (1 << BYTEWIDTH) possible
2746 characters can start a string that matches the pattern. This fastmap
2747 is used by re_search to skip quickly over impossible starting points.
2749 The caller must supply the address of a (1 << BYTEWIDTH)-byte data
2750 area as BUFP->fastmap.
2752 We set the `fastmap', `fastmap_accurate', and `can_be_null' fields in
2755 Returns 0 if we succeed, -2 if an internal error. */
2758 re_compile_fastmap (bufp)
2759 struct re_pattern_buffer *bufp;
2762 #ifdef MATCH_MAY_ALLOCATE
2763 fail_stack_type fail_stack;
2765 #ifndef REGEX_MALLOC
2768 /* We don't push any register information onto the failure stack. */
2769 unsigned num_regs = 0;
2771 register char *fastmap = bufp->fastmap;
2772 unsigned char *pattern = bufp->buffer;
2773 unsigned long size = bufp->used;
2774 unsigned char *p = pattern;
2775 register unsigned char *pend = pattern + size;
2777 /* Assume that each path through the pattern can be null until
2778 proven otherwise. We set this false at the bottom of switch
2779 statement, to which we get only if a particular path doesn't
2780 match the empty string. */
2781 boolean path_can_be_null = true;
2783 /* We aren't doing a `succeed_n' to begin with. */
2784 boolean succeed_n_p = false;
2786 assert (fastmap != NULL && p != NULL);
2789 bzero (fastmap, 1 << BYTEWIDTH); /* Assume nothing's valid. */
2790 bufp->fastmap_accurate = 1; /* It will be when we're done. */
2791 bufp->can_be_null = 0;
2795 if (p == pend || *p == succeed)
2797 /* We have reached the (effective) end of pattern. */
2798 if (!FAIL_STACK_EMPTY ())
2800 bufp->can_be_null |= path_can_be_null;
2802 /* Reset for next path. */
2803 path_can_be_null = true;
2805 p = fail_stack.stack[--fail_stack.avail];
2813 /* We should never be about to go beyond the end of the pattern. */
2816 #ifdef SWITCH_ENUM_BUG
2817 switch ((int) ((re_opcode_t) *p++))
2819 switch ((re_opcode_t) *p++)
2823 /* I guess the idea here is to simply not bother with a fastmap
2824 if a backreference is used, since it's too hard to figure out
2825 the fastmap for the corresponding group. Setting
2826 `can_be_null' stops `re_search_2' from using the fastmap, so
2827 that is all we do. */
2829 bufp->can_be_null = 1;
2833 /* Following are the cases which match a character. These end
2842 for (j = *p++ * BYTEWIDTH - 1; j >= 0; j--)
2843 if (p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH)))
2849 /* Chars beyond end of map must be allowed. */
2850 for (j = *p * BYTEWIDTH; j < (1 << BYTEWIDTH); j++)
2853 for (j = *p++ * BYTEWIDTH - 1; j >= 0; j--)
2854 if (!(p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH))))
2860 for (j = 0; j < (1 << BYTEWIDTH); j++)
2861 if (SYNTAX (j) == Sword)
2867 for (j = 0; j < (1 << BYTEWIDTH); j++)
2868 if (SYNTAX (j) != Sword)
2875 int fastmap_newline = fastmap['\n'];
2877 /* `.' matches anything ... */
2878 for (j = 0; j < (1 << BYTEWIDTH); j++)
2881 /* ... except perhaps newline. */
2882 if (!(bufp->syntax & RE_DOT_NEWLINE))
2883 fastmap['\n'] = fastmap_newline;
2885 /* Return if we have already set `can_be_null'; if we have,
2886 then the fastmap is irrelevant. Something's wrong here. */
2887 else if (bufp->can_be_null)
2890 /* Otherwise, have to check alternative paths. */
2897 for (j = 0; j < (1 << BYTEWIDTH); j++)
2898 if (SYNTAX (j) == (enum syntaxcode) k)
2905 for (j = 0; j < (1 << BYTEWIDTH); j++)
2906 if (SYNTAX (j) != (enum syntaxcode) k)
2911 /* All cases after this match the empty string. These end with
2919 #endif /* not emacs */
2931 case push_dummy_failure:
2936 case pop_failure_jump:
2937 case maybe_pop_jump:
2940 case dummy_failure_jump:
2941 EXTRACT_NUMBER_AND_INCR (j, p);
2946 /* Jump backward implies we just went through the body of a
2947 loop and matched nothing. Opcode jumped to should be
2948 `on_failure_jump' or `succeed_n'. Just treat it like an
2949 ordinary jump. For a * loop, it has pushed its failure
2950 point already; if so, discard that as redundant. */
2951 if ((re_opcode_t) *p != on_failure_jump
2952 && (re_opcode_t) *p != succeed_n)
2956 EXTRACT_NUMBER_AND_INCR (j, p);
2959 /* If what's on the stack is where we are now, pop it. */
2960 if (!FAIL_STACK_EMPTY ()
2961 && fail_stack.stack[fail_stack.avail - 1] == p)
2967 case on_failure_jump:
2968 case on_failure_keep_string_jump:
2969 handle_on_failure_jump:
2970 EXTRACT_NUMBER_AND_INCR (j, p);
2972 /* For some patterns, e.g., `(a?)?', `p+j' here points to the
2973 end of the pattern. We don't want to push such a point,
2974 since when we restore it above, entering the switch will
2975 increment `p' past the end of the pattern. We don't need
2976 to push such a point since we obviously won't find any more
2977 fastmap entries beyond `pend'. Such a pattern can match
2978 the null string, though. */
2981 if (!PUSH_PATTERN_OP (p + j, fail_stack))
2985 bufp->can_be_null = 1;
2989 EXTRACT_NUMBER_AND_INCR (k, p); /* Skip the n. */
2990 succeed_n_p = false;
2997 /* Get to the number of times to succeed. */
3000 /* Increment p past the n for when k != 0. */
3001 EXTRACT_NUMBER_AND_INCR (k, p);
3005 succeed_n_p = true; /* Spaghetti code alert. */
3006 goto handle_on_failure_jump;
3023 abort (); /* We have listed all the cases. */
3026 /* Getting here means we have found the possible starting
3027 characters for one path of the pattern -- and that the empty
3028 string does not match. We need not follow this path further.
3029 Instead, look at the next alternative (remembered on the
3030 stack), or quit if no more. The test at the top of the loop
3031 does these things. */
3032 path_can_be_null = false;
3036 /* Set `can_be_null' for the last path (also the first path, if the
3037 pattern is empty). */
3038 bufp->can_be_null |= path_can_be_null;
3040 } /* re_compile_fastmap */
3042 /* Set REGS to hold NUM_REGS registers, storing them in STARTS and
3043 ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use
3044 this memory for recording register information. STARTS and ENDS
3045 must be allocated using the malloc library routine, and must each
3046 be at least NUM_REGS * sizeof (regoff_t) bytes long.
3048 If NUM_REGS == 0, then subsequent matches should allocate their own
3051 Unless this function is called, the first search or match using
3052 PATTERN_BUFFER will allocate its own register data, without
3053 freeing the old data. */
3056 re_set_registers (bufp, regs, num_regs, starts, ends)
3057 struct re_pattern_buffer *bufp;
3058 struct re_registers *regs;
3060 regoff_t *starts, *ends;
3064 bufp->regs_allocated = REGS_REALLOCATE;
3065 regs->num_regs = num_regs;
3066 regs->start = starts;
3071 bufp->regs_allocated = REGS_UNALLOCATED;
3073 regs->start = regs->end = (regoff_t *) 0;
3077 /* Searching routines. */
3079 /* Like re_search_2, below, but only one string is specified, and
3080 doesn't let you say where to stop matching. */
3083 re_search (bufp, string, size, startpos, range, regs)
3084 struct re_pattern_buffer *bufp;
3086 int size, startpos, range;
3087 struct re_registers *regs;
3089 return re_search_2 (bufp, NULL, 0, string, size, startpos, range,
3094 /* Using the compiled pattern in BUFP->buffer, first tries to match the
3095 virtual concatenation of STRING1 and STRING2, starting first at index
3096 STARTPOS, then at STARTPOS + 1, and so on.
3098 STRING1 and STRING2 have length SIZE1 and SIZE2, respectively.
3100 RANGE is how far to scan while trying to match. RANGE = 0 means try
3101 only at STARTPOS; in general, the last start tried is STARTPOS +
3104 In REGS, return the indices of the virtual concatenation of STRING1
3105 and STRING2 that matched the entire BUFP->buffer and its contained
3108 Do not consider matching one past the index STOP in the virtual
3109 concatenation of STRING1 and STRING2.
3111 We return either the position in the strings at which the match was
3112 found, -1 if no match, or -2 if error (such as failure
3116 re_search_2 (bufp, string1, size1, string2, size2, startpos, range, regs, stop)
3117 struct re_pattern_buffer *bufp;
3118 const char *string1, *string2;
3122 struct re_registers *regs;
3126 register char *fastmap = bufp->fastmap;
3127 register char *translate = bufp->translate;
3128 int total_size = size1 + size2;
3129 int endpos = startpos + range;
3131 /* Check for out-of-range STARTPOS. */
3132 if (startpos < 0 || startpos > total_size)
3135 /* Fix up RANGE if it might eventually take us outside
3136 the virtual concatenation of STRING1 and STRING2. */
3138 range = -1 - startpos;
3139 else if (endpos > total_size)
3140 range = total_size - startpos;
3142 /* If the search isn't to be a backwards one, don't waste time in a
3143 search for a pattern that must be anchored. */
3144 if (bufp->used > 0 && (re_opcode_t) bufp->buffer[0] == begbuf && range > 0)
3152 /* Update the fastmap now if not correct already. */
3153 if (fastmap && !bufp->fastmap_accurate)
3154 if (re_compile_fastmap (bufp) == -2)
3157 /* Loop through the string, looking for a place to start matching. */
3160 /* If a fastmap is supplied, skip quickly over characters that
3161 cannot be the start of a match. If the pattern can match the
3162 null string, however, we don't need to skip characters; we want
3163 the first null string. */
3164 if (fastmap && startpos < total_size && !bufp->can_be_null)
3166 if (range > 0) /* Searching forwards. */
3168 register const char *d;
3169 register int lim = 0;
3172 if (startpos < size1 && startpos + range >= size1)
3173 lim = range - (size1 - startpos);
3175 d = (startpos >= size1 ? string2 - size1 : string1) + startpos;
3177 /* Written out as an if-else to avoid testing `translate'
3181 && !fastmap[(unsigned char)
3182 translate[(unsigned char) *d++]])
3185 while (range > lim && !fastmap[(unsigned char) *d++])
3188 startpos += irange - range;
3190 else /* Searching backwards. */
3192 register char c = (size1 == 0 || startpos >= size1
3193 ? string2[startpos - size1]
3194 : string1[startpos]);
3196 if (!fastmap[(unsigned char) TRANSLATE (c)])
3201 /* If can't match the null string, and that's all we have left, fail. */
3202 if (range >= 0 && startpos == total_size && fastmap
3203 && !bufp->can_be_null)
3206 val = re_match_2_internal (bufp, string1, size1, string2, size2,
3207 startpos, regs, stop);
3208 #ifndef REGEX_MALLOC
3237 /* Declarations and macros for re_match_2. */
3239 static int bcmp_translate ();
3240 static boolean alt_match_null_string_p (),
3241 common_op_match_null_string_p (),
3242 group_match_null_string_p ();
3244 /* This converts PTR, a pointer into one of the search strings `string1'
3245 and `string2' into an offset from the beginning of that string. */
3246 #define POINTER_TO_OFFSET(ptr) \
3247 (FIRST_STRING_P (ptr) \
3248 ? ((regoff_t) ((ptr) - string1)) \
3249 : ((regoff_t) ((ptr) - string2 + size1)))
3251 /* Macros for dealing with the split strings in re_match_2. */
3253 #define MATCHING_IN_FIRST_STRING (dend == end_match_1)
3255 /* Call before fetching a character with *d. This switches over to
3256 string2 if necessary. */
3257 #define PREFETCH() \
3260 /* End of string2 => fail. */ \
3261 if (dend == end_match_2) \
3263 /* End of string1 => advance to string2. */ \
3265 dend = end_match_2; \
3269 /* Test if at very beginning or at very end of the virtual concatenation
3270 of `string1' and `string2'. If only one string, it's `string2'. */
3271 #define AT_STRINGS_BEG(d) ((d) == (size1 ? string1 : string2) || !size2)
3272 #define AT_STRINGS_END(d) ((d) == end2)
3275 /* Test if D points to a character which is word-constituent. We have
3276 two special cases to check for: if past the end of string1, look at
3277 the first character in string2; and if before the beginning of
3278 string2, look at the last character in string1. */
3279 #define WORDCHAR_P(d) \
3280 (SYNTAX ((d) == end1 ? *string2 \
3281 : (d) == string2 - 1 ? *(end1 - 1) : *(d)) \
3284 /* Test if the character before D and the one at D differ with respect
3285 to being word-constituent. */
3286 #define AT_WORD_BOUNDARY(d) \
3287 (AT_STRINGS_BEG (d) || AT_STRINGS_END (d) \
3288 || WORDCHAR_P (d - 1) != WORDCHAR_P (d))
3291 /* Free everything we malloc. */
3292 #ifdef MATCH_MAY_ALLOCATE
3294 #define FREE_VAR(var) if (var) free (var); var = NULL
3295 #define FREE_VARIABLES() \
3297 FREE_VAR (fail_stack.stack); \
3298 FREE_VAR (regstart); \
3299 FREE_VAR (regend); \
3300 FREE_VAR (old_regstart); \
3301 FREE_VAR (old_regend); \
3302 FREE_VAR (best_regstart); \
3303 FREE_VAR (best_regend); \
3304 FREE_VAR (reg_info); \
3305 FREE_VAR (reg_dummy); \
3306 FREE_VAR (reg_info_dummy); \
3308 #else /* not REGEX_MALLOC */
3309 /* This used to do alloca (0), but now we do that in the caller. */
3310 #define FREE_VARIABLES() /* Nothing */
3311 #endif /* not REGEX_MALLOC */
3313 #define FREE_VARIABLES() /* Do nothing! */
3314 #endif /* not MATCH_MAY_ALLOCATE */
3316 /* These values must meet several constraints. They must not be valid
3317 register values; since we have a limit of 255 registers (because
3318 we use only one byte in the pattern for the register number), we can
3319 use numbers larger than 255. They must differ by 1, because of
3320 NUM_FAILURE_ITEMS above. And the value for the lowest register must
3321 be larger than the value for the highest register, so we do not try
3322 to actually save any registers when none are active. */
3323 #define NO_HIGHEST_ACTIVE_REG (1 << BYTEWIDTH)
3324 #define NO_LOWEST_ACTIVE_REG (NO_HIGHEST_ACTIVE_REG + 1)
3326 /* Matching routines. */
3328 #ifndef emacs /* Emacs never uses this. */
3329 /* re_match is like re_match_2 except it takes only a single string. */
3332 re_match (bufp, string, size, pos, regs)
3333 struct re_pattern_buffer *bufp;
3336 struct re_registers *regs;
3338 int result = re_match_2_internal (bufp, NULL, 0, string, size,
3343 #endif /* not emacs */
3346 /* re_match_2 matches the compiled pattern in BUFP against the
3347 the (virtual) concatenation of STRING1 and STRING2 (of length SIZE1
3348 and SIZE2, respectively). We start matching at POS, and stop
3351 If REGS is non-null and the `no_sub' field of BUFP is nonzero, we
3352 store offsets for the substring each group matched in REGS. See the
3353 documentation for exactly how many groups we fill.
3355 We return -1 if no match, -2 if an internal error (such as the
3356 failure stack overflowing). Otherwise, we return the length of the
3357 matched substring. */
3360 re_match_2 (bufp, string1, size1, string2, size2, pos, regs, stop)
3361 struct re_pattern_buffer *bufp;
3362 const char *string1, *string2;
3365 struct re_registers *regs;
3368 int result = re_match_2_internal (bufp, string1, size1, string2, size2,
3374 /* This is a separate function so that we can force an alloca cleanup
3377 re_match_2_internal (bufp, string1, size1, string2, size2, pos, regs, stop)
3378 struct re_pattern_buffer *bufp;
3379 const char *string1, *string2;
3382 struct re_registers *regs;
3385 /* General temporaries. */
3389 /* Just past the end of the corresponding string. */
3390 const char *end1, *end2;
3392 /* Pointers into string1 and string2, just past the last characters in
3393 each to consider matching. */
3394 const char *end_match_1, *end_match_2;
3396 /* Where we are in the data, and the end of the current string. */
3397 const char *d, *dend;
3399 /* Where we are in the pattern, and the end of the pattern. */
3400 unsigned char *p = bufp->buffer;
3401 register unsigned char *pend = p + bufp->used;
3403 /* Mark the opcode just after a start_memory, so we can test for an
3404 empty subpattern when we get to the stop_memory. */
3405 unsigned char *just_past_start_mem = 0;
3407 /* We use this to map every character in the string. */
3408 char *translate = bufp->translate;
3410 /* Failure point stack. Each place that can handle a failure further
3411 down the line pushes a failure point on this stack. It consists of
3412 restart, regend, and reg_info for all registers corresponding to
3413 the subexpressions we're currently inside, plus the number of such
3414 registers, and, finally, two char *'s. The first char * is where
3415 to resume scanning the pattern; the second one is where to resume
3416 scanning the strings. If the latter is zero, the failure point is
3417 a ``dummy''; if a failure happens and the failure point is a dummy,
3418 it gets discarded and the next next one is tried. */
3419 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
3420 fail_stack_type fail_stack;
3423 static unsigned failure_id = 0;
3424 unsigned nfailure_points_pushed = 0, nfailure_points_popped = 0;
3427 /* We fill all the registers internally, independent of what we
3428 return, for use in backreferences. The number here includes
3429 an element for register zero. */
3430 unsigned num_regs = bufp->re_nsub + 1;
3432 /* The currently active registers. */
3433 unsigned lowest_active_reg = NO_LOWEST_ACTIVE_REG;
3434 unsigned highest_active_reg = NO_HIGHEST_ACTIVE_REG;
3436 /* Information on the contents of registers. These are pointers into
3437 the input strings; they record just what was matched (on this
3438 attempt) by a subexpression part of the pattern, that is, the
3439 regnum-th regstart pointer points to where in the pattern we began
3440 matching and the regnum-th regend points to right after where we
3441 stopped matching the regnum-th subexpression. (The zeroth register
3442 keeps track of what the whole pattern matches.) */
3443 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
3444 const char **regstart, **regend;
3447 /* If a group that's operated upon by a repetition operator fails to
3448 match anything, then the register for its start will need to be
3449 restored because it will have been set to wherever in the string we
3450 are when we last see its open-group operator. Similarly for a
3452 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
3453 const char **old_regstart, **old_regend;
3456 /* The is_active field of reg_info helps us keep track of which (possibly
3457 nested) subexpressions we are currently in. The matched_something
3458 field of reg_info[reg_num] helps us tell whether or not we have
3459 matched any of the pattern so far this time through the reg_num-th
3460 subexpression. These two fields get reset each time through any
3461 loop their register is in. */
3462 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
3463 register_info_type *reg_info;
3466 /* The following record the register info as found in the above
3467 variables when we find a match better than any we've seen before.
3468 This happens as we backtrack through the failure points, which in
3469 turn happens only if we have not yet matched the entire string. */
3470 unsigned best_regs_set = false;
3471 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
3472 const char **best_regstart, **best_regend;
3475 /* Logically, this is `best_regend[0]'. But we don't want to have to
3476 allocate space for that if we're not allocating space for anything
3477 else (see below). Also, we never need info about register 0 for
3478 any of the other register vectors, and it seems rather a kludge to
3479 treat `best_regend' differently than the rest. So we keep track of
3480 the end of the best match so far in a separate variable. We
3481 initialize this to NULL so that when we backtrack the first time
3482 and need to test it, it's not garbage. */
3483 const char *match_end = NULL;
3485 /* Used when we pop values we don't care about. */
3486 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
3487 const char **reg_dummy;
3488 register_info_type *reg_info_dummy;
3492 /* Counts the total number of registers pushed. */
3493 unsigned num_regs_pushed = 0;
3496 DEBUG_PRINT1 ("\n\nEntering re_match_2.\n");
3500 #ifdef MATCH_MAY_ALLOCATE
3501 /* Do not bother to initialize all the register variables if there are
3502 no groups in the pattern, as it takes a fair amount of time. If
3503 there are groups, we include space for register 0 (the whole
3504 pattern), even though we never use it, since it simplifies the
3505 array indexing. We should fix this. */
3508 regstart = REGEX_TALLOC (num_regs, const char *);
3509 regend = REGEX_TALLOC (num_regs, const char *);
3510 old_regstart = REGEX_TALLOC (num_regs, const char *);
3511 old_regend = REGEX_TALLOC (num_regs, const char *);
3512 best_regstart = REGEX_TALLOC (num_regs, const char *);
3513 best_regend = REGEX_TALLOC (num_regs, const char *);
3514 reg_info = REGEX_TALLOC (num_regs, register_info_type);
3515 reg_dummy = REGEX_TALLOC (num_regs, const char *);
3516 reg_info_dummy = REGEX_TALLOC (num_regs, register_info_type);
3518 if (!(regstart && regend && old_regstart && old_regend && reg_info
3519 && best_regstart && best_regend && reg_dummy && reg_info_dummy))
3525 #if defined (REGEX_MALLOC)
3528 /* We must initialize all our variables to NULL, so that
3529 `FREE_VARIABLES' doesn't try to free them. */
3530 regstart = regend = old_regstart = old_regend = best_regstart
3531 = best_regend = reg_dummy = NULL;
3532 reg_info = reg_info_dummy = (register_info_type *) NULL;
3534 #endif /* REGEX_MALLOC */
3535 #endif /* MATCH_MAY_ALLOCATE */
3537 /* The starting position is bogus. */
3538 if (pos < 0 || pos > size1 + size2)
3544 /* Initialize subexpression text positions to -1 to mark ones that no
3545 start_memory/stop_memory has been seen for. Also initialize the
3546 register information struct. */
3547 for (mcnt = 1; mcnt < num_regs; mcnt++)
3549 regstart[mcnt] = regend[mcnt]
3550 = old_regstart[mcnt] = old_regend[mcnt] = REG_UNSET_VALUE;
3552 REG_MATCH_NULL_STRING_P (reg_info[mcnt]) = MATCH_NULL_UNSET_VALUE;
3553 IS_ACTIVE (reg_info[mcnt]) = 0;
3554 MATCHED_SOMETHING (reg_info[mcnt]) = 0;
3555 EVER_MATCHED_SOMETHING (reg_info[mcnt]) = 0;
3558 /* We move `string1' into `string2' if the latter's empty -- but not if
3559 `string1' is null. */
3560 if (size2 == 0 && string1 != NULL)
3567 end1 = string1 + size1;
3568 end2 = string2 + size2;
3570 /* Compute where to stop matching, within the two strings. */
3573 end_match_1 = string1 + stop;
3574 end_match_2 = string2;
3579 end_match_2 = string2 + stop - size1;
3582 /* `p' scans through the pattern as `d' scans through the data.
3583 `dend' is the end of the input string that `d' points within. `d'
3584 is advanced into the following input string whenever necessary, but
3585 this happens before fetching; therefore, at the beginning of the
3586 loop, `d' can be pointing at the end of a string, but it cannot
3588 if (size1 > 0 && pos <= size1)
3595 d = string2 + pos - size1;
3599 DEBUG_PRINT1 ("The compiled pattern is: ");
3600 DEBUG_PRINT_COMPILED_PATTERN (bufp, p, pend);
3601 DEBUG_PRINT1 ("The string to match is: `");
3602 DEBUG_PRINT_DOUBLE_STRING (d, string1, size1, string2, size2);
3603 DEBUG_PRINT1 ("'\n");
3605 /* This loops over pattern commands. It exits by returning from the
3606 function if the match is complete, or it drops through if the match
3607 fails at this starting point in the input data. */
3610 DEBUG_PRINT2 ("\n0x%x: ", p);
3613 { /* End of pattern means we might have succeeded. */
3614 DEBUG_PRINT1 ("end of pattern ... ");
3616 /* If we haven't matched the entire string, and we want the
3617 longest match, try backtracking. */
3618 if (d != end_match_2)
3620 /* 1 if this match ends in the same string (string1 or string2)
3621 as the best previous match. */
3622 boolean same_str_p = (FIRST_STRING_P (match_end)
3623 == MATCHING_IN_FIRST_STRING);
3624 /* 1 if this match is the best seen so far. */
3625 boolean best_match_p;
3627 /* AIX compiler got confused when this was combined
3628 with the previous declaration. */
3630 best_match_p = d > match_end;
3632 best_match_p = !MATCHING_IN_FIRST_STRING;
3634 DEBUG_PRINT1 ("backtracking.\n");
3636 if (!FAIL_STACK_EMPTY ())
3637 { /* More failure points to try. */
3639 /* If exceeds best match so far, save it. */
3640 if (!best_regs_set || best_match_p)
3642 best_regs_set = true;
3645 DEBUG_PRINT1 ("\nSAVING match as best so far.\n");
3647 for (mcnt = 1; mcnt < num_regs; mcnt++)
3649 best_regstart[mcnt] = regstart[mcnt];
3650 best_regend[mcnt] = regend[mcnt];
3656 /* If no failure points, don't restore garbage. And if
3657 last match is real best match, don't restore second
3659 else if (best_regs_set && !best_match_p)
3662 /* Restore best match. It may happen that `dend ==
3663 end_match_1' while the restored d is in string2.
3664 For example, the pattern `x.*y.*z' against the
3665 strings `x-' and `y-z-', if the two strings are
3666 not consecutive in memory. */
3667 DEBUG_PRINT1 ("Restoring best registers.\n");
3670 dend = ((d >= string1 && d <= end1)
3671 ? end_match_1 : end_match_2);
3673 for (mcnt = 1; mcnt < num_regs; mcnt++)
3675 regstart[mcnt] = best_regstart[mcnt];
3676 regend[mcnt] = best_regend[mcnt];
3679 } /* d != end_match_2 */
3682 DEBUG_PRINT1 ("Accepting match.\n");
3684 /* If caller wants register contents data back, do it. */
3685 if (regs && !bufp->no_sub)
3687 /* Have the register data arrays been allocated? */
3688 if (bufp->regs_allocated == REGS_UNALLOCATED)
3689 { /* No. So allocate them with malloc. We need one
3690 extra element beyond `num_regs' for the `-1' marker
3692 regs->num_regs = MAX (RE_NREGS, num_regs + 1);
3693 regs->start = TALLOC (regs->num_regs, regoff_t);
3694 regs->end = TALLOC (regs->num_regs, regoff_t);
3695 if (regs->start == NULL || regs->end == NULL)
3697 bufp->regs_allocated = REGS_REALLOCATE;
3699 else if (bufp->regs_allocated == REGS_REALLOCATE)
3700 { /* Yes. If we need more elements than were already
3701 allocated, reallocate them. If we need fewer, just
3703 if (regs->num_regs < num_regs + 1)
3705 regs->num_regs = num_regs + 1;
3706 RETALLOC (regs->start, regs->num_regs, regoff_t);
3707 RETALLOC (regs->end, regs->num_regs, regoff_t);
3708 if (regs->start == NULL || regs->end == NULL)
3714 /* These braces fend off a "empty body in an else-statement"
3715 warning under GCC when assert expands to nothing. */
3716 assert (bufp->regs_allocated == REGS_FIXED);
3719 /* Convert the pointer data in `regstart' and `regend' to
3720 indices. Register zero has to be set differently,
3721 since we haven't kept track of any info for it. */
3722 if (regs->num_regs > 0)
3724 regs->start[0] = pos;
3725 regs->end[0] = (MATCHING_IN_FIRST_STRING
3726 ? ((regoff_t) (d - string1))
3727 : ((regoff_t) (d - string2 + size1)));
3730 /* Go through the first `min (num_regs, regs->num_regs)'
3731 registers, since that is all we initialized. */
3732 for (mcnt = 1; mcnt < MIN (num_regs, regs->num_regs); mcnt++)
3734 if (REG_UNSET (regstart[mcnt]) || REG_UNSET (regend[mcnt]))
3735 regs->start[mcnt] = regs->end[mcnt] = -1;
3739 = (regoff_t) POINTER_TO_OFFSET (regstart[mcnt]);
3741 = (regoff_t) POINTER_TO_OFFSET (regend[mcnt]);
3745 /* If the regs structure we return has more elements than
3746 were in the pattern, set the extra elements to -1. If
3747 we (re)allocated the registers, this is the case,
3748 because we always allocate enough to have at least one
3750 for (mcnt = num_regs; mcnt < regs->num_regs; mcnt++)
3751 regs->start[mcnt] = regs->end[mcnt] = -1;
3752 } /* regs && !bufp->no_sub */
3755 DEBUG_PRINT4 ("%u failure points pushed, %u popped (%u remain).\n",
3756 nfailure_points_pushed, nfailure_points_popped,
3757 nfailure_points_pushed - nfailure_points_popped);
3758 DEBUG_PRINT2 ("%u registers pushed.\n", num_regs_pushed);
3760 mcnt = d - pos - (MATCHING_IN_FIRST_STRING
3764 DEBUG_PRINT2 ("Returning %d from re_match_2.\n", mcnt);
3769 /* Otherwise match next pattern command. */
3770 #ifdef SWITCH_ENUM_BUG
3771 switch ((int) ((re_opcode_t) *p++))
3773 switch ((re_opcode_t) *p++)
3776 /* Ignore these. Used to ignore the n of succeed_n's which
3777 currently have n == 0. */
3779 DEBUG_PRINT1 ("EXECUTING no_op.\n");
3783 DEBUG_PRINT1 ("EXECUTING succeed.\n");
3786 /* Match the next n pattern characters exactly. The following
3787 byte in the pattern defines n, and the n bytes after that
3788 are the characters to match. */
3791 DEBUG_PRINT2 ("EXECUTING exactn %d.\n", mcnt);
3793 /* This is written out as an if-else so we don't waste time
3794 testing `translate' inside the loop. */
3800 if (translate[(unsigned char) *d++] != (char) *p++)
3810 if (*d++ != (char) *p++) goto fail;
3814 SET_REGS_MATCHED ();
3818 /* Match any character except possibly a newline or a null. */
3820 DEBUG_PRINT1 ("EXECUTING anychar.\n");
3824 if ((!(bufp->syntax & RE_DOT_NEWLINE) && TRANSLATE (*d) == '\n')
3825 || (bufp->syntax & RE_DOT_NOT_NULL && TRANSLATE (*d) == '\000'))
3828 SET_REGS_MATCHED ();
3829 DEBUG_PRINT2 (" Matched `%d'.\n", *d);
3837 register unsigned char c;
3838 boolean not = (re_opcode_t) *(p - 1) == charset_not;
3840 DEBUG_PRINT2 ("EXECUTING charset%s.\n", not ? "_not" : "");
3843 c = TRANSLATE (*d); /* The character to match. */
3845 /* Cast to `unsigned' instead of `unsigned char' in case the
3846 bit list is a full 32 bytes long. */
3847 if (c < (unsigned) (*p * BYTEWIDTH)
3848 && p[1 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH)))
3853 if (!not) goto fail;
3855 SET_REGS_MATCHED ();
3861 /* The beginning of a group is represented by start_memory.
3862 The arguments are the register number in the next byte, and the
3863 number of groups inner to this one in the next. The text
3864 matched within the group is recorded (in the internal
3865 registers data structure) under the register number. */
3867 DEBUG_PRINT3 ("EXECUTING start_memory %d (%d):\n", *p, p[1]);
3869 /* Find out if this group can match the empty string. */
3870 p1 = p; /* To send to group_match_null_string_p. */
3872 if (REG_MATCH_NULL_STRING_P (reg_info[*p]) == MATCH_NULL_UNSET_VALUE)
3873 REG_MATCH_NULL_STRING_P (reg_info[*p])
3874 = group_match_null_string_p (&p1, pend, reg_info);
3876 /* Save the position in the string where we were the last time
3877 we were at this open-group operator in case the group is
3878 operated upon by a repetition operator, e.g., with `(a*)*b'
3879 against `ab'; then we want to ignore where we are now in
3880 the string in case this attempt to match fails. */
3881 old_regstart[*p] = REG_MATCH_NULL_STRING_P (reg_info[*p])
3882 ? REG_UNSET (regstart[*p]) ? d : regstart[*p]
3884 DEBUG_PRINT2 (" old_regstart: %d\n",
3885 POINTER_TO_OFFSET (old_regstart[*p]));
3888 DEBUG_PRINT2 (" regstart: %d\n", POINTER_TO_OFFSET (regstart[*p]));
3890 IS_ACTIVE (reg_info[*p]) = 1;
3891 MATCHED_SOMETHING (reg_info[*p]) = 0;
3893 /* This is the new highest active register. */
3894 highest_active_reg = *p;
3896 /* If nothing was active before, this is the new lowest active
3898 if (lowest_active_reg == NO_LOWEST_ACTIVE_REG)
3899 lowest_active_reg = *p;
3901 /* Move past the register number and inner group count. */
3903 just_past_start_mem = p;
3907 /* The stop_memory opcode represents the end of a group. Its
3908 arguments are the same as start_memory's: the register
3909 number, and the number of inner groups. */
3911 DEBUG_PRINT3 ("EXECUTING stop_memory %d (%d):\n", *p, p[1]);
3913 /* We need to save the string position the last time we were at
3914 this close-group operator in case the group is operated
3915 upon by a repetition operator, e.g., with `((a*)*(b*)*)*'
3916 against `aba'; then we want to ignore where we are now in
3917 the string in case this attempt to match fails. */
3918 old_regend[*p] = REG_MATCH_NULL_STRING_P (reg_info[*p])
3919 ? REG_UNSET (regend[*p]) ? d : regend[*p]
3921 DEBUG_PRINT2 (" old_regend: %d\n",
3922 POINTER_TO_OFFSET (old_regend[*p]));
3925 DEBUG_PRINT2 (" regend: %d\n", POINTER_TO_OFFSET (regend[*p]));
3927 /* This register isn't active anymore. */
3928 IS_ACTIVE (reg_info[*p]) = 0;
3930 /* If this was the only register active, nothing is active
3932 if (lowest_active_reg == highest_active_reg)
3934 lowest_active_reg = NO_LOWEST_ACTIVE_REG;
3935 highest_active_reg = NO_HIGHEST_ACTIVE_REG;
3938 { /* We must scan for the new highest active register, since
3939 it isn't necessarily one less than now: consider
3940 (a(b)c(d(e)f)g). When group 3 ends, after the f), the
3941 new highest active register is 1. */
3942 unsigned char r = *p - 1;
3943 while (r > 0 && !IS_ACTIVE (reg_info[r]))
3946 /* If we end up at register zero, that means that we saved
3947 the registers as the result of an `on_failure_jump', not
3948 a `start_memory', and we jumped to past the innermost
3949 `stop_memory'. For example, in ((.)*) we save
3950 registers 1 and 2 as a result of the *, but when we pop
3951 back to the second ), we are at the stop_memory 1.
3952 Thus, nothing is active. */
3955 lowest_active_reg = NO_LOWEST_ACTIVE_REG;
3956 highest_active_reg = NO_HIGHEST_ACTIVE_REG;
3959 highest_active_reg = r;
3962 /* If just failed to match something this time around with a
3963 group that's operated on by a repetition operator, try to
3964 force exit from the ``loop'', and restore the register
3965 information for this group that we had before trying this
3967 if ((!MATCHED_SOMETHING (reg_info[*p])
3968 || just_past_start_mem == p - 1)
3971 boolean is_a_jump_n = false;
3975 switch ((re_opcode_t) *p1++)
3979 case pop_failure_jump:
3980 case maybe_pop_jump:
3982 case dummy_failure_jump:
3983 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
3993 /* If the next operation is a jump backwards in the pattern
3994 to an on_failure_jump right before the start_memory
3995 corresponding to this stop_memory, exit from the loop
3996 by forcing a failure after pushing on the stack the
3997 on_failure_jump's jump in the pattern, and d. */
3998 if (mcnt < 0 && (re_opcode_t) *p1 == on_failure_jump
3999 && (re_opcode_t) p1[3] == start_memory && p1[4] == *p)
4001 /* If this group ever matched anything, then restore
4002 what its registers were before trying this last
4003 failed match, e.g., with `(a*)*b' against `ab' for
4004 regstart[1], and, e.g., with `((a*)*(b*)*)*'
4005 against `aba' for regend[3].
4007 Also restore the registers for inner groups for,
4008 e.g., `((a*)(b*))*' against `aba' (register 3 would
4009 otherwise get trashed). */
4011 if (EVER_MATCHED_SOMETHING (reg_info[*p]))
4015 EVER_MATCHED_SOMETHING (reg_info[*p]) = 0;
4017 /* Restore this and inner groups' (if any) registers. */
4018 for (r = *p; r < *p + *(p + 1); r++)
4020 regstart[r] = old_regstart[r];
4022 /* xx why this test? */
4023 if ((int) old_regend[r] >= (int) regstart[r])
4024 regend[r] = old_regend[r];
4028 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4029 PUSH_FAILURE_POINT (p1 + mcnt, d, -2);
4035 /* Move past the register number and the inner group count. */
4040 /* \<digit> has been turned into a `duplicate' command which is
4041 followed by the numeric value of <digit> as the register number. */
4044 register const char *d2, *dend2;
4045 int regno = *p++; /* Get which register to match against. */
4046 DEBUG_PRINT2 ("EXECUTING duplicate %d.\n", regno);
4048 /* Can't back reference a group which we've never matched. */
4049 if (REG_UNSET (regstart[regno]) || REG_UNSET (regend[regno]))
4052 /* Where in input to try to start matching. */
4053 d2 = regstart[regno];
4055 /* Where to stop matching; if both the place to start and
4056 the place to stop matching are in the same string, then
4057 set to the place to stop, otherwise, for now have to use
4058 the end of the first string. */
4060 dend2 = ((FIRST_STRING_P (regstart[regno])
4061 == FIRST_STRING_P (regend[regno]))
4062 ? regend[regno] : end_match_1);
4065 /* If necessary, advance to next segment in register
4069 if (dend2 == end_match_2) break;
4070 if (dend2 == regend[regno]) break;
4072 /* End of string1 => advance to string2. */
4074 dend2 = regend[regno];
4076 /* At end of register contents => success */
4077 if (d2 == dend2) break;
4079 /* If necessary, advance to next segment in data. */
4082 /* How many characters left in this segment to match. */
4085 /* Want how many consecutive characters we can match in
4086 one shot, so, if necessary, adjust the count. */
4087 if (mcnt > dend2 - d2)
4090 /* Compare that many; failure if mismatch, else move
4093 ? bcmp_translate (d, d2, mcnt, translate)
4094 : bcmp (d, d2, mcnt))
4096 d += mcnt, d2 += mcnt;
4102 /* begline matches the empty string at the beginning of the string
4103 (unless `not_bol' is set in `bufp'), and, if
4104 `newline_anchor' is set, after newlines. */
4106 DEBUG_PRINT1 ("EXECUTING begline.\n");
4108 if (AT_STRINGS_BEG (d))
4110 if (!bufp->not_bol) break;
4112 else if (d[-1] == '\n' && bufp->newline_anchor)
4116 /* In all other cases, we fail. */
4120 /* endline is the dual of begline. */
4122 DEBUG_PRINT1 ("EXECUTING endline.\n");
4124 if (AT_STRINGS_END (d))
4126 if (!bufp->not_eol) break;
4129 /* We have to ``prefetch'' the next character. */
4130 else if ((d == end1 ? *string2 : *d) == '\n'
4131 && bufp->newline_anchor)
4138 /* Match at the very beginning of the data. */
4140 DEBUG_PRINT1 ("EXECUTING begbuf.\n");
4141 if (AT_STRINGS_BEG (d))
4146 /* Match at the very end of the data. */
4148 DEBUG_PRINT1 ("EXECUTING endbuf.\n");
4149 if (AT_STRINGS_END (d))
4154 /* on_failure_keep_string_jump is used to optimize `.*\n'. It
4155 pushes NULL as the value for the string on the stack. Then
4156 `pop_failure_point' will keep the current value for the
4157 string, instead of restoring it. To see why, consider
4158 matching `foo\nbar' against `.*\n'. The .* matches the foo;
4159 then the . fails against the \n. But the next thing we want
4160 to do is match the \n against the \n; if we restored the
4161 string value, we would be back at the foo.
4163 Because this is used only in specific cases, we don't need to
4164 check all the things that `on_failure_jump' does, to make
4165 sure the right things get saved on the stack. Hence we don't
4166 share its code. The only reason to push anything on the
4167 stack at all is that otherwise we would have to change
4168 `anychar's code to do something besides goto fail in this
4169 case; that seems worse than this. */
4170 case on_failure_keep_string_jump:
4171 DEBUG_PRINT1 ("EXECUTING on_failure_keep_string_jump");
4173 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4174 DEBUG_PRINT3 (" %d (to 0x%x):\n", mcnt, p + mcnt);
4176 PUSH_FAILURE_POINT (p + mcnt, NULL, -2);
4180 /* Uses of on_failure_jump:
4182 Each alternative starts with an on_failure_jump that points
4183 to the beginning of the next alternative. Each alternative
4184 except the last ends with a jump that in effect jumps past
4185 the rest of the alternatives. (They really jump to the
4186 ending jump of the following alternative, because tensioning
4187 these jumps is a hassle.)
4189 Repeats start with an on_failure_jump that points past both
4190 the repetition text and either the following jump or
4191 pop_failure_jump back to this on_failure_jump. */
4192 case on_failure_jump:
4194 DEBUG_PRINT1 ("EXECUTING on_failure_jump");
4196 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4197 DEBUG_PRINT3 (" %d (to 0x%x)", mcnt, p + mcnt);
4199 /* If this on_failure_jump comes right before a group (i.e.,
4200 the original * applied to a group), save the information
4201 for that group and all inner ones, so that if we fail back
4202 to this point, the group's information will be correct.
4203 For example, in \(a*\)*\1, we need the preceding group,
4204 and in \(\(a*\)b*\)\2, we need the inner group. */
4206 /* We can't use `p' to check ahead because we push
4207 a failure point to `p + mcnt' after we do this. */
4210 /* We need to skip no_op's before we look for the
4211 start_memory in case this on_failure_jump is happening as
4212 the result of a completed succeed_n, as in \(a\)\{1,3\}b\1
4214 while (p1 < pend && (re_opcode_t) *p1 == no_op)
4217 if (p1 < pend && (re_opcode_t) *p1 == start_memory)
4219 /* We have a new highest active register now. This will
4220 get reset at the start_memory we are about to get to,
4221 but we will have saved all the registers relevant to
4222 this repetition op, as described above. */
4223 highest_active_reg = *(p1 + 1) + *(p1 + 2);
4224 if (lowest_active_reg == NO_LOWEST_ACTIVE_REG)
4225 lowest_active_reg = *(p1 + 1);
4228 DEBUG_PRINT1 (":\n");
4229 PUSH_FAILURE_POINT (p + mcnt, d, -2);
4233 /* A smart repeat ends with `maybe_pop_jump'.
4234 We change it to either `pop_failure_jump' or `jump'. */
4235 case maybe_pop_jump:
4236 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4237 DEBUG_PRINT2 ("EXECUTING maybe_pop_jump %d.\n", mcnt);
4239 register unsigned char *p2 = p;
4241 /* Compare the beginning of the repeat with what in the
4242 pattern follows its end. If we can establish that there
4243 is nothing that they would both match, i.e., that we
4244 would have to backtrack because of (as in, e.g., `a*a')
4245 then we can change to pop_failure_jump, because we'll
4246 never have to backtrack.
4248 This is not true in the case of alternatives: in
4249 `(a|ab)*' we do need to backtrack to the `ab' alternative
4250 (e.g., if the string was `ab'). But instead of trying to
4251 detect that here, the alternative has put on a dummy
4252 failure point which is what we will end up popping. */
4254 /* Skip over open/close-group commands.
4255 If what follows this loop is a ...+ construct,
4256 look at what begins its body, since we will have to
4257 match at least one of that. */
4261 && ((re_opcode_t) *p2 == stop_memory
4262 || (re_opcode_t) *p2 == start_memory))
4264 else if (p2 + 6 < pend
4265 && (re_opcode_t) *p2 == dummy_failure_jump)
4272 /* p1[0] ... p1[2] are the `on_failure_jump' corresponding
4273 to the `maybe_finalize_jump' of this case. Examine what
4276 /* If we're at the end of the pattern, we can change. */
4279 /* Consider what happens when matching ":\(.*\)"
4280 against ":/". I don't really understand this code
4282 p[-3] = (unsigned char) pop_failure_jump;
4284 (" End of pattern: change to `pop_failure_jump'.\n");
4287 else if ((re_opcode_t) *p2 == exactn
4288 || (bufp->newline_anchor && (re_opcode_t) *p2 == endline))
4290 register unsigned char c
4291 = *p2 == (unsigned char) endline ? '\n' : p2[2];
4293 if ((re_opcode_t) p1[3] == exactn && p1[5] != c)
4295 p[-3] = (unsigned char) pop_failure_jump;
4296 DEBUG_PRINT3 (" %c != %c => pop_failure_jump.\n",
4300 else if ((re_opcode_t) p1[3] == charset
4301 || (re_opcode_t) p1[3] == charset_not)
4303 int not = (re_opcode_t) p1[3] == charset_not;
4305 if (c < (unsigned char) (p1[4] * BYTEWIDTH)
4306 && p1[5 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH)))
4309 /* `not' is equal to 1 if c would match, which means
4310 that we can't change to pop_failure_jump. */
4313 p[-3] = (unsigned char) pop_failure_jump;
4314 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
4318 else if ((re_opcode_t) *p2 == charset)
4321 register unsigned char c
4322 = *p2 == (unsigned char) endline ? '\n' : p2[2];
4325 if ((re_opcode_t) p1[3] == exactn
4326 && ! ((int) p2[1] * BYTEWIDTH > (int) p1[4]
4327 && (p2[1 + p1[4] / BYTEWIDTH]
4328 & (1 << (p1[4] % BYTEWIDTH)))))
4330 p[-3] = (unsigned char) pop_failure_jump;
4331 DEBUG_PRINT3 (" %c != %c => pop_failure_jump.\n",
4335 else if ((re_opcode_t) p1[3] == charset_not)
4338 /* We win if the charset_not inside the loop
4339 lists every character listed in the charset after. */
4340 for (idx = 0; idx < (int) p2[1]; idx++)
4341 if (! (p2[2 + idx] == 0
4342 || (idx < (int) p1[4]
4343 && ((p2[2 + idx] & ~ p1[5 + idx]) == 0))))
4348 p[-3] = (unsigned char) pop_failure_jump;
4349 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
4352 else if ((re_opcode_t) p1[3] == charset)
4355 /* We win if the charset inside the loop
4356 has no overlap with the one after the loop. */
4358 idx < (int) p2[1] && idx < (int) p1[4];
4360 if ((p2[2 + idx] & p1[5 + idx]) != 0)
4363 if (idx == p2[1] || idx == p1[4])
4365 p[-3] = (unsigned char) pop_failure_jump;
4366 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
4371 p -= 2; /* Point at relative address again. */
4372 if ((re_opcode_t) p[-1] != pop_failure_jump)
4374 p[-1] = (unsigned char) jump;
4375 DEBUG_PRINT1 (" Match => jump.\n");
4376 goto unconditional_jump;
4378 /* Note fall through. */
4381 /* The end of a simple repeat has a pop_failure_jump back to
4382 its matching on_failure_jump, where the latter will push a
4383 failure point. The pop_failure_jump takes off failure
4384 points put on by this pop_failure_jump's matching
4385 on_failure_jump; we got through the pattern to here from the
4386 matching on_failure_jump, so didn't fail. */
4387 case pop_failure_jump:
4389 /* We need to pass separate storage for the lowest and
4390 highest registers, even though we don't care about the
4391 actual values. Otherwise, we will restore only one
4392 register from the stack, since lowest will == highest in
4393 `pop_failure_point'. */
4394 unsigned dummy_low_reg, dummy_high_reg;
4395 unsigned char *pdummy;
4398 DEBUG_PRINT1 ("EXECUTING pop_failure_jump.\n");
4399 POP_FAILURE_POINT (sdummy, pdummy,
4400 dummy_low_reg, dummy_high_reg,
4401 reg_dummy, reg_dummy, reg_info_dummy);
4403 /* Note fall through. */
4406 /* Unconditionally jump (without popping any failure points). */
4409 EXTRACT_NUMBER_AND_INCR (mcnt, p); /* Get the amount to jump. */
4410 DEBUG_PRINT2 ("EXECUTING jump %d ", mcnt);
4411 p += mcnt; /* Do the jump. */
4412 DEBUG_PRINT2 ("(to 0x%x).\n", p);
4416 /* We need this opcode so we can detect where alternatives end
4417 in `group_match_null_string_p' et al. */
4419 DEBUG_PRINT1 ("EXECUTING jump_past_alt.\n");
4420 goto unconditional_jump;
4423 /* Normally, the on_failure_jump pushes a failure point, which
4424 then gets popped at pop_failure_jump. We will end up at
4425 pop_failure_jump, also, and with a pattern of, say, `a+', we
4426 are skipping over the on_failure_jump, so we have to push
4427 something meaningless for pop_failure_jump to pop. */
4428 case dummy_failure_jump:
4429 DEBUG_PRINT1 ("EXECUTING dummy_failure_jump.\n");
4430 /* It doesn't matter what we push for the string here. What
4431 the code at `fail' tests is the value for the pattern. */
4432 PUSH_FAILURE_POINT (0, 0, -2);
4433 goto unconditional_jump;
4436 /* At the end of an alternative, we need to push a dummy failure
4437 point in case we are followed by a `pop_failure_jump', because
4438 we don't want the failure point for the alternative to be
4439 popped. For example, matching `(a|ab)*' against `aab'
4440 requires that we match the `ab' alternative. */
4441 case push_dummy_failure:
4442 DEBUG_PRINT1 ("EXECUTING push_dummy_failure.\n");
4443 /* See comments just above at `dummy_failure_jump' about the
4445 PUSH_FAILURE_POINT (0, 0, -2);
4448 /* Have to succeed matching what follows at least n times.
4449 After that, handle like `on_failure_jump'. */
4451 EXTRACT_NUMBER (mcnt, p + 2);
4452 DEBUG_PRINT2 ("EXECUTING succeed_n %d.\n", mcnt);
4455 /* Originally, this is how many times we HAVE to succeed. */
4460 STORE_NUMBER_AND_INCR (p, mcnt);
4461 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p, mcnt);
4465 DEBUG_PRINT2 (" Setting two bytes from 0x%x to no_op.\n", p+2);
4466 p[2] = (unsigned char) no_op;
4467 p[3] = (unsigned char) no_op;
4473 EXTRACT_NUMBER (mcnt, p + 2);
4474 DEBUG_PRINT2 ("EXECUTING jump_n %d.\n", mcnt);
4476 /* Originally, this is how many times we CAN jump. */
4480 STORE_NUMBER (p + 2, mcnt);
4481 goto unconditional_jump;
4483 /* If don't have to jump any more, skip over the rest of command. */
4490 DEBUG_PRINT1 ("EXECUTING set_number_at.\n");
4492 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4494 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4495 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p1, mcnt);
4496 STORE_NUMBER (p1, mcnt);
4501 DEBUG_PRINT1 ("EXECUTING wordbound.\n");
4502 if (AT_WORD_BOUNDARY (d))
4507 DEBUG_PRINT1 ("EXECUTING notwordbound.\n");
4508 if (AT_WORD_BOUNDARY (d))
4513 DEBUG_PRINT1 ("EXECUTING wordbeg.\n");
4514 if (WORDCHAR_P (d) && (AT_STRINGS_BEG (d) || !WORDCHAR_P (d - 1)))
4519 DEBUG_PRINT1 ("EXECUTING wordend.\n");
4520 if (!AT_STRINGS_BEG (d) && WORDCHAR_P (d - 1)
4521 && (!WORDCHAR_P (d) || AT_STRINGS_END (d)))
4527 DEBUG_PRINT1 ("EXECUTING before_dot.\n");
4528 if (PTR_CHAR_POS ((unsigned char *) d) >= point)
4533 DEBUG_PRINT1 ("EXECUTING at_dot.\n");
4534 if (PTR_CHAR_POS ((unsigned char *) d) != point)
4539 DEBUG_PRINT1 ("EXECUTING after_dot.\n");
4540 if (PTR_CHAR_POS ((unsigned char *) d) <= point)
4543 #if 0 /* not emacs19 */
4545 DEBUG_PRINT1 ("EXECUTING at_dot.\n");
4546 if (PTR_CHAR_POS ((unsigned char *) d) + 1 != point)
4549 #endif /* not emacs19 */
4552 DEBUG_PRINT2 ("EXECUTING syntaxspec %d.\n", mcnt);
4557 DEBUG_PRINT1 ("EXECUTING Emacs wordchar.\n");
4561 /* Can't use *d++ here; SYNTAX may be an unsafe macro. */
4563 if (SYNTAX (d[-1]) != (enum syntaxcode) mcnt)
4565 SET_REGS_MATCHED ();
4569 DEBUG_PRINT2 ("EXECUTING notsyntaxspec %d.\n", mcnt);
4571 goto matchnotsyntax;
4574 DEBUG_PRINT1 ("EXECUTING Emacs notwordchar.\n");
4578 /* Can't use *d++ here; SYNTAX may be an unsafe macro. */
4580 if (SYNTAX (d[-1]) == (enum syntaxcode) mcnt)
4582 SET_REGS_MATCHED ();
4585 #else /* not emacs */
4587 DEBUG_PRINT1 ("EXECUTING non-Emacs wordchar.\n");
4589 if (!WORDCHAR_P (d))
4591 SET_REGS_MATCHED ();
4596 DEBUG_PRINT1 ("EXECUTING non-Emacs notwordchar.\n");
4600 SET_REGS_MATCHED ();
4603 #endif /* not emacs */
4608 continue; /* Successfully executed one pattern command; keep going. */
4611 /* We goto here if a matching operation fails. */
4613 if (!FAIL_STACK_EMPTY ())
4614 { /* A restart point is known. Restore to that state. */
4615 DEBUG_PRINT1 ("\nFAIL:\n");
4616 POP_FAILURE_POINT (d, p,
4617 lowest_active_reg, highest_active_reg,
4618 regstart, regend, reg_info);
4620 /* If this failure point is a dummy, try the next one. */
4624 /* If we failed to the end of the pattern, don't examine *p. */
4628 boolean is_a_jump_n = false;
4630 /* If failed to a backwards jump that's part of a repetition
4631 loop, need to pop this failure point and use the next one. */
4632 switch ((re_opcode_t) *p)
4636 case maybe_pop_jump:
4637 case pop_failure_jump:
4640 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4643 if ((is_a_jump_n && (re_opcode_t) *p1 == succeed_n)
4645 && (re_opcode_t) *p1 == on_failure_jump))
4653 if (d >= string1 && d <= end1)
4657 break; /* Matching at this starting point really fails. */
4661 goto restore_best_regs;
4665 return -1; /* Failure to match. */
4668 /* Subroutine definitions for re_match_2. */
4671 /* We are passed P pointing to a register number after a start_memory.
4673 Return true if the pattern up to the corresponding stop_memory can
4674 match the empty string, and false otherwise.
4676 If we find the matching stop_memory, sets P to point to one past its number.
4677 Otherwise, sets P to an undefined byte less than or equal to END.
4679 We don't handle duplicates properly (yet). */
4682 group_match_null_string_p (p, end, reg_info)
4683 unsigned char **p, *end;
4684 register_info_type *reg_info;
4687 /* Point to after the args to the start_memory. */
4688 unsigned char *p1 = *p + 2;
4692 /* Skip over opcodes that can match nothing, and return true or
4693 false, as appropriate, when we get to one that can't, or to the
4694 matching stop_memory. */
4696 switch ((re_opcode_t) *p1)
4698 /* Could be either a loop or a series of alternatives. */
4699 case on_failure_jump:
4701 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4703 /* If the next operation is not a jump backwards in the
4708 /* Go through the on_failure_jumps of the alternatives,
4709 seeing if any of the alternatives cannot match nothing.
4710 The last alternative starts with only a jump,
4711 whereas the rest start with on_failure_jump and end
4712 with a jump, e.g., here is the pattern for `a|b|c':
4714 /on_failure_jump/0/6/exactn/1/a/jump_past_alt/0/6
4715 /on_failure_jump/0/6/exactn/1/b/jump_past_alt/0/3
4718 So, we have to first go through the first (n-1)
4719 alternatives and then deal with the last one separately. */
4722 /* Deal with the first (n-1) alternatives, which start
4723 with an on_failure_jump (see above) that jumps to right
4724 past a jump_past_alt. */
4726 while ((re_opcode_t) p1[mcnt-3] == jump_past_alt)
4728 /* `mcnt' holds how many bytes long the alternative
4729 is, including the ending `jump_past_alt' and
4732 if (!alt_match_null_string_p (p1, p1 + mcnt - 3,
4736 /* Move to right after this alternative, including the
4740 /* Break if it's the beginning of an n-th alternative
4741 that doesn't begin with an on_failure_jump. */
4742 if ((re_opcode_t) *p1 != on_failure_jump)
4745 /* Still have to check that it's not an n-th
4746 alternative that starts with an on_failure_jump. */
4748 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4749 if ((re_opcode_t) p1[mcnt-3] != jump_past_alt)
4751 /* Get to the beginning of the n-th alternative. */
4757 /* Deal with the last alternative: go back and get number
4758 of the `jump_past_alt' just before it. `mcnt' contains
4759 the length of the alternative. */
4760 EXTRACT_NUMBER (mcnt, p1 - 2);
4762 if (!alt_match_null_string_p (p1, p1 + mcnt, reg_info))
4765 p1 += mcnt; /* Get past the n-th alternative. */
4771 assert (p1[1] == **p);
4777 if (!common_op_match_null_string_p (&p1, end, reg_info))
4780 } /* while p1 < end */
4783 } /* group_match_null_string_p */
4786 /* Similar to group_match_null_string_p, but doesn't deal with alternatives:
4787 It expects P to be the first byte of a single alternative and END one
4788 byte past the last. The alternative can contain groups. */
4791 alt_match_null_string_p (p, end, reg_info)
4792 unsigned char *p, *end;
4793 register_info_type *reg_info;
4796 unsigned char *p1 = p;
4800 /* Skip over opcodes that can match nothing, and break when we get
4801 to one that can't. */
4803 switch ((re_opcode_t) *p1)
4806 case on_failure_jump:
4808 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4813 if (!common_op_match_null_string_p (&p1, end, reg_info))
4816 } /* while p1 < end */
4819 } /* alt_match_null_string_p */
4822 /* Deals with the ops common to group_match_null_string_p and
4823 alt_match_null_string_p.
4825 Sets P to one after the op and its arguments, if any. */
4828 common_op_match_null_string_p (p, end, reg_info)
4829 unsigned char **p, *end;
4830 register_info_type *reg_info;
4835 unsigned char *p1 = *p;
4837 switch ((re_opcode_t) *p1++)
4857 assert (reg_no > 0 && reg_no <= MAX_REGNUM);
4858 ret = group_match_null_string_p (&p1, end, reg_info);
4860 /* Have to set this here in case we're checking a group which
4861 contains a group and a back reference to it. */
4863 if (REG_MATCH_NULL_STRING_P (reg_info[reg_no]) == MATCH_NULL_UNSET_VALUE)
4864 REG_MATCH_NULL_STRING_P (reg_info[reg_no]) = ret;
4870 /* If this is an optimized succeed_n for zero times, make the jump. */
4872 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4880 /* Get to the number of times to succeed. */
4882 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4887 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4895 if (!REG_MATCH_NULL_STRING_P (reg_info[*p1]))
4903 /* All other opcodes mean we cannot match the empty string. */
4909 } /* common_op_match_null_string_p */
4912 /* Return zero if TRANSLATE[S1] and TRANSLATE[S2] are identical for LEN
4913 bytes; nonzero otherwise. */
4916 bcmp_translate (s1, s2, len, translate)
4917 unsigned char *s1, *s2;
4921 register unsigned char *p1 = s1, *p2 = s2;
4924 if (translate[*p1++] != translate[*p2++]) return 1;
4930 /* Entry points for GNU code. */
4932 /* re_compile_pattern is the GNU regular expression compiler: it
4933 compiles PATTERN (of length SIZE) and puts the result in BUFP.
4934 Returns 0 if the pattern was valid, otherwise an error string.
4936 Assumes the `allocated' (and perhaps `buffer') and `translate' fields
4937 are set in BUFP on entry.
4939 We call regex_compile to do the actual compilation. */
4942 re_compile_pattern (pattern, length, bufp)
4943 const char *pattern;
4945 struct re_pattern_buffer *bufp;
4949 /* GNU code is written to assume at least RE_NREGS registers will be set
4950 (and at least one extra will be -1). */
4951 bufp->regs_allocated = REGS_UNALLOCATED;
4953 /* And GNU code determines whether or not to get register information
4954 by passing null for the REGS argument to re_match, etc., not by
4958 /* Match anchors at newline. */
4959 bufp->newline_anchor = 1;
4961 ret = regex_compile (pattern, length, re_syntax_options, bufp);
4965 return gettext (re_error_msgid[(int) ret]);
4968 /* Entry points compatible with 4.2 BSD regex library. We don't define
4969 them unless specifically requested. */
4971 #ifdef _REGEX_RE_COMP
4973 /* BSD has one and only one pattern buffer. */
4974 static struct re_pattern_buffer re_comp_buf;
4984 if (!re_comp_buf.buffer)
4985 return gettext ("No previous regular expression");
4989 if (!re_comp_buf.buffer)
4991 re_comp_buf.buffer = (unsigned char *) malloc (200);
4992 if (re_comp_buf.buffer == NULL)
4993 return gettext (re_error_msgid[(int) REG_ESPACE]);
4994 re_comp_buf.allocated = 200;
4996 re_comp_buf.fastmap = (char *) malloc (1 << BYTEWIDTH);
4997 if (re_comp_buf.fastmap == NULL)
4998 return gettext (re_error_msgid[(int) REG_ESPACE]);
5001 /* Since `re_exec' always passes NULL for the `regs' argument, we
5002 don't need to initialize the pattern buffer fields which affect it. */
5004 /* Match anchors at newlines. */
5005 re_comp_buf.newline_anchor = 1;
5007 ret = regex_compile (s, strlen (s), re_syntax_options, &re_comp_buf);
5012 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
5013 return (char *) gettext (re_error_msgid[(int) ret]);
5021 const int len = strlen (s);
5023 0 <= re_search (&re_comp_buf, s, len, 0, len, (struct re_registers *) 0);
5025 #endif /* _REGEX_RE_COMP */
5027 /* POSIX.2 functions. Don't define these for Emacs. */
5031 /* regcomp takes a regular expression as a string and compiles it.
5033 PREG is a regex_t *. We do not expect any fields to be initialized,
5034 since POSIX says we shouldn't. Thus, we set
5036 `buffer' to the compiled pattern;
5037 `used' to the length of the compiled pattern;
5038 `syntax' to RE_SYNTAX_POSIX_EXTENDED if the
5039 REG_EXTENDED bit in CFLAGS is set; otherwise, to
5040 RE_SYNTAX_POSIX_BASIC;
5041 `newline_anchor' to REG_NEWLINE being set in CFLAGS;
5042 `fastmap' and `fastmap_accurate' to zero;
5043 `re_nsub' to the number of subexpressions in PATTERN.
5045 PATTERN is the address of the pattern string.
5047 CFLAGS is a series of bits which affect compilation.
5049 If REG_EXTENDED is set, we use POSIX extended syntax; otherwise, we
5050 use POSIX basic syntax.
5052 If REG_NEWLINE is set, then . and [^...] don't match newline.
5053 Also, regexec will try a match beginning after every newline.
5055 If REG_ICASE is set, then we considers upper- and lowercase
5056 versions of letters to be equivalent when matching.
5058 If REG_NOSUB is set, then when PREG is passed to regexec, that
5059 routine will report only success or failure, and nothing about the
5062 It returns 0 if it succeeds, nonzero if it doesn't. (See regex.h for
5063 the return codes and their meanings.) */
5066 regcomp (preg, pattern, cflags)
5068 const char *pattern;
5073 = (cflags & REG_EXTENDED) ?
5074 RE_SYNTAX_POSIX_EXTENDED : RE_SYNTAX_POSIX_BASIC;
5076 /* regex_compile will allocate the space for the compiled pattern. */
5078 preg->allocated = 0;
5081 /* Don't bother to use a fastmap when searching. This simplifies the
5082 REG_NEWLINE case: if we used a fastmap, we'd have to put all the
5083 characters after newlines into the fastmap. This way, we just try
5087 if (cflags & REG_ICASE)
5091 preg->translate = (char *) malloc (CHAR_SET_SIZE);
5092 if (preg->translate == NULL)
5093 return (int) REG_ESPACE;
5095 /* Map uppercase characters to corresponding lowercase ones. */
5096 for (i = 0; i < CHAR_SET_SIZE; i++)
5097 preg->translate[i] = ISUPPER (i) ? tolower (i) : i;
5100 preg->translate = NULL;
5102 /* If REG_NEWLINE is set, newlines are treated differently. */
5103 if (cflags & REG_NEWLINE)
5104 { /* REG_NEWLINE implies neither . nor [^...] match newline. */
5105 syntax &= ~RE_DOT_NEWLINE;
5106 syntax |= RE_HAT_LISTS_NOT_NEWLINE;
5107 /* It also changes the matching behavior. */
5108 preg->newline_anchor = 1;
5111 preg->newline_anchor = 0;
5113 preg->no_sub = !!(cflags & REG_NOSUB);
5115 /* POSIX says a null character in the pattern terminates it, so we
5116 can use strlen here in compiling the pattern. */
5117 ret = regex_compile (pattern, strlen (pattern), syntax, preg);
5119 /* POSIX doesn't distinguish between an unmatched open-group and an
5120 unmatched close-group: both are REG_EPAREN. */
5121 if (ret == REG_ERPAREN) ret = REG_EPAREN;
5127 /* regexec searches for a given pattern, specified by PREG, in the
5130 If NMATCH is zero or REG_NOSUB was set in the cflags argument to
5131 `regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at
5132 least NMATCH elements, and we set them to the offsets of the
5133 corresponding matched substrings.
5135 EFLAGS specifies `execution flags' which affect matching: if
5136 REG_NOTBOL is set, then ^ does not match at the beginning of the
5137 string; if REG_NOTEOL is set, then $ does not match at the end.
5139 We return 0 if we find a match and REG_NOMATCH if not. */
5142 regexec (preg, string, nmatch, pmatch, eflags)
5143 const regex_t *preg;
5146 regmatch_t pmatch[];
5150 struct re_registers regs;
5151 regex_t private_preg;
5152 int len = strlen (string);
5153 boolean want_reg_info = !preg->no_sub && nmatch > 0;
5155 private_preg = *preg;
5157 private_preg.not_bol = !!(eflags & REG_NOTBOL);
5158 private_preg.not_eol = !!(eflags & REG_NOTEOL);
5160 /* The user has told us exactly how many registers to return
5161 information about, via `nmatch'. We have to pass that on to the
5162 matching routines. */
5163 private_preg.regs_allocated = REGS_FIXED;
5167 regs.num_regs = nmatch;
5168 regs.start = TALLOC (nmatch, regoff_t);
5169 regs.end = TALLOC (nmatch, regoff_t);
5170 if (regs.start == NULL || regs.end == NULL)
5171 return (int) REG_NOMATCH;
5174 /* Perform the searching operation. */
5175 ret = re_search (&private_preg, string, len,
5176 /* start: */ 0, /* range: */ len,
5177 want_reg_info ? ®s : (struct re_registers *) 0);
5179 /* Copy the register information to the POSIX structure. */
5186 for (r = 0; r < nmatch; r++)
5188 pmatch[r].rm_so = regs.start[r];
5189 pmatch[r].rm_eo = regs.end[r];
5193 /* If we needed the temporary register info, free the space now. */
5198 /* We want zero return to mean success, unlike `re_search'. */
5199 return ret >= 0 ? (int) REG_NOERROR : (int) REG_NOMATCH;
5203 /* Returns a message corresponding to an error code, ERRCODE, returned
5204 from either regcomp or regexec. We don't use PREG here. */
5207 regerror (errcode, preg, errbuf, errbuf_size)
5209 const regex_t *preg;
5217 || errcode >= (sizeof (re_error_msgid) / sizeof (re_error_msgid[0])))
5218 /* Only error codes returned by the rest of the code should be passed
5219 to this routine. If we are given anything else, or if other regex
5220 code generates an invalid error code, then the program has a bug.
5221 Dump core so we can fix it. */
5224 msg = gettext (re_error_msgid[errcode]);
5226 msg_size = strlen (msg) + 1; /* Includes the null. */
5228 if (errbuf_size != 0)
5230 if (msg_size > errbuf_size)
5232 strncpy (errbuf, msg, errbuf_size - 1);
5233 errbuf[errbuf_size - 1] = 0;
5236 strcpy (errbuf, msg);
5243 /* Free dynamically allocated space used by PREG. */
5249 if (preg->buffer != NULL)
5250 free (preg->buffer);
5251 preg->buffer = NULL;
5253 preg->allocated = 0;
5256 if (preg->fastmap != NULL)
5257 free (preg->fastmap);
5258 preg->fastmap = NULL;
5259 preg->fastmap_accurate = 0;
5261 if (preg->translate != NULL)
5262 free (preg->translate);
5263 preg->translate = NULL;
5266 #endif /* not emacs */
5270 make-backup-files: t
5272 trim-versions-without-asking: nil