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, 1995 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 #if HAVE_LIBINTL_H || defined (_LIBC)
40 # define gettext(msgid) (msgid)
43 /* The `emacs' switch turns on certain matching commands
44 that make sense only in Emacs. */
53 /* If we are not linking with Emacs proper,
54 we can't use the relocating allocator
55 even if config.h says that we can. */
58 #if defined (STDC_HEADERS) || defined (_LIBC)
65 /* When used in Emacs's lib-src, we need to get bzero and bcopy somehow.
66 If nothing else has been done, use the method below. */
67 #ifdef INHIBIT_STRING_HEADER
68 #if !(defined (HAVE_BZERO) && defined (HAVE_BCOPY))
69 #if !defined (bzero) && !defined (bcopy)
70 #undef INHIBIT_STRING_HEADER
75 /* This is the normal way of making sure we have a bcopy and a bzero.
76 This is used in most programs--a few other programs avoid this
77 by defining INHIBIT_STRING_HEADER. */
78 #ifndef INHIBIT_STRING_HEADER
79 #if defined (HAVE_STRING_H) || defined (STDC_HEADERS) || defined (_LIBC)
82 #define bcmp(s1, s2, n) memcmp ((s1), (s2), (n))
85 #define bcopy(s, d, n) memcpy ((d), (s), (n))
88 #define bzero(s, n) memset ((s), 0, (n))
95 /* Define the syntax stuff for \<, \>, etc. */
97 /* This must be nonzero for the wordchar and notwordchar pattern
98 commands in re_match_2. */
103 #ifdef SWITCH_ENUM_BUG
104 #define SWITCH_ENUM_CAST(x) ((int)(x))
106 #define SWITCH_ENUM_CAST(x) (x)
111 extern char *re_syntax_table;
113 #else /* not SYNTAX_TABLE */
115 /* How many characters in the character set. */
116 #define CHAR_SET_SIZE 256
118 static char re_syntax_table[CHAR_SET_SIZE];
129 bzero (re_syntax_table, sizeof re_syntax_table);
131 for (c = 'a'; c <= 'z'; c++)
132 re_syntax_table[c] = Sword;
134 for (c = 'A'; c <= 'Z'; c++)
135 re_syntax_table[c] = Sword;
137 for (c = '0'; c <= '9'; c++)
138 re_syntax_table[c] = Sword;
140 re_syntax_table['_'] = Sword;
145 #endif /* not SYNTAX_TABLE */
147 #define SYNTAX(c) re_syntax_table[c]
149 #endif /* not emacs */
151 /* Get the interface, including the syntax bits. */
154 /* isalpha etc. are used for the character classes. */
157 /* Jim Meyering writes:
159 "... Some ctype macros are valid only for character codes that
160 isascii says are ASCII (SGI's IRIX-4.0.5 is one such system --when
161 using /bin/cc or gcc but without giving an ansi option). So, all
162 ctype uses should be through macros like ISPRINT... If
163 STDC_HEADERS is defined, then autoconf has verified that the ctype
164 macros don't need to be guarded with references to isascii. ...
165 Defining isascii to 1 should let any compiler worth its salt
166 eliminate the && through constant folding." */
168 #if defined (STDC_HEADERS) || (!defined (isascii) && !defined (HAVE_ISASCII))
171 #define ISASCII(c) isascii(c)
175 #define ISBLANK(c) (ISASCII (c) && isblank (c))
177 #define ISBLANK(c) ((c) == ' ' || (c) == '\t')
180 #define ISGRAPH(c) (ISASCII (c) && isgraph (c))
182 #define ISGRAPH(c) (ISASCII (c) && isprint (c) && !isspace (c))
185 #define ISPRINT(c) (ISASCII (c) && isprint (c))
186 #define ISDIGIT(c) (ISASCII (c) && isdigit (c))
187 #define ISALNUM(c) (ISASCII (c) && isalnum (c))
188 #define ISALPHA(c) (ISASCII (c) && isalpha (c))
189 #define ISCNTRL(c) (ISASCII (c) && iscntrl (c))
190 #define ISLOWER(c) (ISASCII (c) && islower (c))
191 #define ISPUNCT(c) (ISASCII (c) && ispunct (c))
192 #define ISSPACE(c) (ISASCII (c) && isspace (c))
193 #define ISUPPER(c) (ISASCII (c) && isupper (c))
194 #define ISXDIGIT(c) (ISASCII (c) && isxdigit (c))
197 #define NULL (void *)0
200 /* We remove any previous definition of `SIGN_EXTEND_CHAR',
201 since ours (we hope) works properly with all combinations of
202 machines, compilers, `char' and `unsigned char' argument types.
203 (Per Bothner suggested the basic approach.) */
204 #undef SIGN_EXTEND_CHAR
206 #define SIGN_EXTEND_CHAR(c) ((signed char) (c))
207 #else /* not __STDC__ */
208 /* As in Harbison and Steele. */
209 #define SIGN_EXTEND_CHAR(c) ((((unsigned char) (c)) ^ 128) - 128)
212 /* Should we use malloc or alloca? If REGEX_MALLOC is not defined, we
213 use `alloca' instead of `malloc'. This is because using malloc in
214 re_search* or re_match* could cause memory leaks when C-g is used in
215 Emacs; also, malloc is slower and causes storage fragmentation. On
216 the other hand, malloc is more portable, and easier to debug.
218 Because we sometimes use alloca, some routines have to be macros,
219 not functions -- `alloca'-allocated space disappears at the end of the
220 function it is called in. */
224 #define REGEX_ALLOCATE malloc
225 #define REGEX_REALLOCATE(source, osize, nsize) realloc (source, nsize)
226 #define REGEX_FREE free
228 #else /* not REGEX_MALLOC */
230 /* Emacs already defines alloca, sometimes. */
233 /* Make alloca work the best possible way. */
235 #define alloca __builtin_alloca
236 #else /* not __GNUC__ */
239 #else /* not __GNUC__ or HAVE_ALLOCA_H */
240 #ifndef _AIX /* Already did AIX, up at the top. */
242 #endif /* not _AIX */
243 #endif /* not HAVE_ALLOCA_H */
244 #endif /* not __GNUC__ */
246 #endif /* not alloca */
248 #define REGEX_ALLOCATE alloca
250 /* Assumes a `char *destination' variable. */
251 #define REGEX_REALLOCATE(source, osize, nsize) \
252 (destination = (char *) alloca (nsize), \
253 bcopy (source, destination, osize), \
256 /* No need to do anything to free, after alloca. */
257 #define REGEX_FREE(arg) ((void)0) /* Do nothing! But inhibit gcc warning. */
259 #endif /* not REGEX_MALLOC */
261 /* Define how to allocate the failure stack. */
263 #if defined (REL_ALLOC) && defined (REGEX_MALLOC)
265 #define REGEX_ALLOCATE_STACK(size) \
266 r_alloc (&failure_stack_ptr, (size))
267 #define REGEX_REALLOCATE_STACK(source, osize, nsize) \
268 r_re_alloc (&failure_stack_ptr, (nsize))
269 #define REGEX_FREE_STACK(ptr) \
270 r_alloc_free (&failure_stack_ptr)
272 #else /* not using relocating allocator */
276 #define REGEX_ALLOCATE_STACK malloc
277 #define REGEX_REALLOCATE_STACK(source, osize, nsize) realloc (source, nsize)
278 #define REGEX_FREE_STACK free
280 #else /* not REGEX_MALLOC */
282 #define REGEX_ALLOCATE_STACK alloca
284 #define REGEX_REALLOCATE_STACK(source, osize, nsize) \
285 REGEX_REALLOCATE (source, osize, nsize)
286 /* No need to explicitly free anything. */
287 #define REGEX_FREE_STACK(arg)
289 #endif /* not REGEX_MALLOC */
290 #endif /* not using relocating allocator */
293 /* True if `size1' is non-NULL and PTR is pointing anywhere inside
294 `string1' or just past its end. This works if PTR is NULL, which is
296 #define FIRST_STRING_P(ptr) \
297 (size1 && string1 <= (ptr) && (ptr) <= string1 + size1)
299 /* (Re)Allocate N items of type T using malloc, or fail. */
300 #define TALLOC(n, t) ((t *) malloc ((n) * sizeof (t)))
301 #define RETALLOC(addr, n, t) ((addr) = (t *) realloc (addr, (n) * sizeof (t)))
302 #define RETALLOC_IF(addr, n, t) \
303 if (addr) RETALLOC((addr), (n), t); else (addr) = TALLOC ((n), t)
304 #define REGEX_TALLOC(n, t) ((t *) REGEX_ALLOCATE ((n) * sizeof (t)))
306 #define BYTEWIDTH 8 /* In bits. */
308 #define STREQ(s1, s2) ((strcmp (s1, s2) == 0))
312 #define MAX(a, b) ((a) > (b) ? (a) : (b))
313 #define MIN(a, b) ((a) < (b) ? (a) : (b))
315 typedef char boolean;
319 static int re_match_2_internal ();
321 /* These are the command codes that appear in compiled regular
322 expressions. Some opcodes are followed by argument bytes. A
323 command code can specify any interpretation whatsoever for its
324 arguments. Zero bytes may appear in the compiled regular expression. */
330 /* Succeed right away--no more backtracking. */
333 /* Followed by one byte giving n, then by n literal bytes. */
336 /* Matches any (more or less) character. */
339 /* Matches any one char belonging to specified set. First
340 following byte is number of bitmap bytes. Then come bytes
341 for a bitmap saying which chars are in. Bits in each byte
342 are ordered low-bit-first. A character is in the set if its
343 bit is 1. A character too large to have a bit in the map is
344 automatically not in the set. */
347 /* Same parameters as charset, but match any character that is
348 not one of those specified. */
351 /* Start remembering the text that is matched, for storing in a
352 register. Followed by one byte with the register number, in
353 the range 0 to one less than the pattern buffer's re_nsub
354 field. Then followed by one byte with the number of groups
355 inner to this one. (This last has to be part of the
356 start_memory only because we need it in the on_failure_jump
360 /* Stop remembering the text that is matched and store it in a
361 memory register. Followed by one byte with the register
362 number, in the range 0 to one less than `re_nsub' in the
363 pattern buffer, and one byte with the number of inner groups,
364 just like `start_memory'. (We need the number of inner
365 groups here because we don't have any easy way of finding the
366 corresponding start_memory when we're at a stop_memory.) */
369 /* Match a duplicate of something remembered. Followed by one
370 byte containing the register number. */
373 /* Fail unless at beginning of line. */
376 /* Fail unless at end of line. */
379 /* Succeeds if at beginning of buffer (if emacs) or at beginning
380 of string to be matched (if not). */
383 /* Analogously, for end of buffer/string. */
386 /* Followed by two byte relative address to which to jump. */
389 /* Same as jump, but marks the end of an alternative. */
392 /* Followed by two-byte relative address of place to resume at
393 in case of failure. */
396 /* Like on_failure_jump, but pushes a placeholder instead of the
397 current string position when executed. */
398 on_failure_keep_string_jump,
400 /* Throw away latest failure point and then jump to following
401 two-byte relative address. */
404 /* Change to pop_failure_jump if know won't have to backtrack to
405 match; otherwise change to jump. This is used to jump
406 back to the beginning of a repeat. If what follows this jump
407 clearly won't match what the repeat does, such that we can be
408 sure that there is no use backtracking out of repetitions
409 already matched, then we change it to a pop_failure_jump.
410 Followed by two-byte address. */
413 /* Jump to following two-byte address, and push a dummy failure
414 point. This failure point will be thrown away if an attempt
415 is made to use it for a failure. A `+' construct makes this
416 before the first repeat. Also used as an intermediary kind
417 of jump when compiling an alternative. */
420 /* Push a dummy failure point and continue. Used at the end of
424 /* Followed by two-byte relative address and two-byte number n.
425 After matching N times, jump to the address upon failure. */
428 /* Followed by two-byte relative address, and two-byte number n.
429 Jump to the address N times, then fail. */
432 /* Set the following two-byte relative address to the
433 subsequent two-byte number. The address *includes* the two
437 wordchar, /* Matches any word-constituent character. */
438 notwordchar, /* Matches any char that is not a word-constituent. */
440 wordbeg, /* Succeeds if at word beginning. */
441 wordend, /* Succeeds if at word end. */
443 wordbound, /* Succeeds if at a word boundary. */
444 notwordbound /* Succeeds if not at a word boundary. */
447 ,before_dot, /* Succeeds if before point. */
448 at_dot, /* Succeeds if at point. */
449 after_dot, /* Succeeds if after point. */
451 /* Matches any character whose syntax is specified. Followed by
452 a byte which contains a syntax code, e.g., Sword. */
455 /* Matches any character whose syntax is not that specified. */
460 /* Common operations on the compiled pattern. */
462 /* Store NUMBER in two contiguous bytes starting at DESTINATION. */
464 #define STORE_NUMBER(destination, number) \
466 (destination)[0] = (number) & 0377; \
467 (destination)[1] = (number) >> 8; \
470 /* Same as STORE_NUMBER, except increment DESTINATION to
471 the byte after where the number is stored. Therefore, DESTINATION
472 must be an lvalue. */
474 #define STORE_NUMBER_AND_INCR(destination, number) \
476 STORE_NUMBER (destination, number); \
477 (destination) += 2; \
480 /* Put into DESTINATION a number stored in two contiguous bytes starting
483 #define EXTRACT_NUMBER(destination, source) \
485 (destination) = *(source) & 0377; \
486 (destination) += SIGN_EXTEND_CHAR (*((source) + 1)) << 8; \
491 extract_number (dest, source)
493 unsigned char *source;
495 int temp = SIGN_EXTEND_CHAR (*(source + 1));
496 *dest = *source & 0377;
500 #ifndef EXTRACT_MACROS /* To debug the macros. */
501 #undef EXTRACT_NUMBER
502 #define EXTRACT_NUMBER(dest, src) extract_number (&dest, src)
503 #endif /* not EXTRACT_MACROS */
507 /* Same as EXTRACT_NUMBER, except increment SOURCE to after the number.
508 SOURCE must be an lvalue. */
510 #define EXTRACT_NUMBER_AND_INCR(destination, source) \
512 EXTRACT_NUMBER (destination, source); \
518 extract_number_and_incr (destination, source)
520 unsigned char **source;
522 extract_number (destination, *source);
526 #ifndef EXTRACT_MACROS
527 #undef EXTRACT_NUMBER_AND_INCR
528 #define EXTRACT_NUMBER_AND_INCR(dest, src) \
529 extract_number_and_incr (&dest, &src)
530 #endif /* not EXTRACT_MACROS */
534 /* If DEBUG is defined, Regex prints many voluminous messages about what
535 it is doing (if the variable `debug' is nonzero). If linked with the
536 main program in `iregex.c', you can enter patterns and strings
537 interactively. And if linked with the main program in `main.c' and
538 the other test files, you can run the already-written tests. */
542 /* We use standard I/O for debugging. */
545 /* It is useful to test things that ``must'' be true when debugging. */
548 static int debug = 0;
550 #define DEBUG_STATEMENT(e) e
551 #define DEBUG_PRINT1(x) if (debug) printf (x)
552 #define DEBUG_PRINT2(x1, x2) if (debug) printf (x1, x2)
553 #define DEBUG_PRINT3(x1, x2, x3) if (debug) printf (x1, x2, x3)
554 #define DEBUG_PRINT4(x1, x2, x3, x4) if (debug) printf (x1, x2, x3, x4)
555 #define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) \
556 if (debug) print_partial_compiled_pattern (s, e)
557 #define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) \
558 if (debug) print_double_string (w, s1, sz1, s2, sz2)
561 /* Print the fastmap in human-readable form. */
564 print_fastmap (fastmap)
567 unsigned was_a_range = 0;
570 while (i < (1 << BYTEWIDTH))
576 while (i < (1 << BYTEWIDTH) && fastmap[i])
592 /* Print a compiled pattern string in human-readable form, starting at
593 the START pointer into it and ending just before the pointer END. */
596 print_partial_compiled_pattern (start, end)
597 unsigned char *start;
601 unsigned char *p = start;
602 unsigned char *pend = end;
610 /* Loop over pattern commands. */
613 printf ("%d:\t", p - start);
615 switch ((re_opcode_t) *p++)
623 printf ("/exactn/%d", mcnt);
634 printf ("/start_memory/%d/%d", mcnt, *p++);
639 printf ("/stop_memory/%d/%d", mcnt, *p++);
643 printf ("/duplicate/%d", *p++);
653 register int c, last = -100;
654 register int in_range = 0;
656 printf ("/charset [%s",
657 (re_opcode_t) *(p - 1) == charset_not ? "^" : "");
659 assert (p + *p < pend);
661 for (c = 0; c < 256; c++)
663 && (p[1 + (c/8)] & (1 << (c % 8))))
665 /* Are we starting a range? */
666 if (last + 1 == c && ! in_range)
671 /* Have we broken a range? */
672 else if (last + 1 != c && in_range)
701 case on_failure_jump:
702 extract_number_and_incr (&mcnt, &p);
703 printf ("/on_failure_jump to %d", p + mcnt - start);
706 case on_failure_keep_string_jump:
707 extract_number_and_incr (&mcnt, &p);
708 printf ("/on_failure_keep_string_jump to %d", p + mcnt - start);
711 case dummy_failure_jump:
712 extract_number_and_incr (&mcnt, &p);
713 printf ("/dummy_failure_jump to %d", p + mcnt - start);
716 case push_dummy_failure:
717 printf ("/push_dummy_failure");
721 extract_number_and_incr (&mcnt, &p);
722 printf ("/maybe_pop_jump to %d", p + mcnt - start);
725 case pop_failure_jump:
726 extract_number_and_incr (&mcnt, &p);
727 printf ("/pop_failure_jump to %d", p + mcnt - start);
731 extract_number_and_incr (&mcnt, &p);
732 printf ("/jump_past_alt to %d", p + mcnt - start);
736 extract_number_and_incr (&mcnt, &p);
737 printf ("/jump to %d", p + mcnt - start);
741 extract_number_and_incr (&mcnt, &p);
742 extract_number_and_incr (&mcnt2, &p);
743 printf ("/succeed_n to %d, %d times", p + mcnt - start, mcnt2);
747 extract_number_and_incr (&mcnt, &p);
748 extract_number_and_incr (&mcnt2, &p);
749 printf ("/jump_n to %d, %d times", p + mcnt - start, mcnt2);
753 extract_number_and_incr (&mcnt, &p);
754 extract_number_and_incr (&mcnt2, &p);
755 printf ("/set_number_at location %d to %d", p + mcnt - start, mcnt2);
759 printf ("/wordbound");
763 printf ("/notwordbound");
775 printf ("/before_dot");
783 printf ("/after_dot");
787 printf ("/syntaxspec");
789 printf ("/%d", mcnt);
793 printf ("/notsyntaxspec");
795 printf ("/%d", mcnt);
800 printf ("/wordchar");
804 printf ("/notwordchar");
816 printf ("?%d", *(p-1));
822 printf ("%d:\tend of pattern.\n", p - start);
827 print_compiled_pattern (bufp)
828 struct re_pattern_buffer *bufp;
830 unsigned char *buffer = bufp->buffer;
832 print_partial_compiled_pattern (buffer, buffer + bufp->used);
833 printf ("%d bytes used/%d bytes allocated.\n", bufp->used, bufp->allocated);
835 if (bufp->fastmap_accurate && bufp->fastmap)
837 printf ("fastmap: ");
838 print_fastmap (bufp->fastmap);
841 printf ("re_nsub: %d\t", bufp->re_nsub);
842 printf ("regs_alloc: %d\t", bufp->regs_allocated);
843 printf ("can_be_null: %d\t", bufp->can_be_null);
844 printf ("newline_anchor: %d\n", bufp->newline_anchor);
845 printf ("no_sub: %d\t", bufp->no_sub);
846 printf ("not_bol: %d\t", bufp->not_bol);
847 printf ("not_eol: %d\t", bufp->not_eol);
848 printf ("syntax: %d\n", bufp->syntax);
849 /* Perhaps we should print the translate table? */
854 print_double_string (where, string1, size1, string2, size2)
867 if (FIRST_STRING_P (where))
869 for (this_char = where - string1; this_char < size1; this_char++)
870 putchar (string1[this_char]);
875 for (this_char = where - string2; this_char < size2; this_char++)
876 putchar (string2[this_char]);
880 #else /* not DEBUG */
885 #define DEBUG_STATEMENT(e)
886 #define DEBUG_PRINT1(x)
887 #define DEBUG_PRINT2(x1, x2)
888 #define DEBUG_PRINT3(x1, x2, x3)
889 #define DEBUG_PRINT4(x1, x2, x3, x4)
890 #define DEBUG_PRINT_COMPILED_PATTERN(p, s, e)
891 #define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2)
893 #endif /* not DEBUG */
895 /* Set by `re_set_syntax' to the current regexp syntax to recognize. Can
896 also be assigned to arbitrarily: each pattern buffer stores its own
897 syntax, so it can be changed between regex compilations. */
898 /* This has no initializer because initialized variables in Emacs
899 become read-only after dumping. */
900 reg_syntax_t re_syntax_options;
903 /* Specify the precise syntax of regexps for compilation. This provides
904 for compatibility for various utilities which historically have
905 different, incompatible syntaxes.
907 The argument SYNTAX is a bit mask comprised of the various bits
908 defined in regex.h. We return the old syntax. */
911 re_set_syntax (syntax)
914 reg_syntax_t ret = re_syntax_options;
916 re_syntax_options = syntax;
920 /* This table gives an error message for each of the error codes listed
921 in regex.h. Obviously the order here has to be same as there.
922 POSIX doesn't require that we do anything for REG_NOERROR,
923 but why not be nice? */
925 static const char *re_error_msgid[] =
926 { "Success", /* REG_NOERROR */
927 "No match", /* REG_NOMATCH */
928 "Invalid regular expression", /* REG_BADPAT */
929 "Invalid collation character", /* REG_ECOLLATE */
930 "Invalid character class name", /* REG_ECTYPE */
931 "Trailing backslash", /* REG_EESCAPE */
932 "Invalid back reference", /* REG_ESUBREG */
933 "Unmatched [ or [^", /* REG_EBRACK */
934 "Unmatched ( or \\(", /* REG_EPAREN */
935 "Unmatched \\{", /* REG_EBRACE */
936 "Invalid content of \\{\\}", /* REG_BADBR */
937 "Invalid range end", /* REG_ERANGE */
938 "Memory exhausted", /* REG_ESPACE */
939 "Invalid preceding regular expression", /* REG_BADRPT */
940 "Premature end of regular expression", /* REG_EEND */
941 "Regular expression too big", /* REG_ESIZE */
942 "Unmatched ) or \\)", /* REG_ERPAREN */
945 /* Avoiding alloca during matching, to placate r_alloc. */
947 /* Define MATCH_MAY_ALLOCATE unless we need to make sure that the
948 searching and matching functions should not call alloca. On some
949 systems, alloca is implemented in terms of malloc, and if we're
950 using the relocating allocator routines, then malloc could cause a
951 relocation, which might (if the strings being searched are in the
952 ralloc heap) shift the data out from underneath the regexp
955 Here's another reason to avoid allocation: Emacs
956 processes input from X in a signal handler; processing X input may
957 call malloc; if input arrives while a matching routine is calling
958 malloc, then we're scrod. But Emacs can't just block input while
959 calling matching routines; then we don't notice interrupts when
960 they come in. So, Emacs blocks input around all regexp calls
961 except the matching calls, which it leaves unprotected, in the
962 faith that they will not malloc. */
964 /* Normally, this is fine. */
965 #define MATCH_MAY_ALLOCATE
967 /* When using GNU C, we are not REALLY using the C alloca, no matter
968 what config.h may say. So don't take precautions for it. */
973 /* The match routines may not allocate if (1) they would do it with malloc
974 and (2) it's not safe for them to use malloc.
975 Note that if REL_ALLOC is defined, matching would not use malloc for the
976 failure stack, but we would still use it for the register vectors;
977 so REL_ALLOC should not affect this. */
978 #if (defined (C_ALLOCA) || defined (REGEX_MALLOC)) && defined (emacs)
979 #undef MATCH_MAY_ALLOCATE
983 /* Failure stack declarations and macros; both re_compile_fastmap and
984 re_match_2 use a failure stack. These have to be macros because of
985 REGEX_ALLOCATE_STACK. */
988 /* Number of failure points for which to initially allocate space
989 when matching. If this number is exceeded, we allocate more
990 space, so it is not a hard limit. */
991 #ifndef INIT_FAILURE_ALLOC
992 #define INIT_FAILURE_ALLOC 5
995 /* Roughly the maximum number of failure points on the stack. Would be
996 exactly that if always used MAX_FAILURE_SPACE each time we failed.
997 This is a variable only so users of regex can assign to it; we never
998 change it ourselves. */
999 #if defined (MATCH_MAY_ALLOCATE)
1000 int re_max_failures = 200000;
1002 int re_max_failures = 2000;
1005 union fail_stack_elt
1007 unsigned char *pointer;
1011 typedef union fail_stack_elt fail_stack_elt_t;
1015 fail_stack_elt_t *stack;
1017 unsigned avail; /* Offset of next open position. */
1020 #define FAIL_STACK_EMPTY() (fail_stack.avail == 0)
1021 #define FAIL_STACK_PTR_EMPTY() (fail_stack_ptr->avail == 0)
1022 #define FAIL_STACK_FULL() (fail_stack.avail == fail_stack.size)
1025 /* Define macros to initialize and free the failure stack.
1026 Do `return -2' if the alloc fails. */
1028 #ifdef MATCH_MAY_ALLOCATE
1029 #define INIT_FAIL_STACK() \
1031 fail_stack.stack = (fail_stack_elt_t *) \
1032 REGEX_ALLOCATE_STACK (INIT_FAILURE_ALLOC * sizeof (fail_stack_elt_t)); \
1034 if (fail_stack.stack == NULL) \
1037 fail_stack.size = INIT_FAILURE_ALLOC; \
1038 fail_stack.avail = 0; \
1041 #define RESET_FAIL_STACK() REGEX_FREE_STACK (fail_stack.stack)
1043 #define INIT_FAIL_STACK() \
1045 fail_stack.avail = 0; \
1048 #define RESET_FAIL_STACK()
1052 /* Double the size of FAIL_STACK, up to approximately `re_max_failures' items.
1054 Return 1 if succeeds, and 0 if either ran out of memory
1055 allocating space for it or it was already too large.
1057 REGEX_REALLOCATE_STACK requires `destination' be declared. */
1059 #define DOUBLE_FAIL_STACK(fail_stack) \
1060 ((fail_stack).size > re_max_failures * MAX_FAILURE_ITEMS \
1062 : ((fail_stack).stack = (fail_stack_elt_t *) \
1063 REGEX_REALLOCATE_STACK ((fail_stack).stack, \
1064 (fail_stack).size * sizeof (fail_stack_elt_t), \
1065 ((fail_stack).size << 1) * sizeof (fail_stack_elt_t)), \
1067 (fail_stack).stack == NULL \
1069 : ((fail_stack).size <<= 1, \
1073 /* Push pointer POINTER on FAIL_STACK.
1074 Return 1 if was able to do so and 0 if ran out of memory allocating
1076 #define PUSH_PATTERN_OP(POINTER, FAIL_STACK) \
1077 ((FAIL_STACK_FULL () \
1078 && !DOUBLE_FAIL_STACK (FAIL_STACK)) \
1080 : ((FAIL_STACK).stack[(FAIL_STACK).avail++].pointer = POINTER, \
1083 /* Push a pointer value onto the failure stack.
1084 Assumes the variable `fail_stack'. Probably should only
1085 be called from within `PUSH_FAILURE_POINT'. */
1086 #define PUSH_FAILURE_POINTER(item) \
1087 fail_stack.stack[fail_stack.avail++].pointer = (unsigned char *) (item)
1089 /* This pushes an integer-valued item onto the failure stack.
1090 Assumes the variable `fail_stack'. Probably should only
1091 be called from within `PUSH_FAILURE_POINT'. */
1092 #define PUSH_FAILURE_INT(item) \
1093 fail_stack.stack[fail_stack.avail++].integer = (item)
1095 /* Push a fail_stack_elt_t value onto the failure stack.
1096 Assumes the variable `fail_stack'. Probably should only
1097 be called from within `PUSH_FAILURE_POINT'. */
1098 #define PUSH_FAILURE_ELT(item) \
1099 fail_stack.stack[fail_stack.avail++] = (item)
1101 /* These three POP... operations complement the three PUSH... operations.
1102 All assume that `fail_stack' is nonempty. */
1103 #define POP_FAILURE_POINTER() fail_stack.stack[--fail_stack.avail].pointer
1104 #define POP_FAILURE_INT() fail_stack.stack[--fail_stack.avail].integer
1105 #define POP_FAILURE_ELT() fail_stack.stack[--fail_stack.avail]
1107 /* Used to omit pushing failure point id's when we're not debugging. */
1109 #define DEBUG_PUSH PUSH_FAILURE_INT
1110 #define DEBUG_POP(item_addr) *(item_addr) = POP_FAILURE_INT ()
1112 #define DEBUG_PUSH(item)
1113 #define DEBUG_POP(item_addr)
1117 /* Push the information about the state we will need
1118 if we ever fail back to it.
1120 Requires variables fail_stack, regstart, regend, reg_info, and
1121 num_regs be declared. DOUBLE_FAIL_STACK requires `destination' be
1124 Does `return FAILURE_CODE' if runs out of memory. */
1126 #define PUSH_FAILURE_POINT(pattern_place, string_place, failure_code) \
1128 char *destination; \
1129 /* Must be int, so when we don't save any registers, the arithmetic \
1130 of 0 + -1 isn't done as unsigned. */ \
1133 DEBUG_STATEMENT (failure_id++); \
1134 DEBUG_STATEMENT (nfailure_points_pushed++); \
1135 DEBUG_PRINT2 ("\nPUSH_FAILURE_POINT #%u:\n", failure_id); \
1136 DEBUG_PRINT2 (" Before push, next avail: %d\n", (fail_stack).avail);\
1137 DEBUG_PRINT2 (" size: %d\n", (fail_stack).size);\
1139 DEBUG_PRINT2 (" slots needed: %d\n", NUM_FAILURE_ITEMS); \
1140 DEBUG_PRINT2 (" available: %d\n", REMAINING_AVAIL_SLOTS); \
1142 /* Ensure we have enough space allocated for what we will push. */ \
1143 while (REMAINING_AVAIL_SLOTS < NUM_FAILURE_ITEMS) \
1145 if (!DOUBLE_FAIL_STACK (fail_stack)) \
1146 return failure_code; \
1148 DEBUG_PRINT2 ("\n Doubled stack; size now: %d\n", \
1149 (fail_stack).size); \
1150 DEBUG_PRINT2 (" slots available: %d\n", REMAINING_AVAIL_SLOTS);\
1153 /* Push the info, starting with the registers. */ \
1154 DEBUG_PRINT1 ("\n"); \
1156 if (!RE_NO_POSIX_BACKTRACKING & bufp->syntax) \
1157 for (this_reg = lowest_active_reg; this_reg <= highest_active_reg; \
1160 DEBUG_PRINT2 (" Pushing reg: %d\n", this_reg); \
1161 DEBUG_STATEMENT (num_regs_pushed++); \
1163 DEBUG_PRINT2 (" start: 0x%x\n", regstart[this_reg]); \
1164 PUSH_FAILURE_POINTER (regstart[this_reg]); \
1166 DEBUG_PRINT2 (" end: 0x%x\n", regend[this_reg]); \
1167 PUSH_FAILURE_POINTER (regend[this_reg]); \
1169 DEBUG_PRINT2 (" info: 0x%x\n ", reg_info[this_reg]); \
1170 DEBUG_PRINT2 (" match_null=%d", \
1171 REG_MATCH_NULL_STRING_P (reg_info[this_reg])); \
1172 DEBUG_PRINT2 (" active=%d", IS_ACTIVE (reg_info[this_reg])); \
1173 DEBUG_PRINT2 (" matched_something=%d", \
1174 MATCHED_SOMETHING (reg_info[this_reg])); \
1175 DEBUG_PRINT2 (" ever_matched=%d", \
1176 EVER_MATCHED_SOMETHING (reg_info[this_reg])); \
1177 DEBUG_PRINT1 ("\n"); \
1178 PUSH_FAILURE_ELT (reg_info[this_reg].word); \
1181 DEBUG_PRINT2 (" Pushing low active reg: %d\n", lowest_active_reg);\
1182 PUSH_FAILURE_INT (lowest_active_reg); \
1184 DEBUG_PRINT2 (" Pushing high active reg: %d\n", highest_active_reg);\
1185 PUSH_FAILURE_INT (highest_active_reg); \
1187 DEBUG_PRINT2 (" Pushing pattern 0x%x: ", pattern_place); \
1188 DEBUG_PRINT_COMPILED_PATTERN (bufp, pattern_place, pend); \
1189 PUSH_FAILURE_POINTER (pattern_place); \
1191 DEBUG_PRINT2 (" Pushing string 0x%x: `", string_place); \
1192 DEBUG_PRINT_DOUBLE_STRING (string_place, string1, size1, string2, \
1194 DEBUG_PRINT1 ("'\n"); \
1195 PUSH_FAILURE_POINTER (string_place); \
1197 DEBUG_PRINT2 (" Pushing failure id: %u\n", failure_id); \
1198 DEBUG_PUSH (failure_id); \
1201 /* This is the number of items that are pushed and popped on the stack
1202 for each register. */
1203 #define NUM_REG_ITEMS 3
1205 /* Individual items aside from the registers. */
1207 #define NUM_NONREG_ITEMS 5 /* Includes failure point id. */
1209 #define NUM_NONREG_ITEMS 4
1212 /* We push at most this many items on the stack. */
1213 #define MAX_FAILURE_ITEMS ((num_regs - 1) * NUM_REG_ITEMS + NUM_NONREG_ITEMS)
1215 /* We actually push this many items. */
1216 #define NUM_FAILURE_ITEMS \
1217 (((RE_NO_POSIX_BACKTRACKING & bufp->syntax \
1218 ? 0 : highest_active_reg - lowest_active_reg + 1) \
1222 /* How many items can still be added to the stack without overflowing it. */
1223 #define REMAINING_AVAIL_SLOTS ((fail_stack).size - (fail_stack).avail)
1226 /* Pops what PUSH_FAIL_STACK pushes.
1228 We restore into the parameters, all of which should be lvalues:
1229 STR -- the saved data position.
1230 PAT -- the saved pattern position.
1231 LOW_REG, HIGH_REG -- the highest and lowest active registers.
1232 REGSTART, REGEND -- arrays of string positions.
1233 REG_INFO -- array of information about each subexpression.
1235 Also assumes the variables `fail_stack' and (if debugging), `bufp',
1236 `pend', `string1', `size1', `string2', and `size2'. */
1238 #define POP_FAILURE_POINT(str, pat, low_reg, high_reg, regstart, regend, reg_info)\
1240 DEBUG_STATEMENT (fail_stack_elt_t failure_id;) \
1242 const unsigned char *string_temp; \
1244 assert (!FAIL_STACK_EMPTY ()); \
1246 /* Remove failure points and point to how many regs pushed. */ \
1247 DEBUG_PRINT1 ("POP_FAILURE_POINT:\n"); \
1248 DEBUG_PRINT2 (" Before pop, next avail: %d\n", fail_stack.avail); \
1249 DEBUG_PRINT2 (" size: %d\n", fail_stack.size); \
1251 assert (fail_stack.avail >= NUM_NONREG_ITEMS); \
1253 DEBUG_POP (&failure_id); \
1254 DEBUG_PRINT2 (" Popping failure id: %u\n", failure_id); \
1256 /* If the saved string location is NULL, it came from an \
1257 on_failure_keep_string_jump opcode, and we want to throw away the \
1258 saved NULL, thus retaining our current position in the string. */ \
1259 string_temp = POP_FAILURE_POINTER (); \
1260 if (string_temp != NULL) \
1261 str = (const char *) string_temp; \
1263 DEBUG_PRINT2 (" Popping string 0x%x: `", str); \
1264 DEBUG_PRINT_DOUBLE_STRING (str, string1, size1, string2, size2); \
1265 DEBUG_PRINT1 ("'\n"); \
1267 pat = (unsigned char *) POP_FAILURE_POINTER (); \
1268 DEBUG_PRINT2 (" Popping pattern 0x%x: ", pat); \
1269 DEBUG_PRINT_COMPILED_PATTERN (bufp, pat, pend); \
1271 /* Restore register info. */ \
1272 high_reg = (unsigned) POP_FAILURE_INT (); \
1273 DEBUG_PRINT2 (" Popping high active reg: %d\n", high_reg); \
1275 low_reg = (unsigned) POP_FAILURE_INT (); \
1276 DEBUG_PRINT2 (" Popping low active reg: %d\n", low_reg); \
1278 if (!RE_NO_POSIX_BACKTRACKING & bufp->syntax) \
1279 for (this_reg = high_reg; this_reg >= low_reg; this_reg--) \
1281 DEBUG_PRINT2 (" Popping reg: %d\n", this_reg); \
1283 reg_info[this_reg].word = POP_FAILURE_ELT (); \
1284 DEBUG_PRINT2 (" info: 0x%x\n", reg_info[this_reg]); \
1286 regend[this_reg] = (const char *) POP_FAILURE_POINTER (); \
1287 DEBUG_PRINT2 (" end: 0x%x\n", regend[this_reg]); \
1289 regstart[this_reg] = (const char *) POP_FAILURE_POINTER (); \
1290 DEBUG_PRINT2 (" start: 0x%x\n", regstart[this_reg]); \
1293 set_regs_matched_done = 0; \
1294 DEBUG_STATEMENT (nfailure_points_popped++); \
1295 } /* POP_FAILURE_POINT */
1299 /* Structure for per-register (a.k.a. per-group) information.
1300 Other register information, such as the
1301 starting and ending positions (which are addresses), and the list of
1302 inner groups (which is a bits list) are maintained in separate
1305 We are making a (strictly speaking) nonportable assumption here: that
1306 the compiler will pack our bit fields into something that fits into
1307 the type of `word', i.e., is something that fits into one item on the
1312 fail_stack_elt_t word;
1315 /* This field is one if this group can match the empty string,
1316 zero if not. If not yet determined, `MATCH_NULL_UNSET_VALUE'. */
1317 #define MATCH_NULL_UNSET_VALUE 3
1318 unsigned match_null_string_p : 2;
1319 unsigned is_active : 1;
1320 unsigned matched_something : 1;
1321 unsigned ever_matched_something : 1;
1323 } register_info_type;
1325 #define REG_MATCH_NULL_STRING_P(R) ((R).bits.match_null_string_p)
1326 #define IS_ACTIVE(R) ((R).bits.is_active)
1327 #define MATCHED_SOMETHING(R) ((R).bits.matched_something)
1328 #define EVER_MATCHED_SOMETHING(R) ((R).bits.ever_matched_something)
1331 /* Call this when have matched a real character; it sets `matched' flags
1332 for the subexpressions which we are currently inside. Also records
1333 that those subexprs have matched. */
1334 #define SET_REGS_MATCHED() \
1337 if (!set_regs_matched_done) \
1340 set_regs_matched_done = 1; \
1341 for (r = lowest_active_reg; r <= highest_active_reg; r++) \
1343 MATCHED_SOMETHING (reg_info[r]) \
1344 = EVER_MATCHED_SOMETHING (reg_info[r]) \
1351 /* Registers are set to a sentinel when they haven't yet matched. */
1352 static char reg_unset_dummy;
1353 #define REG_UNSET_VALUE (®_unset_dummy)
1354 #define REG_UNSET(e) ((e) == REG_UNSET_VALUE)
1356 /* Subroutine declarations and macros for regex_compile. */
1358 static void store_op1 (), store_op2 ();
1359 static void insert_op1 (), insert_op2 ();
1360 static boolean at_begline_loc_p (), at_endline_loc_p ();
1361 static boolean group_in_compile_stack ();
1362 static reg_errcode_t compile_range ();
1364 /* Fetch the next character in the uncompiled pattern---translating it
1365 if necessary. Also cast from a signed character in the constant
1366 string passed to us by the user to an unsigned char that we can use
1367 as an array index (in, e.g., `translate'). */
1368 #define PATFETCH(c) \
1369 do {if (p == pend) return REG_EEND; \
1370 c = (unsigned char) *p++; \
1371 if (translate) c = translate[c]; \
1374 /* Fetch the next character in the uncompiled pattern, with no
1376 #define PATFETCH_RAW(c) \
1377 do {if (p == pend) return REG_EEND; \
1378 c = (unsigned char) *p++; \
1381 /* Go backwards one character in the pattern. */
1382 #define PATUNFETCH p--
1385 /* If `translate' is non-null, return translate[D], else just D. We
1386 cast the subscript to translate because some data is declared as
1387 `char *', to avoid warnings when a string constant is passed. But
1388 when we use a character as a subscript we must make it unsigned. */
1389 #define TRANSLATE(d) (translate ? translate[(unsigned char) (d)] : (d))
1392 /* Macros for outputting the compiled pattern into `buffer'. */
1394 /* If the buffer isn't allocated when it comes in, use this. */
1395 #define INIT_BUF_SIZE 32
1397 /* Make sure we have at least N more bytes of space in buffer. */
1398 #define GET_BUFFER_SPACE(n) \
1399 while (b - bufp->buffer + (n) > bufp->allocated) \
1402 /* Make sure we have one more byte of buffer space and then add C to it. */
1403 #define BUF_PUSH(c) \
1405 GET_BUFFER_SPACE (1); \
1406 *b++ = (unsigned char) (c); \
1410 /* Ensure we have two more bytes of buffer space and then append C1 and C2. */
1411 #define BUF_PUSH_2(c1, c2) \
1413 GET_BUFFER_SPACE (2); \
1414 *b++ = (unsigned char) (c1); \
1415 *b++ = (unsigned char) (c2); \
1419 /* As with BUF_PUSH_2, except for three bytes. */
1420 #define BUF_PUSH_3(c1, c2, c3) \
1422 GET_BUFFER_SPACE (3); \
1423 *b++ = (unsigned char) (c1); \
1424 *b++ = (unsigned char) (c2); \
1425 *b++ = (unsigned char) (c3); \
1429 /* Store a jump with opcode OP at LOC to location TO. We store a
1430 relative address offset by the three bytes the jump itself occupies. */
1431 #define STORE_JUMP(op, loc, to) \
1432 store_op1 (op, loc, (to) - (loc) - 3)
1434 /* Likewise, for a two-argument jump. */
1435 #define STORE_JUMP2(op, loc, to, arg) \
1436 store_op2 (op, loc, (to) - (loc) - 3, arg)
1438 /* Like `STORE_JUMP', but for inserting. Assume `b' is the buffer end. */
1439 #define INSERT_JUMP(op, loc, to) \
1440 insert_op1 (op, loc, (to) - (loc) - 3, b)
1442 /* Like `STORE_JUMP2', but for inserting. Assume `b' is the buffer end. */
1443 #define INSERT_JUMP2(op, loc, to, arg) \
1444 insert_op2 (op, loc, (to) - (loc) - 3, arg, b)
1447 /* This is not an arbitrary limit: the arguments which represent offsets
1448 into the pattern are two bytes long. So if 2^16 bytes turns out to
1449 be too small, many things would have to change. */
1450 #define MAX_BUF_SIZE (1L << 16)
1453 /* Extend the buffer by twice its current size via realloc and
1454 reset the pointers that pointed into the old block to point to the
1455 correct places in the new one. If extending the buffer results in it
1456 being larger than MAX_BUF_SIZE, then flag memory exhausted. */
1457 #define EXTEND_BUFFER() \
1459 unsigned char *old_buffer = bufp->buffer; \
1460 if (bufp->allocated == MAX_BUF_SIZE) \
1462 bufp->allocated <<= 1; \
1463 if (bufp->allocated > MAX_BUF_SIZE) \
1464 bufp->allocated = MAX_BUF_SIZE; \
1465 bufp->buffer = (unsigned char *) realloc (bufp->buffer, bufp->allocated);\
1466 if (bufp->buffer == NULL) \
1467 return REG_ESPACE; \
1468 /* If the buffer moved, move all the pointers into it. */ \
1469 if (old_buffer != bufp->buffer) \
1471 b = (b - old_buffer) + bufp->buffer; \
1472 begalt = (begalt - old_buffer) + bufp->buffer; \
1473 if (fixup_alt_jump) \
1474 fixup_alt_jump = (fixup_alt_jump - old_buffer) + bufp->buffer;\
1476 laststart = (laststart - old_buffer) + bufp->buffer; \
1477 if (pending_exact) \
1478 pending_exact = (pending_exact - old_buffer) + bufp->buffer; \
1483 /* Since we have one byte reserved for the register number argument to
1484 {start,stop}_memory, the maximum number of groups we can report
1485 things about is what fits in that byte. */
1486 #define MAX_REGNUM 255
1488 /* But patterns can have more than `MAX_REGNUM' registers. We just
1489 ignore the excess. */
1490 typedef unsigned regnum_t;
1493 /* Macros for the compile stack. */
1495 /* Since offsets can go either forwards or backwards, this type needs to
1496 be able to hold values from -(MAX_BUF_SIZE - 1) to MAX_BUF_SIZE - 1. */
1497 typedef int pattern_offset_t;
1501 pattern_offset_t begalt_offset;
1502 pattern_offset_t fixup_alt_jump;
1503 pattern_offset_t inner_group_offset;
1504 pattern_offset_t laststart_offset;
1506 } compile_stack_elt_t;
1511 compile_stack_elt_t *stack;
1513 unsigned avail; /* Offset of next open position. */
1514 } compile_stack_type;
1517 #define INIT_COMPILE_STACK_SIZE 32
1519 #define COMPILE_STACK_EMPTY (compile_stack.avail == 0)
1520 #define COMPILE_STACK_FULL (compile_stack.avail == compile_stack.size)
1522 /* The next available element. */
1523 #define COMPILE_STACK_TOP (compile_stack.stack[compile_stack.avail])
1526 /* Set the bit for character C in a list. */
1527 #define SET_LIST_BIT(c) \
1528 (b[((unsigned char) (c)) / BYTEWIDTH] \
1529 |= 1 << (((unsigned char) c) % BYTEWIDTH))
1532 /* Get the next unsigned number in the uncompiled pattern. */
1533 #define GET_UNSIGNED_NUMBER(num) \
1537 while (ISDIGIT (c)) \
1541 num = num * 10 + c - '0'; \
1549 #define CHAR_CLASS_MAX_LENGTH 6 /* Namely, `xdigit'. */
1551 #define IS_CHAR_CLASS(string) \
1552 (STREQ (string, "alpha") || STREQ (string, "upper") \
1553 || STREQ (string, "lower") || STREQ (string, "digit") \
1554 || STREQ (string, "alnum") || STREQ (string, "xdigit") \
1555 || STREQ (string, "space") || STREQ (string, "print") \
1556 || STREQ (string, "punct") || STREQ (string, "graph") \
1557 || STREQ (string, "cntrl") || STREQ (string, "blank"))
1559 #ifndef MATCH_MAY_ALLOCATE
1561 /* If we cannot allocate large objects within re_match_2_internal,
1562 we make the fail stack and register vectors global.
1563 The fail stack, we grow to the maximum size when a regexp
1565 The register vectors, we adjust in size each time we
1566 compile a regexp, according to the number of registers it needs. */
1568 static fail_stack_type fail_stack;
1570 /* Size with which the following vectors are currently allocated.
1571 That is so we can make them bigger as needed,
1572 but never make them smaller. */
1573 static int regs_allocated_size;
1575 static const char ** regstart, ** regend;
1576 static const char ** old_regstart, ** old_regend;
1577 static const char **best_regstart, **best_regend;
1578 static register_info_type *reg_info;
1579 static const char **reg_dummy;
1580 static register_info_type *reg_info_dummy;
1582 /* Make the register vectors big enough for NUM_REGS registers,
1583 but don't make them smaller. */
1586 regex_grow_registers (num_regs)
1589 if (num_regs > regs_allocated_size)
1591 RETALLOC_IF (regstart, num_regs, const char *);
1592 RETALLOC_IF (regend, num_regs, const char *);
1593 RETALLOC_IF (old_regstart, num_regs, const char *);
1594 RETALLOC_IF (old_regend, num_regs, const char *);
1595 RETALLOC_IF (best_regstart, num_regs, const char *);
1596 RETALLOC_IF (best_regend, num_regs, const char *);
1597 RETALLOC_IF (reg_info, num_regs, register_info_type);
1598 RETALLOC_IF (reg_dummy, num_regs, const char *);
1599 RETALLOC_IF (reg_info_dummy, num_regs, register_info_type);
1601 regs_allocated_size = num_regs;
1605 #endif /* not MATCH_MAY_ALLOCATE */
1607 /* `regex_compile' compiles PATTERN (of length SIZE) according to SYNTAX.
1608 Returns one of error codes defined in `regex.h', or zero for success.
1610 Assumes the `allocated' (and perhaps `buffer') and `translate'
1611 fields are set in BUFP on entry.
1613 If it succeeds, results are put in BUFP (if it returns an error, the
1614 contents of BUFP are undefined):
1615 `buffer' is the compiled pattern;
1616 `syntax' is set to SYNTAX;
1617 `used' is set to the length of the compiled pattern;
1618 `fastmap_accurate' is zero;
1619 `re_nsub' is the number of subexpressions in PATTERN;
1620 `not_bol' and `not_eol' are zero;
1622 The `fastmap' and `newline_anchor' fields are neither
1623 examined nor set. */
1625 /* Return, freeing storage we allocated. */
1626 #define FREE_STACK_RETURN(value) \
1627 return (free (compile_stack.stack), value)
1629 static reg_errcode_t
1630 regex_compile (pattern, size, syntax, bufp)
1631 const char *pattern;
1633 reg_syntax_t syntax;
1634 struct re_pattern_buffer *bufp;
1636 /* We fetch characters from PATTERN here. Even though PATTERN is
1637 `char *' (i.e., signed), we declare these variables as unsigned, so
1638 they can be reliably used as array indices. */
1639 register unsigned char c, c1;
1641 /* A random temporary spot in PATTERN. */
1644 /* Points to the end of the buffer, where we should append. */
1645 register unsigned char *b;
1647 /* Keeps track of unclosed groups. */
1648 compile_stack_type compile_stack;
1650 /* Points to the current (ending) position in the pattern. */
1651 const char *p = pattern;
1652 const char *pend = pattern + size;
1654 /* How to translate the characters in the pattern. */
1655 char *translate = bufp->translate;
1657 /* Address of the count-byte of the most recently inserted `exactn'
1658 command. This makes it possible to tell if a new exact-match
1659 character can be added to that command or if the character requires
1660 a new `exactn' command. */
1661 unsigned char *pending_exact = 0;
1663 /* Address of start of the most recently finished expression.
1664 This tells, e.g., postfix * where to find the start of its
1665 operand. Reset at the beginning of groups and alternatives. */
1666 unsigned char *laststart = 0;
1668 /* Address of beginning of regexp, or inside of last group. */
1669 unsigned char *begalt;
1671 /* Place in the uncompiled pattern (i.e., the {) to
1672 which to go back if the interval is invalid. */
1673 const char *beg_interval;
1675 /* Address of the place where a forward jump should go to the end of
1676 the containing expression. Each alternative of an `or' -- except the
1677 last -- ends with a forward jump of this sort. */
1678 unsigned char *fixup_alt_jump = 0;
1680 /* Counts open-groups as they are encountered. Remembered for the
1681 matching close-group on the compile stack, so the same register
1682 number is put in the stop_memory as the start_memory. */
1683 regnum_t regnum = 0;
1686 DEBUG_PRINT1 ("\nCompiling pattern: ");
1689 unsigned debug_count;
1691 for (debug_count = 0; debug_count < size; debug_count++)
1692 putchar (pattern[debug_count]);
1697 /* Initialize the compile stack. */
1698 compile_stack.stack = TALLOC (INIT_COMPILE_STACK_SIZE, compile_stack_elt_t);
1699 if (compile_stack.stack == NULL)
1702 compile_stack.size = INIT_COMPILE_STACK_SIZE;
1703 compile_stack.avail = 0;
1705 /* Initialize the pattern buffer. */
1706 bufp->syntax = syntax;
1707 bufp->fastmap_accurate = 0;
1708 bufp->not_bol = bufp->not_eol = 0;
1710 /* Set `used' to zero, so that if we return an error, the pattern
1711 printer (for debugging) will think there's no pattern. We reset it
1715 /* Always count groups, whether or not bufp->no_sub is set. */
1718 #if !defined (emacs) && !defined (SYNTAX_TABLE)
1719 /* Initialize the syntax table. */
1720 init_syntax_once ();
1723 if (bufp->allocated == 0)
1726 { /* If zero allocated, but buffer is non-null, try to realloc
1727 enough space. This loses if buffer's address is bogus, but
1728 that is the user's responsibility. */
1729 RETALLOC (bufp->buffer, INIT_BUF_SIZE, unsigned char);
1732 { /* Caller did not allocate a buffer. Do it for them. */
1733 bufp->buffer = TALLOC (INIT_BUF_SIZE, unsigned char);
1735 if (!bufp->buffer) FREE_STACK_RETURN (REG_ESPACE);
1737 bufp->allocated = INIT_BUF_SIZE;
1740 begalt = b = bufp->buffer;
1742 /* Loop through the uncompiled pattern until we're at the end. */
1751 if ( /* If at start of pattern, it's an operator. */
1753 /* If context independent, it's an operator. */
1754 || syntax & RE_CONTEXT_INDEP_ANCHORS
1755 /* Otherwise, depends on what's come before. */
1756 || at_begline_loc_p (pattern, p, syntax))
1766 if ( /* If at end of pattern, it's an operator. */
1768 /* If context independent, it's an operator. */
1769 || syntax & RE_CONTEXT_INDEP_ANCHORS
1770 /* Otherwise, depends on what's next. */
1771 || at_endline_loc_p (p, pend, syntax))
1781 if ((syntax & RE_BK_PLUS_QM)
1782 || (syntax & RE_LIMITED_OPS))
1786 /* If there is no previous pattern... */
1789 if (syntax & RE_CONTEXT_INVALID_OPS)
1790 FREE_STACK_RETURN (REG_BADRPT);
1791 else if (!(syntax & RE_CONTEXT_INDEP_OPS))
1796 /* Are we optimizing this jump? */
1797 boolean keep_string_p = false;
1799 /* 1 means zero (many) matches is allowed. */
1800 char zero_times_ok = 0, many_times_ok = 0;
1802 /* If there is a sequence of repetition chars, collapse it
1803 down to just one (the right one). We can't combine
1804 interval operators with these because of, e.g., `a{2}*',
1805 which should only match an even number of `a's. */
1809 zero_times_ok |= c != '+';
1810 many_times_ok |= c != '?';
1818 || (!(syntax & RE_BK_PLUS_QM) && (c == '+' || c == '?')))
1821 else if (syntax & RE_BK_PLUS_QM && c == '\\')
1823 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
1826 if (!(c1 == '+' || c1 == '?'))
1841 /* If we get here, we found another repeat character. */
1844 /* Star, etc. applied to an empty pattern is equivalent
1845 to an empty pattern. */
1849 /* Now we know whether or not zero matches is allowed
1850 and also whether or not two or more matches is allowed. */
1852 { /* More than one repetition is allowed, so put in at the
1853 end a backward relative jump from `b' to before the next
1854 jump we're going to put in below (which jumps from
1855 laststart to after this jump).
1857 But if we are at the `*' in the exact sequence `.*\n',
1858 insert an unconditional jump backwards to the .,
1859 instead of the beginning of the loop. This way we only
1860 push a failure point once, instead of every time
1861 through the loop. */
1862 assert (p - 1 > pattern);
1864 /* Allocate the space for the jump. */
1865 GET_BUFFER_SPACE (3);
1867 /* We know we are not at the first character of the pattern,
1868 because laststart was nonzero. And we've already
1869 incremented `p', by the way, to be the character after
1870 the `*'. Do we have to do something analogous here
1871 for null bytes, because of RE_DOT_NOT_NULL? */
1872 if (TRANSLATE (*(p - 2)) == TRANSLATE ('.')
1874 && p < pend && TRANSLATE (*p) == TRANSLATE ('\n')
1875 && !(syntax & RE_DOT_NEWLINE))
1876 { /* We have .*\n. */
1877 STORE_JUMP (jump, b, laststart);
1878 keep_string_p = true;
1881 /* Anything else. */
1882 STORE_JUMP (maybe_pop_jump, b, laststart - 3);
1884 /* We've added more stuff to the buffer. */
1888 /* On failure, jump from laststart to b + 3, which will be the
1889 end of the buffer after this jump is inserted. */
1890 GET_BUFFER_SPACE (3);
1891 INSERT_JUMP (keep_string_p ? on_failure_keep_string_jump
1899 /* At least one repetition is required, so insert a
1900 `dummy_failure_jump' before the initial
1901 `on_failure_jump' instruction of the loop. This
1902 effects a skip over that instruction the first time
1903 we hit that loop. */
1904 GET_BUFFER_SPACE (3);
1905 INSERT_JUMP (dummy_failure_jump, laststart, laststart + 6);
1920 boolean had_char_class = false;
1922 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
1924 /* Ensure that we have enough space to push a charset: the
1925 opcode, the length count, and the bitset; 34 bytes in all. */
1926 GET_BUFFER_SPACE (34);
1930 /* We test `*p == '^' twice, instead of using an if
1931 statement, so we only need one BUF_PUSH. */
1932 BUF_PUSH (*p == '^' ? charset_not : charset);
1936 /* Remember the first position in the bracket expression. */
1939 /* Push the number of bytes in the bitmap. */
1940 BUF_PUSH ((1 << BYTEWIDTH) / BYTEWIDTH);
1942 /* Clear the whole map. */
1943 bzero (b, (1 << BYTEWIDTH) / BYTEWIDTH);
1945 /* charset_not matches newline according to a syntax bit. */
1946 if ((re_opcode_t) b[-2] == charset_not
1947 && (syntax & RE_HAT_LISTS_NOT_NEWLINE))
1948 SET_LIST_BIT ('\n');
1950 /* Read in characters and ranges, setting map bits. */
1953 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
1957 /* \ might escape characters inside [...] and [^...]. */
1958 if ((syntax & RE_BACKSLASH_ESCAPE_IN_LISTS) && c == '\\')
1960 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
1967 /* Could be the end of the bracket expression. If it's
1968 not (i.e., when the bracket expression is `[]' so
1969 far), the ']' character bit gets set way below. */
1970 if (c == ']' && p != p1 + 1)
1973 /* Look ahead to see if it's a range when the last thing
1974 was a character class. */
1975 if (had_char_class && c == '-' && *p != ']')
1976 FREE_STACK_RETURN (REG_ERANGE);
1978 /* Look ahead to see if it's a range when the last thing
1979 was a character: if this is a hyphen not at the
1980 beginning or the end of a list, then it's the range
1983 && !(p - 2 >= pattern && p[-2] == '[')
1984 && !(p - 3 >= pattern && p[-3] == '[' && p[-2] == '^')
1988 = compile_range (&p, pend, translate, syntax, b);
1989 if (ret != REG_NOERROR) FREE_STACK_RETURN (ret);
1992 else if (p[0] == '-' && p[1] != ']')
1993 { /* This handles ranges made up of characters only. */
1996 /* Move past the `-'. */
1999 ret = compile_range (&p, pend, translate, syntax, b);
2000 if (ret != REG_NOERROR) FREE_STACK_RETURN (ret);
2003 /* See if we're at the beginning of a possible character
2006 else if (syntax & RE_CHAR_CLASSES && c == '[' && *p == ':')
2007 { /* Leave room for the null. */
2008 char str[CHAR_CLASS_MAX_LENGTH + 1];
2013 /* If pattern is `[[:'. */
2014 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2019 if (c == ':' || c == ']' || p == pend
2020 || c1 == CHAR_CLASS_MAX_LENGTH)
2026 /* If isn't a word bracketed by `[:' and:`]':
2027 undo the ending character, the letters, and leave
2028 the leading `:' and `[' (but set bits for them). */
2029 if (c == ':' && *p == ']')
2032 boolean is_alnum = STREQ (str, "alnum");
2033 boolean is_alpha = STREQ (str, "alpha");
2034 boolean is_blank = STREQ (str, "blank");
2035 boolean is_cntrl = STREQ (str, "cntrl");
2036 boolean is_digit = STREQ (str, "digit");
2037 boolean is_graph = STREQ (str, "graph");
2038 boolean is_lower = STREQ (str, "lower");
2039 boolean is_print = STREQ (str, "print");
2040 boolean is_punct = STREQ (str, "punct");
2041 boolean is_space = STREQ (str, "space");
2042 boolean is_upper = STREQ (str, "upper");
2043 boolean is_xdigit = STREQ (str, "xdigit");
2045 if (!IS_CHAR_CLASS (str))
2046 FREE_STACK_RETURN (REG_ECTYPE);
2048 /* Throw away the ] at the end of the character
2052 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2054 for (ch = 0; ch < 1 << BYTEWIDTH; ch++)
2056 /* This was split into 3 if's to
2057 avoid an arbitrary limit in some compiler. */
2058 if ( (is_alnum && ISALNUM (ch))
2059 || (is_alpha && ISALPHA (ch))
2060 || (is_blank && ISBLANK (ch))
2061 || (is_cntrl && ISCNTRL (ch)))
2063 if ( (is_digit && ISDIGIT (ch))
2064 || (is_graph && ISGRAPH (ch))
2065 || (is_lower && ISLOWER (ch))
2066 || (is_print && ISPRINT (ch)))
2068 if ( (is_punct && ISPUNCT (ch))
2069 || (is_space && ISSPACE (ch))
2070 || (is_upper && ISUPPER (ch))
2071 || (is_xdigit && ISXDIGIT (ch)))
2074 had_char_class = true;
2083 had_char_class = false;
2088 had_char_class = false;
2093 /* Discard any (non)matching list bytes that are all 0 at the
2094 end of the map. Decrease the map-length byte too. */
2095 while ((int) b[-1] > 0 && b[b[-1] - 1] == 0)
2103 if (syntax & RE_NO_BK_PARENS)
2110 if (syntax & RE_NO_BK_PARENS)
2117 if (syntax & RE_NEWLINE_ALT)
2124 if (syntax & RE_NO_BK_VBAR)
2131 if (syntax & RE_INTERVALS && syntax & RE_NO_BK_BRACES)
2132 goto handle_interval;
2138 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
2140 /* Do not translate the character after the \, so that we can
2141 distinguish, e.g., \B from \b, even if we normally would
2142 translate, e.g., B to b. */
2148 if (syntax & RE_NO_BK_PARENS)
2149 goto normal_backslash;
2155 if (COMPILE_STACK_FULL)
2157 RETALLOC (compile_stack.stack, compile_stack.size << 1,
2158 compile_stack_elt_t);
2159 if (compile_stack.stack == NULL) return REG_ESPACE;
2161 compile_stack.size <<= 1;
2164 /* These are the values to restore when we hit end of this
2165 group. They are all relative offsets, so that if the
2166 whole pattern moves because of realloc, they will still
2168 COMPILE_STACK_TOP.begalt_offset = begalt - bufp->buffer;
2169 COMPILE_STACK_TOP.fixup_alt_jump
2170 = fixup_alt_jump ? fixup_alt_jump - bufp->buffer + 1 : 0;
2171 COMPILE_STACK_TOP.laststart_offset = b - bufp->buffer;
2172 COMPILE_STACK_TOP.regnum = regnum;
2174 /* We will eventually replace the 0 with the number of
2175 groups inner to this one. But do not push a
2176 start_memory for groups beyond the last one we can
2177 represent in the compiled pattern. */
2178 if (regnum <= MAX_REGNUM)
2180 COMPILE_STACK_TOP.inner_group_offset = b - bufp->buffer + 2;
2181 BUF_PUSH_3 (start_memory, regnum, 0);
2184 compile_stack.avail++;
2189 /* If we've reached MAX_REGNUM groups, then this open
2190 won't actually generate any code, so we'll have to
2191 clear pending_exact explicitly. */
2197 if (syntax & RE_NO_BK_PARENS) goto normal_backslash;
2199 if (COMPILE_STACK_EMPTY)
2200 if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD)
2201 goto normal_backslash;
2203 FREE_STACK_RETURN (REG_ERPAREN);
2207 { /* Push a dummy failure point at the end of the
2208 alternative for a possible future
2209 `pop_failure_jump' to pop. See comments at
2210 `push_dummy_failure' in `re_match_2'. */
2211 BUF_PUSH (push_dummy_failure);
2213 /* We allocated space for this jump when we assigned
2214 to `fixup_alt_jump', in the `handle_alt' case below. */
2215 STORE_JUMP (jump_past_alt, fixup_alt_jump, b - 1);
2218 /* See similar code for backslashed left paren above. */
2219 if (COMPILE_STACK_EMPTY)
2220 if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD)
2223 FREE_STACK_RETURN (REG_ERPAREN);
2225 /* Since we just checked for an empty stack above, this
2226 ``can't happen''. */
2227 assert (compile_stack.avail != 0);
2229 /* We don't just want to restore into `regnum', because
2230 later groups should continue to be numbered higher,
2231 as in `(ab)c(de)' -- the second group is #2. */
2232 regnum_t this_group_regnum;
2234 compile_stack.avail--;
2235 begalt = bufp->buffer + COMPILE_STACK_TOP.begalt_offset;
2237 = COMPILE_STACK_TOP.fixup_alt_jump
2238 ? bufp->buffer + COMPILE_STACK_TOP.fixup_alt_jump - 1
2240 laststart = bufp->buffer + COMPILE_STACK_TOP.laststart_offset;
2241 this_group_regnum = COMPILE_STACK_TOP.regnum;
2242 /* If we've reached MAX_REGNUM groups, then this open
2243 won't actually generate any code, so we'll have to
2244 clear pending_exact explicitly. */
2247 /* We're at the end of the group, so now we know how many
2248 groups were inside this one. */
2249 if (this_group_regnum <= MAX_REGNUM)
2251 unsigned char *inner_group_loc
2252 = bufp->buffer + COMPILE_STACK_TOP.inner_group_offset;
2254 *inner_group_loc = regnum - this_group_regnum;
2255 BUF_PUSH_3 (stop_memory, this_group_regnum,
2256 regnum - this_group_regnum);
2262 case '|': /* `\|'. */
2263 if (syntax & RE_LIMITED_OPS || syntax & RE_NO_BK_VBAR)
2264 goto normal_backslash;
2266 if (syntax & RE_LIMITED_OPS)
2269 /* Insert before the previous alternative a jump which
2270 jumps to this alternative if the former fails. */
2271 GET_BUFFER_SPACE (3);
2272 INSERT_JUMP (on_failure_jump, begalt, b + 6);
2276 /* The alternative before this one has a jump after it
2277 which gets executed if it gets matched. Adjust that
2278 jump so it will jump to this alternative's analogous
2279 jump (put in below, which in turn will jump to the next
2280 (if any) alternative's such jump, etc.). The last such
2281 jump jumps to the correct final destination. A picture:
2287 If we are at `b', then fixup_alt_jump right now points to a
2288 three-byte space after `a'. We'll put in the jump, set
2289 fixup_alt_jump to right after `b', and leave behind three
2290 bytes which we'll fill in when we get to after `c'. */
2293 STORE_JUMP (jump_past_alt, fixup_alt_jump, b);
2295 /* Mark and leave space for a jump after this alternative,
2296 to be filled in later either by next alternative or
2297 when know we're at the end of a series of alternatives. */
2299 GET_BUFFER_SPACE (3);
2308 /* If \{ is a literal. */
2309 if (!(syntax & RE_INTERVALS)
2310 /* If we're at `\{' and it's not the open-interval
2312 || ((syntax & RE_INTERVALS) && (syntax & RE_NO_BK_BRACES))
2313 || (p - 2 == pattern && p == pend))
2314 goto normal_backslash;
2318 /* If got here, then the syntax allows intervals. */
2320 /* At least (most) this many matches must be made. */
2321 int lower_bound = -1, upper_bound = -1;
2323 beg_interval = p - 1;
2327 if (syntax & RE_NO_BK_BRACES)
2328 goto unfetch_interval;
2330 FREE_STACK_RETURN (REG_EBRACE);
2333 GET_UNSIGNED_NUMBER (lower_bound);
2337 GET_UNSIGNED_NUMBER (upper_bound);
2338 if (upper_bound < 0) upper_bound = RE_DUP_MAX;
2341 /* Interval such as `{1}' => match exactly once. */
2342 upper_bound = lower_bound;
2344 if (lower_bound < 0 || upper_bound > RE_DUP_MAX
2345 || lower_bound > upper_bound)
2347 if (syntax & RE_NO_BK_BRACES)
2348 goto unfetch_interval;
2350 FREE_STACK_RETURN (REG_BADBR);
2353 if (!(syntax & RE_NO_BK_BRACES))
2355 if (c != '\\') FREE_STACK_RETURN (REG_EBRACE);
2362 if (syntax & RE_NO_BK_BRACES)
2363 goto unfetch_interval;
2365 FREE_STACK_RETURN (REG_BADBR);
2368 /* We just parsed a valid interval. */
2370 /* If it's invalid to have no preceding re. */
2373 if (syntax & RE_CONTEXT_INVALID_OPS)
2374 FREE_STACK_RETURN (REG_BADRPT);
2375 else if (syntax & RE_CONTEXT_INDEP_OPS)
2378 goto unfetch_interval;
2381 /* If the upper bound is zero, don't want to succeed at
2382 all; jump from `laststart' to `b + 3', which will be
2383 the end of the buffer after we insert the jump. */
2384 if (upper_bound == 0)
2386 GET_BUFFER_SPACE (3);
2387 INSERT_JUMP (jump, laststart, b + 3);
2391 /* Otherwise, we have a nontrivial interval. When
2392 we're all done, the pattern will look like:
2393 set_number_at <jump count> <upper bound>
2394 set_number_at <succeed_n count> <lower bound>
2395 succeed_n <after jump addr> <succeed_n count>
2397 jump_n <succeed_n addr> <jump count>
2398 (The upper bound and `jump_n' are omitted if
2399 `upper_bound' is 1, though.) */
2401 { /* If the upper bound is > 1, we need to insert
2402 more at the end of the loop. */
2403 unsigned nbytes = 10 + (upper_bound > 1) * 10;
2405 GET_BUFFER_SPACE (nbytes);
2407 /* Initialize lower bound of the `succeed_n', even
2408 though it will be set during matching by its
2409 attendant `set_number_at' (inserted next),
2410 because `re_compile_fastmap' needs to know.
2411 Jump to the `jump_n' we might insert below. */
2412 INSERT_JUMP2 (succeed_n, laststart,
2413 b + 5 + (upper_bound > 1) * 5,
2417 /* Code to initialize the lower bound. Insert
2418 before the `succeed_n'. The `5' is the last two
2419 bytes of this `set_number_at', plus 3 bytes of
2420 the following `succeed_n'. */
2421 insert_op2 (set_number_at, laststart, 5, lower_bound, b);
2424 if (upper_bound > 1)
2425 { /* More than one repetition is allowed, so
2426 append a backward jump to the `succeed_n'
2427 that starts this interval.
2429 When we've reached this during matching,
2430 we'll have matched the interval once, so
2431 jump back only `upper_bound - 1' times. */
2432 STORE_JUMP2 (jump_n, b, laststart + 5,
2436 /* The location we want to set is the second
2437 parameter of the `jump_n'; that is `b-2' as
2438 an absolute address. `laststart' will be
2439 the `set_number_at' we're about to insert;
2440 `laststart+3' the number to set, the source
2441 for the relative address. But we are
2442 inserting into the middle of the pattern --
2443 so everything is getting moved up by 5.
2444 Conclusion: (b - 2) - (laststart + 3) + 5,
2445 i.e., b - laststart.
2447 We insert this at the beginning of the loop
2448 so that if we fail during matching, we'll
2449 reinitialize the bounds. */
2450 insert_op2 (set_number_at, laststart, b - laststart,
2451 upper_bound - 1, b);
2456 beg_interval = NULL;
2461 /* If an invalid interval, match the characters as literals. */
2462 assert (beg_interval);
2464 beg_interval = NULL;
2466 /* normal_char and normal_backslash need `c'. */
2469 if (!(syntax & RE_NO_BK_BRACES))
2471 if (p > pattern && p[-1] == '\\')
2472 goto normal_backslash;
2477 /* There is no way to specify the before_dot and after_dot
2478 operators. rms says this is ok. --karl */
2486 BUF_PUSH_2 (syntaxspec, syntax_spec_code[c]);
2492 BUF_PUSH_2 (notsyntaxspec, syntax_spec_code[c]);
2499 BUF_PUSH (wordchar);
2505 BUF_PUSH (notwordchar);
2518 BUF_PUSH (wordbound);
2522 BUF_PUSH (notwordbound);
2533 case '1': case '2': case '3': case '4': case '5':
2534 case '6': case '7': case '8': case '9':
2535 if (syntax & RE_NO_BK_REFS)
2541 FREE_STACK_RETURN (REG_ESUBREG);
2543 /* Can't back reference to a subexpression if inside of it. */
2544 if (group_in_compile_stack (compile_stack, c1))
2548 BUF_PUSH_2 (duplicate, c1);
2554 if (syntax & RE_BK_PLUS_QM)
2557 goto normal_backslash;
2561 /* You might think it would be useful for \ to mean
2562 not to translate; but if we don't translate it
2563 it will never match anything. */
2571 /* Expects the character in `c'. */
2573 /* If no exactn currently being built. */
2576 /* If last exactn not at current position. */
2577 || pending_exact + *pending_exact + 1 != b
2579 /* We have only one byte following the exactn for the count. */
2580 || *pending_exact == (1 << BYTEWIDTH) - 1
2582 /* If followed by a repetition operator. */
2583 || *p == '*' || *p == '^'
2584 || ((syntax & RE_BK_PLUS_QM)
2585 ? *p == '\\' && (p[1] == '+' || p[1] == '?')
2586 : (*p == '+' || *p == '?'))
2587 || ((syntax & RE_INTERVALS)
2588 && ((syntax & RE_NO_BK_BRACES)
2590 : (p[0] == '\\' && p[1] == '{'))))
2592 /* Start building a new exactn. */
2596 BUF_PUSH_2 (exactn, 0);
2597 pending_exact = b - 1;
2604 } /* while p != pend */
2607 /* Through the pattern now. */
2610 STORE_JUMP (jump_past_alt, fixup_alt_jump, b);
2612 if (!COMPILE_STACK_EMPTY)
2613 FREE_STACK_RETURN (REG_EPAREN);
2615 /* If we don't want backtracking, force success
2616 the first time we reach the end of the compiled pattern. */
2617 if (syntax & RE_NO_POSIX_BACKTRACKING)
2620 free (compile_stack.stack);
2622 /* We have succeeded; set the length of the buffer. */
2623 bufp->used = b - bufp->buffer;
2628 DEBUG_PRINT1 ("\nCompiled pattern: \n");
2629 print_compiled_pattern (bufp);
2633 #ifndef MATCH_MAY_ALLOCATE
2634 /* Initialize the failure stack to the largest possible stack. This
2635 isn't necessary unless we're trying to avoid calling alloca in
2636 the search and match routines. */
2638 int num_regs = bufp->re_nsub + 1;
2640 /* Since DOUBLE_FAIL_STACK refuses to double only if the current size
2641 is strictly greater than re_max_failures, the largest possible stack
2642 is 2 * re_max_failures failure points. */
2643 if (fail_stack.size < (2 * re_max_failures * MAX_FAILURE_ITEMS))
2645 fail_stack.size = (2 * re_max_failures * MAX_FAILURE_ITEMS);
2648 if (! fail_stack.stack)
2650 = (fail_stack_elt_t *) xmalloc (fail_stack.size
2651 * sizeof (fail_stack_elt_t));
2654 = (fail_stack_elt_t *) xrealloc (fail_stack.stack,
2656 * sizeof (fail_stack_elt_t)));
2657 #else /* not emacs */
2658 if (! fail_stack.stack)
2660 = (fail_stack_elt_t *) malloc (fail_stack.size
2661 * sizeof (fail_stack_elt_t));
2664 = (fail_stack_elt_t *) realloc (fail_stack.stack,
2666 * sizeof (fail_stack_elt_t)));
2667 #endif /* not emacs */
2670 regex_grow_registers (num_regs);
2672 #endif /* not MATCH_MAY_ALLOCATE */
2675 } /* regex_compile */
2677 /* Subroutines for `regex_compile'. */
2679 /* Store OP at LOC followed by two-byte integer parameter ARG. */
2682 store_op1 (op, loc, arg)
2687 *loc = (unsigned char) op;
2688 STORE_NUMBER (loc + 1, arg);
2692 /* Like `store_op1', but for two two-byte parameters ARG1 and ARG2. */
2695 store_op2 (op, loc, arg1, arg2)
2700 *loc = (unsigned char) op;
2701 STORE_NUMBER (loc + 1, arg1);
2702 STORE_NUMBER (loc + 3, arg2);
2706 /* Copy the bytes from LOC to END to open up three bytes of space at LOC
2707 for OP followed by two-byte integer parameter ARG. */
2710 insert_op1 (op, loc, arg, end)
2716 register unsigned char *pfrom = end;
2717 register unsigned char *pto = end + 3;
2719 while (pfrom != loc)
2722 store_op1 (op, loc, arg);
2726 /* Like `insert_op1', but for two two-byte parameters ARG1 and ARG2. */
2729 insert_op2 (op, loc, arg1, arg2, end)
2735 register unsigned char *pfrom = end;
2736 register unsigned char *pto = end + 5;
2738 while (pfrom != loc)
2741 store_op2 (op, loc, arg1, arg2);
2745 /* P points to just after a ^ in PATTERN. Return true if that ^ comes
2746 after an alternative or a begin-subexpression. We assume there is at
2747 least one character before the ^. */
2750 at_begline_loc_p (pattern, p, syntax)
2751 const char *pattern, *p;
2752 reg_syntax_t syntax;
2754 const char *prev = p - 2;
2755 boolean prev_prev_backslash = prev > pattern && prev[-1] == '\\';
2758 /* After a subexpression? */
2759 (*prev == '(' && (syntax & RE_NO_BK_PARENS || prev_prev_backslash))
2760 /* After an alternative? */
2761 || (*prev == '|' && (syntax & RE_NO_BK_VBAR || prev_prev_backslash));
2765 /* The dual of at_begline_loc_p. This one is for $. We assume there is
2766 at least one character after the $, i.e., `P < PEND'. */
2769 at_endline_loc_p (p, pend, syntax)
2770 const char *p, *pend;
2773 const char *next = p;
2774 boolean next_backslash = *next == '\\';
2775 const char *next_next = p + 1 < pend ? p + 1 : 0;
2778 /* Before a subexpression? */
2779 (syntax & RE_NO_BK_PARENS ? *next == ')'
2780 : next_backslash && next_next && *next_next == ')')
2781 /* Before an alternative? */
2782 || (syntax & RE_NO_BK_VBAR ? *next == '|'
2783 : next_backslash && next_next && *next_next == '|');
2787 /* Returns true if REGNUM is in one of COMPILE_STACK's elements and
2788 false if it's not. */
2791 group_in_compile_stack (compile_stack, regnum)
2792 compile_stack_type compile_stack;
2797 for (this_element = compile_stack.avail - 1;
2800 if (compile_stack.stack[this_element].regnum == regnum)
2807 /* Read the ending character of a range (in a bracket expression) from the
2808 uncompiled pattern *P_PTR (which ends at PEND). We assume the
2809 starting character is in `P[-2]'. (`P[-1]' is the character `-'.)
2810 Then we set the translation of all bits between the starting and
2811 ending characters (inclusive) in the compiled pattern B.
2813 Return an error code.
2815 We use these short variable names so we can use the same macros as
2816 `regex_compile' itself. */
2818 static reg_errcode_t
2819 compile_range (p_ptr, pend, translate, syntax, b)
2820 const char **p_ptr, *pend;
2822 reg_syntax_t syntax;
2827 const char *p = *p_ptr;
2828 int range_start, range_end;
2833 /* Even though the pattern is a signed `char *', we need to fetch
2834 with unsigned char *'s; if the high bit of the pattern character
2835 is set, the range endpoints will be negative if we fetch using a
2838 We also want to fetch the endpoints without translating them; the
2839 appropriate translation is done in the bit-setting loop below. */
2840 /* The SVR4 compiler on the 3B2 had trouble with unsigned const char *. */
2841 range_start = ((const unsigned char *) p)[-2];
2842 range_end = ((const unsigned char *) p)[0];
2844 /* Have to increment the pointer into the pattern string, so the
2845 caller isn't still at the ending character. */
2848 /* If the start is after the end, the range is empty. */
2849 if (range_start > range_end)
2850 return syntax & RE_NO_EMPTY_RANGES ? REG_ERANGE : REG_NOERROR;
2852 /* Here we see why `this_char' has to be larger than an `unsigned
2853 char' -- the range is inclusive, so if `range_end' == 0xff
2854 (assuming 8-bit characters), we would otherwise go into an infinite
2855 loop, since all characters <= 0xff. */
2856 for (this_char = range_start; this_char <= range_end; this_char++)
2858 SET_LIST_BIT (TRANSLATE (this_char));
2864 /* re_compile_fastmap computes a ``fastmap'' for the compiled pattern in
2865 BUFP. A fastmap records which of the (1 << BYTEWIDTH) possible
2866 characters can start a string that matches the pattern. This fastmap
2867 is used by re_search to skip quickly over impossible starting points.
2869 The caller must supply the address of a (1 << BYTEWIDTH)-byte data
2870 area as BUFP->fastmap.
2872 We set the `fastmap', `fastmap_accurate', and `can_be_null' fields in
2875 Returns 0 if we succeed, -2 if an internal error. */
2878 re_compile_fastmap (bufp)
2879 struct re_pattern_buffer *bufp;
2882 #ifdef MATCH_MAY_ALLOCATE
2883 fail_stack_type fail_stack;
2885 #ifndef REGEX_MALLOC
2888 /* We don't push any register information onto the failure stack. */
2889 unsigned num_regs = 0;
2891 register char *fastmap = bufp->fastmap;
2892 unsigned char *pattern = bufp->buffer;
2893 unsigned long size = bufp->used;
2894 unsigned char *p = pattern;
2895 register unsigned char *pend = pattern + size;
2897 /* This holds the pointer to the failure stack, when
2898 it is allocated relocatably. */
2899 fail_stack_elt_t *failure_stack_ptr;
2901 /* Assume that each path through the pattern can be null until
2902 proven otherwise. We set this false at the bottom of switch
2903 statement, to which we get only if a particular path doesn't
2904 match the empty string. */
2905 boolean path_can_be_null = true;
2907 /* We aren't doing a `succeed_n' to begin with. */
2908 boolean succeed_n_p = false;
2910 assert (fastmap != NULL && p != NULL);
2913 bzero (fastmap, 1 << BYTEWIDTH); /* Assume nothing's valid. */
2914 bufp->fastmap_accurate = 1; /* It will be when we're done. */
2915 bufp->can_be_null = 0;
2919 if (p == pend || *p == succeed)
2921 /* We have reached the (effective) end of pattern. */
2922 if (!FAIL_STACK_EMPTY ())
2924 bufp->can_be_null |= path_can_be_null;
2926 /* Reset for next path. */
2927 path_can_be_null = true;
2929 p = fail_stack.stack[--fail_stack.avail].pointer;
2937 /* We should never be about to go beyond the end of the pattern. */
2940 switch (SWITCH_ENUM_CAST ((re_opcode_t) *p++))
2943 /* I guess the idea here is to simply not bother with a fastmap
2944 if a backreference is used, since it's too hard to figure out
2945 the fastmap for the corresponding group. Setting
2946 `can_be_null' stops `re_search_2' from using the fastmap, so
2947 that is all we do. */
2949 bufp->can_be_null = 1;
2953 /* Following are the cases which match a character. These end
2962 for (j = *p++ * BYTEWIDTH - 1; j >= 0; j--)
2963 if (p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH)))
2969 /* Chars beyond end of map must be allowed. */
2970 for (j = *p * BYTEWIDTH; j < (1 << BYTEWIDTH); j++)
2973 for (j = *p++ * BYTEWIDTH - 1; j >= 0; j--)
2974 if (!(p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH))))
2980 for (j = 0; j < (1 << BYTEWIDTH); j++)
2981 if (SYNTAX (j) == Sword)
2987 for (j = 0; j < (1 << BYTEWIDTH); j++)
2988 if (SYNTAX (j) != Sword)
2995 int fastmap_newline = fastmap['\n'];
2997 /* `.' matches anything ... */
2998 for (j = 0; j < (1 << BYTEWIDTH); j++)
3001 /* ... except perhaps newline. */
3002 if (!(bufp->syntax & RE_DOT_NEWLINE))
3003 fastmap['\n'] = fastmap_newline;
3005 /* Return if we have already set `can_be_null'; if we have,
3006 then the fastmap is irrelevant. Something's wrong here. */
3007 else if (bufp->can_be_null)
3010 /* Otherwise, have to check alternative paths. */
3017 for (j = 0; j < (1 << BYTEWIDTH); j++)
3018 if (SYNTAX (j) == (enum syntaxcode) k)
3025 for (j = 0; j < (1 << BYTEWIDTH); j++)
3026 if (SYNTAX (j) != (enum syntaxcode) k)
3031 /* All cases after this match the empty string. These end with
3051 case push_dummy_failure:
3056 case pop_failure_jump:
3057 case maybe_pop_jump:
3060 case dummy_failure_jump:
3061 EXTRACT_NUMBER_AND_INCR (j, p);
3066 /* Jump backward implies we just went through the body of a
3067 loop and matched nothing. Opcode jumped to should be
3068 `on_failure_jump' or `succeed_n'. Just treat it like an
3069 ordinary jump. For a * loop, it has pushed its failure
3070 point already; if so, discard that as redundant. */
3071 if ((re_opcode_t) *p != on_failure_jump
3072 && (re_opcode_t) *p != succeed_n)
3076 EXTRACT_NUMBER_AND_INCR (j, p);
3079 /* If what's on the stack is where we are now, pop it. */
3080 if (!FAIL_STACK_EMPTY ()
3081 && fail_stack.stack[fail_stack.avail - 1].pointer == p)
3087 case on_failure_jump:
3088 case on_failure_keep_string_jump:
3089 handle_on_failure_jump:
3090 EXTRACT_NUMBER_AND_INCR (j, p);
3092 /* For some patterns, e.g., `(a?)?', `p+j' here points to the
3093 end of the pattern. We don't want to push such a point,
3094 since when we restore it above, entering the switch will
3095 increment `p' past the end of the pattern. We don't need
3096 to push such a point since we obviously won't find any more
3097 fastmap entries beyond `pend'. Such a pattern can match
3098 the null string, though. */
3101 if (!PUSH_PATTERN_OP (p + j, fail_stack))
3103 RESET_FAIL_STACK ();
3108 bufp->can_be_null = 1;
3112 EXTRACT_NUMBER_AND_INCR (k, p); /* Skip the n. */
3113 succeed_n_p = false;
3120 /* Get to the number of times to succeed. */
3123 /* Increment p past the n for when k != 0. */
3124 EXTRACT_NUMBER_AND_INCR (k, p);
3128 succeed_n_p = true; /* Spaghetti code alert. */
3129 goto handle_on_failure_jump;
3146 abort (); /* We have listed all the cases. */
3149 /* Getting here means we have found the possible starting
3150 characters for one path of the pattern -- and that the empty
3151 string does not match. We need not follow this path further.
3152 Instead, look at the next alternative (remembered on the
3153 stack), or quit if no more. The test at the top of the loop
3154 does these things. */
3155 path_can_be_null = false;
3159 /* Set `can_be_null' for the last path (also the first path, if the
3160 pattern is empty). */
3161 bufp->can_be_null |= path_can_be_null;
3164 RESET_FAIL_STACK ();
3166 } /* re_compile_fastmap */
3168 /* Set REGS to hold NUM_REGS registers, storing them in STARTS and
3169 ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use
3170 this memory for recording register information. STARTS and ENDS
3171 must be allocated using the malloc library routine, and must each
3172 be at least NUM_REGS * sizeof (regoff_t) bytes long.
3174 If NUM_REGS == 0, then subsequent matches should allocate their own
3177 Unless this function is called, the first search or match using
3178 PATTERN_BUFFER will allocate its own register data, without
3179 freeing the old data. */
3182 re_set_registers (bufp, regs, num_regs, starts, ends)
3183 struct re_pattern_buffer *bufp;
3184 struct re_registers *regs;
3186 regoff_t *starts, *ends;
3190 bufp->regs_allocated = REGS_REALLOCATE;
3191 regs->num_regs = num_regs;
3192 regs->start = starts;
3197 bufp->regs_allocated = REGS_UNALLOCATED;
3199 regs->start = regs->end = (regoff_t *) 0;
3203 /* Searching routines. */
3205 /* Like re_search_2, below, but only one string is specified, and
3206 doesn't let you say where to stop matching. */
3209 re_search (bufp, string, size, startpos, range, regs)
3210 struct re_pattern_buffer *bufp;
3212 int size, startpos, range;
3213 struct re_registers *regs;
3215 return re_search_2 (bufp, NULL, 0, string, size, startpos, range,
3220 /* Using the compiled pattern in BUFP->buffer, first tries to match the
3221 virtual concatenation of STRING1 and STRING2, starting first at index
3222 STARTPOS, then at STARTPOS + 1, and so on.
3224 STRING1 and STRING2 have length SIZE1 and SIZE2, respectively.
3226 RANGE is how far to scan while trying to match. RANGE = 0 means try
3227 only at STARTPOS; in general, the last start tried is STARTPOS +
3230 In REGS, return the indices of the virtual concatenation of STRING1
3231 and STRING2 that matched the entire BUFP->buffer and its contained
3234 Do not consider matching one past the index STOP in the virtual
3235 concatenation of STRING1 and STRING2.
3237 We return either the position in the strings at which the match was
3238 found, -1 if no match, or -2 if error (such as failure
3242 re_search_2 (bufp, string1, size1, string2, size2, startpos, range, regs, stop)
3243 struct re_pattern_buffer *bufp;
3244 const char *string1, *string2;
3248 struct re_registers *regs;
3252 register char *fastmap = bufp->fastmap;
3253 register char *translate = bufp->translate;
3254 int total_size = size1 + size2;
3255 int endpos = startpos + range;
3257 /* Check for out-of-range STARTPOS. */
3258 if (startpos < 0 || startpos > total_size)
3261 /* Fix up RANGE if it might eventually take us outside
3262 the virtual concatenation of STRING1 and STRING2. */
3264 range = -1 - startpos;
3265 else if (endpos > total_size)
3266 range = total_size - startpos;
3268 /* If the search isn't to be a backwards one, don't waste time in a
3269 search for a pattern that must be anchored. */
3270 if (bufp->used > 0 && (re_opcode_t) bufp->buffer[0] == begbuf && range > 0)
3279 /* In a forward search for something that starts with \=.
3280 don't keep searching past point. */
3281 if (bufp->used > 0 && (re_opcode_t) bufp->buffer[0] == at_dot && range > 0)
3283 range = PT - startpos;
3289 /* Update the fastmap now if not correct already. */
3290 if (fastmap && !bufp->fastmap_accurate)
3291 if (re_compile_fastmap (bufp) == -2)
3294 /* Loop through the string, looking for a place to start matching. */
3297 /* If a fastmap is supplied, skip quickly over characters that
3298 cannot be the start of a match. If the pattern can match the
3299 null string, however, we don't need to skip characters; we want
3300 the first null string. */
3301 if (fastmap && startpos < total_size && !bufp->can_be_null)
3303 if (range > 0) /* Searching forwards. */
3305 register const char *d;
3306 register int lim = 0;
3309 if (startpos < size1 && startpos + range >= size1)
3310 lim = range - (size1 - startpos);
3312 d = (startpos >= size1 ? string2 - size1 : string1) + startpos;
3314 /* Written out as an if-else to avoid testing `translate'
3318 && !fastmap[(unsigned char)
3319 translate[(unsigned char) *d++]])
3322 while (range > lim && !fastmap[(unsigned char) *d++])
3325 startpos += irange - range;
3327 else /* Searching backwards. */
3329 register char c = (size1 == 0 || startpos >= size1
3330 ? string2[startpos - size1]
3331 : string1[startpos]);
3333 if (!fastmap[(unsigned char) TRANSLATE (c)])
3338 /* If can't match the null string, and that's all we have left, fail. */
3339 if (range >= 0 && startpos == total_size && fastmap
3340 && !bufp->can_be_null)
3343 val = re_match_2_internal (bufp, string1, size1, string2, size2,
3344 startpos, regs, stop);
3345 #ifndef REGEX_MALLOC
3374 /* Declarations and macros for re_match_2. */
3376 static int bcmp_translate ();
3377 static boolean alt_match_null_string_p (),
3378 common_op_match_null_string_p (),
3379 group_match_null_string_p ();
3381 /* This converts PTR, a pointer into one of the search strings `string1'
3382 and `string2' into an offset from the beginning of that string. */
3383 #define POINTER_TO_OFFSET(ptr) \
3384 (FIRST_STRING_P (ptr) \
3385 ? ((regoff_t) ((ptr) - string1)) \
3386 : ((regoff_t) ((ptr) - string2 + size1)))
3388 /* Macros for dealing with the split strings in re_match_2. */
3390 #define MATCHING_IN_FIRST_STRING (dend == end_match_1)
3392 /* Call before fetching a character with *d. This switches over to
3393 string2 if necessary. */
3394 #define PREFETCH() \
3397 /* End of string2 => fail. */ \
3398 if (dend == end_match_2) \
3400 /* End of string1 => advance to string2. */ \
3402 dend = end_match_2; \
3406 /* Test if at very beginning or at very end of the virtual concatenation
3407 of `string1' and `string2'. If only one string, it's `string2'. */
3408 #define AT_STRINGS_BEG(d) ((d) == (size1 ? string1 : string2) || !size2)
3409 #define AT_STRINGS_END(d) ((d) == end2)
3412 /* Test if D points to a character which is word-constituent. We have
3413 two special cases to check for: if past the end of string1, look at
3414 the first character in string2; and if before the beginning of
3415 string2, look at the last character in string1. */
3416 #define WORDCHAR_P(d) \
3417 (SYNTAX ((d) == end1 ? *string2 \
3418 : (d) == string2 - 1 ? *(end1 - 1) : *(d)) \
3421 /* Test if the character before D and the one at D differ with respect
3422 to being word-constituent. */
3423 #define AT_WORD_BOUNDARY(d) \
3424 (AT_STRINGS_BEG (d) || AT_STRINGS_END (d) \
3425 || WORDCHAR_P (d - 1) != WORDCHAR_P (d))
3428 /* Free everything we malloc. */
3429 #ifdef MATCH_MAY_ALLOCATE
3430 #define FREE_VAR(var) if (var) REGEX_FREE (var); var = NULL
3431 #define FREE_VARIABLES() \
3433 REGEX_FREE_STACK (fail_stack.stack); \
3434 FREE_VAR (regstart); \
3435 FREE_VAR (regend); \
3436 FREE_VAR (old_regstart); \
3437 FREE_VAR (old_regend); \
3438 FREE_VAR (best_regstart); \
3439 FREE_VAR (best_regend); \
3440 FREE_VAR (reg_info); \
3441 FREE_VAR (reg_dummy); \
3442 FREE_VAR (reg_info_dummy); \
3445 #define FREE_VARIABLES() ((void)0) /* Do nothing! But inhibit gcc warning. */
3446 #endif /* not MATCH_MAY_ALLOCATE */
3448 /* These values must meet several constraints. They must not be valid
3449 register values; since we have a limit of 255 registers (because
3450 we use only one byte in the pattern for the register number), we can
3451 use numbers larger than 255. They must differ by 1, because of
3452 NUM_FAILURE_ITEMS above. And the value for the lowest register must
3453 be larger than the value for the highest register, so we do not try
3454 to actually save any registers when none are active. */
3455 #define NO_HIGHEST_ACTIVE_REG (1 << BYTEWIDTH)
3456 #define NO_LOWEST_ACTIVE_REG (NO_HIGHEST_ACTIVE_REG + 1)
3458 /* Matching routines. */
3460 #ifndef emacs /* Emacs never uses this. */
3461 /* re_match is like re_match_2 except it takes only a single string. */
3464 re_match (bufp, string, size, pos, regs)
3465 struct re_pattern_buffer *bufp;
3468 struct re_registers *regs;
3470 int result = re_match_2_internal (bufp, NULL, 0, string, size,
3475 #endif /* not emacs */
3478 /* re_match_2 matches the compiled pattern in BUFP against the
3479 the (virtual) concatenation of STRING1 and STRING2 (of length SIZE1
3480 and SIZE2, respectively). We start matching at POS, and stop
3483 If REGS is non-null and the `no_sub' field of BUFP is nonzero, we
3484 store offsets for the substring each group matched in REGS. See the
3485 documentation for exactly how many groups we fill.
3487 We return -1 if no match, -2 if an internal error (such as the
3488 failure stack overflowing). Otherwise, we return the length of the
3489 matched substring. */
3492 re_match_2 (bufp, string1, size1, string2, size2, pos, regs, stop)
3493 struct re_pattern_buffer *bufp;
3494 const char *string1, *string2;
3497 struct re_registers *regs;
3500 int result = re_match_2_internal (bufp, string1, size1, string2, size2,
3506 /* This is a separate function so that we can force an alloca cleanup
3509 re_match_2_internal (bufp, string1, size1, string2, size2, pos, regs, stop)
3510 struct re_pattern_buffer *bufp;
3511 const char *string1, *string2;
3514 struct re_registers *regs;
3517 /* General temporaries. */
3521 /* Just past the end of the corresponding string. */
3522 const char *end1, *end2;
3524 /* Pointers into string1 and string2, just past the last characters in
3525 each to consider matching. */
3526 const char *end_match_1, *end_match_2;
3528 /* Where we are in the data, and the end of the current string. */
3529 const char *d, *dend;
3531 /* Where we are in the pattern, and the end of the pattern. */
3532 unsigned char *p = bufp->buffer;
3533 register unsigned char *pend = p + bufp->used;
3535 /* Mark the opcode just after a start_memory, so we can test for an
3536 empty subpattern when we get to the stop_memory. */
3537 unsigned char *just_past_start_mem = 0;
3539 /* We use this to map every character in the string. */
3540 char *translate = bufp->translate;
3542 /* Failure point stack. Each place that can handle a failure further
3543 down the line pushes a failure point on this stack. It consists of
3544 restart, regend, and reg_info for all registers corresponding to
3545 the subexpressions we're currently inside, plus the number of such
3546 registers, and, finally, two char *'s. The first char * is where
3547 to resume scanning the pattern; the second one is where to resume
3548 scanning the strings. If the latter is zero, the failure point is
3549 a ``dummy''; if a failure happens and the failure point is a dummy,
3550 it gets discarded and the next next one is tried. */
3551 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
3552 fail_stack_type fail_stack;
3555 static unsigned failure_id = 0;
3556 unsigned nfailure_points_pushed = 0, nfailure_points_popped = 0;
3559 /* This holds the pointer to the failure stack, when
3560 it is allocated relocatably. */
3561 fail_stack_elt_t *failure_stack_ptr;
3563 /* We fill all the registers internally, independent of what we
3564 return, for use in backreferences. The number here includes
3565 an element for register zero. */
3566 unsigned num_regs = bufp->re_nsub + 1;
3568 /* The currently active registers. */
3569 unsigned lowest_active_reg = NO_LOWEST_ACTIVE_REG;
3570 unsigned highest_active_reg = NO_HIGHEST_ACTIVE_REG;
3572 /* Information on the contents of registers. These are pointers into
3573 the input strings; they record just what was matched (on this
3574 attempt) by a subexpression part of the pattern, that is, the
3575 regnum-th regstart pointer points to where in the pattern we began
3576 matching and the regnum-th regend points to right after where we
3577 stopped matching the regnum-th subexpression. (The zeroth register
3578 keeps track of what the whole pattern matches.) */
3579 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
3580 const char **regstart, **regend;
3583 /* If a group that's operated upon by a repetition operator fails to
3584 match anything, then the register for its start will need to be
3585 restored because it will have been set to wherever in the string we
3586 are when we last see its open-group operator. Similarly for a
3588 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
3589 const char **old_regstart, **old_regend;
3592 /* The is_active field of reg_info helps us keep track of which (possibly
3593 nested) subexpressions we are currently in. The matched_something
3594 field of reg_info[reg_num] helps us tell whether or not we have
3595 matched any of the pattern so far this time through the reg_num-th
3596 subexpression. These two fields get reset each time through any
3597 loop their register is in. */
3598 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
3599 register_info_type *reg_info;
3602 /* The following record the register info as found in the above
3603 variables when we find a match better than any we've seen before.
3604 This happens as we backtrack through the failure points, which in
3605 turn happens only if we have not yet matched the entire string. */
3606 unsigned best_regs_set = false;
3607 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
3608 const char **best_regstart, **best_regend;
3611 /* Logically, this is `best_regend[0]'. But we don't want to have to
3612 allocate space for that if we're not allocating space for anything
3613 else (see below). Also, we never need info about register 0 for
3614 any of the other register vectors, and it seems rather a kludge to
3615 treat `best_regend' differently than the rest. So we keep track of
3616 the end of the best match so far in a separate variable. We
3617 initialize this to NULL so that when we backtrack the first time
3618 and need to test it, it's not garbage. */
3619 const char *match_end = NULL;
3621 /* This helps SET_REGS_MATCHED avoid doing redundant work. */
3622 int set_regs_matched_done = 0;
3624 /* Used when we pop values we don't care about. */
3625 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
3626 const char **reg_dummy;
3627 register_info_type *reg_info_dummy;
3631 /* Counts the total number of registers pushed. */
3632 unsigned num_regs_pushed = 0;
3635 DEBUG_PRINT1 ("\n\nEntering re_match_2.\n");
3639 #ifdef MATCH_MAY_ALLOCATE
3640 /* Do not bother to initialize all the register variables if there are
3641 no groups in the pattern, as it takes a fair amount of time. If
3642 there are groups, we include space for register 0 (the whole
3643 pattern), even though we never use it, since it simplifies the
3644 array indexing. We should fix this. */
3647 regstart = REGEX_TALLOC (num_regs, const char *);
3648 regend = REGEX_TALLOC (num_regs, const char *);
3649 old_regstart = REGEX_TALLOC (num_regs, const char *);
3650 old_regend = REGEX_TALLOC (num_regs, const char *);
3651 best_regstart = REGEX_TALLOC (num_regs, const char *);
3652 best_regend = REGEX_TALLOC (num_regs, const char *);
3653 reg_info = REGEX_TALLOC (num_regs, register_info_type);
3654 reg_dummy = REGEX_TALLOC (num_regs, const char *);
3655 reg_info_dummy = REGEX_TALLOC (num_regs, register_info_type);
3657 if (!(regstart && regend && old_regstart && old_regend && reg_info
3658 && best_regstart && best_regend && reg_dummy && reg_info_dummy))
3666 /* We must initialize all our variables to NULL, so that
3667 `FREE_VARIABLES' doesn't try to free them. */
3668 regstart = regend = old_regstart = old_regend = best_regstart
3669 = best_regend = reg_dummy = NULL;
3670 reg_info = reg_info_dummy = (register_info_type *) NULL;
3672 #endif /* MATCH_MAY_ALLOCATE */
3674 /* The starting position is bogus. */
3675 if (pos < 0 || pos > size1 + size2)
3681 /* Initialize subexpression text positions to -1 to mark ones that no
3682 start_memory/stop_memory has been seen for. Also initialize the
3683 register information struct. */
3684 for (mcnt = 1; mcnt < num_regs; mcnt++)
3686 regstart[mcnt] = regend[mcnt]
3687 = old_regstart[mcnt] = old_regend[mcnt] = REG_UNSET_VALUE;
3689 REG_MATCH_NULL_STRING_P (reg_info[mcnt]) = MATCH_NULL_UNSET_VALUE;
3690 IS_ACTIVE (reg_info[mcnt]) = 0;
3691 MATCHED_SOMETHING (reg_info[mcnt]) = 0;
3692 EVER_MATCHED_SOMETHING (reg_info[mcnt]) = 0;
3695 /* We move `string1' into `string2' if the latter's empty -- but not if
3696 `string1' is null. */
3697 if (size2 == 0 && string1 != NULL)
3704 end1 = string1 + size1;
3705 end2 = string2 + size2;
3707 /* Compute where to stop matching, within the two strings. */
3710 end_match_1 = string1 + stop;
3711 end_match_2 = string2;
3716 end_match_2 = string2 + stop - size1;
3719 /* `p' scans through the pattern as `d' scans through the data.
3720 `dend' is the end of the input string that `d' points within. `d'
3721 is advanced into the following input string whenever necessary, but
3722 this happens before fetching; therefore, at the beginning of the
3723 loop, `d' can be pointing at the end of a string, but it cannot
3725 if (size1 > 0 && pos <= size1)
3732 d = string2 + pos - size1;
3736 DEBUG_PRINT1 ("The compiled pattern is: ");
3737 DEBUG_PRINT_COMPILED_PATTERN (bufp, p, pend);
3738 DEBUG_PRINT1 ("The string to match is: `");
3739 DEBUG_PRINT_DOUBLE_STRING (d, string1, size1, string2, size2);
3740 DEBUG_PRINT1 ("'\n");
3742 /* This loops over pattern commands. It exits by returning from the
3743 function if the match is complete, or it drops through if the match
3744 fails at this starting point in the input data. */
3747 DEBUG_PRINT2 ("\n0x%x: ", p);
3750 { /* End of pattern means we might have succeeded. */
3751 DEBUG_PRINT1 ("end of pattern ... ");
3753 /* If we haven't matched the entire string, and we want the
3754 longest match, try backtracking. */
3755 if (d != end_match_2)
3757 /* 1 if this match ends in the same string (string1 or string2)
3758 as the best previous match. */
3759 boolean same_str_p = (FIRST_STRING_P (match_end)
3760 == MATCHING_IN_FIRST_STRING);
3761 /* 1 if this match is the best seen so far. */
3762 boolean best_match_p;
3764 /* AIX compiler got confused when this was combined
3765 with the previous declaration. */
3767 best_match_p = d > match_end;
3769 best_match_p = !MATCHING_IN_FIRST_STRING;
3771 DEBUG_PRINT1 ("backtracking.\n");
3773 if (!FAIL_STACK_EMPTY ())
3774 { /* More failure points to try. */
3776 /* If exceeds best match so far, save it. */
3777 if (!best_regs_set || best_match_p)
3779 best_regs_set = true;
3782 DEBUG_PRINT1 ("\nSAVING match as best so far.\n");
3784 for (mcnt = 1; mcnt < num_regs; mcnt++)
3786 best_regstart[mcnt] = regstart[mcnt];
3787 best_regend[mcnt] = regend[mcnt];
3793 /* If no failure points, don't restore garbage. And if
3794 last match is real best match, don't restore second
3796 else if (best_regs_set && !best_match_p)
3799 /* Restore best match. It may happen that `dend ==
3800 end_match_1' while the restored d is in string2.
3801 For example, the pattern `x.*y.*z' against the
3802 strings `x-' and `y-z-', if the two strings are
3803 not consecutive in memory. */
3804 DEBUG_PRINT1 ("Restoring best registers.\n");
3807 dend = ((d >= string1 && d <= end1)
3808 ? end_match_1 : end_match_2);
3810 for (mcnt = 1; mcnt < num_regs; mcnt++)
3812 regstart[mcnt] = best_regstart[mcnt];
3813 regend[mcnt] = best_regend[mcnt];
3816 } /* d != end_match_2 */
3819 DEBUG_PRINT1 ("Accepting match.\n");
3821 /* If caller wants register contents data back, do it. */
3822 if (regs && !bufp->no_sub)
3824 /* Have the register data arrays been allocated? */
3825 if (bufp->regs_allocated == REGS_UNALLOCATED)
3826 { /* No. So allocate them with malloc. We need one
3827 extra element beyond `num_regs' for the `-1' marker
3829 regs->num_regs = MAX (RE_NREGS, num_regs + 1);
3830 regs->start = TALLOC (regs->num_regs, regoff_t);
3831 regs->end = TALLOC (regs->num_regs, regoff_t);
3832 if (regs->start == NULL || regs->end == NULL)
3837 bufp->regs_allocated = REGS_REALLOCATE;
3839 else if (bufp->regs_allocated == REGS_REALLOCATE)
3840 { /* Yes. If we need more elements than were already
3841 allocated, reallocate them. If we need fewer, just
3843 if (regs->num_regs < num_regs + 1)
3845 regs->num_regs = num_regs + 1;
3846 RETALLOC (regs->start, regs->num_regs, regoff_t);
3847 RETALLOC (regs->end, regs->num_regs, regoff_t);
3848 if (regs->start == NULL || regs->end == NULL)
3857 /* These braces fend off a "empty body in an else-statement"
3858 warning under GCC when assert expands to nothing. */
3859 assert (bufp->regs_allocated == REGS_FIXED);
3862 /* Convert the pointer data in `regstart' and `regend' to
3863 indices. Register zero has to be set differently,
3864 since we haven't kept track of any info for it. */
3865 if (regs->num_regs > 0)
3867 regs->start[0] = pos;
3868 regs->end[0] = (MATCHING_IN_FIRST_STRING
3869 ? ((regoff_t) (d - string1))
3870 : ((regoff_t) (d - string2 + size1)));
3873 /* Go through the first `min (num_regs, regs->num_regs)'
3874 registers, since that is all we initialized. */
3875 for (mcnt = 1; mcnt < MIN (num_regs, regs->num_regs); mcnt++)
3877 if (REG_UNSET (regstart[mcnt]) || REG_UNSET (regend[mcnt]))
3878 regs->start[mcnt] = regs->end[mcnt] = -1;
3882 = (regoff_t) POINTER_TO_OFFSET (regstart[mcnt]);
3884 = (regoff_t) POINTER_TO_OFFSET (regend[mcnt]);
3888 /* If the regs structure we return has more elements than
3889 were in the pattern, set the extra elements to -1. If
3890 we (re)allocated the registers, this is the case,
3891 because we always allocate enough to have at least one
3893 for (mcnt = num_regs; mcnt < regs->num_regs; mcnt++)
3894 regs->start[mcnt] = regs->end[mcnt] = -1;
3895 } /* regs && !bufp->no_sub */
3897 DEBUG_PRINT4 ("%u failure points pushed, %u popped (%u remain).\n",
3898 nfailure_points_pushed, nfailure_points_popped,
3899 nfailure_points_pushed - nfailure_points_popped);
3900 DEBUG_PRINT2 ("%u registers pushed.\n", num_regs_pushed);
3902 mcnt = d - pos - (MATCHING_IN_FIRST_STRING
3906 DEBUG_PRINT2 ("Returning %d from re_match_2.\n", mcnt);
3912 /* Otherwise match next pattern command. */
3913 switch (SWITCH_ENUM_CAST ((re_opcode_t) *p++))
3915 /* Ignore these. Used to ignore the n of succeed_n's which
3916 currently have n == 0. */
3918 DEBUG_PRINT1 ("EXECUTING no_op.\n");
3922 DEBUG_PRINT1 ("EXECUTING succeed.\n");
3925 /* Match the next n pattern characters exactly. The following
3926 byte in the pattern defines n, and the n bytes after that
3927 are the characters to match. */
3930 DEBUG_PRINT2 ("EXECUTING exactn %d.\n", mcnt);
3932 /* This is written out as an if-else so we don't waste time
3933 testing `translate' inside the loop. */
3939 if (translate[(unsigned char) *d++] != (char) *p++)
3949 if (*d++ != (char) *p++) goto fail;
3953 SET_REGS_MATCHED ();
3957 /* Match any character except possibly a newline or a null. */
3959 DEBUG_PRINT1 ("EXECUTING anychar.\n");
3963 if ((!(bufp->syntax & RE_DOT_NEWLINE) && TRANSLATE (*d) == '\n')
3964 || (bufp->syntax & RE_DOT_NOT_NULL && TRANSLATE (*d) == '\000'))
3967 SET_REGS_MATCHED ();
3968 DEBUG_PRINT2 (" Matched `%d'.\n", *d);
3976 register unsigned char c;
3977 boolean not = (re_opcode_t) *(p - 1) == charset_not;
3979 DEBUG_PRINT2 ("EXECUTING charset%s.\n", not ? "_not" : "");
3982 c = TRANSLATE (*d); /* The character to match. */
3984 /* Cast to `unsigned' instead of `unsigned char' in case the
3985 bit list is a full 32 bytes long. */
3986 if (c < (unsigned) (*p * BYTEWIDTH)
3987 && p[1 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH)))
3992 if (!not) goto fail;
3994 SET_REGS_MATCHED ();
4000 /* The beginning of a group is represented by start_memory.
4001 The arguments are the register number in the next byte, and the
4002 number of groups inner to this one in the next. The text
4003 matched within the group is recorded (in the internal
4004 registers data structure) under the register number. */
4006 DEBUG_PRINT3 ("EXECUTING start_memory %d (%d):\n", *p, p[1]);
4008 /* Find out if this group can match the empty string. */
4009 p1 = p; /* To send to group_match_null_string_p. */
4011 if (REG_MATCH_NULL_STRING_P (reg_info[*p]) == MATCH_NULL_UNSET_VALUE)
4012 REG_MATCH_NULL_STRING_P (reg_info[*p])
4013 = group_match_null_string_p (&p1, pend, reg_info);
4015 /* Save the position in the string where we were the last time
4016 we were at this open-group operator in case the group is
4017 operated upon by a repetition operator, e.g., with `(a*)*b'
4018 against `ab'; then we want to ignore where we are now in
4019 the string in case this attempt to match fails. */
4020 old_regstart[*p] = REG_MATCH_NULL_STRING_P (reg_info[*p])
4021 ? REG_UNSET (regstart[*p]) ? d : regstart[*p]
4023 DEBUG_PRINT2 (" old_regstart: %d\n",
4024 POINTER_TO_OFFSET (old_regstart[*p]));
4027 DEBUG_PRINT2 (" regstart: %d\n", POINTER_TO_OFFSET (regstart[*p]));
4029 IS_ACTIVE (reg_info[*p]) = 1;
4030 MATCHED_SOMETHING (reg_info[*p]) = 0;
4032 /* Clear this whenever we change the register activity status. */
4033 set_regs_matched_done = 0;
4035 /* This is the new highest active register. */
4036 highest_active_reg = *p;
4038 /* If nothing was active before, this is the new lowest active
4040 if (lowest_active_reg == NO_LOWEST_ACTIVE_REG)
4041 lowest_active_reg = *p;
4043 /* Move past the register number and inner group count. */
4045 just_past_start_mem = p;
4050 /* The stop_memory opcode represents the end of a group. Its
4051 arguments are the same as start_memory's: the register
4052 number, and the number of inner groups. */
4054 DEBUG_PRINT3 ("EXECUTING stop_memory %d (%d):\n", *p, p[1]);
4056 /* We need to save the string position the last time we were at
4057 this close-group operator in case the group is operated
4058 upon by a repetition operator, e.g., with `((a*)*(b*)*)*'
4059 against `aba'; then we want to ignore where we are now in
4060 the string in case this attempt to match fails. */
4061 old_regend[*p] = REG_MATCH_NULL_STRING_P (reg_info[*p])
4062 ? REG_UNSET (regend[*p]) ? d : regend[*p]
4064 DEBUG_PRINT2 (" old_regend: %d\n",
4065 POINTER_TO_OFFSET (old_regend[*p]));
4068 DEBUG_PRINT2 (" regend: %d\n", POINTER_TO_OFFSET (regend[*p]));
4070 /* This register isn't active anymore. */
4071 IS_ACTIVE (reg_info[*p]) = 0;
4073 /* Clear this whenever we change the register activity status. */
4074 set_regs_matched_done = 0;
4076 /* If this was the only register active, nothing is active
4078 if (lowest_active_reg == highest_active_reg)
4080 lowest_active_reg = NO_LOWEST_ACTIVE_REG;
4081 highest_active_reg = NO_HIGHEST_ACTIVE_REG;
4084 { /* We must scan for the new highest active register, since
4085 it isn't necessarily one less than now: consider
4086 (a(b)c(d(e)f)g). When group 3 ends, after the f), the
4087 new highest active register is 1. */
4088 unsigned char r = *p - 1;
4089 while (r > 0 && !IS_ACTIVE (reg_info[r]))
4092 /* If we end up at register zero, that means that we saved
4093 the registers as the result of an `on_failure_jump', not
4094 a `start_memory', and we jumped to past the innermost
4095 `stop_memory'. For example, in ((.)*) we save
4096 registers 1 and 2 as a result of the *, but when we pop
4097 back to the second ), we are at the stop_memory 1.
4098 Thus, nothing is active. */
4101 lowest_active_reg = NO_LOWEST_ACTIVE_REG;
4102 highest_active_reg = NO_HIGHEST_ACTIVE_REG;
4105 highest_active_reg = r;
4108 /* If just failed to match something this time around with a
4109 group that's operated on by a repetition operator, try to
4110 force exit from the ``loop'', and restore the register
4111 information for this group that we had before trying this
4113 if ((!MATCHED_SOMETHING (reg_info[*p])
4114 || just_past_start_mem == p - 1)
4117 boolean is_a_jump_n = false;
4121 switch ((re_opcode_t) *p1++)
4125 case pop_failure_jump:
4126 case maybe_pop_jump:
4128 case dummy_failure_jump:
4129 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4139 /* If the next operation is a jump backwards in the pattern
4140 to an on_failure_jump right before the start_memory
4141 corresponding to this stop_memory, exit from the loop
4142 by forcing a failure after pushing on the stack the
4143 on_failure_jump's jump in the pattern, and d. */
4144 if (mcnt < 0 && (re_opcode_t) *p1 == on_failure_jump
4145 && (re_opcode_t) p1[3] == start_memory && p1[4] == *p)
4147 /* If this group ever matched anything, then restore
4148 what its registers were before trying this last
4149 failed match, e.g., with `(a*)*b' against `ab' for
4150 regstart[1], and, e.g., with `((a*)*(b*)*)*'
4151 against `aba' for regend[3].
4153 Also restore the registers for inner groups for,
4154 e.g., `((a*)(b*))*' against `aba' (register 3 would
4155 otherwise get trashed). */
4157 if (EVER_MATCHED_SOMETHING (reg_info[*p]))
4161 EVER_MATCHED_SOMETHING (reg_info[*p]) = 0;
4163 /* Restore this and inner groups' (if any) registers. */
4164 for (r = *p; r < *p + *(p + 1); r++)
4166 regstart[r] = old_regstart[r];
4168 /* xx why this test? */
4169 if (old_regend[r] >= regstart[r])
4170 regend[r] = old_regend[r];
4174 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4175 PUSH_FAILURE_POINT (p1 + mcnt, d, -2);
4181 /* Move past the register number and the inner group count. */
4186 /* \<digit> has been turned into a `duplicate' command which is
4187 followed by the numeric value of <digit> as the register number. */
4190 register const char *d2, *dend2;
4191 int regno = *p++; /* Get which register to match against. */
4192 DEBUG_PRINT2 ("EXECUTING duplicate %d.\n", regno);
4194 /* Can't back reference a group which we've never matched. */
4195 if (REG_UNSET (regstart[regno]) || REG_UNSET (regend[regno]))
4198 /* Where in input to try to start matching. */
4199 d2 = regstart[regno];
4201 /* Where to stop matching; if both the place to start and
4202 the place to stop matching are in the same string, then
4203 set to the place to stop, otherwise, for now have to use
4204 the end of the first string. */
4206 dend2 = ((FIRST_STRING_P (regstart[regno])
4207 == FIRST_STRING_P (regend[regno]))
4208 ? regend[regno] : end_match_1);
4211 /* If necessary, advance to next segment in register
4215 if (dend2 == end_match_2) break;
4216 if (dend2 == regend[regno]) break;
4218 /* End of string1 => advance to string2. */
4220 dend2 = regend[regno];
4222 /* At end of register contents => success */
4223 if (d2 == dend2) break;
4225 /* If necessary, advance to next segment in data. */
4228 /* How many characters left in this segment to match. */
4231 /* Want how many consecutive characters we can match in
4232 one shot, so, if necessary, adjust the count. */
4233 if (mcnt > dend2 - d2)
4236 /* Compare that many; failure if mismatch, else move
4239 ? bcmp_translate (d, d2, mcnt, translate)
4240 : bcmp (d, d2, mcnt))
4242 d += mcnt, d2 += mcnt;
4244 /* Do this because we've match some characters. */
4245 SET_REGS_MATCHED ();
4251 /* begline matches the empty string at the beginning of the string
4252 (unless `not_bol' is set in `bufp'), and, if
4253 `newline_anchor' is set, after newlines. */
4255 DEBUG_PRINT1 ("EXECUTING begline.\n");
4257 if (AT_STRINGS_BEG (d))
4259 if (!bufp->not_bol) break;
4261 else if (d[-1] == '\n' && bufp->newline_anchor)
4265 /* In all other cases, we fail. */
4269 /* endline is the dual of begline. */
4271 DEBUG_PRINT1 ("EXECUTING endline.\n");
4273 if (AT_STRINGS_END (d))
4275 if (!bufp->not_eol) break;
4278 /* We have to ``prefetch'' the next character. */
4279 else if ((d == end1 ? *string2 : *d) == '\n'
4280 && bufp->newline_anchor)
4287 /* Match at the very beginning of the data. */
4289 DEBUG_PRINT1 ("EXECUTING begbuf.\n");
4290 if (AT_STRINGS_BEG (d))
4295 /* Match at the very end of the data. */
4297 DEBUG_PRINT1 ("EXECUTING endbuf.\n");
4298 if (AT_STRINGS_END (d))
4303 /* on_failure_keep_string_jump is used to optimize `.*\n'. It
4304 pushes NULL as the value for the string on the stack. Then
4305 `pop_failure_point' will keep the current value for the
4306 string, instead of restoring it. To see why, consider
4307 matching `foo\nbar' against `.*\n'. The .* matches the foo;
4308 then the . fails against the \n. But the next thing we want
4309 to do is match the \n against the \n; if we restored the
4310 string value, we would be back at the foo.
4312 Because this is used only in specific cases, we don't need to
4313 check all the things that `on_failure_jump' does, to make
4314 sure the right things get saved on the stack. Hence we don't
4315 share its code. The only reason to push anything on the
4316 stack at all is that otherwise we would have to change
4317 `anychar's code to do something besides goto fail in this
4318 case; that seems worse than this. */
4319 case on_failure_keep_string_jump:
4320 DEBUG_PRINT1 ("EXECUTING on_failure_keep_string_jump");
4322 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4323 DEBUG_PRINT3 (" %d (to 0x%x):\n", mcnt, p + mcnt);
4325 PUSH_FAILURE_POINT (p + mcnt, NULL, -2);
4329 /* Uses of on_failure_jump:
4331 Each alternative starts with an on_failure_jump that points
4332 to the beginning of the next alternative. Each alternative
4333 except the last ends with a jump that in effect jumps past
4334 the rest of the alternatives. (They really jump to the
4335 ending jump of the following alternative, because tensioning
4336 these jumps is a hassle.)
4338 Repeats start with an on_failure_jump that points past both
4339 the repetition text and either the following jump or
4340 pop_failure_jump back to this on_failure_jump. */
4341 case on_failure_jump:
4343 DEBUG_PRINT1 ("EXECUTING on_failure_jump");
4345 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4346 DEBUG_PRINT3 (" %d (to 0x%x)", mcnt, p + mcnt);
4348 /* If this on_failure_jump comes right before a group (i.e.,
4349 the original * applied to a group), save the information
4350 for that group and all inner ones, so that if we fail back
4351 to this point, the group's information will be correct.
4352 For example, in \(a*\)*\1, we need the preceding group,
4353 and in \(\(a*\)b*\)\2, we need the inner group. */
4355 /* We can't use `p' to check ahead because we push
4356 a failure point to `p + mcnt' after we do this. */
4359 /* We need to skip no_op's before we look for the
4360 start_memory in case this on_failure_jump is happening as
4361 the result of a completed succeed_n, as in \(a\)\{1,3\}b\1
4363 while (p1 < pend && (re_opcode_t) *p1 == no_op)
4366 if (p1 < pend && (re_opcode_t) *p1 == start_memory)
4368 /* We have a new highest active register now. This will
4369 get reset at the start_memory we are about to get to,
4370 but we will have saved all the registers relevant to
4371 this repetition op, as described above. */
4372 highest_active_reg = *(p1 + 1) + *(p1 + 2);
4373 if (lowest_active_reg == NO_LOWEST_ACTIVE_REG)
4374 lowest_active_reg = *(p1 + 1);
4377 DEBUG_PRINT1 (":\n");
4378 PUSH_FAILURE_POINT (p + mcnt, d, -2);
4382 /* A smart repeat ends with `maybe_pop_jump'.
4383 We change it to either `pop_failure_jump' or `jump'. */
4384 case maybe_pop_jump:
4385 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4386 DEBUG_PRINT2 ("EXECUTING maybe_pop_jump %d.\n", mcnt);
4388 register unsigned char *p2 = p;
4390 /* Compare the beginning of the repeat with what in the
4391 pattern follows its end. If we can establish that there
4392 is nothing that they would both match, i.e., that we
4393 would have to backtrack because of (as in, e.g., `a*a')
4394 then we can change to pop_failure_jump, because we'll
4395 never have to backtrack.
4397 This is not true in the case of alternatives: in
4398 `(a|ab)*' we do need to backtrack to the `ab' alternative
4399 (e.g., if the string was `ab'). But instead of trying to
4400 detect that here, the alternative has put on a dummy
4401 failure point which is what we will end up popping. */
4403 /* Skip over open/close-group commands.
4404 If what follows this loop is a ...+ construct,
4405 look at what begins its body, since we will have to
4406 match at least one of that. */
4410 && ((re_opcode_t) *p2 == stop_memory
4411 || (re_opcode_t) *p2 == start_memory))
4413 else if (p2 + 6 < pend
4414 && (re_opcode_t) *p2 == dummy_failure_jump)
4421 /* p1[0] ... p1[2] are the `on_failure_jump' corresponding
4422 to the `maybe_finalize_jump' of this case. Examine what
4425 /* If we're at the end of the pattern, we can change. */
4428 /* Consider what happens when matching ":\(.*\)"
4429 against ":/". I don't really understand this code
4431 p[-3] = (unsigned char) pop_failure_jump;
4433 (" End of pattern: change to `pop_failure_jump'.\n");
4436 else if ((re_opcode_t) *p2 == exactn
4437 || (bufp->newline_anchor && (re_opcode_t) *p2 == endline))
4439 register unsigned char c
4440 = *p2 == (unsigned char) endline ? '\n' : p2[2];
4442 if ((re_opcode_t) p1[3] == exactn && p1[5] != c)
4444 p[-3] = (unsigned char) pop_failure_jump;
4445 DEBUG_PRINT3 (" %c != %c => pop_failure_jump.\n",
4449 else if ((re_opcode_t) p1[3] == charset
4450 || (re_opcode_t) p1[3] == charset_not)
4452 int not = (re_opcode_t) p1[3] == charset_not;
4454 if (c < (unsigned char) (p1[4] * BYTEWIDTH)
4455 && p1[5 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH)))
4458 /* `not' is equal to 1 if c would match, which means
4459 that we can't change to pop_failure_jump. */
4462 p[-3] = (unsigned char) pop_failure_jump;
4463 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
4467 else if ((re_opcode_t) *p2 == charset)
4470 register unsigned char c
4471 = *p2 == (unsigned char) endline ? '\n' : p2[2];
4474 if ((re_opcode_t) p1[3] == exactn
4475 && ! ((int) p2[1] * BYTEWIDTH > (int) p1[4]
4476 && (p2[1 + p1[4] / BYTEWIDTH]
4477 & (1 << (p1[4] % BYTEWIDTH)))))
4479 p[-3] = (unsigned char) pop_failure_jump;
4480 DEBUG_PRINT3 (" %c != %c => pop_failure_jump.\n",
4484 else if ((re_opcode_t) p1[3] == charset_not)
4487 /* We win if the charset_not inside the loop
4488 lists every character listed in the charset after. */
4489 for (idx = 0; idx < (int) p2[1]; idx++)
4490 if (! (p2[2 + idx] == 0
4491 || (idx < (int) p1[4]
4492 && ((p2[2 + idx] & ~ p1[5 + idx]) == 0))))
4497 p[-3] = (unsigned char) pop_failure_jump;
4498 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
4501 else if ((re_opcode_t) p1[3] == charset)
4504 /* We win if the charset inside the loop
4505 has no overlap with the one after the loop. */
4507 idx < (int) p2[1] && idx < (int) p1[4];
4509 if ((p2[2 + idx] & p1[5 + idx]) != 0)
4512 if (idx == p2[1] || idx == p1[4])
4514 p[-3] = (unsigned char) pop_failure_jump;
4515 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
4520 p -= 2; /* Point at relative address again. */
4521 if ((re_opcode_t) p[-1] != pop_failure_jump)
4523 p[-1] = (unsigned char) jump;
4524 DEBUG_PRINT1 (" Match => jump.\n");
4525 goto unconditional_jump;
4527 /* Note fall through. */
4530 /* The end of a simple repeat has a pop_failure_jump back to
4531 its matching on_failure_jump, where the latter will push a
4532 failure point. The pop_failure_jump takes off failure
4533 points put on by this pop_failure_jump's matching
4534 on_failure_jump; we got through the pattern to here from the
4535 matching on_failure_jump, so didn't fail. */
4536 case pop_failure_jump:
4538 /* We need to pass separate storage for the lowest and
4539 highest registers, even though we don't care about the
4540 actual values. Otherwise, we will restore only one
4541 register from the stack, since lowest will == highest in
4542 `pop_failure_point'. */
4543 unsigned dummy_low_reg, dummy_high_reg;
4544 unsigned char *pdummy;
4547 DEBUG_PRINT1 ("EXECUTING pop_failure_jump.\n");
4548 POP_FAILURE_POINT (sdummy, pdummy,
4549 dummy_low_reg, dummy_high_reg,
4550 reg_dummy, reg_dummy, reg_info_dummy);
4552 /* Note fall through. */
4555 /* Unconditionally jump (without popping any failure points). */
4558 EXTRACT_NUMBER_AND_INCR (mcnt, p); /* Get the amount to jump. */
4559 DEBUG_PRINT2 ("EXECUTING jump %d ", mcnt);
4560 p += mcnt; /* Do the jump. */
4561 DEBUG_PRINT2 ("(to 0x%x).\n", p);
4565 /* We need this opcode so we can detect where alternatives end
4566 in `group_match_null_string_p' et al. */
4568 DEBUG_PRINT1 ("EXECUTING jump_past_alt.\n");
4569 goto unconditional_jump;
4572 /* Normally, the on_failure_jump pushes a failure point, which
4573 then gets popped at pop_failure_jump. We will end up at
4574 pop_failure_jump, also, and with a pattern of, say, `a+', we
4575 are skipping over the on_failure_jump, so we have to push
4576 something meaningless for pop_failure_jump to pop. */
4577 case dummy_failure_jump:
4578 DEBUG_PRINT1 ("EXECUTING dummy_failure_jump.\n");
4579 /* It doesn't matter what we push for the string here. What
4580 the code at `fail' tests is the value for the pattern. */
4581 PUSH_FAILURE_POINT (0, 0, -2);
4582 goto unconditional_jump;
4585 /* At the end of an alternative, we need to push a dummy failure
4586 point in case we are followed by a `pop_failure_jump', because
4587 we don't want the failure point for the alternative to be
4588 popped. For example, matching `(a|ab)*' against `aab'
4589 requires that we match the `ab' alternative. */
4590 case push_dummy_failure:
4591 DEBUG_PRINT1 ("EXECUTING push_dummy_failure.\n");
4592 /* See comments just above at `dummy_failure_jump' about the
4594 PUSH_FAILURE_POINT (0, 0, -2);
4597 /* Have to succeed matching what follows at least n times.
4598 After that, handle like `on_failure_jump'. */
4600 EXTRACT_NUMBER (mcnt, p + 2);
4601 DEBUG_PRINT2 ("EXECUTING succeed_n %d.\n", mcnt);
4604 /* Originally, this is how many times we HAVE to succeed. */
4609 STORE_NUMBER_AND_INCR (p, mcnt);
4610 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p, mcnt);
4614 DEBUG_PRINT2 (" Setting two bytes from 0x%x to no_op.\n", p+2);
4615 p[2] = (unsigned char) no_op;
4616 p[3] = (unsigned char) no_op;
4622 EXTRACT_NUMBER (mcnt, p + 2);
4623 DEBUG_PRINT2 ("EXECUTING jump_n %d.\n", mcnt);
4625 /* Originally, this is how many times we CAN jump. */
4629 STORE_NUMBER (p + 2, mcnt);
4630 goto unconditional_jump;
4632 /* If don't have to jump any more, skip over the rest of command. */
4639 DEBUG_PRINT1 ("EXECUTING set_number_at.\n");
4641 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4643 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4644 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p1, mcnt);
4645 STORE_NUMBER (p1, mcnt);
4650 DEBUG_PRINT1 ("EXECUTING wordbound.\n");
4651 if (AT_WORD_BOUNDARY (d))
4656 DEBUG_PRINT1 ("EXECUTING notwordbound.\n");
4657 if (AT_WORD_BOUNDARY (d))
4662 DEBUG_PRINT1 ("EXECUTING wordbeg.\n");
4663 if (WORDCHAR_P (d) && (AT_STRINGS_BEG (d) || !WORDCHAR_P (d - 1)))
4668 DEBUG_PRINT1 ("EXECUTING wordend.\n");
4669 if (!AT_STRINGS_BEG (d) && WORDCHAR_P (d - 1)
4670 && (!WORDCHAR_P (d) || AT_STRINGS_END (d)))
4676 DEBUG_PRINT1 ("EXECUTING before_dot.\n");
4677 if (PTR_CHAR_POS ((unsigned char *) d) >= point)
4682 DEBUG_PRINT1 ("EXECUTING at_dot.\n");
4683 if (PTR_CHAR_POS ((unsigned char *) d) != point)
4688 DEBUG_PRINT1 ("EXECUTING after_dot.\n");
4689 if (PTR_CHAR_POS ((unsigned char *) d) <= point)
4694 DEBUG_PRINT2 ("EXECUTING syntaxspec %d.\n", mcnt);
4699 DEBUG_PRINT1 ("EXECUTING Emacs wordchar.\n");
4703 /* Can't use *d++ here; SYNTAX may be an unsafe macro. */
4705 if (SYNTAX (d[-1]) != (enum syntaxcode) mcnt)
4707 SET_REGS_MATCHED ();
4711 DEBUG_PRINT2 ("EXECUTING notsyntaxspec %d.\n", mcnt);
4713 goto matchnotsyntax;
4716 DEBUG_PRINT1 ("EXECUTING Emacs notwordchar.\n");
4720 /* Can't use *d++ here; SYNTAX may be an unsafe macro. */
4722 if (SYNTAX (d[-1]) == (enum syntaxcode) mcnt)
4724 SET_REGS_MATCHED ();
4727 #else /* not emacs */
4729 DEBUG_PRINT1 ("EXECUTING non-Emacs wordchar.\n");
4731 if (!WORDCHAR_P (d))
4733 SET_REGS_MATCHED ();
4738 DEBUG_PRINT1 ("EXECUTING non-Emacs notwordchar.\n");
4742 SET_REGS_MATCHED ();
4745 #endif /* not emacs */
4750 continue; /* Successfully executed one pattern command; keep going. */
4753 /* We goto here if a matching operation fails. */
4755 if (!FAIL_STACK_EMPTY ())
4756 { /* A restart point is known. Restore to that state. */
4757 DEBUG_PRINT1 ("\nFAIL:\n");
4758 POP_FAILURE_POINT (d, p,
4759 lowest_active_reg, highest_active_reg,
4760 regstart, regend, reg_info);
4762 /* If this failure point is a dummy, try the next one. */
4766 /* If we failed to the end of the pattern, don't examine *p. */
4770 boolean is_a_jump_n = false;
4772 /* If failed to a backwards jump that's part of a repetition
4773 loop, need to pop this failure point and use the next one. */
4774 switch ((re_opcode_t) *p)
4778 case maybe_pop_jump:
4779 case pop_failure_jump:
4782 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4785 if ((is_a_jump_n && (re_opcode_t) *p1 == succeed_n)
4787 && (re_opcode_t) *p1 == on_failure_jump))
4795 if (d >= string1 && d <= end1)
4799 break; /* Matching at this starting point really fails. */
4803 goto restore_best_regs;
4807 return -1; /* Failure to match. */
4810 /* Subroutine definitions for re_match_2. */
4813 /* We are passed P pointing to a register number after a start_memory.
4815 Return true if the pattern up to the corresponding stop_memory can
4816 match the empty string, and false otherwise.
4818 If we find the matching stop_memory, sets P to point to one past its number.
4819 Otherwise, sets P to an undefined byte less than or equal to END.
4821 We don't handle duplicates properly (yet). */
4824 group_match_null_string_p (p, end, reg_info)
4825 unsigned char **p, *end;
4826 register_info_type *reg_info;
4829 /* Point to after the args to the start_memory. */
4830 unsigned char *p1 = *p + 2;
4834 /* Skip over opcodes that can match nothing, and return true or
4835 false, as appropriate, when we get to one that can't, or to the
4836 matching stop_memory. */
4838 switch ((re_opcode_t) *p1)
4840 /* Could be either a loop or a series of alternatives. */
4841 case on_failure_jump:
4843 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4845 /* If the next operation is not a jump backwards in the
4850 /* Go through the on_failure_jumps of the alternatives,
4851 seeing if any of the alternatives cannot match nothing.
4852 The last alternative starts with only a jump,
4853 whereas the rest start with on_failure_jump and end
4854 with a jump, e.g., here is the pattern for `a|b|c':
4856 /on_failure_jump/0/6/exactn/1/a/jump_past_alt/0/6
4857 /on_failure_jump/0/6/exactn/1/b/jump_past_alt/0/3
4860 So, we have to first go through the first (n-1)
4861 alternatives and then deal with the last one separately. */
4864 /* Deal with the first (n-1) alternatives, which start
4865 with an on_failure_jump (see above) that jumps to right
4866 past a jump_past_alt. */
4868 while ((re_opcode_t) p1[mcnt-3] == jump_past_alt)
4870 /* `mcnt' holds how many bytes long the alternative
4871 is, including the ending `jump_past_alt' and
4874 if (!alt_match_null_string_p (p1, p1 + mcnt - 3,
4878 /* Move to right after this alternative, including the
4882 /* Break if it's the beginning of an n-th alternative
4883 that doesn't begin with an on_failure_jump. */
4884 if ((re_opcode_t) *p1 != on_failure_jump)
4887 /* Still have to check that it's not an n-th
4888 alternative that starts with an on_failure_jump. */
4890 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4891 if ((re_opcode_t) p1[mcnt-3] != jump_past_alt)
4893 /* Get to the beginning of the n-th alternative. */
4899 /* Deal with the last alternative: go back and get number
4900 of the `jump_past_alt' just before it. `mcnt' contains
4901 the length of the alternative. */
4902 EXTRACT_NUMBER (mcnt, p1 - 2);
4904 if (!alt_match_null_string_p (p1, p1 + mcnt, reg_info))
4907 p1 += mcnt; /* Get past the n-th alternative. */
4913 assert (p1[1] == **p);
4919 if (!common_op_match_null_string_p (&p1, end, reg_info))
4922 } /* while p1 < end */
4925 } /* group_match_null_string_p */
4928 /* Similar to group_match_null_string_p, but doesn't deal with alternatives:
4929 It expects P to be the first byte of a single alternative and END one
4930 byte past the last. The alternative can contain groups. */
4933 alt_match_null_string_p (p, end, reg_info)
4934 unsigned char *p, *end;
4935 register_info_type *reg_info;
4938 unsigned char *p1 = p;
4942 /* Skip over opcodes that can match nothing, and break when we get
4943 to one that can't. */
4945 switch ((re_opcode_t) *p1)
4948 case on_failure_jump:
4950 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4955 if (!common_op_match_null_string_p (&p1, end, reg_info))
4958 } /* while p1 < end */
4961 } /* alt_match_null_string_p */
4964 /* Deals with the ops common to group_match_null_string_p and
4965 alt_match_null_string_p.
4967 Sets P to one after the op and its arguments, if any. */
4970 common_op_match_null_string_p (p, end, reg_info)
4971 unsigned char **p, *end;
4972 register_info_type *reg_info;
4977 unsigned char *p1 = *p;
4979 switch ((re_opcode_t) *p1++)
4999 assert (reg_no > 0 && reg_no <= MAX_REGNUM);
5000 ret = group_match_null_string_p (&p1, end, reg_info);
5002 /* Have to set this here in case we're checking a group which
5003 contains a group and a back reference to it. */
5005 if (REG_MATCH_NULL_STRING_P (reg_info[reg_no]) == MATCH_NULL_UNSET_VALUE)
5006 REG_MATCH_NULL_STRING_P (reg_info[reg_no]) = ret;
5012 /* If this is an optimized succeed_n for zero times, make the jump. */
5014 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
5022 /* Get to the number of times to succeed. */
5024 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
5029 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
5037 if (!REG_MATCH_NULL_STRING_P (reg_info[*p1]))
5045 /* All other opcodes mean we cannot match the empty string. */
5051 } /* common_op_match_null_string_p */
5054 /* Return zero if TRANSLATE[S1] and TRANSLATE[S2] are identical for LEN
5055 bytes; nonzero otherwise. */
5058 bcmp_translate (s1, s2, len, translate)
5059 unsigned char *s1, *s2;
5063 register unsigned char *p1 = s1, *p2 = s2;
5066 if (translate[*p1++] != translate[*p2++]) return 1;
5072 /* Entry points for GNU code. */
5074 /* re_compile_pattern is the GNU regular expression compiler: it
5075 compiles PATTERN (of length SIZE) and puts the result in BUFP.
5076 Returns 0 if the pattern was valid, otherwise an error string.
5078 Assumes the `allocated' (and perhaps `buffer') and `translate' fields
5079 are set in BUFP on entry.
5081 We call regex_compile to do the actual compilation. */
5084 re_compile_pattern (pattern, length, bufp)
5085 const char *pattern;
5087 struct re_pattern_buffer *bufp;
5091 /* GNU code is written to assume at least RE_NREGS registers will be set
5092 (and at least one extra will be -1). */
5093 bufp->regs_allocated = REGS_UNALLOCATED;
5095 /* And GNU code determines whether or not to get register information
5096 by passing null for the REGS argument to re_match, etc., not by
5100 /* Match anchors at newline. */
5101 bufp->newline_anchor = 1;
5103 ret = regex_compile (pattern, length, re_syntax_options, bufp);
5107 return gettext (re_error_msgid[(int) ret]);
5110 /* Entry points compatible with 4.2 BSD regex library. We don't define
5111 them unless specifically requested. */
5113 #ifdef _REGEX_RE_COMP
5115 /* BSD has one and only one pattern buffer. */
5116 static struct re_pattern_buffer re_comp_buf;
5126 if (!re_comp_buf.buffer)
5127 return gettext ("No previous regular expression");
5131 if (!re_comp_buf.buffer)
5133 re_comp_buf.buffer = (unsigned char *) malloc (200);
5134 if (re_comp_buf.buffer == NULL)
5135 return gettext (re_error_msgid[(int) REG_ESPACE]);
5136 re_comp_buf.allocated = 200;
5138 re_comp_buf.fastmap = (char *) malloc (1 << BYTEWIDTH);
5139 if (re_comp_buf.fastmap == NULL)
5140 return gettext (re_error_msgid[(int) REG_ESPACE]);
5143 /* Since `re_exec' always passes NULL for the `regs' argument, we
5144 don't need to initialize the pattern buffer fields which affect it. */
5146 /* Match anchors at newlines. */
5147 re_comp_buf.newline_anchor = 1;
5149 ret = regex_compile (s, strlen (s), re_syntax_options, &re_comp_buf);
5154 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
5155 return (char *) gettext (re_error_msgid[(int) ret]);
5163 const int len = strlen (s);
5165 0 <= re_search (&re_comp_buf, s, len, 0, len, (struct re_registers *) 0);
5167 #endif /* _REGEX_RE_COMP */
5169 /* POSIX.2 functions. Don't define these for Emacs. */
5173 /* regcomp takes a regular expression as a string and compiles it.
5175 PREG is a regex_t *. We do not expect any fields to be initialized,
5176 since POSIX says we shouldn't. Thus, we set
5178 `buffer' to the compiled pattern;
5179 `used' to the length of the compiled pattern;
5180 `syntax' to RE_SYNTAX_POSIX_EXTENDED if the
5181 REG_EXTENDED bit in CFLAGS is set; otherwise, to
5182 RE_SYNTAX_POSIX_BASIC;
5183 `newline_anchor' to REG_NEWLINE being set in CFLAGS;
5184 `fastmap' and `fastmap_accurate' to zero;
5185 `re_nsub' to the number of subexpressions in PATTERN.
5187 PATTERN is the address of the pattern string.
5189 CFLAGS is a series of bits which affect compilation.
5191 If REG_EXTENDED is set, we use POSIX extended syntax; otherwise, we
5192 use POSIX basic syntax.
5194 If REG_NEWLINE is set, then . and [^...] don't match newline.
5195 Also, regexec will try a match beginning after every newline.
5197 If REG_ICASE is set, then we considers upper- and lowercase
5198 versions of letters to be equivalent when matching.
5200 If REG_NOSUB is set, then when PREG is passed to regexec, that
5201 routine will report only success or failure, and nothing about the
5204 It returns 0 if it succeeds, nonzero if it doesn't. (See regex.h for
5205 the return codes and their meanings.) */
5208 regcomp (preg, pattern, cflags)
5210 const char *pattern;
5215 = (cflags & REG_EXTENDED) ?
5216 RE_SYNTAX_POSIX_EXTENDED : RE_SYNTAX_POSIX_BASIC;
5218 /* regex_compile will allocate the space for the compiled pattern. */
5220 preg->allocated = 0;
5223 /* Don't bother to use a fastmap when searching. This simplifies the
5224 REG_NEWLINE case: if we used a fastmap, we'd have to put all the
5225 characters after newlines into the fastmap. This way, we just try
5229 if (cflags & REG_ICASE)
5233 preg->translate = (char *) malloc (CHAR_SET_SIZE);
5234 if (preg->translate == NULL)
5235 return (int) REG_ESPACE;
5237 /* Map uppercase characters to corresponding lowercase ones. */
5238 for (i = 0; i < CHAR_SET_SIZE; i++)
5239 preg->translate[i] = ISUPPER (i) ? tolower (i) : i;
5242 preg->translate = NULL;
5244 /* If REG_NEWLINE is set, newlines are treated differently. */
5245 if (cflags & REG_NEWLINE)
5246 { /* REG_NEWLINE implies neither . nor [^...] match newline. */
5247 syntax &= ~RE_DOT_NEWLINE;
5248 syntax |= RE_HAT_LISTS_NOT_NEWLINE;
5249 /* It also changes the matching behavior. */
5250 preg->newline_anchor = 1;
5253 preg->newline_anchor = 0;
5255 preg->no_sub = !!(cflags & REG_NOSUB);
5257 /* POSIX says a null character in the pattern terminates it, so we
5258 can use strlen here in compiling the pattern. */
5259 ret = regex_compile (pattern, strlen (pattern), syntax, preg);
5261 /* POSIX doesn't distinguish between an unmatched open-group and an
5262 unmatched close-group: both are REG_EPAREN. */
5263 if (ret == REG_ERPAREN) ret = REG_EPAREN;
5269 /* regexec searches for a given pattern, specified by PREG, in the
5272 If NMATCH is zero or REG_NOSUB was set in the cflags argument to
5273 `regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at
5274 least NMATCH elements, and we set them to the offsets of the
5275 corresponding matched substrings.
5277 EFLAGS specifies `execution flags' which affect matching: if
5278 REG_NOTBOL is set, then ^ does not match at the beginning of the
5279 string; if REG_NOTEOL is set, then $ does not match at the end.
5281 We return 0 if we find a match and REG_NOMATCH if not. */
5284 regexec (preg, string, nmatch, pmatch, eflags)
5285 const regex_t *preg;
5288 regmatch_t pmatch[];
5292 struct re_registers regs;
5293 regex_t private_preg;
5294 int len = strlen (string);
5295 boolean want_reg_info = !preg->no_sub && nmatch > 0;
5297 private_preg = *preg;
5299 private_preg.not_bol = !!(eflags & REG_NOTBOL);
5300 private_preg.not_eol = !!(eflags & REG_NOTEOL);
5302 /* The user has told us exactly how many registers to return
5303 information about, via `nmatch'. We have to pass that on to the
5304 matching routines. */
5305 private_preg.regs_allocated = REGS_FIXED;
5309 regs.num_regs = nmatch;
5310 regs.start = TALLOC (nmatch, regoff_t);
5311 regs.end = TALLOC (nmatch, regoff_t);
5312 if (regs.start == NULL || regs.end == NULL)
5313 return (int) REG_NOMATCH;
5316 /* Perform the searching operation. */
5317 ret = re_search (&private_preg, string, len,
5318 /* start: */ 0, /* range: */ len,
5319 want_reg_info ? ®s : (struct re_registers *) 0);
5321 /* Copy the register information to the POSIX structure. */
5328 for (r = 0; r < nmatch; r++)
5330 pmatch[r].rm_so = regs.start[r];
5331 pmatch[r].rm_eo = regs.end[r];
5335 /* If we needed the temporary register info, free the space now. */
5340 /* We want zero return to mean success, unlike `re_search'. */
5341 return ret >= 0 ? (int) REG_NOERROR : (int) REG_NOMATCH;
5345 /* Returns a message corresponding to an error code, ERRCODE, returned
5346 from either regcomp or regexec. We don't use PREG here. */
5349 regerror (errcode, preg, errbuf, errbuf_size)
5351 const regex_t *preg;
5359 || errcode >= (sizeof (re_error_msgid) / sizeof (re_error_msgid[0])))
5360 /* Only error codes returned by the rest of the code should be passed
5361 to this routine. If we are given anything else, or if other regex
5362 code generates an invalid error code, then the program has a bug.
5363 Dump core so we can fix it. */
5366 msg = gettext (re_error_msgid[errcode]);
5368 msg_size = strlen (msg) + 1; /* Includes the null. */
5370 if (errbuf_size != 0)
5372 if (msg_size > errbuf_size)
5374 strncpy (errbuf, msg, errbuf_size - 1);
5375 errbuf[errbuf_size - 1] = 0;
5378 strcpy (errbuf, msg);
5385 /* Free dynamically allocated space used by PREG. */
5391 if (preg->buffer != NULL)
5392 free (preg->buffer);
5393 preg->buffer = NULL;
5395 preg->allocated = 0;
5398 if (preg->fastmap != NULL)
5399 free (preg->fastmap);
5400 preg->fastmap = NULL;
5401 preg->fastmap_accurate = 0;
5403 if (preg->translate != NULL)
5404 free (preg->translate);
5405 preg->translate = NULL;
5408 #endif /* not emacs */
5412 make-backup-files: t
5414 trim-versions-without-asking: nil