1 /* Extended regular expression matching and search library,
3 * (Implements POSIX draft P1003.2/D11.2, except for some of the
4 * internationalization features.)
6 * Copyright (C) 1993, 1994, 1995, 1996, 1997 Free Software Foundation, Inc.
8 * This file is part of the GNU C Library. Its master source is NOT part of
9 * the C library, however. The master source lives in /gd/gnu/lib.
11 * The GNU C Library is free software; you can redistribute it and/or
12 * modify it under the terms of the GNU Library General Public License as
13 * published by the Free Software Foundation; either version 2 of the
14 * License, or (at your option) any later version.
16 * The GNU C Library is distributed in the hope that it will be useful,
17 * but WITHOUT ANY WARRANTY; without even the implied warranty of
18 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
19 * Library General Public License for more details.
21 * You should have received a copy of the GNU Library General Public
22 * License along with the GNU C Library; see the file COPYING.LIB. If not,
23 * write to the Free Software Foundation, Inc., 59 Temple Place - Suite 330,
24 * Boston, MA 02111-1307, USA.
30 * Use _regex.h instead of regex.h. tlr, 1999-01-06
31 * Make REGEX_MALLOC depend on HAVE_ALLOCA &c.
33 * Don't switch on regex debugging when debugging mutt.
37 /* The following doesn't mix too well with autoconfiguring
38 * the use of alloca. So let's disable it for AIX.
43 /* AIX requires this to be the first thing in the file. */
44 # if defined (_AIX) && !defined (REGEX_MALLOC)
59 #if (defined(HAVE_ALLOCA_H) && !defined(_AIX))
63 #if (!defined(HAVE_ALLOCA) || defined(_AIX))
67 #if defined(STDC_HEADERS) && !defined(emacs)
70 /* We need this for `regex.h', and perhaps for the Emacs include files. */
71 #include <sys/types.h>
74 /* For platform which support the ISO C amendement 1 functionality we
75 support user defined character classes. */
76 #if defined _LIBC || (defined HAVE_WCTYPE_H && defined HAVE_WCHAR_H)
81 /* This is for other GNU distributions with internationalized messages. */
82 #if HAVE_LIBINTL_H || defined (_LIBC)
85 # define gettext(msgid) (msgid)
89 /* This define is so xgettext can find the internationalizable
91 #define gettext_noop(String) String
94 /* The `emacs' switch turns on certain matching commands
95 that make sense only in Emacs. */
102 #else /* not emacs */
104 /* If we are not linking with Emacs proper,
105 we can't use the relocating allocator
106 even if config.h says that we can. */
109 #if defined (STDC_HEADERS) || defined (_LIBC)
112 char *malloc (); /* __MEM_CHECKED__ */
113 char *realloc (); /* __MEM_CHECKED__ */
116 /* When used in Emacs's lib-src, we need to get bzero and bcopy somehow.
117 If nothing else has been done, use the method below. */
118 #ifdef INHIBIT_STRING_HEADER
119 #if !(defined (HAVE_BZERO) && defined (HAVE_BCOPY))
120 #if !defined (bzero) && !defined (bcopy)
121 #undef INHIBIT_STRING_HEADER
126 /* This is the normal way of making sure we have a bcopy and a bzero.
127 This is used in most programs--a few other programs avoid this
128 by defining INHIBIT_STRING_HEADER. */
129 #ifndef INHIBIT_STRING_HEADER
130 #if defined (HAVE_STRING_H) || defined (STDC_HEADERS) || defined (_LIBC)
133 #define bcmp(s1, s2, n) memcmp ((s1), (s2), (n))
136 #define bcopy(s, d, n) memcpy ((d), (s), (n))
139 #define bzero(s, n) memset ((s), 0, (n))
146 /* Define the syntax stuff for \<, \>, etc. */
148 /* This must be nonzero for the wordchar and notwordchar pattern
149 commands in re_match_2. */
154 #ifdef SWITCH_ENUM_BUG
155 #define SWITCH_ENUM_CAST(x) ((int)(x))
157 #define SWITCH_ENUM_CAST(x) (x)
162 extern char *re_syntax_table;
164 #else /* not SYNTAX_TABLE */
166 /* How many characters in the character set. */
167 #define CHAR_SET_SIZE 256
169 static char re_syntax_table[CHAR_SET_SIZE];
180 bzero (re_syntax_table, sizeof re_syntax_table);
182 for (c = 'a'; c <= 'z'; c++)
183 re_syntax_table[c] = Sword;
185 for (c = 'A'; c <= 'Z'; c++)
186 re_syntax_table[c] = Sword;
188 for (c = '0'; c <= '9'; c++)
189 re_syntax_table[c] = Sword;
191 re_syntax_table['_'] = Sword;
196 #endif /* not SYNTAX_TABLE */
198 #define SYNTAX(c) re_syntax_table[c]
200 #endif /* not emacs */
202 /* Get the interface, including the syntax bits. */
204 /* Changed to fit into mutt - tlr, 1999-01-06 */
208 /* isalpha etc. are used for the character classes. */
211 /* Jim Meyering writes:
213 "... Some ctype macros are valid only for character codes that
214 isascii says are ASCII (SGI's IRIX-4.0.5 is one such system --when
215 using /bin/cc or gcc but without giving an ansi option). So, all
216 ctype uses should be through macros like ISPRINT... If
217 STDC_HEADERS is defined, then autoconf has verified that the ctype
218 macros don't need to be guarded with references to isascii. ...
219 Defining isascii to 1 should let any compiler worth its salt
220 eliminate the && through constant folding." */
222 #if defined (STDC_HEADERS) || (!defined (isascii) && !defined (HAVE_ISASCII))
225 #define ISASCII(c) isascii(c)
229 #define ISBLANK(c) (ISASCII (c) && isblank (c))
231 #define ISBLANK(c) ((c) == ' ' || (c) == '\t')
234 #define ISGRAPH(c) (ISASCII (c) && isgraph (c))
236 #define ISGRAPH(c) (ISASCII (c) && isprint (c) && !isspace (c))
239 #define ISPRINT(c) (ISASCII (c) && isprint (c))
240 #define ISDIGIT(c) (ISASCII (c) && isdigit (c))
241 #define ISALNUM(c) (ISASCII (c) && isalnum (c))
242 #define ISALPHA(c) (ISASCII (c) && isalpha (c))
243 #define ISCNTRL(c) (ISASCII (c) && iscntrl (c))
244 #define ISLOWER(c) (ISASCII (c) && islower (c))
245 #define ISPUNCT(c) (ISASCII (c) && ispunct (c))
246 #define ISSPACE(c) (ISASCII (c) && isspace (c))
247 #define ISUPPER(c) (ISASCII (c) && isupper (c))
248 #define ISXDIGIT(c) (ISASCII (c) && isxdigit (c))
251 #define NULL (void *)0
254 /* We remove any previous definition of `SIGN_EXTEND_CHAR',
255 since ours (we hope) works properly with all combinations of
256 machines, compilers, `char' and `unsigned char' argument types.
257 (Per Bothner suggested the basic approach.) */
258 #undef SIGN_EXTEND_CHAR
260 #define SIGN_EXTEND_CHAR(c) ((signed char) (c))
261 #else /* not __STDC__ */
262 /* As in Harbison and Steele. */
263 #define SIGN_EXTEND_CHAR(c) ((((unsigned char) (c)) ^ 128) - 128)
266 /* Should we use malloc or alloca? If REGEX_MALLOC is not defined, we
267 use `alloca' instead of `malloc'. This is because using malloc in
268 re_search* or re_match* could cause memory leaks when C-g is used in
269 Emacs; also, malloc is slower and causes storage fragmentation. On
270 the other hand, malloc is more portable, and easier to debug.
272 Because we sometimes use alloca, some routines have to be macros,
273 not functions -- `alloca'-allocated space disappears at the end of the
274 function it is called in. */
278 #define REGEX_ALLOCATE malloc
279 #define REGEX_REALLOCATE(source, osize, nsize) realloc (source, nsize)
280 #define REGEX_FREE free
282 #else /* not REGEX_MALLOC */
284 /* Emacs already defines alloca, sometimes. */
287 /* Make alloca work the best possible way. */
289 #define alloca __builtin_alloca
290 #else /* not __GNUC__ */
293 #else /* not __GNUC__ or HAVE_ALLOCA_H */
294 #if 0 /* It is a bad idea to declare alloca. We always cast the result. */
295 #ifndef _AIX /* Already did AIX, up at the top. */
297 #endif /* not _AIX */
299 #endif /* not HAVE_ALLOCA_H */
300 #endif /* not __GNUC__ */
302 #endif /* not alloca */
304 #define REGEX_ALLOCATE alloca
306 /* Assumes a `char *destination' variable. */
307 #define REGEX_REALLOCATE(source, osize, nsize) \
308 (destination = (char *) alloca (nsize), \
309 bcopy (source, destination, osize), \
312 /* No need to do anything to free, after alloca. */
313 #define REGEX_FREE(arg) ((void)0) /* Do nothing! But inhibit gcc warning. */
315 #endif /* not REGEX_MALLOC */
317 /* Define how to allocate the failure stack. */
319 #if defined (REL_ALLOC) && defined (REGEX_MALLOC)
321 #define REGEX_ALLOCATE_STACK(size) \
322 r_alloc (&failure_stack_ptr, (size))
323 #define REGEX_REALLOCATE_STACK(source, osize, nsize) \
324 r_re_alloc (&failure_stack_ptr, (nsize))
325 #define REGEX_FREE_STACK(ptr) \
326 r_alloc_free (&failure_stack_ptr)
328 #else /* not using relocating allocator */
332 #define REGEX_ALLOCATE_STACK malloc
333 #define REGEX_REALLOCATE_STACK(source, osize, nsize) realloc (source, nsize)
334 #define REGEX_FREE_STACK free
336 #else /* not REGEX_MALLOC */
338 #define REGEX_ALLOCATE_STACK alloca
340 #define REGEX_REALLOCATE_STACK(source, osize, nsize) \
341 REGEX_REALLOCATE (source, osize, nsize)
342 /* No need to explicitly free anything. */
343 #define REGEX_FREE_STACK(arg)
345 #endif /* not REGEX_MALLOC */
346 #endif /* not using relocating allocator */
349 /* True if `size1' is non-NULL and PTR is pointing anywhere inside
350 `string1' or just past its end. This works if PTR is NULL, which is
352 #define FIRST_STRING_P(ptr) \
353 (size1 && string1 <= (ptr) && (ptr) <= string1 + size1)
355 /* (Re)Allocate N items of type T using malloc, or fail. */
356 #define TALLOC(n, t) ((t *) malloc ((n) * sizeof (t)))
357 #define RETALLOC(addr, n, t) ((addr) = (t *) realloc (addr, (n) * sizeof (t)))
358 #define RETALLOC_IF(addr, n, t) \
359 if (addr) RETALLOC((addr), (n), t); else (addr) = TALLOC ((n), t)
360 #define REGEX_TALLOC(n, t) ((t *) REGEX_ALLOCATE ((n) * sizeof (t)))
362 #define BYTEWIDTH 8 /* In bits. */
364 #define STREQ(s1, s2) ((strcmp (s1, s2) == 0))
368 #define MAX(a, b) ((a) > (b) ? (a) : (b))
369 #define MIN(a, b) ((a) < (b) ? (a) : (b))
371 typedef char boolean;
375 static int re_match_2_internal ();
377 /* These are the command codes that appear in compiled regular
378 expressions. Some opcodes are followed by argument bytes. A
379 command code can specify any interpretation whatsoever for its
380 arguments. Zero bytes may appear in the compiled regular expression. */
386 /* Succeed right away--no more backtracking. */
389 /* Followed by one byte giving n, then by n literal bytes. */
392 /* Matches any (more or less) character. */
395 /* Matches any one char belonging to specified set. First
396 following byte is number of bitmap bytes. Then come bytes
397 for a bitmap saying which chars are in. Bits in each byte
398 are ordered low-bit-first. A character is in the set if its
399 bit is 1. A character too large to have a bit in the map is
400 automatically not in the set. */
403 /* Same parameters as charset, but match any character that is
404 not one of those specified. */
407 /* Start remembering the text that is matched, for storing in a
408 register. Followed by one byte with the register number, in
409 the range 0 to one less than the pattern buffer's re_nsub
410 field. Then followed by one byte with the number of groups
411 inner to this one. (This last has to be part of the
412 start_memory only because we need it in the on_failure_jump
416 /* Stop remembering the text that is matched and store it in a
417 memory register. Followed by one byte with the register
418 number, in the range 0 to one less than `re_nsub' in the
419 pattern buffer, and one byte with the number of inner groups,
420 just like `start_memory'. (We need the number of inner
421 groups here because we don't have any easy way of finding the
422 corresponding start_memory when we're at a stop_memory.) */
425 /* Match a duplicate of something remembered. Followed by one
426 byte containing the register number. */
429 /* Fail unless at beginning of line. */
432 /* Fail unless at end of line. */
435 /* Succeeds if at beginning of buffer (if emacs) or at beginning
436 of string to be matched (if not). */
439 /* Analogously, for end of buffer/string. */
442 /* Followed by two byte relative address to which to jump. */
445 /* Same as jump, but marks the end of an alternative. */
448 /* Followed by two-byte relative address of place to resume at
449 in case of failure. */
452 /* Like on_failure_jump, but pushes a placeholder instead of the
453 current string position when executed. */
454 on_failure_keep_string_jump,
456 /* Throw away latest failure point and then jump to following
457 two-byte relative address. */
460 /* Change to pop_failure_jump if know won't have to backtrack to
461 match; otherwise change to jump. This is used to jump
462 back to the beginning of a repeat. If what follows this jump
463 clearly won't match what the repeat does, such that we can be
464 sure that there is no use backtracking out of repetitions
465 already matched, then we change it to a pop_failure_jump.
466 Followed by two-byte address. */
469 /* Jump to following two-byte address, and push a dummy failure
470 point. This failure point will be thrown away if an attempt
471 is made to use it for a failure. A `+' construct makes this
472 before the first repeat. Also used as an intermediary kind
473 of jump when compiling an alternative. */
476 /* Push a dummy failure point and continue. Used at the end of
480 /* Followed by two-byte relative address and two-byte number n.
481 After matching N times, jump to the address upon failure. */
484 /* Followed by two-byte relative address, and two-byte number n.
485 Jump to the address N times, then fail. */
488 /* Set the following two-byte relative address to the
489 subsequent two-byte number. The address *includes* the two
493 wordchar, /* Matches any word-constituent character. */
494 notwordchar, /* Matches any char that is not a word-constituent. */
496 wordbeg, /* Succeeds if at word beginning. */
497 wordend, /* Succeeds if at word end. */
499 wordbound, /* Succeeds if at a word boundary. */
500 notwordbound /* Succeeds if not at a word boundary. */
503 ,before_dot, /* Succeeds if before point. */
504 at_dot, /* Succeeds if at point. */
505 after_dot, /* Succeeds if after point. */
507 /* Matches any character whose syntax is specified. Followed by
508 a byte which contains a syntax code, e.g., Sword. */
511 /* Matches any character whose syntax is not that specified. */
516 /* Common operations on the compiled pattern. */
518 /* Store NUMBER in two contiguous bytes starting at DESTINATION. */
520 #define STORE_NUMBER(destination, number) \
522 (destination)[0] = (number) & 0377; \
523 (destination)[1] = (number) >> 8; \
526 /* Same as STORE_NUMBER, except increment DESTINATION to
527 the byte after where the number is stored. Therefore, DESTINATION
528 must be an lvalue. */
530 #define STORE_NUMBER_AND_INCR(destination, number) \
532 STORE_NUMBER (destination, number); \
533 (destination) += 2; \
536 /* Put into DESTINATION a number stored in two contiguous bytes starting
539 #define EXTRACT_NUMBER(destination, source) \
541 (destination) = *(source) & 0377; \
542 (destination) += SIGN_EXTEND_CHAR (*((source) + 1)) << 8; \
546 static void extract_number _RE_ARGS ((int *dest, unsigned char *source));
548 extract_number (dest, source)
550 unsigned char *source;
552 int temp = SIGN_EXTEND_CHAR (*(source + 1));
553 *dest = *source & 0377;
557 #ifndef EXTRACT_MACROS /* To debug the macros. */
558 #undef EXTRACT_NUMBER
559 #define EXTRACT_NUMBER(dest, src) extract_number (&dest, src)
560 #endif /* not EXTRACT_MACROS */
564 /* Same as EXTRACT_NUMBER, except increment SOURCE to after the number.
565 SOURCE must be an lvalue. */
567 #define EXTRACT_NUMBER_AND_INCR(destination, source) \
569 EXTRACT_NUMBER (destination, source); \
574 static void extract_number_and_incr _RE_ARGS ((int *destination,
575 unsigned char **source));
577 extract_number_and_incr (destination, source)
579 unsigned char **source;
581 extract_number (destination, *source);
585 #ifndef EXTRACT_MACROS
586 #undef EXTRACT_NUMBER_AND_INCR
587 #define EXTRACT_NUMBER_AND_INCR(dest, src) \
588 extract_number_and_incr (&dest, &src)
589 #endif /* not EXTRACT_MACROS */
593 /* If DEBUG is defined, Regex prints many voluminous messages about what
594 it is doing (if the variable `debug' is nonzero). If linked with the
595 main program in `iregex.c', you can enter patterns and strings
596 interactively. And if linked with the main program in `main.c' and
597 the other test files, you can run the already-written tests. */
601 /* We use standard I/O for debugging. */
604 /* It is useful to test things that ``must'' be true when debugging. */
607 static int debug = 0;
609 #define DEBUG_STATEMENT(e) e
610 #define DEBUG_PRINT1(x) if (debug) printf (x)
611 #define DEBUG_PRINT2(x1, x2) if (debug) printf (x1, x2)
612 #define DEBUG_PRINT3(x1, x2, x3) if (debug) printf (x1, x2, x3)
613 #define DEBUG_PRINT4(x1, x2, x3, x4) if (debug) printf (x1, x2, x3, x4)
614 #define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) \
615 if (debug) print_partial_compiled_pattern (s, e)
616 #define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) \
617 if (debug) print_double_string (w, s1, sz1, s2, sz2)
620 /* Print the fastmap in human-readable form. */
623 print_fastmap (fastmap)
626 unsigned was_a_range = 0;
629 while (i < (1 << BYTEWIDTH))
635 while (i < (1 << BYTEWIDTH) && fastmap[i])
651 /* Print a compiled pattern string in human-readable form, starting at
652 the START pointer into it and ending just before the pointer END. */
655 print_partial_compiled_pattern (start, end)
656 unsigned char *start;
661 unsigned char *p = start;
662 unsigned char *pend = end;
670 /* Loop over pattern commands. */
673 printf ("%d:\t", p - start);
675 switch ((re_opcode_t) *p++)
683 printf ("/exactn/%d", mcnt);
694 printf ("/start_memory/%d/%d", mcnt, *p++);
699 printf ("/stop_memory/%d/%d", mcnt, *p++);
703 printf ("/duplicate/%d", *p++);
713 register int c, last = -100;
714 register int in_range = 0;
716 printf ("/charset [%s",
717 (re_opcode_t) *(p - 1) == charset_not ? "^" : "");
719 assert (p + *p < pend);
721 for (c = 0; c < 256; c++)
723 && (p[1 + (c/8)] & (1 << (c % 8))))
725 /* Are we starting a range? */
726 if (last + 1 == c && ! in_range)
731 /* Have we broken a range? */
732 else if (last + 1 != c && in_range)
761 case on_failure_jump:
762 extract_number_and_incr (&mcnt, &p);
763 printf ("/on_failure_jump to %d", p + mcnt - start);
766 case on_failure_keep_string_jump:
767 extract_number_and_incr (&mcnt, &p);
768 printf ("/on_failure_keep_string_jump to %d", p + mcnt - start);
771 case dummy_failure_jump:
772 extract_number_and_incr (&mcnt, &p);
773 printf ("/dummy_failure_jump to %d", p + mcnt - start);
776 case push_dummy_failure:
777 printf ("/push_dummy_failure");
781 extract_number_and_incr (&mcnt, &p);
782 printf ("/maybe_pop_jump to %d", p + mcnt - start);
785 case pop_failure_jump:
786 extract_number_and_incr (&mcnt, &p);
787 printf ("/pop_failure_jump to %d", p + mcnt - start);
791 extract_number_and_incr (&mcnt, &p);
792 printf ("/jump_past_alt to %d", p + mcnt - start);
796 extract_number_and_incr (&mcnt, &p);
797 printf ("/jump to %d", p + mcnt - start);
801 extract_number_and_incr (&mcnt, &p);
803 extract_number_and_incr (&mcnt2, &p);
804 printf ("/succeed_n to %d, %d times", p1 - start, mcnt2);
808 extract_number_and_incr (&mcnt, &p);
810 extract_number_and_incr (&mcnt2, &p);
811 printf ("/jump_n to %d, %d times", p1 - start, mcnt2);
815 extract_number_and_incr (&mcnt, &p);
817 extract_number_and_incr (&mcnt2, &p);
818 printf ("/set_number_at location %d to %d", p1 - start, mcnt2);
822 printf ("/wordbound");
826 printf ("/notwordbound");
838 printf ("/before_dot");
846 printf ("/after_dot");
850 printf ("/syntaxspec");
852 printf ("/%d", mcnt);
856 printf ("/notsyntaxspec");
858 printf ("/%d", mcnt);
863 printf ("/wordchar");
867 printf ("/notwordchar");
879 printf ("?%d", *(p-1));
885 printf ("%d:\tend of pattern.\n", p - start);
890 print_compiled_pattern (bufp)
891 struct re_pattern_buffer *bufp;
893 unsigned char *buffer = bufp->buffer;
895 print_partial_compiled_pattern (buffer, buffer + bufp->used);
896 printf ("%ld bytes used/%ld bytes allocated.\n",
897 bufp->used, bufp->allocated);
899 if (bufp->fastmap_accurate && bufp->fastmap)
901 printf ("fastmap: ");
902 print_fastmap (bufp->fastmap);
905 printf ("re_nsub: %d\t", bufp->re_nsub);
906 printf ("regs_alloc: %d\t", bufp->regs_allocated);
907 printf ("can_be_null: %d\t", bufp->can_be_null);
908 printf ("newline_anchor: %d\n", bufp->newline_anchor);
909 printf ("no_sub: %d\t", bufp->no_sub);
910 printf ("not_bol: %d\t", bufp->not_bol);
911 printf ("not_eol: %d\t", bufp->not_eol);
912 printf ("syntax: %lx\n", bufp->syntax);
913 /* Perhaps we should print the translate table? */
918 print_double_string (where, string1, size1, string2, size2)
931 if (FIRST_STRING_P (where))
933 for (this_char = where - string1; this_char < size1; this_char++)
934 putchar (string1[this_char]);
939 for (this_char = where - string2; this_char < size2; this_char++)
940 putchar (string2[this_char]);
951 #else /* not DEBUG */
956 #define DEBUG_STATEMENT(e)
957 #define DEBUG_PRINT1(x)
958 #define DEBUG_PRINT2(x1, x2)
959 #define DEBUG_PRINT3(x1, x2, x3)
960 #define DEBUG_PRINT4(x1, x2, x3, x4)
961 #define DEBUG_PRINT_COMPILED_PATTERN(p, s, e)
962 #define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2)
964 #endif /* not DEBUG */
966 /* Set by `re_set_syntax' to the current regexp syntax to recognize. Can
967 also be assigned to arbitrarily: each pattern buffer stores its own
968 syntax, so it can be changed between regex compilations. */
969 /* This has no initializer because initialized variables in Emacs
970 become read-only after dumping. */
971 reg_syntax_t re_syntax_options;
974 /* Specify the precise syntax of regexps for compilation. This provides
975 for compatibility for various utilities which historically have
976 different, incompatible syntaxes.
978 The argument SYNTAX is a bit mask comprised of the various bits
979 defined in regex.h. We return the old syntax. */
982 re_set_syntax (syntax)
985 reg_syntax_t ret = re_syntax_options;
987 re_syntax_options = syntax;
989 if (syntax & RE_DEBUG)
991 else if (debug) /* was on but now is not */
997 /* This table gives an error message for each of the error codes listed
998 in regex.h. Obviously the order here has to be same as there.
999 POSIX doesn't require that we do anything for REG_NOERROR,
1000 but why not be nice? */
1002 static const char *re_error_msgid[] =
1004 gettext_noop ("Success"), /* REG_NOERROR */
1005 gettext_noop ("No match"), /* REG_NOMATCH */
1006 gettext_noop ("Invalid regular expression"), /* REG_BADPAT */
1007 gettext_noop ("Invalid collation character"), /* REG_ECOLLATE */
1008 gettext_noop ("Invalid character class name"), /* REG_ECTYPE */
1009 gettext_noop ("Trailing backslash"), /* REG_EESCAPE */
1010 gettext_noop ("Invalid back reference"), /* REG_ESUBREG */
1011 gettext_noop ("Unmatched [ or [^"), /* REG_EBRACK */
1012 gettext_noop ("Unmatched ( or \\("), /* REG_EPAREN */
1013 gettext_noop ("Unmatched \\{"), /* REG_EBRACE */
1014 gettext_noop ("Invalid content of \\{\\}"), /* REG_BADBR */
1015 gettext_noop ("Invalid range end"), /* REG_ERANGE */
1016 gettext_noop ("Memory exhausted"), /* REG_ESPACE */
1017 gettext_noop ("Invalid preceding regular expression"), /* REG_BADRPT */
1018 gettext_noop ("Premature end of regular expression"), /* REG_EEND */
1019 gettext_noop ("Regular expression too big"), /* REG_ESIZE */
1020 gettext_noop ("Unmatched ) or \\)"), /* REG_ERPAREN */
1023 /* Avoiding alloca during matching, to placate r_alloc. */
1025 /* Define MATCH_MAY_ALLOCATE unless we need to make sure that the
1026 searching and matching functions should not call alloca. On some
1027 systems, alloca is implemented in terms of malloc, and if we're
1028 using the relocating allocator routines, then malloc could cause a
1029 relocation, which might (if the strings being searched are in the
1030 ralloc heap) shift the data out from underneath the regexp
1033 Here's another reason to avoid allocation: Emacs
1034 processes input from X in a signal handler; processing X input may
1035 call malloc; if input arrives while a matching routine is calling
1036 malloc, then we're scrod. But Emacs can't just block input while
1037 calling matching routines; then we don't notice interrupts when
1038 they come in. So, Emacs blocks input around all regexp calls
1039 except the matching calls, which it leaves unprotected, in the
1040 faith that they will not malloc. */
1042 /* Normally, this is fine. */
1043 #define MATCH_MAY_ALLOCATE
1045 /* When using GNU C, we are not REALLY using the C alloca, no matter
1046 what config.h may say. So don't take precautions for it. */
1051 /* The match routines may not allocate if (1) they would do it with malloc
1052 and (2) it's not safe for them to use malloc.
1053 Note that if REL_ALLOC is defined, matching would not use malloc for the
1054 failure stack, but we would still use it for the register vectors;
1055 so REL_ALLOC should not affect this. */
1056 #if (defined (C_ALLOCA) || defined (REGEX_MALLOC)) && defined (emacs)
1057 #undef MATCH_MAY_ALLOCATE
1061 /* Failure stack declarations and macros; both re_compile_fastmap and
1062 re_match_2 use a failure stack. These have to be macros because of
1063 REGEX_ALLOCATE_STACK. */
1066 /* Number of failure points for which to initially allocate space
1067 when matching. If this number is exceeded, we allocate more
1068 space, so it is not a hard limit. */
1069 #ifndef INIT_FAILURE_ALLOC
1070 #define INIT_FAILURE_ALLOC 5
1073 /* Roughly the maximum number of failure points on the stack. Would be
1074 exactly that if always used MAX_FAILURE_ITEMS items each time we failed.
1075 This is a variable only so users of regex can assign to it; we never
1076 change it ourselves. */
1080 #if defined (MATCH_MAY_ALLOCATE)
1081 /* 4400 was enough to cause a crash on Alpha OSF/1,
1082 whose default stack limit is 2mb. */
1083 long int re_max_failures = 4000;
1085 long int re_max_failures = 2000;
1088 union fail_stack_elt
1090 unsigned char *pointer;
1094 typedef union fail_stack_elt fail_stack_elt_t;
1098 fail_stack_elt_t *stack;
1099 unsigned long int size;
1100 unsigned long int avail; /* Offset of next open position. */
1103 #else /* not INT_IS_16BIT */
1105 #if defined (MATCH_MAY_ALLOCATE)
1106 /* 4400 was enough to cause a crash on Alpha OSF/1,
1107 whose default stack limit is 2mb. */
1108 int re_max_failures = 20000;
1110 int re_max_failures = 2000;
1113 union fail_stack_elt
1115 unsigned char *pointer;
1119 typedef union fail_stack_elt fail_stack_elt_t;
1123 fail_stack_elt_t *stack;
1125 unsigned avail; /* Offset of next open position. */
1128 #endif /* INT_IS_16BIT */
1130 #define FAIL_STACK_EMPTY() (fail_stack.avail == 0)
1131 #define FAIL_STACK_PTR_EMPTY() (fail_stack_ptr->avail == 0)
1132 #define FAIL_STACK_FULL() (fail_stack.avail == fail_stack.size)
1135 /* Define macros to initialize and free the failure stack.
1136 Do `return -2' if the alloc fails. */
1138 #ifdef MATCH_MAY_ALLOCATE
1139 #define INIT_FAIL_STACK() \
1141 fail_stack.stack = (fail_stack_elt_t *) \
1142 REGEX_ALLOCATE_STACK (INIT_FAILURE_ALLOC * sizeof (fail_stack_elt_t)); \
1144 if (fail_stack.stack == NULL) \
1147 fail_stack.size = INIT_FAILURE_ALLOC; \
1148 fail_stack.avail = 0; \
1151 #define RESET_FAIL_STACK() REGEX_FREE_STACK (fail_stack.stack)
1153 #define INIT_FAIL_STACK() \
1155 fail_stack.avail = 0; \
1158 #define RESET_FAIL_STACK()
1162 /* Double the size of FAIL_STACK, up to approximately `re_max_failures' items.
1164 Return 1 if succeeds, and 0 if either ran out of memory
1165 allocating space for it or it was already too large.
1167 REGEX_REALLOCATE_STACK requires `destination' be declared. */
1169 #define DOUBLE_FAIL_STACK(fail_stack) \
1170 ((fail_stack).size > (unsigned) (re_max_failures * MAX_FAILURE_ITEMS) \
1172 : ((fail_stack).stack = (fail_stack_elt_t *) \
1173 REGEX_REALLOCATE_STACK ((fail_stack).stack, \
1174 (fail_stack).size * sizeof (fail_stack_elt_t), \
1175 ((fail_stack).size << 1) * sizeof (fail_stack_elt_t)), \
1177 (fail_stack).stack == NULL \
1179 : ((fail_stack).size <<= 1, \
1183 /* Push pointer POINTER on FAIL_STACK.
1184 Return 1 if was able to do so and 0 if ran out of memory allocating
1186 #define PUSH_PATTERN_OP(POINTER, FAIL_STACK) \
1187 ((FAIL_STACK_FULL () \
1188 && !DOUBLE_FAIL_STACK (FAIL_STACK)) \
1190 : ((FAIL_STACK).stack[(FAIL_STACK).avail++].pointer = POINTER, \
1193 /* Push a pointer value onto the failure stack.
1194 Assumes the variable `fail_stack'. Probably should only
1195 be called from within `PUSH_FAILURE_POINT'. */
1196 #define PUSH_FAILURE_POINTER(item) \
1197 fail_stack.stack[fail_stack.avail++].pointer = (unsigned char *) (item)
1199 /* This pushes an integer-valued item onto the failure stack.
1200 Assumes the variable `fail_stack'. Probably should only
1201 be called from within `PUSH_FAILURE_POINT'. */
1202 #define PUSH_FAILURE_INT(item) \
1203 fail_stack.stack[fail_stack.avail++].integer = (item)
1205 /* Push a fail_stack_elt_t value onto the failure stack.
1206 Assumes the variable `fail_stack'. Probably should only
1207 be called from within `PUSH_FAILURE_POINT'. */
1208 #define PUSH_FAILURE_ELT(item) \
1209 fail_stack.stack[fail_stack.avail++] = (item)
1211 /* These three POP... operations complement the three PUSH... operations.
1212 All assume that `fail_stack' is nonempty. */
1213 #define POP_FAILURE_POINTER() fail_stack.stack[--fail_stack.avail].pointer
1214 #define POP_FAILURE_INT() fail_stack.stack[--fail_stack.avail].integer
1215 #define POP_FAILURE_ELT() fail_stack.stack[--fail_stack.avail]
1217 /* Used to omit pushing failure point id's when we're not debugging. */
1219 #define DEBUG_PUSH PUSH_FAILURE_INT
1220 #define DEBUG_POP(item_addr) (item_addr)->integer = POP_FAILURE_INT ()
1222 #define DEBUG_PUSH(item)
1223 #define DEBUG_POP(item_addr)
1227 /* Push the information about the state we will need
1228 if we ever fail back to it.
1230 Requires variables fail_stack, regstart, regend, reg_info, and
1231 num_regs be declared. DOUBLE_FAIL_STACK requires `destination' be
1234 Does `return FAILURE_CODE' if runs out of memory. */
1236 #define PUSH_FAILURE_POINT(pattern_place, string_place, failure_code) \
1238 char *destination; \
1239 /* Must be int, so when we don't save any registers, the arithmetic \
1240 of 0 + -1 isn't done as unsigned. */ \
1241 /* Can't be int, since there is not a shred of a guarantee that int \
1242 is wide enough to hold a value of something to which pointer can \
1246 DEBUG_STATEMENT (failure_id++); \
1247 DEBUG_STATEMENT (nfailure_points_pushed++); \
1248 DEBUG_PRINT2 ("\nPUSH_FAILURE_POINT #%u:\n", failure_id); \
1249 DEBUG_PRINT2 (" Before push, next avail: %d\n", (fail_stack).avail);\
1250 DEBUG_PRINT2 (" size: %d\n", (fail_stack).size);\
1252 DEBUG_PRINT2 (" slots needed: %d\n", NUM_FAILURE_ITEMS); \
1253 DEBUG_PRINT2 (" available: %d\n", REMAINING_AVAIL_SLOTS); \
1255 /* Ensure we have enough space allocated for what we will push. */ \
1256 while (REMAINING_AVAIL_SLOTS < NUM_FAILURE_ITEMS) \
1258 if (!DOUBLE_FAIL_STACK (fail_stack)) \
1259 return failure_code; \
1261 DEBUG_PRINT2 ("\n Doubled stack; size now: %d\n", \
1262 (fail_stack).size); \
1263 DEBUG_PRINT2 (" slots available: %d\n", REMAINING_AVAIL_SLOTS);\
1266 /* Push the info, starting with the registers. */ \
1267 DEBUG_PRINT1 ("\n"); \
1270 for (this_reg = lowest_active_reg; this_reg <= highest_active_reg; \
1273 DEBUG_PRINT2 (" Pushing reg: %d\n", this_reg); \
1274 DEBUG_STATEMENT (num_regs_pushed++); \
1276 DEBUG_PRINT2 (" start: 0x%x\n", regstart[this_reg]); \
1277 PUSH_FAILURE_POINTER (regstart[this_reg]); \
1279 DEBUG_PRINT2 (" end: 0x%x\n", regend[this_reg]); \
1280 PUSH_FAILURE_POINTER (regend[this_reg]); \
1282 DEBUG_PRINT2 (" info: 0x%x\n ", reg_info[this_reg]); \
1283 DEBUG_PRINT2 (" match_null=%d", \
1284 REG_MATCH_NULL_STRING_P (reg_info[this_reg])); \
1285 DEBUG_PRINT2 (" active=%d", IS_ACTIVE (reg_info[this_reg])); \
1286 DEBUG_PRINT2 (" matched_something=%d", \
1287 MATCHED_SOMETHING (reg_info[this_reg])); \
1288 DEBUG_PRINT2 (" ever_matched=%d", \
1289 EVER_MATCHED_SOMETHING (reg_info[this_reg])); \
1290 DEBUG_PRINT1 ("\n"); \
1291 PUSH_FAILURE_ELT (reg_info[this_reg].word); \
1294 DEBUG_PRINT2 (" Pushing low active reg: %d\n", lowest_active_reg);\
1295 PUSH_FAILURE_INT (lowest_active_reg); \
1297 DEBUG_PRINT2 (" Pushing high active reg: %d\n", highest_active_reg);\
1298 PUSH_FAILURE_INT (highest_active_reg); \
1300 DEBUG_PRINT2 (" Pushing pattern 0x%x:\n", pattern_place); \
1301 DEBUG_PRINT_COMPILED_PATTERN (bufp, pattern_place, pend); \
1302 PUSH_FAILURE_POINTER (pattern_place); \
1304 DEBUG_PRINT2 (" Pushing string 0x%x: `", string_place); \
1305 DEBUG_PRINT_DOUBLE_STRING (string_place, string1, size1, string2, \
1307 DEBUG_PRINT1 ("'\n"); \
1308 PUSH_FAILURE_POINTER (string_place); \
1310 DEBUG_PRINT2 (" Pushing failure id: %u\n", failure_id); \
1311 DEBUG_PUSH (failure_id); \
1314 /* This is the number of items that are pushed and popped on the stack
1315 for each register. */
1316 #define NUM_REG_ITEMS 3
1318 /* Individual items aside from the registers. */
1320 #define NUM_NONREG_ITEMS 5 /* Includes failure point id. */
1322 #define NUM_NONREG_ITEMS 4
1325 /* We push at most this many items on the stack. */
1326 /* We used to use (num_regs - 1), which is the number of registers
1327 this regexp will save; but that was changed to 5
1328 to avoid stack overflow for a regexp with lots of parens. */
1329 #define MAX_FAILURE_ITEMS (5 * NUM_REG_ITEMS + NUM_NONREG_ITEMS)
1331 /* We actually push this many items. */
1332 #define NUM_FAILURE_ITEMS \
1334 ? 0 : highest_active_reg - lowest_active_reg + 1) \
1338 /* How many items can still be added to the stack without overflowing it. */
1339 #define REMAINING_AVAIL_SLOTS ((fail_stack).size - (fail_stack).avail)
1342 /* Pops what PUSH_FAIL_STACK pushes.
1344 We restore into the parameters, all of which should be lvalues:
1345 STR -- the saved data position.
1346 PAT -- the saved pattern position.
1347 LOW_REG, HIGH_REG -- the highest and lowest active registers.
1348 REGSTART, REGEND -- arrays of string positions.
1349 REG_INFO -- array of information about each subexpression.
1351 Also assumes the variables `fail_stack' and (if debugging), `bufp',
1352 `pend', `string1', `size1', `string2', and `size2'. */
1354 #define POP_FAILURE_POINT(str, pat, low_reg, high_reg, regstart, regend, reg_info)\
1356 DEBUG_STATEMENT (fail_stack_elt_t failure_id;) \
1358 const unsigned char *string_temp; \
1360 assert (!FAIL_STACK_EMPTY ()); \
1362 /* Remove failure points and point to how many regs pushed. */ \
1363 DEBUG_PRINT1 ("POP_FAILURE_POINT:\n"); \
1364 DEBUG_PRINT2 (" Before pop, next avail: %d\n", fail_stack.avail); \
1365 DEBUG_PRINT2 (" size: %d\n", fail_stack.size); \
1367 assert (fail_stack.avail >= NUM_NONREG_ITEMS); \
1369 DEBUG_POP (&failure_id); \
1370 DEBUG_PRINT2 (" Popping failure id: %u\n", failure_id); \
1372 /* If the saved string location is NULL, it came from an \
1373 on_failure_keep_string_jump opcode, and we want to throw away the \
1374 saved NULL, thus retaining our current position in the string. */ \
1375 string_temp = POP_FAILURE_POINTER (); \
1376 if (string_temp != NULL) \
1377 str = (const char *) string_temp; \
1379 DEBUG_PRINT2 (" Popping string 0x%x: `", str); \
1380 DEBUG_PRINT_DOUBLE_STRING (str, string1, size1, string2, size2); \
1381 DEBUG_PRINT1 ("'\n"); \
1383 pat = (unsigned char *) POP_FAILURE_POINTER (); \
1384 DEBUG_PRINT2 (" Popping pattern 0x%x:\n", pat); \
1385 DEBUG_PRINT_COMPILED_PATTERN (bufp, pat, pend); \
1387 /* Restore register info. */ \
1388 high_reg = (active_reg_t) POP_FAILURE_INT (); \
1389 DEBUG_PRINT2 (" Popping high active reg: %d\n", high_reg); \
1391 low_reg = (active_reg_t) POP_FAILURE_INT (); \
1392 DEBUG_PRINT2 (" Popping low active reg: %d\n", low_reg); \
1395 for (this_reg = high_reg; this_reg >= low_reg; this_reg--) \
1397 DEBUG_PRINT2 (" Popping reg: %d\n", this_reg); \
1399 reg_info[this_reg].word = POP_FAILURE_ELT (); \
1400 DEBUG_PRINT2 (" info: 0x%x\n", reg_info[this_reg]); \
1402 regend[this_reg] = (const char *) POP_FAILURE_POINTER (); \
1403 DEBUG_PRINT2 (" end: 0x%x\n", regend[this_reg]); \
1405 regstart[this_reg] = (const char *) POP_FAILURE_POINTER (); \
1406 DEBUG_PRINT2 (" start: 0x%x\n", regstart[this_reg]); \
1410 for (this_reg = highest_active_reg; this_reg > high_reg; this_reg--) \
1412 reg_info[this_reg].word.integer = 0; \
1413 regend[this_reg] = 0; \
1414 regstart[this_reg] = 0; \
1416 highest_active_reg = high_reg; \
1419 set_regs_matched_done = 0; \
1420 DEBUG_STATEMENT (nfailure_points_popped++); \
1421 } /* POP_FAILURE_POINT */
1425 /* Structure for per-register (a.k.a. per-group) information.
1426 Other register information, such as the
1427 starting and ending positions (which are addresses), and the list of
1428 inner groups (which is a bits list) are maintained in separate
1431 We are making a (strictly speaking) nonportable assumption here: that
1432 the compiler will pack our bit fields into something that fits into
1433 the type of `word', i.e., is something that fits into one item on the
1437 /* Declarations and macros for re_match_2. */
1441 fail_stack_elt_t word;
1444 /* This field is one if this group can match the empty string,
1445 zero if not. If not yet determined, `MATCH_NULL_UNSET_VALUE'. */
1446 #define MATCH_NULL_UNSET_VALUE 3
1447 unsigned match_null_string_p : 2;
1448 unsigned is_active : 1;
1449 unsigned matched_something : 1;
1450 unsigned ever_matched_something : 1;
1452 } register_info_type;
1454 #define REG_MATCH_NULL_STRING_P(R) ((R).bits.match_null_string_p)
1455 #define IS_ACTIVE(R) ((R).bits.is_active)
1456 #define MATCHED_SOMETHING(R) ((R).bits.matched_something)
1457 #define EVER_MATCHED_SOMETHING(R) ((R).bits.ever_matched_something)
1460 /* Call this when have matched a real character; it sets `matched' flags
1461 for the subexpressions which we are currently inside. Also records
1462 that those subexprs have matched. */
1463 #define SET_REGS_MATCHED() \
1466 if (!set_regs_matched_done) \
1469 set_regs_matched_done = 1; \
1470 for (r = lowest_active_reg; r <= highest_active_reg; r++) \
1472 MATCHED_SOMETHING (reg_info[r]) \
1473 = EVER_MATCHED_SOMETHING (reg_info[r]) \
1480 /* Registers are set to a sentinel when they haven't yet matched. */
1481 static char reg_unset_dummy;
1482 #define REG_UNSET_VALUE (®_unset_dummy)
1483 #define REG_UNSET(e) ((e) == REG_UNSET_VALUE)
1485 /* Subroutine declarations and macros for regex_compile. */
1487 static reg_errcode_t regex_compile _RE_ARGS ((const char *pattern, size_t size,
1488 reg_syntax_t syntax,
1489 struct re_pattern_buffer *bufp));
1490 static void store_op1 _RE_ARGS ((re_opcode_t op, unsigned char *loc, int arg));
1491 static void store_op2 _RE_ARGS ((re_opcode_t op, unsigned char *loc,
1492 int arg1, int arg2));
1493 static void insert_op1 _RE_ARGS ((re_opcode_t op, unsigned char *loc,
1494 int arg, unsigned char *end));
1495 static void insert_op2 _RE_ARGS ((re_opcode_t op, unsigned char *loc,
1496 int arg1, int arg2, unsigned char *end));
1497 static boolean at_begline_loc_p _RE_ARGS ((const char *pattern, const char *p,
1498 reg_syntax_t syntax));
1499 static boolean at_endline_loc_p _RE_ARGS ((const char *p, const char *pend,
1500 reg_syntax_t syntax));
1501 static reg_errcode_t compile_range _RE_ARGS ((const char **p_ptr,
1504 reg_syntax_t syntax,
1507 /* Fetch the next character in the uncompiled pattern---translating it
1508 if necessary. Also cast from a signed character in the constant
1509 string passed to us by the user to an unsigned char that we can use
1510 as an array index (in, e.g., `translate'). */
1512 #define PATFETCH(c) \
1513 do {if (p == pend) return REG_EEND; \
1514 c = (unsigned char) *p++; \
1515 if (translate) c = (unsigned char) translate[c]; \
1519 /* Fetch the next character in the uncompiled pattern, with no
1521 #define PATFETCH_RAW(c) \
1522 do {if (p == pend) return REG_EEND; \
1523 c = (unsigned char) *p++; \
1526 /* Go backwards one character in the pattern. */
1527 #define PATUNFETCH p--
1530 /* If `translate' is non-null, return translate[D], else just D. We
1531 cast the subscript to translate because some data is declared as
1532 `char *', to avoid warnings when a string constant is passed. But
1533 when we use a character as a subscript we must make it unsigned. */
1535 #define TRANSLATE(d) \
1536 (translate ? (char) translate[(unsigned char) (d)] : (d))
1540 /* Macros for outputting the compiled pattern into `buffer'. */
1542 /* If the buffer isn't allocated when it comes in, use this. */
1543 #define INIT_BUF_SIZE 32
1545 /* Make sure we have at least N more bytes of space in buffer. */
1546 #define GET_BUFFER_SPACE(n) \
1547 while ((unsigned long) (b - bufp->buffer + (n)) > bufp->allocated) \
1550 /* Make sure we have one more byte of buffer space and then add C to it. */
1551 #define BUF_PUSH(c) \
1553 GET_BUFFER_SPACE (1); \
1554 *b++ = (unsigned char) (c); \
1558 /* Ensure we have two more bytes of buffer space and then append C1 and C2. */
1559 #define BUF_PUSH_2(c1, c2) \
1561 GET_BUFFER_SPACE (2); \
1562 *b++ = (unsigned char) (c1); \
1563 *b++ = (unsigned char) (c2); \
1567 /* As with BUF_PUSH_2, except for three bytes. */
1568 #define BUF_PUSH_3(c1, c2, c3) \
1570 GET_BUFFER_SPACE (3); \
1571 *b++ = (unsigned char) (c1); \
1572 *b++ = (unsigned char) (c2); \
1573 *b++ = (unsigned char) (c3); \
1577 /* Store a jump with opcode OP at LOC to location TO. We store a
1578 relative address offset by the three bytes the jump itself occupies. */
1579 #define STORE_JUMP(op, loc, to) \
1580 store_op1 (op, loc, (int) ((to) - (loc) - 3))
1582 /* Likewise, for a two-argument jump. */
1583 #define STORE_JUMP2(op, loc, to, arg) \
1584 store_op2 (op, loc, (int) ((to) - (loc) - 3), arg)
1586 /* Like `STORE_JUMP', but for inserting. Assume `b' is the buffer end. */
1587 #define INSERT_JUMP(op, loc, to) \
1588 insert_op1 (op, loc, (int) ((to) - (loc) - 3), b)
1590 /* Like `STORE_JUMP2', but for inserting. Assume `b' is the buffer end. */
1591 #define INSERT_JUMP2(op, loc, to, arg) \
1592 insert_op2 (op, loc, (int) ((to) - (loc) - 3), arg, b)
1595 /* This is not an arbitrary limit: the arguments which represent offsets
1596 into the pattern are two bytes long. So if 2^16 bytes turns out to
1597 be too small, many things would have to change. */
1598 /* Any other compiler which, like MSC, has allocation limit below 2^16
1599 bytes will have to use approach similar to what was done below for
1600 MSC and drop MAX_BUF_SIZE a bit. Otherwise you may end up
1601 reallocating to 0 bytes. Such thing is not going to work too well.
1602 You have been warned!! */
1603 #if defined(_MSC_VER) && !defined(WIN32)
1604 /* Microsoft C 16-bit versions limit malloc to approx 65512 bytes.
1605 The REALLOC define eliminates a flurry of conversion warnings,
1606 but is not required. */
1607 #define MAX_BUF_SIZE 65500L
1608 #define REALLOC(p,s) realloc ((p), (size_t) (s))
1610 #define MAX_BUF_SIZE (1L << 16)
1611 #define REALLOC(p,s) realloc ((p), (s))
1614 /* Extend the buffer by twice its current size via realloc and
1615 reset the pointers that pointed into the old block to point to the
1616 correct places in the new one. If extending the buffer results in it
1617 being larger than MAX_BUF_SIZE, then flag memory exhausted. */
1618 #define EXTEND_BUFFER() \
1620 unsigned char *old_buffer = bufp->buffer; \
1621 if (bufp->allocated == MAX_BUF_SIZE) \
1623 bufp->allocated <<= 1; \
1624 if (bufp->allocated > MAX_BUF_SIZE) \
1625 bufp->allocated = MAX_BUF_SIZE; \
1626 bufp->buffer = (unsigned char *) REALLOC (bufp->buffer, bufp->allocated);\
1627 if (bufp->buffer == NULL) \
1628 return REG_ESPACE; \
1629 /* If the buffer moved, move all the pointers into it. */ \
1630 if (old_buffer != bufp->buffer) \
1632 b = (b - old_buffer) + bufp->buffer; \
1633 begalt = (begalt - old_buffer) + bufp->buffer; \
1634 if (fixup_alt_jump) \
1635 fixup_alt_jump = (fixup_alt_jump - old_buffer) + bufp->buffer;\
1637 laststart = (laststart - old_buffer) + bufp->buffer; \
1638 if (pending_exact) \
1639 pending_exact = (pending_exact - old_buffer) + bufp->buffer; \
1644 /* Since we have one byte reserved for the register number argument to
1645 {start,stop}_memory, the maximum number of groups we can report
1646 things about is what fits in that byte. */
1647 #define MAX_REGNUM 255
1649 /* But patterns can have more than `MAX_REGNUM' registers. We just
1650 ignore the excess. */
1651 typedef unsigned regnum_t;
1654 /* Macros for the compile stack. */
1656 /* Since offsets can go either forwards or backwards, this type needs to
1657 be able to hold values from -(MAX_BUF_SIZE - 1) to MAX_BUF_SIZE - 1. */
1658 /* int may be not enough when sizeof(int) == 2. */
1659 typedef long pattern_offset_t;
1663 pattern_offset_t begalt_offset;
1664 pattern_offset_t fixup_alt_jump;
1665 pattern_offset_t inner_group_offset;
1666 pattern_offset_t laststart_offset;
1668 } compile_stack_elt_t;
1673 compile_stack_elt_t *stack;
1675 unsigned avail; /* Offset of next open position. */
1676 } compile_stack_type;
1679 #define INIT_COMPILE_STACK_SIZE 32
1681 #define COMPILE_STACK_EMPTY (compile_stack.avail == 0)
1682 #define COMPILE_STACK_FULL (compile_stack.avail == compile_stack.size)
1684 /* The next available element. */
1685 #define COMPILE_STACK_TOP (compile_stack.stack[compile_stack.avail])
1688 /* Set the bit for character C in a list. */
1689 #define SET_LIST_BIT(c) \
1690 (b[((unsigned char) (c)) / BYTEWIDTH] \
1691 |= 1 << (((unsigned char) c) % BYTEWIDTH))
1694 /* Get the next unsigned number in the uncompiled pattern. */
1695 #define GET_UNSIGNED_NUMBER(num) \
1699 while (ISDIGIT (c)) \
1703 num = num * 10 + c - '0'; \
1711 #if defined _LIBC || (defined HAVE_WCTYPE_H && defined HAVE_WCHAR_H)
1712 /* The GNU C library provides support for user-defined character classes
1713 and the functions from ISO C amendement 1. */
1714 # ifdef CHARCLASS_NAME_MAX
1715 # define CHAR_CLASS_MAX_LENGTH CHARCLASS_NAME_MAX
1717 /* This shouldn't happen but some implementation might still have this
1718 problem. Use a reasonable default value. */
1719 # define CHAR_CLASS_MAX_LENGTH 256
1722 # define IS_CHAR_CLASS(string) wctype (string)
1724 # define CHAR_CLASS_MAX_LENGTH 6 /* Namely, `xdigit'. */
1726 # define IS_CHAR_CLASS(string) \
1727 (STREQ (string, "alpha") || STREQ (string, "upper") \
1728 || STREQ (string, "lower") || STREQ (string, "digit") \
1729 || STREQ (string, "alnum") || STREQ (string, "xdigit") \
1730 || STREQ (string, "space") || STREQ (string, "print") \
1731 || STREQ (string, "punct") || STREQ (string, "graph") \
1732 || STREQ (string, "cntrl") || STREQ (string, "blank"))
1735 #ifndef MATCH_MAY_ALLOCATE
1737 /* If we cannot allocate large objects within re_match_2_internal,
1738 we make the fail stack and register vectors global.
1739 The fail stack, we grow to the maximum size when a regexp
1741 The register vectors, we adjust in size each time we
1742 compile a regexp, according to the number of registers it needs. */
1744 static fail_stack_type fail_stack;
1746 /* Size with which the following vectors are currently allocated.
1747 That is so we can make them bigger as needed,
1748 but never make them smaller. */
1749 static int regs_allocated_size;
1751 static const char ** regstart, ** regend;
1752 static const char ** old_regstart, ** old_regend;
1753 static const char **best_regstart, **best_regend;
1754 static register_info_type *reg_info;
1755 static const char **reg_dummy;
1756 static register_info_type *reg_info_dummy;
1758 /* Make the register vectors big enough for NUM_REGS registers,
1759 but don't make them smaller. */
1762 regex_grow_registers (num_regs)
1765 if (num_regs > regs_allocated_size)
1767 RETALLOC_IF (regstart, num_regs, const char *);
1768 RETALLOC_IF (regend, num_regs, const char *);
1769 RETALLOC_IF (old_regstart, num_regs, const char *);
1770 RETALLOC_IF (old_regend, num_regs, const char *);
1771 RETALLOC_IF (best_regstart, num_regs, const char *);
1772 RETALLOC_IF (best_regend, num_regs, const char *);
1773 RETALLOC_IF (reg_info, num_regs, register_info_type);
1774 RETALLOC_IF (reg_dummy, num_regs, const char *);
1775 RETALLOC_IF (reg_info_dummy, num_regs, register_info_type);
1777 regs_allocated_size = num_regs;
1781 #endif /* not MATCH_MAY_ALLOCATE */
1783 static boolean group_in_compile_stack _RE_ARGS ((compile_stack_type
1787 /* `regex_compile' compiles PATTERN (of length SIZE) according to SYNTAX.
1788 Returns one of error codes defined in `regex.h', or zero for success.
1790 Assumes the `allocated' (and perhaps `buffer') and `translate'
1791 fields are set in BUFP on entry.
1793 If it succeeds, results are put in BUFP (if it returns an error, the
1794 contents of BUFP are undefined):
1795 `buffer' is the compiled pattern;
1796 `syntax' is set to SYNTAX;
1797 `used' is set to the length of the compiled pattern;
1798 `fastmap_accurate' is zero;
1799 `re_nsub' is the number of subexpressions in PATTERN;
1800 `not_bol' and `not_eol' are zero;
1802 The `fastmap' and `newline_anchor' fields are neither
1803 examined nor set. */
1805 /* Return, freeing storage we allocated. */
1806 #define FREE_STACK_RETURN(value) \
1807 return (free (compile_stack.stack), value) /* __MEM_CHECKED__ */
1809 static reg_errcode_t
1810 regex_compile (pattern, size, syntax, bufp)
1811 const char *pattern;
1813 reg_syntax_t syntax;
1814 struct re_pattern_buffer *bufp;
1816 /* We fetch characters from PATTERN here. Even though PATTERN is
1817 `char *' (i.e., signed), we declare these variables as unsigned, so
1818 they can be reliably used as array indices. */
1819 register unsigned char c, c1;
1821 /* A random temporary spot in PATTERN. */
1824 /* Points to the end of the buffer, where we should append. */
1825 register unsigned char *b;
1827 /* Keeps track of unclosed groups. */
1828 compile_stack_type compile_stack;
1830 /* Points to the current (ending) position in the pattern. */
1831 const char *p = pattern;
1832 const char *pend = pattern + size;
1834 /* How to translate the characters in the pattern. */
1835 RE_TRANSLATE_TYPE translate = bufp->translate;
1837 /* Address of the count-byte of the most recently inserted `exactn'
1838 command. This makes it possible to tell if a new exact-match
1839 character can be added to that command or if the character requires
1840 a new `exactn' command. */
1841 unsigned char *pending_exact = 0;
1843 /* Address of start of the most recently finished expression.
1844 This tells, e.g., postfix * where to find the start of its
1845 operand. Reset at the beginning of groups and alternatives. */
1846 unsigned char *laststart = 0;
1848 /* Address of beginning of regexp, or inside of last group. */
1849 unsigned char *begalt;
1851 /* Place in the uncompiled pattern (i.e., the {) to
1852 which to go back if the interval is invalid. */
1853 const char *beg_interval;
1855 /* Address of the place where a forward jump should go to the end of
1856 the containing expression. Each alternative of an `or' -- except the
1857 last -- ends with a forward jump of this sort. */
1858 unsigned char *fixup_alt_jump = 0;
1860 /* Counts open-groups as they are encountered. Remembered for the
1861 matching close-group on the compile stack, so the same register
1862 number is put in the stop_memory as the start_memory. */
1863 regnum_t regnum = 0;
1866 DEBUG_PRINT1 ("\nCompiling pattern: ");
1869 unsigned debug_count;
1871 for (debug_count = 0; debug_count < size; debug_count++)
1872 putchar (pattern[debug_count]);
1877 /* Initialize the compile stack. */
1878 compile_stack.stack = TALLOC (INIT_COMPILE_STACK_SIZE, compile_stack_elt_t);
1879 if (compile_stack.stack == NULL)
1882 compile_stack.size = INIT_COMPILE_STACK_SIZE;
1883 compile_stack.avail = 0;
1885 /* Initialize the pattern buffer. */
1886 bufp->syntax = syntax;
1887 bufp->fastmap_accurate = 0;
1888 bufp->not_bol = bufp->not_eol = 0;
1890 /* Set `used' to zero, so that if we return an error, the pattern
1891 printer (for debugging) will think there's no pattern. We reset it
1895 /* Always count groups, whether or not bufp->no_sub is set. */
1898 #if !defined (emacs) && !defined (SYNTAX_TABLE)
1899 /* Initialize the syntax table. */
1900 init_syntax_once ();
1903 if (bufp->allocated == 0)
1906 { /* If zero allocated, but buffer is non-null, try to realloc
1907 enough space. This loses if buffer's address is bogus, but
1908 that is the user's responsibility. */
1909 RETALLOC (bufp->buffer, INIT_BUF_SIZE, unsigned char);
1912 { /* Caller did not allocate a buffer. Do it for them. */
1913 bufp->buffer = TALLOC (INIT_BUF_SIZE, unsigned char);
1915 if (!bufp->buffer) FREE_STACK_RETURN (REG_ESPACE);
1917 bufp->allocated = INIT_BUF_SIZE;
1920 begalt = b = bufp->buffer;
1922 /* Loop through the uncompiled pattern until we're at the end. */
1931 if ( /* If at start of pattern, it's an operator. */
1933 /* If context independent, it's an operator. */
1934 || syntax & RE_CONTEXT_INDEP_ANCHORS
1935 /* Otherwise, depends on what's come before. */
1936 || at_begline_loc_p (pattern, p, syntax))
1946 if ( /* If at end of pattern, it's an operator. */
1948 /* If context independent, it's an operator. */
1949 || syntax & RE_CONTEXT_INDEP_ANCHORS
1950 /* Otherwise, depends on what's next. */
1951 || at_endline_loc_p (p, pend, syntax))
1961 if ((syntax & RE_BK_PLUS_QM)
1962 || (syntax & RE_LIMITED_OPS))
1966 /* If there is no previous pattern... */
1969 if (syntax & RE_CONTEXT_INVALID_OPS)
1970 FREE_STACK_RETURN (REG_BADRPT);
1971 else if (!(syntax & RE_CONTEXT_INDEP_OPS))
1976 /* Are we optimizing this jump? */
1977 boolean keep_string_p = false;
1979 /* 1 means zero (many) matches is allowed. */
1980 char zero_times_ok = 0, many_times_ok = 0;
1982 /* If there is a sequence of repetition chars, collapse it
1983 down to just one (the right one). We can't combine
1984 interval operators with these because of, e.g., `a{2}*',
1985 which should only match an even number of `a's. */
1989 zero_times_ok |= c != '+';
1990 many_times_ok |= c != '?';
1998 || (!(syntax & RE_BK_PLUS_QM) && (c == '+' || c == '?')))
2001 else if (syntax & RE_BK_PLUS_QM && c == '\\')
2003 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
2006 if (!(c1 == '+' || c1 == '?'))
2021 /* If we get here, we found another repeat character. */
2024 /* Star, etc. applied to an empty pattern is equivalent
2025 to an empty pattern. */
2029 /* Now we know whether or not zero matches is allowed
2030 and also whether or not two or more matches is allowed. */
2032 { /* More than one repetition is allowed, so put in at the
2033 end a backward relative jump from `b' to before the next
2034 jump we're going to put in below (which jumps from
2035 laststart to after this jump).
2037 But if we are at the `*' in the exact sequence `.*\n',
2038 insert an unconditional jump backwards to the .,
2039 instead of the beginning of the loop. This way we only
2040 push a failure point once, instead of every time
2041 through the loop. */
2042 assert (p - 1 > pattern);
2044 /* Allocate the space for the jump. */
2045 GET_BUFFER_SPACE (3);
2047 /* We know we are not at the first character of the pattern,
2048 because laststart was nonzero. And we've already
2049 incremented `p', by the way, to be the character after
2050 the `*'. Do we have to do something analogous here
2051 for null bytes, because of RE_DOT_NOT_NULL? */
2052 if (TRANSLATE (*(p - 2)) == TRANSLATE ('.')
2054 && p < pend && TRANSLATE (*p) == TRANSLATE ('\n')
2055 && !(syntax & RE_DOT_NEWLINE))
2056 { /* We have .*\n. */
2057 STORE_JUMP (jump, b, laststart);
2058 keep_string_p = true;
2061 /* Anything else. */
2062 STORE_JUMP (maybe_pop_jump, b, laststart - 3);
2064 /* We've added more stuff to the buffer. */
2068 /* On failure, jump from laststart to b + 3, which will be the
2069 end of the buffer after this jump is inserted. */
2070 GET_BUFFER_SPACE (3);
2071 INSERT_JUMP (keep_string_p ? on_failure_keep_string_jump
2079 /* At least one repetition is required, so insert a
2080 `dummy_failure_jump' before the initial
2081 `on_failure_jump' instruction of the loop. This
2082 effects a skip over that instruction the first time
2083 we hit that loop. */
2084 GET_BUFFER_SPACE (3);
2085 INSERT_JUMP (dummy_failure_jump, laststart, laststart + 6);
2100 boolean had_char_class = false;
2102 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2104 /* Ensure that we have enough space to push a charset: the
2105 opcode, the length count, and the bitset; 34 bytes in all. */
2106 GET_BUFFER_SPACE (34);
2110 /* We test `*p == '^' twice, instead of using an if
2111 statement, so we only need one BUF_PUSH. */
2112 BUF_PUSH (*p == '^' ? charset_not : charset);
2116 /* Remember the first position in the bracket expression. */
2119 /* Push the number of bytes in the bitmap. */
2120 BUF_PUSH ((1 << BYTEWIDTH) / BYTEWIDTH);
2122 /* Clear the whole map. */
2123 bzero (b, (1 << BYTEWIDTH) / BYTEWIDTH);
2125 /* charset_not matches newline according to a syntax bit. */
2126 if ((re_opcode_t) b[-2] == charset_not
2127 && (syntax & RE_HAT_LISTS_NOT_NEWLINE))
2128 SET_LIST_BIT ('\n');
2130 /* Read in characters and ranges, setting map bits. */
2133 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2137 /* \ might escape characters inside [...] and [^...]. */
2138 if ((syntax & RE_BACKSLASH_ESCAPE_IN_LISTS) && c == '\\')
2140 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
2147 /* Could be the end of the bracket expression. If it's
2148 not (i.e., when the bracket expression is `[]' so
2149 far), the ']' character bit gets set way below. */
2150 if (c == ']' && p != p1 + 1)
2153 /* Look ahead to see if it's a range when the last thing
2154 was a character class. */
2155 if (had_char_class && c == '-' && *p != ']')
2156 FREE_STACK_RETURN (REG_ERANGE);
2158 /* Look ahead to see if it's a range when the last thing
2159 was a character: if this is a hyphen not at the
2160 beginning or the end of a list, then it's the range
2163 && !(p - 2 >= pattern && p[-2] == '[')
2164 && !(p - 3 >= pattern && p[-3] == '[' && p[-2] == '^')
2168 = compile_range (&p, pend, translate, syntax, b);
2169 if (ret != REG_NOERROR) FREE_STACK_RETURN (ret);
2172 else if (p[0] == '-' && p[1] != ']')
2173 { /* This handles ranges made up of characters only. */
2176 /* Move past the `-'. */
2179 ret = compile_range (&p, pend, translate, syntax, b);
2180 if (ret != REG_NOERROR) FREE_STACK_RETURN (ret);
2183 /* See if we're at the beginning of a possible character
2186 else if (syntax & RE_CHAR_CLASSES && c == '[' && *p == ':')
2187 { /* Leave room for the null. */
2188 char str[CHAR_CLASS_MAX_LENGTH + 1];
2193 /* If pattern is `[[:'. */
2194 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2199 if (c == ':' || c == ']' || p == pend
2200 || c1 == CHAR_CLASS_MAX_LENGTH)
2206 /* If isn't a word bracketed by `[:' and:`]':
2207 undo the ending character, the letters, and leave
2208 the leading `:' and `[' (but set bits for them). */
2209 if (c == ':' && *p == ']')
2211 #if defined _LIBC || (defined HAVE_WCTYPE_H && defined HAVE_WCHAR_H)
2212 boolean is_lower = STREQ (str, "lower");
2213 boolean is_upper = STREQ (str, "upper");
2219 FREE_STACK_RETURN (REG_ECTYPE);
2221 /* Throw away the ] at the end of the character
2225 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2227 for (ch = 0; ch < 1 << BYTEWIDTH; ++ch)
2229 if (iswctype (btowc (ch), wt))
2232 if (translate && (is_upper || is_lower)
2233 && (ISUPPER (ch) || ISLOWER (ch)))
2237 had_char_class = true;
2240 boolean is_alnum = STREQ (str, "alnum");
2241 boolean is_alpha = STREQ (str, "alpha");
2242 boolean is_blank = STREQ (str, "blank");
2243 boolean is_cntrl = STREQ (str, "cntrl");
2244 boolean is_digit = STREQ (str, "digit");
2245 boolean is_graph = STREQ (str, "graph");
2246 boolean is_lower = STREQ (str, "lower");
2247 boolean is_print = STREQ (str, "print");
2248 boolean is_punct = STREQ (str, "punct");
2249 boolean is_space = STREQ (str, "space");
2250 boolean is_upper = STREQ (str, "upper");
2251 boolean is_xdigit = STREQ (str, "xdigit");
2253 if (!IS_CHAR_CLASS (str))
2254 FREE_STACK_RETURN (REG_ECTYPE);
2256 /* Throw away the ] at the end of the character
2260 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2262 for (ch = 0; ch < 1 << BYTEWIDTH; ch++)
2264 /* This was split into 3 if's to
2265 avoid an arbitrary limit in some compiler. */
2266 if ( (is_alnum && ISALNUM (ch))
2267 || (is_alpha && ISALPHA (ch))
2268 || (is_blank && ISBLANK (ch))
2269 || (is_cntrl && ISCNTRL (ch)))
2271 if ( (is_digit && ISDIGIT (ch))
2272 || (is_graph && ISGRAPH (ch))
2273 || (is_lower && ISLOWER (ch))
2274 || (is_print && ISPRINT (ch)))
2276 if ( (is_punct && ISPUNCT (ch))
2277 || (is_space && ISSPACE (ch))
2278 || (is_upper && ISUPPER (ch))
2279 || (is_xdigit && ISXDIGIT (ch)))
2281 if ( translate && (is_upper || is_lower)
2282 && (ISUPPER (ch) || ISLOWER (ch)))
2285 had_char_class = true;
2286 #endif /* libc || wctype.h */
2295 had_char_class = false;
2300 had_char_class = false;
2305 /* Discard any (non)matching list bytes that are all 0 at the
2306 end of the map. Decrease the map-length byte too. */
2307 while ((int) b[-1] > 0 && b[b[-1] - 1] == 0)
2315 if (syntax & RE_NO_BK_PARENS)
2322 if (syntax & RE_NO_BK_PARENS)
2329 if (syntax & RE_NEWLINE_ALT)
2336 if (syntax & RE_NO_BK_VBAR)
2343 if (syntax & RE_INTERVALS && syntax & RE_NO_BK_BRACES)
2344 goto handle_interval;
2350 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
2352 /* Do not translate the character after the \, so that we can
2353 distinguish, e.g., \B from \b, even if we normally would
2354 translate, e.g., B to b. */
2360 if (syntax & RE_NO_BK_PARENS)
2361 goto normal_backslash;
2367 if (COMPILE_STACK_FULL)
2369 RETALLOC (compile_stack.stack, compile_stack.size << 1,
2370 compile_stack_elt_t);
2371 if (compile_stack.stack == NULL) return REG_ESPACE;
2373 compile_stack.size <<= 1;
2376 /* These are the values to restore when we hit end of this
2377 group. They are all relative offsets, so that if the
2378 whole pattern moves because of realloc, they will still
2380 COMPILE_STACK_TOP.begalt_offset = begalt - bufp->buffer;
2381 COMPILE_STACK_TOP.fixup_alt_jump
2382 = fixup_alt_jump ? fixup_alt_jump - bufp->buffer + 1 : 0;
2383 COMPILE_STACK_TOP.laststart_offset = b - bufp->buffer;
2384 COMPILE_STACK_TOP.regnum = regnum;
2386 /* We will eventually replace the 0 with the number of
2387 groups inner to this one. But do not push a
2388 start_memory for groups beyond the last one we can
2389 represent in the compiled pattern. */
2390 if (regnum <= MAX_REGNUM)
2392 COMPILE_STACK_TOP.inner_group_offset = b - bufp->buffer + 2;
2393 BUF_PUSH_3 (start_memory, regnum, 0);
2396 compile_stack.avail++;
2401 /* If we've reached MAX_REGNUM groups, then this open
2402 won't actually generate any code, so we'll have to
2403 clear pending_exact explicitly. */
2409 if (syntax & RE_NO_BK_PARENS) goto normal_backslash;
2411 if (COMPILE_STACK_EMPTY)
2412 if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD)
2413 goto normal_backslash;
2415 FREE_STACK_RETURN (REG_ERPAREN);
2419 { /* Push a dummy failure point at the end of the
2420 alternative for a possible future
2421 `pop_failure_jump' to pop. See comments at
2422 `push_dummy_failure' in `re_match_2'. */
2423 BUF_PUSH (push_dummy_failure);
2425 /* We allocated space for this jump when we assigned
2426 to `fixup_alt_jump', in the `handle_alt' case below. */
2427 STORE_JUMP (jump_past_alt, fixup_alt_jump, b - 1);
2430 /* See similar code for backslashed left paren above. */
2431 if (COMPILE_STACK_EMPTY)
2432 if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD)
2435 FREE_STACK_RETURN (REG_ERPAREN);
2437 /* Since we just checked for an empty stack above, this
2438 ``can't happen''. */
2439 assert (compile_stack.avail != 0);
2441 /* We don't just want to restore into `regnum', because
2442 later groups should continue to be numbered higher,
2443 as in `(ab)c(de)' -- the second group is #2. */
2444 regnum_t this_group_regnum;
2446 compile_stack.avail--;
2447 begalt = bufp->buffer + COMPILE_STACK_TOP.begalt_offset;
2449 = COMPILE_STACK_TOP.fixup_alt_jump
2450 ? bufp->buffer + COMPILE_STACK_TOP.fixup_alt_jump - 1
2452 laststart = bufp->buffer + COMPILE_STACK_TOP.laststart_offset;
2453 this_group_regnum = COMPILE_STACK_TOP.regnum;
2454 /* If we've reached MAX_REGNUM groups, then this open
2455 won't actually generate any code, so we'll have to
2456 clear pending_exact explicitly. */
2459 /* We're at the end of the group, so now we know how many
2460 groups were inside this one. */
2461 if (this_group_regnum <= MAX_REGNUM)
2463 unsigned char *inner_group_loc
2464 = bufp->buffer + COMPILE_STACK_TOP.inner_group_offset;
2466 *inner_group_loc = regnum - this_group_regnum;
2467 BUF_PUSH_3 (stop_memory, this_group_regnum,
2468 regnum - this_group_regnum);
2474 case '|': /* `\|'. */
2475 if (syntax & RE_LIMITED_OPS || syntax & RE_NO_BK_VBAR)
2476 goto normal_backslash;
2478 if (syntax & RE_LIMITED_OPS)
2481 /* Insert before the previous alternative a jump which
2482 jumps to this alternative if the former fails. */
2483 GET_BUFFER_SPACE (3);
2484 INSERT_JUMP (on_failure_jump, begalt, b + 6);
2488 /* The alternative before this one has a jump after it
2489 which gets executed if it gets matched. Adjust that
2490 jump so it will jump to this alternative's analogous
2491 jump (put in below, which in turn will jump to the next
2492 (if any) alternative's such jump, etc.). The last such
2493 jump jumps to the correct final destination. A picture:
2499 If we are at `b', then fixup_alt_jump right now points to a
2500 three-byte space after `a'. We'll put in the jump, set
2501 fixup_alt_jump to right after `b', and leave behind three
2502 bytes which we'll fill in when we get to after `c'. */
2505 STORE_JUMP (jump_past_alt, fixup_alt_jump, b);
2507 /* Mark and leave space for a jump after this alternative,
2508 to be filled in later either by next alternative or
2509 when know we're at the end of a series of alternatives. */
2511 GET_BUFFER_SPACE (3);
2520 /* If \{ is a literal. */
2521 if (!(syntax & RE_INTERVALS)
2522 /* If we're at `\{' and it's not the open-interval
2524 || ((syntax & RE_INTERVALS) && (syntax & RE_NO_BK_BRACES))
2525 || (p - 2 == pattern && p == pend))
2526 goto normal_backslash;
2530 /* If got here, then the syntax allows intervals. */
2532 /* At least (most) this many matches must be made. */
2533 int lower_bound = -1, upper_bound = -1;
2535 beg_interval = p - 1;
2539 if (syntax & RE_NO_BK_BRACES)
2540 goto unfetch_interval;
2542 FREE_STACK_RETURN (REG_EBRACE);
2545 GET_UNSIGNED_NUMBER (lower_bound);
2549 GET_UNSIGNED_NUMBER (upper_bound);
2550 if (upper_bound < 0) upper_bound = RE_DUP_MAX;
2553 /* Interval such as `{1}' => match exactly once. */
2554 upper_bound = lower_bound;
2556 if (lower_bound < 0 || upper_bound > RE_DUP_MAX
2557 || lower_bound > upper_bound)
2559 if (syntax & RE_NO_BK_BRACES)
2560 goto unfetch_interval;
2562 FREE_STACK_RETURN (REG_BADBR);
2565 if (!(syntax & RE_NO_BK_BRACES))
2567 if (c != '\\') FREE_STACK_RETURN (REG_EBRACE);
2574 if (syntax & RE_NO_BK_BRACES)
2575 goto unfetch_interval;
2577 FREE_STACK_RETURN (REG_BADBR);
2580 /* We just parsed a valid interval. */
2582 /* If it's invalid to have no preceding re. */
2585 if (syntax & RE_CONTEXT_INVALID_OPS)
2586 FREE_STACK_RETURN (REG_BADRPT);
2587 else if (syntax & RE_CONTEXT_INDEP_OPS)
2590 goto unfetch_interval;
2593 /* If the upper bound is zero, don't want to succeed at
2594 all; jump from `laststart' to `b + 3', which will be
2595 the end of the buffer after we insert the jump. */
2596 if (upper_bound == 0)
2598 GET_BUFFER_SPACE (3);
2599 INSERT_JUMP (jump, laststart, b + 3);
2603 /* Otherwise, we have a nontrivial interval. When
2604 we're all done, the pattern will look like:
2605 set_number_at <jump count> <upper bound>
2606 set_number_at <succeed_n count> <lower bound>
2607 succeed_n <after jump addr> <succeed_n count>
2609 jump_n <succeed_n addr> <jump count>
2610 (The upper bound and `jump_n' are omitted if
2611 `upper_bound' is 1, though.) */
2613 { /* If the upper bound is > 1, we need to insert
2614 more at the end of the loop. */
2615 unsigned nbytes = 10 + (upper_bound > 1) * 10;
2617 GET_BUFFER_SPACE (nbytes);
2619 /* Initialize lower bound of the `succeed_n', even
2620 though it will be set during matching by its
2621 attendant `set_number_at' (inserted next),
2622 because `re_compile_fastmap' needs to know.
2623 Jump to the `jump_n' we might insert below. */
2624 INSERT_JUMP2 (succeed_n, laststart,
2625 b + 5 + (upper_bound > 1) * 5,
2629 /* Code to initialize the lower bound. Insert
2630 before the `succeed_n'. The `5' is the last two
2631 bytes of this `set_number_at', plus 3 bytes of
2632 the following `succeed_n'. */
2633 insert_op2 (set_number_at, laststart, 5, lower_bound, b);
2636 if (upper_bound > 1)
2637 { /* More than one repetition is allowed, so
2638 append a backward jump to the `succeed_n'
2639 that starts this interval.
2641 When we've reached this during matching,
2642 we'll have matched the interval once, so
2643 jump back only `upper_bound - 1' times. */
2644 STORE_JUMP2 (jump_n, b, laststart + 5,
2648 /* The location we want to set is the second
2649 parameter of the `jump_n'; that is `b-2' as
2650 an absolute address. `laststart' will be
2651 the `set_number_at' we're about to insert;
2652 `laststart+3' the number to set, the source
2653 for the relative address. But we are
2654 inserting into the middle of the pattern --
2655 so everything is getting moved up by 5.
2656 Conclusion: (b - 2) - (laststart + 3) + 5,
2657 i.e., b - laststart.
2659 We insert this at the beginning of the loop
2660 so that if we fail during matching, we'll
2661 reinitialize the bounds. */
2662 insert_op2 (set_number_at, laststart, b - laststart,
2663 upper_bound - 1, b);
2668 beg_interval = NULL;
2673 /* If an invalid interval, match the characters as literals. */
2674 assert (beg_interval);
2676 beg_interval = NULL;
2678 /* normal_char and normal_backslash need `c'. */
2681 if (!(syntax & RE_NO_BK_BRACES))
2683 if (p > pattern && p[-1] == '\\')
2684 goto normal_backslash;
2689 /* There is no way to specify the before_dot and after_dot
2690 operators. rms says this is ok. --karl */
2698 BUF_PUSH_2 (syntaxspec, syntax_spec_code[c]);
2704 BUF_PUSH_2 (notsyntaxspec, syntax_spec_code[c]);
2710 if (re_syntax_options & RE_NO_GNU_OPS)
2713 BUF_PUSH (wordchar);
2718 if (re_syntax_options & RE_NO_GNU_OPS)
2721 BUF_PUSH (notwordchar);
2726 if (re_syntax_options & RE_NO_GNU_OPS)
2732 if (re_syntax_options & RE_NO_GNU_OPS)
2738 if (re_syntax_options & RE_NO_GNU_OPS)
2740 BUF_PUSH (wordbound);
2744 if (re_syntax_options & RE_NO_GNU_OPS)
2746 BUF_PUSH (notwordbound);
2750 if (re_syntax_options & RE_NO_GNU_OPS)
2756 if (re_syntax_options & RE_NO_GNU_OPS)
2761 case '1': case '2': case '3': case '4': case '5':
2762 case '6': case '7': case '8': case '9':
2763 if (syntax & RE_NO_BK_REFS)
2769 FREE_STACK_RETURN (REG_ESUBREG);
2771 /* Can't back reference to a subexpression if inside of it. */
2772 if (group_in_compile_stack (compile_stack, (regnum_t) c1))
2776 BUF_PUSH_2 (duplicate, c1);
2782 if (syntax & RE_BK_PLUS_QM)
2785 goto normal_backslash;
2789 /* You might think it would be useful for \ to mean
2790 not to translate; but if we don't translate it
2791 it will never match anything. */
2799 /* Expects the character in `c'. */
2801 /* If no exactn currently being built. */
2804 /* If last exactn not at current position. */
2805 || pending_exact + *pending_exact + 1 != b
2807 /* We have only one byte following the exactn for the count. */
2808 || *pending_exact == (1 << BYTEWIDTH) - 1
2810 /* If followed by a repetition operator. */
2811 || *p == '*' || *p == '^'
2812 || ((syntax & RE_BK_PLUS_QM)
2813 ? *p == '\\' && (p[1] == '+' || p[1] == '?')
2814 : (*p == '+' || *p == '?'))
2815 || ((syntax & RE_INTERVALS)
2816 && ((syntax & RE_NO_BK_BRACES)
2818 : (p[0] == '\\' && p[1] == '{'))))
2820 /* Start building a new exactn. */
2824 BUF_PUSH_2 (exactn, 0);
2825 pending_exact = b - 1;
2832 } /* while p != pend */
2835 /* Through the pattern now. */
2838 STORE_JUMP (jump_past_alt, fixup_alt_jump, b);
2840 if (!COMPILE_STACK_EMPTY)
2841 FREE_STACK_RETURN (REG_EPAREN);
2843 /* If we don't want backtracking, force success
2844 the first time we reach the end of the compiled pattern. */
2845 if (syntax & RE_NO_POSIX_BACKTRACKING)
2848 free (compile_stack.stack); /* __MEM_CHECKED__ */
2850 /* We have succeeded; set the length of the buffer. */
2851 bufp->used = b - bufp->buffer;
2856 DEBUG_PRINT1 ("\nCompiled pattern: \n");
2857 print_compiled_pattern (bufp);
2861 #ifndef MATCH_MAY_ALLOCATE
2862 /* Initialize the failure stack to the largest possible stack. This
2863 isn't necessary unless we're trying to avoid calling alloca in
2864 the search and match routines. */
2866 int num_regs = bufp->re_nsub + 1;
2868 /* Since DOUBLE_FAIL_STACK refuses to double only if the current size
2869 is strictly greater than re_max_failures, the largest possible stack
2870 is 2 * re_max_failures failure points. */
2871 if (fail_stack.size < (2 * re_max_failures * MAX_FAILURE_ITEMS))
2873 fail_stack.size = (2 * re_max_failures * MAX_FAILURE_ITEMS);
2876 if (! fail_stack.stack)
2878 = (fail_stack_elt_t *) xmalloc (fail_stack.size
2879 * sizeof (fail_stack_elt_t));
2882 = (fail_stack_elt_t *) xrealloc (fail_stack.stack,
2884 * sizeof (fail_stack_elt_t)));
2885 #else /* not emacs */
2886 if (! fail_stack.stack)
2888 = (fail_stack_elt_t *) malloc (fail_stack.size /* __MEM_CHECKED__ */
2889 * sizeof (fail_stack_elt_t));
2892 = (fail_stack_elt_t *) realloc (fail_stack.stack, /* __MEM_CHECKED__ */
2894 * sizeof (fail_stack_elt_t)));
2895 #endif /* not emacs */
2898 regex_grow_registers (num_regs);
2900 #endif /* not MATCH_MAY_ALLOCATE */
2903 } /* regex_compile */
2905 /* Subroutines for `regex_compile'. */
2907 /* Store OP at LOC followed by two-byte integer parameter ARG. */
2910 store_op1 (op, loc, arg)
2915 *loc = (unsigned char) op;
2916 STORE_NUMBER (loc + 1, arg);
2920 /* Like `store_op1', but for two two-byte parameters ARG1 and ARG2. */
2923 store_op2 (op, loc, arg1, arg2)
2928 *loc = (unsigned char) op;
2929 STORE_NUMBER (loc + 1, arg1);
2930 STORE_NUMBER (loc + 3, arg2);
2934 /* Copy the bytes from LOC to END to open up three bytes of space at LOC
2935 for OP followed by two-byte integer parameter ARG. */
2938 insert_op1 (op, loc, arg, end)
2944 register unsigned char *pfrom = end;
2945 register unsigned char *pto = end + 3;
2947 while (pfrom != loc)
2950 store_op1 (op, loc, arg);
2954 /* Like `insert_op1', but for two two-byte parameters ARG1 and ARG2. */
2957 insert_op2 (op, loc, arg1, arg2, end)
2963 register unsigned char *pfrom = end;
2964 register unsigned char *pto = end + 5;
2966 while (pfrom != loc)
2969 store_op2 (op, loc, arg1, arg2);
2973 /* P points to just after a ^ in PATTERN. Return true if that ^ comes
2974 after an alternative or a begin-subexpression. We assume there is at
2975 least one character before the ^. */
2978 at_begline_loc_p (pattern, p, syntax)
2979 const char *pattern, *p;
2980 reg_syntax_t syntax;
2982 const char *prev = p - 2;
2983 boolean prev_prev_backslash = prev > pattern && prev[-1] == '\\';
2986 /* After a subexpression? */
2987 (*prev == '(' && (syntax & RE_NO_BK_PARENS || prev_prev_backslash))
2988 /* After an alternative? */
2989 || (*prev == '|' && (syntax & RE_NO_BK_VBAR || prev_prev_backslash));
2993 /* The dual of at_begline_loc_p. This one is for $. We assume there is
2994 at least one character after the $, i.e., `P < PEND'. */
2997 at_endline_loc_p (p, pend, syntax)
2998 const char *p, *pend;
2999 reg_syntax_t syntax;
3001 const char *next = p;
3002 boolean next_backslash = *next == '\\';
3003 const char *next_next = p + 1 < pend ? p + 1 : 0;
3006 /* Before a subexpression? */
3007 (syntax & RE_NO_BK_PARENS ? *next == ')'
3008 : next_backslash && next_next && *next_next == ')')
3009 /* Before an alternative? */
3010 || (syntax & RE_NO_BK_VBAR ? *next == '|'
3011 : next_backslash && next_next && *next_next == '|');
3015 /* Returns true if REGNUM is in one of COMPILE_STACK's elements and
3016 false if it's not. */
3019 group_in_compile_stack (compile_stack, regnum)
3020 compile_stack_type compile_stack;
3025 for (this_element = compile_stack.avail - 1;
3028 if (compile_stack.stack[this_element].regnum == regnum)
3035 /* Read the ending character of a range (in a bracket expression) from the
3036 uncompiled pattern *P_PTR (which ends at PEND). We assume the
3037 starting character is in `P[-2]'. (`P[-1]' is the character `-'.)
3038 Then we set the translation of all bits between the starting and
3039 ending characters (inclusive) in the compiled pattern B.
3041 Return an error code.
3043 We use these short variable names so we can use the same macros as
3044 `regex_compile' itself. */
3046 static reg_errcode_t
3047 compile_range (p_ptr, pend, translate, syntax, b)
3048 const char **p_ptr, *pend;
3049 RE_TRANSLATE_TYPE translate;
3050 reg_syntax_t syntax;
3055 const char *p = *p_ptr;
3056 unsigned int range_start, range_end;
3061 /* Even though the pattern is a signed `char *', we need to fetch
3062 with unsigned char *'s; if the high bit of the pattern character
3063 is set, the range endpoints will be negative if we fetch using a
3066 We also want to fetch the endpoints without translating them; the
3067 appropriate translation is done in the bit-setting loop below. */
3068 /* The SVR4 compiler on the 3B2 had trouble with unsigned const char *. */
3069 range_start = ((const unsigned char *) p)[-2];
3070 range_end = ((const unsigned char *) p)[0];
3072 /* Have to increment the pointer into the pattern string, so the
3073 caller isn't still at the ending character. */
3076 /* If the start is after the end, the range is empty. */
3077 if (range_start > range_end)
3078 return syntax & RE_NO_EMPTY_RANGES ? REG_ERANGE : REG_NOERROR;
3080 /* Here we see why `this_char' has to be larger than an `unsigned
3081 char' -- the range is inclusive, so if `range_end' == 0xff
3082 (assuming 8-bit characters), we would otherwise go into an infinite
3083 loop, since all characters <= 0xff. */
3084 for (this_char = range_start; this_char <= range_end; this_char++)
3086 SET_LIST_BIT (TRANSLATE (this_char));
3092 /* re_compile_fastmap computes a ``fastmap'' for the compiled pattern in
3093 BUFP. A fastmap records which of the (1 << BYTEWIDTH) possible
3094 characters can start a string that matches the pattern. This fastmap
3095 is used by re_search to skip quickly over impossible starting points.
3097 The caller must supply the address of a (1 << BYTEWIDTH)-byte data
3098 area as BUFP->fastmap.
3100 We set the `fastmap', `fastmap_accurate', and `can_be_null' fields in
3103 Returns 0 if we succeed, -2 if an internal error. */
3106 re_compile_fastmap (bufp)
3107 struct re_pattern_buffer *bufp;
3110 #ifdef MATCH_MAY_ALLOCATE
3111 fail_stack_type fail_stack;
3113 #ifndef REGEX_MALLOC
3116 /* We don't push any register information onto the failure stack. */
3117 unsigned num_regs = 0;
3119 register char *fastmap = bufp->fastmap;
3120 unsigned char *pattern = bufp->buffer;
3121 unsigned char *p = pattern;
3122 register unsigned char *pend = pattern + bufp->used;
3125 /* This holds the pointer to the failure stack, when
3126 it is allocated relocatably. */
3127 fail_stack_elt_t *failure_stack_ptr;
3130 /* Assume that each path through the pattern can be null until
3131 proven otherwise. We set this false at the bottom of switch
3132 statement, to which we get only if a particular path doesn't
3133 match the empty string. */
3134 boolean path_can_be_null = true;
3136 /* We aren't doing a `succeed_n' to begin with. */
3137 boolean succeed_n_p = false;
3139 assert (fastmap != NULL && p != NULL);
3142 bzero (fastmap, 1 << BYTEWIDTH); /* Assume nothing's valid. */
3143 bufp->fastmap_accurate = 1; /* It will be when we're done. */
3144 bufp->can_be_null = 0;
3148 if (p == pend || *p == succeed)
3150 /* We have reached the (effective) end of pattern. */
3151 if (!FAIL_STACK_EMPTY ())
3153 bufp->can_be_null |= path_can_be_null;
3155 /* Reset for next path. */
3156 path_can_be_null = true;
3158 p = fail_stack.stack[--fail_stack.avail].pointer;
3166 /* We should never be about to go beyond the end of the pattern. */
3169 switch (SWITCH_ENUM_CAST ((re_opcode_t) *p++))
3172 /* I guess the idea here is to simply not bother with a fastmap
3173 if a backreference is used, since it's too hard to figure out
3174 the fastmap for the corresponding group. Setting
3175 `can_be_null' stops `re_search_2' from using the fastmap, so
3176 that is all we do. */
3178 bufp->can_be_null = 1;
3182 /* Following are the cases which match a character. These end
3191 for (j = *p++ * BYTEWIDTH - 1; j >= 0; j--)
3192 if (p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH)))
3198 /* Chars beyond end of map must be allowed. */
3199 for (j = *p * BYTEWIDTH; j < (1 << BYTEWIDTH); j++)
3202 for (j = *p++ * BYTEWIDTH - 1; j >= 0; j--)
3203 if (!(p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH))))
3209 for (j = 0; j < (1 << BYTEWIDTH); j++)
3210 if (SYNTAX (j) == Sword)
3216 for (j = 0; j < (1 << BYTEWIDTH); j++)
3217 if (SYNTAX (j) != Sword)
3224 int fastmap_newline = fastmap['\n'];
3226 /* `.' matches anything ... */
3227 for (j = 0; j < (1 << BYTEWIDTH); j++)
3230 /* ... except perhaps newline. */
3231 if (!(bufp->syntax & RE_DOT_NEWLINE))
3232 fastmap['\n'] = fastmap_newline;
3234 /* Return if we have already set `can_be_null'; if we have,
3235 then the fastmap is irrelevant. Something's wrong here. */
3236 else if (bufp->can_be_null)
3239 /* Otherwise, have to check alternative paths. */
3246 for (j = 0; j < (1 << BYTEWIDTH); j++)
3247 if (SYNTAX (j) == (enum syntaxcode) k)
3254 for (j = 0; j < (1 << BYTEWIDTH); j++)
3255 if (SYNTAX (j) != (enum syntaxcode) k)
3260 /* All cases after this match the empty string. These end with
3280 case push_dummy_failure:
3285 case pop_failure_jump:
3286 case maybe_pop_jump:
3289 case dummy_failure_jump:
3290 EXTRACT_NUMBER_AND_INCR (j, p);
3295 /* Jump backward implies we just went through the body of a
3296 loop and matched nothing. Opcode jumped to should be
3297 `on_failure_jump' or `succeed_n'. Just treat it like an
3298 ordinary jump. For a * loop, it has pushed its failure
3299 point already; if so, discard that as redundant. */
3300 if ((re_opcode_t) *p != on_failure_jump
3301 && (re_opcode_t) *p != succeed_n)
3305 EXTRACT_NUMBER_AND_INCR (j, p);
3308 /* If what's on the stack is where we are now, pop it. */
3309 if (!FAIL_STACK_EMPTY ()
3310 && fail_stack.stack[fail_stack.avail - 1].pointer == p)
3316 case on_failure_jump:
3317 case on_failure_keep_string_jump:
3318 handle_on_failure_jump:
3319 EXTRACT_NUMBER_AND_INCR (j, p);
3321 /* For some patterns, e.g., `(a?)?', `p+j' here points to the
3322 end of the pattern. We don't want to push such a point,
3323 since when we restore it above, entering the switch will
3324 increment `p' past the end of the pattern. We don't need
3325 to push such a point since we obviously won't find any more
3326 fastmap entries beyond `pend'. Such a pattern can match
3327 the null string, though. */
3330 if (!PUSH_PATTERN_OP (p + j, fail_stack))
3332 RESET_FAIL_STACK ();
3337 bufp->can_be_null = 1;
3341 EXTRACT_NUMBER_AND_INCR (k, p); /* Skip the n. */
3342 succeed_n_p = false;
3349 /* Get to the number of times to succeed. */
3352 /* Increment p past the n for when k != 0. */
3353 EXTRACT_NUMBER_AND_INCR (k, p);
3357 succeed_n_p = true; /* Spaghetti code alert. */
3358 goto handle_on_failure_jump;
3375 abort (); /* We have listed all the cases. */
3378 /* Getting here means we have found the possible starting
3379 characters for one path of the pattern -- and that the empty
3380 string does not match. We need not follow this path further.
3381 Instead, look at the next alternative (remembered on the
3382 stack), or quit if no more. The test at the top of the loop
3383 does these things. */
3384 path_can_be_null = false;
3388 /* Set `can_be_null' for the last path (also the first path, if the
3389 pattern is empty). */
3390 bufp->can_be_null |= path_can_be_null;
3393 RESET_FAIL_STACK ();
3395 } /* re_compile_fastmap */
3397 /* Set REGS to hold NUM_REGS registers, storing them in STARTS and
3398 ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use
3399 this memory for recording register information. STARTS and ENDS
3400 must be allocated using the malloc library routine, and must each
3401 be at least NUM_REGS * sizeof (regoff_t) bytes long.
3403 If NUM_REGS == 0, then subsequent matches should allocate their own
3406 Unless this function is called, the first search or match using
3407 PATTERN_BUFFER will allocate its own register data, without
3408 freeing the old data. */
3411 re_set_registers (bufp, regs, num_regs, starts, ends)
3412 struct re_pattern_buffer *bufp;
3413 struct re_registers *regs;
3415 regoff_t *starts, *ends;
3419 bufp->regs_allocated = REGS_REALLOCATE;
3420 regs->num_regs = num_regs;
3421 regs->start = starts;
3426 bufp->regs_allocated = REGS_UNALLOCATED;
3428 regs->start = regs->end = (regoff_t *) 0;
3432 /* Searching routines. */
3434 /* Like re_search_2, below, but only one string is specified, and
3435 doesn't let you say where to stop matching. */
3438 re_search (bufp, string, size, startpos, range, regs)
3439 struct re_pattern_buffer *bufp;
3441 int size, startpos, range;
3442 struct re_registers *regs;
3444 return re_search_2 (bufp, NULL, 0, string, size, startpos, range,
3449 /* Using the compiled pattern in BUFP->buffer, first tries to match the
3450 virtual concatenation of STRING1 and STRING2, starting first at index
3451 STARTPOS, then at STARTPOS + 1, and so on.
3453 STRING1 and STRING2 have length SIZE1 and SIZE2, respectively.
3455 RANGE is how far to scan while trying to match. RANGE = 0 means try
3456 only at STARTPOS; in general, the last start tried is STARTPOS +
3459 In REGS, return the indices of the virtual concatenation of STRING1
3460 and STRING2 that matched the entire BUFP->buffer and its contained
3463 Do not consider matching one past the index STOP in the virtual
3464 concatenation of STRING1 and STRING2.
3466 We return either the position in the strings at which the match was
3467 found, -1 if no match, or -2 if error (such as failure
3471 re_search_2 (bufp, string1, size1, string2, size2, startpos, range, regs, stop)
3472 struct re_pattern_buffer *bufp;
3473 const char *string1, *string2;
3477 struct re_registers *regs;
3481 register char *fastmap = bufp->fastmap;
3482 register RE_TRANSLATE_TYPE translate = bufp->translate;
3483 int total_size = size1 + size2;
3484 int endpos = startpos + range;
3486 /* Check for out-of-range STARTPOS. */
3487 if (startpos < 0 || startpos > total_size)
3490 /* Fix up RANGE if it might eventually take us outside
3491 the virtual concatenation of STRING1 and STRING2.
3492 Make sure we won't move STARTPOS below 0 or above TOTAL_SIZE. */
3494 range = 0 - startpos;
3495 else if (endpos > total_size)
3496 range = total_size - startpos;
3498 /* If the search isn't to be a backwards one, don't waste time in a
3499 search for a pattern that must be anchored. */
3500 if (bufp->used > 0 && (re_opcode_t) bufp->buffer[0] == begbuf && range > 0)
3509 /* In a forward search for something that starts with \=.
3510 don't keep searching past point. */
3511 if (bufp->used > 0 && (re_opcode_t) bufp->buffer[0] == at_dot && range > 0)
3513 range = PT - startpos;
3519 /* Update the fastmap now if not correct already. */
3520 if (fastmap && !bufp->fastmap_accurate)
3521 if (re_compile_fastmap (bufp) == -2)
3524 /* Loop through the string, looking for a place to start matching. */
3527 /* If a fastmap is supplied, skip quickly over characters that
3528 cannot be the start of a match. If the pattern can match the
3529 null string, however, we don't need to skip characters; we want
3530 the first null string. */
3531 if (fastmap && startpos < total_size && !bufp->can_be_null)
3533 if (range > 0) /* Searching forwards. */
3535 register const char *d;
3536 register int lim = 0;
3539 if (startpos < size1 && startpos + range >= size1)
3540 lim = range - (size1 - startpos);
3542 d = (startpos >= size1 ? string2 - size1 : string1) + startpos;
3544 /* Written out as an if-else to avoid testing `translate'
3548 && !fastmap[(unsigned char)
3549 translate[(unsigned char) *d++]])
3552 while (range > lim && !fastmap[(unsigned char) *d++])
3555 startpos += irange - range;
3557 else /* Searching backwards. */
3559 register char c = (size1 == 0 || startpos >= size1
3560 ? string2[startpos - size1]
3561 : string1[startpos]);
3563 if (!fastmap[(unsigned char) TRANSLATE (c)])
3568 /* If can't match the null string, and that's all we have left, fail. */
3569 if (range >= 0 && startpos == total_size && fastmap
3570 && !bufp->can_be_null)
3573 val = re_match_2_internal (bufp, string1, size1, string2, size2,
3574 startpos, regs, stop);
3575 #ifndef REGEX_MALLOC
3604 /* This converts PTR, a pointer into one of the search strings `string1'
3605 and `string2' into an offset from the beginning of that string. */
3606 #define POINTER_TO_OFFSET(ptr) \
3607 (FIRST_STRING_P (ptr) \
3608 ? ((regoff_t) ((ptr) - string1)) \
3609 : ((regoff_t) ((ptr) - string2 + size1)))
3611 /* Macros for dealing with the split strings in re_match_2. */
3613 #define MATCHING_IN_FIRST_STRING (dend == end_match_1)
3615 /* Call before fetching a character with *d. This switches over to
3616 string2 if necessary. */
3617 #define PREFETCH() \
3620 /* End of string2 => fail. */ \
3621 if (dend == end_match_2) \
3623 /* End of string1 => advance to string2. */ \
3625 dend = end_match_2; \
3629 /* Test if at very beginning or at very end of the virtual concatenation
3630 of `string1' and `string2'. If only one string, it's `string2'. */
3631 #define AT_STRINGS_BEG(d) ((d) == (size1 ? string1 : string2) || !size2)
3632 #define AT_STRINGS_END(d) ((d) == end2)
3635 /* Test if D points to a character which is word-constituent. We have
3636 two special cases to check for: if past the end of string1, look at
3637 the first character in string2; and if before the beginning of
3638 string2, look at the last character in string1. */
3639 #define WORDCHAR_P(d) \
3640 (SYNTAX ((d) == end1 ? *string2 \
3641 : (d) == string2 - 1 ? *(end1 - 1) : *(d)) \
3644 /* Disabled due to a compiler bug -- see comment at case wordbound */
3646 /* Test if the character before D and the one at D differ with respect
3647 to being word-constituent. */
3648 #define AT_WORD_BOUNDARY(d) \
3649 (AT_STRINGS_BEG (d) || AT_STRINGS_END (d) \
3650 || WORDCHAR_P (d - 1) != WORDCHAR_P (d))
3653 /* Free everything we malloc. */
3654 #ifdef MATCH_MAY_ALLOCATE
3655 #define FREE_VAR(var) if (var) REGEX_FREE (var); var = NULL
3656 #define FREE_VARIABLES() \
3658 REGEX_FREE_STACK (fail_stack.stack); \
3659 FREE_VAR (regstart); \
3660 FREE_VAR (regend); \
3661 FREE_VAR (old_regstart); \
3662 FREE_VAR (old_regend); \
3663 FREE_VAR (best_regstart); \
3664 FREE_VAR (best_regend); \
3665 FREE_VAR (reg_info); \
3666 FREE_VAR (reg_dummy); \
3667 FREE_VAR (reg_info_dummy); \
3670 #define FREE_VARIABLES() ((void)0) /* Do nothing! But inhibit gcc warning. */
3671 #endif /* not MATCH_MAY_ALLOCATE */
3673 /* These values must meet several constraints. They must not be valid
3674 register values; since we have a limit of 255 registers (because
3675 we use only one byte in the pattern for the register number), we can
3676 use numbers larger than 255. They must differ by 1, because of
3677 NUM_FAILURE_ITEMS above. And the value for the lowest register must
3678 be larger than the value for the highest register, so we do not try
3679 to actually save any registers when none are active. */
3680 #define NO_HIGHEST_ACTIVE_REG (1 << BYTEWIDTH)
3681 #define NO_LOWEST_ACTIVE_REG (NO_HIGHEST_ACTIVE_REG + 1)
3683 /* Matching routines. */
3685 #ifndef emacs /* Emacs never uses this. */
3686 /* re_match is like re_match_2 except it takes only a single string. */
3689 re_match (bufp, string, size, pos, regs)
3690 struct re_pattern_buffer *bufp;
3693 struct re_registers *regs;
3695 int result = re_match_2_internal (bufp, NULL, 0, string, size,
3697 #ifndef REGEX_MALLOC
3704 #endif /* not emacs */
3706 static boolean group_match_null_string_p _RE_ARGS ((unsigned char **p,
3708 register_info_type *reg_info));
3709 static boolean alt_match_null_string_p _RE_ARGS ((unsigned char *p,
3711 register_info_type *reg_info));
3712 static boolean common_op_match_null_string_p _RE_ARGS ((unsigned char **p,
3714 register_info_type *reg_info));
3715 static int bcmp_translate _RE_ARGS ((const char *s1, const char *s2,
3716 int len, char *translate));
3718 /* re_match_2 matches the compiled pattern in BUFP against the
3719 the (virtual) concatenation of STRING1 and STRING2 (of length SIZE1
3720 and SIZE2, respectively). We start matching at POS, and stop
3723 If REGS is non-null and the `no_sub' field of BUFP is nonzero, we
3724 store offsets for the substring each group matched in REGS. See the
3725 documentation for exactly how many groups we fill.
3727 We return -1 if no match, -2 if an internal error (such as the
3728 failure stack overflowing). Otherwise, we return the length of the
3729 matched substring. */
3732 re_match_2 (bufp, string1, size1, string2, size2, pos, regs, stop)
3733 struct re_pattern_buffer *bufp;
3734 const char *string1, *string2;
3737 struct re_registers *regs;
3740 int result = re_match_2_internal (bufp, string1, size1, string2, size2,
3742 #ifndef REGEX_MALLOC
3750 /* This is a separate function so that we can force an alloca cleanup
3753 re_match_2_internal (bufp, string1, size1, string2, size2, pos, regs, stop)
3754 struct re_pattern_buffer *bufp;
3755 const char *string1, *string2;
3758 struct re_registers *regs;
3761 /* General temporaries. */
3765 /* Just past the end of the corresponding string. */
3766 const char *end1, *end2;
3768 /* Pointers into string1 and string2, just past the last characters in
3769 each to consider matching. */
3770 const char *end_match_1, *end_match_2;
3772 /* Where we are in the data, and the end of the current string. */
3773 const char *d, *dend;
3775 /* Where we are in the pattern, and the end of the pattern. */
3776 unsigned char *p = bufp->buffer;
3777 register unsigned char *pend = p + bufp->used;
3779 /* Mark the opcode just after a start_memory, so we can test for an
3780 empty subpattern when we get to the stop_memory. */
3781 unsigned char *just_past_start_mem = 0;
3783 /* We use this to map every character in the string. */
3784 RE_TRANSLATE_TYPE translate = bufp->translate;
3786 /* Failure point stack. Each place that can handle a failure further
3787 down the line pushes a failure point on this stack. It consists of
3788 restart, regend, and reg_info for all registers corresponding to
3789 the subexpressions we're currently inside, plus the number of such
3790 registers, and, finally, two char *'s. The first char * is where
3791 to resume scanning the pattern; the second one is where to resume
3792 scanning the strings. If the latter is zero, the failure point is
3793 a ``dummy''; if a failure happens and the failure point is a dummy,
3794 it gets discarded and the next next one is tried. */
3795 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
3796 fail_stack_type fail_stack;
3799 static unsigned failure_id = 0;
3800 unsigned nfailure_points_pushed = 0, nfailure_points_popped = 0;
3804 /* This holds the pointer to the failure stack, when
3805 it is allocated relocatably. */
3806 fail_stack_elt_t *failure_stack_ptr;
3809 /* We fill all the registers internally, independent of what we
3810 return, for use in backreferences. The number here includes
3811 an element for register zero. */
3812 size_t num_regs = bufp->re_nsub + 1;
3814 /* The currently active registers. */
3815 active_reg_t lowest_active_reg = NO_LOWEST_ACTIVE_REG;
3816 active_reg_t highest_active_reg = NO_HIGHEST_ACTIVE_REG;
3818 /* Information on the contents of registers. These are pointers into
3819 the input strings; they record just what was matched (on this
3820 attempt) by a subexpression part of the pattern, that is, the
3821 regnum-th regstart pointer points to where in the pattern we began
3822 matching and the regnum-th regend points to right after where we
3823 stopped matching the regnum-th subexpression. (The zeroth register
3824 keeps track of what the whole pattern matches.) */
3825 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
3826 const char **regstart, **regend;
3829 /* If a group that's operated upon by a repetition operator fails to
3830 match anything, then the register for its start will need to be
3831 restored because it will have been set to wherever in the string we
3832 are when we last see its open-group operator. Similarly for a
3834 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
3835 const char **old_regstart, **old_regend;
3838 /* The is_active field of reg_info helps us keep track of which (possibly
3839 nested) subexpressions we are currently in. The matched_something
3840 field of reg_info[reg_num] helps us tell whether or not we have
3841 matched any of the pattern so far this time through the reg_num-th
3842 subexpression. These two fields get reset each time through any
3843 loop their register is in. */
3844 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
3845 register_info_type *reg_info;
3848 /* The following record the register info as found in the above
3849 variables when we find a match better than any we've seen before.
3850 This happens as we backtrack through the failure points, which in
3851 turn happens only if we have not yet matched the entire string. */
3852 unsigned best_regs_set = false;
3853 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
3854 const char **best_regstart, **best_regend;
3857 /* Logically, this is `best_regend[0]'. But we don't want to have to
3858 allocate space for that if we're not allocating space for anything
3859 else (see below). Also, we never need info about register 0 for
3860 any of the other register vectors, and it seems rather a kludge to
3861 treat `best_regend' differently than the rest. So we keep track of
3862 the end of the best match so far in a separate variable. We
3863 initialize this to NULL so that when we backtrack the first time
3864 and need to test it, it's not garbage. */
3865 const char *match_end = NULL;
3867 /* This helps SET_REGS_MATCHED avoid doing redundant work. */
3868 int set_regs_matched_done = 0;
3870 /* Used when we pop values we don't care about. */
3871 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
3872 const char **reg_dummy;
3873 register_info_type *reg_info_dummy;
3877 /* Counts the total number of registers pushed. */
3878 unsigned num_regs_pushed = 0;
3881 DEBUG_PRINT1 ("\n\nEntering re_match_2.\n");
3885 #ifdef MATCH_MAY_ALLOCATE
3886 /* Do not bother to initialize all the register variables if there are
3887 no groups in the pattern, as it takes a fair amount of time. If
3888 there are groups, we include space for register 0 (the whole
3889 pattern), even though we never use it, since it simplifies the
3890 array indexing. We should fix this. */
3893 regstart = REGEX_TALLOC (num_regs, const char *);
3894 regend = REGEX_TALLOC (num_regs, const char *);
3895 old_regstart = REGEX_TALLOC (num_regs, const char *);
3896 old_regend = REGEX_TALLOC (num_regs, const char *);
3897 best_regstart = REGEX_TALLOC (num_regs, const char *);
3898 best_regend = REGEX_TALLOC (num_regs, const char *);
3899 reg_info = REGEX_TALLOC (num_regs, register_info_type);
3900 reg_dummy = REGEX_TALLOC (num_regs, const char *);
3901 reg_info_dummy = REGEX_TALLOC (num_regs, register_info_type);
3903 if (!(regstart && regend && old_regstart && old_regend && reg_info
3904 && best_regstart && best_regend && reg_dummy && reg_info_dummy))
3912 /* We must initialize all our variables to NULL, so that
3913 `FREE_VARIABLES' doesn't try to free them. */
3914 regstart = regend = old_regstart = old_regend = best_regstart
3915 = best_regend = reg_dummy = NULL;
3916 reg_info = reg_info_dummy = (register_info_type *) NULL;
3918 #endif /* MATCH_MAY_ALLOCATE */
3920 /* The starting position is bogus. */
3921 if (pos < 0 || pos > size1 + size2)
3927 /* Initialize subexpression text positions to -1 to mark ones that no
3928 start_memory/stop_memory has been seen for. Also initialize the
3929 register information struct. */
3930 for (mcnt = 1; (unsigned) mcnt < num_regs; mcnt++)
3932 regstart[mcnt] = regend[mcnt]
3933 = old_regstart[mcnt] = old_regend[mcnt] = REG_UNSET_VALUE;
3935 REG_MATCH_NULL_STRING_P (reg_info[mcnt]) = MATCH_NULL_UNSET_VALUE;
3936 IS_ACTIVE (reg_info[mcnt]) = 0;
3937 MATCHED_SOMETHING (reg_info[mcnt]) = 0;
3938 EVER_MATCHED_SOMETHING (reg_info[mcnt]) = 0;
3941 /* We move `string1' into `string2' if the latter's empty -- but not if
3942 `string1' is null. */
3943 if (size2 == 0 && string1 != NULL)
3950 end1 = string1 + size1;
3951 end2 = string2 + size2;
3953 /* Compute where to stop matching, within the two strings. */
3956 end_match_1 = string1 + stop;
3957 end_match_2 = string2;
3962 end_match_2 = string2 + stop - size1;
3965 /* `p' scans through the pattern as `d' scans through the data.
3966 `dend' is the end of the input string that `d' points within. `d'
3967 is advanced into the following input string whenever necessary, but
3968 this happens before fetching; therefore, at the beginning of the
3969 loop, `d' can be pointing at the end of a string, but it cannot
3971 if (size1 > 0 && pos <= size1)
3978 d = string2 + pos - size1;
3982 DEBUG_PRINT1 ("The compiled pattern is:\n");
3983 DEBUG_PRINT_COMPILED_PATTERN (bufp, p, pend);
3984 DEBUG_PRINT1 ("The string to match is: `");
3985 DEBUG_PRINT_DOUBLE_STRING (d, string1, size1, string2, size2);
3986 DEBUG_PRINT1 ("'\n");
3988 /* This loops over pattern commands. It exits by returning from the
3989 function if the match is complete, or it drops through if the match
3990 fails at this starting point in the input data. */
3994 DEBUG_PRINT2 ("\n%p: ", p);
3996 DEBUG_PRINT2 ("\n0x%x: ", p);
4000 { /* End of pattern means we might have succeeded. */
4001 DEBUG_PRINT1 ("end of pattern ... ");
4003 /* If we haven't matched the entire string, and we want the
4004 longest match, try backtracking. */
4005 if (d != end_match_2)
4007 /* 1 if this match ends in the same string (string1 or string2)
4008 as the best previous match. */
4009 boolean same_str_p = (FIRST_STRING_P (match_end)
4010 == MATCHING_IN_FIRST_STRING);
4011 /* 1 if this match is the best seen so far. */
4012 boolean best_match_p;
4014 /* AIX compiler got confused when this was combined
4015 with the previous declaration. */
4017 best_match_p = d > match_end;
4019 best_match_p = !MATCHING_IN_FIRST_STRING;
4021 DEBUG_PRINT1 ("backtracking.\n");
4023 if (!FAIL_STACK_EMPTY ())
4024 { /* More failure points to try. */
4026 /* If exceeds best match so far, save it. */
4027 if (!best_regs_set || best_match_p)
4029 best_regs_set = true;
4032 DEBUG_PRINT1 ("\nSAVING match as best so far.\n");
4034 for (mcnt = 1; (unsigned) mcnt < num_regs; mcnt++)
4036 best_regstart[mcnt] = regstart[mcnt];
4037 best_regend[mcnt] = regend[mcnt];
4043 /* If no failure points, don't restore garbage. And if
4044 last match is real best match, don't restore second
4046 else if (best_regs_set && !best_match_p)
4049 /* Restore best match. It may happen that `dend ==
4050 end_match_1' while the restored d is in string2.
4051 For example, the pattern `x.*y.*z' against the
4052 strings `x-' and `y-z-', if the two strings are
4053 not consecutive in memory. */
4054 DEBUG_PRINT1 ("Restoring best registers.\n");
4057 dend = ((d >= string1 && d <= end1)
4058 ? end_match_1 : end_match_2);
4060 for (mcnt = 1; (unsigned) mcnt < num_regs; mcnt++)
4062 regstart[mcnt] = best_regstart[mcnt];
4063 regend[mcnt] = best_regend[mcnt];
4066 } /* d != end_match_2 */
4069 DEBUG_PRINT1 ("Accepting match.\n");
4071 /* If caller wants register contents data back, do it. */
4072 if (regs && !bufp->no_sub)
4074 /* Have the register data arrays been allocated? */
4075 if (bufp->regs_allocated == REGS_UNALLOCATED)
4076 { /* No. So allocate them with malloc. We need one
4077 extra element beyond `num_regs' for the `-1' marker
4079 regs->num_regs = MAX (RE_NREGS, num_regs + 1);
4080 regs->start = TALLOC (regs->num_regs, regoff_t);
4081 regs->end = TALLOC (regs->num_regs, regoff_t);
4082 if (regs->start == NULL || regs->end == NULL)
4087 bufp->regs_allocated = REGS_REALLOCATE;
4089 else if (bufp->regs_allocated == REGS_REALLOCATE)
4090 { /* Yes. If we need more elements than were already
4091 allocated, reallocate them. If we need fewer, just
4093 if (regs->num_regs < num_regs + 1)
4095 regs->num_regs = num_regs + 1;
4096 RETALLOC (regs->start, regs->num_regs, regoff_t);
4097 RETALLOC (regs->end, regs->num_regs, regoff_t);
4098 if (regs->start == NULL || regs->end == NULL)
4107 /* These braces fend off a "empty body in an else-statement"
4108 warning under GCC when assert expands to nothing. */
4109 assert (bufp->regs_allocated == REGS_FIXED);
4112 /* Convert the pointer data in `regstart' and `regend' to
4113 indices. Register zero has to be set differently,
4114 since we haven't kept track of any info for it. */
4115 if (regs->num_regs > 0)
4117 regs->start[0] = pos;
4118 regs->end[0] = (MATCHING_IN_FIRST_STRING
4119 ? ((regoff_t) (d - string1))
4120 : ((regoff_t) (d - string2 + size1)));
4123 /* Go through the first `min (num_regs, regs->num_regs)'
4124 registers, since that is all we initialized. */
4125 for (mcnt = 1; (unsigned) mcnt < MIN (num_regs, regs->num_regs);
4128 if (REG_UNSET (regstart[mcnt]) || REG_UNSET (regend[mcnt]))
4129 regs->start[mcnt] = regs->end[mcnt] = -1;
4133 = (regoff_t) POINTER_TO_OFFSET (regstart[mcnt]);
4135 = (regoff_t) POINTER_TO_OFFSET (regend[mcnt]);
4139 /* If the regs structure we return has more elements than
4140 were in the pattern, set the extra elements to -1. If
4141 we (re)allocated the registers, this is the case,
4142 because we always allocate enough to have at least one
4144 for (mcnt = num_regs; (unsigned) mcnt < regs->num_regs; mcnt++)
4145 regs->start[mcnt] = regs->end[mcnt] = -1;
4146 } /* regs && !bufp->no_sub */
4148 DEBUG_PRINT4 ("%u failure points pushed, %u popped (%u remain).\n",
4149 nfailure_points_pushed, nfailure_points_popped,
4150 nfailure_points_pushed - nfailure_points_popped);
4151 DEBUG_PRINT2 ("%u registers pushed.\n", num_regs_pushed);
4153 mcnt = d - pos - (MATCHING_IN_FIRST_STRING
4157 DEBUG_PRINT2 ("Returning %d from re_match_2.\n", mcnt);
4163 /* Otherwise match next pattern command. */
4164 switch (SWITCH_ENUM_CAST ((re_opcode_t) *p++))
4166 /* Ignore these. Used to ignore the n of succeed_n's which
4167 currently have n == 0. */
4169 DEBUG_PRINT1 ("EXECUTING no_op.\n");
4173 DEBUG_PRINT1 ("EXECUTING succeed.\n");
4176 /* Match the next n pattern characters exactly. The following
4177 byte in the pattern defines n, and the n bytes after that
4178 are the characters to match. */
4181 DEBUG_PRINT2 ("EXECUTING exactn %d.\n", mcnt);
4183 /* This is written out as an if-else so we don't waste time
4184 testing `translate' inside the loop. */
4190 if ((unsigned char) translate[(unsigned char) *d++]
4191 != (unsigned char) *p++)
4201 if (*d++ != (char) *p++) goto fail;
4205 SET_REGS_MATCHED ();
4209 /* Match any character except possibly a newline or a null. */
4211 DEBUG_PRINT1 ("EXECUTING anychar.\n");
4215 if ((!(bufp->syntax & RE_DOT_NEWLINE) && TRANSLATE (*d) == '\n')
4216 || (bufp->syntax & RE_DOT_NOT_NULL && TRANSLATE (*d) == '\000'))
4219 SET_REGS_MATCHED ();
4220 DEBUG_PRINT2 (" Matched `%d'.\n", *d);
4228 register unsigned char c;
4229 boolean not = (re_opcode_t) *(p - 1) == charset_not;
4231 DEBUG_PRINT2 ("EXECUTING charset%s.\n", not ? "_not" : "");
4234 c = TRANSLATE (*d); /* The character to match. */
4236 /* Cast to `unsigned' instead of `unsigned char' in case the
4237 bit list is a full 32 bytes long. */
4238 if (c < (unsigned) (*p * BYTEWIDTH)
4239 && p[1 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH)))
4244 if (!not) goto fail;
4246 SET_REGS_MATCHED ();
4252 /* The beginning of a group is represented by start_memory.
4253 The arguments are the register number in the next byte, and the
4254 number of groups inner to this one in the next. The text
4255 matched within the group is recorded (in the internal
4256 registers data structure) under the register number. */
4258 DEBUG_PRINT3 ("EXECUTING start_memory %d (%d):\n", *p, p[1]);
4260 /* Find out if this group can match the empty string. */
4261 p1 = p; /* To send to group_match_null_string_p. */
4263 if (REG_MATCH_NULL_STRING_P (reg_info[*p]) == MATCH_NULL_UNSET_VALUE)
4264 REG_MATCH_NULL_STRING_P (reg_info[*p])
4265 = group_match_null_string_p (&p1, pend, reg_info);
4267 /* Save the position in the string where we were the last time
4268 we were at this open-group operator in case the group is
4269 operated upon by a repetition operator, e.g., with `(a*)*b'
4270 against `ab'; then we want to ignore where we are now in
4271 the string in case this attempt to match fails. */
4272 old_regstart[*p] = REG_MATCH_NULL_STRING_P (reg_info[*p])
4273 ? REG_UNSET (regstart[*p]) ? d : regstart[*p]
4275 DEBUG_PRINT2 (" old_regstart: %d\n",
4276 POINTER_TO_OFFSET (old_regstart[*p]));
4279 DEBUG_PRINT2 (" regstart: %d\n", POINTER_TO_OFFSET (regstart[*p]));
4281 IS_ACTIVE (reg_info[*p]) = 1;
4282 MATCHED_SOMETHING (reg_info[*p]) = 0;
4284 /* Clear this whenever we change the register activity status. */
4285 set_regs_matched_done = 0;
4287 /* This is the new highest active register. */
4288 highest_active_reg = *p;
4290 /* If nothing was active before, this is the new lowest active
4292 if (lowest_active_reg == NO_LOWEST_ACTIVE_REG)
4293 lowest_active_reg = *p;
4295 /* Move past the register number and inner group count. */
4297 just_past_start_mem = p;
4302 /* The stop_memory opcode represents the end of a group. Its
4303 arguments are the same as start_memory's: the register
4304 number, and the number of inner groups. */
4306 DEBUG_PRINT3 ("EXECUTING stop_memory %d (%d):\n", *p, p[1]);
4308 /* We need to save the string position the last time we were at
4309 this close-group operator in case the group is operated
4310 upon by a repetition operator, e.g., with `((a*)*(b*)*)*'
4311 against `aba'; then we want to ignore where we are now in
4312 the string in case this attempt to match fails. */
4313 old_regend[*p] = REG_MATCH_NULL_STRING_P (reg_info[*p])
4314 ? REG_UNSET (regend[*p]) ? d : regend[*p]
4316 DEBUG_PRINT2 (" old_regend: %d\n",
4317 POINTER_TO_OFFSET (old_regend[*p]));
4320 DEBUG_PRINT2 (" regend: %d\n", POINTER_TO_OFFSET (regend[*p]));
4322 /* This register isn't active anymore. */
4323 IS_ACTIVE (reg_info[*p]) = 0;
4325 /* Clear this whenever we change the register activity status. */
4326 set_regs_matched_done = 0;
4328 /* If this was the only register active, nothing is active
4330 if (lowest_active_reg == highest_active_reg)
4332 lowest_active_reg = NO_LOWEST_ACTIVE_REG;
4333 highest_active_reg = NO_HIGHEST_ACTIVE_REG;
4336 { /* We must scan for the new highest active register, since
4337 it isn't necessarily one less than now: consider
4338 (a(b)c(d(e)f)g). When group 3 ends, after the f), the
4339 new highest active register is 1. */
4340 unsigned char r = *p - 1;
4341 while (r > 0 && !IS_ACTIVE (reg_info[r]))
4344 /* If we end up at register zero, that means that we saved
4345 the registers as the result of an `on_failure_jump', not
4346 a `start_memory', and we jumped to past the innermost
4347 `stop_memory'. For example, in ((.)*) we save
4348 registers 1 and 2 as a result of the *, but when we pop
4349 back to the second ), we are at the stop_memory 1.
4350 Thus, nothing is active. */
4353 lowest_active_reg = NO_LOWEST_ACTIVE_REG;
4354 highest_active_reg = NO_HIGHEST_ACTIVE_REG;
4357 highest_active_reg = r;
4360 /* If just failed to match something this time around with a
4361 group that's operated on by a repetition operator, try to
4362 force exit from the ``loop'', and restore the register
4363 information for this group that we had before trying this
4365 if ((!MATCHED_SOMETHING (reg_info[*p])
4366 || just_past_start_mem == p - 1)
4369 boolean is_a_jump_n = false;
4373 switch ((re_opcode_t) *p1++)
4377 case pop_failure_jump:
4378 case maybe_pop_jump:
4380 case dummy_failure_jump:
4381 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4391 /* If the next operation is a jump backwards in the pattern
4392 to an on_failure_jump right before the start_memory
4393 corresponding to this stop_memory, exit from the loop
4394 by forcing a failure after pushing on the stack the
4395 on_failure_jump's jump in the pattern, and d. */
4396 if (mcnt < 0 && (re_opcode_t) *p1 == on_failure_jump
4397 && (re_opcode_t) p1[3] == start_memory && p1[4] == *p)
4399 /* If this group ever matched anything, then restore
4400 what its registers were before trying this last
4401 failed match, e.g., with `(a*)*b' against `ab' for
4402 regstart[1], and, e.g., with `((a*)*(b*)*)*'
4403 against `aba' for regend[3].
4405 Also restore the registers for inner groups for,
4406 e.g., `((a*)(b*))*' against `aba' (register 3 would
4407 otherwise get trashed). */
4409 if (EVER_MATCHED_SOMETHING (reg_info[*p]))
4413 EVER_MATCHED_SOMETHING (reg_info[*p]) = 0;
4415 /* Restore this and inner groups' (if any) registers. */
4416 for (r = *p; r < (unsigned) *p + (unsigned) *(p + 1);
4419 regstart[r] = old_regstart[r];
4421 /* xx why this test? */
4422 if (old_regend[r] >= regstart[r])
4423 regend[r] = old_regend[r];
4427 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4428 PUSH_FAILURE_POINT (p1 + mcnt, d, -2);
4434 /* Move past the register number and the inner group count. */
4439 /* \<digit> has been turned into a `duplicate' command which is
4440 followed by the numeric value of <digit> as the register number. */
4443 register const char *d2, *dend2;
4444 int regno = *p++; /* Get which register to match against. */
4445 DEBUG_PRINT2 ("EXECUTING duplicate %d.\n", regno);
4447 /* Can't back reference a group which we've never matched. */
4448 if (REG_UNSET (regstart[regno]) || REG_UNSET (regend[regno]))
4451 /* Where in input to try to start matching. */
4452 d2 = regstart[regno];
4454 /* Where to stop matching; if both the place to start and
4455 the place to stop matching are in the same string, then
4456 set to the place to stop, otherwise, for now have to use
4457 the end of the first string. */
4459 dend2 = ((FIRST_STRING_P (regstart[regno])
4460 == FIRST_STRING_P (regend[regno]))
4461 ? regend[regno] : end_match_1);
4464 /* If necessary, advance to next segment in register
4468 if (dend2 == end_match_2) break;
4469 if (dend2 == regend[regno]) break;
4471 /* End of string1 => advance to string2. */
4473 dend2 = regend[regno];
4475 /* At end of register contents => success */
4476 if (d2 == dend2) break;
4478 /* If necessary, advance to next segment in data. */
4481 /* How many characters left in this segment to match. */
4484 /* Want how many consecutive characters we can match in
4485 one shot, so, if necessary, adjust the count. */
4486 if (mcnt > dend2 - d2)
4489 /* Compare that many; failure if mismatch, else move
4492 ? bcmp_translate (d, d2, mcnt, translate)
4493 : bcmp (d, d2, mcnt))
4495 d += mcnt, d2 += mcnt;
4497 /* Do this because we've match some characters. */
4498 SET_REGS_MATCHED ();
4504 /* begline matches the empty string at the beginning of the string
4505 (unless `not_bol' is set in `bufp'), and, if
4506 `newline_anchor' is set, after newlines. */
4508 DEBUG_PRINT1 ("EXECUTING begline.\n");
4510 if (AT_STRINGS_BEG (d))
4512 if (!bufp->not_bol) break;
4514 else if (d[-1] == '\n' && bufp->newline_anchor)
4518 /* In all other cases, we fail. */
4522 /* endline is the dual of begline. */
4524 DEBUG_PRINT1 ("EXECUTING endline.\n");
4526 if (AT_STRINGS_END (d))
4528 if (!bufp->not_eol) break;
4531 /* We have to ``prefetch'' the next character. */
4532 else if ((d == end1 ? *string2 : *d) == '\n'
4533 && bufp->newline_anchor)
4540 /* Match at the very beginning of the data. */
4542 DEBUG_PRINT1 ("EXECUTING begbuf.\n");
4543 if (AT_STRINGS_BEG (d))
4548 /* Match at the very end of the data. */
4550 DEBUG_PRINT1 ("EXECUTING endbuf.\n");
4551 if (AT_STRINGS_END (d))
4556 /* on_failure_keep_string_jump is used to optimize `.*\n'. It
4557 pushes NULL as the value for the string on the stack. Then
4558 `pop_failure_point' will keep the current value for the
4559 string, instead of restoring it. To see why, consider
4560 matching `foo\nbar' against `.*\n'. The .* matches the foo;
4561 then the . fails against the \n. But the next thing we want
4562 to do is match the \n against the \n; if we restored the
4563 string value, we would be back at the foo.
4565 Because this is used only in specific cases, we don't need to
4566 check all the things that `on_failure_jump' does, to make
4567 sure the right things get saved on the stack. Hence we don't
4568 share its code. The only reason to push anything on the
4569 stack at all is that otherwise we would have to change
4570 `anychar's code to do something besides goto fail in this
4571 case; that seems worse than this. */
4572 case on_failure_keep_string_jump:
4573 DEBUG_PRINT1 ("EXECUTING on_failure_keep_string_jump");
4575 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4577 DEBUG_PRINT3 (" %d (to %p):\n", mcnt, p + mcnt);
4579 DEBUG_PRINT3 (" %d (to 0x%x):\n", mcnt, p + mcnt);
4582 PUSH_FAILURE_POINT (p + mcnt, NULL, -2);
4586 /* Uses of on_failure_jump:
4588 Each alternative starts with an on_failure_jump that points
4589 to the beginning of the next alternative. Each alternative
4590 except the last ends with a jump that in effect jumps past
4591 the rest of the alternatives. (They really jump to the
4592 ending jump of the following alternative, because tensioning
4593 these jumps is a hassle.)
4595 Repeats start with an on_failure_jump that points past both
4596 the repetition text and either the following jump or
4597 pop_failure_jump back to this on_failure_jump. */
4598 case on_failure_jump:
4600 DEBUG_PRINT1 ("EXECUTING on_failure_jump");
4602 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4604 DEBUG_PRINT3 (" %d (to %p)", mcnt, p + mcnt);
4606 DEBUG_PRINT3 (" %d (to 0x%x)", mcnt, p + mcnt);
4609 /* If this on_failure_jump comes right before a group (i.e.,
4610 the original * applied to a group), save the information
4611 for that group and all inner ones, so that if we fail back
4612 to this point, the group's information will be correct.
4613 For example, in \(a*\)*\1, we need the preceding group,
4614 and in \(zz\(a*\)b*\)\2, we need the inner group. */
4616 /* We can't use `p' to check ahead because we push
4617 a failure point to `p + mcnt' after we do this. */
4620 /* We need to skip no_op's before we look for the
4621 start_memory in case this on_failure_jump is happening as
4622 the result of a completed succeed_n, as in \(a\)\{1,3\}b\1
4624 while (p1 < pend && (re_opcode_t) *p1 == no_op)
4627 if (p1 < pend && (re_opcode_t) *p1 == start_memory)
4629 /* We have a new highest active register now. This will
4630 get reset at the start_memory we are about to get to,
4631 but we will have saved all the registers relevant to
4632 this repetition op, as described above. */
4633 highest_active_reg = *(p1 + 1) + *(p1 + 2);
4634 if (lowest_active_reg == NO_LOWEST_ACTIVE_REG)
4635 lowest_active_reg = *(p1 + 1);
4638 DEBUG_PRINT1 (":\n");
4639 PUSH_FAILURE_POINT (p + mcnt, d, -2);
4643 /* A smart repeat ends with `maybe_pop_jump'.
4644 We change it to either `pop_failure_jump' or `jump'. */
4645 case maybe_pop_jump:
4646 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4647 DEBUG_PRINT2 ("EXECUTING maybe_pop_jump %d.\n", mcnt);
4649 register unsigned char *p2 = p;
4651 /* Compare the beginning of the repeat with what in the
4652 pattern follows its end. If we can establish that there
4653 is nothing that they would both match, i.e., that we
4654 would have to backtrack because of (as in, e.g., `a*a')
4655 then we can change to pop_failure_jump, because we'll
4656 never have to backtrack.
4658 This is not true in the case of alternatives: in
4659 `(a|ab)*' we do need to backtrack to the `ab' alternative
4660 (e.g., if the string was `ab'). But instead of trying to
4661 detect that here, the alternative has put on a dummy
4662 failure point which is what we will end up popping. */
4664 /* Skip over open/close-group commands.
4665 If what follows this loop is a ...+ construct,
4666 look at what begins its body, since we will have to
4667 match at least one of that. */
4671 && ((re_opcode_t) *p2 == stop_memory
4672 || (re_opcode_t) *p2 == start_memory))
4674 else if (p2 + 6 < pend
4675 && (re_opcode_t) *p2 == dummy_failure_jump)
4682 /* p1[0] ... p1[2] are the `on_failure_jump' corresponding
4683 to the `maybe_finalize_jump' of this case. Examine what
4686 /* If we're at the end of the pattern, we can change. */
4689 /* Consider what happens when matching ":\(.*\)"
4690 against ":/". I don't really understand this code
4692 p[-3] = (unsigned char) pop_failure_jump;
4694 (" End of pattern: change to `pop_failure_jump'.\n");
4697 else if ((re_opcode_t) *p2 == exactn
4698 || (bufp->newline_anchor && (re_opcode_t) *p2 == endline))
4700 register unsigned char c
4701 = *p2 == (unsigned char) endline ? '\n' : p2[2];
4703 if ((re_opcode_t) p1[3] == exactn && p1[5] != c)
4705 p[-3] = (unsigned char) pop_failure_jump;
4706 DEBUG_PRINT3 (" %c != %c => pop_failure_jump.\n",
4710 else if ((re_opcode_t) p1[3] == charset
4711 || (re_opcode_t) p1[3] == charset_not)
4713 int not = (re_opcode_t) p1[3] == charset_not;
4715 if (c < (unsigned char) (p1[4] * BYTEWIDTH)
4716 && p1[5 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH)))
4719 /* `not' is equal to 1 if c would match, which means
4720 that we can't change to pop_failure_jump. */
4723 p[-3] = (unsigned char) pop_failure_jump;
4724 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
4728 else if ((re_opcode_t) *p2 == charset)
4731 register unsigned char c
4732 = *p2 == (unsigned char) endline ? '\n' : p2[2];
4736 if ((re_opcode_t) p1[3] == exactn
4737 && ! ((int) p2[1] * BYTEWIDTH > (int) p1[5]
4738 && (p2[2 + p1[5] / BYTEWIDTH]
4739 & (1 << (p1[5] % BYTEWIDTH)))))
4741 if ((re_opcode_t) p1[3] == exactn
4742 && ! ((int) p2[1] * BYTEWIDTH > (int) p1[4]
4743 && (p2[2 + p1[4] / BYTEWIDTH]
4744 & (1 << (p1[4] % BYTEWIDTH)))))
4747 p[-3] = (unsigned char) pop_failure_jump;
4748 DEBUG_PRINT3 (" %c != %c => pop_failure_jump.\n",
4752 else if ((re_opcode_t) p1[3] == charset_not)
4755 /* We win if the charset_not inside the loop
4756 lists every character listed in the charset after. */
4757 for (idx = 0; idx < (int) p2[1]; idx++)
4758 if (! (p2[2 + idx] == 0
4759 || (idx < (int) p1[4]
4760 && ((p2[2 + idx] & ~ p1[5 + idx]) == 0))))
4765 p[-3] = (unsigned char) pop_failure_jump;
4766 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
4769 else if ((re_opcode_t) p1[3] == charset)
4772 /* We win if the charset inside the loop
4773 has no overlap with the one after the loop. */
4775 idx < (int) p2[1] && idx < (int) p1[4];
4777 if ((p2[2 + idx] & p1[5 + idx]) != 0)
4780 if (idx == p2[1] || idx == p1[4])
4782 p[-3] = (unsigned char) pop_failure_jump;
4783 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
4788 p -= 2; /* Point at relative address again. */
4789 if ((re_opcode_t) p[-1] != pop_failure_jump)
4791 p[-1] = (unsigned char) jump;
4792 DEBUG_PRINT1 (" Match => jump.\n");
4793 goto unconditional_jump;
4795 /* Note fall through. */
4798 /* The end of a simple repeat has a pop_failure_jump back to
4799 its matching on_failure_jump, where the latter will push a
4800 failure point. The pop_failure_jump takes off failure
4801 points put on by this pop_failure_jump's matching
4802 on_failure_jump; we got through the pattern to here from the
4803 matching on_failure_jump, so didn't fail. */
4804 case pop_failure_jump:
4806 /* We need to pass separate storage for the lowest and
4807 highest registers, even though we don't care about the
4808 actual values. Otherwise, we will restore only one
4809 register from the stack, since lowest will == highest in
4810 `pop_failure_point'. */
4811 active_reg_t dummy_low_reg, dummy_high_reg;
4812 unsigned char *pdummy;
4815 DEBUG_PRINT1 ("EXECUTING pop_failure_jump.\n");
4816 POP_FAILURE_POINT (sdummy, pdummy,
4817 dummy_low_reg, dummy_high_reg,
4818 reg_dummy, reg_dummy, reg_info_dummy);
4820 /* Note fall through. */
4824 DEBUG_PRINT2 ("\n%p: ", p);
4826 DEBUG_PRINT2 ("\n0x%x: ", p);
4828 /* Note fall through. */
4830 /* Unconditionally jump (without popping any failure points). */
4832 EXTRACT_NUMBER_AND_INCR (mcnt, p); /* Get the amount to jump. */
4833 DEBUG_PRINT2 ("EXECUTING jump %d ", mcnt);
4834 p += mcnt; /* Do the jump. */
4836 DEBUG_PRINT2 ("(to %p).\n", p);
4838 DEBUG_PRINT2 ("(to 0x%x).\n", p);
4843 /* We need this opcode so we can detect where alternatives end
4844 in `group_match_null_string_p' et al. */
4846 DEBUG_PRINT1 ("EXECUTING jump_past_alt.\n");
4847 goto unconditional_jump;
4850 /* Normally, the on_failure_jump pushes a failure point, which
4851 then gets popped at pop_failure_jump. We will end up at
4852 pop_failure_jump, also, and with a pattern of, say, `a+', we
4853 are skipping over the on_failure_jump, so we have to push
4854 something meaningless for pop_failure_jump to pop. */
4855 case dummy_failure_jump:
4856 DEBUG_PRINT1 ("EXECUTING dummy_failure_jump.\n");
4857 /* It doesn't matter what we push for the string here. What
4858 the code at `fail' tests is the value for the pattern. */
4859 PUSH_FAILURE_POINT (0, 0, -2);
4860 goto unconditional_jump;
4863 /* At the end of an alternative, we need to push a dummy failure
4864 point in case we are followed by a `pop_failure_jump', because
4865 we don't want the failure point for the alternative to be
4866 popped. For example, matching `(a|ab)*' against `aab'
4867 requires that we match the `ab' alternative. */
4868 case push_dummy_failure:
4869 DEBUG_PRINT1 ("EXECUTING push_dummy_failure.\n");
4870 /* See comments just above at `dummy_failure_jump' about the
4872 PUSH_FAILURE_POINT (0, 0, -2);
4875 /* Have to succeed matching what follows at least n times.
4876 After that, handle like `on_failure_jump'. */
4878 EXTRACT_NUMBER (mcnt, p + 2);
4879 DEBUG_PRINT2 ("EXECUTING succeed_n %d.\n", mcnt);
4882 /* Originally, this is how many times we HAVE to succeed. */
4887 STORE_NUMBER_AND_INCR (p, mcnt);
4889 DEBUG_PRINT3 (" Setting %p to %d.\n", p - 2, mcnt);
4891 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p - 2, mcnt);
4897 DEBUG_PRINT2 (" Setting two bytes from %p to no_op.\n", p+2);
4899 DEBUG_PRINT2 (" Setting two bytes from 0x%x to no_op.\n", p+2);
4901 p[2] = (unsigned char) no_op;
4902 p[3] = (unsigned char) no_op;
4908 EXTRACT_NUMBER (mcnt, p + 2);
4909 DEBUG_PRINT2 ("EXECUTING jump_n %d.\n", mcnt);
4911 /* Originally, this is how many times we CAN jump. */
4915 STORE_NUMBER (p + 2, mcnt);
4917 DEBUG_PRINT3 (" Setting %p to %d.\n", p + 2, mcnt);
4919 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p + 2, mcnt);
4921 goto unconditional_jump;
4923 /* If don't have to jump any more, skip over the rest of command. */
4930 DEBUG_PRINT1 ("EXECUTING set_number_at.\n");
4932 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4934 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4936 DEBUG_PRINT3 (" Setting %p to %d.\n", p1, mcnt);
4938 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p1, mcnt);
4940 STORE_NUMBER (p1, mcnt);
4945 /* The DEC Alpha C compiler 3.x generates incorrect code for the
4946 test WORDCHAR_P (d - 1) != WORDCHAR_P (d) in the expansion of
4947 AT_WORD_BOUNDARY, so this code is disabled. Expanding the
4948 macro and introducing temporary variables works around the bug. */
4951 DEBUG_PRINT1 ("EXECUTING wordbound.\n");
4952 if (AT_WORD_BOUNDARY (d))
4957 DEBUG_PRINT1 ("EXECUTING notwordbound.\n");
4958 if (AT_WORD_BOUNDARY (d))
4964 boolean prevchar, thischar;
4966 DEBUG_PRINT1 ("EXECUTING wordbound.\n");
4967 if (AT_STRINGS_BEG (d) || AT_STRINGS_END (d))
4970 prevchar = WORDCHAR_P (d - 1);
4971 thischar = WORDCHAR_P (d);
4972 if (prevchar != thischar)
4979 boolean prevchar, thischar;
4981 DEBUG_PRINT1 ("EXECUTING notwordbound.\n");
4982 if (AT_STRINGS_BEG (d) || AT_STRINGS_END (d))
4985 prevchar = WORDCHAR_P (d - 1);
4986 thischar = WORDCHAR_P (d);
4987 if (prevchar != thischar)
4994 DEBUG_PRINT1 ("EXECUTING wordbeg.\n");
4995 if (WORDCHAR_P (d) && (AT_STRINGS_BEG (d) || !WORDCHAR_P (d - 1)))
5000 DEBUG_PRINT1 ("EXECUTING wordend.\n");
5001 if (!AT_STRINGS_BEG (d) && WORDCHAR_P (d - 1)
5002 && (!WORDCHAR_P (d) || AT_STRINGS_END (d)))
5008 DEBUG_PRINT1 ("EXECUTING before_dot.\n");
5009 if (PTR_CHAR_POS ((unsigned char *) d) >= point)
5014 DEBUG_PRINT1 ("EXECUTING at_dot.\n");
5015 if (PTR_CHAR_POS ((unsigned char *) d) != point)
5020 DEBUG_PRINT1 ("EXECUTING after_dot.\n");
5021 if (PTR_CHAR_POS ((unsigned char *) d) <= point)
5026 DEBUG_PRINT2 ("EXECUTING syntaxspec %d.\n", mcnt);
5031 DEBUG_PRINT1 ("EXECUTING Emacs wordchar.\n");
5035 /* Can't use *d++ here; SYNTAX may be an unsafe macro. */
5037 if (SYNTAX (d[-1]) != (enum syntaxcode) mcnt)
5039 SET_REGS_MATCHED ();
5043 DEBUG_PRINT2 ("EXECUTING notsyntaxspec %d.\n", mcnt);
5045 goto matchnotsyntax;
5048 DEBUG_PRINT1 ("EXECUTING Emacs notwordchar.\n");
5052 /* Can't use *d++ here; SYNTAX may be an unsafe macro. */
5054 if (SYNTAX (d[-1]) == (enum syntaxcode) mcnt)
5056 SET_REGS_MATCHED ();
5059 #else /* not emacs */
5061 DEBUG_PRINT1 ("EXECUTING non-Emacs wordchar.\n");
5063 if (!WORDCHAR_P (d))
5065 SET_REGS_MATCHED ();
5070 DEBUG_PRINT1 ("EXECUTING non-Emacs notwordchar.\n");
5074 SET_REGS_MATCHED ();
5077 #endif /* not emacs */
5082 continue; /* Successfully executed one pattern command; keep going. */
5085 /* We goto here if a matching operation fails. */
5087 if (!FAIL_STACK_EMPTY ())
5088 { /* A restart point is known. Restore to that state. */
5089 DEBUG_PRINT1 ("\nFAIL:\n");
5090 POP_FAILURE_POINT (d, p,
5091 lowest_active_reg, highest_active_reg,
5092 regstart, regend, reg_info);
5094 /* If this failure point is a dummy, try the next one. */
5098 /* If we failed to the end of the pattern, don't examine *p. */
5102 boolean is_a_jump_n = false;
5104 /* If failed to a backwards jump that's part of a repetition
5105 loop, need to pop this failure point and use the next one. */
5106 switch ((re_opcode_t) *p)
5110 case maybe_pop_jump:
5111 case pop_failure_jump:
5114 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
5117 if ((is_a_jump_n && (re_opcode_t) *p1 == succeed_n)
5119 && (re_opcode_t) *p1 == on_failure_jump))
5127 if (d >= string1 && d <= end1)
5131 break; /* Matching at this starting point really fails. */
5135 goto restore_best_regs;
5139 return -1; /* Failure to match. */
5142 /* Subroutine definitions for re_match_2. */
5145 /* We are passed P pointing to a register number after a start_memory.
5147 Return true if the pattern up to the corresponding stop_memory can
5148 match the empty string, and false otherwise.
5150 If we find the matching stop_memory, sets P to point to one past its number.
5151 Otherwise, sets P to an undefined byte less than or equal to END.
5153 We don't handle duplicates properly (yet). */
5156 group_match_null_string_p (p, end, reg_info)
5157 unsigned char **p, *end;
5158 register_info_type *reg_info;
5161 /* Point to after the args to the start_memory. */
5162 unsigned char *p1 = *p + 2;
5166 /* Skip over opcodes that can match nothing, and return true or
5167 false, as appropriate, when we get to one that can't, or to the
5168 matching stop_memory. */
5170 switch ((re_opcode_t) *p1)
5172 /* Could be either a loop or a series of alternatives. */
5173 case on_failure_jump:
5175 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
5177 /* If the next operation is not a jump backwards in the
5182 /* Go through the on_failure_jumps of the alternatives,
5183 seeing if any of the alternatives cannot match nothing.
5184 The last alternative starts with only a jump,
5185 whereas the rest start with on_failure_jump and end
5186 with a jump, e.g., here is the pattern for `a|b|c':
5188 /on_failure_jump/0/6/exactn/1/a/jump_past_alt/0/6
5189 /on_failure_jump/0/6/exactn/1/b/jump_past_alt/0/3
5192 So, we have to first go through the first (n-1)
5193 alternatives and then deal with the last one separately. */
5196 /* Deal with the first (n-1) alternatives, which start
5197 with an on_failure_jump (see above) that jumps to right
5198 past a jump_past_alt. */
5200 while ((re_opcode_t) p1[mcnt-3] == jump_past_alt)
5202 /* `mcnt' holds how many bytes long the alternative
5203 is, including the ending `jump_past_alt' and
5206 if (!alt_match_null_string_p (p1, p1 + mcnt - 3,
5210 /* Move to right after this alternative, including the
5214 /* Break if it's the beginning of an n-th alternative
5215 that doesn't begin with an on_failure_jump. */
5216 if ((re_opcode_t) *p1 != on_failure_jump)
5219 /* Still have to check that it's not an n-th
5220 alternative that starts with an on_failure_jump. */
5222 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
5223 if ((re_opcode_t) p1[mcnt-3] != jump_past_alt)
5225 /* Get to the beginning of the n-th alternative. */
5231 /* Deal with the last alternative: go back and get number
5232 of the `jump_past_alt' just before it. `mcnt' contains
5233 the length of the alternative. */
5234 EXTRACT_NUMBER (mcnt, p1 - 2);
5236 if (!alt_match_null_string_p (p1, p1 + mcnt, reg_info))
5239 p1 += mcnt; /* Get past the n-th alternative. */
5245 assert (p1[1] == **p);
5251 if (!common_op_match_null_string_p (&p1, end, reg_info))
5254 } /* while p1 < end */
5257 } /* group_match_null_string_p */
5260 /* Similar to group_match_null_string_p, but doesn't deal with alternatives:
5261 It expects P to be the first byte of a single alternative and END one
5262 byte past the last. The alternative can contain groups. */
5265 alt_match_null_string_p (p, end, reg_info)
5266 unsigned char *p, *end;
5267 register_info_type *reg_info;
5270 unsigned char *p1 = p;
5274 /* Skip over opcodes that can match nothing, and break when we get
5275 to one that can't. */
5277 switch ((re_opcode_t) *p1)
5280 case on_failure_jump:
5282 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
5287 if (!common_op_match_null_string_p (&p1, end, reg_info))
5290 } /* while p1 < end */
5293 } /* alt_match_null_string_p */
5296 /* Deals with the ops common to group_match_null_string_p and
5297 alt_match_null_string_p.
5299 Sets P to one after the op and its arguments, if any. */
5302 common_op_match_null_string_p (p, end, reg_info)
5303 unsigned char **p, *end;
5304 register_info_type *reg_info;
5309 unsigned char *p1 = *p;
5311 switch ((re_opcode_t) *p1++)
5331 assert (reg_no > 0 && reg_no <= MAX_REGNUM);
5332 ret = group_match_null_string_p (&p1, end, reg_info);
5334 /* Have to set this here in case we're checking a group which
5335 contains a group and a back reference to it. */
5337 if (REG_MATCH_NULL_STRING_P (reg_info[reg_no]) == MATCH_NULL_UNSET_VALUE)
5338 REG_MATCH_NULL_STRING_P (reg_info[reg_no]) = ret;
5344 /* If this is an optimized succeed_n for zero times, make the jump. */
5346 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
5354 /* Get to the number of times to succeed. */
5356 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
5361 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
5369 if (!REG_MATCH_NULL_STRING_P (reg_info[*p1]))
5377 /* All other opcodes mean we cannot match the empty string. */
5383 } /* common_op_match_null_string_p */
5386 /* Return zero if TRANSLATE[S1] and TRANSLATE[S2] are identical for LEN
5387 bytes; nonzero otherwise. */
5390 bcmp_translate (s1, s2, len, translate)
5391 const char *s1, *s2;
5393 RE_TRANSLATE_TYPE translate;
5395 register const unsigned char *p1 = (const unsigned char *) s1;
5396 register const unsigned char *p2 = (const unsigned char *) s2;
5399 if (translate[*p1++] != translate[*p2++]) return 1;
5405 /* Entry points for GNU code. */
5407 /* re_compile_pattern is the GNU regular expression compiler: it
5408 compiles PATTERN (of length SIZE) and puts the result in BUFP.
5409 Returns 0 if the pattern was valid, otherwise an error string.
5411 Assumes the `allocated' (and perhaps `buffer') and `translate' fields
5412 are set in BUFP on entry.
5414 We call regex_compile to do the actual compilation. */
5417 re_compile_pattern (pattern, length, bufp)
5418 const char *pattern;
5420 struct re_pattern_buffer *bufp;
5424 /* GNU code is written to assume at least RE_NREGS registers will be set
5425 (and at least one extra will be -1). */
5426 bufp->regs_allocated = REGS_UNALLOCATED;
5428 /* And GNU code determines whether or not to get register information
5429 by passing null for the REGS argument to re_match, etc., not by
5433 /* Match anchors at newline. */
5434 bufp->newline_anchor = 1;
5436 ret = regex_compile (pattern, length, re_syntax_options, bufp);
5440 return gettext (re_error_msgid[(int) ret]);
5443 /* Entry points compatible with 4.2 BSD regex library. We don't define
5444 them unless specifically requested. */
5446 #if defined (_REGEX_RE_COMP) || defined (_LIBC)
5448 /* BSD has one and only one pattern buffer. */
5449 static struct re_pattern_buffer re_comp_buf;
5453 /* Make these definitions weak in libc, so POSIX programs can redefine
5454 these names if they don't use our functions, and still use
5455 regcomp/regexec below without link errors. */
5465 if (!re_comp_buf.buffer)
5466 return gettext ("No previous regular expression");
5470 if (!re_comp_buf.buffer)
5472 re_comp_buf.buffer = (unsigned char *) malloc (200); /* __MEM_CHECKED__ */
5473 if (re_comp_buf.buffer == NULL)
5474 return gettext (re_error_msgid[(int) REG_ESPACE]);
5475 re_comp_buf.allocated = 200;
5477 re_comp_buf.fastmap = (char *) malloc (1 << BYTEWIDTH); /* __MEM_CHECKED__ */
5478 if (re_comp_buf.fastmap == NULL)
5479 return gettext (re_error_msgid[(int) REG_ESPACE]);
5482 /* Since `re_exec' always passes NULL for the `regs' argument, we
5483 don't need to initialize the pattern buffer fields which affect it. */
5485 /* Match anchors at newlines. */
5486 re_comp_buf.newline_anchor = 1;
5488 ret = regex_compile (s, strlen (s), re_syntax_options, &re_comp_buf);
5493 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
5494 return (char *) gettext (re_error_msgid[(int) ret]);
5505 const int len = strlen (s);
5507 0 <= re_search (&re_comp_buf, s, len, 0, len, (struct re_registers *) 0);
5510 #endif /* _REGEX_RE_COMP */
5512 /* POSIX.2 functions. Don't define these for Emacs. */
5516 /* regcomp takes a regular expression as a string and compiles it.
5518 PREG is a regex_t *. We do not expect any fields to be initialized,
5519 since POSIX says we shouldn't. Thus, we set
5521 `buffer' to the compiled pattern;
5522 `used' to the length of the compiled pattern;
5523 `syntax' to RE_SYNTAX_POSIX_EXTENDED if the
5524 REG_EXTENDED bit in CFLAGS is set; otherwise, to
5525 RE_SYNTAX_POSIX_BASIC;
5526 `newline_anchor' to REG_NEWLINE being set in CFLAGS;
5527 `fastmap' and `fastmap_accurate' to zero;
5528 `re_nsub' to the number of subexpressions in PATTERN.
5530 PATTERN is the address of the pattern string.
5532 CFLAGS is a series of bits which affect compilation.
5534 If REG_EXTENDED is set, we use POSIX extended syntax; otherwise, we
5535 use POSIX basic syntax.
5537 If REG_NEWLINE is set, then . and [^...] don't match newline.
5538 Also, regexec will try a match beginning after every newline.
5540 If REG_ICASE is set, then we considers upper- and lowercase
5541 versions of letters to be equivalent when matching.
5543 If REG_NOSUB is set, then when PREG is passed to regexec, that
5544 routine will report only success or failure, and nothing about the
5547 It returns 0 if it succeeds, nonzero if it doesn't. (See regex.h for
5548 the return codes and their meanings.) */
5551 regcomp (preg, pattern, cflags)
5553 const char *pattern;
5558 = (cflags & REG_EXTENDED) ?
5559 RE_SYNTAX_POSIX_EXTENDED : RE_SYNTAX_POSIX_BASIC;
5561 /* regex_compile will allocate the space for the compiled pattern. */
5563 preg->allocated = 0;
5566 /* Don't bother to use a fastmap when searching. This simplifies the
5567 REG_NEWLINE case: if we used a fastmap, we'd have to put all the
5568 characters after newlines into the fastmap. This way, we just try
5572 if (cflags & REG_ICASE)
5577 = (RE_TRANSLATE_TYPE) malloc (CHAR_SET_SIZE /* __MEM_CHECKED__ */
5578 * sizeof (*(RE_TRANSLATE_TYPE)0));
5579 if (preg->translate == NULL)
5580 return (int) REG_ESPACE;
5582 /* Map uppercase characters to corresponding lowercase ones. */
5583 for (i = 0; i < CHAR_SET_SIZE; i++)
5584 preg->translate[i] = ISUPPER (i) ? tolower (i) : i;
5587 preg->translate = NULL;
5589 /* If REG_NEWLINE is set, newlines are treated differently. */
5590 if (cflags & REG_NEWLINE)
5591 { /* REG_NEWLINE implies neither . nor [^...] match newline. */
5592 syntax &= ~RE_DOT_NEWLINE;
5593 syntax |= RE_HAT_LISTS_NOT_NEWLINE;
5594 /* It also changes the matching behavior. */
5595 preg->newline_anchor = 1;
5598 preg->newline_anchor = 0;
5600 preg->no_sub = !!(cflags & REG_NOSUB);
5602 /* POSIX says a null character in the pattern terminates it, so we
5603 can use strlen here in compiling the pattern. */
5604 ret = regex_compile (pattern, strlen (pattern), syntax, preg);
5606 /* POSIX doesn't distinguish between an unmatched open-group and an
5607 unmatched close-group: both are REG_EPAREN. */
5608 if (ret == REG_ERPAREN) ret = REG_EPAREN;
5614 /* regexec searches for a given pattern, specified by PREG, in the
5617 If NMATCH is zero or REG_NOSUB was set in the cflags argument to
5618 `regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at
5619 least NMATCH elements, and we set them to the offsets of the
5620 corresponding matched substrings.
5622 EFLAGS specifies `execution flags' which affect matching: if
5623 REG_NOTBOL is set, then ^ does not match at the beginning of the
5624 string; if REG_NOTEOL is set, then $ does not match at the end.
5626 We return 0 if we find a match and REG_NOMATCH if not. */
5629 regexec (preg, string, nmatch, pmatch, eflags)
5630 const regex_t *preg;
5633 regmatch_t pmatch[];
5637 struct re_registers regs;
5638 regex_t private_preg;
5639 int len = strlen (string);
5640 boolean want_reg_info = !preg->no_sub && nmatch > 0;
5642 private_preg = *preg;
5644 private_preg.not_bol = !!(eflags & REG_NOTBOL);
5645 private_preg.not_eol = !!(eflags & REG_NOTEOL);
5647 /* The user has told us exactly how many registers to return
5648 information about, via `nmatch'. We have to pass that on to the
5649 matching routines. */
5650 private_preg.regs_allocated = REGS_FIXED;
5654 regs.num_regs = nmatch;
5655 regs.start = TALLOC (nmatch, regoff_t);
5656 regs.end = TALLOC (nmatch, regoff_t);
5657 if (regs.start == NULL || regs.end == NULL)
5658 return (int) REG_NOMATCH;
5661 /* Perform the searching operation. */
5662 ret = re_search (&private_preg, string, len,
5663 /* start: */ 0, /* range: */ len,
5664 want_reg_info ? ®s : (struct re_registers *) 0);
5666 /* Copy the register information to the POSIX structure. */
5673 for (r = 0; r < nmatch; r++)
5675 pmatch[r].rm_so = regs.start[r];
5676 pmatch[r].rm_eo = regs.end[r];
5680 /* If we needed the temporary register info, free the space now. */
5681 free (regs.start); /* __MEM_CHECKED__ */
5682 free (regs.end); /* __MEM_CHECKED__ */
5685 /* We want zero return to mean success, unlike `re_search'. */
5686 return ret >= 0 ? (int) REG_NOERROR : (int) REG_NOMATCH;
5690 /* Returns a message corresponding to an error code, ERRCODE, returned
5691 from either regcomp or regexec. We don't use PREG here. */
5694 regerror (errcode, preg, errbuf, errbuf_size)
5696 const regex_t *preg;
5704 || errcode >= (int) (sizeof (re_error_msgid)
5705 / sizeof (re_error_msgid[0])))
5706 /* Only error codes returned by the rest of the code should be passed
5707 to this routine. If we are given anything else, or if other regex
5708 code generates an invalid error code, then the program has a bug.
5709 Dump core so we can fix it. */
5712 msg = gettext (re_error_msgid[errcode]);
5714 msg_size = strlen (msg) + 1; /* Includes the null. */
5716 if (errbuf_size != 0)
5718 if (msg_size > errbuf_size)
5720 strncpy (errbuf, msg, errbuf_size - 1);
5721 errbuf[errbuf_size - 1] = 0;
5724 strcpy (errbuf, msg); /* __STRCPY_CHECKED__ */
5731 /* Free dynamically allocated space used by PREG. */
5737 if (preg->buffer != NULL)
5738 free (preg->buffer); /* __MEM_CHECKED__ */
5739 preg->buffer = NULL;
5741 preg->allocated = 0;
5744 if (preg->fastmap != NULL)
5745 free (preg->fastmap); /* __MEM_CHECKED__ */
5746 preg->fastmap = NULL;
5747 preg->fastmap_accurate = 0;
5749 if (preg->translate != NULL)
5750 free (preg->translate); /* __MEM_CHECKED__ */
5751 preg->translate = NULL;
5754 #endif /* not emacs */