1 /* Extended regular expression matching and search library,
3 (Implements POSIX draft P1003.2/D11.2, except for some of the
4 internationalization features.)
5 Copyright (C) 1993-1999, 2000 Free Software Foundation, Inc.
7 The GNU C Library is free software; you can redistribute it and/or
8 modify it under the terms of the GNU Library General Public License as
9 published by the Free Software Foundation; either version 2 of the
10 License, or (at your option) any later version.
12 The GNU C Library is distributed in the hope that it will be useful,
13 but WITHOUT ANY WARRANTY; without even the implied warranty of
14 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 Library General Public License for more details.
17 You should have received a copy of the GNU Library General Public
18 License along with the GNU C Library; see the file COPYING.LIB. If not,
19 write to the Free Software Foundation, Inc., 59 Temple Place - Suite 330,
20 Boston, MA 02111-1307, USA. */
22 /* AIX requires this to be the first thing in the file. */
23 #if defined _AIX && !defined REGEX_MALLOC
33 # define PARAMS(args) args
35 #if defined STDC_HEADERS && !defined emacs
38 /* We need this for `regex.h', and perhaps for the Emacs include files. */
39 # include <sys/types.h>
42 #define WIDE_CHAR_SUPPORT (HAVE_WCTYPE_H && HAVE_WCHAR_H && HAVE_BTOWC)
44 /* For platform which support the ISO C amendement 1 functionality we
45 support user defined character classes. */
46 #if defined _LIBC || WIDE_CHAR_SUPPORT
47 /* Solaris 2.5 has a bug: <wchar.h> must be included before <wctype.h>. */
53 /* We have to keep the namespace clean. */
54 # define regfree(preg) __regfree (preg)
55 # define regexec(pr, st, nm, pm, ef) __regexec (pr, st, nm, pm, ef)
56 # define regcomp(preg, pattern, cflags) __regcomp (preg, pattern, cflags)
57 # define regerror(errcode, preg, errbuf, errbuf_size) \
58 __regerror(errcode, preg, errbuf, errbuf_size)
59 # define re_set_registers(bu, re, nu, st, en) \
60 __re_set_registers (bu, re, nu, st, en)
61 # define re_match_2(bufp, string1, size1, string2, size2, pos, regs, stop) \
62 __re_match_2 (bufp, string1, size1, string2, size2, pos, regs, stop)
63 # define re_match(bufp, string, size, pos, regs) \
64 __re_match (bufp, string, size, pos, regs)
65 # define re_search(bufp, string, size, startpos, range, regs) \
66 __re_search (bufp, string, size, startpos, range, regs)
67 # define re_compile_pattern(pattern, length, bufp) \
68 __re_compile_pattern (pattern, length, bufp)
69 # define re_set_syntax(syntax) __re_set_syntax (syntax)
70 # define re_search_2(bufp, st1, s1, st2, s2, startpos, range, regs, stop) \
71 __re_search_2 (bufp, st1, s1, st2, s2, startpos, range, regs, stop)
72 # define re_compile_fastmap(bufp) __re_compile_fastmap (bufp)
74 # define btowc __btowc
76 /* We are also using some library internals. */
77 # include <locale/localeinfo.h>
78 # include <locale/elem-hash.h>
79 # include <langinfo.h>
82 /* This is for other GNU distributions with internationalized messages. */
83 #if HAVE_LIBINTL_H || defined _LIBC
86 # define gettext(msgid) (msgid)
90 /* This define is so xgettext can find the internationalizable
92 # define gettext_noop(String) String
95 /* The `emacs' switch turns on certain matching commands
96 that make sense only in Emacs. */
103 #else /* not emacs */
105 /* If we are not linking with Emacs proper,
106 we can't use the relocating allocator
107 even if config.h says that we can. */
112 /* This is the normal way of making sure we have a bcopy and a bzero.
113 This is used in most programs--a few other programs avoid this
114 by defining INHIBIT_STRING_HEADER. */
116 # define bzero(s, n) memset (s, 0, n)
118 /* Define the syntax stuff for \<, \>, etc. */
120 /* This must be nonzero for the wordchar and notwordchar pattern
121 commands in re_match_2. */
126 # ifdef SWITCH_ENUM_BUG
127 # define SWITCH_ENUM_CAST(x) ((int)(x))
129 # define SWITCH_ENUM_CAST(x) (x)
132 #endif /* not emacs */
134 #if defined _LIBC || HAVE_LIMITS_H
139 # define MB_LEN_MAX 1
142 /* Get the interface, including the syntax bits. */
145 /* isalpha etc. are used for the character classes. */
148 /* Jim Meyering writes:
150 "... Some ctype macros are valid only for character codes that
151 isascii says are ASCII (SGI's IRIX-4.0.5 is one such system --when
152 using /bin/cc or gcc but without giving an ansi option). So, all
153 ctype uses should be through macros like ISPRINT... If
154 STDC_HEADERS is defined, then autoconf has verified that the ctype
155 macros don't need to be guarded with references to isascii. ...
156 Defining isascii to 1 should let any compiler worth its salt
157 eliminate the && through constant folding."
158 Solaris defines some of these symbols so we must undefine them first. */
161 #if defined STDC_HEADERS || (!defined isascii && !defined HAVE_ISASCII)
162 # define ISASCII(c) 1
164 # define ISASCII(c) isascii(c)
168 # define ISBLANK(c) (ISASCII (c) && isblank (c))
170 # define ISBLANK(c) ((c) == ' ' || (c) == '\t')
173 # define ISGRAPH(c) (ISASCII (c) && isgraph (c))
175 # define ISGRAPH(c) (ISASCII (c) && isprint (c) && !isspace (c))
179 #define ISPRINT(c) (ISASCII (c) && isprint (c))
180 #define ISDIGIT(c) (ISASCII (c) && isdigit (c))
181 #define ISALNUM(c) (ISASCII (c) && isalnum (c))
182 #define ISALPHA(c) (ISASCII (c) && isalpha (c))
183 #define ISCNTRL(c) (ISASCII (c) && iscntrl (c))
184 #define ISLOWER(c) (ISASCII (c) && islower (c))
185 #define ISPUNCT(c) (ISASCII (c) && ispunct (c))
186 #define ISSPACE(c) (ISASCII (c) && isspace (c))
187 #define ISUPPER(c) (ISASCII (c) && isupper (c))
188 #define ISXDIGIT(c) (ISASCII (c) && isxdigit (c))
191 # define TOLOWER(c) _tolower(c)
193 # define TOLOWER(c) tolower(c)
197 # define NULL (void *)0
200 /* We remove any previous definition of `SIGN_EXTEND_CHAR',
201 since ours (we hope) works properly with all combinations of
202 machines, compilers, `char' and `unsigned char' argument types.
203 (Per Bothner suggested the basic approach.) */
204 #undef SIGN_EXTEND_CHAR
206 # define SIGN_EXTEND_CHAR(c) ((signed char) (c))
207 #else /* not __STDC__ */
208 /* As in Harbison and Steele. */
209 # define SIGN_EXTEND_CHAR(c) ((((unsigned char) (c)) ^ 128) - 128)
213 /* How many characters in the character set. */
214 # define CHAR_SET_SIZE 256
218 extern char *re_syntax_table;
220 # else /* not SYNTAX_TABLE */
222 static char re_syntax_table[CHAR_SET_SIZE];
232 bzero(re_syntax_table, sizeof(re_syntax_table));
234 for (c = 0; c < CHAR_SET_SIZE; ++c)
236 re_syntax_table[c] = Sword;
239 re_syntax_table[c] = Sword;
244 # endif /* not SYNTAX_TABLE */
246 # define SYNTAX(c) re_syntax_table[(unsigned char) (c)]
250 /* Should we use malloc or alloca? If REGEX_MALLOC is not defined, we
251 use `alloca' instead of `malloc'. This is because using malloc in
252 re_search* or re_match* could cause memory leaks when C-g is used in
253 Emacs; also, malloc is slower and causes storage fragmentation. On
254 the other hand, malloc is more portable, and easier to debug.
256 Because we sometimes use alloca, some routines have to be macros,
257 not functions -- `alloca'-allocated space disappears at the end of the
258 function it is called in. */
262 # define REGEX_ALLOCATE malloc
263 # define REGEX_REALLOCATE(source, osize, nsize) realloc (source, nsize)
264 # define REGEX_FREE free
266 #else /* not REGEX_MALLOC */
268 /* Emacs already defines alloca, sometimes. */
271 /* Make alloca work the best possible way. */
273 # define alloca __builtin_alloca
274 # else /* not __GNUC__ */
277 # endif /* HAVE_ALLOCA_H */
278 # endif /* not __GNUC__ */
280 # endif /* not alloca */
282 # define REGEX_ALLOCATE alloca
284 /* Assumes a `char *destination' variable. */
285 # define REGEX_REALLOCATE(source, osize, nsize) \
286 (destination = (char *) alloca (nsize), \
287 memcpy (destination, source, osize))
289 /* No need to do anything to free, after alloca. */
290 # define REGEX_FREE(arg) ((void)0) /* Do nothing! But inhibit gcc warning. */
292 #endif /* not REGEX_MALLOC */
294 /* Define how to allocate the failure stack. */
296 #if defined REL_ALLOC && defined REGEX_MALLOC
298 # define REGEX_ALLOCATE_STACK(size) \
299 r_alloc (&failure_stack_ptr, (size))
300 # define REGEX_REALLOCATE_STACK(source, osize, nsize) \
301 r_re_alloc (&failure_stack_ptr, (nsize))
302 # define REGEX_FREE_STACK(ptr) \
303 r_alloc_free (&failure_stack_ptr)
305 #else /* not using relocating allocator */
309 # define REGEX_ALLOCATE_STACK malloc
310 # define REGEX_REALLOCATE_STACK(source, osize, nsize) realloc (source, nsize)
311 # define REGEX_FREE_STACK free
313 # else /* not REGEX_MALLOC */
315 # define REGEX_ALLOCATE_STACK alloca
317 # define REGEX_REALLOCATE_STACK(source, osize, nsize) \
318 REGEX_REALLOCATE (source, osize, nsize)
319 /* No need to explicitly free anything. */
320 # define REGEX_FREE_STACK(arg)
322 # endif /* not REGEX_MALLOC */
323 #endif /* not using relocating allocator */
326 /* True if `size1' is non-NULL and PTR is pointing anywhere inside
327 `string1' or just past its end. This works if PTR is NULL, which is
329 #define FIRST_STRING_P(ptr) \
330 (size1 && string1 <= (ptr) && (ptr) <= string1 + size1)
332 /* (Re)Allocate N items of type T using malloc, or fail. */
333 #define TALLOC(n, t) ((t *) malloc ((n) * sizeof (t)))
334 #define RETALLOC(addr, n, t) ((addr) = (t *) realloc (addr, (n) * sizeof (t)))
335 #define RETALLOC_IF(addr, n, t) \
336 if (addr) RETALLOC((addr), (n), t); else (addr) = TALLOC ((n), t)
337 #define REGEX_TALLOC(n, t) ((t *) REGEX_ALLOCATE ((n) * sizeof (t)))
339 #define BYTEWIDTH 8 /* In bits. */
341 #define STREQ(s1, s2) ((strcmp (s1, s2) == 0))
345 #define MAX(a, b) ((a) > (b) ? (a) : (b))
346 #define MIN(a, b) ((a) < (b) ? (a) : (b))
348 //typedef char boolean;
352 static int re_match_2_internal (
353 struct re_pattern_buffer *bufp,
354 const char *string1, int size1,
355 const char *string2, int size2,
357 struct re_registers *regs,
361 /* These are the command codes that appear in compiled regular
362 expressions. Some opcodes are followed by argument bytes. A
363 command code can specify any interpretation whatsoever for its
364 arguments. Zero bytes may appear in the compiled regular expression. */
370 /* Succeed right away--no more backtracking. */
373 /* Followed by one byte giving n, then by n literal bytes. */
376 /* Matches any (more or less) character. */
379 /* Matches any one char belonging to specified set. First
380 following byte is number of bitmap bytes. Then come bytes
381 for a bitmap saying which chars are in. Bits in each byte
382 are ordered low-bit-first. A character is in the set if its
383 bit is 1. A character too large to have a bit in the map is
384 automatically not in the set. */
387 /* Same parameters as charset, but match any character that is
388 not one of those specified. */
391 /* Start remembering the text that is matched, for storing in a
392 register. Followed by one byte with the register number, in
393 the range 0 to one less than the pattern buffer's re_nsub
394 field. Then followed by one byte with the number of groups
395 inner to this one. (This last has to be part of the
396 start_memory only because we need it in the on_failure_jump
400 /* Stop remembering the text that is matched and store it in a
401 memory register. Followed by one byte with the register
402 number, in the range 0 to one less than `re_nsub' in the
403 pattern buffer, and one byte with the number of inner groups,
404 just like `start_memory'. (We need the number of inner
405 groups here because we don't have any easy way of finding the
406 corresponding start_memory when we're at a stop_memory.) */
409 /* Match a duplicate of something remembered. Followed by one
410 byte containing the register number. */
413 /* Fail unless at beginning of line. */
416 /* Fail unless at end of line. */
419 /* Succeeds if at beginning of buffer (if emacs) or at beginning
420 of string to be matched (if not). */
423 /* Analogously, for end of buffer/string. */
426 /* Followed by two byte relative address to which to jump. */
429 /* Same as jump, but marks the end of an alternative. */
432 /* Followed by two-byte relative address of place to resume at
433 in case of failure. */
436 /* Like on_failure_jump, but pushes a placeholder instead of the
437 current string position when executed. */
438 on_failure_keep_string_jump,
440 /* Throw away latest failure point and then jump to following
441 two-byte relative address. */
444 /* Change to pop_failure_jump if know won't have to backtrack to
445 match; otherwise change to jump. This is used to jump
446 back to the beginning of a repeat. If what follows this jump
447 clearly won't match what the repeat does, such that we can be
448 sure that there is no use backtracking out of repetitions
449 already matched, then we change it to a pop_failure_jump.
450 Followed by two-byte address. */
453 /* Jump to following two-byte address, and push a dummy failure
454 point. This failure point will be thrown away if an attempt
455 is made to use it for a failure. A `+' construct makes this
456 before the first repeat. Also used as an intermediary kind
457 of jump when compiling an alternative. */
460 /* Push a dummy failure point and continue. Used at the end of
464 /* Followed by two-byte relative address and two-byte number n.
465 After matching N times, jump to the address upon failure. */
468 /* Followed by two-byte relative address, and two-byte number n.
469 Jump to the address N times, then fail. */
472 /* Set the following two-byte relative address to the
473 subsequent two-byte number. The address *includes* the two
477 wordchar, /* Matches any word-constituent character. */
478 notwordchar, /* Matches any char that is not a word-constituent. */
480 wordbeg, /* Succeeds if at word beginning. */
481 wordend, /* Succeeds if at word end. */
483 wordbound, /* Succeeds if at a word boundary. */
484 notwordbound /* Succeeds if not at a word boundary. */
487 ,before_dot, /* Succeeds if before point. */
488 at_dot, /* Succeeds if at point. */
489 after_dot, /* Succeeds if after point. */
491 /* Matches any character whose syntax is specified. Followed by
492 a byte which contains a syntax code, e.g., Sword. */
495 /* Matches any character whose syntax is not that specified. */
500 /* Common operations on the compiled pattern. */
502 /* Store NUMBER in two contiguous bytes starting at DESTINATION. */
504 #define STORE_NUMBER(destination, number) \
506 (destination)[0] = (number) & 0377; \
507 (destination)[1] = (number) >> 8; \
510 /* Same as STORE_NUMBER, except increment DESTINATION to
511 the byte after where the number is stored. Therefore, DESTINATION
512 must be an lvalue. */
514 #define STORE_NUMBER_AND_INCR(destination, number) \
516 STORE_NUMBER (destination, number); \
517 (destination) += 2; \
520 /* Put into DESTINATION a number stored in two contiguous bytes starting
523 #define EXTRACT_NUMBER(destination, source) \
525 (destination) = *(source) & 0377; \
526 (destination) += SIGN_EXTEND_CHAR (*((source) + 1)) << 8; \
530 static void extract_number _RE_ARGS ((int *dest, unsigned char *source));
534 unsigned char *source)
536 int temp = SIGN_EXTEND_CHAR (*(source + 1));
537 *dest = *source & 0377;
541 # ifndef EXTRACT_MACROS /* To debug the macros. */
542 # undef EXTRACT_NUMBER
543 # define EXTRACT_NUMBER(dest, src) extract_number (&dest, src)
544 # endif /* not EXTRACT_MACROS */
546 #endif /* REGEX_DEBUG */
548 /* Same as EXTRACT_NUMBER, except increment SOURCE to after the number.
549 SOURCE must be an lvalue. */
551 #define EXTRACT_NUMBER_AND_INCR(destination, source) \
553 EXTRACT_NUMBER (destination, source); \
559 extract_number_and_incr (
561 unsigned char **source)
563 extract_number (destination, *source);
567 # ifndef EXTRACT_MACROS
568 # undef EXTRACT_NUMBER_AND_INCR
569 # define EXTRACT_NUMBER_AND_INCR(dest, src) \
570 extract_number_and_incr (&dest, &src)
571 # endif /* not EXTRACT_MACROS */
573 #endif /* REGEX_DEBUG */
575 /* If DEBUG is defined, Regex prints many voluminous messages about what
576 it is doing (if the variable `debug' is nonzero). If linked with the
577 main program in `iregex.c', you can enter patterns and strings
578 interactively. And if linked with the main program in `main.c' and
579 the other test files, you can run the already-written tests. */
583 /* We use standard I/O for debugging. */
586 /* It is useful to test things that ``must'' be true when debugging. */
591 # define DEBUG_STATEMENT(e) e
592 # define DEBUG_PRINT1(x) if (debug) printf (x)
593 # define DEBUG_PRINT2(x1, x2) if (debug) printf (x1, x2)
594 # define DEBUG_PRINT3(x1, x2, x3) if (debug) printf (x1, x2, x3)
595 # define DEBUG_PRINT4(x1, x2, x3, x4) if (debug) printf (x1, x2, x3, x4)
596 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) \
597 if (debug) print_partial_compiled_pattern (s, e)
598 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) \
599 if (debug) print_double_string (w, s1, sz1, s2, sz2)
602 /* Print the fastmap in human-readable form. */
608 unsigned was_a_range = 0;
611 while (i < (1 << BYTEWIDTH))
617 while (i < (1 << BYTEWIDTH) && fastmap[i])
633 /* Print a compiled pattern string in human-readable form, starting at
634 the START pointer into it and ending just before the pointer END. */
637 print_partial_compiled_pattern (start, end)
638 unsigned char *start;
643 unsigned char *p = start;
644 unsigned char *pend = end;
652 /* Loop over pattern commands. */
656 printf ("%t:\t", p - start);
658 printf ("%ld:\t", (long int) (p - start));
661 switch ((re_opcode_t) *p++)
669 printf ("/exactn/%d", mcnt);
680 printf ("/start_memory/%d/%d", mcnt, *p++);
685 printf ("/stop_memory/%d/%d", mcnt, *p++);
689 printf ("/duplicate/%d", *p++);
699 register int c, last = -100;
700 register int in_range = 0;
702 printf ("/charset [%s",
703 (re_opcode_t) *(p - 1) == charset_not ? "^" : "");
705 assert (p + *p < pend);
707 for (c = 0; c < 256; c++)
709 && (p[1 + (c/8)] & (1 << (c % 8))))
711 /* Are we starting a range? */
712 if (last + 1 == c && ! in_range)
717 /* Have we broken a range? */
718 else if (last + 1 != c && in_range)
747 case on_failure_jump:
748 extract_number_and_incr (&mcnt, &p);
750 printf ("/on_failure_jump to %t", p + mcnt - start);
752 printf ("/on_failure_jump to %ld", (long int) (p + mcnt - start));
756 case on_failure_keep_string_jump:
757 extract_number_and_incr (&mcnt, &p);
759 printf ("/on_failure_keep_string_jump to %t", p + mcnt - start);
761 printf ("/on_failure_keep_string_jump to %ld",
762 (long int) (p + mcnt - start));
766 case dummy_failure_jump:
767 extract_number_and_incr (&mcnt, &p);
769 printf ("/dummy_failure_jump to %t", p + mcnt - start);
771 printf ("/dummy_failure_jump to %ld", (long int) (p + mcnt - start));
775 case push_dummy_failure:
776 printf ("/push_dummy_failure");
780 extract_number_and_incr (&mcnt, &p);
782 printf ("/maybe_pop_jump to %t", p + mcnt - start);
784 printf ("/maybe_pop_jump to %ld", (long int) (p + mcnt - start));
788 case pop_failure_jump:
789 extract_number_and_incr (&mcnt, &p);
791 printf ("/pop_failure_jump to %t", p + mcnt - start);
793 printf ("/pop_failure_jump to %ld", (long int) (p + mcnt - start));
798 extract_number_and_incr (&mcnt, &p);
800 printf ("/jump_past_alt to %t", p + mcnt - start);
802 printf ("/jump_past_alt to %ld", (long int) (p + mcnt - start));
807 extract_number_and_incr (&mcnt, &p);
809 printf ("/jump to %t", p + mcnt - start);
811 printf ("/jump to %ld", (long int) (p + mcnt - start));
816 extract_number_and_incr (&mcnt, &p);
818 extract_number_and_incr (&mcnt2, &p);
820 printf ("/succeed_n to %t, %d times", p1 - start, mcnt2);
822 printf ("/succeed_n to %ld, %d times",
823 (long int) (p1 - start), mcnt2);
828 extract_number_and_incr (&mcnt, &p);
830 extract_number_and_incr (&mcnt2, &p);
831 printf ("/jump_n to %d, %d times", p1 - start, mcnt2);
835 extract_number_and_incr (&mcnt, &p);
837 extract_number_and_incr (&mcnt2, &p);
839 printf ("/set_number_at location %t to %d", p1 - start, mcnt2);
841 printf ("/set_number_at location %ld to %d",
842 (long int) (p1 - start), mcnt2);
847 printf ("/wordbound");
851 printf ("/notwordbound");
863 printf ("/before_dot");
871 printf ("/after_dot");
875 printf ("/syntaxspec");
877 printf ("/%d", mcnt);
881 printf ("/notsyntaxspec");
883 printf ("/%d", mcnt);
888 printf ("/wordchar");
892 printf ("/notwordchar");
904 printf ("?%d", *(p-1));
911 printf ("%t:\tend of pattern.\n", p - start);
913 printf ("%ld:\tend of pattern.\n", (long int) (p - start));
919 print_compiled_pattern (bufp)
920 struct re_pattern_buffer *bufp;
922 unsigned char *buffer = bufp->buffer;
924 print_partial_compiled_pattern (buffer, buffer + bufp->used);
925 printf ("%ld bytes used/%ld bytes allocated.\n",
926 bufp->used, bufp->allocated);
928 if (bufp->fastmap_accurate && bufp->fastmap)
930 printf ("fastmap: ");
931 print_fastmap (bufp->fastmap);
935 printf ("re_nsub: %Zd\t", bufp->re_nsub);
937 printf ("re_nsub: %ld\t", (long int) bufp->re_nsub);
939 printf ("regs_alloc: %d\t", bufp->regs_allocated);
940 printf ("can_be_null: %d\t", bufp->can_be_null);
941 printf ("newline_anchor: %d\n", bufp->newline_anchor);
942 printf ("no_sub: %d\t", bufp->no_sub);
943 printf ("not_bol: %d\t", bufp->not_bol);
944 printf ("not_eol: %d\t", bufp->not_eol);
945 printf ("syntax: %lx\n", bufp->syntax);
946 /* Perhaps we should print the translate table? */
951 print_double_string (where, string1, size1, string2, size2)
964 if (FIRST_STRING_P (where))
966 for (this_char = where - string1; this_char < size1; this_char++)
967 putchar (string1[this_char]);
972 for (this_char = where - string2; this_char < size2; this_char++)
973 putchar (string2[this_char]);
984 #else /* not REGEX_DEBUG */
989 # define DEBUG_STATEMENT(e)
990 # define DEBUG_PRINT1(x)
991 # define DEBUG_PRINT2(x1, x2)
992 # define DEBUG_PRINT3(x1, x2, x3)
993 # define DEBUG_PRINT4(x1, x2, x3, x4)
994 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e)
995 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2)
997 #endif /* not DEBUG */
999 /* Set by `re_set_syntax' to the current regexp syntax to recognize. Can
1000 also be assigned to arbitrarily: each pattern buffer stores its own
1001 syntax, so it can be changed between regex compilations. */
1002 /* This has no initializer because initialized variables in Emacs
1003 become read-only after dumping. */
1004 reg_syntax_t re_syntax_options;
1007 /* Specify the precise syntax of regexps for compilation. This provides
1008 for compatibility for various utilities which historically have
1009 different, incompatible syntaxes.
1011 The argument SYNTAX is a bit mask comprised of the various bits
1012 defined in regex.h. We return the old syntax. */
1015 re_set_syntax (reg_syntax_t syntax)
1017 reg_syntax_t ret = re_syntax_options;
1019 re_syntax_options = syntax;
1021 if (syntax & RE_DEBUG)
1023 else if (debug) /* was on but now is not */
1029 weak_alias (__re_set_syntax, re_set_syntax)
1032 /* This table gives an error message for each of the error codes listed
1033 in regex.h. Obviously the order here has to be same as there.
1034 POSIX doesn't require that we do anything for REG_NOERROR,
1035 but why not be nice? */
1037 static const char re_error_msgid[] =
1039 #define REG_NOERROR_IDX 0
1040 _("Success") /* REG_NOERROR */
1042 #define REG_NOMATCH_IDX (REG_NOERROR_IDX + sizeof "Success")
1043 _("No match") /* REG_NOMATCH */
1045 #define REG_BADPAT_IDX (REG_NOMATCH_IDX + sizeof "No match")
1046 _("Invalid regular expression") /* REG_BADPAT */
1048 #define REG_ECOLLATE_IDX (REG_BADPAT_IDX + sizeof "Invalid regular expression")
1049 _("Invalid collation character") /* REG_ECOLLATE */
1051 #define REG_ECTYPE_IDX (REG_ECOLLATE_IDX + sizeof "Invalid collation character")
1052 _("Invalid character class name") /* REG_ECTYPE */
1054 #define REG_EESCAPE_IDX (REG_ECTYPE_IDX + sizeof "Invalid character class name")
1055 _("Trailing backslash") /* REG_EESCAPE */
1057 #define REG_ESUBREG_IDX (REG_EESCAPE_IDX + sizeof "Trailing backslash")
1058 _("Invalid back reference") /* REG_ESUBREG */
1060 #define REG_EBRACK_IDX (REG_ESUBREG_IDX + sizeof "Invalid back reference")
1061 _("Unmatched [ or [^") /* REG_EBRACK */
1063 #define REG_EPAREN_IDX (REG_EBRACK_IDX + sizeof "Unmatched [ or [^")
1064 _("Unmatched ( or \\(") /* REG_EPAREN */
1066 #define REG_EBRACE_IDX (REG_EPAREN_IDX + sizeof "Unmatched ( or \\(")
1067 _("Unmatched \\{") /* REG_EBRACE */
1069 #define REG_BADBR_IDX (REG_EBRACE_IDX + sizeof "Unmatched \\{")
1070 _("Invalid content of \\{\\}") /* REG_BADBR */
1072 #define REG_ERANGE_IDX (REG_BADBR_IDX + sizeof "Invalid content of \\{\\}")
1073 _("Invalid range end") /* REG_ERANGE */
1075 #define REG_ESPACE_IDX (REG_ERANGE_IDX + sizeof "Invalid range end")
1076 _("Memory exhausted") /* REG_ESPACE */
1078 #define REG_BADRPT_IDX (REG_ESPACE_IDX + sizeof "Memory exhausted")
1079 _("Invalid preceding regular expression") /* REG_BADRPT */
1081 #define REG_EEND_IDX (REG_BADRPT_IDX + sizeof "Invalid preceding regular expression")
1082 _("Premature end of regular expression") /* REG_EEND */
1084 #define REG_ESIZE_IDX (REG_EEND_IDX + sizeof "Premature end of regular expression")
1085 _("Regular expression too big") /* REG_ESIZE */
1087 #define REG_ERPAREN_IDX (REG_ESIZE_IDX + sizeof "Regular expression too big")
1088 _("Unmatched ) or \\)") /* REG_ERPAREN */
1091 static const size_t re_error_msgid_idx[] =
1112 /* Avoiding alloca during matching, to placate r_alloc. */
1114 /* Define MATCH_MAY_ALLOCATE unless we need to make sure that the
1115 searching and matching functions should not call alloca. On some
1116 systems, alloca is implemented in terms of malloc, and if we're
1117 using the relocating allocator routines, then malloc could cause a
1118 relocation, which might (if the strings being searched are in the
1119 ralloc heap) shift the data out from underneath the regexp
1122 Here's another reason to avoid allocation: Emacs
1123 processes input from X in a signal handler; processing X input may
1124 call malloc; if input arrives while a matching routine is calling
1125 malloc, then we're scrod. But Emacs can't just block input while
1126 calling matching routines; then we don't notice interrupts when
1127 they come in. So, Emacs blocks input around all regexp calls
1128 except the matching calls, which it leaves unprotected, in the
1129 faith that they will not malloc. */
1131 /* Normally, this is fine. */
1132 #define MATCH_MAY_ALLOCATE
1134 /* When using GNU C, we are not REALLY using the C alloca, no matter
1135 what config.h may say. So don't take precautions for it. */
1140 /* The match routines may not allocate if (1) they would do it with malloc
1141 and (2) it's not safe for them to use malloc.
1142 Note that if REL_ALLOC is defined, matching would not use malloc for the
1143 failure stack, but we would still use it for the register vectors;
1144 so REL_ALLOC should not affect this. */
1145 #if (defined C_ALLOCA || defined REGEX_MALLOC) && defined emacs
1146 # undef MATCH_MAY_ALLOCATE
1150 /* Failure stack declarations and macros; both re_compile_fastmap and
1151 re_match_2 use a failure stack. These have to be macros because of
1152 REGEX_ALLOCATE_STACK. */
1155 /* Number of failure points for which to initially allocate space
1156 when matching. If this number is exceeded, we allocate more
1157 space, so it is not a hard limit. */
1158 #ifndef INIT_FAILURE_ALLOC
1159 # define INIT_FAILURE_ALLOC 5
1162 /* Roughly the maximum number of failure points on the stack. Would be
1163 exactly that if always used MAX_FAILURE_ITEMS items each time we failed.
1164 This is a variable only so users of regex can assign to it; we never
1165 change it ourselves. */
1169 # if defined MATCH_MAY_ALLOCATE
1170 /* 4400 was enough to cause a crash on Alpha OSF/1,
1171 whose default stack limit is 2mb. */
1172 long int re_max_failures = 4000;
1174 long int re_max_failures = 2000;
1177 union fail_stack_elt
1179 unsigned char *pointer;
1183 typedef union fail_stack_elt fail_stack_elt_t;
1187 fail_stack_elt_t *stack;
1188 unsigned long int size;
1189 unsigned long int avail; /* Offset of next open position. */
1192 #else /* not INT_IS_16BIT */
1194 # if defined MATCH_MAY_ALLOCATE
1195 /* 4400 was enough to cause a crash on Alpha OSF/1,
1196 whose default stack limit is 2mb. */
1197 int re_max_failures = 20000;
1199 int re_max_failures = 2000;
1202 union fail_stack_elt
1204 unsigned char *pointer;
1208 typedef union fail_stack_elt fail_stack_elt_t;
1212 fail_stack_elt_t *stack;
1214 unsigned avail; /* Offset of next open position. */
1217 #endif /* INT_IS_16BIT */
1219 #define FAIL_STACK_EMPTY() (fail_stack.avail == 0)
1220 #define FAIL_STACK_PTR_EMPTY() (fail_stack_ptr->avail == 0)
1221 #define FAIL_STACK_FULL() (fail_stack.avail == fail_stack.size)
1224 /* Define macros to initialize and free the failure stack.
1225 Do `return -2' if the alloc fails. */
1227 #ifdef MATCH_MAY_ALLOCATE
1228 # define INIT_FAIL_STACK() \
1230 fail_stack.stack = (fail_stack_elt_t *) \
1231 REGEX_ALLOCATE_STACK (INIT_FAILURE_ALLOC * sizeof (fail_stack_elt_t)); \
1233 if (fail_stack.stack == NULL) \
1236 fail_stack.size = INIT_FAILURE_ALLOC; \
1237 fail_stack.avail = 0; \
1240 # define RESET_FAIL_STACK() REGEX_FREE_STACK (fail_stack.stack)
1242 # define INIT_FAIL_STACK() \
1244 fail_stack.avail = 0; \
1247 # define RESET_FAIL_STACK()
1251 /* Double the size of FAIL_STACK, up to approximately `re_max_failures' items.
1253 Return 1 if succeeds, and 0 if either ran out of memory
1254 allocating space for it or it was already too large.
1256 REGEX_REALLOCATE_STACK requires `destination' be declared. */
1258 #define DOUBLE_FAIL_STACK(fail_stack) \
1259 ((fail_stack).size > (unsigned) (re_max_failures * MAX_FAILURE_ITEMS) \
1261 : ((fail_stack).stack = (fail_stack_elt_t *) \
1262 REGEX_REALLOCATE_STACK ((fail_stack).stack, \
1263 (fail_stack).size * sizeof (fail_stack_elt_t), \
1264 ((fail_stack).size << 1) * sizeof (fail_stack_elt_t)), \
1266 (fail_stack).stack == NULL \
1268 : ((fail_stack).size <<= 1, \
1272 /* Push pointer POINTER on FAIL_STACK.
1273 Return 1 if was able to do so and 0 if ran out of memory allocating
1275 #define PUSH_PATTERN_OP(POINTER, FAIL_STACK) \
1276 ((FAIL_STACK_FULL () \
1277 && !DOUBLE_FAIL_STACK (FAIL_STACK)) \
1279 : ((FAIL_STACK).stack[(FAIL_STACK).avail++].pointer = POINTER, \
1282 /* Push a pointer value onto the failure stack.
1283 Assumes the variable `fail_stack'. Probably should only
1284 be called from within `PUSH_FAILURE_POINT'. */
1285 #define PUSH_FAILURE_POINTER(item) \
1286 fail_stack.stack[fail_stack.avail++].pointer = (unsigned char *) (item)
1288 /* This pushes an integer-valued item onto the failure stack.
1289 Assumes the variable `fail_stack'. Probably should only
1290 be called from within `PUSH_FAILURE_POINT'. */
1291 #define PUSH_FAILURE_INT(item) \
1292 fail_stack.stack[fail_stack.avail++].integer = (item)
1294 /* Push a fail_stack_elt_t value onto the failure stack.
1295 Assumes the variable `fail_stack'. Probably should only
1296 be called from within `PUSH_FAILURE_POINT'. */
1297 #define PUSH_FAILURE_ELT(item) \
1298 fail_stack.stack[fail_stack.avail++] = (item)
1300 /* These three POP... operations complement the three PUSH... operations.
1301 All assume that `fail_stack' is nonempty. */
1302 #define POP_FAILURE_POINTER() fail_stack.stack[--fail_stack.avail].pointer
1303 #define POP_FAILURE_INT() fail_stack.stack[--fail_stack.avail].integer
1304 #define POP_FAILURE_ELT() fail_stack.stack[--fail_stack.avail]
1306 /* Used to omit pushing failure point id's when we're not debugging. */
1308 # define DEBUG_PUSH PUSH_FAILURE_INT
1309 # define DEBUG_POP(item_addr) *(item_addr) = POP_FAILURE_INT ()
1311 # define DEBUG_PUSH(item)
1312 # define DEBUG_POP(item_addr)
1316 /* Push the information about the state we will need
1317 if we ever fail back to it.
1319 Requires variables fail_stack, regstart, regend, reg_info, and
1320 num_regs_pushed be declared. DOUBLE_FAIL_STACK requires `destination'
1323 Does `return FAILURE_CODE' if runs out of memory. */
1325 #define PUSH_FAILURE_POINT(pattern_place, string_place, failure_code) \
1327 char *destination; \
1328 /* Must be int, so when we don't save any registers, the arithmetic \
1329 of 0 + -1 isn't done as unsigned. */ \
1330 /* Can't be int, since there is not a shred of a guarantee that int \
1331 is wide enough to hold a value of something to which pointer can \
1333 active_reg_t this_reg; \
1335 DEBUG_STATEMENT (failure_id++); \
1336 DEBUG_STATEMENT (nfailure_points_pushed++); \
1337 DEBUG_PRINT2 ("\nPUSH_FAILURE_POINT #%u:\n", failure_id); \
1338 DEBUG_PRINT2 (" Before push, next avail: %d\n", (fail_stack).avail);\
1339 DEBUG_PRINT2 (" size: %d\n", (fail_stack).size);\
1341 DEBUG_PRINT2 (" slots needed: %ld\n", NUM_FAILURE_ITEMS); \
1342 DEBUG_PRINT2 (" available: %d\n", REMAINING_AVAIL_SLOTS); \
1344 /* Ensure we have enough space allocated for what we will push. */ \
1345 while (REMAINING_AVAIL_SLOTS < NUM_FAILURE_ITEMS) \
1347 if (!DOUBLE_FAIL_STACK (fail_stack)) \
1348 return failure_code; \
1350 DEBUG_PRINT2 ("\n Doubled stack; size now: %d\n", \
1351 (fail_stack).size); \
1352 DEBUG_PRINT2 (" slots available: %d\n", REMAINING_AVAIL_SLOTS);\
1355 /* Push the info, starting with the registers. */ \
1356 DEBUG_PRINT1 ("\n"); \
1359 for (this_reg = lowest_active_reg; this_reg <= highest_active_reg; \
1362 DEBUG_PRINT2 (" Pushing reg: %lu\n", this_reg); \
1363 DEBUG_STATEMENT (num_regs_pushed++); \
1365 DEBUG_PRINT2 (" start: %p\n", regstart[this_reg]); \
1366 PUSH_FAILURE_POINTER (regstart[this_reg]); \
1368 DEBUG_PRINT2 (" end: %p\n", regend[this_reg]); \
1369 PUSH_FAILURE_POINTER (regend[this_reg]); \
1371 DEBUG_PRINT2 (" info: %p\n ", \
1372 reg_info[this_reg].word.pointer); \
1373 DEBUG_PRINT2 (" match_null=%d", \
1374 REG_MATCH_NULL_STRING_P (reg_info[this_reg])); \
1375 DEBUG_PRINT2 (" active=%d", IS_ACTIVE (reg_info[this_reg])); \
1376 DEBUG_PRINT2 (" matched_something=%d", \
1377 MATCHED_SOMETHING (reg_info[this_reg])); \
1378 DEBUG_PRINT2 (" ever_matched=%d", \
1379 EVER_MATCHED_SOMETHING (reg_info[this_reg])); \
1380 DEBUG_PRINT1 ("\n"); \
1381 PUSH_FAILURE_ELT (reg_info[this_reg].word); \
1384 DEBUG_PRINT2 (" Pushing low active reg: %ld\n", lowest_active_reg);\
1385 PUSH_FAILURE_INT (lowest_active_reg); \
1387 DEBUG_PRINT2 (" Pushing high active reg: %ld\n", highest_active_reg);\
1388 PUSH_FAILURE_INT (highest_active_reg); \
1390 DEBUG_PRINT2 (" Pushing pattern %p:\n", pattern_place); \
1391 DEBUG_PRINT_COMPILED_PATTERN (bufp, pattern_place, pend); \
1392 PUSH_FAILURE_POINTER (pattern_place); \
1394 DEBUG_PRINT2 (" Pushing string %p: `", string_place); \
1395 DEBUG_PRINT_DOUBLE_STRING (string_place, string1, size1, string2, \
1397 DEBUG_PRINT1 ("'\n"); \
1398 PUSH_FAILURE_POINTER (string_place); \
1400 DEBUG_PRINT2 (" Pushing failure id: %u\n", failure_id); \
1401 DEBUG_PUSH (failure_id); \
1404 /* This is the number of items that are pushed and popped on the stack
1405 for each register. */
1406 #define NUM_REG_ITEMS 3
1408 /* Individual items aside from the registers. */
1410 # define NUM_NONREG_ITEMS 5 /* Includes failure point id. */
1412 # define NUM_NONREG_ITEMS 4
1415 /* We push at most this many items on the stack. */
1416 /* We used to use (num_regs - 1), which is the number of registers
1417 this regexp will save; but that was changed to 5
1418 to avoid stack overflow for a regexp with lots of parens. */
1419 #define MAX_FAILURE_ITEMS (5 * NUM_REG_ITEMS + NUM_NONREG_ITEMS)
1421 /* We actually push this many items. */
1422 #define NUM_FAILURE_ITEMS \
1424 ? 0 : highest_active_reg - lowest_active_reg + 1) \
1428 /* How many items can still be added to the stack without overflowing it. */
1429 #define REMAINING_AVAIL_SLOTS ((fail_stack).size - (fail_stack).avail)
1432 /* Pops what PUSH_FAIL_STACK pushes.
1434 We restore into the parameters, all of which should be lvalues:
1435 STR -- the saved data position.
1436 PAT -- the saved pattern position.
1437 LOW_REG, HIGH_REG -- the highest and lowest active registers.
1438 REGSTART, REGEND -- arrays of string positions.
1439 REG_INFO -- array of information about each subexpression.
1441 Also assumes the variables `fail_stack' and (if debugging), `bufp',
1442 `pend', `string1', `size1', `string2', and `size2'. */
1444 #define POP_FAILURE_POINT(str, pat, low_reg, high_reg, regstart, regend, reg_info)\
1446 DEBUG_STATEMENT (unsigned failure_id;) \
1447 active_reg_t this_reg; \
1448 const unsigned char *string_temp; \
1450 assert (!FAIL_STACK_EMPTY ()); \
1452 /* Remove failure points and point to how many regs pushed. */ \
1453 DEBUG_PRINT1 ("POP_FAILURE_POINT:\n"); \
1454 DEBUG_PRINT2 (" Before pop, next avail: %d\n", fail_stack.avail); \
1455 DEBUG_PRINT2 (" size: %d\n", fail_stack.size); \
1457 assert (fail_stack.avail >= NUM_NONREG_ITEMS); \
1459 DEBUG_POP (&failure_id); \
1460 DEBUG_PRINT2 (" Popping failure id: %u\n", failure_id); \
1462 /* If the saved string location is NULL, it came from an \
1463 on_failure_keep_string_jump opcode, and we want to throw away the \
1464 saved NULL, thus retaining our current position in the string. */ \
1465 string_temp = POP_FAILURE_POINTER (); \
1466 if (string_temp != NULL) \
1467 str = (const char *) string_temp; \
1469 DEBUG_PRINT2 (" Popping string %p: `", str); \
1470 DEBUG_PRINT_DOUBLE_STRING (str, string1, size1, string2, size2); \
1471 DEBUG_PRINT1 ("'\n"); \
1473 pat = (unsigned char *) POP_FAILURE_POINTER (); \
1474 DEBUG_PRINT2 (" Popping pattern %p:\n", pat); \
1475 DEBUG_PRINT_COMPILED_PATTERN (bufp, pat, pend); \
1477 /* Restore register info. */ \
1478 high_reg = (active_reg_t) POP_FAILURE_INT (); \
1479 DEBUG_PRINT2 (" Popping high active reg: %ld\n", high_reg); \
1481 low_reg = (active_reg_t) POP_FAILURE_INT (); \
1482 DEBUG_PRINT2 (" Popping low active reg: %ld\n", low_reg); \
1485 for (this_reg = high_reg; this_reg >= low_reg; this_reg--) \
1487 DEBUG_PRINT2 (" Popping reg: %ld\n", this_reg); \
1489 reg_info[this_reg].word = POP_FAILURE_ELT (); \
1490 DEBUG_PRINT2 (" info: %p\n", \
1491 reg_info[this_reg].word.pointer); \
1493 regend[this_reg] = (const char *) POP_FAILURE_POINTER (); \
1494 DEBUG_PRINT2 (" end: %p\n", regend[this_reg]); \
1496 regstart[this_reg] = (const char *) POP_FAILURE_POINTER (); \
1497 DEBUG_PRINT2 (" start: %p\n", regstart[this_reg]); \
1501 for (this_reg = highest_active_reg; this_reg > high_reg; this_reg--) \
1503 reg_info[this_reg].word.integer = 0; \
1504 regend[this_reg] = 0; \
1505 regstart[this_reg] = 0; \
1507 highest_active_reg = high_reg; \
1510 set_regs_matched_done = 0; \
1511 DEBUG_STATEMENT (nfailure_points_popped++); \
1512 } /* POP_FAILURE_POINT */
1516 /* Structure for per-register (a.k.a. per-group) information.
1517 Other register information, such as the
1518 starting and ending positions (which are addresses), and the list of
1519 inner groups (which is a bits list) are maintained in separate
1522 We are making a (strictly speaking) nonportable assumption here: that
1523 the compiler will pack our bit fields into something that fits into
1524 the type of `word', i.e., is something that fits into one item on the
1528 /* Declarations and macros for re_match_2. */
1532 fail_stack_elt_t word;
1535 /* This field is one if this group can match the empty string,
1536 zero if not. If not yet determined, `MATCH_NULL_UNSET_VALUE'. */
1537 #define MATCH_NULL_UNSET_VALUE 3
1538 unsigned match_null_string_p : 2;
1539 unsigned is_active : 1;
1540 unsigned matched_something : 1;
1541 unsigned ever_matched_something : 1;
1543 } register_info_type;
1545 #define REG_MATCH_NULL_STRING_P(R) ((R).bits.match_null_string_p)
1546 #define IS_ACTIVE(R) ((R).bits.is_active)
1547 #define MATCHED_SOMETHING(R) ((R).bits.matched_something)
1548 #define EVER_MATCHED_SOMETHING(R) ((R).bits.ever_matched_something)
1551 /* Call this when have matched a real character; it sets `matched' flags
1552 for the subexpressions which we are currently inside. Also records
1553 that those subexprs have matched. */
1554 #define SET_REGS_MATCHED() \
1557 if (!set_regs_matched_done) \
1560 set_regs_matched_done = 1; \
1561 for (r = lowest_active_reg; r <= highest_active_reg; r++) \
1563 MATCHED_SOMETHING (reg_info[r]) \
1564 = EVER_MATCHED_SOMETHING (reg_info[r]) \
1571 /* Registers are set to a sentinel when they haven't yet matched. */
1572 static char reg_unset_dummy;
1573 #define REG_UNSET_VALUE (®_unset_dummy)
1574 #define REG_UNSET(e) ((e) == REG_UNSET_VALUE)
1576 /* Subroutine declarations and macros for regex_compile. */
1578 static reg_errcode_t regex_compile _RE_ARGS ((const char *pattern, size_t size,
1579 reg_syntax_t syntax,
1580 struct re_pattern_buffer *bufp));
1581 static void store_op1 _RE_ARGS ((re_opcode_t op, unsigned char *loc, int arg));
1582 static void store_op2 _RE_ARGS ((re_opcode_t op, unsigned char *loc,
1583 int arg1, int arg2));
1584 static void insert_op1 _RE_ARGS ((re_opcode_t op, unsigned char *loc,
1585 int arg, unsigned char *end));
1586 static void insert_op2 _RE_ARGS ((re_opcode_t op, unsigned char *loc,
1587 int arg1, int arg2, unsigned char *end));
1588 static boolean at_begline_loc_p _RE_ARGS ((const char *pattern, const char *p,
1589 reg_syntax_t syntax));
1590 static boolean at_endline_loc_p _RE_ARGS ((const char *p, const char *pend,
1591 reg_syntax_t syntax));
1592 static reg_errcode_t compile_range _RE_ARGS ((unsigned int range_start,
1596 reg_syntax_t syntax,
1599 /* Fetch the next character in the uncompiled pattern---translating it
1600 if necessary. Also cast from a signed character in the constant
1601 string passed to us by the user to an unsigned char that we can use
1602 as an array index (in, e.g., `translate'). */
1604 # define PATFETCH(c) \
1605 do {if (p == pend) return REG_EEND; \
1606 c = (unsigned char) *p++; \
1607 if (translate) c = (unsigned char) translate[c]; \
1611 /* Fetch the next character in the uncompiled pattern, with no
1613 #define PATFETCH_RAW(c) \
1614 do {if (p == pend) return REG_EEND; \
1615 c = (unsigned char) *p++; \
1618 /* Go backwards one character in the pattern. */
1619 #define PATUNFETCH p--
1622 /* If `translate' is non-null, return translate[D], else just D. We
1623 cast the subscript to translate because some data is declared as
1624 `char *', to avoid warnings when a string constant is passed. But
1625 when we use a character as a subscript we must make it unsigned. */
1627 # define TRANSLATE(d) \
1628 (translate ? (char) translate[(unsigned char) (d)] : (d))
1632 /* Macros for outputting the compiled pattern into `buffer'. */
1634 /* If the buffer isn't allocated when it comes in, use this. */
1635 #define INIT_BUF_SIZE 32
1637 /* Make sure we have at least N more bytes of space in buffer. */
1638 #define GET_BUFFER_SPACE(n) \
1639 while ((unsigned long) (b - bufp->buffer + (n)) > bufp->allocated) \
1642 /* Make sure we have one more byte of buffer space and then add C to it. */
1643 #define BUF_PUSH(c) \
1645 GET_BUFFER_SPACE (1); \
1646 *b++ = (unsigned char) (c); \
1650 /* Ensure we have two more bytes of buffer space and then append C1 and C2. */
1651 #define BUF_PUSH_2(c1, c2) \
1653 GET_BUFFER_SPACE (2); \
1654 *b++ = (unsigned char) (c1); \
1655 *b++ = (unsigned char) (c2); \
1659 /* As with BUF_PUSH_2, except for three bytes. */
1660 #define BUF_PUSH_3(c1, c2, c3) \
1662 GET_BUFFER_SPACE (3); \
1663 *b++ = (unsigned char) (c1); \
1664 *b++ = (unsigned char) (c2); \
1665 *b++ = (unsigned char) (c3); \
1669 /* Store a jump with opcode OP at LOC to location TO. We store a
1670 relative address offset by the three bytes the jump itself occupies. */
1671 #define STORE_JUMP(op, loc, to) \
1672 store_op1 (op, loc, (int) ((to) - (loc) - 3))
1674 /* Likewise, for a two-argument jump. */
1675 #define STORE_JUMP2(op, loc, to, arg) \
1676 store_op2 (op, loc, (int) ((to) - (loc) - 3), arg)
1678 /* Like `STORE_JUMP', but for inserting. Assume `b' is the buffer end. */
1679 #define INSERT_JUMP(op, loc, to) \
1680 insert_op1 (op, loc, (int) ((to) - (loc) - 3), b)
1682 /* Like `STORE_JUMP2', but for inserting. Assume `b' is the buffer end. */
1683 #define INSERT_JUMP2(op, loc, to, arg) \
1684 insert_op2 (op, loc, (int) ((to) - (loc) - 3), arg, b)
1687 /* This is not an arbitrary limit: the arguments which represent offsets
1688 into the pattern are two bytes long. So if 2^16 bytes turns out to
1689 be too small, many things would have to change. */
1690 /* Any other compiler which, like MSC, has allocation limit below 2^16
1691 bytes will have to use approach similar to what was done below for
1692 MSC and drop MAX_BUF_SIZE a bit. Otherwise you may end up
1693 reallocating to 0 bytes. Such thing is not going to work too well.
1694 You have been warned!! */
1695 #if defined _MSC_VER && !defined WIN32
1696 /* Microsoft C 16-bit versions limit malloc to approx 65512 bytes.
1697 The REALLOC define eliminates a flurry of conversion warnings,
1698 but is not required. */
1699 # define MAX_BUF_SIZE 65500L
1700 # define REALLOC(p,s) realloc ((p), (size_t) (s))
1702 # define MAX_BUF_SIZE (1L << 16)
1703 # define REALLOC(p,s) realloc ((p), (s))
1706 /* Extend the buffer by twice its current size via realloc and
1707 reset the pointers that pointed into the old block to point to the
1708 correct places in the new one. If extending the buffer results in it
1709 being larger than MAX_BUF_SIZE, then flag memory exhausted. */
1710 #define EXTEND_BUFFER() \
1712 unsigned char *old_buffer = bufp->buffer; \
1713 if (bufp->allocated == MAX_BUF_SIZE) \
1715 bufp->allocated <<= 1; \
1716 if (bufp->allocated > MAX_BUF_SIZE) \
1717 bufp->allocated = MAX_BUF_SIZE; \
1718 bufp->buffer = (unsigned char *) REALLOC (bufp->buffer, bufp->allocated);\
1719 if (bufp->buffer == NULL) \
1720 return REG_ESPACE; \
1721 /* If the buffer moved, move all the pointers into it. */ \
1722 if (old_buffer != bufp->buffer) \
1724 b = (b - old_buffer) + bufp->buffer; \
1725 begalt = (begalt - old_buffer) + bufp->buffer; \
1726 if (fixup_alt_jump) \
1727 fixup_alt_jump = (fixup_alt_jump - old_buffer) + bufp->buffer;\
1729 laststart = (laststart - old_buffer) + bufp->buffer; \
1730 if (pending_exact) \
1731 pending_exact = (pending_exact - old_buffer) + bufp->buffer; \
1736 /* Since we have one byte reserved for the register number argument to
1737 {start,stop}_memory, the maximum number of groups we can report
1738 things about is what fits in that byte. */
1739 #define MAX_REGNUM 255
1741 /* But patterns can have more than `MAX_REGNUM' registers. We just
1742 ignore the excess. */
1743 typedef unsigned regnum_t;
1746 /* Macros for the compile stack. */
1748 /* Since offsets can go either forwards or backwards, this type needs to
1749 be able to hold values from -(MAX_BUF_SIZE - 1) to MAX_BUF_SIZE - 1. */
1750 /* int may be not enough when sizeof(int) == 2. */
1751 typedef long pattern_offset_t;
1755 pattern_offset_t begalt_offset;
1756 pattern_offset_t fixup_alt_jump;
1757 pattern_offset_t inner_group_offset;
1758 pattern_offset_t laststart_offset;
1760 } compile_stack_elt_t;
1765 compile_stack_elt_t *stack;
1767 unsigned avail; /* Offset of next open position. */
1768 } compile_stack_type;
1771 #define INIT_COMPILE_STACK_SIZE 32
1773 #define COMPILE_STACK_EMPTY (compile_stack.avail == 0)
1774 #define COMPILE_STACK_FULL (compile_stack.avail == compile_stack.size)
1776 /* The next available element. */
1777 #define COMPILE_STACK_TOP (compile_stack.stack[compile_stack.avail])
1780 /* Set the bit for character C in a list. */
1781 #define SET_LIST_BIT(c) \
1782 (b[((unsigned char) (c)) / BYTEWIDTH] \
1783 |= 1 << (((unsigned char) c) % BYTEWIDTH))
1786 /* Get the next unsigned number in the uncompiled pattern. */
1787 #define GET_UNSIGNED_NUMBER(num) \
1791 while ('0' <= c && c <= '9') \
1795 num = num * 10 + c - '0'; \
1803 #if defined _LIBC || WIDE_CHAR_SUPPORT
1804 /* The GNU C library provides support for user-defined character classes
1805 and the functions from ISO C amendement 1. */
1806 # ifdef CHARCLASS_NAME_MAX
1807 # define CHAR_CLASS_MAX_LENGTH CHARCLASS_NAME_MAX
1809 /* This shouldn't happen but some implementation might still have this
1810 problem. Use a reasonable default value. */
1811 # define CHAR_CLASS_MAX_LENGTH 256
1815 # define IS_CHAR_CLASS(string) __wctype (string)
1817 # define IS_CHAR_CLASS(string) wctype (string)
1820 # define CHAR_CLASS_MAX_LENGTH 6 /* Namely, `xdigit'. */
1822 # define IS_CHAR_CLASS(string) \
1823 (STREQ (string, "alpha") || STREQ (string, "upper") \
1824 || STREQ (string, "lower") || STREQ (string, "digit") \
1825 || STREQ (string, "alnum") || STREQ (string, "xdigit") \
1826 || STREQ (string, "space") || STREQ (string, "print") \
1827 || STREQ (string, "punct") || STREQ (string, "graph") \
1828 || STREQ (string, "cntrl") || STREQ (string, "blank"))
1831 #ifndef MATCH_MAY_ALLOCATE
1833 /* If we cannot allocate large objects within re_match_2_internal,
1834 we make the fail stack and register vectors global.
1835 The fail stack, we grow to the maximum size when a regexp
1837 The register vectors, we adjust in size each time we
1838 compile a regexp, according to the number of registers it needs. */
1840 static fail_stack_type fail_stack;
1842 /* Size with which the following vectors are currently allocated.
1843 That is so we can make them bigger as needed,
1844 but never make them smaller. */
1845 static int regs_allocated_size;
1847 static const char ** regstart, ** regend;
1848 static const char ** old_regstart, ** old_regend;
1849 static const char **best_regstart, **best_regend;
1850 static register_info_type *reg_info;
1851 static const char **reg_dummy;
1852 static register_info_type *reg_info_dummy;
1854 /* Make the register vectors big enough for NUM_REGS registers,
1855 but don't make them smaller. */
1858 regex_grow_registers (num_regs)
1861 if (num_regs > regs_allocated_size)
1863 RETALLOC_IF (regstart, num_regs, const char *);
1864 RETALLOC_IF (regend, num_regs, const char *);
1865 RETALLOC_IF (old_regstart, num_regs, const char *);
1866 RETALLOC_IF (old_regend, num_regs, const char *);
1867 RETALLOC_IF (best_regstart, num_regs, const char *);
1868 RETALLOC_IF (best_regend, num_regs, const char *);
1869 RETALLOC_IF (reg_info, num_regs, register_info_type);
1870 RETALLOC_IF (reg_dummy, num_regs, const char *);
1871 RETALLOC_IF (reg_info_dummy, num_regs, register_info_type);
1873 regs_allocated_size = num_regs;
1877 #endif /* not MATCH_MAY_ALLOCATE */
1879 static boolean group_in_compile_stack _RE_ARGS ((compile_stack_type
1883 /* `regex_compile' compiles PATTERN (of length SIZE) according to SYNTAX.
1884 Returns one of error codes defined in `regex.h', or zero for success.
1886 Assumes the `allocated' (and perhaps `buffer') and `translate'
1887 fields are set in BUFP on entry.
1889 If it succeeds, results are put in BUFP (if it returns an error, the
1890 contents of BUFP are undefined):
1891 `buffer' is the compiled pattern;
1892 `syntax' is set to SYNTAX;
1893 `used' is set to the length of the compiled pattern;
1894 `fastmap_accurate' is zero;
1895 `re_nsub' is the number of subexpressions in PATTERN;
1896 `not_bol' and `not_eol' are zero;
1898 The `fastmap' and `newline_anchor' fields are neither
1899 examined nor set. */
1901 /* Return, freeing storage we allocated. */
1902 #define FREE_STACK_RETURN(value) \
1903 return (free (compile_stack.stack), value)
1905 static reg_errcode_t
1907 const char *pattern,
1909 reg_syntax_t syntax,
1910 struct re_pattern_buffer *bufp)
1912 /* We fetch characters from PATTERN here. Even though PATTERN is
1913 `char *' (i.e., signed), we declare these variables as unsigned, so
1914 they can be reliably used as array indices. */
1915 register unsigned char c, c1;
1917 /* A random temporary spot in PATTERN. */
1920 /* Points to the end of the buffer, where we should append. */
1921 register unsigned char *b;
1923 /* Keeps track of unclosed groups. */
1924 compile_stack_type compile_stack;
1926 /* Points to the current (ending) position in the pattern. */
1927 const char *p = pattern;
1928 const char *pend = pattern + size;
1930 /* How to translate the characters in the pattern. */
1931 RE_TRANSLATE_TYPE translate = bufp->translate;
1933 /* Address of the count-byte of the most recently inserted `exactn'
1934 command. This makes it possible to tell if a new exact-match
1935 character can be added to that command or if the character requires
1936 a new `exactn' command. */
1937 unsigned char *pending_exact = 0;
1939 /* Address of start of the most recently finished expression.
1940 This tells, e.g., postfix * where to find the start of its
1941 operand. Reset at the beginning of groups and alternatives. */
1942 unsigned char *laststart = 0;
1944 /* Address of beginning of regexp, or inside of last group. */
1945 unsigned char *begalt;
1947 /* Place in the uncompiled pattern (i.e., the {) to
1948 which to go back if the interval is invalid. */
1949 const char *beg_interval;
1951 /* Address of the place where a forward jump should go to the end of
1952 the containing expression. Each alternative of an `or' -- except the
1953 last -- ends with a forward jump of this sort. */
1954 unsigned char *fixup_alt_jump = 0;
1956 /* Counts open-groups as they are encountered. Remembered for the
1957 matching close-group on the compile stack, so the same register
1958 number is put in the stop_memory as the start_memory. */
1959 regnum_t regnum = 0;
1962 DEBUG_PRINT1 ("\nCompiling pattern: ");
1965 unsigned debug_count;
1967 for (debug_count = 0; debug_count < size; debug_count++)
1968 putchar (pattern[debug_count]);
1971 #endif /* REGEX_DEBUG */
1973 /* Initialize the compile stack. */
1974 compile_stack.stack = TALLOC (INIT_COMPILE_STACK_SIZE, compile_stack_elt_t);
1975 if (compile_stack.stack == NULL)
1978 compile_stack.size = INIT_COMPILE_STACK_SIZE;
1979 compile_stack.avail = 0;
1981 /* Initialize the pattern buffer. */
1982 bufp->syntax = syntax;
1983 bufp->fastmap_accurate = 0;
1984 bufp->not_bol = bufp->not_eol = 0;
1986 /* Set `used' to zero, so that if we return an error, the pattern
1987 printer (for debugging) will think there's no pattern. We reset it
1991 /* Always count groups, whether or not bufp->no_sub is set. */
1994 #if !defined emacs && !defined SYNTAX_TABLE
1995 /* Initialize the syntax table. */
1996 init_syntax_once ();
1999 if (bufp->allocated == 0)
2002 { /* If zero allocated, but buffer is non-null, try to realloc
2003 enough space. This loses if buffer's address is bogus, but
2004 that is the user's responsibility. */
2005 RETALLOC (bufp->buffer, INIT_BUF_SIZE, unsigned char);
2008 { /* Caller did not allocate a buffer. Do it for them. */
2009 bufp->buffer = TALLOC (INIT_BUF_SIZE, unsigned char);
2011 if (!bufp->buffer) FREE_STACK_RETURN (REG_ESPACE);
2013 bufp->allocated = INIT_BUF_SIZE;
2016 begalt = b = bufp->buffer;
2018 /* Loop through the uncompiled pattern until we're at the end. */
2027 if ( /* If at start of pattern, it's an operator. */
2029 /* If context independent, it's an operator. */
2030 || syntax & RE_CONTEXT_INDEP_ANCHORS
2031 /* Otherwise, depends on what's come before. */
2032 || at_begline_loc_p (pattern, p, syntax))
2042 if ( /* If at end of pattern, it's an operator. */
2044 /* If context independent, it's an operator. */
2045 || syntax & RE_CONTEXT_INDEP_ANCHORS
2046 /* Otherwise, depends on what's next. */
2047 || at_endline_loc_p (p, pend, syntax))
2057 if ((syntax & RE_BK_PLUS_QM)
2058 || (syntax & RE_LIMITED_OPS))
2062 /* If there is no previous pattern... */
2065 if (syntax & RE_CONTEXT_INVALID_OPS)
2066 FREE_STACK_RETURN (REG_BADRPT);
2067 else if (!(syntax & RE_CONTEXT_INDEP_OPS))
2072 /* Are we optimizing this jump? */
2073 boolean keep_string_p = false;
2075 /* 1 means zero (many) matches is allowed. */
2076 char zero_times_ok = 0, many_times_ok = 0;
2078 /* If there is a sequence of repetition chars, collapse it
2079 down to just one (the right one). We can't combine
2080 interval operators with these because of, e.g., `a{2}*',
2081 which should only match an even number of `a's. */
2085 zero_times_ok |= c != '+';
2086 many_times_ok |= c != '?';
2094 || (!(syntax & RE_BK_PLUS_QM) && (c == '+' || c == '?')))
2097 else if (syntax & RE_BK_PLUS_QM && c == '\\')
2099 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
2102 if (!(c1 == '+' || c1 == '?'))
2117 /* If we get here, we found another repeat character. */
2120 /* Star, etc. applied to an empty pattern is equivalent
2121 to an empty pattern. */
2125 /* Now we know whether or not zero matches is allowed
2126 and also whether or not two or more matches is allowed. */
2128 { /* More than one repetition is allowed, so put in at the
2129 end a backward relative jump from `b' to before the next
2130 jump we're going to put in below (which jumps from
2131 laststart to after this jump).
2133 But if we are at the `*' in the exact sequence `.*\n',
2134 insert an unconditional jump backwards to the .,
2135 instead of the beginning of the loop. This way we only
2136 push a failure point once, instead of every time
2137 through the loop. */
2138 assert (p - 1 > pattern);
2140 /* Allocate the space for the jump. */
2141 GET_BUFFER_SPACE (3);
2143 /* We know we are not at the first character of the pattern,
2144 because laststart was nonzero. And we've already
2145 incremented `p', by the way, to be the character after
2146 the `*'. Do we have to do something analogous here
2147 for null bytes, because of RE_DOT_NOT_NULL? */
2148 if (TRANSLATE (*(p - 2)) == TRANSLATE ('.')
2150 && p < pend && TRANSLATE (*p) == TRANSLATE ('\n')
2151 && !(syntax & RE_DOT_NEWLINE))
2152 { /* We have .*\n. */
2153 STORE_JUMP (jump, b, laststart);
2154 keep_string_p = true;
2157 /* Anything else. */
2158 STORE_JUMP (maybe_pop_jump, b, laststart - 3);
2160 /* We've added more stuff to the buffer. */
2164 /* On failure, jump from laststart to b + 3, which will be the
2165 end of the buffer after this jump is inserted. */
2166 GET_BUFFER_SPACE (3);
2167 INSERT_JUMP (keep_string_p ? on_failure_keep_string_jump
2175 /* At least one repetition is required, so insert a
2176 `dummy_failure_jump' before the initial
2177 `on_failure_jump' instruction of the loop. This
2178 effects a skip over that instruction the first time
2179 we hit that loop. */
2180 GET_BUFFER_SPACE (3);
2181 INSERT_JUMP (dummy_failure_jump, laststart, laststart + 6);
2196 boolean had_char_class = false;
2197 unsigned int range_start = 0xffffffff;
2199 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2201 /* Ensure that we have enough space to push a charset: the
2202 opcode, the length count, and the bitset; 34 bytes in all. */
2203 GET_BUFFER_SPACE (34);
2207 /* We test `*p == '^' twice, instead of using an if
2208 statement, so we only need one BUF_PUSH. */
2209 BUF_PUSH (*p == '^' ? charset_not : charset);
2213 /* Remember the first position in the bracket expression. */
2216 /* Push the number of bytes in the bitmap. */
2217 BUF_PUSH ((1 << BYTEWIDTH) / BYTEWIDTH);
2219 /* Clear the whole map. */
2220 bzero (b, (1 << BYTEWIDTH) / BYTEWIDTH);
2222 /* charset_not matches newline according to a syntax bit. */
2223 if ((re_opcode_t) b[-2] == charset_not
2224 && (syntax & RE_HAT_LISTS_NOT_NEWLINE))
2225 SET_LIST_BIT ('\n');
2227 /* Read in characters and ranges, setting map bits. */
2230 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2234 /* \ might escape characters inside [...] and [^...]. */
2235 if ((syntax & RE_BACKSLASH_ESCAPE_IN_LISTS) && c == '\\')
2237 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
2245 /* Could be the end of the bracket expression. If it's
2246 not (i.e., when the bracket expression is `[]' so
2247 far), the ']' character bit gets set way below. */
2248 if (c == ']' && p != p1 + 1)
2251 /* Look ahead to see if it's a range when the last thing
2252 was a character class. */
2253 if (had_char_class && c == '-' && *p != ']')
2254 FREE_STACK_RETURN (REG_ERANGE);
2256 /* Look ahead to see if it's a range when the last thing
2257 was a character: if this is a hyphen not at the
2258 beginning or the end of a list, then it's the range
2261 && !(p - 2 >= pattern && p[-2] == '[')
2262 && !(p - 3 >= pattern && p[-3] == '[' && p[-2] == '^')
2266 = compile_range (range_start, &p, pend, translate,
2268 if (ret != REG_NOERROR) FREE_STACK_RETURN (ret);
2269 range_start = 0xffffffff;
2272 else if (p[0] == '-' && p[1] != ']')
2273 { /* This handles ranges made up of characters only. */
2276 /* Move past the `-'. */
2279 ret = compile_range (c, &p, pend, translate, syntax, b);
2280 if (ret != REG_NOERROR) FREE_STACK_RETURN (ret);
2281 range_start = 0xffffffff;
2284 /* See if we're at the beginning of a possible character
2287 else if (syntax & RE_CHAR_CLASSES && c == '[' && *p == ':')
2288 { /* Leave room for the null. */
2289 char str[CHAR_CLASS_MAX_LENGTH + 1];
2294 /* If pattern is `[[:'. */
2295 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2300 if ((c == ':' && *p == ']') || p == pend)
2302 if (c1 < CHAR_CLASS_MAX_LENGTH)
2305 /* This is in any case an invalid class name. */
2310 /* If isn't a word bracketed by `[:' and `:]':
2311 undo the ending character, the letters, and leave
2312 the leading `:' and `[' (but set bits for them). */
2313 if (c == ':' && *p == ']')
2315 #if defined _LIBC || WIDE_CHAR_SUPPORT
2316 boolean is_lower = STREQ (str, "lower");
2317 boolean is_upper = STREQ (str, "upper");
2321 wt = IS_CHAR_CLASS (str);
2323 FREE_STACK_RETURN (REG_ECTYPE);
2325 /* Throw away the ] at the end of the character
2329 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2331 for (ch = 0; ch < 1 << BYTEWIDTH; ++ch)
2334 if (__iswctype (__btowc (ch), wt))
2337 if (iswctype (btowc (ch), wt))
2341 if (translate && (is_upper || is_lower)
2342 && (ISUPPER (ch) || ISLOWER (ch)))
2346 had_char_class = true;
2349 boolean is_alnum = STREQ (str, "alnum");
2350 boolean is_alpha = STREQ (str, "alpha");
2351 boolean is_blank = STREQ (str, "blank");
2352 boolean is_cntrl = STREQ (str, "cntrl");
2353 boolean is_digit = STREQ (str, "digit");
2354 boolean is_graph = STREQ (str, "graph");
2355 boolean is_lower = STREQ (str, "lower");
2356 boolean is_print = STREQ (str, "print");
2357 boolean is_punct = STREQ (str, "punct");
2358 boolean is_space = STREQ (str, "space");
2359 boolean is_upper = STREQ (str, "upper");
2360 boolean is_xdigit = STREQ (str, "xdigit");
2362 if (!IS_CHAR_CLASS (str))
2363 FREE_STACK_RETURN (REG_ECTYPE);
2365 /* Throw away the ] at the end of the character
2369 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2371 for (ch = 0; ch < 1 << BYTEWIDTH; ch++)
2373 /* This was split into 3 if's to
2374 avoid an arbitrary limit in some compiler. */
2375 if ( (is_alnum && ISALNUM (ch))
2376 || (is_alpha && ISALPHA (ch))
2377 || (is_blank && ISBLANK (ch))
2378 || (is_cntrl && ISCNTRL (ch)))
2380 if ( (is_digit && ISDIGIT (ch))
2381 || (is_graph && ISGRAPH (ch))
2382 || (is_lower && ISLOWER (ch))
2383 || (is_print && ISPRINT (ch)))
2385 if ( (is_punct && ISPUNCT (ch))
2386 || (is_space && ISSPACE (ch))
2387 || (is_upper && ISUPPER (ch))
2388 || (is_xdigit && ISXDIGIT (ch)))
2390 if ( translate && (is_upper || is_lower)
2391 && (ISUPPER (ch) || ISLOWER (ch)))
2394 had_char_class = true;
2395 #endif /* libc || wctype.h */
2405 had_char_class = false;
2408 else if (syntax & RE_CHAR_CLASSES && c == '[' && *p == '=')
2410 unsigned char str[MB_LEN_MAX + 1];
2413 _NL_CURRENT_WORD (LC_COLLATE, _NL_COLLATE_NRULES);
2419 /* If pattern is `[[='. */
2420 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2425 if ((c == '=' && *p == ']') || p == pend)
2427 if (c1 < MB_LEN_MAX)
2430 /* This is in any case an invalid class name. */
2435 if (c == '=' && *p == ']' && str[0] != '\0')
2437 /* If we have no collation data we use the default
2438 collation in which each character is in a class
2439 by itself. It also means that ASCII is the
2440 character set and therefore we cannot have character
2441 with more than one byte in the multibyte
2448 FREE_STACK_RETURN (REG_ECOLLATE);
2450 /* Throw away the ] at the end of the equivalence
2454 /* Set the bit for the character. */
2455 SET_LIST_BIT (str[0]);
2460 /* Try to match the byte sequence in `str' against
2461 those known to the collate implementation.
2462 First find out whether the bytes in `str' are
2463 actually from exactly one character. */
2464 const int32_t *table;
2465 const unsigned char *weights;
2466 const unsigned char *extra;
2467 const int32_t *indirect;
2469 const unsigned char *cp = str;
2472 /* This #include defines a local function! */
2473 # include <locale/weight.h>
2475 table = (const int32_t *)
2476 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_TABLEMB);
2477 weights = (const unsigned char *)
2478 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_WEIGHTMB);
2479 extra = (const unsigned char *)
2480 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_EXTRAMB);
2481 indirect = (const int32_t *)
2482 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_INDIRECTMB);
2484 idx = findidx (&cp);
2485 if (idx == 0 || cp < str + c1)
2486 /* This is no valid character. */
2487 FREE_STACK_RETURN (REG_ECOLLATE);
2489 /* Throw away the ] at the end of the equivalence
2493 /* Now we have to go throught the whole table
2494 and find all characters which have the same
2497 XXX Note that this is not entirely correct.
2498 we would have to match multibyte sequences
2499 but this is not possible with the current
2501 for (ch = 1; ch < 256; ++ch)
2502 /* XXX This test would have to be changed if we
2503 would allow matching multibyte sequences. */
2506 int32_t idx2 = table[ch];
2507 size_t len = weights[idx2];
2509 /* Test whether the lenghts match. */
2510 if (weights[idx] == len)
2512 /* They do. New compare the bytes of
2517 && (weights[idx + 1 + cnt]
2518 == weights[idx2 + 1 + cnt]))
2522 /* They match. Mark the character as
2529 had_char_class = true;
2539 had_char_class = false;
2542 else if (syntax & RE_CHAR_CLASSES && c == '[' && *p == '.')
2544 unsigned char str[128]; /* Should be large enough. */
2547 _NL_CURRENT_WORD (LC_COLLATE, _NL_COLLATE_NRULES);
2553 /* If pattern is `[[='. */
2554 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2559 if ((c == '.' && *p == ']') || p == pend)
2561 if (c1 < sizeof (str))
2564 /* This is in any case an invalid class name. */
2569 if (c == '.' && *p == ']' && str[0] != '\0')
2571 /* If we have no collation data we use the default
2572 collation in which each character is the name
2573 for its own class which contains only the one
2574 character. It also means that ASCII is the
2575 character set and therefore we cannot have character
2576 with more than one byte in the multibyte
2583 FREE_STACK_RETURN (REG_ECOLLATE);
2585 /* Throw away the ] at the end of the equivalence
2589 /* Set the bit for the character. */
2590 SET_LIST_BIT (str[0]);
2591 range_start = ((const unsigned char *) str)[0];
2596 /* Try to match the byte sequence in `str' against
2597 those known to the collate implementation.
2598 First find out whether the bytes in `str' are
2599 actually from exactly one character. */
2601 const int32_t *symb_table;
2602 const unsigned char *extra;
2609 _NL_CURRENT_WORD (LC_COLLATE,
2610 _NL_COLLATE_SYMB_HASH_SIZEMB);
2611 symb_table = (const int32_t *)
2612 _NL_CURRENT (LC_COLLATE,
2613 _NL_COLLATE_SYMB_TABLEMB);
2614 extra = (const unsigned char *)
2615 _NL_CURRENT (LC_COLLATE,
2616 _NL_COLLATE_SYMB_EXTRAMB);
2618 /* Locate the character in the hashing table. */
2619 hash = elem_hash (str, c1);
2622 elem = hash % table_size;
2623 second = hash % (table_size - 2);
2624 while (symb_table[2 * elem] != 0)
2626 /* First compare the hashing value. */
2627 if (symb_table[2 * elem] == hash
2628 && c1 == extra[symb_table[2 * elem + 1]]
2630 &extra[symb_table[2 * elem + 1]
2634 /* Yep, this is the entry. */
2635 idx = symb_table[2 * elem + 1];
2636 idx += 1 + extra[idx];
2644 if (symb_table[2 * elem] == 0)
2645 /* This is no valid character. */
2646 FREE_STACK_RETURN (REG_ECOLLATE);
2648 /* Throw away the ] at the end of the equivalence
2652 /* Now add the multibyte character(s) we found
2653 to the acceptabed list.
2655 XXX Note that this is not entirely correct.
2656 we would have to match multibyte sequences
2657 but this is not possible with the current
2658 implementation. Also, we have to match
2659 collating symbols, which expand to more than
2660 one file, as a whole and not allow the
2661 individual bytes. */
2664 range_start = extra[idx];
2666 SET_LIST_BIT (extra[idx++]);
2669 had_char_class = false;
2679 had_char_class = false;
2684 had_char_class = false;
2690 /* Discard any (non)matching list bytes that are all 0 at the
2691 end of the map. Decrease the map-length byte too. */
2692 while ((int) b[-1] > 0 && b[b[-1] - 1] == 0)
2700 if (syntax & RE_NO_BK_PARENS)
2707 if (syntax & RE_NO_BK_PARENS)
2714 if (syntax & RE_NEWLINE_ALT)
2721 if (syntax & RE_NO_BK_VBAR)
2728 if (syntax & RE_INTERVALS && syntax & RE_NO_BK_BRACES)
2729 goto handle_interval;
2735 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
2737 /* Do not translate the character after the \, so that we can
2738 distinguish, e.g., \B from \b, even if we normally would
2739 translate, e.g., B to b. */
2745 if (syntax & RE_NO_BK_PARENS)
2746 goto normal_backslash;
2752 if (COMPILE_STACK_FULL)
2754 RETALLOC (compile_stack.stack, compile_stack.size << 1,
2755 compile_stack_elt_t);
2756 if (compile_stack.stack == NULL) return REG_ESPACE;
2758 compile_stack.size <<= 1;
2761 /* These are the values to restore when we hit end of this
2762 group. They are all relative offsets, so that if the
2763 whole pattern moves because of realloc, they will still
2765 COMPILE_STACK_TOP.begalt_offset = begalt - bufp->buffer;
2766 COMPILE_STACK_TOP.fixup_alt_jump
2767 = fixup_alt_jump ? fixup_alt_jump - bufp->buffer + 1 : 0;
2768 COMPILE_STACK_TOP.laststart_offset = b - bufp->buffer;
2769 COMPILE_STACK_TOP.regnum = regnum;
2771 /* We will eventually replace the 0 with the number of
2772 groups inner to this one. But do not push a
2773 start_memory for groups beyond the last one we can
2774 represent in the compiled pattern. */
2775 if (regnum <= MAX_REGNUM)
2777 COMPILE_STACK_TOP.inner_group_offset = b - bufp->buffer + 2;
2778 BUF_PUSH_3 (start_memory, regnum, 0);
2781 compile_stack.avail++;
2786 /* If we've reached MAX_REGNUM groups, then this open
2787 won't actually generate any code, so we'll have to
2788 clear pending_exact explicitly. */
2794 if (syntax & RE_NO_BK_PARENS) goto normal_backslash;
2796 if (COMPILE_STACK_EMPTY)
2798 if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD)
2799 goto normal_backslash;
2801 FREE_STACK_RETURN (REG_ERPAREN);
2806 { /* Push a dummy failure point at the end of the
2807 alternative for a possible future
2808 `pop_failure_jump' to pop. See comments at
2809 `push_dummy_failure' in `re_match_2'. */
2810 BUF_PUSH (push_dummy_failure);
2812 /* We allocated space for this jump when we assigned
2813 to `fixup_alt_jump', in the `handle_alt' case below. */
2814 STORE_JUMP (jump_past_alt, fixup_alt_jump, b - 1);
2817 /* See similar code for backslashed left paren above. */
2818 if (COMPILE_STACK_EMPTY)
2820 if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD)
2823 FREE_STACK_RETURN (REG_ERPAREN);
2826 /* Since we just checked for an empty stack above, this
2827 ``can't happen''. */
2828 assert (compile_stack.avail != 0);
2830 /* We don't just want to restore into `regnum', because
2831 later groups should continue to be numbered higher,
2832 as in `(ab)c(de)' -- the second group is #2. */
2833 regnum_t this_group_regnum;
2835 compile_stack.avail--;
2836 begalt = bufp->buffer + COMPILE_STACK_TOP.begalt_offset;
2838 = COMPILE_STACK_TOP.fixup_alt_jump
2839 ? bufp->buffer + COMPILE_STACK_TOP.fixup_alt_jump - 1
2841 laststart = bufp->buffer + COMPILE_STACK_TOP.laststart_offset;
2842 this_group_regnum = COMPILE_STACK_TOP.regnum;
2843 /* If we've reached MAX_REGNUM groups, then this open
2844 won't actually generate any code, so we'll have to
2845 clear pending_exact explicitly. */
2848 /* We're at the end of the group, so now we know how many
2849 groups were inside this one. */
2850 if (this_group_regnum <= MAX_REGNUM)
2852 unsigned char *inner_group_loc
2853 = bufp->buffer + COMPILE_STACK_TOP.inner_group_offset;
2855 *inner_group_loc = regnum - this_group_regnum;
2856 BUF_PUSH_3 (stop_memory, this_group_regnum,
2857 regnum - this_group_regnum);
2863 case '|': /* `\|'. */
2864 if (syntax & RE_LIMITED_OPS || syntax & RE_NO_BK_VBAR)
2865 goto normal_backslash;
2867 if (syntax & RE_LIMITED_OPS)
2870 /* Insert before the previous alternative a jump which
2871 jumps to this alternative if the former fails. */
2872 GET_BUFFER_SPACE (3);
2873 INSERT_JUMP (on_failure_jump, begalt, b + 6);
2877 /* The alternative before this one has a jump after it
2878 which gets executed if it gets matched. Adjust that
2879 jump so it will jump to this alternative's analogous
2880 jump (put in below, which in turn will jump to the next
2881 (if any) alternative's such jump, etc.). The last such
2882 jump jumps to the correct final destination. A picture:
2888 If we are at `b', then fixup_alt_jump right now points to a
2889 three-byte space after `a'. We'll put in the jump, set
2890 fixup_alt_jump to right after `b', and leave behind three
2891 bytes which we'll fill in when we get to after `c'. */
2894 STORE_JUMP (jump_past_alt, fixup_alt_jump, b);
2896 /* Mark and leave space for a jump after this alternative,
2897 to be filled in later either by next alternative or
2898 when know we're at the end of a series of alternatives. */
2900 GET_BUFFER_SPACE (3);
2909 /* If \{ is a literal. */
2910 if (!(syntax & RE_INTERVALS)
2911 /* If we're at `\{' and it's not the open-interval
2913 || (syntax & RE_NO_BK_BRACES))
2914 goto normal_backslash;
2918 /* If got here, then the syntax allows intervals. */
2920 /* At least (most) this many matches must be made. */
2921 int lower_bound = -1, upper_bound = -1;
2923 beg_interval = p - 1;
2927 if (!(syntax & RE_INTERVALS) && (syntax & RE_NO_BK_BRACES))
2928 goto unfetch_interval;
2930 FREE_STACK_RETURN (REG_EBRACE);
2933 GET_UNSIGNED_NUMBER (lower_bound);
2937 GET_UNSIGNED_NUMBER (upper_bound);
2938 if ((!(syntax & RE_NO_BK_BRACES) && c != '\\')
2939 || ((syntax & RE_NO_BK_BRACES) && c != '}'))
2940 FREE_STACK_RETURN (REG_BADBR);
2942 if (upper_bound < 0)
2943 upper_bound = RE_DUP_MAX;
2946 /* Interval such as `{1}' => match exactly once. */
2947 upper_bound = lower_bound;
2949 if (lower_bound < 0 || upper_bound > RE_DUP_MAX
2950 || lower_bound > upper_bound)
2952 if (!(syntax & RE_INTERVALS) && (syntax & RE_NO_BK_BRACES))
2953 goto unfetch_interval;
2955 FREE_STACK_RETURN (REG_BADBR);
2958 if (!(syntax & RE_NO_BK_BRACES))
2960 if (c != '\\') FREE_STACK_RETURN (REG_EBRACE);
2967 if (!(syntax & RE_INTERVALS) && (syntax & RE_NO_BK_BRACES))
2968 goto unfetch_interval;
2970 FREE_STACK_RETURN (REG_BADBR);
2973 /* We just parsed a valid interval. */
2975 /* If it's invalid to have no preceding re. */
2978 if (syntax & RE_CONTEXT_INVALID_OPS)
2979 FREE_STACK_RETURN (REG_BADRPT);
2980 else if (syntax & RE_CONTEXT_INDEP_OPS)
2983 goto unfetch_interval;
2986 /* If the upper bound is zero, don't want to succeed at
2987 all; jump from `laststart' to `b + 3', which will be
2988 the end of the buffer after we insert the jump. */
2989 if (upper_bound == 0)
2991 GET_BUFFER_SPACE (3);
2992 INSERT_JUMP (jump, laststart, b + 3);
2996 /* Otherwise, we have a nontrivial interval. When
2997 we're all done, the pattern will look like:
2998 set_number_at <jump count> <upper bound>
2999 set_number_at <succeed_n count> <lower bound>
3000 succeed_n <after jump addr> <succeed_n count>
3002 jump_n <succeed_n addr> <jump count>
3003 (The upper bound and `jump_n' are omitted if
3004 `upper_bound' is 1, though.) */
3006 { /* If the upper bound is > 1, we need to insert
3007 more at the end of the loop. */
3008 unsigned nbytes = 10 + (upper_bound > 1) * 10;
3010 GET_BUFFER_SPACE (nbytes);
3012 /* Initialize lower bound of the `succeed_n', even
3013 though it will be set during matching by its
3014 attendant `set_number_at' (inserted next),
3015 because `re_compile_fastmap' needs to know.
3016 Jump to the `jump_n' we might insert below. */
3017 INSERT_JUMP2 (succeed_n, laststart,
3018 b + 5 + (upper_bound > 1) * 5,
3022 /* Code to initialize the lower bound. Insert
3023 before the `succeed_n'. The `5' is the last two
3024 bytes of this `set_number_at', plus 3 bytes of
3025 the following `succeed_n'. */
3026 insert_op2 (set_number_at, laststart, 5, lower_bound, b);
3029 if (upper_bound > 1)
3030 { /* More than one repetition is allowed, so
3031 append a backward jump to the `succeed_n'
3032 that starts this interval.
3034 When we've reached this during matching,
3035 we'll have matched the interval once, so
3036 jump back only `upper_bound - 1' times. */
3037 STORE_JUMP2 (jump_n, b, laststart + 5,
3041 /* The location we want to set is the second
3042 parameter of the `jump_n'; that is `b-2' as
3043 an absolute address. `laststart' will be
3044 the `set_number_at' we're about to insert;
3045 `laststart+3' the number to set, the source
3046 for the relative address. But we are
3047 inserting into the middle of the pattern --
3048 so everything is getting moved up by 5.
3049 Conclusion: (b - 2) - (laststart + 3) + 5,
3050 i.e., b - laststart.
3052 We insert this at the beginning of the loop
3053 so that if we fail during matching, we'll
3054 reinitialize the bounds. */
3055 insert_op2 (set_number_at, laststart, b - laststart,
3056 upper_bound - 1, b);
3061 beg_interval = NULL;
3066 /* If an invalid interval, match the characters as literals. */
3067 assert (beg_interval);
3069 beg_interval = NULL;
3071 /* normal_char and normal_backslash need `c'. */
3074 if (!(syntax & RE_NO_BK_BRACES))
3076 if (p > pattern && p[-1] == '\\')
3077 goto normal_backslash;
3082 /* There is no way to specify the before_dot and after_dot
3083 operators. rms says this is ok. --karl */
3091 BUF_PUSH_2 (syntaxspec, syntax_spec_code[c]);
3097 BUF_PUSH_2 (notsyntaxspec, syntax_spec_code[c]);
3103 if (syntax & RE_NO_GNU_OPS)
3106 BUF_PUSH (wordchar);
3111 if (syntax & RE_NO_GNU_OPS)
3114 BUF_PUSH (notwordchar);
3119 if (syntax & RE_NO_GNU_OPS)
3125 if (syntax & RE_NO_GNU_OPS)
3131 if (syntax & RE_NO_GNU_OPS)
3133 BUF_PUSH (wordbound);
3137 if (syntax & RE_NO_GNU_OPS)
3139 BUF_PUSH (notwordbound);
3143 if (syntax & RE_NO_GNU_OPS)
3149 if (syntax & RE_NO_GNU_OPS)
3154 case '1': case '2': case '3': case '4': case '5':
3155 case '6': case '7': case '8': case '9':
3156 if (syntax & RE_NO_BK_REFS)
3162 FREE_STACK_RETURN (REG_ESUBREG);
3164 /* Can't back reference to a subexpression if inside of it. */
3165 if (group_in_compile_stack (compile_stack, (regnum_t) c1))
3169 BUF_PUSH_2 (duplicate, c1);
3175 if (syntax & RE_BK_PLUS_QM)
3178 goto normal_backslash;
3182 /* You might think it would be useful for \ to mean
3183 not to translate; but if we don't translate it
3184 it will never match anything. */
3192 /* Expects the character in `c'. */
3194 /* If no exactn currently being built. */
3197 /* If last exactn not at current position. */
3198 || pending_exact + *pending_exact + 1 != b
3200 /* We have only one byte following the exactn for the count. */
3201 || *pending_exact == (1 << BYTEWIDTH) - 1
3203 /* If followed by a repetition operator. */
3204 || *p == '*' || *p == '^'
3205 || ((syntax & RE_BK_PLUS_QM)
3206 ? *p == '\\' && (p[1] == '+' || p[1] == '?')
3207 : (*p == '+' || *p == '?'))
3208 || ((syntax & RE_INTERVALS)
3209 && ((syntax & RE_NO_BK_BRACES)
3211 : (p[0] == '\\' && p[1] == '{'))))
3213 /* Start building a new exactn. */
3217 BUF_PUSH_2 (exactn, 0);
3218 pending_exact = b - 1;
3225 } /* while p != pend */
3228 /* Through the pattern now. */
3231 STORE_JUMP (jump_past_alt, fixup_alt_jump, b);
3233 if (!COMPILE_STACK_EMPTY)
3234 FREE_STACK_RETURN (REG_EPAREN);
3236 /* If we don't want backtracking, force success
3237 the first time we reach the end of the compiled pattern. */
3238 if (syntax & RE_NO_POSIX_BACKTRACKING)
3241 free (compile_stack.stack);
3243 /* We have succeeded; set the length of the buffer. */
3244 bufp->used = b - bufp->buffer;
3249 DEBUG_PRINT1 ("\nCompiled pattern: \n");
3250 print_compiled_pattern (bufp);
3252 #endif /* REGEX_DEBUG */
3254 #ifndef MATCH_MAY_ALLOCATE
3255 /* Initialize the failure stack to the largest possible stack. This
3256 isn't necessary unless we're trying to avoid calling alloca in
3257 the search and match routines. */
3259 int num_regs = bufp->re_nsub + 1;
3261 /* Since DOUBLE_FAIL_STACK refuses to double only if the current size
3262 is strictly greater than re_max_failures, the largest possible stack
3263 is 2 * re_max_failures failure points. */
3264 if (fail_stack.size < (2 * re_max_failures * MAX_FAILURE_ITEMS))
3266 fail_stack.size = (2 * re_max_failures * MAX_FAILURE_ITEMS);
3269 if (! fail_stack.stack)
3271 = (fail_stack_elt_t *) xmalloc (fail_stack.size
3272 * sizeof (fail_stack_elt_t));
3275 = (fail_stack_elt_t *) xrealloc (fail_stack.stack,
3277 * sizeof (fail_stack_elt_t)));
3278 # else /* not emacs */
3279 if (! fail_stack.stack)
3281 = (fail_stack_elt_t *) malloc (fail_stack.size
3282 * sizeof (fail_stack_elt_t));
3285 = (fail_stack_elt_t *) realloc (fail_stack.stack,
3287 * sizeof (fail_stack_elt_t)));
3288 # endif /* not emacs */
3291 regex_grow_registers (num_regs);
3293 #endif /* not MATCH_MAY_ALLOCATE */
3296 } /* regex_compile */
3298 /* Subroutines for `regex_compile'. */
3300 /* Store OP at LOC followed by two-byte integer parameter ARG. */
3308 *loc = (unsigned char) op;
3309 STORE_NUMBER (loc + 1, arg);
3313 /* Like `store_op1', but for two two-byte parameters ARG1 and ARG2. */
3321 *loc = (unsigned char) op;
3322 STORE_NUMBER (loc + 1, arg1);
3323 STORE_NUMBER (loc + 3, arg2);
3327 /* Copy the bytes from LOC to END to open up three bytes of space at LOC
3328 for OP followed by two-byte integer parameter ARG. */
3337 register unsigned char *pfrom = end;
3338 register unsigned char *pto = end + 3;
3340 while (pfrom != loc)
3343 store_op1 (op, loc, arg);
3347 /* Like `insert_op1', but for two two-byte parameters ARG1 and ARG2. */
3356 register unsigned char *pfrom = end;
3357 register unsigned char *pto = end + 5;
3359 while (pfrom != loc)
3362 store_op2 (op, loc, arg1, arg2);
3366 /* P points to just after a ^ in PATTERN. Return true if that ^ comes
3367 after an alternative or a begin-subexpression. We assume there is at
3368 least one character before the ^. */
3372 const char *pattern, const char *p,
3373 reg_syntax_t syntax)
3375 const char *prev = p - 2;
3376 boolean prev_prev_backslash = prev > pattern && prev[-1] == '\\';
3379 /* After a subexpression? */
3380 (*prev == '(' && (syntax & RE_NO_BK_PARENS || prev_prev_backslash))
3381 /* After an alternative? */
3382 || (*prev == '|' && (syntax & RE_NO_BK_VBAR || prev_prev_backslash));
3386 /* The dual of at_begline_loc_p. This one is for $. We assume there is
3387 at least one character after the $, i.e., `P < PEND'. */
3391 const char *p, const char *pend,
3392 reg_syntax_t syntax)
3394 const char *next = p;
3395 boolean next_backslash = *next == '\\';
3396 const char *next_next = p + 1 < pend ? p + 1 : 0;
3399 /* Before a subexpression? */
3400 (syntax & RE_NO_BK_PARENS ? *next == ')'
3401 : next_backslash && next_next && *next_next == ')')
3402 /* Before an alternative? */
3403 || (syntax & RE_NO_BK_VBAR ? *next == '|'
3404 : next_backslash && next_next && *next_next == '|');
3408 /* Returns true if REGNUM is in one of COMPILE_STACK's elements and
3409 false if it's not. */
3412 group_in_compile_stack (
3413 compile_stack_type compile_stack,
3418 for (this_element = compile_stack.avail - 1;
3421 if (compile_stack.stack[this_element].regnum == regnum)
3428 /* Read the ending character of a range (in a bracket expression) from the
3429 uncompiled pattern *P_PTR (which ends at PEND). We assume the
3430 starting character is in `P[-2]'. (`P[-1]' is the character `-'.)
3431 Then we set the translation of all bits between the starting and
3432 ending characters (inclusive) in the compiled pattern B.
3434 Return an error code.
3436 We use these short variable names so we can use the same macros as
3437 `regex_compile' itself. */
3439 static reg_errcode_t
3441 unsigned int range_start_char,
3442 const char **p_ptr, const char *pend,
3443 RE_TRANSLATE_TYPE translate,
3444 reg_syntax_t syntax,
3449 const char *p = *p_ptr;
3451 char range_start[2];
3458 /* Fetch the endpoints without translating them; the
3459 appropriate translation is done in the bit-setting loop below. */
3460 range_start[0] = range_start_char;
3461 range_start[1] = '\0';
3462 range_end[0] = p[0];
3463 range_end[1] = '\0';
3465 /* Have to increment the pointer into the pattern string, so the
3466 caller isn't still at the ending character. */
3469 /* Report an error if the range is empty and the syntax prohibits this. */
3470 ret = syntax & RE_NO_EMPTY_RANGES ? REG_ERANGE : REG_NOERROR;
3472 /* Here we see why `this_char' has to be larger than an `unsigned
3473 char' -- we would otherwise go into an infinite loop, since all
3474 characters <= 0xff. */
3476 for (this_char = 0; this_char <= (unsigned char) -1; ++this_char)
3479 if (strcoll (range_start, ch) <= 0 && strcoll (ch, range_end) <= 0)
3481 SET_LIST_BIT (TRANSLATE (this_char));
3489 /* re_compile_fastmap computes a ``fastmap'' for the compiled pattern in
3490 BUFP. A fastmap records which of the (1 << BYTEWIDTH) possible
3491 characters can start a string that matches the pattern. This fastmap
3492 is used by re_search to skip quickly over impossible starting points.
3494 The caller must supply the address of a (1 << BYTEWIDTH)-byte data
3495 area as BUFP->fastmap.
3497 We set the `fastmap', `fastmap_accurate', and `can_be_null' fields in
3500 Returns 0 if we succeed, -2 if an internal error. */
3503 re_compile_fastmap (
3504 struct re_pattern_buffer *bufp)
3507 #ifdef MATCH_MAY_ALLOCATE
3508 fail_stack_type fail_stack;
3510 #ifndef REGEX_MALLOC
3514 register char *fastmap = bufp->fastmap;
3515 unsigned char *pattern = bufp->buffer;
3516 unsigned char *p = pattern;
3517 register unsigned char *pend = pattern + bufp->used;
3520 /* This holds the pointer to the failure stack, when
3521 it is allocated relocatably. */
3522 fail_stack_elt_t *failure_stack_ptr;
3525 /* Assume that each path through the pattern can be null until
3526 proven otherwise. We set this false at the bottom of switch
3527 statement, to which we get only if a particular path doesn't
3528 match the empty string. */
3529 boolean path_can_be_null = true;
3531 /* We aren't doing a `succeed_n' to begin with. */
3532 boolean succeed_n_p = false;
3534 assert (fastmap != NULL && p != NULL);
3537 bzero (fastmap, 1 << BYTEWIDTH); /* Assume nothing's valid. */
3538 bufp->fastmap_accurate = 1; /* It will be when we're done. */
3539 bufp->can_be_null = 0;
3543 if (p == pend || *p == succeed)
3545 /* We have reached the (effective) end of pattern. */
3546 if (!FAIL_STACK_EMPTY ())
3548 bufp->can_be_null |= path_can_be_null;
3550 /* Reset for next path. */
3551 path_can_be_null = true;
3553 p = fail_stack.stack[--fail_stack.avail].pointer;
3561 /* We should never be about to go beyond the end of the pattern. */
3564 switch (SWITCH_ENUM_CAST ((re_opcode_t) *p++))
3567 /* I guess the idea here is to simply not bother with a fastmap
3568 if a backreference is used, since it's too hard to figure out
3569 the fastmap for the corresponding group. Setting
3570 `can_be_null' stops `re_search_2' from using the fastmap, so
3571 that is all we do. */
3573 bufp->can_be_null = 1;
3577 /* Following are the cases which match a character. These end
3586 for (j = *p++ * BYTEWIDTH - 1; j >= 0; j--)
3587 if (p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH)))
3593 /* Chars beyond end of map must be allowed. */
3594 for (j = *p * BYTEWIDTH; j < (1 << BYTEWIDTH); j++)
3597 for (j = *p++ * BYTEWIDTH - 1; j >= 0; j--)
3598 if (!(p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH))))
3604 for (j = 0; j < (1 << BYTEWIDTH); j++)
3605 if (SYNTAX (j) == Sword)
3611 for (j = 0; j < (1 << BYTEWIDTH); j++)
3612 if (SYNTAX (j) != Sword)
3619 int fastmap_newline = fastmap['\n'];
3621 /* `.' matches anything ... */
3622 for (j = 0; j < (1 << BYTEWIDTH); j++)
3625 /* ... except perhaps newline. */
3626 if (!(bufp->syntax & RE_DOT_NEWLINE))
3627 fastmap['\n'] = fastmap_newline;
3629 /* Return if we have already set `can_be_null'; if we have,
3630 then the fastmap is irrelevant. Something's wrong here. */
3631 else if (bufp->can_be_null)
3634 /* Otherwise, have to check alternative paths. */
3641 for (j = 0; j < (1 << BYTEWIDTH); j++)
3642 if (SYNTAX (j) == (enum syntaxcode) k)
3649 for (j = 0; j < (1 << BYTEWIDTH); j++)
3650 if (SYNTAX (j) != (enum syntaxcode) k)
3655 /* All cases after this match the empty string. These end with
3675 case push_dummy_failure:
3680 case pop_failure_jump:
3681 case maybe_pop_jump:
3684 case dummy_failure_jump:
3685 EXTRACT_NUMBER_AND_INCR (j, p);
3690 /* Jump backward implies we just went through the body of a
3691 loop and matched nothing. Opcode jumped to should be
3692 `on_failure_jump' or `succeed_n'. Just treat it like an
3693 ordinary jump. For a * loop, it has pushed its failure
3694 point already; if so, discard that as redundant. */
3695 if ((re_opcode_t) *p != on_failure_jump
3696 && (re_opcode_t) *p != succeed_n)
3700 EXTRACT_NUMBER_AND_INCR (j, p);
3703 /* If what's on the stack is where we are now, pop it. */
3704 if (!FAIL_STACK_EMPTY ()
3705 && fail_stack.stack[fail_stack.avail - 1].pointer == p)
3711 case on_failure_jump:
3712 case on_failure_keep_string_jump:
3713 handle_on_failure_jump:
3714 EXTRACT_NUMBER_AND_INCR (j, p);
3716 /* For some patterns, e.g., `(a?)?', `p+j' here points to the
3717 end of the pattern. We don't want to push such a point,
3718 since when we restore it above, entering the switch will
3719 increment `p' past the end of the pattern. We don't need
3720 to push such a point since we obviously won't find any more
3721 fastmap entries beyond `pend'. Such a pattern can match
3722 the null string, though. */
3725 if (!PUSH_PATTERN_OP (p + j, fail_stack))
3727 RESET_FAIL_STACK ();
3732 bufp->can_be_null = 1;
3736 EXTRACT_NUMBER_AND_INCR (k, p); /* Skip the n. */
3737 succeed_n_p = false;
3744 /* Get to the number of times to succeed. */
3747 /* Increment p past the n for when k != 0. */
3748 EXTRACT_NUMBER_AND_INCR (k, p);
3752 succeed_n_p = true; /* Spaghetti code alert. */
3753 goto handle_on_failure_jump;
3770 abort (); /* We have listed all the cases. */
3773 /* Getting here means we have found the possible starting
3774 characters for one path of the pattern -- and that the empty
3775 string does not match. We need not follow this path further.
3776 Instead, look at the next alternative (remembered on the
3777 stack), or quit if no more. The test at the top of the loop
3778 does these things. */
3779 path_can_be_null = false;
3783 /* Set `can_be_null' for the last path (also the first path, if the
3784 pattern is empty). */
3785 bufp->can_be_null |= path_can_be_null;
3788 RESET_FAIL_STACK ();
3790 } /* re_compile_fastmap */
3792 weak_alias (__re_compile_fastmap, re_compile_fastmap)
3795 /* Set REGS to hold NUM_REGS registers, storing them in STARTS and
3796 ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use
3797 this memory for recording register information. STARTS and ENDS
3798 must be allocated using the malloc library routine, and must each
3799 be at least NUM_REGS * sizeof (regoff_t) bytes long.
3801 If NUM_REGS == 0, then subsequent matches should allocate their own
3804 Unless this function is called, the first search or match using
3805 PATTERN_BUFFER will allocate its own register data, without
3806 freeing the old data. */
3810 struct re_pattern_buffer *bufp,
3811 struct re_registers *regs,
3813 regoff_t *starts, regoff_t *ends)
3817 bufp->regs_allocated = REGS_REALLOCATE;
3818 regs->num_regs = num_regs;
3819 regs->start = starts;
3824 bufp->regs_allocated = REGS_UNALLOCATED;
3826 regs->start = regs->end = (regoff_t *) 0;
3830 weak_alias (__re_set_registers, re_set_registers)
3833 /* Searching routines. */
3835 /* Like re_search_2, below, but only one string is specified, and
3836 doesn't let you say where to stop matching. */
3840 struct re_pattern_buffer *bufp,
3842 int size, int startpos, int range,
3843 struct re_registers *regs)
3845 return re_search_2 (bufp, NULL, 0, string, size, startpos, range,
3849 weak_alias (__re_search, re_search)
3853 /* Using the compiled pattern in BUFP->buffer, first tries to match the
3854 virtual concatenation of STRING1 and STRING2, starting first at index
3855 STARTPOS, then at STARTPOS + 1, and so on.
3857 STRING1 and STRING2 have length SIZE1 and SIZE2, respectively.
3859 RANGE is how far to scan while trying to match. RANGE = 0 means try
3860 only at STARTPOS; in general, the last start tried is STARTPOS +
3863 In REGS, return the indices of the virtual concatenation of STRING1
3864 and STRING2 that matched the entire BUFP->buffer and its contained
3867 Do not consider matching one past the index STOP in the virtual
3868 concatenation of STRING1 and STRING2.
3870 We return either the position in the strings at which the match was
3871 found, -1 if no match, or -2 if error (such as failure
3876 struct re_pattern_buffer *bufp,
3877 const char *string1, const char *string2,
3878 int size1, int size2,
3881 struct re_registers *regs,
3885 register char *fastmap = bufp->fastmap;
3886 register RE_TRANSLATE_TYPE translate = bufp->translate;
3887 int total_size = size1 + size2;
3888 int endpos = startpos + range;
3890 /* Check for out-of-range STARTPOS. */
3891 if (startpos < 0 || startpos > total_size)
3894 /* Fix up RANGE if it might eventually take us outside
3895 the virtual concatenation of STRING1 and STRING2.
3896 Make sure we won't move STARTPOS below 0 or above TOTAL_SIZE. */
3898 range = 0 - startpos;
3899 else if (endpos > total_size)
3900 range = total_size - startpos;
3902 /* If the search isn't to be a backwards one, don't waste time in a
3903 search for a pattern that must be anchored. */
3904 if (bufp->used > 0 && range > 0
3905 && ((re_opcode_t) bufp->buffer[0] == begbuf
3906 /* `begline' is like `begbuf' if it cannot match at newlines. */
3907 || ((re_opcode_t) bufp->buffer[0] == begline
3908 && !bufp->newline_anchor)))
3917 /* In a forward search for something that starts with \=.
3918 don't keep searching past point. */
3919 if (bufp->used > 0 && (re_opcode_t) bufp->buffer[0] == at_dot && range > 0)
3921 range = PT - startpos;
3927 /* Update the fastmap now if not correct already. */
3928 if (fastmap && !bufp->fastmap_accurate)
3929 if (re_compile_fastmap (bufp) == -2)
3932 /* Loop through the string, looking for a place to start matching. */
3935 /* If a fastmap is supplied, skip quickly over characters that
3936 cannot be the start of a match. If the pattern can match the
3937 null string, however, we don't need to skip characters; we want
3938 the first null string. */
3939 if (fastmap && startpos < total_size && !bufp->can_be_null)
3941 if (range > 0) /* Searching forwards. */
3943 register const char *d;
3944 register int lim = 0;
3947 if (startpos < size1 && startpos + range >= size1)
3948 lim = range - (size1 - startpos);
3950 d = (startpos >= size1 ? string2 - size1 : string1) + startpos;
3952 /* Written out as an if-else to avoid testing `translate'
3956 && !fastmap[(unsigned char)
3957 translate[(unsigned char) *d++]])
3960 while (range > lim && !fastmap[(unsigned char) *d++])
3963 startpos += irange - range;
3965 else /* Searching backwards. */
3967 register char c = (size1 == 0 || startpos >= size1
3968 ? string2[startpos - size1]
3969 : string1[startpos]);
3971 if (!fastmap[(unsigned char) TRANSLATE (c)])
3976 /* If can't match the null string, and that's all we have left, fail. */
3977 if (range >= 0 && startpos == total_size && fastmap
3978 && !bufp->can_be_null)
3981 val = re_match_2_internal (bufp, string1, size1, string2, size2,
3982 startpos, regs, stop);
3983 #ifndef REGEX_MALLOC
4012 weak_alias (__re_search_2, re_search_2)
4015 /* This converts PTR, a pointer into one of the search strings `string1'
4016 and `string2' into an offset from the beginning of that string. */
4017 #define POINTER_TO_OFFSET(ptr) \
4018 (FIRST_STRING_P (ptr) \
4019 ? ((regoff_t) ((ptr) - string1)) \
4020 : ((regoff_t) ((ptr) - string2 + size1)))
4022 /* Macros for dealing with the split strings in re_match_2. */
4024 #define MATCHING_IN_FIRST_STRING (dend == end_match_1)
4026 /* Call before fetching a character with *d. This switches over to
4027 string2 if necessary. */
4028 #define PREFETCH() \
4031 /* End of string2 => fail. */ \
4032 if (dend == end_match_2) \
4034 /* End of string1 => advance to string2. */ \
4036 dend = end_match_2; \
4040 /* Test if at very beginning or at very end of the virtual concatenation
4041 of `string1' and `string2'. If only one string, it's `string2'. */
4042 #define AT_STRINGS_BEG(d) ((d) == (size1 ? string1 : string2) || !size2)
4043 #define AT_STRINGS_END(d) ((d) == end2)
4046 /* Test if D points to a character which is word-constituent. We have
4047 two special cases to check for: if past the end of string1, look at
4048 the first character in string2; and if before the beginning of
4049 string2, look at the last character in string1. */
4050 #define WORDCHAR_P(d) \
4051 (SYNTAX ((d) == end1 ? *string2 \
4052 : (d) == string2 - 1 ? *(end1 - 1) : *(d)) \
4055 /* Disabled due to a compiler bug -- see comment at case wordbound */
4057 /* Test if the character before D and the one at D differ with respect
4058 to being word-constituent. */
4059 #define AT_WORD_BOUNDARY(d) \
4060 (AT_STRINGS_BEG (d) || AT_STRINGS_END (d) \
4061 || WORDCHAR_P (d - 1) != WORDCHAR_P (d))
4064 /* Free everything we malloc. */
4065 #ifdef MATCH_MAY_ALLOCATE
4066 # define FREE_VAR(var) if (var) REGEX_FREE (var); var = NULL
4067 # define FREE_VARIABLES() \
4069 REGEX_FREE_STACK (fail_stack.stack); \
4070 FREE_VAR (regstart); \
4071 FREE_VAR (regend); \
4072 FREE_VAR (old_regstart); \
4073 FREE_VAR (old_regend); \
4074 FREE_VAR (best_regstart); \
4075 FREE_VAR (best_regend); \
4076 FREE_VAR (reg_info); \
4077 FREE_VAR (reg_dummy); \
4078 FREE_VAR (reg_info_dummy); \
4081 # define FREE_VARIABLES() ((void)0) /* Do nothing! But inhibit gcc warning. */
4082 #endif /* not MATCH_MAY_ALLOCATE */
4084 /* These values must meet several constraints. They must not be valid
4085 register values; since we have a limit of 255 registers (because
4086 we use only one byte in the pattern for the register number), we can
4087 use numbers larger than 255. They must differ by 1, because of
4088 NUM_FAILURE_ITEMS above. And the value for the lowest register must
4089 be larger than the value for the highest register, so we do not try
4090 to actually save any registers when none are active. */
4091 #define NO_HIGHEST_ACTIVE_REG (1 << BYTEWIDTH)
4092 #define NO_LOWEST_ACTIVE_REG (NO_HIGHEST_ACTIVE_REG + 1)
4094 /* Matching routines. */
4096 #ifndef emacs /* Emacs never uses this. */
4097 /* re_match is like re_match_2 except it takes only a single string. */
4101 struct re_pattern_buffer *bufp,
4104 struct re_registers *regs)
4106 int result = re_match_2_internal (bufp, NULL, 0, string, size,
4108 # ifndef REGEX_MALLOC
4116 weak_alias (__re_match, re_match)
4118 #endif /* not emacs */
4120 static boolean group_match_null_string_p _RE_ARGS ((unsigned char **p,
4122 register_info_type *reg_info));
4123 static boolean alt_match_null_string_p _RE_ARGS ((unsigned char *p,
4125 register_info_type *reg_info));
4126 static boolean common_op_match_null_string_p _RE_ARGS ((unsigned char **p,
4128 register_info_type *reg_info));
4129 static int bcmp_translate _RE_ARGS ((const char *s1, const char *s2,
4130 int len, char *translate));
4132 /* re_match_2 matches the compiled pattern in BUFP against the
4133 the (virtual) concatenation of STRING1 and STRING2 (of length SIZE1
4134 and SIZE2, respectively). We start matching at POS, and stop
4137 If REGS is non-null and the `no_sub' field of BUFP is nonzero, we
4138 store offsets for the substring each group matched in REGS. See the
4139 documentation for exactly how many groups we fill.
4141 We return -1 if no match, -2 if an internal error (such as the
4142 failure stack overflowing). Otherwise, we return the length of the
4143 matched substring. */
4146 struct re_pattern_buffer *bufp,
4147 const char *string1, const char *string2,
4148 int size1, int size2,
4150 struct re_registers *regs,
4153 int result = re_match_2_internal (bufp, string1, size1, string2, size2,
4155 #ifndef REGEX_MALLOC
4163 weak_alias (__re_match_2, re_match_2)
4166 /* This is a separate function so that we can force an alloca cleanup
4168 static int re_match_2_internal (
4169 struct re_pattern_buffer *bufp,
4170 const char *string1, int size1,
4171 const char *string2, int size2,
4173 struct re_registers *regs,
4176 /* General temporaries. */
4180 /* Just past the end of the corresponding string. */
4181 const char *end1, *end2;
4183 /* Pointers into string1 and string2, just past the last characters in
4184 each to consider matching. */
4185 const char *end_match_1, *end_match_2;
4187 /* Where we are in the data, and the end of the current string. */
4188 const char *d, *dend;
4190 /* Where we are in the pattern, and the end of the pattern. */
4191 unsigned char *p = bufp->buffer;
4192 register unsigned char *pend = p + bufp->used;
4194 /* Mark the opcode just after a start_memory, so we can test for an
4195 empty subpattern when we get to the stop_memory. */
4196 unsigned char *just_past_start_mem = 0;
4198 /* We use this to map every character in the string. */
4199 RE_TRANSLATE_TYPE translate = bufp->translate;
4201 /* Failure point stack. Each place that can handle a failure further
4202 down the line pushes a failure point on this stack. It consists of
4203 restart, regend, and reg_info for all registers corresponding to
4204 the subexpressions we're currently inside, plus the number of such
4205 registers, and, finally, two char *'s. The first char * is where
4206 to resume scanning the pattern; the second one is where to resume
4207 scanning the strings. If the latter is zero, the failure point is
4208 a ``dummy''; if a failure happens and the failure point is a dummy,
4209 it gets discarded and the next next one is tried. */
4210 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
4211 fail_stack_type fail_stack;
4214 static unsigned failure_id;
4215 unsigned nfailure_points_pushed = 0, nfailure_points_popped = 0;
4219 /* This holds the pointer to the failure stack, when
4220 it is allocated relocatably. */
4221 fail_stack_elt_t *failure_stack_ptr;
4224 /* We fill all the registers internally, independent of what we
4225 return, for use in backreferences. The number here includes
4226 an element for register zero. */
4227 size_t num_regs = bufp->re_nsub + 1;
4229 /* The currently active registers. */
4230 active_reg_t lowest_active_reg = NO_LOWEST_ACTIVE_REG;
4231 active_reg_t highest_active_reg = NO_HIGHEST_ACTIVE_REG;
4233 /* Information on the contents of registers. These are pointers into
4234 the input strings; they record just what was matched (on this
4235 attempt) by a subexpression part of the pattern, that is, the
4236 regnum-th regstart pointer points to where in the pattern we began
4237 matching and the regnum-th regend points to right after where we
4238 stopped matching the regnum-th subexpression. (The zeroth register
4239 keeps track of what the whole pattern matches.) */
4240 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
4241 const char **regstart, **regend;
4244 /* If a group that's operated upon by a repetition operator fails to
4245 match anything, then the register for its start will need to be
4246 restored because it will have been set to wherever in the string we
4247 are when we last see its open-group operator. Similarly for a
4249 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
4250 const char **old_regstart, **old_regend;
4253 /* The is_active field of reg_info helps us keep track of which (possibly
4254 nested) subexpressions we are currently in. The matched_something
4255 field of reg_info[reg_num] helps us tell whether or not we have
4256 matched any of the pattern so far this time through the reg_num-th
4257 subexpression. These two fields get reset each time through any
4258 loop their register is in. */
4259 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
4260 register_info_type *reg_info;
4263 /* The following record the register info as found in the above
4264 variables when we find a match better than any we've seen before.
4265 This happens as we backtrack through the failure points, which in
4266 turn happens only if we have not yet matched the entire string. */
4267 unsigned best_regs_set = false;
4268 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
4269 const char **best_regstart, **best_regend;
4272 /* Logically, this is `best_regend[0]'. But we don't want to have to
4273 allocate space for that if we're not allocating space for anything
4274 else (see below). Also, we never need info about register 0 for
4275 any of the other register vectors, and it seems rather a kludge to
4276 treat `best_regend' differently than the rest. So we keep track of
4277 the end of the best match so far in a separate variable. We
4278 initialize this to NULL so that when we backtrack the first time
4279 and need to test it, it's not garbage. */
4280 const char *match_end = NULL;
4282 /* This helps SET_REGS_MATCHED avoid doing redundant work. */
4283 int set_regs_matched_done = 0;
4285 /* Used when we pop values we don't care about. */
4286 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
4287 const char **reg_dummy;
4288 register_info_type *reg_info_dummy;
4292 /* Counts the total number of registers pushed. */
4293 unsigned num_regs_pushed = 0;
4296 DEBUG_PRINT1 ("\n\nEntering re_match_2.\n");
4300 #ifdef MATCH_MAY_ALLOCATE
4301 /* Do not bother to initialize all the register variables if there are
4302 no groups in the pattern, as it takes a fair amount of time. If
4303 there are groups, we include space for register 0 (the whole
4304 pattern), even though we never use it, since it simplifies the
4305 array indexing. We should fix this. */
4308 regstart = REGEX_TALLOC (num_regs, const char *);
4309 regend = REGEX_TALLOC (num_regs, const char *);
4310 old_regstart = REGEX_TALLOC (num_regs, const char *);
4311 old_regend = REGEX_TALLOC (num_regs, const char *);
4312 best_regstart = REGEX_TALLOC (num_regs, const char *);
4313 best_regend = REGEX_TALLOC (num_regs, const char *);
4314 reg_info = REGEX_TALLOC (num_regs, register_info_type);
4315 reg_dummy = REGEX_TALLOC (num_regs, const char *);
4316 reg_info_dummy = REGEX_TALLOC (num_regs, register_info_type);
4318 if (!(regstart && regend && old_regstart && old_regend && reg_info
4319 && best_regstart && best_regend && reg_dummy && reg_info_dummy))
4327 /* We must initialize all our variables to NULL, so that
4328 `FREE_VARIABLES' doesn't try to free them. */
4329 regstart = regend = old_regstart = old_regend = best_regstart
4330 = best_regend = reg_dummy = NULL;
4331 reg_info = reg_info_dummy = (register_info_type *) NULL;
4333 #endif /* MATCH_MAY_ALLOCATE */
4335 /* The starting position is bogus. */
4336 if (pos < 0 || pos > size1 + size2)
4342 /* Initialize subexpression text positions to -1 to mark ones that no
4343 start_memory/stop_memory has been seen for. Also initialize the
4344 register information struct. */
4345 for (mcnt = 1; (unsigned) mcnt < num_regs; mcnt++)
4347 regstart[mcnt] = regend[mcnt]
4348 = old_regstart[mcnt] = old_regend[mcnt] = REG_UNSET_VALUE;
4350 REG_MATCH_NULL_STRING_P (reg_info[mcnt]) = MATCH_NULL_UNSET_VALUE;
4351 IS_ACTIVE (reg_info[mcnt]) = 0;
4352 MATCHED_SOMETHING (reg_info[mcnt]) = 0;
4353 EVER_MATCHED_SOMETHING (reg_info[mcnt]) = 0;
4356 /* We move `string1' into `string2' if the latter's empty -- but not if
4357 `string1' is null. */
4358 if (size2 == 0 && string1 != NULL)
4365 end1 = string1 + size1;
4366 end2 = string2 + size2;
4368 /* Compute where to stop matching, within the two strings. */
4371 end_match_1 = string1 + stop;
4372 end_match_2 = string2;
4377 end_match_2 = string2 + stop - size1;
4380 /* `p' scans through the pattern as `d' scans through the data.
4381 `dend' is the end of the input string that `d' points within. `d'
4382 is advanced into the following input string whenever necessary, but
4383 this happens before fetching; therefore, at the beginning of the
4384 loop, `d' can be pointing at the end of a string, but it cannot
4386 if (size1 > 0 && pos <= size1)
4393 d = string2 + pos - size1;
4397 DEBUG_PRINT1 ("The compiled pattern is:\n");
4398 DEBUG_PRINT_COMPILED_PATTERN (bufp, p, pend);
4399 DEBUG_PRINT1 ("The string to match is: `");
4400 DEBUG_PRINT_DOUBLE_STRING (d, string1, size1, string2, size2);
4401 DEBUG_PRINT1 ("'\n");
4403 /* This loops over pattern commands. It exits by returning from the
4404 function if the match is complete, or it drops through if the match
4405 fails at this starting point in the input data. */
4410 DEBUG_PRINT2 ("\n%p: ", p);
4412 DEBUG_PRINT2 ("\n0x%x: ", p);
4416 { /* End of pattern means we might have succeeded. */
4417 DEBUG_PRINT1 ("end of pattern ... ");
4419 /* If we haven't matched the entire string, and we want the
4420 longest match, try backtracking. */
4421 if (d != end_match_2)
4423 /* 1 if this match ends in the same string (string1 or string2)
4424 as the best previous match. */
4425 same_str_p = (FIRST_STRING_P (match_end)
4426 == MATCHING_IN_FIRST_STRING);
4427 /* 1 if this match is the best seen so far. */
4428 boolean best_match_p;
4430 /* AIX compiler got confused when this was combined
4431 with the previous declaration. */
4433 best_match_p = d > match_end;
4435 best_match_p = !MATCHING_IN_FIRST_STRING;
4437 DEBUG_PRINT1 ("backtracking.\n");
4439 if (!FAIL_STACK_EMPTY ())
4440 { /* More failure points to try. */
4442 /* If exceeds best match so far, save it. */
4443 if (!best_regs_set || best_match_p)
4445 best_regs_set = true;
4448 DEBUG_PRINT1 ("\nSAVING match as best so far.\n");
4450 for (mcnt = 1; (unsigned) mcnt < num_regs; mcnt++)
4452 best_regstart[mcnt] = regstart[mcnt];
4453 best_regend[mcnt] = regend[mcnt];
4459 /* If no failure points, don't restore garbage. And if
4460 last match is real best match, don't restore second
4462 else if (best_regs_set && !best_match_p)
4465 /* Restore best match. It may happen that `dend ==
4466 end_match_1' while the restored d is in string2.
4467 For example, the pattern `x.*y.*z' against the
4468 strings `x-' and `y-z-', if the two strings are
4469 not consecutive in memory. */
4470 DEBUG_PRINT1 ("Restoring best registers.\n");
4473 dend = ((d >= string1 && d <= end1)
4474 ? end_match_1 : end_match_2);
4476 for (mcnt = 1; (unsigned) mcnt < num_regs; mcnt++)
4478 regstart[mcnt] = best_regstart[mcnt];
4479 regend[mcnt] = best_regend[mcnt];
4482 } /* d != end_match_2 */
4485 DEBUG_PRINT1 ("Accepting match.\n");
4487 /* If caller wants register contents data back, do it. */
4488 if (regs && !bufp->no_sub)
4490 /* Have the register data arrays been allocated? */
4491 if (bufp->regs_allocated == REGS_UNALLOCATED)
4492 { /* No. So allocate them with malloc. We need one
4493 extra element beyond `num_regs' for the `-1' marker
4495 regs->num_regs = MAX (RE_NREGS, num_regs + 1);
4496 regs->start = TALLOC (regs->num_regs, regoff_t);
4497 regs->end = TALLOC (regs->num_regs, regoff_t);
4498 if (regs->start == NULL || regs->end == NULL)
4503 bufp->regs_allocated = REGS_REALLOCATE;
4505 else if (bufp->regs_allocated == REGS_REALLOCATE)
4506 { /* Yes. If we need more elements than were already
4507 allocated, reallocate them. If we need fewer, just
4509 if (regs->num_regs < num_regs + 1)
4511 regs->num_regs = num_regs + 1;
4512 RETALLOC (regs->start, regs->num_regs, regoff_t);
4513 RETALLOC (regs->end, regs->num_regs, regoff_t);
4514 if (regs->start == NULL || regs->end == NULL)
4523 /* These braces fend off a "empty body in an else-statement"
4524 warning under GCC when assert expands to nothing. */
4525 assert (bufp->regs_allocated == REGS_FIXED);
4528 /* Convert the pointer data in `regstart' and `regend' to
4529 indices. Register zero has to be set differently,
4530 since we haven't kept track of any info for it. */
4531 if (regs->num_regs > 0)
4533 regs->start[0] = pos;
4534 regs->end[0] = (MATCHING_IN_FIRST_STRING
4535 ? ((regoff_t) (d - string1))
4536 : ((regoff_t) (d - string2 + size1)));
4539 /* Go through the first `min (num_regs, regs->num_regs)'
4540 registers, since that is all we initialized. */
4541 for (mcnt = 1; (unsigned) mcnt < MIN (num_regs, regs->num_regs);
4544 if (REG_UNSET (regstart[mcnt]) || REG_UNSET (regend[mcnt]))
4545 regs->start[mcnt] = regs->end[mcnt] = -1;
4549 = (regoff_t) POINTER_TO_OFFSET (regstart[mcnt]);
4551 = (regoff_t) POINTER_TO_OFFSET (regend[mcnt]);
4555 /* If the regs structure we return has more elements than
4556 were in the pattern, set the extra elements to -1. If
4557 we (re)allocated the registers, this is the case,
4558 because we always allocate enough to have at least one
4560 for (mcnt = num_regs; (unsigned) mcnt < regs->num_regs; mcnt++)
4561 regs->start[mcnt] = regs->end[mcnt] = -1;
4562 } /* regs && !bufp->no_sub */
4564 DEBUG_PRINT4 ("%u failure points pushed, %u popped (%u remain).\n",
4565 nfailure_points_pushed, nfailure_points_popped,
4566 nfailure_points_pushed - nfailure_points_popped);
4567 DEBUG_PRINT2 ("%u registers pushed.\n", num_regs_pushed);
4569 mcnt = d - pos - (MATCHING_IN_FIRST_STRING
4573 DEBUG_PRINT2 ("Returning %d from re_match_2.\n", mcnt);
4579 /* Otherwise match next pattern command. */
4580 switch (SWITCH_ENUM_CAST ((re_opcode_t) *p++))
4582 /* Ignore these. Used to ignore the n of succeed_n's which
4583 currently have n == 0. */
4585 DEBUG_PRINT1 ("EXECUTING no_op.\n");
4589 DEBUG_PRINT1 ("EXECUTING succeed.\n");
4592 /* Match the next n pattern characters exactly. The following
4593 byte in the pattern defines n, and the n bytes after that
4594 are the characters to match. */
4597 DEBUG_PRINT2 ("EXECUTING exactn %d.\n", mcnt);
4599 /* This is written out as an if-else so we don't waste time
4600 testing `translate' inside the loop. */
4606 if ((unsigned char) translate[(unsigned char) *d++]
4607 != (unsigned char) *p++)
4617 if (*d++ != (char) *p++) goto fail;
4621 SET_REGS_MATCHED ();
4625 /* Match any character except possibly a newline or a null. */
4627 DEBUG_PRINT1 ("EXECUTING anychar.\n");
4631 if ((!(bufp->syntax & RE_DOT_NEWLINE) && TRANSLATE (*d) == '\n')
4632 || (bufp->syntax & RE_DOT_NOT_NULL && TRANSLATE (*d) == '\000'))
4635 SET_REGS_MATCHED ();
4636 DEBUG_PRINT2 (" Matched `%d'.\n", *d);
4644 register unsigned char c;
4645 boolean bnot = (re_opcode_t) *(p - 1) == charset_not;
4647 DEBUG_PRINT2 ("EXECUTING charset%s.\n", bnot ? "_not" : "");
4650 c = TRANSLATE (*d); /* The character to match. */
4652 /* Cast to `unsigned' instead of `unsigned char' in case the
4653 bit list is a full 32 bytes long. */
4654 if (c < (unsigned) (*p * BYTEWIDTH)
4655 && p[1 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH)))
4660 if (!bnot) goto fail;
4662 SET_REGS_MATCHED ();
4668 /* The beginning of a group is represented by start_memory.
4669 The arguments are the register number in the next byte, and the
4670 number of groups inner to this one in the next. The text
4671 matched within the group is recorded (in the internal
4672 registers data structure) under the register number. */
4674 DEBUG_PRINT3 ("EXECUTING start_memory %d (%d):\n", *p, p[1]);
4676 /* Find out if this group can match the empty string. */
4677 p1 = p; /* To send to group_match_null_string_p. */
4679 if (REG_MATCH_NULL_STRING_P (reg_info[*p]) == MATCH_NULL_UNSET_VALUE)
4680 REG_MATCH_NULL_STRING_P (reg_info[*p])
4681 = group_match_null_string_p (&p1, pend, reg_info);
4683 /* Save the position in the string where we were the last time
4684 we were at this open-group operator in case the group is
4685 operated upon by a repetition operator, e.g., with `(a*)*b'
4686 against `ab'; then we want to ignore where we are now in
4687 the string in case this attempt to match fails. */
4688 old_regstart[*p] = REG_MATCH_NULL_STRING_P (reg_info[*p])
4689 ? REG_UNSET (regstart[*p]) ? d : regstart[*p]
4691 DEBUG_PRINT2 (" old_regstart: %d\n",
4692 POINTER_TO_OFFSET (old_regstart[*p]));
4695 DEBUG_PRINT2 (" regstart: %d\n", POINTER_TO_OFFSET (regstart[*p]));
4697 IS_ACTIVE (reg_info[*p]) = 1;
4698 MATCHED_SOMETHING (reg_info[*p]) = 0;
4700 /* Clear this whenever we change the register activity status. */
4701 set_regs_matched_done = 0;
4703 /* This is the new highest active register. */
4704 highest_active_reg = *p;
4706 /* If nothing was active before, this is the new lowest active
4708 if (lowest_active_reg == NO_LOWEST_ACTIVE_REG)
4709 lowest_active_reg = *p;
4711 /* Move past the register number and inner group count. */
4713 just_past_start_mem = p;
4718 /* The stop_memory opcode represents the end of a group. Its
4719 arguments are the same as start_memory's: the register
4720 number, and the number of inner groups. */
4722 DEBUG_PRINT3 ("EXECUTING stop_memory %d (%d):\n", *p, p[1]);
4724 /* We need to save the string position the last time we were at
4725 this close-group operator in case the group is operated
4726 upon by a repetition operator, e.g., with `((a*)*(b*)*)*'
4727 against `aba'; then we want to ignore where we are now in
4728 the string in case this attempt to match fails. */
4729 old_regend[*p] = REG_MATCH_NULL_STRING_P (reg_info[*p])
4730 ? REG_UNSET (regend[*p]) ? d : regend[*p]
4732 DEBUG_PRINT2 (" old_regend: %d\n",
4733 POINTER_TO_OFFSET (old_regend[*p]));
4736 DEBUG_PRINT2 (" regend: %d\n", POINTER_TO_OFFSET (regend[*p]));
4738 /* This register isn't active anymore. */
4739 IS_ACTIVE (reg_info[*p]) = 0;
4741 /* Clear this whenever we change the register activity status. */
4742 set_regs_matched_done = 0;
4744 /* If this was the only register active, nothing is active
4746 if (lowest_active_reg == highest_active_reg)
4748 lowest_active_reg = NO_LOWEST_ACTIVE_REG;
4749 highest_active_reg = NO_HIGHEST_ACTIVE_REG;
4752 { /* We must scan for the new highest active register, since
4753 it isn't necessarily one less than now: consider
4754 (a(b)c(d(e)f)g). When group 3 ends, after the f), the
4755 new highest active register is 1. */
4756 unsigned char r = *p - 1;
4757 while (r > 0 && !IS_ACTIVE (reg_info[r]))
4760 /* If we end up at register zero, that means that we saved
4761 the registers as the result of an `on_failure_jump', not
4762 a `start_memory', and we jumped to past the innermost
4763 `stop_memory'. For example, in ((.)*) we save
4764 registers 1 and 2 as a result of the *, but when we pop
4765 back to the second ), we are at the stop_memory 1.
4766 Thus, nothing is active. */
4769 lowest_active_reg = NO_LOWEST_ACTIVE_REG;
4770 highest_active_reg = NO_HIGHEST_ACTIVE_REG;
4773 highest_active_reg = r;
4776 /* If just failed to match something this time around with a
4777 group that's operated on by a repetition operator, try to
4778 force exit from the ``loop'', and restore the register
4779 information for this group that we had before trying this
4781 if ((!MATCHED_SOMETHING (reg_info[*p])
4782 || just_past_start_mem == p - 1)
4785 boolean is_a_jump_n = false;
4789 switch ((re_opcode_t) *p1++)
4793 case pop_failure_jump:
4794 case maybe_pop_jump:
4796 case dummy_failure_jump:
4797 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4807 /* If the next operation is a jump backwards in the pattern
4808 to an on_failure_jump right before the start_memory
4809 corresponding to this stop_memory, exit from the loop
4810 by forcing a failure after pushing on the stack the
4811 on_failure_jump's jump in the pattern, and d. */
4812 if (mcnt < 0 && (re_opcode_t) *p1 == on_failure_jump
4813 && (re_opcode_t) p1[3] == start_memory && p1[4] == *p)
4815 /* If this group ever matched anything, then restore
4816 what its registers were before trying this last
4817 failed match, e.g., with `(a*)*b' against `ab' for
4818 regstart[1], and, e.g., with `((a*)*(b*)*)*'
4819 against `aba' for regend[3].
4821 Also restore the registers for inner groups for,
4822 e.g., `((a*)(b*))*' against `aba' (register 3 would
4823 otherwise get trashed). */
4825 if (EVER_MATCHED_SOMETHING (reg_info[*p]))
4829 EVER_MATCHED_SOMETHING (reg_info[*p]) = 0;
4831 /* Restore this and inner groups' (if any) registers. */
4832 for (r = *p; r < (unsigned) *p + (unsigned) *(p + 1);
4835 regstart[r] = old_regstart[r];
4837 /* xx why this test? */
4838 if (old_regend[r] >= regstart[r])
4839 regend[r] = old_regend[r];
4843 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4844 PUSH_FAILURE_POINT (p1 + mcnt, d, -2);
4850 /* Move past the register number and the inner group count. */
4855 /* \<digit> has been turned into a `duplicate' command which is
4856 followed by the numeric value of <digit> as the register number. */
4859 register const char *d2, *dend2;
4860 int regno = *p++; /* Get which register to match against. */
4861 DEBUG_PRINT2 ("EXECUTING duplicate %d.\n", regno);
4863 /* Can't back reference a group which we've never matched. */
4864 if (REG_UNSET (regstart[regno]) || REG_UNSET (regend[regno]))
4867 /* Where in input to try to start matching. */
4868 d2 = regstart[regno];
4870 /* Where to stop matching; if both the place to start and
4871 the place to stop matching are in the same string, then
4872 set to the place to stop, otherwise, for now have to use
4873 the end of the first string. */
4875 dend2 = ((FIRST_STRING_P (regstart[regno])
4876 == FIRST_STRING_P (regend[regno]))
4877 ? regend[regno] : end_match_1);
4880 /* If necessary, advance to next segment in register
4884 if (dend2 == end_match_2) break;
4885 if (dend2 == regend[regno]) break;
4887 /* End of string1 => advance to string2. */
4889 dend2 = regend[regno];
4891 /* At end of register contents => success */
4892 if (d2 == dend2) break;
4894 /* If necessary, advance to next segment in data. */
4897 /* How many characters left in this segment to match. */
4900 /* Want how many consecutive characters we can match in
4901 one shot, so, if necessary, adjust the count. */
4902 if (mcnt > dend2 - d2)
4905 /* Compare that many; failure if mismatch, else move
4908 ? bcmp_translate (d, d2, mcnt, translate)
4909 : memcmp (d, d2, mcnt))
4911 d += mcnt, d2 += mcnt;
4913 /* Do this because we've match some characters. */
4914 SET_REGS_MATCHED ();
4920 /* begline matches the empty string at the beginning of the string
4921 (unless `not_bol' is set in `bufp'), and, if
4922 `newline_anchor' is set, after newlines. */
4924 DEBUG_PRINT1 ("EXECUTING begline.\n");
4926 if (AT_STRINGS_BEG (d))
4928 if (!bufp->not_bol) break;
4930 else if (d[-1] == '\n' && bufp->newline_anchor)
4934 /* In all other cases, we fail. */
4938 /* endline is the dual of begline. */
4940 DEBUG_PRINT1 ("EXECUTING endline.\n");
4942 if (AT_STRINGS_END (d))
4944 if (!bufp->not_eol) break;
4947 /* We have to ``prefetch'' the next character. */
4948 else if ((d == end1 ? *string2 : *d) == '\n'
4949 && bufp->newline_anchor)
4956 /* Match at the very beginning of the data. */
4958 DEBUG_PRINT1 ("EXECUTING begbuf.\n");
4959 if (AT_STRINGS_BEG (d))
4964 /* Match at the very end of the data. */
4966 DEBUG_PRINT1 ("EXECUTING endbuf.\n");
4967 if (AT_STRINGS_END (d))
4972 /* on_failure_keep_string_jump is used to optimize `.*\n'. It
4973 pushes NULL as the value for the string on the stack. Then
4974 `pop_failure_point' will keep the current value for the
4975 string, instead of restoring it. To see why, consider
4976 matching `foo\nbar' against `.*\n'. The .* matches the foo;
4977 then the . fails against the \n. But the next thing we want
4978 to do is match the \n against the \n; if we restored the
4979 string value, we would be back at the foo.
4981 Because this is used only in specific cases, we don't need to
4982 check all the things that `on_failure_jump' does, to make
4983 sure the right things get saved on the stack. Hence we don't
4984 share its code. The only reason to push anything on the
4985 stack at all is that otherwise we would have to change
4986 `anychar's code to do something besides goto fail in this
4987 case; that seems worse than this. */
4988 case on_failure_keep_string_jump:
4989 DEBUG_PRINT1 ("EXECUTING on_failure_keep_string_jump");
4991 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4993 DEBUG_PRINT3 (" %d (to %p):\n", mcnt, p + mcnt);
4995 DEBUG_PRINT3 (" %d (to 0x%x):\n", mcnt, p + mcnt);
4998 PUSH_FAILURE_POINT (p + mcnt, NULL, -2);
5002 /* Uses of on_failure_jump:
5004 Each alternative starts with an on_failure_jump that points
5005 to the beginning of the next alternative. Each alternative
5006 except the last ends with a jump that in effect jumps past
5007 the rest of the alternatives. (They really jump to the
5008 ending jump of the following alternative, because tensioning
5009 these jumps is a hassle.)
5011 Repeats start with an on_failure_jump that points past both
5012 the repetition text and either the following jump or
5013 pop_failure_jump back to this on_failure_jump. */
5014 case on_failure_jump:
5016 DEBUG_PRINT1 ("EXECUTING on_failure_jump");
5018 EXTRACT_NUMBER_AND_INCR (mcnt, p);
5020 DEBUG_PRINT3 (" %d (to %p)", mcnt, p + mcnt);
5022 DEBUG_PRINT3 (" %d (to 0x%x)", mcnt, p + mcnt);
5025 /* If this on_failure_jump comes right before a group (i.e.,
5026 the original * applied to a group), save the information
5027 for that group and all inner ones, so that if we fail back
5028 to this point, the group's information will be correct.
5029 For example, in \(a*\)*\1, we need the preceding group,
5030 and in \(zz\(a*\)b*\)\2, we need the inner group. */
5032 /* We can't use `p' to check ahead because we push
5033 a failure point to `p + mcnt' after we do this. */
5036 /* We need to skip no_op's before we look for the
5037 start_memory in case this on_failure_jump is happening as
5038 the result of a completed succeed_n, as in \(a\)\{1,3\}b\1
5040 while (p1 < pend && (re_opcode_t) *p1 == no_op)
5043 if (p1 < pend && (re_opcode_t) *p1 == start_memory)
5045 /* We have a new highest active register now. This will
5046 get reset at the start_memory we are about to get to,
5047 but we will have saved all the registers relevant to
5048 this repetition op, as described above. */
5049 highest_active_reg = *(p1 + 1) + *(p1 + 2);
5050 if (lowest_active_reg == NO_LOWEST_ACTIVE_REG)
5051 lowest_active_reg = *(p1 + 1);
5054 DEBUG_PRINT1 (":\n");
5055 PUSH_FAILURE_POINT (p + mcnt, d, -2);
5059 /* A smart repeat ends with `maybe_pop_jump'.
5060 We change it to either `pop_failure_jump' or `jump'. */
5061 case maybe_pop_jump:
5062 EXTRACT_NUMBER_AND_INCR (mcnt, p);
5063 DEBUG_PRINT2 ("EXECUTING maybe_pop_jump %d.\n", mcnt);
5065 register unsigned char *p2 = p;
5067 /* Compare the beginning of the repeat with what in the
5068 pattern follows its end. If we can establish that there
5069 is nothing that they would both match, i.e., that we
5070 would have to backtrack because of (as in, e.g., `a*a')
5071 then we can change to pop_failure_jump, because we'll
5072 never have to backtrack.
5074 This is not true in the case of alternatives: in
5075 `(a|ab)*' we do need to backtrack to the `ab' alternative
5076 (e.g., if the string was `ab'). But instead of trying to
5077 detect that here, the alternative has put on a dummy
5078 failure point which is what we will end up popping. */
5080 /* Skip over open/close-group commands.
5081 If what follows this loop is a ...+ construct,
5082 look at what begins its body, since we will have to
5083 match at least one of that. */
5087 && ((re_opcode_t) *p2 == stop_memory
5088 || (re_opcode_t) *p2 == start_memory))
5090 else if (p2 + 6 < pend
5091 && (re_opcode_t) *p2 == dummy_failure_jump)
5098 /* p1[0] ... p1[2] are the `on_failure_jump' corresponding
5099 to the `maybe_finalize_jump' of this case. Examine what
5102 /* If we're at the end of the pattern, we can change. */
5105 /* Consider what happens when matching ":\(.*\)"
5106 against ":/". I don't really understand this code
5108 p[-3] = (unsigned char) pop_failure_jump;
5110 (" End of pattern: change to `pop_failure_jump'.\n");
5113 else if ((re_opcode_t) *p2 == exactn
5114 || (bufp->newline_anchor && (re_opcode_t) *p2 == endline))
5116 register unsigned char c
5117 = *p2 == (unsigned char) endline ? '\n' : p2[2];
5119 if ((re_opcode_t) p1[3] == exactn && p1[5] != c)
5121 p[-3] = (unsigned char) pop_failure_jump;
5122 DEBUG_PRINT3 (" %c != %c => pop_failure_jump.\n",
5126 else if ((re_opcode_t) p1[3] == charset
5127 || (re_opcode_t) p1[3] == charset_not)
5129 int bnot = (re_opcode_t) p1[3] == charset_not;
5131 if (c < (unsigned char) (p1[4] * BYTEWIDTH)
5132 && p1[5 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH)))
5135 /* `not' is equal to 1 if c would match, which means
5136 that we can't change to pop_failure_jump. */
5139 p[-3] = (unsigned char) pop_failure_jump;
5140 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
5144 else if ((re_opcode_t) *p2 == charset)
5146 /* We win if the first character of the loop is not part
5148 if ((re_opcode_t) p1[3] == exactn
5149 && ! ((int) p2[1] * BYTEWIDTH > (int) p1[5]
5150 && (p2[2 + p1[5] / BYTEWIDTH]
5151 & (1 << (p1[5] % BYTEWIDTH)))))
5153 p[-3] = (unsigned char) pop_failure_jump;
5154 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
5157 else if ((re_opcode_t) p1[3] == charset_not)
5160 /* We win if the charset_not inside the loop
5161 lists every character listed in the charset after. */
5162 for (idx = 0; idx < (int) p2[1]; idx++)
5163 if (! (p2[2 + idx] == 0
5164 || (idx < (int) p1[4]
5165 && ((p2[2 + idx] & ~ p1[5 + idx]) == 0))))
5170 p[-3] = (unsigned char) pop_failure_jump;
5171 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
5174 else if ((re_opcode_t) p1[3] == charset)
5177 /* We win if the charset inside the loop
5178 has no overlap with the one after the loop. */
5180 idx < (int) p2[1] && idx < (int) p1[4];
5182 if ((p2[2 + idx] & p1[5 + idx]) != 0)
5185 if (idx == p2[1] || idx == p1[4])
5187 p[-3] = (unsigned char) pop_failure_jump;
5188 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
5193 p -= 2; /* Point at relative address again. */
5194 if ((re_opcode_t) p[-1] != pop_failure_jump)
5196 p[-1] = (unsigned char) jump;
5197 DEBUG_PRINT1 (" Match => jump.\n");
5198 goto unconditional_jump;
5200 /* Note fall through. */
5203 /* The end of a simple repeat has a pop_failure_jump back to
5204 its matching on_failure_jump, where the latter will push a
5205 failure point. The pop_failure_jump takes off failure
5206 points put on by this pop_failure_jump's matching
5207 on_failure_jump; we got through the pattern to here from the
5208 matching on_failure_jump, so didn't fail. */
5209 case pop_failure_jump:
5211 /* We need to pass separate storage for the lowest and
5212 highest registers, even though we don't care about the
5213 actual values. Otherwise, we will restore only one
5214 register from the stack, since lowest will == highest in
5215 `pop_failure_point'. */
5216 active_reg_t dummy_low_reg, dummy_high_reg;
5217 unsigned char *pdummy;
5220 DEBUG_PRINT1 ("EXECUTING pop_failure_jump.\n");
5221 POP_FAILURE_POINT (sdummy, pdummy,
5222 dummy_low_reg, dummy_high_reg,
5223 reg_dummy, reg_dummy, reg_info_dummy);
5225 /* Note fall through. */
5229 DEBUG_PRINT2 ("\n%p: ", p);
5231 DEBUG_PRINT2 ("\n0x%x: ", p);
5233 /* Note fall through. */
5235 /* Unconditionally jump (without popping any failure points). */
5237 EXTRACT_NUMBER_AND_INCR (mcnt, p); /* Get the amount to jump. */
5238 DEBUG_PRINT2 ("EXECUTING jump %d ", mcnt);
5239 p += mcnt; /* Do the jump. */
5241 DEBUG_PRINT2 ("(to %p).\n", p);
5243 DEBUG_PRINT2 ("(to 0x%x).\n", p);
5248 /* We need this opcode so we can detect where alternatives end
5249 in `group_match_null_string_p' et al. */
5251 DEBUG_PRINT1 ("EXECUTING jump_past_alt.\n");
5252 goto unconditional_jump;
5255 /* Normally, the on_failure_jump pushes a failure point, which
5256 then gets popped at pop_failure_jump. We will end up at
5257 pop_failure_jump, also, and with a pattern of, say, `a+', we
5258 are skipping over the on_failure_jump, so we have to push
5259 something meaningless for pop_failure_jump to pop. */
5260 case dummy_failure_jump:
5261 DEBUG_PRINT1 ("EXECUTING dummy_failure_jump.\n");
5262 /* It doesn't matter what we push for the string here. What
5263 the code at `fail' tests is the value for the pattern. */
5264 PUSH_FAILURE_POINT (NULL, NULL, -2);
5265 goto unconditional_jump;
5268 /* At the end of an alternative, we need to push a dummy failure
5269 point in case we are followed by a `pop_failure_jump', because
5270 we don't want the failure point for the alternative to be
5271 popped. For example, matching `(a|ab)*' against `aab'
5272 requires that we match the `ab' alternative. */
5273 case push_dummy_failure:
5274 DEBUG_PRINT1 ("EXECUTING push_dummy_failure.\n");
5275 /* See comments just above at `dummy_failure_jump' about the
5277 PUSH_FAILURE_POINT (NULL, NULL, -2);
5280 /* Have to succeed matching what follows at least n times.
5281 After that, handle like `on_failure_jump'. */
5283 EXTRACT_NUMBER (mcnt, p + 2);
5284 DEBUG_PRINT2 ("EXECUTING succeed_n %d.\n", mcnt);
5287 /* Originally, this is how many times we HAVE to succeed. */
5292 STORE_NUMBER_AND_INCR (p, mcnt);
5294 DEBUG_PRINT3 (" Setting %p to %d.\n", p - 2, mcnt);
5296 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p - 2, mcnt);
5302 DEBUG_PRINT2 (" Setting two bytes from %p to no_op.\n", p+2);
5304 DEBUG_PRINT2 (" Setting two bytes from 0x%x to no_op.\n", p+2);
5306 p[2] = (unsigned char) no_op;
5307 p[3] = (unsigned char) no_op;
5313 EXTRACT_NUMBER (mcnt, p + 2);
5314 DEBUG_PRINT2 ("EXECUTING jump_n %d.\n", mcnt);
5316 /* Originally, this is how many times we CAN jump. */
5320 STORE_NUMBER (p + 2, mcnt);
5322 DEBUG_PRINT3 (" Setting %p to %d.\n", p + 2, mcnt);
5324 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p + 2, mcnt);
5326 goto unconditional_jump;
5328 /* If don't have to jump any more, skip over the rest of command. */
5335 DEBUG_PRINT1 ("EXECUTING set_number_at.\n");
5337 EXTRACT_NUMBER_AND_INCR (mcnt, p);
5339 EXTRACT_NUMBER_AND_INCR (mcnt, p);
5341 DEBUG_PRINT3 (" Setting %p to %d.\n", p1, mcnt);
5343 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p1, mcnt);
5345 STORE_NUMBER (p1, mcnt);
5350 /* The DEC Alpha C compiler 3.x generates incorrect code for the
5351 test WORDCHAR_P (d - 1) != WORDCHAR_P (d) in the expansion of
5352 AT_WORD_BOUNDARY, so this code is disabled. Expanding the
5353 macro and introducing temporary variables works around the bug. */
5356 DEBUG_PRINT1 ("EXECUTING wordbound.\n");
5357 if (AT_WORD_BOUNDARY (d))
5362 DEBUG_PRINT1 ("EXECUTING notwordbound.\n");
5363 if (AT_WORD_BOUNDARY (d))
5369 boolean prevchar, thischar;
5371 DEBUG_PRINT1 ("EXECUTING wordbound.\n");
5372 if (AT_STRINGS_BEG (d) || AT_STRINGS_END (d))
5375 prevchar = WORDCHAR_P (d - 1);
5376 thischar = WORDCHAR_P (d);
5377 if (prevchar != thischar)
5384 boolean prevchar, thischar;
5386 DEBUG_PRINT1 ("EXECUTING notwordbound.\n");
5387 if (AT_STRINGS_BEG (d) || AT_STRINGS_END (d))
5390 prevchar = WORDCHAR_P (d - 1);
5391 thischar = WORDCHAR_P (d);
5392 if (prevchar != thischar)
5399 DEBUG_PRINT1 ("EXECUTING wordbeg.\n");
5400 if (WORDCHAR_P (d) && (AT_STRINGS_BEG (d) || !WORDCHAR_P (d - 1)))
5405 DEBUG_PRINT1 ("EXECUTING wordend.\n");
5406 if (!AT_STRINGS_BEG (d) && WORDCHAR_P (d - 1)
5407 && (!WORDCHAR_P (d) || AT_STRINGS_END (d)))
5413 DEBUG_PRINT1 ("EXECUTING before_dot.\n");
5414 if (PTR_CHAR_POS ((unsigned char *) d) >= point)
5419 DEBUG_PRINT1 ("EXECUTING at_dot.\n");
5420 if (PTR_CHAR_POS ((unsigned char *) d) != point)
5425 DEBUG_PRINT1 ("EXECUTING after_dot.\n");
5426 if (PTR_CHAR_POS ((unsigned char *) d) <= point)
5431 DEBUG_PRINT2 ("EXECUTING syntaxspec %d.\n", mcnt);
5436 DEBUG_PRINT1 ("EXECUTING Emacs wordchar.\n");
5440 /* Can't use *d++ here; SYNTAX may be an unsafe macro. */
5442 if (SYNTAX (d[-1]) != (enum syntaxcode) mcnt)
5444 SET_REGS_MATCHED ();
5448 DEBUG_PRINT2 ("EXECUTING notsyntaxspec %d.\n", mcnt);
5450 goto matchnotsyntax;
5453 DEBUG_PRINT1 ("EXECUTING Emacs notwordchar.\n");
5457 /* Can't use *d++ here; SYNTAX may be an unsafe macro. */
5459 if (SYNTAX (d[-1]) == (enum syntaxcode) mcnt)
5461 SET_REGS_MATCHED ();
5464 #else /* not emacs */
5466 DEBUG_PRINT1 ("EXECUTING non-Emacs wordchar.\n");
5468 if (!WORDCHAR_P (d))
5470 SET_REGS_MATCHED ();
5475 DEBUG_PRINT1 ("EXECUTING non-Emacs notwordchar.\n");
5479 SET_REGS_MATCHED ();
5482 #endif /* not emacs */
5487 continue; /* Successfully executed one pattern command; keep going. */
5490 /* We goto here if a matching operation fails. */
5492 if (!FAIL_STACK_EMPTY ())
5493 { /* A restart point is known. Restore to that state. */
5494 DEBUG_PRINT1 ("\nFAIL:\n");
5495 POP_FAILURE_POINT (d, p,
5496 lowest_active_reg, highest_active_reg,
5497 regstart, regend, reg_info);
5499 /* If this failure point is a dummy, try the next one. */
5503 /* If we failed to the end of the pattern, don't examine *p. */
5507 boolean is_a_jump_n = false;
5509 /* If failed to a backwards jump that's part of a repetition
5510 loop, need to pop this failure point and use the next one. */
5511 switch ((re_opcode_t) *p)
5515 case maybe_pop_jump:
5516 case pop_failure_jump:
5519 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
5522 if ((is_a_jump_n && (re_opcode_t) *p1 == succeed_n)
5524 && (re_opcode_t) *p1 == on_failure_jump))
5532 if (d >= string1 && d <= end1)
5536 break; /* Matching at this starting point really fails. */
5540 goto restore_best_regs;
5544 return -1; /* Failure to match. */
5547 /* Subroutine definitions for re_match_2. */
5550 /* We are passed P pointing to a register number after a start_memory.
5552 Return true if the pattern up to the corresponding stop_memory can
5553 match the empty string, and false otherwise.
5555 If we find the matching stop_memory, sets P to point to one past its number.
5556 Otherwise, sets P to an undefined byte less than or equal to END.
5558 We don't handle duplicates properly (yet). */
5561 group_match_null_string_p (
5562 unsigned char **p, unsigned char *end,
5563 register_info_type *reg_info)
5566 /* Point to after the args to the start_memory. */
5567 unsigned char *p1 = *p + 2;
5571 /* Skip over opcodes that can match nothing, and return true or
5572 false, as appropriate, when we get to one that can't, or to the
5573 matching stop_memory. */
5575 switch ((re_opcode_t) *p1)
5577 /* Could be either a loop or a series of alternatives. */
5578 case on_failure_jump:
5580 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
5582 /* If the next operation is not a jump backwards in the
5587 /* Go through the on_failure_jumps of the alternatives,
5588 seeing if any of the alternatives cannot match nothing.
5589 The last alternative starts with only a jump,
5590 whereas the rest start with on_failure_jump and end
5591 with a jump, e.g., here is the pattern for `a|b|c':
5593 /on_failure_jump/0/6/exactn/1/a/jump_past_alt/0/6
5594 /on_failure_jump/0/6/exactn/1/b/jump_past_alt/0/3
5597 So, we have to first go through the first (n-1)
5598 alternatives and then deal with the last one separately. */
5601 /* Deal with the first (n-1) alternatives, which start
5602 with an on_failure_jump (see above) that jumps to right
5603 past a jump_past_alt. */
5605 while ((re_opcode_t) p1[mcnt-3] == jump_past_alt)
5607 /* `mcnt' holds how many bytes long the alternative
5608 is, including the ending `jump_past_alt' and
5611 if (!alt_match_null_string_p (p1, p1 + mcnt - 3,
5615 /* Move to right after this alternative, including the
5619 /* Break if it's the beginning of an n-th alternative
5620 that doesn't begin with an on_failure_jump. */
5621 if ((re_opcode_t) *p1 != on_failure_jump)
5624 /* Still have to check that it's not an n-th
5625 alternative that starts with an on_failure_jump. */
5627 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
5628 if ((re_opcode_t) p1[mcnt-3] != jump_past_alt)
5630 /* Get to the beginning of the n-th alternative. */
5636 /* Deal with the last alternative: go back and get number
5637 of the `jump_past_alt' just before it. `mcnt' contains
5638 the length of the alternative. */
5639 EXTRACT_NUMBER (mcnt, p1 - 2);
5641 if (!alt_match_null_string_p (p1, p1 + mcnt, reg_info))
5644 p1 += mcnt; /* Get past the n-th alternative. */
5650 assert (p1[1] == **p);
5656 if (!common_op_match_null_string_p (&p1, end, reg_info))
5659 } /* while p1 < end */
5662 } /* group_match_null_string_p */
5665 /* Similar to group_match_null_string_p, but doesn't deal with alternatives:
5666 It expects P to be the first byte of a single alternative and END one
5667 byte past the last. The alternative can contain groups. */
5670 alt_match_null_string_p (
5671 unsigned char *p, unsigned char *end,
5672 register_info_type *reg_info)
5675 unsigned char *p1 = p;
5679 /* Skip over opcodes that can match nothing, and break when we get
5680 to one that can't. */
5682 switch ((re_opcode_t) *p1)
5685 case on_failure_jump:
5687 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
5692 if (!common_op_match_null_string_p (&p1, end, reg_info))
5695 } /* while p1 < end */
5698 } /* alt_match_null_string_p */
5701 /* Deals with the ops common to group_match_null_string_p and
5702 alt_match_null_string_p.
5704 Sets P to one after the op and its arguments, if any. */
5707 common_op_match_null_string_p (
5708 unsigned char **p, unsigned char *end,
5709 register_info_type *reg_info)
5714 unsigned char *p1 = *p;
5716 switch ((re_opcode_t) *p1++)
5736 assert (reg_no > 0 && reg_no <= MAX_REGNUM);
5737 ret = group_match_null_string_p (&p1, end, reg_info);
5739 /* Have to set this here in case we're checking a group which
5740 contains a group and a back reference to it. */
5742 if (REG_MATCH_NULL_STRING_P (reg_info[reg_no]) == MATCH_NULL_UNSET_VALUE)
5743 REG_MATCH_NULL_STRING_P (reg_info[reg_no]) = ret;
5749 /* If this is an optimized succeed_n for zero times, make the jump. */
5751 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
5759 /* Get to the number of times to succeed. */
5761 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
5766 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
5774 if (!REG_MATCH_NULL_STRING_P (reg_info[*p1]))
5782 /* All other opcodes mean we cannot match the empty string. */
5788 } /* common_op_match_null_string_p */
5791 /* Return zero if TRANSLATE[S1] and TRANSLATE[S2] are identical for LEN
5792 bytes; nonzero otherwise. */
5796 const char *s1, const char *s2,
5798 RE_TRANSLATE_TYPE translate)
5800 register const unsigned char *p1 = (const unsigned char *) s1;
5801 register const unsigned char *p2 = (const unsigned char *) s2;
5804 if (translate[*p1++] != translate[*p2++]) return 1;
5810 /* Entry points for GNU code. */
5812 /* re_compile_pattern is the GNU regular expression compiler: it
5813 compiles PATTERN (of length SIZE) and puts the result in BUFP.
5814 Returns 0 if the pattern was valid, otherwise an error string.
5816 Assumes the `allocated' (and perhaps `buffer') and `translate' fields
5817 are set in BUFP on entry.
5819 We call regex_compile to do the actual compilation. */
5822 re_compile_pattern (
5823 const char *pattern,
5825 struct re_pattern_buffer *bufp)
5829 /* GNU code is written to assume at least RE_NREGS registers will be set
5830 (and at least one extra will be -1). */
5831 bufp->regs_allocated = REGS_UNALLOCATED;
5833 /* And GNU code determines whether or not to get register information
5834 by passing null for the REGS argument to re_match, etc., not by
5838 /* Match anchors at newline. */
5839 bufp->newline_anchor = 1;
5841 ret = regex_compile (pattern, length, re_syntax_options, bufp);
5845 return gettext (re_error_msgid + re_error_msgid_idx[(int) ret]);
5848 weak_alias (__re_compile_pattern, re_compile_pattern)
5851 /* Entry points compatible with 4.2 BSD regex library. We don't define
5852 them unless specifically requested. */
5854 #if defined _REGEX_RE_COMP || defined _LIBC
5856 /* BSD has one and only one pattern buffer. */
5857 static struct re_pattern_buffer re_comp_buf;
5861 /* Make these definitions weak in libc, so POSIX programs can redefine
5862 these names if they don't use our functions, and still use
5863 regcomp/regexec below without link errors. */
5873 if (!re_comp_buf.buffer)
5874 return gettext ("No previous regular expression");
5878 if (!re_comp_buf.buffer)
5880 re_comp_buf.buffer = (unsigned char *) malloc (200);
5881 if (re_comp_buf.buffer == NULL)
5882 return (char *) gettext (re_error_msgid
5883 + re_error_msgid_idx[(int) REG_ESPACE]);
5884 re_comp_buf.allocated = 200;
5886 re_comp_buf.fastmap = (char *) malloc (1 << BYTEWIDTH);
5887 if (re_comp_buf.fastmap == NULL)
5888 return (char *) gettext (re_error_msgid
5889 + re_error_msgid_idx[(int) REG_ESPACE]);
5892 /* Since `re_exec' always passes NULL for the `regs' argument, we
5893 don't need to initialize the pattern buffer fields which affect it. */
5895 /* Match anchors at newlines. */
5896 re_comp_buf.newline_anchor = 1;
5898 ret = regex_compile (s, strlen (s), re_syntax_options, &re_comp_buf);
5903 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
5904 return (char *) gettext (re_error_msgid + re_error_msgid_idx[(int) ret]);
5915 const int len = strlen (s);
5917 0 <= re_search (&re_comp_buf, s, len, 0, len, (struct re_registers *) 0);
5920 #endif /* _REGEX_RE_COMP */
5922 /* POSIX.2 functions. Don't define these for Emacs. */
5926 /* regcomp takes a regular expression as a string and compiles it.
5928 PREG is a regex_t *. We do not expect any fields to be initialized,
5929 since POSIX says we shouldn't. Thus, we set
5931 `buffer' to the compiled pattern;
5932 `used' to the length of the compiled pattern;
5933 `syntax' to RE_SYNTAX_POSIX_EXTENDED if the
5934 REG_EXTENDED bit in CFLAGS is set; otherwise, to
5935 RE_SYNTAX_POSIX_BASIC;
5936 `newline_anchor' to REG_NEWLINE being set in CFLAGS;
5937 `fastmap' to an allocated space for the fastmap;
5938 `fastmap_accurate' to zero;
5939 `re_nsub' to the number of subexpressions in PATTERN.
5941 PATTERN is the address of the pattern string.
5943 CFLAGS is a series of bits which affect compilation.
5945 If REG_EXTENDED is set, we use POSIX extended syntax; otherwise, we
5946 use POSIX basic syntax.
5948 If REG_NEWLINE is set, then . and [^...] don't match newline.
5949 Also, regexec will try a match beginning after every newline.
5951 If REG_ICASE is set, then we considers upper- and lowercase
5952 versions of letters to be equivalent when matching.
5954 If REG_NOSUB is set, then when PREG is passed to regexec, that
5955 routine will report only success or failure, and nothing about the
5958 It returns 0 if it succeeds, nonzero if it doesn't. (See regex.h for
5959 the return codes and their meanings.) */
5964 const char *pattern,
5969 = (cflags & REG_EXTENDED) ?
5970 RE_SYNTAX_POSIX_EXTENDED : RE_SYNTAX_POSIX_BASIC;
5972 /* regex_compile will allocate the space for the compiled pattern. */
5974 preg->allocated = 0;
5977 /* Try to allocate space for the fastmap. */
5978 preg->fastmap = (char *) malloc (1 << BYTEWIDTH);
5980 if (cflags & REG_ICASE)
5985 = (RE_TRANSLATE_TYPE) malloc (CHAR_SET_SIZE
5986 * sizeof (*(RE_TRANSLATE_TYPE)0));
5987 if (preg->translate == NULL)
5988 return (int) REG_ESPACE;
5990 /* Map uppercase characters to corresponding lowercase ones. */
5991 for (i = 0; i < CHAR_SET_SIZE; i++)
5992 preg->translate[i] = ISUPPER (i) ? TOLOWER (i) : i;
5995 preg->translate = NULL;
5997 /* If REG_NEWLINE is set, newlines are treated differently. */
5998 if (cflags & REG_NEWLINE)
5999 { /* REG_NEWLINE implies neither . nor [^...] match newline. */
6000 syntax &= ~RE_DOT_NEWLINE;
6001 syntax |= RE_HAT_LISTS_NOT_NEWLINE;
6002 /* It also changes the matching behavior. */
6003 preg->newline_anchor = 1;
6006 preg->newline_anchor = 0;
6008 preg->no_sub = !!(cflags & REG_NOSUB);
6010 /* POSIX says a null character in the pattern terminates it, so we
6011 can use strlen here in compiling the pattern. */
6012 ret = regex_compile (pattern, strlen (pattern), syntax, preg);
6014 /* POSIX doesn't distinguish between an unmatched open-group and an
6015 unmatched close-group: both are REG_EPAREN. */
6016 if (ret == REG_ERPAREN) ret = REG_EPAREN;
6018 if (ret == REG_NOERROR && preg->fastmap)
6020 /* Compute the fastmap now, since regexec cannot modify the pattern
6022 if (re_compile_fastmap (preg) == -2)
6024 /* Some error occurred while computing the fastmap, just forget
6026 free (preg->fastmap);
6027 preg->fastmap = NULL;
6034 weak_alias (__regcomp, regcomp)
6038 /* regexec searches for a given pattern, specified by PREG, in the
6041 If NMATCH is zero or REG_NOSUB was set in the cflags argument to
6042 `regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at
6043 least NMATCH elements, and we set them to the offsets of the
6044 corresponding matched substrings.
6046 EFLAGS specifies `execution flags' which affect matching: if
6047 REG_NOTBOL is set, then ^ does not match at the beginning of the
6048 string; if REG_NOTEOL is set, then $ does not match at the end.
6050 We return 0 if we find a match and REG_NOMATCH if not. */
6054 const regex_t *preg,
6057 regmatch_t pmatch[],
6061 struct re_registers regs;
6062 regex_t private_preg;
6063 int len = strlen (string);
6064 boolean want_reg_info = !preg->no_sub && nmatch > 0;
6066 private_preg = *preg;
6068 private_preg.not_bol = !!(eflags & REG_NOTBOL);
6069 private_preg.not_eol = !!(eflags & REG_NOTEOL);
6071 /* The user has told us exactly how many registers to return
6072 information about, via `nmatch'. We have to pass that on to the
6073 matching routines. */
6074 private_preg.regs_allocated = REGS_FIXED;
6078 regs.num_regs = nmatch;
6079 regs.start = TALLOC (nmatch * 2, regoff_t);
6080 if (regs.start == NULL)
6081 return (int) REG_NOMATCH;
6082 regs.end = regs.start + nmatch;
6085 /* Perform the searching operation. */
6086 ret = re_search (&private_preg, string, len,
6087 /* start: */ 0, /* range: */ len,
6088 want_reg_info ? ®s : (struct re_registers *) 0);
6090 /* Copy the register information to the POSIX structure. */
6097 for (r = 0; r < nmatch; r++)
6099 pmatch[r].rm_so = regs.start[r];
6100 pmatch[r].rm_eo = regs.end[r];
6104 /* If we needed the temporary register info, free the space now. */
6108 /* We want zero return to mean success, unlike `re_search'. */
6109 return ret >= 0 ? (int) REG_NOERROR : (int) REG_NOMATCH;
6112 weak_alias (__regexec, regexec)
6116 /* Returns a message corresponding to an error code, ERRCODE, returned
6117 from either regcomp or regexec. We don't use PREG here. */
6122 const regex_t *preg,
6130 || errcode >= (int) (sizeof (re_error_msgid_idx)
6131 / sizeof (re_error_msgid_idx[0])))
6132 /* Only error codes returned by the rest of the code should be passed
6133 to this routine. If we are given anything else, or if other regex
6134 code generates an invalid error code, then the program has a bug.
6135 Dump core so we can fix it. */
6138 msg = gettext (re_error_msgid + re_error_msgid_idx[errcode]);
6140 msg_size = strlen (msg) + 1; /* Includes the null. */
6142 if (errbuf_size != 0)
6144 if (msg_size > errbuf_size)
6146 #if defined HAVE_MEMPCPY || defined _LIBC
6147 *((char *) __mempcpy (errbuf, msg, errbuf_size - 1)) = '\0';
6149 memcpy (errbuf, msg, errbuf_size - 1);
6150 errbuf[errbuf_size - 1] = 0;
6154 memcpy (errbuf, msg, msg_size);
6160 weak_alias (__regerror, regerror)
6164 /* Free dynamically allocated space used by PREG. */
6170 if (preg->buffer != NULL)
6171 free (preg->buffer);
6172 preg->buffer = NULL;
6174 preg->allocated = 0;
6177 if (preg->fastmap != NULL)
6178 free (preg->fastmap);
6179 preg->fastmap = NULL;
6180 preg->fastmap_accurate = 0;
6182 if (preg->translate != NULL)
6183 free (preg->translate);
6184 preg->translate = NULL;
6188 weak_alias (__regfree, regfree)
6191 #endif /* not emacs */