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
35 # if defined __GNUC__ || (defined __STDC__ && __STDC__)
36 # define PARAMS(args) args
38 # define PARAMS(args) ()
40 #endif /* Not PARAMS. */
42 #if defined STDC_HEADERS && !defined emacs
45 /* We need this for `regex.h', and perhaps for the Emacs include files. */
46 # include <sys/types.h>
49 #define WIDE_CHAR_SUPPORT (HAVE_WCTYPE_H && HAVE_WCHAR_H && HAVE_BTOWC)
51 /* For platform which support the ISO C amendement 1 functionality we
52 support user defined character classes. */
53 #if defined _LIBC || WIDE_CHAR_SUPPORT
54 /* Solaris 2.5 has a bug: <wchar.h> must be included before <wctype.h>. */
60 /* We have to keep the namespace clean. */
61 # define regfree(preg) __regfree (preg)
62 # define regexec(pr, st, nm, pm, ef) __regexec (pr, st, nm, pm, ef)
63 # define regcomp(preg, pattern, cflags) __regcomp (preg, pattern, cflags)
64 # define regerror(errcode, preg, errbuf, errbuf_size) \
65 __regerror(errcode, preg, errbuf, errbuf_size)
66 # define re_set_registers(bu, re, nu, st, en) \
67 __re_set_registers (bu, re, nu, st, en)
68 # define re_match_2(bufp, string1, size1, string2, size2, pos, regs, stop) \
69 __re_match_2 (bufp, string1, size1, string2, size2, pos, regs, stop)
70 # define re_match(bufp, string, size, pos, regs) \
71 __re_match (bufp, string, size, pos, regs)
72 # define re_search(bufp, string, size, startpos, range, regs) \
73 __re_search (bufp, string, size, startpos, range, regs)
74 # define re_compile_pattern(pattern, length, bufp) \
75 __re_compile_pattern (pattern, length, bufp)
76 # define re_set_syntax(syntax) __re_set_syntax (syntax)
77 # define re_search_2(bufp, st1, s1, st2, s2, startpos, range, regs, stop) \
78 __re_search_2 (bufp, st1, s1, st2, s2, startpos, range, regs, stop)
79 # define re_compile_fastmap(bufp) __re_compile_fastmap (bufp)
81 # define btowc __btowc
83 /* We are also using some library internals. */
84 # include <locale/localeinfo.h>
85 # include <locale/elem-hash.h>
86 # include <langinfo.h>
89 /* This is for other GNU distributions with internationalized messages. */
90 #if HAVE_LIBINTL_H || defined _LIBC
93 # define gettext(msgid) (msgid)
97 /* This define is so xgettext can find the internationalizable
99 # define gettext_noop(String) String
102 /* The `emacs' switch turns on certain matching commands
103 that make sense only in Emacs. */
110 #else /* not emacs */
112 /* If we are not linking with Emacs proper,
113 we can't use the relocating allocator
114 even if config.h says that we can. */
117 # if defined STDC_HEADERS || defined _LIBC
124 /* When used in Emacs's lib-src, we need to get bzero and bcopy somehow.
125 If nothing else has been done, use the method below. */
126 # ifdef INHIBIT_STRING_HEADER
127 # if !(defined HAVE_BZERO && defined HAVE_BCOPY)
128 # if !defined bzero && !defined bcopy
129 # undef INHIBIT_STRING_HEADER
134 /* This is the normal way of making sure we have a bcopy and a bzero.
135 This is used in most programs--a few other programs avoid this
136 by defining INHIBIT_STRING_HEADER. */
137 # ifndef INHIBIT_STRING_HEADER
138 # if defined HAVE_STRING_H || defined STDC_HEADERS || defined _LIBC
142 # define bzero(s, n) (memset (s, '\0', n), (s))
144 # define bzero(s, n) __bzero (s, n)
148 # include <strings.h>
150 # define memcmp(s1, s2, n) bcmp (s1, s2, n)
153 # define memcpy(d, s, n) (bcopy (s, d, n), (d))
158 /* Define the syntax stuff for \<, \>, etc. */
160 /* This must be nonzero for the wordchar and notwordchar pattern
161 commands in re_match_2. */
166 # ifdef SWITCH_ENUM_BUG
167 # define SWITCH_ENUM_CAST(x) ((int)(x))
169 # define SWITCH_ENUM_CAST(x) (x)
172 #endif /* not emacs */
174 #if defined _LIBC || HAVE_LIMITS_H
179 # define MB_LEN_MAX 1
182 /* Get the interface, including the syntax bits. */
185 /* isalpha etc. are used for the character classes. */
188 /* Jim Meyering writes:
190 "... Some ctype macros are valid only for character codes that
191 isascii says are ASCII (SGI's IRIX-4.0.5 is one such system --when
192 using /bin/cc or gcc but without giving an ansi option). So, all
193 ctype uses should be through macros like ISPRINT... If
194 STDC_HEADERS is defined, then autoconf has verified that the ctype
195 macros don't need to be guarded with references to isascii. ...
196 Defining isascii to 1 should let any compiler worth its salt
197 eliminate the && through constant folding."
198 Solaris defines some of these symbols so we must undefine them first. */
201 #if defined STDC_HEADERS || (!defined isascii && !defined HAVE_ISASCII)
202 # define ISASCII(c) 1
204 # define ISASCII(c) isascii(c)
208 # define ISBLANK(c) (ISASCII (c) && isblank (c))
210 # define ISBLANK(c) ((c) == ' ' || (c) == '\t')
213 # define ISGRAPH(c) (ISASCII (c) && isgraph (c))
215 # define ISGRAPH(c) (ISASCII (c) && isprint (c) && !isspace (c))
219 #define ISPRINT(c) (ISASCII (c) && isprint (c))
220 #define ISDIGIT(c) (ISASCII (c) && isdigit (c))
221 #define ISALNUM(c) (ISASCII (c) && isalnum (c))
222 #define ISALPHA(c) (ISASCII (c) && isalpha (c))
223 #define ISCNTRL(c) (ISASCII (c) && iscntrl (c))
224 #define ISLOWER(c) (ISASCII (c) && islower (c))
225 #define ISPUNCT(c) (ISASCII (c) && ispunct (c))
226 #define ISSPACE(c) (ISASCII (c) && isspace (c))
227 #define ISUPPER(c) (ISASCII (c) && isupper (c))
228 #define ISXDIGIT(c) (ISASCII (c) && isxdigit (c))
231 # define TOLOWER(c) _tolower(c)
233 # define TOLOWER(c) tolower(c)
237 # define NULL (void *)0
240 /* We remove any previous definition of `SIGN_EXTEND_CHAR',
241 since ours (we hope) works properly with all combinations of
242 machines, compilers, `char' and `unsigned char' argument types.
243 (Per Bothner suggested the basic approach.) */
244 #undef SIGN_EXTEND_CHAR
246 # define SIGN_EXTEND_CHAR(c) ((signed char) (c))
247 #else /* not __STDC__ */
248 /* As in Harbison and Steele. */
249 # define SIGN_EXTEND_CHAR(c) ((((unsigned char) (c)) ^ 128) - 128)
253 /* How many characters in the character set. */
254 # define CHAR_SET_SIZE 256
258 extern char *re_syntax_table;
260 # else /* not SYNTAX_TABLE */
262 static char re_syntax_table[CHAR_SET_SIZE];
272 bzero (re_syntax_table, sizeof re_syntax_table);
274 for (c = 0; c < CHAR_SET_SIZE; ++c)
276 re_syntax_table[c] = Sword;
278 re_syntax_table['_'] = Sword;
283 # endif /* not SYNTAX_TABLE */
285 # define SYNTAX(c) re_syntax_table[(unsigned char) (c)]
289 /* Should we use malloc or alloca? If REGEX_MALLOC is not defined, we
290 use `alloca' instead of `malloc'. This is because using malloc in
291 re_search* or re_match* could cause memory leaks when C-g is used in
292 Emacs; also, malloc is slower and causes storage fragmentation. On
293 the other hand, malloc is more portable, and easier to debug.
295 Because we sometimes use alloca, some routines have to be macros,
296 not functions -- `alloca'-allocated space disappears at the end of the
297 function it is called in. */
301 # define REGEX_ALLOCATE malloc
302 # define REGEX_REALLOCATE(source, osize, nsize) realloc (source, nsize)
303 # define REGEX_FREE free
305 #else /* not REGEX_MALLOC */
307 /* Emacs already defines alloca, sometimes. */
310 /* Make alloca work the best possible way. */
312 # define alloca __builtin_alloca
313 # else /* not __GNUC__ */
316 # endif /* HAVE_ALLOCA_H */
317 # endif /* not __GNUC__ */
319 # endif /* not alloca */
321 # define REGEX_ALLOCATE alloca
323 /* Assumes a `char *destination' variable. */
324 # define REGEX_REALLOCATE(source, osize, nsize) \
325 (destination = (char *) alloca (nsize), \
326 memcpy (destination, source, osize))
328 /* No need to do anything to free, after alloca. */
329 # define REGEX_FREE(arg) ((void)0) /* Do nothing! But inhibit gcc warning. */
331 #endif /* not REGEX_MALLOC */
333 /* Define how to allocate the failure stack. */
335 #if defined REL_ALLOC && defined REGEX_MALLOC
337 # define REGEX_ALLOCATE_STACK(size) \
338 r_alloc (&failure_stack_ptr, (size))
339 # define REGEX_REALLOCATE_STACK(source, osize, nsize) \
340 r_re_alloc (&failure_stack_ptr, (nsize))
341 # define REGEX_FREE_STACK(ptr) \
342 r_alloc_free (&failure_stack_ptr)
344 #else /* not using relocating allocator */
348 # define REGEX_ALLOCATE_STACK malloc
349 # define REGEX_REALLOCATE_STACK(source, osize, nsize) realloc (source, nsize)
350 # define REGEX_FREE_STACK free
352 # else /* not REGEX_MALLOC */
354 # define REGEX_ALLOCATE_STACK alloca
356 # define REGEX_REALLOCATE_STACK(source, osize, nsize) \
357 REGEX_REALLOCATE (source, osize, nsize)
358 /* No need to explicitly free anything. */
359 # define REGEX_FREE_STACK(arg)
361 # endif /* not REGEX_MALLOC */
362 #endif /* not using relocating allocator */
365 /* True if `size1' is non-NULL and PTR is pointing anywhere inside
366 `string1' or just past its end. This works if PTR is NULL, which is
368 #define FIRST_STRING_P(ptr) \
369 (size1 && string1 <= (ptr) && (ptr) <= string1 + size1)
371 /* (Re)Allocate N items of type T using malloc, or fail. */
372 #define TALLOC(n, t) ((t *) malloc ((n) * sizeof (t)))
373 #define RETALLOC(addr, n, t) ((addr) = (t *) realloc (addr, (n) * sizeof (t)))
374 #define RETALLOC_IF(addr, n, t) \
375 if (addr) RETALLOC((addr), (n), t); else (addr) = TALLOC ((n), t)
376 #define REGEX_TALLOC(n, t) ((t *) REGEX_ALLOCATE ((n) * sizeof (t)))
378 #define BYTEWIDTH 8 /* In bits. */
380 #define STREQ(s1, s2) ((strcmp (s1, s2) == 0))
384 #define MAX(a, b) ((a) > (b) ? (a) : (b))
385 #define MIN(a, b) ((a) < (b) ? (a) : (b))
387 typedef char boolean;
391 static int re_match_2_internal PARAMS ((struct re_pattern_buffer *bufp,
392 const char *string1, int size1,
393 const char *string2, int size2,
395 struct re_registers *regs,
398 /* These are the command codes that appear in compiled regular
399 expressions. Some opcodes are followed by argument bytes. A
400 command code can specify any interpretation whatsoever for its
401 arguments. Zero bytes may appear in the compiled regular expression. */
407 /* Succeed right away--no more backtracking. */
410 /* Followed by one byte giving n, then by n literal bytes. */
413 /* Matches any (more or less) character. */
416 /* Matches any one char belonging to specified set. First
417 following byte is number of bitmap bytes. Then come bytes
418 for a bitmap saying which chars are in. Bits in each byte
419 are ordered low-bit-first. A character is in the set if its
420 bit is 1. A character too large to have a bit in the map is
421 automatically not in the set. */
424 /* Same parameters as charset, but match any character that is
425 not one of those specified. */
428 /* Start remembering the text that is matched, for storing in a
429 register. Followed by one byte with the register number, in
430 the range 0 to one less than the pattern buffer's re_nsub
431 field. Then followed by one byte with the number of groups
432 inner to this one. (This last has to be part of the
433 start_memory only because we need it in the on_failure_jump
437 /* Stop remembering the text that is matched and store it in a
438 memory register. Followed by one byte with the register
439 number, in the range 0 to one less than `re_nsub' in the
440 pattern buffer, and one byte with the number of inner groups,
441 just like `start_memory'. (We need the number of inner
442 groups here because we don't have any easy way of finding the
443 corresponding start_memory when we're at a stop_memory.) */
446 /* Match a duplicate of something remembered. Followed by one
447 byte containing the register number. */
450 /* Fail unless at beginning of line. */
453 /* Fail unless at end of line. */
456 /* Succeeds if at beginning of buffer (if emacs) or at beginning
457 of string to be matched (if not). */
460 /* Analogously, for end of buffer/string. */
463 /* Followed by two byte relative address to which to jump. */
466 /* Same as jump, but marks the end of an alternative. */
469 /* Followed by two-byte relative address of place to resume at
470 in case of failure. */
473 /* Like on_failure_jump, but pushes a placeholder instead of the
474 current string position when executed. */
475 on_failure_keep_string_jump,
477 /* Throw away latest failure point and then jump to following
478 two-byte relative address. */
481 /* Change to pop_failure_jump if know won't have to backtrack to
482 match; otherwise change to jump. This is used to jump
483 back to the beginning of a repeat. If what follows this jump
484 clearly won't match what the repeat does, such that we can be
485 sure that there is no use backtracking out of repetitions
486 already matched, then we change it to a pop_failure_jump.
487 Followed by two-byte address. */
490 /* Jump to following two-byte address, and push a dummy failure
491 point. This failure point will be thrown away if an attempt
492 is made to use it for a failure. A `+' construct makes this
493 before the first repeat. Also used as an intermediary kind
494 of jump when compiling an alternative. */
497 /* Push a dummy failure point and continue. Used at the end of
501 /* Followed by two-byte relative address and two-byte number n.
502 After matching N times, jump to the address upon failure. */
505 /* Followed by two-byte relative address, and two-byte number n.
506 Jump to the address N times, then fail. */
509 /* Set the following two-byte relative address to the
510 subsequent two-byte number. The address *includes* the two
514 wordchar, /* Matches any word-constituent character. */
515 notwordchar, /* Matches any char that is not a word-constituent. */
517 wordbeg, /* Succeeds if at word beginning. */
518 wordend, /* Succeeds if at word end. */
520 wordbound, /* Succeeds if at a word boundary. */
521 notwordbound /* Succeeds if not at a word boundary. */
524 ,before_dot, /* Succeeds if before point. */
525 at_dot, /* Succeeds if at point. */
526 after_dot, /* Succeeds if after point. */
528 /* Matches any character whose syntax is specified. Followed by
529 a byte which contains a syntax code, e.g., Sword. */
532 /* Matches any character whose syntax is not that specified. */
537 /* Common operations on the compiled pattern. */
539 /* Store NUMBER in two contiguous bytes starting at DESTINATION. */
541 #define STORE_NUMBER(destination, number) \
543 (destination)[0] = (number) & 0377; \
544 (destination)[1] = (number) >> 8; \
547 /* Same as STORE_NUMBER, except increment DESTINATION to
548 the byte after where the number is stored. Therefore, DESTINATION
549 must be an lvalue. */
551 #define STORE_NUMBER_AND_INCR(destination, number) \
553 STORE_NUMBER (destination, number); \
554 (destination) += 2; \
557 /* Put into DESTINATION a number stored in two contiguous bytes starting
560 #define EXTRACT_NUMBER(destination, source) \
562 (destination) = *(source) & 0377; \
563 (destination) += SIGN_EXTEND_CHAR (*((source) + 1)) << 8; \
567 static void extract_number _RE_ARGS ((int *dest, unsigned char *source));
569 extract_number (dest, source)
571 unsigned char *source;
573 int temp = SIGN_EXTEND_CHAR (*(source + 1));
574 *dest = *source & 0377;
578 # ifndef EXTRACT_MACROS /* To debug the macros. */
579 # undef EXTRACT_NUMBER
580 # define EXTRACT_NUMBER(dest, src) extract_number (&dest, src)
581 # endif /* not EXTRACT_MACROS */
585 /* Same as EXTRACT_NUMBER, except increment SOURCE to after the number.
586 SOURCE must be an lvalue. */
588 #define EXTRACT_NUMBER_AND_INCR(destination, source) \
590 EXTRACT_NUMBER (destination, source); \
595 static void extract_number_and_incr _RE_ARGS ((int *destination,
596 unsigned char **source));
598 extract_number_and_incr (destination, source)
600 unsigned char **source;
602 extract_number (destination, *source);
606 # ifndef EXTRACT_MACROS
607 # undef EXTRACT_NUMBER_AND_INCR
608 # define EXTRACT_NUMBER_AND_INCR(dest, src) \
609 extract_number_and_incr (&dest, &src)
610 # endif /* not EXTRACT_MACROS */
614 /* If DEBUG is defined, Regex prints many voluminous messages about what
615 it is doing (if the variable `debug' is nonzero). If linked with the
616 main program in `iregex.c', you can enter patterns and strings
617 interactively. And if linked with the main program in `main.c' and
618 the other test files, you can run the already-written tests. */
622 /* We use standard I/O for debugging. */
625 /* It is useful to test things that ``must'' be true when debugging. */
630 # define DEBUG_STATEMENT(e) e
631 # define DEBUG_PRINT1(x) if (debug) printf (x)
632 # define DEBUG_PRINT2(x1, x2) if (debug) printf (x1, x2)
633 # define DEBUG_PRINT3(x1, x2, x3) if (debug) printf (x1, x2, x3)
634 # define DEBUG_PRINT4(x1, x2, x3, x4) if (debug) printf (x1, x2, x3, x4)
635 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) \
636 if (debug) print_partial_compiled_pattern (s, e)
637 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) \
638 if (debug) print_double_string (w, s1, sz1, s2, sz2)
641 /* Print the fastmap in human-readable form. */
644 print_fastmap (fastmap)
647 unsigned was_a_range = 0;
650 while (i < (1 << BYTEWIDTH))
656 while (i < (1 << BYTEWIDTH) && fastmap[i])
672 /* Print a compiled pattern string in human-readable form, starting at
673 the START pointer into it and ending just before the pointer END. */
676 print_partial_compiled_pattern (start, end)
677 unsigned char *start;
682 unsigned char *p = start;
683 unsigned char *pend = end;
691 /* Loop over pattern commands. */
695 printf ("%t:\t", p - start);
697 printf ("%ld:\t", (long int) (p - start));
700 switch ((re_opcode_t) *p++)
708 printf ("/exactn/%d", mcnt);
719 printf ("/start_memory/%d/%d", mcnt, *p++);
724 printf ("/stop_memory/%d/%d", mcnt, *p++);
728 printf ("/duplicate/%d", *p++);
738 register int c, last = -100;
739 register int in_range = 0;
741 printf ("/charset [%s",
742 (re_opcode_t) *(p - 1) == charset_not ? "^" : "");
744 assert (p + *p < pend);
746 for (c = 0; c < 256; c++)
748 && (p[1 + (c/8)] & (1 << (c % 8))))
750 /* Are we starting a range? */
751 if (last + 1 == c && ! in_range)
756 /* Have we broken a range? */
757 else if (last + 1 != c && in_range)
786 case on_failure_jump:
787 extract_number_and_incr (&mcnt, &p);
789 printf ("/on_failure_jump to %t", p + mcnt - start);
791 printf ("/on_failure_jump to %ld", (long int) (p + mcnt - start));
795 case on_failure_keep_string_jump:
796 extract_number_and_incr (&mcnt, &p);
798 printf ("/on_failure_keep_string_jump to %t", p + mcnt - start);
800 printf ("/on_failure_keep_string_jump to %ld",
801 (long int) (p + mcnt - start));
805 case dummy_failure_jump:
806 extract_number_and_incr (&mcnt, &p);
808 printf ("/dummy_failure_jump to %t", p + mcnt - start);
810 printf ("/dummy_failure_jump to %ld", (long int) (p + mcnt - start));
814 case push_dummy_failure:
815 printf ("/push_dummy_failure");
819 extract_number_and_incr (&mcnt, &p);
821 printf ("/maybe_pop_jump to %t", p + mcnt - start);
823 printf ("/maybe_pop_jump to %ld", (long int) (p + mcnt - start));
827 case pop_failure_jump:
828 extract_number_and_incr (&mcnt, &p);
830 printf ("/pop_failure_jump to %t", p + mcnt - start);
832 printf ("/pop_failure_jump to %ld", (long int) (p + mcnt - start));
837 extract_number_and_incr (&mcnt, &p);
839 printf ("/jump_past_alt to %t", p + mcnt - start);
841 printf ("/jump_past_alt to %ld", (long int) (p + mcnt - start));
846 extract_number_and_incr (&mcnt, &p);
848 printf ("/jump to %t", p + mcnt - start);
850 printf ("/jump to %ld", (long int) (p + mcnt - start));
855 extract_number_and_incr (&mcnt, &p);
857 extract_number_and_incr (&mcnt2, &p);
859 printf ("/succeed_n to %t, %d times", p1 - start, mcnt2);
861 printf ("/succeed_n to %ld, %d times",
862 (long int) (p1 - start), mcnt2);
867 extract_number_and_incr (&mcnt, &p);
869 extract_number_and_incr (&mcnt2, &p);
870 printf ("/jump_n to %d, %d times", p1 - start, mcnt2);
874 extract_number_and_incr (&mcnt, &p);
876 extract_number_and_incr (&mcnt2, &p);
878 printf ("/set_number_at location %t to %d", p1 - start, mcnt2);
880 printf ("/set_number_at location %ld to %d",
881 (long int) (p1 - start), mcnt2);
886 printf ("/wordbound");
890 printf ("/notwordbound");
902 printf ("/before_dot");
910 printf ("/after_dot");
914 printf ("/syntaxspec");
916 printf ("/%d", mcnt);
920 printf ("/notsyntaxspec");
922 printf ("/%d", mcnt);
927 printf ("/wordchar");
931 printf ("/notwordchar");
943 printf ("?%d", *(p-1));
950 printf ("%t:\tend of pattern.\n", p - start);
952 printf ("%ld:\tend of pattern.\n", (long int) (p - start));
958 print_compiled_pattern (bufp)
959 struct re_pattern_buffer *bufp;
961 unsigned char *buffer = bufp->buffer;
963 print_partial_compiled_pattern (buffer, buffer + bufp->used);
964 printf ("%ld bytes used/%ld bytes allocated.\n",
965 bufp->used, bufp->allocated);
967 if (bufp->fastmap_accurate && bufp->fastmap)
969 printf ("fastmap: ");
970 print_fastmap (bufp->fastmap);
974 printf ("re_nsub: %Zd\t", bufp->re_nsub);
976 printf ("re_nsub: %ld\t", (long int) bufp->re_nsub);
978 printf ("regs_alloc: %d\t", bufp->regs_allocated);
979 printf ("can_be_null: %d\t", bufp->can_be_null);
980 printf ("newline_anchor: %d\n", bufp->newline_anchor);
981 printf ("no_sub: %d\t", bufp->no_sub);
982 printf ("not_bol: %d\t", bufp->not_bol);
983 printf ("not_eol: %d\t", bufp->not_eol);
984 printf ("syntax: %lx\n", bufp->syntax);
985 /* Perhaps we should print the translate table? */
990 print_double_string (where, string1, size1, string2, size2)
1003 if (FIRST_STRING_P (where))
1005 for (this_char = where - string1; this_char < size1; this_char++)
1006 putchar (string1[this_char]);
1011 for (this_char = where - string2; this_char < size2; this_char++)
1012 putchar (string2[this_char]);
1023 #else /* not DEBUG */
1028 # define DEBUG_STATEMENT(e)
1029 # define DEBUG_PRINT1(x)
1030 # define DEBUG_PRINT2(x1, x2)
1031 # define DEBUG_PRINT3(x1, x2, x3)
1032 # define DEBUG_PRINT4(x1, x2, x3, x4)
1033 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e)
1034 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2)
1036 #endif /* not DEBUG */
1038 /* Set by `re_set_syntax' to the current regexp syntax to recognize. Can
1039 also be assigned to arbitrarily: each pattern buffer stores its own
1040 syntax, so it can be changed between regex compilations. */
1041 /* This has no initializer because initialized variables in Emacs
1042 become read-only after dumping. */
1043 reg_syntax_t re_syntax_options;
1046 /* Specify the precise syntax of regexps for compilation. This provides
1047 for compatibility for various utilities which historically have
1048 different, incompatible syntaxes.
1050 The argument SYNTAX is a bit mask comprised of the various bits
1051 defined in regex.h. We return the old syntax. */
1054 re_set_syntax (syntax)
1055 reg_syntax_t syntax;
1057 reg_syntax_t ret = re_syntax_options;
1059 re_syntax_options = syntax;
1061 if (syntax & RE_DEBUG)
1063 else if (debug) /* was on but now is not */
1069 weak_alias (__re_set_syntax, re_set_syntax)
1072 /* This table gives an error message for each of the error codes listed
1073 in regex.h. Obviously the order here has to be same as there.
1074 POSIX doesn't require that we do anything for REG_NOERROR,
1075 but why not be nice? */
1077 static const char re_error_msgid[] =
1079 #define REG_NOERROR_IDX 0
1080 gettext_noop ("Success") /* REG_NOERROR */
1082 #define REG_NOMATCH_IDX (REG_NOERROR_IDX + sizeof "Success")
1083 gettext_noop ("No match") /* REG_NOMATCH */
1085 #define REG_BADPAT_IDX (REG_NOMATCH_IDX + sizeof "No match")
1086 gettext_noop ("Invalid regular expression") /* REG_BADPAT */
1088 #define REG_ECOLLATE_IDX (REG_BADPAT_IDX + sizeof "Invalid regular expression")
1089 gettext_noop ("Invalid collation character") /* REG_ECOLLATE */
1091 #define REG_ECTYPE_IDX (REG_ECOLLATE_IDX + sizeof "Invalid collation character")
1092 gettext_noop ("Invalid character class name") /* REG_ECTYPE */
1094 #define REG_EESCAPE_IDX (REG_ECTYPE_IDX + sizeof "Invalid character class name")
1095 gettext_noop ("Trailing backslash") /* REG_EESCAPE */
1097 #define REG_ESUBREG_IDX (REG_EESCAPE_IDX + sizeof "Trailing backslash")
1098 gettext_noop ("Invalid back reference") /* REG_ESUBREG */
1100 #define REG_EBRACK_IDX (REG_ESUBREG_IDX + sizeof "Invalid back reference")
1101 gettext_noop ("Unmatched [ or [^") /* REG_EBRACK */
1103 #define REG_EPAREN_IDX (REG_EBRACK_IDX + sizeof "Unmatched [ or [^")
1104 gettext_noop ("Unmatched ( or \\(") /* REG_EPAREN */
1106 #define REG_EBRACE_IDX (REG_EPAREN_IDX + sizeof "Unmatched ( or \\(")
1107 gettext_noop ("Unmatched \\{") /* REG_EBRACE */
1109 #define REG_BADBR_IDX (REG_EBRACE_IDX + sizeof "Unmatched \\{")
1110 gettext_noop ("Invalid content of \\{\\}") /* REG_BADBR */
1112 #define REG_ERANGE_IDX (REG_BADBR_IDX + sizeof "Invalid content of \\{\\}")
1113 gettext_noop ("Invalid range end") /* REG_ERANGE */
1115 #define REG_ESPACE_IDX (REG_ERANGE_IDX + sizeof "Invalid range end")
1116 gettext_noop ("Memory exhausted") /* REG_ESPACE */
1118 #define REG_BADRPT_IDX (REG_ESPACE_IDX + sizeof "Memory exhausted")
1119 gettext_noop ("Invalid preceding regular expression") /* REG_BADRPT */
1121 #define REG_EEND_IDX (REG_BADRPT_IDX + sizeof "Invalid preceding regular expression")
1122 gettext_noop ("Premature end of regular expression") /* REG_EEND */
1124 #define REG_ESIZE_IDX (REG_EEND_IDX + sizeof "Premature end of regular expression")
1125 gettext_noop ("Regular expression too big") /* REG_ESIZE */
1127 #define REG_ERPAREN_IDX (REG_ESIZE_IDX + sizeof "Regular expression too big")
1128 gettext_noop ("Unmatched ) or \\)") /* REG_ERPAREN */
1131 static const size_t re_error_msgid_idx[] =
1152 /* Avoiding alloca during matching, to placate r_alloc. */
1154 /* Define MATCH_MAY_ALLOCATE unless we need to make sure that the
1155 searching and matching functions should not call alloca. On some
1156 systems, alloca is implemented in terms of malloc, and if we're
1157 using the relocating allocator routines, then malloc could cause a
1158 relocation, which might (if the strings being searched are in the
1159 ralloc heap) shift the data out from underneath the regexp
1162 Here's another reason to avoid allocation: Emacs
1163 processes input from X in a signal handler; processing X input may
1164 call malloc; if input arrives while a matching routine is calling
1165 malloc, then we're scrod. But Emacs can't just block input while
1166 calling matching routines; then we don't notice interrupts when
1167 they come in. So, Emacs blocks input around all regexp calls
1168 except the matching calls, which it leaves unprotected, in the
1169 faith that they will not malloc. */
1171 /* Normally, this is fine. */
1172 #define MATCH_MAY_ALLOCATE
1174 /* When using GNU C, we are not REALLY using the C alloca, no matter
1175 what config.h may say. So don't take precautions for it. */
1180 /* The match routines may not allocate if (1) they would do it with malloc
1181 and (2) it's not safe for them to use malloc.
1182 Note that if REL_ALLOC is defined, matching would not use malloc for the
1183 failure stack, but we would still use it for the register vectors;
1184 so REL_ALLOC should not affect this. */
1185 #if (defined C_ALLOCA || defined REGEX_MALLOC) && defined emacs
1186 # undef MATCH_MAY_ALLOCATE
1190 /* Failure stack declarations and macros; both re_compile_fastmap and
1191 re_match_2 use a failure stack. These have to be macros because of
1192 REGEX_ALLOCATE_STACK. */
1195 /* Number of failure points for which to initially allocate space
1196 when matching. If this number is exceeded, we allocate more
1197 space, so it is not a hard limit. */
1198 #ifndef INIT_FAILURE_ALLOC
1199 # define INIT_FAILURE_ALLOC 5
1202 /* Roughly the maximum number of failure points on the stack. Would be
1203 exactly that if always used MAX_FAILURE_ITEMS items each time we failed.
1204 This is a variable only so users of regex can assign to it; we never
1205 change it ourselves. */
1209 # if defined MATCH_MAY_ALLOCATE
1210 /* 4400 was enough to cause a crash on Alpha OSF/1,
1211 whose default stack limit is 2mb. */
1212 long int re_max_failures = 4000;
1214 long int re_max_failures = 2000;
1217 union fail_stack_elt
1219 unsigned char *pointer;
1223 typedef union fail_stack_elt fail_stack_elt_t;
1227 fail_stack_elt_t *stack;
1228 unsigned long int size;
1229 unsigned long int avail; /* Offset of next open position. */
1232 #else /* not INT_IS_16BIT */
1234 # if defined MATCH_MAY_ALLOCATE
1235 /* 4400 was enough to cause a crash on Alpha OSF/1,
1236 whose default stack limit is 2mb. */
1237 int re_max_failures = 4000;
1239 int re_max_failures = 2000;
1242 union fail_stack_elt
1244 unsigned char *pointer;
1248 typedef union fail_stack_elt fail_stack_elt_t;
1252 fail_stack_elt_t *stack;
1254 unsigned avail; /* Offset of next open position. */
1257 #endif /* INT_IS_16BIT */
1259 #define FAIL_STACK_EMPTY() (fail_stack.avail == 0)
1260 #define FAIL_STACK_PTR_EMPTY() (fail_stack_ptr->avail == 0)
1261 #define FAIL_STACK_FULL() (fail_stack.avail == fail_stack.size)
1264 /* Define macros to initialize and free the failure stack.
1265 Do `return -2' if the alloc fails. */
1267 #ifdef MATCH_MAY_ALLOCATE
1268 # define INIT_FAIL_STACK() \
1270 fail_stack.stack = (fail_stack_elt_t *) \
1271 REGEX_ALLOCATE_STACK (INIT_FAILURE_ALLOC * sizeof (fail_stack_elt_t)); \
1273 if (fail_stack.stack == NULL) \
1276 fail_stack.size = INIT_FAILURE_ALLOC; \
1277 fail_stack.avail = 0; \
1280 # define RESET_FAIL_STACK() REGEX_FREE_STACK (fail_stack.stack)
1282 # define INIT_FAIL_STACK() \
1284 fail_stack.avail = 0; \
1287 # define RESET_FAIL_STACK()
1291 /* Double the size of FAIL_STACK, up to approximately `re_max_failures' items.
1293 Return 1 if succeeds, and 0 if either ran out of memory
1294 allocating space for it or it was already too large.
1296 REGEX_REALLOCATE_STACK requires `destination' be declared. */
1298 #define DOUBLE_FAIL_STACK(fail_stack) \
1299 ((fail_stack).size > (unsigned) (re_max_failures * MAX_FAILURE_ITEMS) \
1301 : ((fail_stack).stack = (fail_stack_elt_t *) \
1302 REGEX_REALLOCATE_STACK ((fail_stack).stack, \
1303 (fail_stack).size * sizeof (fail_stack_elt_t), \
1304 ((fail_stack).size << 1) * sizeof (fail_stack_elt_t)), \
1306 (fail_stack).stack == NULL \
1308 : ((fail_stack).size <<= 1, \
1312 /* Push pointer POINTER on FAIL_STACK.
1313 Return 1 if was able to do so and 0 if ran out of memory allocating
1315 #define PUSH_PATTERN_OP(POINTER, FAIL_STACK) \
1316 ((FAIL_STACK_FULL () \
1317 && !DOUBLE_FAIL_STACK (FAIL_STACK)) \
1319 : ((FAIL_STACK).stack[(FAIL_STACK).avail++].pointer = POINTER, \
1322 /* Push a pointer value onto the failure stack.
1323 Assumes the variable `fail_stack'. Probably should only
1324 be called from within `PUSH_FAILURE_POINT'. */
1325 #define PUSH_FAILURE_POINTER(item) \
1326 fail_stack.stack[fail_stack.avail++].pointer = (unsigned char *) (item)
1328 /* This pushes an integer-valued item onto the failure stack.
1329 Assumes the variable `fail_stack'. Probably should only
1330 be called from within `PUSH_FAILURE_POINT'. */
1331 #define PUSH_FAILURE_INT(item) \
1332 fail_stack.stack[fail_stack.avail++].integer = (item)
1334 /* Push a fail_stack_elt_t value onto the failure stack.
1335 Assumes the variable `fail_stack'. Probably should only
1336 be called from within `PUSH_FAILURE_POINT'. */
1337 #define PUSH_FAILURE_ELT(item) \
1338 fail_stack.stack[fail_stack.avail++] = (item)
1340 /* These three POP... operations complement the three PUSH... operations.
1341 All assume that `fail_stack' is nonempty. */
1342 #define POP_FAILURE_POINTER() fail_stack.stack[--fail_stack.avail].pointer
1343 #define POP_FAILURE_INT() fail_stack.stack[--fail_stack.avail].integer
1344 #define POP_FAILURE_ELT() fail_stack.stack[--fail_stack.avail]
1346 /* Used to omit pushing failure point id's when we're not debugging. */
1348 # define DEBUG_PUSH PUSH_FAILURE_INT
1349 # define DEBUG_POP(item_addr) *(item_addr) = POP_FAILURE_INT ()
1351 # define DEBUG_PUSH(item)
1352 # define DEBUG_POP(item_addr)
1356 /* Push the information about the state we will need
1357 if we ever fail back to it.
1359 Requires variables fail_stack, regstart, regend, reg_info, and
1360 num_regs_pushed be declared. DOUBLE_FAIL_STACK requires `destination'
1363 Does `return FAILURE_CODE' if runs out of memory. */
1365 #define PUSH_FAILURE_POINT(pattern_place, string_place, failure_code) \
1367 char *destination; \
1368 /* Must be int, so when we don't save any registers, the arithmetic \
1369 of 0 + -1 isn't done as unsigned. */ \
1370 /* Can't be int, since there is not a shred of a guarantee that int \
1371 is wide enough to hold a value of something to which pointer can \
1373 active_reg_t this_reg; \
1375 DEBUG_STATEMENT (failure_id++); \
1376 DEBUG_STATEMENT (nfailure_points_pushed++); \
1377 DEBUG_PRINT2 ("\nPUSH_FAILURE_POINT #%u:\n", failure_id); \
1378 DEBUG_PRINT2 (" Before push, next avail: %d\n", (fail_stack).avail);\
1379 DEBUG_PRINT2 (" size: %d\n", (fail_stack).size);\
1381 DEBUG_PRINT2 (" slots needed: %ld\n", NUM_FAILURE_ITEMS); \
1382 DEBUG_PRINT2 (" available: %d\n", REMAINING_AVAIL_SLOTS); \
1384 /* Ensure we have enough space allocated for what we will push. */ \
1385 while (REMAINING_AVAIL_SLOTS < NUM_FAILURE_ITEMS) \
1387 if (!DOUBLE_FAIL_STACK (fail_stack)) \
1388 return failure_code; \
1390 DEBUG_PRINT2 ("\n Doubled stack; size now: %d\n", \
1391 (fail_stack).size); \
1392 DEBUG_PRINT2 (" slots available: %d\n", REMAINING_AVAIL_SLOTS);\
1395 /* Push the info, starting with the registers. */ \
1396 DEBUG_PRINT1 ("\n"); \
1399 for (this_reg = lowest_active_reg; this_reg <= highest_active_reg; \
1402 DEBUG_PRINT2 (" Pushing reg: %lu\n", this_reg); \
1403 DEBUG_STATEMENT (num_regs_pushed++); \
1405 DEBUG_PRINT2 (" start: %p\n", regstart[this_reg]); \
1406 PUSH_FAILURE_POINTER (regstart[this_reg]); \
1408 DEBUG_PRINT2 (" end: %p\n", regend[this_reg]); \
1409 PUSH_FAILURE_POINTER (regend[this_reg]); \
1411 DEBUG_PRINT2 (" info: %p\n ", \
1412 reg_info[this_reg].word.pointer); \
1413 DEBUG_PRINT2 (" match_null=%d", \
1414 REG_MATCH_NULL_STRING_P (reg_info[this_reg])); \
1415 DEBUG_PRINT2 (" active=%d", IS_ACTIVE (reg_info[this_reg])); \
1416 DEBUG_PRINT2 (" matched_something=%d", \
1417 MATCHED_SOMETHING (reg_info[this_reg])); \
1418 DEBUG_PRINT2 (" ever_matched=%d", \
1419 EVER_MATCHED_SOMETHING (reg_info[this_reg])); \
1420 DEBUG_PRINT1 ("\n"); \
1421 PUSH_FAILURE_ELT (reg_info[this_reg].word); \
1424 DEBUG_PRINT2 (" Pushing low active reg: %ld\n", lowest_active_reg);\
1425 PUSH_FAILURE_INT (lowest_active_reg); \
1427 DEBUG_PRINT2 (" Pushing high active reg: %ld\n", highest_active_reg);\
1428 PUSH_FAILURE_INT (highest_active_reg); \
1430 DEBUG_PRINT2 (" Pushing pattern %p:\n", pattern_place); \
1431 DEBUG_PRINT_COMPILED_PATTERN (bufp, pattern_place, pend); \
1432 PUSH_FAILURE_POINTER (pattern_place); \
1434 DEBUG_PRINT2 (" Pushing string %p: `", string_place); \
1435 DEBUG_PRINT_DOUBLE_STRING (string_place, string1, size1, string2, \
1437 DEBUG_PRINT1 ("'\n"); \
1438 PUSH_FAILURE_POINTER (string_place); \
1440 DEBUG_PRINT2 (" Pushing failure id: %u\n", failure_id); \
1441 DEBUG_PUSH (failure_id); \
1444 /* This is the number of items that are pushed and popped on the stack
1445 for each register. */
1446 #define NUM_REG_ITEMS 3
1448 /* Individual items aside from the registers. */
1450 # define NUM_NONREG_ITEMS 5 /* Includes failure point id. */
1452 # define NUM_NONREG_ITEMS 4
1455 /* We push at most this many items on the stack. */
1456 /* We used to use (num_regs - 1), which is the number of registers
1457 this regexp will save; but that was changed to 5
1458 to avoid stack overflow for a regexp with lots of parens. */
1459 #define MAX_FAILURE_ITEMS (5 * NUM_REG_ITEMS + NUM_NONREG_ITEMS)
1461 /* We actually push this many items. */
1462 #define NUM_FAILURE_ITEMS \
1464 ? 0 : highest_active_reg - lowest_active_reg + 1) \
1468 /* How many items can still be added to the stack without overflowing it. */
1469 #define REMAINING_AVAIL_SLOTS ((fail_stack).size - (fail_stack).avail)
1472 /* Pops what PUSH_FAIL_STACK pushes.
1474 We restore into the parameters, all of which should be lvalues:
1475 STR -- the saved data position.
1476 PAT -- the saved pattern position.
1477 LOW_REG, HIGH_REG -- the highest and lowest active registers.
1478 REGSTART, REGEND -- arrays of string positions.
1479 REG_INFO -- array of information about each subexpression.
1481 Also assumes the variables `fail_stack' and (if debugging), `bufp',
1482 `pend', `string1', `size1', `string2', and `size2'. */
1484 #define POP_FAILURE_POINT(str, pat, low_reg, high_reg, regstart, regend, reg_info)\
1486 DEBUG_STATEMENT (unsigned failure_id;) \
1487 active_reg_t this_reg; \
1488 const unsigned char *string_temp; \
1490 assert (!FAIL_STACK_EMPTY ()); \
1492 /* Remove failure points and point to how many regs pushed. */ \
1493 DEBUG_PRINT1 ("POP_FAILURE_POINT:\n"); \
1494 DEBUG_PRINT2 (" Before pop, next avail: %d\n", fail_stack.avail); \
1495 DEBUG_PRINT2 (" size: %d\n", fail_stack.size); \
1497 assert (fail_stack.avail >= NUM_NONREG_ITEMS); \
1499 DEBUG_POP (&failure_id); \
1500 DEBUG_PRINT2 (" Popping failure id: %u\n", failure_id); \
1502 /* If the saved string location is NULL, it came from an \
1503 on_failure_keep_string_jump opcode, and we want to throw away the \
1504 saved NULL, thus retaining our current position in the string. */ \
1505 string_temp = POP_FAILURE_POINTER (); \
1506 if (string_temp != NULL) \
1507 str = (const char *) string_temp; \
1509 DEBUG_PRINT2 (" Popping string %p: `", str); \
1510 DEBUG_PRINT_DOUBLE_STRING (str, string1, size1, string2, size2); \
1511 DEBUG_PRINT1 ("'\n"); \
1513 pat = (unsigned char *) POP_FAILURE_POINTER (); \
1514 DEBUG_PRINT2 (" Popping pattern %p:\n", pat); \
1515 DEBUG_PRINT_COMPILED_PATTERN (bufp, pat, pend); \
1517 /* Restore register info. */ \
1518 high_reg = (active_reg_t) POP_FAILURE_INT (); \
1519 DEBUG_PRINT2 (" Popping high active reg: %ld\n", high_reg); \
1521 low_reg = (active_reg_t) POP_FAILURE_INT (); \
1522 DEBUG_PRINT2 (" Popping low active reg: %ld\n", low_reg); \
1525 for (this_reg = high_reg; this_reg >= low_reg; this_reg--) \
1527 DEBUG_PRINT2 (" Popping reg: %ld\n", this_reg); \
1529 reg_info[this_reg].word = POP_FAILURE_ELT (); \
1530 DEBUG_PRINT2 (" info: %p\n", \
1531 reg_info[this_reg].word.pointer); \
1533 regend[this_reg] = (const char *) POP_FAILURE_POINTER (); \
1534 DEBUG_PRINT2 (" end: %p\n", regend[this_reg]); \
1536 regstart[this_reg] = (const char *) POP_FAILURE_POINTER (); \
1537 DEBUG_PRINT2 (" start: %p\n", regstart[this_reg]); \
1541 for (this_reg = highest_active_reg; this_reg > high_reg; this_reg--) \
1543 reg_info[this_reg].word.integer = 0; \
1544 regend[this_reg] = 0; \
1545 regstart[this_reg] = 0; \
1547 highest_active_reg = high_reg; \
1550 set_regs_matched_done = 0; \
1551 DEBUG_STATEMENT (nfailure_points_popped++); \
1552 } /* POP_FAILURE_POINT */
1556 /* Structure for per-register (a.k.a. per-group) information.
1557 Other register information, such as the
1558 starting and ending positions (which are addresses), and the list of
1559 inner groups (which is a bits list) are maintained in separate
1562 We are making a (strictly speaking) nonportable assumption here: that
1563 the compiler will pack our bit fields into something that fits into
1564 the type of `word', i.e., is something that fits into one item on the
1568 /* Declarations and macros for re_match_2. */
1572 fail_stack_elt_t word;
1575 /* This field is one if this group can match the empty string,
1576 zero if not. If not yet determined, `MATCH_NULL_UNSET_VALUE'. */
1577 #define MATCH_NULL_UNSET_VALUE 3
1578 unsigned match_null_string_p : 2;
1579 unsigned is_active : 1;
1580 unsigned matched_something : 1;
1581 unsigned ever_matched_something : 1;
1583 } register_info_type;
1585 #define REG_MATCH_NULL_STRING_P(R) ((R).bits.match_null_string_p)
1586 #define IS_ACTIVE(R) ((R).bits.is_active)
1587 #define MATCHED_SOMETHING(R) ((R).bits.matched_something)
1588 #define EVER_MATCHED_SOMETHING(R) ((R).bits.ever_matched_something)
1591 /* Call this when have matched a real character; it sets `matched' flags
1592 for the subexpressions which we are currently inside. Also records
1593 that those subexprs have matched. */
1594 #define SET_REGS_MATCHED() \
1597 if (!set_regs_matched_done) \
1600 set_regs_matched_done = 1; \
1601 for (r = lowest_active_reg; r <= highest_active_reg; r++) \
1603 MATCHED_SOMETHING (reg_info[r]) \
1604 = EVER_MATCHED_SOMETHING (reg_info[r]) \
1611 /* Registers are set to a sentinel when they haven't yet matched. */
1612 static char reg_unset_dummy;
1613 #define REG_UNSET_VALUE (®_unset_dummy)
1614 #define REG_UNSET(e) ((e) == REG_UNSET_VALUE)
1616 /* Subroutine declarations and macros for regex_compile. */
1618 static reg_errcode_t regex_compile _RE_ARGS ((const char *pattern, size_t size,
1619 reg_syntax_t syntax,
1620 struct re_pattern_buffer *bufp));
1621 static void store_op1 _RE_ARGS ((re_opcode_t op, unsigned char *loc, int arg));
1622 static void store_op2 _RE_ARGS ((re_opcode_t op, unsigned char *loc,
1623 int arg1, int arg2));
1624 static void insert_op1 _RE_ARGS ((re_opcode_t op, unsigned char *loc,
1625 int arg, unsigned char *end));
1626 static void insert_op2 _RE_ARGS ((re_opcode_t op, unsigned char *loc,
1627 int arg1, int arg2, unsigned char *end));
1628 static boolean at_begline_loc_p _RE_ARGS ((const char *pattern, const char *p,
1629 reg_syntax_t syntax));
1630 static boolean at_endline_loc_p _RE_ARGS ((const char *p, const char *pend,
1631 reg_syntax_t syntax));
1632 static reg_errcode_t compile_range _RE_ARGS ((unsigned int range_start,
1636 reg_syntax_t syntax,
1639 /* Fetch the next character in the uncompiled pattern---translating it
1640 if necessary. Also cast from a signed character in the constant
1641 string passed to us by the user to an unsigned char that we can use
1642 as an array index (in, e.g., `translate'). */
1644 # define PATFETCH(c) \
1645 do {if (p == pend) return REG_EEND; \
1646 c = (unsigned char) *p++; \
1647 if (translate) c = (unsigned char) translate[c]; \
1651 /* Fetch the next character in the uncompiled pattern, with no
1653 #define PATFETCH_RAW(c) \
1654 do {if (p == pend) return REG_EEND; \
1655 c = (unsigned char) *p++; \
1658 /* Go backwards one character in the pattern. */
1659 #define PATUNFETCH p--
1662 /* If `translate' is non-null, return translate[D], else just D. We
1663 cast the subscript to translate because some data is declared as
1664 `char *', to avoid warnings when a string constant is passed. But
1665 when we use a character as a subscript we must make it unsigned. */
1667 # define TRANSLATE(d) \
1668 (translate ? (char) translate[(unsigned char) (d)] : (d))
1672 /* Macros for outputting the compiled pattern into `buffer'. */
1674 /* If the buffer isn't allocated when it comes in, use this. */
1675 #define INIT_BUF_SIZE 32
1677 /* Make sure we have at least N more bytes of space in buffer. */
1678 #define GET_BUFFER_SPACE(n) \
1679 while ((unsigned long) (b - bufp->buffer + (n)) > bufp->allocated) \
1682 /* Make sure we have one more byte of buffer space and then add C to it. */
1683 #define BUF_PUSH(c) \
1685 GET_BUFFER_SPACE (1); \
1686 *b++ = (unsigned char) (c); \
1690 /* Ensure we have two more bytes of buffer space and then append C1 and C2. */
1691 #define BUF_PUSH_2(c1, c2) \
1693 GET_BUFFER_SPACE (2); \
1694 *b++ = (unsigned char) (c1); \
1695 *b++ = (unsigned char) (c2); \
1699 /* As with BUF_PUSH_2, except for three bytes. */
1700 #define BUF_PUSH_3(c1, c2, c3) \
1702 GET_BUFFER_SPACE (3); \
1703 *b++ = (unsigned char) (c1); \
1704 *b++ = (unsigned char) (c2); \
1705 *b++ = (unsigned char) (c3); \
1709 /* Store a jump with opcode OP at LOC to location TO. We store a
1710 relative address offset by the three bytes the jump itself occupies. */
1711 #define STORE_JUMP(op, loc, to) \
1712 store_op1 (op, loc, (int) ((to) - (loc) - 3))
1714 /* Likewise, for a two-argument jump. */
1715 #define STORE_JUMP2(op, loc, to, arg) \
1716 store_op2 (op, loc, (int) ((to) - (loc) - 3), arg)
1718 /* Like `STORE_JUMP', but for inserting. Assume `b' is the buffer end. */
1719 #define INSERT_JUMP(op, loc, to) \
1720 insert_op1 (op, loc, (int) ((to) - (loc) - 3), b)
1722 /* Like `STORE_JUMP2', but for inserting. Assume `b' is the buffer end. */
1723 #define INSERT_JUMP2(op, loc, to, arg) \
1724 insert_op2 (op, loc, (int) ((to) - (loc) - 3), arg, b)
1727 /* This is not an arbitrary limit: the arguments which represent offsets
1728 into the pattern are two bytes long. So if 2^16 bytes turns out to
1729 be too small, many things would have to change. */
1730 /* Any other compiler which, like MSC, has allocation limit below 2^16
1731 bytes will have to use approach similar to what was done below for
1732 MSC and drop MAX_BUF_SIZE a bit. Otherwise you may end up
1733 reallocating to 0 bytes. Such thing is not going to work too well.
1734 You have been warned!! */
1735 #if defined _MSC_VER && !defined WIN32
1736 /* Microsoft C 16-bit versions limit malloc to approx 65512 bytes.
1737 The REALLOC define eliminates a flurry of conversion warnings,
1738 but is not required. */
1739 # define MAX_BUF_SIZE 65500L
1740 # define REALLOC(p,s) realloc ((p), (size_t) (s))
1742 # define MAX_BUF_SIZE (1L << 16)
1743 # define REALLOC(p,s) realloc ((p), (s))
1746 /* Extend the buffer by twice its current size via realloc and
1747 reset the pointers that pointed into the old block to point to the
1748 correct places in the new one. If extending the buffer results in it
1749 being larger than MAX_BUF_SIZE, then flag memory exhausted. */
1750 #if __BOUNDED_POINTERS__
1751 # define MOVE_BUFFER_POINTER(P) \
1752 (__ptrhigh (P) = (__ptrlow (P) += incr) + bufp->allocated, \
1753 __ptrvalue (P) += incr)
1755 # define MOVE_BUFFER_POINTER(P) (P) += incr
1757 #define EXTEND_BUFFER() \
1759 unsigned char *old_buffer = bufp->buffer; \
1760 if (bufp->allocated == MAX_BUF_SIZE) \
1762 bufp->allocated <<= 1; \
1763 if (bufp->allocated > MAX_BUF_SIZE) \
1764 bufp->allocated = MAX_BUF_SIZE; \
1765 bufp->buffer = (unsigned char *) REALLOC (bufp->buffer, bufp->allocated);\
1766 if (bufp->buffer == NULL) \
1767 return REG_ESPACE; \
1768 /* If the buffer moved, move all the pointers into it. */ \
1769 if (old_buffer != bufp->buffer) \
1771 int incr = bufp->buffer - old_buffer; \
1772 MOVE_BUFFER_POINTER (b); \
1773 MOVE_BUFFER_POINTER (begalt); \
1774 if (fixup_alt_jump) \
1775 MOVE_BUFFER_POINTER (fixup_alt_jump); \
1777 MOVE_BUFFER_POINTER (laststart); \
1778 if (pending_exact) \
1779 MOVE_BUFFER_POINTER (pending_exact); \
1784 /* Since we have one byte reserved for the register number argument to
1785 {start,stop}_memory, the maximum number of groups we can report
1786 things about is what fits in that byte. */
1787 #define MAX_REGNUM 255
1789 /* But patterns can have more than `MAX_REGNUM' registers. We just
1790 ignore the excess. */
1791 typedef unsigned regnum_t;
1794 /* Macros for the compile stack. */
1796 /* Since offsets can go either forwards or backwards, this type needs to
1797 be able to hold values from -(MAX_BUF_SIZE - 1) to MAX_BUF_SIZE - 1. */
1798 /* int may be not enough when sizeof(int) == 2. */
1799 typedef long pattern_offset_t;
1803 pattern_offset_t begalt_offset;
1804 pattern_offset_t fixup_alt_jump;
1805 pattern_offset_t inner_group_offset;
1806 pattern_offset_t laststart_offset;
1808 } compile_stack_elt_t;
1813 compile_stack_elt_t *stack;
1815 unsigned avail; /* Offset of next open position. */
1816 } compile_stack_type;
1819 #define INIT_COMPILE_STACK_SIZE 32
1821 #define COMPILE_STACK_EMPTY (compile_stack.avail == 0)
1822 #define COMPILE_STACK_FULL (compile_stack.avail == compile_stack.size)
1824 /* The next available element. */
1825 #define COMPILE_STACK_TOP (compile_stack.stack[compile_stack.avail])
1828 /* Set the bit for character C in a list. */
1829 #define SET_LIST_BIT(c) \
1830 (b[((unsigned char) (c)) / BYTEWIDTH] \
1831 |= 1 << (((unsigned char) c) % BYTEWIDTH))
1834 /* Get the next unsigned number in the uncompiled pattern. */
1835 #define GET_UNSIGNED_NUMBER(num) \
1839 while ('0' <= c && c <= '9') \
1843 num = num * 10 + c - '0'; \
1851 #if defined _LIBC || WIDE_CHAR_SUPPORT
1852 /* The GNU C library provides support for user-defined character classes
1853 and the functions from ISO C amendement 1. */
1854 # ifdef CHARCLASS_NAME_MAX
1855 # define CHAR_CLASS_MAX_LENGTH CHARCLASS_NAME_MAX
1857 /* This shouldn't happen but some implementation might still have this
1858 problem. Use a reasonable default value. */
1859 # define CHAR_CLASS_MAX_LENGTH 256
1863 # define IS_CHAR_CLASS(string) __wctype (string)
1865 # define IS_CHAR_CLASS(string) wctype (string)
1868 # define CHAR_CLASS_MAX_LENGTH 6 /* Namely, `xdigit'. */
1870 # define IS_CHAR_CLASS(string) \
1871 (STREQ (string, "alpha") || STREQ (string, "upper") \
1872 || STREQ (string, "lower") || STREQ (string, "digit") \
1873 || STREQ (string, "alnum") || STREQ (string, "xdigit") \
1874 || STREQ (string, "space") || STREQ (string, "print") \
1875 || STREQ (string, "punct") || STREQ (string, "graph") \
1876 || STREQ (string, "cntrl") || STREQ (string, "blank"))
1879 #ifndef MATCH_MAY_ALLOCATE
1881 /* If we cannot allocate large objects within re_match_2_internal,
1882 we make the fail stack and register vectors global.
1883 The fail stack, we grow to the maximum size when a regexp
1885 The register vectors, we adjust in size each time we
1886 compile a regexp, according to the number of registers it needs. */
1888 static fail_stack_type fail_stack;
1890 /* Size with which the following vectors are currently allocated.
1891 That is so we can make them bigger as needed,
1892 but never make them smaller. */
1893 static int regs_allocated_size;
1895 static const char ** regstart, ** regend;
1896 static const char ** old_regstart, ** old_regend;
1897 static const char **best_regstart, **best_regend;
1898 static register_info_type *reg_info;
1899 static const char **reg_dummy;
1900 static register_info_type *reg_info_dummy;
1902 /* Make the register vectors big enough for NUM_REGS registers,
1903 but don't make them smaller. */
1906 regex_grow_registers (num_regs)
1909 if (num_regs > regs_allocated_size)
1911 RETALLOC_IF (regstart, num_regs, const char *);
1912 RETALLOC_IF (regend, num_regs, const char *);
1913 RETALLOC_IF (old_regstart, num_regs, const char *);
1914 RETALLOC_IF (old_regend, num_regs, const char *);
1915 RETALLOC_IF (best_regstart, num_regs, const char *);
1916 RETALLOC_IF (best_regend, num_regs, const char *);
1917 RETALLOC_IF (reg_info, num_regs, register_info_type);
1918 RETALLOC_IF (reg_dummy, num_regs, const char *);
1919 RETALLOC_IF (reg_info_dummy, num_regs, register_info_type);
1921 regs_allocated_size = num_regs;
1925 #endif /* not MATCH_MAY_ALLOCATE */
1927 static boolean group_in_compile_stack _RE_ARGS ((compile_stack_type
1931 /* `regex_compile' compiles PATTERN (of length SIZE) according to SYNTAX.
1932 Returns one of error codes defined in `regex.h', or zero for success.
1934 Assumes the `allocated' (and perhaps `buffer') and `translate'
1935 fields are set in BUFP on entry.
1937 If it succeeds, results are put in BUFP (if it returns an error, the
1938 contents of BUFP are undefined):
1939 `buffer' is the compiled pattern;
1940 `syntax' is set to SYNTAX;
1941 `used' is set to the length of the compiled pattern;
1942 `fastmap_accurate' is zero;
1943 `re_nsub' is the number of subexpressions in PATTERN;
1944 `not_bol' and `not_eol' are zero;
1946 The `fastmap' and `newline_anchor' fields are neither
1947 examined nor set. */
1949 /* Return, freeing storage we allocated. */
1950 #define FREE_STACK_RETURN(value) \
1951 return (free (compile_stack.stack), value)
1953 static reg_errcode_t
1954 regex_compile (pattern, size, syntax, bufp)
1955 const char *pattern;
1957 reg_syntax_t syntax;
1958 struct re_pattern_buffer *bufp;
1960 /* We fetch characters from PATTERN here. Even though PATTERN is
1961 `char *' (i.e., signed), we declare these variables as unsigned, so
1962 they can be reliably used as array indices. */
1963 register unsigned char c, c1;
1965 /* A random temporary spot in PATTERN. */
1968 /* Points to the end of the buffer, where we should append. */
1969 register unsigned char *b;
1971 /* Keeps track of unclosed groups. */
1972 compile_stack_type compile_stack;
1974 /* Points to the current (ending) position in the pattern. */
1975 const char *p = pattern;
1976 const char *pend = pattern + size;
1978 /* How to translate the characters in the pattern. */
1979 RE_TRANSLATE_TYPE translate = bufp->translate;
1981 /* Address of the count-byte of the most recently inserted `exactn'
1982 command. This makes it possible to tell if a new exact-match
1983 character can be added to that command or if the character requires
1984 a new `exactn' command. */
1985 unsigned char *pending_exact = 0;
1987 /* Address of start of the most recently finished expression.
1988 This tells, e.g., postfix * where to find the start of its
1989 operand. Reset at the beginning of groups and alternatives. */
1990 unsigned char *laststart = 0;
1992 /* Address of beginning of regexp, or inside of last group. */
1993 unsigned char *begalt;
1995 /* Place in the uncompiled pattern (i.e., the {) to
1996 which to go back if the interval is invalid. */
1997 const char *beg_interval;
1999 /* Address of the place where a forward jump should go to the end of
2000 the containing expression. Each alternative of an `or' -- except the
2001 last -- ends with a forward jump of this sort. */
2002 unsigned char *fixup_alt_jump = 0;
2004 /* Counts open-groups as they are encountered. Remembered for the
2005 matching close-group on the compile stack, so the same register
2006 number is put in the stop_memory as the start_memory. */
2007 regnum_t regnum = 0;
2010 DEBUG_PRINT1 ("\nCompiling pattern: ");
2013 unsigned debug_count;
2015 for (debug_count = 0; debug_count < size; debug_count++)
2016 putchar (pattern[debug_count]);
2021 /* Initialize the compile stack. */
2022 compile_stack.stack = TALLOC (INIT_COMPILE_STACK_SIZE, compile_stack_elt_t);
2023 if (compile_stack.stack == NULL)
2026 compile_stack.size = INIT_COMPILE_STACK_SIZE;
2027 compile_stack.avail = 0;
2029 /* Initialize the pattern buffer. */
2030 bufp->syntax = syntax;
2031 bufp->fastmap_accurate = 0;
2032 bufp->not_bol = bufp->not_eol = 0;
2034 /* Set `used' to zero, so that if we return an error, the pattern
2035 printer (for debugging) will think there's no pattern. We reset it
2039 /* Always count groups, whether or not bufp->no_sub is set. */
2042 #if !defined emacs && !defined SYNTAX_TABLE
2043 /* Initialize the syntax table. */
2044 init_syntax_once ();
2047 if (bufp->allocated == 0)
2050 { /* If zero allocated, but buffer is non-null, try to realloc
2051 enough space. This loses if buffer's address is bogus, but
2052 that is the user's responsibility. */
2053 RETALLOC (bufp->buffer, INIT_BUF_SIZE, unsigned char);
2056 { /* Caller did not allocate a buffer. Do it for them. */
2057 bufp->buffer = TALLOC (INIT_BUF_SIZE, unsigned char);
2059 if (!bufp->buffer) FREE_STACK_RETURN (REG_ESPACE);
2061 bufp->allocated = INIT_BUF_SIZE;
2064 begalt = b = bufp->buffer;
2066 /* Loop through the uncompiled pattern until we're at the end. */
2075 if ( /* If at start of pattern, it's an operator. */
2077 /* If context independent, it's an operator. */
2078 || syntax & RE_CONTEXT_INDEP_ANCHORS
2079 /* Otherwise, depends on what's come before. */
2080 || at_begline_loc_p (pattern, p, syntax))
2090 if ( /* If at end of pattern, it's an operator. */
2092 /* If context independent, it's an operator. */
2093 || syntax & RE_CONTEXT_INDEP_ANCHORS
2094 /* Otherwise, depends on what's next. */
2095 || at_endline_loc_p (p, pend, syntax))
2105 if ((syntax & RE_BK_PLUS_QM)
2106 || (syntax & RE_LIMITED_OPS))
2110 /* If there is no previous pattern... */
2113 if (syntax & RE_CONTEXT_INVALID_OPS)
2114 FREE_STACK_RETURN (REG_BADRPT);
2115 else if (!(syntax & RE_CONTEXT_INDEP_OPS))
2120 /* Are we optimizing this jump? */
2121 boolean keep_string_p = false;
2123 /* 1 means zero (many) matches is allowed. */
2124 char zero_times_ok = 0, many_times_ok = 0;
2126 /* If there is a sequence of repetition chars, collapse it
2127 down to just one (the right one). We can't combine
2128 interval operators with these because of, e.g., `a{2}*',
2129 which should only match an even number of `a's. */
2133 zero_times_ok |= c != '+';
2134 many_times_ok |= c != '?';
2142 || (!(syntax & RE_BK_PLUS_QM) && (c == '+' || c == '?')))
2145 else if (syntax & RE_BK_PLUS_QM && c == '\\')
2147 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
2150 if (!(c1 == '+' || c1 == '?'))
2165 /* If we get here, we found another repeat character. */
2168 /* Star, etc. applied to an empty pattern is equivalent
2169 to an empty pattern. */
2173 /* Now we know whether or not zero matches is allowed
2174 and also whether or not two or more matches is allowed. */
2176 { /* More than one repetition is allowed, so put in at the
2177 end a backward relative jump from `b' to before the next
2178 jump we're going to put in below (which jumps from
2179 laststart to after this jump).
2181 But if we are at the `*' in the exact sequence `.*\n',
2182 insert an unconditional jump backwards to the .,
2183 instead of the beginning of the loop. This way we only
2184 push a failure point once, instead of every time
2185 through the loop. */
2186 assert (p - 1 > pattern);
2188 /* Allocate the space for the jump. */
2189 GET_BUFFER_SPACE (3);
2191 /* We know we are not at the first character of the pattern,
2192 because laststart was nonzero. And we've already
2193 incremented `p', by the way, to be the character after
2194 the `*'. Do we have to do something analogous here
2195 for null bytes, because of RE_DOT_NOT_NULL? */
2196 if (TRANSLATE (*(p - 2)) == TRANSLATE ('.')
2198 && p < pend && TRANSLATE (*p) == TRANSLATE ('\n')
2199 && !(syntax & RE_DOT_NEWLINE))
2200 { /* We have .*\n. */
2201 STORE_JUMP (jump, b, laststart);
2202 keep_string_p = true;
2205 /* Anything else. */
2206 STORE_JUMP (maybe_pop_jump, b, laststart - 3);
2208 /* We've added more stuff to the buffer. */
2212 /* On failure, jump from laststart to b + 3, which will be the
2213 end of the buffer after this jump is inserted. */
2214 GET_BUFFER_SPACE (3);
2215 INSERT_JUMP (keep_string_p ? on_failure_keep_string_jump
2223 /* At least one repetition is required, so insert a
2224 `dummy_failure_jump' before the initial
2225 `on_failure_jump' instruction of the loop. This
2226 effects a skip over that instruction the first time
2227 we hit that loop. */
2228 GET_BUFFER_SPACE (3);
2229 INSERT_JUMP (dummy_failure_jump, laststart, laststart + 6);
2244 boolean had_char_class = false;
2245 unsigned int range_start = 0xffffffff;
2247 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2249 /* Ensure that we have enough space to push a charset: the
2250 opcode, the length count, and the bitset; 34 bytes in all. */
2251 GET_BUFFER_SPACE (34);
2255 /* We test `*p == '^' twice, instead of using an if
2256 statement, so we only need one BUF_PUSH. */
2257 BUF_PUSH (*p == '^' ? charset_not : charset);
2261 /* Remember the first position in the bracket expression. */
2264 /* Push the number of bytes in the bitmap. */
2265 BUF_PUSH ((1 << BYTEWIDTH) / BYTEWIDTH);
2267 /* Clear the whole map. */
2268 bzero (b, (1 << BYTEWIDTH) / BYTEWIDTH);
2270 /* charset_not matches newline according to a syntax bit. */
2271 if ((re_opcode_t) b[-2] == charset_not
2272 && (syntax & RE_HAT_LISTS_NOT_NEWLINE))
2273 SET_LIST_BIT ('\n');
2275 /* Read in characters and ranges, setting map bits. */
2278 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2282 /* \ might escape characters inside [...] and [^...]. */
2283 if ((syntax & RE_BACKSLASH_ESCAPE_IN_LISTS) && c == '\\')
2285 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
2293 /* Could be the end of the bracket expression. If it's
2294 not (i.e., when the bracket expression is `[]' so
2295 far), the ']' character bit gets set way below. */
2296 if (c == ']' && p != p1 + 1)
2299 /* Look ahead to see if it's a range when the last thing
2300 was a character class. */
2301 if (had_char_class && c == '-' && *p != ']')
2302 FREE_STACK_RETURN (REG_ERANGE);
2304 /* Look ahead to see if it's a range when the last thing
2305 was a character: if this is a hyphen not at the
2306 beginning or the end of a list, then it's the range
2309 && !(p - 2 >= pattern && p[-2] == '[')
2310 && !(p - 3 >= pattern && p[-3] == '[' && p[-2] == '^')
2314 = compile_range (range_start, &p, pend, translate,
2316 if (ret != REG_NOERROR) FREE_STACK_RETURN (ret);
2317 range_start = 0xffffffff;
2320 else if (p[0] == '-' && p[1] != ']')
2321 { /* This handles ranges made up of characters only. */
2324 /* Move past the `-'. */
2327 ret = compile_range (c, &p, pend, translate, syntax, b);
2328 if (ret != REG_NOERROR) FREE_STACK_RETURN (ret);
2329 range_start = 0xffffffff;
2332 /* See if we're at the beginning of a possible character
2335 else if (syntax & RE_CHAR_CLASSES && c == '[' && *p == ':')
2336 { /* Leave room for the null. */
2337 char str[CHAR_CLASS_MAX_LENGTH + 1];
2342 /* If pattern is `[[:'. */
2343 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2348 if ((c == ':' && *p == ']') || p == pend)
2350 if (c1 < CHAR_CLASS_MAX_LENGTH)
2353 /* This is in any case an invalid class name. */
2358 /* If isn't a word bracketed by `[:' and `:]':
2359 undo the ending character, the letters, and leave
2360 the leading `:' and `[' (but set bits for them). */
2361 if (c == ':' && *p == ']')
2363 #if defined _LIBC || WIDE_CHAR_SUPPORT
2364 boolean is_lower = STREQ (str, "lower");
2365 boolean is_upper = STREQ (str, "upper");
2369 wt = IS_CHAR_CLASS (str);
2371 FREE_STACK_RETURN (REG_ECTYPE);
2373 /* Throw away the ] at the end of the character
2377 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2379 for (ch = 0; ch < 1 << BYTEWIDTH; ++ch)
2382 if (__iswctype (__btowc (ch), wt))
2385 if (iswctype (btowc (ch), wt))
2389 if (translate && (is_upper || is_lower)
2390 && (ISUPPER (ch) || ISLOWER (ch)))
2394 had_char_class = true;
2397 boolean is_alnum = STREQ (str, "alnum");
2398 boolean is_alpha = STREQ (str, "alpha");
2399 boolean is_blank = STREQ (str, "blank");
2400 boolean is_cntrl = STREQ (str, "cntrl");
2401 boolean is_digit = STREQ (str, "digit");
2402 boolean is_graph = STREQ (str, "graph");
2403 boolean is_lower = STREQ (str, "lower");
2404 boolean is_print = STREQ (str, "print");
2405 boolean is_punct = STREQ (str, "punct");
2406 boolean is_space = STREQ (str, "space");
2407 boolean is_upper = STREQ (str, "upper");
2408 boolean is_xdigit = STREQ (str, "xdigit");
2410 if (!IS_CHAR_CLASS (str))
2411 FREE_STACK_RETURN (REG_ECTYPE);
2413 /* Throw away the ] at the end of the character
2417 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2419 for (ch = 0; ch < 1 << BYTEWIDTH; ch++)
2421 /* This was split into 3 if's to
2422 avoid an arbitrary limit in some compiler. */
2423 if ( (is_alnum && ISALNUM (ch))
2424 || (is_alpha && ISALPHA (ch))
2425 || (is_blank && ISBLANK (ch))
2426 || (is_cntrl && ISCNTRL (ch)))
2428 if ( (is_digit && ISDIGIT (ch))
2429 || (is_graph && ISGRAPH (ch))
2430 || (is_lower && ISLOWER (ch))
2431 || (is_print && ISPRINT (ch)))
2433 if ( (is_punct && ISPUNCT (ch))
2434 || (is_space && ISSPACE (ch))
2435 || (is_upper && ISUPPER (ch))
2436 || (is_xdigit && ISXDIGIT (ch)))
2438 if ( translate && (is_upper || is_lower)
2439 && (ISUPPER (ch) || ISLOWER (ch)))
2442 had_char_class = true;
2443 #endif /* libc || wctype.h */
2453 had_char_class = false;
2456 else if (syntax & RE_CHAR_CLASSES && c == '[' && *p == '=')
2458 unsigned char str[MB_LEN_MAX + 1];
2461 _NL_CURRENT_WORD (LC_COLLATE, _NL_COLLATE_NRULES);
2467 /* If pattern is `[[='. */
2468 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2473 if ((c == '=' && *p == ']') || p == pend)
2475 if (c1 < MB_LEN_MAX)
2478 /* This is in any case an invalid class name. */
2483 if (c == '=' && *p == ']' && str[0] != '\0')
2485 /* If we have no collation data we use the default
2486 collation in which each character is in a class
2487 by itself. It also means that ASCII is the
2488 character set and therefore we cannot have character
2489 with more than one byte in the multibyte
2496 FREE_STACK_RETURN (REG_ECOLLATE);
2498 /* Throw away the ] at the end of the equivalence
2502 /* Set the bit for the character. */
2503 SET_LIST_BIT (str[0]);
2508 /* Try to match the byte sequence in `str' against
2509 those known to the collate implementation.
2510 First find out whether the bytes in `str' are
2511 actually from exactly one character. */
2512 const int32_t *table;
2513 const unsigned char *weights;
2514 const unsigned char *extra;
2515 const int32_t *indirect;
2517 const unsigned char *cp = str;
2520 /* This #include defines a local function! */
2521 # include <locale/weight.h>
2523 table = (const int32_t *)
2524 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_TABLEMB);
2525 weights = (const unsigned char *)
2526 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_WEIGHTMB);
2527 extra = (const unsigned char *)
2528 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_EXTRAMB);
2529 indirect = (const int32_t *)
2530 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_INDIRECTMB);
2532 idx = findidx (&cp);
2533 if (idx == 0 || cp < str + c1)
2534 /* This is no valid character. */
2535 FREE_STACK_RETURN (REG_ECOLLATE);
2537 /* Throw away the ] at the end of the equivalence
2541 /* Now we have to go throught the whole table
2542 and find all characters which have the same
2545 XXX Note that this is not entirely correct.
2546 we would have to match multibyte sequences
2547 but this is not possible with the current
2549 for (ch = 1; ch < 256; ++ch)
2550 /* XXX This test would have to be changed if we
2551 would allow matching multibyte sequences. */
2554 int32_t idx2 = table[ch];
2555 size_t len = weights[idx2];
2557 /* Test whether the lenghts match. */
2558 if (weights[idx] == len)
2560 /* They do. New compare the bytes of
2565 && (weights[idx + 1 + cnt]
2566 == weights[idx2 + 1 + cnt]))
2570 /* They match. Mark the character as
2577 had_char_class = true;
2587 had_char_class = false;
2590 else if (syntax & RE_CHAR_CLASSES && c == '[' && *p == '.')
2592 unsigned char str[128]; /* Should be large enough. */
2595 _NL_CURRENT_WORD (LC_COLLATE, _NL_COLLATE_NRULES);
2601 /* If pattern is `[[='. */
2602 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2607 if ((c == '.' && *p == ']') || p == pend)
2609 if (c1 < sizeof (str))
2612 /* This is in any case an invalid class name. */
2617 if (c == '.' && *p == ']' && str[0] != '\0')
2619 /* If we have no collation data we use the default
2620 collation in which each character is the name
2621 for its own class which contains only the one
2622 character. It also means that ASCII is the
2623 character set and therefore we cannot have character
2624 with more than one byte in the multibyte
2631 FREE_STACK_RETURN (REG_ECOLLATE);
2633 /* Throw away the ] at the end of the equivalence
2637 /* Set the bit for the character. */
2638 SET_LIST_BIT (str[0]);
2639 range_start = ((const unsigned char *) str)[0];
2644 /* Try to match the byte sequence in `str' against
2645 those known to the collate implementation.
2646 First find out whether the bytes in `str' are
2647 actually from exactly one character. */
2649 const int32_t *symb_table;
2650 const unsigned char *extra;
2657 _NL_CURRENT_WORD (LC_COLLATE,
2658 _NL_COLLATE_SYMB_HASH_SIZEMB);
2659 symb_table = (const int32_t *)
2660 _NL_CURRENT (LC_COLLATE,
2661 _NL_COLLATE_SYMB_TABLEMB);
2662 extra = (const unsigned char *)
2663 _NL_CURRENT (LC_COLLATE,
2664 _NL_COLLATE_SYMB_EXTRAMB);
2666 /* Locate the character in the hashing table. */
2667 hash = elem_hash (str, c1);
2670 elem = hash % table_size;
2671 second = hash % (table_size - 2);
2672 while (symb_table[2 * elem] != 0)
2674 /* First compare the hashing value. */
2675 if (symb_table[2 * elem] == hash
2676 && c1 == extra[symb_table[2 * elem + 1]]
2678 &extra[symb_table[2 * elem + 1]
2682 /* Yep, this is the entry. */
2683 idx = symb_table[2 * elem + 1];
2684 idx += 1 + extra[idx];
2692 if (symb_table[2 * elem] == 0)
2693 /* This is no valid character. */
2694 FREE_STACK_RETURN (REG_ECOLLATE);
2696 /* Throw away the ] at the end of the equivalence
2700 /* Now add the multibyte character(s) we found
2703 XXX Note that this is not entirely correct.
2704 we would have to match multibyte sequences
2705 but this is not possible with the current
2706 implementation. Also, we have to match
2707 collating symbols, which expand to more than
2708 one file, as a whole and not allow the
2709 individual bytes. */
2712 range_start = extra[idx];
2714 SET_LIST_BIT (extra[idx++]);
2717 had_char_class = false;
2727 had_char_class = false;
2732 had_char_class = false;
2738 /* Discard any (non)matching list bytes that are all 0 at the
2739 end of the map. Decrease the map-length byte too. */
2740 while ((int) b[-1] > 0 && b[b[-1] - 1] == 0)
2748 if (syntax & RE_NO_BK_PARENS)
2755 if (syntax & RE_NO_BK_PARENS)
2762 if (syntax & RE_NEWLINE_ALT)
2769 if (syntax & RE_NO_BK_VBAR)
2776 if (syntax & RE_INTERVALS && syntax & RE_NO_BK_BRACES)
2777 goto handle_interval;
2783 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
2785 /* Do not translate the character after the \, so that we can
2786 distinguish, e.g., \B from \b, even if we normally would
2787 translate, e.g., B to b. */
2793 if (syntax & RE_NO_BK_PARENS)
2794 goto normal_backslash;
2800 if (COMPILE_STACK_FULL)
2802 RETALLOC (compile_stack.stack, compile_stack.size << 1,
2803 compile_stack_elt_t);
2804 if (compile_stack.stack == NULL) return REG_ESPACE;
2806 compile_stack.size <<= 1;
2809 /* These are the values to restore when we hit end of this
2810 group. They are all relative offsets, so that if the
2811 whole pattern moves because of realloc, they will still
2813 COMPILE_STACK_TOP.begalt_offset = begalt - bufp->buffer;
2814 COMPILE_STACK_TOP.fixup_alt_jump
2815 = fixup_alt_jump ? fixup_alt_jump - bufp->buffer + 1 : 0;
2816 COMPILE_STACK_TOP.laststart_offset = b - bufp->buffer;
2817 COMPILE_STACK_TOP.regnum = regnum;
2819 /* We will eventually replace the 0 with the number of
2820 groups inner to this one. But do not push a
2821 start_memory for groups beyond the last one we can
2822 represent in the compiled pattern. */
2823 if (regnum <= MAX_REGNUM)
2825 COMPILE_STACK_TOP.inner_group_offset = b - bufp->buffer + 2;
2826 BUF_PUSH_3 (start_memory, regnum, 0);
2829 compile_stack.avail++;
2834 /* If we've reached MAX_REGNUM groups, then this open
2835 won't actually generate any code, so we'll have to
2836 clear pending_exact explicitly. */
2842 if (syntax & RE_NO_BK_PARENS) goto normal_backslash;
2844 if (COMPILE_STACK_EMPTY)
2846 if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD)
2847 goto normal_backslash;
2849 FREE_STACK_RETURN (REG_ERPAREN);
2854 { /* Push a dummy failure point at the end of the
2855 alternative for a possible future
2856 `pop_failure_jump' to pop. See comments at
2857 `push_dummy_failure' in `re_match_2'. */
2858 BUF_PUSH (push_dummy_failure);
2860 /* We allocated space for this jump when we assigned
2861 to `fixup_alt_jump', in the `handle_alt' case below. */
2862 STORE_JUMP (jump_past_alt, fixup_alt_jump, b - 1);
2865 /* See similar code for backslashed left paren above. */
2866 if (COMPILE_STACK_EMPTY)
2868 if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD)
2871 FREE_STACK_RETURN (REG_ERPAREN);
2874 /* Since we just checked for an empty stack above, this
2875 ``can't happen''. */
2876 assert (compile_stack.avail != 0);
2878 /* We don't just want to restore into `regnum', because
2879 later groups should continue to be numbered higher,
2880 as in `(ab)c(de)' -- the second group is #2. */
2881 regnum_t this_group_regnum;
2883 compile_stack.avail--;
2884 begalt = bufp->buffer + COMPILE_STACK_TOP.begalt_offset;
2886 = COMPILE_STACK_TOP.fixup_alt_jump
2887 ? bufp->buffer + COMPILE_STACK_TOP.fixup_alt_jump - 1
2889 laststart = bufp->buffer + COMPILE_STACK_TOP.laststart_offset;
2890 this_group_regnum = COMPILE_STACK_TOP.regnum;
2891 /* If we've reached MAX_REGNUM groups, then this open
2892 won't actually generate any code, so we'll have to
2893 clear pending_exact explicitly. */
2896 /* We're at the end of the group, so now we know how many
2897 groups were inside this one. */
2898 if (this_group_regnum <= MAX_REGNUM)
2900 unsigned char *inner_group_loc
2901 = bufp->buffer + COMPILE_STACK_TOP.inner_group_offset;
2903 *inner_group_loc = regnum - this_group_regnum;
2904 BUF_PUSH_3 (stop_memory, this_group_regnum,
2905 regnum - this_group_regnum);
2911 case '|': /* `\|'. */
2912 if (syntax & RE_LIMITED_OPS || syntax & RE_NO_BK_VBAR)
2913 goto normal_backslash;
2915 if (syntax & RE_LIMITED_OPS)
2918 /* Insert before the previous alternative a jump which
2919 jumps to this alternative if the former fails. */
2920 GET_BUFFER_SPACE (3);
2921 INSERT_JUMP (on_failure_jump, begalt, b + 6);
2925 /* The alternative before this one has a jump after it
2926 which gets executed if it gets matched. Adjust that
2927 jump so it will jump to this alternative's analogous
2928 jump (put in below, which in turn will jump to the next
2929 (if any) alternative's such jump, etc.). The last such
2930 jump jumps to the correct final destination. A picture:
2936 If we are at `b', then fixup_alt_jump right now points to a
2937 three-byte space after `a'. We'll put in the jump, set
2938 fixup_alt_jump to right after `b', and leave behind three
2939 bytes which we'll fill in when we get to after `c'. */
2942 STORE_JUMP (jump_past_alt, fixup_alt_jump, b);
2944 /* Mark and leave space for a jump after this alternative,
2945 to be filled in later either by next alternative or
2946 when know we're at the end of a series of alternatives. */
2948 GET_BUFFER_SPACE (3);
2957 /* If \{ is a literal. */
2958 if (!(syntax & RE_INTERVALS)
2959 /* If we're at `\{' and it's not the open-interval
2961 || (syntax & RE_NO_BK_BRACES))
2962 goto normal_backslash;
2966 /* If got here, then the syntax allows intervals. */
2968 /* At least (most) this many matches must be made. */
2969 int lower_bound = -1, upper_bound = -1;
2971 beg_interval = p - 1;
2975 if (!(syntax & RE_INTERVALS) && (syntax & RE_NO_BK_BRACES))
2976 goto unfetch_interval;
2978 FREE_STACK_RETURN (REG_EBRACE);
2981 GET_UNSIGNED_NUMBER (lower_bound);
2985 GET_UNSIGNED_NUMBER (upper_bound);
2986 if ((!(syntax & RE_NO_BK_BRACES) && c != '\\')
2987 || ((syntax & RE_NO_BK_BRACES) && c != '}'))
2988 FREE_STACK_RETURN (REG_BADBR);
2990 if (upper_bound < 0)
2991 upper_bound = RE_DUP_MAX;
2994 /* Interval such as `{1}' => match exactly once. */
2995 upper_bound = lower_bound;
2997 if (lower_bound < 0 || upper_bound > RE_DUP_MAX
2998 || lower_bound > upper_bound)
3000 if (!(syntax & RE_INTERVALS) && (syntax & RE_NO_BK_BRACES))
3001 goto unfetch_interval;
3003 FREE_STACK_RETURN (REG_BADBR);
3006 if (!(syntax & RE_NO_BK_BRACES))
3008 if (c != '\\') FREE_STACK_RETURN (REG_EBRACE);
3015 if (!(syntax & RE_INTERVALS) && (syntax & RE_NO_BK_BRACES))
3016 goto unfetch_interval;
3018 FREE_STACK_RETURN (REG_BADBR);
3021 /* We just parsed a valid interval. */
3023 /* If it's invalid to have no preceding re. */
3026 if (syntax & RE_CONTEXT_INVALID_OPS)
3027 FREE_STACK_RETURN (REG_BADRPT);
3028 else if (syntax & RE_CONTEXT_INDEP_OPS)
3031 goto unfetch_interval;
3034 /* If the upper bound is zero, don't want to succeed at
3035 all; jump from `laststart' to `b + 3', which will be
3036 the end of the buffer after we insert the jump. */
3037 if (upper_bound == 0)
3039 GET_BUFFER_SPACE (3);
3040 INSERT_JUMP (jump, laststart, b + 3);
3044 /* Otherwise, we have a nontrivial interval. When
3045 we're all done, the pattern will look like:
3046 set_number_at <jump count> <upper bound>
3047 set_number_at <succeed_n count> <lower bound>
3048 succeed_n <after jump addr> <succeed_n count>
3050 jump_n <succeed_n addr> <jump count>
3051 (The upper bound and `jump_n' are omitted if
3052 `upper_bound' is 1, though.) */
3054 { /* If the upper bound is > 1, we need to insert
3055 more at the end of the loop. */
3056 unsigned nbytes = 10 + (upper_bound > 1) * 10;
3058 GET_BUFFER_SPACE (nbytes);
3060 /* Initialize lower bound of the `succeed_n', even
3061 though it will be set during matching by its
3062 attendant `set_number_at' (inserted next),
3063 because `re_compile_fastmap' needs to know.
3064 Jump to the `jump_n' we might insert below. */
3065 INSERT_JUMP2 (succeed_n, laststart,
3066 b + 5 + (upper_bound > 1) * 5,
3070 /* Code to initialize the lower bound. Insert
3071 before the `succeed_n'. The `5' is the last two
3072 bytes of this `set_number_at', plus 3 bytes of
3073 the following `succeed_n'. */
3074 insert_op2 (set_number_at, laststart, 5, lower_bound, b);
3077 if (upper_bound > 1)
3078 { /* More than one repetition is allowed, so
3079 append a backward jump to the `succeed_n'
3080 that starts this interval.
3082 When we've reached this during matching,
3083 we'll have matched the interval once, so
3084 jump back only `upper_bound - 1' times. */
3085 STORE_JUMP2 (jump_n, b, laststart + 5,
3089 /* The location we want to set is the second
3090 parameter of the `jump_n'; that is `b-2' as
3091 an absolute address. `laststart' will be
3092 the `set_number_at' we're about to insert;
3093 `laststart+3' the number to set, the source
3094 for the relative address. But we are
3095 inserting into the middle of the pattern --
3096 so everything is getting moved up by 5.
3097 Conclusion: (b - 2) - (laststart + 3) + 5,
3098 i.e., b - laststart.
3100 We insert this at the beginning of the loop
3101 so that if we fail during matching, we'll
3102 reinitialize the bounds. */
3103 insert_op2 (set_number_at, laststart, b - laststart,
3104 upper_bound - 1, b);
3109 beg_interval = NULL;
3114 /* If an invalid interval, match the characters as literals. */
3115 assert (beg_interval);
3117 beg_interval = NULL;
3119 /* normal_char and normal_backslash need `c'. */
3122 if (!(syntax & RE_NO_BK_BRACES))
3124 if (p > pattern && p[-1] == '\\')
3125 goto normal_backslash;
3130 /* There is no way to specify the before_dot and after_dot
3131 operators. rms says this is ok. --karl */
3139 BUF_PUSH_2 (syntaxspec, syntax_spec_code[c]);
3145 BUF_PUSH_2 (notsyntaxspec, syntax_spec_code[c]);
3151 if (syntax & RE_NO_GNU_OPS)
3154 BUF_PUSH (wordchar);
3159 if (syntax & RE_NO_GNU_OPS)
3162 BUF_PUSH (notwordchar);
3167 if (syntax & RE_NO_GNU_OPS)
3173 if (syntax & RE_NO_GNU_OPS)
3179 if (syntax & RE_NO_GNU_OPS)
3181 BUF_PUSH (wordbound);
3185 if (syntax & RE_NO_GNU_OPS)
3187 BUF_PUSH (notwordbound);
3191 if (syntax & RE_NO_GNU_OPS)
3197 if (syntax & RE_NO_GNU_OPS)
3202 case '1': case '2': case '3': case '4': case '5':
3203 case '6': case '7': case '8': case '9':
3204 if (syntax & RE_NO_BK_REFS)
3210 FREE_STACK_RETURN (REG_ESUBREG);
3212 /* Can't back reference to a subexpression if inside of it. */
3213 if (group_in_compile_stack (compile_stack, (regnum_t) c1))
3217 BUF_PUSH_2 (duplicate, c1);
3223 if (syntax & RE_BK_PLUS_QM)
3226 goto normal_backslash;
3230 /* You might think it would be useful for \ to mean
3231 not to translate; but if we don't translate it
3232 it will never match anything. */
3240 /* Expects the character in `c'. */
3242 /* If no exactn currently being built. */
3245 /* If last exactn not at current position. */
3246 || pending_exact + *pending_exact + 1 != b
3248 /* We have only one byte following the exactn for the count. */
3249 || *pending_exact == (1 << BYTEWIDTH) - 1
3251 /* If followed by a repetition operator. */
3252 || *p == '*' || *p == '^'
3253 || ((syntax & RE_BK_PLUS_QM)
3254 ? *p == '\\' && (p[1] == '+' || p[1] == '?')
3255 : (*p == '+' || *p == '?'))
3256 || ((syntax & RE_INTERVALS)
3257 && ((syntax & RE_NO_BK_BRACES)
3259 : (p[0] == '\\' && p[1] == '{'))))
3261 /* Start building a new exactn. */
3265 BUF_PUSH_2 (exactn, 0);
3266 pending_exact = b - 1;
3273 } /* while p != pend */
3276 /* Through the pattern now. */
3279 STORE_JUMP (jump_past_alt, fixup_alt_jump, b);
3281 if (!COMPILE_STACK_EMPTY)
3282 FREE_STACK_RETURN (REG_EPAREN);
3284 /* If we don't want backtracking, force success
3285 the first time we reach the end of the compiled pattern. */
3286 if (syntax & RE_NO_POSIX_BACKTRACKING)
3289 free (compile_stack.stack);
3291 /* We have succeeded; set the length of the buffer. */
3292 bufp->used = b - bufp->buffer;
3297 DEBUG_PRINT1 ("\nCompiled pattern: \n");
3298 print_compiled_pattern (bufp);
3302 #ifndef MATCH_MAY_ALLOCATE
3303 /* Initialize the failure stack to the largest possible stack. This
3304 isn't necessary unless we're trying to avoid calling alloca in
3305 the search and match routines. */
3307 int num_regs = bufp->re_nsub + 1;
3309 /* Since DOUBLE_FAIL_STACK refuses to double only if the current size
3310 is strictly greater than re_max_failures, the largest possible stack
3311 is 2 * re_max_failures failure points. */
3312 if (fail_stack.size < (2 * re_max_failures * MAX_FAILURE_ITEMS))
3314 fail_stack.size = (2 * re_max_failures * MAX_FAILURE_ITEMS);
3317 if (! fail_stack.stack)
3319 = (fail_stack_elt_t *) xmalloc (fail_stack.size
3320 * sizeof (fail_stack_elt_t));
3323 = (fail_stack_elt_t *) xrealloc (fail_stack.stack,
3325 * sizeof (fail_stack_elt_t)));
3326 # else /* not emacs */
3327 if (! fail_stack.stack)
3329 = (fail_stack_elt_t *) malloc (fail_stack.size
3330 * sizeof (fail_stack_elt_t));
3333 = (fail_stack_elt_t *) realloc (fail_stack.stack,
3335 * sizeof (fail_stack_elt_t)));
3336 # endif /* not emacs */
3339 regex_grow_registers (num_regs);
3341 #endif /* not MATCH_MAY_ALLOCATE */
3344 } /* regex_compile */
3346 /* Subroutines for `regex_compile'. */
3348 /* Store OP at LOC followed by two-byte integer parameter ARG. */
3351 store_op1 (op, loc, arg)
3356 *loc = (unsigned char) op;
3357 STORE_NUMBER (loc + 1, arg);
3361 /* Like `store_op1', but for two two-byte parameters ARG1 and ARG2. */
3364 store_op2 (op, loc, arg1, arg2)
3369 *loc = (unsigned char) op;
3370 STORE_NUMBER (loc + 1, arg1);
3371 STORE_NUMBER (loc + 3, arg2);
3375 /* Copy the bytes from LOC to END to open up three bytes of space at LOC
3376 for OP followed by two-byte integer parameter ARG. */
3379 insert_op1 (op, loc, arg, end)
3385 register unsigned char *pfrom = end;
3386 register unsigned char *pto = end + 3;
3388 while (pfrom != loc)
3391 store_op1 (op, loc, arg);
3395 /* Like `insert_op1', but for two two-byte parameters ARG1 and ARG2. */
3398 insert_op2 (op, loc, arg1, arg2, end)
3404 register unsigned char *pfrom = end;
3405 register unsigned char *pto = end + 5;
3407 while (pfrom != loc)
3410 store_op2 (op, loc, arg1, arg2);
3414 /* P points to just after a ^ in PATTERN. Return true if that ^ comes
3415 after an alternative or a begin-subexpression. We assume there is at
3416 least one character before the ^. */
3419 at_begline_loc_p (pattern, p, syntax)
3420 const char *pattern, *p;
3421 reg_syntax_t syntax;
3423 const char *prev = p - 2;
3424 boolean prev_prev_backslash = prev > pattern && prev[-1] == '\\';
3427 /* After a subexpression? */
3428 (*prev == '(' && (syntax & RE_NO_BK_PARENS || prev_prev_backslash))
3429 /* After an alternative? */
3430 || (*prev == '|' && (syntax & RE_NO_BK_VBAR || prev_prev_backslash));
3434 /* The dual of at_begline_loc_p. This one is for $. We assume there is
3435 at least one character after the $, i.e., `P < PEND'. */
3438 at_endline_loc_p (p, pend, syntax)
3439 const char *p, *pend;
3440 reg_syntax_t syntax;
3442 const char *next = p;
3443 boolean next_backslash = *next == '\\';
3444 const char *next_next = p + 1 < pend ? p + 1 : 0;
3447 /* Before a subexpression? */
3448 (syntax & RE_NO_BK_PARENS ? *next == ')'
3449 : next_backslash && next_next && *next_next == ')')
3450 /* Before an alternative? */
3451 || (syntax & RE_NO_BK_VBAR ? *next == '|'
3452 : next_backslash && next_next && *next_next == '|');
3456 /* Returns true if REGNUM is in one of COMPILE_STACK's elements and
3457 false if it's not. */
3460 group_in_compile_stack (compile_stack, regnum)
3461 compile_stack_type compile_stack;
3466 for (this_element = compile_stack.avail - 1;
3469 if (compile_stack.stack[this_element].regnum == regnum)
3476 /* Read the ending character of a range (in a bracket expression) from the
3477 uncompiled pattern *P_PTR (which ends at PEND). We assume the
3478 starting character is in `P[-2]'. (`P[-1]' is the character `-'.)
3479 Then we set the translation of all bits between the starting and
3480 ending characters (inclusive) in the compiled pattern B.
3482 Return an error code.
3484 We use these short variable names so we can use the same macros as
3485 `regex_compile' itself. */
3487 static reg_errcode_t
3488 compile_range (range_start_char, p_ptr, pend, translate, syntax, b)
3489 unsigned int range_start_char;
3490 const char **p_ptr, *pend;
3491 RE_TRANSLATE_TYPE translate;
3492 reg_syntax_t syntax;
3497 const char *p = *p_ptr;
3499 char range_start[2];
3506 /* Fetch the endpoints without translating them; the
3507 appropriate translation is done in the bit-setting loop below. */
3508 range_start[0] = TRANSLATE (range_start_char);
3509 range_start[1] = '\0';
3510 range_end[0] = TRANSLATE (p[0]);
3511 range_end[1] = '\0';
3513 /* Have to increment the pointer into the pattern string, so the
3514 caller isn't still at the ending character. */
3517 /* Report an error if the range is empty and the syntax prohibits this. */
3518 ret = syntax & RE_NO_EMPTY_RANGES ? REG_ERANGE : REG_NOERROR;
3520 /* Here we see why `this_char' has to be larger than an `unsigned
3521 char' -- we would otherwise go into an infinite loop, since all
3522 characters <= 0xff. */
3524 for (this_char = 0; this_char <= (unsigned char) -1; ++this_char)
3527 if (strcoll (range_start, ch) <= 0 && strcoll (ch, range_end) <= 0)
3529 SET_LIST_BIT (TRANSLATE (this_char));
3537 /* re_compile_fastmap computes a ``fastmap'' for the compiled pattern in
3538 BUFP. A fastmap records which of the (1 << BYTEWIDTH) possible
3539 characters can start a string that matches the pattern. This fastmap
3540 is used by re_search to skip quickly over impossible starting points.
3542 The caller must supply the address of a (1 << BYTEWIDTH)-byte data
3543 area as BUFP->fastmap.
3545 We set the `fastmap', `fastmap_accurate', and `can_be_null' fields in
3548 Returns 0 if we succeed, -2 if an internal error. */
3551 re_compile_fastmap (bufp)
3552 struct re_pattern_buffer *bufp;
3555 #ifdef MATCH_MAY_ALLOCATE
3556 fail_stack_type fail_stack;
3558 #ifndef REGEX_MALLOC
3562 register char *fastmap = bufp->fastmap;
3563 unsigned char *pattern = bufp->buffer;
3564 unsigned char *p = pattern;
3565 register unsigned char *pend = pattern + bufp->used;
3568 /* This holds the pointer to the failure stack, when
3569 it is allocated relocatably. */
3570 fail_stack_elt_t *failure_stack_ptr;
3573 /* Assume that each path through the pattern can be null until
3574 proven otherwise. We set this false at the bottom of switch
3575 statement, to which we get only if a particular path doesn't
3576 match the empty string. */
3577 boolean path_can_be_null = true;
3579 /* We aren't doing a `succeed_n' to begin with. */
3580 boolean succeed_n_p = false;
3582 assert (fastmap != NULL && p != NULL);
3585 bzero (fastmap, 1 << BYTEWIDTH); /* Assume nothing's valid. */
3586 bufp->fastmap_accurate = 1; /* It will be when we're done. */
3587 bufp->can_be_null = 0;
3591 if (p == pend || *p == succeed)
3593 /* We have reached the (effective) end of pattern. */
3594 if (!FAIL_STACK_EMPTY ())
3596 bufp->can_be_null |= path_can_be_null;
3598 /* Reset for next path. */
3599 path_can_be_null = true;
3601 p = fail_stack.stack[--fail_stack.avail].pointer;
3609 /* We should never be about to go beyond the end of the pattern. */
3612 switch (SWITCH_ENUM_CAST ((re_opcode_t) *p++))
3615 /* I guess the idea here is to simply not bother with a fastmap
3616 if a backreference is used, since it's too hard to figure out
3617 the fastmap for the corresponding group. Setting
3618 `can_be_null' stops `re_search_2' from using the fastmap, so
3619 that is all we do. */
3621 bufp->can_be_null = 1;
3625 /* Following are the cases which match a character. These end
3634 for (j = *p++ * BYTEWIDTH - 1; j >= 0; j--)
3635 if (p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH)))
3641 /* Chars beyond end of map must be allowed. */
3642 for (j = *p * BYTEWIDTH; j < (1 << BYTEWIDTH); j++)
3645 for (j = *p++ * BYTEWIDTH - 1; j >= 0; j--)
3646 if (!(p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH))))
3652 for (j = 0; j < (1 << BYTEWIDTH); j++)
3653 if (SYNTAX (j) == Sword)
3659 for (j = 0; j < (1 << BYTEWIDTH); j++)
3660 if (SYNTAX (j) != Sword)
3667 int fastmap_newline = fastmap['\n'];
3669 /* `.' matches anything ... */
3670 for (j = 0; j < (1 << BYTEWIDTH); j++)
3673 /* ... except perhaps newline. */
3674 if (!(bufp->syntax & RE_DOT_NEWLINE))
3675 fastmap['\n'] = fastmap_newline;
3677 /* Return if we have already set `can_be_null'; if we have,
3678 then the fastmap is irrelevant. Something's wrong here. */
3679 else if (bufp->can_be_null)
3682 /* Otherwise, have to check alternative paths. */
3689 for (j = 0; j < (1 << BYTEWIDTH); j++)
3690 if (SYNTAX (j) == (enum syntaxcode) k)
3697 for (j = 0; j < (1 << BYTEWIDTH); j++)
3698 if (SYNTAX (j) != (enum syntaxcode) k)
3703 /* All cases after this match the empty string. These end with
3723 case push_dummy_failure:
3728 case pop_failure_jump:
3729 case maybe_pop_jump:
3732 case dummy_failure_jump:
3733 EXTRACT_NUMBER_AND_INCR (j, p);
3738 /* Jump backward implies we just went through the body of a
3739 loop and matched nothing. Opcode jumped to should be
3740 `on_failure_jump' or `succeed_n'. Just treat it like an
3741 ordinary jump. For a * loop, it has pushed its failure
3742 point already; if so, discard that as redundant. */
3743 if ((re_opcode_t) *p != on_failure_jump
3744 && (re_opcode_t) *p != succeed_n)
3748 EXTRACT_NUMBER_AND_INCR (j, p);
3751 /* If what's on the stack is where we are now, pop it. */
3752 if (!FAIL_STACK_EMPTY ()
3753 && fail_stack.stack[fail_stack.avail - 1].pointer == p)
3759 case on_failure_jump:
3760 case on_failure_keep_string_jump:
3761 handle_on_failure_jump:
3762 EXTRACT_NUMBER_AND_INCR (j, p);
3764 /* For some patterns, e.g., `(a?)?', `p+j' here points to the
3765 end of the pattern. We don't want to push such a point,
3766 since when we restore it above, entering the switch will
3767 increment `p' past the end of the pattern. We don't need
3768 to push such a point since we obviously won't find any more
3769 fastmap entries beyond `pend'. Such a pattern can match
3770 the null string, though. */
3773 if (!PUSH_PATTERN_OP (p + j, fail_stack))
3775 RESET_FAIL_STACK ();
3780 bufp->can_be_null = 1;
3784 EXTRACT_NUMBER_AND_INCR (k, p); /* Skip the n. */
3785 succeed_n_p = false;
3792 /* Get to the number of times to succeed. */
3795 /* Increment p past the n for when k != 0. */
3796 EXTRACT_NUMBER_AND_INCR (k, p);
3800 succeed_n_p = true; /* Spaghetti code alert. */
3801 goto handle_on_failure_jump;
3818 abort (); /* We have listed all the cases. */
3821 /* Getting here means we have found the possible starting
3822 characters for one path of the pattern -- and that the empty
3823 string does not match. We need not follow this path further.
3824 Instead, look at the next alternative (remembered on the
3825 stack), or quit if no more. The test at the top of the loop
3826 does these things. */
3827 path_can_be_null = false;
3831 /* Set `can_be_null' for the last path (also the first path, if the
3832 pattern is empty). */
3833 bufp->can_be_null |= path_can_be_null;
3836 RESET_FAIL_STACK ();
3838 } /* re_compile_fastmap */
3840 weak_alias (__re_compile_fastmap, re_compile_fastmap)
3843 /* Set REGS to hold NUM_REGS registers, storing them in STARTS and
3844 ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use
3845 this memory for recording register information. STARTS and ENDS
3846 must be allocated using the malloc library routine, and must each
3847 be at least NUM_REGS * sizeof (regoff_t) bytes long.
3849 If NUM_REGS == 0, then subsequent matches should allocate their own
3852 Unless this function is called, the first search or match using
3853 PATTERN_BUFFER will allocate its own register data, without
3854 freeing the old data. */
3857 re_set_registers (bufp, regs, num_regs, starts, ends)
3858 struct re_pattern_buffer *bufp;
3859 struct re_registers *regs;
3861 regoff_t *starts, *ends;
3865 bufp->regs_allocated = REGS_REALLOCATE;
3866 regs->num_regs = num_regs;
3867 regs->start = starts;
3872 bufp->regs_allocated = REGS_UNALLOCATED;
3874 regs->start = regs->end = (regoff_t *) 0;
3878 weak_alias (__re_set_registers, re_set_registers)
3881 /* Searching routines. */
3883 /* Like re_search_2, below, but only one string is specified, and
3884 doesn't let you say where to stop matching. */
3887 re_search (bufp, string, size, startpos, range, regs)
3888 struct re_pattern_buffer *bufp;
3890 int size, startpos, range;
3891 struct re_registers *regs;
3893 return re_search_2 (bufp, NULL, 0, string, size, startpos, range,
3897 weak_alias (__re_search, re_search)
3901 /* Using the compiled pattern in BUFP->buffer, first tries to match the
3902 virtual concatenation of STRING1 and STRING2, starting first at index
3903 STARTPOS, then at STARTPOS + 1, and so on.
3905 STRING1 and STRING2 have length SIZE1 and SIZE2, respectively.
3907 RANGE is how far to scan while trying to match. RANGE = 0 means try
3908 only at STARTPOS; in general, the last start tried is STARTPOS +
3911 In REGS, return the indices of the virtual concatenation of STRING1
3912 and STRING2 that matched the entire BUFP->buffer and its contained
3915 Do not consider matching one past the index STOP in the virtual
3916 concatenation of STRING1 and STRING2.
3918 We return either the position in the strings at which the match was
3919 found, -1 if no match, or -2 if error (such as failure
3923 re_search_2 (bufp, string1, size1, string2, size2, startpos, range, regs, stop)
3924 struct re_pattern_buffer *bufp;
3925 const char *string1, *string2;
3929 struct re_registers *regs;
3933 register char *fastmap = bufp->fastmap;
3934 register RE_TRANSLATE_TYPE translate = bufp->translate;
3935 int total_size = size1 + size2;
3936 int endpos = startpos + range;
3938 /* Check for out-of-range STARTPOS. */
3939 if (startpos < 0 || startpos > total_size)
3942 /* Fix up RANGE if it might eventually take us outside
3943 the virtual concatenation of STRING1 and STRING2.
3944 Make sure we won't move STARTPOS below 0 or above TOTAL_SIZE. */
3946 range = 0 - startpos;
3947 else if (endpos > total_size)
3948 range = total_size - startpos;
3950 /* If the search isn't to be a backwards one, don't waste time in a
3951 search for a pattern that must be anchored. */
3952 if (bufp->used > 0 && range > 0
3953 && ((re_opcode_t) bufp->buffer[0] == begbuf
3954 /* `begline' is like `begbuf' if it cannot match at newlines. */
3955 || ((re_opcode_t) bufp->buffer[0] == begline
3956 && !bufp->newline_anchor)))
3965 /* In a forward search for something that starts with \=.
3966 don't keep searching past point. */
3967 if (bufp->used > 0 && (re_opcode_t) bufp->buffer[0] == at_dot && range > 0)
3969 range = PT - startpos;
3975 /* Update the fastmap now if not correct already. */
3976 if (fastmap && !bufp->fastmap_accurate)
3977 if (re_compile_fastmap (bufp) == -2)
3980 /* Loop through the string, looking for a place to start matching. */
3983 /* If a fastmap is supplied, skip quickly over characters that
3984 cannot be the start of a match. If the pattern can match the
3985 null string, however, we don't need to skip characters; we want
3986 the first null string. */
3987 if (fastmap && startpos < total_size && !bufp->can_be_null)
3989 if (range > 0) /* Searching forwards. */
3991 register const char *d;
3992 register int lim = 0;
3995 if (startpos < size1 && startpos + range >= size1)
3996 lim = range - (size1 - startpos);
3998 d = (startpos >= size1 ? string2 - size1 : string1) + startpos;
4000 /* Written out as an if-else to avoid testing `translate'
4004 && !fastmap[(unsigned char)
4005 translate[(unsigned char) *d++]])
4008 while (range > lim && !fastmap[(unsigned char) *d++])
4011 startpos += irange - range;
4013 else /* Searching backwards. */
4015 register char c = (size1 == 0 || startpos >= size1
4016 ? string2[startpos - size1]
4017 : string1[startpos]);
4019 if (!fastmap[(unsigned char) TRANSLATE (c)])
4024 /* If can't match the null string, and that's all we have left, fail. */
4025 if (range >= 0 && startpos == total_size && fastmap
4026 && !bufp->can_be_null)
4029 val = re_match_2_internal (bufp, string1, size1, string2, size2,
4030 startpos, regs, stop);
4031 #ifndef REGEX_MALLOC
4060 weak_alias (__re_search_2, re_search_2)
4063 /* This converts PTR, a pointer into one of the search strings `string1'
4064 and `string2' into an offset from the beginning of that string. */
4065 #define POINTER_TO_OFFSET(ptr) \
4066 (FIRST_STRING_P (ptr) \
4067 ? ((regoff_t) ((ptr) - string1)) \
4068 : ((regoff_t) ((ptr) - string2 + size1)))
4070 /* Macros for dealing with the split strings in re_match_2. */
4072 #define MATCHING_IN_FIRST_STRING (dend == end_match_1)
4074 /* Call before fetching a character with *d. This switches over to
4075 string2 if necessary. */
4076 #define PREFETCH() \
4079 /* End of string2 => fail. */ \
4080 if (dend == end_match_2) \
4082 /* End of string1 => advance to string2. */ \
4084 dend = end_match_2; \
4088 /* Test if at very beginning or at very end of the virtual concatenation
4089 of `string1' and `string2'. If only one string, it's `string2'. */
4090 #define AT_STRINGS_BEG(d) ((d) == (size1 ? string1 : string2) || !size2)
4091 #define AT_STRINGS_END(d) ((d) == end2)
4094 /* Test if D points to a character which is word-constituent. We have
4095 two special cases to check for: if past the end of string1, look at
4096 the first character in string2; and if before the beginning of
4097 string2, look at the last character in string1. */
4098 #define WORDCHAR_P(d) \
4099 (SYNTAX ((d) == end1 ? *string2 \
4100 : (d) == string2 - 1 ? *(end1 - 1) : *(d)) \
4103 /* Disabled due to a compiler bug -- see comment at case wordbound */
4105 /* Test if the character before D and the one at D differ with respect
4106 to being word-constituent. */
4107 #define AT_WORD_BOUNDARY(d) \
4108 (AT_STRINGS_BEG (d) || AT_STRINGS_END (d) \
4109 || WORDCHAR_P (d - 1) != WORDCHAR_P (d))
4112 /* Free everything we malloc. */
4113 #ifdef MATCH_MAY_ALLOCATE
4114 # define FREE_VAR(var) if (var) REGEX_FREE (var); var = NULL
4115 # define FREE_VARIABLES() \
4117 REGEX_FREE_STACK (fail_stack.stack); \
4118 FREE_VAR (regstart); \
4119 FREE_VAR (regend); \
4120 FREE_VAR (old_regstart); \
4121 FREE_VAR (old_regend); \
4122 FREE_VAR (best_regstart); \
4123 FREE_VAR (best_regend); \
4124 FREE_VAR (reg_info); \
4125 FREE_VAR (reg_dummy); \
4126 FREE_VAR (reg_info_dummy); \
4129 # define FREE_VARIABLES() ((void)0) /* Do nothing! But inhibit gcc warning. */
4130 #endif /* not MATCH_MAY_ALLOCATE */
4132 /* These values must meet several constraints. They must not be valid
4133 register values; since we have a limit of 255 registers (because
4134 we use only one byte in the pattern for the register number), we can
4135 use numbers larger than 255. They must differ by 1, because of
4136 NUM_FAILURE_ITEMS above. And the value for the lowest register must
4137 be larger than the value for the highest register, so we do not try
4138 to actually save any registers when none are active. */
4139 #define NO_HIGHEST_ACTIVE_REG (1 << BYTEWIDTH)
4140 #define NO_LOWEST_ACTIVE_REG (NO_HIGHEST_ACTIVE_REG + 1)
4142 /* Matching routines. */
4144 #ifndef emacs /* Emacs never uses this. */
4145 /* re_match is like re_match_2 except it takes only a single string. */
4148 re_match (bufp, string, size, pos, regs)
4149 struct re_pattern_buffer *bufp;
4152 struct re_registers *regs;
4154 int result = re_match_2_internal (bufp, NULL, 0, string, size,
4156 # ifndef REGEX_MALLOC
4164 weak_alias (__re_match, re_match)
4166 #endif /* not emacs */
4168 static boolean group_match_null_string_p _RE_ARGS ((unsigned char **p,
4170 register_info_type *reg_info));
4171 static boolean alt_match_null_string_p _RE_ARGS ((unsigned char *p,
4173 register_info_type *reg_info));
4174 static boolean common_op_match_null_string_p _RE_ARGS ((unsigned char **p,
4176 register_info_type *reg_info));
4177 static int bcmp_translate _RE_ARGS ((const char *s1, const char *s2,
4178 int len, char *translate));
4180 /* re_match_2 matches the compiled pattern in BUFP against the
4181 the (virtual) concatenation of STRING1 and STRING2 (of length SIZE1
4182 and SIZE2, respectively). We start matching at POS, and stop
4185 If REGS is non-null and the `no_sub' field of BUFP is nonzero, we
4186 store offsets for the substring each group matched in REGS. See the
4187 documentation for exactly how many groups we fill.
4189 We return -1 if no match, -2 if an internal error (such as the
4190 failure stack overflowing). Otherwise, we return the length of the
4191 matched substring. */
4194 re_match_2 (bufp, string1, size1, string2, size2, pos, regs, stop)
4195 struct re_pattern_buffer *bufp;
4196 const char *string1, *string2;
4199 struct re_registers *regs;
4202 int result = re_match_2_internal (bufp, string1, size1, string2, size2,
4204 #ifndef REGEX_MALLOC
4212 weak_alias (__re_match_2, re_match_2)
4215 /* This is a separate function so that we can force an alloca cleanup
4218 re_match_2_internal (bufp, string1, size1, string2, size2, pos, regs, stop)
4219 struct re_pattern_buffer *bufp;
4220 const char *string1, *string2;
4223 struct re_registers *regs;
4226 /* General temporaries. */
4230 /* Just past the end of the corresponding string. */
4231 const char *end1, *end2;
4233 /* Pointers into string1 and string2, just past the last characters in
4234 each to consider matching. */
4235 const char *end_match_1, *end_match_2;
4237 /* Where we are in the data, and the end of the current string. */
4238 const char *d, *dend;
4240 /* Where we are in the pattern, and the end of the pattern. */
4241 unsigned char *p = bufp->buffer;
4242 register unsigned char *pend = p + bufp->used;
4244 /* Mark the opcode just after a start_memory, so we can test for an
4245 empty subpattern when we get to the stop_memory. */
4246 unsigned char *just_past_start_mem = 0;
4248 /* We use this to map every character in the string. */
4249 RE_TRANSLATE_TYPE translate = bufp->translate;
4251 /* Failure point stack. Each place that can handle a failure further
4252 down the line pushes a failure point on this stack. It consists of
4253 restart, regend, and reg_info for all registers corresponding to
4254 the subexpressions we're currently inside, plus the number of such
4255 registers, and, finally, two char *'s. The first char * is where
4256 to resume scanning the pattern; the second one is where to resume
4257 scanning the strings. If the latter is zero, the failure point is
4258 a ``dummy''; if a failure happens and the failure point is a dummy,
4259 it gets discarded and the next next one is tried. */
4260 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
4261 fail_stack_type fail_stack;
4264 static unsigned failure_id;
4265 unsigned nfailure_points_pushed = 0, nfailure_points_popped = 0;
4269 /* This holds the pointer to the failure stack, when
4270 it is allocated relocatably. */
4271 fail_stack_elt_t *failure_stack_ptr;
4274 /* We fill all the registers internally, independent of what we
4275 return, for use in backreferences. The number here includes
4276 an element for register zero. */
4277 size_t num_regs = bufp->re_nsub + 1;
4279 /* The currently active registers. */
4280 active_reg_t lowest_active_reg = NO_LOWEST_ACTIVE_REG;
4281 active_reg_t highest_active_reg = NO_HIGHEST_ACTIVE_REG;
4283 /* Information on the contents of registers. These are pointers into
4284 the input strings; they record just what was matched (on this
4285 attempt) by a subexpression part of the pattern, that is, the
4286 regnum-th regstart pointer points to where in the pattern we began
4287 matching and the regnum-th regend points to right after where we
4288 stopped matching the regnum-th subexpression. (The zeroth register
4289 keeps track of what the whole pattern matches.) */
4290 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
4291 const char **regstart, **regend;
4294 /* If a group that's operated upon by a repetition operator fails to
4295 match anything, then the register for its start will need to be
4296 restored because it will have been set to wherever in the string we
4297 are when we last see its open-group operator. Similarly for a
4299 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
4300 const char **old_regstart, **old_regend;
4303 /* The is_active field of reg_info helps us keep track of which (possibly
4304 nested) subexpressions we are currently in. The matched_something
4305 field of reg_info[reg_num] helps us tell whether or not we have
4306 matched any of the pattern so far this time through the reg_num-th
4307 subexpression. These two fields get reset each time through any
4308 loop their register is in. */
4309 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
4310 register_info_type *reg_info;
4313 /* The following record the register info as found in the above
4314 variables when we find a match better than any we've seen before.
4315 This happens as we backtrack through the failure points, which in
4316 turn happens only if we have not yet matched the entire string. */
4317 unsigned best_regs_set = false;
4318 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
4319 const char **best_regstart, **best_regend;
4322 /* Logically, this is `best_regend[0]'. But we don't want to have to
4323 allocate space for that if we're not allocating space for anything
4324 else (see below). Also, we never need info about register 0 for
4325 any of the other register vectors, and it seems rather a kludge to
4326 treat `best_regend' differently than the rest. So we keep track of
4327 the end of the best match so far in a separate variable. We
4328 initialize this to NULL so that when we backtrack the first time
4329 and need to test it, it's not garbage. */
4330 const char *match_end = NULL;
4332 /* This helps SET_REGS_MATCHED avoid doing redundant work. */
4333 int set_regs_matched_done = 0;
4335 /* Used when we pop values we don't care about. */
4336 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
4337 const char **reg_dummy;
4338 register_info_type *reg_info_dummy;
4342 /* Counts the total number of registers pushed. */
4343 unsigned num_regs_pushed = 0;
4346 DEBUG_PRINT1 ("\n\nEntering re_match_2.\n");
4350 #ifdef MATCH_MAY_ALLOCATE
4351 /* Do not bother to initialize all the register variables if there are
4352 no groups in the pattern, as it takes a fair amount of time. If
4353 there are groups, we include space for register 0 (the whole
4354 pattern), even though we never use it, since it simplifies the
4355 array indexing. We should fix this. */
4358 regstart = REGEX_TALLOC (num_regs, const char *);
4359 regend = REGEX_TALLOC (num_regs, const char *);
4360 old_regstart = REGEX_TALLOC (num_regs, const char *);
4361 old_regend = REGEX_TALLOC (num_regs, const char *);
4362 best_regstart = REGEX_TALLOC (num_regs, const char *);
4363 best_regend = REGEX_TALLOC (num_regs, const char *);
4364 reg_info = REGEX_TALLOC (num_regs, register_info_type);
4365 reg_dummy = REGEX_TALLOC (num_regs, const char *);
4366 reg_info_dummy = REGEX_TALLOC (num_regs, register_info_type);
4368 if (!(regstart && regend && old_regstart && old_regend && reg_info
4369 && best_regstart && best_regend && reg_dummy && reg_info_dummy))
4377 /* We must initialize all our variables to NULL, so that
4378 `FREE_VARIABLES' doesn't try to free them. */
4379 regstart = regend = old_regstart = old_regend = best_regstart
4380 = best_regend = reg_dummy = NULL;
4381 reg_info = reg_info_dummy = (register_info_type *) NULL;
4383 #endif /* MATCH_MAY_ALLOCATE */
4385 /* The starting position is bogus. */
4386 if (pos < 0 || pos > size1 + size2)
4392 /* Initialize subexpression text positions to -1 to mark ones that no
4393 start_memory/stop_memory has been seen for. Also initialize the
4394 register information struct. */
4395 for (mcnt = 1; (unsigned) mcnt < num_regs; mcnt++)
4397 regstart[mcnt] = regend[mcnt]
4398 = old_regstart[mcnt] = old_regend[mcnt] = REG_UNSET_VALUE;
4400 REG_MATCH_NULL_STRING_P (reg_info[mcnt]) = MATCH_NULL_UNSET_VALUE;
4401 IS_ACTIVE (reg_info[mcnt]) = 0;
4402 MATCHED_SOMETHING (reg_info[mcnt]) = 0;
4403 EVER_MATCHED_SOMETHING (reg_info[mcnt]) = 0;
4406 /* We move `string1' into `string2' if the latter's empty -- but not if
4407 `string1' is null. */
4408 if (size2 == 0 && string1 != NULL)
4415 end1 = string1 + size1;
4416 end2 = string2 + size2;
4418 /* Compute where to stop matching, within the two strings. */
4421 end_match_1 = string1 + stop;
4422 end_match_2 = string2;
4427 end_match_2 = string2 + stop - size1;
4430 /* `p' scans through the pattern as `d' scans through the data.
4431 `dend' is the end of the input string that `d' points within. `d'
4432 is advanced into the following input string whenever necessary, but
4433 this happens before fetching; therefore, at the beginning of the
4434 loop, `d' can be pointing at the end of a string, but it cannot
4436 if (size1 > 0 && pos <= size1)
4443 d = string2 + pos - size1;
4447 DEBUG_PRINT1 ("The compiled pattern is:\n");
4448 DEBUG_PRINT_COMPILED_PATTERN (bufp, p, pend);
4449 DEBUG_PRINT1 ("The string to match is: `");
4450 DEBUG_PRINT_DOUBLE_STRING (d, string1, size1, string2, size2);
4451 DEBUG_PRINT1 ("'\n");
4453 /* This loops over pattern commands. It exits by returning from the
4454 function if the match is complete, or it drops through if the match
4455 fails at this starting point in the input data. */
4459 DEBUG_PRINT2 ("\n%p: ", p);
4461 DEBUG_PRINT2 ("\n0x%x: ", p);
4465 { /* End of pattern means we might have succeeded. */
4466 DEBUG_PRINT1 ("end of pattern ... ");
4468 /* If we haven't matched the entire string, and we want the
4469 longest match, try backtracking. */
4470 if (d != end_match_2)
4472 /* 1 if this match ends in the same string (string1 or string2)
4473 as the best previous match. */
4474 boolean same_str_p = (FIRST_STRING_P (match_end)
4475 == MATCHING_IN_FIRST_STRING);
4476 /* 1 if this match is the best seen so far. */
4477 boolean best_match_p;
4479 /* AIX compiler got confused when this was combined
4480 with the previous declaration. */
4482 best_match_p = d > match_end;
4484 best_match_p = !MATCHING_IN_FIRST_STRING;
4486 DEBUG_PRINT1 ("backtracking.\n");
4488 if (!FAIL_STACK_EMPTY ())
4489 { /* More failure points to try. */
4491 /* If exceeds best match so far, save it. */
4492 if (!best_regs_set || best_match_p)
4494 best_regs_set = true;
4497 DEBUG_PRINT1 ("\nSAVING match as best so far.\n");
4499 for (mcnt = 1; (unsigned) mcnt < num_regs; mcnt++)
4501 best_regstart[mcnt] = regstart[mcnt];
4502 best_regend[mcnt] = regend[mcnt];
4508 /* If no failure points, don't restore garbage. And if
4509 last match is real best match, don't restore second
4511 else if (best_regs_set && !best_match_p)
4514 /* Restore best match. It may happen that `dend ==
4515 end_match_1' while the restored d is in string2.
4516 For example, the pattern `x.*y.*z' against the
4517 strings `x-' and `y-z-', if the two strings are
4518 not consecutive in memory. */
4519 DEBUG_PRINT1 ("Restoring best registers.\n");
4522 dend = ((d >= string1 && d <= end1)
4523 ? end_match_1 : end_match_2);
4525 for (mcnt = 1; (unsigned) mcnt < num_regs; mcnt++)
4527 regstart[mcnt] = best_regstart[mcnt];
4528 regend[mcnt] = best_regend[mcnt];
4531 } /* d != end_match_2 */
4534 DEBUG_PRINT1 ("Accepting match.\n");
4536 /* If caller wants register contents data back, do it. */
4537 if (regs && !bufp->no_sub)
4539 /* Have the register data arrays been allocated? */
4540 if (bufp->regs_allocated == REGS_UNALLOCATED)
4541 { /* No. So allocate them with malloc. We need one
4542 extra element beyond `num_regs' for the `-1' marker
4544 regs->num_regs = MAX (RE_NREGS, num_regs + 1);
4545 regs->start = TALLOC (regs->num_regs, regoff_t);
4546 regs->end = TALLOC (regs->num_regs, regoff_t);
4547 if (regs->start == NULL || regs->end == NULL)
4552 bufp->regs_allocated = REGS_REALLOCATE;
4554 else if (bufp->regs_allocated == REGS_REALLOCATE)
4555 { /* Yes. If we need more elements than were already
4556 allocated, reallocate them. If we need fewer, just
4558 if (regs->num_regs < num_regs + 1)
4560 regs->num_regs = num_regs + 1;
4561 RETALLOC (regs->start, regs->num_regs, regoff_t);
4562 RETALLOC (regs->end, regs->num_regs, regoff_t);
4563 if (regs->start == NULL || regs->end == NULL)
4572 /* These braces fend off a "empty body in an else-statement"
4573 warning under GCC when assert expands to nothing. */
4574 assert (bufp->regs_allocated == REGS_FIXED);
4577 /* Convert the pointer data in `regstart' and `regend' to
4578 indices. Register zero has to be set differently,
4579 since we haven't kept track of any info for it. */
4580 if (regs->num_regs > 0)
4582 regs->start[0] = pos;
4583 regs->end[0] = (MATCHING_IN_FIRST_STRING
4584 ? ((regoff_t) (d - string1))
4585 : ((regoff_t) (d - string2 + size1)));
4588 /* Go through the first `min (num_regs, regs->num_regs)'
4589 registers, since that is all we initialized. */
4590 for (mcnt = 1; (unsigned) mcnt < MIN (num_regs, regs->num_regs);
4593 if (REG_UNSET (regstart[mcnt]) || REG_UNSET (regend[mcnt]))
4594 regs->start[mcnt] = regs->end[mcnt] = -1;
4598 = (regoff_t) POINTER_TO_OFFSET (regstart[mcnt]);
4600 = (regoff_t) POINTER_TO_OFFSET (regend[mcnt]);
4604 /* If the regs structure we return has more elements than
4605 were in the pattern, set the extra elements to -1. If
4606 we (re)allocated the registers, this is the case,
4607 because we always allocate enough to have at least one
4609 for (mcnt = num_regs; (unsigned) mcnt < regs->num_regs; mcnt++)
4610 regs->start[mcnt] = regs->end[mcnt] = -1;
4611 } /* regs && !bufp->no_sub */
4613 DEBUG_PRINT4 ("%u failure points pushed, %u popped (%u remain).\n",
4614 nfailure_points_pushed, nfailure_points_popped,
4615 nfailure_points_pushed - nfailure_points_popped);
4616 DEBUG_PRINT2 ("%u registers pushed.\n", num_regs_pushed);
4618 mcnt = d - pos - (MATCHING_IN_FIRST_STRING
4622 DEBUG_PRINT2 ("Returning %d from re_match_2.\n", mcnt);
4628 /* Otherwise match next pattern command. */
4629 switch (SWITCH_ENUM_CAST ((re_opcode_t) *p++))
4631 /* Ignore these. Used to ignore the n of succeed_n's which
4632 currently have n == 0. */
4634 DEBUG_PRINT1 ("EXECUTING no_op.\n");
4638 DEBUG_PRINT1 ("EXECUTING succeed.\n");
4641 /* Match the next n pattern characters exactly. The following
4642 byte in the pattern defines n, and the n bytes after that
4643 are the characters to match. */
4646 DEBUG_PRINT2 ("EXECUTING exactn %d.\n", mcnt);
4648 /* This is written out as an if-else so we don't waste time
4649 testing `translate' inside the loop. */
4655 if ((unsigned char) translate[(unsigned char) *d++]
4656 != (unsigned char) *p++)
4666 if (*d++ != (char) *p++) goto fail;
4670 SET_REGS_MATCHED ();
4674 /* Match any character except possibly a newline or a null. */
4676 DEBUG_PRINT1 ("EXECUTING anychar.\n");
4680 if ((!(bufp->syntax & RE_DOT_NEWLINE) && TRANSLATE (*d) == '\n')
4681 || (bufp->syntax & RE_DOT_NOT_NULL && TRANSLATE (*d) == '\000'))
4684 SET_REGS_MATCHED ();
4685 DEBUG_PRINT2 (" Matched `%d'.\n", *d);
4693 register unsigned char c;
4694 boolean not = (re_opcode_t) *(p - 1) == charset_not;
4696 DEBUG_PRINT2 ("EXECUTING charset%s.\n", not ? "_not" : "");
4699 c = TRANSLATE (*d); /* The character to match. */
4701 /* Cast to `unsigned' instead of `unsigned char' in case the
4702 bit list is a full 32 bytes long. */
4703 if (c < (unsigned) (*p * BYTEWIDTH)
4704 && p[1 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH)))
4709 if (!not) goto fail;
4711 SET_REGS_MATCHED ();
4717 /* The beginning of a group is represented by start_memory.
4718 The arguments are the register number in the next byte, and the
4719 number of groups inner to this one in the next. The text
4720 matched within the group is recorded (in the internal
4721 registers data structure) under the register number. */
4723 DEBUG_PRINT3 ("EXECUTING start_memory %d (%d):\n", *p, p[1]);
4725 /* Find out if this group can match the empty string. */
4726 p1 = p; /* To send to group_match_null_string_p. */
4728 if (REG_MATCH_NULL_STRING_P (reg_info[*p]) == MATCH_NULL_UNSET_VALUE)
4729 REG_MATCH_NULL_STRING_P (reg_info[*p])
4730 = group_match_null_string_p (&p1, pend, reg_info);
4732 /* Save the position in the string where we were the last time
4733 we were at this open-group operator in case the group is
4734 operated upon by a repetition operator, e.g., with `(a*)*b'
4735 against `ab'; then we want to ignore where we are now in
4736 the string in case this attempt to match fails. */
4737 old_regstart[*p] = REG_MATCH_NULL_STRING_P (reg_info[*p])
4738 ? REG_UNSET (regstart[*p]) ? d : regstart[*p]
4740 DEBUG_PRINT2 (" old_regstart: %d\n",
4741 POINTER_TO_OFFSET (old_regstart[*p]));
4744 DEBUG_PRINT2 (" regstart: %d\n", POINTER_TO_OFFSET (regstart[*p]));
4746 IS_ACTIVE (reg_info[*p]) = 1;
4747 MATCHED_SOMETHING (reg_info[*p]) = 0;
4749 /* Clear this whenever we change the register activity status. */
4750 set_regs_matched_done = 0;
4752 /* This is the new highest active register. */
4753 highest_active_reg = *p;
4755 /* If nothing was active before, this is the new lowest active
4757 if (lowest_active_reg == NO_LOWEST_ACTIVE_REG)
4758 lowest_active_reg = *p;
4760 /* Move past the register number and inner group count. */
4762 just_past_start_mem = p;
4767 /* The stop_memory opcode represents the end of a group. Its
4768 arguments are the same as start_memory's: the register
4769 number, and the number of inner groups. */
4771 DEBUG_PRINT3 ("EXECUTING stop_memory %d (%d):\n", *p, p[1]);
4773 /* We need to save the string position the last time we were at
4774 this close-group operator in case the group is operated
4775 upon by a repetition operator, e.g., with `((a*)*(b*)*)*'
4776 against `aba'; then we want to ignore where we are now in
4777 the string in case this attempt to match fails. */
4778 old_regend[*p] = REG_MATCH_NULL_STRING_P (reg_info[*p])
4779 ? REG_UNSET (regend[*p]) ? d : regend[*p]
4781 DEBUG_PRINT2 (" old_regend: %d\n",
4782 POINTER_TO_OFFSET (old_regend[*p]));
4785 DEBUG_PRINT2 (" regend: %d\n", POINTER_TO_OFFSET (regend[*p]));
4787 /* This register isn't active anymore. */
4788 IS_ACTIVE (reg_info[*p]) = 0;
4790 /* Clear this whenever we change the register activity status. */
4791 set_regs_matched_done = 0;
4793 /* If this was the only register active, nothing is active
4795 if (lowest_active_reg == highest_active_reg)
4797 lowest_active_reg = NO_LOWEST_ACTIVE_REG;
4798 highest_active_reg = NO_HIGHEST_ACTIVE_REG;
4801 { /* We must scan for the new highest active register, since
4802 it isn't necessarily one less than now: consider
4803 (a(b)c(d(e)f)g). When group 3 ends, after the f), the
4804 new highest active register is 1. */
4805 unsigned char r = *p - 1;
4806 while (r > 0 && !IS_ACTIVE (reg_info[r]))
4809 /* If we end up at register zero, that means that we saved
4810 the registers as the result of an `on_failure_jump', not
4811 a `start_memory', and we jumped to past the innermost
4812 `stop_memory'. For example, in ((.)*) we save
4813 registers 1 and 2 as a result of the *, but when we pop
4814 back to the second ), we are at the stop_memory 1.
4815 Thus, nothing is active. */
4818 lowest_active_reg = NO_LOWEST_ACTIVE_REG;
4819 highest_active_reg = NO_HIGHEST_ACTIVE_REG;
4822 highest_active_reg = r;
4825 /* If just failed to match something this time around with a
4826 group that's operated on by a repetition operator, try to
4827 force exit from the ``loop'', and restore the register
4828 information for this group that we had before trying this
4830 if ((!MATCHED_SOMETHING (reg_info[*p])
4831 || just_past_start_mem == p - 1)
4834 boolean is_a_jump_n = false;
4838 switch ((re_opcode_t) *p1++)
4842 case pop_failure_jump:
4843 case maybe_pop_jump:
4845 case dummy_failure_jump:
4846 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4856 /* If the next operation is a jump backwards in the pattern
4857 to an on_failure_jump right before the start_memory
4858 corresponding to this stop_memory, exit from the loop
4859 by forcing a failure after pushing on the stack the
4860 on_failure_jump's jump in the pattern, and d. */
4861 if (mcnt < 0 && (re_opcode_t) *p1 == on_failure_jump
4862 && (re_opcode_t) p1[3] == start_memory && p1[4] == *p)
4864 /* If this group ever matched anything, then restore
4865 what its registers were before trying this last
4866 failed match, e.g., with `(a*)*b' against `ab' for
4867 regstart[1], and, e.g., with `((a*)*(b*)*)*'
4868 against `aba' for regend[3].
4870 Also restore the registers for inner groups for,
4871 e.g., `((a*)(b*))*' against `aba' (register 3 would
4872 otherwise get trashed). */
4874 if (EVER_MATCHED_SOMETHING (reg_info[*p]))
4878 EVER_MATCHED_SOMETHING (reg_info[*p]) = 0;
4880 /* Restore this and inner groups' (if any) registers. */
4881 for (r = *p; r < (unsigned) *p + (unsigned) *(p + 1);
4884 regstart[r] = old_regstart[r];
4886 /* xx why this test? */
4887 if (old_regend[r] >= regstart[r])
4888 regend[r] = old_regend[r];
4892 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4893 PUSH_FAILURE_POINT (p1 + mcnt, d, -2);
4899 /* Move past the register number and the inner group count. */
4904 /* \<digit> has been turned into a `duplicate' command which is
4905 followed by the numeric value of <digit> as the register number. */
4908 register const char *d2, *dend2;
4909 int regno = *p++; /* Get which register to match against. */
4910 DEBUG_PRINT2 ("EXECUTING duplicate %d.\n", regno);
4912 /* Can't back reference a group which we've never matched. */
4913 if (REG_UNSET (regstart[regno]) || REG_UNSET (regend[regno]))
4916 /* Where in input to try to start matching. */
4917 d2 = regstart[regno];
4919 /* Where to stop matching; if both the place to start and
4920 the place to stop matching are in the same string, then
4921 set to the place to stop, otherwise, for now have to use
4922 the end of the first string. */
4924 dend2 = ((FIRST_STRING_P (regstart[regno])
4925 == FIRST_STRING_P (regend[regno]))
4926 ? regend[regno] : end_match_1);
4929 /* If necessary, advance to next segment in register
4933 if (dend2 == end_match_2) break;
4934 if (dend2 == regend[regno]) break;
4936 /* End of string1 => advance to string2. */
4938 dend2 = regend[regno];
4940 /* At end of register contents => success */
4941 if (d2 == dend2) break;
4943 /* If necessary, advance to next segment in data. */
4946 /* How many characters left in this segment to match. */
4949 /* Want how many consecutive characters we can match in
4950 one shot, so, if necessary, adjust the count. */
4951 if (mcnt > dend2 - d2)
4954 /* Compare that many; failure if mismatch, else move
4957 ? bcmp_translate (d, d2, mcnt, translate)
4958 : memcmp (d, d2, mcnt))
4960 d += mcnt, d2 += mcnt;
4962 /* Do this because we've match some characters. */
4963 SET_REGS_MATCHED ();
4969 /* begline matches the empty string at the beginning of the string
4970 (unless `not_bol' is set in `bufp'), and, if
4971 `newline_anchor' is set, after newlines. */
4973 DEBUG_PRINT1 ("EXECUTING begline.\n");
4975 if (AT_STRINGS_BEG (d))
4977 if (!bufp->not_bol) break;
4979 else if (d[-1] == '\n' && bufp->newline_anchor)
4983 /* In all other cases, we fail. */
4987 /* endline is the dual of begline. */
4989 DEBUG_PRINT1 ("EXECUTING endline.\n");
4991 if (AT_STRINGS_END (d))
4993 if (!bufp->not_eol) break;
4996 /* We have to ``prefetch'' the next character. */
4997 else if ((d == end1 ? *string2 : *d) == '\n'
4998 && bufp->newline_anchor)
5005 /* Match at the very beginning of the data. */
5007 DEBUG_PRINT1 ("EXECUTING begbuf.\n");
5008 if (AT_STRINGS_BEG (d))
5013 /* Match at the very end of the data. */
5015 DEBUG_PRINT1 ("EXECUTING endbuf.\n");
5016 if (AT_STRINGS_END (d))
5021 /* on_failure_keep_string_jump is used to optimize `.*\n'. It
5022 pushes NULL as the value for the string on the stack. Then
5023 `pop_failure_point' will keep the current value for the
5024 string, instead of restoring it. To see why, consider
5025 matching `foo\nbar' against `.*\n'. The .* matches the foo;
5026 then the . fails against the \n. But the next thing we want
5027 to do is match the \n against the \n; if we restored the
5028 string value, we would be back at the foo.
5030 Because this is used only in specific cases, we don't need to
5031 check all the things that `on_failure_jump' does, to make
5032 sure the right things get saved on the stack. Hence we don't
5033 share its code. The only reason to push anything on the
5034 stack at all is that otherwise we would have to change
5035 `anychar's code to do something besides goto fail in this
5036 case; that seems worse than this. */
5037 case on_failure_keep_string_jump:
5038 DEBUG_PRINT1 ("EXECUTING on_failure_keep_string_jump");
5040 EXTRACT_NUMBER_AND_INCR (mcnt, p);
5042 DEBUG_PRINT3 (" %d (to %p):\n", mcnt, p + mcnt);
5044 DEBUG_PRINT3 (" %d (to 0x%x):\n", mcnt, p + mcnt);
5047 PUSH_FAILURE_POINT (p + mcnt, NULL, -2);
5051 /* Uses of on_failure_jump:
5053 Each alternative starts with an on_failure_jump that points
5054 to the beginning of the next alternative. Each alternative
5055 except the last ends with a jump that in effect jumps past
5056 the rest of the alternatives. (They really jump to the
5057 ending jump of the following alternative, because tensioning
5058 these jumps is a hassle.)
5060 Repeats start with an on_failure_jump that points past both
5061 the repetition text and either the following jump or
5062 pop_failure_jump back to this on_failure_jump. */
5063 case on_failure_jump:
5065 DEBUG_PRINT1 ("EXECUTING on_failure_jump");
5067 EXTRACT_NUMBER_AND_INCR (mcnt, p);
5069 DEBUG_PRINT3 (" %d (to %p)", mcnt, p + mcnt);
5071 DEBUG_PRINT3 (" %d (to 0x%x)", mcnt, p + mcnt);
5074 /* If this on_failure_jump comes right before a group (i.e.,
5075 the original * applied to a group), save the information
5076 for that group and all inner ones, so that if we fail back
5077 to this point, the group's information will be correct.
5078 For example, in \(a*\)*\1, we need the preceding group,
5079 and in \(zz\(a*\)b*\)\2, we need the inner group. */
5081 /* We can't use `p' to check ahead because we push
5082 a failure point to `p + mcnt' after we do this. */
5085 /* We need to skip no_op's before we look for the
5086 start_memory in case this on_failure_jump is happening as
5087 the result of a completed succeed_n, as in \(a\)\{1,3\}b\1
5089 while (p1 < pend && (re_opcode_t) *p1 == no_op)
5092 if (p1 < pend && (re_opcode_t) *p1 == start_memory)
5094 /* We have a new highest active register now. This will
5095 get reset at the start_memory we are about to get to,
5096 but we will have saved all the registers relevant to
5097 this repetition op, as described above. */
5098 highest_active_reg = *(p1 + 1) + *(p1 + 2);
5099 if (lowest_active_reg == NO_LOWEST_ACTIVE_REG)
5100 lowest_active_reg = *(p1 + 1);
5103 DEBUG_PRINT1 (":\n");
5104 PUSH_FAILURE_POINT (p + mcnt, d, -2);
5108 /* A smart repeat ends with `maybe_pop_jump'.
5109 We change it to either `pop_failure_jump' or `jump'. */
5110 case maybe_pop_jump:
5111 EXTRACT_NUMBER_AND_INCR (mcnt, p);
5112 DEBUG_PRINT2 ("EXECUTING maybe_pop_jump %d.\n", mcnt);
5114 register unsigned char *p2 = p;
5116 /* Compare the beginning of the repeat with what in the
5117 pattern follows its end. If we can establish that there
5118 is nothing that they would both match, i.e., that we
5119 would have to backtrack because of (as in, e.g., `a*a')
5120 then we can change to pop_failure_jump, because we'll
5121 never have to backtrack.
5123 This is not true in the case of alternatives: in
5124 `(a|ab)*' we do need to backtrack to the `ab' alternative
5125 (e.g., if the string was `ab'). But instead of trying to
5126 detect that here, the alternative has put on a dummy
5127 failure point which is what we will end up popping. */
5129 /* Skip over open/close-group commands.
5130 If what follows this loop is a ...+ construct,
5131 look at what begins its body, since we will have to
5132 match at least one of that. */
5136 && ((re_opcode_t) *p2 == stop_memory
5137 || (re_opcode_t) *p2 == start_memory))
5139 else if (p2 + 6 < pend
5140 && (re_opcode_t) *p2 == dummy_failure_jump)
5147 /* p1[0] ... p1[2] are the `on_failure_jump' corresponding
5148 to the `maybe_finalize_jump' of this case. Examine what
5151 /* If we're at the end of the pattern, we can change. */
5154 /* Consider what happens when matching ":\(.*\)"
5155 against ":/". I don't really understand this code
5157 p[-3] = (unsigned char) pop_failure_jump;
5159 (" End of pattern: change to `pop_failure_jump'.\n");
5162 else if ((re_opcode_t) *p2 == exactn
5163 || (bufp->newline_anchor && (re_opcode_t) *p2 == endline))
5165 register unsigned char c
5166 = *p2 == (unsigned char) endline ? '\n' : p2[2];
5168 if ((re_opcode_t) p1[3] == exactn && p1[5] != c)
5170 p[-3] = (unsigned char) pop_failure_jump;
5171 DEBUG_PRINT3 (" %c != %c => pop_failure_jump.\n",
5175 else if ((re_opcode_t) p1[3] == charset
5176 || (re_opcode_t) p1[3] == charset_not)
5178 int not = (re_opcode_t) p1[3] == charset_not;
5180 if (c < (unsigned char) (p1[4] * BYTEWIDTH)
5181 && p1[5 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH)))
5184 /* `not' is equal to 1 if c would match, which means
5185 that we can't change to pop_failure_jump. */
5188 p[-3] = (unsigned char) pop_failure_jump;
5189 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
5193 else if ((re_opcode_t) *p2 == charset)
5195 /* We win if the first character of the loop is not part
5197 if ((re_opcode_t) p1[3] == exactn
5198 && ! ((int) p2[1] * BYTEWIDTH > (int) p1[5]
5199 && (p2[2 + p1[5] / BYTEWIDTH]
5200 & (1 << (p1[5] % BYTEWIDTH)))))
5202 p[-3] = (unsigned char) pop_failure_jump;
5203 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
5206 else if ((re_opcode_t) p1[3] == charset_not)
5209 /* We win if the charset_not inside the loop
5210 lists every character listed in the charset after. */
5211 for (idx = 0; idx < (int) p2[1]; idx++)
5212 if (! (p2[2 + idx] == 0
5213 || (idx < (int) p1[4]
5214 && ((p2[2 + idx] & ~ p1[5 + idx]) == 0))))
5219 p[-3] = (unsigned char) pop_failure_jump;
5220 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
5223 else if ((re_opcode_t) p1[3] == charset)
5226 /* We win if the charset inside the loop
5227 has no overlap with the one after the loop. */
5229 idx < (int) p2[1] && idx < (int) p1[4];
5231 if ((p2[2 + idx] & p1[5 + idx]) != 0)
5234 if (idx == p2[1] || idx == p1[4])
5236 p[-3] = (unsigned char) pop_failure_jump;
5237 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
5242 p -= 2; /* Point at relative address again. */
5243 if ((re_opcode_t) p[-1] != pop_failure_jump)
5245 p[-1] = (unsigned char) jump;
5246 DEBUG_PRINT1 (" Match => jump.\n");
5247 goto unconditional_jump;
5249 /* Note fall through. */
5252 /* The end of a simple repeat has a pop_failure_jump back to
5253 its matching on_failure_jump, where the latter will push a
5254 failure point. The pop_failure_jump takes off failure
5255 points put on by this pop_failure_jump's matching
5256 on_failure_jump; we got through the pattern to here from the
5257 matching on_failure_jump, so didn't fail. */
5258 case pop_failure_jump:
5260 /* We need to pass separate storage for the lowest and
5261 highest registers, even though we don't care about the
5262 actual values. Otherwise, we will restore only one
5263 register from the stack, since lowest will == highest in
5264 `pop_failure_point'. */
5265 active_reg_t dummy_low_reg, dummy_high_reg;
5266 unsigned char *pdummy;
5269 DEBUG_PRINT1 ("EXECUTING pop_failure_jump.\n");
5270 POP_FAILURE_POINT (sdummy, pdummy,
5271 dummy_low_reg, dummy_high_reg,
5272 reg_dummy, reg_dummy, reg_info_dummy);
5274 /* Note fall through. */
5278 DEBUG_PRINT2 ("\n%p: ", p);
5280 DEBUG_PRINT2 ("\n0x%x: ", p);
5282 /* Note fall through. */
5284 /* Unconditionally jump (without popping any failure points). */
5286 EXTRACT_NUMBER_AND_INCR (mcnt, p); /* Get the amount to jump. */
5287 DEBUG_PRINT2 ("EXECUTING jump %d ", mcnt);
5288 p += mcnt; /* Do the jump. */
5290 DEBUG_PRINT2 ("(to %p).\n", p);
5292 DEBUG_PRINT2 ("(to 0x%x).\n", p);
5297 /* We need this opcode so we can detect where alternatives end
5298 in `group_match_null_string_p' et al. */
5300 DEBUG_PRINT1 ("EXECUTING jump_past_alt.\n");
5301 goto unconditional_jump;
5304 /* Normally, the on_failure_jump pushes a failure point, which
5305 then gets popped at pop_failure_jump. We will end up at
5306 pop_failure_jump, also, and with a pattern of, say, `a+', we
5307 are skipping over the on_failure_jump, so we have to push
5308 something meaningless for pop_failure_jump to pop. */
5309 case dummy_failure_jump:
5310 DEBUG_PRINT1 ("EXECUTING dummy_failure_jump.\n");
5311 /* It doesn't matter what we push for the string here. What
5312 the code at `fail' tests is the value for the pattern. */
5313 PUSH_FAILURE_POINT (NULL, NULL, -2);
5314 goto unconditional_jump;
5317 /* At the end of an alternative, we need to push a dummy failure
5318 point in case we are followed by a `pop_failure_jump', because
5319 we don't want the failure point for the alternative to be
5320 popped. For example, matching `(a|ab)*' against `aab'
5321 requires that we match the `ab' alternative. */
5322 case push_dummy_failure:
5323 DEBUG_PRINT1 ("EXECUTING push_dummy_failure.\n");
5324 /* See comments just above at `dummy_failure_jump' about the
5326 PUSH_FAILURE_POINT (NULL, NULL, -2);
5329 /* Have to succeed matching what follows at least n times.
5330 After that, handle like `on_failure_jump'. */
5332 EXTRACT_NUMBER (mcnt, p + 2);
5333 DEBUG_PRINT2 ("EXECUTING succeed_n %d.\n", mcnt);
5336 /* Originally, this is how many times we HAVE to succeed. */
5341 STORE_NUMBER_AND_INCR (p, mcnt);
5343 DEBUG_PRINT3 (" Setting %p to %d.\n", p - 2, mcnt);
5345 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p - 2, mcnt);
5351 DEBUG_PRINT2 (" Setting two bytes from %p to no_op.\n", p+2);
5353 DEBUG_PRINT2 (" Setting two bytes from 0x%x to no_op.\n", p+2);
5355 p[2] = (unsigned char) no_op;
5356 p[3] = (unsigned char) no_op;
5362 EXTRACT_NUMBER (mcnt, p + 2);
5363 DEBUG_PRINT2 ("EXECUTING jump_n %d.\n", mcnt);
5365 /* Originally, this is how many times we CAN jump. */
5369 STORE_NUMBER (p + 2, mcnt);
5371 DEBUG_PRINT3 (" Setting %p to %d.\n", p + 2, mcnt);
5373 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p + 2, mcnt);
5375 goto unconditional_jump;
5377 /* If don't have to jump any more, skip over the rest of command. */
5384 DEBUG_PRINT1 ("EXECUTING set_number_at.\n");
5386 EXTRACT_NUMBER_AND_INCR (mcnt, p);
5388 EXTRACT_NUMBER_AND_INCR (mcnt, p);
5390 DEBUG_PRINT3 (" Setting %p to %d.\n", p1, mcnt);
5392 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p1, mcnt);
5394 STORE_NUMBER (p1, mcnt);
5399 /* The DEC Alpha C compiler 3.x generates incorrect code for the
5400 test WORDCHAR_P (d - 1) != WORDCHAR_P (d) in the expansion of
5401 AT_WORD_BOUNDARY, so this code is disabled. Expanding the
5402 macro and introducing temporary variables works around the bug. */
5405 DEBUG_PRINT1 ("EXECUTING wordbound.\n");
5406 if (AT_WORD_BOUNDARY (d))
5411 DEBUG_PRINT1 ("EXECUTING notwordbound.\n");
5412 if (AT_WORD_BOUNDARY (d))
5418 boolean prevchar, thischar;
5420 DEBUG_PRINT1 ("EXECUTING wordbound.\n");
5421 if (AT_STRINGS_BEG (d) || AT_STRINGS_END (d))
5424 prevchar = WORDCHAR_P (d - 1);
5425 thischar = WORDCHAR_P (d);
5426 if (prevchar != thischar)
5433 boolean prevchar, thischar;
5435 DEBUG_PRINT1 ("EXECUTING notwordbound.\n");
5436 if (AT_STRINGS_BEG (d) || AT_STRINGS_END (d))
5439 prevchar = WORDCHAR_P (d - 1);
5440 thischar = WORDCHAR_P (d);
5441 if (prevchar != thischar)
5448 DEBUG_PRINT1 ("EXECUTING wordbeg.\n");
5449 if (WORDCHAR_P (d) && (AT_STRINGS_BEG (d) || !WORDCHAR_P (d - 1)))
5454 DEBUG_PRINT1 ("EXECUTING wordend.\n");
5455 if (!AT_STRINGS_BEG (d) && WORDCHAR_P (d - 1)
5456 && (!WORDCHAR_P (d) || AT_STRINGS_END (d)))
5462 DEBUG_PRINT1 ("EXECUTING before_dot.\n");
5463 if (PTR_CHAR_POS ((unsigned char *) d) >= point)
5468 DEBUG_PRINT1 ("EXECUTING at_dot.\n");
5469 if (PTR_CHAR_POS ((unsigned char *) d) != point)
5474 DEBUG_PRINT1 ("EXECUTING after_dot.\n");
5475 if (PTR_CHAR_POS ((unsigned char *) d) <= point)
5480 DEBUG_PRINT2 ("EXECUTING syntaxspec %d.\n", mcnt);
5485 DEBUG_PRINT1 ("EXECUTING Emacs wordchar.\n");
5489 /* Can't use *d++ here; SYNTAX may be an unsafe macro. */
5491 if (SYNTAX (d[-1]) != (enum syntaxcode) mcnt)
5493 SET_REGS_MATCHED ();
5497 DEBUG_PRINT2 ("EXECUTING notsyntaxspec %d.\n", mcnt);
5499 goto matchnotsyntax;
5502 DEBUG_PRINT1 ("EXECUTING Emacs notwordchar.\n");
5506 /* Can't use *d++ here; SYNTAX may be an unsafe macro. */
5508 if (SYNTAX (d[-1]) == (enum syntaxcode) mcnt)
5510 SET_REGS_MATCHED ();
5513 #else /* not emacs */
5515 DEBUG_PRINT1 ("EXECUTING non-Emacs wordchar.\n");
5517 if (!WORDCHAR_P (d))
5519 SET_REGS_MATCHED ();
5524 DEBUG_PRINT1 ("EXECUTING non-Emacs notwordchar.\n");
5528 SET_REGS_MATCHED ();
5531 #endif /* not emacs */
5536 continue; /* Successfully executed one pattern command; keep going. */
5539 /* We goto here if a matching operation fails. */
5541 if (!FAIL_STACK_EMPTY ())
5542 { /* A restart point is known. Restore to that state. */
5543 DEBUG_PRINT1 ("\nFAIL:\n");
5544 POP_FAILURE_POINT (d, p,
5545 lowest_active_reg, highest_active_reg,
5546 regstart, regend, reg_info);
5548 /* If this failure point is a dummy, try the next one. */
5552 /* If we failed to the end of the pattern, don't examine *p. */
5556 boolean is_a_jump_n = false;
5558 /* If failed to a backwards jump that's part of a repetition
5559 loop, need to pop this failure point and use the next one. */
5560 switch ((re_opcode_t) *p)
5564 case maybe_pop_jump:
5565 case pop_failure_jump:
5568 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
5571 if ((is_a_jump_n && (re_opcode_t) *p1 == succeed_n)
5573 && (re_opcode_t) *p1 == on_failure_jump))
5581 if (d >= string1 && d <= end1)
5585 break; /* Matching at this starting point really fails. */
5589 goto restore_best_regs;
5593 return -1; /* Failure to match. */
5596 /* Subroutine definitions for re_match_2. */
5599 /* We are passed P pointing to a register number after a start_memory.
5601 Return true if the pattern up to the corresponding stop_memory can
5602 match the empty string, and false otherwise.
5604 If we find the matching stop_memory, sets P to point to one past its number.
5605 Otherwise, sets P to an undefined byte less than or equal to END.
5607 We don't handle duplicates properly (yet). */
5610 group_match_null_string_p (p, end, reg_info)
5611 unsigned char **p, *end;
5612 register_info_type *reg_info;
5615 /* Point to after the args to the start_memory. */
5616 unsigned char *p1 = *p + 2;
5620 /* Skip over opcodes that can match nothing, and return true or
5621 false, as appropriate, when we get to one that can't, or to the
5622 matching stop_memory. */
5624 switch ((re_opcode_t) *p1)
5626 /* Could be either a loop or a series of alternatives. */
5627 case on_failure_jump:
5629 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
5631 /* If the next operation is not a jump backwards in the
5636 /* Go through the on_failure_jumps of the alternatives,
5637 seeing if any of the alternatives cannot match nothing.
5638 The last alternative starts with only a jump,
5639 whereas the rest start with on_failure_jump and end
5640 with a jump, e.g., here is the pattern for `a|b|c':
5642 /on_failure_jump/0/6/exactn/1/a/jump_past_alt/0/6
5643 /on_failure_jump/0/6/exactn/1/b/jump_past_alt/0/3
5646 So, we have to first go through the first (n-1)
5647 alternatives and then deal with the last one separately. */
5650 /* Deal with the first (n-1) alternatives, which start
5651 with an on_failure_jump (see above) that jumps to right
5652 past a jump_past_alt. */
5654 while ((re_opcode_t) p1[mcnt-3] == jump_past_alt)
5656 /* `mcnt' holds how many bytes long the alternative
5657 is, including the ending `jump_past_alt' and
5660 if (!alt_match_null_string_p (p1, p1 + mcnt - 3,
5664 /* Move to right after this alternative, including the
5668 /* Break if it's the beginning of an n-th alternative
5669 that doesn't begin with an on_failure_jump. */
5670 if ((re_opcode_t) *p1 != on_failure_jump)
5673 /* Still have to check that it's not an n-th
5674 alternative that starts with an on_failure_jump. */
5676 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
5677 if ((re_opcode_t) p1[mcnt-3] != jump_past_alt)
5679 /* Get to the beginning of the n-th alternative. */
5685 /* Deal with the last alternative: go back and get number
5686 of the `jump_past_alt' just before it. `mcnt' contains
5687 the length of the alternative. */
5688 EXTRACT_NUMBER (mcnt, p1 - 2);
5690 if (!alt_match_null_string_p (p1, p1 + mcnt, reg_info))
5693 p1 += mcnt; /* Get past the n-th alternative. */
5699 assert (p1[1] == **p);
5705 if (!common_op_match_null_string_p (&p1, end, reg_info))
5708 } /* while p1 < end */
5711 } /* group_match_null_string_p */
5714 /* Similar to group_match_null_string_p, but doesn't deal with alternatives:
5715 It expects P to be the first byte of a single alternative and END one
5716 byte past the last. The alternative can contain groups. */
5719 alt_match_null_string_p (p, end, reg_info)
5720 unsigned char *p, *end;
5721 register_info_type *reg_info;
5724 unsigned char *p1 = p;
5728 /* Skip over opcodes that can match nothing, and break when we get
5729 to one that can't. */
5731 switch ((re_opcode_t) *p1)
5734 case on_failure_jump:
5736 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
5741 if (!common_op_match_null_string_p (&p1, end, reg_info))
5744 } /* while p1 < end */
5747 } /* alt_match_null_string_p */
5750 /* Deals with the ops common to group_match_null_string_p and
5751 alt_match_null_string_p.
5753 Sets P to one after the op and its arguments, if any. */
5756 common_op_match_null_string_p (p, end, reg_info)
5757 unsigned char **p, *end;
5758 register_info_type *reg_info;
5763 unsigned char *p1 = *p;
5765 switch ((re_opcode_t) *p1++)
5785 assert (reg_no > 0 && reg_no <= MAX_REGNUM);
5786 ret = group_match_null_string_p (&p1, end, reg_info);
5788 /* Have to set this here in case we're checking a group which
5789 contains a group and a back reference to it. */
5791 if (REG_MATCH_NULL_STRING_P (reg_info[reg_no]) == MATCH_NULL_UNSET_VALUE)
5792 REG_MATCH_NULL_STRING_P (reg_info[reg_no]) = ret;
5798 /* If this is an optimized succeed_n for zero times, make the jump. */
5800 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
5808 /* Get to the number of times to succeed. */
5810 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
5815 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
5823 if (!REG_MATCH_NULL_STRING_P (reg_info[*p1]))
5831 /* All other opcodes mean we cannot match the empty string. */
5837 } /* common_op_match_null_string_p */
5840 /* Return zero if TRANSLATE[S1] and TRANSLATE[S2] are identical for LEN
5841 bytes; nonzero otherwise. */
5844 bcmp_translate (s1, s2, len, translate)
5845 const char *s1, *s2;
5847 RE_TRANSLATE_TYPE translate;
5849 register const unsigned char *p1 = (const unsigned char *) s1;
5850 register const unsigned char *p2 = (const unsigned char *) s2;
5853 if (translate[*p1++] != translate[*p2++]) return 1;
5859 /* Entry points for GNU code. */
5861 /* re_compile_pattern is the GNU regular expression compiler: it
5862 compiles PATTERN (of length SIZE) and puts the result in BUFP.
5863 Returns 0 if the pattern was valid, otherwise an error string.
5865 Assumes the `allocated' (and perhaps `buffer') and `translate' fields
5866 are set in BUFP on entry.
5868 We call regex_compile to do the actual compilation. */
5871 re_compile_pattern (pattern, length, bufp)
5872 const char *pattern;
5874 struct re_pattern_buffer *bufp;
5878 /* GNU code is written to assume at least RE_NREGS registers will be set
5879 (and at least one extra will be -1). */
5880 bufp->regs_allocated = REGS_UNALLOCATED;
5882 /* And GNU code determines whether or not to get register information
5883 by passing null for the REGS argument to re_match, etc., not by
5887 /* Match anchors at newline. */
5888 bufp->newline_anchor = 1;
5890 ret = regex_compile (pattern, length, re_syntax_options, bufp);
5894 return gettext (re_error_msgid + re_error_msgid_idx[(int) ret]);
5897 weak_alias (__re_compile_pattern, re_compile_pattern)
5900 /* Entry points compatible with 4.2 BSD regex library. We don't define
5901 them unless specifically requested. */
5903 #if defined _REGEX_RE_COMP || defined _LIBC
5905 /* BSD has one and only one pattern buffer. */
5906 static struct re_pattern_buffer re_comp_buf;
5910 /* Make these definitions weak in libc, so POSIX programs can redefine
5911 these names if they don't use our functions, and still use
5912 regcomp/regexec below without link errors. */
5922 if (!re_comp_buf.buffer)
5923 return gettext ("No previous regular expression");
5927 if (!re_comp_buf.buffer)
5929 re_comp_buf.buffer = (unsigned char *) malloc (200);
5930 if (re_comp_buf.buffer == NULL)
5931 return (char *) gettext (re_error_msgid
5932 + re_error_msgid_idx[(int) REG_ESPACE]);
5933 re_comp_buf.allocated = 200;
5935 re_comp_buf.fastmap = (char *) malloc (1 << BYTEWIDTH);
5936 if (re_comp_buf.fastmap == NULL)
5937 return (char *) gettext (re_error_msgid
5938 + re_error_msgid_idx[(int) REG_ESPACE]);
5941 /* Since `re_exec' always passes NULL for the `regs' argument, we
5942 don't need to initialize the pattern buffer fields which affect it. */
5944 /* Match anchors at newlines. */
5945 re_comp_buf.newline_anchor = 1;
5947 ret = regex_compile (s, strlen (s), re_syntax_options, &re_comp_buf);
5952 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
5953 return (char *) gettext (re_error_msgid + re_error_msgid_idx[(int) ret]);
5964 const int len = strlen (s);
5966 0 <= re_search (&re_comp_buf, s, len, 0, len, (struct re_registers *) 0);
5969 #endif /* _REGEX_RE_COMP */
5971 /* POSIX.2 functions. Don't define these for Emacs. */
5975 /* regcomp takes a regular expression as a string and compiles it.
5977 PREG is a regex_t *. We do not expect any fields to be initialized,
5978 since POSIX says we shouldn't. Thus, we set
5980 `buffer' to the compiled pattern;
5981 `used' to the length of the compiled pattern;
5982 `syntax' to RE_SYNTAX_POSIX_EXTENDED if the
5983 REG_EXTENDED bit in CFLAGS is set; otherwise, to
5984 RE_SYNTAX_POSIX_BASIC;
5985 `newline_anchor' to REG_NEWLINE being set in CFLAGS;
5986 `fastmap' to an allocated space for the fastmap;
5987 `fastmap_accurate' to zero;
5988 `re_nsub' to the number of subexpressions in PATTERN.
5990 PATTERN is the address of the pattern string.
5992 CFLAGS is a series of bits which affect compilation.
5994 If REG_EXTENDED is set, we use POSIX extended syntax; otherwise, we
5995 use POSIX basic syntax.
5997 If REG_NEWLINE is set, then . and [^...] don't match newline.
5998 Also, regexec will try a match beginning after every newline.
6000 If REG_ICASE is set, then we considers upper- and lowercase
6001 versions of letters to be equivalent when matching.
6003 If REG_NOSUB is set, then when PREG is passed to regexec, that
6004 routine will report only success or failure, and nothing about the
6007 It returns 0 if it succeeds, nonzero if it doesn't. (See regex.h for
6008 the return codes and their meanings.) */
6011 regcomp (preg, pattern, cflags)
6013 const char *pattern;
6018 = (cflags & REG_EXTENDED) ?
6019 RE_SYNTAX_POSIX_EXTENDED : RE_SYNTAX_POSIX_BASIC;
6021 /* regex_compile will allocate the space for the compiled pattern. */
6023 preg->allocated = 0;
6026 /* Try to allocate space for the fastmap. */
6027 preg->fastmap = (char *) malloc (1 << BYTEWIDTH);
6029 if (cflags & REG_ICASE)
6034 = (RE_TRANSLATE_TYPE) malloc (CHAR_SET_SIZE
6035 * sizeof (*(RE_TRANSLATE_TYPE)0));
6036 if (preg->translate == NULL)
6037 return (int) REG_ESPACE;
6039 /* Map uppercase characters to corresponding lowercase ones. */
6040 for (i = 0; i < CHAR_SET_SIZE; i++)
6041 preg->translate[i] = ISUPPER (i) ? TOLOWER (i) : i;
6044 preg->translate = NULL;
6046 /* If REG_NEWLINE is set, newlines are treated differently. */
6047 if (cflags & REG_NEWLINE)
6048 { /* REG_NEWLINE implies neither . nor [^...] match newline. */
6049 syntax &= ~RE_DOT_NEWLINE;
6050 syntax |= RE_HAT_LISTS_NOT_NEWLINE;
6051 /* It also changes the matching behavior. */
6052 preg->newline_anchor = 1;
6055 preg->newline_anchor = 0;
6057 preg->no_sub = !!(cflags & REG_NOSUB);
6059 /* POSIX says a null character in the pattern terminates it, so we
6060 can use strlen here in compiling the pattern. */
6061 ret = regex_compile (pattern, strlen (pattern), syntax, preg);
6063 /* POSIX doesn't distinguish between an unmatched open-group and an
6064 unmatched close-group: both are REG_EPAREN. */
6065 if (ret == REG_ERPAREN) ret = REG_EPAREN;
6067 if (ret == REG_NOERROR && preg->fastmap)
6069 /* Compute the fastmap now, since regexec cannot modify the pattern
6071 if (re_compile_fastmap (preg) == -2)
6073 /* Some error occurred while computing the fastmap, just forget
6075 free (preg->fastmap);
6076 preg->fastmap = NULL;
6083 weak_alias (__regcomp, regcomp)
6087 /* regexec searches for a given pattern, specified by PREG, in the
6090 If NMATCH is zero or REG_NOSUB was set in the cflags argument to
6091 `regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at
6092 least NMATCH elements, and we set them to the offsets of the
6093 corresponding matched substrings.
6095 EFLAGS specifies `execution flags' which affect matching: if
6096 REG_NOTBOL is set, then ^ does not match at the beginning of the
6097 string; if REG_NOTEOL is set, then $ does not match at the end.
6099 We return 0 if we find a match and REG_NOMATCH if not. */
6102 regexec (preg, string, nmatch, pmatch, eflags)
6103 const regex_t *preg;
6106 regmatch_t pmatch[];
6110 struct re_registers regs;
6111 regex_t private_preg;
6112 int len = strlen (string);
6113 boolean want_reg_info = !preg->no_sub && nmatch > 0;
6115 private_preg = *preg;
6117 private_preg.not_bol = !!(eflags & REG_NOTBOL);
6118 private_preg.not_eol = !!(eflags & REG_NOTEOL);
6120 /* The user has told us exactly how many registers to return
6121 information about, via `nmatch'. We have to pass that on to the
6122 matching routines. */
6123 private_preg.regs_allocated = REGS_FIXED;
6127 regs.num_regs = nmatch;
6128 regs.start = TALLOC (nmatch * 2, regoff_t);
6129 if (regs.start == NULL)
6130 return (int) REG_NOMATCH;
6131 regs.end = regs.start + nmatch;
6134 /* Perform the searching operation. */
6135 ret = re_search (&private_preg, string, len,
6136 /* start: */ 0, /* range: */ len,
6137 want_reg_info ? ®s : (struct re_registers *) 0);
6139 /* Copy the register information to the POSIX structure. */
6146 for (r = 0; r < nmatch; r++)
6148 pmatch[r].rm_so = regs.start[r];
6149 pmatch[r].rm_eo = regs.end[r];
6153 /* If we needed the temporary register info, free the space now. */
6157 /* We want zero return to mean success, unlike `re_search'. */
6158 return ret >= 0 ? (int) REG_NOERROR : (int) REG_NOMATCH;
6161 weak_alias (__regexec, regexec)
6165 /* Returns a message corresponding to an error code, ERRCODE, returned
6166 from either regcomp or regexec. We don't use PREG here. */
6169 regerror (errcode, preg, errbuf, errbuf_size)
6171 const regex_t *preg;
6179 || errcode >= (int) (sizeof (re_error_msgid_idx)
6180 / sizeof (re_error_msgid_idx[0])))
6181 /* Only error codes returned by the rest of the code should be passed
6182 to this routine. If we are given anything else, or if other regex
6183 code generates an invalid error code, then the program has a bug.
6184 Dump core so we can fix it. */
6187 msg = gettext (re_error_msgid + re_error_msgid_idx[errcode]);
6189 msg_size = strlen (msg) + 1; /* Includes the null. */
6191 if (errbuf_size != 0)
6193 if (msg_size > errbuf_size)
6195 #if defined HAVE_MEMPCPY || defined _LIBC
6196 *((char *) __mempcpy (errbuf, msg, errbuf_size - 1)) = '\0';
6198 memcpy (errbuf, msg, errbuf_size - 1);
6199 errbuf[errbuf_size - 1] = 0;
6203 memcpy (errbuf, msg, msg_size);
6209 weak_alias (__regerror, regerror)
6213 /* Free dynamically allocated space used by PREG. */
6219 if (preg->buffer != NULL)
6220 free (preg->buffer);
6221 preg->buffer = NULL;
6223 preg->allocated = 0;
6226 if (preg->fastmap != NULL)
6227 free (preg->fastmap);
6228 preg->fastmap = NULL;
6229 preg->fastmap_accurate = 0;
6231 if (preg->translate != NULL)
6232 free (preg->translate);
6233 preg->translate = NULL;
6236 weak_alias (__regfree, regfree)
6239 #endif /* not emacs */