1 /* Extended regular expression matching and search library,
3 (Implements POSIX draft P1003.2/D11.2, except for some of the
4 internationalization features.)
6 Copyright (C) 1993, 1994, 1995, 1996, 1997 Free Software Foundation, Inc.
8 the C library, however. The master source lives in /gd/gnu/lib.
10 NOTE: The canonical source of this file is maintained with the
11 GNU C Library. Bugs can be reported to bug-glibc@prep.ai.mit.edu.
13 This program is free software; you can redistribute it and/or modify it
14 under the terms of the GNU General Public License as published by the
15 Free Software Foundation; either version 2, or (at your option) any
18 This program is distributed in the hope that it will be useful,
19 but WITHOUT ANY WARRANTY; without even the implied warranty of
20 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
21 GNU General Public License for more details.
23 You should have received a copy of the GNU General Public License
24 along with this program; if not, write to the Free Software Foundation,
25 Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. */
27 /* AIX requires this to be the first thing in the file. */
28 #if defined (_AIX) && !defined (REGEX_MALLOC)
39 #if defined(STDC_HEADERS) && !defined(emacs)
42 /* We need this for `regex.h', and perhaps for the Emacs include files. */
43 #include <sys/types.h>
46 /* For platform which support the ISO C amendement 1 functionality we
47 support user defined character classes. */
48 #if defined _LIBC || (defined HAVE_WCTYPE_H && defined HAVE_WCHAR_H)
53 /* This is for other GNU distributions with internationalized messages. */
54 #if HAVE_LIBINTL_H || defined (_LIBC)
57 # define gettext(msgid) (msgid)
61 /* This define is so xgettext can find the internationalizable
63 #define gettext_noop(String) String
66 /* The `emacs' switch turns on certain matching commands
67 that make sense only in Emacs. */
76 /* If we are not linking with Emacs proper,
77 we can't use the relocating allocator
78 even if config.h says that we can. */
81 #if defined (STDC_HEADERS) || defined (_LIBC)
88 /* When used in Emacs's lib-src, we need to get bzero and bcopy somehow.
89 If nothing else has been done, use the method below. */
90 #ifdef INHIBIT_STRING_HEADER
91 #if !(defined (HAVE_BZERO) && defined (HAVE_BCOPY))
92 #if !defined (bzero) && !defined (bcopy)
93 #undef INHIBIT_STRING_HEADER
98 /* This is the normal way of making sure we have a bcopy and a bzero.
99 This is used in most programs--a few other programs avoid this
100 by defining INHIBIT_STRING_HEADER. */
101 #ifndef INHIBIT_STRING_HEADER
102 #if defined (HAVE_STRING_H) || defined (STDC_HEADERS) || defined (_LIBC)
105 #define bcmp(s1, s2, n) memcmp ((s1), (s2), (n))
108 #define bcopy(s, d, n) memcpy ((d), (s), (n))
111 #define bzero(s, n) memset ((s), 0, (n))
118 /* Define the syntax stuff for \<, \>, etc. */
120 /* This must be nonzero for the wordchar and notwordchar pattern
121 commands in re_match_2. */
126 #ifdef SWITCH_ENUM_BUG
127 #define SWITCH_ENUM_CAST(x) ((int)(x))
129 #define SWITCH_ENUM_CAST(x) (x)
134 extern char *re_syntax_table;
136 #else /* not SYNTAX_TABLE */
138 /* How many characters in the character set. */
139 #define CHAR_SET_SIZE 256
141 static char re_syntax_table[CHAR_SET_SIZE];
152 bzero (re_syntax_table, sizeof re_syntax_table);
154 for (c = 'a'; c <= 'z'; c++)
155 re_syntax_table[c] = Sword;
157 for (c = 'A'; c <= 'Z'; c++)
158 re_syntax_table[c] = Sword;
160 for (c = '0'; c <= '9'; c++)
161 re_syntax_table[c] = Sword;
163 re_syntax_table['_'] = Sword;
168 #endif /* not SYNTAX_TABLE */
170 #define SYNTAX(c) re_syntax_table[c]
172 #endif /* not emacs */
174 /* Get the interface, including the syntax bits. */
177 /* isalpha etc. are used for the character classes. */
180 /* Jim Meyering writes:
182 "... Some ctype macros are valid only for character codes that
183 isascii says are ASCII (SGI's IRIX-4.0.5 is one such system --when
184 using /bin/cc or gcc but without giving an ansi option). So, all
185 ctype uses should be through macros like ISPRINT... If
186 STDC_HEADERS is defined, then autoconf has verified that the ctype
187 macros don't need to be guarded with references to isascii. ...
188 Defining isascii to 1 should let any compiler worth its salt
189 eliminate the && through constant folding." */
191 #if defined (STDC_HEADERS) || (!defined (isascii) && !defined (HAVE_ISASCII))
194 #define ISASCII(c) isascii(c)
198 #define ISBLANK(c) (ISASCII (c) && isblank (c))
200 #define ISBLANK(c) ((c) == ' ' || (c) == '\t')
203 #define ISGRAPH(c) (ISASCII (c) && isgraph (c))
205 #define ISGRAPH(c) (ISASCII (c) && isprint (c) && !isspace (c))
208 #define ISPRINT(c) (ISASCII (c) && isprint (c))
209 #define ISDIGIT(c) (ISASCII (c) && isdigit (c))
210 #define ISALNUM(c) (ISASCII (c) && isalnum (c))
211 #define ISALPHA(c) (ISASCII (c) && isalpha (c))
212 #define ISCNTRL(c) (ISASCII (c) && iscntrl (c))
213 #define ISLOWER(c) (ISASCII (c) && islower (c))
214 #define ISPUNCT(c) (ISASCII (c) && ispunct (c))
215 #define ISSPACE(c) (ISASCII (c) && isspace (c))
216 #define ISUPPER(c) (ISASCII (c) && isupper (c))
217 #define ISXDIGIT(c) (ISASCII (c) && isxdigit (c))
220 #define NULL (void *)0
223 /* We remove any previous definition of `SIGN_EXTEND_CHAR',
224 since ours (we hope) works properly with all combinations of
225 machines, compilers, `char' and `unsigned char' argument types.
226 (Per Bothner suggested the basic approach.) */
227 #undef SIGN_EXTEND_CHAR
229 #define SIGN_EXTEND_CHAR(c) ((signed char) (c))
230 #else /* not __STDC__ */
231 /* As in Harbison and Steele. */
232 #define SIGN_EXTEND_CHAR(c) ((((unsigned char) (c)) ^ 128) - 128)
235 /* Should we use malloc or alloca? If REGEX_MALLOC is not defined, we
236 use `alloca' instead of `malloc'. This is because using malloc in
237 re_search* or re_match* could cause memory leaks when C-g is used in
238 Emacs; also, malloc is slower and causes storage fragmentation. On
239 the other hand, malloc is more portable, and easier to debug.
241 Because we sometimes use alloca, some routines have to be macros,
242 not functions -- `alloca'-allocated space disappears at the end of the
243 function it is called in. */
247 #define REGEX_ALLOCATE malloc
248 #define REGEX_REALLOCATE(source, osize, nsize) realloc (source, nsize)
249 #define REGEX_FREE free
251 #else /* not REGEX_MALLOC */
253 /* Emacs already defines alloca, sometimes. */
256 /* Make alloca work the best possible way. */
258 #define alloca __builtin_alloca
259 #else /* not __GNUC__ */
262 #else /* not __GNUC__ or HAVE_ALLOCA_H */
263 #if 0 /* It is a bad idea to declare alloca. We always cast the result. */
264 #ifndef _AIX /* Already did AIX, up at the top. */
266 #endif /* not _AIX */
268 #endif /* not HAVE_ALLOCA_H */
269 #endif /* not __GNUC__ */
271 #endif /* not alloca */
273 #define REGEX_ALLOCATE alloca
275 /* Assumes a `char *destination' variable. */
276 #define REGEX_REALLOCATE(source, osize, nsize) \
277 (destination = (char *) alloca (nsize), \
278 bcopy (source, destination, osize), \
281 /* No need to do anything to free, after alloca. */
282 #define REGEX_FREE(arg) ((void)0) /* Do nothing! But inhibit gcc warning. */
284 #endif /* not REGEX_MALLOC */
286 /* Define how to allocate the failure stack. */
288 #if defined (REL_ALLOC) && defined (REGEX_MALLOC)
290 #define REGEX_ALLOCATE_STACK(size) \
291 r_alloc (&failure_stack_ptr, (size))
292 #define REGEX_REALLOCATE_STACK(source, osize, nsize) \
293 r_re_alloc (&failure_stack_ptr, (nsize))
294 #define REGEX_FREE_STACK(ptr) \
295 r_alloc_free (&failure_stack_ptr)
297 #else /* not using relocating allocator */
301 #define REGEX_ALLOCATE_STACK malloc
302 #define REGEX_REALLOCATE_STACK(source, osize, nsize) realloc (source, nsize)
303 #define REGEX_FREE_STACK free
305 #else /* not REGEX_MALLOC */
307 #define REGEX_ALLOCATE_STACK alloca
309 #define REGEX_REALLOCATE_STACK(source, osize, nsize) \
310 REGEX_REALLOCATE (source, osize, nsize)
311 /* No need to explicitly free anything. */
312 #define REGEX_FREE_STACK(arg)
314 #endif /* not REGEX_MALLOC */
315 #endif /* not using relocating allocator */
318 /* True if `size1' is non-NULL and PTR is pointing anywhere inside
319 `string1' or just past its end. This works if PTR is NULL, which is
321 #define FIRST_STRING_P(ptr) \
322 (size1 && string1 <= (ptr) && (ptr) <= string1 + size1)
324 /* (Re)Allocate N items of type T using malloc, or fail. */
325 #define TALLOC(n, t) ((t *) malloc ((n) * sizeof (t)))
326 #define RETALLOC(addr, n, t) ((addr) = (t *) realloc (addr, (n) * sizeof (t)))
327 #define RETALLOC_IF(addr, n, t) \
328 if (addr) RETALLOC((addr), (n), t); else (addr) = TALLOC ((n), t)
329 #define REGEX_TALLOC(n, t) ((t *) REGEX_ALLOCATE ((n) * sizeof (t)))
331 #define BYTEWIDTH 8 /* In bits. */
333 #define STREQ(s1, s2) ((strcmp (s1, s2) == 0))
337 #define MAX(a, b) ((a) > (b) ? (a) : (b))
338 #define MIN(a, b) ((a) < (b) ? (a) : (b))
340 typedef char boolean;
344 static int re_match_2_internal ();
346 /* These are the command codes that appear in compiled regular
347 expressions. Some opcodes are followed by argument bytes. A
348 command code can specify any interpretation whatsoever for its
349 arguments. Zero bytes may appear in the compiled regular expression. */
355 /* Succeed right away--no more backtracking. */
358 /* Followed by one byte giving n, then by n literal bytes. */
361 /* Matches any (more or less) character. */
364 /* Matches any one char belonging to specified set. First
365 following byte is number of bitmap bytes. Then come bytes
366 for a bitmap saying which chars are in. Bits in each byte
367 are ordered low-bit-first. A character is in the set if its
368 bit is 1. A character too large to have a bit in the map is
369 automatically not in the set. */
372 /* Same parameters as charset, but match any character that is
373 not one of those specified. */
376 /* Start remembering the text that is matched, for storing in a
377 register. Followed by one byte with the register number, in
378 the range 0 to one less than the pattern buffer's re_nsub
379 field. Then followed by one byte with the number of groups
380 inner to this one. (This last has to be part of the
381 start_memory only because we need it in the on_failure_jump
385 /* Stop remembering the text that is matched and store it in a
386 memory register. Followed by one byte with the register
387 number, in the range 0 to one less than `re_nsub' in the
388 pattern buffer, and one byte with the number of inner groups,
389 just like `start_memory'. (We need the number of inner
390 groups here because we don't have any easy way of finding the
391 corresponding start_memory when we're at a stop_memory.) */
394 /* Match a duplicate of something remembered. Followed by one
395 byte containing the register number. */
398 /* Fail unless at beginning of line. */
401 /* Fail unless at end of line. */
404 /* Succeeds if at beginning of buffer (if emacs) or at beginning
405 of string to be matched (if not). */
408 /* Analogously, for end of buffer/string. */
411 /* Followed by two byte relative address to which to jump. */
414 /* Same as jump, but marks the end of an alternative. */
417 /* Followed by two-byte relative address of place to resume at
418 in case of failure. */
421 /* Like on_failure_jump, but pushes a placeholder instead of the
422 current string position when executed. */
423 on_failure_keep_string_jump,
425 /* Throw away latest failure point and then jump to following
426 two-byte relative address. */
429 /* Change to pop_failure_jump if know won't have to backtrack to
430 match; otherwise change to jump. This is used to jump
431 back to the beginning of a repeat. If what follows this jump
432 clearly won't match what the repeat does, such that we can be
433 sure that there is no use backtracking out of repetitions
434 already matched, then we change it to a pop_failure_jump.
435 Followed by two-byte address. */
438 /* Jump to following two-byte address, and push a dummy failure
439 point. This failure point will be thrown away if an attempt
440 is made to use it for a failure. A `+' construct makes this
441 before the first repeat. Also used as an intermediary kind
442 of jump when compiling an alternative. */
445 /* Push a dummy failure point and continue. Used at the end of
449 /* Followed by two-byte relative address and two-byte number n.
450 After matching N times, jump to the address upon failure. */
453 /* Followed by two-byte relative address, and two-byte number n.
454 Jump to the address N times, then fail. */
457 /* Set the following two-byte relative address to the
458 subsequent two-byte number. The address *includes* the two
462 wordchar, /* Matches any word-constituent character. */
463 notwordchar, /* Matches any char that is not a word-constituent. */
465 wordbeg, /* Succeeds if at word beginning. */
466 wordend, /* Succeeds if at word end. */
468 wordbound, /* Succeeds if at a word boundary. */
469 notwordbound /* Succeeds if not at a word boundary. */
472 ,before_dot, /* Succeeds if before point. */
473 at_dot, /* Succeeds if at point. */
474 after_dot, /* Succeeds if after point. */
476 /* Matches any character whose syntax is specified. Followed by
477 a byte which contains a syntax code, e.g., Sword. */
480 /* Matches any character whose syntax is not that specified. */
485 /* Common operations on the compiled pattern. */
487 /* Store NUMBER in two contiguous bytes starting at DESTINATION. */
489 #define STORE_NUMBER(destination, number) \
491 (destination)[0] = (number) & 0377; \
492 (destination)[1] = (number) >> 8; \
495 /* Same as STORE_NUMBER, except increment DESTINATION to
496 the byte after where the number is stored. Therefore, DESTINATION
497 must be an lvalue. */
499 #define STORE_NUMBER_AND_INCR(destination, number) \
501 STORE_NUMBER (destination, number); \
502 (destination) += 2; \
505 /* Put into DESTINATION a number stored in two contiguous bytes starting
508 #define EXTRACT_NUMBER(destination, source) \
510 (destination) = *(source) & 0377; \
511 (destination) += SIGN_EXTEND_CHAR (*((source) + 1)) << 8; \
515 static void extract_number _RE_ARGS ((int *dest, unsigned char *source));
517 extract_number (dest, source)
519 unsigned char *source;
521 int temp = SIGN_EXTEND_CHAR (*(source + 1));
522 *dest = *source & 0377;
526 #ifndef EXTRACT_MACROS /* To debug the macros. */
527 #undef EXTRACT_NUMBER
528 #define EXTRACT_NUMBER(dest, src) extract_number (&dest, src)
529 #endif /* not EXTRACT_MACROS */
533 /* Same as EXTRACT_NUMBER, except increment SOURCE to after the number.
534 SOURCE must be an lvalue. */
536 #define EXTRACT_NUMBER_AND_INCR(destination, source) \
538 EXTRACT_NUMBER (destination, source); \
543 static void extract_number_and_incr _RE_ARGS ((int *destination,
544 unsigned char **source));
546 extract_number_and_incr (destination, source)
548 unsigned char **source;
550 extract_number (destination, *source);
554 #ifndef EXTRACT_MACROS
555 #undef EXTRACT_NUMBER_AND_INCR
556 #define EXTRACT_NUMBER_AND_INCR(dest, src) \
557 extract_number_and_incr (&dest, &src)
558 #endif /* not EXTRACT_MACROS */
562 /* If DEBUG is defined, Regex prints many voluminous messages about what
563 it is doing (if the variable `debug' is nonzero). If linked with the
564 main program in `iregex.c', you can enter patterns and strings
565 interactively. And if linked with the main program in `main.c' and
566 the other test files, you can run the already-written tests. */
570 /* We use standard I/O for debugging. */
573 /* It is useful to test things that ``must'' be true when debugging. */
576 static int debug = 0;
578 #define DEBUG_STATEMENT(e) e
579 #define DEBUG_PRINT1(x) if (debug) printf (x)
580 #define DEBUG_PRINT2(x1, x2) if (debug) printf (x1, x2)
581 #define DEBUG_PRINT3(x1, x2, x3) if (debug) printf (x1, x2, x3)
582 #define DEBUG_PRINT4(x1, x2, x3, x4) if (debug) printf (x1, x2, x3, x4)
583 #define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) \
584 if (debug) print_partial_compiled_pattern (s, e)
585 #define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) \
586 if (debug) print_double_string (w, s1, sz1, s2, sz2)
589 /* Print the fastmap in human-readable form. */
592 print_fastmap (fastmap)
595 unsigned was_a_range = 0;
598 while (i < (1 << BYTEWIDTH))
604 while (i < (1 << BYTEWIDTH) && fastmap[i])
620 /* Print a compiled pattern string in human-readable form, starting at
621 the START pointer into it and ending just before the pointer END. */
624 print_partial_compiled_pattern (start, end)
625 unsigned char *start;
630 unsigned char *p = start;
631 unsigned char *pend = end;
639 /* Loop over pattern commands. */
642 printf ("%d:\t", p - start);
644 switch ((re_opcode_t) *p++)
652 printf ("/exactn/%d", mcnt);
663 printf ("/start_memory/%d/%d", mcnt, *p++);
668 printf ("/stop_memory/%d/%d", mcnt, *p++);
672 printf ("/duplicate/%d", *p++);
682 register int c, last = -100;
683 register int in_range = 0;
685 printf ("/charset [%s",
686 (re_opcode_t) *(p - 1) == charset_not ? "^" : "");
688 assert (p + *p < pend);
690 for (c = 0; c < 256; c++)
692 && (p[1 + (c/8)] & (1 << (c % 8))))
694 /* Are we starting a range? */
695 if (last + 1 == c && ! in_range)
700 /* Have we broken a range? */
701 else if (last + 1 != c && in_range)
730 case on_failure_jump:
731 extract_number_and_incr (&mcnt, &p);
732 printf ("/on_failure_jump to %d", p + mcnt - start);
735 case on_failure_keep_string_jump:
736 extract_number_and_incr (&mcnt, &p);
737 printf ("/on_failure_keep_string_jump to %d", p + mcnt - start);
740 case dummy_failure_jump:
741 extract_number_and_incr (&mcnt, &p);
742 printf ("/dummy_failure_jump to %d", p + mcnt - start);
745 case push_dummy_failure:
746 printf ("/push_dummy_failure");
750 extract_number_and_incr (&mcnt, &p);
751 printf ("/maybe_pop_jump to %d", p + mcnt - start);
754 case pop_failure_jump:
755 extract_number_and_incr (&mcnt, &p);
756 printf ("/pop_failure_jump to %d", p + mcnt - start);
760 extract_number_and_incr (&mcnt, &p);
761 printf ("/jump_past_alt to %d", p + mcnt - start);
765 extract_number_and_incr (&mcnt, &p);
766 printf ("/jump to %d", p + mcnt - start);
770 extract_number_and_incr (&mcnt, &p);
772 extract_number_and_incr (&mcnt2, &p);
773 printf ("/succeed_n to %d, %d times", p1 - start, mcnt2);
777 extract_number_and_incr (&mcnt, &p);
779 extract_number_and_incr (&mcnt2, &p);
780 printf ("/jump_n to %d, %d times", p1 - start, mcnt2);
784 extract_number_and_incr (&mcnt, &p);
786 extract_number_and_incr (&mcnt2, &p);
787 printf ("/set_number_at location %d to %d", p1 - start, mcnt2);
791 printf ("/wordbound");
795 printf ("/notwordbound");
807 printf ("/before_dot");
815 printf ("/after_dot");
819 printf ("/syntaxspec");
821 printf ("/%d", mcnt);
825 printf ("/notsyntaxspec");
827 printf ("/%d", mcnt);
832 printf ("/wordchar");
836 printf ("/notwordchar");
848 printf ("?%d", *(p-1));
854 printf ("%d:\tend of pattern.\n", p - start);
859 print_compiled_pattern (bufp)
860 struct re_pattern_buffer *bufp;
862 unsigned char *buffer = bufp->buffer;
864 print_partial_compiled_pattern (buffer, buffer + bufp->used);
865 printf ("%ld bytes used/%ld bytes allocated.\n",
866 bufp->used, bufp->allocated);
868 if (bufp->fastmap_accurate && bufp->fastmap)
870 printf ("fastmap: ");
871 print_fastmap (bufp->fastmap);
874 printf ("re_nsub: %d\t", bufp->re_nsub);
875 printf ("regs_alloc: %d\t", bufp->regs_allocated);
876 printf ("can_be_null: %d\t", bufp->can_be_null);
877 printf ("newline_anchor: %d\n", bufp->newline_anchor);
878 printf ("no_sub: %d\t", bufp->no_sub);
879 printf ("not_bol: %d\t", bufp->not_bol);
880 printf ("not_eol: %d\t", bufp->not_eol);
881 printf ("syntax: %lx\n", bufp->syntax);
882 /* Perhaps we should print the translate table? */
887 print_double_string (where, string1, size1, string2, size2)
900 if (FIRST_STRING_P (where))
902 for (this_char = where - string1; this_char < size1; this_char++)
903 putchar (string1[this_char]);
908 for (this_char = where - string2; this_char < size2; this_char++)
909 putchar (string2[this_char]);
920 #else /* not DEBUG */
925 #define DEBUG_STATEMENT(e)
926 #define DEBUG_PRINT1(x)
927 #define DEBUG_PRINT2(x1, x2)
928 #define DEBUG_PRINT3(x1, x2, x3)
929 #define DEBUG_PRINT4(x1, x2, x3, x4)
930 #define DEBUG_PRINT_COMPILED_PATTERN(p, s, e)
931 #define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2)
933 #endif /* not DEBUG */
935 /* Set by `re_set_syntax' to the current regexp syntax to recognize. Can
936 also be assigned to arbitrarily: each pattern buffer stores its own
937 syntax, so it can be changed between regex compilations. */
938 /* This has no initializer because initialized variables in Emacs
939 become read-only after dumping. */
940 reg_syntax_t re_syntax_options;
943 /* Specify the precise syntax of regexps for compilation. This provides
944 for compatibility for various utilities which historically have
945 different, incompatible syntaxes.
947 The argument SYNTAX is a bit mask comprised of the various bits
948 defined in regex.h. We return the old syntax. */
951 re_set_syntax (syntax)
954 reg_syntax_t ret = re_syntax_options;
956 re_syntax_options = syntax;
958 if (syntax & RE_DEBUG)
960 else if (debug) /* was on but now is not */
966 /* This table gives an error message for each of the error codes listed
967 in regex.h. Obviously the order here has to be same as there.
968 POSIX doesn't require that we do anything for REG_NOERROR,
969 but why not be nice? */
971 static const char *re_error_msgid[] =
973 gettext_noop ("Success"), /* REG_NOERROR */
974 gettext_noop ("No match"), /* REG_NOMATCH */
975 gettext_noop ("Invalid regular expression"), /* REG_BADPAT */
976 gettext_noop ("Invalid collation character"), /* REG_ECOLLATE */
977 gettext_noop ("Invalid character class name"), /* REG_ECTYPE */
978 gettext_noop ("Trailing backslash"), /* REG_EESCAPE */
979 gettext_noop ("Invalid back reference"), /* REG_ESUBREG */
980 gettext_noop ("Unmatched [ or [^"), /* REG_EBRACK */
981 gettext_noop ("Unmatched ( or \\("), /* REG_EPAREN */
982 gettext_noop ("Unmatched \\{"), /* REG_EBRACE */
983 gettext_noop ("Invalid content of \\{\\}"), /* REG_BADBR */
984 gettext_noop ("Invalid range end"), /* REG_ERANGE */
985 gettext_noop ("Memory exhausted"), /* REG_ESPACE */
986 gettext_noop ("Invalid preceding regular expression"), /* REG_BADRPT */
987 gettext_noop ("Premature end of regular expression"), /* REG_EEND */
988 gettext_noop ("Regular expression too big"), /* REG_ESIZE */
989 gettext_noop ("Unmatched ) or \\)"), /* REG_ERPAREN */
992 /* Avoiding alloca during matching, to placate r_alloc. */
994 /* Define MATCH_MAY_ALLOCATE unless we need to make sure that the
995 searching and matching functions should not call alloca. On some
996 systems, alloca is implemented in terms of malloc, and if we're
997 using the relocating allocator routines, then malloc could cause a
998 relocation, which might (if the strings being searched are in the
999 ralloc heap) shift the data out from underneath the regexp
1002 Here's another reason to avoid allocation: Emacs
1003 processes input from X in a signal handler; processing X input may
1004 call malloc; if input arrives while a matching routine is calling
1005 malloc, then we're scrod. But Emacs can't just block input while
1006 calling matching routines; then we don't notice interrupts when
1007 they come in. So, Emacs blocks input around all regexp calls
1008 except the matching calls, which it leaves unprotected, in the
1009 faith that they will not malloc. */
1011 /* Normally, this is fine. */
1012 #define MATCH_MAY_ALLOCATE
1014 /* When using GNU C, we are not REALLY using the C alloca, no matter
1015 what config.h may say. So don't take precautions for it. */
1020 /* The match routines may not allocate if (1) they would do it with malloc
1021 and (2) it's not safe for them to use malloc.
1022 Note that if REL_ALLOC is defined, matching would not use malloc for the
1023 failure stack, but we would still use it for the register vectors;
1024 so REL_ALLOC should not affect this. */
1025 #if (defined (C_ALLOCA) || defined (REGEX_MALLOC)) && defined (emacs)
1026 #undef MATCH_MAY_ALLOCATE
1030 /* Failure stack declarations and macros; both re_compile_fastmap and
1031 re_match_2 use a failure stack. These have to be macros because of
1032 REGEX_ALLOCATE_STACK. */
1035 /* Number of failure points for which to initially allocate space
1036 when matching. If this number is exceeded, we allocate more
1037 space, so it is not a hard limit. */
1038 #ifndef INIT_FAILURE_ALLOC
1039 #define INIT_FAILURE_ALLOC 5
1042 /* Roughly the maximum number of failure points on the stack. Would be
1043 exactly that if always used MAX_FAILURE_ITEMS items each time we failed.
1044 This is a variable only so users of regex can assign to it; we never
1045 change it ourselves. */
1049 #if defined (MATCH_MAY_ALLOCATE)
1050 /* 4400 was enough to cause a crash on Alpha OSF/1,
1051 whose default stack limit is 2mb. */
1052 long int re_max_failures = 4000;
1054 long int re_max_failures = 2000;
1057 union fail_stack_elt
1059 unsigned char *pointer;
1063 typedef union fail_stack_elt fail_stack_elt_t;
1067 fail_stack_elt_t *stack;
1068 unsigned long int size;
1069 unsigned long int avail; /* Offset of next open position. */
1072 #else /* not INT_IS_16BIT */
1074 #if defined (MATCH_MAY_ALLOCATE)
1075 /* 4400 was enough to cause a crash on Alpha OSF/1,
1076 whose default stack limit is 2mb. */
1077 int re_max_failures = 20000;
1079 int re_max_failures = 2000;
1082 union fail_stack_elt
1084 unsigned char *pointer;
1088 typedef union fail_stack_elt fail_stack_elt_t;
1092 fail_stack_elt_t *stack;
1094 unsigned avail; /* Offset of next open position. */
1097 #endif /* INT_IS_16BIT */
1099 #define FAIL_STACK_EMPTY() (fail_stack.avail == 0)
1100 #define FAIL_STACK_PTR_EMPTY() (fail_stack_ptr->avail == 0)
1101 #define FAIL_STACK_FULL() (fail_stack.avail == fail_stack.size)
1104 /* Define macros to initialize and free the failure stack.
1105 Do `return -2' if the alloc fails. */
1107 #ifdef MATCH_MAY_ALLOCATE
1108 #define INIT_FAIL_STACK() \
1110 fail_stack.stack = (fail_stack_elt_t *) \
1111 REGEX_ALLOCATE_STACK (INIT_FAILURE_ALLOC * sizeof (fail_stack_elt_t)); \
1113 if (fail_stack.stack == NULL) \
1116 fail_stack.size = INIT_FAILURE_ALLOC; \
1117 fail_stack.avail = 0; \
1120 #define RESET_FAIL_STACK() REGEX_FREE_STACK (fail_stack.stack)
1122 #define INIT_FAIL_STACK() \
1124 fail_stack.avail = 0; \
1127 #define RESET_FAIL_STACK()
1131 /* Double the size of FAIL_STACK, up to approximately `re_max_failures' items.
1133 Return 1 if succeeds, and 0 if either ran out of memory
1134 allocating space for it or it was already too large.
1136 REGEX_REALLOCATE_STACK requires `destination' be declared. */
1138 #define DOUBLE_FAIL_STACK(fail_stack) \
1139 ((fail_stack).size > (unsigned) (re_max_failures * MAX_FAILURE_ITEMS) \
1141 : ((fail_stack).stack = (fail_stack_elt_t *) \
1142 REGEX_REALLOCATE_STACK ((fail_stack).stack, \
1143 (fail_stack).size * sizeof (fail_stack_elt_t), \
1144 ((fail_stack).size << 1) * sizeof (fail_stack_elt_t)), \
1146 (fail_stack).stack == NULL \
1148 : ((fail_stack).size <<= 1, \
1152 /* Push pointer POINTER on FAIL_STACK.
1153 Return 1 if was able to do so and 0 if ran out of memory allocating
1155 #define PUSH_PATTERN_OP(POINTER, FAIL_STACK) \
1156 ((FAIL_STACK_FULL () \
1157 && !DOUBLE_FAIL_STACK (FAIL_STACK)) \
1159 : ((FAIL_STACK).stack[(FAIL_STACK).avail++].pointer = POINTER, \
1162 /* Push a pointer value onto the failure stack.
1163 Assumes the variable `fail_stack'. Probably should only
1164 be called from within `PUSH_FAILURE_POINT'. */
1165 #define PUSH_FAILURE_POINTER(item) \
1166 fail_stack.stack[fail_stack.avail++].pointer = (unsigned char *) (item)
1168 /* This pushes an integer-valued item onto the failure stack.
1169 Assumes the variable `fail_stack'. Probably should only
1170 be called from within `PUSH_FAILURE_POINT'. */
1171 #define PUSH_FAILURE_INT(item) \
1172 fail_stack.stack[fail_stack.avail++].integer = (item)
1174 /* Push a fail_stack_elt_t value onto the failure stack.
1175 Assumes the variable `fail_stack'. Probably should only
1176 be called from within `PUSH_FAILURE_POINT'. */
1177 #define PUSH_FAILURE_ELT(item) \
1178 fail_stack.stack[fail_stack.avail++] = (item)
1180 /* These three POP... operations complement the three PUSH... operations.
1181 All assume that `fail_stack' is nonempty. */
1182 #define POP_FAILURE_POINTER() fail_stack.stack[--fail_stack.avail].pointer
1183 #define POP_FAILURE_INT() fail_stack.stack[--fail_stack.avail].integer
1184 #define POP_FAILURE_ELT() fail_stack.stack[--fail_stack.avail]
1186 /* Used to omit pushing failure point id's when we're not debugging. */
1188 #define DEBUG_PUSH PUSH_FAILURE_INT
1189 #define DEBUG_POP(item_addr) (item_addr)->integer = POP_FAILURE_INT ()
1191 #define DEBUG_PUSH(item)
1192 #define DEBUG_POP(item_addr)
1196 /* Push the information about the state we will need
1197 if we ever fail back to it.
1199 Requires variables fail_stack, regstart, regend, reg_info, and
1200 num_regs be declared. DOUBLE_FAIL_STACK requires `destination' be
1203 Does `return FAILURE_CODE' if runs out of memory. */
1205 #define PUSH_FAILURE_POINT(pattern_place, string_place, failure_code) \
1207 char *destination; \
1208 /* Must be int, so when we don't save any registers, the arithmetic \
1209 of 0 + -1 isn't done as unsigned. */ \
1210 /* Can't be int, since there is not a shred of a guarantee that int \
1211 is wide enough to hold a value of something to which pointer can \
1215 DEBUG_STATEMENT (failure_id++); \
1216 DEBUG_STATEMENT (nfailure_points_pushed++); \
1217 DEBUG_PRINT2 ("\nPUSH_FAILURE_POINT #%u:\n", failure_id); \
1218 DEBUG_PRINT2 (" Before push, next avail: %d\n", (fail_stack).avail);\
1219 DEBUG_PRINT2 (" size: %d\n", (fail_stack).size);\
1221 DEBUG_PRINT2 (" slots needed: %d\n", NUM_FAILURE_ITEMS); \
1222 DEBUG_PRINT2 (" available: %d\n", REMAINING_AVAIL_SLOTS); \
1224 /* Ensure we have enough space allocated for what we will push. */ \
1225 while (REMAINING_AVAIL_SLOTS < NUM_FAILURE_ITEMS) \
1227 if (!DOUBLE_FAIL_STACK (fail_stack)) \
1228 return failure_code; \
1230 DEBUG_PRINT2 ("\n Doubled stack; size now: %d\n", \
1231 (fail_stack).size); \
1232 DEBUG_PRINT2 (" slots available: %d\n", REMAINING_AVAIL_SLOTS);\
1235 /* Push the info, starting with the registers. */ \
1236 DEBUG_PRINT1 ("\n"); \
1239 for (this_reg = lowest_active_reg; this_reg <= highest_active_reg; \
1242 DEBUG_PRINT2 (" Pushing reg: %d\n", this_reg); \
1243 DEBUG_STATEMENT (num_regs_pushed++); \
1245 DEBUG_PRINT2 (" start: 0x%x\n", regstart[this_reg]); \
1246 PUSH_FAILURE_POINTER (regstart[this_reg]); \
1248 DEBUG_PRINT2 (" end: 0x%x\n", regend[this_reg]); \
1249 PUSH_FAILURE_POINTER (regend[this_reg]); \
1251 DEBUG_PRINT2 (" info: 0x%x\n ", reg_info[this_reg]); \
1252 DEBUG_PRINT2 (" match_null=%d", \
1253 REG_MATCH_NULL_STRING_P (reg_info[this_reg])); \
1254 DEBUG_PRINT2 (" active=%d", IS_ACTIVE (reg_info[this_reg])); \
1255 DEBUG_PRINT2 (" matched_something=%d", \
1256 MATCHED_SOMETHING (reg_info[this_reg])); \
1257 DEBUG_PRINT2 (" ever_matched=%d", \
1258 EVER_MATCHED_SOMETHING (reg_info[this_reg])); \
1259 DEBUG_PRINT1 ("\n"); \
1260 PUSH_FAILURE_ELT (reg_info[this_reg].word); \
1263 DEBUG_PRINT2 (" Pushing low active reg: %d\n", lowest_active_reg);\
1264 PUSH_FAILURE_INT (lowest_active_reg); \
1266 DEBUG_PRINT2 (" Pushing high active reg: %d\n", highest_active_reg);\
1267 PUSH_FAILURE_INT (highest_active_reg); \
1269 DEBUG_PRINT2 (" Pushing pattern 0x%x:\n", pattern_place); \
1270 DEBUG_PRINT_COMPILED_PATTERN (bufp, pattern_place, pend); \
1271 PUSH_FAILURE_POINTER (pattern_place); \
1273 DEBUG_PRINT2 (" Pushing string 0x%x: `", string_place); \
1274 DEBUG_PRINT_DOUBLE_STRING (string_place, string1, size1, string2, \
1276 DEBUG_PRINT1 ("'\n"); \
1277 PUSH_FAILURE_POINTER (string_place); \
1279 DEBUG_PRINT2 (" Pushing failure id: %u\n", failure_id); \
1280 DEBUG_PUSH (failure_id); \
1283 /* This is the number of items that are pushed and popped on the stack
1284 for each register. */
1285 #define NUM_REG_ITEMS 3
1287 /* Individual items aside from the registers. */
1289 #define NUM_NONREG_ITEMS 5 /* Includes failure point id. */
1291 #define NUM_NONREG_ITEMS 4
1294 /* We push at most this many items on the stack. */
1295 /* We used to use (num_regs - 1), which is the number of registers
1296 this regexp will save; but that was changed to 5
1297 to avoid stack overflow for a regexp with lots of parens. */
1298 #define MAX_FAILURE_ITEMS (5 * NUM_REG_ITEMS + NUM_NONREG_ITEMS)
1300 /* We actually push this many items. */
1301 #define NUM_FAILURE_ITEMS \
1303 ? 0 : highest_active_reg - lowest_active_reg + 1) \
1307 /* How many items can still be added to the stack without overflowing it. */
1308 #define REMAINING_AVAIL_SLOTS ((fail_stack).size - (fail_stack).avail)
1311 /* Pops what PUSH_FAIL_STACK pushes.
1313 We restore into the parameters, all of which should be lvalues:
1314 STR -- the saved data position.
1315 PAT -- the saved pattern position.
1316 LOW_REG, HIGH_REG -- the highest and lowest active registers.
1317 REGSTART, REGEND -- arrays of string positions.
1318 REG_INFO -- array of information about each subexpression.
1320 Also assumes the variables `fail_stack' and (if debugging), `bufp',
1321 `pend', `string1', `size1', `string2', and `size2'. */
1323 #define POP_FAILURE_POINT(str, pat, low_reg, high_reg, regstart, regend, reg_info)\
1325 DEBUG_STATEMENT (fail_stack_elt_t failure_id;) \
1327 const unsigned char *string_temp; \
1329 assert (!FAIL_STACK_EMPTY ()); \
1331 /* Remove failure points and point to how many regs pushed. */ \
1332 DEBUG_PRINT1 ("POP_FAILURE_POINT:\n"); \
1333 DEBUG_PRINT2 (" Before pop, next avail: %d\n", fail_stack.avail); \
1334 DEBUG_PRINT2 (" size: %d\n", fail_stack.size); \
1336 assert (fail_stack.avail >= NUM_NONREG_ITEMS); \
1338 DEBUG_POP (&failure_id); \
1339 DEBUG_PRINT2 (" Popping failure id: %u\n", failure_id); \
1341 /* If the saved string location is NULL, it came from an \
1342 on_failure_keep_string_jump opcode, and we want to throw away the \
1343 saved NULL, thus retaining our current position in the string. */ \
1344 string_temp = POP_FAILURE_POINTER (); \
1345 if (string_temp != NULL) \
1346 str = (const char *) string_temp; \
1348 DEBUG_PRINT2 (" Popping string 0x%x: `", str); \
1349 DEBUG_PRINT_DOUBLE_STRING (str, string1, size1, string2, size2); \
1350 DEBUG_PRINT1 ("'\n"); \
1352 pat = (unsigned char *) POP_FAILURE_POINTER (); \
1353 DEBUG_PRINT2 (" Popping pattern 0x%x:\n", pat); \
1354 DEBUG_PRINT_COMPILED_PATTERN (bufp, pat, pend); \
1356 /* Restore register info. */ \
1357 high_reg = (active_reg_t) POP_FAILURE_INT (); \
1358 DEBUG_PRINT2 (" Popping high active reg: %d\n", high_reg); \
1360 low_reg = (active_reg_t) POP_FAILURE_INT (); \
1361 DEBUG_PRINT2 (" Popping low active reg: %d\n", low_reg); \
1364 for (this_reg = high_reg; this_reg >= low_reg; this_reg--) \
1366 DEBUG_PRINT2 (" Popping reg: %d\n", this_reg); \
1368 reg_info[this_reg].word = POP_FAILURE_ELT (); \
1369 DEBUG_PRINT2 (" info: 0x%x\n", reg_info[this_reg]); \
1371 regend[this_reg] = (const char *) POP_FAILURE_POINTER (); \
1372 DEBUG_PRINT2 (" end: 0x%x\n", regend[this_reg]); \
1374 regstart[this_reg] = (const char *) POP_FAILURE_POINTER (); \
1375 DEBUG_PRINT2 (" start: 0x%x\n", regstart[this_reg]); \
1379 for (this_reg = highest_active_reg; this_reg > high_reg; this_reg--) \
1381 reg_info[this_reg].word.integer = 0; \
1382 regend[this_reg] = 0; \
1383 regstart[this_reg] = 0; \
1385 highest_active_reg = high_reg; \
1388 set_regs_matched_done = 0; \
1389 DEBUG_STATEMENT (nfailure_points_popped++); \
1390 } /* POP_FAILURE_POINT */
1394 /* Structure for per-register (a.k.a. per-group) information.
1395 Other register information, such as the
1396 starting and ending positions (which are addresses), and the list of
1397 inner groups (which is a bits list) are maintained in separate
1400 We are making a (strictly speaking) nonportable assumption here: that
1401 the compiler will pack our bit fields into something that fits into
1402 the type of `word', i.e., is something that fits into one item on the
1406 /* Declarations and macros for re_match_2. */
1410 fail_stack_elt_t word;
1413 /* This field is one if this group can match the empty string,
1414 zero if not. If not yet determined, `MATCH_NULL_UNSET_VALUE'. */
1415 #define MATCH_NULL_UNSET_VALUE 3
1416 unsigned match_null_string_p : 2;
1417 unsigned is_active : 1;
1418 unsigned matched_something : 1;
1419 unsigned ever_matched_something : 1;
1421 } register_info_type;
1423 #define REG_MATCH_NULL_STRING_P(R) ((R).bits.match_null_string_p)
1424 #define IS_ACTIVE(R) ((R).bits.is_active)
1425 #define MATCHED_SOMETHING(R) ((R).bits.matched_something)
1426 #define EVER_MATCHED_SOMETHING(R) ((R).bits.ever_matched_something)
1429 /* Call this when have matched a real character; it sets `matched' flags
1430 for the subexpressions which we are currently inside. Also records
1431 that those subexprs have matched. */
1432 #define SET_REGS_MATCHED() \
1435 if (!set_regs_matched_done) \
1438 set_regs_matched_done = 1; \
1439 for (r = lowest_active_reg; r <= highest_active_reg; r++) \
1441 MATCHED_SOMETHING (reg_info[r]) \
1442 = EVER_MATCHED_SOMETHING (reg_info[r]) \
1449 /* Registers are set to a sentinel when they haven't yet matched. */
1450 static char reg_unset_dummy;
1451 #define REG_UNSET_VALUE (®_unset_dummy)
1452 #define REG_UNSET(e) ((e) == REG_UNSET_VALUE)
1454 /* Subroutine declarations and macros for regex_compile. */
1456 static reg_errcode_t regex_compile _RE_ARGS ((const char *pattern, size_t size,
1457 reg_syntax_t syntax,
1458 struct re_pattern_buffer *bufp));
1459 static void store_op1 _RE_ARGS ((re_opcode_t op, unsigned char *loc, int arg));
1460 static void store_op2 _RE_ARGS ((re_opcode_t op, unsigned char *loc,
1461 int arg1, int arg2));
1462 static void insert_op1 _RE_ARGS ((re_opcode_t op, unsigned char *loc,
1463 int arg, unsigned char *end));
1464 static void insert_op2 _RE_ARGS ((re_opcode_t op, unsigned char *loc,
1465 int arg1, int arg2, unsigned char *end));
1466 static boolean at_begline_loc_p _RE_ARGS ((const char *pattern, const char *p,
1467 reg_syntax_t syntax));
1468 static boolean at_endline_loc_p _RE_ARGS ((const char *p, const char *pend,
1469 reg_syntax_t syntax));
1470 static reg_errcode_t compile_range _RE_ARGS ((const char **p_ptr,
1473 reg_syntax_t syntax,
1476 /* Fetch the next character in the uncompiled pattern---translating it
1477 if necessary. Also cast from a signed character in the constant
1478 string passed to us by the user to an unsigned char that we can use
1479 as an array index (in, e.g., `translate'). */
1481 #define PATFETCH(c) \
1482 do {if (p == pend) return REG_EEND; \
1483 c = (unsigned char) *p++; \
1484 if (translate) c = (unsigned char) translate[c]; \
1488 /* Fetch the next character in the uncompiled pattern, with no
1490 #define PATFETCH_RAW(c) \
1491 do {if (p == pend) return REG_EEND; \
1492 c = (unsigned char) *p++; \
1495 /* Go backwards one character in the pattern. */
1496 #define PATUNFETCH p--
1499 /* If `translate' is non-null, return translate[D], else just D. We
1500 cast the subscript to translate because some data is declared as
1501 `char *', to avoid warnings when a string constant is passed. But
1502 when we use a character as a subscript we must make it unsigned. */
1504 #define TRANSLATE(d) \
1505 (translate ? (char) translate[(unsigned char) (d)] : (d))
1509 /* Macros for outputting the compiled pattern into `buffer'. */
1511 /* If the buffer isn't allocated when it comes in, use this. */
1512 #define INIT_BUF_SIZE 32
1514 /* Make sure we have at least N more bytes of space in buffer. */
1515 #define GET_BUFFER_SPACE(n) \
1516 while ((unsigned long) (b - bufp->buffer + (n)) > bufp->allocated) \
1519 /* Make sure we have one more byte of buffer space and then add C to it. */
1520 #define BUF_PUSH(c) \
1522 GET_BUFFER_SPACE (1); \
1523 *b++ = (unsigned char) (c); \
1527 /* Ensure we have two more bytes of buffer space and then append C1 and C2. */
1528 #define BUF_PUSH_2(c1, c2) \
1530 GET_BUFFER_SPACE (2); \
1531 *b++ = (unsigned char) (c1); \
1532 *b++ = (unsigned char) (c2); \
1536 /* As with BUF_PUSH_2, except for three bytes. */
1537 #define BUF_PUSH_3(c1, c2, c3) \
1539 GET_BUFFER_SPACE (3); \
1540 *b++ = (unsigned char) (c1); \
1541 *b++ = (unsigned char) (c2); \
1542 *b++ = (unsigned char) (c3); \
1546 /* Store a jump with opcode OP at LOC to location TO. We store a
1547 relative address offset by the three bytes the jump itself occupies. */
1548 #define STORE_JUMP(op, loc, to) \
1549 store_op1 (op, loc, (int) ((to) - (loc) - 3))
1551 /* Likewise, for a two-argument jump. */
1552 #define STORE_JUMP2(op, loc, to, arg) \
1553 store_op2 (op, loc, (int) ((to) - (loc) - 3), arg)
1555 /* Like `STORE_JUMP', but for inserting. Assume `b' is the buffer end. */
1556 #define INSERT_JUMP(op, loc, to) \
1557 insert_op1 (op, loc, (int) ((to) - (loc) - 3), b)
1559 /* Like `STORE_JUMP2', but for inserting. Assume `b' is the buffer end. */
1560 #define INSERT_JUMP2(op, loc, to, arg) \
1561 insert_op2 (op, loc, (int) ((to) - (loc) - 3), arg, b)
1564 /* This is not an arbitrary limit: the arguments which represent offsets
1565 into the pattern are two bytes long. So if 2^16 bytes turns out to
1566 be too small, many things would have to change. */
1567 /* Any other compiler which, like MSC, has allocation limit below 2^16
1568 bytes will have to use approach similar to what was done below for
1569 MSC and drop MAX_BUF_SIZE a bit. Otherwise you may end up
1570 reallocating to 0 bytes. Such thing is not going to work too well.
1571 You have been warned!! */
1572 #if defined(_MSC_VER) && !defined(WIN32)
1573 /* Microsoft C 16-bit versions limit malloc to approx 65512 bytes.
1574 The REALLOC define eliminates a flurry of conversion warnings,
1575 but is not required. */
1576 #define MAX_BUF_SIZE 65500L
1577 #define REALLOC(p,s) realloc ((p), (size_t) (s))
1579 #define MAX_BUF_SIZE (1L << 16)
1580 #define REALLOC(p,s) realloc ((p), (s))
1583 /* Extend the buffer by twice its current size via realloc and
1584 reset the pointers that pointed into the old block to point to the
1585 correct places in the new one. If extending the buffer results in it
1586 being larger than MAX_BUF_SIZE, then flag memory exhausted. */
1587 #define EXTEND_BUFFER() \
1589 unsigned char *old_buffer = bufp->buffer; \
1590 if (bufp->allocated == MAX_BUF_SIZE) \
1592 bufp->allocated <<= 1; \
1593 if (bufp->allocated > MAX_BUF_SIZE) \
1594 bufp->allocated = MAX_BUF_SIZE; \
1595 bufp->buffer = (unsigned char *) REALLOC (bufp->buffer, bufp->allocated);\
1596 if (bufp->buffer == NULL) \
1597 return REG_ESPACE; \
1598 /* If the buffer moved, move all the pointers into it. */ \
1599 if (old_buffer != bufp->buffer) \
1601 b = (b - old_buffer) + bufp->buffer; \
1602 begalt = (begalt - old_buffer) + bufp->buffer; \
1603 if (fixup_alt_jump) \
1604 fixup_alt_jump = (fixup_alt_jump - old_buffer) + bufp->buffer;\
1606 laststart = (laststart - old_buffer) + bufp->buffer; \
1607 if (pending_exact) \
1608 pending_exact = (pending_exact - old_buffer) + bufp->buffer; \
1613 /* Since we have one byte reserved for the register number argument to
1614 {start,stop}_memory, the maximum number of groups we can report
1615 things about is what fits in that byte. */
1616 #define MAX_REGNUM 255
1618 /* But patterns can have more than `MAX_REGNUM' registers. We just
1619 ignore the excess. */
1620 typedef unsigned regnum_t;
1623 /* Macros for the compile stack. */
1625 /* Since offsets can go either forwards or backwards, this type needs to
1626 be able to hold values from -(MAX_BUF_SIZE - 1) to MAX_BUF_SIZE - 1. */
1627 /* int may be not enough when sizeof(int) == 2. */
1628 typedef long pattern_offset_t;
1632 pattern_offset_t begalt_offset;
1633 pattern_offset_t fixup_alt_jump;
1634 pattern_offset_t inner_group_offset;
1635 pattern_offset_t laststart_offset;
1637 } compile_stack_elt_t;
1642 compile_stack_elt_t *stack;
1644 unsigned avail; /* Offset of next open position. */
1645 } compile_stack_type;
1648 #define INIT_COMPILE_STACK_SIZE 32
1650 #define COMPILE_STACK_EMPTY (compile_stack.avail == 0)
1651 #define COMPILE_STACK_FULL (compile_stack.avail == compile_stack.size)
1653 /* The next available element. */
1654 #define COMPILE_STACK_TOP (compile_stack.stack[compile_stack.avail])
1657 /* Set the bit for character C in a list. */
1658 #define SET_LIST_BIT(c) \
1659 (b[((unsigned char) (c)) / BYTEWIDTH] \
1660 |= 1 << (((unsigned char) c) % BYTEWIDTH))
1663 /* Get the next unsigned number in the uncompiled pattern. */
1664 #define GET_UNSIGNED_NUMBER(num) \
1668 while (ISDIGIT (c)) \
1672 num = num * 10 + c - '0'; \
1680 #if defined _LIBC || (defined HAVE_WCTYPE_H && defined HAVE_WCHAR_H)
1681 /* The GNU C library provides support for user-defined character classes
1682 and the functions from ISO C amendement 1. */
1683 # ifdef CHARCLASS_NAME_MAX
1684 # define CHAR_CLASS_MAX_LENGTH CHARCLASS_NAME_MAX
1686 /* This shouldn't happen but some implementation might still have this
1687 problem. Use a reasonable default value. */
1688 # define CHAR_CLASS_MAX_LENGTH 256
1691 # define IS_CHAR_CLASS(string) wctype (string)
1693 # define CHAR_CLASS_MAX_LENGTH 6 /* Namely, `xdigit'. */
1695 # define IS_CHAR_CLASS(string) \
1696 (STREQ (string, "alpha") || STREQ (string, "upper") \
1697 || STREQ (string, "lower") || STREQ (string, "digit") \
1698 || STREQ (string, "alnum") || STREQ (string, "xdigit") \
1699 || STREQ (string, "space") || STREQ (string, "print") \
1700 || STREQ (string, "punct") || STREQ (string, "graph") \
1701 || STREQ (string, "cntrl") || STREQ (string, "blank"))
1704 #ifndef MATCH_MAY_ALLOCATE
1706 /* If we cannot allocate large objects within re_match_2_internal,
1707 we make the fail stack and register vectors global.
1708 The fail stack, we grow to the maximum size when a regexp
1710 The register vectors, we adjust in size each time we
1711 compile a regexp, according to the number of registers it needs. */
1713 static fail_stack_type fail_stack;
1715 /* Size with which the following vectors are currently allocated.
1716 That is so we can make them bigger as needed,
1717 but never make them smaller. */
1718 static int regs_allocated_size;
1720 static const char ** regstart, ** regend;
1721 static const char ** old_regstart, ** old_regend;
1722 static const char **best_regstart, **best_regend;
1723 static register_info_type *reg_info;
1724 static const char **reg_dummy;
1725 static register_info_type *reg_info_dummy;
1727 /* Make the register vectors big enough for NUM_REGS registers,
1728 but don't make them smaller. */
1731 regex_grow_registers (num_regs)
1734 if (num_regs > regs_allocated_size)
1736 RETALLOC_IF (regstart, num_regs, const char *);
1737 RETALLOC_IF (regend, num_regs, const char *);
1738 RETALLOC_IF (old_regstart, num_regs, const char *);
1739 RETALLOC_IF (old_regend, num_regs, const char *);
1740 RETALLOC_IF (best_regstart, num_regs, const char *);
1741 RETALLOC_IF (best_regend, num_regs, const char *);
1742 RETALLOC_IF (reg_info, num_regs, register_info_type);
1743 RETALLOC_IF (reg_dummy, num_regs, const char *);
1744 RETALLOC_IF (reg_info_dummy, num_regs, register_info_type);
1746 regs_allocated_size = num_regs;
1750 #endif /* not MATCH_MAY_ALLOCATE */
1752 static boolean group_in_compile_stack _RE_ARGS ((compile_stack_type
1756 /* `regex_compile' compiles PATTERN (of length SIZE) according to SYNTAX.
1757 Returns one of error codes defined in `regex.h', or zero for success.
1759 Assumes the `allocated' (and perhaps `buffer') and `translate'
1760 fields are set in BUFP on entry.
1762 If it succeeds, results are put in BUFP (if it returns an error, the
1763 contents of BUFP are undefined):
1764 `buffer' is the compiled pattern;
1765 `syntax' is set to SYNTAX;
1766 `used' is set to the length of the compiled pattern;
1767 `fastmap_accurate' is zero;
1768 `re_nsub' is the number of subexpressions in PATTERN;
1769 `not_bol' and `not_eol' are zero;
1771 The `fastmap' and `newline_anchor' fields are neither
1772 examined nor set. */
1774 /* Return, freeing storage we allocated. */
1775 #define FREE_STACK_RETURN(value) \
1776 return (free (compile_stack.stack), value)
1778 static reg_errcode_t
1779 regex_compile (pattern, size, syntax, bufp)
1780 const char *pattern;
1782 reg_syntax_t syntax;
1783 struct re_pattern_buffer *bufp;
1785 /* We fetch characters from PATTERN here. Even though PATTERN is
1786 `char *' (i.e., signed), we declare these variables as unsigned, so
1787 they can be reliably used as array indices. */
1788 register unsigned char c, c1;
1790 /* A random temporary spot in PATTERN. */
1793 /* Points to the end of the buffer, where we should append. */
1794 register unsigned char *b;
1796 /* Keeps track of unclosed groups. */
1797 compile_stack_type compile_stack;
1799 /* Points to the current (ending) position in the pattern. */
1800 const char *p = pattern;
1801 const char *pend = pattern + size;
1803 /* How to translate the characters in the pattern. */
1804 RE_TRANSLATE_TYPE translate = bufp->translate;
1806 /* Address of the count-byte of the most recently inserted `exactn'
1807 command. This makes it possible to tell if a new exact-match
1808 character can be added to that command or if the character requires
1809 a new `exactn' command. */
1810 unsigned char *pending_exact = 0;
1812 /* Address of start of the most recently finished expression.
1813 This tells, e.g., postfix * where to find the start of its
1814 operand. Reset at the beginning of groups and alternatives. */
1815 unsigned char *laststart = 0;
1817 /* Address of beginning of regexp, or inside of last group. */
1818 unsigned char *begalt;
1820 /* Place in the uncompiled pattern (i.e., the {) to
1821 which to go back if the interval is invalid. */
1822 const char *beg_interval;
1824 /* Address of the place where a forward jump should go to the end of
1825 the containing expression. Each alternative of an `or' -- except the
1826 last -- ends with a forward jump of this sort. */
1827 unsigned char *fixup_alt_jump = 0;
1829 /* Counts open-groups as they are encountered. Remembered for the
1830 matching close-group on the compile stack, so the same register
1831 number is put in the stop_memory as the start_memory. */
1832 regnum_t regnum = 0;
1835 DEBUG_PRINT1 ("\nCompiling pattern: ");
1838 unsigned debug_count;
1840 for (debug_count = 0; debug_count < size; debug_count++)
1841 putchar (pattern[debug_count]);
1846 /* Initialize the compile stack. */
1847 compile_stack.stack = TALLOC (INIT_COMPILE_STACK_SIZE, compile_stack_elt_t);
1848 if (compile_stack.stack == NULL)
1851 compile_stack.size = INIT_COMPILE_STACK_SIZE;
1852 compile_stack.avail = 0;
1854 /* Initialize the pattern buffer. */
1855 bufp->syntax = syntax;
1856 bufp->fastmap_accurate = 0;
1857 bufp->not_bol = bufp->not_eol = 0;
1859 /* Set `used' to zero, so that if we return an error, the pattern
1860 printer (for debugging) will think there's no pattern. We reset it
1864 /* Always count groups, whether or not bufp->no_sub is set. */
1867 #if !defined (emacs) && !defined (SYNTAX_TABLE)
1868 /* Initialize the syntax table. */
1869 init_syntax_once ();
1872 if (bufp->allocated == 0)
1875 { /* If zero allocated, but buffer is non-null, try to realloc
1876 enough space. This loses if buffer's address is bogus, but
1877 that is the user's responsibility. */
1878 RETALLOC (bufp->buffer, INIT_BUF_SIZE, unsigned char);
1881 { /* Caller did not allocate a buffer. Do it for them. */
1882 bufp->buffer = TALLOC (INIT_BUF_SIZE, unsigned char);
1884 if (!bufp->buffer) FREE_STACK_RETURN (REG_ESPACE);
1886 bufp->allocated = INIT_BUF_SIZE;
1889 begalt = b = bufp->buffer;
1891 /* Loop through the uncompiled pattern until we're at the end. */
1900 if ( /* If at start of pattern, it's an operator. */
1902 /* If context independent, it's an operator. */
1903 || syntax & RE_CONTEXT_INDEP_ANCHORS
1904 /* Otherwise, depends on what's come before. */
1905 || at_begline_loc_p (pattern, p, syntax))
1915 if ( /* If at end of pattern, it's an operator. */
1917 /* If context independent, it's an operator. */
1918 || syntax & RE_CONTEXT_INDEP_ANCHORS
1919 /* Otherwise, depends on what's next. */
1920 || at_endline_loc_p (p, pend, syntax))
1930 if ((syntax & RE_BK_PLUS_QM)
1931 || (syntax & RE_LIMITED_OPS))
1935 /* If there is no previous pattern... */
1938 if (syntax & RE_CONTEXT_INVALID_OPS)
1939 FREE_STACK_RETURN (REG_BADRPT);
1940 else if (!(syntax & RE_CONTEXT_INDEP_OPS))
1945 /* Are we optimizing this jump? */
1946 boolean keep_string_p = false;
1948 /* 1 means zero (many) matches is allowed. */
1949 char zero_times_ok = 0, many_times_ok = 0;
1951 /* If there is a sequence of repetition chars, collapse it
1952 down to just one (the right one). We can't combine
1953 interval operators with these because of, e.g., `a{2}*',
1954 which should only match an even number of `a's. */
1958 zero_times_ok |= c != '+';
1959 many_times_ok |= c != '?';
1967 || (!(syntax & RE_BK_PLUS_QM) && (c == '+' || c == '?')))
1970 else if (syntax & RE_BK_PLUS_QM && c == '\\')
1972 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
1975 if (!(c1 == '+' || c1 == '?'))
1990 /* If we get here, we found another repeat character. */
1993 /* Star, etc. applied to an empty pattern is equivalent
1994 to an empty pattern. */
1998 /* Now we know whether or not zero matches is allowed
1999 and also whether or not two or more matches is allowed. */
2001 { /* More than one repetition is allowed, so put in at the
2002 end a backward relative jump from `b' to before the next
2003 jump we're going to put in below (which jumps from
2004 laststart to after this jump).
2006 But if we are at the `*' in the exact sequence `.*\n',
2007 insert an unconditional jump backwards to the .,
2008 instead of the beginning of the loop. This way we only
2009 push a failure point once, instead of every time
2010 through the loop. */
2011 assert (p - 1 > pattern);
2013 /* Allocate the space for the jump. */
2014 GET_BUFFER_SPACE (3);
2016 /* We know we are not at the first character of the pattern,
2017 because laststart was nonzero. And we've already
2018 incremented `p', by the way, to be the character after
2019 the `*'. Do we have to do something analogous here
2020 for null bytes, because of RE_DOT_NOT_NULL? */
2021 if (TRANSLATE (*(p - 2)) == TRANSLATE ('.')
2023 && p < pend && TRANSLATE (*p) == TRANSLATE ('\n')
2024 && !(syntax & RE_DOT_NEWLINE))
2025 { /* We have .*\n. */
2026 STORE_JUMP (jump, b, laststart);
2027 keep_string_p = true;
2030 /* Anything else. */
2031 STORE_JUMP (maybe_pop_jump, b, laststart - 3);
2033 /* We've added more stuff to the buffer. */
2037 /* On failure, jump from laststart to b + 3, which will be the
2038 end of the buffer after this jump is inserted. */
2039 GET_BUFFER_SPACE (3);
2040 INSERT_JUMP (keep_string_p ? on_failure_keep_string_jump
2048 /* At least one repetition is required, so insert a
2049 `dummy_failure_jump' before the initial
2050 `on_failure_jump' instruction of the loop. This
2051 effects a skip over that instruction the first time
2052 we hit that loop. */
2053 GET_BUFFER_SPACE (3);
2054 INSERT_JUMP (dummy_failure_jump, laststart, laststart + 6);
2069 boolean had_char_class = false;
2071 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2073 /* Ensure that we have enough space to push a charset: the
2074 opcode, the length count, and the bitset; 34 bytes in all. */
2075 GET_BUFFER_SPACE (34);
2079 /* We test `*p == '^' twice, instead of using an if
2080 statement, so we only need one BUF_PUSH. */
2081 BUF_PUSH (*p == '^' ? charset_not : charset);
2085 /* Remember the first position in the bracket expression. */
2088 /* Push the number of bytes in the bitmap. */
2089 BUF_PUSH ((1 << BYTEWIDTH) / BYTEWIDTH);
2091 /* Clear the whole map. */
2092 bzero (b, (1 << BYTEWIDTH) / BYTEWIDTH);
2094 /* charset_not matches newline according to a syntax bit. */
2095 if ((re_opcode_t) b[-2] == charset_not
2096 && (syntax & RE_HAT_LISTS_NOT_NEWLINE))
2097 SET_LIST_BIT ('\n');
2099 /* Read in characters and ranges, setting map bits. */
2102 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2106 /* \ might escape characters inside [...] and [^...]. */
2107 if ((syntax & RE_BACKSLASH_ESCAPE_IN_LISTS) && c == '\\')
2109 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
2116 /* Could be the end of the bracket expression. If it's
2117 not (i.e., when the bracket expression is `[]' so
2118 far), the ']' character bit gets set way below. */
2119 if (c == ']' && p != p1 + 1)
2122 /* Look ahead to see if it's a range when the last thing
2123 was a character class. */
2124 if (had_char_class && c == '-' && *p != ']')
2125 FREE_STACK_RETURN (REG_ERANGE);
2127 /* Look ahead to see if it's a range when the last thing
2128 was a character: if this is a hyphen not at the
2129 beginning or the end of a list, then it's the range
2132 && !(p - 2 >= pattern && p[-2] == '[')
2133 && !(p - 3 >= pattern && p[-3] == '[' && p[-2] == '^')
2137 = compile_range (&p, pend, translate, syntax, b);
2138 if (ret != REG_NOERROR) FREE_STACK_RETURN (ret);
2141 else if (p[0] == '-' && p[1] != ']')
2142 { /* This handles ranges made up of characters only. */
2145 /* Move past the `-'. */
2148 ret = compile_range (&p, pend, translate, syntax, b);
2149 if (ret != REG_NOERROR) FREE_STACK_RETURN (ret);
2152 /* See if we're at the beginning of a possible character
2155 else if (syntax & RE_CHAR_CLASSES && c == '[' && *p == ':')
2156 { /* Leave room for the null. */
2157 char str[CHAR_CLASS_MAX_LENGTH + 1];
2162 /* If pattern is `[[:'. */
2163 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2168 if (c == ':' || c == ']' || p == pend
2169 || c1 == CHAR_CLASS_MAX_LENGTH)
2175 /* If isn't a word bracketed by `[:' and:`]':
2176 undo the ending character, the letters, and leave
2177 the leading `:' and `[' (but set bits for them). */
2178 if (c == ':' && *p == ']')
2180 #if defined _LIBC || (defined HAVE_WCTYPE_H && defined HAVE_WCHAR_H)
2181 boolean is_lower = STREQ (str, "lower");
2182 boolean is_upper = STREQ (str, "upper");
2188 FREE_STACK_RETURN (REG_ECTYPE);
2190 /* Throw away the ] at the end of the character
2194 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2196 for (ch = 0; ch < 1 << BYTEWIDTH; ++ch)
2198 if (iswctype (btowc (ch), wt))
2201 if (translate && (is_upper || is_lower)
2202 && (ISUPPER (ch) || ISLOWER (ch)))
2206 had_char_class = true;
2209 boolean is_alnum = STREQ (str, "alnum");
2210 boolean is_alpha = STREQ (str, "alpha");
2211 boolean is_blank = STREQ (str, "blank");
2212 boolean is_cntrl = STREQ (str, "cntrl");
2213 boolean is_digit = STREQ (str, "digit");
2214 boolean is_graph = STREQ (str, "graph");
2215 boolean is_lower = STREQ (str, "lower");
2216 boolean is_print = STREQ (str, "print");
2217 boolean is_punct = STREQ (str, "punct");
2218 boolean is_space = STREQ (str, "space");
2219 boolean is_upper = STREQ (str, "upper");
2220 boolean is_xdigit = STREQ (str, "xdigit");
2222 if (!IS_CHAR_CLASS (str))
2223 FREE_STACK_RETURN (REG_ECTYPE);
2225 /* Throw away the ] at the end of the character
2229 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2231 for (ch = 0; ch < 1 << BYTEWIDTH; ch++)
2233 /* This was split into 3 if's to
2234 avoid an arbitrary limit in some compiler. */
2235 if ( (is_alnum && ISALNUM (ch))
2236 || (is_alpha && ISALPHA (ch))
2237 || (is_blank && ISBLANK (ch))
2238 || (is_cntrl && ISCNTRL (ch)))
2240 if ( (is_digit && ISDIGIT (ch))
2241 || (is_graph && ISGRAPH (ch))
2242 || (is_lower && ISLOWER (ch))
2243 || (is_print && ISPRINT (ch)))
2245 if ( (is_punct && ISPUNCT (ch))
2246 || (is_space && ISSPACE (ch))
2247 || (is_upper && ISUPPER (ch))
2248 || (is_xdigit && ISXDIGIT (ch)))
2250 if ( translate && (is_upper || is_lower)
2251 && (ISUPPER (ch) || ISLOWER (ch)))
2254 had_char_class = true;
2255 #endif /* libc || wctype.h */
2264 had_char_class = false;
2269 had_char_class = false;
2274 /* Discard any (non)matching list bytes that are all 0 at the
2275 end of the map. Decrease the map-length byte too. */
2276 while ((int) b[-1] > 0 && b[b[-1] - 1] == 0)
2284 if (syntax & RE_NO_BK_PARENS)
2291 if (syntax & RE_NO_BK_PARENS)
2298 if (syntax & RE_NEWLINE_ALT)
2305 if (syntax & RE_NO_BK_VBAR)
2312 if (syntax & RE_INTERVALS && syntax & RE_NO_BK_BRACES)
2313 goto handle_interval;
2319 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
2321 /* Do not translate the character after the \, so that we can
2322 distinguish, e.g., \B from \b, even if we normally would
2323 translate, e.g., B to b. */
2329 if (syntax & RE_NO_BK_PARENS)
2330 goto normal_backslash;
2336 if (COMPILE_STACK_FULL)
2338 RETALLOC (compile_stack.stack, compile_stack.size << 1,
2339 compile_stack_elt_t);
2340 if (compile_stack.stack == NULL) return REG_ESPACE;
2342 compile_stack.size <<= 1;
2345 /* These are the values to restore when we hit end of this
2346 group. They are all relative offsets, so that if the
2347 whole pattern moves because of realloc, they will still
2349 COMPILE_STACK_TOP.begalt_offset = begalt - bufp->buffer;
2350 COMPILE_STACK_TOP.fixup_alt_jump
2351 = fixup_alt_jump ? fixup_alt_jump - bufp->buffer + 1 : 0;
2352 COMPILE_STACK_TOP.laststart_offset = b - bufp->buffer;
2353 COMPILE_STACK_TOP.regnum = regnum;
2355 /* We will eventually replace the 0 with the number of
2356 groups inner to this one. But do not push a
2357 start_memory for groups beyond the last one we can
2358 represent in the compiled pattern. */
2359 if (regnum <= MAX_REGNUM)
2361 COMPILE_STACK_TOP.inner_group_offset = b - bufp->buffer + 2;
2362 BUF_PUSH_3 (start_memory, regnum, 0);
2365 compile_stack.avail++;
2370 /* If we've reached MAX_REGNUM groups, then this open
2371 won't actually generate any code, so we'll have to
2372 clear pending_exact explicitly. */
2378 if (syntax & RE_NO_BK_PARENS) goto normal_backslash;
2380 if (COMPILE_STACK_EMPTY)
2381 if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD)
2382 goto normal_backslash;
2384 FREE_STACK_RETURN (REG_ERPAREN);
2388 { /* Push a dummy failure point at the end of the
2389 alternative for a possible future
2390 `pop_failure_jump' to pop. See comments at
2391 `push_dummy_failure' in `re_match_2'. */
2392 BUF_PUSH (push_dummy_failure);
2394 /* We allocated space for this jump when we assigned
2395 to `fixup_alt_jump', in the `handle_alt' case below. */
2396 STORE_JUMP (jump_past_alt, fixup_alt_jump, b - 1);
2399 /* See similar code for backslashed left paren above. */
2400 if (COMPILE_STACK_EMPTY)
2401 if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD)
2404 FREE_STACK_RETURN (REG_ERPAREN);
2406 /* Since we just checked for an empty stack above, this
2407 ``can't happen''. */
2408 assert (compile_stack.avail != 0);
2410 /* We don't just want to restore into `regnum', because
2411 later groups should continue to be numbered higher,
2412 as in `(ab)c(de)' -- the second group is #2. */
2413 regnum_t this_group_regnum;
2415 compile_stack.avail--;
2416 begalt = bufp->buffer + COMPILE_STACK_TOP.begalt_offset;
2418 = COMPILE_STACK_TOP.fixup_alt_jump
2419 ? bufp->buffer + COMPILE_STACK_TOP.fixup_alt_jump - 1
2421 laststart = bufp->buffer + COMPILE_STACK_TOP.laststart_offset;
2422 this_group_regnum = COMPILE_STACK_TOP.regnum;
2423 /* If we've reached MAX_REGNUM groups, then this open
2424 won't actually generate any code, so we'll have to
2425 clear pending_exact explicitly. */
2428 /* We're at the end of the group, so now we know how many
2429 groups were inside this one. */
2430 if (this_group_regnum <= MAX_REGNUM)
2432 unsigned char *inner_group_loc
2433 = bufp->buffer + COMPILE_STACK_TOP.inner_group_offset;
2435 *inner_group_loc = regnum - this_group_regnum;
2436 BUF_PUSH_3 (stop_memory, this_group_regnum,
2437 regnum - this_group_regnum);
2443 case '|': /* `\|'. */
2444 if (syntax & RE_LIMITED_OPS || syntax & RE_NO_BK_VBAR)
2445 goto normal_backslash;
2447 if (syntax & RE_LIMITED_OPS)
2450 /* Insert before the previous alternative a jump which
2451 jumps to this alternative if the former fails. */
2452 GET_BUFFER_SPACE (3);
2453 INSERT_JUMP (on_failure_jump, begalt, b + 6);
2457 /* The alternative before this one has a jump after it
2458 which gets executed if it gets matched. Adjust that
2459 jump so it will jump to this alternative's analogous
2460 jump (put in below, which in turn will jump to the next
2461 (if any) alternative's such jump, etc.). The last such
2462 jump jumps to the correct final destination. A picture:
2468 If we are at `b', then fixup_alt_jump right now points to a
2469 three-byte space after `a'. We'll put in the jump, set
2470 fixup_alt_jump to right after `b', and leave behind three
2471 bytes which we'll fill in when we get to after `c'. */
2474 STORE_JUMP (jump_past_alt, fixup_alt_jump, b);
2476 /* Mark and leave space for a jump after this alternative,
2477 to be filled in later either by next alternative or
2478 when know we're at the end of a series of alternatives. */
2480 GET_BUFFER_SPACE (3);
2489 /* If \{ is a literal. */
2490 if (!(syntax & RE_INTERVALS)
2491 /* If we're at `\{' and it's not the open-interval
2493 || ((syntax & RE_INTERVALS) && (syntax & RE_NO_BK_BRACES))
2494 || (p - 2 == pattern && p == pend))
2495 goto normal_backslash;
2499 /* If got here, then the syntax allows intervals. */
2501 /* At least (most) this many matches must be made. */
2502 int lower_bound = -1, upper_bound = -1;
2504 beg_interval = p - 1;
2508 if (syntax & RE_NO_BK_BRACES)
2509 goto unfetch_interval;
2511 FREE_STACK_RETURN (REG_EBRACE);
2514 GET_UNSIGNED_NUMBER (lower_bound);
2518 GET_UNSIGNED_NUMBER (upper_bound);
2519 if (upper_bound < 0) upper_bound = RE_DUP_MAX;
2522 /* Interval such as `{1}' => match exactly once. */
2523 upper_bound = lower_bound;
2525 if (lower_bound < 0 || upper_bound > RE_DUP_MAX
2526 || lower_bound > upper_bound)
2528 if (syntax & RE_NO_BK_BRACES)
2529 goto unfetch_interval;
2531 FREE_STACK_RETURN (REG_BADBR);
2534 if (!(syntax & RE_NO_BK_BRACES))
2536 if (c != '\\') FREE_STACK_RETURN (REG_EBRACE);
2543 if (syntax & RE_NO_BK_BRACES)
2544 goto unfetch_interval;
2546 FREE_STACK_RETURN (REG_BADBR);
2549 /* We just parsed a valid interval. */
2551 /* If it's invalid to have no preceding re. */
2554 if (syntax & RE_CONTEXT_INVALID_OPS)
2555 FREE_STACK_RETURN (REG_BADRPT);
2556 else if (syntax & RE_CONTEXT_INDEP_OPS)
2559 goto unfetch_interval;
2562 /* If the upper bound is zero, don't want to succeed at
2563 all; jump from `laststart' to `b + 3', which will be
2564 the end of the buffer after we insert the jump. */
2565 if (upper_bound == 0)
2567 GET_BUFFER_SPACE (3);
2568 INSERT_JUMP (jump, laststart, b + 3);
2572 /* Otherwise, we have a nontrivial interval. When
2573 we're all done, the pattern will look like:
2574 set_number_at <jump count> <upper bound>
2575 set_number_at <succeed_n count> <lower bound>
2576 succeed_n <after jump addr> <succeed_n count>
2578 jump_n <succeed_n addr> <jump count>
2579 (The upper bound and `jump_n' are omitted if
2580 `upper_bound' is 1, though.) */
2582 { /* If the upper bound is > 1, we need to insert
2583 more at the end of the loop. */
2584 unsigned nbytes = 10 + (upper_bound > 1) * 10;
2586 GET_BUFFER_SPACE (nbytes);
2588 /* Initialize lower bound of the `succeed_n', even
2589 though it will be set during matching by its
2590 attendant `set_number_at' (inserted next),
2591 because `re_compile_fastmap' needs to know.
2592 Jump to the `jump_n' we might insert below. */
2593 INSERT_JUMP2 (succeed_n, laststart,
2594 b + 5 + (upper_bound > 1) * 5,
2598 /* Code to initialize the lower bound. Insert
2599 before the `succeed_n'. The `5' is the last two
2600 bytes of this `set_number_at', plus 3 bytes of
2601 the following `succeed_n'. */
2602 insert_op2 (set_number_at, laststart, 5, lower_bound, b);
2605 if (upper_bound > 1)
2606 { /* More than one repetition is allowed, so
2607 append a backward jump to the `succeed_n'
2608 that starts this interval.
2610 When we've reached this during matching,
2611 we'll have matched the interval once, so
2612 jump back only `upper_bound - 1' times. */
2613 STORE_JUMP2 (jump_n, b, laststart + 5,
2617 /* The location we want to set is the second
2618 parameter of the `jump_n'; that is `b-2' as
2619 an absolute address. `laststart' will be
2620 the `set_number_at' we're about to insert;
2621 `laststart+3' the number to set, the source
2622 for the relative address. But we are
2623 inserting into the middle of the pattern --
2624 so everything is getting moved up by 5.
2625 Conclusion: (b - 2) - (laststart + 3) + 5,
2626 i.e., b - laststart.
2628 We insert this at the beginning of the loop
2629 so that if we fail during matching, we'll
2630 reinitialize the bounds. */
2631 insert_op2 (set_number_at, laststart, b - laststart,
2632 upper_bound - 1, b);
2637 beg_interval = NULL;
2642 /* If an invalid interval, match the characters as literals. */
2643 assert (beg_interval);
2645 beg_interval = NULL;
2647 /* normal_char and normal_backslash need `c'. */
2650 if (!(syntax & RE_NO_BK_BRACES))
2652 if (p > pattern && p[-1] == '\\')
2653 goto normal_backslash;
2658 /* There is no way to specify the before_dot and after_dot
2659 operators. rms says this is ok. --karl */
2667 BUF_PUSH_2 (syntaxspec, syntax_spec_code[c]);
2673 BUF_PUSH_2 (notsyntaxspec, syntax_spec_code[c]);
2679 if (re_syntax_options & RE_NO_GNU_OPS)
2682 BUF_PUSH (wordchar);
2687 if (re_syntax_options & RE_NO_GNU_OPS)
2690 BUF_PUSH (notwordchar);
2695 if (re_syntax_options & RE_NO_GNU_OPS)
2701 if (re_syntax_options & RE_NO_GNU_OPS)
2707 if (re_syntax_options & RE_NO_GNU_OPS)
2709 BUF_PUSH (wordbound);
2713 if (re_syntax_options & RE_NO_GNU_OPS)
2715 BUF_PUSH (notwordbound);
2719 if (re_syntax_options & RE_NO_GNU_OPS)
2725 if (re_syntax_options & RE_NO_GNU_OPS)
2730 case '1': case '2': case '3': case '4': case '5':
2731 case '6': case '7': case '8': case '9':
2732 if (syntax & RE_NO_BK_REFS)
2738 FREE_STACK_RETURN (REG_ESUBREG);
2740 /* Can't back reference to a subexpression if inside of it. */
2741 if (group_in_compile_stack (compile_stack, (regnum_t) c1))
2745 BUF_PUSH_2 (duplicate, c1);
2751 if (syntax & RE_BK_PLUS_QM)
2754 goto normal_backslash;
2758 /* You might think it would be useful for \ to mean
2759 not to translate; but if we don't translate it
2760 it will never match anything. */
2768 /* Expects the character in `c'. */
2770 /* If no exactn currently being built. */
2773 /* If last exactn not at current position. */
2774 || pending_exact + *pending_exact + 1 != b
2776 /* We have only one byte following the exactn for the count. */
2777 || *pending_exact == (1 << BYTEWIDTH) - 1
2779 /* If followed by a repetition operator. */
2780 || *p == '*' || *p == '^'
2781 || ((syntax & RE_BK_PLUS_QM)
2782 ? *p == '\\' && (p[1] == '+' || p[1] == '?')
2783 : (*p == '+' || *p == '?'))
2784 || ((syntax & RE_INTERVALS)
2785 && ((syntax & RE_NO_BK_BRACES)
2787 : (p[0] == '\\' && p[1] == '{'))))
2789 /* Start building a new exactn. */
2793 BUF_PUSH_2 (exactn, 0);
2794 pending_exact = b - 1;
2801 } /* while p != pend */
2804 /* Through the pattern now. */
2807 STORE_JUMP (jump_past_alt, fixup_alt_jump, b);
2809 if (!COMPILE_STACK_EMPTY)
2810 FREE_STACK_RETURN (REG_EPAREN);
2812 /* If we don't want backtracking, force success
2813 the first time we reach the end of the compiled pattern. */
2814 if (syntax & RE_NO_POSIX_BACKTRACKING)
2817 free (compile_stack.stack);
2819 /* We have succeeded; set the length of the buffer. */
2820 bufp->used = b - bufp->buffer;
2825 DEBUG_PRINT1 ("\nCompiled pattern: \n");
2826 print_compiled_pattern (bufp);
2830 #ifndef MATCH_MAY_ALLOCATE
2831 /* Initialize the failure stack to the largest possible stack. This
2832 isn't necessary unless we're trying to avoid calling alloca in
2833 the search and match routines. */
2835 int num_regs = bufp->re_nsub + 1;
2837 /* Since DOUBLE_FAIL_STACK refuses to double only if the current size
2838 is strictly greater than re_max_failures, the largest possible stack
2839 is 2 * re_max_failures failure points. */
2840 if (fail_stack.size < (2 * re_max_failures * MAX_FAILURE_ITEMS))
2842 fail_stack.size = (2 * re_max_failures * MAX_FAILURE_ITEMS);
2845 if (! fail_stack.stack)
2847 = (fail_stack_elt_t *) xmalloc (fail_stack.size
2848 * sizeof (fail_stack_elt_t));
2851 = (fail_stack_elt_t *) xrealloc (fail_stack.stack,
2853 * sizeof (fail_stack_elt_t)));
2854 #else /* not emacs */
2855 if (! fail_stack.stack)
2857 = (fail_stack_elt_t *) malloc (fail_stack.size
2858 * sizeof (fail_stack_elt_t));
2861 = (fail_stack_elt_t *) realloc (fail_stack.stack,
2863 * sizeof (fail_stack_elt_t)));
2864 #endif /* not emacs */
2867 regex_grow_registers (num_regs);
2869 #endif /* not MATCH_MAY_ALLOCATE */
2872 } /* regex_compile */
2874 /* Subroutines for `regex_compile'. */
2876 /* Store OP at LOC followed by two-byte integer parameter ARG. */
2879 store_op1 (op, loc, arg)
2884 *loc = (unsigned char) op;
2885 STORE_NUMBER (loc + 1, arg);
2889 /* Like `store_op1', but for two two-byte parameters ARG1 and ARG2. */
2892 store_op2 (op, loc, arg1, arg2)
2897 *loc = (unsigned char) op;
2898 STORE_NUMBER (loc + 1, arg1);
2899 STORE_NUMBER (loc + 3, arg2);
2903 /* Copy the bytes from LOC to END to open up three bytes of space at LOC
2904 for OP followed by two-byte integer parameter ARG. */
2907 insert_op1 (op, loc, arg, end)
2913 register unsigned char *pfrom = end;
2914 register unsigned char *pto = end + 3;
2916 while (pfrom != loc)
2919 store_op1 (op, loc, arg);
2923 /* Like `insert_op1', but for two two-byte parameters ARG1 and ARG2. */
2926 insert_op2 (op, loc, arg1, arg2, end)
2932 register unsigned char *pfrom = end;
2933 register unsigned char *pto = end + 5;
2935 while (pfrom != loc)
2938 store_op2 (op, loc, arg1, arg2);
2942 /* P points to just after a ^ in PATTERN. Return true if that ^ comes
2943 after an alternative or a begin-subexpression. We assume there is at
2944 least one character before the ^. */
2947 at_begline_loc_p (pattern, p, syntax)
2948 const char *pattern, *p;
2949 reg_syntax_t syntax;
2951 const char *prev = p - 2;
2952 boolean prev_prev_backslash = prev > pattern && prev[-1] == '\\';
2955 /* After a subexpression? */
2956 (*prev == '(' && (syntax & RE_NO_BK_PARENS || prev_prev_backslash))
2957 /* After an alternative? */
2958 || (*prev == '|' && (syntax & RE_NO_BK_VBAR || prev_prev_backslash));
2962 /* The dual of at_begline_loc_p. This one is for $. We assume there is
2963 at least one character after the $, i.e., `P < PEND'. */
2966 at_endline_loc_p (p, pend, syntax)
2967 const char *p, *pend;
2968 reg_syntax_t syntax;
2970 const char *next = p;
2971 boolean next_backslash = *next == '\\';
2972 const char *next_next = p + 1 < pend ? p + 1 : 0;
2975 /* Before a subexpression? */
2976 (syntax & RE_NO_BK_PARENS ? *next == ')'
2977 : next_backslash && next_next && *next_next == ')')
2978 /* Before an alternative? */
2979 || (syntax & RE_NO_BK_VBAR ? *next == '|'
2980 : next_backslash && next_next && *next_next == '|');
2984 /* Returns true if REGNUM is in one of COMPILE_STACK's elements and
2985 false if it's not. */
2988 group_in_compile_stack (compile_stack, regnum)
2989 compile_stack_type compile_stack;
2994 for (this_element = compile_stack.avail - 1;
2997 if (compile_stack.stack[this_element].regnum == regnum)
3004 /* Read the ending character of a range (in a bracket expression) from the
3005 uncompiled pattern *P_PTR (which ends at PEND). We assume the
3006 starting character is in `P[-2]'. (`P[-1]' is the character `-'.)
3007 Then we set the translation of all bits between the starting and
3008 ending characters (inclusive) in the compiled pattern B.
3010 Return an error code.
3012 We use these short variable names so we can use the same macros as
3013 `regex_compile' itself. */
3015 static reg_errcode_t
3016 compile_range (p_ptr, pend, translate, syntax, b)
3017 const char **p_ptr, *pend;
3018 RE_TRANSLATE_TYPE translate;
3019 reg_syntax_t syntax;
3024 const char *p = *p_ptr;
3025 unsigned int range_start, range_end;
3030 /* Even though the pattern is a signed `char *', we need to fetch
3031 with unsigned char *'s; if the high bit of the pattern character
3032 is set, the range endpoints will be negative if we fetch using a
3035 We also want to fetch the endpoints without translating them; the
3036 appropriate translation is done in the bit-setting loop below. */
3037 /* The SVR4 compiler on the 3B2 had trouble with unsigned const char *. */
3038 range_start = ((const unsigned char *) p)[-2];
3039 range_end = ((const unsigned char *) p)[0];
3041 /* Have to increment the pointer into the pattern string, so the
3042 caller isn't still at the ending character. */
3045 /* If the start is after the end, the range is empty. */
3046 if (range_start > range_end)
3047 return syntax & RE_NO_EMPTY_RANGES ? REG_ERANGE : REG_NOERROR;
3049 /* Here we see why `this_char' has to be larger than an `unsigned
3050 char' -- the range is inclusive, so if `range_end' == 0xff
3051 (assuming 8-bit characters), we would otherwise go into an infinite
3052 loop, since all characters <= 0xff. */
3053 for (this_char = range_start; this_char <= range_end; this_char++)
3055 SET_LIST_BIT (TRANSLATE (this_char));
3061 /* re_compile_fastmap computes a ``fastmap'' for the compiled pattern in
3062 BUFP. A fastmap records which of the (1 << BYTEWIDTH) possible
3063 characters can start a string that matches the pattern. This fastmap
3064 is used by re_search to skip quickly over impossible starting points.
3066 The caller must supply the address of a (1 << BYTEWIDTH)-byte data
3067 area as BUFP->fastmap.
3069 We set the `fastmap', `fastmap_accurate', and `can_be_null' fields in
3072 Returns 0 if we succeed, -2 if an internal error. */
3075 re_compile_fastmap (bufp)
3076 struct re_pattern_buffer *bufp;
3079 #ifdef MATCH_MAY_ALLOCATE
3080 fail_stack_type fail_stack;
3082 #ifndef REGEX_MALLOC
3085 /* We don't push any register information onto the failure stack. */
3086 unsigned num_regs = 0;
3088 register char *fastmap = bufp->fastmap;
3089 unsigned char *pattern = bufp->buffer;
3090 unsigned char *p = pattern;
3091 register unsigned char *pend = pattern + bufp->used;
3094 /* This holds the pointer to the failure stack, when
3095 it is allocated relocatably. */
3096 fail_stack_elt_t *failure_stack_ptr;
3099 /* Assume that each path through the pattern can be null until
3100 proven otherwise. We set this false at the bottom of switch
3101 statement, to which we get only if a particular path doesn't
3102 match the empty string. */
3103 boolean path_can_be_null = true;
3105 /* We aren't doing a `succeed_n' to begin with. */
3106 boolean succeed_n_p = false;
3108 assert (fastmap != NULL && p != NULL);
3111 bzero (fastmap, 1 << BYTEWIDTH); /* Assume nothing's valid. */
3112 bufp->fastmap_accurate = 1; /* It will be when we're done. */
3113 bufp->can_be_null = 0;
3117 if (p == pend || *p == succeed)
3119 /* We have reached the (effective) end of pattern. */
3120 if (!FAIL_STACK_EMPTY ())
3122 bufp->can_be_null |= path_can_be_null;
3124 /* Reset for next path. */
3125 path_can_be_null = true;
3127 p = fail_stack.stack[--fail_stack.avail].pointer;
3135 /* We should never be about to go beyond the end of the pattern. */
3138 switch (SWITCH_ENUM_CAST ((re_opcode_t) *p++))
3141 /* I guess the idea here is to simply not bother with a fastmap
3142 if a backreference is used, since it's too hard to figure out
3143 the fastmap for the corresponding group. Setting
3144 `can_be_null' stops `re_search_2' from using the fastmap, so
3145 that is all we do. */
3147 bufp->can_be_null = 1;
3151 /* Following are the cases which match a character. These end
3160 for (j = *p++ * BYTEWIDTH - 1; j >= 0; j--)
3161 if (p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH)))
3167 /* Chars beyond end of map must be allowed. */
3168 for (j = *p * BYTEWIDTH; j < (1 << BYTEWIDTH); j++)
3171 for (j = *p++ * BYTEWIDTH - 1; j >= 0; j--)
3172 if (!(p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH))))
3178 for (j = 0; j < (1 << BYTEWIDTH); j++)
3179 if (SYNTAX (j) == Sword)
3185 for (j = 0; j < (1 << BYTEWIDTH); j++)
3186 if (SYNTAX (j) != Sword)
3193 int fastmap_newline = fastmap['\n'];
3195 /* `.' matches anything ... */
3196 for (j = 0; j < (1 << BYTEWIDTH); j++)
3199 /* ... except perhaps newline. */
3200 if (!(bufp->syntax & RE_DOT_NEWLINE))
3201 fastmap['\n'] = fastmap_newline;
3203 /* Return if we have already set `can_be_null'; if we have,
3204 then the fastmap is irrelevant. Something's wrong here. */
3205 else if (bufp->can_be_null)
3208 /* Otherwise, have to check alternative paths. */
3215 for (j = 0; j < (1 << BYTEWIDTH); j++)
3216 if (SYNTAX (j) == (enum syntaxcode) k)
3223 for (j = 0; j < (1 << BYTEWIDTH); j++)
3224 if (SYNTAX (j) != (enum syntaxcode) k)
3229 /* All cases after this match the empty string. These end with
3249 case push_dummy_failure:
3254 case pop_failure_jump:
3255 case maybe_pop_jump:
3258 case dummy_failure_jump:
3259 EXTRACT_NUMBER_AND_INCR (j, p);
3264 /* Jump backward implies we just went through the body of a
3265 loop and matched nothing. Opcode jumped to should be
3266 `on_failure_jump' or `succeed_n'. Just treat it like an
3267 ordinary jump. For a * loop, it has pushed its failure
3268 point already; if so, discard that as redundant. */
3269 if ((re_opcode_t) *p != on_failure_jump
3270 && (re_opcode_t) *p != succeed_n)
3274 EXTRACT_NUMBER_AND_INCR (j, p);
3277 /* If what's on the stack is where we are now, pop it. */
3278 if (!FAIL_STACK_EMPTY ()
3279 && fail_stack.stack[fail_stack.avail - 1].pointer == p)
3285 case on_failure_jump:
3286 case on_failure_keep_string_jump:
3287 handle_on_failure_jump:
3288 EXTRACT_NUMBER_AND_INCR (j, p);
3290 /* For some patterns, e.g., `(a?)?', `p+j' here points to the
3291 end of the pattern. We don't want to push such a point,
3292 since when we restore it above, entering the switch will
3293 increment `p' past the end of the pattern. We don't need
3294 to push such a point since we obviously won't find any more
3295 fastmap entries beyond `pend'. Such a pattern can match
3296 the null string, though. */
3299 if (!PUSH_PATTERN_OP (p + j, fail_stack))
3301 RESET_FAIL_STACK ();
3306 bufp->can_be_null = 1;
3310 EXTRACT_NUMBER_AND_INCR (k, p); /* Skip the n. */
3311 succeed_n_p = false;
3318 /* Get to the number of times to succeed. */
3321 /* Increment p past the n for when k != 0. */
3322 EXTRACT_NUMBER_AND_INCR (k, p);
3326 succeed_n_p = true; /* Spaghetti code alert. */
3327 goto handle_on_failure_jump;
3344 abort (); /* We have listed all the cases. */
3347 /* Getting here means we have found the possible starting
3348 characters for one path of the pattern -- and that the empty
3349 string does not match. We need not follow this path further.
3350 Instead, look at the next alternative (remembered on the
3351 stack), or quit if no more. The test at the top of the loop
3352 does these things. */
3353 path_can_be_null = false;
3357 /* Set `can_be_null' for the last path (also the first path, if the
3358 pattern is empty). */
3359 bufp->can_be_null |= path_can_be_null;
3362 RESET_FAIL_STACK ();
3364 } /* re_compile_fastmap */
3366 /* Set REGS to hold NUM_REGS registers, storing them in STARTS and
3367 ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use
3368 this memory for recording register information. STARTS and ENDS
3369 must be allocated using the malloc library routine, and must each
3370 be at least NUM_REGS * sizeof (regoff_t) bytes long.
3372 If NUM_REGS == 0, then subsequent matches should allocate their own
3375 Unless this function is called, the first search or match using
3376 PATTERN_BUFFER will allocate its own register data, without
3377 freeing the old data. */
3380 re_set_registers (bufp, regs, num_regs, starts, ends)
3381 struct re_pattern_buffer *bufp;
3382 struct re_registers *regs;
3384 regoff_t *starts, *ends;
3388 bufp->regs_allocated = REGS_REALLOCATE;
3389 regs->num_regs = num_regs;
3390 regs->start = starts;
3395 bufp->regs_allocated = REGS_UNALLOCATED;
3397 regs->start = regs->end = (regoff_t *) 0;
3401 /* Searching routines. */
3403 /* Like re_search_2, below, but only one string is specified, and
3404 doesn't let you say where to stop matching. */
3407 re_search (bufp, string, size, startpos, range, regs)
3408 struct re_pattern_buffer *bufp;
3410 int size, startpos, range;
3411 struct re_registers *regs;
3413 return re_search_2 (bufp, NULL, 0, string, size, startpos, range,
3418 /* Using the compiled pattern in BUFP->buffer, first tries to match the
3419 virtual concatenation of STRING1 and STRING2, starting first at index
3420 STARTPOS, then at STARTPOS + 1, and so on.
3422 STRING1 and STRING2 have length SIZE1 and SIZE2, respectively.
3424 RANGE is how far to scan while trying to match. RANGE = 0 means try
3425 only at STARTPOS; in general, the last start tried is STARTPOS +
3428 In REGS, return the indices of the virtual concatenation of STRING1
3429 and STRING2 that matched the entire BUFP->buffer and its contained
3432 Do not consider matching one past the index STOP in the virtual
3433 concatenation of STRING1 and STRING2.
3435 We return either the position in the strings at which the match was
3436 found, -1 if no match, or -2 if error (such as failure
3440 re_search_2 (bufp, string1, size1, string2, size2, startpos, range, regs, stop)
3441 struct re_pattern_buffer *bufp;
3442 const char *string1, *string2;
3446 struct re_registers *regs;
3450 register char *fastmap = bufp->fastmap;
3451 register RE_TRANSLATE_TYPE translate = bufp->translate;
3452 int total_size = size1 + size2;
3453 int endpos = startpos + range;
3455 /* Check for out-of-range STARTPOS. */
3456 if (startpos < 0 || startpos > total_size)
3459 /* Fix up RANGE if it might eventually take us outside
3460 the virtual concatenation of STRING1 and STRING2.
3461 Make sure we won't move STARTPOS below 0 or above TOTAL_SIZE. */
3463 range = 0 - startpos;
3464 else if (endpos > total_size)
3465 range = total_size - startpos;
3467 /* If the search isn't to be a backwards one, don't waste time in a
3468 search for a pattern that must be anchored. */
3469 if (bufp->used > 0 && (re_opcode_t) bufp->buffer[0] == begbuf && range > 0)
3478 /* In a forward search for something that starts with \=.
3479 don't keep searching past point. */
3480 if (bufp->used > 0 && (re_opcode_t) bufp->buffer[0] == at_dot && range > 0)
3482 range = PT - startpos;
3488 /* Update the fastmap now if not correct already. */
3489 if (fastmap && !bufp->fastmap_accurate)
3490 if (re_compile_fastmap (bufp) == -2)
3493 /* Loop through the string, looking for a place to start matching. */
3496 /* If a fastmap is supplied, skip quickly over characters that
3497 cannot be the start of a match. If the pattern can match the
3498 null string, however, we don't need to skip characters; we want
3499 the first null string. */
3500 if (fastmap && startpos < total_size && !bufp->can_be_null)
3502 if (range > 0) /* Searching forwards. */
3504 register const char *d;
3505 register int lim = 0;
3508 if (startpos < size1 && startpos + range >= size1)
3509 lim = range - (size1 - startpos);
3511 d = (startpos >= size1 ? string2 - size1 : string1) + startpos;
3513 /* Written out as an if-else to avoid testing `translate'
3517 && !fastmap[(unsigned char)
3518 translate[(unsigned char) *d++]])
3521 while (range > lim && !fastmap[(unsigned char) *d++])
3524 startpos += irange - range;
3526 else /* Searching backwards. */
3528 register char c = (size1 == 0 || startpos >= size1
3529 ? string2[startpos - size1]
3530 : string1[startpos]);
3532 if (!fastmap[(unsigned char) TRANSLATE (c)])
3537 /* If can't match the null string, and that's all we have left, fail. */
3538 if (range >= 0 && startpos == total_size && fastmap
3539 && !bufp->can_be_null)
3542 val = re_match_2_internal (bufp, string1, size1, string2, size2,
3543 startpos, regs, stop);
3544 #ifndef REGEX_MALLOC
3573 /* This converts PTR, a pointer into one of the search strings `string1'
3574 and `string2' into an offset from the beginning of that string. */
3575 #define POINTER_TO_OFFSET(ptr) \
3576 (FIRST_STRING_P (ptr) \
3577 ? ((regoff_t) ((ptr) - string1)) \
3578 : ((regoff_t) ((ptr) - string2 + size1)))
3580 /* Macros for dealing with the split strings in re_match_2. */
3582 #define MATCHING_IN_FIRST_STRING (dend == end_match_1)
3584 /* Call before fetching a character with *d. This switches over to
3585 string2 if necessary. */
3586 #define PREFETCH() \
3589 /* End of string2 => fail. */ \
3590 if (dend == end_match_2) \
3592 /* End of string1 => advance to string2. */ \
3594 dend = end_match_2; \
3598 /* Test if at very beginning or at very end of the virtual concatenation
3599 of `string1' and `string2'. If only one string, it's `string2'. */
3600 #define AT_STRINGS_BEG(d) ((d) == (size1 ? string1 : string2) || !size2)
3601 #define AT_STRINGS_END(d) ((d) == end2)
3604 /* Test if D points to a character which is word-constituent. We have
3605 two special cases to check for: if past the end of string1, look at
3606 the first character in string2; and if before the beginning of
3607 string2, look at the last character in string1. */
3608 #define WORDCHAR_P(d) \
3609 (SYNTAX ((d) == end1 ? *string2 \
3610 : (d) == string2 - 1 ? *(end1 - 1) : *(d)) \
3613 /* Disabled due to a compiler bug -- see comment at case wordbound */
3615 /* Test if the character before D and the one at D differ with respect
3616 to being word-constituent. */
3617 #define AT_WORD_BOUNDARY(d) \
3618 (AT_STRINGS_BEG (d) || AT_STRINGS_END (d) \
3619 || WORDCHAR_P (d - 1) != WORDCHAR_P (d))
3622 /* Free everything we malloc. */
3623 #ifdef MATCH_MAY_ALLOCATE
3624 #define FREE_VAR(var) if (var) REGEX_FREE (var); var = NULL
3625 #define FREE_VARIABLES() \
3627 REGEX_FREE_STACK (fail_stack.stack); \
3628 FREE_VAR (regstart); \
3629 FREE_VAR (regend); \
3630 FREE_VAR (old_regstart); \
3631 FREE_VAR (old_regend); \
3632 FREE_VAR (best_regstart); \
3633 FREE_VAR (best_regend); \
3634 FREE_VAR (reg_info); \
3635 FREE_VAR (reg_dummy); \
3636 FREE_VAR (reg_info_dummy); \
3639 #define FREE_VARIABLES() ((void)0) /* Do nothing! But inhibit gcc warning. */
3640 #endif /* not MATCH_MAY_ALLOCATE */
3642 /* These values must meet several constraints. They must not be valid
3643 register values; since we have a limit of 255 registers (because
3644 we use only one byte in the pattern for the register number), we can
3645 use numbers larger than 255. They must differ by 1, because of
3646 NUM_FAILURE_ITEMS above. And the value for the lowest register must
3647 be larger than the value for the highest register, so we do not try
3648 to actually save any registers when none are active. */
3649 #define NO_HIGHEST_ACTIVE_REG (1 << BYTEWIDTH)
3650 #define NO_LOWEST_ACTIVE_REG (NO_HIGHEST_ACTIVE_REG + 1)
3652 /* Matching routines. */
3654 #ifndef emacs /* Emacs never uses this. */
3655 /* re_match is like re_match_2 except it takes only a single string. */
3658 re_match (bufp, string, size, pos, regs)
3659 struct re_pattern_buffer *bufp;
3662 struct re_registers *regs;
3664 int result = re_match_2_internal (bufp, NULL, 0, string, size,
3666 #ifndef REGEX_MALLOC
3673 #endif /* not emacs */
3675 static boolean group_match_null_string_p _RE_ARGS ((unsigned char **p,
3677 register_info_type *reg_info));
3678 static boolean alt_match_null_string_p _RE_ARGS ((unsigned char *p,
3680 register_info_type *reg_info));
3681 static boolean common_op_match_null_string_p _RE_ARGS ((unsigned char **p,
3683 register_info_type *reg_info));
3684 static int bcmp_translate _RE_ARGS ((const char *s1, const char *s2,
3685 int len, char *translate));
3687 /* re_match_2 matches the compiled pattern in BUFP against the
3688 the (virtual) concatenation of STRING1 and STRING2 (of length SIZE1
3689 and SIZE2, respectively). We start matching at POS, and stop
3692 If REGS is non-null and the `no_sub' field of BUFP is nonzero, we
3693 store offsets for the substring each group matched in REGS. See the
3694 documentation for exactly how many groups we fill.
3696 We return -1 if no match, -2 if an internal error (such as the
3697 failure stack overflowing). Otherwise, we return the length of the
3698 matched substring. */
3701 re_match_2 (bufp, string1, size1, string2, size2, pos, regs, stop)
3702 struct re_pattern_buffer *bufp;
3703 const char *string1, *string2;
3706 struct re_registers *regs;
3709 int result = re_match_2_internal (bufp, string1, size1, string2, size2,
3711 #ifndef REGEX_MALLOC
3719 /* This is a separate function so that we can force an alloca cleanup
3722 re_match_2_internal (bufp, string1, size1, string2, size2, pos, regs, stop)
3723 struct re_pattern_buffer *bufp;
3724 const char *string1, *string2;
3727 struct re_registers *regs;
3730 /* General temporaries. */
3734 /* Just past the end of the corresponding string. */
3735 const char *end1, *end2;
3737 /* Pointers into string1 and string2, just past the last characters in
3738 each to consider matching. */
3739 const char *end_match_1, *end_match_2;
3741 /* Where we are in the data, and the end of the current string. */
3742 const char *d, *dend;
3744 /* Where we are in the pattern, and the end of the pattern. */
3745 unsigned char *p = bufp->buffer;
3746 register unsigned char *pend = p + bufp->used;
3748 /* Mark the opcode just after a start_memory, so we can test for an
3749 empty subpattern when we get to the stop_memory. */
3750 unsigned char *just_past_start_mem = 0;
3752 /* We use this to map every character in the string. */
3753 RE_TRANSLATE_TYPE translate = bufp->translate;
3755 /* Failure point stack. Each place that can handle a failure further
3756 down the line pushes a failure point on this stack. It consists of
3757 restart, regend, and reg_info for all registers corresponding to
3758 the subexpressions we're currently inside, plus the number of such
3759 registers, and, finally, two char *'s. The first char * is where
3760 to resume scanning the pattern; the second one is where to resume
3761 scanning the strings. If the latter is zero, the failure point is
3762 a ``dummy''; if a failure happens and the failure point is a dummy,
3763 it gets discarded and the next next one is tried. */
3764 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
3765 fail_stack_type fail_stack;
3768 static unsigned failure_id = 0;
3769 unsigned nfailure_points_pushed = 0, nfailure_points_popped = 0;
3773 /* This holds the pointer to the failure stack, when
3774 it is allocated relocatably. */
3775 fail_stack_elt_t *failure_stack_ptr;
3778 /* We fill all the registers internally, independent of what we
3779 return, for use in backreferences. The number here includes
3780 an element for register zero. */
3781 size_t num_regs = bufp->re_nsub + 1;
3783 /* The currently active registers. */
3784 active_reg_t lowest_active_reg = NO_LOWEST_ACTIVE_REG;
3785 active_reg_t highest_active_reg = NO_HIGHEST_ACTIVE_REG;
3787 /* Information on the contents of registers. These are pointers into
3788 the input strings; they record just what was matched (on this
3789 attempt) by a subexpression part of the pattern, that is, the
3790 regnum-th regstart pointer points to where in the pattern we began
3791 matching and the regnum-th regend points to right after where we
3792 stopped matching the regnum-th subexpression. (The zeroth register
3793 keeps track of what the whole pattern matches.) */
3794 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
3795 const char **regstart, **regend;
3798 /* If a group that's operated upon by a repetition operator fails to
3799 match anything, then the register for its start will need to be
3800 restored because it will have been set to wherever in the string we
3801 are when we last see its open-group operator. Similarly for a
3803 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
3804 const char **old_regstart, **old_regend;
3807 /* The is_active field of reg_info helps us keep track of which (possibly
3808 nested) subexpressions we are currently in. The matched_something
3809 field of reg_info[reg_num] helps us tell whether or not we have
3810 matched any of the pattern so far this time through the reg_num-th
3811 subexpression. These two fields get reset each time through any
3812 loop their register is in. */
3813 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
3814 register_info_type *reg_info;
3817 /* The following record the register info as found in the above
3818 variables when we find a match better than any we've seen before.
3819 This happens as we backtrack through the failure points, which in
3820 turn happens only if we have not yet matched the entire string. */
3821 unsigned best_regs_set = false;
3822 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
3823 const char **best_regstart, **best_regend;
3826 /* Logically, this is `best_regend[0]'. But we don't want to have to
3827 allocate space for that if we're not allocating space for anything
3828 else (see below). Also, we never need info about register 0 for
3829 any of the other register vectors, and it seems rather a kludge to
3830 treat `best_regend' differently than the rest. So we keep track of
3831 the end of the best match so far in a separate variable. We
3832 initialize this to NULL so that when we backtrack the first time
3833 and need to test it, it's not garbage. */
3834 const char *match_end = NULL;
3836 /* This helps SET_REGS_MATCHED avoid doing redundant work. */
3837 int set_regs_matched_done = 0;
3839 /* Used when we pop values we don't care about. */
3840 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
3841 const char **reg_dummy;
3842 register_info_type *reg_info_dummy;
3846 /* Counts the total number of registers pushed. */
3847 unsigned num_regs_pushed = 0;
3850 DEBUG_PRINT1 ("\n\nEntering re_match_2.\n");
3854 #ifdef MATCH_MAY_ALLOCATE
3855 /* Do not bother to initialize all the register variables if there are
3856 no groups in the pattern, as it takes a fair amount of time. If
3857 there are groups, we include space for register 0 (the whole
3858 pattern), even though we never use it, since it simplifies the
3859 array indexing. We should fix this. */
3862 regstart = REGEX_TALLOC (num_regs, const char *);
3863 regend = REGEX_TALLOC (num_regs, const char *);
3864 old_regstart = REGEX_TALLOC (num_regs, const char *);
3865 old_regend = REGEX_TALLOC (num_regs, const char *);
3866 best_regstart = REGEX_TALLOC (num_regs, const char *);
3867 best_regend = REGEX_TALLOC (num_regs, const char *);
3868 reg_info = REGEX_TALLOC (num_regs, register_info_type);
3869 reg_dummy = REGEX_TALLOC (num_regs, const char *);
3870 reg_info_dummy = REGEX_TALLOC (num_regs, register_info_type);
3872 if (!(regstart && regend && old_regstart && old_regend && reg_info
3873 && best_regstart && best_regend && reg_dummy && reg_info_dummy))
3881 /* We must initialize all our variables to NULL, so that
3882 `FREE_VARIABLES' doesn't try to free them. */
3883 regstart = regend = old_regstart = old_regend = best_regstart
3884 = best_regend = reg_dummy = NULL;
3885 reg_info = reg_info_dummy = (register_info_type *) NULL;
3887 #endif /* MATCH_MAY_ALLOCATE */
3889 /* The starting position is bogus. */
3890 if (pos < 0 || pos > size1 + size2)
3896 /* Initialize subexpression text positions to -1 to mark ones that no
3897 start_memory/stop_memory has been seen for. Also initialize the
3898 register information struct. */
3899 for (mcnt = 1; (unsigned) mcnt < num_regs; mcnt++)
3901 regstart[mcnt] = regend[mcnt]
3902 = old_regstart[mcnt] = old_regend[mcnt] = REG_UNSET_VALUE;
3904 REG_MATCH_NULL_STRING_P (reg_info[mcnt]) = MATCH_NULL_UNSET_VALUE;
3905 IS_ACTIVE (reg_info[mcnt]) = 0;
3906 MATCHED_SOMETHING (reg_info[mcnt]) = 0;
3907 EVER_MATCHED_SOMETHING (reg_info[mcnt]) = 0;
3910 /* We move `string1' into `string2' if the latter's empty -- but not if
3911 `string1' is null. */
3912 if (size2 == 0 && string1 != NULL)
3919 end1 = string1 + size1;
3920 end2 = string2 + size2;
3922 /* Compute where to stop matching, within the two strings. */
3925 end_match_1 = string1 + stop;
3926 end_match_2 = string2;
3931 end_match_2 = string2 + stop - size1;
3934 /* `p' scans through the pattern as `d' scans through the data.
3935 `dend' is the end of the input string that `d' points within. `d'
3936 is advanced into the following input string whenever necessary, but
3937 this happens before fetching; therefore, at the beginning of the
3938 loop, `d' can be pointing at the end of a string, but it cannot
3940 if (size1 > 0 && pos <= size1)
3947 d = string2 + pos - size1;
3951 DEBUG_PRINT1 ("The compiled pattern is:\n");
3952 DEBUG_PRINT_COMPILED_PATTERN (bufp, p, pend);
3953 DEBUG_PRINT1 ("The string to match is: `");
3954 DEBUG_PRINT_DOUBLE_STRING (d, string1, size1, string2, size2);
3955 DEBUG_PRINT1 ("'\n");
3957 /* This loops over pattern commands. It exits by returning from the
3958 function if the match is complete, or it drops through if the match
3959 fails at this starting point in the input data. */
3963 DEBUG_PRINT2 ("\n%p: ", p);
3965 DEBUG_PRINT2 ("\n0x%x: ", p);
3969 { /* End of pattern means we might have succeeded. */
3970 DEBUG_PRINT1 ("end of pattern ... ");
3972 /* If we haven't matched the entire string, and we want the
3973 longest match, try backtracking. */
3974 if (d != end_match_2)
3976 /* 1 if this match ends in the same string (string1 or string2)
3977 as the best previous match. */
3978 boolean same_str_p = (FIRST_STRING_P (match_end)
3979 == MATCHING_IN_FIRST_STRING);
3980 /* 1 if this match is the best seen so far. */
3981 boolean best_match_p;
3983 /* AIX compiler got confused when this was combined
3984 with the previous declaration. */
3986 best_match_p = d > match_end;
3988 best_match_p = !MATCHING_IN_FIRST_STRING;
3990 DEBUG_PRINT1 ("backtracking.\n");
3992 if (!FAIL_STACK_EMPTY ())
3993 { /* More failure points to try. */
3995 /* If exceeds best match so far, save it. */
3996 if (!best_regs_set || best_match_p)
3998 best_regs_set = true;
4001 DEBUG_PRINT1 ("\nSAVING match as best so far.\n");
4003 for (mcnt = 1; (unsigned) mcnt < num_regs; mcnt++)
4005 best_regstart[mcnt] = regstart[mcnt];
4006 best_regend[mcnt] = regend[mcnt];
4012 /* If no failure points, don't restore garbage. And if
4013 last match is real best match, don't restore second
4015 else if (best_regs_set && !best_match_p)
4018 /* Restore best match. It may happen that `dend ==
4019 end_match_1' while the restored d is in string2.
4020 For example, the pattern `x.*y.*z' against the
4021 strings `x-' and `y-z-', if the two strings are
4022 not consecutive in memory. */
4023 DEBUG_PRINT1 ("Restoring best registers.\n");
4026 dend = ((d >= string1 && d <= end1)
4027 ? end_match_1 : end_match_2);
4029 for (mcnt = 1; (unsigned) mcnt < num_regs; mcnt++)
4031 regstart[mcnt] = best_regstart[mcnt];
4032 regend[mcnt] = best_regend[mcnt];
4035 } /* d != end_match_2 */
4038 DEBUG_PRINT1 ("Accepting match.\n");
4040 /* If caller wants register contents data back, do it. */
4041 if (regs && !bufp->no_sub)
4043 /* Have the register data arrays been allocated? */
4044 if (bufp->regs_allocated == REGS_UNALLOCATED)
4045 { /* No. So allocate them with malloc. We need one
4046 extra element beyond `num_regs' for the `-1' marker
4048 regs->num_regs = MAX (RE_NREGS, num_regs + 1);
4049 regs->start = TALLOC (regs->num_regs, regoff_t);
4050 regs->end = TALLOC (regs->num_regs, regoff_t);
4051 if (regs->start == NULL || regs->end == NULL)
4056 bufp->regs_allocated = REGS_REALLOCATE;
4058 else if (bufp->regs_allocated == REGS_REALLOCATE)
4059 { /* Yes. If we need more elements than were already
4060 allocated, reallocate them. If we need fewer, just
4062 if (regs->num_regs < num_regs + 1)
4064 regs->num_regs = num_regs + 1;
4065 RETALLOC (regs->start, regs->num_regs, regoff_t);
4066 RETALLOC (regs->end, regs->num_regs, regoff_t);
4067 if (regs->start == NULL || regs->end == NULL)
4076 /* These braces fend off a "empty body in an else-statement"
4077 warning under GCC when assert expands to nothing. */
4078 assert (bufp->regs_allocated == REGS_FIXED);
4081 /* Convert the pointer data in `regstart' and `regend' to
4082 indices. Register zero has to be set differently,
4083 since we haven't kept track of any info for it. */
4084 if (regs->num_regs > 0)
4086 regs->start[0] = pos;
4087 regs->end[0] = (MATCHING_IN_FIRST_STRING
4088 ? ((regoff_t) (d - string1))
4089 : ((regoff_t) (d - string2 + size1)));
4092 /* Go through the first `min (num_regs, regs->num_regs)'
4093 registers, since that is all we initialized. */
4094 for (mcnt = 1; (unsigned) mcnt < MIN (num_regs, regs->num_regs);
4097 if (REG_UNSET (regstart[mcnt]) || REG_UNSET (regend[mcnt]))
4098 regs->start[mcnt] = regs->end[mcnt] = -1;
4102 = (regoff_t) POINTER_TO_OFFSET (regstart[mcnt]);
4104 = (regoff_t) POINTER_TO_OFFSET (regend[mcnt]);
4108 /* If the regs structure we return has more elements than
4109 were in the pattern, set the extra elements to -1. If
4110 we (re)allocated the registers, this is the case,
4111 because we always allocate enough to have at least one
4113 for (mcnt = num_regs; (unsigned) mcnt < regs->num_regs; mcnt++)
4114 regs->start[mcnt] = regs->end[mcnt] = -1;
4115 } /* regs && !bufp->no_sub */
4117 DEBUG_PRINT4 ("%u failure points pushed, %u popped (%u remain).\n",
4118 nfailure_points_pushed, nfailure_points_popped,
4119 nfailure_points_pushed - nfailure_points_popped);
4120 DEBUG_PRINT2 ("%u registers pushed.\n", num_regs_pushed);
4122 mcnt = d - pos - (MATCHING_IN_FIRST_STRING
4126 DEBUG_PRINT2 ("Returning %d from re_match_2.\n", mcnt);
4132 /* Otherwise match next pattern command. */
4133 switch (SWITCH_ENUM_CAST ((re_opcode_t) *p++))
4135 /* Ignore these. Used to ignore the n of succeed_n's which
4136 currently have n == 0. */
4138 DEBUG_PRINT1 ("EXECUTING no_op.\n");
4142 DEBUG_PRINT1 ("EXECUTING succeed.\n");
4145 /* Match the next n pattern characters exactly. The following
4146 byte in the pattern defines n, and the n bytes after that
4147 are the characters to match. */
4150 DEBUG_PRINT2 ("EXECUTING exactn %d.\n", mcnt);
4152 /* This is written out as an if-else so we don't waste time
4153 testing `translate' inside the loop. */
4159 if ((unsigned char) translate[(unsigned char) *d++]
4160 != (unsigned char) *p++)
4170 if (*d++ != (char) *p++) goto fail;
4174 SET_REGS_MATCHED ();
4178 /* Match any character except possibly a newline or a null. */
4180 DEBUG_PRINT1 ("EXECUTING anychar.\n");
4184 if ((!(bufp->syntax & RE_DOT_NEWLINE) && TRANSLATE (*d) == '\n')
4185 || (bufp->syntax & RE_DOT_NOT_NULL && TRANSLATE (*d) == '\000'))
4188 SET_REGS_MATCHED ();
4189 DEBUG_PRINT2 (" Matched `%d'.\n", *d);
4197 register unsigned char c;
4198 boolean not = (re_opcode_t) *(p - 1) == charset_not;
4200 DEBUG_PRINT2 ("EXECUTING charset%s.\n", not ? "_not" : "");
4203 c = TRANSLATE (*d); /* The character to match. */
4205 /* Cast to `unsigned' instead of `unsigned char' in case the
4206 bit list is a full 32 bytes long. */
4207 if (c < (unsigned) (*p * BYTEWIDTH)
4208 && p[1 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH)))
4213 if (!not) goto fail;
4215 SET_REGS_MATCHED ();
4221 /* The beginning of a group is represented by start_memory.
4222 The arguments are the register number in the next byte, and the
4223 number of groups inner to this one in the next. The text
4224 matched within the group is recorded (in the internal
4225 registers data structure) under the register number. */
4227 DEBUG_PRINT3 ("EXECUTING start_memory %d (%d):\n", *p, p[1]);
4229 /* Find out if this group can match the empty string. */
4230 p1 = p; /* To send to group_match_null_string_p. */
4232 if (REG_MATCH_NULL_STRING_P (reg_info[*p]) == MATCH_NULL_UNSET_VALUE)
4233 REG_MATCH_NULL_STRING_P (reg_info[*p])
4234 = group_match_null_string_p (&p1, pend, reg_info);
4236 /* Save the position in the string where we were the last time
4237 we were at this open-group operator in case the group is
4238 operated upon by a repetition operator, e.g., with `(a*)*b'
4239 against `ab'; then we want to ignore where we are now in
4240 the string in case this attempt to match fails. */
4241 old_regstart[*p] = REG_MATCH_NULL_STRING_P (reg_info[*p])
4242 ? REG_UNSET (regstart[*p]) ? d : regstart[*p]
4244 DEBUG_PRINT2 (" old_regstart: %d\n",
4245 POINTER_TO_OFFSET (old_regstart[*p]));
4248 DEBUG_PRINT2 (" regstart: %d\n", POINTER_TO_OFFSET (regstart[*p]));
4250 IS_ACTIVE (reg_info[*p]) = 1;
4251 MATCHED_SOMETHING (reg_info[*p]) = 0;
4253 /* Clear this whenever we change the register activity status. */
4254 set_regs_matched_done = 0;
4256 /* This is the new highest active register. */
4257 highest_active_reg = *p;
4259 /* If nothing was active before, this is the new lowest active
4261 if (lowest_active_reg == NO_LOWEST_ACTIVE_REG)
4262 lowest_active_reg = *p;
4264 /* Move past the register number and inner group count. */
4266 just_past_start_mem = p;
4271 /* The stop_memory opcode represents the end of a group. Its
4272 arguments are the same as start_memory's: the register
4273 number, and the number of inner groups. */
4275 DEBUG_PRINT3 ("EXECUTING stop_memory %d (%d):\n", *p, p[1]);
4277 /* We need to save the string position the last time we were at
4278 this close-group operator in case the group is operated
4279 upon by a repetition operator, e.g., with `((a*)*(b*)*)*'
4280 against `aba'; then we want to ignore where we are now in
4281 the string in case this attempt to match fails. */
4282 old_regend[*p] = REG_MATCH_NULL_STRING_P (reg_info[*p])
4283 ? REG_UNSET (regend[*p]) ? d : regend[*p]
4285 DEBUG_PRINT2 (" old_regend: %d\n",
4286 POINTER_TO_OFFSET (old_regend[*p]));
4289 DEBUG_PRINT2 (" regend: %d\n", POINTER_TO_OFFSET (regend[*p]));
4291 /* This register isn't active anymore. */
4292 IS_ACTIVE (reg_info[*p]) = 0;
4294 /* Clear this whenever we change the register activity status. */
4295 set_regs_matched_done = 0;
4297 /* If this was the only register active, nothing is active
4299 if (lowest_active_reg == highest_active_reg)
4301 lowest_active_reg = NO_LOWEST_ACTIVE_REG;
4302 highest_active_reg = NO_HIGHEST_ACTIVE_REG;
4305 { /* We must scan for the new highest active register, since
4306 it isn't necessarily one less than now: consider
4307 (a(b)c(d(e)f)g). When group 3 ends, after the f), the
4308 new highest active register is 1. */
4309 unsigned char r = *p - 1;
4310 while (r > 0 && !IS_ACTIVE (reg_info[r]))
4313 /* If we end up at register zero, that means that we saved
4314 the registers as the result of an `on_failure_jump', not
4315 a `start_memory', and we jumped to past the innermost
4316 `stop_memory'. For example, in ((.)*) we save
4317 registers 1 and 2 as a result of the *, but when we pop
4318 back to the second ), we are at the stop_memory 1.
4319 Thus, nothing is active. */
4322 lowest_active_reg = NO_LOWEST_ACTIVE_REG;
4323 highest_active_reg = NO_HIGHEST_ACTIVE_REG;
4326 highest_active_reg = r;
4329 /* If just failed to match something this time around with a
4330 group that's operated on by a repetition operator, try to
4331 force exit from the ``loop'', and restore the register
4332 information for this group that we had before trying this
4334 if ((!MATCHED_SOMETHING (reg_info[*p])
4335 || just_past_start_mem == p - 1)
4338 boolean is_a_jump_n = false;
4342 switch ((re_opcode_t) *p1++)
4346 case pop_failure_jump:
4347 case maybe_pop_jump:
4349 case dummy_failure_jump:
4350 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4360 /* If the next operation is a jump backwards in the pattern
4361 to an on_failure_jump right before the start_memory
4362 corresponding to this stop_memory, exit from the loop
4363 by forcing a failure after pushing on the stack the
4364 on_failure_jump's jump in the pattern, and d. */
4365 if (mcnt < 0 && (re_opcode_t) *p1 == on_failure_jump
4366 && (re_opcode_t) p1[3] == start_memory && p1[4] == *p)
4368 /* If this group ever matched anything, then restore
4369 what its registers were before trying this last
4370 failed match, e.g., with `(a*)*b' against `ab' for
4371 regstart[1], and, e.g., with `((a*)*(b*)*)*'
4372 against `aba' for regend[3].
4374 Also restore the registers for inner groups for,
4375 e.g., `((a*)(b*))*' against `aba' (register 3 would
4376 otherwise get trashed). */
4378 if (EVER_MATCHED_SOMETHING (reg_info[*p]))
4382 EVER_MATCHED_SOMETHING (reg_info[*p]) = 0;
4384 /* Restore this and inner groups' (if any) registers. */
4385 for (r = *p; r < (unsigned) *p + (unsigned) *(p + 1);
4388 regstart[r] = old_regstart[r];
4390 /* xx why this test? */
4391 if (old_regend[r] >= regstart[r])
4392 regend[r] = old_regend[r];
4396 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4397 PUSH_FAILURE_POINT (p1 + mcnt, d, -2);
4403 /* Move past the register number and the inner group count. */
4408 /* \<digit> has been turned into a `duplicate' command which is
4409 followed by the numeric value of <digit> as the register number. */
4412 register const char *d2, *dend2;
4413 int regno = *p++; /* Get which register to match against. */
4414 DEBUG_PRINT2 ("EXECUTING duplicate %d.\n", regno);
4416 /* Can't back reference a group which we've never matched. */
4417 if (REG_UNSET (regstart[regno]) || REG_UNSET (regend[regno]))
4420 /* Where in input to try to start matching. */
4421 d2 = regstart[regno];
4423 /* Where to stop matching; if both the place to start and
4424 the place to stop matching are in the same string, then
4425 set to the place to stop, otherwise, for now have to use
4426 the end of the first string. */
4428 dend2 = ((FIRST_STRING_P (regstart[regno])
4429 == FIRST_STRING_P (regend[regno]))
4430 ? regend[regno] : end_match_1);
4433 /* If necessary, advance to next segment in register
4437 if (dend2 == end_match_2) break;
4438 if (dend2 == regend[regno]) break;
4440 /* End of string1 => advance to string2. */
4442 dend2 = regend[regno];
4444 /* At end of register contents => success */
4445 if (d2 == dend2) break;
4447 /* If necessary, advance to next segment in data. */
4450 /* How many characters left in this segment to match. */
4453 /* Want how many consecutive characters we can match in
4454 one shot, so, if necessary, adjust the count. */
4455 if (mcnt > dend2 - d2)
4458 /* Compare that many; failure if mismatch, else move
4461 ? bcmp_translate (d, d2, mcnt, translate)
4462 : bcmp (d, d2, mcnt))
4464 d += mcnt, d2 += mcnt;
4466 /* Do this because we've match some characters. */
4467 SET_REGS_MATCHED ();
4473 /* begline matches the empty string at the beginning of the string
4474 (unless `not_bol' is set in `bufp'), and, if
4475 `newline_anchor' is set, after newlines. */
4477 DEBUG_PRINT1 ("EXECUTING begline.\n");
4479 if (AT_STRINGS_BEG (d))
4481 if (!bufp->not_bol) break;
4483 else if (d[-1] == '\n' && bufp->newline_anchor)
4487 /* In all other cases, we fail. */
4491 /* endline is the dual of begline. */
4493 DEBUG_PRINT1 ("EXECUTING endline.\n");
4495 if (AT_STRINGS_END (d))
4497 if (!bufp->not_eol) break;
4500 /* We have to ``prefetch'' the next character. */
4501 else if ((d == end1 ? *string2 : *d) == '\n'
4502 && bufp->newline_anchor)
4509 /* Match at the very beginning of the data. */
4511 DEBUG_PRINT1 ("EXECUTING begbuf.\n");
4512 if (AT_STRINGS_BEG (d))
4517 /* Match at the very end of the data. */
4519 DEBUG_PRINT1 ("EXECUTING endbuf.\n");
4520 if (AT_STRINGS_END (d))
4525 /* on_failure_keep_string_jump is used to optimize `.*\n'. It
4526 pushes NULL as the value for the string on the stack. Then
4527 `pop_failure_point' will keep the current value for the
4528 string, instead of restoring it. To see why, consider
4529 matching `foo\nbar' against `.*\n'. The .* matches the foo;
4530 then the . fails against the \n. But the next thing we want
4531 to do is match the \n against the \n; if we restored the
4532 string value, we would be back at the foo.
4534 Because this is used only in specific cases, we don't need to
4535 check all the things that `on_failure_jump' does, to make
4536 sure the right things get saved on the stack. Hence we don't
4537 share its code. The only reason to push anything on the
4538 stack at all is that otherwise we would have to change
4539 `anychar's code to do something besides goto fail in this
4540 case; that seems worse than this. */
4541 case on_failure_keep_string_jump:
4542 DEBUG_PRINT1 ("EXECUTING on_failure_keep_string_jump");
4544 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4546 DEBUG_PRINT3 (" %d (to %p):\n", mcnt, p + mcnt);
4548 DEBUG_PRINT3 (" %d (to 0x%x):\n", mcnt, p + mcnt);
4551 PUSH_FAILURE_POINT (p + mcnt, NULL, -2);
4555 /* Uses of on_failure_jump:
4557 Each alternative starts with an on_failure_jump that points
4558 to the beginning of the next alternative. Each alternative
4559 except the last ends with a jump that in effect jumps past
4560 the rest of the alternatives. (They really jump to the
4561 ending jump of the following alternative, because tensioning
4562 these jumps is a hassle.)
4564 Repeats start with an on_failure_jump that points past both
4565 the repetition text and either the following jump or
4566 pop_failure_jump back to this on_failure_jump. */
4567 case on_failure_jump:
4569 DEBUG_PRINT1 ("EXECUTING on_failure_jump");
4571 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4573 DEBUG_PRINT3 (" %d (to %p)", mcnt, p + mcnt);
4575 DEBUG_PRINT3 (" %d (to 0x%x)", mcnt, p + mcnt);
4578 /* If this on_failure_jump comes right before a group (i.e.,
4579 the original * applied to a group), save the information
4580 for that group and all inner ones, so that if we fail back
4581 to this point, the group's information will be correct.
4582 For example, in \(a*\)*\1, we need the preceding group,
4583 and in \(zz\(a*\)b*\)\2, we need the inner group. */
4585 /* We can't use `p' to check ahead because we push
4586 a failure point to `p + mcnt' after we do this. */
4589 /* We need to skip no_op's before we look for the
4590 start_memory in case this on_failure_jump is happening as
4591 the result of a completed succeed_n, as in \(a\)\{1,3\}b\1
4593 while (p1 < pend && (re_opcode_t) *p1 == no_op)
4596 if (p1 < pend && (re_opcode_t) *p1 == start_memory)
4598 /* We have a new highest active register now. This will
4599 get reset at the start_memory we are about to get to,
4600 but we will have saved all the registers relevant to
4601 this repetition op, as described above. */
4602 highest_active_reg = *(p1 + 1) + *(p1 + 2);
4603 if (lowest_active_reg == NO_LOWEST_ACTIVE_REG)
4604 lowest_active_reg = *(p1 + 1);
4607 DEBUG_PRINT1 (":\n");
4608 PUSH_FAILURE_POINT (p + mcnt, d, -2);
4612 /* A smart repeat ends with `maybe_pop_jump'.
4613 We change it to either `pop_failure_jump' or `jump'. */
4614 case maybe_pop_jump:
4615 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4616 DEBUG_PRINT2 ("EXECUTING maybe_pop_jump %d.\n", mcnt);
4618 register unsigned char *p2 = p;
4620 /* Compare the beginning of the repeat with what in the
4621 pattern follows its end. If we can establish that there
4622 is nothing that they would both match, i.e., that we
4623 would have to backtrack because of (as in, e.g., `a*a')
4624 then we can change to pop_failure_jump, because we'll
4625 never have to backtrack.
4627 This is not true in the case of alternatives: in
4628 `(a|ab)*' we do need to backtrack to the `ab' alternative
4629 (e.g., if the string was `ab'). But instead of trying to
4630 detect that here, the alternative has put on a dummy
4631 failure point which is what we will end up popping. */
4633 /* Skip over open/close-group commands.
4634 If what follows this loop is a ...+ construct,
4635 look at what begins its body, since we will have to
4636 match at least one of that. */
4640 && ((re_opcode_t) *p2 == stop_memory
4641 || (re_opcode_t) *p2 == start_memory))
4643 else if (p2 + 6 < pend
4644 && (re_opcode_t) *p2 == dummy_failure_jump)
4651 /* p1[0] ... p1[2] are the `on_failure_jump' corresponding
4652 to the `maybe_finalize_jump' of this case. Examine what
4655 /* If we're at the end of the pattern, we can change. */
4658 /* Consider what happens when matching ":\(.*\)"
4659 against ":/". I don't really understand this code
4661 p[-3] = (unsigned char) pop_failure_jump;
4663 (" End of pattern: change to `pop_failure_jump'.\n");
4666 else if ((re_opcode_t) *p2 == exactn
4667 || (bufp->newline_anchor && (re_opcode_t) *p2 == endline))
4669 register unsigned char c
4670 = *p2 == (unsigned char) endline ? '\n' : p2[2];
4672 if ((re_opcode_t) p1[3] == exactn && p1[5] != c)
4674 p[-3] = (unsigned char) pop_failure_jump;
4675 DEBUG_PRINT3 (" %c != %c => pop_failure_jump.\n",
4679 else if ((re_opcode_t) p1[3] == charset
4680 || (re_opcode_t) p1[3] == charset_not)
4682 int not = (re_opcode_t) p1[3] == charset_not;
4684 if (c < (unsigned char) (p1[4] * BYTEWIDTH)
4685 && p1[5 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH)))
4688 /* `not' is equal to 1 if c would match, which means
4689 that we can't change to pop_failure_jump. */
4692 p[-3] = (unsigned char) pop_failure_jump;
4693 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
4697 else if ((re_opcode_t) *p2 == charset)
4700 register unsigned char c
4701 = *p2 == (unsigned char) endline ? '\n' : p2[2];
4705 if ((re_opcode_t) p1[3] == exactn
4706 && ! ((int) p2[1] * BYTEWIDTH > (int) p1[5]
4707 && (p2[2 + p1[5] / BYTEWIDTH]
4708 & (1 << (p1[5] % BYTEWIDTH)))))
4710 if ((re_opcode_t) p1[3] == exactn
4711 && ! ((int) p2[1] * BYTEWIDTH > (int) p1[4]
4712 && (p2[2 + p1[4] / BYTEWIDTH]
4713 & (1 << (p1[4] % BYTEWIDTH)))))
4716 p[-3] = (unsigned char) pop_failure_jump;
4717 DEBUG_PRINT3 (" %c != %c => pop_failure_jump.\n",
4721 else if ((re_opcode_t) p1[3] == charset_not)
4724 /* We win if the charset_not inside the loop
4725 lists every character listed in the charset after. */
4726 for (idx = 0; idx < (int) p2[1]; idx++)
4727 if (! (p2[2 + idx] == 0
4728 || (idx < (int) p1[4]
4729 && ((p2[2 + idx] & ~ p1[5 + idx]) == 0))))
4734 p[-3] = (unsigned char) pop_failure_jump;
4735 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
4738 else if ((re_opcode_t) p1[3] == charset)
4741 /* We win if the charset inside the loop
4742 has no overlap with the one after the loop. */
4744 idx < (int) p2[1] && idx < (int) p1[4];
4746 if ((p2[2 + idx] & p1[5 + idx]) != 0)
4749 if (idx == p2[1] || idx == p1[4])
4751 p[-3] = (unsigned char) pop_failure_jump;
4752 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
4757 p -= 2; /* Point at relative address again. */
4758 if ((re_opcode_t) p[-1] != pop_failure_jump)
4760 p[-1] = (unsigned char) jump;
4761 DEBUG_PRINT1 (" Match => jump.\n");
4762 goto unconditional_jump;
4764 /* Note fall through. */
4767 /* The end of a simple repeat has a pop_failure_jump back to
4768 its matching on_failure_jump, where the latter will push a
4769 failure point. The pop_failure_jump takes off failure
4770 points put on by this pop_failure_jump's matching
4771 on_failure_jump; we got through the pattern to here from the
4772 matching on_failure_jump, so didn't fail. */
4773 case pop_failure_jump:
4775 /* We need to pass separate storage for the lowest and
4776 highest registers, even though we don't care about the
4777 actual values. Otherwise, we will restore only one
4778 register from the stack, since lowest will == highest in
4779 `pop_failure_point'. */
4780 active_reg_t dummy_low_reg, dummy_high_reg;
4781 unsigned char *pdummy;
4784 DEBUG_PRINT1 ("EXECUTING pop_failure_jump.\n");
4785 POP_FAILURE_POINT (sdummy, pdummy,
4786 dummy_low_reg, dummy_high_reg,
4787 reg_dummy, reg_dummy, reg_info_dummy);
4789 /* Note fall through. */
4793 DEBUG_PRINT2 ("\n%p: ", p);
4795 DEBUG_PRINT2 ("\n0x%x: ", p);
4797 /* Note fall through. */
4799 /* Unconditionally jump (without popping any failure points). */
4801 EXTRACT_NUMBER_AND_INCR (mcnt, p); /* Get the amount to jump. */
4802 DEBUG_PRINT2 ("EXECUTING jump %d ", mcnt);
4803 p += mcnt; /* Do the jump. */
4805 DEBUG_PRINT2 ("(to %p).\n", p);
4807 DEBUG_PRINT2 ("(to 0x%x).\n", p);
4812 /* We need this opcode so we can detect where alternatives end
4813 in `group_match_null_string_p' et al. */
4815 DEBUG_PRINT1 ("EXECUTING jump_past_alt.\n");
4816 goto unconditional_jump;
4819 /* Normally, the on_failure_jump pushes a failure point, which
4820 then gets popped at pop_failure_jump. We will end up at
4821 pop_failure_jump, also, and with a pattern of, say, `a+', we
4822 are skipping over the on_failure_jump, so we have to push
4823 something meaningless for pop_failure_jump to pop. */
4824 case dummy_failure_jump:
4825 DEBUG_PRINT1 ("EXECUTING dummy_failure_jump.\n");
4826 /* It doesn't matter what we push for the string here. What
4827 the code at `fail' tests is the value for the pattern. */
4828 PUSH_FAILURE_POINT (0, 0, -2);
4829 goto unconditional_jump;
4832 /* At the end of an alternative, we need to push a dummy failure
4833 point in case we are followed by a `pop_failure_jump', because
4834 we don't want the failure point for the alternative to be
4835 popped. For example, matching `(a|ab)*' against `aab'
4836 requires that we match the `ab' alternative. */
4837 case push_dummy_failure:
4838 DEBUG_PRINT1 ("EXECUTING push_dummy_failure.\n");
4839 /* See comments just above at `dummy_failure_jump' about the
4841 PUSH_FAILURE_POINT (0, 0, -2);
4844 /* Have to succeed matching what follows at least n times.
4845 After that, handle like `on_failure_jump'. */
4847 EXTRACT_NUMBER (mcnt, p + 2);
4848 DEBUG_PRINT2 ("EXECUTING succeed_n %d.\n", mcnt);
4851 /* Originally, this is how many times we HAVE to succeed. */
4856 STORE_NUMBER_AND_INCR (p, mcnt);
4858 DEBUG_PRINT3 (" Setting %p to %d.\n", p - 2, mcnt);
4860 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p - 2, mcnt);
4866 DEBUG_PRINT2 (" Setting two bytes from %p to no_op.\n", p+2);
4868 DEBUG_PRINT2 (" Setting two bytes from 0x%x to no_op.\n", p+2);
4870 p[2] = (unsigned char) no_op;
4871 p[3] = (unsigned char) no_op;
4877 EXTRACT_NUMBER (mcnt, p + 2);
4878 DEBUG_PRINT2 ("EXECUTING jump_n %d.\n", mcnt);
4880 /* Originally, this is how many times we CAN jump. */
4884 STORE_NUMBER (p + 2, mcnt);
4886 DEBUG_PRINT3 (" Setting %p to %d.\n", p + 2, mcnt);
4888 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p + 2, mcnt);
4890 goto unconditional_jump;
4892 /* If don't have to jump any more, skip over the rest of command. */
4899 DEBUG_PRINT1 ("EXECUTING set_number_at.\n");
4901 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4903 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4905 DEBUG_PRINT3 (" Setting %p to %d.\n", p1, mcnt);
4907 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p1, mcnt);
4909 STORE_NUMBER (p1, mcnt);
4914 /* The DEC Alpha C compiler 3.x generates incorrect code for the
4915 test WORDCHAR_P (d - 1) != WORDCHAR_P (d) in the expansion of
4916 AT_WORD_BOUNDARY, so this code is disabled. Expanding the
4917 macro and introducing temporary variables works around the bug. */
4920 DEBUG_PRINT1 ("EXECUTING wordbound.\n");
4921 if (AT_WORD_BOUNDARY (d))
4926 DEBUG_PRINT1 ("EXECUTING notwordbound.\n");
4927 if (AT_WORD_BOUNDARY (d))
4933 boolean prevchar, thischar;
4935 DEBUG_PRINT1 ("EXECUTING wordbound.\n");
4936 if (AT_STRINGS_BEG (d) || AT_STRINGS_END (d))
4939 prevchar = WORDCHAR_P (d - 1);
4940 thischar = WORDCHAR_P (d);
4941 if (prevchar != thischar)
4948 boolean prevchar, thischar;
4950 DEBUG_PRINT1 ("EXECUTING notwordbound.\n");
4951 if (AT_STRINGS_BEG (d) || AT_STRINGS_END (d))
4954 prevchar = WORDCHAR_P (d - 1);
4955 thischar = WORDCHAR_P (d);
4956 if (prevchar != thischar)
4963 DEBUG_PRINT1 ("EXECUTING wordbeg.\n");
4964 if (WORDCHAR_P (d) && (AT_STRINGS_BEG (d) || !WORDCHAR_P (d - 1)))
4969 DEBUG_PRINT1 ("EXECUTING wordend.\n");
4970 if (!AT_STRINGS_BEG (d) && WORDCHAR_P (d - 1)
4971 && (!WORDCHAR_P (d) || AT_STRINGS_END (d)))
4977 DEBUG_PRINT1 ("EXECUTING before_dot.\n");
4978 if (PTR_CHAR_POS ((unsigned char *) d) >= point)
4983 DEBUG_PRINT1 ("EXECUTING at_dot.\n");
4984 if (PTR_CHAR_POS ((unsigned char *) d) != point)
4989 DEBUG_PRINT1 ("EXECUTING after_dot.\n");
4990 if (PTR_CHAR_POS ((unsigned char *) d) <= point)
4995 DEBUG_PRINT2 ("EXECUTING syntaxspec %d.\n", mcnt);
5000 DEBUG_PRINT1 ("EXECUTING Emacs wordchar.\n");
5004 /* Can't use *d++ here; SYNTAX may be an unsafe macro. */
5006 if (SYNTAX (d[-1]) != (enum syntaxcode) mcnt)
5008 SET_REGS_MATCHED ();
5012 DEBUG_PRINT2 ("EXECUTING notsyntaxspec %d.\n", mcnt);
5014 goto matchnotsyntax;
5017 DEBUG_PRINT1 ("EXECUTING Emacs notwordchar.\n");
5021 /* Can't use *d++ here; SYNTAX may be an unsafe macro. */
5023 if (SYNTAX (d[-1]) == (enum syntaxcode) mcnt)
5025 SET_REGS_MATCHED ();
5028 #else /* not emacs */
5030 DEBUG_PRINT1 ("EXECUTING non-Emacs wordchar.\n");
5032 if (!WORDCHAR_P (d))
5034 SET_REGS_MATCHED ();
5039 DEBUG_PRINT1 ("EXECUTING non-Emacs notwordchar.\n");
5043 SET_REGS_MATCHED ();
5046 #endif /* not emacs */
5051 continue; /* Successfully executed one pattern command; keep going. */
5054 /* We goto here if a matching operation fails. */
5056 if (!FAIL_STACK_EMPTY ())
5057 { /* A restart point is known. Restore to that state. */
5058 DEBUG_PRINT1 ("\nFAIL:\n");
5059 POP_FAILURE_POINT (d, p,
5060 lowest_active_reg, highest_active_reg,
5061 regstart, regend, reg_info);
5063 /* If this failure point is a dummy, try the next one. */
5067 /* If we failed to the end of the pattern, don't examine *p. */
5071 boolean is_a_jump_n = false;
5073 /* If failed to a backwards jump that's part of a repetition
5074 loop, need to pop this failure point and use the next one. */
5075 switch ((re_opcode_t) *p)
5079 case maybe_pop_jump:
5080 case pop_failure_jump:
5083 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
5086 if ((is_a_jump_n && (re_opcode_t) *p1 == succeed_n)
5088 && (re_opcode_t) *p1 == on_failure_jump))
5096 if (d >= string1 && d <= end1)
5100 break; /* Matching at this starting point really fails. */
5104 goto restore_best_regs;
5108 return -1; /* Failure to match. */
5111 /* Subroutine definitions for re_match_2. */
5114 /* We are passed P pointing to a register number after a start_memory.
5116 Return true if the pattern up to the corresponding stop_memory can
5117 match the empty string, and false otherwise.
5119 If we find the matching stop_memory, sets P to point to one past its number.
5120 Otherwise, sets P to an undefined byte less than or equal to END.
5122 We don't handle duplicates properly (yet). */
5125 group_match_null_string_p (p, end, reg_info)
5126 unsigned char **p, *end;
5127 register_info_type *reg_info;
5130 /* Point to after the args to the start_memory. */
5131 unsigned char *p1 = *p + 2;
5135 /* Skip over opcodes that can match nothing, and return true or
5136 false, as appropriate, when we get to one that can't, or to the
5137 matching stop_memory. */
5139 switch ((re_opcode_t) *p1)
5141 /* Could be either a loop or a series of alternatives. */
5142 case on_failure_jump:
5144 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
5146 /* If the next operation is not a jump backwards in the
5151 /* Go through the on_failure_jumps of the alternatives,
5152 seeing if any of the alternatives cannot match nothing.
5153 The last alternative starts with only a jump,
5154 whereas the rest start with on_failure_jump and end
5155 with a jump, e.g., here is the pattern for `a|b|c':
5157 /on_failure_jump/0/6/exactn/1/a/jump_past_alt/0/6
5158 /on_failure_jump/0/6/exactn/1/b/jump_past_alt/0/3
5161 So, we have to first go through the first (n-1)
5162 alternatives and then deal with the last one separately. */
5165 /* Deal with the first (n-1) alternatives, which start
5166 with an on_failure_jump (see above) that jumps to right
5167 past a jump_past_alt. */
5169 while ((re_opcode_t) p1[mcnt-3] == jump_past_alt)
5171 /* `mcnt' holds how many bytes long the alternative
5172 is, including the ending `jump_past_alt' and
5175 if (!alt_match_null_string_p (p1, p1 + mcnt - 3,
5179 /* Move to right after this alternative, including the
5183 /* Break if it's the beginning of an n-th alternative
5184 that doesn't begin with an on_failure_jump. */
5185 if ((re_opcode_t) *p1 != on_failure_jump)
5188 /* Still have to check that it's not an n-th
5189 alternative that starts with an on_failure_jump. */
5191 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
5192 if ((re_opcode_t) p1[mcnt-3] != jump_past_alt)
5194 /* Get to the beginning of the n-th alternative. */
5200 /* Deal with the last alternative: go back and get number
5201 of the `jump_past_alt' just before it. `mcnt' contains
5202 the length of the alternative. */
5203 EXTRACT_NUMBER (mcnt, p1 - 2);
5205 if (!alt_match_null_string_p (p1, p1 + mcnt, reg_info))
5208 p1 += mcnt; /* Get past the n-th alternative. */
5214 assert (p1[1] == **p);
5220 if (!common_op_match_null_string_p (&p1, end, reg_info))
5223 } /* while p1 < end */
5226 } /* group_match_null_string_p */
5229 /* Similar to group_match_null_string_p, but doesn't deal with alternatives:
5230 It expects P to be the first byte of a single alternative and END one
5231 byte past the last. The alternative can contain groups. */
5234 alt_match_null_string_p (p, end, reg_info)
5235 unsigned char *p, *end;
5236 register_info_type *reg_info;
5239 unsigned char *p1 = p;
5243 /* Skip over opcodes that can match nothing, and break when we get
5244 to one that can't. */
5246 switch ((re_opcode_t) *p1)
5249 case on_failure_jump:
5251 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
5256 if (!common_op_match_null_string_p (&p1, end, reg_info))
5259 } /* while p1 < end */
5262 } /* alt_match_null_string_p */
5265 /* Deals with the ops common to group_match_null_string_p and
5266 alt_match_null_string_p.
5268 Sets P to one after the op and its arguments, if any. */
5271 common_op_match_null_string_p (p, end, reg_info)
5272 unsigned char **p, *end;
5273 register_info_type *reg_info;
5278 unsigned char *p1 = *p;
5280 switch ((re_opcode_t) *p1++)
5300 assert (reg_no > 0 && reg_no <= MAX_REGNUM);
5301 ret = group_match_null_string_p (&p1, end, reg_info);
5303 /* Have to set this here in case we're checking a group which
5304 contains a group and a back reference to it. */
5306 if (REG_MATCH_NULL_STRING_P (reg_info[reg_no]) == MATCH_NULL_UNSET_VALUE)
5307 REG_MATCH_NULL_STRING_P (reg_info[reg_no]) = ret;
5313 /* If this is an optimized succeed_n for zero times, make the jump. */
5315 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
5323 /* Get to the number of times to succeed. */
5325 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
5330 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
5338 if (!REG_MATCH_NULL_STRING_P (reg_info[*p1]))
5346 /* All other opcodes mean we cannot match the empty string. */
5352 } /* common_op_match_null_string_p */
5355 /* Return zero if TRANSLATE[S1] and TRANSLATE[S2] are identical for LEN
5356 bytes; nonzero otherwise. */
5359 bcmp_translate (s1, s2, len, translate)
5360 const char *s1, *s2;
5362 RE_TRANSLATE_TYPE translate;
5364 register const unsigned char *p1 = (const unsigned char *) s1;
5365 register const unsigned char *p2 = (const unsigned char *) s2;
5368 if (translate[*p1++] != translate[*p2++]) return 1;
5374 /* Entry points for GNU code. */
5376 /* re_compile_pattern is the GNU regular expression compiler: it
5377 compiles PATTERN (of length SIZE) and puts the result in BUFP.
5378 Returns 0 if the pattern was valid, otherwise an error string.
5380 Assumes the `allocated' (and perhaps `buffer') and `translate' fields
5381 are set in BUFP on entry.
5383 We call regex_compile to do the actual compilation. */
5386 re_compile_pattern (pattern, length, bufp)
5387 const char *pattern;
5389 struct re_pattern_buffer *bufp;
5393 /* GNU code is written to assume at least RE_NREGS registers will be set
5394 (and at least one extra will be -1). */
5395 bufp->regs_allocated = REGS_UNALLOCATED;
5397 /* And GNU code determines whether or not to get register information
5398 by passing null for the REGS argument to re_match, etc., not by
5402 /* Match anchors at newline. */
5403 bufp->newline_anchor = 1;
5405 ret = regex_compile (pattern, length, re_syntax_options, bufp);
5409 return gettext (re_error_msgid[(int) ret]);
5412 /* Entry points compatible with 4.2 BSD regex library. We don't define
5413 them unless specifically requested. */
5415 #if defined (_REGEX_RE_COMP) || defined (_LIBC)
5417 /* BSD has one and only one pattern buffer. */
5418 static struct re_pattern_buffer re_comp_buf;
5422 /* Make these definitions weak in libc, so POSIX programs can redefine
5423 these names if they don't use our functions, and still use
5424 regcomp/regexec below without link errors. */
5434 if (!re_comp_buf.buffer)
5435 return gettext ("No previous regular expression");
5439 if (!re_comp_buf.buffer)
5441 re_comp_buf.buffer = (unsigned char *) malloc (200);
5442 if (re_comp_buf.buffer == NULL)
5443 return gettext (re_error_msgid[(int) REG_ESPACE]);
5444 re_comp_buf.allocated = 200;
5446 re_comp_buf.fastmap = (char *) malloc (1 << BYTEWIDTH);
5447 if (re_comp_buf.fastmap == NULL)
5448 return gettext (re_error_msgid[(int) REG_ESPACE]);
5451 /* Since `re_exec' always passes NULL for the `regs' argument, we
5452 don't need to initialize the pattern buffer fields which affect it. */
5454 /* Match anchors at newlines. */
5455 re_comp_buf.newline_anchor = 1;
5457 ret = regex_compile (s, strlen (s), re_syntax_options, &re_comp_buf);
5462 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
5463 return (char *) gettext (re_error_msgid[(int) ret]);
5474 const int len = strlen (s);
5476 0 <= re_search (&re_comp_buf, s, len, 0, len, (struct re_registers *) 0);
5479 #endif /* _REGEX_RE_COMP */
5481 /* POSIX.2 functions. Don't define these for Emacs. */
5485 /* regcomp takes a regular expression as a string and compiles it.
5487 PREG is a regex_t *. We do not expect any fields to be initialized,
5488 since POSIX says we shouldn't. Thus, we set
5490 `buffer' to the compiled pattern;
5491 `used' to the length of the compiled pattern;
5492 `syntax' to RE_SYNTAX_POSIX_EXTENDED if the
5493 REG_EXTENDED bit in CFLAGS is set; otherwise, to
5494 RE_SYNTAX_POSIX_BASIC;
5495 `newline_anchor' to REG_NEWLINE being set in CFLAGS;
5496 `fastmap' and `fastmap_accurate' to zero;
5497 `re_nsub' to the number of subexpressions in PATTERN.
5499 PATTERN is the address of the pattern string.
5501 CFLAGS is a series of bits which affect compilation.
5503 If REG_EXTENDED is set, we use POSIX extended syntax; otherwise, we
5504 use POSIX basic syntax.
5506 If REG_NEWLINE is set, then . and [^...] don't match newline.
5507 Also, regexec will try a match beginning after every newline.
5509 If REG_ICASE is set, then we considers upper- and lowercase
5510 versions of letters to be equivalent when matching.
5512 If REG_NOSUB is set, then when PREG is passed to regexec, that
5513 routine will report only success or failure, and nothing about the
5516 It returns 0 if it succeeds, nonzero if it doesn't. (See regex.h for
5517 the return codes and their meanings.) */
5520 regcomp (preg, pattern, cflags)
5522 const char *pattern;
5527 = (cflags & REG_EXTENDED) ?
5528 RE_SYNTAX_POSIX_EXTENDED : RE_SYNTAX_POSIX_BASIC;
5530 /* regex_compile will allocate the space for the compiled pattern. */
5532 preg->allocated = 0;
5535 /* Don't bother to use a fastmap when searching. This simplifies the
5536 REG_NEWLINE case: if we used a fastmap, we'd have to put all the
5537 characters after newlines into the fastmap. This way, we just try
5541 if (cflags & REG_ICASE)
5546 = (RE_TRANSLATE_TYPE) malloc (CHAR_SET_SIZE
5547 * sizeof (*(RE_TRANSLATE_TYPE)0));
5548 if (preg->translate == NULL)
5549 return (int) REG_ESPACE;
5551 /* Map uppercase characters to corresponding lowercase ones. */
5552 for (i = 0; i < CHAR_SET_SIZE; i++)
5553 preg->translate[i] = ISUPPER (i) ? tolower (i) : i;
5556 preg->translate = NULL;
5558 /* If REG_NEWLINE is set, newlines are treated differently. */
5559 if (cflags & REG_NEWLINE)
5560 { /* REG_NEWLINE implies neither . nor [^...] match newline. */
5561 syntax &= ~RE_DOT_NEWLINE;
5562 syntax |= RE_HAT_LISTS_NOT_NEWLINE;
5563 /* It also changes the matching behavior. */
5564 preg->newline_anchor = 1;
5567 preg->newline_anchor = 0;
5569 preg->no_sub = !!(cflags & REG_NOSUB);
5571 /* POSIX says a null character in the pattern terminates it, so we
5572 can use strlen here in compiling the pattern. */
5573 ret = regex_compile (pattern, strlen (pattern), syntax, preg);
5575 /* POSIX doesn't distinguish between an unmatched open-group and an
5576 unmatched close-group: both are REG_EPAREN. */
5577 if (ret == REG_ERPAREN) ret = REG_EPAREN;
5583 /* regexec searches for a given pattern, specified by PREG, in the
5586 If NMATCH is zero or REG_NOSUB was set in the cflags argument to
5587 `regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at
5588 least NMATCH elements, and we set them to the offsets of the
5589 corresponding matched substrings.
5591 EFLAGS specifies `execution flags' which affect matching: if
5592 REG_NOTBOL is set, then ^ does not match at the beginning of the
5593 string; if REG_NOTEOL is set, then $ does not match at the end.
5595 We return 0 if we find a match and REG_NOMATCH if not. */
5598 regexec (preg, string, nmatch, pmatch, eflags)
5599 const regex_t *preg;
5602 regmatch_t pmatch[];
5606 struct re_registers regs;
5607 regex_t private_preg;
5608 int len = strlen (string);
5609 boolean want_reg_info = !preg->no_sub && nmatch > 0;
5611 private_preg = *preg;
5613 private_preg.not_bol = !!(eflags & REG_NOTBOL);
5614 private_preg.not_eol = !!(eflags & REG_NOTEOL);
5616 /* The user has told us exactly how many registers to return
5617 information about, via `nmatch'. We have to pass that on to the
5618 matching routines. */
5619 private_preg.regs_allocated = REGS_FIXED;
5623 regs.num_regs = nmatch;
5624 regs.start = TALLOC (nmatch, regoff_t);
5625 regs.end = TALLOC (nmatch, regoff_t);
5626 if (regs.start == NULL || regs.end == NULL)
5627 return (int) REG_NOMATCH;
5630 /* Perform the searching operation. */
5631 ret = re_search (&private_preg, string, len,
5632 /* start: */ 0, /* range: */ len,
5633 want_reg_info ? ®s : (struct re_registers *) 0);
5635 /* Copy the register information to the POSIX structure. */
5642 for (r = 0; r < nmatch; r++)
5644 pmatch[r].rm_so = regs.start[r];
5645 pmatch[r].rm_eo = regs.end[r];
5649 /* If we needed the temporary register info, free the space now. */
5654 /* We want zero return to mean success, unlike `re_search'. */
5655 return ret >= 0 ? (int) REG_NOERROR : (int) REG_NOMATCH;
5659 /* Returns a message corresponding to an error code, ERRCODE, returned
5660 from either regcomp or regexec. We don't use PREG here. */
5663 regerror (errcode, preg, errbuf, errbuf_size)
5665 const regex_t *preg;
5673 || errcode >= (int) (sizeof (re_error_msgid)
5674 / sizeof (re_error_msgid[0])))
5675 /* Only error codes returned by the rest of the code should be passed
5676 to this routine. If we are given anything else, or if other regex
5677 code generates an invalid error code, then the program has a bug.
5678 Dump core so we can fix it. */
5681 msg = gettext (re_error_msgid[errcode]);
5683 msg_size = strlen (msg) + 1; /* Includes the null. */
5685 if (errbuf_size != 0)
5687 if (msg_size > errbuf_size)
5689 strncpy (errbuf, msg, errbuf_size - 1);
5690 errbuf[errbuf_size - 1] = 0;
5693 strcpy (errbuf, msg);
5700 /* Free dynamically allocated space used by PREG. */
5706 if (preg->buffer != NULL)
5707 free (preg->buffer);
5708 preg->buffer = NULL;
5710 preg->allocated = 0;
5713 if (preg->fastmap != NULL)
5714 free (preg->fastmap);
5715 preg->fastmap = NULL;
5716 preg->fastmap_accurate = 0;
5718 if (preg->translate != NULL)
5719 free (preg->translate);
5720 preg->translate = NULL;
5723 #endif /* not emacs */