1 /* Extended regular expression matching and search library,
3 (Implements POSIX draft P1003.2/D11.2, except for some of the
4 internationalization features.)
5 Copyright (C) 1993, 94, 95, 96, 97, 98 Free Software Foundation, Inc.
7 NOTE: The canonical source of this file is maintained with the GNU C Library.
8 Bugs can be reported to bug-glibc@prep.ai.mit.edu.
10 This program is free software; you can redistribute it and/or modify it
11 under the terms of the GNU General Public License as published by the
12 Free Software Foundation; either version 2, or (at your option) any
15 This program is distributed in the hope that it will be useful,
16 but WITHOUT ANY WARRANTY; without even the implied warranty of
17 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
18 GNU General Public License for more details.
20 You should have received a copy of the GNU General Public License
21 along with this program; if not, write to the Free Software
22 Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307,
25 /* AIX requires this to be the first thing in the file. */
26 #if defined _AIX && !defined REGEX_MALLOC
38 # if defined __GNUC__ || (defined __STDC__ && __STDC__)
39 # define PARAMS(args) args
41 # define PARAMS(args) ()
43 #endif /* Not PARAMS. */
45 #if defined STDC_HEADERS && !defined emacs
48 /* We need this for `regex.h', and perhaps for the Emacs include files. */
49 # include <sys/types.h>
52 #define WIDE_CHAR_SUPPORT \
53 defined _LIBC || (HAVE_WCTYPE_H && HAVE_WCHAR_H && HAVE_BTOWC)
55 /* For platform which support the ISO C amendement 1 functionality we
56 support user defined character classes. */
62 /* This is for other GNU distributions with internationalized messages. */
63 #if HAVE_LIBINTL_H || defined _LIBC
66 # define gettext(msgid) (msgid)
70 /* This define is so xgettext can find the internationalizable
72 # define gettext_noop(String) String
75 /* The `emacs' switch turns on certain matching commands
76 that make sense only in Emacs. */
85 /* If we are not linking with Emacs proper,
86 we can't use the relocating allocator
87 even if config.h says that we can. */
90 # if defined STDC_HEADERS || defined _LIBC
97 /* When used in Emacs's lib-src, we need to get bzero and bcopy somehow.
98 If nothing else has been done, use the method below. */
99 # ifdef INHIBIT_STRING_HEADER
100 # if !(defined HAVE_BZERO && defined HAVE_BCOPY)
101 # if !defined bzero && !defined bcopy
102 # undef INHIBIT_STRING_HEADER
107 /* This is the normal way of making sure we have a bcopy and a bzero.
108 This is used in most programs--a few other programs avoid this
109 by defining INHIBIT_STRING_HEADER. */
110 # ifndef INHIBIT_STRING_HEADER
111 # if defined HAVE_STRING_H || defined STDC_HEADERS || defined _LIBC
113 # if !defined bzero && !defined _LIBC
114 # define bzero(s, n) (memset (s, '\0', n), (s))
117 # include <strings.h>
119 # define memcmp(s1, s2, n) bcmp (s1, s2, n)
122 # define memcpy(d, s, n) (bcopy (s, d, n), (d))
127 /* Define the syntax stuff for \<, \>, etc. */
129 /* This must be nonzero for the wordchar and notwordchar pattern
130 commands in re_match_2. */
135 # ifdef SWITCH_ENUM_BUG
136 # define SWITCH_ENUM_CAST(x) ((int)(x))
138 # define SWITCH_ENUM_CAST(x) (x)
141 /* How many characters in the character set. */
142 # define CHAR_SET_SIZE 256
146 extern char *re_syntax_table;
148 # else /* not SYNTAX_TABLE */
150 static char re_syntax_table[CHAR_SET_SIZE];
161 bzero (re_syntax_table, sizeof re_syntax_table);
163 for (c = 'a'; c <= 'z'; c++)
164 re_syntax_table[c] = Sword;
166 for (c = 'A'; c <= 'Z'; c++)
167 re_syntax_table[c] = Sword;
169 for (c = '0'; c <= '9'; c++)
170 re_syntax_table[c] = Sword;
172 re_syntax_table['_'] = Sword;
177 # endif /* not SYNTAX_TABLE */
179 # define SYNTAX(c) re_syntax_table[c]
181 #endif /* not emacs */
183 /* Get the interface, including the syntax bits. */
186 /* isalpha etc. are used for the character classes. */
189 /* Jim Meyering writes:
191 "... Some ctype macros are valid only for character codes that
192 isascii says are ASCII (SGI's IRIX-4.0.5 is one such system --when
193 using /bin/cc or gcc but without giving an ansi option). So, all
194 ctype uses should be through macros like ISPRINT... If
195 STDC_HEADERS is defined, then autoconf has verified that the ctype
196 macros don't need to be guarded with references to isascii. ...
197 Defining isascii to 1 should let any compiler worth its salt
198 eliminate the && through constant folding." */
201 #if defined STDC_HEADERS || (!defined isascii && !defined HAVE_ISASCII)
202 # define ISASCII(c) 1
204 # define ISASCII(c) isascii(c)
208 # define ISBLANK(c) (ISASCII (c) && isblank (c))
210 # define ISBLANK(c) ((c) == ' ' || (c) == '\t')
213 # define ISGRAPH(c) (ISASCII (c) && isgraph (c))
215 # define ISGRAPH(c) (ISASCII (c) && isprint (c) && !isspace (c))
219 #define ISPRINT(c) (ISASCII (c) && isprint (c))
220 #define ISDIGIT(c) (ISASCII (c) && isdigit (c))
221 #define ISALNUM(c) (ISASCII (c) && isalnum (c))
222 #define ISALPHA(c) (ISASCII (c) && isalpha (c))
223 #define ISCNTRL(c) (ISASCII (c) && iscntrl (c))
224 #define ISLOWER(c) (ISASCII (c) && islower (c))
225 #define ISPUNCT(c) (ISASCII (c) && ispunct (c))
226 #define ISSPACE(c) (ISASCII (c) && isspace (c))
227 #define ISUPPER(c) (ISASCII (c) && isupper (c))
228 #define ISXDIGIT(c) (ISASCII (c) && isxdigit (c))
231 # define NULL (void *)0
234 /* We remove any previous definition of `SIGN_EXTEND_CHAR',
235 since ours (we hope) works properly with all combinations of
236 machines, compilers, `char' and `unsigned char' argument types.
237 (Per Bothner suggested the basic approach.) */
238 #undef SIGN_EXTEND_CHAR
240 # define SIGN_EXTEND_CHAR(c) ((signed char) (c))
241 #else /* not __STDC__ */
242 /* As in Harbison and Steele. */
243 # define SIGN_EXTEND_CHAR(c) ((((unsigned char) (c)) ^ 128) - 128)
246 /* Should we use malloc or alloca? If REGEX_MALLOC is not defined, we
247 use `alloca' instead of `malloc'. This is because using malloc in
248 re_search* or re_match* could cause memory leaks when C-g is used in
249 Emacs; also, malloc is slower and causes storage fragmentation. On
250 the other hand, malloc is more portable, and easier to debug.
252 Because we sometimes use alloca, some routines have to be macros,
253 not functions -- `alloca'-allocated space disappears at the end of the
254 function it is called in. */
258 # define REGEX_ALLOCATE malloc
259 # define REGEX_REALLOCATE(source, osize, nsize) realloc (source, nsize)
260 # define REGEX_FREE free
262 #else /* not REGEX_MALLOC */
264 /* Emacs already defines alloca, sometimes. */
267 /* Make alloca work the best possible way. */
269 # define alloca __builtin_alloca
270 # else /* not __GNUC__ */
273 # endif /* HAVE_ALLOCA_H */
274 # endif /* not __GNUC__ */
276 # endif /* not alloca */
278 # define REGEX_ALLOCATE alloca
280 /* Assumes a `char *destination' variable. */
281 # define REGEX_REALLOCATE(source, osize, nsize) \
282 (destination = (char *) alloca (nsize), \
283 memcpy (destination, source, osize))
285 /* No need to do anything to free, after alloca. */
286 # define REGEX_FREE(arg) ((void)0) /* Do nothing! But inhibit gcc warning. */
288 #endif /* not REGEX_MALLOC */
290 /* Define how to allocate the failure stack. */
292 #if defined REL_ALLOC && defined REGEX_MALLOC
294 # define REGEX_ALLOCATE_STACK(size) \
295 r_alloc (&failure_stack_ptr, (size))
296 # define REGEX_REALLOCATE_STACK(source, osize, nsize) \
297 r_re_alloc (&failure_stack_ptr, (nsize))
298 # define REGEX_FREE_STACK(ptr) \
299 r_alloc_free (&failure_stack_ptr)
301 #else /* not using relocating allocator */
305 # define REGEX_ALLOCATE_STACK malloc
306 # define REGEX_REALLOCATE_STACK(source, osize, nsize) realloc (source, nsize)
307 # define REGEX_FREE_STACK free
309 # else /* not REGEX_MALLOC */
311 # define REGEX_ALLOCATE_STACK alloca
313 # define REGEX_REALLOCATE_STACK(source, osize, nsize) \
314 REGEX_REALLOCATE (source, osize, nsize)
315 /* No need to explicitly free anything. */
316 # define REGEX_FREE_STACK(arg)
318 # endif /* not REGEX_MALLOC */
319 #endif /* not using relocating allocator */
322 /* True if `size1' is non-NULL and PTR is pointing anywhere inside
323 `string1' or just past its end. This works if PTR is NULL, which is
325 #define FIRST_STRING_P(ptr) \
326 (size1 && string1 <= (ptr) && (ptr) <= string1 + size1)
328 /* (Re)Allocate N items of type T using malloc, or fail. */
329 #define TALLOC(n, t) ((t *) malloc ((n) * sizeof (t)))
330 #define RETALLOC(addr, n, t) ((addr) = (t *) realloc (addr, (n) * sizeof (t)))
331 #define RETALLOC_IF(addr, n, t) \
332 if (addr) RETALLOC((addr), (n), t); else (addr) = TALLOC ((n), t)
333 #define REGEX_TALLOC(n, t) ((t *) REGEX_ALLOCATE ((n) * sizeof (t)))
335 #define BYTEWIDTH 8 /* In bits. */
337 #define STREQ(s1, s2) ((strcmp (s1, s2) == 0))
341 #define MAX(a, b) ((a) > (b) ? (a) : (b))
342 #define MIN(a, b) ((a) < (b) ? (a) : (b))
344 typedef char boolean;
348 static int re_match_2_internal PARAMS ((struct re_pattern_buffer *bufp,
349 const char *string1, int size1,
350 const char *string2, int size2,
352 struct re_registers *regs,
355 /* These are the command codes that appear in compiled regular
356 expressions. Some opcodes are followed by argument bytes. A
357 command code can specify any interpretation whatsoever for its
358 arguments. Zero bytes may appear in the compiled regular expression. */
364 /* Succeed right away--no more backtracking. */
367 /* Followed by one byte giving n, then by n literal bytes. */
370 /* Matches any (more or less) character. */
373 /* Matches any one char belonging to specified set. First
374 following byte is number of bitmap bytes. Then come bytes
375 for a bitmap saying which chars are in. Bits in each byte
376 are ordered low-bit-first. A character is in the set if its
377 bit is 1. A character too large to have a bit in the map is
378 automatically not in the set. */
381 /* Same parameters as charset, but match any character that is
382 not one of those specified. */
385 /* Start remembering the text that is matched, for storing in a
386 register. Followed by one byte with the register number, in
387 the range 0 to one less than the pattern buffer's re_nsub
388 field. Then followed by one byte with the number of groups
389 inner to this one. (This last has to be part of the
390 start_memory only because we need it in the on_failure_jump
394 /* Stop remembering the text that is matched and store it in a
395 memory register. Followed by one byte with the register
396 number, in the range 0 to one less than `re_nsub' in the
397 pattern buffer, and one byte with the number of inner groups,
398 just like `start_memory'. (We need the number of inner
399 groups here because we don't have any easy way of finding the
400 corresponding start_memory when we're at a stop_memory.) */
403 /* Match a duplicate of something remembered. Followed by one
404 byte containing the register number. */
407 /* Fail unless at beginning of line. */
410 /* Fail unless at end of line. */
413 /* Succeeds if at beginning of buffer (if emacs) or at beginning
414 of string to be matched (if not). */
417 /* Analogously, for end of buffer/string. */
420 /* Followed by two byte relative address to which to jump. */
423 /* Same as jump, but marks the end of an alternative. */
426 /* Followed by two-byte relative address of place to resume at
427 in case of failure. */
430 /* Like on_failure_jump, but pushes a placeholder instead of the
431 current string position when executed. */
432 on_failure_keep_string_jump,
434 /* Throw away latest failure point and then jump to following
435 two-byte relative address. */
438 /* Change to pop_failure_jump if know won't have to backtrack to
439 match; otherwise change to jump. This is used to jump
440 back to the beginning of a repeat. If what follows this jump
441 clearly won't match what the repeat does, such that we can be
442 sure that there is no use backtracking out of repetitions
443 already matched, then we change it to a pop_failure_jump.
444 Followed by two-byte address. */
447 /* Jump to following two-byte address, and push a dummy failure
448 point. This failure point will be thrown away if an attempt
449 is made to use it for a failure. A `+' construct makes this
450 before the first repeat. Also used as an intermediary kind
451 of jump when compiling an alternative. */
454 /* Push a dummy failure point and continue. Used at the end of
458 /* Followed by two-byte relative address and two-byte number n.
459 After matching N times, jump to the address upon failure. */
462 /* Followed by two-byte relative address, and two-byte number n.
463 Jump to the address N times, then fail. */
466 /* Set the following two-byte relative address to the
467 subsequent two-byte number. The address *includes* the two
471 wordchar, /* Matches any word-constituent character. */
472 notwordchar, /* Matches any char that is not a word-constituent. */
474 wordbeg, /* Succeeds if at word beginning. */
475 wordend, /* Succeeds if at word end. */
477 wordbound, /* Succeeds if at a word boundary. */
478 notwordbound /* Succeeds if not at a word boundary. */
481 ,before_dot, /* Succeeds if before point. */
482 at_dot, /* Succeeds if at point. */
483 after_dot, /* Succeeds if after point. */
485 /* Matches any character whose syntax is specified. Followed by
486 a byte which contains a syntax code, e.g., Sword. */
489 /* Matches any character whose syntax is not that specified. */
494 /* Common operations on the compiled pattern. */
496 /* Store NUMBER in two contiguous bytes starting at DESTINATION. */
498 #define STORE_NUMBER(destination, number) \
500 (destination)[0] = (number) & 0377; \
501 (destination)[1] = (number) >> 8; \
504 /* Same as STORE_NUMBER, except increment DESTINATION to
505 the byte after where the number is stored. Therefore, DESTINATION
506 must be an lvalue. */
508 #define STORE_NUMBER_AND_INCR(destination, number) \
510 STORE_NUMBER (destination, number); \
511 (destination) += 2; \
514 /* Put into DESTINATION a number stored in two contiguous bytes starting
517 #define EXTRACT_NUMBER(destination, source) \
519 (destination) = *(source) & 0377; \
520 (destination) += SIGN_EXTEND_CHAR (*((source) + 1)) << 8; \
524 static void extract_number _RE_ARGS ((int *dest, unsigned char *source));
526 extract_number (dest, source)
528 unsigned char *source;
530 int temp = SIGN_EXTEND_CHAR (*(source + 1));
531 *dest = *source & 0377;
535 # ifndef EXTRACT_MACROS /* To debug the macros. */
536 # undef EXTRACT_NUMBER
537 # define EXTRACT_NUMBER(dest, src) extract_number (&dest, src)
538 # endif /* not EXTRACT_MACROS */
542 /* Same as EXTRACT_NUMBER, except increment SOURCE to after the number.
543 SOURCE must be an lvalue. */
545 #define EXTRACT_NUMBER_AND_INCR(destination, source) \
547 EXTRACT_NUMBER (destination, source); \
552 static void extract_number_and_incr _RE_ARGS ((int *destination,
553 unsigned char **source));
555 extract_number_and_incr (destination, source)
557 unsigned char **source;
559 extract_number (destination, *source);
563 # ifndef EXTRACT_MACROS
564 # undef EXTRACT_NUMBER_AND_INCR
565 # define EXTRACT_NUMBER_AND_INCR(dest, src) \
566 extract_number_and_incr (&dest, &src)
567 # endif /* not EXTRACT_MACROS */
571 /* If DEBUG is defined, Regex prints many voluminous messages about what
572 it is doing (if the variable `debug' is nonzero). If linked with the
573 main program in `iregex.c', you can enter patterns and strings
574 interactively. And if linked with the main program in `main.c' and
575 the other test files, you can run the already-written tests. */
579 /* We use standard I/O for debugging. */
582 /* It is useful to test things that ``must'' be true when debugging. */
585 static int debug = 0;
587 # define DEBUG_STATEMENT(e) e
588 # define DEBUG_PRINT1(x) if (debug) printf (x)
589 # define DEBUG_PRINT2(x1, x2) if (debug) printf (x1, x2)
590 # define DEBUG_PRINT3(x1, x2, x3) if (debug) printf (x1, x2, x3)
591 # define DEBUG_PRINT4(x1, x2, x3, x4) if (debug) printf (x1, x2, x3, x4)
592 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) \
593 if (debug) print_partial_compiled_pattern (s, e)
594 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) \
595 if (debug) print_double_string (w, s1, sz1, s2, sz2)
598 /* Print the fastmap in human-readable form. */
601 print_fastmap (fastmap)
604 unsigned was_a_range = 0;
607 while (i < (1 << BYTEWIDTH))
613 while (i < (1 << BYTEWIDTH) && fastmap[i])
629 /* Print a compiled pattern string in human-readable form, starting at
630 the START pointer into it and ending just before the pointer END. */
633 print_partial_compiled_pattern (start, end)
634 unsigned char *start;
639 unsigned char *p = start;
640 unsigned char *pend = end;
648 /* Loop over pattern commands. */
651 printf ("%d:\t", p - start);
653 switch ((re_opcode_t) *p++)
661 printf ("/exactn/%d", mcnt);
672 printf ("/start_memory/%d/%d", mcnt, *p++);
677 printf ("/stop_memory/%d/%d", mcnt, *p++);
681 printf ("/duplicate/%d", *p++);
691 register int c, last = -100;
692 register int in_range = 0;
694 printf ("/charset [%s",
695 (re_opcode_t) *(p - 1) == charset_not ? "^" : "");
697 assert (p + *p < pend);
699 for (c = 0; c < 256; c++)
701 && (p[1 + (c/8)] & (1 << (c % 8))))
703 /* Are we starting a range? */
704 if (last + 1 == c && ! in_range)
709 /* Have we broken a range? */
710 else if (last + 1 != c && in_range)
739 case on_failure_jump:
740 extract_number_and_incr (&mcnt, &p);
741 printf ("/on_failure_jump to %d", p + mcnt - start);
744 case on_failure_keep_string_jump:
745 extract_number_and_incr (&mcnt, &p);
746 printf ("/on_failure_keep_string_jump to %d", p + mcnt - start);
749 case dummy_failure_jump:
750 extract_number_and_incr (&mcnt, &p);
751 printf ("/dummy_failure_jump to %d", p + mcnt - start);
754 case push_dummy_failure:
755 printf ("/push_dummy_failure");
759 extract_number_and_incr (&mcnt, &p);
760 printf ("/maybe_pop_jump to %d", p + mcnt - start);
763 case pop_failure_jump:
764 extract_number_and_incr (&mcnt, &p);
765 printf ("/pop_failure_jump to %d", p + mcnt - start);
769 extract_number_and_incr (&mcnt, &p);
770 printf ("/jump_past_alt to %d", p + mcnt - start);
774 extract_number_and_incr (&mcnt, &p);
775 printf ("/jump to %d", p + mcnt - start);
779 extract_number_and_incr (&mcnt, &p);
781 extract_number_and_incr (&mcnt2, &p);
782 printf ("/succeed_n to %d, %d times", p1 - start, mcnt2);
786 extract_number_and_incr (&mcnt, &p);
788 extract_number_and_incr (&mcnt2, &p);
789 printf ("/jump_n to %d, %d times", p1 - start, mcnt2);
793 extract_number_and_incr (&mcnt, &p);
795 extract_number_and_incr (&mcnt2, &p);
796 printf ("/set_number_at location %d to %d", p1 - start, mcnt2);
800 printf ("/wordbound");
804 printf ("/notwordbound");
816 printf ("/before_dot");
824 printf ("/after_dot");
828 printf ("/syntaxspec");
830 printf ("/%d", mcnt);
834 printf ("/notsyntaxspec");
836 printf ("/%d", mcnt);
841 printf ("/wordchar");
845 printf ("/notwordchar");
857 printf ("?%d", *(p-1));
863 printf ("%d:\tend of pattern.\n", p - start);
868 print_compiled_pattern (bufp)
869 struct re_pattern_buffer *bufp;
871 unsigned char *buffer = bufp->buffer;
873 print_partial_compiled_pattern (buffer, buffer + bufp->used);
874 printf ("%ld bytes used/%ld bytes allocated.\n",
875 bufp->used, bufp->allocated);
877 if (bufp->fastmap_accurate && bufp->fastmap)
879 printf ("fastmap: ");
880 print_fastmap (bufp->fastmap);
883 printf ("re_nsub: %d\t", bufp->re_nsub);
884 printf ("regs_alloc: %d\t", bufp->regs_allocated);
885 printf ("can_be_null: %d\t", bufp->can_be_null);
886 printf ("newline_anchor: %d\n", bufp->newline_anchor);
887 printf ("no_sub: %d\t", bufp->no_sub);
888 printf ("not_bol: %d\t", bufp->not_bol);
889 printf ("not_eol: %d\t", bufp->not_eol);
890 printf ("syntax: %lx\n", bufp->syntax);
891 /* Perhaps we should print the translate table? */
896 print_double_string (where, string1, size1, string2, size2)
909 if (FIRST_STRING_P (where))
911 for (this_char = where - string1; this_char < size1; this_char++)
912 putchar (string1[this_char]);
917 for (this_char = where - string2; this_char < size2; this_char++)
918 putchar (string2[this_char]);
929 #else /* not DEBUG */
934 # define DEBUG_STATEMENT(e)
935 # define DEBUG_PRINT1(x)
936 # define DEBUG_PRINT2(x1, x2)
937 # define DEBUG_PRINT3(x1, x2, x3)
938 # define DEBUG_PRINT4(x1, x2, x3, x4)
939 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e)
940 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2)
942 #endif /* not DEBUG */
944 /* Set by `re_set_syntax' to the current regexp syntax to recognize. Can
945 also be assigned to arbitrarily: each pattern buffer stores its own
946 syntax, so it can be changed between regex compilations. */
947 /* This has no initializer because initialized variables in Emacs
948 become read-only after dumping. */
949 reg_syntax_t re_syntax_options;
952 /* Specify the precise syntax of regexps for compilation. This provides
953 for compatibility for various utilities which historically have
954 different, incompatible syntaxes.
956 The argument SYNTAX is a bit mask comprised of the various bits
957 defined in regex.h. We return the old syntax. */
960 re_set_syntax (syntax)
963 reg_syntax_t ret = re_syntax_options;
965 re_syntax_options = syntax;
967 if (syntax & RE_DEBUG)
969 else if (debug) /* was on but now is not */
975 /* This table gives an error message for each of the error codes listed
976 in regex.h. Obviously the order here has to be same as there.
977 POSIX doesn't require that we do anything for REG_NOERROR,
978 but why not be nice? */
980 static const char *re_error_msgid[] =
982 gettext_noop ("Success"), /* REG_NOERROR */
983 gettext_noop ("No match"), /* REG_NOMATCH */
984 gettext_noop ("Invalid regular expression"), /* REG_BADPAT */
985 gettext_noop ("Invalid collation character"), /* REG_ECOLLATE */
986 gettext_noop ("Invalid character class name"), /* REG_ECTYPE */
987 gettext_noop ("Trailing backslash"), /* REG_EESCAPE */
988 gettext_noop ("Invalid back reference"), /* REG_ESUBREG */
989 gettext_noop ("Unmatched [ or [^"), /* REG_EBRACK */
990 gettext_noop ("Unmatched ( or \\("), /* REG_EPAREN */
991 gettext_noop ("Unmatched \\{"), /* REG_EBRACE */
992 gettext_noop ("Invalid content of \\{\\}"), /* REG_BADBR */
993 gettext_noop ("Invalid range end"), /* REG_ERANGE */
994 gettext_noop ("Memory exhausted"), /* REG_ESPACE */
995 gettext_noop ("Invalid preceding regular expression"), /* REG_BADRPT */
996 gettext_noop ("Premature end of regular expression"), /* REG_EEND */
997 gettext_noop ("Regular expression too big"), /* REG_ESIZE */
998 gettext_noop ("Unmatched ) or \\)"), /* REG_ERPAREN */
1001 /* Avoiding alloca during matching, to placate r_alloc. */
1003 /* Define MATCH_MAY_ALLOCATE unless we need to make sure that the
1004 searching and matching functions should not call alloca. On some
1005 systems, alloca is implemented in terms of malloc, and if we're
1006 using the relocating allocator routines, then malloc could cause a
1007 relocation, which might (if the strings being searched are in the
1008 ralloc heap) shift the data out from underneath the regexp
1011 Here's another reason to avoid allocation: Emacs
1012 processes input from X in a signal handler; processing X input may
1013 call malloc; if input arrives while a matching routine is calling
1014 malloc, then we're scrod. But Emacs can't just block input while
1015 calling matching routines; then we don't notice interrupts when
1016 they come in. So, Emacs blocks input around all regexp calls
1017 except the matching calls, which it leaves unprotected, in the
1018 faith that they will not malloc. */
1020 /* Normally, this is fine. */
1021 #define MATCH_MAY_ALLOCATE
1023 /* When using GNU C, we are not REALLY using the C alloca, no matter
1024 what config.h may say. So don't take precautions for it. */
1029 /* The match routines may not allocate if (1) they would do it with malloc
1030 and (2) it's not safe for them to use malloc.
1031 Note that if REL_ALLOC is defined, matching would not use malloc for the
1032 failure stack, but we would still use it for the register vectors;
1033 so REL_ALLOC should not affect this. */
1034 #if (defined C_ALLOCA || defined REGEX_MALLOC) && defined emacs
1035 # undef MATCH_MAY_ALLOCATE
1039 /* Failure stack declarations and macros; both re_compile_fastmap and
1040 re_match_2 use a failure stack. These have to be macros because of
1041 REGEX_ALLOCATE_STACK. */
1044 /* Number of failure points for which to initially allocate space
1045 when matching. If this number is exceeded, we allocate more
1046 space, so it is not a hard limit. */
1047 #ifndef INIT_FAILURE_ALLOC
1048 # define INIT_FAILURE_ALLOC 5
1051 /* Roughly the maximum number of failure points on the stack. Would be
1052 exactly that if always used MAX_FAILURE_ITEMS items each time we failed.
1053 This is a variable only so users of regex can assign to it; we never
1054 change it ourselves. */
1058 # if defined MATCH_MAY_ALLOCATE
1059 /* 4400 was enough to cause a crash on Alpha OSF/1,
1060 whose default stack limit is 2mb. */
1061 long int re_max_failures = 4000;
1063 long int re_max_failures = 2000;
1066 union fail_stack_elt
1068 unsigned char *pointer;
1072 typedef union fail_stack_elt fail_stack_elt_t;
1076 fail_stack_elt_t *stack;
1077 unsigned long int size;
1078 unsigned long int avail; /* Offset of next open position. */
1081 #else /* not INT_IS_16BIT */
1083 # if defined MATCH_MAY_ALLOCATE
1084 /* 4400 was enough to cause a crash on Alpha OSF/1,
1085 whose default stack limit is 2mb. */
1086 int re_max_failures = 20000;
1088 int re_max_failures = 2000;
1091 union fail_stack_elt
1093 unsigned char *pointer;
1097 typedef union fail_stack_elt fail_stack_elt_t;
1101 fail_stack_elt_t *stack;
1103 unsigned avail; /* Offset of next open position. */
1106 #endif /* INT_IS_16BIT */
1108 #define FAIL_STACK_EMPTY() (fail_stack.avail == 0)
1109 #define FAIL_STACK_PTR_EMPTY() (fail_stack_ptr->avail == 0)
1110 #define FAIL_STACK_FULL() (fail_stack.avail == fail_stack.size)
1113 /* Define macros to initialize and free the failure stack.
1114 Do `return -2' if the alloc fails. */
1116 #ifdef MATCH_MAY_ALLOCATE
1117 # define INIT_FAIL_STACK() \
1119 fail_stack.stack = (fail_stack_elt_t *) \
1120 REGEX_ALLOCATE_STACK (INIT_FAILURE_ALLOC * sizeof (fail_stack_elt_t)); \
1122 if (fail_stack.stack == NULL) \
1125 fail_stack.size = INIT_FAILURE_ALLOC; \
1126 fail_stack.avail = 0; \
1129 # define RESET_FAIL_STACK() REGEX_FREE_STACK (fail_stack.stack)
1131 # define INIT_FAIL_STACK() \
1133 fail_stack.avail = 0; \
1136 # define RESET_FAIL_STACK()
1140 /* Double the size of FAIL_STACK, up to approximately `re_max_failures' items.
1142 Return 1 if succeeds, and 0 if either ran out of memory
1143 allocating space for it or it was already too large.
1145 REGEX_REALLOCATE_STACK requires `destination' be declared. */
1147 #define DOUBLE_FAIL_STACK(fail_stack) \
1148 ((fail_stack).size > (unsigned) (re_max_failures * MAX_FAILURE_ITEMS) \
1150 : ((fail_stack).stack = (fail_stack_elt_t *) \
1151 REGEX_REALLOCATE_STACK ((fail_stack).stack, \
1152 (fail_stack).size * sizeof (fail_stack_elt_t), \
1153 ((fail_stack).size << 1) * sizeof (fail_stack_elt_t)), \
1155 (fail_stack).stack == NULL \
1157 : ((fail_stack).size <<= 1, \
1161 /* Push pointer POINTER on FAIL_STACK.
1162 Return 1 if was able to do so and 0 if ran out of memory allocating
1164 #define PUSH_PATTERN_OP(POINTER, FAIL_STACK) \
1165 ((FAIL_STACK_FULL () \
1166 && !DOUBLE_FAIL_STACK (FAIL_STACK)) \
1168 : ((FAIL_STACK).stack[(FAIL_STACK).avail++].pointer = POINTER, \
1171 /* Push a pointer value onto the failure stack.
1172 Assumes the variable `fail_stack'. Probably should only
1173 be called from within `PUSH_FAILURE_POINT'. */
1174 #define PUSH_FAILURE_POINTER(item) \
1175 fail_stack.stack[fail_stack.avail++].pointer = (unsigned char *) (item)
1177 /* This pushes an integer-valued item onto the failure stack.
1178 Assumes the variable `fail_stack'. Probably should only
1179 be called from within `PUSH_FAILURE_POINT'. */
1180 #define PUSH_FAILURE_INT(item) \
1181 fail_stack.stack[fail_stack.avail++].integer = (item)
1183 /* Push a fail_stack_elt_t value onto the failure stack.
1184 Assumes the variable `fail_stack'. Probably should only
1185 be called from within `PUSH_FAILURE_POINT'. */
1186 #define PUSH_FAILURE_ELT(item) \
1187 fail_stack.stack[fail_stack.avail++] = (item)
1189 /* These three POP... operations complement the three PUSH... operations.
1190 All assume that `fail_stack' is nonempty. */
1191 #define POP_FAILURE_POINTER() fail_stack.stack[--fail_stack.avail].pointer
1192 #define POP_FAILURE_INT() fail_stack.stack[--fail_stack.avail].integer
1193 #define POP_FAILURE_ELT() fail_stack.stack[--fail_stack.avail]
1195 /* Used to omit pushing failure point id's when we're not debugging. */
1197 # define DEBUG_PUSH PUSH_FAILURE_INT
1198 # define DEBUG_POP(item_addr) *(item_addr) = POP_FAILURE_INT ()
1200 # define DEBUG_PUSH(item)
1201 # define DEBUG_POP(item_addr)
1205 /* Push the information about the state we will need
1206 if we ever fail back to it.
1208 Requires variables fail_stack, regstart, regend, reg_info, and
1209 num_regs_pushed be declared. DOUBLE_FAIL_STACK requires `destination'
1212 Does `return FAILURE_CODE' if runs out of memory. */
1214 #define PUSH_FAILURE_POINT(pattern_place, string_place, failure_code) \
1216 char *destination; \
1217 /* Must be int, so when we don't save any registers, the arithmetic \
1218 of 0 + -1 isn't done as unsigned. */ \
1219 /* Can't be int, since there is not a shred of a guarantee that int \
1220 is wide enough to hold a value of something to which pointer can \
1222 active_reg_t this_reg; \
1224 DEBUG_STATEMENT (failure_id++); \
1225 DEBUG_STATEMENT (nfailure_points_pushed++); \
1226 DEBUG_PRINT2 ("\nPUSH_FAILURE_POINT #%u:\n", failure_id); \
1227 DEBUG_PRINT2 (" Before push, next avail: %d\n", (fail_stack).avail);\
1228 DEBUG_PRINT2 (" size: %d\n", (fail_stack).size);\
1230 DEBUG_PRINT2 (" slots needed: %ld\n", NUM_FAILURE_ITEMS); \
1231 DEBUG_PRINT2 (" available: %d\n", REMAINING_AVAIL_SLOTS); \
1233 /* Ensure we have enough space allocated for what we will push. */ \
1234 while (REMAINING_AVAIL_SLOTS < NUM_FAILURE_ITEMS) \
1236 if (!DOUBLE_FAIL_STACK (fail_stack)) \
1237 return failure_code; \
1239 DEBUG_PRINT2 ("\n Doubled stack; size now: %d\n", \
1240 (fail_stack).size); \
1241 DEBUG_PRINT2 (" slots available: %d\n", REMAINING_AVAIL_SLOTS);\
1244 /* Push the info, starting with the registers. */ \
1245 DEBUG_PRINT1 ("\n"); \
1248 for (this_reg = lowest_active_reg; this_reg <= highest_active_reg; \
1251 DEBUG_PRINT2 (" Pushing reg: %lu\n", this_reg); \
1252 DEBUG_STATEMENT (num_regs_pushed++); \
1254 DEBUG_PRINT2 (" start: %p\n", regstart[this_reg]); \
1255 PUSH_FAILURE_POINTER (regstart[this_reg]); \
1257 DEBUG_PRINT2 (" end: %p\n", regend[this_reg]); \
1258 PUSH_FAILURE_POINTER (regend[this_reg]); \
1260 DEBUG_PRINT2 (" info: %p\n ", \
1261 reg_info[this_reg].word.pointer); \
1262 DEBUG_PRINT2 (" match_null=%d", \
1263 REG_MATCH_NULL_STRING_P (reg_info[this_reg])); \
1264 DEBUG_PRINT2 (" active=%d", IS_ACTIVE (reg_info[this_reg])); \
1265 DEBUG_PRINT2 (" matched_something=%d", \
1266 MATCHED_SOMETHING (reg_info[this_reg])); \
1267 DEBUG_PRINT2 (" ever_matched=%d", \
1268 EVER_MATCHED_SOMETHING (reg_info[this_reg])); \
1269 DEBUG_PRINT1 ("\n"); \
1270 PUSH_FAILURE_ELT (reg_info[this_reg].word); \
1273 DEBUG_PRINT2 (" Pushing low active reg: %ld\n", lowest_active_reg);\
1274 PUSH_FAILURE_INT (lowest_active_reg); \
1276 DEBUG_PRINT2 (" Pushing high active reg: %ld\n", highest_active_reg);\
1277 PUSH_FAILURE_INT (highest_active_reg); \
1279 DEBUG_PRINT2 (" Pushing pattern %p:\n", pattern_place); \
1280 DEBUG_PRINT_COMPILED_PATTERN (bufp, pattern_place, pend); \
1281 PUSH_FAILURE_POINTER (pattern_place); \
1283 DEBUG_PRINT2 (" Pushing string %p: `", string_place); \
1284 DEBUG_PRINT_DOUBLE_STRING (string_place, string1, size1, string2, \
1286 DEBUG_PRINT1 ("'\n"); \
1287 PUSH_FAILURE_POINTER (string_place); \
1289 DEBUG_PRINT2 (" Pushing failure id: %u\n", failure_id); \
1290 DEBUG_PUSH (failure_id); \
1293 /* This is the number of items that are pushed and popped on the stack
1294 for each register. */
1295 #define NUM_REG_ITEMS 3
1297 /* Individual items aside from the registers. */
1299 # define NUM_NONREG_ITEMS 5 /* Includes failure point id. */
1301 # define NUM_NONREG_ITEMS 4
1304 /* We push at most this many items on the stack. */
1305 /* We used to use (num_regs - 1), which is the number of registers
1306 this regexp will save; but that was changed to 5
1307 to avoid stack overflow for a regexp with lots of parens. */
1308 #define MAX_FAILURE_ITEMS (5 * NUM_REG_ITEMS + NUM_NONREG_ITEMS)
1310 /* We actually push this many items. */
1311 #define NUM_FAILURE_ITEMS \
1313 ? 0 : highest_active_reg - lowest_active_reg + 1) \
1317 /* How many items can still be added to the stack without overflowing it. */
1318 #define REMAINING_AVAIL_SLOTS ((fail_stack).size - (fail_stack).avail)
1321 /* Pops what PUSH_FAIL_STACK pushes.
1323 We restore into the parameters, all of which should be lvalues:
1324 STR -- the saved data position.
1325 PAT -- the saved pattern position.
1326 LOW_REG, HIGH_REG -- the highest and lowest active registers.
1327 REGSTART, REGEND -- arrays of string positions.
1328 REG_INFO -- array of information about each subexpression.
1330 Also assumes the variables `fail_stack' and (if debugging), `bufp',
1331 `pend', `string1', `size1', `string2', and `size2'. */
1333 #define POP_FAILURE_POINT(str, pat, low_reg, high_reg, regstart, regend, reg_info)\
1335 DEBUG_STATEMENT (unsigned failure_id;) \
1336 active_reg_t this_reg; \
1337 const unsigned char *string_temp; \
1339 assert (!FAIL_STACK_EMPTY ()); \
1341 /* Remove failure points and point to how many regs pushed. */ \
1342 DEBUG_PRINT1 ("POP_FAILURE_POINT:\n"); \
1343 DEBUG_PRINT2 (" Before pop, next avail: %d\n", fail_stack.avail); \
1344 DEBUG_PRINT2 (" size: %d\n", fail_stack.size); \
1346 assert (fail_stack.avail >= NUM_NONREG_ITEMS); \
1348 DEBUG_POP (&failure_id); \
1349 DEBUG_PRINT2 (" Popping failure id: %u\n", failure_id); \
1351 /* If the saved string location is NULL, it came from an \
1352 on_failure_keep_string_jump opcode, and we want to throw away the \
1353 saved NULL, thus retaining our current position in the string. */ \
1354 string_temp = POP_FAILURE_POINTER (); \
1355 if (string_temp != NULL) \
1356 str = (const char *) string_temp; \
1358 DEBUG_PRINT2 (" Popping string %p: `", str); \
1359 DEBUG_PRINT_DOUBLE_STRING (str, string1, size1, string2, size2); \
1360 DEBUG_PRINT1 ("'\n"); \
1362 pat = (unsigned char *) POP_FAILURE_POINTER (); \
1363 DEBUG_PRINT2 (" Popping pattern %p:\n", pat); \
1364 DEBUG_PRINT_COMPILED_PATTERN (bufp, pat, pend); \
1366 /* Restore register info. */ \
1367 high_reg = (active_reg_t) POP_FAILURE_INT (); \
1368 DEBUG_PRINT2 (" Popping high active reg: %ld\n", high_reg); \
1370 low_reg = (active_reg_t) POP_FAILURE_INT (); \
1371 DEBUG_PRINT2 (" Popping low active reg: %ld\n", low_reg); \
1374 for (this_reg = high_reg; this_reg >= low_reg; this_reg--) \
1376 DEBUG_PRINT2 (" Popping reg: %ld\n", this_reg); \
1378 reg_info[this_reg].word = POP_FAILURE_ELT (); \
1379 DEBUG_PRINT2 (" info: %p\n", \
1380 reg_info[this_reg].word.pointer); \
1382 regend[this_reg] = (const char *) POP_FAILURE_POINTER (); \
1383 DEBUG_PRINT2 (" end: %p\n", regend[this_reg]); \
1385 regstart[this_reg] = (const char *) POP_FAILURE_POINTER (); \
1386 DEBUG_PRINT2 (" start: %p\n", regstart[this_reg]); \
1390 for (this_reg = highest_active_reg; this_reg > high_reg; this_reg--) \
1392 reg_info[this_reg].word.integer = 0; \
1393 regend[this_reg] = 0; \
1394 regstart[this_reg] = 0; \
1396 highest_active_reg = high_reg; \
1399 set_regs_matched_done = 0; \
1400 DEBUG_STATEMENT (nfailure_points_popped++); \
1401 } /* POP_FAILURE_POINT */
1405 /* Structure for per-register (a.k.a. per-group) information.
1406 Other register information, such as the
1407 starting and ending positions (which are addresses), and the list of
1408 inner groups (which is a bits list) are maintained in separate
1411 We are making a (strictly speaking) nonportable assumption here: that
1412 the compiler will pack our bit fields into something that fits into
1413 the type of `word', i.e., is something that fits into one item on the
1417 /* Declarations and macros for re_match_2. */
1421 fail_stack_elt_t word;
1424 /* This field is one if this group can match the empty string,
1425 zero if not. If not yet determined, `MATCH_NULL_UNSET_VALUE'. */
1426 #define MATCH_NULL_UNSET_VALUE 3
1427 unsigned match_null_string_p : 2;
1428 unsigned is_active : 1;
1429 unsigned matched_something : 1;
1430 unsigned ever_matched_something : 1;
1432 } register_info_type;
1434 #define REG_MATCH_NULL_STRING_P(R) ((R).bits.match_null_string_p)
1435 #define IS_ACTIVE(R) ((R).bits.is_active)
1436 #define MATCHED_SOMETHING(R) ((R).bits.matched_something)
1437 #define EVER_MATCHED_SOMETHING(R) ((R).bits.ever_matched_something)
1440 /* Call this when have matched a real character; it sets `matched' flags
1441 for the subexpressions which we are currently inside. Also records
1442 that those subexprs have matched. */
1443 #define SET_REGS_MATCHED() \
1446 if (!set_regs_matched_done) \
1449 set_regs_matched_done = 1; \
1450 for (r = lowest_active_reg; r <= highest_active_reg; r++) \
1452 MATCHED_SOMETHING (reg_info[r]) \
1453 = EVER_MATCHED_SOMETHING (reg_info[r]) \
1460 /* Registers are set to a sentinel when they haven't yet matched. */
1461 static char reg_unset_dummy;
1462 #define REG_UNSET_VALUE (®_unset_dummy)
1463 #define REG_UNSET(e) ((e) == REG_UNSET_VALUE)
1465 /* Subroutine declarations and macros for regex_compile. */
1467 static reg_errcode_t regex_compile _RE_ARGS ((const char *pattern, size_t size,
1468 reg_syntax_t syntax,
1469 struct re_pattern_buffer *bufp));
1470 static void store_op1 _RE_ARGS ((re_opcode_t op, unsigned char *loc, int arg));
1471 static void store_op2 _RE_ARGS ((re_opcode_t op, unsigned char *loc,
1472 int arg1, int arg2));
1473 static void insert_op1 _RE_ARGS ((re_opcode_t op, unsigned char *loc,
1474 int arg, unsigned char *end));
1475 static void insert_op2 _RE_ARGS ((re_opcode_t op, unsigned char *loc,
1476 int arg1, int arg2, unsigned char *end));
1477 static boolean at_begline_loc_p _RE_ARGS ((const char *pattern, const char *p,
1478 reg_syntax_t syntax));
1479 static boolean at_endline_loc_p _RE_ARGS ((const char *p, const char *pend,
1480 reg_syntax_t syntax));
1481 static reg_errcode_t compile_range _RE_ARGS ((const char **p_ptr,
1484 reg_syntax_t syntax,
1487 /* Fetch the next character in the uncompiled pattern---translating it
1488 if necessary. Also cast from a signed character in the constant
1489 string passed to us by the user to an unsigned char that we can use
1490 as an array index (in, e.g., `translate'). */
1492 # define PATFETCH(c) \
1493 do {if (p == pend) return REG_EEND; \
1494 c = (unsigned char) *p++; \
1495 if (translate) c = (unsigned char) translate[c]; \
1499 /* Fetch the next character in the uncompiled pattern, with no
1501 #define PATFETCH_RAW(c) \
1502 do {if (p == pend) return REG_EEND; \
1503 c = (unsigned char) *p++; \
1506 /* Go backwards one character in the pattern. */
1507 #define PATUNFETCH p--
1510 /* If `translate' is non-null, return translate[D], else just D. We
1511 cast the subscript to translate because some data is declared as
1512 `char *', to avoid warnings when a string constant is passed. But
1513 when we use a character as a subscript we must make it unsigned. */
1515 # define TRANSLATE(d) \
1516 (translate ? (char) translate[(unsigned char) (d)] : (d))
1520 /* Macros for outputting the compiled pattern into `buffer'. */
1522 /* If the buffer isn't allocated when it comes in, use this. */
1523 #define INIT_BUF_SIZE 32
1525 /* Make sure we have at least N more bytes of space in buffer. */
1526 #define GET_BUFFER_SPACE(n) \
1527 while ((unsigned long) (b - bufp->buffer + (n)) > bufp->allocated) \
1530 /* Make sure we have one more byte of buffer space and then add C to it. */
1531 #define BUF_PUSH(c) \
1533 GET_BUFFER_SPACE (1); \
1534 *b++ = (unsigned char) (c); \
1538 /* Ensure we have two more bytes of buffer space and then append C1 and C2. */
1539 #define BUF_PUSH_2(c1, c2) \
1541 GET_BUFFER_SPACE (2); \
1542 *b++ = (unsigned char) (c1); \
1543 *b++ = (unsigned char) (c2); \
1547 /* As with BUF_PUSH_2, except for three bytes. */
1548 #define BUF_PUSH_3(c1, c2, c3) \
1550 GET_BUFFER_SPACE (3); \
1551 *b++ = (unsigned char) (c1); \
1552 *b++ = (unsigned char) (c2); \
1553 *b++ = (unsigned char) (c3); \
1557 /* Store a jump with opcode OP at LOC to location TO. We store a
1558 relative address offset by the three bytes the jump itself occupies. */
1559 #define STORE_JUMP(op, loc, to) \
1560 store_op1 (op, loc, (int) ((to) - (loc) - 3))
1562 /* Likewise, for a two-argument jump. */
1563 #define STORE_JUMP2(op, loc, to, arg) \
1564 store_op2 (op, loc, (int) ((to) - (loc) - 3), arg)
1566 /* Like `STORE_JUMP', but for inserting. Assume `b' is the buffer end. */
1567 #define INSERT_JUMP(op, loc, to) \
1568 insert_op1 (op, loc, (int) ((to) - (loc) - 3), b)
1570 /* Like `STORE_JUMP2', but for inserting. Assume `b' is the buffer end. */
1571 #define INSERT_JUMP2(op, loc, to, arg) \
1572 insert_op2 (op, loc, (int) ((to) - (loc) - 3), arg, b)
1575 /* This is not an arbitrary limit: the arguments which represent offsets
1576 into the pattern are two bytes long. So if 2^16 bytes turns out to
1577 be too small, many things would have to change. */
1578 /* Any other compiler which, like MSC, has allocation limit below 2^16
1579 bytes will have to use approach similar to what was done below for
1580 MSC and drop MAX_BUF_SIZE a bit. Otherwise you may end up
1581 reallocating to 0 bytes. Such thing is not going to work too well.
1582 You have been warned!! */
1583 #if defined _MSC_VER && !defined WIN32
1584 /* Microsoft C 16-bit versions limit malloc to approx 65512 bytes.
1585 The REALLOC define eliminates a flurry of conversion warnings,
1586 but is not required. */
1587 # define MAX_BUF_SIZE 65500L
1588 # define REALLOC(p,s) realloc ((p), (size_t) (s))
1590 # define MAX_BUF_SIZE (1L << 16)
1591 # define REALLOC(p,s) realloc ((p), (s))
1594 /* Extend the buffer by twice its current size via realloc and
1595 reset the pointers that pointed into the old block to point to the
1596 correct places in the new one. If extending the buffer results in it
1597 being larger than MAX_BUF_SIZE, then flag memory exhausted. */
1598 #define EXTEND_BUFFER() \
1600 unsigned char *old_buffer = bufp->buffer; \
1601 if (bufp->allocated == MAX_BUF_SIZE) \
1603 bufp->allocated <<= 1; \
1604 if (bufp->allocated > MAX_BUF_SIZE) \
1605 bufp->allocated = MAX_BUF_SIZE; \
1606 bufp->buffer = (unsigned char *) REALLOC (bufp->buffer, bufp->allocated);\
1607 if (bufp->buffer == NULL) \
1608 return REG_ESPACE; \
1609 /* If the buffer moved, move all the pointers into it. */ \
1610 if (old_buffer != bufp->buffer) \
1612 b = (b - old_buffer) + bufp->buffer; \
1613 begalt = (begalt - old_buffer) + bufp->buffer; \
1614 if (fixup_alt_jump) \
1615 fixup_alt_jump = (fixup_alt_jump - old_buffer) + bufp->buffer;\
1617 laststart = (laststart - old_buffer) + bufp->buffer; \
1618 if (pending_exact) \
1619 pending_exact = (pending_exact - old_buffer) + bufp->buffer; \
1624 /* Since we have one byte reserved for the register number argument to
1625 {start,stop}_memory, the maximum number of groups we can report
1626 things about is what fits in that byte. */
1627 #define MAX_REGNUM 255
1629 /* But patterns can have more than `MAX_REGNUM' registers. We just
1630 ignore the excess. */
1631 typedef unsigned regnum_t;
1634 /* Macros for the compile stack. */
1636 /* Since offsets can go either forwards or backwards, this type needs to
1637 be able to hold values from -(MAX_BUF_SIZE - 1) to MAX_BUF_SIZE - 1. */
1638 /* int may be not enough when sizeof(int) == 2. */
1639 typedef long pattern_offset_t;
1643 pattern_offset_t begalt_offset;
1644 pattern_offset_t fixup_alt_jump;
1645 pattern_offset_t inner_group_offset;
1646 pattern_offset_t laststart_offset;
1648 } compile_stack_elt_t;
1653 compile_stack_elt_t *stack;
1655 unsigned avail; /* Offset of next open position. */
1656 } compile_stack_type;
1659 #define INIT_COMPILE_STACK_SIZE 32
1661 #define COMPILE_STACK_EMPTY (compile_stack.avail == 0)
1662 #define COMPILE_STACK_FULL (compile_stack.avail == compile_stack.size)
1664 /* The next available element. */
1665 #define COMPILE_STACK_TOP (compile_stack.stack[compile_stack.avail])
1668 /* Set the bit for character C in a list. */
1669 #define SET_LIST_BIT(c) \
1670 (b[((unsigned char) (c)) / BYTEWIDTH] \
1671 |= 1 << (((unsigned char) c) % BYTEWIDTH))
1674 /* Get the next unsigned number in the uncompiled pattern. */
1675 #define GET_UNSIGNED_NUMBER(num) \
1679 while (ISDIGIT (c)) \
1683 num = num * 10 + c - '0'; \
1691 #if WIDE_CHAR_SUPPORT
1692 /* The GNU C library provides support for user-defined character classes
1693 and the functions from ISO C amendement 1. */
1694 # ifdef CHARCLASS_NAME_MAX
1695 # define CHAR_CLASS_MAX_LENGTH CHARCLASS_NAME_MAX
1697 /* This shouldn't happen but some implementation might still have this
1698 problem. Use a reasonable default value. */
1699 # define CHAR_CLASS_MAX_LENGTH 256
1702 # define IS_CHAR_CLASS(string) wctype (string)
1704 # define CHAR_CLASS_MAX_LENGTH 6 /* Namely, `xdigit'. */
1706 # define IS_CHAR_CLASS(string) \
1707 (STREQ (string, "alpha") || STREQ (string, "upper") \
1708 || STREQ (string, "lower") || STREQ (string, "digit") \
1709 || STREQ (string, "alnum") || STREQ (string, "xdigit") \
1710 || STREQ (string, "space") || STREQ (string, "print") \
1711 || STREQ (string, "punct") || STREQ (string, "graph") \
1712 || STREQ (string, "cntrl") || STREQ (string, "blank"))
1715 #ifndef MATCH_MAY_ALLOCATE
1717 /* If we cannot allocate large objects within re_match_2_internal,
1718 we make the fail stack and register vectors global.
1719 The fail stack, we grow to the maximum size when a regexp
1721 The register vectors, we adjust in size each time we
1722 compile a regexp, according to the number of registers it needs. */
1724 static fail_stack_type fail_stack;
1726 /* Size with which the following vectors are currently allocated.
1727 That is so we can make them bigger as needed,
1728 but never make them smaller. */
1729 static int regs_allocated_size;
1731 static const char ** regstart, ** regend;
1732 static const char ** old_regstart, ** old_regend;
1733 static const char **best_regstart, **best_regend;
1734 static register_info_type *reg_info;
1735 static const char **reg_dummy;
1736 static register_info_type *reg_info_dummy;
1738 /* Make the register vectors big enough for NUM_REGS registers,
1739 but don't make them smaller. */
1742 regex_grow_registers (num_regs)
1745 if (num_regs > regs_allocated_size)
1747 RETALLOC_IF (regstart, num_regs, const char *);
1748 RETALLOC_IF (regend, num_regs, const char *);
1749 RETALLOC_IF (old_regstart, num_regs, const char *);
1750 RETALLOC_IF (old_regend, num_regs, const char *);
1751 RETALLOC_IF (best_regstart, num_regs, const char *);
1752 RETALLOC_IF (best_regend, num_regs, const char *);
1753 RETALLOC_IF (reg_info, num_regs, register_info_type);
1754 RETALLOC_IF (reg_dummy, num_regs, const char *);
1755 RETALLOC_IF (reg_info_dummy, num_regs, register_info_type);
1757 regs_allocated_size = num_regs;
1761 #endif /* not MATCH_MAY_ALLOCATE */
1763 static boolean group_in_compile_stack _RE_ARGS ((compile_stack_type
1767 /* `regex_compile' compiles PATTERN (of length SIZE) according to SYNTAX.
1768 Returns one of error codes defined in `regex.h', or zero for success.
1770 Assumes the `allocated' (and perhaps `buffer') and `translate'
1771 fields are set in BUFP on entry.
1773 If it succeeds, results are put in BUFP (if it returns an error, the
1774 contents of BUFP are undefined):
1775 `buffer' is the compiled pattern;
1776 `syntax' is set to SYNTAX;
1777 `used' is set to the length of the compiled pattern;
1778 `fastmap_accurate' is zero;
1779 `re_nsub' is the number of subexpressions in PATTERN;
1780 `not_bol' and `not_eol' are zero;
1782 The `fastmap' and `newline_anchor' fields are neither
1783 examined nor set. */
1785 /* Return, freeing storage we allocated. */
1786 #define FREE_STACK_RETURN(value) \
1787 return (free (compile_stack.stack), value)
1789 static reg_errcode_t
1790 regex_compile (pattern, size, syntax, bufp)
1791 const char *pattern;
1793 reg_syntax_t syntax;
1794 struct re_pattern_buffer *bufp;
1796 /* We fetch characters from PATTERN here. Even though PATTERN is
1797 `char *' (i.e., signed), we declare these variables as unsigned, so
1798 they can be reliably used as array indices. */
1799 register unsigned char c, c1;
1801 /* A random temporary spot in PATTERN. */
1804 /* Points to the end of the buffer, where we should append. */
1805 register unsigned char *b;
1807 /* Keeps track of unclosed groups. */
1808 compile_stack_type compile_stack;
1810 /* Points to the current (ending) position in the pattern. */
1811 const char *p = pattern;
1812 const char *pend = pattern + size;
1814 /* How to translate the characters in the pattern. */
1815 RE_TRANSLATE_TYPE translate = bufp->translate;
1817 /* Address of the count-byte of the most recently inserted `exactn'
1818 command. This makes it possible to tell if a new exact-match
1819 character can be added to that command or if the character requires
1820 a new `exactn' command. */
1821 unsigned char *pending_exact = 0;
1823 /* Address of start of the most recently finished expression.
1824 This tells, e.g., postfix * where to find the start of its
1825 operand. Reset at the beginning of groups and alternatives. */
1826 unsigned char *laststart = 0;
1828 /* Address of beginning of regexp, or inside of last group. */
1829 unsigned char *begalt;
1831 /* Place in the uncompiled pattern (i.e., the {) to
1832 which to go back if the interval is invalid. */
1833 const char *beg_interval;
1835 /* Address of the place where a forward jump should go to the end of
1836 the containing expression. Each alternative of an `or' -- except the
1837 last -- ends with a forward jump of this sort. */
1838 unsigned char *fixup_alt_jump = 0;
1840 /* Counts open-groups as they are encountered. Remembered for the
1841 matching close-group on the compile stack, so the same register
1842 number is put in the stop_memory as the start_memory. */
1843 regnum_t regnum = 0;
1846 DEBUG_PRINT1 ("\nCompiling pattern: ");
1849 unsigned debug_count;
1851 for (debug_count = 0; debug_count < size; debug_count++)
1852 putchar (pattern[debug_count]);
1857 /* Initialize the compile stack. */
1858 compile_stack.stack = TALLOC (INIT_COMPILE_STACK_SIZE, compile_stack_elt_t);
1859 if (compile_stack.stack == NULL)
1862 compile_stack.size = INIT_COMPILE_STACK_SIZE;
1863 compile_stack.avail = 0;
1865 /* Initialize the pattern buffer. */
1866 bufp->syntax = syntax;
1867 bufp->fastmap_accurate = 0;
1868 bufp->not_bol = bufp->not_eol = 0;
1870 /* Set `used' to zero, so that if we return an error, the pattern
1871 printer (for debugging) will think there's no pattern. We reset it
1875 /* Always count groups, whether or not bufp->no_sub is set. */
1878 #if !defined emacs && !defined SYNTAX_TABLE
1879 /* Initialize the syntax table. */
1880 init_syntax_once ();
1883 if (bufp->allocated == 0)
1886 { /* If zero allocated, but buffer is non-null, try to realloc
1887 enough space. This loses if buffer's address is bogus, but
1888 that is the user's responsibility. */
1889 RETALLOC (bufp->buffer, INIT_BUF_SIZE, unsigned char);
1892 { /* Caller did not allocate a buffer. Do it for them. */
1893 bufp->buffer = TALLOC (INIT_BUF_SIZE, unsigned char);
1895 if (!bufp->buffer) FREE_STACK_RETURN (REG_ESPACE);
1897 bufp->allocated = INIT_BUF_SIZE;
1900 begalt = b = bufp->buffer;
1902 /* Loop through the uncompiled pattern until we're at the end. */
1911 if ( /* If at start of pattern, it's an operator. */
1913 /* If context independent, it's an operator. */
1914 || syntax & RE_CONTEXT_INDEP_ANCHORS
1915 /* Otherwise, depends on what's come before. */
1916 || at_begline_loc_p (pattern, p, syntax))
1926 if ( /* If at end of pattern, it's an operator. */
1928 /* If context independent, it's an operator. */
1929 || syntax & RE_CONTEXT_INDEP_ANCHORS
1930 /* Otherwise, depends on what's next. */
1931 || at_endline_loc_p (p, pend, syntax))
1941 if ((syntax & RE_BK_PLUS_QM)
1942 || (syntax & RE_LIMITED_OPS))
1946 /* If there is no previous pattern... */
1949 if (syntax & RE_CONTEXT_INVALID_OPS)
1950 FREE_STACK_RETURN (REG_BADRPT);
1951 else if (!(syntax & RE_CONTEXT_INDEP_OPS))
1956 /* Are we optimizing this jump? */
1957 boolean keep_string_p = false;
1959 /* 1 means zero (many) matches is allowed. */
1960 char zero_times_ok = 0, many_times_ok = 0;
1962 /* If there is a sequence of repetition chars, collapse it
1963 down to just one (the right one). We can't combine
1964 interval operators with these because of, e.g., `a{2}*',
1965 which should only match an even number of `a's. */
1969 zero_times_ok |= c != '+';
1970 many_times_ok |= c != '?';
1978 || (!(syntax & RE_BK_PLUS_QM) && (c == '+' || c == '?')))
1981 else if (syntax & RE_BK_PLUS_QM && c == '\\')
1983 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
1986 if (!(c1 == '+' || c1 == '?'))
2001 /* If we get here, we found another repeat character. */
2004 /* Star, etc. applied to an empty pattern is equivalent
2005 to an empty pattern. */
2009 /* Now we know whether or not zero matches is allowed
2010 and also whether or not two or more matches is allowed. */
2012 { /* More than one repetition is allowed, so put in at the
2013 end a backward relative jump from `b' to before the next
2014 jump we're going to put in below (which jumps from
2015 laststart to after this jump).
2017 But if we are at the `*' in the exact sequence `.*\n',
2018 insert an unconditional jump backwards to the .,
2019 instead of the beginning of the loop. This way we only
2020 push a failure point once, instead of every time
2021 through the loop. */
2022 assert (p - 1 > pattern);
2024 /* Allocate the space for the jump. */
2025 GET_BUFFER_SPACE (3);
2027 /* We know we are not at the first character of the pattern,
2028 because laststart was nonzero. And we've already
2029 incremented `p', by the way, to be the character after
2030 the `*'. Do we have to do something analogous here
2031 for null bytes, because of RE_DOT_NOT_NULL? */
2032 if (TRANSLATE (*(p - 2)) == TRANSLATE ('.')
2034 && p < pend && TRANSLATE (*p) == TRANSLATE ('\n')
2035 && !(syntax & RE_DOT_NEWLINE))
2036 { /* We have .*\n. */
2037 STORE_JUMP (jump, b, laststart);
2038 keep_string_p = true;
2041 /* Anything else. */
2042 STORE_JUMP (maybe_pop_jump, b, laststart - 3);
2044 /* We've added more stuff to the buffer. */
2048 /* On failure, jump from laststart to b + 3, which will be the
2049 end of the buffer after this jump is inserted. */
2050 GET_BUFFER_SPACE (3);
2051 INSERT_JUMP (keep_string_p ? on_failure_keep_string_jump
2059 /* At least one repetition is required, so insert a
2060 `dummy_failure_jump' before the initial
2061 `on_failure_jump' instruction of the loop. This
2062 effects a skip over that instruction the first time
2063 we hit that loop. */
2064 GET_BUFFER_SPACE (3);
2065 INSERT_JUMP (dummy_failure_jump, laststart, laststart + 6);
2080 boolean had_char_class = false;
2082 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2084 /* Ensure that we have enough space to push a charset: the
2085 opcode, the length count, and the bitset; 34 bytes in all. */
2086 GET_BUFFER_SPACE (34);
2090 /* We test `*p == '^' twice, instead of using an if
2091 statement, so we only need one BUF_PUSH. */
2092 BUF_PUSH (*p == '^' ? charset_not : charset);
2096 /* Remember the first position in the bracket expression. */
2099 /* Push the number of bytes in the bitmap. */
2100 BUF_PUSH ((1 << BYTEWIDTH) / BYTEWIDTH);
2102 /* Clear the whole map. */
2103 bzero (b, (1 << BYTEWIDTH) / BYTEWIDTH);
2105 /* charset_not matches newline according to a syntax bit. */
2106 if ((re_opcode_t) b[-2] == charset_not
2107 && (syntax & RE_HAT_LISTS_NOT_NEWLINE))
2108 SET_LIST_BIT ('\n');
2110 /* Read in characters and ranges, setting map bits. */
2113 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2117 /* \ might escape characters inside [...] and [^...]. */
2118 if ((syntax & RE_BACKSLASH_ESCAPE_IN_LISTS) && c == '\\')
2120 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
2127 /* Could be the end of the bracket expression. If it's
2128 not (i.e., when the bracket expression is `[]' so
2129 far), the ']' character bit gets set way below. */
2130 if (c == ']' && p != p1 + 1)
2133 /* Look ahead to see if it's a range when the last thing
2134 was a character class. */
2135 if (had_char_class && c == '-' && *p != ']')
2136 FREE_STACK_RETURN (REG_ERANGE);
2138 /* Look ahead to see if it's a range when the last thing
2139 was a character: if this is a hyphen not at the
2140 beginning or the end of a list, then it's the range
2143 && !(p - 2 >= pattern && p[-2] == '[')
2144 && !(p - 3 >= pattern && p[-3] == '[' && p[-2] == '^')
2148 = compile_range (&p, pend, translate, syntax, b);
2149 if (ret != REG_NOERROR) FREE_STACK_RETURN (ret);
2152 else if (p[0] == '-' && p[1] != ']')
2153 { /* This handles ranges made up of characters only. */
2156 /* Move past the `-'. */
2159 ret = compile_range (&p, pend, translate, syntax, b);
2160 if (ret != REG_NOERROR) FREE_STACK_RETURN (ret);
2163 /* See if we're at the beginning of a possible character
2166 else if (syntax & RE_CHAR_CLASSES && c == '[' && *p == ':')
2167 { /* Leave room for the null. */
2168 char str[CHAR_CLASS_MAX_LENGTH + 1];
2173 /* If pattern is `[[:'. */
2174 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2179 if (c == ':' || c == ']' || p == pend
2180 || c1 == CHAR_CLASS_MAX_LENGTH)
2186 /* If isn't a word bracketed by `[:' and:`]':
2187 undo the ending character, the letters, and leave
2188 the leading `:' and `[' (but set bits for them). */
2189 if (c == ':' && *p == ']')
2191 #if WIDE_CHAR_SUPPORT
2192 boolean is_lower = STREQ (str, "lower");
2193 boolean is_upper = STREQ (str, "upper");
2199 FREE_STACK_RETURN (REG_ECTYPE);
2201 /* Throw away the ] at the end of the character
2205 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2207 for (ch = 0; ch < 1 << BYTEWIDTH; ++ch)
2209 if (iswctype (btowc (ch), wt))
2212 if (translate && (is_upper || is_lower)
2213 && (ISUPPER (ch) || ISLOWER (ch)))
2217 had_char_class = true;
2220 boolean is_alnum = STREQ (str, "alnum");
2221 boolean is_alpha = STREQ (str, "alpha");
2222 boolean is_blank = STREQ (str, "blank");
2223 boolean is_cntrl = STREQ (str, "cntrl");
2224 boolean is_digit = STREQ (str, "digit");
2225 boolean is_graph = STREQ (str, "graph");
2226 boolean is_lower = STREQ (str, "lower");
2227 boolean is_print = STREQ (str, "print");
2228 boolean is_punct = STREQ (str, "punct");
2229 boolean is_space = STREQ (str, "space");
2230 boolean is_upper = STREQ (str, "upper");
2231 boolean is_xdigit = STREQ (str, "xdigit");
2233 if (!IS_CHAR_CLASS (str))
2234 FREE_STACK_RETURN (REG_ECTYPE);
2236 /* Throw away the ] at the end of the character
2240 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2242 for (ch = 0; ch < 1 << BYTEWIDTH; ch++)
2244 /* This was split into 3 if's to
2245 avoid an arbitrary limit in some compiler. */
2246 if ( (is_alnum && ISALNUM (ch))
2247 || (is_alpha && ISALPHA (ch))
2248 || (is_blank && ISBLANK (ch))
2249 || (is_cntrl && ISCNTRL (ch)))
2251 if ( (is_digit && ISDIGIT (ch))
2252 || (is_graph && ISGRAPH (ch))
2253 || (is_lower && ISLOWER (ch))
2254 || (is_print && ISPRINT (ch)))
2256 if ( (is_punct && ISPUNCT (ch))
2257 || (is_space && ISSPACE (ch))
2258 || (is_upper && ISUPPER (ch))
2259 || (is_xdigit && ISXDIGIT (ch)))
2261 if ( translate && (is_upper || is_lower)
2262 && (ISUPPER (ch) || ISLOWER (ch)))
2265 had_char_class = true;
2266 #endif /* libc || wctype.h */
2275 had_char_class = false;
2280 had_char_class = false;
2285 /* Discard any (non)matching list bytes that are all 0 at the
2286 end of the map. Decrease the map-length byte too. */
2287 while ((int) b[-1] > 0 && b[b[-1] - 1] == 0)
2295 if (syntax & RE_NO_BK_PARENS)
2302 if (syntax & RE_NO_BK_PARENS)
2309 if (syntax & RE_NEWLINE_ALT)
2316 if (syntax & RE_NO_BK_VBAR)
2323 if (syntax & RE_INTERVALS && syntax & RE_NO_BK_BRACES)
2324 goto handle_interval;
2330 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
2332 /* Do not translate the character after the \, so that we can
2333 distinguish, e.g., \B from \b, even if we normally would
2334 translate, e.g., B to b. */
2340 if (syntax & RE_NO_BK_PARENS)
2341 goto normal_backslash;
2347 if (COMPILE_STACK_FULL)
2349 RETALLOC (compile_stack.stack, compile_stack.size << 1,
2350 compile_stack_elt_t);
2351 if (compile_stack.stack == NULL) return REG_ESPACE;
2353 compile_stack.size <<= 1;
2356 /* These are the values to restore when we hit end of this
2357 group. They are all relative offsets, so that if the
2358 whole pattern moves because of realloc, they will still
2360 COMPILE_STACK_TOP.begalt_offset = begalt - bufp->buffer;
2361 COMPILE_STACK_TOP.fixup_alt_jump
2362 = fixup_alt_jump ? fixup_alt_jump - bufp->buffer + 1 : 0;
2363 COMPILE_STACK_TOP.laststart_offset = b - bufp->buffer;
2364 COMPILE_STACK_TOP.regnum = regnum;
2366 /* We will eventually replace the 0 with the number of
2367 groups inner to this one. But do not push a
2368 start_memory for groups beyond the last one we can
2369 represent in the compiled pattern. */
2370 if (regnum <= MAX_REGNUM)
2372 COMPILE_STACK_TOP.inner_group_offset = b - bufp->buffer + 2;
2373 BUF_PUSH_3 (start_memory, regnum, 0);
2376 compile_stack.avail++;
2381 /* If we've reached MAX_REGNUM groups, then this open
2382 won't actually generate any code, so we'll have to
2383 clear pending_exact explicitly. */
2389 if (syntax & RE_NO_BK_PARENS) goto normal_backslash;
2391 if (COMPILE_STACK_EMPTY)
2393 if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD)
2394 goto normal_backslash;
2396 FREE_STACK_RETURN (REG_ERPAREN);
2401 { /* Push a dummy failure point at the end of the
2402 alternative for a possible future
2403 `pop_failure_jump' to pop. See comments at
2404 `push_dummy_failure' in `re_match_2'. */
2405 BUF_PUSH (push_dummy_failure);
2407 /* We allocated space for this jump when we assigned
2408 to `fixup_alt_jump', in the `handle_alt' case below. */
2409 STORE_JUMP (jump_past_alt, fixup_alt_jump, b - 1);
2412 /* See similar code for backslashed left paren above. */
2413 if (COMPILE_STACK_EMPTY)
2415 if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD)
2418 FREE_STACK_RETURN (REG_ERPAREN);
2421 /* Since we just checked for an empty stack above, this
2422 ``can't happen''. */
2423 assert (compile_stack.avail != 0);
2425 /* We don't just want to restore into `regnum', because
2426 later groups should continue to be numbered higher,
2427 as in `(ab)c(de)' -- the second group is #2. */
2428 regnum_t this_group_regnum;
2430 compile_stack.avail--;
2431 begalt = bufp->buffer + COMPILE_STACK_TOP.begalt_offset;
2433 = COMPILE_STACK_TOP.fixup_alt_jump
2434 ? bufp->buffer + COMPILE_STACK_TOP.fixup_alt_jump - 1
2436 laststart = bufp->buffer + COMPILE_STACK_TOP.laststart_offset;
2437 this_group_regnum = COMPILE_STACK_TOP.regnum;
2438 /* If we've reached MAX_REGNUM groups, then this open
2439 won't actually generate any code, so we'll have to
2440 clear pending_exact explicitly. */
2443 /* We're at the end of the group, so now we know how many
2444 groups were inside this one. */
2445 if (this_group_regnum <= MAX_REGNUM)
2447 unsigned char *inner_group_loc
2448 = bufp->buffer + COMPILE_STACK_TOP.inner_group_offset;
2450 *inner_group_loc = regnum - this_group_regnum;
2451 BUF_PUSH_3 (stop_memory, this_group_regnum,
2452 regnum - this_group_regnum);
2458 case '|': /* `\|'. */
2459 if (syntax & RE_LIMITED_OPS || syntax & RE_NO_BK_VBAR)
2460 goto normal_backslash;
2462 if (syntax & RE_LIMITED_OPS)
2465 /* Insert before the previous alternative a jump which
2466 jumps to this alternative if the former fails. */
2467 GET_BUFFER_SPACE (3);
2468 INSERT_JUMP (on_failure_jump, begalt, b + 6);
2472 /* The alternative before this one has a jump after it
2473 which gets executed if it gets matched. Adjust that
2474 jump so it will jump to this alternative's analogous
2475 jump (put in below, which in turn will jump to the next
2476 (if any) alternative's such jump, etc.). The last such
2477 jump jumps to the correct final destination. A picture:
2483 If we are at `b', then fixup_alt_jump right now points to a
2484 three-byte space after `a'. We'll put in the jump, set
2485 fixup_alt_jump to right after `b', and leave behind three
2486 bytes which we'll fill in when we get to after `c'. */
2489 STORE_JUMP (jump_past_alt, fixup_alt_jump, b);
2491 /* Mark and leave space for a jump after this alternative,
2492 to be filled in later either by next alternative or
2493 when know we're at the end of a series of alternatives. */
2495 GET_BUFFER_SPACE (3);
2504 /* If \{ is a literal. */
2505 if (!(syntax & RE_INTERVALS)
2506 /* If we're at `\{' and it's not the open-interval
2508 || ((syntax & RE_INTERVALS) && (syntax & RE_NO_BK_BRACES))
2509 || (p - 2 == pattern && p == pend))
2510 goto normal_backslash;
2514 /* If got here, then the syntax allows intervals. */
2516 /* At least (most) this many matches must be made. */
2517 int lower_bound = -1, upper_bound = -1;
2519 beg_interval = p - 1;
2523 if (syntax & RE_NO_BK_BRACES)
2524 goto unfetch_interval;
2526 FREE_STACK_RETURN (REG_EBRACE);
2529 GET_UNSIGNED_NUMBER (lower_bound);
2533 GET_UNSIGNED_NUMBER (upper_bound);
2534 if (upper_bound < 0) upper_bound = RE_DUP_MAX;
2537 /* Interval such as `{1}' => match exactly once. */
2538 upper_bound = lower_bound;
2540 if (lower_bound < 0 || upper_bound > RE_DUP_MAX
2541 || lower_bound > upper_bound)
2543 if (syntax & RE_NO_BK_BRACES)
2544 goto unfetch_interval;
2546 FREE_STACK_RETURN (REG_BADBR);
2549 if (!(syntax & RE_NO_BK_BRACES))
2551 if (c != '\\') FREE_STACK_RETURN (REG_EBRACE);
2558 if (syntax & RE_NO_BK_BRACES)
2559 goto unfetch_interval;
2561 FREE_STACK_RETURN (REG_BADBR);
2564 /* We just parsed a valid interval. */
2566 /* If it's invalid to have no preceding re. */
2569 if (syntax & RE_CONTEXT_INVALID_OPS)
2570 FREE_STACK_RETURN (REG_BADRPT);
2571 else if (syntax & RE_CONTEXT_INDEP_OPS)
2574 goto unfetch_interval;
2577 /* If the upper bound is zero, don't want to succeed at
2578 all; jump from `laststart' to `b + 3', which will be
2579 the end of the buffer after we insert the jump. */
2580 if (upper_bound == 0)
2582 GET_BUFFER_SPACE (3);
2583 INSERT_JUMP (jump, laststart, b + 3);
2587 /* Otherwise, we have a nontrivial interval. When
2588 we're all done, the pattern will look like:
2589 set_number_at <jump count> <upper bound>
2590 set_number_at <succeed_n count> <lower bound>
2591 succeed_n <after jump addr> <succeed_n count>
2593 jump_n <succeed_n addr> <jump count>
2594 (The upper bound and `jump_n' are omitted if
2595 `upper_bound' is 1, though.) */
2597 { /* If the upper bound is > 1, we need to insert
2598 more at the end of the loop. */
2599 unsigned nbytes = 10 + (upper_bound > 1) * 10;
2601 GET_BUFFER_SPACE (nbytes);
2603 /* Initialize lower bound of the `succeed_n', even
2604 though it will be set during matching by its
2605 attendant `set_number_at' (inserted next),
2606 because `re_compile_fastmap' needs to know.
2607 Jump to the `jump_n' we might insert below. */
2608 INSERT_JUMP2 (succeed_n, laststart,
2609 b + 5 + (upper_bound > 1) * 5,
2613 /* Code to initialize the lower bound. Insert
2614 before the `succeed_n'. The `5' is the last two
2615 bytes of this `set_number_at', plus 3 bytes of
2616 the following `succeed_n'. */
2617 insert_op2 (set_number_at, laststart, 5, lower_bound, b);
2620 if (upper_bound > 1)
2621 { /* More than one repetition is allowed, so
2622 append a backward jump to the `succeed_n'
2623 that starts this interval.
2625 When we've reached this during matching,
2626 we'll have matched the interval once, so
2627 jump back only `upper_bound - 1' times. */
2628 STORE_JUMP2 (jump_n, b, laststart + 5,
2632 /* The location we want to set is the second
2633 parameter of the `jump_n'; that is `b-2' as
2634 an absolute address. `laststart' will be
2635 the `set_number_at' we're about to insert;
2636 `laststart+3' the number to set, the source
2637 for the relative address. But we are
2638 inserting into the middle of the pattern --
2639 so everything is getting moved up by 5.
2640 Conclusion: (b - 2) - (laststart + 3) + 5,
2641 i.e., b - laststart.
2643 We insert this at the beginning of the loop
2644 so that if we fail during matching, we'll
2645 reinitialize the bounds. */
2646 insert_op2 (set_number_at, laststart, b - laststart,
2647 upper_bound - 1, b);
2652 beg_interval = NULL;
2657 /* If an invalid interval, match the characters as literals. */
2658 assert (beg_interval);
2660 beg_interval = NULL;
2662 /* normal_char and normal_backslash need `c'. */
2665 if (!(syntax & RE_NO_BK_BRACES))
2667 if (p > pattern && p[-1] == '\\')
2668 goto normal_backslash;
2673 /* There is no way to specify the before_dot and after_dot
2674 operators. rms says this is ok. --karl */
2682 BUF_PUSH_2 (syntaxspec, syntax_spec_code[c]);
2688 BUF_PUSH_2 (notsyntaxspec, syntax_spec_code[c]);
2694 if (re_syntax_options & RE_NO_GNU_OPS)
2697 BUF_PUSH (wordchar);
2702 if (re_syntax_options & RE_NO_GNU_OPS)
2705 BUF_PUSH (notwordchar);
2710 if (re_syntax_options & RE_NO_GNU_OPS)
2716 if (re_syntax_options & RE_NO_GNU_OPS)
2722 if (re_syntax_options & RE_NO_GNU_OPS)
2724 BUF_PUSH (wordbound);
2728 if (re_syntax_options & RE_NO_GNU_OPS)
2730 BUF_PUSH (notwordbound);
2734 if (re_syntax_options & RE_NO_GNU_OPS)
2740 if (re_syntax_options & RE_NO_GNU_OPS)
2745 case '1': case '2': case '3': case '4': case '5':
2746 case '6': case '7': case '8': case '9':
2747 if (syntax & RE_NO_BK_REFS)
2753 FREE_STACK_RETURN (REG_ESUBREG);
2755 /* Can't back reference to a subexpression if inside of it. */
2756 if (group_in_compile_stack (compile_stack, (regnum_t) c1))
2760 BUF_PUSH_2 (duplicate, c1);
2766 if (syntax & RE_BK_PLUS_QM)
2769 goto normal_backslash;
2773 /* You might think it would be useful for \ to mean
2774 not to translate; but if we don't translate it
2775 it will never match anything. */
2783 /* Expects the character in `c'. */
2785 /* If no exactn currently being built. */
2788 /* If last exactn not at current position. */
2789 || pending_exact + *pending_exact + 1 != b
2791 /* We have only one byte following the exactn for the count. */
2792 || *pending_exact == (1 << BYTEWIDTH) - 1
2794 /* If followed by a repetition operator. */
2795 || *p == '*' || *p == '^'
2796 || ((syntax & RE_BK_PLUS_QM)
2797 ? *p == '\\' && (p[1] == '+' || p[1] == '?')
2798 : (*p == '+' || *p == '?'))
2799 || ((syntax & RE_INTERVALS)
2800 && ((syntax & RE_NO_BK_BRACES)
2802 : (p[0] == '\\' && p[1] == '{'))))
2804 /* Start building a new exactn. */
2808 BUF_PUSH_2 (exactn, 0);
2809 pending_exact = b - 1;
2816 } /* while p != pend */
2819 /* Through the pattern now. */
2822 STORE_JUMP (jump_past_alt, fixup_alt_jump, b);
2824 if (!COMPILE_STACK_EMPTY)
2825 FREE_STACK_RETURN (REG_EPAREN);
2827 /* If we don't want backtracking, force success
2828 the first time we reach the end of the compiled pattern. */
2829 if (syntax & RE_NO_POSIX_BACKTRACKING)
2832 free (compile_stack.stack);
2834 /* We have succeeded; set the length of the buffer. */
2835 bufp->used = b - bufp->buffer;
2840 DEBUG_PRINT1 ("\nCompiled pattern: \n");
2841 print_compiled_pattern (bufp);
2845 #ifndef MATCH_MAY_ALLOCATE
2846 /* Initialize the failure stack to the largest possible stack. This
2847 isn't necessary unless we're trying to avoid calling alloca in
2848 the search and match routines. */
2850 int num_regs = bufp->re_nsub + 1;
2852 /* Since DOUBLE_FAIL_STACK refuses to double only if the current size
2853 is strictly greater than re_max_failures, the largest possible stack
2854 is 2 * re_max_failures failure points. */
2855 if (fail_stack.size < (2 * re_max_failures * MAX_FAILURE_ITEMS))
2857 fail_stack.size = (2 * re_max_failures * MAX_FAILURE_ITEMS);
2860 if (! fail_stack.stack)
2862 = (fail_stack_elt_t *) xmalloc (fail_stack.size
2863 * sizeof (fail_stack_elt_t));
2866 = (fail_stack_elt_t *) xrealloc (fail_stack.stack,
2868 * sizeof (fail_stack_elt_t)));
2869 # else /* not emacs */
2870 if (! fail_stack.stack)
2872 = (fail_stack_elt_t *) malloc (fail_stack.size
2873 * sizeof (fail_stack_elt_t));
2876 = (fail_stack_elt_t *) realloc (fail_stack.stack,
2878 * sizeof (fail_stack_elt_t)));
2879 # endif /* not emacs */
2882 regex_grow_registers (num_regs);
2884 #endif /* not MATCH_MAY_ALLOCATE */
2887 } /* regex_compile */
2889 /* Subroutines for `regex_compile'. */
2891 /* Store OP at LOC followed by two-byte integer parameter ARG. */
2894 store_op1 (op, loc, arg)
2899 *loc = (unsigned char) op;
2900 STORE_NUMBER (loc + 1, arg);
2904 /* Like `store_op1', but for two two-byte parameters ARG1 and ARG2. */
2907 store_op2 (op, loc, arg1, arg2)
2912 *loc = (unsigned char) op;
2913 STORE_NUMBER (loc + 1, arg1);
2914 STORE_NUMBER (loc + 3, arg2);
2918 /* Copy the bytes from LOC to END to open up three bytes of space at LOC
2919 for OP followed by two-byte integer parameter ARG. */
2922 insert_op1 (op, loc, arg, end)
2928 register unsigned char *pfrom = end;
2929 register unsigned char *pto = end + 3;
2931 while (pfrom != loc)
2934 store_op1 (op, loc, arg);
2938 /* Like `insert_op1', but for two two-byte parameters ARG1 and ARG2. */
2941 insert_op2 (op, loc, arg1, arg2, end)
2947 register unsigned char *pfrom = end;
2948 register unsigned char *pto = end + 5;
2950 while (pfrom != loc)
2953 store_op2 (op, loc, arg1, arg2);
2957 /* P points to just after a ^ in PATTERN. Return true if that ^ comes
2958 after an alternative or a begin-subexpression. We assume there is at
2959 least one character before the ^. */
2962 at_begline_loc_p (pattern, p, syntax)
2963 const char *pattern, *p;
2964 reg_syntax_t syntax;
2966 const char *prev = p - 2;
2967 boolean prev_prev_backslash = prev > pattern && prev[-1] == '\\';
2970 /* After a subexpression? */
2971 (*prev == '(' && (syntax & RE_NO_BK_PARENS || prev_prev_backslash))
2972 /* After an alternative? */
2973 || (*prev == '|' && (syntax & RE_NO_BK_VBAR || prev_prev_backslash));
2977 /* The dual of at_begline_loc_p. This one is for $. We assume there is
2978 at least one character after the $, i.e., `P < PEND'. */
2981 at_endline_loc_p (p, pend, syntax)
2982 const char *p, *pend;
2983 reg_syntax_t syntax;
2985 const char *next = p;
2986 boolean next_backslash = *next == '\\';
2987 const char *next_next = p + 1 < pend ? p + 1 : 0;
2990 /* Before a subexpression? */
2991 (syntax & RE_NO_BK_PARENS ? *next == ')'
2992 : next_backslash && next_next && *next_next == ')')
2993 /* Before an alternative? */
2994 || (syntax & RE_NO_BK_VBAR ? *next == '|'
2995 : next_backslash && next_next && *next_next == '|');
2999 /* Returns true if REGNUM is in one of COMPILE_STACK's elements and
3000 false if it's not. */
3003 group_in_compile_stack (compile_stack, regnum)
3004 compile_stack_type compile_stack;
3009 for (this_element = compile_stack.avail - 1;
3012 if (compile_stack.stack[this_element].regnum == regnum)
3019 /* Read the ending character of a range (in a bracket expression) from the
3020 uncompiled pattern *P_PTR (which ends at PEND). We assume the
3021 starting character is in `P[-2]'. (`P[-1]' is the character `-'.)
3022 Then we set the translation of all bits between the starting and
3023 ending characters (inclusive) in the compiled pattern B.
3025 Return an error code.
3027 We use these short variable names so we can use the same macros as
3028 `regex_compile' itself. */
3030 static reg_errcode_t
3031 compile_range (p_ptr, pend, translate, syntax, b)
3032 const char **p_ptr, *pend;
3033 RE_TRANSLATE_TYPE translate;
3034 reg_syntax_t syntax;
3039 const char *p = *p_ptr;
3040 unsigned int range_start, range_end;
3045 /* Even though the pattern is a signed `char *', we need to fetch
3046 with unsigned char *'s; if the high bit of the pattern character
3047 is set, the range endpoints will be negative if we fetch using a
3050 We also want to fetch the endpoints without translating them; the
3051 appropriate translation is done in the bit-setting loop below. */
3052 /* The SVR4 compiler on the 3B2 had trouble with unsigned const char *. */
3053 range_start = ((const unsigned char *) p)[-2];
3054 range_end = ((const unsigned char *) p)[0];
3056 /* Have to increment the pointer into the pattern string, so the
3057 caller isn't still at the ending character. */
3060 /* If the start is after the end, the range is empty. */
3061 if (range_start > range_end)
3062 return syntax & RE_NO_EMPTY_RANGES ? REG_ERANGE : REG_NOERROR;
3064 /* Here we see why `this_char' has to be larger than an `unsigned
3065 char' -- the range is inclusive, so if `range_end' == 0xff
3066 (assuming 8-bit characters), we would otherwise go into an infinite
3067 loop, since all characters <= 0xff. */
3068 for (this_char = range_start; this_char <= range_end; this_char++)
3070 SET_LIST_BIT (TRANSLATE (this_char));
3076 /* re_compile_fastmap computes a ``fastmap'' for the compiled pattern in
3077 BUFP. A fastmap records which of the (1 << BYTEWIDTH) possible
3078 characters can start a string that matches the pattern. This fastmap
3079 is used by re_search to skip quickly over impossible starting points.
3081 The caller must supply the address of a (1 << BYTEWIDTH)-byte data
3082 area as BUFP->fastmap.
3084 We set the `fastmap', `fastmap_accurate', and `can_be_null' fields in
3087 Returns 0 if we succeed, -2 if an internal error. */
3090 re_compile_fastmap (bufp)
3091 struct re_pattern_buffer *bufp;
3094 #ifdef MATCH_MAY_ALLOCATE
3095 fail_stack_type fail_stack;
3097 #ifndef REGEX_MALLOC
3101 register char *fastmap = bufp->fastmap;
3102 unsigned char *pattern = bufp->buffer;
3103 unsigned char *p = pattern;
3104 register unsigned char *pend = pattern + bufp->used;
3107 /* This holds the pointer to the failure stack, when
3108 it is allocated relocatably. */
3109 fail_stack_elt_t *failure_stack_ptr;
3112 /* Assume that each path through the pattern can be null until
3113 proven otherwise. We set this false at the bottom of switch
3114 statement, to which we get only if a particular path doesn't
3115 match the empty string. */
3116 boolean path_can_be_null = true;
3118 /* We aren't doing a `succeed_n' to begin with. */
3119 boolean succeed_n_p = false;
3121 assert (fastmap != NULL && p != NULL);
3124 bzero (fastmap, 1 << BYTEWIDTH); /* Assume nothing's valid. */
3125 bufp->fastmap_accurate = 1; /* It will be when we're done. */
3126 bufp->can_be_null = 0;
3130 if (p == pend || *p == succeed)
3132 /* We have reached the (effective) end of pattern. */
3133 if (!FAIL_STACK_EMPTY ())
3135 bufp->can_be_null |= path_can_be_null;
3137 /* Reset for next path. */
3138 path_can_be_null = true;
3140 p = fail_stack.stack[--fail_stack.avail].pointer;
3148 /* We should never be about to go beyond the end of the pattern. */
3151 switch (SWITCH_ENUM_CAST ((re_opcode_t) *p++))
3154 /* I guess the idea here is to simply not bother with a fastmap
3155 if a backreference is used, since it's too hard to figure out
3156 the fastmap for the corresponding group. Setting
3157 `can_be_null' stops `re_search_2' from using the fastmap, so
3158 that is all we do. */
3160 bufp->can_be_null = 1;
3164 /* Following are the cases which match a character. These end
3173 for (j = *p++ * BYTEWIDTH - 1; j >= 0; j--)
3174 if (p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH)))
3180 /* Chars beyond end of map must be allowed. */
3181 for (j = *p * BYTEWIDTH; j < (1 << BYTEWIDTH); j++)
3184 for (j = *p++ * BYTEWIDTH - 1; j >= 0; j--)
3185 if (!(p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH))))
3191 for (j = 0; j < (1 << BYTEWIDTH); j++)
3192 if (SYNTAX (j) == Sword)
3198 for (j = 0; j < (1 << BYTEWIDTH); j++)
3199 if (SYNTAX (j) != Sword)
3206 int fastmap_newline = fastmap['\n'];
3208 /* `.' matches anything ... */
3209 for (j = 0; j < (1 << BYTEWIDTH); j++)
3212 /* ... except perhaps newline. */
3213 if (!(bufp->syntax & RE_DOT_NEWLINE))
3214 fastmap['\n'] = fastmap_newline;
3216 /* Return if we have already set `can_be_null'; if we have,
3217 then the fastmap is irrelevant. Something's wrong here. */
3218 else if (bufp->can_be_null)
3221 /* Otherwise, have to check alternative paths. */
3228 for (j = 0; j < (1 << BYTEWIDTH); j++)
3229 if (SYNTAX (j) == (enum syntaxcode) k)
3236 for (j = 0; j < (1 << BYTEWIDTH); j++)
3237 if (SYNTAX (j) != (enum syntaxcode) k)
3242 /* All cases after this match the empty string. These end with
3262 case push_dummy_failure:
3267 case pop_failure_jump:
3268 case maybe_pop_jump:
3271 case dummy_failure_jump:
3272 EXTRACT_NUMBER_AND_INCR (j, p);
3277 /* Jump backward implies we just went through the body of a
3278 loop and matched nothing. Opcode jumped to should be
3279 `on_failure_jump' or `succeed_n'. Just treat it like an
3280 ordinary jump. For a * loop, it has pushed its failure
3281 point already; if so, discard that as redundant. */
3282 if ((re_opcode_t) *p != on_failure_jump
3283 && (re_opcode_t) *p != succeed_n)
3287 EXTRACT_NUMBER_AND_INCR (j, p);
3290 /* If what's on the stack is where we are now, pop it. */
3291 if (!FAIL_STACK_EMPTY ()
3292 && fail_stack.stack[fail_stack.avail - 1].pointer == p)
3298 case on_failure_jump:
3299 case on_failure_keep_string_jump:
3300 handle_on_failure_jump:
3301 EXTRACT_NUMBER_AND_INCR (j, p);
3303 /* For some patterns, e.g., `(a?)?', `p+j' here points to the
3304 end of the pattern. We don't want to push such a point,
3305 since when we restore it above, entering the switch will
3306 increment `p' past the end of the pattern. We don't need
3307 to push such a point since we obviously won't find any more
3308 fastmap entries beyond `pend'. Such a pattern can match
3309 the null string, though. */
3312 if (!PUSH_PATTERN_OP (p + j, fail_stack))
3314 RESET_FAIL_STACK ();
3319 bufp->can_be_null = 1;
3323 EXTRACT_NUMBER_AND_INCR (k, p); /* Skip the n. */
3324 succeed_n_p = false;
3331 /* Get to the number of times to succeed. */
3334 /* Increment p past the n for when k != 0. */
3335 EXTRACT_NUMBER_AND_INCR (k, p);
3339 succeed_n_p = true; /* Spaghetti code alert. */
3340 goto handle_on_failure_jump;
3357 abort (); /* We have listed all the cases. */
3360 /* Getting here means we have found the possible starting
3361 characters for one path of the pattern -- and that the empty
3362 string does not match. We need not follow this path further.
3363 Instead, look at the next alternative (remembered on the
3364 stack), or quit if no more. The test at the top of the loop
3365 does these things. */
3366 path_can_be_null = false;
3370 /* Set `can_be_null' for the last path (also the first path, if the
3371 pattern is empty). */
3372 bufp->can_be_null |= path_can_be_null;
3375 RESET_FAIL_STACK ();
3377 } /* re_compile_fastmap */
3379 /* Set REGS to hold NUM_REGS registers, storing them in STARTS and
3380 ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use
3381 this memory for recording register information. STARTS and ENDS
3382 must be allocated using the malloc library routine, and must each
3383 be at least NUM_REGS * sizeof (regoff_t) bytes long.
3385 If NUM_REGS == 0, then subsequent matches should allocate their own
3388 Unless this function is called, the first search or match using
3389 PATTERN_BUFFER will allocate its own register data, without
3390 freeing the old data. */
3393 re_set_registers (bufp, regs, num_regs, starts, ends)
3394 struct re_pattern_buffer *bufp;
3395 struct re_registers *regs;
3397 regoff_t *starts, *ends;
3401 bufp->regs_allocated = REGS_REALLOCATE;
3402 regs->num_regs = num_regs;
3403 regs->start = starts;
3408 bufp->regs_allocated = REGS_UNALLOCATED;
3410 regs->start = regs->end = (regoff_t *) 0;
3414 /* Searching routines. */
3416 /* Like re_search_2, below, but only one string is specified, and
3417 doesn't let you say where to stop matching. */
3420 re_search (bufp, string, size, startpos, range, regs)
3421 struct re_pattern_buffer *bufp;
3423 int size, startpos, range;
3424 struct re_registers *regs;
3426 return re_search_2 (bufp, NULL, 0, string, size, startpos, range,
3431 /* Using the compiled pattern in BUFP->buffer, first tries to match the
3432 virtual concatenation of STRING1 and STRING2, starting first at index
3433 STARTPOS, then at STARTPOS + 1, and so on.
3435 STRING1 and STRING2 have length SIZE1 and SIZE2, respectively.
3437 RANGE is how far to scan while trying to match. RANGE = 0 means try
3438 only at STARTPOS; in general, the last start tried is STARTPOS +
3441 In REGS, return the indices of the virtual concatenation of STRING1
3442 and STRING2 that matched the entire BUFP->buffer and its contained
3445 Do not consider matching one past the index STOP in the virtual
3446 concatenation of STRING1 and STRING2.
3448 We return either the position in the strings at which the match was
3449 found, -1 if no match, or -2 if error (such as failure
3453 re_search_2 (bufp, string1, size1, string2, size2, startpos, range, regs, stop)
3454 struct re_pattern_buffer *bufp;
3455 const char *string1, *string2;
3459 struct re_registers *regs;
3463 register char *fastmap = bufp->fastmap;
3464 register RE_TRANSLATE_TYPE translate = bufp->translate;
3465 int total_size = size1 + size2;
3466 int endpos = startpos + range;
3468 /* Check for out-of-range STARTPOS. */
3469 if (startpos < 0 || startpos > total_size)
3472 /* Fix up RANGE if it might eventually take us outside
3473 the virtual concatenation of STRING1 and STRING2.
3474 Make sure we won't move STARTPOS below 0 or above TOTAL_SIZE. */
3476 range = 0 - startpos;
3477 else if (endpos > total_size)
3478 range = total_size - startpos;
3480 /* If the search isn't to be a backwards one, don't waste time in a
3481 search for a pattern that must be anchored. */
3482 if (bufp->used > 0 && (re_opcode_t) bufp->buffer[0] == begbuf && range > 0)
3491 /* In a forward search for something that starts with \=.
3492 don't keep searching past point. */
3493 if (bufp->used > 0 && (re_opcode_t) bufp->buffer[0] == at_dot && range > 0)
3495 range = PT - startpos;
3501 /* Update the fastmap now if not correct already. */
3502 if (fastmap && !bufp->fastmap_accurate)
3503 if (re_compile_fastmap (bufp) == -2)
3506 /* Loop through the string, looking for a place to start matching. */
3509 /* If a fastmap is supplied, skip quickly over characters that
3510 cannot be the start of a match. If the pattern can match the
3511 null string, however, we don't need to skip characters; we want
3512 the first null string. */
3513 if (fastmap && startpos < total_size && !bufp->can_be_null)
3515 if (range > 0) /* Searching forwards. */
3517 register const char *d;
3518 register int lim = 0;
3521 if (startpos < size1 && startpos + range >= size1)
3522 lim = range - (size1 - startpos);
3524 d = (startpos >= size1 ? string2 - size1 : string1) + startpos;
3526 /* Written out as an if-else to avoid testing `translate'
3530 && !fastmap[(unsigned char)
3531 translate[(unsigned char) *d++]])
3534 while (range > lim && !fastmap[(unsigned char) *d++])
3537 startpos += irange - range;
3539 else /* Searching backwards. */
3541 register char c = (size1 == 0 || startpos >= size1
3542 ? string2[startpos - size1]
3543 : string1[startpos]);
3545 if (!fastmap[(unsigned char) TRANSLATE (c)])
3550 /* If can't match the null string, and that's all we have left, fail. */
3551 if (range >= 0 && startpos == total_size && fastmap
3552 && !bufp->can_be_null)
3555 val = re_match_2_internal (bufp, string1, size1, string2, size2,
3556 startpos, regs, stop);
3557 #ifndef REGEX_MALLOC
3586 /* This converts PTR, a pointer into one of the search strings `string1'
3587 and `string2' into an offset from the beginning of that string. */
3588 #define POINTER_TO_OFFSET(ptr) \
3589 (FIRST_STRING_P (ptr) \
3590 ? ((regoff_t) ((ptr) - string1)) \
3591 : ((regoff_t) ((ptr) - string2 + size1)))
3593 /* Macros for dealing with the split strings in re_match_2. */
3595 #define MATCHING_IN_FIRST_STRING (dend == end_match_1)
3597 /* Call before fetching a character with *d. This switches over to
3598 string2 if necessary. */
3599 #define PREFETCH() \
3602 /* End of string2 => fail. */ \
3603 if (dend == end_match_2) \
3605 /* End of string1 => advance to string2. */ \
3607 dend = end_match_2; \
3611 /* Test if at very beginning or at very end of the virtual concatenation
3612 of `string1' and `string2'. If only one string, it's `string2'. */
3613 #define AT_STRINGS_BEG(d) ((d) == (size1 ? string1 : string2) || !size2)
3614 #define AT_STRINGS_END(d) ((d) == end2)
3617 /* Test if D points to a character which is word-constituent. We have
3618 two special cases to check for: if past the end of string1, look at
3619 the first character in string2; and if before the beginning of
3620 string2, look at the last character in string1. */
3621 #define WORDCHAR_P(d) \
3622 (SYNTAX ((d) == end1 ? *string2 \
3623 : (d) == string2 - 1 ? *(end1 - 1) : *(d)) \
3626 /* Disabled due to a compiler bug -- see comment at case wordbound */
3628 /* Test if the character before D and the one at D differ with respect
3629 to being word-constituent. */
3630 #define AT_WORD_BOUNDARY(d) \
3631 (AT_STRINGS_BEG (d) || AT_STRINGS_END (d) \
3632 || WORDCHAR_P (d - 1) != WORDCHAR_P (d))
3635 /* Free everything we malloc. */
3636 #ifdef MATCH_MAY_ALLOCATE
3637 # define FREE_VAR(var) if (var) REGEX_FREE (var); var = NULL
3638 # define FREE_VARIABLES() \
3640 REGEX_FREE_STACK (fail_stack.stack); \
3641 FREE_VAR (regstart); \
3642 FREE_VAR (regend); \
3643 FREE_VAR (old_regstart); \
3644 FREE_VAR (old_regend); \
3645 FREE_VAR (best_regstart); \
3646 FREE_VAR (best_regend); \
3647 FREE_VAR (reg_info); \
3648 FREE_VAR (reg_dummy); \
3649 FREE_VAR (reg_info_dummy); \
3652 # define FREE_VARIABLES() ((void)0) /* Do nothing! But inhibit gcc warning. */
3653 #endif /* not MATCH_MAY_ALLOCATE */
3655 /* These values must meet several constraints. They must not be valid
3656 register values; since we have a limit of 255 registers (because
3657 we use only one byte in the pattern for the register number), we can
3658 use numbers larger than 255. They must differ by 1, because of
3659 NUM_FAILURE_ITEMS above. And the value for the lowest register must
3660 be larger than the value for the highest register, so we do not try
3661 to actually save any registers when none are active. */
3662 #define NO_HIGHEST_ACTIVE_REG (1 << BYTEWIDTH)
3663 #define NO_LOWEST_ACTIVE_REG (NO_HIGHEST_ACTIVE_REG + 1)
3665 /* Matching routines. */
3667 #ifndef emacs /* Emacs never uses this. */
3668 /* re_match is like re_match_2 except it takes only a single string. */
3671 re_match (bufp, string, size, pos, regs)
3672 struct re_pattern_buffer *bufp;
3675 struct re_registers *regs;
3677 int result = re_match_2_internal (bufp, NULL, 0, string, size,
3679 # ifndef REGEX_MALLOC
3686 #endif /* not emacs */
3688 static boolean group_match_null_string_p _RE_ARGS ((unsigned char **p,
3690 register_info_type *reg_info));
3691 static boolean alt_match_null_string_p _RE_ARGS ((unsigned char *p,
3693 register_info_type *reg_info));
3694 static boolean common_op_match_null_string_p _RE_ARGS ((unsigned char **p,
3696 register_info_type *reg_info));
3697 static int bcmp_translate _RE_ARGS ((const char *s1, const char *s2,
3698 int len, char *translate));
3700 /* re_match_2 matches the compiled pattern in BUFP against the
3701 the (virtual) concatenation of STRING1 and STRING2 (of length SIZE1
3702 and SIZE2, respectively). We start matching at POS, and stop
3705 If REGS is non-null and the `no_sub' field of BUFP is nonzero, we
3706 store offsets for the substring each group matched in REGS. See the
3707 documentation for exactly how many groups we fill.
3709 We return -1 if no match, -2 if an internal error (such as the
3710 failure stack overflowing). Otherwise, we return the length of the
3711 matched substring. */
3714 re_match_2 (bufp, string1, size1, string2, size2, pos, regs, stop)
3715 struct re_pattern_buffer *bufp;
3716 const char *string1, *string2;
3719 struct re_registers *regs;
3722 int result = re_match_2_internal (bufp, string1, size1, string2, size2,
3724 #ifndef REGEX_MALLOC
3732 /* This is a separate function so that we can force an alloca cleanup
3735 re_match_2_internal (bufp, string1, size1, string2, size2, pos, regs, stop)
3736 struct re_pattern_buffer *bufp;
3737 const char *string1, *string2;
3740 struct re_registers *regs;
3743 /* General temporaries. */
3747 /* Just past the end of the corresponding string. */
3748 const char *end1, *end2;
3750 /* Pointers into string1 and string2, just past the last characters in
3751 each to consider matching. */
3752 const char *end_match_1, *end_match_2;
3754 /* Where we are in the data, and the end of the current string. */
3755 const char *d, *dend;
3757 /* Where we are in the pattern, and the end of the pattern. */
3758 unsigned char *p = bufp->buffer;
3759 register unsigned char *pend = p + bufp->used;
3761 /* Mark the opcode just after a start_memory, so we can test for an
3762 empty subpattern when we get to the stop_memory. */
3763 unsigned char *just_past_start_mem = 0;
3765 /* We use this to map every character in the string. */
3766 RE_TRANSLATE_TYPE translate = bufp->translate;
3768 /* Failure point stack. Each place that can handle a failure further
3769 down the line pushes a failure point on this stack. It consists of
3770 restart, regend, and reg_info for all registers corresponding to
3771 the subexpressions we're currently inside, plus the number of such
3772 registers, and, finally, two char *'s. The first char * is where
3773 to resume scanning the pattern; the second one is where to resume
3774 scanning the strings. If the latter is zero, the failure point is
3775 a ``dummy''; if a failure happens and the failure point is a dummy,
3776 it gets discarded and the next next one is tried. */
3777 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
3778 fail_stack_type fail_stack;
3781 static unsigned failure_id = 0;
3782 unsigned nfailure_points_pushed = 0, nfailure_points_popped = 0;
3786 /* This holds the pointer to the failure stack, when
3787 it is allocated relocatably. */
3788 fail_stack_elt_t *failure_stack_ptr;
3791 /* We fill all the registers internally, independent of what we
3792 return, for use in backreferences. The number here includes
3793 an element for register zero. */
3794 size_t num_regs = bufp->re_nsub + 1;
3796 /* The currently active registers. */
3797 active_reg_t lowest_active_reg = NO_LOWEST_ACTIVE_REG;
3798 active_reg_t highest_active_reg = NO_HIGHEST_ACTIVE_REG;
3800 /* Information on the contents of registers. These are pointers into
3801 the input strings; they record just what was matched (on this
3802 attempt) by a subexpression part of the pattern, that is, the
3803 regnum-th regstart pointer points to where in the pattern we began
3804 matching and the regnum-th regend points to right after where we
3805 stopped matching the regnum-th subexpression. (The zeroth register
3806 keeps track of what the whole pattern matches.) */
3807 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
3808 const char **regstart, **regend;
3811 /* If a group that's operated upon by a repetition operator fails to
3812 match anything, then the register for its start will need to be
3813 restored because it will have been set to wherever in the string we
3814 are when we last see its open-group operator. Similarly for a
3816 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
3817 const char **old_regstart, **old_regend;
3820 /* The is_active field of reg_info helps us keep track of which (possibly
3821 nested) subexpressions we are currently in. The matched_something
3822 field of reg_info[reg_num] helps us tell whether or not we have
3823 matched any of the pattern so far this time through the reg_num-th
3824 subexpression. These two fields get reset each time through any
3825 loop their register is in. */
3826 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
3827 register_info_type *reg_info;
3830 /* The following record the register info as found in the above
3831 variables when we find a match better than any we've seen before.
3832 This happens as we backtrack through the failure points, which in
3833 turn happens only if we have not yet matched the entire string. */
3834 unsigned best_regs_set = false;
3835 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
3836 const char **best_regstart, **best_regend;
3839 /* Logically, this is `best_regend[0]'. But we don't want to have to
3840 allocate space for that if we're not allocating space for anything
3841 else (see below). Also, we never need info about register 0 for
3842 any of the other register vectors, and it seems rather a kludge to
3843 treat `best_regend' differently than the rest. So we keep track of
3844 the end of the best match so far in a separate variable. We
3845 initialize this to NULL so that when we backtrack the first time
3846 and need to test it, it's not garbage. */
3847 const char *match_end = NULL;
3849 /* This helps SET_REGS_MATCHED avoid doing redundant work. */
3850 int set_regs_matched_done = 0;
3852 /* Used when we pop values we don't care about. */
3853 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
3854 const char **reg_dummy;
3855 register_info_type *reg_info_dummy;
3859 /* Counts the total number of registers pushed. */
3860 unsigned num_regs_pushed = 0;
3863 DEBUG_PRINT1 ("\n\nEntering re_match_2.\n");
3867 #ifdef MATCH_MAY_ALLOCATE
3868 /* Do not bother to initialize all the register variables if there are
3869 no groups in the pattern, as it takes a fair amount of time. If
3870 there are groups, we include space for register 0 (the whole
3871 pattern), even though we never use it, since it simplifies the
3872 array indexing. We should fix this. */
3875 regstart = REGEX_TALLOC (num_regs, const char *);
3876 regend = REGEX_TALLOC (num_regs, const char *);
3877 old_regstart = REGEX_TALLOC (num_regs, const char *);
3878 old_regend = REGEX_TALLOC (num_regs, const char *);
3879 best_regstart = REGEX_TALLOC (num_regs, const char *);
3880 best_regend = REGEX_TALLOC (num_regs, const char *);
3881 reg_info = REGEX_TALLOC (num_regs, register_info_type);
3882 reg_dummy = REGEX_TALLOC (num_regs, const char *);
3883 reg_info_dummy = REGEX_TALLOC (num_regs, register_info_type);
3885 if (!(regstart && regend && old_regstart && old_regend && reg_info
3886 && best_regstart && best_regend && reg_dummy && reg_info_dummy))
3894 /* We must initialize all our variables to NULL, so that
3895 `FREE_VARIABLES' doesn't try to free them. */
3896 regstart = regend = old_regstart = old_regend = best_regstart
3897 = best_regend = reg_dummy = NULL;
3898 reg_info = reg_info_dummy = (register_info_type *) NULL;
3900 #endif /* MATCH_MAY_ALLOCATE */
3902 /* The starting position is bogus. */
3903 if (pos < 0 || pos > size1 + size2)
3909 /* Initialize subexpression text positions to -1 to mark ones that no
3910 start_memory/stop_memory has been seen for. Also initialize the
3911 register information struct. */
3912 for (mcnt = 1; (unsigned) mcnt < num_regs; mcnt++)
3914 regstart[mcnt] = regend[mcnt]
3915 = old_regstart[mcnt] = old_regend[mcnt] = REG_UNSET_VALUE;
3917 REG_MATCH_NULL_STRING_P (reg_info[mcnt]) = MATCH_NULL_UNSET_VALUE;
3918 IS_ACTIVE (reg_info[mcnt]) = 0;
3919 MATCHED_SOMETHING (reg_info[mcnt]) = 0;
3920 EVER_MATCHED_SOMETHING (reg_info[mcnt]) = 0;
3923 /* We move `string1' into `string2' if the latter's empty -- but not if
3924 `string1' is null. */
3925 if (size2 == 0 && string1 != NULL)
3932 end1 = string1 + size1;
3933 end2 = string2 + size2;
3935 /* Compute where to stop matching, within the two strings. */
3938 end_match_1 = string1 + stop;
3939 end_match_2 = string2;
3944 end_match_2 = string2 + stop - size1;
3947 /* `p' scans through the pattern as `d' scans through the data.
3948 `dend' is the end of the input string that `d' points within. `d'
3949 is advanced into the following input string whenever necessary, but
3950 this happens before fetching; therefore, at the beginning of the
3951 loop, `d' can be pointing at the end of a string, but it cannot
3953 if (size1 > 0 && pos <= size1)
3960 d = string2 + pos - size1;
3964 DEBUG_PRINT1 ("The compiled pattern is:\n");
3965 DEBUG_PRINT_COMPILED_PATTERN (bufp, p, pend);
3966 DEBUG_PRINT1 ("The string to match is: `");
3967 DEBUG_PRINT_DOUBLE_STRING (d, string1, size1, string2, size2);
3968 DEBUG_PRINT1 ("'\n");
3970 /* This loops over pattern commands. It exits by returning from the
3971 function if the match is complete, or it drops through if the match
3972 fails at this starting point in the input data. */
3976 DEBUG_PRINT2 ("\n%p: ", p);
3978 DEBUG_PRINT2 ("\n0x%x: ", p);
3982 { /* End of pattern means we might have succeeded. */
3983 DEBUG_PRINT1 ("end of pattern ... ");
3985 /* If we haven't matched the entire string, and we want the
3986 longest match, try backtracking. */
3987 if (d != end_match_2)
3989 /* 1 if this match ends in the same string (string1 or string2)
3990 as the best previous match. */
3991 boolean same_str_p = (FIRST_STRING_P (match_end)
3992 == MATCHING_IN_FIRST_STRING);
3993 /* 1 if this match is the best seen so far. */
3994 boolean best_match_p;
3996 /* AIX compiler got confused when this was combined
3997 with the previous declaration. */
3999 best_match_p = d > match_end;
4001 best_match_p = !MATCHING_IN_FIRST_STRING;
4003 DEBUG_PRINT1 ("backtracking.\n");
4005 if (!FAIL_STACK_EMPTY ())
4006 { /* More failure points to try. */
4008 /* If exceeds best match so far, save it. */
4009 if (!best_regs_set || best_match_p)
4011 best_regs_set = true;
4014 DEBUG_PRINT1 ("\nSAVING match as best so far.\n");
4016 for (mcnt = 1; (unsigned) mcnt < num_regs; mcnt++)
4018 best_regstart[mcnt] = regstart[mcnt];
4019 best_regend[mcnt] = regend[mcnt];
4025 /* If no failure points, don't restore garbage. And if
4026 last match is real best match, don't restore second
4028 else if (best_regs_set && !best_match_p)
4031 /* Restore best match. It may happen that `dend ==
4032 end_match_1' while the restored d is in string2.
4033 For example, the pattern `x.*y.*z' against the
4034 strings `x-' and `y-z-', if the two strings are
4035 not consecutive in memory. */
4036 DEBUG_PRINT1 ("Restoring best registers.\n");
4039 dend = ((d >= string1 && d <= end1)
4040 ? end_match_1 : end_match_2);
4042 for (mcnt = 1; (unsigned) mcnt < num_regs; mcnt++)
4044 regstart[mcnt] = best_regstart[mcnt];
4045 regend[mcnt] = best_regend[mcnt];
4048 } /* d != end_match_2 */
4051 DEBUG_PRINT1 ("Accepting match.\n");
4053 /* If caller wants register contents data back, do it. */
4054 if (regs && !bufp->no_sub)
4056 /* Have the register data arrays been allocated? */
4057 if (bufp->regs_allocated == REGS_UNALLOCATED)
4058 { /* No. So allocate them with malloc. We need one
4059 extra element beyond `num_regs' for the `-1' marker
4061 regs->num_regs = MAX (RE_NREGS, num_regs + 1);
4062 regs->start = TALLOC (regs->num_regs, regoff_t);
4063 regs->end = TALLOC (regs->num_regs, regoff_t);
4064 if (regs->start == NULL || regs->end == NULL)
4069 bufp->regs_allocated = REGS_REALLOCATE;
4071 else if (bufp->regs_allocated == REGS_REALLOCATE)
4072 { /* Yes. If we need more elements than were already
4073 allocated, reallocate them. If we need fewer, just
4075 if (regs->num_regs < num_regs + 1)
4077 regs->num_regs = num_regs + 1;
4078 RETALLOC (regs->start, regs->num_regs, regoff_t);
4079 RETALLOC (regs->end, regs->num_regs, regoff_t);
4080 if (regs->start == NULL || regs->end == NULL)
4089 /* These braces fend off a "empty body in an else-statement"
4090 warning under GCC when assert expands to nothing. */
4091 assert (bufp->regs_allocated == REGS_FIXED);
4094 /* Convert the pointer data in `regstart' and `regend' to
4095 indices. Register zero has to be set differently,
4096 since we haven't kept track of any info for it. */
4097 if (regs->num_regs > 0)
4099 regs->start[0] = pos;
4100 regs->end[0] = (MATCHING_IN_FIRST_STRING
4101 ? ((regoff_t) (d - string1))
4102 : ((regoff_t) (d - string2 + size1)));
4105 /* Go through the first `min (num_regs, regs->num_regs)'
4106 registers, since that is all we initialized. */
4107 for (mcnt = 1; (unsigned) mcnt < MIN (num_regs, regs->num_regs);
4110 if (REG_UNSET (regstart[mcnt]) || REG_UNSET (regend[mcnt]))
4111 regs->start[mcnt] = regs->end[mcnt] = -1;
4115 = (regoff_t) POINTER_TO_OFFSET (regstart[mcnt]);
4117 = (regoff_t) POINTER_TO_OFFSET (regend[mcnt]);
4121 /* If the regs structure we return has more elements than
4122 were in the pattern, set the extra elements to -1. If
4123 we (re)allocated the registers, this is the case,
4124 because we always allocate enough to have at least one
4126 for (mcnt = num_regs; (unsigned) mcnt < regs->num_regs; mcnt++)
4127 regs->start[mcnt] = regs->end[mcnt] = -1;
4128 } /* regs && !bufp->no_sub */
4130 DEBUG_PRINT4 ("%u failure points pushed, %u popped (%u remain).\n",
4131 nfailure_points_pushed, nfailure_points_popped,
4132 nfailure_points_pushed - nfailure_points_popped);
4133 DEBUG_PRINT2 ("%u registers pushed.\n", num_regs_pushed);
4135 mcnt = d - pos - (MATCHING_IN_FIRST_STRING
4139 DEBUG_PRINT2 ("Returning %d from re_match_2.\n", mcnt);
4145 /* Otherwise match next pattern command. */
4146 switch (SWITCH_ENUM_CAST ((re_opcode_t) *p++))
4148 /* Ignore these. Used to ignore the n of succeed_n's which
4149 currently have n == 0. */
4151 DEBUG_PRINT1 ("EXECUTING no_op.\n");
4155 DEBUG_PRINT1 ("EXECUTING succeed.\n");
4158 /* Match the next n pattern characters exactly. The following
4159 byte in the pattern defines n, and the n bytes after that
4160 are the characters to match. */
4163 DEBUG_PRINT2 ("EXECUTING exactn %d.\n", mcnt);
4165 /* This is written out as an if-else so we don't waste time
4166 testing `translate' inside the loop. */
4172 if ((unsigned char) translate[(unsigned char) *d++]
4173 != (unsigned char) *p++)
4183 if (*d++ != (char) *p++) goto fail;
4187 SET_REGS_MATCHED ();
4191 /* Match any character except possibly a newline or a null. */
4193 DEBUG_PRINT1 ("EXECUTING anychar.\n");
4197 if ((!(bufp->syntax & RE_DOT_NEWLINE) && TRANSLATE (*d) == '\n')
4198 || (bufp->syntax & RE_DOT_NOT_NULL && TRANSLATE (*d) == '\000'))
4201 SET_REGS_MATCHED ();
4202 DEBUG_PRINT2 (" Matched `%d'.\n", *d);
4210 register unsigned char c;
4211 boolean not = (re_opcode_t) *(p - 1) == charset_not;
4213 DEBUG_PRINT2 ("EXECUTING charset%s.\n", not ? "_not" : "");
4216 c = TRANSLATE (*d); /* The character to match. */
4218 /* Cast to `unsigned' instead of `unsigned char' in case the
4219 bit list is a full 32 bytes long. */
4220 if (c < (unsigned) (*p * BYTEWIDTH)
4221 && p[1 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH)))
4226 if (!not) goto fail;
4228 SET_REGS_MATCHED ();
4234 /* The beginning of a group is represented by start_memory.
4235 The arguments are the register number in the next byte, and the
4236 number of groups inner to this one in the next. The text
4237 matched within the group is recorded (in the internal
4238 registers data structure) under the register number. */
4240 DEBUG_PRINT3 ("EXECUTING start_memory %d (%d):\n", *p, p[1]);
4242 /* Find out if this group can match the empty string. */
4243 p1 = p; /* To send to group_match_null_string_p. */
4245 if (REG_MATCH_NULL_STRING_P (reg_info[*p]) == MATCH_NULL_UNSET_VALUE)
4246 REG_MATCH_NULL_STRING_P (reg_info[*p])
4247 = group_match_null_string_p (&p1, pend, reg_info);
4249 /* Save the position in the string where we were the last time
4250 we were at this open-group operator in case the group is
4251 operated upon by a repetition operator, e.g., with `(a*)*b'
4252 against `ab'; then we want to ignore where we are now in
4253 the string in case this attempt to match fails. */
4254 old_regstart[*p] = REG_MATCH_NULL_STRING_P (reg_info[*p])
4255 ? REG_UNSET (regstart[*p]) ? d : regstart[*p]
4257 DEBUG_PRINT2 (" old_regstart: %d\n",
4258 POINTER_TO_OFFSET (old_regstart[*p]));
4261 DEBUG_PRINT2 (" regstart: %d\n", POINTER_TO_OFFSET (regstart[*p]));
4263 IS_ACTIVE (reg_info[*p]) = 1;
4264 MATCHED_SOMETHING (reg_info[*p]) = 0;
4266 /* Clear this whenever we change the register activity status. */
4267 set_regs_matched_done = 0;
4269 /* This is the new highest active register. */
4270 highest_active_reg = *p;
4272 /* If nothing was active before, this is the new lowest active
4274 if (lowest_active_reg == NO_LOWEST_ACTIVE_REG)
4275 lowest_active_reg = *p;
4277 /* Move past the register number and inner group count. */
4279 just_past_start_mem = p;
4284 /* The stop_memory opcode represents the end of a group. Its
4285 arguments are the same as start_memory's: the register
4286 number, and the number of inner groups. */
4288 DEBUG_PRINT3 ("EXECUTING stop_memory %d (%d):\n", *p, p[1]);
4290 /* We need to save the string position the last time we were at
4291 this close-group operator in case the group is operated
4292 upon by a repetition operator, e.g., with `((a*)*(b*)*)*'
4293 against `aba'; then we want to ignore where we are now in
4294 the string in case this attempt to match fails. */
4295 old_regend[*p] = REG_MATCH_NULL_STRING_P (reg_info[*p])
4296 ? REG_UNSET (regend[*p]) ? d : regend[*p]
4298 DEBUG_PRINT2 (" old_regend: %d\n",
4299 POINTER_TO_OFFSET (old_regend[*p]));
4302 DEBUG_PRINT2 (" regend: %d\n", POINTER_TO_OFFSET (regend[*p]));
4304 /* This register isn't active anymore. */
4305 IS_ACTIVE (reg_info[*p]) = 0;
4307 /* Clear this whenever we change the register activity status. */
4308 set_regs_matched_done = 0;
4310 /* If this was the only register active, nothing is active
4312 if (lowest_active_reg == highest_active_reg)
4314 lowest_active_reg = NO_LOWEST_ACTIVE_REG;
4315 highest_active_reg = NO_HIGHEST_ACTIVE_REG;
4318 { /* We must scan for the new highest active register, since
4319 it isn't necessarily one less than now: consider
4320 (a(b)c(d(e)f)g). When group 3 ends, after the f), the
4321 new highest active register is 1. */
4322 unsigned char r = *p - 1;
4323 while (r > 0 && !IS_ACTIVE (reg_info[r]))
4326 /* If we end up at register zero, that means that we saved
4327 the registers as the result of an `on_failure_jump', not
4328 a `start_memory', and we jumped to past the innermost
4329 `stop_memory'. For example, in ((.)*) we save
4330 registers 1 and 2 as a result of the *, but when we pop
4331 back to the second ), we are at the stop_memory 1.
4332 Thus, nothing is active. */
4335 lowest_active_reg = NO_LOWEST_ACTIVE_REG;
4336 highest_active_reg = NO_HIGHEST_ACTIVE_REG;
4339 highest_active_reg = r;
4342 /* If just failed to match something this time around with a
4343 group that's operated on by a repetition operator, try to
4344 force exit from the ``loop'', and restore the register
4345 information for this group that we had before trying this
4347 if ((!MATCHED_SOMETHING (reg_info[*p])
4348 || just_past_start_mem == p - 1)
4351 boolean is_a_jump_n = false;
4355 switch ((re_opcode_t) *p1++)
4359 case pop_failure_jump:
4360 case maybe_pop_jump:
4362 case dummy_failure_jump:
4363 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4373 /* If the next operation is a jump backwards in the pattern
4374 to an on_failure_jump right before the start_memory
4375 corresponding to this stop_memory, exit from the loop
4376 by forcing a failure after pushing on the stack the
4377 on_failure_jump's jump in the pattern, and d. */
4378 if (mcnt < 0 && (re_opcode_t) *p1 == on_failure_jump
4379 && (re_opcode_t) p1[3] == start_memory && p1[4] == *p)
4381 /* If this group ever matched anything, then restore
4382 what its registers were before trying this last
4383 failed match, e.g., with `(a*)*b' against `ab' for
4384 regstart[1], and, e.g., with `((a*)*(b*)*)*'
4385 against `aba' for regend[3].
4387 Also restore the registers for inner groups for,
4388 e.g., `((a*)(b*))*' against `aba' (register 3 would
4389 otherwise get trashed). */
4391 if (EVER_MATCHED_SOMETHING (reg_info[*p]))
4395 EVER_MATCHED_SOMETHING (reg_info[*p]) = 0;
4397 /* Restore this and inner groups' (if any) registers. */
4398 for (r = *p; r < (unsigned) *p + (unsigned) *(p + 1);
4401 regstart[r] = old_regstart[r];
4403 /* xx why this test? */
4404 if (old_regend[r] >= regstart[r])
4405 regend[r] = old_regend[r];
4409 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4410 PUSH_FAILURE_POINT (p1 + mcnt, d, -2);
4416 /* Move past the register number and the inner group count. */
4421 /* \<digit> has been turned into a `duplicate' command which is
4422 followed by the numeric value of <digit> as the register number. */
4425 register const char *d2, *dend2;
4426 int regno = *p++; /* Get which register to match against. */
4427 DEBUG_PRINT2 ("EXECUTING duplicate %d.\n", regno);
4429 /* Can't back reference a group which we've never matched. */
4430 if (REG_UNSET (regstart[regno]) || REG_UNSET (regend[regno]))
4433 /* Where in input to try to start matching. */
4434 d2 = regstart[regno];
4436 /* Where to stop matching; if both the place to start and
4437 the place to stop matching are in the same string, then
4438 set to the place to stop, otherwise, for now have to use
4439 the end of the first string. */
4441 dend2 = ((FIRST_STRING_P (regstart[regno])
4442 == FIRST_STRING_P (regend[regno]))
4443 ? regend[regno] : end_match_1);
4446 /* If necessary, advance to next segment in register
4450 if (dend2 == end_match_2) break;
4451 if (dend2 == regend[regno]) break;
4453 /* End of string1 => advance to string2. */
4455 dend2 = regend[regno];
4457 /* At end of register contents => success */
4458 if (d2 == dend2) break;
4460 /* If necessary, advance to next segment in data. */
4463 /* How many characters left in this segment to match. */
4466 /* Want how many consecutive characters we can match in
4467 one shot, so, if necessary, adjust the count. */
4468 if (mcnt > dend2 - d2)
4471 /* Compare that many; failure if mismatch, else move
4474 ? bcmp_translate (d, d2, mcnt, translate)
4475 : memcmp (d, d2, mcnt))
4477 d += mcnt, d2 += mcnt;
4479 /* Do this because we've match some characters. */
4480 SET_REGS_MATCHED ();
4486 /* begline matches the empty string at the beginning of the string
4487 (unless `not_bol' is set in `bufp'), and, if
4488 `newline_anchor' is set, after newlines. */
4490 DEBUG_PRINT1 ("EXECUTING begline.\n");
4492 if (AT_STRINGS_BEG (d))
4494 if (!bufp->not_bol) break;
4496 else if (d[-1] == '\n' && bufp->newline_anchor)
4500 /* In all other cases, we fail. */
4504 /* endline is the dual of begline. */
4506 DEBUG_PRINT1 ("EXECUTING endline.\n");
4508 if (AT_STRINGS_END (d))
4510 if (!bufp->not_eol) break;
4513 /* We have to ``prefetch'' the next character. */
4514 else if ((d == end1 ? *string2 : *d) == '\n'
4515 && bufp->newline_anchor)
4522 /* Match at the very beginning of the data. */
4524 DEBUG_PRINT1 ("EXECUTING begbuf.\n");
4525 if (AT_STRINGS_BEG (d))
4530 /* Match at the very end of the data. */
4532 DEBUG_PRINT1 ("EXECUTING endbuf.\n");
4533 if (AT_STRINGS_END (d))
4538 /* on_failure_keep_string_jump is used to optimize `.*\n'. It
4539 pushes NULL as the value for the string on the stack. Then
4540 `pop_failure_point' will keep the current value for the
4541 string, instead of restoring it. To see why, consider
4542 matching `foo\nbar' against `.*\n'. The .* matches the foo;
4543 then the . fails against the \n. But the next thing we want
4544 to do is match the \n against the \n; if we restored the
4545 string value, we would be back at the foo.
4547 Because this is used only in specific cases, we don't need to
4548 check all the things that `on_failure_jump' does, to make
4549 sure the right things get saved on the stack. Hence we don't
4550 share its code. The only reason to push anything on the
4551 stack at all is that otherwise we would have to change
4552 `anychar's code to do something besides goto fail in this
4553 case; that seems worse than this. */
4554 case on_failure_keep_string_jump:
4555 DEBUG_PRINT1 ("EXECUTING on_failure_keep_string_jump");
4557 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4559 DEBUG_PRINT3 (" %d (to %p):\n", mcnt, p + mcnt);
4561 DEBUG_PRINT3 (" %d (to 0x%x):\n", mcnt, p + mcnt);
4564 PUSH_FAILURE_POINT (p + mcnt, NULL, -2);
4568 /* Uses of on_failure_jump:
4570 Each alternative starts with an on_failure_jump that points
4571 to the beginning of the next alternative. Each alternative
4572 except the last ends with a jump that in effect jumps past
4573 the rest of the alternatives. (They really jump to the
4574 ending jump of the following alternative, because tensioning
4575 these jumps is a hassle.)
4577 Repeats start with an on_failure_jump that points past both
4578 the repetition text and either the following jump or
4579 pop_failure_jump back to this on_failure_jump. */
4580 case on_failure_jump:
4582 DEBUG_PRINT1 ("EXECUTING on_failure_jump");
4584 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4586 DEBUG_PRINT3 (" %d (to %p)", mcnt, p + mcnt);
4588 DEBUG_PRINT3 (" %d (to 0x%x)", mcnt, p + mcnt);
4591 /* If this on_failure_jump comes right before a group (i.e.,
4592 the original * applied to a group), save the information
4593 for that group and all inner ones, so that if we fail back
4594 to this point, the group's information will be correct.
4595 For example, in \(a*\)*\1, we need the preceding group,
4596 and in \(zz\(a*\)b*\)\2, we need the inner group. */
4598 /* We can't use `p' to check ahead because we push
4599 a failure point to `p + mcnt' after we do this. */
4602 /* We need to skip no_op's before we look for the
4603 start_memory in case this on_failure_jump is happening as
4604 the result of a completed succeed_n, as in \(a\)\{1,3\}b\1
4606 while (p1 < pend && (re_opcode_t) *p1 == no_op)
4609 if (p1 < pend && (re_opcode_t) *p1 == start_memory)
4611 /* We have a new highest active register now. This will
4612 get reset at the start_memory we are about to get to,
4613 but we will have saved all the registers relevant to
4614 this repetition op, as described above. */
4615 highest_active_reg = *(p1 + 1) + *(p1 + 2);
4616 if (lowest_active_reg == NO_LOWEST_ACTIVE_REG)
4617 lowest_active_reg = *(p1 + 1);
4620 DEBUG_PRINT1 (":\n");
4621 PUSH_FAILURE_POINT (p + mcnt, d, -2);
4625 /* A smart repeat ends with `maybe_pop_jump'.
4626 We change it to either `pop_failure_jump' or `jump'. */
4627 case maybe_pop_jump:
4628 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4629 DEBUG_PRINT2 ("EXECUTING maybe_pop_jump %d.\n", mcnt);
4631 register unsigned char *p2 = p;
4633 /* Compare the beginning of the repeat with what in the
4634 pattern follows its end. If we can establish that there
4635 is nothing that they would both match, i.e., that we
4636 would have to backtrack because of (as in, e.g., `a*a')
4637 then we can change to pop_failure_jump, because we'll
4638 never have to backtrack.
4640 This is not true in the case of alternatives: in
4641 `(a|ab)*' we do need to backtrack to the `ab' alternative
4642 (e.g., if the string was `ab'). But instead of trying to
4643 detect that here, the alternative has put on a dummy
4644 failure point which is what we will end up popping. */
4646 /* Skip over open/close-group commands.
4647 If what follows this loop is a ...+ construct,
4648 look at what begins its body, since we will have to
4649 match at least one of that. */
4653 && ((re_opcode_t) *p2 == stop_memory
4654 || (re_opcode_t) *p2 == start_memory))
4656 else if (p2 + 6 < pend
4657 && (re_opcode_t) *p2 == dummy_failure_jump)
4664 /* p1[0] ... p1[2] are the `on_failure_jump' corresponding
4665 to the `maybe_finalize_jump' of this case. Examine what
4668 /* If we're at the end of the pattern, we can change. */
4671 /* Consider what happens when matching ":\(.*\)"
4672 against ":/". I don't really understand this code
4674 p[-3] = (unsigned char) pop_failure_jump;
4676 (" End of pattern: change to `pop_failure_jump'.\n");
4679 else if ((re_opcode_t) *p2 == exactn
4680 || (bufp->newline_anchor && (re_opcode_t) *p2 == endline))
4682 register unsigned char c
4683 = *p2 == (unsigned char) endline ? '\n' : p2[2];
4685 if ((re_opcode_t) p1[3] == exactn && p1[5] != c)
4687 p[-3] = (unsigned char) pop_failure_jump;
4688 DEBUG_PRINT3 (" %c != %c => pop_failure_jump.\n",
4692 else if ((re_opcode_t) p1[3] == charset
4693 || (re_opcode_t) p1[3] == charset_not)
4695 int not = (re_opcode_t) p1[3] == charset_not;
4697 if (c < (unsigned char) (p1[4] * BYTEWIDTH)
4698 && p1[5 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH)))
4701 /* `not' is equal to 1 if c would match, which means
4702 that we can't change to pop_failure_jump. */
4705 p[-3] = (unsigned char) pop_failure_jump;
4706 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
4710 else if ((re_opcode_t) *p2 == charset)
4713 register unsigned char c
4714 = *p2 == (unsigned char) endline ? '\n' : p2[2];
4718 if ((re_opcode_t) p1[3] == exactn
4719 && ! ((int) p2[1] * BYTEWIDTH > (int) p1[5]
4720 && (p2[2 + p1[5] / BYTEWIDTH]
4721 & (1 << (p1[5] % BYTEWIDTH)))))
4723 if ((re_opcode_t) p1[3] == exactn
4724 && ! ((int) p2[1] * BYTEWIDTH > (int) p1[4]
4725 && (p2[2 + p1[4] / BYTEWIDTH]
4726 & (1 << (p1[4] % BYTEWIDTH)))))
4729 p[-3] = (unsigned char) pop_failure_jump;
4730 DEBUG_PRINT3 (" %c != %c => pop_failure_jump.\n",
4734 else if ((re_opcode_t) p1[3] == charset_not)
4737 /* We win if the charset_not inside the loop
4738 lists every character listed in the charset after. */
4739 for (idx = 0; idx < (int) p2[1]; idx++)
4740 if (! (p2[2 + idx] == 0
4741 || (idx < (int) p1[4]
4742 && ((p2[2 + idx] & ~ p1[5 + idx]) == 0))))
4747 p[-3] = (unsigned char) pop_failure_jump;
4748 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
4751 else if ((re_opcode_t) p1[3] == charset)
4754 /* We win if the charset inside the loop
4755 has no overlap with the one after the loop. */
4757 idx < (int) p2[1] && idx < (int) p1[4];
4759 if ((p2[2 + idx] & p1[5 + idx]) != 0)
4762 if (idx == p2[1] || idx == p1[4])
4764 p[-3] = (unsigned char) pop_failure_jump;
4765 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
4770 p -= 2; /* Point at relative address again. */
4771 if ((re_opcode_t) p[-1] != pop_failure_jump)
4773 p[-1] = (unsigned char) jump;
4774 DEBUG_PRINT1 (" Match => jump.\n");
4775 goto unconditional_jump;
4777 /* Note fall through. */
4780 /* The end of a simple repeat has a pop_failure_jump back to
4781 its matching on_failure_jump, where the latter will push a
4782 failure point. The pop_failure_jump takes off failure
4783 points put on by this pop_failure_jump's matching
4784 on_failure_jump; we got through the pattern to here from the
4785 matching on_failure_jump, so didn't fail. */
4786 case pop_failure_jump:
4788 /* We need to pass separate storage for the lowest and
4789 highest registers, even though we don't care about the
4790 actual values. Otherwise, we will restore only one
4791 register from the stack, since lowest will == highest in
4792 `pop_failure_point'. */
4793 active_reg_t dummy_low_reg, dummy_high_reg;
4794 unsigned char *pdummy;
4797 DEBUG_PRINT1 ("EXECUTING pop_failure_jump.\n");
4798 POP_FAILURE_POINT (sdummy, pdummy,
4799 dummy_low_reg, dummy_high_reg,
4800 reg_dummy, reg_dummy, reg_info_dummy);
4802 /* Note fall through. */
4806 DEBUG_PRINT2 ("\n%p: ", p);
4808 DEBUG_PRINT2 ("\n0x%x: ", p);
4810 /* Note fall through. */
4812 /* Unconditionally jump (without popping any failure points). */
4814 EXTRACT_NUMBER_AND_INCR (mcnt, p); /* Get the amount to jump. */
4815 DEBUG_PRINT2 ("EXECUTING jump %d ", mcnt);
4816 p += mcnt; /* Do the jump. */
4818 DEBUG_PRINT2 ("(to %p).\n", p);
4820 DEBUG_PRINT2 ("(to 0x%x).\n", p);
4825 /* We need this opcode so we can detect where alternatives end
4826 in `group_match_null_string_p' et al. */
4828 DEBUG_PRINT1 ("EXECUTING jump_past_alt.\n");
4829 goto unconditional_jump;
4832 /* Normally, the on_failure_jump pushes a failure point, which
4833 then gets popped at pop_failure_jump. We will end up at
4834 pop_failure_jump, also, and with a pattern of, say, `a+', we
4835 are skipping over the on_failure_jump, so we have to push
4836 something meaningless for pop_failure_jump to pop. */
4837 case dummy_failure_jump:
4838 DEBUG_PRINT1 ("EXECUTING dummy_failure_jump.\n");
4839 /* It doesn't matter what we push for the string here. What
4840 the code at `fail' tests is the value for the pattern. */
4841 PUSH_FAILURE_POINT (NULL, NULL, -2);
4842 goto unconditional_jump;
4845 /* At the end of an alternative, we need to push a dummy failure
4846 point in case we are followed by a `pop_failure_jump', because
4847 we don't want the failure point for the alternative to be
4848 popped. For example, matching `(a|ab)*' against `aab'
4849 requires that we match the `ab' alternative. */
4850 case push_dummy_failure:
4851 DEBUG_PRINT1 ("EXECUTING push_dummy_failure.\n");
4852 /* See comments just above at `dummy_failure_jump' about the
4854 PUSH_FAILURE_POINT (NULL, NULL, -2);
4857 /* Have to succeed matching what follows at least n times.
4858 After that, handle like `on_failure_jump'. */
4860 EXTRACT_NUMBER (mcnt, p + 2);
4861 DEBUG_PRINT2 ("EXECUTING succeed_n %d.\n", mcnt);
4864 /* Originally, this is how many times we HAVE to succeed. */
4869 STORE_NUMBER_AND_INCR (p, mcnt);
4871 DEBUG_PRINT3 (" Setting %p to %d.\n", p - 2, mcnt);
4873 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p - 2, mcnt);
4879 DEBUG_PRINT2 (" Setting two bytes from %p to no_op.\n", p+2);
4881 DEBUG_PRINT2 (" Setting two bytes from 0x%x to no_op.\n", p+2);
4883 p[2] = (unsigned char) no_op;
4884 p[3] = (unsigned char) no_op;
4890 EXTRACT_NUMBER (mcnt, p + 2);
4891 DEBUG_PRINT2 ("EXECUTING jump_n %d.\n", mcnt);
4893 /* Originally, this is how many times we CAN jump. */
4897 STORE_NUMBER (p + 2, mcnt);
4899 DEBUG_PRINT3 (" Setting %p to %d.\n", p + 2, mcnt);
4901 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p + 2, mcnt);
4903 goto unconditional_jump;
4905 /* If don't have to jump any more, skip over the rest of command. */
4912 DEBUG_PRINT1 ("EXECUTING set_number_at.\n");
4914 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4916 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4918 DEBUG_PRINT3 (" Setting %p to %d.\n", p1, mcnt);
4920 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p1, mcnt);
4922 STORE_NUMBER (p1, mcnt);
4927 /* The DEC Alpha C compiler 3.x generates incorrect code for the
4928 test WORDCHAR_P (d - 1) != WORDCHAR_P (d) in the expansion of
4929 AT_WORD_BOUNDARY, so this code is disabled. Expanding the
4930 macro and introducing temporary variables works around the bug. */
4933 DEBUG_PRINT1 ("EXECUTING wordbound.\n");
4934 if (AT_WORD_BOUNDARY (d))
4939 DEBUG_PRINT1 ("EXECUTING notwordbound.\n");
4940 if (AT_WORD_BOUNDARY (d))
4946 boolean prevchar, thischar;
4948 DEBUG_PRINT1 ("EXECUTING wordbound.\n");
4949 if (AT_STRINGS_BEG (d) || AT_STRINGS_END (d))
4952 prevchar = WORDCHAR_P (d - 1);
4953 thischar = WORDCHAR_P (d);
4954 if (prevchar != thischar)
4961 boolean prevchar, thischar;
4963 DEBUG_PRINT1 ("EXECUTING notwordbound.\n");
4964 if (AT_STRINGS_BEG (d) || AT_STRINGS_END (d))
4967 prevchar = WORDCHAR_P (d - 1);
4968 thischar = WORDCHAR_P (d);
4969 if (prevchar != thischar)
4976 DEBUG_PRINT1 ("EXECUTING wordbeg.\n");
4977 if (WORDCHAR_P (d) && (AT_STRINGS_BEG (d) || !WORDCHAR_P (d - 1)))
4982 DEBUG_PRINT1 ("EXECUTING wordend.\n");
4983 if (!AT_STRINGS_BEG (d) && WORDCHAR_P (d - 1)
4984 && (!WORDCHAR_P (d) || AT_STRINGS_END (d)))
4990 DEBUG_PRINT1 ("EXECUTING before_dot.\n");
4991 if (PTR_CHAR_POS ((unsigned char *) d) >= point)
4996 DEBUG_PRINT1 ("EXECUTING at_dot.\n");
4997 if (PTR_CHAR_POS ((unsigned char *) d) != point)
5002 DEBUG_PRINT1 ("EXECUTING after_dot.\n");
5003 if (PTR_CHAR_POS ((unsigned char *) d) <= point)
5008 DEBUG_PRINT2 ("EXECUTING syntaxspec %d.\n", mcnt);
5013 DEBUG_PRINT1 ("EXECUTING Emacs wordchar.\n");
5017 /* Can't use *d++ here; SYNTAX may be an unsafe macro. */
5019 if (SYNTAX (d[-1]) != (enum syntaxcode) mcnt)
5021 SET_REGS_MATCHED ();
5025 DEBUG_PRINT2 ("EXECUTING notsyntaxspec %d.\n", mcnt);
5027 goto matchnotsyntax;
5030 DEBUG_PRINT1 ("EXECUTING Emacs notwordchar.\n");
5034 /* Can't use *d++ here; SYNTAX may be an unsafe macro. */
5036 if (SYNTAX (d[-1]) == (enum syntaxcode) mcnt)
5038 SET_REGS_MATCHED ();
5041 #else /* not emacs */
5043 DEBUG_PRINT1 ("EXECUTING non-Emacs wordchar.\n");
5045 if (!WORDCHAR_P (d))
5047 SET_REGS_MATCHED ();
5052 DEBUG_PRINT1 ("EXECUTING non-Emacs notwordchar.\n");
5056 SET_REGS_MATCHED ();
5059 #endif /* not emacs */
5064 continue; /* Successfully executed one pattern command; keep going. */
5067 /* We goto here if a matching operation fails. */
5069 if (!FAIL_STACK_EMPTY ())
5070 { /* A restart point is known. Restore to that state. */
5071 DEBUG_PRINT1 ("\nFAIL:\n");
5072 POP_FAILURE_POINT (d, p,
5073 lowest_active_reg, highest_active_reg,
5074 regstart, regend, reg_info);
5076 /* If this failure point is a dummy, try the next one. */
5080 /* If we failed to the end of the pattern, don't examine *p. */
5084 boolean is_a_jump_n = false;
5086 /* If failed to a backwards jump that's part of a repetition
5087 loop, need to pop this failure point and use the next one. */
5088 switch ((re_opcode_t) *p)
5092 case maybe_pop_jump:
5093 case pop_failure_jump:
5096 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
5099 if ((is_a_jump_n && (re_opcode_t) *p1 == succeed_n)
5101 && (re_opcode_t) *p1 == on_failure_jump))
5109 if (d >= string1 && d <= end1)
5113 break; /* Matching at this starting point really fails. */
5117 goto restore_best_regs;
5121 return -1; /* Failure to match. */
5124 /* Subroutine definitions for re_match_2. */
5127 /* We are passed P pointing to a register number after a start_memory.
5129 Return true if the pattern up to the corresponding stop_memory can
5130 match the empty string, and false otherwise.
5132 If we find the matching stop_memory, sets P to point to one past its number.
5133 Otherwise, sets P to an undefined byte less than or equal to END.
5135 We don't handle duplicates properly (yet). */
5138 group_match_null_string_p (p, end, reg_info)
5139 unsigned char **p, *end;
5140 register_info_type *reg_info;
5143 /* Point to after the args to the start_memory. */
5144 unsigned char *p1 = *p + 2;
5148 /* Skip over opcodes that can match nothing, and return true or
5149 false, as appropriate, when we get to one that can't, or to the
5150 matching stop_memory. */
5152 switch ((re_opcode_t) *p1)
5154 /* Could be either a loop or a series of alternatives. */
5155 case on_failure_jump:
5157 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
5159 /* If the next operation is not a jump backwards in the
5164 /* Go through the on_failure_jumps of the alternatives,
5165 seeing if any of the alternatives cannot match nothing.
5166 The last alternative starts with only a jump,
5167 whereas the rest start with on_failure_jump and end
5168 with a jump, e.g., here is the pattern for `a|b|c':
5170 /on_failure_jump/0/6/exactn/1/a/jump_past_alt/0/6
5171 /on_failure_jump/0/6/exactn/1/b/jump_past_alt/0/3
5174 So, we have to first go through the first (n-1)
5175 alternatives and then deal with the last one separately. */
5178 /* Deal with the first (n-1) alternatives, which start
5179 with an on_failure_jump (see above) that jumps to right
5180 past a jump_past_alt. */
5182 while ((re_opcode_t) p1[mcnt-3] == jump_past_alt)
5184 /* `mcnt' holds how many bytes long the alternative
5185 is, including the ending `jump_past_alt' and
5188 if (!alt_match_null_string_p (p1, p1 + mcnt - 3,
5192 /* Move to right after this alternative, including the
5196 /* Break if it's the beginning of an n-th alternative
5197 that doesn't begin with an on_failure_jump. */
5198 if ((re_opcode_t) *p1 != on_failure_jump)
5201 /* Still have to check that it's not an n-th
5202 alternative that starts with an on_failure_jump. */
5204 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
5205 if ((re_opcode_t) p1[mcnt-3] != jump_past_alt)
5207 /* Get to the beginning of the n-th alternative. */
5213 /* Deal with the last alternative: go back and get number
5214 of the `jump_past_alt' just before it. `mcnt' contains
5215 the length of the alternative. */
5216 EXTRACT_NUMBER (mcnt, p1 - 2);
5218 if (!alt_match_null_string_p (p1, p1 + mcnt, reg_info))
5221 p1 += mcnt; /* Get past the n-th alternative. */
5227 assert (p1[1] == **p);
5233 if (!common_op_match_null_string_p (&p1, end, reg_info))
5236 } /* while p1 < end */
5239 } /* group_match_null_string_p */
5242 /* Similar to group_match_null_string_p, but doesn't deal with alternatives:
5243 It expects P to be the first byte of a single alternative and END one
5244 byte past the last. The alternative can contain groups. */
5247 alt_match_null_string_p (p, end, reg_info)
5248 unsigned char *p, *end;
5249 register_info_type *reg_info;
5252 unsigned char *p1 = p;
5256 /* Skip over opcodes that can match nothing, and break when we get
5257 to one that can't. */
5259 switch ((re_opcode_t) *p1)
5262 case on_failure_jump:
5264 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
5269 if (!common_op_match_null_string_p (&p1, end, reg_info))
5272 } /* while p1 < end */
5275 } /* alt_match_null_string_p */
5278 /* Deals with the ops common to group_match_null_string_p and
5279 alt_match_null_string_p.
5281 Sets P to one after the op and its arguments, if any. */
5284 common_op_match_null_string_p (p, end, reg_info)
5285 unsigned char **p, *end;
5286 register_info_type *reg_info;
5291 unsigned char *p1 = *p;
5293 switch ((re_opcode_t) *p1++)
5313 assert (reg_no > 0 && reg_no <= MAX_REGNUM);
5314 ret = group_match_null_string_p (&p1, end, reg_info);
5316 /* Have to set this here in case we're checking a group which
5317 contains a group and a back reference to it. */
5319 if (REG_MATCH_NULL_STRING_P (reg_info[reg_no]) == MATCH_NULL_UNSET_VALUE)
5320 REG_MATCH_NULL_STRING_P (reg_info[reg_no]) = ret;
5326 /* If this is an optimized succeed_n for zero times, make the jump. */
5328 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
5336 /* Get to the number of times to succeed. */
5338 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
5343 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
5351 if (!REG_MATCH_NULL_STRING_P (reg_info[*p1]))
5359 /* All other opcodes mean we cannot match the empty string. */
5365 } /* common_op_match_null_string_p */
5368 /* Return zero if TRANSLATE[S1] and TRANSLATE[S2] are identical for LEN
5369 bytes; nonzero otherwise. */
5372 bcmp_translate (s1, s2, len, translate)
5373 const char *s1, *s2;
5375 RE_TRANSLATE_TYPE translate;
5377 register const unsigned char *p1 = (const unsigned char *) s1;
5378 register const unsigned char *p2 = (const unsigned char *) s2;
5381 if (translate[*p1++] != translate[*p2++]) return 1;
5387 /* Entry points for GNU code. */
5389 /* re_compile_pattern is the GNU regular expression compiler: it
5390 compiles PATTERN (of length SIZE) and puts the result in BUFP.
5391 Returns 0 if the pattern was valid, otherwise an error string.
5393 Assumes the `allocated' (and perhaps `buffer') and `translate' fields
5394 are set in BUFP on entry.
5396 We call regex_compile to do the actual compilation. */
5399 re_compile_pattern (pattern, length, bufp)
5400 const char *pattern;
5402 struct re_pattern_buffer *bufp;
5406 /* GNU code is written to assume at least RE_NREGS registers will be set
5407 (and at least one extra will be -1). */
5408 bufp->regs_allocated = REGS_UNALLOCATED;
5410 /* And GNU code determines whether or not to get register information
5411 by passing null for the REGS argument to re_match, etc., not by
5415 /* Match anchors at newline. */
5416 bufp->newline_anchor = 1;
5418 ret = regex_compile (pattern, length, re_syntax_options, bufp);
5422 return gettext (re_error_msgid[(int) ret]);
5425 /* Entry points compatible with 4.2 BSD regex library. We don't define
5426 them unless specifically requested. */
5428 #if defined _REGEX_RE_COMP || defined _LIBC
5430 /* BSD has one and only one pattern buffer. */
5431 static struct re_pattern_buffer re_comp_buf;
5435 /* Make these definitions weak in libc, so POSIX programs can redefine
5436 these names if they don't use our functions, and still use
5437 regcomp/regexec below without link errors. */
5447 if (!re_comp_buf.buffer)
5448 return gettext ("No previous regular expression");
5452 if (!re_comp_buf.buffer)
5454 re_comp_buf.buffer = (unsigned char *) malloc (200);
5455 if (re_comp_buf.buffer == NULL)
5456 return gettext (re_error_msgid[(int) REG_ESPACE]);
5457 re_comp_buf.allocated = 200;
5459 re_comp_buf.fastmap = (char *) malloc (1 << BYTEWIDTH);
5460 if (re_comp_buf.fastmap == NULL)
5461 return gettext (re_error_msgid[(int) REG_ESPACE]);
5464 /* Since `re_exec' always passes NULL for the `regs' argument, we
5465 don't need to initialize the pattern buffer fields which affect it. */
5467 /* Match anchors at newlines. */
5468 re_comp_buf.newline_anchor = 1;
5470 ret = regex_compile (s, strlen (s), re_syntax_options, &re_comp_buf);
5475 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
5476 return (char *) gettext (re_error_msgid[(int) ret]);
5487 const int len = strlen (s);
5489 0 <= re_search (&re_comp_buf, s, len, 0, len, (struct re_registers *) 0);
5492 #endif /* _REGEX_RE_COMP */
5494 /* POSIX.2 functions. Don't define these for Emacs. */
5498 /* regcomp takes a regular expression as a string and compiles it.
5500 PREG is a regex_t *. We do not expect any fields to be initialized,
5501 since POSIX says we shouldn't. Thus, we set
5503 `buffer' to the compiled pattern;
5504 `used' to the length of the compiled pattern;
5505 `syntax' to RE_SYNTAX_POSIX_EXTENDED if the
5506 REG_EXTENDED bit in CFLAGS is set; otherwise, to
5507 RE_SYNTAX_POSIX_BASIC;
5508 `newline_anchor' to REG_NEWLINE being set in CFLAGS;
5509 `fastmap' and `fastmap_accurate' to zero;
5510 `re_nsub' to the number of subexpressions in PATTERN.
5512 PATTERN is the address of the pattern string.
5514 CFLAGS is a series of bits which affect compilation.
5516 If REG_EXTENDED is set, we use POSIX extended syntax; otherwise, we
5517 use POSIX basic syntax.
5519 If REG_NEWLINE is set, then . and [^...] don't match newline.
5520 Also, regexec will try a match beginning after every newline.
5522 If REG_ICASE is set, then we considers upper- and lowercase
5523 versions of letters to be equivalent when matching.
5525 If REG_NOSUB is set, then when PREG is passed to regexec, that
5526 routine will report only success or failure, and nothing about the
5529 It returns 0 if it succeeds, nonzero if it doesn't. (See regex.h for
5530 the return codes and their meanings.) */
5533 regcomp (preg, pattern, cflags)
5535 const char *pattern;
5540 = (cflags & REG_EXTENDED) ?
5541 RE_SYNTAX_POSIX_EXTENDED : RE_SYNTAX_POSIX_BASIC;
5543 /* regex_compile will allocate the space for the compiled pattern. */
5545 preg->allocated = 0;
5548 /* Don't bother to use a fastmap when searching. This simplifies the
5549 REG_NEWLINE case: if we used a fastmap, we'd have to put all the
5550 characters after newlines into the fastmap. This way, we just try
5554 if (cflags & REG_ICASE)
5559 = (RE_TRANSLATE_TYPE) malloc (CHAR_SET_SIZE
5560 * sizeof (*(RE_TRANSLATE_TYPE)0));
5561 if (preg->translate == NULL)
5562 return (int) REG_ESPACE;
5564 /* Map uppercase characters to corresponding lowercase ones. */
5565 for (i = 0; i < CHAR_SET_SIZE; i++)
5566 preg->translate[i] = ISUPPER (i) ? tolower (i) : i;
5569 preg->translate = NULL;
5571 /* If REG_NEWLINE is set, newlines are treated differently. */
5572 if (cflags & REG_NEWLINE)
5573 { /* REG_NEWLINE implies neither . nor [^...] match newline. */
5574 syntax &= ~RE_DOT_NEWLINE;
5575 syntax |= RE_HAT_LISTS_NOT_NEWLINE;
5576 /* It also changes the matching behavior. */
5577 preg->newline_anchor = 1;
5580 preg->newline_anchor = 0;
5582 preg->no_sub = !!(cflags & REG_NOSUB);
5584 /* POSIX says a null character in the pattern terminates it, so we
5585 can use strlen here in compiling the pattern. */
5586 ret = regex_compile (pattern, strlen (pattern), syntax, preg);
5588 /* POSIX doesn't distinguish between an unmatched open-group and an
5589 unmatched close-group: both are REG_EPAREN. */
5590 if (ret == REG_ERPAREN) ret = REG_EPAREN;
5596 /* regexec searches for a given pattern, specified by PREG, in the
5599 If NMATCH is zero or REG_NOSUB was set in the cflags argument to
5600 `regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at
5601 least NMATCH elements, and we set them to the offsets of the
5602 corresponding matched substrings.
5604 EFLAGS specifies `execution flags' which affect matching: if
5605 REG_NOTBOL is set, then ^ does not match at the beginning of the
5606 string; if REG_NOTEOL is set, then $ does not match at the end.
5608 We return 0 if we find a match and REG_NOMATCH if not. */
5611 regexec (preg, string, nmatch, pmatch, eflags)
5612 const regex_t *preg;
5615 regmatch_t pmatch[];
5619 struct re_registers regs;
5620 regex_t private_preg;
5621 int len = strlen (string);
5622 boolean want_reg_info = !preg->no_sub && nmatch > 0;
5624 private_preg = *preg;
5626 private_preg.not_bol = !!(eflags & REG_NOTBOL);
5627 private_preg.not_eol = !!(eflags & REG_NOTEOL);
5629 /* The user has told us exactly how many registers to return
5630 information about, via `nmatch'. We have to pass that on to the
5631 matching routines. */
5632 private_preg.regs_allocated = REGS_FIXED;
5636 regs.num_regs = nmatch;
5637 regs.start = TALLOC (nmatch, regoff_t);
5638 regs.end = TALLOC (nmatch, regoff_t);
5639 if (regs.start == NULL || regs.end == NULL)
5640 return (int) REG_NOMATCH;
5643 /* Perform the searching operation. */
5644 ret = re_search (&private_preg, string, len,
5645 /* start: */ 0, /* range: */ len,
5646 want_reg_info ? ®s : (struct re_registers *) 0);
5648 /* Copy the register information to the POSIX structure. */
5655 for (r = 0; r < nmatch; r++)
5657 pmatch[r].rm_so = regs.start[r];
5658 pmatch[r].rm_eo = regs.end[r];
5662 /* If we needed the temporary register info, free the space now. */
5667 /* We want zero return to mean success, unlike `re_search'. */
5668 return ret >= 0 ? (int) REG_NOERROR : (int) REG_NOMATCH;
5672 /* Returns a message corresponding to an error code, ERRCODE, returned
5673 from either regcomp or regexec. We don't use PREG here. */
5676 regerror (errcode, preg, errbuf, errbuf_size)
5678 const regex_t *preg;
5686 || errcode >= (int) (sizeof (re_error_msgid)
5687 / sizeof (re_error_msgid[0])))
5688 /* Only error codes returned by the rest of the code should be passed
5689 to this routine. If we are given anything else, or if other regex
5690 code generates an invalid error code, then the program has a bug.
5691 Dump core so we can fix it. */
5694 msg = gettext (re_error_msgid[errcode]);
5696 msg_size = strlen (msg) + 1; /* Includes the null. */
5698 if (errbuf_size != 0)
5700 if (msg_size > errbuf_size)
5702 #if defined HAVE_MEMPCPY || defined _LIBC
5703 *((char *) __mempcpy (errbuf, msg, errbuf_size - 1)) = '\0';
5705 memcpy (errbuf, msg, errbuf_size - 1);
5706 errbuf[errbuf_size - 1] = 0;
5710 memcpy (errbuf, msg, msg_size);
5717 /* Free dynamically allocated space used by PREG. */
5723 if (preg->buffer != NULL)
5724 free (preg->buffer);
5725 preg->buffer = NULL;
5727 preg->allocated = 0;
5730 if (preg->fastmap != NULL)
5731 free (preg->fastmap);
5732 preg->fastmap = NULL;
5733 preg->fastmap_accurate = 0;
5735 if (preg->translate != NULL)
5736 free (preg->translate);
5737 preg->translate = NULL;
5740 #endif /* not emacs */