1 /* Extended regular expression matching and search library,
3 (Implements POSIX draft P10003.2/D11.2, except for
4 internationalization features.)
6 Copyright (C) 1993 Free Software Foundation, Inc.
8 This program is free software; you can redistribute it and/or modify
9 it under the terms of the GNU General Public License as published by
10 the Free Software Foundation; either version 2, or (at your option)
13 This program is distributed in the hope that it will be useful,
14 but WITHOUT ANY WARRANTY; without even the implied warranty of
15 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
16 GNU General Public License for more details.
18 You should have received a copy of the GNU General Public License
19 along with this program; if not, write to the Free Software
20 Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. */
22 /* AIX requires this to be the first thing in the file. */
23 #if defined (_AIX) && !defined (REGEX_MALLOC)
30 #if defined (emacs) || defined (CONFIG_BROKETS)
31 /* We use <config.h> instead of "config.h" so that a compilation
32 using -I. -I$srcdir will use ./config.h rather than $srcdir/config.h
33 (which it would do because it found this file in $srcdir). */
40 /* We need this for `regex.h', and perhaps for the Emacs include files. */
41 #include <sys/types.h>
43 /* The `emacs' switch turns on certain matching commands
44 that make sense only in Emacs. */
51 /* Emacs uses `NULL' as a predicate. */
64 /* We used to test for `BSTRING' here, but only GCC and Emacs define
65 `BSTRING', as far as I know, and neither of them use this code. */
66 #ifndef INHIBIT_STRING_HEADER
67 #if HAVE_STRING_H || STDC_HEADERS
70 #define bcmp(s1, s2, n) memcmp ((s1), (s2), (n))
73 #define bcopy(s, d, n) memcpy ((d), (s), (n))
76 #define bzero(s, n) memset ((s), 0, (n))
83 /* Define the syntax stuff for \<, \>, etc. */
85 /* This must be nonzero for the wordchar and notwordchar pattern
86 commands in re_match_2. */
93 extern char *re_syntax_table;
95 #else /* not SYNTAX_TABLE */
97 /* How many characters in the character set. */
98 #define CHAR_SET_SIZE 256
100 static char re_syntax_table[CHAR_SET_SIZE];
111 bzero (re_syntax_table, sizeof re_syntax_table);
113 for (c = 'a'; c <= 'z'; c++)
114 re_syntax_table[c] = Sword;
116 for (c = 'A'; c <= 'Z'; c++)
117 re_syntax_table[c] = Sword;
119 for (c = '0'; c <= '9'; c++)
120 re_syntax_table[c] = Sword;
122 re_syntax_table['_'] = Sword;
127 #endif /* not SYNTAX_TABLE */
129 #define SYNTAX(c) re_syntax_table[c]
131 #endif /* not emacs */
133 /* Get the interface, including the syntax bits. */
136 /* isalpha etc. are used for the character classes. */
139 /* Jim Meyering writes:
141 "... Some ctype macros are valid only for character codes that
142 isascii says are ASCII (SGI's IRIX-4.0.5 is one such system --when
143 using /bin/cc or gcc but without giving an ansi option). So, all
144 ctype uses should be through macros like ISPRINT... If
145 STDC_HEADERS is defined, then autoconf has verified that the ctype
146 macros don't need to be guarded with references to isascii. ...
147 Defining isascii to 1 should let any compiler worth its salt
148 eliminate the && through constant folding." */
150 #if defined (STDC_HEADERS) || (!defined (isascii) && !defined (HAVE_ISASCII))
153 #define ISASCII(c) isascii(c)
157 #define ISBLANK(c) (ISASCII (c) && isblank (c))
159 #define ISBLANK(c) ((c) == ' ' || (c) == '\t')
162 #define ISGRAPH(c) (ISASCII (c) && isgraph (c))
164 #define ISGRAPH(c) (ISASCII (c) && isprint (c) && !isspace (c))
167 #define ISPRINT(c) (ISASCII (c) && isprint (c))
168 #define ISDIGIT(c) (ISASCII (c) && isdigit (c))
169 #define ISALNUM(c) (ISASCII (c) && isalnum (c))
170 #define ISALPHA(c) (ISASCII (c) && isalpha (c))
171 #define ISCNTRL(c) (ISASCII (c) && iscntrl (c))
172 #define ISLOWER(c) (ISASCII (c) && islower (c))
173 #define ISPUNCT(c) (ISASCII (c) && ispunct (c))
174 #define ISSPACE(c) (ISASCII (c) && isspace (c))
175 #define ISUPPER(c) (ISASCII (c) && isupper (c))
176 #define ISXDIGIT(c) (ISASCII (c) && isxdigit (c))
182 /* We remove any previous definition of `SIGN_EXTEND_CHAR',
183 since ours (we hope) works properly with all combinations of
184 machines, compilers, `char' and `unsigned char' argument types.
185 (Per Bothner suggested the basic approach.) */
186 #undef SIGN_EXTEND_CHAR
188 #define SIGN_EXTEND_CHAR(c) ((signed char) (c))
189 #else /* not __STDC__ */
190 /* As in Harbison and Steele. */
191 #define SIGN_EXTEND_CHAR(c) ((((unsigned char) (c)) ^ 128) - 128)
194 /* Should we use malloc or alloca? If REGEX_MALLOC is not defined, we
195 use `alloca' instead of `malloc'. This is because using malloc in
196 re_search* or re_match* could cause memory leaks when C-g is used in
197 Emacs; also, malloc is slower and causes storage fragmentation. On
198 the other hand, malloc is more portable, and easier to debug.
200 Because we sometimes use alloca, some routines have to be macros,
201 not functions -- `alloca'-allocated space disappears at the end of the
202 function it is called in. */
206 #define REGEX_ALLOCATE malloc
207 #define REGEX_REALLOCATE(source, osize, nsize) realloc (source, nsize)
209 #else /* not REGEX_MALLOC */
211 /* Emacs already defines alloca, sometimes. */
214 /* Make alloca work the best possible way. */
216 #define alloca __builtin_alloca
217 #else /* not __GNUC__ */
220 #else /* not __GNUC__ or HAVE_ALLOCA_H */
221 #ifndef _AIX /* Already did AIX, up at the top. */
223 #endif /* not _AIX */
224 #endif /* not HAVE_ALLOCA_H */
225 #endif /* not __GNUC__ */
227 #endif /* not alloca */
229 #define REGEX_ALLOCATE alloca
231 /* Assumes a `char *destination' variable. */
232 #define REGEX_REALLOCATE(source, osize, nsize) \
233 (destination = (char *) alloca (nsize), \
234 bcopy (source, destination, osize), \
237 #endif /* not REGEX_MALLOC */
240 /* True if `size1' is non-NULL and PTR is pointing anywhere inside
241 `string1' or just past its end. This works if PTR is NULL, which is
243 #define FIRST_STRING_P(ptr) \
244 (size1 && string1 <= (ptr) && (ptr) <= string1 + size1)
246 /* (Re)Allocate N items of type T using malloc, or fail. */
247 #define TALLOC(n, t) ((t *) malloc ((n) * sizeof (t)))
248 #define RETALLOC(addr, n, t) ((addr) = (t *) realloc (addr, (n) * sizeof (t)))
249 #define RETALLOC_IF(addr, n, t) \
250 if (addr) RETALLOC((addr), (n), t); else (addr) = TALLOC ((n), t)
251 #define REGEX_TALLOC(n, t) ((t *) REGEX_ALLOCATE ((n) * sizeof (t)))
253 #define BYTEWIDTH 8 /* In bits. */
255 #define STREQ(s1, s2) ((strcmp (s1, s2) == 0))
259 #define MAX(a, b) ((a) > (b) ? (a) : (b))
260 #define MIN(a, b) ((a) < (b) ? (a) : (b))
262 typedef char boolean;
266 static int re_match_2_internal ();
268 /* These are the command codes that appear in compiled regular
269 expressions. Some opcodes are followed by argument bytes. A
270 command code can specify any interpretation whatsoever for its
271 arguments. Zero bytes may appear in the compiled regular expression.
273 The value of `exactn' is needed in search.c (search_buffer) in Emacs.
274 So regex.h defines a symbol `RE_EXACTN_VALUE' to be 1; the value of
275 `exactn' we use here must also be 1. */
281 /* Followed by one byte giving n, then by n literal bytes. */
284 /* Matches any (more or less) character. */
287 /* Matches any one char belonging to specified set. First
288 following byte is number of bitmap bytes. Then come bytes
289 for a bitmap saying which chars are in. Bits in each byte
290 are ordered low-bit-first. A character is in the set if its
291 bit is 1. A character too large to have a bit in the map is
292 automatically not in the set. */
295 /* Same parameters as charset, but match any character that is
296 not one of those specified. */
299 /* Start remembering the text that is matched, for storing in a
300 register. Followed by one byte with the register number, in
301 the range 0 to one less than the pattern buffer's re_nsub
302 field. Then followed by one byte with the number of groups
303 inner to this one. (This last has to be part of the
304 start_memory only because we need it in the on_failure_jump
308 /* Stop remembering the text that is matched and store it in a
309 memory register. Followed by one byte with the register
310 number, in the range 0 to one less than `re_nsub' in the
311 pattern buffer, and one byte with the number of inner groups,
312 just like `start_memory'. (We need the number of inner
313 groups here because we don't have any easy way of finding the
314 corresponding start_memory when we're at a stop_memory.) */
317 /* Match a duplicate of something remembered. Followed by one
318 byte containing the register number. */
321 /* Fail unless at beginning of line. */
324 /* Fail unless at end of line. */
327 /* Succeeds if at beginning of buffer (if emacs) or at beginning
328 of string to be matched (if not). */
331 /* Analogously, for end of buffer/string. */
334 /* Followed by two byte relative address to which to jump. */
337 /* Same as jump, but marks the end of an alternative. */
340 /* Followed by two-byte relative address of place to resume at
341 in case of failure. */
344 /* Like on_failure_jump, but pushes a placeholder instead of the
345 current string position when executed. */
346 on_failure_keep_string_jump,
348 /* Throw away latest failure point and then jump to following
349 two-byte relative address. */
352 /* Change to pop_failure_jump if know won't have to backtrack to
353 match; otherwise change to jump. This is used to jump
354 back to the beginning of a repeat. If what follows this jump
355 clearly won't match what the repeat does, such that we can be
356 sure that there is no use backtracking out of repetitions
357 already matched, then we change it to a pop_failure_jump.
358 Followed by two-byte address. */
361 /* Jump to following two-byte address, and push a dummy failure
362 point. This failure point will be thrown away if an attempt
363 is made to use it for a failure. A `+' construct makes this
364 before the first repeat. Also used as an intermediary kind
365 of jump when compiling an alternative. */
368 /* Push a dummy failure point and continue. Used at the end of
372 /* Followed by two-byte relative address and two-byte number n.
373 After matching N times, jump to the address upon failure. */
376 /* Followed by two-byte relative address, and two-byte number n.
377 Jump to the address N times, then fail. */
380 /* Set the following two-byte relative address to the
381 subsequent two-byte number. The address *includes* the two
385 wordchar, /* Matches any word-constituent character. */
386 notwordchar, /* Matches any char that is not a word-constituent. */
388 wordbeg, /* Succeeds if at word beginning. */
389 wordend, /* Succeeds if at word end. */
391 wordbound, /* Succeeds if at a word boundary. */
392 notwordbound /* Succeeds if not at a word boundary. */
395 ,before_dot, /* Succeeds if before point. */
396 at_dot, /* Succeeds if at point. */
397 after_dot, /* Succeeds if after point. */
399 /* Matches any character whose syntax is specified. Followed by
400 a byte which contains a syntax code, e.g., Sword. */
403 /* Matches any character whose syntax is not that specified. */
408 /* Common operations on the compiled pattern. */
410 /* Store NUMBER in two contiguous bytes starting at DESTINATION. */
412 #define STORE_NUMBER(destination, number) \
414 (destination)[0] = (number) & 0377; \
415 (destination)[1] = (number) >> 8; \
418 /* Same as STORE_NUMBER, except increment DESTINATION to
419 the byte after where the number is stored. Therefore, DESTINATION
420 must be an lvalue. */
422 #define STORE_NUMBER_AND_INCR(destination, number) \
424 STORE_NUMBER (destination, number); \
425 (destination) += 2; \
428 /* Put into DESTINATION a number stored in two contiguous bytes starting
431 #define EXTRACT_NUMBER(destination, source) \
433 (destination) = *(source) & 0377; \
434 (destination) += SIGN_EXTEND_CHAR (*((source) + 1)) << 8; \
439 extract_number (dest, source)
441 unsigned char *source;
443 int temp = SIGN_EXTEND_CHAR (*(source + 1));
444 *dest = *source & 0377;
448 #ifndef EXTRACT_MACROS /* To debug the macros. */
449 #undef EXTRACT_NUMBER
450 #define EXTRACT_NUMBER(dest, src) extract_number (&dest, src)
451 #endif /* not EXTRACT_MACROS */
455 /* Same as EXTRACT_NUMBER, except increment SOURCE to after the number.
456 SOURCE must be an lvalue. */
458 #define EXTRACT_NUMBER_AND_INCR(destination, source) \
460 EXTRACT_NUMBER (destination, source); \
466 extract_number_and_incr (destination, source)
468 unsigned char **source;
470 extract_number (destination, *source);
474 #ifndef EXTRACT_MACROS
475 #undef EXTRACT_NUMBER_AND_INCR
476 #define EXTRACT_NUMBER_AND_INCR(dest, src) \
477 extract_number_and_incr (&dest, &src)
478 #endif /* not EXTRACT_MACROS */
482 /* If DEBUG is defined, Regex prints many voluminous messages about what
483 it is doing (if the variable `debug' is nonzero). If linked with the
484 main program in `iregex.c', you can enter patterns and strings
485 interactively. And if linked with the main program in `main.c' and
486 the other test files, you can run the already-written tests. */
490 /* We use standard I/O for debugging. */
493 /* It is useful to test things that ``must'' be true when debugging. */
496 static int debug = 0;
498 #define DEBUG_STATEMENT(e) e
499 #define DEBUG_PRINT1(x) if (debug) printf (x)
500 #define DEBUG_PRINT2(x1, x2) if (debug) printf (x1, x2)
501 #define DEBUG_PRINT3(x1, x2, x3) if (debug) printf (x1, x2, x3)
502 #define DEBUG_PRINT4(x1, x2, x3, x4) if (debug) printf (x1, x2, x3, x4)
503 #define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) \
504 if (debug) print_partial_compiled_pattern (s, e)
505 #define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) \
506 if (debug) print_double_string (w, s1, sz1, s2, sz2)
509 extern void printchar ();
511 /* Print the fastmap in human-readable form. */
514 print_fastmap (fastmap)
517 unsigned was_a_range = 0;
520 while (i < (1 << BYTEWIDTH))
526 while (i < (1 << BYTEWIDTH) && fastmap[i])
542 /* Print a compiled pattern string in human-readable form, starting at
543 the START pointer into it and ending just before the pointer END. */
546 print_partial_compiled_pattern (start, end)
547 unsigned char *start;
551 unsigned char *p = start;
552 unsigned char *pend = end;
560 /* Loop over pattern commands. */
563 printf ("%d:\t", p - start);
565 switch ((re_opcode_t) *p++)
573 printf ("/exactn/%d", mcnt);
584 printf ("/start_memory/%d/%d", mcnt, *p++);
589 printf ("/stop_memory/%d/%d", mcnt, *p++);
593 printf ("/duplicate/%d", *p++);
603 register int c, last = -100;
604 register int in_range = 0;
606 printf ("/charset [%s",
607 (re_opcode_t) *(p - 1) == charset_not ? "^" : "");
609 assert (p + *p < pend);
611 for (c = 0; c < 256; c++)
613 && (p[1 + (c/8)] & (1 << (c % 8))))
615 /* Are we starting a range? */
616 if (last + 1 == c && ! in_range)
621 /* Have we broken a range? */
622 else if (last + 1 != c && in_range)
651 case on_failure_jump:
652 extract_number_and_incr (&mcnt, &p);
653 printf ("/on_failure_jump to %d", p + mcnt - start);
656 case on_failure_keep_string_jump:
657 extract_number_and_incr (&mcnt, &p);
658 printf ("/on_failure_keep_string_jump to %d", p + mcnt - start);
661 case dummy_failure_jump:
662 extract_number_and_incr (&mcnt, &p);
663 printf ("/dummy_failure_jump to %d", p + mcnt - start);
666 case push_dummy_failure:
667 printf ("/push_dummy_failure");
671 extract_number_and_incr (&mcnt, &p);
672 printf ("/maybe_pop_jump to %d", p + mcnt - start);
675 case pop_failure_jump:
676 extract_number_and_incr (&mcnt, &p);
677 printf ("/pop_failure_jump to %d", p + mcnt - start);
681 extract_number_and_incr (&mcnt, &p);
682 printf ("/jump_past_alt to %d", p + mcnt - start);
686 extract_number_and_incr (&mcnt, &p);
687 printf ("/jump to %d", p + mcnt - start);
691 extract_number_and_incr (&mcnt, &p);
692 extract_number_and_incr (&mcnt2, &p);
693 printf ("/succeed_n to %d, %d times", p + mcnt - start, mcnt2);
697 extract_number_and_incr (&mcnt, &p);
698 extract_number_and_incr (&mcnt2, &p);
699 printf ("/jump_n to %d, %d times", p + mcnt - start, mcnt2);
703 extract_number_and_incr (&mcnt, &p);
704 extract_number_and_incr (&mcnt2, &p);
705 printf ("/set_number_at location %d to %d", p + mcnt - start, mcnt2);
709 printf ("/wordbound");
713 printf ("/notwordbound");
725 printf ("/before_dot");
733 printf ("/after_dot");
737 printf ("/syntaxspec");
739 printf ("/%d", mcnt);
743 printf ("/notsyntaxspec");
745 printf ("/%d", mcnt);
750 printf ("/wordchar");
754 printf ("/notwordchar");
766 printf ("?%d", *(p-1));
772 printf ("%d:\tend of pattern.\n", p - start);
777 print_compiled_pattern (bufp)
778 struct re_pattern_buffer *bufp;
780 unsigned char *buffer = bufp->buffer;
782 print_partial_compiled_pattern (buffer, buffer + bufp->used);
783 printf ("%ld bytes used/%ld bytes allocated.\n", bufp->used, bufp->allocated);
785 if (bufp->fastmap_accurate && bufp->fastmap)
787 printf ("fastmap: ");
788 print_fastmap (bufp->fastmap);
791 printf ("re_nsub: %d\t", bufp->re_nsub);
792 printf ("regs_alloc: %d\t", bufp->regs_allocated);
793 printf ("can_be_null: %d\t", bufp->can_be_null);
794 printf ("newline_anchor: %d\n", bufp->newline_anchor);
795 printf ("no_sub: %d\t", bufp->no_sub);
796 printf ("not_bol: %d\t", bufp->not_bol);
797 printf ("not_eol: %d\t", bufp->not_eol);
798 printf ("syntax: %d\n", bufp->syntax);
799 /* Perhaps we should print the translate table? */
804 print_double_string (where, string1, size1, string2, size2)
817 if (FIRST_STRING_P (where))
819 for (this_char = where - string1; this_char < size1; this_char++)
820 printchar (string1[this_char]);
825 for (this_char = where - string2; this_char < size2; this_char++)
826 printchar (string2[this_char]);
830 #else /* not DEBUG */
835 #define DEBUG_STATEMENT(e)
836 #define DEBUG_PRINT1(x)
837 #define DEBUG_PRINT2(x1, x2)
838 #define DEBUG_PRINT3(x1, x2, x3)
839 #define DEBUG_PRINT4(x1, x2, x3, x4)
840 #define DEBUG_PRINT_COMPILED_PATTERN(p, s, e)
841 #define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2)
843 #endif /* not DEBUG */
845 /* Set by `re_set_syntax' to the current regexp syntax to recognize. Can
846 also be assigned to arbitrarily: each pattern buffer stores its own
847 syntax, so it can be changed between regex compilations. */
848 reg_syntax_t re_syntax_options = RE_SYNTAX_EMACS;
851 /* Specify the precise syntax of regexps for compilation. This provides
852 for compatibility for various utilities which historically have
853 different, incompatible syntaxes.
855 The argument SYNTAX is a bit mask comprised of the various bits
856 defined in regex.h. We return the old syntax. */
859 re_set_syntax (syntax)
862 reg_syntax_t ret = re_syntax_options;
864 re_syntax_options = syntax;
868 /* This table gives an error message for each of the error codes listed
869 in regex.h. Obviously the order here has to be same as there. */
871 static const char *re_error_msg[] =
872 { NULL, /* REG_NOERROR */
873 "No match", /* REG_NOMATCH */
874 "Invalid regular expression", /* REG_BADPAT */
875 "Invalid collation character", /* REG_ECOLLATE */
876 "Invalid character class name", /* REG_ECTYPE */
877 "Trailing backslash", /* REG_EESCAPE */
878 "Invalid back reference", /* REG_ESUBREG */
879 "Unmatched [ or [^", /* REG_EBRACK */
880 "Unmatched ( or \\(", /* REG_EPAREN */
881 "Unmatched \\{", /* REG_EBRACE */
882 "Invalid content of \\{\\}", /* REG_BADBR */
883 "Invalid range end", /* REG_ERANGE */
884 "Memory exhausted", /* REG_ESPACE */
885 "Invalid preceding regular expression", /* REG_BADRPT */
886 "Premature end of regular expression", /* REG_EEND */
887 "Regular expression too big", /* REG_ESIZE */
888 "Unmatched ) or \\)", /* REG_ERPAREN */
891 /* Avoiding alloca during matching, to placate r_alloc. */
893 /* Define MATCH_MAY_ALLOCATE unless we need to make sure that the
894 searching and matching functions should not call alloca. On some
895 systems, alloca is implemented in terms of malloc, and if we're
896 using the relocating allocator routines, then malloc could cause a
897 relocation, which might (if the strings being searched are in the
898 ralloc heap) shift the data out from underneath the regexp
901 Here's another reason to avoid allocation: Emacs insists on
902 processing input from X in a signal handler; processing X input may
903 call malloc; if input arrives while a matching routine is calling
904 malloc, then we're scrod. But Emacs can't just block input while
905 calling matching routines; then we don't notice interrupts when
906 they come in. So, Emacs blocks input around all regexp calls
907 except the matching calls, which it leaves unprotected, in the
908 faith that they will not malloc. */
910 /* Normally, this is fine. */
911 #define MATCH_MAY_ALLOCATE
913 /* But under some circumstances, it's not. */
914 #if defined (emacs) || (defined (REL_ALLOC) && defined (C_ALLOCA))
915 #undef MATCH_MAY_ALLOCATE
919 /* Failure stack declarations and macros; both re_compile_fastmap and
920 re_match_2 use a failure stack. These have to be macros because of
924 /* Number of failure points for which to initially allocate space
925 when matching. If this number is exceeded, we allocate more
926 space, so it is not a hard limit. */
927 #ifndef INIT_FAILURE_ALLOC
928 #define INIT_FAILURE_ALLOC 5
931 /* Roughly the maximum number of failure points on the stack. Would be
932 exactly that if always used MAX_FAILURE_SPACE each time we failed.
933 This is a variable only so users of regex can assign to it; we never
934 change it ourselves. */
935 int re_max_failures = 2000;
937 typedef unsigned char *fail_stack_elt_t;
941 fail_stack_elt_t *stack;
943 unsigned avail; /* Offset of next open position. */
946 #define FAIL_STACK_EMPTY() (fail_stack.avail == 0)
947 #define FAIL_STACK_PTR_EMPTY() (fail_stack_ptr->avail == 0)
948 #define FAIL_STACK_FULL() (fail_stack.avail == fail_stack.size)
949 #define FAIL_STACK_TOP() (fail_stack.stack[fail_stack.avail])
952 /* Initialize `fail_stack'. Do `return -2' if the alloc fails. */
954 #ifdef MATCH_MAY_ALLOCATE
955 #define INIT_FAIL_STACK() \
957 fail_stack.stack = (fail_stack_elt_t *) \
958 REGEX_ALLOCATE (INIT_FAILURE_ALLOC * sizeof (fail_stack_elt_t)); \
960 if (fail_stack.stack == NULL) \
963 fail_stack.size = INIT_FAILURE_ALLOC; \
964 fail_stack.avail = 0; \
967 #define INIT_FAIL_STACK() \
969 fail_stack.avail = 0; \
974 /* Double the size of FAIL_STACK, up to approximately `re_max_failures' items.
976 Return 1 if succeeds, and 0 if either ran out of memory
977 allocating space for it or it was already too large.
979 REGEX_REALLOCATE requires `destination' be declared. */
981 #define DOUBLE_FAIL_STACK(fail_stack) \
982 ((fail_stack).size > re_max_failures * MAX_FAILURE_ITEMS \
984 : ((fail_stack).stack = (fail_stack_elt_t *) \
985 REGEX_REALLOCATE ((fail_stack).stack, \
986 (fail_stack).size * sizeof (fail_stack_elt_t), \
987 ((fail_stack).size << 1) * sizeof (fail_stack_elt_t)), \
989 (fail_stack).stack == NULL \
991 : ((fail_stack).size <<= 1, \
995 /* Push PATTERN_OP on FAIL_STACK.
997 Return 1 if was able to do so and 0 if ran out of memory allocating
999 #define PUSH_PATTERN_OP(pattern_op, fail_stack) \
1000 ((FAIL_STACK_FULL () \
1001 && !DOUBLE_FAIL_STACK (fail_stack)) \
1003 : ((fail_stack).stack[(fail_stack).avail++] = pattern_op, \
1006 /* This pushes an item onto the failure stack. sizeof(ITEM) must be no
1007 larger than sizeof (unsigned char *). Assumes the variable `fail_stack'.
1008 Probably should only be called from within `PUSH_FAILURE_POINT'. */
1009 #define PUSH_FAILURE_ITEM(item) \
1012 fail_stack.stack[fail_stack.avail++] = (fail_stack_elt_t) item; \
1016 /* The complement operation. Assumes `fail_stack' is nonempty. */
1017 #define POP_FAILURE_ITEM() fail_stack.stack[--fail_stack.avail]
1019 /* Used to omit pushing failure point id's when we're not debugging. */
1021 #define DEBUG_PUSH PUSH_FAILURE_ITEM
1022 #define DEBUG_POP(item_addr) *(item_addr) = POP_FAILURE_ITEM ()
1024 #define DEBUG_PUSH(item)
1025 #define DEBUG_POP(item_addr)
1029 /* Push the information about the state we will need
1030 if we ever fail back to it.
1032 Requires variables fail_stack, regstart, regend, reg_info, and
1033 num_regs be declared. DOUBLE_FAIL_STACK requires `destination' be
1036 Does `return FAILURE_CODE' if runs out of memory. */
1038 #define PUSH_FAILURE_POINT(pattern_place, string_place, failure_code) \
1040 char *destination; \
1041 /* Must be int, so when we don't save any registers, the arithmetic \
1042 of 0 + -1 isn't done as unsigned. */ \
1045 DEBUG_STATEMENT (failure_id++); \
1046 DEBUG_STATEMENT (nfailure_points_pushed++); \
1047 DEBUG_PRINT2 ("\nPUSH_FAILURE_POINT #%u:\n", failure_id); \
1048 DEBUG_PRINT2 (" Before push, next avail: %d\n", (fail_stack).avail);\
1049 DEBUG_PRINT2 (" size: %d\n", (fail_stack).size);\
1051 DEBUG_PRINT2 (" slots needed: %d\n", NUM_FAILURE_ITEMS); \
1052 DEBUG_PRINT2 (" available: %d\n", REMAINING_AVAIL_SLOTS); \
1054 /* Ensure we have enough space allocated for what we will push. */ \
1055 while (REMAINING_AVAIL_SLOTS < NUM_FAILURE_ITEMS) \
1057 if (!DOUBLE_FAIL_STACK (fail_stack)) \
1058 return failure_code; \
1060 DEBUG_PRINT2 ("\n Doubled stack; size now: %d\n", \
1061 (fail_stack).size); \
1062 DEBUG_PRINT2 (" slots available: %d\n", REMAINING_AVAIL_SLOTS);\
1065 /* Push the info, starting with the registers. */ \
1066 DEBUG_PRINT1 ("\n"); \
1068 for (this_reg = lowest_active_reg; this_reg <= highest_active_reg; \
1071 DEBUG_PRINT2 (" Pushing reg: %d\n", this_reg); \
1072 DEBUG_STATEMENT (num_regs_pushed++); \
1074 DEBUG_PRINT2 (" start: 0x%x\n", regstart[this_reg]); \
1075 PUSH_FAILURE_ITEM (regstart[this_reg]); \
1077 DEBUG_PRINT2 (" end: 0x%x\n", regend[this_reg]); \
1078 PUSH_FAILURE_ITEM (regend[this_reg]); \
1080 DEBUG_PRINT2 (" info: 0x%x\n ", reg_info[this_reg]); \
1081 DEBUG_PRINT2 (" match_null=%d", \
1082 REG_MATCH_NULL_STRING_P (reg_info[this_reg])); \
1083 DEBUG_PRINT2 (" active=%d", IS_ACTIVE (reg_info[this_reg])); \
1084 DEBUG_PRINT2 (" matched_something=%d", \
1085 MATCHED_SOMETHING (reg_info[this_reg])); \
1086 DEBUG_PRINT2 (" ever_matched=%d", \
1087 EVER_MATCHED_SOMETHING (reg_info[this_reg])); \
1088 DEBUG_PRINT1 ("\n"); \
1089 PUSH_FAILURE_ITEM (reg_info[this_reg].word); \
1092 DEBUG_PRINT2 (" Pushing low active reg: %d\n", lowest_active_reg);\
1093 PUSH_FAILURE_ITEM (lowest_active_reg); \
1095 DEBUG_PRINT2 (" Pushing high active reg: %d\n", highest_active_reg);\
1096 PUSH_FAILURE_ITEM (highest_active_reg); \
1098 DEBUG_PRINT2 (" Pushing pattern 0x%x: ", pattern_place); \
1099 DEBUG_PRINT_COMPILED_PATTERN (bufp, pattern_place, pend); \
1100 PUSH_FAILURE_ITEM (pattern_place); \
1102 DEBUG_PRINT2 (" Pushing string 0x%x: `", string_place); \
1103 DEBUG_PRINT_DOUBLE_STRING (string_place, string1, size1, string2, \
1105 DEBUG_PRINT1 ("'\n"); \
1106 PUSH_FAILURE_ITEM (string_place); \
1108 DEBUG_PRINT2 (" Pushing failure id: %u\n", failure_id); \
1109 DEBUG_PUSH (failure_id); \
1112 /* This is the number of items that are pushed and popped on the stack
1113 for each register. */
1114 #define NUM_REG_ITEMS 3
1116 /* Individual items aside from the registers. */
1118 #define NUM_NONREG_ITEMS 5 /* Includes failure point id. */
1120 #define NUM_NONREG_ITEMS 4
1123 /* We push at most this many items on the stack. */
1124 #define MAX_FAILURE_ITEMS ((num_regs - 1) * NUM_REG_ITEMS + NUM_NONREG_ITEMS)
1126 /* We actually push this many items. */
1127 #define NUM_FAILURE_ITEMS \
1128 ((highest_active_reg - lowest_active_reg + 1) * NUM_REG_ITEMS \
1131 /* How many items can still be added to the stack without overflowing it. */
1132 #define REMAINING_AVAIL_SLOTS ((fail_stack).size - (fail_stack).avail)
1135 /* Pops what PUSH_FAIL_STACK pushes.
1137 We restore into the parameters, all of which should be lvalues:
1138 STR -- the saved data position.
1139 PAT -- the saved pattern position.
1140 LOW_REG, HIGH_REG -- the highest and lowest active registers.
1141 REGSTART, REGEND -- arrays of string positions.
1142 REG_INFO -- array of information about each subexpression.
1144 Also assumes the variables `fail_stack' and (if debugging), `bufp',
1145 `pend', `string1', `size1', `string2', and `size2'. */
1147 #define POP_FAILURE_POINT(str, pat, low_reg, high_reg, regstart, regend, reg_info)\
1149 DEBUG_STATEMENT (fail_stack_elt_t failure_id;) \
1151 const unsigned char *string_temp; \
1153 assert (!FAIL_STACK_EMPTY ()); \
1155 /* Remove failure points and point to how many regs pushed. */ \
1156 DEBUG_PRINT1 ("POP_FAILURE_POINT:\n"); \
1157 DEBUG_PRINT2 (" Before pop, next avail: %d\n", fail_stack.avail); \
1158 DEBUG_PRINT2 (" size: %d\n", fail_stack.size); \
1160 assert (fail_stack.avail >= NUM_NONREG_ITEMS); \
1162 DEBUG_POP (&failure_id); \
1163 DEBUG_PRINT2 (" Popping failure id: %u\n", failure_id); \
1165 /* If the saved string location is NULL, it came from an \
1166 on_failure_keep_string_jump opcode, and we want to throw away the \
1167 saved NULL, thus retaining our current position in the string. */ \
1168 string_temp = POP_FAILURE_ITEM (); \
1169 if (string_temp != NULL) \
1170 str = (const char *) string_temp; \
1172 DEBUG_PRINT2 (" Popping string 0x%x: `", str); \
1173 DEBUG_PRINT_DOUBLE_STRING (str, string1, size1, string2, size2); \
1174 DEBUG_PRINT1 ("'\n"); \
1176 pat = (unsigned char *) POP_FAILURE_ITEM (); \
1177 DEBUG_PRINT2 (" Popping pattern 0x%x: ", pat); \
1178 DEBUG_PRINT_COMPILED_PATTERN (bufp, pat, pend); \
1180 /* Restore register info. */ \
1181 high_reg = (unsigned long) POP_FAILURE_ITEM (); \
1182 DEBUG_PRINT2 (" Popping high active reg: %d\n", high_reg); \
1184 low_reg = (unsigned long) POP_FAILURE_ITEM (); \
1185 DEBUG_PRINT2 (" Popping low active reg: %d\n", low_reg); \
1187 for (this_reg = high_reg; this_reg >= low_reg; this_reg--) \
1189 DEBUG_PRINT2 (" Popping reg: %d\n", this_reg); \
1191 reg_info[this_reg].word = POP_FAILURE_ITEM (); \
1192 DEBUG_PRINT2 (" info: 0x%x\n", reg_info[this_reg]); \
1194 regend[this_reg] = (const char *) POP_FAILURE_ITEM (); \
1195 DEBUG_PRINT2 (" end: 0x%x\n", regend[this_reg]); \
1197 regstart[this_reg] = (const char *) POP_FAILURE_ITEM (); \
1198 DEBUG_PRINT2 (" start: 0x%x\n", regstart[this_reg]); \
1201 DEBUG_STATEMENT (nfailure_points_popped++); \
1202 } /* POP_FAILURE_POINT */
1206 /* Structure for per-register (a.k.a. per-group) information.
1207 This must not be longer than one word, because we push this value
1208 onto the failure stack. Other register information, such as the
1209 starting and ending positions (which are addresses), and the list of
1210 inner groups (which is a bits list) are maintained in separate
1213 We are making a (strictly speaking) nonportable assumption here: that
1214 the compiler will pack our bit fields into something that fits into
1215 the type of `word', i.e., is something that fits into one item on the
1219 fail_stack_elt_t word;
1222 /* This field is one if this group can match the empty string,
1223 zero if not. If not yet determined, `MATCH_NULL_UNSET_VALUE'. */
1224 #define MATCH_NULL_UNSET_VALUE 3
1225 unsigned match_null_string_p : 2;
1226 unsigned is_active : 1;
1227 unsigned matched_something : 1;
1228 unsigned ever_matched_something : 1;
1230 } register_info_type;
1232 #define REG_MATCH_NULL_STRING_P(R) ((R).bits.match_null_string_p)
1233 #define IS_ACTIVE(R) ((R).bits.is_active)
1234 #define MATCHED_SOMETHING(R) ((R).bits.matched_something)
1235 #define EVER_MATCHED_SOMETHING(R) ((R).bits.ever_matched_something)
1238 /* Call this when have matched a real character; it sets `matched' flags
1239 for the subexpressions which we are currently inside. Also records
1240 that those subexprs have matched. */
1241 #define SET_REGS_MATCHED() \
1245 for (r = lowest_active_reg; r <= highest_active_reg; r++) \
1247 MATCHED_SOMETHING (reg_info[r]) \
1248 = EVER_MATCHED_SOMETHING (reg_info[r]) \
1255 /* Registers are set to a sentinel when they haven't yet matched. */
1256 #define REG_UNSET_VALUE ((char *) -1)
1257 #define REG_UNSET(e) ((e) == REG_UNSET_VALUE)
1261 /* How do we implement a missing MATCH_MAY_ALLOCATE?
1262 We make the fail stack a global thing, and then grow it to
1263 re_max_failures when we compile. */
1264 #ifndef MATCH_MAY_ALLOCATE
1265 static fail_stack_type fail_stack;
1267 static const char ** regstart, ** regend;
1268 static const char ** old_regstart, ** old_regend;
1269 static const char **best_regstart, **best_regend;
1270 static register_info_type *reg_info;
1271 static const char **reg_dummy;
1272 static register_info_type *reg_info_dummy;
1276 /* Subroutine declarations and macros for regex_compile. */
1278 static void store_op1 (), store_op2 ();
1279 static void insert_op1 (), insert_op2 ();
1280 static boolean at_begline_loc_p (), at_endline_loc_p ();
1281 static boolean group_in_compile_stack ();
1282 static reg_errcode_t compile_range ();
1284 /* Fetch the next character in the uncompiled pattern---translating it
1285 if necessary. Also cast from a signed character in the constant
1286 string passed to us by the user to an unsigned char that we can use
1287 as an array index (in, e.g., `translate'). */
1288 #define PATFETCH(c) \
1289 do {if (p == pend) return REG_EEND; \
1290 c = (unsigned char) *p++; \
1291 if (translate) c = translate[c]; \
1294 /* Fetch the next character in the uncompiled pattern, with no
1296 #define PATFETCH_RAW(c) \
1297 do {if (p == pend) return REG_EEND; \
1298 c = (unsigned char) *p++; \
1301 /* Go backwards one character in the pattern. */
1302 #define PATUNFETCH p--
1305 /* If `translate' is non-null, return translate[D], else just D. We
1306 cast the subscript to translate because some data is declared as
1307 `char *', to avoid warnings when a string constant is passed. But
1308 when we use a character as a subscript we must make it unsigned. */
1309 #define TRANSLATE(d) (translate ? translate[(unsigned char) (d)] : (d))
1312 /* Macros for outputting the compiled pattern into `buffer'. */
1314 /* If the buffer isn't allocated when it comes in, use this. */
1315 #define INIT_BUF_SIZE 32
1317 /* Make sure we have at least N more bytes of space in buffer. */
1318 #define GET_BUFFER_SPACE(n) \
1319 while (b - bufp->buffer + (n) > bufp->allocated) \
1322 /* Make sure we have one more byte of buffer space and then add C to it. */
1323 #define BUF_PUSH(c) \
1325 GET_BUFFER_SPACE (1); \
1326 *b++ = (unsigned char) (c); \
1330 /* Ensure we have two more bytes of buffer space and then append C1 and C2. */
1331 #define BUF_PUSH_2(c1, c2) \
1333 GET_BUFFER_SPACE (2); \
1334 *b++ = (unsigned char) (c1); \
1335 *b++ = (unsigned char) (c2); \
1339 /* As with BUF_PUSH_2, except for three bytes. */
1340 #define BUF_PUSH_3(c1, c2, c3) \
1342 GET_BUFFER_SPACE (3); \
1343 *b++ = (unsigned char) (c1); \
1344 *b++ = (unsigned char) (c2); \
1345 *b++ = (unsigned char) (c3); \
1349 /* Store a jump with opcode OP at LOC to location TO. We store a
1350 relative address offset by the three bytes the jump itself occupies. */
1351 #define STORE_JUMP(op, loc, to) \
1352 store_op1 (op, loc, (to) - (loc) - 3)
1354 /* Likewise, for a two-argument jump. */
1355 #define STORE_JUMP2(op, loc, to, arg) \
1356 store_op2 (op, loc, (to) - (loc) - 3, arg)
1358 /* Like `STORE_JUMP', but for inserting. Assume `b' is the buffer end. */
1359 #define INSERT_JUMP(op, loc, to) \
1360 insert_op1 (op, loc, (to) - (loc) - 3, b)
1362 /* Like `STORE_JUMP2', but for inserting. Assume `b' is the buffer end. */
1363 #define INSERT_JUMP2(op, loc, to, arg) \
1364 insert_op2 (op, loc, (to) - (loc) - 3, arg, b)
1367 /* This is not an arbitrary limit: the arguments which represent offsets
1368 into the pattern are two bytes long. So if 2^16 bytes turns out to
1369 be too small, many things would have to change. */
1370 #define MAX_BUF_SIZE (1L << 16)
1373 /* Extend the buffer by twice its current size via realloc and
1374 reset the pointers that pointed into the old block to point to the
1375 correct places in the new one. If extending the buffer results in it
1376 being larger than MAX_BUF_SIZE, then flag memory exhausted. */
1377 #define EXTEND_BUFFER() \
1379 unsigned char *old_buffer = bufp->buffer; \
1380 if (bufp->allocated == MAX_BUF_SIZE) \
1382 bufp->allocated <<= 1; \
1383 if (bufp->allocated > MAX_BUF_SIZE) \
1384 bufp->allocated = MAX_BUF_SIZE; \
1385 bufp->buffer = (unsigned char *) realloc (bufp->buffer, bufp->allocated);\
1386 if (bufp->buffer == NULL) \
1387 return REG_ESPACE; \
1388 /* If the buffer moved, move all the pointers into it. */ \
1389 if (old_buffer != bufp->buffer) \
1391 b = (b - old_buffer) + bufp->buffer; \
1392 begalt = (begalt - old_buffer) + bufp->buffer; \
1393 if (fixup_alt_jump) \
1394 fixup_alt_jump = (fixup_alt_jump - old_buffer) + bufp->buffer;\
1396 laststart = (laststart - old_buffer) + bufp->buffer; \
1397 if (pending_exact) \
1398 pending_exact = (pending_exact - old_buffer) + bufp->buffer; \
1403 /* Since we have one byte reserved for the register number argument to
1404 {start,stop}_memory, the maximum number of groups we can report
1405 things about is what fits in that byte. */
1406 #define MAX_REGNUM 255
1408 /* But patterns can have more than `MAX_REGNUM' registers. We just
1409 ignore the excess. */
1410 typedef unsigned regnum_t;
1413 /* Macros for the compile stack. */
1415 /* Since offsets can go either forwards or backwards, this type needs to
1416 be able to hold values from -(MAX_BUF_SIZE - 1) to MAX_BUF_SIZE - 1. */
1417 typedef int pattern_offset_t;
1421 pattern_offset_t begalt_offset;
1422 pattern_offset_t fixup_alt_jump;
1423 pattern_offset_t inner_group_offset;
1424 pattern_offset_t laststart_offset;
1426 } compile_stack_elt_t;
1431 compile_stack_elt_t *stack;
1433 unsigned avail; /* Offset of next open position. */
1434 } compile_stack_type;
1437 #define INIT_COMPILE_STACK_SIZE 32
1439 #define COMPILE_STACK_EMPTY (compile_stack.avail == 0)
1440 #define COMPILE_STACK_FULL (compile_stack.avail == compile_stack.size)
1442 /* The next available element. */
1443 #define COMPILE_STACK_TOP (compile_stack.stack[compile_stack.avail])
1446 /* Set the bit for character C in a list. */
1447 #define SET_LIST_BIT(c) \
1448 (b[((unsigned char) (c)) / BYTEWIDTH] \
1449 |= 1 << (((unsigned char) c) % BYTEWIDTH))
1452 /* Get the next unsigned number in the uncompiled pattern. */
1453 #define GET_UNSIGNED_NUMBER(num) \
1457 while (ISDIGIT (c)) \
1461 num = num * 10 + c - '0'; \
1469 #define CHAR_CLASS_MAX_LENGTH 6 /* Namely, `xdigit'. */
1471 #define IS_CHAR_CLASS(string) \
1472 (STREQ (string, "alpha") || STREQ (string, "upper") \
1473 || STREQ (string, "lower") || STREQ (string, "digit") \
1474 || STREQ (string, "alnum") || STREQ (string, "xdigit") \
1475 || STREQ (string, "space") || STREQ (string, "print") \
1476 || STREQ (string, "punct") || STREQ (string, "graph") \
1477 || STREQ (string, "cntrl") || STREQ (string, "blank"))
1479 /* `regex_compile' compiles PATTERN (of length SIZE) according to SYNTAX.
1480 Returns one of error codes defined in `regex.h', or zero for success.
1482 Assumes the `allocated' (and perhaps `buffer') and `translate'
1483 fields are set in BUFP on entry.
1485 If it succeeds, results are put in BUFP (if it returns an error, the
1486 contents of BUFP are undefined):
1487 `buffer' is the compiled pattern;
1488 `syntax' is set to SYNTAX;
1489 `used' is set to the length of the compiled pattern;
1490 `fastmap_accurate' is zero;
1491 `re_nsub' is the number of subexpressions in PATTERN;
1492 `not_bol' and `not_eol' are zero;
1494 The `fastmap' and `newline_anchor' fields are neither
1495 examined nor set. */
1497 static reg_errcode_t
1498 regex_compile (pattern, size, syntax, bufp)
1499 const char *pattern;
1501 reg_syntax_t syntax;
1502 struct re_pattern_buffer *bufp;
1504 /* We fetch characters from PATTERN here. Even though PATTERN is
1505 `char *' (i.e., signed), we declare these variables as unsigned, so
1506 they can be reliably used as array indices. */
1507 register unsigned char c, c1;
1509 /* A random temporary spot in PATTERN. */
1512 /* Points to the end of the buffer, where we should append. */
1513 register unsigned char *b;
1515 /* Keeps track of unclosed groups. */
1516 compile_stack_type compile_stack;
1518 /* Points to the current (ending) position in the pattern. */
1519 const char *p = pattern;
1520 const char *pend = pattern + size;
1522 /* How to translate the characters in the pattern. */
1523 char *translate = bufp->translate;
1525 /* Address of the count-byte of the most recently inserted `exactn'
1526 command. This makes it possible to tell if a new exact-match
1527 character can be added to that command or if the character requires
1528 a new `exactn' command. */
1529 unsigned char *pending_exact = 0;
1531 /* Address of start of the most recently finished expression.
1532 This tells, e.g., postfix * where to find the start of its
1533 operand. Reset at the beginning of groups and alternatives. */
1534 unsigned char *laststart = 0;
1536 /* Address of beginning of regexp, or inside of last group. */
1537 unsigned char *begalt;
1539 /* Place in the uncompiled pattern (i.e., the {) to
1540 which to go back if the interval is invalid. */
1541 const char *beg_interval;
1543 /* Address of the place where a forward jump should go to the end of
1544 the containing expression. Each alternative of an `or' -- except the
1545 last -- ends with a forward jump of this sort. */
1546 unsigned char *fixup_alt_jump = 0;
1548 /* Counts open-groups as they are encountered. Remembered for the
1549 matching close-group on the compile stack, so the same register
1550 number is put in the stop_memory as the start_memory. */
1551 regnum_t regnum = 0;
1554 DEBUG_PRINT1 ("\nCompiling pattern: ");
1557 unsigned debug_count;
1559 for (debug_count = 0; debug_count < size; debug_count++)
1560 printchar (pattern[debug_count]);
1565 /* Initialize the compile stack. */
1566 compile_stack.stack = TALLOC (INIT_COMPILE_STACK_SIZE, compile_stack_elt_t);
1567 if (compile_stack.stack == NULL)
1570 compile_stack.size = INIT_COMPILE_STACK_SIZE;
1571 compile_stack.avail = 0;
1573 /* Initialize the pattern buffer. */
1574 bufp->syntax = syntax;
1575 bufp->fastmap_accurate = 0;
1576 bufp->not_bol = bufp->not_eol = 0;
1578 /* Set `used' to zero, so that if we return an error, the pattern
1579 printer (for debugging) will think there's no pattern. We reset it
1583 /* Always count groups, whether or not bufp->no_sub is set. */
1586 #if !defined (emacs) && !defined (SYNTAX_TABLE)
1587 /* Initialize the syntax table. */
1588 init_syntax_once ();
1591 if (bufp->allocated == 0)
1594 { /* If zero allocated, but buffer is non-null, try to realloc
1595 enough space. This loses if buffer's address is bogus, but
1596 that is the user's responsibility. */
1597 RETALLOC (bufp->buffer, INIT_BUF_SIZE, unsigned char);
1600 { /* Caller did not allocate a buffer. Do it for them. */
1601 bufp->buffer = TALLOC (INIT_BUF_SIZE, unsigned char);
1603 if (!bufp->buffer) return REG_ESPACE;
1605 bufp->allocated = INIT_BUF_SIZE;
1608 begalt = b = bufp->buffer;
1610 /* Loop through the uncompiled pattern until we're at the end. */
1619 if ( /* If at start of pattern, it's an operator. */
1621 /* If context independent, it's an operator. */
1622 || syntax & RE_CONTEXT_INDEP_ANCHORS
1623 /* Otherwise, depends on what's come before. */
1624 || at_begline_loc_p (pattern, p, syntax))
1634 if ( /* If at end of pattern, it's an operator. */
1636 /* If context independent, it's an operator. */
1637 || syntax & RE_CONTEXT_INDEP_ANCHORS
1638 /* Otherwise, depends on what's next. */
1639 || at_endline_loc_p (p, pend, syntax))
1649 if ((syntax & RE_BK_PLUS_QM)
1650 || (syntax & RE_LIMITED_OPS))
1654 /* If there is no previous pattern... */
1657 if (syntax & RE_CONTEXT_INVALID_OPS)
1659 else if (!(syntax & RE_CONTEXT_INDEP_OPS))
1664 /* Are we optimizing this jump? */
1665 boolean keep_string_p = false;
1667 /* 1 means zero (many) matches is allowed. */
1668 char zero_times_ok = 0, many_times_ok = 0;
1670 /* If there is a sequence of repetition chars, collapse it
1671 down to just one (the right one). We can't combine
1672 interval operators with these because of, e.g., `a{2}*',
1673 which should only match an even number of `a's. */
1677 zero_times_ok |= c != '+';
1678 many_times_ok |= c != '?';
1686 || (!(syntax & RE_BK_PLUS_QM) && (c == '+' || c == '?')))
1689 else if (syntax & RE_BK_PLUS_QM && c == '\\')
1691 if (p == pend) return REG_EESCAPE;
1694 if (!(c1 == '+' || c1 == '?'))
1709 /* If we get here, we found another repeat character. */
1712 /* Star, etc. applied to an empty pattern is equivalent
1713 to an empty pattern. */
1717 /* Now we know whether or not zero matches is allowed
1718 and also whether or not two or more matches is allowed. */
1720 { /* More than one repetition is allowed, so put in at the
1721 end a backward relative jump from `b' to before the next
1722 jump we're going to put in below (which jumps from
1723 laststart to after this jump).
1725 But if we are at the `*' in the exact sequence `.*\n',
1726 insert an unconditional jump backwards to the .,
1727 instead of the beginning of the loop. This way we only
1728 push a failure point once, instead of every time
1729 through the loop. */
1730 assert (p - 1 > pattern);
1732 /* Allocate the space for the jump. */
1733 GET_BUFFER_SPACE (3);
1735 /* We know we are not at the first character of the pattern,
1736 because laststart was nonzero. And we've already
1737 incremented `p', by the way, to be the character after
1738 the `*'. Do we have to do something analogous here
1739 for null bytes, because of RE_DOT_NOT_NULL? */
1740 if (TRANSLATE (*(p - 2)) == TRANSLATE ('.')
1742 && p < pend && TRANSLATE (*p) == TRANSLATE ('\n')
1743 && !(syntax & RE_DOT_NEWLINE))
1744 { /* We have .*\n. */
1745 STORE_JUMP (jump, b, laststart);
1746 keep_string_p = true;
1749 /* Anything else. */
1750 STORE_JUMP (maybe_pop_jump, b, laststart - 3);
1752 /* We've added more stuff to the buffer. */
1756 /* On failure, jump from laststart to b + 3, which will be the
1757 end of the buffer after this jump is inserted. */
1758 GET_BUFFER_SPACE (3);
1759 INSERT_JUMP (keep_string_p ? on_failure_keep_string_jump
1767 /* At least one repetition is required, so insert a
1768 `dummy_failure_jump' before the initial
1769 `on_failure_jump' instruction of the loop. This
1770 effects a skip over that instruction the first time
1771 we hit that loop. */
1772 GET_BUFFER_SPACE (3);
1773 INSERT_JUMP (dummy_failure_jump, laststart, laststart + 6);
1788 boolean had_char_class = false;
1790 if (p == pend) return REG_EBRACK;
1792 /* Ensure that we have enough space to push a charset: the
1793 opcode, the length count, and the bitset; 34 bytes in all. */
1794 GET_BUFFER_SPACE (34);
1798 /* We test `*p == '^' twice, instead of using an if
1799 statement, so we only need one BUF_PUSH. */
1800 BUF_PUSH (*p == '^' ? charset_not : charset);
1804 /* Remember the first position in the bracket expression. */
1807 /* Push the number of bytes in the bitmap. */
1808 BUF_PUSH ((1 << BYTEWIDTH) / BYTEWIDTH);
1810 /* Clear the whole map. */
1811 bzero (b, (1 << BYTEWIDTH) / BYTEWIDTH);
1813 /* charset_not matches newline according to a syntax bit. */
1814 if ((re_opcode_t) b[-2] == charset_not
1815 && (syntax & RE_HAT_LISTS_NOT_NEWLINE))
1816 SET_LIST_BIT ('\n');
1818 /* Read in characters and ranges, setting map bits. */
1821 if (p == pend) return REG_EBRACK;
1825 /* \ might escape characters inside [...] and [^...]. */
1826 if ((syntax & RE_BACKSLASH_ESCAPE_IN_LISTS) && c == '\\')
1828 if (p == pend) return REG_EESCAPE;
1835 /* Could be the end of the bracket expression. If it's
1836 not (i.e., when the bracket expression is `[]' so
1837 far), the ']' character bit gets set way below. */
1838 if (c == ']' && p != p1 + 1)
1841 /* Look ahead to see if it's a range when the last thing
1842 was a character class. */
1843 if (had_char_class && c == '-' && *p != ']')
1846 /* Look ahead to see if it's a range when the last thing
1847 was a character: if this is a hyphen not at the
1848 beginning or the end of a list, then it's the range
1851 && !(p - 2 >= pattern && p[-2] == '[')
1852 && !(p - 3 >= pattern && p[-3] == '[' && p[-2] == '^')
1856 = compile_range (&p, pend, translate, syntax, b);
1857 if (ret != REG_NOERROR) return ret;
1860 else if (p[0] == '-' && p[1] != ']')
1861 { /* This handles ranges made up of characters only. */
1864 /* Move past the `-'. */
1867 ret = compile_range (&p, pend, translate, syntax, b);
1868 if (ret != REG_NOERROR) return ret;
1871 /* See if we're at the beginning of a possible character
1874 else if (syntax & RE_CHAR_CLASSES && c == '[' && *p == ':')
1875 { /* Leave room for the null. */
1876 char str[CHAR_CLASS_MAX_LENGTH + 1];
1881 /* If pattern is `[[:'. */
1882 if (p == pend) return REG_EBRACK;
1887 if (c == ':' || c == ']' || p == pend
1888 || c1 == CHAR_CLASS_MAX_LENGTH)
1894 /* If isn't a word bracketed by `[:' and:`]':
1895 undo the ending character, the letters, and leave
1896 the leading `:' and `[' (but set bits for them). */
1897 if (c == ':' && *p == ']')
1900 boolean is_alnum = STREQ (str, "alnum");
1901 boolean is_alpha = STREQ (str, "alpha");
1902 boolean is_blank = STREQ (str, "blank");
1903 boolean is_cntrl = STREQ (str, "cntrl");
1904 boolean is_digit = STREQ (str, "digit");
1905 boolean is_graph = STREQ (str, "graph");
1906 boolean is_lower = STREQ (str, "lower");
1907 boolean is_print = STREQ (str, "print");
1908 boolean is_punct = STREQ (str, "punct");
1909 boolean is_space = STREQ (str, "space");
1910 boolean is_upper = STREQ (str, "upper");
1911 boolean is_xdigit = STREQ (str, "xdigit");
1913 if (!IS_CHAR_CLASS (str)) return REG_ECTYPE;
1915 /* Throw away the ] at the end of the character
1919 if (p == pend) return REG_EBRACK;
1921 for (ch = 0; ch < 1 << BYTEWIDTH; ch++)
1923 if ( (is_alnum && ISALNUM (ch))
1924 || (is_alpha && ISALPHA (ch))
1925 || (is_blank && ISBLANK (ch))
1926 || (is_cntrl && ISCNTRL (ch))
1927 || (is_digit && ISDIGIT (ch))
1928 || (is_graph && ISGRAPH (ch))
1929 || (is_lower && ISLOWER (ch))
1930 || (is_print && ISPRINT (ch))
1931 || (is_punct && ISPUNCT (ch))
1932 || (is_space && ISSPACE (ch))
1933 || (is_upper && ISUPPER (ch))
1934 || (is_xdigit && ISXDIGIT (ch)))
1937 had_char_class = true;
1946 had_char_class = false;
1951 had_char_class = false;
1956 /* Discard any (non)matching list bytes that are all 0 at the
1957 end of the map. Decrease the map-length byte too. */
1958 while ((int) b[-1] > 0 && b[b[-1] - 1] == 0)
1966 if (syntax & RE_NO_BK_PARENS)
1973 if (syntax & RE_NO_BK_PARENS)
1980 if (syntax & RE_NEWLINE_ALT)
1987 if (syntax & RE_NO_BK_VBAR)
1994 if (syntax & RE_INTERVALS && syntax & RE_NO_BK_BRACES)
1995 goto handle_interval;
2001 if (p == pend) return REG_EESCAPE;
2003 /* Do not translate the character after the \, so that we can
2004 distinguish, e.g., \B from \b, even if we normally would
2005 translate, e.g., B to b. */
2011 if (syntax & RE_NO_BK_PARENS)
2012 goto normal_backslash;
2018 if (COMPILE_STACK_FULL)
2020 RETALLOC (compile_stack.stack, compile_stack.size << 1,
2021 compile_stack_elt_t);
2022 if (compile_stack.stack == NULL) return REG_ESPACE;
2024 compile_stack.size <<= 1;
2027 /* These are the values to restore when we hit end of this
2028 group. They are all relative offsets, so that if the
2029 whole pattern moves because of realloc, they will still
2031 COMPILE_STACK_TOP.begalt_offset = begalt - bufp->buffer;
2032 COMPILE_STACK_TOP.fixup_alt_jump
2033 = fixup_alt_jump ? fixup_alt_jump - bufp->buffer + 1 : 0;
2034 COMPILE_STACK_TOP.laststart_offset = b - bufp->buffer;
2035 COMPILE_STACK_TOP.regnum = regnum;
2037 /* We will eventually replace the 0 with the number of
2038 groups inner to this one. But do not push a
2039 start_memory for groups beyond the last one we can
2040 represent in the compiled pattern. */
2041 if (regnum <= MAX_REGNUM)
2043 COMPILE_STACK_TOP.inner_group_offset = b - bufp->buffer + 2;
2044 BUF_PUSH_3 (start_memory, regnum, 0);
2047 compile_stack.avail++;
2052 /* If we've reached MAX_REGNUM groups, then this open
2053 won't actually generate any code, so we'll have to
2054 clear pending_exact explicitly. */
2060 if (syntax & RE_NO_BK_PARENS) goto normal_backslash;
2062 if (COMPILE_STACK_EMPTY)
2063 if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD)
2064 goto normal_backslash;
2070 { /* Push a dummy failure point at the end of the
2071 alternative for a possible future
2072 `pop_failure_jump' to pop. See comments at
2073 `push_dummy_failure' in `re_match_2'. */
2074 BUF_PUSH (push_dummy_failure);
2076 /* We allocated space for this jump when we assigned
2077 to `fixup_alt_jump', in the `handle_alt' case below. */
2078 STORE_JUMP (jump_past_alt, fixup_alt_jump, b - 1);
2081 /* See similar code for backslashed left paren above. */
2082 if (COMPILE_STACK_EMPTY)
2083 if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD)
2088 /* Since we just checked for an empty stack above, this
2089 ``can't happen''. */
2090 assert (compile_stack.avail != 0);
2092 /* We don't just want to restore into `regnum', because
2093 later groups should continue to be numbered higher,
2094 as in `(ab)c(de)' -- the second group is #2. */
2095 regnum_t this_group_regnum;
2097 compile_stack.avail--;
2098 begalt = bufp->buffer + COMPILE_STACK_TOP.begalt_offset;
2100 = COMPILE_STACK_TOP.fixup_alt_jump
2101 ? bufp->buffer + COMPILE_STACK_TOP.fixup_alt_jump - 1
2103 laststart = bufp->buffer + COMPILE_STACK_TOP.laststart_offset;
2104 this_group_regnum = COMPILE_STACK_TOP.regnum;
2105 /* If we've reached MAX_REGNUM groups, then this open
2106 won't actually generate any code, so we'll have to
2107 clear pending_exact explicitly. */
2110 /* We're at the end of the group, so now we know how many
2111 groups were inside this one. */
2112 if (this_group_regnum <= MAX_REGNUM)
2114 unsigned char *inner_group_loc
2115 = bufp->buffer + COMPILE_STACK_TOP.inner_group_offset;
2117 *inner_group_loc = regnum - this_group_regnum;
2118 BUF_PUSH_3 (stop_memory, this_group_regnum,
2119 regnum - this_group_regnum);
2125 case '|': /* `\|'. */
2126 if (syntax & RE_LIMITED_OPS || syntax & RE_NO_BK_VBAR)
2127 goto normal_backslash;
2129 if (syntax & RE_LIMITED_OPS)
2132 /* Insert before the previous alternative a jump which
2133 jumps to this alternative if the former fails. */
2134 GET_BUFFER_SPACE (3);
2135 INSERT_JUMP (on_failure_jump, begalt, b + 6);
2139 /* The alternative before this one has a jump after it
2140 which gets executed if it gets matched. Adjust that
2141 jump so it will jump to this alternative's analogous
2142 jump (put in below, which in turn will jump to the next
2143 (if any) alternative's such jump, etc.). The last such
2144 jump jumps to the correct final destination. A picture:
2150 If we are at `b', then fixup_alt_jump right now points to a
2151 three-byte space after `a'. We'll put in the jump, set
2152 fixup_alt_jump to right after `b', and leave behind three
2153 bytes which we'll fill in when we get to after `c'. */
2156 STORE_JUMP (jump_past_alt, fixup_alt_jump, b);
2158 /* Mark and leave space for a jump after this alternative,
2159 to be filled in later either by next alternative or
2160 when know we're at the end of a series of alternatives. */
2162 GET_BUFFER_SPACE (3);
2171 /* If \{ is a literal. */
2172 if (!(syntax & RE_INTERVALS)
2173 /* If we're at `\{' and it's not the open-interval
2175 || ((syntax & RE_INTERVALS) && (syntax & RE_NO_BK_BRACES))
2176 || (p - 2 == pattern && p == pend))
2177 goto normal_backslash;
2181 /* If got here, then the syntax allows intervals. */
2183 /* At least (most) this many matches must be made. */
2184 int lower_bound = -1, upper_bound = -1;
2186 beg_interval = p - 1;
2190 if (syntax & RE_NO_BK_BRACES)
2191 goto unfetch_interval;
2196 GET_UNSIGNED_NUMBER (lower_bound);
2200 GET_UNSIGNED_NUMBER (upper_bound);
2201 if (upper_bound < 0) upper_bound = RE_DUP_MAX;
2204 /* Interval such as `{1}' => match exactly once. */
2205 upper_bound = lower_bound;
2207 if (lower_bound < 0 || upper_bound > RE_DUP_MAX
2208 || lower_bound > upper_bound)
2210 if (syntax & RE_NO_BK_BRACES)
2211 goto unfetch_interval;
2216 if (!(syntax & RE_NO_BK_BRACES))
2218 if (c != '\\') return REG_EBRACE;
2225 if (syntax & RE_NO_BK_BRACES)
2226 goto unfetch_interval;
2231 /* We just parsed a valid interval. */
2233 /* If it's invalid to have no preceding re. */
2236 if (syntax & RE_CONTEXT_INVALID_OPS)
2238 else if (syntax & RE_CONTEXT_INDEP_OPS)
2241 goto unfetch_interval;
2244 /* If the upper bound is zero, don't want to succeed at
2245 all; jump from `laststart' to `b + 3', which will be
2246 the end of the buffer after we insert the jump. */
2247 if (upper_bound == 0)
2249 GET_BUFFER_SPACE (3);
2250 INSERT_JUMP (jump, laststart, b + 3);
2254 /* Otherwise, we have a nontrivial interval. When
2255 we're all done, the pattern will look like:
2256 set_number_at <jump count> <upper bound>
2257 set_number_at <succeed_n count> <lower bound>
2258 succeed_n <after jump addr> <succeed_n count>
2260 jump_n <succeed_n addr> <jump count>
2261 (The upper bound and `jump_n' are omitted if
2262 `upper_bound' is 1, though.) */
2264 { /* If the upper bound is > 1, we need to insert
2265 more at the end of the loop. */
2266 unsigned nbytes = 10 + (upper_bound > 1) * 10;
2268 GET_BUFFER_SPACE (nbytes);
2270 /* Initialize lower bound of the `succeed_n', even
2271 though it will be set during matching by its
2272 attendant `set_number_at' (inserted next),
2273 because `re_compile_fastmap' needs to know.
2274 Jump to the `jump_n' we might insert below. */
2275 INSERT_JUMP2 (succeed_n, laststart,
2276 b + 5 + (upper_bound > 1) * 5,
2280 /* Code to initialize the lower bound. Insert
2281 before the `succeed_n'. The `5' is the last two
2282 bytes of this `set_number_at', plus 3 bytes of
2283 the following `succeed_n'. */
2284 insert_op2 (set_number_at, laststart, 5, lower_bound, b);
2287 if (upper_bound > 1)
2288 { /* More than one repetition is allowed, so
2289 append a backward jump to the `succeed_n'
2290 that starts this interval.
2292 When we've reached this during matching,
2293 we'll have matched the interval once, so
2294 jump back only `upper_bound - 1' times. */
2295 STORE_JUMP2 (jump_n, b, laststart + 5,
2299 /* The location we want to set is the second
2300 parameter of the `jump_n'; that is `b-2' as
2301 an absolute address. `laststart' will be
2302 the `set_number_at' we're about to insert;
2303 `laststart+3' the number to set, the source
2304 for the relative address. But we are
2305 inserting into the middle of the pattern --
2306 so everything is getting moved up by 5.
2307 Conclusion: (b - 2) - (laststart + 3) + 5,
2308 i.e., b - laststart.
2310 We insert this at the beginning of the loop
2311 so that if we fail during matching, we'll
2312 reinitialize the bounds. */
2313 insert_op2 (set_number_at, laststart, b - laststart,
2314 upper_bound - 1, b);
2319 beg_interval = NULL;
2324 /* If an invalid interval, match the characters as literals. */
2325 assert (beg_interval);
2327 beg_interval = NULL;
2329 /* normal_char and normal_backslash need `c'. */
2332 if (!(syntax & RE_NO_BK_BRACES))
2334 if (p > pattern && p[-1] == '\\')
2335 goto normal_backslash;
2340 /* There is no way to specify the before_dot and after_dot
2341 operators. rms says this is ok. --karl */
2349 BUF_PUSH_2 (syntaxspec, syntax_spec_code[c]);
2355 BUF_PUSH_2 (notsyntaxspec, syntax_spec_code[c]);
2362 BUF_PUSH (wordchar);
2368 BUF_PUSH (notwordchar);
2381 BUF_PUSH (wordbound);
2385 BUF_PUSH (notwordbound);
2396 case '1': case '2': case '3': case '4': case '5':
2397 case '6': case '7': case '8': case '9':
2398 if (syntax & RE_NO_BK_REFS)
2406 /* Can't back reference to a subexpression if inside of it. */
2407 if (group_in_compile_stack (compile_stack, c1))
2411 BUF_PUSH_2 (duplicate, c1);
2417 if (syntax & RE_BK_PLUS_QM)
2420 goto normal_backslash;
2424 /* You might think it would be useful for \ to mean
2425 not to translate; but if we don't translate it
2426 it will never match anything. */
2434 /* Expects the character in `c'. */
2436 /* If no exactn currently being built. */
2439 /* If last exactn not at current position. */
2440 || pending_exact + *pending_exact + 1 != b
2442 /* We have only one byte following the exactn for the count. */
2443 || *pending_exact == (1 << BYTEWIDTH) - 1
2445 /* If followed by a repetition operator. */
2446 || *p == '*' || *p == '^'
2447 || ((syntax & RE_BK_PLUS_QM)
2448 ? *p == '\\' && (p[1] == '+' || p[1] == '?')
2449 : (*p == '+' || *p == '?'))
2450 || ((syntax & RE_INTERVALS)
2451 && ((syntax & RE_NO_BK_BRACES)
2453 : (p[0] == '\\' && p[1] == '{'))))
2455 /* Start building a new exactn. */
2459 BUF_PUSH_2 (exactn, 0);
2460 pending_exact = b - 1;
2467 } /* while p != pend */
2470 /* Through the pattern now. */
2473 STORE_JUMP (jump_past_alt, fixup_alt_jump, b);
2475 if (!COMPILE_STACK_EMPTY)
2478 free (compile_stack.stack);
2480 /* We have succeeded; set the length of the buffer. */
2481 bufp->used = b - bufp->buffer;
2486 DEBUG_PRINT1 ("\nCompiled pattern: \n");
2487 print_compiled_pattern (bufp);
2491 #ifndef MATCH_MAY_ALLOCATE
2492 /* Initialize the failure stack to the largest possible stack. This
2493 isn't necessary unless we're trying to avoid calling alloca in
2494 the search and match routines. */
2496 int num_regs = bufp->re_nsub + 1;
2498 /* Since DOUBLE_FAIL_STACK refuses to double only if the current size
2499 is strictly greater than re_max_failures, the largest possible stack
2500 is 2 * re_max_failures failure points. */
2501 if (fail_stack.size < (2 * re_max_failures * MAX_FAILURE_ITEMS))
2503 fail_stack.size = (2 * re_max_failures * MAX_FAILURE_ITEMS);
2506 if (! fail_stack.stack)
2508 = (fail_stack_elt_t *) xmalloc (fail_stack.size
2509 * sizeof (fail_stack_elt_t));
2512 = (fail_stack_elt_t *) xrealloc (fail_stack.stack,
2514 * sizeof (fail_stack_elt_t)));
2515 #else /* not emacs */
2516 if (! fail_stack.stack)
2518 = (fail_stack_elt_t *) malloc (fail_stack.size
2519 * sizeof (fail_stack_elt_t));
2522 = (fail_stack_elt_t *) realloc (fail_stack.stack,
2524 * sizeof (fail_stack_elt_t)));
2525 #endif /* not emacs */
2528 /* Initialize some other variables the matcher uses. */
2529 RETALLOC_IF (regstart, num_regs, const char *);
2530 RETALLOC_IF (regend, num_regs, const char *);
2531 RETALLOC_IF (old_regstart, num_regs, const char *);
2532 RETALLOC_IF (old_regend, num_regs, const char *);
2533 RETALLOC_IF (best_regstart, num_regs, const char *);
2534 RETALLOC_IF (best_regend, num_regs, const char *);
2535 RETALLOC_IF (reg_info, num_regs, register_info_type);
2536 RETALLOC_IF (reg_dummy, num_regs, const char *);
2537 RETALLOC_IF (reg_info_dummy, num_regs, register_info_type);
2542 } /* regex_compile */
2544 /* Subroutines for `regex_compile'. */
2546 /* Store OP at LOC followed by two-byte integer parameter ARG. */
2549 store_op1 (op, loc, arg)
2554 *loc = (unsigned char) op;
2555 STORE_NUMBER (loc + 1, arg);
2559 /* Like `store_op1', but for two two-byte parameters ARG1 and ARG2. */
2562 store_op2 (op, loc, arg1, arg2)
2567 *loc = (unsigned char) op;
2568 STORE_NUMBER (loc + 1, arg1);
2569 STORE_NUMBER (loc + 3, arg2);
2573 /* Copy the bytes from LOC to END to open up three bytes of space at LOC
2574 for OP followed by two-byte integer parameter ARG. */
2577 insert_op1 (op, loc, arg, end)
2583 register unsigned char *pfrom = end;
2584 register unsigned char *pto = end + 3;
2586 while (pfrom != loc)
2589 store_op1 (op, loc, arg);
2593 /* Like `insert_op1', but for two two-byte parameters ARG1 and ARG2. */
2596 insert_op2 (op, loc, arg1, arg2, end)
2602 register unsigned char *pfrom = end;
2603 register unsigned char *pto = end + 5;
2605 while (pfrom != loc)
2608 store_op2 (op, loc, arg1, arg2);
2612 /* P points to just after a ^ in PATTERN. Return true if that ^ comes
2613 after an alternative or a begin-subexpression. We assume there is at
2614 least one character before the ^. */
2617 at_begline_loc_p (pattern, p, syntax)
2618 const char *pattern, *p;
2619 reg_syntax_t syntax;
2621 const char *prev = p - 2;
2622 boolean prev_prev_backslash = prev > pattern && prev[-1] == '\\';
2625 /* After a subexpression? */
2626 (*prev == '(' && (syntax & RE_NO_BK_PARENS || prev_prev_backslash))
2627 /* After an alternative? */
2628 || (*prev == '|' && (syntax & RE_NO_BK_VBAR || prev_prev_backslash));
2632 /* The dual of at_begline_loc_p. This one is for $. We assume there is
2633 at least one character after the $, i.e., `P < PEND'. */
2636 at_endline_loc_p (p, pend, syntax)
2637 const char *p, *pend;
2640 const char *next = p;
2641 boolean next_backslash = *next == '\\';
2642 const char *next_next = p + 1 < pend ? p + 1 : NULL;
2645 /* Before a subexpression? */
2646 (syntax & RE_NO_BK_PARENS ? *next == ')'
2647 : next_backslash && next_next && *next_next == ')')
2648 /* Before an alternative? */
2649 || (syntax & RE_NO_BK_VBAR ? *next == '|'
2650 : next_backslash && next_next && *next_next == '|');
2654 /* Returns true if REGNUM is in one of COMPILE_STACK's elements and
2655 false if it's not. */
2658 group_in_compile_stack (compile_stack, regnum)
2659 compile_stack_type compile_stack;
2664 for (this_element = compile_stack.avail - 1;
2667 if (compile_stack.stack[this_element].regnum == regnum)
2674 /* Read the ending character of a range (in a bracket expression) from the
2675 uncompiled pattern *P_PTR (which ends at PEND). We assume the
2676 starting character is in `P[-2]'. (`P[-1]' is the character `-'.)
2677 Then we set the translation of all bits between the starting and
2678 ending characters (inclusive) in the compiled pattern B.
2680 Return an error code.
2682 We use these short variable names so we can use the same macros as
2683 `regex_compile' itself. */
2685 static reg_errcode_t
2686 compile_range (p_ptr, pend, translate, syntax, b)
2687 const char **p_ptr, *pend;
2689 reg_syntax_t syntax;
2694 const char *p = *p_ptr;
2695 int range_start, range_end;
2700 /* Even though the pattern is a signed `char *', we need to fetch
2701 with unsigned char *'s; if the high bit of the pattern character
2702 is set, the range endpoints will be negative if we fetch using a
2705 We also want to fetch the endpoints without translating them; the
2706 appropriate translation is done in the bit-setting loop below. */
2707 range_start = ((unsigned char *) p)[-2];
2708 range_end = ((unsigned char *) p)[0];
2710 /* Have to increment the pointer into the pattern string, so the
2711 caller isn't still at the ending character. */
2714 /* If the start is after the end, the range is empty. */
2715 if (range_start > range_end)
2716 return syntax & RE_NO_EMPTY_RANGES ? REG_ERANGE : REG_NOERROR;
2718 /* Here we see why `this_char' has to be larger than an `unsigned
2719 char' -- the range is inclusive, so if `range_end' == 0xff
2720 (assuming 8-bit characters), we would otherwise go into an infinite
2721 loop, since all characters <= 0xff. */
2722 for (this_char = range_start; this_char <= range_end; this_char++)
2724 SET_LIST_BIT (TRANSLATE (this_char));
2730 /* re_compile_fastmap computes a ``fastmap'' for the compiled pattern in
2731 BUFP. A fastmap records which of the (1 << BYTEWIDTH) possible
2732 characters can start a string that matches the pattern. This fastmap
2733 is used by re_search to skip quickly over impossible starting points.
2735 The caller must supply the address of a (1 << BYTEWIDTH)-byte data
2736 area as BUFP->fastmap.
2738 We set the `fastmap', `fastmap_accurate', and `can_be_null' fields in
2741 Returns 0 if we succeed, -2 if an internal error. */
2744 re_compile_fastmap (bufp)
2745 struct re_pattern_buffer *bufp;
2748 #ifdef MATCH_MAY_ALLOCATE
2749 fail_stack_type fail_stack;
2751 #ifndef REGEX_MALLOC
2754 /* We don't push any register information onto the failure stack. */
2755 unsigned num_regs = 0;
2757 register char *fastmap = bufp->fastmap;
2758 unsigned char *pattern = bufp->buffer;
2759 unsigned long size = bufp->used;
2760 unsigned char *p = pattern;
2761 register unsigned char *pend = pattern + size;
2763 /* Assume that each path through the pattern can be null until
2764 proven otherwise. We set this false at the bottom of switch
2765 statement, to which we get only if a particular path doesn't
2766 match the empty string. */
2767 boolean path_can_be_null = true;
2769 /* We aren't doing a `succeed_n' to begin with. */
2770 boolean succeed_n_p = false;
2772 assert (fastmap != NULL && p != NULL);
2775 bzero (fastmap, 1 << BYTEWIDTH); /* Assume nothing's valid. */
2776 bufp->fastmap_accurate = 1; /* It will be when we're done. */
2777 bufp->can_be_null = 0;
2779 while (p != pend || !FAIL_STACK_EMPTY ())
2783 bufp->can_be_null |= path_can_be_null;
2785 /* Reset for next path. */
2786 path_can_be_null = true;
2788 p = fail_stack.stack[--fail_stack.avail];
2791 /* We should never be about to go beyond the end of the pattern. */
2794 #ifdef SWITCH_ENUM_BUG
2795 switch ((int) ((re_opcode_t) *p++))
2797 switch ((re_opcode_t) *p++)
2801 /* I guess the idea here is to simply not bother with a fastmap
2802 if a backreference is used, since it's too hard to figure out
2803 the fastmap for the corresponding group. Setting
2804 `can_be_null' stops `re_search_2' from using the fastmap, so
2805 that is all we do. */
2807 bufp->can_be_null = 1;
2811 /* Following are the cases which match a character. These end
2820 for (j = *p++ * BYTEWIDTH - 1; j >= 0; j--)
2821 if (p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH)))
2827 /* Chars beyond end of map must be allowed. */
2828 for (j = *p * BYTEWIDTH; j < (1 << BYTEWIDTH); j++)
2831 for (j = *p++ * BYTEWIDTH - 1; j >= 0; j--)
2832 if (!(p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH))))
2838 for (j = 0; j < (1 << BYTEWIDTH); j++)
2839 if (SYNTAX (j) == Sword)
2845 for (j = 0; j < (1 << BYTEWIDTH); j++)
2846 if (SYNTAX (j) != Sword)
2852 /* `.' matches anything ... */
2853 for (j = 0; j < (1 << BYTEWIDTH); j++)
2856 /* ... except perhaps newline. */
2857 if (!(bufp->syntax & RE_DOT_NEWLINE))
2860 /* Return if we have already set `can_be_null'; if we have,
2861 then the fastmap is irrelevant. Something's wrong here. */
2862 else if (bufp->can_be_null)
2865 /* Otherwise, have to check alternative paths. */
2872 for (j = 0; j < (1 << BYTEWIDTH); j++)
2873 if (SYNTAX (j) == (enum syntaxcode) k)
2880 for (j = 0; j < (1 << BYTEWIDTH); j++)
2881 if (SYNTAX (j) != (enum syntaxcode) k)
2886 /* All cases after this match the empty string. These end with
2894 #endif /* not emacs */
2906 case push_dummy_failure:
2911 case pop_failure_jump:
2912 case maybe_pop_jump:
2915 case dummy_failure_jump:
2916 EXTRACT_NUMBER_AND_INCR (j, p);
2921 /* Jump backward implies we just went through the body of a
2922 loop and matched nothing. Opcode jumped to should be
2923 `on_failure_jump' or `succeed_n'. Just treat it like an
2924 ordinary jump. For a * loop, it has pushed its failure
2925 point already; if so, discard that as redundant. */
2926 if ((re_opcode_t) *p != on_failure_jump
2927 && (re_opcode_t) *p != succeed_n)
2931 EXTRACT_NUMBER_AND_INCR (j, p);
2934 /* If what's on the stack is where we are now, pop it. */
2935 if (!FAIL_STACK_EMPTY ()
2936 && fail_stack.stack[fail_stack.avail - 1] == p)
2942 case on_failure_jump:
2943 case on_failure_keep_string_jump:
2944 handle_on_failure_jump:
2945 EXTRACT_NUMBER_AND_INCR (j, p);
2947 /* For some patterns, e.g., `(a?)?', `p+j' here points to the
2948 end of the pattern. We don't want to push such a point,
2949 since when we restore it above, entering the switch will
2950 increment `p' past the end of the pattern. We don't need
2951 to push such a point since we obviously won't find any more
2952 fastmap entries beyond `pend'. Such a pattern can match
2953 the null string, though. */
2956 if (!PUSH_PATTERN_OP (p + j, fail_stack))
2960 bufp->can_be_null = 1;
2964 EXTRACT_NUMBER_AND_INCR (k, p); /* Skip the n. */
2965 succeed_n_p = false;
2972 /* Get to the number of times to succeed. */
2975 /* Increment p past the n for when k != 0. */
2976 EXTRACT_NUMBER_AND_INCR (k, p);
2980 succeed_n_p = true; /* Spaghetti code alert. */
2981 goto handle_on_failure_jump;
2998 abort (); /* We have listed all the cases. */
3001 /* Getting here means we have found the possible starting
3002 characters for one path of the pattern -- and that the empty
3003 string does not match. We need not follow this path further.
3004 Instead, look at the next alternative (remembered on the
3005 stack), or quit if no more. The test at the top of the loop
3006 does these things. */
3007 path_can_be_null = false;
3011 /* Set `can_be_null' for the last path (also the first path, if the
3012 pattern is empty). */
3013 bufp->can_be_null |= path_can_be_null;
3015 } /* re_compile_fastmap */
3017 /* Set REGS to hold NUM_REGS registers, storing them in STARTS and
3018 ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use
3019 this memory for recording register information. STARTS and ENDS
3020 must be allocated using the malloc library routine, and must each
3021 be at least NUM_REGS * sizeof (regoff_t) bytes long.
3023 If NUM_REGS == 0, then subsequent matches should allocate their own
3026 Unless this function is called, the first search or match using
3027 PATTERN_BUFFER will allocate its own register data, without
3028 freeing the old data. */
3031 re_set_registers (bufp, regs, num_regs, starts, ends)
3032 struct re_pattern_buffer *bufp;
3033 struct re_registers *regs;
3035 regoff_t *starts, *ends;
3039 bufp->regs_allocated = REGS_REALLOCATE;
3040 regs->num_regs = num_regs;
3041 regs->start = starts;
3046 bufp->regs_allocated = REGS_UNALLOCATED;
3048 regs->start = regs->end = (regoff_t *) 0;
3052 /* Searching routines. */
3054 /* Like re_search_2, below, but only one string is specified, and
3055 doesn't let you say where to stop matching. */
3058 re_search (bufp, string, size, startpos, range, regs)
3059 struct re_pattern_buffer *bufp;
3061 int size, startpos, range;
3062 struct re_registers *regs;
3064 return re_search_2 (bufp, NULL, 0, string, size, startpos, range,
3069 /* Using the compiled pattern in BUFP->buffer, first tries to match the
3070 virtual concatenation of STRING1 and STRING2, starting first at index
3071 STARTPOS, then at STARTPOS + 1, and so on.
3073 STRING1 and STRING2 have length SIZE1 and SIZE2, respectively.
3075 RANGE is how far to scan while trying to match. RANGE = 0 means try
3076 only at STARTPOS; in general, the last start tried is STARTPOS +
3079 In REGS, return the indices of the virtual concatenation of STRING1
3080 and STRING2 that matched the entire BUFP->buffer and its contained
3083 Do not consider matching one past the index STOP in the virtual
3084 concatenation of STRING1 and STRING2.
3086 We return either the position in the strings at which the match was
3087 found, -1 if no match, or -2 if error (such as failure
3091 re_search_2 (bufp, string1, size1, string2, size2, startpos, range, regs, stop)
3092 struct re_pattern_buffer *bufp;
3093 const char *string1, *string2;
3097 struct re_registers *regs;
3101 register char *fastmap = bufp->fastmap;
3102 register char *translate = bufp->translate;
3103 int total_size = size1 + size2;
3104 int endpos = startpos + range;
3106 /* Check for out-of-range STARTPOS. */
3107 if (startpos < 0 || startpos > total_size)
3110 /* Fix up RANGE if it might eventually take us outside
3111 the virtual concatenation of STRING1 and STRING2. */
3113 range = -1 - startpos;
3114 else if (endpos > total_size)
3115 range = total_size - startpos;
3117 /* If the search isn't to be a backwards one, don't waste time in a
3118 search for a pattern that must be anchored. */
3119 if (bufp->used > 0 && (re_opcode_t) bufp->buffer[0] == begbuf && range > 0)
3127 /* Update the fastmap now if not correct already. */
3128 if (fastmap && !bufp->fastmap_accurate)
3129 if (re_compile_fastmap (bufp) == -2)
3132 /* Loop through the string, looking for a place to start matching. */
3135 /* If a fastmap is supplied, skip quickly over characters that
3136 cannot be the start of a match. If the pattern can match the
3137 null string, however, we don't need to skip characters; we want
3138 the first null string. */
3139 if (fastmap && startpos < total_size && !bufp->can_be_null)
3141 if (range > 0) /* Searching forwards. */
3143 register const char *d;
3144 register int lim = 0;
3147 if (startpos < size1 && startpos + range >= size1)
3148 lim = range - (size1 - startpos);
3150 d = (startpos >= size1 ? string2 - size1 : string1) + startpos;
3152 /* Written out as an if-else to avoid testing `translate'
3156 && !fastmap[(unsigned char)
3157 translate[(unsigned char) *d++]])
3160 while (range > lim && !fastmap[(unsigned char) *d++])
3163 startpos += irange - range;
3165 else /* Searching backwards. */
3167 register char c = (size1 == 0 || startpos >= size1
3168 ? string2[startpos - size1]
3169 : string1[startpos]);
3171 if (!fastmap[(unsigned char) TRANSLATE (c)])
3176 /* If can't match the null string, and that's all we have left, fail. */
3177 if (range >= 0 && startpos == total_size && fastmap
3178 && !bufp->can_be_null)
3181 val = re_match_2_internal (bufp, string1, size1, string2, size2,
3182 startpos, regs, stop);
3208 /* Declarations and macros for re_match_2. */
3210 static int bcmp_translate ();
3211 static boolean alt_match_null_string_p (),
3212 common_op_match_null_string_p (),
3213 group_match_null_string_p ();
3215 /* This converts PTR, a pointer into one of the search strings `string1'
3216 and `string2' into an offset from the beginning of that string. */
3217 #define POINTER_TO_OFFSET(ptr) \
3218 (FIRST_STRING_P (ptr) \
3219 ? ((regoff_t) ((ptr) - string1)) \
3220 : ((regoff_t) ((ptr) - string2 + size1)))
3222 /* Macros for dealing with the split strings in re_match_2. */
3224 #define MATCHING_IN_FIRST_STRING (dend == end_match_1)
3226 /* Call before fetching a character with *d. This switches over to
3227 string2 if necessary. */
3228 #define PREFETCH() \
3231 /* End of string2 => fail. */ \
3232 if (dend == end_match_2) \
3234 /* End of string1 => advance to string2. */ \
3236 dend = end_match_2; \
3240 /* Test if at very beginning or at very end of the virtual concatenation
3241 of `string1' and `string2'. If only one string, it's `string2'. */
3242 #define AT_STRINGS_BEG(d) ((d) == (size1 ? string1 : string2) || !size2)
3243 #define AT_STRINGS_END(d) ((d) == end2)
3246 /* Test if D points to a character which is word-constituent. We have
3247 two special cases to check for: if past the end of string1, look at
3248 the first character in string2; and if before the beginning of
3249 string2, look at the last character in string1. */
3250 #define WORDCHAR_P(d) \
3251 (SYNTAX ((d) == end1 ? *string2 \
3252 : (d) == string2 - 1 ? *(end1 - 1) : *(d)) \
3255 /* Test if the character before D and the one at D differ with respect
3256 to being word-constituent. */
3257 #define AT_WORD_BOUNDARY(d) \
3258 (AT_STRINGS_BEG (d) || AT_STRINGS_END (d) \
3259 || WORDCHAR_P (d - 1) != WORDCHAR_P (d))
3262 /* Free everything we malloc. */
3263 #ifdef MATCH_MAY_ALLOCATE
3265 #define FREE_VAR(var) if (var) free (var); var = NULL
3266 #define FREE_VARIABLES() \
3268 FREE_VAR (fail_stack.stack); \
3269 FREE_VAR (regstart); \
3270 FREE_VAR (regend); \
3271 FREE_VAR (old_regstart); \
3272 FREE_VAR (old_regend); \
3273 FREE_VAR (best_regstart); \
3274 FREE_VAR (best_regend); \
3275 FREE_VAR (reg_info); \
3276 FREE_VAR (reg_dummy); \
3277 FREE_VAR (reg_info_dummy); \
3279 #else /* not REGEX_MALLOC */
3280 /* This used to do alloca (0), but now we do that in the caller. */
3281 #define FREE_VARIABLES() /* Nothing */
3282 #endif /* not REGEX_MALLOC */
3284 #define FREE_VARIABLES() /* Do nothing! */
3285 #endif /* not MATCH_MAY_ALLOCATE */
3287 /* These values must meet several constraints. They must not be valid
3288 register values; since we have a limit of 255 registers (because
3289 we use only one byte in the pattern for the register number), we can
3290 use numbers larger than 255. They must differ by 1, because of
3291 NUM_FAILURE_ITEMS above. And the value for the lowest register must
3292 be larger than the value for the highest register, so we do not try
3293 to actually save any registers when none are active. */
3294 #define NO_HIGHEST_ACTIVE_REG (1 << BYTEWIDTH)
3295 #define NO_LOWEST_ACTIVE_REG (NO_HIGHEST_ACTIVE_REG + 1)
3297 /* Matching routines. */
3299 #ifndef emacs /* Emacs never uses this. */
3300 /* re_match is like re_match_2 except it takes only a single string. */
3303 re_match (bufp, string, size, pos, regs)
3304 struct re_pattern_buffer *bufp;
3307 struct re_registers *regs;
3309 int result = re_match_2_internal (bufp, NULL, 0, string, size,
3314 #endif /* not emacs */
3317 /* re_match_2 matches the compiled pattern in BUFP against the
3318 the (virtual) concatenation of STRING1 and STRING2 (of length SIZE1
3319 and SIZE2, respectively). We start matching at POS, and stop
3322 If REGS is non-null and the `no_sub' field of BUFP is nonzero, we
3323 store offsets for the substring each group matched in REGS. See the
3324 documentation for exactly how many groups we fill.
3326 We return -1 if no match, -2 if an internal error (such as the
3327 failure stack overflowing). Otherwise, we return the length of the
3328 matched substring. */
3331 re_match_2 (bufp, string1, size1, string2, size2, pos, regs, stop)
3332 struct re_pattern_buffer *bufp;
3333 const char *string1, *string2;
3336 struct re_registers *regs;
3339 int result = re_match_2_internal (bufp, string1, size1, string2, size2,
3345 /* This is a separate function so that we can force an alloca cleanup
3348 re_match_2_internal (bufp, string1, size1, string2, size2, pos, regs, stop)
3349 struct re_pattern_buffer *bufp;
3350 const char *string1, *string2;
3353 struct re_registers *regs;
3356 /* General temporaries. */
3360 /* Just past the end of the corresponding string. */
3361 const char *end1, *end2;
3363 /* Pointers into string1 and string2, just past the last characters in
3364 each to consider matching. */
3365 const char *end_match_1, *end_match_2;
3367 /* Where we are in the data, and the end of the current string. */
3368 const char *d, *dend;
3370 /* Where we are in the pattern, and the end of the pattern. */
3371 unsigned char *p = bufp->buffer;
3372 register unsigned char *pend = p + bufp->used;
3374 /* Mark the opcode just after a start_memory, so we can test for an
3375 empty subpattern when we get to the stop_memory. */
3376 unsigned char *just_past_start_mem = 0;
3378 /* We use this to map every character in the string. */
3379 char *translate = bufp->translate;
3381 /* Failure point stack. Each place that can handle a failure further
3382 down the line pushes a failure point on this stack. It consists of
3383 restart, regend, and reg_info for all registers corresponding to
3384 the subexpressions we're currently inside, plus the number of such
3385 registers, and, finally, two char *'s. The first char * is where
3386 to resume scanning the pattern; the second one is where to resume
3387 scanning the strings. If the latter is zero, the failure point is
3388 a ``dummy''; if a failure happens and the failure point is a dummy,
3389 it gets discarded and the next next one is tried. */
3390 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
3391 fail_stack_type fail_stack;
3394 static unsigned failure_id = 0;
3395 unsigned nfailure_points_pushed = 0, nfailure_points_popped = 0;
3398 /* We fill all the registers internally, independent of what we
3399 return, for use in backreferences. The number here includes
3400 an element for register zero. */
3401 unsigned num_regs = bufp->re_nsub + 1;
3403 /* The currently active registers. */
3404 unsigned lowest_active_reg = NO_LOWEST_ACTIVE_REG;
3405 unsigned highest_active_reg = NO_HIGHEST_ACTIVE_REG;
3407 /* Information on the contents of registers. These are pointers into
3408 the input strings; they record just what was matched (on this
3409 attempt) by a subexpression part of the pattern, that is, the
3410 regnum-th regstart pointer points to where in the pattern we began
3411 matching and the regnum-th regend points to right after where we
3412 stopped matching the regnum-th subexpression. (The zeroth register
3413 keeps track of what the whole pattern matches.) */
3414 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
3415 const char **regstart, **regend;
3418 /* If a group that's operated upon by a repetition operator fails to
3419 match anything, then the register for its start will need to be
3420 restored because it will have been set to wherever in the string we
3421 are when we last see its open-group operator. Similarly for a
3423 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
3424 const char **old_regstart, **old_regend;
3427 /* The is_active field of reg_info helps us keep track of which (possibly
3428 nested) subexpressions we are currently in. The matched_something
3429 field of reg_info[reg_num] helps us tell whether or not we have
3430 matched any of the pattern so far this time through the reg_num-th
3431 subexpression. These two fields get reset each time through any
3432 loop their register is in. */
3433 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
3434 register_info_type *reg_info;
3437 /* The following record the register info as found in the above
3438 variables when we find a match better than any we've seen before.
3439 This happens as we backtrack through the failure points, which in
3440 turn happens only if we have not yet matched the entire string. */
3441 unsigned best_regs_set = false;
3442 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
3443 const char **best_regstart, **best_regend;
3446 /* Logically, this is `best_regend[0]'. But we don't want to have to
3447 allocate space for that if we're not allocating space for anything
3448 else (see below). Also, we never need info about register 0 for
3449 any of the other register vectors, and it seems rather a kludge to
3450 treat `best_regend' differently than the rest. So we keep track of
3451 the end of the best match so far in a separate variable. We
3452 initialize this to NULL so that when we backtrack the first time
3453 and need to test it, it's not garbage. */
3454 const char *match_end = NULL;
3456 /* Used when we pop values we don't care about. */
3457 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
3458 const char **reg_dummy;
3459 register_info_type *reg_info_dummy;
3463 /* Counts the total number of registers pushed. */
3464 unsigned num_regs_pushed = 0;
3467 DEBUG_PRINT1 ("\n\nEntering re_match_2.\n");
3471 #ifdef MATCH_MAY_ALLOCATE
3472 /* Do not bother to initialize all the register variables if there are
3473 no groups in the pattern, as it takes a fair amount of time. If
3474 there are groups, we include space for register 0 (the whole
3475 pattern), even though we never use it, since it simplifies the
3476 array indexing. We should fix this. */
3479 regstart = REGEX_TALLOC (num_regs, const char *);
3480 regend = REGEX_TALLOC (num_regs, const char *);
3481 old_regstart = REGEX_TALLOC (num_regs, const char *);
3482 old_regend = REGEX_TALLOC (num_regs, const char *);
3483 best_regstart = REGEX_TALLOC (num_regs, const char *);
3484 best_regend = REGEX_TALLOC (num_regs, const char *);
3485 reg_info = REGEX_TALLOC (num_regs, register_info_type);
3486 reg_dummy = REGEX_TALLOC (num_regs, const char *);
3487 reg_info_dummy = REGEX_TALLOC (num_regs, register_info_type);
3489 if (!(regstart && regend && old_regstart && old_regend && reg_info
3490 && best_regstart && best_regend && reg_dummy && reg_info_dummy))
3496 #if defined (REGEX_MALLOC)
3499 /* We must initialize all our variables to NULL, so that
3500 `FREE_VARIABLES' doesn't try to free them. */
3501 regstart = regend = old_regstart = old_regend = best_regstart
3502 = best_regend = reg_dummy = NULL;
3503 reg_info = reg_info_dummy = (register_info_type *) NULL;
3505 #endif /* REGEX_MALLOC */
3506 #endif /* MATCH_MAY_ALLOCATE */
3508 /* The starting position is bogus. */
3509 if (pos < 0 || pos > size1 + size2)
3515 /* Initialize subexpression text positions to -1 to mark ones that no
3516 start_memory/stop_memory has been seen for. Also initialize the
3517 register information struct. */
3518 for (mcnt = 1; mcnt < num_regs; mcnt++)
3520 regstart[mcnt] = regend[mcnt]
3521 = old_regstart[mcnt] = old_regend[mcnt] = REG_UNSET_VALUE;
3523 REG_MATCH_NULL_STRING_P (reg_info[mcnt]) = MATCH_NULL_UNSET_VALUE;
3524 IS_ACTIVE (reg_info[mcnt]) = 0;
3525 MATCHED_SOMETHING (reg_info[mcnt]) = 0;
3526 EVER_MATCHED_SOMETHING (reg_info[mcnt]) = 0;
3529 /* We move `string1' into `string2' if the latter's empty -- but not if
3530 `string1' is null. */
3531 if (size2 == 0 && string1 != NULL)
3538 end1 = string1 + size1;
3539 end2 = string2 + size2;
3541 /* Compute where to stop matching, within the two strings. */
3544 end_match_1 = string1 + stop;
3545 end_match_2 = string2;
3550 end_match_2 = string2 + stop - size1;
3553 /* `p' scans through the pattern as `d' scans through the data.
3554 `dend' is the end of the input string that `d' points within. `d'
3555 is advanced into the following input string whenever necessary, but
3556 this happens before fetching; therefore, at the beginning of the
3557 loop, `d' can be pointing at the end of a string, but it cannot
3559 if (size1 > 0 && pos <= size1)
3566 d = string2 + pos - size1;
3570 DEBUG_PRINT1 ("The compiled pattern is: ");
3571 DEBUG_PRINT_COMPILED_PATTERN (bufp, p, pend);
3572 DEBUG_PRINT1 ("The string to match is: `");
3573 DEBUG_PRINT_DOUBLE_STRING (d, string1, size1, string2, size2);
3574 DEBUG_PRINT1 ("'\n");
3576 /* This loops over pattern commands. It exits by returning from the
3577 function if the match is complete, or it drops through if the match
3578 fails at this starting point in the input data. */
3581 DEBUG_PRINT2 ("\n0x%x: ", p);
3584 { /* End of pattern means we might have succeeded. */
3585 DEBUG_PRINT1 ("end of pattern ... ");
3587 /* If we haven't matched the entire string, and we want the
3588 longest match, try backtracking. */
3589 if (d != end_match_2)
3591 DEBUG_PRINT1 ("backtracking.\n");
3593 if (!FAIL_STACK_EMPTY ())
3594 { /* More failure points to try. */
3595 boolean same_str_p = (FIRST_STRING_P (match_end)
3596 == MATCHING_IN_FIRST_STRING);
3598 /* If exceeds best match so far, save it. */
3600 || (same_str_p && d > match_end)
3601 || (!same_str_p && !MATCHING_IN_FIRST_STRING))
3603 best_regs_set = true;
3606 DEBUG_PRINT1 ("\nSAVING match as best so far.\n");
3608 for (mcnt = 1; mcnt < num_regs; mcnt++)
3610 best_regstart[mcnt] = regstart[mcnt];
3611 best_regend[mcnt] = regend[mcnt];
3617 /* If no failure points, don't restore garbage. */
3618 else if (best_regs_set)
3621 /* Restore best match. It may happen that `dend ==
3622 end_match_1' while the restored d is in string2.
3623 For example, the pattern `x.*y.*z' against the
3624 strings `x-' and `y-z-', if the two strings are
3625 not consecutive in memory. */
3626 DEBUG_PRINT1 ("Restoring best registers.\n");
3629 dend = ((d >= string1 && d <= end1)
3630 ? end_match_1 : end_match_2);
3632 for (mcnt = 1; mcnt < num_regs; mcnt++)
3634 regstart[mcnt] = best_regstart[mcnt];
3635 regend[mcnt] = best_regend[mcnt];
3638 } /* d != end_match_2 */
3640 DEBUG_PRINT1 ("Accepting match.\n");
3642 /* If caller wants register contents data back, do it. */
3643 if (regs && !bufp->no_sub)
3645 /* Have the register data arrays been allocated? */
3646 if (bufp->regs_allocated == REGS_UNALLOCATED)
3647 { /* No. So allocate them with malloc. We need one
3648 extra element beyond `num_regs' for the `-1' marker
3650 regs->num_regs = MAX (RE_NREGS, num_regs + 1);
3651 regs->start = TALLOC (regs->num_regs, regoff_t);
3652 regs->end = TALLOC (regs->num_regs, regoff_t);
3653 if (regs->start == NULL || regs->end == NULL)
3655 bufp->regs_allocated = REGS_REALLOCATE;
3657 else if (bufp->regs_allocated == REGS_REALLOCATE)
3658 { /* Yes. If we need more elements than were already
3659 allocated, reallocate them. If we need fewer, just
3661 if (regs->num_regs < num_regs + 1)
3663 regs->num_regs = num_regs + 1;
3664 RETALLOC (regs->start, regs->num_regs, regoff_t);
3665 RETALLOC (regs->end, regs->num_regs, regoff_t);
3666 if (regs->start == NULL || regs->end == NULL)
3672 /* These braces fend off a "empty body in an else-statement"
3673 warning under GCC when assert expands to nothing. */
3674 assert (bufp->regs_allocated == REGS_FIXED);
3677 /* Convert the pointer data in `regstart' and `regend' to
3678 indices. Register zero has to be set differently,
3679 since we haven't kept track of any info for it. */
3680 if (regs->num_regs > 0)
3682 regs->start[0] = pos;
3683 regs->end[0] = (MATCHING_IN_FIRST_STRING
3684 ? ((regoff_t) (d - string1))
3685 : ((regoff_t) (d - string2 + size1)));
3688 /* Go through the first `min (num_regs, regs->num_regs)'
3689 registers, since that is all we initialized. */
3690 for (mcnt = 1; mcnt < MIN (num_regs, regs->num_regs); mcnt++)
3692 if (REG_UNSET (regstart[mcnt]) || REG_UNSET (regend[mcnt]))
3693 regs->start[mcnt] = regs->end[mcnt] = -1;
3697 = (regoff_t) POINTER_TO_OFFSET (regstart[mcnt]);
3699 = (regoff_t) POINTER_TO_OFFSET (regend[mcnt]);
3703 /* If the regs structure we return has more elements than
3704 were in the pattern, set the extra elements to -1. If
3705 we (re)allocated the registers, this is the case,
3706 because we always allocate enough to have at least one
3708 for (mcnt = num_regs; mcnt < regs->num_regs; mcnt++)
3709 regs->start[mcnt] = regs->end[mcnt] = -1;
3710 } /* regs && !bufp->no_sub */
3713 DEBUG_PRINT4 ("%u failure points pushed, %u popped (%u remain).\n",
3714 nfailure_points_pushed, nfailure_points_popped,
3715 nfailure_points_pushed - nfailure_points_popped);
3716 DEBUG_PRINT2 ("%u registers pushed.\n", num_regs_pushed);
3718 mcnt = d - pos - (MATCHING_IN_FIRST_STRING
3722 DEBUG_PRINT2 ("Returning %d from re_match_2.\n", mcnt);
3727 /* Otherwise match next pattern command. */
3728 #ifdef SWITCH_ENUM_BUG
3729 switch ((int) ((re_opcode_t) *p++))
3731 switch ((re_opcode_t) *p++)
3734 /* Ignore these. Used to ignore the n of succeed_n's which
3735 currently have n == 0. */
3737 DEBUG_PRINT1 ("EXECUTING no_op.\n");
3741 /* Match the next n pattern characters exactly. The following
3742 byte in the pattern defines n, and the n bytes after that
3743 are the characters to match. */
3746 DEBUG_PRINT2 ("EXECUTING exactn %d.\n", mcnt);
3748 /* This is written out as an if-else so we don't waste time
3749 testing `translate' inside the loop. */
3755 if (translate[(unsigned char) *d++] != (char) *p++)
3765 if (*d++ != (char) *p++) goto fail;
3769 SET_REGS_MATCHED ();
3773 /* Match any character except possibly a newline or a null. */
3775 DEBUG_PRINT1 ("EXECUTING anychar.\n");
3779 if ((!(bufp->syntax & RE_DOT_NEWLINE) && TRANSLATE (*d) == '\n')
3780 || (bufp->syntax & RE_DOT_NOT_NULL && TRANSLATE (*d) == '\000'))
3783 SET_REGS_MATCHED ();
3784 DEBUG_PRINT2 (" Matched `%d'.\n", *d);
3792 register unsigned char c;
3793 boolean not = (re_opcode_t) *(p - 1) == charset_not;
3795 DEBUG_PRINT2 ("EXECUTING charset%s.\n", not ? "_not" : "");
3798 c = TRANSLATE (*d); /* The character to match. */
3800 /* Cast to `unsigned' instead of `unsigned char' in case the
3801 bit list is a full 32 bytes long. */
3802 if (c < (unsigned) (*p * BYTEWIDTH)
3803 && p[1 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH)))
3808 if (!not) goto fail;
3810 SET_REGS_MATCHED ();
3816 /* The beginning of a group is represented by start_memory.
3817 The arguments are the register number in the next byte, and the
3818 number of groups inner to this one in the next. The text
3819 matched within the group is recorded (in the internal
3820 registers data structure) under the register number. */
3822 DEBUG_PRINT3 ("EXECUTING start_memory %d (%d):\n", *p, p[1]);
3824 /* Find out if this group can match the empty string. */
3825 p1 = p; /* To send to group_match_null_string_p. */
3827 if (REG_MATCH_NULL_STRING_P (reg_info[*p]) == MATCH_NULL_UNSET_VALUE)
3828 REG_MATCH_NULL_STRING_P (reg_info[*p])
3829 = group_match_null_string_p (&p1, pend, reg_info);
3831 /* Save the position in the string where we were the last time
3832 we were at this open-group operator in case the group is
3833 operated upon by a repetition operator, e.g., with `(a*)*b'
3834 against `ab'; then we want to ignore where we are now in
3835 the string in case this attempt to match fails. */
3836 old_regstart[*p] = REG_MATCH_NULL_STRING_P (reg_info[*p])
3837 ? REG_UNSET (regstart[*p]) ? d : regstart[*p]
3839 DEBUG_PRINT2 (" old_regstart: %d\n",
3840 POINTER_TO_OFFSET (old_regstart[*p]));
3843 DEBUG_PRINT2 (" regstart: %d\n", POINTER_TO_OFFSET (regstart[*p]));
3845 IS_ACTIVE (reg_info[*p]) = 1;
3846 MATCHED_SOMETHING (reg_info[*p]) = 0;
3848 /* This is the new highest active register. */
3849 highest_active_reg = *p;
3851 /* If nothing was active before, this is the new lowest active
3853 if (lowest_active_reg == NO_LOWEST_ACTIVE_REG)
3854 lowest_active_reg = *p;
3856 /* Move past the register number and inner group count. */
3858 just_past_start_mem = p;
3862 /* The stop_memory opcode represents the end of a group. Its
3863 arguments are the same as start_memory's: the register
3864 number, and the number of inner groups. */
3866 DEBUG_PRINT3 ("EXECUTING stop_memory %d (%d):\n", *p, p[1]);
3868 /* We need to save the string position the last time we were at
3869 this close-group operator in case the group is operated
3870 upon by a repetition operator, e.g., with `((a*)*(b*)*)*'
3871 against `aba'; then we want to ignore where we are now in
3872 the string in case this attempt to match fails. */
3873 old_regend[*p] = REG_MATCH_NULL_STRING_P (reg_info[*p])
3874 ? REG_UNSET (regend[*p]) ? d : regend[*p]
3876 DEBUG_PRINT2 (" old_regend: %d\n",
3877 POINTER_TO_OFFSET (old_regend[*p]));
3880 DEBUG_PRINT2 (" regend: %d\n", POINTER_TO_OFFSET (regend[*p]));
3882 /* This register isn't active anymore. */
3883 IS_ACTIVE (reg_info[*p]) = 0;
3885 /* If this was the only register active, nothing is active
3887 if (lowest_active_reg == highest_active_reg)
3889 lowest_active_reg = NO_LOWEST_ACTIVE_REG;
3890 highest_active_reg = NO_HIGHEST_ACTIVE_REG;
3893 { /* We must scan for the new highest active register, since
3894 it isn't necessarily one less than now: consider
3895 (a(b)c(d(e)f)g). When group 3 ends, after the f), the
3896 new highest active register is 1. */
3897 unsigned char r = *p - 1;
3898 while (r > 0 && !IS_ACTIVE (reg_info[r]))
3901 /* If we end up at register zero, that means that we saved
3902 the registers as the result of an `on_failure_jump', not
3903 a `start_memory', and we jumped to past the innermost
3904 `stop_memory'. For example, in ((.)*) we save
3905 registers 1 and 2 as a result of the *, but when we pop
3906 back to the second ), we are at the stop_memory 1.
3907 Thus, nothing is active. */
3910 lowest_active_reg = NO_LOWEST_ACTIVE_REG;
3911 highest_active_reg = NO_HIGHEST_ACTIVE_REG;
3914 highest_active_reg = r;
3917 /* If just failed to match something this time around with a
3918 group that's operated on by a repetition operator, try to
3919 force exit from the ``loop'', and restore the register
3920 information for this group that we had before trying this
3922 if ((!MATCHED_SOMETHING (reg_info[*p])
3923 || just_past_start_mem == p - 1)
3926 boolean is_a_jump_n = false;
3930 switch ((re_opcode_t) *p1++)
3934 case pop_failure_jump:
3935 case maybe_pop_jump:
3937 case dummy_failure_jump:
3938 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
3948 /* If the next operation is a jump backwards in the pattern
3949 to an on_failure_jump right before the start_memory
3950 corresponding to this stop_memory, exit from the loop
3951 by forcing a failure after pushing on the stack the
3952 on_failure_jump's jump in the pattern, and d. */
3953 if (mcnt < 0 && (re_opcode_t) *p1 == on_failure_jump
3954 && (re_opcode_t) p1[3] == start_memory && p1[4] == *p)
3956 /* If this group ever matched anything, then restore
3957 what its registers were before trying this last
3958 failed match, e.g., with `(a*)*b' against `ab' for
3959 regstart[1], and, e.g., with `((a*)*(b*)*)*'
3960 against `aba' for regend[3].
3962 Also restore the registers for inner groups for,
3963 e.g., `((a*)(b*))*' against `aba' (register 3 would
3964 otherwise get trashed). */
3966 if (EVER_MATCHED_SOMETHING (reg_info[*p]))
3970 EVER_MATCHED_SOMETHING (reg_info[*p]) = 0;
3972 /* Restore this and inner groups' (if any) registers. */
3973 for (r = *p; r < *p + *(p + 1); r++)
3975 regstart[r] = old_regstart[r];
3977 /* xx why this test? */
3978 if (old_regend[r] >= regstart[r])
3979 regend[r] = old_regend[r];
3983 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
3984 PUSH_FAILURE_POINT (p1 + mcnt, d, -2);
3990 /* Move past the register number and the inner group count. */
3995 /* \<digit> has been turned into a `duplicate' command which is
3996 followed by the numeric value of <digit> as the register number. */
3999 register const char *d2, *dend2;
4000 int regno = *p++; /* Get which register to match against. */
4001 DEBUG_PRINT2 ("EXECUTING duplicate %d.\n", regno);
4003 /* Can't back reference a group which we've never matched. */
4004 if (REG_UNSET (regstart[regno]) || REG_UNSET (regend[regno]))
4007 /* Where in input to try to start matching. */
4008 d2 = regstart[regno];
4010 /* Where to stop matching; if both the place to start and
4011 the place to stop matching are in the same string, then
4012 set to the place to stop, otherwise, for now have to use
4013 the end of the first string. */
4015 dend2 = ((FIRST_STRING_P (regstart[regno])
4016 == FIRST_STRING_P (regend[regno]))
4017 ? regend[regno] : end_match_1);
4020 /* If necessary, advance to next segment in register
4024 if (dend2 == end_match_2) break;
4025 if (dend2 == regend[regno]) break;
4027 /* End of string1 => advance to string2. */
4029 dend2 = regend[regno];
4031 /* At end of register contents => success */
4032 if (d2 == dend2) break;
4034 /* If necessary, advance to next segment in data. */
4037 /* How many characters left in this segment to match. */
4040 /* Want how many consecutive characters we can match in
4041 one shot, so, if necessary, adjust the count. */
4042 if (mcnt > dend2 - d2)
4045 /* Compare that many; failure if mismatch, else move
4048 ? bcmp_translate (d, d2, mcnt, translate)
4049 : bcmp (d, d2, mcnt))
4051 d += mcnt, d2 += mcnt;
4057 /* begline matches the empty string at the beginning of the string
4058 (unless `not_bol' is set in `bufp'), and, if
4059 `newline_anchor' is set, after newlines. */
4061 DEBUG_PRINT1 ("EXECUTING begline.\n");
4063 if (AT_STRINGS_BEG (d))
4065 if (!bufp->not_bol) break;
4067 else if (d[-1] == '\n' && bufp->newline_anchor)
4071 /* In all other cases, we fail. */
4075 /* endline is the dual of begline. */
4077 DEBUG_PRINT1 ("EXECUTING endline.\n");
4079 if (AT_STRINGS_END (d))
4081 if (!bufp->not_eol) break;
4084 /* We have to ``prefetch'' the next character. */
4085 else if ((d == end1 ? *string2 : *d) == '\n'
4086 && bufp->newline_anchor)
4093 /* Match at the very beginning of the data. */
4095 DEBUG_PRINT1 ("EXECUTING begbuf.\n");
4096 if (AT_STRINGS_BEG (d))
4101 /* Match at the very end of the data. */
4103 DEBUG_PRINT1 ("EXECUTING endbuf.\n");
4104 if (AT_STRINGS_END (d))
4109 /* on_failure_keep_string_jump is used to optimize `.*\n'. It
4110 pushes NULL as the value for the string on the stack. Then
4111 `pop_failure_point' will keep the current value for the
4112 string, instead of restoring it. To see why, consider
4113 matching `foo\nbar' against `.*\n'. The .* matches the foo;
4114 then the . fails against the \n. But the next thing we want
4115 to do is match the \n against the \n; if we restored the
4116 string value, we would be back at the foo.
4118 Because this is used only in specific cases, we don't need to
4119 check all the things that `on_failure_jump' does, to make
4120 sure the right things get saved on the stack. Hence we don't
4121 share its code. The only reason to push anything on the
4122 stack at all is that otherwise we would have to change
4123 `anychar's code to do something besides goto fail in this
4124 case; that seems worse than this. */
4125 case on_failure_keep_string_jump:
4126 DEBUG_PRINT1 ("EXECUTING on_failure_keep_string_jump");
4128 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4129 DEBUG_PRINT3 (" %d (to 0x%x):\n", mcnt, p + mcnt);
4131 PUSH_FAILURE_POINT (p + mcnt, NULL, -2);
4135 /* Uses of on_failure_jump:
4137 Each alternative starts with an on_failure_jump that points
4138 to the beginning of the next alternative. Each alternative
4139 except the last ends with a jump that in effect jumps past
4140 the rest of the alternatives. (They really jump to the
4141 ending jump of the following alternative, because tensioning
4142 these jumps is a hassle.)
4144 Repeats start with an on_failure_jump that points past both
4145 the repetition text and either the following jump or
4146 pop_failure_jump back to this on_failure_jump. */
4147 case on_failure_jump:
4149 DEBUG_PRINT1 ("EXECUTING on_failure_jump");
4151 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4152 DEBUG_PRINT3 (" %d (to 0x%x)", mcnt, p + mcnt);
4154 /* If this on_failure_jump comes right before a group (i.e.,
4155 the original * applied to a group), save the information
4156 for that group and all inner ones, so that if we fail back
4157 to this point, the group's information will be correct.
4158 For example, in \(a*\)*\1, we need the preceding group,
4159 and in \(\(a*\)b*\)\2, we need the inner group. */
4161 /* We can't use `p' to check ahead because we push
4162 a failure point to `p + mcnt' after we do this. */
4165 /* We need to skip no_op's before we look for the
4166 start_memory in case this on_failure_jump is happening as
4167 the result of a completed succeed_n, as in \(a\)\{1,3\}b\1
4169 while (p1 < pend && (re_opcode_t) *p1 == no_op)
4172 if (p1 < pend && (re_opcode_t) *p1 == start_memory)
4174 /* We have a new highest active register now. This will
4175 get reset at the start_memory we are about to get to,
4176 but we will have saved all the registers relevant to
4177 this repetition op, as described above. */
4178 highest_active_reg = *(p1 + 1) + *(p1 + 2);
4179 if (lowest_active_reg == NO_LOWEST_ACTIVE_REG)
4180 lowest_active_reg = *(p1 + 1);
4183 DEBUG_PRINT1 (":\n");
4184 PUSH_FAILURE_POINT (p + mcnt, d, -2);
4188 /* A smart repeat ends with `maybe_pop_jump'.
4189 We change it to either `pop_failure_jump' or `jump'. */
4190 case maybe_pop_jump:
4191 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4192 DEBUG_PRINT2 ("EXECUTING maybe_pop_jump %d.\n", mcnt);
4194 register unsigned char *p2 = p;
4196 /* Compare the beginning of the repeat with what in the
4197 pattern follows its end. If we can establish that there
4198 is nothing that they would both match, i.e., that we
4199 would have to backtrack because of (as in, e.g., `a*a')
4200 then we can change to pop_failure_jump, because we'll
4201 never have to backtrack.
4203 This is not true in the case of alternatives: in
4204 `(a|ab)*' we do need to backtrack to the `ab' alternative
4205 (e.g., if the string was `ab'). But instead of trying to
4206 detect that here, the alternative has put on a dummy
4207 failure point which is what we will end up popping. */
4209 /* Skip over open/close-group commands.
4210 If what follows this loop is a ...+ construct,
4211 look at what begins its body, since we will have to
4212 match at least one of that. */
4216 && ((re_opcode_t) *p2 == stop_memory
4217 || (re_opcode_t) *p2 == start_memory))
4219 else if (p2 + 6 < pend
4220 && (re_opcode_t) *p2 == dummy_failure_jump)
4227 /* p1[0] ... p1[2] are the `on_failure_jump' corresponding
4228 to the `maybe_finalize_jump' of this case. Examine what
4231 /* If we're at the end of the pattern, we can change. */
4234 /* Consider what happens when matching ":\(.*\)"
4235 against ":/". I don't really understand this code
4237 p[-3] = (unsigned char) pop_failure_jump;
4239 (" End of pattern: change to `pop_failure_jump'.\n");
4242 else if ((re_opcode_t) *p2 == exactn
4243 || (bufp->newline_anchor && (re_opcode_t) *p2 == endline))
4245 register unsigned char c
4246 = *p2 == (unsigned char) endline ? '\n' : p2[2];
4248 if ((re_opcode_t) p1[3] == exactn && p1[5] != c)
4250 p[-3] = (unsigned char) pop_failure_jump;
4251 DEBUG_PRINT3 (" %c != %c => pop_failure_jump.\n",
4255 else if ((re_opcode_t) p1[3] == charset
4256 || (re_opcode_t) p1[3] == charset_not)
4258 int not = (re_opcode_t) p1[3] == charset_not;
4260 if (c < (unsigned char) (p1[4] * BYTEWIDTH)
4261 && p1[5 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH)))
4264 /* `not' is equal to 1 if c would match, which means
4265 that we can't change to pop_failure_jump. */
4268 p[-3] = (unsigned char) pop_failure_jump;
4269 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
4273 else if ((re_opcode_t) *p2 == charset)
4276 register unsigned char c
4277 = *p2 == (unsigned char) endline ? '\n' : p2[2];
4280 if ((re_opcode_t) p1[3] == exactn
4281 && ! (p2[1] * BYTEWIDTH > p1[4]
4282 && (p2[1 + p1[4] / BYTEWIDTH]
4283 & (1 << (p1[4] % BYTEWIDTH)))))
4285 p[-3] = (unsigned char) pop_failure_jump;
4286 DEBUG_PRINT3 (" %c != %c => pop_failure_jump.\n",
4290 else if ((re_opcode_t) p1[3] == charset_not)
4293 /* We win if the charset_not inside the loop
4294 lists every character listed in the charset after. */
4295 for (idx = 0; idx < p2[1]; idx++)
4296 if (! (p2[2 + idx] == 0
4298 && ((p2[2 + idx] & ~ p1[5 + idx]) == 0))))
4303 p[-3] = (unsigned char) pop_failure_jump;
4304 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
4307 else if ((re_opcode_t) p1[3] == charset)
4310 /* We win if the charset inside the loop
4311 has no overlap with the one after the loop. */
4312 for (idx = 0; idx < p2[1] && idx < p1[4]; idx++)
4313 if ((p2[2 + idx] & p1[5 + idx]) != 0)
4316 if (idx == p2[1] || idx == p1[4])
4318 p[-3] = (unsigned char) pop_failure_jump;
4319 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
4324 p -= 2; /* Point at relative address again. */
4325 if ((re_opcode_t) p[-1] != pop_failure_jump)
4327 p[-1] = (unsigned char) jump;
4328 DEBUG_PRINT1 (" Match => jump.\n");
4329 goto unconditional_jump;
4331 /* Note fall through. */
4334 /* The end of a simple repeat has a pop_failure_jump back to
4335 its matching on_failure_jump, where the latter will push a
4336 failure point. The pop_failure_jump takes off failure
4337 points put on by this pop_failure_jump's matching
4338 on_failure_jump; we got through the pattern to here from the
4339 matching on_failure_jump, so didn't fail. */
4340 case pop_failure_jump:
4342 /* We need to pass separate storage for the lowest and
4343 highest registers, even though we don't care about the
4344 actual values. Otherwise, we will restore only one
4345 register from the stack, since lowest will == highest in
4346 `pop_failure_point'. */
4347 unsigned long dummy_low_reg, dummy_high_reg;
4348 unsigned char *pdummy;
4351 DEBUG_PRINT1 ("EXECUTING pop_failure_jump.\n");
4352 POP_FAILURE_POINT (sdummy, pdummy,
4353 dummy_low_reg, dummy_high_reg,
4354 reg_dummy, reg_dummy, reg_info_dummy);
4356 /* Note fall through. */
4359 /* Unconditionally jump (without popping any failure points). */
4362 EXTRACT_NUMBER_AND_INCR (mcnt, p); /* Get the amount to jump. */
4363 DEBUG_PRINT2 ("EXECUTING jump %d ", mcnt);
4364 p += mcnt; /* Do the jump. */
4365 DEBUG_PRINT2 ("(to 0x%x).\n", p);
4369 /* We need this opcode so we can detect where alternatives end
4370 in `group_match_null_string_p' et al. */
4372 DEBUG_PRINT1 ("EXECUTING jump_past_alt.\n");
4373 goto unconditional_jump;
4376 /* Normally, the on_failure_jump pushes a failure point, which
4377 then gets popped at pop_failure_jump. We will end up at
4378 pop_failure_jump, also, and with a pattern of, say, `a+', we
4379 are skipping over the on_failure_jump, so we have to push
4380 something meaningless for pop_failure_jump to pop. */
4381 case dummy_failure_jump:
4382 DEBUG_PRINT1 ("EXECUTING dummy_failure_jump.\n");
4383 /* It doesn't matter what we push for the string here. What
4384 the code at `fail' tests is the value for the pattern. */
4385 PUSH_FAILURE_POINT (0, 0, -2);
4386 goto unconditional_jump;
4389 /* At the end of an alternative, we need to push a dummy failure
4390 point in case we are followed by a `pop_failure_jump', because
4391 we don't want the failure point for the alternative to be
4392 popped. For example, matching `(a|ab)*' against `aab'
4393 requires that we match the `ab' alternative. */
4394 case push_dummy_failure:
4395 DEBUG_PRINT1 ("EXECUTING push_dummy_failure.\n");
4396 /* See comments just above at `dummy_failure_jump' about the
4398 PUSH_FAILURE_POINT (0, 0, -2);
4401 /* Have to succeed matching what follows at least n times.
4402 After that, handle like `on_failure_jump'. */
4404 EXTRACT_NUMBER (mcnt, p + 2);
4405 DEBUG_PRINT2 ("EXECUTING succeed_n %d.\n", mcnt);
4408 /* Originally, this is how many times we HAVE to succeed. */
4413 STORE_NUMBER_AND_INCR (p, mcnt);
4414 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p, mcnt);
4418 DEBUG_PRINT2 (" Setting two bytes from 0x%x to no_op.\n", p+2);
4419 p[2] = (unsigned char) no_op;
4420 p[3] = (unsigned char) no_op;
4426 EXTRACT_NUMBER (mcnt, p + 2);
4427 DEBUG_PRINT2 ("EXECUTING jump_n %d.\n", mcnt);
4429 /* Originally, this is how many times we CAN jump. */
4433 STORE_NUMBER (p + 2, mcnt);
4434 goto unconditional_jump;
4436 /* If don't have to jump any more, skip over the rest of command. */
4443 DEBUG_PRINT1 ("EXECUTING set_number_at.\n");
4445 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4447 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4448 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p1, mcnt);
4449 STORE_NUMBER (p1, mcnt);
4454 DEBUG_PRINT1 ("EXECUTING wordbound.\n");
4455 if (AT_WORD_BOUNDARY (d))
4460 DEBUG_PRINT1 ("EXECUTING notwordbound.\n");
4461 if (AT_WORD_BOUNDARY (d))
4466 DEBUG_PRINT1 ("EXECUTING wordbeg.\n");
4467 if (WORDCHAR_P (d) && (AT_STRINGS_BEG (d) || !WORDCHAR_P (d - 1)))
4472 DEBUG_PRINT1 ("EXECUTING wordend.\n");
4473 if (!AT_STRINGS_BEG (d) && WORDCHAR_P (d - 1)
4474 && (!WORDCHAR_P (d) || AT_STRINGS_END (d)))
4480 DEBUG_PRINT1 ("EXECUTING before_dot.\n");
4481 if (PTR_CHAR_POS ((unsigned char *) d) >= point)
4486 DEBUG_PRINT1 ("EXECUTING at_dot.\n");
4487 if (PTR_CHAR_POS ((unsigned char *) d) != point)
4492 DEBUG_PRINT1 ("EXECUTING after_dot.\n");
4493 if (PTR_CHAR_POS ((unsigned char *) d) <= point)
4496 #if 0 /* not emacs19 */
4498 DEBUG_PRINT1 ("EXECUTING at_dot.\n");
4499 if (PTR_CHAR_POS ((unsigned char *) d) + 1 != point)
4502 #endif /* not emacs19 */
4505 DEBUG_PRINT2 ("EXECUTING syntaxspec %d.\n", mcnt);
4510 DEBUG_PRINT1 ("EXECUTING Emacs wordchar.\n");
4514 /* Can't use *d++ here; SYNTAX may be an unsafe macro. */
4516 if (SYNTAX (d[-1]) != (enum syntaxcode) mcnt)
4518 SET_REGS_MATCHED ();
4522 DEBUG_PRINT2 ("EXECUTING notsyntaxspec %d.\n", mcnt);
4524 goto matchnotsyntax;
4527 DEBUG_PRINT1 ("EXECUTING Emacs notwordchar.\n");
4531 /* Can't use *d++ here; SYNTAX may be an unsafe macro. */
4533 if (SYNTAX (d[-1]) == (enum syntaxcode) mcnt)
4535 SET_REGS_MATCHED ();
4538 #else /* not emacs */
4540 DEBUG_PRINT1 ("EXECUTING non-Emacs wordchar.\n");
4542 if (!WORDCHAR_P (d))
4544 SET_REGS_MATCHED ();
4549 DEBUG_PRINT1 ("EXECUTING non-Emacs notwordchar.\n");
4553 SET_REGS_MATCHED ();
4556 #endif /* not emacs */
4561 continue; /* Successfully executed one pattern command; keep going. */
4564 /* We goto here if a matching operation fails. */
4566 if (!FAIL_STACK_EMPTY ())
4567 { /* A restart point is known. Restore to that state. */
4568 DEBUG_PRINT1 ("\nFAIL:\n");
4569 POP_FAILURE_POINT (d, p,
4570 lowest_active_reg, highest_active_reg,
4571 regstart, regend, reg_info);
4573 /* If this failure point is a dummy, try the next one. */
4577 /* If we failed to the end of the pattern, don't examine *p. */
4581 boolean is_a_jump_n = false;
4583 /* If failed to a backwards jump that's part of a repetition
4584 loop, need to pop this failure point and use the next one. */
4585 switch ((re_opcode_t) *p)
4589 case maybe_pop_jump:
4590 case pop_failure_jump:
4593 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4596 if ((is_a_jump_n && (re_opcode_t) *p1 == succeed_n)
4598 && (re_opcode_t) *p1 == on_failure_jump))
4606 if (d >= string1 && d <= end1)
4610 break; /* Matching at this starting point really fails. */
4614 goto restore_best_regs;
4618 return -1; /* Failure to match. */
4621 /* Subroutine definitions for re_match_2. */
4624 /* We are passed P pointing to a register number after a start_memory.
4626 Return true if the pattern up to the corresponding stop_memory can
4627 match the empty string, and false otherwise.
4629 If we find the matching stop_memory, sets P to point to one past its number.
4630 Otherwise, sets P to an undefined byte less than or equal to END.
4632 We don't handle duplicates properly (yet). */
4635 group_match_null_string_p (p, end, reg_info)
4636 unsigned char **p, *end;
4637 register_info_type *reg_info;
4640 /* Point to after the args to the start_memory. */
4641 unsigned char *p1 = *p + 2;
4645 /* Skip over opcodes that can match nothing, and return true or
4646 false, as appropriate, when we get to one that can't, or to the
4647 matching stop_memory. */
4649 switch ((re_opcode_t) *p1)
4651 /* Could be either a loop or a series of alternatives. */
4652 case on_failure_jump:
4654 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4656 /* If the next operation is not a jump backwards in the
4661 /* Go through the on_failure_jumps of the alternatives,
4662 seeing if any of the alternatives cannot match nothing.
4663 The last alternative starts with only a jump,
4664 whereas the rest start with on_failure_jump and end
4665 with a jump, e.g., here is the pattern for `a|b|c':
4667 /on_failure_jump/0/6/exactn/1/a/jump_past_alt/0/6
4668 /on_failure_jump/0/6/exactn/1/b/jump_past_alt/0/3
4671 So, we have to first go through the first (n-1)
4672 alternatives and then deal with the last one separately. */
4675 /* Deal with the first (n-1) alternatives, which start
4676 with an on_failure_jump (see above) that jumps to right
4677 past a jump_past_alt. */
4679 while ((re_opcode_t) p1[mcnt-3] == jump_past_alt)
4681 /* `mcnt' holds how many bytes long the alternative
4682 is, including the ending `jump_past_alt' and
4685 if (!alt_match_null_string_p (p1, p1 + mcnt - 3,
4689 /* Move to right after this alternative, including the
4693 /* Break if it's the beginning of an n-th alternative
4694 that doesn't begin with an on_failure_jump. */
4695 if ((re_opcode_t) *p1 != on_failure_jump)
4698 /* Still have to check that it's not an n-th
4699 alternative that starts with an on_failure_jump. */
4701 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4702 if ((re_opcode_t) p1[mcnt-3] != jump_past_alt)
4704 /* Get to the beginning of the n-th alternative. */
4710 /* Deal with the last alternative: go back and get number
4711 of the `jump_past_alt' just before it. `mcnt' contains
4712 the length of the alternative. */
4713 EXTRACT_NUMBER (mcnt, p1 - 2);
4715 if (!alt_match_null_string_p (p1, p1 + mcnt, reg_info))
4718 p1 += mcnt; /* Get past the n-th alternative. */
4724 assert (p1[1] == **p);
4730 if (!common_op_match_null_string_p (&p1, end, reg_info))
4733 } /* while p1 < end */
4736 } /* group_match_null_string_p */
4739 /* Similar to group_match_null_string_p, but doesn't deal with alternatives:
4740 It expects P to be the first byte of a single alternative and END one
4741 byte past the last. The alternative can contain groups. */
4744 alt_match_null_string_p (p, end, reg_info)
4745 unsigned char *p, *end;
4746 register_info_type *reg_info;
4749 unsigned char *p1 = p;
4753 /* Skip over opcodes that can match nothing, and break when we get
4754 to one that can't. */
4756 switch ((re_opcode_t) *p1)
4759 case on_failure_jump:
4761 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4766 if (!common_op_match_null_string_p (&p1, end, reg_info))
4769 } /* while p1 < end */
4772 } /* alt_match_null_string_p */
4775 /* Deals with the ops common to group_match_null_string_p and
4776 alt_match_null_string_p.
4778 Sets P to one after the op and its arguments, if any. */
4781 common_op_match_null_string_p (p, end, reg_info)
4782 unsigned char **p, *end;
4783 register_info_type *reg_info;
4788 unsigned char *p1 = *p;
4790 switch ((re_opcode_t) *p1++)
4810 assert (reg_no > 0 && reg_no <= MAX_REGNUM);
4811 ret = group_match_null_string_p (&p1, end, reg_info);
4813 /* Have to set this here in case we're checking a group which
4814 contains a group and a back reference to it. */
4816 if (REG_MATCH_NULL_STRING_P (reg_info[reg_no]) == MATCH_NULL_UNSET_VALUE)
4817 REG_MATCH_NULL_STRING_P (reg_info[reg_no]) = ret;
4823 /* If this is an optimized succeed_n for zero times, make the jump. */
4825 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4833 /* Get to the number of times to succeed. */
4835 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4840 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4848 if (!REG_MATCH_NULL_STRING_P (reg_info[*p1]))
4856 /* All other opcodes mean we cannot match the empty string. */
4862 } /* common_op_match_null_string_p */
4865 /* Return zero if TRANSLATE[S1] and TRANSLATE[S2] are identical for LEN
4866 bytes; nonzero otherwise. */
4869 bcmp_translate (s1, s2, len, translate)
4870 unsigned char *s1, *s2;
4874 register unsigned char *p1 = s1, *p2 = s2;
4877 if (translate[*p1++] != translate[*p2++]) return 1;
4883 /* Entry points for GNU code. */
4885 /* re_compile_pattern is the GNU regular expression compiler: it
4886 compiles PATTERN (of length SIZE) and puts the result in BUFP.
4887 Returns 0 if the pattern was valid, otherwise an error string.
4889 Assumes the `allocated' (and perhaps `buffer') and `translate' fields
4890 are set in BUFP on entry.
4892 We call regex_compile to do the actual compilation. */
4895 re_compile_pattern (pattern, length, bufp)
4896 const char *pattern;
4898 struct re_pattern_buffer *bufp;
4902 /* GNU code is written to assume at least RE_NREGS registers will be set
4903 (and at least one extra will be -1). */
4904 bufp->regs_allocated = REGS_UNALLOCATED;
4906 /* And GNU code determines whether or not to get register information
4907 by passing null for the REGS argument to re_match, etc., not by
4911 /* Match anchors at newline. */
4912 bufp->newline_anchor = 1;
4914 ret = regex_compile (pattern, length, re_syntax_options, bufp);
4916 return re_error_msg[(int) ret];
4919 /* Entry points compatible with 4.2 BSD regex library. We don't define
4920 them if this is an Emacs or POSIX compilation. */
4922 #if !defined (emacs) && !defined (_POSIX_SOURCE)
4924 /* BSD has one and only one pattern buffer. */
4925 static struct re_pattern_buffer re_comp_buf;
4935 if (!re_comp_buf.buffer)
4936 return "No previous regular expression";
4940 if (!re_comp_buf.buffer)
4942 re_comp_buf.buffer = (unsigned char *) malloc (200);
4943 if (re_comp_buf.buffer == NULL)
4944 return "Memory exhausted";
4945 re_comp_buf.allocated = 200;
4947 re_comp_buf.fastmap = (char *) malloc (1 << BYTEWIDTH);
4948 if (re_comp_buf.fastmap == NULL)
4949 return "Memory exhausted";
4952 /* Since `re_exec' always passes NULL for the `regs' argument, we
4953 don't need to initialize the pattern buffer fields which affect it. */
4955 /* Match anchors at newlines. */
4956 re_comp_buf.newline_anchor = 1;
4958 ret = regex_compile (s, strlen (s), re_syntax_options, &re_comp_buf);
4960 /* Yes, we're discarding `const' here. */
4961 return (char *) re_error_msg[(int) ret];
4969 const int len = strlen (s);
4971 0 <= re_search (&re_comp_buf, s, len, 0, len, (struct re_registers *) 0);
4973 #endif /* not emacs and not _POSIX_SOURCE */
4975 /* POSIX.2 functions. Don't define these for Emacs. */
4979 /* regcomp takes a regular expression as a string and compiles it.
4981 PREG is a regex_t *. We do not expect any fields to be initialized,
4982 since POSIX says we shouldn't. Thus, we set
4984 `buffer' to the compiled pattern;
4985 `used' to the length of the compiled pattern;
4986 `syntax' to RE_SYNTAX_POSIX_EXTENDED if the
4987 REG_EXTENDED bit in CFLAGS is set; otherwise, to
4988 RE_SYNTAX_POSIX_BASIC;
4989 `newline_anchor' to REG_NEWLINE being set in CFLAGS;
4990 `fastmap' and `fastmap_accurate' to zero;
4991 `re_nsub' to the number of subexpressions in PATTERN.
4993 PATTERN is the address of the pattern string.
4995 CFLAGS is a series of bits which affect compilation.
4997 If REG_EXTENDED is set, we use POSIX extended syntax; otherwise, we
4998 use POSIX basic syntax.
5000 If REG_NEWLINE is set, then . and [^...] don't match newline.
5001 Also, regexec will try a match beginning after every newline.
5003 If REG_ICASE is set, then we considers upper- and lowercase
5004 versions of letters to be equivalent when matching.
5006 If REG_NOSUB is set, then when PREG is passed to regexec, that
5007 routine will report only success or failure, and nothing about the
5010 It returns 0 if it succeeds, nonzero if it doesn't. (See regex.h for
5011 the return codes and their meanings.) */
5014 regcomp (preg, pattern, cflags)
5016 const char *pattern;
5021 = (cflags & REG_EXTENDED) ?
5022 RE_SYNTAX_POSIX_EXTENDED : RE_SYNTAX_POSIX_BASIC;
5024 /* regex_compile will allocate the space for the compiled pattern. */
5026 preg->allocated = 0;
5029 /* Don't bother to use a fastmap when searching. This simplifies the
5030 REG_NEWLINE case: if we used a fastmap, we'd have to put all the
5031 characters after newlines into the fastmap. This way, we just try
5035 if (cflags & REG_ICASE)
5039 preg->translate = (char *) malloc (CHAR_SET_SIZE);
5040 if (preg->translate == NULL)
5041 return (int) REG_ESPACE;
5043 /* Map uppercase characters to corresponding lowercase ones. */
5044 for (i = 0; i < CHAR_SET_SIZE; i++)
5045 preg->translate[i] = ISUPPER (i) ? tolower (i) : i;
5048 preg->translate = NULL;
5050 /* If REG_NEWLINE is set, newlines are treated differently. */
5051 if (cflags & REG_NEWLINE)
5052 { /* REG_NEWLINE implies neither . nor [^...] match newline. */
5053 syntax &= ~RE_DOT_NEWLINE;
5054 syntax |= RE_HAT_LISTS_NOT_NEWLINE;
5055 /* It also changes the matching behavior. */
5056 preg->newline_anchor = 1;
5059 preg->newline_anchor = 0;
5061 preg->no_sub = !!(cflags & REG_NOSUB);
5063 /* POSIX says a null character in the pattern terminates it, so we
5064 can use strlen here in compiling the pattern. */
5065 ret = regex_compile (pattern, strlen (pattern), syntax, preg);
5067 /* POSIX doesn't distinguish between an unmatched open-group and an
5068 unmatched close-group: both are REG_EPAREN. */
5069 if (ret == REG_ERPAREN) ret = REG_EPAREN;
5075 /* regexec searches for a given pattern, specified by PREG, in the
5078 If NMATCH is zero or REG_NOSUB was set in the cflags argument to
5079 `regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at
5080 least NMATCH elements, and we set them to the offsets of the
5081 corresponding matched substrings.
5083 EFLAGS specifies `execution flags' which affect matching: if
5084 REG_NOTBOL is set, then ^ does not match at the beginning of the
5085 string; if REG_NOTEOL is set, then $ does not match at the end.
5087 We return 0 if we find a match and REG_NOMATCH if not. */
5090 regexec (preg, string, nmatch, pmatch, eflags)
5091 const regex_t *preg;
5094 regmatch_t pmatch[];
5098 struct re_registers regs;
5099 regex_t private_preg;
5100 int len = strlen (string);
5101 boolean want_reg_info = !preg->no_sub && nmatch > 0;
5103 private_preg = *preg;
5105 private_preg.not_bol = !!(eflags & REG_NOTBOL);
5106 private_preg.not_eol = !!(eflags & REG_NOTEOL);
5108 /* The user has told us exactly how many registers to return
5109 information about, via `nmatch'. We have to pass that on to the
5110 matching routines. */
5111 private_preg.regs_allocated = REGS_FIXED;
5115 regs.num_regs = nmatch;
5116 regs.start = TALLOC (nmatch, regoff_t);
5117 regs.end = TALLOC (nmatch, regoff_t);
5118 if (regs.start == NULL || regs.end == NULL)
5119 return (int) REG_NOMATCH;
5122 /* Perform the searching operation. */
5123 ret = re_search (&private_preg, string, len,
5124 /* start: */ 0, /* range: */ len,
5125 want_reg_info ? ®s : (struct re_registers *) 0);
5127 /* Copy the register information to the POSIX structure. */
5134 for (r = 0; r < nmatch; r++)
5136 pmatch[r].rm_so = regs.start[r];
5137 pmatch[r].rm_eo = regs.end[r];
5141 /* If we needed the temporary register info, free the space now. */
5146 /* We want zero return to mean success, unlike `re_search'. */
5147 return ret >= 0 ? (int) REG_NOERROR : (int) REG_NOMATCH;
5151 /* Returns a message corresponding to an error code, ERRCODE, returned
5152 from either regcomp or regexec. We don't use PREG here. */
5155 regerror (errcode, preg, errbuf, errbuf_size)
5157 const regex_t *preg;
5165 || errcode >= (sizeof (re_error_msg) / sizeof (re_error_msg[0])))
5166 /* Only error codes returned by the rest of the code should be passed
5167 to this routine. If we are given anything else, or if other regex
5168 code generates an invalid error code, then the program has a bug.
5169 Dump core so we can fix it. */
5172 msg = re_error_msg[errcode];
5174 /* POSIX doesn't require that we do anything in this case, but why
5179 msg_size = strlen (msg) + 1; /* Includes the null. */
5181 if (errbuf_size != 0)
5183 if (msg_size > errbuf_size)
5185 strncpy (errbuf, msg, errbuf_size - 1);
5186 errbuf[errbuf_size - 1] = 0;
5189 strcpy (errbuf, msg);
5196 /* Free dynamically allocated space used by PREG. */
5202 if (preg->buffer != NULL)
5203 free (preg->buffer);
5204 preg->buffer = NULL;
5206 preg->allocated = 0;
5209 if (preg->fastmap != NULL)
5210 free (preg->fastmap);
5211 preg->fastmap = NULL;
5212 preg->fastmap_accurate = 0;
5214 if (preg->translate != NULL)
5215 free (preg->translate);
5216 preg->translate = NULL;
5219 #endif /* not emacs */
5223 make-backup-files: t
5225 trim-versions-without-asking: nil