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, 1994, 1995 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)
33 /* We need this for `regex.h', and perhaps for the Emacs include files. */
34 #include <sys/types.h>
36 /* This is for other GNU distributions with internationalized messages. */
37 #if HAVE_LIBINTL_H || defined (_LIBC)
40 # define gettext(msgid) (msgid)
43 /* The `emacs' switch turns on certain matching commands
44 that make sense only in Emacs. */
53 /* If we are not linking with Emacs proper,
54 we can't use the relocating allocator
55 even if config.h says that we can. */
58 #if defined (STDC_HEADERS) || defined (_LIBC)
65 /* We used to test for `BSTRING' here, but only GCC and Emacs define
66 `BSTRING', as far as I know, and neither of them use this code. */
67 #ifndef INHIBIT_STRING_HEADER
68 #if HAVE_STRING_H || STDC_HEADERS || defined (_LIBC)
71 #define bcmp(s1, s2, n) memcmp ((s1), (s2), (n))
74 #define bcopy(s, d, n) memcpy ((d), (s), (n))
77 #define bzero(s, n) memset ((s), 0, (n))
84 /* Define the syntax stuff for \<, \>, etc. */
86 /* This must be nonzero for the wordchar and notwordchar pattern
87 commands in re_match_2. */
92 #ifdef SWITCH_ENUM_BUG
93 #define SWITCH_ENUM_CAST(x) ((int)(x))
95 #define SWITCH_ENUM_CAST(x) (x)
100 extern char *re_syntax_table;
102 #else /* not SYNTAX_TABLE */
104 /* How many characters in the character set. */
105 #define CHAR_SET_SIZE 256
107 static char re_syntax_table[CHAR_SET_SIZE];
118 bzero (re_syntax_table, sizeof re_syntax_table);
120 for (c = 'a'; c <= 'z'; c++)
121 re_syntax_table[c] = Sword;
123 for (c = 'A'; c <= 'Z'; c++)
124 re_syntax_table[c] = Sword;
126 for (c = '0'; c <= '9'; c++)
127 re_syntax_table[c] = Sword;
129 re_syntax_table['_'] = Sword;
134 #endif /* not SYNTAX_TABLE */
136 #define SYNTAX(c) re_syntax_table[c]
138 #endif /* not emacs */
140 /* Get the interface, including the syntax bits. */
143 /* isalpha etc. are used for the character classes. */
146 /* Jim Meyering writes:
148 "... Some ctype macros are valid only for character codes that
149 isascii says are ASCII (SGI's IRIX-4.0.5 is one such system --when
150 using /bin/cc or gcc but without giving an ansi option). So, all
151 ctype uses should be through macros like ISPRINT... If
152 STDC_HEADERS is defined, then autoconf has verified that the ctype
153 macros don't need to be guarded with references to isascii. ...
154 Defining isascii to 1 should let any compiler worth its salt
155 eliminate the && through constant folding." */
157 #if defined (STDC_HEADERS) || (!defined (isascii) && !defined (HAVE_ISASCII))
160 #define ISASCII(c) isascii(c)
164 #define ISBLANK(c) (ISASCII (c) && isblank (c))
166 #define ISBLANK(c) ((c) == ' ' || (c) == '\t')
169 #define ISGRAPH(c) (ISASCII (c) && isgraph (c))
171 #define ISGRAPH(c) (ISASCII (c) && isprint (c) && !isspace (c))
174 #define ISPRINT(c) (ISASCII (c) && isprint (c))
175 #define ISDIGIT(c) (ISASCII (c) && isdigit (c))
176 #define ISALNUM(c) (ISASCII (c) && isalnum (c))
177 #define ISALPHA(c) (ISASCII (c) && isalpha (c))
178 #define ISCNTRL(c) (ISASCII (c) && iscntrl (c))
179 #define ISLOWER(c) (ISASCII (c) && islower (c))
180 #define ISPUNCT(c) (ISASCII (c) && ispunct (c))
181 #define ISSPACE(c) (ISASCII (c) && isspace (c))
182 #define ISUPPER(c) (ISASCII (c) && isupper (c))
183 #define ISXDIGIT(c) (ISASCII (c) && isxdigit (c))
186 #define NULL (void *)0
189 /* We remove any previous definition of `SIGN_EXTEND_CHAR',
190 since ours (we hope) works properly with all combinations of
191 machines, compilers, `char' and `unsigned char' argument types.
192 (Per Bothner suggested the basic approach.) */
193 #undef SIGN_EXTEND_CHAR
195 #define SIGN_EXTEND_CHAR(c) ((signed char) (c))
196 #else /* not __STDC__ */
197 /* As in Harbison and Steele. */
198 #define SIGN_EXTEND_CHAR(c) ((((unsigned char) (c)) ^ 128) - 128)
201 /* Should we use malloc or alloca? If REGEX_MALLOC is not defined, we
202 use `alloca' instead of `malloc'. This is because using malloc in
203 re_search* or re_match* could cause memory leaks when C-g is used in
204 Emacs; also, malloc is slower and causes storage fragmentation. On
205 the other hand, malloc is more portable, and easier to debug.
207 Because we sometimes use alloca, some routines have to be macros,
208 not functions -- `alloca'-allocated space disappears at the end of the
209 function it is called in. */
213 #define REGEX_ALLOCATE malloc
214 #define REGEX_REALLOCATE(source, osize, nsize) realloc (source, nsize)
215 #define REGEX_FREE free
217 #else /* not REGEX_MALLOC */
219 /* Emacs already defines alloca, sometimes. */
222 /* Make alloca work the best possible way. */
224 #define alloca __builtin_alloca
225 #else /* not __GNUC__ */
228 #else /* not __GNUC__ or HAVE_ALLOCA_H */
229 #ifndef _AIX /* Already did AIX, up at the top. */
230 #if defined (__STDC__) && __STDC__
235 #endif /* not _AIX */
236 #endif /* not HAVE_ALLOCA_H */
237 #endif /* not __GNUC__ */
239 #endif /* not alloca */
241 #define REGEX_ALLOCATE alloca
243 /* Assumes a `char *destination' variable. */
244 #define REGEX_REALLOCATE(source, osize, nsize) \
245 (destination = (char *) alloca (nsize), \
246 bcopy (source, destination, osize), \
249 /* No need to do anything to free, after alloca. */
250 #define REGEX_FREE(arg) ((void)0) /* Do nothing! But inhibit gcc warning. */
252 #endif /* not REGEX_MALLOC */
254 /* Define how to allocate the failure stack. */
257 #define REGEX_ALLOCATE_STACK(size) \
258 r_alloc (&failure_stack_ptr, (size))
259 #define REGEX_REALLOCATE_STACK(source, osize, nsize) \
260 r_re_alloc (&failure_stack_ptr, (nsize))
261 #define REGEX_FREE_STACK(ptr) \
262 r_alloc_free (&failure_stack_ptr)
264 #else /* not REL_ALLOC */
268 #define REGEX_ALLOCATE_STACK malloc
269 #define REGEX_REALLOCATE_STACK(source, osize, nsize) realloc (source, nsize)
270 #define REGEX_FREE_STACK free
272 #else /* not REGEX_MALLOC */
274 #define REGEX_ALLOCATE_STACK alloca
276 #define REGEX_REALLOCATE_STACK(source, osize, nsize) \
277 REGEX_REALLOCATE (source, osize, nsize)
278 /* No need to explicitly free anything. */
279 #define REGEX_FREE_STACK(arg)
281 #endif /* not REGEX_MALLOC */
282 #endif /* not REL_ALLOC */
285 /* True if `size1' is non-NULL and PTR is pointing anywhere inside
286 `string1' or just past its end. This works if PTR is NULL, which is
288 #define FIRST_STRING_P(ptr) \
289 (size1 && string1 <= (ptr) && (ptr) <= string1 + size1)
291 /* (Re)Allocate N items of type T using malloc, or fail. */
292 #define TALLOC(n, t) ((t *) malloc ((n) * sizeof (t)))
293 #define RETALLOC(addr, n, t) ((addr) = (t *) realloc (addr, (n) * sizeof (t)))
294 #define RETALLOC_IF(addr, n, t) \
295 if (addr) RETALLOC((addr), (n), t); else (addr) = TALLOC ((n), t)
296 #define REGEX_TALLOC(n, t) ((t *) REGEX_ALLOCATE ((n) * sizeof (t)))
298 #define BYTEWIDTH 8 /* In bits. */
300 #define STREQ(s1, s2) ((strcmp (s1, s2) == 0))
304 #define MAX(a, b) ((a) > (b) ? (a) : (b))
305 #define MIN(a, b) ((a) < (b) ? (a) : (b))
307 typedef char boolean;
311 static int re_match_2_internal ();
313 /* These are the command codes that appear in compiled regular
314 expressions. Some opcodes are followed by argument bytes. A
315 command code can specify any interpretation whatsoever for its
316 arguments. Zero bytes may appear in the compiled regular expression. */
322 /* Succeed right away--no more backtracking. */
325 /* Followed by one byte giving n, then by n literal bytes. */
328 /* Matches any (more or less) character. */
331 /* Matches any one char belonging to specified set. First
332 following byte is number of bitmap bytes. Then come bytes
333 for a bitmap saying which chars are in. Bits in each byte
334 are ordered low-bit-first. A character is in the set if its
335 bit is 1. A character too large to have a bit in the map is
336 automatically not in the set. */
339 /* Same parameters as charset, but match any character that is
340 not one of those specified. */
343 /* Start remembering the text that is matched, for storing in a
344 register. Followed by one byte with the register number, in
345 the range 0 to one less than the pattern buffer's re_nsub
346 field. Then followed by one byte with the number of groups
347 inner to this one. (This last has to be part of the
348 start_memory only because we need it in the on_failure_jump
352 /* Stop remembering the text that is matched and store it in a
353 memory register. Followed by one byte with the register
354 number, in the range 0 to one less than `re_nsub' in the
355 pattern buffer, and one byte with the number of inner groups,
356 just like `start_memory'. (We need the number of inner
357 groups here because we don't have any easy way of finding the
358 corresponding start_memory when we're at a stop_memory.) */
361 /* Match a duplicate of something remembered. Followed by one
362 byte containing the register number. */
365 /* Fail unless at beginning of line. */
368 /* Fail unless at end of line. */
371 /* Succeeds if at beginning of buffer (if emacs) or at beginning
372 of string to be matched (if not). */
375 /* Analogously, for end of buffer/string. */
378 /* Followed by two byte relative address to which to jump. */
381 /* Same as jump, but marks the end of an alternative. */
384 /* Followed by two-byte relative address of place to resume at
385 in case of failure. */
388 /* Like on_failure_jump, but pushes a placeholder instead of the
389 current string position when executed. */
390 on_failure_keep_string_jump,
392 /* Throw away latest failure point and then jump to following
393 two-byte relative address. */
396 /* Change to pop_failure_jump if know won't have to backtrack to
397 match; otherwise change to jump. This is used to jump
398 back to the beginning of a repeat. If what follows this jump
399 clearly won't match what the repeat does, such that we can be
400 sure that there is no use backtracking out of repetitions
401 already matched, then we change it to a pop_failure_jump.
402 Followed by two-byte address. */
405 /* Jump to following two-byte address, and push a dummy failure
406 point. This failure point will be thrown away if an attempt
407 is made to use it for a failure. A `+' construct makes this
408 before the first repeat. Also used as an intermediary kind
409 of jump when compiling an alternative. */
412 /* Push a dummy failure point and continue. Used at the end of
416 /* Followed by two-byte relative address and two-byte number n.
417 After matching N times, jump to the address upon failure. */
420 /* Followed by two-byte relative address, and two-byte number n.
421 Jump to the address N times, then fail. */
424 /* Set the following two-byte relative address to the
425 subsequent two-byte number. The address *includes* the two
429 wordchar, /* Matches any word-constituent character. */
430 notwordchar, /* Matches any char that is not a word-constituent. */
432 wordbeg, /* Succeeds if at word beginning. */
433 wordend, /* Succeeds if at word end. */
435 wordbound, /* Succeeds if at a word boundary. */
436 notwordbound /* Succeeds if not at a word boundary. */
439 ,before_dot, /* Succeeds if before point. */
440 at_dot, /* Succeeds if at point. */
441 after_dot, /* Succeeds if after point. */
443 /* Matches any character whose syntax is specified. Followed by
444 a byte which contains a syntax code, e.g., Sword. */
447 /* Matches any character whose syntax is not that specified. */
452 /* Common operations on the compiled pattern. */
454 /* Store NUMBER in two contiguous bytes starting at DESTINATION. */
456 #define STORE_NUMBER(destination, number) \
458 (destination)[0] = (number) & 0377; \
459 (destination)[1] = (number) >> 8; \
462 /* Same as STORE_NUMBER, except increment DESTINATION to
463 the byte after where the number is stored. Therefore, DESTINATION
464 must be an lvalue. */
466 #define STORE_NUMBER_AND_INCR(destination, number) \
468 STORE_NUMBER (destination, number); \
469 (destination) += 2; \
472 /* Put into DESTINATION a number stored in two contiguous bytes starting
475 #define EXTRACT_NUMBER(destination, source) \
477 (destination) = *(source) & 0377; \
478 (destination) += SIGN_EXTEND_CHAR (*((source) + 1)) << 8; \
483 extract_number (dest, source)
485 unsigned char *source;
487 int temp = SIGN_EXTEND_CHAR (*(source + 1));
488 *dest = *source & 0377;
492 #ifndef EXTRACT_MACROS /* To debug the macros. */
493 #undef EXTRACT_NUMBER
494 #define EXTRACT_NUMBER(dest, src) extract_number (&dest, src)
495 #endif /* not EXTRACT_MACROS */
499 /* Same as EXTRACT_NUMBER, except increment SOURCE to after the number.
500 SOURCE must be an lvalue. */
502 #define EXTRACT_NUMBER_AND_INCR(destination, source) \
504 EXTRACT_NUMBER (destination, source); \
510 extract_number_and_incr (destination, source)
512 unsigned char **source;
514 extract_number (destination, *source);
518 #ifndef EXTRACT_MACROS
519 #undef EXTRACT_NUMBER_AND_INCR
520 #define EXTRACT_NUMBER_AND_INCR(dest, src) \
521 extract_number_and_incr (&dest, &src)
522 #endif /* not EXTRACT_MACROS */
526 /* If DEBUG is defined, Regex prints many voluminous messages about what
527 it is doing (if the variable `debug' is nonzero). If linked with the
528 main program in `iregex.c', you can enter patterns and strings
529 interactively. And if linked with the main program in `main.c' and
530 the other test files, you can run the already-written tests. */
534 /* We use standard I/O for debugging. */
537 /* It is useful to test things that ``must'' be true when debugging. */
540 static int debug = 0;
542 #define DEBUG_STATEMENT(e) e
543 #define DEBUG_PRINT1(x) if (debug) printf (x)
544 #define DEBUG_PRINT2(x1, x2) if (debug) printf (x1, x2)
545 #define DEBUG_PRINT3(x1, x2, x3) if (debug) printf (x1, x2, x3)
546 #define DEBUG_PRINT4(x1, x2, x3, x4) if (debug) printf (x1, x2, x3, x4)
547 #define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) \
548 if (debug) print_partial_compiled_pattern (s, e)
549 #define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) \
550 if (debug) print_double_string (w, s1, sz1, s2, sz2)
553 /* Print the fastmap in human-readable form. */
556 print_fastmap (fastmap)
559 unsigned was_a_range = 0;
562 while (i < (1 << BYTEWIDTH))
568 while (i < (1 << BYTEWIDTH) && fastmap[i])
584 /* Print a compiled pattern string in human-readable form, starting at
585 the START pointer into it and ending just before the pointer END. */
588 print_partial_compiled_pattern (start, end)
589 unsigned char *start;
593 unsigned char *p = start;
594 unsigned char *pend = end;
602 /* Loop over pattern commands. */
605 printf ("%d:\t", p - start);
607 switch ((re_opcode_t) *p++)
615 printf ("/exactn/%d", mcnt);
626 printf ("/start_memory/%d/%d", mcnt, *p++);
631 printf ("/stop_memory/%d/%d", mcnt, *p++);
635 printf ("/duplicate/%d", *p++);
645 register int c, last = -100;
646 register int in_range = 0;
648 printf ("/charset [%s",
649 (re_opcode_t) *(p - 1) == charset_not ? "^" : "");
651 assert (p + *p < pend);
653 for (c = 0; c < 256; c++)
655 && (p[1 + (c/8)] & (1 << (c % 8))))
657 /* Are we starting a range? */
658 if (last + 1 == c && ! in_range)
663 /* Have we broken a range? */
664 else if (last + 1 != c && in_range)
693 case on_failure_jump:
694 extract_number_and_incr (&mcnt, &p);
695 printf ("/on_failure_jump to %d", p + mcnt - start);
698 case on_failure_keep_string_jump:
699 extract_number_and_incr (&mcnt, &p);
700 printf ("/on_failure_keep_string_jump to %d", p + mcnt - start);
703 case dummy_failure_jump:
704 extract_number_and_incr (&mcnt, &p);
705 printf ("/dummy_failure_jump to %d", p + mcnt - start);
708 case push_dummy_failure:
709 printf ("/push_dummy_failure");
713 extract_number_and_incr (&mcnt, &p);
714 printf ("/maybe_pop_jump to %d", p + mcnt - start);
717 case pop_failure_jump:
718 extract_number_and_incr (&mcnt, &p);
719 printf ("/pop_failure_jump to %d", p + mcnt - start);
723 extract_number_and_incr (&mcnt, &p);
724 printf ("/jump_past_alt to %d", p + mcnt - start);
728 extract_number_and_incr (&mcnt, &p);
729 printf ("/jump to %d", p + mcnt - start);
733 extract_number_and_incr (&mcnt, &p);
734 extract_number_and_incr (&mcnt2, &p);
735 printf ("/succeed_n to %d, %d times", p + mcnt - start, mcnt2);
739 extract_number_and_incr (&mcnt, &p);
740 extract_number_and_incr (&mcnt2, &p);
741 printf ("/jump_n to %d, %d times", p + mcnt - start, mcnt2);
745 extract_number_and_incr (&mcnt, &p);
746 extract_number_and_incr (&mcnt2, &p);
747 printf ("/set_number_at location %d to %d", p + mcnt - start, mcnt2);
751 printf ("/wordbound");
755 printf ("/notwordbound");
767 printf ("/before_dot");
775 printf ("/after_dot");
779 printf ("/syntaxspec");
781 printf ("/%d", mcnt);
785 printf ("/notsyntaxspec");
787 printf ("/%d", mcnt);
792 printf ("/wordchar");
796 printf ("/notwordchar");
808 printf ("?%d", *(p-1));
814 printf ("%d:\tend of pattern.\n", p - start);
819 print_compiled_pattern (bufp)
820 struct re_pattern_buffer *bufp;
822 unsigned char *buffer = bufp->buffer;
824 print_partial_compiled_pattern (buffer, buffer + bufp->used);
825 printf ("%d bytes used/%d bytes allocated.\n", bufp->used, bufp->allocated);
827 if (bufp->fastmap_accurate && bufp->fastmap)
829 printf ("fastmap: ");
830 print_fastmap (bufp->fastmap);
833 printf ("re_nsub: %d\t", bufp->re_nsub);
834 printf ("regs_alloc: %d\t", bufp->regs_allocated);
835 printf ("can_be_null: %d\t", bufp->can_be_null);
836 printf ("newline_anchor: %d\n", bufp->newline_anchor);
837 printf ("no_sub: %d\t", bufp->no_sub);
838 printf ("not_bol: %d\t", bufp->not_bol);
839 printf ("not_eol: %d\t", bufp->not_eol);
840 printf ("syntax: %d\n", bufp->syntax);
841 /* Perhaps we should print the translate table? */
846 print_double_string (where, string1, size1, string2, size2)
859 if (FIRST_STRING_P (where))
861 for (this_char = where - string1; this_char < size1; this_char++)
862 putchar (string1[this_char]);
867 for (this_char = where - string2; this_char < size2; this_char++)
868 putchar (string2[this_char]);
872 #else /* not DEBUG */
877 #define DEBUG_STATEMENT(e)
878 #define DEBUG_PRINT1(x)
879 #define DEBUG_PRINT2(x1, x2)
880 #define DEBUG_PRINT3(x1, x2, x3)
881 #define DEBUG_PRINT4(x1, x2, x3, x4)
882 #define DEBUG_PRINT_COMPILED_PATTERN(p, s, e)
883 #define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2)
885 #endif /* not DEBUG */
887 /* Set by `re_set_syntax' to the current regexp syntax to recognize. Can
888 also be assigned to arbitrarily: each pattern buffer stores its own
889 syntax, so it can be changed between regex compilations. */
890 /* This has no initializer because initialized variables in Emacs
891 become read-only after dumping. */
892 reg_syntax_t re_syntax_options;
895 /* Specify the precise syntax of regexps for compilation. This provides
896 for compatibility for various utilities which historically have
897 different, incompatible syntaxes.
899 The argument SYNTAX is a bit mask comprised of the various bits
900 defined in regex.h. We return the old syntax. */
903 re_set_syntax (syntax)
906 reg_syntax_t ret = re_syntax_options;
908 re_syntax_options = syntax;
912 /* This table gives an error message for each of the error codes listed
913 in regex.h. Obviously the order here has to be same as there.
914 POSIX doesn't require that we do anything for REG_NOERROR,
915 but why not be nice? */
917 static const char *re_error_msgid[] =
918 { "Success", /* REG_NOERROR */
919 "No match", /* REG_NOMATCH */
920 "Invalid regular expression", /* REG_BADPAT */
921 "Invalid collation character", /* REG_ECOLLATE */
922 "Invalid character class name", /* REG_ECTYPE */
923 "Trailing backslash", /* REG_EESCAPE */
924 "Invalid back reference", /* REG_ESUBREG */
925 "Unmatched [ or [^", /* REG_EBRACK */
926 "Unmatched ( or \\(", /* REG_EPAREN */
927 "Unmatched \\{", /* REG_EBRACE */
928 "Invalid content of \\{\\}", /* REG_BADBR */
929 "Invalid range end", /* REG_ERANGE */
930 "Memory exhausted", /* REG_ESPACE */
931 "Invalid preceding regular expression", /* REG_BADRPT */
932 "Premature end of regular expression", /* REG_EEND */
933 "Regular expression too big", /* REG_ESIZE */
934 "Unmatched ) or \\)", /* REG_ERPAREN */
937 /* Avoiding alloca during matching, to placate r_alloc. */
939 /* Define MATCH_MAY_ALLOCATE unless we need to make sure that the
940 searching and matching functions should not call alloca. On some
941 systems, alloca is implemented in terms of malloc, and if we're
942 using the relocating allocator routines, then malloc could cause a
943 relocation, which might (if the strings being searched are in the
944 ralloc heap) shift the data out from underneath the regexp
947 Here's another reason to avoid allocation: Emacs
948 processes input from X in a signal handler; processing X input may
949 call malloc; if input arrives while a matching routine is calling
950 malloc, then we're scrod. But Emacs can't just block input while
951 calling matching routines; then we don't notice interrupts when
952 they come in. So, Emacs blocks input around all regexp calls
953 except the matching calls, which it leaves unprotected, in the
954 faith that they will not malloc. */
956 /* Normally, this is fine. */
957 #define MATCH_MAY_ALLOCATE
959 /* When using GNU C, we are not REALLY using the C alloca, no matter
960 what config.h may say. So don't take precautions for it. */
965 /* The match routines may not allocate if (1) they would do it with malloc
966 and (2) it's not safe for them to use malloc.
967 Note that if REL_ALLOC is defined, matching would not use malloc for the
968 failure stack, but we would still use it for the register vectors;
969 so REL_ALLOC should not affect this. */
970 #if (defined (C_ALLOCA) || defined (REGEX_MALLOC)) && defined (emacs)
971 #undef MATCH_MAY_ALLOCATE
975 /* Failure stack declarations and macros; both re_compile_fastmap and
976 re_match_2 use a failure stack. These have to be macros because of
977 REGEX_ALLOCATE_STACK. */
980 /* Number of failure points for which to initially allocate space
981 when matching. If this number is exceeded, we allocate more
982 space, so it is not a hard limit. */
983 #ifndef INIT_FAILURE_ALLOC
984 #define INIT_FAILURE_ALLOC 5
987 /* Roughly the maximum number of failure points on the stack. Would be
988 exactly that if always used MAX_FAILURE_SPACE each time we failed.
989 This is a variable only so users of regex can assign to it; we never
990 change it ourselves. */
991 #if defined (MATCH_MAY_ALLOCATE)
992 int re_max_failures = 200000;
994 int re_max_failures = 2000;
999 unsigned char *pointer;
1003 typedef union fail_stack_elt fail_stack_elt_t;
1007 fail_stack_elt_t *stack;
1009 unsigned avail; /* Offset of next open position. */
1012 #define FAIL_STACK_EMPTY() (fail_stack.avail == 0)
1013 #define FAIL_STACK_PTR_EMPTY() (fail_stack_ptr->avail == 0)
1014 #define FAIL_STACK_FULL() (fail_stack.avail == fail_stack.size)
1017 /* Define macros to initialize and free the failure stack.
1018 Do `return -2' if the alloc fails. */
1020 #ifdef MATCH_MAY_ALLOCATE
1021 #define INIT_FAIL_STACK() \
1023 fail_stack.stack = (fail_stack_elt_t *) \
1024 REGEX_ALLOCATE_STACK (INIT_FAILURE_ALLOC * sizeof (fail_stack_elt_t)); \
1026 if (fail_stack.stack == NULL) \
1029 fail_stack.size = INIT_FAILURE_ALLOC; \
1030 fail_stack.avail = 0; \
1033 #define RESET_FAIL_STACK() REGEX_FREE_STACK (fail_stack.stack)
1035 #define INIT_FAIL_STACK() \
1037 fail_stack.avail = 0; \
1040 #define RESET_FAIL_STACK()
1044 /* Double the size of FAIL_STACK, up to approximately `re_max_failures' items.
1046 Return 1 if succeeds, and 0 if either ran out of memory
1047 allocating space for it or it was already too large.
1049 REGEX_REALLOCATE_STACK requires `destination' be declared. */
1051 #define DOUBLE_FAIL_STACK(fail_stack) \
1052 ((fail_stack).size > re_max_failures * MAX_FAILURE_ITEMS \
1054 : ((fail_stack).stack = (fail_stack_elt_t *) \
1055 REGEX_REALLOCATE_STACK ((fail_stack).stack, \
1056 (fail_stack).size * sizeof (fail_stack_elt_t), \
1057 ((fail_stack).size << 1) * sizeof (fail_stack_elt_t)), \
1059 (fail_stack).stack == NULL \
1061 : ((fail_stack).size <<= 1, \
1065 /* Push pointer POINTER on FAIL_STACK.
1066 Return 1 if was able to do so and 0 if ran out of memory allocating
1068 #define PUSH_PATTERN_OP(POINTER, FAIL_STACK) \
1069 ((FAIL_STACK_FULL () \
1070 && !DOUBLE_FAIL_STACK (FAIL_STACK)) \
1072 : ((FAIL_STACK).stack[(FAIL_STACK).avail++].pointer = POINTER, \
1075 /* Push a pointer value onto the failure stack.
1076 Assumes the variable `fail_stack'. Probably should only
1077 be called from within `PUSH_FAILURE_POINT'. */
1078 #define PUSH_FAILURE_POINTER(item) \
1079 fail_stack.stack[fail_stack.avail++].pointer = (unsigned char *) (item)
1081 /* This pushes an integer-valued item onto the failure stack.
1082 Assumes the variable `fail_stack'. Probably should only
1083 be called from within `PUSH_FAILURE_POINT'. */
1084 #define PUSH_FAILURE_INT(item) \
1085 fail_stack.stack[fail_stack.avail++].integer = (item)
1087 /* Push a fail_stack_elt_t value onto the failure stack.
1088 Assumes the variable `fail_stack'. Probably should only
1089 be called from within `PUSH_FAILURE_POINT'. */
1090 #define PUSH_FAILURE_ELT(item) \
1091 fail_stack.stack[fail_stack.avail++] = (item)
1093 /* These three POP... operations complement the three PUSH... operations.
1094 All assume that `fail_stack' is nonempty. */
1095 #define POP_FAILURE_POINTER() fail_stack.stack[--fail_stack.avail].pointer
1096 #define POP_FAILURE_INT() fail_stack.stack[--fail_stack.avail].integer
1097 #define POP_FAILURE_ELT() fail_stack.stack[--fail_stack.avail]
1099 /* Used to omit pushing failure point id's when we're not debugging. */
1101 #define DEBUG_PUSH PUSH_FAILURE_INT
1102 #define DEBUG_POP(item_addr) *(item_addr) = POP_FAILURE_INT ()
1104 #define DEBUG_PUSH(item)
1105 #define DEBUG_POP(item_addr)
1109 /* Push the information about the state we will need
1110 if we ever fail back to it.
1112 Requires variables fail_stack, regstart, regend, reg_info, and
1113 num_regs be declared. DOUBLE_FAIL_STACK requires `destination' be
1116 Does `return FAILURE_CODE' if runs out of memory. */
1118 #define PUSH_FAILURE_POINT(pattern_place, string_place, failure_code) \
1120 char *destination; \
1121 /* Must be int, so when we don't save any registers, the arithmetic \
1122 of 0 + -1 isn't done as unsigned. */ \
1125 DEBUG_STATEMENT (failure_id++); \
1126 DEBUG_STATEMENT (nfailure_points_pushed++); \
1127 DEBUG_PRINT2 ("\nPUSH_FAILURE_POINT #%u:\n", failure_id); \
1128 DEBUG_PRINT2 (" Before push, next avail: %d\n", (fail_stack).avail);\
1129 DEBUG_PRINT2 (" size: %d\n", (fail_stack).size);\
1131 DEBUG_PRINT2 (" slots needed: %d\n", NUM_FAILURE_ITEMS); \
1132 DEBUG_PRINT2 (" available: %d\n", REMAINING_AVAIL_SLOTS); \
1134 /* Ensure we have enough space allocated for what we will push. */ \
1135 while (REMAINING_AVAIL_SLOTS < NUM_FAILURE_ITEMS) \
1137 if (!DOUBLE_FAIL_STACK (fail_stack)) \
1138 return failure_code; \
1140 DEBUG_PRINT2 ("\n Doubled stack; size now: %d\n", \
1141 (fail_stack).size); \
1142 DEBUG_PRINT2 (" slots available: %d\n", REMAINING_AVAIL_SLOTS);\
1145 /* Push the info, starting with the registers. */ \
1146 DEBUG_PRINT1 ("\n"); \
1148 for (this_reg = lowest_active_reg; this_reg <= highest_active_reg; \
1151 DEBUG_PRINT2 (" Pushing reg: %d\n", this_reg); \
1152 DEBUG_STATEMENT (num_regs_pushed++); \
1154 DEBUG_PRINT2 (" start: 0x%x\n", regstart[this_reg]); \
1155 PUSH_FAILURE_POINTER (regstart[this_reg]); \
1157 DEBUG_PRINT2 (" end: 0x%x\n", regend[this_reg]); \
1158 PUSH_FAILURE_POINTER (regend[this_reg]); \
1160 DEBUG_PRINT2 (" info: 0x%x\n ", reg_info[this_reg]); \
1161 DEBUG_PRINT2 (" match_null=%d", \
1162 REG_MATCH_NULL_STRING_P (reg_info[this_reg])); \
1163 DEBUG_PRINT2 (" active=%d", IS_ACTIVE (reg_info[this_reg])); \
1164 DEBUG_PRINT2 (" matched_something=%d", \
1165 MATCHED_SOMETHING (reg_info[this_reg])); \
1166 DEBUG_PRINT2 (" ever_matched=%d", \
1167 EVER_MATCHED_SOMETHING (reg_info[this_reg])); \
1168 DEBUG_PRINT1 ("\n"); \
1169 PUSH_FAILURE_ELT (reg_info[this_reg].word); \
1172 DEBUG_PRINT2 (" Pushing low active reg: %d\n", lowest_active_reg);\
1173 PUSH_FAILURE_INT (lowest_active_reg); \
1175 DEBUG_PRINT2 (" Pushing high active reg: %d\n", highest_active_reg);\
1176 PUSH_FAILURE_INT (highest_active_reg); \
1178 DEBUG_PRINT2 (" Pushing pattern 0x%x: ", pattern_place); \
1179 DEBUG_PRINT_COMPILED_PATTERN (bufp, pattern_place, pend); \
1180 PUSH_FAILURE_POINTER (pattern_place); \
1182 DEBUG_PRINT2 (" Pushing string 0x%x: `", string_place); \
1183 DEBUG_PRINT_DOUBLE_STRING (string_place, string1, size1, string2, \
1185 DEBUG_PRINT1 ("'\n"); \
1186 PUSH_FAILURE_POINTER (string_place); \
1188 DEBUG_PRINT2 (" Pushing failure id: %u\n", failure_id); \
1189 DEBUG_PUSH (failure_id); \
1192 /* This is the number of items that are pushed and popped on the stack
1193 for each register. */
1194 #define NUM_REG_ITEMS 3
1196 /* Individual items aside from the registers. */
1198 #define NUM_NONREG_ITEMS 5 /* Includes failure point id. */
1200 #define NUM_NONREG_ITEMS 4
1203 /* We push at most this many items on the stack. */
1204 #define MAX_FAILURE_ITEMS ((num_regs - 1) * NUM_REG_ITEMS + NUM_NONREG_ITEMS)
1206 /* We actually push this many items. */
1207 #define NUM_FAILURE_ITEMS \
1208 ((highest_active_reg - lowest_active_reg + 1) * NUM_REG_ITEMS \
1211 /* How many items can still be added to the stack without overflowing it. */
1212 #define REMAINING_AVAIL_SLOTS ((fail_stack).size - (fail_stack).avail)
1215 /* Pops what PUSH_FAIL_STACK pushes.
1217 We restore into the parameters, all of which should be lvalues:
1218 STR -- the saved data position.
1219 PAT -- the saved pattern position.
1220 LOW_REG, HIGH_REG -- the highest and lowest active registers.
1221 REGSTART, REGEND -- arrays of string positions.
1222 REG_INFO -- array of information about each subexpression.
1224 Also assumes the variables `fail_stack' and (if debugging), `bufp',
1225 `pend', `string1', `size1', `string2', and `size2'. */
1227 #define POP_FAILURE_POINT(str, pat, low_reg, high_reg, regstart, regend, reg_info)\
1229 DEBUG_STATEMENT (fail_stack_elt_t failure_id;) \
1231 const unsigned char *string_temp; \
1233 assert (!FAIL_STACK_EMPTY ()); \
1235 /* Remove failure points and point to how many regs pushed. */ \
1236 DEBUG_PRINT1 ("POP_FAILURE_POINT:\n"); \
1237 DEBUG_PRINT2 (" Before pop, next avail: %d\n", fail_stack.avail); \
1238 DEBUG_PRINT2 (" size: %d\n", fail_stack.size); \
1240 assert (fail_stack.avail >= NUM_NONREG_ITEMS); \
1242 DEBUG_POP (&failure_id); \
1243 DEBUG_PRINT2 (" Popping failure id: %u\n", failure_id); \
1245 /* If the saved string location is NULL, it came from an \
1246 on_failure_keep_string_jump opcode, and we want to throw away the \
1247 saved NULL, thus retaining our current position in the string. */ \
1248 string_temp = POP_FAILURE_POINTER (); \
1249 if (string_temp != NULL) \
1250 str = (const char *) string_temp; \
1252 DEBUG_PRINT2 (" Popping string 0x%x: `", str); \
1253 DEBUG_PRINT_DOUBLE_STRING (str, string1, size1, string2, size2); \
1254 DEBUG_PRINT1 ("'\n"); \
1256 pat = (unsigned char *) POP_FAILURE_POINTER (); \
1257 DEBUG_PRINT2 (" Popping pattern 0x%x: ", pat); \
1258 DEBUG_PRINT_COMPILED_PATTERN (bufp, pat, pend); \
1260 /* Restore register info. */ \
1261 high_reg = (unsigned) POP_FAILURE_INT (); \
1262 DEBUG_PRINT2 (" Popping high active reg: %d\n", high_reg); \
1264 low_reg = (unsigned) POP_FAILURE_INT (); \
1265 DEBUG_PRINT2 (" Popping low active reg: %d\n", low_reg); \
1267 for (this_reg = high_reg; this_reg >= low_reg; this_reg--) \
1269 DEBUG_PRINT2 (" Popping reg: %d\n", this_reg); \
1271 reg_info[this_reg].word = POP_FAILURE_ELT (); \
1272 DEBUG_PRINT2 (" info: 0x%x\n", reg_info[this_reg]); \
1274 regend[this_reg] = (const char *) POP_FAILURE_POINTER (); \
1275 DEBUG_PRINT2 (" end: 0x%x\n", regend[this_reg]); \
1277 regstart[this_reg] = (const char *) POP_FAILURE_POINTER (); \
1278 DEBUG_PRINT2 (" start: 0x%x\n", regstart[this_reg]); \
1281 set_regs_matched_done = 0; \
1282 DEBUG_STATEMENT (nfailure_points_popped++); \
1283 } /* POP_FAILURE_POINT */
1287 /* Structure for per-register (a.k.a. per-group) information.
1288 Other register information, such as the
1289 starting and ending positions (which are addresses), and the list of
1290 inner groups (which is a bits list) are maintained in separate
1293 We are making a (strictly speaking) nonportable assumption here: that
1294 the compiler will pack our bit fields into something that fits into
1295 the type of `word', i.e., is something that fits into one item on the
1300 fail_stack_elt_t word;
1303 /* This field is one if this group can match the empty string,
1304 zero if not. If not yet determined, `MATCH_NULL_UNSET_VALUE'. */
1305 #define MATCH_NULL_UNSET_VALUE 3
1306 unsigned match_null_string_p : 2;
1307 unsigned is_active : 1;
1308 unsigned matched_something : 1;
1309 unsigned ever_matched_something : 1;
1311 } register_info_type;
1313 #define REG_MATCH_NULL_STRING_P(R) ((R).bits.match_null_string_p)
1314 #define IS_ACTIVE(R) ((R).bits.is_active)
1315 #define MATCHED_SOMETHING(R) ((R).bits.matched_something)
1316 #define EVER_MATCHED_SOMETHING(R) ((R).bits.ever_matched_something)
1319 /* Call this when have matched a real character; it sets `matched' flags
1320 for the subexpressions which we are currently inside. Also records
1321 that those subexprs have matched. */
1322 #define SET_REGS_MATCHED() \
1325 if (!set_regs_matched_done) \
1328 set_regs_matched_done = 1; \
1329 for (r = lowest_active_reg; r <= highest_active_reg; r++) \
1331 MATCHED_SOMETHING (reg_info[r]) \
1332 = EVER_MATCHED_SOMETHING (reg_info[r]) \
1339 /* Registers are set to a sentinel when they haven't yet matched. */
1340 static char reg_unset_dummy;
1341 #define REG_UNSET_VALUE (®_unset_dummy)
1342 #define REG_UNSET(e) ((e) == REG_UNSET_VALUE)
1344 /* Subroutine declarations and macros for regex_compile. */
1346 static void store_op1 (), store_op2 ();
1347 static void insert_op1 (), insert_op2 ();
1348 static boolean at_begline_loc_p (), at_endline_loc_p ();
1349 static boolean group_in_compile_stack ();
1350 static reg_errcode_t compile_range ();
1352 /* Fetch the next character in the uncompiled pattern---translating it
1353 if necessary. Also cast from a signed character in the constant
1354 string passed to us by the user to an unsigned char that we can use
1355 as an array index (in, e.g., `translate'). */
1356 #define PATFETCH(c) \
1357 do {if (p == pend) return REG_EEND; \
1358 c = (unsigned char) *p++; \
1359 if (translate) c = translate[c]; \
1362 /* Fetch the next character in the uncompiled pattern, with no
1364 #define PATFETCH_RAW(c) \
1365 do {if (p == pend) return REG_EEND; \
1366 c = (unsigned char) *p++; \
1369 /* Go backwards one character in the pattern. */
1370 #define PATUNFETCH p--
1373 /* If `translate' is non-null, return translate[D], else just D. We
1374 cast the subscript to translate because some data is declared as
1375 `char *', to avoid warnings when a string constant is passed. But
1376 when we use a character as a subscript we must make it unsigned. */
1377 #define TRANSLATE(d) (translate ? translate[(unsigned char) (d)] : (d))
1380 /* Macros for outputting the compiled pattern into `buffer'. */
1382 /* If the buffer isn't allocated when it comes in, use this. */
1383 #define INIT_BUF_SIZE 32
1385 /* Make sure we have at least N more bytes of space in buffer. */
1386 #define GET_BUFFER_SPACE(n) \
1387 while (b - bufp->buffer + (n) > bufp->allocated) \
1390 /* Make sure we have one more byte of buffer space and then add C to it. */
1391 #define BUF_PUSH(c) \
1393 GET_BUFFER_SPACE (1); \
1394 *b++ = (unsigned char) (c); \
1398 /* Ensure we have two more bytes of buffer space and then append C1 and C2. */
1399 #define BUF_PUSH_2(c1, c2) \
1401 GET_BUFFER_SPACE (2); \
1402 *b++ = (unsigned char) (c1); \
1403 *b++ = (unsigned char) (c2); \
1407 /* As with BUF_PUSH_2, except for three bytes. */
1408 #define BUF_PUSH_3(c1, c2, c3) \
1410 GET_BUFFER_SPACE (3); \
1411 *b++ = (unsigned char) (c1); \
1412 *b++ = (unsigned char) (c2); \
1413 *b++ = (unsigned char) (c3); \
1417 /* Store a jump with opcode OP at LOC to location TO. We store a
1418 relative address offset by the three bytes the jump itself occupies. */
1419 #define STORE_JUMP(op, loc, to) \
1420 store_op1 (op, loc, (to) - (loc) - 3)
1422 /* Likewise, for a two-argument jump. */
1423 #define STORE_JUMP2(op, loc, to, arg) \
1424 store_op2 (op, loc, (to) - (loc) - 3, arg)
1426 /* Like `STORE_JUMP', but for inserting. Assume `b' is the buffer end. */
1427 #define INSERT_JUMP(op, loc, to) \
1428 insert_op1 (op, loc, (to) - (loc) - 3, b)
1430 /* Like `STORE_JUMP2', but for inserting. Assume `b' is the buffer end. */
1431 #define INSERT_JUMP2(op, loc, to, arg) \
1432 insert_op2 (op, loc, (to) - (loc) - 3, arg, b)
1435 /* This is not an arbitrary limit: the arguments which represent offsets
1436 into the pattern are two bytes long. So if 2^16 bytes turns out to
1437 be too small, many things would have to change. */
1438 #define MAX_BUF_SIZE (1L << 16)
1441 /* Extend the buffer by twice its current size via realloc and
1442 reset the pointers that pointed into the old block to point to the
1443 correct places in the new one. If extending the buffer results in it
1444 being larger than MAX_BUF_SIZE, then flag memory exhausted. */
1445 #define EXTEND_BUFFER() \
1447 unsigned char *old_buffer = bufp->buffer; \
1448 if (bufp->allocated == MAX_BUF_SIZE) \
1450 bufp->allocated <<= 1; \
1451 if (bufp->allocated > MAX_BUF_SIZE) \
1452 bufp->allocated = MAX_BUF_SIZE; \
1453 bufp->buffer = (unsigned char *) realloc (bufp->buffer, bufp->allocated);\
1454 if (bufp->buffer == NULL) \
1455 return REG_ESPACE; \
1456 /* If the buffer moved, move all the pointers into it. */ \
1457 if (old_buffer != bufp->buffer) \
1459 b = (b - old_buffer) + bufp->buffer; \
1460 begalt = (begalt - old_buffer) + bufp->buffer; \
1461 if (fixup_alt_jump) \
1462 fixup_alt_jump = (fixup_alt_jump - old_buffer) + bufp->buffer;\
1464 laststart = (laststart - old_buffer) + bufp->buffer; \
1465 if (pending_exact) \
1466 pending_exact = (pending_exact - old_buffer) + bufp->buffer; \
1471 /* Since we have one byte reserved for the register number argument to
1472 {start,stop}_memory, the maximum number of groups we can report
1473 things about is what fits in that byte. */
1474 #define MAX_REGNUM 255
1476 /* But patterns can have more than `MAX_REGNUM' registers. We just
1477 ignore the excess. */
1478 typedef unsigned regnum_t;
1481 /* Macros for the compile stack. */
1483 /* Since offsets can go either forwards or backwards, this type needs to
1484 be able to hold values from -(MAX_BUF_SIZE - 1) to MAX_BUF_SIZE - 1. */
1485 typedef int pattern_offset_t;
1489 pattern_offset_t begalt_offset;
1490 pattern_offset_t fixup_alt_jump;
1491 pattern_offset_t inner_group_offset;
1492 pattern_offset_t laststart_offset;
1494 } compile_stack_elt_t;
1499 compile_stack_elt_t *stack;
1501 unsigned avail; /* Offset of next open position. */
1502 } compile_stack_type;
1505 #define INIT_COMPILE_STACK_SIZE 32
1507 #define COMPILE_STACK_EMPTY (compile_stack.avail == 0)
1508 #define COMPILE_STACK_FULL (compile_stack.avail == compile_stack.size)
1510 /* The next available element. */
1511 #define COMPILE_STACK_TOP (compile_stack.stack[compile_stack.avail])
1514 /* Set the bit for character C in a list. */
1515 #define SET_LIST_BIT(c) \
1516 (b[((unsigned char) (c)) / BYTEWIDTH] \
1517 |= 1 << (((unsigned char) c) % BYTEWIDTH))
1520 /* Get the next unsigned number in the uncompiled pattern. */
1521 #define GET_UNSIGNED_NUMBER(num) \
1525 while (ISDIGIT (c)) \
1529 num = num * 10 + c - '0'; \
1537 #define CHAR_CLASS_MAX_LENGTH 6 /* Namely, `xdigit'. */
1539 #define IS_CHAR_CLASS(string) \
1540 (STREQ (string, "alpha") || STREQ (string, "upper") \
1541 || STREQ (string, "lower") || STREQ (string, "digit") \
1542 || STREQ (string, "alnum") || STREQ (string, "xdigit") \
1543 || STREQ (string, "space") || STREQ (string, "print") \
1544 || STREQ (string, "punct") || STREQ (string, "graph") \
1545 || STREQ (string, "cntrl") || STREQ (string, "blank"))
1547 #ifndef MATCH_MAY_ALLOCATE
1549 /* If we cannot allocate large objects within re_match_2_internal,
1550 we make the fail stack and register vectors global.
1551 The fail stack, we grow to the maximum size when a regexp
1553 The register vectors, we adjust in size each time we
1554 compile a regexp, according to the number of registers it needs. */
1556 static fail_stack_type fail_stack;
1558 /* Size with which the following vectors are currently allocated.
1559 That is so we can make them bigger as needed,
1560 but never make them smaller. */
1561 static int regs_allocated_size;
1563 static const char ** regstart, ** regend;
1564 static const char ** old_regstart, ** old_regend;
1565 static const char **best_regstart, **best_regend;
1566 static register_info_type *reg_info;
1567 static const char **reg_dummy;
1568 static register_info_type *reg_info_dummy;
1570 /* Make the register vectors big enough for NUM_REGS registers,
1571 but don't make them smaller. */
1574 regex_grow_registers (num_regs)
1577 if (num_regs > regs_allocated_size)
1579 RETALLOC_IF (regstart, num_regs, const char *);
1580 RETALLOC_IF (regend, num_regs, const char *);
1581 RETALLOC_IF (old_regstart, num_regs, const char *);
1582 RETALLOC_IF (old_regend, num_regs, const char *);
1583 RETALLOC_IF (best_regstart, num_regs, const char *);
1584 RETALLOC_IF (best_regend, num_regs, const char *);
1585 RETALLOC_IF (reg_info, num_regs, register_info_type);
1586 RETALLOC_IF (reg_dummy, num_regs, const char *);
1587 RETALLOC_IF (reg_info_dummy, num_regs, register_info_type);
1589 regs_allocated_size = num_regs;
1593 #endif /* not MATCH_MAY_ALLOCATE */
1595 /* `regex_compile' compiles PATTERN (of length SIZE) according to SYNTAX.
1596 Returns one of error codes defined in `regex.h', or zero for success.
1598 Assumes the `allocated' (and perhaps `buffer') and `translate'
1599 fields are set in BUFP on entry.
1601 If it succeeds, results are put in BUFP (if it returns an error, the
1602 contents of BUFP are undefined):
1603 `buffer' is the compiled pattern;
1604 `syntax' is set to SYNTAX;
1605 `used' is set to the length of the compiled pattern;
1606 `fastmap_accurate' is zero;
1607 `re_nsub' is the number of subexpressions in PATTERN;
1608 `not_bol' and `not_eol' are zero;
1610 The `fastmap' and `newline_anchor' fields are neither
1611 examined nor set. */
1613 /* Return, freeing storage we allocated. */
1614 #define FREE_STACK_RETURN(value) \
1615 return (free (compile_stack.stack), value)
1617 static reg_errcode_t
1618 regex_compile (pattern, size, syntax, bufp)
1619 const char *pattern;
1621 reg_syntax_t syntax;
1622 struct re_pattern_buffer *bufp;
1624 /* We fetch characters from PATTERN here. Even though PATTERN is
1625 `char *' (i.e., signed), we declare these variables as unsigned, so
1626 they can be reliably used as array indices. */
1627 register unsigned char c, c1;
1629 /* A random temporary spot in PATTERN. */
1632 /* Points to the end of the buffer, where we should append. */
1633 register unsigned char *b;
1635 /* Keeps track of unclosed groups. */
1636 compile_stack_type compile_stack;
1638 /* Points to the current (ending) position in the pattern. */
1639 const char *p = pattern;
1640 const char *pend = pattern + size;
1642 /* How to translate the characters in the pattern. */
1643 char *translate = bufp->translate;
1645 /* Address of the count-byte of the most recently inserted `exactn'
1646 command. This makes it possible to tell if a new exact-match
1647 character can be added to that command or if the character requires
1648 a new `exactn' command. */
1649 unsigned char *pending_exact = 0;
1651 /* Address of start of the most recently finished expression.
1652 This tells, e.g., postfix * where to find the start of its
1653 operand. Reset at the beginning of groups and alternatives. */
1654 unsigned char *laststart = 0;
1656 /* Address of beginning of regexp, or inside of last group. */
1657 unsigned char *begalt;
1659 /* Place in the uncompiled pattern (i.e., the {) to
1660 which to go back if the interval is invalid. */
1661 const char *beg_interval;
1663 /* Address of the place where a forward jump should go to the end of
1664 the containing expression. Each alternative of an `or' -- except the
1665 last -- ends with a forward jump of this sort. */
1666 unsigned char *fixup_alt_jump = 0;
1668 /* Counts open-groups as they are encountered. Remembered for the
1669 matching close-group on the compile stack, so the same register
1670 number is put in the stop_memory as the start_memory. */
1671 regnum_t regnum = 0;
1674 DEBUG_PRINT1 ("\nCompiling pattern: ");
1677 unsigned debug_count;
1679 for (debug_count = 0; debug_count < size; debug_count++)
1680 putchar (pattern[debug_count]);
1685 /* Initialize the compile stack. */
1686 compile_stack.stack = TALLOC (INIT_COMPILE_STACK_SIZE, compile_stack_elt_t);
1687 if (compile_stack.stack == NULL)
1690 compile_stack.size = INIT_COMPILE_STACK_SIZE;
1691 compile_stack.avail = 0;
1693 /* Initialize the pattern buffer. */
1694 bufp->syntax = syntax;
1695 bufp->fastmap_accurate = 0;
1696 bufp->not_bol = bufp->not_eol = 0;
1698 /* Set `used' to zero, so that if we return an error, the pattern
1699 printer (for debugging) will think there's no pattern. We reset it
1703 /* Always count groups, whether or not bufp->no_sub is set. */
1706 #if !defined (emacs) && !defined (SYNTAX_TABLE)
1707 /* Initialize the syntax table. */
1708 init_syntax_once ();
1711 if (bufp->allocated == 0)
1714 { /* If zero allocated, but buffer is non-null, try to realloc
1715 enough space. This loses if buffer's address is bogus, but
1716 that is the user's responsibility. */
1717 RETALLOC (bufp->buffer, INIT_BUF_SIZE, unsigned char);
1720 { /* Caller did not allocate a buffer. Do it for them. */
1721 bufp->buffer = TALLOC (INIT_BUF_SIZE, unsigned char);
1723 if (!bufp->buffer) FREE_STACK_RETURN (REG_ESPACE);
1725 bufp->allocated = INIT_BUF_SIZE;
1728 begalt = b = bufp->buffer;
1730 /* Loop through the uncompiled pattern until we're at the end. */
1739 if ( /* If at start of pattern, it's an operator. */
1741 /* If context independent, it's an operator. */
1742 || syntax & RE_CONTEXT_INDEP_ANCHORS
1743 /* Otherwise, depends on what's come before. */
1744 || at_begline_loc_p (pattern, p, syntax))
1754 if ( /* If at end of pattern, it's an operator. */
1756 /* If context independent, it's an operator. */
1757 || syntax & RE_CONTEXT_INDEP_ANCHORS
1758 /* Otherwise, depends on what's next. */
1759 || at_endline_loc_p (p, pend, syntax))
1769 if ((syntax & RE_BK_PLUS_QM)
1770 || (syntax & RE_LIMITED_OPS))
1774 /* If there is no previous pattern... */
1777 if (syntax & RE_CONTEXT_INVALID_OPS)
1778 FREE_STACK_RETURN (REG_BADRPT);
1779 else if (!(syntax & RE_CONTEXT_INDEP_OPS))
1784 /* Are we optimizing this jump? */
1785 boolean keep_string_p = false;
1787 /* 1 means zero (many) matches is allowed. */
1788 char zero_times_ok = 0, many_times_ok = 0;
1790 /* If there is a sequence of repetition chars, collapse it
1791 down to just one (the right one). We can't combine
1792 interval operators with these because of, e.g., `a{2}*',
1793 which should only match an even number of `a's. */
1797 zero_times_ok |= c != '+';
1798 many_times_ok |= c != '?';
1806 || (!(syntax & RE_BK_PLUS_QM) && (c == '+' || c == '?')))
1809 else if (syntax & RE_BK_PLUS_QM && c == '\\')
1811 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
1814 if (!(c1 == '+' || c1 == '?'))
1829 /* If we get here, we found another repeat character. */
1832 /* Star, etc. applied to an empty pattern is equivalent
1833 to an empty pattern. */
1837 /* Now we know whether or not zero matches is allowed
1838 and also whether or not two or more matches is allowed. */
1840 { /* More than one repetition is allowed, so put in at the
1841 end a backward relative jump from `b' to before the next
1842 jump we're going to put in below (which jumps from
1843 laststart to after this jump).
1845 But if we are at the `*' in the exact sequence `.*\n',
1846 insert an unconditional jump backwards to the .,
1847 instead of the beginning of the loop. This way we only
1848 push a failure point once, instead of every time
1849 through the loop. */
1850 assert (p - 1 > pattern);
1852 /* Allocate the space for the jump. */
1853 GET_BUFFER_SPACE (3);
1855 /* We know we are not at the first character of the pattern,
1856 because laststart was nonzero. And we've already
1857 incremented `p', by the way, to be the character after
1858 the `*'. Do we have to do something analogous here
1859 for null bytes, because of RE_DOT_NOT_NULL? */
1860 if (TRANSLATE (*(p - 2)) == TRANSLATE ('.')
1862 && p < pend && TRANSLATE (*p) == TRANSLATE ('\n')
1863 && !(syntax & RE_DOT_NEWLINE))
1864 { /* We have .*\n. */
1865 STORE_JUMP (jump, b, laststart);
1866 keep_string_p = true;
1869 /* Anything else. */
1870 STORE_JUMP (maybe_pop_jump, b, laststart - 3);
1872 /* We've added more stuff to the buffer. */
1876 /* On failure, jump from laststart to b + 3, which will be the
1877 end of the buffer after this jump is inserted. */
1878 GET_BUFFER_SPACE (3);
1879 INSERT_JUMP (keep_string_p ? on_failure_keep_string_jump
1887 /* At least one repetition is required, so insert a
1888 `dummy_failure_jump' before the initial
1889 `on_failure_jump' instruction of the loop. This
1890 effects a skip over that instruction the first time
1891 we hit that loop. */
1892 GET_BUFFER_SPACE (3);
1893 INSERT_JUMP (dummy_failure_jump, laststart, laststart + 6);
1908 boolean had_char_class = false;
1910 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
1912 /* Ensure that we have enough space to push a charset: the
1913 opcode, the length count, and the bitset; 34 bytes in all. */
1914 GET_BUFFER_SPACE (34);
1918 /* We test `*p == '^' twice, instead of using an if
1919 statement, so we only need one BUF_PUSH. */
1920 BUF_PUSH (*p == '^' ? charset_not : charset);
1924 /* Remember the first position in the bracket expression. */
1927 /* Push the number of bytes in the bitmap. */
1928 BUF_PUSH ((1 << BYTEWIDTH) / BYTEWIDTH);
1930 /* Clear the whole map. */
1931 bzero (b, (1 << BYTEWIDTH) / BYTEWIDTH);
1933 /* charset_not matches newline according to a syntax bit. */
1934 if ((re_opcode_t) b[-2] == charset_not
1935 && (syntax & RE_HAT_LISTS_NOT_NEWLINE))
1936 SET_LIST_BIT ('\n');
1938 /* Read in characters and ranges, setting map bits. */
1941 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
1945 /* \ might escape characters inside [...] and [^...]. */
1946 if ((syntax & RE_BACKSLASH_ESCAPE_IN_LISTS) && c == '\\')
1948 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
1955 /* Could be the end of the bracket expression. If it's
1956 not (i.e., when the bracket expression is `[]' so
1957 far), the ']' character bit gets set way below. */
1958 if (c == ']' && p != p1 + 1)
1961 /* Look ahead to see if it's a range when the last thing
1962 was a character class. */
1963 if (had_char_class && c == '-' && *p != ']')
1964 FREE_STACK_RETURN (REG_ERANGE);
1966 /* Look ahead to see if it's a range when the last thing
1967 was a character: if this is a hyphen not at the
1968 beginning or the end of a list, then it's the range
1971 && !(p - 2 >= pattern && p[-2] == '[')
1972 && !(p - 3 >= pattern && p[-3] == '[' && p[-2] == '^')
1976 = compile_range (&p, pend, translate, syntax, b);
1977 if (ret != REG_NOERROR) FREE_STACK_RETURN (ret);
1980 else if (p[0] == '-' && p[1] != ']')
1981 { /* This handles ranges made up of characters only. */
1984 /* Move past the `-'. */
1987 ret = compile_range (&p, pend, translate, syntax, b);
1988 if (ret != REG_NOERROR) FREE_STACK_RETURN (ret);
1991 /* See if we're at the beginning of a possible character
1994 else if (syntax & RE_CHAR_CLASSES && c == '[' && *p == ':')
1995 { /* Leave room for the null. */
1996 char str[CHAR_CLASS_MAX_LENGTH + 1];
2001 /* If pattern is `[[:'. */
2002 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2007 if (c == ':' || c == ']' || p == pend
2008 || c1 == CHAR_CLASS_MAX_LENGTH)
2014 /* If isn't a word bracketed by `[:' and:`]':
2015 undo the ending character, the letters, and leave
2016 the leading `:' and `[' (but set bits for them). */
2017 if (c == ':' && *p == ']')
2020 boolean is_alnum = STREQ (str, "alnum");
2021 boolean is_alpha = STREQ (str, "alpha");
2022 boolean is_blank = STREQ (str, "blank");
2023 boolean is_cntrl = STREQ (str, "cntrl");
2024 boolean is_digit = STREQ (str, "digit");
2025 boolean is_graph = STREQ (str, "graph");
2026 boolean is_lower = STREQ (str, "lower");
2027 boolean is_print = STREQ (str, "print");
2028 boolean is_punct = STREQ (str, "punct");
2029 boolean is_space = STREQ (str, "space");
2030 boolean is_upper = STREQ (str, "upper");
2031 boolean is_xdigit = STREQ (str, "xdigit");
2033 if (!IS_CHAR_CLASS (str))
2034 FREE_STACK_RETURN (REG_ECTYPE);
2036 /* Throw away the ] at the end of the character
2040 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2042 for (ch = 0; ch < 1 << BYTEWIDTH; ch++)
2044 /* This was split into 3 if's to
2045 avoid an arbitrary limit in some compiler. */
2046 if ( (is_alnum && ISALNUM (ch))
2047 || (is_alpha && ISALPHA (ch))
2048 || (is_blank && ISBLANK (ch))
2049 || (is_cntrl && ISCNTRL (ch)))
2051 if ( (is_digit && ISDIGIT (ch))
2052 || (is_graph && ISGRAPH (ch))
2053 || (is_lower && ISLOWER (ch))
2054 || (is_print && ISPRINT (ch)))
2056 if ( (is_punct && ISPUNCT (ch))
2057 || (is_space && ISSPACE (ch))
2058 || (is_upper && ISUPPER (ch))
2059 || (is_xdigit && ISXDIGIT (ch)))
2062 had_char_class = true;
2071 had_char_class = false;
2076 had_char_class = false;
2081 /* Discard any (non)matching list bytes that are all 0 at the
2082 end of the map. Decrease the map-length byte too. */
2083 while ((int) b[-1] > 0 && b[b[-1] - 1] == 0)
2091 if (syntax & RE_NO_BK_PARENS)
2098 if (syntax & RE_NO_BK_PARENS)
2105 if (syntax & RE_NEWLINE_ALT)
2112 if (syntax & RE_NO_BK_VBAR)
2119 if (syntax & RE_INTERVALS && syntax & RE_NO_BK_BRACES)
2120 goto handle_interval;
2126 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
2128 /* Do not translate the character after the \, so that we can
2129 distinguish, e.g., \B from \b, even if we normally would
2130 translate, e.g., B to b. */
2136 if (syntax & RE_NO_BK_PARENS)
2137 goto normal_backslash;
2143 if (COMPILE_STACK_FULL)
2145 RETALLOC (compile_stack.stack, compile_stack.size << 1,
2146 compile_stack_elt_t);
2147 if (compile_stack.stack == NULL) return REG_ESPACE;
2149 compile_stack.size <<= 1;
2152 /* These are the values to restore when we hit end of this
2153 group. They are all relative offsets, so that if the
2154 whole pattern moves because of realloc, they will still
2156 COMPILE_STACK_TOP.begalt_offset = begalt - bufp->buffer;
2157 COMPILE_STACK_TOP.fixup_alt_jump
2158 = fixup_alt_jump ? fixup_alt_jump - bufp->buffer + 1 : 0;
2159 COMPILE_STACK_TOP.laststart_offset = b - bufp->buffer;
2160 COMPILE_STACK_TOP.regnum = regnum;
2162 /* We will eventually replace the 0 with the number of
2163 groups inner to this one. But do not push a
2164 start_memory for groups beyond the last one we can
2165 represent in the compiled pattern. */
2166 if (regnum <= MAX_REGNUM)
2168 COMPILE_STACK_TOP.inner_group_offset = b - bufp->buffer + 2;
2169 BUF_PUSH_3 (start_memory, regnum, 0);
2172 compile_stack.avail++;
2177 /* If we've reached MAX_REGNUM groups, then this open
2178 won't actually generate any code, so we'll have to
2179 clear pending_exact explicitly. */
2185 if (syntax & RE_NO_BK_PARENS) goto normal_backslash;
2187 if (COMPILE_STACK_EMPTY)
2188 if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD)
2189 goto normal_backslash;
2191 FREE_STACK_RETURN (REG_ERPAREN);
2195 { /* Push a dummy failure point at the end of the
2196 alternative for a possible future
2197 `pop_failure_jump' to pop. See comments at
2198 `push_dummy_failure' in `re_match_2'. */
2199 BUF_PUSH (push_dummy_failure);
2201 /* We allocated space for this jump when we assigned
2202 to `fixup_alt_jump', in the `handle_alt' case below. */
2203 STORE_JUMP (jump_past_alt, fixup_alt_jump, b - 1);
2206 /* See similar code for backslashed left paren above. */
2207 if (COMPILE_STACK_EMPTY)
2208 if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD)
2211 FREE_STACK_RETURN (REG_ERPAREN);
2213 /* Since we just checked for an empty stack above, this
2214 ``can't happen''. */
2215 assert (compile_stack.avail != 0);
2217 /* We don't just want to restore into `regnum', because
2218 later groups should continue to be numbered higher,
2219 as in `(ab)c(de)' -- the second group is #2. */
2220 regnum_t this_group_regnum;
2222 compile_stack.avail--;
2223 begalt = bufp->buffer + COMPILE_STACK_TOP.begalt_offset;
2225 = COMPILE_STACK_TOP.fixup_alt_jump
2226 ? bufp->buffer + COMPILE_STACK_TOP.fixup_alt_jump - 1
2228 laststart = bufp->buffer + COMPILE_STACK_TOP.laststart_offset;
2229 this_group_regnum = COMPILE_STACK_TOP.regnum;
2230 /* If we've reached MAX_REGNUM groups, then this open
2231 won't actually generate any code, so we'll have to
2232 clear pending_exact explicitly. */
2235 /* We're at the end of the group, so now we know how many
2236 groups were inside this one. */
2237 if (this_group_regnum <= MAX_REGNUM)
2239 unsigned char *inner_group_loc
2240 = bufp->buffer + COMPILE_STACK_TOP.inner_group_offset;
2242 *inner_group_loc = regnum - this_group_regnum;
2243 BUF_PUSH_3 (stop_memory, this_group_regnum,
2244 regnum - this_group_regnum);
2250 case '|': /* `\|'. */
2251 if (syntax & RE_LIMITED_OPS || syntax & RE_NO_BK_VBAR)
2252 goto normal_backslash;
2254 if (syntax & RE_LIMITED_OPS)
2257 /* Insert before the previous alternative a jump which
2258 jumps to this alternative if the former fails. */
2259 GET_BUFFER_SPACE (3);
2260 INSERT_JUMP (on_failure_jump, begalt, b + 6);
2264 /* The alternative before this one has a jump after it
2265 which gets executed if it gets matched. Adjust that
2266 jump so it will jump to this alternative's analogous
2267 jump (put in below, which in turn will jump to the next
2268 (if any) alternative's such jump, etc.). The last such
2269 jump jumps to the correct final destination. A picture:
2275 If we are at `b', then fixup_alt_jump right now points to a
2276 three-byte space after `a'. We'll put in the jump, set
2277 fixup_alt_jump to right after `b', and leave behind three
2278 bytes which we'll fill in when we get to after `c'. */
2281 STORE_JUMP (jump_past_alt, fixup_alt_jump, b);
2283 /* Mark and leave space for a jump after this alternative,
2284 to be filled in later either by next alternative or
2285 when know we're at the end of a series of alternatives. */
2287 GET_BUFFER_SPACE (3);
2296 /* If \{ is a literal. */
2297 if (!(syntax & RE_INTERVALS)
2298 /* If we're at `\{' and it's not the open-interval
2300 || ((syntax & RE_INTERVALS) && (syntax & RE_NO_BK_BRACES))
2301 || (p - 2 == pattern && p == pend))
2302 goto normal_backslash;
2306 /* If got here, then the syntax allows intervals. */
2308 /* At least (most) this many matches must be made. */
2309 int lower_bound = -1, upper_bound = -1;
2311 beg_interval = p - 1;
2315 if (syntax & RE_NO_BK_BRACES)
2316 goto unfetch_interval;
2318 FREE_STACK_RETURN (REG_EBRACE);
2321 GET_UNSIGNED_NUMBER (lower_bound);
2325 GET_UNSIGNED_NUMBER (upper_bound);
2326 if (upper_bound < 0) upper_bound = RE_DUP_MAX;
2329 /* Interval such as `{1}' => match exactly once. */
2330 upper_bound = lower_bound;
2332 if (lower_bound < 0 || upper_bound > RE_DUP_MAX
2333 || lower_bound > upper_bound)
2335 if (syntax & RE_NO_BK_BRACES)
2336 goto unfetch_interval;
2338 FREE_STACK_RETURN (REG_BADBR);
2341 if (!(syntax & RE_NO_BK_BRACES))
2343 if (c != '\\') FREE_STACK_RETURN (REG_EBRACE);
2350 if (syntax & RE_NO_BK_BRACES)
2351 goto unfetch_interval;
2353 FREE_STACK_RETURN (REG_BADBR);
2356 /* We just parsed a valid interval. */
2358 /* If it's invalid to have no preceding re. */
2361 if (syntax & RE_CONTEXT_INVALID_OPS)
2362 FREE_STACK_RETURN (REG_BADRPT);
2363 else if (syntax & RE_CONTEXT_INDEP_OPS)
2366 goto unfetch_interval;
2369 /* If the upper bound is zero, don't want to succeed at
2370 all; jump from `laststart' to `b + 3', which will be
2371 the end of the buffer after we insert the jump. */
2372 if (upper_bound == 0)
2374 GET_BUFFER_SPACE (3);
2375 INSERT_JUMP (jump, laststart, b + 3);
2379 /* Otherwise, we have a nontrivial interval. When
2380 we're all done, the pattern will look like:
2381 set_number_at <jump count> <upper bound>
2382 set_number_at <succeed_n count> <lower bound>
2383 succeed_n <after jump addr> <succeed_n count>
2385 jump_n <succeed_n addr> <jump count>
2386 (The upper bound and `jump_n' are omitted if
2387 `upper_bound' is 1, though.) */
2389 { /* If the upper bound is > 1, we need to insert
2390 more at the end of the loop. */
2391 unsigned nbytes = 10 + (upper_bound > 1) * 10;
2393 GET_BUFFER_SPACE (nbytes);
2395 /* Initialize lower bound of the `succeed_n', even
2396 though it will be set during matching by its
2397 attendant `set_number_at' (inserted next),
2398 because `re_compile_fastmap' needs to know.
2399 Jump to the `jump_n' we might insert below. */
2400 INSERT_JUMP2 (succeed_n, laststart,
2401 b + 5 + (upper_bound > 1) * 5,
2405 /* Code to initialize the lower bound. Insert
2406 before the `succeed_n'. The `5' is the last two
2407 bytes of this `set_number_at', plus 3 bytes of
2408 the following `succeed_n'. */
2409 insert_op2 (set_number_at, laststart, 5, lower_bound, b);
2412 if (upper_bound > 1)
2413 { /* More than one repetition is allowed, so
2414 append a backward jump to the `succeed_n'
2415 that starts this interval.
2417 When we've reached this during matching,
2418 we'll have matched the interval once, so
2419 jump back only `upper_bound - 1' times. */
2420 STORE_JUMP2 (jump_n, b, laststart + 5,
2424 /* The location we want to set is the second
2425 parameter of the `jump_n'; that is `b-2' as
2426 an absolute address. `laststart' will be
2427 the `set_number_at' we're about to insert;
2428 `laststart+3' the number to set, the source
2429 for the relative address. But we are
2430 inserting into the middle of the pattern --
2431 so everything is getting moved up by 5.
2432 Conclusion: (b - 2) - (laststart + 3) + 5,
2433 i.e., b - laststart.
2435 We insert this at the beginning of the loop
2436 so that if we fail during matching, we'll
2437 reinitialize the bounds. */
2438 insert_op2 (set_number_at, laststart, b - laststart,
2439 upper_bound - 1, b);
2444 beg_interval = NULL;
2449 /* If an invalid interval, match the characters as literals. */
2450 assert (beg_interval);
2452 beg_interval = NULL;
2454 /* normal_char and normal_backslash need `c'. */
2457 if (!(syntax & RE_NO_BK_BRACES))
2459 if (p > pattern && p[-1] == '\\')
2460 goto normal_backslash;
2465 /* There is no way to specify the before_dot and after_dot
2466 operators. rms says this is ok. --karl */
2474 BUF_PUSH_2 (syntaxspec, syntax_spec_code[c]);
2480 BUF_PUSH_2 (notsyntaxspec, syntax_spec_code[c]);
2487 BUF_PUSH (wordchar);
2493 BUF_PUSH (notwordchar);
2506 BUF_PUSH (wordbound);
2510 BUF_PUSH (notwordbound);
2521 case '1': case '2': case '3': case '4': case '5':
2522 case '6': case '7': case '8': case '9':
2523 if (syntax & RE_NO_BK_REFS)
2529 FREE_STACK_RETURN (REG_ESUBREG);
2531 /* Can't back reference to a subexpression if inside of it. */
2532 if (group_in_compile_stack (compile_stack, c1))
2536 BUF_PUSH_2 (duplicate, c1);
2542 if (syntax & RE_BK_PLUS_QM)
2545 goto normal_backslash;
2549 /* You might think it would be useful for \ to mean
2550 not to translate; but if we don't translate it
2551 it will never match anything. */
2559 /* Expects the character in `c'. */
2561 /* If no exactn currently being built. */
2564 /* If last exactn not at current position. */
2565 || pending_exact + *pending_exact + 1 != b
2567 /* We have only one byte following the exactn for the count. */
2568 || *pending_exact == (1 << BYTEWIDTH) - 1
2570 /* If followed by a repetition operator. */
2571 || *p == '*' || *p == '^'
2572 || ((syntax & RE_BK_PLUS_QM)
2573 ? *p == '\\' && (p[1] == '+' || p[1] == '?')
2574 : (*p == '+' || *p == '?'))
2575 || ((syntax & RE_INTERVALS)
2576 && ((syntax & RE_NO_BK_BRACES)
2578 : (p[0] == '\\' && p[1] == '{'))))
2580 /* Start building a new exactn. */
2584 BUF_PUSH_2 (exactn, 0);
2585 pending_exact = b - 1;
2592 } /* while p != pend */
2595 /* Through the pattern now. */
2598 STORE_JUMP (jump_past_alt, fixup_alt_jump, b);
2600 if (!COMPILE_STACK_EMPTY)
2601 FREE_STACK_RETURN (REG_EPAREN);
2603 /* If we don't want backtracking, force success
2604 the first time we reach the end of the compiled pattern. */
2605 if (syntax & RE_NO_POSIX_BACKTRACKING)
2608 free (compile_stack.stack);
2610 /* We have succeeded; set the length of the buffer. */
2611 bufp->used = b - bufp->buffer;
2616 DEBUG_PRINT1 ("\nCompiled pattern: \n");
2617 print_compiled_pattern (bufp);
2621 #ifndef MATCH_MAY_ALLOCATE
2622 /* Initialize the failure stack to the largest possible stack. This
2623 isn't necessary unless we're trying to avoid calling alloca in
2624 the search and match routines. */
2626 int num_regs = bufp->re_nsub + 1;
2628 /* Since DOUBLE_FAIL_STACK refuses to double only if the current size
2629 is strictly greater than re_max_failures, the largest possible stack
2630 is 2 * re_max_failures failure points. */
2631 if (fail_stack.size < (2 * re_max_failures * MAX_FAILURE_ITEMS))
2633 fail_stack.size = (2 * re_max_failures * MAX_FAILURE_ITEMS);
2636 if (! fail_stack.stack)
2638 = (fail_stack_elt_t *) xmalloc (fail_stack.size
2639 * sizeof (fail_stack_elt_t));
2642 = (fail_stack_elt_t *) xrealloc (fail_stack.stack,
2644 * sizeof (fail_stack_elt_t)));
2645 #else /* not emacs */
2646 if (! fail_stack.stack)
2648 = (fail_stack_elt_t *) malloc (fail_stack.size
2649 * sizeof (fail_stack_elt_t));
2652 = (fail_stack_elt_t *) realloc (fail_stack.stack,
2654 * sizeof (fail_stack_elt_t)));
2655 #endif /* not emacs */
2658 regex_grow_registers (num_regs);
2660 #endif /* not MATCH_MAY_ALLOCATE */
2663 } /* regex_compile */
2665 /* Subroutines for `regex_compile'. */
2667 /* Store OP at LOC followed by two-byte integer parameter ARG. */
2670 store_op1 (op, loc, arg)
2675 *loc = (unsigned char) op;
2676 STORE_NUMBER (loc + 1, arg);
2680 /* Like `store_op1', but for two two-byte parameters ARG1 and ARG2. */
2683 store_op2 (op, loc, arg1, arg2)
2688 *loc = (unsigned char) op;
2689 STORE_NUMBER (loc + 1, arg1);
2690 STORE_NUMBER (loc + 3, arg2);
2694 /* Copy the bytes from LOC to END to open up three bytes of space at LOC
2695 for OP followed by two-byte integer parameter ARG. */
2698 insert_op1 (op, loc, arg, end)
2704 register unsigned char *pfrom = end;
2705 register unsigned char *pto = end + 3;
2707 while (pfrom != loc)
2710 store_op1 (op, loc, arg);
2714 /* Like `insert_op1', but for two two-byte parameters ARG1 and ARG2. */
2717 insert_op2 (op, loc, arg1, arg2, end)
2723 register unsigned char *pfrom = end;
2724 register unsigned char *pto = end + 5;
2726 while (pfrom != loc)
2729 store_op2 (op, loc, arg1, arg2);
2733 /* P points to just after a ^ in PATTERN. Return true if that ^ comes
2734 after an alternative or a begin-subexpression. We assume there is at
2735 least one character before the ^. */
2738 at_begline_loc_p (pattern, p, syntax)
2739 const char *pattern, *p;
2740 reg_syntax_t syntax;
2742 const char *prev = p - 2;
2743 boolean prev_prev_backslash = prev > pattern && prev[-1] == '\\';
2746 /* After a subexpression? */
2747 (*prev == '(' && (syntax & RE_NO_BK_PARENS || prev_prev_backslash))
2748 /* After an alternative? */
2749 || (*prev == '|' && (syntax & RE_NO_BK_VBAR || prev_prev_backslash));
2753 /* The dual of at_begline_loc_p. This one is for $. We assume there is
2754 at least one character after the $, i.e., `P < PEND'. */
2757 at_endline_loc_p (p, pend, syntax)
2758 const char *p, *pend;
2761 const char *next = p;
2762 boolean next_backslash = *next == '\\';
2763 const char *next_next = p + 1 < pend ? p + 1 : 0;
2766 /* Before a subexpression? */
2767 (syntax & RE_NO_BK_PARENS ? *next == ')'
2768 : next_backslash && next_next && *next_next == ')')
2769 /* Before an alternative? */
2770 || (syntax & RE_NO_BK_VBAR ? *next == '|'
2771 : next_backslash && next_next && *next_next == '|');
2775 /* Returns true if REGNUM is in one of COMPILE_STACK's elements and
2776 false if it's not. */
2779 group_in_compile_stack (compile_stack, regnum)
2780 compile_stack_type compile_stack;
2785 for (this_element = compile_stack.avail - 1;
2788 if (compile_stack.stack[this_element].regnum == regnum)
2795 /* Read the ending character of a range (in a bracket expression) from the
2796 uncompiled pattern *P_PTR (which ends at PEND). We assume the
2797 starting character is in `P[-2]'. (`P[-1]' is the character `-'.)
2798 Then we set the translation of all bits between the starting and
2799 ending characters (inclusive) in the compiled pattern B.
2801 Return an error code.
2803 We use these short variable names so we can use the same macros as
2804 `regex_compile' itself. */
2806 static reg_errcode_t
2807 compile_range (p_ptr, pend, translate, syntax, b)
2808 const char **p_ptr, *pend;
2810 reg_syntax_t syntax;
2815 const char *p = *p_ptr;
2816 int range_start, range_end;
2821 /* Even though the pattern is a signed `char *', we need to fetch
2822 with unsigned char *'s; if the high bit of the pattern character
2823 is set, the range endpoints will be negative if we fetch using a
2826 We also want to fetch the endpoints without translating them; the
2827 appropriate translation is done in the bit-setting loop below. */
2828 /* The SVR4 compiler on the 3B2 had trouble with unsigned const char *. */
2829 range_start = ((const unsigned char *) p)[-2];
2830 range_end = ((const unsigned char *) p)[0];
2832 /* Have to increment the pointer into the pattern string, so the
2833 caller isn't still at the ending character. */
2836 /* If the start is after the end, the range is empty. */
2837 if (range_start > range_end)
2838 return syntax & RE_NO_EMPTY_RANGES ? REG_ERANGE : REG_NOERROR;
2840 /* Here we see why `this_char' has to be larger than an `unsigned
2841 char' -- the range is inclusive, so if `range_end' == 0xff
2842 (assuming 8-bit characters), we would otherwise go into an infinite
2843 loop, since all characters <= 0xff. */
2844 for (this_char = range_start; this_char <= range_end; this_char++)
2846 SET_LIST_BIT (TRANSLATE (this_char));
2852 /* re_compile_fastmap computes a ``fastmap'' for the compiled pattern in
2853 BUFP. A fastmap records which of the (1 << BYTEWIDTH) possible
2854 characters can start a string that matches the pattern. This fastmap
2855 is used by re_search to skip quickly over impossible starting points.
2857 The caller must supply the address of a (1 << BYTEWIDTH)-byte data
2858 area as BUFP->fastmap.
2860 We set the `fastmap', `fastmap_accurate', and `can_be_null' fields in
2863 Returns 0 if we succeed, -2 if an internal error. */
2866 re_compile_fastmap (bufp)
2867 struct re_pattern_buffer *bufp;
2870 #ifdef MATCH_MAY_ALLOCATE
2871 fail_stack_type fail_stack;
2873 #ifndef REGEX_MALLOC
2876 /* We don't push any register information onto the failure stack. */
2877 unsigned num_regs = 0;
2879 register char *fastmap = bufp->fastmap;
2880 unsigned char *pattern = bufp->buffer;
2881 unsigned long size = bufp->used;
2882 unsigned char *p = pattern;
2883 register unsigned char *pend = pattern + size;
2885 /* This holds the pointer to the failure stack, when
2886 it is allocated relocatably. */
2888 fail_stack_elt_t *failure_stack_ptr;
2891 /* Assume that each path through the pattern can be null until
2892 proven otherwise. We set this false at the bottom of switch
2893 statement, to which we get only if a particular path doesn't
2894 match the empty string. */
2895 boolean path_can_be_null = true;
2897 /* We aren't doing a `succeed_n' to begin with. */
2898 boolean succeed_n_p = false;
2900 assert (fastmap != NULL && p != NULL);
2903 bzero (fastmap, 1 << BYTEWIDTH); /* Assume nothing's valid. */
2904 bufp->fastmap_accurate = 1; /* It will be when we're done. */
2905 bufp->can_be_null = 0;
2909 if (p == pend || *p == succeed)
2911 /* We have reached the (effective) end of pattern. */
2912 if (!FAIL_STACK_EMPTY ())
2914 bufp->can_be_null |= path_can_be_null;
2916 /* Reset for next path. */
2917 path_can_be_null = true;
2919 p = fail_stack.stack[--fail_stack.avail].pointer;
2927 /* We should never be about to go beyond the end of the pattern. */
2930 switch (SWITCH_ENUM_CAST ((re_opcode_t) *p++))
2933 /* I guess the idea here is to simply not bother with a fastmap
2934 if a backreference is used, since it's too hard to figure out
2935 the fastmap for the corresponding group. Setting
2936 `can_be_null' stops `re_search_2' from using the fastmap, so
2937 that is all we do. */
2939 bufp->can_be_null = 1;
2943 /* Following are the cases which match a character. These end
2952 for (j = *p++ * BYTEWIDTH - 1; j >= 0; j--)
2953 if (p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH)))
2959 /* Chars beyond end of map must be allowed. */
2960 for (j = *p * BYTEWIDTH; j < (1 << BYTEWIDTH); j++)
2963 for (j = *p++ * BYTEWIDTH - 1; j >= 0; j--)
2964 if (!(p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH))))
2970 for (j = 0; j < (1 << BYTEWIDTH); j++)
2971 if (SYNTAX (j) == Sword)
2977 for (j = 0; j < (1 << BYTEWIDTH); j++)
2978 if (SYNTAX (j) != Sword)
2985 int fastmap_newline = fastmap['\n'];
2987 /* `.' matches anything ... */
2988 for (j = 0; j < (1 << BYTEWIDTH); j++)
2991 /* ... except perhaps newline. */
2992 if (!(bufp->syntax & RE_DOT_NEWLINE))
2993 fastmap['\n'] = fastmap_newline;
2995 /* Return if we have already set `can_be_null'; if we have,
2996 then the fastmap is irrelevant. Something's wrong here. */
2997 else if (bufp->can_be_null)
3000 /* Otherwise, have to check alternative paths. */
3007 for (j = 0; j < (1 << BYTEWIDTH); j++)
3008 if (SYNTAX (j) == (enum syntaxcode) k)
3015 for (j = 0; j < (1 << BYTEWIDTH); j++)
3016 if (SYNTAX (j) != (enum syntaxcode) k)
3021 /* All cases after this match the empty string. These end with
3029 #endif /* not emacs */
3041 case push_dummy_failure:
3046 case pop_failure_jump:
3047 case maybe_pop_jump:
3050 case dummy_failure_jump:
3051 EXTRACT_NUMBER_AND_INCR (j, p);
3056 /* Jump backward implies we just went through the body of a
3057 loop and matched nothing. Opcode jumped to should be
3058 `on_failure_jump' or `succeed_n'. Just treat it like an
3059 ordinary jump. For a * loop, it has pushed its failure
3060 point already; if so, discard that as redundant. */
3061 if ((re_opcode_t) *p != on_failure_jump
3062 && (re_opcode_t) *p != succeed_n)
3066 EXTRACT_NUMBER_AND_INCR (j, p);
3069 /* If what's on the stack is where we are now, pop it. */
3070 if (!FAIL_STACK_EMPTY ()
3071 && fail_stack.stack[fail_stack.avail - 1].pointer == p)
3077 case on_failure_jump:
3078 case on_failure_keep_string_jump:
3079 handle_on_failure_jump:
3080 EXTRACT_NUMBER_AND_INCR (j, p);
3082 /* For some patterns, e.g., `(a?)?', `p+j' here points to the
3083 end of the pattern. We don't want to push such a point,
3084 since when we restore it above, entering the switch will
3085 increment `p' past the end of the pattern. We don't need
3086 to push such a point since we obviously won't find any more
3087 fastmap entries beyond `pend'. Such a pattern can match
3088 the null string, though. */
3091 if (!PUSH_PATTERN_OP (p + j, fail_stack))
3093 RESET_FAIL_STACK ();
3098 bufp->can_be_null = 1;
3102 EXTRACT_NUMBER_AND_INCR (k, p); /* Skip the n. */
3103 succeed_n_p = false;
3110 /* Get to the number of times to succeed. */
3113 /* Increment p past the n for when k != 0. */
3114 EXTRACT_NUMBER_AND_INCR (k, p);
3118 succeed_n_p = true; /* Spaghetti code alert. */
3119 goto handle_on_failure_jump;
3136 abort (); /* We have listed all the cases. */
3139 /* Getting here means we have found the possible starting
3140 characters for one path of the pattern -- and that the empty
3141 string does not match. We need not follow this path further.
3142 Instead, look at the next alternative (remembered on the
3143 stack), or quit if no more. The test at the top of the loop
3144 does these things. */
3145 path_can_be_null = false;
3149 /* Set `can_be_null' for the last path (also the first path, if the
3150 pattern is empty). */
3151 bufp->can_be_null |= path_can_be_null;
3154 RESET_FAIL_STACK ();
3156 } /* re_compile_fastmap */
3158 /* Set REGS to hold NUM_REGS registers, storing them in STARTS and
3159 ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use
3160 this memory for recording register information. STARTS and ENDS
3161 must be allocated using the malloc library routine, and must each
3162 be at least NUM_REGS * sizeof (regoff_t) bytes long.
3164 If NUM_REGS == 0, then subsequent matches should allocate their own
3167 Unless this function is called, the first search or match using
3168 PATTERN_BUFFER will allocate its own register data, without
3169 freeing the old data. */
3172 re_set_registers (bufp, regs, num_regs, starts, ends)
3173 struct re_pattern_buffer *bufp;
3174 struct re_registers *regs;
3176 regoff_t *starts, *ends;
3180 bufp->regs_allocated = REGS_REALLOCATE;
3181 regs->num_regs = num_regs;
3182 regs->start = starts;
3187 bufp->regs_allocated = REGS_UNALLOCATED;
3189 regs->start = regs->end = (regoff_t *) 0;
3193 /* Searching routines. */
3195 /* Like re_search_2, below, but only one string is specified, and
3196 doesn't let you say where to stop matching. */
3199 re_search (bufp, string, size, startpos, range, regs)
3200 struct re_pattern_buffer *bufp;
3202 int size, startpos, range;
3203 struct re_registers *regs;
3205 return re_search_2 (bufp, NULL, 0, string, size, startpos, range,
3210 /* Using the compiled pattern in BUFP->buffer, first tries to match the
3211 virtual concatenation of STRING1 and STRING2, starting first at index
3212 STARTPOS, then at STARTPOS + 1, and so on.
3214 STRING1 and STRING2 have length SIZE1 and SIZE2, respectively.
3216 RANGE is how far to scan while trying to match. RANGE = 0 means try
3217 only at STARTPOS; in general, the last start tried is STARTPOS +
3220 In REGS, return the indices of the virtual concatenation of STRING1
3221 and STRING2 that matched the entire BUFP->buffer and its contained
3224 Do not consider matching one past the index STOP in the virtual
3225 concatenation of STRING1 and STRING2.
3227 We return either the position in the strings at which the match was
3228 found, -1 if no match, or -2 if error (such as failure
3232 re_search_2 (bufp, string1, size1, string2, size2, startpos, range, regs, stop)
3233 struct re_pattern_buffer *bufp;
3234 const char *string1, *string2;
3238 struct re_registers *regs;
3242 register char *fastmap = bufp->fastmap;
3243 register char *translate = bufp->translate;
3244 int total_size = size1 + size2;
3245 int endpos = startpos + range;
3247 /* Check for out-of-range STARTPOS. */
3248 if (startpos < 0 || startpos > total_size)
3251 /* Fix up RANGE if it might eventually take us outside
3252 the virtual concatenation of STRING1 and STRING2. */
3254 range = -1 - startpos;
3255 else if (endpos > total_size)
3256 range = total_size - startpos;
3258 /* If the search isn't to be a backwards one, don't waste time in a
3259 search for a pattern that must be anchored. */
3260 if (bufp->used > 0 && (re_opcode_t) bufp->buffer[0] == begbuf && range > 0)
3268 /* Update the fastmap now if not correct already. */
3269 if (fastmap && !bufp->fastmap_accurate)
3270 if (re_compile_fastmap (bufp) == -2)
3273 /* Loop through the string, looking for a place to start matching. */
3276 /* If a fastmap is supplied, skip quickly over characters that
3277 cannot be the start of a match. If the pattern can match the
3278 null string, however, we don't need to skip characters; we want
3279 the first null string. */
3280 if (fastmap && startpos < total_size && !bufp->can_be_null)
3282 if (range > 0) /* Searching forwards. */
3284 register const char *d;
3285 register int lim = 0;
3288 if (startpos < size1 && startpos + range >= size1)
3289 lim = range - (size1 - startpos);
3291 d = (startpos >= size1 ? string2 - size1 : string1) + startpos;
3293 /* Written out as an if-else to avoid testing `translate'
3297 && !fastmap[(unsigned char)
3298 translate[(unsigned char) *d++]])
3301 while (range > lim && !fastmap[(unsigned char) *d++])
3304 startpos += irange - range;
3306 else /* Searching backwards. */
3308 register char c = (size1 == 0 || startpos >= size1
3309 ? string2[startpos - size1]
3310 : string1[startpos]);
3312 if (!fastmap[(unsigned char) TRANSLATE (c)])
3317 /* If can't match the null string, and that's all we have left, fail. */
3318 if (range >= 0 && startpos == total_size && fastmap
3319 && !bufp->can_be_null)
3322 val = re_match_2_internal (bufp, string1, size1, string2, size2,
3323 startpos, regs, stop);
3324 #ifndef REGEX_MALLOC
3353 /* Declarations and macros for re_match_2. */
3355 static int bcmp_translate ();
3356 static boolean alt_match_null_string_p (),
3357 common_op_match_null_string_p (),
3358 group_match_null_string_p ();
3360 /* This converts PTR, a pointer into one of the search strings `string1'
3361 and `string2' into an offset from the beginning of that string. */
3362 #define POINTER_TO_OFFSET(ptr) \
3363 (FIRST_STRING_P (ptr) \
3364 ? ((regoff_t) ((ptr) - string1)) \
3365 : ((regoff_t) ((ptr) - string2 + size1)))
3367 /* Macros for dealing with the split strings in re_match_2. */
3369 #define MATCHING_IN_FIRST_STRING (dend == end_match_1)
3371 /* Call before fetching a character with *d. This switches over to
3372 string2 if necessary. */
3373 #define PREFETCH() \
3376 /* End of string2 => fail. */ \
3377 if (dend == end_match_2) \
3379 /* End of string1 => advance to string2. */ \
3381 dend = end_match_2; \
3385 /* Test if at very beginning or at very end of the virtual concatenation
3386 of `string1' and `string2'. If only one string, it's `string2'. */
3387 #define AT_STRINGS_BEG(d) ((d) == (size1 ? string1 : string2) || !size2)
3388 #define AT_STRINGS_END(d) ((d) == end2)
3391 /* Test if D points to a character which is word-constituent. We have
3392 two special cases to check for: if past the end of string1, look at
3393 the first character in string2; and if before the beginning of
3394 string2, look at the last character in string1. */
3395 #define WORDCHAR_P(d) \
3396 (SYNTAX ((d) == end1 ? *string2 \
3397 : (d) == string2 - 1 ? *(end1 - 1) : *(d)) \
3400 /* Test if the character before D and the one at D differ with respect
3401 to being word-constituent. */
3402 #define AT_WORD_BOUNDARY(d) \
3403 (AT_STRINGS_BEG (d) || AT_STRINGS_END (d) \
3404 || WORDCHAR_P (d - 1) != WORDCHAR_P (d))
3407 /* Free everything we malloc. */
3408 #ifdef MATCH_MAY_ALLOCATE
3409 #define FREE_VAR(var) if (var) REGEX_FREE (var); var = NULL
3410 #define FREE_VARIABLES() \
3412 REGEX_FREE_STACK (fail_stack.stack); \
3413 FREE_VAR (regstart); \
3414 FREE_VAR (regend); \
3415 FREE_VAR (old_regstart); \
3416 FREE_VAR (old_regend); \
3417 FREE_VAR (best_regstart); \
3418 FREE_VAR (best_regend); \
3419 FREE_VAR (reg_info); \
3420 FREE_VAR (reg_dummy); \
3421 FREE_VAR (reg_info_dummy); \
3424 #define FREE_VARIABLES() ((void)0) /* Do nothing! But inhibit gcc warning. */
3425 #endif /* not MATCH_MAY_ALLOCATE */
3427 /* These values must meet several constraints. They must not be valid
3428 register values; since we have a limit of 255 registers (because
3429 we use only one byte in the pattern for the register number), we can
3430 use numbers larger than 255. They must differ by 1, because of
3431 NUM_FAILURE_ITEMS above. And the value for the lowest register must
3432 be larger than the value for the highest register, so we do not try
3433 to actually save any registers when none are active. */
3434 #define NO_HIGHEST_ACTIVE_REG (1 << BYTEWIDTH)
3435 #define NO_LOWEST_ACTIVE_REG (NO_HIGHEST_ACTIVE_REG + 1)
3437 /* Matching routines. */
3439 #ifndef emacs /* Emacs never uses this. */
3440 /* re_match is like re_match_2 except it takes only a single string. */
3443 re_match (bufp, string, size, pos, regs)
3444 struct re_pattern_buffer *bufp;
3447 struct re_registers *regs;
3449 int result = re_match_2_internal (bufp, NULL, 0, string, size,
3454 #endif /* not emacs */
3457 /* re_match_2 matches the compiled pattern in BUFP against the
3458 the (virtual) concatenation of STRING1 and STRING2 (of length SIZE1
3459 and SIZE2, respectively). We start matching at POS, and stop
3462 If REGS is non-null and the `no_sub' field of BUFP is nonzero, we
3463 store offsets for the substring each group matched in REGS. See the
3464 documentation for exactly how many groups we fill.
3466 We return -1 if no match, -2 if an internal error (such as the
3467 failure stack overflowing). Otherwise, we return the length of the
3468 matched substring. */
3471 re_match_2 (bufp, string1, size1, string2, size2, pos, regs, stop)
3472 struct re_pattern_buffer *bufp;
3473 const char *string1, *string2;
3476 struct re_registers *regs;
3479 int result = re_match_2_internal (bufp, string1, size1, string2, size2,
3485 /* This is a separate function so that we can force an alloca cleanup
3488 re_match_2_internal (bufp, string1, size1, string2, size2, pos, regs, stop)
3489 struct re_pattern_buffer *bufp;
3490 const char *string1, *string2;
3493 struct re_registers *regs;
3496 /* General temporaries. */
3500 /* Just past the end of the corresponding string. */
3501 const char *end1, *end2;
3503 /* Pointers into string1 and string2, just past the last characters in
3504 each to consider matching. */
3505 const char *end_match_1, *end_match_2;
3507 /* Where we are in the data, and the end of the current string. */
3508 const char *d, *dend;
3510 /* Where we are in the pattern, and the end of the pattern. */
3511 unsigned char *p = bufp->buffer;
3512 register unsigned char *pend = p + bufp->used;
3514 /* Mark the opcode just after a start_memory, so we can test for an
3515 empty subpattern when we get to the stop_memory. */
3516 unsigned char *just_past_start_mem = 0;
3518 /* We use this to map every character in the string. */
3519 char *translate = bufp->translate;
3521 /* Failure point stack. Each place that can handle a failure further
3522 down the line pushes a failure point on this stack. It consists of
3523 restart, regend, and reg_info for all registers corresponding to
3524 the subexpressions we're currently inside, plus the number of such
3525 registers, and, finally, two char *'s. The first char * is where
3526 to resume scanning the pattern; the second one is where to resume
3527 scanning the strings. If the latter is zero, the failure point is
3528 a ``dummy''; if a failure happens and the failure point is a dummy,
3529 it gets discarded and the next next one is tried. */
3530 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
3531 fail_stack_type fail_stack;
3534 static unsigned failure_id = 0;
3535 unsigned nfailure_points_pushed = 0, nfailure_points_popped = 0;
3538 /* This holds the pointer to the failure stack, when
3539 it is allocated relocatably. */
3541 fail_stack_elt_t *failure_stack_ptr;
3544 /* We fill all the registers internally, independent of what we
3545 return, for use in backreferences. The number here includes
3546 an element for register zero. */
3547 unsigned num_regs = bufp->re_nsub + 1;
3549 /* The currently active registers. */
3550 unsigned lowest_active_reg = NO_LOWEST_ACTIVE_REG;
3551 unsigned highest_active_reg = NO_HIGHEST_ACTIVE_REG;
3553 /* Information on the contents of registers. These are pointers into
3554 the input strings; they record just what was matched (on this
3555 attempt) by a subexpression part of the pattern, that is, the
3556 regnum-th regstart pointer points to where in the pattern we began
3557 matching and the regnum-th regend points to right after where we
3558 stopped matching the regnum-th subexpression. (The zeroth register
3559 keeps track of what the whole pattern matches.) */
3560 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
3561 const char **regstart, **regend;
3564 /* If a group that's operated upon by a repetition operator fails to
3565 match anything, then the register for its start will need to be
3566 restored because it will have been set to wherever in the string we
3567 are when we last see its open-group operator. Similarly for a
3569 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
3570 const char **old_regstart, **old_regend;
3573 /* The is_active field of reg_info helps us keep track of which (possibly
3574 nested) subexpressions we are currently in. The matched_something
3575 field of reg_info[reg_num] helps us tell whether or not we have
3576 matched any of the pattern so far this time through the reg_num-th
3577 subexpression. These two fields get reset each time through any
3578 loop their register is in. */
3579 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
3580 register_info_type *reg_info;
3583 /* The following record the register info as found in the above
3584 variables when we find a match better than any we've seen before.
3585 This happens as we backtrack through the failure points, which in
3586 turn happens only if we have not yet matched the entire string. */
3587 unsigned best_regs_set = false;
3588 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
3589 const char **best_regstart, **best_regend;
3592 /* Logically, this is `best_regend[0]'. But we don't want to have to
3593 allocate space for that if we're not allocating space for anything
3594 else (see below). Also, we never need info about register 0 for
3595 any of the other register vectors, and it seems rather a kludge to
3596 treat `best_regend' differently than the rest. So we keep track of
3597 the end of the best match so far in a separate variable. We
3598 initialize this to NULL so that when we backtrack the first time
3599 and need to test it, it's not garbage. */
3600 const char *match_end = NULL;
3602 /* This helps SET_REGS_MATCHED avoid doing redundant work. */
3603 int set_regs_matched_done = 0;
3605 /* Used when we pop values we don't care about. */
3606 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
3607 const char **reg_dummy;
3608 register_info_type *reg_info_dummy;
3612 /* Counts the total number of registers pushed. */
3613 unsigned num_regs_pushed = 0;
3616 DEBUG_PRINT1 ("\n\nEntering re_match_2.\n");
3620 #ifdef MATCH_MAY_ALLOCATE
3621 /* Do not bother to initialize all the register variables if there are
3622 no groups in the pattern, as it takes a fair amount of time. If
3623 there are groups, we include space for register 0 (the whole
3624 pattern), even though we never use it, since it simplifies the
3625 array indexing. We should fix this. */
3628 regstart = REGEX_TALLOC (num_regs, const char *);
3629 regend = REGEX_TALLOC (num_regs, const char *);
3630 old_regstart = REGEX_TALLOC (num_regs, const char *);
3631 old_regend = REGEX_TALLOC (num_regs, const char *);
3632 best_regstart = REGEX_TALLOC (num_regs, const char *);
3633 best_regend = REGEX_TALLOC (num_regs, const char *);
3634 reg_info = REGEX_TALLOC (num_regs, register_info_type);
3635 reg_dummy = REGEX_TALLOC (num_regs, const char *);
3636 reg_info_dummy = REGEX_TALLOC (num_regs, register_info_type);
3638 if (!(regstart && regend && old_regstart && old_regend && reg_info
3639 && best_regstart && best_regend && reg_dummy && reg_info_dummy))
3647 /* We must initialize all our variables to NULL, so that
3648 `FREE_VARIABLES' doesn't try to free them. */
3649 regstart = regend = old_regstart = old_regend = best_regstart
3650 = best_regend = reg_dummy = NULL;
3651 reg_info = reg_info_dummy = (register_info_type *) NULL;
3653 #endif /* MATCH_MAY_ALLOCATE */
3655 /* The starting position is bogus. */
3656 if (pos < 0 || pos > size1 + size2)
3662 /* Initialize subexpression text positions to -1 to mark ones that no
3663 start_memory/stop_memory has been seen for. Also initialize the
3664 register information struct. */
3665 for (mcnt = 1; mcnt < num_regs; mcnt++)
3667 regstart[mcnt] = regend[mcnt]
3668 = old_regstart[mcnt] = old_regend[mcnt] = REG_UNSET_VALUE;
3670 REG_MATCH_NULL_STRING_P (reg_info[mcnt]) = MATCH_NULL_UNSET_VALUE;
3671 IS_ACTIVE (reg_info[mcnt]) = 0;
3672 MATCHED_SOMETHING (reg_info[mcnt]) = 0;
3673 EVER_MATCHED_SOMETHING (reg_info[mcnt]) = 0;
3676 /* We move `string1' into `string2' if the latter's empty -- but not if
3677 `string1' is null. */
3678 if (size2 == 0 && string1 != NULL)
3685 end1 = string1 + size1;
3686 end2 = string2 + size2;
3688 /* Compute where to stop matching, within the two strings. */
3691 end_match_1 = string1 + stop;
3692 end_match_2 = string2;
3697 end_match_2 = string2 + stop - size1;
3700 /* `p' scans through the pattern as `d' scans through the data.
3701 `dend' is the end of the input string that `d' points within. `d'
3702 is advanced into the following input string whenever necessary, but
3703 this happens before fetching; therefore, at the beginning of the
3704 loop, `d' can be pointing at the end of a string, but it cannot
3706 if (size1 > 0 && pos <= size1)
3713 d = string2 + pos - size1;
3717 DEBUG_PRINT1 ("The compiled pattern is: ");
3718 DEBUG_PRINT_COMPILED_PATTERN (bufp, p, pend);
3719 DEBUG_PRINT1 ("The string to match is: `");
3720 DEBUG_PRINT_DOUBLE_STRING (d, string1, size1, string2, size2);
3721 DEBUG_PRINT1 ("'\n");
3723 /* This loops over pattern commands. It exits by returning from the
3724 function if the match is complete, or it drops through if the match
3725 fails at this starting point in the input data. */
3728 DEBUG_PRINT2 ("\n0x%x: ", p);
3731 { /* End of pattern means we might have succeeded. */
3732 DEBUG_PRINT1 ("end of pattern ... ");
3734 /* If we haven't matched the entire string, and we want the
3735 longest match, try backtracking. */
3736 if (d != end_match_2)
3738 /* 1 if this match ends in the same string (string1 or string2)
3739 as the best previous match. */
3740 boolean same_str_p = (FIRST_STRING_P (match_end)
3741 == MATCHING_IN_FIRST_STRING);
3742 /* 1 if this match is the best seen so far. */
3743 boolean best_match_p;
3745 /* AIX compiler got confused when this was combined
3746 with the previous declaration. */
3748 best_match_p = d > match_end;
3750 best_match_p = !MATCHING_IN_FIRST_STRING;
3752 DEBUG_PRINT1 ("backtracking.\n");
3754 if (!FAIL_STACK_EMPTY ())
3755 { /* More failure points to try. */
3757 /* If exceeds best match so far, save it. */
3758 if (!best_regs_set || best_match_p)
3760 best_regs_set = true;
3763 DEBUG_PRINT1 ("\nSAVING match as best so far.\n");
3765 for (mcnt = 1; mcnt < num_regs; mcnt++)
3767 best_regstart[mcnt] = regstart[mcnt];
3768 best_regend[mcnt] = regend[mcnt];
3774 /* If no failure points, don't restore garbage. And if
3775 last match is real best match, don't restore second
3777 else if (best_regs_set && !best_match_p)
3780 /* Restore best match. It may happen that `dend ==
3781 end_match_1' while the restored d is in string2.
3782 For example, the pattern `x.*y.*z' against the
3783 strings `x-' and `y-z-', if the two strings are
3784 not consecutive in memory. */
3785 DEBUG_PRINT1 ("Restoring best registers.\n");
3788 dend = ((d >= string1 && d <= end1)
3789 ? end_match_1 : end_match_2);
3791 for (mcnt = 1; mcnt < num_regs; mcnt++)
3793 regstart[mcnt] = best_regstart[mcnt];
3794 regend[mcnt] = best_regend[mcnt];
3797 } /* d != end_match_2 */
3800 DEBUG_PRINT1 ("Accepting match.\n");
3802 /* If caller wants register contents data back, do it. */
3803 if (regs && !bufp->no_sub)
3805 /* Have the register data arrays been allocated? */
3806 if (bufp->regs_allocated == REGS_UNALLOCATED)
3807 { /* No. So allocate them with malloc. We need one
3808 extra element beyond `num_regs' for the `-1' marker
3810 regs->num_regs = MAX (RE_NREGS, num_regs + 1);
3811 regs->start = TALLOC (regs->num_regs, regoff_t);
3812 regs->end = TALLOC (regs->num_regs, regoff_t);
3813 if (regs->start == NULL || regs->end == NULL)
3818 bufp->regs_allocated = REGS_REALLOCATE;
3820 else if (bufp->regs_allocated == REGS_REALLOCATE)
3821 { /* Yes. If we need more elements than were already
3822 allocated, reallocate them. If we need fewer, just
3824 if (regs->num_regs < num_regs + 1)
3826 regs->num_regs = num_regs + 1;
3827 RETALLOC (regs->start, regs->num_regs, regoff_t);
3828 RETALLOC (regs->end, regs->num_regs, regoff_t);
3829 if (regs->start == NULL || regs->end == NULL)
3838 /* These braces fend off a "empty body in an else-statement"
3839 warning under GCC when assert expands to nothing. */
3840 assert (bufp->regs_allocated == REGS_FIXED);
3843 /* Convert the pointer data in `regstart' and `regend' to
3844 indices. Register zero has to be set differently,
3845 since we haven't kept track of any info for it. */
3846 if (regs->num_regs > 0)
3848 regs->start[0] = pos;
3849 regs->end[0] = (MATCHING_IN_FIRST_STRING
3850 ? ((regoff_t) (d - string1))
3851 : ((regoff_t) (d - string2 + size1)));
3854 /* Go through the first `min (num_regs, regs->num_regs)'
3855 registers, since that is all we initialized. */
3856 for (mcnt = 1; mcnt < MIN (num_regs, regs->num_regs); mcnt++)
3858 if (REG_UNSET (regstart[mcnt]) || REG_UNSET (regend[mcnt]))
3859 regs->start[mcnt] = regs->end[mcnt] = -1;
3863 = (regoff_t) POINTER_TO_OFFSET (regstart[mcnt]);
3865 = (regoff_t) POINTER_TO_OFFSET (regend[mcnt]);
3869 /* If the regs structure we return has more elements than
3870 were in the pattern, set the extra elements to -1. If
3871 we (re)allocated the registers, this is the case,
3872 because we always allocate enough to have at least one
3874 for (mcnt = num_regs; mcnt < regs->num_regs; mcnt++)
3875 regs->start[mcnt] = regs->end[mcnt] = -1;
3876 } /* regs && !bufp->no_sub */
3878 DEBUG_PRINT4 ("%u failure points pushed, %u popped (%u remain).\n",
3879 nfailure_points_pushed, nfailure_points_popped,
3880 nfailure_points_pushed - nfailure_points_popped);
3881 DEBUG_PRINT2 ("%u registers pushed.\n", num_regs_pushed);
3883 mcnt = d - pos - (MATCHING_IN_FIRST_STRING
3887 DEBUG_PRINT2 ("Returning %d from re_match_2.\n", mcnt);
3893 /* Otherwise match next pattern command. */
3894 switch (SWITCH_ENUM_CAST ((re_opcode_t) *p++))
3896 /* Ignore these. Used to ignore the n of succeed_n's which
3897 currently have n == 0. */
3899 DEBUG_PRINT1 ("EXECUTING no_op.\n");
3903 DEBUG_PRINT1 ("EXECUTING succeed.\n");
3906 /* Match the next n pattern characters exactly. The following
3907 byte in the pattern defines n, and the n bytes after that
3908 are the characters to match. */
3911 DEBUG_PRINT2 ("EXECUTING exactn %d.\n", mcnt);
3913 /* This is written out as an if-else so we don't waste time
3914 testing `translate' inside the loop. */
3920 if (translate[(unsigned char) *d++] != (char) *p++)
3930 if (*d++ != (char) *p++) goto fail;
3934 SET_REGS_MATCHED ();
3938 /* Match any character except possibly a newline or a null. */
3940 DEBUG_PRINT1 ("EXECUTING anychar.\n");
3944 if ((!(bufp->syntax & RE_DOT_NEWLINE) && TRANSLATE (*d) == '\n')
3945 || (bufp->syntax & RE_DOT_NOT_NULL && TRANSLATE (*d) == '\000'))
3948 SET_REGS_MATCHED ();
3949 DEBUG_PRINT2 (" Matched `%d'.\n", *d);
3957 register unsigned char c;
3958 boolean not = (re_opcode_t) *(p - 1) == charset_not;
3960 DEBUG_PRINT2 ("EXECUTING charset%s.\n", not ? "_not" : "");
3963 c = TRANSLATE (*d); /* The character to match. */
3965 /* Cast to `unsigned' instead of `unsigned char' in case the
3966 bit list is a full 32 bytes long. */
3967 if (c < (unsigned) (*p * BYTEWIDTH)
3968 && p[1 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH)))
3973 if (!not) goto fail;
3975 SET_REGS_MATCHED ();
3981 /* The beginning of a group is represented by start_memory.
3982 The arguments are the register number in the next byte, and the
3983 number of groups inner to this one in the next. The text
3984 matched within the group is recorded (in the internal
3985 registers data structure) under the register number. */
3987 DEBUG_PRINT3 ("EXECUTING start_memory %d (%d):\n", *p, p[1]);
3989 /* Find out if this group can match the empty string. */
3990 p1 = p; /* To send to group_match_null_string_p. */
3992 if (REG_MATCH_NULL_STRING_P (reg_info[*p]) == MATCH_NULL_UNSET_VALUE)
3993 REG_MATCH_NULL_STRING_P (reg_info[*p])
3994 = group_match_null_string_p (&p1, pend, reg_info);
3996 /* Save the position in the string where we were the last time
3997 we were at this open-group operator in case the group is
3998 operated upon by a repetition operator, e.g., with `(a*)*b'
3999 against `ab'; then we want to ignore where we are now in
4000 the string in case this attempt to match fails. */
4001 old_regstart[*p] = REG_MATCH_NULL_STRING_P (reg_info[*p])
4002 ? REG_UNSET (regstart[*p]) ? d : regstart[*p]
4004 DEBUG_PRINT2 (" old_regstart: %d\n",
4005 POINTER_TO_OFFSET (old_regstart[*p]));
4008 DEBUG_PRINT2 (" regstart: %d\n", POINTER_TO_OFFSET (regstart[*p]));
4010 IS_ACTIVE (reg_info[*p]) = 1;
4011 MATCHED_SOMETHING (reg_info[*p]) = 0;
4013 /* Clear this whenever we change the register activity status. */
4014 set_regs_matched_done = 0;
4016 /* This is the new highest active register. */
4017 highest_active_reg = *p;
4019 /* If nothing was active before, this is the new lowest active
4021 if (lowest_active_reg == NO_LOWEST_ACTIVE_REG)
4022 lowest_active_reg = *p;
4024 /* Move past the register number and inner group count. */
4026 just_past_start_mem = p;
4031 /* The stop_memory opcode represents the end of a group. Its
4032 arguments are the same as start_memory's: the register
4033 number, and the number of inner groups. */
4035 DEBUG_PRINT3 ("EXECUTING stop_memory %d (%d):\n", *p, p[1]);
4037 /* We need to save the string position the last time we were at
4038 this close-group operator in case the group is operated
4039 upon by a repetition operator, e.g., with `((a*)*(b*)*)*'
4040 against `aba'; then we want to ignore where we are now in
4041 the string in case this attempt to match fails. */
4042 old_regend[*p] = REG_MATCH_NULL_STRING_P (reg_info[*p])
4043 ? REG_UNSET (regend[*p]) ? d : regend[*p]
4045 DEBUG_PRINT2 (" old_regend: %d\n",
4046 POINTER_TO_OFFSET (old_regend[*p]));
4049 DEBUG_PRINT2 (" regend: %d\n", POINTER_TO_OFFSET (regend[*p]));
4051 /* This register isn't active anymore. */
4052 IS_ACTIVE (reg_info[*p]) = 0;
4054 /* Clear this whenever we change the register activity status. */
4055 set_regs_matched_done = 0;
4057 /* If this was the only register active, nothing is active
4059 if (lowest_active_reg == highest_active_reg)
4061 lowest_active_reg = NO_LOWEST_ACTIVE_REG;
4062 highest_active_reg = NO_HIGHEST_ACTIVE_REG;
4065 { /* We must scan for the new highest active register, since
4066 it isn't necessarily one less than now: consider
4067 (a(b)c(d(e)f)g). When group 3 ends, after the f), the
4068 new highest active register is 1. */
4069 unsigned char r = *p - 1;
4070 while (r > 0 && !IS_ACTIVE (reg_info[r]))
4073 /* If we end up at register zero, that means that we saved
4074 the registers as the result of an `on_failure_jump', not
4075 a `start_memory', and we jumped to past the innermost
4076 `stop_memory'. For example, in ((.)*) we save
4077 registers 1 and 2 as a result of the *, but when we pop
4078 back to the second ), we are at the stop_memory 1.
4079 Thus, nothing is active. */
4082 lowest_active_reg = NO_LOWEST_ACTIVE_REG;
4083 highest_active_reg = NO_HIGHEST_ACTIVE_REG;
4086 highest_active_reg = r;
4089 /* If just failed to match something this time around with a
4090 group that's operated on by a repetition operator, try to
4091 force exit from the ``loop'', and restore the register
4092 information for this group that we had before trying this
4094 if ((!MATCHED_SOMETHING (reg_info[*p])
4095 || just_past_start_mem == p - 1)
4098 boolean is_a_jump_n = false;
4102 switch ((re_opcode_t) *p1++)
4106 case pop_failure_jump:
4107 case maybe_pop_jump:
4109 case dummy_failure_jump:
4110 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4120 /* If the next operation is a jump backwards in the pattern
4121 to an on_failure_jump right before the start_memory
4122 corresponding to this stop_memory, exit from the loop
4123 by forcing a failure after pushing on the stack the
4124 on_failure_jump's jump in the pattern, and d. */
4125 if (mcnt < 0 && (re_opcode_t) *p1 == on_failure_jump
4126 && (re_opcode_t) p1[3] == start_memory && p1[4] == *p)
4128 /* If this group ever matched anything, then restore
4129 what its registers were before trying this last
4130 failed match, e.g., with `(a*)*b' against `ab' for
4131 regstart[1], and, e.g., with `((a*)*(b*)*)*'
4132 against `aba' for regend[3].
4134 Also restore the registers for inner groups for,
4135 e.g., `((a*)(b*))*' against `aba' (register 3 would
4136 otherwise get trashed). */
4138 if (EVER_MATCHED_SOMETHING (reg_info[*p]))
4142 EVER_MATCHED_SOMETHING (reg_info[*p]) = 0;
4144 /* Restore this and inner groups' (if any) registers. */
4145 for (r = *p; r < *p + *(p + 1); r++)
4147 regstart[r] = old_regstart[r];
4149 /* xx why this test? */
4150 if (old_regend[r] >= regstart[r])
4151 regend[r] = old_regend[r];
4155 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4156 PUSH_FAILURE_POINT (p1 + mcnt, d, -2);
4162 /* Move past the register number and the inner group count. */
4167 /* \<digit> has been turned into a `duplicate' command which is
4168 followed by the numeric value of <digit> as the register number. */
4171 register const char *d2, *dend2;
4172 int regno = *p++; /* Get which register to match against. */
4173 DEBUG_PRINT2 ("EXECUTING duplicate %d.\n", regno);
4175 /* Can't back reference a group which we've never matched. */
4176 if (REG_UNSET (regstart[regno]) || REG_UNSET (regend[regno]))
4179 /* Where in input to try to start matching. */
4180 d2 = regstart[regno];
4182 /* Where to stop matching; if both the place to start and
4183 the place to stop matching are in the same string, then
4184 set to the place to stop, otherwise, for now have to use
4185 the end of the first string. */
4187 dend2 = ((FIRST_STRING_P (regstart[regno])
4188 == FIRST_STRING_P (regend[regno]))
4189 ? regend[regno] : end_match_1);
4192 /* If necessary, advance to next segment in register
4196 if (dend2 == end_match_2) break;
4197 if (dend2 == regend[regno]) break;
4199 /* End of string1 => advance to string2. */
4201 dend2 = regend[regno];
4203 /* At end of register contents => success */
4204 if (d2 == dend2) break;
4206 /* If necessary, advance to next segment in data. */
4209 /* How many characters left in this segment to match. */
4212 /* Want how many consecutive characters we can match in
4213 one shot, so, if necessary, adjust the count. */
4214 if (mcnt > dend2 - d2)
4217 /* Compare that many; failure if mismatch, else move
4220 ? bcmp_translate (d, d2, mcnt, translate)
4221 : bcmp (d, d2, mcnt))
4223 d += mcnt, d2 += mcnt;
4225 /* Do this because we've match some characters. */
4226 SET_REGS_MATCHED ();
4232 /* begline matches the empty string at the beginning of the string
4233 (unless `not_bol' is set in `bufp'), and, if
4234 `newline_anchor' is set, after newlines. */
4236 DEBUG_PRINT1 ("EXECUTING begline.\n");
4238 if (AT_STRINGS_BEG (d))
4240 if (!bufp->not_bol) break;
4242 else if (d[-1] == '\n' && bufp->newline_anchor)
4246 /* In all other cases, we fail. */
4250 /* endline is the dual of begline. */
4252 DEBUG_PRINT1 ("EXECUTING endline.\n");
4254 if (AT_STRINGS_END (d))
4256 if (!bufp->not_eol) break;
4259 /* We have to ``prefetch'' the next character. */
4260 else if ((d == end1 ? *string2 : *d) == '\n'
4261 && bufp->newline_anchor)
4268 /* Match at the very beginning of the data. */
4270 DEBUG_PRINT1 ("EXECUTING begbuf.\n");
4271 if (AT_STRINGS_BEG (d))
4276 /* Match at the very end of the data. */
4278 DEBUG_PRINT1 ("EXECUTING endbuf.\n");
4279 if (AT_STRINGS_END (d))
4284 /* on_failure_keep_string_jump is used to optimize `.*\n'. It
4285 pushes NULL as the value for the string on the stack. Then
4286 `pop_failure_point' will keep the current value for the
4287 string, instead of restoring it. To see why, consider
4288 matching `foo\nbar' against `.*\n'. The .* matches the foo;
4289 then the . fails against the \n. But the next thing we want
4290 to do is match the \n against the \n; if we restored the
4291 string value, we would be back at the foo.
4293 Because this is used only in specific cases, we don't need to
4294 check all the things that `on_failure_jump' does, to make
4295 sure the right things get saved on the stack. Hence we don't
4296 share its code. The only reason to push anything on the
4297 stack at all is that otherwise we would have to change
4298 `anychar's code to do something besides goto fail in this
4299 case; that seems worse than this. */
4300 case on_failure_keep_string_jump:
4301 DEBUG_PRINT1 ("EXECUTING on_failure_keep_string_jump");
4303 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4304 DEBUG_PRINT3 (" %d (to 0x%x):\n", mcnt, p + mcnt);
4306 PUSH_FAILURE_POINT (p + mcnt, NULL, -2);
4310 /* Uses of on_failure_jump:
4312 Each alternative starts with an on_failure_jump that points
4313 to the beginning of the next alternative. Each alternative
4314 except the last ends with a jump that in effect jumps past
4315 the rest of the alternatives. (They really jump to the
4316 ending jump of the following alternative, because tensioning
4317 these jumps is a hassle.)
4319 Repeats start with an on_failure_jump that points past both
4320 the repetition text and either the following jump or
4321 pop_failure_jump back to this on_failure_jump. */
4322 case on_failure_jump:
4324 DEBUG_PRINT1 ("EXECUTING on_failure_jump");
4326 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4327 DEBUG_PRINT3 (" %d (to 0x%x)", mcnt, p + mcnt);
4329 /* If this on_failure_jump comes right before a group (i.e.,
4330 the original * applied to a group), save the information
4331 for that group and all inner ones, so that if we fail back
4332 to this point, the group's information will be correct.
4333 For example, in \(a*\)*\1, we need the preceding group,
4334 and in \(\(a*\)b*\)\2, we need the inner group. */
4336 /* We can't use `p' to check ahead because we push
4337 a failure point to `p + mcnt' after we do this. */
4340 /* We need to skip no_op's before we look for the
4341 start_memory in case this on_failure_jump is happening as
4342 the result of a completed succeed_n, as in \(a\)\{1,3\}b\1
4344 while (p1 < pend && (re_opcode_t) *p1 == no_op)
4347 if (p1 < pend && (re_opcode_t) *p1 == start_memory)
4349 /* We have a new highest active register now. This will
4350 get reset at the start_memory we are about to get to,
4351 but we will have saved all the registers relevant to
4352 this repetition op, as described above. */
4353 highest_active_reg = *(p1 + 1) + *(p1 + 2);
4354 if (lowest_active_reg == NO_LOWEST_ACTIVE_REG)
4355 lowest_active_reg = *(p1 + 1);
4358 DEBUG_PRINT1 (":\n");
4359 PUSH_FAILURE_POINT (p + mcnt, d, -2);
4363 /* A smart repeat ends with `maybe_pop_jump'.
4364 We change it to either `pop_failure_jump' or `jump'. */
4365 case maybe_pop_jump:
4366 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4367 DEBUG_PRINT2 ("EXECUTING maybe_pop_jump %d.\n", mcnt);
4369 register unsigned char *p2 = p;
4371 /* Compare the beginning of the repeat with what in the
4372 pattern follows its end. If we can establish that there
4373 is nothing that they would both match, i.e., that we
4374 would have to backtrack because of (as in, e.g., `a*a')
4375 then we can change to pop_failure_jump, because we'll
4376 never have to backtrack.
4378 This is not true in the case of alternatives: in
4379 `(a|ab)*' we do need to backtrack to the `ab' alternative
4380 (e.g., if the string was `ab'). But instead of trying to
4381 detect that here, the alternative has put on a dummy
4382 failure point which is what we will end up popping. */
4384 /* Skip over open/close-group commands.
4385 If what follows this loop is a ...+ construct,
4386 look at what begins its body, since we will have to
4387 match at least one of that. */
4391 && ((re_opcode_t) *p2 == stop_memory
4392 || (re_opcode_t) *p2 == start_memory))
4394 else if (p2 + 6 < pend
4395 && (re_opcode_t) *p2 == dummy_failure_jump)
4402 /* p1[0] ... p1[2] are the `on_failure_jump' corresponding
4403 to the `maybe_finalize_jump' of this case. Examine what
4406 /* If we're at the end of the pattern, we can change. */
4409 /* Consider what happens when matching ":\(.*\)"
4410 against ":/". I don't really understand this code
4412 p[-3] = (unsigned char) pop_failure_jump;
4414 (" End of pattern: change to `pop_failure_jump'.\n");
4417 else if ((re_opcode_t) *p2 == exactn
4418 || (bufp->newline_anchor && (re_opcode_t) *p2 == endline))
4420 register unsigned char c
4421 = *p2 == (unsigned char) endline ? '\n' : p2[2];
4423 if ((re_opcode_t) p1[3] == exactn && p1[5] != c)
4425 p[-3] = (unsigned char) pop_failure_jump;
4426 DEBUG_PRINT3 (" %c != %c => pop_failure_jump.\n",
4430 else if ((re_opcode_t) p1[3] == charset
4431 || (re_opcode_t) p1[3] == charset_not)
4433 int not = (re_opcode_t) p1[3] == charset_not;
4435 if (c < (unsigned char) (p1[4] * BYTEWIDTH)
4436 && p1[5 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH)))
4439 /* `not' is equal to 1 if c would match, which means
4440 that we can't change to pop_failure_jump. */
4443 p[-3] = (unsigned char) pop_failure_jump;
4444 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
4448 else if ((re_opcode_t) *p2 == charset)
4451 register unsigned char c
4452 = *p2 == (unsigned char) endline ? '\n' : p2[2];
4455 if ((re_opcode_t) p1[3] == exactn
4456 && ! ((int) p2[1] * BYTEWIDTH > (int) p1[4]
4457 && (p2[1 + p1[4] / BYTEWIDTH]
4458 & (1 << (p1[4] % BYTEWIDTH)))))
4460 p[-3] = (unsigned char) pop_failure_jump;
4461 DEBUG_PRINT3 (" %c != %c => pop_failure_jump.\n",
4465 else if ((re_opcode_t) p1[3] == charset_not)
4468 /* We win if the charset_not inside the loop
4469 lists every character listed in the charset after. */
4470 for (idx = 0; idx < (int) p2[1]; idx++)
4471 if (! (p2[2 + idx] == 0
4472 || (idx < (int) p1[4]
4473 && ((p2[2 + idx] & ~ p1[5 + idx]) == 0))))
4478 p[-3] = (unsigned char) pop_failure_jump;
4479 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
4482 else if ((re_opcode_t) p1[3] == charset)
4485 /* We win if the charset inside the loop
4486 has no overlap with the one after the loop. */
4488 idx < (int) p2[1] && idx < (int) p1[4];
4490 if ((p2[2 + idx] & p1[5 + idx]) != 0)
4493 if (idx == p2[1] || idx == p1[4])
4495 p[-3] = (unsigned char) pop_failure_jump;
4496 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
4501 p -= 2; /* Point at relative address again. */
4502 if ((re_opcode_t) p[-1] != pop_failure_jump)
4504 p[-1] = (unsigned char) jump;
4505 DEBUG_PRINT1 (" Match => jump.\n");
4506 goto unconditional_jump;
4508 /* Note fall through. */
4511 /* The end of a simple repeat has a pop_failure_jump back to
4512 its matching on_failure_jump, where the latter will push a
4513 failure point. The pop_failure_jump takes off failure
4514 points put on by this pop_failure_jump's matching
4515 on_failure_jump; we got through the pattern to here from the
4516 matching on_failure_jump, so didn't fail. */
4517 case pop_failure_jump:
4519 /* We need to pass separate storage for the lowest and
4520 highest registers, even though we don't care about the
4521 actual values. Otherwise, we will restore only one
4522 register from the stack, since lowest will == highest in
4523 `pop_failure_point'. */
4524 unsigned dummy_low_reg, dummy_high_reg;
4525 unsigned char *pdummy;
4528 DEBUG_PRINT1 ("EXECUTING pop_failure_jump.\n");
4529 POP_FAILURE_POINT (sdummy, pdummy,
4530 dummy_low_reg, dummy_high_reg,
4531 reg_dummy, reg_dummy, reg_info_dummy);
4533 /* Note fall through. */
4536 /* Unconditionally jump (without popping any failure points). */
4539 EXTRACT_NUMBER_AND_INCR (mcnt, p); /* Get the amount to jump. */
4540 DEBUG_PRINT2 ("EXECUTING jump %d ", mcnt);
4541 p += mcnt; /* Do the jump. */
4542 DEBUG_PRINT2 ("(to 0x%x).\n", p);
4546 /* We need this opcode so we can detect where alternatives end
4547 in `group_match_null_string_p' et al. */
4549 DEBUG_PRINT1 ("EXECUTING jump_past_alt.\n");
4550 goto unconditional_jump;
4553 /* Normally, the on_failure_jump pushes a failure point, which
4554 then gets popped at pop_failure_jump. We will end up at
4555 pop_failure_jump, also, and with a pattern of, say, `a+', we
4556 are skipping over the on_failure_jump, so we have to push
4557 something meaningless for pop_failure_jump to pop. */
4558 case dummy_failure_jump:
4559 DEBUG_PRINT1 ("EXECUTING dummy_failure_jump.\n");
4560 /* It doesn't matter what we push for the string here. What
4561 the code at `fail' tests is the value for the pattern. */
4562 PUSH_FAILURE_POINT (0, 0, -2);
4563 goto unconditional_jump;
4566 /* At the end of an alternative, we need to push a dummy failure
4567 point in case we are followed by a `pop_failure_jump', because
4568 we don't want the failure point for the alternative to be
4569 popped. For example, matching `(a|ab)*' against `aab'
4570 requires that we match the `ab' alternative. */
4571 case push_dummy_failure:
4572 DEBUG_PRINT1 ("EXECUTING push_dummy_failure.\n");
4573 /* See comments just above at `dummy_failure_jump' about the
4575 PUSH_FAILURE_POINT (0, 0, -2);
4578 /* Have to succeed matching what follows at least n times.
4579 After that, handle like `on_failure_jump'. */
4581 EXTRACT_NUMBER (mcnt, p + 2);
4582 DEBUG_PRINT2 ("EXECUTING succeed_n %d.\n", mcnt);
4585 /* Originally, this is how many times we HAVE to succeed. */
4590 STORE_NUMBER_AND_INCR (p, mcnt);
4591 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p, mcnt);
4595 DEBUG_PRINT2 (" Setting two bytes from 0x%x to no_op.\n", p+2);
4596 p[2] = (unsigned char) no_op;
4597 p[3] = (unsigned char) no_op;
4603 EXTRACT_NUMBER (mcnt, p + 2);
4604 DEBUG_PRINT2 ("EXECUTING jump_n %d.\n", mcnt);
4606 /* Originally, this is how many times we CAN jump. */
4610 STORE_NUMBER (p + 2, mcnt);
4611 goto unconditional_jump;
4613 /* If don't have to jump any more, skip over the rest of command. */
4620 DEBUG_PRINT1 ("EXECUTING set_number_at.\n");
4622 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4624 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4625 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p1, mcnt);
4626 STORE_NUMBER (p1, mcnt);
4631 DEBUG_PRINT1 ("EXECUTING wordbound.\n");
4632 if (AT_WORD_BOUNDARY (d))
4637 DEBUG_PRINT1 ("EXECUTING notwordbound.\n");
4638 if (AT_WORD_BOUNDARY (d))
4643 DEBUG_PRINT1 ("EXECUTING wordbeg.\n");
4644 if (WORDCHAR_P (d) && (AT_STRINGS_BEG (d) || !WORDCHAR_P (d - 1)))
4649 DEBUG_PRINT1 ("EXECUTING wordend.\n");
4650 if (!AT_STRINGS_BEG (d) && WORDCHAR_P (d - 1)
4651 && (!WORDCHAR_P (d) || AT_STRINGS_END (d)))
4657 DEBUG_PRINT1 ("EXECUTING before_dot.\n");
4658 if (PTR_CHAR_POS ((unsigned char *) d) >= point)
4663 DEBUG_PRINT1 ("EXECUTING at_dot.\n");
4664 if (PTR_CHAR_POS ((unsigned char *) d) != point)
4669 DEBUG_PRINT1 ("EXECUTING after_dot.\n");
4670 if (PTR_CHAR_POS ((unsigned char *) d) <= point)
4673 #if 0 /* not emacs19 */
4675 DEBUG_PRINT1 ("EXECUTING at_dot.\n");
4676 if (PTR_CHAR_POS ((unsigned char *) d) + 1 != point)
4679 #endif /* not emacs19 */
4682 DEBUG_PRINT2 ("EXECUTING syntaxspec %d.\n", mcnt);
4687 DEBUG_PRINT1 ("EXECUTING Emacs wordchar.\n");
4691 /* Can't use *d++ here; SYNTAX may be an unsafe macro. */
4693 if (SYNTAX (d[-1]) != (enum syntaxcode) mcnt)
4695 SET_REGS_MATCHED ();
4699 DEBUG_PRINT2 ("EXECUTING notsyntaxspec %d.\n", mcnt);
4701 goto matchnotsyntax;
4704 DEBUG_PRINT1 ("EXECUTING Emacs notwordchar.\n");
4708 /* Can't use *d++ here; SYNTAX may be an unsafe macro. */
4710 if (SYNTAX (d[-1]) == (enum syntaxcode) mcnt)
4712 SET_REGS_MATCHED ();
4715 #else /* not emacs */
4717 DEBUG_PRINT1 ("EXECUTING non-Emacs wordchar.\n");
4719 if (!WORDCHAR_P (d))
4721 SET_REGS_MATCHED ();
4726 DEBUG_PRINT1 ("EXECUTING non-Emacs notwordchar.\n");
4730 SET_REGS_MATCHED ();
4733 #endif /* not emacs */
4738 continue; /* Successfully executed one pattern command; keep going. */
4741 /* We goto here if a matching operation fails. */
4743 if (!FAIL_STACK_EMPTY ())
4744 { /* A restart point is known. Restore to that state. */
4745 DEBUG_PRINT1 ("\nFAIL:\n");
4746 POP_FAILURE_POINT (d, p,
4747 lowest_active_reg, highest_active_reg,
4748 regstart, regend, reg_info);
4750 /* If this failure point is a dummy, try the next one. */
4754 /* If we failed to the end of the pattern, don't examine *p. */
4758 boolean is_a_jump_n = false;
4760 /* If failed to a backwards jump that's part of a repetition
4761 loop, need to pop this failure point and use the next one. */
4762 switch ((re_opcode_t) *p)
4766 case maybe_pop_jump:
4767 case pop_failure_jump:
4770 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4773 if ((is_a_jump_n && (re_opcode_t) *p1 == succeed_n)
4775 && (re_opcode_t) *p1 == on_failure_jump))
4783 if (d >= string1 && d <= end1)
4787 break; /* Matching at this starting point really fails. */
4791 goto restore_best_regs;
4795 return -1; /* Failure to match. */
4798 /* Subroutine definitions for re_match_2. */
4801 /* We are passed P pointing to a register number after a start_memory.
4803 Return true if the pattern up to the corresponding stop_memory can
4804 match the empty string, and false otherwise.
4806 If we find the matching stop_memory, sets P to point to one past its number.
4807 Otherwise, sets P to an undefined byte less than or equal to END.
4809 We don't handle duplicates properly (yet). */
4812 group_match_null_string_p (p, end, reg_info)
4813 unsigned char **p, *end;
4814 register_info_type *reg_info;
4817 /* Point to after the args to the start_memory. */
4818 unsigned char *p1 = *p + 2;
4822 /* Skip over opcodes that can match nothing, and return true or
4823 false, as appropriate, when we get to one that can't, or to the
4824 matching stop_memory. */
4826 switch ((re_opcode_t) *p1)
4828 /* Could be either a loop or a series of alternatives. */
4829 case on_failure_jump:
4831 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4833 /* If the next operation is not a jump backwards in the
4838 /* Go through the on_failure_jumps of the alternatives,
4839 seeing if any of the alternatives cannot match nothing.
4840 The last alternative starts with only a jump,
4841 whereas the rest start with on_failure_jump and end
4842 with a jump, e.g., here is the pattern for `a|b|c':
4844 /on_failure_jump/0/6/exactn/1/a/jump_past_alt/0/6
4845 /on_failure_jump/0/6/exactn/1/b/jump_past_alt/0/3
4848 So, we have to first go through the first (n-1)
4849 alternatives and then deal with the last one separately. */
4852 /* Deal with the first (n-1) alternatives, which start
4853 with an on_failure_jump (see above) that jumps to right
4854 past a jump_past_alt. */
4856 while ((re_opcode_t) p1[mcnt-3] == jump_past_alt)
4858 /* `mcnt' holds how many bytes long the alternative
4859 is, including the ending `jump_past_alt' and
4862 if (!alt_match_null_string_p (p1, p1 + mcnt - 3,
4866 /* Move to right after this alternative, including the
4870 /* Break if it's the beginning of an n-th alternative
4871 that doesn't begin with an on_failure_jump. */
4872 if ((re_opcode_t) *p1 != on_failure_jump)
4875 /* Still have to check that it's not an n-th
4876 alternative that starts with an on_failure_jump. */
4878 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4879 if ((re_opcode_t) p1[mcnt-3] != jump_past_alt)
4881 /* Get to the beginning of the n-th alternative. */
4887 /* Deal with the last alternative: go back and get number
4888 of the `jump_past_alt' just before it. `mcnt' contains
4889 the length of the alternative. */
4890 EXTRACT_NUMBER (mcnt, p1 - 2);
4892 if (!alt_match_null_string_p (p1, p1 + mcnt, reg_info))
4895 p1 += mcnt; /* Get past the n-th alternative. */
4901 assert (p1[1] == **p);
4907 if (!common_op_match_null_string_p (&p1, end, reg_info))
4910 } /* while p1 < end */
4913 } /* group_match_null_string_p */
4916 /* Similar to group_match_null_string_p, but doesn't deal with alternatives:
4917 It expects P to be the first byte of a single alternative and END one
4918 byte past the last. The alternative can contain groups. */
4921 alt_match_null_string_p (p, end, reg_info)
4922 unsigned char *p, *end;
4923 register_info_type *reg_info;
4926 unsigned char *p1 = p;
4930 /* Skip over opcodes that can match nothing, and break when we get
4931 to one that can't. */
4933 switch ((re_opcode_t) *p1)
4936 case on_failure_jump:
4938 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4943 if (!common_op_match_null_string_p (&p1, end, reg_info))
4946 } /* while p1 < end */
4949 } /* alt_match_null_string_p */
4952 /* Deals with the ops common to group_match_null_string_p and
4953 alt_match_null_string_p.
4955 Sets P to one after the op and its arguments, if any. */
4958 common_op_match_null_string_p (p, end, reg_info)
4959 unsigned char **p, *end;
4960 register_info_type *reg_info;
4965 unsigned char *p1 = *p;
4967 switch ((re_opcode_t) *p1++)
4987 assert (reg_no > 0 && reg_no <= MAX_REGNUM);
4988 ret = group_match_null_string_p (&p1, end, reg_info);
4990 /* Have to set this here in case we're checking a group which
4991 contains a group and a back reference to it. */
4993 if (REG_MATCH_NULL_STRING_P (reg_info[reg_no]) == MATCH_NULL_UNSET_VALUE)
4994 REG_MATCH_NULL_STRING_P (reg_info[reg_no]) = ret;
5000 /* If this is an optimized succeed_n for zero times, make the jump. */
5002 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
5010 /* Get to the number of times to succeed. */
5012 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
5017 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
5025 if (!REG_MATCH_NULL_STRING_P (reg_info[*p1]))
5033 /* All other opcodes mean we cannot match the empty string. */
5039 } /* common_op_match_null_string_p */
5042 /* Return zero if TRANSLATE[S1] and TRANSLATE[S2] are identical for LEN
5043 bytes; nonzero otherwise. */
5046 bcmp_translate (s1, s2, len, translate)
5047 unsigned char *s1, *s2;
5051 register unsigned char *p1 = s1, *p2 = s2;
5054 if (translate[*p1++] != translate[*p2++]) return 1;
5060 /* Entry points for GNU code. */
5062 /* re_compile_pattern is the GNU regular expression compiler: it
5063 compiles PATTERN (of length SIZE) and puts the result in BUFP.
5064 Returns 0 if the pattern was valid, otherwise an error string.
5066 Assumes the `allocated' (and perhaps `buffer') and `translate' fields
5067 are set in BUFP on entry.
5069 We call regex_compile to do the actual compilation. */
5072 re_compile_pattern (pattern, length, bufp)
5073 const char *pattern;
5075 struct re_pattern_buffer *bufp;
5079 /* GNU code is written to assume at least RE_NREGS registers will be set
5080 (and at least one extra will be -1). */
5081 bufp->regs_allocated = REGS_UNALLOCATED;
5083 /* And GNU code determines whether or not to get register information
5084 by passing null for the REGS argument to re_match, etc., not by
5088 /* Match anchors at newline. */
5089 bufp->newline_anchor = 1;
5091 ret = regex_compile (pattern, length, re_syntax_options, bufp);
5095 return gettext (re_error_msgid[(int) ret]);
5098 /* Entry points compatible with 4.2 BSD regex library. We don't define
5099 them unless specifically requested. */
5101 #ifdef _REGEX_RE_COMP
5103 /* BSD has one and only one pattern buffer. */
5104 static struct re_pattern_buffer re_comp_buf;
5114 if (!re_comp_buf.buffer)
5115 return gettext ("No previous regular expression");
5119 if (!re_comp_buf.buffer)
5121 re_comp_buf.buffer = (unsigned char *) malloc (200);
5122 if (re_comp_buf.buffer == NULL)
5123 return gettext (re_error_msgid[(int) REG_ESPACE]);
5124 re_comp_buf.allocated = 200;
5126 re_comp_buf.fastmap = (char *) malloc (1 << BYTEWIDTH);
5127 if (re_comp_buf.fastmap == NULL)
5128 return gettext (re_error_msgid[(int) REG_ESPACE]);
5131 /* Since `re_exec' always passes NULL for the `regs' argument, we
5132 don't need to initialize the pattern buffer fields which affect it. */
5134 /* Match anchors at newlines. */
5135 re_comp_buf.newline_anchor = 1;
5137 ret = regex_compile (s, strlen (s), re_syntax_options, &re_comp_buf);
5142 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
5143 return (char *) gettext (re_error_msgid[(int) ret]);
5151 const int len = strlen (s);
5153 0 <= re_search (&re_comp_buf, s, len, 0, len, (struct re_registers *) 0);
5155 #endif /* _REGEX_RE_COMP */
5157 /* POSIX.2 functions. Don't define these for Emacs. */
5161 /* regcomp takes a regular expression as a string and compiles it.
5163 PREG is a regex_t *. We do not expect any fields to be initialized,
5164 since POSIX says we shouldn't. Thus, we set
5166 `buffer' to the compiled pattern;
5167 `used' to the length of the compiled pattern;
5168 `syntax' to RE_SYNTAX_POSIX_EXTENDED if the
5169 REG_EXTENDED bit in CFLAGS is set; otherwise, to
5170 RE_SYNTAX_POSIX_BASIC;
5171 `newline_anchor' to REG_NEWLINE being set in CFLAGS;
5172 `fastmap' and `fastmap_accurate' to zero;
5173 `re_nsub' to the number of subexpressions in PATTERN.
5175 PATTERN is the address of the pattern string.
5177 CFLAGS is a series of bits which affect compilation.
5179 If REG_EXTENDED is set, we use POSIX extended syntax; otherwise, we
5180 use POSIX basic syntax.
5182 If REG_NEWLINE is set, then . and [^...] don't match newline.
5183 Also, regexec will try a match beginning after every newline.
5185 If REG_ICASE is set, then we considers upper- and lowercase
5186 versions of letters to be equivalent when matching.
5188 If REG_NOSUB is set, then when PREG is passed to regexec, that
5189 routine will report only success or failure, and nothing about the
5192 It returns 0 if it succeeds, nonzero if it doesn't. (See regex.h for
5193 the return codes and their meanings.) */
5196 regcomp (preg, pattern, cflags)
5198 const char *pattern;
5203 = (cflags & REG_EXTENDED) ?
5204 RE_SYNTAX_POSIX_EXTENDED : RE_SYNTAX_POSIX_BASIC;
5206 /* regex_compile will allocate the space for the compiled pattern. */
5208 preg->allocated = 0;
5211 /* Don't bother to use a fastmap when searching. This simplifies the
5212 REG_NEWLINE case: if we used a fastmap, we'd have to put all the
5213 characters after newlines into the fastmap. This way, we just try
5217 if (cflags & REG_ICASE)
5221 preg->translate = (char *) malloc (CHAR_SET_SIZE);
5222 if (preg->translate == NULL)
5223 return (int) REG_ESPACE;
5225 /* Map uppercase characters to corresponding lowercase ones. */
5226 for (i = 0; i < CHAR_SET_SIZE; i++)
5227 preg->translate[i] = ISUPPER (i) ? tolower (i) : i;
5230 preg->translate = NULL;
5232 /* If REG_NEWLINE is set, newlines are treated differently. */
5233 if (cflags & REG_NEWLINE)
5234 { /* REG_NEWLINE implies neither . nor [^...] match newline. */
5235 syntax &= ~RE_DOT_NEWLINE;
5236 syntax |= RE_HAT_LISTS_NOT_NEWLINE;
5237 /* It also changes the matching behavior. */
5238 preg->newline_anchor = 1;
5241 preg->newline_anchor = 0;
5243 preg->no_sub = !!(cflags & REG_NOSUB);
5245 /* POSIX says a null character in the pattern terminates it, so we
5246 can use strlen here in compiling the pattern. */
5247 ret = regex_compile (pattern, strlen (pattern), syntax, preg);
5249 /* POSIX doesn't distinguish between an unmatched open-group and an
5250 unmatched close-group: both are REG_EPAREN. */
5251 if (ret == REG_ERPAREN) ret = REG_EPAREN;
5257 /* regexec searches for a given pattern, specified by PREG, in the
5260 If NMATCH is zero or REG_NOSUB was set in the cflags argument to
5261 `regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at
5262 least NMATCH elements, and we set them to the offsets of the
5263 corresponding matched substrings.
5265 EFLAGS specifies `execution flags' which affect matching: if
5266 REG_NOTBOL is set, then ^ does not match at the beginning of the
5267 string; if REG_NOTEOL is set, then $ does not match at the end.
5269 We return 0 if we find a match and REG_NOMATCH if not. */
5272 regexec (preg, string, nmatch, pmatch, eflags)
5273 const regex_t *preg;
5276 regmatch_t pmatch[];
5280 struct re_registers regs;
5281 regex_t private_preg;
5282 int len = strlen (string);
5283 boolean want_reg_info = !preg->no_sub && nmatch > 0;
5285 private_preg = *preg;
5287 private_preg.not_bol = !!(eflags & REG_NOTBOL);
5288 private_preg.not_eol = !!(eflags & REG_NOTEOL);
5290 /* The user has told us exactly how many registers to return
5291 information about, via `nmatch'. We have to pass that on to the
5292 matching routines. */
5293 private_preg.regs_allocated = REGS_FIXED;
5297 regs.num_regs = nmatch;
5298 regs.start = TALLOC (nmatch, regoff_t);
5299 regs.end = TALLOC (nmatch, regoff_t);
5300 if (regs.start == NULL || regs.end == NULL)
5301 return (int) REG_NOMATCH;
5304 /* Perform the searching operation. */
5305 ret = re_search (&private_preg, string, len,
5306 /* start: */ 0, /* range: */ len,
5307 want_reg_info ? ®s : (struct re_registers *) 0);
5309 /* Copy the register information to the POSIX structure. */
5316 for (r = 0; r < nmatch; r++)
5318 pmatch[r].rm_so = regs.start[r];
5319 pmatch[r].rm_eo = regs.end[r];
5323 /* If we needed the temporary register info, free the space now. */
5328 /* We want zero return to mean success, unlike `re_search'. */
5329 return ret >= 0 ? (int) REG_NOERROR : (int) REG_NOMATCH;
5333 /* Returns a message corresponding to an error code, ERRCODE, returned
5334 from either regcomp or regexec. We don't use PREG here. */
5337 regerror (errcode, preg, errbuf, errbuf_size)
5339 const regex_t *preg;
5347 || errcode >= (sizeof (re_error_msgid) / sizeof (re_error_msgid[0])))
5348 /* Only error codes returned by the rest of the code should be passed
5349 to this routine. If we are given anything else, or if other regex
5350 code generates an invalid error code, then the program has a bug.
5351 Dump core so we can fix it. */
5354 msg = gettext (re_error_msgid[errcode]);
5356 msg_size = strlen (msg) + 1; /* Includes the null. */
5358 if (errbuf_size != 0)
5360 if (msg_size > errbuf_size)
5362 strncpy (errbuf, msg, errbuf_size - 1);
5363 errbuf[errbuf_size - 1] = 0;
5366 strcpy (errbuf, msg);
5373 /* Free dynamically allocated space used by PREG. */
5379 if (preg->buffer != NULL)
5380 free (preg->buffer);
5381 preg->buffer = NULL;
5383 preg->allocated = 0;
5386 if (preg->fastmap != NULL)
5387 free (preg->fastmap);
5388 preg->fastmap = NULL;
5389 preg->fastmap_accurate = 0;
5391 if (preg->translate != NULL)
5392 free (preg->translate);
5393 preg->translate = NULL;
5396 #endif /* not emacs */
5400 make-backup-files: t
5402 trim-versions-without-asking: nil