1 /* Extended regular expression matching and search library,
3 (Implements POSIX draft P10003.2/D11.2, except for
4 internationalization features.)
6 Copyright (C) 1993 Free Software Foundation, Inc.
8 This program is free software; you can redistribute it and/or modify
9 it under the terms of the GNU General Public License as published by
10 the Free Software Foundation; either version 2, or (at your option)
13 This program is distributed in the hope that it will be useful,
14 but WITHOUT ANY WARRANTY; without even the implied warranty of
15 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
16 GNU General Public License for more details.
18 You should have received a copy of the GNU General Public License
19 along with this program; if not, write to the Free Software
20 Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. */
22 /* AIX requires this to be the first thing in the file. */
23 #if defined (_AIX) && !defined (REGEX_MALLOC)
30 #if defined (CONFIG_BROKETS)
31 /* We use <config.h> instead of "config.h" so that a compilation
32 using -I. -I$srcdir will use ./config.h rather than $srcdir/config.h
33 (which it would do because it found this file in $srcdir). */
40 /* We need this for `regex.h', and perhaps for the Emacs include files. */
41 #include <sys/types.h>
47 /* The `emacs' switch turns on certain matching commands
48 that make sense only in Emacs. */
55 /* Emacs uses `NULL' as a predicate. */
68 /* We used to test for `BSTRING' here, but only GCC and Emacs define
69 `BSTRING', as far as I know, and neither of them use this code. */
70 #if HAVE_STRING_H || STDC_HEADERS
73 #define bcmp(s1, s2, n) memcmp ((s1), (s2), (n))
76 #define bcopy(s, d, n) memcpy ((d), (s), (n))
79 #define bzero(s, n) memset ((s), 0, (n))
85 /* Define the syntax stuff for \<, \>, etc. */
87 /* This must be nonzero for the wordchar and notwordchar pattern
88 commands in re_match_2. */
95 extern char *re_syntax_table;
97 #else /* not SYNTAX_TABLE */
99 /* How many characters in the character set. */
100 #define CHAR_SET_SIZE 256
102 static char re_syntax_table[CHAR_SET_SIZE];
113 bzero (re_syntax_table, sizeof re_syntax_table);
115 for (c = 'a'; c <= 'z'; c++)
116 re_syntax_table[c] = Sword;
118 for (c = 'A'; c <= 'Z'; c++)
119 re_syntax_table[c] = Sword;
121 for (c = '0'; c <= '9'; c++)
122 re_syntax_table[c] = Sword;
124 re_syntax_table['_'] = Sword;
129 #endif /* not SYNTAX_TABLE */
131 #define SYNTAX(c) re_syntax_table[c]
133 #endif /* not emacs */
135 /* Get the interface, including the syntax bits. */
138 /* isalpha etc. are used for the character classes. */
141 /* Jim Meyering writes:
143 "... Some ctype macros are valid only for character codes that
144 isascii says are ASCII (SGI's IRIX-4.0.5 is one such system --when
145 using /bin/cc or gcc but without giving an ansi option). So, all
146 ctype uses should be through macros like ISPRINT... If
147 STDC_HEADERS is defined, then autoconf has verified that the ctype
148 macros don't need to be guarded with references to isascii. ...
149 Defining isascii to 1 should let any compiler worth its salt
150 eliminate the && through constant folding." */
152 #if defined (STDC_HEADERS) || (!defined (isascii) && !defined (HAVE_ISASCII))
155 #define ISASCII(c) isascii(c)
159 #define ISBLANK(c) (ISASCII (c) && isblank (c))
161 #define ISBLANK(c) ((c) == ' ' || (c) == '\t')
164 #define ISGRAPH(c) (ISASCII (c) && isgraph (c))
166 #define ISGRAPH(c) (ISASCII (c) && isprint (c) && !isspace (c))
169 #define ISPRINT(c) (ISASCII (c) && isprint (c))
170 #define ISDIGIT(c) (ISASCII (c) && isdigit (c))
171 #define ISALNUM(c) (ISASCII (c) && isalnum (c))
172 #define ISALPHA(c) (ISASCII (c) && isalpha (c))
173 #define ISCNTRL(c) (ISASCII (c) && iscntrl (c))
174 #define ISLOWER(c) (ISASCII (c) && islower (c))
175 #define ISPUNCT(c) (ISASCII (c) && ispunct (c))
176 #define ISSPACE(c) (ISASCII (c) && isspace (c))
177 #define ISUPPER(c) (ISASCII (c) && isupper (c))
178 #define ISXDIGIT(c) (ISASCII (c) && isxdigit (c))
184 /* We remove any previous definition of `SIGN_EXTEND_CHAR',
185 since ours (we hope) works properly with all combinations of
186 machines, compilers, `char' and `unsigned char' argument types.
187 (Per Bothner suggested the basic approach.) */
188 #undef SIGN_EXTEND_CHAR
190 #define SIGN_EXTEND_CHAR(c) ((signed char) (c))
191 #else /* not __STDC__ */
192 /* As in Harbison and Steele. */
193 #define SIGN_EXTEND_CHAR(c) ((((unsigned char) (c)) ^ 128) - 128)
196 /* Should we use malloc or alloca? If REGEX_MALLOC is not defined, we
197 use `alloca' instead of `malloc'. This is because using malloc in
198 re_search* or re_match* could cause memory leaks when C-g is used in
199 Emacs; also, malloc is slower and causes storage fragmentation. On
200 the other hand, malloc is more portable, and easier to debug.
202 Because we sometimes use alloca, some routines have to be macros,
203 not functions -- `alloca'-allocated space disappears at the end of the
204 function it is called in. */
208 #define REGEX_ALLOCATE malloc
209 #define REGEX_REALLOCATE(source, osize, nsize) realloc (source, nsize)
211 #else /* not REGEX_MALLOC */
213 /* Emacs already defines alloca, sometimes. */
216 /* Make alloca work the best possible way. */
218 #define alloca __builtin_alloca
219 #else /* not __GNUC__ */
222 #else /* not __GNUC__ or HAVE_ALLOCA_H */
223 #ifndef _AIX /* Already did AIX, up at the top. */
225 #endif /* not _AIX */
226 #endif /* not HAVE_ALLOCA_H */
227 #endif /* not __GNUC__ */
229 #endif /* not alloca */
231 #define REGEX_ALLOCATE alloca
233 /* Assumes a `char *destination' variable. */
234 #define REGEX_REALLOCATE(source, osize, nsize) \
235 (destination = (char *) alloca (nsize), \
236 bcopy (source, destination, osize), \
239 #endif /* not REGEX_MALLOC */
242 /* True if `size1' is non-NULL and PTR is pointing anywhere inside
243 `string1' or just past its end. This works if PTR is NULL, which is
245 #define FIRST_STRING_P(ptr) \
246 (size1 && string1 <= (ptr) && (ptr) <= string1 + size1)
248 /* (Re)Allocate N items of type T using malloc, or fail. */
249 #define TALLOC(n, t) ((t *) malloc ((n) * sizeof (t)))
250 #define RETALLOC(addr, n, t) ((addr) = (t *) realloc (addr, (n) * sizeof (t)))
251 #define RETALLOC_IF(addr, n, t) \
252 if (addr) RETALLOC((addr), (n), t); else (addr) = TALLOC ((n), t)
253 #define REGEX_TALLOC(n, t) ((t *) REGEX_ALLOCATE ((n) * sizeof (t)))
255 #define BYTEWIDTH 8 /* In bits. */
257 #define STREQ(s1, s2) ((strcmp (s1, s2) == 0))
261 #define MAX(a, b) ((a) > (b) ? (a) : (b))
262 #define MIN(a, b) ((a) < (b) ? (a) : (b))
264 typedef char boolean;
268 /* These are the command codes that appear in compiled regular
269 expressions. Some opcodes are followed by argument bytes. A
270 command code can specify any interpretation whatsoever for its
271 arguments. Zero bytes may appear in the compiled regular expression.
273 The value of `exactn' is needed in search.c (search_buffer) in Emacs.
274 So regex.h defines a symbol `RE_EXACTN_VALUE' to be 1; the value of
275 `exactn' we use here must also be 1. */
281 /* Followed by one byte giving n, then by n literal bytes. */
284 /* Matches any (more or less) character. */
287 /* Matches any one char belonging to specified set. First
288 following byte is number of bitmap bytes. Then come bytes
289 for a bitmap saying which chars are in. Bits in each byte
290 are ordered low-bit-first. A character is in the set if its
291 bit is 1. A character too large to have a bit in the map is
292 automatically not in the set. */
295 /* Same parameters as charset, but match any character that is
296 not one of those specified. */
299 /* Start remembering the text that is matched, for storing in a
300 register. Followed by one byte with the register number, in
301 the range 0 to one less than the pattern buffer's re_nsub
302 field. Then followed by one byte with the number of groups
303 inner to this one. (This last has to be part of the
304 start_memory only because we need it in the on_failure_jump
308 /* Stop remembering the text that is matched and store it in a
309 memory register. Followed by one byte with the register
310 number, in the range 0 to one less than `re_nsub' in the
311 pattern buffer, and one byte with the number of inner groups,
312 just like `start_memory'. (We need the number of inner
313 groups here because we don't have any easy way of finding the
314 corresponding start_memory when we're at a stop_memory.) */
317 /* Match a duplicate of something remembered. Followed by one
318 byte containing the register number. */
321 /* Fail unless at beginning of line. */
324 /* Fail unless at end of line. */
327 /* Succeeds if at beginning of buffer (if emacs) or at beginning
328 of string to be matched (if not). */
331 /* Analogously, for end of buffer/string. */
334 /* Followed by two byte relative address to which to jump. */
337 /* Same as jump, but marks the end of an alternative. */
340 /* Followed by two-byte relative address of place to resume at
341 in case of failure. */
344 /* Like on_failure_jump, but pushes a placeholder instead of the
345 current string position when executed. */
346 on_failure_keep_string_jump,
348 /* Throw away latest failure point and then jump to following
349 two-byte relative address. */
352 /* Change to pop_failure_jump if know won't have to backtrack to
353 match; otherwise change to jump. This is used to jump
354 back to the beginning of a repeat. If what follows this jump
355 clearly won't match what the repeat does, such that we can be
356 sure that there is no use backtracking out of repetitions
357 already matched, then we change it to a pop_failure_jump.
358 Followed by two-byte address. */
361 /* Jump to following two-byte address, and push a dummy failure
362 point. This failure point will be thrown away if an attempt
363 is made to use it for a failure. A `+' construct makes this
364 before the first repeat. Also used as an intermediary kind
365 of jump when compiling an alternative. */
368 /* Push a dummy failure point and continue. Used at the end of
372 /* Followed by two-byte relative address and two-byte number n.
373 After matching N times, jump to the address upon failure. */
376 /* Followed by two-byte relative address, and two-byte number n.
377 Jump to the address N times, then fail. */
380 /* Set the following two-byte relative address to the
381 subsequent two-byte number. The address *includes* the two
385 wordchar, /* Matches any word-constituent character. */
386 notwordchar, /* Matches any char that is not a word-constituent. */
388 wordbeg, /* Succeeds if at word beginning. */
389 wordend, /* Succeeds if at word end. */
391 wordbound, /* Succeeds if at a word boundary. */
392 notwordbound /* Succeeds if not at a word boundary. */
395 ,before_dot, /* Succeeds if before point. */
396 at_dot, /* Succeeds if at point. */
397 after_dot, /* Succeeds if after point. */
399 /* Matches any character whose syntax is specified. Followed by
400 a byte which contains a syntax code, e.g., Sword. */
403 /* Matches any character whose syntax is not that specified. */
408 /* Common operations on the compiled pattern. */
410 /* Store NUMBER in two contiguous bytes starting at DESTINATION. */
412 #define STORE_NUMBER(destination, number) \
414 (destination)[0] = (number) & 0377; \
415 (destination)[1] = (number) >> 8; \
418 /* Same as STORE_NUMBER, except increment DESTINATION to
419 the byte after where the number is stored. Therefore, DESTINATION
420 must be an lvalue. */
422 #define STORE_NUMBER_AND_INCR(destination, number) \
424 STORE_NUMBER (destination, number); \
425 (destination) += 2; \
428 /* Put into DESTINATION a number stored in two contiguous bytes starting
431 #define EXTRACT_NUMBER(destination, source) \
433 (destination) = *(source) & 0377; \
434 (destination) += SIGN_EXTEND_CHAR (*((source) + 1)) << 8; \
439 extract_number (dest, source)
441 unsigned char *source;
443 int temp = SIGN_EXTEND_CHAR (*(source + 1));
444 *dest = *source & 0377;
448 #ifndef EXTRACT_MACROS /* To debug the macros. */
449 #undef EXTRACT_NUMBER
450 #define EXTRACT_NUMBER(dest, src) extract_number (&dest, src)
451 #endif /* not EXTRACT_MACROS */
455 /* Same as EXTRACT_NUMBER, except increment SOURCE to after the number.
456 SOURCE must be an lvalue. */
458 #define EXTRACT_NUMBER_AND_INCR(destination, source) \
460 EXTRACT_NUMBER (destination, source); \
466 extract_number_and_incr (destination, source)
468 unsigned char **source;
470 extract_number (destination, *source);
474 #ifndef EXTRACT_MACROS
475 #undef EXTRACT_NUMBER_AND_INCR
476 #define EXTRACT_NUMBER_AND_INCR(dest, src) \
477 extract_number_and_incr (&dest, &src)
478 #endif /* not EXTRACT_MACROS */
482 /* If DEBUG is defined, Regex prints many voluminous messages about what
483 it is doing (if the variable `debug' is nonzero). If linked with the
484 main program in `iregex.c', you can enter patterns and strings
485 interactively. And if linked with the main program in `main.c' and
486 the other test files, you can run the already-written tests. */
490 /* We use standard I/O for debugging. */
493 /* It is useful to test things that ``must'' be true when debugging. */
496 static int debug = 0;
498 #define DEBUG_STATEMENT(e) e
499 #define DEBUG_PRINT1(x) if (debug) printf (x)
500 #define DEBUG_PRINT2(x1, x2) if (debug) printf (x1, x2)
501 #define DEBUG_PRINT3(x1, x2, x3) if (debug) printf (x1, x2, x3)
502 #define DEBUG_PRINT4(x1, x2, x3, x4) if (debug) printf (x1, x2, x3, x4)
503 #define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) \
504 if (debug) print_partial_compiled_pattern (s, e)
505 #define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) \
506 if (debug) print_double_string (w, s1, sz1, s2, sz2)
509 extern void printchar ();
511 /* Print the fastmap in human-readable form. */
514 print_fastmap (fastmap)
517 unsigned was_a_range = 0;
520 while (i < (1 << BYTEWIDTH))
526 while (i < (1 << BYTEWIDTH) && fastmap[i])
542 /* Print a compiled pattern string in human-readable form, starting at
543 the START pointer into it and ending just before the pointer END. */
546 print_partial_compiled_pattern (start, end)
547 unsigned char *start;
551 unsigned char *p = start;
552 unsigned char *pend = end;
560 /* Loop over pattern commands. */
563 printf ("%d:\t", p - start);
565 switch ((re_opcode_t) *p++)
573 printf ("/exactn/%d", mcnt);
584 printf ("/start_memory/%d/%d", mcnt, *p++);
589 printf ("/stop_memory/%d/%d", mcnt, *p++);
593 printf ("/duplicate/%d", *p++);
603 register int c, last = -100;
604 register int in_range = 0;
606 printf ("/charset [%s",
607 (re_opcode_t) *(p - 1) == charset_not ? "^" : "");
609 assert (p + *p < pend);
611 for (c = 0; c < 256; c++)
613 && (p[1 + (c/8)] & (1 << (c % 8))))
615 /* Are we starting a range? */
616 if (last + 1 == c && ! in_range)
621 /* Have we broken a range? */
622 else if (last + 1 != c && in_range)
651 case on_failure_jump:
652 extract_number_and_incr (&mcnt, &p);
653 printf ("/on_failure_jump to %d", p + mcnt - start);
656 case on_failure_keep_string_jump:
657 extract_number_and_incr (&mcnt, &p);
658 printf ("/on_failure_keep_string_jump to %d", p + mcnt - start);
661 case dummy_failure_jump:
662 extract_number_and_incr (&mcnt, &p);
663 printf ("/dummy_failure_jump to %d", p + mcnt - start);
666 case push_dummy_failure:
667 printf ("/push_dummy_failure");
671 extract_number_and_incr (&mcnt, &p);
672 printf ("/maybe_pop_jump to %d", p + mcnt - start);
675 case pop_failure_jump:
676 extract_number_and_incr (&mcnt, &p);
677 printf ("/pop_failure_jump to %d", p + mcnt - start);
681 extract_number_and_incr (&mcnt, &p);
682 printf ("/jump_past_alt to %d", p + mcnt - start);
686 extract_number_and_incr (&mcnt, &p);
687 printf ("/jump to %d", p + mcnt - start);
691 extract_number_and_incr (&mcnt, &p);
692 extract_number_and_incr (&mcnt2, &p);
693 printf ("/succeed_n to %d, %d times", p + mcnt - start, mcnt2);
697 extract_number_and_incr (&mcnt, &p);
698 extract_number_and_incr (&mcnt2, &p);
699 printf ("/jump_n to %d, %d times", p + mcnt - start, mcnt2);
703 extract_number_and_incr (&mcnt, &p);
704 extract_number_and_incr (&mcnt2, &p);
705 printf ("/set_number_at location %d to %d", p + mcnt - start, mcnt2);
709 printf ("/wordbound");
713 printf ("/notwordbound");
725 printf ("/before_dot");
733 printf ("/after_dot");
737 printf ("/syntaxspec");
739 printf ("/%d", mcnt);
743 printf ("/notsyntaxspec");
745 printf ("/%d", mcnt);
750 printf ("/wordchar");
754 printf ("/notwordchar");
766 printf ("?%d", *(p-1));
772 printf ("%d:\tend of pattern.\n", p - start);
777 print_compiled_pattern (bufp)
778 struct re_pattern_buffer *bufp;
780 unsigned char *buffer = bufp->buffer;
782 print_partial_compiled_pattern (buffer, buffer + bufp->used);
783 printf ("%d bytes used/%d bytes allocated.\n", bufp->used, bufp->allocated);
785 if (bufp->fastmap_accurate && bufp->fastmap)
787 printf ("fastmap: ");
788 print_fastmap (bufp->fastmap);
791 printf ("re_nsub: %d\t", bufp->re_nsub);
792 printf ("regs_alloc: %d\t", bufp->regs_allocated);
793 printf ("can_be_null: %d\t", bufp->can_be_null);
794 printf ("newline_anchor: %d\n", bufp->newline_anchor);
795 printf ("no_sub: %d\t", bufp->no_sub);
796 printf ("not_bol: %d\t", bufp->not_bol);
797 printf ("not_eol: %d\t", bufp->not_eol);
798 printf ("syntax: %d\n", bufp->syntax);
799 /* Perhaps we should print the translate table? */
804 print_double_string (where, string1, size1, string2, size2)
817 if (FIRST_STRING_P (where))
819 for (this_char = where - string1; this_char < size1; this_char++)
820 printchar (string1[this_char]);
825 for (this_char = where - string2; this_char < size2; this_char++)
826 printchar (string2[this_char]);
830 #else /* not DEBUG */
835 #define DEBUG_STATEMENT(e)
836 #define DEBUG_PRINT1(x)
837 #define DEBUG_PRINT2(x1, x2)
838 #define DEBUG_PRINT3(x1, x2, x3)
839 #define DEBUG_PRINT4(x1, x2, x3, x4)
840 #define DEBUG_PRINT_COMPILED_PATTERN(p, s, e)
841 #define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2)
843 #endif /* not DEBUG */
845 /* Set by `re_set_syntax' to the current regexp syntax to recognize. Can
846 also be assigned to arbitrarily: each pattern buffer stores its own
847 syntax, so it can be changed between regex compilations. */
848 reg_syntax_t re_syntax_options = RE_SYNTAX_EMACS;
851 /* Specify the precise syntax of regexps for compilation. This provides
852 for compatibility for various utilities which historically have
853 different, incompatible syntaxes.
855 The argument SYNTAX is a bit mask comprised of the various bits
856 defined in regex.h. We return the old syntax. */
859 re_set_syntax (syntax)
862 reg_syntax_t ret = re_syntax_options;
864 re_syntax_options = syntax;
868 /* This table gives an error message for each of the error codes listed
869 in regex.h. Obviously the order here has to be same as there. */
871 static const char *re_error_msg[] =
872 { NULL, /* REG_NOERROR */
873 "No match", /* REG_NOMATCH */
874 "Invalid regular expression", /* REG_BADPAT */
875 "Invalid collation character", /* REG_ECOLLATE */
876 "Invalid character class name", /* REG_ECTYPE */
877 "Trailing backslash", /* REG_EESCAPE */
878 "Invalid back reference", /* REG_ESUBREG */
879 "Unmatched [ or [^", /* REG_EBRACK */
880 "Unmatched ( or \\(", /* REG_EPAREN */
881 "Unmatched \\{", /* REG_EBRACE */
882 "Invalid content of \\{\\}", /* REG_BADBR */
883 "Invalid range end", /* REG_ERANGE */
884 "Memory exhausted", /* REG_ESPACE */
885 "Invalid preceding regular expression", /* REG_BADRPT */
886 "Premature end of regular expression", /* REG_EEND */
887 "Regular expression too big", /* REG_ESIZE */
888 "Unmatched ) or \\)", /* REG_ERPAREN */
891 /* Avoiding alloca during matching, to placate r_alloc. */
893 /* Define MATCH_MAY_ALLOCATE if we need to make sure that the
894 searching and matching functions should not call alloca. On some
895 systems, alloca is implemented in terms of malloc, and if we're
896 using the relocating allocator routines, then malloc could cause a
897 relocation, which might (if the strings being searched are in the
898 ralloc heap) shift the data out from underneath the regexp
901 Here's another reason to avoid allocation: Emacs insists on
902 processing input from X in a signal handler; processing X input may
903 call malloc; if input arrives while a matching routine is calling
904 malloc, then we're scrod. But Emacs can't just block input while
905 calling matching routines; then we don't notice interrupts when
906 they come in. So, Emacs blocks input around all regexp calls
907 except the matching calls, which it leaves unprotected, in the
908 faith that they will not malloc. */
910 /* Normally, this is fine. */
911 #define MATCH_MAY_ALLOCATE
913 /* But under some circumstances, it's not. */
914 #if defined (emacs) || (defined (REL_ALLOC) && defined (C_ALLOCA))
915 #undef MATCH_MAY_ALLOCATE
919 /* Failure stack declarations and macros; both re_compile_fastmap and
920 re_match_2 use a failure stack. These have to be macros because of
924 /* Number of failure points for which to initially allocate space
925 when matching. If this number is exceeded, we allocate more
926 space, so it is not a hard limit. */
927 #ifndef INIT_FAILURE_ALLOC
928 #define INIT_FAILURE_ALLOC 5
931 /* Roughly the maximum number of failure points on the stack. Would be
932 exactly that if always used MAX_FAILURE_SPACE each time we failed.
933 This is a variable only so users of regex can assign to it; we never
934 change it ourselves. */
935 int re_max_failures = 2000;
937 typedef unsigned char *fail_stack_elt_t;
941 fail_stack_elt_t *stack;
943 unsigned avail; /* Offset of next open position. */
946 #define FAIL_STACK_EMPTY() (fail_stack.avail == 0)
947 #define FAIL_STACK_PTR_EMPTY() (fail_stack_ptr->avail == 0)
948 #define FAIL_STACK_FULL() (fail_stack.avail == fail_stack.size)
949 #define FAIL_STACK_TOP() (fail_stack.stack[fail_stack.avail])
952 /* Initialize `fail_stack'. Do `return -2' if the alloc fails. */
954 #ifdef MATCH_MAY_ALLOCATE
955 #define INIT_FAIL_STACK() \
957 fail_stack.stack = (fail_stack_elt_t *) \
958 REGEX_ALLOCATE (INIT_FAILURE_ALLOC * sizeof (fail_stack_elt_t)); \
960 if (fail_stack.stack == NULL) \
963 fail_stack.size = INIT_FAILURE_ALLOC; \
964 fail_stack.avail = 0; \
967 #define INIT_FAIL_STACK() \
969 fail_stack.avail = 0; \
974 /* Double the size of FAIL_STACK, up to approximately `re_max_failures' items.
976 Return 1 if succeeds, and 0 if either ran out of memory
977 allocating space for it or it was already too large.
979 REGEX_REALLOCATE requires `destination' be declared. */
981 #define DOUBLE_FAIL_STACK(fail_stack) \
982 ((fail_stack).size > re_max_failures * MAX_FAILURE_ITEMS \
984 : ((fail_stack).stack = (fail_stack_elt_t *) \
985 REGEX_REALLOCATE ((fail_stack).stack, \
986 (fail_stack).size * sizeof (fail_stack_elt_t), \
987 ((fail_stack).size << 1) * sizeof (fail_stack_elt_t)), \
989 (fail_stack).stack == NULL \
991 : ((fail_stack).size <<= 1, \
995 /* Push PATTERN_OP on FAIL_STACK.
997 Return 1 if was able to do so and 0 if ran out of memory allocating
999 #define PUSH_PATTERN_OP(pattern_op, fail_stack) \
1000 ((FAIL_STACK_FULL () \
1001 && !DOUBLE_FAIL_STACK (fail_stack)) \
1003 : ((fail_stack).stack[(fail_stack).avail++] = pattern_op, \
1006 /* This pushes an item onto the failure stack. Must be a four-byte
1007 value. Assumes the variable `fail_stack'. Probably should only
1008 be called from within `PUSH_FAILURE_POINT'. */
1009 #define PUSH_FAILURE_ITEM(item) \
1010 fail_stack.stack[fail_stack.avail++] = (fail_stack_elt_t) item
1012 /* The complement operation. Assumes `fail_stack' is nonempty. */
1013 #define POP_FAILURE_ITEM() fail_stack.stack[--fail_stack.avail]
1015 /* Used to omit pushing failure point id's when we're not debugging. */
1017 #define DEBUG_PUSH PUSH_FAILURE_ITEM
1018 #define DEBUG_POP(item_addr) *(item_addr) = POP_FAILURE_ITEM ()
1020 #define DEBUG_PUSH(item)
1021 #define DEBUG_POP(item_addr)
1025 /* Push the information about the state we will need
1026 if we ever fail back to it.
1028 Requires variables fail_stack, regstart, regend, reg_info, and
1029 num_regs be declared. DOUBLE_FAIL_STACK requires `destination' be
1032 Does `return FAILURE_CODE' if runs out of memory. */
1034 #define PUSH_FAILURE_POINT(pattern_place, string_place, failure_code) \
1036 char *destination; \
1037 /* Must be int, so when we don't save any registers, the arithmetic \
1038 of 0 + -1 isn't done as unsigned. */ \
1041 DEBUG_STATEMENT (failure_id++); \
1042 DEBUG_STATEMENT (nfailure_points_pushed++); \
1043 DEBUG_PRINT2 ("\nPUSH_FAILURE_POINT #%u:\n", failure_id); \
1044 DEBUG_PRINT2 (" Before push, next avail: %d\n", (fail_stack).avail);\
1045 DEBUG_PRINT2 (" size: %d\n", (fail_stack).size);\
1047 DEBUG_PRINT2 (" slots needed: %d\n", NUM_FAILURE_ITEMS); \
1048 DEBUG_PRINT2 (" available: %d\n", REMAINING_AVAIL_SLOTS); \
1050 /* Ensure we have enough space allocated for what we will push. */ \
1051 while (REMAINING_AVAIL_SLOTS < NUM_FAILURE_ITEMS) \
1053 if (!DOUBLE_FAIL_STACK (fail_stack)) \
1054 return failure_code; \
1056 DEBUG_PRINT2 ("\n Doubled stack; size now: %d\n", \
1057 (fail_stack).size); \
1058 DEBUG_PRINT2 (" slots available: %d\n", REMAINING_AVAIL_SLOTS);\
1061 /* Push the info, starting with the registers. */ \
1062 DEBUG_PRINT1 ("\n"); \
1064 for (this_reg = lowest_active_reg; this_reg <= highest_active_reg; \
1067 DEBUG_PRINT2 (" Pushing reg: %d\n", this_reg); \
1068 DEBUG_STATEMENT (num_regs_pushed++); \
1070 DEBUG_PRINT2 (" start: 0x%x\n", regstart[this_reg]); \
1071 PUSH_FAILURE_ITEM (regstart[this_reg]); \
1073 DEBUG_PRINT2 (" end: 0x%x\n", regend[this_reg]); \
1074 PUSH_FAILURE_ITEM (regend[this_reg]); \
1076 DEBUG_PRINT2 (" info: 0x%x\n ", reg_info[this_reg]); \
1077 DEBUG_PRINT2 (" match_null=%d", \
1078 REG_MATCH_NULL_STRING_P (reg_info[this_reg])); \
1079 DEBUG_PRINT2 (" active=%d", IS_ACTIVE (reg_info[this_reg])); \
1080 DEBUG_PRINT2 (" matched_something=%d", \
1081 MATCHED_SOMETHING (reg_info[this_reg])); \
1082 DEBUG_PRINT2 (" ever_matched=%d", \
1083 EVER_MATCHED_SOMETHING (reg_info[this_reg])); \
1084 DEBUG_PRINT1 ("\n"); \
1085 PUSH_FAILURE_ITEM (reg_info[this_reg].word); \
1088 DEBUG_PRINT2 (" Pushing low active reg: %d\n", lowest_active_reg);\
1089 PUSH_FAILURE_ITEM (lowest_active_reg); \
1091 DEBUG_PRINT2 (" Pushing high active reg: %d\n", highest_active_reg);\
1092 PUSH_FAILURE_ITEM (highest_active_reg); \
1094 DEBUG_PRINT2 (" Pushing pattern 0x%x: ", pattern_place); \
1095 DEBUG_PRINT_COMPILED_PATTERN (bufp, pattern_place, pend); \
1096 PUSH_FAILURE_ITEM (pattern_place); \
1098 DEBUG_PRINT2 (" Pushing string 0x%x: `", string_place); \
1099 DEBUG_PRINT_DOUBLE_STRING (string_place, string1, size1, string2, \
1101 DEBUG_PRINT1 ("'\n"); \
1102 PUSH_FAILURE_ITEM (string_place); \
1104 DEBUG_PRINT2 (" Pushing failure id: %u\n", failure_id); \
1105 DEBUG_PUSH (failure_id); \
1108 /* This is the number of items that are pushed and popped on the stack
1109 for each register. */
1110 #define NUM_REG_ITEMS 3
1112 /* Individual items aside from the registers. */
1114 #define NUM_NONREG_ITEMS 5 /* Includes failure point id. */
1116 #define NUM_NONREG_ITEMS 4
1119 /* We push at most this many items on the stack. */
1120 #define MAX_FAILURE_ITEMS ((num_regs - 1) * NUM_REG_ITEMS + NUM_NONREG_ITEMS)
1122 /* We actually push this many items. */
1123 #define NUM_FAILURE_ITEMS \
1124 ((highest_active_reg - lowest_active_reg + 1) * NUM_REG_ITEMS \
1127 /* How many items can still be added to the stack without overflowing it. */
1128 #define REMAINING_AVAIL_SLOTS ((fail_stack).size - (fail_stack).avail)
1131 /* Pops what PUSH_FAIL_STACK pushes.
1133 We restore into the parameters, all of which should be lvalues:
1134 STR -- the saved data position.
1135 PAT -- the saved pattern position.
1136 LOW_REG, HIGH_REG -- the highest and lowest active registers.
1137 REGSTART, REGEND -- arrays of string positions.
1138 REG_INFO -- array of information about each subexpression.
1140 Also assumes the variables `fail_stack' and (if debugging), `bufp',
1141 `pend', `string1', `size1', `string2', and `size2'. */
1143 #define POP_FAILURE_POINT(str, pat, low_reg, high_reg, regstart, regend, reg_info)\
1145 DEBUG_STATEMENT (fail_stack_elt_t failure_id;) \
1147 const unsigned char *string_temp; \
1149 assert (!FAIL_STACK_EMPTY ()); \
1151 /* Remove failure points and point to how many regs pushed. */ \
1152 DEBUG_PRINT1 ("POP_FAILURE_POINT:\n"); \
1153 DEBUG_PRINT2 (" Before pop, next avail: %d\n", fail_stack.avail); \
1154 DEBUG_PRINT2 (" size: %d\n", fail_stack.size); \
1156 assert (fail_stack.avail >= NUM_NONREG_ITEMS); \
1158 DEBUG_POP (&failure_id); \
1159 DEBUG_PRINT2 (" Popping failure id: %u\n", failure_id); \
1161 /* If the saved string location is NULL, it came from an \
1162 on_failure_keep_string_jump opcode, and we want to throw away the \
1163 saved NULL, thus retaining our current position in the string. */ \
1164 string_temp = POP_FAILURE_ITEM (); \
1165 if (string_temp != NULL) \
1166 str = (const char *) string_temp; \
1168 DEBUG_PRINT2 (" Popping string 0x%x: `", str); \
1169 DEBUG_PRINT_DOUBLE_STRING (str, string1, size1, string2, size2); \
1170 DEBUG_PRINT1 ("'\n"); \
1172 pat = (unsigned char *) POP_FAILURE_ITEM (); \
1173 DEBUG_PRINT2 (" Popping pattern 0x%x: ", pat); \
1174 DEBUG_PRINT_COMPILED_PATTERN (bufp, pat, pend); \
1176 /* Restore register info. */ \
1177 high_reg = (unsigned) POP_FAILURE_ITEM (); \
1178 DEBUG_PRINT2 (" Popping high active reg: %d\n", high_reg); \
1180 low_reg = (unsigned) POP_FAILURE_ITEM (); \
1181 DEBUG_PRINT2 (" Popping low active reg: %d\n", low_reg); \
1183 for (this_reg = high_reg; this_reg >= low_reg; this_reg--) \
1185 DEBUG_PRINT2 (" Popping reg: %d\n", this_reg); \
1187 reg_info[this_reg].word = POP_FAILURE_ITEM (); \
1188 DEBUG_PRINT2 (" info: 0x%x\n", reg_info[this_reg]); \
1190 regend[this_reg] = (const char *) POP_FAILURE_ITEM (); \
1191 DEBUG_PRINT2 (" end: 0x%x\n", regend[this_reg]); \
1193 regstart[this_reg] = (const char *) POP_FAILURE_ITEM (); \
1194 DEBUG_PRINT2 (" start: 0x%x\n", regstart[this_reg]); \
1197 DEBUG_STATEMENT (nfailure_points_popped++); \
1198 } /* POP_FAILURE_POINT */
1202 /* Structure for per-register (a.k.a. per-group) information.
1203 This must not be longer than one word, because we push this value
1204 onto the failure stack. Other register information, such as the
1205 starting and ending positions (which are addresses), and the list of
1206 inner groups (which is a bits list) are maintained in separate
1209 We are making a (strictly speaking) nonportable assumption here: that
1210 the compiler will pack our bit fields into something that fits into
1211 the type of `word', i.e., is something that fits into one item on the
1215 fail_stack_elt_t word;
1218 /* This field is one if this group can match the empty string,
1219 zero if not. If not yet determined, `MATCH_NULL_UNSET_VALUE'. */
1220 #define MATCH_NULL_UNSET_VALUE 3
1221 unsigned match_null_string_p : 2;
1222 unsigned is_active : 1;
1223 unsigned matched_something : 1;
1224 unsigned ever_matched_something : 1;
1226 } register_info_type;
1228 #define REG_MATCH_NULL_STRING_P(R) ((R).bits.match_null_string_p)
1229 #define IS_ACTIVE(R) ((R).bits.is_active)
1230 #define MATCHED_SOMETHING(R) ((R).bits.matched_something)
1231 #define EVER_MATCHED_SOMETHING(R) ((R).bits.ever_matched_something)
1234 /* Call this when have matched a real character; it sets `matched' flags
1235 for the subexpressions which we are currently inside. Also records
1236 that those subexprs have matched. */
1237 #define SET_REGS_MATCHED() \
1241 for (r = lowest_active_reg; r <= highest_active_reg; r++) \
1243 MATCHED_SOMETHING (reg_info[r]) \
1244 = EVER_MATCHED_SOMETHING (reg_info[r]) \
1251 /* Registers are set to a sentinel when they haven't yet matched. */
1252 #define REG_UNSET_VALUE ((char *) -1)
1253 #define REG_UNSET(e) ((e) == REG_UNSET_VALUE)
1257 /* How do we implement a missing MATCH_MAY_ALLOCATE?
1258 We make the fail stack a global thing, and then grow it to
1259 re_max_failures when we compile. */
1260 #ifndef MATCH_MAY_ALLOCATE
1261 static fail_stack_type fail_stack;
1263 static const char ** regstart, ** regend;
1264 static const char ** old_regstart, ** old_regend;
1265 static const char **best_regstart, **best_regend;
1266 static register_info_type *reg_info;
1267 static const char **reg_dummy;
1268 static register_info_type *reg_info_dummy;
1272 /* Subroutine declarations and macros for regex_compile. */
1274 static void store_op1 (), store_op2 ();
1275 static void insert_op1 (), insert_op2 ();
1276 static boolean at_begline_loc_p (), at_endline_loc_p ();
1277 static boolean group_in_compile_stack ();
1278 static reg_errcode_t compile_range ();
1280 /* Fetch the next character in the uncompiled pattern---translating it
1281 if necessary. Also cast from a signed character in the constant
1282 string passed to us by the user to an unsigned char that we can use
1283 as an array index (in, e.g., `translate'). */
1284 #define PATFETCH(c) \
1285 do {if (p == pend) return REG_EEND; \
1286 c = (unsigned char) *p++; \
1287 if (translate) c = translate[c]; \
1290 /* Fetch the next character in the uncompiled pattern, with no
1292 #define PATFETCH_RAW(c) \
1293 do {if (p == pend) return REG_EEND; \
1294 c = (unsigned char) *p++; \
1297 /* Go backwards one character in the pattern. */
1298 #define PATUNFETCH p--
1301 /* If `translate' is non-null, return translate[D], else just D. We
1302 cast the subscript to translate because some data is declared as
1303 `char *', to avoid warnings when a string constant is passed. But
1304 when we use a character as a subscript we must make it unsigned. */
1305 #define TRANSLATE(d) (translate ? translate[(unsigned char) (d)] : (d))
1308 /* Macros for outputting the compiled pattern into `buffer'. */
1310 /* If the buffer isn't allocated when it comes in, use this. */
1311 #define INIT_BUF_SIZE 32
1313 /* Make sure we have at least N more bytes of space in buffer. */
1314 #define GET_BUFFER_SPACE(n) \
1315 while (b - bufp->buffer + (n) > bufp->allocated) \
1318 /* Make sure we have one more byte of buffer space and then add C to it. */
1319 #define BUF_PUSH(c) \
1321 GET_BUFFER_SPACE (1); \
1322 *b++ = (unsigned char) (c); \
1326 /* Ensure we have two more bytes of buffer space and then append C1 and C2. */
1327 #define BUF_PUSH_2(c1, c2) \
1329 GET_BUFFER_SPACE (2); \
1330 *b++ = (unsigned char) (c1); \
1331 *b++ = (unsigned char) (c2); \
1335 /* As with BUF_PUSH_2, except for three bytes. */
1336 #define BUF_PUSH_3(c1, c2, c3) \
1338 GET_BUFFER_SPACE (3); \
1339 *b++ = (unsigned char) (c1); \
1340 *b++ = (unsigned char) (c2); \
1341 *b++ = (unsigned char) (c3); \
1345 /* Store a jump with opcode OP at LOC to location TO. We store a
1346 relative address offset by the three bytes the jump itself occupies. */
1347 #define STORE_JUMP(op, loc, to) \
1348 store_op1 (op, loc, (to) - (loc) - 3)
1350 /* Likewise, for a two-argument jump. */
1351 #define STORE_JUMP2(op, loc, to, arg) \
1352 store_op2 (op, loc, (to) - (loc) - 3, arg)
1354 /* Like `STORE_JUMP', but for inserting. Assume `b' is the buffer end. */
1355 #define INSERT_JUMP(op, loc, to) \
1356 insert_op1 (op, loc, (to) - (loc) - 3, b)
1358 /* Like `STORE_JUMP2', but for inserting. Assume `b' is the buffer end. */
1359 #define INSERT_JUMP2(op, loc, to, arg) \
1360 insert_op2 (op, loc, (to) - (loc) - 3, arg, b)
1363 /* This is not an arbitrary limit: the arguments which represent offsets
1364 into the pattern are two bytes long. So if 2^16 bytes turns out to
1365 be too small, many things would have to change. */
1366 #define MAX_BUF_SIZE (1L << 16)
1369 /* Extend the buffer by twice its current size via realloc and
1370 reset the pointers that pointed into the old block to point to the
1371 correct places in the new one. If extending the buffer results in it
1372 being larger than MAX_BUF_SIZE, then flag memory exhausted. */
1373 #define EXTEND_BUFFER() \
1375 unsigned char *old_buffer = bufp->buffer; \
1376 if (bufp->allocated == MAX_BUF_SIZE) \
1378 bufp->allocated <<= 1; \
1379 if (bufp->allocated > MAX_BUF_SIZE) \
1380 bufp->allocated = MAX_BUF_SIZE; \
1381 bufp->buffer = (unsigned char *) realloc (bufp->buffer, bufp->allocated);\
1382 if (bufp->buffer == NULL) \
1383 return REG_ESPACE; \
1384 /* If the buffer moved, move all the pointers into it. */ \
1385 if (old_buffer != bufp->buffer) \
1387 b = (b - old_buffer) + bufp->buffer; \
1388 begalt = (begalt - old_buffer) + bufp->buffer; \
1389 if (fixup_alt_jump) \
1390 fixup_alt_jump = (fixup_alt_jump - old_buffer) + bufp->buffer;\
1392 laststart = (laststart - old_buffer) + bufp->buffer; \
1393 if (pending_exact) \
1394 pending_exact = (pending_exact - old_buffer) + bufp->buffer; \
1399 /* Since we have one byte reserved for the register number argument to
1400 {start,stop}_memory, the maximum number of groups we can report
1401 things about is what fits in that byte. */
1402 #define MAX_REGNUM 255
1404 /* But patterns can have more than `MAX_REGNUM' registers. We just
1405 ignore the excess. */
1406 typedef unsigned regnum_t;
1409 /* Macros for the compile stack. */
1411 /* Since offsets can go either forwards or backwards, this type needs to
1412 be able to hold values from -(MAX_BUF_SIZE - 1) to MAX_BUF_SIZE - 1. */
1413 typedef int pattern_offset_t;
1417 pattern_offset_t begalt_offset;
1418 pattern_offset_t fixup_alt_jump;
1419 pattern_offset_t inner_group_offset;
1420 pattern_offset_t laststart_offset;
1422 } compile_stack_elt_t;
1427 compile_stack_elt_t *stack;
1429 unsigned avail; /* Offset of next open position. */
1430 } compile_stack_type;
1433 #define INIT_COMPILE_STACK_SIZE 32
1435 #define COMPILE_STACK_EMPTY (compile_stack.avail == 0)
1436 #define COMPILE_STACK_FULL (compile_stack.avail == compile_stack.size)
1438 /* The next available element. */
1439 #define COMPILE_STACK_TOP (compile_stack.stack[compile_stack.avail])
1442 /* Set the bit for character C in a list. */
1443 #define SET_LIST_BIT(c) \
1444 (b[((unsigned char) (c)) / BYTEWIDTH] \
1445 |= 1 << (((unsigned char) c) % BYTEWIDTH))
1448 /* Get the next unsigned number in the uncompiled pattern. */
1449 #define GET_UNSIGNED_NUMBER(num) \
1453 while (ISDIGIT (c)) \
1457 num = num * 10 + c - '0'; \
1465 #define CHAR_CLASS_MAX_LENGTH 6 /* Namely, `xdigit'. */
1467 #define IS_CHAR_CLASS(string) \
1468 (STREQ (string, "alpha") || STREQ (string, "upper") \
1469 || STREQ (string, "lower") || STREQ (string, "digit") \
1470 || STREQ (string, "alnum") || STREQ (string, "xdigit") \
1471 || STREQ (string, "space") || STREQ (string, "print") \
1472 || STREQ (string, "punct") || STREQ (string, "graph") \
1473 || STREQ (string, "cntrl") || STREQ (string, "blank"))
1475 /* `regex_compile' compiles PATTERN (of length SIZE) according to SYNTAX.
1476 Returns one of error codes defined in `regex.h', or zero for success.
1478 Assumes the `allocated' (and perhaps `buffer') and `translate'
1479 fields are set in BUFP on entry.
1481 If it succeeds, results are put in BUFP (if it returns an error, the
1482 contents of BUFP are undefined):
1483 `buffer' is the compiled pattern;
1484 `syntax' is set to SYNTAX;
1485 `used' is set to the length of the compiled pattern;
1486 `fastmap_accurate' is zero;
1487 `re_nsub' is the number of subexpressions in PATTERN;
1488 `not_bol' and `not_eol' are zero;
1490 The `fastmap' and `newline_anchor' fields are neither
1491 examined nor set. */
1493 static reg_errcode_t
1494 regex_compile (pattern, size, syntax, bufp)
1495 const char *pattern;
1497 reg_syntax_t syntax;
1498 struct re_pattern_buffer *bufp;
1500 /* We fetch characters from PATTERN here. Even though PATTERN is
1501 `char *' (i.e., signed), we declare these variables as unsigned, so
1502 they can be reliably used as array indices. */
1503 register unsigned char c, c1;
1505 /* A random temporary spot in PATTERN. */
1508 /* Points to the end of the buffer, where we should append. */
1509 register unsigned char *b;
1511 /* Keeps track of unclosed groups. */
1512 compile_stack_type compile_stack;
1514 /* Points to the current (ending) position in the pattern. */
1515 const char *p = pattern;
1516 const char *pend = pattern + size;
1518 /* How to translate the characters in the pattern. */
1519 char *translate = bufp->translate;
1521 /* Address of the count-byte of the most recently inserted `exactn'
1522 command. This makes it possible to tell if a new exact-match
1523 character can be added to that command or if the character requires
1524 a new `exactn' command. */
1525 unsigned char *pending_exact = 0;
1527 /* Address of start of the most recently finished expression.
1528 This tells, e.g., postfix * where to find the start of its
1529 operand. Reset at the beginning of groups and alternatives. */
1530 unsigned char *laststart = 0;
1532 /* Address of beginning of regexp, or inside of last group. */
1533 unsigned char *begalt;
1535 /* Place in the uncompiled pattern (i.e., the {) to
1536 which to go back if the interval is invalid. */
1537 const char *beg_interval;
1539 /* Address of the place where a forward jump should go to the end of
1540 the containing expression. Each alternative of an `or' -- except the
1541 last -- ends with a forward jump of this sort. */
1542 unsigned char *fixup_alt_jump = 0;
1544 /* Counts open-groups as they are encountered. Remembered for the
1545 matching close-group on the compile stack, so the same register
1546 number is put in the stop_memory as the start_memory. */
1547 regnum_t regnum = 0;
1550 DEBUG_PRINT1 ("\nCompiling pattern: ");
1553 unsigned debug_count;
1555 for (debug_count = 0; debug_count < size; debug_count++)
1556 printchar (pattern[debug_count]);
1561 /* Initialize the compile stack. */
1562 compile_stack.stack = TALLOC (INIT_COMPILE_STACK_SIZE, compile_stack_elt_t);
1563 if (compile_stack.stack == NULL)
1566 compile_stack.size = INIT_COMPILE_STACK_SIZE;
1567 compile_stack.avail = 0;
1569 /* Initialize the pattern buffer. */
1570 bufp->syntax = syntax;
1571 bufp->fastmap_accurate = 0;
1572 bufp->not_bol = bufp->not_eol = 0;
1574 /* Set `used' to zero, so that if we return an error, the pattern
1575 printer (for debugging) will think there's no pattern. We reset it
1579 /* Always count groups, whether or not bufp->no_sub is set. */
1582 #if !defined (emacs) && !defined (SYNTAX_TABLE)
1583 /* Initialize the syntax table. */
1584 init_syntax_once ();
1587 if (bufp->allocated == 0)
1590 { /* If zero allocated, but buffer is non-null, try to realloc
1591 enough space. This loses if buffer's address is bogus, but
1592 that is the user's responsibility. */
1593 RETALLOC (bufp->buffer, INIT_BUF_SIZE, unsigned char);
1596 { /* Caller did not allocate a buffer. Do it for them. */
1597 bufp->buffer = TALLOC (INIT_BUF_SIZE, unsigned char);
1599 if (!bufp->buffer) return REG_ESPACE;
1601 bufp->allocated = INIT_BUF_SIZE;
1604 begalt = b = bufp->buffer;
1606 /* Loop through the uncompiled pattern until we're at the end. */
1615 if ( /* If at start of pattern, it's an operator. */
1617 /* If context independent, it's an operator. */
1618 || syntax & RE_CONTEXT_INDEP_ANCHORS
1619 /* Otherwise, depends on what's come before. */
1620 || at_begline_loc_p (pattern, p, syntax))
1630 if ( /* If at end of pattern, it's an operator. */
1632 /* If context independent, it's an operator. */
1633 || syntax & RE_CONTEXT_INDEP_ANCHORS
1634 /* Otherwise, depends on what's next. */
1635 || at_endline_loc_p (p, pend, syntax))
1645 if ((syntax & RE_BK_PLUS_QM)
1646 || (syntax & RE_LIMITED_OPS))
1650 /* If there is no previous pattern... */
1653 if (syntax & RE_CONTEXT_INVALID_OPS)
1655 else if (!(syntax & RE_CONTEXT_INDEP_OPS))
1660 /* Are we optimizing this jump? */
1661 boolean keep_string_p = false;
1663 /* 1 means zero (many) matches is allowed. */
1664 char zero_times_ok = 0, many_times_ok = 0;
1666 /* If there is a sequence of repetition chars, collapse it
1667 down to just one (the right one). We can't combine
1668 interval operators with these because of, e.g., `a{2}*',
1669 which should only match an even number of `a's. */
1673 zero_times_ok |= c != '+';
1674 many_times_ok |= c != '?';
1682 || (!(syntax & RE_BK_PLUS_QM) && (c == '+' || c == '?')))
1685 else if (syntax & RE_BK_PLUS_QM && c == '\\')
1687 if (p == pend) return REG_EESCAPE;
1690 if (!(c1 == '+' || c1 == '?'))
1705 /* If we get here, we found another repeat character. */
1708 /* Star, etc. applied to an empty pattern is equivalent
1709 to an empty pattern. */
1713 /* Now we know whether or not zero matches is allowed
1714 and also whether or not two or more matches is allowed. */
1716 { /* More than one repetition is allowed, so put in at the
1717 end a backward relative jump from `b' to before the next
1718 jump we're going to put in below (which jumps from
1719 laststart to after this jump).
1721 But if we are at the `*' in the exact sequence `.*\n',
1722 insert an unconditional jump backwards to the .,
1723 instead of the beginning of the loop. This way we only
1724 push a failure point once, instead of every time
1725 through the loop. */
1726 assert (p - 1 > pattern);
1728 /* Allocate the space for the jump. */
1729 GET_BUFFER_SPACE (3);
1731 /* We know we are not at the first character of the pattern,
1732 because laststart was nonzero. And we've already
1733 incremented `p', by the way, to be the character after
1734 the `*'. Do we have to do something analogous here
1735 for null bytes, because of RE_DOT_NOT_NULL? */
1736 if (TRANSLATE (*(p - 2)) == TRANSLATE ('.')
1738 && p < pend && TRANSLATE (*p) == TRANSLATE ('\n')
1739 && !(syntax & RE_DOT_NEWLINE))
1740 { /* We have .*\n. */
1741 STORE_JUMP (jump, b, laststart);
1742 keep_string_p = true;
1745 /* Anything else. */
1746 STORE_JUMP (maybe_pop_jump, b, laststart - 3);
1748 /* We've added more stuff to the buffer. */
1752 /* On failure, jump from laststart to b + 3, which will be the
1753 end of the buffer after this jump is inserted. */
1754 GET_BUFFER_SPACE (3);
1755 INSERT_JUMP (keep_string_p ? on_failure_keep_string_jump
1763 /* At least one repetition is required, so insert a
1764 `dummy_failure_jump' before the initial
1765 `on_failure_jump' instruction of the loop. This
1766 effects a skip over that instruction the first time
1767 we hit that loop. */
1768 GET_BUFFER_SPACE (3);
1769 INSERT_JUMP (dummy_failure_jump, laststart, laststart + 6);
1784 boolean had_char_class = false;
1786 if (p == pend) return REG_EBRACK;
1788 /* Ensure that we have enough space to push a charset: the
1789 opcode, the length count, and the bitset; 34 bytes in all. */
1790 GET_BUFFER_SPACE (34);
1794 /* We test `*p == '^' twice, instead of using an if
1795 statement, so we only need one BUF_PUSH. */
1796 BUF_PUSH (*p == '^' ? charset_not : charset);
1800 /* Remember the first position in the bracket expression. */
1803 /* Push the number of bytes in the bitmap. */
1804 BUF_PUSH ((1 << BYTEWIDTH) / BYTEWIDTH);
1806 /* Clear the whole map. */
1807 bzero (b, (1 << BYTEWIDTH) / BYTEWIDTH);
1809 /* charset_not matches newline according to a syntax bit. */
1810 if ((re_opcode_t) b[-2] == charset_not
1811 && (syntax & RE_HAT_LISTS_NOT_NEWLINE))
1812 SET_LIST_BIT ('\n');
1814 /* Read in characters and ranges, setting map bits. */
1817 if (p == pend) return REG_EBRACK;
1821 /* \ might escape characters inside [...] and [^...]. */
1822 if ((syntax & RE_BACKSLASH_ESCAPE_IN_LISTS) && c == '\\')
1824 if (p == pend) return REG_EESCAPE;
1831 /* Could be the end of the bracket expression. If it's
1832 not (i.e., when the bracket expression is `[]' so
1833 far), the ']' character bit gets set way below. */
1834 if (c == ']' && p != p1 + 1)
1837 /* Look ahead to see if it's a range when the last thing
1838 was a character class. */
1839 if (had_char_class && c == '-' && *p != ']')
1842 /* Look ahead to see if it's a range when the last thing
1843 was a character: if this is a hyphen not at the
1844 beginning or the end of a list, then it's the range
1847 && !(p - 2 >= pattern && p[-2] == '[')
1848 && !(p - 3 >= pattern && p[-3] == '[' && p[-2] == '^')
1852 = compile_range (&p, pend, translate, syntax, b);
1853 if (ret != REG_NOERROR) return ret;
1856 else if (p[0] == '-' && p[1] != ']')
1857 { /* This handles ranges made up of characters only. */
1860 /* Move past the `-'. */
1863 ret = compile_range (&p, pend, translate, syntax, b);
1864 if (ret != REG_NOERROR) return ret;
1867 /* See if we're at the beginning of a possible character
1870 else if (syntax & RE_CHAR_CLASSES && c == '[' && *p == ':')
1871 { /* Leave room for the null. */
1872 char str[CHAR_CLASS_MAX_LENGTH + 1];
1877 /* If pattern is `[[:'. */
1878 if (p == pend) return REG_EBRACK;
1883 if (c == ':' || c == ']' || p == pend
1884 || c1 == CHAR_CLASS_MAX_LENGTH)
1890 /* If isn't a word bracketed by `[:' and:`]':
1891 undo the ending character, the letters, and leave
1892 the leading `:' and `[' (but set bits for them). */
1893 if (c == ':' && *p == ']')
1896 boolean is_alnum = STREQ (str, "alnum");
1897 boolean is_alpha = STREQ (str, "alpha");
1898 boolean is_blank = STREQ (str, "blank");
1899 boolean is_cntrl = STREQ (str, "cntrl");
1900 boolean is_digit = STREQ (str, "digit");
1901 boolean is_graph = STREQ (str, "graph");
1902 boolean is_lower = STREQ (str, "lower");
1903 boolean is_print = STREQ (str, "print");
1904 boolean is_punct = STREQ (str, "punct");
1905 boolean is_space = STREQ (str, "space");
1906 boolean is_upper = STREQ (str, "upper");
1907 boolean is_xdigit = STREQ (str, "xdigit");
1909 if (!IS_CHAR_CLASS (str)) return REG_ECTYPE;
1911 /* Throw away the ] at the end of the character
1915 if (p == pend) return REG_EBRACK;
1917 for (ch = 0; ch < 1 << BYTEWIDTH; ch++)
1919 if ( (is_alnum && ISALNUM (ch))
1920 || (is_alpha && ISALPHA (ch))
1921 || (is_blank && ISBLANK (ch))
1922 || (is_cntrl && ISCNTRL (ch))
1923 || (is_digit && ISDIGIT (ch))
1924 || (is_graph && ISGRAPH (ch))
1925 || (is_lower && ISLOWER (ch))
1926 || (is_print && ISPRINT (ch))
1927 || (is_punct && ISPUNCT (ch))
1928 || (is_space && ISSPACE (ch))
1929 || (is_upper && ISUPPER (ch))
1930 || (is_xdigit && ISXDIGIT (ch)))
1933 had_char_class = true;
1942 had_char_class = false;
1947 had_char_class = false;
1952 /* Discard any (non)matching list bytes that are all 0 at the
1953 end of the map. Decrease the map-length byte too. */
1954 while ((int) b[-1] > 0 && b[b[-1] - 1] == 0)
1962 if (syntax & RE_NO_BK_PARENS)
1969 if (syntax & RE_NO_BK_PARENS)
1976 if (syntax & RE_NEWLINE_ALT)
1983 if (syntax & RE_NO_BK_VBAR)
1990 if (syntax & RE_INTERVALS && syntax & RE_NO_BK_BRACES)
1991 goto handle_interval;
1997 if (p == pend) return REG_EESCAPE;
1999 /* Do not translate the character after the \, so that we can
2000 distinguish, e.g., \B from \b, even if we normally would
2001 translate, e.g., B to b. */
2007 if (syntax & RE_NO_BK_PARENS)
2008 goto normal_backslash;
2014 if (COMPILE_STACK_FULL)
2016 RETALLOC (compile_stack.stack, compile_stack.size << 1,
2017 compile_stack_elt_t);
2018 if (compile_stack.stack == NULL) return REG_ESPACE;
2020 compile_stack.size <<= 1;
2023 /* These are the values to restore when we hit end of this
2024 group. They are all relative offsets, so that if the
2025 whole pattern moves because of realloc, they will still
2027 COMPILE_STACK_TOP.begalt_offset = begalt - bufp->buffer;
2028 COMPILE_STACK_TOP.fixup_alt_jump
2029 = fixup_alt_jump ? fixup_alt_jump - bufp->buffer + 1 : 0;
2030 COMPILE_STACK_TOP.laststart_offset = b - bufp->buffer;
2031 COMPILE_STACK_TOP.regnum = regnum;
2033 /* We will eventually replace the 0 with the number of
2034 groups inner to this one. But do not push a
2035 start_memory for groups beyond the last one we can
2036 represent in the compiled pattern. */
2037 if (regnum <= MAX_REGNUM)
2039 COMPILE_STACK_TOP.inner_group_offset = b - bufp->buffer + 2;
2040 BUF_PUSH_3 (start_memory, regnum, 0);
2043 compile_stack.avail++;
2048 /* If we've reached MAX_REGNUM groups, then this open
2049 won't actually generate any code, so we'll have to
2050 clear pending_exact explicitly. */
2056 if (syntax & RE_NO_BK_PARENS) goto normal_backslash;
2058 if (COMPILE_STACK_EMPTY)
2059 if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD)
2060 goto normal_backslash;
2066 { /* Push a dummy failure point at the end of the
2067 alternative for a possible future
2068 `pop_failure_jump' to pop. See comments at
2069 `push_dummy_failure' in `re_match_2'. */
2070 BUF_PUSH (push_dummy_failure);
2072 /* We allocated space for this jump when we assigned
2073 to `fixup_alt_jump', in the `handle_alt' case below. */
2074 STORE_JUMP (jump_past_alt, fixup_alt_jump, b - 1);
2077 /* See similar code for backslashed left paren above. */
2078 if (COMPILE_STACK_EMPTY)
2079 if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD)
2084 /* Since we just checked for an empty stack above, this
2085 ``can't happen''. */
2086 assert (compile_stack.avail != 0);
2088 /* We don't just want to restore into `regnum', because
2089 later groups should continue to be numbered higher,
2090 as in `(ab)c(de)' -- the second group is #2. */
2091 regnum_t this_group_regnum;
2093 compile_stack.avail--;
2094 begalt = bufp->buffer + COMPILE_STACK_TOP.begalt_offset;
2096 = COMPILE_STACK_TOP.fixup_alt_jump
2097 ? bufp->buffer + COMPILE_STACK_TOP.fixup_alt_jump - 1
2099 laststart = bufp->buffer + COMPILE_STACK_TOP.laststart_offset;
2100 this_group_regnum = COMPILE_STACK_TOP.regnum;
2101 /* If we've reached MAX_REGNUM groups, then this open
2102 won't actually generate any code, so we'll have to
2103 clear pending_exact explicitly. */
2106 /* We're at the end of the group, so now we know how many
2107 groups were inside this one. */
2108 if (this_group_regnum <= MAX_REGNUM)
2110 unsigned char *inner_group_loc
2111 = bufp->buffer + COMPILE_STACK_TOP.inner_group_offset;
2113 *inner_group_loc = regnum - this_group_regnum;
2114 BUF_PUSH_3 (stop_memory, this_group_regnum,
2115 regnum - this_group_regnum);
2121 case '|': /* `\|'. */
2122 if (syntax & RE_LIMITED_OPS || syntax & RE_NO_BK_VBAR)
2123 goto normal_backslash;
2125 if (syntax & RE_LIMITED_OPS)
2128 /* Insert before the previous alternative a jump which
2129 jumps to this alternative if the former fails. */
2130 GET_BUFFER_SPACE (3);
2131 INSERT_JUMP (on_failure_jump, begalt, b + 6);
2135 /* The alternative before this one has a jump after it
2136 which gets executed if it gets matched. Adjust that
2137 jump so it will jump to this alternative's analogous
2138 jump (put in below, which in turn will jump to the next
2139 (if any) alternative's such jump, etc.). The last such
2140 jump jumps to the correct final destination. A picture:
2146 If we are at `b', then fixup_alt_jump right now points to a
2147 three-byte space after `a'. We'll put in the jump, set
2148 fixup_alt_jump to right after `b', and leave behind three
2149 bytes which we'll fill in when we get to after `c'. */
2152 STORE_JUMP (jump_past_alt, fixup_alt_jump, b);
2154 /* Mark and leave space for a jump after this alternative,
2155 to be filled in later either by next alternative or
2156 when know we're at the end of a series of alternatives. */
2158 GET_BUFFER_SPACE (3);
2167 /* If \{ is a literal. */
2168 if (!(syntax & RE_INTERVALS)
2169 /* If we're at `\{' and it's not the open-interval
2171 || ((syntax & RE_INTERVALS) && (syntax & RE_NO_BK_BRACES))
2172 || (p - 2 == pattern && p == pend))
2173 goto normal_backslash;
2177 /* If got here, then the syntax allows intervals. */
2179 /* At least (most) this many matches must be made. */
2180 int lower_bound = -1, upper_bound = -1;
2182 beg_interval = p - 1;
2186 if (syntax & RE_NO_BK_BRACES)
2187 goto unfetch_interval;
2192 GET_UNSIGNED_NUMBER (lower_bound);
2196 GET_UNSIGNED_NUMBER (upper_bound);
2197 if (upper_bound < 0) upper_bound = RE_DUP_MAX;
2200 /* Interval such as `{1}' => match exactly once. */
2201 upper_bound = lower_bound;
2203 if (lower_bound < 0 || upper_bound > RE_DUP_MAX
2204 || lower_bound > upper_bound)
2206 if (syntax & RE_NO_BK_BRACES)
2207 goto unfetch_interval;
2212 if (!(syntax & RE_NO_BK_BRACES))
2214 if (c != '\\') return REG_EBRACE;
2221 if (syntax & RE_NO_BK_BRACES)
2222 goto unfetch_interval;
2227 /* We just parsed a valid interval. */
2229 /* If it's invalid to have no preceding re. */
2232 if (syntax & RE_CONTEXT_INVALID_OPS)
2234 else if (syntax & RE_CONTEXT_INDEP_OPS)
2237 goto unfetch_interval;
2240 /* If the upper bound is zero, don't want to succeed at
2241 all; jump from `laststart' to `b + 3', which will be
2242 the end of the buffer after we insert the jump. */
2243 if (upper_bound == 0)
2245 GET_BUFFER_SPACE (3);
2246 INSERT_JUMP (jump, laststart, b + 3);
2250 /* Otherwise, we have a nontrivial interval. When
2251 we're all done, the pattern will look like:
2252 set_number_at <jump count> <upper bound>
2253 set_number_at <succeed_n count> <lower bound>
2254 succeed_n <after jump addr> <succeed_n count>
2256 jump_n <succeed_n addr> <jump count>
2257 (The upper bound and `jump_n' are omitted if
2258 `upper_bound' is 1, though.) */
2260 { /* If the upper bound is > 1, we need to insert
2261 more at the end of the loop. */
2262 unsigned nbytes = 10 + (upper_bound > 1) * 10;
2264 GET_BUFFER_SPACE (nbytes);
2266 /* Initialize lower bound of the `succeed_n', even
2267 though it will be set during matching by its
2268 attendant `set_number_at' (inserted next),
2269 because `re_compile_fastmap' needs to know.
2270 Jump to the `jump_n' we might insert below. */
2271 INSERT_JUMP2 (succeed_n, laststart,
2272 b + 5 + (upper_bound > 1) * 5,
2276 /* Code to initialize the lower bound. Insert
2277 before the `succeed_n'. The `5' is the last two
2278 bytes of this `set_number_at', plus 3 bytes of
2279 the following `succeed_n'. */
2280 insert_op2 (set_number_at, laststart, 5, lower_bound, b);
2283 if (upper_bound > 1)
2284 { /* More than one repetition is allowed, so
2285 append a backward jump to the `succeed_n'
2286 that starts this interval.
2288 When we've reached this during matching,
2289 we'll have matched the interval once, so
2290 jump back only `upper_bound - 1' times. */
2291 STORE_JUMP2 (jump_n, b, laststart + 5,
2295 /* The location we want to set is the second
2296 parameter of the `jump_n'; that is `b-2' as
2297 an absolute address. `laststart' will be
2298 the `set_number_at' we're about to insert;
2299 `laststart+3' the number to set, the source
2300 for the relative address. But we are
2301 inserting into the middle of the pattern --
2302 so everything is getting moved up by 5.
2303 Conclusion: (b - 2) - (laststart + 3) + 5,
2304 i.e., b - laststart.
2306 We insert this at the beginning of the loop
2307 so that if we fail during matching, we'll
2308 reinitialize the bounds. */
2309 insert_op2 (set_number_at, laststart, b - laststart,
2310 upper_bound - 1, b);
2315 beg_interval = NULL;
2320 /* If an invalid interval, match the characters as literals. */
2321 assert (beg_interval);
2323 beg_interval = NULL;
2325 /* normal_char and normal_backslash need `c'. */
2328 if (!(syntax & RE_NO_BK_BRACES))
2330 if (p > pattern && p[-1] == '\\')
2331 goto normal_backslash;
2336 /* There is no way to specify the before_dot and after_dot
2337 operators. rms says this is ok. --karl */
2345 BUF_PUSH_2 (syntaxspec, syntax_spec_code[c]);
2351 BUF_PUSH_2 (notsyntaxspec, syntax_spec_code[c]);
2358 BUF_PUSH (wordchar);
2364 BUF_PUSH (notwordchar);
2377 BUF_PUSH (wordbound);
2381 BUF_PUSH (notwordbound);
2392 case '1': case '2': case '3': case '4': case '5':
2393 case '6': case '7': case '8': case '9':
2394 if (syntax & RE_NO_BK_REFS)
2402 /* Can't back reference to a subexpression if inside of it. */
2403 if (group_in_compile_stack (compile_stack, c1))
2407 BUF_PUSH_2 (duplicate, c1);
2413 if (syntax & RE_BK_PLUS_QM)
2416 goto normal_backslash;
2420 /* You might think it would be useful for \ to mean
2421 not to translate; but if we don't translate it
2422 it will never match anything. */
2430 /* Expects the character in `c'. */
2432 /* If no exactn currently being built. */
2435 /* If last exactn not at current position. */
2436 || pending_exact + *pending_exact + 1 != b
2438 /* We have only one byte following the exactn for the count. */
2439 || *pending_exact == (1 << BYTEWIDTH) - 1
2441 /* If followed by a repetition operator. */
2442 || *p == '*' || *p == '^'
2443 || ((syntax & RE_BK_PLUS_QM)
2444 ? *p == '\\' && (p[1] == '+' || p[1] == '?')
2445 : (*p == '+' || *p == '?'))
2446 || ((syntax & RE_INTERVALS)
2447 && ((syntax & RE_NO_BK_BRACES)
2449 : (p[0] == '\\' && p[1] == '{'))))
2451 /* Start building a new exactn. */
2455 BUF_PUSH_2 (exactn, 0);
2456 pending_exact = b - 1;
2463 } /* while p != pend */
2466 /* Through the pattern now. */
2469 STORE_JUMP (jump_past_alt, fixup_alt_jump, b);
2471 if (!COMPILE_STACK_EMPTY)
2474 free (compile_stack.stack);
2476 /* We have succeeded; set the length of the buffer. */
2477 bufp->used = b - bufp->buffer;
2482 DEBUG_PRINT1 ("\nCompiled pattern: \n");
2483 print_compiled_pattern (bufp);
2487 #ifndef MATCH_MAY_ALLOCATE
2488 /* Initialize the failure stack to the largest possible stack. This
2489 isn't necessary unless we're trying to avoid calling alloca in
2490 the search and match routines. */
2492 int num_regs = bufp->re_nsub + 1;
2494 /* Since DOUBLE_FAIL_STACK refuses to double only if the current size
2495 is strictly greater than re_max_failures, the largest possible stack
2496 is 2 * re_max_failures failure points. */
2497 fail_stack.size = (2 * re_max_failures * MAX_FAILURE_ITEMS);
2498 if (! fail_stack.stack)
2500 (fail_stack_elt_t *) malloc (fail_stack.size
2501 * sizeof (fail_stack_elt_t));
2503 /* Initialize some other variables the matcher uses. */
2504 RETALLOC_IF (regstart, num_regs, const char *);
2505 RETALLOC_IF (regend, num_regs, const char *);
2506 RETALLOC_IF (old_regstart, num_regs, const char *);
2507 RETALLOC_IF (old_regend, num_regs, const char *);
2508 RETALLOC_IF (best_regstart, num_regs, const char *);
2509 RETALLOC_IF (best_regend, num_regs, const char *);
2510 RETALLOC_IF (reg_info, num_regs, register_info_type);
2511 RETALLOC_IF (reg_dummy, num_regs, const char *);
2512 RETALLOC_IF (reg_info_dummy, num_regs, register_info_type);
2517 } /* regex_compile */
2519 /* Subroutines for `regex_compile'. */
2521 /* Store OP at LOC followed by two-byte integer parameter ARG. */
2524 store_op1 (op, loc, arg)
2529 *loc = (unsigned char) op;
2530 STORE_NUMBER (loc + 1, arg);
2534 /* Like `store_op1', but for two two-byte parameters ARG1 and ARG2. */
2537 store_op2 (op, loc, arg1, arg2)
2542 *loc = (unsigned char) op;
2543 STORE_NUMBER (loc + 1, arg1);
2544 STORE_NUMBER (loc + 3, arg2);
2548 /* Copy the bytes from LOC to END to open up three bytes of space at LOC
2549 for OP followed by two-byte integer parameter ARG. */
2552 insert_op1 (op, loc, arg, end)
2558 register unsigned char *pfrom = end;
2559 register unsigned char *pto = end + 3;
2561 while (pfrom != loc)
2564 store_op1 (op, loc, arg);
2568 /* Like `insert_op1', but for two two-byte parameters ARG1 and ARG2. */
2571 insert_op2 (op, loc, arg1, arg2, end)
2577 register unsigned char *pfrom = end;
2578 register unsigned char *pto = end + 5;
2580 while (pfrom != loc)
2583 store_op2 (op, loc, arg1, arg2);
2587 /* P points to just after a ^ in PATTERN. Return true if that ^ comes
2588 after an alternative or a begin-subexpression. We assume there is at
2589 least one character before the ^. */
2592 at_begline_loc_p (pattern, p, syntax)
2593 const char *pattern, *p;
2594 reg_syntax_t syntax;
2596 const char *prev = p - 2;
2597 boolean prev_prev_backslash = prev > pattern && prev[-1] == '\\';
2600 /* After a subexpression? */
2601 (*prev == '(' && (syntax & RE_NO_BK_PARENS || prev_prev_backslash))
2602 /* After an alternative? */
2603 || (*prev == '|' && (syntax & RE_NO_BK_VBAR || prev_prev_backslash));
2607 /* The dual of at_begline_loc_p. This one is for $. We assume there is
2608 at least one character after the $, i.e., `P < PEND'. */
2611 at_endline_loc_p (p, pend, syntax)
2612 const char *p, *pend;
2615 const char *next = p;
2616 boolean next_backslash = *next == '\\';
2617 const char *next_next = p + 1 < pend ? p + 1 : NULL;
2620 /* Before a subexpression? */
2621 (syntax & RE_NO_BK_PARENS ? *next == ')'
2622 : next_backslash && next_next && *next_next == ')')
2623 /* Before an alternative? */
2624 || (syntax & RE_NO_BK_VBAR ? *next == '|'
2625 : next_backslash && next_next && *next_next == '|');
2629 /* Returns true if REGNUM is in one of COMPILE_STACK's elements and
2630 false if it's not. */
2633 group_in_compile_stack (compile_stack, regnum)
2634 compile_stack_type compile_stack;
2639 for (this_element = compile_stack.avail - 1;
2642 if (compile_stack.stack[this_element].regnum == regnum)
2649 /* Read the ending character of a range (in a bracket expression) from the
2650 uncompiled pattern *P_PTR (which ends at PEND). We assume the
2651 starting character is in `P[-2]'. (`P[-1]' is the character `-'.)
2652 Then we set the translation of all bits between the starting and
2653 ending characters (inclusive) in the compiled pattern B.
2655 Return an error code.
2657 We use these short variable names so we can use the same macros as
2658 `regex_compile' itself. */
2660 static reg_errcode_t
2661 compile_range (p_ptr, pend, translate, syntax, b)
2662 const char **p_ptr, *pend;
2664 reg_syntax_t syntax;
2669 const char *p = *p_ptr;
2670 int range_start, range_end;
2675 /* Even though the pattern is a signed `char *', we need to fetch
2676 with unsigned char *'s; if the high bit of the pattern character
2677 is set, the range endpoints will be negative if we fetch using a
2680 We also want to fetch the endpoints without translating them; the
2681 appropriate translation is done in the bit-setting loop below. */
2682 range_start = ((unsigned char *) p)[-2];
2683 range_end = ((unsigned char *) p)[0];
2685 /* Have to increment the pointer into the pattern string, so the
2686 caller isn't still at the ending character. */
2689 /* If the start is after the end, the range is empty. */
2690 if (range_start > range_end)
2691 return syntax & RE_NO_EMPTY_RANGES ? REG_ERANGE : REG_NOERROR;
2693 /* Here we see why `this_char' has to be larger than an `unsigned
2694 char' -- the range is inclusive, so if `range_end' == 0xff
2695 (assuming 8-bit characters), we would otherwise go into an infinite
2696 loop, since all characters <= 0xff. */
2697 for (this_char = range_start; this_char <= range_end; this_char++)
2699 SET_LIST_BIT (TRANSLATE (this_char));
2705 /* re_compile_fastmap computes a ``fastmap'' for the compiled pattern in
2706 BUFP. A fastmap records which of the (1 << BYTEWIDTH) possible
2707 characters can start a string that matches the pattern. This fastmap
2708 is used by re_search to skip quickly over impossible starting points.
2710 The caller must supply the address of a (1 << BYTEWIDTH)-byte data
2711 area as BUFP->fastmap.
2713 We set the `fastmap', `fastmap_accurate', and `can_be_null' fields in
2716 Returns 0 if we succeed, -2 if an internal error. */
2719 re_compile_fastmap (bufp)
2720 struct re_pattern_buffer *bufp;
2723 #ifdef MATCH_MAY_ALLOCATE
2724 fail_stack_type fail_stack;
2726 #ifndef REGEX_MALLOC
2729 /* We don't push any register information onto the failure stack. */
2730 unsigned num_regs = 0;
2732 register char *fastmap = bufp->fastmap;
2733 unsigned char *pattern = bufp->buffer;
2734 unsigned long size = bufp->used;
2735 unsigned char *p = pattern;
2736 register unsigned char *pend = pattern + size;
2738 /* Assume that each path through the pattern can be null until
2739 proven otherwise. We set this false at the bottom of switch
2740 statement, to which we get only if a particular path doesn't
2741 match the empty string. */
2742 boolean path_can_be_null = true;
2744 /* We aren't doing a `succeed_n' to begin with. */
2745 boolean succeed_n_p = false;
2747 assert (fastmap != NULL && p != NULL);
2750 bzero (fastmap, 1 << BYTEWIDTH); /* Assume nothing's valid. */
2751 bufp->fastmap_accurate = 1; /* It will be when we're done. */
2752 bufp->can_be_null = 0;
2754 while (p != pend || !FAIL_STACK_EMPTY ())
2758 bufp->can_be_null |= path_can_be_null;
2760 /* Reset for next path. */
2761 path_can_be_null = true;
2763 p = fail_stack.stack[--fail_stack.avail];
2766 /* We should never be about to go beyond the end of the pattern. */
2769 #ifdef SWITCH_ENUM_BUG
2770 switch ((int) ((re_opcode_t) *p++))
2772 switch ((re_opcode_t) *p++)
2776 /* I guess the idea here is to simply not bother with a fastmap
2777 if a backreference is used, since it's too hard to figure out
2778 the fastmap for the corresponding group. Setting
2779 `can_be_null' stops `re_search_2' from using the fastmap, so
2780 that is all we do. */
2782 bufp->can_be_null = 1;
2786 /* Following are the cases which match a character. These end
2795 for (j = *p++ * BYTEWIDTH - 1; j >= 0; j--)
2796 if (p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH)))
2802 /* Chars beyond end of map must be allowed. */
2803 for (j = *p * BYTEWIDTH; j < (1 << BYTEWIDTH); j++)
2806 for (j = *p++ * BYTEWIDTH - 1; j >= 0; j--)
2807 if (!(p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH))))
2813 for (j = 0; j < (1 << BYTEWIDTH); j++)
2814 if (SYNTAX (j) == Sword)
2820 for (j = 0; j < (1 << BYTEWIDTH); j++)
2821 if (SYNTAX (j) != Sword)
2827 /* `.' matches anything ... */
2828 for (j = 0; j < (1 << BYTEWIDTH); j++)
2831 /* ... except perhaps newline. */
2832 if (!(bufp->syntax & RE_DOT_NEWLINE))
2835 /* Return if we have already set `can_be_null'; if we have,
2836 then the fastmap is irrelevant. Something's wrong here. */
2837 else if (bufp->can_be_null)
2840 /* Otherwise, have to check alternative paths. */
2847 for (j = 0; j < (1 << BYTEWIDTH); j++)
2848 if (SYNTAX (j) == (enum syntaxcode) k)
2855 for (j = 0; j < (1 << BYTEWIDTH); j++)
2856 if (SYNTAX (j) != (enum syntaxcode) k)
2861 /* All cases after this match the empty string. These end with
2869 #endif /* not emacs */
2881 case push_dummy_failure:
2886 case pop_failure_jump:
2887 case maybe_pop_jump:
2890 case dummy_failure_jump:
2891 EXTRACT_NUMBER_AND_INCR (j, p);
2896 /* Jump backward implies we just went through the body of a
2897 loop and matched nothing. Opcode jumped to should be
2898 `on_failure_jump' or `succeed_n'. Just treat it like an
2899 ordinary jump. For a * loop, it has pushed its failure
2900 point already; if so, discard that as redundant. */
2901 if ((re_opcode_t) *p != on_failure_jump
2902 && (re_opcode_t) *p != succeed_n)
2906 EXTRACT_NUMBER_AND_INCR (j, p);
2909 /* If what's on the stack is where we are now, pop it. */
2910 if (!FAIL_STACK_EMPTY ()
2911 && fail_stack.stack[fail_stack.avail - 1] == p)
2917 case on_failure_jump:
2918 case on_failure_keep_string_jump:
2919 handle_on_failure_jump:
2920 EXTRACT_NUMBER_AND_INCR (j, p);
2922 /* For some patterns, e.g., `(a?)?', `p+j' here points to the
2923 end of the pattern. We don't want to push such a point,
2924 since when we restore it above, entering the switch will
2925 increment `p' past the end of the pattern. We don't need
2926 to push such a point since we obviously won't find any more
2927 fastmap entries beyond `pend'. Such a pattern can match
2928 the null string, though. */
2931 if (!PUSH_PATTERN_OP (p + j, fail_stack))
2935 bufp->can_be_null = 1;
2939 EXTRACT_NUMBER_AND_INCR (k, p); /* Skip the n. */
2940 succeed_n_p = false;
2947 /* Get to the number of times to succeed. */
2950 /* Increment p past the n for when k != 0. */
2951 EXTRACT_NUMBER_AND_INCR (k, p);
2955 succeed_n_p = true; /* Spaghetti code alert. */
2956 goto handle_on_failure_jump;
2973 abort (); /* We have listed all the cases. */
2976 /* Getting here means we have found the possible starting
2977 characters for one path of the pattern -- and that the empty
2978 string does not match. We need not follow this path further.
2979 Instead, look at the next alternative (remembered on the
2980 stack), or quit if no more. The test at the top of the loop
2981 does these things. */
2982 path_can_be_null = false;
2986 /* Set `can_be_null' for the last path (also the first path, if the
2987 pattern is empty). */
2988 bufp->can_be_null |= path_can_be_null;
2990 } /* re_compile_fastmap */
2992 /* Set REGS to hold NUM_REGS registers, storing them in STARTS and
2993 ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use
2994 this memory for recording register information. STARTS and ENDS
2995 must be allocated using the malloc library routine, and must each
2996 be at least NUM_REGS * sizeof (regoff_t) bytes long.
2998 If NUM_REGS == 0, then subsequent matches should allocate their own
3001 Unless this function is called, the first search or match using
3002 PATTERN_BUFFER will allocate its own register data, without
3003 freeing the old data. */
3006 re_set_registers (bufp, regs, num_regs, starts, ends)
3007 struct re_pattern_buffer *bufp;
3008 struct re_registers *regs;
3010 regoff_t *starts, *ends;
3014 bufp->regs_allocated = REGS_REALLOCATE;
3015 regs->num_regs = num_regs;
3016 regs->start = starts;
3021 bufp->regs_allocated = REGS_UNALLOCATED;
3023 regs->start = regs->end = (regoff_t *) 0;
3027 /* Searching routines. */
3029 /* Like re_search_2, below, but only one string is specified, and
3030 doesn't let you say where to stop matching. */
3033 re_search (bufp, string, size, startpos, range, regs)
3034 struct re_pattern_buffer *bufp;
3036 int size, startpos, range;
3037 struct re_registers *regs;
3039 return re_search_2 (bufp, NULL, 0, string, size, startpos, range,
3044 /* Using the compiled pattern in BUFP->buffer, first tries to match the
3045 virtual concatenation of STRING1 and STRING2, starting first at index
3046 STARTPOS, then at STARTPOS + 1, and so on.
3048 STRING1 and STRING2 have length SIZE1 and SIZE2, respectively.
3050 RANGE is how far to scan while trying to match. RANGE = 0 means try
3051 only at STARTPOS; in general, the last start tried is STARTPOS +
3054 In REGS, return the indices of the virtual concatenation of STRING1
3055 and STRING2 that matched the entire BUFP->buffer and its contained
3058 Do not consider matching one past the index STOP in the virtual
3059 concatenation of STRING1 and STRING2.
3061 We return either the position in the strings at which the match was
3062 found, -1 if no match, or -2 if error (such as failure
3066 re_search_2 (bufp, string1, size1, string2, size2, startpos, range, regs, stop)
3067 struct re_pattern_buffer *bufp;
3068 const char *string1, *string2;
3072 struct re_registers *regs;
3076 register char *fastmap = bufp->fastmap;
3077 register char *translate = bufp->translate;
3078 int total_size = size1 + size2;
3079 int endpos = startpos + range;
3081 /* Check for out-of-range STARTPOS. */
3082 if (startpos < 0 || startpos > total_size)
3085 /* Fix up RANGE if it might eventually take us outside
3086 the virtual concatenation of STRING1 and STRING2. */
3088 range = -1 - startpos;
3089 else if (endpos > total_size)
3090 range = total_size - startpos;
3092 /* If the search isn't to be a backwards one, don't waste time in a
3093 search for a pattern that must be anchored. */
3094 if (bufp->used > 0 && (re_opcode_t) bufp->buffer[0] == begbuf && range > 0)
3102 /* Update the fastmap now if not correct already. */
3103 if (fastmap && !bufp->fastmap_accurate)
3104 if (re_compile_fastmap (bufp) == -2)
3107 /* Loop through the string, looking for a place to start matching. */
3110 /* If a fastmap is supplied, skip quickly over characters that
3111 cannot be the start of a match. If the pattern can match the
3112 null string, however, we don't need to skip characters; we want
3113 the first null string. */
3114 if (fastmap && startpos < total_size && !bufp->can_be_null)
3116 if (range > 0) /* Searching forwards. */
3118 register const char *d;
3119 register int lim = 0;
3122 if (startpos < size1 && startpos + range >= size1)
3123 lim = range - (size1 - startpos);
3125 d = (startpos >= size1 ? string2 - size1 : string1) + startpos;
3127 /* Written out as an if-else to avoid testing `translate'
3131 && !fastmap[(unsigned char)
3132 translate[(unsigned char) *d++]])
3135 while (range > lim && !fastmap[(unsigned char) *d++])
3138 startpos += irange - range;
3140 else /* Searching backwards. */
3142 register char c = (size1 == 0 || startpos >= size1
3143 ? string2[startpos - size1]
3144 : string1[startpos]);
3146 if (!fastmap[(unsigned char) TRANSLATE (c)])
3151 /* If can't match the null string, and that's all we have left, fail. */
3152 if (range >= 0 && startpos == total_size && fastmap
3153 && !bufp->can_be_null)
3156 val = re_match_2 (bufp, string1, size1, string2, size2,
3157 startpos, regs, stop);
3181 /* Declarations and macros for re_match_2. */
3183 static int bcmp_translate ();
3184 static boolean alt_match_null_string_p (),
3185 common_op_match_null_string_p (),
3186 group_match_null_string_p ();
3188 /* This converts PTR, a pointer into one of the search strings `string1'
3189 and `string2' into an offset from the beginning of that string. */
3190 #define POINTER_TO_OFFSET(ptr) \
3191 (FIRST_STRING_P (ptr) \
3192 ? ((regoff_t) ((ptr) - string1)) \
3193 : ((regoff_t) ((ptr) - string2 + size1)))
3195 /* Macros for dealing with the split strings in re_match_2. */
3197 #define MATCHING_IN_FIRST_STRING (dend == end_match_1)
3199 /* Call before fetching a character with *d. This switches over to
3200 string2 if necessary. */
3201 #define PREFETCH() \
3204 /* End of string2 => fail. */ \
3205 if (dend == end_match_2) \
3207 /* End of string1 => advance to string2. */ \
3209 dend = end_match_2; \
3213 /* Test if at very beginning or at very end of the virtual concatenation
3214 of `string1' and `string2'. If only one string, it's `string2'. */
3215 #define AT_STRINGS_BEG(d) ((d) == (size1 ? string1 : string2) || !size2)
3216 #define AT_STRINGS_END(d) ((d) == end2)
3219 /* Test if D points to a character which is word-constituent. We have
3220 two special cases to check for: if past the end of string1, look at
3221 the first character in string2; and if before the beginning of
3222 string2, look at the last character in string1. */
3223 #define WORDCHAR_P(d) \
3224 (SYNTAX ((d) == end1 ? *string2 \
3225 : (d) == string2 - 1 ? *(end1 - 1) : *(d)) \
3228 /* Test if the character before D and the one at D differ with respect
3229 to being word-constituent. */
3230 #define AT_WORD_BOUNDARY(d) \
3231 (AT_STRINGS_BEG (d) || AT_STRINGS_END (d) \
3232 || WORDCHAR_P (d - 1) != WORDCHAR_P (d))
3235 /* Free everything we malloc. */
3236 #ifdef MATCH_MAY_ALLOCATE
3238 #define FREE_VAR(var) if (var) free (var); var = NULL
3239 #define FREE_VARIABLES() \
3241 FREE_VAR (fail_stack.stack); \
3242 FREE_VAR (regstart); \
3243 FREE_VAR (regend); \
3244 FREE_VAR (old_regstart); \
3245 FREE_VAR (old_regend); \
3246 FREE_VAR (best_regstart); \
3247 FREE_VAR (best_regend); \
3248 FREE_VAR (reg_info); \
3249 FREE_VAR (reg_dummy); \
3250 FREE_VAR (reg_info_dummy); \
3252 #else /* not REGEX_MALLOC */
3253 /* Some MIPS systems (at least) want this to free alloca'd storage. */
3254 #define FREE_VARIABLES() alloca (0)
3255 #endif /* not REGEX_MALLOC */
3257 #define FREE_VARIABLES() /* Do nothing! */
3258 #endif /* not MATCH_MAY_ALLOCATE */
3260 /* These values must meet several constraints. They must not be valid
3261 register values; since we have a limit of 255 registers (because
3262 we use only one byte in the pattern for the register number), we can
3263 use numbers larger than 255. They must differ by 1, because of
3264 NUM_FAILURE_ITEMS above. And the value for the lowest register must
3265 be larger than the value for the highest register, so we do not try
3266 to actually save any registers when none are active. */
3267 #define NO_HIGHEST_ACTIVE_REG (1 << BYTEWIDTH)
3268 #define NO_LOWEST_ACTIVE_REG (NO_HIGHEST_ACTIVE_REG + 1)
3270 /* Matching routines. */
3272 #ifndef emacs /* Emacs never uses this. */
3273 /* re_match is like re_match_2 except it takes only a single string. */
3276 re_match (bufp, string, size, pos, regs)
3277 struct re_pattern_buffer *bufp;
3280 struct re_registers *regs;
3282 return re_match_2 (bufp, NULL, 0, string, size, pos, regs, size);
3284 #endif /* not emacs */
3287 /* re_match_2 matches the compiled pattern in BUFP against the
3288 the (virtual) concatenation of STRING1 and STRING2 (of length SIZE1
3289 and SIZE2, respectively). We start matching at POS, and stop
3292 If REGS is non-null and the `no_sub' field of BUFP is nonzero, we
3293 store offsets for the substring each group matched in REGS. See the
3294 documentation for exactly how many groups we fill.
3296 We return -1 if no match, -2 if an internal error (such as the
3297 failure stack overflowing). Otherwise, we return the length of the
3298 matched substring. */
3301 re_match_2 (bufp, string1, size1, string2, size2, pos, regs, stop)
3302 struct re_pattern_buffer *bufp;
3303 const char *string1, *string2;
3306 struct re_registers *regs;
3309 /* General temporaries. */
3313 /* Just past the end of the corresponding string. */
3314 const char *end1, *end2;
3316 /* Pointers into string1 and string2, just past the last characters in
3317 each to consider matching. */
3318 const char *end_match_1, *end_match_2;
3320 /* Where we are in the data, and the end of the current string. */
3321 const char *d, *dend;
3323 /* Where we are in the pattern, and the end of the pattern. */
3324 unsigned char *p = bufp->buffer;
3325 register unsigned char *pend = p + bufp->used;
3327 /* Mark the opcode just after a start_memory, so we can test for an
3328 empty subpattern when we get to the stop_memory. */
3329 unsigned char *just_past_start_mem = 0;
3331 /* We use this to map every character in the string. */
3332 char *translate = bufp->translate;
3334 /* Failure point stack. Each place that can handle a failure further
3335 down the line pushes a failure point on this stack. It consists of
3336 restart, regend, and reg_info for all registers corresponding to
3337 the subexpressions we're currently inside, plus the number of such
3338 registers, and, finally, two char *'s. The first char * is where
3339 to resume scanning the pattern; the second one is where to resume
3340 scanning the strings. If the latter is zero, the failure point is
3341 a ``dummy''; if a failure happens and the failure point is a dummy,
3342 it gets discarded and the next next one is tried. */
3343 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
3344 fail_stack_type fail_stack;
3347 static unsigned failure_id = 0;
3348 unsigned nfailure_points_pushed = 0, nfailure_points_popped = 0;
3351 /* We fill all the registers internally, independent of what we
3352 return, for use in backreferences. The number here includes
3353 an element for register zero. */
3354 unsigned num_regs = bufp->re_nsub + 1;
3356 /* The currently active registers. */
3357 unsigned lowest_active_reg = NO_LOWEST_ACTIVE_REG;
3358 unsigned highest_active_reg = NO_HIGHEST_ACTIVE_REG;
3360 /* Information on the contents of registers. These are pointers into
3361 the input strings; they record just what was matched (on this
3362 attempt) by a subexpression part of the pattern, that is, the
3363 regnum-th regstart pointer points to where in the pattern we began
3364 matching and the regnum-th regend points to right after where we
3365 stopped matching the regnum-th subexpression. (The zeroth register
3366 keeps track of what the whole pattern matches.) */
3367 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
3368 const char **regstart, **regend;
3371 /* If a group that's operated upon by a repetition operator fails to
3372 match anything, then the register for its start will need to be
3373 restored because it will have been set to wherever in the string we
3374 are when we last see its open-group operator. Similarly for a
3376 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
3377 const char **old_regstart, **old_regend;
3380 /* The is_active field of reg_info helps us keep track of which (possibly
3381 nested) subexpressions we are currently in. The matched_something
3382 field of reg_info[reg_num] helps us tell whether or not we have
3383 matched any of the pattern so far this time through the reg_num-th
3384 subexpression. These two fields get reset each time through any
3385 loop their register is in. */
3386 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
3387 register_info_type *reg_info;
3390 /* The following record the register info as found in the above
3391 variables when we find a match better than any we've seen before.
3392 This happens as we backtrack through the failure points, which in
3393 turn happens only if we have not yet matched the entire string. */
3394 unsigned best_regs_set = false;
3395 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
3396 const char **best_regstart, **best_regend;
3399 /* Logically, this is `best_regend[0]'. But we don't want to have to
3400 allocate space for that if we're not allocating space for anything
3401 else (see below). Also, we never need info about register 0 for
3402 any of the other register vectors, and it seems rather a kludge to
3403 treat `best_regend' differently than the rest. So we keep track of
3404 the end of the best match so far in a separate variable. We
3405 initialize this to NULL so that when we backtrack the first time
3406 and need to test it, it's not garbage. */
3407 const char *match_end = NULL;
3409 /* Used when we pop values we don't care about. */
3410 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
3411 const char **reg_dummy;
3412 register_info_type *reg_info_dummy;
3416 /* Counts the total number of registers pushed. */
3417 unsigned num_regs_pushed = 0;
3420 DEBUG_PRINT1 ("\n\nEntering re_match_2.\n");
3424 #ifdef MATCH_MAY_ALLOCATE
3425 /* Do not bother to initialize all the register variables if there are
3426 no groups in the pattern, as it takes a fair amount of time. If
3427 there are groups, we include space for register 0 (the whole
3428 pattern), even though we never use it, since it simplifies the
3429 array indexing. We should fix this. */
3432 regstart = REGEX_TALLOC (num_regs, const char *);
3433 regend = REGEX_TALLOC (num_regs, const char *);
3434 old_regstart = REGEX_TALLOC (num_regs, const char *);
3435 old_regend = REGEX_TALLOC (num_regs, const char *);
3436 best_regstart = REGEX_TALLOC (num_regs, const char *);
3437 best_regend = REGEX_TALLOC (num_regs, const char *);
3438 reg_info = REGEX_TALLOC (num_regs, register_info_type);
3439 reg_dummy = REGEX_TALLOC (num_regs, const char *);
3440 reg_info_dummy = REGEX_TALLOC (num_regs, register_info_type);
3442 if (!(regstart && regend && old_regstart && old_regend && reg_info
3443 && best_regstart && best_regend && reg_dummy && reg_info_dummy))
3449 #if defined (REGEX_MALLOC)
3452 /* We must initialize all our variables to NULL, so that
3453 `FREE_VARIABLES' doesn't try to free them. */
3454 regstart = regend = old_regstart = old_regend = best_regstart
3455 = best_regend = reg_dummy = NULL;
3456 reg_info = reg_info_dummy = (register_info_type *) NULL;
3458 #endif /* REGEX_MALLOC */
3459 #endif /* MATCH_MAY_ALLOCATE */
3461 /* The starting position is bogus. */
3462 if (pos < 0 || pos > size1 + size2)
3468 /* Initialize subexpression text positions to -1 to mark ones that no
3469 start_memory/stop_memory has been seen for. Also initialize the
3470 register information struct. */
3471 for (mcnt = 1; mcnt < num_regs; mcnt++)
3473 regstart[mcnt] = regend[mcnt]
3474 = old_regstart[mcnt] = old_regend[mcnt] = REG_UNSET_VALUE;
3476 REG_MATCH_NULL_STRING_P (reg_info[mcnt]) = MATCH_NULL_UNSET_VALUE;
3477 IS_ACTIVE (reg_info[mcnt]) = 0;
3478 MATCHED_SOMETHING (reg_info[mcnt]) = 0;
3479 EVER_MATCHED_SOMETHING (reg_info[mcnt]) = 0;
3482 /* We move `string1' into `string2' if the latter's empty -- but not if
3483 `string1' is null. */
3484 if (size2 == 0 && string1 != NULL)
3491 end1 = string1 + size1;
3492 end2 = string2 + size2;
3494 /* Compute where to stop matching, within the two strings. */
3497 end_match_1 = string1 + stop;
3498 end_match_2 = string2;
3503 end_match_2 = string2 + stop - size1;
3506 /* `p' scans through the pattern as `d' scans through the data.
3507 `dend' is the end of the input string that `d' points within. `d'
3508 is advanced into the following input string whenever necessary, but
3509 this happens before fetching; therefore, at the beginning of the
3510 loop, `d' can be pointing at the end of a string, but it cannot
3512 if (size1 > 0 && pos <= size1)
3519 d = string2 + pos - size1;
3523 DEBUG_PRINT1 ("The compiled pattern is: ");
3524 DEBUG_PRINT_COMPILED_PATTERN (bufp, p, pend);
3525 DEBUG_PRINT1 ("The string to match is: `");
3526 DEBUG_PRINT_DOUBLE_STRING (d, string1, size1, string2, size2);
3527 DEBUG_PRINT1 ("'\n");
3529 /* This loops over pattern commands. It exits by returning from the
3530 function if the match is complete, or it drops through if the match
3531 fails at this starting point in the input data. */
3534 DEBUG_PRINT2 ("\n0x%x: ", p);
3537 { /* End of pattern means we might have succeeded. */
3538 DEBUG_PRINT1 ("end of pattern ... ");
3540 /* If we haven't matched the entire string, and we want the
3541 longest match, try backtracking. */
3542 if (d != end_match_2)
3544 DEBUG_PRINT1 ("backtracking.\n");
3546 if (!FAIL_STACK_EMPTY ())
3547 { /* More failure points to try. */
3548 boolean same_str_p = (FIRST_STRING_P (match_end)
3549 == MATCHING_IN_FIRST_STRING);
3551 /* If exceeds best match so far, save it. */
3553 || (same_str_p && d > match_end)
3554 || (!same_str_p && !MATCHING_IN_FIRST_STRING))
3556 best_regs_set = true;
3559 DEBUG_PRINT1 ("\nSAVING match as best so far.\n");
3561 for (mcnt = 1; mcnt < num_regs; mcnt++)
3563 best_regstart[mcnt] = regstart[mcnt];
3564 best_regend[mcnt] = regend[mcnt];
3570 /* If no failure points, don't restore garbage. */
3571 else if (best_regs_set)
3574 /* Restore best match. It may happen that `dend ==
3575 end_match_1' while the restored d is in string2.
3576 For example, the pattern `x.*y.*z' against the
3577 strings `x-' and `y-z-', if the two strings are
3578 not consecutive in memory. */
3579 DEBUG_PRINT1 ("Restoring best registers.\n");
3582 dend = ((d >= string1 && d <= end1)
3583 ? end_match_1 : end_match_2);
3585 for (mcnt = 1; mcnt < num_regs; mcnt++)
3587 regstart[mcnt] = best_regstart[mcnt];
3588 regend[mcnt] = best_regend[mcnt];
3591 } /* d != end_match_2 */
3593 DEBUG_PRINT1 ("Accepting match.\n");
3595 /* If caller wants register contents data back, do it. */
3596 if (regs && !bufp->no_sub)
3598 /* Have the register data arrays been allocated? */
3599 if (bufp->regs_allocated == REGS_UNALLOCATED)
3600 { /* No. So allocate them with malloc. We need one
3601 extra element beyond `num_regs' for the `-1' marker
3603 regs->num_regs = MAX (RE_NREGS, num_regs + 1);
3604 regs->start = TALLOC (regs->num_regs, regoff_t);
3605 regs->end = TALLOC (regs->num_regs, regoff_t);
3606 if (regs->start == NULL || regs->end == NULL)
3608 bufp->regs_allocated = REGS_REALLOCATE;
3610 else if (bufp->regs_allocated == REGS_REALLOCATE)
3611 { /* Yes. If we need more elements than were already
3612 allocated, reallocate them. If we need fewer, just
3614 if (regs->num_regs < num_regs + 1)
3616 regs->num_regs = num_regs + 1;
3617 RETALLOC (regs->start, regs->num_regs, regoff_t);
3618 RETALLOC (regs->end, regs->num_regs, regoff_t);
3619 if (regs->start == NULL || regs->end == NULL)
3625 /* These braces fend off a "empty body in an else-statement"
3626 warning under GCC when assert expands to nothing. */
3627 assert (bufp->regs_allocated == REGS_FIXED);
3630 /* Convert the pointer data in `regstart' and `regend' to
3631 indices. Register zero has to be set differently,
3632 since we haven't kept track of any info for it. */
3633 if (regs->num_regs > 0)
3635 regs->start[0] = pos;
3636 regs->end[0] = (MATCHING_IN_FIRST_STRING
3637 ? ((regoff_t) (d - string1))
3638 : ((regoff_t) (d - string2 + size1)));
3641 /* Go through the first `min (num_regs, regs->num_regs)'
3642 registers, since that is all we initialized. */
3643 for (mcnt = 1; mcnt < MIN (num_regs, regs->num_regs); mcnt++)
3645 if (REG_UNSET (regstart[mcnt]) || REG_UNSET (regend[mcnt]))
3646 regs->start[mcnt] = regs->end[mcnt] = -1;
3650 = (regoff_t) POINTER_TO_OFFSET (regstart[mcnt]);
3652 = (regoff_t) POINTER_TO_OFFSET (regend[mcnt]);
3656 /* If the regs structure we return has more elements than
3657 were in the pattern, set the extra elements to -1. If
3658 we (re)allocated the registers, this is the case,
3659 because we always allocate enough to have at least one
3661 for (mcnt = num_regs; mcnt < regs->num_regs; mcnt++)
3662 regs->start[mcnt] = regs->end[mcnt] = -1;
3663 } /* regs && !bufp->no_sub */
3666 DEBUG_PRINT4 ("%u failure points pushed, %u popped (%u remain).\n",
3667 nfailure_points_pushed, nfailure_points_popped,
3668 nfailure_points_pushed - nfailure_points_popped);
3669 DEBUG_PRINT2 ("%u registers pushed.\n", num_regs_pushed);
3671 mcnt = d - pos - (MATCHING_IN_FIRST_STRING
3675 DEBUG_PRINT2 ("Returning %d from re_match_2.\n", mcnt);
3680 /* Otherwise match next pattern command. */
3681 #ifdef SWITCH_ENUM_BUG
3682 switch ((int) ((re_opcode_t) *p++))
3684 switch ((re_opcode_t) *p++)
3687 /* Ignore these. Used to ignore the n of succeed_n's which
3688 currently have n == 0. */
3690 DEBUG_PRINT1 ("EXECUTING no_op.\n");
3694 /* Match the next n pattern characters exactly. The following
3695 byte in the pattern defines n, and the n bytes after that
3696 are the characters to match. */
3699 DEBUG_PRINT2 ("EXECUTING exactn %d.\n", mcnt);
3701 /* This is written out as an if-else so we don't waste time
3702 testing `translate' inside the loop. */
3708 if (translate[(unsigned char) *d++] != (char) *p++)
3718 if (*d++ != (char) *p++) goto fail;
3722 SET_REGS_MATCHED ();
3726 /* Match any character except possibly a newline or a null. */
3728 DEBUG_PRINT1 ("EXECUTING anychar.\n");
3732 if ((!(bufp->syntax & RE_DOT_NEWLINE) && TRANSLATE (*d) == '\n')
3733 || (bufp->syntax & RE_DOT_NOT_NULL && TRANSLATE (*d) == '\000'))
3736 SET_REGS_MATCHED ();
3737 DEBUG_PRINT2 (" Matched `%d'.\n", *d);
3745 register unsigned char c;
3746 boolean not = (re_opcode_t) *(p - 1) == charset_not;
3748 DEBUG_PRINT2 ("EXECUTING charset%s.\n", not ? "_not" : "");
3751 c = TRANSLATE (*d); /* The character to match. */
3753 /* Cast to `unsigned' instead of `unsigned char' in case the
3754 bit list is a full 32 bytes long. */
3755 if (c < (unsigned) (*p * BYTEWIDTH)
3756 && p[1 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH)))
3761 if (!not) goto fail;
3763 SET_REGS_MATCHED ();
3769 /* The beginning of a group is represented by start_memory.
3770 The arguments are the register number in the next byte, and the
3771 number of groups inner to this one in the next. The text
3772 matched within the group is recorded (in the internal
3773 registers data structure) under the register number. */
3775 DEBUG_PRINT3 ("EXECUTING start_memory %d (%d):\n", *p, p[1]);
3777 /* Find out if this group can match the empty string. */
3778 p1 = p; /* To send to group_match_null_string_p. */
3780 if (REG_MATCH_NULL_STRING_P (reg_info[*p]) == MATCH_NULL_UNSET_VALUE)
3781 REG_MATCH_NULL_STRING_P (reg_info[*p])
3782 = group_match_null_string_p (&p1, pend, reg_info);
3784 /* Save the position in the string where we were the last time
3785 we were at this open-group operator in case the group is
3786 operated upon by a repetition operator, e.g., with `(a*)*b'
3787 against `ab'; then we want to ignore where we are now in
3788 the string in case this attempt to match fails. */
3789 old_regstart[*p] = REG_MATCH_NULL_STRING_P (reg_info[*p])
3790 ? REG_UNSET (regstart[*p]) ? d : regstart[*p]
3792 DEBUG_PRINT2 (" old_regstart: %d\n",
3793 POINTER_TO_OFFSET (old_regstart[*p]));
3796 DEBUG_PRINT2 (" regstart: %d\n", POINTER_TO_OFFSET (regstart[*p]));
3798 IS_ACTIVE (reg_info[*p]) = 1;
3799 MATCHED_SOMETHING (reg_info[*p]) = 0;
3801 /* This is the new highest active register. */
3802 highest_active_reg = *p;
3804 /* If nothing was active before, this is the new lowest active
3806 if (lowest_active_reg == NO_LOWEST_ACTIVE_REG)
3807 lowest_active_reg = *p;
3809 /* Move past the register number and inner group count. */
3811 just_past_start_mem = p;
3815 /* The stop_memory opcode represents the end of a group. Its
3816 arguments are the same as start_memory's: the register
3817 number, and the number of inner groups. */
3819 DEBUG_PRINT3 ("EXECUTING stop_memory %d (%d):\n", *p, p[1]);
3821 /* We need to save the string position the last time we were at
3822 this close-group operator in case the group is operated
3823 upon by a repetition operator, e.g., with `((a*)*(b*)*)*'
3824 against `aba'; then we want to ignore where we are now in
3825 the string in case this attempt to match fails. */
3826 old_regend[*p] = REG_MATCH_NULL_STRING_P (reg_info[*p])
3827 ? REG_UNSET (regend[*p]) ? d : regend[*p]
3829 DEBUG_PRINT2 (" old_regend: %d\n",
3830 POINTER_TO_OFFSET (old_regend[*p]));
3833 DEBUG_PRINT2 (" regend: %d\n", POINTER_TO_OFFSET (regend[*p]));
3835 /* This register isn't active anymore. */
3836 IS_ACTIVE (reg_info[*p]) = 0;
3838 /* If this was the only register active, nothing is active
3840 if (lowest_active_reg == highest_active_reg)
3842 lowest_active_reg = NO_LOWEST_ACTIVE_REG;
3843 highest_active_reg = NO_HIGHEST_ACTIVE_REG;
3846 { /* We must scan for the new highest active register, since
3847 it isn't necessarily one less than now: consider
3848 (a(b)c(d(e)f)g). When group 3 ends, after the f), the
3849 new highest active register is 1. */
3850 unsigned char r = *p - 1;
3851 while (r > 0 && !IS_ACTIVE (reg_info[r]))
3854 /* If we end up at register zero, that means that we saved
3855 the registers as the result of an `on_failure_jump', not
3856 a `start_memory', and we jumped to past the innermost
3857 `stop_memory'. For example, in ((.)*) we save
3858 registers 1 and 2 as a result of the *, but when we pop
3859 back to the second ), we are at the stop_memory 1.
3860 Thus, nothing is active. */
3863 lowest_active_reg = NO_LOWEST_ACTIVE_REG;
3864 highest_active_reg = NO_HIGHEST_ACTIVE_REG;
3867 highest_active_reg = r;
3870 /* If just failed to match something this time around with a
3871 group that's operated on by a repetition operator, try to
3872 force exit from the ``loop'', and restore the register
3873 information for this group that we had before trying this
3875 if ((!MATCHED_SOMETHING (reg_info[*p])
3876 || just_past_start_mem == p - 1)
3879 boolean is_a_jump_n = false;
3883 switch ((re_opcode_t) *p1++)
3887 case pop_failure_jump:
3888 case maybe_pop_jump:
3890 case dummy_failure_jump:
3891 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
3901 /* If the next operation is a jump backwards in the pattern
3902 to an on_failure_jump right before the start_memory
3903 corresponding to this stop_memory, exit from the loop
3904 by forcing a failure after pushing on the stack the
3905 on_failure_jump's jump in the pattern, and d. */
3906 if (mcnt < 0 && (re_opcode_t) *p1 == on_failure_jump
3907 && (re_opcode_t) p1[3] == start_memory && p1[4] == *p)
3909 /* If this group ever matched anything, then restore
3910 what its registers were before trying this last
3911 failed match, e.g., with `(a*)*b' against `ab' for
3912 regstart[1], and, e.g., with `((a*)*(b*)*)*'
3913 against `aba' for regend[3].
3915 Also restore the registers for inner groups for,
3916 e.g., `((a*)(b*))*' against `aba' (register 3 would
3917 otherwise get trashed). */
3919 if (EVER_MATCHED_SOMETHING (reg_info[*p]))
3923 EVER_MATCHED_SOMETHING (reg_info[*p]) = 0;
3925 /* Restore this and inner groups' (if any) registers. */
3926 for (r = *p; r < *p + *(p + 1); r++)
3928 regstart[r] = old_regstart[r];
3930 /* xx why this test? */
3931 if ((int) old_regend[r] >= (int) regstart[r])
3932 regend[r] = old_regend[r];
3936 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
3937 PUSH_FAILURE_POINT (p1 + mcnt, d, -2);
3943 /* Move past the register number and the inner group count. */
3948 /* \<digit> has been turned into a `duplicate' command which is
3949 followed by the numeric value of <digit> as the register number. */
3952 register const char *d2, *dend2;
3953 int regno = *p++; /* Get which register to match against. */
3954 DEBUG_PRINT2 ("EXECUTING duplicate %d.\n", regno);
3956 /* Can't back reference a group which we've never matched. */
3957 if (REG_UNSET (regstart[regno]) || REG_UNSET (regend[regno]))
3960 /* Where in input to try to start matching. */
3961 d2 = regstart[regno];
3963 /* Where to stop matching; if both the place to start and
3964 the place to stop matching are in the same string, then
3965 set to the place to stop, otherwise, for now have to use
3966 the end of the first string. */
3968 dend2 = ((FIRST_STRING_P (regstart[regno])
3969 == FIRST_STRING_P (regend[regno]))
3970 ? regend[regno] : end_match_1);
3973 /* If necessary, advance to next segment in register
3977 if (dend2 == end_match_2) break;
3978 if (dend2 == regend[regno]) break;
3980 /* End of string1 => advance to string2. */
3982 dend2 = regend[regno];
3984 /* At end of register contents => success */
3985 if (d2 == dend2) break;
3987 /* If necessary, advance to next segment in data. */
3990 /* How many characters left in this segment to match. */
3993 /* Want how many consecutive characters we can match in
3994 one shot, so, if necessary, adjust the count. */
3995 if (mcnt > dend2 - d2)
3998 /* Compare that many; failure if mismatch, else move
4001 ? bcmp_translate (d, d2, mcnt, translate)
4002 : bcmp (d, d2, mcnt))
4004 d += mcnt, d2 += mcnt;
4010 /* begline matches the empty string at the beginning of the string
4011 (unless `not_bol' is set in `bufp'), and, if
4012 `newline_anchor' is set, after newlines. */
4014 DEBUG_PRINT1 ("EXECUTING begline.\n");
4016 if (AT_STRINGS_BEG (d))
4018 if (!bufp->not_bol) break;
4020 else if (d[-1] == '\n' && bufp->newline_anchor)
4024 /* In all other cases, we fail. */
4028 /* endline is the dual of begline. */
4030 DEBUG_PRINT1 ("EXECUTING endline.\n");
4032 if (AT_STRINGS_END (d))
4034 if (!bufp->not_eol) break;
4037 /* We have to ``prefetch'' the next character. */
4038 else if ((d == end1 ? *string2 : *d) == '\n'
4039 && bufp->newline_anchor)
4046 /* Match at the very beginning of the data. */
4048 DEBUG_PRINT1 ("EXECUTING begbuf.\n");
4049 if (AT_STRINGS_BEG (d))
4054 /* Match at the very end of the data. */
4056 DEBUG_PRINT1 ("EXECUTING endbuf.\n");
4057 if (AT_STRINGS_END (d))
4062 /* on_failure_keep_string_jump is used to optimize `.*\n'. It
4063 pushes NULL as the value for the string on the stack. Then
4064 `pop_failure_point' will keep the current value for the
4065 string, instead of restoring it. To see why, consider
4066 matching `foo\nbar' against `.*\n'. The .* matches the foo;
4067 then the . fails against the \n. But the next thing we want
4068 to do is match the \n against the \n; if we restored the
4069 string value, we would be back at the foo.
4071 Because this is used only in specific cases, we don't need to
4072 check all the things that `on_failure_jump' does, to make
4073 sure the right things get saved on the stack. Hence we don't
4074 share its code. The only reason to push anything on the
4075 stack at all is that otherwise we would have to change
4076 `anychar's code to do something besides goto fail in this
4077 case; that seems worse than this. */
4078 case on_failure_keep_string_jump:
4079 DEBUG_PRINT1 ("EXECUTING on_failure_keep_string_jump");
4081 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4082 DEBUG_PRINT3 (" %d (to 0x%x):\n", mcnt, p + mcnt);
4084 PUSH_FAILURE_POINT (p + mcnt, NULL, -2);
4088 /* Uses of on_failure_jump:
4090 Each alternative starts with an on_failure_jump that points
4091 to the beginning of the next alternative. Each alternative
4092 except the last ends with a jump that in effect jumps past
4093 the rest of the alternatives. (They really jump to the
4094 ending jump of the following alternative, because tensioning
4095 these jumps is a hassle.)
4097 Repeats start with an on_failure_jump that points past both
4098 the repetition text and either the following jump or
4099 pop_failure_jump back to this on_failure_jump. */
4100 case on_failure_jump:
4102 DEBUG_PRINT1 ("EXECUTING on_failure_jump");
4104 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4105 DEBUG_PRINT3 (" %d (to 0x%x)", mcnt, p + mcnt);
4107 /* If this on_failure_jump comes right before a group (i.e.,
4108 the original * applied to a group), save the information
4109 for that group and all inner ones, so that if we fail back
4110 to this point, the group's information will be correct.
4111 For example, in \(a*\)*\1, we need the preceding group,
4112 and in \(\(a*\)b*\)\2, we need the inner group. */
4114 /* We can't use `p' to check ahead because we push
4115 a failure point to `p + mcnt' after we do this. */
4118 /* We need to skip no_op's before we look for the
4119 start_memory in case this on_failure_jump is happening as
4120 the result of a completed succeed_n, as in \(a\)\{1,3\}b\1
4122 while (p1 < pend && (re_opcode_t) *p1 == no_op)
4125 if (p1 < pend && (re_opcode_t) *p1 == start_memory)
4127 /* We have a new highest active register now. This will
4128 get reset at the start_memory we are about to get to,
4129 but we will have saved all the registers relevant to
4130 this repetition op, as described above. */
4131 highest_active_reg = *(p1 + 1) + *(p1 + 2);
4132 if (lowest_active_reg == NO_LOWEST_ACTIVE_REG)
4133 lowest_active_reg = *(p1 + 1);
4136 DEBUG_PRINT1 (":\n");
4137 PUSH_FAILURE_POINT (p + mcnt, d, -2);
4141 /* A smart repeat ends with `maybe_pop_jump'.
4142 We change it to either `pop_failure_jump' or `jump'. */
4143 case maybe_pop_jump:
4144 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4145 DEBUG_PRINT2 ("EXECUTING maybe_pop_jump %d.\n", mcnt);
4147 register unsigned char *p2 = p;
4149 /* Compare the beginning of the repeat with what in the
4150 pattern follows its end. If we can establish that there
4151 is nothing that they would both match, i.e., that we
4152 would have to backtrack because of (as in, e.g., `a*a')
4153 then we can change to pop_failure_jump, because we'll
4154 never have to backtrack.
4156 This is not true in the case of alternatives: in
4157 `(a|ab)*' we do need to backtrack to the `ab' alternative
4158 (e.g., if the string was `ab'). But instead of trying to
4159 detect that here, the alternative has put on a dummy
4160 failure point which is what we will end up popping. */
4162 /* Skip over open/close-group commands.
4163 If what follows this loop is a ...+ construct,
4164 look at what begins its body, since we will have to
4165 match at least one of that. */
4169 && ((re_opcode_t) *p2 == stop_memory
4170 || (re_opcode_t) *p2 == start_memory))
4172 else if (p2 + 6 < pend
4173 && (re_opcode_t) *p2 == dummy_failure_jump)
4180 /* p1[0] ... p1[2] are the `on_failure_jump' corresponding
4181 to the `maybe_finalize_jump' of this case. Examine what
4184 /* If we're at the end of the pattern, we can change. */
4187 /* Consider what happens when matching ":\(.*\)"
4188 against ":/". I don't really understand this code
4190 p[-3] = (unsigned char) pop_failure_jump;
4192 (" End of pattern: change to `pop_failure_jump'.\n");
4195 else if ((re_opcode_t) *p2 == exactn
4196 || (bufp->newline_anchor && (re_opcode_t) *p2 == endline))
4198 register unsigned char c
4199 = *p2 == (unsigned char) endline ? '\n' : p2[2];
4201 if ((re_opcode_t) p1[3] == exactn && p1[5] != c)
4203 p[-3] = (unsigned char) pop_failure_jump;
4204 DEBUG_PRINT3 (" %c != %c => pop_failure_jump.\n",
4208 else if ((re_opcode_t) p1[3] == charset
4209 || (re_opcode_t) p1[3] == charset_not)
4211 int not = (re_opcode_t) p1[3] == charset_not;
4213 if (c < (unsigned char) (p1[4] * BYTEWIDTH)
4214 && p1[5 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH)))
4217 /* `not' is equal to 1 if c would match, which means
4218 that we can't change to pop_failure_jump. */
4221 p[-3] = (unsigned char) pop_failure_jump;
4222 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
4226 else if ((re_opcode_t) *p2 == charset)
4228 register unsigned char c
4229 = *p2 == (unsigned char) endline ? '\n' : p2[2];
4231 if ((re_opcode_t) p1[3] == exactn
4232 && ! (p2[1] * BYTEWIDTH > p1[4]
4233 && (p2[1 + p1[4] / BYTEWIDTH]
4234 & (1 << (p1[4] % BYTEWIDTH)))))
4236 p[-3] = (unsigned char) pop_failure_jump;
4237 DEBUG_PRINT3 (" %c != %c => pop_failure_jump.\n",
4241 else if ((re_opcode_t) p1[3] == charset_not)
4244 /* We win if the charset_not inside the loop
4245 lists every character listed in the charset after. */
4246 for (idx = 0; idx < p2[1]; idx++)
4247 if (! (p2[2 + idx] == 0
4249 && ((p2[2 + idx] & ~ p1[5 + idx]) == 0))))
4254 p[-3] = (unsigned char) pop_failure_jump;
4255 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
4258 else if ((re_opcode_t) p1[3] == charset)
4261 /* We win if the charset inside the loop
4262 has no overlap with the one after the loop. */
4263 for (idx = 0; idx < p2[1] && idx < p1[4]; idx++)
4264 if ((p2[2 + idx] & p1[5 + idx]) != 0)
4267 if (idx == p2[1] || idx == p1[4])
4269 p[-3] = (unsigned char) pop_failure_jump;
4270 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
4275 p -= 2; /* Point at relative address again. */
4276 if ((re_opcode_t) p[-1] != pop_failure_jump)
4278 p[-1] = (unsigned char) jump;
4279 DEBUG_PRINT1 (" Match => jump.\n");
4280 goto unconditional_jump;
4282 /* Note fall through. */
4285 /* The end of a simple repeat has a pop_failure_jump back to
4286 its matching on_failure_jump, where the latter will push a
4287 failure point. The pop_failure_jump takes off failure
4288 points put on by this pop_failure_jump's matching
4289 on_failure_jump; we got through the pattern to here from the
4290 matching on_failure_jump, so didn't fail. */
4291 case pop_failure_jump:
4293 /* We need to pass separate storage for the lowest and
4294 highest registers, even though we don't care about the
4295 actual values. Otherwise, we will restore only one
4296 register from the stack, since lowest will == highest in
4297 `pop_failure_point'. */
4298 unsigned dummy_low_reg, dummy_high_reg;
4299 unsigned char *pdummy;
4302 DEBUG_PRINT1 ("EXECUTING pop_failure_jump.\n");
4303 POP_FAILURE_POINT (sdummy, pdummy,
4304 dummy_low_reg, dummy_high_reg,
4305 reg_dummy, reg_dummy, reg_info_dummy);
4307 /* Note fall through. */
4310 /* Unconditionally jump (without popping any failure points). */
4313 EXTRACT_NUMBER_AND_INCR (mcnt, p); /* Get the amount to jump. */
4314 DEBUG_PRINT2 ("EXECUTING jump %d ", mcnt);
4315 p += mcnt; /* Do the jump. */
4316 DEBUG_PRINT2 ("(to 0x%x).\n", p);
4320 /* We need this opcode so we can detect where alternatives end
4321 in `group_match_null_string_p' et al. */
4323 DEBUG_PRINT1 ("EXECUTING jump_past_alt.\n");
4324 goto unconditional_jump;
4327 /* Normally, the on_failure_jump pushes a failure point, which
4328 then gets popped at pop_failure_jump. We will end up at
4329 pop_failure_jump, also, and with a pattern of, say, `a+', we
4330 are skipping over the on_failure_jump, so we have to push
4331 something meaningless for pop_failure_jump to pop. */
4332 case dummy_failure_jump:
4333 DEBUG_PRINT1 ("EXECUTING dummy_failure_jump.\n");
4334 /* It doesn't matter what we push for the string here. What
4335 the code at `fail' tests is the value for the pattern. */
4336 PUSH_FAILURE_POINT (0, 0, -2);
4337 goto unconditional_jump;
4340 /* At the end of an alternative, we need to push a dummy failure
4341 point in case we are followed by a `pop_failure_jump', because
4342 we don't want the failure point for the alternative to be
4343 popped. For example, matching `(a|ab)*' against `aab'
4344 requires that we match the `ab' alternative. */
4345 case push_dummy_failure:
4346 DEBUG_PRINT1 ("EXECUTING push_dummy_failure.\n");
4347 /* See comments just above at `dummy_failure_jump' about the
4349 PUSH_FAILURE_POINT (0, 0, -2);
4352 /* Have to succeed matching what follows at least n times.
4353 After that, handle like `on_failure_jump'. */
4355 EXTRACT_NUMBER (mcnt, p + 2);
4356 DEBUG_PRINT2 ("EXECUTING succeed_n %d.\n", mcnt);
4359 /* Originally, this is how many times we HAVE to succeed. */
4364 STORE_NUMBER_AND_INCR (p, mcnt);
4365 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p, mcnt);
4369 DEBUG_PRINT2 (" Setting two bytes from 0x%x to no_op.\n", p+2);
4370 p[2] = (unsigned char) no_op;
4371 p[3] = (unsigned char) no_op;
4377 EXTRACT_NUMBER (mcnt, p + 2);
4378 DEBUG_PRINT2 ("EXECUTING jump_n %d.\n", mcnt);
4380 /* Originally, this is how many times we CAN jump. */
4384 STORE_NUMBER (p + 2, mcnt);
4385 goto unconditional_jump;
4387 /* If don't have to jump any more, skip over the rest of command. */
4394 DEBUG_PRINT1 ("EXECUTING set_number_at.\n");
4396 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4398 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4399 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p1, mcnt);
4400 STORE_NUMBER (p1, mcnt);
4405 DEBUG_PRINT1 ("EXECUTING wordbound.\n");
4406 if (AT_WORD_BOUNDARY (d))
4411 DEBUG_PRINT1 ("EXECUTING notwordbound.\n");
4412 if (AT_WORD_BOUNDARY (d))
4417 DEBUG_PRINT1 ("EXECUTING wordbeg.\n");
4418 if (WORDCHAR_P (d) && (AT_STRINGS_BEG (d) || !WORDCHAR_P (d - 1)))
4423 DEBUG_PRINT1 ("EXECUTING wordend.\n");
4424 if (!AT_STRINGS_BEG (d) && WORDCHAR_P (d - 1)
4425 && (!WORDCHAR_P (d) || AT_STRINGS_END (d)))
4432 DEBUG_PRINT1 ("EXECUTING before_dot.\n");
4433 if (PTR_CHAR_POS ((unsigned char *) d) >= point)
4438 DEBUG_PRINT1 ("EXECUTING at_dot.\n");
4439 if (PTR_CHAR_POS ((unsigned char *) d) != point)
4444 DEBUG_PRINT1 ("EXECUTING after_dot.\n");
4445 if (PTR_CHAR_POS ((unsigned char *) d) <= point)
4448 #else /* not emacs19 */
4450 DEBUG_PRINT1 ("EXECUTING at_dot.\n");
4451 if (PTR_CHAR_POS ((unsigned char *) d) + 1 != point)
4454 #endif /* not emacs19 */
4457 DEBUG_PRINT2 ("EXECUTING syntaxspec %d.\n", mcnt);
4462 DEBUG_PRINT1 ("EXECUTING Emacs wordchar.\n");
4466 /* Can't use *d++ here; SYNTAX may be an unsafe macro. */
4468 if (SYNTAX (d[-1]) != (enum syntaxcode) mcnt)
4470 SET_REGS_MATCHED ();
4474 DEBUG_PRINT2 ("EXECUTING notsyntaxspec %d.\n", mcnt);
4476 goto matchnotsyntax;
4479 DEBUG_PRINT1 ("EXECUTING Emacs notwordchar.\n");
4483 /* Can't use *d++ here; SYNTAX may be an unsafe macro. */
4485 if (SYNTAX (d[-1]) == (enum syntaxcode) mcnt)
4487 SET_REGS_MATCHED ();
4490 #else /* not emacs */
4492 DEBUG_PRINT1 ("EXECUTING non-Emacs wordchar.\n");
4494 if (!WORDCHAR_P (d))
4496 SET_REGS_MATCHED ();
4501 DEBUG_PRINT1 ("EXECUTING non-Emacs notwordchar.\n");
4505 SET_REGS_MATCHED ();
4508 #endif /* not emacs */
4513 continue; /* Successfully executed one pattern command; keep going. */
4516 /* We goto here if a matching operation fails. */
4518 if (!FAIL_STACK_EMPTY ())
4519 { /* A restart point is known. Restore to that state. */
4520 DEBUG_PRINT1 ("\nFAIL:\n");
4521 POP_FAILURE_POINT (d, p,
4522 lowest_active_reg, highest_active_reg,
4523 regstart, regend, reg_info);
4525 /* If this failure point is a dummy, try the next one. */
4529 /* If we failed to the end of the pattern, don't examine *p. */
4533 boolean is_a_jump_n = false;
4535 /* If failed to a backwards jump that's part of a repetition
4536 loop, need to pop this failure point and use the next one. */
4537 switch ((re_opcode_t) *p)
4541 case maybe_pop_jump:
4542 case pop_failure_jump:
4545 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4548 if ((is_a_jump_n && (re_opcode_t) *p1 == succeed_n)
4550 && (re_opcode_t) *p1 == on_failure_jump))
4558 if (d >= string1 && d <= end1)
4562 break; /* Matching at this starting point really fails. */
4566 goto restore_best_regs;
4570 return -1; /* Failure to match. */
4573 /* Subroutine definitions for re_match_2. */
4576 /* We are passed P pointing to a register number after a start_memory.
4578 Return true if the pattern up to the corresponding stop_memory can
4579 match the empty string, and false otherwise.
4581 If we find the matching stop_memory, sets P to point to one past its number.
4582 Otherwise, sets P to an undefined byte less than or equal to END.
4584 We don't handle duplicates properly (yet). */
4587 group_match_null_string_p (p, end, reg_info)
4588 unsigned char **p, *end;
4589 register_info_type *reg_info;
4592 /* Point to after the args to the start_memory. */
4593 unsigned char *p1 = *p + 2;
4597 /* Skip over opcodes that can match nothing, and return true or
4598 false, as appropriate, when we get to one that can't, or to the
4599 matching stop_memory. */
4601 switch ((re_opcode_t) *p1)
4603 /* Could be either a loop or a series of alternatives. */
4604 case on_failure_jump:
4606 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4608 /* If the next operation is not a jump backwards in the
4613 /* Go through the on_failure_jumps of the alternatives,
4614 seeing if any of the alternatives cannot match nothing.
4615 The last alternative starts with only a jump,
4616 whereas the rest start with on_failure_jump and end
4617 with a jump, e.g., here is the pattern for `a|b|c':
4619 /on_failure_jump/0/6/exactn/1/a/jump_past_alt/0/6
4620 /on_failure_jump/0/6/exactn/1/b/jump_past_alt/0/3
4623 So, we have to first go through the first (n-1)
4624 alternatives and then deal with the last one separately. */
4627 /* Deal with the first (n-1) alternatives, which start
4628 with an on_failure_jump (see above) that jumps to right
4629 past a jump_past_alt. */
4631 while ((re_opcode_t) p1[mcnt-3] == jump_past_alt)
4633 /* `mcnt' holds how many bytes long the alternative
4634 is, including the ending `jump_past_alt' and
4637 if (!alt_match_null_string_p (p1, p1 + mcnt - 3,
4641 /* Move to right after this alternative, including the
4645 /* Break if it's the beginning of an n-th alternative
4646 that doesn't begin with an on_failure_jump. */
4647 if ((re_opcode_t) *p1 != on_failure_jump)
4650 /* Still have to check that it's not an n-th
4651 alternative that starts with an on_failure_jump. */
4653 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4654 if ((re_opcode_t) p1[mcnt-3] != jump_past_alt)
4656 /* Get to the beginning of the n-th alternative. */
4662 /* Deal with the last alternative: go back and get number
4663 of the `jump_past_alt' just before it. `mcnt' contains
4664 the length of the alternative. */
4665 EXTRACT_NUMBER (mcnt, p1 - 2);
4667 if (!alt_match_null_string_p (p1, p1 + mcnt, reg_info))
4670 p1 += mcnt; /* Get past the n-th alternative. */
4676 assert (p1[1] == **p);
4682 if (!common_op_match_null_string_p (&p1, end, reg_info))
4685 } /* while p1 < end */
4688 } /* group_match_null_string_p */
4691 /* Similar to group_match_null_string_p, but doesn't deal with alternatives:
4692 It expects P to be the first byte of a single alternative and END one
4693 byte past the last. The alternative can contain groups. */
4696 alt_match_null_string_p (p, end, reg_info)
4697 unsigned char *p, *end;
4698 register_info_type *reg_info;
4701 unsigned char *p1 = p;
4705 /* Skip over opcodes that can match nothing, and break when we get
4706 to one that can't. */
4708 switch ((re_opcode_t) *p1)
4711 case on_failure_jump:
4713 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4718 if (!common_op_match_null_string_p (&p1, end, reg_info))
4721 } /* while p1 < end */
4724 } /* alt_match_null_string_p */
4727 /* Deals with the ops common to group_match_null_string_p and
4728 alt_match_null_string_p.
4730 Sets P to one after the op and its arguments, if any. */
4733 common_op_match_null_string_p (p, end, reg_info)
4734 unsigned char **p, *end;
4735 register_info_type *reg_info;
4740 unsigned char *p1 = *p;
4742 switch ((re_opcode_t) *p1++)
4762 assert (reg_no > 0 && reg_no <= MAX_REGNUM);
4763 ret = group_match_null_string_p (&p1, end, reg_info);
4765 /* Have to set this here in case we're checking a group which
4766 contains a group and a back reference to it. */
4768 if (REG_MATCH_NULL_STRING_P (reg_info[reg_no]) == MATCH_NULL_UNSET_VALUE)
4769 REG_MATCH_NULL_STRING_P (reg_info[reg_no]) = ret;
4775 /* If this is an optimized succeed_n for zero times, make the jump. */
4777 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4785 /* Get to the number of times to succeed. */
4787 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4792 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4800 if (!REG_MATCH_NULL_STRING_P (reg_info[*p1]))
4808 /* All other opcodes mean we cannot match the empty string. */
4814 } /* common_op_match_null_string_p */
4817 /* Return zero if TRANSLATE[S1] and TRANSLATE[S2] are identical for LEN
4818 bytes; nonzero otherwise. */
4821 bcmp_translate (s1, s2, len, translate)
4822 unsigned char *s1, *s2;
4826 register unsigned char *p1 = s1, *p2 = s2;
4829 if (translate[*p1++] != translate[*p2++]) return 1;
4835 /* Entry points for GNU code. */
4837 /* re_compile_pattern is the GNU regular expression compiler: it
4838 compiles PATTERN (of length SIZE) and puts the result in BUFP.
4839 Returns 0 if the pattern was valid, otherwise an error string.
4841 Assumes the `allocated' (and perhaps `buffer') and `translate' fields
4842 are set in BUFP on entry.
4844 We call regex_compile to do the actual compilation. */
4847 re_compile_pattern (pattern, length, bufp)
4848 const char *pattern;
4850 struct re_pattern_buffer *bufp;
4854 /* GNU code is written to assume at least RE_NREGS registers will be set
4855 (and at least one extra will be -1). */
4856 bufp->regs_allocated = REGS_UNALLOCATED;
4858 /* And GNU code determines whether or not to get register information
4859 by passing null for the REGS argument to re_match, etc., not by
4863 /* Match anchors at newline. */
4864 bufp->newline_anchor = 1;
4866 ret = regex_compile (pattern, length, re_syntax_options, bufp);
4868 return re_error_msg[(int) ret];
4871 /* Entry points compatible with 4.2 BSD regex library. We don't define
4872 them if this is an Emacs or POSIX compilation. */
4874 #if !defined (emacs) && !defined (_POSIX_SOURCE)
4876 /* BSD has one and only one pattern buffer. */
4877 static struct re_pattern_buffer re_comp_buf;
4887 if (!re_comp_buf.buffer)
4888 return "No previous regular expression";
4892 if (!re_comp_buf.buffer)
4894 re_comp_buf.buffer = (unsigned char *) malloc (200);
4895 if (re_comp_buf.buffer == NULL)
4896 return "Memory exhausted";
4897 re_comp_buf.allocated = 200;
4899 re_comp_buf.fastmap = (char *) malloc (1 << BYTEWIDTH);
4900 if (re_comp_buf.fastmap == NULL)
4901 return "Memory exhausted";
4904 /* Since `re_exec' always passes NULL for the `regs' argument, we
4905 don't need to initialize the pattern buffer fields which affect it. */
4907 /* Match anchors at newlines. */
4908 re_comp_buf.newline_anchor = 1;
4910 ret = regex_compile (s, strlen (s), re_syntax_options, &re_comp_buf);
4912 /* Yes, we're discarding `const' here. */
4913 return (char *) re_error_msg[(int) ret];
4921 const int len = strlen (s);
4923 0 <= re_search (&re_comp_buf, s, len, 0, len, (struct re_registers *) 0);
4925 #endif /* not emacs and not _POSIX_SOURCE */
4927 /* POSIX.2 functions. Don't define these for Emacs. */
4931 /* regcomp takes a regular expression as a string and compiles it.
4933 PREG is a regex_t *. We do not expect any fields to be initialized,
4934 since POSIX says we shouldn't. Thus, we set
4936 `buffer' to the compiled pattern;
4937 `used' to the length of the compiled pattern;
4938 `syntax' to RE_SYNTAX_POSIX_EXTENDED if the
4939 REG_EXTENDED bit in CFLAGS is set; otherwise, to
4940 RE_SYNTAX_POSIX_BASIC;
4941 `newline_anchor' to REG_NEWLINE being set in CFLAGS;
4942 `fastmap' and `fastmap_accurate' to zero;
4943 `re_nsub' to the number of subexpressions in PATTERN.
4945 PATTERN is the address of the pattern string.
4947 CFLAGS is a series of bits which affect compilation.
4949 If REG_EXTENDED is set, we use POSIX extended syntax; otherwise, we
4950 use POSIX basic syntax.
4952 If REG_NEWLINE is set, then . and [^...] don't match newline.
4953 Also, regexec will try a match beginning after every newline.
4955 If REG_ICASE is set, then we considers upper- and lowercase
4956 versions of letters to be equivalent when matching.
4958 If REG_NOSUB is set, then when PREG is passed to regexec, that
4959 routine will report only success or failure, and nothing about the
4962 It returns 0 if it succeeds, nonzero if it doesn't. (See regex.h for
4963 the return codes and their meanings.) */
4966 regcomp (preg, pattern, cflags)
4968 const char *pattern;
4973 = (cflags & REG_EXTENDED) ?
4974 RE_SYNTAX_POSIX_EXTENDED : RE_SYNTAX_POSIX_BASIC;
4976 /* regex_compile will allocate the space for the compiled pattern. */
4978 preg->allocated = 0;
4981 /* Don't bother to use a fastmap when searching. This simplifies the
4982 REG_NEWLINE case: if we used a fastmap, we'd have to put all the
4983 characters after newlines into the fastmap. This way, we just try
4987 if (cflags & REG_ICASE)
4991 preg->translate = (char *) malloc (CHAR_SET_SIZE);
4992 if (preg->translate == NULL)
4993 return (int) REG_ESPACE;
4995 /* Map uppercase characters to corresponding lowercase ones. */
4996 for (i = 0; i < CHAR_SET_SIZE; i++)
4997 preg->translate[i] = ISUPPER (i) ? tolower (i) : i;
5000 preg->translate = NULL;
5002 /* If REG_NEWLINE is set, newlines are treated differently. */
5003 if (cflags & REG_NEWLINE)
5004 { /* REG_NEWLINE implies neither . nor [^...] match newline. */
5005 syntax &= ~RE_DOT_NEWLINE;
5006 syntax |= RE_HAT_LISTS_NOT_NEWLINE;
5007 /* It also changes the matching behavior. */
5008 preg->newline_anchor = 1;
5011 preg->newline_anchor = 0;
5013 preg->no_sub = !!(cflags & REG_NOSUB);
5015 /* POSIX says a null character in the pattern terminates it, so we
5016 can use strlen here in compiling the pattern. */
5017 ret = regex_compile (pattern, strlen (pattern), syntax, preg);
5019 /* POSIX doesn't distinguish between an unmatched open-group and an
5020 unmatched close-group: both are REG_EPAREN. */
5021 if (ret == REG_ERPAREN) ret = REG_EPAREN;
5027 /* regexec searches for a given pattern, specified by PREG, in the
5030 If NMATCH is zero or REG_NOSUB was set in the cflags argument to
5031 `regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at
5032 least NMATCH elements, and we set them to the offsets of the
5033 corresponding matched substrings.
5035 EFLAGS specifies `execution flags' which affect matching: if
5036 REG_NOTBOL is set, then ^ does not match at the beginning of the
5037 string; if REG_NOTEOL is set, then $ does not match at the end.
5039 We return 0 if we find a match and REG_NOMATCH if not. */
5042 regexec (preg, string, nmatch, pmatch, eflags)
5043 const regex_t *preg;
5046 regmatch_t pmatch[];
5050 struct re_registers regs;
5051 regex_t private_preg;
5052 int len = strlen (string);
5053 boolean want_reg_info = !preg->no_sub && nmatch > 0;
5055 private_preg = *preg;
5057 private_preg.not_bol = !!(eflags & REG_NOTBOL);
5058 private_preg.not_eol = !!(eflags & REG_NOTEOL);
5060 /* The user has told us exactly how many registers to return
5061 information about, via `nmatch'. We have to pass that on to the
5062 matching routines. */
5063 private_preg.regs_allocated = REGS_FIXED;
5067 regs.num_regs = nmatch;
5068 regs.start = TALLOC (nmatch, regoff_t);
5069 regs.end = TALLOC (nmatch, regoff_t);
5070 if (regs.start == NULL || regs.end == NULL)
5071 return (int) REG_NOMATCH;
5074 /* Perform the searching operation. */
5075 ret = re_search (&private_preg, string, len,
5076 /* start: */ 0, /* range: */ len,
5077 want_reg_info ? ®s : (struct re_registers *) 0);
5079 /* Copy the register information to the POSIX structure. */
5086 for (r = 0; r < nmatch; r++)
5088 pmatch[r].rm_so = regs.start[r];
5089 pmatch[r].rm_eo = regs.end[r];
5093 /* If we needed the temporary register info, free the space now. */
5098 /* We want zero return to mean success, unlike `re_search'. */
5099 return ret >= 0 ? (int) REG_NOERROR : (int) REG_NOMATCH;
5103 /* Returns a message corresponding to an error code, ERRCODE, returned
5104 from either regcomp or regexec. We don't use PREG here. */
5107 regerror (errcode, preg, errbuf, errbuf_size)
5109 const regex_t *preg;
5117 || errcode >= (sizeof (re_error_msg) / sizeof (re_error_msg[0])))
5118 /* Only error codes returned by the rest of the code should be passed
5119 to this routine. If we are given anything else, or if other regex
5120 code generates an invalid error code, then the program has a bug.
5121 Dump core so we can fix it. */
5124 msg = re_error_msg[errcode];
5126 /* POSIX doesn't require that we do anything in this case, but why
5131 msg_size = strlen (msg) + 1; /* Includes the null. */
5133 if (errbuf_size != 0)
5135 if (msg_size > errbuf_size)
5137 strncpy (errbuf, msg, errbuf_size - 1);
5138 errbuf[errbuf_size - 1] = 0;
5141 strcpy (errbuf, msg);
5148 /* Free dynamically allocated space used by PREG. */
5154 if (preg->buffer != NULL)
5155 free (preg->buffer);
5156 preg->buffer = NULL;
5158 preg->allocated = 0;
5161 if (preg->fastmap != NULL)
5162 free (preg->fastmap);
5163 preg->fastmap = NULL;
5164 preg->fastmap_accurate = 0;
5166 if (preg->translate != NULL)
5167 free (preg->translate);
5168 preg->translate = NULL;
5171 #endif /* not emacs */
5175 make-backup-files: t
5177 trim-versions-without-asking: nil