1 /* Extended regular expression matching and search library,
3 (Implements POSIX draft P10003.2/D11.2, except for
4 internationalization features.)
6 Copyright (C) 1993, 1994, 1995 Free Software Foundation, Inc.
8 This program is free software; you can redistribute it and/or modify
9 it under the terms of the GNU General Public License as published by
10 the Free Software Foundation; either version 2, or (at your option)
13 This program is distributed in the hope that it will be useful,
14 but WITHOUT ANY WARRANTY; without even the implied warranty of
15 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
16 GNU General Public License for more details.
18 You should have received a copy of the GNU General Public License
19 along with this program; if not, write to the Free Software
20 Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. */
22 /* AIX requires this to be the first thing in the file. */
23 #if defined (_AIX) && !defined (REGEX_MALLOC)
33 /* We need this for `regex.h', and perhaps for the Emacs include files. */
34 #include <sys/types.h>
36 /* This is for other GNU distributions with internationalized messages. */
37 #if HAVE_LIBINTL_H || defined (_LIBC)
40 # define gettext(msgid) (msgid)
43 /* The `emacs' switch turns on certain matching commands
44 that make sense only in Emacs. */
53 /* If we are not linking with Emacs proper,
54 we can't use the relocating allocator
55 even if config.h says that we can. */
58 #if defined (STDC_HEADERS) || defined (_LIBC)
65 /* We used to test for `BSTRING' here, but only GCC and Emacs define
66 `BSTRING', as far as I know, and neither of them use this code. */
67 #ifndef INHIBIT_STRING_HEADER
68 #if HAVE_STRING_H || STDC_HEADERS || defined (_LIBC)
71 #define bcmp(s1, s2, n) memcmp ((s1), (s2), (n))
74 #define bcopy(s, d, n) memcpy ((d), (s), (n))
77 #define bzero(s, n) memset ((s), 0, (n))
84 /* Define the syntax stuff for \<, \>, etc. */
86 /* This must be nonzero for the wordchar and notwordchar pattern
87 commands in re_match_2. */
92 #ifdef SWITCH_ENUM_BUG
93 #define SWITCH_ENUM_CAST(x) ((int)(x))
95 #define SWITCH_ENUM_CAST(x) (x)
100 extern char *re_syntax_table;
102 #else /* not SYNTAX_TABLE */
104 /* How many characters in the character set. */
105 #define CHAR_SET_SIZE 256
107 static char re_syntax_table[CHAR_SET_SIZE];
118 bzero (re_syntax_table, sizeof re_syntax_table);
120 for (c = 'a'; c <= 'z'; c++)
121 re_syntax_table[c] = Sword;
123 for (c = 'A'; c <= 'Z'; c++)
124 re_syntax_table[c] = Sword;
126 for (c = '0'; c <= '9'; c++)
127 re_syntax_table[c] = Sword;
129 re_syntax_table['_'] = Sword;
134 #endif /* not SYNTAX_TABLE */
136 #define SYNTAX(c) re_syntax_table[c]
138 #endif /* not emacs */
140 /* Get the interface, including the syntax bits. */
143 /* isalpha etc. are used for the character classes. */
146 /* Jim Meyering writes:
148 "... Some ctype macros are valid only for character codes that
149 isascii says are ASCII (SGI's IRIX-4.0.5 is one such system --when
150 using /bin/cc or gcc but without giving an ansi option). So, all
151 ctype uses should be through macros like ISPRINT... If
152 STDC_HEADERS is defined, then autoconf has verified that the ctype
153 macros don't need to be guarded with references to isascii. ...
154 Defining isascii to 1 should let any compiler worth its salt
155 eliminate the && through constant folding." */
157 #if defined (STDC_HEADERS) || (!defined (isascii) && !defined (HAVE_ISASCII))
160 #define ISASCII(c) isascii(c)
164 #define ISBLANK(c) (ISASCII (c) && isblank (c))
166 #define ISBLANK(c) ((c) == ' ' || (c) == '\t')
169 #define ISGRAPH(c) (ISASCII (c) && isgraph (c))
171 #define ISGRAPH(c) (ISASCII (c) && isprint (c) && !isspace (c))
174 #define ISPRINT(c) (ISASCII (c) && isprint (c))
175 #define ISDIGIT(c) (ISASCII (c) && isdigit (c))
176 #define ISALNUM(c) (ISASCII (c) && isalnum (c))
177 #define ISALPHA(c) (ISASCII (c) && isalpha (c))
178 #define ISCNTRL(c) (ISASCII (c) && iscntrl (c))
179 #define ISLOWER(c) (ISASCII (c) && islower (c))
180 #define ISPUNCT(c) (ISASCII (c) && ispunct (c))
181 #define ISSPACE(c) (ISASCII (c) && isspace (c))
182 #define ISUPPER(c) (ISASCII (c) && isupper (c))
183 #define ISXDIGIT(c) (ISASCII (c) && isxdigit (c))
189 /* We remove any previous definition of `SIGN_EXTEND_CHAR',
190 since ours (we hope) works properly with all combinations of
191 machines, compilers, `char' and `unsigned char' argument types.
192 (Per Bothner suggested the basic approach.) */
193 #undef SIGN_EXTEND_CHAR
195 #define SIGN_EXTEND_CHAR(c) ((signed char) (c))
196 #else /* not __STDC__ */
197 /* As in Harbison and Steele. */
198 #define SIGN_EXTEND_CHAR(c) ((((unsigned char) (c)) ^ 128) - 128)
201 /* Should we use malloc or alloca? If REGEX_MALLOC is not defined, we
202 use `alloca' instead of `malloc'. This is because using malloc in
203 re_search* or re_match* could cause memory leaks when C-g is used in
204 Emacs; also, malloc is slower and causes storage fragmentation. On
205 the other hand, malloc is more portable, and easier to debug.
207 Because we sometimes use alloca, some routines have to be macros,
208 not functions -- `alloca'-allocated space disappears at the end of the
209 function it is called in. */
213 #define REGEX_ALLOCATE malloc
214 #define REGEX_REALLOCATE(source, osize, nsize) realloc (source, nsize)
215 #define REGEX_FREE free
217 #else /* not REGEX_MALLOC */
219 /* Emacs already defines alloca, sometimes. */
222 /* Make alloca work the best possible way. */
224 #define alloca __builtin_alloca
225 #else /* not __GNUC__ */
228 #else /* not __GNUC__ or HAVE_ALLOCA_H */
229 #ifndef _AIX /* Already did AIX, up at the top. */
231 #endif /* not _AIX */
232 #endif /* not HAVE_ALLOCA_H */
233 #endif /* not __GNUC__ */
235 #endif /* not alloca */
237 #define REGEX_ALLOCATE alloca
239 /* Assumes a `char *destination' variable. */
240 #define REGEX_REALLOCATE(source, osize, nsize) \
241 (destination = (char *) alloca (nsize), \
242 bcopy (source, destination, osize), \
245 /* No need to do anything to free, after alloca. */
246 #define REGEX_FREE(arg) ((void)0) /* Do nothing! But inhibit gcc warning. */
248 #endif /* not REGEX_MALLOC */
250 /* Define how to allocate the failure stack. */
253 #define REGEX_ALLOCATE_STACK(size) \
254 r_alloc (&failure_stack_ptr, (size))
255 #define REGEX_REALLOCATE_STACK(source, osize, nsize) \
256 r_re_alloc (&failure_stack_ptr, (nsize))
257 #define REGEX_FREE_STACK(ptr) \
258 r_alloc_free (&failure_stack_ptr)
260 #else /* not REL_ALLOC */
264 #define REGEX_ALLOCATE_STACK malloc
265 #define REGEX_REALLOCATE_STACK(source, osize, nsize) realloc (source, nsize)
266 #define REGEX_FREE_STACK free
268 #else /* not REGEX_MALLOC */
270 #define REGEX_ALLOCATE_STACK alloca
272 #define REGEX_REALLOCATE_STACK(source, osize, nsize) \
273 REGEX_REALLOCATE (source, osize, nsize)
274 /* No need to explicitly free anything. */
275 #define REGEX_FREE_STACK(arg)
277 #endif /* not REGEX_MALLOC */
278 #endif /* not REL_ALLOC */
281 /* True if `size1' is non-NULL and PTR is pointing anywhere inside
282 `string1' or just past its end. This works if PTR is NULL, which is
284 #define FIRST_STRING_P(ptr) \
285 (size1 && string1 <= (ptr) && (ptr) <= string1 + size1)
287 /* (Re)Allocate N items of type T using malloc, or fail. */
288 #define TALLOC(n, t) ((t *) malloc ((n) * sizeof (t)))
289 #define RETALLOC(addr, n, t) ((addr) = (t *) realloc (addr, (n) * sizeof (t)))
290 #define RETALLOC_IF(addr, n, t) \
291 if (addr) RETALLOC((addr), (n), t); else (addr) = TALLOC ((n), t)
292 #define REGEX_TALLOC(n, t) ((t *) REGEX_ALLOCATE ((n) * sizeof (t)))
294 #define BYTEWIDTH 8 /* In bits. */
296 #define STREQ(s1, s2) ((strcmp (s1, s2) == 0))
300 #define MAX(a, b) ((a) > (b) ? (a) : (b))
301 #define MIN(a, b) ((a) < (b) ? (a) : (b))
303 typedef char boolean;
307 static int re_match_2_internal ();
309 /* These are the command codes that appear in compiled regular
310 expressions. Some opcodes are followed by argument bytes. A
311 command code can specify any interpretation whatsoever for its
312 arguments. Zero bytes may appear in the compiled regular expression. */
318 /* Succeed right away--no more backtracking. */
321 /* Followed by one byte giving n, then by n literal bytes. */
324 /* Matches any (more or less) character. */
327 /* Matches any one char belonging to specified set. First
328 following byte is number of bitmap bytes. Then come bytes
329 for a bitmap saying which chars are in. Bits in each byte
330 are ordered low-bit-first. A character is in the set if its
331 bit is 1. A character too large to have a bit in the map is
332 automatically not in the set. */
335 /* Same parameters as charset, but match any character that is
336 not one of those specified. */
339 /* Start remembering the text that is matched, for storing in a
340 register. Followed by one byte with the register number, in
341 the range 0 to one less than the pattern buffer's re_nsub
342 field. Then followed by one byte with the number of groups
343 inner to this one. (This last has to be part of the
344 start_memory only because we need it in the on_failure_jump
348 /* Stop remembering the text that is matched and store it in a
349 memory register. Followed by one byte with the register
350 number, in the range 0 to one less than `re_nsub' in the
351 pattern buffer, and one byte with the number of inner groups,
352 just like `start_memory'. (We need the number of inner
353 groups here because we don't have any easy way of finding the
354 corresponding start_memory when we're at a stop_memory.) */
357 /* Match a duplicate of something remembered. Followed by one
358 byte containing the register number. */
361 /* Fail unless at beginning of line. */
364 /* Fail unless at end of line. */
367 /* Succeeds if at beginning of buffer (if emacs) or at beginning
368 of string to be matched (if not). */
371 /* Analogously, for end of buffer/string. */
374 /* Followed by two byte relative address to which to jump. */
377 /* Same as jump, but marks the end of an alternative. */
380 /* Followed by two-byte relative address of place to resume at
381 in case of failure. */
384 /* Like on_failure_jump, but pushes a placeholder instead of the
385 current string position when executed. */
386 on_failure_keep_string_jump,
388 /* Throw away latest failure point and then jump to following
389 two-byte relative address. */
392 /* Change to pop_failure_jump if know won't have to backtrack to
393 match; otherwise change to jump. This is used to jump
394 back to the beginning of a repeat. If what follows this jump
395 clearly won't match what the repeat does, such that we can be
396 sure that there is no use backtracking out of repetitions
397 already matched, then we change it to a pop_failure_jump.
398 Followed by two-byte address. */
401 /* Jump to following two-byte address, and push a dummy failure
402 point. This failure point will be thrown away if an attempt
403 is made to use it for a failure. A `+' construct makes this
404 before the first repeat. Also used as an intermediary kind
405 of jump when compiling an alternative. */
408 /* Push a dummy failure point and continue. Used at the end of
412 /* Followed by two-byte relative address and two-byte number n.
413 After matching N times, jump to the address upon failure. */
416 /* Followed by two-byte relative address, and two-byte number n.
417 Jump to the address N times, then fail. */
420 /* Set the following two-byte relative address to the
421 subsequent two-byte number. The address *includes* the two
425 wordchar, /* Matches any word-constituent character. */
426 notwordchar, /* Matches any char that is not a word-constituent. */
428 wordbeg, /* Succeeds if at word beginning. */
429 wordend, /* Succeeds if at word end. */
431 wordbound, /* Succeeds if at a word boundary. */
432 notwordbound /* Succeeds if not at a word boundary. */
435 ,before_dot, /* Succeeds if before point. */
436 at_dot, /* Succeeds if at point. */
437 after_dot, /* Succeeds if after point. */
439 /* Matches any character whose syntax is specified. Followed by
440 a byte which contains a syntax code, e.g., Sword. */
443 /* Matches any character whose syntax is not that specified. */
448 /* Common operations on the compiled pattern. */
450 /* Store NUMBER in two contiguous bytes starting at DESTINATION. */
452 #define STORE_NUMBER(destination, number) \
454 (destination)[0] = (number) & 0377; \
455 (destination)[1] = (number) >> 8; \
458 /* Same as STORE_NUMBER, except increment DESTINATION to
459 the byte after where the number is stored. Therefore, DESTINATION
460 must be an lvalue. */
462 #define STORE_NUMBER_AND_INCR(destination, number) \
464 STORE_NUMBER (destination, number); \
465 (destination) += 2; \
468 /* Put into DESTINATION a number stored in two contiguous bytes starting
471 #define EXTRACT_NUMBER(destination, source) \
473 (destination) = *(source) & 0377; \
474 (destination) += SIGN_EXTEND_CHAR (*((source) + 1)) << 8; \
479 extract_number (dest, source)
481 unsigned char *source;
483 int temp = SIGN_EXTEND_CHAR (*(source + 1));
484 *dest = *source & 0377;
488 #ifndef EXTRACT_MACROS /* To debug the macros. */
489 #undef EXTRACT_NUMBER
490 #define EXTRACT_NUMBER(dest, src) extract_number (&dest, src)
491 #endif /* not EXTRACT_MACROS */
495 /* Same as EXTRACT_NUMBER, except increment SOURCE to after the number.
496 SOURCE must be an lvalue. */
498 #define EXTRACT_NUMBER_AND_INCR(destination, source) \
500 EXTRACT_NUMBER (destination, source); \
506 extract_number_and_incr (destination, source)
508 unsigned char **source;
510 extract_number (destination, *source);
514 #ifndef EXTRACT_MACROS
515 #undef EXTRACT_NUMBER_AND_INCR
516 #define EXTRACT_NUMBER_AND_INCR(dest, src) \
517 extract_number_and_incr (&dest, &src)
518 #endif /* not EXTRACT_MACROS */
522 /* If DEBUG is defined, Regex prints many voluminous messages about what
523 it is doing (if the variable `debug' is nonzero). If linked with the
524 main program in `iregex.c', you can enter patterns and strings
525 interactively. And if linked with the main program in `main.c' and
526 the other test files, you can run the already-written tests. */
530 /* We use standard I/O for debugging. */
533 /* It is useful to test things that ``must'' be true when debugging. */
536 static int debug = 0;
538 #define DEBUG_STATEMENT(e) e
539 #define DEBUG_PRINT1(x) if (debug) printf (x)
540 #define DEBUG_PRINT2(x1, x2) if (debug) printf (x1, x2)
541 #define DEBUG_PRINT3(x1, x2, x3) if (debug) printf (x1, x2, x3)
542 #define DEBUG_PRINT4(x1, x2, x3, x4) if (debug) printf (x1, x2, x3, x4)
543 #define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) \
544 if (debug) print_partial_compiled_pattern (s, e)
545 #define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) \
546 if (debug) print_double_string (w, s1, sz1, s2, sz2)
549 /* Print the fastmap in human-readable form. */
552 print_fastmap (fastmap)
555 unsigned was_a_range = 0;
558 while (i < (1 << BYTEWIDTH))
564 while (i < (1 << BYTEWIDTH) && fastmap[i])
580 /* Print a compiled pattern string in human-readable form, starting at
581 the START pointer into it and ending just before the pointer END. */
584 print_partial_compiled_pattern (start, end)
585 unsigned char *start;
589 unsigned char *p = start;
590 unsigned char *pend = end;
598 /* Loop over pattern commands. */
601 printf ("%d:\t", p - start);
603 switch ((re_opcode_t) *p++)
611 printf ("/exactn/%d", mcnt);
622 printf ("/start_memory/%d/%d", mcnt, *p++);
627 printf ("/stop_memory/%d/%d", mcnt, *p++);
631 printf ("/duplicate/%d", *p++);
641 register int c, last = -100;
642 register int in_range = 0;
644 printf ("/charset [%s",
645 (re_opcode_t) *(p - 1) == charset_not ? "^" : "");
647 assert (p + *p < pend);
649 for (c = 0; c < 256; c++)
651 && (p[1 + (c/8)] & (1 << (c % 8))))
653 /* Are we starting a range? */
654 if (last + 1 == c && ! in_range)
659 /* Have we broken a range? */
660 else if (last + 1 != c && in_range)
689 case on_failure_jump:
690 extract_number_and_incr (&mcnt, &p);
691 printf ("/on_failure_jump to %d", p + mcnt - start);
694 case on_failure_keep_string_jump:
695 extract_number_and_incr (&mcnt, &p);
696 printf ("/on_failure_keep_string_jump to %d", p + mcnt - start);
699 case dummy_failure_jump:
700 extract_number_and_incr (&mcnt, &p);
701 printf ("/dummy_failure_jump to %d", p + mcnt - start);
704 case push_dummy_failure:
705 printf ("/push_dummy_failure");
709 extract_number_and_incr (&mcnt, &p);
710 printf ("/maybe_pop_jump to %d", p + mcnt - start);
713 case pop_failure_jump:
714 extract_number_and_incr (&mcnt, &p);
715 printf ("/pop_failure_jump to %d", p + mcnt - start);
719 extract_number_and_incr (&mcnt, &p);
720 printf ("/jump_past_alt to %d", p + mcnt - start);
724 extract_number_and_incr (&mcnt, &p);
725 printf ("/jump to %d", p + mcnt - start);
729 extract_number_and_incr (&mcnt, &p);
730 extract_number_and_incr (&mcnt2, &p);
731 printf ("/succeed_n to %d, %d times", p + mcnt - start, mcnt2);
735 extract_number_and_incr (&mcnt, &p);
736 extract_number_and_incr (&mcnt2, &p);
737 printf ("/jump_n to %d, %d times", p + mcnt - start, mcnt2);
741 extract_number_and_incr (&mcnt, &p);
742 extract_number_and_incr (&mcnt2, &p);
743 printf ("/set_number_at location %d to %d", p + mcnt - start, mcnt2);
747 printf ("/wordbound");
751 printf ("/notwordbound");
763 printf ("/before_dot");
771 printf ("/after_dot");
775 printf ("/syntaxspec");
777 printf ("/%d", mcnt);
781 printf ("/notsyntaxspec");
783 printf ("/%d", mcnt);
788 printf ("/wordchar");
792 printf ("/notwordchar");
804 printf ("?%d", *(p-1));
810 printf ("%d:\tend of pattern.\n", p - start);
815 print_compiled_pattern (bufp)
816 struct re_pattern_buffer *bufp;
818 unsigned char *buffer = bufp->buffer;
820 print_partial_compiled_pattern (buffer, buffer + bufp->used);
821 printf ("%d bytes used/%d bytes allocated.\n", bufp->used, bufp->allocated);
823 if (bufp->fastmap_accurate && bufp->fastmap)
825 printf ("fastmap: ");
826 print_fastmap (bufp->fastmap);
829 printf ("re_nsub: %d\t", bufp->re_nsub);
830 printf ("regs_alloc: %d\t", bufp->regs_allocated);
831 printf ("can_be_null: %d\t", bufp->can_be_null);
832 printf ("newline_anchor: %d\n", bufp->newline_anchor);
833 printf ("no_sub: %d\t", bufp->no_sub);
834 printf ("not_bol: %d\t", bufp->not_bol);
835 printf ("not_eol: %d\t", bufp->not_eol);
836 printf ("syntax: %d\n", bufp->syntax);
837 /* Perhaps we should print the translate table? */
842 print_double_string (where, string1, size1, string2, size2)
855 if (FIRST_STRING_P (where))
857 for (this_char = where - string1; this_char < size1; this_char++)
858 putchar (string1[this_char]);
863 for (this_char = where - string2; this_char < size2; this_char++)
864 putchar (string2[this_char]);
868 #else /* not DEBUG */
873 #define DEBUG_STATEMENT(e)
874 #define DEBUG_PRINT1(x)
875 #define DEBUG_PRINT2(x1, x2)
876 #define DEBUG_PRINT3(x1, x2, x3)
877 #define DEBUG_PRINT4(x1, x2, x3, x4)
878 #define DEBUG_PRINT_COMPILED_PATTERN(p, s, e)
879 #define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2)
881 #endif /* not DEBUG */
883 /* Set by `re_set_syntax' to the current regexp syntax to recognize. Can
884 also be assigned to arbitrarily: each pattern buffer stores its own
885 syntax, so it can be changed between regex compilations. */
886 /* This has no initializer because initialized variables in Emacs
887 become read-only after dumping. */
888 reg_syntax_t re_syntax_options;
891 /* Specify the precise syntax of regexps for compilation. This provides
892 for compatibility for various utilities which historically have
893 different, incompatible syntaxes.
895 The argument SYNTAX is a bit mask comprised of the various bits
896 defined in regex.h. We return the old syntax. */
899 re_set_syntax (syntax)
902 reg_syntax_t ret = re_syntax_options;
904 re_syntax_options = syntax;
908 /* This table gives an error message for each of the error codes listed
909 in regex.h. Obviously the order here has to be same as there.
910 POSIX doesn't require that we do anything for REG_NOERROR,
911 but why not be nice? */
913 static const char *re_error_msgid[] =
914 { "Success", /* REG_NOERROR */
915 "No match", /* REG_NOMATCH */
916 "Invalid regular expression", /* REG_BADPAT */
917 "Invalid collation character", /* REG_ECOLLATE */
918 "Invalid character class name", /* REG_ECTYPE */
919 "Trailing backslash", /* REG_EESCAPE */
920 "Invalid back reference", /* REG_ESUBREG */
921 "Unmatched [ or [^", /* REG_EBRACK */
922 "Unmatched ( or \\(", /* REG_EPAREN */
923 "Unmatched \\{", /* REG_EBRACE */
924 "Invalid content of \\{\\}", /* REG_BADBR */
925 "Invalid range end", /* REG_ERANGE */
926 "Memory exhausted", /* REG_ESPACE */
927 "Invalid preceding regular expression", /* REG_BADRPT */
928 "Premature end of regular expression", /* REG_EEND */
929 "Regular expression too big", /* REG_ESIZE */
930 "Unmatched ) or \\)", /* REG_ERPAREN */
933 /* Avoiding alloca during matching, to placate r_alloc. */
935 /* Define MATCH_MAY_ALLOCATE unless we need to make sure that the
936 searching and matching functions should not call alloca. On some
937 systems, alloca is implemented in terms of malloc, and if we're
938 using the relocating allocator routines, then malloc could cause a
939 relocation, which might (if the strings being searched are in the
940 ralloc heap) shift the data out from underneath the regexp
943 Here's another reason to avoid allocation: Emacs
944 processes input from X in a signal handler; processing X input may
945 call malloc; if input arrives while a matching routine is calling
946 malloc, then we're scrod. But Emacs can't just block input while
947 calling matching routines; then we don't notice interrupts when
948 they come in. So, Emacs blocks input around all regexp calls
949 except the matching calls, which it leaves unprotected, in the
950 faith that they will not malloc. */
952 /* Normally, this is fine. */
953 #define MATCH_MAY_ALLOCATE
955 /* When using GNU C, we are not REALLY using the C alloca, no matter
956 what config.h may say. So don't take precautions for it. */
961 /* The match routines may not allocate if (1) they would do it with malloc
962 and (2) it's not safe for them to use malloc.
963 Note that if REL_ALLOC is defined, matching would not use malloc for the
964 failure stack, but we would still use it for the register vectors;
965 so REL_ALLOC should not affect this. */
966 #if (defined (C_ALLOCA) || defined (REGEX_MALLOC)) && defined (emacs)
967 #undef MATCH_MAY_ALLOCATE
971 /* Failure stack declarations and macros; both re_compile_fastmap and
972 re_match_2 use a failure stack. These have to be macros because of
973 REGEX_ALLOCATE_STACK. */
976 /* Number of failure points for which to initially allocate space
977 when matching. If this number is exceeded, we allocate more
978 space, so it is not a hard limit. */
979 #ifndef INIT_FAILURE_ALLOC
980 #define INIT_FAILURE_ALLOC 5
983 /* Roughly the maximum number of failure points on the stack. Would be
984 exactly that if always used MAX_FAILURE_SPACE each time we failed.
985 This is a variable only so users of regex can assign to it; we never
986 change it ourselves. */
987 #if defined (MATCH_MAY_ALLOCATE)
988 int re_max_failures = 200000;
990 int re_max_failures = 2000;
995 unsigned char *pointer;
999 typedef union fail_stack_elt fail_stack_elt_t;
1003 fail_stack_elt_t *stack;
1005 unsigned avail; /* Offset of next open position. */
1008 #define FAIL_STACK_EMPTY() (fail_stack.avail == 0)
1009 #define FAIL_STACK_PTR_EMPTY() (fail_stack_ptr->avail == 0)
1010 #define FAIL_STACK_FULL() (fail_stack.avail == fail_stack.size)
1013 /* Define macros to initialize and free the failure stack.
1014 Do `return -2' if the alloc fails. */
1016 #ifdef MATCH_MAY_ALLOCATE
1017 #define INIT_FAIL_STACK() \
1019 fail_stack.stack = (fail_stack_elt_t *) \
1020 REGEX_ALLOCATE_STACK (INIT_FAILURE_ALLOC * sizeof (fail_stack_elt_t)); \
1022 if (fail_stack.stack == NULL) \
1025 fail_stack.size = INIT_FAILURE_ALLOC; \
1026 fail_stack.avail = 0; \
1029 #define RESET_FAIL_STACK() REGEX_FREE_STACK (fail_stack.stack)
1031 #define INIT_FAIL_STACK() \
1033 fail_stack.avail = 0; \
1036 #define RESET_FAIL_STACK()
1040 /* Double the size of FAIL_STACK, up to approximately `re_max_failures' items.
1042 Return 1 if succeeds, and 0 if either ran out of memory
1043 allocating space for it or it was already too large.
1045 REGEX_REALLOCATE_STACK requires `destination' be declared. */
1047 #define DOUBLE_FAIL_STACK(fail_stack) \
1048 ((fail_stack).size > re_max_failures * MAX_FAILURE_ITEMS \
1050 : ((fail_stack).stack = (fail_stack_elt_t *) \
1051 REGEX_REALLOCATE_STACK ((fail_stack).stack, \
1052 (fail_stack).size * sizeof (fail_stack_elt_t), \
1053 ((fail_stack).size << 1) * sizeof (fail_stack_elt_t)), \
1055 (fail_stack).stack == NULL \
1057 : ((fail_stack).size <<= 1, \
1061 /* Push pointer POINTER on FAIL_STACK.
1062 Return 1 if was able to do so and 0 if ran out of memory allocating
1064 #define PUSH_PATTERN_OP(POINTER, FAIL_STACK) \
1065 ((FAIL_STACK_FULL () \
1066 && !DOUBLE_FAIL_STACK (FAIL_STACK)) \
1068 : ((FAIL_STACK).stack[(FAIL_STACK).avail++].pointer = POINTER, \
1071 /* Push a pointer value onto the failure stack.
1072 Assumes the variable `fail_stack'. Probably should only
1073 be called from within `PUSH_FAILURE_POINT'. */
1074 #define PUSH_FAILURE_POINTER(item) \
1075 fail_stack.stack[fail_stack.avail++].pointer = (unsigned char *) (item)
1077 /* This pushes an integer-valued item onto the failure stack.
1078 Assumes the variable `fail_stack'. Probably should only
1079 be called from within `PUSH_FAILURE_POINT'. */
1080 #define PUSH_FAILURE_INT(item) \
1081 fail_stack.stack[fail_stack.avail++].integer = (item)
1083 /* Push a fail_stack_elt_t value onto the failure stack.
1084 Assumes the variable `fail_stack'. Probably should only
1085 be called from within `PUSH_FAILURE_POINT'. */
1086 #define PUSH_FAILURE_ELT(item) \
1087 fail_stack.stack[fail_stack.avail++] = (item)
1089 /* These three POP... operations complement the three PUSH... operations.
1090 All assume that `fail_stack' is nonempty. */
1091 #define POP_FAILURE_POINTER() fail_stack.stack[--fail_stack.avail].pointer
1092 #define POP_FAILURE_INT() fail_stack.stack[--fail_stack.avail].integer
1093 #define POP_FAILURE_ELT() fail_stack.stack[--fail_stack.avail]
1095 /* Used to omit pushing failure point id's when we're not debugging. */
1097 #define DEBUG_PUSH PUSH_FAILURE_INT
1098 #define DEBUG_POP(item_addr) *(item_addr) = POP_FAILURE_INT ()
1100 #define DEBUG_PUSH(item)
1101 #define DEBUG_POP(item_addr)
1105 /* Push the information about the state we will need
1106 if we ever fail back to it.
1108 Requires variables fail_stack, regstart, regend, reg_info, and
1109 num_regs be declared. DOUBLE_FAIL_STACK requires `destination' be
1112 Does `return FAILURE_CODE' if runs out of memory. */
1114 #define PUSH_FAILURE_POINT(pattern_place, string_place, failure_code) \
1116 char *destination; \
1117 /* Must be int, so when we don't save any registers, the arithmetic \
1118 of 0 + -1 isn't done as unsigned. */ \
1121 DEBUG_STATEMENT (failure_id++); \
1122 DEBUG_STATEMENT (nfailure_points_pushed++); \
1123 DEBUG_PRINT2 ("\nPUSH_FAILURE_POINT #%u:\n", failure_id); \
1124 DEBUG_PRINT2 (" Before push, next avail: %d\n", (fail_stack).avail);\
1125 DEBUG_PRINT2 (" size: %d\n", (fail_stack).size);\
1127 DEBUG_PRINT2 (" slots needed: %d\n", NUM_FAILURE_ITEMS); \
1128 DEBUG_PRINT2 (" available: %d\n", REMAINING_AVAIL_SLOTS); \
1130 /* Ensure we have enough space allocated for what we will push. */ \
1131 while (REMAINING_AVAIL_SLOTS < NUM_FAILURE_ITEMS) \
1133 if (!DOUBLE_FAIL_STACK (fail_stack)) \
1134 return failure_code; \
1136 DEBUG_PRINT2 ("\n Doubled stack; size now: %d\n", \
1137 (fail_stack).size); \
1138 DEBUG_PRINT2 (" slots available: %d\n", REMAINING_AVAIL_SLOTS);\
1141 /* Push the info, starting with the registers. */ \
1142 DEBUG_PRINT1 ("\n"); \
1144 for (this_reg = lowest_active_reg; this_reg <= highest_active_reg; \
1147 DEBUG_PRINT2 (" Pushing reg: %d\n", this_reg); \
1148 DEBUG_STATEMENT (num_regs_pushed++); \
1150 DEBUG_PRINT2 (" start: 0x%x\n", regstart[this_reg]); \
1151 PUSH_FAILURE_POINTER (regstart[this_reg]); \
1153 DEBUG_PRINT2 (" end: 0x%x\n", regend[this_reg]); \
1154 PUSH_FAILURE_POINTER (regend[this_reg]); \
1156 DEBUG_PRINT2 (" info: 0x%x\n ", reg_info[this_reg]); \
1157 DEBUG_PRINT2 (" match_null=%d", \
1158 REG_MATCH_NULL_STRING_P (reg_info[this_reg])); \
1159 DEBUG_PRINT2 (" active=%d", IS_ACTIVE (reg_info[this_reg])); \
1160 DEBUG_PRINT2 (" matched_something=%d", \
1161 MATCHED_SOMETHING (reg_info[this_reg])); \
1162 DEBUG_PRINT2 (" ever_matched=%d", \
1163 EVER_MATCHED_SOMETHING (reg_info[this_reg])); \
1164 DEBUG_PRINT1 ("\n"); \
1165 PUSH_FAILURE_ELT (reg_info[this_reg].word); \
1168 DEBUG_PRINT2 (" Pushing low active reg: %d\n", lowest_active_reg);\
1169 PUSH_FAILURE_INT (lowest_active_reg); \
1171 DEBUG_PRINT2 (" Pushing high active reg: %d\n", highest_active_reg);\
1172 PUSH_FAILURE_INT (highest_active_reg); \
1174 DEBUG_PRINT2 (" Pushing pattern 0x%x: ", pattern_place); \
1175 DEBUG_PRINT_COMPILED_PATTERN (bufp, pattern_place, pend); \
1176 PUSH_FAILURE_POINTER (pattern_place); \
1178 DEBUG_PRINT2 (" Pushing string 0x%x: `", string_place); \
1179 DEBUG_PRINT_DOUBLE_STRING (string_place, string1, size1, string2, \
1181 DEBUG_PRINT1 ("'\n"); \
1182 PUSH_FAILURE_POINTER (string_place); \
1184 DEBUG_PRINT2 (" Pushing failure id: %u\n", failure_id); \
1185 DEBUG_PUSH (failure_id); \
1188 /* This is the number of items that are pushed and popped on the stack
1189 for each register. */
1190 #define NUM_REG_ITEMS 3
1192 /* Individual items aside from the registers. */
1194 #define NUM_NONREG_ITEMS 5 /* Includes failure point id. */
1196 #define NUM_NONREG_ITEMS 4
1199 /* We push at most this many items on the stack. */
1200 #define MAX_FAILURE_ITEMS ((num_regs - 1) * NUM_REG_ITEMS + NUM_NONREG_ITEMS)
1202 /* We actually push this many items. */
1203 #define NUM_FAILURE_ITEMS \
1204 ((highest_active_reg - lowest_active_reg + 1) * NUM_REG_ITEMS \
1207 /* How many items can still be added to the stack without overflowing it. */
1208 #define REMAINING_AVAIL_SLOTS ((fail_stack).size - (fail_stack).avail)
1211 /* Pops what PUSH_FAIL_STACK pushes.
1213 We restore into the parameters, all of which should be lvalues:
1214 STR -- the saved data position.
1215 PAT -- the saved pattern position.
1216 LOW_REG, HIGH_REG -- the highest and lowest active registers.
1217 REGSTART, REGEND -- arrays of string positions.
1218 REG_INFO -- array of information about each subexpression.
1220 Also assumes the variables `fail_stack' and (if debugging), `bufp',
1221 `pend', `string1', `size1', `string2', and `size2'. */
1223 #define POP_FAILURE_POINT(str, pat, low_reg, high_reg, regstart, regend, reg_info)\
1225 DEBUG_STATEMENT (fail_stack_elt_t failure_id;) \
1227 const unsigned char *string_temp; \
1229 assert (!FAIL_STACK_EMPTY ()); \
1231 /* Remove failure points and point to how many regs pushed. */ \
1232 DEBUG_PRINT1 ("POP_FAILURE_POINT:\n"); \
1233 DEBUG_PRINT2 (" Before pop, next avail: %d\n", fail_stack.avail); \
1234 DEBUG_PRINT2 (" size: %d\n", fail_stack.size); \
1236 assert (fail_stack.avail >= NUM_NONREG_ITEMS); \
1238 DEBUG_POP (&failure_id); \
1239 DEBUG_PRINT2 (" Popping failure id: %u\n", failure_id); \
1241 /* If the saved string location is NULL, it came from an \
1242 on_failure_keep_string_jump opcode, and we want to throw away the \
1243 saved NULL, thus retaining our current position in the string. */ \
1244 string_temp = POP_FAILURE_POINTER (); \
1245 if (string_temp != NULL) \
1246 str = (const char *) string_temp; \
1248 DEBUG_PRINT2 (" Popping string 0x%x: `", str); \
1249 DEBUG_PRINT_DOUBLE_STRING (str, string1, size1, string2, size2); \
1250 DEBUG_PRINT1 ("'\n"); \
1252 pat = (unsigned char *) POP_FAILURE_POINTER (); \
1253 DEBUG_PRINT2 (" Popping pattern 0x%x: ", pat); \
1254 DEBUG_PRINT_COMPILED_PATTERN (bufp, pat, pend); \
1256 /* Restore register info. */ \
1257 high_reg = (unsigned) POP_FAILURE_INT (); \
1258 DEBUG_PRINT2 (" Popping high active reg: %d\n", high_reg); \
1260 low_reg = (unsigned) POP_FAILURE_INT (); \
1261 DEBUG_PRINT2 (" Popping low active reg: %d\n", low_reg); \
1263 for (this_reg = high_reg; this_reg >= low_reg; this_reg--) \
1265 DEBUG_PRINT2 (" Popping reg: %d\n", this_reg); \
1267 reg_info[this_reg].word = POP_FAILURE_ELT (); \
1268 DEBUG_PRINT2 (" info: 0x%x\n", reg_info[this_reg]); \
1270 regend[this_reg] = (const char *) POP_FAILURE_POINTER (); \
1271 DEBUG_PRINT2 (" end: 0x%x\n", regend[this_reg]); \
1273 regstart[this_reg] = (const char *) POP_FAILURE_POINTER (); \
1274 DEBUG_PRINT2 (" start: 0x%x\n", regstart[this_reg]); \
1277 set_regs_matched_done = 0; \
1278 DEBUG_STATEMENT (nfailure_points_popped++); \
1279 } /* POP_FAILURE_POINT */
1283 /* Structure for per-register (a.k.a. per-group) information.
1284 Other register information, such as the
1285 starting and ending positions (which are addresses), and the list of
1286 inner groups (which is a bits list) are maintained in separate
1289 We are making a (strictly speaking) nonportable assumption here: that
1290 the compiler will pack our bit fields into something that fits into
1291 the type of `word', i.e., is something that fits into one item on the
1296 fail_stack_elt_t word;
1299 /* This field is one if this group can match the empty string,
1300 zero if not. If not yet determined, `MATCH_NULL_UNSET_VALUE'. */
1301 #define MATCH_NULL_UNSET_VALUE 3
1302 unsigned match_null_string_p : 2;
1303 unsigned is_active : 1;
1304 unsigned matched_something : 1;
1305 unsigned ever_matched_something : 1;
1307 } register_info_type;
1309 #define REG_MATCH_NULL_STRING_P(R) ((R).bits.match_null_string_p)
1310 #define IS_ACTIVE(R) ((R).bits.is_active)
1311 #define MATCHED_SOMETHING(R) ((R).bits.matched_something)
1312 #define EVER_MATCHED_SOMETHING(R) ((R).bits.ever_matched_something)
1315 /* Call this when have matched a real character; it sets `matched' flags
1316 for the subexpressions which we are currently inside. Also records
1317 that those subexprs have matched. */
1318 #define SET_REGS_MATCHED() \
1321 if (!set_regs_matched_done) \
1324 set_regs_matched_done = 1; \
1325 for (r = lowest_active_reg; r <= highest_active_reg; r++) \
1327 MATCHED_SOMETHING (reg_info[r]) \
1328 = EVER_MATCHED_SOMETHING (reg_info[r]) \
1335 /* Registers are set to a sentinel when they haven't yet matched. */
1336 static char reg_unset_dummy;
1337 #define REG_UNSET_VALUE (®_unset_dummy)
1338 #define REG_UNSET(e) ((e) == REG_UNSET_VALUE)
1340 /* Subroutine declarations and macros for regex_compile. */
1342 static void store_op1 (), store_op2 ();
1343 static void insert_op1 (), insert_op2 ();
1344 static boolean at_begline_loc_p (), at_endline_loc_p ();
1345 static boolean group_in_compile_stack ();
1346 static reg_errcode_t compile_range ();
1348 /* Fetch the next character in the uncompiled pattern---translating it
1349 if necessary. Also cast from a signed character in the constant
1350 string passed to us by the user to an unsigned char that we can use
1351 as an array index (in, e.g., `translate'). */
1352 #define PATFETCH(c) \
1353 do {if (p == pend) return REG_EEND; \
1354 c = (unsigned char) *p++; \
1355 if (translate) c = translate[c]; \
1358 /* Fetch the next character in the uncompiled pattern, with no
1360 #define PATFETCH_RAW(c) \
1361 do {if (p == pend) return REG_EEND; \
1362 c = (unsigned char) *p++; \
1365 /* Go backwards one character in the pattern. */
1366 #define PATUNFETCH p--
1369 /* If `translate' is non-null, return translate[D], else just D. We
1370 cast the subscript to translate because some data is declared as
1371 `char *', to avoid warnings when a string constant is passed. But
1372 when we use a character as a subscript we must make it unsigned. */
1373 #define TRANSLATE(d) (translate ? translate[(unsigned char) (d)] : (d))
1376 /* Macros for outputting the compiled pattern into `buffer'. */
1378 /* If the buffer isn't allocated when it comes in, use this. */
1379 #define INIT_BUF_SIZE 32
1381 /* Make sure we have at least N more bytes of space in buffer. */
1382 #define GET_BUFFER_SPACE(n) \
1383 while (b - bufp->buffer + (n) > bufp->allocated) \
1386 /* Make sure we have one more byte of buffer space and then add C to it. */
1387 #define BUF_PUSH(c) \
1389 GET_BUFFER_SPACE (1); \
1390 *b++ = (unsigned char) (c); \
1394 /* Ensure we have two more bytes of buffer space and then append C1 and C2. */
1395 #define BUF_PUSH_2(c1, c2) \
1397 GET_BUFFER_SPACE (2); \
1398 *b++ = (unsigned char) (c1); \
1399 *b++ = (unsigned char) (c2); \
1403 /* As with BUF_PUSH_2, except for three bytes. */
1404 #define BUF_PUSH_3(c1, c2, c3) \
1406 GET_BUFFER_SPACE (3); \
1407 *b++ = (unsigned char) (c1); \
1408 *b++ = (unsigned char) (c2); \
1409 *b++ = (unsigned char) (c3); \
1413 /* Store a jump with opcode OP at LOC to location TO. We store a
1414 relative address offset by the three bytes the jump itself occupies. */
1415 #define STORE_JUMP(op, loc, to) \
1416 store_op1 (op, loc, (to) - (loc) - 3)
1418 /* Likewise, for a two-argument jump. */
1419 #define STORE_JUMP2(op, loc, to, arg) \
1420 store_op2 (op, loc, (to) - (loc) - 3, arg)
1422 /* Like `STORE_JUMP', but for inserting. Assume `b' is the buffer end. */
1423 #define INSERT_JUMP(op, loc, to) \
1424 insert_op1 (op, loc, (to) - (loc) - 3, b)
1426 /* Like `STORE_JUMP2', but for inserting. Assume `b' is the buffer end. */
1427 #define INSERT_JUMP2(op, loc, to, arg) \
1428 insert_op2 (op, loc, (to) - (loc) - 3, arg, b)
1431 /* This is not an arbitrary limit: the arguments which represent offsets
1432 into the pattern are two bytes long. So if 2^16 bytes turns out to
1433 be too small, many things would have to change. */
1434 #define MAX_BUF_SIZE (1L << 16)
1437 /* Extend the buffer by twice its current size via realloc and
1438 reset the pointers that pointed into the old block to point to the
1439 correct places in the new one. If extending the buffer results in it
1440 being larger than MAX_BUF_SIZE, then flag memory exhausted. */
1441 #define EXTEND_BUFFER() \
1443 unsigned char *old_buffer = bufp->buffer; \
1444 if (bufp->allocated == MAX_BUF_SIZE) \
1446 bufp->allocated <<= 1; \
1447 if (bufp->allocated > MAX_BUF_SIZE) \
1448 bufp->allocated = MAX_BUF_SIZE; \
1449 bufp->buffer = (unsigned char *) realloc (bufp->buffer, bufp->allocated);\
1450 if (bufp->buffer == NULL) \
1451 return REG_ESPACE; \
1452 /* If the buffer moved, move all the pointers into it. */ \
1453 if (old_buffer != bufp->buffer) \
1455 b = (b - old_buffer) + bufp->buffer; \
1456 begalt = (begalt - old_buffer) + bufp->buffer; \
1457 if (fixup_alt_jump) \
1458 fixup_alt_jump = (fixup_alt_jump - old_buffer) + bufp->buffer;\
1460 laststart = (laststart - old_buffer) + bufp->buffer; \
1461 if (pending_exact) \
1462 pending_exact = (pending_exact - old_buffer) + bufp->buffer; \
1467 /* Since we have one byte reserved for the register number argument to
1468 {start,stop}_memory, the maximum number of groups we can report
1469 things about is what fits in that byte. */
1470 #define MAX_REGNUM 255
1472 /* But patterns can have more than `MAX_REGNUM' registers. We just
1473 ignore the excess. */
1474 typedef unsigned regnum_t;
1477 /* Macros for the compile stack. */
1479 /* Since offsets can go either forwards or backwards, this type needs to
1480 be able to hold values from -(MAX_BUF_SIZE - 1) to MAX_BUF_SIZE - 1. */
1481 typedef int pattern_offset_t;
1485 pattern_offset_t begalt_offset;
1486 pattern_offset_t fixup_alt_jump;
1487 pattern_offset_t inner_group_offset;
1488 pattern_offset_t laststart_offset;
1490 } compile_stack_elt_t;
1495 compile_stack_elt_t *stack;
1497 unsigned avail; /* Offset of next open position. */
1498 } compile_stack_type;
1501 #define INIT_COMPILE_STACK_SIZE 32
1503 #define COMPILE_STACK_EMPTY (compile_stack.avail == 0)
1504 #define COMPILE_STACK_FULL (compile_stack.avail == compile_stack.size)
1506 /* The next available element. */
1507 #define COMPILE_STACK_TOP (compile_stack.stack[compile_stack.avail])
1510 /* Set the bit for character C in a list. */
1511 #define SET_LIST_BIT(c) \
1512 (b[((unsigned char) (c)) / BYTEWIDTH] \
1513 |= 1 << (((unsigned char) c) % BYTEWIDTH))
1516 /* Get the next unsigned number in the uncompiled pattern. */
1517 #define GET_UNSIGNED_NUMBER(num) \
1521 while (ISDIGIT (c)) \
1525 num = num * 10 + c - '0'; \
1533 #define CHAR_CLASS_MAX_LENGTH 6 /* Namely, `xdigit'. */
1535 #define IS_CHAR_CLASS(string) \
1536 (STREQ (string, "alpha") || STREQ (string, "upper") \
1537 || STREQ (string, "lower") || STREQ (string, "digit") \
1538 || STREQ (string, "alnum") || STREQ (string, "xdigit") \
1539 || STREQ (string, "space") || STREQ (string, "print") \
1540 || STREQ (string, "punct") || STREQ (string, "graph") \
1541 || STREQ (string, "cntrl") || STREQ (string, "blank"))
1543 #ifndef MATCH_MAY_ALLOCATE
1545 /* If we cannot allocate large objects within re_match_2_internal,
1546 we make the fail stack and register vectors global.
1547 The fail stack, we grow to the maximum size when a regexp
1549 The register vectors, we adjust in size each time we
1550 compile a regexp, according to the number of registers it needs. */
1552 static fail_stack_type fail_stack;
1554 /* Size with which the following vectors are currently allocated.
1555 That is so we can make them bigger as needed,
1556 but never make them smaller. */
1557 static int regs_allocated_size;
1559 static const char ** regstart, ** regend;
1560 static const char ** old_regstart, ** old_regend;
1561 static const char **best_regstart, **best_regend;
1562 static register_info_type *reg_info;
1563 static const char **reg_dummy;
1564 static register_info_type *reg_info_dummy;
1566 /* Make the register vectors big enough for NUM_REGS registers,
1567 but don't make them smaller. */
1570 regex_grow_registers (num_regs)
1573 if (num_regs > regs_allocated_size)
1575 RETALLOC_IF (regstart, num_regs, const char *);
1576 RETALLOC_IF (regend, num_regs, const char *);
1577 RETALLOC_IF (old_regstart, num_regs, const char *);
1578 RETALLOC_IF (old_regend, num_regs, const char *);
1579 RETALLOC_IF (best_regstart, num_regs, const char *);
1580 RETALLOC_IF (best_regend, num_regs, const char *);
1581 RETALLOC_IF (reg_info, num_regs, register_info_type);
1582 RETALLOC_IF (reg_dummy, num_regs, const char *);
1583 RETALLOC_IF (reg_info_dummy, num_regs, register_info_type);
1585 regs_allocated_size = num_regs;
1589 #endif /* not MATCH_MAY_ALLOCATE */
1591 /* `regex_compile' compiles PATTERN (of length SIZE) according to SYNTAX.
1592 Returns one of error codes defined in `regex.h', or zero for success.
1594 Assumes the `allocated' (and perhaps `buffer') and `translate'
1595 fields are set in BUFP on entry.
1597 If it succeeds, results are put in BUFP (if it returns an error, the
1598 contents of BUFP are undefined):
1599 `buffer' is the compiled pattern;
1600 `syntax' is set to SYNTAX;
1601 `used' is set to the length of the compiled pattern;
1602 `fastmap_accurate' is zero;
1603 `re_nsub' is the number of subexpressions in PATTERN;
1604 `not_bol' and `not_eol' are zero;
1606 The `fastmap' and `newline_anchor' fields are neither
1607 examined nor set. */
1609 /* Return, freeing storage we allocated. */
1610 #define FREE_STACK_RETURN(value) \
1611 return (free (compile_stack.stack), value)
1613 static reg_errcode_t
1614 regex_compile (pattern, size, syntax, bufp)
1615 const char *pattern;
1617 reg_syntax_t syntax;
1618 struct re_pattern_buffer *bufp;
1620 /* We fetch characters from PATTERN here. Even though PATTERN is
1621 `char *' (i.e., signed), we declare these variables as unsigned, so
1622 they can be reliably used as array indices. */
1623 register unsigned char c, c1;
1625 /* A random temporary spot in PATTERN. */
1628 /* Points to the end of the buffer, where we should append. */
1629 register unsigned char *b;
1631 /* Keeps track of unclosed groups. */
1632 compile_stack_type compile_stack;
1634 /* Points to the current (ending) position in the pattern. */
1635 const char *p = pattern;
1636 const char *pend = pattern + size;
1638 /* How to translate the characters in the pattern. */
1639 char *translate = bufp->translate;
1641 /* Address of the count-byte of the most recently inserted `exactn'
1642 command. This makes it possible to tell if a new exact-match
1643 character can be added to that command or if the character requires
1644 a new `exactn' command. */
1645 unsigned char *pending_exact = 0;
1647 /* Address of start of the most recently finished expression.
1648 This tells, e.g., postfix * where to find the start of its
1649 operand. Reset at the beginning of groups and alternatives. */
1650 unsigned char *laststart = 0;
1652 /* Address of beginning of regexp, or inside of last group. */
1653 unsigned char *begalt;
1655 /* Place in the uncompiled pattern (i.e., the {) to
1656 which to go back if the interval is invalid. */
1657 const char *beg_interval;
1659 /* Address of the place where a forward jump should go to the end of
1660 the containing expression. Each alternative of an `or' -- except the
1661 last -- ends with a forward jump of this sort. */
1662 unsigned char *fixup_alt_jump = 0;
1664 /* Counts open-groups as they are encountered. Remembered for the
1665 matching close-group on the compile stack, so the same register
1666 number is put in the stop_memory as the start_memory. */
1667 regnum_t regnum = 0;
1670 DEBUG_PRINT1 ("\nCompiling pattern: ");
1673 unsigned debug_count;
1675 for (debug_count = 0; debug_count < size; debug_count++)
1676 putchar (pattern[debug_count]);
1681 /* Initialize the compile stack. */
1682 compile_stack.stack = TALLOC (INIT_COMPILE_STACK_SIZE, compile_stack_elt_t);
1683 if (compile_stack.stack == NULL)
1686 compile_stack.size = INIT_COMPILE_STACK_SIZE;
1687 compile_stack.avail = 0;
1689 /* Initialize the pattern buffer. */
1690 bufp->syntax = syntax;
1691 bufp->fastmap_accurate = 0;
1692 bufp->not_bol = bufp->not_eol = 0;
1694 /* Set `used' to zero, so that if we return an error, the pattern
1695 printer (for debugging) will think there's no pattern. We reset it
1699 /* Always count groups, whether or not bufp->no_sub is set. */
1702 #if !defined (emacs) && !defined (SYNTAX_TABLE)
1703 /* Initialize the syntax table. */
1704 init_syntax_once ();
1707 if (bufp->allocated == 0)
1710 { /* If zero allocated, but buffer is non-null, try to realloc
1711 enough space. This loses if buffer's address is bogus, but
1712 that is the user's responsibility. */
1713 RETALLOC (bufp->buffer, INIT_BUF_SIZE, unsigned char);
1716 { /* Caller did not allocate a buffer. Do it for them. */
1717 bufp->buffer = TALLOC (INIT_BUF_SIZE, unsigned char);
1719 if (!bufp->buffer) FREE_STACK_RETURN (REG_ESPACE);
1721 bufp->allocated = INIT_BUF_SIZE;
1724 begalt = b = bufp->buffer;
1726 /* Loop through the uncompiled pattern until we're at the end. */
1735 if ( /* If at start of pattern, it's an operator. */
1737 /* If context independent, it's an operator. */
1738 || syntax & RE_CONTEXT_INDEP_ANCHORS
1739 /* Otherwise, depends on what's come before. */
1740 || at_begline_loc_p (pattern, p, syntax))
1750 if ( /* If at end of pattern, it's an operator. */
1752 /* If context independent, it's an operator. */
1753 || syntax & RE_CONTEXT_INDEP_ANCHORS
1754 /* Otherwise, depends on what's next. */
1755 || at_endline_loc_p (p, pend, syntax))
1765 if ((syntax & RE_BK_PLUS_QM)
1766 || (syntax & RE_LIMITED_OPS))
1770 /* If there is no previous pattern... */
1773 if (syntax & RE_CONTEXT_INVALID_OPS)
1774 FREE_STACK_RETURN (REG_BADRPT);
1775 else if (!(syntax & RE_CONTEXT_INDEP_OPS))
1780 /* Are we optimizing this jump? */
1781 boolean keep_string_p = false;
1783 /* 1 means zero (many) matches is allowed. */
1784 char zero_times_ok = 0, many_times_ok = 0;
1786 /* If there is a sequence of repetition chars, collapse it
1787 down to just one (the right one). We can't combine
1788 interval operators with these because of, e.g., `a{2}*',
1789 which should only match an even number of `a's. */
1793 zero_times_ok |= c != '+';
1794 many_times_ok |= c != '?';
1802 || (!(syntax & RE_BK_PLUS_QM) && (c == '+' || c == '?')))
1805 else if (syntax & RE_BK_PLUS_QM && c == '\\')
1807 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
1810 if (!(c1 == '+' || c1 == '?'))
1825 /* If we get here, we found another repeat character. */
1828 /* Star, etc. applied to an empty pattern is equivalent
1829 to an empty pattern. */
1833 /* Now we know whether or not zero matches is allowed
1834 and also whether or not two or more matches is allowed. */
1836 { /* More than one repetition is allowed, so put in at the
1837 end a backward relative jump from `b' to before the next
1838 jump we're going to put in below (which jumps from
1839 laststart to after this jump).
1841 But if we are at the `*' in the exact sequence `.*\n',
1842 insert an unconditional jump backwards to the .,
1843 instead of the beginning of the loop. This way we only
1844 push a failure point once, instead of every time
1845 through the loop. */
1846 assert (p - 1 > pattern);
1848 /* Allocate the space for the jump. */
1849 GET_BUFFER_SPACE (3);
1851 /* We know we are not at the first character of the pattern,
1852 because laststart was nonzero. And we've already
1853 incremented `p', by the way, to be the character after
1854 the `*'. Do we have to do something analogous here
1855 for null bytes, because of RE_DOT_NOT_NULL? */
1856 if (TRANSLATE (*(p - 2)) == TRANSLATE ('.')
1858 && p < pend && TRANSLATE (*p) == TRANSLATE ('\n')
1859 && !(syntax & RE_DOT_NEWLINE))
1860 { /* We have .*\n. */
1861 STORE_JUMP (jump, b, laststart);
1862 keep_string_p = true;
1865 /* Anything else. */
1866 STORE_JUMP (maybe_pop_jump, b, laststart - 3);
1868 /* We've added more stuff to the buffer. */
1872 /* On failure, jump from laststart to b + 3, which will be the
1873 end of the buffer after this jump is inserted. */
1874 GET_BUFFER_SPACE (3);
1875 INSERT_JUMP (keep_string_p ? on_failure_keep_string_jump
1883 /* At least one repetition is required, so insert a
1884 `dummy_failure_jump' before the initial
1885 `on_failure_jump' instruction of the loop. This
1886 effects a skip over that instruction the first time
1887 we hit that loop. */
1888 GET_BUFFER_SPACE (3);
1889 INSERT_JUMP (dummy_failure_jump, laststart, laststart + 6);
1904 boolean had_char_class = false;
1906 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
1908 /* Ensure that we have enough space to push a charset: the
1909 opcode, the length count, and the bitset; 34 bytes in all. */
1910 GET_BUFFER_SPACE (34);
1914 /* We test `*p == '^' twice, instead of using an if
1915 statement, so we only need one BUF_PUSH. */
1916 BUF_PUSH (*p == '^' ? charset_not : charset);
1920 /* Remember the first position in the bracket expression. */
1923 /* Push the number of bytes in the bitmap. */
1924 BUF_PUSH ((1 << BYTEWIDTH) / BYTEWIDTH);
1926 /* Clear the whole map. */
1927 bzero (b, (1 << BYTEWIDTH) / BYTEWIDTH);
1929 /* charset_not matches newline according to a syntax bit. */
1930 if ((re_opcode_t) b[-2] == charset_not
1931 && (syntax & RE_HAT_LISTS_NOT_NEWLINE))
1932 SET_LIST_BIT ('\n');
1934 /* Read in characters and ranges, setting map bits. */
1937 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
1941 /* \ might escape characters inside [...] and [^...]. */
1942 if ((syntax & RE_BACKSLASH_ESCAPE_IN_LISTS) && c == '\\')
1944 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
1951 /* Could be the end of the bracket expression. If it's
1952 not (i.e., when the bracket expression is `[]' so
1953 far), the ']' character bit gets set way below. */
1954 if (c == ']' && p != p1 + 1)
1957 /* Look ahead to see if it's a range when the last thing
1958 was a character class. */
1959 if (had_char_class && c == '-' && *p != ']')
1960 FREE_STACK_RETURN (REG_ERANGE);
1962 /* Look ahead to see if it's a range when the last thing
1963 was a character: if this is a hyphen not at the
1964 beginning or the end of a list, then it's the range
1967 && !(p - 2 >= pattern && p[-2] == '[')
1968 && !(p - 3 >= pattern && p[-3] == '[' && p[-2] == '^')
1972 = compile_range (&p, pend, translate, syntax, b);
1973 if (ret != REG_NOERROR) FREE_STACK_RETURN (ret);
1976 else if (p[0] == '-' && p[1] != ']')
1977 { /* This handles ranges made up of characters only. */
1980 /* Move past the `-'. */
1983 ret = compile_range (&p, pend, translate, syntax, b);
1984 if (ret != REG_NOERROR) FREE_STACK_RETURN (ret);
1987 /* See if we're at the beginning of a possible character
1990 else if (syntax & RE_CHAR_CLASSES && c == '[' && *p == ':')
1991 { /* Leave room for the null. */
1992 char str[CHAR_CLASS_MAX_LENGTH + 1];
1997 /* If pattern is `[[:'. */
1998 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2003 if (c == ':' || c == ']' || p == pend
2004 || c1 == CHAR_CLASS_MAX_LENGTH)
2010 /* If isn't a word bracketed by `[:' and:`]':
2011 undo the ending character, the letters, and leave
2012 the leading `:' and `[' (but set bits for them). */
2013 if (c == ':' && *p == ']')
2016 boolean is_alnum = STREQ (str, "alnum");
2017 boolean is_alpha = STREQ (str, "alpha");
2018 boolean is_blank = STREQ (str, "blank");
2019 boolean is_cntrl = STREQ (str, "cntrl");
2020 boolean is_digit = STREQ (str, "digit");
2021 boolean is_graph = STREQ (str, "graph");
2022 boolean is_lower = STREQ (str, "lower");
2023 boolean is_print = STREQ (str, "print");
2024 boolean is_punct = STREQ (str, "punct");
2025 boolean is_space = STREQ (str, "space");
2026 boolean is_upper = STREQ (str, "upper");
2027 boolean is_xdigit = STREQ (str, "xdigit");
2029 if (!IS_CHAR_CLASS (str))
2030 FREE_STACK_RETURN (REG_ECTYPE);
2032 /* Throw away the ] at the end of the character
2036 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2038 for (ch = 0; ch < 1 << BYTEWIDTH; ch++)
2040 /* This was split into 3 if's to
2041 avoid an arbitrary limit in some compiler. */
2042 if ( (is_alnum && ISALNUM (ch))
2043 || (is_alpha && ISALPHA (ch))
2044 || (is_blank && ISBLANK (ch))
2045 || (is_cntrl && ISCNTRL (ch)))
2047 if ( (is_digit && ISDIGIT (ch))
2048 || (is_graph && ISGRAPH (ch))
2049 || (is_lower && ISLOWER (ch))
2050 || (is_print && ISPRINT (ch)))
2052 if ( (is_punct && ISPUNCT (ch))
2053 || (is_space && ISSPACE (ch))
2054 || (is_upper && ISUPPER (ch))
2055 || (is_xdigit && ISXDIGIT (ch)))
2058 had_char_class = true;
2067 had_char_class = false;
2072 had_char_class = false;
2077 /* Discard any (non)matching list bytes that are all 0 at the
2078 end of the map. Decrease the map-length byte too. */
2079 while ((int) b[-1] > 0 && b[b[-1] - 1] == 0)
2087 if (syntax & RE_NO_BK_PARENS)
2094 if (syntax & RE_NO_BK_PARENS)
2101 if (syntax & RE_NEWLINE_ALT)
2108 if (syntax & RE_NO_BK_VBAR)
2115 if (syntax & RE_INTERVALS && syntax & RE_NO_BK_BRACES)
2116 goto handle_interval;
2122 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
2124 /* Do not translate the character after the \, so that we can
2125 distinguish, e.g., \B from \b, even if we normally would
2126 translate, e.g., B to b. */
2132 if (syntax & RE_NO_BK_PARENS)
2133 goto normal_backslash;
2139 if (COMPILE_STACK_FULL)
2141 RETALLOC (compile_stack.stack, compile_stack.size << 1,
2142 compile_stack_elt_t);
2143 if (compile_stack.stack == NULL) return REG_ESPACE;
2145 compile_stack.size <<= 1;
2148 /* These are the values to restore when we hit end of this
2149 group. They are all relative offsets, so that if the
2150 whole pattern moves because of realloc, they will still
2152 COMPILE_STACK_TOP.begalt_offset = begalt - bufp->buffer;
2153 COMPILE_STACK_TOP.fixup_alt_jump
2154 = fixup_alt_jump ? fixup_alt_jump - bufp->buffer + 1 : 0;
2155 COMPILE_STACK_TOP.laststart_offset = b - bufp->buffer;
2156 COMPILE_STACK_TOP.regnum = regnum;
2158 /* We will eventually replace the 0 with the number of
2159 groups inner to this one. But do not push a
2160 start_memory for groups beyond the last one we can
2161 represent in the compiled pattern. */
2162 if (regnum <= MAX_REGNUM)
2164 COMPILE_STACK_TOP.inner_group_offset = b - bufp->buffer + 2;
2165 BUF_PUSH_3 (start_memory, regnum, 0);
2168 compile_stack.avail++;
2173 /* If we've reached MAX_REGNUM groups, then this open
2174 won't actually generate any code, so we'll have to
2175 clear pending_exact explicitly. */
2181 if (syntax & RE_NO_BK_PARENS) goto normal_backslash;
2183 if (COMPILE_STACK_EMPTY)
2184 if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD)
2185 goto normal_backslash;
2187 FREE_STACK_RETURN (REG_ERPAREN);
2191 { /* Push a dummy failure point at the end of the
2192 alternative for a possible future
2193 `pop_failure_jump' to pop. See comments at
2194 `push_dummy_failure' in `re_match_2'. */
2195 BUF_PUSH (push_dummy_failure);
2197 /* We allocated space for this jump when we assigned
2198 to `fixup_alt_jump', in the `handle_alt' case below. */
2199 STORE_JUMP (jump_past_alt, fixup_alt_jump, b - 1);
2202 /* See similar code for backslashed left paren above. */
2203 if (COMPILE_STACK_EMPTY)
2204 if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD)
2207 FREE_STACK_RETURN (REG_ERPAREN);
2209 /* Since we just checked for an empty stack above, this
2210 ``can't happen''. */
2211 assert (compile_stack.avail != 0);
2213 /* We don't just want to restore into `regnum', because
2214 later groups should continue to be numbered higher,
2215 as in `(ab)c(de)' -- the second group is #2. */
2216 regnum_t this_group_regnum;
2218 compile_stack.avail--;
2219 begalt = bufp->buffer + COMPILE_STACK_TOP.begalt_offset;
2221 = COMPILE_STACK_TOP.fixup_alt_jump
2222 ? bufp->buffer + COMPILE_STACK_TOP.fixup_alt_jump - 1
2224 laststart = bufp->buffer + COMPILE_STACK_TOP.laststart_offset;
2225 this_group_regnum = COMPILE_STACK_TOP.regnum;
2226 /* If we've reached MAX_REGNUM groups, then this open
2227 won't actually generate any code, so we'll have to
2228 clear pending_exact explicitly. */
2231 /* We're at the end of the group, so now we know how many
2232 groups were inside this one. */
2233 if (this_group_regnum <= MAX_REGNUM)
2235 unsigned char *inner_group_loc
2236 = bufp->buffer + COMPILE_STACK_TOP.inner_group_offset;
2238 *inner_group_loc = regnum - this_group_regnum;
2239 BUF_PUSH_3 (stop_memory, this_group_regnum,
2240 regnum - this_group_regnum);
2246 case '|': /* `\|'. */
2247 if (syntax & RE_LIMITED_OPS || syntax & RE_NO_BK_VBAR)
2248 goto normal_backslash;
2250 if (syntax & RE_LIMITED_OPS)
2253 /* Insert before the previous alternative a jump which
2254 jumps to this alternative if the former fails. */
2255 GET_BUFFER_SPACE (3);
2256 INSERT_JUMP (on_failure_jump, begalt, b + 6);
2260 /* The alternative before this one has a jump after it
2261 which gets executed if it gets matched. Adjust that
2262 jump so it will jump to this alternative's analogous
2263 jump (put in below, which in turn will jump to the next
2264 (if any) alternative's such jump, etc.). The last such
2265 jump jumps to the correct final destination. A picture:
2271 If we are at `b', then fixup_alt_jump right now points to a
2272 three-byte space after `a'. We'll put in the jump, set
2273 fixup_alt_jump to right after `b', and leave behind three
2274 bytes which we'll fill in when we get to after `c'. */
2277 STORE_JUMP (jump_past_alt, fixup_alt_jump, b);
2279 /* Mark and leave space for a jump after this alternative,
2280 to be filled in later either by next alternative or
2281 when know we're at the end of a series of alternatives. */
2283 GET_BUFFER_SPACE (3);
2292 /* If \{ is a literal. */
2293 if (!(syntax & RE_INTERVALS)
2294 /* If we're at `\{' and it's not the open-interval
2296 || ((syntax & RE_INTERVALS) && (syntax & RE_NO_BK_BRACES))
2297 || (p - 2 == pattern && p == pend))
2298 goto normal_backslash;
2302 /* If got here, then the syntax allows intervals. */
2304 /* At least (most) this many matches must be made. */
2305 int lower_bound = -1, upper_bound = -1;
2307 beg_interval = p - 1;
2311 if (syntax & RE_NO_BK_BRACES)
2312 goto unfetch_interval;
2314 FREE_STACK_RETURN (REG_EBRACE);
2317 GET_UNSIGNED_NUMBER (lower_bound);
2321 GET_UNSIGNED_NUMBER (upper_bound);
2322 if (upper_bound < 0) upper_bound = RE_DUP_MAX;
2325 /* Interval such as `{1}' => match exactly once. */
2326 upper_bound = lower_bound;
2328 if (lower_bound < 0 || upper_bound > RE_DUP_MAX
2329 || lower_bound > upper_bound)
2331 if (syntax & RE_NO_BK_BRACES)
2332 goto unfetch_interval;
2334 FREE_STACK_RETURN (REG_BADBR);
2337 if (!(syntax & RE_NO_BK_BRACES))
2339 if (c != '\\') FREE_STACK_RETURN (REG_EBRACE);
2346 if (syntax & RE_NO_BK_BRACES)
2347 goto unfetch_interval;
2349 FREE_STACK_RETURN (REG_BADBR);
2352 /* We just parsed a valid interval. */
2354 /* If it's invalid to have no preceding re. */
2357 if (syntax & RE_CONTEXT_INVALID_OPS)
2358 FREE_STACK_RETURN (REG_BADRPT);
2359 else if (syntax & RE_CONTEXT_INDEP_OPS)
2362 goto unfetch_interval;
2365 /* If the upper bound is zero, don't want to succeed at
2366 all; jump from `laststart' to `b + 3', which will be
2367 the end of the buffer after we insert the jump. */
2368 if (upper_bound == 0)
2370 GET_BUFFER_SPACE (3);
2371 INSERT_JUMP (jump, laststart, b + 3);
2375 /* Otherwise, we have a nontrivial interval. When
2376 we're all done, the pattern will look like:
2377 set_number_at <jump count> <upper bound>
2378 set_number_at <succeed_n count> <lower bound>
2379 succeed_n <after jump addr> <succeed_n count>
2381 jump_n <succeed_n addr> <jump count>
2382 (The upper bound and `jump_n' are omitted if
2383 `upper_bound' is 1, though.) */
2385 { /* If the upper bound is > 1, we need to insert
2386 more at the end of the loop. */
2387 unsigned nbytes = 10 + (upper_bound > 1) * 10;
2389 GET_BUFFER_SPACE (nbytes);
2391 /* Initialize lower bound of the `succeed_n', even
2392 though it will be set during matching by its
2393 attendant `set_number_at' (inserted next),
2394 because `re_compile_fastmap' needs to know.
2395 Jump to the `jump_n' we might insert below. */
2396 INSERT_JUMP2 (succeed_n, laststart,
2397 b + 5 + (upper_bound > 1) * 5,
2401 /* Code to initialize the lower bound. Insert
2402 before the `succeed_n'. The `5' is the last two
2403 bytes of this `set_number_at', plus 3 bytes of
2404 the following `succeed_n'. */
2405 insert_op2 (set_number_at, laststart, 5, lower_bound, b);
2408 if (upper_bound > 1)
2409 { /* More than one repetition is allowed, so
2410 append a backward jump to the `succeed_n'
2411 that starts this interval.
2413 When we've reached this during matching,
2414 we'll have matched the interval once, so
2415 jump back only `upper_bound - 1' times. */
2416 STORE_JUMP2 (jump_n, b, laststart + 5,
2420 /* The location we want to set is the second
2421 parameter of the `jump_n'; that is `b-2' as
2422 an absolute address. `laststart' will be
2423 the `set_number_at' we're about to insert;
2424 `laststart+3' the number to set, the source
2425 for the relative address. But we are
2426 inserting into the middle of the pattern --
2427 so everything is getting moved up by 5.
2428 Conclusion: (b - 2) - (laststart + 3) + 5,
2429 i.e., b - laststart.
2431 We insert this at the beginning of the loop
2432 so that if we fail during matching, we'll
2433 reinitialize the bounds. */
2434 insert_op2 (set_number_at, laststart, b - laststart,
2435 upper_bound - 1, b);
2440 beg_interval = NULL;
2445 /* If an invalid interval, match the characters as literals. */
2446 assert (beg_interval);
2448 beg_interval = NULL;
2450 /* normal_char and normal_backslash need `c'. */
2453 if (!(syntax & RE_NO_BK_BRACES))
2455 if (p > pattern && p[-1] == '\\')
2456 goto normal_backslash;
2461 /* There is no way to specify the before_dot and after_dot
2462 operators. rms says this is ok. --karl */
2470 BUF_PUSH_2 (syntaxspec, syntax_spec_code[c]);
2476 BUF_PUSH_2 (notsyntaxspec, syntax_spec_code[c]);
2483 BUF_PUSH (wordchar);
2489 BUF_PUSH (notwordchar);
2502 BUF_PUSH (wordbound);
2506 BUF_PUSH (notwordbound);
2517 case '1': case '2': case '3': case '4': case '5':
2518 case '6': case '7': case '8': case '9':
2519 if (syntax & RE_NO_BK_REFS)
2525 FREE_STACK_RETURN (REG_ESUBREG);
2527 /* Can't back reference to a subexpression if inside of it. */
2528 if (group_in_compile_stack (compile_stack, c1))
2532 BUF_PUSH_2 (duplicate, c1);
2538 if (syntax & RE_BK_PLUS_QM)
2541 goto normal_backslash;
2545 /* You might think it would be useful for \ to mean
2546 not to translate; but if we don't translate it
2547 it will never match anything. */
2555 /* Expects the character in `c'. */
2557 /* If no exactn currently being built. */
2560 /* If last exactn not at current position. */
2561 || pending_exact + *pending_exact + 1 != b
2563 /* We have only one byte following the exactn for the count. */
2564 || *pending_exact == (1 << BYTEWIDTH) - 1
2566 /* If followed by a repetition operator. */
2567 || *p == '*' || *p == '^'
2568 || ((syntax & RE_BK_PLUS_QM)
2569 ? *p == '\\' && (p[1] == '+' || p[1] == '?')
2570 : (*p == '+' || *p == '?'))
2571 || ((syntax & RE_INTERVALS)
2572 && ((syntax & RE_NO_BK_BRACES)
2574 : (p[0] == '\\' && p[1] == '{'))))
2576 /* Start building a new exactn. */
2580 BUF_PUSH_2 (exactn, 0);
2581 pending_exact = b - 1;
2588 } /* while p != pend */
2591 /* Through the pattern now. */
2594 STORE_JUMP (jump_past_alt, fixup_alt_jump, b);
2596 if (!COMPILE_STACK_EMPTY)
2597 FREE_STACK_RETURN (REG_EPAREN);
2599 /* If we don't want backtracking, force success
2600 the first time we reach the end of the compiled pattern. */
2601 if (syntax & RE_NO_POSIX_BACKTRACKING)
2604 free (compile_stack.stack);
2606 /* We have succeeded; set the length of the buffer. */
2607 bufp->used = b - bufp->buffer;
2612 DEBUG_PRINT1 ("\nCompiled pattern: \n");
2613 print_compiled_pattern (bufp);
2617 #ifndef MATCH_MAY_ALLOCATE
2618 /* Initialize the failure stack to the largest possible stack. This
2619 isn't necessary unless we're trying to avoid calling alloca in
2620 the search and match routines. */
2622 int num_regs = bufp->re_nsub + 1;
2624 /* Since DOUBLE_FAIL_STACK refuses to double only if the current size
2625 is strictly greater than re_max_failures, the largest possible stack
2626 is 2 * re_max_failures failure points. */
2627 if (fail_stack.size < (2 * re_max_failures * MAX_FAILURE_ITEMS))
2629 fail_stack.size = (2 * re_max_failures * MAX_FAILURE_ITEMS);
2632 if (! fail_stack.stack)
2634 = (fail_stack_elt_t *) xmalloc (fail_stack.size
2635 * sizeof (fail_stack_elt_t));
2638 = (fail_stack_elt_t *) xrealloc (fail_stack.stack,
2640 * sizeof (fail_stack_elt_t)));
2641 #else /* not emacs */
2642 if (! fail_stack.stack)
2644 = (fail_stack_elt_t *) malloc (fail_stack.size
2645 * sizeof (fail_stack_elt_t));
2648 = (fail_stack_elt_t *) realloc (fail_stack.stack,
2650 * sizeof (fail_stack_elt_t)));
2651 #endif /* not emacs */
2654 regex_grow_registers (num_regs);
2656 #endif /* not MATCH_MAY_ALLOCATE */
2659 } /* regex_compile */
2661 /* Subroutines for `regex_compile'. */
2663 /* Store OP at LOC followed by two-byte integer parameter ARG. */
2666 store_op1 (op, loc, arg)
2671 *loc = (unsigned char) op;
2672 STORE_NUMBER (loc + 1, arg);
2676 /* Like `store_op1', but for two two-byte parameters ARG1 and ARG2. */
2679 store_op2 (op, loc, arg1, arg2)
2684 *loc = (unsigned char) op;
2685 STORE_NUMBER (loc + 1, arg1);
2686 STORE_NUMBER (loc + 3, arg2);
2690 /* Copy the bytes from LOC to END to open up three bytes of space at LOC
2691 for OP followed by two-byte integer parameter ARG. */
2694 insert_op1 (op, loc, arg, end)
2700 register unsigned char *pfrom = end;
2701 register unsigned char *pto = end + 3;
2703 while (pfrom != loc)
2706 store_op1 (op, loc, arg);
2710 /* Like `insert_op1', but for two two-byte parameters ARG1 and ARG2. */
2713 insert_op2 (op, loc, arg1, arg2, end)
2719 register unsigned char *pfrom = end;
2720 register unsigned char *pto = end + 5;
2722 while (pfrom != loc)
2725 store_op2 (op, loc, arg1, arg2);
2729 /* P points to just after a ^ in PATTERN. Return true if that ^ comes
2730 after an alternative or a begin-subexpression. We assume there is at
2731 least one character before the ^. */
2734 at_begline_loc_p (pattern, p, syntax)
2735 const char *pattern, *p;
2736 reg_syntax_t syntax;
2738 const char *prev = p - 2;
2739 boolean prev_prev_backslash = prev > pattern && prev[-1] == '\\';
2742 /* After a subexpression? */
2743 (*prev == '(' && (syntax & RE_NO_BK_PARENS || prev_prev_backslash))
2744 /* After an alternative? */
2745 || (*prev == '|' && (syntax & RE_NO_BK_VBAR || prev_prev_backslash));
2749 /* The dual of at_begline_loc_p. This one is for $. We assume there is
2750 at least one character after the $, i.e., `P < PEND'. */
2753 at_endline_loc_p (p, pend, syntax)
2754 const char *p, *pend;
2757 const char *next = p;
2758 boolean next_backslash = *next == '\\';
2759 const char *next_next = p + 1 < pend ? p + 1 : NULL;
2762 /* Before a subexpression? */
2763 (syntax & RE_NO_BK_PARENS ? *next == ')'
2764 : next_backslash && next_next && *next_next == ')')
2765 /* Before an alternative? */
2766 || (syntax & RE_NO_BK_VBAR ? *next == '|'
2767 : next_backslash && next_next && *next_next == '|');
2771 /* Returns true if REGNUM is in one of COMPILE_STACK's elements and
2772 false if it's not. */
2775 group_in_compile_stack (compile_stack, regnum)
2776 compile_stack_type compile_stack;
2781 for (this_element = compile_stack.avail - 1;
2784 if (compile_stack.stack[this_element].regnum == regnum)
2791 /* Read the ending character of a range (in a bracket expression) from the
2792 uncompiled pattern *P_PTR (which ends at PEND). We assume the
2793 starting character is in `P[-2]'. (`P[-1]' is the character `-'.)
2794 Then we set the translation of all bits between the starting and
2795 ending characters (inclusive) in the compiled pattern B.
2797 Return an error code.
2799 We use these short variable names so we can use the same macros as
2800 `regex_compile' itself. */
2802 static reg_errcode_t
2803 compile_range (p_ptr, pend, translate, syntax, b)
2804 const char **p_ptr, *pend;
2806 reg_syntax_t syntax;
2811 const char *p = *p_ptr;
2812 int range_start, range_end;
2817 /* Even though the pattern is a signed `char *', we need to fetch
2818 with unsigned char *'s; if the high bit of the pattern character
2819 is set, the range endpoints will be negative if we fetch using a
2822 We also want to fetch the endpoints without translating them; the
2823 appropriate translation is done in the bit-setting loop below. */
2824 /* The SVR4 compiler on the 3B2 had trouble with unsigned const char *. */
2825 range_start = ((const unsigned char *) p)[-2];
2826 range_end = ((const unsigned char *) p)[0];
2828 /* Have to increment the pointer into the pattern string, so the
2829 caller isn't still at the ending character. */
2832 /* If the start is after the end, the range is empty. */
2833 if (range_start > range_end)
2834 return syntax & RE_NO_EMPTY_RANGES ? REG_ERANGE : REG_NOERROR;
2836 /* Here we see why `this_char' has to be larger than an `unsigned
2837 char' -- the range is inclusive, so if `range_end' == 0xff
2838 (assuming 8-bit characters), we would otherwise go into an infinite
2839 loop, since all characters <= 0xff. */
2840 for (this_char = range_start; this_char <= range_end; this_char++)
2842 SET_LIST_BIT (TRANSLATE (this_char));
2848 /* re_compile_fastmap computes a ``fastmap'' for the compiled pattern in
2849 BUFP. A fastmap records which of the (1 << BYTEWIDTH) possible
2850 characters can start a string that matches the pattern. This fastmap
2851 is used by re_search to skip quickly over impossible starting points.
2853 The caller must supply the address of a (1 << BYTEWIDTH)-byte data
2854 area as BUFP->fastmap.
2856 We set the `fastmap', `fastmap_accurate', and `can_be_null' fields in
2859 Returns 0 if we succeed, -2 if an internal error. */
2862 re_compile_fastmap (bufp)
2863 struct re_pattern_buffer *bufp;
2866 #ifdef MATCH_MAY_ALLOCATE
2867 fail_stack_type fail_stack;
2869 #ifndef REGEX_MALLOC
2872 /* We don't push any register information onto the failure stack. */
2873 unsigned num_regs = 0;
2875 register char *fastmap = bufp->fastmap;
2876 unsigned char *pattern = bufp->buffer;
2877 unsigned long size = bufp->used;
2878 unsigned char *p = pattern;
2879 register unsigned char *pend = pattern + size;
2881 /* This holds the pointer to the failure stack, when
2882 it is allocated relocatably. */
2884 fail_stack_elt_t *failure_stack_ptr;
2887 /* Assume that each path through the pattern can be null until
2888 proven otherwise. We set this false at the bottom of switch
2889 statement, to which we get only if a particular path doesn't
2890 match the empty string. */
2891 boolean path_can_be_null = true;
2893 /* We aren't doing a `succeed_n' to begin with. */
2894 boolean succeed_n_p = false;
2896 assert (fastmap != NULL && p != NULL);
2899 bzero (fastmap, 1 << BYTEWIDTH); /* Assume nothing's valid. */
2900 bufp->fastmap_accurate = 1; /* It will be when we're done. */
2901 bufp->can_be_null = 0;
2905 if (p == pend || *p == succeed)
2907 /* We have reached the (effective) end of pattern. */
2908 if (!FAIL_STACK_EMPTY ())
2910 bufp->can_be_null |= path_can_be_null;
2912 /* Reset for next path. */
2913 path_can_be_null = true;
2915 p = fail_stack.stack[--fail_stack.avail].pointer;
2923 /* We should never be about to go beyond the end of the pattern. */
2926 switch (SWITCH_ENUM_CAST ((re_opcode_t) *p++))
2929 /* I guess the idea here is to simply not bother with a fastmap
2930 if a backreference is used, since it's too hard to figure out
2931 the fastmap for the corresponding group. Setting
2932 `can_be_null' stops `re_search_2' from using the fastmap, so
2933 that is all we do. */
2935 bufp->can_be_null = 1;
2939 /* Following are the cases which match a character. These end
2948 for (j = *p++ * BYTEWIDTH - 1; j >= 0; j--)
2949 if (p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH)))
2955 /* Chars beyond end of map must be allowed. */
2956 for (j = *p * BYTEWIDTH; j < (1 << BYTEWIDTH); j++)
2959 for (j = *p++ * BYTEWIDTH - 1; j >= 0; j--)
2960 if (!(p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH))))
2966 for (j = 0; j < (1 << BYTEWIDTH); j++)
2967 if (SYNTAX (j) == Sword)
2973 for (j = 0; j < (1 << BYTEWIDTH); j++)
2974 if (SYNTAX (j) != Sword)
2981 int fastmap_newline = fastmap['\n'];
2983 /* `.' matches anything ... */
2984 for (j = 0; j < (1 << BYTEWIDTH); j++)
2987 /* ... except perhaps newline. */
2988 if (!(bufp->syntax & RE_DOT_NEWLINE))
2989 fastmap['\n'] = fastmap_newline;
2991 /* Return if we have already set `can_be_null'; if we have,
2992 then the fastmap is irrelevant. Something's wrong here. */
2993 else if (bufp->can_be_null)
2996 /* Otherwise, have to check alternative paths. */
3003 for (j = 0; j < (1 << BYTEWIDTH); j++)
3004 if (SYNTAX (j) == (enum syntaxcode) k)
3011 for (j = 0; j < (1 << BYTEWIDTH); j++)
3012 if (SYNTAX (j) != (enum syntaxcode) k)
3017 /* All cases after this match the empty string. These end with
3025 #endif /* not emacs */
3037 case push_dummy_failure:
3042 case pop_failure_jump:
3043 case maybe_pop_jump:
3046 case dummy_failure_jump:
3047 EXTRACT_NUMBER_AND_INCR (j, p);
3052 /* Jump backward implies we just went through the body of a
3053 loop and matched nothing. Opcode jumped to should be
3054 `on_failure_jump' or `succeed_n'. Just treat it like an
3055 ordinary jump. For a * loop, it has pushed its failure
3056 point already; if so, discard that as redundant. */
3057 if ((re_opcode_t) *p != on_failure_jump
3058 && (re_opcode_t) *p != succeed_n)
3062 EXTRACT_NUMBER_AND_INCR (j, p);
3065 /* If what's on the stack is where we are now, pop it. */
3066 if (!FAIL_STACK_EMPTY ()
3067 && fail_stack.stack[fail_stack.avail - 1].pointer == p)
3073 case on_failure_jump:
3074 case on_failure_keep_string_jump:
3075 handle_on_failure_jump:
3076 EXTRACT_NUMBER_AND_INCR (j, p);
3078 /* For some patterns, e.g., `(a?)?', `p+j' here points to the
3079 end of the pattern. We don't want to push such a point,
3080 since when we restore it above, entering the switch will
3081 increment `p' past the end of the pattern. We don't need
3082 to push such a point since we obviously won't find any more
3083 fastmap entries beyond `pend'. Such a pattern can match
3084 the null string, though. */
3087 if (!PUSH_PATTERN_OP (p + j, fail_stack))
3089 RESET_FAIL_STACK ();
3094 bufp->can_be_null = 1;
3098 EXTRACT_NUMBER_AND_INCR (k, p); /* Skip the n. */
3099 succeed_n_p = false;
3106 /* Get to the number of times to succeed. */
3109 /* Increment p past the n for when k != 0. */
3110 EXTRACT_NUMBER_AND_INCR (k, p);
3114 succeed_n_p = true; /* Spaghetti code alert. */
3115 goto handle_on_failure_jump;
3132 abort (); /* We have listed all the cases. */
3135 /* Getting here means we have found the possible starting
3136 characters for one path of the pattern -- and that the empty
3137 string does not match. We need not follow this path further.
3138 Instead, look at the next alternative (remembered on the
3139 stack), or quit if no more. The test at the top of the loop
3140 does these things. */
3141 path_can_be_null = false;
3145 /* Set `can_be_null' for the last path (also the first path, if the
3146 pattern is empty). */
3147 bufp->can_be_null |= path_can_be_null;
3150 RESET_FAIL_STACK ();
3152 } /* re_compile_fastmap */
3154 /* Set REGS to hold NUM_REGS registers, storing them in STARTS and
3155 ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use
3156 this memory for recording register information. STARTS and ENDS
3157 must be allocated using the malloc library routine, and must each
3158 be at least NUM_REGS * sizeof (regoff_t) bytes long.
3160 If NUM_REGS == 0, then subsequent matches should allocate their own
3163 Unless this function is called, the first search or match using
3164 PATTERN_BUFFER will allocate its own register data, without
3165 freeing the old data. */
3168 re_set_registers (bufp, regs, num_regs, starts, ends)
3169 struct re_pattern_buffer *bufp;
3170 struct re_registers *regs;
3172 regoff_t *starts, *ends;
3176 bufp->regs_allocated = REGS_REALLOCATE;
3177 regs->num_regs = num_regs;
3178 regs->start = starts;
3183 bufp->regs_allocated = REGS_UNALLOCATED;
3185 regs->start = regs->end = (regoff_t *) 0;
3189 /* Searching routines. */
3191 /* Like re_search_2, below, but only one string is specified, and
3192 doesn't let you say where to stop matching. */
3195 re_search (bufp, string, size, startpos, range, regs)
3196 struct re_pattern_buffer *bufp;
3198 int size, startpos, range;
3199 struct re_registers *regs;
3201 return re_search_2 (bufp, NULL, 0, string, size, startpos, range,
3206 /* Using the compiled pattern in BUFP->buffer, first tries to match the
3207 virtual concatenation of STRING1 and STRING2, starting first at index
3208 STARTPOS, then at STARTPOS + 1, and so on.
3210 STRING1 and STRING2 have length SIZE1 and SIZE2, respectively.
3212 RANGE is how far to scan while trying to match. RANGE = 0 means try
3213 only at STARTPOS; in general, the last start tried is STARTPOS +
3216 In REGS, return the indices of the virtual concatenation of STRING1
3217 and STRING2 that matched the entire BUFP->buffer and its contained
3220 Do not consider matching one past the index STOP in the virtual
3221 concatenation of STRING1 and STRING2.
3223 We return either the position in the strings at which the match was
3224 found, -1 if no match, or -2 if error (such as failure
3228 re_search_2 (bufp, string1, size1, string2, size2, startpos, range, regs, stop)
3229 struct re_pattern_buffer *bufp;
3230 const char *string1, *string2;
3234 struct re_registers *regs;
3238 register char *fastmap = bufp->fastmap;
3239 register char *translate = bufp->translate;
3240 int total_size = size1 + size2;
3241 int endpos = startpos + range;
3243 /* Check for out-of-range STARTPOS. */
3244 if (startpos < 0 || startpos > total_size)
3247 /* Fix up RANGE if it might eventually take us outside
3248 the virtual concatenation of STRING1 and STRING2. */
3250 range = -1 - startpos;
3251 else if (endpos > total_size)
3252 range = total_size - startpos;
3254 /* If the search isn't to be a backwards one, don't waste time in a
3255 search for a pattern that must be anchored. */
3256 if (bufp->used > 0 && (re_opcode_t) bufp->buffer[0] == begbuf && range > 0)
3264 /* Update the fastmap now if not correct already. */
3265 if (fastmap && !bufp->fastmap_accurate)
3266 if (re_compile_fastmap (bufp) == -2)
3269 /* Loop through the string, looking for a place to start matching. */
3272 /* If a fastmap is supplied, skip quickly over characters that
3273 cannot be the start of a match. If the pattern can match the
3274 null string, however, we don't need to skip characters; we want
3275 the first null string. */
3276 if (fastmap && startpos < total_size && !bufp->can_be_null)
3278 if (range > 0) /* Searching forwards. */
3280 register const char *d;
3281 register int lim = 0;
3284 if (startpos < size1 && startpos + range >= size1)
3285 lim = range - (size1 - startpos);
3287 d = (startpos >= size1 ? string2 - size1 : string1) + startpos;
3289 /* Written out as an if-else to avoid testing `translate'
3293 && !fastmap[(unsigned char)
3294 translate[(unsigned char) *d++]])
3297 while (range > lim && !fastmap[(unsigned char) *d++])
3300 startpos += irange - range;
3302 else /* Searching backwards. */
3304 register char c = (size1 == 0 || startpos >= size1
3305 ? string2[startpos - size1]
3306 : string1[startpos]);
3308 if (!fastmap[(unsigned char) TRANSLATE (c)])
3313 /* If can't match the null string, and that's all we have left, fail. */
3314 if (range >= 0 && startpos == total_size && fastmap
3315 && !bufp->can_be_null)
3318 val = re_match_2_internal (bufp, string1, size1, string2, size2,
3319 startpos, regs, stop);
3320 #ifndef REGEX_MALLOC
3349 /* Declarations and macros for re_match_2. */
3351 static int bcmp_translate ();
3352 static boolean alt_match_null_string_p (),
3353 common_op_match_null_string_p (),
3354 group_match_null_string_p ();
3356 /* This converts PTR, a pointer into one of the search strings `string1'
3357 and `string2' into an offset from the beginning of that string. */
3358 #define POINTER_TO_OFFSET(ptr) \
3359 (FIRST_STRING_P (ptr) \
3360 ? ((regoff_t) ((ptr) - string1)) \
3361 : ((regoff_t) ((ptr) - string2 + size1)))
3363 /* Macros for dealing with the split strings in re_match_2. */
3365 #define MATCHING_IN_FIRST_STRING (dend == end_match_1)
3367 /* Call before fetching a character with *d. This switches over to
3368 string2 if necessary. */
3369 #define PREFETCH() \
3372 /* End of string2 => fail. */ \
3373 if (dend == end_match_2) \
3375 /* End of string1 => advance to string2. */ \
3377 dend = end_match_2; \
3381 /* Test if at very beginning or at very end of the virtual concatenation
3382 of `string1' and `string2'. If only one string, it's `string2'. */
3383 #define AT_STRINGS_BEG(d) ((d) == (size1 ? string1 : string2) || !size2)
3384 #define AT_STRINGS_END(d) ((d) == end2)
3387 /* Test if D points to a character which is word-constituent. We have
3388 two special cases to check for: if past the end of string1, look at
3389 the first character in string2; and if before the beginning of
3390 string2, look at the last character in string1. */
3391 #define WORDCHAR_P(d) \
3392 (SYNTAX ((d) == end1 ? *string2 \
3393 : (d) == string2 - 1 ? *(end1 - 1) : *(d)) \
3396 /* Test if the character before D and the one at D differ with respect
3397 to being word-constituent. */
3398 #define AT_WORD_BOUNDARY(d) \
3399 (AT_STRINGS_BEG (d) || AT_STRINGS_END (d) \
3400 || WORDCHAR_P (d - 1) != WORDCHAR_P (d))
3403 /* Free everything we malloc. */
3404 #ifdef MATCH_MAY_ALLOCATE
3405 #define FREE_VAR(var) if (var) REGEX_FREE (var); var = NULL
3406 #define FREE_VARIABLES() \
3408 REGEX_FREE_STACK (fail_stack.stack); \
3409 FREE_VAR (regstart); \
3410 FREE_VAR (regend); \
3411 FREE_VAR (old_regstart); \
3412 FREE_VAR (old_regend); \
3413 FREE_VAR (best_regstart); \
3414 FREE_VAR (best_regend); \
3415 FREE_VAR (reg_info); \
3416 FREE_VAR (reg_dummy); \
3417 FREE_VAR (reg_info_dummy); \
3420 #define FREE_VARIABLES() ((void)0) /* Do nothing! But inhibit gcc warning. */
3421 #endif /* not MATCH_MAY_ALLOCATE */
3423 /* These values must meet several constraints. They must not be valid
3424 register values; since we have a limit of 255 registers (because
3425 we use only one byte in the pattern for the register number), we can
3426 use numbers larger than 255. They must differ by 1, because of
3427 NUM_FAILURE_ITEMS above. And the value for the lowest register must
3428 be larger than the value for the highest register, so we do not try
3429 to actually save any registers when none are active. */
3430 #define NO_HIGHEST_ACTIVE_REG (1 << BYTEWIDTH)
3431 #define NO_LOWEST_ACTIVE_REG (NO_HIGHEST_ACTIVE_REG + 1)
3433 /* Matching routines. */
3435 #ifndef emacs /* Emacs never uses this. */
3436 /* re_match is like re_match_2 except it takes only a single string. */
3439 re_match (bufp, string, size, pos, regs)
3440 struct re_pattern_buffer *bufp;
3443 struct re_registers *regs;
3445 int result = re_match_2_internal (bufp, NULL, 0, string, size,
3450 #endif /* not emacs */
3453 /* re_match_2 matches the compiled pattern in BUFP against the
3454 the (virtual) concatenation of STRING1 and STRING2 (of length SIZE1
3455 and SIZE2, respectively). We start matching at POS, and stop
3458 If REGS is non-null and the `no_sub' field of BUFP is nonzero, we
3459 store offsets for the substring each group matched in REGS. See the
3460 documentation for exactly how many groups we fill.
3462 We return -1 if no match, -2 if an internal error (such as the
3463 failure stack overflowing). Otherwise, we return the length of the
3464 matched substring. */
3467 re_match_2 (bufp, string1, size1, string2, size2, pos, regs, stop)
3468 struct re_pattern_buffer *bufp;
3469 const char *string1, *string2;
3472 struct re_registers *regs;
3475 int result = re_match_2_internal (bufp, string1, size1, string2, size2,
3481 /* This is a separate function so that we can force an alloca cleanup
3484 re_match_2_internal (bufp, string1, size1, string2, size2, pos, regs, stop)
3485 struct re_pattern_buffer *bufp;
3486 const char *string1, *string2;
3489 struct re_registers *regs;
3492 /* General temporaries. */
3496 /* Just past the end of the corresponding string. */
3497 const char *end1, *end2;
3499 /* Pointers into string1 and string2, just past the last characters in
3500 each to consider matching. */
3501 const char *end_match_1, *end_match_2;
3503 /* Where we are in the data, and the end of the current string. */
3504 const char *d, *dend;
3506 /* Where we are in the pattern, and the end of the pattern. */
3507 unsigned char *p = bufp->buffer;
3508 register unsigned char *pend = p + bufp->used;
3510 /* Mark the opcode just after a start_memory, so we can test for an
3511 empty subpattern when we get to the stop_memory. */
3512 unsigned char *just_past_start_mem = 0;
3514 /* We use this to map every character in the string. */
3515 char *translate = bufp->translate;
3517 /* Failure point stack. Each place that can handle a failure further
3518 down the line pushes a failure point on this stack. It consists of
3519 restart, regend, and reg_info for all registers corresponding to
3520 the subexpressions we're currently inside, plus the number of such
3521 registers, and, finally, two char *'s. The first char * is where
3522 to resume scanning the pattern; the second one is where to resume
3523 scanning the strings. If the latter is zero, the failure point is
3524 a ``dummy''; if a failure happens and the failure point is a dummy,
3525 it gets discarded and the next next one is tried. */
3526 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
3527 fail_stack_type fail_stack;
3530 static unsigned failure_id = 0;
3531 unsigned nfailure_points_pushed = 0, nfailure_points_popped = 0;
3534 /* This holds the pointer to the failure stack, when
3535 it is allocated relocatably. */
3537 fail_stack_elt_t *failure_stack_ptr;
3540 /* We fill all the registers internally, independent of what we
3541 return, for use in backreferences. The number here includes
3542 an element for register zero. */
3543 unsigned num_regs = bufp->re_nsub + 1;
3545 /* The currently active registers. */
3546 unsigned lowest_active_reg = NO_LOWEST_ACTIVE_REG;
3547 unsigned highest_active_reg = NO_HIGHEST_ACTIVE_REG;
3549 /* Information on the contents of registers. These are pointers into
3550 the input strings; they record just what was matched (on this
3551 attempt) by a subexpression part of the pattern, that is, the
3552 regnum-th regstart pointer points to where in the pattern we began
3553 matching and the regnum-th regend points to right after where we
3554 stopped matching the regnum-th subexpression. (The zeroth register
3555 keeps track of what the whole pattern matches.) */
3556 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
3557 const char **regstart, **regend;
3560 /* If a group that's operated upon by a repetition operator fails to
3561 match anything, then the register for its start will need to be
3562 restored because it will have been set to wherever in the string we
3563 are when we last see its open-group operator. Similarly for a
3565 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
3566 const char **old_regstart, **old_regend;
3569 /* The is_active field of reg_info helps us keep track of which (possibly
3570 nested) subexpressions we are currently in. The matched_something
3571 field of reg_info[reg_num] helps us tell whether or not we have
3572 matched any of the pattern so far this time through the reg_num-th
3573 subexpression. These two fields get reset each time through any
3574 loop their register is in. */
3575 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
3576 register_info_type *reg_info;
3579 /* The following record the register info as found in the above
3580 variables when we find a match better than any we've seen before.
3581 This happens as we backtrack through the failure points, which in
3582 turn happens only if we have not yet matched the entire string. */
3583 unsigned best_regs_set = false;
3584 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
3585 const char **best_regstart, **best_regend;
3588 /* Logically, this is `best_regend[0]'. But we don't want to have to
3589 allocate space for that if we're not allocating space for anything
3590 else (see below). Also, we never need info about register 0 for
3591 any of the other register vectors, and it seems rather a kludge to
3592 treat `best_regend' differently than the rest. So we keep track of
3593 the end of the best match so far in a separate variable. We
3594 initialize this to NULL so that when we backtrack the first time
3595 and need to test it, it's not garbage. */
3596 const char *match_end = NULL;
3598 /* This helps SET_REGS_MATCHED avoid doing redundant work. */
3599 int set_regs_matched_done = 0;
3601 /* Used when we pop values we don't care about. */
3602 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
3603 const char **reg_dummy;
3604 register_info_type *reg_info_dummy;
3608 /* Counts the total number of registers pushed. */
3609 unsigned num_regs_pushed = 0;
3612 DEBUG_PRINT1 ("\n\nEntering re_match_2.\n");
3616 #ifdef MATCH_MAY_ALLOCATE
3617 /* Do not bother to initialize all the register variables if there are
3618 no groups in the pattern, as it takes a fair amount of time. If
3619 there are groups, we include space for register 0 (the whole
3620 pattern), even though we never use it, since it simplifies the
3621 array indexing. We should fix this. */
3624 regstart = REGEX_TALLOC (num_regs, const char *);
3625 regend = REGEX_TALLOC (num_regs, const char *);
3626 old_regstart = REGEX_TALLOC (num_regs, const char *);
3627 old_regend = REGEX_TALLOC (num_regs, const char *);
3628 best_regstart = REGEX_TALLOC (num_regs, const char *);
3629 best_regend = REGEX_TALLOC (num_regs, const char *);
3630 reg_info = REGEX_TALLOC (num_regs, register_info_type);
3631 reg_dummy = REGEX_TALLOC (num_regs, const char *);
3632 reg_info_dummy = REGEX_TALLOC (num_regs, register_info_type);
3634 if (!(regstart && regend && old_regstart && old_regend && reg_info
3635 && best_regstart && best_regend && reg_dummy && reg_info_dummy))
3643 /* We must initialize all our variables to NULL, so that
3644 `FREE_VARIABLES' doesn't try to free them. */
3645 regstart = regend = old_regstart = old_regend = best_regstart
3646 = best_regend = reg_dummy = NULL;
3647 reg_info = reg_info_dummy = (register_info_type *) NULL;
3649 #endif /* MATCH_MAY_ALLOCATE */
3651 /* The starting position is bogus. */
3652 if (pos < 0 || pos > size1 + size2)
3658 /* Initialize subexpression text positions to -1 to mark ones that no
3659 start_memory/stop_memory has been seen for. Also initialize the
3660 register information struct. */
3661 for (mcnt = 1; mcnt < num_regs; mcnt++)
3663 regstart[mcnt] = regend[mcnt]
3664 = old_regstart[mcnt] = old_regend[mcnt] = REG_UNSET_VALUE;
3666 REG_MATCH_NULL_STRING_P (reg_info[mcnt]) = MATCH_NULL_UNSET_VALUE;
3667 IS_ACTIVE (reg_info[mcnt]) = 0;
3668 MATCHED_SOMETHING (reg_info[mcnt]) = 0;
3669 EVER_MATCHED_SOMETHING (reg_info[mcnt]) = 0;
3672 /* We move `string1' into `string2' if the latter's empty -- but not if
3673 `string1' is null. */
3674 if (size2 == 0 && string1 != NULL)
3681 end1 = string1 + size1;
3682 end2 = string2 + size2;
3684 /* Compute where to stop matching, within the two strings. */
3687 end_match_1 = string1 + stop;
3688 end_match_2 = string2;
3693 end_match_2 = string2 + stop - size1;
3696 /* `p' scans through the pattern as `d' scans through the data.
3697 `dend' is the end of the input string that `d' points within. `d'
3698 is advanced into the following input string whenever necessary, but
3699 this happens before fetching; therefore, at the beginning of the
3700 loop, `d' can be pointing at the end of a string, but it cannot
3702 if (size1 > 0 && pos <= size1)
3709 d = string2 + pos - size1;
3713 DEBUG_PRINT1 ("The compiled pattern is: ");
3714 DEBUG_PRINT_COMPILED_PATTERN (bufp, p, pend);
3715 DEBUG_PRINT1 ("The string to match is: `");
3716 DEBUG_PRINT_DOUBLE_STRING (d, string1, size1, string2, size2);
3717 DEBUG_PRINT1 ("'\n");
3719 /* This loops over pattern commands. It exits by returning from the
3720 function if the match is complete, or it drops through if the match
3721 fails at this starting point in the input data. */
3724 DEBUG_PRINT2 ("\n0x%x: ", p);
3727 { /* End of pattern means we might have succeeded. */
3728 DEBUG_PRINT1 ("end of pattern ... ");
3730 /* If we haven't matched the entire string, and we want the
3731 longest match, try backtracking. */
3732 if (d != end_match_2)
3734 /* 1 if this match ends in the same string (string1 or string2)
3735 as the best previous match. */
3736 boolean same_str_p = (FIRST_STRING_P (match_end)
3737 == MATCHING_IN_FIRST_STRING);
3738 /* 1 if this match is the best seen so far. */
3739 boolean best_match_p;
3741 /* AIX compiler got confused when this was combined
3742 with the previous declaration. */
3744 best_match_p = d > match_end;
3746 best_match_p = !MATCHING_IN_FIRST_STRING;
3748 DEBUG_PRINT1 ("backtracking.\n");
3750 if (!FAIL_STACK_EMPTY ())
3751 { /* More failure points to try. */
3753 /* If exceeds best match so far, save it. */
3754 if (!best_regs_set || best_match_p)
3756 best_regs_set = true;
3759 DEBUG_PRINT1 ("\nSAVING match as best so far.\n");
3761 for (mcnt = 1; mcnt < num_regs; mcnt++)
3763 best_regstart[mcnt] = regstart[mcnt];
3764 best_regend[mcnt] = regend[mcnt];
3770 /* If no failure points, don't restore garbage. And if
3771 last match is real best match, don't restore second
3773 else if (best_regs_set && !best_match_p)
3776 /* Restore best match. It may happen that `dend ==
3777 end_match_1' while the restored d is in string2.
3778 For example, the pattern `x.*y.*z' against the
3779 strings `x-' and `y-z-', if the two strings are
3780 not consecutive in memory. */
3781 DEBUG_PRINT1 ("Restoring best registers.\n");
3784 dend = ((d >= string1 && d <= end1)
3785 ? end_match_1 : end_match_2);
3787 for (mcnt = 1; mcnt < num_regs; mcnt++)
3789 regstart[mcnt] = best_regstart[mcnt];
3790 regend[mcnt] = best_regend[mcnt];
3793 } /* d != end_match_2 */
3796 DEBUG_PRINT1 ("Accepting match.\n");
3798 /* If caller wants register contents data back, do it. */
3799 if (regs && !bufp->no_sub)
3801 /* Have the register data arrays been allocated? */
3802 if (bufp->regs_allocated == REGS_UNALLOCATED)
3803 { /* No. So allocate them with malloc. We need one
3804 extra element beyond `num_regs' for the `-1' marker
3806 regs->num_regs = MAX (RE_NREGS, num_regs + 1);
3807 regs->start = TALLOC (regs->num_regs, regoff_t);
3808 regs->end = TALLOC (regs->num_regs, regoff_t);
3809 if (regs->start == NULL || regs->end == NULL)
3814 bufp->regs_allocated = REGS_REALLOCATE;
3816 else if (bufp->regs_allocated == REGS_REALLOCATE)
3817 { /* Yes. If we need more elements than were already
3818 allocated, reallocate them. If we need fewer, just
3820 if (regs->num_regs < num_regs + 1)
3822 regs->num_regs = num_regs + 1;
3823 RETALLOC (regs->start, regs->num_regs, regoff_t);
3824 RETALLOC (regs->end, regs->num_regs, regoff_t);
3825 if (regs->start == NULL || regs->end == NULL)
3834 /* These braces fend off a "empty body in an else-statement"
3835 warning under GCC when assert expands to nothing. */
3836 assert (bufp->regs_allocated == REGS_FIXED);
3839 /* Convert the pointer data in `regstart' and `regend' to
3840 indices. Register zero has to be set differently,
3841 since we haven't kept track of any info for it. */
3842 if (regs->num_regs > 0)
3844 regs->start[0] = pos;
3845 regs->end[0] = (MATCHING_IN_FIRST_STRING
3846 ? ((regoff_t) (d - string1))
3847 : ((regoff_t) (d - string2 + size1)));
3850 /* Go through the first `min (num_regs, regs->num_regs)'
3851 registers, since that is all we initialized. */
3852 for (mcnt = 1; mcnt < MIN (num_regs, regs->num_regs); mcnt++)
3854 if (REG_UNSET (regstart[mcnt]) || REG_UNSET (regend[mcnt]))
3855 regs->start[mcnt] = regs->end[mcnt] = -1;
3859 = (regoff_t) POINTER_TO_OFFSET (regstart[mcnt]);
3861 = (regoff_t) POINTER_TO_OFFSET (regend[mcnt]);
3865 /* If the regs structure we return has more elements than
3866 were in the pattern, set the extra elements to -1. If
3867 we (re)allocated the registers, this is the case,
3868 because we always allocate enough to have at least one
3870 for (mcnt = num_regs; mcnt < regs->num_regs; mcnt++)
3871 regs->start[mcnt] = regs->end[mcnt] = -1;
3872 } /* regs && !bufp->no_sub */
3874 DEBUG_PRINT4 ("%u failure points pushed, %u popped (%u remain).\n",
3875 nfailure_points_pushed, nfailure_points_popped,
3876 nfailure_points_pushed - nfailure_points_popped);
3877 DEBUG_PRINT2 ("%u registers pushed.\n", num_regs_pushed);
3879 mcnt = d - pos - (MATCHING_IN_FIRST_STRING
3883 DEBUG_PRINT2 ("Returning %d from re_match_2.\n", mcnt);
3889 /* Otherwise match next pattern command. */
3890 switch (SWITCH_ENUM_CAST ((re_opcode_t) *p++))
3892 /* Ignore these. Used to ignore the n of succeed_n's which
3893 currently have n == 0. */
3895 DEBUG_PRINT1 ("EXECUTING no_op.\n");
3899 DEBUG_PRINT1 ("EXECUTING succeed.\n");
3902 /* Match the next n pattern characters exactly. The following
3903 byte in the pattern defines n, and the n bytes after that
3904 are the characters to match. */
3907 DEBUG_PRINT2 ("EXECUTING exactn %d.\n", mcnt);
3909 /* This is written out as an if-else so we don't waste time
3910 testing `translate' inside the loop. */
3916 if (translate[(unsigned char) *d++] != (char) *p++)
3926 if (*d++ != (char) *p++) goto fail;
3930 SET_REGS_MATCHED ();
3934 /* Match any character except possibly a newline or a null. */
3936 DEBUG_PRINT1 ("EXECUTING anychar.\n");
3940 if ((!(bufp->syntax & RE_DOT_NEWLINE) && TRANSLATE (*d) == '\n')
3941 || (bufp->syntax & RE_DOT_NOT_NULL && TRANSLATE (*d) == '\000'))
3944 SET_REGS_MATCHED ();
3945 DEBUG_PRINT2 (" Matched `%d'.\n", *d);
3953 register unsigned char c;
3954 boolean not = (re_opcode_t) *(p - 1) == charset_not;
3956 DEBUG_PRINT2 ("EXECUTING charset%s.\n", not ? "_not" : "");
3959 c = TRANSLATE (*d); /* The character to match. */
3961 /* Cast to `unsigned' instead of `unsigned char' in case the
3962 bit list is a full 32 bytes long. */
3963 if (c < (unsigned) (*p * BYTEWIDTH)
3964 && p[1 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH)))
3969 if (!not) goto fail;
3971 SET_REGS_MATCHED ();
3977 /* The beginning of a group is represented by start_memory.
3978 The arguments are the register number in the next byte, and the
3979 number of groups inner to this one in the next. The text
3980 matched within the group is recorded (in the internal
3981 registers data structure) under the register number. */
3983 DEBUG_PRINT3 ("EXECUTING start_memory %d (%d):\n", *p, p[1]);
3985 /* Find out if this group can match the empty string. */
3986 p1 = p; /* To send to group_match_null_string_p. */
3988 if (REG_MATCH_NULL_STRING_P (reg_info[*p]) == MATCH_NULL_UNSET_VALUE)
3989 REG_MATCH_NULL_STRING_P (reg_info[*p])
3990 = group_match_null_string_p (&p1, pend, reg_info);
3992 /* Save the position in the string where we were the last time
3993 we were at this open-group operator in case the group is
3994 operated upon by a repetition operator, e.g., with `(a*)*b'
3995 against `ab'; then we want to ignore where we are now in
3996 the string in case this attempt to match fails. */
3997 old_regstart[*p] = REG_MATCH_NULL_STRING_P (reg_info[*p])
3998 ? REG_UNSET (regstart[*p]) ? d : regstart[*p]
4000 DEBUG_PRINT2 (" old_regstart: %d\n",
4001 POINTER_TO_OFFSET (old_regstart[*p]));
4004 DEBUG_PRINT2 (" regstart: %d\n", POINTER_TO_OFFSET (regstart[*p]));
4006 IS_ACTIVE (reg_info[*p]) = 1;
4007 MATCHED_SOMETHING (reg_info[*p]) = 0;
4009 /* Clear this whenever we change the register activity status. */
4010 set_regs_matched_done = 0;
4012 /* This is the new highest active register. */
4013 highest_active_reg = *p;
4015 /* If nothing was active before, this is the new lowest active
4017 if (lowest_active_reg == NO_LOWEST_ACTIVE_REG)
4018 lowest_active_reg = *p;
4020 /* Move past the register number and inner group count. */
4022 just_past_start_mem = p;
4027 /* The stop_memory opcode represents the end of a group. Its
4028 arguments are the same as start_memory's: the register
4029 number, and the number of inner groups. */
4031 DEBUG_PRINT3 ("EXECUTING stop_memory %d (%d):\n", *p, p[1]);
4033 /* We need to save the string position the last time we were at
4034 this close-group operator in case the group is operated
4035 upon by a repetition operator, e.g., with `((a*)*(b*)*)*'
4036 against `aba'; then we want to ignore where we are now in
4037 the string in case this attempt to match fails. */
4038 old_regend[*p] = REG_MATCH_NULL_STRING_P (reg_info[*p])
4039 ? REG_UNSET (regend[*p]) ? d : regend[*p]
4041 DEBUG_PRINT2 (" old_regend: %d\n",
4042 POINTER_TO_OFFSET (old_regend[*p]));
4045 DEBUG_PRINT2 (" regend: %d\n", POINTER_TO_OFFSET (regend[*p]));
4047 /* This register isn't active anymore. */
4048 IS_ACTIVE (reg_info[*p]) = 0;
4050 /* Clear this whenever we change the register activity status. */
4051 set_regs_matched_done = 0;
4053 /* If this was the only register active, nothing is active
4055 if (lowest_active_reg == highest_active_reg)
4057 lowest_active_reg = NO_LOWEST_ACTIVE_REG;
4058 highest_active_reg = NO_HIGHEST_ACTIVE_REG;
4061 { /* We must scan for the new highest active register, since
4062 it isn't necessarily one less than now: consider
4063 (a(b)c(d(e)f)g). When group 3 ends, after the f), the
4064 new highest active register is 1. */
4065 unsigned char r = *p - 1;
4066 while (r > 0 && !IS_ACTIVE (reg_info[r]))
4069 /* If we end up at register zero, that means that we saved
4070 the registers as the result of an `on_failure_jump', not
4071 a `start_memory', and we jumped to past the innermost
4072 `stop_memory'. For example, in ((.)*) we save
4073 registers 1 and 2 as a result of the *, but when we pop
4074 back to the second ), we are at the stop_memory 1.
4075 Thus, nothing is active. */
4078 lowest_active_reg = NO_LOWEST_ACTIVE_REG;
4079 highest_active_reg = NO_HIGHEST_ACTIVE_REG;
4082 highest_active_reg = r;
4085 /* If just failed to match something this time around with a
4086 group that's operated on by a repetition operator, try to
4087 force exit from the ``loop'', and restore the register
4088 information for this group that we had before trying this
4090 if ((!MATCHED_SOMETHING (reg_info[*p])
4091 || just_past_start_mem == p - 1)
4094 boolean is_a_jump_n = false;
4098 switch ((re_opcode_t) *p1++)
4102 case pop_failure_jump:
4103 case maybe_pop_jump:
4105 case dummy_failure_jump:
4106 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4116 /* If the next operation is a jump backwards in the pattern
4117 to an on_failure_jump right before the start_memory
4118 corresponding to this stop_memory, exit from the loop
4119 by forcing a failure after pushing on the stack the
4120 on_failure_jump's jump in the pattern, and d. */
4121 if (mcnt < 0 && (re_opcode_t) *p1 == on_failure_jump
4122 && (re_opcode_t) p1[3] == start_memory && p1[4] == *p)
4124 /* If this group ever matched anything, then restore
4125 what its registers were before trying this last
4126 failed match, e.g., with `(a*)*b' against `ab' for
4127 regstart[1], and, e.g., with `((a*)*(b*)*)*'
4128 against `aba' for regend[3].
4130 Also restore the registers for inner groups for,
4131 e.g., `((a*)(b*))*' against `aba' (register 3 would
4132 otherwise get trashed). */
4134 if (EVER_MATCHED_SOMETHING (reg_info[*p]))
4138 EVER_MATCHED_SOMETHING (reg_info[*p]) = 0;
4140 /* Restore this and inner groups' (if any) registers. */
4141 for (r = *p; r < *p + *(p + 1); r++)
4143 regstart[r] = old_regstart[r];
4145 /* xx why this test? */
4146 if (old_regend[r] >= regstart[r])
4147 regend[r] = old_regend[r];
4151 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4152 PUSH_FAILURE_POINT (p1 + mcnt, d, -2);
4158 /* Move past the register number and the inner group count. */
4163 /* \<digit> has been turned into a `duplicate' command which is
4164 followed by the numeric value of <digit> as the register number. */
4167 register const char *d2, *dend2;
4168 int regno = *p++; /* Get which register to match against. */
4169 DEBUG_PRINT2 ("EXECUTING duplicate %d.\n", regno);
4171 /* Can't back reference a group which we've never matched. */
4172 if (REG_UNSET (regstart[regno]) || REG_UNSET (regend[regno]))
4175 /* Where in input to try to start matching. */
4176 d2 = regstart[regno];
4178 /* Where to stop matching; if both the place to start and
4179 the place to stop matching are in the same string, then
4180 set to the place to stop, otherwise, for now have to use
4181 the end of the first string. */
4183 dend2 = ((FIRST_STRING_P (regstart[regno])
4184 == FIRST_STRING_P (regend[regno]))
4185 ? regend[regno] : end_match_1);
4188 /* If necessary, advance to next segment in register
4192 if (dend2 == end_match_2) break;
4193 if (dend2 == regend[regno]) break;
4195 /* End of string1 => advance to string2. */
4197 dend2 = regend[regno];
4199 /* At end of register contents => success */
4200 if (d2 == dend2) break;
4202 /* If necessary, advance to next segment in data. */
4205 /* How many characters left in this segment to match. */
4208 /* Want how many consecutive characters we can match in
4209 one shot, so, if necessary, adjust the count. */
4210 if (mcnt > dend2 - d2)
4213 /* Compare that many; failure if mismatch, else move
4216 ? bcmp_translate (d, d2, mcnt, translate)
4217 : bcmp (d, d2, mcnt))
4219 d += mcnt, d2 += mcnt;
4221 /* Do this because we've match some characters. */
4222 SET_REGS_MATCHED ();
4228 /* begline matches the empty string at the beginning of the string
4229 (unless `not_bol' is set in `bufp'), and, if
4230 `newline_anchor' is set, after newlines. */
4232 DEBUG_PRINT1 ("EXECUTING begline.\n");
4234 if (AT_STRINGS_BEG (d))
4236 if (!bufp->not_bol) break;
4238 else if (d[-1] == '\n' && bufp->newline_anchor)
4242 /* In all other cases, we fail. */
4246 /* endline is the dual of begline. */
4248 DEBUG_PRINT1 ("EXECUTING endline.\n");
4250 if (AT_STRINGS_END (d))
4252 if (!bufp->not_eol) break;
4255 /* We have to ``prefetch'' the next character. */
4256 else if ((d == end1 ? *string2 : *d) == '\n'
4257 && bufp->newline_anchor)
4264 /* Match at the very beginning of the data. */
4266 DEBUG_PRINT1 ("EXECUTING begbuf.\n");
4267 if (AT_STRINGS_BEG (d))
4272 /* Match at the very end of the data. */
4274 DEBUG_PRINT1 ("EXECUTING endbuf.\n");
4275 if (AT_STRINGS_END (d))
4280 /* on_failure_keep_string_jump is used to optimize `.*\n'. It
4281 pushes NULL as the value for the string on the stack. Then
4282 `pop_failure_point' will keep the current value for the
4283 string, instead of restoring it. To see why, consider
4284 matching `foo\nbar' against `.*\n'. The .* matches the foo;
4285 then the . fails against the \n. But the next thing we want
4286 to do is match the \n against the \n; if we restored the
4287 string value, we would be back at the foo.
4289 Because this is used only in specific cases, we don't need to
4290 check all the things that `on_failure_jump' does, to make
4291 sure the right things get saved on the stack. Hence we don't
4292 share its code. The only reason to push anything on the
4293 stack at all is that otherwise we would have to change
4294 `anychar's code to do something besides goto fail in this
4295 case; that seems worse than this. */
4296 case on_failure_keep_string_jump:
4297 DEBUG_PRINT1 ("EXECUTING on_failure_keep_string_jump");
4299 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4300 DEBUG_PRINT3 (" %d (to 0x%x):\n", mcnt, p + mcnt);
4302 PUSH_FAILURE_POINT (p + mcnt, NULL, -2);
4306 /* Uses of on_failure_jump:
4308 Each alternative starts with an on_failure_jump that points
4309 to the beginning of the next alternative. Each alternative
4310 except the last ends with a jump that in effect jumps past
4311 the rest of the alternatives. (They really jump to the
4312 ending jump of the following alternative, because tensioning
4313 these jumps is a hassle.)
4315 Repeats start with an on_failure_jump that points past both
4316 the repetition text and either the following jump or
4317 pop_failure_jump back to this on_failure_jump. */
4318 case on_failure_jump:
4320 DEBUG_PRINT1 ("EXECUTING on_failure_jump");
4322 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4323 DEBUG_PRINT3 (" %d (to 0x%x)", mcnt, p + mcnt);
4325 /* If this on_failure_jump comes right before a group (i.e.,
4326 the original * applied to a group), save the information
4327 for that group and all inner ones, so that if we fail back
4328 to this point, the group's information will be correct.
4329 For example, in \(a*\)*\1, we need the preceding group,
4330 and in \(\(a*\)b*\)\2, we need the inner group. */
4332 /* We can't use `p' to check ahead because we push
4333 a failure point to `p + mcnt' after we do this. */
4336 /* We need to skip no_op's before we look for the
4337 start_memory in case this on_failure_jump is happening as
4338 the result of a completed succeed_n, as in \(a\)\{1,3\}b\1
4340 while (p1 < pend && (re_opcode_t) *p1 == no_op)
4343 if (p1 < pend && (re_opcode_t) *p1 == start_memory)
4345 /* We have a new highest active register now. This will
4346 get reset at the start_memory we are about to get to,
4347 but we will have saved all the registers relevant to
4348 this repetition op, as described above. */
4349 highest_active_reg = *(p1 + 1) + *(p1 + 2);
4350 if (lowest_active_reg == NO_LOWEST_ACTIVE_REG)
4351 lowest_active_reg = *(p1 + 1);
4354 DEBUG_PRINT1 (":\n");
4355 PUSH_FAILURE_POINT (p + mcnt, d, -2);
4359 /* A smart repeat ends with `maybe_pop_jump'.
4360 We change it to either `pop_failure_jump' or `jump'. */
4361 case maybe_pop_jump:
4362 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4363 DEBUG_PRINT2 ("EXECUTING maybe_pop_jump %d.\n", mcnt);
4365 register unsigned char *p2 = p;
4367 /* Compare the beginning of the repeat with what in the
4368 pattern follows its end. If we can establish that there
4369 is nothing that they would both match, i.e., that we
4370 would have to backtrack because of (as in, e.g., `a*a')
4371 then we can change to pop_failure_jump, because we'll
4372 never have to backtrack.
4374 This is not true in the case of alternatives: in
4375 `(a|ab)*' we do need to backtrack to the `ab' alternative
4376 (e.g., if the string was `ab'). But instead of trying to
4377 detect that here, the alternative has put on a dummy
4378 failure point which is what we will end up popping. */
4380 /* Skip over open/close-group commands.
4381 If what follows this loop is a ...+ construct,
4382 look at what begins its body, since we will have to
4383 match at least one of that. */
4387 && ((re_opcode_t) *p2 == stop_memory
4388 || (re_opcode_t) *p2 == start_memory))
4390 else if (p2 + 6 < pend
4391 && (re_opcode_t) *p2 == dummy_failure_jump)
4398 /* p1[0] ... p1[2] are the `on_failure_jump' corresponding
4399 to the `maybe_finalize_jump' of this case. Examine what
4402 /* If we're at the end of the pattern, we can change. */
4405 /* Consider what happens when matching ":\(.*\)"
4406 against ":/". I don't really understand this code
4408 p[-3] = (unsigned char) pop_failure_jump;
4410 (" End of pattern: change to `pop_failure_jump'.\n");
4413 else if ((re_opcode_t) *p2 == exactn
4414 || (bufp->newline_anchor && (re_opcode_t) *p2 == endline))
4416 register unsigned char c
4417 = *p2 == (unsigned char) endline ? '\n' : p2[2];
4419 if ((re_opcode_t) p1[3] == exactn && p1[5] != c)
4421 p[-3] = (unsigned char) pop_failure_jump;
4422 DEBUG_PRINT3 (" %c != %c => pop_failure_jump.\n",
4426 else if ((re_opcode_t) p1[3] == charset
4427 || (re_opcode_t) p1[3] == charset_not)
4429 int not = (re_opcode_t) p1[3] == charset_not;
4431 if (c < (unsigned char) (p1[4] * BYTEWIDTH)
4432 && p1[5 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH)))
4435 /* `not' is equal to 1 if c would match, which means
4436 that we can't change to pop_failure_jump. */
4439 p[-3] = (unsigned char) pop_failure_jump;
4440 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
4444 else if ((re_opcode_t) *p2 == charset)
4447 register unsigned char c
4448 = *p2 == (unsigned char) endline ? '\n' : p2[2];
4451 if ((re_opcode_t) p1[3] == exactn
4452 && ! ((int) p2[1] * BYTEWIDTH > (int) p1[4]
4453 && (p2[1 + p1[4] / BYTEWIDTH]
4454 & (1 << (p1[4] % BYTEWIDTH)))))
4456 p[-3] = (unsigned char) pop_failure_jump;
4457 DEBUG_PRINT3 (" %c != %c => pop_failure_jump.\n",
4461 else if ((re_opcode_t) p1[3] == charset_not)
4464 /* We win if the charset_not inside the loop
4465 lists every character listed in the charset after. */
4466 for (idx = 0; idx < (int) p2[1]; idx++)
4467 if (! (p2[2 + idx] == 0
4468 || (idx < (int) p1[4]
4469 && ((p2[2 + idx] & ~ p1[5 + idx]) == 0))))
4474 p[-3] = (unsigned char) pop_failure_jump;
4475 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
4478 else if ((re_opcode_t) p1[3] == charset)
4481 /* We win if the charset inside the loop
4482 has no overlap with the one after the loop. */
4484 idx < (int) p2[1] && idx < (int) p1[4];
4486 if ((p2[2 + idx] & p1[5 + idx]) != 0)
4489 if (idx == p2[1] || idx == p1[4])
4491 p[-3] = (unsigned char) pop_failure_jump;
4492 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
4497 p -= 2; /* Point at relative address again. */
4498 if ((re_opcode_t) p[-1] != pop_failure_jump)
4500 p[-1] = (unsigned char) jump;
4501 DEBUG_PRINT1 (" Match => jump.\n");
4502 goto unconditional_jump;
4504 /* Note fall through. */
4507 /* The end of a simple repeat has a pop_failure_jump back to
4508 its matching on_failure_jump, where the latter will push a
4509 failure point. The pop_failure_jump takes off failure
4510 points put on by this pop_failure_jump's matching
4511 on_failure_jump; we got through the pattern to here from the
4512 matching on_failure_jump, so didn't fail. */
4513 case pop_failure_jump:
4515 /* We need to pass separate storage for the lowest and
4516 highest registers, even though we don't care about the
4517 actual values. Otherwise, we will restore only one
4518 register from the stack, since lowest will == highest in
4519 `pop_failure_point'. */
4520 unsigned dummy_low_reg, dummy_high_reg;
4521 unsigned char *pdummy;
4524 DEBUG_PRINT1 ("EXECUTING pop_failure_jump.\n");
4525 POP_FAILURE_POINT (sdummy, pdummy,
4526 dummy_low_reg, dummy_high_reg,
4527 reg_dummy, reg_dummy, reg_info_dummy);
4529 /* Note fall through. */
4532 /* Unconditionally jump (without popping any failure points). */
4535 EXTRACT_NUMBER_AND_INCR (mcnt, p); /* Get the amount to jump. */
4536 DEBUG_PRINT2 ("EXECUTING jump %d ", mcnt);
4537 p += mcnt; /* Do the jump. */
4538 DEBUG_PRINT2 ("(to 0x%x).\n", p);
4542 /* We need this opcode so we can detect where alternatives end
4543 in `group_match_null_string_p' et al. */
4545 DEBUG_PRINT1 ("EXECUTING jump_past_alt.\n");
4546 goto unconditional_jump;
4549 /* Normally, the on_failure_jump pushes a failure point, which
4550 then gets popped at pop_failure_jump. We will end up at
4551 pop_failure_jump, also, and with a pattern of, say, `a+', we
4552 are skipping over the on_failure_jump, so we have to push
4553 something meaningless for pop_failure_jump to pop. */
4554 case dummy_failure_jump:
4555 DEBUG_PRINT1 ("EXECUTING dummy_failure_jump.\n");
4556 /* It doesn't matter what we push for the string here. What
4557 the code at `fail' tests is the value for the pattern. */
4558 PUSH_FAILURE_POINT (0, 0, -2);
4559 goto unconditional_jump;
4562 /* At the end of an alternative, we need to push a dummy failure
4563 point in case we are followed by a `pop_failure_jump', because
4564 we don't want the failure point for the alternative to be
4565 popped. For example, matching `(a|ab)*' against `aab'
4566 requires that we match the `ab' alternative. */
4567 case push_dummy_failure:
4568 DEBUG_PRINT1 ("EXECUTING push_dummy_failure.\n");
4569 /* See comments just above at `dummy_failure_jump' about the
4571 PUSH_FAILURE_POINT (0, 0, -2);
4574 /* Have to succeed matching what follows at least n times.
4575 After that, handle like `on_failure_jump'. */
4577 EXTRACT_NUMBER (mcnt, p + 2);
4578 DEBUG_PRINT2 ("EXECUTING succeed_n %d.\n", mcnt);
4581 /* Originally, this is how many times we HAVE to succeed. */
4586 STORE_NUMBER_AND_INCR (p, mcnt);
4587 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p, mcnt);
4591 DEBUG_PRINT2 (" Setting two bytes from 0x%x to no_op.\n", p+2);
4592 p[2] = (unsigned char) no_op;
4593 p[3] = (unsigned char) no_op;
4599 EXTRACT_NUMBER (mcnt, p + 2);
4600 DEBUG_PRINT2 ("EXECUTING jump_n %d.\n", mcnt);
4602 /* Originally, this is how many times we CAN jump. */
4606 STORE_NUMBER (p + 2, mcnt);
4607 goto unconditional_jump;
4609 /* If don't have to jump any more, skip over the rest of command. */
4616 DEBUG_PRINT1 ("EXECUTING set_number_at.\n");
4618 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4620 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4621 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p1, mcnt);
4622 STORE_NUMBER (p1, mcnt);
4627 DEBUG_PRINT1 ("EXECUTING wordbound.\n");
4628 if (AT_WORD_BOUNDARY (d))
4633 DEBUG_PRINT1 ("EXECUTING notwordbound.\n");
4634 if (AT_WORD_BOUNDARY (d))
4639 DEBUG_PRINT1 ("EXECUTING wordbeg.\n");
4640 if (WORDCHAR_P (d) && (AT_STRINGS_BEG (d) || !WORDCHAR_P (d - 1)))
4645 DEBUG_PRINT1 ("EXECUTING wordend.\n");
4646 if (!AT_STRINGS_BEG (d) && WORDCHAR_P (d - 1)
4647 && (!WORDCHAR_P (d) || AT_STRINGS_END (d)))
4653 DEBUG_PRINT1 ("EXECUTING before_dot.\n");
4654 if (PTR_CHAR_POS ((unsigned char *) d) >= point)
4659 DEBUG_PRINT1 ("EXECUTING at_dot.\n");
4660 if (PTR_CHAR_POS ((unsigned char *) d) != point)
4665 DEBUG_PRINT1 ("EXECUTING after_dot.\n");
4666 if (PTR_CHAR_POS ((unsigned char *) d) <= point)
4669 #if 0 /* not emacs19 */
4671 DEBUG_PRINT1 ("EXECUTING at_dot.\n");
4672 if (PTR_CHAR_POS ((unsigned char *) d) + 1 != point)
4675 #endif /* not emacs19 */
4678 DEBUG_PRINT2 ("EXECUTING syntaxspec %d.\n", mcnt);
4683 DEBUG_PRINT1 ("EXECUTING Emacs wordchar.\n");
4687 /* Can't use *d++ here; SYNTAX may be an unsafe macro. */
4689 if (SYNTAX (d[-1]) != (enum syntaxcode) mcnt)
4691 SET_REGS_MATCHED ();
4695 DEBUG_PRINT2 ("EXECUTING notsyntaxspec %d.\n", mcnt);
4697 goto matchnotsyntax;
4700 DEBUG_PRINT1 ("EXECUTING Emacs notwordchar.\n");
4704 /* Can't use *d++ here; SYNTAX may be an unsafe macro. */
4706 if (SYNTAX (d[-1]) == (enum syntaxcode) mcnt)
4708 SET_REGS_MATCHED ();
4711 #else /* not emacs */
4713 DEBUG_PRINT1 ("EXECUTING non-Emacs wordchar.\n");
4715 if (!WORDCHAR_P (d))
4717 SET_REGS_MATCHED ();
4722 DEBUG_PRINT1 ("EXECUTING non-Emacs notwordchar.\n");
4726 SET_REGS_MATCHED ();
4729 #endif /* not emacs */
4734 continue; /* Successfully executed one pattern command; keep going. */
4737 /* We goto here if a matching operation fails. */
4739 if (!FAIL_STACK_EMPTY ())
4740 { /* A restart point is known. Restore to that state. */
4741 DEBUG_PRINT1 ("\nFAIL:\n");
4742 POP_FAILURE_POINT (d, p,
4743 lowest_active_reg, highest_active_reg,
4744 regstart, regend, reg_info);
4746 /* If this failure point is a dummy, try the next one. */
4750 /* If we failed to the end of the pattern, don't examine *p. */
4754 boolean is_a_jump_n = false;
4756 /* If failed to a backwards jump that's part of a repetition
4757 loop, need to pop this failure point and use the next one. */
4758 switch ((re_opcode_t) *p)
4762 case maybe_pop_jump:
4763 case pop_failure_jump:
4766 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4769 if ((is_a_jump_n && (re_opcode_t) *p1 == succeed_n)
4771 && (re_opcode_t) *p1 == on_failure_jump))
4779 if (d >= string1 && d <= end1)
4783 break; /* Matching at this starting point really fails. */
4787 goto restore_best_regs;
4791 return -1; /* Failure to match. */
4794 /* Subroutine definitions for re_match_2. */
4797 /* We are passed P pointing to a register number after a start_memory.
4799 Return true if the pattern up to the corresponding stop_memory can
4800 match the empty string, and false otherwise.
4802 If we find the matching stop_memory, sets P to point to one past its number.
4803 Otherwise, sets P to an undefined byte less than or equal to END.
4805 We don't handle duplicates properly (yet). */
4808 group_match_null_string_p (p, end, reg_info)
4809 unsigned char **p, *end;
4810 register_info_type *reg_info;
4813 /* Point to after the args to the start_memory. */
4814 unsigned char *p1 = *p + 2;
4818 /* Skip over opcodes that can match nothing, and return true or
4819 false, as appropriate, when we get to one that can't, or to the
4820 matching stop_memory. */
4822 switch ((re_opcode_t) *p1)
4824 /* Could be either a loop or a series of alternatives. */
4825 case on_failure_jump:
4827 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4829 /* If the next operation is not a jump backwards in the
4834 /* Go through the on_failure_jumps of the alternatives,
4835 seeing if any of the alternatives cannot match nothing.
4836 The last alternative starts with only a jump,
4837 whereas the rest start with on_failure_jump and end
4838 with a jump, e.g., here is the pattern for `a|b|c':
4840 /on_failure_jump/0/6/exactn/1/a/jump_past_alt/0/6
4841 /on_failure_jump/0/6/exactn/1/b/jump_past_alt/0/3
4844 So, we have to first go through the first (n-1)
4845 alternatives and then deal with the last one separately. */
4848 /* Deal with the first (n-1) alternatives, which start
4849 with an on_failure_jump (see above) that jumps to right
4850 past a jump_past_alt. */
4852 while ((re_opcode_t) p1[mcnt-3] == jump_past_alt)
4854 /* `mcnt' holds how many bytes long the alternative
4855 is, including the ending `jump_past_alt' and
4858 if (!alt_match_null_string_p (p1, p1 + mcnt - 3,
4862 /* Move to right after this alternative, including the
4866 /* Break if it's the beginning of an n-th alternative
4867 that doesn't begin with an on_failure_jump. */
4868 if ((re_opcode_t) *p1 != on_failure_jump)
4871 /* Still have to check that it's not an n-th
4872 alternative that starts with an on_failure_jump. */
4874 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4875 if ((re_opcode_t) p1[mcnt-3] != jump_past_alt)
4877 /* Get to the beginning of the n-th alternative. */
4883 /* Deal with the last alternative: go back and get number
4884 of the `jump_past_alt' just before it. `mcnt' contains
4885 the length of the alternative. */
4886 EXTRACT_NUMBER (mcnt, p1 - 2);
4888 if (!alt_match_null_string_p (p1, p1 + mcnt, reg_info))
4891 p1 += mcnt; /* Get past the n-th alternative. */
4897 assert (p1[1] == **p);
4903 if (!common_op_match_null_string_p (&p1, end, reg_info))
4906 } /* while p1 < end */
4909 } /* group_match_null_string_p */
4912 /* Similar to group_match_null_string_p, but doesn't deal with alternatives:
4913 It expects P to be the first byte of a single alternative and END one
4914 byte past the last. The alternative can contain groups. */
4917 alt_match_null_string_p (p, end, reg_info)
4918 unsigned char *p, *end;
4919 register_info_type *reg_info;
4922 unsigned char *p1 = p;
4926 /* Skip over opcodes that can match nothing, and break when we get
4927 to one that can't. */
4929 switch ((re_opcode_t) *p1)
4932 case on_failure_jump:
4934 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4939 if (!common_op_match_null_string_p (&p1, end, reg_info))
4942 } /* while p1 < end */
4945 } /* alt_match_null_string_p */
4948 /* Deals with the ops common to group_match_null_string_p and
4949 alt_match_null_string_p.
4951 Sets P to one after the op and its arguments, if any. */
4954 common_op_match_null_string_p (p, end, reg_info)
4955 unsigned char **p, *end;
4956 register_info_type *reg_info;
4961 unsigned char *p1 = *p;
4963 switch ((re_opcode_t) *p1++)
4983 assert (reg_no > 0 && reg_no <= MAX_REGNUM);
4984 ret = group_match_null_string_p (&p1, end, reg_info);
4986 /* Have to set this here in case we're checking a group which
4987 contains a group and a back reference to it. */
4989 if (REG_MATCH_NULL_STRING_P (reg_info[reg_no]) == MATCH_NULL_UNSET_VALUE)
4990 REG_MATCH_NULL_STRING_P (reg_info[reg_no]) = ret;
4996 /* If this is an optimized succeed_n for zero times, make the jump. */
4998 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
5006 /* Get to the number of times to succeed. */
5008 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
5013 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
5021 if (!REG_MATCH_NULL_STRING_P (reg_info[*p1]))
5029 /* All other opcodes mean we cannot match the empty string. */
5035 } /* common_op_match_null_string_p */
5038 /* Return zero if TRANSLATE[S1] and TRANSLATE[S2] are identical for LEN
5039 bytes; nonzero otherwise. */
5042 bcmp_translate (s1, s2, len, translate)
5043 unsigned char *s1, *s2;
5047 register unsigned char *p1 = s1, *p2 = s2;
5050 if (translate[*p1++] != translate[*p2++]) return 1;
5056 /* Entry points for GNU code. */
5058 /* re_compile_pattern is the GNU regular expression compiler: it
5059 compiles PATTERN (of length SIZE) and puts the result in BUFP.
5060 Returns 0 if the pattern was valid, otherwise an error string.
5062 Assumes the `allocated' (and perhaps `buffer') and `translate' fields
5063 are set in BUFP on entry.
5065 We call regex_compile to do the actual compilation. */
5068 re_compile_pattern (pattern, length, bufp)
5069 const char *pattern;
5071 struct re_pattern_buffer *bufp;
5075 /* GNU code is written to assume at least RE_NREGS registers will be set
5076 (and at least one extra will be -1). */
5077 bufp->regs_allocated = REGS_UNALLOCATED;
5079 /* And GNU code determines whether or not to get register information
5080 by passing null for the REGS argument to re_match, etc., not by
5084 /* Match anchors at newline. */
5085 bufp->newline_anchor = 1;
5087 ret = regex_compile (pattern, length, re_syntax_options, bufp);
5091 return gettext (re_error_msgid[(int) ret]);
5094 /* Entry points compatible with 4.2 BSD regex library. We don't define
5095 them unless specifically requested. */
5097 #ifdef _REGEX_RE_COMP
5099 /* BSD has one and only one pattern buffer. */
5100 static struct re_pattern_buffer re_comp_buf;
5110 if (!re_comp_buf.buffer)
5111 return gettext ("No previous regular expression");
5115 if (!re_comp_buf.buffer)
5117 re_comp_buf.buffer = (unsigned char *) malloc (200);
5118 if (re_comp_buf.buffer == NULL)
5119 return gettext (re_error_msgid[(int) REG_ESPACE]);
5120 re_comp_buf.allocated = 200;
5122 re_comp_buf.fastmap = (char *) malloc (1 << BYTEWIDTH);
5123 if (re_comp_buf.fastmap == NULL)
5124 return gettext (re_error_msgid[(int) REG_ESPACE]);
5127 /* Since `re_exec' always passes NULL for the `regs' argument, we
5128 don't need to initialize the pattern buffer fields which affect it. */
5130 /* Match anchors at newlines. */
5131 re_comp_buf.newline_anchor = 1;
5133 ret = regex_compile (s, strlen (s), re_syntax_options, &re_comp_buf);
5138 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
5139 return (char *) gettext (re_error_msgid[(int) ret]);
5147 const int len = strlen (s);
5149 0 <= re_search (&re_comp_buf, s, len, 0, len, (struct re_registers *) 0);
5151 #endif /* _REGEX_RE_COMP */
5153 /* POSIX.2 functions. Don't define these for Emacs. */
5157 /* regcomp takes a regular expression as a string and compiles it.
5159 PREG is a regex_t *. We do not expect any fields to be initialized,
5160 since POSIX says we shouldn't. Thus, we set
5162 `buffer' to the compiled pattern;
5163 `used' to the length of the compiled pattern;
5164 `syntax' to RE_SYNTAX_POSIX_EXTENDED if the
5165 REG_EXTENDED bit in CFLAGS is set; otherwise, to
5166 RE_SYNTAX_POSIX_BASIC;
5167 `newline_anchor' to REG_NEWLINE being set in CFLAGS;
5168 `fastmap' and `fastmap_accurate' to zero;
5169 `re_nsub' to the number of subexpressions in PATTERN.
5171 PATTERN is the address of the pattern string.
5173 CFLAGS is a series of bits which affect compilation.
5175 If REG_EXTENDED is set, we use POSIX extended syntax; otherwise, we
5176 use POSIX basic syntax.
5178 If REG_NEWLINE is set, then . and [^...] don't match newline.
5179 Also, regexec will try a match beginning after every newline.
5181 If REG_ICASE is set, then we considers upper- and lowercase
5182 versions of letters to be equivalent when matching.
5184 If REG_NOSUB is set, then when PREG is passed to regexec, that
5185 routine will report only success or failure, and nothing about the
5188 It returns 0 if it succeeds, nonzero if it doesn't. (See regex.h for
5189 the return codes and their meanings.) */
5192 regcomp (preg, pattern, cflags)
5194 const char *pattern;
5199 = (cflags & REG_EXTENDED) ?
5200 RE_SYNTAX_POSIX_EXTENDED : RE_SYNTAX_POSIX_BASIC;
5202 /* regex_compile will allocate the space for the compiled pattern. */
5204 preg->allocated = 0;
5207 /* Don't bother to use a fastmap when searching. This simplifies the
5208 REG_NEWLINE case: if we used a fastmap, we'd have to put all the
5209 characters after newlines into the fastmap. This way, we just try
5213 if (cflags & REG_ICASE)
5217 preg->translate = (char *) malloc (CHAR_SET_SIZE);
5218 if (preg->translate == NULL)
5219 return (int) REG_ESPACE;
5221 /* Map uppercase characters to corresponding lowercase ones. */
5222 for (i = 0; i < CHAR_SET_SIZE; i++)
5223 preg->translate[i] = ISUPPER (i) ? tolower (i) : i;
5226 preg->translate = NULL;
5228 /* If REG_NEWLINE is set, newlines are treated differently. */
5229 if (cflags & REG_NEWLINE)
5230 { /* REG_NEWLINE implies neither . nor [^...] match newline. */
5231 syntax &= ~RE_DOT_NEWLINE;
5232 syntax |= RE_HAT_LISTS_NOT_NEWLINE;
5233 /* It also changes the matching behavior. */
5234 preg->newline_anchor = 1;
5237 preg->newline_anchor = 0;
5239 preg->no_sub = !!(cflags & REG_NOSUB);
5241 /* POSIX says a null character in the pattern terminates it, so we
5242 can use strlen here in compiling the pattern. */
5243 ret = regex_compile (pattern, strlen (pattern), syntax, preg);
5245 /* POSIX doesn't distinguish between an unmatched open-group and an
5246 unmatched close-group: both are REG_EPAREN. */
5247 if (ret == REG_ERPAREN) ret = REG_EPAREN;
5253 /* regexec searches for a given pattern, specified by PREG, in the
5256 If NMATCH is zero or REG_NOSUB was set in the cflags argument to
5257 `regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at
5258 least NMATCH elements, and we set them to the offsets of the
5259 corresponding matched substrings.
5261 EFLAGS specifies `execution flags' which affect matching: if
5262 REG_NOTBOL is set, then ^ does not match at the beginning of the
5263 string; if REG_NOTEOL is set, then $ does not match at the end.
5265 We return 0 if we find a match and REG_NOMATCH if not. */
5268 regexec (preg, string, nmatch, pmatch, eflags)
5269 const regex_t *preg;
5272 regmatch_t pmatch[];
5276 struct re_registers regs;
5277 regex_t private_preg;
5278 int len = strlen (string);
5279 boolean want_reg_info = !preg->no_sub && nmatch > 0;
5281 private_preg = *preg;
5283 private_preg.not_bol = !!(eflags & REG_NOTBOL);
5284 private_preg.not_eol = !!(eflags & REG_NOTEOL);
5286 /* The user has told us exactly how many registers to return
5287 information about, via `nmatch'. We have to pass that on to the
5288 matching routines. */
5289 private_preg.regs_allocated = REGS_FIXED;
5293 regs.num_regs = nmatch;
5294 regs.start = TALLOC (nmatch, regoff_t);
5295 regs.end = TALLOC (nmatch, regoff_t);
5296 if (regs.start == NULL || regs.end == NULL)
5297 return (int) REG_NOMATCH;
5300 /* Perform the searching operation. */
5301 ret = re_search (&private_preg, string, len,
5302 /* start: */ 0, /* range: */ len,
5303 want_reg_info ? ®s : (struct re_registers *) 0);
5305 /* Copy the register information to the POSIX structure. */
5312 for (r = 0; r < nmatch; r++)
5314 pmatch[r].rm_so = regs.start[r];
5315 pmatch[r].rm_eo = regs.end[r];
5319 /* If we needed the temporary register info, free the space now. */
5324 /* We want zero return to mean success, unlike `re_search'. */
5325 return ret >= 0 ? (int) REG_NOERROR : (int) REG_NOMATCH;
5329 /* Returns a message corresponding to an error code, ERRCODE, returned
5330 from either regcomp or regexec. We don't use PREG here. */
5333 regerror (errcode, preg, errbuf, errbuf_size)
5335 const regex_t *preg;
5343 || errcode >= (sizeof (re_error_msgid) / sizeof (re_error_msgid[0])))
5344 /* Only error codes returned by the rest of the code should be passed
5345 to this routine. If we are given anything else, or if other regex
5346 code generates an invalid error code, then the program has a bug.
5347 Dump core so we can fix it. */
5350 msg = gettext (re_error_msgid[errcode]);
5352 msg_size = strlen (msg) + 1; /* Includes the null. */
5354 if (errbuf_size != 0)
5356 if (msg_size > errbuf_size)
5358 strncpy (errbuf, msg, errbuf_size - 1);
5359 errbuf[errbuf_size - 1] = 0;
5362 strcpy (errbuf, msg);
5369 /* Free dynamically allocated space used by PREG. */
5375 if (preg->buffer != NULL)
5376 free (preg->buffer);
5377 preg->buffer = NULL;
5379 preg->allocated = 0;
5382 if (preg->fastmap != NULL)
5383 free (preg->fastmap);
5384 preg->fastmap = NULL;
5385 preg->fastmap_accurate = 0;
5387 if (preg->translate != NULL)
5388 free (preg->translate);
5389 preg->translate = NULL;
5392 #endif /* not emacs */
5396 make-backup-files: t
5398 trim-versions-without-asking: nil