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 /* When used in Emacs's lib-src, we need to get bzero and bcopy somehow.
66 If nothing else has been done, use the method below. */
67 #ifdef INHIBIT_STRING_HEADER
68 #if !(defined (HAVE_BZERO) && defined (HAVE_BCOPY))
69 #if !defined (bzero) && !defined (bcopy)
70 #undef INHIBIT_STRING_HEADER
75 /* This is the normal way of making sure we have a bcopy and a bzero.
76 This is used in most programs--a few other programs avoid this
77 by defining INHIBIT_STRING_HEADER. */
78 #ifndef INHIBIT_STRING_HEADER
79 #if defined (HAVE_STRING_H) || defined (STDC_HEADERS) || defined (_LIBC)
82 #define bcmp(s1, s2, n) memcmp ((s1), (s2), (n))
85 #define bcopy(s, d, n) memcpy ((d), (s), (n))
88 #define bzero(s, n) memset ((s), 0, (n))
95 /* Define the syntax stuff for \<, \>, etc. */
97 /* This must be nonzero for the wordchar and notwordchar pattern
98 commands in re_match_2. */
103 #ifdef SWITCH_ENUM_BUG
104 #define SWITCH_ENUM_CAST(x) ((int)(x))
106 #define SWITCH_ENUM_CAST(x) (x)
111 extern char *re_syntax_table;
113 #else /* not SYNTAX_TABLE */
115 /* How many characters in the character set. */
116 #define CHAR_SET_SIZE 256
118 static char re_syntax_table[CHAR_SET_SIZE];
129 bzero (re_syntax_table, sizeof re_syntax_table);
131 for (c = 'a'; c <= 'z'; c++)
132 re_syntax_table[c] = Sword;
134 for (c = 'A'; c <= 'Z'; c++)
135 re_syntax_table[c] = Sword;
137 for (c = '0'; c <= '9'; c++)
138 re_syntax_table[c] = Sword;
140 re_syntax_table['_'] = Sword;
145 #endif /* not SYNTAX_TABLE */
147 #define SYNTAX(c) re_syntax_table[c]
149 #endif /* not emacs */
151 /* Get the interface, including the syntax bits. */
154 /* isalpha etc. are used for the character classes. */
157 /* Jim Meyering writes:
159 "... Some ctype macros are valid only for character codes that
160 isascii says are ASCII (SGI's IRIX-4.0.5 is one such system --when
161 using /bin/cc or gcc but without giving an ansi option). So, all
162 ctype uses should be through macros like ISPRINT... If
163 STDC_HEADERS is defined, then autoconf has verified that the ctype
164 macros don't need to be guarded with references to isascii. ...
165 Defining isascii to 1 should let any compiler worth its salt
166 eliminate the && through constant folding." */
168 #if defined (STDC_HEADERS) || (!defined (isascii) && !defined (HAVE_ISASCII))
171 #define ISASCII(c) isascii(c)
175 #define ISBLANK(c) (ISASCII (c) && isblank (c))
177 #define ISBLANK(c) ((c) == ' ' || (c) == '\t')
180 #define ISGRAPH(c) (ISASCII (c) && isgraph (c))
182 #define ISGRAPH(c) (ISASCII (c) && isprint (c) && !isspace (c))
185 #define ISPRINT(c) (ISASCII (c) && isprint (c))
186 #define ISDIGIT(c) (ISASCII (c) && isdigit (c))
187 #define ISALNUM(c) (ISASCII (c) && isalnum (c))
188 #define ISALPHA(c) (ISASCII (c) && isalpha (c))
189 #define ISCNTRL(c) (ISASCII (c) && iscntrl (c))
190 #define ISLOWER(c) (ISASCII (c) && islower (c))
191 #define ISPUNCT(c) (ISASCII (c) && ispunct (c))
192 #define ISSPACE(c) (ISASCII (c) && isspace (c))
193 #define ISUPPER(c) (ISASCII (c) && isupper (c))
194 #define ISXDIGIT(c) (ISASCII (c) && isxdigit (c))
197 #define NULL (void *)0
200 /* We remove any previous definition of `SIGN_EXTEND_CHAR',
201 since ours (we hope) works properly with all combinations of
202 machines, compilers, `char' and `unsigned char' argument types.
203 (Per Bothner suggested the basic approach.) */
204 #undef SIGN_EXTEND_CHAR
206 #define SIGN_EXTEND_CHAR(c) ((signed char) (c))
207 #else /* not __STDC__ */
208 /* As in Harbison and Steele. */
209 #define SIGN_EXTEND_CHAR(c) ((((unsigned char) (c)) ^ 128) - 128)
212 /* Should we use malloc or alloca? If REGEX_MALLOC is not defined, we
213 use `alloca' instead of `malloc'. This is because using malloc in
214 re_search* or re_match* could cause memory leaks when C-g is used in
215 Emacs; also, malloc is slower and causes storage fragmentation. On
216 the other hand, malloc is more portable, and easier to debug.
218 Because we sometimes use alloca, some routines have to be macros,
219 not functions -- `alloca'-allocated space disappears at the end of the
220 function it is called in. */
224 #define REGEX_ALLOCATE malloc
225 #define REGEX_REALLOCATE(source, osize, nsize) realloc (source, nsize)
226 #define REGEX_FREE free
228 #else /* not REGEX_MALLOC */
230 /* Emacs already defines alloca, sometimes. */
233 /* Make alloca work the best possible way. */
235 #define alloca __builtin_alloca
236 #else /* not __GNUC__ */
239 #else /* not __GNUC__ or HAVE_ALLOCA_H */
240 #ifndef _AIX /* Already did AIX, up at the top. */
241 #if defined (__STDC__) && __STDC__
246 #endif /* not _AIX */
247 #endif /* not HAVE_ALLOCA_H */
248 #endif /* not __GNUC__ */
250 #endif /* not alloca */
252 #define REGEX_ALLOCATE alloca
254 /* Assumes a `char *destination' variable. */
255 #define REGEX_REALLOCATE(source, osize, nsize) \
256 (destination = (char *) alloca (nsize), \
257 bcopy (source, destination, osize), \
260 /* No need to do anything to free, after alloca. */
261 #define REGEX_FREE(arg) ((void)0) /* Do nothing! But inhibit gcc warning. */
263 #endif /* not REGEX_MALLOC */
265 /* Define how to allocate the failure stack. */
267 #if defined (REL_ALLOC) && defined (REGEX_MALLOC)
269 #define REGEX_ALLOCATE_STACK(size) \
270 r_alloc (&failure_stack_ptr, (size))
271 #define REGEX_REALLOCATE_STACK(source, osize, nsize) \
272 r_re_alloc (&failure_stack_ptr, (nsize))
273 #define REGEX_FREE_STACK(ptr) \
274 r_alloc_free (&failure_stack_ptr)
276 #else /* not using relocating allocator */
280 #define REGEX_ALLOCATE_STACK malloc
281 #define REGEX_REALLOCATE_STACK(source, osize, nsize) realloc (source, nsize)
282 #define REGEX_FREE_STACK free
284 #else /* not REGEX_MALLOC */
286 #define REGEX_ALLOCATE_STACK alloca
288 #define REGEX_REALLOCATE_STACK(source, osize, nsize) \
289 REGEX_REALLOCATE (source, osize, nsize)
290 /* No need to explicitly free anything. */
291 #define REGEX_FREE_STACK(arg)
293 #endif /* not REGEX_MALLOC */
294 #endif /* not using relocating allocator */
297 /* True if `size1' is non-NULL and PTR is pointing anywhere inside
298 `string1' or just past its end. This works if PTR is NULL, which is
300 #define FIRST_STRING_P(ptr) \
301 (size1 && string1 <= (ptr) && (ptr) <= string1 + size1)
303 /* (Re)Allocate N items of type T using malloc, or fail. */
304 #define TALLOC(n, t) ((t *) malloc ((n) * sizeof (t)))
305 #define RETALLOC(addr, n, t) ((addr) = (t *) realloc (addr, (n) * sizeof (t)))
306 #define RETALLOC_IF(addr, n, t) \
307 if (addr) RETALLOC((addr), (n), t); else (addr) = TALLOC ((n), t)
308 #define REGEX_TALLOC(n, t) ((t *) REGEX_ALLOCATE ((n) * sizeof (t)))
310 #define BYTEWIDTH 8 /* In bits. */
312 #define STREQ(s1, s2) ((strcmp (s1, s2) == 0))
316 #define MAX(a, b) ((a) > (b) ? (a) : (b))
317 #define MIN(a, b) ((a) < (b) ? (a) : (b))
319 typedef char boolean;
323 static int re_match_2_internal ();
325 /* These are the command codes that appear in compiled regular
326 expressions. Some opcodes are followed by argument bytes. A
327 command code can specify any interpretation whatsoever for its
328 arguments. Zero bytes may appear in the compiled regular expression. */
334 /* Succeed right away--no more backtracking. */
337 /* Followed by one byte giving n, then by n literal bytes. */
340 /* Matches any (more or less) character. */
343 /* Matches any one char belonging to specified set. First
344 following byte is number of bitmap bytes. Then come bytes
345 for a bitmap saying which chars are in. Bits in each byte
346 are ordered low-bit-first. A character is in the set if its
347 bit is 1. A character too large to have a bit in the map is
348 automatically not in the set. */
351 /* Same parameters as charset, but match any character that is
352 not one of those specified. */
355 /* Start remembering the text that is matched, for storing in a
356 register. Followed by one byte with the register number, in
357 the range 0 to one less than the pattern buffer's re_nsub
358 field. Then followed by one byte with the number of groups
359 inner to this one. (This last has to be part of the
360 start_memory only because we need it in the on_failure_jump
364 /* Stop remembering the text that is matched and store it in a
365 memory register. Followed by one byte with the register
366 number, in the range 0 to one less than `re_nsub' in the
367 pattern buffer, and one byte with the number of inner groups,
368 just like `start_memory'. (We need the number of inner
369 groups here because we don't have any easy way of finding the
370 corresponding start_memory when we're at a stop_memory.) */
373 /* Match a duplicate of something remembered. Followed by one
374 byte containing the register number. */
377 /* Fail unless at beginning of line. */
380 /* Fail unless at end of line. */
383 /* Succeeds if at beginning of buffer (if emacs) or at beginning
384 of string to be matched (if not). */
387 /* Analogously, for end of buffer/string. */
390 /* Followed by two byte relative address to which to jump. */
393 /* Same as jump, but marks the end of an alternative. */
396 /* Followed by two-byte relative address of place to resume at
397 in case of failure. */
400 /* Like on_failure_jump, but pushes a placeholder instead of the
401 current string position when executed. */
402 on_failure_keep_string_jump,
404 /* Throw away latest failure point and then jump to following
405 two-byte relative address. */
408 /* Change to pop_failure_jump if know won't have to backtrack to
409 match; otherwise change to jump. This is used to jump
410 back to the beginning of a repeat. If what follows this jump
411 clearly won't match what the repeat does, such that we can be
412 sure that there is no use backtracking out of repetitions
413 already matched, then we change it to a pop_failure_jump.
414 Followed by two-byte address. */
417 /* Jump to following two-byte address, and push a dummy failure
418 point. This failure point will be thrown away if an attempt
419 is made to use it for a failure. A `+' construct makes this
420 before the first repeat. Also used as an intermediary kind
421 of jump when compiling an alternative. */
424 /* Push a dummy failure point and continue. Used at the end of
428 /* Followed by two-byte relative address and two-byte number n.
429 After matching N times, jump to the address upon failure. */
432 /* Followed by two-byte relative address, and two-byte number n.
433 Jump to the address N times, then fail. */
436 /* Set the following two-byte relative address to the
437 subsequent two-byte number. The address *includes* the two
441 wordchar, /* Matches any word-constituent character. */
442 notwordchar, /* Matches any char that is not a word-constituent. */
444 wordbeg, /* Succeeds if at word beginning. */
445 wordend, /* Succeeds if at word end. */
447 wordbound, /* Succeeds if at a word boundary. */
448 notwordbound /* Succeeds if not at a word boundary. */
451 ,before_dot, /* Succeeds if before point. */
452 at_dot, /* Succeeds if at point. */
453 after_dot, /* Succeeds if after point. */
455 /* Matches any character whose syntax is specified. Followed by
456 a byte which contains a syntax code, e.g., Sword. */
459 /* Matches any character whose syntax is not that specified. */
464 /* Common operations on the compiled pattern. */
466 /* Store NUMBER in two contiguous bytes starting at DESTINATION. */
468 #define STORE_NUMBER(destination, number) \
470 (destination)[0] = (number) & 0377; \
471 (destination)[1] = (number) >> 8; \
474 /* Same as STORE_NUMBER, except increment DESTINATION to
475 the byte after where the number is stored. Therefore, DESTINATION
476 must be an lvalue. */
478 #define STORE_NUMBER_AND_INCR(destination, number) \
480 STORE_NUMBER (destination, number); \
481 (destination) += 2; \
484 /* Put into DESTINATION a number stored in two contiguous bytes starting
487 #define EXTRACT_NUMBER(destination, source) \
489 (destination) = *(source) & 0377; \
490 (destination) += SIGN_EXTEND_CHAR (*((source) + 1)) << 8; \
495 extract_number (dest, source)
497 unsigned char *source;
499 int temp = SIGN_EXTEND_CHAR (*(source + 1));
500 *dest = *source & 0377;
504 #ifndef EXTRACT_MACROS /* To debug the macros. */
505 #undef EXTRACT_NUMBER
506 #define EXTRACT_NUMBER(dest, src) extract_number (&dest, src)
507 #endif /* not EXTRACT_MACROS */
511 /* Same as EXTRACT_NUMBER, except increment SOURCE to after the number.
512 SOURCE must be an lvalue. */
514 #define EXTRACT_NUMBER_AND_INCR(destination, source) \
516 EXTRACT_NUMBER (destination, source); \
522 extract_number_and_incr (destination, source)
524 unsigned char **source;
526 extract_number (destination, *source);
530 #ifndef EXTRACT_MACROS
531 #undef EXTRACT_NUMBER_AND_INCR
532 #define EXTRACT_NUMBER_AND_INCR(dest, src) \
533 extract_number_and_incr (&dest, &src)
534 #endif /* not EXTRACT_MACROS */
538 /* If DEBUG is defined, Regex prints many voluminous messages about what
539 it is doing (if the variable `debug' is nonzero). If linked with the
540 main program in `iregex.c', you can enter patterns and strings
541 interactively. And if linked with the main program in `main.c' and
542 the other test files, you can run the already-written tests. */
546 /* We use standard I/O for debugging. */
549 /* It is useful to test things that ``must'' be true when debugging. */
552 static int debug = 0;
554 #define DEBUG_STATEMENT(e) e
555 #define DEBUG_PRINT1(x) if (debug) printf (x)
556 #define DEBUG_PRINT2(x1, x2) if (debug) printf (x1, x2)
557 #define DEBUG_PRINT3(x1, x2, x3) if (debug) printf (x1, x2, x3)
558 #define DEBUG_PRINT4(x1, x2, x3, x4) if (debug) printf (x1, x2, x3, x4)
559 #define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) \
560 if (debug) print_partial_compiled_pattern (s, e)
561 #define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) \
562 if (debug) print_double_string (w, s1, sz1, s2, sz2)
565 /* Print the fastmap in human-readable form. */
568 print_fastmap (fastmap)
571 unsigned was_a_range = 0;
574 while (i < (1 << BYTEWIDTH))
580 while (i < (1 << BYTEWIDTH) && fastmap[i])
596 /* Print a compiled pattern string in human-readable form, starting at
597 the START pointer into it and ending just before the pointer END. */
600 print_partial_compiled_pattern (start, end)
601 unsigned char *start;
605 unsigned char *p = start;
606 unsigned char *pend = end;
614 /* Loop over pattern commands. */
617 printf ("%d:\t", p - start);
619 switch ((re_opcode_t) *p++)
627 printf ("/exactn/%d", mcnt);
638 printf ("/start_memory/%d/%d", mcnt, *p++);
643 printf ("/stop_memory/%d/%d", mcnt, *p++);
647 printf ("/duplicate/%d", *p++);
657 register int c, last = -100;
658 register int in_range = 0;
660 printf ("/charset [%s",
661 (re_opcode_t) *(p - 1) == charset_not ? "^" : "");
663 assert (p + *p < pend);
665 for (c = 0; c < 256; c++)
667 && (p[1 + (c/8)] & (1 << (c % 8))))
669 /* Are we starting a range? */
670 if (last + 1 == c && ! in_range)
675 /* Have we broken a range? */
676 else if (last + 1 != c && in_range)
705 case on_failure_jump:
706 extract_number_and_incr (&mcnt, &p);
707 printf ("/on_failure_jump to %d", p + mcnt - start);
710 case on_failure_keep_string_jump:
711 extract_number_and_incr (&mcnt, &p);
712 printf ("/on_failure_keep_string_jump to %d", p + mcnt - start);
715 case dummy_failure_jump:
716 extract_number_and_incr (&mcnt, &p);
717 printf ("/dummy_failure_jump to %d", p + mcnt - start);
720 case push_dummy_failure:
721 printf ("/push_dummy_failure");
725 extract_number_and_incr (&mcnt, &p);
726 printf ("/maybe_pop_jump to %d", p + mcnt - start);
729 case pop_failure_jump:
730 extract_number_and_incr (&mcnt, &p);
731 printf ("/pop_failure_jump to %d", p + mcnt - start);
735 extract_number_and_incr (&mcnt, &p);
736 printf ("/jump_past_alt to %d", p + mcnt - start);
740 extract_number_and_incr (&mcnt, &p);
741 printf ("/jump to %d", p + mcnt - start);
745 extract_number_and_incr (&mcnt, &p);
746 extract_number_and_incr (&mcnt2, &p);
747 printf ("/succeed_n to %d, %d times", p + mcnt - start, mcnt2);
751 extract_number_and_incr (&mcnt, &p);
752 extract_number_and_incr (&mcnt2, &p);
753 printf ("/jump_n to %d, %d times", p + mcnt - start, mcnt2);
757 extract_number_and_incr (&mcnt, &p);
758 extract_number_and_incr (&mcnt2, &p);
759 printf ("/set_number_at location %d to %d", p + mcnt - start, mcnt2);
763 printf ("/wordbound");
767 printf ("/notwordbound");
779 printf ("/before_dot");
787 printf ("/after_dot");
791 printf ("/syntaxspec");
793 printf ("/%d", mcnt);
797 printf ("/notsyntaxspec");
799 printf ("/%d", mcnt);
804 printf ("/wordchar");
808 printf ("/notwordchar");
820 printf ("?%d", *(p-1));
826 printf ("%d:\tend of pattern.\n", p - start);
831 print_compiled_pattern (bufp)
832 struct re_pattern_buffer *bufp;
834 unsigned char *buffer = bufp->buffer;
836 print_partial_compiled_pattern (buffer, buffer + bufp->used);
837 printf ("%d bytes used/%d bytes allocated.\n", bufp->used, bufp->allocated);
839 if (bufp->fastmap_accurate && bufp->fastmap)
841 printf ("fastmap: ");
842 print_fastmap (bufp->fastmap);
845 printf ("re_nsub: %d\t", bufp->re_nsub);
846 printf ("regs_alloc: %d\t", bufp->regs_allocated);
847 printf ("can_be_null: %d\t", bufp->can_be_null);
848 printf ("newline_anchor: %d\n", bufp->newline_anchor);
849 printf ("no_sub: %d\t", bufp->no_sub);
850 printf ("not_bol: %d\t", bufp->not_bol);
851 printf ("not_eol: %d\t", bufp->not_eol);
852 printf ("syntax: %d\n", bufp->syntax);
853 /* Perhaps we should print the translate table? */
858 print_double_string (where, string1, size1, string2, size2)
871 if (FIRST_STRING_P (where))
873 for (this_char = where - string1; this_char < size1; this_char++)
874 putchar (string1[this_char]);
879 for (this_char = where - string2; this_char < size2; this_char++)
880 putchar (string2[this_char]);
884 #else /* not DEBUG */
889 #define DEBUG_STATEMENT(e)
890 #define DEBUG_PRINT1(x)
891 #define DEBUG_PRINT2(x1, x2)
892 #define DEBUG_PRINT3(x1, x2, x3)
893 #define DEBUG_PRINT4(x1, x2, x3, x4)
894 #define DEBUG_PRINT_COMPILED_PATTERN(p, s, e)
895 #define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2)
897 #endif /* not DEBUG */
899 /* Set by `re_set_syntax' to the current regexp syntax to recognize. Can
900 also be assigned to arbitrarily: each pattern buffer stores its own
901 syntax, so it can be changed between regex compilations. */
902 /* This has no initializer because initialized variables in Emacs
903 become read-only after dumping. */
904 reg_syntax_t re_syntax_options;
907 /* Specify the precise syntax of regexps for compilation. This provides
908 for compatibility for various utilities which historically have
909 different, incompatible syntaxes.
911 The argument SYNTAX is a bit mask comprised of the various bits
912 defined in regex.h. We return the old syntax. */
915 re_set_syntax (syntax)
918 reg_syntax_t ret = re_syntax_options;
920 re_syntax_options = syntax;
924 /* This table gives an error message for each of the error codes listed
925 in regex.h. Obviously the order here has to be same as there.
926 POSIX doesn't require that we do anything for REG_NOERROR,
927 but why not be nice? */
929 static const char *re_error_msgid[] =
930 { "Success", /* REG_NOERROR */
931 "No match", /* REG_NOMATCH */
932 "Invalid regular expression", /* REG_BADPAT */
933 "Invalid collation character", /* REG_ECOLLATE */
934 "Invalid character class name", /* REG_ECTYPE */
935 "Trailing backslash", /* REG_EESCAPE */
936 "Invalid back reference", /* REG_ESUBREG */
937 "Unmatched [ or [^", /* REG_EBRACK */
938 "Unmatched ( or \\(", /* REG_EPAREN */
939 "Unmatched \\{", /* REG_EBRACE */
940 "Invalid content of \\{\\}", /* REG_BADBR */
941 "Invalid range end", /* REG_ERANGE */
942 "Memory exhausted", /* REG_ESPACE */
943 "Invalid preceding regular expression", /* REG_BADRPT */
944 "Premature end of regular expression", /* REG_EEND */
945 "Regular expression too big", /* REG_ESIZE */
946 "Unmatched ) or \\)", /* REG_ERPAREN */
949 /* Avoiding alloca during matching, to placate r_alloc. */
951 /* Define MATCH_MAY_ALLOCATE unless we need to make sure that the
952 searching and matching functions should not call alloca. On some
953 systems, alloca is implemented in terms of malloc, and if we're
954 using the relocating allocator routines, then malloc could cause a
955 relocation, which might (if the strings being searched are in the
956 ralloc heap) shift the data out from underneath the regexp
959 Here's another reason to avoid allocation: Emacs
960 processes input from X in a signal handler; processing X input may
961 call malloc; if input arrives while a matching routine is calling
962 malloc, then we're scrod. But Emacs can't just block input while
963 calling matching routines; then we don't notice interrupts when
964 they come in. So, Emacs blocks input around all regexp calls
965 except the matching calls, which it leaves unprotected, in the
966 faith that they will not malloc. */
968 /* Normally, this is fine. */
969 #define MATCH_MAY_ALLOCATE
971 /* When using GNU C, we are not REALLY using the C alloca, no matter
972 what config.h may say. So don't take precautions for it. */
977 /* The match routines may not allocate if (1) they would do it with malloc
978 and (2) it's not safe for them to use malloc.
979 Note that if REL_ALLOC is defined, matching would not use malloc for the
980 failure stack, but we would still use it for the register vectors;
981 so REL_ALLOC should not affect this. */
982 #if (defined (C_ALLOCA) || defined (REGEX_MALLOC)) && defined (emacs)
983 #undef MATCH_MAY_ALLOCATE
987 /* Failure stack declarations and macros; both re_compile_fastmap and
988 re_match_2 use a failure stack. These have to be macros because of
989 REGEX_ALLOCATE_STACK. */
992 /* Number of failure points for which to initially allocate space
993 when matching. If this number is exceeded, we allocate more
994 space, so it is not a hard limit. */
995 #ifndef INIT_FAILURE_ALLOC
996 #define INIT_FAILURE_ALLOC 5
999 /* Roughly the maximum number of failure points on the stack. Would be
1000 exactly that if always used MAX_FAILURE_SPACE each time we failed.
1001 This is a variable only so users of regex can assign to it; we never
1002 change it ourselves. */
1003 #if defined (MATCH_MAY_ALLOCATE)
1004 int re_max_failures = 200000;
1006 int re_max_failures = 2000;
1009 union fail_stack_elt
1011 unsigned char *pointer;
1015 typedef union fail_stack_elt fail_stack_elt_t;
1019 fail_stack_elt_t *stack;
1021 unsigned avail; /* Offset of next open position. */
1024 #define FAIL_STACK_EMPTY() (fail_stack.avail == 0)
1025 #define FAIL_STACK_PTR_EMPTY() (fail_stack_ptr->avail == 0)
1026 #define FAIL_STACK_FULL() (fail_stack.avail == fail_stack.size)
1029 /* Define macros to initialize and free the failure stack.
1030 Do `return -2' if the alloc fails. */
1032 #ifdef MATCH_MAY_ALLOCATE
1033 #define INIT_FAIL_STACK() \
1035 fail_stack.stack = (fail_stack_elt_t *) \
1036 REGEX_ALLOCATE_STACK (INIT_FAILURE_ALLOC * sizeof (fail_stack_elt_t)); \
1038 if (fail_stack.stack == NULL) \
1041 fail_stack.size = INIT_FAILURE_ALLOC; \
1042 fail_stack.avail = 0; \
1045 #define RESET_FAIL_STACK() REGEX_FREE_STACK (fail_stack.stack)
1047 #define INIT_FAIL_STACK() \
1049 fail_stack.avail = 0; \
1052 #define RESET_FAIL_STACK()
1056 /* Double the size of FAIL_STACK, up to approximately `re_max_failures' items.
1058 Return 1 if succeeds, and 0 if either ran out of memory
1059 allocating space for it or it was already too large.
1061 REGEX_REALLOCATE_STACK requires `destination' be declared. */
1063 #define DOUBLE_FAIL_STACK(fail_stack) \
1064 ((fail_stack).size > re_max_failures * MAX_FAILURE_ITEMS \
1066 : ((fail_stack).stack = (fail_stack_elt_t *) \
1067 REGEX_REALLOCATE_STACK ((fail_stack).stack, \
1068 (fail_stack).size * sizeof (fail_stack_elt_t), \
1069 ((fail_stack).size << 1) * sizeof (fail_stack_elt_t)), \
1071 (fail_stack).stack == NULL \
1073 : ((fail_stack).size <<= 1, \
1077 /* Push pointer POINTER on FAIL_STACK.
1078 Return 1 if was able to do so and 0 if ran out of memory allocating
1080 #define PUSH_PATTERN_OP(POINTER, FAIL_STACK) \
1081 ((FAIL_STACK_FULL () \
1082 && !DOUBLE_FAIL_STACK (FAIL_STACK)) \
1084 : ((FAIL_STACK).stack[(FAIL_STACK).avail++].pointer = POINTER, \
1087 /* Push a pointer value onto the failure stack.
1088 Assumes the variable `fail_stack'. Probably should only
1089 be called from within `PUSH_FAILURE_POINT'. */
1090 #define PUSH_FAILURE_POINTER(item) \
1091 fail_stack.stack[fail_stack.avail++].pointer = (unsigned char *) (item)
1093 /* This pushes an integer-valued item onto the failure stack.
1094 Assumes the variable `fail_stack'. Probably should only
1095 be called from within `PUSH_FAILURE_POINT'. */
1096 #define PUSH_FAILURE_INT(item) \
1097 fail_stack.stack[fail_stack.avail++].integer = (item)
1099 /* Push a fail_stack_elt_t value onto the failure stack.
1100 Assumes the variable `fail_stack'. Probably should only
1101 be called from within `PUSH_FAILURE_POINT'. */
1102 #define PUSH_FAILURE_ELT(item) \
1103 fail_stack.stack[fail_stack.avail++] = (item)
1105 /* These three POP... operations complement the three PUSH... operations.
1106 All assume that `fail_stack' is nonempty. */
1107 #define POP_FAILURE_POINTER() fail_stack.stack[--fail_stack.avail].pointer
1108 #define POP_FAILURE_INT() fail_stack.stack[--fail_stack.avail].integer
1109 #define POP_FAILURE_ELT() fail_stack.stack[--fail_stack.avail]
1111 /* Used to omit pushing failure point id's when we're not debugging. */
1113 #define DEBUG_PUSH PUSH_FAILURE_INT
1114 #define DEBUG_POP(item_addr) *(item_addr) = POP_FAILURE_INT ()
1116 #define DEBUG_PUSH(item)
1117 #define DEBUG_POP(item_addr)
1121 /* Push the information about the state we will need
1122 if we ever fail back to it.
1124 Requires variables fail_stack, regstart, regend, reg_info, and
1125 num_regs be declared. DOUBLE_FAIL_STACK requires `destination' be
1128 Does `return FAILURE_CODE' if runs out of memory. */
1130 #define PUSH_FAILURE_POINT(pattern_place, string_place, failure_code) \
1132 char *destination; \
1133 /* Must be int, so when we don't save any registers, the arithmetic \
1134 of 0 + -1 isn't done as unsigned. */ \
1137 DEBUG_STATEMENT (failure_id++); \
1138 DEBUG_STATEMENT (nfailure_points_pushed++); \
1139 DEBUG_PRINT2 ("\nPUSH_FAILURE_POINT #%u:\n", failure_id); \
1140 DEBUG_PRINT2 (" Before push, next avail: %d\n", (fail_stack).avail);\
1141 DEBUG_PRINT2 (" size: %d\n", (fail_stack).size);\
1143 DEBUG_PRINT2 (" slots needed: %d\n", NUM_FAILURE_ITEMS); \
1144 DEBUG_PRINT2 (" available: %d\n", REMAINING_AVAIL_SLOTS); \
1146 /* Ensure we have enough space allocated for what we will push. */ \
1147 while (REMAINING_AVAIL_SLOTS < NUM_FAILURE_ITEMS) \
1149 if (!DOUBLE_FAIL_STACK (fail_stack)) \
1150 return failure_code; \
1152 DEBUG_PRINT2 ("\n Doubled stack; size now: %d\n", \
1153 (fail_stack).size); \
1154 DEBUG_PRINT2 (" slots available: %d\n", REMAINING_AVAIL_SLOTS);\
1157 /* Push the info, starting with the registers. */ \
1158 DEBUG_PRINT1 ("\n"); \
1160 if (!RE_NO_POSIX_BACKTRACKING & bufp->syntax) \
1161 for (this_reg = lowest_active_reg; this_reg <= highest_active_reg; \
1164 DEBUG_PRINT2 (" Pushing reg: %d\n", this_reg); \
1165 DEBUG_STATEMENT (num_regs_pushed++); \
1167 DEBUG_PRINT2 (" start: 0x%x\n", regstart[this_reg]); \
1168 PUSH_FAILURE_POINTER (regstart[this_reg]); \
1170 DEBUG_PRINT2 (" end: 0x%x\n", regend[this_reg]); \
1171 PUSH_FAILURE_POINTER (regend[this_reg]); \
1173 DEBUG_PRINT2 (" info: 0x%x\n ", reg_info[this_reg]); \
1174 DEBUG_PRINT2 (" match_null=%d", \
1175 REG_MATCH_NULL_STRING_P (reg_info[this_reg])); \
1176 DEBUG_PRINT2 (" active=%d", IS_ACTIVE (reg_info[this_reg])); \
1177 DEBUG_PRINT2 (" matched_something=%d", \
1178 MATCHED_SOMETHING (reg_info[this_reg])); \
1179 DEBUG_PRINT2 (" ever_matched=%d", \
1180 EVER_MATCHED_SOMETHING (reg_info[this_reg])); \
1181 DEBUG_PRINT1 ("\n"); \
1182 PUSH_FAILURE_ELT (reg_info[this_reg].word); \
1185 DEBUG_PRINT2 (" Pushing low active reg: %d\n", lowest_active_reg);\
1186 PUSH_FAILURE_INT (lowest_active_reg); \
1188 DEBUG_PRINT2 (" Pushing high active reg: %d\n", highest_active_reg);\
1189 PUSH_FAILURE_INT (highest_active_reg); \
1191 DEBUG_PRINT2 (" Pushing pattern 0x%x: ", pattern_place); \
1192 DEBUG_PRINT_COMPILED_PATTERN (bufp, pattern_place, pend); \
1193 PUSH_FAILURE_POINTER (pattern_place); \
1195 DEBUG_PRINT2 (" Pushing string 0x%x: `", string_place); \
1196 DEBUG_PRINT_DOUBLE_STRING (string_place, string1, size1, string2, \
1198 DEBUG_PRINT1 ("'\n"); \
1199 PUSH_FAILURE_POINTER (string_place); \
1201 DEBUG_PRINT2 (" Pushing failure id: %u\n", failure_id); \
1202 DEBUG_PUSH (failure_id); \
1205 /* This is the number of items that are pushed and popped on the stack
1206 for each register. */
1207 #define NUM_REG_ITEMS 3
1209 /* Individual items aside from the registers. */
1211 #define NUM_NONREG_ITEMS 5 /* Includes failure point id. */
1213 #define NUM_NONREG_ITEMS 4
1216 /* We push at most this many items on the stack. */
1217 #define MAX_FAILURE_ITEMS ((num_regs - 1) * NUM_REG_ITEMS + NUM_NONREG_ITEMS)
1219 /* We actually push this many items. */
1220 #define NUM_FAILURE_ITEMS \
1221 (((RE_NO_POSIX_BACKTRACKING & bufp->syntax \
1222 ? 0 : highest_active_reg - lowest_active_reg + 1) \
1226 /* How many items can still be added to the stack without overflowing it. */
1227 #define REMAINING_AVAIL_SLOTS ((fail_stack).size - (fail_stack).avail)
1230 /* Pops what PUSH_FAIL_STACK pushes.
1232 We restore into the parameters, all of which should be lvalues:
1233 STR -- the saved data position.
1234 PAT -- the saved pattern position.
1235 LOW_REG, HIGH_REG -- the highest and lowest active registers.
1236 REGSTART, REGEND -- arrays of string positions.
1237 REG_INFO -- array of information about each subexpression.
1239 Also assumes the variables `fail_stack' and (if debugging), `bufp',
1240 `pend', `string1', `size1', `string2', and `size2'. */
1242 #define POP_FAILURE_POINT(str, pat, low_reg, high_reg, regstart, regend, reg_info)\
1244 DEBUG_STATEMENT (fail_stack_elt_t failure_id;) \
1246 const unsigned char *string_temp; \
1248 assert (!FAIL_STACK_EMPTY ()); \
1250 /* Remove failure points and point to how many regs pushed. */ \
1251 DEBUG_PRINT1 ("POP_FAILURE_POINT:\n"); \
1252 DEBUG_PRINT2 (" Before pop, next avail: %d\n", fail_stack.avail); \
1253 DEBUG_PRINT2 (" size: %d\n", fail_stack.size); \
1255 assert (fail_stack.avail >= NUM_NONREG_ITEMS); \
1257 DEBUG_POP (&failure_id); \
1258 DEBUG_PRINT2 (" Popping failure id: %u\n", failure_id); \
1260 /* If the saved string location is NULL, it came from an \
1261 on_failure_keep_string_jump opcode, and we want to throw away the \
1262 saved NULL, thus retaining our current position in the string. */ \
1263 string_temp = POP_FAILURE_POINTER (); \
1264 if (string_temp != NULL) \
1265 str = (const char *) string_temp; \
1267 DEBUG_PRINT2 (" Popping string 0x%x: `", str); \
1268 DEBUG_PRINT_DOUBLE_STRING (str, string1, size1, string2, size2); \
1269 DEBUG_PRINT1 ("'\n"); \
1271 pat = (unsigned char *) POP_FAILURE_POINTER (); \
1272 DEBUG_PRINT2 (" Popping pattern 0x%x: ", pat); \
1273 DEBUG_PRINT_COMPILED_PATTERN (bufp, pat, pend); \
1275 /* Restore register info. */ \
1276 high_reg = (unsigned) POP_FAILURE_INT (); \
1277 DEBUG_PRINT2 (" Popping high active reg: %d\n", high_reg); \
1279 low_reg = (unsigned) POP_FAILURE_INT (); \
1280 DEBUG_PRINT2 (" Popping low active reg: %d\n", low_reg); \
1282 if (!RE_NO_POSIX_BACKTRACKING & bufp->syntax) \
1283 for (this_reg = high_reg; this_reg >= low_reg; this_reg--) \
1285 DEBUG_PRINT2 (" Popping reg: %d\n", this_reg); \
1287 reg_info[this_reg].word = POP_FAILURE_ELT (); \
1288 DEBUG_PRINT2 (" info: 0x%x\n", reg_info[this_reg]); \
1290 regend[this_reg] = (const char *) POP_FAILURE_POINTER (); \
1291 DEBUG_PRINT2 (" end: 0x%x\n", regend[this_reg]); \
1293 regstart[this_reg] = (const char *) POP_FAILURE_POINTER (); \
1294 DEBUG_PRINT2 (" start: 0x%x\n", regstart[this_reg]); \
1297 set_regs_matched_done = 0; \
1298 DEBUG_STATEMENT (nfailure_points_popped++); \
1299 } /* POP_FAILURE_POINT */
1303 /* Structure for per-register (a.k.a. per-group) information.
1304 Other register information, such as the
1305 starting and ending positions (which are addresses), and the list of
1306 inner groups (which is a bits list) are maintained in separate
1309 We are making a (strictly speaking) nonportable assumption here: that
1310 the compiler will pack our bit fields into something that fits into
1311 the type of `word', i.e., is something that fits into one item on the
1316 fail_stack_elt_t word;
1319 /* This field is one if this group can match the empty string,
1320 zero if not. If not yet determined, `MATCH_NULL_UNSET_VALUE'. */
1321 #define MATCH_NULL_UNSET_VALUE 3
1322 unsigned match_null_string_p : 2;
1323 unsigned is_active : 1;
1324 unsigned matched_something : 1;
1325 unsigned ever_matched_something : 1;
1327 } register_info_type;
1329 #define REG_MATCH_NULL_STRING_P(R) ((R).bits.match_null_string_p)
1330 #define IS_ACTIVE(R) ((R).bits.is_active)
1331 #define MATCHED_SOMETHING(R) ((R).bits.matched_something)
1332 #define EVER_MATCHED_SOMETHING(R) ((R).bits.ever_matched_something)
1335 /* Call this when have matched a real character; it sets `matched' flags
1336 for the subexpressions which we are currently inside. Also records
1337 that those subexprs have matched. */
1338 #define SET_REGS_MATCHED() \
1341 if (!set_regs_matched_done) \
1344 set_regs_matched_done = 1; \
1345 for (r = lowest_active_reg; r <= highest_active_reg; r++) \
1347 MATCHED_SOMETHING (reg_info[r]) \
1348 = EVER_MATCHED_SOMETHING (reg_info[r]) \
1355 /* Registers are set to a sentinel when they haven't yet matched. */
1356 static char reg_unset_dummy;
1357 #define REG_UNSET_VALUE (®_unset_dummy)
1358 #define REG_UNSET(e) ((e) == REG_UNSET_VALUE)
1360 /* Subroutine declarations and macros for regex_compile. */
1362 static void store_op1 (), store_op2 ();
1363 static void insert_op1 (), insert_op2 ();
1364 static boolean at_begline_loc_p (), at_endline_loc_p ();
1365 static boolean group_in_compile_stack ();
1366 static reg_errcode_t compile_range ();
1368 /* Fetch the next character in the uncompiled pattern---translating it
1369 if necessary. Also cast from a signed character in the constant
1370 string passed to us by the user to an unsigned char that we can use
1371 as an array index (in, e.g., `translate'). */
1372 #define PATFETCH(c) \
1373 do {if (p == pend) return REG_EEND; \
1374 c = (unsigned char) *p++; \
1375 if (translate) c = translate[c]; \
1378 /* Fetch the next character in the uncompiled pattern, with no
1380 #define PATFETCH_RAW(c) \
1381 do {if (p == pend) return REG_EEND; \
1382 c = (unsigned char) *p++; \
1385 /* Go backwards one character in the pattern. */
1386 #define PATUNFETCH p--
1389 /* If `translate' is non-null, return translate[D], else just D. We
1390 cast the subscript to translate because some data is declared as
1391 `char *', to avoid warnings when a string constant is passed. But
1392 when we use a character as a subscript we must make it unsigned. */
1393 #define TRANSLATE(d) (translate ? translate[(unsigned char) (d)] : (d))
1396 /* Macros for outputting the compiled pattern into `buffer'. */
1398 /* If the buffer isn't allocated when it comes in, use this. */
1399 #define INIT_BUF_SIZE 32
1401 /* Make sure we have at least N more bytes of space in buffer. */
1402 #define GET_BUFFER_SPACE(n) \
1403 while (b - bufp->buffer + (n) > bufp->allocated) \
1406 /* Make sure we have one more byte of buffer space and then add C to it. */
1407 #define BUF_PUSH(c) \
1409 GET_BUFFER_SPACE (1); \
1410 *b++ = (unsigned char) (c); \
1414 /* Ensure we have two more bytes of buffer space and then append C1 and C2. */
1415 #define BUF_PUSH_2(c1, c2) \
1417 GET_BUFFER_SPACE (2); \
1418 *b++ = (unsigned char) (c1); \
1419 *b++ = (unsigned char) (c2); \
1423 /* As with BUF_PUSH_2, except for three bytes. */
1424 #define BUF_PUSH_3(c1, c2, c3) \
1426 GET_BUFFER_SPACE (3); \
1427 *b++ = (unsigned char) (c1); \
1428 *b++ = (unsigned char) (c2); \
1429 *b++ = (unsigned char) (c3); \
1433 /* Store a jump with opcode OP at LOC to location TO. We store a
1434 relative address offset by the three bytes the jump itself occupies. */
1435 #define STORE_JUMP(op, loc, to) \
1436 store_op1 (op, loc, (to) - (loc) - 3)
1438 /* Likewise, for a two-argument jump. */
1439 #define STORE_JUMP2(op, loc, to, arg) \
1440 store_op2 (op, loc, (to) - (loc) - 3, arg)
1442 /* Like `STORE_JUMP', but for inserting. Assume `b' is the buffer end. */
1443 #define INSERT_JUMP(op, loc, to) \
1444 insert_op1 (op, loc, (to) - (loc) - 3, b)
1446 /* Like `STORE_JUMP2', but for inserting. Assume `b' is the buffer end. */
1447 #define INSERT_JUMP2(op, loc, to, arg) \
1448 insert_op2 (op, loc, (to) - (loc) - 3, arg, b)
1451 /* This is not an arbitrary limit: the arguments which represent offsets
1452 into the pattern are two bytes long. So if 2^16 bytes turns out to
1453 be too small, many things would have to change. */
1454 #define MAX_BUF_SIZE (1L << 16)
1457 /* Extend the buffer by twice its current size via realloc and
1458 reset the pointers that pointed into the old block to point to the
1459 correct places in the new one. If extending the buffer results in it
1460 being larger than MAX_BUF_SIZE, then flag memory exhausted. */
1461 #define EXTEND_BUFFER() \
1463 unsigned char *old_buffer = bufp->buffer; \
1464 if (bufp->allocated == MAX_BUF_SIZE) \
1466 bufp->allocated <<= 1; \
1467 if (bufp->allocated > MAX_BUF_SIZE) \
1468 bufp->allocated = MAX_BUF_SIZE; \
1469 bufp->buffer = (unsigned char *) realloc (bufp->buffer, bufp->allocated);\
1470 if (bufp->buffer == NULL) \
1471 return REG_ESPACE; \
1472 /* If the buffer moved, move all the pointers into it. */ \
1473 if (old_buffer != bufp->buffer) \
1475 b = (b - old_buffer) + bufp->buffer; \
1476 begalt = (begalt - old_buffer) + bufp->buffer; \
1477 if (fixup_alt_jump) \
1478 fixup_alt_jump = (fixup_alt_jump - old_buffer) + bufp->buffer;\
1480 laststart = (laststart - old_buffer) + bufp->buffer; \
1481 if (pending_exact) \
1482 pending_exact = (pending_exact - old_buffer) + bufp->buffer; \
1487 /* Since we have one byte reserved for the register number argument to
1488 {start,stop}_memory, the maximum number of groups we can report
1489 things about is what fits in that byte. */
1490 #define MAX_REGNUM 255
1492 /* But patterns can have more than `MAX_REGNUM' registers. We just
1493 ignore the excess. */
1494 typedef unsigned regnum_t;
1497 /* Macros for the compile stack. */
1499 /* Since offsets can go either forwards or backwards, this type needs to
1500 be able to hold values from -(MAX_BUF_SIZE - 1) to MAX_BUF_SIZE - 1. */
1501 typedef int pattern_offset_t;
1505 pattern_offset_t begalt_offset;
1506 pattern_offset_t fixup_alt_jump;
1507 pattern_offset_t inner_group_offset;
1508 pattern_offset_t laststart_offset;
1510 } compile_stack_elt_t;
1515 compile_stack_elt_t *stack;
1517 unsigned avail; /* Offset of next open position. */
1518 } compile_stack_type;
1521 #define INIT_COMPILE_STACK_SIZE 32
1523 #define COMPILE_STACK_EMPTY (compile_stack.avail == 0)
1524 #define COMPILE_STACK_FULL (compile_stack.avail == compile_stack.size)
1526 /* The next available element. */
1527 #define COMPILE_STACK_TOP (compile_stack.stack[compile_stack.avail])
1530 /* Set the bit for character C in a list. */
1531 #define SET_LIST_BIT(c) \
1532 (b[((unsigned char) (c)) / BYTEWIDTH] \
1533 |= 1 << (((unsigned char) c) % BYTEWIDTH))
1536 /* Get the next unsigned number in the uncompiled pattern. */
1537 #define GET_UNSIGNED_NUMBER(num) \
1541 while (ISDIGIT (c)) \
1545 num = num * 10 + c - '0'; \
1553 #define CHAR_CLASS_MAX_LENGTH 6 /* Namely, `xdigit'. */
1555 #define IS_CHAR_CLASS(string) \
1556 (STREQ (string, "alpha") || STREQ (string, "upper") \
1557 || STREQ (string, "lower") || STREQ (string, "digit") \
1558 || STREQ (string, "alnum") || STREQ (string, "xdigit") \
1559 || STREQ (string, "space") || STREQ (string, "print") \
1560 || STREQ (string, "punct") || STREQ (string, "graph") \
1561 || STREQ (string, "cntrl") || STREQ (string, "blank"))
1563 #ifndef MATCH_MAY_ALLOCATE
1565 /* If we cannot allocate large objects within re_match_2_internal,
1566 we make the fail stack and register vectors global.
1567 The fail stack, we grow to the maximum size when a regexp
1569 The register vectors, we adjust in size each time we
1570 compile a regexp, according to the number of registers it needs. */
1572 static fail_stack_type fail_stack;
1574 /* Size with which the following vectors are currently allocated.
1575 That is so we can make them bigger as needed,
1576 but never make them smaller. */
1577 static int regs_allocated_size;
1579 static const char ** regstart, ** regend;
1580 static const char ** old_regstart, ** old_regend;
1581 static const char **best_regstart, **best_regend;
1582 static register_info_type *reg_info;
1583 static const char **reg_dummy;
1584 static register_info_type *reg_info_dummy;
1586 /* Make the register vectors big enough for NUM_REGS registers,
1587 but don't make them smaller. */
1590 regex_grow_registers (num_regs)
1593 if (num_regs > regs_allocated_size)
1595 RETALLOC_IF (regstart, num_regs, const char *);
1596 RETALLOC_IF (regend, num_regs, const char *);
1597 RETALLOC_IF (old_regstart, num_regs, const char *);
1598 RETALLOC_IF (old_regend, num_regs, const char *);
1599 RETALLOC_IF (best_regstart, num_regs, const char *);
1600 RETALLOC_IF (best_regend, num_regs, const char *);
1601 RETALLOC_IF (reg_info, num_regs, register_info_type);
1602 RETALLOC_IF (reg_dummy, num_regs, const char *);
1603 RETALLOC_IF (reg_info_dummy, num_regs, register_info_type);
1605 regs_allocated_size = num_regs;
1609 #endif /* not MATCH_MAY_ALLOCATE */
1611 /* `regex_compile' compiles PATTERN (of length SIZE) according to SYNTAX.
1612 Returns one of error codes defined in `regex.h', or zero for success.
1614 Assumes the `allocated' (and perhaps `buffer') and `translate'
1615 fields are set in BUFP on entry.
1617 If it succeeds, results are put in BUFP (if it returns an error, the
1618 contents of BUFP are undefined):
1619 `buffer' is the compiled pattern;
1620 `syntax' is set to SYNTAX;
1621 `used' is set to the length of the compiled pattern;
1622 `fastmap_accurate' is zero;
1623 `re_nsub' is the number of subexpressions in PATTERN;
1624 `not_bol' and `not_eol' are zero;
1626 The `fastmap' and `newline_anchor' fields are neither
1627 examined nor set. */
1629 /* Return, freeing storage we allocated. */
1630 #define FREE_STACK_RETURN(value) \
1631 return (free (compile_stack.stack), value)
1633 static reg_errcode_t
1634 regex_compile (pattern, size, syntax, bufp)
1635 const char *pattern;
1637 reg_syntax_t syntax;
1638 struct re_pattern_buffer *bufp;
1640 /* We fetch characters from PATTERN here. Even though PATTERN is
1641 `char *' (i.e., signed), we declare these variables as unsigned, so
1642 they can be reliably used as array indices. */
1643 register unsigned char c, c1;
1645 /* A random temporary spot in PATTERN. */
1648 /* Points to the end of the buffer, where we should append. */
1649 register unsigned char *b;
1651 /* Keeps track of unclosed groups. */
1652 compile_stack_type compile_stack;
1654 /* Points to the current (ending) position in the pattern. */
1655 const char *p = pattern;
1656 const char *pend = pattern + size;
1658 /* How to translate the characters in the pattern. */
1659 char *translate = bufp->translate;
1661 /* Address of the count-byte of the most recently inserted `exactn'
1662 command. This makes it possible to tell if a new exact-match
1663 character can be added to that command or if the character requires
1664 a new `exactn' command. */
1665 unsigned char *pending_exact = 0;
1667 /* Address of start of the most recently finished expression.
1668 This tells, e.g., postfix * where to find the start of its
1669 operand. Reset at the beginning of groups and alternatives. */
1670 unsigned char *laststart = 0;
1672 /* Address of beginning of regexp, or inside of last group. */
1673 unsigned char *begalt;
1675 /* Place in the uncompiled pattern (i.e., the {) to
1676 which to go back if the interval is invalid. */
1677 const char *beg_interval;
1679 /* Address of the place where a forward jump should go to the end of
1680 the containing expression. Each alternative of an `or' -- except the
1681 last -- ends with a forward jump of this sort. */
1682 unsigned char *fixup_alt_jump = 0;
1684 /* Counts open-groups as they are encountered. Remembered for the
1685 matching close-group on the compile stack, so the same register
1686 number is put in the stop_memory as the start_memory. */
1687 regnum_t regnum = 0;
1690 DEBUG_PRINT1 ("\nCompiling pattern: ");
1693 unsigned debug_count;
1695 for (debug_count = 0; debug_count < size; debug_count++)
1696 putchar (pattern[debug_count]);
1701 /* Initialize the compile stack. */
1702 compile_stack.stack = TALLOC (INIT_COMPILE_STACK_SIZE, compile_stack_elt_t);
1703 if (compile_stack.stack == NULL)
1706 compile_stack.size = INIT_COMPILE_STACK_SIZE;
1707 compile_stack.avail = 0;
1709 /* Initialize the pattern buffer. */
1710 bufp->syntax = syntax;
1711 bufp->fastmap_accurate = 0;
1712 bufp->not_bol = bufp->not_eol = 0;
1714 /* Set `used' to zero, so that if we return an error, the pattern
1715 printer (for debugging) will think there's no pattern. We reset it
1719 /* Always count groups, whether or not bufp->no_sub is set. */
1722 #if !defined (emacs) && !defined (SYNTAX_TABLE)
1723 /* Initialize the syntax table. */
1724 init_syntax_once ();
1727 if (bufp->allocated == 0)
1730 { /* If zero allocated, but buffer is non-null, try to realloc
1731 enough space. This loses if buffer's address is bogus, but
1732 that is the user's responsibility. */
1733 RETALLOC (bufp->buffer, INIT_BUF_SIZE, unsigned char);
1736 { /* Caller did not allocate a buffer. Do it for them. */
1737 bufp->buffer = TALLOC (INIT_BUF_SIZE, unsigned char);
1739 if (!bufp->buffer) FREE_STACK_RETURN (REG_ESPACE);
1741 bufp->allocated = INIT_BUF_SIZE;
1744 begalt = b = bufp->buffer;
1746 /* Loop through the uncompiled pattern until we're at the end. */
1755 if ( /* If at start of pattern, it's an operator. */
1757 /* If context independent, it's an operator. */
1758 || syntax & RE_CONTEXT_INDEP_ANCHORS
1759 /* Otherwise, depends on what's come before. */
1760 || at_begline_loc_p (pattern, p, syntax))
1770 if ( /* If at end of pattern, it's an operator. */
1772 /* If context independent, it's an operator. */
1773 || syntax & RE_CONTEXT_INDEP_ANCHORS
1774 /* Otherwise, depends on what's next. */
1775 || at_endline_loc_p (p, pend, syntax))
1785 if ((syntax & RE_BK_PLUS_QM)
1786 || (syntax & RE_LIMITED_OPS))
1790 /* If there is no previous pattern... */
1793 if (syntax & RE_CONTEXT_INVALID_OPS)
1794 FREE_STACK_RETURN (REG_BADRPT);
1795 else if (!(syntax & RE_CONTEXT_INDEP_OPS))
1800 /* Are we optimizing this jump? */
1801 boolean keep_string_p = false;
1803 /* 1 means zero (many) matches is allowed. */
1804 char zero_times_ok = 0, many_times_ok = 0;
1806 /* If there is a sequence of repetition chars, collapse it
1807 down to just one (the right one). We can't combine
1808 interval operators with these because of, e.g., `a{2}*',
1809 which should only match an even number of `a's. */
1813 zero_times_ok |= c != '+';
1814 many_times_ok |= c != '?';
1822 || (!(syntax & RE_BK_PLUS_QM) && (c == '+' || c == '?')))
1825 else if (syntax & RE_BK_PLUS_QM && c == '\\')
1827 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
1830 if (!(c1 == '+' || c1 == '?'))
1845 /* If we get here, we found another repeat character. */
1848 /* Star, etc. applied to an empty pattern is equivalent
1849 to an empty pattern. */
1853 /* Now we know whether or not zero matches is allowed
1854 and also whether or not two or more matches is allowed. */
1856 { /* More than one repetition is allowed, so put in at the
1857 end a backward relative jump from `b' to before the next
1858 jump we're going to put in below (which jumps from
1859 laststart to after this jump).
1861 But if we are at the `*' in the exact sequence `.*\n',
1862 insert an unconditional jump backwards to the .,
1863 instead of the beginning of the loop. This way we only
1864 push a failure point once, instead of every time
1865 through the loop. */
1866 assert (p - 1 > pattern);
1868 /* Allocate the space for the jump. */
1869 GET_BUFFER_SPACE (3);
1871 /* We know we are not at the first character of the pattern,
1872 because laststart was nonzero. And we've already
1873 incremented `p', by the way, to be the character after
1874 the `*'. Do we have to do something analogous here
1875 for null bytes, because of RE_DOT_NOT_NULL? */
1876 if (TRANSLATE (*(p - 2)) == TRANSLATE ('.')
1878 && p < pend && TRANSLATE (*p) == TRANSLATE ('\n')
1879 && !(syntax & RE_DOT_NEWLINE))
1880 { /* We have .*\n. */
1881 STORE_JUMP (jump, b, laststart);
1882 keep_string_p = true;
1885 /* Anything else. */
1886 STORE_JUMP (maybe_pop_jump, b, laststart - 3);
1888 /* We've added more stuff to the buffer. */
1892 /* On failure, jump from laststart to b + 3, which will be the
1893 end of the buffer after this jump is inserted. */
1894 GET_BUFFER_SPACE (3);
1895 INSERT_JUMP (keep_string_p ? on_failure_keep_string_jump
1903 /* At least one repetition is required, so insert a
1904 `dummy_failure_jump' before the initial
1905 `on_failure_jump' instruction of the loop. This
1906 effects a skip over that instruction the first time
1907 we hit that loop. */
1908 GET_BUFFER_SPACE (3);
1909 INSERT_JUMP (dummy_failure_jump, laststart, laststart + 6);
1924 boolean had_char_class = false;
1926 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
1928 /* Ensure that we have enough space to push a charset: the
1929 opcode, the length count, and the bitset; 34 bytes in all. */
1930 GET_BUFFER_SPACE (34);
1934 /* We test `*p == '^' twice, instead of using an if
1935 statement, so we only need one BUF_PUSH. */
1936 BUF_PUSH (*p == '^' ? charset_not : charset);
1940 /* Remember the first position in the bracket expression. */
1943 /* Push the number of bytes in the bitmap. */
1944 BUF_PUSH ((1 << BYTEWIDTH) / BYTEWIDTH);
1946 /* Clear the whole map. */
1947 bzero (b, (1 << BYTEWIDTH) / BYTEWIDTH);
1949 /* charset_not matches newline according to a syntax bit. */
1950 if ((re_opcode_t) b[-2] == charset_not
1951 && (syntax & RE_HAT_LISTS_NOT_NEWLINE))
1952 SET_LIST_BIT ('\n');
1954 /* Read in characters and ranges, setting map bits. */
1957 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
1961 /* \ might escape characters inside [...] and [^...]. */
1962 if ((syntax & RE_BACKSLASH_ESCAPE_IN_LISTS) && c == '\\')
1964 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
1971 /* Could be the end of the bracket expression. If it's
1972 not (i.e., when the bracket expression is `[]' so
1973 far), the ']' character bit gets set way below. */
1974 if (c == ']' && p != p1 + 1)
1977 /* Look ahead to see if it's a range when the last thing
1978 was a character class. */
1979 if (had_char_class && c == '-' && *p != ']')
1980 FREE_STACK_RETURN (REG_ERANGE);
1982 /* Look ahead to see if it's a range when the last thing
1983 was a character: if this is a hyphen not at the
1984 beginning or the end of a list, then it's the range
1987 && !(p - 2 >= pattern && p[-2] == '[')
1988 && !(p - 3 >= pattern && p[-3] == '[' && p[-2] == '^')
1992 = compile_range (&p, pend, translate, syntax, b);
1993 if (ret != REG_NOERROR) FREE_STACK_RETURN (ret);
1996 else if (p[0] == '-' && p[1] != ']')
1997 { /* This handles ranges made up of characters only. */
2000 /* Move past the `-'. */
2003 ret = compile_range (&p, pend, translate, syntax, b);
2004 if (ret != REG_NOERROR) FREE_STACK_RETURN (ret);
2007 /* See if we're at the beginning of a possible character
2010 else if (syntax & RE_CHAR_CLASSES && c == '[' && *p == ':')
2011 { /* Leave room for the null. */
2012 char str[CHAR_CLASS_MAX_LENGTH + 1];
2017 /* If pattern is `[[:'. */
2018 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2023 if (c == ':' || c == ']' || p == pend
2024 || c1 == CHAR_CLASS_MAX_LENGTH)
2030 /* If isn't a word bracketed by `[:' and:`]':
2031 undo the ending character, the letters, and leave
2032 the leading `:' and `[' (but set bits for them). */
2033 if (c == ':' && *p == ']')
2036 boolean is_alnum = STREQ (str, "alnum");
2037 boolean is_alpha = STREQ (str, "alpha");
2038 boolean is_blank = STREQ (str, "blank");
2039 boolean is_cntrl = STREQ (str, "cntrl");
2040 boolean is_digit = STREQ (str, "digit");
2041 boolean is_graph = STREQ (str, "graph");
2042 boolean is_lower = STREQ (str, "lower");
2043 boolean is_print = STREQ (str, "print");
2044 boolean is_punct = STREQ (str, "punct");
2045 boolean is_space = STREQ (str, "space");
2046 boolean is_upper = STREQ (str, "upper");
2047 boolean is_xdigit = STREQ (str, "xdigit");
2049 if (!IS_CHAR_CLASS (str))
2050 FREE_STACK_RETURN (REG_ECTYPE);
2052 /* Throw away the ] at the end of the character
2056 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2058 for (ch = 0; ch < 1 << BYTEWIDTH; ch++)
2060 /* This was split into 3 if's to
2061 avoid an arbitrary limit in some compiler. */
2062 if ( (is_alnum && ISALNUM (ch))
2063 || (is_alpha && ISALPHA (ch))
2064 || (is_blank && ISBLANK (ch))
2065 || (is_cntrl && ISCNTRL (ch)))
2067 if ( (is_digit && ISDIGIT (ch))
2068 || (is_graph && ISGRAPH (ch))
2069 || (is_lower && ISLOWER (ch))
2070 || (is_print && ISPRINT (ch)))
2072 if ( (is_punct && ISPUNCT (ch))
2073 || (is_space && ISSPACE (ch))
2074 || (is_upper && ISUPPER (ch))
2075 || (is_xdigit && ISXDIGIT (ch)))
2078 had_char_class = true;
2087 had_char_class = false;
2092 had_char_class = false;
2097 /* Discard any (non)matching list bytes that are all 0 at the
2098 end of the map. Decrease the map-length byte too. */
2099 while ((int) b[-1] > 0 && b[b[-1] - 1] == 0)
2107 if (syntax & RE_NO_BK_PARENS)
2114 if (syntax & RE_NO_BK_PARENS)
2121 if (syntax & RE_NEWLINE_ALT)
2128 if (syntax & RE_NO_BK_VBAR)
2135 if (syntax & RE_INTERVALS && syntax & RE_NO_BK_BRACES)
2136 goto handle_interval;
2142 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
2144 /* Do not translate the character after the \, so that we can
2145 distinguish, e.g., \B from \b, even if we normally would
2146 translate, e.g., B to b. */
2152 if (syntax & RE_NO_BK_PARENS)
2153 goto normal_backslash;
2159 if (COMPILE_STACK_FULL)
2161 RETALLOC (compile_stack.stack, compile_stack.size << 1,
2162 compile_stack_elt_t);
2163 if (compile_stack.stack == NULL) return REG_ESPACE;
2165 compile_stack.size <<= 1;
2168 /* These are the values to restore when we hit end of this
2169 group. They are all relative offsets, so that if the
2170 whole pattern moves because of realloc, they will still
2172 COMPILE_STACK_TOP.begalt_offset = begalt - bufp->buffer;
2173 COMPILE_STACK_TOP.fixup_alt_jump
2174 = fixup_alt_jump ? fixup_alt_jump - bufp->buffer + 1 : 0;
2175 COMPILE_STACK_TOP.laststart_offset = b - bufp->buffer;
2176 COMPILE_STACK_TOP.regnum = regnum;
2178 /* We will eventually replace the 0 with the number of
2179 groups inner to this one. But do not push a
2180 start_memory for groups beyond the last one we can
2181 represent in the compiled pattern. */
2182 if (regnum <= MAX_REGNUM)
2184 COMPILE_STACK_TOP.inner_group_offset = b - bufp->buffer + 2;
2185 BUF_PUSH_3 (start_memory, regnum, 0);
2188 compile_stack.avail++;
2193 /* If we've reached MAX_REGNUM groups, then this open
2194 won't actually generate any code, so we'll have to
2195 clear pending_exact explicitly. */
2201 if (syntax & RE_NO_BK_PARENS) goto normal_backslash;
2203 if (COMPILE_STACK_EMPTY)
2204 if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD)
2205 goto normal_backslash;
2207 FREE_STACK_RETURN (REG_ERPAREN);
2211 { /* Push a dummy failure point at the end of the
2212 alternative for a possible future
2213 `pop_failure_jump' to pop. See comments at
2214 `push_dummy_failure' in `re_match_2'. */
2215 BUF_PUSH (push_dummy_failure);
2217 /* We allocated space for this jump when we assigned
2218 to `fixup_alt_jump', in the `handle_alt' case below. */
2219 STORE_JUMP (jump_past_alt, fixup_alt_jump, b - 1);
2222 /* See similar code for backslashed left paren above. */
2223 if (COMPILE_STACK_EMPTY)
2224 if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD)
2227 FREE_STACK_RETURN (REG_ERPAREN);
2229 /* Since we just checked for an empty stack above, this
2230 ``can't happen''. */
2231 assert (compile_stack.avail != 0);
2233 /* We don't just want to restore into `regnum', because
2234 later groups should continue to be numbered higher,
2235 as in `(ab)c(de)' -- the second group is #2. */
2236 regnum_t this_group_regnum;
2238 compile_stack.avail--;
2239 begalt = bufp->buffer + COMPILE_STACK_TOP.begalt_offset;
2241 = COMPILE_STACK_TOP.fixup_alt_jump
2242 ? bufp->buffer + COMPILE_STACK_TOP.fixup_alt_jump - 1
2244 laststart = bufp->buffer + COMPILE_STACK_TOP.laststart_offset;
2245 this_group_regnum = COMPILE_STACK_TOP.regnum;
2246 /* If we've reached MAX_REGNUM groups, then this open
2247 won't actually generate any code, so we'll have to
2248 clear pending_exact explicitly. */
2251 /* We're at the end of the group, so now we know how many
2252 groups were inside this one. */
2253 if (this_group_regnum <= MAX_REGNUM)
2255 unsigned char *inner_group_loc
2256 = bufp->buffer + COMPILE_STACK_TOP.inner_group_offset;
2258 *inner_group_loc = regnum - this_group_regnum;
2259 BUF_PUSH_3 (stop_memory, this_group_regnum,
2260 regnum - this_group_regnum);
2266 case '|': /* `\|'. */
2267 if (syntax & RE_LIMITED_OPS || syntax & RE_NO_BK_VBAR)
2268 goto normal_backslash;
2270 if (syntax & RE_LIMITED_OPS)
2273 /* Insert before the previous alternative a jump which
2274 jumps to this alternative if the former fails. */
2275 GET_BUFFER_SPACE (3);
2276 INSERT_JUMP (on_failure_jump, begalt, b + 6);
2280 /* The alternative before this one has a jump after it
2281 which gets executed if it gets matched. Adjust that
2282 jump so it will jump to this alternative's analogous
2283 jump (put in below, which in turn will jump to the next
2284 (if any) alternative's such jump, etc.). The last such
2285 jump jumps to the correct final destination. A picture:
2291 If we are at `b', then fixup_alt_jump right now points to a
2292 three-byte space after `a'. We'll put in the jump, set
2293 fixup_alt_jump to right after `b', and leave behind three
2294 bytes which we'll fill in when we get to after `c'. */
2297 STORE_JUMP (jump_past_alt, fixup_alt_jump, b);
2299 /* Mark and leave space for a jump after this alternative,
2300 to be filled in later either by next alternative or
2301 when know we're at the end of a series of alternatives. */
2303 GET_BUFFER_SPACE (3);
2312 /* If \{ is a literal. */
2313 if (!(syntax & RE_INTERVALS)
2314 /* If we're at `\{' and it's not the open-interval
2316 || ((syntax & RE_INTERVALS) && (syntax & RE_NO_BK_BRACES))
2317 || (p - 2 == pattern && p == pend))
2318 goto normal_backslash;
2322 /* If got here, then the syntax allows intervals. */
2324 /* At least (most) this many matches must be made. */
2325 int lower_bound = -1, upper_bound = -1;
2327 beg_interval = p - 1;
2331 if (syntax & RE_NO_BK_BRACES)
2332 goto unfetch_interval;
2334 FREE_STACK_RETURN (REG_EBRACE);
2337 GET_UNSIGNED_NUMBER (lower_bound);
2341 GET_UNSIGNED_NUMBER (upper_bound);
2342 if (upper_bound < 0) upper_bound = RE_DUP_MAX;
2345 /* Interval such as `{1}' => match exactly once. */
2346 upper_bound = lower_bound;
2348 if (lower_bound < 0 || upper_bound > RE_DUP_MAX
2349 || lower_bound > upper_bound)
2351 if (syntax & RE_NO_BK_BRACES)
2352 goto unfetch_interval;
2354 FREE_STACK_RETURN (REG_BADBR);
2357 if (!(syntax & RE_NO_BK_BRACES))
2359 if (c != '\\') FREE_STACK_RETURN (REG_EBRACE);
2366 if (syntax & RE_NO_BK_BRACES)
2367 goto unfetch_interval;
2369 FREE_STACK_RETURN (REG_BADBR);
2372 /* We just parsed a valid interval. */
2374 /* If it's invalid to have no preceding re. */
2377 if (syntax & RE_CONTEXT_INVALID_OPS)
2378 FREE_STACK_RETURN (REG_BADRPT);
2379 else if (syntax & RE_CONTEXT_INDEP_OPS)
2382 goto unfetch_interval;
2385 /* If the upper bound is zero, don't want to succeed at
2386 all; jump from `laststart' to `b + 3', which will be
2387 the end of the buffer after we insert the jump. */
2388 if (upper_bound == 0)
2390 GET_BUFFER_SPACE (3);
2391 INSERT_JUMP (jump, laststart, b + 3);
2395 /* Otherwise, we have a nontrivial interval. When
2396 we're all done, the pattern will look like:
2397 set_number_at <jump count> <upper bound>
2398 set_number_at <succeed_n count> <lower bound>
2399 succeed_n <after jump addr> <succeed_n count>
2401 jump_n <succeed_n addr> <jump count>
2402 (The upper bound and `jump_n' are omitted if
2403 `upper_bound' is 1, though.) */
2405 { /* If the upper bound is > 1, we need to insert
2406 more at the end of the loop. */
2407 unsigned nbytes = 10 + (upper_bound > 1) * 10;
2409 GET_BUFFER_SPACE (nbytes);
2411 /* Initialize lower bound of the `succeed_n', even
2412 though it will be set during matching by its
2413 attendant `set_number_at' (inserted next),
2414 because `re_compile_fastmap' needs to know.
2415 Jump to the `jump_n' we might insert below. */
2416 INSERT_JUMP2 (succeed_n, laststart,
2417 b + 5 + (upper_bound > 1) * 5,
2421 /* Code to initialize the lower bound. Insert
2422 before the `succeed_n'. The `5' is the last two
2423 bytes of this `set_number_at', plus 3 bytes of
2424 the following `succeed_n'. */
2425 insert_op2 (set_number_at, laststart, 5, lower_bound, b);
2428 if (upper_bound > 1)
2429 { /* More than one repetition is allowed, so
2430 append a backward jump to the `succeed_n'
2431 that starts this interval.
2433 When we've reached this during matching,
2434 we'll have matched the interval once, so
2435 jump back only `upper_bound - 1' times. */
2436 STORE_JUMP2 (jump_n, b, laststart + 5,
2440 /* The location we want to set is the second
2441 parameter of the `jump_n'; that is `b-2' as
2442 an absolute address. `laststart' will be
2443 the `set_number_at' we're about to insert;
2444 `laststart+3' the number to set, the source
2445 for the relative address. But we are
2446 inserting into the middle of the pattern --
2447 so everything is getting moved up by 5.
2448 Conclusion: (b - 2) - (laststart + 3) + 5,
2449 i.e., b - laststart.
2451 We insert this at the beginning of the loop
2452 so that if we fail during matching, we'll
2453 reinitialize the bounds. */
2454 insert_op2 (set_number_at, laststart, b - laststart,
2455 upper_bound - 1, b);
2460 beg_interval = NULL;
2465 /* If an invalid interval, match the characters as literals. */
2466 assert (beg_interval);
2468 beg_interval = NULL;
2470 /* normal_char and normal_backslash need `c'. */
2473 if (!(syntax & RE_NO_BK_BRACES))
2475 if (p > pattern && p[-1] == '\\')
2476 goto normal_backslash;
2481 /* There is no way to specify the before_dot and after_dot
2482 operators. rms says this is ok. --karl */
2490 BUF_PUSH_2 (syntaxspec, syntax_spec_code[c]);
2496 BUF_PUSH_2 (notsyntaxspec, syntax_spec_code[c]);
2503 BUF_PUSH (wordchar);
2509 BUF_PUSH (notwordchar);
2522 BUF_PUSH (wordbound);
2526 BUF_PUSH (notwordbound);
2537 case '1': case '2': case '3': case '4': case '5':
2538 case '6': case '7': case '8': case '9':
2539 if (syntax & RE_NO_BK_REFS)
2545 FREE_STACK_RETURN (REG_ESUBREG);
2547 /* Can't back reference to a subexpression if inside of it. */
2548 if (group_in_compile_stack (compile_stack, c1))
2552 BUF_PUSH_2 (duplicate, c1);
2558 if (syntax & RE_BK_PLUS_QM)
2561 goto normal_backslash;
2565 /* You might think it would be useful for \ to mean
2566 not to translate; but if we don't translate it
2567 it will never match anything. */
2575 /* Expects the character in `c'. */
2577 /* If no exactn currently being built. */
2580 /* If last exactn not at current position. */
2581 || pending_exact + *pending_exact + 1 != b
2583 /* We have only one byte following the exactn for the count. */
2584 || *pending_exact == (1 << BYTEWIDTH) - 1
2586 /* If followed by a repetition operator. */
2587 || *p == '*' || *p == '^'
2588 || ((syntax & RE_BK_PLUS_QM)
2589 ? *p == '\\' && (p[1] == '+' || p[1] == '?')
2590 : (*p == '+' || *p == '?'))
2591 || ((syntax & RE_INTERVALS)
2592 && ((syntax & RE_NO_BK_BRACES)
2594 : (p[0] == '\\' && p[1] == '{'))))
2596 /* Start building a new exactn. */
2600 BUF_PUSH_2 (exactn, 0);
2601 pending_exact = b - 1;
2608 } /* while p != pend */
2611 /* Through the pattern now. */
2614 STORE_JUMP (jump_past_alt, fixup_alt_jump, b);
2616 if (!COMPILE_STACK_EMPTY)
2617 FREE_STACK_RETURN (REG_EPAREN);
2619 /* If we don't want backtracking, force success
2620 the first time we reach the end of the compiled pattern. */
2621 if (syntax & RE_NO_POSIX_BACKTRACKING)
2624 free (compile_stack.stack);
2626 /* We have succeeded; set the length of the buffer. */
2627 bufp->used = b - bufp->buffer;
2632 DEBUG_PRINT1 ("\nCompiled pattern: \n");
2633 print_compiled_pattern (bufp);
2637 #ifndef MATCH_MAY_ALLOCATE
2638 /* Initialize the failure stack to the largest possible stack. This
2639 isn't necessary unless we're trying to avoid calling alloca in
2640 the search and match routines. */
2642 int num_regs = bufp->re_nsub + 1;
2644 /* Since DOUBLE_FAIL_STACK refuses to double only if the current size
2645 is strictly greater than re_max_failures, the largest possible stack
2646 is 2 * re_max_failures failure points. */
2647 if (fail_stack.size < (2 * re_max_failures * MAX_FAILURE_ITEMS))
2649 fail_stack.size = (2 * re_max_failures * MAX_FAILURE_ITEMS);
2652 if (! fail_stack.stack)
2654 = (fail_stack_elt_t *) xmalloc (fail_stack.size
2655 * sizeof (fail_stack_elt_t));
2658 = (fail_stack_elt_t *) xrealloc (fail_stack.stack,
2660 * sizeof (fail_stack_elt_t)));
2661 #else /* not emacs */
2662 if (! fail_stack.stack)
2664 = (fail_stack_elt_t *) malloc (fail_stack.size
2665 * sizeof (fail_stack_elt_t));
2668 = (fail_stack_elt_t *) realloc (fail_stack.stack,
2670 * sizeof (fail_stack_elt_t)));
2671 #endif /* not emacs */
2674 regex_grow_registers (num_regs);
2676 #endif /* not MATCH_MAY_ALLOCATE */
2679 } /* regex_compile */
2681 /* Subroutines for `regex_compile'. */
2683 /* Store OP at LOC followed by two-byte integer parameter ARG. */
2686 store_op1 (op, loc, arg)
2691 *loc = (unsigned char) op;
2692 STORE_NUMBER (loc + 1, arg);
2696 /* Like `store_op1', but for two two-byte parameters ARG1 and ARG2. */
2699 store_op2 (op, loc, arg1, arg2)
2704 *loc = (unsigned char) op;
2705 STORE_NUMBER (loc + 1, arg1);
2706 STORE_NUMBER (loc + 3, arg2);
2710 /* Copy the bytes from LOC to END to open up three bytes of space at LOC
2711 for OP followed by two-byte integer parameter ARG. */
2714 insert_op1 (op, loc, arg, end)
2720 register unsigned char *pfrom = end;
2721 register unsigned char *pto = end + 3;
2723 while (pfrom != loc)
2726 store_op1 (op, loc, arg);
2730 /* Like `insert_op1', but for two two-byte parameters ARG1 and ARG2. */
2733 insert_op2 (op, loc, arg1, arg2, end)
2739 register unsigned char *pfrom = end;
2740 register unsigned char *pto = end + 5;
2742 while (pfrom != loc)
2745 store_op2 (op, loc, arg1, arg2);
2749 /* P points to just after a ^ in PATTERN. Return true if that ^ comes
2750 after an alternative or a begin-subexpression. We assume there is at
2751 least one character before the ^. */
2754 at_begline_loc_p (pattern, p, syntax)
2755 const char *pattern, *p;
2756 reg_syntax_t syntax;
2758 const char *prev = p - 2;
2759 boolean prev_prev_backslash = prev > pattern && prev[-1] == '\\';
2762 /* After a subexpression? */
2763 (*prev == '(' && (syntax & RE_NO_BK_PARENS || prev_prev_backslash))
2764 /* After an alternative? */
2765 || (*prev == '|' && (syntax & RE_NO_BK_VBAR || prev_prev_backslash));
2769 /* The dual of at_begline_loc_p. This one is for $. We assume there is
2770 at least one character after the $, i.e., `P < PEND'. */
2773 at_endline_loc_p (p, pend, syntax)
2774 const char *p, *pend;
2777 const char *next = p;
2778 boolean next_backslash = *next == '\\';
2779 const char *next_next = p + 1 < pend ? p + 1 : 0;
2782 /* Before a subexpression? */
2783 (syntax & RE_NO_BK_PARENS ? *next == ')'
2784 : next_backslash && next_next && *next_next == ')')
2785 /* Before an alternative? */
2786 || (syntax & RE_NO_BK_VBAR ? *next == '|'
2787 : next_backslash && next_next && *next_next == '|');
2791 /* Returns true if REGNUM is in one of COMPILE_STACK's elements and
2792 false if it's not. */
2795 group_in_compile_stack (compile_stack, regnum)
2796 compile_stack_type compile_stack;
2801 for (this_element = compile_stack.avail - 1;
2804 if (compile_stack.stack[this_element].regnum == regnum)
2811 /* Read the ending character of a range (in a bracket expression) from the
2812 uncompiled pattern *P_PTR (which ends at PEND). We assume the
2813 starting character is in `P[-2]'. (`P[-1]' is the character `-'.)
2814 Then we set the translation of all bits between the starting and
2815 ending characters (inclusive) in the compiled pattern B.
2817 Return an error code.
2819 We use these short variable names so we can use the same macros as
2820 `regex_compile' itself. */
2822 static reg_errcode_t
2823 compile_range (p_ptr, pend, translate, syntax, b)
2824 const char **p_ptr, *pend;
2826 reg_syntax_t syntax;
2831 const char *p = *p_ptr;
2832 int range_start, range_end;
2837 /* Even though the pattern is a signed `char *', we need to fetch
2838 with unsigned char *'s; if the high bit of the pattern character
2839 is set, the range endpoints will be negative if we fetch using a
2842 We also want to fetch the endpoints without translating them; the
2843 appropriate translation is done in the bit-setting loop below. */
2844 /* The SVR4 compiler on the 3B2 had trouble with unsigned const char *. */
2845 range_start = ((const unsigned char *) p)[-2];
2846 range_end = ((const unsigned char *) p)[0];
2848 /* Have to increment the pointer into the pattern string, so the
2849 caller isn't still at the ending character. */
2852 /* If the start is after the end, the range is empty. */
2853 if (range_start > range_end)
2854 return syntax & RE_NO_EMPTY_RANGES ? REG_ERANGE : REG_NOERROR;
2856 /* Here we see why `this_char' has to be larger than an `unsigned
2857 char' -- the range is inclusive, so if `range_end' == 0xff
2858 (assuming 8-bit characters), we would otherwise go into an infinite
2859 loop, since all characters <= 0xff. */
2860 for (this_char = range_start; this_char <= range_end; this_char++)
2862 SET_LIST_BIT (TRANSLATE (this_char));
2868 /* re_compile_fastmap computes a ``fastmap'' for the compiled pattern in
2869 BUFP. A fastmap records which of the (1 << BYTEWIDTH) possible
2870 characters can start a string that matches the pattern. This fastmap
2871 is used by re_search to skip quickly over impossible starting points.
2873 The caller must supply the address of a (1 << BYTEWIDTH)-byte data
2874 area as BUFP->fastmap.
2876 We set the `fastmap', `fastmap_accurate', and `can_be_null' fields in
2879 Returns 0 if we succeed, -2 if an internal error. */
2882 re_compile_fastmap (bufp)
2883 struct re_pattern_buffer *bufp;
2886 #ifdef MATCH_MAY_ALLOCATE
2887 fail_stack_type fail_stack;
2889 #ifndef REGEX_MALLOC
2892 /* We don't push any register information onto the failure stack. */
2893 unsigned num_regs = 0;
2895 register char *fastmap = bufp->fastmap;
2896 unsigned char *pattern = bufp->buffer;
2897 unsigned long size = bufp->used;
2898 unsigned char *p = pattern;
2899 register unsigned char *pend = pattern + size;
2901 /* This holds the pointer to the failure stack, when
2902 it is allocated relocatably. */
2904 fail_stack_elt_t *failure_stack_ptr;
2907 /* Assume that each path through the pattern can be null until
2908 proven otherwise. We set this false at the bottom of switch
2909 statement, to which we get only if a particular path doesn't
2910 match the empty string. */
2911 boolean path_can_be_null = true;
2913 /* We aren't doing a `succeed_n' to begin with. */
2914 boolean succeed_n_p = false;
2916 assert (fastmap != NULL && p != NULL);
2919 bzero (fastmap, 1 << BYTEWIDTH); /* Assume nothing's valid. */
2920 bufp->fastmap_accurate = 1; /* It will be when we're done. */
2921 bufp->can_be_null = 0;
2925 if (p == pend || *p == succeed)
2927 /* We have reached the (effective) end of pattern. */
2928 if (!FAIL_STACK_EMPTY ())
2930 bufp->can_be_null |= path_can_be_null;
2932 /* Reset for next path. */
2933 path_can_be_null = true;
2935 p = fail_stack.stack[--fail_stack.avail].pointer;
2943 /* We should never be about to go beyond the end of the pattern. */
2946 switch (SWITCH_ENUM_CAST ((re_opcode_t) *p++))
2949 /* I guess the idea here is to simply not bother with a fastmap
2950 if a backreference is used, since it's too hard to figure out
2951 the fastmap for the corresponding group. Setting
2952 `can_be_null' stops `re_search_2' from using the fastmap, so
2953 that is all we do. */
2955 bufp->can_be_null = 1;
2959 /* Following are the cases which match a character. These end
2968 for (j = *p++ * BYTEWIDTH - 1; j >= 0; j--)
2969 if (p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH)))
2975 /* Chars beyond end of map must be allowed. */
2976 for (j = *p * BYTEWIDTH; j < (1 << BYTEWIDTH); j++)
2979 for (j = *p++ * BYTEWIDTH - 1; j >= 0; j--)
2980 if (!(p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH))))
2986 for (j = 0; j < (1 << BYTEWIDTH); j++)
2987 if (SYNTAX (j) == Sword)
2993 for (j = 0; j < (1 << BYTEWIDTH); j++)
2994 if (SYNTAX (j) != Sword)
3001 int fastmap_newline = fastmap['\n'];
3003 /* `.' matches anything ... */
3004 for (j = 0; j < (1 << BYTEWIDTH); j++)
3007 /* ... except perhaps newline. */
3008 if (!(bufp->syntax & RE_DOT_NEWLINE))
3009 fastmap['\n'] = fastmap_newline;
3011 /* Return if we have already set `can_be_null'; if we have,
3012 then the fastmap is irrelevant. Something's wrong here. */
3013 else if (bufp->can_be_null)
3016 /* Otherwise, have to check alternative paths. */
3023 for (j = 0; j < (1 << BYTEWIDTH); j++)
3024 if (SYNTAX (j) == (enum syntaxcode) k)
3031 for (j = 0; j < (1 << BYTEWIDTH); j++)
3032 if (SYNTAX (j) != (enum syntaxcode) k)
3037 /* All cases after this match the empty string. These end with
3057 case push_dummy_failure:
3062 case pop_failure_jump:
3063 case maybe_pop_jump:
3066 case dummy_failure_jump:
3067 EXTRACT_NUMBER_AND_INCR (j, p);
3072 /* Jump backward implies we just went through the body of a
3073 loop and matched nothing. Opcode jumped to should be
3074 `on_failure_jump' or `succeed_n'. Just treat it like an
3075 ordinary jump. For a * loop, it has pushed its failure
3076 point already; if so, discard that as redundant. */
3077 if ((re_opcode_t) *p != on_failure_jump
3078 && (re_opcode_t) *p != succeed_n)
3082 EXTRACT_NUMBER_AND_INCR (j, p);
3085 /* If what's on the stack is where we are now, pop it. */
3086 if (!FAIL_STACK_EMPTY ()
3087 && fail_stack.stack[fail_stack.avail - 1].pointer == p)
3093 case on_failure_jump:
3094 case on_failure_keep_string_jump:
3095 handle_on_failure_jump:
3096 EXTRACT_NUMBER_AND_INCR (j, p);
3098 /* For some patterns, e.g., `(a?)?', `p+j' here points to the
3099 end of the pattern. We don't want to push such a point,
3100 since when we restore it above, entering the switch will
3101 increment `p' past the end of the pattern. We don't need
3102 to push such a point since we obviously won't find any more
3103 fastmap entries beyond `pend'. Such a pattern can match
3104 the null string, though. */
3107 if (!PUSH_PATTERN_OP (p + j, fail_stack))
3109 RESET_FAIL_STACK ();
3114 bufp->can_be_null = 1;
3118 EXTRACT_NUMBER_AND_INCR (k, p); /* Skip the n. */
3119 succeed_n_p = false;
3126 /* Get to the number of times to succeed. */
3129 /* Increment p past the n for when k != 0. */
3130 EXTRACT_NUMBER_AND_INCR (k, p);
3134 succeed_n_p = true; /* Spaghetti code alert. */
3135 goto handle_on_failure_jump;
3152 abort (); /* We have listed all the cases. */
3155 /* Getting here means we have found the possible starting
3156 characters for one path of the pattern -- and that the empty
3157 string does not match. We need not follow this path further.
3158 Instead, look at the next alternative (remembered on the
3159 stack), or quit if no more. The test at the top of the loop
3160 does these things. */
3161 path_can_be_null = false;
3165 /* Set `can_be_null' for the last path (also the first path, if the
3166 pattern is empty). */
3167 bufp->can_be_null |= path_can_be_null;
3170 RESET_FAIL_STACK ();
3172 } /* re_compile_fastmap */
3174 /* Set REGS to hold NUM_REGS registers, storing them in STARTS and
3175 ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use
3176 this memory for recording register information. STARTS and ENDS
3177 must be allocated using the malloc library routine, and must each
3178 be at least NUM_REGS * sizeof (regoff_t) bytes long.
3180 If NUM_REGS == 0, then subsequent matches should allocate their own
3183 Unless this function is called, the first search or match using
3184 PATTERN_BUFFER will allocate its own register data, without
3185 freeing the old data. */
3188 re_set_registers (bufp, regs, num_regs, starts, ends)
3189 struct re_pattern_buffer *bufp;
3190 struct re_registers *regs;
3192 regoff_t *starts, *ends;
3196 bufp->regs_allocated = REGS_REALLOCATE;
3197 regs->num_regs = num_regs;
3198 regs->start = starts;
3203 bufp->regs_allocated = REGS_UNALLOCATED;
3205 regs->start = regs->end = (regoff_t *) 0;
3209 /* Searching routines. */
3211 /* Like re_search_2, below, but only one string is specified, and
3212 doesn't let you say where to stop matching. */
3215 re_search (bufp, string, size, startpos, range, regs)
3216 struct re_pattern_buffer *bufp;
3218 int size, startpos, range;
3219 struct re_registers *regs;
3221 return re_search_2 (bufp, NULL, 0, string, size, startpos, range,
3226 /* Using the compiled pattern in BUFP->buffer, first tries to match the
3227 virtual concatenation of STRING1 and STRING2, starting first at index
3228 STARTPOS, then at STARTPOS + 1, and so on.
3230 STRING1 and STRING2 have length SIZE1 and SIZE2, respectively.
3232 RANGE is how far to scan while trying to match. RANGE = 0 means try
3233 only at STARTPOS; in general, the last start tried is STARTPOS +
3236 In REGS, return the indices of the virtual concatenation of STRING1
3237 and STRING2 that matched the entire BUFP->buffer and its contained
3240 Do not consider matching one past the index STOP in the virtual
3241 concatenation of STRING1 and STRING2.
3243 We return either the position in the strings at which the match was
3244 found, -1 if no match, or -2 if error (such as failure
3248 re_search_2 (bufp, string1, size1, string2, size2, startpos, range, regs, stop)
3249 struct re_pattern_buffer *bufp;
3250 const char *string1, *string2;
3254 struct re_registers *regs;
3258 register char *fastmap = bufp->fastmap;
3259 register char *translate = bufp->translate;
3260 int total_size = size1 + size2;
3261 int endpos = startpos + range;
3263 /* Check for out-of-range STARTPOS. */
3264 if (startpos < 0 || startpos > total_size)
3267 /* Fix up RANGE if it might eventually take us outside
3268 the virtual concatenation of STRING1 and STRING2. */
3270 range = -1 - startpos;
3271 else if (endpos > total_size)
3272 range = total_size - startpos;
3274 /* If the search isn't to be a backwards one, don't waste time in a
3275 search for a pattern that must be anchored. */
3276 if (bufp->used > 0 && (re_opcode_t) bufp->buffer[0] == begbuf && range > 0)
3285 /* In a forward search for something that starts with \=.
3286 don't keep searching past point. */
3287 if (bufp->used > 0 && (re_opcode_t) bufp->buffer[0] == at_dot && range > 0)
3289 range = PT - startpos;
3295 /* Update the fastmap now if not correct already. */
3296 if (fastmap && !bufp->fastmap_accurate)
3297 if (re_compile_fastmap (bufp) == -2)
3300 /* Loop through the string, looking for a place to start matching. */
3303 /* If a fastmap is supplied, skip quickly over characters that
3304 cannot be the start of a match. If the pattern can match the
3305 null string, however, we don't need to skip characters; we want
3306 the first null string. */
3307 if (fastmap && startpos < total_size && !bufp->can_be_null)
3309 if (range > 0) /* Searching forwards. */
3311 register const char *d;
3312 register int lim = 0;
3315 if (startpos < size1 && startpos + range >= size1)
3316 lim = range - (size1 - startpos);
3318 d = (startpos >= size1 ? string2 - size1 : string1) + startpos;
3320 /* Written out as an if-else to avoid testing `translate'
3324 && !fastmap[(unsigned char)
3325 translate[(unsigned char) *d++]])
3328 while (range > lim && !fastmap[(unsigned char) *d++])
3331 startpos += irange - range;
3333 else /* Searching backwards. */
3335 register char c = (size1 == 0 || startpos >= size1
3336 ? string2[startpos - size1]
3337 : string1[startpos]);
3339 if (!fastmap[(unsigned char) TRANSLATE (c)])
3344 /* If can't match the null string, and that's all we have left, fail. */
3345 if (range >= 0 && startpos == total_size && fastmap
3346 && !bufp->can_be_null)
3349 val = re_match_2_internal (bufp, string1, size1, string2, size2,
3350 startpos, regs, stop);
3351 #ifndef REGEX_MALLOC
3380 /* Declarations and macros for re_match_2. */
3382 static int bcmp_translate ();
3383 static boolean alt_match_null_string_p (),
3384 common_op_match_null_string_p (),
3385 group_match_null_string_p ();
3387 /* This converts PTR, a pointer into one of the search strings `string1'
3388 and `string2' into an offset from the beginning of that string. */
3389 #define POINTER_TO_OFFSET(ptr) \
3390 (FIRST_STRING_P (ptr) \
3391 ? ((regoff_t) ((ptr) - string1)) \
3392 : ((regoff_t) ((ptr) - string2 + size1)))
3394 /* Macros for dealing with the split strings in re_match_2. */
3396 #define MATCHING_IN_FIRST_STRING (dend == end_match_1)
3398 /* Call before fetching a character with *d. This switches over to
3399 string2 if necessary. */
3400 #define PREFETCH() \
3403 /* End of string2 => fail. */ \
3404 if (dend == end_match_2) \
3406 /* End of string1 => advance to string2. */ \
3408 dend = end_match_2; \
3412 /* Test if at very beginning or at very end of the virtual concatenation
3413 of `string1' and `string2'. If only one string, it's `string2'. */
3414 #define AT_STRINGS_BEG(d) ((d) == (size1 ? string1 : string2) || !size2)
3415 #define AT_STRINGS_END(d) ((d) == end2)
3418 /* Test if D points to a character which is word-constituent. We have
3419 two special cases to check for: if past the end of string1, look at
3420 the first character in string2; and if before the beginning of
3421 string2, look at the last character in string1. */
3422 #define WORDCHAR_P(d) \
3423 (SYNTAX ((d) == end1 ? *string2 \
3424 : (d) == string2 - 1 ? *(end1 - 1) : *(d)) \
3427 /* Test if the character before D and the one at D differ with respect
3428 to being word-constituent. */
3429 #define AT_WORD_BOUNDARY(d) \
3430 (AT_STRINGS_BEG (d) || AT_STRINGS_END (d) \
3431 || WORDCHAR_P (d - 1) != WORDCHAR_P (d))
3434 /* Free everything we malloc. */
3435 #ifdef MATCH_MAY_ALLOCATE
3436 #define FREE_VAR(var) if (var) REGEX_FREE (var); var = NULL
3437 #define FREE_VARIABLES() \
3439 REGEX_FREE_STACK (fail_stack.stack); \
3440 FREE_VAR (regstart); \
3441 FREE_VAR (regend); \
3442 FREE_VAR (old_regstart); \
3443 FREE_VAR (old_regend); \
3444 FREE_VAR (best_regstart); \
3445 FREE_VAR (best_regend); \
3446 FREE_VAR (reg_info); \
3447 FREE_VAR (reg_dummy); \
3448 FREE_VAR (reg_info_dummy); \
3451 #define FREE_VARIABLES() ((void)0) /* Do nothing! But inhibit gcc warning. */
3452 #endif /* not MATCH_MAY_ALLOCATE */
3454 /* These values must meet several constraints. They must not be valid
3455 register values; since we have a limit of 255 registers (because
3456 we use only one byte in the pattern for the register number), we can
3457 use numbers larger than 255. They must differ by 1, because of
3458 NUM_FAILURE_ITEMS above. And the value for the lowest register must
3459 be larger than the value for the highest register, so we do not try
3460 to actually save any registers when none are active. */
3461 #define NO_HIGHEST_ACTIVE_REG (1 << BYTEWIDTH)
3462 #define NO_LOWEST_ACTIVE_REG (NO_HIGHEST_ACTIVE_REG + 1)
3464 /* Matching routines. */
3466 #ifndef emacs /* Emacs never uses this. */
3467 /* re_match is like re_match_2 except it takes only a single string. */
3470 re_match (bufp, string, size, pos, regs)
3471 struct re_pattern_buffer *bufp;
3474 struct re_registers *regs;
3476 int result = re_match_2_internal (bufp, NULL, 0, string, size,
3481 #endif /* not emacs */
3484 /* re_match_2 matches the compiled pattern in BUFP against the
3485 the (virtual) concatenation of STRING1 and STRING2 (of length SIZE1
3486 and SIZE2, respectively). We start matching at POS, and stop
3489 If REGS is non-null and the `no_sub' field of BUFP is nonzero, we
3490 store offsets for the substring each group matched in REGS. See the
3491 documentation for exactly how many groups we fill.
3493 We return -1 if no match, -2 if an internal error (such as the
3494 failure stack overflowing). Otherwise, we return the length of the
3495 matched substring. */
3498 re_match_2 (bufp, string1, size1, string2, size2, pos, regs, stop)
3499 struct re_pattern_buffer *bufp;
3500 const char *string1, *string2;
3503 struct re_registers *regs;
3506 int result = re_match_2_internal (bufp, string1, size1, string2, size2,
3512 /* This is a separate function so that we can force an alloca cleanup
3515 re_match_2_internal (bufp, string1, size1, string2, size2, pos, regs, stop)
3516 struct re_pattern_buffer *bufp;
3517 const char *string1, *string2;
3520 struct re_registers *regs;
3523 /* General temporaries. */
3527 /* Just past the end of the corresponding string. */
3528 const char *end1, *end2;
3530 /* Pointers into string1 and string2, just past the last characters in
3531 each to consider matching. */
3532 const char *end_match_1, *end_match_2;
3534 /* Where we are in the data, and the end of the current string. */
3535 const char *d, *dend;
3537 /* Where we are in the pattern, and the end of the pattern. */
3538 unsigned char *p = bufp->buffer;
3539 register unsigned char *pend = p + bufp->used;
3541 /* Mark the opcode just after a start_memory, so we can test for an
3542 empty subpattern when we get to the stop_memory. */
3543 unsigned char *just_past_start_mem = 0;
3545 /* We use this to map every character in the string. */
3546 char *translate = bufp->translate;
3548 /* Failure point stack. Each place that can handle a failure further
3549 down the line pushes a failure point on this stack. It consists of
3550 restart, regend, and reg_info for all registers corresponding to
3551 the subexpressions we're currently inside, plus the number of such
3552 registers, and, finally, two char *'s. The first char * is where
3553 to resume scanning the pattern; the second one is where to resume
3554 scanning the strings. If the latter is zero, the failure point is
3555 a ``dummy''; if a failure happens and the failure point is a dummy,
3556 it gets discarded and the next next one is tried. */
3557 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
3558 fail_stack_type fail_stack;
3561 static unsigned failure_id = 0;
3562 unsigned nfailure_points_pushed = 0, nfailure_points_popped = 0;
3565 /* This holds the pointer to the failure stack, when
3566 it is allocated relocatably. */
3568 fail_stack_elt_t *failure_stack_ptr;
3571 /* We fill all the registers internally, independent of what we
3572 return, for use in backreferences. The number here includes
3573 an element for register zero. */
3574 unsigned num_regs = bufp->re_nsub + 1;
3576 /* The currently active registers. */
3577 unsigned lowest_active_reg = NO_LOWEST_ACTIVE_REG;
3578 unsigned highest_active_reg = NO_HIGHEST_ACTIVE_REG;
3580 /* Information on the contents of registers. These are pointers into
3581 the input strings; they record just what was matched (on this
3582 attempt) by a subexpression part of the pattern, that is, the
3583 regnum-th regstart pointer points to where in the pattern we began
3584 matching and the regnum-th regend points to right after where we
3585 stopped matching the regnum-th subexpression. (The zeroth register
3586 keeps track of what the whole pattern matches.) */
3587 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
3588 const char **regstart, **regend;
3591 /* If a group that's operated upon by a repetition operator fails to
3592 match anything, then the register for its start will need to be
3593 restored because it will have been set to wherever in the string we
3594 are when we last see its open-group operator. Similarly for a
3596 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
3597 const char **old_regstart, **old_regend;
3600 /* The is_active field of reg_info helps us keep track of which (possibly
3601 nested) subexpressions we are currently in. The matched_something
3602 field of reg_info[reg_num] helps us tell whether or not we have
3603 matched any of the pattern so far this time through the reg_num-th
3604 subexpression. These two fields get reset each time through any
3605 loop their register is in. */
3606 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
3607 register_info_type *reg_info;
3610 /* The following record the register info as found in the above
3611 variables when we find a match better than any we've seen before.
3612 This happens as we backtrack through the failure points, which in
3613 turn happens only if we have not yet matched the entire string. */
3614 unsigned best_regs_set = false;
3615 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
3616 const char **best_regstart, **best_regend;
3619 /* Logically, this is `best_regend[0]'. But we don't want to have to
3620 allocate space for that if we're not allocating space for anything
3621 else (see below). Also, we never need info about register 0 for
3622 any of the other register vectors, and it seems rather a kludge to
3623 treat `best_regend' differently than the rest. So we keep track of
3624 the end of the best match so far in a separate variable. We
3625 initialize this to NULL so that when we backtrack the first time
3626 and need to test it, it's not garbage. */
3627 const char *match_end = NULL;
3629 /* This helps SET_REGS_MATCHED avoid doing redundant work. */
3630 int set_regs_matched_done = 0;
3632 /* Used when we pop values we don't care about. */
3633 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
3634 const char **reg_dummy;
3635 register_info_type *reg_info_dummy;
3639 /* Counts the total number of registers pushed. */
3640 unsigned num_regs_pushed = 0;
3643 DEBUG_PRINT1 ("\n\nEntering re_match_2.\n");
3647 #ifdef MATCH_MAY_ALLOCATE
3648 /* Do not bother to initialize all the register variables if there are
3649 no groups in the pattern, as it takes a fair amount of time. If
3650 there are groups, we include space for register 0 (the whole
3651 pattern), even though we never use it, since it simplifies the
3652 array indexing. We should fix this. */
3655 regstart = REGEX_TALLOC (num_regs, const char *);
3656 regend = REGEX_TALLOC (num_regs, const char *);
3657 old_regstart = REGEX_TALLOC (num_regs, const char *);
3658 old_regend = REGEX_TALLOC (num_regs, const char *);
3659 best_regstart = REGEX_TALLOC (num_regs, const char *);
3660 best_regend = REGEX_TALLOC (num_regs, const char *);
3661 reg_info = REGEX_TALLOC (num_regs, register_info_type);
3662 reg_dummy = REGEX_TALLOC (num_regs, const char *);
3663 reg_info_dummy = REGEX_TALLOC (num_regs, register_info_type);
3665 if (!(regstart && regend && old_regstart && old_regend && reg_info
3666 && best_regstart && best_regend && reg_dummy && reg_info_dummy))
3674 /* We must initialize all our variables to NULL, so that
3675 `FREE_VARIABLES' doesn't try to free them. */
3676 regstart = regend = old_regstart = old_regend = best_regstart
3677 = best_regend = reg_dummy = NULL;
3678 reg_info = reg_info_dummy = (register_info_type *) NULL;
3680 #endif /* MATCH_MAY_ALLOCATE */
3682 /* The starting position is bogus. */
3683 if (pos < 0 || pos > size1 + size2)
3689 /* Initialize subexpression text positions to -1 to mark ones that no
3690 start_memory/stop_memory has been seen for. Also initialize the
3691 register information struct. */
3692 for (mcnt = 1; mcnt < num_regs; mcnt++)
3694 regstart[mcnt] = regend[mcnt]
3695 = old_regstart[mcnt] = old_regend[mcnt] = REG_UNSET_VALUE;
3697 REG_MATCH_NULL_STRING_P (reg_info[mcnt]) = MATCH_NULL_UNSET_VALUE;
3698 IS_ACTIVE (reg_info[mcnt]) = 0;
3699 MATCHED_SOMETHING (reg_info[mcnt]) = 0;
3700 EVER_MATCHED_SOMETHING (reg_info[mcnt]) = 0;
3703 /* We move `string1' into `string2' if the latter's empty -- but not if
3704 `string1' is null. */
3705 if (size2 == 0 && string1 != NULL)
3712 end1 = string1 + size1;
3713 end2 = string2 + size2;
3715 /* Compute where to stop matching, within the two strings. */
3718 end_match_1 = string1 + stop;
3719 end_match_2 = string2;
3724 end_match_2 = string2 + stop - size1;
3727 /* `p' scans through the pattern as `d' scans through the data.
3728 `dend' is the end of the input string that `d' points within. `d'
3729 is advanced into the following input string whenever necessary, but
3730 this happens before fetching; therefore, at the beginning of the
3731 loop, `d' can be pointing at the end of a string, but it cannot
3733 if (size1 > 0 && pos <= size1)
3740 d = string2 + pos - size1;
3744 DEBUG_PRINT1 ("The compiled pattern is: ");
3745 DEBUG_PRINT_COMPILED_PATTERN (bufp, p, pend);
3746 DEBUG_PRINT1 ("The string to match is: `");
3747 DEBUG_PRINT_DOUBLE_STRING (d, string1, size1, string2, size2);
3748 DEBUG_PRINT1 ("'\n");
3750 /* This loops over pattern commands. It exits by returning from the
3751 function if the match is complete, or it drops through if the match
3752 fails at this starting point in the input data. */
3755 DEBUG_PRINT2 ("\n0x%x: ", p);
3758 { /* End of pattern means we might have succeeded. */
3759 DEBUG_PRINT1 ("end of pattern ... ");
3761 /* If we haven't matched the entire string, and we want the
3762 longest match, try backtracking. */
3763 if (d != end_match_2)
3765 /* 1 if this match ends in the same string (string1 or string2)
3766 as the best previous match. */
3767 boolean same_str_p = (FIRST_STRING_P (match_end)
3768 == MATCHING_IN_FIRST_STRING);
3769 /* 1 if this match is the best seen so far. */
3770 boolean best_match_p;
3772 /* AIX compiler got confused when this was combined
3773 with the previous declaration. */
3775 best_match_p = d > match_end;
3777 best_match_p = !MATCHING_IN_FIRST_STRING;
3779 DEBUG_PRINT1 ("backtracking.\n");
3781 if (!FAIL_STACK_EMPTY ())
3782 { /* More failure points to try. */
3784 /* If exceeds best match so far, save it. */
3785 if (!best_regs_set || best_match_p)
3787 best_regs_set = true;
3790 DEBUG_PRINT1 ("\nSAVING match as best so far.\n");
3792 for (mcnt = 1; mcnt < num_regs; mcnt++)
3794 best_regstart[mcnt] = regstart[mcnt];
3795 best_regend[mcnt] = regend[mcnt];
3801 /* If no failure points, don't restore garbage. And if
3802 last match is real best match, don't restore second
3804 else if (best_regs_set && !best_match_p)
3807 /* Restore best match. It may happen that `dend ==
3808 end_match_1' while the restored d is in string2.
3809 For example, the pattern `x.*y.*z' against the
3810 strings `x-' and `y-z-', if the two strings are
3811 not consecutive in memory. */
3812 DEBUG_PRINT1 ("Restoring best registers.\n");
3815 dend = ((d >= string1 && d <= end1)
3816 ? end_match_1 : end_match_2);
3818 for (mcnt = 1; mcnt < num_regs; mcnt++)
3820 regstart[mcnt] = best_regstart[mcnt];
3821 regend[mcnt] = best_regend[mcnt];
3824 } /* d != end_match_2 */
3827 DEBUG_PRINT1 ("Accepting match.\n");
3829 /* If caller wants register contents data back, do it. */
3830 if (regs && !bufp->no_sub)
3832 /* Have the register data arrays been allocated? */
3833 if (bufp->regs_allocated == REGS_UNALLOCATED)
3834 { /* No. So allocate them with malloc. We need one
3835 extra element beyond `num_regs' for the `-1' marker
3837 regs->num_regs = MAX (RE_NREGS, num_regs + 1);
3838 regs->start = TALLOC (regs->num_regs, regoff_t);
3839 regs->end = TALLOC (regs->num_regs, regoff_t);
3840 if (regs->start == NULL || regs->end == NULL)
3845 bufp->regs_allocated = REGS_REALLOCATE;
3847 else if (bufp->regs_allocated == REGS_REALLOCATE)
3848 { /* Yes. If we need more elements than were already
3849 allocated, reallocate them. If we need fewer, just
3851 if (regs->num_regs < num_regs + 1)
3853 regs->num_regs = num_regs + 1;
3854 RETALLOC (regs->start, regs->num_regs, regoff_t);
3855 RETALLOC (regs->end, regs->num_regs, regoff_t);
3856 if (regs->start == NULL || regs->end == NULL)
3865 /* These braces fend off a "empty body in an else-statement"
3866 warning under GCC when assert expands to nothing. */
3867 assert (bufp->regs_allocated == REGS_FIXED);
3870 /* Convert the pointer data in `regstart' and `regend' to
3871 indices. Register zero has to be set differently,
3872 since we haven't kept track of any info for it. */
3873 if (regs->num_regs > 0)
3875 regs->start[0] = pos;
3876 regs->end[0] = (MATCHING_IN_FIRST_STRING
3877 ? ((regoff_t) (d - string1))
3878 : ((regoff_t) (d - string2 + size1)));
3881 /* Go through the first `min (num_regs, regs->num_regs)'
3882 registers, since that is all we initialized. */
3883 for (mcnt = 1; mcnt < MIN (num_regs, regs->num_regs); mcnt++)
3885 if (REG_UNSET (regstart[mcnt]) || REG_UNSET (regend[mcnt]))
3886 regs->start[mcnt] = regs->end[mcnt] = -1;
3890 = (regoff_t) POINTER_TO_OFFSET (regstart[mcnt]);
3892 = (regoff_t) POINTER_TO_OFFSET (regend[mcnt]);
3896 /* If the regs structure we return has more elements than
3897 were in the pattern, set the extra elements to -1. If
3898 we (re)allocated the registers, this is the case,
3899 because we always allocate enough to have at least one
3901 for (mcnt = num_regs; mcnt < regs->num_regs; mcnt++)
3902 regs->start[mcnt] = regs->end[mcnt] = -1;
3903 } /* regs && !bufp->no_sub */
3905 DEBUG_PRINT4 ("%u failure points pushed, %u popped (%u remain).\n",
3906 nfailure_points_pushed, nfailure_points_popped,
3907 nfailure_points_pushed - nfailure_points_popped);
3908 DEBUG_PRINT2 ("%u registers pushed.\n", num_regs_pushed);
3910 mcnt = d - pos - (MATCHING_IN_FIRST_STRING
3914 DEBUG_PRINT2 ("Returning %d from re_match_2.\n", mcnt);
3920 /* Otherwise match next pattern command. */
3921 switch (SWITCH_ENUM_CAST ((re_opcode_t) *p++))
3923 /* Ignore these. Used to ignore the n of succeed_n's which
3924 currently have n == 0. */
3926 DEBUG_PRINT1 ("EXECUTING no_op.\n");
3930 DEBUG_PRINT1 ("EXECUTING succeed.\n");
3933 /* Match the next n pattern characters exactly. The following
3934 byte in the pattern defines n, and the n bytes after that
3935 are the characters to match. */
3938 DEBUG_PRINT2 ("EXECUTING exactn %d.\n", mcnt);
3940 /* This is written out as an if-else so we don't waste time
3941 testing `translate' inside the loop. */
3947 if (translate[(unsigned char) *d++] != (char) *p++)
3957 if (*d++ != (char) *p++) goto fail;
3961 SET_REGS_MATCHED ();
3965 /* Match any character except possibly a newline or a null. */
3967 DEBUG_PRINT1 ("EXECUTING anychar.\n");
3971 if ((!(bufp->syntax & RE_DOT_NEWLINE) && TRANSLATE (*d) == '\n')
3972 || (bufp->syntax & RE_DOT_NOT_NULL && TRANSLATE (*d) == '\000'))
3975 SET_REGS_MATCHED ();
3976 DEBUG_PRINT2 (" Matched `%d'.\n", *d);
3984 register unsigned char c;
3985 boolean not = (re_opcode_t) *(p - 1) == charset_not;
3987 DEBUG_PRINT2 ("EXECUTING charset%s.\n", not ? "_not" : "");
3990 c = TRANSLATE (*d); /* The character to match. */
3992 /* Cast to `unsigned' instead of `unsigned char' in case the
3993 bit list is a full 32 bytes long. */
3994 if (c < (unsigned) (*p * BYTEWIDTH)
3995 && p[1 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH)))
4000 if (!not) goto fail;
4002 SET_REGS_MATCHED ();
4008 /* The beginning of a group is represented by start_memory.
4009 The arguments are the register number in the next byte, and the
4010 number of groups inner to this one in the next. The text
4011 matched within the group is recorded (in the internal
4012 registers data structure) under the register number. */
4014 DEBUG_PRINT3 ("EXECUTING start_memory %d (%d):\n", *p, p[1]);
4016 /* Find out if this group can match the empty string. */
4017 p1 = p; /* To send to group_match_null_string_p. */
4019 if (REG_MATCH_NULL_STRING_P (reg_info[*p]) == MATCH_NULL_UNSET_VALUE)
4020 REG_MATCH_NULL_STRING_P (reg_info[*p])
4021 = group_match_null_string_p (&p1, pend, reg_info);
4023 /* Save the position in the string where we were the last time
4024 we were at this open-group operator in case the group is
4025 operated upon by a repetition operator, e.g., with `(a*)*b'
4026 against `ab'; then we want to ignore where we are now in
4027 the string in case this attempt to match fails. */
4028 old_regstart[*p] = REG_MATCH_NULL_STRING_P (reg_info[*p])
4029 ? REG_UNSET (regstart[*p]) ? d : regstart[*p]
4031 DEBUG_PRINT2 (" old_regstart: %d\n",
4032 POINTER_TO_OFFSET (old_regstart[*p]));
4035 DEBUG_PRINT2 (" regstart: %d\n", POINTER_TO_OFFSET (regstart[*p]));
4037 IS_ACTIVE (reg_info[*p]) = 1;
4038 MATCHED_SOMETHING (reg_info[*p]) = 0;
4040 /* Clear this whenever we change the register activity status. */
4041 set_regs_matched_done = 0;
4043 /* This is the new highest active register. */
4044 highest_active_reg = *p;
4046 /* If nothing was active before, this is the new lowest active
4048 if (lowest_active_reg == NO_LOWEST_ACTIVE_REG)
4049 lowest_active_reg = *p;
4051 /* Move past the register number and inner group count. */
4053 just_past_start_mem = p;
4058 /* The stop_memory opcode represents the end of a group. Its
4059 arguments are the same as start_memory's: the register
4060 number, and the number of inner groups. */
4062 DEBUG_PRINT3 ("EXECUTING stop_memory %d (%d):\n", *p, p[1]);
4064 /* We need to save the string position the last time we were at
4065 this close-group operator in case the group is operated
4066 upon by a repetition operator, e.g., with `((a*)*(b*)*)*'
4067 against `aba'; then we want to ignore where we are now in
4068 the string in case this attempt to match fails. */
4069 old_regend[*p] = REG_MATCH_NULL_STRING_P (reg_info[*p])
4070 ? REG_UNSET (regend[*p]) ? d : regend[*p]
4072 DEBUG_PRINT2 (" old_regend: %d\n",
4073 POINTER_TO_OFFSET (old_regend[*p]));
4076 DEBUG_PRINT2 (" regend: %d\n", POINTER_TO_OFFSET (regend[*p]));
4078 /* This register isn't active anymore. */
4079 IS_ACTIVE (reg_info[*p]) = 0;
4081 /* Clear this whenever we change the register activity status. */
4082 set_regs_matched_done = 0;
4084 /* If this was the only register active, nothing is active
4086 if (lowest_active_reg == highest_active_reg)
4088 lowest_active_reg = NO_LOWEST_ACTIVE_REG;
4089 highest_active_reg = NO_HIGHEST_ACTIVE_REG;
4092 { /* We must scan for the new highest active register, since
4093 it isn't necessarily one less than now: consider
4094 (a(b)c(d(e)f)g). When group 3 ends, after the f), the
4095 new highest active register is 1. */
4096 unsigned char r = *p - 1;
4097 while (r > 0 && !IS_ACTIVE (reg_info[r]))
4100 /* If we end up at register zero, that means that we saved
4101 the registers as the result of an `on_failure_jump', not
4102 a `start_memory', and we jumped to past the innermost
4103 `stop_memory'. For example, in ((.)*) we save
4104 registers 1 and 2 as a result of the *, but when we pop
4105 back to the second ), we are at the stop_memory 1.
4106 Thus, nothing is active. */
4109 lowest_active_reg = NO_LOWEST_ACTIVE_REG;
4110 highest_active_reg = NO_HIGHEST_ACTIVE_REG;
4113 highest_active_reg = r;
4116 /* If just failed to match something this time around with a
4117 group that's operated on by a repetition operator, try to
4118 force exit from the ``loop'', and restore the register
4119 information for this group that we had before trying this
4121 if ((!MATCHED_SOMETHING (reg_info[*p])
4122 || just_past_start_mem == p - 1)
4125 boolean is_a_jump_n = false;
4129 switch ((re_opcode_t) *p1++)
4133 case pop_failure_jump:
4134 case maybe_pop_jump:
4136 case dummy_failure_jump:
4137 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4147 /* If the next operation is a jump backwards in the pattern
4148 to an on_failure_jump right before the start_memory
4149 corresponding to this stop_memory, exit from the loop
4150 by forcing a failure after pushing on the stack the
4151 on_failure_jump's jump in the pattern, and d. */
4152 if (mcnt < 0 && (re_opcode_t) *p1 == on_failure_jump
4153 && (re_opcode_t) p1[3] == start_memory && p1[4] == *p)
4155 /* If this group ever matched anything, then restore
4156 what its registers were before trying this last
4157 failed match, e.g., with `(a*)*b' against `ab' for
4158 regstart[1], and, e.g., with `((a*)*(b*)*)*'
4159 against `aba' for regend[3].
4161 Also restore the registers for inner groups for,
4162 e.g., `((a*)(b*))*' against `aba' (register 3 would
4163 otherwise get trashed). */
4165 if (EVER_MATCHED_SOMETHING (reg_info[*p]))
4169 EVER_MATCHED_SOMETHING (reg_info[*p]) = 0;
4171 /* Restore this and inner groups' (if any) registers. */
4172 for (r = *p; r < *p + *(p + 1); r++)
4174 regstart[r] = old_regstart[r];
4176 /* xx why this test? */
4177 if (old_regend[r] >= regstart[r])
4178 regend[r] = old_regend[r];
4182 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4183 PUSH_FAILURE_POINT (p1 + mcnt, d, -2);
4189 /* Move past the register number and the inner group count. */
4194 /* \<digit> has been turned into a `duplicate' command which is
4195 followed by the numeric value of <digit> as the register number. */
4198 register const char *d2, *dend2;
4199 int regno = *p++; /* Get which register to match against. */
4200 DEBUG_PRINT2 ("EXECUTING duplicate %d.\n", regno);
4202 /* Can't back reference a group which we've never matched. */
4203 if (REG_UNSET (regstart[regno]) || REG_UNSET (regend[regno]))
4206 /* Where in input to try to start matching. */
4207 d2 = regstart[regno];
4209 /* Where to stop matching; if both the place to start and
4210 the place to stop matching are in the same string, then
4211 set to the place to stop, otherwise, for now have to use
4212 the end of the first string. */
4214 dend2 = ((FIRST_STRING_P (regstart[regno])
4215 == FIRST_STRING_P (regend[regno]))
4216 ? regend[regno] : end_match_1);
4219 /* If necessary, advance to next segment in register
4223 if (dend2 == end_match_2) break;
4224 if (dend2 == regend[regno]) break;
4226 /* End of string1 => advance to string2. */
4228 dend2 = regend[regno];
4230 /* At end of register contents => success */
4231 if (d2 == dend2) break;
4233 /* If necessary, advance to next segment in data. */
4236 /* How many characters left in this segment to match. */
4239 /* Want how many consecutive characters we can match in
4240 one shot, so, if necessary, adjust the count. */
4241 if (mcnt > dend2 - d2)
4244 /* Compare that many; failure if mismatch, else move
4247 ? bcmp_translate (d, d2, mcnt, translate)
4248 : bcmp (d, d2, mcnt))
4250 d += mcnt, d2 += mcnt;
4252 /* Do this because we've match some characters. */
4253 SET_REGS_MATCHED ();
4259 /* begline matches the empty string at the beginning of the string
4260 (unless `not_bol' is set in `bufp'), and, if
4261 `newline_anchor' is set, after newlines. */
4263 DEBUG_PRINT1 ("EXECUTING begline.\n");
4265 if (AT_STRINGS_BEG (d))
4267 if (!bufp->not_bol) break;
4269 else if (d[-1] == '\n' && bufp->newline_anchor)
4273 /* In all other cases, we fail. */
4277 /* endline is the dual of begline. */
4279 DEBUG_PRINT1 ("EXECUTING endline.\n");
4281 if (AT_STRINGS_END (d))
4283 if (!bufp->not_eol) break;
4286 /* We have to ``prefetch'' the next character. */
4287 else if ((d == end1 ? *string2 : *d) == '\n'
4288 && bufp->newline_anchor)
4295 /* Match at the very beginning of the data. */
4297 DEBUG_PRINT1 ("EXECUTING begbuf.\n");
4298 if (AT_STRINGS_BEG (d))
4303 /* Match at the very end of the data. */
4305 DEBUG_PRINT1 ("EXECUTING endbuf.\n");
4306 if (AT_STRINGS_END (d))
4311 /* on_failure_keep_string_jump is used to optimize `.*\n'. It
4312 pushes NULL as the value for the string on the stack. Then
4313 `pop_failure_point' will keep the current value for the
4314 string, instead of restoring it. To see why, consider
4315 matching `foo\nbar' against `.*\n'. The .* matches the foo;
4316 then the . fails against the \n. But the next thing we want
4317 to do is match the \n against the \n; if we restored the
4318 string value, we would be back at the foo.
4320 Because this is used only in specific cases, we don't need to
4321 check all the things that `on_failure_jump' does, to make
4322 sure the right things get saved on the stack. Hence we don't
4323 share its code. The only reason to push anything on the
4324 stack at all is that otherwise we would have to change
4325 `anychar's code to do something besides goto fail in this
4326 case; that seems worse than this. */
4327 case on_failure_keep_string_jump:
4328 DEBUG_PRINT1 ("EXECUTING on_failure_keep_string_jump");
4330 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4331 DEBUG_PRINT3 (" %d (to 0x%x):\n", mcnt, p + mcnt);
4333 PUSH_FAILURE_POINT (p + mcnt, NULL, -2);
4337 /* Uses of on_failure_jump:
4339 Each alternative starts with an on_failure_jump that points
4340 to the beginning of the next alternative. Each alternative
4341 except the last ends with a jump that in effect jumps past
4342 the rest of the alternatives. (They really jump to the
4343 ending jump of the following alternative, because tensioning
4344 these jumps is a hassle.)
4346 Repeats start with an on_failure_jump that points past both
4347 the repetition text and either the following jump or
4348 pop_failure_jump back to this on_failure_jump. */
4349 case on_failure_jump:
4351 DEBUG_PRINT1 ("EXECUTING on_failure_jump");
4353 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4354 DEBUG_PRINT3 (" %d (to 0x%x)", mcnt, p + mcnt);
4356 /* If this on_failure_jump comes right before a group (i.e.,
4357 the original * applied to a group), save the information
4358 for that group and all inner ones, so that if we fail back
4359 to this point, the group's information will be correct.
4360 For example, in \(a*\)*\1, we need the preceding group,
4361 and in \(\(a*\)b*\)\2, we need the inner group. */
4363 /* We can't use `p' to check ahead because we push
4364 a failure point to `p + mcnt' after we do this. */
4367 /* We need to skip no_op's before we look for the
4368 start_memory in case this on_failure_jump is happening as
4369 the result of a completed succeed_n, as in \(a\)\{1,3\}b\1
4371 while (p1 < pend && (re_opcode_t) *p1 == no_op)
4374 if (p1 < pend && (re_opcode_t) *p1 == start_memory)
4376 /* We have a new highest active register now. This will
4377 get reset at the start_memory we are about to get to,
4378 but we will have saved all the registers relevant to
4379 this repetition op, as described above. */
4380 highest_active_reg = *(p1 + 1) + *(p1 + 2);
4381 if (lowest_active_reg == NO_LOWEST_ACTIVE_REG)
4382 lowest_active_reg = *(p1 + 1);
4385 DEBUG_PRINT1 (":\n");
4386 PUSH_FAILURE_POINT (p + mcnt, d, -2);
4390 /* A smart repeat ends with `maybe_pop_jump'.
4391 We change it to either `pop_failure_jump' or `jump'. */
4392 case maybe_pop_jump:
4393 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4394 DEBUG_PRINT2 ("EXECUTING maybe_pop_jump %d.\n", mcnt);
4396 register unsigned char *p2 = p;
4398 /* Compare the beginning of the repeat with what in the
4399 pattern follows its end. If we can establish that there
4400 is nothing that they would both match, i.e., that we
4401 would have to backtrack because of (as in, e.g., `a*a')
4402 then we can change to pop_failure_jump, because we'll
4403 never have to backtrack.
4405 This is not true in the case of alternatives: in
4406 `(a|ab)*' we do need to backtrack to the `ab' alternative
4407 (e.g., if the string was `ab'). But instead of trying to
4408 detect that here, the alternative has put on a dummy
4409 failure point which is what we will end up popping. */
4411 /* Skip over open/close-group commands.
4412 If what follows this loop is a ...+ construct,
4413 look at what begins its body, since we will have to
4414 match at least one of that. */
4418 && ((re_opcode_t) *p2 == stop_memory
4419 || (re_opcode_t) *p2 == start_memory))
4421 else if (p2 + 6 < pend
4422 && (re_opcode_t) *p2 == dummy_failure_jump)
4429 /* p1[0] ... p1[2] are the `on_failure_jump' corresponding
4430 to the `maybe_finalize_jump' of this case. Examine what
4433 /* If we're at the end of the pattern, we can change. */
4436 /* Consider what happens when matching ":\(.*\)"
4437 against ":/". I don't really understand this code
4439 p[-3] = (unsigned char) pop_failure_jump;
4441 (" End of pattern: change to `pop_failure_jump'.\n");
4444 else if ((re_opcode_t) *p2 == exactn
4445 || (bufp->newline_anchor && (re_opcode_t) *p2 == endline))
4447 register unsigned char c
4448 = *p2 == (unsigned char) endline ? '\n' : p2[2];
4450 if ((re_opcode_t) p1[3] == exactn && p1[5] != c)
4452 p[-3] = (unsigned char) pop_failure_jump;
4453 DEBUG_PRINT3 (" %c != %c => pop_failure_jump.\n",
4457 else if ((re_opcode_t) p1[3] == charset
4458 || (re_opcode_t) p1[3] == charset_not)
4460 int not = (re_opcode_t) p1[3] == charset_not;
4462 if (c < (unsigned char) (p1[4] * BYTEWIDTH)
4463 && p1[5 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH)))
4466 /* `not' is equal to 1 if c would match, which means
4467 that we can't change to pop_failure_jump. */
4470 p[-3] = (unsigned char) pop_failure_jump;
4471 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
4475 else if ((re_opcode_t) *p2 == charset)
4478 register unsigned char c
4479 = *p2 == (unsigned char) endline ? '\n' : p2[2];
4482 if ((re_opcode_t) p1[3] == exactn
4483 && ! ((int) p2[1] * BYTEWIDTH > (int) p1[4]
4484 && (p2[1 + p1[4] / BYTEWIDTH]
4485 & (1 << (p1[4] % BYTEWIDTH)))))
4487 p[-3] = (unsigned char) pop_failure_jump;
4488 DEBUG_PRINT3 (" %c != %c => pop_failure_jump.\n",
4492 else if ((re_opcode_t) p1[3] == charset_not)
4495 /* We win if the charset_not inside the loop
4496 lists every character listed in the charset after. */
4497 for (idx = 0; idx < (int) p2[1]; idx++)
4498 if (! (p2[2 + idx] == 0
4499 || (idx < (int) p1[4]
4500 && ((p2[2 + idx] & ~ p1[5 + idx]) == 0))))
4505 p[-3] = (unsigned char) pop_failure_jump;
4506 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
4509 else if ((re_opcode_t) p1[3] == charset)
4512 /* We win if the charset inside the loop
4513 has no overlap with the one after the loop. */
4515 idx < (int) p2[1] && idx < (int) p1[4];
4517 if ((p2[2 + idx] & p1[5 + idx]) != 0)
4520 if (idx == p2[1] || idx == p1[4])
4522 p[-3] = (unsigned char) pop_failure_jump;
4523 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
4528 p -= 2; /* Point at relative address again. */
4529 if ((re_opcode_t) p[-1] != pop_failure_jump)
4531 p[-1] = (unsigned char) jump;
4532 DEBUG_PRINT1 (" Match => jump.\n");
4533 goto unconditional_jump;
4535 /* Note fall through. */
4538 /* The end of a simple repeat has a pop_failure_jump back to
4539 its matching on_failure_jump, where the latter will push a
4540 failure point. The pop_failure_jump takes off failure
4541 points put on by this pop_failure_jump's matching
4542 on_failure_jump; we got through the pattern to here from the
4543 matching on_failure_jump, so didn't fail. */
4544 case pop_failure_jump:
4546 /* We need to pass separate storage for the lowest and
4547 highest registers, even though we don't care about the
4548 actual values. Otherwise, we will restore only one
4549 register from the stack, since lowest will == highest in
4550 `pop_failure_point'. */
4551 unsigned dummy_low_reg, dummy_high_reg;
4552 unsigned char *pdummy;
4555 DEBUG_PRINT1 ("EXECUTING pop_failure_jump.\n");
4556 POP_FAILURE_POINT (sdummy, pdummy,
4557 dummy_low_reg, dummy_high_reg,
4558 reg_dummy, reg_dummy, reg_info_dummy);
4560 /* Note fall through. */
4563 /* Unconditionally jump (without popping any failure points). */
4566 EXTRACT_NUMBER_AND_INCR (mcnt, p); /* Get the amount to jump. */
4567 DEBUG_PRINT2 ("EXECUTING jump %d ", mcnt);
4568 p += mcnt; /* Do the jump. */
4569 DEBUG_PRINT2 ("(to 0x%x).\n", p);
4573 /* We need this opcode so we can detect where alternatives end
4574 in `group_match_null_string_p' et al. */
4576 DEBUG_PRINT1 ("EXECUTING jump_past_alt.\n");
4577 goto unconditional_jump;
4580 /* Normally, the on_failure_jump pushes a failure point, which
4581 then gets popped at pop_failure_jump. We will end up at
4582 pop_failure_jump, also, and with a pattern of, say, `a+', we
4583 are skipping over the on_failure_jump, so we have to push
4584 something meaningless for pop_failure_jump to pop. */
4585 case dummy_failure_jump:
4586 DEBUG_PRINT1 ("EXECUTING dummy_failure_jump.\n");
4587 /* It doesn't matter what we push for the string here. What
4588 the code at `fail' tests is the value for the pattern. */
4589 PUSH_FAILURE_POINT (0, 0, -2);
4590 goto unconditional_jump;
4593 /* At the end of an alternative, we need to push a dummy failure
4594 point in case we are followed by a `pop_failure_jump', because
4595 we don't want the failure point for the alternative to be
4596 popped. For example, matching `(a|ab)*' against `aab'
4597 requires that we match the `ab' alternative. */
4598 case push_dummy_failure:
4599 DEBUG_PRINT1 ("EXECUTING push_dummy_failure.\n");
4600 /* See comments just above at `dummy_failure_jump' about the
4602 PUSH_FAILURE_POINT (0, 0, -2);
4605 /* Have to succeed matching what follows at least n times.
4606 After that, handle like `on_failure_jump'. */
4608 EXTRACT_NUMBER (mcnt, p + 2);
4609 DEBUG_PRINT2 ("EXECUTING succeed_n %d.\n", mcnt);
4612 /* Originally, this is how many times we HAVE to succeed. */
4617 STORE_NUMBER_AND_INCR (p, mcnt);
4618 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p, mcnt);
4622 DEBUG_PRINT2 (" Setting two bytes from 0x%x to no_op.\n", p+2);
4623 p[2] = (unsigned char) no_op;
4624 p[3] = (unsigned char) no_op;
4630 EXTRACT_NUMBER (mcnt, p + 2);
4631 DEBUG_PRINT2 ("EXECUTING jump_n %d.\n", mcnt);
4633 /* Originally, this is how many times we CAN jump. */
4637 STORE_NUMBER (p + 2, mcnt);
4638 goto unconditional_jump;
4640 /* If don't have to jump any more, skip over the rest of command. */
4647 DEBUG_PRINT1 ("EXECUTING set_number_at.\n");
4649 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4651 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4652 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p1, mcnt);
4653 STORE_NUMBER (p1, mcnt);
4658 DEBUG_PRINT1 ("EXECUTING wordbound.\n");
4659 if (AT_WORD_BOUNDARY (d))
4664 DEBUG_PRINT1 ("EXECUTING notwordbound.\n");
4665 if (AT_WORD_BOUNDARY (d))
4670 DEBUG_PRINT1 ("EXECUTING wordbeg.\n");
4671 if (WORDCHAR_P (d) && (AT_STRINGS_BEG (d) || !WORDCHAR_P (d - 1)))
4676 DEBUG_PRINT1 ("EXECUTING wordend.\n");
4677 if (!AT_STRINGS_BEG (d) && WORDCHAR_P (d - 1)
4678 && (!WORDCHAR_P (d) || AT_STRINGS_END (d)))
4684 DEBUG_PRINT1 ("EXECUTING before_dot.\n");
4685 if (PTR_CHAR_POS ((unsigned char *) d) >= point)
4690 DEBUG_PRINT1 ("EXECUTING at_dot.\n");
4691 if (PTR_CHAR_POS ((unsigned char *) d) != point)
4696 DEBUG_PRINT1 ("EXECUTING after_dot.\n");
4697 if (PTR_CHAR_POS ((unsigned char *) d) <= point)
4702 DEBUG_PRINT2 ("EXECUTING syntaxspec %d.\n", mcnt);
4707 DEBUG_PRINT1 ("EXECUTING Emacs wordchar.\n");
4711 /* Can't use *d++ here; SYNTAX may be an unsafe macro. */
4713 if (SYNTAX (d[-1]) != (enum syntaxcode) mcnt)
4715 SET_REGS_MATCHED ();
4719 DEBUG_PRINT2 ("EXECUTING notsyntaxspec %d.\n", mcnt);
4721 goto matchnotsyntax;
4724 DEBUG_PRINT1 ("EXECUTING Emacs notwordchar.\n");
4728 /* Can't use *d++ here; SYNTAX may be an unsafe macro. */
4730 if (SYNTAX (d[-1]) == (enum syntaxcode) mcnt)
4732 SET_REGS_MATCHED ();
4735 #else /* not emacs */
4737 DEBUG_PRINT1 ("EXECUTING non-Emacs wordchar.\n");
4739 if (!WORDCHAR_P (d))
4741 SET_REGS_MATCHED ();
4746 DEBUG_PRINT1 ("EXECUTING non-Emacs notwordchar.\n");
4750 SET_REGS_MATCHED ();
4753 #endif /* not emacs */
4758 continue; /* Successfully executed one pattern command; keep going. */
4761 /* We goto here if a matching operation fails. */
4763 if (!FAIL_STACK_EMPTY ())
4764 { /* A restart point is known. Restore to that state. */
4765 DEBUG_PRINT1 ("\nFAIL:\n");
4766 POP_FAILURE_POINT (d, p,
4767 lowest_active_reg, highest_active_reg,
4768 regstart, regend, reg_info);
4770 /* If this failure point is a dummy, try the next one. */
4774 /* If we failed to the end of the pattern, don't examine *p. */
4778 boolean is_a_jump_n = false;
4780 /* If failed to a backwards jump that's part of a repetition
4781 loop, need to pop this failure point and use the next one. */
4782 switch ((re_opcode_t) *p)
4786 case maybe_pop_jump:
4787 case pop_failure_jump:
4790 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4793 if ((is_a_jump_n && (re_opcode_t) *p1 == succeed_n)
4795 && (re_opcode_t) *p1 == on_failure_jump))
4803 if (d >= string1 && d <= end1)
4807 break; /* Matching at this starting point really fails. */
4811 goto restore_best_regs;
4815 return -1; /* Failure to match. */
4818 /* Subroutine definitions for re_match_2. */
4821 /* We are passed P pointing to a register number after a start_memory.
4823 Return true if the pattern up to the corresponding stop_memory can
4824 match the empty string, and false otherwise.
4826 If we find the matching stop_memory, sets P to point to one past its number.
4827 Otherwise, sets P to an undefined byte less than or equal to END.
4829 We don't handle duplicates properly (yet). */
4832 group_match_null_string_p (p, end, reg_info)
4833 unsigned char **p, *end;
4834 register_info_type *reg_info;
4837 /* Point to after the args to the start_memory. */
4838 unsigned char *p1 = *p + 2;
4842 /* Skip over opcodes that can match nothing, and return true or
4843 false, as appropriate, when we get to one that can't, or to the
4844 matching stop_memory. */
4846 switch ((re_opcode_t) *p1)
4848 /* Could be either a loop or a series of alternatives. */
4849 case on_failure_jump:
4851 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4853 /* If the next operation is not a jump backwards in the
4858 /* Go through the on_failure_jumps of the alternatives,
4859 seeing if any of the alternatives cannot match nothing.
4860 The last alternative starts with only a jump,
4861 whereas the rest start with on_failure_jump and end
4862 with a jump, e.g., here is the pattern for `a|b|c':
4864 /on_failure_jump/0/6/exactn/1/a/jump_past_alt/0/6
4865 /on_failure_jump/0/6/exactn/1/b/jump_past_alt/0/3
4868 So, we have to first go through the first (n-1)
4869 alternatives and then deal with the last one separately. */
4872 /* Deal with the first (n-1) alternatives, which start
4873 with an on_failure_jump (see above) that jumps to right
4874 past a jump_past_alt. */
4876 while ((re_opcode_t) p1[mcnt-3] == jump_past_alt)
4878 /* `mcnt' holds how many bytes long the alternative
4879 is, including the ending `jump_past_alt' and
4882 if (!alt_match_null_string_p (p1, p1 + mcnt - 3,
4886 /* Move to right after this alternative, including the
4890 /* Break if it's the beginning of an n-th alternative
4891 that doesn't begin with an on_failure_jump. */
4892 if ((re_opcode_t) *p1 != on_failure_jump)
4895 /* Still have to check that it's not an n-th
4896 alternative that starts with an on_failure_jump. */
4898 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4899 if ((re_opcode_t) p1[mcnt-3] != jump_past_alt)
4901 /* Get to the beginning of the n-th alternative. */
4907 /* Deal with the last alternative: go back and get number
4908 of the `jump_past_alt' just before it. `mcnt' contains
4909 the length of the alternative. */
4910 EXTRACT_NUMBER (mcnt, p1 - 2);
4912 if (!alt_match_null_string_p (p1, p1 + mcnt, reg_info))
4915 p1 += mcnt; /* Get past the n-th alternative. */
4921 assert (p1[1] == **p);
4927 if (!common_op_match_null_string_p (&p1, end, reg_info))
4930 } /* while p1 < end */
4933 } /* group_match_null_string_p */
4936 /* Similar to group_match_null_string_p, but doesn't deal with alternatives:
4937 It expects P to be the first byte of a single alternative and END one
4938 byte past the last. The alternative can contain groups. */
4941 alt_match_null_string_p (p, end, reg_info)
4942 unsigned char *p, *end;
4943 register_info_type *reg_info;
4946 unsigned char *p1 = p;
4950 /* Skip over opcodes that can match nothing, and break when we get
4951 to one that can't. */
4953 switch ((re_opcode_t) *p1)
4956 case on_failure_jump:
4958 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4963 if (!common_op_match_null_string_p (&p1, end, reg_info))
4966 } /* while p1 < end */
4969 } /* alt_match_null_string_p */
4972 /* Deals with the ops common to group_match_null_string_p and
4973 alt_match_null_string_p.
4975 Sets P to one after the op and its arguments, if any. */
4978 common_op_match_null_string_p (p, end, reg_info)
4979 unsigned char **p, *end;
4980 register_info_type *reg_info;
4985 unsigned char *p1 = *p;
4987 switch ((re_opcode_t) *p1++)
5007 assert (reg_no > 0 && reg_no <= MAX_REGNUM);
5008 ret = group_match_null_string_p (&p1, end, reg_info);
5010 /* Have to set this here in case we're checking a group which
5011 contains a group and a back reference to it. */
5013 if (REG_MATCH_NULL_STRING_P (reg_info[reg_no]) == MATCH_NULL_UNSET_VALUE)
5014 REG_MATCH_NULL_STRING_P (reg_info[reg_no]) = ret;
5020 /* If this is an optimized succeed_n for zero times, make the jump. */
5022 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
5030 /* Get to the number of times to succeed. */
5032 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
5037 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
5045 if (!REG_MATCH_NULL_STRING_P (reg_info[*p1]))
5053 /* All other opcodes mean we cannot match the empty string. */
5059 } /* common_op_match_null_string_p */
5062 /* Return zero if TRANSLATE[S1] and TRANSLATE[S2] are identical for LEN
5063 bytes; nonzero otherwise. */
5066 bcmp_translate (s1, s2, len, translate)
5067 unsigned char *s1, *s2;
5071 register unsigned char *p1 = s1, *p2 = s2;
5074 if (translate[*p1++] != translate[*p2++]) return 1;
5080 /* Entry points for GNU code. */
5082 /* re_compile_pattern is the GNU regular expression compiler: it
5083 compiles PATTERN (of length SIZE) and puts the result in BUFP.
5084 Returns 0 if the pattern was valid, otherwise an error string.
5086 Assumes the `allocated' (and perhaps `buffer') and `translate' fields
5087 are set in BUFP on entry.
5089 We call regex_compile to do the actual compilation. */
5092 re_compile_pattern (pattern, length, bufp)
5093 const char *pattern;
5095 struct re_pattern_buffer *bufp;
5099 /* GNU code is written to assume at least RE_NREGS registers will be set
5100 (and at least one extra will be -1). */
5101 bufp->regs_allocated = REGS_UNALLOCATED;
5103 /* And GNU code determines whether or not to get register information
5104 by passing null for the REGS argument to re_match, etc., not by
5108 /* Match anchors at newline. */
5109 bufp->newline_anchor = 1;
5111 ret = regex_compile (pattern, length, re_syntax_options, bufp);
5115 return gettext (re_error_msgid[(int) ret]);
5118 /* Entry points compatible with 4.2 BSD regex library. We don't define
5119 them unless specifically requested. */
5121 #ifdef _REGEX_RE_COMP
5123 /* BSD has one and only one pattern buffer. */
5124 static struct re_pattern_buffer re_comp_buf;
5134 if (!re_comp_buf.buffer)
5135 return gettext ("No previous regular expression");
5139 if (!re_comp_buf.buffer)
5141 re_comp_buf.buffer = (unsigned char *) malloc (200);
5142 if (re_comp_buf.buffer == NULL)
5143 return gettext (re_error_msgid[(int) REG_ESPACE]);
5144 re_comp_buf.allocated = 200;
5146 re_comp_buf.fastmap = (char *) malloc (1 << BYTEWIDTH);
5147 if (re_comp_buf.fastmap == NULL)
5148 return gettext (re_error_msgid[(int) REG_ESPACE]);
5151 /* Since `re_exec' always passes NULL for the `regs' argument, we
5152 don't need to initialize the pattern buffer fields which affect it. */
5154 /* Match anchors at newlines. */
5155 re_comp_buf.newline_anchor = 1;
5157 ret = regex_compile (s, strlen (s), re_syntax_options, &re_comp_buf);
5162 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
5163 return (char *) gettext (re_error_msgid[(int) ret]);
5171 const int len = strlen (s);
5173 0 <= re_search (&re_comp_buf, s, len, 0, len, (struct re_registers *) 0);
5175 #endif /* _REGEX_RE_COMP */
5177 /* POSIX.2 functions. Don't define these for Emacs. */
5181 /* regcomp takes a regular expression as a string and compiles it.
5183 PREG is a regex_t *. We do not expect any fields to be initialized,
5184 since POSIX says we shouldn't. Thus, we set
5186 `buffer' to the compiled pattern;
5187 `used' to the length of the compiled pattern;
5188 `syntax' to RE_SYNTAX_POSIX_EXTENDED if the
5189 REG_EXTENDED bit in CFLAGS is set; otherwise, to
5190 RE_SYNTAX_POSIX_BASIC;
5191 `newline_anchor' to REG_NEWLINE being set in CFLAGS;
5192 `fastmap' and `fastmap_accurate' to zero;
5193 `re_nsub' to the number of subexpressions in PATTERN.
5195 PATTERN is the address of the pattern string.
5197 CFLAGS is a series of bits which affect compilation.
5199 If REG_EXTENDED is set, we use POSIX extended syntax; otherwise, we
5200 use POSIX basic syntax.
5202 If REG_NEWLINE is set, then . and [^...] don't match newline.
5203 Also, regexec will try a match beginning after every newline.
5205 If REG_ICASE is set, then we considers upper- and lowercase
5206 versions of letters to be equivalent when matching.
5208 If REG_NOSUB is set, then when PREG is passed to regexec, that
5209 routine will report only success or failure, and nothing about the
5212 It returns 0 if it succeeds, nonzero if it doesn't. (See regex.h for
5213 the return codes and their meanings.) */
5216 regcomp (preg, pattern, cflags)
5218 const char *pattern;
5223 = (cflags & REG_EXTENDED) ?
5224 RE_SYNTAX_POSIX_EXTENDED : RE_SYNTAX_POSIX_BASIC;
5226 /* regex_compile will allocate the space for the compiled pattern. */
5228 preg->allocated = 0;
5231 /* Don't bother to use a fastmap when searching. This simplifies the
5232 REG_NEWLINE case: if we used a fastmap, we'd have to put all the
5233 characters after newlines into the fastmap. This way, we just try
5237 if (cflags & REG_ICASE)
5241 preg->translate = (char *) malloc (CHAR_SET_SIZE);
5242 if (preg->translate == NULL)
5243 return (int) REG_ESPACE;
5245 /* Map uppercase characters to corresponding lowercase ones. */
5246 for (i = 0; i < CHAR_SET_SIZE; i++)
5247 preg->translate[i] = ISUPPER (i) ? tolower (i) : i;
5250 preg->translate = NULL;
5252 /* If REG_NEWLINE is set, newlines are treated differently. */
5253 if (cflags & REG_NEWLINE)
5254 { /* REG_NEWLINE implies neither . nor [^...] match newline. */
5255 syntax &= ~RE_DOT_NEWLINE;
5256 syntax |= RE_HAT_LISTS_NOT_NEWLINE;
5257 /* It also changes the matching behavior. */
5258 preg->newline_anchor = 1;
5261 preg->newline_anchor = 0;
5263 preg->no_sub = !!(cflags & REG_NOSUB);
5265 /* POSIX says a null character in the pattern terminates it, so we
5266 can use strlen here in compiling the pattern. */
5267 ret = regex_compile (pattern, strlen (pattern), syntax, preg);
5269 /* POSIX doesn't distinguish between an unmatched open-group and an
5270 unmatched close-group: both are REG_EPAREN. */
5271 if (ret == REG_ERPAREN) ret = REG_EPAREN;
5277 /* regexec searches for a given pattern, specified by PREG, in the
5280 If NMATCH is zero or REG_NOSUB was set in the cflags argument to
5281 `regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at
5282 least NMATCH elements, and we set them to the offsets of the
5283 corresponding matched substrings.
5285 EFLAGS specifies `execution flags' which affect matching: if
5286 REG_NOTBOL is set, then ^ does not match at the beginning of the
5287 string; if REG_NOTEOL is set, then $ does not match at the end.
5289 We return 0 if we find a match and REG_NOMATCH if not. */
5292 regexec (preg, string, nmatch, pmatch, eflags)
5293 const regex_t *preg;
5296 regmatch_t pmatch[];
5300 struct re_registers regs;
5301 regex_t private_preg;
5302 int len = strlen (string);
5303 boolean want_reg_info = !preg->no_sub && nmatch > 0;
5305 private_preg = *preg;
5307 private_preg.not_bol = !!(eflags & REG_NOTBOL);
5308 private_preg.not_eol = !!(eflags & REG_NOTEOL);
5310 /* The user has told us exactly how many registers to return
5311 information about, via `nmatch'. We have to pass that on to the
5312 matching routines. */
5313 private_preg.regs_allocated = REGS_FIXED;
5317 regs.num_regs = nmatch;
5318 regs.start = TALLOC (nmatch, regoff_t);
5319 regs.end = TALLOC (nmatch, regoff_t);
5320 if (regs.start == NULL || regs.end == NULL)
5321 return (int) REG_NOMATCH;
5324 /* Perform the searching operation. */
5325 ret = re_search (&private_preg, string, len,
5326 /* start: */ 0, /* range: */ len,
5327 want_reg_info ? ®s : (struct re_registers *) 0);
5329 /* Copy the register information to the POSIX structure. */
5336 for (r = 0; r < nmatch; r++)
5338 pmatch[r].rm_so = regs.start[r];
5339 pmatch[r].rm_eo = regs.end[r];
5343 /* If we needed the temporary register info, free the space now. */
5348 /* We want zero return to mean success, unlike `re_search'. */
5349 return ret >= 0 ? (int) REG_NOERROR : (int) REG_NOMATCH;
5353 /* Returns a message corresponding to an error code, ERRCODE, returned
5354 from either regcomp or regexec. We don't use PREG here. */
5357 regerror (errcode, preg, errbuf, errbuf_size)
5359 const regex_t *preg;
5367 || errcode >= (sizeof (re_error_msgid) / sizeof (re_error_msgid[0])))
5368 /* Only error codes returned by the rest of the code should be passed
5369 to this routine. If we are given anything else, or if other regex
5370 code generates an invalid error code, then the program has a bug.
5371 Dump core so we can fix it. */
5374 msg = gettext (re_error_msgid[errcode]);
5376 msg_size = strlen (msg) + 1; /* Includes the null. */
5378 if (errbuf_size != 0)
5380 if (msg_size > errbuf_size)
5382 strncpy (errbuf, msg, errbuf_size - 1);
5383 errbuf[errbuf_size - 1] = 0;
5386 strcpy (errbuf, msg);
5393 /* Free dynamically allocated space used by PREG. */
5399 if (preg->buffer != NULL)
5400 free (preg->buffer);
5401 preg->buffer = NULL;
5403 preg->allocated = 0;
5406 if (preg->fastmap != NULL)
5407 free (preg->fastmap);
5408 preg->fastmap = NULL;
5409 preg->fastmap_accurate = 0;
5411 if (preg->translate != NULL)
5412 free (preg->translate);
5413 preg->translate = NULL;
5416 #endif /* not emacs */
5420 make-backup-files: t
5422 trim-versions-without-asking: nil