1 /* Extended regular expression matching and search library, version
2 0.12. (Implements POSIX draft P10003.2/D11.2, except for
3 internationalization features.)
5 Copyright (C) 1993, 1994, 1995, 1996 Free Software Foundation, Inc.
7 This program is free software; you can redistribute it and/or modify
8 it under the terms of the GNU General Public License as published by
9 the Free Software Foundation; either version 2, or (at your option)
12 This program is distributed in the hope that it will be useful,
13 but WITHOUT ANY WARRANTY; without even the implied warranty of
14 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 GNU General Public License for more details.
17 You should have received a copy of the GNU General Public License
18 along with this program; if not, write to the Free Software
19 Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307,
22 /* AIX requires this to be the first thing in the file. */
23 #if defined (_AIX) && !defined (REGEX_MALLOC)
34 /* We need this for `regex.h', and perhaps for the Emacs include files. */
35 #include <sys/types.h>
37 /* This is for other GNU distributions with internationalized messages. */
38 #if HAVE_LIBINTL_H || defined (_LIBC)
41 # define gettext(msgid) (msgid)
45 /* This define is so xgettext can find the internationalizable
47 #define gettext_noop(String) String
50 /* The `emacs' switch turns on certain matching commands
51 that make sense only in Emacs. */
60 /* If we are not linking with Emacs proper,
61 we can't use the relocating allocator
62 even if config.h says that we can. */
65 #if defined (STDC_HEADERS) || defined (_LIBC)
72 /* When used in Emacs's lib-src, we need to get bzero and bcopy somehow.
73 If nothing else has been done, use the method below. */
74 #ifdef INHIBIT_STRING_HEADER
75 #if !(defined (HAVE_BZERO) && defined (HAVE_BCOPY))
76 #if !defined (bzero) && !defined (bcopy)
77 #undef INHIBIT_STRING_HEADER
82 /* This is the normal way of making sure we have a bcopy and a bzero.
83 This is used in most programs--a few other programs avoid this
84 by defining INHIBIT_STRING_HEADER. */
85 #ifndef INHIBIT_STRING_HEADER
86 #if defined (HAVE_STRING_H) || defined (STDC_HEADERS) || defined (_LIBC)
89 #define bcmp(s1, s2, n) memcmp ((s1), (s2), (n))
92 #define bcopy(s, d, n) memcpy ((d), (s), (n))
95 #define bzero(s, n) memset ((s), 0, (n))
102 /* Define the syntax stuff for \<, \>, etc. */
104 /* This must be nonzero for the wordchar and notwordchar pattern
105 commands in re_match_2. */
110 #ifdef SWITCH_ENUM_BUG
111 #define SWITCH_ENUM_CAST(x) ((int)(x))
113 #define SWITCH_ENUM_CAST(x) (x)
118 extern char *re_syntax_table;
120 #else /* not SYNTAX_TABLE */
122 /* How many characters in the character set. */
123 #define CHAR_SET_SIZE 256
125 static char re_syntax_table[CHAR_SET_SIZE];
136 bzero (re_syntax_table, sizeof re_syntax_table);
138 for (c = 'a'; c <= 'z'; c++)
139 re_syntax_table[c] = Sword;
141 for (c = 'A'; c <= 'Z'; c++)
142 re_syntax_table[c] = Sword;
144 for (c = '0'; c <= '9'; c++)
145 re_syntax_table[c] = Sword;
147 re_syntax_table['_'] = Sword;
152 #endif /* not SYNTAX_TABLE */
154 #define SYNTAX(c) re_syntax_table[c]
156 #endif /* not emacs */
158 /* Get the interface, including the syntax bits. */
161 /* isalpha etc. are used for the character classes. */
164 /* Jim Meyering writes:
166 "... Some ctype macros are valid only for character codes that
167 isascii says are ASCII (SGI's IRIX-4.0.5 is one such system --when
168 using /bin/cc or gcc but without giving an ansi option). So, all
169 ctype uses should be through macros like ISPRINT... If
170 STDC_HEADERS is defined, then autoconf has verified that the ctype
171 macros don't need to be guarded with references to isascii. ...
172 Defining isascii to 1 should let any compiler worth its salt
173 eliminate the && through constant folding." */
175 #if defined (STDC_HEADERS) || (!defined (isascii) && !defined (HAVE_ISASCII))
178 #define ISASCII(c) isascii(c)
182 #define ISBLANK(c) (ISASCII (c) && isblank (c))
184 #define ISBLANK(c) ((c) == ' ' || (c) == '\t')
187 #define ISGRAPH(c) (ISASCII (c) && isgraph (c))
189 #define ISGRAPH(c) (ISASCII (c) && isprint (c) && !isspace (c))
192 #define ISPRINT(c) (ISASCII (c) && isprint (c))
193 #define ISDIGIT(c) (ISASCII (c) && isdigit (c))
194 #define ISALNUM(c) (ISASCII (c) && isalnum (c))
195 #define ISALPHA(c) (ISASCII (c) && isalpha (c))
196 #define ISCNTRL(c) (ISASCII (c) && iscntrl (c))
197 #define ISLOWER(c) (ISASCII (c) && islower (c))
198 #define ISPUNCT(c) (ISASCII (c) && ispunct (c))
199 #define ISSPACE(c) (ISASCII (c) && isspace (c))
200 #define ISUPPER(c) (ISASCII (c) && isupper (c))
201 #define ISXDIGIT(c) (ISASCII (c) && isxdigit (c))
204 #define NULL (void *)0
207 /* We remove any previous definition of `SIGN_EXTEND_CHAR',
208 since ours (we hope) works properly with all combinations of
209 machines, compilers, `char' and `unsigned char' argument types.
210 (Per Bothner suggested the basic approach.) */
211 #undef SIGN_EXTEND_CHAR
213 #define SIGN_EXTEND_CHAR(c) ((signed char) (c))
214 #else /* not __STDC__ */
215 /* As in Harbison and Steele. */
216 #define SIGN_EXTEND_CHAR(c) ((((unsigned char) (c)) ^ 128) - 128)
219 /* Should we use malloc or alloca? If REGEX_MALLOC is not defined, we
220 use `alloca' instead of `malloc'. This is because using malloc in
221 re_search* or re_match* could cause memory leaks when C-g is used in
222 Emacs; also, malloc is slower and causes storage fragmentation. On
223 the other hand, malloc is more portable, and easier to debug.
225 Because we sometimes use alloca, some routines have to be macros,
226 not functions -- `alloca'-allocated space disappears at the end of the
227 function it is called in. */
231 #define REGEX_ALLOCATE malloc
232 #define REGEX_REALLOCATE(source, osize, nsize) realloc (source, nsize)
233 #define REGEX_FREE free
235 #else /* not REGEX_MALLOC */
237 /* Emacs already defines alloca, sometimes. */
240 /* Make alloca work the best possible way. */
242 #define alloca __builtin_alloca
243 #else /* not __GNUC__ */
246 #else /* not __GNUC__ or HAVE_ALLOCA_H */
247 #if 0 /* It is a bad idea to declare alloca. We always cast the result. */
248 #ifndef _AIX /* Already did AIX, up at the top. */
250 #endif /* not _AIX */
252 #endif /* not HAVE_ALLOCA_H */
253 #endif /* not __GNUC__ */
255 #endif /* not alloca */
257 #define REGEX_ALLOCATE alloca
259 /* Assumes a `char *destination' variable. */
260 #define REGEX_REALLOCATE(source, osize, nsize) \
261 (destination = (char *) alloca (nsize), \
262 bcopy (source, destination, osize), \
265 /* No need to do anything to free, after alloca. */
266 #define REGEX_FREE(arg) ((void)0) /* Do nothing! But inhibit gcc warning. */
268 #endif /* not REGEX_MALLOC */
270 /* Define how to allocate the failure stack. */
272 #if defined (REL_ALLOC) && defined (REGEX_MALLOC)
274 #define REGEX_ALLOCATE_STACK(size) \
275 r_alloc (&failure_stack_ptr, (size))
276 #define REGEX_REALLOCATE_STACK(source, osize, nsize) \
277 r_re_alloc (&failure_stack_ptr, (nsize))
278 #define REGEX_FREE_STACK(ptr) \
279 r_alloc_free (&failure_stack_ptr)
281 #else /* not using relocating allocator */
285 #define REGEX_ALLOCATE_STACK malloc
286 #define REGEX_REALLOCATE_STACK(source, osize, nsize) realloc (source, nsize)
287 #define REGEX_FREE_STACK free
289 #else /* not REGEX_MALLOC */
291 #define REGEX_ALLOCATE_STACK alloca
293 #define REGEX_REALLOCATE_STACK(source, osize, nsize) \
294 REGEX_REALLOCATE (source, osize, nsize)
295 /* No need to explicitly free anything. */
296 #define REGEX_FREE_STACK(arg)
298 #endif /* not REGEX_MALLOC */
299 #endif /* not using relocating allocator */
302 /* True if `size1' is non-NULL and PTR is pointing anywhere inside
303 `string1' or just past its end. This works if PTR is NULL, which is
305 #define FIRST_STRING_P(ptr) \
306 (size1 && string1 <= (ptr) && (ptr) <= string1 + size1)
308 /* (Re)Allocate N items of type T using malloc, or fail. */
309 #define TALLOC(n, t) ((t *) malloc ((n) * sizeof (t)))
310 #define RETALLOC(addr, n, t) ((addr) = (t *) realloc (addr, (n) * sizeof (t)))
311 #define RETALLOC_IF(addr, n, t) \
312 if (addr) RETALLOC((addr), (n), t); else (addr) = TALLOC ((n), t)
313 #define REGEX_TALLOC(n, t) ((t *) REGEX_ALLOCATE ((n) * sizeof (t)))
315 #define BYTEWIDTH 8 /* In bits. */
317 #define STREQ(s1, s2) ((strcmp (s1, s2) == 0))
321 #define MAX(a, b) ((a) > (b) ? (a) : (b))
322 #define MIN(a, b) ((a) < (b) ? (a) : (b))
324 typedef char boolean;
328 static int re_match_2_internal ();
330 /* These are the command codes that appear in compiled regular
331 expressions. Some opcodes are followed by argument bytes. A
332 command code can specify any interpretation whatsoever for its
333 arguments. Zero bytes may appear in the compiled regular expression. */
339 /* Succeed right away--no more backtracking. */
342 /* Followed by one byte giving n, then by n literal bytes. */
345 /* Matches any (more or less) character. */
348 /* Matches any one char belonging to specified set. First
349 following byte is number of bitmap bytes. Then come bytes
350 for a bitmap saying which chars are in. Bits in each byte
351 are ordered low-bit-first. A character is in the set if its
352 bit is 1. A character too large to have a bit in the map is
353 automatically not in the set. */
356 /* Same parameters as charset, but match any character that is
357 not one of those specified. */
360 /* Start remembering the text that is matched, for storing in a
361 register. Followed by one byte with the register number, in
362 the range 0 to one less than the pattern buffer's re_nsub
363 field. Then followed by one byte with the number of groups
364 inner to this one. (This last has to be part of the
365 start_memory only because we need it in the on_failure_jump
369 /* Stop remembering the text that is matched and store it in a
370 memory register. Followed by one byte with the register
371 number, in the range 0 to one less than `re_nsub' in the
372 pattern buffer, and one byte with the number of inner groups,
373 just like `start_memory'. (We need the number of inner
374 groups here because we don't have any easy way of finding the
375 corresponding start_memory when we're at a stop_memory.) */
378 /* Match a duplicate of something remembered. Followed by one
379 byte containing the register number. */
382 /* Fail unless at beginning of line. */
385 /* Fail unless at end of line. */
388 /* Succeeds if at beginning of buffer (if emacs) or at beginning
389 of string to be matched (if not). */
392 /* Analogously, for end of buffer/string. */
395 /* Followed by two byte relative address to which to jump. */
398 /* Same as jump, but marks the end of an alternative. */
401 /* Followed by two-byte relative address of place to resume at
402 in case of failure. */
405 /* Like on_failure_jump, but pushes a placeholder instead of the
406 current string position when executed. */
407 on_failure_keep_string_jump,
409 /* Throw away latest failure point and then jump to following
410 two-byte relative address. */
413 /* Change to pop_failure_jump if know won't have to backtrack to
414 match; otherwise change to jump. This is used to jump
415 back to the beginning of a repeat. If what follows this jump
416 clearly won't match what the repeat does, such that we can be
417 sure that there is no use backtracking out of repetitions
418 already matched, then we change it to a pop_failure_jump.
419 Followed by two-byte address. */
422 /* Jump to following two-byte address, and push a dummy failure
423 point. This failure point will be thrown away if an attempt
424 is made to use it for a failure. A `+' construct makes this
425 before the first repeat. Also used as an intermediary kind
426 of jump when compiling an alternative. */
429 /* Push a dummy failure point and continue. Used at the end of
433 /* Followed by two-byte relative address and two-byte number n.
434 After matching N times, jump to the address upon failure. */
437 /* Followed by two-byte relative address, and two-byte number n.
438 Jump to the address N times, then fail. */
441 /* Set the following two-byte relative address to the
442 subsequent two-byte number. The address *includes* the two
446 wordchar, /* Matches any word-constituent character. */
447 notwordchar, /* Matches any char that is not a word-constituent. */
449 wordbeg, /* Succeeds if at word beginning. */
450 wordend, /* Succeeds if at word end. */
452 wordbound, /* Succeeds if at a word boundary. */
453 notwordbound /* Succeeds if not at a word boundary. */
456 ,before_dot, /* Succeeds if before point. */
457 at_dot, /* Succeeds if at point. */
458 after_dot, /* Succeeds if after point. */
460 /* Matches any character whose syntax is specified. Followed by
461 a byte which contains a syntax code, e.g., Sword. */
464 /* Matches any character whose syntax is not that specified. */
469 /* Common operations on the compiled pattern. */
471 /* Store NUMBER in two contiguous bytes starting at DESTINATION. */
473 #define STORE_NUMBER(destination, number) \
475 (destination)[0] = (number) & 0377; \
476 (destination)[1] = (number) >> 8; \
479 /* Same as STORE_NUMBER, except increment DESTINATION to
480 the byte after where the number is stored. Therefore, DESTINATION
481 must be an lvalue. */
483 #define STORE_NUMBER_AND_INCR(destination, number) \
485 STORE_NUMBER (destination, number); \
486 (destination) += 2; \
489 /* Put into DESTINATION a number stored in two contiguous bytes starting
492 #define EXTRACT_NUMBER(destination, source) \
494 (destination) = *(source) & 0377; \
495 (destination) += SIGN_EXTEND_CHAR (*((source) + 1)) << 8; \
500 extract_number (dest, source)
502 unsigned char *source;
504 int temp = SIGN_EXTEND_CHAR (*(source + 1));
505 *dest = *source & 0377;
509 #ifndef EXTRACT_MACROS /* To debug the macros. */
510 #undef EXTRACT_NUMBER
511 #define EXTRACT_NUMBER(dest, src) extract_number (&dest, src)
512 #endif /* not EXTRACT_MACROS */
516 /* Same as EXTRACT_NUMBER, except increment SOURCE to after the number.
517 SOURCE must be an lvalue. */
519 #define EXTRACT_NUMBER_AND_INCR(destination, source) \
521 EXTRACT_NUMBER (destination, source); \
527 extract_number_and_incr (destination, source)
529 unsigned char **source;
531 extract_number (destination, *source);
535 #ifndef EXTRACT_MACROS
536 #undef EXTRACT_NUMBER_AND_INCR
537 #define EXTRACT_NUMBER_AND_INCR(dest, src) \
538 extract_number_and_incr (&dest, &src)
539 #endif /* not EXTRACT_MACROS */
543 /* If DEBUG is defined, Regex prints many voluminous messages about what
544 it is doing (if the variable `debug' is nonzero). If linked with the
545 main program in `iregex.c', you can enter patterns and strings
546 interactively. And if linked with the main program in `main.c' and
547 the other test files, you can run the already-written tests. */
551 /* We use standard I/O for debugging. */
554 /* It is useful to test things that ``must'' be true when debugging. */
557 static int debug = 0;
559 #define DEBUG_STATEMENT(e) e
560 #define DEBUG_PRINT1(x) if (debug) printf (x)
561 #define DEBUG_PRINT2(x1, x2) if (debug) printf (x1, x2)
562 #define DEBUG_PRINT3(x1, x2, x3) if (debug) printf (x1, x2, x3)
563 #define DEBUG_PRINT4(x1, x2, x3, x4) if (debug) printf (x1, x2, x3, x4)
564 #define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) \
565 if (debug) print_partial_compiled_pattern (s, e)
566 #define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) \
567 if (debug) print_double_string (w, s1, sz1, s2, sz2)
570 /* Print the fastmap in human-readable form. */
573 print_fastmap (fastmap)
576 unsigned was_a_range = 0;
579 while (i < (1 << BYTEWIDTH))
585 while (i < (1 << BYTEWIDTH) && fastmap[i])
601 /* Print a compiled pattern string in human-readable form, starting at
602 the START pointer into it and ending just before the pointer END. */
605 print_partial_compiled_pattern (start, end)
606 unsigned char *start;
610 unsigned char *p = start;
611 unsigned char *pend = end;
619 /* Loop over pattern commands. */
622 printf ("%d:\t", p - start);
624 switch ((re_opcode_t) *p++)
632 printf ("/exactn/%d", mcnt);
643 printf ("/start_memory/%d/%d", mcnt, *p++);
648 printf ("/stop_memory/%d/%d", mcnt, *p++);
652 printf ("/duplicate/%d", *p++);
662 register int c, last = -100;
663 register int in_range = 0;
665 printf ("/charset [%s",
666 (re_opcode_t) *(p - 1) == charset_not ? "^" : "");
668 assert (p + *p < pend);
670 for (c = 0; c < 256; c++)
672 && (p[1 + (c/8)] & (1 << (c % 8))))
674 /* Are we starting a range? */
675 if (last + 1 == c && ! in_range)
680 /* Have we broken a range? */
681 else if (last + 1 != c && in_range)
710 case on_failure_jump:
711 extract_number_and_incr (&mcnt, &p);
712 printf ("/on_failure_jump to %d", p + mcnt - start);
715 case on_failure_keep_string_jump:
716 extract_number_and_incr (&mcnt, &p);
717 printf ("/on_failure_keep_string_jump to %d", p + mcnt - start);
720 case dummy_failure_jump:
721 extract_number_and_incr (&mcnt, &p);
722 printf ("/dummy_failure_jump to %d", p + mcnt - start);
725 case push_dummy_failure:
726 printf ("/push_dummy_failure");
730 extract_number_and_incr (&mcnt, &p);
731 printf ("/maybe_pop_jump to %d", p + mcnt - start);
734 case pop_failure_jump:
735 extract_number_and_incr (&mcnt, &p);
736 printf ("/pop_failure_jump to %d", p + mcnt - start);
740 extract_number_and_incr (&mcnt, &p);
741 printf ("/jump_past_alt to %d", p + mcnt - start);
745 extract_number_and_incr (&mcnt, &p);
746 printf ("/jump to %d", p + mcnt - start);
750 extract_number_and_incr (&mcnt, &p);
751 extract_number_and_incr (&mcnt2, &p);
752 printf ("/succeed_n to %d, %d times", p + mcnt - start, mcnt2);
756 extract_number_and_incr (&mcnt, &p);
757 extract_number_and_incr (&mcnt2, &p);
758 printf ("/jump_n to %d, %d times", p + mcnt - start, mcnt2);
762 extract_number_and_incr (&mcnt, &p);
763 extract_number_and_incr (&mcnt2, &p);
764 printf ("/set_number_at location %d to %d", p + mcnt - start, mcnt2);
768 printf ("/wordbound");
772 printf ("/notwordbound");
784 printf ("/before_dot");
792 printf ("/after_dot");
796 printf ("/syntaxspec");
798 printf ("/%d", mcnt);
802 printf ("/notsyntaxspec");
804 printf ("/%d", mcnt);
809 printf ("/wordchar");
813 printf ("/notwordchar");
825 printf ("?%d", *(p-1));
831 printf ("%d:\tend of pattern.\n", p - start);
836 print_compiled_pattern (bufp)
837 struct re_pattern_buffer *bufp;
839 unsigned char *buffer = bufp->buffer;
841 print_partial_compiled_pattern (buffer, buffer + bufp->used);
842 printf ("%d bytes used/%d bytes allocated.\n", bufp->used, bufp->allocated);
844 if (bufp->fastmap_accurate && bufp->fastmap)
846 printf ("fastmap: ");
847 print_fastmap (bufp->fastmap);
850 printf ("re_nsub: %d\t", bufp->re_nsub);
851 printf ("regs_alloc: %d\t", bufp->regs_allocated);
852 printf ("can_be_null: %d\t", bufp->can_be_null);
853 printf ("newline_anchor: %d\n", bufp->newline_anchor);
854 printf ("no_sub: %d\t", bufp->no_sub);
855 printf ("not_bol: %d\t", bufp->not_bol);
856 printf ("not_eol: %d\t", bufp->not_eol);
857 printf ("syntax: %d\n", bufp->syntax);
858 /* Perhaps we should print the translate table? */
863 print_double_string (where, string1, size1, string2, size2)
876 if (FIRST_STRING_P (where))
878 for (this_char = where - string1; this_char < size1; this_char++)
879 putchar (string1[this_char]);
884 for (this_char = where - string2; this_char < size2; this_char++)
885 putchar (string2[this_char]);
889 #else /* not DEBUG */
894 #define DEBUG_STATEMENT(e)
895 #define DEBUG_PRINT1(x)
896 #define DEBUG_PRINT2(x1, x2)
897 #define DEBUG_PRINT3(x1, x2, x3)
898 #define DEBUG_PRINT4(x1, x2, x3, x4)
899 #define DEBUG_PRINT_COMPILED_PATTERN(p, s, e)
900 #define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2)
902 #endif /* not DEBUG */
904 /* Set by `re_set_syntax' to the current regexp syntax to recognize. Can
905 also be assigned to arbitrarily: each pattern buffer stores its own
906 syntax, so it can be changed between regex compilations. */
907 /* This has no initializer because initialized variables in Emacs
908 become read-only after dumping. */
909 reg_syntax_t re_syntax_options;
912 /* Specify the precise syntax of regexps for compilation. This provides
913 for compatibility for various utilities which historically have
914 different, incompatible syntaxes.
916 The argument SYNTAX is a bit mask comprised of the various bits
917 defined in regex.h. We return the old syntax. */
920 re_set_syntax (syntax)
923 reg_syntax_t ret = re_syntax_options;
925 re_syntax_options = syntax;
929 /* This table gives an error message for each of the error codes listed
930 in regex.h. Obviously the order here has to be same as there.
931 POSIX doesn't require that we do anything for REG_NOERROR,
932 but why not be nice? */
934 static const char *re_error_msgid[] =
936 gettext_noop ("Success"), /* REG_NOERROR */
937 gettext_noop ("No match"), /* REG_NOMATCH */
938 gettext_noop ("Invalid regular expression"), /* REG_BADPAT */
939 gettext_noop ("Invalid collation character"), /* REG_ECOLLATE */
940 gettext_noop ("Invalid character class name"), /* REG_ECTYPE */
941 gettext_noop ("Trailing backslash"), /* REG_EESCAPE */
942 gettext_noop ("Invalid back reference"), /* REG_ESUBREG */
943 gettext_noop ("Unmatched [ or [^"), /* REG_EBRACK */
944 gettext_noop ("Unmatched ( or \\("), /* REG_EPAREN */
945 gettext_noop ("Unmatched \\{"), /* REG_EBRACE */
946 gettext_noop ("Invalid content of \\{\\}"), /* REG_BADBR */
947 gettext_noop ("Invalid range end"), /* REG_ERANGE */
948 gettext_noop ("Memory exhausted"), /* REG_ESPACE */
949 gettext_noop ("Invalid preceding regular expression"), /* REG_BADRPT */
950 gettext_noop ("Premature end of regular expression"), /* REG_EEND */
951 gettext_noop ("Regular expression too big"), /* REG_ESIZE */
952 gettext_noop ("Unmatched ) or \\)"), /* REG_ERPAREN */
955 /* Avoiding alloca during matching, to placate r_alloc. */
957 /* Define MATCH_MAY_ALLOCATE unless we need to make sure that the
958 searching and matching functions should not call alloca. On some
959 systems, alloca is implemented in terms of malloc, and if we're
960 using the relocating allocator routines, then malloc could cause a
961 relocation, which might (if the strings being searched are in the
962 ralloc heap) shift the data out from underneath the regexp
965 Here's another reason to avoid allocation: Emacs
966 processes input from X in a signal handler; processing X input may
967 call malloc; if input arrives while a matching routine is calling
968 malloc, then we're scrod. But Emacs can't just block input while
969 calling matching routines; then we don't notice interrupts when
970 they come in. So, Emacs blocks input around all regexp calls
971 except the matching calls, which it leaves unprotected, in the
972 faith that they will not malloc. */
974 /* Normally, this is fine. */
975 #define MATCH_MAY_ALLOCATE
977 /* When using GNU C, we are not REALLY using the C alloca, no matter
978 what config.h may say. So don't take precautions for it. */
983 /* The match routines may not allocate if (1) they would do it with malloc
984 and (2) it's not safe for them to use malloc.
985 Note that if REL_ALLOC is defined, matching would not use malloc for the
986 failure stack, but we would still use it for the register vectors;
987 so REL_ALLOC should not affect this. */
988 #if (defined (C_ALLOCA) || defined (REGEX_MALLOC)) && defined (emacs)
989 #undef MATCH_MAY_ALLOCATE
993 /* Failure stack declarations and macros; both re_compile_fastmap and
994 re_match_2 use a failure stack. These have to be macros because of
995 REGEX_ALLOCATE_STACK. */
998 /* Number of failure points for which to initially allocate space
999 when matching. If this number is exceeded, we allocate more
1000 space, so it is not a hard limit. */
1001 #ifndef INIT_FAILURE_ALLOC
1002 #define INIT_FAILURE_ALLOC 5
1005 /* Roughly the maximum number of failure points on the stack. Would be
1006 exactly that if always used MAX_FAILURE_ITEMS items each time we failed.
1007 This is a variable only so users of regex can assign to it; we never
1008 change it ourselves. */
1009 #if defined (MATCH_MAY_ALLOCATE)
1010 /* 4400 was enough to cause a crash on Alpha OSF/1,
1011 whose default stack limit is 2mb. */
1012 int re_max_failures = 20000;
1014 int re_max_failures = 2000;
1017 union fail_stack_elt
1019 unsigned char *pointer;
1023 typedef union fail_stack_elt fail_stack_elt_t;
1027 fail_stack_elt_t *stack;
1029 unsigned avail; /* Offset of next open position. */
1032 #define FAIL_STACK_EMPTY() (fail_stack.avail == 0)
1033 #define FAIL_STACK_PTR_EMPTY() (fail_stack_ptr->avail == 0)
1034 #define FAIL_STACK_FULL() (fail_stack.avail == fail_stack.size)
1037 /* Define macros to initialize and free the failure stack.
1038 Do `return -2' if the alloc fails. */
1040 #ifdef MATCH_MAY_ALLOCATE
1041 #define INIT_FAIL_STACK() \
1043 fail_stack.stack = (fail_stack_elt_t *) \
1044 REGEX_ALLOCATE_STACK (INIT_FAILURE_ALLOC * sizeof (fail_stack_elt_t)); \
1046 if (fail_stack.stack == NULL) \
1049 fail_stack.size = INIT_FAILURE_ALLOC; \
1050 fail_stack.avail = 0; \
1053 #define RESET_FAIL_STACK() REGEX_FREE_STACK (fail_stack.stack)
1055 #define INIT_FAIL_STACK() \
1057 fail_stack.avail = 0; \
1060 #define RESET_FAIL_STACK()
1064 /* Double the size of FAIL_STACK, up to approximately `re_max_failures' items.
1066 Return 1 if succeeds, and 0 if either ran out of memory
1067 allocating space for it or it was already too large.
1069 REGEX_REALLOCATE_STACK requires `destination' be declared. */
1071 #define DOUBLE_FAIL_STACK(fail_stack) \
1072 ((fail_stack).size > re_max_failures * MAX_FAILURE_ITEMS \
1074 : ((fail_stack).stack = (fail_stack_elt_t *) \
1075 REGEX_REALLOCATE_STACK ((fail_stack).stack, \
1076 (fail_stack).size * sizeof (fail_stack_elt_t), \
1077 ((fail_stack).size << 1) * sizeof (fail_stack_elt_t)), \
1079 (fail_stack).stack == NULL \
1081 : ((fail_stack).size <<= 1, \
1085 /* Push pointer POINTER on FAIL_STACK.
1086 Return 1 if was able to do so and 0 if ran out of memory allocating
1088 #define PUSH_PATTERN_OP(POINTER, FAIL_STACK) \
1089 ((FAIL_STACK_FULL () \
1090 && !DOUBLE_FAIL_STACK (FAIL_STACK)) \
1092 : ((FAIL_STACK).stack[(FAIL_STACK).avail++].pointer = POINTER, \
1095 /* Push a pointer value onto the failure stack.
1096 Assumes the variable `fail_stack'. Probably should only
1097 be called from within `PUSH_FAILURE_POINT'. */
1098 #define PUSH_FAILURE_POINTER(item) \
1099 fail_stack.stack[fail_stack.avail++].pointer = (unsigned char *) (item)
1101 /* This pushes an integer-valued item onto the failure stack.
1102 Assumes the variable `fail_stack'. Probably should only
1103 be called from within `PUSH_FAILURE_POINT'. */
1104 #define PUSH_FAILURE_INT(item) \
1105 fail_stack.stack[fail_stack.avail++].integer = (item)
1107 /* Push a fail_stack_elt_t value onto the failure stack.
1108 Assumes the variable `fail_stack'. Probably should only
1109 be called from within `PUSH_FAILURE_POINT'. */
1110 #define PUSH_FAILURE_ELT(item) \
1111 fail_stack.stack[fail_stack.avail++] = (item)
1113 /* These three POP... operations complement the three PUSH... operations.
1114 All assume that `fail_stack' is nonempty. */
1115 #define POP_FAILURE_POINTER() fail_stack.stack[--fail_stack.avail].pointer
1116 #define POP_FAILURE_INT() fail_stack.stack[--fail_stack.avail].integer
1117 #define POP_FAILURE_ELT() fail_stack.stack[--fail_stack.avail]
1119 /* Used to omit pushing failure point id's when we're not debugging. */
1121 #define DEBUG_PUSH PUSH_FAILURE_INT
1122 #define DEBUG_POP(item_addr) *(item_addr) = POP_FAILURE_INT ()
1124 #define DEBUG_PUSH(item)
1125 #define DEBUG_POP(item_addr)
1129 /* Push the information about the state we will need
1130 if we ever fail back to it.
1132 Requires variables fail_stack, regstart, regend, reg_info, and
1133 num_regs be declared. DOUBLE_FAIL_STACK requires `destination' be
1136 Does `return FAILURE_CODE' if runs out of memory. */
1138 #define PUSH_FAILURE_POINT(pattern_place, string_place, failure_code) \
1140 char *destination; \
1141 /* Must be int, so when we don't save any registers, the arithmetic \
1142 of 0 + -1 isn't done as unsigned. */ \
1145 DEBUG_STATEMENT (failure_id++); \
1146 DEBUG_STATEMENT (nfailure_points_pushed++); \
1147 DEBUG_PRINT2 ("\nPUSH_FAILURE_POINT #%u:\n", failure_id); \
1148 DEBUG_PRINT2 (" Before push, next avail: %d\n", (fail_stack).avail);\
1149 DEBUG_PRINT2 (" size: %d\n", (fail_stack).size);\
1151 DEBUG_PRINT2 (" slots needed: %d\n", NUM_FAILURE_ITEMS); \
1152 DEBUG_PRINT2 (" available: %d\n", REMAINING_AVAIL_SLOTS); \
1154 /* Ensure we have enough space allocated for what we will push. */ \
1155 while (REMAINING_AVAIL_SLOTS < NUM_FAILURE_ITEMS) \
1157 if (!DOUBLE_FAIL_STACK (fail_stack)) \
1158 return failure_code; \
1160 DEBUG_PRINT2 ("\n Doubled stack; size now: %d\n", \
1161 (fail_stack).size); \
1162 DEBUG_PRINT2 (" slots available: %d\n", REMAINING_AVAIL_SLOTS);\
1165 /* Push the info, starting with the registers. */ \
1166 DEBUG_PRINT1 ("\n"); \
1169 for (this_reg = lowest_active_reg; this_reg <= highest_active_reg; \
1172 DEBUG_PRINT2 (" Pushing reg: %d\n", this_reg); \
1173 DEBUG_STATEMENT (num_regs_pushed++); \
1175 DEBUG_PRINT2 (" start: 0x%x\n", regstart[this_reg]); \
1176 PUSH_FAILURE_POINTER (regstart[this_reg]); \
1178 DEBUG_PRINT2 (" end: 0x%x\n", regend[this_reg]); \
1179 PUSH_FAILURE_POINTER (regend[this_reg]); \
1181 DEBUG_PRINT2 (" info: 0x%x\n ", reg_info[this_reg]); \
1182 DEBUG_PRINT2 (" match_null=%d", \
1183 REG_MATCH_NULL_STRING_P (reg_info[this_reg])); \
1184 DEBUG_PRINT2 (" active=%d", IS_ACTIVE (reg_info[this_reg])); \
1185 DEBUG_PRINT2 (" matched_something=%d", \
1186 MATCHED_SOMETHING (reg_info[this_reg])); \
1187 DEBUG_PRINT2 (" ever_matched=%d", \
1188 EVER_MATCHED_SOMETHING (reg_info[this_reg])); \
1189 DEBUG_PRINT1 ("\n"); \
1190 PUSH_FAILURE_ELT (reg_info[this_reg].word); \
1193 DEBUG_PRINT2 (" Pushing low active reg: %d\n", lowest_active_reg);\
1194 PUSH_FAILURE_INT (lowest_active_reg); \
1196 DEBUG_PRINT2 (" Pushing high active reg: %d\n", highest_active_reg);\
1197 PUSH_FAILURE_INT (highest_active_reg); \
1199 DEBUG_PRINT2 (" Pushing pattern 0x%x: ", pattern_place); \
1200 DEBUG_PRINT_COMPILED_PATTERN (bufp, pattern_place, pend); \
1201 PUSH_FAILURE_POINTER (pattern_place); \
1203 DEBUG_PRINT2 (" Pushing string 0x%x: `", string_place); \
1204 DEBUG_PRINT_DOUBLE_STRING (string_place, string1, size1, string2, \
1206 DEBUG_PRINT1 ("'\n"); \
1207 PUSH_FAILURE_POINTER (string_place); \
1209 DEBUG_PRINT2 (" Pushing failure id: %u\n", failure_id); \
1210 DEBUG_PUSH (failure_id); \
1213 /* This is the number of items that are pushed and popped on the stack
1214 for each register. */
1215 #define NUM_REG_ITEMS 3
1217 /* Individual items aside from the registers. */
1219 #define NUM_NONREG_ITEMS 5 /* Includes failure point id. */
1221 #define NUM_NONREG_ITEMS 4
1224 /* We push at most this many items on the stack. */
1225 /* We used to use (num_regs - 1), which is the number of registers
1226 this regexp will save; but that was changed to 5
1227 to avoid stack overflow for a regexp with lots of parens. */
1228 #define MAX_FAILURE_ITEMS (5 * NUM_REG_ITEMS + NUM_NONREG_ITEMS)
1230 /* We actually push this many items. */
1231 #define NUM_FAILURE_ITEMS \
1233 ? 0 : highest_active_reg - lowest_active_reg + 1) \
1237 /* How many items can still be added to the stack without overflowing it. */
1238 #define REMAINING_AVAIL_SLOTS ((fail_stack).size - (fail_stack).avail)
1241 /* Pops what PUSH_FAIL_STACK pushes.
1243 We restore into the parameters, all of which should be lvalues:
1244 STR -- the saved data position.
1245 PAT -- the saved pattern position.
1246 LOW_REG, HIGH_REG -- the highest and lowest active registers.
1247 REGSTART, REGEND -- arrays of string positions.
1248 REG_INFO -- array of information about each subexpression.
1250 Also assumes the variables `fail_stack' and (if debugging), `bufp',
1251 `pend', `string1', `size1', `string2', and `size2'. */
1253 #define POP_FAILURE_POINT(str, pat, low_reg, high_reg, regstart, regend, reg_info)\
1255 DEBUG_STATEMENT (fail_stack_elt_t failure_id;) \
1257 const unsigned char *string_temp; \
1259 assert (!FAIL_STACK_EMPTY ()); \
1261 /* Remove failure points and point to how many regs pushed. */ \
1262 DEBUG_PRINT1 ("POP_FAILURE_POINT:\n"); \
1263 DEBUG_PRINT2 (" Before pop, next avail: %d\n", fail_stack.avail); \
1264 DEBUG_PRINT2 (" size: %d\n", fail_stack.size); \
1266 assert (fail_stack.avail >= NUM_NONREG_ITEMS); \
1268 DEBUG_POP (&failure_id); \
1269 DEBUG_PRINT2 (" Popping failure id: %u\n", failure_id); \
1271 /* If the saved string location is NULL, it came from an \
1272 on_failure_keep_string_jump opcode, and we want to throw away the \
1273 saved NULL, thus retaining our current position in the string. */ \
1274 string_temp = POP_FAILURE_POINTER (); \
1275 if (string_temp != NULL) \
1276 str = (const char *) string_temp; \
1278 DEBUG_PRINT2 (" Popping string 0x%x: `", str); \
1279 DEBUG_PRINT_DOUBLE_STRING (str, string1, size1, string2, size2); \
1280 DEBUG_PRINT1 ("'\n"); \
1282 pat = (unsigned char *) POP_FAILURE_POINTER (); \
1283 DEBUG_PRINT2 (" Popping pattern 0x%x: ", pat); \
1284 DEBUG_PRINT_COMPILED_PATTERN (bufp, pat, pend); \
1286 /* Restore register info. */ \
1287 high_reg = (unsigned) POP_FAILURE_INT (); \
1288 DEBUG_PRINT2 (" Popping high active reg: %d\n", high_reg); \
1290 low_reg = (unsigned) POP_FAILURE_INT (); \
1291 DEBUG_PRINT2 (" Popping low active reg: %d\n", low_reg); \
1294 for (this_reg = high_reg; this_reg >= low_reg; this_reg--) \
1296 DEBUG_PRINT2 (" Popping reg: %d\n", this_reg); \
1298 reg_info[this_reg].word = POP_FAILURE_ELT (); \
1299 DEBUG_PRINT2 (" info: 0x%x\n", reg_info[this_reg]); \
1301 regend[this_reg] = (const char *) POP_FAILURE_POINTER (); \
1302 DEBUG_PRINT2 (" end: 0x%x\n", regend[this_reg]); \
1304 regstart[this_reg] = (const char *) POP_FAILURE_POINTER (); \
1305 DEBUG_PRINT2 (" start: 0x%x\n", regstart[this_reg]); \
1309 for (this_reg = highest_active_reg; this_reg > high_reg; this_reg--) \
1311 reg_info[this_reg].word.integer = 0; \
1312 regend[this_reg] = 0; \
1313 regstart[this_reg] = 0; \
1315 highest_active_reg = high_reg; \
1318 set_regs_matched_done = 0; \
1319 DEBUG_STATEMENT (nfailure_points_popped++); \
1320 } /* POP_FAILURE_POINT */
1324 /* Structure for per-register (a.k.a. per-group) information.
1325 Other register information, such as the
1326 starting and ending positions (which are addresses), and the list of
1327 inner groups (which is a bits list) are maintained in separate
1330 We are making a (strictly speaking) nonportable assumption here: that
1331 the compiler will pack our bit fields into something that fits into
1332 the type of `word', i.e., is something that fits into one item on the
1337 fail_stack_elt_t word;
1340 /* This field is one if this group can match the empty string,
1341 zero if not. If not yet determined, `MATCH_NULL_UNSET_VALUE'. */
1342 #define MATCH_NULL_UNSET_VALUE 3
1343 unsigned match_null_string_p : 2;
1344 unsigned is_active : 1;
1345 unsigned matched_something : 1;
1346 unsigned ever_matched_something : 1;
1348 } register_info_type;
1350 #define REG_MATCH_NULL_STRING_P(R) ((R).bits.match_null_string_p)
1351 #define IS_ACTIVE(R) ((R).bits.is_active)
1352 #define MATCHED_SOMETHING(R) ((R).bits.matched_something)
1353 #define EVER_MATCHED_SOMETHING(R) ((R).bits.ever_matched_something)
1356 /* Call this when have matched a real character; it sets `matched' flags
1357 for the subexpressions which we are currently inside. Also records
1358 that those subexprs have matched. */
1359 #define SET_REGS_MATCHED() \
1362 if (!set_regs_matched_done) \
1365 set_regs_matched_done = 1; \
1366 for (r = lowest_active_reg; r <= highest_active_reg; r++) \
1368 MATCHED_SOMETHING (reg_info[r]) \
1369 = EVER_MATCHED_SOMETHING (reg_info[r]) \
1376 /* Registers are set to a sentinel when they haven't yet matched. */
1377 static char reg_unset_dummy;
1378 #define REG_UNSET_VALUE (®_unset_dummy)
1379 #define REG_UNSET(e) ((e) == REG_UNSET_VALUE)
1381 /* Subroutine declarations and macros for regex_compile. */
1383 static void store_op1 (), store_op2 ();
1384 static void insert_op1 (), insert_op2 ();
1385 static boolean at_begline_loc_p (), at_endline_loc_p ();
1386 static boolean group_in_compile_stack ();
1387 static reg_errcode_t compile_range ();
1389 /* Fetch the next character in the uncompiled pattern---translating it
1390 if necessary. Also cast from a signed character in the constant
1391 string passed to us by the user to an unsigned char that we can use
1392 as an array index (in, e.g., `translate'). */
1394 #define PATFETCH(c) \
1395 do {if (p == pend) return REG_EEND; \
1396 c = (unsigned char) *p++; \
1397 if (translate) c = (unsigned char) translate[c]; \
1401 /* Fetch the next character in the uncompiled pattern, with no
1403 #define PATFETCH_RAW(c) \
1404 do {if (p == pend) return REG_EEND; \
1405 c = (unsigned char) *p++; \
1408 /* Go backwards one character in the pattern. */
1409 #define PATUNFETCH p--
1412 /* If `translate' is non-null, return translate[D], else just D. We
1413 cast the subscript to translate because some data is declared as
1414 `char *', to avoid warnings when a string constant is passed. But
1415 when we use a character as a subscript we must make it unsigned. */
1417 #define TRANSLATE(d) \
1418 (translate ? (char) translate[(unsigned char) (d)] : (d))
1422 /* Macros for outputting the compiled pattern into `buffer'. */
1424 /* If the buffer isn't allocated when it comes in, use this. */
1425 #define INIT_BUF_SIZE 32
1427 /* Make sure we have at least N more bytes of space in buffer. */
1428 #define GET_BUFFER_SPACE(n) \
1429 while (b - bufp->buffer + (n) > bufp->allocated) \
1432 /* Make sure we have one more byte of buffer space and then add C to it. */
1433 #define BUF_PUSH(c) \
1435 GET_BUFFER_SPACE (1); \
1436 *b++ = (unsigned char) (c); \
1440 /* Ensure we have two more bytes of buffer space and then append C1 and C2. */
1441 #define BUF_PUSH_2(c1, c2) \
1443 GET_BUFFER_SPACE (2); \
1444 *b++ = (unsigned char) (c1); \
1445 *b++ = (unsigned char) (c2); \
1449 /* As with BUF_PUSH_2, except for three bytes. */
1450 #define BUF_PUSH_3(c1, c2, c3) \
1452 GET_BUFFER_SPACE (3); \
1453 *b++ = (unsigned char) (c1); \
1454 *b++ = (unsigned char) (c2); \
1455 *b++ = (unsigned char) (c3); \
1459 /* Store a jump with opcode OP at LOC to location TO. We store a
1460 relative address offset by the three bytes the jump itself occupies. */
1461 #define STORE_JUMP(op, loc, to) \
1462 store_op1 (op, loc, (to) - (loc) - 3)
1464 /* Likewise, for a two-argument jump. */
1465 #define STORE_JUMP2(op, loc, to, arg) \
1466 store_op2 (op, loc, (to) - (loc) - 3, arg)
1468 /* Like `STORE_JUMP', but for inserting. Assume `b' is the buffer end. */
1469 #define INSERT_JUMP(op, loc, to) \
1470 insert_op1 (op, loc, (to) - (loc) - 3, b)
1472 /* Like `STORE_JUMP2', but for inserting. Assume `b' is the buffer end. */
1473 #define INSERT_JUMP2(op, loc, to, arg) \
1474 insert_op2 (op, loc, (to) - (loc) - 3, arg, b)
1477 /* This is not an arbitrary limit: the arguments which represent offsets
1478 into the pattern are two bytes long. So if 2^16 bytes turns out to
1479 be too small, many things would have to change. */
1480 #define MAX_BUF_SIZE (1L << 16)
1483 /* Extend the buffer by twice its current size via realloc and
1484 reset the pointers that pointed into the old block to point to the
1485 correct places in the new one. If extending the buffer results in it
1486 being larger than MAX_BUF_SIZE, then flag memory exhausted. */
1487 #define EXTEND_BUFFER() \
1489 unsigned char *old_buffer = bufp->buffer; \
1490 if (bufp->allocated == MAX_BUF_SIZE) \
1492 bufp->allocated <<= 1; \
1493 if (bufp->allocated > MAX_BUF_SIZE) \
1494 bufp->allocated = MAX_BUF_SIZE; \
1495 bufp->buffer = (unsigned char *) realloc (bufp->buffer, bufp->allocated);\
1496 if (bufp->buffer == NULL) \
1497 return REG_ESPACE; \
1498 /* If the buffer moved, move all the pointers into it. */ \
1499 if (old_buffer != bufp->buffer) \
1501 b = (b - old_buffer) + bufp->buffer; \
1502 begalt = (begalt - old_buffer) + bufp->buffer; \
1503 if (fixup_alt_jump) \
1504 fixup_alt_jump = (fixup_alt_jump - old_buffer) + bufp->buffer;\
1506 laststart = (laststart - old_buffer) + bufp->buffer; \
1507 if (pending_exact) \
1508 pending_exact = (pending_exact - old_buffer) + bufp->buffer; \
1513 /* Since we have one byte reserved for the register number argument to
1514 {start,stop}_memory, the maximum number of groups we can report
1515 things about is what fits in that byte. */
1516 #define MAX_REGNUM 255
1518 /* But patterns can have more than `MAX_REGNUM' registers. We just
1519 ignore the excess. */
1520 typedef unsigned regnum_t;
1523 /* Macros for the compile stack. */
1525 /* Since offsets can go either forwards or backwards, this type needs to
1526 be able to hold values from -(MAX_BUF_SIZE - 1) to MAX_BUF_SIZE - 1. */
1527 typedef int pattern_offset_t;
1531 pattern_offset_t begalt_offset;
1532 pattern_offset_t fixup_alt_jump;
1533 pattern_offset_t inner_group_offset;
1534 pattern_offset_t laststart_offset;
1536 } compile_stack_elt_t;
1541 compile_stack_elt_t *stack;
1543 unsigned avail; /* Offset of next open position. */
1544 } compile_stack_type;
1547 #define INIT_COMPILE_STACK_SIZE 32
1549 #define COMPILE_STACK_EMPTY (compile_stack.avail == 0)
1550 #define COMPILE_STACK_FULL (compile_stack.avail == compile_stack.size)
1552 /* The next available element. */
1553 #define COMPILE_STACK_TOP (compile_stack.stack[compile_stack.avail])
1556 /* Set the bit for character C in a list. */
1557 #define SET_LIST_BIT(c) \
1558 (b[((unsigned char) (c)) / BYTEWIDTH] \
1559 |= 1 << (((unsigned char) c) % BYTEWIDTH))
1562 /* Get the next unsigned number in the uncompiled pattern. */
1563 #define GET_UNSIGNED_NUMBER(num) \
1567 while (ISDIGIT (c)) \
1571 num = num * 10 + c - '0'; \
1579 #define CHAR_CLASS_MAX_LENGTH 6 /* Namely, `xdigit'. */
1581 #define IS_CHAR_CLASS(string) \
1582 (STREQ (string, "alpha") || STREQ (string, "upper") \
1583 || STREQ (string, "lower") || STREQ (string, "digit") \
1584 || STREQ (string, "alnum") || STREQ (string, "xdigit") \
1585 || STREQ (string, "space") || STREQ (string, "print") \
1586 || STREQ (string, "punct") || STREQ (string, "graph") \
1587 || STREQ (string, "cntrl") || STREQ (string, "blank"))
1589 #ifndef MATCH_MAY_ALLOCATE
1591 /* If we cannot allocate large objects within re_match_2_internal,
1592 we make the fail stack and register vectors global.
1593 The fail stack, we grow to the maximum size when a regexp
1595 The register vectors, we adjust in size each time we
1596 compile a regexp, according to the number of registers it needs. */
1598 static fail_stack_type fail_stack;
1600 /* Size with which the following vectors are currently allocated.
1601 That is so we can make them bigger as needed,
1602 but never make them smaller. */
1603 static int regs_allocated_size;
1605 static const char ** regstart, ** regend;
1606 static const char ** old_regstart, ** old_regend;
1607 static const char **best_regstart, **best_regend;
1608 static register_info_type *reg_info;
1609 static const char **reg_dummy;
1610 static register_info_type *reg_info_dummy;
1612 /* Make the register vectors big enough for NUM_REGS registers,
1613 but don't make them smaller. */
1616 regex_grow_registers (num_regs)
1619 if (num_regs > regs_allocated_size)
1621 RETALLOC_IF (regstart, num_regs, const char *);
1622 RETALLOC_IF (regend, num_regs, const char *);
1623 RETALLOC_IF (old_regstart, num_regs, const char *);
1624 RETALLOC_IF (old_regend, num_regs, const char *);
1625 RETALLOC_IF (best_regstart, num_regs, const char *);
1626 RETALLOC_IF (best_regend, num_regs, const char *);
1627 RETALLOC_IF (reg_info, num_regs, register_info_type);
1628 RETALLOC_IF (reg_dummy, num_regs, const char *);
1629 RETALLOC_IF (reg_info_dummy, num_regs, register_info_type);
1631 regs_allocated_size = num_regs;
1635 #endif /* not MATCH_MAY_ALLOCATE */
1637 /* `regex_compile' compiles PATTERN (of length SIZE) according to SYNTAX.
1638 Returns one of error codes defined in `regex.h', or zero for success.
1640 Assumes the `allocated' (and perhaps `buffer') and `translate'
1641 fields are set in BUFP on entry.
1643 If it succeeds, results are put in BUFP (if it returns an error, the
1644 contents of BUFP are undefined):
1645 `buffer' is the compiled pattern;
1646 `syntax' is set to SYNTAX;
1647 `used' is set to the length of the compiled pattern;
1648 `fastmap_accurate' is zero;
1649 `re_nsub' is the number of subexpressions in PATTERN;
1650 `not_bol' and `not_eol' are zero;
1652 The `fastmap' and `newline_anchor' fields are neither
1653 examined nor set. */
1655 /* Return, freeing storage we allocated. */
1656 #define FREE_STACK_RETURN(value) \
1657 return (free (compile_stack.stack), value)
1659 static reg_errcode_t
1660 regex_compile (pattern, size, syntax, bufp)
1661 const char *pattern;
1663 reg_syntax_t syntax;
1664 struct re_pattern_buffer *bufp;
1666 /* We fetch characters from PATTERN here. Even though PATTERN is
1667 `char *' (i.e., signed), we declare these variables as unsigned, so
1668 they can be reliably used as array indices. */
1669 register unsigned char c, c1;
1671 /* A random temporary spot in PATTERN. */
1674 /* Points to the end of the buffer, where we should append. */
1675 register unsigned char *b;
1677 /* Keeps track of unclosed groups. */
1678 compile_stack_type compile_stack;
1680 /* Points to the current (ending) position in the pattern. */
1681 const char *p = pattern;
1682 const char *pend = pattern + size;
1684 /* How to translate the characters in the pattern. */
1685 RE_TRANSLATE_TYPE translate = bufp->translate;
1687 /* Address of the count-byte of the most recently inserted `exactn'
1688 command. This makes it possible to tell if a new exact-match
1689 character can be added to that command or if the character requires
1690 a new `exactn' command. */
1691 unsigned char *pending_exact = 0;
1693 /* Address of start of the most recently finished expression.
1694 This tells, e.g., postfix * where to find the start of its
1695 operand. Reset at the beginning of groups and alternatives. */
1696 unsigned char *laststart = 0;
1698 /* Address of beginning of regexp, or inside of last group. */
1699 unsigned char *begalt;
1701 /* Place in the uncompiled pattern (i.e., the {) to
1702 which to go back if the interval is invalid. */
1703 const char *beg_interval;
1705 /* Address of the place where a forward jump should go to the end of
1706 the containing expression. Each alternative of an `or' -- except the
1707 last -- ends with a forward jump of this sort. */
1708 unsigned char *fixup_alt_jump = 0;
1710 /* Counts open-groups as they are encountered. Remembered for the
1711 matching close-group on the compile stack, so the same register
1712 number is put in the stop_memory as the start_memory. */
1713 regnum_t regnum = 0;
1716 DEBUG_PRINT1 ("\nCompiling pattern: ");
1719 unsigned debug_count;
1721 for (debug_count = 0; debug_count < size; debug_count++)
1722 putchar (pattern[debug_count]);
1727 /* Initialize the compile stack. */
1728 compile_stack.stack = TALLOC (INIT_COMPILE_STACK_SIZE, compile_stack_elt_t);
1729 if (compile_stack.stack == NULL)
1732 compile_stack.size = INIT_COMPILE_STACK_SIZE;
1733 compile_stack.avail = 0;
1735 /* Initialize the pattern buffer. */
1736 bufp->syntax = syntax;
1737 bufp->fastmap_accurate = 0;
1738 bufp->not_bol = bufp->not_eol = 0;
1740 /* Set `used' to zero, so that if we return an error, the pattern
1741 printer (for debugging) will think there's no pattern. We reset it
1745 /* Always count groups, whether or not bufp->no_sub is set. */
1748 #if !defined (emacs) && !defined (SYNTAX_TABLE)
1749 /* Initialize the syntax table. */
1750 init_syntax_once ();
1753 if (bufp->allocated == 0)
1756 { /* If zero allocated, but buffer is non-null, try to realloc
1757 enough space. This loses if buffer's address is bogus, but
1758 that is the user's responsibility. */
1759 RETALLOC (bufp->buffer, INIT_BUF_SIZE, unsigned char);
1762 { /* Caller did not allocate a buffer. Do it for them. */
1763 bufp->buffer = TALLOC (INIT_BUF_SIZE, unsigned char);
1765 if (!bufp->buffer) FREE_STACK_RETURN (REG_ESPACE);
1767 bufp->allocated = INIT_BUF_SIZE;
1770 begalt = b = bufp->buffer;
1772 /* Loop through the uncompiled pattern until we're at the end. */
1781 if ( /* If at start of pattern, it's an operator. */
1783 /* If context independent, it's an operator. */
1784 || syntax & RE_CONTEXT_INDEP_ANCHORS
1785 /* Otherwise, depends on what's come before. */
1786 || at_begline_loc_p (pattern, p, syntax))
1796 if ( /* If at end of pattern, it's an operator. */
1798 /* If context independent, it's an operator. */
1799 || syntax & RE_CONTEXT_INDEP_ANCHORS
1800 /* Otherwise, depends on what's next. */
1801 || at_endline_loc_p (p, pend, syntax))
1811 if ((syntax & RE_BK_PLUS_QM)
1812 || (syntax & RE_LIMITED_OPS))
1816 /* If there is no previous pattern... */
1819 if (syntax & RE_CONTEXT_INVALID_OPS)
1820 FREE_STACK_RETURN (REG_BADRPT);
1821 else if (!(syntax & RE_CONTEXT_INDEP_OPS))
1826 /* Are we optimizing this jump? */
1827 boolean keep_string_p = false;
1829 /* 1 means zero (many) matches is allowed. */
1830 char zero_times_ok = 0, many_times_ok = 0;
1832 /* If there is a sequence of repetition chars, collapse it
1833 down to just one (the right one). We can't combine
1834 interval operators with these because of, e.g., `a{2}*',
1835 which should only match an even number of `a's. */
1839 zero_times_ok |= c != '+';
1840 many_times_ok |= c != '?';
1848 || (!(syntax & RE_BK_PLUS_QM) && (c == '+' || c == '?')))
1851 else if (syntax & RE_BK_PLUS_QM && c == '\\')
1853 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
1856 if (!(c1 == '+' || c1 == '?'))
1871 /* If we get here, we found another repeat character. */
1874 /* Star, etc. applied to an empty pattern is equivalent
1875 to an empty pattern. */
1879 /* Now we know whether or not zero matches is allowed
1880 and also whether or not two or more matches is allowed. */
1882 { /* More than one repetition is allowed, so put in at the
1883 end a backward relative jump from `b' to before the next
1884 jump we're going to put in below (which jumps from
1885 laststart to after this jump).
1887 But if we are at the `*' in the exact sequence `.*\n',
1888 insert an unconditional jump backwards to the .,
1889 instead of the beginning of the loop. This way we only
1890 push a failure point once, instead of every time
1891 through the loop. */
1892 assert (p - 1 > pattern);
1894 /* Allocate the space for the jump. */
1895 GET_BUFFER_SPACE (3);
1897 /* We know we are not at the first character of the pattern,
1898 because laststart was nonzero. And we've already
1899 incremented `p', by the way, to be the character after
1900 the `*'. Do we have to do something analogous here
1901 for null bytes, because of RE_DOT_NOT_NULL? */
1902 if (TRANSLATE (*(p - 2)) == TRANSLATE ('.')
1904 && p < pend && TRANSLATE (*p) == TRANSLATE ('\n')
1905 && !(syntax & RE_DOT_NEWLINE))
1906 { /* We have .*\n. */
1907 STORE_JUMP (jump, b, laststart);
1908 keep_string_p = true;
1911 /* Anything else. */
1912 STORE_JUMP (maybe_pop_jump, b, laststart - 3);
1914 /* We've added more stuff to the buffer. */
1918 /* On failure, jump from laststart to b + 3, which will be the
1919 end of the buffer after this jump is inserted. */
1920 GET_BUFFER_SPACE (3);
1921 INSERT_JUMP (keep_string_p ? on_failure_keep_string_jump
1929 /* At least one repetition is required, so insert a
1930 `dummy_failure_jump' before the initial
1931 `on_failure_jump' instruction of the loop. This
1932 effects a skip over that instruction the first time
1933 we hit that loop. */
1934 GET_BUFFER_SPACE (3);
1935 INSERT_JUMP (dummy_failure_jump, laststart, laststart + 6);
1950 boolean had_char_class = false;
1952 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
1954 /* Ensure that we have enough space to push a charset: the
1955 opcode, the length count, and the bitset; 34 bytes in all. */
1956 GET_BUFFER_SPACE (34);
1960 /* We test `*p == '^' twice, instead of using an if
1961 statement, so we only need one BUF_PUSH. */
1962 BUF_PUSH (*p == '^' ? charset_not : charset);
1966 /* Remember the first position in the bracket expression. */
1969 /* Push the number of bytes in the bitmap. */
1970 BUF_PUSH ((1 << BYTEWIDTH) / BYTEWIDTH);
1972 /* Clear the whole map. */
1973 bzero (b, (1 << BYTEWIDTH) / BYTEWIDTH);
1975 /* charset_not matches newline according to a syntax bit. */
1976 if ((re_opcode_t) b[-2] == charset_not
1977 && (syntax & RE_HAT_LISTS_NOT_NEWLINE))
1978 SET_LIST_BIT ('\n');
1980 /* Read in characters and ranges, setting map bits. */
1983 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
1987 /* \ might escape characters inside [...] and [^...]. */
1988 if ((syntax & RE_BACKSLASH_ESCAPE_IN_LISTS) && c == '\\')
1990 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
1997 /* Could be the end of the bracket expression. If it's
1998 not (i.e., when the bracket expression is `[]' so
1999 far), the ']' character bit gets set way below. */
2000 if (c == ']' && p != p1 + 1)
2003 /* Look ahead to see if it's a range when the last thing
2004 was a character class. */
2005 if (had_char_class && c == '-' && *p != ']')
2006 FREE_STACK_RETURN (REG_ERANGE);
2008 /* Look ahead to see if it's a range when the last thing
2009 was a character: if this is a hyphen not at the
2010 beginning or the end of a list, then it's the range
2013 && !(p - 2 >= pattern && p[-2] == '[')
2014 && !(p - 3 >= pattern && p[-3] == '[' && p[-2] == '^')
2018 = compile_range (&p, pend, translate, syntax, b);
2019 if (ret != REG_NOERROR) FREE_STACK_RETURN (ret);
2022 else if (p[0] == '-' && p[1] != ']')
2023 { /* This handles ranges made up of characters only. */
2026 /* Move past the `-'. */
2029 ret = compile_range (&p, pend, translate, syntax, b);
2030 if (ret != REG_NOERROR) FREE_STACK_RETURN (ret);
2033 /* See if we're at the beginning of a possible character
2036 else if (syntax & RE_CHAR_CLASSES && c == '[' && *p == ':')
2037 { /* Leave room for the null. */
2038 char str[CHAR_CLASS_MAX_LENGTH + 1];
2043 /* If pattern is `[[:'. */
2044 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2049 if (c == ':' || c == ']' || p == pend
2050 || c1 == CHAR_CLASS_MAX_LENGTH)
2056 /* If isn't a word bracketed by `[:' and:`]':
2057 undo the ending character, the letters, and leave
2058 the leading `:' and `[' (but set bits for them). */
2059 if (c == ':' && *p == ']')
2062 boolean is_alnum = STREQ (str, "alnum");
2063 boolean is_alpha = STREQ (str, "alpha");
2064 boolean is_blank = STREQ (str, "blank");
2065 boolean is_cntrl = STREQ (str, "cntrl");
2066 boolean is_digit = STREQ (str, "digit");
2067 boolean is_graph = STREQ (str, "graph");
2068 boolean is_lower = STREQ (str, "lower");
2069 boolean is_print = STREQ (str, "print");
2070 boolean is_punct = STREQ (str, "punct");
2071 boolean is_space = STREQ (str, "space");
2072 boolean is_upper = STREQ (str, "upper");
2073 boolean is_xdigit = STREQ (str, "xdigit");
2075 if (!IS_CHAR_CLASS (str))
2076 FREE_STACK_RETURN (REG_ECTYPE);
2078 /* Throw away the ] at the end of the character
2082 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2084 for (ch = 0; ch < 1 << BYTEWIDTH; ch++)
2086 /* This was split into 3 if's to
2087 avoid an arbitrary limit in some compiler. */
2088 if ( (is_alnum && ISALNUM (ch))
2089 || (is_alpha && ISALPHA (ch))
2090 || (is_blank && ISBLANK (ch))
2091 || (is_cntrl && ISCNTRL (ch)))
2093 if ( (is_digit && ISDIGIT (ch))
2094 || (is_graph && ISGRAPH (ch))
2095 || (is_lower && ISLOWER (ch))
2096 || (is_print && ISPRINT (ch)))
2098 if ( (is_punct && ISPUNCT (ch))
2099 || (is_space && ISSPACE (ch))
2100 || (is_upper && ISUPPER (ch))
2101 || (is_xdigit && ISXDIGIT (ch)))
2104 had_char_class = true;
2113 had_char_class = false;
2118 had_char_class = false;
2123 /* Discard any (non)matching list bytes that are all 0 at the
2124 end of the map. Decrease the map-length byte too. */
2125 while ((int) b[-1] > 0 && b[b[-1] - 1] == 0)
2133 if (syntax & RE_NO_BK_PARENS)
2140 if (syntax & RE_NO_BK_PARENS)
2147 if (syntax & RE_NEWLINE_ALT)
2154 if (syntax & RE_NO_BK_VBAR)
2161 if (syntax & RE_INTERVALS && syntax & RE_NO_BK_BRACES)
2162 goto handle_interval;
2168 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
2170 /* Do not translate the character after the \, so that we can
2171 distinguish, e.g., \B from \b, even if we normally would
2172 translate, e.g., B to b. */
2178 if (syntax & RE_NO_BK_PARENS)
2179 goto normal_backslash;
2185 if (COMPILE_STACK_FULL)
2187 RETALLOC (compile_stack.stack, compile_stack.size << 1,
2188 compile_stack_elt_t);
2189 if (compile_stack.stack == NULL) return REG_ESPACE;
2191 compile_stack.size <<= 1;
2194 /* These are the values to restore when we hit end of this
2195 group. They are all relative offsets, so that if the
2196 whole pattern moves because of realloc, they will still
2198 COMPILE_STACK_TOP.begalt_offset = begalt - bufp->buffer;
2199 COMPILE_STACK_TOP.fixup_alt_jump
2200 = fixup_alt_jump ? fixup_alt_jump - bufp->buffer + 1 : 0;
2201 COMPILE_STACK_TOP.laststart_offset = b - bufp->buffer;
2202 COMPILE_STACK_TOP.regnum = regnum;
2204 /* We will eventually replace the 0 with the number of
2205 groups inner to this one. But do not push a
2206 start_memory for groups beyond the last one we can
2207 represent in the compiled pattern. */
2208 if (regnum <= MAX_REGNUM)
2210 COMPILE_STACK_TOP.inner_group_offset = b - bufp->buffer + 2;
2211 BUF_PUSH_3 (start_memory, regnum, 0);
2214 compile_stack.avail++;
2219 /* If we've reached MAX_REGNUM groups, then this open
2220 won't actually generate any code, so we'll have to
2221 clear pending_exact explicitly. */
2227 if (syntax & RE_NO_BK_PARENS) goto normal_backslash;
2229 if (COMPILE_STACK_EMPTY)
2230 if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD)
2231 goto normal_backslash;
2233 FREE_STACK_RETURN (REG_ERPAREN);
2237 { /* Push a dummy failure point at the end of the
2238 alternative for a possible future
2239 `pop_failure_jump' to pop. See comments at
2240 `push_dummy_failure' in `re_match_2'. */
2241 BUF_PUSH (push_dummy_failure);
2243 /* We allocated space for this jump when we assigned
2244 to `fixup_alt_jump', in the `handle_alt' case below. */
2245 STORE_JUMP (jump_past_alt, fixup_alt_jump, b - 1);
2248 /* See similar code for backslashed left paren above. */
2249 if (COMPILE_STACK_EMPTY)
2250 if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD)
2253 FREE_STACK_RETURN (REG_ERPAREN);
2255 /* Since we just checked for an empty stack above, this
2256 ``can't happen''. */
2257 assert (compile_stack.avail != 0);
2259 /* We don't just want to restore into `regnum', because
2260 later groups should continue to be numbered higher,
2261 as in `(ab)c(de)' -- the second group is #2. */
2262 regnum_t this_group_regnum;
2264 compile_stack.avail--;
2265 begalt = bufp->buffer + COMPILE_STACK_TOP.begalt_offset;
2267 = COMPILE_STACK_TOP.fixup_alt_jump
2268 ? bufp->buffer + COMPILE_STACK_TOP.fixup_alt_jump - 1
2270 laststart = bufp->buffer + COMPILE_STACK_TOP.laststart_offset;
2271 this_group_regnum = COMPILE_STACK_TOP.regnum;
2272 /* If we've reached MAX_REGNUM groups, then this open
2273 won't actually generate any code, so we'll have to
2274 clear pending_exact explicitly. */
2277 /* We're at the end of the group, so now we know how many
2278 groups were inside this one. */
2279 if (this_group_regnum <= MAX_REGNUM)
2281 unsigned char *inner_group_loc
2282 = bufp->buffer + COMPILE_STACK_TOP.inner_group_offset;
2284 *inner_group_loc = regnum - this_group_regnum;
2285 BUF_PUSH_3 (stop_memory, this_group_regnum,
2286 regnum - this_group_regnum);
2292 case '|': /* `\|'. */
2293 if (syntax & RE_LIMITED_OPS || syntax & RE_NO_BK_VBAR)
2294 goto normal_backslash;
2296 if (syntax & RE_LIMITED_OPS)
2299 /* Insert before the previous alternative a jump which
2300 jumps to this alternative if the former fails. */
2301 GET_BUFFER_SPACE (3);
2302 INSERT_JUMP (on_failure_jump, begalt, b + 6);
2306 /* The alternative before this one has a jump after it
2307 which gets executed if it gets matched. Adjust that
2308 jump so it will jump to this alternative's analogous
2309 jump (put in below, which in turn will jump to the next
2310 (if any) alternative's such jump, etc.). The last such
2311 jump jumps to the correct final destination. A picture:
2317 If we are at `b', then fixup_alt_jump right now points to a
2318 three-byte space after `a'. We'll put in the jump, set
2319 fixup_alt_jump to right after `b', and leave behind three
2320 bytes which we'll fill in when we get to after `c'. */
2323 STORE_JUMP (jump_past_alt, fixup_alt_jump, b);
2325 /* Mark and leave space for a jump after this alternative,
2326 to be filled in later either by next alternative or
2327 when know we're at the end of a series of alternatives. */
2329 GET_BUFFER_SPACE (3);
2338 /* If \{ is a literal. */
2339 if (!(syntax & RE_INTERVALS)
2340 /* If we're at `\{' and it's not the open-interval
2342 || ((syntax & RE_INTERVALS) && (syntax & RE_NO_BK_BRACES))
2343 || (p - 2 == pattern && p == pend))
2344 goto normal_backslash;
2348 /* If got here, then the syntax allows intervals. */
2350 /* At least (most) this many matches must be made. */
2351 int lower_bound = -1, upper_bound = -1;
2353 beg_interval = p - 1;
2357 if (syntax & RE_NO_BK_BRACES)
2358 goto unfetch_interval;
2360 FREE_STACK_RETURN (REG_EBRACE);
2363 GET_UNSIGNED_NUMBER (lower_bound);
2367 GET_UNSIGNED_NUMBER (upper_bound);
2368 if (upper_bound < 0) upper_bound = RE_DUP_MAX;
2371 /* Interval such as `{1}' => match exactly once. */
2372 upper_bound = lower_bound;
2374 if (lower_bound < 0 || upper_bound > RE_DUP_MAX
2375 || lower_bound > upper_bound)
2377 if (syntax & RE_NO_BK_BRACES)
2378 goto unfetch_interval;
2380 FREE_STACK_RETURN (REG_BADBR);
2383 if (!(syntax & RE_NO_BK_BRACES))
2385 if (c != '\\') FREE_STACK_RETURN (REG_EBRACE);
2392 if (syntax & RE_NO_BK_BRACES)
2393 goto unfetch_interval;
2395 FREE_STACK_RETURN (REG_BADBR);
2398 /* We just parsed a valid interval. */
2400 /* If it's invalid to have no preceding re. */
2403 if (syntax & RE_CONTEXT_INVALID_OPS)
2404 FREE_STACK_RETURN (REG_BADRPT);
2405 else if (syntax & RE_CONTEXT_INDEP_OPS)
2408 goto unfetch_interval;
2411 /* If the upper bound is zero, don't want to succeed at
2412 all; jump from `laststart' to `b + 3', which will be
2413 the end of the buffer after we insert the jump. */
2414 if (upper_bound == 0)
2416 GET_BUFFER_SPACE (3);
2417 INSERT_JUMP (jump, laststart, b + 3);
2421 /* Otherwise, we have a nontrivial interval. When
2422 we're all done, the pattern will look like:
2423 set_number_at <jump count> <upper bound>
2424 set_number_at <succeed_n count> <lower bound>
2425 succeed_n <after jump addr> <succeed_n count>
2427 jump_n <succeed_n addr> <jump count>
2428 (The upper bound and `jump_n' are omitted if
2429 `upper_bound' is 1, though.) */
2431 { /* If the upper bound is > 1, we need to insert
2432 more at the end of the loop. */
2433 unsigned nbytes = 10 + (upper_bound > 1) * 10;
2435 GET_BUFFER_SPACE (nbytes);
2437 /* Initialize lower bound of the `succeed_n', even
2438 though it will be set during matching by its
2439 attendant `set_number_at' (inserted next),
2440 because `re_compile_fastmap' needs to know.
2441 Jump to the `jump_n' we might insert below. */
2442 INSERT_JUMP2 (succeed_n, laststart,
2443 b + 5 + (upper_bound > 1) * 5,
2447 /* Code to initialize the lower bound. Insert
2448 before the `succeed_n'. The `5' is the last two
2449 bytes of this `set_number_at', plus 3 bytes of
2450 the following `succeed_n'. */
2451 insert_op2 (set_number_at, laststart, 5, lower_bound, b);
2454 if (upper_bound > 1)
2455 { /* More than one repetition is allowed, so
2456 append a backward jump to the `succeed_n'
2457 that starts this interval.
2459 When we've reached this during matching,
2460 we'll have matched the interval once, so
2461 jump back only `upper_bound - 1' times. */
2462 STORE_JUMP2 (jump_n, b, laststart + 5,
2466 /* The location we want to set is the second
2467 parameter of the `jump_n'; that is `b-2' as
2468 an absolute address. `laststart' will be
2469 the `set_number_at' we're about to insert;
2470 `laststart+3' the number to set, the source
2471 for the relative address. But we are
2472 inserting into the middle of the pattern --
2473 so everything is getting moved up by 5.
2474 Conclusion: (b - 2) - (laststart + 3) + 5,
2475 i.e., b - laststart.
2477 We insert this at the beginning of the loop
2478 so that if we fail during matching, we'll
2479 reinitialize the bounds. */
2480 insert_op2 (set_number_at, laststart, b - laststart,
2481 upper_bound - 1, b);
2486 beg_interval = NULL;
2491 /* If an invalid interval, match the characters as literals. */
2492 assert (beg_interval);
2494 beg_interval = NULL;
2496 /* normal_char and normal_backslash need `c'. */
2499 if (!(syntax & RE_NO_BK_BRACES))
2501 if (p > pattern && p[-1] == '\\')
2502 goto normal_backslash;
2507 /* There is no way to specify the before_dot and after_dot
2508 operators. rms says this is ok. --karl */
2516 BUF_PUSH_2 (syntaxspec, syntax_spec_code[c]);
2522 BUF_PUSH_2 (notsyntaxspec, syntax_spec_code[c]);
2529 BUF_PUSH (wordchar);
2535 BUF_PUSH (notwordchar);
2548 BUF_PUSH (wordbound);
2552 BUF_PUSH (notwordbound);
2563 case '1': case '2': case '3': case '4': case '5':
2564 case '6': case '7': case '8': case '9':
2565 if (syntax & RE_NO_BK_REFS)
2571 FREE_STACK_RETURN (REG_ESUBREG);
2573 /* Can't back reference to a subexpression if inside of it. */
2574 if (group_in_compile_stack (compile_stack, c1))
2578 BUF_PUSH_2 (duplicate, c1);
2584 if (syntax & RE_BK_PLUS_QM)
2587 goto normal_backslash;
2591 /* You might think it would be useful for \ to mean
2592 not to translate; but if we don't translate it
2593 it will never match anything. */
2601 /* Expects the character in `c'. */
2603 /* If no exactn currently being built. */
2606 /* If last exactn not at current position. */
2607 || pending_exact + *pending_exact + 1 != b
2609 /* We have only one byte following the exactn for the count. */
2610 || *pending_exact == (1 << BYTEWIDTH) - 1
2612 /* If followed by a repetition operator. */
2613 || *p == '*' || *p == '^'
2614 || ((syntax & RE_BK_PLUS_QM)
2615 ? *p == '\\' && (p[1] == '+' || p[1] == '?')
2616 : (*p == '+' || *p == '?'))
2617 || ((syntax & RE_INTERVALS)
2618 && ((syntax & RE_NO_BK_BRACES)
2620 : (p[0] == '\\' && p[1] == '{'))))
2622 /* Start building a new exactn. */
2626 BUF_PUSH_2 (exactn, 0);
2627 pending_exact = b - 1;
2634 } /* while p != pend */
2637 /* Through the pattern now. */
2640 STORE_JUMP (jump_past_alt, fixup_alt_jump, b);
2642 if (!COMPILE_STACK_EMPTY)
2643 FREE_STACK_RETURN (REG_EPAREN);
2645 /* If we don't want backtracking, force success
2646 the first time we reach the end of the compiled pattern. */
2647 if (syntax & RE_NO_POSIX_BACKTRACKING)
2650 free (compile_stack.stack);
2652 /* We have succeeded; set the length of the buffer. */
2653 bufp->used = b - bufp->buffer;
2658 DEBUG_PRINT1 ("\nCompiled pattern: \n");
2659 print_compiled_pattern (bufp);
2663 #ifndef MATCH_MAY_ALLOCATE
2664 /* Initialize the failure stack to the largest possible stack. This
2665 isn't necessary unless we're trying to avoid calling alloca in
2666 the search and match routines. */
2668 int num_regs = bufp->re_nsub + 1;
2670 /* Since DOUBLE_FAIL_STACK refuses to double only if the current size
2671 is strictly greater than re_max_failures, the largest possible stack
2672 is 2 * re_max_failures failure points. */
2673 if (fail_stack.size < (2 * re_max_failures * MAX_FAILURE_ITEMS))
2675 fail_stack.size = (2 * re_max_failures * MAX_FAILURE_ITEMS);
2678 if (! fail_stack.stack)
2680 = (fail_stack_elt_t *) xmalloc (fail_stack.size
2681 * sizeof (fail_stack_elt_t));
2684 = (fail_stack_elt_t *) xrealloc (fail_stack.stack,
2686 * sizeof (fail_stack_elt_t)));
2687 #else /* not emacs */
2688 if (! fail_stack.stack)
2690 = (fail_stack_elt_t *) malloc (fail_stack.size
2691 * sizeof (fail_stack_elt_t));
2694 = (fail_stack_elt_t *) realloc (fail_stack.stack,
2696 * sizeof (fail_stack_elt_t)));
2697 #endif /* not emacs */
2700 regex_grow_registers (num_regs);
2702 #endif /* not MATCH_MAY_ALLOCATE */
2705 } /* regex_compile */
2707 /* Subroutines for `regex_compile'. */
2709 /* Store OP at LOC followed by two-byte integer parameter ARG. */
2712 store_op1 (op, loc, arg)
2717 *loc = (unsigned char) op;
2718 STORE_NUMBER (loc + 1, arg);
2722 /* Like `store_op1', but for two two-byte parameters ARG1 and ARG2. */
2725 store_op2 (op, loc, arg1, arg2)
2730 *loc = (unsigned char) op;
2731 STORE_NUMBER (loc + 1, arg1);
2732 STORE_NUMBER (loc + 3, arg2);
2736 /* Copy the bytes from LOC to END to open up three bytes of space at LOC
2737 for OP followed by two-byte integer parameter ARG. */
2740 insert_op1 (op, loc, arg, end)
2746 register unsigned char *pfrom = end;
2747 register unsigned char *pto = end + 3;
2749 while (pfrom != loc)
2752 store_op1 (op, loc, arg);
2756 /* Like `insert_op1', but for two two-byte parameters ARG1 and ARG2. */
2759 insert_op2 (op, loc, arg1, arg2, end)
2765 register unsigned char *pfrom = end;
2766 register unsigned char *pto = end + 5;
2768 while (pfrom != loc)
2771 store_op2 (op, loc, arg1, arg2);
2775 /* P points to just after a ^ in PATTERN. Return true if that ^ comes
2776 after an alternative or a begin-subexpression. We assume there is at
2777 least one character before the ^. */
2780 at_begline_loc_p (pattern, p, syntax)
2781 const char *pattern, *p;
2782 reg_syntax_t syntax;
2784 const char *prev = p - 2;
2785 boolean prev_prev_backslash = prev > pattern && prev[-1] == '\\';
2788 /* After a subexpression? */
2789 (*prev == '(' && (syntax & RE_NO_BK_PARENS || prev_prev_backslash))
2790 /* After an alternative? */
2791 || (*prev == '|' && (syntax & RE_NO_BK_VBAR || prev_prev_backslash));
2795 /* The dual of at_begline_loc_p. This one is for $. We assume there is
2796 at least one character after the $, i.e., `P < PEND'. */
2799 at_endline_loc_p (p, pend, syntax)
2800 const char *p, *pend;
2803 const char *next = p;
2804 boolean next_backslash = *next == '\\';
2805 const char *next_next = p + 1 < pend ? p + 1 : 0;
2808 /* Before a subexpression? */
2809 (syntax & RE_NO_BK_PARENS ? *next == ')'
2810 : next_backslash && next_next && *next_next == ')')
2811 /* Before an alternative? */
2812 || (syntax & RE_NO_BK_VBAR ? *next == '|'
2813 : next_backslash && next_next && *next_next == '|');
2817 /* Returns true if REGNUM is in one of COMPILE_STACK's elements and
2818 false if it's not. */
2821 group_in_compile_stack (compile_stack, regnum)
2822 compile_stack_type compile_stack;
2827 for (this_element = compile_stack.avail - 1;
2830 if (compile_stack.stack[this_element].regnum == regnum)
2837 /* Read the ending character of a range (in a bracket expression) from the
2838 uncompiled pattern *P_PTR (which ends at PEND). We assume the
2839 starting character is in `P[-2]'. (`P[-1]' is the character `-'.)
2840 Then we set the translation of all bits between the starting and
2841 ending characters (inclusive) in the compiled pattern B.
2843 Return an error code.
2845 We use these short variable names so we can use the same macros as
2846 `regex_compile' itself. */
2848 static reg_errcode_t
2849 compile_range (p_ptr, pend, translate, syntax, b)
2850 const char **p_ptr, *pend;
2851 RE_TRANSLATE_TYPE translate;
2852 reg_syntax_t syntax;
2857 const char *p = *p_ptr;
2858 int range_start, range_end;
2863 /* Even though the pattern is a signed `char *', we need to fetch
2864 with unsigned char *'s; if the high bit of the pattern character
2865 is set, the range endpoints will be negative if we fetch using a
2868 We also want to fetch the endpoints without translating them; the
2869 appropriate translation is done in the bit-setting loop below. */
2870 /* The SVR4 compiler on the 3B2 had trouble with unsigned const char *. */
2871 range_start = ((const unsigned char *) p)[-2];
2872 range_end = ((const unsigned char *) p)[0];
2874 /* Have to increment the pointer into the pattern string, so the
2875 caller isn't still at the ending character. */
2878 /* If the start is after the end, the range is empty. */
2879 if (range_start > range_end)
2880 return syntax & RE_NO_EMPTY_RANGES ? REG_ERANGE : REG_NOERROR;
2882 /* Here we see why `this_char' has to be larger than an `unsigned
2883 char' -- the range is inclusive, so if `range_end' == 0xff
2884 (assuming 8-bit characters), we would otherwise go into an infinite
2885 loop, since all characters <= 0xff. */
2886 for (this_char = range_start; this_char <= range_end; this_char++)
2888 SET_LIST_BIT (TRANSLATE (this_char));
2894 /* re_compile_fastmap computes a ``fastmap'' for the compiled pattern in
2895 BUFP. A fastmap records which of the (1 << BYTEWIDTH) possible
2896 characters can start a string that matches the pattern. This fastmap
2897 is used by re_search to skip quickly over impossible starting points.
2899 The caller must supply the address of a (1 << BYTEWIDTH)-byte data
2900 area as BUFP->fastmap.
2902 We set the `fastmap', `fastmap_accurate', and `can_be_null' fields in
2905 Returns 0 if we succeed, -2 if an internal error. */
2908 re_compile_fastmap (bufp)
2909 struct re_pattern_buffer *bufp;
2912 #ifdef MATCH_MAY_ALLOCATE
2913 fail_stack_type fail_stack;
2915 #ifndef REGEX_MALLOC
2918 /* We don't push any register information onto the failure stack. */
2919 unsigned num_regs = 0;
2921 register char *fastmap = bufp->fastmap;
2922 unsigned char *pattern = bufp->buffer;
2923 unsigned long size = bufp->used;
2924 unsigned char *p = pattern;
2925 register unsigned char *pend = pattern + size;
2927 /* This holds the pointer to the failure stack, when
2928 it is allocated relocatably. */
2929 fail_stack_elt_t *failure_stack_ptr;
2931 /* Assume that each path through the pattern can be null until
2932 proven otherwise. We set this false at the bottom of switch
2933 statement, to which we get only if a particular path doesn't
2934 match the empty string. */
2935 boolean path_can_be_null = true;
2937 /* We aren't doing a `succeed_n' to begin with. */
2938 boolean succeed_n_p = false;
2940 assert (fastmap != NULL && p != NULL);
2943 bzero (fastmap, 1 << BYTEWIDTH); /* Assume nothing's valid. */
2944 bufp->fastmap_accurate = 1; /* It will be when we're done. */
2945 bufp->can_be_null = 0;
2949 if (p == pend || *p == succeed)
2951 /* We have reached the (effective) end of pattern. */
2952 if (!FAIL_STACK_EMPTY ())
2954 bufp->can_be_null |= path_can_be_null;
2956 /* Reset for next path. */
2957 path_can_be_null = true;
2959 p = fail_stack.stack[--fail_stack.avail].pointer;
2967 /* We should never be about to go beyond the end of the pattern. */
2970 switch (SWITCH_ENUM_CAST ((re_opcode_t) *p++))
2973 /* I guess the idea here is to simply not bother with a fastmap
2974 if a backreference is used, since it's too hard to figure out
2975 the fastmap for the corresponding group. Setting
2976 `can_be_null' stops `re_search_2' from using the fastmap, so
2977 that is all we do. */
2979 bufp->can_be_null = 1;
2983 /* Following are the cases which match a character. These end
2992 for (j = *p++ * BYTEWIDTH - 1; j >= 0; j--)
2993 if (p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH)))
2999 /* Chars beyond end of map must be allowed. */
3000 for (j = *p * BYTEWIDTH; j < (1 << BYTEWIDTH); j++)
3003 for (j = *p++ * BYTEWIDTH - 1; j >= 0; j--)
3004 if (!(p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH))))
3010 for (j = 0; j < (1 << BYTEWIDTH); j++)
3011 if (SYNTAX (j) == Sword)
3017 for (j = 0; j < (1 << BYTEWIDTH); j++)
3018 if (SYNTAX (j) != Sword)
3025 int fastmap_newline = fastmap['\n'];
3027 /* `.' matches anything ... */
3028 for (j = 0; j < (1 << BYTEWIDTH); j++)
3031 /* ... except perhaps newline. */
3032 if (!(bufp->syntax & RE_DOT_NEWLINE))
3033 fastmap['\n'] = fastmap_newline;
3035 /* Return if we have already set `can_be_null'; if we have,
3036 then the fastmap is irrelevant. Something's wrong here. */
3037 else if (bufp->can_be_null)
3040 /* Otherwise, have to check alternative paths. */
3047 for (j = 0; j < (1 << BYTEWIDTH); j++)
3048 if (SYNTAX (j) == (enum syntaxcode) k)
3055 for (j = 0; j < (1 << BYTEWIDTH); j++)
3056 if (SYNTAX (j) != (enum syntaxcode) k)
3061 /* All cases after this match the empty string. These end with
3081 case push_dummy_failure:
3086 case pop_failure_jump:
3087 case maybe_pop_jump:
3090 case dummy_failure_jump:
3091 EXTRACT_NUMBER_AND_INCR (j, p);
3096 /* Jump backward implies we just went through the body of a
3097 loop and matched nothing. Opcode jumped to should be
3098 `on_failure_jump' or `succeed_n'. Just treat it like an
3099 ordinary jump. For a * loop, it has pushed its failure
3100 point already; if so, discard that as redundant. */
3101 if ((re_opcode_t) *p != on_failure_jump
3102 && (re_opcode_t) *p != succeed_n)
3106 EXTRACT_NUMBER_AND_INCR (j, p);
3109 /* If what's on the stack is where we are now, pop it. */
3110 if (!FAIL_STACK_EMPTY ()
3111 && fail_stack.stack[fail_stack.avail - 1].pointer == p)
3117 case on_failure_jump:
3118 case on_failure_keep_string_jump:
3119 handle_on_failure_jump:
3120 EXTRACT_NUMBER_AND_INCR (j, p);
3122 /* For some patterns, e.g., `(a?)?', `p+j' here points to the
3123 end of the pattern. We don't want to push such a point,
3124 since when we restore it above, entering the switch will
3125 increment `p' past the end of the pattern. We don't need
3126 to push such a point since we obviously won't find any more
3127 fastmap entries beyond `pend'. Such a pattern can match
3128 the null string, though. */
3131 if (!PUSH_PATTERN_OP (p + j, fail_stack))
3133 RESET_FAIL_STACK ();
3138 bufp->can_be_null = 1;
3142 EXTRACT_NUMBER_AND_INCR (k, p); /* Skip the n. */
3143 succeed_n_p = false;
3150 /* Get to the number of times to succeed. */
3153 /* Increment p past the n for when k != 0. */
3154 EXTRACT_NUMBER_AND_INCR (k, p);
3158 succeed_n_p = true; /* Spaghetti code alert. */
3159 goto handle_on_failure_jump;
3176 abort (); /* We have listed all the cases. */
3179 /* Getting here means we have found the possible starting
3180 characters for one path of the pattern -- and that the empty
3181 string does not match. We need not follow this path further.
3182 Instead, look at the next alternative (remembered on the
3183 stack), or quit if no more. The test at the top of the loop
3184 does these things. */
3185 path_can_be_null = false;
3189 /* Set `can_be_null' for the last path (also the first path, if the
3190 pattern is empty). */
3191 bufp->can_be_null |= path_can_be_null;
3194 RESET_FAIL_STACK ();
3196 } /* re_compile_fastmap */
3198 /* Set REGS to hold NUM_REGS registers, storing them in STARTS and
3199 ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use
3200 this memory for recording register information. STARTS and ENDS
3201 must be allocated using the malloc library routine, and must each
3202 be at least NUM_REGS * sizeof (regoff_t) bytes long.
3204 If NUM_REGS == 0, then subsequent matches should allocate their own
3207 Unless this function is called, the first search or match using
3208 PATTERN_BUFFER will allocate its own register data, without
3209 freeing the old data. */
3212 re_set_registers (bufp, regs, num_regs, starts, ends)
3213 struct re_pattern_buffer *bufp;
3214 struct re_registers *regs;
3216 regoff_t *starts, *ends;
3220 bufp->regs_allocated = REGS_REALLOCATE;
3221 regs->num_regs = num_regs;
3222 regs->start = starts;
3227 bufp->regs_allocated = REGS_UNALLOCATED;
3229 regs->start = regs->end = (regoff_t *) 0;
3233 /* Searching routines. */
3235 /* Like re_search_2, below, but only one string is specified, and
3236 doesn't let you say where to stop matching. */
3239 re_search (bufp, string, size, startpos, range, regs)
3240 struct re_pattern_buffer *bufp;
3242 int size, startpos, range;
3243 struct re_registers *regs;
3245 return re_search_2 (bufp, NULL, 0, string, size, startpos, range,
3250 /* Using the compiled pattern in BUFP->buffer, first tries to match the
3251 virtual concatenation of STRING1 and STRING2, starting first at index
3252 STARTPOS, then at STARTPOS + 1, and so on.
3254 STRING1 and STRING2 have length SIZE1 and SIZE2, respectively.
3256 RANGE is how far to scan while trying to match. RANGE = 0 means try
3257 only at STARTPOS; in general, the last start tried is STARTPOS +
3260 In REGS, return the indices of the virtual concatenation of STRING1
3261 and STRING2 that matched the entire BUFP->buffer and its contained
3264 Do not consider matching one past the index STOP in the virtual
3265 concatenation of STRING1 and STRING2.
3267 We return either the position in the strings at which the match was
3268 found, -1 if no match, or -2 if error (such as failure
3272 re_search_2 (bufp, string1, size1, string2, size2, startpos, range, regs, stop)
3273 struct re_pattern_buffer *bufp;
3274 const char *string1, *string2;
3278 struct re_registers *regs;
3282 register char *fastmap = bufp->fastmap;
3283 register RE_TRANSLATE_TYPE translate = bufp->translate;
3284 int total_size = size1 + size2;
3285 int endpos = startpos + range;
3286 int anchored_start = 0;
3288 /* Check for out-of-range STARTPOS. */
3289 if (startpos < 0 || startpos > total_size)
3292 /* Fix up RANGE if it might eventually take us outside
3293 the virtual concatenation of STRING1 and STRING2.
3294 Make sure we won't move STARTPOS below 0 or above TOTAL_SIZE. */
3296 range = 0 - startpos;
3297 else if (endpos > total_size)
3298 range = total_size - startpos;
3300 /* If the search isn't to be a backwards one, don't waste time in a
3301 search for a pattern that must be anchored. */
3302 if (bufp->used > 0 && (re_opcode_t) bufp->buffer[0] == begbuf && range > 0)
3311 /* In a forward search for something that starts with \=.
3312 don't keep searching past point. */
3313 if (bufp->used > 0 && (re_opcode_t) bufp->buffer[0] == at_dot && range > 0)
3315 range = PT - startpos;
3321 /* Update the fastmap now if not correct already. */
3322 if (fastmap && !bufp->fastmap_accurate)
3323 if (re_compile_fastmap (bufp) == -2)
3326 /* See whether the pattern is anchored. */
3327 if (bufp->buffer[0] == begline)
3330 /* Loop through the string, looking for a place to start matching. */
3333 /* If the pattern is anchored,
3334 skip quickly past places we cannot match.
3335 We don't bother to treat startpos == 0 specially
3336 because that case doesn't repeat. */
3337 if (anchored_start && startpos > 0)
3339 if (! (bufp->newline_anchor
3340 && ((startpos <= size1 ? string1[startpos - 1]
3341 : string2[startpos - size1 - 1])
3346 /* If a fastmap is supplied, skip quickly over characters that
3347 cannot be the start of a match. If the pattern can match the
3348 null string, however, we don't need to skip characters; we want
3349 the first null string. */
3350 if (fastmap && startpos < total_size && !bufp->can_be_null)
3352 if (range > 0) /* Searching forwards. */
3354 register const char *d;
3355 register int lim = 0;
3358 if (startpos < size1 && startpos + range >= size1)
3359 lim = range - (size1 - startpos);
3361 d = (startpos >= size1 ? string2 - size1 : string1) + startpos;
3363 /* Written out as an if-else to avoid testing `translate'
3367 && !fastmap[(unsigned char)
3368 translate[(unsigned char) *d++]])
3371 while (range > lim && !fastmap[(unsigned char) *d++])
3374 startpos += irange - range;
3376 else /* Searching backwards. */
3378 register char c = (size1 == 0 || startpos >= size1
3379 ? string2[startpos - size1]
3380 : string1[startpos]);
3382 if (!fastmap[(unsigned char) TRANSLATE (c)])
3387 /* If can't match the null string, and that's all we have left, fail. */
3388 if (range >= 0 && startpos == total_size && fastmap
3389 && !bufp->can_be_null)
3392 val = re_match_2_internal (bufp, string1, size1, string2, size2,
3393 startpos, regs, stop);
3394 #ifndef REGEX_MALLOC
3423 /* Declarations and macros for re_match_2. */
3425 static int bcmp_translate ();
3426 static boolean alt_match_null_string_p (),
3427 common_op_match_null_string_p (),
3428 group_match_null_string_p ();
3430 /* This converts PTR, a pointer into one of the search strings `string1'
3431 and `string2' into an offset from the beginning of that string. */
3432 #define POINTER_TO_OFFSET(ptr) \
3433 (FIRST_STRING_P (ptr) \
3434 ? ((regoff_t) ((ptr) - string1)) \
3435 : ((regoff_t) ((ptr) - string2 + size1)))
3437 /* Macros for dealing with the split strings in re_match_2. */
3439 #define MATCHING_IN_FIRST_STRING (dend == end_match_1)
3441 /* Call before fetching a character with *d. This switches over to
3442 string2 if necessary. */
3443 #define PREFETCH() \
3446 /* End of string2 => fail. */ \
3447 if (dend == end_match_2) \
3449 /* End of string1 => advance to string2. */ \
3451 dend = end_match_2; \
3455 /* Test if at very beginning or at very end of the virtual concatenation
3456 of `string1' and `string2'. If only one string, it's `string2'. */
3457 #define AT_STRINGS_BEG(d) ((d) == (size1 ? string1 : string2) || !size2)
3458 #define AT_STRINGS_END(d) ((d) == end2)
3461 /* Test if D points to a character which is word-constituent. We have
3462 two special cases to check for: if past the end of string1, look at
3463 the first character in string2; and if before the beginning of
3464 string2, look at the last character in string1. */
3465 #define WORDCHAR_P(d) \
3466 (SYNTAX ((d) == end1 ? *string2 \
3467 : (d) == string2 - 1 ? *(end1 - 1) : *(d)) \
3470 /* Disabled due to a compiler bug -- see comment at case wordbound */
3472 /* Test if the character before D and the one at D differ with respect
3473 to being word-constituent. */
3474 #define AT_WORD_BOUNDARY(d) \
3475 (AT_STRINGS_BEG (d) || AT_STRINGS_END (d) \
3476 || WORDCHAR_P (d - 1) != WORDCHAR_P (d))
3479 /* Free everything we malloc. */
3480 #ifdef MATCH_MAY_ALLOCATE
3481 #define FREE_VAR(var) if (var) { REGEX_FREE (var); var = NULL; } else
3482 #define FREE_VARIABLES() \
3484 REGEX_FREE_STACK (fail_stack.stack); \
3485 FREE_VAR (regstart); \
3486 FREE_VAR (regend); \
3487 FREE_VAR (old_regstart); \
3488 FREE_VAR (old_regend); \
3489 FREE_VAR (best_regstart); \
3490 FREE_VAR (best_regend); \
3491 FREE_VAR (reg_info); \
3492 FREE_VAR (reg_dummy); \
3493 FREE_VAR (reg_info_dummy); \
3496 #define FREE_VARIABLES() ((void)0) /* Do nothing! But inhibit gcc warning. */
3497 #endif /* not MATCH_MAY_ALLOCATE */
3499 /* These values must meet several constraints. They must not be valid
3500 register values; since we have a limit of 255 registers (because
3501 we use only one byte in the pattern for the register number), we can
3502 use numbers larger than 255. They must differ by 1, because of
3503 NUM_FAILURE_ITEMS above. And the value for the lowest register must
3504 be larger than the value for the highest register, so we do not try
3505 to actually save any registers when none are active. */
3506 #define NO_HIGHEST_ACTIVE_REG (1 << BYTEWIDTH)
3507 #define NO_LOWEST_ACTIVE_REG (NO_HIGHEST_ACTIVE_REG + 1)
3509 /* Matching routines. */
3511 #ifndef emacs /* Emacs never uses this. */
3512 /* re_match is like re_match_2 except it takes only a single string. */
3515 re_match (bufp, string, size, pos, regs)
3516 struct re_pattern_buffer *bufp;
3519 struct re_registers *regs;
3521 int result = re_match_2_internal (bufp, NULL, 0, string, size,
3526 #endif /* not emacs */
3529 /* re_match_2 matches the compiled pattern in BUFP against the
3530 the (virtual) concatenation of STRING1 and STRING2 (of length SIZE1
3531 and SIZE2, respectively). We start matching at POS, and stop
3534 If REGS is non-null and the `no_sub' field of BUFP is nonzero, we
3535 store offsets for the substring each group matched in REGS. See the
3536 documentation for exactly how many groups we fill.
3538 We return -1 if no match, -2 if an internal error (such as the
3539 failure stack overflowing). Otherwise, we return the length of the
3540 matched substring. */
3543 re_match_2 (bufp, string1, size1, string2, size2, pos, regs, stop)
3544 struct re_pattern_buffer *bufp;
3545 const char *string1, *string2;
3548 struct re_registers *regs;
3551 int result = re_match_2_internal (bufp, string1, size1, string2, size2,
3557 /* This is a separate function so that we can force an alloca cleanup
3560 re_match_2_internal (bufp, string1, size1, string2, size2, pos, regs, stop)
3561 struct re_pattern_buffer *bufp;
3562 const char *string1, *string2;
3565 struct re_registers *regs;
3568 /* General temporaries. */
3572 /* Just past the end of the corresponding string. */
3573 const char *end1, *end2;
3575 /* Pointers into string1 and string2, just past the last characters in
3576 each to consider matching. */
3577 const char *end_match_1, *end_match_2;
3579 /* Where we are in the data, and the end of the current string. */
3580 const char *d, *dend;
3582 /* Where we are in the pattern, and the end of the pattern. */
3583 unsigned char *p = bufp->buffer;
3584 register unsigned char *pend = p + bufp->used;
3586 /* Mark the opcode just after a start_memory, so we can test for an
3587 empty subpattern when we get to the stop_memory. */
3588 unsigned char *just_past_start_mem = 0;
3590 /* We use this to map every character in the string. */
3591 RE_TRANSLATE_TYPE translate = bufp->translate;
3593 /* Failure point stack. Each place that can handle a failure further
3594 down the line pushes a failure point on this stack. It consists of
3595 restart, regend, and reg_info for all registers corresponding to
3596 the subexpressions we're currently inside, plus the number of such
3597 registers, and, finally, two char *'s. The first char * is where
3598 to resume scanning the pattern; the second one is where to resume
3599 scanning the strings. If the latter is zero, the failure point is
3600 a ``dummy''; if a failure happens and the failure point is a dummy,
3601 it gets discarded and the next next one is tried. */
3602 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
3603 fail_stack_type fail_stack;
3606 static unsigned failure_id = 0;
3607 unsigned nfailure_points_pushed = 0, nfailure_points_popped = 0;
3610 /* This holds the pointer to the failure stack, when
3611 it is allocated relocatably. */
3612 fail_stack_elt_t *failure_stack_ptr;
3614 /* We fill all the registers internally, independent of what we
3615 return, for use in backreferences. The number here includes
3616 an element for register zero. */
3617 unsigned num_regs = bufp->re_nsub + 1;
3619 /* The currently active registers. */
3620 unsigned lowest_active_reg = NO_LOWEST_ACTIVE_REG;
3621 unsigned highest_active_reg = NO_HIGHEST_ACTIVE_REG;
3623 /* Information on the contents of registers. These are pointers into
3624 the input strings; they record just what was matched (on this
3625 attempt) by a subexpression part of the pattern, that is, the
3626 regnum-th regstart pointer points to where in the pattern we began
3627 matching and the regnum-th regend points to right after where we
3628 stopped matching the regnum-th subexpression. (The zeroth register
3629 keeps track of what the whole pattern matches.) */
3630 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
3631 const char **regstart, **regend;
3634 /* If a group that's operated upon by a repetition operator fails to
3635 match anything, then the register for its start will need to be
3636 restored because it will have been set to wherever in the string we
3637 are when we last see its open-group operator. Similarly for a
3639 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
3640 const char **old_regstart, **old_regend;
3643 /* The is_active field of reg_info helps us keep track of which (possibly
3644 nested) subexpressions we are currently in. The matched_something
3645 field of reg_info[reg_num] helps us tell whether or not we have
3646 matched any of the pattern so far this time through the reg_num-th
3647 subexpression. These two fields get reset each time through any
3648 loop their register is in. */
3649 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
3650 register_info_type *reg_info;
3653 /* The following record the register info as found in the above
3654 variables when we find a match better than any we've seen before.
3655 This happens as we backtrack through the failure points, which in
3656 turn happens only if we have not yet matched the entire string. */
3657 unsigned best_regs_set = false;
3658 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
3659 const char **best_regstart, **best_regend;
3662 /* Logically, this is `best_regend[0]'. But we don't want to have to
3663 allocate space for that if we're not allocating space for anything
3664 else (see below). Also, we never need info about register 0 for
3665 any of the other register vectors, and it seems rather a kludge to
3666 treat `best_regend' differently than the rest. So we keep track of
3667 the end of the best match so far in a separate variable. We
3668 initialize this to NULL so that when we backtrack the first time
3669 and need to test it, it's not garbage. */
3670 const char *match_end = NULL;
3672 /* This helps SET_REGS_MATCHED avoid doing redundant work. */
3673 int set_regs_matched_done = 0;
3675 /* Used when we pop values we don't care about. */
3676 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
3677 const char **reg_dummy;
3678 register_info_type *reg_info_dummy;
3682 /* Counts the total number of registers pushed. */
3683 unsigned num_regs_pushed = 0;
3686 DEBUG_PRINT1 ("\n\nEntering re_match_2.\n");
3690 #ifdef MATCH_MAY_ALLOCATE
3691 /* Do not bother to initialize all the register variables if there are
3692 no groups in the pattern, as it takes a fair amount of time. If
3693 there are groups, we include space for register 0 (the whole
3694 pattern), even though we never use it, since it simplifies the
3695 array indexing. We should fix this. */
3698 regstart = REGEX_TALLOC (num_regs, const char *);
3699 regend = REGEX_TALLOC (num_regs, const char *);
3700 old_regstart = REGEX_TALLOC (num_regs, const char *);
3701 old_regend = REGEX_TALLOC (num_regs, const char *);
3702 best_regstart = REGEX_TALLOC (num_regs, const char *);
3703 best_regend = REGEX_TALLOC (num_regs, const char *);
3704 reg_info = REGEX_TALLOC (num_regs, register_info_type);
3705 reg_dummy = REGEX_TALLOC (num_regs, const char *);
3706 reg_info_dummy = REGEX_TALLOC (num_regs, register_info_type);
3708 if (!(regstart && regend && old_regstart && old_regend && reg_info
3709 && best_regstart && best_regend && reg_dummy && reg_info_dummy))
3717 /* We must initialize all our variables to NULL, so that
3718 `FREE_VARIABLES' doesn't try to free them. */
3719 regstart = regend = old_regstart = old_regend = best_regstart
3720 = best_regend = reg_dummy = NULL;
3721 reg_info = reg_info_dummy = (register_info_type *) NULL;
3723 #endif /* MATCH_MAY_ALLOCATE */
3725 /* The starting position is bogus. */
3726 if (pos < 0 || pos > size1 + size2)
3732 /* Initialize subexpression text positions to -1 to mark ones that no
3733 start_memory/stop_memory has been seen for. Also initialize the
3734 register information struct. */
3735 for (mcnt = 1; mcnt < num_regs; mcnt++)
3737 regstart[mcnt] = regend[mcnt]
3738 = old_regstart[mcnt] = old_regend[mcnt] = REG_UNSET_VALUE;
3740 REG_MATCH_NULL_STRING_P (reg_info[mcnt]) = MATCH_NULL_UNSET_VALUE;
3741 IS_ACTIVE (reg_info[mcnt]) = 0;
3742 MATCHED_SOMETHING (reg_info[mcnt]) = 0;
3743 EVER_MATCHED_SOMETHING (reg_info[mcnt]) = 0;
3746 /* We move `string1' into `string2' if the latter's empty -- but not if
3747 `string1' is null. */
3748 if (size2 == 0 && string1 != NULL)
3755 end1 = string1 + size1;
3756 end2 = string2 + size2;
3758 /* Compute where to stop matching, within the two strings. */
3761 end_match_1 = string1 + stop;
3762 end_match_2 = string2;
3767 end_match_2 = string2 + stop - size1;
3770 /* `p' scans through the pattern as `d' scans through the data.
3771 `dend' is the end of the input string that `d' points within. `d'
3772 is advanced into the following input string whenever necessary, but
3773 this happens before fetching; therefore, at the beginning of the
3774 loop, `d' can be pointing at the end of a string, but it cannot
3776 if (size1 > 0 && pos <= size1)
3783 d = string2 + pos - size1;
3787 DEBUG_PRINT1 ("The compiled pattern is: ");
3788 DEBUG_PRINT_COMPILED_PATTERN (bufp, p, pend);
3789 DEBUG_PRINT1 ("The string to match is: `");
3790 DEBUG_PRINT_DOUBLE_STRING (d, string1, size1, string2, size2);
3791 DEBUG_PRINT1 ("'\n");
3793 /* This loops over pattern commands. It exits by returning from the
3794 function if the match is complete, or it drops through if the match
3795 fails at this starting point in the input data. */
3798 DEBUG_PRINT2 ("\n0x%x: ", p);
3801 { /* End of pattern means we might have succeeded. */
3802 DEBUG_PRINT1 ("end of pattern ... ");
3804 /* If we haven't matched the entire string, and we want the
3805 longest match, try backtracking. */
3806 if (d != end_match_2)
3808 /* 1 if this match ends in the same string (string1 or string2)
3809 as the best previous match. */
3810 boolean same_str_p = (FIRST_STRING_P (match_end)
3811 == MATCHING_IN_FIRST_STRING);
3812 /* 1 if this match is the best seen so far. */
3813 boolean best_match_p;
3815 /* AIX compiler got confused when this was combined
3816 with the previous declaration. */
3818 best_match_p = d > match_end;
3820 best_match_p = !MATCHING_IN_FIRST_STRING;
3822 DEBUG_PRINT1 ("backtracking.\n");
3824 if (!FAIL_STACK_EMPTY ())
3825 { /* More failure points to try. */
3827 /* If exceeds best match so far, save it. */
3828 if (!best_regs_set || best_match_p)
3830 best_regs_set = true;
3833 DEBUG_PRINT1 ("\nSAVING match as best so far.\n");
3835 for (mcnt = 1; mcnt < num_regs; mcnt++)
3837 best_regstart[mcnt] = regstart[mcnt];
3838 best_regend[mcnt] = regend[mcnt];
3844 /* If no failure points, don't restore garbage. And if
3845 last match is real best match, don't restore second
3847 else if (best_regs_set && !best_match_p)
3850 /* Restore best match. It may happen that `dend ==
3851 end_match_1' while the restored d is in string2.
3852 For example, the pattern `x.*y.*z' against the
3853 strings `x-' and `y-z-', if the two strings are
3854 not consecutive in memory. */
3855 DEBUG_PRINT1 ("Restoring best registers.\n");
3858 dend = ((d >= string1 && d <= end1)
3859 ? end_match_1 : end_match_2);
3861 for (mcnt = 1; mcnt < num_regs; mcnt++)
3863 regstart[mcnt] = best_regstart[mcnt];
3864 regend[mcnt] = best_regend[mcnt];
3867 } /* d != end_match_2 */
3870 DEBUG_PRINT1 ("Accepting match.\n");
3872 /* If caller wants register contents data back, do it. */
3873 if (regs && !bufp->no_sub)
3875 /* Have the register data arrays been allocated? */
3876 if (bufp->regs_allocated == REGS_UNALLOCATED)
3877 { /* No. So allocate them with malloc. We need one
3878 extra element beyond `num_regs' for the `-1' marker
3880 regs->num_regs = MAX (RE_NREGS, num_regs + 1);
3881 regs->start = TALLOC (regs->num_regs, regoff_t);
3882 regs->end = TALLOC (regs->num_regs, regoff_t);
3883 if (regs->start == NULL || regs->end == NULL)
3888 bufp->regs_allocated = REGS_REALLOCATE;
3890 else if (bufp->regs_allocated == REGS_REALLOCATE)
3891 { /* Yes. If we need more elements than were already
3892 allocated, reallocate them. If we need fewer, just
3894 if (regs->num_regs < num_regs + 1)
3896 regs->num_regs = num_regs + 1;
3897 RETALLOC (regs->start, regs->num_regs, regoff_t);
3898 RETALLOC (regs->end, regs->num_regs, regoff_t);
3899 if (regs->start == NULL || regs->end == NULL)
3908 /* These braces fend off a "empty body in an else-statement"
3909 warning under GCC when assert expands to nothing. */
3910 assert (bufp->regs_allocated == REGS_FIXED);
3913 /* Convert the pointer data in `regstart' and `regend' to
3914 indices. Register zero has to be set differently,
3915 since we haven't kept track of any info for it. */
3916 if (regs->num_regs > 0)
3918 regs->start[0] = pos;
3919 regs->end[0] = (MATCHING_IN_FIRST_STRING
3920 ? ((regoff_t) (d - string1))
3921 : ((regoff_t) (d - string2 + size1)));
3924 /* Go through the first `min (num_regs, regs->num_regs)'
3925 registers, since that is all we initialized. */
3926 for (mcnt = 1; mcnt < MIN (num_regs, regs->num_regs); mcnt++)
3928 if (REG_UNSET (regstart[mcnt]) || REG_UNSET (regend[mcnt]))
3929 regs->start[mcnt] = regs->end[mcnt] = -1;
3933 = (regoff_t) POINTER_TO_OFFSET (regstart[mcnt]);
3935 = (regoff_t) POINTER_TO_OFFSET (regend[mcnt]);
3939 /* If the regs structure we return has more elements than
3940 were in the pattern, set the extra elements to -1. If
3941 we (re)allocated the registers, this is the case,
3942 because we always allocate enough to have at least one
3944 for (mcnt = num_regs; mcnt < regs->num_regs; mcnt++)
3945 regs->start[mcnt] = regs->end[mcnt] = -1;
3946 } /* regs && !bufp->no_sub */
3948 DEBUG_PRINT4 ("%u failure points pushed, %u popped (%u remain).\n",
3949 nfailure_points_pushed, nfailure_points_popped,
3950 nfailure_points_pushed - nfailure_points_popped);
3951 DEBUG_PRINT2 ("%u registers pushed.\n", num_regs_pushed);
3953 mcnt = d - pos - (MATCHING_IN_FIRST_STRING
3957 DEBUG_PRINT2 ("Returning %d from re_match_2.\n", mcnt);
3963 /* Otherwise match next pattern command. */
3964 switch (SWITCH_ENUM_CAST ((re_opcode_t) *p++))
3966 /* Ignore these. Used to ignore the n of succeed_n's which
3967 currently have n == 0. */
3969 DEBUG_PRINT1 ("EXECUTING no_op.\n");
3973 DEBUG_PRINT1 ("EXECUTING succeed.\n");
3976 /* Match the next n pattern characters exactly. The following
3977 byte in the pattern defines n, and the n bytes after that
3978 are the characters to match. */
3981 DEBUG_PRINT2 ("EXECUTING exactn %d.\n", mcnt);
3983 /* This is written out as an if-else so we don't waste time
3984 testing `translate' inside the loop. */
3990 if ((unsigned char) translate[(unsigned char) *d++]
3991 != (unsigned char) *p++)
4001 if (*d++ != (char) *p++) goto fail;
4005 SET_REGS_MATCHED ();
4009 /* Match any character except possibly a newline or a null. */
4011 DEBUG_PRINT1 ("EXECUTING anychar.\n");
4015 if ((!(bufp->syntax & RE_DOT_NEWLINE) && TRANSLATE (*d) == '\n')
4016 || (bufp->syntax & RE_DOT_NOT_NULL && TRANSLATE (*d) == '\000'))
4019 SET_REGS_MATCHED ();
4020 DEBUG_PRINT2 (" Matched `%d'.\n", *d);
4028 register unsigned char c;
4029 boolean not = (re_opcode_t) *(p - 1) == charset_not;
4031 DEBUG_PRINT2 ("EXECUTING charset%s.\n", not ? "_not" : "");
4034 c = TRANSLATE (*d); /* The character to match. */
4036 /* Cast to `unsigned' instead of `unsigned char' in case the
4037 bit list is a full 32 bytes long. */
4038 if (c < (unsigned) (*p * BYTEWIDTH)
4039 && p[1 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH)))
4044 if (!not) goto fail;
4046 SET_REGS_MATCHED ();
4052 /* The beginning of a group is represented by start_memory.
4053 The arguments are the register number in the next byte, and the
4054 number of groups inner to this one in the next. The text
4055 matched within the group is recorded (in the internal
4056 registers data structure) under the register number. */
4058 DEBUG_PRINT3 ("EXECUTING start_memory %d (%d):\n", *p, p[1]);
4060 /* Find out if this group can match the empty string. */
4061 p1 = p; /* To send to group_match_null_string_p. */
4063 if (REG_MATCH_NULL_STRING_P (reg_info[*p]) == MATCH_NULL_UNSET_VALUE)
4064 REG_MATCH_NULL_STRING_P (reg_info[*p])
4065 = group_match_null_string_p (&p1, pend, reg_info);
4067 /* Save the position in the string where we were the last time
4068 we were at this open-group operator in case the group is
4069 operated upon by a repetition operator, e.g., with `(a*)*b'
4070 against `ab'; then we want to ignore where we are now in
4071 the string in case this attempt to match fails. */
4072 old_regstart[*p] = REG_MATCH_NULL_STRING_P (reg_info[*p])
4073 ? REG_UNSET (regstart[*p]) ? d : regstart[*p]
4075 DEBUG_PRINT2 (" old_regstart: %d\n",
4076 POINTER_TO_OFFSET (old_regstart[*p]));
4079 DEBUG_PRINT2 (" regstart: %d\n", POINTER_TO_OFFSET (regstart[*p]));
4081 IS_ACTIVE (reg_info[*p]) = 1;
4082 MATCHED_SOMETHING (reg_info[*p]) = 0;
4084 /* Clear this whenever we change the register activity status. */
4085 set_regs_matched_done = 0;
4087 /* This is the new highest active register. */
4088 highest_active_reg = *p;
4090 /* If nothing was active before, this is the new lowest active
4092 if (lowest_active_reg == NO_LOWEST_ACTIVE_REG)
4093 lowest_active_reg = *p;
4095 /* Move past the register number and inner group count. */
4097 just_past_start_mem = p;
4102 /* The stop_memory opcode represents the end of a group. Its
4103 arguments are the same as start_memory's: the register
4104 number, and the number of inner groups. */
4106 DEBUG_PRINT3 ("EXECUTING stop_memory %d (%d):\n", *p, p[1]);
4108 /* We need to save the string position the last time we were at
4109 this close-group operator in case the group is operated
4110 upon by a repetition operator, e.g., with `((a*)*(b*)*)*'
4111 against `aba'; then we want to ignore where we are now in
4112 the string in case this attempt to match fails. */
4113 old_regend[*p] = REG_MATCH_NULL_STRING_P (reg_info[*p])
4114 ? REG_UNSET (regend[*p]) ? d : regend[*p]
4116 DEBUG_PRINT2 (" old_regend: %d\n",
4117 POINTER_TO_OFFSET (old_regend[*p]));
4120 DEBUG_PRINT2 (" regend: %d\n", POINTER_TO_OFFSET (regend[*p]));
4122 /* This register isn't active anymore. */
4123 IS_ACTIVE (reg_info[*p]) = 0;
4125 /* Clear this whenever we change the register activity status. */
4126 set_regs_matched_done = 0;
4128 /* If this was the only register active, nothing is active
4130 if (lowest_active_reg == highest_active_reg)
4132 lowest_active_reg = NO_LOWEST_ACTIVE_REG;
4133 highest_active_reg = NO_HIGHEST_ACTIVE_REG;
4136 { /* We must scan for the new highest active register, since
4137 it isn't necessarily one less than now: consider
4138 (a(b)c(d(e)f)g). When group 3 ends, after the f), the
4139 new highest active register is 1. */
4140 unsigned char r = *p - 1;
4141 while (r > 0 && !IS_ACTIVE (reg_info[r]))
4144 /* If we end up at register zero, that means that we saved
4145 the registers as the result of an `on_failure_jump', not
4146 a `start_memory', and we jumped to past the innermost
4147 `stop_memory'. For example, in ((.)*) we save
4148 registers 1 and 2 as a result of the *, but when we pop
4149 back to the second ), we are at the stop_memory 1.
4150 Thus, nothing is active. */
4153 lowest_active_reg = NO_LOWEST_ACTIVE_REG;
4154 highest_active_reg = NO_HIGHEST_ACTIVE_REG;
4157 highest_active_reg = r;
4160 /* If just failed to match something this time around with a
4161 group that's operated on by a repetition operator, try to
4162 force exit from the ``loop'', and restore the register
4163 information for this group that we had before trying this
4165 if ((!MATCHED_SOMETHING (reg_info[*p])
4166 || just_past_start_mem == p - 1)
4169 boolean is_a_jump_n = false;
4173 switch ((re_opcode_t) *p1++)
4177 case pop_failure_jump:
4178 case maybe_pop_jump:
4180 case dummy_failure_jump:
4181 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4191 /* If the next operation is a jump backwards in the pattern
4192 to an on_failure_jump right before the start_memory
4193 corresponding to this stop_memory, exit from the loop
4194 by forcing a failure after pushing on the stack the
4195 on_failure_jump's jump in the pattern, and d. */
4196 if (mcnt < 0 && (re_opcode_t) *p1 == on_failure_jump
4197 && (re_opcode_t) p1[3] == start_memory && p1[4] == *p)
4199 /* If this group ever matched anything, then restore
4200 what its registers were before trying this last
4201 failed match, e.g., with `(a*)*b' against `ab' for
4202 regstart[1], and, e.g., with `((a*)*(b*)*)*'
4203 against `aba' for regend[3].
4205 Also restore the registers for inner groups for,
4206 e.g., `((a*)(b*))*' against `aba' (register 3 would
4207 otherwise get trashed). */
4209 if (EVER_MATCHED_SOMETHING (reg_info[*p]))
4213 EVER_MATCHED_SOMETHING (reg_info[*p]) = 0;
4215 /* Restore this and inner groups' (if any) registers. */
4216 for (r = *p; r < *p + *(p + 1); r++)
4218 regstart[r] = old_regstart[r];
4220 /* xx why this test? */
4221 if (old_regend[r] >= regstart[r])
4222 regend[r] = old_regend[r];
4226 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4227 PUSH_FAILURE_POINT (p1 + mcnt, d, -2);
4233 /* Move past the register number and the inner group count. */
4238 /* \<digit> has been turned into a `duplicate' command which is
4239 followed by the numeric value of <digit> as the register number. */
4242 register const char *d2, *dend2;
4243 int regno = *p++; /* Get which register to match against. */
4244 DEBUG_PRINT2 ("EXECUTING duplicate %d.\n", regno);
4246 /* Can't back reference a group which we've never matched. */
4247 if (REG_UNSET (regstart[regno]) || REG_UNSET (regend[regno]))
4250 /* Where in input to try to start matching. */
4251 d2 = regstart[regno];
4253 /* Where to stop matching; if both the place to start and
4254 the place to stop matching are in the same string, then
4255 set to the place to stop, otherwise, for now have to use
4256 the end of the first string. */
4258 dend2 = ((FIRST_STRING_P (regstart[regno])
4259 == FIRST_STRING_P (regend[regno]))
4260 ? regend[regno] : end_match_1);
4263 /* If necessary, advance to next segment in register
4267 if (dend2 == end_match_2) break;
4268 if (dend2 == regend[regno]) break;
4270 /* End of string1 => advance to string2. */
4272 dend2 = regend[regno];
4274 /* At end of register contents => success */
4275 if (d2 == dend2) break;
4277 /* If necessary, advance to next segment in data. */
4280 /* How many characters left in this segment to match. */
4283 /* Want how many consecutive characters we can match in
4284 one shot, so, if necessary, adjust the count. */
4285 if (mcnt > dend2 - d2)
4288 /* Compare that many; failure if mismatch, else move
4291 ? bcmp_translate (d, d2, mcnt, translate)
4292 : bcmp (d, d2, mcnt))
4294 d += mcnt, d2 += mcnt;
4296 /* Do this because we've match some characters. */
4297 SET_REGS_MATCHED ();
4303 /* begline matches the empty string at the beginning of the string
4304 (unless `not_bol' is set in `bufp'), and, if
4305 `newline_anchor' is set, after newlines. */
4307 DEBUG_PRINT1 ("EXECUTING begline.\n");
4309 if (AT_STRINGS_BEG (d))
4311 if (!bufp->not_bol) break;
4313 else if (d[-1] == '\n' && bufp->newline_anchor)
4317 /* In all other cases, we fail. */
4321 /* endline is the dual of begline. */
4323 DEBUG_PRINT1 ("EXECUTING endline.\n");
4325 if (AT_STRINGS_END (d))
4327 if (!bufp->not_eol) break;
4330 /* We have to ``prefetch'' the next character. */
4331 else if ((d == end1 ? *string2 : *d) == '\n'
4332 && bufp->newline_anchor)
4339 /* Match at the very beginning of the data. */
4341 DEBUG_PRINT1 ("EXECUTING begbuf.\n");
4342 if (AT_STRINGS_BEG (d))
4347 /* Match at the very end of the data. */
4349 DEBUG_PRINT1 ("EXECUTING endbuf.\n");
4350 if (AT_STRINGS_END (d))
4355 /* on_failure_keep_string_jump is used to optimize `.*\n'. It
4356 pushes NULL as the value for the string on the stack. Then
4357 `pop_failure_point' will keep the current value for the
4358 string, instead of restoring it. To see why, consider
4359 matching `foo\nbar' against `.*\n'. The .* matches the foo;
4360 then the . fails against the \n. But the next thing we want
4361 to do is match the \n against the \n; if we restored the
4362 string value, we would be back at the foo.
4364 Because this is used only in specific cases, we don't need to
4365 check all the things that `on_failure_jump' does, to make
4366 sure the right things get saved on the stack. Hence we don't
4367 share its code. The only reason to push anything on the
4368 stack at all is that otherwise we would have to change
4369 `anychar's code to do something besides goto fail in this
4370 case; that seems worse than this. */
4371 case on_failure_keep_string_jump:
4372 DEBUG_PRINT1 ("EXECUTING on_failure_keep_string_jump");
4374 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4375 DEBUG_PRINT3 (" %d (to 0x%x):\n", mcnt, p + mcnt);
4377 PUSH_FAILURE_POINT (p + mcnt, NULL, -2);
4381 /* Uses of on_failure_jump:
4383 Each alternative starts with an on_failure_jump that points
4384 to the beginning of the next alternative. Each alternative
4385 except the last ends with a jump that in effect jumps past
4386 the rest of the alternatives. (They really jump to the
4387 ending jump of the following alternative, because tensioning
4388 these jumps is a hassle.)
4390 Repeats start with an on_failure_jump that points past both
4391 the repetition text and either the following jump or
4392 pop_failure_jump back to this on_failure_jump. */
4393 case on_failure_jump:
4395 DEBUG_PRINT1 ("EXECUTING on_failure_jump");
4397 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4398 DEBUG_PRINT3 (" %d (to 0x%x)", mcnt, p + mcnt);
4400 /* If this on_failure_jump comes right before a group (i.e.,
4401 the original * applied to a group), save the information
4402 for that group and all inner ones, so that if we fail back
4403 to this point, the group's information will be correct.
4404 For example, in \(a*\)*\1, we need the preceding group,
4405 and in \(zz\(a*\)b*\)\2, we need the inner group. */
4407 /* We can't use `p' to check ahead because we push
4408 a failure point to `p + mcnt' after we do this. */
4411 /* We need to skip no_op's before we look for the
4412 start_memory in case this on_failure_jump is happening as
4413 the result of a completed succeed_n, as in \(a\)\{1,3\}b\1
4415 while (p1 < pend && (re_opcode_t) *p1 == no_op)
4418 if (p1 < pend && (re_opcode_t) *p1 == start_memory)
4420 /* We have a new highest active register now. This will
4421 get reset at the start_memory we are about to get to,
4422 but we will have saved all the registers relevant to
4423 this repetition op, as described above. */
4424 highest_active_reg = *(p1 + 1) + *(p1 + 2);
4425 if (lowest_active_reg == NO_LOWEST_ACTIVE_REG)
4426 lowest_active_reg = *(p1 + 1);
4429 DEBUG_PRINT1 (":\n");
4430 PUSH_FAILURE_POINT (p + mcnt, d, -2);
4434 /* A smart repeat ends with `maybe_pop_jump'.
4435 We change it to either `pop_failure_jump' or `jump'. */
4436 case maybe_pop_jump:
4437 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4438 DEBUG_PRINT2 ("EXECUTING maybe_pop_jump %d.\n", mcnt);
4440 register unsigned char *p2 = p;
4442 /* Compare the beginning of the repeat with what in the
4443 pattern follows its end. If we can establish that there
4444 is nothing that they would both match, i.e., that we
4445 would have to backtrack because of (as in, e.g., `a*a')
4446 then we can change to pop_failure_jump, because we'll
4447 never have to backtrack.
4449 This is not true in the case of alternatives: in
4450 `(a|ab)*' we do need to backtrack to the `ab' alternative
4451 (e.g., if the string was `ab'). But instead of trying to
4452 detect that here, the alternative has put on a dummy
4453 failure point which is what we will end up popping. */
4455 /* Skip over open/close-group commands.
4456 If what follows this loop is a ...+ construct,
4457 look at what begins its body, since we will have to
4458 match at least one of that. */
4462 && ((re_opcode_t) *p2 == stop_memory
4463 || (re_opcode_t) *p2 == start_memory))
4465 else if (p2 + 6 < pend
4466 && (re_opcode_t) *p2 == dummy_failure_jump)
4473 /* p1[0] ... p1[2] are the `on_failure_jump' corresponding
4474 to the `maybe_finalize_jump' of this case. Examine what
4477 /* If we're at the end of the pattern, we can change. */
4480 /* Consider what happens when matching ":\(.*\)"
4481 against ":/". I don't really understand this code
4483 p[-3] = (unsigned char) pop_failure_jump;
4485 (" End of pattern: change to `pop_failure_jump'.\n");
4488 else if ((re_opcode_t) *p2 == exactn
4489 || (bufp->newline_anchor && (re_opcode_t) *p2 == endline))
4491 register unsigned char c
4492 = *p2 == (unsigned char) endline ? '\n' : p2[2];
4494 if ((re_opcode_t) p1[3] == exactn && p1[5] != c)
4496 p[-3] = (unsigned char) pop_failure_jump;
4497 DEBUG_PRINT3 (" %c != %c => pop_failure_jump.\n",
4501 else if ((re_opcode_t) p1[3] == charset
4502 || (re_opcode_t) p1[3] == charset_not)
4504 int not = (re_opcode_t) p1[3] == charset_not;
4506 if (c < (unsigned char) (p1[4] * BYTEWIDTH)
4507 && p1[5 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH)))
4510 /* `not' is equal to 1 if c would match, which means
4511 that we can't change to pop_failure_jump. */
4514 p[-3] = (unsigned char) pop_failure_jump;
4515 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
4519 else if ((re_opcode_t) *p2 == charset)
4522 register unsigned char c
4523 = *p2 == (unsigned char) endline ? '\n' : p2[2];
4526 if ((re_opcode_t) p1[3] == exactn
4527 && ! ((int) p2[1] * BYTEWIDTH > (int) p1[5]
4528 && (p2[2 + p1[5] / BYTEWIDTH]
4529 & (1 << (p1[5] % BYTEWIDTH)))))
4531 p[-3] = (unsigned char) pop_failure_jump;
4532 DEBUG_PRINT3 (" %c != %c => pop_failure_jump.\n",
4536 else if ((re_opcode_t) p1[3] == charset_not)
4539 /* We win if the charset_not inside the loop
4540 lists every character listed in the charset after. */
4541 for (idx = 0; idx < (int) p2[1]; idx++)
4542 if (! (p2[2 + idx] == 0
4543 || (idx < (int) p1[4]
4544 && ((p2[2 + idx] & ~ p1[5 + idx]) == 0))))
4549 p[-3] = (unsigned char) pop_failure_jump;
4550 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
4553 else if ((re_opcode_t) p1[3] == charset)
4556 /* We win if the charset inside the loop
4557 has no overlap with the one after the loop. */
4559 idx < (int) p2[1] && idx < (int) p1[4];
4561 if ((p2[2 + idx] & p1[5 + idx]) != 0)
4564 if (idx == p2[1] || idx == p1[4])
4566 p[-3] = (unsigned char) pop_failure_jump;
4567 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
4572 p -= 2; /* Point at relative address again. */
4573 if ((re_opcode_t) p[-1] != pop_failure_jump)
4575 p[-1] = (unsigned char) jump;
4576 DEBUG_PRINT1 (" Match => jump.\n");
4577 goto unconditional_jump;
4579 /* Note fall through. */
4582 /* The end of a simple repeat has a pop_failure_jump back to
4583 its matching on_failure_jump, where the latter will push a
4584 failure point. The pop_failure_jump takes off failure
4585 points put on by this pop_failure_jump's matching
4586 on_failure_jump; we got through the pattern to here from the
4587 matching on_failure_jump, so didn't fail. */
4588 case pop_failure_jump:
4590 /* We need to pass separate storage for the lowest and
4591 highest registers, even though we don't care about the
4592 actual values. Otherwise, we will restore only one
4593 register from the stack, since lowest will == highest in
4594 `pop_failure_point'. */
4595 unsigned dummy_low_reg, dummy_high_reg;
4596 unsigned char *pdummy;
4599 DEBUG_PRINT1 ("EXECUTING pop_failure_jump.\n");
4600 POP_FAILURE_POINT (sdummy, pdummy,
4601 dummy_low_reg, dummy_high_reg,
4602 reg_dummy, reg_dummy, reg_info_dummy);
4604 /* Note fall through. */
4607 /* Unconditionally jump (without popping any failure points). */
4610 EXTRACT_NUMBER_AND_INCR (mcnt, p); /* Get the amount to jump. */
4611 DEBUG_PRINT2 ("EXECUTING jump %d ", mcnt);
4612 p += mcnt; /* Do the jump. */
4613 DEBUG_PRINT2 ("(to 0x%x).\n", p);
4617 /* We need this opcode so we can detect where alternatives end
4618 in `group_match_null_string_p' et al. */
4620 DEBUG_PRINT1 ("EXECUTING jump_past_alt.\n");
4621 goto unconditional_jump;
4624 /* Normally, the on_failure_jump pushes a failure point, which
4625 then gets popped at pop_failure_jump. We will end up at
4626 pop_failure_jump, also, and with a pattern of, say, `a+', we
4627 are skipping over the on_failure_jump, so we have to push
4628 something meaningless for pop_failure_jump to pop. */
4629 case dummy_failure_jump:
4630 DEBUG_PRINT1 ("EXECUTING dummy_failure_jump.\n");
4631 /* It doesn't matter what we push for the string here. What
4632 the code at `fail' tests is the value for the pattern. */
4633 PUSH_FAILURE_POINT (0, 0, -2);
4634 goto unconditional_jump;
4637 /* At the end of an alternative, we need to push a dummy failure
4638 point in case we are followed by a `pop_failure_jump', because
4639 we don't want the failure point for the alternative to be
4640 popped. For example, matching `(a|ab)*' against `aab'
4641 requires that we match the `ab' alternative. */
4642 case push_dummy_failure:
4643 DEBUG_PRINT1 ("EXECUTING push_dummy_failure.\n");
4644 /* See comments just above at `dummy_failure_jump' about the
4646 PUSH_FAILURE_POINT (0, 0, -2);
4649 /* Have to succeed matching what follows at least n times.
4650 After that, handle like `on_failure_jump'. */
4652 EXTRACT_NUMBER (mcnt, p + 2);
4653 DEBUG_PRINT2 ("EXECUTING succeed_n %d.\n", mcnt);
4656 /* Originally, this is how many times we HAVE to succeed. */
4661 STORE_NUMBER_AND_INCR (p, mcnt);
4662 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p, mcnt);
4666 DEBUG_PRINT2 (" Setting two bytes from 0x%x to no_op.\n", p+2);
4667 p[2] = (unsigned char) no_op;
4668 p[3] = (unsigned char) no_op;
4674 EXTRACT_NUMBER (mcnt, p + 2);
4675 DEBUG_PRINT2 ("EXECUTING jump_n %d.\n", mcnt);
4677 /* Originally, this is how many times we CAN jump. */
4681 STORE_NUMBER (p + 2, mcnt);
4682 goto unconditional_jump;
4684 /* If don't have to jump any more, skip over the rest of command. */
4691 DEBUG_PRINT1 ("EXECUTING set_number_at.\n");
4693 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4695 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4696 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p1, mcnt);
4697 STORE_NUMBER (p1, mcnt);
4702 /* The DEC Alpha C compiler 3.x generates incorrect code for the
4703 test WORDCHAR_P (d - 1) != WORDCHAR_P (d) in the expansion of
4704 AT_WORD_BOUNDARY, so this code is disabled. Expanding the
4705 macro and introducing temporary variables works around the bug. */
4708 DEBUG_PRINT1 ("EXECUTING wordbound.\n");
4709 if (AT_WORD_BOUNDARY (d))
4714 DEBUG_PRINT1 ("EXECUTING notwordbound.\n");
4715 if (AT_WORD_BOUNDARY (d))
4721 boolean prevchar, thischar;
4723 DEBUG_PRINT1 ("EXECUTING wordbound.\n");
4724 if (AT_STRINGS_BEG (d) || AT_STRINGS_END (d))
4727 prevchar = WORDCHAR_P (d - 1);
4728 thischar = WORDCHAR_P (d);
4729 if (prevchar != thischar)
4736 boolean prevchar, thischar;
4738 DEBUG_PRINT1 ("EXECUTING notwordbound.\n");
4739 if (AT_STRINGS_BEG (d) || AT_STRINGS_END (d))
4742 prevchar = WORDCHAR_P (d - 1);
4743 thischar = WORDCHAR_P (d);
4744 if (prevchar != thischar)
4751 DEBUG_PRINT1 ("EXECUTING wordbeg.\n");
4752 if (WORDCHAR_P (d) && (AT_STRINGS_BEG (d) || !WORDCHAR_P (d - 1)))
4757 DEBUG_PRINT1 ("EXECUTING wordend.\n");
4758 if (!AT_STRINGS_BEG (d) && WORDCHAR_P (d - 1)
4759 && (!WORDCHAR_P (d) || AT_STRINGS_END (d)))
4765 DEBUG_PRINT1 ("EXECUTING before_dot.\n");
4766 if (PTR_CHAR_POS ((unsigned char *) d) >= PT)
4771 DEBUG_PRINT1 ("EXECUTING at_dot.\n");
4772 if (PTR_CHAR_POS ((unsigned char *) d) != PT)
4777 DEBUG_PRINT1 ("EXECUTING after_dot.\n");
4778 if (PTR_CHAR_POS ((unsigned char *) d) <= PT)
4783 DEBUG_PRINT2 ("EXECUTING syntaxspec %d.\n", mcnt);
4788 DEBUG_PRINT1 ("EXECUTING Emacs wordchar.\n");
4792 /* Can't use *d++ here; SYNTAX may be an unsafe macro. */
4794 if (SYNTAX (d[-1]) != (enum syntaxcode) mcnt)
4796 SET_REGS_MATCHED ();
4800 DEBUG_PRINT2 ("EXECUTING notsyntaxspec %d.\n", mcnt);
4802 goto matchnotsyntax;
4805 DEBUG_PRINT1 ("EXECUTING Emacs notwordchar.\n");
4809 /* Can't use *d++ here; SYNTAX may be an unsafe macro. */
4811 if (SYNTAX (d[-1]) == (enum syntaxcode) mcnt)
4813 SET_REGS_MATCHED ();
4816 #else /* not emacs */
4818 DEBUG_PRINT1 ("EXECUTING non-Emacs wordchar.\n");
4820 if (!WORDCHAR_P (d))
4822 SET_REGS_MATCHED ();
4827 DEBUG_PRINT1 ("EXECUTING non-Emacs notwordchar.\n");
4831 SET_REGS_MATCHED ();
4834 #endif /* not emacs */
4839 continue; /* Successfully executed one pattern command; keep going. */
4842 /* We goto here if a matching operation fails. */
4844 if (!FAIL_STACK_EMPTY ())
4845 { /* A restart point is known. Restore to that state. */
4846 DEBUG_PRINT1 ("\nFAIL:\n");
4847 POP_FAILURE_POINT (d, p,
4848 lowest_active_reg, highest_active_reg,
4849 regstart, regend, reg_info);
4851 /* If this failure point is a dummy, try the next one. */
4855 /* If we failed to the end of the pattern, don't examine *p. */
4859 boolean is_a_jump_n = false;
4861 /* If failed to a backwards jump that's part of a repetition
4862 loop, need to pop this failure point and use the next one. */
4863 switch ((re_opcode_t) *p)
4867 case maybe_pop_jump:
4868 case pop_failure_jump:
4871 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4874 if ((is_a_jump_n && (re_opcode_t) *p1 == succeed_n)
4876 && (re_opcode_t) *p1 == on_failure_jump))
4884 if (d >= string1 && d <= end1)
4888 break; /* Matching at this starting point really fails. */
4892 goto restore_best_regs;
4896 return -1; /* Failure to match. */
4899 /* Subroutine definitions for re_match_2. */
4902 /* We are passed P pointing to a register number after a start_memory.
4904 Return true if the pattern up to the corresponding stop_memory can
4905 match the empty string, and false otherwise.
4907 If we find the matching stop_memory, sets P to point to one past its number.
4908 Otherwise, sets P to an undefined byte less than or equal to END.
4910 We don't handle duplicates properly (yet). */
4913 group_match_null_string_p (p, end, reg_info)
4914 unsigned char **p, *end;
4915 register_info_type *reg_info;
4918 /* Point to after the args to the start_memory. */
4919 unsigned char *p1 = *p + 2;
4923 /* Skip over opcodes that can match nothing, and return true or
4924 false, as appropriate, when we get to one that can't, or to the
4925 matching stop_memory. */
4927 switch ((re_opcode_t) *p1)
4929 /* Could be either a loop or a series of alternatives. */
4930 case on_failure_jump:
4932 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4934 /* If the next operation is not a jump backwards in the
4939 /* Go through the on_failure_jumps of the alternatives,
4940 seeing if any of the alternatives cannot match nothing.
4941 The last alternative starts with only a jump,
4942 whereas the rest start with on_failure_jump and end
4943 with a jump, e.g., here is the pattern for `a|b|c':
4945 /on_failure_jump/0/6/exactn/1/a/jump_past_alt/0/6
4946 /on_failure_jump/0/6/exactn/1/b/jump_past_alt/0/3
4949 So, we have to first go through the first (n-1)
4950 alternatives and then deal with the last one separately. */
4953 /* Deal with the first (n-1) alternatives, which start
4954 with an on_failure_jump (see above) that jumps to right
4955 past a jump_past_alt. */
4957 while ((re_opcode_t) p1[mcnt-3] == jump_past_alt)
4959 /* `mcnt' holds how many bytes long the alternative
4960 is, including the ending `jump_past_alt' and
4963 if (!alt_match_null_string_p (p1, p1 + mcnt - 3,
4967 /* Move to right after this alternative, including the
4971 /* Break if it's the beginning of an n-th alternative
4972 that doesn't begin with an on_failure_jump. */
4973 if ((re_opcode_t) *p1 != on_failure_jump)
4976 /* Still have to check that it's not an n-th
4977 alternative that starts with an on_failure_jump. */
4979 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4980 if ((re_opcode_t) p1[mcnt-3] != jump_past_alt)
4982 /* Get to the beginning of the n-th alternative. */
4988 /* Deal with the last alternative: go back and get number
4989 of the `jump_past_alt' just before it. `mcnt' contains
4990 the length of the alternative. */
4991 EXTRACT_NUMBER (mcnt, p1 - 2);
4993 if (!alt_match_null_string_p (p1, p1 + mcnt, reg_info))
4996 p1 += mcnt; /* Get past the n-th alternative. */
5002 assert (p1[1] == **p);
5008 if (!common_op_match_null_string_p (&p1, end, reg_info))
5011 } /* while p1 < end */
5014 } /* group_match_null_string_p */
5017 /* Similar to group_match_null_string_p, but doesn't deal with alternatives:
5018 It expects P to be the first byte of a single alternative and END one
5019 byte past the last. The alternative can contain groups. */
5022 alt_match_null_string_p (p, end, reg_info)
5023 unsigned char *p, *end;
5024 register_info_type *reg_info;
5027 unsigned char *p1 = p;
5031 /* Skip over opcodes that can match nothing, and break when we get
5032 to one that can't. */
5034 switch ((re_opcode_t) *p1)
5037 case on_failure_jump:
5039 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
5044 if (!common_op_match_null_string_p (&p1, end, reg_info))
5047 } /* while p1 < end */
5050 } /* alt_match_null_string_p */
5053 /* Deals with the ops common to group_match_null_string_p and
5054 alt_match_null_string_p.
5056 Sets P to one after the op and its arguments, if any. */
5059 common_op_match_null_string_p (p, end, reg_info)
5060 unsigned char **p, *end;
5061 register_info_type *reg_info;
5066 unsigned char *p1 = *p;
5068 switch ((re_opcode_t) *p1++)
5088 assert (reg_no > 0 && reg_no <= MAX_REGNUM);
5089 ret = group_match_null_string_p (&p1, end, reg_info);
5091 /* Have to set this here in case we're checking a group which
5092 contains a group and a back reference to it. */
5094 if (REG_MATCH_NULL_STRING_P (reg_info[reg_no]) == MATCH_NULL_UNSET_VALUE)
5095 REG_MATCH_NULL_STRING_P (reg_info[reg_no]) = ret;
5101 /* If this is an optimized succeed_n for zero times, make the jump. */
5103 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
5111 /* Get to the number of times to succeed. */
5113 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
5118 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
5126 if (!REG_MATCH_NULL_STRING_P (reg_info[*p1]))
5134 /* All other opcodes mean we cannot match the empty string. */
5140 } /* common_op_match_null_string_p */
5143 /* Return zero if TRANSLATE[S1] and TRANSLATE[S2] are identical for LEN
5144 bytes; nonzero otherwise. */
5147 bcmp_translate (s1, s2, len, translate)
5148 unsigned char *s1, *s2;
5150 RE_TRANSLATE_TYPE translate;
5152 register unsigned char *p1 = s1, *p2 = s2;
5155 if (translate[*p1++] != translate[*p2++]) return 1;
5161 /* Entry points for GNU code. */
5163 /* re_compile_pattern is the GNU regular expression compiler: it
5164 compiles PATTERN (of length SIZE) and puts the result in BUFP.
5165 Returns 0 if the pattern was valid, otherwise an error string.
5167 Assumes the `allocated' (and perhaps `buffer') and `translate' fields
5168 are set in BUFP on entry.
5170 We call regex_compile to do the actual compilation. */
5173 re_compile_pattern (pattern, length, bufp)
5174 const char *pattern;
5176 struct re_pattern_buffer *bufp;
5180 /* GNU code is written to assume at least RE_NREGS registers will be set
5181 (and at least one extra will be -1). */
5182 bufp->regs_allocated = REGS_UNALLOCATED;
5184 /* And GNU code determines whether or not to get register information
5185 by passing null for the REGS argument to re_match, etc., not by
5189 /* Match anchors at newline. */
5190 bufp->newline_anchor = 1;
5192 ret = regex_compile (pattern, length, re_syntax_options, bufp);
5196 return gettext (re_error_msgid[(int) ret]);
5199 /* Entry points compatible with 4.2 BSD regex library. We don't define
5200 them unless specifically requested. */
5202 #if defined (_REGEX_RE_COMP) || defined (_LIBC)
5204 /* BSD has one and only one pattern buffer. */
5205 static struct re_pattern_buffer re_comp_buf;
5209 /* Make these definitions weak in libc, so POSIX programs can redefine
5210 these names if they don't use our functions, and still use
5211 regcomp/regexec below without link errors. */
5221 if (!re_comp_buf.buffer)
5222 return gettext ("No previous regular expression");
5226 if (!re_comp_buf.buffer)
5228 re_comp_buf.buffer = (unsigned char *) malloc (200);
5229 if (re_comp_buf.buffer == NULL)
5230 return gettext (re_error_msgid[(int) REG_ESPACE]);
5231 re_comp_buf.allocated = 200;
5233 re_comp_buf.fastmap = (char *) malloc (1 << BYTEWIDTH);
5234 if (re_comp_buf.fastmap == NULL)
5235 return gettext (re_error_msgid[(int) REG_ESPACE]);
5238 /* Since `re_exec' always passes NULL for the `regs' argument, we
5239 don't need to initialize the pattern buffer fields which affect it. */
5241 /* Match anchors at newlines. */
5242 re_comp_buf.newline_anchor = 1;
5244 ret = regex_compile (s, strlen (s), re_syntax_options, &re_comp_buf);
5249 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
5250 return (char *) gettext (re_error_msgid[(int) ret]);
5261 const int len = strlen (s);
5263 0 <= re_search (&re_comp_buf, s, len, 0, len, (struct re_registers *) 0);
5265 #endif /* _REGEX_RE_COMP */
5267 /* POSIX.2 functions. Don't define these for Emacs. */
5271 /* regcomp takes a regular expression as a string and compiles it.
5273 PREG is a regex_t *. We do not expect any fields to be initialized,
5274 since POSIX says we shouldn't. Thus, we set
5276 `buffer' to the compiled pattern;
5277 `used' to the length of the compiled pattern;
5278 `syntax' to RE_SYNTAX_POSIX_EXTENDED if the
5279 REG_EXTENDED bit in CFLAGS is set; otherwise, to
5280 RE_SYNTAX_POSIX_BASIC;
5281 `newline_anchor' to REG_NEWLINE being set in CFLAGS;
5282 `fastmap' and `fastmap_accurate' to zero;
5283 `re_nsub' to the number of subexpressions in PATTERN.
5285 PATTERN is the address of the pattern string.
5287 CFLAGS is a series of bits which affect compilation.
5289 If REG_EXTENDED is set, we use POSIX extended syntax; otherwise, we
5290 use POSIX basic syntax.
5292 If REG_NEWLINE is set, then . and [^...] don't match newline.
5293 Also, regexec will try a match beginning after every newline.
5295 If REG_ICASE is set, then we considers upper- and lowercase
5296 versions of letters to be equivalent when matching.
5298 If REG_NOSUB is set, then when PREG is passed to regexec, that
5299 routine will report only success or failure, and nothing about the
5302 It returns 0 if it succeeds, nonzero if it doesn't. (See regex.h for
5303 the return codes and their meanings.) */
5306 regcomp (preg, pattern, cflags)
5308 const char *pattern;
5313 = (cflags & REG_EXTENDED) ?
5314 RE_SYNTAX_POSIX_EXTENDED : RE_SYNTAX_POSIX_BASIC;
5316 /* regex_compile will allocate the space for the compiled pattern. */
5318 preg->allocated = 0;
5321 /* Don't bother to use a fastmap when searching. This simplifies the
5322 REG_NEWLINE case: if we used a fastmap, we'd have to put all the
5323 characters after newlines into the fastmap. This way, we just try
5327 if (cflags & REG_ICASE)
5332 = (RE_TRANSLATE_TYPE) malloc (CHAR_SET_SIZE
5333 * sizeof (*(RE_TRANSLATE_TYPE)0));
5334 if (preg->translate == NULL)
5335 return (int) REG_ESPACE;
5337 /* Map uppercase characters to corresponding lowercase ones. */
5338 for (i = 0; i < CHAR_SET_SIZE; i++)
5339 preg->translate[i] = ISUPPER (i) ? tolower (i) : i;
5342 preg->translate = NULL;
5344 /* If REG_NEWLINE is set, newlines are treated differently. */
5345 if (cflags & REG_NEWLINE)
5346 { /* REG_NEWLINE implies neither . nor [^...] match newline. */
5347 syntax &= ~RE_DOT_NEWLINE;
5348 syntax |= RE_HAT_LISTS_NOT_NEWLINE;
5349 /* It also changes the matching behavior. */
5350 preg->newline_anchor = 1;
5353 preg->newline_anchor = 0;
5355 preg->no_sub = !!(cflags & REG_NOSUB);
5357 /* POSIX says a null character in the pattern terminates it, so we
5358 can use strlen here in compiling the pattern. */
5359 ret = regex_compile (pattern, strlen (pattern), syntax, preg);
5361 /* POSIX doesn't distinguish between an unmatched open-group and an
5362 unmatched close-group: both are REG_EPAREN. */
5363 if (ret == REG_ERPAREN) ret = REG_EPAREN;
5369 /* regexec searches for a given pattern, specified by PREG, in the
5372 If NMATCH is zero or REG_NOSUB was set in the cflags argument to
5373 `regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at
5374 least NMATCH elements, and we set them to the offsets of the
5375 corresponding matched substrings.
5377 EFLAGS specifies `execution flags' which affect matching: if
5378 REG_NOTBOL is set, then ^ does not match at the beginning of the
5379 string; if REG_NOTEOL is set, then $ does not match at the end.
5381 We return 0 if we find a match and REG_NOMATCH if not. */
5384 regexec (preg, string, nmatch, pmatch, eflags)
5385 const regex_t *preg;
5388 regmatch_t pmatch[];
5392 struct re_registers regs;
5393 regex_t private_preg;
5394 int len = strlen (string);
5395 boolean want_reg_info = !preg->no_sub && nmatch > 0;
5397 private_preg = *preg;
5399 private_preg.not_bol = !!(eflags & REG_NOTBOL);
5400 private_preg.not_eol = !!(eflags & REG_NOTEOL);
5402 /* The user has told us exactly how many registers to return
5403 information about, via `nmatch'. We have to pass that on to the
5404 matching routines. */
5405 private_preg.regs_allocated = REGS_FIXED;
5409 regs.num_regs = nmatch;
5410 regs.start = TALLOC (nmatch, regoff_t);
5411 regs.end = TALLOC (nmatch, regoff_t);
5412 if (regs.start == NULL || regs.end == NULL)
5413 return (int) REG_NOMATCH;
5416 /* Perform the searching operation. */
5417 ret = re_search (&private_preg, string, len,
5418 /* start: */ 0, /* range: */ len,
5419 want_reg_info ? ®s : (struct re_registers *) 0);
5421 /* Copy the register information to the POSIX structure. */
5428 for (r = 0; r < nmatch; r++)
5430 pmatch[r].rm_so = regs.start[r];
5431 pmatch[r].rm_eo = regs.end[r];
5435 /* If we needed the temporary register info, free the space now. */
5440 /* We want zero return to mean success, unlike `re_search'. */
5441 return ret >= 0 ? (int) REG_NOERROR : (int) REG_NOMATCH;
5445 /* Returns a message corresponding to an error code, ERRCODE, returned
5446 from either regcomp or regexec. We don't use PREG here. */
5449 regerror (errcode, preg, errbuf, errbuf_size)
5451 const regex_t *preg;
5459 || errcode >= (sizeof (re_error_msgid) / sizeof (re_error_msgid[0])))
5460 /* Only error codes returned by the rest of the code should be passed
5461 to this routine. If we are given anything else, or if other regex
5462 code generates an invalid error code, then the program has a bug.
5463 Dump core so we can fix it. */
5466 msg = gettext (re_error_msgid[errcode]);
5468 msg_size = strlen (msg) + 1; /* Includes the null. */
5470 if (errbuf_size != 0)
5472 if (msg_size > errbuf_size)
5474 strncpy (errbuf, msg, errbuf_size - 1);
5475 errbuf[errbuf_size - 1] = 0;
5478 strcpy (errbuf, msg);
5485 /* Free dynamically allocated space used by PREG. */
5491 if (preg->buffer != NULL)
5492 free (preg->buffer);
5493 preg->buffer = NULL;
5495 preg->allocated = 0;
5498 if (preg->fastmap != NULL)
5499 free (preg->fastmap);
5500 preg->fastmap = NULL;
5501 preg->fastmap_accurate = 0;
5503 if (preg->translate != NULL)
5504 free (preg->translate);
5505 preg->translate = NULL;
5508 #endif /* not emacs */