1 /* Extended regular expression matching and search library,
3 (Implements POSIX draft P1003.2/D11.2, except for some of the
4 internationalization features.)
5 Copyright (C) 1993, 94, 95, 96, 97, 98 Free Software Foundation, Inc.
7 NOTE: The canonical source of this file is maintained with the GNU C Library.
8 Bugs can be reported to bug-glibc@prep.ai.mit.edu.
10 This program is free software; you can redistribute it and/or modify it
11 under the terms of the GNU General Public License as published by the
12 Free Software Foundation; either version 2, or (at your option) any
15 This program is distributed in the hope that it will be useful,
16 but WITHOUT ANY WARRANTY; without even the implied warranty of
17 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
18 GNU General Public License for more details.
20 You should have received a copy of the GNU General Public License
21 along with this program; if not, write to the Free Software
22 Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307,
25 /* AIX requires this to be the first thing in the file. */
26 #if defined _AIX && !defined REGEX_MALLOC
38 # if defined __GNUC__ || (defined __STDC__ && __STDC__)
39 # define PARAMS(args) args
41 # define PARAMS(args) ()
43 #endif /* Not PARAMS. */
45 #if defined STDC_HEADERS && !defined emacs
48 /* We need this for `regex.h', and perhaps for the Emacs include files. */
49 # include <sys/types.h>
52 #define WIDE_CHAR_SUPPORT \
53 defined _LIBC || (HAVE_WCTYPE_H && HAVE_WCHAR_H && HAVE_BTOWC)
55 /* For platform which support the ISO C amendement 1 functionality we
56 support user defined character classes. */
58 /* Solaris 2.5 has a bug: <wchar.h> must be included before <wctype.h>. */
64 /* We have to keep the namespace clean. */
65 # define regfree(preg) __regfree (preg)
66 # define regexec(pr, st, nm, pm, ef) __regexec (pr, st, nm, pm, ef)
67 # define regcomp(preg, pattern, cflags) __regcomp (preg, pattern, cflags)
68 # define regerror(errcode, preg, errbuf, errbuf_size) \
69 __regerror(errcode, preg, errbuf, errbuf_size)
70 # define re_set_registers(bu, re, nu, st, en) \
71 __re_set_registers (bu, re, nu, st, en)
72 # define re_match_2(bufp, string1, size1, string2, size2, pos, regs, stop) \
73 __re_match_2 (bufp, string1, size1, string2, size2, pos, regs, stop)
74 # define re_match(bufp, string, size, pos, regs) \
75 __re_match (bufp, string, size, pos, regs)
76 # define re_search(bufp, string, size, startpos, range, regs) \
77 __re_search (bufp, string, size, startpos, range, regs)
78 # define re_compile_pattern(pattern, length, bufp) \
79 __re_compile_pattern (pattern, length, bufp)
80 # define re_set_syntax(syntax) __re_set_syntax (syntax)
81 # define re_search_2(bufp, st1, s1, st2, s2, startpos, range, regs, stop) \
82 __re_search_2 (bufp, st1, s1, st2, s2, startpos, range, regs, stop)
83 # define re_compile_fastmap(bufp) __re_compile_fastmap (bufp)
88 /* This is for other GNU distributions with internationalized messages. */
89 #if HAVE_LIBINTL_H || defined _LIBC
92 # define gettext(msgid) (msgid)
96 /* This define is so xgettext can find the internationalizable
98 # define gettext_noop(String) String
101 /* The `emacs' switch turns on certain matching commands
102 that make sense only in Emacs. */
109 #else /* not emacs */
111 /* If we are not linking with Emacs proper,
112 we can't use the relocating allocator
113 even if config.h says that we can. */
116 # if defined STDC_HEADERS || defined _LIBC
123 /* When used in Emacs's lib-src, we need to get bzero and bcopy somehow.
124 If nothing else has been done, use the method below. */
125 # ifdef INHIBIT_STRING_HEADER
126 # if !(defined HAVE_BZERO && defined HAVE_BCOPY)
127 # if !defined bzero && !defined bcopy
128 # undef INHIBIT_STRING_HEADER
133 /* This is the normal way of making sure we have a bcopy and a bzero.
134 This is used in most programs--a few other programs avoid this
135 by defining INHIBIT_STRING_HEADER. */
136 # ifndef INHIBIT_STRING_HEADER
137 # if defined HAVE_STRING_H || defined STDC_HEADERS || defined _LIBC
141 # define bzero(s, n) (memset (s, '\0', n), (s))
143 # define bzero(s, n) __bzero (s, n)
147 # include <strings.h>
149 # define memcmp(s1, s2, n) bcmp (s1, s2, n)
152 # define memcpy(d, s, n) (bcopy (s, d, n), (d))
157 /* Define the syntax stuff for \<, \>, etc. */
159 /* This must be nonzero for the wordchar and notwordchar pattern
160 commands in re_match_2. */
165 # ifdef SWITCH_ENUM_BUG
166 # define SWITCH_ENUM_CAST(x) ((int)(x))
168 # define SWITCH_ENUM_CAST(x) (x)
171 /* How many characters in the character set. */
172 # define CHAR_SET_SIZE 256
176 extern char *re_syntax_table;
178 # else /* not SYNTAX_TABLE */
180 static char re_syntax_table[CHAR_SET_SIZE];
191 bzero (re_syntax_table, sizeof re_syntax_table);
193 for (c = 'a'; c <= 'z'; c++)
194 re_syntax_table[c] = Sword;
196 for (c = 'A'; c <= 'Z'; c++)
197 re_syntax_table[c] = Sword;
199 for (c = '0'; c <= '9'; c++)
200 re_syntax_table[c] = Sword;
202 re_syntax_table['_'] = Sword;
207 # endif /* not SYNTAX_TABLE */
209 # define SYNTAX(c) re_syntax_table[c]
211 #endif /* not emacs */
213 /* Get the interface, including the syntax bits. */
216 /* isalpha etc. are used for the character classes. */
219 /* Jim Meyering writes:
221 "... Some ctype macros are valid only for character codes that
222 isascii says are ASCII (SGI's IRIX-4.0.5 is one such system --when
223 using /bin/cc or gcc but without giving an ansi option). So, all
224 ctype uses should be through macros like ISPRINT... If
225 STDC_HEADERS is defined, then autoconf has verified that the ctype
226 macros don't need to be guarded with references to isascii. ...
227 Defining isascii to 1 should let any compiler worth its salt
228 eliminate the && through constant folding."
229 Solaris defines some of these symbols so we must undefine them first. */
232 #if defined STDC_HEADERS || (!defined isascii && !defined HAVE_ISASCII)
233 # define ISASCII(c) 1
235 # define ISASCII(c) isascii(c)
239 # define ISBLANK(c) (ISASCII (c) && isblank (c))
241 # define ISBLANK(c) ((c) == ' ' || (c) == '\t')
244 # define ISGRAPH(c) (ISASCII (c) && isgraph (c))
246 # define ISGRAPH(c) (ISASCII (c) && isprint (c) && !isspace (c))
250 #define ISPRINT(c) (ISASCII (c) && isprint (c))
251 #define ISDIGIT(c) (ISASCII (c) && isdigit (c))
252 #define ISALNUM(c) (ISASCII (c) && isalnum (c))
253 #define ISALPHA(c) (ISASCII (c) && isalpha (c))
254 #define ISCNTRL(c) (ISASCII (c) && iscntrl (c))
255 #define ISLOWER(c) (ISASCII (c) && islower (c))
256 #define ISPUNCT(c) (ISASCII (c) && ispunct (c))
257 #define ISSPACE(c) (ISASCII (c) && isspace (c))
258 #define ISUPPER(c) (ISASCII (c) && isupper (c))
259 #define ISXDIGIT(c) (ISASCII (c) && isxdigit (c))
262 # define NULL (void *)0
265 /* We remove any previous definition of `SIGN_EXTEND_CHAR',
266 since ours (we hope) works properly with all combinations of
267 machines, compilers, `char' and `unsigned char' argument types.
268 (Per Bothner suggested the basic approach.) */
269 #undef SIGN_EXTEND_CHAR
271 # define SIGN_EXTEND_CHAR(c) ((signed char) (c))
272 #else /* not __STDC__ */
273 /* As in Harbison and Steele. */
274 # define SIGN_EXTEND_CHAR(c) ((((unsigned char) (c)) ^ 128) - 128)
277 /* Should we use malloc or alloca? If REGEX_MALLOC is not defined, we
278 use `alloca' instead of `malloc'. This is because using malloc in
279 re_search* or re_match* could cause memory leaks when C-g is used in
280 Emacs; also, malloc is slower and causes storage fragmentation. On
281 the other hand, malloc is more portable, and easier to debug.
283 Because we sometimes use alloca, some routines have to be macros,
284 not functions -- `alloca'-allocated space disappears at the end of the
285 function it is called in. */
289 # define REGEX_ALLOCATE malloc
290 # define REGEX_REALLOCATE(source, osize, nsize) realloc (source, nsize)
291 # define REGEX_FREE free
293 #else /* not REGEX_MALLOC */
295 /* Emacs already defines alloca, sometimes. */
298 /* Make alloca work the best possible way. */
300 # define alloca __builtin_alloca
301 # else /* not __GNUC__ */
304 # endif /* HAVE_ALLOCA_H */
305 # endif /* not __GNUC__ */
307 # endif /* not alloca */
309 # define REGEX_ALLOCATE alloca
311 /* Assumes a `char *destination' variable. */
312 # define REGEX_REALLOCATE(source, osize, nsize) \
313 (destination = (char *) alloca (nsize), \
314 memcpy (destination, source, osize))
316 /* No need to do anything to free, after alloca. */
317 # define REGEX_FREE(arg) ((void)0) /* Do nothing! But inhibit gcc warning. */
319 #endif /* not REGEX_MALLOC */
321 /* Define how to allocate the failure stack. */
323 #if defined REL_ALLOC && defined REGEX_MALLOC
325 # define REGEX_ALLOCATE_STACK(size) \
326 r_alloc (&failure_stack_ptr, (size))
327 # define REGEX_REALLOCATE_STACK(source, osize, nsize) \
328 r_re_alloc (&failure_stack_ptr, (nsize))
329 # define REGEX_FREE_STACK(ptr) \
330 r_alloc_free (&failure_stack_ptr)
332 #else /* not using relocating allocator */
336 # define REGEX_ALLOCATE_STACK malloc
337 # define REGEX_REALLOCATE_STACK(source, osize, nsize) realloc (source, nsize)
338 # define REGEX_FREE_STACK free
340 # else /* not REGEX_MALLOC */
342 # define REGEX_ALLOCATE_STACK alloca
344 # define REGEX_REALLOCATE_STACK(source, osize, nsize) \
345 REGEX_REALLOCATE (source, osize, nsize)
346 /* No need to explicitly free anything. */
347 # define REGEX_FREE_STACK(arg)
349 # endif /* not REGEX_MALLOC */
350 #endif /* not using relocating allocator */
353 /* True if `size1' is non-NULL and PTR is pointing anywhere inside
354 `string1' or just past its end. This works if PTR is NULL, which is
356 #define FIRST_STRING_P(ptr) \
357 (size1 && string1 <= (ptr) && (ptr) <= string1 + size1)
359 /* (Re)Allocate N items of type T using malloc, or fail. */
360 #define TALLOC(n, t) ((t *) malloc ((n) * sizeof (t)))
361 #define RETALLOC(addr, n, t) ((addr) = (t *) realloc (addr, (n) * sizeof (t)))
362 #define RETALLOC_IF(addr, n, t) \
363 if (addr) RETALLOC((addr), (n), t); else (addr) = TALLOC ((n), t)
364 #define REGEX_TALLOC(n, t) ((t *) REGEX_ALLOCATE ((n) * sizeof (t)))
366 #define BYTEWIDTH 8 /* In bits. */
368 #define STREQ(s1, s2) ((strcmp (s1, s2) == 0))
372 #define MAX(a, b) ((a) > (b) ? (a) : (b))
373 #define MIN(a, b) ((a) < (b) ? (a) : (b))
375 typedef char boolean;
379 static int re_match_2_internal PARAMS ((struct re_pattern_buffer *bufp,
380 const char *string1, int size1,
381 const char *string2, int size2,
383 struct re_registers *regs,
386 /* These are the command codes that appear in compiled regular
387 expressions. Some opcodes are followed by argument bytes. A
388 command code can specify any interpretation whatsoever for its
389 arguments. Zero bytes may appear in the compiled regular expression. */
395 /* Succeed right away--no more backtracking. */
398 /* Followed by one byte giving n, then by n literal bytes. */
401 /* Matches any (more or less) character. */
404 /* Matches any one char belonging to specified set. First
405 following byte is number of bitmap bytes. Then come bytes
406 for a bitmap saying which chars are in. Bits in each byte
407 are ordered low-bit-first. A character is in the set if its
408 bit is 1. A character too large to have a bit in the map is
409 automatically not in the set. */
412 /* Same parameters as charset, but match any character that is
413 not one of those specified. */
416 /* Start remembering the text that is matched, for storing in a
417 register. Followed by one byte with the register number, in
418 the range 0 to one less than the pattern buffer's re_nsub
419 field. Then followed by one byte with the number of groups
420 inner to this one. (This last has to be part of the
421 start_memory only because we need it in the on_failure_jump
425 /* Stop remembering the text that is matched and store it in a
426 memory register. Followed by one byte with the register
427 number, in the range 0 to one less than `re_nsub' in the
428 pattern buffer, and one byte with the number of inner groups,
429 just like `start_memory'. (We need the number of inner
430 groups here because we don't have any easy way of finding the
431 corresponding start_memory when we're at a stop_memory.) */
434 /* Match a duplicate of something remembered. Followed by one
435 byte containing the register number. */
438 /* Fail unless at beginning of line. */
441 /* Fail unless at end of line. */
444 /* Succeeds if at beginning of buffer (if emacs) or at beginning
445 of string to be matched (if not). */
448 /* Analogously, for end of buffer/string. */
451 /* Followed by two byte relative address to which to jump. */
454 /* Same as jump, but marks the end of an alternative. */
457 /* Followed by two-byte relative address of place to resume at
458 in case of failure. */
461 /* Like on_failure_jump, but pushes a placeholder instead of the
462 current string position when executed. */
463 on_failure_keep_string_jump,
465 /* Throw away latest failure point and then jump to following
466 two-byte relative address. */
469 /* Change to pop_failure_jump if know won't have to backtrack to
470 match; otherwise change to jump. This is used to jump
471 back to the beginning of a repeat. If what follows this jump
472 clearly won't match what the repeat does, such that we can be
473 sure that there is no use backtracking out of repetitions
474 already matched, then we change it to a pop_failure_jump.
475 Followed by two-byte address. */
478 /* Jump to following two-byte address, and push a dummy failure
479 point. This failure point will be thrown away if an attempt
480 is made to use it for a failure. A `+' construct makes this
481 before the first repeat. Also used as an intermediary kind
482 of jump when compiling an alternative. */
485 /* Push a dummy failure point and continue. Used at the end of
489 /* Followed by two-byte relative address and two-byte number n.
490 After matching N times, jump to the address upon failure. */
493 /* Followed by two-byte relative address, and two-byte number n.
494 Jump to the address N times, then fail. */
497 /* Set the following two-byte relative address to the
498 subsequent two-byte number. The address *includes* the two
502 wordchar, /* Matches any word-constituent character. */
503 notwordchar, /* Matches any char that is not a word-constituent. */
505 wordbeg, /* Succeeds if at word beginning. */
506 wordend, /* Succeeds if at word end. */
508 wordbound, /* Succeeds if at a word boundary. */
509 notwordbound /* Succeeds if not at a word boundary. */
512 ,before_dot, /* Succeeds if before point. */
513 at_dot, /* Succeeds if at point. */
514 after_dot, /* Succeeds if after point. */
516 /* Matches any character whose syntax is specified. Followed by
517 a byte which contains a syntax code, e.g., Sword. */
520 /* Matches any character whose syntax is not that specified. */
525 /* Common operations on the compiled pattern. */
527 /* Store NUMBER in two contiguous bytes starting at DESTINATION. */
529 #define STORE_NUMBER(destination, number) \
531 (destination)[0] = (number) & 0377; \
532 (destination)[1] = (number) >> 8; \
535 /* Same as STORE_NUMBER, except increment DESTINATION to
536 the byte after where the number is stored. Therefore, DESTINATION
537 must be an lvalue. */
539 #define STORE_NUMBER_AND_INCR(destination, number) \
541 STORE_NUMBER (destination, number); \
542 (destination) += 2; \
545 /* Put into DESTINATION a number stored in two contiguous bytes starting
548 #define EXTRACT_NUMBER(destination, source) \
550 (destination) = *(source) & 0377; \
551 (destination) += SIGN_EXTEND_CHAR (*((source) + 1)) << 8; \
555 static void extract_number _RE_ARGS ((int *dest, unsigned char *source));
557 extract_number (dest, source)
559 unsigned char *source;
561 int temp = SIGN_EXTEND_CHAR (*(source + 1));
562 *dest = *source & 0377;
566 # ifndef EXTRACT_MACROS /* To debug the macros. */
567 # undef EXTRACT_NUMBER
568 # define EXTRACT_NUMBER(dest, src) extract_number (&dest, src)
569 # endif /* not EXTRACT_MACROS */
573 /* Same as EXTRACT_NUMBER, except increment SOURCE to after the number.
574 SOURCE must be an lvalue. */
576 #define EXTRACT_NUMBER_AND_INCR(destination, source) \
578 EXTRACT_NUMBER (destination, source); \
583 static void extract_number_and_incr _RE_ARGS ((int *destination,
584 unsigned char **source));
586 extract_number_and_incr (destination, source)
588 unsigned char **source;
590 extract_number (destination, *source);
594 # ifndef EXTRACT_MACROS
595 # undef EXTRACT_NUMBER_AND_INCR
596 # define EXTRACT_NUMBER_AND_INCR(dest, src) \
597 extract_number_and_incr (&dest, &src)
598 # endif /* not EXTRACT_MACROS */
602 /* If DEBUG is defined, Regex prints many voluminous messages about what
603 it is doing (if the variable `debug' is nonzero). If linked with the
604 main program in `iregex.c', you can enter patterns and strings
605 interactively. And if linked with the main program in `main.c' and
606 the other test files, you can run the already-written tests. */
610 /* We use standard I/O for debugging. */
613 /* It is useful to test things that ``must'' be true when debugging. */
616 static int debug = 0;
618 # define DEBUG_STATEMENT(e) e
619 # define DEBUG_PRINT1(x) if (debug) printf (x)
620 # define DEBUG_PRINT2(x1, x2) if (debug) printf (x1, x2)
621 # define DEBUG_PRINT3(x1, x2, x3) if (debug) printf (x1, x2, x3)
622 # define DEBUG_PRINT4(x1, x2, x3, x4) if (debug) printf (x1, x2, x3, x4)
623 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) \
624 if (debug) print_partial_compiled_pattern (s, e)
625 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) \
626 if (debug) print_double_string (w, s1, sz1, s2, sz2)
629 /* Print the fastmap in human-readable form. */
632 print_fastmap (fastmap)
635 unsigned was_a_range = 0;
638 while (i < (1 << BYTEWIDTH))
644 while (i < (1 << BYTEWIDTH) && fastmap[i])
660 /* Print a compiled pattern string in human-readable form, starting at
661 the START pointer into it and ending just before the pointer END. */
664 print_partial_compiled_pattern (start, end)
665 unsigned char *start;
670 unsigned char *p = start;
671 unsigned char *pend = end;
679 /* Loop over pattern commands. */
682 printf ("%d:\t", p - start);
684 switch ((re_opcode_t) *p++)
692 printf ("/exactn/%d", mcnt);
703 printf ("/start_memory/%d/%d", mcnt, *p++);
708 printf ("/stop_memory/%d/%d", mcnt, *p++);
712 printf ("/duplicate/%d", *p++);
722 register int c, last = -100;
723 register int in_range = 0;
725 printf ("/charset [%s",
726 (re_opcode_t) *(p - 1) == charset_not ? "^" : "");
728 assert (p + *p < pend);
730 for (c = 0; c < 256; c++)
732 && (p[1 + (c/8)] & (1 << (c % 8))))
734 /* Are we starting a range? */
735 if (last + 1 == c && ! in_range)
740 /* Have we broken a range? */
741 else if (last + 1 != c && in_range)
770 case on_failure_jump:
771 extract_number_and_incr (&mcnt, &p);
772 printf ("/on_failure_jump to %d", p + mcnt - start);
775 case on_failure_keep_string_jump:
776 extract_number_and_incr (&mcnt, &p);
777 printf ("/on_failure_keep_string_jump to %d", p + mcnt - start);
780 case dummy_failure_jump:
781 extract_number_and_incr (&mcnt, &p);
782 printf ("/dummy_failure_jump to %d", p + mcnt - start);
785 case push_dummy_failure:
786 printf ("/push_dummy_failure");
790 extract_number_and_incr (&mcnt, &p);
791 printf ("/maybe_pop_jump to %d", p + mcnt - start);
794 case pop_failure_jump:
795 extract_number_and_incr (&mcnt, &p);
796 printf ("/pop_failure_jump to %d", p + mcnt - start);
800 extract_number_and_incr (&mcnt, &p);
801 printf ("/jump_past_alt to %d", p + mcnt - start);
805 extract_number_and_incr (&mcnt, &p);
806 printf ("/jump to %d", p + mcnt - start);
810 extract_number_and_incr (&mcnt, &p);
812 extract_number_and_incr (&mcnt2, &p);
813 printf ("/succeed_n to %d, %d times", p1 - start, mcnt2);
817 extract_number_and_incr (&mcnt, &p);
819 extract_number_and_incr (&mcnt2, &p);
820 printf ("/jump_n to %d, %d times", p1 - start, mcnt2);
824 extract_number_and_incr (&mcnt, &p);
826 extract_number_and_incr (&mcnt2, &p);
827 printf ("/set_number_at location %d to %d", p1 - start, mcnt2);
831 printf ("/wordbound");
835 printf ("/notwordbound");
847 printf ("/before_dot");
855 printf ("/after_dot");
859 printf ("/syntaxspec");
861 printf ("/%d", mcnt);
865 printf ("/notsyntaxspec");
867 printf ("/%d", mcnt);
872 printf ("/wordchar");
876 printf ("/notwordchar");
888 printf ("?%d", *(p-1));
894 printf ("%d:\tend of pattern.\n", p - start);
899 print_compiled_pattern (bufp)
900 struct re_pattern_buffer *bufp;
902 unsigned char *buffer = bufp->buffer;
904 print_partial_compiled_pattern (buffer, buffer + bufp->used);
905 printf ("%ld bytes used/%ld bytes allocated.\n",
906 bufp->used, bufp->allocated);
908 if (bufp->fastmap_accurate && bufp->fastmap)
910 printf ("fastmap: ");
911 print_fastmap (bufp->fastmap);
914 printf ("re_nsub: %d\t", bufp->re_nsub);
915 printf ("regs_alloc: %d\t", bufp->regs_allocated);
916 printf ("can_be_null: %d\t", bufp->can_be_null);
917 printf ("newline_anchor: %d\n", bufp->newline_anchor);
918 printf ("no_sub: %d\t", bufp->no_sub);
919 printf ("not_bol: %d\t", bufp->not_bol);
920 printf ("not_eol: %d\t", bufp->not_eol);
921 printf ("syntax: %lx\n", bufp->syntax);
922 /* Perhaps we should print the translate table? */
927 print_double_string (where, string1, size1, string2, size2)
940 if (FIRST_STRING_P (where))
942 for (this_char = where - string1; this_char < size1; this_char++)
943 putchar (string1[this_char]);
948 for (this_char = where - string2; this_char < size2; this_char++)
949 putchar (string2[this_char]);
960 #else /* not DEBUG */
965 # define DEBUG_STATEMENT(e)
966 # define DEBUG_PRINT1(x)
967 # define DEBUG_PRINT2(x1, x2)
968 # define DEBUG_PRINT3(x1, x2, x3)
969 # define DEBUG_PRINT4(x1, x2, x3, x4)
970 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e)
971 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2)
973 #endif /* not DEBUG */
975 /* Set by `re_set_syntax' to the current regexp syntax to recognize. Can
976 also be assigned to arbitrarily: each pattern buffer stores its own
977 syntax, so it can be changed between regex compilations. */
978 /* This has no initializer because initialized variables in Emacs
979 become read-only after dumping. */
980 reg_syntax_t re_syntax_options;
983 /* Specify the precise syntax of regexps for compilation. This provides
984 for compatibility for various utilities which historically have
985 different, incompatible syntaxes.
987 The argument SYNTAX is a bit mask comprised of the various bits
988 defined in regex.h. We return the old syntax. */
991 re_set_syntax (syntax)
994 reg_syntax_t ret = re_syntax_options;
996 re_syntax_options = syntax;
998 if (syntax & RE_DEBUG)
1000 else if (debug) /* was on but now is not */
1006 weak_alias (__re_set_syntax, re_set_syntax)
1009 /* This table gives an error message for each of the error codes listed
1010 in regex.h. Obviously the order here has to be same as there.
1011 POSIX doesn't require that we do anything for REG_NOERROR,
1012 but why not be nice? */
1014 static const char *re_error_msgid[] =
1016 gettext_noop ("Success"), /* REG_NOERROR */
1017 gettext_noop ("No match"), /* REG_NOMATCH */
1018 gettext_noop ("Invalid regular expression"), /* REG_BADPAT */
1019 gettext_noop ("Invalid collation character"), /* REG_ECOLLATE */
1020 gettext_noop ("Invalid character class name"), /* REG_ECTYPE */
1021 gettext_noop ("Trailing backslash"), /* REG_EESCAPE */
1022 gettext_noop ("Invalid back reference"), /* REG_ESUBREG */
1023 gettext_noop ("Unmatched [ or [^"), /* REG_EBRACK */
1024 gettext_noop ("Unmatched ( or \\("), /* REG_EPAREN */
1025 gettext_noop ("Unmatched \\{"), /* REG_EBRACE */
1026 gettext_noop ("Invalid content of \\{\\}"), /* REG_BADBR */
1027 gettext_noop ("Invalid range end"), /* REG_ERANGE */
1028 gettext_noop ("Memory exhausted"), /* REG_ESPACE */
1029 gettext_noop ("Invalid preceding regular expression"), /* REG_BADRPT */
1030 gettext_noop ("Premature end of regular expression"), /* REG_EEND */
1031 gettext_noop ("Regular expression too big"), /* REG_ESIZE */
1032 gettext_noop ("Unmatched ) or \\)"), /* REG_ERPAREN */
1035 /* Avoiding alloca during matching, to placate r_alloc. */
1037 /* Define MATCH_MAY_ALLOCATE unless we need to make sure that the
1038 searching and matching functions should not call alloca. On some
1039 systems, alloca is implemented in terms of malloc, and if we're
1040 using the relocating allocator routines, then malloc could cause a
1041 relocation, which might (if the strings being searched are in the
1042 ralloc heap) shift the data out from underneath the regexp
1045 Here's another reason to avoid allocation: Emacs
1046 processes input from X in a signal handler; processing X input may
1047 call malloc; if input arrives while a matching routine is calling
1048 malloc, then we're scrod. But Emacs can't just block input while
1049 calling matching routines; then we don't notice interrupts when
1050 they come in. So, Emacs blocks input around all regexp calls
1051 except the matching calls, which it leaves unprotected, in the
1052 faith that they will not malloc. */
1054 /* Normally, this is fine. */
1055 #define MATCH_MAY_ALLOCATE
1057 /* When using GNU C, we are not REALLY using the C alloca, no matter
1058 what config.h may say. So don't take precautions for it. */
1063 /* The match routines may not allocate if (1) they would do it with malloc
1064 and (2) it's not safe for them to use malloc.
1065 Note that if REL_ALLOC is defined, matching would not use malloc for the
1066 failure stack, but we would still use it for the register vectors;
1067 so REL_ALLOC should not affect this. */
1068 #if (defined C_ALLOCA || defined REGEX_MALLOC) && defined emacs
1069 # undef MATCH_MAY_ALLOCATE
1073 /* Failure stack declarations and macros; both re_compile_fastmap and
1074 re_match_2 use a failure stack. These have to be macros because of
1075 REGEX_ALLOCATE_STACK. */
1078 /* Number of failure points for which to initially allocate space
1079 when matching. If this number is exceeded, we allocate more
1080 space, so it is not a hard limit. */
1081 #ifndef INIT_FAILURE_ALLOC
1082 # define INIT_FAILURE_ALLOC 5
1085 /* Roughly the maximum number of failure points on the stack. Would be
1086 exactly that if always used MAX_FAILURE_ITEMS items each time we failed.
1087 This is a variable only so users of regex can assign to it; we never
1088 change it ourselves. */
1092 # if defined MATCH_MAY_ALLOCATE
1093 /* 4400 was enough to cause a crash on Alpha OSF/1,
1094 whose default stack limit is 2mb. */
1095 long int re_max_failures = 4000;
1097 long int re_max_failures = 2000;
1100 union fail_stack_elt
1102 unsigned char *pointer;
1106 typedef union fail_stack_elt fail_stack_elt_t;
1110 fail_stack_elt_t *stack;
1111 unsigned long int size;
1112 unsigned long int avail; /* Offset of next open position. */
1115 #else /* not INT_IS_16BIT */
1117 # if defined MATCH_MAY_ALLOCATE
1118 /* 4400 was enough to cause a crash on Alpha OSF/1,
1119 whose default stack limit is 2mb. */
1120 int re_max_failures = 20000;
1122 int re_max_failures = 2000;
1125 union fail_stack_elt
1127 unsigned char *pointer;
1131 typedef union fail_stack_elt fail_stack_elt_t;
1135 fail_stack_elt_t *stack;
1137 unsigned avail; /* Offset of next open position. */
1140 #endif /* INT_IS_16BIT */
1142 #define FAIL_STACK_EMPTY() (fail_stack.avail == 0)
1143 #define FAIL_STACK_PTR_EMPTY() (fail_stack_ptr->avail == 0)
1144 #define FAIL_STACK_FULL() (fail_stack.avail == fail_stack.size)
1147 /* Define macros to initialize and free the failure stack.
1148 Do `return -2' if the alloc fails. */
1150 #ifdef MATCH_MAY_ALLOCATE
1151 # define INIT_FAIL_STACK() \
1153 fail_stack.stack = (fail_stack_elt_t *) \
1154 REGEX_ALLOCATE_STACK (INIT_FAILURE_ALLOC * sizeof (fail_stack_elt_t)); \
1156 if (fail_stack.stack == NULL) \
1159 fail_stack.size = INIT_FAILURE_ALLOC; \
1160 fail_stack.avail = 0; \
1163 # define RESET_FAIL_STACK() REGEX_FREE_STACK (fail_stack.stack)
1165 # define INIT_FAIL_STACK() \
1167 fail_stack.avail = 0; \
1170 # define RESET_FAIL_STACK()
1174 /* Double the size of FAIL_STACK, up to approximately `re_max_failures' items.
1176 Return 1 if succeeds, and 0 if either ran out of memory
1177 allocating space for it or it was already too large.
1179 REGEX_REALLOCATE_STACK requires `destination' be declared. */
1181 #define DOUBLE_FAIL_STACK(fail_stack) \
1182 ((fail_stack).size > (unsigned) (re_max_failures * MAX_FAILURE_ITEMS) \
1184 : ((fail_stack).stack = (fail_stack_elt_t *) \
1185 REGEX_REALLOCATE_STACK ((fail_stack).stack, \
1186 (fail_stack).size * sizeof (fail_stack_elt_t), \
1187 ((fail_stack).size << 1) * sizeof (fail_stack_elt_t)), \
1189 (fail_stack).stack == NULL \
1191 : ((fail_stack).size <<= 1, \
1195 /* Push pointer POINTER on FAIL_STACK.
1196 Return 1 if was able to do so and 0 if ran out of memory allocating
1198 #define PUSH_PATTERN_OP(POINTER, FAIL_STACK) \
1199 ((FAIL_STACK_FULL () \
1200 && !DOUBLE_FAIL_STACK (FAIL_STACK)) \
1202 : ((FAIL_STACK).stack[(FAIL_STACK).avail++].pointer = POINTER, \
1205 /* Push a pointer value onto the failure stack.
1206 Assumes the variable `fail_stack'. Probably should only
1207 be called from within `PUSH_FAILURE_POINT'. */
1208 #define PUSH_FAILURE_POINTER(item) \
1209 fail_stack.stack[fail_stack.avail++].pointer = (unsigned char *) (item)
1211 /* This pushes an integer-valued item onto the failure stack.
1212 Assumes the variable `fail_stack'. Probably should only
1213 be called from within `PUSH_FAILURE_POINT'. */
1214 #define PUSH_FAILURE_INT(item) \
1215 fail_stack.stack[fail_stack.avail++].integer = (item)
1217 /* Push a fail_stack_elt_t value onto the failure stack.
1218 Assumes the variable `fail_stack'. Probably should only
1219 be called from within `PUSH_FAILURE_POINT'. */
1220 #define PUSH_FAILURE_ELT(item) \
1221 fail_stack.stack[fail_stack.avail++] = (item)
1223 /* These three POP... operations complement the three PUSH... operations.
1224 All assume that `fail_stack' is nonempty. */
1225 #define POP_FAILURE_POINTER() fail_stack.stack[--fail_stack.avail].pointer
1226 #define POP_FAILURE_INT() fail_stack.stack[--fail_stack.avail].integer
1227 #define POP_FAILURE_ELT() fail_stack.stack[--fail_stack.avail]
1229 /* Used to omit pushing failure point id's when we're not debugging. */
1231 # define DEBUG_PUSH PUSH_FAILURE_INT
1232 # define DEBUG_POP(item_addr) *(item_addr) = POP_FAILURE_INT ()
1234 # define DEBUG_PUSH(item)
1235 # define DEBUG_POP(item_addr)
1239 /* Push the information about the state we will need
1240 if we ever fail back to it.
1242 Requires variables fail_stack, regstart, regend, reg_info, and
1243 num_regs_pushed be declared. DOUBLE_FAIL_STACK requires `destination'
1246 Does `return FAILURE_CODE' if runs out of memory. */
1248 #define PUSH_FAILURE_POINT(pattern_place, string_place, failure_code) \
1250 char *destination; \
1251 /* Must be int, so when we don't save any registers, the arithmetic \
1252 of 0 + -1 isn't done as unsigned. */ \
1253 /* Can't be int, since there is not a shred of a guarantee that int \
1254 is wide enough to hold a value of something to which pointer can \
1256 active_reg_t this_reg; \
1258 DEBUG_STATEMENT (failure_id++); \
1259 DEBUG_STATEMENT (nfailure_points_pushed++); \
1260 DEBUG_PRINT2 ("\nPUSH_FAILURE_POINT #%u:\n", failure_id); \
1261 DEBUG_PRINT2 (" Before push, next avail: %d\n", (fail_stack).avail);\
1262 DEBUG_PRINT2 (" size: %d\n", (fail_stack).size);\
1264 DEBUG_PRINT2 (" slots needed: %ld\n", NUM_FAILURE_ITEMS); \
1265 DEBUG_PRINT2 (" available: %d\n", REMAINING_AVAIL_SLOTS); \
1267 /* Ensure we have enough space allocated for what we will push. */ \
1268 while (REMAINING_AVAIL_SLOTS < NUM_FAILURE_ITEMS) \
1270 if (!DOUBLE_FAIL_STACK (fail_stack)) \
1271 return failure_code; \
1273 DEBUG_PRINT2 ("\n Doubled stack; size now: %d\n", \
1274 (fail_stack).size); \
1275 DEBUG_PRINT2 (" slots available: %d\n", REMAINING_AVAIL_SLOTS);\
1278 /* Push the info, starting with the registers. */ \
1279 DEBUG_PRINT1 ("\n"); \
1282 for (this_reg = lowest_active_reg; this_reg <= highest_active_reg; \
1285 DEBUG_PRINT2 (" Pushing reg: %lu\n", this_reg); \
1286 DEBUG_STATEMENT (num_regs_pushed++); \
1288 DEBUG_PRINT2 (" start: %p\n", regstart[this_reg]); \
1289 PUSH_FAILURE_POINTER (regstart[this_reg]); \
1291 DEBUG_PRINT2 (" end: %p\n", regend[this_reg]); \
1292 PUSH_FAILURE_POINTER (regend[this_reg]); \
1294 DEBUG_PRINT2 (" info: %p\n ", \
1295 reg_info[this_reg].word.pointer); \
1296 DEBUG_PRINT2 (" match_null=%d", \
1297 REG_MATCH_NULL_STRING_P (reg_info[this_reg])); \
1298 DEBUG_PRINT2 (" active=%d", IS_ACTIVE (reg_info[this_reg])); \
1299 DEBUG_PRINT2 (" matched_something=%d", \
1300 MATCHED_SOMETHING (reg_info[this_reg])); \
1301 DEBUG_PRINT2 (" ever_matched=%d", \
1302 EVER_MATCHED_SOMETHING (reg_info[this_reg])); \
1303 DEBUG_PRINT1 ("\n"); \
1304 PUSH_FAILURE_ELT (reg_info[this_reg].word); \
1307 DEBUG_PRINT2 (" Pushing low active reg: %ld\n", lowest_active_reg);\
1308 PUSH_FAILURE_INT (lowest_active_reg); \
1310 DEBUG_PRINT2 (" Pushing high active reg: %ld\n", highest_active_reg);\
1311 PUSH_FAILURE_INT (highest_active_reg); \
1313 DEBUG_PRINT2 (" Pushing pattern %p:\n", pattern_place); \
1314 DEBUG_PRINT_COMPILED_PATTERN (bufp, pattern_place, pend); \
1315 PUSH_FAILURE_POINTER (pattern_place); \
1317 DEBUG_PRINT2 (" Pushing string %p: `", string_place); \
1318 DEBUG_PRINT_DOUBLE_STRING (string_place, string1, size1, string2, \
1320 DEBUG_PRINT1 ("'\n"); \
1321 PUSH_FAILURE_POINTER (string_place); \
1323 DEBUG_PRINT2 (" Pushing failure id: %u\n", failure_id); \
1324 DEBUG_PUSH (failure_id); \
1327 /* This is the number of items that are pushed and popped on the stack
1328 for each register. */
1329 #define NUM_REG_ITEMS 3
1331 /* Individual items aside from the registers. */
1333 # define NUM_NONREG_ITEMS 5 /* Includes failure point id. */
1335 # define NUM_NONREG_ITEMS 4
1338 /* We push at most this many items on the stack. */
1339 /* We used to use (num_regs - 1), which is the number of registers
1340 this regexp will save; but that was changed to 5
1341 to avoid stack overflow for a regexp with lots of parens. */
1342 #define MAX_FAILURE_ITEMS (5 * NUM_REG_ITEMS + NUM_NONREG_ITEMS)
1344 /* We actually push this many items. */
1345 #define NUM_FAILURE_ITEMS \
1347 ? 0 : highest_active_reg - lowest_active_reg + 1) \
1351 /* How many items can still be added to the stack without overflowing it. */
1352 #define REMAINING_AVAIL_SLOTS ((fail_stack).size - (fail_stack).avail)
1355 /* Pops what PUSH_FAIL_STACK pushes.
1357 We restore into the parameters, all of which should be lvalues:
1358 STR -- the saved data position.
1359 PAT -- the saved pattern position.
1360 LOW_REG, HIGH_REG -- the highest and lowest active registers.
1361 REGSTART, REGEND -- arrays of string positions.
1362 REG_INFO -- array of information about each subexpression.
1364 Also assumes the variables `fail_stack' and (if debugging), `bufp',
1365 `pend', `string1', `size1', `string2', and `size2'. */
1367 #define POP_FAILURE_POINT(str, pat, low_reg, high_reg, regstart, regend, reg_info)\
1369 DEBUG_STATEMENT (unsigned failure_id;) \
1370 active_reg_t this_reg; \
1371 const unsigned char *string_temp; \
1373 assert (!FAIL_STACK_EMPTY ()); \
1375 /* Remove failure points and point to how many regs pushed. */ \
1376 DEBUG_PRINT1 ("POP_FAILURE_POINT:\n"); \
1377 DEBUG_PRINT2 (" Before pop, next avail: %d\n", fail_stack.avail); \
1378 DEBUG_PRINT2 (" size: %d\n", fail_stack.size); \
1380 assert (fail_stack.avail >= NUM_NONREG_ITEMS); \
1382 DEBUG_POP (&failure_id); \
1383 DEBUG_PRINT2 (" Popping failure id: %u\n", failure_id); \
1385 /* If the saved string location is NULL, it came from an \
1386 on_failure_keep_string_jump opcode, and we want to throw away the \
1387 saved NULL, thus retaining our current position in the string. */ \
1388 string_temp = POP_FAILURE_POINTER (); \
1389 if (string_temp != NULL) \
1390 str = (const char *) string_temp; \
1392 DEBUG_PRINT2 (" Popping string %p: `", str); \
1393 DEBUG_PRINT_DOUBLE_STRING (str, string1, size1, string2, size2); \
1394 DEBUG_PRINT1 ("'\n"); \
1396 pat = (unsigned char *) POP_FAILURE_POINTER (); \
1397 DEBUG_PRINT2 (" Popping pattern %p:\n", pat); \
1398 DEBUG_PRINT_COMPILED_PATTERN (bufp, pat, pend); \
1400 /* Restore register info. */ \
1401 high_reg = (active_reg_t) POP_FAILURE_INT (); \
1402 DEBUG_PRINT2 (" Popping high active reg: %ld\n", high_reg); \
1404 low_reg = (active_reg_t) POP_FAILURE_INT (); \
1405 DEBUG_PRINT2 (" Popping low active reg: %ld\n", low_reg); \
1408 for (this_reg = high_reg; this_reg >= low_reg; this_reg--) \
1410 DEBUG_PRINT2 (" Popping reg: %ld\n", this_reg); \
1412 reg_info[this_reg].word = POP_FAILURE_ELT (); \
1413 DEBUG_PRINT2 (" info: %p\n", \
1414 reg_info[this_reg].word.pointer); \
1416 regend[this_reg] = (const char *) POP_FAILURE_POINTER (); \
1417 DEBUG_PRINT2 (" end: %p\n", regend[this_reg]); \
1419 regstart[this_reg] = (const char *) POP_FAILURE_POINTER (); \
1420 DEBUG_PRINT2 (" start: %p\n", regstart[this_reg]); \
1424 for (this_reg = highest_active_reg; this_reg > high_reg; this_reg--) \
1426 reg_info[this_reg].word.integer = 0; \
1427 regend[this_reg] = 0; \
1428 regstart[this_reg] = 0; \
1430 highest_active_reg = high_reg; \
1433 set_regs_matched_done = 0; \
1434 DEBUG_STATEMENT (nfailure_points_popped++); \
1435 } /* POP_FAILURE_POINT */
1439 /* Structure for per-register (a.k.a. per-group) information.
1440 Other register information, such as the
1441 starting and ending positions (which are addresses), and the list of
1442 inner groups (which is a bits list) are maintained in separate
1445 We are making a (strictly speaking) nonportable assumption here: that
1446 the compiler will pack our bit fields into something that fits into
1447 the type of `word', i.e., is something that fits into one item on the
1451 /* Declarations and macros for re_match_2. */
1455 fail_stack_elt_t word;
1458 /* This field is one if this group can match the empty string,
1459 zero if not. If not yet determined, `MATCH_NULL_UNSET_VALUE'. */
1460 #define MATCH_NULL_UNSET_VALUE 3
1461 unsigned match_null_string_p : 2;
1462 unsigned is_active : 1;
1463 unsigned matched_something : 1;
1464 unsigned ever_matched_something : 1;
1466 } register_info_type;
1468 #define REG_MATCH_NULL_STRING_P(R) ((R).bits.match_null_string_p)
1469 #define IS_ACTIVE(R) ((R).bits.is_active)
1470 #define MATCHED_SOMETHING(R) ((R).bits.matched_something)
1471 #define EVER_MATCHED_SOMETHING(R) ((R).bits.ever_matched_something)
1474 /* Call this when have matched a real character; it sets `matched' flags
1475 for the subexpressions which we are currently inside. Also records
1476 that those subexprs have matched. */
1477 #define SET_REGS_MATCHED() \
1480 if (!set_regs_matched_done) \
1483 set_regs_matched_done = 1; \
1484 for (r = lowest_active_reg; r <= highest_active_reg; r++) \
1486 MATCHED_SOMETHING (reg_info[r]) \
1487 = EVER_MATCHED_SOMETHING (reg_info[r]) \
1494 /* Registers are set to a sentinel when they haven't yet matched. */
1495 static char reg_unset_dummy;
1496 #define REG_UNSET_VALUE (®_unset_dummy)
1497 #define REG_UNSET(e) ((e) == REG_UNSET_VALUE)
1499 /* Subroutine declarations and macros for regex_compile. */
1501 static reg_errcode_t regex_compile _RE_ARGS ((const char *pattern, size_t size,
1502 reg_syntax_t syntax,
1503 struct re_pattern_buffer *bufp));
1504 static void store_op1 _RE_ARGS ((re_opcode_t op, unsigned char *loc, int arg));
1505 static void store_op2 _RE_ARGS ((re_opcode_t op, unsigned char *loc,
1506 int arg1, int arg2));
1507 static void insert_op1 _RE_ARGS ((re_opcode_t op, unsigned char *loc,
1508 int arg, unsigned char *end));
1509 static void insert_op2 _RE_ARGS ((re_opcode_t op, unsigned char *loc,
1510 int arg1, int arg2, unsigned char *end));
1511 static boolean at_begline_loc_p _RE_ARGS ((const char *pattern, const char *p,
1512 reg_syntax_t syntax));
1513 static boolean at_endline_loc_p _RE_ARGS ((const char *p, const char *pend,
1514 reg_syntax_t syntax));
1515 static reg_errcode_t compile_range _RE_ARGS ((const char **p_ptr,
1518 reg_syntax_t syntax,
1521 /* Fetch the next character in the uncompiled pattern---translating it
1522 if necessary. Also cast from a signed character in the constant
1523 string passed to us by the user to an unsigned char that we can use
1524 as an array index (in, e.g., `translate'). */
1526 # define PATFETCH(c) \
1527 do {if (p == pend) return REG_EEND; \
1528 c = (unsigned char) *p++; \
1529 if (translate) c = (unsigned char) translate[c]; \
1533 /* Fetch the next character in the uncompiled pattern, with no
1535 #define PATFETCH_RAW(c) \
1536 do {if (p == pend) return REG_EEND; \
1537 c = (unsigned char) *p++; \
1540 /* Go backwards one character in the pattern. */
1541 #define PATUNFETCH p--
1544 /* If `translate' is non-null, return translate[D], else just D. We
1545 cast the subscript to translate because some data is declared as
1546 `char *', to avoid warnings when a string constant is passed. But
1547 when we use a character as a subscript we must make it unsigned. */
1549 # define TRANSLATE(d) \
1550 (translate ? (char) translate[(unsigned char) (d)] : (d))
1554 /* Macros for outputting the compiled pattern into `buffer'. */
1556 /* If the buffer isn't allocated when it comes in, use this. */
1557 #define INIT_BUF_SIZE 32
1559 /* Make sure we have at least N more bytes of space in buffer. */
1560 #define GET_BUFFER_SPACE(n) \
1561 while ((unsigned long) (b - bufp->buffer + (n)) > bufp->allocated) \
1564 /* Make sure we have one more byte of buffer space and then add C to it. */
1565 #define BUF_PUSH(c) \
1567 GET_BUFFER_SPACE (1); \
1568 *b++ = (unsigned char) (c); \
1572 /* Ensure we have two more bytes of buffer space and then append C1 and C2. */
1573 #define BUF_PUSH_2(c1, c2) \
1575 GET_BUFFER_SPACE (2); \
1576 *b++ = (unsigned char) (c1); \
1577 *b++ = (unsigned char) (c2); \
1581 /* As with BUF_PUSH_2, except for three bytes. */
1582 #define BUF_PUSH_3(c1, c2, c3) \
1584 GET_BUFFER_SPACE (3); \
1585 *b++ = (unsigned char) (c1); \
1586 *b++ = (unsigned char) (c2); \
1587 *b++ = (unsigned char) (c3); \
1591 /* Store a jump with opcode OP at LOC to location TO. We store a
1592 relative address offset by the three bytes the jump itself occupies. */
1593 #define STORE_JUMP(op, loc, to) \
1594 store_op1 (op, loc, (int) ((to) - (loc) - 3))
1596 /* Likewise, for a two-argument jump. */
1597 #define STORE_JUMP2(op, loc, to, arg) \
1598 store_op2 (op, loc, (int) ((to) - (loc) - 3), arg)
1600 /* Like `STORE_JUMP', but for inserting. Assume `b' is the buffer end. */
1601 #define INSERT_JUMP(op, loc, to) \
1602 insert_op1 (op, loc, (int) ((to) - (loc) - 3), b)
1604 /* Like `STORE_JUMP2', but for inserting. Assume `b' is the buffer end. */
1605 #define INSERT_JUMP2(op, loc, to, arg) \
1606 insert_op2 (op, loc, (int) ((to) - (loc) - 3), arg, b)
1609 /* This is not an arbitrary limit: the arguments which represent offsets
1610 into the pattern are two bytes long. So if 2^16 bytes turns out to
1611 be too small, many things would have to change. */
1612 /* Any other compiler which, like MSC, has allocation limit below 2^16
1613 bytes will have to use approach similar to what was done below for
1614 MSC and drop MAX_BUF_SIZE a bit. Otherwise you may end up
1615 reallocating to 0 bytes. Such thing is not going to work too well.
1616 You have been warned!! */
1617 #if defined _MSC_VER && !defined WIN32
1618 /* Microsoft C 16-bit versions limit malloc to approx 65512 bytes.
1619 The REALLOC define eliminates a flurry of conversion warnings,
1620 but is not required. */
1621 # define MAX_BUF_SIZE 65500L
1622 # define REALLOC(p,s) realloc ((p), (size_t) (s))
1624 # define MAX_BUF_SIZE (1L << 16)
1625 # define REALLOC(p,s) realloc ((p), (s))
1628 /* Extend the buffer by twice its current size via realloc and
1629 reset the pointers that pointed into the old block to point to the
1630 correct places in the new one. If extending the buffer results in it
1631 being larger than MAX_BUF_SIZE, then flag memory exhausted. */
1632 #define EXTEND_BUFFER() \
1634 unsigned char *old_buffer = bufp->buffer; \
1635 if (bufp->allocated == MAX_BUF_SIZE) \
1637 bufp->allocated <<= 1; \
1638 if (bufp->allocated > MAX_BUF_SIZE) \
1639 bufp->allocated = MAX_BUF_SIZE; \
1640 bufp->buffer = (unsigned char *) REALLOC (bufp->buffer, bufp->allocated);\
1641 if (bufp->buffer == NULL) \
1642 return REG_ESPACE; \
1643 /* If the buffer moved, move all the pointers into it. */ \
1644 if (old_buffer != bufp->buffer) \
1646 b = (b - old_buffer) + bufp->buffer; \
1647 begalt = (begalt - old_buffer) + bufp->buffer; \
1648 if (fixup_alt_jump) \
1649 fixup_alt_jump = (fixup_alt_jump - old_buffer) + bufp->buffer;\
1651 laststart = (laststart - old_buffer) + bufp->buffer; \
1652 if (pending_exact) \
1653 pending_exact = (pending_exact - old_buffer) + bufp->buffer; \
1658 /* Since we have one byte reserved for the register number argument to
1659 {start,stop}_memory, the maximum number of groups we can report
1660 things about is what fits in that byte. */
1661 #define MAX_REGNUM 255
1663 /* But patterns can have more than `MAX_REGNUM' registers. We just
1664 ignore the excess. */
1665 typedef unsigned regnum_t;
1668 /* Macros for the compile stack. */
1670 /* Since offsets can go either forwards or backwards, this type needs to
1671 be able to hold values from -(MAX_BUF_SIZE - 1) to MAX_BUF_SIZE - 1. */
1672 /* int may be not enough when sizeof(int) == 2. */
1673 typedef long pattern_offset_t;
1677 pattern_offset_t begalt_offset;
1678 pattern_offset_t fixup_alt_jump;
1679 pattern_offset_t inner_group_offset;
1680 pattern_offset_t laststart_offset;
1682 } compile_stack_elt_t;
1687 compile_stack_elt_t *stack;
1689 unsigned avail; /* Offset of next open position. */
1690 } compile_stack_type;
1693 #define INIT_COMPILE_STACK_SIZE 32
1695 #define COMPILE_STACK_EMPTY (compile_stack.avail == 0)
1696 #define COMPILE_STACK_FULL (compile_stack.avail == compile_stack.size)
1698 /* The next available element. */
1699 #define COMPILE_STACK_TOP (compile_stack.stack[compile_stack.avail])
1702 /* Set the bit for character C in a list. */
1703 #define SET_LIST_BIT(c) \
1704 (b[((unsigned char) (c)) / BYTEWIDTH] \
1705 |= 1 << (((unsigned char) c) % BYTEWIDTH))
1708 /* Get the next unsigned number in the uncompiled pattern. */
1709 #define GET_UNSIGNED_NUMBER(num) \
1713 while (ISDIGIT (c)) \
1717 num = num * 10 + c - '0'; \
1725 #if WIDE_CHAR_SUPPORT
1726 /* The GNU C library provides support for user-defined character classes
1727 and the functions from ISO C amendement 1. */
1728 # ifdef CHARCLASS_NAME_MAX
1729 # define CHAR_CLASS_MAX_LENGTH CHARCLASS_NAME_MAX
1731 /* This shouldn't happen but some implementation might still have this
1732 problem. Use a reasonable default value. */
1733 # define CHAR_CLASS_MAX_LENGTH 256
1737 # define IS_CHAR_CLASS(string) __wctype (string)
1739 # define IS_CHAR_CLASS(string) wctype (string)
1742 # define CHAR_CLASS_MAX_LENGTH 6 /* Namely, `xdigit'. */
1744 # define IS_CHAR_CLASS(string) \
1745 (STREQ (string, "alpha") || STREQ (string, "upper") \
1746 || STREQ (string, "lower") || STREQ (string, "digit") \
1747 || STREQ (string, "alnum") || STREQ (string, "xdigit") \
1748 || STREQ (string, "space") || STREQ (string, "print") \
1749 || STREQ (string, "punct") || STREQ (string, "graph") \
1750 || STREQ (string, "cntrl") || STREQ (string, "blank"))
1753 #ifndef MATCH_MAY_ALLOCATE
1755 /* If we cannot allocate large objects within re_match_2_internal,
1756 we make the fail stack and register vectors global.
1757 The fail stack, we grow to the maximum size when a regexp
1759 The register vectors, we adjust in size each time we
1760 compile a regexp, according to the number of registers it needs. */
1762 static fail_stack_type fail_stack;
1764 /* Size with which the following vectors are currently allocated.
1765 That is so we can make them bigger as needed,
1766 but never make them smaller. */
1767 static int regs_allocated_size;
1769 static const char ** regstart, ** regend;
1770 static const char ** old_regstart, ** old_regend;
1771 static const char **best_regstart, **best_regend;
1772 static register_info_type *reg_info;
1773 static const char **reg_dummy;
1774 static register_info_type *reg_info_dummy;
1776 /* Make the register vectors big enough for NUM_REGS registers,
1777 but don't make them smaller. */
1780 regex_grow_registers (num_regs)
1783 if (num_regs > regs_allocated_size)
1785 RETALLOC_IF (regstart, num_regs, const char *);
1786 RETALLOC_IF (regend, num_regs, const char *);
1787 RETALLOC_IF (old_regstart, num_regs, const char *);
1788 RETALLOC_IF (old_regend, num_regs, const char *);
1789 RETALLOC_IF (best_regstart, num_regs, const char *);
1790 RETALLOC_IF (best_regend, num_regs, const char *);
1791 RETALLOC_IF (reg_info, num_regs, register_info_type);
1792 RETALLOC_IF (reg_dummy, num_regs, const char *);
1793 RETALLOC_IF (reg_info_dummy, num_regs, register_info_type);
1795 regs_allocated_size = num_regs;
1799 #endif /* not MATCH_MAY_ALLOCATE */
1801 static boolean group_in_compile_stack _RE_ARGS ((compile_stack_type
1805 /* `regex_compile' compiles PATTERN (of length SIZE) according to SYNTAX.
1806 Returns one of error codes defined in `regex.h', or zero for success.
1808 Assumes the `allocated' (and perhaps `buffer') and `translate'
1809 fields are set in BUFP on entry.
1811 If it succeeds, results are put in BUFP (if it returns an error, the
1812 contents of BUFP are undefined):
1813 `buffer' is the compiled pattern;
1814 `syntax' is set to SYNTAX;
1815 `used' is set to the length of the compiled pattern;
1816 `fastmap_accurate' is zero;
1817 `re_nsub' is the number of subexpressions in PATTERN;
1818 `not_bol' and `not_eol' are zero;
1820 The `fastmap' and `newline_anchor' fields are neither
1821 examined nor set. */
1823 /* Return, freeing storage we allocated. */
1824 #define FREE_STACK_RETURN(value) \
1825 return (free (compile_stack.stack), value)
1827 static reg_errcode_t
1828 regex_compile (pattern, size, syntax, bufp)
1829 const char *pattern;
1831 reg_syntax_t syntax;
1832 struct re_pattern_buffer *bufp;
1834 /* We fetch characters from PATTERN here. Even though PATTERN is
1835 `char *' (i.e., signed), we declare these variables as unsigned, so
1836 they can be reliably used as array indices. */
1837 register unsigned char c, c1;
1839 /* A random temporary spot in PATTERN. */
1842 /* Points to the end of the buffer, where we should append. */
1843 register unsigned char *b;
1845 /* Keeps track of unclosed groups. */
1846 compile_stack_type compile_stack;
1848 /* Points to the current (ending) position in the pattern. */
1849 const char *p = pattern;
1850 const char *pend = pattern + size;
1852 /* How to translate the characters in the pattern. */
1853 RE_TRANSLATE_TYPE translate = bufp->translate;
1855 /* Address of the count-byte of the most recently inserted `exactn'
1856 command. This makes it possible to tell if a new exact-match
1857 character can be added to that command or if the character requires
1858 a new `exactn' command. */
1859 unsigned char *pending_exact = 0;
1861 /* Address of start of the most recently finished expression.
1862 This tells, e.g., postfix * where to find the start of its
1863 operand. Reset at the beginning of groups and alternatives. */
1864 unsigned char *laststart = 0;
1866 /* Address of beginning of regexp, or inside of last group. */
1867 unsigned char *begalt;
1869 /* Place in the uncompiled pattern (i.e., the {) to
1870 which to go back if the interval is invalid. */
1871 const char *beg_interval;
1873 /* Address of the place where a forward jump should go to the end of
1874 the containing expression. Each alternative of an `or' -- except the
1875 last -- ends with a forward jump of this sort. */
1876 unsigned char *fixup_alt_jump = 0;
1878 /* Counts open-groups as they are encountered. Remembered for the
1879 matching close-group on the compile stack, so the same register
1880 number is put in the stop_memory as the start_memory. */
1881 regnum_t regnum = 0;
1884 DEBUG_PRINT1 ("\nCompiling pattern: ");
1887 unsigned debug_count;
1889 for (debug_count = 0; debug_count < size; debug_count++)
1890 putchar (pattern[debug_count]);
1895 /* Initialize the compile stack. */
1896 compile_stack.stack = TALLOC (INIT_COMPILE_STACK_SIZE, compile_stack_elt_t);
1897 if (compile_stack.stack == NULL)
1900 compile_stack.size = INIT_COMPILE_STACK_SIZE;
1901 compile_stack.avail = 0;
1903 /* Initialize the pattern buffer. */
1904 bufp->syntax = syntax;
1905 bufp->fastmap_accurate = 0;
1906 bufp->not_bol = bufp->not_eol = 0;
1908 /* Set `used' to zero, so that if we return an error, the pattern
1909 printer (for debugging) will think there's no pattern. We reset it
1913 /* Always count groups, whether or not bufp->no_sub is set. */
1916 #if !defined emacs && !defined SYNTAX_TABLE
1917 /* Initialize the syntax table. */
1918 init_syntax_once ();
1921 if (bufp->allocated == 0)
1924 { /* If zero allocated, but buffer is non-null, try to realloc
1925 enough space. This loses if buffer's address is bogus, but
1926 that is the user's responsibility. */
1927 RETALLOC (bufp->buffer, INIT_BUF_SIZE, unsigned char);
1930 { /* Caller did not allocate a buffer. Do it for them. */
1931 bufp->buffer = TALLOC (INIT_BUF_SIZE, unsigned char);
1933 if (!bufp->buffer) FREE_STACK_RETURN (REG_ESPACE);
1935 bufp->allocated = INIT_BUF_SIZE;
1938 begalt = b = bufp->buffer;
1940 /* Loop through the uncompiled pattern until we're at the end. */
1949 if ( /* If at start of pattern, it's an operator. */
1951 /* If context independent, it's an operator. */
1952 || syntax & RE_CONTEXT_INDEP_ANCHORS
1953 /* Otherwise, depends on what's come before. */
1954 || at_begline_loc_p (pattern, p, syntax))
1964 if ( /* If at end of pattern, it's an operator. */
1966 /* If context independent, it's an operator. */
1967 || syntax & RE_CONTEXT_INDEP_ANCHORS
1968 /* Otherwise, depends on what's next. */
1969 || at_endline_loc_p (p, pend, syntax))
1979 if ((syntax & RE_BK_PLUS_QM)
1980 || (syntax & RE_LIMITED_OPS))
1984 /* If there is no previous pattern... */
1987 if (syntax & RE_CONTEXT_INVALID_OPS)
1988 FREE_STACK_RETURN (REG_BADRPT);
1989 else if (!(syntax & RE_CONTEXT_INDEP_OPS))
1994 /* Are we optimizing this jump? */
1995 boolean keep_string_p = false;
1997 /* 1 means zero (many) matches is allowed. */
1998 char zero_times_ok = 0, many_times_ok = 0;
2000 /* If there is a sequence of repetition chars, collapse it
2001 down to just one (the right one). We can't combine
2002 interval operators with these because of, e.g., `a{2}*',
2003 which should only match an even number of `a's. */
2007 zero_times_ok |= c != '+';
2008 many_times_ok |= c != '?';
2016 || (!(syntax & RE_BK_PLUS_QM) && (c == '+' || c == '?')))
2019 else if (syntax & RE_BK_PLUS_QM && c == '\\')
2021 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
2024 if (!(c1 == '+' || c1 == '?'))
2039 /* If we get here, we found another repeat character. */
2042 /* Star, etc. applied to an empty pattern is equivalent
2043 to an empty pattern. */
2047 /* Now we know whether or not zero matches is allowed
2048 and also whether or not two or more matches is allowed. */
2050 { /* More than one repetition is allowed, so put in at the
2051 end a backward relative jump from `b' to before the next
2052 jump we're going to put in below (which jumps from
2053 laststart to after this jump).
2055 But if we are at the `*' in the exact sequence `.*\n',
2056 insert an unconditional jump backwards to the .,
2057 instead of the beginning of the loop. This way we only
2058 push a failure point once, instead of every time
2059 through the loop. */
2060 assert (p - 1 > pattern);
2062 /* Allocate the space for the jump. */
2063 GET_BUFFER_SPACE (3);
2065 /* We know we are not at the first character of the pattern,
2066 because laststart was nonzero. And we've already
2067 incremented `p', by the way, to be the character after
2068 the `*'. Do we have to do something analogous here
2069 for null bytes, because of RE_DOT_NOT_NULL? */
2070 if (TRANSLATE (*(p - 2)) == TRANSLATE ('.')
2072 && p < pend && TRANSLATE (*p) == TRANSLATE ('\n')
2073 && !(syntax & RE_DOT_NEWLINE))
2074 { /* We have .*\n. */
2075 STORE_JUMP (jump, b, laststart);
2076 keep_string_p = true;
2079 /* Anything else. */
2080 STORE_JUMP (maybe_pop_jump, b, laststart - 3);
2082 /* We've added more stuff to the buffer. */
2086 /* On failure, jump from laststart to b + 3, which will be the
2087 end of the buffer after this jump is inserted. */
2088 GET_BUFFER_SPACE (3);
2089 INSERT_JUMP (keep_string_p ? on_failure_keep_string_jump
2097 /* At least one repetition is required, so insert a
2098 `dummy_failure_jump' before the initial
2099 `on_failure_jump' instruction of the loop. This
2100 effects a skip over that instruction the first time
2101 we hit that loop. */
2102 GET_BUFFER_SPACE (3);
2103 INSERT_JUMP (dummy_failure_jump, laststart, laststart + 6);
2118 boolean had_char_class = false;
2120 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2122 /* Ensure that we have enough space to push a charset: the
2123 opcode, the length count, and the bitset; 34 bytes in all. */
2124 GET_BUFFER_SPACE (34);
2128 /* We test `*p == '^' twice, instead of using an if
2129 statement, so we only need one BUF_PUSH. */
2130 BUF_PUSH (*p == '^' ? charset_not : charset);
2134 /* Remember the first position in the bracket expression. */
2137 /* Push the number of bytes in the bitmap. */
2138 BUF_PUSH ((1 << BYTEWIDTH) / BYTEWIDTH);
2140 /* Clear the whole map. */
2141 bzero (b, (1 << BYTEWIDTH) / BYTEWIDTH);
2143 /* charset_not matches newline according to a syntax bit. */
2144 if ((re_opcode_t) b[-2] == charset_not
2145 && (syntax & RE_HAT_LISTS_NOT_NEWLINE))
2146 SET_LIST_BIT ('\n');
2148 /* Read in characters and ranges, setting map bits. */
2151 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2155 /* \ might escape characters inside [...] and [^...]. */
2156 if ((syntax & RE_BACKSLASH_ESCAPE_IN_LISTS) && c == '\\')
2158 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
2165 /* Could be the end of the bracket expression. If it's
2166 not (i.e., when the bracket expression is `[]' so
2167 far), the ']' character bit gets set way below. */
2168 if (c == ']' && p != p1 + 1)
2171 /* Look ahead to see if it's a range when the last thing
2172 was a character class. */
2173 if (had_char_class && c == '-' && *p != ']')
2174 FREE_STACK_RETURN (REG_ERANGE);
2176 /* Look ahead to see if it's a range when the last thing
2177 was a character: if this is a hyphen not at the
2178 beginning or the end of a list, then it's the range
2181 && !(p - 2 >= pattern && p[-2] == '[')
2182 && !(p - 3 >= pattern && p[-3] == '[' && p[-2] == '^')
2186 = compile_range (&p, pend, translate, syntax, b);
2187 if (ret != REG_NOERROR) FREE_STACK_RETURN (ret);
2190 else if (p[0] == '-' && p[1] != ']')
2191 { /* This handles ranges made up of characters only. */
2194 /* Move past the `-'. */
2197 ret = compile_range (&p, pend, translate, syntax, b);
2198 if (ret != REG_NOERROR) FREE_STACK_RETURN (ret);
2201 /* See if we're at the beginning of a possible character
2204 else if (syntax & RE_CHAR_CLASSES && c == '[' && *p == ':')
2205 { /* Leave room for the null. */
2206 char str[CHAR_CLASS_MAX_LENGTH + 1];
2211 /* If pattern is `[[:'. */
2212 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2217 if ((c == ':' && *p == ']') || p == pend
2218 || c1 == CHAR_CLASS_MAX_LENGTH)
2224 /* If isn't a word bracketed by `[:' and `:]':
2225 undo the ending character, the letters, and leave
2226 the leading `:' and `[' (but set bits for them). */
2227 if (c == ':' && *p == ']')
2229 #if WIDE_CHAR_SUPPORT
2230 boolean is_lower = STREQ (str, "lower");
2231 boolean is_upper = STREQ (str, "upper");
2235 wt = IS_CHAR_CLASS (str);
2237 FREE_STACK_RETURN (REG_ECTYPE);
2239 /* Throw away the ] at the end of the character
2243 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2245 for (ch = 0; ch < 1 << BYTEWIDTH; ++ch)
2248 if (__iswctype (__btowc (ch), wt))
2251 if (iswctype (btowc (ch), wt))
2255 if (translate && (is_upper || is_lower)
2256 && (ISUPPER (ch) || ISLOWER (ch)))
2260 had_char_class = true;
2263 boolean is_alnum = STREQ (str, "alnum");
2264 boolean is_alpha = STREQ (str, "alpha");
2265 boolean is_blank = STREQ (str, "blank");
2266 boolean is_cntrl = STREQ (str, "cntrl");
2267 boolean is_digit = STREQ (str, "digit");
2268 boolean is_graph = STREQ (str, "graph");
2269 boolean is_lower = STREQ (str, "lower");
2270 boolean is_print = STREQ (str, "print");
2271 boolean is_punct = STREQ (str, "punct");
2272 boolean is_space = STREQ (str, "space");
2273 boolean is_upper = STREQ (str, "upper");
2274 boolean is_xdigit = STREQ (str, "xdigit");
2276 if (!IS_CHAR_CLASS (str))
2277 FREE_STACK_RETURN (REG_ECTYPE);
2279 /* Throw away the ] at the end of the character
2283 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2285 for (ch = 0; ch < 1 << BYTEWIDTH; ch++)
2287 /* This was split into 3 if's to
2288 avoid an arbitrary limit in some compiler. */
2289 if ( (is_alnum && ISALNUM (ch))
2290 || (is_alpha && ISALPHA (ch))
2291 || (is_blank && ISBLANK (ch))
2292 || (is_cntrl && ISCNTRL (ch)))
2294 if ( (is_digit && ISDIGIT (ch))
2295 || (is_graph && ISGRAPH (ch))
2296 || (is_lower && ISLOWER (ch))
2297 || (is_print && ISPRINT (ch)))
2299 if ( (is_punct && ISPUNCT (ch))
2300 || (is_space && ISSPACE (ch))
2301 || (is_upper && ISUPPER (ch))
2302 || (is_xdigit && ISXDIGIT (ch)))
2304 if ( translate && (is_upper || is_lower)
2305 && (ISUPPER (ch) || ISLOWER (ch)))
2308 had_char_class = true;
2309 #endif /* libc || wctype.h */
2318 had_char_class = false;
2323 had_char_class = false;
2328 /* Discard any (non)matching list bytes that are all 0 at the
2329 end of the map. Decrease the map-length byte too. */
2330 while ((int) b[-1] > 0 && b[b[-1] - 1] == 0)
2338 if (syntax & RE_NO_BK_PARENS)
2345 if (syntax & RE_NO_BK_PARENS)
2352 if (syntax & RE_NEWLINE_ALT)
2359 if (syntax & RE_NO_BK_VBAR)
2366 if (syntax & RE_INTERVALS && syntax & RE_NO_BK_BRACES)
2367 goto handle_interval;
2373 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
2375 /* Do not translate the character after the \, so that we can
2376 distinguish, e.g., \B from \b, even if we normally would
2377 translate, e.g., B to b. */
2383 if (syntax & RE_NO_BK_PARENS)
2384 goto normal_backslash;
2390 if (COMPILE_STACK_FULL)
2392 RETALLOC (compile_stack.stack, compile_stack.size << 1,
2393 compile_stack_elt_t);
2394 if (compile_stack.stack == NULL) return REG_ESPACE;
2396 compile_stack.size <<= 1;
2399 /* These are the values to restore when we hit end of this
2400 group. They are all relative offsets, so that if the
2401 whole pattern moves because of realloc, they will still
2403 COMPILE_STACK_TOP.begalt_offset = begalt - bufp->buffer;
2404 COMPILE_STACK_TOP.fixup_alt_jump
2405 = fixup_alt_jump ? fixup_alt_jump - bufp->buffer + 1 : 0;
2406 COMPILE_STACK_TOP.laststart_offset = b - bufp->buffer;
2407 COMPILE_STACK_TOP.regnum = regnum;
2409 /* We will eventually replace the 0 with the number of
2410 groups inner to this one. But do not push a
2411 start_memory for groups beyond the last one we can
2412 represent in the compiled pattern. */
2413 if (regnum <= MAX_REGNUM)
2415 COMPILE_STACK_TOP.inner_group_offset = b - bufp->buffer + 2;
2416 BUF_PUSH_3 (start_memory, regnum, 0);
2419 compile_stack.avail++;
2424 /* If we've reached MAX_REGNUM groups, then this open
2425 won't actually generate any code, so we'll have to
2426 clear pending_exact explicitly. */
2432 if (syntax & RE_NO_BK_PARENS) goto normal_backslash;
2434 if (COMPILE_STACK_EMPTY)
2436 if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD)
2437 goto normal_backslash;
2439 FREE_STACK_RETURN (REG_ERPAREN);
2444 { /* Push a dummy failure point at the end of the
2445 alternative for a possible future
2446 `pop_failure_jump' to pop. See comments at
2447 `push_dummy_failure' in `re_match_2'. */
2448 BUF_PUSH (push_dummy_failure);
2450 /* We allocated space for this jump when we assigned
2451 to `fixup_alt_jump', in the `handle_alt' case below. */
2452 STORE_JUMP (jump_past_alt, fixup_alt_jump, b - 1);
2455 /* See similar code for backslashed left paren above. */
2456 if (COMPILE_STACK_EMPTY)
2458 if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD)
2461 FREE_STACK_RETURN (REG_ERPAREN);
2464 /* Since we just checked for an empty stack above, this
2465 ``can't happen''. */
2466 assert (compile_stack.avail != 0);
2468 /* We don't just want to restore into `regnum', because
2469 later groups should continue to be numbered higher,
2470 as in `(ab)c(de)' -- the second group is #2. */
2471 regnum_t this_group_regnum;
2473 compile_stack.avail--;
2474 begalt = bufp->buffer + COMPILE_STACK_TOP.begalt_offset;
2476 = COMPILE_STACK_TOP.fixup_alt_jump
2477 ? bufp->buffer + COMPILE_STACK_TOP.fixup_alt_jump - 1
2479 laststart = bufp->buffer + COMPILE_STACK_TOP.laststart_offset;
2480 this_group_regnum = COMPILE_STACK_TOP.regnum;
2481 /* If we've reached MAX_REGNUM groups, then this open
2482 won't actually generate any code, so we'll have to
2483 clear pending_exact explicitly. */
2486 /* We're at the end of the group, so now we know how many
2487 groups were inside this one. */
2488 if (this_group_regnum <= MAX_REGNUM)
2490 unsigned char *inner_group_loc
2491 = bufp->buffer + COMPILE_STACK_TOP.inner_group_offset;
2493 *inner_group_loc = regnum - this_group_regnum;
2494 BUF_PUSH_3 (stop_memory, this_group_regnum,
2495 regnum - this_group_regnum);
2501 case '|': /* `\|'. */
2502 if (syntax & RE_LIMITED_OPS || syntax & RE_NO_BK_VBAR)
2503 goto normal_backslash;
2505 if (syntax & RE_LIMITED_OPS)
2508 /* Insert before the previous alternative a jump which
2509 jumps to this alternative if the former fails. */
2510 GET_BUFFER_SPACE (3);
2511 INSERT_JUMP (on_failure_jump, begalt, b + 6);
2515 /* The alternative before this one has a jump after it
2516 which gets executed if it gets matched. Adjust that
2517 jump so it will jump to this alternative's analogous
2518 jump (put in below, which in turn will jump to the next
2519 (if any) alternative's such jump, etc.). The last such
2520 jump jumps to the correct final destination. A picture:
2526 If we are at `b', then fixup_alt_jump right now points to a
2527 three-byte space after `a'. We'll put in the jump, set
2528 fixup_alt_jump to right after `b', and leave behind three
2529 bytes which we'll fill in when we get to after `c'. */
2532 STORE_JUMP (jump_past_alt, fixup_alt_jump, b);
2534 /* Mark and leave space for a jump after this alternative,
2535 to be filled in later either by next alternative or
2536 when know we're at the end of a series of alternatives. */
2538 GET_BUFFER_SPACE (3);
2547 /* If \{ is a literal. */
2548 if (!(syntax & RE_INTERVALS)
2549 /* If we're at `\{' and it's not the open-interval
2551 || ((syntax & RE_INTERVALS) && (syntax & RE_NO_BK_BRACES))
2552 || (p - 2 == pattern && p == pend))
2553 goto normal_backslash;
2557 /* If got here, then the syntax allows intervals. */
2559 /* At least (most) this many matches must be made. */
2560 int lower_bound = -1, upper_bound = -1;
2562 beg_interval = p - 1;
2566 if (syntax & RE_NO_BK_BRACES)
2567 goto unfetch_interval;
2569 FREE_STACK_RETURN (REG_EBRACE);
2572 GET_UNSIGNED_NUMBER (lower_bound);
2576 GET_UNSIGNED_NUMBER (upper_bound);
2577 if (upper_bound < 0) upper_bound = RE_DUP_MAX;
2580 /* Interval such as `{1}' => match exactly once. */
2581 upper_bound = lower_bound;
2583 if (lower_bound < 0 || upper_bound > RE_DUP_MAX
2584 || lower_bound > upper_bound)
2586 if (syntax & RE_NO_BK_BRACES)
2587 goto unfetch_interval;
2589 FREE_STACK_RETURN (REG_BADBR);
2592 if (!(syntax & RE_NO_BK_BRACES))
2594 if (c != '\\') FREE_STACK_RETURN (REG_EBRACE);
2601 if (syntax & RE_NO_BK_BRACES)
2602 goto unfetch_interval;
2604 FREE_STACK_RETURN (REG_BADBR);
2607 /* We just parsed a valid interval. */
2609 /* If it's invalid to have no preceding re. */
2612 if (syntax & RE_CONTEXT_INVALID_OPS)
2613 FREE_STACK_RETURN (REG_BADRPT);
2614 else if (syntax & RE_CONTEXT_INDEP_OPS)
2617 goto unfetch_interval;
2620 /* If the upper bound is zero, don't want to succeed at
2621 all; jump from `laststart' to `b + 3', which will be
2622 the end of the buffer after we insert the jump. */
2623 if (upper_bound == 0)
2625 GET_BUFFER_SPACE (3);
2626 INSERT_JUMP (jump, laststart, b + 3);
2630 /* Otherwise, we have a nontrivial interval. When
2631 we're all done, the pattern will look like:
2632 set_number_at <jump count> <upper bound>
2633 set_number_at <succeed_n count> <lower bound>
2634 succeed_n <after jump addr> <succeed_n count>
2636 jump_n <succeed_n addr> <jump count>
2637 (The upper bound and `jump_n' are omitted if
2638 `upper_bound' is 1, though.) */
2640 { /* If the upper bound is > 1, we need to insert
2641 more at the end of the loop. */
2642 unsigned nbytes = 10 + (upper_bound > 1) * 10;
2644 GET_BUFFER_SPACE (nbytes);
2646 /* Initialize lower bound of the `succeed_n', even
2647 though it will be set during matching by its
2648 attendant `set_number_at' (inserted next),
2649 because `re_compile_fastmap' needs to know.
2650 Jump to the `jump_n' we might insert below. */
2651 INSERT_JUMP2 (succeed_n, laststart,
2652 b + 5 + (upper_bound > 1) * 5,
2656 /* Code to initialize the lower bound. Insert
2657 before the `succeed_n'. The `5' is the last two
2658 bytes of this `set_number_at', plus 3 bytes of
2659 the following `succeed_n'. */
2660 insert_op2 (set_number_at, laststart, 5, lower_bound, b);
2663 if (upper_bound > 1)
2664 { /* More than one repetition is allowed, so
2665 append a backward jump to the `succeed_n'
2666 that starts this interval.
2668 When we've reached this during matching,
2669 we'll have matched the interval once, so
2670 jump back only `upper_bound - 1' times. */
2671 STORE_JUMP2 (jump_n, b, laststart + 5,
2675 /* The location we want to set is the second
2676 parameter of the `jump_n'; that is `b-2' as
2677 an absolute address. `laststart' will be
2678 the `set_number_at' we're about to insert;
2679 `laststart+3' the number to set, the source
2680 for the relative address. But we are
2681 inserting into the middle of the pattern --
2682 so everything is getting moved up by 5.
2683 Conclusion: (b - 2) - (laststart + 3) + 5,
2684 i.e., b - laststart.
2686 We insert this at the beginning of the loop
2687 so that if we fail during matching, we'll
2688 reinitialize the bounds. */
2689 insert_op2 (set_number_at, laststart, b - laststart,
2690 upper_bound - 1, b);
2695 beg_interval = NULL;
2700 /* If an invalid interval, match the characters as literals. */
2701 assert (beg_interval);
2703 beg_interval = NULL;
2705 /* normal_char and normal_backslash need `c'. */
2708 if (!(syntax & RE_NO_BK_BRACES))
2710 if (p > pattern && p[-1] == '\\')
2711 goto normal_backslash;
2716 /* There is no way to specify the before_dot and after_dot
2717 operators. rms says this is ok. --karl */
2725 BUF_PUSH_2 (syntaxspec, syntax_spec_code[c]);
2731 BUF_PUSH_2 (notsyntaxspec, syntax_spec_code[c]);
2737 if (syntax & RE_NO_GNU_OPS)
2740 BUF_PUSH (wordchar);
2745 if (syntax & RE_NO_GNU_OPS)
2748 BUF_PUSH (notwordchar);
2753 if (syntax & RE_NO_GNU_OPS)
2759 if (syntax & RE_NO_GNU_OPS)
2765 if (syntax & RE_NO_GNU_OPS)
2767 BUF_PUSH (wordbound);
2771 if (syntax & RE_NO_GNU_OPS)
2773 BUF_PUSH (notwordbound);
2777 if (syntax & RE_NO_GNU_OPS)
2783 if (syntax & RE_NO_GNU_OPS)
2788 case '1': case '2': case '3': case '4': case '5':
2789 case '6': case '7': case '8': case '9':
2790 if (syntax & RE_NO_BK_REFS)
2796 FREE_STACK_RETURN (REG_ESUBREG);
2798 /* Can't back reference to a subexpression if inside of it. */
2799 if (group_in_compile_stack (compile_stack, (regnum_t) c1))
2803 BUF_PUSH_2 (duplicate, c1);
2809 if (syntax & RE_BK_PLUS_QM)
2812 goto normal_backslash;
2816 /* You might think it would be useful for \ to mean
2817 not to translate; but if we don't translate it
2818 it will never match anything. */
2826 /* Expects the character in `c'. */
2828 /* If no exactn currently being built. */
2831 /* If last exactn not at current position. */
2832 || pending_exact + *pending_exact + 1 != b
2834 /* We have only one byte following the exactn for the count. */
2835 || *pending_exact == (1 << BYTEWIDTH) - 1
2837 /* If followed by a repetition operator. */
2838 || *p == '*' || *p == '^'
2839 || ((syntax & RE_BK_PLUS_QM)
2840 ? *p == '\\' && (p[1] == '+' || p[1] == '?')
2841 : (*p == '+' || *p == '?'))
2842 || ((syntax & RE_INTERVALS)
2843 && ((syntax & RE_NO_BK_BRACES)
2845 : (p[0] == '\\' && p[1] == '{'))))
2847 /* Start building a new exactn. */
2851 BUF_PUSH_2 (exactn, 0);
2852 pending_exact = b - 1;
2859 } /* while p != pend */
2862 /* Through the pattern now. */
2865 STORE_JUMP (jump_past_alt, fixup_alt_jump, b);
2867 if (!COMPILE_STACK_EMPTY)
2868 FREE_STACK_RETURN (REG_EPAREN);
2870 /* If we don't want backtracking, force success
2871 the first time we reach the end of the compiled pattern. */
2872 if (syntax & RE_NO_POSIX_BACKTRACKING)
2875 free (compile_stack.stack);
2877 /* We have succeeded; set the length of the buffer. */
2878 bufp->used = b - bufp->buffer;
2883 DEBUG_PRINT1 ("\nCompiled pattern: \n");
2884 print_compiled_pattern (bufp);
2888 #ifndef MATCH_MAY_ALLOCATE
2889 /* Initialize the failure stack to the largest possible stack. This
2890 isn't necessary unless we're trying to avoid calling alloca in
2891 the search and match routines. */
2893 int num_regs = bufp->re_nsub + 1;
2895 /* Since DOUBLE_FAIL_STACK refuses to double only if the current size
2896 is strictly greater than re_max_failures, the largest possible stack
2897 is 2 * re_max_failures failure points. */
2898 if (fail_stack.size < (2 * re_max_failures * MAX_FAILURE_ITEMS))
2900 fail_stack.size = (2 * re_max_failures * MAX_FAILURE_ITEMS);
2903 if (! fail_stack.stack)
2905 = (fail_stack_elt_t *) xmalloc (fail_stack.size
2906 * sizeof (fail_stack_elt_t));
2909 = (fail_stack_elt_t *) xrealloc (fail_stack.stack,
2911 * sizeof (fail_stack_elt_t)));
2912 # else /* not emacs */
2913 if (! fail_stack.stack)
2915 = (fail_stack_elt_t *) malloc (fail_stack.size
2916 * sizeof (fail_stack_elt_t));
2919 = (fail_stack_elt_t *) realloc (fail_stack.stack,
2921 * sizeof (fail_stack_elt_t)));
2922 # endif /* not emacs */
2925 regex_grow_registers (num_regs);
2927 #endif /* not MATCH_MAY_ALLOCATE */
2930 } /* regex_compile */
2932 /* Subroutines for `regex_compile'. */
2934 /* Store OP at LOC followed by two-byte integer parameter ARG. */
2937 store_op1 (op, loc, arg)
2942 *loc = (unsigned char) op;
2943 STORE_NUMBER (loc + 1, arg);
2947 /* Like `store_op1', but for two two-byte parameters ARG1 and ARG2. */
2950 store_op2 (op, loc, arg1, arg2)
2955 *loc = (unsigned char) op;
2956 STORE_NUMBER (loc + 1, arg1);
2957 STORE_NUMBER (loc + 3, arg2);
2961 /* Copy the bytes from LOC to END to open up three bytes of space at LOC
2962 for OP followed by two-byte integer parameter ARG. */
2965 insert_op1 (op, loc, arg, end)
2971 register unsigned char *pfrom = end;
2972 register unsigned char *pto = end + 3;
2974 while (pfrom != loc)
2977 store_op1 (op, loc, arg);
2981 /* Like `insert_op1', but for two two-byte parameters ARG1 and ARG2. */
2984 insert_op2 (op, loc, arg1, arg2, end)
2990 register unsigned char *pfrom = end;
2991 register unsigned char *pto = end + 5;
2993 while (pfrom != loc)
2996 store_op2 (op, loc, arg1, arg2);
3000 /* P points to just after a ^ in PATTERN. Return true if that ^ comes
3001 after an alternative or a begin-subexpression. We assume there is at
3002 least one character before the ^. */
3005 at_begline_loc_p (pattern, p, syntax)
3006 const char *pattern, *p;
3007 reg_syntax_t syntax;
3009 const char *prev = p - 2;
3010 boolean prev_prev_backslash = prev > pattern && prev[-1] == '\\';
3013 /* After a subexpression? */
3014 (*prev == '(' && (syntax & RE_NO_BK_PARENS || prev_prev_backslash))
3015 /* After an alternative? */
3016 || (*prev == '|' && (syntax & RE_NO_BK_VBAR || prev_prev_backslash));
3020 /* The dual of at_begline_loc_p. This one is for $. We assume there is
3021 at least one character after the $, i.e., `P < PEND'. */
3024 at_endline_loc_p (p, pend, syntax)
3025 const char *p, *pend;
3026 reg_syntax_t syntax;
3028 const char *next = p;
3029 boolean next_backslash = *next == '\\';
3030 const char *next_next = p + 1 < pend ? p + 1 : 0;
3033 /* Before a subexpression? */
3034 (syntax & RE_NO_BK_PARENS ? *next == ')'
3035 : next_backslash && next_next && *next_next == ')')
3036 /* Before an alternative? */
3037 || (syntax & RE_NO_BK_VBAR ? *next == '|'
3038 : next_backslash && next_next && *next_next == '|');
3042 /* Returns true if REGNUM is in one of COMPILE_STACK's elements and
3043 false if it's not. */
3046 group_in_compile_stack (compile_stack, regnum)
3047 compile_stack_type compile_stack;
3052 for (this_element = compile_stack.avail - 1;
3055 if (compile_stack.stack[this_element].regnum == regnum)
3062 /* Read the ending character of a range (in a bracket expression) from the
3063 uncompiled pattern *P_PTR (which ends at PEND). We assume the
3064 starting character is in `P[-2]'. (`P[-1]' is the character `-'.)
3065 Then we set the translation of all bits between the starting and
3066 ending characters (inclusive) in the compiled pattern B.
3068 Return an error code.
3070 We use these short variable names so we can use the same macros as
3071 `regex_compile' itself. */
3073 static reg_errcode_t
3074 compile_range (p_ptr, pend, translate, syntax, b)
3075 const char **p_ptr, *pend;
3076 RE_TRANSLATE_TYPE translate;
3077 reg_syntax_t syntax;
3082 const char *p = *p_ptr;
3083 unsigned int range_start, range_end;
3088 /* Even though the pattern is a signed `char *', we need to fetch
3089 with unsigned char *'s; if the high bit of the pattern character
3090 is set, the range endpoints will be negative if we fetch using a
3093 We also want to fetch the endpoints without translating them; the
3094 appropriate translation is done in the bit-setting loop below. */
3095 /* The SVR4 compiler on the 3B2 had trouble with unsigned const char *. */
3096 range_start = ((const unsigned char *) p)[-2];
3097 range_end = ((const unsigned char *) p)[0];
3099 /* Have to increment the pointer into the pattern string, so the
3100 caller isn't still at the ending character. */
3103 /* If the start is after the end, the range is empty. */
3104 if (range_start > range_end)
3105 return syntax & RE_NO_EMPTY_RANGES ? REG_ERANGE : REG_NOERROR;
3107 /* Here we see why `this_char' has to be larger than an `unsigned
3108 char' -- the range is inclusive, so if `range_end' == 0xff
3109 (assuming 8-bit characters), we would otherwise go into an infinite
3110 loop, since all characters <= 0xff. */
3111 for (this_char = range_start; this_char <= range_end; this_char++)
3113 SET_LIST_BIT (TRANSLATE (this_char));
3119 /* re_compile_fastmap computes a ``fastmap'' for the compiled pattern in
3120 BUFP. A fastmap records which of the (1 << BYTEWIDTH) possible
3121 characters can start a string that matches the pattern. This fastmap
3122 is used by re_search to skip quickly over impossible starting points.
3124 The caller must supply the address of a (1 << BYTEWIDTH)-byte data
3125 area as BUFP->fastmap.
3127 We set the `fastmap', `fastmap_accurate', and `can_be_null' fields in
3130 Returns 0 if we succeed, -2 if an internal error. */
3133 re_compile_fastmap (bufp)
3134 struct re_pattern_buffer *bufp;
3137 #ifdef MATCH_MAY_ALLOCATE
3138 fail_stack_type fail_stack;
3140 #ifndef REGEX_MALLOC
3144 register char *fastmap = bufp->fastmap;
3145 unsigned char *pattern = bufp->buffer;
3146 unsigned char *p = pattern;
3147 register unsigned char *pend = pattern + bufp->used;
3150 /* This holds the pointer to the failure stack, when
3151 it is allocated relocatably. */
3152 fail_stack_elt_t *failure_stack_ptr;
3155 /* Assume that each path through the pattern can be null until
3156 proven otherwise. We set this false at the bottom of switch
3157 statement, to which we get only if a particular path doesn't
3158 match the empty string. */
3159 boolean path_can_be_null = true;
3161 /* We aren't doing a `succeed_n' to begin with. */
3162 boolean succeed_n_p = false;
3164 assert (fastmap != NULL && p != NULL);
3167 bzero (fastmap, 1 << BYTEWIDTH); /* Assume nothing's valid. */
3168 bufp->fastmap_accurate = 1; /* It will be when we're done. */
3169 bufp->can_be_null = 0;
3173 if (p == pend || *p == succeed)
3175 /* We have reached the (effective) end of pattern. */
3176 if (!FAIL_STACK_EMPTY ())
3178 bufp->can_be_null |= path_can_be_null;
3180 /* Reset for next path. */
3181 path_can_be_null = true;
3183 p = fail_stack.stack[--fail_stack.avail].pointer;
3191 /* We should never be about to go beyond the end of the pattern. */
3194 switch (SWITCH_ENUM_CAST ((re_opcode_t) *p++))
3197 /* I guess the idea here is to simply not bother with a fastmap
3198 if a backreference is used, since it's too hard to figure out
3199 the fastmap for the corresponding group. Setting
3200 `can_be_null' stops `re_search_2' from using the fastmap, so
3201 that is all we do. */
3203 bufp->can_be_null = 1;
3207 /* Following are the cases which match a character. These end
3216 for (j = *p++ * BYTEWIDTH - 1; j >= 0; j--)
3217 if (p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH)))
3223 /* Chars beyond end of map must be allowed. */
3224 for (j = *p * BYTEWIDTH; j < (1 << BYTEWIDTH); j++)
3227 for (j = *p++ * BYTEWIDTH - 1; j >= 0; j--)
3228 if (!(p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH))))
3234 for (j = 0; j < (1 << BYTEWIDTH); j++)
3235 if (SYNTAX (j) == Sword)
3241 for (j = 0; j < (1 << BYTEWIDTH); j++)
3242 if (SYNTAX (j) != Sword)
3249 int fastmap_newline = fastmap['\n'];
3251 /* `.' matches anything ... */
3252 for (j = 0; j < (1 << BYTEWIDTH); j++)
3255 /* ... except perhaps newline. */
3256 if (!(bufp->syntax & RE_DOT_NEWLINE))
3257 fastmap['\n'] = fastmap_newline;
3259 /* Return if we have already set `can_be_null'; if we have,
3260 then the fastmap is irrelevant. Something's wrong here. */
3261 else if (bufp->can_be_null)
3264 /* Otherwise, have to check alternative paths. */
3271 for (j = 0; j < (1 << BYTEWIDTH); j++)
3272 if (SYNTAX (j) == (enum syntaxcode) k)
3279 for (j = 0; j < (1 << BYTEWIDTH); j++)
3280 if (SYNTAX (j) != (enum syntaxcode) k)
3285 /* All cases after this match the empty string. These end with
3305 case push_dummy_failure:
3310 case pop_failure_jump:
3311 case maybe_pop_jump:
3314 case dummy_failure_jump:
3315 EXTRACT_NUMBER_AND_INCR (j, p);
3320 /* Jump backward implies we just went through the body of a
3321 loop and matched nothing. Opcode jumped to should be
3322 `on_failure_jump' or `succeed_n'. Just treat it like an
3323 ordinary jump. For a * loop, it has pushed its failure
3324 point already; if so, discard that as redundant. */
3325 if ((re_opcode_t) *p != on_failure_jump
3326 && (re_opcode_t) *p != succeed_n)
3330 EXTRACT_NUMBER_AND_INCR (j, p);
3333 /* If what's on the stack is where we are now, pop it. */
3334 if (!FAIL_STACK_EMPTY ()
3335 && fail_stack.stack[fail_stack.avail - 1].pointer == p)
3341 case on_failure_jump:
3342 case on_failure_keep_string_jump:
3343 handle_on_failure_jump:
3344 EXTRACT_NUMBER_AND_INCR (j, p);
3346 /* For some patterns, e.g., `(a?)?', `p+j' here points to the
3347 end of the pattern. We don't want to push such a point,
3348 since when we restore it above, entering the switch will
3349 increment `p' past the end of the pattern. We don't need
3350 to push such a point since we obviously won't find any more
3351 fastmap entries beyond `pend'. Such a pattern can match
3352 the null string, though. */
3355 if (!PUSH_PATTERN_OP (p + j, fail_stack))
3357 RESET_FAIL_STACK ();
3362 bufp->can_be_null = 1;
3366 EXTRACT_NUMBER_AND_INCR (k, p); /* Skip the n. */
3367 succeed_n_p = false;
3374 /* Get to the number of times to succeed. */
3377 /* Increment p past the n for when k != 0. */
3378 EXTRACT_NUMBER_AND_INCR (k, p);
3382 succeed_n_p = true; /* Spaghetti code alert. */
3383 goto handle_on_failure_jump;
3400 abort (); /* We have listed all the cases. */
3403 /* Getting here means we have found the possible starting
3404 characters for one path of the pattern -- and that the empty
3405 string does not match. We need not follow this path further.
3406 Instead, look at the next alternative (remembered on the
3407 stack), or quit if no more. The test at the top of the loop
3408 does these things. */
3409 path_can_be_null = false;
3413 /* Set `can_be_null' for the last path (also the first path, if the
3414 pattern is empty). */
3415 bufp->can_be_null |= path_can_be_null;
3418 RESET_FAIL_STACK ();
3420 } /* re_compile_fastmap */
3422 weak_alias (__re_compile_fastmap, re_compile_fastmap)
3425 /* Set REGS to hold NUM_REGS registers, storing them in STARTS and
3426 ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use
3427 this memory for recording register information. STARTS and ENDS
3428 must be allocated using the malloc library routine, and must each
3429 be at least NUM_REGS * sizeof (regoff_t) bytes long.
3431 If NUM_REGS == 0, then subsequent matches should allocate their own
3434 Unless this function is called, the first search or match using
3435 PATTERN_BUFFER will allocate its own register data, without
3436 freeing the old data. */
3439 re_set_registers (bufp, regs, num_regs, starts, ends)
3440 struct re_pattern_buffer *bufp;
3441 struct re_registers *regs;
3443 regoff_t *starts, *ends;
3447 bufp->regs_allocated = REGS_REALLOCATE;
3448 regs->num_regs = num_regs;
3449 regs->start = starts;
3454 bufp->regs_allocated = REGS_UNALLOCATED;
3456 regs->start = regs->end = (regoff_t *) 0;
3460 weak_alias (__re_set_registers, re_set_registers)
3463 /* Searching routines. */
3465 /* Like re_search_2, below, but only one string is specified, and
3466 doesn't let you say where to stop matching. */
3469 re_search (bufp, string, size, startpos, range, regs)
3470 struct re_pattern_buffer *bufp;
3472 int size, startpos, range;
3473 struct re_registers *regs;
3475 return re_search_2 (bufp, NULL, 0, string, size, startpos, range,
3479 weak_alias (__re_search, re_search)
3483 /* Using the compiled pattern in BUFP->buffer, first tries to match the
3484 virtual concatenation of STRING1 and STRING2, starting first at index
3485 STARTPOS, then at STARTPOS + 1, and so on.
3487 STRING1 and STRING2 have length SIZE1 and SIZE2, respectively.
3489 RANGE is how far to scan while trying to match. RANGE = 0 means try
3490 only at STARTPOS; in general, the last start tried is STARTPOS +
3493 In REGS, return the indices of the virtual concatenation of STRING1
3494 and STRING2 that matched the entire BUFP->buffer and its contained
3497 Do not consider matching one past the index STOP in the virtual
3498 concatenation of STRING1 and STRING2.
3500 We return either the position in the strings at which the match was
3501 found, -1 if no match, or -2 if error (such as failure
3505 re_search_2 (bufp, string1, size1, string2, size2, startpos, range, regs, stop)
3506 struct re_pattern_buffer *bufp;
3507 const char *string1, *string2;
3511 struct re_registers *regs;
3515 register char *fastmap = bufp->fastmap;
3516 register RE_TRANSLATE_TYPE translate = bufp->translate;
3517 int total_size = size1 + size2;
3518 int endpos = startpos + range;
3520 /* Check for out-of-range STARTPOS. */
3521 if (startpos < 0 || startpos > total_size)
3524 /* Fix up RANGE if it might eventually take us outside
3525 the virtual concatenation of STRING1 and STRING2.
3526 Make sure we won't move STARTPOS below 0 or above TOTAL_SIZE. */
3528 range = 0 - startpos;
3529 else if (endpos > total_size)
3530 range = total_size - startpos;
3532 /* If the search isn't to be a backwards one, don't waste time in a
3533 search for a pattern that must be anchored. */
3534 if (bufp->used > 0 && range > 0
3535 && ((re_opcode_t) bufp->buffer[0] == begbuf
3536 /* `begline' is like `begbuf' if it cannot match at newlines. */
3537 || ((re_opcode_t) bufp->buffer[0] == begline
3538 && !bufp->newline_anchor)))
3547 /* In a forward search for something that starts with \=.
3548 don't keep searching past point. */
3549 if (bufp->used > 0 && (re_opcode_t) bufp->buffer[0] == at_dot && range > 0)
3551 range = PT - startpos;
3557 /* Update the fastmap now if not correct already. */
3558 if (fastmap && !bufp->fastmap_accurate)
3559 if (re_compile_fastmap (bufp) == -2)
3562 /* Loop through the string, looking for a place to start matching. */
3565 /* If a fastmap is supplied, skip quickly over characters that
3566 cannot be the start of a match. If the pattern can match the
3567 null string, however, we don't need to skip characters; we want
3568 the first null string. */
3569 if (fastmap && startpos < total_size && !bufp->can_be_null)
3571 if (range > 0) /* Searching forwards. */
3573 register const char *d;
3574 register int lim = 0;
3577 if (startpos < size1 && startpos + range >= size1)
3578 lim = range - (size1 - startpos);
3580 d = (startpos >= size1 ? string2 - size1 : string1) + startpos;
3582 /* Written out as an if-else to avoid testing `translate'
3586 && !fastmap[(unsigned char)
3587 translate[(unsigned char) *d++]])
3590 while (range > lim && !fastmap[(unsigned char) *d++])
3593 startpos += irange - range;
3595 else /* Searching backwards. */
3597 register char c = (size1 == 0 || startpos >= size1
3598 ? string2[startpos - size1]
3599 : string1[startpos]);
3601 if (!fastmap[(unsigned char) TRANSLATE (c)])
3606 /* If can't match the null string, and that's all we have left, fail. */
3607 if (range >= 0 && startpos == total_size && fastmap
3608 && !bufp->can_be_null)
3611 val = re_match_2_internal (bufp, string1, size1, string2, size2,
3612 startpos, regs, stop);
3613 #ifndef REGEX_MALLOC
3642 weak_alias (__re_search_2, re_search_2)
3645 /* This converts PTR, a pointer into one of the search strings `string1'
3646 and `string2' into an offset from the beginning of that string. */
3647 #define POINTER_TO_OFFSET(ptr) \
3648 (FIRST_STRING_P (ptr) \
3649 ? ((regoff_t) ((ptr) - string1)) \
3650 : ((regoff_t) ((ptr) - string2 + size1)))
3652 /* Macros for dealing with the split strings in re_match_2. */
3654 #define MATCHING_IN_FIRST_STRING (dend == end_match_1)
3656 /* Call before fetching a character with *d. This switches over to
3657 string2 if necessary. */
3658 #define PREFETCH() \
3661 /* End of string2 => fail. */ \
3662 if (dend == end_match_2) \
3664 /* End of string1 => advance to string2. */ \
3666 dend = end_match_2; \
3670 /* Test if at very beginning or at very end of the virtual concatenation
3671 of `string1' and `string2'. If only one string, it's `string2'. */
3672 #define AT_STRINGS_BEG(d) ((d) == (size1 ? string1 : string2) || !size2)
3673 #define AT_STRINGS_END(d) ((d) == end2)
3676 /* Test if D points to a character which is word-constituent. We have
3677 two special cases to check for: if past the end of string1, look at
3678 the first character in string2; and if before the beginning of
3679 string2, look at the last character in string1. */
3680 #define WORDCHAR_P(d) \
3681 (SYNTAX ((d) == end1 ? *string2 \
3682 : (d) == string2 - 1 ? *(end1 - 1) : *(d)) \
3685 /* Disabled due to a compiler bug -- see comment at case wordbound */
3687 /* Test if the character before D and the one at D differ with respect
3688 to being word-constituent. */
3689 #define AT_WORD_BOUNDARY(d) \
3690 (AT_STRINGS_BEG (d) || AT_STRINGS_END (d) \
3691 || WORDCHAR_P (d - 1) != WORDCHAR_P (d))
3694 /* Free everything we malloc. */
3695 #ifdef MATCH_MAY_ALLOCATE
3696 # define FREE_VAR(var) if (var) REGEX_FREE (var); var = NULL
3697 # define FREE_VARIABLES() \
3699 REGEX_FREE_STACK (fail_stack.stack); \
3700 FREE_VAR (regstart); \
3701 FREE_VAR (regend); \
3702 FREE_VAR (old_regstart); \
3703 FREE_VAR (old_regend); \
3704 FREE_VAR (best_regstart); \
3705 FREE_VAR (best_regend); \
3706 FREE_VAR (reg_info); \
3707 FREE_VAR (reg_dummy); \
3708 FREE_VAR (reg_info_dummy); \
3711 # define FREE_VARIABLES() ((void)0) /* Do nothing! But inhibit gcc warning. */
3712 #endif /* not MATCH_MAY_ALLOCATE */
3714 /* These values must meet several constraints. They must not be valid
3715 register values; since we have a limit of 255 registers (because
3716 we use only one byte in the pattern for the register number), we can
3717 use numbers larger than 255. They must differ by 1, because of
3718 NUM_FAILURE_ITEMS above. And the value for the lowest register must
3719 be larger than the value for the highest register, so we do not try
3720 to actually save any registers when none are active. */
3721 #define NO_HIGHEST_ACTIVE_REG (1 << BYTEWIDTH)
3722 #define NO_LOWEST_ACTIVE_REG (NO_HIGHEST_ACTIVE_REG + 1)
3724 /* Matching routines. */
3726 #ifndef emacs /* Emacs never uses this. */
3727 /* re_match is like re_match_2 except it takes only a single string. */
3730 re_match (bufp, string, size, pos, regs)
3731 struct re_pattern_buffer *bufp;
3734 struct re_registers *regs;
3736 int result = re_match_2_internal (bufp, NULL, 0, string, size,
3738 # ifndef REGEX_MALLOC
3746 weak_alias (__re_match, re_match)
3748 #endif /* not emacs */
3750 static boolean group_match_null_string_p _RE_ARGS ((unsigned char **p,
3752 register_info_type *reg_info));
3753 static boolean alt_match_null_string_p _RE_ARGS ((unsigned char *p,
3755 register_info_type *reg_info));
3756 static boolean common_op_match_null_string_p _RE_ARGS ((unsigned char **p,
3758 register_info_type *reg_info));
3759 static int bcmp_translate _RE_ARGS ((const char *s1, const char *s2,
3760 int len, char *translate));
3762 /* re_match_2 matches the compiled pattern in BUFP against the
3763 the (virtual) concatenation of STRING1 and STRING2 (of length SIZE1
3764 and SIZE2, respectively). We start matching at POS, and stop
3767 If REGS is non-null and the `no_sub' field of BUFP is nonzero, we
3768 store offsets for the substring each group matched in REGS. See the
3769 documentation for exactly how many groups we fill.
3771 We return -1 if no match, -2 if an internal error (such as the
3772 failure stack overflowing). Otherwise, we return the length of the
3773 matched substring. */
3776 re_match_2 (bufp, string1, size1, string2, size2, pos, regs, stop)
3777 struct re_pattern_buffer *bufp;
3778 const char *string1, *string2;
3781 struct re_registers *regs;
3784 int result = re_match_2_internal (bufp, string1, size1, string2, size2,
3786 #ifndef REGEX_MALLOC
3794 weak_alias (__re_match_2, re_match_2)
3797 /* This is a separate function so that we can force an alloca cleanup
3800 re_match_2_internal (bufp, string1, size1, string2, size2, pos, regs, stop)
3801 struct re_pattern_buffer *bufp;
3802 const char *string1, *string2;
3805 struct re_registers *regs;
3808 /* General temporaries. */
3812 /* Just past the end of the corresponding string. */
3813 const char *end1, *end2;
3815 /* Pointers into string1 and string2, just past the last characters in
3816 each to consider matching. */
3817 const char *end_match_1, *end_match_2;
3819 /* Where we are in the data, and the end of the current string. */
3820 const char *d, *dend;
3822 /* Where we are in the pattern, and the end of the pattern. */
3823 unsigned char *p = bufp->buffer;
3824 register unsigned char *pend = p + bufp->used;
3826 /* Mark the opcode just after a start_memory, so we can test for an
3827 empty subpattern when we get to the stop_memory. */
3828 unsigned char *just_past_start_mem = 0;
3830 /* We use this to map every character in the string. */
3831 RE_TRANSLATE_TYPE translate = bufp->translate;
3833 /* Failure point stack. Each place that can handle a failure further
3834 down the line pushes a failure point on this stack. It consists of
3835 restart, regend, and reg_info for all registers corresponding to
3836 the subexpressions we're currently inside, plus the number of such
3837 registers, and, finally, two char *'s. The first char * is where
3838 to resume scanning the pattern; the second one is where to resume
3839 scanning the strings. If the latter is zero, the failure point is
3840 a ``dummy''; if a failure happens and the failure point is a dummy,
3841 it gets discarded and the next next one is tried. */
3842 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
3843 fail_stack_type fail_stack;
3846 static unsigned failure_id = 0;
3847 unsigned nfailure_points_pushed = 0, nfailure_points_popped = 0;
3851 /* This holds the pointer to the failure stack, when
3852 it is allocated relocatably. */
3853 fail_stack_elt_t *failure_stack_ptr;
3856 /* We fill all the registers internally, independent of what we
3857 return, for use in backreferences. The number here includes
3858 an element for register zero. */
3859 size_t num_regs = bufp->re_nsub + 1;
3861 /* The currently active registers. */
3862 active_reg_t lowest_active_reg = NO_LOWEST_ACTIVE_REG;
3863 active_reg_t highest_active_reg = NO_HIGHEST_ACTIVE_REG;
3865 /* Information on the contents of registers. These are pointers into
3866 the input strings; they record just what was matched (on this
3867 attempt) by a subexpression part of the pattern, that is, the
3868 regnum-th regstart pointer points to where in the pattern we began
3869 matching and the regnum-th regend points to right after where we
3870 stopped matching the regnum-th subexpression. (The zeroth register
3871 keeps track of what the whole pattern matches.) */
3872 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
3873 const char **regstart, **regend;
3876 /* If a group that's operated upon by a repetition operator fails to
3877 match anything, then the register for its start will need to be
3878 restored because it will have been set to wherever in the string we
3879 are when we last see its open-group operator. Similarly for a
3881 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
3882 const char **old_regstart, **old_regend;
3885 /* The is_active field of reg_info helps us keep track of which (possibly
3886 nested) subexpressions we are currently in. The matched_something
3887 field of reg_info[reg_num] helps us tell whether or not we have
3888 matched any of the pattern so far this time through the reg_num-th
3889 subexpression. These two fields get reset each time through any
3890 loop their register is in. */
3891 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
3892 register_info_type *reg_info;
3895 /* The following record the register info as found in the above
3896 variables when we find a match better than any we've seen before.
3897 This happens as we backtrack through the failure points, which in
3898 turn happens only if we have not yet matched the entire string. */
3899 unsigned best_regs_set = false;
3900 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
3901 const char **best_regstart, **best_regend;
3904 /* Logically, this is `best_regend[0]'. But we don't want to have to
3905 allocate space for that if we're not allocating space for anything
3906 else (see below). Also, we never need info about register 0 for
3907 any of the other register vectors, and it seems rather a kludge to
3908 treat `best_regend' differently than the rest. So we keep track of
3909 the end of the best match so far in a separate variable. We
3910 initialize this to NULL so that when we backtrack the first time
3911 and need to test it, it's not garbage. */
3912 const char *match_end = NULL;
3914 /* This helps SET_REGS_MATCHED avoid doing redundant work. */
3915 int set_regs_matched_done = 0;
3917 /* Used when we pop values we don't care about. */
3918 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
3919 const char **reg_dummy;
3920 register_info_type *reg_info_dummy;
3924 /* Counts the total number of registers pushed. */
3925 unsigned num_regs_pushed = 0;
3928 DEBUG_PRINT1 ("\n\nEntering re_match_2.\n");
3932 #ifdef MATCH_MAY_ALLOCATE
3933 /* Do not bother to initialize all the register variables if there are
3934 no groups in the pattern, as it takes a fair amount of time. If
3935 there are groups, we include space for register 0 (the whole
3936 pattern), even though we never use it, since it simplifies the
3937 array indexing. We should fix this. */
3940 regstart = REGEX_TALLOC (num_regs, const char *);
3941 regend = REGEX_TALLOC (num_regs, const char *);
3942 old_regstart = REGEX_TALLOC (num_regs, const char *);
3943 old_regend = REGEX_TALLOC (num_regs, const char *);
3944 best_regstart = REGEX_TALLOC (num_regs, const char *);
3945 best_regend = REGEX_TALLOC (num_regs, const char *);
3946 reg_info = REGEX_TALLOC (num_regs, register_info_type);
3947 reg_dummy = REGEX_TALLOC (num_regs, const char *);
3948 reg_info_dummy = REGEX_TALLOC (num_regs, register_info_type);
3950 if (!(regstart && regend && old_regstart && old_regend && reg_info
3951 && best_regstart && best_regend && reg_dummy && reg_info_dummy))
3959 /* We must initialize all our variables to NULL, so that
3960 `FREE_VARIABLES' doesn't try to free them. */
3961 regstart = regend = old_regstart = old_regend = best_regstart
3962 = best_regend = reg_dummy = NULL;
3963 reg_info = reg_info_dummy = (register_info_type *) NULL;
3965 #endif /* MATCH_MAY_ALLOCATE */
3967 /* The starting position is bogus. */
3968 if (pos < 0 || pos > size1 + size2)
3974 /* Initialize subexpression text positions to -1 to mark ones that no
3975 start_memory/stop_memory has been seen for. Also initialize the
3976 register information struct. */
3977 for (mcnt = 1; (unsigned) mcnt < num_regs; mcnt++)
3979 regstart[mcnt] = regend[mcnt]
3980 = old_regstart[mcnt] = old_regend[mcnt] = REG_UNSET_VALUE;
3982 REG_MATCH_NULL_STRING_P (reg_info[mcnt]) = MATCH_NULL_UNSET_VALUE;
3983 IS_ACTIVE (reg_info[mcnt]) = 0;
3984 MATCHED_SOMETHING (reg_info[mcnt]) = 0;
3985 EVER_MATCHED_SOMETHING (reg_info[mcnt]) = 0;
3988 /* We move `string1' into `string2' if the latter's empty -- but not if
3989 `string1' is null. */
3990 if (size2 == 0 && string1 != NULL)
3997 end1 = string1 + size1;
3998 end2 = string2 + size2;
4000 /* Compute where to stop matching, within the two strings. */
4003 end_match_1 = string1 + stop;
4004 end_match_2 = string2;
4009 end_match_2 = string2 + stop - size1;
4012 /* `p' scans through the pattern as `d' scans through the data.
4013 `dend' is the end of the input string that `d' points within. `d'
4014 is advanced into the following input string whenever necessary, but
4015 this happens before fetching; therefore, at the beginning of the
4016 loop, `d' can be pointing at the end of a string, but it cannot
4018 if (size1 > 0 && pos <= size1)
4025 d = string2 + pos - size1;
4029 DEBUG_PRINT1 ("The compiled pattern is:\n");
4030 DEBUG_PRINT_COMPILED_PATTERN (bufp, p, pend);
4031 DEBUG_PRINT1 ("The string to match is: `");
4032 DEBUG_PRINT_DOUBLE_STRING (d, string1, size1, string2, size2);
4033 DEBUG_PRINT1 ("'\n");
4035 /* This loops over pattern commands. It exits by returning from the
4036 function if the match is complete, or it drops through if the match
4037 fails at this starting point in the input data. */
4041 DEBUG_PRINT2 ("\n%p: ", p);
4043 DEBUG_PRINT2 ("\n0x%x: ", p);
4047 { /* End of pattern means we might have succeeded. */
4048 DEBUG_PRINT1 ("end of pattern ... ");
4050 /* If we haven't matched the entire string, and we want the
4051 longest match, try backtracking. */
4052 if (d != end_match_2)
4054 /* 1 if this match ends in the same string (string1 or string2)
4055 as the best previous match. */
4056 boolean same_str_p = (FIRST_STRING_P (match_end)
4057 == MATCHING_IN_FIRST_STRING);
4058 /* 1 if this match is the best seen so far. */
4059 boolean best_match_p;
4061 /* AIX compiler got confused when this was combined
4062 with the previous declaration. */
4064 best_match_p = d > match_end;
4066 best_match_p = !MATCHING_IN_FIRST_STRING;
4068 DEBUG_PRINT1 ("backtracking.\n");
4070 if (!FAIL_STACK_EMPTY ())
4071 { /* More failure points to try. */
4073 /* If exceeds best match so far, save it. */
4074 if (!best_regs_set || best_match_p)
4076 best_regs_set = true;
4079 DEBUG_PRINT1 ("\nSAVING match as best so far.\n");
4081 for (mcnt = 1; (unsigned) mcnt < num_regs; mcnt++)
4083 best_regstart[mcnt] = regstart[mcnt];
4084 best_regend[mcnt] = regend[mcnt];
4090 /* If no failure points, don't restore garbage. And if
4091 last match is real best match, don't restore second
4093 else if (best_regs_set && !best_match_p)
4096 /* Restore best match. It may happen that `dend ==
4097 end_match_1' while the restored d is in string2.
4098 For example, the pattern `x.*y.*z' against the
4099 strings `x-' and `y-z-', if the two strings are
4100 not consecutive in memory. */
4101 DEBUG_PRINT1 ("Restoring best registers.\n");
4104 dend = ((d >= string1 && d <= end1)
4105 ? end_match_1 : end_match_2);
4107 for (mcnt = 1; (unsigned) mcnt < num_regs; mcnt++)
4109 regstart[mcnt] = best_regstart[mcnt];
4110 regend[mcnt] = best_regend[mcnt];
4113 } /* d != end_match_2 */
4116 DEBUG_PRINT1 ("Accepting match.\n");
4118 /* If caller wants register contents data back, do it. */
4119 if (regs && !bufp->no_sub)
4121 /* Have the register data arrays been allocated? */
4122 if (bufp->regs_allocated == REGS_UNALLOCATED)
4123 { /* No. So allocate them with malloc. We need one
4124 extra element beyond `num_regs' for the `-1' marker
4126 regs->num_regs = MAX (RE_NREGS, num_regs + 1);
4127 regs->start = TALLOC (regs->num_regs, regoff_t);
4128 regs->end = TALLOC (regs->num_regs, regoff_t);
4129 if (regs->start == NULL || regs->end == NULL)
4134 bufp->regs_allocated = REGS_REALLOCATE;
4136 else if (bufp->regs_allocated == REGS_REALLOCATE)
4137 { /* Yes. If we need more elements than were already
4138 allocated, reallocate them. If we need fewer, just
4140 if (regs->num_regs < num_regs + 1)
4142 regs->num_regs = num_regs + 1;
4143 RETALLOC (regs->start, regs->num_regs, regoff_t);
4144 RETALLOC (regs->end, regs->num_regs, regoff_t);
4145 if (regs->start == NULL || regs->end == NULL)
4154 /* These braces fend off a "empty body in an else-statement"
4155 warning under GCC when assert expands to nothing. */
4156 assert (bufp->regs_allocated == REGS_FIXED);
4159 /* Convert the pointer data in `regstart' and `regend' to
4160 indices. Register zero has to be set differently,
4161 since we haven't kept track of any info for it. */
4162 if (regs->num_regs > 0)
4164 regs->start[0] = pos;
4165 regs->end[0] = (MATCHING_IN_FIRST_STRING
4166 ? ((regoff_t) (d - string1))
4167 : ((regoff_t) (d - string2 + size1)));
4170 /* Go through the first `min (num_regs, regs->num_regs)'
4171 registers, since that is all we initialized. */
4172 for (mcnt = 1; (unsigned) mcnt < MIN (num_regs, regs->num_regs);
4175 if (REG_UNSET (regstart[mcnt]) || REG_UNSET (regend[mcnt]))
4176 regs->start[mcnt] = regs->end[mcnt] = -1;
4180 = (regoff_t) POINTER_TO_OFFSET (regstart[mcnt]);
4182 = (regoff_t) POINTER_TO_OFFSET (regend[mcnt]);
4186 /* If the regs structure we return has more elements than
4187 were in the pattern, set the extra elements to -1. If
4188 we (re)allocated the registers, this is the case,
4189 because we always allocate enough to have at least one
4191 for (mcnt = num_regs; (unsigned) mcnt < regs->num_regs; mcnt++)
4192 regs->start[mcnt] = regs->end[mcnt] = -1;
4193 } /* regs && !bufp->no_sub */
4195 DEBUG_PRINT4 ("%u failure points pushed, %u popped (%u remain).\n",
4196 nfailure_points_pushed, nfailure_points_popped,
4197 nfailure_points_pushed - nfailure_points_popped);
4198 DEBUG_PRINT2 ("%u registers pushed.\n", num_regs_pushed);
4200 mcnt = d - pos - (MATCHING_IN_FIRST_STRING
4204 DEBUG_PRINT2 ("Returning %d from re_match_2.\n", mcnt);
4210 /* Otherwise match next pattern command. */
4211 switch (SWITCH_ENUM_CAST ((re_opcode_t) *p++))
4213 /* Ignore these. Used to ignore the n of succeed_n's which
4214 currently have n == 0. */
4216 DEBUG_PRINT1 ("EXECUTING no_op.\n");
4220 DEBUG_PRINT1 ("EXECUTING succeed.\n");
4223 /* Match the next n pattern characters exactly. The following
4224 byte in the pattern defines n, and the n bytes after that
4225 are the characters to match. */
4228 DEBUG_PRINT2 ("EXECUTING exactn %d.\n", mcnt);
4230 /* This is written out as an if-else so we don't waste time
4231 testing `translate' inside the loop. */
4237 if ((unsigned char) translate[(unsigned char) *d++]
4238 != (unsigned char) *p++)
4248 if (*d++ != (char) *p++) goto fail;
4252 SET_REGS_MATCHED ();
4256 /* Match any character except possibly a newline or a null. */
4258 DEBUG_PRINT1 ("EXECUTING anychar.\n");
4262 if ((!(bufp->syntax & RE_DOT_NEWLINE) && TRANSLATE (*d) == '\n')
4263 || (bufp->syntax & RE_DOT_NOT_NULL && TRANSLATE (*d) == '\000'))
4266 SET_REGS_MATCHED ();
4267 DEBUG_PRINT2 (" Matched `%d'.\n", *d);
4275 register unsigned char c;
4276 boolean not = (re_opcode_t) *(p - 1) == charset_not;
4278 DEBUG_PRINT2 ("EXECUTING charset%s.\n", not ? "_not" : "");
4281 c = TRANSLATE (*d); /* The character to match. */
4283 /* Cast to `unsigned' instead of `unsigned char' in case the
4284 bit list is a full 32 bytes long. */
4285 if (c < (unsigned) (*p * BYTEWIDTH)
4286 && p[1 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH)))
4291 if (!not) goto fail;
4293 SET_REGS_MATCHED ();
4299 /* The beginning of a group is represented by start_memory.
4300 The arguments are the register number in the next byte, and the
4301 number of groups inner to this one in the next. The text
4302 matched within the group is recorded (in the internal
4303 registers data structure) under the register number. */
4305 DEBUG_PRINT3 ("EXECUTING start_memory %d (%d):\n", *p, p[1]);
4307 /* Find out if this group can match the empty string. */
4308 p1 = p; /* To send to group_match_null_string_p. */
4310 if (REG_MATCH_NULL_STRING_P (reg_info[*p]) == MATCH_NULL_UNSET_VALUE)
4311 REG_MATCH_NULL_STRING_P (reg_info[*p])
4312 = group_match_null_string_p (&p1, pend, reg_info);
4314 /* Save the position in the string where we were the last time
4315 we were at this open-group operator in case the group is
4316 operated upon by a repetition operator, e.g., with `(a*)*b'
4317 against `ab'; then we want to ignore where we are now in
4318 the string in case this attempt to match fails. */
4319 old_regstart[*p] = REG_MATCH_NULL_STRING_P (reg_info[*p])
4320 ? REG_UNSET (regstart[*p]) ? d : regstart[*p]
4322 DEBUG_PRINT2 (" old_regstart: %d\n",
4323 POINTER_TO_OFFSET (old_regstart[*p]));
4326 DEBUG_PRINT2 (" regstart: %d\n", POINTER_TO_OFFSET (regstart[*p]));
4328 IS_ACTIVE (reg_info[*p]) = 1;
4329 MATCHED_SOMETHING (reg_info[*p]) = 0;
4331 /* Clear this whenever we change the register activity status. */
4332 set_regs_matched_done = 0;
4334 /* This is the new highest active register. */
4335 highest_active_reg = *p;
4337 /* If nothing was active before, this is the new lowest active
4339 if (lowest_active_reg == NO_LOWEST_ACTIVE_REG)
4340 lowest_active_reg = *p;
4342 /* Move past the register number and inner group count. */
4344 just_past_start_mem = p;
4349 /* The stop_memory opcode represents the end of a group. Its
4350 arguments are the same as start_memory's: the register
4351 number, and the number of inner groups. */
4353 DEBUG_PRINT3 ("EXECUTING stop_memory %d (%d):\n", *p, p[1]);
4355 /* We need to save the string position the last time we were at
4356 this close-group operator in case the group is operated
4357 upon by a repetition operator, e.g., with `((a*)*(b*)*)*'
4358 against `aba'; then we want to ignore where we are now in
4359 the string in case this attempt to match fails. */
4360 old_regend[*p] = REG_MATCH_NULL_STRING_P (reg_info[*p])
4361 ? REG_UNSET (regend[*p]) ? d : regend[*p]
4363 DEBUG_PRINT2 (" old_regend: %d\n",
4364 POINTER_TO_OFFSET (old_regend[*p]));
4367 DEBUG_PRINT2 (" regend: %d\n", POINTER_TO_OFFSET (regend[*p]));
4369 /* This register isn't active anymore. */
4370 IS_ACTIVE (reg_info[*p]) = 0;
4372 /* Clear this whenever we change the register activity status. */
4373 set_regs_matched_done = 0;
4375 /* If this was the only register active, nothing is active
4377 if (lowest_active_reg == highest_active_reg)
4379 lowest_active_reg = NO_LOWEST_ACTIVE_REG;
4380 highest_active_reg = NO_HIGHEST_ACTIVE_REG;
4383 { /* We must scan for the new highest active register, since
4384 it isn't necessarily one less than now: consider
4385 (a(b)c(d(e)f)g). When group 3 ends, after the f), the
4386 new highest active register is 1. */
4387 unsigned char r = *p - 1;
4388 while (r > 0 && !IS_ACTIVE (reg_info[r]))
4391 /* If we end up at register zero, that means that we saved
4392 the registers as the result of an `on_failure_jump', not
4393 a `start_memory', and we jumped to past the innermost
4394 `stop_memory'. For example, in ((.)*) we save
4395 registers 1 and 2 as a result of the *, but when we pop
4396 back to the second ), we are at the stop_memory 1.
4397 Thus, nothing is active. */
4400 lowest_active_reg = NO_LOWEST_ACTIVE_REG;
4401 highest_active_reg = NO_HIGHEST_ACTIVE_REG;
4404 highest_active_reg = r;
4407 /* If just failed to match something this time around with a
4408 group that's operated on by a repetition operator, try to
4409 force exit from the ``loop'', and restore the register
4410 information for this group that we had before trying this
4412 if ((!MATCHED_SOMETHING (reg_info[*p])
4413 || just_past_start_mem == p - 1)
4416 boolean is_a_jump_n = false;
4420 switch ((re_opcode_t) *p1++)
4424 case pop_failure_jump:
4425 case maybe_pop_jump:
4427 case dummy_failure_jump:
4428 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4438 /* If the next operation is a jump backwards in the pattern
4439 to an on_failure_jump right before the start_memory
4440 corresponding to this stop_memory, exit from the loop
4441 by forcing a failure after pushing on the stack the
4442 on_failure_jump's jump in the pattern, and d. */
4443 if (mcnt < 0 && (re_opcode_t) *p1 == on_failure_jump
4444 && (re_opcode_t) p1[3] == start_memory && p1[4] == *p)
4446 /* If this group ever matched anything, then restore
4447 what its registers were before trying this last
4448 failed match, e.g., with `(a*)*b' against `ab' for
4449 regstart[1], and, e.g., with `((a*)*(b*)*)*'
4450 against `aba' for regend[3].
4452 Also restore the registers for inner groups for,
4453 e.g., `((a*)(b*))*' against `aba' (register 3 would
4454 otherwise get trashed). */
4456 if (EVER_MATCHED_SOMETHING (reg_info[*p]))
4460 EVER_MATCHED_SOMETHING (reg_info[*p]) = 0;
4462 /* Restore this and inner groups' (if any) registers. */
4463 for (r = *p; r < (unsigned) *p + (unsigned) *(p + 1);
4466 regstart[r] = old_regstart[r];
4468 /* xx why this test? */
4469 if (old_regend[r] >= regstart[r])
4470 regend[r] = old_regend[r];
4474 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4475 PUSH_FAILURE_POINT (p1 + mcnt, d, -2);
4481 /* Move past the register number and the inner group count. */
4486 /* \<digit> has been turned into a `duplicate' command which is
4487 followed by the numeric value of <digit> as the register number. */
4490 register const char *d2, *dend2;
4491 int regno = *p++; /* Get which register to match against. */
4492 DEBUG_PRINT2 ("EXECUTING duplicate %d.\n", regno);
4494 /* Can't back reference a group which we've never matched. */
4495 if (REG_UNSET (regstart[regno]) || REG_UNSET (regend[regno]))
4498 /* Where in input to try to start matching. */
4499 d2 = regstart[regno];
4501 /* Where to stop matching; if both the place to start and
4502 the place to stop matching are in the same string, then
4503 set to the place to stop, otherwise, for now have to use
4504 the end of the first string. */
4506 dend2 = ((FIRST_STRING_P (regstart[regno])
4507 == FIRST_STRING_P (regend[regno]))
4508 ? regend[regno] : end_match_1);
4511 /* If necessary, advance to next segment in register
4515 if (dend2 == end_match_2) break;
4516 if (dend2 == regend[regno]) break;
4518 /* End of string1 => advance to string2. */
4520 dend2 = regend[regno];
4522 /* At end of register contents => success */
4523 if (d2 == dend2) break;
4525 /* If necessary, advance to next segment in data. */
4528 /* How many characters left in this segment to match. */
4531 /* Want how many consecutive characters we can match in
4532 one shot, so, if necessary, adjust the count. */
4533 if (mcnt > dend2 - d2)
4536 /* Compare that many; failure if mismatch, else move
4539 ? bcmp_translate (d, d2, mcnt, translate)
4540 : memcmp (d, d2, mcnt))
4542 d += mcnt, d2 += mcnt;
4544 /* Do this because we've match some characters. */
4545 SET_REGS_MATCHED ();
4551 /* begline matches the empty string at the beginning of the string
4552 (unless `not_bol' is set in `bufp'), and, if
4553 `newline_anchor' is set, after newlines. */
4555 DEBUG_PRINT1 ("EXECUTING begline.\n");
4557 if (AT_STRINGS_BEG (d))
4559 if (!bufp->not_bol) break;
4561 else if (d[-1] == '\n' && bufp->newline_anchor)
4565 /* In all other cases, we fail. */
4569 /* endline is the dual of begline. */
4571 DEBUG_PRINT1 ("EXECUTING endline.\n");
4573 if (AT_STRINGS_END (d))
4575 if (!bufp->not_eol) break;
4578 /* We have to ``prefetch'' the next character. */
4579 else if ((d == end1 ? *string2 : *d) == '\n'
4580 && bufp->newline_anchor)
4587 /* Match at the very beginning of the data. */
4589 DEBUG_PRINT1 ("EXECUTING begbuf.\n");
4590 if (AT_STRINGS_BEG (d))
4595 /* Match at the very end of the data. */
4597 DEBUG_PRINT1 ("EXECUTING endbuf.\n");
4598 if (AT_STRINGS_END (d))
4603 /* on_failure_keep_string_jump is used to optimize `.*\n'. It
4604 pushes NULL as the value for the string on the stack. Then
4605 `pop_failure_point' will keep the current value for the
4606 string, instead of restoring it. To see why, consider
4607 matching `foo\nbar' against `.*\n'. The .* matches the foo;
4608 then the . fails against the \n. But the next thing we want
4609 to do is match the \n against the \n; if we restored the
4610 string value, we would be back at the foo.
4612 Because this is used only in specific cases, we don't need to
4613 check all the things that `on_failure_jump' does, to make
4614 sure the right things get saved on the stack. Hence we don't
4615 share its code. The only reason to push anything on the
4616 stack at all is that otherwise we would have to change
4617 `anychar's code to do something besides goto fail in this
4618 case; that seems worse than this. */
4619 case on_failure_keep_string_jump:
4620 DEBUG_PRINT1 ("EXECUTING on_failure_keep_string_jump");
4622 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4624 DEBUG_PRINT3 (" %d (to %p):\n", mcnt, p + mcnt);
4626 DEBUG_PRINT3 (" %d (to 0x%x):\n", mcnt, p + mcnt);
4629 PUSH_FAILURE_POINT (p + mcnt, NULL, -2);
4633 /* Uses of on_failure_jump:
4635 Each alternative starts with an on_failure_jump that points
4636 to the beginning of the next alternative. Each alternative
4637 except the last ends with a jump that in effect jumps past
4638 the rest of the alternatives. (They really jump to the
4639 ending jump of the following alternative, because tensioning
4640 these jumps is a hassle.)
4642 Repeats start with an on_failure_jump that points past both
4643 the repetition text and either the following jump or
4644 pop_failure_jump back to this on_failure_jump. */
4645 case on_failure_jump:
4647 DEBUG_PRINT1 ("EXECUTING on_failure_jump");
4649 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4651 DEBUG_PRINT3 (" %d (to %p)", mcnt, p + mcnt);
4653 DEBUG_PRINT3 (" %d (to 0x%x)", mcnt, p + mcnt);
4656 /* If this on_failure_jump comes right before a group (i.e.,
4657 the original * applied to a group), save the information
4658 for that group and all inner ones, so that if we fail back
4659 to this point, the group's information will be correct.
4660 For example, in \(a*\)*\1, we need the preceding group,
4661 and in \(zz\(a*\)b*\)\2, we need the inner group. */
4663 /* We can't use `p' to check ahead because we push
4664 a failure point to `p + mcnt' after we do this. */
4667 /* We need to skip no_op's before we look for the
4668 start_memory in case this on_failure_jump is happening as
4669 the result of a completed succeed_n, as in \(a\)\{1,3\}b\1
4671 while (p1 < pend && (re_opcode_t) *p1 == no_op)
4674 if (p1 < pend && (re_opcode_t) *p1 == start_memory)
4676 /* We have a new highest active register now. This will
4677 get reset at the start_memory we are about to get to,
4678 but we will have saved all the registers relevant to
4679 this repetition op, as described above. */
4680 highest_active_reg = *(p1 + 1) + *(p1 + 2);
4681 if (lowest_active_reg == NO_LOWEST_ACTIVE_REG)
4682 lowest_active_reg = *(p1 + 1);
4685 DEBUG_PRINT1 (":\n");
4686 PUSH_FAILURE_POINT (p + mcnt, d, -2);
4690 /* A smart repeat ends with `maybe_pop_jump'.
4691 We change it to either `pop_failure_jump' or `jump'. */
4692 case maybe_pop_jump:
4693 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4694 DEBUG_PRINT2 ("EXECUTING maybe_pop_jump %d.\n", mcnt);
4696 register unsigned char *p2 = p;
4698 /* Compare the beginning of the repeat with what in the
4699 pattern follows its end. If we can establish that there
4700 is nothing that they would both match, i.e., that we
4701 would have to backtrack because of (as in, e.g., `a*a')
4702 then we can change to pop_failure_jump, because we'll
4703 never have to backtrack.
4705 This is not true in the case of alternatives: in
4706 `(a|ab)*' we do need to backtrack to the `ab' alternative
4707 (e.g., if the string was `ab'). But instead of trying to
4708 detect that here, the alternative has put on a dummy
4709 failure point which is what we will end up popping. */
4711 /* Skip over open/close-group commands.
4712 If what follows this loop is a ...+ construct,
4713 look at what begins its body, since we will have to
4714 match at least one of that. */
4718 && ((re_opcode_t) *p2 == stop_memory
4719 || (re_opcode_t) *p2 == start_memory))
4721 else if (p2 + 6 < pend
4722 && (re_opcode_t) *p2 == dummy_failure_jump)
4729 /* p1[0] ... p1[2] are the `on_failure_jump' corresponding
4730 to the `maybe_finalize_jump' of this case. Examine what
4733 /* If we're at the end of the pattern, we can change. */
4736 /* Consider what happens when matching ":\(.*\)"
4737 against ":/". I don't really understand this code
4739 p[-3] = (unsigned char) pop_failure_jump;
4741 (" End of pattern: change to `pop_failure_jump'.\n");
4744 else if ((re_opcode_t) *p2 == exactn
4745 || (bufp->newline_anchor && (re_opcode_t) *p2 == endline))
4747 register unsigned char c
4748 = *p2 == (unsigned char) endline ? '\n' : p2[2];
4750 if ((re_opcode_t) p1[3] == exactn && p1[5] != c)
4752 p[-3] = (unsigned char) pop_failure_jump;
4753 DEBUG_PRINT3 (" %c != %c => pop_failure_jump.\n",
4757 else if ((re_opcode_t) p1[3] == charset
4758 || (re_opcode_t) p1[3] == charset_not)
4760 int not = (re_opcode_t) p1[3] == charset_not;
4762 if (c < (unsigned char) (p1[4] * BYTEWIDTH)
4763 && p1[5 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH)))
4766 /* `not' is equal to 1 if c would match, which means
4767 that we can't change to pop_failure_jump. */
4770 p[-3] = (unsigned char) pop_failure_jump;
4771 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
4775 else if ((re_opcode_t) *p2 == charset)
4778 register unsigned char c
4779 = *p2 == (unsigned char) endline ? '\n' : p2[2];
4783 if ((re_opcode_t) p1[3] == exactn
4784 && ! ((int) p2[1] * BYTEWIDTH > (int) p1[5]
4785 && (p2[2 + p1[5] / BYTEWIDTH]
4786 & (1 << (p1[5] % BYTEWIDTH)))))
4788 if ((re_opcode_t) p1[3] == exactn
4789 && ! ((int) p2[1] * BYTEWIDTH > (int) p1[4]
4790 && (p2[2 + p1[4] / BYTEWIDTH]
4791 & (1 << (p1[4] % BYTEWIDTH)))))
4794 p[-3] = (unsigned char) pop_failure_jump;
4795 DEBUG_PRINT3 (" %c != %c => pop_failure_jump.\n",
4799 else if ((re_opcode_t) p1[3] == charset_not)
4802 /* We win if the charset_not inside the loop
4803 lists every character listed in the charset after. */
4804 for (idx = 0; idx < (int) p2[1]; idx++)
4805 if (! (p2[2 + idx] == 0
4806 || (idx < (int) p1[4]
4807 && ((p2[2 + idx] & ~ p1[5 + idx]) == 0))))
4812 p[-3] = (unsigned char) pop_failure_jump;
4813 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
4816 else if ((re_opcode_t) p1[3] == charset)
4819 /* We win if the charset inside the loop
4820 has no overlap with the one after the loop. */
4822 idx < (int) p2[1] && idx < (int) p1[4];
4824 if ((p2[2 + idx] & p1[5 + idx]) != 0)
4827 if (idx == p2[1] || idx == p1[4])
4829 p[-3] = (unsigned char) pop_failure_jump;
4830 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
4835 p -= 2; /* Point at relative address again. */
4836 if ((re_opcode_t) p[-1] != pop_failure_jump)
4838 p[-1] = (unsigned char) jump;
4839 DEBUG_PRINT1 (" Match => jump.\n");
4840 goto unconditional_jump;
4842 /* Note fall through. */
4845 /* The end of a simple repeat has a pop_failure_jump back to
4846 its matching on_failure_jump, where the latter will push a
4847 failure point. The pop_failure_jump takes off failure
4848 points put on by this pop_failure_jump's matching
4849 on_failure_jump; we got through the pattern to here from the
4850 matching on_failure_jump, so didn't fail. */
4851 case pop_failure_jump:
4853 /* We need to pass separate storage for the lowest and
4854 highest registers, even though we don't care about the
4855 actual values. Otherwise, we will restore only one
4856 register from the stack, since lowest will == highest in
4857 `pop_failure_point'. */
4858 active_reg_t dummy_low_reg, dummy_high_reg;
4859 unsigned char *pdummy;
4862 DEBUG_PRINT1 ("EXECUTING pop_failure_jump.\n");
4863 POP_FAILURE_POINT (sdummy, pdummy,
4864 dummy_low_reg, dummy_high_reg,
4865 reg_dummy, reg_dummy, reg_info_dummy);
4867 /* Note fall through. */
4871 DEBUG_PRINT2 ("\n%p: ", p);
4873 DEBUG_PRINT2 ("\n0x%x: ", p);
4875 /* Note fall through. */
4877 /* Unconditionally jump (without popping any failure points). */
4879 EXTRACT_NUMBER_AND_INCR (mcnt, p); /* Get the amount to jump. */
4880 DEBUG_PRINT2 ("EXECUTING jump %d ", mcnt);
4881 p += mcnt; /* Do the jump. */
4883 DEBUG_PRINT2 ("(to %p).\n", p);
4885 DEBUG_PRINT2 ("(to 0x%x).\n", p);
4890 /* We need this opcode so we can detect where alternatives end
4891 in `group_match_null_string_p' et al. */
4893 DEBUG_PRINT1 ("EXECUTING jump_past_alt.\n");
4894 goto unconditional_jump;
4897 /* Normally, the on_failure_jump pushes a failure point, which
4898 then gets popped at pop_failure_jump. We will end up at
4899 pop_failure_jump, also, and with a pattern of, say, `a+', we
4900 are skipping over the on_failure_jump, so we have to push
4901 something meaningless for pop_failure_jump to pop. */
4902 case dummy_failure_jump:
4903 DEBUG_PRINT1 ("EXECUTING dummy_failure_jump.\n");
4904 /* It doesn't matter what we push for the string here. What
4905 the code at `fail' tests is the value for the pattern. */
4906 PUSH_FAILURE_POINT (NULL, NULL, -2);
4907 goto unconditional_jump;
4910 /* At the end of an alternative, we need to push a dummy failure
4911 point in case we are followed by a `pop_failure_jump', because
4912 we don't want the failure point for the alternative to be
4913 popped. For example, matching `(a|ab)*' against `aab'
4914 requires that we match the `ab' alternative. */
4915 case push_dummy_failure:
4916 DEBUG_PRINT1 ("EXECUTING push_dummy_failure.\n");
4917 /* See comments just above at `dummy_failure_jump' about the
4919 PUSH_FAILURE_POINT (NULL, NULL, -2);
4922 /* Have to succeed matching what follows at least n times.
4923 After that, handle like `on_failure_jump'. */
4925 EXTRACT_NUMBER (mcnt, p + 2);
4926 DEBUG_PRINT2 ("EXECUTING succeed_n %d.\n", mcnt);
4929 /* Originally, this is how many times we HAVE to succeed. */
4934 STORE_NUMBER_AND_INCR (p, mcnt);
4936 DEBUG_PRINT3 (" Setting %p to %d.\n", p - 2, mcnt);
4938 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p - 2, mcnt);
4944 DEBUG_PRINT2 (" Setting two bytes from %p to no_op.\n", p+2);
4946 DEBUG_PRINT2 (" Setting two bytes from 0x%x to no_op.\n", p+2);
4948 p[2] = (unsigned char) no_op;
4949 p[3] = (unsigned char) no_op;
4955 EXTRACT_NUMBER (mcnt, p + 2);
4956 DEBUG_PRINT2 ("EXECUTING jump_n %d.\n", mcnt);
4958 /* Originally, this is how many times we CAN jump. */
4962 STORE_NUMBER (p + 2, mcnt);
4964 DEBUG_PRINT3 (" Setting %p to %d.\n", p + 2, mcnt);
4966 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p + 2, mcnt);
4968 goto unconditional_jump;
4970 /* If don't have to jump any more, skip over the rest of command. */
4977 DEBUG_PRINT1 ("EXECUTING set_number_at.\n");
4979 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4981 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4983 DEBUG_PRINT3 (" Setting %p to %d.\n", p1, mcnt);
4985 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p1, mcnt);
4987 STORE_NUMBER (p1, mcnt);
4992 /* The DEC Alpha C compiler 3.x generates incorrect code for the
4993 test WORDCHAR_P (d - 1) != WORDCHAR_P (d) in the expansion of
4994 AT_WORD_BOUNDARY, so this code is disabled. Expanding the
4995 macro and introducing temporary variables works around the bug. */
4998 DEBUG_PRINT1 ("EXECUTING wordbound.\n");
4999 if (AT_WORD_BOUNDARY (d))
5004 DEBUG_PRINT1 ("EXECUTING notwordbound.\n");
5005 if (AT_WORD_BOUNDARY (d))
5011 boolean prevchar, thischar;
5013 DEBUG_PRINT1 ("EXECUTING wordbound.\n");
5014 if (AT_STRINGS_BEG (d) || AT_STRINGS_END (d))
5017 prevchar = WORDCHAR_P (d - 1);
5018 thischar = WORDCHAR_P (d);
5019 if (prevchar != thischar)
5026 boolean prevchar, thischar;
5028 DEBUG_PRINT1 ("EXECUTING notwordbound.\n");
5029 if (AT_STRINGS_BEG (d) || AT_STRINGS_END (d))
5032 prevchar = WORDCHAR_P (d - 1);
5033 thischar = WORDCHAR_P (d);
5034 if (prevchar != thischar)
5041 DEBUG_PRINT1 ("EXECUTING wordbeg.\n");
5042 if (WORDCHAR_P (d) && (AT_STRINGS_BEG (d) || !WORDCHAR_P (d - 1)))
5047 DEBUG_PRINT1 ("EXECUTING wordend.\n");
5048 if (!AT_STRINGS_BEG (d) && WORDCHAR_P (d - 1)
5049 && (!WORDCHAR_P (d) || AT_STRINGS_END (d)))
5055 DEBUG_PRINT1 ("EXECUTING before_dot.\n");
5056 if (PTR_CHAR_POS ((unsigned char *) d) >= point)
5061 DEBUG_PRINT1 ("EXECUTING at_dot.\n");
5062 if (PTR_CHAR_POS ((unsigned char *) d) != point)
5067 DEBUG_PRINT1 ("EXECUTING after_dot.\n");
5068 if (PTR_CHAR_POS ((unsigned char *) d) <= point)
5073 DEBUG_PRINT2 ("EXECUTING syntaxspec %d.\n", mcnt);
5078 DEBUG_PRINT1 ("EXECUTING Emacs wordchar.\n");
5082 /* Can't use *d++ here; SYNTAX may be an unsafe macro. */
5084 if (SYNTAX (d[-1]) != (enum syntaxcode) mcnt)
5086 SET_REGS_MATCHED ();
5090 DEBUG_PRINT2 ("EXECUTING notsyntaxspec %d.\n", mcnt);
5092 goto matchnotsyntax;
5095 DEBUG_PRINT1 ("EXECUTING Emacs notwordchar.\n");
5099 /* Can't use *d++ here; SYNTAX may be an unsafe macro. */
5101 if (SYNTAX (d[-1]) == (enum syntaxcode) mcnt)
5103 SET_REGS_MATCHED ();
5106 #else /* not emacs */
5108 DEBUG_PRINT1 ("EXECUTING non-Emacs wordchar.\n");
5110 if (!WORDCHAR_P (d))
5112 SET_REGS_MATCHED ();
5117 DEBUG_PRINT1 ("EXECUTING non-Emacs notwordchar.\n");
5121 SET_REGS_MATCHED ();
5124 #endif /* not emacs */
5129 continue; /* Successfully executed one pattern command; keep going. */
5132 /* We goto here if a matching operation fails. */
5134 if (!FAIL_STACK_EMPTY ())
5135 { /* A restart point is known. Restore to that state. */
5136 DEBUG_PRINT1 ("\nFAIL:\n");
5137 POP_FAILURE_POINT (d, p,
5138 lowest_active_reg, highest_active_reg,
5139 regstart, regend, reg_info);
5141 /* If this failure point is a dummy, try the next one. */
5145 /* If we failed to the end of the pattern, don't examine *p. */
5149 boolean is_a_jump_n = false;
5151 /* If failed to a backwards jump that's part of a repetition
5152 loop, need to pop this failure point and use the next one. */
5153 switch ((re_opcode_t) *p)
5157 case maybe_pop_jump:
5158 case pop_failure_jump:
5161 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
5164 if ((is_a_jump_n && (re_opcode_t) *p1 == succeed_n)
5166 && (re_opcode_t) *p1 == on_failure_jump))
5174 if (d >= string1 && d <= end1)
5178 break; /* Matching at this starting point really fails. */
5182 goto restore_best_regs;
5186 return -1; /* Failure to match. */
5189 /* Subroutine definitions for re_match_2. */
5192 /* We are passed P pointing to a register number after a start_memory.
5194 Return true if the pattern up to the corresponding stop_memory can
5195 match the empty string, and false otherwise.
5197 If we find the matching stop_memory, sets P to point to one past its number.
5198 Otherwise, sets P to an undefined byte less than or equal to END.
5200 We don't handle duplicates properly (yet). */
5203 group_match_null_string_p (p, end, reg_info)
5204 unsigned char **p, *end;
5205 register_info_type *reg_info;
5208 /* Point to after the args to the start_memory. */
5209 unsigned char *p1 = *p + 2;
5213 /* Skip over opcodes that can match nothing, and return true or
5214 false, as appropriate, when we get to one that can't, or to the
5215 matching stop_memory. */
5217 switch ((re_opcode_t) *p1)
5219 /* Could be either a loop or a series of alternatives. */
5220 case on_failure_jump:
5222 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
5224 /* If the next operation is not a jump backwards in the
5229 /* Go through the on_failure_jumps of the alternatives,
5230 seeing if any of the alternatives cannot match nothing.
5231 The last alternative starts with only a jump,
5232 whereas the rest start with on_failure_jump and end
5233 with a jump, e.g., here is the pattern for `a|b|c':
5235 /on_failure_jump/0/6/exactn/1/a/jump_past_alt/0/6
5236 /on_failure_jump/0/6/exactn/1/b/jump_past_alt/0/3
5239 So, we have to first go through the first (n-1)
5240 alternatives and then deal with the last one separately. */
5243 /* Deal with the first (n-1) alternatives, which start
5244 with an on_failure_jump (see above) that jumps to right
5245 past a jump_past_alt. */
5247 while ((re_opcode_t) p1[mcnt-3] == jump_past_alt)
5249 /* `mcnt' holds how many bytes long the alternative
5250 is, including the ending `jump_past_alt' and
5253 if (!alt_match_null_string_p (p1, p1 + mcnt - 3,
5257 /* Move to right after this alternative, including the
5261 /* Break if it's the beginning of an n-th alternative
5262 that doesn't begin with an on_failure_jump. */
5263 if ((re_opcode_t) *p1 != on_failure_jump)
5266 /* Still have to check that it's not an n-th
5267 alternative that starts with an on_failure_jump. */
5269 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
5270 if ((re_opcode_t) p1[mcnt-3] != jump_past_alt)
5272 /* Get to the beginning of the n-th alternative. */
5278 /* Deal with the last alternative: go back and get number
5279 of the `jump_past_alt' just before it. `mcnt' contains
5280 the length of the alternative. */
5281 EXTRACT_NUMBER (mcnt, p1 - 2);
5283 if (!alt_match_null_string_p (p1, p1 + mcnt, reg_info))
5286 p1 += mcnt; /* Get past the n-th alternative. */
5292 assert (p1[1] == **p);
5298 if (!common_op_match_null_string_p (&p1, end, reg_info))
5301 } /* while p1 < end */
5304 } /* group_match_null_string_p */
5307 /* Similar to group_match_null_string_p, but doesn't deal with alternatives:
5308 It expects P to be the first byte of a single alternative and END one
5309 byte past the last. The alternative can contain groups. */
5312 alt_match_null_string_p (p, end, reg_info)
5313 unsigned char *p, *end;
5314 register_info_type *reg_info;
5317 unsigned char *p1 = p;
5321 /* Skip over opcodes that can match nothing, and break when we get
5322 to one that can't. */
5324 switch ((re_opcode_t) *p1)
5327 case on_failure_jump:
5329 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
5334 if (!common_op_match_null_string_p (&p1, end, reg_info))
5337 } /* while p1 < end */
5340 } /* alt_match_null_string_p */
5343 /* Deals with the ops common to group_match_null_string_p and
5344 alt_match_null_string_p.
5346 Sets P to one after the op and its arguments, if any. */
5349 common_op_match_null_string_p (p, end, reg_info)
5350 unsigned char **p, *end;
5351 register_info_type *reg_info;
5356 unsigned char *p1 = *p;
5358 switch ((re_opcode_t) *p1++)
5378 assert (reg_no > 0 && reg_no <= MAX_REGNUM);
5379 ret = group_match_null_string_p (&p1, end, reg_info);
5381 /* Have to set this here in case we're checking a group which
5382 contains a group and a back reference to it. */
5384 if (REG_MATCH_NULL_STRING_P (reg_info[reg_no]) == MATCH_NULL_UNSET_VALUE)
5385 REG_MATCH_NULL_STRING_P (reg_info[reg_no]) = ret;
5391 /* If this is an optimized succeed_n for zero times, make the jump. */
5393 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
5401 /* Get to the number of times to succeed. */
5403 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
5408 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
5416 if (!REG_MATCH_NULL_STRING_P (reg_info[*p1]))
5424 /* All other opcodes mean we cannot match the empty string. */
5430 } /* common_op_match_null_string_p */
5433 /* Return zero if TRANSLATE[S1] and TRANSLATE[S2] are identical for LEN
5434 bytes; nonzero otherwise. */
5437 bcmp_translate (s1, s2, len, translate)
5438 const char *s1, *s2;
5440 RE_TRANSLATE_TYPE translate;
5442 register const unsigned char *p1 = (const unsigned char *) s1;
5443 register const unsigned char *p2 = (const unsigned char *) s2;
5446 if (translate[*p1++] != translate[*p2++]) return 1;
5452 /* Entry points for GNU code. */
5454 /* re_compile_pattern is the GNU regular expression compiler: it
5455 compiles PATTERN (of length SIZE) and puts the result in BUFP.
5456 Returns 0 if the pattern was valid, otherwise an error string.
5458 Assumes the `allocated' (and perhaps `buffer') and `translate' fields
5459 are set in BUFP on entry.
5461 We call regex_compile to do the actual compilation. */
5464 re_compile_pattern (pattern, length, bufp)
5465 const char *pattern;
5467 struct re_pattern_buffer *bufp;
5471 /* GNU code is written to assume at least RE_NREGS registers will be set
5472 (and at least one extra will be -1). */
5473 bufp->regs_allocated = REGS_UNALLOCATED;
5475 /* And GNU code determines whether or not to get register information
5476 by passing null for the REGS argument to re_match, etc., not by
5480 /* Match anchors at newline. */
5481 bufp->newline_anchor = 1;
5483 ret = regex_compile (pattern, length, re_syntax_options, bufp);
5487 return gettext (re_error_msgid[(int) ret]);
5490 weak_alias (__re_compile_pattern, re_compile_pattern)
5493 /* Entry points compatible with 4.2 BSD regex library. We don't define
5494 them unless specifically requested. */
5496 #if defined _REGEX_RE_COMP || defined _LIBC
5498 /* BSD has one and only one pattern buffer. */
5499 static struct re_pattern_buffer re_comp_buf;
5503 /* Make these definitions weak in libc, so POSIX programs can redefine
5504 these names if they don't use our functions, and still use
5505 regcomp/regexec below without link errors. */
5515 if (!re_comp_buf.buffer)
5516 return gettext ("No previous regular expression");
5520 if (!re_comp_buf.buffer)
5522 re_comp_buf.buffer = (unsigned char *) malloc (200);
5523 if (re_comp_buf.buffer == NULL)
5524 return (char *) gettext (re_error_msgid[(int) REG_ESPACE]);
5525 re_comp_buf.allocated = 200;
5527 re_comp_buf.fastmap = (char *) malloc (1 << BYTEWIDTH);
5528 if (re_comp_buf.fastmap == NULL)
5529 return (char *) gettext (re_error_msgid[(int) REG_ESPACE]);
5532 /* Since `re_exec' always passes NULL for the `regs' argument, we
5533 don't need to initialize the pattern buffer fields which affect it. */
5535 /* Match anchors at newlines. */
5536 re_comp_buf.newline_anchor = 1;
5538 ret = regex_compile (s, strlen (s), re_syntax_options, &re_comp_buf);
5543 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
5544 return (char *) gettext (re_error_msgid[(int) ret]);
5555 const int len = strlen (s);
5557 0 <= re_search (&re_comp_buf, s, len, 0, len, (struct re_registers *) 0);
5560 #endif /* _REGEX_RE_COMP */
5562 /* POSIX.2 functions. Don't define these for Emacs. */
5566 /* regcomp takes a regular expression as a string and compiles it.
5568 PREG is a regex_t *. We do not expect any fields to be initialized,
5569 since POSIX says we shouldn't. Thus, we set
5571 `buffer' to the compiled pattern;
5572 `used' to the length of the compiled pattern;
5573 `syntax' to RE_SYNTAX_POSIX_EXTENDED if the
5574 REG_EXTENDED bit in CFLAGS is set; otherwise, to
5575 RE_SYNTAX_POSIX_BASIC;
5576 `newline_anchor' to REG_NEWLINE being set in CFLAGS;
5577 `fastmap' and `fastmap_accurate' to zero;
5578 `re_nsub' to the number of subexpressions in PATTERN.
5580 PATTERN is the address of the pattern string.
5582 CFLAGS is a series of bits which affect compilation.
5584 If REG_EXTENDED is set, we use POSIX extended syntax; otherwise, we
5585 use POSIX basic syntax.
5587 If REG_NEWLINE is set, then . and [^...] don't match newline.
5588 Also, regexec will try a match beginning after every newline.
5590 If REG_ICASE is set, then we considers upper- and lowercase
5591 versions of letters to be equivalent when matching.
5593 If REG_NOSUB is set, then when PREG is passed to regexec, that
5594 routine will report only success or failure, and nothing about the
5597 It returns 0 if it succeeds, nonzero if it doesn't. (See regex.h for
5598 the return codes and their meanings.) */
5601 regcomp (preg, pattern, cflags)
5603 const char *pattern;
5608 = (cflags & REG_EXTENDED) ?
5609 RE_SYNTAX_POSIX_EXTENDED : RE_SYNTAX_POSIX_BASIC;
5611 /* regex_compile will allocate the space for the compiled pattern. */
5613 preg->allocated = 0;
5616 /* Don't bother to use a fastmap when searching. This simplifies the
5617 REG_NEWLINE case: if we used a fastmap, we'd have to put all the
5618 characters after newlines into the fastmap. This way, we just try
5622 if (cflags & REG_ICASE)
5627 = (RE_TRANSLATE_TYPE) malloc (CHAR_SET_SIZE
5628 * sizeof (*(RE_TRANSLATE_TYPE)0));
5629 if (preg->translate == NULL)
5630 return (int) REG_ESPACE;
5632 /* Map uppercase characters to corresponding lowercase ones. */
5633 for (i = 0; i < CHAR_SET_SIZE; i++)
5634 preg->translate[i] = ISUPPER (i) ? tolower (i) : i;
5637 preg->translate = NULL;
5639 /* If REG_NEWLINE is set, newlines are treated differently. */
5640 if (cflags & REG_NEWLINE)
5641 { /* REG_NEWLINE implies neither . nor [^...] match newline. */
5642 syntax &= ~RE_DOT_NEWLINE;
5643 syntax |= RE_HAT_LISTS_NOT_NEWLINE;
5644 /* It also changes the matching behavior. */
5645 preg->newline_anchor = 1;
5648 preg->newline_anchor = 0;
5650 preg->no_sub = !!(cflags & REG_NOSUB);
5652 /* POSIX says a null character in the pattern terminates it, so we
5653 can use strlen here in compiling the pattern. */
5654 ret = regex_compile (pattern, strlen (pattern), syntax, preg);
5656 /* POSIX doesn't distinguish between an unmatched open-group and an
5657 unmatched close-group: both are REG_EPAREN. */
5658 if (ret == REG_ERPAREN) ret = REG_EPAREN;
5663 weak_alias (__regcomp, regcomp)
5667 /* regexec searches for a given pattern, specified by PREG, in the
5670 If NMATCH is zero or REG_NOSUB was set in the cflags argument to
5671 `regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at
5672 least NMATCH elements, and we set them to the offsets of the
5673 corresponding matched substrings.
5675 EFLAGS specifies `execution flags' which affect matching: if
5676 REG_NOTBOL is set, then ^ does not match at the beginning of the
5677 string; if REG_NOTEOL is set, then $ does not match at the end.
5679 We return 0 if we find a match and REG_NOMATCH if not. */
5682 regexec (preg, string, nmatch, pmatch, eflags)
5683 const regex_t *preg;
5686 regmatch_t pmatch[];
5690 struct re_registers regs;
5691 regex_t private_preg;
5692 int len = strlen (string);
5693 boolean want_reg_info = !preg->no_sub && nmatch > 0;
5695 private_preg = *preg;
5697 private_preg.not_bol = !!(eflags & REG_NOTBOL);
5698 private_preg.not_eol = !!(eflags & REG_NOTEOL);
5700 /* The user has told us exactly how many registers to return
5701 information about, via `nmatch'. We have to pass that on to the
5702 matching routines. */
5703 private_preg.regs_allocated = REGS_FIXED;
5707 regs.num_regs = nmatch;
5708 regs.start = TALLOC (nmatch, regoff_t);
5709 regs.end = TALLOC (nmatch, regoff_t);
5710 if (regs.start == NULL || regs.end == NULL)
5711 return (int) REG_NOMATCH;
5714 /* Perform the searching operation. */
5715 ret = re_search (&private_preg, string, len,
5716 /* start: */ 0, /* range: */ len,
5717 want_reg_info ? ®s : (struct re_registers *) 0);
5719 /* Copy the register information to the POSIX structure. */
5726 for (r = 0; r < nmatch; r++)
5728 pmatch[r].rm_so = regs.start[r];
5729 pmatch[r].rm_eo = regs.end[r];
5733 /* If we needed the temporary register info, free the space now. */
5738 /* We want zero return to mean success, unlike `re_search'. */
5739 return ret >= 0 ? (int) REG_NOERROR : (int) REG_NOMATCH;
5742 weak_alias (__regexec, regexec)
5746 /* Returns a message corresponding to an error code, ERRCODE, returned
5747 from either regcomp or regexec. We don't use PREG here. */
5750 regerror (errcode, preg, errbuf, errbuf_size)
5752 const regex_t *preg;
5760 || errcode >= (int) (sizeof (re_error_msgid)
5761 / sizeof (re_error_msgid[0])))
5762 /* Only error codes returned by the rest of the code should be passed
5763 to this routine. If we are given anything else, or if other regex
5764 code generates an invalid error code, then the program has a bug.
5765 Dump core so we can fix it. */
5768 msg = gettext (re_error_msgid[errcode]);
5770 msg_size = strlen (msg) + 1; /* Includes the null. */
5772 if (errbuf_size != 0)
5774 if (msg_size > errbuf_size)
5776 #if defined HAVE_MEMPCPY || defined _LIBC
5777 *((char *) __mempcpy (errbuf, msg, errbuf_size - 1)) = '\0';
5779 memcpy (errbuf, msg, errbuf_size - 1);
5780 errbuf[errbuf_size - 1] = 0;
5784 memcpy (errbuf, msg, msg_size);
5790 weak_alias (__regerror, regerror)
5794 /* Free dynamically allocated space used by PREG. */
5800 if (preg->buffer != NULL)
5801 free (preg->buffer);
5802 preg->buffer = NULL;
5804 preg->allocated = 0;
5807 if (preg->fastmap != NULL)
5808 free (preg->fastmap);
5809 preg->fastmap = NULL;
5810 preg->fastmap_accurate = 0;
5812 if (preg->translate != NULL)
5813 free (preg->translate);
5814 preg->translate = NULL;
5817 weak_alias (__regfree, regfree)
5820 #endif /* not emacs */