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) 1993-1999, 2000, 2001 Free Software Foundation, Inc.
6 This file is part of the GNU C Library.
8 The GNU C Library is free software; you can redistribute it and/or
9 modify it under the terms of the GNU Lesser General Public
10 License as published by the Free Software Foundation; either
11 version 2.1 of the License, or (at your option) any later version.
13 The GNU C Library is distributed in the hope that it will be useful,
14 but WITHOUT ANY WARRANTY; without even the implied warranty of
15 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
16 Lesser General Public License for more details.
18 You should have received a copy of the GNU Lesser General Public
19 License along with the GNU C Library; if not, write to the Free
20 Software Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA
23 /* AIX requires this to be the first thing in the file. */
24 #if defined _AIX && !defined REGEX_MALLOC
36 # if defined __GNUC__ || (defined __STDC__ && __STDC__)
37 # define PARAMS(args) args
39 # define PARAMS(args) ()
41 #endif /* Not PARAMS. */
43 #ifndef INSIDE_RECURSION
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 (HAVE_WCTYPE_H && HAVE_WCHAR_H && HAVE_BTOWC)
54 /* For platform which support the ISO C amendement 1 functionality we
55 support user defined character classes. */
56 # if defined _LIBC || WIDE_CHAR_SUPPORT
57 /* Solaris 2.5 has a bug: <wchar.h> must be included before <wctype.h>. */
63 /* We have to keep the namespace clean. */
64 # define regfree(preg) __regfree (preg)
65 # define regexec(pr, st, nm, pm, ef) __regexec (pr, st, nm, pm, ef)
66 # define regcomp(preg, pattern, cflags) __regcomp (preg, pattern, cflags)
67 # define regerror(errcode, preg, errbuf, errbuf_size) \
68 __regerror(errcode, preg, errbuf, errbuf_size)
69 # define re_set_registers(bu, re, nu, st, en) \
70 __re_set_registers (bu, re, nu, st, en)
71 # define re_match_2(bufp, string1, size1, string2, size2, pos, regs, stop) \
72 __re_match_2 (bufp, string1, size1, string2, size2, pos, regs, stop)
73 # define re_match(bufp, string, size, pos, regs) \
74 __re_match (bufp, string, size, pos, regs)
75 # define re_search(bufp, string, size, startpos, range, regs) \
76 __re_search (bufp, string, size, startpos, range, regs)
77 # define re_compile_pattern(pattern, length, bufp) \
78 __re_compile_pattern (pattern, length, bufp)
79 # define re_set_syntax(syntax) __re_set_syntax (syntax)
80 # define re_search_2(bufp, st1, s1, st2, s2, startpos, range, regs, stop) \
81 __re_search_2 (bufp, st1, s1, st2, s2, startpos, range, regs, stop)
82 # define re_compile_fastmap(bufp) __re_compile_fastmap (bufp)
84 # define btowc __btowc
86 /* We are also using some library internals. */
87 # include <locale/localeinfo.h>
88 # include <locale/elem-hash.h>
89 # include <langinfo.h>
90 # include <locale/coll-lookup.h>
93 /* This is for other GNU distributions with internationalized messages. */
94 # if HAVE_LIBINTL_H || defined _LIBC
98 # define gettext(msgid) __dcgettext ("libc", msgid, LC_MESSAGES)
101 # define gettext(msgid) (msgid)
104 # ifndef gettext_noop
105 /* This define is so xgettext can find the internationalizable
107 # define gettext_noop(String) String
110 /* The `emacs' switch turns on certain matching commands
111 that make sense only in Emacs. */
118 # else /* not emacs */
120 /* If we are not linking with Emacs proper,
121 we can't use the relocating allocator
122 even if config.h says that we can. */
125 # if defined STDC_HEADERS || defined _LIBC
132 /* When used in Emacs's lib-src, we need to get bzero and bcopy somehow.
133 If nothing else has been done, use the method below. */
134 # ifdef INHIBIT_STRING_HEADER
135 # if !(defined HAVE_BZERO && defined HAVE_BCOPY)
136 # if !defined bzero && !defined bcopy
137 # undef INHIBIT_STRING_HEADER
142 /* This is the normal way of making sure we have a bcopy and a bzero.
143 This is used in most programs--a few other programs avoid this
144 by defining INHIBIT_STRING_HEADER. */
145 # ifndef INHIBIT_STRING_HEADER
146 # if defined HAVE_STRING_H || defined STDC_HEADERS || defined _LIBC
150 # define bzero(s, n) (memset (s, '\0', n), (s))
152 # define bzero(s, n) __bzero (s, n)
156 # include <strings.h>
158 # define memcmp(s1, s2, n) bcmp (s1, s2, n)
161 # define memcpy(d, s, n) (bcopy (s, d, n), (d))
166 /* Define the syntax stuff for \<, \>, etc. */
168 /* This must be nonzero for the wordchar and notwordchar pattern
169 commands in re_match_2. */
174 # ifdef SWITCH_ENUM_BUG
175 # define SWITCH_ENUM_CAST(x) ((int)(x))
177 # define SWITCH_ENUM_CAST(x) (x)
180 # endif /* not emacs */
182 # if defined _LIBC || HAVE_LIMITS_H
187 # define MB_LEN_MAX 1
190 /* Get the interface, including the syntax bits. */
193 /* isalpha etc. are used for the character classes. */
196 /* Jim Meyering writes:
198 "... Some ctype macros are valid only for character codes that
199 isascii says are ASCII (SGI's IRIX-4.0.5 is one such system --when
200 using /bin/cc or gcc but without giving an ansi option). So, all
201 ctype uses should be through macros like ISPRINT... If
202 STDC_HEADERS is defined, then autoconf has verified that the ctype
203 macros don't need to be guarded with references to isascii. ...
204 Defining isascii to 1 should let any compiler worth its salt
205 eliminate the && through constant folding."
206 Solaris defines some of these symbols so we must undefine them first. */
208 # if defined STDC_HEADERS || (!defined isascii && !defined HAVE_ISASCII)
209 # define IN_CTYPE_DOMAIN(c) 1
211 # define IN_CTYPE_DOMAIN(c) isascii(c)
215 # define ISBLANK(c) (IN_CTYPE_DOMAIN (c) && isblank (c))
217 # define ISBLANK(c) ((c) == ' ' || (c) == '\t')
220 # define ISGRAPH(c) (IN_CTYPE_DOMAIN (c) && isgraph (c))
222 # define ISGRAPH(c) (IN_CTYPE_DOMAIN (c) && isprint (c) && !isspace (c))
226 # define ISPRINT(c) (IN_CTYPE_DOMAIN (c) && isprint (c))
227 # define ISDIGIT(c) (IN_CTYPE_DOMAIN (c) && isdigit (c))
228 # define ISALNUM(c) (IN_CTYPE_DOMAIN (c) && isalnum (c))
229 # define ISALPHA(c) (IN_CTYPE_DOMAIN (c) && isalpha (c))
230 # define ISCNTRL(c) (IN_CTYPE_DOMAIN (c) && iscntrl (c))
231 # define ISLOWER(c) (IN_CTYPE_DOMAIN (c) && islower (c))
232 # define ISPUNCT(c) (IN_CTYPE_DOMAIN (c) && ispunct (c))
233 # define ISSPACE(c) (IN_CTYPE_DOMAIN (c) && isspace (c))
234 # define ISUPPER(c) (IN_CTYPE_DOMAIN (c) && isupper (c))
235 # define ISXDIGIT(c) (IN_CTYPE_DOMAIN (c) && isxdigit (c))
238 # define TOLOWER(c) _tolower(c)
240 # define TOLOWER(c) tolower(c)
244 # define NULL (void *)0
247 /* We remove any previous definition of `SIGN_EXTEND_CHAR',
248 since ours (we hope) works properly with all combinations of
249 machines, compilers, `char' and `unsigned char' argument types.
250 (Per Bothner suggested the basic approach.) */
251 # undef SIGN_EXTEND_CHAR
253 # define SIGN_EXTEND_CHAR(c) ((signed char) (c))
254 # else /* not __STDC__ */
255 /* As in Harbison and Steele. */
256 # define SIGN_EXTEND_CHAR(c) ((((unsigned char) (c)) ^ 128) - 128)
260 /* How many characters in the character set. */
261 # define CHAR_SET_SIZE 256
265 extern char *re_syntax_table;
267 # else /* not SYNTAX_TABLE */
269 static char re_syntax_table[CHAR_SET_SIZE];
271 static void init_syntax_once PARAMS ((void));
281 bzero (re_syntax_table, sizeof re_syntax_table);
283 for (c = 0; c < CHAR_SET_SIZE; ++c)
285 re_syntax_table[c] = Sword;
287 re_syntax_table['_'] = Sword;
292 # endif /* not SYNTAX_TABLE */
294 # define SYNTAX(c) re_syntax_table[(unsigned char) (c)]
298 /* Integer type for pointers. */
300 #define uintptr_t private_uintptr_t
301 typedef unsigned long int uintptr_t;
304 /* Should we use malloc or alloca? If REGEX_MALLOC is not defined, we
305 use `alloca' instead of `malloc'. This is because using malloc in
306 re_search* or re_match* could cause memory leaks when C-g is used in
307 Emacs; also, malloc is slower and causes storage fragmentation. On
308 the other hand, malloc is more portable, and easier to debug.
310 Because we sometimes use alloca, some routines have to be macros,
311 not functions -- `alloca'-allocated space disappears at the end of the
312 function it is called in. */
316 # define REGEX_ALLOCATE malloc
317 # define REGEX_REALLOCATE(source, osize, nsize) realloc (source, nsize)
318 # define REGEX_FREE free
320 # else /* not REGEX_MALLOC */
322 /* Emacs already defines alloca, sometimes. */
325 /* Make alloca work the best possible way. */
327 # define alloca __builtin_alloca
328 # else /* not __GNUC__ */
331 # endif /* HAVE_ALLOCA_H */
332 # endif /* not __GNUC__ */
334 # endif /* not alloca */
336 # define REGEX_ALLOCATE alloca
338 /* Assumes a `char *destination' variable. */
339 # define REGEX_REALLOCATE(source, osize, nsize) \
340 (destination = (char *) alloca (nsize), \
341 memcpy (destination, source, osize))
343 /* No need to do anything to free, after alloca. */
344 # define REGEX_FREE(arg) ((void)0) /* Do nothing! But inhibit gcc warning. */
346 # endif /* not REGEX_MALLOC */
348 /* Define how to allocate the failure stack. */
350 # if defined REL_ALLOC && defined REGEX_MALLOC
352 # define REGEX_ALLOCATE_STACK(size) \
353 r_alloc (&failure_stack_ptr, (size))
354 # define REGEX_REALLOCATE_STACK(source, osize, nsize) \
355 r_re_alloc (&failure_stack_ptr, (nsize))
356 # define REGEX_FREE_STACK(ptr) \
357 r_alloc_free (&failure_stack_ptr)
359 # else /* not using relocating allocator */
363 # define REGEX_ALLOCATE_STACK malloc
364 # define REGEX_REALLOCATE_STACK(source, osize, nsize) realloc (source, nsize)
365 # define REGEX_FREE_STACK free
367 # else /* not REGEX_MALLOC */
369 # define REGEX_ALLOCATE_STACK alloca
371 # define REGEX_REALLOCATE_STACK(source, osize, nsize) \
372 REGEX_REALLOCATE (source, osize, nsize)
373 /* No need to explicitly free anything. */
374 # define REGEX_FREE_STACK(arg)
376 # endif /* not REGEX_MALLOC */
377 # endif /* not using relocating allocator */
380 /* True if `size1' is non-NULL and PTR is pointing anywhere inside
381 `string1' or just past its end. This works if PTR is NULL, which is
383 # define FIRST_STRING_P(ptr) \
384 (size1 && string1 <= (ptr) && (ptr) <= string1 + size1)
386 /* (Re)Allocate N items of type T using malloc, or fail. */
387 # define TALLOC(n, t) ((t *) malloc ((n) * sizeof (t)))
388 # define RETALLOC(addr, n, t) ((addr) = (t *) realloc (addr, (n) * sizeof (t)))
389 # define RETALLOC_IF(addr, n, t) \
390 if (addr) RETALLOC((addr), (n), t); else (addr) = TALLOC ((n), t)
391 # define REGEX_TALLOC(n, t) ((t *) REGEX_ALLOCATE ((n) * sizeof (t)))
393 # define BYTEWIDTH 8 /* In bits. */
395 # define STREQ(s1, s2) ((strcmp (s1, s2) == 0))
399 # define MAX(a, b) ((a) > (b) ? (a) : (b))
400 # define MIN(a, b) ((a) < (b) ? (a) : (b))
402 typedef char boolean;
406 static reg_errcode_t byte_regex_compile _RE_ARGS ((const char *pattern, size_t size,
408 struct re_pattern_buffer *bufp));
410 static int byte_re_match_2_internal PARAMS ((struct re_pattern_buffer *bufp,
411 const char *string1, int size1,
412 const char *string2, int size2,
414 struct re_registers *regs,
416 static int byte_re_search_2 PARAMS ((struct re_pattern_buffer *bufp,
417 const char *string1, int size1,
418 const char *string2, int size2,
419 int startpos, int range,
420 struct re_registers *regs, int stop));
421 static int byte_re_compile_fastmap PARAMS ((struct re_pattern_buffer *bufp));
424 static reg_errcode_t wcs_regex_compile _RE_ARGS ((const char *pattern, size_t size,
426 struct re_pattern_buffer *bufp));
429 static int wcs_re_match_2_internal PARAMS ((struct re_pattern_buffer *bufp,
430 const char *cstring1, int csize1,
431 const char *cstring2, int csize2,
433 struct re_registers *regs,
435 wchar_t *string1, int size1,
436 wchar_t *string2, int size2,
437 int *mbs_offset1, int *mbs_offset2));
438 static int wcs_re_search_2 PARAMS ((struct re_pattern_buffer *bufp,
439 const char *string1, int size1,
440 const char *string2, int size2,
441 int startpos, int range,
442 struct re_registers *regs, int stop));
443 static int wcs_re_compile_fastmap PARAMS ((struct re_pattern_buffer *bufp));
446 /* These are the command codes that appear in compiled regular
447 expressions. Some opcodes are followed by argument bytes. A
448 command code can specify any interpretation whatsoever for its
449 arguments. Zero bytes may appear in the compiled regular expression. */
455 /* Succeed right away--no more backtracking. */
458 /* Followed by one byte giving n, then by n literal bytes. */
462 /* Same as exactn, but contains binary data. */
466 /* Matches any (more or less) character. */
469 /* Matches any one char belonging to specified set. First
470 following byte is number of bitmap bytes. Then come bytes
471 for a bitmap saying which chars are in. Bits in each byte
472 are ordered low-bit-first. A character is in the set if its
473 bit is 1. A character too large to have a bit in the map is
474 automatically not in the set. */
475 /* ifdef MBS_SUPPORT, following element is length of character
476 classes, length of collating symbols, length of equivalence
477 classes, length of character ranges, and length of characters.
478 Next, character class element, collating symbols elements,
479 equivalence class elements, range elements, and character
481 See regex_compile function. */
484 /* Same parameters as charset, but match any character that is
485 not one of those specified. */
488 /* Start remembering the text that is matched, for storing in a
489 register. Followed by one byte with the register number, in
490 the range 0 to one less than the pattern buffer's re_nsub
491 field. Then followed by one byte with the number of groups
492 inner to this one. (This last has to be part of the
493 start_memory only because we need it in the on_failure_jump
497 /* Stop remembering the text that is matched and store it in a
498 memory register. Followed by one byte with the register
499 number, in the range 0 to one less than `re_nsub' in the
500 pattern buffer, and one byte with the number of inner groups,
501 just like `start_memory'. (We need the number of inner
502 groups here because we don't have any easy way of finding the
503 corresponding start_memory when we're at a stop_memory.) */
506 /* Match a duplicate of something remembered. Followed by one
507 byte containing the register number. */
510 /* Fail unless at beginning of line. */
513 /* Fail unless at end of line. */
516 /* Succeeds if at beginning of buffer (if emacs) or at beginning
517 of string to be matched (if not). */
520 /* Analogously, for end of buffer/string. */
523 /* Followed by two byte relative address to which to jump. */
526 /* Same as jump, but marks the end of an alternative. */
529 /* Followed by two-byte relative address of place to resume at
530 in case of failure. */
531 /* ifdef MBS_SUPPORT, the size of address is 1. */
534 /* Like on_failure_jump, but pushes a placeholder instead of the
535 current string position when executed. */
536 on_failure_keep_string_jump,
538 /* Throw away latest failure point and then jump to following
539 two-byte relative address. */
540 /* ifdef MBS_SUPPORT, the size of address is 1. */
543 /* Change to pop_failure_jump if know won't have to backtrack to
544 match; otherwise change to jump. This is used to jump
545 back to the beginning of a repeat. If what follows this jump
546 clearly won't match what the repeat does, such that we can be
547 sure that there is no use backtracking out of repetitions
548 already matched, then we change it to a pop_failure_jump.
549 Followed by two-byte address. */
550 /* ifdef MBS_SUPPORT, the size of address is 1. */
553 /* Jump to following two-byte address, and push a dummy failure
554 point. This failure point will be thrown away if an attempt
555 is made to use it for a failure. A `+' construct makes this
556 before the first repeat. Also used as an intermediary kind
557 of jump when compiling an alternative. */
558 /* ifdef MBS_SUPPORT, the size of address is 1. */
561 /* Push a dummy failure point and continue. Used at the end of
565 /* Followed by two-byte relative address and two-byte number n.
566 After matching N times, jump to the address upon failure. */
567 /* ifdef MBS_SUPPORT, the size of address is 1. */
570 /* Followed by two-byte relative address, and two-byte number n.
571 Jump to the address N times, then fail. */
572 /* ifdef MBS_SUPPORT, the size of address is 1. */
575 /* Set the following two-byte relative address to the
576 subsequent two-byte number. The address *includes* the two
578 /* ifdef MBS_SUPPORT, the size of address is 1. */
581 wordchar, /* Matches any word-constituent character. */
582 notwordchar, /* Matches any char that is not a word-constituent. */
584 wordbeg, /* Succeeds if at word beginning. */
585 wordend, /* Succeeds if at word end. */
587 wordbound, /* Succeeds if at a word boundary. */
588 notwordbound /* Succeeds if not at a word boundary. */
591 ,before_dot, /* Succeeds if before point. */
592 at_dot, /* Succeeds if at point. */
593 after_dot, /* Succeeds if after point. */
595 /* Matches any character whose syntax is specified. Followed by
596 a byte which contains a syntax code, e.g., Sword. */
599 /* Matches any character whose syntax is not that specified. */
603 #endif /* not INSIDE_RECURSION */
608 # define UCHAR_T unsigned char
609 # define COMPILED_BUFFER_VAR bufp->buffer
610 # define OFFSET_ADDRESS_SIZE 2
611 # define PREFIX(name) byte_##name
612 # define ARG_PREFIX(name) name
613 # define PUT_CHAR(c) putchar (c)
616 # define CHAR_T wchar_t
617 # define UCHAR_T wchar_t
618 # define COMPILED_BUFFER_VAR wc_buffer
619 # define OFFSET_ADDRESS_SIZE 1 /* the size which STORE_NUMBER macro use */
620 # define CHAR_CLASS_SIZE ((__alignof__(wctype_t)+sizeof(wctype_t))/sizeof(CHAR_T)+1)
621 # define PREFIX(name) wcs_##name
622 # define ARG_PREFIX(name) c##name
623 /* Should we use wide stream?? */
624 # define PUT_CHAR(c) printf ("%C", c);
630 # define INSIDE_RECURSION
632 # undef INSIDE_RECURSION
635 # define INSIDE_RECURSION
637 # undef INSIDE_RECURSION
641 #ifdef INSIDE_RECURSION
642 /* Common operations on the compiled pattern. */
644 /* Store NUMBER in two contiguous bytes starting at DESTINATION. */
645 /* ifdef MBS_SUPPORT, we store NUMBER in 1 element. */
648 # define STORE_NUMBER(destination, number) \
650 *(destination) = (UCHAR_T)(number); \
653 # define STORE_NUMBER(destination, number) \
655 (destination)[0] = (number) & 0377; \
656 (destination)[1] = (number) >> 8; \
660 /* Same as STORE_NUMBER, except increment DESTINATION to
661 the byte after where the number is stored. Therefore, DESTINATION
662 must be an lvalue. */
663 /* ifdef MBS_SUPPORT, we store NUMBER in 1 element. */
665 # define STORE_NUMBER_AND_INCR(destination, number) \
667 STORE_NUMBER (destination, number); \
668 (destination) += OFFSET_ADDRESS_SIZE; \
671 /* Put into DESTINATION a number stored in two contiguous bytes starting
673 /* ifdef MBS_SUPPORT, we store NUMBER in 1 element. */
676 # define EXTRACT_NUMBER(destination, source) \
678 (destination) = *(source); \
681 # define EXTRACT_NUMBER(destination, source) \
683 (destination) = *(source) & 0377; \
684 (destination) += SIGN_EXTEND_CHAR (*((source) + 1)) << 8; \
689 static void PREFIX(extract_number) _RE_ARGS ((int *dest, UCHAR_T *source));
691 PREFIX(extract_number) (dest, source)
698 int temp = SIGN_EXTEND_CHAR (*(source + 1));
699 *dest = *source & 0377;
704 # ifndef EXTRACT_MACROS /* To debug the macros. */
705 # undef EXTRACT_NUMBER
706 # define EXTRACT_NUMBER(dest, src) PREFIX(extract_number) (&dest, src)
707 # endif /* not EXTRACT_MACROS */
711 /* Same as EXTRACT_NUMBER, except increment SOURCE to after the number.
712 SOURCE must be an lvalue. */
714 # define EXTRACT_NUMBER_AND_INCR(destination, source) \
716 EXTRACT_NUMBER (destination, source); \
717 (source) += OFFSET_ADDRESS_SIZE; \
721 static void PREFIX(extract_number_and_incr) _RE_ARGS ((int *destination,
724 PREFIX(extract_number_and_incr) (destination, source)
728 PREFIX(extract_number) (destination, *source);
729 *source += OFFSET_ADDRESS_SIZE;
732 # ifndef EXTRACT_MACROS
733 # undef EXTRACT_NUMBER_AND_INCR
734 # define EXTRACT_NUMBER_AND_INCR(dest, src) \
735 PREFIX(extract_number_and_incr) (&dest, &src)
736 # endif /* not EXTRACT_MACROS */
742 /* If DEBUG is defined, Regex prints many voluminous messages about what
743 it is doing (if the variable `debug' is nonzero). If linked with the
744 main program in `iregex.c', you can enter patterns and strings
745 interactively. And if linked with the main program in `main.c' and
746 the other test files, you can run the already-written tests. */
750 # ifndef DEFINED_ONCE
752 /* We use standard I/O for debugging. */
755 /* It is useful to test things that ``must'' be true when debugging. */
760 # define DEBUG_STATEMENT(e) e
761 # define DEBUG_PRINT1(x) if (debug) printf (x)
762 # define DEBUG_PRINT2(x1, x2) if (debug) printf (x1, x2)
763 # define DEBUG_PRINT3(x1, x2, x3) if (debug) printf (x1, x2, x3)
764 # define DEBUG_PRINT4(x1, x2, x3, x4) if (debug) printf (x1, x2, x3, x4)
765 # endif /* not DEFINED_ONCE */
767 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) \
768 if (debug) PREFIX(print_partial_compiled_pattern) (s, e)
769 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) \
770 if (debug) PREFIX(print_double_string) (w, s1, sz1, s2, sz2)
773 /* Print the fastmap in human-readable form. */
775 # ifndef DEFINED_ONCE
777 print_fastmap (fastmap)
780 unsigned was_a_range = 0;
783 while (i < (1 << BYTEWIDTH))
789 while (i < (1 << BYTEWIDTH) && fastmap[i])
803 # endif /* not DEFINED_ONCE */
806 /* Print a compiled pattern string in human-readable form, starting at
807 the START pointer into it and ending just before the pointer END. */
810 PREFIX(print_partial_compiled_pattern) (start, end)
825 /* Loop over pattern commands. */
829 printf ("%td:\t", p - start);
831 printf ("%ld:\t", (long int) (p - start));
834 switch ((re_opcode_t) *p++)
842 printf ("/exactn/%d", mcnt);
854 printf ("/exactn_bin/%d", mcnt);
857 printf("/%lx", (long int) *p++);
861 # endif /* MBS_SUPPORT */
865 printf ("/start_memory/%d/%ld", mcnt, (long int) *p++);
870 printf ("/stop_memory/%d/%ld", mcnt, (long int) *p++);
874 printf ("/duplicate/%ld", (long int) *p++);
887 printf ("/charset [%s",
888 (re_opcode_t) *(workp - 1) == charset_not ? "^" : "");
890 length = *workp++; /* the length of char_classes */
891 for (i=0 ; i<length ; i++)
892 printf("[:%lx:]", (long int) *p++);
893 length = *workp++; /* the length of collating_symbol */
894 for (i=0 ; i<length ;)
898 PUT_CHAR((i++,*p++));
902 length = *workp++; /* the length of equivalence_class */
903 for (i=0 ; i<length ;)
907 PUT_CHAR((i++,*p++));
911 length = *workp++; /* the length of char_range */
912 for (i=0 ; i<length ; i++)
914 wchar_t range_start = *p++;
915 wchar_t range_end = *p++;
916 printf("%C-%C", range_start, range_end);
918 length = *workp++; /* the length of char */
919 for (i=0 ; i<length ; i++)
923 register int c, last = -100;
924 register int in_range = 0;
926 printf ("/charset [%s",
927 (re_opcode_t) *(p - 1) == charset_not ? "^" : "");
929 assert (p + *p < pend);
931 for (c = 0; c < 256; c++)
933 && (p[1 + (c/8)] & (1 << (c % 8))))
935 /* Are we starting a range? */
936 if (last + 1 == c && ! in_range)
941 /* Have we broken a range? */
942 else if (last + 1 != c && in_range)
972 case on_failure_jump:
973 PREFIX(extract_number_and_incr) (&mcnt, &p);
975 printf ("/on_failure_jump to %td", p + mcnt - start);
977 printf ("/on_failure_jump to %ld", (long int) (p + mcnt - start));
981 case on_failure_keep_string_jump:
982 PREFIX(extract_number_and_incr) (&mcnt, &p);
984 printf ("/on_failure_keep_string_jump to %td", p + mcnt - start);
986 printf ("/on_failure_keep_string_jump to %ld",
987 (long int) (p + mcnt - start));
991 case dummy_failure_jump:
992 PREFIX(extract_number_and_incr) (&mcnt, &p);
994 printf ("/dummy_failure_jump to %td", p + mcnt - start);
996 printf ("/dummy_failure_jump to %ld", (long int) (p + mcnt - start));
1000 case push_dummy_failure:
1001 printf ("/push_dummy_failure");
1004 case maybe_pop_jump:
1005 PREFIX(extract_number_and_incr) (&mcnt, &p);
1007 printf ("/maybe_pop_jump to %td", p + mcnt - start);
1009 printf ("/maybe_pop_jump to %ld", (long int) (p + mcnt - start));
1013 case pop_failure_jump:
1014 PREFIX(extract_number_and_incr) (&mcnt, &p);
1016 printf ("/pop_failure_jump to %td", p + mcnt - start);
1018 printf ("/pop_failure_jump to %ld", (long int) (p + mcnt - start));
1023 PREFIX(extract_number_and_incr) (&mcnt, &p);
1025 printf ("/jump_past_alt to %td", p + mcnt - start);
1027 printf ("/jump_past_alt to %ld", (long int) (p + mcnt - start));
1032 PREFIX(extract_number_and_incr) (&mcnt, &p);
1034 printf ("/jump to %td", p + mcnt - start);
1036 printf ("/jump to %ld", (long int) (p + mcnt - start));
1041 PREFIX(extract_number_and_incr) (&mcnt, &p);
1043 PREFIX(extract_number_and_incr) (&mcnt2, &p);
1045 printf ("/succeed_n to %td, %d times", p1 - start, mcnt2);
1047 printf ("/succeed_n to %ld, %d times",
1048 (long int) (p1 - start), mcnt2);
1053 PREFIX(extract_number_and_incr) (&mcnt, &p);
1055 PREFIX(extract_number_and_incr) (&mcnt2, &p);
1056 printf ("/jump_n to %d, %d times", p1 - start, mcnt2);
1060 PREFIX(extract_number_and_incr) (&mcnt, &p);
1062 PREFIX(extract_number_and_incr) (&mcnt2, &p);
1064 printf ("/set_number_at location %td to %d", p1 - start, mcnt2);
1066 printf ("/set_number_at location %ld to %d",
1067 (long int) (p1 - start), mcnt2);
1072 printf ("/wordbound");
1076 printf ("/notwordbound");
1080 printf ("/wordbeg");
1084 printf ("/wordend");
1089 printf ("/before_dot");
1097 printf ("/after_dot");
1101 printf ("/syntaxspec");
1103 printf ("/%d", mcnt);
1107 printf ("/notsyntaxspec");
1109 printf ("/%d", mcnt);
1114 printf ("/wordchar");
1118 printf ("/notwordchar");
1130 printf ("?%ld", (long int) *(p-1));
1137 printf ("%td:\tend of pattern.\n", p - start);
1139 printf ("%ld:\tend of pattern.\n", (long int) (p - start));
1145 PREFIX(print_compiled_pattern) (bufp)
1146 struct re_pattern_buffer *bufp;
1148 UCHAR_T *buffer = (UCHAR_T*) bufp->buffer;
1150 PREFIX(print_partial_compiled_pattern) (buffer, buffer
1151 + bufp->used / sizeof(UCHAR_T));
1152 printf ("%ld bytes used/%ld bytes allocated.\n",
1153 bufp->used, bufp->allocated);
1155 if (bufp->fastmap_accurate && bufp->fastmap)
1157 printf ("fastmap: ");
1158 print_fastmap (bufp->fastmap);
1162 printf ("re_nsub: %Zd\t", bufp->re_nsub);
1164 printf ("re_nsub: %ld\t", (long int) bufp->re_nsub);
1166 printf ("regs_alloc: %d\t", bufp->regs_allocated);
1167 printf ("can_be_null: %d\t", bufp->can_be_null);
1168 printf ("newline_anchor: %d\n", bufp->newline_anchor);
1169 printf ("no_sub: %d\t", bufp->no_sub);
1170 printf ("not_bol: %d\t", bufp->not_bol);
1171 printf ("not_eol: %d\t", bufp->not_eol);
1172 printf ("syntax: %lx\n", bufp->syntax);
1173 /* Perhaps we should print the translate table? */
1178 PREFIX(print_double_string) (where, string1, size1, string2, size2)
1179 const CHAR_T *where;
1180 const CHAR_T *string1;
1181 const CHAR_T *string2;
1193 if (FIRST_STRING_P (where))
1195 for (this_char = where - string1; this_char < size1; this_char++)
1196 PUT_CHAR (string1[this_char]);
1202 for (this_char = where - string2; this_char < size2; this_char++)
1204 PUT_CHAR (string2[this_char]);
1207 fputs ("...", stdout);
1214 # ifndef DEFINED_ONCE
1223 # else /* not DEBUG */
1225 # ifndef DEFINED_ONCE
1229 # define DEBUG_STATEMENT(e)
1230 # define DEBUG_PRINT1(x)
1231 # define DEBUG_PRINT2(x1, x2)
1232 # define DEBUG_PRINT3(x1, x2, x3)
1233 # define DEBUG_PRINT4(x1, x2, x3, x4)
1234 # endif /* not DEFINED_ONCE */
1235 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e)
1236 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2)
1238 # endif /* not DEBUG */
1243 /* This convert a multibyte string to a wide character string.
1244 And write their correspondances to offset_buffer(see below)
1245 and write whether each wchar_t is binary data to is_binary.
1246 This assume invalid multibyte sequences as binary data.
1247 We assume offset_buffer and is_binary is already allocated
1250 static size_t convert_mbs_to_wcs (CHAR_T *dest, const unsigned char* src,
1251 size_t len, int *offset_buffer,
1254 convert_mbs_to_wcs (dest, src, len, offset_buffer, is_binary)
1256 const unsigned char* src;
1257 size_t len; /* the length of multibyte string. */
1259 /* It hold correspondances between src(char string) and
1260 dest(wchar_t string) for optimization.
1262 dest = {'X', 'Y', 'Z'}
1263 (each "xxx", "y" and "zz" represent one multibyte character
1264 corresponding to 'X', 'Y' and 'Z'.)
1265 offset_buffer = {0, 0+3("xxx"), 0+3+1("y"), 0+3+1+2("zz")}
1271 wchar_t *pdest = dest;
1272 const unsigned char *psrc = src;
1273 size_t wc_count = 0;
1277 size_t mb_remain = len;
1278 size_t mb_count = 0;
1280 /* Initialize the conversion state. */
1281 memset (&mbs, 0, sizeof (mbstate_t));
1283 offset_buffer[0] = 0;
1284 for( ; mb_remain > 0 ; ++wc_count, ++pdest, mb_remain -= consumed,
1287 consumed = mbrtowc (pdest, psrc, mb_remain, &mbs);
1290 /* failed to convert. maybe src contains binary data.
1291 So we consume 1 byte manualy. */
1295 is_binary[wc_count] = TRUE;
1298 is_binary[wc_count] = FALSE;
1299 /* In sjis encoding, we use yen sign as escape character in
1300 place of reverse solidus. So we convert 0x5c(yen sign in
1301 sjis) to not 0xa5(yen sign in UCS2) but 0x5c(reverse
1302 solidus in UCS2). */
1303 if (consumed == 1 && (int) *psrc == 0x5c && (int) *pdest == 0xa5)
1304 *pdest = (wchar_t) *psrc;
1306 offset_buffer[wc_count + 1] = mb_count += consumed;
1309 /* Fill remain of the buffer with sentinel. */
1310 for (i = wc_count + 1 ; i <= len ; i++)
1311 offset_buffer[i] = mb_count + 1;
1318 #else /* not INSIDE_RECURSION */
1320 /* Set by `re_set_syntax' to the current regexp syntax to recognize. Can
1321 also be assigned to arbitrarily: each pattern buffer stores its own
1322 syntax, so it can be changed between regex compilations. */
1323 /* This has no initializer because initialized variables in Emacs
1324 become read-only after dumping. */
1325 reg_syntax_t re_syntax_options;
1328 /* Specify the precise syntax of regexps for compilation. This provides
1329 for compatibility for various utilities which historically have
1330 different, incompatible syntaxes.
1332 The argument SYNTAX is a bit mask comprised of the various bits
1333 defined in regex.h. We return the old syntax. */
1336 re_set_syntax (syntax)
1337 reg_syntax_t syntax;
1339 reg_syntax_t ret = re_syntax_options;
1341 re_syntax_options = syntax;
1343 if (syntax & RE_DEBUG)
1345 else if (debug) /* was on but now is not */
1351 weak_alias (__re_set_syntax, re_set_syntax)
1354 /* This table gives an error message for each of the error codes listed
1355 in regex.h. Obviously the order here has to be same as there.
1356 POSIX doesn't require that we do anything for REG_NOERROR,
1357 but why not be nice? */
1359 static const char re_error_msgid[] =
1361 # define REG_NOERROR_IDX 0
1362 gettext_noop ("Success") /* REG_NOERROR */
1364 # define REG_NOMATCH_IDX (REG_NOERROR_IDX + sizeof "Success")
1365 gettext_noop ("No match") /* REG_NOMATCH */
1367 # define REG_BADPAT_IDX (REG_NOMATCH_IDX + sizeof "No match")
1368 gettext_noop ("Invalid regular expression") /* REG_BADPAT */
1370 # define REG_ECOLLATE_IDX (REG_BADPAT_IDX + sizeof "Invalid regular expression")
1371 gettext_noop ("Invalid collation character") /* REG_ECOLLATE */
1373 # define REG_ECTYPE_IDX (REG_ECOLLATE_IDX + sizeof "Invalid collation character")
1374 gettext_noop ("Invalid character class name") /* REG_ECTYPE */
1376 # define REG_EESCAPE_IDX (REG_ECTYPE_IDX + sizeof "Invalid character class name")
1377 gettext_noop ("Trailing backslash") /* REG_EESCAPE */
1379 # define REG_ESUBREG_IDX (REG_EESCAPE_IDX + sizeof "Trailing backslash")
1380 gettext_noop ("Invalid back reference") /* REG_ESUBREG */
1382 # define REG_EBRACK_IDX (REG_ESUBREG_IDX + sizeof "Invalid back reference")
1383 gettext_noop ("Unmatched [ or [^") /* REG_EBRACK */
1385 # define REG_EPAREN_IDX (REG_EBRACK_IDX + sizeof "Unmatched [ or [^")
1386 gettext_noop ("Unmatched ( or \\(") /* REG_EPAREN */
1388 # define REG_EBRACE_IDX (REG_EPAREN_IDX + sizeof "Unmatched ( or \\(")
1389 gettext_noop ("Unmatched \\{") /* REG_EBRACE */
1391 # define REG_BADBR_IDX (REG_EBRACE_IDX + sizeof "Unmatched \\{")
1392 gettext_noop ("Invalid content of \\{\\}") /* REG_BADBR */
1394 # define REG_ERANGE_IDX (REG_BADBR_IDX + sizeof "Invalid content of \\{\\}")
1395 gettext_noop ("Invalid range end") /* REG_ERANGE */
1397 # define REG_ESPACE_IDX (REG_ERANGE_IDX + sizeof "Invalid range end")
1398 gettext_noop ("Memory exhausted") /* REG_ESPACE */
1400 # define REG_BADRPT_IDX (REG_ESPACE_IDX + sizeof "Memory exhausted")
1401 gettext_noop ("Invalid preceding regular expression") /* REG_BADRPT */
1403 # define REG_EEND_IDX (REG_BADRPT_IDX + sizeof "Invalid preceding regular expression")
1404 gettext_noop ("Premature end of regular expression") /* REG_EEND */
1406 # define REG_ESIZE_IDX (REG_EEND_IDX + sizeof "Premature end of regular expression")
1407 gettext_noop ("Regular expression too big") /* REG_ESIZE */
1409 # define REG_ERPAREN_IDX (REG_ESIZE_IDX + sizeof "Regular expression too big")
1410 gettext_noop ("Unmatched ) or \\)") /* REG_ERPAREN */
1413 static const size_t re_error_msgid_idx[] =
1434 #endif /* INSIDE_RECURSION */
1436 #ifndef DEFINED_ONCE
1437 /* Avoiding alloca during matching, to placate r_alloc. */
1439 /* Define MATCH_MAY_ALLOCATE unless we need to make sure that the
1440 searching and matching functions should not call alloca. On some
1441 systems, alloca is implemented in terms of malloc, and if we're
1442 using the relocating allocator routines, then malloc could cause a
1443 relocation, which might (if the strings being searched are in the
1444 ralloc heap) shift the data out from underneath the regexp
1447 Here's another reason to avoid allocation: Emacs
1448 processes input from X in a signal handler; processing X input may
1449 call malloc; if input arrives while a matching routine is calling
1450 malloc, then we're scrod. But Emacs can't just block input while
1451 calling matching routines; then we don't notice interrupts when
1452 they come in. So, Emacs blocks input around all regexp calls
1453 except the matching calls, which it leaves unprotected, in the
1454 faith that they will not malloc. */
1456 /* Normally, this is fine. */
1457 # define MATCH_MAY_ALLOCATE
1459 /* When using GNU C, we are not REALLY using the C alloca, no matter
1460 what config.h may say. So don't take precautions for it. */
1465 /* The match routines may not allocate if (1) they would do it with malloc
1466 and (2) it's not safe for them to use malloc.
1467 Note that if REL_ALLOC is defined, matching would not use malloc for the
1468 failure stack, but we would still use it for the register vectors;
1469 so REL_ALLOC should not affect this. */
1470 # if (defined C_ALLOCA || defined REGEX_MALLOC) && defined emacs
1471 # undef MATCH_MAY_ALLOCATE
1473 #endif /* not DEFINED_ONCE */
1475 #ifdef INSIDE_RECURSION
1476 /* Failure stack declarations and macros; both re_compile_fastmap and
1477 re_match_2 use a failure stack. These have to be macros because of
1478 REGEX_ALLOCATE_STACK. */
1481 /* Number of failure points for which to initially allocate space
1482 when matching. If this number is exceeded, we allocate more
1483 space, so it is not a hard limit. */
1484 # ifndef INIT_FAILURE_ALLOC
1485 # define INIT_FAILURE_ALLOC 5
1488 /* Roughly the maximum number of failure points on the stack. Would be
1489 exactly that if always used MAX_FAILURE_ITEMS items each time we failed.
1490 This is a variable only so users of regex can assign to it; we never
1491 change it ourselves. */
1493 # ifdef INT_IS_16BIT
1495 # ifndef DEFINED_ONCE
1496 # if defined MATCH_MAY_ALLOCATE
1497 /* 4400 was enough to cause a crash on Alpha OSF/1,
1498 whose default stack limit is 2mb. */
1499 long int re_max_failures = 4000;
1501 long int re_max_failures = 2000;
1505 union PREFIX(fail_stack_elt)
1511 typedef union PREFIX(fail_stack_elt) PREFIX(fail_stack_elt_t);
1515 PREFIX(fail_stack_elt_t) *stack;
1516 unsigned long int size;
1517 unsigned long int avail; /* Offset of next open position. */
1518 } PREFIX(fail_stack_type);
1520 # else /* not INT_IS_16BIT */
1522 # ifndef DEFINED_ONCE
1523 # if defined MATCH_MAY_ALLOCATE
1524 /* 4400 was enough to cause a crash on Alpha OSF/1,
1525 whose default stack limit is 2mb. */
1526 int re_max_failures = 4000;
1528 int re_max_failures = 2000;
1532 union PREFIX(fail_stack_elt)
1538 typedef union PREFIX(fail_stack_elt) PREFIX(fail_stack_elt_t);
1542 PREFIX(fail_stack_elt_t) *stack;
1544 unsigned avail; /* Offset of next open position. */
1545 } PREFIX(fail_stack_type);
1547 # endif /* INT_IS_16BIT */
1549 # ifndef DEFINED_ONCE
1550 # define FAIL_STACK_EMPTY() (fail_stack.avail == 0)
1551 # define FAIL_STACK_PTR_EMPTY() (fail_stack_ptr->avail == 0)
1552 # define FAIL_STACK_FULL() (fail_stack.avail == fail_stack.size)
1556 /* Define macros to initialize and free the failure stack.
1557 Do `return -2' if the alloc fails. */
1559 # ifdef MATCH_MAY_ALLOCATE
1560 # define INIT_FAIL_STACK() \
1562 fail_stack.stack = (PREFIX(fail_stack_elt_t) *) \
1563 REGEX_ALLOCATE_STACK (INIT_FAILURE_ALLOC * sizeof (PREFIX(fail_stack_elt_t))); \
1565 if (fail_stack.stack == NULL) \
1568 fail_stack.size = INIT_FAILURE_ALLOC; \
1569 fail_stack.avail = 0; \
1572 # define RESET_FAIL_STACK() REGEX_FREE_STACK (fail_stack.stack)
1574 # define INIT_FAIL_STACK() \
1576 fail_stack.avail = 0; \
1579 # define RESET_FAIL_STACK()
1583 /* Double the size of FAIL_STACK, up to approximately `re_max_failures' items.
1585 Return 1 if succeeds, and 0 if either ran out of memory
1586 allocating space for it or it was already too large.
1588 REGEX_REALLOCATE_STACK requires `destination' be declared. */
1590 # define DOUBLE_FAIL_STACK(fail_stack) \
1591 ((fail_stack).size > (unsigned) (re_max_failures * MAX_FAILURE_ITEMS) \
1593 : ((fail_stack).stack = (PREFIX(fail_stack_elt_t) *) \
1594 REGEX_REALLOCATE_STACK ((fail_stack).stack, \
1595 (fail_stack).size * sizeof (PREFIX(fail_stack_elt_t)), \
1596 ((fail_stack).size << 1) * sizeof (PREFIX(fail_stack_elt_t))),\
1598 (fail_stack).stack == NULL \
1600 : ((fail_stack).size <<= 1, \
1604 /* Push pointer POINTER on FAIL_STACK.
1605 Return 1 if was able to do so and 0 if ran out of memory allocating
1607 # define PUSH_PATTERN_OP(POINTER, FAIL_STACK) \
1608 ((FAIL_STACK_FULL () \
1609 && !DOUBLE_FAIL_STACK (FAIL_STACK)) \
1611 : ((FAIL_STACK).stack[(FAIL_STACK).avail++].pointer = POINTER, \
1614 /* Push a pointer value onto the failure stack.
1615 Assumes the variable `fail_stack'. Probably should only
1616 be called from within `PUSH_FAILURE_POINT'. */
1617 # define PUSH_FAILURE_POINTER(item) \
1618 fail_stack.stack[fail_stack.avail++].pointer = (UCHAR_T *) (item)
1620 /* This pushes an integer-valued item onto the failure stack.
1621 Assumes the variable `fail_stack'. Probably should only
1622 be called from within `PUSH_FAILURE_POINT'. */
1623 # define PUSH_FAILURE_INT(item) \
1624 fail_stack.stack[fail_stack.avail++].integer = (item)
1626 /* Push a fail_stack_elt_t value onto the failure stack.
1627 Assumes the variable `fail_stack'. Probably should only
1628 be called from within `PUSH_FAILURE_POINT'. */
1629 # define PUSH_FAILURE_ELT(item) \
1630 fail_stack.stack[fail_stack.avail++] = (item)
1632 /* These three POP... operations complement the three PUSH... operations.
1633 All assume that `fail_stack' is nonempty. */
1634 # define POP_FAILURE_POINTER() fail_stack.stack[--fail_stack.avail].pointer
1635 # define POP_FAILURE_INT() fail_stack.stack[--fail_stack.avail].integer
1636 # define POP_FAILURE_ELT() fail_stack.stack[--fail_stack.avail]
1638 /* Used to omit pushing failure point id's when we're not debugging. */
1640 # define DEBUG_PUSH PUSH_FAILURE_INT
1641 # define DEBUG_POP(item_addr) *(item_addr) = POP_FAILURE_INT ()
1643 # define DEBUG_PUSH(item)
1644 # define DEBUG_POP(item_addr)
1648 /* Push the information about the state we will need
1649 if we ever fail back to it.
1651 Requires variables fail_stack, regstart, regend, reg_info, and
1652 num_regs_pushed be declared. DOUBLE_FAIL_STACK requires `destination'
1655 Does `return FAILURE_CODE' if runs out of memory. */
1657 # define PUSH_FAILURE_POINT(pattern_place, string_place, failure_code) \
1659 char *destination; \
1660 /* Must be int, so when we don't save any registers, the arithmetic \
1661 of 0 + -1 isn't done as unsigned. */ \
1662 /* Can't be int, since there is not a shred of a guarantee that int \
1663 is wide enough to hold a value of something to which pointer can \
1665 active_reg_t this_reg; \
1667 DEBUG_STATEMENT (failure_id++); \
1668 DEBUG_STATEMENT (nfailure_points_pushed++); \
1669 DEBUG_PRINT2 ("\nPUSH_FAILURE_POINT #%u:\n", failure_id); \
1670 DEBUG_PRINT2 (" Before push, next avail: %d\n", (fail_stack).avail);\
1671 DEBUG_PRINT2 (" size: %d\n", (fail_stack).size);\
1673 DEBUG_PRINT2 (" slots needed: %ld\n", NUM_FAILURE_ITEMS); \
1674 DEBUG_PRINT2 (" available: %d\n", REMAINING_AVAIL_SLOTS); \
1676 /* Ensure we have enough space allocated for what we will push. */ \
1677 while (REMAINING_AVAIL_SLOTS < NUM_FAILURE_ITEMS) \
1679 if (!DOUBLE_FAIL_STACK (fail_stack)) \
1680 return failure_code; \
1682 DEBUG_PRINT2 ("\n Doubled stack; size now: %d\n", \
1683 (fail_stack).size); \
1684 DEBUG_PRINT2 (" slots available: %d\n", REMAINING_AVAIL_SLOTS);\
1687 /* Push the info, starting with the registers. */ \
1688 DEBUG_PRINT1 ("\n"); \
1691 for (this_reg = lowest_active_reg; this_reg <= highest_active_reg; \
1694 DEBUG_PRINT2 (" Pushing reg: %lu\n", this_reg); \
1695 DEBUG_STATEMENT (num_regs_pushed++); \
1697 DEBUG_PRINT2 (" start: %p\n", regstart[this_reg]); \
1698 PUSH_FAILURE_POINTER (regstart[this_reg]); \
1700 DEBUG_PRINT2 (" end: %p\n", regend[this_reg]); \
1701 PUSH_FAILURE_POINTER (regend[this_reg]); \
1703 DEBUG_PRINT2 (" info: %p\n ", \
1704 reg_info[this_reg].word.pointer); \
1705 DEBUG_PRINT2 (" match_null=%d", \
1706 REG_MATCH_NULL_STRING_P (reg_info[this_reg])); \
1707 DEBUG_PRINT2 (" active=%d", IS_ACTIVE (reg_info[this_reg])); \
1708 DEBUG_PRINT2 (" matched_something=%d", \
1709 MATCHED_SOMETHING (reg_info[this_reg])); \
1710 DEBUG_PRINT2 (" ever_matched=%d", \
1711 EVER_MATCHED_SOMETHING (reg_info[this_reg])); \
1712 DEBUG_PRINT1 ("\n"); \
1713 PUSH_FAILURE_ELT (reg_info[this_reg].word); \
1716 DEBUG_PRINT2 (" Pushing low active reg: %ld\n", lowest_active_reg);\
1717 PUSH_FAILURE_INT (lowest_active_reg); \
1719 DEBUG_PRINT2 (" Pushing high active reg: %ld\n", highest_active_reg);\
1720 PUSH_FAILURE_INT (highest_active_reg); \
1722 DEBUG_PRINT2 (" Pushing pattern %p:\n", pattern_place); \
1723 DEBUG_PRINT_COMPILED_PATTERN (bufp, pattern_place, pend); \
1724 PUSH_FAILURE_POINTER (pattern_place); \
1726 DEBUG_PRINT2 (" Pushing string %p: `", string_place); \
1727 DEBUG_PRINT_DOUBLE_STRING (string_place, string1, size1, string2, \
1729 DEBUG_PRINT1 ("'\n"); \
1730 PUSH_FAILURE_POINTER (string_place); \
1732 DEBUG_PRINT2 (" Pushing failure id: %u\n", failure_id); \
1733 DEBUG_PUSH (failure_id); \
1736 # ifndef DEFINED_ONCE
1737 /* This is the number of items that are pushed and popped on the stack
1738 for each register. */
1739 # define NUM_REG_ITEMS 3
1741 /* Individual items aside from the registers. */
1743 # define NUM_NONREG_ITEMS 5 /* Includes failure point id. */
1745 # define NUM_NONREG_ITEMS 4
1748 /* We push at most this many items on the stack. */
1749 /* We used to use (num_regs - 1), which is the number of registers
1750 this regexp will save; but that was changed to 5
1751 to avoid stack overflow for a regexp with lots of parens. */
1752 # define MAX_FAILURE_ITEMS (5 * NUM_REG_ITEMS + NUM_NONREG_ITEMS)
1754 /* We actually push this many items. */
1755 # define NUM_FAILURE_ITEMS \
1757 ? 0 : highest_active_reg - lowest_active_reg + 1) \
1761 /* How many items can still be added to the stack without overflowing it. */
1762 # define REMAINING_AVAIL_SLOTS ((fail_stack).size - (fail_stack).avail)
1763 # endif /* not DEFINED_ONCE */
1766 /* Pops what PUSH_FAIL_STACK pushes.
1768 We restore into the parameters, all of which should be lvalues:
1769 STR -- the saved data position.
1770 PAT -- the saved pattern position.
1771 LOW_REG, HIGH_REG -- the highest and lowest active registers.
1772 REGSTART, REGEND -- arrays of string positions.
1773 REG_INFO -- array of information about each subexpression.
1775 Also assumes the variables `fail_stack' and (if debugging), `bufp',
1776 `pend', `string1', `size1', `string2', and `size2'. */
1777 # define POP_FAILURE_POINT(str, pat, low_reg, high_reg, regstart, regend, reg_info)\
1779 DEBUG_STATEMENT (unsigned failure_id;) \
1780 active_reg_t this_reg; \
1781 const UCHAR_T *string_temp; \
1783 assert (!FAIL_STACK_EMPTY ()); \
1785 /* Remove failure points and point to how many regs pushed. */ \
1786 DEBUG_PRINT1 ("POP_FAILURE_POINT:\n"); \
1787 DEBUG_PRINT2 (" Before pop, next avail: %d\n", fail_stack.avail); \
1788 DEBUG_PRINT2 (" size: %d\n", fail_stack.size); \
1790 assert (fail_stack.avail >= NUM_NONREG_ITEMS); \
1792 DEBUG_POP (&failure_id); \
1793 DEBUG_PRINT2 (" Popping failure id: %u\n", failure_id); \
1795 /* If the saved string location is NULL, it came from an \
1796 on_failure_keep_string_jump opcode, and we want to throw away the \
1797 saved NULL, thus retaining our current position in the string. */ \
1798 string_temp = POP_FAILURE_POINTER (); \
1799 if (string_temp != NULL) \
1800 str = (const CHAR_T *) string_temp; \
1802 DEBUG_PRINT2 (" Popping string %p: `", str); \
1803 DEBUG_PRINT_DOUBLE_STRING (str, string1, size1, string2, size2); \
1804 DEBUG_PRINT1 ("'\n"); \
1806 pat = (UCHAR_T *) POP_FAILURE_POINTER (); \
1807 DEBUG_PRINT2 (" Popping pattern %p:\n", pat); \
1808 DEBUG_PRINT_COMPILED_PATTERN (bufp, pat, pend); \
1810 /* Restore register info. */ \
1811 high_reg = (active_reg_t) POP_FAILURE_INT (); \
1812 DEBUG_PRINT2 (" Popping high active reg: %ld\n", high_reg); \
1814 low_reg = (active_reg_t) POP_FAILURE_INT (); \
1815 DEBUG_PRINT2 (" Popping low active reg: %ld\n", low_reg); \
1818 for (this_reg = high_reg; this_reg >= low_reg; this_reg--) \
1820 DEBUG_PRINT2 (" Popping reg: %ld\n", this_reg); \
1822 reg_info[this_reg].word = POP_FAILURE_ELT (); \
1823 DEBUG_PRINT2 (" info: %p\n", \
1824 reg_info[this_reg].word.pointer); \
1826 regend[this_reg] = (const CHAR_T *) POP_FAILURE_POINTER (); \
1827 DEBUG_PRINT2 (" end: %p\n", regend[this_reg]); \
1829 regstart[this_reg] = (const CHAR_T *) POP_FAILURE_POINTER (); \
1830 DEBUG_PRINT2 (" start: %p\n", regstart[this_reg]); \
1834 for (this_reg = highest_active_reg; this_reg > high_reg; this_reg--) \
1836 reg_info[this_reg].word.integer = 0; \
1837 regend[this_reg] = 0; \
1838 regstart[this_reg] = 0; \
1840 highest_active_reg = high_reg; \
1843 set_regs_matched_done = 0; \
1844 DEBUG_STATEMENT (nfailure_points_popped++); \
1845 } /* POP_FAILURE_POINT */
1847 /* Structure for per-register (a.k.a. per-group) information.
1848 Other register information, such as the
1849 starting and ending positions (which are addresses), and the list of
1850 inner groups (which is a bits list) are maintained in separate
1853 We are making a (strictly speaking) nonportable assumption here: that
1854 the compiler will pack our bit fields into something that fits into
1855 the type of `word', i.e., is something that fits into one item on the
1859 /* Declarations and macros for re_match_2. */
1863 PREFIX(fail_stack_elt_t) word;
1866 /* This field is one if this group can match the empty string,
1867 zero if not. If not yet determined, `MATCH_NULL_UNSET_VALUE'. */
1868 # define MATCH_NULL_UNSET_VALUE 3
1869 unsigned match_null_string_p : 2;
1870 unsigned is_active : 1;
1871 unsigned matched_something : 1;
1872 unsigned ever_matched_something : 1;
1874 } PREFIX(register_info_type);
1876 # ifndef DEFINED_ONCE
1877 # define REG_MATCH_NULL_STRING_P(R) ((R).bits.match_null_string_p)
1878 # define IS_ACTIVE(R) ((R).bits.is_active)
1879 # define MATCHED_SOMETHING(R) ((R).bits.matched_something)
1880 # define EVER_MATCHED_SOMETHING(R) ((R).bits.ever_matched_something)
1883 /* Call this when have matched a real character; it sets `matched' flags
1884 for the subexpressions which we are currently inside. Also records
1885 that those subexprs have matched. */
1886 # define SET_REGS_MATCHED() \
1889 if (!set_regs_matched_done) \
1892 set_regs_matched_done = 1; \
1893 for (r = lowest_active_reg; r <= highest_active_reg; r++) \
1895 MATCHED_SOMETHING (reg_info[r]) \
1896 = EVER_MATCHED_SOMETHING (reg_info[r]) \
1902 # endif /* not DEFINED_ONCE */
1904 /* Registers are set to a sentinel when they haven't yet matched. */
1905 static CHAR_T PREFIX(reg_unset_dummy);
1906 # define REG_UNSET_VALUE (&PREFIX(reg_unset_dummy))
1907 # define REG_UNSET(e) ((e) == REG_UNSET_VALUE)
1909 /* Subroutine declarations and macros for regex_compile. */
1910 static void PREFIX(store_op1) _RE_ARGS ((re_opcode_t op, UCHAR_T *loc, int arg));
1911 static void PREFIX(store_op2) _RE_ARGS ((re_opcode_t op, UCHAR_T *loc,
1912 int arg1, int arg2));
1913 static void PREFIX(insert_op1) _RE_ARGS ((re_opcode_t op, UCHAR_T *loc,
1914 int arg, UCHAR_T *end));
1915 static void PREFIX(insert_op2) _RE_ARGS ((re_opcode_t op, UCHAR_T *loc,
1916 int arg1, int arg2, UCHAR_T *end));
1917 static boolean PREFIX(at_begline_loc_p) _RE_ARGS ((const CHAR_T *pattern,
1919 reg_syntax_t syntax));
1920 static boolean PREFIX(at_endline_loc_p) _RE_ARGS ((const CHAR_T *p,
1922 reg_syntax_t syntax));
1924 static reg_errcode_t wcs_compile_range _RE_ARGS ((CHAR_T range_start,
1925 const CHAR_T **p_ptr,
1928 reg_syntax_t syntax,
1931 static void insert_space _RE_ARGS ((int num, CHAR_T *loc, CHAR_T *end));
1933 static reg_errcode_t byte_compile_range _RE_ARGS ((unsigned int range_start,
1937 reg_syntax_t syntax,
1941 /* Fetch the next character in the uncompiled pattern---translating it
1942 if necessary. Also cast from a signed character in the constant
1943 string passed to us by the user to an unsigned char that we can use
1944 as an array index (in, e.g., `translate'). */
1945 /* ifdef MBS_SUPPORT, we translate only if character <= 0xff,
1946 because it is impossible to allocate 4GB array for some encodings
1947 which have 4 byte character_set like UCS4. */
1950 # define PATFETCH(c) \
1951 do {if (p == pend) return REG_EEND; \
1952 c = (UCHAR_T) *p++; \
1953 if (translate && (c <= 0xff)) c = (UCHAR_T) translate[c]; \
1956 # define PATFETCH(c) \
1957 do {if (p == pend) return REG_EEND; \
1958 c = (unsigned char) *p++; \
1959 if (translate) c = (unsigned char) translate[c]; \
1964 /* Fetch the next character in the uncompiled pattern, with no
1966 # define PATFETCH_RAW(c) \
1967 do {if (p == pend) return REG_EEND; \
1968 c = (UCHAR_T) *p++; \
1971 /* Go backwards one character in the pattern. */
1972 # define PATUNFETCH p--
1975 /* If `translate' is non-null, return translate[D], else just D. We
1976 cast the subscript to translate because some data is declared as
1977 `char *', to avoid warnings when a string constant is passed. But
1978 when we use a character as a subscript we must make it unsigned. */
1979 /* ifdef MBS_SUPPORT, we translate only if character <= 0xff,
1980 because it is impossible to allocate 4GB array for some encodings
1981 which have 4 byte character_set like UCS4. */
1985 # define TRANSLATE(d) \
1986 ((translate && ((UCHAR_T) (d)) <= 0xff) \
1987 ? (char) translate[(unsigned char) (d)] : (d))
1989 # define TRANSLATE(d) \
1990 (translate ? (char) translate[(unsigned char) (d)] : (d))
1995 /* Macros for outputting the compiled pattern into `buffer'. */
1997 /* If the buffer isn't allocated when it comes in, use this. */
1998 # define INIT_BUF_SIZE (32 * sizeof(UCHAR_T))
2000 /* Make sure we have at least N more bytes of space in buffer. */
2002 # define GET_BUFFER_SPACE(n) \
2003 while (((unsigned long)b - (unsigned long)COMPILED_BUFFER_VAR \
2004 + (n)*sizeof(CHAR_T)) > bufp->allocated) \
2007 # define GET_BUFFER_SPACE(n) \
2008 while ((unsigned long) (b - bufp->buffer + (n)) > bufp->allocated) \
2012 /* Make sure we have one more byte of buffer space and then add C to it. */
2013 # define BUF_PUSH(c) \
2015 GET_BUFFER_SPACE (1); \
2016 *b++ = (UCHAR_T) (c); \
2020 /* Ensure we have two more bytes of buffer space and then append C1 and C2. */
2021 # define BUF_PUSH_2(c1, c2) \
2023 GET_BUFFER_SPACE (2); \
2024 *b++ = (UCHAR_T) (c1); \
2025 *b++ = (UCHAR_T) (c2); \
2029 /* As with BUF_PUSH_2, except for three bytes. */
2030 # define BUF_PUSH_3(c1, c2, c3) \
2032 GET_BUFFER_SPACE (3); \
2033 *b++ = (UCHAR_T) (c1); \
2034 *b++ = (UCHAR_T) (c2); \
2035 *b++ = (UCHAR_T) (c3); \
2038 /* Store a jump with opcode OP at LOC to location TO. We store a
2039 relative address offset by the three bytes the jump itself occupies. */
2040 # define STORE_JUMP(op, loc, to) \
2041 PREFIX(store_op1) (op, loc, (int) ((to) - (loc) - (1 + OFFSET_ADDRESS_SIZE)))
2043 /* Likewise, for a two-argument jump. */
2044 # define STORE_JUMP2(op, loc, to, arg) \
2045 PREFIX(store_op2) (op, loc, (int) ((to) - (loc) - (1 + OFFSET_ADDRESS_SIZE)), arg)
2047 /* Like `STORE_JUMP', but for inserting. Assume `b' is the buffer end. */
2048 # define INSERT_JUMP(op, loc, to) \
2049 PREFIX(insert_op1) (op, loc, (int) ((to) - (loc) - (1 + OFFSET_ADDRESS_SIZE)), b)
2051 /* Like `STORE_JUMP2', but for inserting. Assume `b' is the buffer end. */
2052 # define INSERT_JUMP2(op, loc, to, arg) \
2053 PREFIX(insert_op2) (op, loc, (int) ((to) - (loc) - (1 + OFFSET_ADDRESS_SIZE)),\
2056 /* This is not an arbitrary limit: the arguments which represent offsets
2057 into the pattern are two bytes long. So if 2^16 bytes turns out to
2058 be too small, many things would have to change. */
2059 /* Any other compiler which, like MSC, has allocation limit below 2^16
2060 bytes will have to use approach similar to what was done below for
2061 MSC and drop MAX_BUF_SIZE a bit. Otherwise you may end up
2062 reallocating to 0 bytes. Such thing is not going to work too well.
2063 You have been warned!! */
2064 # ifndef DEFINED_ONCE
2065 # if defined _MSC_VER && !defined WIN32
2066 /* Microsoft C 16-bit versions limit malloc to approx 65512 bytes.
2067 The REALLOC define eliminates a flurry of conversion warnings,
2068 but is not required. */
2069 # define MAX_BUF_SIZE 65500L
2070 # define REALLOC(p,s) realloc ((p), (size_t) (s))
2072 # define MAX_BUF_SIZE (1L << 16)
2073 # define REALLOC(p,s) realloc ((p), (s))
2076 /* Extend the buffer by twice its current size via realloc and
2077 reset the pointers that pointed into the old block to point to the
2078 correct places in the new one. If extending the buffer results in it
2079 being larger than MAX_BUF_SIZE, then flag memory exhausted. */
2080 # if __BOUNDED_POINTERS__
2081 # define SET_HIGH_BOUND(P) (__ptrhigh (P) = __ptrlow (P) + bufp->allocated)
2082 # define MOVE_BUFFER_POINTER(P) \
2083 (__ptrlow (P) += incr, SET_HIGH_BOUND (P), __ptrvalue (P) += incr)
2084 # define ELSE_EXTEND_BUFFER_HIGH_BOUND \
2087 SET_HIGH_BOUND (b); \
2088 SET_HIGH_BOUND (begalt); \
2089 if (fixup_alt_jump) \
2090 SET_HIGH_BOUND (fixup_alt_jump); \
2092 SET_HIGH_BOUND (laststart); \
2093 if (pending_exact) \
2094 SET_HIGH_BOUND (pending_exact); \
2097 # define MOVE_BUFFER_POINTER(P) (P) += incr
2098 # define ELSE_EXTEND_BUFFER_HIGH_BOUND
2100 # endif /* not DEFINED_ONCE */
2103 # define EXTEND_BUFFER() \
2105 UCHAR_T *old_buffer = COMPILED_BUFFER_VAR; \
2107 if (bufp->allocated + sizeof(UCHAR_T) > MAX_BUF_SIZE) \
2109 bufp->allocated <<= 1; \
2110 if (bufp->allocated > MAX_BUF_SIZE) \
2111 bufp->allocated = MAX_BUF_SIZE; \
2112 /* How many characters the new buffer can have? */ \
2113 wchar_count = bufp->allocated / sizeof(UCHAR_T); \
2114 if (wchar_count == 0) wchar_count = 1; \
2115 /* Truncate the buffer to CHAR_T align. */ \
2116 bufp->allocated = wchar_count * sizeof(UCHAR_T); \
2117 RETALLOC (COMPILED_BUFFER_VAR, wchar_count, UCHAR_T); \
2118 bufp->buffer = (char*)COMPILED_BUFFER_VAR; \
2119 if (COMPILED_BUFFER_VAR == NULL) \
2120 return REG_ESPACE; \
2121 /* If the buffer moved, move all the pointers into it. */ \
2122 if (old_buffer != COMPILED_BUFFER_VAR) \
2124 int incr = COMPILED_BUFFER_VAR - old_buffer; \
2125 MOVE_BUFFER_POINTER (b); \
2126 MOVE_BUFFER_POINTER (begalt); \
2127 if (fixup_alt_jump) \
2128 MOVE_BUFFER_POINTER (fixup_alt_jump); \
2130 MOVE_BUFFER_POINTER (laststart); \
2131 if (pending_exact) \
2132 MOVE_BUFFER_POINTER (pending_exact); \
2134 ELSE_EXTEND_BUFFER_HIGH_BOUND \
2137 # define EXTEND_BUFFER() \
2139 UCHAR_T *old_buffer = COMPILED_BUFFER_VAR; \
2140 if (bufp->allocated == MAX_BUF_SIZE) \
2142 bufp->allocated <<= 1; \
2143 if (bufp->allocated > MAX_BUF_SIZE) \
2144 bufp->allocated = MAX_BUF_SIZE; \
2145 bufp->buffer = (UCHAR_T *) REALLOC (COMPILED_BUFFER_VAR, \
2147 if (COMPILED_BUFFER_VAR == NULL) \
2148 return REG_ESPACE; \
2149 /* If the buffer moved, move all the pointers into it. */ \
2150 if (old_buffer != COMPILED_BUFFER_VAR) \
2152 int incr = COMPILED_BUFFER_VAR - old_buffer; \
2153 MOVE_BUFFER_POINTER (b); \
2154 MOVE_BUFFER_POINTER (begalt); \
2155 if (fixup_alt_jump) \
2156 MOVE_BUFFER_POINTER (fixup_alt_jump); \
2158 MOVE_BUFFER_POINTER (laststart); \
2159 if (pending_exact) \
2160 MOVE_BUFFER_POINTER (pending_exact); \
2162 ELSE_EXTEND_BUFFER_HIGH_BOUND \
2166 # ifndef DEFINED_ONCE
2167 /* Since we have one byte reserved for the register number argument to
2168 {start,stop}_memory, the maximum number of groups we can report
2169 things about is what fits in that byte. */
2170 # define MAX_REGNUM 255
2172 /* But patterns can have more than `MAX_REGNUM' registers. We just
2173 ignore the excess. */
2174 typedef unsigned regnum_t;
2177 /* Macros for the compile stack. */
2179 /* Since offsets can go either forwards or backwards, this type needs to
2180 be able to hold values from -(MAX_BUF_SIZE - 1) to MAX_BUF_SIZE - 1. */
2181 /* int may be not enough when sizeof(int) == 2. */
2182 typedef long pattern_offset_t;
2186 pattern_offset_t begalt_offset;
2187 pattern_offset_t fixup_alt_jump;
2188 pattern_offset_t inner_group_offset;
2189 pattern_offset_t laststart_offset;
2191 } compile_stack_elt_t;
2196 compile_stack_elt_t *stack;
2198 unsigned avail; /* Offset of next open position. */
2199 } compile_stack_type;
2202 # define INIT_COMPILE_STACK_SIZE 32
2204 # define COMPILE_STACK_EMPTY (compile_stack.avail == 0)
2205 # define COMPILE_STACK_FULL (compile_stack.avail == compile_stack.size)
2207 /* The next available element. */
2208 # define COMPILE_STACK_TOP (compile_stack.stack[compile_stack.avail])
2210 # endif /* not DEFINED_ONCE */
2212 /* Set the bit for character C in a list. */
2213 # ifndef DEFINED_ONCE
2214 # define SET_LIST_BIT(c) \
2215 (b[((unsigned char) (c)) / BYTEWIDTH] \
2216 |= 1 << (((unsigned char) c) % BYTEWIDTH))
2217 # endif /* DEFINED_ONCE */
2219 /* Get the next unsigned number in the uncompiled pattern. */
2220 # define GET_UNSIGNED_NUMBER(num) \
2225 if (c < '0' || c > '9') \
2227 if (num <= RE_DUP_MAX) \
2231 num = num * 10 + c - '0'; \
2236 # ifndef DEFINED_ONCE
2237 # if defined _LIBC || WIDE_CHAR_SUPPORT
2238 /* The GNU C library provides support for user-defined character classes
2239 and the functions from ISO C amendement 1. */
2240 # ifdef CHARCLASS_NAME_MAX
2241 # define CHAR_CLASS_MAX_LENGTH CHARCLASS_NAME_MAX
2243 /* This shouldn't happen but some implementation might still have this
2244 problem. Use a reasonable default value. */
2245 # define CHAR_CLASS_MAX_LENGTH 256
2249 # define IS_CHAR_CLASS(string) __wctype (string)
2251 # define IS_CHAR_CLASS(string) wctype (string)
2254 # define CHAR_CLASS_MAX_LENGTH 6 /* Namely, `xdigit'. */
2256 # define IS_CHAR_CLASS(string) \
2257 (STREQ (string, "alpha") || STREQ (string, "upper") \
2258 || STREQ (string, "lower") || STREQ (string, "digit") \
2259 || STREQ (string, "alnum") || STREQ (string, "xdigit") \
2260 || STREQ (string, "space") || STREQ (string, "print") \
2261 || STREQ (string, "punct") || STREQ (string, "graph") \
2262 || STREQ (string, "cntrl") || STREQ (string, "blank"))
2264 # endif /* DEFINED_ONCE */
2266 # ifndef MATCH_MAY_ALLOCATE
2268 /* If we cannot allocate large objects within re_match_2_internal,
2269 we make the fail stack and register vectors global.
2270 The fail stack, we grow to the maximum size when a regexp
2272 The register vectors, we adjust in size each time we
2273 compile a regexp, according to the number of registers it needs. */
2275 static PREFIX(fail_stack_type) fail_stack;
2277 /* Size with which the following vectors are currently allocated.
2278 That is so we can make them bigger as needed,
2279 but never make them smaller. */
2280 # ifdef DEFINED_ONCE
2281 static int regs_allocated_size;
2283 static const char ** regstart, ** regend;
2284 static const char ** old_regstart, ** old_regend;
2285 static const char **best_regstart, **best_regend;
2286 static const char **reg_dummy;
2287 # endif /* DEFINED_ONCE */
2289 static PREFIX(register_info_type) *PREFIX(reg_info);
2290 static PREFIX(register_info_type) *PREFIX(reg_info_dummy);
2292 /* Make the register vectors big enough for NUM_REGS registers,
2293 but don't make them smaller. */
2296 PREFIX(regex_grow_registers) (num_regs)
2299 if (num_regs > regs_allocated_size)
2301 RETALLOC_IF (regstart, num_regs, const char *);
2302 RETALLOC_IF (regend, num_regs, const char *);
2303 RETALLOC_IF (old_regstart, num_regs, const char *);
2304 RETALLOC_IF (old_regend, num_regs, const char *);
2305 RETALLOC_IF (best_regstart, num_regs, const char *);
2306 RETALLOC_IF (best_regend, num_regs, const char *);
2307 RETALLOC_IF (PREFIX(reg_info), num_regs, PREFIX(register_info_type));
2308 RETALLOC_IF (reg_dummy, num_regs, const char *);
2309 RETALLOC_IF (PREFIX(reg_info_dummy), num_regs, PREFIX(register_info_type));
2311 regs_allocated_size = num_regs;
2315 # endif /* not MATCH_MAY_ALLOCATE */
2317 # ifndef DEFINED_ONCE
2318 static boolean group_in_compile_stack _RE_ARGS ((compile_stack_type
2321 # endif /* not DEFINED_ONCE */
2323 /* `regex_compile' compiles PATTERN (of length SIZE) according to SYNTAX.
2324 Returns one of error codes defined in `regex.h', or zero for success.
2326 Assumes the `allocated' (and perhaps `buffer') and `translate'
2327 fields are set in BUFP on entry.
2329 If it succeeds, results are put in BUFP (if it returns an error, the
2330 contents of BUFP are undefined):
2331 `buffer' is the compiled pattern;
2332 `syntax' is set to SYNTAX;
2333 `used' is set to the length of the compiled pattern;
2334 `fastmap_accurate' is zero;
2335 `re_nsub' is the number of subexpressions in PATTERN;
2336 `not_bol' and `not_eol' are zero;
2338 The `fastmap' and `newline_anchor' fields are neither
2339 examined nor set. */
2341 /* Return, freeing storage we allocated. */
2343 # define FREE_STACK_RETURN(value) \
2344 return (free(pattern), free(mbs_offset), free(is_binary), free (compile_stack.stack), value)
2346 # define FREE_STACK_RETURN(value) \
2347 return (free (compile_stack.stack), value)
2350 static reg_errcode_t
2351 PREFIX(regex_compile) (ARG_PREFIX(pattern), ARG_PREFIX(size), syntax, bufp)
2352 const char *ARG_PREFIX(pattern);
2353 size_t ARG_PREFIX(size);
2354 reg_syntax_t syntax;
2355 struct re_pattern_buffer *bufp;
2357 /* We fetch characters from PATTERN here. Even though PATTERN is
2358 `char *' (i.e., signed), we declare these variables as unsigned, so
2359 they can be reliably used as array indices. */
2360 register UCHAR_T c, c1;
2363 /* A temporary space to keep wchar_t pattern and compiled pattern. */
2364 CHAR_T *pattern, *COMPILED_BUFFER_VAR;
2366 /* offset buffer for optimization. See convert_mbs_to_wc. */
2367 int *mbs_offset = NULL;
2368 /* It hold whether each wchar_t is binary data or not. */
2369 char *is_binary = NULL;
2370 /* A flag whether exactn is handling binary data or not. */
2371 char is_exactn_bin = FALSE;
2374 /* A random temporary spot in PATTERN. */
2377 /* Points to the end of the buffer, where we should append. */
2378 register UCHAR_T *b;
2380 /* Keeps track of unclosed groups. */
2381 compile_stack_type compile_stack;
2383 /* Points to the current (ending) position in the pattern. */
2388 const CHAR_T *p = pattern;
2389 const CHAR_T *pend = pattern + size;
2392 /* How to translate the characters in the pattern. */
2393 RE_TRANSLATE_TYPE translate = bufp->translate;
2395 /* Address of the count-byte of the most recently inserted `exactn'
2396 command. This makes it possible to tell if a new exact-match
2397 character can be added to that command or if the character requires
2398 a new `exactn' command. */
2399 UCHAR_T *pending_exact = 0;
2401 /* Address of start of the most recently finished expression.
2402 This tells, e.g., postfix * where to find the start of its
2403 operand. Reset at the beginning of groups and alternatives. */
2404 UCHAR_T *laststart = 0;
2406 /* Address of beginning of regexp, or inside of last group. */
2409 /* Address of the place where a forward jump should go to the end of
2410 the containing expression. Each alternative of an `or' -- except the
2411 last -- ends with a forward jump of this sort. */
2412 UCHAR_T *fixup_alt_jump = 0;
2414 /* Counts open-groups as they are encountered. Remembered for the
2415 matching close-group on the compile stack, so the same register
2416 number is put in the stop_memory as the start_memory. */
2417 regnum_t regnum = 0;
2420 /* Initialize the wchar_t PATTERN and offset_buffer. */
2421 p = pend = pattern = TALLOC(csize + 1, CHAR_T);
2422 mbs_offset = TALLOC(csize + 1, int);
2423 is_binary = TALLOC(csize + 1, char);
2424 if (pattern == NULL || mbs_offset == NULL || is_binary == NULL)
2431 pattern[csize] = L'\0'; /* sentinel */
2432 size = convert_mbs_to_wcs(pattern, cpattern, csize, mbs_offset, is_binary);
2444 DEBUG_PRINT1 ("\nCompiling pattern: ");
2447 unsigned debug_count;
2449 for (debug_count = 0; debug_count < size; debug_count++)
2450 PUT_CHAR (pattern[debug_count]);
2455 /* Initialize the compile stack. */
2456 compile_stack.stack = TALLOC (INIT_COMPILE_STACK_SIZE, compile_stack_elt_t);
2457 if (compile_stack.stack == NULL)
2467 compile_stack.size = INIT_COMPILE_STACK_SIZE;
2468 compile_stack.avail = 0;
2470 /* Initialize the pattern buffer. */
2471 bufp->syntax = syntax;
2472 bufp->fastmap_accurate = 0;
2473 bufp->not_bol = bufp->not_eol = 0;
2475 /* Set `used' to zero, so that if we return an error, the pattern
2476 printer (for debugging) will think there's no pattern. We reset it
2480 /* Always count groups, whether or not bufp->no_sub is set. */
2483 #if !defined emacs && !defined SYNTAX_TABLE
2484 /* Initialize the syntax table. */
2485 init_syntax_once ();
2488 if (bufp->allocated == 0)
2491 { /* If zero allocated, but buffer is non-null, try to realloc
2492 enough space. This loses if buffer's address is bogus, but
2493 that is the user's responsibility. */
2495 /* Free bufp->buffer and allocate an array for wchar_t pattern
2498 COMPILED_BUFFER_VAR = TALLOC (INIT_BUF_SIZE/sizeof(UCHAR_T),
2501 RETALLOC (COMPILED_BUFFER_VAR, INIT_BUF_SIZE, UCHAR_T);
2505 { /* Caller did not allocate a buffer. Do it for them. */
2506 COMPILED_BUFFER_VAR = TALLOC (INIT_BUF_SIZE / sizeof(UCHAR_T),
2510 if (!COMPILED_BUFFER_VAR) FREE_STACK_RETURN (REG_ESPACE);
2512 bufp->buffer = (char*)COMPILED_BUFFER_VAR;
2514 bufp->allocated = INIT_BUF_SIZE;
2518 COMPILED_BUFFER_VAR = (UCHAR_T*) bufp->buffer;
2521 begalt = b = COMPILED_BUFFER_VAR;
2523 /* Loop through the uncompiled pattern until we're at the end. */
2532 if ( /* If at start of pattern, it's an operator. */
2534 /* If context independent, it's an operator. */
2535 || syntax & RE_CONTEXT_INDEP_ANCHORS
2536 /* Otherwise, depends on what's come before. */
2537 || PREFIX(at_begline_loc_p) (pattern, p, syntax))
2547 if ( /* If at end of pattern, it's an operator. */
2549 /* If context independent, it's an operator. */
2550 || syntax & RE_CONTEXT_INDEP_ANCHORS
2551 /* Otherwise, depends on what's next. */
2552 || PREFIX(at_endline_loc_p) (p, pend, syntax))
2562 if ((syntax & RE_BK_PLUS_QM)
2563 || (syntax & RE_LIMITED_OPS))
2567 /* If there is no previous pattern... */
2570 if (syntax & RE_CONTEXT_INVALID_OPS)
2571 FREE_STACK_RETURN (REG_BADRPT);
2572 else if (!(syntax & RE_CONTEXT_INDEP_OPS))
2577 /* Are we optimizing this jump? */
2578 boolean keep_string_p = false;
2580 /* 1 means zero (many) matches is allowed. */
2581 char zero_times_ok = 0, many_times_ok = 0;
2583 /* If there is a sequence of repetition chars, collapse it
2584 down to just one (the right one). We can't combine
2585 interval operators with these because of, e.g., `a{2}*',
2586 which should only match an even number of `a's. */
2590 zero_times_ok |= c != '+';
2591 many_times_ok |= c != '?';
2599 || (!(syntax & RE_BK_PLUS_QM) && (c == '+' || c == '?')))
2602 else if (syntax & RE_BK_PLUS_QM && c == '\\')
2604 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
2607 if (!(c1 == '+' || c1 == '?'))
2622 /* If we get here, we found another repeat character. */
2625 /* Star, etc. applied to an empty pattern is equivalent
2626 to an empty pattern. */
2630 /* Now we know whether or not zero matches is allowed
2631 and also whether or not two or more matches is allowed. */
2633 { /* More than one repetition is allowed, so put in at the
2634 end a backward relative jump from `b' to before the next
2635 jump we're going to put in below (which jumps from
2636 laststart to after this jump).
2638 But if we are at the `*' in the exact sequence `.*\n',
2639 insert an unconditional jump backwards to the .,
2640 instead of the beginning of the loop. This way we only
2641 push a failure point once, instead of every time
2642 through the loop. */
2643 assert (p - 1 > pattern);
2645 /* Allocate the space for the jump. */
2646 GET_BUFFER_SPACE (1 + OFFSET_ADDRESS_SIZE);
2648 /* We know we are not at the first character of the pattern,
2649 because laststart was nonzero. And we've already
2650 incremented `p', by the way, to be the character after
2651 the `*'. Do we have to do something analogous here
2652 for null bytes, because of RE_DOT_NOT_NULL? */
2653 if (TRANSLATE (*(p - 2)) == TRANSLATE ('.')
2655 && p < pend && TRANSLATE (*p) == TRANSLATE ('\n')
2656 && !(syntax & RE_DOT_NEWLINE))
2657 { /* We have .*\n. */
2658 STORE_JUMP (jump, b, laststart);
2659 keep_string_p = true;
2662 /* Anything else. */
2663 STORE_JUMP (maybe_pop_jump, b, laststart -
2664 (1 + OFFSET_ADDRESS_SIZE));
2666 /* We've added more stuff to the buffer. */
2667 b += 1 + OFFSET_ADDRESS_SIZE;
2670 /* On failure, jump from laststart to b + 3, which will be the
2671 end of the buffer after this jump is inserted. */
2672 /* ifdef WCHAR, 'b + 1 + OFFSET_ADDRESS_SIZE' instead of
2674 GET_BUFFER_SPACE (1 + OFFSET_ADDRESS_SIZE);
2675 INSERT_JUMP (keep_string_p ? on_failure_keep_string_jump
2677 laststart, b + 1 + OFFSET_ADDRESS_SIZE);
2679 b += 1 + OFFSET_ADDRESS_SIZE;
2683 /* At least one repetition is required, so insert a
2684 `dummy_failure_jump' before the initial
2685 `on_failure_jump' instruction of the loop. This
2686 effects a skip over that instruction the first time
2687 we hit that loop. */
2688 GET_BUFFER_SPACE (1 + OFFSET_ADDRESS_SIZE);
2689 INSERT_JUMP (dummy_failure_jump, laststart, laststart +
2690 2 + 2 * OFFSET_ADDRESS_SIZE);
2691 b += 1 + OFFSET_ADDRESS_SIZE;
2705 boolean had_char_class = false;
2707 CHAR_T range_start = 0xffffffff;
2709 unsigned int range_start = 0xffffffff;
2711 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2714 /* We assume a charset(_not) structure as a wchar_t array.
2715 charset[0] = (re_opcode_t) charset(_not)
2716 charset[1] = l (= length of char_classes)
2717 charset[2] = m (= length of collating_symbols)
2718 charset[3] = n (= length of equivalence_classes)
2719 charset[4] = o (= length of char_ranges)
2720 charset[5] = p (= length of chars)
2722 charset[6] = char_class (wctype_t)
2723 charset[6+CHAR_CLASS_SIZE] = char_class (wctype_t)
2725 charset[l+5] = char_class (wctype_t)
2727 charset[l+6] = collating_symbol (wchar_t)
2729 charset[l+m+5] = collating_symbol (wchar_t)
2730 ifdef _LIBC we use the index if
2731 _NL_COLLATE_SYMB_EXTRAMB instead of
2734 charset[l+m+6] = equivalence_classes (wchar_t)
2736 charset[l+m+n+5] = equivalence_classes (wchar_t)
2737 ifdef _LIBC we use the index in
2738 _NL_COLLATE_WEIGHT instead of
2741 charset[l+m+n+6] = range_start
2742 charset[l+m+n+7] = range_end
2744 charset[l+m+n+2o+4] = range_start
2745 charset[l+m+n+2o+5] = range_end
2746 ifdef _LIBC we use the value looked up
2747 in _NL_COLLATE_COLLSEQ instead of
2750 charset[l+m+n+2o+6] = char
2752 charset[l+m+n+2o+p+5] = char
2756 /* We need at least 6 spaces: the opcode, the length of
2757 char_classes, the length of collating_symbols, the length of
2758 equivalence_classes, the length of char_ranges, the length of
2760 GET_BUFFER_SPACE (6);
2762 /* Save b as laststart. And We use laststart as the pointer
2763 to the first element of the charset here.
2764 In other words, laststart[i] indicates charset[i]. */
2767 /* We test `*p == '^' twice, instead of using an if
2768 statement, so we only need one BUF_PUSH. */
2769 BUF_PUSH (*p == '^' ? charset_not : charset);
2773 /* Push the length of char_classes, the length of
2774 collating_symbols, the length of equivalence_classes, the
2775 length of char_ranges and the length of chars. */
2776 BUF_PUSH_3 (0, 0, 0);
2779 /* Remember the first position in the bracket expression. */
2782 /* charset_not matches newline according to a syntax bit. */
2783 if ((re_opcode_t) b[-6] == charset_not
2784 && (syntax & RE_HAT_LISTS_NOT_NEWLINE))
2787 laststart[5]++; /* Update the length of characters */
2790 /* Read in characters and ranges, setting map bits. */
2793 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2797 /* \ might escape characters inside [...] and [^...]. */
2798 if ((syntax & RE_BACKSLASH_ESCAPE_IN_LISTS) && c == '\\')
2800 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
2804 laststart[5]++; /* Update the length of chars */
2809 /* Could be the end of the bracket expression. If it's
2810 not (i.e., when the bracket expression is `[]' so
2811 far), the ']' character bit gets set way below. */
2812 if (c == ']' && p != p1 + 1)
2815 /* Look ahead to see if it's a range when the last thing
2816 was a character class. */
2817 if (had_char_class && c == '-' && *p != ']')
2818 FREE_STACK_RETURN (REG_ERANGE);
2820 /* Look ahead to see if it's a range when the last thing
2821 was a character: if this is a hyphen not at the
2822 beginning or the end of a list, then it's the range
2825 && !(p - 2 >= pattern && p[-2] == '[')
2826 && !(p - 3 >= pattern && p[-3] == '[' && p[-2] == '^')
2830 /* Allocate the space for range_start and range_end. */
2831 GET_BUFFER_SPACE (2);
2832 /* Update the pointer to indicate end of buffer. */
2834 ret = wcs_compile_range (range_start, &p, pend, translate,
2835 syntax, b, laststart);
2836 if (ret != REG_NOERROR) FREE_STACK_RETURN (ret);
2837 range_start = 0xffffffff;
2839 else if (p[0] == '-' && p[1] != ']')
2840 { /* This handles ranges made up of characters only. */
2843 /* Move past the `-'. */
2845 /* Allocate the space for range_start and range_end. */
2846 GET_BUFFER_SPACE (2);
2847 /* Update the pointer to indicate end of buffer. */
2849 ret = wcs_compile_range (c, &p, pend, translate, syntax, b,
2851 if (ret != REG_NOERROR) FREE_STACK_RETURN (ret);
2852 range_start = 0xffffffff;
2855 /* See if we're at the beginning of a possible character
2857 else if (syntax & RE_CHAR_CLASSES && c == '[' && *p == ':')
2858 { /* Leave room for the null. */
2859 char str[CHAR_CLASS_MAX_LENGTH + 1];
2864 /* If pattern is `[[:'. */
2865 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2870 if ((c == ':' && *p == ']') || p == pend)
2872 if (c1 < CHAR_CLASS_MAX_LENGTH)
2875 /* This is in any case an invalid class name. */
2880 /* If isn't a word bracketed by `[:' and `:]':
2881 undo the ending character, the letters, and leave
2882 the leading `:' and `[' (but store them as character). */
2883 if (c == ':' && *p == ']')
2888 /* Query the character class as wctype_t. */
2889 wt = IS_CHAR_CLASS (str);
2891 FREE_STACK_RETURN (REG_ECTYPE);
2893 /* Throw away the ] at the end of the character
2897 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2899 /* Allocate the space for character class. */
2900 GET_BUFFER_SPACE(CHAR_CLASS_SIZE);
2901 /* Update the pointer to indicate end of buffer. */
2902 b += CHAR_CLASS_SIZE;
2903 /* Move data which follow character classes
2904 not to violate the data. */
2905 insert_space(CHAR_CLASS_SIZE,
2906 laststart + 6 + laststart[1],
2908 alignedp = ((uintptr_t)(laststart + 6 + laststart[1])
2909 + __alignof__(wctype_t) - 1)
2910 & ~(uintptr_t)(__alignof__(wctype_t) - 1);
2911 /* Store the character class. */
2912 *((wctype_t*)alignedp) = wt;
2913 /* Update length of char_classes */
2914 laststart[1] += CHAR_CLASS_SIZE;
2916 had_char_class = true;
2925 laststart[5] += 2; /* Update the length of characters */
2927 had_char_class = false;
2930 else if (syntax & RE_CHAR_CLASSES && c == '[' && (*p == '='
2933 CHAR_T str[128]; /* Should be large enough. */
2934 CHAR_T delim = *p; /* '=' or '.' */
2937 _NL_CURRENT_WORD (LC_COLLATE, _NL_COLLATE_NRULES);
2942 /* If pattern is `[[=' or '[[.'. */
2943 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2948 if ((c == delim && *p == ']') || p == pend)
2950 if (c1 < sizeof (str) - 1)
2953 /* This is in any case an invalid class name. */
2958 if (c == delim && *p == ']' && str[0] != '\0')
2960 unsigned int i, offset;
2961 /* If we have no collation data we use the default
2962 collation in which each character is in a class
2963 by itself. It also means that ASCII is the
2964 character set and therefore we cannot have character
2965 with more than one byte in the multibyte
2968 /* If not defined _LIBC, we push the name and
2969 `\0' for the sake of matching performance. */
2970 int datasize = c1 + 1;
2978 FREE_STACK_RETURN (REG_ECOLLATE);
2983 const int32_t *table;
2984 const int32_t *weights;
2985 const int32_t *extra;
2986 const int32_t *indirect;
2989 /* This #include defines a local function! */
2990 # include <locale/weightwc.h>
2994 /* We push the index for equivalence class. */
2997 table = (const int32_t *)
2998 _NL_CURRENT (LC_COLLATE,
2999 _NL_COLLATE_TABLEWC);
3000 weights = (const int32_t *)
3001 _NL_CURRENT (LC_COLLATE,
3002 _NL_COLLATE_WEIGHTWC);
3003 extra = (const int32_t *)
3004 _NL_CURRENT (LC_COLLATE,
3005 _NL_COLLATE_EXTRAWC);
3006 indirect = (const int32_t *)
3007 _NL_CURRENT (LC_COLLATE,
3008 _NL_COLLATE_INDIRECTWC);
3010 idx = findidx ((const wint_t**)&cp);
3011 if (idx == 0 || cp < (wint_t*) str + c1)
3012 /* This is no valid character. */
3013 FREE_STACK_RETURN (REG_ECOLLATE);
3015 str[0] = (wchar_t)idx;
3017 else /* delim == '.' */
3019 /* We push collation sequence value
3020 for collating symbol. */
3022 const int32_t *symb_table;
3023 const unsigned char *extra;
3030 /* We have to convert the name to a single-byte
3031 string. This is possible since the names
3032 consist of ASCII characters and the internal
3033 representation is UCS4. */
3034 for (i = 0; i < c1; ++i)
3035 char_str[i] = str[i];
3038 _NL_CURRENT_WORD (LC_COLLATE,
3039 _NL_COLLATE_SYMB_HASH_SIZEMB);
3040 symb_table = (const int32_t *)
3041 _NL_CURRENT (LC_COLLATE,
3042 _NL_COLLATE_SYMB_TABLEMB);
3043 extra = (const unsigned char *)
3044 _NL_CURRENT (LC_COLLATE,
3045 _NL_COLLATE_SYMB_EXTRAMB);
3047 /* Locate the character in the hashing table. */
3048 hash = elem_hash (char_str, c1);
3051 elem = hash % table_size;
3052 second = hash % (table_size - 2);
3053 while (symb_table[2 * elem] != 0)
3055 /* First compare the hashing value. */
3056 if (symb_table[2 * elem] == hash
3057 && c1 == extra[symb_table[2 * elem + 1]]
3058 && memcmp (char_str,
3059 &extra[symb_table[2 * elem + 1]
3062 /* Yep, this is the entry. */
3063 idx = symb_table[2 * elem + 1];
3064 idx += 1 + extra[idx];
3072 if (symb_table[2 * elem] != 0)
3074 /* Compute the index of the byte sequence
3076 idx += 1 + extra[idx];
3077 /* Adjust for the alignment. */
3078 idx = (idx + 3) & ~3;
3080 str[0] = (wchar_t) idx + 4;
3082 else if (symb_table[2 * elem] == 0 && c1 == 1)
3084 /* No valid character. Match it as a
3085 single byte character. */
3086 had_char_class = false;
3088 /* Update the length of characters */
3090 range_start = str[0];
3092 /* Throw away the ] at the end of the
3093 collating symbol. */
3095 /* exit from the switch block. */
3099 FREE_STACK_RETURN (REG_ECOLLATE);
3104 /* Throw away the ] at the end of the equivalence
3105 class (or collating symbol). */
3108 /* Allocate the space for the equivalence class
3109 (or collating symbol) (and '\0' if needed). */
3110 GET_BUFFER_SPACE(datasize);
3111 /* Update the pointer to indicate end of buffer. */
3115 { /* equivalence class */
3116 /* Calculate the offset of char_ranges,
3117 which is next to equivalence_classes. */
3118 offset = laststart[1] + laststart[2]
3121 insert_space(datasize, laststart + offset, b - 1);
3123 /* Write the equivalence_class and \0. */
3124 for (i = 0 ; i < datasize ; i++)
3125 laststart[offset + i] = str[i];
3127 /* Update the length of equivalence_classes. */
3128 laststart[3] += datasize;
3129 had_char_class = true;
3131 else /* delim == '.' */
3132 { /* collating symbol */
3133 /* Calculate the offset of the equivalence_classes,
3134 which is next to collating_symbols. */
3135 offset = laststart[1] + laststart[2] + 6;
3136 /* Insert space and write the collationg_symbol
3138 insert_space(datasize, laststart + offset, b-1);
3139 for (i = 0 ; i < datasize ; i++)
3140 laststart[offset + i] = str[i];
3142 /* In re_match_2_internal if range_start < -1, we
3143 assume -range_start is the offset of the
3144 collating symbol which is specified as
3145 the character of the range start. So we assign
3146 -(laststart[1] + laststart[2] + 6) to
3148 range_start = -(laststart[1] + laststart[2] + 6);
3149 /* Update the length of collating_symbol. */
3150 laststart[2] += datasize;
3151 had_char_class = false;
3161 laststart[5] += 2; /* Update the length of characters */
3162 range_start = delim;
3163 had_char_class = false;
3168 had_char_class = false;
3170 laststart[5]++; /* Update the length of characters */
3176 /* Ensure that we have enough space to push a charset: the
3177 opcode, the length count, and the bitset; 34 bytes in all. */
3178 GET_BUFFER_SPACE (34);
3182 /* We test `*p == '^' twice, instead of using an if
3183 statement, so we only need one BUF_PUSH. */
3184 BUF_PUSH (*p == '^' ? charset_not : charset);
3188 /* Remember the first position in the bracket expression. */
3191 /* Push the number of bytes in the bitmap. */
3192 BUF_PUSH ((1 << BYTEWIDTH) / BYTEWIDTH);
3194 /* Clear the whole map. */
3195 bzero (b, (1 << BYTEWIDTH) / BYTEWIDTH);
3197 /* charset_not matches newline according to a syntax bit. */
3198 if ((re_opcode_t) b[-2] == charset_not
3199 && (syntax & RE_HAT_LISTS_NOT_NEWLINE))
3200 SET_LIST_BIT ('\n');
3202 /* Read in characters and ranges, setting map bits. */
3205 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
3209 /* \ might escape characters inside [...] and [^...]. */
3210 if ((syntax & RE_BACKSLASH_ESCAPE_IN_LISTS) && c == '\\')
3212 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
3220 /* Could be the end of the bracket expression. If it's
3221 not (i.e., when the bracket expression is `[]' so
3222 far), the ']' character bit gets set way below. */
3223 if (c == ']' && p != p1 + 1)
3226 /* Look ahead to see if it's a range when the last thing
3227 was a character class. */
3228 if (had_char_class && c == '-' && *p != ']')
3229 FREE_STACK_RETURN (REG_ERANGE);
3231 /* Look ahead to see if it's a range when the last thing
3232 was a character: if this is a hyphen not at the
3233 beginning or the end of a list, then it's the range
3236 && !(p - 2 >= pattern && p[-2] == '[')
3237 && !(p - 3 >= pattern && p[-3] == '[' && p[-2] == '^')
3241 = byte_compile_range (range_start, &p, pend, translate,
3243 if (ret != REG_NOERROR) FREE_STACK_RETURN (ret);
3244 range_start = 0xffffffff;
3247 else if (p[0] == '-' && p[1] != ']')
3248 { /* This handles ranges made up of characters only. */
3251 /* Move past the `-'. */
3254 ret = byte_compile_range (c, &p, pend, translate, syntax, b);
3255 if (ret != REG_NOERROR) FREE_STACK_RETURN (ret);
3256 range_start = 0xffffffff;
3259 /* See if we're at the beginning of a possible character
3262 else if (syntax & RE_CHAR_CLASSES && c == '[' && *p == ':')
3263 { /* Leave room for the null. */
3264 char str[CHAR_CLASS_MAX_LENGTH + 1];
3269 /* If pattern is `[[:'. */
3270 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
3275 if ((c == ':' && *p == ']') || p == pend)
3277 if (c1 < CHAR_CLASS_MAX_LENGTH)
3280 /* This is in any case an invalid class name. */
3285 /* If isn't a word bracketed by `[:' and `:]':
3286 undo the ending character, the letters, and leave
3287 the leading `:' and `[' (but set bits for them). */
3288 if (c == ':' && *p == ']')
3290 # if defined _LIBC || WIDE_CHAR_SUPPORT
3291 boolean is_lower = STREQ (str, "lower");
3292 boolean is_upper = STREQ (str, "upper");
3296 wt = IS_CHAR_CLASS (str);
3298 FREE_STACK_RETURN (REG_ECTYPE);
3300 /* Throw away the ] at the end of the character
3304 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
3306 for (ch = 0; ch < 1 << BYTEWIDTH; ++ch)
3309 if (__iswctype (__btowc (ch), wt))
3312 if (iswctype (btowc (ch), wt))
3316 if (translate && (is_upper || is_lower)
3317 && (ISUPPER (ch) || ISLOWER (ch)))
3321 had_char_class = true;
3324 boolean is_alnum = STREQ (str, "alnum");
3325 boolean is_alpha = STREQ (str, "alpha");
3326 boolean is_blank = STREQ (str, "blank");
3327 boolean is_cntrl = STREQ (str, "cntrl");
3328 boolean is_digit = STREQ (str, "digit");
3329 boolean is_graph = STREQ (str, "graph");
3330 boolean is_lower = STREQ (str, "lower");
3331 boolean is_print = STREQ (str, "print");
3332 boolean is_punct = STREQ (str, "punct");
3333 boolean is_space = STREQ (str, "space");
3334 boolean is_upper = STREQ (str, "upper");
3335 boolean is_xdigit = STREQ (str, "xdigit");
3337 if (!IS_CHAR_CLASS (str))
3338 FREE_STACK_RETURN (REG_ECTYPE);
3340 /* Throw away the ] at the end of the character
3344 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
3346 for (ch = 0; ch < 1 << BYTEWIDTH; ch++)
3348 /* This was split into 3 if's to
3349 avoid an arbitrary limit in some compiler. */
3350 if ( (is_alnum && ISALNUM (ch))
3351 || (is_alpha && ISALPHA (ch))
3352 || (is_blank && ISBLANK (ch))
3353 || (is_cntrl && ISCNTRL (ch)))
3355 if ( (is_digit && ISDIGIT (ch))
3356 || (is_graph && ISGRAPH (ch))
3357 || (is_lower && ISLOWER (ch))
3358 || (is_print && ISPRINT (ch)))
3360 if ( (is_punct && ISPUNCT (ch))
3361 || (is_space && ISSPACE (ch))
3362 || (is_upper && ISUPPER (ch))
3363 || (is_xdigit && ISXDIGIT (ch)))
3365 if ( translate && (is_upper || is_lower)
3366 && (ISUPPER (ch) || ISLOWER (ch)))
3369 had_char_class = true;
3370 # endif /* libc || wctype.h */
3380 had_char_class = false;
3383 else if (syntax & RE_CHAR_CLASSES && c == '[' && *p == '=')
3385 unsigned char str[MB_LEN_MAX + 1];
3388 _NL_CURRENT_WORD (LC_COLLATE, _NL_COLLATE_NRULES);
3394 /* If pattern is `[[='. */
3395 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
3400 if ((c == '=' && *p == ']') || p == pend)
3402 if (c1 < MB_LEN_MAX)
3405 /* This is in any case an invalid class name. */
3410 if (c == '=' && *p == ']' && str[0] != '\0')
3412 /* If we have no collation data we use the default
3413 collation in which each character is in a class
3414 by itself. It also means that ASCII is the
3415 character set and therefore we cannot have character
3416 with more than one byte in the multibyte
3423 FREE_STACK_RETURN (REG_ECOLLATE);
3425 /* Throw away the ] at the end of the equivalence
3429 /* Set the bit for the character. */
3430 SET_LIST_BIT (str[0]);
3435 /* Try to match the byte sequence in `str' against
3436 those known to the collate implementation.
3437 First find out whether the bytes in `str' are
3438 actually from exactly one character. */
3439 const int32_t *table;
3440 const unsigned char *weights;
3441 const unsigned char *extra;
3442 const int32_t *indirect;
3444 const unsigned char *cp = str;
3447 /* This #include defines a local function! */
3448 # include <locale/weight.h>
3450 table = (const int32_t *)
3451 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_TABLEMB);
3452 weights = (const unsigned char *)
3453 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_WEIGHTMB);
3454 extra = (const unsigned char *)
3455 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_EXTRAMB);
3456 indirect = (const int32_t *)
3457 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_INDIRECTMB);
3459 idx = findidx (&cp);
3460 if (idx == 0 || cp < str + c1)
3461 /* This is no valid character. */
3462 FREE_STACK_RETURN (REG_ECOLLATE);
3464 /* Throw away the ] at the end of the equivalence
3468 /* Now we have to go throught the whole table
3469 and find all characters which have the same
3472 XXX Note that this is not entirely correct.
3473 we would have to match multibyte sequences
3474 but this is not possible with the current
3476 for (ch = 1; ch < 256; ++ch)
3477 /* XXX This test would have to be changed if we
3478 would allow matching multibyte sequences. */
3481 int32_t idx2 = table[ch];
3482 size_t len = weights[idx2];
3484 /* Test whether the lenghts match. */
3485 if (weights[idx] == len)
3487 /* They do. New compare the bytes of
3492 && (weights[idx + 1 + cnt]
3493 == weights[idx2 + 1 + cnt]))
3497 /* They match. Mark the character as
3504 had_char_class = true;
3514 had_char_class = false;
3517 else if (syntax & RE_CHAR_CLASSES && c == '[' && *p == '.')
3519 unsigned char str[128]; /* Should be large enough. */
3522 _NL_CURRENT_WORD (LC_COLLATE, _NL_COLLATE_NRULES);
3528 /* If pattern is `[[.'. */
3529 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
3534 if ((c == '.' && *p == ']') || p == pend)
3536 if (c1 < sizeof (str))
3539 /* This is in any case an invalid class name. */
3544 if (c == '.' && *p == ']' && str[0] != '\0')
3546 /* If we have no collation data we use the default
3547 collation in which each character is the name
3548 for its own class which contains only the one
3549 character. It also means that ASCII is the
3550 character set and therefore we cannot have character
3551 with more than one byte in the multibyte
3558 FREE_STACK_RETURN (REG_ECOLLATE);
3560 /* Throw away the ] at the end of the equivalence
3564 /* Set the bit for the character. */
3565 SET_LIST_BIT (str[0]);
3566 range_start = ((const unsigned char *) str)[0];
3571 /* Try to match the byte sequence in `str' against
3572 those known to the collate implementation.
3573 First find out whether the bytes in `str' are
3574 actually from exactly one character. */
3576 const int32_t *symb_table;
3577 const unsigned char *extra;
3584 _NL_CURRENT_WORD (LC_COLLATE,
3585 _NL_COLLATE_SYMB_HASH_SIZEMB);
3586 symb_table = (const int32_t *)
3587 _NL_CURRENT (LC_COLLATE,
3588 _NL_COLLATE_SYMB_TABLEMB);
3589 extra = (const unsigned char *)
3590 _NL_CURRENT (LC_COLLATE,
3591 _NL_COLLATE_SYMB_EXTRAMB);
3593 /* Locate the character in the hashing table. */
3594 hash = elem_hash (str, c1);
3597 elem = hash % table_size;
3598 second = hash % (table_size - 2);
3599 while (symb_table[2 * elem] != 0)
3601 /* First compare the hashing value. */
3602 if (symb_table[2 * elem] == hash
3603 && c1 == extra[symb_table[2 * elem + 1]]
3605 &extra[symb_table[2 * elem + 1]
3609 /* Yep, this is the entry. */
3610 idx = symb_table[2 * elem + 1];
3611 idx += 1 + extra[idx];
3619 if (symb_table[2 * elem] == 0)
3620 /* This is no valid character. */
3621 FREE_STACK_RETURN (REG_ECOLLATE);
3623 /* Throw away the ] at the end of the equivalence
3627 /* Now add the multibyte character(s) we found
3630 XXX Note that this is not entirely correct.
3631 we would have to match multibyte sequences
3632 but this is not possible with the current
3633 implementation. Also, we have to match
3634 collating symbols, which expand to more than
3635 one file, as a whole and not allow the
3636 individual bytes. */
3639 range_start = extra[idx];
3642 SET_LIST_BIT (extra[idx]);
3647 had_char_class = false;
3657 had_char_class = false;
3662 had_char_class = false;
3668 /* Discard any (non)matching list bytes that are all 0 at the
3669 end of the map. Decrease the map-length byte too. */
3670 while ((int) b[-1] > 0 && b[b[-1] - 1] == 0)
3679 if (syntax & RE_NO_BK_PARENS)
3686 if (syntax & RE_NO_BK_PARENS)
3693 if (syntax & RE_NEWLINE_ALT)
3700 if (syntax & RE_NO_BK_VBAR)
3707 if (syntax & RE_INTERVALS && syntax & RE_NO_BK_BRACES)
3708 goto handle_interval;
3714 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
3716 /* Do not translate the character after the \, so that we can
3717 distinguish, e.g., \B from \b, even if we normally would
3718 translate, e.g., B to b. */
3724 if (syntax & RE_NO_BK_PARENS)
3725 goto normal_backslash;
3731 if (COMPILE_STACK_FULL)
3733 RETALLOC (compile_stack.stack, compile_stack.size << 1,
3734 compile_stack_elt_t);
3735 if (compile_stack.stack == NULL) return REG_ESPACE;
3737 compile_stack.size <<= 1;
3740 /* These are the values to restore when we hit end of this
3741 group. They are all relative offsets, so that if the
3742 whole pattern moves because of realloc, they will still
3744 COMPILE_STACK_TOP.begalt_offset = begalt - COMPILED_BUFFER_VAR;
3745 COMPILE_STACK_TOP.fixup_alt_jump
3746 = fixup_alt_jump ? fixup_alt_jump - COMPILED_BUFFER_VAR + 1 : 0;
3747 COMPILE_STACK_TOP.laststart_offset = b - COMPILED_BUFFER_VAR;
3748 COMPILE_STACK_TOP.regnum = regnum;
3750 /* We will eventually replace the 0 with the number of
3751 groups inner to this one. But do not push a
3752 start_memory for groups beyond the last one we can
3753 represent in the compiled pattern. */
3754 if (regnum <= MAX_REGNUM)
3756 COMPILE_STACK_TOP.inner_group_offset = b
3757 - COMPILED_BUFFER_VAR + 2;
3758 BUF_PUSH_3 (start_memory, regnum, 0);
3761 compile_stack.avail++;
3766 /* If we've reached MAX_REGNUM groups, then this open
3767 won't actually generate any code, so we'll have to
3768 clear pending_exact explicitly. */
3774 if (syntax & RE_NO_BK_PARENS) goto normal_backslash;
3776 if (COMPILE_STACK_EMPTY)
3778 if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD)
3779 goto normal_backslash;
3781 FREE_STACK_RETURN (REG_ERPAREN);
3786 { /* Push a dummy failure point at the end of the
3787 alternative for a possible future
3788 `pop_failure_jump' to pop. See comments at
3789 `push_dummy_failure' in `re_match_2'. */
3790 BUF_PUSH (push_dummy_failure);
3792 /* We allocated space for this jump when we assigned
3793 to `fixup_alt_jump', in the `handle_alt' case below. */
3794 STORE_JUMP (jump_past_alt, fixup_alt_jump, b - 1);
3797 /* See similar code for backslashed left paren above. */
3798 if (COMPILE_STACK_EMPTY)
3800 if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD)
3803 FREE_STACK_RETURN (REG_ERPAREN);
3806 /* Since we just checked for an empty stack above, this
3807 ``can't happen''. */
3808 assert (compile_stack.avail != 0);
3810 /* We don't just want to restore into `regnum', because
3811 later groups should continue to be numbered higher,
3812 as in `(ab)c(de)' -- the second group is #2. */
3813 regnum_t this_group_regnum;
3815 compile_stack.avail--;
3816 begalt = COMPILED_BUFFER_VAR + COMPILE_STACK_TOP.begalt_offset;
3818 = COMPILE_STACK_TOP.fixup_alt_jump
3819 ? COMPILED_BUFFER_VAR + COMPILE_STACK_TOP.fixup_alt_jump - 1
3821 laststart = COMPILED_BUFFER_VAR + COMPILE_STACK_TOP.laststart_offset;
3822 this_group_regnum = COMPILE_STACK_TOP.regnum;
3823 /* If we've reached MAX_REGNUM groups, then this open
3824 won't actually generate any code, so we'll have to
3825 clear pending_exact explicitly. */
3828 /* We're at the end of the group, so now we know how many
3829 groups were inside this one. */
3830 if (this_group_regnum <= MAX_REGNUM)
3832 UCHAR_T *inner_group_loc
3833 = COMPILED_BUFFER_VAR + COMPILE_STACK_TOP.inner_group_offset;
3835 *inner_group_loc = regnum - this_group_regnum;
3836 BUF_PUSH_3 (stop_memory, this_group_regnum,
3837 regnum - this_group_regnum);
3843 case '|': /* `\|'. */
3844 if (syntax & RE_LIMITED_OPS || syntax & RE_NO_BK_VBAR)
3845 goto normal_backslash;
3847 if (syntax & RE_LIMITED_OPS)
3850 /* Insert before the previous alternative a jump which
3851 jumps to this alternative if the former fails. */
3852 GET_BUFFER_SPACE (1 + OFFSET_ADDRESS_SIZE);
3853 INSERT_JUMP (on_failure_jump, begalt,
3854 b + 2 + 2 * OFFSET_ADDRESS_SIZE);
3856 b += 1 + OFFSET_ADDRESS_SIZE;
3858 /* The alternative before this one has a jump after it
3859 which gets executed if it gets matched. Adjust that
3860 jump so it will jump to this alternative's analogous
3861 jump (put in below, which in turn will jump to the next
3862 (if any) alternative's such jump, etc.). The last such
3863 jump jumps to the correct final destination. A picture:
3869 If we are at `b', then fixup_alt_jump right now points to a
3870 three-byte space after `a'. We'll put in the jump, set
3871 fixup_alt_jump to right after `b', and leave behind three
3872 bytes which we'll fill in when we get to after `c'. */
3875 STORE_JUMP (jump_past_alt, fixup_alt_jump, b);
3877 /* Mark and leave space for a jump after this alternative,
3878 to be filled in later either by next alternative or
3879 when know we're at the end of a series of alternatives. */
3881 GET_BUFFER_SPACE (1 + OFFSET_ADDRESS_SIZE);
3882 b += 1 + OFFSET_ADDRESS_SIZE;
3890 /* If \{ is a literal. */
3891 if (!(syntax & RE_INTERVALS)
3892 /* If we're at `\{' and it's not the open-interval
3894 || (syntax & RE_NO_BK_BRACES))
3895 goto normal_backslash;
3899 /* If got here, then the syntax allows intervals. */
3901 /* At least (most) this many matches must be made. */
3902 int lower_bound = -1, upper_bound = -1;
3904 /* Place in the uncompiled pattern (i.e., just after
3905 the '{') to go back to if the interval is invalid. */
3906 const CHAR_T *beg_interval = p;
3909 goto invalid_interval;
3911 GET_UNSIGNED_NUMBER (lower_bound);
3915 GET_UNSIGNED_NUMBER (upper_bound);
3916 if (upper_bound < 0)
3917 upper_bound = RE_DUP_MAX;
3920 /* Interval such as `{1}' => match exactly once. */
3921 upper_bound = lower_bound;
3923 if (! (0 <= lower_bound && lower_bound <= upper_bound))
3924 goto invalid_interval;
3926 if (!(syntax & RE_NO_BK_BRACES))
3928 if (c != '\\' || p == pend)
3929 goto invalid_interval;
3934 goto invalid_interval;
3936 /* If it's invalid to have no preceding re. */
3939 if (syntax & RE_CONTEXT_INVALID_OPS
3940 && !(syntax & RE_INVALID_INTERVAL_ORD))
3941 FREE_STACK_RETURN (REG_BADRPT);
3942 else if (syntax & RE_CONTEXT_INDEP_OPS)
3945 goto unfetch_interval;
3948 /* We just parsed a valid interval. */
3950 if (RE_DUP_MAX < upper_bound)
3951 FREE_STACK_RETURN (REG_BADBR);
3953 /* If the upper bound is zero, don't want to succeed at
3954 all; jump from `laststart' to `b + 3', which will be
3955 the end of the buffer after we insert the jump. */
3956 /* ifdef WCHAR, 'b + 1 + OFFSET_ADDRESS_SIZE'
3957 instead of 'b + 3'. */
3958 if (upper_bound == 0)
3960 GET_BUFFER_SPACE (1 + OFFSET_ADDRESS_SIZE);
3961 INSERT_JUMP (jump, laststart, b + 1
3962 + OFFSET_ADDRESS_SIZE);
3963 b += 1 + OFFSET_ADDRESS_SIZE;
3966 /* Otherwise, we have a nontrivial interval. When
3967 we're all done, the pattern will look like:
3968 set_number_at <jump count> <upper bound>
3969 set_number_at <succeed_n count> <lower bound>
3970 succeed_n <after jump addr> <succeed_n count>
3972 jump_n <succeed_n addr> <jump count>
3973 (The upper bound and `jump_n' are omitted if
3974 `upper_bound' is 1, though.) */
3976 { /* If the upper bound is > 1, we need to insert
3977 more at the end of the loop. */
3978 unsigned nbytes = 2 + 4 * OFFSET_ADDRESS_SIZE +
3979 (upper_bound > 1) * (2 + 4 * OFFSET_ADDRESS_SIZE);
3981 GET_BUFFER_SPACE (nbytes);
3983 /* Initialize lower bound of the `succeed_n', even
3984 though it will be set during matching by its
3985 attendant `set_number_at' (inserted next),
3986 because `re_compile_fastmap' needs to know.
3987 Jump to the `jump_n' we might insert below. */
3988 INSERT_JUMP2 (succeed_n, laststart,
3989 b + 1 + 2 * OFFSET_ADDRESS_SIZE
3990 + (upper_bound > 1) * (1 + 2 * OFFSET_ADDRESS_SIZE)
3992 b += 1 + 2 * OFFSET_ADDRESS_SIZE;
3994 /* Code to initialize the lower bound. Insert
3995 before the `succeed_n'. The `5' is the last two
3996 bytes of this `set_number_at', plus 3 bytes of
3997 the following `succeed_n'. */
3998 /* ifdef WCHAR, The '1+2*OFFSET_ADDRESS_SIZE'
3999 is the 'set_number_at', plus '1+OFFSET_ADDRESS_SIZE'
4000 of the following `succeed_n'. */
4001 PREFIX(insert_op2) (set_number_at, laststart, 1
4002 + 2 * OFFSET_ADDRESS_SIZE, lower_bound, b);
4003 b += 1 + 2 * OFFSET_ADDRESS_SIZE;
4005 if (upper_bound > 1)
4006 { /* More than one repetition is allowed, so
4007 append a backward jump to the `succeed_n'
4008 that starts this interval.
4010 When we've reached this during matching,
4011 we'll have matched the interval once, so
4012 jump back only `upper_bound - 1' times. */
4013 STORE_JUMP2 (jump_n, b, laststart
4014 + 2 * OFFSET_ADDRESS_SIZE + 1,
4016 b += 1 + 2 * OFFSET_ADDRESS_SIZE;
4018 /* The location we want to set is the second
4019 parameter of the `jump_n'; that is `b-2' as
4020 an absolute address. `laststart' will be
4021 the `set_number_at' we're about to insert;
4022 `laststart+3' the number to set, the source
4023 for the relative address. But we are
4024 inserting into the middle of the pattern --
4025 so everything is getting moved up by 5.
4026 Conclusion: (b - 2) - (laststart + 3) + 5,
4027 i.e., b - laststart.
4029 We insert this at the beginning of the loop
4030 so that if we fail during matching, we'll
4031 reinitialize the bounds. */
4032 PREFIX(insert_op2) (set_number_at, laststart,
4034 upper_bound - 1, b);
4035 b += 1 + 2 * OFFSET_ADDRESS_SIZE;
4042 if (!(syntax & RE_INVALID_INTERVAL_ORD))
4043 FREE_STACK_RETURN (p == pend ? REG_EBRACE : REG_BADBR);
4045 /* Match the characters as literals. */
4048 if (syntax & RE_NO_BK_BRACES)
4051 goto normal_backslash;
4055 /* There is no way to specify the before_dot and after_dot
4056 operators. rms says this is ok. --karl */
4064 BUF_PUSH_2 (syntaxspec, syntax_spec_code[c]);
4070 BUF_PUSH_2 (notsyntaxspec, syntax_spec_code[c]);
4076 if (syntax & RE_NO_GNU_OPS)
4079 BUF_PUSH (wordchar);
4084 if (syntax & RE_NO_GNU_OPS)
4087 BUF_PUSH (notwordchar);
4092 if (syntax & RE_NO_GNU_OPS)
4098 if (syntax & RE_NO_GNU_OPS)
4104 if (syntax & RE_NO_GNU_OPS)
4106 BUF_PUSH (wordbound);
4110 if (syntax & RE_NO_GNU_OPS)
4112 BUF_PUSH (notwordbound);
4116 if (syntax & RE_NO_GNU_OPS)
4122 if (syntax & RE_NO_GNU_OPS)
4127 case '1': case '2': case '3': case '4': case '5':
4128 case '6': case '7': case '8': case '9':
4129 if (syntax & RE_NO_BK_REFS)
4135 FREE_STACK_RETURN (REG_ESUBREG);
4137 /* Can't back reference to a subexpression if inside of it. */
4138 if (group_in_compile_stack (compile_stack, (regnum_t) c1))
4142 BUF_PUSH_2 (duplicate, c1);
4148 if (syntax & RE_BK_PLUS_QM)
4151 goto normal_backslash;
4155 /* You might think it would be useful for \ to mean
4156 not to translate; but if we don't translate it
4157 it will never match anything. */
4165 /* Expects the character in `c'. */
4167 /* If no exactn currently being built. */
4170 /* If last exactn handle binary(or character) and
4171 new exactn handle character(or binary). */
4172 || is_exactn_bin != is_binary[p - 1 - pattern]
4175 /* If last exactn not at current position. */
4176 || pending_exact + *pending_exact + 1 != b
4178 /* We have only one byte following the exactn for the count. */
4179 || *pending_exact == (1 << BYTEWIDTH) - 1
4181 /* If followed by a repetition operator. */
4182 || *p == '*' || *p == '^'
4183 || ((syntax & RE_BK_PLUS_QM)
4184 ? *p == '\\' && (p[1] == '+' || p[1] == '?')
4185 : (*p == '+' || *p == '?'))
4186 || ((syntax & RE_INTERVALS)
4187 && ((syntax & RE_NO_BK_BRACES)
4189 : (p[0] == '\\' && p[1] == '{'))))
4191 /* Start building a new exactn. */
4196 /* Is this exactn binary data or character? */
4197 is_exactn_bin = is_binary[p - 1 - pattern];
4199 BUF_PUSH_2 (exactn_bin, 0);
4201 BUF_PUSH_2 (exactn, 0);
4203 BUF_PUSH_2 (exactn, 0);
4205 pending_exact = b - 1;
4212 } /* while p != pend */
4215 /* Through the pattern now. */
4218 STORE_JUMP (jump_past_alt, fixup_alt_jump, b);
4220 if (!COMPILE_STACK_EMPTY)
4221 FREE_STACK_RETURN (REG_EPAREN);
4223 /* If we don't want backtracking, force success
4224 the first time we reach the end of the compiled pattern. */
4225 if (syntax & RE_NO_POSIX_BACKTRACKING)
4233 free (compile_stack.stack);
4235 /* We have succeeded; set the length of the buffer. */
4237 bufp->used = (uintptr_t) b - (uintptr_t) COMPILED_BUFFER_VAR;
4239 bufp->used = b - bufp->buffer;
4245 DEBUG_PRINT1 ("\nCompiled pattern: \n");
4246 PREFIX(print_compiled_pattern) (bufp);
4250 #ifndef MATCH_MAY_ALLOCATE
4251 /* Initialize the failure stack to the largest possible stack. This
4252 isn't necessary unless we're trying to avoid calling alloca in
4253 the search and match routines. */
4255 int num_regs = bufp->re_nsub + 1;
4257 /* Since DOUBLE_FAIL_STACK refuses to double only if the current size
4258 is strictly greater than re_max_failures, the largest possible stack
4259 is 2 * re_max_failures failure points. */
4260 if (fail_stack.size < (2 * re_max_failures * MAX_FAILURE_ITEMS))
4262 fail_stack.size = (2 * re_max_failures * MAX_FAILURE_ITEMS);
4265 if (! fail_stack.stack)
4267 = (PREFIX(fail_stack_elt_t) *) xmalloc (fail_stack.size
4268 * sizeof (PREFIX(fail_stack_elt_t)));
4271 = (PREFIX(fail_stack_elt_t) *) xrealloc (fail_stack.stack,
4273 * sizeof (PREFIX(fail_stack_elt_t))));
4274 # else /* not emacs */
4275 if (! fail_stack.stack)
4277 = (PREFIX(fail_stack_elt_t) *) malloc (fail_stack.size
4278 * sizeof (PREFIX(fail_stack_elt_t)));
4281 = (PREFIX(fail_stack_elt_t) *) realloc (fail_stack.stack,
4283 * sizeof (PREFIX(fail_stack_elt_t))));
4284 # endif /* not emacs */
4287 PREFIX(regex_grow_registers) (num_regs);
4289 #endif /* not MATCH_MAY_ALLOCATE */
4292 } /* regex_compile */
4294 /* Subroutines for `regex_compile'. */
4296 /* Store OP at LOC followed by two-byte integer parameter ARG. */
4297 /* ifdef WCHAR, integer parameter is 1 wchar_t. */
4300 PREFIX(store_op1) (op, loc, arg)
4305 *loc = (UCHAR_T) op;
4306 STORE_NUMBER (loc + 1, arg);
4310 /* Like `store_op1', but for two two-byte parameters ARG1 and ARG2. */
4311 /* ifdef WCHAR, integer parameter is 1 wchar_t. */
4314 PREFIX(store_op2) (op, loc, arg1, arg2)
4319 *loc = (UCHAR_T) op;
4320 STORE_NUMBER (loc + 1, arg1);
4321 STORE_NUMBER (loc + 1 + OFFSET_ADDRESS_SIZE, arg2);
4325 /* Copy the bytes from LOC to END to open up three bytes of space at LOC
4326 for OP followed by two-byte integer parameter ARG. */
4327 /* ifdef WCHAR, integer parameter is 1 wchar_t. */
4330 PREFIX(insert_op1) (op, loc, arg, end)
4336 register UCHAR_T *pfrom = end;
4337 register UCHAR_T *pto = end + 1 + OFFSET_ADDRESS_SIZE;
4339 while (pfrom != loc)
4342 PREFIX(store_op1) (op, loc, arg);
4346 /* Like `insert_op1', but for two two-byte parameters ARG1 and ARG2. */
4347 /* ifdef WCHAR, integer parameter is 1 wchar_t. */
4350 PREFIX(insert_op2) (op, loc, arg1, arg2, end)
4356 register UCHAR_T *pfrom = end;
4357 register UCHAR_T *pto = end + 1 + 2 * OFFSET_ADDRESS_SIZE;
4359 while (pfrom != loc)
4362 PREFIX(store_op2) (op, loc, arg1, arg2);
4366 /* P points to just after a ^ in PATTERN. Return true if that ^ comes
4367 after an alternative or a begin-subexpression. We assume there is at
4368 least one character before the ^. */
4371 PREFIX(at_begline_loc_p) (pattern, p, syntax)
4372 const CHAR_T *pattern, *p;
4373 reg_syntax_t syntax;
4375 const CHAR_T *prev = p - 2;
4376 boolean prev_prev_backslash = prev > pattern && prev[-1] == '\\';
4379 /* After a subexpression? */
4380 (*prev == '(' && (syntax & RE_NO_BK_PARENS || prev_prev_backslash))
4381 /* After an alternative? */
4382 || (*prev == '|' && (syntax & RE_NO_BK_VBAR || prev_prev_backslash));
4386 /* The dual of at_begline_loc_p. This one is for $. We assume there is
4387 at least one character after the $, i.e., `P < PEND'. */
4390 PREFIX(at_endline_loc_p) (p, pend, syntax)
4391 const CHAR_T *p, *pend;
4392 reg_syntax_t syntax;
4394 const CHAR_T *next = p;
4395 boolean next_backslash = *next == '\\';
4396 const CHAR_T *next_next = p + 1 < pend ? p + 1 : 0;
4399 /* Before a subexpression? */
4400 (syntax & RE_NO_BK_PARENS ? *next == ')'
4401 : next_backslash && next_next && *next_next == ')')
4402 /* Before an alternative? */
4403 || (syntax & RE_NO_BK_VBAR ? *next == '|'
4404 : next_backslash && next_next && *next_next == '|');
4407 #else /* not INSIDE_RECURSION */
4409 /* Returns true if REGNUM is in one of COMPILE_STACK's elements and
4410 false if it's not. */
4413 group_in_compile_stack (compile_stack, regnum)
4414 compile_stack_type compile_stack;
4419 for (this_element = compile_stack.avail - 1;
4422 if (compile_stack.stack[this_element].regnum == regnum)
4427 #endif /* not INSIDE_RECURSION */
4429 #ifdef INSIDE_RECURSION
4432 /* This insert space, which size is "num", into the pattern at "loc".
4433 "end" must point the end of the allocated buffer. */
4435 insert_space (num, loc, end)
4440 register CHAR_T *pto = end;
4441 register CHAR_T *pfrom = end - num;
4443 while (pfrom >= loc)
4449 static reg_errcode_t
4450 wcs_compile_range (range_start_char, p_ptr, pend, translate, syntax, b,
4452 CHAR_T range_start_char;
4453 const CHAR_T **p_ptr, *pend;
4454 CHAR_T *char_set, *b;
4455 RE_TRANSLATE_TYPE translate;
4456 reg_syntax_t syntax;
4458 const CHAR_T *p = *p_ptr;
4459 CHAR_T range_start, range_end;
4463 uint32_t start_val, end_val;
4469 nrules = _NL_CURRENT_WORD (LC_COLLATE, _NL_COLLATE_NRULES);
4472 const char *collseq = (const char *) _NL_CURRENT(LC_COLLATE,
4473 _NL_COLLATE_COLLSEQWC);
4474 const unsigned char *extra = (const unsigned char *)
4475 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_SYMB_EXTRAMB);
4477 if (range_start_char < -1)
4479 /* range_start is a collating symbol. */
4481 /* Retreive the index and get collation sequence value. */
4482 wextra = (int32_t*)(extra + char_set[-range_start_char]);
4483 start_val = wextra[1 + *wextra];
4486 start_val = collseq_table_lookup(collseq, TRANSLATE(range_start_char));
4488 end_val = collseq_table_lookup (collseq, TRANSLATE (p[0]));
4490 /* Report an error if the range is empty and the syntax prohibits
4492 ret = ((syntax & RE_NO_EMPTY_RANGES)
4493 && (start_val > end_val))? REG_ERANGE : REG_NOERROR;
4495 /* Insert space to the end of the char_ranges. */
4496 insert_space(2, b - char_set[5] - 2, b - 1);
4497 *(b - char_set[5] - 2) = (wchar_t)start_val;
4498 *(b - char_set[5] - 1) = (wchar_t)end_val;
4499 char_set[4]++; /* ranges_index */
4504 range_start = (range_start_char >= 0)? TRANSLATE (range_start_char):
4506 range_end = TRANSLATE (p[0]);
4507 /* Report an error if the range is empty and the syntax prohibits
4509 ret = ((syntax & RE_NO_EMPTY_RANGES)
4510 && (range_start > range_end))? REG_ERANGE : REG_NOERROR;
4512 /* Insert space to the end of the char_ranges. */
4513 insert_space(2, b - char_set[5] - 2, b - 1);
4514 *(b - char_set[5] - 2) = range_start;
4515 *(b - char_set[5] - 1) = range_end;
4516 char_set[4]++; /* ranges_index */
4518 /* Have to increment the pointer into the pattern string, so the
4519 caller isn't still at the ending character. */
4525 /* Read the ending character of a range (in a bracket expression) from the
4526 uncompiled pattern *P_PTR (which ends at PEND). We assume the
4527 starting character is in `P[-2]'. (`P[-1]' is the character `-'.)
4528 Then we set the translation of all bits between the starting and
4529 ending characters (inclusive) in the compiled pattern B.
4531 Return an error code.
4533 We use these short variable names so we can use the same macros as
4534 `regex_compile' itself. */
4536 static reg_errcode_t
4537 byte_compile_range (range_start_char, p_ptr, pend, translate, syntax, b)
4538 unsigned int range_start_char;
4539 const char **p_ptr, *pend;
4540 RE_TRANSLATE_TYPE translate;
4541 reg_syntax_t syntax;
4545 const char *p = *p_ptr;
4548 const unsigned char *collseq;
4549 unsigned int start_colseq;
4550 unsigned int end_colseq;
4558 /* Have to increment the pointer into the pattern string, so the
4559 caller isn't still at the ending character. */
4562 /* Report an error if the range is empty and the syntax prohibits this. */
4563 ret = syntax & RE_NO_EMPTY_RANGES ? REG_ERANGE : REG_NOERROR;
4566 collseq = (const unsigned char *) _NL_CURRENT (LC_COLLATE,
4567 _NL_COLLATE_COLLSEQMB);
4569 start_colseq = collseq[(unsigned char) TRANSLATE (range_start_char)];
4570 end_colseq = collseq[(unsigned char) TRANSLATE (p[0])];
4571 for (this_char = 0; this_char <= (unsigned char) -1; ++this_char)
4573 unsigned int this_colseq = collseq[(unsigned char) TRANSLATE (this_char)];
4575 if (start_colseq <= this_colseq && this_colseq <= end_colseq)
4577 SET_LIST_BIT (TRANSLATE (this_char));
4582 /* Here we see why `this_char' has to be larger than an `unsigned
4583 char' -- we would otherwise go into an infinite loop, since all
4584 characters <= 0xff. */
4585 range_start_char = TRANSLATE (range_start_char);
4586 /* TRANSLATE(p[0]) is casted to char (not unsigned char) in TRANSLATE,
4587 and some compilers cast it to int implicitly, so following for_loop
4588 may fall to (almost) infinite loop.
4589 e.g. If translate[p[0]] = 0xff, end_char may equals to 0xffffffff.
4590 To avoid this, we cast p[0] to unsigned int and truncate it. */
4591 end_char = ((unsigned)TRANSLATE(p[0]) & ((1 << BYTEWIDTH) - 1));
4593 for (this_char = range_start_char; this_char <= end_char; ++this_char)
4595 SET_LIST_BIT (TRANSLATE (this_char));
4604 /* re_compile_fastmap computes a ``fastmap'' for the compiled pattern in
4605 BUFP. A fastmap records which of the (1 << BYTEWIDTH) possible
4606 characters can start a string that matches the pattern. This fastmap
4607 is used by re_search to skip quickly over impossible starting points.
4609 The caller must supply the address of a (1 << BYTEWIDTH)-byte data
4610 area as BUFP->fastmap.
4612 We set the `fastmap', `fastmap_accurate', and `can_be_null' fields in
4615 Returns 0 if we succeed, -2 if an internal error. */
4618 /* local function for re_compile_fastmap.
4619 truncate wchar_t character to char. */
4620 static unsigned char truncate_wchar (CHAR_T c);
4622 static unsigned char
4626 unsigned char buf[MB_LEN_MAX];
4627 int retval = wctomb(buf, c);
4628 return retval > 0 ? buf[0] : (unsigned char)c;
4633 PREFIX(re_compile_fastmap) (bufp)
4634 struct re_pattern_buffer *bufp;
4637 #ifdef MATCH_MAY_ALLOCATE
4638 PREFIX(fail_stack_type) fail_stack;
4640 #ifndef REGEX_MALLOC
4644 register char *fastmap = bufp->fastmap;
4647 /* We need to cast pattern to (wchar_t*), because we casted this compiled
4648 pattern to (char*) in regex_compile. */
4649 UCHAR_T *pattern = (UCHAR_T*)bufp->buffer;
4650 register UCHAR_T *pend = (UCHAR_T*) (bufp->buffer + bufp->used);
4652 UCHAR_T *pattern = bufp->buffer;
4653 register UCHAR_T *pend = pattern + bufp->used;
4655 UCHAR_T *p = pattern;
4658 /* This holds the pointer to the failure stack, when
4659 it is allocated relocatably. */
4660 fail_stack_elt_t *failure_stack_ptr;
4663 /* Assume that each path through the pattern can be null until
4664 proven otherwise. We set this false at the bottom of switch
4665 statement, to which we get only if a particular path doesn't
4666 match the empty string. */
4667 boolean path_can_be_null = true;
4669 /* We aren't doing a `succeed_n' to begin with. */
4670 boolean succeed_n_p = false;
4672 assert (fastmap != NULL && p != NULL);
4675 bzero (fastmap, 1 << BYTEWIDTH); /* Assume nothing's valid. */
4676 bufp->fastmap_accurate = 1; /* It will be when we're done. */
4677 bufp->can_be_null = 0;
4681 if (p == pend || *p == succeed)
4683 /* We have reached the (effective) end of pattern. */
4684 if (!FAIL_STACK_EMPTY ())
4686 bufp->can_be_null |= path_can_be_null;
4688 /* Reset for next path. */
4689 path_can_be_null = true;
4691 p = fail_stack.stack[--fail_stack.avail].pointer;
4699 /* We should never be about to go beyond the end of the pattern. */
4702 switch (SWITCH_ENUM_CAST ((re_opcode_t) *p++))
4705 /* I guess the idea here is to simply not bother with a fastmap
4706 if a backreference is used, since it's too hard to figure out
4707 the fastmap for the corresponding group. Setting
4708 `can_be_null' stops `re_search_2' from using the fastmap, so
4709 that is all we do. */
4711 bufp->can_be_null = 1;
4715 /* Following are the cases which match a character. These end
4720 fastmap[truncate_wchar(p[1])] = 1;
4734 /* It is hard to distinguish fastmap from (multi byte) characters
4735 which depends on current locale. */
4740 bufp->can_be_null = 1;
4744 for (j = *p++ * BYTEWIDTH - 1; j >= 0; j--)
4745 if (p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH)))
4751 /* Chars beyond end of map must be allowed. */
4752 for (j = *p * BYTEWIDTH; j < (1 << BYTEWIDTH); j++)
4755 for (j = *p++ * BYTEWIDTH - 1; j >= 0; j--)
4756 if (!(p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH))))
4762 for (j = 0; j < (1 << BYTEWIDTH); j++)
4763 if (SYNTAX (j) == Sword)
4769 for (j = 0; j < (1 << BYTEWIDTH); j++)
4770 if (SYNTAX (j) != Sword)
4777 int fastmap_newline = fastmap['\n'];
4779 /* `.' matches anything ... */
4780 for (j = 0; j < (1 << BYTEWIDTH); j++)
4783 /* ... except perhaps newline. */
4784 if (!(bufp->syntax & RE_DOT_NEWLINE))
4785 fastmap['\n'] = fastmap_newline;
4787 /* Return if we have already set `can_be_null'; if we have,
4788 then the fastmap is irrelevant. Something's wrong here. */
4789 else if (bufp->can_be_null)
4792 /* Otherwise, have to check alternative paths. */
4799 for (j = 0; j < (1 << BYTEWIDTH); j++)
4800 if (SYNTAX (j) == (enum syntaxcode) k)
4807 for (j = 0; j < (1 << BYTEWIDTH); j++)
4808 if (SYNTAX (j) != (enum syntaxcode) k)
4813 /* All cases after this match the empty string. These end with
4833 case push_dummy_failure:
4838 case pop_failure_jump:
4839 case maybe_pop_jump:
4842 case dummy_failure_jump:
4843 EXTRACT_NUMBER_AND_INCR (j, p);
4848 /* Jump backward implies we just went through the body of a
4849 loop and matched nothing. Opcode jumped to should be
4850 `on_failure_jump' or `succeed_n'. Just treat it like an
4851 ordinary jump. For a * loop, it has pushed its failure
4852 point already; if so, discard that as redundant. */
4853 if ((re_opcode_t) *p != on_failure_jump
4854 && (re_opcode_t) *p != succeed_n)
4858 EXTRACT_NUMBER_AND_INCR (j, p);
4861 /* If what's on the stack is where we are now, pop it. */
4862 if (!FAIL_STACK_EMPTY ()
4863 && fail_stack.stack[fail_stack.avail - 1].pointer == p)
4869 case on_failure_jump:
4870 case on_failure_keep_string_jump:
4871 handle_on_failure_jump:
4872 EXTRACT_NUMBER_AND_INCR (j, p);
4874 /* For some patterns, e.g., `(a?)?', `p+j' here points to the
4875 end of the pattern. We don't want to push such a point,
4876 since when we restore it above, entering the switch will
4877 increment `p' past the end of the pattern. We don't need
4878 to push such a point since we obviously won't find any more
4879 fastmap entries beyond `pend'. Such a pattern can match
4880 the null string, though. */
4883 if (!PUSH_PATTERN_OP (p + j, fail_stack))
4885 RESET_FAIL_STACK ();
4890 bufp->can_be_null = 1;
4894 EXTRACT_NUMBER_AND_INCR (k, p); /* Skip the n. */
4895 succeed_n_p = false;
4902 /* Get to the number of times to succeed. */
4903 p += OFFSET_ADDRESS_SIZE;
4905 /* Increment p past the n for when k != 0. */
4906 EXTRACT_NUMBER_AND_INCR (k, p);
4909 p -= 2 * OFFSET_ADDRESS_SIZE;
4910 succeed_n_p = true; /* Spaghetti code alert. */
4911 goto handle_on_failure_jump;
4917 p += 2 * OFFSET_ADDRESS_SIZE;
4928 abort (); /* We have listed all the cases. */
4931 /* Getting here means we have found the possible starting
4932 characters for one path of the pattern -- and that the empty
4933 string does not match. We need not follow this path further.
4934 Instead, look at the next alternative (remembered on the
4935 stack), or quit if no more. The test at the top of the loop
4936 does these things. */
4937 path_can_be_null = false;
4941 /* Set `can_be_null' for the last path (also the first path, if the
4942 pattern is empty). */
4943 bufp->can_be_null |= path_can_be_null;
4946 RESET_FAIL_STACK ();
4950 #else /* not INSIDE_RECURSION */
4953 re_compile_fastmap (bufp)
4954 struct re_pattern_buffer *bufp;
4957 if (MB_CUR_MAX != 1)
4958 return wcs_re_compile_fastmap(bufp);
4961 return byte_re_compile_fastmap(bufp);
4962 } /* re_compile_fastmap */
4964 weak_alias (__re_compile_fastmap, re_compile_fastmap)
4968 /* Set REGS to hold NUM_REGS registers, storing them in STARTS and
4969 ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use
4970 this memory for recording register information. STARTS and ENDS
4971 must be allocated using the malloc library routine, and must each
4972 be at least NUM_REGS * sizeof (regoff_t) bytes long.
4974 If NUM_REGS == 0, then subsequent matches should allocate their own
4977 Unless this function is called, the first search or match using
4978 PATTERN_BUFFER will allocate its own register data, without
4979 freeing the old data. */
4982 re_set_registers (bufp, regs, num_regs, starts, ends)
4983 struct re_pattern_buffer *bufp;
4984 struct re_registers *regs;
4986 regoff_t *starts, *ends;
4990 bufp->regs_allocated = REGS_REALLOCATE;
4991 regs->num_regs = num_regs;
4992 regs->start = starts;
4997 bufp->regs_allocated = REGS_UNALLOCATED;
4999 regs->start = regs->end = (regoff_t *) 0;
5003 weak_alias (__re_set_registers, re_set_registers)
5006 /* Searching routines. */
5008 /* Like re_search_2, below, but only one string is specified, and
5009 doesn't let you say where to stop matching. */
5012 re_search (bufp, string, size, startpos, range, regs)
5013 struct re_pattern_buffer *bufp;
5015 int size, startpos, range;
5016 struct re_registers *regs;
5018 return re_search_2 (bufp, NULL, 0, string, size, startpos, range,
5022 weak_alias (__re_search, re_search)
5026 /* Using the compiled pattern in BUFP->buffer, first tries to match the
5027 virtual concatenation of STRING1 and STRING2, starting first at index
5028 STARTPOS, then at STARTPOS + 1, and so on.
5030 STRING1 and STRING2 have length SIZE1 and SIZE2, respectively.
5032 RANGE is how far to scan while trying to match. RANGE = 0 means try
5033 only at STARTPOS; in general, the last start tried is STARTPOS +
5036 In REGS, return the indices of the virtual concatenation of STRING1
5037 and STRING2 that matched the entire BUFP->buffer and its contained
5040 Do not consider matching one past the index STOP in the virtual
5041 concatenation of STRING1 and STRING2.
5043 We return either the position in the strings at which the match was
5044 found, -1 if no match, or -2 if error (such as failure
5048 re_search_2 (bufp, string1, size1, string2, size2, startpos, range, regs, stop)
5049 struct re_pattern_buffer *bufp;
5050 const char *string1, *string2;
5054 struct re_registers *regs;
5058 if (MB_CUR_MAX != 1)
5059 return wcs_re_search_2 (bufp, string1, size1, string2, size2, startpos,
5063 return byte_re_search_2 (bufp, string1, size1, string2, size2, startpos,
5067 weak_alias (__re_search_2, re_search_2)
5070 #endif /* not INSIDE_RECURSION */
5072 #ifdef INSIDE_RECURSION
5074 #ifdef MATCH_MAY_ALLOCATE
5075 # define FREE_VAR(var) if (var) REGEX_FREE (var); var = NULL
5077 # define FREE_VAR(var) if (var) free (var); var = NULL
5081 # define MAX_ALLOCA_SIZE 2000
5083 # define FREE_WCS_BUFFERS() \
5085 if (size1 > MAX_ALLOCA_SIZE) \
5087 free (wcs_string1); \
5088 free (mbs_offset1); \
5092 FREE_VAR (wcs_string1); \
5093 FREE_VAR (mbs_offset1); \
5095 if (size2 > MAX_ALLOCA_SIZE) \
5097 free (wcs_string2); \
5098 free (mbs_offset2); \
5102 FREE_VAR (wcs_string2); \
5103 FREE_VAR (mbs_offset2); \
5111 PREFIX(re_search_2) (bufp, string1, size1, string2, size2, startpos, range,
5113 struct re_pattern_buffer *bufp;
5114 const char *string1, *string2;
5118 struct re_registers *regs;
5122 register char *fastmap = bufp->fastmap;
5123 register RE_TRANSLATE_TYPE translate = bufp->translate;
5124 int total_size = size1 + size2;
5125 int endpos = startpos + range;
5127 /* We need wchar_t* buffers correspond to cstring1, cstring2. */
5128 wchar_t *wcs_string1 = NULL, *wcs_string2 = NULL;
5129 /* We need the size of wchar_t buffers correspond to csize1, csize2. */
5130 int wcs_size1 = 0, wcs_size2 = 0;
5131 /* offset buffer for optimizatoin. See convert_mbs_to_wc. */
5132 int *mbs_offset1 = NULL, *mbs_offset2 = NULL;
5133 /* They hold whether each wchar_t is binary data or not. */
5134 char *is_binary = NULL;
5137 /* Check for out-of-range STARTPOS. */
5138 if (startpos < 0 || startpos > total_size)
5141 /* Fix up RANGE if it might eventually take us outside
5142 the virtual concatenation of STRING1 and STRING2.
5143 Make sure we won't move STARTPOS below 0 or above TOTAL_SIZE. */
5145 range = 0 - startpos;
5146 else if (endpos > total_size)
5147 range = total_size - startpos;
5149 /* If the search isn't to be a backwards one, don't waste time in a
5150 search for a pattern that must be anchored. */
5151 if (bufp->used > 0 && range > 0
5152 && ((re_opcode_t) bufp->buffer[0] == begbuf
5153 /* `begline' is like `begbuf' if it cannot match at newlines. */
5154 || ((re_opcode_t) bufp->buffer[0] == begline
5155 && !bufp->newline_anchor)))
5164 /* In a forward search for something that starts with \=.
5165 don't keep searching past point. */
5166 if (bufp->used > 0 && (re_opcode_t) bufp->buffer[0] == at_dot && range > 0)
5168 range = PT - startpos;
5174 /* Update the fastmap now if not correct already. */
5175 if (fastmap && !bufp->fastmap_accurate)
5176 if (re_compile_fastmap (bufp) == -2)
5180 /* Allocate wchar_t array for wcs_string1 and wcs_string2 and
5181 fill them with converted string. */
5184 if (size1 > MAX_ALLOCA_SIZE)
5186 wcs_string1 = TALLOC (size1 + 1, CHAR_T);
5187 mbs_offset1 = TALLOC (size1 + 1, int);
5188 is_binary = TALLOC (size1 + 1, char);
5192 wcs_string1 = REGEX_TALLOC (size1 + 1, CHAR_T);
5193 mbs_offset1 = REGEX_TALLOC (size1 + 1, int);
5194 is_binary = REGEX_TALLOC (size1 + 1, char);
5196 if (!wcs_string1 || !mbs_offset1 || !is_binary)
5198 if (size1 > MAX_ALLOCA_SIZE)
5206 FREE_VAR (wcs_string1);
5207 FREE_VAR (mbs_offset1);
5208 FREE_VAR (is_binary);
5212 wcs_size1 = convert_mbs_to_wcs(wcs_string1, string1, size1,
5213 mbs_offset1, is_binary);
5214 wcs_string1[wcs_size1] = L'\0'; /* for a sentinel */
5215 if (size1 > MAX_ALLOCA_SIZE)
5218 FREE_VAR (is_binary);
5222 if (size2 > MAX_ALLOCA_SIZE)
5224 wcs_string2 = TALLOC (size2 + 1, CHAR_T);
5225 mbs_offset2 = TALLOC (size2 + 1, int);
5226 is_binary = TALLOC (size2 + 1, char);
5230 wcs_string2 = REGEX_TALLOC (size2 + 1, CHAR_T);
5231 mbs_offset2 = REGEX_TALLOC (size2 + 1, int);
5232 is_binary = REGEX_TALLOC (size2 + 1, char);
5234 if (!wcs_string2 || !mbs_offset2 || !is_binary)
5236 FREE_WCS_BUFFERS ();
5237 if (size2 > MAX_ALLOCA_SIZE)
5240 FREE_VAR (is_binary);
5243 wcs_size2 = convert_mbs_to_wcs(wcs_string2, string2, size2,
5244 mbs_offset2, is_binary);
5245 wcs_string2[wcs_size2] = L'\0'; /* for a sentinel */
5246 if (size2 > MAX_ALLOCA_SIZE)
5249 FREE_VAR (is_binary);
5254 /* Loop through the string, looking for a place to start matching. */
5257 /* If a fastmap is supplied, skip quickly over characters that
5258 cannot be the start of a match. If the pattern can match the
5259 null string, however, we don't need to skip characters; we want
5260 the first null string. */
5261 if (fastmap && startpos < total_size && !bufp->can_be_null)
5263 if (range > 0) /* Searching forwards. */
5265 register const char *d;
5266 register int lim = 0;
5269 if (startpos < size1 && startpos + range >= size1)
5270 lim = range - (size1 - startpos);
5272 d = (startpos >= size1 ? string2 - size1 : string1) + startpos;
5274 /* Written out as an if-else to avoid testing `translate'
5278 && !fastmap[(unsigned char)
5279 translate[(unsigned char) *d++]])
5282 while (range > lim && !fastmap[(unsigned char) *d++])
5285 startpos += irange - range;
5287 else /* Searching backwards. */
5289 register CHAR_T c = (size1 == 0 || startpos >= size1
5290 ? string2[startpos - size1]
5291 : string1[startpos]);
5293 if (!fastmap[(unsigned char) TRANSLATE (c)])
5298 /* If can't match the null string, and that's all we have left, fail. */
5299 if (range >= 0 && startpos == total_size && fastmap
5300 && !bufp->can_be_null)
5303 FREE_WCS_BUFFERS ();
5309 val = wcs_re_match_2_internal (bufp, string1, size1, string2,
5310 size2, startpos, regs, stop,
5311 wcs_string1, wcs_size1,
5312 wcs_string2, wcs_size2,
5313 mbs_offset1, mbs_offset2);
5315 val = byte_re_match_2_internal (bufp, string1, size1, string2,
5316 size2, startpos, regs, stop);
5319 #ifndef REGEX_MALLOC
5328 FREE_WCS_BUFFERS ();
5336 FREE_WCS_BUFFERS ();
5356 FREE_WCS_BUFFERS ();
5362 /* This converts PTR, a pointer into one of the search wchar_t strings
5363 `string1' and `string2' into an multibyte string offset from the
5364 beginning of that string. We use mbs_offset to optimize.
5365 See convert_mbs_to_wcs. */
5366 # define POINTER_TO_OFFSET(ptr) \
5367 (FIRST_STRING_P (ptr) \
5368 ? ((regoff_t)(mbs_offset1 != NULL? mbs_offset1[(ptr)-string1] : 0)) \
5369 : ((regoff_t)((mbs_offset2 != NULL? mbs_offset2[(ptr)-string2] : 0) \
5372 /* This converts PTR, a pointer into one of the search strings `string1'
5373 and `string2' into an offset from the beginning of that string. */
5374 # define POINTER_TO_OFFSET(ptr) \
5375 (FIRST_STRING_P (ptr) \
5376 ? ((regoff_t) ((ptr) - string1)) \
5377 : ((regoff_t) ((ptr) - string2 + size1)))
5380 /* Macros for dealing with the split strings in re_match_2. */
5382 #define MATCHING_IN_FIRST_STRING (dend == end_match_1)
5384 /* Call before fetching a character with *d. This switches over to
5385 string2 if necessary. */
5386 #define PREFETCH() \
5389 /* End of string2 => fail. */ \
5390 if (dend == end_match_2) \
5392 /* End of string1 => advance to string2. */ \
5394 dend = end_match_2; \
5397 /* Test if at very beginning or at very end of the virtual concatenation
5398 of `string1' and `string2'. If only one string, it's `string2'. */
5399 #define AT_STRINGS_BEG(d) ((d) == (size1 ? string1 : string2) || !size2)
5400 #define AT_STRINGS_END(d) ((d) == end2)
5403 /* Test if D points to a character which is word-constituent. We have
5404 two special cases to check for: if past the end of string1, look at
5405 the first character in string2; and if before the beginning of
5406 string2, look at the last character in string1. */
5408 /* Use internationalized API instead of SYNTAX. */
5409 # define WORDCHAR_P(d) \
5410 (iswalnum ((wint_t)((d) == end1 ? *string2 \
5411 : (d) == string2 - 1 ? *(end1 - 1) : *(d))) != 0 \
5412 || ((d) == end1 ? *string2 \
5413 : (d) == string2 - 1 ? *(end1 - 1) : *(d)) == L'_')
5415 # define WORDCHAR_P(d) \
5416 (SYNTAX ((d) == end1 ? *string2 \
5417 : (d) == string2 - 1 ? *(end1 - 1) : *(d)) \
5421 /* Disabled due to a compiler bug -- see comment at case wordbound */
5423 /* Test if the character before D and the one at D differ with respect
5424 to being word-constituent. */
5425 #define AT_WORD_BOUNDARY(d) \
5426 (AT_STRINGS_BEG (d) || AT_STRINGS_END (d) \
5427 || WORDCHAR_P (d - 1) != WORDCHAR_P (d))
5430 /* Free everything we malloc. */
5431 #ifdef MATCH_MAY_ALLOCATE
5433 # define FREE_VARIABLES() \
5435 REGEX_FREE_STACK (fail_stack.stack); \
5436 FREE_VAR (regstart); \
5437 FREE_VAR (regend); \
5438 FREE_VAR (old_regstart); \
5439 FREE_VAR (old_regend); \
5440 FREE_VAR (best_regstart); \
5441 FREE_VAR (best_regend); \
5442 FREE_VAR (reg_info); \
5443 FREE_VAR (reg_dummy); \
5444 FREE_VAR (reg_info_dummy); \
5445 if (!cant_free_wcs_buf) \
5447 FREE_VAR (string1); \
5448 FREE_VAR (string2); \
5449 FREE_VAR (mbs_offset1); \
5450 FREE_VAR (mbs_offset2); \
5454 # define FREE_VARIABLES() \
5456 REGEX_FREE_STACK (fail_stack.stack); \
5457 FREE_VAR (regstart); \
5458 FREE_VAR (regend); \
5459 FREE_VAR (old_regstart); \
5460 FREE_VAR (old_regend); \
5461 FREE_VAR (best_regstart); \
5462 FREE_VAR (best_regend); \
5463 FREE_VAR (reg_info); \
5464 FREE_VAR (reg_dummy); \
5465 FREE_VAR (reg_info_dummy); \
5470 # define FREE_VARIABLES() \
5472 if (!cant_free_wcs_buf) \
5474 FREE_VAR (string1); \
5475 FREE_VAR (string2); \
5476 FREE_VAR (mbs_offset1); \
5477 FREE_VAR (mbs_offset2); \
5481 # define FREE_VARIABLES() ((void)0) /* Do nothing! But inhibit gcc warning. */
5483 #endif /* not MATCH_MAY_ALLOCATE */
5485 /* These values must meet several constraints. They must not be valid
5486 register values; since we have a limit of 255 registers (because
5487 we use only one byte in the pattern for the register number), we can
5488 use numbers larger than 255. They must differ by 1, because of
5489 NUM_FAILURE_ITEMS above. And the value for the lowest register must
5490 be larger than the value for the highest register, so we do not try
5491 to actually save any registers when none are active. */
5492 #define NO_HIGHEST_ACTIVE_REG (1 << BYTEWIDTH)
5493 #define NO_LOWEST_ACTIVE_REG (NO_HIGHEST_ACTIVE_REG + 1)
5495 #else /* not INSIDE_RECURSION */
5496 /* Matching routines. */
5498 #ifndef emacs /* Emacs never uses this. */
5499 /* re_match is like re_match_2 except it takes only a single string. */
5502 re_match (bufp, string, size, pos, regs)
5503 struct re_pattern_buffer *bufp;
5506 struct re_registers *regs;
5510 if (MB_CUR_MAX != 1)
5511 result = wcs_re_match_2_internal (bufp, NULL, 0, string, size,
5513 NULL, 0, NULL, 0, NULL, NULL);
5516 result = byte_re_match_2_internal (bufp, NULL, 0, string, size,
5518 # ifndef REGEX_MALLOC
5526 weak_alias (__re_match, re_match)
5528 #endif /* not emacs */
5530 #endif /* not INSIDE_RECURSION */
5532 #ifdef INSIDE_RECURSION
5533 static boolean PREFIX(group_match_null_string_p) _RE_ARGS ((UCHAR_T **p,
5535 PREFIX(register_info_type) *reg_info));
5536 static boolean PREFIX(alt_match_null_string_p) _RE_ARGS ((UCHAR_T *p,
5538 PREFIX(register_info_type) *reg_info));
5539 static boolean PREFIX(common_op_match_null_string_p) _RE_ARGS ((UCHAR_T **p,
5541 PREFIX(register_info_type) *reg_info));
5542 static int PREFIX(bcmp_translate) _RE_ARGS ((const CHAR_T *s1, const CHAR_T *s2,
5543 int len, char *translate));
5544 #else /* not INSIDE_RECURSION */
5546 /* re_match_2 matches the compiled pattern in BUFP against the
5547 the (virtual) concatenation of STRING1 and STRING2 (of length SIZE1
5548 and SIZE2, respectively). We start matching at POS, and stop
5551 If REGS is non-null and the `no_sub' field of BUFP is nonzero, we
5552 store offsets for the substring each group matched in REGS. See the
5553 documentation for exactly how many groups we fill.
5555 We return -1 if no match, -2 if an internal error (such as the
5556 failure stack overflowing). Otherwise, we return the length of the
5557 matched substring. */
5560 re_match_2 (bufp, string1, size1, string2, size2, pos, regs, stop)
5561 struct re_pattern_buffer *bufp;
5562 const char *string1, *string2;
5565 struct re_registers *regs;
5570 if (MB_CUR_MAX != 1)
5571 result = wcs_re_match_2_internal (bufp, string1, size1, string2, size2,
5573 NULL, 0, NULL, 0, NULL, NULL);
5576 result = byte_re_match_2_internal (bufp, string1, size1, string2, size2,
5579 #ifndef REGEX_MALLOC
5587 weak_alias (__re_match_2, re_match_2)
5590 #endif /* not INSIDE_RECURSION */
5592 #ifdef INSIDE_RECURSION
5595 static int count_mbs_length PARAMS ((int *, int));
5597 /* This check the substring (from 0, to length) of the multibyte string,
5598 to which offset_buffer correspond. And count how many wchar_t_characters
5599 the substring occupy. We use offset_buffer to optimization.
5600 See convert_mbs_to_wcs. */
5603 count_mbs_length(offset_buffer, length)
5609 /* Check whether the size is valid. */
5613 if (offset_buffer == NULL)
5616 /* If there are no multibyte character, offset_buffer[i] == i.
5617 Optmize for this case. */
5618 if (offset_buffer[length] == length)
5621 /* Set up upper with length. (because for all i, offset_buffer[i] >= i) */
5627 int middle = (lower + upper) / 2;
5628 if (middle == lower || middle == upper)
5630 if (offset_buffer[middle] > length)
5632 else if (offset_buffer[middle] < length)
5642 /* This is a separate function so that we can force an alloca cleanup
5646 wcs_re_match_2_internal (bufp, cstring1, csize1, cstring2, csize2, pos,
5647 regs, stop, string1, size1, string2, size2,
5648 mbs_offset1, mbs_offset2)
5649 struct re_pattern_buffer *bufp;
5650 const char *cstring1, *cstring2;
5653 struct re_registers *regs;
5655 /* string1 == string2 == NULL means string1/2, size1/2 and
5656 mbs_offset1/2 need seting up in this function. */
5657 /* We need wchar_t* buffers correspond to cstring1, cstring2. */
5658 wchar_t *string1, *string2;
5659 /* We need the size of wchar_t buffers correspond to csize1, csize2. */
5661 /* offset buffer for optimizatoin. See convert_mbs_to_wc. */
5662 int *mbs_offset1, *mbs_offset2;
5665 byte_re_match_2_internal (bufp, string1, size1,string2, size2, pos,
5667 struct re_pattern_buffer *bufp;
5668 const char *string1, *string2;
5671 struct re_registers *regs;
5675 /* General temporaries. */
5679 /* They hold whether each wchar_t is binary data or not. */
5680 char *is_binary = NULL;
5681 /* If true, we can't free string1/2, mbs_offset1/2. */
5682 int cant_free_wcs_buf = 1;
5685 /* Just past the end of the corresponding string. */
5686 const CHAR_T *end1, *end2;
5688 /* Pointers into string1 and string2, just past the last characters in
5689 each to consider matching. */
5690 const CHAR_T *end_match_1, *end_match_2;
5692 /* Where we are in the data, and the end of the current string. */
5693 const CHAR_T *d, *dend;
5695 /* Where we are in the pattern, and the end of the pattern. */
5697 UCHAR_T *pattern, *p;
5698 register UCHAR_T *pend;
5700 UCHAR_T *p = bufp->buffer;
5701 register UCHAR_T *pend = p + bufp->used;
5704 /* Mark the opcode just after a start_memory, so we can test for an
5705 empty subpattern when we get to the stop_memory. */
5706 UCHAR_T *just_past_start_mem = 0;
5708 /* We use this to map every character in the string. */
5709 RE_TRANSLATE_TYPE translate = bufp->translate;
5711 /* Failure point stack. Each place that can handle a failure further
5712 down the line pushes a failure point on this stack. It consists of
5713 restart, regend, and reg_info for all registers corresponding to
5714 the subexpressions we're currently inside, plus the number of such
5715 registers, and, finally, two char *'s. The first char * is where
5716 to resume scanning the pattern; the second one is where to resume
5717 scanning the strings. If the latter is zero, the failure point is
5718 a ``dummy''; if a failure happens and the failure point is a dummy,
5719 it gets discarded and the next next one is tried. */
5720 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
5721 PREFIX(fail_stack_type) fail_stack;
5724 static unsigned failure_id;
5725 unsigned nfailure_points_pushed = 0, nfailure_points_popped = 0;
5729 /* This holds the pointer to the failure stack, when
5730 it is allocated relocatably. */
5731 fail_stack_elt_t *failure_stack_ptr;
5734 /* We fill all the registers internally, independent of what we
5735 return, for use in backreferences. The number here includes
5736 an element for register zero. */
5737 size_t num_regs = bufp->re_nsub + 1;
5739 /* The currently active registers. */
5740 active_reg_t lowest_active_reg = NO_LOWEST_ACTIVE_REG;
5741 active_reg_t highest_active_reg = NO_HIGHEST_ACTIVE_REG;
5743 /* Information on the contents of registers. These are pointers into
5744 the input strings; they record just what was matched (on this
5745 attempt) by a subexpression part of the pattern, that is, the
5746 regnum-th regstart pointer points to where in the pattern we began
5747 matching and the regnum-th regend points to right after where we
5748 stopped matching the regnum-th subexpression. (The zeroth register
5749 keeps track of what the whole pattern matches.) */
5750 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
5751 const CHAR_T **regstart, **regend;
5754 /* If a group that's operated upon by a repetition operator fails to
5755 match anything, then the register for its start will need to be
5756 restored because it will have been set to wherever in the string we
5757 are when we last see its open-group operator. Similarly for a
5759 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
5760 const CHAR_T **old_regstart, **old_regend;
5763 /* The is_active field of reg_info helps us keep track of which (possibly
5764 nested) subexpressions we are currently in. The matched_something
5765 field of reg_info[reg_num] helps us tell whether or not we have
5766 matched any of the pattern so far this time through the reg_num-th
5767 subexpression. These two fields get reset each time through any
5768 loop their register is in. */
5769 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
5770 PREFIX(register_info_type) *reg_info;
5773 /* The following record the register info as found in the above
5774 variables when we find a match better than any we've seen before.
5775 This happens as we backtrack through the failure points, which in
5776 turn happens only if we have not yet matched the entire string. */
5777 unsigned best_regs_set = false;
5778 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
5779 const CHAR_T **best_regstart, **best_regend;
5782 /* Logically, this is `best_regend[0]'. But we don't want to have to
5783 allocate space for that if we're not allocating space for anything
5784 else (see below). Also, we never need info about register 0 for
5785 any of the other register vectors, and it seems rather a kludge to
5786 treat `best_regend' differently than the rest. So we keep track of
5787 the end of the best match so far in a separate variable. We
5788 initialize this to NULL so that when we backtrack the first time
5789 and need to test it, it's not garbage. */
5790 const CHAR_T *match_end = NULL;
5792 /* This helps SET_REGS_MATCHED avoid doing redundant work. */
5793 int set_regs_matched_done = 0;
5795 /* Used when we pop values we don't care about. */
5796 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
5797 const CHAR_T **reg_dummy;
5798 PREFIX(register_info_type) *reg_info_dummy;
5802 /* Counts the total number of registers pushed. */
5803 unsigned num_regs_pushed = 0;
5806 DEBUG_PRINT1 ("\n\nEntering re_match_2.\n");
5810 #ifdef MATCH_MAY_ALLOCATE
5811 /* Do not bother to initialize all the register variables if there are
5812 no groups in the pattern, as it takes a fair amount of time. If
5813 there are groups, we include space for register 0 (the whole
5814 pattern), even though we never use it, since it simplifies the
5815 array indexing. We should fix this. */
5818 regstart = REGEX_TALLOC (num_regs, const CHAR_T *);
5819 regend = REGEX_TALLOC (num_regs, const CHAR_T *);
5820 old_regstart = REGEX_TALLOC (num_regs, const CHAR_T *);
5821 old_regend = REGEX_TALLOC (num_regs, const CHAR_T *);
5822 best_regstart = REGEX_TALLOC (num_regs, const CHAR_T *);
5823 best_regend = REGEX_TALLOC (num_regs, const CHAR_T *);
5824 reg_info = REGEX_TALLOC (num_regs, PREFIX(register_info_type));
5825 reg_dummy = REGEX_TALLOC (num_regs, const CHAR_T *);
5826 reg_info_dummy = REGEX_TALLOC (num_regs, PREFIX(register_info_type));
5828 if (!(regstart && regend && old_regstart && old_regend && reg_info
5829 && best_regstart && best_regend && reg_dummy && reg_info_dummy))
5837 /* We must initialize all our variables to NULL, so that
5838 `FREE_VARIABLES' doesn't try to free them. */
5839 regstart = regend = old_regstart = old_regend = best_regstart
5840 = best_regend = reg_dummy = NULL;
5841 reg_info = reg_info_dummy = (PREFIX(register_info_type) *) NULL;
5843 #endif /* MATCH_MAY_ALLOCATE */
5845 /* The starting position is bogus. */
5847 if (pos < 0 || pos > csize1 + csize2)
5849 if (pos < 0 || pos > size1 + size2)
5857 /* Allocate wchar_t array for string1 and string2 and
5858 fill them with converted string. */
5859 if (string1 == NULL && string2 == NULL)
5861 /* We need seting up buffers here. */
5863 /* We must free wcs buffers in this function. */
5864 cant_free_wcs_buf = 0;
5868 string1 = REGEX_TALLOC (csize1 + 1, CHAR_T);
5869 mbs_offset1 = REGEX_TALLOC (csize1 + 1, int);
5870 is_binary = REGEX_TALLOC (csize1 + 1, char);
5871 if (!string1 || !mbs_offset1 || !is_binary)
5874 FREE_VAR (mbs_offset1);
5875 FREE_VAR (is_binary);
5881 string2 = REGEX_TALLOC (csize2 + 1, CHAR_T);
5882 mbs_offset2 = REGEX_TALLOC (csize2 + 1, int);
5883 is_binary = REGEX_TALLOC (csize2 + 1, char);
5884 if (!string2 || !mbs_offset2 || !is_binary)
5887 FREE_VAR (mbs_offset1);
5889 FREE_VAR (mbs_offset2);
5890 FREE_VAR (is_binary);
5893 size2 = convert_mbs_to_wcs(string2, cstring2, csize2,
5894 mbs_offset2, is_binary);
5895 string2[size2] = L'\0'; /* for a sentinel */
5896 FREE_VAR (is_binary);
5900 /* We need to cast pattern to (wchar_t*), because we casted this compiled
5901 pattern to (char*) in regex_compile. */
5902 p = pattern = (CHAR_T*)bufp->buffer;
5903 pend = (CHAR_T*)(bufp->buffer + bufp->used);
5907 /* Initialize subexpression text positions to -1 to mark ones that no
5908 start_memory/stop_memory has been seen for. Also initialize the
5909 register information struct. */
5910 for (mcnt = 1; (unsigned) mcnt < num_regs; mcnt++)
5912 regstart[mcnt] = regend[mcnt]
5913 = old_regstart[mcnt] = old_regend[mcnt] = REG_UNSET_VALUE;
5915 REG_MATCH_NULL_STRING_P (reg_info[mcnt]) = MATCH_NULL_UNSET_VALUE;
5916 IS_ACTIVE (reg_info[mcnt]) = 0;
5917 MATCHED_SOMETHING (reg_info[mcnt]) = 0;
5918 EVER_MATCHED_SOMETHING (reg_info[mcnt]) = 0;
5921 /* We move `string1' into `string2' if the latter's empty -- but not if
5922 `string1' is null. */
5923 if (size2 == 0 && string1 != NULL)
5930 mbs_offset2 = mbs_offset1;
5936 end1 = string1 + size1;
5937 end2 = string2 + size2;
5939 /* Compute where to stop matching, within the two strings. */
5943 mcnt = count_mbs_length(mbs_offset1, stop);
5944 end_match_1 = string1 + mcnt;
5945 end_match_2 = string2;
5949 if (stop > csize1 + csize2)
5950 stop = csize1 + csize2;
5952 mcnt = count_mbs_length(mbs_offset2, stop-csize1);
5953 end_match_2 = string2 + mcnt;
5956 { /* count_mbs_length return error. */
5963 end_match_1 = string1 + stop;
5964 end_match_2 = string2;
5969 end_match_2 = string2 + stop - size1;
5973 /* `p' scans through the pattern as `d' scans through the data.
5974 `dend' is the end of the input string that `d' points within. `d'
5975 is advanced into the following input string whenever necessary, but
5976 this happens before fetching; therefore, at the beginning of the
5977 loop, `d' can be pointing at the end of a string, but it cannot
5980 if (size1 > 0 && pos <= csize1)
5982 mcnt = count_mbs_length(mbs_offset1, pos);
5988 mcnt = count_mbs_length(mbs_offset2, pos-csize1);
5994 { /* count_mbs_length return error. */
5999 if (size1 > 0 && pos <= size1)
6006 d = string2 + pos - size1;
6011 DEBUG_PRINT1 ("The compiled pattern is:\n");
6012 DEBUG_PRINT_COMPILED_PATTERN (bufp, p, pend);
6013 DEBUG_PRINT1 ("The string to match is: `");
6014 DEBUG_PRINT_DOUBLE_STRING (d, string1, size1, string2, size2);
6015 DEBUG_PRINT1 ("'\n");
6017 /* This loops over pattern commands. It exits by returning from the
6018 function if the match is complete, or it drops through if the match
6019 fails at this starting point in the input data. */
6023 DEBUG_PRINT2 ("\n%p: ", p);
6025 DEBUG_PRINT2 ("\n0x%x: ", p);
6029 { /* End of pattern means we might have succeeded. */
6030 DEBUG_PRINT1 ("end of pattern ... ");
6032 /* If we haven't matched the entire string, and we want the
6033 longest match, try backtracking. */
6034 if (d != end_match_2)
6036 /* 1 if this match ends in the same string (string1 or string2)
6037 as the best previous match. */
6038 boolean same_str_p = (FIRST_STRING_P (match_end)
6039 == MATCHING_IN_FIRST_STRING);
6040 /* 1 if this match is the best seen so far. */
6041 boolean best_match_p;
6043 /* AIX compiler got confused when this was combined
6044 with the previous declaration. */
6046 best_match_p = d > match_end;
6048 best_match_p = !MATCHING_IN_FIRST_STRING;
6050 DEBUG_PRINT1 ("backtracking.\n");
6052 if (!FAIL_STACK_EMPTY ())
6053 { /* More failure points to try. */
6055 /* If exceeds best match so far, save it. */
6056 if (!best_regs_set || best_match_p)
6058 best_regs_set = true;
6061 DEBUG_PRINT1 ("\nSAVING match as best so far.\n");
6063 for (mcnt = 1; (unsigned) mcnt < num_regs; mcnt++)
6065 best_regstart[mcnt] = regstart[mcnt];
6066 best_regend[mcnt] = regend[mcnt];
6072 /* If no failure points, don't restore garbage. And if
6073 last match is real best match, don't restore second
6075 else if (best_regs_set && !best_match_p)
6078 /* Restore best match. It may happen that `dend ==
6079 end_match_1' while the restored d is in string2.
6080 For example, the pattern `x.*y.*z' against the
6081 strings `x-' and `y-z-', if the two strings are
6082 not consecutive in memory. */
6083 DEBUG_PRINT1 ("Restoring best registers.\n");
6086 dend = ((d >= string1 && d <= end1)
6087 ? end_match_1 : end_match_2);
6089 for (mcnt = 1; (unsigned) mcnt < num_regs; mcnt++)
6091 regstart[mcnt] = best_regstart[mcnt];
6092 regend[mcnt] = best_regend[mcnt];
6095 } /* d != end_match_2 */
6098 DEBUG_PRINT1 ("Accepting match.\n");
6099 /* If caller wants register contents data back, do it. */
6100 if (regs && !bufp->no_sub)
6102 /* Have the register data arrays been allocated? */
6103 if (bufp->regs_allocated == REGS_UNALLOCATED)
6104 { /* No. So allocate them with malloc. We need one
6105 extra element beyond `num_regs' for the `-1' marker
6107 regs->num_regs = MAX (RE_NREGS, num_regs + 1);
6108 regs->start = TALLOC (regs->num_regs, regoff_t);
6109 regs->end = TALLOC (regs->num_regs, regoff_t);
6110 if (regs->start == NULL || regs->end == NULL)
6115 bufp->regs_allocated = REGS_REALLOCATE;
6117 else if (bufp->regs_allocated == REGS_REALLOCATE)
6118 { /* Yes. If we need more elements than were already
6119 allocated, reallocate them. If we need fewer, just
6121 if (regs->num_regs < num_regs + 1)
6123 regs->num_regs = num_regs + 1;
6124 RETALLOC (regs->start, regs->num_regs, regoff_t);
6125 RETALLOC (regs->end, regs->num_regs, regoff_t);
6126 if (regs->start == NULL || regs->end == NULL)
6135 /* These braces fend off a "empty body in an else-statement"
6136 warning under GCC when assert expands to nothing. */
6137 assert (bufp->regs_allocated == REGS_FIXED);
6140 /* Convert the pointer data in `regstart' and `regend' to
6141 indices. Register zero has to be set differently,
6142 since we haven't kept track of any info for it. */
6143 if (regs->num_regs > 0)
6145 regs->start[0] = pos;
6147 if (MATCHING_IN_FIRST_STRING)
6148 regs->end[0] = mbs_offset1 != NULL ?
6149 mbs_offset1[d-string1] : 0;
6151 regs->end[0] = csize1 + (mbs_offset2 != NULL ?
6152 mbs_offset2[d-string2] : 0);
6154 regs->end[0] = (MATCHING_IN_FIRST_STRING
6155 ? ((regoff_t) (d - string1))
6156 : ((regoff_t) (d - string2 + size1)));
6160 /* Go through the first `min (num_regs, regs->num_regs)'
6161 registers, since that is all we initialized. */
6162 for (mcnt = 1; (unsigned) mcnt < MIN (num_regs, regs->num_regs);
6165 if (REG_UNSET (regstart[mcnt]) || REG_UNSET (regend[mcnt]))
6166 regs->start[mcnt] = regs->end[mcnt] = -1;
6170 = (regoff_t) POINTER_TO_OFFSET (regstart[mcnt]);
6172 = (regoff_t) POINTER_TO_OFFSET (regend[mcnt]);
6176 /* If the regs structure we return has more elements than
6177 were in the pattern, set the extra elements to -1. If
6178 we (re)allocated the registers, this is the case,
6179 because we always allocate enough to have at least one
6181 for (mcnt = num_regs; (unsigned) mcnt < regs->num_regs; mcnt++)
6182 regs->start[mcnt] = regs->end[mcnt] = -1;
6183 } /* regs && !bufp->no_sub */
6185 DEBUG_PRINT4 ("%u failure points pushed, %u popped (%u remain).\n",
6186 nfailure_points_pushed, nfailure_points_popped,
6187 nfailure_points_pushed - nfailure_points_popped);
6188 DEBUG_PRINT2 ("%u registers pushed.\n", num_regs_pushed);
6191 if (MATCHING_IN_FIRST_STRING)
6192 mcnt = mbs_offset1 != NULL ? mbs_offset1[d-string1] : 0;
6194 mcnt = (mbs_offset2 != NULL ? mbs_offset2[d-string2] : 0) +
6198 mcnt = d - pos - (MATCHING_IN_FIRST_STRING
6203 DEBUG_PRINT2 ("Returning %d from re_match_2.\n", mcnt);
6209 /* Otherwise match next pattern command. */
6210 switch (SWITCH_ENUM_CAST ((re_opcode_t) *p++))
6212 /* Ignore these. Used to ignore the n of succeed_n's which
6213 currently have n == 0. */
6215 DEBUG_PRINT1 ("EXECUTING no_op.\n");
6219 DEBUG_PRINT1 ("EXECUTING succeed.\n");
6222 /* Match the next n pattern characters exactly. The following
6223 byte in the pattern defines n, and the n bytes after that
6224 are the characters to match. */
6230 DEBUG_PRINT2 ("EXECUTING exactn %d.\n", mcnt);
6232 /* This is written out as an if-else so we don't waste time
6233 testing `translate' inside the loop. */
6242 if ((UCHAR_T) translate[(unsigned char) *d++]
6248 if (*d++ != (CHAR_T) *p++)
6252 if ((UCHAR_T) translate[(unsigned char) *d++]
6264 if (*d++ != (CHAR_T) *p++) goto fail;
6268 SET_REGS_MATCHED ();
6272 /* Match any character except possibly a newline or a null. */
6274 DEBUG_PRINT1 ("EXECUTING anychar.\n");
6278 if ((!(bufp->syntax & RE_DOT_NEWLINE) && TRANSLATE (*d) == '\n')
6279 || (bufp->syntax & RE_DOT_NOT_NULL && TRANSLATE (*d) == '\000'))
6282 SET_REGS_MATCHED ();
6283 DEBUG_PRINT2 (" Matched `%ld'.\n", (long int) *d);
6293 unsigned int i, char_class_length, coll_symbol_length,
6294 equiv_class_length, ranges_length, chars_length, length;
6295 CHAR_T *workp, *workp2, *charset_top;
6296 #define WORK_BUFFER_SIZE 128
6297 CHAR_T str_buf[WORK_BUFFER_SIZE];
6302 boolean not = (re_opcode_t) *(p - 1) == charset_not;
6304 DEBUG_PRINT2 ("EXECUTING charset%s.\n", not ? "_not" : "");
6306 c = TRANSLATE (*d); /* The character to match. */
6309 nrules = _NL_CURRENT_WORD (LC_COLLATE, _NL_COLLATE_NRULES);
6311 charset_top = p - 1;
6312 char_class_length = *p++;
6313 coll_symbol_length = *p++;
6314 equiv_class_length = *p++;
6315 ranges_length = *p++;
6316 chars_length = *p++;
6317 /* p points charset[6], so the address of the next instruction
6318 (charset[l+m+n+2o+k+p']) equals p[l+m+n+2*o+p'],
6319 where l=length of char_classes, m=length of collating_symbol,
6320 n=equivalence_class, o=length of char_range,
6321 p'=length of character. */
6323 /* Update p to indicate the next instruction. */
6324 p += char_class_length + coll_symbol_length+ equiv_class_length +
6325 2*ranges_length + chars_length;
6327 /* match with char_class? */
6328 for (i = 0; i < char_class_length ; i += CHAR_CLASS_SIZE)
6331 uintptr_t alignedp = ((uintptr_t)workp
6332 + __alignof__(wctype_t) - 1)
6333 & ~(uintptr_t)(__alignof__(wctype_t) - 1);
6334 wctype = *((wctype_t*)alignedp);
6335 workp += CHAR_CLASS_SIZE;
6336 if (iswctype((wint_t)c, wctype))
6337 goto char_set_matched;
6340 /* match with collating_symbol? */
6344 const unsigned char *extra = (const unsigned char *)
6345 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_SYMB_EXTRAMB);
6347 for (workp2 = workp + coll_symbol_length ; workp < workp2 ;
6351 wextra = (int32_t*)(extra + *workp++);
6352 for (i = 0; i < *wextra; ++i)
6353 if (TRANSLATE(d[i]) != wextra[1 + i])
6358 /* Update d, however d will be incremented at
6359 char_set_matched:, we decrement d here. */
6361 goto char_set_matched;
6365 else /* (nrules == 0) */
6367 /* If we can't look up collation data, we use wcscoll
6370 for (workp2 = workp + coll_symbol_length ; workp < workp2 ;)
6372 const CHAR_T *backup_d = d, *backup_dend = dend;
6373 length = wcslen(workp);
6375 /* If wcscoll(the collating symbol, whole string) > 0,
6376 any substring of the string never match with the
6377 collating symbol. */
6378 if (wcscoll(workp, d) > 0)
6380 workp += length + 1;
6384 /* First, we compare the collating symbol with
6385 the first character of the string.
6386 If it don't match, we add the next character to
6387 the compare buffer in turn. */
6388 for (i = 0 ; i < WORK_BUFFER_SIZE-1 ; i++, d++)
6393 if (dend == end_match_2)
6399 /* add next character to the compare buffer. */
6400 str_buf[i] = TRANSLATE(*d);
6401 str_buf[i+1] = '\0';
6403 match = wcscoll(workp, str_buf);
6405 goto char_set_matched;
6408 /* (str_buf > workp) indicate (str_buf + X > workp),
6409 because for all X (str_buf + X > str_buf).
6410 So we don't need continue this loop. */
6413 /* Otherwise(str_buf < workp),
6414 (str_buf+next_character) may equals (workp).
6415 So we continue this loop. */
6420 workp += length + 1;
6423 /* match with equivalence_class? */
6427 const CHAR_T *backup_d = d, *backup_dend = dend;
6428 /* Try to match the equivalence class against
6429 those known to the collate implementation. */
6430 const int32_t *table;
6431 const int32_t *weights;
6432 const int32_t *extra;
6433 const int32_t *indirect;
6438 /* This #include defines a local function! */
6439 # include <locale/weightwc.h>
6441 table = (const int32_t *)
6442 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_TABLEWC);
6443 weights = (const wint_t *)
6444 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_WEIGHTWC);
6445 extra = (const wint_t *)
6446 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_EXTRAWC);
6447 indirect = (const int32_t *)
6448 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_INDIRECTWC);
6450 /* Write 1 collating element to str_buf, and
6454 for (i = 0 ; idx2 == 0 && i < WORK_BUFFER_SIZE - 1; i++)
6456 cp = (wint_t*)str_buf;
6459 if (dend == end_match_2)
6464 str_buf[i] = TRANSLATE(*(d+i));
6465 str_buf[i+1] = '\0'; /* sentinel */
6466 idx2 = findidx ((const wint_t**)&cp);
6469 /* Update d, however d will be incremented at
6470 char_set_matched:, we decrement d here. */
6471 d = backup_d + ((wchar_t*)cp - (wchar_t*)str_buf - 1);
6474 if (dend == end_match_2)
6483 len = weights[idx2];
6485 for (workp2 = workp + equiv_class_length ; workp < workp2 ;
6488 idx = (int32_t)*workp;
6489 /* We already checked idx != 0 in regex_compile. */
6491 if (idx2 != 0 && len == weights[idx])
6494 while (cnt < len && (weights[idx + 1 + cnt]
6495 == weights[idx2 + 1 + cnt]))
6499 goto char_set_matched;
6506 else /* (nrules == 0) */
6508 /* If we can't look up collation data, we use wcscoll
6511 for (workp2 = workp + equiv_class_length ; workp < workp2 ;)
6513 const CHAR_T *backup_d = d, *backup_dend = dend;
6514 length = wcslen(workp);
6516 /* If wcscoll(the collating symbol, whole string) > 0,
6517 any substring of the string never match with the
6518 collating symbol. */
6519 if (wcscoll(workp, d) > 0)
6521 workp += length + 1;
6525 /* First, we compare the equivalence class with
6526 the first character of the string.
6527 If it don't match, we add the next character to
6528 the compare buffer in turn. */
6529 for (i = 0 ; i < WORK_BUFFER_SIZE - 1 ; i++, d++)
6534 if (dend == end_match_2)
6540 /* add next character to the compare buffer. */
6541 str_buf[i] = TRANSLATE(*d);
6542 str_buf[i+1] = '\0';
6544 match = wcscoll(workp, str_buf);
6547 goto char_set_matched;
6550 /* (str_buf > workp) indicate (str_buf + X > workp),
6551 because for all X (str_buf + X > str_buf).
6552 So we don't need continue this loop. */
6555 /* Otherwise(str_buf < workp),
6556 (str_buf+next_character) may equals (workp).
6557 So we continue this loop. */
6562 workp += length + 1;
6566 /* match with char_range? */
6570 uint32_t collseqval;
6571 const char *collseq = (const char *)
6572 _NL_CURRENT(LC_COLLATE, _NL_COLLATE_COLLSEQWC);
6574 collseqval = collseq_table_lookup (collseq, c);
6576 for (; workp < p - chars_length ;)
6578 uint32_t start_val, end_val;
6580 /* We already compute the collation sequence value
6581 of the characters (or collating symbols). */
6582 start_val = (uint32_t) *workp++; /* range_start */
6583 end_val = (uint32_t) *workp++; /* range_end */
6585 if (start_val <= collseqval && collseqval <= end_val)
6586 goto char_set_matched;
6592 /* We set range_start_char at str_buf[0], range_end_char
6593 at str_buf[4], and compared char at str_buf[2]. */
6598 for (; workp < p - chars_length ;)
6600 wchar_t *range_start_char, *range_end_char;
6602 /* match if (range_start_char <= c <= range_end_char). */
6604 /* If range_start(or end) < 0, we assume -range_start(end)
6605 is the offset of the collating symbol which is specified
6606 as the character of the range start(end). */
6610 range_start_char = charset_top - (*workp++);
6613 str_buf[0] = *workp++;
6614 range_start_char = str_buf;
6619 range_end_char = charset_top - (*workp++);
6622 str_buf[4] = *workp++;
6623 range_end_char = str_buf + 4;
6626 if (wcscoll(range_start_char, str_buf+2) <= 0 &&
6627 wcscoll(str_buf+2, range_end_char) <= 0)
6629 goto char_set_matched;
6633 /* match with char? */
6634 for (; workp < p ; workp++)
6636 goto char_set_matched;
6643 /* Cast to `unsigned' instead of `unsigned char' in case the
6644 bit list is a full 32 bytes long. */
6645 if (c < (unsigned) (*p * BYTEWIDTH)
6646 && p[1 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH)))
6651 if (!not) goto fail;
6652 #undef WORK_BUFFER_SIZE
6654 SET_REGS_MATCHED ();
6660 /* The beginning of a group is represented by start_memory.
6661 The arguments are the register number in the next byte, and the
6662 number of groups inner to this one in the next. The text
6663 matched within the group is recorded (in the internal
6664 registers data structure) under the register number. */
6666 DEBUG_PRINT3 ("EXECUTING start_memory %ld (%ld):\n",
6667 (long int) *p, (long int) p[1]);
6669 /* Find out if this group can match the empty string. */
6670 p1 = p; /* To send to group_match_null_string_p. */
6672 if (REG_MATCH_NULL_STRING_P (reg_info[*p]) == MATCH_NULL_UNSET_VALUE)
6673 REG_MATCH_NULL_STRING_P (reg_info[*p])
6674 = PREFIX(group_match_null_string_p) (&p1, pend, reg_info);
6676 /* Save the position in the string where we were the last time
6677 we were at this open-group operator in case the group is
6678 operated upon by a repetition operator, e.g., with `(a*)*b'
6679 against `ab'; then we want to ignore where we are now in
6680 the string in case this attempt to match fails. */
6681 old_regstart[*p] = REG_MATCH_NULL_STRING_P (reg_info[*p])
6682 ? REG_UNSET (regstart[*p]) ? d : regstart[*p]
6684 DEBUG_PRINT2 (" old_regstart: %d\n",
6685 POINTER_TO_OFFSET (old_regstart[*p]));
6688 DEBUG_PRINT2 (" regstart: %d\n", POINTER_TO_OFFSET (regstart[*p]));
6690 IS_ACTIVE (reg_info[*p]) = 1;
6691 MATCHED_SOMETHING (reg_info[*p]) = 0;
6693 /* Clear this whenever we change the register activity status. */
6694 set_regs_matched_done = 0;
6696 /* This is the new highest active register. */
6697 highest_active_reg = *p;
6699 /* If nothing was active before, this is the new lowest active
6701 if (lowest_active_reg == NO_LOWEST_ACTIVE_REG)
6702 lowest_active_reg = *p;
6704 /* Move past the register number and inner group count. */
6706 just_past_start_mem = p;
6711 /* The stop_memory opcode represents the end of a group. Its
6712 arguments are the same as start_memory's: the register
6713 number, and the number of inner groups. */
6715 DEBUG_PRINT3 ("EXECUTING stop_memory %ld (%ld):\n",
6716 (long int) *p, (long int) p[1]);
6718 /* We need to save the string position the last time we were at
6719 this close-group operator in case the group is operated
6720 upon by a repetition operator, e.g., with `((a*)*(b*)*)*'
6721 against `aba'; then we want to ignore where we are now in
6722 the string in case this attempt to match fails. */
6723 old_regend[*p] = REG_MATCH_NULL_STRING_P (reg_info[*p])
6724 ? REG_UNSET (regend[*p]) ? d : regend[*p]
6726 DEBUG_PRINT2 (" old_regend: %d\n",
6727 POINTER_TO_OFFSET (old_regend[*p]));
6730 DEBUG_PRINT2 (" regend: %d\n", POINTER_TO_OFFSET (regend[*p]));
6732 /* This register isn't active anymore. */
6733 IS_ACTIVE (reg_info[*p]) = 0;
6735 /* Clear this whenever we change the register activity status. */
6736 set_regs_matched_done = 0;
6738 /* If this was the only register active, nothing is active
6740 if (lowest_active_reg == highest_active_reg)
6742 lowest_active_reg = NO_LOWEST_ACTIVE_REG;
6743 highest_active_reg = NO_HIGHEST_ACTIVE_REG;
6746 { /* We must scan for the new highest active register, since
6747 it isn't necessarily one less than now: consider
6748 (a(b)c(d(e)f)g). When group 3 ends, after the f), the
6749 new highest active register is 1. */
6751 while (r > 0 && !IS_ACTIVE (reg_info[r]))
6754 /* If we end up at register zero, that means that we saved
6755 the registers as the result of an `on_failure_jump', not
6756 a `start_memory', and we jumped to past the innermost
6757 `stop_memory'. For example, in ((.)*) we save
6758 registers 1 and 2 as a result of the *, but when we pop
6759 back to the second ), we are at the stop_memory 1.
6760 Thus, nothing is active. */
6763 lowest_active_reg = NO_LOWEST_ACTIVE_REG;
6764 highest_active_reg = NO_HIGHEST_ACTIVE_REG;
6767 highest_active_reg = r;
6770 /* If just failed to match something this time around with a
6771 group that's operated on by a repetition operator, try to
6772 force exit from the ``loop'', and restore the register
6773 information for this group that we had before trying this
6775 if ((!MATCHED_SOMETHING (reg_info[*p])
6776 || just_past_start_mem == p - 1)
6779 boolean is_a_jump_n = false;
6783 switch ((re_opcode_t) *p1++)
6787 case pop_failure_jump:
6788 case maybe_pop_jump:
6790 case dummy_failure_jump:
6791 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
6793 p1 += OFFSET_ADDRESS_SIZE;
6801 /* If the next operation is a jump backwards in the pattern
6802 to an on_failure_jump right before the start_memory
6803 corresponding to this stop_memory, exit from the loop
6804 by forcing a failure after pushing on the stack the
6805 on_failure_jump's jump in the pattern, and d. */
6806 if (mcnt < 0 && (re_opcode_t) *p1 == on_failure_jump
6807 && (re_opcode_t) p1[1+OFFSET_ADDRESS_SIZE] == start_memory
6808 && p1[2+OFFSET_ADDRESS_SIZE] == *p)
6810 /* If this group ever matched anything, then restore
6811 what its registers were before trying this last
6812 failed match, e.g., with `(a*)*b' against `ab' for
6813 regstart[1], and, e.g., with `((a*)*(b*)*)*'
6814 against `aba' for regend[3].
6816 Also restore the registers for inner groups for,
6817 e.g., `((a*)(b*))*' against `aba' (register 3 would
6818 otherwise get trashed). */
6820 if (EVER_MATCHED_SOMETHING (reg_info[*p]))
6824 EVER_MATCHED_SOMETHING (reg_info[*p]) = 0;
6826 /* Restore this and inner groups' (if any) registers. */
6827 for (r = *p; r < (unsigned) *p + (unsigned) *(p + 1);
6830 regstart[r] = old_regstart[r];
6832 /* xx why this test? */
6833 if (old_regend[r] >= regstart[r])
6834 regend[r] = old_regend[r];
6838 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
6839 PUSH_FAILURE_POINT (p1 + mcnt, d, -2);
6845 /* Move past the register number and the inner group count. */
6850 /* \<digit> has been turned into a `duplicate' command which is
6851 followed by the numeric value of <digit> as the register number. */
6854 register const CHAR_T *d2, *dend2;
6855 int regno = *p++; /* Get which register to match against. */
6856 DEBUG_PRINT2 ("EXECUTING duplicate %d.\n", regno);
6858 /* Can't back reference a group which we've never matched. */
6859 if (REG_UNSET (regstart[regno]) || REG_UNSET (regend[regno]))
6862 /* Where in input to try to start matching. */
6863 d2 = regstart[regno];
6865 /* Where to stop matching; if both the place to start and
6866 the place to stop matching are in the same string, then
6867 set to the place to stop, otherwise, for now have to use
6868 the end of the first string. */
6870 dend2 = ((FIRST_STRING_P (regstart[regno])
6871 == FIRST_STRING_P (regend[regno]))
6872 ? regend[regno] : end_match_1);
6875 /* If necessary, advance to next segment in register
6879 if (dend2 == end_match_2) break;
6880 if (dend2 == regend[regno]) break;
6882 /* End of string1 => advance to string2. */
6884 dend2 = regend[regno];
6886 /* At end of register contents => success */
6887 if (d2 == dend2) break;
6889 /* If necessary, advance to next segment in data. */
6892 /* How many characters left in this segment to match. */
6895 /* Want how many consecutive characters we can match in
6896 one shot, so, if necessary, adjust the count. */
6897 if (mcnt > dend2 - d2)
6900 /* Compare that many; failure if mismatch, else move
6903 ? PREFIX(bcmp_translate) (d, d2, mcnt, translate)
6904 : memcmp (d, d2, mcnt*sizeof(UCHAR_T)))
6906 d += mcnt, d2 += mcnt;
6908 /* Do this because we've match some characters. */
6909 SET_REGS_MATCHED ();
6915 /* begline matches the empty string at the beginning of the string
6916 (unless `not_bol' is set in `bufp'), and, if
6917 `newline_anchor' is set, after newlines. */
6919 DEBUG_PRINT1 ("EXECUTING begline.\n");
6921 if (AT_STRINGS_BEG (d))
6923 if (!bufp->not_bol) break;
6925 else if (d[-1] == '\n' && bufp->newline_anchor)
6929 /* In all other cases, we fail. */
6933 /* endline is the dual of begline. */
6935 DEBUG_PRINT1 ("EXECUTING endline.\n");
6937 if (AT_STRINGS_END (d))
6939 if (!bufp->not_eol) break;
6942 /* We have to ``prefetch'' the next character. */
6943 else if ((d == end1 ? *string2 : *d) == '\n'
6944 && bufp->newline_anchor)
6951 /* Match at the very beginning of the data. */
6953 DEBUG_PRINT1 ("EXECUTING begbuf.\n");
6954 if (AT_STRINGS_BEG (d))
6959 /* Match at the very end of the data. */
6961 DEBUG_PRINT1 ("EXECUTING endbuf.\n");
6962 if (AT_STRINGS_END (d))
6967 /* on_failure_keep_string_jump is used to optimize `.*\n'. It
6968 pushes NULL as the value for the string on the stack. Then
6969 `pop_failure_point' will keep the current value for the
6970 string, instead of restoring it. To see why, consider
6971 matching `foo\nbar' against `.*\n'. The .* matches the foo;
6972 then the . fails against the \n. But the next thing we want
6973 to do is match the \n against the \n; if we restored the
6974 string value, we would be back at the foo.
6976 Because this is used only in specific cases, we don't need to
6977 check all the things that `on_failure_jump' does, to make
6978 sure the right things get saved on the stack. Hence we don't
6979 share its code. The only reason to push anything on the
6980 stack at all is that otherwise we would have to change
6981 `anychar's code to do something besides goto fail in this
6982 case; that seems worse than this. */
6983 case on_failure_keep_string_jump:
6984 DEBUG_PRINT1 ("EXECUTING on_failure_keep_string_jump");
6986 EXTRACT_NUMBER_AND_INCR (mcnt, p);
6988 DEBUG_PRINT3 (" %d (to %p):\n", mcnt, p + mcnt);
6990 DEBUG_PRINT3 (" %d (to 0x%x):\n", mcnt, p + mcnt);
6993 PUSH_FAILURE_POINT (p + mcnt, NULL, -2);
6997 /* Uses of on_failure_jump:
6999 Each alternative starts with an on_failure_jump that points
7000 to the beginning of the next alternative. Each alternative
7001 except the last ends with a jump that in effect jumps past
7002 the rest of the alternatives. (They really jump to the
7003 ending jump of the following alternative, because tensioning
7004 these jumps is a hassle.)
7006 Repeats start with an on_failure_jump that points past both
7007 the repetition text and either the following jump or
7008 pop_failure_jump back to this on_failure_jump. */
7009 case on_failure_jump:
7011 DEBUG_PRINT1 ("EXECUTING on_failure_jump");
7013 EXTRACT_NUMBER_AND_INCR (mcnt, p);
7015 DEBUG_PRINT3 (" %d (to %p)", mcnt, p + mcnt);
7017 DEBUG_PRINT3 (" %d (to 0x%x)", mcnt, p + mcnt);
7020 /* If this on_failure_jump comes right before a group (i.e.,
7021 the original * applied to a group), save the information
7022 for that group and all inner ones, so that if we fail back
7023 to this point, the group's information will be correct.
7024 For example, in \(a*\)*\1, we need the preceding group,
7025 and in \(zz\(a*\)b*\)\2, we need the inner group. */
7027 /* We can't use `p' to check ahead because we push
7028 a failure point to `p + mcnt' after we do this. */
7031 /* We need to skip no_op's before we look for the
7032 start_memory in case this on_failure_jump is happening as
7033 the result of a completed succeed_n, as in \(a\)\{1,3\}b\1
7035 while (p1 < pend && (re_opcode_t) *p1 == no_op)
7038 if (p1 < pend && (re_opcode_t) *p1 == start_memory)
7040 /* We have a new highest active register now. This will
7041 get reset at the start_memory we are about to get to,
7042 but we will have saved all the registers relevant to
7043 this repetition op, as described above. */
7044 highest_active_reg = *(p1 + 1) + *(p1 + 2);
7045 if (lowest_active_reg == NO_LOWEST_ACTIVE_REG)
7046 lowest_active_reg = *(p1 + 1);
7049 DEBUG_PRINT1 (":\n");
7050 PUSH_FAILURE_POINT (p + mcnt, d, -2);
7054 /* A smart repeat ends with `maybe_pop_jump'.
7055 We change it to either `pop_failure_jump' or `jump'. */
7056 case maybe_pop_jump:
7057 EXTRACT_NUMBER_AND_INCR (mcnt, p);
7058 DEBUG_PRINT2 ("EXECUTING maybe_pop_jump %d.\n", mcnt);
7060 register UCHAR_T *p2 = p;
7062 /* Compare the beginning of the repeat with what in the
7063 pattern follows its end. If we can establish that there
7064 is nothing that they would both match, i.e., that we
7065 would have to backtrack because of (as in, e.g., `a*a')
7066 then we can change to pop_failure_jump, because we'll
7067 never have to backtrack.
7069 This is not true in the case of alternatives: in
7070 `(a|ab)*' we do need to backtrack to the `ab' alternative
7071 (e.g., if the string was `ab'). But instead of trying to
7072 detect that here, the alternative has put on a dummy
7073 failure point which is what we will end up popping. */
7075 /* Skip over open/close-group commands.
7076 If what follows this loop is a ...+ construct,
7077 look at what begins its body, since we will have to
7078 match at least one of that. */
7082 && ((re_opcode_t) *p2 == stop_memory
7083 || (re_opcode_t) *p2 == start_memory))
7085 else if (p2 + 2 + 2 * OFFSET_ADDRESS_SIZE < pend
7086 && (re_opcode_t) *p2 == dummy_failure_jump)
7087 p2 += 2 + 2 * OFFSET_ADDRESS_SIZE;
7093 /* p1[0] ... p1[2] are the `on_failure_jump' corresponding
7094 to the `maybe_finalize_jump' of this case. Examine what
7097 /* If we're at the end of the pattern, we can change. */
7100 /* Consider what happens when matching ":\(.*\)"
7101 against ":/". I don't really understand this code
7103 p[-(1+OFFSET_ADDRESS_SIZE)] = (UCHAR_T)
7106 (" End of pattern: change to `pop_failure_jump'.\n");
7109 else if ((re_opcode_t) *p2 == exactn
7111 || (re_opcode_t) *p2 == exactn_bin
7113 || (bufp->newline_anchor && (re_opcode_t) *p2 == endline))
7116 = *p2 == (UCHAR_T) endline ? '\n' : p2[2];
7118 if (((re_opcode_t) p1[1+OFFSET_ADDRESS_SIZE] == exactn
7120 || (re_opcode_t) p1[1+OFFSET_ADDRESS_SIZE] == exactn_bin
7122 ) && p1[3+OFFSET_ADDRESS_SIZE] != c)
7124 p[-(1+OFFSET_ADDRESS_SIZE)] = (UCHAR_T)
7127 DEBUG_PRINT3 (" %C != %C => pop_failure_jump.\n",
7129 (wint_t) p1[3+OFFSET_ADDRESS_SIZE]);
7131 DEBUG_PRINT3 (" %c != %c => pop_failure_jump.\n",
7133 (char) p1[3+OFFSET_ADDRESS_SIZE]);
7138 else if ((re_opcode_t) p1[3] == charset
7139 || (re_opcode_t) p1[3] == charset_not)
7141 int not = (re_opcode_t) p1[3] == charset_not;
7143 if (c < (unsigned) (p1[4] * BYTEWIDTH)
7144 && p1[5 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH)))
7147 /* `not' is equal to 1 if c would match, which means
7148 that we can't change to pop_failure_jump. */
7151 p[-3] = (unsigned char) pop_failure_jump;
7152 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
7155 #endif /* not WCHAR */
7158 else if ((re_opcode_t) *p2 == charset)
7160 /* We win if the first character of the loop is not part
7162 if ((re_opcode_t) p1[3] == exactn
7163 && ! ((int) p2[1] * BYTEWIDTH > (int) p1[5]
7164 && (p2[2 + p1[5] / BYTEWIDTH]
7165 & (1 << (p1[5] % BYTEWIDTH)))))
7167 p[-3] = (unsigned char) pop_failure_jump;
7168 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
7171 else if ((re_opcode_t) p1[3] == charset_not)
7174 /* We win if the charset_not inside the loop
7175 lists every character listed in the charset after. */
7176 for (idx = 0; idx < (int) p2[1]; idx++)
7177 if (! (p2[2 + idx] == 0
7178 || (idx < (int) p1[4]
7179 && ((p2[2 + idx] & ~ p1[5 + idx]) == 0))))
7184 p[-3] = (unsigned char) pop_failure_jump;
7185 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
7188 else if ((re_opcode_t) p1[3] == charset)
7191 /* We win if the charset inside the loop
7192 has no overlap with the one after the loop. */
7194 idx < (int) p2[1] && idx < (int) p1[4];
7196 if ((p2[2 + idx] & p1[5 + idx]) != 0)
7199 if (idx == p2[1] || idx == p1[4])
7201 p[-3] = (unsigned char) pop_failure_jump;
7202 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
7206 #endif /* not WCHAR */
7208 p -= OFFSET_ADDRESS_SIZE; /* Point at relative address again. */
7209 if ((re_opcode_t) p[-1] != pop_failure_jump)
7211 p[-1] = (UCHAR_T) jump;
7212 DEBUG_PRINT1 (" Match => jump.\n");
7213 goto unconditional_jump;
7215 /* Note fall through. */
7218 /* The end of a simple repeat has a pop_failure_jump back to
7219 its matching on_failure_jump, where the latter will push a
7220 failure point. The pop_failure_jump takes off failure
7221 points put on by this pop_failure_jump's matching
7222 on_failure_jump; we got through the pattern to here from the
7223 matching on_failure_jump, so didn't fail. */
7224 case pop_failure_jump:
7226 /* We need to pass separate storage for the lowest and
7227 highest registers, even though we don't care about the
7228 actual values. Otherwise, we will restore only one
7229 register from the stack, since lowest will == highest in
7230 `pop_failure_point'. */
7231 active_reg_t dummy_low_reg, dummy_high_reg;
7232 UCHAR_T *pdummy = NULL;
7233 const CHAR_T *sdummy = NULL;
7235 DEBUG_PRINT1 ("EXECUTING pop_failure_jump.\n");
7236 POP_FAILURE_POINT (sdummy, pdummy,
7237 dummy_low_reg, dummy_high_reg,
7238 reg_dummy, reg_dummy, reg_info_dummy);
7240 /* Note fall through. */
7244 DEBUG_PRINT2 ("\n%p: ", p);
7246 DEBUG_PRINT2 ("\n0x%x: ", p);
7248 /* Note fall through. */
7250 /* Unconditionally jump (without popping any failure points). */
7252 EXTRACT_NUMBER_AND_INCR (mcnt, p); /* Get the amount to jump. */
7253 DEBUG_PRINT2 ("EXECUTING jump %d ", mcnt);
7254 p += mcnt; /* Do the jump. */
7256 DEBUG_PRINT2 ("(to %p).\n", p);
7258 DEBUG_PRINT2 ("(to 0x%x).\n", p);
7263 /* We need this opcode so we can detect where alternatives end
7264 in `group_match_null_string_p' et al. */
7266 DEBUG_PRINT1 ("EXECUTING jump_past_alt.\n");
7267 goto unconditional_jump;
7270 /* Normally, the on_failure_jump pushes a failure point, which
7271 then gets popped at pop_failure_jump. We will end up at
7272 pop_failure_jump, also, and with a pattern of, say, `a+', we
7273 are skipping over the on_failure_jump, so we have to push
7274 something meaningless for pop_failure_jump to pop. */
7275 case dummy_failure_jump:
7276 DEBUG_PRINT1 ("EXECUTING dummy_failure_jump.\n");
7277 /* It doesn't matter what we push for the string here. What
7278 the code at `fail' tests is the value for the pattern. */
7279 PUSH_FAILURE_POINT (NULL, NULL, -2);
7280 goto unconditional_jump;
7283 /* At the end of an alternative, we need to push a dummy failure
7284 point in case we are followed by a `pop_failure_jump', because
7285 we don't want the failure point for the alternative to be
7286 popped. For example, matching `(a|ab)*' against `aab'
7287 requires that we match the `ab' alternative. */
7288 case push_dummy_failure:
7289 DEBUG_PRINT1 ("EXECUTING push_dummy_failure.\n");
7290 /* See comments just above at `dummy_failure_jump' about the
7292 PUSH_FAILURE_POINT (NULL, NULL, -2);
7295 /* Have to succeed matching what follows at least n times.
7296 After that, handle like `on_failure_jump'. */
7298 EXTRACT_NUMBER (mcnt, p + OFFSET_ADDRESS_SIZE);
7299 DEBUG_PRINT2 ("EXECUTING succeed_n %d.\n", mcnt);
7302 /* Originally, this is how many times we HAVE to succeed. */
7306 p += OFFSET_ADDRESS_SIZE;
7307 STORE_NUMBER_AND_INCR (p, mcnt);
7309 DEBUG_PRINT3 (" Setting %p to %d.\n", p - OFFSET_ADDRESS_SIZE
7312 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p - OFFSET_ADDRESS_SIZE
7319 DEBUG_PRINT2 (" Setting two bytes from %p to no_op.\n",
7320 p + OFFSET_ADDRESS_SIZE);
7322 DEBUG_PRINT2 (" Setting two bytes from 0x%x to no_op.\n",
7323 p + OFFSET_ADDRESS_SIZE);
7327 p[1] = (UCHAR_T) no_op;
7329 p[2] = (UCHAR_T) no_op;
7330 p[3] = (UCHAR_T) no_op;
7337 EXTRACT_NUMBER (mcnt, p + OFFSET_ADDRESS_SIZE);
7338 DEBUG_PRINT2 ("EXECUTING jump_n %d.\n", mcnt);
7340 /* Originally, this is how many times we CAN jump. */
7344 STORE_NUMBER (p + OFFSET_ADDRESS_SIZE, mcnt);
7347 DEBUG_PRINT3 (" Setting %p to %d.\n", p + OFFSET_ADDRESS_SIZE,
7350 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p + OFFSET_ADDRESS_SIZE,
7353 goto unconditional_jump;
7355 /* If don't have to jump any more, skip over the rest of command. */
7357 p += 2 * OFFSET_ADDRESS_SIZE;
7362 DEBUG_PRINT1 ("EXECUTING set_number_at.\n");
7364 EXTRACT_NUMBER_AND_INCR (mcnt, p);
7366 EXTRACT_NUMBER_AND_INCR (mcnt, p);
7368 DEBUG_PRINT3 (" Setting %p to %d.\n", p1, mcnt);
7370 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p1, mcnt);
7372 STORE_NUMBER (p1, mcnt);
7377 /* The DEC Alpha C compiler 3.x generates incorrect code for the
7378 test WORDCHAR_P (d - 1) != WORDCHAR_P (d) in the expansion of
7379 AT_WORD_BOUNDARY, so this code is disabled. Expanding the
7380 macro and introducing temporary variables works around the bug. */
7383 DEBUG_PRINT1 ("EXECUTING wordbound.\n");
7384 if (AT_WORD_BOUNDARY (d))
7389 DEBUG_PRINT1 ("EXECUTING notwordbound.\n");
7390 if (AT_WORD_BOUNDARY (d))
7396 boolean prevchar, thischar;
7398 DEBUG_PRINT1 ("EXECUTING wordbound.\n");
7399 if (AT_STRINGS_BEG (d) || AT_STRINGS_END (d))
7402 prevchar = WORDCHAR_P (d - 1);
7403 thischar = WORDCHAR_P (d);
7404 if (prevchar != thischar)
7411 boolean prevchar, thischar;
7413 DEBUG_PRINT1 ("EXECUTING notwordbound.\n");
7414 if (AT_STRINGS_BEG (d) || AT_STRINGS_END (d))
7417 prevchar = WORDCHAR_P (d - 1);
7418 thischar = WORDCHAR_P (d);
7419 if (prevchar != thischar)
7426 DEBUG_PRINT1 ("EXECUTING wordbeg.\n");
7427 if (!AT_STRINGS_END (d) && WORDCHAR_P (d)
7428 && (AT_STRINGS_BEG (d) || !WORDCHAR_P (d - 1)))
7433 DEBUG_PRINT1 ("EXECUTING wordend.\n");
7434 if (!AT_STRINGS_BEG (d) && WORDCHAR_P (d - 1)
7435 && (AT_STRINGS_END (d) || !WORDCHAR_P (d)))
7441 DEBUG_PRINT1 ("EXECUTING before_dot.\n");
7442 if (PTR_CHAR_POS ((unsigned char *) d) >= point)
7447 DEBUG_PRINT1 ("EXECUTING at_dot.\n");
7448 if (PTR_CHAR_POS ((unsigned char *) d) != point)
7453 DEBUG_PRINT1 ("EXECUTING after_dot.\n");
7454 if (PTR_CHAR_POS ((unsigned char *) d) <= point)
7459 DEBUG_PRINT2 ("EXECUTING syntaxspec %d.\n", mcnt);
7464 DEBUG_PRINT1 ("EXECUTING Emacs wordchar.\n");
7468 /* Can't use *d++ here; SYNTAX may be an unsafe macro. */
7470 if (SYNTAX (d[-1]) != (enum syntaxcode) mcnt)
7472 SET_REGS_MATCHED ();
7476 DEBUG_PRINT2 ("EXECUTING notsyntaxspec %d.\n", mcnt);
7478 goto matchnotsyntax;
7481 DEBUG_PRINT1 ("EXECUTING Emacs notwordchar.\n");
7485 /* Can't use *d++ here; SYNTAX may be an unsafe macro. */
7487 if (SYNTAX (d[-1]) == (enum syntaxcode) mcnt)
7489 SET_REGS_MATCHED ();
7492 #else /* not emacs */
7494 DEBUG_PRINT1 ("EXECUTING non-Emacs wordchar.\n");
7496 if (!WORDCHAR_P (d))
7498 SET_REGS_MATCHED ();
7503 DEBUG_PRINT1 ("EXECUTING non-Emacs notwordchar.\n");
7507 SET_REGS_MATCHED ();
7510 #endif /* not emacs */
7515 continue; /* Successfully executed one pattern command; keep going. */
7518 /* We goto here if a matching operation fails. */
7520 if (!FAIL_STACK_EMPTY ())
7521 { /* A restart point is known. Restore to that state. */
7522 DEBUG_PRINT1 ("\nFAIL:\n");
7523 POP_FAILURE_POINT (d, p,
7524 lowest_active_reg, highest_active_reg,
7525 regstart, regend, reg_info);
7527 /* If this failure point is a dummy, try the next one. */
7531 /* If we failed to the end of the pattern, don't examine *p. */
7535 boolean is_a_jump_n = false;
7537 /* If failed to a backwards jump that's part of a repetition
7538 loop, need to pop this failure point and use the next one. */
7539 switch ((re_opcode_t) *p)
7543 case maybe_pop_jump:
7544 case pop_failure_jump:
7547 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
7550 if ((is_a_jump_n && (re_opcode_t) *p1 == succeed_n)
7552 && (re_opcode_t) *p1 == on_failure_jump))
7560 if (d >= string1 && d <= end1)
7564 break; /* Matching at this starting point really fails. */
7568 goto restore_best_regs;
7572 return -1; /* Failure to match. */
7575 /* Subroutine definitions for re_match_2. */
7578 /* We are passed P pointing to a register number after a start_memory.
7580 Return true if the pattern up to the corresponding stop_memory can
7581 match the empty string, and false otherwise.
7583 If we find the matching stop_memory, sets P to point to one past its number.
7584 Otherwise, sets P to an undefined byte less than or equal to END.
7586 We don't handle duplicates properly (yet). */
7589 PREFIX(group_match_null_string_p) (p, end, reg_info)
7591 PREFIX(register_info_type) *reg_info;
7594 /* Point to after the args to the start_memory. */
7595 UCHAR_T *p1 = *p + 2;
7599 /* Skip over opcodes that can match nothing, and return true or
7600 false, as appropriate, when we get to one that can't, or to the
7601 matching stop_memory. */
7603 switch ((re_opcode_t) *p1)
7605 /* Could be either a loop or a series of alternatives. */
7606 case on_failure_jump:
7608 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
7610 /* If the next operation is not a jump backwards in the
7615 /* Go through the on_failure_jumps of the alternatives,
7616 seeing if any of the alternatives cannot match nothing.
7617 The last alternative starts with only a jump,
7618 whereas the rest start with on_failure_jump and end
7619 with a jump, e.g., here is the pattern for `a|b|c':
7621 /on_failure_jump/0/6/exactn/1/a/jump_past_alt/0/6
7622 /on_failure_jump/0/6/exactn/1/b/jump_past_alt/0/3
7625 So, we have to first go through the first (n-1)
7626 alternatives and then deal with the last one separately. */
7629 /* Deal with the first (n-1) alternatives, which start
7630 with an on_failure_jump (see above) that jumps to right
7631 past a jump_past_alt. */
7633 while ((re_opcode_t) p1[mcnt-(1+OFFSET_ADDRESS_SIZE)] ==
7636 /* `mcnt' holds how many bytes long the alternative
7637 is, including the ending `jump_past_alt' and
7640 if (!PREFIX(alt_match_null_string_p) (p1, p1 + mcnt -
7641 (1 + OFFSET_ADDRESS_SIZE),
7645 /* Move to right after this alternative, including the
7649 /* Break if it's the beginning of an n-th alternative
7650 that doesn't begin with an on_failure_jump. */
7651 if ((re_opcode_t) *p1 != on_failure_jump)
7654 /* Still have to check that it's not an n-th
7655 alternative that starts with an on_failure_jump. */
7657 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
7658 if ((re_opcode_t) p1[mcnt-(1+OFFSET_ADDRESS_SIZE)] !=
7661 /* Get to the beginning of the n-th alternative. */
7662 p1 -= 1 + OFFSET_ADDRESS_SIZE;
7667 /* Deal with the last alternative: go back and get number
7668 of the `jump_past_alt' just before it. `mcnt' contains
7669 the length of the alternative. */
7670 EXTRACT_NUMBER (mcnt, p1 - OFFSET_ADDRESS_SIZE);
7672 if (!PREFIX(alt_match_null_string_p) (p1, p1 + mcnt, reg_info))
7675 p1 += mcnt; /* Get past the n-th alternative. */
7681 assert (p1[1] == **p);
7687 if (!PREFIX(common_op_match_null_string_p) (&p1, end, reg_info))
7690 } /* while p1 < end */
7693 } /* group_match_null_string_p */
7696 /* Similar to group_match_null_string_p, but doesn't deal with alternatives:
7697 It expects P to be the first byte of a single alternative and END one
7698 byte past the last. The alternative can contain groups. */
7701 PREFIX(alt_match_null_string_p) (p, end, reg_info)
7703 PREFIX(register_info_type) *reg_info;
7710 /* Skip over opcodes that can match nothing, and break when we get
7711 to one that can't. */
7713 switch ((re_opcode_t) *p1)
7716 case on_failure_jump:
7718 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
7723 if (!PREFIX(common_op_match_null_string_p) (&p1, end, reg_info))
7726 } /* while p1 < end */
7729 } /* alt_match_null_string_p */
7732 /* Deals with the ops common to group_match_null_string_p and
7733 alt_match_null_string_p.
7735 Sets P to one after the op and its arguments, if any. */
7738 PREFIX(common_op_match_null_string_p) (p, end, reg_info)
7740 PREFIX(register_info_type) *reg_info;
7747 switch ((re_opcode_t) *p1++)
7767 assert (reg_no > 0 && reg_no <= MAX_REGNUM);
7768 ret = PREFIX(group_match_null_string_p) (&p1, end, reg_info);
7770 /* Have to set this here in case we're checking a group which
7771 contains a group and a back reference to it. */
7773 if (REG_MATCH_NULL_STRING_P (reg_info[reg_no]) == MATCH_NULL_UNSET_VALUE)
7774 REG_MATCH_NULL_STRING_P (reg_info[reg_no]) = ret;
7780 /* If this is an optimized succeed_n for zero times, make the jump. */
7782 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
7790 /* Get to the number of times to succeed. */
7791 p1 += OFFSET_ADDRESS_SIZE;
7792 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
7796 p1 -= 2 * OFFSET_ADDRESS_SIZE;
7797 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
7805 if (!REG_MATCH_NULL_STRING_P (reg_info[*p1]))
7810 p1 += 2 * OFFSET_ADDRESS_SIZE;
7813 /* All other opcodes mean we cannot match the empty string. */
7819 } /* common_op_match_null_string_p */
7822 /* Return zero if TRANSLATE[S1] and TRANSLATE[S2] are identical for LEN
7823 bytes; nonzero otherwise. */
7826 PREFIX(bcmp_translate) (s1, s2, len, translate)
7827 const CHAR_T *s1, *s2;
7829 RE_TRANSLATE_TYPE translate;
7831 register const UCHAR_T *p1 = (const UCHAR_T *) s1;
7832 register const UCHAR_T *p2 = (const UCHAR_T *) s2;
7836 if (((*p1<=0xff)?translate[*p1++]:*p1++)
7837 != ((*p2<=0xff)?translate[*p2++]:*p2++))
7840 if (translate[*p1++] != translate[*p2++]) return 1;
7848 #else /* not INSIDE_RECURSION */
7850 /* Entry points for GNU code. */
7852 /* re_compile_pattern is the GNU regular expression compiler: it
7853 compiles PATTERN (of length SIZE) and puts the result in BUFP.
7854 Returns 0 if the pattern was valid, otherwise an error string.
7856 Assumes the `allocated' (and perhaps `buffer') and `translate' fields
7857 are set in BUFP on entry.
7859 We call regex_compile to do the actual compilation. */
7862 re_compile_pattern (pattern, length, bufp)
7863 const char *pattern;
7865 struct re_pattern_buffer *bufp;
7869 /* GNU code is written to assume at least RE_NREGS registers will be set
7870 (and at least one extra will be -1). */
7871 bufp->regs_allocated = REGS_UNALLOCATED;
7873 /* And GNU code determines whether or not to get register information
7874 by passing null for the REGS argument to re_match, etc., not by
7878 /* Match anchors at newline. */
7879 bufp->newline_anchor = 1;
7882 if (MB_CUR_MAX != 1)
7883 ret = wcs_regex_compile (pattern, length, re_syntax_options, bufp);
7886 ret = byte_regex_compile (pattern, length, re_syntax_options, bufp);
7890 return gettext (re_error_msgid + re_error_msgid_idx[(int) ret]);
7893 weak_alias (__re_compile_pattern, re_compile_pattern)
7896 /* Entry points compatible with 4.2 BSD regex library. We don't define
7897 them unless specifically requested. */
7899 #if defined _REGEX_RE_COMP || defined _LIBC
7901 /* BSD has one and only one pattern buffer. */
7902 static struct re_pattern_buffer re_comp_buf;
7906 /* Make these definitions weak in libc, so POSIX programs can redefine
7907 these names if they don't use our functions, and still use
7908 regcomp/regexec below without link errors. */
7918 if (!re_comp_buf.buffer)
7919 return gettext ("No previous regular expression");
7923 if (!re_comp_buf.buffer)
7925 re_comp_buf.buffer = (unsigned char *) malloc (200);
7926 if (re_comp_buf.buffer == NULL)
7927 return (char *) gettext (re_error_msgid
7928 + re_error_msgid_idx[(int) REG_ESPACE]);
7929 re_comp_buf.allocated = 200;
7931 re_comp_buf.fastmap = (char *) malloc (1 << BYTEWIDTH);
7932 if (re_comp_buf.fastmap == NULL)
7933 return (char *) gettext (re_error_msgid
7934 + re_error_msgid_idx[(int) REG_ESPACE]);
7937 /* Since `re_exec' always passes NULL for the `regs' argument, we
7938 don't need to initialize the pattern buffer fields which affect it. */
7940 /* Match anchors at newlines. */
7941 re_comp_buf.newline_anchor = 1;
7944 if (MB_CUR_MAX != 1)
7945 ret = wcs_regex_compile (s, strlen (s), re_syntax_options, &re_comp_buf);
7948 ret = byte_regex_compile (s, strlen (s), re_syntax_options, &re_comp_buf);
7953 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
7954 return (char *) gettext (re_error_msgid + re_error_msgid_idx[(int) ret]);
7965 const int len = strlen (s);
7967 0 <= re_search (&re_comp_buf, s, len, 0, len, (struct re_registers *) 0);
7970 #endif /* _REGEX_RE_COMP */
7972 /* POSIX.2 functions. Don't define these for Emacs. */
7976 /* regcomp takes a regular expression as a string and compiles it.
7978 PREG is a regex_t *. We do not expect any fields to be initialized,
7979 since POSIX says we shouldn't. Thus, we set
7981 `buffer' to the compiled pattern;
7982 `used' to the length of the compiled pattern;
7983 `syntax' to RE_SYNTAX_POSIX_EXTENDED if the
7984 REG_EXTENDED bit in CFLAGS is set; otherwise, to
7985 RE_SYNTAX_POSIX_BASIC;
7986 `newline_anchor' to REG_NEWLINE being set in CFLAGS;
7987 `fastmap' to an allocated space for the fastmap;
7988 `fastmap_accurate' to zero;
7989 `re_nsub' to the number of subexpressions in PATTERN.
7991 PATTERN is the address of the pattern string.
7993 CFLAGS is a series of bits which affect compilation.
7995 If REG_EXTENDED is set, we use POSIX extended syntax; otherwise, we
7996 use POSIX basic syntax.
7998 If REG_NEWLINE is set, then . and [^...] don't match newline.
7999 Also, regexec will try a match beginning after every newline.
8001 If REG_ICASE is set, then we considers upper- and lowercase
8002 versions of letters to be equivalent when matching.
8004 If REG_NOSUB is set, then when PREG is passed to regexec, that
8005 routine will report only success or failure, and nothing about the
8008 It returns 0 if it succeeds, nonzero if it doesn't. (See regex.h for
8009 the return codes and their meanings.) */
8012 regcomp (preg, pattern, cflags)
8014 const char *pattern;
8019 = (cflags & REG_EXTENDED) ?
8020 RE_SYNTAX_POSIX_EXTENDED : RE_SYNTAX_POSIX_BASIC;
8022 /* regex_compile will allocate the space for the compiled pattern. */
8024 preg->allocated = 0;
8027 /* Try to allocate space for the fastmap. */
8028 preg->fastmap = (char *) malloc (1 << BYTEWIDTH);
8030 if (cflags & REG_ICASE)
8035 = (RE_TRANSLATE_TYPE) malloc (CHAR_SET_SIZE
8036 * sizeof (*(RE_TRANSLATE_TYPE)0));
8037 if (preg->translate == NULL)
8038 return (int) REG_ESPACE;
8040 /* Map uppercase characters to corresponding lowercase ones. */
8041 for (i = 0; i < CHAR_SET_SIZE; i++)
8042 preg->translate[i] = ISUPPER (i) ? TOLOWER (i) : i;
8045 preg->translate = NULL;
8047 /* If REG_NEWLINE is set, newlines are treated differently. */
8048 if (cflags & REG_NEWLINE)
8049 { /* REG_NEWLINE implies neither . nor [^...] match newline. */
8050 syntax &= ~RE_DOT_NEWLINE;
8051 syntax |= RE_HAT_LISTS_NOT_NEWLINE;
8052 /* It also changes the matching behavior. */
8053 preg->newline_anchor = 1;
8056 preg->newline_anchor = 0;
8058 preg->no_sub = !!(cflags & REG_NOSUB);
8060 /* POSIX says a null character in the pattern terminates it, so we
8061 can use strlen here in compiling the pattern. */
8063 if (MB_CUR_MAX != 1)
8064 ret = wcs_regex_compile (pattern, strlen (pattern), syntax, preg);
8067 ret = byte_regex_compile (pattern, strlen (pattern), syntax, preg);
8069 /* POSIX doesn't distinguish between an unmatched open-group and an
8070 unmatched close-group: both are REG_EPAREN. */
8071 if (ret == REG_ERPAREN) ret = REG_EPAREN;
8073 if (ret == REG_NOERROR && preg->fastmap)
8075 /* Compute the fastmap now, since regexec cannot modify the pattern
8077 if (re_compile_fastmap (preg) == -2)
8079 /* Some error occurred while computing the fastmap, just forget
8081 free (preg->fastmap);
8082 preg->fastmap = NULL;
8089 weak_alias (__regcomp, regcomp)
8093 /* regexec searches for a given pattern, specified by PREG, in the
8096 If NMATCH is zero or REG_NOSUB was set in the cflags argument to
8097 `regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at
8098 least NMATCH elements, and we set them to the offsets of the
8099 corresponding matched substrings.
8101 EFLAGS specifies `execution flags' which affect matching: if
8102 REG_NOTBOL is set, then ^ does not match at the beginning of the
8103 string; if REG_NOTEOL is set, then $ does not match at the end.
8105 We return 0 if we find a match and REG_NOMATCH if not. */
8108 regexec (preg, string, nmatch, pmatch, eflags)
8109 const regex_t *preg;
8112 regmatch_t pmatch[];
8116 struct re_registers regs;
8117 regex_t private_preg;
8118 int len = strlen (string);
8119 boolean want_reg_info = !preg->no_sub && nmatch > 0;
8121 private_preg = *preg;
8123 private_preg.not_bol = !!(eflags & REG_NOTBOL);
8124 private_preg.not_eol = !!(eflags & REG_NOTEOL);
8126 /* The user has told us exactly how many registers to return
8127 information about, via `nmatch'. We have to pass that on to the
8128 matching routines. */
8129 private_preg.regs_allocated = REGS_FIXED;
8133 regs.num_regs = nmatch;
8134 regs.start = TALLOC (nmatch * 2, regoff_t);
8135 if (regs.start == NULL)
8136 return (int) REG_NOMATCH;
8137 regs.end = regs.start + nmatch;
8140 /* Perform the searching operation. */
8141 ret = re_search (&private_preg, string, len,
8142 /* start: */ 0, /* range: */ len,
8143 want_reg_info ? ®s : (struct re_registers *) 0);
8145 /* Copy the register information to the POSIX structure. */
8152 for (r = 0; r < nmatch; r++)
8154 pmatch[r].rm_so = regs.start[r];
8155 pmatch[r].rm_eo = regs.end[r];
8159 /* If we needed the temporary register info, free the space now. */
8163 /* We want zero return to mean success, unlike `re_search'. */
8164 return ret >= 0 ? (int) REG_NOERROR : (int) REG_NOMATCH;
8167 weak_alias (__regexec, regexec)
8171 /* Returns a message corresponding to an error code, ERRCODE, returned
8172 from either regcomp or regexec. We don't use PREG here. */
8175 regerror (errcode, preg, errbuf, errbuf_size)
8177 const regex_t *preg;
8185 || errcode >= (int) (sizeof (re_error_msgid_idx)
8186 / sizeof (re_error_msgid_idx[0])))
8187 /* Only error codes returned by the rest of the code should be passed
8188 to this routine. If we are given anything else, or if other regex
8189 code generates an invalid error code, then the program has a bug.
8190 Dump core so we can fix it. */
8193 msg = gettext (re_error_msgid + re_error_msgid_idx[errcode]);
8195 msg_size = strlen (msg) + 1; /* Includes the null. */
8197 if (errbuf_size != 0)
8199 if (msg_size > errbuf_size)
8201 #if defined HAVE_MEMPCPY || defined _LIBC
8202 *((char *) __mempcpy (errbuf, msg, errbuf_size - 1)) = '\0';
8204 memcpy (errbuf, msg, errbuf_size - 1);
8205 errbuf[errbuf_size - 1] = 0;
8209 memcpy (errbuf, msg, msg_size);
8215 weak_alias (__regerror, regerror)
8219 /* Free dynamically allocated space used by PREG. */
8225 if (preg->buffer != NULL)
8226 free (preg->buffer);
8227 preg->buffer = NULL;
8229 preg->allocated = 0;
8232 if (preg->fastmap != NULL)
8233 free (preg->fastmap);
8234 preg->fastmap = NULL;
8235 preg->fastmap_accurate = 0;
8237 if (preg->translate != NULL)
8238 free (preg->translate);
8239 preg->translate = NULL;
8242 weak_alias (__regfree, regfree)
8245 #endif /* not emacs */
8247 #endif /* not INSIDE_RECURSION */
8251 #undef STORE_NUMBER_AND_INCR
8252 #undef EXTRACT_NUMBER
8253 #undef EXTRACT_NUMBER_AND_INCR
8255 #undef DEBUG_PRINT_COMPILED_PATTERN
8256 #undef DEBUG_PRINT_DOUBLE_STRING
8258 #undef INIT_FAIL_STACK
8259 #undef RESET_FAIL_STACK
8260 #undef DOUBLE_FAIL_STACK
8261 #undef PUSH_PATTERN_OP
8262 #undef PUSH_FAILURE_POINTER
8263 #undef PUSH_FAILURE_INT
8264 #undef PUSH_FAILURE_ELT
8265 #undef POP_FAILURE_POINTER
8266 #undef POP_FAILURE_INT
8267 #undef POP_FAILURE_ELT
8270 #undef PUSH_FAILURE_POINT
8271 #undef POP_FAILURE_POINT
8273 #undef REG_UNSET_VALUE
8281 #undef INIT_BUF_SIZE
8282 #undef GET_BUFFER_SPACE
8290 #undef EXTEND_BUFFER
8291 #undef GET_UNSIGNED_NUMBER
8292 #undef FREE_STACK_RETURN
8294 # undef POINTER_TO_OFFSET
8295 # undef MATCHING_IN_FRST_STRING
8297 # undef AT_STRINGS_BEG
8298 # undef AT_STRINGS_END
8301 # undef FREE_VARIABLES
8302 # undef NO_HIGHEST_ACTIVE_REG
8303 # undef NO_LOWEST_ACTIVE_REG
8307 # undef COMPILED_BUFFER_VAR
8308 # undef OFFSET_ADDRESS_SIZE
8309 # undef CHAR_CLASS_SIZE
8316 # define DEFINED_ONCE