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, 2002 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 /* This file has been modified for usage in libiberty. It includes "xregex.h"
24 instead of <regex.h>. The "xregex.h" header file renames all external
25 routines with an "x" prefix so they do not collide with the native regex
26 routines or with other components regex routines. */
27 /* AIX requires this to be the first thing in the file. */
28 #if defined _AIX && !defined __GNUC__ && !defined REGEX_MALLOC
35 #ifndef INSIDE_RECURSION
44 # if defined __GNUC__ || (defined __STDC__ && __STDC__)
45 # define PARAMS(args) args
47 # define PARAMS(args) ()
49 #endif /* Not PARAMS. */
51 #ifndef INSIDE_RECURSION
53 # if defined STDC_HEADERS && !defined emacs
56 /* We need this for `regex.h', and perhaps for the Emacs include files. */
57 # include <sys/types.h>
60 # define WIDE_CHAR_SUPPORT (HAVE_WCTYPE_H && HAVE_WCHAR_H && HAVE_BTOWC)
62 /* For platform which support the ISO C amendement 1 functionality we
63 support user defined character classes. */
64 # if defined _LIBC || WIDE_CHAR_SUPPORT
65 /* Solaris 2.5 has a bug: <wchar.h> must be included before <wctype.h>. */
71 /* We have to keep the namespace clean. */
72 # define regfree(preg) __regfree (preg)
73 # define regexec(pr, st, nm, pm, ef) __regexec (pr, st, nm, pm, ef)
74 # define regcomp(preg, pattern, cflags) __regcomp (preg, pattern, cflags)
75 # define regerror(errcode, preg, errbuf, errbuf_size) \
76 __regerror(errcode, preg, errbuf, errbuf_size)
77 # define re_set_registers(bu, re, nu, st, en) \
78 __re_set_registers (bu, re, nu, st, en)
79 # define re_match_2(bufp, string1, size1, string2, size2, pos, regs, stop) \
80 __re_match_2 (bufp, string1, size1, string2, size2, pos, regs, stop)
81 # define re_match(bufp, string, size, pos, regs) \
82 __re_match (bufp, string, size, pos, regs)
83 # define re_search(bufp, string, size, startpos, range, regs) \
84 __re_search (bufp, string, size, startpos, range, regs)
85 # define re_compile_pattern(pattern, length, bufp) \
86 __re_compile_pattern (pattern, length, bufp)
87 # define re_set_syntax(syntax) __re_set_syntax (syntax)
88 # define re_search_2(bufp, st1, s1, st2, s2, startpos, range, regs, stop) \
89 __re_search_2 (bufp, st1, s1, st2, s2, startpos, range, regs, stop)
90 # define re_compile_fastmap(bufp) __re_compile_fastmap (bufp)
92 # define btowc __btowc
94 /* We are also using some library internals. */
95 # include <locale/localeinfo.h>
96 # include <locale/elem-hash.h>
97 # include <langinfo.h>
98 # include <locale/coll-lookup.h>
101 /* This is for other GNU distributions with internationalized messages. */
102 # if (HAVE_LIBINTL_H && ENABLE_NLS) || defined _LIBC
103 # include <libintl.h>
106 # define gettext(msgid) __dcgettext ("libc", msgid, LC_MESSAGES)
109 # define gettext(msgid) (msgid)
112 # ifndef gettext_noop
113 /* This define is so xgettext can find the internationalizable
115 # define gettext_noop(String) String
118 /* The `emacs' switch turns on certain matching commands
119 that make sense only in Emacs. */
126 # else /* not emacs */
128 /* If we are not linking with Emacs proper,
129 we can't use the relocating allocator
130 even if config.h says that we can. */
133 # if defined STDC_HEADERS || defined _LIBC
140 /* When used in Emacs's lib-src, we need to get bzero and bcopy somehow.
141 If nothing else has been done, use the method below. */
142 # ifdef INHIBIT_STRING_HEADER
143 # if !(defined HAVE_BZERO && defined HAVE_BCOPY)
144 # if !defined bzero && !defined bcopy
145 # undef INHIBIT_STRING_HEADER
150 /* This is the normal way of making sure we have a bcopy and a bzero.
151 This is used in most programs--a few other programs avoid this
152 by defining INHIBIT_STRING_HEADER. */
153 # ifndef INHIBIT_STRING_HEADER
154 # if defined HAVE_STRING_H || defined STDC_HEADERS || defined _LIBC
158 # define bzero(s, n) (memset (s, '\0', n), (s))
160 # define bzero(s, n) __bzero (s, n)
164 # include <strings.h>
166 # define memcmp(s1, s2, n) bcmp (s1, s2, n)
169 # define memcpy(d, s, n) (bcopy (s, d, n), (d))
174 /* Define the syntax stuff for \<, \>, etc. */
176 /* This must be nonzero for the wordchar and notwordchar pattern
177 commands in re_match_2. */
182 # ifdef SWITCH_ENUM_BUG
183 # define SWITCH_ENUM_CAST(x) ((int)(x))
185 # define SWITCH_ENUM_CAST(x) (x)
188 # endif /* not emacs */
190 # if defined _LIBC || HAVE_LIMITS_H
195 # define MB_LEN_MAX 1
198 /* Get the interface, including the syntax bits. */
199 # include "xregex.h" /* change for libiberty */
201 /* isalpha etc. are used for the character classes. */
204 /* Jim Meyering writes:
206 "... Some ctype macros are valid only for character codes that
207 isascii says are ASCII (SGI's IRIX-4.0.5 is one such system --when
208 using /bin/cc or gcc but without giving an ansi option). So, all
209 ctype uses should be through macros like ISPRINT... If
210 STDC_HEADERS is defined, then autoconf has verified that the ctype
211 macros don't need to be guarded with references to isascii. ...
212 Defining isascii to 1 should let any compiler worth its salt
213 eliminate the && through constant folding."
214 Solaris defines some of these symbols so we must undefine them first. */
217 # if defined STDC_HEADERS || (!defined isascii && !defined HAVE_ISASCII)
218 # define ISASCII(c) 1
220 # define ISASCII(c) isascii(c)
224 # define ISBLANK(c) (ISASCII (c) && isblank (c))
226 # define ISBLANK(c) ((c) == ' ' || (c) == '\t')
229 # define ISGRAPH(c) (ISASCII (c) && isgraph (c))
231 # define ISGRAPH(c) (ISASCII (c) && isprint (c) && !isspace (c))
235 # define ISPRINT(c) (ISASCII (c) && isprint (c))
236 # define ISDIGIT(c) (ISASCII (c) && isdigit (c))
237 # define ISALNUM(c) (ISASCII (c) && isalnum (c))
238 # define ISALPHA(c) (ISASCII (c) && isalpha (c))
239 # define ISCNTRL(c) (ISASCII (c) && iscntrl (c))
240 # define ISLOWER(c) (ISASCII (c) && islower (c))
241 # define ISPUNCT(c) (ISASCII (c) && ispunct (c))
242 # define ISSPACE(c) (ISASCII (c) && isspace (c))
243 # define ISUPPER(c) (ISASCII (c) && isupper (c))
244 # define ISXDIGIT(c) (ISASCII (c) && isxdigit (c))
247 # define TOLOWER(c) _tolower(c)
249 # define TOLOWER(c) tolower(c)
253 # define NULL (void *)0
256 /* We remove any previous definition of `SIGN_EXTEND_CHAR',
257 since ours (we hope) works properly with all combinations of
258 machines, compilers, `char' and `unsigned char' argument types.
259 (Per Bothner suggested the basic approach.) */
260 # undef SIGN_EXTEND_CHAR
262 # define SIGN_EXTEND_CHAR(c) ((signed char) (c))
263 # else /* not __STDC__ */
264 /* As in Harbison and Steele. */
265 # define SIGN_EXTEND_CHAR(c) ((((unsigned char) (c)) ^ 128) - 128)
269 /* How many characters in the character set. */
270 # define CHAR_SET_SIZE 256
274 extern char *re_syntax_table
;
276 # else /* not SYNTAX_TABLE */
278 static char re_syntax_table
[CHAR_SET_SIZE
];
280 static void init_syntax_once
PARAMS ((void));
290 bzero (re_syntax_table
, sizeof re_syntax_table
);
292 for (c
= 0; c
< CHAR_SET_SIZE
; ++c
)
294 re_syntax_table
[c
] = Sword
;
296 re_syntax_table
['_'] = Sword
;
301 # endif /* not SYNTAX_TABLE */
303 # define SYNTAX(c) re_syntax_table[(unsigned char) (c)]
307 /* Integer type for pointers. */
308 # if !defined _LIBC && !defined HAVE_UINTPTR_T
309 typedef unsigned long int uintptr_t;
312 /* Should we use malloc or alloca? If REGEX_MALLOC is not defined, we
313 use `alloca' instead of `malloc'. This is because using malloc in
314 re_search* or re_match* could cause memory leaks when C-g is used in
315 Emacs; also, malloc is slower and causes storage fragmentation. On
316 the other hand, malloc is more portable, and easier to debug.
318 Because we sometimes use alloca, some routines have to be macros,
319 not functions -- `alloca'-allocated space disappears at the end of the
320 function it is called in. */
324 # define REGEX_ALLOCATE malloc
325 # define REGEX_REALLOCATE(source, osize, nsize) realloc (source, nsize)
326 # define REGEX_FREE free
328 # else /* not REGEX_MALLOC */
330 /* Emacs already defines alloca, sometimes. */
333 /* Make alloca work the best possible way. */
335 # define alloca __builtin_alloca
336 # else /* not __GNUC__ */
339 # endif /* HAVE_ALLOCA_H */
340 # endif /* not __GNUC__ */
342 # endif /* not alloca */
344 # define REGEX_ALLOCATE alloca
346 /* Assumes a `char *destination' variable. */
347 # define REGEX_REALLOCATE(source, osize, nsize) \
348 (destination = (char *) alloca (nsize), \
349 memcpy (destination, source, osize))
351 /* No need to do anything to free, after alloca. */
352 # define REGEX_FREE(arg) ((void)0) /* Do nothing! But inhibit gcc warning. */
354 # endif /* not REGEX_MALLOC */
356 /* Define how to allocate the failure stack. */
358 # if defined REL_ALLOC && defined REGEX_MALLOC
360 # define REGEX_ALLOCATE_STACK(size) \
361 r_alloc (&failure_stack_ptr, (size))
362 # define REGEX_REALLOCATE_STACK(source, osize, nsize) \
363 r_re_alloc (&failure_stack_ptr, (nsize))
364 # define REGEX_FREE_STACK(ptr) \
365 r_alloc_free (&failure_stack_ptr)
367 # else /* not using relocating allocator */
371 # define REGEX_ALLOCATE_STACK malloc
372 # define REGEX_REALLOCATE_STACK(source, osize, nsize) realloc (source, nsize)
373 # define REGEX_FREE_STACK free
375 # else /* not REGEX_MALLOC */
377 # define REGEX_ALLOCATE_STACK alloca
379 # define REGEX_REALLOCATE_STACK(source, osize, nsize) \
380 REGEX_REALLOCATE (source, osize, nsize)
381 /* No need to explicitly free anything. */
382 # define REGEX_FREE_STACK(arg)
384 # endif /* not REGEX_MALLOC */
385 # endif /* not using relocating allocator */
388 /* True if `size1' is non-NULL and PTR is pointing anywhere inside
389 `string1' or just past its end. This works if PTR is NULL, which is
391 # define FIRST_STRING_P(ptr) \
392 (size1 && string1 <= (ptr) && (ptr) <= string1 + size1)
394 /* (Re)Allocate N items of type T using malloc, or fail. */
395 # define TALLOC(n, t) ((t *) malloc ((n) * sizeof (t)))
396 # define RETALLOC(addr, n, t) ((addr) = (t *) realloc (addr, (n) * sizeof (t)))
397 # define RETALLOC_IF(addr, n, t) \
398 if (addr) RETALLOC((addr), (n), t); else (addr) = TALLOC ((n), t)
399 # define REGEX_TALLOC(n, t) ((t *) REGEX_ALLOCATE ((n) * sizeof (t)))
401 # define BYTEWIDTH 8 /* In bits. */
403 # define STREQ(s1, s2) ((strcmp (s1, s2) == 0))
407 # define MAX(a, b) ((a) > (b) ? (a) : (b))
408 # define MIN(a, b) ((a) < (b) ? (a) : (b))
410 typedef char boolean
;
414 static reg_errcode_t byte_regex_compile
_RE_ARGS ((const char *pattern
, size_t size
,
416 struct re_pattern_buffer
*bufp
));
418 static int byte_re_match_2_internal
PARAMS ((struct re_pattern_buffer
*bufp
,
419 const char *string1
, int size1
,
420 const char *string2
, int size2
,
422 struct re_registers
*regs
,
424 static int byte_re_search_2
PARAMS ((struct re_pattern_buffer
*bufp
,
425 const char *string1
, int size1
,
426 const char *string2
, int size2
,
427 int startpos
, int range
,
428 struct re_registers
*regs
, int stop
));
429 static int byte_re_compile_fastmap
PARAMS ((struct re_pattern_buffer
*bufp
));
432 static reg_errcode_t wcs_regex_compile
_RE_ARGS ((const char *pattern
, size_t size
,
434 struct re_pattern_buffer
*bufp
));
437 static int wcs_re_match_2_internal
PARAMS ((struct re_pattern_buffer
*bufp
,
438 const char *cstring1
, int csize1
,
439 const char *cstring2
, int csize2
,
441 struct re_registers
*regs
,
443 wchar_t *string1
, int size1
,
444 wchar_t *string2
, int size2
,
445 int *mbs_offset1
, int *mbs_offset2
));
446 static int wcs_re_search_2
PARAMS ((struct re_pattern_buffer
*bufp
,
447 const char *string1
, int size1
,
448 const char *string2
, int size2
,
449 int startpos
, int range
,
450 struct re_registers
*regs
, int stop
));
451 static int wcs_re_compile_fastmap
PARAMS ((struct re_pattern_buffer
*bufp
));
454 /* These are the command codes that appear in compiled regular
455 expressions. Some opcodes are followed by argument bytes. A
456 command code can specify any interpretation whatsoever for its
457 arguments. Zero bytes may appear in the compiled regular expression. */
463 /* Succeed right away--no more backtracking. */
466 /* Followed by one byte giving n, then by n literal bytes. */
470 /* Same as exactn, but contains binary data. */
474 /* Matches any (more or less) character. */
477 /* Matches any one char belonging to specified set. First
478 following byte is number of bitmap bytes. Then come bytes
479 for a bitmap saying which chars are in. Bits in each byte
480 are ordered low-bit-first. A character is in the set if its
481 bit is 1. A character too large to have a bit in the map is
482 automatically not in the set. */
483 /* ifdef MBS_SUPPORT, following element is length of character
484 classes, length of collating symbols, length of equivalence
485 classes, length of character ranges, and length of characters.
486 Next, character class element, collating symbols elements,
487 equivalence class elements, range elements, and character
489 See regex_compile function. */
492 /* Same parameters as charset, but match any character that is
493 not one of those specified. */
496 /* Start remembering the text that is matched, for storing in a
497 register. Followed by one byte with the register number, in
498 the range 0 to one less than the pattern buffer's re_nsub
499 field. Then followed by one byte with the number of groups
500 inner to this one. (This last has to be part of the
501 start_memory only because we need it in the on_failure_jump
505 /* Stop remembering the text that is matched and store it in a
506 memory register. Followed by one byte with the register
507 number, in the range 0 to one less than `re_nsub' in the
508 pattern buffer, and one byte with the number of inner groups,
509 just like `start_memory'. (We need the number of inner
510 groups here because we don't have any easy way of finding the
511 corresponding start_memory when we're at a stop_memory.) */
514 /* Match a duplicate of something remembered. Followed by one
515 byte containing the register number. */
518 /* Fail unless at beginning of line. */
521 /* Fail unless at end of line. */
524 /* Succeeds if at beginning of buffer (if emacs) or at beginning
525 of string to be matched (if not). */
528 /* Analogously, for end of buffer/string. */
531 /* Followed by two byte relative address to which to jump. */
534 /* Same as jump, but marks the end of an alternative. */
537 /* Followed by two-byte relative address of place to resume at
538 in case of failure. */
539 /* ifdef MBS_SUPPORT, the size of address is 1. */
542 /* Like on_failure_jump, but pushes a placeholder instead of the
543 current string position when executed. */
544 on_failure_keep_string_jump
,
546 /* Throw away latest failure point and then jump to following
547 two-byte relative address. */
548 /* ifdef MBS_SUPPORT, the size of address is 1. */
551 /* Change to pop_failure_jump if know won't have to backtrack to
552 match; otherwise change to jump. This is used to jump
553 back to the beginning of a repeat. If what follows this jump
554 clearly won't match what the repeat does, such that we can be
555 sure that there is no use backtracking out of repetitions
556 already matched, then we change it to a pop_failure_jump.
557 Followed by two-byte address. */
558 /* ifdef MBS_SUPPORT, the size of address is 1. */
561 /* Jump to following two-byte address, and push a dummy failure
562 point. This failure point will be thrown away if an attempt
563 is made to use it for a failure. A `+' construct makes this
564 before the first repeat. Also used as an intermediary kind
565 of jump when compiling an alternative. */
566 /* ifdef MBS_SUPPORT, the size of address is 1. */
569 /* Push a dummy failure point and continue. Used at the end of
573 /* Followed by two-byte relative address and two-byte number n.
574 After matching N times, jump to the address upon failure. */
575 /* ifdef MBS_SUPPORT, the size of address is 1. */
578 /* Followed by two-byte relative address, and two-byte number n.
579 Jump to the address N times, then fail. */
580 /* ifdef MBS_SUPPORT, the size of address is 1. */
583 /* Set the following two-byte relative address to the
584 subsequent two-byte number. The address *includes* the two
586 /* ifdef MBS_SUPPORT, the size of address is 1. */
589 wordchar
, /* Matches any word-constituent character. */
590 notwordchar
, /* Matches any char that is not a word-constituent. */
592 wordbeg
, /* Succeeds if at word beginning. */
593 wordend
, /* Succeeds if at word end. */
595 wordbound
, /* Succeeds if at a word boundary. */
596 notwordbound
/* Succeeds if not at a word boundary. */
599 ,before_dot
, /* Succeeds if before point. */
600 at_dot
, /* Succeeds if at point. */
601 after_dot
, /* Succeeds if after point. */
603 /* Matches any character whose syntax is specified. Followed by
604 a byte which contains a syntax code, e.g., Sword. */
607 /* Matches any character whose syntax is not that specified. */
611 #endif /* not INSIDE_RECURSION */
616 # define UCHAR_T unsigned char
617 # define COMPILED_BUFFER_VAR bufp->buffer
618 # define OFFSET_ADDRESS_SIZE 2
619 # if defined (__STDC__) || defined (ALMOST_STDC) || defined (HAVE_STRINGIZE)
620 # define PREFIX(name) byte_##name
622 # define PREFIX(name) byte_/**/name
624 # define ARG_PREFIX(name) name
625 # define PUT_CHAR(c) putchar (c)
628 # define CHAR_T wchar_t
629 # define UCHAR_T wchar_t
630 # define COMPILED_BUFFER_VAR wc_buffer
631 # define OFFSET_ADDRESS_SIZE 1 /* the size which STORE_NUMBER macro use */
632 # define CHAR_CLASS_SIZE ((__alignof__(wctype_t)+sizeof(wctype_t))/sizeof(CHAR_T)+1)
633 # if defined (__STDC__) || defined (ALMOST_STDC) || defined (HAVE_STRINGIZE)
634 # define PREFIX(name) wcs_##name
635 # define ARG_PREFIX(name) c##name
637 # define PREFIX(name) wcs_/**/name
638 # define ARG_PREFIX(name) c/**/name
640 /* Should we use wide stream?? */
641 # define PUT_CHAR(c) printf ("%C", c);
647 # define INSIDE_RECURSION
649 # undef INSIDE_RECURSION
652 # define INSIDE_RECURSION
654 # undef INSIDE_RECURSION
658 #ifdef INSIDE_RECURSION
659 /* Common operations on the compiled pattern. */
661 /* Store NUMBER in two contiguous bytes starting at DESTINATION. */
662 /* ifdef MBS_SUPPORT, we store NUMBER in 1 element. */
665 # define STORE_NUMBER(destination, number) \
667 *(destination) = (UCHAR_T)(number); \
670 # define STORE_NUMBER(destination, number) \
672 (destination)[0] = (number) & 0377; \
673 (destination)[1] = (number) >> 8; \
677 /* Same as STORE_NUMBER, except increment DESTINATION to
678 the byte after where the number is stored. Therefore, DESTINATION
679 must be an lvalue. */
680 /* ifdef MBS_SUPPORT, we store NUMBER in 1 element. */
682 # define STORE_NUMBER_AND_INCR(destination, number) \
684 STORE_NUMBER (destination, number); \
685 (destination) += OFFSET_ADDRESS_SIZE; \
688 /* Put into DESTINATION a number stored in two contiguous bytes starting
690 /* ifdef MBS_SUPPORT, we store NUMBER in 1 element. */
693 # define EXTRACT_NUMBER(destination, source) \
695 (destination) = *(source); \
698 # define EXTRACT_NUMBER(destination, source) \
700 (destination) = *(source) & 0377; \
701 (destination) += SIGN_EXTEND_CHAR (*((source) + 1)) << 8; \
706 static void PREFIX(extract_number
) _RE_ARGS ((int *dest
, UCHAR_T
*source
));
708 PREFIX(extract_number
) (dest
, source
)
715 int temp
= SIGN_EXTEND_CHAR (*(source
+ 1));
716 *dest
= *source
& 0377;
721 # ifndef EXTRACT_MACROS /* To debug the macros. */
722 # undef EXTRACT_NUMBER
723 # define EXTRACT_NUMBER(dest, src) PREFIX(extract_number) (&dest, src)
724 # endif /* not EXTRACT_MACROS */
728 /* Same as EXTRACT_NUMBER, except increment SOURCE to after the number.
729 SOURCE must be an lvalue. */
731 # define EXTRACT_NUMBER_AND_INCR(destination, source) \
733 EXTRACT_NUMBER (destination, source); \
734 (source) += OFFSET_ADDRESS_SIZE; \
738 static void PREFIX(extract_number_and_incr
) _RE_ARGS ((int *destination
,
741 PREFIX(extract_number_and_incr
) (destination
, source
)
745 PREFIX(extract_number
) (destination
, *source
);
746 *source
+= OFFSET_ADDRESS_SIZE
;
749 # ifndef EXTRACT_MACROS
750 # undef EXTRACT_NUMBER_AND_INCR
751 # define EXTRACT_NUMBER_AND_INCR(dest, src) \
752 PREFIX(extract_number_and_incr) (&dest, &src)
753 # endif /* not EXTRACT_MACROS */
759 /* If DEBUG is defined, Regex prints many voluminous messages about what
760 it is doing (if the variable `debug' is nonzero). If linked with the
761 main program in `iregex.c', you can enter patterns and strings
762 interactively. And if linked with the main program in `main.c' and
763 the other test files, you can run the already-written tests. */
767 # ifndef DEFINED_ONCE
769 /* We use standard I/O for debugging. */
772 /* It is useful to test things that ``must'' be true when debugging. */
777 # define DEBUG_STATEMENT(e) e
778 # define DEBUG_PRINT1(x) if (debug) printf (x)
779 # define DEBUG_PRINT2(x1, x2) if (debug) printf (x1, x2)
780 # define DEBUG_PRINT3(x1, x2, x3) if (debug) printf (x1, x2, x3)
781 # define DEBUG_PRINT4(x1, x2, x3, x4) if (debug) printf (x1, x2, x3, x4)
782 # endif /* not DEFINED_ONCE */
784 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) \
785 if (debug) PREFIX(print_partial_compiled_pattern) (s, e)
786 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) \
787 if (debug) PREFIX(print_double_string) (w, s1, sz1, s2, sz2)
790 /* Print the fastmap in human-readable form. */
792 # ifndef DEFINED_ONCE
794 print_fastmap (fastmap
)
797 unsigned was_a_range
= 0;
800 while (i
< (1 << BYTEWIDTH
))
806 while (i
< (1 << BYTEWIDTH
) && fastmap
[i
])
820 # endif /* not DEFINED_ONCE */
823 /* Print a compiled pattern string in human-readable form, starting at
824 the START pointer into it and ending just before the pointer END. */
827 PREFIX(print_partial_compiled_pattern
) (start
, end
)
842 /* Loop over pattern commands. */
846 printf ("%td:\t", p
- start
);
848 printf ("%ld:\t", (long int) (p
- start
));
851 switch ((re_opcode_t
) *p
++)
859 printf ("/exactn/%d", mcnt
);
871 printf ("/exactn_bin/%d", mcnt
);
874 printf("/%lx", (long int) *p
++);
878 # endif /* MBS_SUPPORT */
882 printf ("/start_memory/%d/%ld", mcnt
, (long int) *p
++);
887 printf ("/stop_memory/%d/%ld", mcnt
, (long int) *p
++);
891 printf ("/duplicate/%ld", (long int) *p
++);
904 printf ("/charset [%s",
905 (re_opcode_t
) *(workp
- 1) == charset_not
? "^" : "");
907 length
= *workp
++; /* the length of char_classes */
908 for (i
=0 ; i
<length
; i
++)
909 printf("[:%lx:]", (long int) *p
++);
910 length
= *workp
++; /* the length of collating_symbol */
911 for (i
=0 ; i
<length
;)
915 PUT_CHAR((i
++,*p
++));
919 length
= *workp
++; /* the length of equivalence_class */
920 for (i
=0 ; i
<length
;)
924 PUT_CHAR((i
++,*p
++));
928 length
= *workp
++; /* the length of char_range */
929 for (i
=0 ; i
<length
; i
++)
931 wchar_t range_start
= *p
++;
932 wchar_t range_end
= *p
++;
933 printf("%C-%C", range_start
, range_end
);
935 length
= *workp
++; /* the length of char */
936 for (i
=0 ; i
<length
; i
++)
940 register int c
, last
= -100;
941 register int in_range
= 0;
943 printf ("/charset [%s",
944 (re_opcode_t
) *(p
- 1) == charset_not
? "^" : "");
946 assert (p
+ *p
< pend
);
948 for (c
= 0; c
< 256; c
++)
950 && (p
[1 + (c
/8)] & (1 << (c
% 8))))
952 /* Are we starting a range? */
953 if (last
+ 1 == c
&& ! in_range
)
958 /* Have we broken a range? */
959 else if (last
+ 1 != c
&& in_range
)
989 case on_failure_jump
:
990 PREFIX(extract_number_and_incr
) (&mcnt
, &p
);
992 printf ("/on_failure_jump to %td", p
+ mcnt
- start
);
994 printf ("/on_failure_jump to %ld", (long int) (p
+ mcnt
- start
));
998 case on_failure_keep_string_jump
:
999 PREFIX(extract_number_and_incr
) (&mcnt
, &p
);
1001 printf ("/on_failure_keep_string_jump to %td", p
+ mcnt
- start
);
1003 printf ("/on_failure_keep_string_jump to %ld",
1004 (long int) (p
+ mcnt
- start
));
1008 case dummy_failure_jump
:
1009 PREFIX(extract_number_and_incr
) (&mcnt
, &p
);
1011 printf ("/dummy_failure_jump to %td", p
+ mcnt
- start
);
1013 printf ("/dummy_failure_jump to %ld", (long int) (p
+ mcnt
- start
));
1017 case push_dummy_failure
:
1018 printf ("/push_dummy_failure");
1021 case maybe_pop_jump
:
1022 PREFIX(extract_number_and_incr
) (&mcnt
, &p
);
1024 printf ("/maybe_pop_jump to %td", p
+ mcnt
- start
);
1026 printf ("/maybe_pop_jump to %ld", (long int) (p
+ mcnt
- start
));
1030 case pop_failure_jump
:
1031 PREFIX(extract_number_and_incr
) (&mcnt
, &p
);
1033 printf ("/pop_failure_jump to %td", p
+ mcnt
- start
);
1035 printf ("/pop_failure_jump to %ld", (long int) (p
+ mcnt
- start
));
1040 PREFIX(extract_number_and_incr
) (&mcnt
, &p
);
1042 printf ("/jump_past_alt to %td", p
+ mcnt
- start
);
1044 printf ("/jump_past_alt to %ld", (long int) (p
+ mcnt
- start
));
1049 PREFIX(extract_number_and_incr
) (&mcnt
, &p
);
1051 printf ("/jump to %td", p
+ mcnt
- start
);
1053 printf ("/jump to %ld", (long int) (p
+ mcnt
- start
));
1058 PREFIX(extract_number_and_incr
) (&mcnt
, &p
);
1060 PREFIX(extract_number_and_incr
) (&mcnt2
, &p
);
1062 printf ("/succeed_n to %td, %d times", p1
- start
, mcnt2
);
1064 printf ("/succeed_n to %ld, %d times",
1065 (long int) (p1
- start
), mcnt2
);
1070 PREFIX(extract_number_and_incr
) (&mcnt
, &p
);
1072 PREFIX(extract_number_and_incr
) (&mcnt2
, &p
);
1073 printf ("/jump_n to %d, %d times", p1
- start
, mcnt2
);
1077 PREFIX(extract_number_and_incr
) (&mcnt
, &p
);
1079 PREFIX(extract_number_and_incr
) (&mcnt2
, &p
);
1081 printf ("/set_number_at location %td to %d", p1
- start
, mcnt2
);
1083 printf ("/set_number_at location %ld to %d",
1084 (long int) (p1
- start
), mcnt2
);
1089 printf ("/wordbound");
1093 printf ("/notwordbound");
1097 printf ("/wordbeg");
1101 printf ("/wordend");
1106 printf ("/before_dot");
1114 printf ("/after_dot");
1118 printf ("/syntaxspec");
1120 printf ("/%d", mcnt
);
1124 printf ("/notsyntaxspec");
1126 printf ("/%d", mcnt
);
1131 printf ("/wordchar");
1135 printf ("/notwordchar");
1147 printf ("?%ld", (long int) *(p
-1));
1154 printf ("%td:\tend of pattern.\n", p
- start
);
1156 printf ("%ld:\tend of pattern.\n", (long int) (p
- start
));
1162 PREFIX(print_compiled_pattern
) (bufp
)
1163 struct re_pattern_buffer
*bufp
;
1165 UCHAR_T
*buffer
= (UCHAR_T
*) bufp
->buffer
;
1167 PREFIX(print_partial_compiled_pattern
) (buffer
, buffer
1168 + bufp
->used
/ sizeof(UCHAR_T
));
1169 printf ("%ld bytes used/%ld bytes allocated.\n",
1170 bufp
->used
, bufp
->allocated
);
1172 if (bufp
->fastmap_accurate
&& bufp
->fastmap
)
1174 printf ("fastmap: ");
1175 print_fastmap (bufp
->fastmap
);
1179 printf ("re_nsub: %Zd\t", bufp
->re_nsub
);
1181 printf ("re_nsub: %ld\t", (long int) bufp
->re_nsub
);
1183 printf ("regs_alloc: %d\t", bufp
->regs_allocated
);
1184 printf ("can_be_null: %d\t", bufp
->can_be_null
);
1185 printf ("newline_anchor: %d\n", bufp
->newline_anchor
);
1186 printf ("no_sub: %d\t", bufp
->no_sub
);
1187 printf ("not_bol: %d\t", bufp
->not_bol
);
1188 printf ("not_eol: %d\t", bufp
->not_eol
);
1189 printf ("syntax: %lx\n", bufp
->syntax
);
1190 /* Perhaps we should print the translate table? */
1195 PREFIX(print_double_string
) (where
, string1
, size1
, string2
, size2
)
1196 const CHAR_T
*where
;
1197 const CHAR_T
*string1
;
1198 const CHAR_T
*string2
;
1210 if (FIRST_STRING_P (where
))
1212 for (this_char
= where
- string1
; this_char
< size1
; this_char
++)
1213 PUT_CHAR (string1
[this_char
]);
1219 for (this_char
= where
- string2
; this_char
< size2
; this_char
++)
1221 PUT_CHAR (string2
[this_char
]);
1224 fputs ("...", stdout
);
1231 # ifndef DEFINED_ONCE
1240 # else /* not DEBUG */
1242 # ifndef DEFINED_ONCE
1246 # define DEBUG_STATEMENT(e)
1247 # define DEBUG_PRINT1(x)
1248 # define DEBUG_PRINT2(x1, x2)
1249 # define DEBUG_PRINT3(x1, x2, x3)
1250 # define DEBUG_PRINT4(x1, x2, x3, x4)
1251 # endif /* not DEFINED_ONCE */
1252 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e)
1253 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2)
1255 # endif /* not DEBUG */
1260 /* This convert a multibyte string to a wide character string.
1261 And write their correspondances to offset_buffer(see below)
1262 and write whether each wchar_t is binary data to is_binary.
1263 This assume invalid multibyte sequences as binary data.
1264 We assume offset_buffer and is_binary is already allocated
1267 static size_t convert_mbs_to_wcs (CHAR_T
*dest
, const unsigned char* src
,
1268 size_t len
, int *offset_buffer
,
1271 convert_mbs_to_wcs (dest
, src
, len
, offset_buffer
, is_binary
)
1273 const unsigned char* src
;
1274 size_t len
; /* the length of multibyte string. */
1276 /* It hold correspondances between src(char string) and
1277 dest(wchar_t string) for optimization.
1279 dest = {'X', 'Y', 'Z'}
1280 (each "xxx", "y" and "zz" represent one multibyte character
1281 corresponding to 'X', 'Y' and 'Z'.)
1282 offset_buffer = {0, 0+3("xxx"), 0+3+1("y"), 0+3+1+2("zz")}
1288 wchar_t *pdest
= dest
;
1289 const unsigned char *psrc
= src
;
1290 size_t wc_count
= 0;
1294 size_t mb_remain
= len
;
1295 size_t mb_count
= 0;
1297 /* Initialize the conversion state. */
1298 memset (&mbs
, 0, sizeof (mbstate_t));
1300 offset_buffer
[0] = 0;
1301 for( ; mb_remain
> 0 ; ++wc_count
, ++pdest
, mb_remain
-= consumed
,
1305 consumed
= __mbrtowc (pdest
, psrc
, mb_remain
, &mbs
);
1307 consumed
= mbrtowc (pdest
, psrc
, mb_remain
, &mbs
);
1311 /* failed to convert. maybe src contains binary data.
1312 So we consume 1 byte manualy. */
1316 is_binary
[wc_count
] = TRUE
;
1319 is_binary
[wc_count
] = FALSE
;
1320 /* In sjis encoding, we use yen sign as escape character in
1321 place of reverse solidus. So we convert 0x5c(yen sign in
1322 sjis) to not 0xa5(yen sign in UCS2) but 0x5c(reverse
1323 solidus in UCS2). */
1324 if (consumed
== 1 && (int) *psrc
== 0x5c && (int) *pdest
== 0xa5)
1325 *pdest
= (wchar_t) *psrc
;
1327 offset_buffer
[wc_count
+ 1] = mb_count
+= consumed
;
1330 /* Fill remain of the buffer with sentinel. */
1331 for (i
= wc_count
+ 1 ; i
<= len
; i
++)
1332 offset_buffer
[i
] = mb_count
+ 1;
1339 #else /* not INSIDE_RECURSION */
1341 /* Set by `re_set_syntax' to the current regexp syntax to recognize. Can
1342 also be assigned to arbitrarily: each pattern buffer stores its own
1343 syntax, so it can be changed between regex compilations. */
1344 /* This has no initializer because initialized variables in Emacs
1345 become read-only after dumping. */
1346 reg_syntax_t re_syntax_options
;
1349 /* Specify the precise syntax of regexps for compilation. This provides
1350 for compatibility for various utilities which historically have
1351 different, incompatible syntaxes.
1353 The argument SYNTAX is a bit mask comprised of the various bits
1354 defined in regex.h. We return the old syntax. */
1357 re_set_syntax (syntax
)
1358 reg_syntax_t syntax
;
1360 reg_syntax_t ret
= re_syntax_options
;
1362 re_syntax_options
= syntax
;
1364 if (syntax
& RE_DEBUG
)
1366 else if (debug
) /* was on but now is not */
1372 weak_alias (__re_set_syntax
, re_set_syntax
)
1375 /* This table gives an error message for each of the error codes listed
1376 in regex.h. Obviously the order here has to be same as there.
1377 POSIX doesn't require that we do anything for REG_NOERROR,
1378 but why not be nice? */
1380 static const char *re_error_msgid
[] =
1382 gettext_noop ("Success"), /* REG_NOERROR */
1383 gettext_noop ("No match"), /* REG_NOMATCH */
1384 gettext_noop ("Invalid regular expression"), /* REG_BADPAT */
1385 gettext_noop ("Invalid collation character"), /* REG_ECOLLATE */
1386 gettext_noop ("Invalid character class name"), /* REG_ECTYPE */
1387 gettext_noop ("Trailing backslash"), /* REG_EESCAPE */
1388 gettext_noop ("Invalid back reference"), /* REG_ESUBREG */
1389 gettext_noop ("Unmatched [ or [^"), /* REG_EBRACK */
1390 gettext_noop ("Unmatched ( or \\("), /* REG_EPAREN */
1391 gettext_noop ("Unmatched \\{"), /* REG_EBRACE */
1392 gettext_noop ("Invalid content of \\{\\}"), /* REG_BADBR */
1393 gettext_noop ("Invalid range end"), /* REG_ERANGE */
1394 gettext_noop ("Memory exhausted"), /* REG_ESPACE */
1395 gettext_noop ("Invalid preceding regular expression"), /* REG_BADRPT */
1396 gettext_noop ("Premature end of regular expression"), /* REG_EEND */
1397 gettext_noop ("Regular expression too big"), /* REG_ESIZE */
1398 gettext_noop ("Unmatched ) or \\)") /* REG_ERPAREN */
1401 #endif /* INSIDE_RECURSION */
1403 #ifndef DEFINED_ONCE
1404 /* Avoiding alloca during matching, to placate r_alloc. */
1406 /* Define MATCH_MAY_ALLOCATE unless we need to make sure that the
1407 searching and matching functions should not call alloca. On some
1408 systems, alloca is implemented in terms of malloc, and if we're
1409 using the relocating allocator routines, then malloc could cause a
1410 relocation, which might (if the strings being searched are in the
1411 ralloc heap) shift the data out from underneath the regexp
1414 Here's another reason to avoid allocation: Emacs
1415 processes input from X in a signal handler; processing X input may
1416 call malloc; if input arrives while a matching routine is calling
1417 malloc, then we're scrod. But Emacs can't just block input while
1418 calling matching routines; then we don't notice interrupts when
1419 they come in. So, Emacs blocks input around all regexp calls
1420 except the matching calls, which it leaves unprotected, in the
1421 faith that they will not malloc. */
1423 /* Normally, this is fine. */
1424 # define MATCH_MAY_ALLOCATE
1426 /* When using GNU C, we are not REALLY using the C alloca, no matter
1427 what config.h may say. So don't take precautions for it. */
1432 /* The match routines may not allocate if (1) they would do it with malloc
1433 and (2) it's not safe for them to use malloc.
1434 Note that if REL_ALLOC is defined, matching would not use malloc for the
1435 failure stack, but we would still use it for the register vectors;
1436 so REL_ALLOC should not affect this. */
1437 # if (defined C_ALLOCA || defined REGEX_MALLOC) && defined emacs
1438 # undef MATCH_MAY_ALLOCATE
1440 #endif /* not DEFINED_ONCE */
1442 #ifdef INSIDE_RECURSION
1443 /* Failure stack declarations and macros; both re_compile_fastmap and
1444 re_match_2 use a failure stack. These have to be macros because of
1445 REGEX_ALLOCATE_STACK. */
1448 /* Number of failure points for which to initially allocate space
1449 when matching. If this number is exceeded, we allocate more
1450 space, so it is not a hard limit. */
1451 # ifndef INIT_FAILURE_ALLOC
1452 # define INIT_FAILURE_ALLOC 5
1455 /* Roughly the maximum number of failure points on the stack. Would be
1456 exactly that if always used MAX_FAILURE_ITEMS items each time we failed.
1457 This is a variable only so users of regex can assign to it; we never
1458 change it ourselves. */
1460 # ifdef INT_IS_16BIT
1462 # ifndef DEFINED_ONCE
1463 # if defined MATCH_MAY_ALLOCATE
1464 /* 4400 was enough to cause a crash on Alpha OSF/1,
1465 whose default stack limit is 2mb. */
1466 long int re_max_failures
= 4000;
1468 long int re_max_failures
= 2000;
1472 union PREFIX(fail_stack_elt
)
1478 typedef union PREFIX(fail_stack_elt
) PREFIX(fail_stack_elt_t
);
1482 PREFIX(fail_stack_elt_t
) *stack
;
1483 unsigned long int size
;
1484 unsigned long int avail
; /* Offset of next open position. */
1485 } PREFIX(fail_stack_type
);
1487 # else /* not INT_IS_16BIT */
1489 # ifndef DEFINED_ONCE
1490 # if defined MATCH_MAY_ALLOCATE
1491 /* 4400 was enough to cause a crash on Alpha OSF/1,
1492 whose default stack limit is 2mb. */
1493 int re_max_failures
= 4000;
1495 int re_max_failures
= 2000;
1499 union PREFIX(fail_stack_elt
)
1505 typedef union PREFIX(fail_stack_elt
) PREFIX(fail_stack_elt_t
);
1509 PREFIX(fail_stack_elt_t
) *stack
;
1511 unsigned avail
; /* Offset of next open position. */
1512 } PREFIX(fail_stack_type
);
1514 # endif /* INT_IS_16BIT */
1516 # ifndef DEFINED_ONCE
1517 # define FAIL_STACK_EMPTY() (fail_stack.avail == 0)
1518 # define FAIL_STACK_PTR_EMPTY() (fail_stack_ptr->avail == 0)
1519 # define FAIL_STACK_FULL() (fail_stack.avail == fail_stack.size)
1523 /* Define macros to initialize and free the failure stack.
1524 Do `return -2' if the alloc fails. */
1526 # ifdef MATCH_MAY_ALLOCATE
1527 # define INIT_FAIL_STACK() \
1529 fail_stack.stack = (PREFIX(fail_stack_elt_t) *) \
1530 REGEX_ALLOCATE_STACK (INIT_FAILURE_ALLOC * sizeof (PREFIX(fail_stack_elt_t))); \
1532 if (fail_stack.stack == NULL) \
1535 fail_stack.size = INIT_FAILURE_ALLOC; \
1536 fail_stack.avail = 0; \
1539 # define RESET_FAIL_STACK() REGEX_FREE_STACK (fail_stack.stack)
1541 # define INIT_FAIL_STACK() \
1543 fail_stack.avail = 0; \
1546 # define RESET_FAIL_STACK()
1550 /* Double the size of FAIL_STACK, up to approximately `re_max_failures' items.
1552 Return 1 if succeeds, and 0 if either ran out of memory
1553 allocating space for it or it was already too large.
1555 REGEX_REALLOCATE_STACK requires `destination' be declared. */
1557 # define DOUBLE_FAIL_STACK(fail_stack) \
1558 ((fail_stack).size > (unsigned) (re_max_failures * MAX_FAILURE_ITEMS) \
1560 : ((fail_stack).stack = (PREFIX(fail_stack_elt_t) *) \
1561 REGEX_REALLOCATE_STACK ((fail_stack).stack, \
1562 (fail_stack).size * sizeof (PREFIX(fail_stack_elt_t)), \
1563 ((fail_stack).size << 1) * sizeof (PREFIX(fail_stack_elt_t))),\
1565 (fail_stack).stack == NULL \
1567 : ((fail_stack).size <<= 1, \
1571 /* Push pointer POINTER on FAIL_STACK.
1572 Return 1 if was able to do so and 0 if ran out of memory allocating
1574 # define PUSH_PATTERN_OP(POINTER, FAIL_STACK) \
1575 ((FAIL_STACK_FULL () \
1576 && !DOUBLE_FAIL_STACK (FAIL_STACK)) \
1578 : ((FAIL_STACK).stack[(FAIL_STACK).avail++].pointer = POINTER, \
1581 /* Push a pointer value onto the failure stack.
1582 Assumes the variable `fail_stack'. Probably should only
1583 be called from within `PUSH_FAILURE_POINT'. */
1584 # define PUSH_FAILURE_POINTER(item) \
1585 fail_stack.stack[fail_stack.avail++].pointer = (UCHAR_T *) (item)
1587 /* This pushes an integer-valued item onto the failure stack.
1588 Assumes the variable `fail_stack'. Probably should only
1589 be called from within `PUSH_FAILURE_POINT'. */
1590 # define PUSH_FAILURE_INT(item) \
1591 fail_stack.stack[fail_stack.avail++].integer = (item)
1593 /* Push a fail_stack_elt_t value onto the failure stack.
1594 Assumes the variable `fail_stack'. Probably should only
1595 be called from within `PUSH_FAILURE_POINT'. */
1596 # define PUSH_FAILURE_ELT(item) \
1597 fail_stack.stack[fail_stack.avail++] = (item)
1599 /* These three POP... operations complement the three PUSH... operations.
1600 All assume that `fail_stack' is nonempty. */
1601 # define POP_FAILURE_POINTER() fail_stack.stack[--fail_stack.avail].pointer
1602 # define POP_FAILURE_INT() fail_stack.stack[--fail_stack.avail].integer
1603 # define POP_FAILURE_ELT() fail_stack.stack[--fail_stack.avail]
1605 /* Used to omit pushing failure point id's when we're not debugging. */
1607 # define DEBUG_PUSH PUSH_FAILURE_INT
1608 # define DEBUG_POP(item_addr) *(item_addr) = POP_FAILURE_INT ()
1610 # define DEBUG_PUSH(item)
1611 # define DEBUG_POP(item_addr)
1615 /* Push the information about the state we will need
1616 if we ever fail back to it.
1618 Requires variables fail_stack, regstart, regend, reg_info, and
1619 num_regs_pushed be declared. DOUBLE_FAIL_STACK requires `destination'
1622 Does `return FAILURE_CODE' if runs out of memory. */
1624 # define PUSH_FAILURE_POINT(pattern_place, string_place, failure_code) \
1626 char *destination; \
1627 /* Must be int, so when we don't save any registers, the arithmetic \
1628 of 0 + -1 isn't done as unsigned. */ \
1629 /* Can't be int, since there is not a shred of a guarantee that int \
1630 is wide enough to hold a value of something to which pointer can \
1632 active_reg_t this_reg; \
1634 DEBUG_STATEMENT (failure_id++); \
1635 DEBUG_STATEMENT (nfailure_points_pushed++); \
1636 DEBUG_PRINT2 ("\nPUSH_FAILURE_POINT #%u:\n", failure_id); \
1637 DEBUG_PRINT2 (" Before push, next avail: %d\n", (fail_stack).avail);\
1638 DEBUG_PRINT2 (" size: %d\n", (fail_stack).size);\
1640 DEBUG_PRINT2 (" slots needed: %ld\n", NUM_FAILURE_ITEMS); \
1641 DEBUG_PRINT2 (" available: %d\n", REMAINING_AVAIL_SLOTS); \
1643 /* Ensure we have enough space allocated for what we will push. */ \
1644 while (REMAINING_AVAIL_SLOTS < NUM_FAILURE_ITEMS) \
1646 if (!DOUBLE_FAIL_STACK (fail_stack)) \
1647 return failure_code; \
1649 DEBUG_PRINT2 ("\n Doubled stack; size now: %d\n", \
1650 (fail_stack).size); \
1651 DEBUG_PRINT2 (" slots available: %d\n", REMAINING_AVAIL_SLOTS);\
1654 /* Push the info, starting with the registers. */ \
1655 DEBUG_PRINT1 ("\n"); \
1658 for (this_reg = lowest_active_reg; this_reg <= highest_active_reg; \
1661 DEBUG_PRINT2 (" Pushing reg: %lu\n", this_reg); \
1662 DEBUG_STATEMENT (num_regs_pushed++); \
1664 DEBUG_PRINT2 (" start: %p\n", regstart[this_reg]); \
1665 PUSH_FAILURE_POINTER (regstart[this_reg]); \
1667 DEBUG_PRINT2 (" end: %p\n", regend[this_reg]); \
1668 PUSH_FAILURE_POINTER (regend[this_reg]); \
1670 DEBUG_PRINT2 (" info: %p\n ", \
1671 reg_info[this_reg].word.pointer); \
1672 DEBUG_PRINT2 (" match_null=%d", \
1673 REG_MATCH_NULL_STRING_P (reg_info[this_reg])); \
1674 DEBUG_PRINT2 (" active=%d", IS_ACTIVE (reg_info[this_reg])); \
1675 DEBUG_PRINT2 (" matched_something=%d", \
1676 MATCHED_SOMETHING (reg_info[this_reg])); \
1677 DEBUG_PRINT2 (" ever_matched=%d", \
1678 EVER_MATCHED_SOMETHING (reg_info[this_reg])); \
1679 DEBUG_PRINT1 ("\n"); \
1680 PUSH_FAILURE_ELT (reg_info[this_reg].word); \
1683 DEBUG_PRINT2 (" Pushing low active reg: %ld\n", lowest_active_reg);\
1684 PUSH_FAILURE_INT (lowest_active_reg); \
1686 DEBUG_PRINT2 (" Pushing high active reg: %ld\n", highest_active_reg);\
1687 PUSH_FAILURE_INT (highest_active_reg); \
1689 DEBUG_PRINT2 (" Pushing pattern %p:\n", pattern_place); \
1690 DEBUG_PRINT_COMPILED_PATTERN (bufp, pattern_place, pend); \
1691 PUSH_FAILURE_POINTER (pattern_place); \
1693 DEBUG_PRINT2 (" Pushing string %p: `", string_place); \
1694 DEBUG_PRINT_DOUBLE_STRING (string_place, string1, size1, string2, \
1696 DEBUG_PRINT1 ("'\n"); \
1697 PUSH_FAILURE_POINTER (string_place); \
1699 DEBUG_PRINT2 (" Pushing failure id: %u\n", failure_id); \
1700 DEBUG_PUSH (failure_id); \
1703 # ifndef DEFINED_ONCE
1704 /* This is the number of items that are pushed and popped on the stack
1705 for each register. */
1706 # define NUM_REG_ITEMS 3
1708 /* Individual items aside from the registers. */
1710 # define NUM_NONREG_ITEMS 5 /* Includes failure point id. */
1712 # define NUM_NONREG_ITEMS 4
1715 /* We push at most this many items on the stack. */
1716 /* We used to use (num_regs - 1), which is the number of registers
1717 this regexp will save; but that was changed to 5
1718 to avoid stack overflow for a regexp with lots of parens. */
1719 # define MAX_FAILURE_ITEMS (5 * NUM_REG_ITEMS + NUM_NONREG_ITEMS)
1721 /* We actually push this many items. */
1722 # define NUM_FAILURE_ITEMS \
1724 ? 0 : highest_active_reg - lowest_active_reg + 1) \
1728 /* How many items can still be added to the stack without overflowing it. */
1729 # define REMAINING_AVAIL_SLOTS ((fail_stack).size - (fail_stack).avail)
1730 # endif /* not DEFINED_ONCE */
1733 /* Pops what PUSH_FAIL_STACK pushes.
1735 We restore into the parameters, all of which should be lvalues:
1736 STR -- the saved data position.
1737 PAT -- the saved pattern position.
1738 LOW_REG, HIGH_REG -- the highest and lowest active registers.
1739 REGSTART, REGEND -- arrays of string positions.
1740 REG_INFO -- array of information about each subexpression.
1742 Also assumes the variables `fail_stack' and (if debugging), `bufp',
1743 `pend', `string1', `size1', `string2', and `size2'. */
1744 # define POP_FAILURE_POINT(str, pat, low_reg, high_reg, regstart, regend, reg_info)\
1746 DEBUG_STATEMENT (unsigned failure_id;) \
1747 active_reg_t this_reg; \
1748 const UCHAR_T *string_temp; \
1750 assert (!FAIL_STACK_EMPTY ()); \
1752 /* Remove failure points and point to how many regs pushed. */ \
1753 DEBUG_PRINT1 ("POP_FAILURE_POINT:\n"); \
1754 DEBUG_PRINT2 (" Before pop, next avail: %d\n", fail_stack.avail); \
1755 DEBUG_PRINT2 (" size: %d\n", fail_stack.size); \
1757 assert (fail_stack.avail >= NUM_NONREG_ITEMS); \
1759 DEBUG_POP (&failure_id); \
1760 DEBUG_PRINT2 (" Popping failure id: %u\n", failure_id); \
1762 /* If the saved string location is NULL, it came from an \
1763 on_failure_keep_string_jump opcode, and we want to throw away the \
1764 saved NULL, thus retaining our current position in the string. */ \
1765 string_temp = POP_FAILURE_POINTER (); \
1766 if (string_temp != NULL) \
1767 str = (const CHAR_T *) string_temp; \
1769 DEBUG_PRINT2 (" Popping string %p: `", str); \
1770 DEBUG_PRINT_DOUBLE_STRING (str, string1, size1, string2, size2); \
1771 DEBUG_PRINT1 ("'\n"); \
1773 pat = (UCHAR_T *) POP_FAILURE_POINTER (); \
1774 DEBUG_PRINT2 (" Popping pattern %p:\n", pat); \
1775 DEBUG_PRINT_COMPILED_PATTERN (bufp, pat, pend); \
1777 /* Restore register info. */ \
1778 high_reg = (active_reg_t) POP_FAILURE_INT (); \
1779 DEBUG_PRINT2 (" Popping high active reg: %ld\n", high_reg); \
1781 low_reg = (active_reg_t) POP_FAILURE_INT (); \
1782 DEBUG_PRINT2 (" Popping low active reg: %ld\n", low_reg); \
1785 for (this_reg = high_reg; this_reg >= low_reg; this_reg--) \
1787 DEBUG_PRINT2 (" Popping reg: %ld\n", this_reg); \
1789 reg_info[this_reg].word = POP_FAILURE_ELT (); \
1790 DEBUG_PRINT2 (" info: %p\n", \
1791 reg_info[this_reg].word.pointer); \
1793 regend[this_reg] = (const CHAR_T *) POP_FAILURE_POINTER (); \
1794 DEBUG_PRINT2 (" end: %p\n", regend[this_reg]); \
1796 regstart[this_reg] = (const CHAR_T *) POP_FAILURE_POINTER (); \
1797 DEBUG_PRINT2 (" start: %p\n", regstart[this_reg]); \
1801 for (this_reg = highest_active_reg; this_reg > high_reg; this_reg--) \
1803 reg_info[this_reg].word.integer = 0; \
1804 regend[this_reg] = 0; \
1805 regstart[this_reg] = 0; \
1807 highest_active_reg = high_reg; \
1810 set_regs_matched_done = 0; \
1811 DEBUG_STATEMENT (nfailure_points_popped++); \
1812 } /* POP_FAILURE_POINT */
1814 /* Structure for per-register (a.k.a. per-group) information.
1815 Other register information, such as the
1816 starting and ending positions (which are addresses), and the list of
1817 inner groups (which is a bits list) are maintained in separate
1820 We are making a (strictly speaking) nonportable assumption here: that
1821 the compiler will pack our bit fields into something that fits into
1822 the type of `word', i.e., is something that fits into one item on the
1826 /* Declarations and macros for re_match_2. */
1830 PREFIX(fail_stack_elt_t
) word
;
1833 /* This field is one if this group can match the empty string,
1834 zero if not. If not yet determined, `MATCH_NULL_UNSET_VALUE'. */
1835 # define MATCH_NULL_UNSET_VALUE 3
1836 unsigned match_null_string_p
: 2;
1837 unsigned is_active
: 1;
1838 unsigned matched_something
: 1;
1839 unsigned ever_matched_something
: 1;
1841 } PREFIX(register_info_type
);
1843 # ifndef DEFINED_ONCE
1844 # define REG_MATCH_NULL_STRING_P(R) ((R).bits.match_null_string_p)
1845 # define IS_ACTIVE(R) ((R).bits.is_active)
1846 # define MATCHED_SOMETHING(R) ((R).bits.matched_something)
1847 # define EVER_MATCHED_SOMETHING(R) ((R).bits.ever_matched_something)
1850 /* Call this when have matched a real character; it sets `matched' flags
1851 for the subexpressions which we are currently inside. Also records
1852 that those subexprs have matched. */
1853 # define SET_REGS_MATCHED() \
1856 if (!set_regs_matched_done) \
1859 set_regs_matched_done = 1; \
1860 for (r = lowest_active_reg; r <= highest_active_reg; r++) \
1862 MATCHED_SOMETHING (reg_info[r]) \
1863 = EVER_MATCHED_SOMETHING (reg_info[r]) \
1869 # endif /* not DEFINED_ONCE */
1871 /* Registers are set to a sentinel when they haven't yet matched. */
1872 static CHAR_T
PREFIX(reg_unset_dummy
);
1873 # define REG_UNSET_VALUE (&PREFIX(reg_unset_dummy))
1874 # define REG_UNSET(e) ((e) == REG_UNSET_VALUE)
1876 /* Subroutine declarations and macros for regex_compile. */
1877 static void PREFIX(store_op1
) _RE_ARGS ((re_opcode_t op
, UCHAR_T
*loc
, int arg
));
1878 static void PREFIX(store_op2
) _RE_ARGS ((re_opcode_t op
, UCHAR_T
*loc
,
1879 int arg1
, int arg2
));
1880 static void PREFIX(insert_op1
) _RE_ARGS ((re_opcode_t op
, UCHAR_T
*loc
,
1881 int arg
, UCHAR_T
*end
));
1882 static void PREFIX(insert_op2
) _RE_ARGS ((re_opcode_t op
, UCHAR_T
*loc
,
1883 int arg1
, int arg2
, UCHAR_T
*end
));
1884 static boolean
PREFIX(at_begline_loc_p
) _RE_ARGS ((const CHAR_T
*pattern
,
1886 reg_syntax_t syntax
));
1887 static boolean
PREFIX(at_endline_loc_p
) _RE_ARGS ((const CHAR_T
*p
,
1889 reg_syntax_t syntax
));
1891 static reg_errcode_t wcs_compile_range
_RE_ARGS ((CHAR_T range_start
,
1892 const CHAR_T
**p_ptr
,
1895 reg_syntax_t syntax
,
1898 static void insert_space
_RE_ARGS ((int num
, CHAR_T
*loc
, CHAR_T
*end
));
1900 static reg_errcode_t byte_compile_range
_RE_ARGS ((unsigned int range_start
,
1904 reg_syntax_t syntax
,
1908 /* Fetch the next character in the uncompiled pattern---translating it
1909 if necessary. Also cast from a signed character in the constant
1910 string passed to us by the user to an unsigned char that we can use
1911 as an array index (in, e.g., `translate'). */
1912 /* ifdef MBS_SUPPORT, we translate only if character <= 0xff,
1913 because it is impossible to allocate 4GB array for some encodings
1914 which have 4 byte character_set like UCS4. */
1917 # define PATFETCH(c) \
1918 do {if (p == pend) return REG_EEND; \
1919 c = (UCHAR_T) *p++; \
1920 if (translate && (c <= 0xff)) c = (UCHAR_T) translate[c]; \
1923 # define PATFETCH(c) \
1924 do {if (p == pend) return REG_EEND; \
1925 c = (unsigned char) *p++; \
1926 if (translate) c = (unsigned char) translate[c]; \
1931 /* Fetch the next character in the uncompiled pattern, with no
1933 # define PATFETCH_RAW(c) \
1934 do {if (p == pend) return REG_EEND; \
1935 c = (UCHAR_T) *p++; \
1938 /* Go backwards one character in the pattern. */
1939 # define PATUNFETCH p--
1942 /* If `translate' is non-null, return translate[D], else just D. We
1943 cast the subscript to translate because some data is declared as
1944 `char *', to avoid warnings when a string constant is passed. But
1945 when we use a character as a subscript we must make it unsigned. */
1946 /* ifdef MBS_SUPPORT, we translate only if character <= 0xff,
1947 because it is impossible to allocate 4GB array for some encodings
1948 which have 4 byte character_set like UCS4. */
1952 # define TRANSLATE(d) \
1953 ((translate && ((UCHAR_T) (d)) <= 0xff) \
1954 ? (char) translate[(unsigned char) (d)] : (d))
1956 # define TRANSLATE(d) \
1957 (translate ? (char) translate[(unsigned char) (d)] : (d))
1962 /* Macros for outputting the compiled pattern into `buffer'. */
1964 /* If the buffer isn't allocated when it comes in, use this. */
1965 # define INIT_BUF_SIZE (32 * sizeof(UCHAR_T))
1967 /* Make sure we have at least N more bytes of space in buffer. */
1969 # define GET_BUFFER_SPACE(n) \
1970 while (((unsigned long)b - (unsigned long)COMPILED_BUFFER_VAR \
1971 + (n)*sizeof(CHAR_T)) > bufp->allocated) \
1974 # define GET_BUFFER_SPACE(n) \
1975 while ((unsigned long) (b - bufp->buffer + (n)) > bufp->allocated) \
1979 /* Make sure we have one more byte of buffer space and then add C to it. */
1980 # define BUF_PUSH(c) \
1982 GET_BUFFER_SPACE (1); \
1983 *b++ = (UCHAR_T) (c); \
1987 /* Ensure we have two more bytes of buffer space and then append C1 and C2. */
1988 # define BUF_PUSH_2(c1, c2) \
1990 GET_BUFFER_SPACE (2); \
1991 *b++ = (UCHAR_T) (c1); \
1992 *b++ = (UCHAR_T) (c2); \
1996 /* As with BUF_PUSH_2, except for three bytes. */
1997 # define BUF_PUSH_3(c1, c2, c3) \
1999 GET_BUFFER_SPACE (3); \
2000 *b++ = (UCHAR_T) (c1); \
2001 *b++ = (UCHAR_T) (c2); \
2002 *b++ = (UCHAR_T) (c3); \
2005 /* Store a jump with opcode OP at LOC to location TO. We store a
2006 relative address offset by the three bytes the jump itself occupies. */
2007 # define STORE_JUMP(op, loc, to) \
2008 PREFIX(store_op1) (op, loc, (int) ((to) - (loc) - (1 + OFFSET_ADDRESS_SIZE)))
2010 /* Likewise, for a two-argument jump. */
2011 # define STORE_JUMP2(op, loc, to, arg) \
2012 PREFIX(store_op2) (op, loc, (int) ((to) - (loc) - (1 + OFFSET_ADDRESS_SIZE)), arg)
2014 /* Like `STORE_JUMP', but for inserting. Assume `b' is the buffer end. */
2015 # define INSERT_JUMP(op, loc, to) \
2016 PREFIX(insert_op1) (op, loc, (int) ((to) - (loc) - (1 + OFFSET_ADDRESS_SIZE)), b)
2018 /* Like `STORE_JUMP2', but for inserting. Assume `b' is the buffer end. */
2019 # define INSERT_JUMP2(op, loc, to, arg) \
2020 PREFIX(insert_op2) (op, loc, (int) ((to) - (loc) - (1 + OFFSET_ADDRESS_SIZE)),\
2023 /* This is not an arbitrary limit: the arguments which represent offsets
2024 into the pattern are two bytes long. So if 2^16 bytes turns out to
2025 be too small, many things would have to change. */
2026 /* Any other compiler which, like MSC, has allocation limit below 2^16
2027 bytes will have to use approach similar to what was done below for
2028 MSC and drop MAX_BUF_SIZE a bit. Otherwise you may end up
2029 reallocating to 0 bytes. Such thing is not going to work too well.
2030 You have been warned!! */
2031 # ifndef DEFINED_ONCE
2032 # if defined _MSC_VER && !defined WIN32
2033 /* Microsoft C 16-bit versions limit malloc to approx 65512 bytes.
2034 The REALLOC define eliminates a flurry of conversion warnings,
2035 but is not required. */
2036 # define MAX_BUF_SIZE 65500L
2037 # define REALLOC(p,s) realloc ((p), (size_t) (s))
2039 # define MAX_BUF_SIZE (1L << 16)
2040 # define REALLOC(p,s) realloc ((p), (s))
2043 /* Extend the buffer by twice its current size via realloc and
2044 reset the pointers that pointed into the old block to point to the
2045 correct places in the new one. If extending the buffer results in it
2046 being larger than MAX_BUF_SIZE, then flag memory exhausted. */
2047 # if __BOUNDED_POINTERS__
2048 # define SET_HIGH_BOUND(P) (__ptrhigh (P) = __ptrlow (P) + bufp->allocated)
2049 # define MOVE_BUFFER_POINTER(P) \
2050 (__ptrlow (P) += incr, SET_HIGH_BOUND (P), __ptrvalue (P) += incr)
2051 # define ELSE_EXTEND_BUFFER_HIGH_BOUND \
2054 SET_HIGH_BOUND (b); \
2055 SET_HIGH_BOUND (begalt); \
2056 if (fixup_alt_jump) \
2057 SET_HIGH_BOUND (fixup_alt_jump); \
2059 SET_HIGH_BOUND (laststart); \
2060 if (pending_exact) \
2061 SET_HIGH_BOUND (pending_exact); \
2064 # define MOVE_BUFFER_POINTER(P) (P) += incr
2065 # define ELSE_EXTEND_BUFFER_HIGH_BOUND
2067 # endif /* not DEFINED_ONCE */
2070 # define EXTEND_BUFFER() \
2072 UCHAR_T *old_buffer = COMPILED_BUFFER_VAR; \
2074 if (bufp->allocated + sizeof(UCHAR_T) > MAX_BUF_SIZE) \
2076 bufp->allocated <<= 1; \
2077 if (bufp->allocated > MAX_BUF_SIZE) \
2078 bufp->allocated = MAX_BUF_SIZE; \
2079 /* How many characters the new buffer can have? */ \
2080 wchar_count = bufp->allocated / sizeof(UCHAR_T); \
2081 if (wchar_count == 0) wchar_count = 1; \
2082 /* Truncate the buffer to CHAR_T align. */ \
2083 bufp->allocated = wchar_count * sizeof(UCHAR_T); \
2084 RETALLOC (COMPILED_BUFFER_VAR, wchar_count, UCHAR_T); \
2085 bufp->buffer = (char*)COMPILED_BUFFER_VAR; \
2086 if (COMPILED_BUFFER_VAR == NULL) \
2087 return REG_ESPACE; \
2088 /* If the buffer moved, move all the pointers into it. */ \
2089 if (old_buffer != COMPILED_BUFFER_VAR) \
2091 int incr = COMPILED_BUFFER_VAR - old_buffer; \
2092 MOVE_BUFFER_POINTER (b); \
2093 MOVE_BUFFER_POINTER (begalt); \
2094 if (fixup_alt_jump) \
2095 MOVE_BUFFER_POINTER (fixup_alt_jump); \
2097 MOVE_BUFFER_POINTER (laststart); \
2098 if (pending_exact) \
2099 MOVE_BUFFER_POINTER (pending_exact); \
2101 ELSE_EXTEND_BUFFER_HIGH_BOUND \
2104 # define EXTEND_BUFFER() \
2106 UCHAR_T *old_buffer = COMPILED_BUFFER_VAR; \
2107 if (bufp->allocated == MAX_BUF_SIZE) \
2109 bufp->allocated <<= 1; \
2110 if (bufp->allocated > MAX_BUF_SIZE) \
2111 bufp->allocated = MAX_BUF_SIZE; \
2112 bufp->buffer = (UCHAR_T *) REALLOC (COMPILED_BUFFER_VAR, \
2114 if (COMPILED_BUFFER_VAR == NULL) \
2115 return REG_ESPACE; \
2116 /* If the buffer moved, move all the pointers into it. */ \
2117 if (old_buffer != COMPILED_BUFFER_VAR) \
2119 int incr = COMPILED_BUFFER_VAR - old_buffer; \
2120 MOVE_BUFFER_POINTER (b); \
2121 MOVE_BUFFER_POINTER (begalt); \
2122 if (fixup_alt_jump) \
2123 MOVE_BUFFER_POINTER (fixup_alt_jump); \
2125 MOVE_BUFFER_POINTER (laststart); \
2126 if (pending_exact) \
2127 MOVE_BUFFER_POINTER (pending_exact); \
2129 ELSE_EXTEND_BUFFER_HIGH_BOUND \
2133 # ifndef DEFINED_ONCE
2134 /* Since we have one byte reserved for the register number argument to
2135 {start,stop}_memory, the maximum number of groups we can report
2136 things about is what fits in that byte. */
2137 # define MAX_REGNUM 255
2139 /* But patterns can have more than `MAX_REGNUM' registers. We just
2140 ignore the excess. */
2141 typedef unsigned regnum_t
;
2144 /* Macros for the compile stack. */
2146 /* Since offsets can go either forwards or backwards, this type needs to
2147 be able to hold values from -(MAX_BUF_SIZE - 1) to MAX_BUF_SIZE - 1. */
2148 /* int may be not enough when sizeof(int) == 2. */
2149 typedef long pattern_offset_t
;
2153 pattern_offset_t begalt_offset
;
2154 pattern_offset_t fixup_alt_jump
;
2155 pattern_offset_t inner_group_offset
;
2156 pattern_offset_t laststart_offset
;
2158 } compile_stack_elt_t
;
2163 compile_stack_elt_t
*stack
;
2165 unsigned avail
; /* Offset of next open position. */
2166 } compile_stack_type
;
2169 # define INIT_COMPILE_STACK_SIZE 32
2171 # define COMPILE_STACK_EMPTY (compile_stack.avail == 0)
2172 # define COMPILE_STACK_FULL (compile_stack.avail == compile_stack.size)
2174 /* The next available element. */
2175 # define COMPILE_STACK_TOP (compile_stack.stack[compile_stack.avail])
2177 # endif /* not DEFINED_ONCE */
2179 /* Set the bit for character C in a list. */
2180 # ifndef DEFINED_ONCE
2181 # define SET_LIST_BIT(c) \
2182 (b[((unsigned char) (c)) / BYTEWIDTH] \
2183 |= 1 << (((unsigned char) c) % BYTEWIDTH))
2184 # endif /* DEFINED_ONCE */
2186 /* Get the next unsigned number in the uncompiled pattern. */
2187 # define GET_UNSIGNED_NUMBER(num) \
2192 if (c < '0' || c > '9') \
2194 if (num <= RE_DUP_MAX) \
2198 num = num * 10 + c - '0'; \
2203 # ifndef DEFINED_ONCE
2204 # if defined _LIBC || WIDE_CHAR_SUPPORT
2205 /* The GNU C library provides support for user-defined character classes
2206 and the functions from ISO C amendement 1. */
2207 # ifdef CHARCLASS_NAME_MAX
2208 # define CHAR_CLASS_MAX_LENGTH CHARCLASS_NAME_MAX
2210 /* This shouldn't happen but some implementation might still have this
2211 problem. Use a reasonable default value. */
2212 # define CHAR_CLASS_MAX_LENGTH 256
2216 # define IS_CHAR_CLASS(string) __wctype (string)
2218 # define IS_CHAR_CLASS(string) wctype (string)
2221 # define CHAR_CLASS_MAX_LENGTH 6 /* Namely, `xdigit'. */
2223 # define IS_CHAR_CLASS(string) \
2224 (STREQ (string, "alpha") || STREQ (string, "upper") \
2225 || STREQ (string, "lower") || STREQ (string, "digit") \
2226 || STREQ (string, "alnum") || STREQ (string, "xdigit") \
2227 || STREQ (string, "space") || STREQ (string, "print") \
2228 || STREQ (string, "punct") || STREQ (string, "graph") \
2229 || STREQ (string, "cntrl") || STREQ (string, "blank"))
2231 # endif /* DEFINED_ONCE */
2233 # ifndef MATCH_MAY_ALLOCATE
2235 /* If we cannot allocate large objects within re_match_2_internal,
2236 we make the fail stack and register vectors global.
2237 The fail stack, we grow to the maximum size when a regexp
2239 The register vectors, we adjust in size each time we
2240 compile a regexp, according to the number of registers it needs. */
2242 static PREFIX(fail_stack_type
) fail_stack
;
2244 /* Size with which the following vectors are currently allocated.
2245 That is so we can make them bigger as needed,
2246 but never make them smaller. */
2247 # ifdef DEFINED_ONCE
2248 static int regs_allocated_size
;
2250 static const char ** regstart
, ** regend
;
2251 static const char ** old_regstart
, ** old_regend
;
2252 static const char **best_regstart
, **best_regend
;
2253 static const char **reg_dummy
;
2254 # endif /* DEFINED_ONCE */
2256 static PREFIX(register_info_type
) *PREFIX(reg_info
);
2257 static PREFIX(register_info_type
) *PREFIX(reg_info_dummy
);
2259 /* Make the register vectors big enough for NUM_REGS registers,
2260 but don't make them smaller. */
2263 PREFIX(regex_grow_registers
) (num_regs
)
2266 if (num_regs
> regs_allocated_size
)
2268 RETALLOC_IF (regstart
, num_regs
, const char *);
2269 RETALLOC_IF (regend
, num_regs
, const char *);
2270 RETALLOC_IF (old_regstart
, num_regs
, const char *);
2271 RETALLOC_IF (old_regend
, num_regs
, const char *);
2272 RETALLOC_IF (best_regstart
, num_regs
, const char *);
2273 RETALLOC_IF (best_regend
, num_regs
, const char *);
2274 RETALLOC_IF (PREFIX(reg_info
), num_regs
, PREFIX(register_info_type
));
2275 RETALLOC_IF (reg_dummy
, num_regs
, const char *);
2276 RETALLOC_IF (PREFIX(reg_info_dummy
), num_regs
, PREFIX(register_info_type
));
2278 regs_allocated_size
= num_regs
;
2282 # endif /* not MATCH_MAY_ALLOCATE */
2284 # ifndef DEFINED_ONCE
2285 static boolean group_in_compile_stack
_RE_ARGS ((compile_stack_type
2288 # endif /* not DEFINED_ONCE */
2290 /* `regex_compile' compiles PATTERN (of length SIZE) according to SYNTAX.
2291 Returns one of error codes defined in `regex.h', or zero for success.
2293 Assumes the `allocated' (and perhaps `buffer') and `translate'
2294 fields are set in BUFP on entry.
2296 If it succeeds, results are put in BUFP (if it returns an error, the
2297 contents of BUFP are undefined):
2298 `buffer' is the compiled pattern;
2299 `syntax' is set to SYNTAX;
2300 `used' is set to the length of the compiled pattern;
2301 `fastmap_accurate' is zero;
2302 `re_nsub' is the number of subexpressions in PATTERN;
2303 `not_bol' and `not_eol' are zero;
2305 The `fastmap' and `newline_anchor' fields are neither
2306 examined nor set. */
2308 /* Return, freeing storage we allocated. */
2310 # define FREE_STACK_RETURN(value) \
2311 return (free(pattern), free(mbs_offset), free(is_binary), free (compile_stack.stack), value)
2313 # define FREE_STACK_RETURN(value) \
2314 return (free (compile_stack.stack), value)
2317 static reg_errcode_t
2318 PREFIX(regex_compile
) (ARG_PREFIX(pattern
), ARG_PREFIX(size
), syntax
, bufp
)
2319 const char *ARG_PREFIX(pattern
);
2320 size_t ARG_PREFIX(size
);
2321 reg_syntax_t syntax
;
2322 struct re_pattern_buffer
*bufp
;
2324 /* We fetch characters from PATTERN here. Even though PATTERN is
2325 `char *' (i.e., signed), we declare these variables as unsigned, so
2326 they can be reliably used as array indices. */
2327 register UCHAR_T c
, c1
;
2330 /* A temporary space to keep wchar_t pattern and compiled pattern. */
2331 CHAR_T
*pattern
, *COMPILED_BUFFER_VAR
;
2333 /* offset buffer for optimization. See convert_mbs_to_wc. */
2334 int *mbs_offset
= NULL
;
2335 /* It hold whether each wchar_t is binary data or not. */
2336 char *is_binary
= NULL
;
2337 /* A flag whether exactn is handling binary data or not. */
2338 char is_exactn_bin
= FALSE
;
2341 /* A random temporary spot in PATTERN. */
2344 /* Points to the end of the buffer, where we should append. */
2345 register UCHAR_T
*b
;
2347 /* Keeps track of unclosed groups. */
2348 compile_stack_type compile_stack
;
2350 /* Points to the current (ending) position in the pattern. */
2355 const CHAR_T
*p
= pattern
;
2356 const CHAR_T
*pend
= pattern
+ size
;
2359 /* How to translate the characters in the pattern. */
2360 RE_TRANSLATE_TYPE translate
= bufp
->translate
;
2362 /* Address of the count-byte of the most recently inserted `exactn'
2363 command. This makes it possible to tell if a new exact-match
2364 character can be added to that command or if the character requires
2365 a new `exactn' command. */
2366 UCHAR_T
*pending_exact
= 0;
2368 /* Address of start of the most recently finished expression.
2369 This tells, e.g., postfix * where to find the start of its
2370 operand. Reset at the beginning of groups and alternatives. */
2371 UCHAR_T
*laststart
= 0;
2373 /* Address of beginning of regexp, or inside of last group. */
2376 /* Address of the place where a forward jump should go to the end of
2377 the containing expression. Each alternative of an `or' -- except the
2378 last -- ends with a forward jump of this sort. */
2379 UCHAR_T
*fixup_alt_jump
= 0;
2381 /* Counts open-groups as they are encountered. Remembered for the
2382 matching close-group on the compile stack, so the same register
2383 number is put in the stop_memory as the start_memory. */
2384 regnum_t regnum
= 0;
2387 /* Initialize the wchar_t PATTERN and offset_buffer. */
2388 p
= pend
= pattern
= TALLOC(csize
+ 1, CHAR_T
);
2389 mbs_offset
= TALLOC(csize
+ 1, int);
2390 is_binary
= TALLOC(csize
+ 1, char);
2391 if (pattern
== NULL
|| mbs_offset
== NULL
|| is_binary
== NULL
)
2398 pattern
[csize
] = L
'\0'; /* sentinel */
2399 size
= convert_mbs_to_wcs(pattern
, cpattern
, csize
, mbs_offset
, is_binary
);
2411 DEBUG_PRINT1 ("\nCompiling pattern: ");
2414 unsigned debug_count
;
2416 for (debug_count
= 0; debug_count
< size
; debug_count
++)
2417 PUT_CHAR (pattern
[debug_count
]);
2422 /* Initialize the compile stack. */
2423 compile_stack
.stack
= TALLOC (INIT_COMPILE_STACK_SIZE
, compile_stack_elt_t
);
2424 if (compile_stack
.stack
== NULL
)
2434 compile_stack
.size
= INIT_COMPILE_STACK_SIZE
;
2435 compile_stack
.avail
= 0;
2437 /* Initialize the pattern buffer. */
2438 bufp
->syntax
= syntax
;
2439 bufp
->fastmap_accurate
= 0;
2440 bufp
->not_bol
= bufp
->not_eol
= 0;
2442 /* Set `used' to zero, so that if we return an error, the pattern
2443 printer (for debugging) will think there's no pattern. We reset it
2447 /* Always count groups, whether or not bufp->no_sub is set. */
2450 #if !defined emacs && !defined SYNTAX_TABLE
2451 /* Initialize the syntax table. */
2452 init_syntax_once ();
2455 if (bufp
->allocated
== 0)
2458 { /* If zero allocated, but buffer is non-null, try to realloc
2459 enough space. This loses if buffer's address is bogus, but
2460 that is the user's responsibility. */
2462 /* Free bufp->buffer and allocate an array for wchar_t pattern
2465 COMPILED_BUFFER_VAR
= TALLOC (INIT_BUF_SIZE
/sizeof(UCHAR_T
),
2468 RETALLOC (COMPILED_BUFFER_VAR
, INIT_BUF_SIZE
, UCHAR_T
);
2472 { /* Caller did not allocate a buffer. Do it for them. */
2473 COMPILED_BUFFER_VAR
= TALLOC (INIT_BUF_SIZE
/ sizeof(UCHAR_T
),
2477 if (!COMPILED_BUFFER_VAR
) FREE_STACK_RETURN (REG_ESPACE
);
2479 bufp
->buffer
= (char*)COMPILED_BUFFER_VAR
;
2481 bufp
->allocated
= INIT_BUF_SIZE
;
2485 COMPILED_BUFFER_VAR
= (UCHAR_T
*) bufp
->buffer
;
2488 begalt
= b
= COMPILED_BUFFER_VAR
;
2490 /* Loop through the uncompiled pattern until we're at the end. */
2499 if ( /* If at start of pattern, it's an operator. */
2501 /* If context independent, it's an operator. */
2502 || syntax
& RE_CONTEXT_INDEP_ANCHORS
2503 /* Otherwise, depends on what's come before. */
2504 || PREFIX(at_begline_loc_p
) (pattern
, p
, syntax
))
2514 if ( /* If at end of pattern, it's an operator. */
2516 /* If context independent, it's an operator. */
2517 || syntax
& RE_CONTEXT_INDEP_ANCHORS
2518 /* Otherwise, depends on what's next. */
2519 || PREFIX(at_endline_loc_p
) (p
, pend
, syntax
))
2529 if ((syntax
& RE_BK_PLUS_QM
)
2530 || (syntax
& RE_LIMITED_OPS
))
2534 /* If there is no previous pattern... */
2537 if (syntax
& RE_CONTEXT_INVALID_OPS
)
2538 FREE_STACK_RETURN (REG_BADRPT
);
2539 else if (!(syntax
& RE_CONTEXT_INDEP_OPS
))
2544 /* Are we optimizing this jump? */
2545 boolean keep_string_p
= false;
2547 /* 1 means zero (many) matches is allowed. */
2548 char zero_times_ok
= 0, many_times_ok
= 0;
2550 /* If there is a sequence of repetition chars, collapse it
2551 down to just one (the right one). We can't combine
2552 interval operators with these because of, e.g., `a{2}*',
2553 which should only match an even number of `a's. */
2557 zero_times_ok
|= c
!= '+';
2558 many_times_ok
|= c
!= '?';
2566 || (!(syntax
& RE_BK_PLUS_QM
) && (c
== '+' || c
== '?')))
2569 else if (syntax
& RE_BK_PLUS_QM
&& c
== '\\')
2571 if (p
== pend
) FREE_STACK_RETURN (REG_EESCAPE
);
2574 if (!(c1
== '+' || c1
== '?'))
2589 /* If we get here, we found another repeat character. */
2592 /* Star, etc. applied to an empty pattern is equivalent
2593 to an empty pattern. */
2597 /* Now we know whether or not zero matches is allowed
2598 and also whether or not two or more matches is allowed. */
2600 { /* More than one repetition is allowed, so put in at the
2601 end a backward relative jump from `b' to before the next
2602 jump we're going to put in below (which jumps from
2603 laststart to after this jump).
2605 But if we are at the `*' in the exact sequence `.*\n',
2606 insert an unconditional jump backwards to the .,
2607 instead of the beginning of the loop. This way we only
2608 push a failure point once, instead of every time
2609 through the loop. */
2610 assert (p
- 1 > pattern
);
2612 /* Allocate the space for the jump. */
2613 GET_BUFFER_SPACE (1 + OFFSET_ADDRESS_SIZE
);
2615 /* We know we are not at the first character of the pattern,
2616 because laststart was nonzero. And we've already
2617 incremented `p', by the way, to be the character after
2618 the `*'. Do we have to do something analogous here
2619 for null bytes, because of RE_DOT_NOT_NULL? */
2620 if (TRANSLATE (*(p
- 2)) == TRANSLATE ('.')
2622 && p
< pend
&& TRANSLATE (*p
) == TRANSLATE ('\n')
2623 && !(syntax
& RE_DOT_NEWLINE
))
2624 { /* We have .*\n. */
2625 STORE_JUMP (jump
, b
, laststart
);
2626 keep_string_p
= true;
2629 /* Anything else. */
2630 STORE_JUMP (maybe_pop_jump
, b
, laststart
-
2631 (1 + OFFSET_ADDRESS_SIZE
));
2633 /* We've added more stuff to the buffer. */
2634 b
+= 1 + OFFSET_ADDRESS_SIZE
;
2637 /* On failure, jump from laststart to b + 3, which will be the
2638 end of the buffer after this jump is inserted. */
2639 /* ifdef WCHAR, 'b + 1 + OFFSET_ADDRESS_SIZE' instead of
2641 GET_BUFFER_SPACE (1 + OFFSET_ADDRESS_SIZE
);
2642 INSERT_JUMP (keep_string_p
? on_failure_keep_string_jump
2644 laststart
, b
+ 1 + OFFSET_ADDRESS_SIZE
);
2646 b
+= 1 + OFFSET_ADDRESS_SIZE
;
2650 /* At least one repetition is required, so insert a
2651 `dummy_failure_jump' before the initial
2652 `on_failure_jump' instruction of the loop. This
2653 effects a skip over that instruction the first time
2654 we hit that loop. */
2655 GET_BUFFER_SPACE (1 + OFFSET_ADDRESS_SIZE
);
2656 INSERT_JUMP (dummy_failure_jump
, laststart
, laststart
+
2657 2 + 2 * OFFSET_ADDRESS_SIZE
);
2658 b
+= 1 + OFFSET_ADDRESS_SIZE
;
2672 boolean had_char_class
= false;
2674 CHAR_T range_start
= 0xffffffff;
2676 unsigned int range_start
= 0xffffffff;
2678 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2681 /* We assume a charset(_not) structure as a wchar_t array.
2682 charset[0] = (re_opcode_t) charset(_not)
2683 charset[1] = l (= length of char_classes)
2684 charset[2] = m (= length of collating_symbols)
2685 charset[3] = n (= length of equivalence_classes)
2686 charset[4] = o (= length of char_ranges)
2687 charset[5] = p (= length of chars)
2689 charset[6] = char_class (wctype_t)
2690 charset[6+CHAR_CLASS_SIZE] = char_class (wctype_t)
2692 charset[l+5] = char_class (wctype_t)
2694 charset[l+6] = collating_symbol (wchar_t)
2696 charset[l+m+5] = collating_symbol (wchar_t)
2697 ifdef _LIBC we use the index if
2698 _NL_COLLATE_SYMB_EXTRAMB instead of
2701 charset[l+m+6] = equivalence_classes (wchar_t)
2703 charset[l+m+n+5] = equivalence_classes (wchar_t)
2704 ifdef _LIBC we use the index in
2705 _NL_COLLATE_WEIGHT instead of
2708 charset[l+m+n+6] = range_start
2709 charset[l+m+n+7] = range_end
2711 charset[l+m+n+2o+4] = range_start
2712 charset[l+m+n+2o+5] = range_end
2713 ifdef _LIBC we use the value looked up
2714 in _NL_COLLATE_COLLSEQ instead of
2717 charset[l+m+n+2o+6] = char
2719 charset[l+m+n+2o+p+5] = char
2723 /* We need at least 6 spaces: the opcode, the length of
2724 char_classes, the length of collating_symbols, the length of
2725 equivalence_classes, the length of char_ranges, the length of
2727 GET_BUFFER_SPACE (6);
2729 /* Save b as laststart. And We use laststart as the pointer
2730 to the first element of the charset here.
2731 In other words, laststart[i] indicates charset[i]. */
2734 /* We test `*p == '^' twice, instead of using an if
2735 statement, so we only need one BUF_PUSH. */
2736 BUF_PUSH (*p
== '^' ? charset_not
: charset
);
2740 /* Push the length of char_classes, the length of
2741 collating_symbols, the length of equivalence_classes, the
2742 length of char_ranges and the length of chars. */
2743 BUF_PUSH_3 (0, 0, 0);
2746 /* Remember the first position in the bracket expression. */
2749 /* charset_not matches newline according to a syntax bit. */
2750 if ((re_opcode_t
) b
[-6] == charset_not
2751 && (syntax
& RE_HAT_LISTS_NOT_NEWLINE
))
2754 laststart
[5]++; /* Update the length of characters */
2757 /* Read in characters and ranges, setting map bits. */
2760 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2764 /* \ might escape characters inside [...] and [^...]. */
2765 if ((syntax
& RE_BACKSLASH_ESCAPE_IN_LISTS
) && c
== '\\')
2767 if (p
== pend
) FREE_STACK_RETURN (REG_EESCAPE
);
2771 laststart
[5]++; /* Update the length of chars */
2776 /* Could be the end of the bracket expression. If it's
2777 not (i.e., when the bracket expression is `[]' so
2778 far), the ']' character bit gets set way below. */
2779 if (c
== ']' && p
!= p1
+ 1)
2782 /* Look ahead to see if it's a range when the last thing
2783 was a character class. */
2784 if (had_char_class
&& c
== '-' && *p
!= ']')
2785 FREE_STACK_RETURN (REG_ERANGE
);
2787 /* Look ahead to see if it's a range when the last thing
2788 was a character: if this is a hyphen not at the
2789 beginning or the end of a list, then it's the range
2792 && !(p
- 2 >= pattern
&& p
[-2] == '[')
2793 && !(p
- 3 >= pattern
&& p
[-3] == '[' && p
[-2] == '^')
2797 /* Allocate the space for range_start and range_end. */
2798 GET_BUFFER_SPACE (2);
2799 /* Update the pointer to indicate end of buffer. */
2801 ret
= wcs_compile_range (range_start
, &p
, pend
, translate
,
2802 syntax
, b
, laststart
);
2803 if (ret
!= REG_NOERROR
) FREE_STACK_RETURN (ret
);
2804 range_start
= 0xffffffff;
2806 else if (p
[0] == '-' && p
[1] != ']')
2807 { /* This handles ranges made up of characters only. */
2810 /* Move past the `-'. */
2812 /* Allocate the space for range_start and range_end. */
2813 GET_BUFFER_SPACE (2);
2814 /* Update the pointer to indicate end of buffer. */
2816 ret
= wcs_compile_range (c
, &p
, pend
, translate
, syntax
, b
,
2818 if (ret
!= REG_NOERROR
) FREE_STACK_RETURN (ret
);
2819 range_start
= 0xffffffff;
2822 /* See if we're at the beginning of a possible character
2824 else if (syntax
& RE_CHAR_CLASSES
&& c
== '[' && *p
== ':')
2825 { /* Leave room for the null. */
2826 char str
[CHAR_CLASS_MAX_LENGTH
+ 1];
2831 /* If pattern is `[[:'. */
2832 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2837 if ((c
== ':' && *p
== ']') || p
== pend
)
2839 if (c1
< CHAR_CLASS_MAX_LENGTH
)
2842 /* This is in any case an invalid class name. */
2847 /* If isn't a word bracketed by `[:' and `:]':
2848 undo the ending character, the letters, and leave
2849 the leading `:' and `[' (but store them as character). */
2850 if (c
== ':' && *p
== ']')
2855 /* Query the character class as wctype_t. */
2856 wt
= IS_CHAR_CLASS (str
);
2858 FREE_STACK_RETURN (REG_ECTYPE
);
2860 /* Throw away the ] at the end of the character
2864 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2866 /* Allocate the space for character class. */
2867 GET_BUFFER_SPACE(CHAR_CLASS_SIZE
);
2868 /* Update the pointer to indicate end of buffer. */
2869 b
+= CHAR_CLASS_SIZE
;
2870 /* Move data which follow character classes
2871 not to violate the data. */
2872 insert_space(CHAR_CLASS_SIZE
,
2873 laststart
+ 6 + laststart
[1],
2875 alignedp
= ((uintptr_t)(laststart
+ 6 + laststart
[1])
2876 + __alignof__(wctype_t) - 1)
2877 & ~(uintptr_t)(__alignof__(wctype_t) - 1);
2878 /* Store the character class. */
2879 *((wctype_t*)alignedp
) = wt
;
2880 /* Update length of char_classes */
2881 laststart
[1] += CHAR_CLASS_SIZE
;
2883 had_char_class
= true;
2892 laststart
[5] += 2; /* Update the length of characters */
2894 had_char_class
= false;
2897 else if (syntax
& RE_CHAR_CLASSES
&& c
== '[' && (*p
== '='
2900 CHAR_T str
[128]; /* Should be large enough. */
2901 CHAR_T delim
= *p
; /* '=' or '.' */
2904 _NL_CURRENT_WORD (LC_COLLATE
, _NL_COLLATE_NRULES
);
2909 /* If pattern is `[[=' or '[[.'. */
2910 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
2915 if ((c
== delim
&& *p
== ']') || p
== pend
)
2917 if (c1
< sizeof (str
) - 1)
2920 /* This is in any case an invalid class name. */
2925 if (c
== delim
&& *p
== ']' && str
[0] != '\0')
2927 unsigned int i
, offset
;
2928 /* If we have no collation data we use the default
2929 collation in which each character is in a class
2930 by itself. It also means that ASCII is the
2931 character set and therefore we cannot have character
2932 with more than one byte in the multibyte
2935 /* If not defined _LIBC, we push the name and
2936 `\0' for the sake of matching performance. */
2937 int datasize
= c1
+ 1;
2945 FREE_STACK_RETURN (REG_ECOLLATE
);
2950 const int32_t *table
;
2951 const int32_t *weights
;
2952 const int32_t *extra
;
2953 const int32_t *indirect
;
2956 /* This #include defines a local function! */
2957 # include <locale/weightwc.h>
2961 /* We push the index for equivalence class. */
2964 table
= (const int32_t *)
2965 _NL_CURRENT (LC_COLLATE
,
2966 _NL_COLLATE_TABLEWC
);
2967 weights
= (const int32_t *)
2968 _NL_CURRENT (LC_COLLATE
,
2969 _NL_COLLATE_WEIGHTWC
);
2970 extra
= (const int32_t *)
2971 _NL_CURRENT (LC_COLLATE
,
2972 _NL_COLLATE_EXTRAWC
);
2973 indirect
= (const int32_t *)
2974 _NL_CURRENT (LC_COLLATE
,
2975 _NL_COLLATE_INDIRECTWC
);
2977 idx
= findidx ((const wint_t**)&cp
);
2978 if (idx
== 0 || cp
< (wint_t*) str
+ c1
)
2979 /* This is no valid character. */
2980 FREE_STACK_RETURN (REG_ECOLLATE
);
2982 str
[0] = (wchar_t)idx
;
2984 else /* delim == '.' */
2986 /* We push collation sequence value
2987 for collating symbol. */
2989 const int32_t *symb_table
;
2990 const unsigned char *extra
;
2997 /* We have to convert the name to a single-byte
2998 string. This is possible since the names
2999 consist of ASCII characters and the internal
3000 representation is UCS4. */
3001 for (i
= 0; i
< c1
; ++i
)
3002 char_str
[i
] = str
[i
];
3005 _NL_CURRENT_WORD (LC_COLLATE
,
3006 _NL_COLLATE_SYMB_HASH_SIZEMB
);
3007 symb_table
= (const int32_t *)
3008 _NL_CURRENT (LC_COLLATE
,
3009 _NL_COLLATE_SYMB_TABLEMB
);
3010 extra
= (const unsigned char *)
3011 _NL_CURRENT (LC_COLLATE
,
3012 _NL_COLLATE_SYMB_EXTRAMB
);
3014 /* Locate the character in the hashing table. */
3015 hash
= elem_hash (char_str
, c1
);
3018 elem
= hash
% table_size
;
3019 second
= hash
% (table_size
- 2);
3020 while (symb_table
[2 * elem
] != 0)
3022 /* First compare the hashing value. */
3023 if (symb_table
[2 * elem
] == hash
3024 && c1
== extra
[symb_table
[2 * elem
+ 1]]
3025 && memcmp (char_str
,
3026 &extra
[symb_table
[2 * elem
+ 1]
3029 /* Yep, this is the entry. */
3030 idx
= symb_table
[2 * elem
+ 1];
3031 idx
+= 1 + extra
[idx
];
3039 if (symb_table
[2 * elem
] != 0)
3041 /* Compute the index of the byte sequence
3043 idx
+= 1 + extra
[idx
];
3044 /* Adjust for the alignment. */
3045 idx
= (idx
+ 3) & ~3;
3047 str
[0] = (wchar_t) idx
+ 4;
3049 else if (symb_table
[2 * elem
] == 0 && c1
== 1)
3051 /* No valid character. Match it as a
3052 single byte character. */
3053 had_char_class
= false;
3055 /* Update the length of characters */
3057 range_start
= str
[0];
3059 /* Throw away the ] at the end of the
3060 collating symbol. */
3062 /* exit from the switch block. */
3066 FREE_STACK_RETURN (REG_ECOLLATE
);
3071 /* Throw away the ] at the end of the equivalence
3072 class (or collating symbol). */
3075 /* Allocate the space for the equivalence class
3076 (or collating symbol) (and '\0' if needed). */
3077 GET_BUFFER_SPACE(datasize
);
3078 /* Update the pointer to indicate end of buffer. */
3082 { /* equivalence class */
3083 /* Calculate the offset of char_ranges,
3084 which is next to equivalence_classes. */
3085 offset
= laststart
[1] + laststart
[2]
3088 insert_space(datasize
, laststart
+ offset
, b
- 1);
3090 /* Write the equivalence_class and \0. */
3091 for (i
= 0 ; i
< datasize
; i
++)
3092 laststart
[offset
+ i
] = str
[i
];
3094 /* Update the length of equivalence_classes. */
3095 laststart
[3] += datasize
;
3096 had_char_class
= true;
3098 else /* delim == '.' */
3099 { /* collating symbol */
3100 /* Calculate the offset of the equivalence_classes,
3101 which is next to collating_symbols. */
3102 offset
= laststart
[1] + laststart
[2] + 6;
3103 /* Insert space and write the collationg_symbol
3105 insert_space(datasize
, laststart
+ offset
, b
-1);
3106 for (i
= 0 ; i
< datasize
; i
++)
3107 laststart
[offset
+ i
] = str
[i
];
3109 /* In re_match_2_internal if range_start < -1, we
3110 assume -range_start is the offset of the
3111 collating symbol which is specified as
3112 the character of the range start. So we assign
3113 -(laststart[1] + laststart[2] + 6) to
3115 range_start
= -(laststart
[1] + laststart
[2] + 6);
3116 /* Update the length of collating_symbol. */
3117 laststart
[2] += datasize
;
3118 had_char_class
= false;
3128 laststart
[5] += 2; /* Update the length of characters */
3129 range_start
= delim
;
3130 had_char_class
= false;
3135 had_char_class
= false;
3137 laststart
[5]++; /* Update the length of characters */
3143 /* Ensure that we have enough space to push a charset: the
3144 opcode, the length count, and the bitset; 34 bytes in all. */
3145 GET_BUFFER_SPACE (34);
3149 /* We test `*p == '^' twice, instead of using an if
3150 statement, so we only need one BUF_PUSH. */
3151 BUF_PUSH (*p
== '^' ? charset_not
: charset
);
3155 /* Remember the first position in the bracket expression. */
3158 /* Push the number of bytes in the bitmap. */
3159 BUF_PUSH ((1 << BYTEWIDTH
) / BYTEWIDTH
);
3161 /* Clear the whole map. */
3162 bzero (b
, (1 << BYTEWIDTH
) / BYTEWIDTH
);
3164 /* charset_not matches newline according to a syntax bit. */
3165 if ((re_opcode_t
) b
[-2] == charset_not
3166 && (syntax
& RE_HAT_LISTS_NOT_NEWLINE
))
3167 SET_LIST_BIT ('\n');
3169 /* Read in characters and ranges, setting map bits. */
3172 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
3176 /* \ might escape characters inside [...] and [^...]. */
3177 if ((syntax
& RE_BACKSLASH_ESCAPE_IN_LISTS
) && c
== '\\')
3179 if (p
== pend
) FREE_STACK_RETURN (REG_EESCAPE
);
3187 /* Could be the end of the bracket expression. If it's
3188 not (i.e., when the bracket expression is `[]' so
3189 far), the ']' character bit gets set way below. */
3190 if (c
== ']' && p
!= p1
+ 1)
3193 /* Look ahead to see if it's a range when the last thing
3194 was a character class. */
3195 if (had_char_class
&& c
== '-' && *p
!= ']')
3196 FREE_STACK_RETURN (REG_ERANGE
);
3198 /* Look ahead to see if it's a range when the last thing
3199 was a character: if this is a hyphen not at the
3200 beginning or the end of a list, then it's the range
3203 && !(p
- 2 >= pattern
&& p
[-2] == '[')
3204 && !(p
- 3 >= pattern
&& p
[-3] == '[' && p
[-2] == '^')
3208 = byte_compile_range (range_start
, &p
, pend
, translate
,
3210 if (ret
!= REG_NOERROR
) FREE_STACK_RETURN (ret
);
3211 range_start
= 0xffffffff;
3214 else if (p
[0] == '-' && p
[1] != ']')
3215 { /* This handles ranges made up of characters only. */
3218 /* Move past the `-'. */
3221 ret
= byte_compile_range (c
, &p
, pend
, translate
, syntax
, b
);
3222 if (ret
!= REG_NOERROR
) FREE_STACK_RETURN (ret
);
3223 range_start
= 0xffffffff;
3226 /* See if we're at the beginning of a possible character
3229 else if (syntax
& RE_CHAR_CLASSES
&& c
== '[' && *p
== ':')
3230 { /* Leave room for the null. */
3231 char str
[CHAR_CLASS_MAX_LENGTH
+ 1];
3236 /* If pattern is `[[:'. */
3237 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
3242 if ((c
== ':' && *p
== ']') || p
== pend
)
3244 if (c1
< CHAR_CLASS_MAX_LENGTH
)
3247 /* This is in any case an invalid class name. */
3252 /* If isn't a word bracketed by `[:' and `:]':
3253 undo the ending character, the letters, and leave
3254 the leading `:' and `[' (but set bits for them). */
3255 if (c
== ':' && *p
== ']')
3257 # if defined _LIBC || WIDE_CHAR_SUPPORT
3258 boolean is_lower
= STREQ (str
, "lower");
3259 boolean is_upper
= STREQ (str
, "upper");
3263 wt
= IS_CHAR_CLASS (str
);
3265 FREE_STACK_RETURN (REG_ECTYPE
);
3267 /* Throw away the ] at the end of the character
3271 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
3273 for (ch
= 0; ch
< 1 << BYTEWIDTH
; ++ch
)
3276 if (__iswctype (__btowc (ch
), wt
))
3279 if (iswctype (btowc (ch
), wt
))
3283 if (translate
&& (is_upper
|| is_lower
)
3284 && (ISUPPER (ch
) || ISLOWER (ch
)))
3288 had_char_class
= true;
3291 boolean is_alnum
= STREQ (str
, "alnum");
3292 boolean is_alpha
= STREQ (str
, "alpha");
3293 boolean is_blank
= STREQ (str
, "blank");
3294 boolean is_cntrl
= STREQ (str
, "cntrl");
3295 boolean is_digit
= STREQ (str
, "digit");
3296 boolean is_graph
= STREQ (str
, "graph");
3297 boolean is_lower
= STREQ (str
, "lower");
3298 boolean is_print
= STREQ (str
, "print");
3299 boolean is_punct
= STREQ (str
, "punct");
3300 boolean is_space
= STREQ (str
, "space");
3301 boolean is_upper
= STREQ (str
, "upper");
3302 boolean is_xdigit
= STREQ (str
, "xdigit");
3304 if (!IS_CHAR_CLASS (str
))
3305 FREE_STACK_RETURN (REG_ECTYPE
);
3307 /* Throw away the ] at the end of the character
3311 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
3313 for (ch
= 0; ch
< 1 << BYTEWIDTH
; ch
++)
3315 /* This was split into 3 if's to
3316 avoid an arbitrary limit in some compiler. */
3317 if ( (is_alnum
&& ISALNUM (ch
))
3318 || (is_alpha
&& ISALPHA (ch
))
3319 || (is_blank
&& ISBLANK (ch
))
3320 || (is_cntrl
&& ISCNTRL (ch
)))
3322 if ( (is_digit
&& ISDIGIT (ch
))
3323 || (is_graph
&& ISGRAPH (ch
))
3324 || (is_lower
&& ISLOWER (ch
))
3325 || (is_print
&& ISPRINT (ch
)))
3327 if ( (is_punct
&& ISPUNCT (ch
))
3328 || (is_space
&& ISSPACE (ch
))
3329 || (is_upper
&& ISUPPER (ch
))
3330 || (is_xdigit
&& ISXDIGIT (ch
)))
3332 if ( translate
&& (is_upper
|| is_lower
)
3333 && (ISUPPER (ch
) || ISLOWER (ch
)))
3336 had_char_class
= true;
3337 # endif /* libc || wctype.h */
3347 had_char_class
= false;
3350 else if (syntax
& RE_CHAR_CLASSES
&& c
== '[' && *p
== '=')
3352 unsigned char str
[MB_LEN_MAX
+ 1];
3355 _NL_CURRENT_WORD (LC_COLLATE
, _NL_COLLATE_NRULES
);
3361 /* If pattern is `[[='. */
3362 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
3367 if ((c
== '=' && *p
== ']') || p
== pend
)
3369 if (c1
< MB_LEN_MAX
)
3372 /* This is in any case an invalid class name. */
3377 if (c
== '=' && *p
== ']' && str
[0] != '\0')
3379 /* If we have no collation data we use the default
3380 collation in which each character is in a class
3381 by itself. It also means that ASCII is the
3382 character set and therefore we cannot have character
3383 with more than one byte in the multibyte
3390 FREE_STACK_RETURN (REG_ECOLLATE
);
3392 /* Throw away the ] at the end of the equivalence
3396 /* Set the bit for the character. */
3397 SET_LIST_BIT (str
[0]);
3402 /* Try to match the byte sequence in `str' against
3403 those known to the collate implementation.
3404 First find out whether the bytes in `str' are
3405 actually from exactly one character. */
3406 const int32_t *table
;
3407 const unsigned char *weights
;
3408 const unsigned char *extra
;
3409 const int32_t *indirect
;
3411 const unsigned char *cp
= str
;
3414 /* This #include defines a local function! */
3415 # include <locale/weight.h>
3417 table
= (const int32_t *)
3418 _NL_CURRENT (LC_COLLATE
, _NL_COLLATE_TABLEMB
);
3419 weights
= (const unsigned char *)
3420 _NL_CURRENT (LC_COLLATE
, _NL_COLLATE_WEIGHTMB
);
3421 extra
= (const unsigned char *)
3422 _NL_CURRENT (LC_COLLATE
, _NL_COLLATE_EXTRAMB
);
3423 indirect
= (const int32_t *)
3424 _NL_CURRENT (LC_COLLATE
, _NL_COLLATE_INDIRECTMB
);
3426 idx
= findidx (&cp
);
3427 if (idx
== 0 || cp
< str
+ c1
)
3428 /* This is no valid character. */
3429 FREE_STACK_RETURN (REG_ECOLLATE
);
3431 /* Throw away the ] at the end of the equivalence
3435 /* Now we have to go throught the whole table
3436 and find all characters which have the same
3439 XXX Note that this is not entirely correct.
3440 we would have to match multibyte sequences
3441 but this is not possible with the current
3443 for (ch
= 1; ch
< 256; ++ch
)
3444 /* XXX This test would have to be changed if we
3445 would allow matching multibyte sequences. */
3448 int32_t idx2
= table
[ch
];
3449 size_t len
= weights
[idx2
];
3451 /* Test whether the lenghts match. */
3452 if (weights
[idx
] == len
)
3454 /* They do. New compare the bytes of
3459 && (weights
[idx
+ 1 + cnt
]
3460 == weights
[idx2
+ 1 + cnt
]))
3464 /* They match. Mark the character as
3471 had_char_class
= true;
3481 had_char_class
= false;
3484 else if (syntax
& RE_CHAR_CLASSES
&& c
== '[' && *p
== '.')
3486 unsigned char str
[128]; /* Should be large enough. */
3489 _NL_CURRENT_WORD (LC_COLLATE
, _NL_COLLATE_NRULES
);
3495 /* If pattern is `[[.'. */
3496 if (p
== pend
) FREE_STACK_RETURN (REG_EBRACK
);
3501 if ((c
== '.' && *p
== ']') || p
== pend
)
3503 if (c1
< sizeof (str
))
3506 /* This is in any case an invalid class name. */
3511 if (c
== '.' && *p
== ']' && str
[0] != '\0')
3513 /* If we have no collation data we use the default
3514 collation in which each character is the name
3515 for its own class which contains only the one
3516 character. It also means that ASCII is the
3517 character set and therefore we cannot have character
3518 with more than one byte in the multibyte
3525 FREE_STACK_RETURN (REG_ECOLLATE
);
3527 /* Throw away the ] at the end of the equivalence
3531 /* Set the bit for the character. */
3532 SET_LIST_BIT (str
[0]);
3533 range_start
= ((const unsigned char *) str
)[0];
3538 /* Try to match the byte sequence in `str' against
3539 those known to the collate implementation.
3540 First find out whether the bytes in `str' are
3541 actually from exactly one character. */
3543 const int32_t *symb_table
;
3544 const unsigned char *extra
;
3551 _NL_CURRENT_WORD (LC_COLLATE
,
3552 _NL_COLLATE_SYMB_HASH_SIZEMB
);
3553 symb_table
= (const int32_t *)
3554 _NL_CURRENT (LC_COLLATE
,
3555 _NL_COLLATE_SYMB_TABLEMB
);
3556 extra
= (const unsigned char *)
3557 _NL_CURRENT (LC_COLLATE
,
3558 _NL_COLLATE_SYMB_EXTRAMB
);
3560 /* Locate the character in the hashing table. */
3561 hash
= elem_hash (str
, c1
);
3564 elem
= hash
% table_size
;
3565 second
= hash
% (table_size
- 2);
3566 while (symb_table
[2 * elem
] != 0)
3568 /* First compare the hashing value. */
3569 if (symb_table
[2 * elem
] == hash
3570 && c1
== extra
[symb_table
[2 * elem
+ 1]]
3572 &extra
[symb_table
[2 * elem
+ 1]
3576 /* Yep, this is the entry. */
3577 idx
= symb_table
[2 * elem
+ 1];
3578 idx
+= 1 + extra
[idx
];
3586 if (symb_table
[2 * elem
] == 0)
3587 /* This is no valid character. */
3588 FREE_STACK_RETURN (REG_ECOLLATE
);
3590 /* Throw away the ] at the end of the equivalence
3594 /* Now add the multibyte character(s) we found
3597 XXX Note that this is not entirely correct.
3598 we would have to match multibyte sequences
3599 but this is not possible with the current
3600 implementation. Also, we have to match
3601 collating symbols, which expand to more than
3602 one file, as a whole and not allow the
3603 individual bytes. */
3606 range_start
= extra
[idx
];
3609 SET_LIST_BIT (extra
[idx
]);
3614 had_char_class
= false;
3624 had_char_class
= false;
3629 had_char_class
= false;
3635 /* Discard any (non)matching list bytes that are all 0 at the
3636 end of the map. Decrease the map-length byte too. */
3637 while ((int) b
[-1] > 0 && b
[b
[-1] - 1] == 0)
3646 if (syntax
& RE_NO_BK_PARENS
)
3653 if (syntax
& RE_NO_BK_PARENS
)
3660 if (syntax
& RE_NEWLINE_ALT
)
3667 if (syntax
& RE_NO_BK_VBAR
)
3674 if (syntax
& RE_INTERVALS
&& syntax
& RE_NO_BK_BRACES
)
3675 goto handle_interval
;
3681 if (p
== pend
) FREE_STACK_RETURN (REG_EESCAPE
);
3683 /* Do not translate the character after the \, so that we can
3684 distinguish, e.g., \B from \b, even if we normally would
3685 translate, e.g., B to b. */
3691 if (syntax
& RE_NO_BK_PARENS
)
3692 goto normal_backslash
;
3698 if (COMPILE_STACK_FULL
)
3700 RETALLOC (compile_stack
.stack
, compile_stack
.size
<< 1,
3701 compile_stack_elt_t
);
3702 if (compile_stack
.stack
== NULL
) return REG_ESPACE
;
3704 compile_stack
.size
<<= 1;
3707 /* These are the values to restore when we hit end of this
3708 group. They are all relative offsets, so that if the
3709 whole pattern moves because of realloc, they will still
3711 COMPILE_STACK_TOP
.begalt_offset
= begalt
- COMPILED_BUFFER_VAR
;
3712 COMPILE_STACK_TOP
.fixup_alt_jump
3713 = fixup_alt_jump
? fixup_alt_jump
- COMPILED_BUFFER_VAR
+ 1 : 0;
3714 COMPILE_STACK_TOP
.laststart_offset
= b
- COMPILED_BUFFER_VAR
;
3715 COMPILE_STACK_TOP
.regnum
= regnum
;
3717 /* We will eventually replace the 0 with the number of
3718 groups inner to this one. But do not push a
3719 start_memory for groups beyond the last one we can
3720 represent in the compiled pattern. */
3721 if (regnum
<= MAX_REGNUM
)
3723 COMPILE_STACK_TOP
.inner_group_offset
= b
3724 - COMPILED_BUFFER_VAR
+ 2;
3725 BUF_PUSH_3 (start_memory
, regnum
, 0);
3728 compile_stack
.avail
++;
3733 /* If we've reached MAX_REGNUM groups, then this open
3734 won't actually generate any code, so we'll have to
3735 clear pending_exact explicitly. */
3741 if (syntax
& RE_NO_BK_PARENS
) goto normal_backslash
;
3743 if (COMPILE_STACK_EMPTY
)
3745 if (syntax
& RE_UNMATCHED_RIGHT_PAREN_ORD
)
3746 goto normal_backslash
;
3748 FREE_STACK_RETURN (REG_ERPAREN
);
3753 { /* Push a dummy failure point at the end of the
3754 alternative for a possible future
3755 `pop_failure_jump' to pop. See comments at
3756 `push_dummy_failure' in `re_match_2'. */
3757 BUF_PUSH (push_dummy_failure
);
3759 /* We allocated space for this jump when we assigned
3760 to `fixup_alt_jump', in the `handle_alt' case below. */
3761 STORE_JUMP (jump_past_alt
, fixup_alt_jump
, b
- 1);
3764 /* See similar code for backslashed left paren above. */
3765 if (COMPILE_STACK_EMPTY
)
3767 if (syntax
& RE_UNMATCHED_RIGHT_PAREN_ORD
)
3770 FREE_STACK_RETURN (REG_ERPAREN
);
3773 /* Since we just checked for an empty stack above, this
3774 ``can't happen''. */
3775 assert (compile_stack
.avail
!= 0);
3777 /* We don't just want to restore into `regnum', because
3778 later groups should continue to be numbered higher,
3779 as in `(ab)c(de)' -- the second group is #2. */
3780 regnum_t this_group_regnum
;
3782 compile_stack
.avail
--;
3783 begalt
= COMPILED_BUFFER_VAR
+ COMPILE_STACK_TOP
.begalt_offset
;
3785 = COMPILE_STACK_TOP
.fixup_alt_jump
3786 ? COMPILED_BUFFER_VAR
+ COMPILE_STACK_TOP
.fixup_alt_jump
- 1
3788 laststart
= COMPILED_BUFFER_VAR
+ COMPILE_STACK_TOP
.laststart_offset
;
3789 this_group_regnum
= COMPILE_STACK_TOP
.regnum
;
3790 /* If we've reached MAX_REGNUM groups, then this open
3791 won't actually generate any code, so we'll have to
3792 clear pending_exact explicitly. */
3795 /* We're at the end of the group, so now we know how many
3796 groups were inside this one. */
3797 if (this_group_regnum
<= MAX_REGNUM
)
3799 UCHAR_T
*inner_group_loc
3800 = COMPILED_BUFFER_VAR
+ COMPILE_STACK_TOP
.inner_group_offset
;
3802 *inner_group_loc
= regnum
- this_group_regnum
;
3803 BUF_PUSH_3 (stop_memory
, this_group_regnum
,
3804 regnum
- this_group_regnum
);
3810 case '|': /* `\|'. */
3811 if (syntax
& RE_LIMITED_OPS
|| syntax
& RE_NO_BK_VBAR
)
3812 goto normal_backslash
;
3814 if (syntax
& RE_LIMITED_OPS
)
3817 /* Insert before the previous alternative a jump which
3818 jumps to this alternative if the former fails. */
3819 GET_BUFFER_SPACE (1 + OFFSET_ADDRESS_SIZE
);
3820 INSERT_JUMP (on_failure_jump
, begalt
,
3821 b
+ 2 + 2 * OFFSET_ADDRESS_SIZE
);
3823 b
+= 1 + OFFSET_ADDRESS_SIZE
;
3825 /* The alternative before this one has a jump after it
3826 which gets executed if it gets matched. Adjust that
3827 jump so it will jump to this alternative's analogous
3828 jump (put in below, which in turn will jump to the next
3829 (if any) alternative's such jump, etc.). The last such
3830 jump jumps to the correct final destination. A picture:
3836 If we are at `b', then fixup_alt_jump right now points to a
3837 three-byte space after `a'. We'll put in the jump, set
3838 fixup_alt_jump to right after `b', and leave behind three
3839 bytes which we'll fill in when we get to after `c'. */
3842 STORE_JUMP (jump_past_alt
, fixup_alt_jump
, b
);
3844 /* Mark and leave space for a jump after this alternative,
3845 to be filled in later either by next alternative or
3846 when know we're at the end of a series of alternatives. */
3848 GET_BUFFER_SPACE (1 + OFFSET_ADDRESS_SIZE
);
3849 b
+= 1 + OFFSET_ADDRESS_SIZE
;
3857 /* If \{ is a literal. */
3858 if (!(syntax
& RE_INTERVALS
)
3859 /* If we're at `\{' and it's not the open-interval
3861 || (syntax
& RE_NO_BK_BRACES
))
3862 goto normal_backslash
;
3866 /* If got here, then the syntax allows intervals. */
3868 /* At least (most) this many matches must be made. */
3869 int lower_bound
= -1, upper_bound
= -1;
3871 /* Place in the uncompiled pattern (i.e., just after
3872 the '{') to go back to if the interval is invalid. */
3873 const CHAR_T
*beg_interval
= p
;
3876 goto invalid_interval
;
3878 GET_UNSIGNED_NUMBER (lower_bound
);
3882 GET_UNSIGNED_NUMBER (upper_bound
);
3883 if (upper_bound
< 0)
3884 upper_bound
= RE_DUP_MAX
;
3887 /* Interval such as `{1}' => match exactly once. */
3888 upper_bound
= lower_bound
;
3890 if (! (0 <= lower_bound
&& lower_bound
<= upper_bound
))
3891 goto invalid_interval
;
3893 if (!(syntax
& RE_NO_BK_BRACES
))
3895 if (c
!= '\\' || p
== pend
)
3896 goto invalid_interval
;
3901 goto invalid_interval
;
3903 /* If it's invalid to have no preceding re. */
3906 if (syntax
& RE_CONTEXT_INVALID_OPS
3907 && !(syntax
& RE_INVALID_INTERVAL_ORD
))
3908 FREE_STACK_RETURN (REG_BADRPT
);
3909 else if (syntax
& RE_CONTEXT_INDEP_OPS
)
3912 goto unfetch_interval
;
3915 /* We just parsed a valid interval. */
3917 if (RE_DUP_MAX
< upper_bound
)
3918 FREE_STACK_RETURN (REG_BADBR
);
3920 /* If the upper bound is zero, don't want to succeed at
3921 all; jump from `laststart' to `b + 3', which will be
3922 the end of the buffer after we insert the jump. */
3923 /* ifdef WCHAR, 'b + 1 + OFFSET_ADDRESS_SIZE'
3924 instead of 'b + 3'. */
3925 if (upper_bound
== 0)
3927 GET_BUFFER_SPACE (1 + OFFSET_ADDRESS_SIZE
);
3928 INSERT_JUMP (jump
, laststart
, b
+ 1
3929 + OFFSET_ADDRESS_SIZE
);
3930 b
+= 1 + OFFSET_ADDRESS_SIZE
;
3933 /* Otherwise, we have a nontrivial interval. When
3934 we're all done, the pattern will look like:
3935 set_number_at <jump count> <upper bound>
3936 set_number_at <succeed_n count> <lower bound>
3937 succeed_n <after jump addr> <succeed_n count>
3939 jump_n <succeed_n addr> <jump count>
3940 (The upper bound and `jump_n' are omitted if
3941 `upper_bound' is 1, though.) */
3943 { /* If the upper bound is > 1, we need to insert
3944 more at the end of the loop. */
3945 unsigned nbytes
= 2 + 4 * OFFSET_ADDRESS_SIZE
+
3946 (upper_bound
> 1) * (2 + 4 * OFFSET_ADDRESS_SIZE
);
3948 GET_BUFFER_SPACE (nbytes
);
3950 /* Initialize lower bound of the `succeed_n', even
3951 though it will be set during matching by its
3952 attendant `set_number_at' (inserted next),
3953 because `re_compile_fastmap' needs to know.
3954 Jump to the `jump_n' we might insert below. */
3955 INSERT_JUMP2 (succeed_n
, laststart
,
3956 b
+ 1 + 2 * OFFSET_ADDRESS_SIZE
3957 + (upper_bound
> 1) * (1 + 2 * OFFSET_ADDRESS_SIZE
)
3959 b
+= 1 + 2 * OFFSET_ADDRESS_SIZE
;
3961 /* Code to initialize the lower bound. Insert
3962 before the `succeed_n'. The `5' is the last two
3963 bytes of this `set_number_at', plus 3 bytes of
3964 the following `succeed_n'. */
3965 /* ifdef WCHAR, The '1+2*OFFSET_ADDRESS_SIZE'
3966 is the 'set_number_at', plus '1+OFFSET_ADDRESS_SIZE'
3967 of the following `succeed_n'. */
3968 PREFIX(insert_op2
) (set_number_at
, laststart
, 1
3969 + 2 * OFFSET_ADDRESS_SIZE
, lower_bound
, b
);
3970 b
+= 1 + 2 * OFFSET_ADDRESS_SIZE
;
3972 if (upper_bound
> 1)
3973 { /* More than one repetition is allowed, so
3974 append a backward jump to the `succeed_n'
3975 that starts this interval.
3977 When we've reached this during matching,
3978 we'll have matched the interval once, so
3979 jump back only `upper_bound - 1' times. */
3980 STORE_JUMP2 (jump_n
, b
, laststart
3981 + 2 * OFFSET_ADDRESS_SIZE
+ 1,
3983 b
+= 1 + 2 * OFFSET_ADDRESS_SIZE
;
3985 /* The location we want to set is the second
3986 parameter of the `jump_n'; that is `b-2' as
3987 an absolute address. `laststart' will be
3988 the `set_number_at' we're about to insert;
3989 `laststart+3' the number to set, the source
3990 for the relative address. But we are
3991 inserting into the middle of the pattern --
3992 so everything is getting moved up by 5.
3993 Conclusion: (b - 2) - (laststart + 3) + 5,
3994 i.e., b - laststart.
3996 We insert this at the beginning of the loop
3997 so that if we fail during matching, we'll
3998 reinitialize the bounds. */
3999 PREFIX(insert_op2
) (set_number_at
, laststart
,
4001 upper_bound
- 1, b
);
4002 b
+= 1 + 2 * OFFSET_ADDRESS_SIZE
;
4009 if (!(syntax
& RE_INVALID_INTERVAL_ORD
))
4010 FREE_STACK_RETURN (p
== pend
? REG_EBRACE
: REG_BADBR
);
4012 /* Match the characters as literals. */
4015 if (syntax
& RE_NO_BK_BRACES
)
4018 goto normal_backslash
;
4022 /* There is no way to specify the before_dot and after_dot
4023 operators. rms says this is ok. --karl */
4031 BUF_PUSH_2 (syntaxspec
, syntax_spec_code
[c
]);
4037 BUF_PUSH_2 (notsyntaxspec
, syntax_spec_code
[c
]);
4043 if (syntax
& RE_NO_GNU_OPS
)
4046 BUF_PUSH (wordchar
);
4051 if (syntax
& RE_NO_GNU_OPS
)
4054 BUF_PUSH (notwordchar
);
4059 if (syntax
& RE_NO_GNU_OPS
)
4065 if (syntax
& RE_NO_GNU_OPS
)
4071 if (syntax
& RE_NO_GNU_OPS
)
4073 BUF_PUSH (wordbound
);
4077 if (syntax
& RE_NO_GNU_OPS
)
4079 BUF_PUSH (notwordbound
);
4083 if (syntax
& RE_NO_GNU_OPS
)
4089 if (syntax
& RE_NO_GNU_OPS
)
4094 case '1': case '2': case '3': case '4': case '5':
4095 case '6': case '7': case '8': case '9':
4096 if (syntax
& RE_NO_BK_REFS
)
4102 FREE_STACK_RETURN (REG_ESUBREG
);
4104 /* Can't back reference to a subexpression if inside of it. */
4105 if (group_in_compile_stack (compile_stack
, (regnum_t
) c1
))
4109 BUF_PUSH_2 (duplicate
, c1
);
4115 if (syntax
& RE_BK_PLUS_QM
)
4118 goto normal_backslash
;
4122 /* You might think it would be useful for \ to mean
4123 not to translate; but if we don't translate it
4124 it will never match anything. */
4132 /* Expects the character in `c'. */
4134 /* If no exactn currently being built. */
4137 /* If last exactn handle binary(or character) and
4138 new exactn handle character(or binary). */
4139 || is_exactn_bin
!= is_binary
[p
- 1 - pattern
]
4142 /* If last exactn not at current position. */
4143 || pending_exact
+ *pending_exact
+ 1 != b
4145 /* We have only one byte following the exactn for the count. */
4146 || *pending_exact
== (1 << BYTEWIDTH
) - 1
4148 /* If followed by a repetition operator. */
4149 || *p
== '*' || *p
== '^'
4150 || ((syntax
& RE_BK_PLUS_QM
)
4151 ? *p
== '\\' && (p
[1] == '+' || p
[1] == '?')
4152 : (*p
== '+' || *p
== '?'))
4153 || ((syntax
& RE_INTERVALS
)
4154 && ((syntax
& RE_NO_BK_BRACES
)
4156 : (p
[0] == '\\' && p
[1] == '{'))))
4158 /* Start building a new exactn. */
4163 /* Is this exactn binary data or character? */
4164 is_exactn_bin
= is_binary
[p
- 1 - pattern
];
4166 BUF_PUSH_2 (exactn_bin
, 0);
4168 BUF_PUSH_2 (exactn
, 0);
4170 BUF_PUSH_2 (exactn
, 0);
4172 pending_exact
= b
- 1;
4179 } /* while p != pend */
4182 /* Through the pattern now. */
4185 STORE_JUMP (jump_past_alt
, fixup_alt_jump
, b
);
4187 if (!COMPILE_STACK_EMPTY
)
4188 FREE_STACK_RETURN (REG_EPAREN
);
4190 /* If we don't want backtracking, force success
4191 the first time we reach the end of the compiled pattern. */
4192 if (syntax
& RE_NO_POSIX_BACKTRACKING
)
4200 free (compile_stack
.stack
);
4202 /* We have succeeded; set the length of the buffer. */
4204 bufp
->used
= (uintptr_t) b
- (uintptr_t) COMPILED_BUFFER_VAR
;
4206 bufp
->used
= b
- bufp
->buffer
;
4212 DEBUG_PRINT1 ("\nCompiled pattern: \n");
4213 PREFIX(print_compiled_pattern
) (bufp
);
4217 #ifndef MATCH_MAY_ALLOCATE
4218 /* Initialize the failure stack to the largest possible stack. This
4219 isn't necessary unless we're trying to avoid calling alloca in
4220 the search and match routines. */
4222 int num_regs
= bufp
->re_nsub
+ 1;
4224 /* Since DOUBLE_FAIL_STACK refuses to double only if the current size
4225 is strictly greater than re_max_failures, the largest possible stack
4226 is 2 * re_max_failures failure points. */
4227 if (fail_stack
.size
< (2 * re_max_failures
* MAX_FAILURE_ITEMS
))
4229 fail_stack
.size
= (2 * re_max_failures
* MAX_FAILURE_ITEMS
);
4232 if (! fail_stack
.stack
)
4234 = (PREFIX(fail_stack_elt_t
) *) xmalloc (fail_stack
.size
4235 * sizeof (PREFIX(fail_stack_elt_t
)));
4238 = (PREFIX(fail_stack_elt_t
) *) xrealloc (fail_stack
.stack
,
4240 * sizeof (PREFIX(fail_stack_elt_t
))));
4241 # else /* not emacs */
4242 if (! fail_stack
.stack
)
4244 = (PREFIX(fail_stack_elt_t
) *) malloc (fail_stack
.size
4245 * sizeof (PREFIX(fail_stack_elt_t
)));
4248 = (PREFIX(fail_stack_elt_t
) *) realloc (fail_stack
.stack
,
4250 * sizeof (PREFIX(fail_stack_elt_t
))));
4251 # endif /* not emacs */
4254 PREFIX(regex_grow_registers
) (num_regs
);
4256 #endif /* not MATCH_MAY_ALLOCATE */
4259 } /* regex_compile */
4261 /* Subroutines for `regex_compile'. */
4263 /* Store OP at LOC followed by two-byte integer parameter ARG. */
4264 /* ifdef WCHAR, integer parameter is 1 wchar_t. */
4267 PREFIX(store_op1
) (op
, loc
, arg
)
4272 *loc
= (UCHAR_T
) op
;
4273 STORE_NUMBER (loc
+ 1, arg
);
4277 /* Like `store_op1', but for two two-byte parameters ARG1 and ARG2. */
4278 /* ifdef WCHAR, integer parameter is 1 wchar_t. */
4281 PREFIX(store_op2
) (op
, loc
, arg1
, arg2
)
4286 *loc
= (UCHAR_T
) op
;
4287 STORE_NUMBER (loc
+ 1, arg1
);
4288 STORE_NUMBER (loc
+ 1 + OFFSET_ADDRESS_SIZE
, arg2
);
4292 /* Copy the bytes from LOC to END to open up three bytes of space at LOC
4293 for OP followed by two-byte integer parameter ARG. */
4294 /* ifdef WCHAR, integer parameter is 1 wchar_t. */
4297 PREFIX(insert_op1
) (op
, loc
, arg
, end
)
4303 register UCHAR_T
*pfrom
= end
;
4304 register UCHAR_T
*pto
= end
+ 1 + OFFSET_ADDRESS_SIZE
;
4306 while (pfrom
!= loc
)
4309 PREFIX(store_op1
) (op
, loc
, arg
);
4313 /* Like `insert_op1', but for two two-byte parameters ARG1 and ARG2. */
4314 /* ifdef WCHAR, integer parameter is 1 wchar_t. */
4317 PREFIX(insert_op2
) (op
, loc
, arg1
, arg2
, end
)
4323 register UCHAR_T
*pfrom
= end
;
4324 register UCHAR_T
*pto
= end
+ 1 + 2 * OFFSET_ADDRESS_SIZE
;
4326 while (pfrom
!= loc
)
4329 PREFIX(store_op2
) (op
, loc
, arg1
, arg2
);
4333 /* P points to just after a ^ in PATTERN. Return true if that ^ comes
4334 after an alternative or a begin-subexpression. We assume there is at
4335 least one character before the ^. */
4338 PREFIX(at_begline_loc_p
) (pattern
, p
, syntax
)
4339 const CHAR_T
*pattern
, *p
;
4340 reg_syntax_t syntax
;
4342 const CHAR_T
*prev
= p
- 2;
4343 boolean prev_prev_backslash
= prev
> pattern
&& prev
[-1] == '\\';
4346 /* After a subexpression? */
4347 (*prev
== '(' && (syntax
& RE_NO_BK_PARENS
|| prev_prev_backslash
))
4348 /* After an alternative? */
4349 || (*prev
== '|' && (syntax
& RE_NO_BK_VBAR
|| prev_prev_backslash
));
4353 /* The dual of at_begline_loc_p. This one is for $. We assume there is
4354 at least one character after the $, i.e., `P < PEND'. */
4357 PREFIX(at_endline_loc_p
) (p
, pend
, syntax
)
4358 const CHAR_T
*p
, *pend
;
4359 reg_syntax_t syntax
;
4361 const CHAR_T
*next
= p
;
4362 boolean next_backslash
= *next
== '\\';
4363 const CHAR_T
*next_next
= p
+ 1 < pend
? p
+ 1 : 0;
4366 /* Before a subexpression? */
4367 (syntax
& RE_NO_BK_PARENS
? *next
== ')'
4368 : next_backslash
&& next_next
&& *next_next
== ')')
4369 /* Before an alternative? */
4370 || (syntax
& RE_NO_BK_VBAR
? *next
== '|'
4371 : next_backslash
&& next_next
&& *next_next
== '|');
4374 #else /* not INSIDE_RECURSION */
4376 /* Returns true if REGNUM is in one of COMPILE_STACK's elements and
4377 false if it's not. */
4380 group_in_compile_stack (compile_stack
, regnum
)
4381 compile_stack_type compile_stack
;
4386 for (this_element
= compile_stack
.avail
- 1;
4389 if (compile_stack
.stack
[this_element
].regnum
== regnum
)
4394 #endif /* not INSIDE_RECURSION */
4396 #ifdef INSIDE_RECURSION
4399 /* This insert space, which size is "num", into the pattern at "loc".
4400 "end" must point the end of the allocated buffer. */
4402 insert_space (num
, loc
, end
)
4407 register CHAR_T
*pto
= end
;
4408 register CHAR_T
*pfrom
= end
- num
;
4410 while (pfrom
>= loc
)
4416 static reg_errcode_t
4417 wcs_compile_range (range_start_char
, p_ptr
, pend
, translate
, syntax
, b
,
4419 CHAR_T range_start_char
;
4420 const CHAR_T
**p_ptr
, *pend
;
4421 CHAR_T
*char_set
, *b
;
4422 RE_TRANSLATE_TYPE translate
;
4423 reg_syntax_t syntax
;
4425 const CHAR_T
*p
= *p_ptr
;
4426 CHAR_T range_start
, range_end
;
4430 uint32_t start_val
, end_val
;
4436 nrules
= _NL_CURRENT_WORD (LC_COLLATE
, _NL_COLLATE_NRULES
);
4439 const char *collseq
= (const char *) _NL_CURRENT(LC_COLLATE
,
4440 _NL_COLLATE_COLLSEQWC
);
4441 const unsigned char *extra
= (const unsigned char *)
4442 _NL_CURRENT (LC_COLLATE
, _NL_COLLATE_SYMB_EXTRAMB
);
4444 if (range_start_char
< -1)
4446 /* range_start is a collating symbol. */
4448 /* Retreive the index and get collation sequence value. */
4449 wextra
= (int32_t*)(extra
+ char_set
[-range_start_char
]);
4450 start_val
= wextra
[1 + *wextra
];
4453 start_val
= collseq_table_lookup(collseq
, TRANSLATE(range_start_char
));
4455 end_val
= collseq_table_lookup (collseq
, TRANSLATE (p
[0]));
4457 /* Report an error if the range is empty and the syntax prohibits
4459 ret
= ((syntax
& RE_NO_EMPTY_RANGES
)
4460 && (start_val
> end_val
))? REG_ERANGE
: REG_NOERROR
;
4462 /* Insert space to the end of the char_ranges. */
4463 insert_space(2, b
- char_set
[5] - 2, b
- 1);
4464 *(b
- char_set
[5] - 2) = (wchar_t)start_val
;
4465 *(b
- char_set
[5] - 1) = (wchar_t)end_val
;
4466 char_set
[4]++; /* ranges_index */
4471 range_start
= (range_start_char
>= 0)? TRANSLATE (range_start_char
):
4473 range_end
= TRANSLATE (p
[0]);
4474 /* Report an error if the range is empty and the syntax prohibits
4476 ret
= ((syntax
& RE_NO_EMPTY_RANGES
)
4477 && (range_start
> range_end
))? REG_ERANGE
: REG_NOERROR
;
4479 /* Insert space to the end of the char_ranges. */
4480 insert_space(2, b
- char_set
[5] - 2, b
- 1);
4481 *(b
- char_set
[5] - 2) = range_start
;
4482 *(b
- char_set
[5] - 1) = range_end
;
4483 char_set
[4]++; /* ranges_index */
4485 /* Have to increment the pointer into the pattern string, so the
4486 caller isn't still at the ending character. */
4492 /* Read the ending character of a range (in a bracket expression) from the
4493 uncompiled pattern *P_PTR (which ends at PEND). We assume the
4494 starting character is in `P[-2]'. (`P[-1]' is the character `-'.)
4495 Then we set the translation of all bits between the starting and
4496 ending characters (inclusive) in the compiled pattern B.
4498 Return an error code.
4500 We use these short variable names so we can use the same macros as
4501 `regex_compile' itself. */
4503 static reg_errcode_t
4504 byte_compile_range (range_start_char
, p_ptr
, pend
, translate
, syntax
, b
)
4505 unsigned int range_start_char
;
4506 const char **p_ptr
, *pend
;
4507 RE_TRANSLATE_TYPE translate
;
4508 reg_syntax_t syntax
;
4512 const char *p
= *p_ptr
;
4515 const unsigned char *collseq
;
4516 unsigned int start_colseq
;
4517 unsigned int end_colseq
;
4525 /* Have to increment the pointer into the pattern string, so the
4526 caller isn't still at the ending character. */
4529 /* Report an error if the range is empty and the syntax prohibits this. */
4530 ret
= syntax
& RE_NO_EMPTY_RANGES
? REG_ERANGE
: REG_NOERROR
;
4533 collseq
= (const unsigned char *) _NL_CURRENT (LC_COLLATE
,
4534 _NL_COLLATE_COLLSEQMB
);
4536 start_colseq
= collseq
[(unsigned char) TRANSLATE (range_start_char
)];
4537 end_colseq
= collseq
[(unsigned char) TRANSLATE (p
[0])];
4538 for (this_char
= 0; this_char
<= (unsigned char) -1; ++this_char
)
4540 unsigned int this_colseq
= collseq
[(unsigned char) TRANSLATE (this_char
)];
4542 if (start_colseq
<= this_colseq
&& this_colseq
<= end_colseq
)
4544 SET_LIST_BIT (TRANSLATE (this_char
));
4549 /* Here we see why `this_char' has to be larger than an `unsigned
4550 char' -- we would otherwise go into an infinite loop, since all
4551 characters <= 0xff. */
4552 range_start_char
= TRANSLATE (range_start_char
);
4553 /* TRANSLATE(p[0]) is casted to char (not unsigned char) in TRANSLATE,
4554 and some compilers cast it to int implicitly, so following for_loop
4555 may fall to (almost) infinite loop.
4556 e.g. If translate[p[0]] = 0xff, end_char may equals to 0xffffffff.
4557 To avoid this, we cast p[0] to unsigned int and truncate it. */
4558 end_char
= ((unsigned)TRANSLATE(p
[0]) & ((1 << BYTEWIDTH
) - 1));
4560 for (this_char
= range_start_char
; this_char
<= end_char
; ++this_char
)
4562 SET_LIST_BIT (TRANSLATE (this_char
));
4571 /* re_compile_fastmap computes a ``fastmap'' for the compiled pattern in
4572 BUFP. A fastmap records which of the (1 << BYTEWIDTH) possible
4573 characters can start a string that matches the pattern. This fastmap
4574 is used by re_search to skip quickly over impossible starting points.
4576 The caller must supply the address of a (1 << BYTEWIDTH)-byte data
4577 area as BUFP->fastmap.
4579 We set the `fastmap', `fastmap_accurate', and `can_be_null' fields in
4582 Returns 0 if we succeed, -2 if an internal error. */
4585 /* local function for re_compile_fastmap.
4586 truncate wchar_t character to char. */
4587 static unsigned char truncate_wchar (CHAR_T c
);
4589 static unsigned char
4593 unsigned char buf
[MB_CUR_MAX
];
4596 memset (&state
, '\0', sizeof (state
));
4598 retval
= __wcrtomb (buf
, c
, &state
);
4600 retval
= wcrtomb (buf
, c
, &state
);
4602 return retval
> 0 ? buf
[0] : (unsigned char) c
;
4607 PREFIX(re_compile_fastmap
) (bufp
)
4608 struct re_pattern_buffer
*bufp
;
4611 #ifdef MATCH_MAY_ALLOCATE
4612 PREFIX(fail_stack_type
) fail_stack
;
4614 #ifndef REGEX_MALLOC
4618 register char *fastmap
= bufp
->fastmap
;
4621 /* We need to cast pattern to (wchar_t*), because we casted this compiled
4622 pattern to (char*) in regex_compile. */
4623 UCHAR_T
*pattern
= (UCHAR_T
*)bufp
->buffer
;
4624 register UCHAR_T
*pend
= (UCHAR_T
*) (bufp
->buffer
+ bufp
->used
);
4626 UCHAR_T
*pattern
= bufp
->buffer
;
4627 register UCHAR_T
*pend
= pattern
+ bufp
->used
;
4629 UCHAR_T
*p
= pattern
;
4632 /* This holds the pointer to the failure stack, when
4633 it is allocated relocatably. */
4634 fail_stack_elt_t
*failure_stack_ptr
;
4637 /* Assume that each path through the pattern can be null until
4638 proven otherwise. We set this false at the bottom of switch
4639 statement, to which we get only if a particular path doesn't
4640 match the empty string. */
4641 boolean path_can_be_null
= true;
4643 /* We aren't doing a `succeed_n' to begin with. */
4644 boolean succeed_n_p
= false;
4646 assert (fastmap
!= NULL
&& p
!= NULL
);
4649 bzero (fastmap
, 1 << BYTEWIDTH
); /* Assume nothing's valid. */
4650 bufp
->fastmap_accurate
= 1; /* It will be when we're done. */
4651 bufp
->can_be_null
= 0;
4655 if (p
== pend
|| *p
== (UCHAR_T
) succeed
)
4657 /* We have reached the (effective) end of pattern. */
4658 if (!FAIL_STACK_EMPTY ())
4660 bufp
->can_be_null
|= path_can_be_null
;
4662 /* Reset for next path. */
4663 path_can_be_null
= true;
4665 p
= fail_stack
.stack
[--fail_stack
.avail
].pointer
;
4673 /* We should never be about to go beyond the end of the pattern. */
4676 switch (SWITCH_ENUM_CAST ((re_opcode_t
) *p
++))
4679 /* I guess the idea here is to simply not bother with a fastmap
4680 if a backreference is used, since it's too hard to figure out
4681 the fastmap for the corresponding group. Setting
4682 `can_be_null' stops `re_search_2' from using the fastmap, so
4683 that is all we do. */
4685 bufp
->can_be_null
= 1;
4689 /* Following are the cases which match a character. These end
4694 fastmap
[truncate_wchar(p
[1])] = 1;
4708 /* It is hard to distinguish fastmap from (multi byte) characters
4709 which depends on current locale. */
4714 bufp
->can_be_null
= 1;
4718 for (j
= *p
++ * BYTEWIDTH
- 1; j
>= 0; j
--)
4719 if (p
[j
/ BYTEWIDTH
] & (1 << (j
% BYTEWIDTH
)))
4725 /* Chars beyond end of map must be allowed. */
4726 for (j
= *p
* BYTEWIDTH
; j
< (1 << BYTEWIDTH
); j
++)
4729 for (j
= *p
++ * BYTEWIDTH
- 1; j
>= 0; j
--)
4730 if (!(p
[j
/ BYTEWIDTH
] & (1 << (j
% BYTEWIDTH
))))
4736 for (j
= 0; j
< (1 << BYTEWIDTH
); j
++)
4737 if (SYNTAX (j
) == Sword
)
4743 for (j
= 0; j
< (1 << BYTEWIDTH
); j
++)
4744 if (SYNTAX (j
) != Sword
)
4751 int fastmap_newline
= fastmap
['\n'];
4753 /* `.' matches anything ... */
4754 for (j
= 0; j
< (1 << BYTEWIDTH
); j
++)
4757 /* ... except perhaps newline. */
4758 if (!(bufp
->syntax
& RE_DOT_NEWLINE
))
4759 fastmap
['\n'] = fastmap_newline
;
4761 /* Return if we have already set `can_be_null'; if we have,
4762 then the fastmap is irrelevant. Something's wrong here. */
4763 else if (bufp
->can_be_null
)
4766 /* Otherwise, have to check alternative paths. */
4773 for (j
= 0; j
< (1 << BYTEWIDTH
); j
++)
4774 if (SYNTAX (j
) == (enum syntaxcode
) k
)
4781 for (j
= 0; j
< (1 << BYTEWIDTH
); j
++)
4782 if (SYNTAX (j
) != (enum syntaxcode
) k
)
4787 /* All cases after this match the empty string. These end with
4807 case push_dummy_failure
:
4812 case pop_failure_jump
:
4813 case maybe_pop_jump
:
4816 case dummy_failure_jump
:
4817 EXTRACT_NUMBER_AND_INCR (j
, p
);
4822 /* Jump backward implies we just went through the body of a
4823 loop and matched nothing. Opcode jumped to should be
4824 `on_failure_jump' or `succeed_n'. Just treat it like an
4825 ordinary jump. For a * loop, it has pushed its failure
4826 point already; if so, discard that as redundant. */
4827 if ((re_opcode_t
) *p
!= on_failure_jump
4828 && (re_opcode_t
) *p
!= succeed_n
)
4832 EXTRACT_NUMBER_AND_INCR (j
, p
);
4835 /* If what's on the stack is where we are now, pop it. */
4836 if (!FAIL_STACK_EMPTY ()
4837 && fail_stack
.stack
[fail_stack
.avail
- 1].pointer
== p
)
4843 case on_failure_jump
:
4844 case on_failure_keep_string_jump
:
4845 handle_on_failure_jump
:
4846 EXTRACT_NUMBER_AND_INCR (j
, p
);
4848 /* For some patterns, e.g., `(a?)?', `p+j' here points to the
4849 end of the pattern. We don't want to push such a point,
4850 since when we restore it above, entering the switch will
4851 increment `p' past the end of the pattern. We don't need
4852 to push such a point since we obviously won't find any more
4853 fastmap entries beyond `pend'. Such a pattern can match
4854 the null string, though. */
4857 if (!PUSH_PATTERN_OP (p
+ j
, fail_stack
))
4859 RESET_FAIL_STACK ();
4864 bufp
->can_be_null
= 1;
4868 EXTRACT_NUMBER_AND_INCR (k
, p
); /* Skip the n. */
4869 succeed_n_p
= false;
4876 /* Get to the number of times to succeed. */
4877 p
+= OFFSET_ADDRESS_SIZE
;
4879 /* Increment p past the n for when k != 0. */
4880 EXTRACT_NUMBER_AND_INCR (k
, p
);
4883 p
-= 2 * OFFSET_ADDRESS_SIZE
;
4884 succeed_n_p
= true; /* Spaghetti code alert. */
4885 goto handle_on_failure_jump
;
4891 p
+= 2 * OFFSET_ADDRESS_SIZE
;
4902 abort (); /* We have listed all the cases. */
4905 /* Getting here means we have found the possible starting
4906 characters for one path of the pattern -- and that the empty
4907 string does not match. We need not follow this path further.
4908 Instead, look at the next alternative (remembered on the
4909 stack), or quit if no more. The test at the top of the loop
4910 does these things. */
4911 path_can_be_null
= false;
4915 /* Set `can_be_null' for the last path (also the first path, if the
4916 pattern is empty). */
4917 bufp
->can_be_null
|= path_can_be_null
;
4920 RESET_FAIL_STACK ();
4924 #else /* not INSIDE_RECURSION */
4927 re_compile_fastmap (bufp
)
4928 struct re_pattern_buffer
*bufp
;
4931 if (MB_CUR_MAX
!= 1)
4932 return wcs_re_compile_fastmap(bufp
);
4935 return byte_re_compile_fastmap(bufp
);
4936 } /* re_compile_fastmap */
4938 weak_alias (__re_compile_fastmap
, re_compile_fastmap
)
4942 /* Set REGS to hold NUM_REGS registers, storing them in STARTS and
4943 ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use
4944 this memory for recording register information. STARTS and ENDS
4945 must be allocated using the malloc library routine, and must each
4946 be at least NUM_REGS * sizeof (regoff_t) bytes long.
4948 If NUM_REGS == 0, then subsequent matches should allocate their own
4951 Unless this function is called, the first search or match using
4952 PATTERN_BUFFER will allocate its own register data, without
4953 freeing the old data. */
4956 re_set_registers (bufp
, regs
, num_regs
, starts
, ends
)
4957 struct re_pattern_buffer
*bufp
;
4958 struct re_registers
*regs
;
4960 regoff_t
*starts
, *ends
;
4964 bufp
->regs_allocated
= REGS_REALLOCATE
;
4965 regs
->num_regs
= num_regs
;
4966 regs
->start
= starts
;
4971 bufp
->regs_allocated
= REGS_UNALLOCATED
;
4973 regs
->start
= regs
->end
= (regoff_t
*) 0;
4977 weak_alias (__re_set_registers
, re_set_registers
)
4980 /* Searching routines. */
4982 /* Like re_search_2, below, but only one string is specified, and
4983 doesn't let you say where to stop matching. */
4986 re_search (bufp
, string
, size
, startpos
, range
, regs
)
4987 struct re_pattern_buffer
*bufp
;
4989 int size
, startpos
, range
;
4990 struct re_registers
*regs
;
4992 return re_search_2 (bufp
, NULL
, 0, string
, size
, startpos
, range
,
4996 weak_alias (__re_search
, re_search
)
5000 /* Using the compiled pattern in BUFP->buffer, first tries to match the
5001 virtual concatenation of STRING1 and STRING2, starting first at index
5002 STARTPOS, then at STARTPOS + 1, and so on.
5004 STRING1 and STRING2 have length SIZE1 and SIZE2, respectively.
5006 RANGE is how far to scan while trying to match. RANGE = 0 means try
5007 only at STARTPOS; in general, the last start tried is STARTPOS +
5010 In REGS, return the indices of the virtual concatenation of STRING1
5011 and STRING2 that matched the entire BUFP->buffer and its contained
5014 Do not consider matching one past the index STOP in the virtual
5015 concatenation of STRING1 and STRING2.
5017 We return either the position in the strings at which the match was
5018 found, -1 if no match, or -2 if error (such as failure
5022 re_search_2 (bufp
, string1
, size1
, string2
, size2
, startpos
, range
, regs
, stop
)
5023 struct re_pattern_buffer
*bufp
;
5024 const char *string1
, *string2
;
5028 struct re_registers
*regs
;
5032 if (MB_CUR_MAX
!= 1)
5033 return wcs_re_search_2 (bufp
, string1
, size1
, string2
, size2
, startpos
,
5037 return byte_re_search_2 (bufp
, string1
, size1
, string2
, size2
, startpos
,
5041 weak_alias (__re_search_2
, re_search_2
)
5044 #endif /* not INSIDE_RECURSION */
5046 #ifdef INSIDE_RECURSION
5048 #ifdef MATCH_MAY_ALLOCATE
5049 # define FREE_VAR(var) if (var) REGEX_FREE (var); var = NULL
5051 # define FREE_VAR(var) if (var) free (var); var = NULL
5055 # define MAX_ALLOCA_SIZE 2000
5057 # define FREE_WCS_BUFFERS() \
5059 if (size1 > MAX_ALLOCA_SIZE) \
5061 free (wcs_string1); \
5062 free (mbs_offset1); \
5066 FREE_VAR (wcs_string1); \
5067 FREE_VAR (mbs_offset1); \
5069 if (size2 > MAX_ALLOCA_SIZE) \
5071 free (wcs_string2); \
5072 free (mbs_offset2); \
5076 FREE_VAR (wcs_string2); \
5077 FREE_VAR (mbs_offset2); \
5085 PREFIX(re_search_2
) (bufp
, string1
, size1
, string2
, size2
, startpos
, range
,
5087 struct re_pattern_buffer
*bufp
;
5088 const char *string1
, *string2
;
5092 struct re_registers
*regs
;
5096 register char *fastmap
= bufp
->fastmap
;
5097 register RE_TRANSLATE_TYPE translate
= bufp
->translate
;
5098 int total_size
= size1
+ size2
;
5099 int endpos
= startpos
+ range
;
5101 /* We need wchar_t* buffers correspond to cstring1, cstring2. */
5102 wchar_t *wcs_string1
= NULL
, *wcs_string2
= NULL
;
5103 /* We need the size of wchar_t buffers correspond to csize1, csize2. */
5104 int wcs_size1
= 0, wcs_size2
= 0;
5105 /* offset buffer for optimizatoin. See convert_mbs_to_wc. */
5106 int *mbs_offset1
= NULL
, *mbs_offset2
= NULL
;
5107 /* They hold whether each wchar_t is binary data or not. */
5108 char *is_binary
= NULL
;
5111 /* Check for out-of-range STARTPOS. */
5112 if (startpos
< 0 || startpos
> total_size
)
5115 /* Fix up RANGE if it might eventually take us outside
5116 the virtual concatenation of STRING1 and STRING2.
5117 Make sure we won't move STARTPOS below 0 or above TOTAL_SIZE. */
5119 range
= 0 - startpos
;
5120 else if (endpos
> total_size
)
5121 range
= total_size
- startpos
;
5123 /* If the search isn't to be a backwards one, don't waste time in a
5124 search for a pattern that must be anchored. */
5125 if (bufp
->used
> 0 && range
> 0
5126 && ((re_opcode_t
) bufp
->buffer
[0] == begbuf
5127 /* `begline' is like `begbuf' if it cannot match at newlines. */
5128 || ((re_opcode_t
) bufp
->buffer
[0] == begline
5129 && !bufp
->newline_anchor
)))
5138 /* In a forward search for something that starts with \=.
5139 don't keep searching past point. */
5140 if (bufp
->used
> 0 && (re_opcode_t
) bufp
->buffer
[0] == at_dot
&& range
> 0)
5142 range
= PT
- startpos
;
5148 /* Update the fastmap now if not correct already. */
5149 if (fastmap
&& !bufp
->fastmap_accurate
)
5150 if (re_compile_fastmap (bufp
) == -2)
5154 /* Allocate wchar_t array for wcs_string1 and wcs_string2 and
5155 fill them with converted string. */
5158 if (size1
> MAX_ALLOCA_SIZE
)
5160 wcs_string1
= TALLOC (size1
+ 1, CHAR_T
);
5161 mbs_offset1
= TALLOC (size1
+ 1, int);
5162 is_binary
= TALLOC (size1
+ 1, char);
5166 wcs_string1
= REGEX_TALLOC (size1
+ 1, CHAR_T
);
5167 mbs_offset1
= REGEX_TALLOC (size1
+ 1, int);
5168 is_binary
= REGEX_TALLOC (size1
+ 1, char);
5170 if (!wcs_string1
|| !mbs_offset1
|| !is_binary
)
5172 if (size1
> MAX_ALLOCA_SIZE
)
5180 FREE_VAR (wcs_string1
);
5181 FREE_VAR (mbs_offset1
);
5182 FREE_VAR (is_binary
);
5186 wcs_size1
= convert_mbs_to_wcs(wcs_string1
, string1
, size1
,
5187 mbs_offset1
, is_binary
);
5188 wcs_string1
[wcs_size1
] = L
'\0'; /* for a sentinel */
5189 if (size1
> MAX_ALLOCA_SIZE
)
5192 FREE_VAR (is_binary
);
5196 if (size2
> MAX_ALLOCA_SIZE
)
5198 wcs_string2
= TALLOC (size2
+ 1, CHAR_T
);
5199 mbs_offset2
= TALLOC (size2
+ 1, int);
5200 is_binary
= TALLOC (size2
+ 1, char);
5204 wcs_string2
= REGEX_TALLOC (size2
+ 1, CHAR_T
);
5205 mbs_offset2
= REGEX_TALLOC (size2
+ 1, int);
5206 is_binary
= REGEX_TALLOC (size2
+ 1, char);
5208 if (!wcs_string2
|| !mbs_offset2
|| !is_binary
)
5210 FREE_WCS_BUFFERS ();
5211 if (size2
> MAX_ALLOCA_SIZE
)
5214 FREE_VAR (is_binary
);
5217 wcs_size2
= convert_mbs_to_wcs(wcs_string2
, string2
, size2
,
5218 mbs_offset2
, is_binary
);
5219 wcs_string2
[wcs_size2
] = L
'\0'; /* for a sentinel */
5220 if (size2
> MAX_ALLOCA_SIZE
)
5223 FREE_VAR (is_binary
);
5228 /* Loop through the string, looking for a place to start matching. */
5231 /* If a fastmap is supplied, skip quickly over characters that
5232 cannot be the start of a match. If the pattern can match the
5233 null string, however, we don't need to skip characters; we want
5234 the first null string. */
5235 if (fastmap
&& startpos
< total_size
&& !bufp
->can_be_null
)
5237 if (range
> 0) /* Searching forwards. */
5239 register const char *d
;
5240 register int lim
= 0;
5243 if (startpos
< size1
&& startpos
+ range
>= size1
)
5244 lim
= range
- (size1
- startpos
);
5246 d
= (startpos
>= size1
? string2
- size1
: string1
) + startpos
;
5248 /* Written out as an if-else to avoid testing `translate'
5252 && !fastmap
[(unsigned char)
5253 translate
[(unsigned char) *d
++]])
5256 while (range
> lim
&& !fastmap
[(unsigned char) *d
++])
5259 startpos
+= irange
- range
;
5261 else /* Searching backwards. */
5263 register CHAR_T c
= (size1
== 0 || startpos
>= size1
5264 ? string2
[startpos
- size1
]
5265 : string1
[startpos
]);
5267 if (!fastmap
[(unsigned char) TRANSLATE (c
)])
5272 /* If can't match the null string, and that's all we have left, fail. */
5273 if (range
>= 0 && startpos
== total_size
&& fastmap
5274 && !bufp
->can_be_null
)
5277 FREE_WCS_BUFFERS ();
5283 val
= wcs_re_match_2_internal (bufp
, string1
, size1
, string2
,
5284 size2
, startpos
, regs
, stop
,
5285 wcs_string1
, wcs_size1
,
5286 wcs_string2
, wcs_size2
,
5287 mbs_offset1
, mbs_offset2
);
5289 val
= byte_re_match_2_internal (bufp
, string1
, size1
, string2
,
5290 size2
, startpos
, regs
, stop
);
5293 #ifndef REGEX_MALLOC
5302 FREE_WCS_BUFFERS ();
5310 FREE_WCS_BUFFERS ();
5330 FREE_WCS_BUFFERS ();
5336 /* This converts PTR, a pointer into one of the search wchar_t strings
5337 `string1' and `string2' into an multibyte string offset from the
5338 beginning of that string. We use mbs_offset to optimize.
5339 See convert_mbs_to_wcs. */
5340 # define POINTER_TO_OFFSET(ptr) \
5341 (FIRST_STRING_P (ptr) \
5342 ? ((regoff_t)(mbs_offset1 != NULL? mbs_offset1[(ptr)-string1] : 0)) \
5343 : ((regoff_t)((mbs_offset2 != NULL? mbs_offset2[(ptr)-string2] : 0) \
5346 /* This converts PTR, a pointer into one of the search strings `string1'
5347 and `string2' into an offset from the beginning of that string. */
5348 # define POINTER_TO_OFFSET(ptr) \
5349 (FIRST_STRING_P (ptr) \
5350 ? ((regoff_t) ((ptr) - string1)) \
5351 : ((regoff_t) ((ptr) - string2 + size1)))
5354 /* Macros for dealing with the split strings in re_match_2. */
5356 #define MATCHING_IN_FIRST_STRING (dend == end_match_1)
5358 /* Call before fetching a character with *d. This switches over to
5359 string2 if necessary. */
5360 #define PREFETCH() \
5363 /* End of string2 => fail. */ \
5364 if (dend == end_match_2) \
5366 /* End of string1 => advance to string2. */ \
5368 dend = end_match_2; \
5371 /* Test if at very beginning or at very end of the virtual concatenation
5372 of `string1' and `string2'. If only one string, it's `string2'. */
5373 #define AT_STRINGS_BEG(d) ((d) == (size1 ? string1 : string2) || !size2)
5374 #define AT_STRINGS_END(d) ((d) == end2)
5377 /* Test if D points to a character which is word-constituent. We have
5378 two special cases to check for: if past the end of string1, look at
5379 the first character in string2; and if before the beginning of
5380 string2, look at the last character in string1. */
5382 /* Use internationalized API instead of SYNTAX. */
5383 # define WORDCHAR_P(d) \
5384 (iswalnum ((wint_t)((d) == end1 ? *string2 \
5385 : (d) == string2 - 1 ? *(end1 - 1) : *(d))) != 0 \
5386 || ((d) == end1 ? *string2 \
5387 : (d) == string2 - 1 ? *(end1 - 1) : *(d)) == L'_')
5389 # define WORDCHAR_P(d) \
5390 (SYNTAX ((d) == end1 ? *string2 \
5391 : (d) == string2 - 1 ? *(end1 - 1) : *(d)) \
5395 /* Disabled due to a compiler bug -- see comment at case wordbound */
5397 /* Test if the character before D and the one at D differ with respect
5398 to being word-constituent. */
5399 #define AT_WORD_BOUNDARY(d) \
5400 (AT_STRINGS_BEG (d) || AT_STRINGS_END (d) \
5401 || WORDCHAR_P (d - 1) != WORDCHAR_P (d))
5404 /* Free everything we malloc. */
5405 #ifdef MATCH_MAY_ALLOCATE
5407 # define FREE_VARIABLES() \
5409 REGEX_FREE_STACK (fail_stack.stack); \
5410 FREE_VAR (regstart); \
5411 FREE_VAR (regend); \
5412 FREE_VAR (old_regstart); \
5413 FREE_VAR (old_regend); \
5414 FREE_VAR (best_regstart); \
5415 FREE_VAR (best_regend); \
5416 FREE_VAR (reg_info); \
5417 FREE_VAR (reg_dummy); \
5418 FREE_VAR (reg_info_dummy); \
5419 if (!cant_free_wcs_buf) \
5421 FREE_VAR (string1); \
5422 FREE_VAR (string2); \
5423 FREE_VAR (mbs_offset1); \
5424 FREE_VAR (mbs_offset2); \
5428 # define FREE_VARIABLES() \
5430 REGEX_FREE_STACK (fail_stack.stack); \
5431 FREE_VAR (regstart); \
5432 FREE_VAR (regend); \
5433 FREE_VAR (old_regstart); \
5434 FREE_VAR (old_regend); \
5435 FREE_VAR (best_regstart); \
5436 FREE_VAR (best_regend); \
5437 FREE_VAR (reg_info); \
5438 FREE_VAR (reg_dummy); \
5439 FREE_VAR (reg_info_dummy); \
5444 # define FREE_VARIABLES() \
5446 if (!cant_free_wcs_buf) \
5448 FREE_VAR (string1); \
5449 FREE_VAR (string2); \
5450 FREE_VAR (mbs_offset1); \
5451 FREE_VAR (mbs_offset2); \
5455 # define FREE_VARIABLES() ((void)0) /* Do nothing! But inhibit gcc warning. */
5457 #endif /* not MATCH_MAY_ALLOCATE */
5459 /* These values must meet several constraints. They must not be valid
5460 register values; since we have a limit of 255 registers (because
5461 we use only one byte in the pattern for the register number), we can
5462 use numbers larger than 255. They must differ by 1, because of
5463 NUM_FAILURE_ITEMS above. And the value for the lowest register must
5464 be larger than the value for the highest register, so we do not try
5465 to actually save any registers when none are active. */
5466 #define NO_HIGHEST_ACTIVE_REG (1 << BYTEWIDTH)
5467 #define NO_LOWEST_ACTIVE_REG (NO_HIGHEST_ACTIVE_REG + 1)
5469 #else /* not INSIDE_RECURSION */
5470 /* Matching routines. */
5472 #ifndef emacs /* Emacs never uses this. */
5473 /* re_match is like re_match_2 except it takes only a single string. */
5476 re_match (bufp
, string
, size
, pos
, regs
)
5477 struct re_pattern_buffer
*bufp
;
5480 struct re_registers
*regs
;
5484 if (MB_CUR_MAX
!= 1)
5485 result
= wcs_re_match_2_internal (bufp
, NULL
, 0, string
, size
,
5487 NULL
, 0, NULL
, 0, NULL
, NULL
);
5490 result
= byte_re_match_2_internal (bufp
, NULL
, 0, string
, size
,
5492 # ifndef REGEX_MALLOC
5500 weak_alias (__re_match
, re_match
)
5502 #endif /* not emacs */
5504 #endif /* not INSIDE_RECURSION */
5506 #ifdef INSIDE_RECURSION
5507 static boolean
PREFIX(group_match_null_string_p
) _RE_ARGS ((UCHAR_T
**p
,
5509 PREFIX(register_info_type
) *reg_info
));
5510 static boolean
PREFIX(alt_match_null_string_p
) _RE_ARGS ((UCHAR_T
*p
,
5512 PREFIX(register_info_type
) *reg_info
));
5513 static boolean
PREFIX(common_op_match_null_string_p
) _RE_ARGS ((UCHAR_T
**p
,
5515 PREFIX(register_info_type
) *reg_info
));
5516 static int PREFIX(bcmp_translate
) _RE_ARGS ((const CHAR_T
*s1
, const CHAR_T
*s2
,
5517 int len
, char *translate
));
5518 #else /* not INSIDE_RECURSION */
5520 /* re_match_2 matches the compiled pattern in BUFP against the
5521 the (virtual) concatenation of STRING1 and STRING2 (of length SIZE1
5522 and SIZE2, respectively). We start matching at POS, and stop
5525 If REGS is non-null and the `no_sub' field of BUFP is nonzero, we
5526 store offsets for the substring each group matched in REGS. See the
5527 documentation for exactly how many groups we fill.
5529 We return -1 if no match, -2 if an internal error (such as the
5530 failure stack overflowing). Otherwise, we return the length of the
5531 matched substring. */
5534 re_match_2 (bufp
, string1
, size1
, string2
, size2
, pos
, regs
, stop
)
5535 struct re_pattern_buffer
*bufp
;
5536 const char *string1
, *string2
;
5539 struct re_registers
*regs
;
5544 if (MB_CUR_MAX
!= 1)
5545 result
= wcs_re_match_2_internal (bufp
, string1
, size1
, string2
, size2
,
5547 NULL
, 0, NULL
, 0, NULL
, NULL
);
5550 result
= byte_re_match_2_internal (bufp
, string1
, size1
, string2
, size2
,
5553 #ifndef REGEX_MALLOC
5561 weak_alias (__re_match_2
, re_match_2
)
5564 #endif /* not INSIDE_RECURSION */
5566 #ifdef INSIDE_RECURSION
5569 static int count_mbs_length
PARAMS ((int *, int));
5571 /* This check the substring (from 0, to length) of the multibyte string,
5572 to which offset_buffer correspond. And count how many wchar_t_characters
5573 the substring occupy. We use offset_buffer to optimization.
5574 See convert_mbs_to_wcs. */
5577 count_mbs_length(offset_buffer
, length
)
5583 /* Check whether the size is valid. */
5587 if (offset_buffer
== NULL
)
5590 /* If there are no multibyte character, offset_buffer[i] == i.
5591 Optmize for this case. */
5592 if (offset_buffer
[length
] == length
)
5595 /* Set up upper with length. (because for all i, offset_buffer[i] >= i) */
5601 int middle
= (lower
+ upper
) / 2;
5602 if (middle
== lower
|| middle
== upper
)
5604 if (offset_buffer
[middle
] > length
)
5606 else if (offset_buffer
[middle
] < length
)
5616 /* This is a separate function so that we can force an alloca cleanup
5620 wcs_re_match_2_internal (bufp
, cstring1
, csize1
, cstring2
, csize2
, pos
,
5621 regs
, stop
, string1
, size1
, string2
, size2
,
5622 mbs_offset1
, mbs_offset2
)
5623 struct re_pattern_buffer
*bufp
;
5624 const char *cstring1
, *cstring2
;
5627 struct re_registers
*regs
;
5629 /* string1 == string2 == NULL means string1/2, size1/2 and
5630 mbs_offset1/2 need seting up in this function. */
5631 /* We need wchar_t* buffers correspond to cstring1, cstring2. */
5632 wchar_t *string1
, *string2
;
5633 /* We need the size of wchar_t buffers correspond to csize1, csize2. */
5635 /* offset buffer for optimizatoin. See convert_mbs_to_wc. */
5636 int *mbs_offset1
, *mbs_offset2
;
5639 byte_re_match_2_internal (bufp
, string1
, size1
,string2
, size2
, pos
,
5641 struct re_pattern_buffer
*bufp
;
5642 const char *string1
, *string2
;
5645 struct re_registers
*regs
;
5649 /* General temporaries. */
5653 /* They hold whether each wchar_t is binary data or not. */
5654 char *is_binary
= NULL
;
5655 /* If true, we can't free string1/2, mbs_offset1/2. */
5656 int cant_free_wcs_buf
= 1;
5659 /* Just past the end of the corresponding string. */
5660 const CHAR_T
*end1
, *end2
;
5662 /* Pointers into string1 and string2, just past the last characters in
5663 each to consider matching. */
5664 const CHAR_T
*end_match_1
, *end_match_2
;
5666 /* Where we are in the data, and the end of the current string. */
5667 const CHAR_T
*d
, *dend
;
5669 /* Where we are in the pattern, and the end of the pattern. */
5671 UCHAR_T
*pattern
, *p
;
5672 register UCHAR_T
*pend
;
5674 UCHAR_T
*p
= bufp
->buffer
;
5675 register UCHAR_T
*pend
= p
+ bufp
->used
;
5678 /* Mark the opcode just after a start_memory, so we can test for an
5679 empty subpattern when we get to the stop_memory. */
5680 UCHAR_T
*just_past_start_mem
= 0;
5682 /* We use this to map every character in the string. */
5683 RE_TRANSLATE_TYPE translate
= bufp
->translate
;
5685 /* Failure point stack. Each place that can handle a failure further
5686 down the line pushes a failure point on this stack. It consists of
5687 restart, regend, and reg_info for all registers corresponding to
5688 the subexpressions we're currently inside, plus the number of such
5689 registers, and, finally, two char *'s. The first char * is where
5690 to resume scanning the pattern; the second one is where to resume
5691 scanning the strings. If the latter is zero, the failure point is
5692 a ``dummy''; if a failure happens and the failure point is a dummy,
5693 it gets discarded and the next next one is tried. */
5694 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
5695 PREFIX(fail_stack_type
) fail_stack
;
5698 static unsigned failure_id
;
5699 unsigned nfailure_points_pushed
= 0, nfailure_points_popped
= 0;
5703 /* This holds the pointer to the failure stack, when
5704 it is allocated relocatably. */
5705 fail_stack_elt_t
*failure_stack_ptr
;
5708 /* We fill all the registers internally, independent of what we
5709 return, for use in backreferences. The number here includes
5710 an element for register zero. */
5711 size_t num_regs
= bufp
->re_nsub
+ 1;
5713 /* The currently active registers. */
5714 active_reg_t lowest_active_reg
= NO_LOWEST_ACTIVE_REG
;
5715 active_reg_t highest_active_reg
= NO_HIGHEST_ACTIVE_REG
;
5717 /* Information on the contents of registers. These are pointers into
5718 the input strings; they record just what was matched (on this
5719 attempt) by a subexpression part of the pattern, that is, the
5720 regnum-th regstart pointer points to where in the pattern we began
5721 matching and the regnum-th regend points to right after where we
5722 stopped matching the regnum-th subexpression. (The zeroth register
5723 keeps track of what the whole pattern matches.) */
5724 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
5725 const CHAR_T
**regstart
, **regend
;
5728 /* If a group that's operated upon by a repetition operator fails to
5729 match anything, then the register for its start will need to be
5730 restored because it will have been set to wherever in the string we
5731 are when we last see its open-group operator. Similarly for a
5733 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
5734 const CHAR_T
**old_regstart
, **old_regend
;
5737 /* The is_active field of reg_info helps us keep track of which (possibly
5738 nested) subexpressions we are currently in. The matched_something
5739 field of reg_info[reg_num] helps us tell whether or not we have
5740 matched any of the pattern so far this time through the reg_num-th
5741 subexpression. These two fields get reset each time through any
5742 loop their register is in. */
5743 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
5744 PREFIX(register_info_type
) *reg_info
;
5747 /* The following record the register info as found in the above
5748 variables when we find a match better than any we've seen before.
5749 This happens as we backtrack through the failure points, which in
5750 turn happens only if we have not yet matched the entire string. */
5751 unsigned best_regs_set
= false;
5752 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
5753 const CHAR_T
**best_regstart
, **best_regend
;
5756 /* Logically, this is `best_regend[0]'. But we don't want to have to
5757 allocate space for that if we're not allocating space for anything
5758 else (see below). Also, we never need info about register 0 for
5759 any of the other register vectors, and it seems rather a kludge to
5760 treat `best_regend' differently than the rest. So we keep track of
5761 the end of the best match so far in a separate variable. We
5762 initialize this to NULL so that when we backtrack the first time
5763 and need to test it, it's not garbage. */
5764 const CHAR_T
*match_end
= NULL
;
5766 /* This helps SET_REGS_MATCHED avoid doing redundant work. */
5767 int set_regs_matched_done
= 0;
5769 /* Used when we pop values we don't care about. */
5770 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
5771 const CHAR_T
**reg_dummy
;
5772 PREFIX(register_info_type
) *reg_info_dummy
;
5776 /* Counts the total number of registers pushed. */
5777 unsigned num_regs_pushed
= 0;
5780 DEBUG_PRINT1 ("\n\nEntering re_match_2.\n");
5784 #ifdef MATCH_MAY_ALLOCATE
5785 /* Do not bother to initialize all the register variables if there are
5786 no groups in the pattern, as it takes a fair amount of time. If
5787 there are groups, we include space for register 0 (the whole
5788 pattern), even though we never use it, since it simplifies the
5789 array indexing. We should fix this. */
5792 regstart
= REGEX_TALLOC (num_regs
, const CHAR_T
*);
5793 regend
= REGEX_TALLOC (num_regs
, const CHAR_T
*);
5794 old_regstart
= REGEX_TALLOC (num_regs
, const CHAR_T
*);
5795 old_regend
= REGEX_TALLOC (num_regs
, const CHAR_T
*);
5796 best_regstart
= REGEX_TALLOC (num_regs
, const CHAR_T
*);
5797 best_regend
= REGEX_TALLOC (num_regs
, const CHAR_T
*);
5798 reg_info
= REGEX_TALLOC (num_regs
, PREFIX(register_info_type
));
5799 reg_dummy
= REGEX_TALLOC (num_regs
, const CHAR_T
*);
5800 reg_info_dummy
= REGEX_TALLOC (num_regs
, PREFIX(register_info_type
));
5802 if (!(regstart
&& regend
&& old_regstart
&& old_regend
&& reg_info
5803 && best_regstart
&& best_regend
&& reg_dummy
&& reg_info_dummy
))
5811 /* We must initialize all our variables to NULL, so that
5812 `FREE_VARIABLES' doesn't try to free them. */
5813 regstart
= regend
= old_regstart
= old_regend
= best_regstart
5814 = best_regend
= reg_dummy
= NULL
;
5815 reg_info
= reg_info_dummy
= (PREFIX(register_info_type
) *) NULL
;
5817 #endif /* MATCH_MAY_ALLOCATE */
5819 /* The starting position is bogus. */
5821 if (pos
< 0 || pos
> csize1
+ csize2
)
5823 if (pos
< 0 || pos
> size1
+ size2
)
5831 /* Allocate wchar_t array for string1 and string2 and
5832 fill them with converted string. */
5833 if (string1
== NULL
&& string2
== NULL
)
5835 /* We need seting up buffers here. */
5837 /* We must free wcs buffers in this function. */
5838 cant_free_wcs_buf
= 0;
5842 string1
= REGEX_TALLOC (csize1
+ 1, CHAR_T
);
5843 mbs_offset1
= REGEX_TALLOC (csize1
+ 1, int);
5844 is_binary
= REGEX_TALLOC (csize1
+ 1, char);
5845 if (!string1
|| !mbs_offset1
|| !is_binary
)
5848 FREE_VAR (mbs_offset1
);
5849 FREE_VAR (is_binary
);
5855 string2
= REGEX_TALLOC (csize2
+ 1, CHAR_T
);
5856 mbs_offset2
= REGEX_TALLOC (csize2
+ 1, int);
5857 is_binary
= REGEX_TALLOC (csize2
+ 1, char);
5858 if (!string2
|| !mbs_offset2
|| !is_binary
)
5861 FREE_VAR (mbs_offset1
);
5863 FREE_VAR (mbs_offset2
);
5864 FREE_VAR (is_binary
);
5867 size2
= convert_mbs_to_wcs(string2
, cstring2
, csize2
,
5868 mbs_offset2
, is_binary
);
5869 string2
[size2
] = L
'\0'; /* for a sentinel */
5870 FREE_VAR (is_binary
);
5874 /* We need to cast pattern to (wchar_t*), because we casted this compiled
5875 pattern to (char*) in regex_compile. */
5876 p
= pattern
= (CHAR_T
*)bufp
->buffer
;
5877 pend
= (CHAR_T
*)(bufp
->buffer
+ bufp
->used
);
5881 /* Initialize subexpression text positions to -1 to mark ones that no
5882 start_memory/stop_memory has been seen for. Also initialize the
5883 register information struct. */
5884 for (mcnt
= 1; (unsigned) mcnt
< num_regs
; mcnt
++)
5886 regstart
[mcnt
] = regend
[mcnt
]
5887 = old_regstart
[mcnt
] = old_regend
[mcnt
] = REG_UNSET_VALUE
;
5889 REG_MATCH_NULL_STRING_P (reg_info
[mcnt
]) = MATCH_NULL_UNSET_VALUE
;
5890 IS_ACTIVE (reg_info
[mcnt
]) = 0;
5891 MATCHED_SOMETHING (reg_info
[mcnt
]) = 0;
5892 EVER_MATCHED_SOMETHING (reg_info
[mcnt
]) = 0;
5895 /* We move `string1' into `string2' if the latter's empty -- but not if
5896 `string1' is null. */
5897 if (size2
== 0 && string1
!= NULL
)
5904 mbs_offset2
= mbs_offset1
;
5910 end1
= string1
+ size1
;
5911 end2
= string2
+ size2
;
5913 /* Compute where to stop matching, within the two strings. */
5917 mcnt
= count_mbs_length(mbs_offset1
, stop
);
5918 end_match_1
= string1
+ mcnt
;
5919 end_match_2
= string2
;
5923 if (stop
> csize1
+ csize2
)
5924 stop
= csize1
+ csize2
;
5926 mcnt
= count_mbs_length(mbs_offset2
, stop
-csize1
);
5927 end_match_2
= string2
+ mcnt
;
5930 { /* count_mbs_length return error. */
5937 end_match_1
= string1
+ stop
;
5938 end_match_2
= string2
;
5943 end_match_2
= string2
+ stop
- size1
;
5947 /* `p' scans through the pattern as `d' scans through the data.
5948 `dend' is the end of the input string that `d' points within. `d'
5949 is advanced into the following input string whenever necessary, but
5950 this happens before fetching; therefore, at the beginning of the
5951 loop, `d' can be pointing at the end of a string, but it cannot
5954 if (size1
> 0 && pos
<= csize1
)
5956 mcnt
= count_mbs_length(mbs_offset1
, pos
);
5962 mcnt
= count_mbs_length(mbs_offset2
, pos
-csize1
);
5968 { /* count_mbs_length return error. */
5973 if (size1
> 0 && pos
<= size1
)
5980 d
= string2
+ pos
- size1
;
5985 DEBUG_PRINT1 ("The compiled pattern is:\n");
5986 DEBUG_PRINT_COMPILED_PATTERN (bufp
, p
, pend
);
5987 DEBUG_PRINT1 ("The string to match is: `");
5988 DEBUG_PRINT_DOUBLE_STRING (d
, string1
, size1
, string2
, size2
);
5989 DEBUG_PRINT1 ("'\n");
5991 /* This loops over pattern commands. It exits by returning from the
5992 function if the match is complete, or it drops through if the match
5993 fails at this starting point in the input data. */
5997 DEBUG_PRINT2 ("\n%p: ", p
);
5999 DEBUG_PRINT2 ("\n0x%x: ", p
);
6003 { /* End of pattern means we might have succeeded. */
6004 DEBUG_PRINT1 ("end of pattern ... ");
6006 /* If we haven't matched the entire string, and we want the
6007 longest match, try backtracking. */
6008 if (d
!= end_match_2
)
6010 /* 1 if this match ends in the same string (string1 or string2)
6011 as the best previous match. */
6012 boolean same_str_p
= (FIRST_STRING_P (match_end
)
6013 == MATCHING_IN_FIRST_STRING
);
6014 /* 1 if this match is the best seen so far. */
6015 boolean best_match_p
;
6017 /* AIX compiler got confused when this was combined
6018 with the previous declaration. */
6020 best_match_p
= d
> match_end
;
6022 best_match_p
= !MATCHING_IN_FIRST_STRING
;
6024 DEBUG_PRINT1 ("backtracking.\n");
6026 if (!FAIL_STACK_EMPTY ())
6027 { /* More failure points to try. */
6029 /* If exceeds best match so far, save it. */
6030 if (!best_regs_set
|| best_match_p
)
6032 best_regs_set
= true;
6035 DEBUG_PRINT1 ("\nSAVING match as best so far.\n");
6037 for (mcnt
= 1; (unsigned) mcnt
< num_regs
; mcnt
++)
6039 best_regstart
[mcnt
] = regstart
[mcnt
];
6040 best_regend
[mcnt
] = regend
[mcnt
];
6046 /* If no failure points, don't restore garbage. And if
6047 last match is real best match, don't restore second
6049 else if (best_regs_set
&& !best_match_p
)
6052 /* Restore best match. It may happen that `dend ==
6053 end_match_1' while the restored d is in string2.
6054 For example, the pattern `x.*y.*z' against the
6055 strings `x-' and `y-z-', if the two strings are
6056 not consecutive in memory. */
6057 DEBUG_PRINT1 ("Restoring best registers.\n");
6060 dend
= ((d
>= string1
&& d
<= end1
)
6061 ? end_match_1
: end_match_2
);
6063 for (mcnt
= 1; (unsigned) mcnt
< num_regs
; mcnt
++)
6065 regstart
[mcnt
] = best_regstart
[mcnt
];
6066 regend
[mcnt
] = best_regend
[mcnt
];
6069 } /* d != end_match_2 */
6072 DEBUG_PRINT1 ("Accepting match.\n");
6073 /* If caller wants register contents data back, do it. */
6074 if (regs
&& !bufp
->no_sub
)
6076 /* Have the register data arrays been allocated? */
6077 if (bufp
->regs_allocated
== REGS_UNALLOCATED
)
6078 { /* No. So allocate them with malloc. We need one
6079 extra element beyond `num_regs' for the `-1' marker
6081 regs
->num_regs
= MAX (RE_NREGS
, num_regs
+ 1);
6082 regs
->start
= TALLOC (regs
->num_regs
, regoff_t
);
6083 regs
->end
= TALLOC (regs
->num_regs
, regoff_t
);
6084 if (regs
->start
== NULL
|| regs
->end
== NULL
)
6089 bufp
->regs_allocated
= REGS_REALLOCATE
;
6091 else if (bufp
->regs_allocated
== REGS_REALLOCATE
)
6092 { /* Yes. If we need more elements than were already
6093 allocated, reallocate them. If we need fewer, just
6095 if (regs
->num_regs
< num_regs
+ 1)
6097 regs
->num_regs
= num_regs
+ 1;
6098 RETALLOC (regs
->start
, regs
->num_regs
, regoff_t
);
6099 RETALLOC (regs
->end
, regs
->num_regs
, regoff_t
);
6100 if (regs
->start
== NULL
|| regs
->end
== NULL
)
6109 /* These braces fend off a "empty body in an else-statement"
6110 warning under GCC when assert expands to nothing. */
6111 assert (bufp
->regs_allocated
== REGS_FIXED
);
6114 /* Convert the pointer data in `regstart' and `regend' to
6115 indices. Register zero has to be set differently,
6116 since we haven't kept track of any info for it. */
6117 if (regs
->num_regs
> 0)
6119 regs
->start
[0] = pos
;
6121 if (MATCHING_IN_FIRST_STRING
)
6122 regs
->end
[0] = mbs_offset1
!= NULL
?
6123 mbs_offset1
[d
-string1
] : 0;
6125 regs
->end
[0] = csize1
+ (mbs_offset2
!= NULL
?
6126 mbs_offset2
[d
-string2
] : 0);
6128 regs
->end
[0] = (MATCHING_IN_FIRST_STRING
6129 ? ((regoff_t
) (d
- string1
))
6130 : ((regoff_t
) (d
- string2
+ size1
)));
6134 /* Go through the first `min (num_regs, regs->num_regs)'
6135 registers, since that is all we initialized. */
6136 for (mcnt
= 1; (unsigned) mcnt
< MIN (num_regs
, regs
->num_regs
);
6139 if (REG_UNSET (regstart
[mcnt
]) || REG_UNSET (regend
[mcnt
]))
6140 regs
->start
[mcnt
] = regs
->end
[mcnt
] = -1;
6144 = (regoff_t
) POINTER_TO_OFFSET (regstart
[mcnt
]);
6146 = (regoff_t
) POINTER_TO_OFFSET (regend
[mcnt
]);
6150 /* If the regs structure we return has more elements than
6151 were in the pattern, set the extra elements to -1. If
6152 we (re)allocated the registers, this is the case,
6153 because we always allocate enough to have at least one
6155 for (mcnt
= num_regs
; (unsigned) mcnt
< regs
->num_regs
; mcnt
++)
6156 regs
->start
[mcnt
] = regs
->end
[mcnt
] = -1;
6157 } /* regs && !bufp->no_sub */
6159 DEBUG_PRINT4 ("%u failure points pushed, %u popped (%u remain).\n",
6160 nfailure_points_pushed
, nfailure_points_popped
,
6161 nfailure_points_pushed
- nfailure_points_popped
);
6162 DEBUG_PRINT2 ("%u registers pushed.\n", num_regs_pushed
);
6165 if (MATCHING_IN_FIRST_STRING
)
6166 mcnt
= mbs_offset1
!= NULL
? mbs_offset1
[d
-string1
] : 0;
6168 mcnt
= (mbs_offset2
!= NULL
? mbs_offset2
[d
-string2
] : 0) +
6172 mcnt
= d
- pos
- (MATCHING_IN_FIRST_STRING
6177 DEBUG_PRINT2 ("Returning %d from re_match_2.\n", mcnt
);
6183 /* Otherwise match next pattern command. */
6184 switch (SWITCH_ENUM_CAST ((re_opcode_t
) *p
++))
6186 /* Ignore these. Used to ignore the n of succeed_n's which
6187 currently have n == 0. */
6189 DEBUG_PRINT1 ("EXECUTING no_op.\n");
6193 DEBUG_PRINT1 ("EXECUTING succeed.\n");
6196 /* Match the next n pattern characters exactly. The following
6197 byte in the pattern defines n, and the n bytes after that
6198 are the characters to match. */
6204 DEBUG_PRINT2 ("EXECUTING exactn %d.\n", mcnt
);
6206 /* This is written out as an if-else so we don't waste time
6207 testing `translate' inside the loop. */
6216 if ((UCHAR_T
) translate
[(unsigned char) *d
++]
6222 if (*d
++ != (CHAR_T
) *p
++)
6226 if ((UCHAR_T
) translate
[(unsigned char) *d
++]
6238 if (*d
++ != (CHAR_T
) *p
++) goto fail
;
6242 SET_REGS_MATCHED ();
6246 /* Match any character except possibly a newline or a null. */
6248 DEBUG_PRINT1 ("EXECUTING anychar.\n");
6252 if ((!(bufp
->syntax
& RE_DOT_NEWLINE
) && TRANSLATE (*d
) == '\n')
6253 || (bufp
->syntax
& RE_DOT_NOT_NULL
&& TRANSLATE (*d
) == '\000'))
6256 SET_REGS_MATCHED ();
6257 DEBUG_PRINT2 (" Matched `%ld'.\n", (long int) *d
);
6267 unsigned int i
, char_class_length
, coll_symbol_length
,
6268 equiv_class_length
, ranges_length
, chars_length
, length
;
6269 CHAR_T
*workp
, *workp2
, *charset_top
;
6270 #define WORK_BUFFER_SIZE 128
6271 CHAR_T str_buf
[WORK_BUFFER_SIZE
];
6276 boolean
not = (re_opcode_t
) *(p
- 1) == charset_not
;
6278 DEBUG_PRINT2 ("EXECUTING charset%s.\n", not ? "_not" : "");
6280 c
= TRANSLATE (*d
); /* The character to match. */
6283 nrules
= _NL_CURRENT_WORD (LC_COLLATE
, _NL_COLLATE_NRULES
);
6285 charset_top
= p
- 1;
6286 char_class_length
= *p
++;
6287 coll_symbol_length
= *p
++;
6288 equiv_class_length
= *p
++;
6289 ranges_length
= *p
++;
6290 chars_length
= *p
++;
6291 /* p points charset[6], so the address of the next instruction
6292 (charset[l+m+n+2o+k+p']) equals p[l+m+n+2*o+p'],
6293 where l=length of char_classes, m=length of collating_symbol,
6294 n=equivalence_class, o=length of char_range,
6295 p'=length of character. */
6297 /* Update p to indicate the next instruction. */
6298 p
+= char_class_length
+ coll_symbol_length
+ equiv_class_length
+
6299 2*ranges_length
+ chars_length
;
6301 /* match with char_class? */
6302 for (i
= 0; i
< char_class_length
; i
+= CHAR_CLASS_SIZE
)
6305 uintptr_t alignedp
= ((uintptr_t)workp
6306 + __alignof__(wctype_t) - 1)
6307 & ~(uintptr_t)(__alignof__(wctype_t) - 1);
6308 wctype
= *((wctype_t*)alignedp
);
6309 workp
+= CHAR_CLASS_SIZE
;
6311 if (__iswctype((wint_t)c
, wctype
))
6312 goto char_set_matched
;
6314 if (iswctype((wint_t)c
, wctype
))
6315 goto char_set_matched
;
6319 /* match with collating_symbol? */
6323 const unsigned char *extra
= (const unsigned char *)
6324 _NL_CURRENT (LC_COLLATE
, _NL_COLLATE_SYMB_EXTRAMB
);
6326 for (workp2
= workp
+ coll_symbol_length
; workp
< workp2
;
6330 wextra
= (int32_t*)(extra
+ *workp
++);
6331 for (i
= 0; i
< *wextra
; ++i
)
6332 if (TRANSLATE(d
[i
]) != wextra
[1 + i
])
6337 /* Update d, however d will be incremented at
6338 char_set_matched:, we decrement d here. */
6340 goto char_set_matched
;
6344 else /* (nrules == 0) */
6346 /* If we can't look up collation data, we use wcscoll
6349 for (workp2
= workp
+ coll_symbol_length
; workp
< workp2
;)
6351 const CHAR_T
*backup_d
= d
, *backup_dend
= dend
;
6353 length
= __wcslen (workp
);
6355 length
= wcslen (workp
);
6358 /* If wcscoll(the collating symbol, whole string) > 0,
6359 any substring of the string never match with the
6360 collating symbol. */
6362 if (__wcscoll (workp
, d
) > 0)
6364 if (wcscoll (workp
, d
) > 0)
6367 workp
+= length
+ 1;
6371 /* First, we compare the collating symbol with
6372 the first character of the string.
6373 If it don't match, we add the next character to
6374 the compare buffer in turn. */
6375 for (i
= 0 ; i
< WORK_BUFFER_SIZE
-1 ; i
++, d
++)
6380 if (dend
== end_match_2
)
6386 /* add next character to the compare buffer. */
6387 str_buf
[i
] = TRANSLATE(*d
);
6388 str_buf
[i
+1] = '\0';
6391 match
= __wcscoll (workp
, str_buf
);
6393 match
= wcscoll (workp
, str_buf
);
6396 goto char_set_matched
;
6399 /* (str_buf > workp) indicate (str_buf + X > workp),
6400 because for all X (str_buf + X > str_buf).
6401 So we don't need continue this loop. */
6404 /* Otherwise(str_buf < workp),
6405 (str_buf+next_character) may equals (workp).
6406 So we continue this loop. */
6411 workp
+= length
+ 1;
6414 /* match with equivalence_class? */
6418 const CHAR_T
*backup_d
= d
, *backup_dend
= dend
;
6419 /* Try to match the equivalence class against
6420 those known to the collate implementation. */
6421 const int32_t *table
;
6422 const int32_t *weights
;
6423 const int32_t *extra
;
6424 const int32_t *indirect
;
6429 /* This #include defines a local function! */
6430 # include <locale/weightwc.h>
6432 table
= (const int32_t *)
6433 _NL_CURRENT (LC_COLLATE
, _NL_COLLATE_TABLEWC
);
6434 weights
= (const wint_t *)
6435 _NL_CURRENT (LC_COLLATE
, _NL_COLLATE_WEIGHTWC
);
6436 extra
= (const wint_t *)
6437 _NL_CURRENT (LC_COLLATE
, _NL_COLLATE_EXTRAWC
);
6438 indirect
= (const int32_t *)
6439 _NL_CURRENT (LC_COLLATE
, _NL_COLLATE_INDIRECTWC
);
6441 /* Write 1 collating element to str_buf, and
6445 for (i
= 0 ; idx2
== 0 && i
< WORK_BUFFER_SIZE
- 1; i
++)
6447 cp
= (wint_t*)str_buf
;
6450 if (dend
== end_match_2
)
6455 str_buf
[i
] = TRANSLATE(*(d
+i
));
6456 str_buf
[i
+1] = '\0'; /* sentinel */
6457 idx2
= findidx ((const wint_t**)&cp
);
6460 /* Update d, however d will be incremented at
6461 char_set_matched:, we decrement d here. */
6462 d
= backup_d
+ ((wchar_t*)cp
- (wchar_t*)str_buf
- 1);
6465 if (dend
== end_match_2
)
6474 len
= weights
[idx2
];
6476 for (workp2
= workp
+ equiv_class_length
; workp
< workp2
;
6479 idx
= (int32_t)*workp
;
6480 /* We already checked idx != 0 in regex_compile. */
6482 if (idx2
!= 0 && len
== weights
[idx
])
6485 while (cnt
< len
&& (weights
[idx
+ 1 + cnt
]
6486 == weights
[idx2
+ 1 + cnt
]))
6490 goto char_set_matched
;
6497 else /* (nrules == 0) */
6499 /* If we can't look up collation data, we use wcscoll
6502 for (workp2
= workp
+ equiv_class_length
; workp
< workp2
;)
6504 const CHAR_T
*backup_d
= d
, *backup_dend
= dend
;
6506 length
= __wcslen (workp
);
6508 length
= wcslen (workp
);
6511 /* If wcscoll(the collating symbol, whole string) > 0,
6512 any substring of the string never match with the
6513 collating symbol. */
6515 if (__wcscoll (workp
, d
) > 0)
6517 if (wcscoll (workp
, d
) > 0)
6520 workp
+= length
+ 1;
6524 /* First, we compare the equivalence class with
6525 the first character of the string.
6526 If it don't match, we add the next character to
6527 the compare buffer in turn. */
6528 for (i
= 0 ; i
< WORK_BUFFER_SIZE
- 1 ; i
++, d
++)
6533 if (dend
== end_match_2
)
6539 /* add next character to the compare buffer. */
6540 str_buf
[i
] = TRANSLATE(*d
);
6541 str_buf
[i
+1] = '\0';
6544 match
= __wcscoll (workp
, str_buf
);
6546 match
= wcscoll (workp
, str_buf
);
6550 goto char_set_matched
;
6553 /* (str_buf > workp) indicate (str_buf + X > workp),
6554 because for all X (str_buf + X > str_buf).
6555 So we don't need continue this loop. */
6558 /* Otherwise(str_buf < workp),
6559 (str_buf+next_character) may equals (workp).
6560 So we continue this loop. */
6565 workp
+= length
+ 1;
6569 /* match with char_range? */
6573 uint32_t collseqval
;
6574 const char *collseq
= (const char *)
6575 _NL_CURRENT(LC_COLLATE
, _NL_COLLATE_COLLSEQWC
);
6577 collseqval
= collseq_table_lookup (collseq
, c
);
6579 for (; workp
< p
- chars_length
;)
6581 uint32_t start_val
, end_val
;
6583 /* We already compute the collation sequence value
6584 of the characters (or collating symbols). */
6585 start_val
= (uint32_t) *workp
++; /* range_start */
6586 end_val
= (uint32_t) *workp
++; /* range_end */
6588 if (start_val
<= collseqval
&& collseqval
<= end_val
)
6589 goto char_set_matched
;
6595 /* We set range_start_char at str_buf[0], range_end_char
6596 at str_buf[4], and compared char at str_buf[2]. */
6601 for (; workp
< p
- chars_length
;)
6603 wchar_t *range_start_char
, *range_end_char
;
6605 /* match if (range_start_char <= c <= range_end_char). */
6607 /* If range_start(or end) < 0, we assume -range_start(end)
6608 is the offset of the collating symbol which is specified
6609 as the character of the range start(end). */
6613 range_start_char
= charset_top
- (*workp
++);
6616 str_buf
[0] = *workp
++;
6617 range_start_char
= str_buf
;
6622 range_end_char
= charset_top
- (*workp
++);
6625 str_buf
[4] = *workp
++;
6626 range_end_char
= str_buf
+ 4;
6630 if (__wcscoll (range_start_char
, str_buf
+2) <= 0
6631 && __wcscoll (str_buf
+2, range_end_char
) <= 0)
6633 if (wcscoll (range_start_char
, str_buf
+2) <= 0
6634 && wcscoll (str_buf
+2, range_end_char
) <= 0)
6636 goto char_set_matched
;
6640 /* match with char? */
6641 for (; workp
< p
; workp
++)
6643 goto char_set_matched
;
6650 /* Cast to `unsigned' instead of `unsigned char' in case the
6651 bit list is a full 32 bytes long. */
6652 if (c
< (unsigned) (*p
* BYTEWIDTH
)
6653 && p
[1 + c
/ BYTEWIDTH
] & (1 << (c
% BYTEWIDTH
)))
6658 if (!not) goto fail
;
6659 #undef WORK_BUFFER_SIZE
6661 SET_REGS_MATCHED ();
6667 /* The beginning of a group is represented by start_memory.
6668 The arguments are the register number in the next byte, and the
6669 number of groups inner to this one in the next. The text
6670 matched within the group is recorded (in the internal
6671 registers data structure) under the register number. */
6673 DEBUG_PRINT3 ("EXECUTING start_memory %ld (%ld):\n",
6674 (long int) *p
, (long int) p
[1]);
6676 /* Find out if this group can match the empty string. */
6677 p1
= p
; /* To send to group_match_null_string_p. */
6679 if (REG_MATCH_NULL_STRING_P (reg_info
[*p
]) == MATCH_NULL_UNSET_VALUE
)
6680 REG_MATCH_NULL_STRING_P (reg_info
[*p
])
6681 = PREFIX(group_match_null_string_p
) (&p1
, pend
, reg_info
);
6683 /* Save the position in the string where we were the last time
6684 we were at this open-group operator in case the group is
6685 operated upon by a repetition operator, e.g., with `(a*)*b'
6686 against `ab'; then we want to ignore where we are now in
6687 the string in case this attempt to match fails. */
6688 old_regstart
[*p
] = REG_MATCH_NULL_STRING_P (reg_info
[*p
])
6689 ? REG_UNSET (regstart
[*p
]) ? d
: regstart
[*p
]
6691 DEBUG_PRINT2 (" old_regstart: %d\n",
6692 POINTER_TO_OFFSET (old_regstart
[*p
]));
6695 DEBUG_PRINT2 (" regstart: %d\n", POINTER_TO_OFFSET (regstart
[*p
]));
6697 IS_ACTIVE (reg_info
[*p
]) = 1;
6698 MATCHED_SOMETHING (reg_info
[*p
]) = 0;
6700 /* Clear this whenever we change the register activity status. */
6701 set_regs_matched_done
= 0;
6703 /* This is the new highest active register. */
6704 highest_active_reg
= *p
;
6706 /* If nothing was active before, this is the new lowest active
6708 if (lowest_active_reg
== NO_LOWEST_ACTIVE_REG
)
6709 lowest_active_reg
= *p
;
6711 /* Move past the register number and inner group count. */
6713 just_past_start_mem
= p
;
6718 /* The stop_memory opcode represents the end of a group. Its
6719 arguments are the same as start_memory's: the register
6720 number, and the number of inner groups. */
6722 DEBUG_PRINT3 ("EXECUTING stop_memory %ld (%ld):\n",
6723 (long int) *p
, (long int) p
[1]);
6725 /* We need to save the string position the last time we were at
6726 this close-group operator in case the group is operated
6727 upon by a repetition operator, e.g., with `((a*)*(b*)*)*'
6728 against `aba'; then we want to ignore where we are now in
6729 the string in case this attempt to match fails. */
6730 old_regend
[*p
] = REG_MATCH_NULL_STRING_P (reg_info
[*p
])
6731 ? REG_UNSET (regend
[*p
]) ? d
: regend
[*p
]
6733 DEBUG_PRINT2 (" old_regend: %d\n",
6734 POINTER_TO_OFFSET (old_regend
[*p
]));
6737 DEBUG_PRINT2 (" regend: %d\n", POINTER_TO_OFFSET (regend
[*p
]));
6739 /* This register isn't active anymore. */
6740 IS_ACTIVE (reg_info
[*p
]) = 0;
6742 /* Clear this whenever we change the register activity status. */
6743 set_regs_matched_done
= 0;
6745 /* If this was the only register active, nothing is active
6747 if (lowest_active_reg
== highest_active_reg
)
6749 lowest_active_reg
= NO_LOWEST_ACTIVE_REG
;
6750 highest_active_reg
= NO_HIGHEST_ACTIVE_REG
;
6753 { /* We must scan for the new highest active register, since
6754 it isn't necessarily one less than now: consider
6755 (a(b)c(d(e)f)g). When group 3 ends, after the f), the
6756 new highest active register is 1. */
6758 while (r
> 0 && !IS_ACTIVE (reg_info
[r
]))
6761 /* If we end up at register zero, that means that we saved
6762 the registers as the result of an `on_failure_jump', not
6763 a `start_memory', and we jumped to past the innermost
6764 `stop_memory'. For example, in ((.)*) we save
6765 registers 1 and 2 as a result of the *, but when we pop
6766 back to the second ), we are at the stop_memory 1.
6767 Thus, nothing is active. */
6770 lowest_active_reg
= NO_LOWEST_ACTIVE_REG
;
6771 highest_active_reg
= NO_HIGHEST_ACTIVE_REG
;
6774 highest_active_reg
= r
;
6777 /* If just failed to match something this time around with a
6778 group that's operated on by a repetition operator, try to
6779 force exit from the ``loop'', and restore the register
6780 information for this group that we had before trying this
6782 if ((!MATCHED_SOMETHING (reg_info
[*p
])
6783 || just_past_start_mem
== p
- 1)
6786 boolean is_a_jump_n
= false;
6790 switch ((re_opcode_t
) *p1
++)
6794 case pop_failure_jump
:
6795 case maybe_pop_jump
:
6797 case dummy_failure_jump
:
6798 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
6800 p1
+= OFFSET_ADDRESS_SIZE
;
6808 /* If the next operation is a jump backwards in the pattern
6809 to an on_failure_jump right before the start_memory
6810 corresponding to this stop_memory, exit from the loop
6811 by forcing a failure after pushing on the stack the
6812 on_failure_jump's jump in the pattern, and d. */
6813 if (mcnt
< 0 && (re_opcode_t
) *p1
== on_failure_jump
6814 && (re_opcode_t
) p1
[1+OFFSET_ADDRESS_SIZE
] == start_memory
6815 && p1
[2+OFFSET_ADDRESS_SIZE
] == *p
)
6817 /* If this group ever matched anything, then restore
6818 what its registers were before trying this last
6819 failed match, e.g., with `(a*)*b' against `ab' for
6820 regstart[1], and, e.g., with `((a*)*(b*)*)*'
6821 against `aba' for regend[3].
6823 Also restore the registers for inner groups for,
6824 e.g., `((a*)(b*))*' against `aba' (register 3 would
6825 otherwise get trashed). */
6827 if (EVER_MATCHED_SOMETHING (reg_info
[*p
]))
6831 EVER_MATCHED_SOMETHING (reg_info
[*p
]) = 0;
6833 /* Restore this and inner groups' (if any) registers. */
6834 for (r
= *p
; r
< (unsigned) *p
+ (unsigned) *(p
+ 1);
6837 regstart
[r
] = old_regstart
[r
];
6839 /* xx why this test? */
6840 if (old_regend
[r
] >= regstart
[r
])
6841 regend
[r
] = old_regend
[r
];
6845 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
6846 PUSH_FAILURE_POINT (p1
+ mcnt
, d
, -2);
6852 /* Move past the register number and the inner group count. */
6857 /* \<digit> has been turned into a `duplicate' command which is
6858 followed by the numeric value of <digit> as the register number. */
6861 register const CHAR_T
*d2
, *dend2
;
6862 int regno
= *p
++; /* Get which register to match against. */
6863 DEBUG_PRINT2 ("EXECUTING duplicate %d.\n", regno
);
6865 /* Can't back reference a group which we've never matched. */
6866 if (REG_UNSET (regstart
[regno
]) || REG_UNSET (regend
[regno
]))
6869 /* Where in input to try to start matching. */
6870 d2
= regstart
[regno
];
6872 /* Where to stop matching; if both the place to start and
6873 the place to stop matching are in the same string, then
6874 set to the place to stop, otherwise, for now have to use
6875 the end of the first string. */
6877 dend2
= ((FIRST_STRING_P (regstart
[regno
])
6878 == FIRST_STRING_P (regend
[regno
]))
6879 ? regend
[regno
] : end_match_1
);
6882 /* If necessary, advance to next segment in register
6886 if (dend2
== end_match_2
) break;
6887 if (dend2
== regend
[regno
]) break;
6889 /* End of string1 => advance to string2. */
6891 dend2
= regend
[regno
];
6893 /* At end of register contents => success */
6894 if (d2
== dend2
) break;
6896 /* If necessary, advance to next segment in data. */
6899 /* How many characters left in this segment to match. */
6902 /* Want how many consecutive characters we can match in
6903 one shot, so, if necessary, adjust the count. */
6904 if (mcnt
> dend2
- d2
)
6907 /* Compare that many; failure if mismatch, else move
6910 ? PREFIX(bcmp_translate
) (d
, d2
, mcnt
, translate
)
6911 : memcmp (d
, d2
, mcnt
*sizeof(UCHAR_T
)))
6913 d
+= mcnt
, d2
+= mcnt
;
6915 /* Do this because we've match some characters. */
6916 SET_REGS_MATCHED ();
6922 /* begline matches the empty string at the beginning of the string
6923 (unless `not_bol' is set in `bufp'), and, if
6924 `newline_anchor' is set, after newlines. */
6926 DEBUG_PRINT1 ("EXECUTING begline.\n");
6928 if (AT_STRINGS_BEG (d
))
6930 if (!bufp
->not_bol
) break;
6932 else if (d
[-1] == '\n' && bufp
->newline_anchor
)
6936 /* In all other cases, we fail. */
6940 /* endline is the dual of begline. */
6942 DEBUG_PRINT1 ("EXECUTING endline.\n");
6944 if (AT_STRINGS_END (d
))
6946 if (!bufp
->not_eol
) break;
6949 /* We have to ``prefetch'' the next character. */
6950 else if ((d
== end1
? *string2
: *d
) == '\n'
6951 && bufp
->newline_anchor
)
6958 /* Match at the very beginning of the data. */
6960 DEBUG_PRINT1 ("EXECUTING begbuf.\n");
6961 if (AT_STRINGS_BEG (d
))
6966 /* Match at the very end of the data. */
6968 DEBUG_PRINT1 ("EXECUTING endbuf.\n");
6969 if (AT_STRINGS_END (d
))
6974 /* on_failure_keep_string_jump is used to optimize `.*\n'. It
6975 pushes NULL as the value for the string on the stack. Then
6976 `pop_failure_point' will keep the current value for the
6977 string, instead of restoring it. To see why, consider
6978 matching `foo\nbar' against `.*\n'. The .* matches the foo;
6979 then the . fails against the \n. But the next thing we want
6980 to do is match the \n against the \n; if we restored the
6981 string value, we would be back at the foo.
6983 Because this is used only in specific cases, we don't need to
6984 check all the things that `on_failure_jump' does, to make
6985 sure the right things get saved on the stack. Hence we don't
6986 share its code. The only reason to push anything on the
6987 stack at all is that otherwise we would have to change
6988 `anychar's code to do something besides goto fail in this
6989 case; that seems worse than this. */
6990 case on_failure_keep_string_jump
:
6991 DEBUG_PRINT1 ("EXECUTING on_failure_keep_string_jump");
6993 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
6995 DEBUG_PRINT3 (" %d (to %p):\n", mcnt
, p
+ mcnt
);
6997 DEBUG_PRINT3 (" %d (to 0x%x):\n", mcnt
, p
+ mcnt
);
7000 PUSH_FAILURE_POINT (p
+ mcnt
, NULL
, -2);
7004 /* Uses of on_failure_jump:
7006 Each alternative starts with an on_failure_jump that points
7007 to the beginning of the next alternative. Each alternative
7008 except the last ends with a jump that in effect jumps past
7009 the rest of the alternatives. (They really jump to the
7010 ending jump of the following alternative, because tensioning
7011 these jumps is a hassle.)
7013 Repeats start with an on_failure_jump that points past both
7014 the repetition text and either the following jump or
7015 pop_failure_jump back to this on_failure_jump. */
7016 case on_failure_jump
:
7018 DEBUG_PRINT1 ("EXECUTING on_failure_jump");
7020 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
7022 DEBUG_PRINT3 (" %d (to %p)", mcnt
, p
+ mcnt
);
7024 DEBUG_PRINT3 (" %d (to 0x%x)", mcnt
, p
+ mcnt
);
7027 /* If this on_failure_jump comes right before a group (i.e.,
7028 the original * applied to a group), save the information
7029 for that group and all inner ones, so that if we fail back
7030 to this point, the group's information will be correct.
7031 For example, in \(a*\)*\1, we need the preceding group,
7032 and in \(zz\(a*\)b*\)\2, we need the inner group. */
7034 /* We can't use `p' to check ahead because we push
7035 a failure point to `p + mcnt' after we do this. */
7038 /* We need to skip no_op's before we look for the
7039 start_memory in case this on_failure_jump is happening as
7040 the result of a completed succeed_n, as in \(a\)\{1,3\}b\1
7042 while (p1
< pend
&& (re_opcode_t
) *p1
== no_op
)
7045 if (p1
< pend
&& (re_opcode_t
) *p1
== start_memory
)
7047 /* We have a new highest active register now. This will
7048 get reset at the start_memory we are about to get to,
7049 but we will have saved all the registers relevant to
7050 this repetition op, as described above. */
7051 highest_active_reg
= *(p1
+ 1) + *(p1
+ 2);
7052 if (lowest_active_reg
== NO_LOWEST_ACTIVE_REG
)
7053 lowest_active_reg
= *(p1
+ 1);
7056 DEBUG_PRINT1 (":\n");
7057 PUSH_FAILURE_POINT (p
+ mcnt
, d
, -2);
7061 /* A smart repeat ends with `maybe_pop_jump'.
7062 We change it to either `pop_failure_jump' or `jump'. */
7063 case maybe_pop_jump
:
7064 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
7065 DEBUG_PRINT2 ("EXECUTING maybe_pop_jump %d.\n", mcnt
);
7067 register UCHAR_T
*p2
= p
;
7069 /* Compare the beginning of the repeat with what in the
7070 pattern follows its end. If we can establish that there
7071 is nothing that they would both match, i.e., that we
7072 would have to backtrack because of (as in, e.g., `a*a')
7073 then we can change to pop_failure_jump, because we'll
7074 never have to backtrack.
7076 This is not true in the case of alternatives: in
7077 `(a|ab)*' we do need to backtrack to the `ab' alternative
7078 (e.g., if the string was `ab'). But instead of trying to
7079 detect that here, the alternative has put on a dummy
7080 failure point which is what we will end up popping. */
7082 /* Skip over open/close-group commands.
7083 If what follows this loop is a ...+ construct,
7084 look at what begins its body, since we will have to
7085 match at least one of that. */
7089 && ((re_opcode_t
) *p2
== stop_memory
7090 || (re_opcode_t
) *p2
== start_memory
))
7092 else if (p2
+ 2 + 2 * OFFSET_ADDRESS_SIZE
< pend
7093 && (re_opcode_t
) *p2
== dummy_failure_jump
)
7094 p2
+= 2 + 2 * OFFSET_ADDRESS_SIZE
;
7100 /* p1[0] ... p1[2] are the `on_failure_jump' corresponding
7101 to the `maybe_finalize_jump' of this case. Examine what
7104 /* If we're at the end of the pattern, we can change. */
7107 /* Consider what happens when matching ":\(.*\)"
7108 against ":/". I don't really understand this code
7110 p
[-(1+OFFSET_ADDRESS_SIZE
)] = (UCHAR_T
)
7113 (" End of pattern: change to `pop_failure_jump'.\n");
7116 else if ((re_opcode_t
) *p2
== exactn
7118 || (re_opcode_t
) *p2
== exactn_bin
7120 || (bufp
->newline_anchor
&& (re_opcode_t
) *p2
== endline
))
7123 = *p2
== (UCHAR_T
) endline
? '\n' : p2
[2];
7125 if (((re_opcode_t
) p1
[1+OFFSET_ADDRESS_SIZE
] == exactn
7127 || (re_opcode_t
) p1
[1+OFFSET_ADDRESS_SIZE
] == exactn_bin
7129 ) && p1
[3+OFFSET_ADDRESS_SIZE
] != c
)
7131 p
[-(1+OFFSET_ADDRESS_SIZE
)] = (UCHAR_T
)
7134 DEBUG_PRINT3 (" %C != %C => pop_failure_jump.\n",
7136 (wint_t) p1
[3+OFFSET_ADDRESS_SIZE
]);
7138 DEBUG_PRINT3 (" %c != %c => pop_failure_jump.\n",
7140 (char) p1
[3+OFFSET_ADDRESS_SIZE
]);
7145 else if ((re_opcode_t
) p1
[3] == charset
7146 || (re_opcode_t
) p1
[3] == charset_not
)
7148 int not = (re_opcode_t
) p1
[3] == charset_not
;
7150 if (c
< (unsigned) (p1
[4] * BYTEWIDTH
)
7151 && p1
[5 + c
/ BYTEWIDTH
] & (1 << (c
% BYTEWIDTH
)))
7154 /* `not' is equal to 1 if c would match, which means
7155 that we can't change to pop_failure_jump. */
7158 p
[-3] = (unsigned char) pop_failure_jump
;
7159 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
7162 #endif /* not WCHAR */
7165 else if ((re_opcode_t
) *p2
== charset
)
7167 /* We win if the first character of the loop is not part
7169 if ((re_opcode_t
) p1
[3] == exactn
7170 && ! ((int) p2
[1] * BYTEWIDTH
> (int) p1
[5]
7171 && (p2
[2 + p1
[5] / BYTEWIDTH
]
7172 & (1 << (p1
[5] % BYTEWIDTH
)))))
7174 p
[-3] = (unsigned char) pop_failure_jump
;
7175 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
7178 else if ((re_opcode_t
) p1
[3] == charset_not
)
7181 /* We win if the charset_not inside the loop
7182 lists every character listed in the charset after. */
7183 for (idx
= 0; idx
< (int) p2
[1]; idx
++)
7184 if (! (p2
[2 + idx
] == 0
7185 || (idx
< (int) p1
[4]
7186 && ((p2
[2 + idx
] & ~ p1
[5 + idx
]) == 0))))
7191 p
[-3] = (unsigned char) pop_failure_jump
;
7192 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
7195 else if ((re_opcode_t
) p1
[3] == charset
)
7198 /* We win if the charset inside the loop
7199 has no overlap with the one after the loop. */
7201 idx
< (int) p2
[1] && idx
< (int) p1
[4];
7203 if ((p2
[2 + idx
] & p1
[5 + idx
]) != 0)
7206 if (idx
== p2
[1] || idx
== p1
[4])
7208 p
[-3] = (unsigned char) pop_failure_jump
;
7209 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
7213 #endif /* not WCHAR */
7215 p
-= OFFSET_ADDRESS_SIZE
; /* Point at relative address again. */
7216 if ((re_opcode_t
) p
[-1] != pop_failure_jump
)
7218 p
[-1] = (UCHAR_T
) jump
;
7219 DEBUG_PRINT1 (" Match => jump.\n");
7220 goto unconditional_jump
;
7222 /* Note fall through. */
7225 /* The end of a simple repeat has a pop_failure_jump back to
7226 its matching on_failure_jump, where the latter will push a
7227 failure point. The pop_failure_jump takes off failure
7228 points put on by this pop_failure_jump's matching
7229 on_failure_jump; we got through the pattern to here from the
7230 matching on_failure_jump, so didn't fail. */
7231 case pop_failure_jump
:
7233 /* We need to pass separate storage for the lowest and
7234 highest registers, even though we don't care about the
7235 actual values. Otherwise, we will restore only one
7236 register from the stack, since lowest will == highest in
7237 `pop_failure_point'. */
7238 active_reg_t dummy_low_reg
, dummy_high_reg
;
7239 UCHAR_T
*pdummy
= NULL
;
7240 const CHAR_T
*sdummy
= NULL
;
7242 DEBUG_PRINT1 ("EXECUTING pop_failure_jump.\n");
7243 POP_FAILURE_POINT (sdummy
, pdummy
,
7244 dummy_low_reg
, dummy_high_reg
,
7245 reg_dummy
, reg_dummy
, reg_info_dummy
);
7247 /* Note fall through. */
7251 DEBUG_PRINT2 ("\n%p: ", p
);
7253 DEBUG_PRINT2 ("\n0x%x: ", p
);
7255 /* Note fall through. */
7257 /* Unconditionally jump (without popping any failure points). */
7259 EXTRACT_NUMBER_AND_INCR (mcnt
, p
); /* Get the amount to jump. */
7260 DEBUG_PRINT2 ("EXECUTING jump %d ", mcnt
);
7261 p
+= mcnt
; /* Do the jump. */
7263 DEBUG_PRINT2 ("(to %p).\n", p
);
7265 DEBUG_PRINT2 ("(to 0x%x).\n", p
);
7270 /* We need this opcode so we can detect where alternatives end
7271 in `group_match_null_string_p' et al. */
7273 DEBUG_PRINT1 ("EXECUTING jump_past_alt.\n");
7274 goto unconditional_jump
;
7277 /* Normally, the on_failure_jump pushes a failure point, which
7278 then gets popped at pop_failure_jump. We will end up at
7279 pop_failure_jump, also, and with a pattern of, say, `a+', we
7280 are skipping over the on_failure_jump, so we have to push
7281 something meaningless for pop_failure_jump to pop. */
7282 case dummy_failure_jump
:
7283 DEBUG_PRINT1 ("EXECUTING dummy_failure_jump.\n");
7284 /* It doesn't matter what we push for the string here. What
7285 the code at `fail' tests is the value for the pattern. */
7286 PUSH_FAILURE_POINT (NULL
, NULL
, -2);
7287 goto unconditional_jump
;
7290 /* At the end of an alternative, we need to push a dummy failure
7291 point in case we are followed by a `pop_failure_jump', because
7292 we don't want the failure point for the alternative to be
7293 popped. For example, matching `(a|ab)*' against `aab'
7294 requires that we match the `ab' alternative. */
7295 case push_dummy_failure
:
7296 DEBUG_PRINT1 ("EXECUTING push_dummy_failure.\n");
7297 /* See comments just above at `dummy_failure_jump' about the
7299 PUSH_FAILURE_POINT (NULL
, NULL
, -2);
7302 /* Have to succeed matching what follows at least n times.
7303 After that, handle like `on_failure_jump'. */
7305 EXTRACT_NUMBER (mcnt
, p
+ OFFSET_ADDRESS_SIZE
);
7306 DEBUG_PRINT2 ("EXECUTING succeed_n %d.\n", mcnt
);
7309 /* Originally, this is how many times we HAVE to succeed. */
7313 p
+= OFFSET_ADDRESS_SIZE
;
7314 STORE_NUMBER_AND_INCR (p
, mcnt
);
7316 DEBUG_PRINT3 (" Setting %p to %d.\n", p
- OFFSET_ADDRESS_SIZE
7319 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p
- OFFSET_ADDRESS_SIZE
7326 DEBUG_PRINT2 (" Setting two bytes from %p to no_op.\n",
7327 p
+ OFFSET_ADDRESS_SIZE
);
7329 DEBUG_PRINT2 (" Setting two bytes from 0x%x to no_op.\n",
7330 p
+ OFFSET_ADDRESS_SIZE
);
7334 p
[1] = (UCHAR_T
) no_op
;
7336 p
[2] = (UCHAR_T
) no_op
;
7337 p
[3] = (UCHAR_T
) no_op
;
7344 EXTRACT_NUMBER (mcnt
, p
+ OFFSET_ADDRESS_SIZE
);
7345 DEBUG_PRINT2 ("EXECUTING jump_n %d.\n", mcnt
);
7347 /* Originally, this is how many times we CAN jump. */
7351 STORE_NUMBER (p
+ OFFSET_ADDRESS_SIZE
, mcnt
);
7354 DEBUG_PRINT3 (" Setting %p to %d.\n", p
+ OFFSET_ADDRESS_SIZE
,
7357 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p
+ OFFSET_ADDRESS_SIZE
,
7360 goto unconditional_jump
;
7362 /* If don't have to jump any more, skip over the rest of command. */
7364 p
+= 2 * OFFSET_ADDRESS_SIZE
;
7369 DEBUG_PRINT1 ("EXECUTING set_number_at.\n");
7371 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
7373 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
7375 DEBUG_PRINT3 (" Setting %p to %d.\n", p1
, mcnt
);
7377 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p1
, mcnt
);
7379 STORE_NUMBER (p1
, mcnt
);
7384 /* The DEC Alpha C compiler 3.x generates incorrect code for the
7385 test WORDCHAR_P (d - 1) != WORDCHAR_P (d) in the expansion of
7386 AT_WORD_BOUNDARY, so this code is disabled. Expanding the
7387 macro and introducing temporary variables works around the bug. */
7390 DEBUG_PRINT1 ("EXECUTING wordbound.\n");
7391 if (AT_WORD_BOUNDARY (d
))
7396 DEBUG_PRINT1 ("EXECUTING notwordbound.\n");
7397 if (AT_WORD_BOUNDARY (d
))
7403 boolean prevchar
, thischar
;
7405 DEBUG_PRINT1 ("EXECUTING wordbound.\n");
7406 if (AT_STRINGS_BEG (d
) || AT_STRINGS_END (d
))
7409 prevchar
= WORDCHAR_P (d
- 1);
7410 thischar
= WORDCHAR_P (d
);
7411 if (prevchar
!= thischar
)
7418 boolean prevchar
, thischar
;
7420 DEBUG_PRINT1 ("EXECUTING notwordbound.\n");
7421 if (AT_STRINGS_BEG (d
) || AT_STRINGS_END (d
))
7424 prevchar
= WORDCHAR_P (d
- 1);
7425 thischar
= WORDCHAR_P (d
);
7426 if (prevchar
!= thischar
)
7433 DEBUG_PRINT1 ("EXECUTING wordbeg.\n");
7434 if (!AT_STRINGS_END (d
) && WORDCHAR_P (d
)
7435 && (AT_STRINGS_BEG (d
) || !WORDCHAR_P (d
- 1)))
7440 DEBUG_PRINT1 ("EXECUTING wordend.\n");
7441 if (!AT_STRINGS_BEG (d
) && WORDCHAR_P (d
- 1)
7442 && (AT_STRINGS_END (d
) || !WORDCHAR_P (d
)))
7448 DEBUG_PRINT1 ("EXECUTING before_dot.\n");
7449 if (PTR_CHAR_POS ((unsigned char *) d
) >= point
)
7454 DEBUG_PRINT1 ("EXECUTING at_dot.\n");
7455 if (PTR_CHAR_POS ((unsigned char *) d
) != point
)
7460 DEBUG_PRINT1 ("EXECUTING after_dot.\n");
7461 if (PTR_CHAR_POS ((unsigned char *) d
) <= point
)
7466 DEBUG_PRINT2 ("EXECUTING syntaxspec %d.\n", mcnt
);
7471 DEBUG_PRINT1 ("EXECUTING Emacs wordchar.\n");
7475 /* Can't use *d++ here; SYNTAX may be an unsafe macro. */
7477 if (SYNTAX (d
[-1]) != (enum syntaxcode
) mcnt
)
7479 SET_REGS_MATCHED ();
7483 DEBUG_PRINT2 ("EXECUTING notsyntaxspec %d.\n", mcnt
);
7485 goto matchnotsyntax
;
7488 DEBUG_PRINT1 ("EXECUTING Emacs notwordchar.\n");
7492 /* Can't use *d++ here; SYNTAX may be an unsafe macro. */
7494 if (SYNTAX (d
[-1]) == (enum syntaxcode
) mcnt
)
7496 SET_REGS_MATCHED ();
7499 #else /* not emacs */
7501 DEBUG_PRINT1 ("EXECUTING non-Emacs wordchar.\n");
7503 if (!WORDCHAR_P (d
))
7505 SET_REGS_MATCHED ();
7510 DEBUG_PRINT1 ("EXECUTING non-Emacs notwordchar.\n");
7514 SET_REGS_MATCHED ();
7517 #endif /* not emacs */
7522 continue; /* Successfully executed one pattern command; keep going. */
7525 /* We goto here if a matching operation fails. */
7527 if (!FAIL_STACK_EMPTY ())
7528 { /* A restart point is known. Restore to that state. */
7529 DEBUG_PRINT1 ("\nFAIL:\n");
7530 POP_FAILURE_POINT (d
, p
,
7531 lowest_active_reg
, highest_active_reg
,
7532 regstart
, regend
, reg_info
);
7534 /* If this failure point is a dummy, try the next one. */
7538 /* If we failed to the end of the pattern, don't examine *p. */
7542 boolean is_a_jump_n
= false;
7544 /* If failed to a backwards jump that's part of a repetition
7545 loop, need to pop this failure point and use the next one. */
7546 switch ((re_opcode_t
) *p
)
7550 case maybe_pop_jump
:
7551 case pop_failure_jump
:
7554 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
7557 if ((is_a_jump_n
&& (re_opcode_t
) *p1
== succeed_n
)
7559 && (re_opcode_t
) *p1
== on_failure_jump
))
7567 if (d
>= string1
&& d
<= end1
)
7571 break; /* Matching at this starting point really fails. */
7575 goto restore_best_regs
;
7579 return -1; /* Failure to match. */
7582 /* Subroutine definitions for re_match_2. */
7585 /* We are passed P pointing to a register number after a start_memory.
7587 Return true if the pattern up to the corresponding stop_memory can
7588 match the empty string, and false otherwise.
7590 If we find the matching stop_memory, sets P to point to one past its number.
7591 Otherwise, sets P to an undefined byte less than or equal to END.
7593 We don't handle duplicates properly (yet). */
7596 PREFIX(group_match_null_string_p
) (p
, end
, reg_info
)
7598 PREFIX(register_info_type
) *reg_info
;
7601 /* Point to after the args to the start_memory. */
7602 UCHAR_T
*p1
= *p
+ 2;
7606 /* Skip over opcodes that can match nothing, and return true or
7607 false, as appropriate, when we get to one that can't, or to the
7608 matching stop_memory. */
7610 switch ((re_opcode_t
) *p1
)
7612 /* Could be either a loop or a series of alternatives. */
7613 case on_failure_jump
:
7615 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
7617 /* If the next operation is not a jump backwards in the
7622 /* Go through the on_failure_jumps of the alternatives,
7623 seeing if any of the alternatives cannot match nothing.
7624 The last alternative starts with only a jump,
7625 whereas the rest start with on_failure_jump and end
7626 with a jump, e.g., here is the pattern for `a|b|c':
7628 /on_failure_jump/0/6/exactn/1/a/jump_past_alt/0/6
7629 /on_failure_jump/0/6/exactn/1/b/jump_past_alt/0/3
7632 So, we have to first go through the first (n-1)
7633 alternatives and then deal with the last one separately. */
7636 /* Deal with the first (n-1) alternatives, which start
7637 with an on_failure_jump (see above) that jumps to right
7638 past a jump_past_alt. */
7640 while ((re_opcode_t
) p1
[mcnt
-(1+OFFSET_ADDRESS_SIZE
)] ==
7643 /* `mcnt' holds how many bytes long the alternative
7644 is, including the ending `jump_past_alt' and
7647 if (!PREFIX(alt_match_null_string_p
) (p1
, p1
+ mcnt
-
7648 (1 + OFFSET_ADDRESS_SIZE
),
7652 /* Move to right after this alternative, including the
7656 /* Break if it's the beginning of an n-th alternative
7657 that doesn't begin with an on_failure_jump. */
7658 if ((re_opcode_t
) *p1
!= on_failure_jump
)
7661 /* Still have to check that it's not an n-th
7662 alternative that starts with an on_failure_jump. */
7664 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
7665 if ((re_opcode_t
) p1
[mcnt
-(1+OFFSET_ADDRESS_SIZE
)] !=
7668 /* Get to the beginning of the n-th alternative. */
7669 p1
-= 1 + OFFSET_ADDRESS_SIZE
;
7674 /* Deal with the last alternative: go back and get number
7675 of the `jump_past_alt' just before it. `mcnt' contains
7676 the length of the alternative. */
7677 EXTRACT_NUMBER (mcnt
, p1
- OFFSET_ADDRESS_SIZE
);
7679 if (!PREFIX(alt_match_null_string_p
) (p1
, p1
+ mcnt
, reg_info
))
7682 p1
+= mcnt
; /* Get past the n-th alternative. */
7688 assert (p1
[1] == **p
);
7694 if (!PREFIX(common_op_match_null_string_p
) (&p1
, end
, reg_info
))
7697 } /* while p1 < end */
7700 } /* group_match_null_string_p */
7703 /* Similar to group_match_null_string_p, but doesn't deal with alternatives:
7704 It expects P to be the first byte of a single alternative and END one
7705 byte past the last. The alternative can contain groups. */
7708 PREFIX(alt_match_null_string_p
) (p
, end
, reg_info
)
7710 PREFIX(register_info_type
) *reg_info
;
7717 /* Skip over opcodes that can match nothing, and break when we get
7718 to one that can't. */
7720 switch ((re_opcode_t
) *p1
)
7723 case on_failure_jump
:
7725 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
7730 if (!PREFIX(common_op_match_null_string_p
) (&p1
, end
, reg_info
))
7733 } /* while p1 < end */
7736 } /* alt_match_null_string_p */
7739 /* Deals with the ops common to group_match_null_string_p and
7740 alt_match_null_string_p.
7742 Sets P to one after the op and its arguments, if any. */
7745 PREFIX(common_op_match_null_string_p
) (p
, end
, reg_info
)
7747 PREFIX(register_info_type
) *reg_info
;
7754 switch ((re_opcode_t
) *p1
++)
7774 assert (reg_no
> 0 && reg_no
<= MAX_REGNUM
);
7775 ret
= PREFIX(group_match_null_string_p
) (&p1
, end
, reg_info
);
7777 /* Have to set this here in case we're checking a group which
7778 contains a group and a back reference to it. */
7780 if (REG_MATCH_NULL_STRING_P (reg_info
[reg_no
]) == MATCH_NULL_UNSET_VALUE
)
7781 REG_MATCH_NULL_STRING_P (reg_info
[reg_no
]) = ret
;
7787 /* If this is an optimized succeed_n for zero times, make the jump. */
7789 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
7797 /* Get to the number of times to succeed. */
7798 p1
+= OFFSET_ADDRESS_SIZE
;
7799 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
7803 p1
-= 2 * OFFSET_ADDRESS_SIZE
;
7804 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
7812 if (!REG_MATCH_NULL_STRING_P (reg_info
[*p1
]))
7817 p1
+= 2 * OFFSET_ADDRESS_SIZE
;
7820 /* All other opcodes mean we cannot match the empty string. */
7826 } /* common_op_match_null_string_p */
7829 /* Return zero if TRANSLATE[S1] and TRANSLATE[S2] are identical for LEN
7830 bytes; nonzero otherwise. */
7833 PREFIX(bcmp_translate
) (s1
, s2
, len
, translate
)
7834 const CHAR_T
*s1
, *s2
;
7836 RE_TRANSLATE_TYPE translate
;
7838 register const UCHAR_T
*p1
= (const UCHAR_T
*) s1
;
7839 register const UCHAR_T
*p2
= (const UCHAR_T
*) s2
;
7843 if (((*p1
<=0xff)?translate
[*p1
++]:*p1
++)
7844 != ((*p2
<=0xff)?translate
[*p2
++]:*p2
++))
7847 if (translate
[*p1
++] != translate
[*p2
++]) return 1;
7855 #else /* not INSIDE_RECURSION */
7857 /* Entry points for GNU code. */
7859 /* re_compile_pattern is the GNU regular expression compiler: it
7860 compiles PATTERN (of length SIZE) and puts the result in BUFP.
7861 Returns 0 if the pattern was valid, otherwise an error string.
7863 Assumes the `allocated' (and perhaps `buffer') and `translate' fields
7864 are set in BUFP on entry.
7866 We call regex_compile to do the actual compilation. */
7869 re_compile_pattern (pattern
, length
, bufp
)
7870 const char *pattern
;
7872 struct re_pattern_buffer
*bufp
;
7876 /* GNU code is written to assume at least RE_NREGS registers will be set
7877 (and at least one extra will be -1). */
7878 bufp
->regs_allocated
= REGS_UNALLOCATED
;
7880 /* And GNU code determines whether or not to get register information
7881 by passing null for the REGS argument to re_match, etc., not by
7885 /* Match anchors at newline. */
7886 bufp
->newline_anchor
= 1;
7889 if (MB_CUR_MAX
!= 1)
7890 ret
= wcs_regex_compile (pattern
, length
, re_syntax_options
, bufp
);
7893 ret
= byte_regex_compile (pattern
, length
, re_syntax_options
, bufp
);
7897 return gettext (re_error_msgid
[(int) ret
]);
7900 weak_alias (__re_compile_pattern
, re_compile_pattern
)
7903 /* Entry points compatible with 4.2 BSD regex library. We don't define
7904 them unless specifically requested. */
7906 #if defined _REGEX_RE_COMP || defined _LIBC
7908 /* BSD has one and only one pattern buffer. */
7909 static struct re_pattern_buffer re_comp_buf
;
7913 /* Make these definitions weak in libc, so POSIX programs can redefine
7914 these names if they don't use our functions, and still use
7915 regcomp/regexec below without link errors. */
7925 if (!re_comp_buf
.buffer
)
7926 return gettext ("No previous regular expression");
7930 if (!re_comp_buf
.buffer
)
7932 re_comp_buf
.buffer
= (unsigned char *) malloc (200);
7933 if (re_comp_buf
.buffer
== NULL
)
7934 return (char *) gettext (re_error_msgid
[(int) REG_ESPACE
]);
7935 re_comp_buf
.allocated
= 200;
7937 re_comp_buf
.fastmap
= (char *) malloc (1 << BYTEWIDTH
);
7938 if (re_comp_buf
.fastmap
== NULL
)
7939 return (char *) gettext (re_error_msgid
[(int) REG_ESPACE
]);
7942 /* Since `re_exec' always passes NULL for the `regs' argument, we
7943 don't need to initialize the pattern buffer fields which affect it. */
7945 /* Match anchors at newlines. */
7946 re_comp_buf
.newline_anchor
= 1;
7949 if (MB_CUR_MAX
!= 1)
7950 ret
= wcs_regex_compile (s
, strlen (s
), re_syntax_options
, &re_comp_buf
);
7953 ret
= byte_regex_compile (s
, strlen (s
), re_syntax_options
, &re_comp_buf
);
7958 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
7959 return (char *) gettext (re_error_msgid
[(int) ret
]);
7970 const int len
= strlen (s
);
7972 0 <= re_search (&re_comp_buf
, s
, len
, 0, len
, (struct re_registers
*) 0);
7975 #endif /* _REGEX_RE_COMP */
7977 /* POSIX.2 functions. Don't define these for Emacs. */
7981 /* regcomp takes a regular expression as a string and compiles it.
7983 PREG is a regex_t *. We do not expect any fields to be initialized,
7984 since POSIX says we shouldn't. Thus, we set
7986 `buffer' to the compiled pattern;
7987 `used' to the length of the compiled pattern;
7988 `syntax' to RE_SYNTAX_POSIX_EXTENDED if the
7989 REG_EXTENDED bit in CFLAGS is set; otherwise, to
7990 RE_SYNTAX_POSIX_BASIC;
7991 `newline_anchor' to REG_NEWLINE being set in CFLAGS;
7992 `fastmap' to an allocated space for the fastmap;
7993 `fastmap_accurate' to zero;
7994 `re_nsub' to the number of subexpressions in PATTERN.
7996 PATTERN is the address of the pattern string.
7998 CFLAGS is a series of bits which affect compilation.
8000 If REG_EXTENDED is set, we use POSIX extended syntax; otherwise, we
8001 use POSIX basic syntax.
8003 If REG_NEWLINE is set, then . and [^...] don't match newline.
8004 Also, regexec will try a match beginning after every newline.
8006 If REG_ICASE is set, then we considers upper- and lowercase
8007 versions of letters to be equivalent when matching.
8009 If REG_NOSUB is set, then when PREG is passed to regexec, that
8010 routine will report only success or failure, and nothing about the
8013 It returns 0 if it succeeds, nonzero if it doesn't. (See regex.h for
8014 the return codes and their meanings.) */
8017 regcomp (preg
, pattern
, cflags
)
8019 const char *pattern
;
8024 = (cflags
& REG_EXTENDED
) ?
8025 RE_SYNTAX_POSIX_EXTENDED
: RE_SYNTAX_POSIX_BASIC
;
8027 /* regex_compile will allocate the space for the compiled pattern. */
8029 preg
->allocated
= 0;
8032 /* Try to allocate space for the fastmap. */
8033 preg
->fastmap
= (char *) malloc (1 << BYTEWIDTH
);
8035 if (cflags
& REG_ICASE
)
8040 = (RE_TRANSLATE_TYPE
) malloc (CHAR_SET_SIZE
8041 * sizeof (*(RE_TRANSLATE_TYPE
)0));
8042 if (preg
->translate
== NULL
)
8043 return (int) REG_ESPACE
;
8045 /* Map uppercase characters to corresponding lowercase ones. */
8046 for (i
= 0; i
< CHAR_SET_SIZE
; i
++)
8047 preg
->translate
[i
] = ISUPPER (i
) ? TOLOWER (i
) : (int) i
;
8050 preg
->translate
= NULL
;
8052 /* If REG_NEWLINE is set, newlines are treated differently. */
8053 if (cflags
& REG_NEWLINE
)
8054 { /* REG_NEWLINE implies neither . nor [^...] match newline. */
8055 syntax
&= ~RE_DOT_NEWLINE
;
8056 syntax
|= RE_HAT_LISTS_NOT_NEWLINE
;
8057 /* It also changes the matching behavior. */
8058 preg
->newline_anchor
= 1;
8061 preg
->newline_anchor
= 0;
8063 preg
->no_sub
= !!(cflags
& REG_NOSUB
);
8065 /* POSIX says a null character in the pattern terminates it, so we
8066 can use strlen here in compiling the pattern. */
8068 if (MB_CUR_MAX
!= 1)
8069 ret
= wcs_regex_compile (pattern
, strlen (pattern
), syntax
, preg
);
8072 ret
= byte_regex_compile (pattern
, strlen (pattern
), syntax
, preg
);
8074 /* POSIX doesn't distinguish between an unmatched open-group and an
8075 unmatched close-group: both are REG_EPAREN. */
8076 if (ret
== REG_ERPAREN
) ret
= REG_EPAREN
;
8078 if (ret
== REG_NOERROR
&& preg
->fastmap
)
8080 /* Compute the fastmap now, since regexec cannot modify the pattern
8082 if (re_compile_fastmap (preg
) == -2)
8084 /* Some error occurred while computing the fastmap, just forget
8086 free (preg
->fastmap
);
8087 preg
->fastmap
= NULL
;
8094 weak_alias (__regcomp
, regcomp
)
8098 /* regexec searches for a given pattern, specified by PREG, in the
8101 If NMATCH is zero or REG_NOSUB was set in the cflags argument to
8102 `regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at
8103 least NMATCH elements, and we set them to the offsets of the
8104 corresponding matched substrings.
8106 EFLAGS specifies `execution flags' which affect matching: if
8107 REG_NOTBOL is set, then ^ does not match at the beginning of the
8108 string; if REG_NOTEOL is set, then $ does not match at the end.
8110 We return 0 if we find a match and REG_NOMATCH if not. */
8113 regexec (preg
, string
, nmatch
, pmatch
, eflags
)
8114 const regex_t
*preg
;
8117 regmatch_t pmatch
[];
8121 struct re_registers regs
;
8122 regex_t private_preg
;
8123 int len
= strlen (string
);
8124 boolean want_reg_info
= !preg
->no_sub
&& nmatch
> 0;
8126 private_preg
= *preg
;
8128 private_preg
.not_bol
= !!(eflags
& REG_NOTBOL
);
8129 private_preg
.not_eol
= !!(eflags
& REG_NOTEOL
);
8131 /* The user has told us exactly how many registers to return
8132 information about, via `nmatch'. We have to pass that on to the
8133 matching routines. */
8134 private_preg
.regs_allocated
= REGS_FIXED
;
8138 regs
.num_regs
= nmatch
;
8139 regs
.start
= TALLOC (nmatch
* 2, regoff_t
);
8140 if (regs
.start
== NULL
)
8141 return (int) REG_NOMATCH
;
8142 regs
.end
= regs
.start
+ nmatch
;
8145 /* Perform the searching operation. */
8146 ret
= re_search (&private_preg
, string
, len
,
8147 /* start: */ 0, /* range: */ len
,
8148 want_reg_info
? ®s
: (struct re_registers
*) 0);
8150 /* Copy the register information to the POSIX structure. */
8157 for (r
= 0; r
< nmatch
; r
++)
8159 pmatch
[r
].rm_so
= regs
.start
[r
];
8160 pmatch
[r
].rm_eo
= regs
.end
[r
];
8164 /* If we needed the temporary register info, free the space now. */
8168 /* We want zero return to mean success, unlike `re_search'. */
8169 return ret
>= 0 ? (int) REG_NOERROR
: (int) REG_NOMATCH
;
8172 weak_alias (__regexec
, regexec
)
8176 /* Returns a message corresponding to an error code, ERRCODE, returned
8177 from either regcomp or regexec. We don't use PREG here. */
8180 regerror (errcode
, preg
, errbuf
, errbuf_size
)
8182 const regex_t
*preg ATTRIBUTE_UNUSED
;
8190 || errcode
>= (int) (sizeof (re_error_msgid
)
8191 / sizeof (re_error_msgid
[0])))
8192 /* Only error codes returned by the rest of the code should be passed
8193 to this routine. If we are given anything else, or if other regex
8194 code generates an invalid error code, then the program has a bug.
8195 Dump core so we can fix it. */
8198 msg
= gettext (re_error_msgid
[errcode
]);
8200 msg_size
= strlen (msg
) + 1; /* Includes the null. */
8202 if (errbuf_size
!= 0)
8204 if (msg_size
> errbuf_size
)
8206 #if defined HAVE_MEMPCPY || defined _LIBC
8207 *((char *) mempcpy (errbuf
, msg
, errbuf_size
- 1)) = '\0';
8209 memcpy (errbuf
, msg
, errbuf_size
- 1);
8210 errbuf
[errbuf_size
- 1] = 0;
8214 memcpy (errbuf
, msg
, msg_size
);
8220 weak_alias (__regerror
, regerror
)
8224 /* Free dynamically allocated space used by PREG. */
8230 if (preg
->buffer
!= NULL
)
8231 free (preg
->buffer
);
8232 preg
->buffer
= NULL
;
8234 preg
->allocated
= 0;
8237 if (preg
->fastmap
!= NULL
)
8238 free (preg
->fastmap
);
8239 preg
->fastmap
= NULL
;
8240 preg
->fastmap_accurate
= 0;
8242 if (preg
->translate
!= NULL
)
8243 free (preg
->translate
);
8244 preg
->translate
= NULL
;
8247 weak_alias (__regfree
, regfree
)
8250 #endif /* not emacs */
8252 #endif /* not INSIDE_RECURSION */
8256 #undef STORE_NUMBER_AND_INCR
8257 #undef EXTRACT_NUMBER
8258 #undef EXTRACT_NUMBER_AND_INCR
8260 #undef DEBUG_PRINT_COMPILED_PATTERN
8261 #undef DEBUG_PRINT_DOUBLE_STRING
8263 #undef INIT_FAIL_STACK
8264 #undef RESET_FAIL_STACK
8265 #undef DOUBLE_FAIL_STACK
8266 #undef PUSH_PATTERN_OP
8267 #undef PUSH_FAILURE_POINTER
8268 #undef PUSH_FAILURE_INT
8269 #undef PUSH_FAILURE_ELT
8270 #undef POP_FAILURE_POINTER
8271 #undef POP_FAILURE_INT
8272 #undef POP_FAILURE_ELT
8275 #undef PUSH_FAILURE_POINT
8276 #undef POP_FAILURE_POINT
8278 #undef REG_UNSET_VALUE
8286 #undef INIT_BUF_SIZE
8287 #undef GET_BUFFER_SPACE
8295 #undef EXTEND_BUFFER
8296 #undef GET_UNSIGNED_NUMBER
8297 #undef FREE_STACK_RETURN
8299 # undef POINTER_TO_OFFSET
8300 # undef MATCHING_IN_FRST_STRING
8302 # undef AT_STRINGS_BEG
8303 # undef AT_STRINGS_END
8306 # undef FREE_VARIABLES
8307 # undef NO_HIGHEST_ACTIVE_REG
8308 # undef NO_LOWEST_ACTIVE_REG
8312 # undef COMPILED_BUFFER_VAR
8313 # undef OFFSET_ADDRESS_SIZE
8314 # undef CHAR_CLASS_SIZE
8321 # define DEFINED_ONCE